xref: /linux/drivers/scsi/scsi_lib.c (revision fe833e4397fbdc3ae13a60202dfc7c335b032499)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 1999 Eric Youngdale
4  * Copyright (C) 2014 Christoph Hellwig
5  *
6  *  SCSI queueing library.
7  *      Initial versions: Eric Youngdale (eric@andante.org).
8  *                        Based upon conversations with large numbers
9  *                        of people at Linux Expo.
10  */
11 
12 #include <linux/bio.h>
13 #include <linux/bitops.h>
14 #include <linux/blkdev.h>
15 #include <linux/completion.h>
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/init.h>
19 #include <linux/pci.h>
20 #include <linux/delay.h>
21 #include <linux/hardirq.h>
22 #include <linux/scatterlist.h>
23 #include <linux/blk-mq.h>
24 #include <linux/blk-integrity.h>
25 #include <linux/ratelimit.h>
26 #include <asm/unaligned.h>
27 
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_driver.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_transport.h> /* scsi_init_limits() */
36 #include <scsi/scsi_dh.h>
37 
38 #include <trace/events/scsi.h>
39 
40 #include "scsi_debugfs.h"
41 #include "scsi_priv.h"
42 #include "scsi_logging.h"
43 
44 /*
45  * Size of integrity metadata is usually small, 1 inline sg should
46  * cover normal cases.
47  */
48 #ifdef CONFIG_ARCH_NO_SG_CHAIN
49 #define  SCSI_INLINE_PROT_SG_CNT  0
50 #define  SCSI_INLINE_SG_CNT  0
51 #else
52 #define  SCSI_INLINE_PROT_SG_CNT  1
53 #define  SCSI_INLINE_SG_CNT  2
54 #endif
55 
56 static struct kmem_cache *scsi_sense_cache;
57 static DEFINE_MUTEX(scsi_sense_cache_mutex);
58 
59 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
60 
61 int scsi_init_sense_cache(struct Scsi_Host *shost)
62 {
63 	int ret = 0;
64 
65 	mutex_lock(&scsi_sense_cache_mutex);
66 	if (!scsi_sense_cache) {
67 		scsi_sense_cache =
68 			kmem_cache_create_usercopy("scsi_sense_cache",
69 				SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
70 				0, SCSI_SENSE_BUFFERSIZE, NULL);
71 		if (!scsi_sense_cache)
72 			ret = -ENOMEM;
73 	}
74 	mutex_unlock(&scsi_sense_cache_mutex);
75 	return ret;
76 }
77 
78 static void
79 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
80 {
81 	struct Scsi_Host *host = cmd->device->host;
82 	struct scsi_device *device = cmd->device;
83 	struct scsi_target *starget = scsi_target(device);
84 
85 	/*
86 	 * Set the appropriate busy bit for the device/host.
87 	 *
88 	 * If the host/device isn't busy, assume that something actually
89 	 * completed, and that we should be able to queue a command now.
90 	 *
91 	 * Note that the prior mid-layer assumption that any host could
92 	 * always queue at least one command is now broken.  The mid-layer
93 	 * will implement a user specifiable stall (see
94 	 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
95 	 * if a command is requeued with no other commands outstanding
96 	 * either for the device or for the host.
97 	 */
98 	switch (reason) {
99 	case SCSI_MLQUEUE_HOST_BUSY:
100 		atomic_set(&host->host_blocked, host->max_host_blocked);
101 		break;
102 	case SCSI_MLQUEUE_DEVICE_BUSY:
103 	case SCSI_MLQUEUE_EH_RETRY:
104 		atomic_set(&device->device_blocked,
105 			   device->max_device_blocked);
106 		break;
107 	case SCSI_MLQUEUE_TARGET_BUSY:
108 		atomic_set(&starget->target_blocked,
109 			   starget->max_target_blocked);
110 		break;
111 	}
112 }
113 
114 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd, unsigned long msecs)
115 {
116 	struct request *rq = scsi_cmd_to_rq(cmd);
117 
118 	if (rq->rq_flags & RQF_DONTPREP) {
119 		rq->rq_flags &= ~RQF_DONTPREP;
120 		scsi_mq_uninit_cmd(cmd);
121 	} else {
122 		WARN_ON_ONCE(true);
123 	}
124 
125 	blk_mq_requeue_request(rq, false);
126 	if (!scsi_host_in_recovery(cmd->device->host))
127 		blk_mq_delay_kick_requeue_list(rq->q, msecs);
128 }
129 
130 /**
131  * __scsi_queue_insert - private queue insertion
132  * @cmd: The SCSI command being requeued
133  * @reason:  The reason for the requeue
134  * @unbusy: Whether the queue should be unbusied
135  *
136  * This is a private queue insertion.  The public interface
137  * scsi_queue_insert() always assumes the queue should be unbusied
138  * because it's always called before the completion.  This function is
139  * for a requeue after completion, which should only occur in this
140  * file.
141  */
142 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
143 {
144 	struct scsi_device *device = cmd->device;
145 
146 	SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
147 		"Inserting command %p into mlqueue\n", cmd));
148 
149 	scsi_set_blocked(cmd, reason);
150 
151 	/*
152 	 * Decrement the counters, since these commands are no longer
153 	 * active on the host/device.
154 	 */
155 	if (unbusy)
156 		scsi_device_unbusy(device, cmd);
157 
158 	/*
159 	 * Requeue this command.  It will go before all other commands
160 	 * that are already in the queue. Schedule requeue work under
161 	 * lock such that the kblockd_schedule_work() call happens
162 	 * before blk_mq_destroy_queue() finishes.
163 	 */
164 	cmd->result = 0;
165 
166 	blk_mq_requeue_request(scsi_cmd_to_rq(cmd),
167 			       !scsi_host_in_recovery(cmd->device->host));
168 }
169 
170 /**
171  * scsi_queue_insert - Reinsert a command in the queue.
172  * @cmd:    command that we are adding to queue.
173  * @reason: why we are inserting command to queue.
174  *
175  * We do this for one of two cases. Either the host is busy and it cannot accept
176  * any more commands for the time being, or the device returned QUEUE_FULL and
177  * can accept no more commands.
178  *
179  * Context: This could be called either from an interrupt context or a normal
180  * process context.
181  */
182 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
183 {
184 	__scsi_queue_insert(cmd, reason, true);
185 }
186 
187 void scsi_failures_reset_retries(struct scsi_failures *failures)
188 {
189 	struct scsi_failure *failure;
190 
191 	failures->total_retries = 0;
192 
193 	for (failure = failures->failure_definitions; failure->result;
194 	     failure++)
195 		failure->retries = 0;
196 }
197 EXPORT_SYMBOL_GPL(scsi_failures_reset_retries);
198 
199 /**
200  * scsi_check_passthrough - Determine if passthrough scsi_cmnd needs a retry.
201  * @scmd: scsi_cmnd to check.
202  * @failures: scsi_failures struct that lists failures to check for.
203  *
204  * Returns -EAGAIN if the caller should retry else 0.
205  */
206 static int scsi_check_passthrough(struct scsi_cmnd *scmd,
207 				  struct scsi_failures *failures)
208 {
209 	struct scsi_failure *failure;
210 	struct scsi_sense_hdr sshdr;
211 	enum sam_status status;
212 
213 	if (!failures)
214 		return 0;
215 
216 	for (failure = failures->failure_definitions; failure->result;
217 	     failure++) {
218 		if (failure->result == SCMD_FAILURE_RESULT_ANY)
219 			goto maybe_retry;
220 
221 		if (host_byte(scmd->result) &&
222 		    host_byte(scmd->result) == host_byte(failure->result))
223 			goto maybe_retry;
224 
225 		status = status_byte(scmd->result);
226 		if (!status)
227 			continue;
228 
229 		if (failure->result == SCMD_FAILURE_STAT_ANY &&
230 		    !scsi_status_is_good(scmd->result))
231 			goto maybe_retry;
232 
233 		if (status != status_byte(failure->result))
234 			continue;
235 
236 		if (status_byte(failure->result) != SAM_STAT_CHECK_CONDITION ||
237 		    failure->sense == SCMD_FAILURE_SENSE_ANY)
238 			goto maybe_retry;
239 
240 		if (!scsi_command_normalize_sense(scmd, &sshdr))
241 			return 0;
242 
243 		if (failure->sense != sshdr.sense_key)
244 			continue;
245 
246 		if (failure->asc == SCMD_FAILURE_ASC_ANY)
247 			goto maybe_retry;
248 
249 		if (failure->asc != sshdr.asc)
250 			continue;
251 
252 		if (failure->ascq == SCMD_FAILURE_ASCQ_ANY ||
253 		    failure->ascq == sshdr.ascq)
254 			goto maybe_retry;
255 	}
256 
257 	return 0;
258 
259 maybe_retry:
260 	if (failure->allowed) {
261 		if (failure->allowed == SCMD_FAILURE_NO_LIMIT ||
262 		    ++failure->retries <= failure->allowed)
263 			return -EAGAIN;
264 	} else {
265 		if (failures->total_allowed == SCMD_FAILURE_NO_LIMIT ||
266 		    ++failures->total_retries <= failures->total_allowed)
267 			return -EAGAIN;
268 	}
269 
270 	return 0;
271 }
272 
273 /**
274  * scsi_execute_cmd - insert request and wait for the result
275  * @sdev:	scsi_device
276  * @cmd:	scsi command
277  * @opf:	block layer request cmd_flags
278  * @buffer:	data buffer
279  * @bufflen:	len of buffer
280  * @timeout:	request timeout in HZ
281  * @ml_retries:	number of times SCSI midlayer will retry request
282  * @args:	Optional args. See struct definition for field descriptions
283  *
284  * Returns the scsi_cmnd result field if a command was executed, or a negative
285  * Linux error code if we didn't get that far.
286  */
287 int scsi_execute_cmd(struct scsi_device *sdev, const unsigned char *cmd,
288 		     blk_opf_t opf, void *buffer, unsigned int bufflen,
289 		     int timeout, int ml_retries,
290 		     const struct scsi_exec_args *args)
291 {
292 	static const struct scsi_exec_args default_args;
293 	struct request *req;
294 	struct scsi_cmnd *scmd;
295 	int ret;
296 
297 	if (!args)
298 		args = &default_args;
299 	else if (WARN_ON_ONCE(args->sense &&
300 			      args->sense_len != SCSI_SENSE_BUFFERSIZE))
301 		return -EINVAL;
302 
303 retry:
304 	req = scsi_alloc_request(sdev->request_queue, opf, args->req_flags);
305 	if (IS_ERR(req))
306 		return PTR_ERR(req);
307 
308 	if (bufflen) {
309 		ret = blk_rq_map_kern(sdev->request_queue, req,
310 				      buffer, bufflen, GFP_NOIO);
311 		if (ret)
312 			goto out;
313 	}
314 	scmd = blk_mq_rq_to_pdu(req);
315 	scmd->cmd_len = COMMAND_SIZE(cmd[0]);
316 	memcpy(scmd->cmnd, cmd, scmd->cmd_len);
317 	scmd->allowed = ml_retries;
318 	scmd->flags |= args->scmd_flags;
319 	req->timeout = timeout;
320 	req->rq_flags |= RQF_QUIET;
321 
322 	/*
323 	 * head injection *required* here otherwise quiesce won't work
324 	 */
325 	blk_execute_rq(req, true);
326 
327 	if (scsi_check_passthrough(scmd, args->failures) == -EAGAIN) {
328 		blk_mq_free_request(req);
329 		goto retry;
330 	}
331 
332 	/*
333 	 * Some devices (USB mass-storage in particular) may transfer
334 	 * garbage data together with a residue indicating that the data
335 	 * is invalid.  Prevent the garbage from being misinterpreted
336 	 * and prevent security leaks by zeroing out the excess data.
337 	 */
338 	if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
339 		memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);
340 
341 	if (args->resid)
342 		*args->resid = scmd->resid_len;
343 	if (args->sense)
344 		memcpy(args->sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
345 	if (args->sshdr)
346 		scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,
347 				     args->sshdr);
348 
349 	ret = scmd->result;
350  out:
351 	blk_mq_free_request(req);
352 
353 	return ret;
354 }
355 EXPORT_SYMBOL(scsi_execute_cmd);
356 
357 /*
358  * Wake up the error handler if necessary. Avoid as follows that the error
359  * handler is not woken up if host in-flight requests number ==
360  * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
361  * with an RCU read lock in this function to ensure that this function in
362  * its entirety either finishes before scsi_eh_scmd_add() increases the
363  * host_failed counter or that it notices the shost state change made by
364  * scsi_eh_scmd_add().
365  */
366 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
367 {
368 	unsigned long flags;
369 
370 	rcu_read_lock();
371 	__clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
372 	if (unlikely(scsi_host_in_recovery(shost))) {
373 		unsigned int busy = scsi_host_busy(shost);
374 
375 		spin_lock_irqsave(shost->host_lock, flags);
376 		if (shost->host_failed || shost->host_eh_scheduled)
377 			scsi_eh_wakeup(shost, busy);
378 		spin_unlock_irqrestore(shost->host_lock, flags);
379 	}
380 	rcu_read_unlock();
381 }
382 
383 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
384 {
385 	struct Scsi_Host *shost = sdev->host;
386 	struct scsi_target *starget = scsi_target(sdev);
387 
388 	scsi_dec_host_busy(shost, cmd);
389 
390 	if (starget->can_queue > 0)
391 		atomic_dec(&starget->target_busy);
392 
393 	sbitmap_put(&sdev->budget_map, cmd->budget_token);
394 	cmd->budget_token = -1;
395 }
396 
397 /*
398  * Kick the queue of SCSI device @sdev if @sdev != current_sdev. Called with
399  * interrupts disabled.
400  */
401 static void scsi_kick_sdev_queue(struct scsi_device *sdev, void *data)
402 {
403 	struct scsi_device *current_sdev = data;
404 
405 	if (sdev != current_sdev)
406 		blk_mq_run_hw_queues(sdev->request_queue, true);
407 }
408 
409 /*
410  * Called for single_lun devices on IO completion. Clear starget_sdev_user,
411  * and call blk_run_queue for all the scsi_devices on the target -
412  * including current_sdev first.
413  *
414  * Called with *no* scsi locks held.
