Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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/ctype.h>
17#include <linux/kernel.h>
18#include <linux/export.h>
19#include <linux/init.h>
20#include <linux/pci.h>
21#include <linux/delay.h>
22#include <linux/hardirq.h>
23#include <linux/scatterlist.h>
24#include <linux/blk-mq.h>
25#include <linux/blk-integrity.h>
26#include <linux/ratelimit.h>
27#include <linux/unaligned.h>
28
29#include <scsi/scsi.h>
30#include <scsi/scsi_cmnd.h>
31#include <scsi/scsi_dbg.h>
32#include <scsi/scsi_device.h>
33#include <scsi/scsi_driver.h>
34#include <scsi/scsi_eh.h>
35#include <scsi/scsi_host.h>
36#include <scsi/scsi_transport.h> /* scsi_init_limits() */
37#include <scsi/scsi_dh.h>
38
39#include <trace/events/scsi.h>
40
41#include "scsi_debugfs.h"
42#include "scsi_priv.h"
43#include "scsi_logging.h"
44
45/*
46 * Size of integrity metadata is usually small, 1 inline sg should
47 * cover normal cases.
48 */
49#ifdef CONFIG_ARCH_NO_SG_CHAIN
50#define SCSI_INLINE_PROT_SG_CNT 0
51#define SCSI_INLINE_SG_CNT 0
52#else
53#define SCSI_INLINE_PROT_SG_CNT 1
54#define SCSI_INLINE_SG_CNT 2
55#endif
56
57static struct kmem_cache *scsi_sense_cache;
58static DEFINE_MUTEX(scsi_sense_cache_mutex);
59
60static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
61
62int scsi_init_sense_cache(struct Scsi_Host *shost)
63{
64 int ret = 0;
65
66 mutex_lock(&scsi_sense_cache_mutex);
67 if (!scsi_sense_cache) {
68 scsi_sense_cache =
69 kmem_cache_create_usercopy("scsi_sense_cache",
70 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
71 0, SCSI_SENSE_BUFFERSIZE, NULL);
72 if (!scsi_sense_cache)
73 ret = -ENOMEM;
74 }
75 mutex_unlock(&scsi_sense_cache_mutex);
76 return ret;
77}
78
79static void
80scsi_set_blocked(struct scsi_cmnd *cmd, enum scsi_qc_status reason)
81{
82 struct Scsi_Host *host = cmd->device->host;
83 struct scsi_device *device = cmd->device;
84 struct scsi_target *starget = scsi_target(device);
85
86 /*
87 * Set the appropriate busy bit for the device/host.
88 *
89 * If the host/device isn't busy, assume that something actually
90 * completed, and that we should be able to queue a command now.
91 *
92 * Note that the prior mid-layer assumption that any host could
93 * always queue at least one command is now broken. The mid-layer
94 * will implement a user specifiable stall (see
95 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
96 * if a command is requeued with no other commands outstanding
97 * either for the device or for the host.
98 */
99 switch (reason) {
100 case SCSI_MLQUEUE_HOST_BUSY:
101 atomic_set(&host->host_blocked, host->max_host_blocked);
102 break;
103 case SCSI_MLQUEUE_DEVICE_BUSY:
104 case SCSI_MLQUEUE_EH_RETRY:
105 atomic_set(&device->device_blocked,
106 device->max_device_blocked);
107 break;
108 case SCSI_MLQUEUE_TARGET_BUSY:
109 atomic_set(&starget->target_blocked,
110 starget->max_target_blocked);
111 break;
112 }
113}
114
115static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd, unsigned long msecs)
116{
117 struct request *rq = scsi_cmd_to_rq(cmd);
118
119 if (rq->rq_flags & RQF_DONTPREP) {
120 rq->rq_flags &= ~RQF_DONTPREP;
121 scsi_mq_uninit_cmd(cmd);
122 } else {
123 WARN_ON_ONCE(true);
124 }
125
126 blk_mq_requeue_request(rq, false);
127 if (!scsi_host_in_recovery(cmd->device->host))
128 blk_mq_delay_kick_requeue_list(rq->q, msecs);
129}
130
131/**
132 * __scsi_queue_insert - private queue insertion
133 * @cmd: The SCSI command being requeued
134 * @reason: The reason for the requeue
135 * @unbusy: Whether the queue should be unbusied
136 *
137 * This is a private queue insertion. The public interface
138 * scsi_queue_insert() always assumes the queue should be unbusied
139 * because it's always called before the completion. This function is
140 * for a requeue after completion, which should only occur in this
141 * file.
142 */
143static void __scsi_queue_insert(struct scsi_cmnd *cmd,
144 enum scsi_qc_status reason, bool unbusy)
145{
146 struct scsi_device *device = cmd->device;
147
148 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
149 "Inserting command %p into mlqueue\n", cmd));
150
151 scsi_set_blocked(cmd, reason);
152
153 /*
154 * Decrement the counters, since these commands are no longer
155 * active on the host/device.
156 */
157 if (unbusy)
158 scsi_device_unbusy(device, cmd);
159
160 /*
161 * Requeue this command. It will go before all other commands
162 * that are already in the queue. Schedule requeue work under
163 * lock such that the kblockd_schedule_work() call happens
164 * before blk_mq_destroy_queue() finishes.
165 */
166 cmd->result = 0;
167
168 blk_mq_requeue_request(scsi_cmd_to_rq(cmd),
169 !scsi_host_in_recovery(cmd->device->host));
170}
171
172/**
173 * scsi_queue_insert - Reinsert a command in the queue.
174 * @cmd: command that we are adding to queue.
175 * @reason: why we are inserting command to queue.
176 *
177 * We do this for one of two cases. Either the host is busy and it cannot accept
178 * any more commands for the time being, or the device returned QUEUE_FULL and
179 * can accept no more commands.
180 *
181 * Context: This could be called either from an interrupt context or a normal
182 * process context.
183 */
184void scsi_queue_insert(struct scsi_cmnd *cmd, enum scsi_qc_status reason)
185{
186 __scsi_queue_insert(cmd, reason, true);
187}
188
189/**
190 * scsi_failures_reset_retries - reset all failures to zero
191 * @failures: &struct scsi_failures with specific failure modes set
192 */
193void scsi_failures_reset_retries(struct scsi_failures *failures)
194{
195 struct scsi_failure *failure;
196
197 failures->total_retries = 0;
198
199 for (failure = failures->failure_definitions; failure->result;
200 failure++)
201 failure->retries = 0;
202}
203EXPORT_SYMBOL_GPL(scsi_failures_reset_retries);
204
205/**
206 * scsi_check_passthrough - Determine if passthrough scsi_cmnd needs a retry.
207 * @scmd: scsi_cmnd to check.
208 * @failures: scsi_failures struct that lists failures to check for.
209 *
210 * Returns -EAGAIN if the caller should retry else 0.
211 */
212static int scsi_check_passthrough(struct scsi_cmnd *scmd,
213 struct scsi_failures *failures)
214{
215 struct scsi_failure *failure;
216 struct scsi_sense_hdr sshdr;
217 enum sam_status status;
218
219 if (!scmd->result)
220 return 0;
221
222 if (!failures)
223 return 0;
224
225 for (failure = failures->failure_definitions; failure->result;
226 failure++) {
227 if (failure->result == SCMD_FAILURE_RESULT_ANY)
228 goto maybe_retry;
229
230 if (host_byte(scmd->result) &&
231 host_byte(scmd->result) == host_byte(failure->result))
232 goto maybe_retry;
233
234 status = status_byte(scmd->result);
235 if (!status)
236 continue;
237
238 if (failure->result == SCMD_FAILURE_STAT_ANY &&
239 !scsi_status_is_good(scmd->result))
240 goto maybe_retry;
241
242 if (status != status_byte(failure->result))
243 continue;
244
245 if (status_byte(failure->result) != SAM_STAT_CHECK_CONDITION ||
246 failure->sense == SCMD_FAILURE_SENSE_ANY)
247 goto maybe_retry;
248
249 if (!scsi_command_normalize_sense(scmd, &sshdr))
250 return 0;
251
252 if (failure->sense != sshdr.sense_key)
253 continue;
254
255 if (failure->asc == SCMD_FAILURE_ASC_ANY)
256 goto maybe_retry;
257
258 if (failure->asc != sshdr.asc)
259 continue;
260
261 if (failure->ascq == SCMD_FAILURE_ASCQ_ANY ||
262 failure->ascq == sshdr.ascq)
263 goto maybe_retry;
264 }
265
266 return 0;
267
268maybe_retry:
269 if (failure->allowed) {
270 if (failure->allowed == SCMD_FAILURE_NO_LIMIT ||
271 ++failure->retries <= failure->allowed)
272 return -EAGAIN;
273 } else {
274 if (failures->total_allowed == SCMD_FAILURE_NO_LIMIT ||
275 ++failures->total_retries <= failures->total_allowed)
276 return -EAGAIN;
277 }
278
279 return 0;
280}
281
282/**
283 * scsi_execute_cmd - insert request and wait for the result
284 * @sdev: scsi_device
285 * @cmd: scsi command
286 * @opf: block layer request cmd_flags
287 * @buffer: data buffer
288 * @bufflen: len of buffer
289 * @timeout: request timeout in HZ
290 * @ml_retries: number of times SCSI midlayer will retry request
291 * @args: Optional args. See struct definition for field descriptions
292 *
293 * Returns the scsi_cmnd result field if a command was executed, or a negative
294 * Linux error code if we didn't get that far.
295 */
296int scsi_execute_cmd(struct scsi_device *sdev, const unsigned char *cmd,
297 blk_opf_t opf, void *buffer, unsigned int bufflen,
298 int timeout, int ml_retries,
299 const struct scsi_exec_args *args)
300{
301 static const struct scsi_exec_args default_args;
302 struct request *req;
303 struct scsi_cmnd *scmd;
304 int ret;
305
306 if (!args)
307 args = &default_args;
308 else if (WARN_ON_ONCE(args->sense &&
309 args->sense_len != SCSI_SENSE_BUFFERSIZE))
310 return -EINVAL;
311
312retry:
313 req = scsi_alloc_request(sdev->request_queue, opf, args->req_flags);
314 if (IS_ERR(req))
315 return PTR_ERR(req);
316
317 if (bufflen) {
318 ret = blk_rq_map_kern(req, buffer, bufflen, GFP_NOIO);
319 if (ret)
320 goto out;
321 }
322 scmd = blk_mq_rq_to_pdu(req);
323 scmd->cmd_len = COMMAND_SIZE(cmd[0]);
324 memcpy(scmd->cmnd, cmd, scmd->cmd_len);
325 scmd->allowed = ml_retries;
326 scmd->flags |= args->scmd_flags;
327 req->timeout = timeout;
328 req->rq_flags |= RQF_QUIET;
329
330 /*
331 * head injection *required* here otherwise quiesce won't work
332 */
333 blk_execute_rq(req, true);
334
335 if (scsi_check_passthrough(scmd, args->failures) == -EAGAIN) {
336 blk_mq_free_request(req);
337 goto retry;
338 }
339
340 /*
341 * Some devices (USB mass-storage in particular) may transfer
342 * garbage data together with a residue indicating that the data
343 * is invalid. Prevent the garbage from being misinterpreted
344 * and prevent security leaks by zeroing out the excess data.
345 */
346 if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
347 memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);
348
349 if (args->resid)
350 *args->resid = scmd->resid_len;
351 if (args->sense)
352 memcpy(args->sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
353 if (args->sshdr)
354 scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,
355 args->sshdr);
356
357 ret = scmd->result;
358 out:
359 blk_mq_free_request(req);
360
361 return ret;
362}
363EXPORT_SYMBOL(scsi_execute_cmd);
364
365/*
366 * Wake up the error handler if necessary. Avoid as follows that the error
367 * handler is not woken up if host in-flight requests number ==
368 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
369 * with an RCU read lock in this function to ensure that this function in
370 * its entirety either finishes before scsi_eh_scmd_add() increases the
371 * host_failed counter or that it notices the shost state change made by
372 * scsi_eh_scmd_add().
373 */
374static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
375{
376 unsigned long flags;
377
378 rcu_read_lock();
379 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
380 if (unlikely(scsi_host_in_recovery(shost))) {
381 /*
382 * Ensure the clear of SCMD_STATE_INFLIGHT is visible to
383 * other CPUs before counting busy requests. Otherwise,
384 * reordering can cause CPUs to race and miss an eh wakeup
385 * when no CPU sees all busy requests as done or timed out.
386 */
387 smp_mb();
388
389 unsigned int busy = scsi_host_busy(shost);
390
391 spin_lock_irqsave(shost->host_lock, flags);
392 if (shost->host_failed || shost->host_eh_scheduled)
393 scsi_eh_wakeup(shost, busy);
394 spin_unlock_irqrestore(shost->host_lock, flags);
395 }
396 rcu_read_unlock();
397}
398
399void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
400{
401 struct Scsi_Host *shost = sdev->host;
402 struct scsi_target *starget = scsi_target(sdev);
403
404 scsi_dec_host_busy(shost, cmd);
405
406 if (starget->can_queue > 0)
407 atomic_dec(&starget->target_busy);
408
409 if (sdev->budget_map.map)
410 sbitmap_put(&sdev->budget_map, cmd->budget_token);
411 cmd->budget_token = -1;
412}
413
414/*
415 * Kick the queue of SCSI device @sdev if @sdev != current_sdev. Called with
416 * interrupts disabled.
417 */
418static void scsi_kick_sdev_queue(struct scsi_device *sdev, void *data)
419{
420 struct scsi_device *current_sdev = data;
421
422 if (sdev != current_sdev)
423 blk_mq_run_hw_queues(sdev->request_queue, true);
424}
425
426/*
427 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
428 * and call blk_run_queue for all the scsi_devices on the target -
429 * including current_sdev first.
430 *
431 * Called with *no* scsi locks held.
432 */
433static void scsi_single_lun_run(struct scsi_device *current_sdev)
434{
435 struct Scsi_Host *shost = current_sdev->host;
436 struct scsi_target *starget = scsi_target(current_sdev);
437 unsigned long flags;
438
439 spin_lock_irqsave(shost->host_lock, flags);
440 starget->starget_sdev_user = NULL;
441 spin_unlock_irqrestore(shost->host_lock, flags);
442
443 /*
444 * Call blk_run_queue for all LUNs on the target, starting with
445 * current_sdev. We race with others (to set starget_sdev_user),
446 * but in most cases, we will be first. Ideally, each LU on the
447 * target would get some limited time or requests on the target.
