xref: /linux/drivers/scsi/scsi_lib.c (revision 6ca80638b90cec66547011ee1ef79e534589989a)
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_queue() */
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 /**
188  * scsi_execute_cmd - insert request and wait for the result
189  * @sdev:	scsi_device
190  * @cmd:	scsi command
191  * @opf:	block layer request cmd_flags
192  * @buffer:	data buffer
193  * @bufflen:	len of buffer
194  * @timeout:	request timeout in HZ
195  * @retries:	number of times to retry request
196  * @args:	Optional args. See struct definition for field descriptions
197  *
198  * Returns the scsi_cmnd result field if a command was executed, or a negative
199  * Linux error code if we didn't get that far.
200  */
201 int scsi_execute_cmd(struct scsi_device *sdev, const unsigned char *cmd,
202 		     blk_opf_t opf, void *buffer, unsigned int bufflen,
203 		     int timeout, int retries,
204 		     const struct scsi_exec_args *args)
205 {
206 	static const struct scsi_exec_args default_args;
207 	struct request *req;
208 	struct scsi_cmnd *scmd;
209 	int ret;
210 
211 	if (!args)
212 		args = &default_args;
213 	else if (WARN_ON_ONCE(args->sense &&
214 			      args->sense_len != SCSI_SENSE_BUFFERSIZE))
215 		return -EINVAL;
216 
217 	req = scsi_alloc_request(sdev->request_queue, opf, args->req_flags);
218 	if (IS_ERR(req))
219 		return PTR_ERR(req);
220 
221 	if (bufflen) {
222 		ret = blk_rq_map_kern(sdev->request_queue, req,
223 				      buffer, bufflen, GFP_NOIO);
224 		if (ret)
225 			goto out;
226 	}
227 	scmd = blk_mq_rq_to_pdu(req);
228 	scmd->cmd_len = COMMAND_SIZE(cmd[0]);
229 	memcpy(scmd->cmnd, cmd, scmd->cmd_len);
230 	scmd->allowed = retries;
231 	scmd->flags |= args->scmd_flags;
232 	req->timeout = timeout;
233 	req->rq_flags |= RQF_QUIET;
234 
235 	/*
236 	 * head injection *required* here otherwise quiesce won't work
237 	 */
238 	blk_execute_rq(req, true);
239 
240 	/*
241 	 * Some devices (USB mass-storage in particular) may transfer
242 	 * garbage data together with a residue indicating that the data
243 	 * is invalid.  Prevent the garbage from being misinterpreted
244 	 * and prevent security leaks by zeroing out the excess data.
245 	 */
246 	if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
247 		memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);
248 
249 	if (args->resid)
250 		*args->resid = scmd->resid_len;
251 	if (args->sense)
252 		memcpy(args->sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
253 	if (args->sshdr)
254 		scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,
255 				     args->sshdr);
256 
257 	ret = scmd->result;
258  out:
259 	blk_mq_free_request(req);
260 
261 	return ret;
262 }
263 EXPORT_SYMBOL(scsi_execute_cmd);
264 
265 /*
266  * Wake up the error handler if necessary. Avoid as follows that the error
267  * handler is not woken up if host in-flight requests number ==
268  * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
269  * with an RCU read lock in this function to ensure that this function in
270  * its entirety either finishes before scsi_eh_scmd_add() increases the
271  * host_failed counter or that it notices the shost state change made by
272  * scsi_eh_scmd_add().
273  */
274 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
275 {
276 	unsigned long flags;
277 
278 	rcu_read_lock();
279 	__clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
280 	if (unlikely(scsi_host_in_recovery(shost))) {
281 		spin_lock_irqsave(shost->host_lock, flags);
282 		if (shost->host_failed || shost->host_eh_scheduled)
283 			scsi_eh_wakeup(shost);
284 		spin_unlock_irqrestore(shost->host_lock, flags);
285 	}
286 	rcu_read_unlock();
287 }
288 
289 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
290 {
291 	struct Scsi_Host *shost = sdev->host;
292 	struct scsi_target *starget = scsi_target(sdev);
293 
294 	scsi_dec_host_busy(shost, cmd);
295 
296 	if (starget->can_queue > 0)
297 		atomic_dec(&starget->target_busy);
298 
299 	sbitmap_put(&sdev->budget_map, cmd->budget_token);
300 	cmd->budget_token = -1;
301 }
302 
303 /*
304  * Kick the queue of SCSI device @sdev if @sdev != current_sdev. Called with
305  * interrupts disabled.
306  */
307 static void scsi_kick_sdev_queue(struct scsi_device *sdev, void *data)
308 {
309 	struct scsi_device *current_sdev = data;
310 
311 	if (sdev != current_sdev)
312 		blk_mq_run_hw_queues(sdev->request_queue, true);
313 }
314 
315 /*
316  * Called for single_lun devices on IO completion. Clear starget_sdev_user,
317  * and call blk_run_queue for all the scsi_devices on the target -
318  * including current_sdev first.
319  *
320  * Called with *no* scsi locks held.
321  */
322 static void scsi_single_lun_run(struct scsi_device *current_sdev)
323 {
324 	struct Scsi_Host *shost = current_sdev->host;
325 	struct scsi_target *starget = scsi_target(current_sdev);
326 	unsigned long flags;
327 
328 	spin_lock_irqsave(shost->host_lock, flags);
329 	starget->starget_sdev_user = NULL;
330 	spin_unlock_irqrestore(shost->host_lock, flags);
331 
332 	/*
333 	 * Call blk_run_queue for all LUNs on the target, starting with
334 	 * current_sdev. We race with others (to set starget_sdev_user),
335 	 * but in most cases, we will be first. Ideally, each LU on the
336 	 * target would get some limited time or requests on the target.
337 	 */
338 	blk_mq_run_hw_queues(current_sdev->request_queue,
339 			     shost->queuecommand_may_block);
340 
341 	spin_lock_irqsave(shost->host_lock, flags);
342 	if (!starget->starget_sdev_user)
343 		__starget_for_each_device(starget, current_sdev,
344 					  scsi_kick_sdev_queue);
345 	spin_unlock_irqrestore(shost->host_lock, flags);
346 }
347 
348 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
349 {
350 	if (scsi_device_busy(sdev) >= sdev->queue_depth)
351 		return true;
352 	if (atomic_read(&sdev->device_blocked) > 0)
353 		return true;
354 	return false;
355 }
356 
357 static inline bool scsi_target_is_busy(struct scsi_target *starget)
358 {
359 	if (starget->can_queue > 0) {
360 		if (atomic_read(&starget->target_busy) >= starget->can_queue)
361 			return true;
362 		if (atomic_read(&starget->target_blocked) > 0)
363 			return true;
364 	}
365 	return false;
366 }
367 
368 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
369 {
370 	if (atomic_read(&shost->host_blocked) > 0)
371 		return true;
372 	if (shost->host_self_blocked)
373 		return true;
374 	return false;
375 }
376 
377 static void scsi_starved_list_run(struct Scsi_Host *shost)
378 {
379 	LIST_HEAD(starved_list);
380 	struct scsi_device *sdev;
381 	unsigned long flags;
382 
383 	spin_lock_irqsave(shost->host_lock, flags);
384 	list_splice_init(&shost->starved_list, &starved_list);
385 
386 	while (!list_empty(&starved_list)) {
387 		struct request_queue *slq;
388 
389 		/*
390 		 * As long as shost is accepting commands and we have
391 		 * starved queues, call blk_run_queue. scsi_request_fn
392 		 * drops the queue_lock and can add us back to the
393 		 * starved_list.
394 		 *
395 		 * host_lock protects the starved_list and starved_entry.
396 		 * scsi_request_fn must get the host_lock before checking
397 		 * or modifying starved_list or starved_entry.
398 		 */
399 		if (scsi_host_is_busy(shost))
400 			break;
401 
402 		sdev = list_entry(starved_list.next,
403 				  struct scsi_device, starved_entry);
404 		list_del_init(&sdev->starved_entry);
405 		if (scsi_target_is_busy(scsi_target(sdev))) {
406 			list_move_tail(&sdev->starved_entry,
407 				       &shost->starved_list);
408 			continue;
409 		}
410 
411 		/*
412 		 * Once we drop the host lock, a racing scsi_remove_device()
413 		 * call may remove the sdev from the starved list and destroy
414 		 * it and the queue.  Mitigate by taking a reference to the
415 		 * queue and never touching the sdev again after we drop the
416 		 * host lock.  Note: if __scsi_remove_device() invokes
417 		 * blk_mq_destroy_queue() before the queue is run from this
418 		 * function then blk_run_queue() will return immediately since
419 		 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
420 		 */
421 		slq = sdev->request_queue;
422 		if (!blk_get_queue(slq))
423 			continue;
424 		spin_unlock_irqrestore(shost->host_lock, flags);
425 
426 		blk_mq_run_hw_queues(slq, false);
427 		blk_put_queue(slq);
428 
429 		spin_lock_irqsave(shost->host_lock, flags);
430 	}
431 	/* put any unprocessed entries back */
432 	list_splice(&starved_list, &shost->starved_list);
433 	spin_unlock_irqrestore(shost->host_lock, flags);
434 }
435 
436 /**
437  * scsi_run_queue - Select a proper request queue to serve next.
438  * @q:  last request's queue
439  *
440  * The previous command was completely finished, start a new one if possible.