415  */
416 static void scsi_single_lun_run(struct scsi_device *current_sdev)
417 {
418 	struct Scsi_Host *shost = current_sdev->host;
419 	struct scsi_target *starget = scsi_target(current_sdev);
420 	unsigned long flags;
421 
422 	spin_lock_irqsave(shost->host_lock, flags);
423 	starget->starget_sdev_user = NULL;
424 	spin_unlock_irqrestore(shost->host_lock, flags);
425 
426 	/*
427 	 * Call blk_run_queue for all LUNs on the target, starting with
428 	 * current_sdev. We race with others (to set starget_sdev_user),
429 	 * but in most cases, we will be first. Ideally, each LU on the
430 	 * target would get some limited time or requests on the target.
431 	 */
432 	blk_mq_run_hw_queues(current_sdev->request_queue,
433 			     shost->queuecommand_may_block);
434 
435 	spin_lock_irqsave(shost->host_lock, flags);
436 	if (!starget->starget_sdev_user)
437 		__starget_for_each_device(starget, current_sdev,
438 					  scsi_kick_sdev_queue);
439 	spin_unlock_irqrestore(shost->host_lock, flags);
440 }
441 
442 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
443 {
444 	if (scsi_device_busy(sdev) >= sdev->queue_depth)
445 		return true;
446 	if (atomic_read(&sdev->device_blocked) > 0)
447 		return true;
448 	return false;
449 }
450 
451 static inline bool scsi_target_is_busy(struct scsi_target *starget)
452 {
453 	if (starget->can_queue > 0) {
454 		if (atomic_read(&starget->target_busy) >= starget->can_queue)
455 			return true;
456 		if (atomic_read(&starget->target_blocked) > 0)
457 			return true;
458 	}
459 	return false;
460 }
461 
462 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
463 {
464 	if (atomic_read(&shost->host_blocked) > 0)
465 		return true;
466 	if (shost->host_self_blocked)
467 		return true;
468 	return false;
469 }
470 
471 static void scsi_starved_list_run(struct Scsi_Host *shost)
472 {
473 	LIST_HEAD(starved_list);
474 	struct scsi_device *sdev;
475 	unsigned long flags;
476 
477 	spin_lock_irqsave(shost->host_lock, flags);
478 	list_splice_init(&shost->starved_list, &starved_list);
479 
480 	while (!list_empty(&starved_list)) {
481 		struct request_queue *slq;
482 
483 		/*
484 		 * As long as shost is accepting commands and we have
485 		 * starved queues, call blk_run_queue. scsi_request_fn
486 		 * drops the queue_lock and can add us back to the
487 		 * starved_list.
488 		 *
489 		 * host_lock protects the starved_list and starved_entry.
490 		 * scsi_request_fn must get the host_lock before checking
491 		 * or modifying starved_list or starved_entry.
492 		 */
493 		if (scsi_host_is_busy(shost))
494 			break;
495 
496 		sdev = list_entry(starved_list.next,
497 				  struct scsi_device, starved_entry);
498 		list_del_init(&sdev->starved_entry);
499 		if (scsi_target_is_busy(scsi_target(sdev))) {
500 			list_move_tail(&sdev->starved_entry,
501 				       &shost->starved_list);
502 			continue;
503 		}
504 
505 		/*
506 		 * Once we drop the host lock, a racing scsi_remove_device()
507 		 * call may remove the sdev from the starved list and destroy
508 		 * it and the queue.  Mitigate by taking a reference to the
509 		 * queue and never touching the sdev again after we drop the
510 		 * host lock.  Note: if __scsi_remove_device() invokes
511 		 * blk_mq_destroy_queue() before the queue is run from this
512 		 * function then blk_run_queue() will return immediately since
513 		 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
514 		 */
515 		slq = sdev->request_queue;
516 		if (!blk_get_queue(slq))
517 			continue;
518 		spin_unlock_irqrestore(shost->host_lock, flags);
519 
520 		blk_mq_run_hw_queues(slq, false);
521 		blk_put_queue(slq);
522 
523 		spin_lock_irqsave(shost->host_lock, flags);
524 	}
525 	/* put any unprocessed entries back */
526 	list_splice(&starved_list, &shost->starved_list);
527 	spin_unlock_irqrestore(shost->host_lock, flags);
528 }
529 
530 /**
531  * scsi_run_queue - Select a proper request queue to serve next.
532  * @q:  last request's queue
533  *
534  * The previous command was completely finished, start a new one if possible.
535  */
536 static void scsi_run_queue(struct request_queue *q)
537 {
538 	struct scsi_device *sdev = q->queuedata;
539 
540 	if (scsi_target(sdev)->single_lun)
541 		scsi_single_lun_run(sdev);
542 	if (!list_empty(&sdev->host->starved_list))
543 		scsi_starved_list_run(sdev->host);
544 
545 	/* Note: blk_mq_kick_requeue_list() runs the queue asynchronously. */
546 	blk_mq_kick_requeue_list(q);
547 }
548 
549 void scsi_requeue_run_queue(struct work_struct *work)
550 {
551 	struct scsi_device *sdev;
552 	struct request_queue *q;
553 
554 	sdev = container_of(work, struct scsi_device, requeue_work);
555 	q = sdev->request_queue;
556 	scsi_run_queue(q);
557 }
558 
559 void scsi_run_host_queues(struct Scsi_Host *shost)
560 {
561 	struct scsi_device *sdev;
562 
563 	shost_for_each_device(sdev, shost)
564 		scsi_run_queue(sdev->request_queue);
565 }
566 
567 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
568 {
569 	if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
570 		struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
571 
572 		if (drv->uninit_command)
573 			drv->uninit_command(cmd);
574 	}
575 }
576 
577 void scsi_free_sgtables(struct scsi_cmnd *cmd)
578 {
579 	if (cmd->sdb.table.nents)
580 		sg_free_table_chained(&cmd->sdb.table,
581 				SCSI_INLINE_SG_CNT);
582 	if (scsi_prot_sg_count(cmd))
583 		sg_free_table_chained(&cmd->prot_sdb->table,
584 				SCSI_INLINE_PROT_SG_CNT);
585 }
586 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
587 
588 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
589 {
590 	scsi_free_sgtables(cmd);
591 	scsi_uninit_cmd(cmd);
592 }
593 
594 static void scsi_run_queue_async(struct scsi_device *sdev)
595 {
596 	if (scsi_host_in_recovery(sdev->host))
597 		return;
598 
599 	if (scsi_target(sdev)->single_lun ||
600 	    !list_empty(&sdev->host->starved_list)) {
601 		kblockd_schedule_work(&sdev->requeue_work);
602 	} else {
603 		/*
604 		 * smp_mb() present in sbitmap_queue_clear() or implied in
605 		 * .end_io is for ordering writing .device_busy in
606 		 * scsi_device_unbusy() and reading sdev->restarts.
607 		 */
608 		int old = atomic_read(&sdev->restarts);
609 
610 		/*
611 		 * ->restarts has to be kept as non-zero if new budget
612 		 *  contention occurs.
613 		 *
614 		 *  No need to run queue when either another re-run
615 		 *  queue wins in updating ->restarts or a new budget
616 		 *  contention occurs.
617 		 */
618 		if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
619 			blk_mq_run_hw_queues(sdev->request_queue, true);
620 	}
621 }
622 
623 /* Returns false when no more bytes to process, true if there are more */
624 static bool scsi_end_request(struct request *req, blk_status_t error,
625 		unsigned int bytes)
626 {
627 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
628 	struct scsi_device *sdev = cmd->device;
629 	struct request_queue *q = sdev->request_queue;
630 
631 	if (blk_update_request(req, error, bytes))
632 		return true;
633 
634 	// XXX:
635 	if (blk_queue_add_random(q))
636 		add_disk_randomness(req->q->disk);
637 
638 	WARN_ON_ONCE(!blk_rq_is_passthrough(req) &&
639 		     !(cmd->flags & SCMD_INITIALIZED));
640 	cmd->flags = 0;
641 
642 	/*
643 	 * Calling rcu_barrier() is not necessary here because the
644 	 * SCSI error handler guarantees that the function called by
645 	 * call_rcu() has been called before scsi_end_request() is
646 	 * called.
647 	 */
648 	destroy_rcu_head(&cmd->rcu);
649 
650 	/*
651 	 * In the MQ case the command gets freed by __blk_mq_end_request,
652 	 * so we have to do all cleanup that depends on it earlier.
653 	 *
654 	 * We also can't kick the queues from irq context, so we
655 	 * will have to defer it to a workqueue.
656 	 */
657 	scsi_mq_uninit_cmd(cmd);
658 
659 	/*
660 	 * queue is still alive, so grab the ref for preventing it
661 	 * from being cleaned up during running queue.
662 	 */
663 	percpu_ref_get(&q->q_usage_counter);
664 
665 	__blk_mq_end_request(req, error);
666 
667 	scsi_run_queue_async(sdev);
668 
669 	percpu_ref_put(&q->q_usage_counter);
670 	return false;
671 }
672 
673 /**
674  * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
675  * @result:	scsi error code
676  *
677  * Translate a SCSI result code into a blk_status_t value.
678  */
679 static blk_status_t scsi_result_to_blk_status(int result)
680 {
681 	/*
682 	 * Check the scsi-ml byte first in case we converted a host or status
683 	 * byte.
684 	 */
685 	switch (scsi_ml_byte(result)) {
686 	case SCSIML_STAT_OK:
687 		break;
688 	case SCSIML_STAT_RESV_CONFLICT:
689 		return BLK_STS_RESV_CONFLICT;
690 	case SCSIML_STAT_NOSPC:
691 		return BLK_STS_NOSPC;
692 	case SCSIML_STAT_MED_ERROR:
693 		return BLK_STS_MEDIUM;
694 	case SCSIML_STAT_TGT_FAILURE:
695 		return BLK_STS_TARGET;
696 	case SCSIML_STAT_DL_TIMEOUT:
697 		return BLK_STS_DURATION_LIMIT;
698 	}
699 
700 	switch (host_byte(result)) {
701 	case DID_OK:
702 		if (scsi_status_is_good(result))
703 			return BLK_STS_OK;
704 		return BLK_STS_IOERR;
705 	case DID_TRANSPORT_FAILFAST:
706 	case DID_TRANSPORT_MARGINAL:
707 		return BLK_STS_TRANSPORT;
708 	default:
709 		return BLK_STS_IOERR;
710 	}
711 }
712 
713 /**
714  * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
715  * @rq: request to examine
716  *
717  * Description:
718  *     A request could be merge of IOs which require different failure
719  *     handling.  This function determines the number of bytes which
720  *     can be failed from the beginning of the request without
721  *     crossing into area which need to be retried further.
722  *
723  * Return:
724  *     The number of bytes to fail.
725  */
726 static unsigned int scsi_rq_err_bytes(const struct request *rq)
727 {
728 	blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
729 	unsigned int bytes = 0;
730 	struct bio *bio;
731 
732 	if (!(rq->rq_flags & RQF_MIXED_MERGE))
733 		return blk_rq_bytes(rq);
734 
735 	/*
736 	 * Currently the only 'mixing' which can happen is between
737 	 * different fastfail types.  We can safely fail portions
738 	 * which have all the failfast bits that the first one has -
739 	 * the ones which are at least as eager to fail as the first
740 	 * one.
741 	 */
742 	for (bio = rq->bio; bio; bio = bio->bi_next) {
743 		if ((bio->bi_opf & ff) != ff)
744 			break;
745 		bytes += bio->bi_iter.bi_size;
746 	}
747 
748 	/* this could lead to infinite loop */
749 	BUG_ON(blk_rq_bytes(rq) && !bytes);
750 	return bytes;
751 }
752 
753 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
754 {
755 	struct request *req = scsi_cmd_to_rq(cmd);
756 	unsigned long wait_for;
757 
758 	if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
759 		return false;
760 
761 	wait_for = (cmd->allowed + 1) * req->timeout;
762 	if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
763 		scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
764 			    wait_for/HZ);
765 		return true;
766 	}
767 	return false;
768 }
769 
770 /*
771  * When ALUA transition state is returned, reprep the cmd to
772  * use the ALUA handler's transition timeout. Delay the reprep
773  * 1 sec to avoid aggressive retries of the target in that
774  * state.
775  */
776 #define ALUA_TRANSITION_REPREP_DELAY	1000
777 
778 /* Helper for scsi_io_completion() when special action required. */
779 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
780 {
781 	struct request *req = scsi_cmd_to_rq(cmd);
782 	int level = 0;
783 	enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
784 	      ACTION_RETRY, ACTION_DELAYED_RETRY} action;
785 	struct scsi_sense_hdr sshdr;
786 	bool sense_valid;
787 	bool sense_current = true;      /* false implies "deferred sense" */
788 	blk_status_t blk_stat;
789 
790 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
791 	if (sense_valid)
792 		sense_current = !scsi_sense_is_deferred(&sshdr);
793 
794 	blk_stat = scsi_result_to_blk_status(result);
795 
796 	if (host_byte(result) == DID_RESET) {
797 		/* Third party bus reset or reset for error recovery
798 		 * reasons.  Just retry the command and see what
799 		 * happens.
800 		 */
801 		action = ACTION_RETRY;
802 	} else if (sense_valid && sense_current) {
803 		switch (sshdr.sense_key) {
804 		case UNIT_ATTENTION:
805 			if (cmd->device->removable) {
806 				/* Detected disc change.  Set a bit
807 				 * and quietly refuse further access.
808 				 */
809 				cmd->device->changed = 1;
810 				action = ACTION_FAIL;
811 			} else {
812 				/* Must have been a power glitch, or a
813 				 * bus reset.  Could not have been a
814 				 * media change, so we just retry the
815 				 * command and see what happens.
816 				 */
817 				action = ACTION_RETRY;
818 			}
819 			break;
820 		case ILLEGAL_REQUEST:
821 			/* If we had an ILLEGAL REQUEST returned, then
822 			 * we may have performed an unsupported
823 			 * command.  The only thing this should be
824 			 * would be a ten byte read where only a six
825 			 * byte read was supported.  Also, on a system
826 			 * where READ CAPACITY failed, we may have
827 			 * read past the end of the disk.
828 			 */
829 			if ((cmd->device->use_10_for_rw &&
830 			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
831 			    (cmd->cmnd[0] == READ_10 ||
832 			     cmd->cmnd[0] == WRITE_10)) {
833 				/* This will issue a new 6-byte command. */
834 				cmd->device->use_10_for_rw = 0;
835 				action = ACTION_REPREP;
836 			} else if (sshdr.asc == 0x10) /* DIX */ {
837 				action = ACTION_FAIL;
838 				blk_stat = BLK_STS_PROTECTION;
839 			/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
840 			} else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
841 				action = ACTION_FAIL;
842 				blk_stat = BLK_STS_TARGET;
843 			} else
844 				action = ACTION_FAIL;
845 			break;
846 		case ABORTED_COMMAND:
847 			action = ACTION_FAIL;
848 			if (sshdr.asc == 0x10) /* DIF */
849 				blk_stat = BLK_STS_PROTECTION;
850 			break;
851 		case NOT_READY:
852 			/* If the device is in the process of becoming
853 			 * ready, or has a temporary blockage, retry.