448 */
449 blk_mq_run_hw_queues(current_sdev->request_queue,
450 shost->queuecommand_may_block);
451
452 spin_lock_irqsave(shost->host_lock, flags);
453 if (!starget->starget_sdev_user)
454 __starget_for_each_device(starget, current_sdev,
455 scsi_kick_sdev_queue);
456 spin_unlock_irqrestore(shost->host_lock, flags);
457}
458
459static inline bool scsi_device_is_busy(struct scsi_device *sdev)
460{
461 if (scsi_device_busy(sdev) >= sdev->queue_depth)
462 return true;
463 if (atomic_read(&sdev->device_blocked) > 0)
464 return true;
465 return false;
466}
467
468static inline bool scsi_target_is_busy(struct scsi_target *starget)
469{
470 if (starget->can_queue > 0) {
471 if (atomic_read(&starget->target_busy) >= starget->can_queue)
472 return true;
473 if (atomic_read(&starget->target_blocked) > 0)
474 return true;
475 }
476 return false;
477}
478
479static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
480{
481 if (atomic_read(&shost->host_blocked) > 0)
482 return true;
483 if (shost->host_self_blocked)
484 return true;
485 return false;
486}
487
488static void scsi_starved_list_run(struct Scsi_Host *shost)
489{
490 LIST_HEAD(starved_list);
491 struct scsi_device *sdev;
492 unsigned long flags;
493
494 spin_lock_irqsave(shost->host_lock, flags);
495 list_splice_init(&shost->starved_list, &starved_list);
496
497 while (!list_empty(&starved_list)) {
498 struct request_queue *slq;
499
500 /*
501 * As long as shost is accepting commands and we have
502 * starved queues, call blk_run_queue. scsi_request_fn
503 * drops the queue_lock and can add us back to the
504 * starved_list.
505 *
506 * host_lock protects the starved_list and starved_entry.
507 * scsi_request_fn must get the host_lock before checking
508 * or modifying starved_list or starved_entry.
509 */
510 if (scsi_host_is_busy(shost))
511 break;
512
513 sdev = list_entry(starved_list.next,
514 struct scsi_device, starved_entry);
515 list_del_init(&sdev->starved_entry);
516 if (scsi_target_is_busy(scsi_target(sdev))) {
517 list_move_tail(&sdev->starved_entry,
518 &shost->starved_list);
519 continue;
520 }
521
522 /*
523 * Once we drop the host lock, a racing scsi_remove_device()
524 * call may remove the sdev from the starved list and destroy
525 * it and the queue. Mitigate by taking a reference to the
526 * queue and never touching the sdev again after we drop the
527 * host lock. Note: if __scsi_remove_device() invokes
528 * blk_mq_destroy_queue() before the queue is run from this
529 * function then blk_run_queue() will return immediately since
530 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
531 */
532 slq = sdev->request_queue;
533 if (!blk_get_queue(slq))
534 continue;
535 spin_unlock_irqrestore(shost->host_lock, flags);
536
537 blk_mq_run_hw_queues(slq, false);
538 blk_put_queue(slq);
539
540 spin_lock_irqsave(shost->host_lock, flags);
541 }
542 /* put any unprocessed entries back */
543 list_splice(&starved_list, &shost->starved_list);
544 spin_unlock_irqrestore(shost->host_lock, flags);
545}
546
547/**
548 * scsi_run_queue - Select a proper request queue to serve next.
549 * @q: last request's queue
550 *
551 * The previous command was completely finished, start a new one if possible.
552 */
553static void scsi_run_queue(struct request_queue *q)
554{
555 struct scsi_device *sdev = q->queuedata;
556
557 if (scsi_target(sdev)->single_lun)
558 scsi_single_lun_run(sdev);
559 if (!list_empty(&sdev->host->starved_list))
560 scsi_starved_list_run(sdev->host);
561
562 /* Note: blk_mq_kick_requeue_list() runs the queue asynchronously. */
563 blk_mq_kick_requeue_list(q);
564}
565
566void scsi_requeue_run_queue(struct work_struct *work)
567{
568 struct scsi_device *sdev;
569 struct request_queue *q;
570
571 sdev = container_of(work, struct scsi_device, requeue_work);
572 q = sdev->request_queue;
573 scsi_run_queue(q);
574}
575
576void scsi_run_host_queues(struct Scsi_Host *shost)
577{
578 struct scsi_device *sdev;
579
580 shost_for_each_device(sdev, shost)
581 scsi_run_queue(sdev->request_queue);
582}
583
584static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
585{
586 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
587 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
588
589 if (drv->uninit_command)
590 drv->uninit_command(cmd);
591 }
592}
593
594void scsi_free_sgtables(struct scsi_cmnd *cmd)
595{
596 if (cmd->sdb.table.nents)
597 sg_free_table_chained(&cmd->sdb.table,
598 SCSI_INLINE_SG_CNT);
599 if (scsi_prot_sg_count(cmd))
600 sg_free_table_chained(&cmd->prot_sdb->table,
601 SCSI_INLINE_PROT_SG_CNT);
602}
603EXPORT_SYMBOL_GPL(scsi_free_sgtables);
604
605static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
606{
607 scsi_free_sgtables(cmd);
608 scsi_uninit_cmd(cmd);
609}
610
611static void scsi_run_queue_async(struct scsi_device *sdev)
612{
613 if (scsi_host_in_recovery(sdev->host))
614 return;
615
616 if (scsi_target(sdev)->single_lun ||
617 !list_empty(&sdev->host->starved_list)) {
618 kblockd_schedule_work(&sdev->requeue_work);
619 } else {
620 /*
621 * smp_mb() present in sbitmap_queue_clear() or implied in
622 * .end_io is for ordering writing .device_busy in
623 * scsi_device_unbusy() and reading sdev->restarts.
624 */
625 int old = atomic_read(&sdev->restarts);
626
627 /*
628 * ->restarts has to be kept as non-zero if new budget
629 * contention occurs.
630 *
631 * No need to run queue when either another re-run
632 * queue wins in updating ->restarts or a new budget
633 * contention occurs.
634 */
635 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
636 blk_mq_run_hw_queues(sdev->request_queue, true);
637 }
638}
639
640/* Returns false when no more bytes to process, true if there are more */
641static bool scsi_end_request(struct request *req, blk_status_t error,
642 unsigned int bytes)
643{
644 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
645 struct scsi_device *sdev = cmd->device;
646 struct request_queue *q = sdev->request_queue;
647
648 if (blk_update_request(req, error, bytes))
649 return true;
650
651 if (q->limits.features & BLK_FEAT_ADD_RANDOM)
652 add_disk_randomness(req->q->disk);
653
654 WARN_ON_ONCE(!blk_rq_is_passthrough(req) &&
655 !(cmd->flags & SCMD_INITIALIZED));
656 cmd->flags = 0;
657
658 /*
659 * Calling rcu_barrier() is not necessary here because the
660 * SCSI error handler guarantees that the function called by
661 * call_rcu() has been called before scsi_end_request() is
662 * called.
663 */
664 destroy_rcu_head(&cmd->rcu);
665
666 /*
667 * In the MQ case the command gets freed by __blk_mq_end_request,
668 * so we have to do all cleanup that depends on it earlier.
669 *
670 * We also can't kick the queues from irq context, so we
671 * will have to defer it to a workqueue.
672 */
673 scsi_mq_uninit_cmd(cmd);
674
675 /*
676 * queue is still alive, so grab the ref for preventing it
677 * from being cleaned up during running queue.
678 */
679 percpu_ref_get(&q->q_usage_counter);
680
681 __blk_mq_end_request(req, error);
682
683 scsi_run_queue_async(sdev);
684
685 percpu_ref_put(&q->q_usage_counter);
686 return false;
687}
688
689/**
690 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
691 * @result: scsi error code
692 *
693 * Translate a SCSI result code into a blk_status_t value.
694 */
695static blk_status_t scsi_result_to_blk_status(int result)
696{
697 /*
698 * Check the scsi-ml byte first in case we converted a host or status
699 * byte.
700 */
701 switch (scsi_ml_byte(result)) {
702 case SCSIML_STAT_OK:
703 break;
704 case SCSIML_STAT_RESV_CONFLICT:
705 return BLK_STS_RESV_CONFLICT;
706 case SCSIML_STAT_NOSPC:
707 return BLK_STS_NOSPC;
708 case SCSIML_STAT_MED_ERROR:
709 return BLK_STS_MEDIUM;
710 case SCSIML_STAT_TGT_FAILURE:
711 return BLK_STS_TARGET;
712 case SCSIML_STAT_DL_TIMEOUT:
713 return BLK_STS_DURATION_LIMIT;
714 }
715
716 switch (host_byte(result)) {
717 case DID_OK:
718 if (scsi_status_is_good(result))
719 return BLK_STS_OK;
720 return BLK_STS_IOERR;
721 case DID_TRANSPORT_FAILFAST:
722 case DID_TRANSPORT_MARGINAL:
723 return BLK_STS_TRANSPORT;
724 default:
725 return BLK_STS_IOERR;
726 }
727}
728
729/**
730 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
731 * @rq: request to examine
732 *
733 * Description:
734 * A request could be merge of IOs which require different failure
735 * handling. This function determines the number of bytes which
736 * can be failed from the beginning of the request without
737 * crossing into area which need to be retried further.
738 *
739 * Return:
740 * The number of bytes to fail.
741 */
742static unsigned int scsi_rq_err_bytes(const struct request *rq)
743{
744 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
745 unsigned int bytes = 0;
746 struct bio *bio;
747
748 if (!(rq->rq_flags & RQF_MIXED_MERGE))
749 return blk_rq_bytes(rq);
750
751 /*
752 * Currently the only 'mixing' which can happen is between
753 * different fastfail types. We can safely fail portions
754 * which have all the failfast bits that the first one has -
755 * the ones which are at least as eager to fail as the first
756 * one.
757 */
758 for (bio = rq->bio; bio; bio = bio->bi_next) {
759 if ((bio->bi_opf & ff) != ff)
760 break;
761 bytes += bio->bi_iter.bi_size;
762 }
763
764 /* this could lead to infinite loop */
765 BUG_ON(blk_rq_bytes(rq) && !bytes);
766 return bytes;
767}
768
769static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
770{
771 struct request *req = scsi_cmd_to_rq(cmd);
772 unsigned long wait_for;
773
774 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
775 return false;
776
777 wait_for = (cmd->allowed + 1) * req->timeout;
778 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
779 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
780 wait_for/HZ);
781 return true;
782 }
783 return false;
784}
785
786/*
787 * When ALUA transition state is returned, reprep the cmd to
788 * use the ALUA handler's transition timeout. Delay the reprep
789 * 1 sec to avoid aggressive retries of the target in that
790 * state.
791 */
792#define ALUA_TRANSITION_REPREP_DELAY 1000
793
794/* Helper for scsi_io_completion() when special action required. */
795static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
796{
797 struct request *req = scsi_cmd_to_rq(cmd);
798 int level = 0;
799 enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
800 ACTION_RETRY, ACTION_DELAYED_RETRY} action;
801 struct scsi_sense_hdr sshdr;
802 bool sense_valid;
803 bool sense_current = true; /* false implies "deferred sense" */
804 blk_status_t blk_stat;
805
806 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
807 if (sense_valid)
808 sense_current = !scsi_sense_is_deferred(&sshdr);
809
810 blk_stat = scsi_result_to_blk_status(result);
811
812 if (host_byte(result) == DID_RESET) {
813 /* Third party bus reset or reset for error recovery
814 * reasons. Just retry the command and see what
815 * happens.
816 */
817 action = ACTION_RETRY;
818 } else if (sense_valid && sense_current) {
819 switch (sshdr.sense_key) {
820 case UNIT_ATTENTION:
821 if (cmd->device->removable) {
822 /* Detected disc change. Set a bit
823 * and quietly refuse further access.
824 */
825 cmd->device->changed = 1;
826 action = ACTION_FAIL;
827 } else {
828 /* Must have been a power glitch, or a
829 * bus reset. Could not have been a
830 * media change, so we just retry the
831 * command and see what happens.
832 */
833 action = ACTION_RETRY;
834 }
835 break;
836 case ILLEGAL_REQUEST:
837 /* If we had an ILLEGAL REQUEST returned, then
838 * we may have performed an unsupported
839 * command. The only thing this should be
840 * would be a ten byte read where only a six
841 * byte read was supported. Also, on a system
842 * where READ CAPACITY failed, we may have
843 * read past the end of the disk.
844 */
845 if ((cmd->device->use_10_for_rw &&
846 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
847 (cmd->cmnd[0] == READ_10 ||
848 cmd->cmnd[0] == WRITE_10)) {
849 /* This will issue a new 6-byte command. */
850 cmd->device->use_10_for_rw = 0;
851 action = ACTION_REPREP;
852 } else if (sshdr.asc == 0x10) /* DIX */ {
853 action = ACTION_FAIL;
854 blk_stat = BLK_STS_PROTECTION;
855 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
856 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
857 action = ACTION_FAIL;
858 blk_stat = BLK_STS_TARGET;
859 } else
860 action = ACTION_FAIL;
861 break;
862 case ABORTED_COMMAND:
863 action = ACTION_FAIL;
864 if (sshdr.asc == 0x10) /* DIF */
865 blk_stat = BLK_STS_PROTECTION;
866 break;
867 case NOT_READY:
868 /* If the device is in the process of becoming
869 * ready, or has a temporary blockage, retry.
870 */
871 if (sshdr.asc == 0x04) {
872 switch (sshdr.ascq) {
873 case 0x01: /* becoming ready */
874 case 0x04: /* format in progress */
875 case 0x05: /* rebuild in progress */
876 case 0x06: /* recalculation in progress */
877 case 0x07: /* operation in progress */
878 case 0x08: /* Long write in progress */
879 case 0x09: /* self test in progress */
880 case 0x11: /* notify (enable spinup) required */
881 case 0x14: /* space allocation in progress */
882 case 0x1a: /* start stop unit in progress */
883 case 0x1b: /* sanitize in progress */
884 case 0x1d: /* configuration in progress */
885 action = ACTION_DELAYED_RETRY;
886 break;
887 case 0x0a: /* ALUA state transition */
888 action = ACTION_DELAYED_REPREP;
889 break;
890 /*
891 * Depopulation might take many hours,
892 * thus it is not worthwhile to retry.