441  */
442 static void scsi_run_queue(struct request_queue *q)
443 {
444 	struct scsi_device *sdev = q->queuedata;
445 
446 	if (scsi_target(sdev)->single_lun)
447 		scsi_single_lun_run(sdev);
448 	if (!list_empty(&sdev->host->starved_list))
449 		scsi_starved_list_run(sdev->host);
450 
451 	/* Note: blk_mq_kick_requeue_list() runs the queue asynchronously. */
452 	blk_mq_kick_requeue_list(q);
453 }
454 
455 void scsi_requeue_run_queue(struct work_struct *work)
456 {
457 	struct scsi_device *sdev;
458 	struct request_queue *q;
459 
460 	sdev = container_of(work, struct scsi_device, requeue_work);
461 	q = sdev->request_queue;
462 	scsi_run_queue(q);
463 }
464 
465 void scsi_run_host_queues(struct Scsi_Host *shost)
466 {
467 	struct scsi_device *sdev;
468 
469 	shost_for_each_device(sdev, shost)
470 		scsi_run_queue(sdev->request_queue);
471 }
472 
473 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
474 {
475 	if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
476 		struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
477 
478 		if (drv->uninit_command)
479 			drv->uninit_command(cmd);
480 	}
481 }
482 
483 void scsi_free_sgtables(struct scsi_cmnd *cmd)
484 {
485 	if (cmd->sdb.table.nents)
486 		sg_free_table_chained(&cmd->sdb.table,
487 				SCSI_INLINE_SG_CNT);
488 	if (scsi_prot_sg_count(cmd))
489 		sg_free_table_chained(&cmd->prot_sdb->table,
490 				SCSI_INLINE_PROT_SG_CNT);
491 }
492 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
493 
494 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
495 {
496 	scsi_free_sgtables(cmd);
497 	scsi_uninit_cmd(cmd);
498 }
499 
500 static void scsi_run_queue_async(struct scsi_device *sdev)
501 {
502 	if (scsi_host_in_recovery(sdev->host))
503 		return;
504 
505 	if (scsi_target(sdev)->single_lun ||
506 	    !list_empty(&sdev->host->starved_list)) {
507 		kblockd_schedule_work(&sdev->requeue_work);
508 	} else {
509 		/*
510 		 * smp_mb() present in sbitmap_queue_clear() or implied in
511 		 * .end_io is for ordering writing .device_busy in
512 		 * scsi_device_unbusy() and reading sdev->restarts.
513 		 */
514 		int old = atomic_read(&sdev->restarts);
515 
516 		/*
517 		 * ->restarts has to be kept as non-zero if new budget
518 		 *  contention occurs.
519 		 *
520 		 *  No need to run queue when either another re-run
521 		 *  queue wins in updating ->restarts or a new budget
522 		 *  contention occurs.
523 		 */
524 		if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
525 			blk_mq_run_hw_queues(sdev->request_queue, true);
526 	}
527 }
528 
529 /* Returns false when no more bytes to process, true if there are more */
530 static bool scsi_end_request(struct request *req, blk_status_t error,
531 		unsigned int bytes)
532 {
533 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
534 	struct scsi_device *sdev = cmd->device;
535 	struct request_queue *q = sdev->request_queue;
536 
537 	if (blk_update_request(req, error, bytes))
538 		return true;
539 
540 	// XXX:
541 	if (blk_queue_add_random(q))
542 		add_disk_randomness(req->q->disk);
543 
544 	if (!blk_rq_is_passthrough(req)) {
545 		WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
546 		cmd->flags &= ~SCMD_INITIALIZED;
547 	}
548 
549 	/*
550 	 * Calling rcu_barrier() is not necessary here because the
551 	 * SCSI error handler guarantees that the function called by
552 	 * call_rcu() has been called before scsi_end_request() is
553 	 * called.
554 	 */
555 	destroy_rcu_head(&cmd->rcu);
556 
557 	/*
558 	 * In the MQ case the command gets freed by __blk_mq_end_request,
559 	 * so we have to do all cleanup that depends on it earlier.
560 	 *
561 	 * We also can't kick the queues from irq context, so we
562 	 * will have to defer it to a workqueue.
563 	 */
564 	scsi_mq_uninit_cmd(cmd);
565 
566 	/*
567 	 * queue is still alive, so grab the ref for preventing it
568 	 * from being cleaned up during running queue.
569 	 */
570 	percpu_ref_get(&q->q_usage_counter);
571 
572 	__blk_mq_end_request(req, error);
573 
574 	scsi_run_queue_async(sdev);
575 
576 	percpu_ref_put(&q->q_usage_counter);
577 	return false;
578 }
579 
580 /**
581  * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
582  * @result:	scsi error code
583  *
584  * Translate a SCSI result code into a blk_status_t value.
585  */
586 static blk_status_t scsi_result_to_blk_status(int result)
587 {
588 	/*
589 	 * Check the scsi-ml byte first in case we converted a host or status
590 	 * byte.
591 	 */
592 	switch (scsi_ml_byte(result)) {
593 	case SCSIML_STAT_OK:
594 		break;
595 	case SCSIML_STAT_RESV_CONFLICT:
596 		return BLK_STS_RESV_CONFLICT;
597 	case SCSIML_STAT_NOSPC:
598 		return BLK_STS_NOSPC;
599 	case SCSIML_STAT_MED_ERROR:
600 		return BLK_STS_MEDIUM;
601 	case SCSIML_STAT_TGT_FAILURE:
602 		return BLK_STS_TARGET;
603 	case SCSIML_STAT_DL_TIMEOUT:
604 		return BLK_STS_DURATION_LIMIT;
605 	}
606 
607 	switch (host_byte(result)) {
608 	case DID_OK:
609 		if (scsi_status_is_good(result))
610 			return BLK_STS_OK;
611 		return BLK_STS_IOERR;
612 	case DID_TRANSPORT_FAILFAST:
613 	case DID_TRANSPORT_MARGINAL:
614 		return BLK_STS_TRANSPORT;
615 	default:
616 		return BLK_STS_IOERR;
617 	}
618 }
619 
620 /**
621  * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
622  * @rq: request to examine
623  *
624  * Description:
625  *     A request could be merge of IOs which require different failure
626  *     handling.  This function determines the number of bytes which
627  *     can be failed from the beginning of the request without
628  *     crossing into area which need to be retried further.
629  *
630  * Return:
631  *     The number of bytes to fail.
632  */
633 static unsigned int scsi_rq_err_bytes(const struct request *rq)
634 {
635 	blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
636 	unsigned int bytes = 0;
637 	struct bio *bio;
638 
639 	if (!(rq->rq_flags & RQF_MIXED_MERGE))
640 		return blk_rq_bytes(rq);
641 
642 	/*
643 	 * Currently the only 'mixing' which can happen is between
644 	 * different fastfail types.  We can safely fail portions
645 	 * which have all the failfast bits that the first one has -
646 	 * the ones which are at least as eager to fail as the first
647 	 * one.
648 	 */
649 	for (bio = rq->bio; bio; bio = bio->bi_next) {
650 		if ((bio->bi_opf & ff) != ff)
651 			break;
652 		bytes += bio->bi_iter.bi_size;
653 	}
654 
655 	/* this could lead to infinite loop */
656 	BUG_ON(blk_rq_bytes(rq) && !bytes);
657 	return bytes;
658 }
659 
660 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
661 {
662 	struct request *req = scsi_cmd_to_rq(cmd);
663 	unsigned long wait_for;
664 
665 	if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
666 		return false;
667 
668 	wait_for = (cmd->allowed + 1) * req->timeout;
669 	if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
670 		scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
671 			    wait_for/HZ);
672 		return true;
673 	}
674 	return false;
675 }
676 
677 /*
678  * When ALUA transition state is returned, reprep the cmd to
679  * use the ALUA handler's transition timeout. Delay the reprep
680  * 1 sec to avoid aggressive retries of the target in that
681  * state.
682  */
683 #define ALUA_TRANSITION_REPREP_DELAY	1000
684 
685 /* Helper for scsi_io_completion() when special action required. */
686 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
687 {
688 	struct request *req = scsi_cmd_to_rq(cmd);
689 	int level = 0;
690 	enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
691 	      ACTION_RETRY, ACTION_DELAYED_RETRY} action;
692 	struct scsi_sense_hdr sshdr;
693 	bool sense_valid;
694 	bool sense_current = true;      /* false implies "deferred sense" */
695 	blk_status_t blk_stat;
696 
697 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
698 	if (sense_valid)
699 		sense_current = !scsi_sense_is_deferred(&sshdr);
700 
701 	blk_stat = scsi_result_to_blk_status(result);
702 
703 	if (host_byte(result) == DID_RESET) {
704 		/* Third party bus reset or reset for error recovery
705 		 * reasons.  Just retry the command and see what
706 		 * happens.
707 		 */
708 		action = ACTION_RETRY;
709 	} else if (sense_valid && sense_current) {
710 		switch (sshdr.sense_key) {
711 		case UNIT_ATTENTION:
712 			if (cmd->device->removable) {
713 				/* Detected disc change.  Set a bit
714 				 * and quietly refuse further access.
715 				 */
716 				cmd->device->changed = 1;
717 				action = ACTION_FAIL;
718 			} else {
719 				/* Must have been a power glitch, or a
720 				 * bus reset.  Could not have been a
721 				 * media change, so we just retry the
722 				 * command and see what happens.
723 				 */
724 				action = ACTION_RETRY;
725 			}
726 			break;
727 		case ILLEGAL_REQUEST:
728 			/* If we had an ILLEGAL REQUEST returned, then
729 			 * we may have performed an unsupported
730 			 * command.  The only thing this should be
731 			 * would be a ten byte read where only a six
732 			 * byte read was supported.  Also, on a system
733 			 * where READ CAPACITY failed, we may have
734 			 * read past the end of the disk.
735 			 */
736 			if ((cmd->device->use_10_for_rw &&
737 			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
738 			    (cmd->cmnd[0] == READ_10 ||
739 			     cmd->cmnd[0] == WRITE_10)) {
740 				/* This will issue a new 6-byte command. */
741 				cmd->device->use_10_for_rw = 0;
742 				action = ACTION_REPREP;
743 			} else if (sshdr.asc == 0x10) /* DIX */ {
744 				action = ACTION_FAIL;
745 				blk_stat = BLK_STS_PROTECTION;
746 			/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
747 			} else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
748 				action = ACTION_FAIL;
749 				blk_stat = BLK_STS_TARGET;
750 			} else
751 				action = ACTION_FAIL;
752 			break;
753 		case ABORTED_COMMAND:
754 			action = ACTION_FAIL;
755 			if (sshdr.asc == 0x10) /* DIF */
756 				blk_stat = BLK_STS_PROTECTION;
757 			break;
758 		case NOT_READY:
759 			/* If the device is in the process of becoming
760 			 * ready, or has a temporary blockage, retry.