854 			 */
855 			if (sshdr.asc == 0x04) {
856 				switch (sshdr.ascq) {
857 				case 0x01: /* becoming ready */
858 				case 0x04: /* format in progress */
859 				case 0x05: /* rebuild in progress */
860 				case 0x06: /* recalculation in progress */
861 				case 0x07: /* operation in progress */
862 				case 0x08: /* Long write in progress */
863 				case 0x09: /* self test in progress */
864 				case 0x11: /* notify (enable spinup) required */
865 				case 0x14: /* space allocation in progress */
866 				case 0x1a: /* start stop unit in progress */
867 				case 0x1b: /* sanitize in progress */
868 				case 0x1d: /* configuration in progress */
869 				case 0x24: /* depopulation in progress */
870 				case 0x25: /* depopulation restore in progress */
871 					action = ACTION_DELAYED_RETRY;
872 					break;
873 				case 0x0a: /* ALUA state transition */
874 					action = ACTION_DELAYED_REPREP;
875 					break;
876 				default:
877 					action = ACTION_FAIL;
878 					break;
879 				}
880 			} else
881 				action = ACTION_FAIL;
882 			break;
883 		case VOLUME_OVERFLOW:
884 			/* See SSC3rXX or current. */
885 			action = ACTION_FAIL;
886 			break;
887 		case DATA_PROTECT:
888 			action = ACTION_FAIL;
889 			if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
890 			    (sshdr.asc == 0x55 &&
891 			     (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
892 				/* Insufficient zone resources */
893 				blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
894 			}
895 			break;
896 		case COMPLETED:
897 			fallthrough;
898 		default:
899 			action = ACTION_FAIL;
900 			break;
901 		}
902 	} else
903 		action = ACTION_FAIL;
904 
905 	if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
906 		action = ACTION_FAIL;
907 
908 	switch (action) {
909 	case ACTION_FAIL:
910 		/* Give up and fail the remainder of the request */
911 		if (!(req->rq_flags & RQF_QUIET)) {
912 			static DEFINE_RATELIMIT_STATE(_rs,
913 					DEFAULT_RATELIMIT_INTERVAL,
914 					DEFAULT_RATELIMIT_BURST);
915 
916 			if (unlikely(scsi_logging_level))
917 				level =
918 				     SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
919 						    SCSI_LOG_MLCOMPLETE_BITS);
920 
921 			/*
922 			 * if logging is enabled the failure will be printed
923 			 * in scsi_log_completion(), so avoid duplicate messages
924 			 */
925 			if (!level && __ratelimit(&_rs)) {
926 				scsi_print_result(cmd, NULL, FAILED);
927 				if (sense_valid)
928 					scsi_print_sense(cmd);
929 				scsi_print_command(cmd);
930 			}
931 		}
932 		if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
933 			return;
934 		fallthrough;
935 	case ACTION_REPREP:
936 		scsi_mq_requeue_cmd(cmd, 0);
937 		break;
938 	case ACTION_DELAYED_REPREP:
939 		scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
940 		break;
941 	case ACTION_RETRY:
942 		/* Retry the same command immediately */
943 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
944 		break;
945 	case ACTION_DELAYED_RETRY:
946 		/* Retry the same command after a delay */
947 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
948 		break;
949 	}
950 }
951 
952 /*
953  * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
954  * new result that may suppress further error checking. Also modifies
955  * *blk_statp in some cases.
956  */
957 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
958 					blk_status_t *blk_statp)
959 {
960 	bool sense_valid;
961 	bool sense_current = true;	/* false implies "deferred sense" */
962 	struct request *req = scsi_cmd_to_rq(cmd);
963 	struct scsi_sense_hdr sshdr;
964 
965 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
966 	if (sense_valid)
967 		sense_current = !scsi_sense_is_deferred(&sshdr);
968 
969 	if (blk_rq_is_passthrough(req)) {
970 		if (sense_valid) {
971 			/*
972 			 * SG_IO wants current and deferred errors
973 			 */
974 			cmd->sense_len = min(8 + cmd->sense_buffer[7],
975 					     SCSI_SENSE_BUFFERSIZE);
976 		}
977 		if (sense_current)
978 			*blk_statp = scsi_result_to_blk_status(result);
979 	} else if (blk_rq_bytes(req) == 0 && sense_current) {
980 		/*
981 		 * Flush commands do not transfers any data, and thus cannot use
982 		 * good_bytes != blk_rq_bytes(req) as the signal for an error.
983 		 * This sets *blk_statp explicitly for the problem case.
984 		 */
985 		*blk_statp = scsi_result_to_blk_status(result);
986 	}
987 	/*
988 	 * Recovered errors need reporting, but they're always treated as
989 	 * success, so fiddle the result code here.  For passthrough requests
990 	 * we already took a copy of the original into sreq->result which
991 	 * is what gets returned to the user
992 	 */
993 	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
994 		bool do_print = true;
995 		/*
996 		 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
997 		 * skip print since caller wants ATA registers. Only occurs
998 		 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
999 		 */
1000 		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
1001 			do_print = false;
1002 		else if (req->rq_flags & RQF_QUIET)
1003 			do_print = false;
1004 		if (do_print)
1005 			scsi_print_sense(cmd);
1006 		result = 0;
1007 		/* for passthrough, *blk_statp may be set */
1008 		*blk_statp = BLK_STS_OK;
1009 	}
1010 	/*
1011 	 * Another corner case: the SCSI status byte is non-zero but 'good'.
1012 	 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
1013 	 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
1014 	 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
1015 	 * intermediate statuses (both obsolete in SAM-4) as good.
1016 	 */
1017 	if ((result & 0xff) && scsi_status_is_good(result)) {
1018 		result = 0;
1019 		*blk_statp = BLK_STS_OK;
1020 	}
1021 	return result;
1022 }
1023 
1024 /**
1025  * scsi_io_completion - Completion processing for SCSI commands.
1026  * @cmd:	command that is finished.
1027  * @good_bytes:	number of processed bytes.
1028  *
1029  * We will finish off the specified number of sectors. If we are done, the
1030  * command block will be released and the queue function will be goosed. If we
1031  * are not done then we have to figure out what to do next:
1032  *
1033  *   a) We can call scsi_mq_requeue_cmd().  The request will be
1034  *	unprepared and put back on the queue.  Then a new command will
1035  *	be created for it.  This should be used if we made forward
1036  *	progress, or if we want to switch from READ(10) to READ(6) for
1037  *	example.
1038  *
1039  *   b) We can call scsi_io_completion_action().  The request will be
1040  *	put back on the queue and retried using the same command as
1041  *	before, possibly after a delay.
1042  *
1043  *   c) We can call scsi_end_request() with blk_stat other than
1044  *	BLK_STS_OK, to fail the remainder of the request.
1045  */
1046 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
1047 {
1048 	int result = cmd->result;
1049 	struct request *req = scsi_cmd_to_rq(cmd);
1050 	blk_status_t blk_stat = BLK_STS_OK;
1051 
1052 	if (unlikely(result))	/* a nz result may or may not be an error */
1053 		result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
1054 
1055 	/*
1056 	 * Next deal with any sectors which we were able to correctly
1057 	 * handle.
1058 	 */
1059 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
1060 		"%u sectors total, %d bytes done.\n",
1061 		blk_rq_sectors(req), good_bytes));
1062 
1063 	/*
1064 	 * Failed, zero length commands always need to drop down
1065 	 * to retry code. Fast path should return in this block.
1066 	 */
1067 	if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
1068 		if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
1069 			return; /* no bytes remaining */
1070 	}
1071 
1072 	/* Kill remainder if no retries. */
1073 	if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
1074 		if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
1075 			WARN_ONCE(true,
1076 			    "Bytes remaining after failed, no-retry command");
1077 		return;
1078 	}
1079 
1080 	/*
1081 	 * If there had been no error, but we have leftover bytes in the
1082 	 * request just queue the command up again.
1083 	 */
1084 	if (likely(result == 0))
1085 		scsi_mq_requeue_cmd(cmd, 0);
1086 	else
1087 		scsi_io_completion_action(cmd, result);
1088 }
1089 
1090 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1091 		struct request *rq)
1092 {
1093 	return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1094 	       !op_is_write(req_op(rq)) &&
1095 	       sdev->host->hostt->dma_need_drain(rq);
1096 }
1097 
1098 /**
1099  * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1100  * @cmd: SCSI command data structure to initialize.
1101  *
1102  * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1103  * for @cmd.
1104  *
1105  * Returns:
1106  * * BLK_STS_OK       - on success
1107  * * BLK_STS_RESOURCE - if the failure is retryable
1108  * * BLK_STS_IOERR    - if the failure is fatal
1109  */
1110 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1111 {
1112 	struct scsi_device *sdev = cmd->device;
1113 	struct request *rq = scsi_cmd_to_rq(cmd);
1114 	unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1115 	struct scatterlist *last_sg = NULL;
1116 	blk_status_t ret;
1117 	bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1118 	int count;
1119 
1120 	if (WARN_ON_ONCE(!nr_segs))
1121 		return BLK_STS_IOERR;
1122 
1123 	/*
1124 	 * Make sure there is space for the drain.  The driver must adjust
1125 	 * max_hw_segments to be prepared for this.
1126 	 */
1127 	if (need_drain)
1128 		nr_segs++;
1129 
1130 	/*
1131 	 * If sg table allocation fails, requeue request later.
1132 	 */
1133 	if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1134 			cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1135 		return BLK_STS_RESOURCE;
1136 
1137 	/*
1138 	 * Next, walk the list, and fill in the addresses and sizes of
1139 	 * each segment.
1140 	 */
1141 	count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1142 
1143 	if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1144 		unsigned int pad_len =
1145 			(rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1146 
1147 		last_sg->length += pad_len;
1148 		cmd->extra_len += pad_len;
1149 	}
1150 
1151 	if (need_drain) {
1152 		sg_unmark_end(last_sg);
1153 		last_sg = sg_next(last_sg);
1154 		sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1155 		sg_mark_end(last_sg);
1156 
1157 		cmd->extra_len += sdev->dma_drain_len;
1158 		count++;
1159 	}
1160 
1161 	BUG_ON(count > cmd->sdb.table.nents);
1162 	cmd->sdb.table.nents = count;
1163 	cmd->sdb.length = blk_rq_payload_bytes(rq);
1164 
1165 	if (blk_integrity_rq(rq)) {
1166 		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1167 		int ivecs;
1168 
1169 		if (WARN_ON_ONCE(!prot_sdb)) {
1170 			/*
1171 			 * This can happen if someone (e.g. multipath)
1172 			 * queues a command to a device on an adapter
1173 			 * that does not support DIX.
1174 			 */
1175 			ret = BLK_STS_IOERR;
1176 			goto out_free_sgtables;
1177 		}
1178 
1179 		ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1180 
1181 		if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1182 				prot_sdb->table.sgl,
1183 				SCSI_INLINE_PROT_SG_CNT)) {
1184 			ret = BLK_STS_RESOURCE;
1185 			goto out_free_sgtables;
1186 		}
1187 
1188 		count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1189 						prot_sdb->table.sgl);
1190 		BUG_ON(count > ivecs);
1191 		BUG_ON(count > queue_max_integrity_segments(rq->q));
1192 
1193 		cmd->prot_sdb = prot_sdb;
1194 		cmd->prot_sdb->table.nents = count;
1195 	}
1196 
1197 	return BLK_STS_OK;
1198 out_free_sgtables:
1199 	scsi_free_sgtables(cmd);
1200 	return ret;
1201 }
1202 EXPORT_SYMBOL(scsi_alloc_sgtables);
1203 
1204 /**
1205  * scsi_initialize_rq - initialize struct scsi_cmnd partially
1206  * @rq: Request associated with the SCSI command to be initialized.
1207  *
1208  * This function initializes the members of struct scsi_cmnd that must be
1209  * initialized before request processing starts and that won't be
1210  * reinitialized if a SCSI command is requeued.
1211  */
1212 static void scsi_initialize_rq(struct request *rq)
1213 {
1214 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1215 
1216 	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1217 	cmd->cmd_len = MAX_COMMAND_SIZE;
1218 	cmd->sense_len = 0;
1219 	init_rcu_head(&cmd->rcu);
1220 	cmd->jiffies_at_alloc = jiffies;
1221 	cmd->retries = 0;
1222 }
1223 
1224 struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1225 				   blk_mq_req_flags_t flags)
1226 {
1227 	struct request *rq;
1228 
1229 	rq = blk_mq_alloc_request(q, opf, flags);
1230 	if (!IS_ERR(rq))
1231 		scsi_initialize_rq(rq);
1232 	return rq;
1233 }
1234 EXPORT_SYMBOL_GPL(scsi_alloc_request);
1235 
1236 /*
1237  * Only called when the request isn't completed by SCSI, and not freed by
1238  * SCSI
1239  */
1240 static void scsi_cleanup_rq(struct request *rq)
1241 {
1242 	if (rq->rq_flags & RQF_DONTPREP) {
1243 		scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1244 		rq->rq_flags &= ~RQF_DONTPREP;
1245 	}
1246 }
1247 
1248 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1249 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1250 {
1251 	struct request *rq = scsi_cmd_to_rq(cmd);
1252 
1253 	if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1254 		cmd->flags |= SCMD_INITIALIZED;
1255 		scsi_initialize_rq(rq);
1256 	}
1257 
1258 	cmd->device = dev;
1259 	INIT_LIST_HEAD(&cmd->eh_entry);
1260 	INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1261 }
1262 
1263 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1264 		struct request *req)
1265 {
1266 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1267 
1268 	/*
1269 	 * Passthrough requests may transfer data, in which case they must
1270 	 * a bio attached to them.  Or they might contain a SCSI command
1271 	 * that does not transfer data, in which case they may optionally
1272 	 * submit a request without an attached bio.
1273 	 */
1274 	if (req->bio) {
1275 		blk_status_t ret = scsi_alloc_sgtables(cmd);
1276 		if (unlikely(ret != BLK_STS_OK))
1277 			return ret;
1278 	} else {
1279 		BUG_ON(blk_rq_bytes(req));
1280 
1281 		memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1282 	}
1283 
1284 	cmd->transfersize = blk_rq_bytes(req);
1285 	return BLK_STS_OK;
1286 }
1287 
1288 static blk_status_t
1289 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1290 {
1291 	switch (sdev->sdev_state) {
1292 	case SDEV_CREATED:
1293 		return BLK_STS_OK;
1294 	case SDEV_OFFLINE:
1295 	case SDEV_TRANSPORT_OFFLINE:
1296 		/*
1297 		 * If the device is offline we refuse to process any
1298 		 * commands.  The device must be brought online
1299 		 * before trying any recovery commands.