893 */
894 case 0x24: /* depopulation in progress */
895 case 0x25: /* depopulation restore in progress */
896 fallthrough;
897 default:
898 action = ACTION_FAIL;
899 break;
900 }
901 } else
902 action = ACTION_FAIL;
903 break;
904 case VOLUME_OVERFLOW:
905 /* See SSC3rXX or current. */
906 action = ACTION_FAIL;
907 break;
908 case DATA_PROTECT:
909 action = ACTION_FAIL;
910 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
911 (sshdr.asc == 0x55 &&
912 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
913 /* Insufficient zone resources */
914 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
915 }
916 break;
917 case COMPLETED:
918 fallthrough;
919 default:
920 action = ACTION_FAIL;
921 break;
922 }
923 } else
924 action = ACTION_FAIL;
925
926 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
927 action = ACTION_FAIL;
928
929 switch (action) {
930 case ACTION_FAIL:
931 /* Give up and fail the remainder of the request */
932 if (!(req->rq_flags & RQF_QUIET)) {
933 static DEFINE_RATELIMIT_STATE(_rs,
934 DEFAULT_RATELIMIT_INTERVAL,
935 DEFAULT_RATELIMIT_BURST);
936
937 if (unlikely(scsi_logging_level))
938 level =
939 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
940 SCSI_LOG_MLCOMPLETE_BITS);
941
942 /*
943 * if logging is enabled the failure will be printed
944 * in scsi_log_completion(), so avoid duplicate messages
945 */
946 if (!level && __ratelimit(&_rs)) {
947 scsi_print_result(cmd, NULL, FAILED);
948 if (sense_valid)
949 scsi_print_sense(cmd);
950 scsi_print_command(cmd);
951 }
952 }
953 if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
954 return;
955 fallthrough;
956 case ACTION_REPREP:
957 scsi_mq_requeue_cmd(cmd, 0);
958 break;
959 case ACTION_DELAYED_REPREP:
960 scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
961 break;
962 case ACTION_RETRY:
963 /* Retry the same command immediately */
964 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
965 break;
966 case ACTION_DELAYED_RETRY:
967 /* Retry the same command after a delay */
968 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
969 break;
970 }
971}
972
973/*
974 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
975 * new result that may suppress further error checking. Also modifies
976 * *blk_statp in some cases.
977 */
978static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
979 blk_status_t *blk_statp)
980{
981 bool sense_valid;
982 bool sense_current = true; /* false implies "deferred sense" */
983 struct request *req = scsi_cmd_to_rq(cmd);
984 struct scsi_sense_hdr sshdr;
985
986 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
987 if (sense_valid)
988 sense_current = !scsi_sense_is_deferred(&sshdr);
989
990 if (blk_rq_is_passthrough(req)) {
991 if (sense_valid) {
992 /*
993 * SG_IO wants current and deferred errors
994 */
995 cmd->sense_len = min(8 + cmd->sense_buffer[7],
996 SCSI_SENSE_BUFFERSIZE);
997 }
998 if (sense_current)
999 *blk_statp = scsi_result_to_blk_status(result);
1000 } else if (blk_rq_bytes(req) == 0 && sense_current) {
1001 /*
1002 * Flush commands do not transfers any data, and thus cannot use
1003 * good_bytes != blk_rq_bytes(req) as the signal for an error.
1004 * This sets *blk_statp explicitly for the problem case.
1005 */
1006 *blk_statp = scsi_result_to_blk_status(result);
1007 }
1008 /*
1009 * Recovered errors need reporting, but they're always treated as
1010 * success, so fiddle the result code here. For passthrough requests
1011 * we already took a copy of the original into sreq->result which
1012 * is what gets returned to the user
1013 */
1014 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
1015 bool do_print = true;
1016 /*
1017 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
1018 * skip print since caller wants ATA registers. Only occurs
1019 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
1020 */
1021 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
1022 do_print = false;
1023 else if (req->rq_flags & RQF_QUIET)
1024 do_print = false;
1025 if (do_print)
1026 scsi_print_sense(cmd);
1027 result = 0;
1028 /* for passthrough, *blk_statp may be set */
1029 *blk_statp = BLK_STS_OK;
1030 }
1031 /*
1032 * Another corner case: the SCSI status byte is non-zero but 'good'.
1033 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
1034 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
1035 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
1036 * intermediate statuses (both obsolete in SAM-4) as good.
1037 */
1038 if ((result & 0xff) && scsi_status_is_good(result)) {
1039 result = 0;
1040 *blk_statp = BLK_STS_OK;
1041 }
1042 return result;
1043}
1044
1045/**
1046 * scsi_io_completion - Completion processing for SCSI commands.
1047 * @cmd: command that is finished.
1048 * @good_bytes: number of processed bytes.
1049 *
1050 * We will finish off the specified number of sectors. If we are done, the
1051 * command block will be released and the queue function will be goosed. If we
1052 * are not done then we have to figure out what to do next:
1053 *
1054 * a) We can call scsi_mq_requeue_cmd(). The request will be
1055 * unprepared and put back on the queue. Then a new command will
1056 * be created for it. This should be used if we made forward
1057 * progress, or if we want to switch from READ(10) to READ(6) for
1058 * example.
1059 *
1060 * b) We can call scsi_io_completion_action(). The request will be
1061 * put back on the queue and retried using the same command as
1062 * before, possibly after a delay.
1063 *
1064 * c) We can call scsi_end_request() with blk_stat other than
1065 * BLK_STS_OK, to fail the remainder of the request.
1066 */
1067void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
1068{
1069 int result = cmd->result;
1070 struct request *req = scsi_cmd_to_rq(cmd);
1071 blk_status_t blk_stat = BLK_STS_OK;
1072
1073 if (unlikely(result)) /* a nz result may or may not be an error */
1074 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
1075
1076 /*
1077 * Next deal with any sectors which we were able to correctly
1078 * handle.
1079 */
1080 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
1081 "%u sectors total, %d bytes done.\n",
1082 blk_rq_sectors(req), good_bytes));
1083
1084 /*
1085 * Failed, zero length commands always need to drop down
1086 * to retry code. Fast path should return in this block.
1087 */
1088 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
1089 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
1090 return; /* no bytes remaining */
1091 }
1092
1093 /* Kill remainder if no retries. */
1094 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
1095 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
1096 WARN_ONCE(true,
1097 "Bytes remaining after failed, no-retry command");
1098 return;
1099 }
1100
1101 /*
1102 * If there had been no error, but we have leftover bytes in the
1103 * request just queue the command up again.
1104 */
1105 if (likely(result == 0))
1106 scsi_mq_requeue_cmd(cmd, 0);
1107 else
1108 scsi_io_completion_action(cmd, result);
1109}
1110
1111static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1112 struct request *rq)
1113{
1114 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1115 !op_is_write(req_op(rq)) &&
1116 sdev->host->hostt->dma_need_drain(rq);
1117}
1118
1119/**
1120 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1121 * @cmd: SCSI command data structure to initialize.
1122 *
1123 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1124 * for @cmd.
1125 *
1126 * Returns:
1127 * * BLK_STS_OK - on success
1128 * * BLK_STS_RESOURCE - if the failure is retryable
1129 * * BLK_STS_IOERR - if the failure is fatal
1130 */
1131blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1132{
1133 struct scsi_device *sdev = cmd->device;
1134 struct request *rq = scsi_cmd_to_rq(cmd);
1135 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1136 struct scatterlist *last_sg = NULL;
1137 blk_status_t ret;
1138 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1139 int count;
1140
1141 if (WARN_ON_ONCE(!nr_segs))
1142 return BLK_STS_IOERR;
1143
1144 /*
1145 * Make sure there is space for the drain. The driver must adjust
1146 * max_hw_segments to be prepared for this.
1147 */
1148 if (need_drain)
1149 nr_segs++;
1150
1151 /*
1152 * If sg table allocation fails, requeue request later.
1153 */
1154 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1155 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1156 return BLK_STS_RESOURCE;
1157
1158 /*
1159 * Next, walk the list, and fill in the addresses and sizes of
1160 * each segment.
1161 */
1162 count = __blk_rq_map_sg(rq, cmd->sdb.table.sgl, &last_sg);
1163
1164 if (blk_rq_bytes(rq) & rq->q->limits.dma_pad_mask) {
1165 unsigned int pad_len =
1166 (rq->q->limits.dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1167
1168 last_sg->length += pad_len;
1169 cmd->extra_len += pad_len;
1170 }
1171
1172 if (need_drain) {
1173 sg_unmark_end(last_sg);
1174 last_sg = sg_next(last_sg);
1175 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1176 sg_mark_end(last_sg);
1177
1178 cmd->extra_len += sdev->dma_drain_len;
1179 count++;
1180 }
1181
1182 BUG_ON(count > cmd->sdb.table.nents);
1183 cmd->sdb.table.nents = count;
1184 cmd->sdb.length = blk_rq_payload_bytes(rq);
1185
1186 if (blk_integrity_rq(rq)) {
1187 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1188
1189 if (WARN_ON_ONCE(!prot_sdb)) {
1190 /*
1191 * This can happen if someone (e.g. multipath)
1192 * queues a command to a device on an adapter
1193 * that does not support DIX.
1194 */
1195 ret = BLK_STS_IOERR;
1196 goto out_free_sgtables;
1197 }
1198
1199 if (sg_alloc_table_chained(&prot_sdb->table,
1200 rq->nr_integrity_segments,
1201 prot_sdb->table.sgl,
1202 SCSI_INLINE_PROT_SG_CNT)) {
1203 ret = BLK_STS_RESOURCE;
1204 goto out_free_sgtables;
1205 }
1206
1207 count = blk_rq_map_integrity_sg(rq, prot_sdb->table.sgl);
1208 cmd->prot_sdb = prot_sdb;
1209 cmd->prot_sdb->table.nents = count;
1210 }
1211
1212 return BLK_STS_OK;
1213out_free_sgtables:
1214 scsi_free_sgtables(cmd);
1215 return ret;
1216}
1217EXPORT_SYMBOL(scsi_alloc_sgtables);
1218
1219/**
1220 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1221 * @rq: Request associated with the SCSI command to be initialized.
1222 *
1223 * This function initializes the members of struct scsi_cmnd that must be
1224 * initialized before request processing starts and that won't be
1225 * reinitialized if a SCSI command is requeued.
1226 */
1227static void scsi_initialize_rq(struct request *rq)
1228{
1229 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1230
1231 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1232 cmd->cmd_len = MAX_COMMAND_SIZE;
1233 cmd->sense_len = 0;
1234 init_rcu_head(&cmd->rcu);
1235 cmd->jiffies_at_alloc = jiffies;
1236 cmd->retries = 0;
1237}
1238
1239/**
1240 * scsi_alloc_request - allocate a block request and partially
1241 * initialize its &scsi_cmnd
1242 * @q: the device's request queue
1243 * @opf: the request operation code
1244 * @flags: block layer allocation flags
1245 *
1246 * Return: &struct request pointer on success or %NULL on failure
1247 */
1248struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1249 blk_mq_req_flags_t flags)
1250{
1251 struct request *rq;
1252
1253 rq = blk_mq_alloc_request(q, opf, flags);
1254 if (!IS_ERR(rq))
1255 scsi_initialize_rq(rq);
1256 return rq;
1257}
1258EXPORT_SYMBOL_GPL(scsi_alloc_request);
1259
1260/*
1261 * Only called when the request isn't completed by SCSI, and not freed by
1262 * SCSI
1263 */
1264static void scsi_cleanup_rq(struct request *rq)
1265{
1266 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1267
1268 cmd->flags = 0;
1269
1270 if (rq->rq_flags & RQF_DONTPREP) {
1271 scsi_mq_uninit_cmd(cmd);
1272 rq->rq_flags &= ~RQF_DONTPREP;
1273 }
1274}
1275
1276/* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1277void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1278{
1279 struct request *rq = scsi_cmd_to_rq(cmd);
1280
1281 if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1282 cmd->flags |= SCMD_INITIALIZED;
1283 scsi_initialize_rq(rq);
1284 }
1285
1286 cmd->device = dev;
1287 INIT_LIST_HEAD(&cmd->eh_entry);
1288 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1289}
1290
1291static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1292 struct request *req)
1293{
1294 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1295
1296 /*
1297 * Passthrough requests may transfer data, in which case they must
1298 * a bio attached to them. Or they might contain a SCSI command
1299 * that does not transfer data, in which case they may optionally
1300 * submit a request without an attached bio.
1301 */
1302 if (req->bio) {
1303 blk_status_t ret = scsi_alloc_sgtables(cmd);
1304 if (unlikely(ret != BLK_STS_OK))
1305 return ret;
1306 } else {
1307 BUG_ON(blk_rq_bytes(req));
1308
1309 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1310 }
1311
1312 cmd->transfersize = blk_rq_bytes(req);
1313 return BLK_STS_OK;
1314}
1315
1316static blk_status_t
1317scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1318{
1319 switch (sdev->sdev_state) {
1320 case SDEV_CREATED:
1321 return BLK_STS_OK;
1322 case SDEV_OFFLINE:
1323 case SDEV_TRANSPORT_OFFLINE:
1324 /*
1325 * If the device is offline we refuse to process any
1326 * commands. The device must be brought online
1327 * before trying any recovery commands.
1328 */
1329 if (!sdev->offline_already) {
1330 sdev->offline_already = true;
1331 sdev_printk(KERN_ERR, sdev,
1332 "rejecting I/O to offline device\n");
1333 }
1334 return BLK_STS_IOERR;
1335 case SDEV_DEL:
1336 /*
1337 * If the device is fully deleted, we refuse to
1338 * process any commands as well.