761 			 */
762 			if (sshdr.asc == 0x04) {
763 				switch (sshdr.ascq) {
764 				case 0x01: /* becoming ready */
765 				case 0x04: /* format in progress */
766 				case 0x05: /* rebuild in progress */
767 				case 0x06: /* recalculation in progress */
768 				case 0x07: /* operation in progress */
769 				case 0x08: /* Long write in progress */
770 				case 0x09: /* self test in progress */
771 				case 0x11: /* notify (enable spinup) required */
772 				case 0x14: /* space allocation in progress */
773 				case 0x1a: /* start stop unit in progress */
774 				case 0x1b: /* sanitize in progress */
775 				case 0x1d: /* configuration in progress */
776 				case 0x24: /* depopulation in progress */
777 					action = ACTION_DELAYED_RETRY;
778 					break;
779 				case 0x0a: /* ALUA state transition */
780 					action = ACTION_DELAYED_REPREP;
781 					break;
782 				default:
783 					action = ACTION_FAIL;
784 					break;
785 				}
786 			} else
787 				action = ACTION_FAIL;
788 			break;
789 		case VOLUME_OVERFLOW:
790 			/* See SSC3rXX or current. */
791 			action = ACTION_FAIL;
792 			break;
793 		case DATA_PROTECT:
794 			action = ACTION_FAIL;
795 			if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
796 			    (sshdr.asc == 0x55 &&
797 			     (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
798 				/* Insufficient zone resources */
799 				blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
800 			}
801 			break;
802 		case COMPLETED:
803 			fallthrough;
804 		default:
805 			action = ACTION_FAIL;
806 			break;
807 		}
808 	} else
809 		action = ACTION_FAIL;
810 
811 	if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
812 		action = ACTION_FAIL;
813 
814 	switch (action) {
815 	case ACTION_FAIL:
816 		/* Give up and fail the remainder of the request */
817 		if (!(req->rq_flags & RQF_QUIET)) {
818 			static DEFINE_RATELIMIT_STATE(_rs,
819 					DEFAULT_RATELIMIT_INTERVAL,
820 					DEFAULT_RATELIMIT_BURST);
821 
822 			if (unlikely(scsi_logging_level))
823 				level =
824 				     SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
825 						    SCSI_LOG_MLCOMPLETE_BITS);
826 
827 			/*
828 			 * if logging is enabled the failure will be printed
829 			 * in scsi_log_completion(), so avoid duplicate messages
830 			 */
831 			if (!level && __ratelimit(&_rs)) {
832 				scsi_print_result(cmd, NULL, FAILED);
833 				if (sense_valid)
834 					scsi_print_sense(cmd);
835 				scsi_print_command(cmd);
836 			}
837 		}
838 		if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
839 			return;
840 		fallthrough;
841 	case ACTION_REPREP:
842 		scsi_mq_requeue_cmd(cmd, 0);
843 		break;
844 	case ACTION_DELAYED_REPREP:
845 		scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
846 		break;
847 	case ACTION_RETRY:
848 		/* Retry the same command immediately */
849 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
850 		break;
851 	case ACTION_DELAYED_RETRY:
852 		/* Retry the same command after a delay */
853 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
854 		break;
855 	}
856 }
857 
858 /*
859  * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
860  * new result that may suppress further error checking. Also modifies
861  * *blk_statp in some cases.
862  */
863 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
864 					blk_status_t *blk_statp)
865 {
866 	bool sense_valid;
867 	bool sense_current = true;	/* false implies "deferred sense" */
868 	struct request *req = scsi_cmd_to_rq(cmd);
869 	struct scsi_sense_hdr sshdr;
870 
871 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
872 	if (sense_valid)
873 		sense_current = !scsi_sense_is_deferred(&sshdr);
874 
875 	if (blk_rq_is_passthrough(req)) {
876 		if (sense_valid) {
877 			/*
878 			 * SG_IO wants current and deferred errors
879 			 */
880 			cmd->sense_len = min(8 + cmd->sense_buffer[7],
881 					     SCSI_SENSE_BUFFERSIZE);
882 		}
883 		if (sense_current)
884 			*blk_statp = scsi_result_to_blk_status(result);
885 	} else if (blk_rq_bytes(req) == 0 && sense_current) {
886 		/*
887 		 * Flush commands do not transfers any data, and thus cannot use
888 		 * good_bytes != blk_rq_bytes(req) as the signal for an error.
889 		 * This sets *blk_statp explicitly for the problem case.
890 		 */
891 		*blk_statp = scsi_result_to_blk_status(result);
892 	}
893 	/*
894 	 * Recovered errors need reporting, but they're always treated as
895 	 * success, so fiddle the result code here.  For passthrough requests
896 	 * we already took a copy of the original into sreq->result which
897 	 * is what gets returned to the user
898 	 */
899 	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
900 		bool do_print = true;
901 		/*
902 		 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
903 		 * skip print since caller wants ATA registers. Only occurs
904 		 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
905 		 */
906 		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
907 			do_print = false;
908 		else if (req->rq_flags & RQF_QUIET)
909 			do_print = false;
910 		if (do_print)
911 			scsi_print_sense(cmd);
912 		result = 0;
913 		/* for passthrough, *blk_statp may be set */
914 		*blk_statp = BLK_STS_OK;
915 	}
916 	/*
917 	 * Another corner case: the SCSI status byte is non-zero but 'good'.
918 	 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
919 	 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
920 	 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
921 	 * intermediate statuses (both obsolete in SAM-4) as good.
922 	 */
923 	if ((result & 0xff) && scsi_status_is_good(result)) {
924 		result = 0;
925 		*blk_statp = BLK_STS_OK;
926 	}
927 	return result;
928 }
929 
930 /**
931  * scsi_io_completion - Completion processing for SCSI commands.
932  * @cmd:	command that is finished.
933  * @good_bytes:	number of processed bytes.
934  *
935  * We will finish off the specified number of sectors. If we are done, the
936  * command block will be released and the queue function will be goosed. If we
937  * are not done then we have to figure out what to do next:
938  *
939  *   a) We can call scsi_mq_requeue_cmd().  The request will be
940  *	unprepared and put back on the queue.  Then a new command will
941  *	be created for it.  This should be used if we made forward
942  *	progress, or if we want to switch from READ(10) to READ(6) for
943  *	example.
944  *
945  *   b) We can call scsi_io_completion_action().  The request will be
946  *	put back on the queue and retried using the same command as
947  *	before, possibly after a delay.
948  *
949  *   c) We can call scsi_end_request() with blk_stat other than
950  *	BLK_STS_OK, to fail the remainder of the request.
951  */
952 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
953 {
954 	int result = cmd->result;
955 	struct request *req = scsi_cmd_to_rq(cmd);
956 	blk_status_t blk_stat = BLK_STS_OK;
957 
958 	if (unlikely(result))	/* a nz result may or may not be an error */
959 		result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
960 
961 	/*
962 	 * Next deal with any sectors which we were able to correctly
963 	 * handle.
964 	 */
965 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
966 		"%u sectors total, %d bytes done.\n",
967 		blk_rq_sectors(req), good_bytes));
968 
969 	/*
970 	 * Failed, zero length commands always need to drop down
971 	 * to retry code. Fast path should return in this block.
972 	 */
973 	if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
974 		if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
975 			return; /* no bytes remaining */
976 	}
977 
978 	/* Kill remainder if no retries. */
979 	if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
980 		if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
981 			WARN_ONCE(true,
982 			    "Bytes remaining after failed, no-retry command");
983 		return;
984 	}
985 
986 	/*
987 	 * If there had been no error, but we have leftover bytes in the
988 	 * request just queue the command up again.
989 	 */
990 	if (likely(result == 0))
991 		scsi_mq_requeue_cmd(cmd, 0);
992 	else
993 		scsi_io_completion_action(cmd, result);
994 }
995 
996 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
997 		struct request *rq)
998 {
999 	return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1000 	       !op_is_write(req_op(rq)) &&
1001 	       sdev->host->hostt->dma_need_drain(rq);
1002 }
1003 
1004 /**
1005  * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1006  * @cmd: SCSI command data structure to initialize.
1007  *
1008  * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1009  * for @cmd.
1010  *
1011  * Returns:
1012  * * BLK_STS_OK       - on success
1013  * * BLK_STS_RESOURCE - if the failure is retryable
1014  * * BLK_STS_IOERR    - if the failure is fatal
1015  */
1016 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1017 {
1018 	struct scsi_device *sdev = cmd->device;
1019 	struct request *rq = scsi_cmd_to_rq(cmd);
1020 	unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1021 	struct scatterlist *last_sg = NULL;
1022 	blk_status_t ret;
1023 	bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1024 	int count;
1025 
1026 	if (WARN_ON_ONCE(!nr_segs))
1027 		return BLK_STS_IOERR;
1028 
1029 	/*
1030 	 * Make sure there is space for the drain.  The driver must adjust
1031 	 * max_hw_segments to be prepared for this.
1032 	 */
1033 	if (need_drain)
1034 		nr_segs++;
1035 
1036 	/*
1037 	 * If sg table allocation fails, requeue request later.
1038 	 */
1039 	if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1040 			cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1041 		return BLK_STS_RESOURCE;
1042 
1043 	/*
1044 	 * Next, walk the list, and fill in the addresses and sizes of
1045 	 * each segment.
1046 	 */
1047 	count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1048 
1049 	if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1050 		unsigned int pad_len =
1051 			(rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1052 
1053 		last_sg->length += pad_len;
1054 		cmd->extra_len += pad_len;
1055 	}
1056 
1057 	if (need_drain) {
1058 		sg_unmark_end(last_sg);
1059 		last_sg = sg_next(last_sg);
1060 		sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1061 		sg_mark_end(last_sg);
1062 
1063 		cmd->extra_len += sdev->dma_drain_len;
1064 		count++;
1065 	}
1066 
1067 	BUG_ON(count > cmd->sdb.table.nents);
1068 	cmd->sdb.table.nents = count;
1069 	cmd->sdb.length = blk_rq_payload_bytes(rq);
1070 
1071 	if (blk_integrity_rq(rq)) {
1072 		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1073 		int ivecs;
1074 
1075 		if (WARN_ON_ONCE(!prot_sdb)) {
1076 			/*
1077 			 * This can happen if someone (e.g. multipath)
1078 			 * queues a command to a device on an adapter
1079 			 * that does not support DIX.
1080 			 */
1081 			ret = BLK_STS_IOERR;
1082 			goto out_free_sgtables;
1083 		}
1084 
1085 		ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1086 
1087 		if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1088 				prot_sdb->table.sgl,
1089 				SCSI_INLINE_PROT_SG_CNT)) {
1090 			ret = BLK_STS_RESOURCE;
1091 			goto out_free_sgtables;
1092 		}
1093 
1094 		count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1095 						prot_sdb->table.sgl);
1096 		BUG_ON(count > ivecs);
1097 		BUG_ON(count > queue_max_integrity_segments(rq->q));
1098 
1099 		cmd->prot_sdb = prot_sdb;
1100 		cmd->prot_sdb->table.nents = count;
1101 	}
1102 
1103 	return BLK_STS_OK;
1104 out_free_sgtables:
1105 	scsi_free_sgtables(cmd);
1106 	return ret;
1107 }
1108 EXPORT_SYMBOL(scsi_alloc_sgtables);
1109 
1110 /**
1111  * scsi_initialize_rq - initialize struct scsi_cmnd partially
1112  * @rq: Request associated with the SCSI command to be initialized.