1300 		 */
1301 		if (!sdev->offline_already) {
1302 			sdev->offline_already = true;
1303 			sdev_printk(KERN_ERR, sdev,
1304 				    "rejecting I/O to offline device\n");
1305 		}
1306 		return BLK_STS_IOERR;
1307 	case SDEV_DEL:
1308 		/*
1309 		 * If the device is fully deleted, we refuse to
1310 		 * process any commands as well.
1311 		 */
1312 		sdev_printk(KERN_ERR, sdev,
1313 			    "rejecting I/O to dead device\n");
1314 		return BLK_STS_IOERR;
1315 	case SDEV_BLOCK:
1316 	case SDEV_CREATED_BLOCK:
1317 		return BLK_STS_RESOURCE;
1318 	case SDEV_QUIESCE:
1319 		/*
1320 		 * If the device is blocked we only accept power management
1321 		 * commands.
1322 		 */
1323 		if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1324 			return BLK_STS_RESOURCE;
1325 		return BLK_STS_OK;
1326 	default:
1327 		/*
1328 		 * For any other not fully online state we only allow
1329 		 * power management commands.
1330 		 */
1331 		if (req && !(req->rq_flags & RQF_PM))
1332 			return BLK_STS_OFFLINE;
1333 		return BLK_STS_OK;
1334 	}
1335 }
1336 
1337 /*
1338  * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1339  * and return the token else return -1.
1340  */
1341 static inline int scsi_dev_queue_ready(struct request_queue *q,
1342 				  struct scsi_device *sdev)
1343 {
1344 	int token;
1345 
1346 	token = sbitmap_get(&sdev->budget_map);
1347 	if (token < 0)
1348 		return -1;
1349 
1350 	if (!atomic_read(&sdev->device_blocked))
1351 		return token;
1352 
1353 	/*
1354 	 * Only unblock if no other commands are pending and
1355 	 * if device_blocked has decreased to zero
1356 	 */
1357 	if (scsi_device_busy(sdev) > 1 ||
1358 	    atomic_dec_return(&sdev->device_blocked) > 0) {
1359 		sbitmap_put(&sdev->budget_map, token);
1360 		return -1;
1361 	}
1362 
1363 	SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1364 			 "unblocking device at zero depth\n"));
1365 
1366 	return token;
1367 }
1368 
1369 /*
1370  * scsi_target_queue_ready: checks if there we can send commands to target
1371  * @sdev: scsi device on starget to check.
1372  */
1373 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1374 					   struct scsi_device *sdev)
1375 {
1376 	struct scsi_target *starget = scsi_target(sdev);
1377 	unsigned int busy;
1378 
1379 	if (starget->single_lun) {
1380 		spin_lock_irq(shost->host_lock);
1381 		if (starget->starget_sdev_user &&
1382 		    starget->starget_sdev_user != sdev) {
1383 			spin_unlock_irq(shost->host_lock);
1384 			return 0;
1385 		}
1386 		starget->starget_sdev_user = sdev;
1387 		spin_unlock_irq(shost->host_lock);
1388 	}
1389 
1390 	if (starget->can_queue <= 0)
1391 		return 1;
1392 
1393 	busy = atomic_inc_return(&starget->target_busy) - 1;
1394 	if (atomic_read(&starget->target_blocked) > 0) {
1395 		if (busy)
1396 			goto starved;
1397 
1398 		/*
1399 		 * unblock after target_blocked iterates to zero
1400 		 */
1401 		if (atomic_dec_return(&starget->target_blocked) > 0)
1402 			goto out_dec;
1403 
1404 		SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1405 				 "unblocking target at zero depth\n"));
1406 	}
1407 
1408 	if (busy >= starget->can_queue)
1409 		goto starved;
1410 
1411 	return 1;
1412 
1413 starved:
1414 	spin_lock_irq(shost->host_lock);
1415 	list_move_tail(&sdev->starved_entry, &shost->starved_list);
1416 	spin_unlock_irq(shost->host_lock);
1417 out_dec:
1418 	if (starget->can_queue > 0)
1419 		atomic_dec(&starget->target_busy);
1420 	return 0;
1421 }
1422 
1423 /*
1424  * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1425  * return 0. We must end up running the queue again whenever 0 is
1426  * returned, else IO can hang.
1427  */
1428 static inline int scsi_host_queue_ready(struct request_queue *q,
1429 				   struct Scsi_Host *shost,
1430 				   struct scsi_device *sdev,
1431 				   struct scsi_cmnd *cmd)
1432 {
1433 	if (atomic_read(&shost->host_blocked) > 0) {
1434 		if (scsi_host_busy(shost) > 0)
1435 			goto starved;
1436 
1437 		/*
1438 		 * unblock after host_blocked iterates to zero
1439 		 */
1440 		if (atomic_dec_return(&shost->host_blocked) > 0)
1441 			goto out_dec;
1442 
1443 		SCSI_LOG_MLQUEUE(3,
1444 			shost_printk(KERN_INFO, shost,
1445 				     "unblocking host at zero depth\n"));
1446 	}
1447 
1448 	if (shost->host_self_blocked)
1449 		goto starved;
1450 
1451 	/* We're OK to process the command, so we can't be starved */
1452 	if (!list_empty(&sdev->starved_entry)) {
1453 		spin_lock_irq(shost->host_lock);
1454 		if (!list_empty(&sdev->starved_entry))
1455 			list_del_init(&sdev->starved_entry);
1456 		spin_unlock_irq(shost->host_lock);
1457 	}
1458 
1459 	__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1460 
1461 	return 1;
1462 
1463 starved:
1464 	spin_lock_irq(shost->host_lock);
1465 	if (list_empty(&sdev->starved_entry))
1466 		list_add_tail(&sdev->starved_entry, &shost->starved_list);
1467 	spin_unlock_irq(shost->host_lock);
1468 out_dec:
1469 	scsi_dec_host_busy(shost, cmd);
1470 	return 0;
1471 }
1472 
1473 /*
1474  * Busy state exporting function for request stacking drivers.
1475  *
1476  * For efficiency, no lock is taken to check the busy state of
1477  * shost/starget/sdev, since the returned value is not guaranteed and
1478  * may be changed after request stacking drivers call the function,
1479  * regardless of taking lock or not.
1480  *
1481  * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1482  * needs to return 'not busy'. Otherwise, request stacking drivers
1483  * may hold requests forever.
1484  */
1485 static bool scsi_mq_lld_busy(struct request_queue *q)
1486 {
1487 	struct scsi_device *sdev = q->queuedata;
1488 	struct Scsi_Host *shost;
1489 
1490 	if (blk_queue_dying(q))
1491 		return false;
1492 
1493 	shost = sdev->host;
1494 
1495 	/*
1496 	 * Ignore host/starget busy state.
1497 	 * Since block layer does not have a concept of fairness across
1498 	 * multiple queues, congestion of host/starget needs to be handled
1499 	 * in SCSI layer.
1500 	 */
1501 	if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1502 		return true;
1503 
1504 	return false;
1505 }
1506 
1507 /*
1508  * Block layer request completion callback. May be called from interrupt
1509  * context.
1510  */
1511 static void scsi_complete(struct request *rq)
1512 {
1513 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1514 	enum scsi_disposition disposition;
1515 
1516 	INIT_LIST_HEAD(&cmd->eh_entry);
1517 
1518 	atomic_inc(&cmd->device->iodone_cnt);
1519 	if (cmd->result)
1520 		atomic_inc(&cmd->device->ioerr_cnt);
1521 
1522 	disposition = scsi_decide_disposition(cmd);
1523 	if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1524 		disposition = SUCCESS;
1525 
1526 	scsi_log_completion(cmd, disposition);
1527 
1528 	switch (disposition) {
1529 	case SUCCESS:
1530 		scsi_finish_command(cmd);
1531 		break;
1532 	case NEEDS_RETRY:
1533 		scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1534 		break;
1535 	case ADD_TO_MLQUEUE:
1536 		scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1537 		break;
1538 	default:
1539 		scsi_eh_scmd_add(cmd);
1540 		break;
1541 	}
1542 }
1543 
1544 /**
1545  * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1546  * @cmd: command block we are dispatching.
1547  *
1548  * Return: nonzero return request was rejected and device's queue needs to be
1549  * plugged.
1550  */
1551 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1552 {
1553 	struct Scsi_Host *host = cmd->device->host;
1554 	int rtn = 0;
1555 
1556 	atomic_inc(&cmd->device->iorequest_cnt);
1557 
1558 	/* check if the device is still usable */
1559 	if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1560 		/* in SDEV_DEL we error all commands. DID_NO_CONNECT
1561 		 * returns an immediate error upwards, and signals
1562 		 * that the device is no longer present */
1563 		cmd->result = DID_NO_CONNECT << 16;
1564 		goto done;
1565 	}
1566 
1567 	/* Check to see if the scsi lld made this device blocked. */
1568 	if (unlikely(scsi_device_blocked(cmd->device))) {
1569 		/*
1570 		 * in blocked state, the command is just put back on
1571 		 * the device queue.  The suspend state has already
1572 		 * blocked the queue so future requests should not
1573 		 * occur until the device transitions out of the
1574 		 * suspend state.
1575 		 */
1576 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1577 			"queuecommand : device blocked\n"));
1578 		atomic_dec(&cmd->device->iorequest_cnt);
1579 		return SCSI_MLQUEUE_DEVICE_BUSY;
1580 	}
1581 
1582 	/* Store the LUN value in cmnd, if needed. */
1583 	if (cmd->device->lun_in_cdb)
1584 		cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1585 			       (cmd->device->lun << 5 & 0xe0);
1586 
1587 	scsi_log_send(cmd);
1588 
1589 	/*
1590 	 * Before we queue this command, check if the command
1591 	 * length exceeds what the host adapter can handle.
1592 	 */
1593 	if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1594 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1595 			       "queuecommand : command too long. "
1596 			       "cdb_size=%d host->max_cmd_len=%d\n",
1597 			       cmd->cmd_len, cmd->device->host->max_cmd_len));
1598 		cmd->result = (DID_ABORT << 16);
1599 		goto done;
1600 	}
1601 
1602 	if (unlikely(host->shost_state == SHOST_DEL)) {
1603 		cmd->result = (DID_NO_CONNECT << 16);
1604 		goto done;
1605 
1606 	}
1607 
1608 	trace_scsi_dispatch_cmd_start(cmd);
1609 	rtn = host->hostt->queuecommand(host, cmd);
1610 	if (rtn) {
1611 		atomic_dec(&cmd->device->iorequest_cnt);
1612 		trace_scsi_dispatch_cmd_error(cmd, rtn);
1613 		if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1614 		    rtn != SCSI_MLQUEUE_TARGET_BUSY)
1615 			rtn = SCSI_MLQUEUE_HOST_BUSY;
1616 
1617 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1618 			"queuecommand : request rejected\n"));
1619 	}
1620 
1621 	return rtn;
1622  done:
1623 	scsi_done(cmd);
1624 	return 0;
1625 }
1626 
1627 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1628 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1629 {
1630 	return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1631 		sizeof(struct scatterlist);
1632 }
1633 
1634 static blk_status_t scsi_prepare_cmd(struct request *req)
1635 {
1636 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1637 	struct scsi_device *sdev = req->q->queuedata;
1638 	struct Scsi_Host *shost = sdev->host;
1639 	bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1640 	struct scatterlist *sg;
1641 
1642 	scsi_init_command(sdev, cmd);
1643 
1644 	cmd->eh_eflags = 0;
1645 	cmd->prot_type = 0;
1646 	cmd->prot_flags = 0;
1647 	cmd->submitter = 0;
1648 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1649 	cmd->underflow = 0;
1650 	cmd->transfersize = 0;
1651 	cmd->host_scribble = NULL;
1652 	cmd->result = 0;
1653 	cmd->extra_len = 0;
1654 	cmd->state = 0;
1655 	if (in_flight)
1656 		__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1657 
1658 	/*
1659 	 * Only clear the driver-private command data if the LLD does not supply
1660 	 * a function to initialize that data.
1661 	 */
1662 	if (!shost->hostt->init_cmd_priv)
1663 		memset(cmd + 1, 0, shost->hostt->cmd_size);
1664 
1665 	cmd->prot_op = SCSI_PROT_NORMAL;
1666 	if (blk_rq_bytes(req))
1667 		cmd->sc_data_direction = rq_dma_dir(req);
1668 	else
1669 		cmd->sc_data_direction = DMA_NONE;
1670 
1671 	sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1672 	cmd->sdb.table.sgl = sg;
1673 
1674 	if (scsi_host_get_prot(shost)) {
1675 		memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1676 
1677 		cmd->prot_sdb->table.sgl =
1678 			(struct scatterlist *)(cmd->prot_sdb + 1);
1679 	}
1680 
1681 	/*
1682 	 * Special handling for passthrough commands, which don't go to the ULP
1683 	 * at all:
1684 	 */
1685 	if (blk_rq_is_passthrough(req))
1686 		return scsi_setup_scsi_cmnd(sdev, req);
1687 
1688 	if (sdev->handler && sdev->handler->prep_fn) {
1689 		blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1690 
1691 		if (ret != BLK_STS_OK)
1692 			return ret;
1693 	}
1694 
1695 	/* Usually overridden by the ULP */
1696 	cmd->allowed = 0;
1697 	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1698 	return scsi_cmd_to_driver(cmd)->init_command(cmd);
1699 }
1700 
1701 static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1702 {
1703 	struct request *req = scsi_cmd_to_rq(cmd);
1704 
1705 	switch (cmd->submitter) {
1706 	case SUBMITTED_BY_BLOCK_LAYER:
1707 		break;
1708 	case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1709 		return scsi_eh_done(cmd);
1710 	case SUBMITTED_BY_SCSI_RESET_IOCTL:
1711 		return;
1712 	}
1713 
1714 	if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1715 		return;
1716 	if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1717 		return;
1718 	trace_scsi_dispatch_cmd_done(cmd);
1719 
1720 	if (complete_directly)
1721 		blk_mq_complete_request_direct(req, scsi_complete);
1722 	else
1723 		blk_mq_complete_request(req);
1724 }
1725 
1726 void scsi_done(struct scsi_cmnd *cmd)
1727 {
1728 	scsi_done_internal(cmd, false);
1729 }
1730 EXPORT_SYMBOL(scsi_done);
1731 
1732 void scsi_done_direct(struct scsi_cmnd *cmd)
1733 {
1734 	scsi_done_internal(cmd, true);
1735 }
1736 EXPORT_SYMBOL(scsi_done_direct);
1737 
1738 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1739 {
1740 	struct scsi_device *sdev = q->queuedata;
1741 
1742 	sbitmap_put(&sdev->budget_map, budget_token);
1743 }
1744 
1745 /*
1746  * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1747  * not change behaviour from the previous unplug mechanism, experimentation
1748  * may prove this needs changing.