1339 */
1340 sdev_printk(KERN_ERR, sdev,
1341 "rejecting I/O to dead device\n");
1342 return BLK_STS_IOERR;
1343 case SDEV_BLOCK:
1344 case SDEV_CREATED_BLOCK:
1345 return BLK_STS_RESOURCE;
1346 case SDEV_QUIESCE:
1347 /*
1348 * If the device is blocked we only accept power management
1349 * commands.
1350 */
1351 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1352 return BLK_STS_RESOURCE;
1353 return BLK_STS_OK;
1354 default:
1355 /*
1356 * For any other not fully online state we only allow
1357 * power management commands.
1358 */
1359 if (req && !(req->rq_flags & RQF_PM))
1360 return BLK_STS_OFFLINE;
1361 return BLK_STS_OK;
1362 }
1363}
1364
1365/*
1366 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1367 * and return the token else return -1.
1368 */
1369static inline int scsi_dev_queue_ready(struct request_queue *q,
1370 struct scsi_device *sdev)
1371{
1372 int token;
1373
1374 if (!sdev->budget_map.map)
1375 return INT_MAX;
1376
1377 token = sbitmap_get(&sdev->budget_map);
1378 if (token < 0)
1379 return -1;
1380
1381 if (!atomic_read(&sdev->device_blocked))
1382 return token;
1383
1384 /*
1385 * Only unblock if no other commands are pending and
1386 * if device_blocked has decreased to zero
1387 */
1388 if (scsi_device_busy(sdev) > 1 ||
1389 atomic_dec_return(&sdev->device_blocked) > 0) {
1390 sbitmap_put(&sdev->budget_map, token);
1391 return -1;
1392 }
1393
1394 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1395 "unblocking device at zero depth\n"));
1396
1397 return token;
1398}
1399
1400/*
1401 * scsi_target_queue_ready: checks if there we can send commands to target
1402 * @sdev: scsi device on starget to check.
1403 */
1404static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1405 struct scsi_device *sdev)
1406{
1407 struct scsi_target *starget = scsi_target(sdev);
1408 unsigned int busy;
1409
1410 if (starget->single_lun) {
1411 spin_lock_irq(shost->host_lock);
1412 if (starget->starget_sdev_user &&
1413 starget->starget_sdev_user != sdev) {
1414 spin_unlock_irq(shost->host_lock);
1415 return 0;
1416 }
1417 starget->starget_sdev_user = sdev;
1418 spin_unlock_irq(shost->host_lock);
1419 }
1420
1421 if (starget->can_queue <= 0)
1422 return 1;
1423
1424 busy = atomic_inc_return(&starget->target_busy) - 1;
1425 if (atomic_read(&starget->target_blocked) > 0) {
1426 if (busy)
1427 goto starved;
1428
1429 /*
1430 * unblock after target_blocked iterates to zero
1431 */
1432 if (atomic_dec_return(&starget->target_blocked) > 0)
1433 goto out_dec;
1434
1435 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1436 "unblocking target at zero depth\n"));
1437 }
1438
1439 if (busy >= starget->can_queue)
1440 goto starved;
1441
1442 return 1;
1443
1444starved:
1445 spin_lock_irq(shost->host_lock);
1446 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1447 spin_unlock_irq(shost->host_lock);
1448out_dec:
1449 if (starget->can_queue > 0)
1450 atomic_dec(&starget->target_busy);
1451 return 0;
1452}
1453
1454/*
1455 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1456 * return 0. We must end up running the queue again whenever 0 is
1457 * returned, else IO can hang.
1458 */
1459static inline int scsi_host_queue_ready(struct request_queue *q,
1460 struct Scsi_Host *shost,
1461 struct scsi_device *sdev,
1462 struct scsi_cmnd *cmd)
1463{
1464 if (atomic_read(&shost->host_blocked) > 0) {
1465 if (scsi_host_busy(shost) > 0)
1466 goto starved;
1467
1468 /*
1469 * unblock after host_blocked iterates to zero
1470 */
1471 if (atomic_dec_return(&shost->host_blocked) > 0)
1472 goto out_dec;
1473
1474 SCSI_LOG_MLQUEUE(3,
1475 shost_printk(KERN_INFO, shost,
1476 "unblocking host at zero depth\n"));
1477 }
1478
1479 if (shost->host_self_blocked)
1480 goto starved;
1481
1482 /* We're OK to process the command, so we can't be starved */
1483 if (!list_empty(&sdev->starved_entry)) {
1484 spin_lock_irq(shost->host_lock);
1485 if (!list_empty(&sdev->starved_entry))
1486 list_del_init(&sdev->starved_entry);
1487 spin_unlock_irq(shost->host_lock);
1488 }
1489
1490 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1491
1492 return 1;
1493
1494starved:
1495 spin_lock_irq(shost->host_lock);
1496 if (list_empty(&sdev->starved_entry))
1497 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1498 spin_unlock_irq(shost->host_lock);
1499out_dec:
1500 scsi_dec_host_busy(shost, cmd);
1501 return 0;
1502}
1503
1504/*
1505 * Busy state exporting function for request stacking drivers.
1506 *
1507 * For efficiency, no lock is taken to check the busy state of
1508 * shost/starget/sdev, since the returned value is not guaranteed and
1509 * may be changed after request stacking drivers call the function,
1510 * regardless of taking lock or not.
1511 *
1512 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1513 * needs to return 'not busy'. Otherwise, request stacking drivers
1514 * may hold requests forever.
1515 */
1516static bool scsi_mq_lld_busy(struct request_queue *q)
1517{
1518 struct scsi_device *sdev = q->queuedata;
1519 struct Scsi_Host *shost;
1520
1521 if (blk_queue_dying(q))
1522 return false;
1523
1524 shost = sdev->host;
1525
1526 /*
1527 * Ignore host/starget busy state.
1528 * Since block layer does not have a concept of fairness across
1529 * multiple queues, congestion of host/starget needs to be handled
1530 * in SCSI layer.
1531 */
1532 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1533 return true;
1534
1535 return false;
1536}
1537
1538/*
1539 * Block layer request completion callback. May be called from interrupt
1540 * context.
1541 */
1542static void scsi_complete(struct request *rq)
1543{
1544 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1545 enum scsi_disposition disposition;
1546
1547 if (blk_mq_is_reserved_rq(rq)) {
1548 /* Only pass-through requests are supported in this code path. */
1549 WARN_ON_ONCE(!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd)));
1550 scsi_mq_uninit_cmd(cmd);
1551 __blk_mq_end_request(rq, scsi_result_to_blk_status(cmd->result));
1552 return;
1553 }
1554
1555 INIT_LIST_HEAD(&cmd->eh_entry);
1556
1557 atomic_inc(&cmd->device->iodone_cnt);
1558 if (cmd->result)
1559 atomic_inc(&cmd->device->ioerr_cnt);
1560
1561 disposition = scsi_decide_disposition(cmd);
1562 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1563 disposition = SUCCESS;
1564
1565 scsi_log_completion(cmd, disposition);
1566
1567 switch (disposition) {
1568 case SUCCESS:
1569 scsi_finish_command(cmd);
1570 break;
1571 case NEEDS_RETRY:
1572 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1573 break;
1574 case ADD_TO_MLQUEUE:
1575 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1576 break;
1577 default:
1578 scsi_eh_scmd_add(cmd);
1579 break;
1580 }
1581}
1582
1583/**
1584 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1585 * @cmd: command block we are dispatching.
1586 *
1587 * Return: nonzero return request was rejected and device's queue needs to be
1588 * plugged.
1589 */
1590static enum scsi_qc_status scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1591{
1592 struct Scsi_Host *host = cmd->device->host;
1593 int rtn = 0;
1594
1595 atomic_inc(&cmd->device->iorequest_cnt);
1596
1597 /* check if the device is still usable */
1598 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1599 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1600 * returns an immediate error upwards, and signals
1601 * that the device is no longer present */
1602 cmd->result = DID_NO_CONNECT << 16;
1603 goto done;
1604 }
1605
1606 /* Check to see if the scsi lld made this device blocked. */
1607 if (unlikely(scsi_device_blocked(cmd->device))) {
1608 /*
1609 * in blocked state, the command is just put back on
1610 * the device queue. The suspend state has already
1611 * blocked the queue so future requests should not
1612 * occur until the device transitions out of the
1613 * suspend state.
1614 */
1615 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1616 "queuecommand : device blocked\n"));
1617 atomic_dec(&cmd->device->iorequest_cnt);
1618 return SCSI_MLQUEUE_DEVICE_BUSY;
1619 }
1620
1621 /* Store the LUN value in cmnd, if needed. */
1622 if (cmd->device->lun_in_cdb)
1623 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1624 (cmd->device->lun << 5 & 0xe0);
1625
1626 scsi_log_send(cmd);
1627
1628 /*
1629 * Before we queue this command, check if the command
1630 * length exceeds what the host adapter can handle.
1631 */
1632 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1633 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1634 "queuecommand : command too long. "
1635 "cdb_size=%d host->max_cmd_len=%d\n",
1636 cmd->cmd_len, cmd->device->host->max_cmd_len));
1637 cmd->result = (DID_ABORT << 16);
1638 goto done;
1639 }
1640
1641 if (unlikely(host->shost_state == SHOST_DEL)) {
1642 cmd->result = (DID_NO_CONNECT << 16);
1643 goto done;
1644
1645 }
1646
1647 trace_scsi_dispatch_cmd_start(cmd);
1648 rtn = host->hostt->queuecommand(host, cmd);
1649 if (rtn) {
1650 atomic_dec(&cmd->device->iorequest_cnt);
1651 trace_scsi_dispatch_cmd_error(cmd, rtn);
1652 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1653 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1654 rtn = SCSI_MLQUEUE_HOST_BUSY;
1655
1656 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1657 "queuecommand : request rejected\n"));
1658 }
1659
1660 return rtn;
1661 done:
1662 scsi_done(cmd);
1663 return 0;
1664}
1665
1666/* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1667static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1668{
1669 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1670 sizeof(struct scatterlist);
1671}
1672
1673static blk_status_t scsi_prepare_cmd(struct request *req)
1674{
1675 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1676 struct scsi_device *sdev = req->q->queuedata;
1677 struct Scsi_Host *shost = sdev->host;
1678 bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1679 struct scatterlist *sg;
1680
1681 scsi_init_command(sdev, cmd);
1682
1683 cmd->eh_eflags = 0;
1684 cmd->prot_type = 0;
1685 cmd->prot_flags = 0;
1686 cmd->submitter = 0;
1687 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1688 cmd->underflow = 0;
1689 cmd->transfersize = 0;
1690 cmd->host_scribble = NULL;
1691 cmd->result = 0;
1692 cmd->extra_len = 0;
1693 cmd->state = 0;
1694 if (in_flight)
1695 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1696
1697 cmd->prot_op = SCSI_PROT_NORMAL;
1698 if (blk_rq_bytes(req))
1699 cmd->sc_data_direction = rq_dma_dir(req);
1700 else
1701 cmd->sc_data_direction = DMA_NONE;
1702
1703 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1704 cmd->sdb.table.sgl = sg;
1705
1706 if (scsi_host_get_prot(shost)) {
1707 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1708
1709 cmd->prot_sdb->table.sgl =
1710 (struct scatterlist *)(cmd->prot_sdb + 1);
1711 }
1712
1713 /*
1714 * Special handling for passthrough commands, which don't go to the ULP
1715 * at all:
1716 */
1717 if (blk_rq_is_passthrough(req))
1718 return scsi_setup_scsi_cmnd(sdev, req);
1719
1720 if (sdev->handler && sdev->handler->prep_fn) {
1721 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1722
1723 if (ret != BLK_STS_OK)
1724 return ret;
1725 }
1726
1727 /* Usually overridden by the ULP */
1728 cmd->allowed = 0;
1729 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1730 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1731}
1732
1733static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1734{
1735 struct request *req = scsi_cmd_to_rq(cmd);
1736
1737 switch (cmd->submitter) {
1738 case SUBMITTED_BY_BLOCK_LAYER:
1739 break;
1740 case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1741 return scsi_eh_done(cmd);
1742 case SUBMITTED_BY_SCSI_RESET_IOCTL:
1743 return;
1744 }
1745
1746 if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1747 return;
1748 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1749 return;
1750 trace_scsi_dispatch_cmd_done(cmd);
1751
1752 if (complete_directly)
1753 blk_mq_complete_request_direct(req, scsi_complete);
1754 else
1755 blk_mq_complete_request(req);
1756}
1757
1758void scsi_done(struct scsi_cmnd *cmd)
1759{
1760 scsi_done_internal(cmd, false);
1761}
1762EXPORT_SYMBOL(scsi_done);
1763
1764void scsi_done_direct(struct scsi_cmnd *cmd)
1765{
1766 scsi_done_internal(cmd, true);
1767}
1768EXPORT_SYMBOL(scsi_done_direct);
1769
1770static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1771{
1772 struct scsi_device *sdev = q->queuedata;
1773
1774 if (sdev->budget_map.map)
1775 sbitmap_put(&sdev->budget_map, budget_token);
1776}
1777
1778/*
1779 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1780 * not change behaviour from the previous unplug mechanism, experimentation
1781 * may prove this needs changing.
1782 */
1783#define SCSI_QUEUE_DELAY 3
1784
1785static int scsi_mq_get_budget(struct request_queue *q)
1786{
1787 struct scsi_device *sdev = q->queuedata;
1788 int token = scsi_dev_queue_ready(q, sdev);
1789
1790 if (token >= 0)
1791 return token;
1792
1793 atomic_inc(&sdev->restarts);
1794
1795 /*
1796 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1797 * .restarts must be incremented before .device_busy is read because the
1798 * code in scsi_run_queue_async() depends on the order of these operations.
1799 */
1800 smp_mb__after_atomic();
1801
1802 /*
1803 * If all in-flight requests originated from this LUN are completed
1804 * before reading .device_busy, sdev->device_busy will be observed as
1805 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1806 * soon. Otherwise, completion of one of these requests will observe
1807 * the .restarts flag, and the request queue will be run for handling
1808 * this request, see scsi_end_request().