1113  *
1114  * This function initializes the members of struct scsi_cmnd that must be
1115  * initialized before request processing starts and that won't be
1116  * reinitialized if a SCSI command is requeued.
1117  */
1118 static void scsi_initialize_rq(struct request *rq)
1119 {
1120 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1121 
1122 	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1123 	cmd->cmd_len = MAX_COMMAND_SIZE;
1124 	cmd->sense_len = 0;
1125 	init_rcu_head(&cmd->rcu);
1126 	cmd->jiffies_at_alloc = jiffies;
1127 	cmd->retries = 0;
1128 }
1129 
1130 struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1131 				   blk_mq_req_flags_t flags)
1132 {
1133 	struct request *rq;
1134 
1135 	rq = blk_mq_alloc_request(q, opf, flags);
1136 	if (!IS_ERR(rq))
1137 		scsi_initialize_rq(rq);
1138 	return rq;
1139 }
1140 EXPORT_SYMBOL_GPL(scsi_alloc_request);
1141 
1142 /*
1143  * Only called when the request isn't completed by SCSI, and not freed by
1144  * SCSI
1145  */
1146 static void scsi_cleanup_rq(struct request *rq)
1147 {
1148 	if (rq->rq_flags & RQF_DONTPREP) {
1149 		scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1150 		rq->rq_flags &= ~RQF_DONTPREP;
1151 	}
1152 }
1153 
1154 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1155 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1156 {
1157 	struct request *rq = scsi_cmd_to_rq(cmd);
1158 
1159 	if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1160 		cmd->flags |= SCMD_INITIALIZED;
1161 		scsi_initialize_rq(rq);
1162 	}
1163 
1164 	cmd->device = dev;
1165 	INIT_LIST_HEAD(&cmd->eh_entry);
1166 	INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1167 }
1168 
1169 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1170 		struct request *req)
1171 {
1172 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1173 
1174 	/*
1175 	 * Passthrough requests may transfer data, in which case they must
1176 	 * a bio attached to them.  Or they might contain a SCSI command
1177 	 * that does not transfer data, in which case they may optionally
1178 	 * submit a request without an attached bio.
1179 	 */
1180 	if (req->bio) {
1181 		blk_status_t ret = scsi_alloc_sgtables(cmd);
1182 		if (unlikely(ret != BLK_STS_OK))
1183 			return ret;
1184 	} else {
1185 		BUG_ON(blk_rq_bytes(req));
1186 
1187 		memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1188 	}
1189 
1190 	cmd->transfersize = blk_rq_bytes(req);
1191 	return BLK_STS_OK;
1192 }
1193 
1194 static blk_status_t
1195 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1196 {
1197 	switch (sdev->sdev_state) {
1198 	case SDEV_CREATED:
1199 		return BLK_STS_OK;
1200 	case SDEV_OFFLINE:
1201 	case SDEV_TRANSPORT_OFFLINE:
1202 		/*
1203 		 * If the device is offline we refuse to process any
1204 		 * commands.  The device must be brought online
1205 		 * before trying any recovery commands.
1206 		 */
1207 		if (!sdev->offline_already) {
1208 			sdev->offline_already = true;
1209 			sdev_printk(KERN_ERR, sdev,
1210 				    "rejecting I/O to offline device\n");
1211 		}
1212 		return BLK_STS_IOERR;
1213 	case SDEV_DEL:
1214 		/*
1215 		 * If the device is fully deleted, we refuse to
1216 		 * process any commands as well.
1217 		 */
1218 		sdev_printk(KERN_ERR, sdev,
1219 			    "rejecting I/O to dead device\n");
1220 		return BLK_STS_IOERR;
1221 	case SDEV_BLOCK:
1222 	case SDEV_CREATED_BLOCK:
1223 		return BLK_STS_RESOURCE;
1224 	case SDEV_QUIESCE:
1225 		/*
1226 		 * If the device is blocked we only accept power management
1227 		 * commands.
1228 		 */
1229 		if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1230 			return BLK_STS_RESOURCE;
1231 		return BLK_STS_OK;
1232 	default:
1233 		/*
1234 		 * For any other not fully online state we only allow
1235 		 * power management commands.
1236 		 */
1237 		if (req && !(req->rq_flags & RQF_PM))
1238 			return BLK_STS_OFFLINE;
1239 		return BLK_STS_OK;
1240 	}
1241 }
1242 
1243 /*
1244  * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1245  * and return the token else return -1.
1246  */
1247 static inline int scsi_dev_queue_ready(struct request_queue *q,
1248 				  struct scsi_device *sdev)
1249 {
1250 	int token;
1251 
1252 	token = sbitmap_get(&sdev->budget_map);
1253 	if (atomic_read(&sdev->device_blocked)) {
1254 		if (token < 0)
1255 			goto out;
1256 
1257 		if (scsi_device_busy(sdev) > 1)
1258 			goto out_dec;
1259 
1260 		/*
1261 		 * unblock after device_blocked iterates to zero
1262 		 */
1263 		if (atomic_dec_return(&sdev->device_blocked) > 0)
1264 			goto out_dec;
1265 		SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1266 				   "unblocking device at zero depth\n"));
1267 	}
1268 
1269 	return token;
1270 out_dec:
1271 	if (token >= 0)
1272 		sbitmap_put(&sdev->budget_map, token);
1273 out:
1274 	return -1;
1275 }
1276 
1277 /*
1278  * scsi_target_queue_ready: checks if there we can send commands to target
1279  * @sdev: scsi device on starget to check.
1280  */
1281 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1282 					   struct scsi_device *sdev)
1283 {
1284 	struct scsi_target *starget = scsi_target(sdev);
1285 	unsigned int busy;
1286 
1287 	if (starget->single_lun) {
1288 		spin_lock_irq(shost->host_lock);
1289 		if (starget->starget_sdev_user &&
1290 		    starget->starget_sdev_user != sdev) {
1291 			spin_unlock_irq(shost->host_lock);
1292 			return 0;
1293 		}
1294 		starget->starget_sdev_user = sdev;
1295 		spin_unlock_irq(shost->host_lock);
1296 	}
1297 
1298 	if (starget->can_queue <= 0)
1299 		return 1;
1300 
1301 	busy = atomic_inc_return(&starget->target_busy) - 1;
1302 	if (atomic_read(&starget->target_blocked) > 0) {
1303 		if (busy)
1304 			goto starved;
1305 
1306 		/*
1307 		 * unblock after target_blocked iterates to zero
1308 		 */
1309 		if (atomic_dec_return(&starget->target_blocked) > 0)
1310 			goto out_dec;
1311 
1312 		SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1313 				 "unblocking target at zero depth\n"));
1314 	}
1315 
1316 	if (busy >= starget->can_queue)
1317 		goto starved;
1318 
1319 	return 1;
1320 
1321 starved:
1322 	spin_lock_irq(shost->host_lock);
1323 	list_move_tail(&sdev->starved_entry, &shost->starved_list);
1324 	spin_unlock_irq(shost->host_lock);
1325 out_dec:
1326 	if (starget->can_queue > 0)
1327 		atomic_dec(&starget->target_busy);
1328 	return 0;
1329 }
1330 
1331 /*
1332  * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1333  * return 0. We must end up running the queue again whenever 0 is
1334  * returned, else IO can hang.
1335  */
1336 static inline int scsi_host_queue_ready(struct request_queue *q,
1337 				   struct Scsi_Host *shost,
1338 				   struct scsi_device *sdev,
1339 				   struct scsi_cmnd *cmd)
1340 {
1341 	if (atomic_read(&shost->host_blocked) > 0) {
1342 		if (scsi_host_busy(shost) > 0)
1343 			goto starved;
1344 
1345 		/*
1346 		 * unblock after host_blocked iterates to zero
1347 		 */
1348 		if (atomic_dec_return(&shost->host_blocked) > 0)
1349 			goto out_dec;
1350 
1351 		SCSI_LOG_MLQUEUE(3,
1352 			shost_printk(KERN_INFO, shost,
1353 				     "unblocking host at zero depth\n"));
1354 	}
1355 
1356 	if (shost->host_self_blocked)
1357 		goto starved;
1358 
1359 	/* We're OK to process the command, so we can't be starved */
1360 	if (!list_empty(&sdev->starved_entry)) {
1361 		spin_lock_irq(shost->host_lock);
1362 		if (!list_empty(&sdev->starved_entry))
1363 			list_del_init(&sdev->starved_entry);
1364 		spin_unlock_irq(shost->host_lock);
1365 	}
1366 
1367 	__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1368 
1369 	return 1;
1370 
1371 starved:
1372 	spin_lock_irq(shost->host_lock);
1373 	if (list_empty(&sdev->starved_entry))
1374 		list_add_tail(&sdev->starved_entry, &shost->starved_list);
1375 	spin_unlock_irq(shost->host_lock);
1376 out_dec:
1377 	scsi_dec_host_busy(shost, cmd);
1378 	return 0;
1379 }
1380 
1381 /*
1382  * Busy state exporting function for request stacking drivers.
1383  *
1384  * For efficiency, no lock is taken to check the busy state of
1385  * shost/starget/sdev, since the returned value is not guaranteed and
1386  * may be changed after request stacking drivers call the function,
1387  * regardless of taking lock or not.
1388  *
1389  * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1390  * needs to return 'not busy'. Otherwise, request stacking drivers
1391  * may hold requests forever.
1392  */
1393 static bool scsi_mq_lld_busy(struct request_queue *q)
1394 {
1395 	struct scsi_device *sdev = q->queuedata;
1396 	struct Scsi_Host *shost;
1397 
1398 	if (blk_queue_dying(q))
1399 		return false;
1400 
1401 	shost = sdev->host;
1402 
1403 	/*
1404 	 * Ignore host/starget busy state.
1405 	 * Since block layer does not have a concept of fairness across
1406 	 * multiple queues, congestion of host/starget needs to be handled
1407 	 * in SCSI layer.