1749  */
1750 #define SCSI_QUEUE_DELAY 3
1751 
1752 static int scsi_mq_get_budget(struct request_queue *q)
1753 {
1754 	struct scsi_device *sdev = q->queuedata;
1755 	int token = scsi_dev_queue_ready(q, sdev);
1756 
1757 	if (token >= 0)
1758 		return token;
1759 
1760 	atomic_inc(&sdev->restarts);
1761 
1762 	/*
1763 	 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1764 	 * .restarts must be incremented before .device_busy is read because the
1765 	 * code in scsi_run_queue_async() depends on the order of these operations.
1766 	 */
1767 	smp_mb__after_atomic();
1768 
1769 	/*
1770 	 * If all in-flight requests originated from this LUN are completed
1771 	 * before reading .device_busy, sdev->device_busy will be observed as
1772 	 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1773 	 * soon. Otherwise, completion of one of these requests will observe
1774 	 * the .restarts flag, and the request queue will be run for handling
1775 	 * this request, see scsi_end_request().
1776 	 */
1777 	if (unlikely(scsi_device_busy(sdev) == 0 &&
1778 				!scsi_device_blocked(sdev)))
1779 		blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1780 	return -1;
1781 }
1782 
1783 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1784 {
1785 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1786 
1787 	cmd->budget_token = token;
1788 }
1789 
1790 static int scsi_mq_get_rq_budget_token(struct request *req)
1791 {
1792 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1793 
1794 	return cmd->budget_token;
1795 }
1796 
1797 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1798 			 const struct blk_mq_queue_data *bd)
1799 {
1800 	struct request *req = bd->rq;
1801 	struct request_queue *q = req->q;
1802 	struct scsi_device *sdev = q->queuedata;
1803 	struct Scsi_Host *shost = sdev->host;
1804 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1805 	blk_status_t ret;
1806 	int reason;
1807 
1808 	WARN_ON_ONCE(cmd->budget_token < 0);
1809 
1810 	/*
1811 	 * If the device is not in running state we will reject some or all
1812 	 * commands.
1813 	 */
1814 	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1815 		ret = scsi_device_state_check(sdev, req);
1816 		if (ret != BLK_STS_OK)
1817 			goto out_put_budget;
1818 	}
1819 
1820 	ret = BLK_STS_RESOURCE;
1821 	if (!scsi_target_queue_ready(shost, sdev))
1822 		goto out_put_budget;
1823 	if (unlikely(scsi_host_in_recovery(shost))) {
1824 		if (cmd->flags & SCMD_FAIL_IF_RECOVERING)
1825 			ret = BLK_STS_OFFLINE;
1826 		goto out_dec_target_busy;
1827 	}
1828 	if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1829 		goto out_dec_target_busy;
1830 
1831 	if (!(req->rq_flags & RQF_DONTPREP)) {
1832 		ret = scsi_prepare_cmd(req);
1833 		if (ret != BLK_STS_OK)
1834 			goto out_dec_host_busy;
1835 		req->rq_flags |= RQF_DONTPREP;
1836 	} else {
1837 		clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1838 	}
1839 
1840 	cmd->flags &= SCMD_PRESERVED_FLAGS;
1841 	if (sdev->simple_tags)
1842 		cmd->flags |= SCMD_TAGGED;
1843 	if (bd->last)
1844 		cmd->flags |= SCMD_LAST;
1845 
1846 	scsi_set_resid(cmd, 0);
1847 	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1848 	cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1849 
1850 	blk_mq_start_request(req);
1851 	reason = scsi_dispatch_cmd(cmd);
1852 	if (reason) {
1853 		scsi_set_blocked(cmd, reason);
1854 		ret = BLK_STS_RESOURCE;
1855 		goto out_dec_host_busy;
1856 	}
1857 
1858 	return BLK_STS_OK;
1859 
1860 out_dec_host_busy:
1861 	scsi_dec_host_busy(shost, cmd);
1862 out_dec_target_busy:
1863 	if (scsi_target(sdev)->can_queue > 0)
1864 		atomic_dec(&scsi_target(sdev)->target_busy);
1865 out_put_budget:
1866 	scsi_mq_put_budget(q, cmd->budget_token);
1867 	cmd->budget_token = -1;
1868 	switch (ret) {
1869 	case BLK_STS_OK:
1870 		break;
1871 	case BLK_STS_RESOURCE:
1872 		if (scsi_device_blocked(sdev))
1873 			ret = BLK_STS_DEV_RESOURCE;
1874 		break;
1875 	case BLK_STS_AGAIN:
1876 		cmd->result = DID_BUS_BUSY << 16;
1877 		if (req->rq_flags & RQF_DONTPREP)
1878 			scsi_mq_uninit_cmd(cmd);
1879 		break;
1880 	default:
1881 		if (unlikely(!scsi_device_online(sdev)))
1882 			cmd->result = DID_NO_CONNECT << 16;
1883 		else
1884 			cmd->result = DID_ERROR << 16;
1885 		/*
1886 		 * Make sure to release all allocated resources when
1887 		 * we hit an error, as we will never see this command
1888 		 * again.
1889 		 */
1890 		if (req->rq_flags & RQF_DONTPREP)
1891 			scsi_mq_uninit_cmd(cmd);
1892 		scsi_run_queue_async(sdev);
1893 		break;
1894 	}
1895 	return ret;
1896 }
1897 
1898 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1899 				unsigned int hctx_idx, unsigned int numa_node)
1900 {
1901 	struct Scsi_Host *shost = set->driver_data;
1902 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1903 	struct scatterlist *sg;
1904 	int ret = 0;
1905 
1906 	cmd->sense_buffer =
1907 		kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1908 	if (!cmd->sense_buffer)
1909 		return -ENOMEM;
1910 
1911 	if (scsi_host_get_prot(shost)) {
1912 		sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1913 			shost->hostt->cmd_size;
1914 		cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1915 	}
1916 
1917 	if (shost->hostt->init_cmd_priv) {
1918 		ret = shost->hostt->init_cmd_priv(shost, cmd);
1919 		if (ret < 0)
1920 			kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1921 	}
1922 
1923 	return ret;
1924 }
1925 
1926 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1927 				 unsigned int hctx_idx)
1928 {
1929 	struct Scsi_Host *shost = set->driver_data;
1930 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1931 
1932 	if (shost->hostt->exit_cmd_priv)
1933 		shost->hostt->exit_cmd_priv(shost, cmd);
1934 	kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1935 }
1936 
1937 
1938 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1939 {
1940 	struct Scsi_Host *shost = hctx->driver_data;
1941 
1942 	if (shost->hostt->mq_poll)
1943 		return shost->hostt->mq_poll(shost, hctx->queue_num);
1944 
1945 	return 0;
1946 }
1947 
1948 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1949 			  unsigned int hctx_idx)
1950 {
1951 	struct Scsi_Host *shost = data;
1952 
1953 	hctx->driver_data = shost;
1954 	return 0;
1955 }
1956 
1957 static void scsi_map_queues(struct blk_mq_tag_set *set)
1958 {
1959 	struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1960 
1961 	if (shost->hostt->map_queues)
1962 		return shost->hostt->map_queues(shost);
1963 	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1964 }
1965 
1966 void scsi_init_limits(struct Scsi_Host *shost, struct queue_limits *lim)
1967 {
1968 	struct device *dev = shost->dma_dev;
1969 
1970 	memset(lim, 0, sizeof(*lim));
1971 	lim->max_segments =
1972 		min_t(unsigned short, shost->sg_tablesize, SG_MAX_SEGMENTS);
1973 
1974 	if (scsi_host_prot_dma(shost)) {
1975 		shost->sg_prot_tablesize =
1976 			min_not_zero(shost->sg_prot_tablesize,
1977 				     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1978 		BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1979 		lim->max_integrity_segments = shost->sg_prot_tablesize;
1980 	}
1981 
1982 	lim->max_hw_sectors = shost->max_sectors;
1983 	lim->seg_boundary_mask = shost->dma_boundary;
1984 	lim->max_segment_size = shost->max_segment_size;
1985 	lim->virt_boundary_mask = shost->virt_boundary_mask;
1986 	lim->dma_alignment = max_t(unsigned int,
1987 		shost->dma_alignment, dma_get_cache_alignment() - 1);
1988 
1989 	if (shost->no_highmem)
1990 		lim->bounce = BLK_BOUNCE_HIGH;
1991 
1992 	dma_set_seg_boundary(dev, shost->dma_boundary);
1993 	dma_set_max_seg_size(dev, shost->max_segment_size);
1994 }
1995 EXPORT_SYMBOL_GPL(scsi_init_limits);
1996 
1997 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1998 	.get_budget	= scsi_mq_get_budget,
1999 	.put_budget	= scsi_mq_put_budget,
2000 	.queue_rq	= scsi_queue_rq,
2001 	.complete	= scsi_complete,
2002 	.timeout	= scsi_timeout,
2003 #ifdef CONFIG_BLK_DEBUG_FS
2004 	.show_rq	= scsi_show_rq,
2005 #endif
2006 	.init_request	= scsi_mq_init_request,
2007 	.exit_request	= scsi_mq_exit_request,
2008 	.cleanup_rq	= scsi_cleanup_rq,
2009 	.busy		= scsi_mq_lld_busy,
2010 	.map_queues	= scsi_map_queues,
2011 	.init_hctx	= scsi_init_hctx,
2012 	.poll		= scsi_mq_poll,
2013 	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
2014 	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
2015 };
2016 
2017 
2018 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
2019 {
2020 	struct Scsi_Host *shost = hctx->driver_data;
2021 
2022 	shost->hostt->commit_rqs(shost, hctx->queue_num);
2023 }
2024 
2025 static const struct blk_mq_ops scsi_mq_ops = {
2026 	.get_budget	= scsi_mq_get_budget,
2027 	.put_budget	= scsi_mq_put_budget,
2028 	.queue_rq	= scsi_queue_rq,
2029 	.commit_rqs	= scsi_commit_rqs,
2030 	.complete	= scsi_complete,
2031 	.timeout	= scsi_timeout,
2032 #ifdef CONFIG_BLK_DEBUG_FS
2033 	.show_rq	= scsi_show_rq,
2034 #endif
2035 	.init_request	= scsi_mq_init_request,
2036 	.exit_request	= scsi_mq_exit_request,
2037 	.cleanup_rq	= scsi_cleanup_rq,
2038 	.busy		= scsi_mq_lld_busy,
2039 	.map_queues	= scsi_map_queues,
2040 	.init_hctx	= scsi_init_hctx,
2041 	.poll		= scsi_mq_poll,
2042 	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
2043 	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
2044 };
2045 
2046 int scsi_mq_setup_tags(struct Scsi_Host *shost)
2047 {
2048 	unsigned int cmd_size, sgl_size;
2049 	struct blk_mq_tag_set *tag_set = &shost->tag_set;
2050 
2051 	sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
2052 				scsi_mq_inline_sgl_size(shost));
2053 	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2054 	if (scsi_host_get_prot(shost))
2055 		cmd_size += sizeof(struct scsi_data_buffer) +
2056 			sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
2057 
2058 	memset(tag_set, 0, sizeof(*tag_set));
2059 	if (shost->hostt->commit_rqs)
2060 		tag_set->ops = &scsi_mq_ops;
2061 	else
2062 		tag_set->ops = &scsi_mq_ops_no_commit;
2063 	tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
2064 	tag_set->nr_maps = shost->nr_maps ? : 1;
2065 	tag_set->queue_depth = shost->can_queue;
2066 	tag_set->cmd_size = cmd_size;
2067 	tag_set->numa_node = dev_to_node(shost->dma_dev);
2068 	tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
2069 	tag_set->flags |=
2070 		BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2071 	if (shost->queuecommand_may_block)
2072 		tag_set->flags |= BLK_MQ_F_BLOCKING;
2073 	tag_set->driver_data = shost;
2074 	if (shost->host_tagset)
2075 		tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
2076 
2077 	return blk_mq_alloc_tag_set(tag_set);
2078 }
2079 
2080 void scsi_mq_free_tags(struct kref *kref)
2081 {
2082 	struct Scsi_Host *shost = container_of(kref, typeof(*shost),
2083 					       tagset_refcnt);
2084 
2085 	blk_mq_free_tag_set(&shost->tag_set);
2086 	complete(&shost->tagset_freed);
2087 }
2088 
2089 /**
2090  * scsi_device_from_queue - return sdev associated with a request_queue
2091  * @q: The request queue to return the sdev from
2092  *
2093  * Return the sdev associated with a request queue or NULL if the
2094  * request_queue does not reference a SCSI device.
2095  */
2096 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2097 {
2098 	struct scsi_device *sdev = NULL;
2099 
2100 	if (q->mq_ops == &scsi_mq_ops_no_commit ||
2101 	    q->mq_ops == &scsi_mq_ops)
2102 		sdev = q->queuedata;
2103 	if (!sdev || !get_device(&sdev->sdev_gendev))
2104 		sdev = NULL;
2105 
2106 	return sdev;
2107 }
2108 /*
2109  * pktcdvd should have been integrated into the SCSI layers, but for historical
2110  * reasons like the old IDE driver it isn't.  This export allows it to safely
2111  * probe if a given device is a SCSI one and only attach to that.
2112  */
2113 #ifdef CONFIG_CDROM_PKTCDVD_MODULE
2114 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2115 #endif
2116 
2117 /**
2118  * scsi_block_requests - Utility function used by low-level drivers to prevent
2119  * further commands from being queued to the device.
2120  * @shost:  host in question
2121  *
2122  * There is no timer nor any other means by which the requests get unblocked
2123  * other than the low-level driver calling scsi_unblock_requests().
2124  */
2125 void scsi_block_requests(struct Scsi_Host *shost)
2126 {
2127 	shost->host_self_blocked = 1;
2128 }
2129 EXPORT_SYMBOL(scsi_block_requests);
2130 
2131 /**
2132  * scsi_unblock_requests - Utility function used by low-level drivers to allow
2133  * further commands to be queued to the device.