1809 */
1810 if (unlikely(scsi_device_busy(sdev) == 0 &&
1811 !scsi_device_blocked(sdev)))
1812 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1813 return -1;
1814}
1815
1816static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1817{
1818 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1819
1820 cmd->budget_token = token;
1821}
1822
1823static int scsi_mq_get_rq_budget_token(struct request *req)
1824{
1825 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1826
1827 return cmd->budget_token;
1828}
1829
1830static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1831 const struct blk_mq_queue_data *bd)
1832{
1833 struct request *req = bd->rq;
1834 struct request_queue *q = req->q;
1835 struct scsi_device *sdev = q->queuedata;
1836 struct Scsi_Host *shost = sdev->host;
1837 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1838 blk_status_t ret;
1839 enum scsi_qc_status reason;
1840
1841 WARN_ON_ONCE(cmd->budget_token < 0);
1842
1843 /*
1844 * Bypass the SCSI device, SCSI target and SCSI host checks for
1845 * reserved commands.
1846 */
1847 if (!blk_mq_is_reserved_rq(req)) {
1848 /*
1849 * If the device is not in running state we will reject some or
1850 * all commands.
1851 */
1852 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1853 ret = scsi_device_state_check(sdev, req);
1854 if (ret != BLK_STS_OK)
1855 goto out_put_budget;
1856 }
1857
1858 ret = BLK_STS_RESOURCE;
1859 if (!scsi_target_queue_ready(shost, sdev))
1860 goto out_put_budget;
1861 if (unlikely(scsi_host_in_recovery(shost))) {
1862 if (cmd->flags & SCMD_FAIL_IF_RECOVERING)
1863 ret = BLK_STS_OFFLINE;
1864 goto out_dec_target_busy;
1865 }
1866 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1867 goto out_dec_target_busy;
1868 }
1869
1870 /*
1871 * Only clear the driver-private command data if the LLD does not supply
1872 * a function to initialize that data.
1873 */
1874 if (shost->hostt->cmd_size && !shost->hostt->init_cmd_priv)
1875 memset(scsi_cmd_priv(cmd), 0, shost->hostt->cmd_size);
1876
1877 if (!(req->rq_flags & RQF_DONTPREP)) {
1878 ret = scsi_prepare_cmd(req);
1879 if (ret != BLK_STS_OK)
1880 goto out_dec_host_busy;
1881 req->rq_flags |= RQF_DONTPREP;
1882 } else {
1883 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1884 }
1885
1886 cmd->flags &= SCMD_PRESERVED_FLAGS;
1887 if (sdev->simple_tags)
1888 cmd->flags |= SCMD_TAGGED;
1889 if (bd->last)
1890 cmd->flags |= SCMD_LAST;
1891
1892 scsi_set_resid(cmd, 0);
1893 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1894 cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1895
1896 blk_mq_start_request(req);
1897 if (blk_mq_is_reserved_rq(req)) {
1898 reason = shost->hostt->queue_reserved_command(shost, cmd);
1899 if (reason) {
1900 ret = BLK_STS_RESOURCE;
1901 goto out_put_budget;
1902 }
1903 return BLK_STS_OK;
1904 }
1905 reason = scsi_dispatch_cmd(cmd);
1906 if (reason) {
1907 scsi_set_blocked(cmd, reason);
1908 ret = BLK_STS_RESOURCE;
1909 goto out_dec_host_busy;
1910 }
1911
1912 return BLK_STS_OK;
1913
1914out_dec_host_busy:
1915 scsi_dec_host_busy(shost, cmd);
1916out_dec_target_busy:
1917 if (scsi_target(sdev)->can_queue > 0)
1918 atomic_dec(&scsi_target(sdev)->target_busy);
1919out_put_budget:
1920 scsi_mq_put_budget(q, cmd->budget_token);
1921 cmd->budget_token = -1;
1922 switch (ret) {
1923 case BLK_STS_OK:
1924 break;
1925 case BLK_STS_RESOURCE:
1926 if (scsi_device_blocked(sdev))
1927 ret = BLK_STS_DEV_RESOURCE;
1928 break;
1929 case BLK_STS_AGAIN:
1930 cmd->result = DID_BUS_BUSY << 16;
1931 if (req->rq_flags & RQF_DONTPREP)
1932 scsi_mq_uninit_cmd(cmd);
1933 break;
1934 default:
1935 if (unlikely(!scsi_device_online(sdev)))
1936 cmd->result = DID_NO_CONNECT << 16;
1937 else
1938 cmd->result = DID_ERROR << 16;
1939 /*
1940 * Make sure to release all allocated resources when
1941 * we hit an error, as we will never see this command
1942 * again.
1943 */
1944 if (req->rq_flags & RQF_DONTPREP)
1945 scsi_mq_uninit_cmd(cmd);
1946 scsi_run_queue_async(sdev);
1947 break;
1948 }
1949 return ret;
1950}
1951
1952static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1953 unsigned int hctx_idx, unsigned int numa_node)
1954{
1955 struct Scsi_Host *shost = set->driver_data;
1956 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1957 struct scatterlist *sg;
1958 int ret = 0;
1959
1960 cmd->sense_buffer =
1961 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1962 if (!cmd->sense_buffer)
1963 return -ENOMEM;
1964
1965 if (scsi_host_get_prot(shost)) {
1966 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1967 shost->hostt->cmd_size;
1968 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1969 }
1970
1971 if (shost->hostt->init_cmd_priv) {
1972 ret = shost->hostt->init_cmd_priv(shost, cmd);
1973 if (ret < 0)
1974 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1975 }
1976
1977 return ret;
1978}
1979
1980static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1981 unsigned int hctx_idx)
1982{
1983 struct Scsi_Host *shost = set->driver_data;
1984 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1985
1986 if (shost->hostt->exit_cmd_priv)
1987 shost->hostt->exit_cmd_priv(shost, cmd);
1988 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1989}
1990
1991
1992static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1993{
1994 struct Scsi_Host *shost = hctx->driver_data;
1995
1996 if (shost->hostt->mq_poll)
1997 return shost->hostt->mq_poll(shost, hctx->queue_num);
1998
1999 return 0;
2000}
2001
2002static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
2003 unsigned int hctx_idx)
2004{
2005 struct Scsi_Host *shost = data;
2006
2007 hctx->driver_data = shost;
2008 return 0;
2009}
2010
2011static void scsi_map_queues(struct blk_mq_tag_set *set)
2012{
2013 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
2014
2015 if (shost->hostt->map_queues)
2016 return shost->hostt->map_queues(shost);
2017 blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
2018}
2019
2020void scsi_init_limits(struct Scsi_Host *shost, struct queue_limits *lim)
2021{
2022 struct device *dev = shost->dma_dev;
2023
2024 memset(lim, 0, sizeof(*lim));
2025 lim->max_segments =
2026 min_t(unsigned short, shost->sg_tablesize, SG_MAX_SEGMENTS);
2027
2028 if (scsi_host_prot_dma(shost)) {
2029 shost->sg_prot_tablesize =
2030 min_not_zero(shost->sg_prot_tablesize,
2031 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2032 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2033 lim->max_integrity_segments = shost->sg_prot_tablesize;
2034 }
2035
2036 lim->max_hw_sectors = shost->max_sectors;
2037 lim->seg_boundary_mask = shost->dma_boundary;
2038 lim->max_segment_size = shost->max_segment_size;
2039 lim->virt_boundary_mask = shost->virt_boundary_mask;
2040 lim->dma_alignment = max_t(unsigned int,
2041 shost->dma_alignment, dma_get_cache_alignment() - 1);
2042
2043 /*
2044 * Propagate the DMA formation properties to the dma-mapping layer as
2045 * a courtesy service to the LLDDs. This needs to check that the buses
2046 * actually support the DMA API first, though.
2047 */
2048 if (dev->dma_parms) {
2049 dma_set_seg_boundary(dev, shost->dma_boundary);
2050 dma_set_max_seg_size(dev, shost->max_segment_size);
2051 }
2052}
2053EXPORT_SYMBOL_GPL(scsi_init_limits);
2054
2055static const struct blk_mq_ops scsi_mq_ops_no_commit = {
2056 .get_budget = scsi_mq_get_budget,
2057 .put_budget = scsi_mq_put_budget,
2058 .queue_rq = scsi_queue_rq,
2059 .complete = scsi_complete,
2060 .timeout = scsi_timeout,
2061#ifdef CONFIG_BLK_DEBUG_FS
2062 .show_rq = scsi_show_rq,
2063#endif
2064 .init_request = scsi_mq_init_request,
2065 .exit_request = scsi_mq_exit_request,
2066 .cleanup_rq = scsi_cleanup_rq,
2067 .busy = scsi_mq_lld_busy,
2068 .map_queues = scsi_map_queues,
2069 .init_hctx = scsi_init_hctx,
2070 .poll = scsi_mq_poll,
2071 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
2072 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
2073};
2074
2075
2076static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
2077{
2078 struct Scsi_Host *shost = hctx->driver_data;
2079
2080 shost->hostt->commit_rqs(shost, hctx->queue_num);
2081}
2082
2083static const struct blk_mq_ops scsi_mq_ops = {
2084 .get_budget = scsi_mq_get_budget,
2085 .put_budget = scsi_mq_put_budget,
2086 .queue_rq = scsi_queue_rq,
2087 .commit_rqs = scsi_commit_rqs,
2088 .complete = scsi_complete,
2089 .timeout = scsi_timeout,
2090#ifdef CONFIG_BLK_DEBUG_FS
2091 .show_rq = scsi_show_rq,
2092#endif
2093 .init_request = scsi_mq_init_request,
2094 .exit_request = scsi_mq_exit_request,
2095 .cleanup_rq = scsi_cleanup_rq,
2096 .busy = scsi_mq_lld_busy,
2097 .map_queues = scsi_map_queues,
2098 .init_hctx = scsi_init_hctx,
2099 .poll = scsi_mq_poll,
2100 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
2101 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
2102};
2103
2104int scsi_mq_setup_tags(struct Scsi_Host *shost)
2105{
2106 unsigned int cmd_size, sgl_size;
2107 struct blk_mq_tag_set *tag_set = &shost->tag_set;
2108
2109 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
2110 scsi_mq_inline_sgl_size(shost));
2111 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2112 if (scsi_host_get_prot(shost))
2113 cmd_size += sizeof(struct scsi_data_buffer) +
2114 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
2115
2116 memset(tag_set, 0, sizeof(*tag_set));
2117 if (shost->hostt->commit_rqs)
2118 tag_set->ops = &scsi_mq_ops;
2119 else
2120 tag_set->ops = &scsi_mq_ops_no_commit;
2121 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
2122 tag_set->nr_maps = shost->nr_maps ? : 1;
2123 tag_set->queue_depth = shost->can_queue + shost->nr_reserved_cmds;
2124 tag_set->reserved_tags = shost->nr_reserved_cmds;
2125 tag_set->cmd_size = cmd_size;
2126 tag_set->numa_node = dev_to_node(shost->dma_dev);
2127 if (shost->hostt->tag_alloc_policy_rr)
2128 tag_set->flags |= BLK_MQ_F_TAG_RR;
2129 if (shost->queuecommand_may_block)
2130 tag_set->flags |= BLK_MQ_F_BLOCKING;
2131 tag_set->driver_data = shost;
2132 if (shost->host_tagset)
2133 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
2134
2135 return blk_mq_alloc_tag_set(tag_set);
2136}
2137
2138void scsi_mq_free_tags(struct kref *kref)
2139{
2140 struct Scsi_Host *shost = container_of(kref, typeof(*shost),
2141 tagset_refcnt);
2142
2143 blk_mq_free_tag_set(&shost->tag_set);
2144 complete(&shost->tagset_freed);
2145}
2146
2147/**
2148 * scsi_get_internal_cmd() - Allocate an internal SCSI command.
2149 * @sdev: SCSI device from which to allocate the command
2150 * @data_direction: Data direction for the allocated command
2151 * @flags: request allocation flags, e.g. BLK_MQ_REQ_RESERVED or
2152 * BLK_MQ_REQ_NOWAIT.
2153 *
2154 * Allocates a SCSI command for internal LLDD use.
2155 */
2156struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
2157 enum dma_data_direction data_direction,
2158 blk_mq_req_flags_t flags)
2159{
2160 enum req_op op = data_direction == DMA_TO_DEVICE ? REQ_OP_DRV_OUT :
2161 REQ_OP_DRV_IN;
2162 struct scsi_cmnd *scmd;
2163 struct request *rq;
2164
2165 rq = scsi_alloc_request(sdev->request_queue, op, flags);
2166 if (IS_ERR(rq))
2167 return NULL;
2168 scmd = blk_mq_rq_to_pdu(rq);
2169 scmd->device = sdev;
2170
2171 return scmd;
2172}
2173EXPORT_SYMBOL_GPL(scsi_get_internal_cmd);
2174
2175/**
2176 * scsi_put_internal_cmd() - Free an internal SCSI command.
2177 * @scmd: SCSI command to be freed
2178 */
2179void scsi_put_internal_cmd(struct scsi_cmnd *scmd)
2180{
2181 blk_mq_free_request(blk_mq_rq_from_pdu(scmd));
2182}
2183EXPORT_SYMBOL_GPL(scsi_put_internal_cmd);
2184
2185/**
2186 * scsi_device_from_queue - return sdev associated with a request_queue
2187 * @q: The request queue to return the sdev from
2188 *
2189 * Return the sdev associated with a request queue or NULL if the
2190 * request_queue does not reference a SCSI device.
2191 */
2192struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2193{
2194 struct scsi_device *sdev = NULL;
2195
2196 if (q->mq_ops == &scsi_mq_ops_no_commit ||
2197 q->mq_ops == &scsi_mq_ops)
2198 sdev = q->queuedata;
2199 if (!sdev || !get_device(&sdev->sdev_gendev))
2200 sdev = NULL;
2201
2202 return sdev;
2203}
2204/*
2205 * pktcdvd should have been integrated into the SCSI layers, but for historical
2206 * reasons like the old IDE driver it isn't. This export allows it to safely
2207 * probe if a given device is a SCSI one and only attach to that.
2208 */
2209#ifdef CONFIG_CDROM_PKTCDVD_MODULE
2210EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2211#endif
2212
2213/**
2214 * scsi_block_requests - Utility function used by low-level drivers to prevent
2215 * further commands from being queued to the device.