1408 	 */
1409 	if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1410 		return true;
1411 
1412 	return false;
1413 }
1414 
1415 /*
1416  * Block layer request completion callback. May be called from interrupt
1417  * context.
1418  */
1419 static void scsi_complete(struct request *rq)
1420 {
1421 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1422 	enum scsi_disposition disposition;
1423 
1424 	INIT_LIST_HEAD(&cmd->eh_entry);
1425 
1426 	atomic_inc(&cmd->device->iodone_cnt);
1427 	if (cmd->result)
1428 		atomic_inc(&cmd->device->ioerr_cnt);
1429 
1430 	disposition = scsi_decide_disposition(cmd);
1431 	if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1432 		disposition = SUCCESS;
1433 
1434 	scsi_log_completion(cmd, disposition);
1435 
1436 	switch (disposition) {
1437 	case SUCCESS:
1438 		scsi_finish_command(cmd);
1439 		break;
1440 	case NEEDS_RETRY:
1441 		scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1442 		break;
1443 	case ADD_TO_MLQUEUE:
1444 		scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1445 		break;
1446 	default:
1447 		scsi_eh_scmd_add(cmd);
1448 		break;
1449 	}
1450 }
1451 
1452 /**
1453  * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1454  * @cmd: command block we are dispatching.
1455  *
1456  * Return: nonzero return request was rejected and device's queue needs to be
1457  * plugged.
1458  */
1459 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1460 {
1461 	struct Scsi_Host *host = cmd->device->host;
1462 	int rtn = 0;
1463 
1464 	atomic_inc(&cmd->device->iorequest_cnt);
1465 
1466 	/* check if the device is still usable */
1467 	if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1468 		/* in SDEV_DEL we error all commands. DID_NO_CONNECT
1469 		 * returns an immediate error upwards, and signals
1470 		 * that the device is no longer present */
1471 		cmd->result = DID_NO_CONNECT << 16;
1472 		goto done;
1473 	}
1474 
1475 	/* Check to see if the scsi lld made this device blocked. */
1476 	if (unlikely(scsi_device_blocked(cmd->device))) {
1477 		/*
1478 		 * in blocked state, the command is just put back on
1479 		 * the device queue.  The suspend state has already
1480 		 * blocked the queue so future requests should not
1481 		 * occur until the device transitions out of the
1482 		 * suspend state.
1483 		 */
1484 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1485 			"queuecommand : device blocked\n"));
1486 		atomic_dec(&cmd->device->iorequest_cnt);
1487 		return SCSI_MLQUEUE_DEVICE_BUSY;
1488 	}
1489 
1490 	/* Store the LUN value in cmnd, if needed. */
1491 	if (cmd->device->lun_in_cdb)
1492 		cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1493 			       (cmd->device->lun << 5 & 0xe0);
1494 
1495 	scsi_log_send(cmd);
1496 
1497 	/*
1498 	 * Before we queue this command, check if the command
1499 	 * length exceeds what the host adapter can handle.
1500 	 */
1501 	if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1502 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1503 			       "queuecommand : command too long. "
1504 			       "cdb_size=%d host->max_cmd_len=%d\n",
1505 			       cmd->cmd_len, cmd->device->host->max_cmd_len));
1506 		cmd->result = (DID_ABORT << 16);
1507 		goto done;
1508 	}
1509 
1510 	if (unlikely(host->shost_state == SHOST_DEL)) {
1511 		cmd->result = (DID_NO_CONNECT << 16);
1512 		goto done;
1513 
1514 	}
1515 
1516 	trace_scsi_dispatch_cmd_start(cmd);
1517 	rtn = host->hostt->queuecommand(host, cmd);
1518 	if (rtn) {
1519 		atomic_dec(&cmd->device->iorequest_cnt);
1520 		trace_scsi_dispatch_cmd_error(cmd, rtn);
1521 		if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1522 		    rtn != SCSI_MLQUEUE_TARGET_BUSY)
1523 			rtn = SCSI_MLQUEUE_HOST_BUSY;
1524 
1525 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1526 			"queuecommand : request rejected\n"));
1527 	}
1528 
1529 	return rtn;
1530  done:
1531 	scsi_done(cmd);
1532 	return 0;
1533 }
1534 
1535 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1536 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1537 {
1538 	return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1539 		sizeof(struct scatterlist);
1540 }
1541 
1542 static blk_status_t scsi_prepare_cmd(struct request *req)
1543 {
1544 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1545 	struct scsi_device *sdev = req->q->queuedata;
1546 	struct Scsi_Host *shost = sdev->host;
1547 	bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1548 	struct scatterlist *sg;
1549 
1550 	scsi_init_command(sdev, cmd);
1551 
1552 	cmd->eh_eflags = 0;
1553 	cmd->prot_type = 0;
1554 	cmd->prot_flags = 0;
1555 	cmd->submitter = 0;
1556 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1557 	cmd->underflow = 0;
1558 	cmd->transfersize = 0;
1559 	cmd->host_scribble = NULL;
1560 	cmd->result = 0;
1561 	cmd->extra_len = 0;
1562 	cmd->state = 0;
1563 	if (in_flight)
1564 		__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1565 
1566 	/*
1567 	 * Only clear the driver-private command data if the LLD does not supply
1568 	 * a function to initialize that data.
1569 	 */
1570 	if (!shost->hostt->init_cmd_priv)
1571 		memset(cmd + 1, 0, shost->hostt->cmd_size);
1572 
1573 	cmd->prot_op = SCSI_PROT_NORMAL;
1574 	if (blk_rq_bytes(req))
1575 		cmd->sc_data_direction = rq_dma_dir(req);
1576 	else
1577 		cmd->sc_data_direction = DMA_NONE;
1578 
1579 	sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1580 	cmd->sdb.table.sgl = sg;
1581 
1582 	if (scsi_host_get_prot(shost)) {
1583 		memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1584 
1585 		cmd->prot_sdb->table.sgl =
1586 			(struct scatterlist *)(cmd->prot_sdb + 1);
1587 	}
1588 
1589 	/*
1590 	 * Special handling for passthrough commands, which don't go to the ULP
1591 	 * at all:
1592 	 */
1593 	if (blk_rq_is_passthrough(req))
1594 		return scsi_setup_scsi_cmnd(sdev, req);
1595 
1596 	if (sdev->handler && sdev->handler->prep_fn) {
1597 		blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1598 
1599 		if (ret != BLK_STS_OK)
1600 			return ret;
1601 	}
1602 
1603 	/* Usually overridden by the ULP */
1604 	cmd->allowed = 0;
1605 	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1606 	return scsi_cmd_to_driver(cmd)->init_command(cmd);
1607 }
1608 
1609 static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1610 {
1611 	struct request *req = scsi_cmd_to_rq(cmd);
1612 
1613 	switch (cmd->submitter) {
1614 	case SUBMITTED_BY_BLOCK_LAYER:
1615 		break;
1616 	case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1617 		return scsi_eh_done(cmd);
1618 	case SUBMITTED_BY_SCSI_RESET_IOCTL:
1619 		return;
1620 	}
1621 
1622 	if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1623 		return;
1624 	if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1625 		return;
1626 	trace_scsi_dispatch_cmd_done(cmd);
1627 
1628 	if (complete_directly)
1629 		blk_mq_complete_request_direct(req, scsi_complete);
1630 	else
1631 		blk_mq_complete_request(req);
1632 }
1633 
1634 void scsi_done(struct scsi_cmnd *cmd)
1635 {
1636 	scsi_done_internal(cmd, false);
1637 }
1638 EXPORT_SYMBOL(scsi_done);
1639 
1640 void scsi_done_direct(struct scsi_cmnd *cmd)
1641 {
1642 	scsi_done_internal(cmd, true);
1643 }
1644 EXPORT_SYMBOL(scsi_done_direct);
1645 
1646 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1647 {
1648 	struct scsi_device *sdev = q->queuedata;
1649 
1650 	sbitmap_put(&sdev->budget_map, budget_token);
1651 }
1652 
1653 /*
1654  * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1655  * not change behaviour from the previous unplug mechanism, experimentation
1656  * may prove this needs changing.
1657  */
1658 #define SCSI_QUEUE_DELAY 3
1659 
1660 static int scsi_mq_get_budget(struct request_queue *q)
1661 {
1662 	struct scsi_device *sdev = q->queuedata;
1663 	int token = scsi_dev_queue_ready(q, sdev);
1664 
1665 	if (token >= 0)
1666 		return token;
1667 
1668 	atomic_inc(&sdev->restarts);
1669 
1670 	/*
1671 	 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1672 	 * .restarts must be incremented before .device_busy is read because the
1673 	 * code in scsi_run_queue_async() depends on the order of these operations.
1674 	 */
1675 	smp_mb__after_atomic();
1676 
1677 	/*
1678 	 * If all in-flight requests originated from this LUN are completed
1679 	 * before reading .device_busy, sdev->device_busy will be observed as
1680 	 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1681 	 * soon. Otherwise, completion of one of these requests will observe
1682 	 * the .restarts flag, and the request queue will be run for handling
1683 	 * this request, see scsi_end_request().
1684 	 */
1685 	if (unlikely(scsi_device_busy(sdev) == 0 &&
1686 				!scsi_device_blocked(sdev)))
1687 		blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1688 	return -1;
1689 }
1690 
1691 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1692 {
1693 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1694 
1695 	cmd->budget_token = token;
1696 }
1697 
1698 static int scsi_mq_get_rq_budget_token(struct request *req)
1699 {
1700 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1701 
1702 	return cmd->budget_token;
1703 }
1704 
1705 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1706 			 const struct blk_mq_queue_data *bd)
1707 {
1708 	struct request *req = bd->rq;
1709 	struct request_queue *q = req->q;
1710 	struct scsi_device *sdev = q->queuedata;
1711 	struct Scsi_Host *shost = sdev->host;
1712 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1713 	blk_status_t ret;
1714 	int reason;
1715 
1716 	WARN_ON_ONCE(cmd->budget_token < 0);
1717 
1718 	/*
1719 	 * If the device is not in running state we will reject some or all
1720 	 * commands.