2134  * @shost:  host in question
2135  *
2136  * There is no timer nor any other means by which the requests get unblocked
2137  * other than the low-level driver calling scsi_unblock_requests(). This is done
2138  * as an API function so that changes to the internals of the scsi mid-layer
2139  * won't require wholesale changes to drivers that use this feature.
2140  */
2141 void scsi_unblock_requests(struct Scsi_Host *shost)
2142 {
2143 	shost->host_self_blocked = 0;
2144 	scsi_run_host_queues(shost);
2145 }
2146 EXPORT_SYMBOL(scsi_unblock_requests);
2147 
2148 void scsi_exit_queue(void)
2149 {
2150 	kmem_cache_destroy(scsi_sense_cache);
2151 }
2152 
2153 /**
2154  *	scsi_mode_select - issue a mode select
2155  *	@sdev:	SCSI device to be queried
2156  *	@pf:	Page format bit (1 == standard, 0 == vendor specific)
2157  *	@sp:	Save page bit (0 == don't save, 1 == save)
2158  *	@buffer: request buffer (may not be smaller than eight bytes)
2159  *	@len:	length of request buffer.
2160  *	@timeout: command timeout
2161  *	@retries: number of retries before failing
2162  *	@data: returns a structure abstracting the mode header data
2163  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2164  *		must be SCSI_SENSE_BUFFERSIZE big.
2165  *
2166  *	Returns zero if successful; negative error number or scsi
2167  *	status on error
2168  *
2169  */
2170 int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2171 		     unsigned char *buffer, int len, int timeout, int retries,
2172 		     struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2173 {
2174 	unsigned char cmd[10];
2175 	unsigned char *real_buffer;
2176 	const struct scsi_exec_args exec_args = {
2177 		.sshdr = sshdr,
2178 	};
2179 	int ret;
2180 
2181 	memset(cmd, 0, sizeof(cmd));
2182 	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2183 
2184 	/*
2185 	 * Use MODE SELECT(10) if the device asked for it or if the mode page
2186 	 * and the mode select header cannot fit within the maximumm 255 bytes
2187 	 * of the MODE SELECT(6) command.
2188 	 */
2189 	if (sdev->use_10_for_ms ||
2190 	    len + 4 > 255 ||
2191 	    data->block_descriptor_length > 255) {
2192 		if (len > 65535 - 8)
2193 			return -EINVAL;
2194 		real_buffer = kmalloc(8 + len, GFP_KERNEL);
2195 		if (!real_buffer)
2196 			return -ENOMEM;
2197 		memcpy(real_buffer + 8, buffer, len);
2198 		len += 8;
2199 		real_buffer[0] = 0;
2200 		real_buffer[1] = 0;
2201 		real_buffer[2] = data->medium_type;
2202 		real_buffer[3] = data->device_specific;
2203 		real_buffer[4] = data->longlba ? 0x01 : 0;
2204 		real_buffer[5] = 0;
2205 		put_unaligned_be16(data->block_descriptor_length,
2206 				   &real_buffer[6]);
2207 
2208 		cmd[0] = MODE_SELECT_10;
2209 		put_unaligned_be16(len, &cmd[7]);
2210 	} else {
2211 		if (data->longlba)
2212 			return -EINVAL;
2213 
2214 		real_buffer = kmalloc(4 + len, GFP_KERNEL);
2215 		if (!real_buffer)
2216 			return -ENOMEM;
2217 		memcpy(real_buffer + 4, buffer, len);
2218 		len += 4;
2219 		real_buffer[0] = 0;
2220 		real_buffer[1] = data->medium_type;
2221 		real_buffer[2] = data->device_specific;
2222 		real_buffer[3] = data->block_descriptor_length;
2223 
2224 		cmd[0] = MODE_SELECT;
2225 		cmd[4] = len;
2226 	}
2227 
2228 	ret = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, real_buffer, len,
2229 			       timeout, retries, &exec_args);
2230 	kfree(real_buffer);
2231 	return ret;
2232 }
2233 EXPORT_SYMBOL_GPL(scsi_mode_select);
2234 
2235 /**
2236  *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2237  *	@sdev:	SCSI device to be queried
2238  *	@dbd:	set to prevent mode sense from returning block descriptors
2239  *	@modepage: mode page being requested
2240  *	@subpage: sub-page of the mode page being requested
2241  *	@buffer: request buffer (may not be smaller than eight bytes)
2242  *	@len:	length of request buffer.
2243  *	@timeout: command timeout
2244  *	@retries: number of retries before failing
2245  *	@data: returns a structure abstracting the mode header data
2246  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2247  *		must be SCSI_SENSE_BUFFERSIZE big.
2248  *
2249  *	Returns zero if successful, or a negative error number on failure
2250  */
2251 int
2252 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, int subpage,
2253 		  unsigned char *buffer, int len, int timeout, int retries,
2254 		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2255 {
2256 	unsigned char cmd[12];
2257 	int use_10_for_ms;
2258 	int header_length;
2259 	int result;
2260 	struct scsi_sense_hdr my_sshdr;
2261 	struct scsi_failure failure_defs[] = {
2262 		{
2263 			.sense = UNIT_ATTENTION,
2264 			.asc = SCMD_FAILURE_ASC_ANY,
2265 			.ascq = SCMD_FAILURE_ASCQ_ANY,
2266 			.allowed = retries,
2267 			.result = SAM_STAT_CHECK_CONDITION,
2268 		},
2269 		{}
2270 	};
2271 	struct scsi_failures failures = {
2272 		.failure_definitions = failure_defs,
2273 	};
2274 	const struct scsi_exec_args exec_args = {
2275 		/* caller might not be interested in sense, but we need it */
2276 		.sshdr = sshdr ? : &my_sshdr,
2277 		.failures = &failures,
2278 	};
2279 
2280 	memset(data, 0, sizeof(*data));
2281 	memset(&cmd[0], 0, 12);
2282 
2283 	dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2284 	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */
2285 	cmd[2] = modepage;
2286 	cmd[3] = subpage;
2287 
2288 	sshdr = exec_args.sshdr;
2289 
2290  retry:
2291 	use_10_for_ms = sdev->use_10_for_ms || len > 255;
2292 
2293 	if (use_10_for_ms) {
2294 		if (len < 8 || len > 65535)
2295 			return -EINVAL;
2296 
2297 		cmd[0] = MODE_SENSE_10;
2298 		put_unaligned_be16(len, &cmd[7]);
2299 		header_length = 8;
2300 	} else {
2301 		if (len < 4)
2302 			return -EINVAL;
2303 
2304 		cmd[0] = MODE_SENSE;
2305 		cmd[4] = len;
2306 		header_length = 4;
2307 	}
2308 
2309 	memset(buffer, 0, len);
2310 
2311 	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, buffer, len,
2312 				  timeout, retries, &exec_args);
2313 	if (result < 0)
2314 		return result;
2315 
2316 	/* This code looks awful: what it's doing is making sure an
2317 	 * ILLEGAL REQUEST sense return identifies the actual command
2318 	 * byte as the problem.  MODE_SENSE commands can return
2319 	 * ILLEGAL REQUEST if the code page isn't supported */
2320 
2321 	if (!scsi_status_is_good(result)) {
2322 		if (scsi_sense_valid(sshdr)) {
2323 			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2324 			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2325 				/*
2326 				 * Invalid command operation code: retry using
2327 				 * MODE SENSE(6) if this was a MODE SENSE(10)
2328 				 * request, except if the request mode page is
2329 				 * too large for MODE SENSE single byte
2330 				 * allocation length field.
2331 				 */
2332 				if (use_10_for_ms) {
2333 					if (len > 255)
2334 						return -EIO;
2335 					sdev->use_10_for_ms = 0;
2336 					goto retry;
2337 				}
2338 			}
2339 		}
2340 		return -EIO;
2341 	}
2342 	if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2343 		     (modepage == 6 || modepage == 8))) {
2344 		/* Initio breakage? */
2345 		header_length = 0;
2346 		data->length = 13;
2347 		data->medium_type = 0;
2348 		data->device_specific = 0;
2349 		data->longlba = 0;
2350 		data->block_descriptor_length = 0;
2351 	} else if (use_10_for_ms) {
2352 		data->length = get_unaligned_be16(&buffer[0]) + 2;
2353 		data->medium_type = buffer[2];
2354 		data->device_specific = buffer[3];
2355 		data->longlba = buffer[4] & 0x01;
2356 		data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2357 	} else {
2358 		data->length = buffer[0] + 1;
2359 		data->medium_type = buffer[1];
2360 		data->device_specific = buffer[2];
2361 		data->block_descriptor_length = buffer[3];
2362 	}
2363 	data->header_length = header_length;
2364 
2365 	return 0;
2366 }
2367 EXPORT_SYMBOL(scsi_mode_sense);
2368 
2369 /**
2370  *	scsi_test_unit_ready - test if unit is ready
2371  *	@sdev:	scsi device to change the state of.
2372  *	@timeout: command timeout
2373  *	@retries: number of retries before failing
2374  *	@sshdr: outpout pointer for decoded sense information.
2375  *
2376  *	Returns zero if unsuccessful or an error if TUR failed.  For
2377  *	removable media, UNIT_ATTENTION sets ->changed flag.
2378  **/
2379 int
2380 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2381 		     struct scsi_sense_hdr *sshdr)
2382 {
2383 	char cmd[] = {
2384 		TEST_UNIT_READY, 0, 0, 0, 0, 0,
2385 	};
2386 	const struct scsi_exec_args exec_args = {
2387 		.sshdr = sshdr,
2388 	};
2389 	int result;
2390 
2391 	/* try to eat the UNIT_ATTENTION if there are enough retries */
2392 	do {
2393 		result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, NULL, 0,
2394 					  timeout, 1, &exec_args);
2395 		if (sdev->removable && result > 0 && scsi_sense_valid(sshdr) &&
2396 		    sshdr->sense_key == UNIT_ATTENTION)
2397 			sdev->changed = 1;
2398 	} while (result > 0 && scsi_sense_valid(sshdr) &&
2399 		 sshdr->sense_key == UNIT_ATTENTION && --retries);
2400 
2401 	return result;
2402 }
2403 EXPORT_SYMBOL(scsi_test_unit_ready);
2404 
2405 /**
2406  *	scsi_device_set_state - Take the given device through the device state model.
2407  *	@sdev:	scsi device to change the state of.
2408  *	@state:	state to change to.
2409  *
2410  *	Returns zero if successful or an error if the requested
2411  *	transition is illegal.
2412  */
2413 int
2414 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2415 {
2416 	enum scsi_device_state oldstate = sdev->sdev_state;
2417 
2418 	if (state == oldstate)
2419 		return 0;
2420 
2421 	switch (state) {
2422 	case SDEV_CREATED:
2423 		switch (oldstate) {
2424 		case SDEV_CREATED_BLOCK:
2425 			break;
2426 		default:
2427 			goto illegal;
2428 		}
2429 		break;
2430 
2431 	case SDEV_RUNNING:
2432 		switch (oldstate) {
2433 		case SDEV_CREATED:
2434 		case SDEV_OFFLINE:
2435 		case SDEV_TRANSPORT_OFFLINE:
2436 		case SDEV_QUIESCE:
2437 		case SDEV_BLOCK:
2438 			break;
2439 		default:
2440 			goto illegal;
2441 		}
2442 		break;
2443 
2444 	case SDEV_QUIESCE:
2445 		switch (oldstate) {
2446 		case SDEV_RUNNING:
2447 		case SDEV_OFFLINE:
2448 		case SDEV_TRANSPORT_OFFLINE:
2449 			break;
2450 		default:
2451 			goto illegal;
2452 		}
2453 		break;
2454 
2455 	case SDEV_OFFLINE:
2456 	case SDEV_TRANSPORT_OFFLINE:
2457 		switch (oldstate) {
2458 		case SDEV_CREATED:
2459 		case SDEV_RUNNING:
2460 		case SDEV_QUIESCE:
2461 		case SDEV_BLOCK:
2462 			break;
2463 		default:
2464 			goto illegal;
2465 		}
2466 		break;
2467 
2468 	case SDEV_BLOCK:
2469 		switch (oldstate) {
2470 		case SDEV_RUNNING:
2471 		case SDEV_CREATED_BLOCK:
2472 		case SDEV_QUIESCE:
2473 		case SDEV_OFFLINE:
2474 			break;
2475 		default:
2476 			goto illegal;
2477 		}
2478 		break;
2479 
2480 	case SDEV_CREATED_BLOCK:
2481 		switch (oldstate) {
2482 		case SDEV_CREATED:
2483 			break;
2484 		default:
2485 			goto illegal;
2486 		}
2487 		break;
2488 
2489 	case SDEV_CANCEL:
2490 		switch (oldstate) {
2491 		case SDEV_CREATED:
2492 		case SDEV_RUNNING:
2493 		case SDEV_QUIESCE:
2494 		case SDEV_OFFLINE:
2495 		case SDEV_TRANSPORT_OFFLINE:
2496 			break;
2497 		default:
2498 			goto illegal;
2499 		}
2500 		break;
2501 
2502 	case SDEV_DEL:
2503 		switch (oldstate) {
2504 		case SDEV_CREATED:
2505 		case SDEV_RUNNING:
2506 		case SDEV_OFFLINE:
2507 		case SDEV_TRANSPORT_OFFLINE:
2508 		case SDEV_CANCEL:
2509 		case SDEV_BLOCK:
2510 		case SDEV_CREATED_BLOCK:
2511 			break;
2512 		default:
2513 			goto illegal;
2514 		}
2515 		break;
2516 
2517 	}
2518 	sdev->offline_already = false;
2519 	sdev->sdev_state = state;
2520 	return 0;
2521 
2522  illegal:
2523 	SCSI_LOG_ERROR_RECOVERY(1,
2524 				sdev_printk(KERN_ERR, sdev,
2525 					    "Illegal state transition %s->%s",
2526 					    scsi_device_state_name(oldstate),
2527 					    scsi_device_state_name(state))
2528 				);
2529 	return -EINVAL;
2530 }
2531 EXPORT_SYMBOL(scsi_device_set_state);
2532 
2533 /**
2534  *	scsi_evt_emit - emit a single SCSI device uevent
2535  *	@sdev: associated SCSI device
2536  *	@evt: event to emit
2537  *
2538  *	Send a single uevent (scsi_event) to the associated scsi_device.