2216 * @shost: host in question
2217 *
2218 * There is no timer nor any other means by which the requests get unblocked
2219 * other than the low-level driver calling scsi_unblock_requests().
2220 */
2221void scsi_block_requests(struct Scsi_Host *shost)
2222{
2223 shost->host_self_blocked = 1;
2224}
2225EXPORT_SYMBOL(scsi_block_requests);
2226
2227/**
2228 * scsi_unblock_requests - Utility function used by low-level drivers to allow
2229 * further commands to be queued to the device.
2230 * @shost: host in question
2231 *
2232 * There is no timer nor any other means by which the requests get unblocked
2233 * other than the low-level driver calling scsi_unblock_requests(). This is done
2234 * as an API function so that changes to the internals of the scsi mid-layer
2235 * won't require wholesale changes to drivers that use this feature.
2236 */
2237void scsi_unblock_requests(struct Scsi_Host *shost)
2238{
2239 shost->host_self_blocked = 0;
2240 scsi_run_host_queues(shost);
2241}
2242EXPORT_SYMBOL(scsi_unblock_requests);
2243
2244void scsi_exit_queue(void)
2245{
2246 kmem_cache_destroy(scsi_sense_cache);
2247}
2248
2249/**
2250 * scsi_mode_select - issue a mode select
2251 * @sdev: SCSI device to be queried
2252 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2253 * @sp: Save page bit (0 == don't save, 1 == save)
2254 * @buffer: request buffer (may not be smaller than eight bytes)
2255 * @len: length of request buffer.
2256 * @timeout: command timeout
2257 * @retries: number of retries before failing
2258 * @data: returns a structure abstracting the mode header data
2259 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2260 * must be SCSI_SENSE_BUFFERSIZE big.
2261 *
2262 * Returns zero if successful; negative error number or scsi
2263 * status on error
2264 *
2265 */
2266int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2267 unsigned char *buffer, int len, int timeout, int retries,
2268 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2269{
2270 unsigned char cmd[10];
2271 unsigned char *real_buffer;
2272 const struct scsi_exec_args exec_args = {
2273 .sshdr = sshdr,
2274 };
2275 int ret;
2276
2277 memset(cmd, 0, sizeof(cmd));
2278 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2279
2280 /*
2281 * Use MODE SELECT(10) if the device asked for it or if the mode page
2282 * and the mode select header cannot fit within the maximumm 255 bytes
2283 * of the MODE SELECT(6) command.
2284 */
2285 if (sdev->use_10_for_ms ||
2286 len + 4 > 255 ||
2287 data->block_descriptor_length > 255) {
2288 if (len > 65535 - 8)
2289 return -EINVAL;
2290 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2291 if (!real_buffer)
2292 return -ENOMEM;
2293 memcpy(real_buffer + 8, buffer, len);
2294 len += 8;
2295 real_buffer[0] = 0;
2296 real_buffer[1] = 0;
2297 real_buffer[2] = data->medium_type;
2298 real_buffer[3] = data->device_specific;
2299 real_buffer[4] = data->longlba ? 0x01 : 0;
2300 real_buffer[5] = 0;
2301 put_unaligned_be16(data->block_descriptor_length,
2302 &real_buffer[6]);
2303
2304 cmd[0] = MODE_SELECT_10;
2305 put_unaligned_be16(len, &cmd[7]);
2306 } else {
2307 if (data->longlba)
2308 return -EINVAL;
2309
2310 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2311 if (!real_buffer)
2312 return -ENOMEM;
2313 memcpy(real_buffer + 4, buffer, len);
2314 len += 4;
2315 real_buffer[0] = 0;
2316 real_buffer[1] = data->medium_type;
2317 real_buffer[2] = data->device_specific;
2318 real_buffer[3] = data->block_descriptor_length;
2319
2320 cmd[0] = MODE_SELECT;
2321 cmd[4] = len;
2322 }
2323
2324 ret = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, real_buffer, len,
2325 timeout, retries, &exec_args);
2326 kfree(real_buffer);
2327 return ret;
2328}
2329EXPORT_SYMBOL_GPL(scsi_mode_select);
2330
2331/**
2332 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2333 * @sdev: SCSI device to be queried
2334 * @dbd: set to prevent mode sense from returning block descriptors
2335 * @modepage: mode page being requested
2336 * @subpage: sub-page of the mode page being requested
2337 * @buffer: request buffer (may not be smaller than eight bytes)
2338 * @len: length of request buffer.
2339 * @timeout: command timeout
2340 * @retries: number of retries before failing
2341 * @data: returns a structure abstracting the mode header data
2342 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2343 * must be SCSI_SENSE_BUFFERSIZE big.
2344 *
2345 * Returns zero if successful, or a negative error number on failure
2346 */
2347int
2348scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, int subpage,
2349 unsigned char *buffer, int len, int timeout, int retries,
2350 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2351{
2352 unsigned char cmd[12];
2353 int use_10_for_ms;
2354 int header_length;
2355 int result;
2356 struct scsi_sense_hdr my_sshdr;
2357 struct scsi_failure failure_defs[] = {
2358 {
2359 .sense = UNIT_ATTENTION,
2360 .asc = SCMD_FAILURE_ASC_ANY,
2361 .ascq = SCMD_FAILURE_ASCQ_ANY,
2362 .allowed = retries,
2363 .result = SAM_STAT_CHECK_CONDITION,
2364 },
2365 {}
2366 };
2367 struct scsi_failures failures = {
2368 .failure_definitions = failure_defs,
2369 };
2370 const struct scsi_exec_args exec_args = {
2371 /* caller might not be interested in sense, but we need it */
2372 .sshdr = sshdr ? : &my_sshdr,
2373 .failures = &failures,
2374 };
2375
2376 memset(data, 0, sizeof(*data));
2377 memset(&cmd[0], 0, 12);
2378
2379 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2380 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2381 cmd[2] = modepage;
2382 cmd[3] = subpage;
2383
2384 sshdr = exec_args.sshdr;
2385
2386 retry:
2387 use_10_for_ms = sdev->use_10_for_ms || len > 255;
2388
2389 if (use_10_for_ms) {
2390 if (len < 8 || len > 65535)
2391 return -EINVAL;
2392
2393 cmd[0] = MODE_SENSE_10;
2394 put_unaligned_be16(len, &cmd[7]);
2395 header_length = 8;
2396 } else {
2397 if (len < 4)
2398 return -EINVAL;
2399
2400 cmd[0] = MODE_SENSE;
2401 cmd[4] = len;
2402 header_length = 4;
2403 }
2404
2405 memset(buffer, 0, len);
2406
2407 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, buffer, len,
2408 timeout, retries, &exec_args);
2409 if (result < 0)
2410 return result;
2411
2412 /* This code looks awful: what it's doing is making sure an
2413 * ILLEGAL REQUEST sense return identifies the actual command
2414 * byte as the problem. MODE_SENSE commands can return
2415 * ILLEGAL REQUEST if the code page isn't supported */
2416
2417 if (!scsi_status_is_good(result)) {
2418 if (scsi_sense_valid(sshdr)) {
2419 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2420 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2421 /*
2422 * Invalid command operation code: retry using
2423 * MODE SENSE(6) if this was a MODE SENSE(10)
2424 * request, except if the request mode page is
2425 * too large for MODE SENSE single byte
2426 * allocation length field.
2427 */
2428 if (use_10_for_ms) {
2429 if (len > 255)
2430 return -EIO;
2431 sdev->use_10_for_ms = 0;
2432 goto retry;
2433 }
2434 }
2435 }
2436 return -EIO;
2437 }
2438 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2439 (modepage == 6 || modepage == 8))) {
2440 /* Initio breakage? */
2441 header_length = 0;
2442 data->length = 13;
2443 data->medium_type = 0;
2444 data->device_specific = 0;
2445 data->longlba = 0;
2446 data->block_descriptor_length = 0;
2447 } else if (use_10_for_ms) {
2448 data->length = get_unaligned_be16(&buffer[0]) + 2;
2449 data->medium_type = buffer[2];
2450 data->device_specific = buffer[3];
2451 data->longlba = buffer[4] & 0x01;
2452 data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2453 } else {
2454 data->length = buffer[0] + 1;
2455 data->medium_type = buffer[1];
2456 data->device_specific = buffer[2];
2457 data->block_descriptor_length = buffer[3];
2458 }
2459 data->header_length = header_length;
2460
2461 return 0;
2462}
2463EXPORT_SYMBOL(scsi_mode_sense);
2464
2465/**
2466 * scsi_test_unit_ready - test if unit is ready
2467 * @sdev: scsi device to change the state of.
2468 * @timeout: command timeout
2469 * @retries: number of retries before failing
2470 * @sshdr: outpout pointer for decoded sense information.
2471 *
2472 * Returns zero if successful or an error if TUR failed. For
2473 * removable media, UNIT_ATTENTION sets ->changed flag.
2474 **/
2475int
2476scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2477 struct scsi_sense_hdr *sshdr)
2478{
2479 char cmd[] = {
2480 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2481 };
2482 const struct scsi_exec_args exec_args = {
2483 .sshdr = sshdr,
2484 };
2485 int result;
2486
2487 /* try to eat the UNIT_ATTENTION if there are enough retries */
2488 do {
2489 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, NULL, 0,
2490 timeout, 1, &exec_args);
2491 if (sdev->removable && result > 0 && scsi_sense_valid(sshdr) &&
2492 sshdr->sense_key == UNIT_ATTENTION)
2493 sdev->changed = 1;
2494 } while (result > 0 && scsi_sense_valid(sshdr) &&
2495 sshdr->sense_key == UNIT_ATTENTION && --retries);
2496
2497 return result;
2498}
2499EXPORT_SYMBOL(scsi_test_unit_ready);
2500
2501/**
2502 * scsi_device_set_state - Take the given device through the device state model.
2503 * @sdev: scsi device to change the state of.
2504 * @state: state to change to.
2505 *
2506 * Returns zero if successful or an error if the requested
2507 * transition is illegal.
2508 */
2509int
2510scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2511{
2512 enum scsi_device_state oldstate = sdev->sdev_state;
2513
2514 if (state == oldstate)
2515 return 0;
2516
2517 switch (state) {
2518 case SDEV_CREATED:
2519 switch (oldstate) {
2520 case SDEV_CREATED_BLOCK:
2521 break;
2522 default:
2523 goto illegal;
2524 }
2525 break;
2526
2527 case SDEV_RUNNING:
2528 switch (oldstate) {
2529 case SDEV_CREATED:
2530 case SDEV_OFFLINE:
2531 case SDEV_TRANSPORT_OFFLINE:
2532 case SDEV_QUIESCE:
2533 case SDEV_BLOCK:
2534 break;
2535 default:
2536 goto illegal;
2537 }
2538 break;
2539
2540 case SDEV_QUIESCE:
2541 switch (oldstate) {
2542 case SDEV_RUNNING:
2543 case SDEV_OFFLINE:
2544 case SDEV_TRANSPORT_OFFLINE:
2545 break;
2546 default:
2547 goto illegal;
2548 }
2549 break;
2550
2551 case SDEV_OFFLINE:
2552 case SDEV_TRANSPORT_OFFLINE:
2553 switch (oldstate) {
2554 case SDEV_CREATED:
2555 case SDEV_RUNNING:
2556 case SDEV_QUIESCE:
2557 case SDEV_BLOCK:
2558 break;
2559 default:
2560 goto illegal;
2561 }
2562 break;
2563
2564 case SDEV_BLOCK:
2565 switch (oldstate) {
2566 case SDEV_RUNNING:
2567 case SDEV_CREATED_BLOCK:
2568 case SDEV_QUIESCE:
2569 case SDEV_OFFLINE:
2570 break;
2571 default:
2572 goto illegal;
2573 }
2574 break;
2575
2576 case SDEV_CREATED_BLOCK:
2577 switch (oldstate) {
2578 case SDEV_CREATED:
2579 break;
2580 default:
2581 goto illegal;
2582 }
2583 break;
2584
2585 case SDEV_CANCEL:
2586 switch (oldstate) {
2587 case SDEV_CREATED:
2588 case SDEV_RUNNING:
2589 case SDEV_QUIESCE:
2590 case SDEV_OFFLINE:
2591 case SDEV_TRANSPORT_OFFLINE:
2592 break;
2593 default:
2594 goto illegal;
2595 }
2596 break;
2597
2598 case SDEV_DEL:
2599 switch (oldstate) {
2600 case SDEV_CREATED:
2601 case SDEV_RUNNING:
2602 case SDEV_OFFLINE:
2603 case SDEV_TRANSPORT_OFFLINE:
2604 case SDEV_CANCEL:
2605 case SDEV_BLOCK:
2606 case SDEV_CREATED_BLOCK:
2607 break;
2608 default:
2609 goto illegal;
2610 }
2611 break;
2612
2613 }
2614 sdev->offline_already = false;
2615 sdev->sdev_state = state;
2616 return 0;
2617
2618 illegal:
2619 SCSI_LOG_ERROR_RECOVERY(1,
2620 sdev_printk(KERN_ERR, sdev,
2621 "Illegal state transition %s->%s",
2622 scsi_device_state_name(oldstate),
2623 scsi_device_state_name(state))
2624 );
2625 return -EINVAL;
2626}
2627EXPORT_SYMBOL(scsi_device_set_state);
2628
2629/**
2630 * scsi_evt_emit - emit a single SCSI device uevent
2631 * @sdev: associated SCSI device
2632 * @evt: event to emit
2633 *
2634 * Send a single uevent (scsi_event) to the associated scsi_device.
2635 */
2636static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2637{
2638 int idx = 0;
2639 char *envp[3];
2640
2641 switch (evt->evt_type) {
2642 case SDEV_EVT_MEDIA_CHANGE:
2643 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2644 break;
2645 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2646 scsi_rescan_device(sdev);
2647 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2648 break;
2649 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2650 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2651 break;
2652 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2653 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2654 break;
2655 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2656 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2657 break;
2658 case SDEV_EVT_LUN_CHANGE_REPORTED:
2659 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2660 break;
2661 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2662 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2663 break;
2664 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2665 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2666 break;
2667 default:
2668 /* do nothing */
2669 break;
2670 }
2671
2672 envp[idx++] = NULL;
2673
2674 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2675}
2676
2677/**
2678 * scsi_evt_thread - send a uevent for each scsi event
2679 * @work: work struct for scsi_device
2680 *
2681 * Dispatch queued events to their associated scsi_device kobjects
2682 * as uevents.