1721 	 */
1722 	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1723 		ret = scsi_device_state_check(sdev, req);
1724 		if (ret != BLK_STS_OK)
1725 			goto out_put_budget;
1726 	}
1727 
1728 	ret = BLK_STS_RESOURCE;
1729 	if (!scsi_target_queue_ready(shost, sdev))
1730 		goto out_put_budget;
1731 	if (unlikely(scsi_host_in_recovery(shost))) {
1732 		if (cmd->flags & SCMD_FAIL_IF_RECOVERING)
1733 			ret = BLK_STS_OFFLINE;
1734 		goto out_dec_target_busy;
1735 	}
1736 	if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1737 		goto out_dec_target_busy;
1738 
1739 	if (!(req->rq_flags & RQF_DONTPREP)) {
1740 		ret = scsi_prepare_cmd(req);
1741 		if (ret != BLK_STS_OK)
1742 			goto out_dec_host_busy;
1743 		req->rq_flags |= RQF_DONTPREP;
1744 	} else {
1745 		clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1746 	}
1747 
1748 	cmd->flags &= SCMD_PRESERVED_FLAGS;
1749 	if (sdev->simple_tags)
1750 		cmd->flags |= SCMD_TAGGED;
1751 	if (bd->last)
1752 		cmd->flags |= SCMD_LAST;
1753 
1754 	scsi_set_resid(cmd, 0);
1755 	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1756 	cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1757 
1758 	blk_mq_start_request(req);
1759 	reason = scsi_dispatch_cmd(cmd);
1760 	if (reason) {
1761 		scsi_set_blocked(cmd, reason);
1762 		ret = BLK_STS_RESOURCE;
1763 		goto out_dec_host_busy;
1764 	}
1765 
1766 	return BLK_STS_OK;
1767 
1768 out_dec_host_busy:
1769 	scsi_dec_host_busy(shost, cmd);
1770 out_dec_target_busy:
1771 	if (scsi_target(sdev)->can_queue > 0)
1772 		atomic_dec(&scsi_target(sdev)->target_busy);
1773 out_put_budget:
1774 	scsi_mq_put_budget(q, cmd->budget_token);
1775 	cmd->budget_token = -1;
1776 	switch (ret) {
1777 	case BLK_STS_OK:
1778 		break;
1779 	case BLK_STS_RESOURCE:
1780 	case BLK_STS_ZONE_RESOURCE:
1781 		if (scsi_device_blocked(sdev))
1782 			ret = BLK_STS_DEV_RESOURCE;
1783 		break;
1784 	case BLK_STS_AGAIN:
1785 		cmd->result = DID_BUS_BUSY << 16;
1786 		if (req->rq_flags & RQF_DONTPREP)
1787 			scsi_mq_uninit_cmd(cmd);
1788 		break;
1789 	default:
1790 		if (unlikely(!scsi_device_online(sdev)))
1791 			cmd->result = DID_NO_CONNECT << 16;
1792 		else
1793 			cmd->result = DID_ERROR << 16;
1794 		/*
1795 		 * Make sure to release all allocated resources when
1796 		 * we hit an error, as we will never see this command
1797 		 * again.
1798 		 */
1799 		if (req->rq_flags & RQF_DONTPREP)
1800 			scsi_mq_uninit_cmd(cmd);
1801 		scsi_run_queue_async(sdev);
1802 		break;
1803 	}
1804 	return ret;
1805 }
1806 
1807 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1808 				unsigned int hctx_idx, unsigned int numa_node)
1809 {
1810 	struct Scsi_Host *shost = set->driver_data;
1811 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1812 	struct scatterlist *sg;
1813 	int ret = 0;
1814 
1815 	cmd->sense_buffer =
1816 		kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1817 	if (!cmd->sense_buffer)
1818 		return -ENOMEM;
1819 
1820 	if (scsi_host_get_prot(shost)) {
1821 		sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1822 			shost->hostt->cmd_size;
1823 		cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1824 	}
1825 
1826 	if (shost->hostt->init_cmd_priv) {
1827 		ret = shost->hostt->init_cmd_priv(shost, cmd);
1828 		if (ret < 0)
1829 			kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1830 	}
1831 
1832 	return ret;
1833 }
1834 
1835 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1836 				 unsigned int hctx_idx)
1837 {
1838 	struct Scsi_Host *shost = set->driver_data;
1839 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1840 
1841 	if (shost->hostt->exit_cmd_priv)
1842 		shost->hostt->exit_cmd_priv(shost, cmd);
1843 	kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1844 }
1845 
1846 
1847 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1848 {
1849 	struct Scsi_Host *shost = hctx->driver_data;
1850 
1851 	if (shost->hostt->mq_poll)
1852 		return shost->hostt->mq_poll(shost, hctx->queue_num);
1853 
1854 	return 0;
1855 }
1856 
1857 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1858 			  unsigned int hctx_idx)
1859 {
1860 	struct Scsi_Host *shost = data;
1861 
1862 	hctx->driver_data = shost;
1863 	return 0;
1864 }
1865 
1866 static void scsi_map_queues(struct blk_mq_tag_set *set)
1867 {
1868 	struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1869 
1870 	if (shost->hostt->map_queues)
1871 		return shost->hostt->map_queues(shost);
1872 	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1873 }
1874 
1875 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1876 {
1877 	struct device *dev = shost->dma_dev;
1878 
1879 	/*
1880 	 * this limit is imposed by hardware restrictions
1881 	 */
1882 	blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1883 					SG_MAX_SEGMENTS));
1884 
1885 	if (scsi_host_prot_dma(shost)) {
1886 		shost->sg_prot_tablesize =
1887 			min_not_zero(shost->sg_prot_tablesize,
1888 				     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1889 		BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1890 		blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1891 	}
1892 
1893 	blk_queue_max_hw_sectors(q, shost->max_sectors);
1894 	blk_queue_segment_boundary(q, shost->dma_boundary);
1895 	dma_set_seg_boundary(dev, shost->dma_boundary);
1896 
1897 	blk_queue_max_segment_size(q, shost->max_segment_size);
1898 	blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1899 	dma_set_max_seg_size(dev, queue_max_segment_size(q));
1900 
1901 	/*
1902 	 * Set a reasonable default alignment:  The larger of 32-byte (dword),
1903 	 * which is a common minimum for HBAs, and the minimum DMA alignment,
1904 	 * which is set by the platform.
1905 	 *
1906 	 * Devices that require a bigger alignment can increase it later.
1907 	 */
1908 	blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1909 }
1910 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1911 
1912 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1913 	.get_budget	= scsi_mq_get_budget,
1914 	.put_budget	= scsi_mq_put_budget,
1915 	.queue_rq	= scsi_queue_rq,
1916 	.complete	= scsi_complete,
1917 	.timeout	= scsi_timeout,
1918 #ifdef CONFIG_BLK_DEBUG_FS
1919 	.show_rq	= scsi_show_rq,
1920 #endif
1921 	.init_request	= scsi_mq_init_request,
1922 	.exit_request	= scsi_mq_exit_request,
1923 	.cleanup_rq	= scsi_cleanup_rq,
1924 	.busy		= scsi_mq_lld_busy,
1925 	.map_queues	= scsi_map_queues,
1926 	.init_hctx	= scsi_init_hctx,
1927 	.poll		= scsi_mq_poll,
1928 	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
1929 	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
1930 };
1931 
1932 
1933 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1934 {
1935 	struct Scsi_Host *shost = hctx->driver_data;
1936 
1937 	shost->hostt->commit_rqs(shost, hctx->queue_num);
1938 }
1939 
1940 static const struct blk_mq_ops scsi_mq_ops = {
1941 	.get_budget	= scsi_mq_get_budget,
1942 	.put_budget	= scsi_mq_put_budget,
1943 	.queue_rq	= scsi_queue_rq,
1944 	.commit_rqs	= scsi_commit_rqs,
1945 	.complete	= scsi_complete,
1946 	.timeout	= scsi_timeout,
1947 #ifdef CONFIG_BLK_DEBUG_FS
1948 	.show_rq	= scsi_show_rq,
1949 #endif
1950 	.init_request	= scsi_mq_init_request,
1951 	.exit_request	= scsi_mq_exit_request,
1952 	.cleanup_rq	= scsi_cleanup_rq,
1953 	.busy		= scsi_mq_lld_busy,
1954 	.map_queues	= scsi_map_queues,
1955 	.init_hctx	= scsi_init_hctx,
1956 	.poll		= scsi_mq_poll,
1957 	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
1958 	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
1959 };
1960 
1961 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1962 {
1963 	unsigned int cmd_size, sgl_size;
1964 	struct blk_mq_tag_set *tag_set = &shost->tag_set;
1965 
1966 	sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1967 				scsi_mq_inline_sgl_size(shost));
1968 	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1969 	if (scsi_host_get_prot(shost))
1970 		cmd_size += sizeof(struct scsi_data_buffer) +
1971 			sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1972 
1973 	memset(tag_set, 0, sizeof(*tag_set));
1974 	if (shost->hostt->commit_rqs)
1975 		tag_set->ops = &scsi_mq_ops;
1976 	else
1977 		tag_set->ops = &scsi_mq_ops_no_commit;
1978 	tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1979 	tag_set->nr_maps = shost->nr_maps ? : 1;
1980 	tag_set->queue_depth = shost->can_queue;
1981 	tag_set->cmd_size = cmd_size;
1982 	tag_set->numa_node = dev_to_node(shost->dma_dev);
1983 	tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1984 	tag_set->flags |=
1985 		BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1986 	if (shost->queuecommand_may_block)
1987 		tag_set->flags |= BLK_MQ_F_BLOCKING;
1988 	tag_set->driver_data = shost;
1989 	if (shost->host_tagset)
1990 		tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1991 
1992 	return blk_mq_alloc_tag_set(tag_set);
1993 }
1994 
1995 void scsi_mq_free_tags(struct kref *kref)
1996 {
1997 	struct Scsi_Host *shost = container_of(kref, typeof(*shost),
1998 					       tagset_refcnt);
1999 
2000 	blk_mq_free_tag_set(&shost->tag_set);
2001 	complete(&shost->tagset_freed);
2002 }
2003 
2004 /**
2005  * scsi_device_from_queue - return sdev associated with a request_queue
2006  * @q: The request queue to return the sdev from
2007  *
2008  * Return the sdev associated with a request queue or NULL if the
2009  * request_queue does not reference a SCSI device.
2010  */
2011 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2012 {
2013 	struct scsi_device *sdev = NULL;
2014 
2015 	if (q->mq_ops == &scsi_mq_ops_no_commit ||
2016 	    q->mq_ops == &scsi_mq_ops)
2017 		sdev = q->queuedata;
2018 	if (!sdev || !get_device(&sdev->sdev_gendev))
2019 		sdev = NULL;
2020 
2021 	return sdev;
2022 }
2023 /*
2024  * pktcdvd should have been integrated into the SCSI layers, but for historical
2025  * reasons like the old IDE driver it isn't.  This export allows it to safely
2026  * probe if a given device is a SCSI one and only attach to that.