2539  */
2540 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2541 {
2542 	int idx = 0;
2543 	char *envp[3];
2544 
2545 	switch (evt->evt_type) {
2546 	case SDEV_EVT_MEDIA_CHANGE:
2547 		envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2548 		break;
2549 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2550 		scsi_rescan_device(sdev);
2551 		envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2552 		break;
2553 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2554 		envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2555 		break;
2556 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2557 	       envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2558 		break;
2559 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2560 		envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2561 		break;
2562 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2563 		envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2564 		break;
2565 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2566 		envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2567 		break;
2568 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2569 		envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2570 		break;
2571 	default:
2572 		/* do nothing */
2573 		break;
2574 	}
2575 
2576 	envp[idx++] = NULL;
2577 
2578 	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2579 }
2580 
2581 /**
2582  *	scsi_evt_thread - send a uevent for each scsi event
2583  *	@work: work struct for scsi_device
2584  *
2585  *	Dispatch queued events to their associated scsi_device kobjects
2586  *	as uevents.
2587  */
2588 void scsi_evt_thread(struct work_struct *work)
2589 {
2590 	struct scsi_device *sdev;
2591 	enum scsi_device_event evt_type;
2592 	LIST_HEAD(event_list);
2593 
2594 	sdev = container_of(work, struct scsi_device, event_work);
2595 
2596 	for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2597 		if (test_and_clear_bit(evt_type, sdev->pending_events))
2598 			sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2599 
2600 	while (1) {
2601 		struct scsi_event *evt;
2602 		struct list_head *this, *tmp;
2603 		unsigned long flags;
2604 
2605 		spin_lock_irqsave(&sdev->list_lock, flags);
2606 		list_splice_init(&sdev->event_list, &event_list);
2607 		spin_unlock_irqrestore(&sdev->list_lock, flags);
2608 
2609 		if (list_empty(&event_list))
2610 			break;
2611 
2612 		list_for_each_safe(this, tmp, &event_list) {
2613 			evt = list_entry(this, struct scsi_event, node);
2614 			list_del(&evt->node);
2615 			scsi_evt_emit(sdev, evt);
2616 			kfree(evt);
2617 		}
2618 	}
2619 }
2620 
2621 /**
2622  * 	sdev_evt_send - send asserted event to uevent thread
2623  *	@sdev: scsi_device event occurred on
2624  *	@evt: event to send
2625  *
2626  *	Assert scsi device event asynchronously.
2627  */
2628 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2629 {
2630 	unsigned long flags;
2631 
2632 #if 0
2633 	/* FIXME: currently this check eliminates all media change events
2634 	 * for polled devices.  Need to update to discriminate between AN
2635 	 * and polled events */
2636 	if (!test_bit(evt->evt_type, sdev->supported_events)) {
2637 		kfree(evt);
2638 		return;
2639 	}
2640 #endif
2641 
2642 	spin_lock_irqsave(&sdev->list_lock, flags);
2643 	list_add_tail(&evt->node, &sdev->event_list);
2644 	schedule_work(&sdev->event_work);
2645 	spin_unlock_irqrestore(&sdev->list_lock, flags);
2646 }
2647 EXPORT_SYMBOL_GPL(sdev_evt_send);
2648 
2649 /**
2650  * 	sdev_evt_alloc - allocate a new scsi event
2651  *	@evt_type: type of event to allocate
2652  *	@gfpflags: GFP flags for allocation
2653  *
2654  *	Allocates and returns a new scsi_event.
2655  */
2656 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2657 				  gfp_t gfpflags)
2658 {
2659 	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2660 	if (!evt)
2661 		return NULL;
2662 
2663 	evt->evt_type = evt_type;
2664 	INIT_LIST_HEAD(&evt->node);
2665 
2666 	/* evt_type-specific initialization, if any */
2667 	switch (evt_type) {
2668 	case SDEV_EVT_MEDIA_CHANGE:
2669 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2670 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2671 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2672 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2673 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2674 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2675 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2676 	default:
2677 		/* do nothing */
2678 		break;
2679 	}
2680 
2681 	return evt;
2682 }
2683 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2684 
2685 /**
2686  * 	sdev_evt_send_simple - send asserted event to uevent thread
2687  *	@sdev: scsi_device event occurred on
2688  *	@evt_type: type of event to send
2689  *	@gfpflags: GFP flags for allocation
2690  *
2691  *	Assert scsi device event asynchronously, given an event type.
2692  */
2693 void sdev_evt_send_simple(struct scsi_device *sdev,
2694 			  enum scsi_device_event evt_type, gfp_t gfpflags)
2695 {
2696 	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2697 	if (!evt) {
2698 		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2699 			    evt_type);
2700 		return;
2701 	}
2702 
2703 	sdev_evt_send(sdev, evt);
2704 }
2705 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2706 
2707 /**
2708  *	scsi_device_quiesce - Block all commands except power management.
2709  *	@sdev:	scsi device to quiesce.
2710  *
2711  *	This works by trying to transition to the SDEV_QUIESCE state
2712  *	(which must be a legal transition).  When the device is in this
2713  *	state, only power management requests will be accepted, all others will
2714  *	be deferred.
2715  *
2716  *	Must be called with user context, may sleep.
2717  *
2718  *	Returns zero if unsuccessful or an error if not.
2719  */
2720 int
2721 scsi_device_quiesce(struct scsi_device *sdev)
2722 {
2723 	struct request_queue *q = sdev->request_queue;
2724 	int err;
2725 
2726 	/*
2727 	 * It is allowed to call scsi_device_quiesce() multiple times from
2728 	 * the same context but concurrent scsi_device_quiesce() calls are
2729 	 * not allowed.
2730 	 */
2731 	WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2732 
2733 	if (sdev->quiesced_by == current)
2734 		return 0;
2735 
2736 	blk_set_pm_only(q);
2737 
2738 	blk_mq_freeze_queue(q);
2739 	/*
2740 	 * Ensure that the effect of blk_set_pm_only() will be visible
2741 	 * for percpu_ref_tryget() callers that occur after the queue
2742 	 * unfreeze even if the queue was already frozen before this function
2743 	 * was called. See also https://lwn.net/Articles/573497/.
2744 	 */
2745 	synchronize_rcu();
2746 	blk_mq_unfreeze_queue(q);
2747 
2748 	mutex_lock(&sdev->state_mutex);
2749 	err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2750 	if (err == 0)
2751 		sdev->quiesced_by = current;
2752 	else
2753 		blk_clear_pm_only(q);
2754 	mutex_unlock(&sdev->state_mutex);
2755 
2756 	return err;
2757 }
2758 EXPORT_SYMBOL(scsi_device_quiesce);
2759 
2760 /**
2761  *	scsi_device_resume - Restart user issued commands to a quiesced device.
2762  *	@sdev:	scsi device to resume.
2763  *
2764  *	Moves the device from quiesced back to running and restarts the
2765  *	queues.
2766  *
2767  *	Must be called with user context, may sleep.
2768  */
2769 void scsi_device_resume(struct scsi_device *sdev)
2770 {
2771 	/* check if the device state was mutated prior to resume, and if
2772 	 * so assume the state is being managed elsewhere (for example
2773 	 * device deleted during suspend)
2774 	 */
2775 	mutex_lock(&sdev->state_mutex);
2776 	if (sdev->sdev_state == SDEV_QUIESCE)
2777 		scsi_device_set_state(sdev, SDEV_RUNNING);
2778 	if (sdev->quiesced_by) {
2779 		sdev->quiesced_by = NULL;
2780 		blk_clear_pm_only(sdev->request_queue);
2781 	}
2782 	mutex_unlock(&sdev->state_mutex);
2783 }
2784 EXPORT_SYMBOL(scsi_device_resume);
2785 
2786 static void
2787 device_quiesce_fn(struct scsi_device *sdev, void *data)
2788 {
2789 	scsi_device_quiesce(sdev);
2790 }
2791 
2792 void
2793 scsi_target_quiesce(struct scsi_target *starget)
2794 {
2795 	starget_for_each_device(starget, NULL, device_quiesce_fn);
2796 }
2797 EXPORT_SYMBOL(scsi_target_quiesce);
2798 
2799 static void
2800 device_resume_fn(struct scsi_device *sdev, void *data)
2801 {
2802 	scsi_device_resume(sdev);
2803 }
2804 
2805 void
2806 scsi_target_resume(struct scsi_target *starget)
2807 {
2808 	starget_for_each_device(starget, NULL, device_resume_fn);
2809 }
2810 EXPORT_SYMBOL(scsi_target_resume);
2811 
2812 static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2813 {
2814 	if (scsi_device_set_state(sdev, SDEV_BLOCK))
2815 		return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2816 
2817 	return 0;
2818 }
2819 
2820 void scsi_start_queue(struct scsi_device *sdev)
2821 {
2822 	if (cmpxchg(&sdev->queue_stopped, 1, 0))
2823 		blk_mq_unquiesce_queue(sdev->request_queue);
2824 }
2825 
2826 static void scsi_stop_queue(struct scsi_device *sdev)
2827 {
2828 	/*
2829 	 * The atomic variable of ->queue_stopped covers that
2830 	 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2831 	 *
2832 	 * The caller needs to wait until quiesce is done.
2833 	 */
2834 	if (!cmpxchg(&sdev->queue_stopped, 0, 1))
2835 		blk_mq_quiesce_queue_nowait(sdev->request_queue);
2836 }
2837 
2838 /**
2839  * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2840  * @sdev: device to block
2841  *
2842  * Pause SCSI command processing on the specified device. Does not sleep.
2843  *
2844  * Returns zero if successful or a negative error code upon failure.
2845  *
2846  * Notes:
2847  * This routine transitions the device to the SDEV_BLOCK state (which must be
2848  * a legal transition). When the device is in this state, command processing
2849  * is paused until the device leaves the SDEV_BLOCK state. See also
2850  * scsi_internal_device_unblock_nowait().
2851  */
2852 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2853 {
2854 	int ret = __scsi_internal_device_block_nowait(sdev);
2855 
2856 	/*
2857 	 * The device has transitioned to SDEV_BLOCK.  Stop the
2858 	 * block layer from calling the midlayer with this device's
2859 	 * request queue.
2860 	 */
2861 	if (!ret)
2862 		scsi_stop_queue(sdev);
2863 	return ret;
2864 }
2865 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2866 
2867 /**
2868  * scsi_device_block - try to transition to the SDEV_BLOCK state
2869  * @sdev: device to block
2870  * @data: dummy argument, ignored
2871  *
2872  * Pause SCSI command processing on the specified device. Callers must wait
2873  * until all ongoing scsi_queue_rq() calls have finished after this function
2874  * returns.
2875  *
2876  * Note:
2877  * This routine transitions the device to the SDEV_BLOCK state (which must be
2878  * a legal transition). When the device is in this state, command processing
2879  * is paused until the device leaves the SDEV_BLOCK state. See also
2880  * scsi_internal_device_unblock().
2881  */
2882 static void scsi_device_block(struct scsi_device *sdev, void *data)
2883 {
2884 	int err;
2885 	enum scsi_device_state state;
2886 
2887 	mutex_lock(&sdev->state_mutex);
2888 	err = __scsi_internal_device_block_nowait(sdev);
2889 	state = sdev->sdev_state;
2890 	if (err == 0)
2891 		/*
2892 		 * scsi_stop_queue() must be called with the state_mutex
2893 		 * held. Otherwise a simultaneous scsi_start_queue() call
2894 		 * might unquiesce the queue before we quiesce it.
2895 		 */
2896 		scsi_stop_queue(sdev);
2897 
2898 	mutex_unlock(&sdev->state_mutex);
2899 
2900 	WARN_ONCE(err, "%s: failed to block %s in state %d\n",
2901 		  __func__, dev_name(&sdev->sdev_gendev), state);
2902 }
2903 
2904 /**
2905  * scsi_internal_device_unblock_nowait - resume a device after a block request
2906  * @sdev:	device to resume
2907  * @new_state:	state to set the device to after unblocking
2908  *
2909  * Restart the device queue for a previously suspended SCSI device. Does not
2910  * sleep.
2911  *
2912  * Returns zero if successful or a negative error code upon failure.
2913  *
2914  * Notes:
2915  * This routine transitions the device to the SDEV_RUNNING state or to one of
2916  * the offline states (which must be a legal transition) allowing the midlayer
2917  * to goose the queue for this device.
2918  */
2919 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2920 					enum scsi_device_state new_state)
2921 {
2922 	switch (new_state) {
2923 	case SDEV_RUNNING:
2924 	case SDEV_TRANSPORT_OFFLINE:
2925 		break;
2926 	default:
2927 		return -EINVAL;
2928 	}
2929 
2930 	/*
2931 	 * Try to transition the scsi device to SDEV_RUNNING or one of the
2932 	 * offlined states and goose the device queue if successful.
2933 	 */
2934 	switch (sdev->sdev_state) {
2935 	case SDEV_BLOCK:
2936 	case SDEV_TRANSPORT_OFFLINE:
2937 		sdev->sdev_state = new_state;
2938 		break;
2939 	case SDEV_CREATED_BLOCK:
2940 		if (new_state == SDEV_TRANSPORT_OFFLINE ||
2941 		    new_state == SDEV_OFFLINE)
2942 			sdev->sdev_state = new_state;
2943 		else
2944 			sdev->sdev_state = SDEV_CREATED;
2945 		break;
2946 	case SDEV_CANCEL:
2947 	case SDEV_OFFLINE:
2948 		break;
2949 	default:
2950 		return -EINVAL;
2951 	}
2952 	scsi_start_queue(sdev);
2953 
2954 	return 0;
2955 }
2956 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2957 
2958 /**
2959  * scsi_internal_device_unblock - resume a device after a block request
2960  * @sdev:	device to resume
2961  * @new_state:	state to set the device to after unblocking
2962  *
2963  * Restart the device queue for a previously suspended SCSI device. May sleep.
2964  *
2965  * Returns zero if successful or a negative error code upon failure.
2966  *
2967  * Notes:
2968  * This routine transitions the device to the SDEV_RUNNING state or to one of
2969  * the offline states (which must be a legal transition) allowing the midlayer
2970  * to goose the queue for this device.