2683 */
2684void scsi_evt_thread(struct work_struct *work)
2685{
2686 struct scsi_device *sdev;
2687 enum scsi_device_event evt_type;
2688 LIST_HEAD(event_list);
2689
2690 sdev = container_of(work, struct scsi_device, event_work);
2691
2692 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2693 if (test_and_clear_bit(evt_type, sdev->pending_events))
2694 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2695
2696 while (1) {
2697 struct scsi_event *evt;
2698 struct list_head *this, *tmp;
2699 unsigned long flags;
2700
2701 spin_lock_irqsave(&sdev->list_lock, flags);
2702 list_splice_init(&sdev->event_list, &event_list);
2703 spin_unlock_irqrestore(&sdev->list_lock, flags);
2704
2705 if (list_empty(&event_list))
2706 break;
2707
2708 list_for_each_safe(this, tmp, &event_list) {
2709 evt = list_entry(this, struct scsi_event, node);
2710 list_del(&evt->node);
2711 scsi_evt_emit(sdev, evt);
2712 kfree(evt);
2713 }
2714 }
2715}
2716
2717/**
2718 * sdev_evt_send - send asserted event to uevent thread
2719 * @sdev: scsi_device event occurred on
2720 * @evt: event to send
2721 *
2722 * Assert scsi device event asynchronously.
2723 */
2724void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2725{
2726 unsigned long flags;
2727
2728#if 0
2729 /* FIXME: currently this check eliminates all media change events
2730 * for polled devices. Need to update to discriminate between AN
2731 * and polled events */
2732 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2733 kfree(evt);
2734 return;
2735 }
2736#endif
2737
2738 spin_lock_irqsave(&sdev->list_lock, flags);
2739 list_add_tail(&evt->node, &sdev->event_list);
2740 schedule_work(&sdev->event_work);
2741 spin_unlock_irqrestore(&sdev->list_lock, flags);
2742}
2743EXPORT_SYMBOL_GPL(sdev_evt_send);
2744
2745/**
2746 * sdev_evt_alloc - allocate a new scsi event
2747 * @evt_type: type of event to allocate
2748 * @gfpflags: GFP flags for allocation
2749 *
2750 * Allocates and returns a new scsi_event.
2751 */
2752struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2753 gfp_t gfpflags)
2754{
2755 struct scsi_event *evt = kzalloc_obj(struct scsi_event, gfpflags);
2756 if (!evt)
2757 return NULL;
2758
2759 evt->evt_type = evt_type;
2760 INIT_LIST_HEAD(&evt->node);
2761
2762 /* evt_type-specific initialization, if any */
2763 switch (evt_type) {
2764 case SDEV_EVT_MEDIA_CHANGE:
2765 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2766 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2767 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2768 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2769 case SDEV_EVT_LUN_CHANGE_REPORTED:
2770 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2771 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2772 default:
2773 /* do nothing */
2774 break;
2775 }
2776
2777 return evt;
2778}
2779EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2780
2781/**
2782 * sdev_evt_send_simple - send asserted event to uevent thread
2783 * @sdev: scsi_device event occurred on
2784 * @evt_type: type of event to send
2785 * @gfpflags: GFP flags for allocation
2786 *
2787 * Assert scsi device event asynchronously, given an event type.
2788 */
2789void sdev_evt_send_simple(struct scsi_device *sdev,
2790 enum scsi_device_event evt_type, gfp_t gfpflags)
2791{
2792 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2793 if (!evt) {
2794 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2795 evt_type);
2796 return;
2797 }
2798
2799 sdev_evt_send(sdev, evt);
2800}
2801EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2802
2803/**
2804 * scsi_device_quiesce - Block all commands except power management.
2805 * @sdev: scsi device to quiesce.
2806 *
2807 * This works by trying to transition to the SDEV_QUIESCE state
2808 * (which must be a legal transition). When the device is in this
2809 * state, only power management requests will be accepted, all others will
2810 * be deferred.
2811 *
2812 * Must be called with user context, may sleep.
2813 *
2814 * Returns zero if successful or an error if not.
2815 */
2816int
2817scsi_device_quiesce(struct scsi_device *sdev)
2818{
2819 struct request_queue *q = sdev->request_queue;
2820 unsigned int memflags;
2821 int err;
2822
2823 /*
2824 * It is allowed to call scsi_device_quiesce() multiple times from
2825 * the same context but concurrent scsi_device_quiesce() calls are
2826 * not allowed.
2827 */
2828 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2829
2830 if (sdev->quiesced_by == current)
2831 return 0;
2832
2833 blk_set_pm_only(q);
2834
2835 memflags = blk_mq_freeze_queue(q);
2836 /*
2837 * Ensure that the effect of blk_set_pm_only() will be visible
2838 * for percpu_ref_tryget() callers that occur after the queue
2839 * unfreeze even if the queue was already frozen before this function
2840 * was called. See also https://lwn.net/Articles/573497/.
2841 */
2842 synchronize_rcu();
2843 blk_mq_unfreeze_queue(q, memflags);
2844
2845 mutex_lock(&sdev->state_mutex);
2846 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2847 if (err == 0)
2848 sdev->quiesced_by = current;
2849 else
2850 blk_clear_pm_only(q);
2851 mutex_unlock(&sdev->state_mutex);
2852
2853 return err;
2854}
2855EXPORT_SYMBOL(scsi_device_quiesce);
2856
2857/**
2858 * scsi_device_resume - Restart user issued commands to a quiesced device.
2859 * @sdev: scsi device to resume.
2860 *
2861 * Moves the device from quiesced back to running and restarts the
2862 * queues.
2863 *
2864 * Must be called with user context, may sleep.
2865 */
2866void scsi_device_resume(struct scsi_device *sdev)
2867{
2868 /* check if the device state was mutated prior to resume, and if
2869 * so assume the state is being managed elsewhere (for example
2870 * device deleted during suspend)
2871 */
2872 mutex_lock(&sdev->state_mutex);
2873 if (sdev->sdev_state == SDEV_QUIESCE)
2874 scsi_device_set_state(sdev, SDEV_RUNNING);
2875 if (sdev->quiesced_by) {
2876 sdev->quiesced_by = NULL;
2877 blk_clear_pm_only(sdev->request_queue);
2878 }
2879 mutex_unlock(&sdev->state_mutex);
2880}
2881EXPORT_SYMBOL(scsi_device_resume);
2882
2883static void
2884device_quiesce_fn(struct scsi_device *sdev, void *data)
2885{
2886 scsi_device_quiesce(sdev);
2887}
2888
2889void
2890scsi_target_quiesce(struct scsi_target *starget)
2891{
2892 starget_for_each_device(starget, NULL, device_quiesce_fn);
2893}
2894EXPORT_SYMBOL(scsi_target_quiesce);
2895
2896static void
2897device_resume_fn(struct scsi_device *sdev, void *data)
2898{
2899 scsi_device_resume(sdev);
2900}
2901
2902void
2903scsi_target_resume(struct scsi_target *starget)
2904{
2905 starget_for_each_device(starget, NULL, device_resume_fn);
2906}
2907EXPORT_SYMBOL(scsi_target_resume);
2908
2909static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2910{
2911 if (scsi_device_set_state(sdev, SDEV_BLOCK))
2912 return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2913
2914 return 0;
2915}
2916
2917void scsi_start_queue(struct scsi_device *sdev)
2918{
2919 if (cmpxchg(&sdev->queue_stopped, 1, 0))
2920 blk_mq_unquiesce_queue(sdev->request_queue);
2921}
2922
2923static void scsi_stop_queue(struct scsi_device *sdev)
2924{
2925 /*
2926 * The atomic variable of ->queue_stopped covers that
2927 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2928 *
2929 * The caller needs to wait until quiesce is done.
2930 */
2931 if (!cmpxchg(&sdev->queue_stopped, 0, 1))
2932 blk_mq_quiesce_queue_nowait(sdev->request_queue);
2933}
2934
2935/**
2936 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2937 * @sdev: device to block
2938 *
2939 * Pause SCSI command processing on the specified device. Does not sleep.
2940 *
2941 * Returns zero if successful or a negative error code upon failure.
2942 *
2943 * Notes:
2944 * This routine transitions the device to the SDEV_BLOCK state (which must be
2945 * a legal transition). When the device is in this state, command processing
2946 * is paused until the device leaves the SDEV_BLOCK state. See also
2947 * scsi_internal_device_unblock_nowait().
2948 */
2949int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2950{
2951 int ret = __scsi_internal_device_block_nowait(sdev);
2952
2953 /*
2954 * The device has transitioned to SDEV_BLOCK. Stop the
2955 * block layer from calling the midlayer with this device's
2956 * request queue.
2957 */
2958 if (!ret)
2959 scsi_stop_queue(sdev);
2960 return ret;
2961}
2962EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2963
2964/**
2965 * scsi_device_block - try to transition to the SDEV_BLOCK state
2966 * @sdev: device to block
2967 * @data: dummy argument, ignored
2968 *
2969 * Pause SCSI command processing on the specified device. Callers must wait
2970 * until all ongoing scsi_queue_rq() calls have finished after this function
2971 * returns.
2972 *
2973 * Note:
2974 * This routine transitions the device to the SDEV_BLOCK state (which must be
2975 * a legal transition). When the device is in this state, command processing
2976 * is paused until the device leaves the SDEV_BLOCK state. See also
2977 * scsi_internal_device_unblock().
2978 */
2979static void scsi_device_block(struct scsi_device *sdev, void *data)
2980{
2981 int err;
2982 enum scsi_device_state state;
2983
2984 mutex_lock(&sdev->state_mutex);
2985 err = __scsi_internal_device_block_nowait(sdev);
2986 state = sdev->sdev_state;
2987 if (err == 0)
2988 /*
2989 * scsi_stop_queue() must be called with the state_mutex
2990 * held. Otherwise a simultaneous scsi_start_queue() call
2991 * might unquiesce the queue before we quiesce it.
2992 */
2993 scsi_stop_queue(sdev);
2994
2995 mutex_unlock(&sdev->state_mutex);
2996
2997 WARN_ONCE(err, "%s: failed to block %s in state %d\n",
2998 __func__, dev_name(&sdev->sdev_gendev), state);
2999}
3000
3001/**
3002 * scsi_internal_device_unblock_nowait - resume a device after a block request
3003 * @sdev: device to resume
3004 * @new_state: state to set the device to after unblocking
3005 *
3006 * Restart the device queue for a previously suspended SCSI device. Does not
3007 * sleep.
3008 *
3009 * Returns zero if successful or a negative error code upon failure.
3010 *
3011 * Notes:
3012 * This routine transitions the device to the SDEV_RUNNING state or to one of
3013 * the offline states (which must be a legal transition) allowing the midlayer
3014 * to goose the queue for this device.
3015 */
3016int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
3017 enum scsi_device_state new_state)
3018{
3019 switch (new_state) {
3020 case SDEV_RUNNING:
3021 case SDEV_TRANSPORT_OFFLINE:
3022 break;
3023 default:
3024 return -EINVAL;
3025 }
3026
3027 /*
3028 * Try to transition the scsi device to SDEV_RUNNING or one of the
3029 * offlined states and goose the device queue if successful.
3030 */
3031 switch (sdev->sdev_state) {
3032 case SDEV_BLOCK:
3033 case SDEV_TRANSPORT_OFFLINE:
3034 sdev->sdev_state = new_state;
3035 break;
3036 case SDEV_CREATED_BLOCK:
3037 if (new_state == SDEV_TRANSPORT_OFFLINE ||
3038 new_state == SDEV_OFFLINE)
3039 sdev->sdev_state = new_state;
3040 else
3041 sdev->sdev_state = SDEV_CREATED;
3042 break;
3043 case SDEV_CANCEL:
3044 case SDEV_OFFLINE:
3045 break;
3046 default:
3047 return -EINVAL;
3048 }
3049 scsi_start_queue(sdev);
3050
3051 return 0;
3052}
3053EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
3054
3055/**
3056 * scsi_internal_device_unblock - resume a device after a block request
3057 * @sdev: device to resume
3058 * @new_state: state to set the device to after unblocking
3059 *
3060 * Restart the device queue for a previously suspended SCSI device. May sleep.
3061 *
3062 * Returns zero if successful or a negative error code upon failure.
3063 *
3064 * Notes:
3065 * This routine transitions the device to the SDEV_RUNNING state or to one of
3066 * the offline states (which must be a legal transition) allowing the midlayer
3067 * to goose the queue for this device.
3068 */
3069static int scsi_internal_device_unblock(struct scsi_device *sdev,
3070 enum scsi_device_state new_state)
3071{
3072 int ret;
3073
3074 mutex_lock(&sdev->state_mutex);
3075 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
3076 mutex_unlock(&sdev->state_mutex);
3077
3078 return ret;
3079}
3080
3081static int
3082target_block(struct device *dev, void *data)
3083{
3084 if (scsi_is_target_device(dev))
3085 starget_for_each_device(to_scsi_target(dev), NULL,
3086 scsi_device_block);
3087 return 0;
3088}
3089
3090/**
3091 * scsi_block_targets - transition all SCSI child devices to SDEV_BLOCK state
3092 * @dev: a parent device of one or more scsi_target devices
3093 * @shost: the Scsi_Host to which this device belongs
3094 *
3095 * Iterate over all children of @dev, which should be scsi_target devices,
3096 * and switch all subordinate scsi devices to SDEV_BLOCK state. Wait for
3097 * ongoing scsi_queue_rq() calls to finish. May sleep.
3098 *
3099 * Note:
3100 * @dev must not itself be a scsi_target device.