2027  */
2028 #ifdef CONFIG_CDROM_PKTCDVD_MODULE
2029 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2030 #endif
2031 
2032 /**
2033  * scsi_block_requests - Utility function used by low-level drivers to prevent
2034  * further commands from being queued to the device.
2035  * @shost:  host in question
2036  *
2037  * There is no timer nor any other means by which the requests get unblocked
2038  * other than the low-level driver calling scsi_unblock_requests().
2039  */
2040 void scsi_block_requests(struct Scsi_Host *shost)
2041 {
2042 	shost->host_self_blocked = 1;
2043 }
2044 EXPORT_SYMBOL(scsi_block_requests);
2045 
2046 /**
2047  * scsi_unblock_requests - Utility function used by low-level drivers to allow
2048  * further commands to be queued to the device.
2049  * @shost:  host in question
2050  *
2051  * There is no timer nor any other means by which the requests get unblocked
2052  * other than the low-level driver calling scsi_unblock_requests(). This is done
2053  * as an API function so that changes to the internals of the scsi mid-layer
2054  * won't require wholesale changes to drivers that use this feature.
2055  */
2056 void scsi_unblock_requests(struct Scsi_Host *shost)
2057 {
2058 	shost->host_self_blocked = 0;
2059 	scsi_run_host_queues(shost);
2060 }
2061 EXPORT_SYMBOL(scsi_unblock_requests);
2062 
2063 void scsi_exit_queue(void)
2064 {
2065 	kmem_cache_destroy(scsi_sense_cache);
2066 }
2067 
2068 /**
2069  *	scsi_mode_select - issue a mode select
2070  *	@sdev:	SCSI device to be queried
2071  *	@pf:	Page format bit (1 == standard, 0 == vendor specific)
2072  *	@sp:	Save page bit (0 == don't save, 1 == save)
2073  *	@buffer: request buffer (may not be smaller than eight bytes)
2074  *	@len:	length of request buffer.
2075  *	@timeout: command timeout
2076  *	@retries: number of retries before failing
2077  *	@data: returns a structure abstracting the mode header data
2078  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2079  *		must be SCSI_SENSE_BUFFERSIZE big.
2080  *
2081  *	Returns zero if successful; negative error number or scsi
2082  *	status on error
2083  *
2084  */
2085 int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2086 		     unsigned char *buffer, int len, int timeout, int retries,
2087 		     struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2088 {
2089 	unsigned char cmd[10];
2090 	unsigned char *real_buffer;
2091 	const struct scsi_exec_args exec_args = {
2092 		.sshdr = sshdr,
2093 	};
2094 	int ret;
2095 
2096 	memset(cmd, 0, sizeof(cmd));
2097 	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2098 
2099 	/*
2100 	 * Use MODE SELECT(10) if the device asked for it or if the mode page
2101 	 * and the mode select header cannot fit within the maximumm 255 bytes
2102 	 * of the MODE SELECT(6) command.
2103 	 */
2104 	if (sdev->use_10_for_ms ||
2105 	    len + 4 > 255 ||
2106 	    data->block_descriptor_length > 255) {
2107 		if (len > 65535 - 8)
2108 			return -EINVAL;
2109 		real_buffer = kmalloc(8 + len, GFP_KERNEL);
2110 		if (!real_buffer)
2111 			return -ENOMEM;
2112 		memcpy(real_buffer + 8, buffer, len);
2113 		len += 8;
2114 		real_buffer[0] = 0;
2115 		real_buffer[1] = 0;
2116 		real_buffer[2] = data->medium_type;
2117 		real_buffer[3] = data->device_specific;
2118 		real_buffer[4] = data->longlba ? 0x01 : 0;
2119 		real_buffer[5] = 0;
2120 		put_unaligned_be16(data->block_descriptor_length,
2121 				   &real_buffer[6]);
2122 
2123 		cmd[0] = MODE_SELECT_10;
2124 		put_unaligned_be16(len, &cmd[7]);
2125 	} else {
2126 		if (data->longlba)
2127 			return -EINVAL;
2128 
2129 		real_buffer = kmalloc(4 + len, GFP_KERNEL);
2130 		if (!real_buffer)
2131 			return -ENOMEM;
2132 		memcpy(real_buffer + 4, buffer, len);
2133 		len += 4;
2134 		real_buffer[0] = 0;
2135 		real_buffer[1] = data->medium_type;
2136 		real_buffer[2] = data->device_specific;
2137 		real_buffer[3] = data->block_descriptor_length;
2138 
2139 		cmd[0] = MODE_SELECT;
2140 		cmd[4] = len;
2141 	}
2142 
2143 	ret = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, real_buffer, len,
2144 			       timeout, retries, &exec_args);
2145 	kfree(real_buffer);
2146 	return ret;
2147 }
2148 EXPORT_SYMBOL_GPL(scsi_mode_select);
2149 
2150 /**
2151  *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2152  *	@sdev:	SCSI device to be queried
2153  *	@dbd:	set to prevent mode sense from returning block descriptors
2154  *	@modepage: mode page being requested
2155  *	@subpage: sub-page of the mode page being requested
2156  *	@buffer: request buffer (may not be smaller than eight bytes)
2157  *	@len:	length of request buffer.
2158  *	@timeout: command timeout
2159  *	@retries: number of retries before failing
2160  *	@data: returns a structure abstracting the mode header data
2161  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2162  *		must be SCSI_SENSE_BUFFERSIZE big.
2163  *
2164  *	Returns zero if successful, or a negative error number on failure
2165  */
2166 int
2167 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, int subpage,
2168 		  unsigned char *buffer, int len, int timeout, int retries,
2169 		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2170 {
2171 	unsigned char cmd[12];
2172 	int use_10_for_ms;
2173 	int header_length;
2174 	int result, retry_count = retries;
2175 	struct scsi_sense_hdr my_sshdr;
2176 	const struct scsi_exec_args exec_args = {
2177 		/* caller might not be interested in sense, but we need it */
2178 		.sshdr = sshdr ? : &my_sshdr,
2179 	};
2180 
2181 	memset(data, 0, sizeof(*data));
2182 	memset(&cmd[0], 0, 12);
2183 
2184 	dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2185 	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */
2186 	cmd[2] = modepage;
2187 	cmd[3] = subpage;
2188 
2189 	sshdr = exec_args.sshdr;
2190 
2191  retry:
2192 	use_10_for_ms = sdev->use_10_for_ms || len > 255;
2193 
2194 	if (use_10_for_ms) {
2195 		if (len < 8 || len > 65535)
2196 			return -EINVAL;
2197 
2198 		cmd[0] = MODE_SENSE_10;
2199 		put_unaligned_be16(len, &cmd[7]);
2200 		header_length = 8;
2201 	} else {
2202 		if (len < 4)
2203 			return -EINVAL;
2204 
2205 		cmd[0] = MODE_SENSE;
2206 		cmd[4] = len;
2207 		header_length = 4;
2208 	}
2209 
2210 	memset(buffer, 0, len);
2211 
2212 	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, buffer, len,
2213 				  timeout, retries, &exec_args);
2214 	if (result < 0)
2215 		return result;
2216 
2217 	/* This code looks awful: what it's doing is making sure an
2218 	 * ILLEGAL REQUEST sense return identifies the actual command
2219 	 * byte as the problem.  MODE_SENSE commands can return
2220 	 * ILLEGAL REQUEST if the code page isn't supported */
2221 
2222 	if (!scsi_status_is_good(result)) {
2223 		if (scsi_sense_valid(sshdr)) {
2224 			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2225 			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2226 				/*
2227 				 * Invalid command operation code: retry using
2228 				 * MODE SENSE(6) if this was a MODE SENSE(10)
2229 				 * request, except if the request mode page is
2230 				 * too large for MODE SENSE single byte
2231 				 * allocation length field.
2232 				 */
2233 				if (use_10_for_ms) {
2234 					if (len > 255)
2235 						return -EIO;
2236 					sdev->use_10_for_ms = 0;
2237 					goto retry;
2238 				}
2239 			}
2240 			if (scsi_status_is_check_condition(result) &&
2241 			    sshdr->sense_key == UNIT_ATTENTION &&
2242 			    retry_count) {
2243 				retry_count--;
2244 				goto retry;
2245 			}
2246 		}
2247 		return -EIO;
2248 	}
2249 	if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2250 		     (modepage == 6 || modepage == 8))) {
2251 		/* Initio breakage? */
2252 		header_length = 0;
2253 		data->length = 13;
2254 		data->medium_type = 0;
2255 		data->device_specific = 0;
2256 		data->longlba = 0;
2257 		data->block_descriptor_length = 0;
2258 	} else if (use_10_for_ms) {
2259 		data->length = get_unaligned_be16(&buffer[0]) + 2;
2260 		data->medium_type = buffer[2];
2261 		data->device_specific = buffer[3];
2262 		data->longlba = buffer[4] & 0x01;
2263 		data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2264 	} else {
2265 		data->length = buffer[0] + 1;
2266 		data->medium_type = buffer[1];
2267 		data->device_specific = buffer[2];
2268 		data->block_descriptor_length = buffer[3];
2269 	}
2270 	data->header_length = header_length;
2271 
2272 	return 0;
2273 }
2274 EXPORT_SYMBOL(scsi_mode_sense);
2275 
2276 /**
2277  *	scsi_test_unit_ready - test if unit is ready
2278  *	@sdev:	scsi device to change the state of.
2279  *	@timeout: command timeout
2280  *	@retries: number of retries before failing
2281  *	@sshdr: outpout pointer for decoded sense information.
2282  *
2283  *	Returns zero if unsuccessful or an error if TUR failed.  For
2284  *	removable media, UNIT_ATTENTION sets ->changed flag.