2971  */
2972 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2973 					enum scsi_device_state new_state)
2974 {
2975 	int ret;
2976 
2977 	mutex_lock(&sdev->state_mutex);
2978 	ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2979 	mutex_unlock(&sdev->state_mutex);
2980 
2981 	return ret;
2982 }
2983 
2984 static int
2985 target_block(struct device *dev, void *data)
2986 {
2987 	if (scsi_is_target_device(dev))
2988 		starget_for_each_device(to_scsi_target(dev), NULL,
2989 					scsi_device_block);
2990 	return 0;
2991 }
2992 
2993 /**
2994  * scsi_block_targets - transition all SCSI child devices to SDEV_BLOCK state
2995  * @dev: a parent device of one or more scsi_target devices
2996  * @shost: the Scsi_Host to which this device belongs
2997  *
2998  * Iterate over all children of @dev, which should be scsi_target devices,
2999  * and switch all subordinate scsi devices to SDEV_BLOCK state. Wait for
3000  * ongoing scsi_queue_rq() calls to finish. May sleep.
3001  *
3002  * Note:
3003  * @dev must not itself be a scsi_target device.
3004  */
3005 void
3006 scsi_block_targets(struct Scsi_Host *shost, struct device *dev)
3007 {
3008 	WARN_ON_ONCE(scsi_is_target_device(dev));
3009 	device_for_each_child(dev, NULL, target_block);
3010 	blk_mq_wait_quiesce_done(&shost->tag_set);
3011 }
3012 EXPORT_SYMBOL_GPL(scsi_block_targets);
3013 
3014 static void
3015 device_unblock(struct scsi_device *sdev, void *data)
3016 {
3017 	scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3018 }
3019 
3020 static int
3021 target_unblock(struct device *dev, void *data)
3022 {
3023 	if (scsi_is_target_device(dev))
3024 		starget_for_each_device(to_scsi_target(dev), data,
3025 					device_unblock);
3026 	return 0;
3027 }
3028 
3029 void
3030 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3031 {
3032 	if (scsi_is_target_device(dev))
3033 		starget_for_each_device(to_scsi_target(dev), &new_state,
3034 					device_unblock);
3035 	else
3036 		device_for_each_child(dev, &new_state, target_unblock);
3037 }
3038 EXPORT_SYMBOL_GPL(scsi_target_unblock);
3039 
3040 /**
3041  * scsi_host_block - Try to transition all logical units to the SDEV_BLOCK state
3042  * @shost: device to block
3043  *
3044  * Pause SCSI command processing for all logical units associated with the SCSI
3045  * host and wait until pending scsi_queue_rq() calls have finished.
3046  *
3047  * Returns zero if successful or a negative error code upon failure.
3048  */
3049 int
3050 scsi_host_block(struct Scsi_Host *shost)
3051 {
3052 	struct scsi_device *sdev;
3053 	int ret;
3054 
3055 	/*
3056 	 * Call scsi_internal_device_block_nowait so we can avoid
3057 	 * calling synchronize_rcu() for each LUN.
3058 	 */
3059 	shost_for_each_device(sdev, shost) {
3060 		mutex_lock(&sdev->state_mutex);
3061 		ret = scsi_internal_device_block_nowait(sdev);
3062 		mutex_unlock(&sdev->state_mutex);
3063 		if (ret) {
3064 			scsi_device_put(sdev);
3065 			return ret;
3066 		}
3067 	}
3068 
3069 	/* Wait for ongoing scsi_queue_rq() calls to finish. */
3070 	blk_mq_wait_quiesce_done(&shost->tag_set);
3071 
3072 	return 0;
3073 }
3074 EXPORT_SYMBOL_GPL(scsi_host_block);
3075 
3076 int
3077 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
3078 {
3079 	struct scsi_device *sdev;
3080 	int ret = 0;
3081 
3082 	shost_for_each_device(sdev, shost) {
3083 		ret = scsi_internal_device_unblock(sdev, new_state);
3084 		if (ret) {
3085 			scsi_device_put(sdev);
3086 			break;
3087 		}
3088 	}
3089 	return ret;
3090 }
3091 EXPORT_SYMBOL_GPL(scsi_host_unblock);
3092 
3093 /**
3094  * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3095  * @sgl:	scatter-gather list
3096  * @sg_count:	number of segments in sg
3097  * @offset:	offset in bytes into sg, on return offset into the mapped area
3098  * @len:	bytes to map, on return number of bytes mapped
3099  *
3100  * Returns virtual address of the start of the mapped page
3101  */
3102 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3103 			  size_t *offset, size_t *len)
3104 {
3105 	int i;
3106 	size_t sg_len = 0, len_complete = 0;
3107 	struct scatterlist *sg;
3108 	struct page *page;
3109 
3110 	WARN_ON(!irqs_disabled());
3111 
3112 	for_each_sg(sgl, sg, sg_count, i) {
3113 		len_complete = sg_len; /* Complete sg-entries */
3114 		sg_len += sg->length;
3115 		if (sg_len > *offset)
3116 			break;
3117 	}
3118 
3119 	if (unlikely(i == sg_count)) {
3120 		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3121 			"elements %d\n",
3122 		       __func__, sg_len, *offset, sg_count);
3123 		WARN_ON(1);
3124 		return NULL;
3125 	}
3126 
3127 	/* Offset starting from the beginning of first page in this sg-entry */
3128 	*offset = *offset - len_complete + sg->offset;
3129 
3130 	/* Assumption: contiguous pages can be accessed as "page + i" */
3131 	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3132 	*offset &= ~PAGE_MASK;
3133 
3134 	/* Bytes in this sg-entry from *offset to the end of the page */
3135 	sg_len = PAGE_SIZE - *offset;
3136 	if (*len > sg_len)
3137 		*len = sg_len;
3138 
3139 	return kmap_atomic(page);
3140 }
3141 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3142 
3143 /**
3144  * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3145  * @virt:	virtual address to be unmapped
3146  */
3147 void scsi_kunmap_atomic_sg(void *virt)
3148 {
3149 	kunmap_atomic(virt);
3150 }
3151 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3152 
3153 void sdev_disable_disk_events(struct scsi_device *sdev)
3154 {
3155 	atomic_inc(&sdev->disk_events_disable_depth);
3156 }
3157 EXPORT_SYMBOL(sdev_disable_disk_events);
3158 
3159 void sdev_enable_disk_events(struct scsi_device *sdev)
3160 {
3161 	if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3162 		return;
3163 	atomic_dec(&sdev->disk_events_disable_depth);
3164 }
3165 EXPORT_SYMBOL(sdev_enable_disk_events);
3166 
3167 static unsigned char designator_prio(const unsigned char *d)
3168 {
3169 	if (d[1] & 0x30)
3170 		/* not associated with LUN */
3171 		return 0;
3172 
3173 	if (d[3] == 0)
3174 		/* invalid length */
3175 		return 0;
3176 
3177 	/*
3178 	 * Order of preference for lun descriptor:
3179 	 * - SCSI name string
3180 	 * - NAA IEEE Registered Extended
3181 	 * - EUI-64 based 16-byte
3182 	 * - EUI-64 based 12-byte
3183 	 * - NAA IEEE Registered
3184 	 * - NAA IEEE Extended
3185 	 * - EUI-64 based 8-byte
3186 	 * - SCSI name string (truncated)
3187 	 * - T10 Vendor ID
3188 	 * as longer descriptors reduce the likelyhood
3189 	 * of identification clashes.
3190 	 */
3191 
3192 	switch (d[1] & 0xf) {
3193 	case 8:
3194 		/* SCSI name string, variable-length UTF-8 */
3195 		return 9;
3196 	case 3:
3197 		switch (d[4] >> 4) {
3198 		case 6:
3199 			/* NAA registered extended */
3200 			return 8;
3201 		case 5:
3202 			/* NAA registered */
3203 			return 5;
3204 		case 4:
3205 			/* NAA extended */
3206 			return 4;
3207 		case 3:
3208 			/* NAA locally assigned */
3209 			return 1;
3210 		default:
3211 			break;
3212 		}
3213 		break;
3214 	case 2:
3215 		switch (d[3]) {
3216 		case 16:
3217 			/* EUI64-based, 16 byte */
3218 			return 7;
3219 		case 12:
3220 			/* EUI64-based, 12 byte */
3221 			return 6;
3222 		case 8:
3223 			/* EUI64-based, 8 byte */
3224 			return 3;
3225 		default:
3226 			break;
3227 		}
3228 		break;
3229 	case 1:
3230 		/* T10 vendor ID */
3231 		return 1;
3232 	default:
3233 		break;
3234 	}
3235 
3236 	return 0;
3237 }
3238 
3239 /**
3240  * scsi_vpd_lun_id - return a unique device identification
3241  * @sdev: SCSI device
3242  * @id:   buffer for the identification
3243  * @id_len:  length of the buffer
3244  *
3245  * Copies a unique device identification into @id based
3246  * on the information in the VPD page 0x83 of the device.
3247  * The string will be formatted as a SCSI name string.
3248  *
3249  * Returns the length of the identification or error on failure.
3250  * If the identifier is longer than the supplied buffer the actual
3251  * identifier length is returned and the buffer is not zero-padded.
3252  */
3253 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3254 {
3255 	u8 cur_id_prio = 0;
3256 	u8 cur_id_size = 0;
3257 	const unsigned char *d, *cur_id_str;
3258 	const struct scsi_vpd *vpd_pg83;
3259 	int id_size = -EINVAL;
3260 
3261 	rcu_read_lock();
3262 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3263 	if (!vpd_pg83) {
3264 		rcu_read_unlock();
3265 		return -ENXIO;
3266 	}
3267 
3268 	/* The id string must be at least 20 bytes + terminating NULL byte */
3269 	if (id_len < 21) {
3270 		rcu_read_unlock();
3271 		return -EINVAL;
3272 	}
3273 
3274 	memset(id, 0, id_len);
3275 	for (d = vpd_pg83->data + 4;
3276 	     d < vpd_pg83->data + vpd_pg83->len;
3277 	     d += d[3] + 4) {
3278 		u8 prio = designator_prio(d);
3279 
3280 		if (prio == 0 || cur_id_prio > prio)
3281 			continue;
3282 
3283 		switch (d[1] & 0xf) {
3284 		case 0x1:
3285 			/* T10 Vendor ID */
3286 			if (cur_id_size > d[3])
3287 				break;
3288 			cur_id_prio = prio;
3289 			cur_id_size = d[3];
3290 			if (cur_id_size + 4 > id_len)
3291 				cur_id_size = id_len - 4;
3292 			cur_id_str = d + 4;
3293 			id_size = snprintf(id, id_len, "t10.%*pE",
3294 					   cur_id_size, cur_id_str);
3295 			break;
3296 		case 0x2:
3297 			/* EUI-64 */
3298 			cur_id_prio = prio;
3299 			cur_id_size = d[3];
3300 			cur_id_str = d + 4;
3301 			switch (cur_id_size) {
3302 			case 8:
3303 				id_size = snprintf(id, id_len,
3304 						   "eui.%8phN",
3305 						   cur_id_str);
3306 				break;
3307 			case 12:
3308 				id_size = snprintf(id, id_len,
3309 						   "eui.%12phN",
3310 						   cur_id_str);
3311 				break;
3312 			case 16:
3313 				id_size = snprintf(id, id_len,
3314 						   "eui.%16phN",
3315 						   cur_id_str);
3316 				break;
3317 			default:
3318 				break;
3319 			}
3320 			break;
3321 		case 0x3:
3322 			/* NAA */
3323 			cur_id_prio = prio;
3324 			cur_id_size = d[3];
3325 			cur_id_str = d + 4;
3326 			switch (cur_id_size) {
3327 			case 8:
3328 				id_size = snprintf(id, id_len,
3329 						   "naa.%8phN",
3330 						   cur_id_str);
3331 				break;
3332 			case 16:
3333 				id_size = snprintf(id, id_len,
3334 						   "naa.%16phN",
3335 						   cur_id_str);
3336 				break;
3337 			default:
3338 				break;
3339 			}
3340 			break;
3341 		case 0x8:
3342 			/* SCSI name string */
3343 			if (cur_id_size > d[3])
3344 				break;
3345 			/* Prefer others for truncated descriptor */
3346 			if (d[3] > id_len) {
3347 				prio = 2;
3348 				if (cur_id_prio > prio)
3349 					break;
3350 			}
3351 			cur_id_prio = prio;
3352 			cur_id_size = id_size = d[3];
3353 			cur_id_str = d + 4;
3354 			if (cur_id_size >= id_len)
3355 				cur_id_size = id_len - 1;
3356 			memcpy(id, cur_id_str, cur_id_size);
3357 			break;
3358 		default:
3359 			break;
3360 		}
3361 	}
3362 	rcu_read_unlock();
3363 
3364 	return id_size;
3365 }
3366 EXPORT_SYMBOL(scsi_vpd_lun_id);
3367 
3368 /*
3369  * scsi_vpd_tpg_id - return a target port group identifier
3370  * @sdev: SCSI device
3371  *
3372  * Returns the Target Port Group identifier from the information
3373  * froom VPD page 0x83 of the device.
3374  *
3375  * Returns the identifier or error on failure.
3376  */
3377 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3378 {
3379 	const unsigned char *d;
3380 	const struct scsi_vpd *vpd_pg83;
3381 	int group_id = -EAGAIN, rel_port = -1;
3382 
3383 	rcu_read_lock();
3384 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3385 	if (!vpd_pg83) {
3386 		rcu_read_unlock();
3387 		return -ENXIO;
3388 	}
3389 
3390 	d = vpd_pg83->data + 4;
3391 	while (d < vpd_pg83->data + vpd_pg83->len) {
3392 		switch (d[1] & 0xf) {
3393 		case 0x4:
3394 			/* Relative target port */
3395 			rel_port = get_unaligned_be16(&d[6]);
3396 			break;
3397 		case 0x5:
3398 			/* Target port group */
3399 			group_id = get_unaligned_be16(&d[6]);
3400 			break;
3401 		default:
3402 			break;
3403 		}
3404 		d += d[3] + 4;
3405 	}
3406 	rcu_read_unlock();
3407 
3408 	if (group_id >= 0 && rel_id && rel_port != -1)
3409 		*rel_id = rel_port;
3410 
3411 	return group_id;
3412 }
3413 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3414 
3415 /**
3416  * scsi_build_sense - build sense data for a command
3417  * @scmd:	scsi command for which the sense should be formatted
3418  * @desc:	Sense format (non-zero == descriptor format,
3419  *              0 == fixed format)
3420  * @key:	Sense key
3421  * @asc:	Additional sense code
3422  * @ascq:	Additional sense code qualifier
3423  *
3424  **/
3425 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3426 {
3427 	scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3428 	scmd->result = SAM_STAT_CHECK_CONDITION;
3429 }
3430 EXPORT_SYMBOL_GPL(scsi_build_sense);
3431 
3432 #ifdef CONFIG_SCSI_LIB_KUNIT_TEST
3433 #include "scsi_lib_test.c"
3434 #endif
3435