3101 */
3102void
3103scsi_block_targets(struct Scsi_Host *shost, struct device *dev)
3104{
3105 WARN_ON_ONCE(scsi_is_target_device(dev));
3106 device_for_each_child(dev, NULL, target_block);
3107 blk_mq_wait_quiesce_done(&shost->tag_set);
3108}
3109EXPORT_SYMBOL_GPL(scsi_block_targets);
3110
3111static void
3112device_unblock(struct scsi_device *sdev, void *data)
3113{
3114 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3115}
3116
3117static int
3118target_unblock(struct device *dev, void *data)
3119{
3120 if (scsi_is_target_device(dev))
3121 starget_for_each_device(to_scsi_target(dev), data,
3122 device_unblock);
3123 return 0;
3124}
3125
3126void
3127scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3128{
3129 if (scsi_is_target_device(dev))
3130 starget_for_each_device(to_scsi_target(dev), &new_state,
3131 device_unblock);
3132 else
3133 device_for_each_child(dev, &new_state, target_unblock);
3134}
3135EXPORT_SYMBOL_GPL(scsi_target_unblock);
3136
3137/**
3138 * scsi_host_block - Try to transition all logical units to the SDEV_BLOCK state
3139 * @shost: device to block
3140 *
3141 * Pause SCSI command processing for all logical units associated with the SCSI
3142 * host and wait until pending scsi_queue_rq() calls have finished.
3143 *
3144 * Returns zero if successful or a negative error code upon failure.
3145 */
3146int
3147scsi_host_block(struct Scsi_Host *shost)
3148{
3149 struct scsi_device *sdev;
3150 int ret;
3151
3152 /*
3153 * Call scsi_internal_device_block_nowait so we can avoid
3154 * calling synchronize_rcu() for each LUN.
3155 */
3156 shost_for_each_device(sdev, shost) {
3157 mutex_lock(&sdev->state_mutex);
3158 ret = scsi_internal_device_block_nowait(sdev);
3159 mutex_unlock(&sdev->state_mutex);
3160 if (ret) {
3161 scsi_device_put(sdev);
3162 return ret;
3163 }
3164 }
3165
3166 /* Wait for ongoing scsi_queue_rq() calls to finish. */
3167 blk_mq_wait_quiesce_done(&shost->tag_set);
3168
3169 return 0;
3170}
3171EXPORT_SYMBOL_GPL(scsi_host_block);
3172
3173int
3174scsi_host_unblock(struct Scsi_Host *shost, int new_state)
3175{
3176 struct scsi_device *sdev;
3177 int ret = 0;
3178
3179 shost_for_each_device(sdev, shost) {
3180 ret = scsi_internal_device_unblock(sdev, new_state);
3181 if (ret) {
3182 scsi_device_put(sdev);
3183 break;
3184 }
3185 }
3186 return ret;
3187}
3188EXPORT_SYMBOL_GPL(scsi_host_unblock);
3189
3190/**
3191 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3192 * @sgl: scatter-gather list
3193 * @sg_count: number of segments in sg
3194 * @offset: offset in bytes into sg, on return offset into the mapped area
3195 * @len: bytes to map, on return number of bytes mapped
3196 *
3197 * Returns virtual address of the start of the mapped page
3198 */
3199void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3200 size_t *offset, size_t *len)
3201{
3202 int i;
3203 size_t sg_len = 0, len_complete = 0;
3204 struct scatterlist *sg;
3205 struct page *page;
3206
3207 WARN_ON(!irqs_disabled());
3208
3209 for_each_sg(sgl, sg, sg_count, i) {
3210 len_complete = sg_len; /* Complete sg-entries */
3211 sg_len += sg->length;
3212 if (sg_len > *offset)
3213 break;
3214 }
3215
3216 if (unlikely(i == sg_count)) {
3217 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3218 "elements %d\n",
3219 __func__, sg_len, *offset, sg_count);
3220 WARN_ON(1);
3221 return NULL;
3222 }
3223
3224 /* Offset starting from the beginning of first page in this sg-entry */
3225 *offset = *offset - len_complete + sg->offset;
3226
3227 page = sg_page(sg) + (*offset >> PAGE_SHIFT);
3228 *offset &= ~PAGE_MASK;
3229
3230 /* Bytes in this sg-entry from *offset to the end of the page */
3231 sg_len = PAGE_SIZE - *offset;
3232 if (*len > sg_len)
3233 *len = sg_len;
3234
3235 return kmap_atomic(page);
3236}
3237EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3238
3239/**
3240 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3241 * @virt: virtual address to be unmapped
3242 */
3243void scsi_kunmap_atomic_sg(void *virt)
3244{
3245 kunmap_atomic(virt);
3246}
3247EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3248
3249void sdev_disable_disk_events(struct scsi_device *sdev)
3250{
3251 atomic_inc(&sdev->disk_events_disable_depth);
3252}
3253EXPORT_SYMBOL(sdev_disable_disk_events);
3254
3255void sdev_enable_disk_events(struct scsi_device *sdev)
3256{
3257 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3258 return;
3259 atomic_dec(&sdev->disk_events_disable_depth);
3260}
3261EXPORT_SYMBOL(sdev_enable_disk_events);
3262
3263static unsigned char designator_prio(const unsigned char *d)
3264{
3265 if (d[1] & 0x30)
3266 /* not associated with LUN */
3267 return 0;
3268
3269 if (d[3] == 0)
3270 /* invalid length */
3271 return 0;
3272
3273 /*
3274 * Order of preference for lun descriptor:
3275 * - SCSI name string
3276 * - NAA IEEE Registered Extended
3277 * - EUI-64 based 16-byte
3278 * - EUI-64 based 12-byte
3279 * - NAA IEEE Registered
3280 * - NAA IEEE Extended
3281 * - EUI-64 based 8-byte
3282 * - SCSI name string (truncated)
3283 * - T10 Vendor ID
3284 * as longer descriptors reduce the likelyhood
3285 * of identification clashes.
3286 */
3287
3288 switch (d[1] & 0xf) {
3289 case 8:
3290 /* SCSI name string, variable-length UTF-8 */
3291 return 9;
3292 case 3:
3293 switch (d[4] >> 4) {
3294 case 6:
3295 /* NAA registered extended */
3296 return 8;
3297 case 5:
3298 /* NAA registered */
3299 return 5;
3300 case 4:
3301 /* NAA extended */
3302 return 4;
3303 case 3:
3304 /* NAA locally assigned */
3305 return 1;
3306 default:
3307 break;
3308 }
3309 break;
3310 case 2:
3311 switch (d[3]) {
3312 case 16:
3313 /* EUI64-based, 16 byte */
3314 return 7;
3315 case 12:
3316 /* EUI64-based, 12 byte */
3317 return 6;
3318 case 8:
3319 /* EUI64-based, 8 byte */
3320 return 3;
3321 default:
3322 break;
3323 }
3324 break;
3325 case 1:
3326 /* T10 vendor ID */
3327 return 1;
3328 default:
3329 break;
3330 }
3331
3332 return 0;
3333}
3334
3335/**
3336 * scsi_vpd_lun_id - return a unique device identification
3337 * @sdev: SCSI device
3338 * @id: buffer for the identification
3339 * @id_len: length of the buffer
3340 *
3341 * Copies a unique device identification into @id based
3342 * on the information in the VPD page 0x83 of the device.
3343 * The string will be formatted as a SCSI name string.
3344 *
3345 * Returns the length of the identification or error on failure.
3346 * If the identifier is longer than the supplied buffer the actual
3347 * identifier length is returned and the buffer is not zero-padded.
3348 */
3349int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3350{
3351 u8 cur_id_prio = 0;
3352 u8 cur_id_size = 0;
3353 const unsigned char *d, *cur_id_str;
3354 const struct scsi_vpd *vpd_pg83;
3355 int id_size = -EINVAL;
3356
3357 rcu_read_lock();
3358 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3359 if (!vpd_pg83) {
3360 rcu_read_unlock();
3361 return -ENXIO;
3362 }
3363
3364 /* The id string must be at least 20 bytes + terminating NULL byte */
3365 if (id_len < 21) {
3366 rcu_read_unlock();
3367 return -EINVAL;
3368 }
3369
3370 memset(id, 0, id_len);
3371 for (d = vpd_pg83->data + 4;
3372 d < vpd_pg83->data + vpd_pg83->len;
3373 d += d[3] + 4) {
3374 u8 prio = designator_prio(d);
3375
3376 if (prio == 0 || cur_id_prio > prio)
3377 continue;
3378
3379 switch (d[1] & 0xf) {
3380 case 0x1:
3381 /* T10 Vendor ID */
3382 if (cur_id_size > d[3])
3383 break;
3384 cur_id_prio = prio;
3385 cur_id_size = d[3];
3386 if (cur_id_size + 4 > id_len)
3387 cur_id_size = id_len - 4;
3388 cur_id_str = d + 4;
3389 id_size = snprintf(id, id_len, "t10.%*pE",
3390 cur_id_size, cur_id_str);
3391 break;
3392 case 0x2:
3393 /* EUI-64 */
3394 cur_id_prio = prio;
3395 cur_id_size = d[3];
3396 cur_id_str = d + 4;
3397 switch (cur_id_size) {
3398 case 8:
3399 id_size = snprintf(id, id_len,
3400 "eui.%8phN",
3401 cur_id_str);
3402 break;
3403 case 12:
3404 id_size = snprintf(id, id_len,
3405 "eui.%12phN",
3406 cur_id_str);
3407 break;
3408 case 16:
3409 id_size = snprintf(id, id_len,
3410 "eui.%16phN",
3411 cur_id_str);
3412 break;
3413 default:
3414 break;
3415 }
3416 break;
3417 case 0x3:
3418 /* NAA */
3419 cur_id_prio = prio;
3420 cur_id_size = d[3];
3421 cur_id_str = d + 4;
3422 switch (cur_id_size) {
3423 case 8:
3424 id_size = snprintf(id, id_len,
3425 "naa.%8phN",
3426 cur_id_str);
3427 break;
3428 case 16:
3429 id_size = snprintf(id, id_len,
3430 "naa.%16phN",
3431 cur_id_str);
3432 break;
3433 default:
3434 break;
3435 }
3436 break;
3437 case 0x8:
3438 /* SCSI name string */
3439 if (cur_id_size > d[3])
3440 break;
3441 /* Prefer others for truncated descriptor */
3442 if (d[3] > id_len) {
3443 prio = 2;
3444 if (cur_id_prio > prio)
3445 break;
3446 }
3447 cur_id_prio = prio;
3448 cur_id_size = id_size = d[3];
3449 cur_id_str = d + 4;
3450 if (cur_id_size >= id_len)
3451 cur_id_size = id_len - 1;
3452 memcpy(id, cur_id_str, cur_id_size);
3453 break;
3454 default:
3455 break;
3456 }
3457 }
3458 rcu_read_unlock();
3459
3460 return id_size;
3461}
3462EXPORT_SYMBOL(scsi_vpd_lun_id);
3463
3464/**
3465 * scsi_vpd_lun_serial - return a unique device serial number
3466 * @sdev: SCSI device
3467 * @sn: buffer for the serial number
3468 * @sn_size: size of the buffer
3469 *
3470 * Copies the device serial number into @sn based on the information in
3471 * the VPD page 0x80 of the device. The string will be null terminated
3472 * and have leading and trailing whitespace stripped.
3473 *
3474 * Returns the length of the serial number or error on failure.
3475 */
3476int scsi_vpd_lun_serial(struct scsi_device *sdev, char *sn, size_t sn_size)
3477{
3478 const struct scsi_vpd *vpd_pg80;
3479 const unsigned char *d;
3480 int len;
3481
3482 guard(rcu)();
3483 vpd_pg80 = rcu_dereference(sdev->vpd_pg80);
3484 if (!vpd_pg80)
3485 return -ENXIO;
3486
3487 len = vpd_pg80->len - 4;
3488 d = vpd_pg80->data + 4;
3489
3490 /* Skip leading spaces */
3491 while (len > 0 && isspace(*d)) {
3492 len--;
3493 d++;
3494 }
3495
3496 /* Skip trailing spaces */
3497 while (len > 0 && isspace(d[len - 1]))
3498 len--;
3499
3500 if (sn_size < len + 1)
3501 return -EINVAL;
3502
3503 memcpy(sn, d, len);
3504 sn[len] = '\0';
3505
3506 return len;
3507}
3508EXPORT_SYMBOL(scsi_vpd_lun_serial);
3509
3510/**
3511 * scsi_vpd_tpg_id - return a target port group identifier
3512 * @sdev: SCSI device
3513 * @rel_id: pointer to return relative target port in if not %NULL
3514 *
3515 * Returns the Target Port Group identifier from the information
3516 * from VPD page 0x83 of the device.
3517 * Optionally sets @rel_id to the relative target port on success.
3518 *
3519 * Return: the identifier or error on failure.
3520 */
3521int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3522{
3523 const unsigned char *d;
3524 const struct scsi_vpd *vpd_pg83;
3525 int group_id = -EAGAIN, rel_port = -1;
3526
3527 rcu_read_lock();
3528 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3529 if (!vpd_pg83) {
3530 rcu_read_unlock();
3531 return -ENXIO;
3532 }
3533
3534 d = vpd_pg83->data + 4;
3535 while (d < vpd_pg83->data + vpd_pg83->len) {
3536 switch (d[1] & 0xf) {
3537 case 0x4:
3538 /* Relative target port */
3539 rel_port = get_unaligned_be16(&d[6]);
3540 break;
3541 case 0x5:
3542 /* Target port group */
3543 group_id = get_unaligned_be16(&d[6]);
3544 break;
3545 default:
3546 break;
3547 }
3548 d += d[3] + 4;
3549 }
3550 rcu_read_unlock();
3551
3552 if (group_id >= 0 && rel_id && rel_port != -1)
3553 *rel_id = rel_port;
3554
3555 return group_id;
3556}
3557EXPORT_SYMBOL(scsi_vpd_tpg_id);
3558
3559/**
3560 * scsi_build_sense - build sense data for a command
3561 * @scmd: scsi command for which the sense should be formatted
3562 * @desc: Sense format (non-zero == descriptor format,
3563 * 0 == fixed format)
3564 * @key: Sense key
3565 * @asc: Additional sense code
3566 * @ascq: Additional sense code qualifier
3567 *
3568 **/
3569void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3570{
3571 scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3572 scmd->result = SAM_STAT_CHECK_CONDITION;
3573}
3574EXPORT_SYMBOL_GPL(scsi_build_sense);
3575
3576#ifdef CONFIG_SCSI_LIB_KUNIT_TEST
3577#include "scsi_lib_test.c"
3578#endif