2285  **/
2286 int
2287 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2288 		     struct scsi_sense_hdr *sshdr)
2289 {
2290 	char cmd[] = {
2291 		TEST_UNIT_READY, 0, 0, 0, 0, 0,
2292 	};
2293 	const struct scsi_exec_args exec_args = {
2294 		.sshdr = sshdr,
2295 	};
2296 	int result;
2297 
2298 	/* try to eat the UNIT_ATTENTION if there are enough retries */
2299 	do {
2300 		result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, NULL, 0,
2301 					  timeout, 1, &exec_args);
2302 		if (sdev->removable && scsi_sense_valid(sshdr) &&
2303 		    sshdr->sense_key == UNIT_ATTENTION)
2304 			sdev->changed = 1;
2305 	} while (scsi_sense_valid(sshdr) &&
2306 		 sshdr->sense_key == UNIT_ATTENTION && --retries);
2307 
2308 	return result;
2309 }
2310 EXPORT_SYMBOL(scsi_test_unit_ready);
2311 
2312 /**
2313  *	scsi_device_set_state - Take the given device through the device state model.
2314  *	@sdev:	scsi device to change the state of.
2315  *	@state:	state to change to.
2316  *
2317  *	Returns zero if successful or an error if the requested
2318  *	transition is illegal.
2319  */
2320 int
2321 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2322 {
2323 	enum scsi_device_state oldstate = sdev->sdev_state;
2324 
2325 	if (state == oldstate)
2326 		return 0;
2327 
2328 	switch (state) {
2329 	case SDEV_CREATED:
2330 		switch (oldstate) {
2331 		case SDEV_CREATED_BLOCK:
2332 			break;
2333 		default:
2334 			goto illegal;
2335 		}
2336 		break;
2337 
2338 	case SDEV_RUNNING:
2339 		switch (oldstate) {
2340 		case SDEV_CREATED:
2341 		case SDEV_OFFLINE:
2342 		case SDEV_TRANSPORT_OFFLINE:
2343 		case SDEV_QUIESCE:
2344 		case SDEV_BLOCK:
2345 			break;
2346 		default:
2347 			goto illegal;
2348 		}
2349 		break;
2350 
2351 	case SDEV_QUIESCE:
2352 		switch (oldstate) {
2353 		case SDEV_RUNNING:
2354 		case SDEV_OFFLINE:
2355 		case SDEV_TRANSPORT_OFFLINE:
2356 			break;
2357 		default:
2358 			goto illegal;
2359 		}
2360 		break;
2361 
2362 	case SDEV_OFFLINE:
2363 	case SDEV_TRANSPORT_OFFLINE:
2364 		switch (oldstate) {
2365 		case SDEV_CREATED:
2366 		case SDEV_RUNNING:
2367 		case SDEV_QUIESCE:
2368 		case SDEV_BLOCK:
2369 			break;
2370 		default:
2371 			goto illegal;
2372 		}
2373 		break;
2374 
2375 	case SDEV_BLOCK:
2376 		switch (oldstate) {
2377 		case SDEV_RUNNING:
2378 		case SDEV_CREATED_BLOCK:
2379 		case SDEV_QUIESCE:
2380 		case SDEV_OFFLINE:
2381 			break;
2382 		default:
2383 			goto illegal;
2384 		}
2385 		break;
2386 
2387 	case SDEV_CREATED_BLOCK:
2388 		switch (oldstate) {
2389 		case SDEV_CREATED:
2390 			break;
2391 		default:
2392 			goto illegal;
2393 		}
2394 		break;
2395 
2396 	case SDEV_CANCEL:
2397 		switch (oldstate) {
2398 		case SDEV_CREATED:
2399 		case SDEV_RUNNING:
2400 		case SDEV_QUIESCE:
2401 		case SDEV_OFFLINE:
2402 		case SDEV_TRANSPORT_OFFLINE:
2403 			break;
2404 		default:
2405 			goto illegal;
2406 		}
2407 		break;
2408 
2409 	case SDEV_DEL:
2410 		switch (oldstate) {
2411 		case SDEV_CREATED:
2412 		case SDEV_RUNNING:
2413 		case SDEV_OFFLINE:
2414 		case SDEV_TRANSPORT_OFFLINE:
2415 		case SDEV_CANCEL:
2416 		case SDEV_BLOCK:
2417 		case SDEV_CREATED_BLOCK:
2418 			break;
2419 		default:
2420 			goto illegal;
2421 		}
2422 		break;
2423 
2424 	}
2425 	sdev->offline_already = false;
2426 	sdev->sdev_state = state;
2427 	return 0;
2428 
2429  illegal:
2430 	SCSI_LOG_ERROR_RECOVERY(1,
2431 				sdev_printk(KERN_ERR, sdev,
2432 					    "Illegal state transition %s->%s",
2433 					    scsi_device_state_name(oldstate),
2434 					    scsi_device_state_name(state))
2435 				);
2436 	return -EINVAL;
2437 }
2438 EXPORT_SYMBOL(scsi_device_set_state);
2439 
2440 /**
2441  *	scsi_evt_emit - emit a single SCSI device uevent
2442  *	@sdev: associated SCSI device
2443  *	@evt: event to emit
2444  *
2445  *	Send a single uevent (scsi_event) to the associated scsi_device.
2446  */
2447 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2448 {
2449 	int idx = 0;
2450 	char *envp[3];
2451 
2452 	switch (evt->evt_type) {
2453 	case SDEV_EVT_MEDIA_CHANGE:
2454 		envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2455 		break;
2456 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2457 		scsi_rescan_device(sdev);
2458 		envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2459 		break;
2460 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2461 		envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2462 		break;
2463 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2464 	       envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2465 		break;
2466 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2467 		envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2468 		break;
2469 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2470 		envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2471 		break;
2472 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2473 		envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2474 		break;
2475 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2476 		envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2477 		break;
2478 	default:
2479 		/* do nothing */
2480 		break;
2481 	}
2482 
2483 	envp[idx++] = NULL;
2484 
2485 	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2486 }
2487 
2488 /**
2489  *	scsi_evt_thread - send a uevent for each scsi event
2490  *	@work: work struct for scsi_device
2491  *
2492  *	Dispatch queued events to their associated scsi_device kobjects
2493  *	as uevents.
2494  */
2495 void scsi_evt_thread(struct work_struct *work)
2496 {
2497 	struct scsi_device *sdev;
2498 	enum scsi_device_event evt_type;
2499 	LIST_HEAD(event_list);
2500 
2501 	sdev = container_of(work, struct scsi_device, event_work);
2502 
2503 	for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2504 		if (test_and_clear_bit(evt_type, sdev->pending_events))
2505 			sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2506 
2507 	while (1) {
2508 		struct scsi_event *evt;
2509 		struct list_head *this, *tmp;
2510 		unsigned long flags;
2511 
2512 		spin_lock_irqsave(&sdev->list_lock, flags);
2513 		list_splice_init(&sdev->event_list, &event_list);
2514 		spin_unlock_irqrestore(&sdev->list_lock, flags);
2515 
2516 		if (list_empty(&event_list))
2517 			break;
2518 
2519 		list_for_each_safe(this, tmp, &event_list) {
2520 			evt = list_entry(this, struct scsi_event, node);
2521 			list_del(&evt->node);
2522 			scsi_evt_emit(sdev, evt);
2523 			kfree(evt);
2524 		}
2525 	}
2526 }
2527 
2528 /**
2529  * 	sdev_evt_send - send asserted event to uevent thread
2530  *	@sdev: scsi_device event occurred on
2531  *	@evt: event to send
2532  *
2533  *	Assert scsi device event asynchronously.
2534  */
2535 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2536 {
2537 	unsigned long flags;
2538 
2539 #if 0
2540 	/* FIXME: currently this check eliminates all media change events
2541 	 * for polled devices.  Need to update to discriminate between AN
2542 	 * and polled events */
2543 	if (!test_bit(evt->evt_type, sdev->supported_events)) {
2544 		kfree(evt);
2545 		return;
2546 	}
2547 #endif
2548 
2549 	spin_lock_irqsave(&sdev->list_lock, flags);
2550 	list_add_tail(&evt->node, &sdev->event_list);
2551 	schedule_work(&sdev->event_work);
2552 	spin_unlock_irqrestore(&sdev->list_lock, flags);
2553 }
2554 EXPORT_SYMBOL_GPL(sdev_evt_send);
2555 
2556 /**
2557  * 	sdev_evt_alloc - allocate a new scsi event
2558  *	@evt_type: type of event to allocate
2559  *	@gfpflags: GFP flags for allocation
2560  *
2561  *	Allocates and returns a new scsi_event.
2562  */
2563 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2564 				  gfp_t gfpflags)
2565 {
2566 	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2567 	if (!evt)
2568 		return NULL;
2569 
2570 	evt->evt_type = evt_type;
2571 	INIT_LIST_HEAD(&evt->node);
2572 
2573 	/* evt_type-specific initialization, if any */
2574 	switch (evt_type) {
2575 	case SDEV_EVT_MEDIA_CHANGE:
2576 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2577 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2578 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2579 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2580 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2581 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2582 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2583 	default:
2584 		/* do nothing */
2585 		break;
2586 	}
2587 
2588 	return evt;
2589 }
2590 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2591 
2592 /**
2593  * 	sdev_evt_send_simple - send asserted event to uevent thread
2594  *	@sdev: scsi_device event occurred on
2595  *	@evt_type: type of event to send
2596  *	@gfpflags: GFP flags for allocation
2597  *
2598  *	Assert scsi device event asynchronously, given an event type.
2599  */
2600 void sdev_evt_send_simple(struct scsi_device *sdev,
2601 			  enum scsi_device_event evt_type, gfp_t gfpflags)
2602 {
2603 	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2604 	if (!evt) {
2605 		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2606 			    evt_type);
2607 		return;
2608 	}
2609 
2610 	sdev_evt_send(sdev, evt);
2611 }
2612 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2613 
2614 /**
2615  *	scsi_device_quiesce - Block all commands except power management.
2616  *	@sdev:	scsi device to quiesce.
2617  *
2618  *	This works by trying to transition to the SDEV_QUIESCE state
2619  *	(which must be a legal transition).  When the device is in this
2620  *	state, only power management requests will be accepted, all others will
2621  *	be deferred.
2622  *
2623  *	Must be called with user context, may sleep.
2624  *
2625  *	Returns zero if unsuccessful or an error if not.
2626  */
2627 int
2628 scsi_device_quiesce(struct scsi_device *sdev)
2629 {
2630 	struct request_queue *q = sdev->request_queue;
2631 	int err;
2632 
2633 	/*
2634 	 * It is allowed to call scsi_device_quiesce() multiple times from
2635 	 * the same context but concurrent scsi_device_quiesce() calls are
2636 	 * not allowed.
2637 	 */
2638 	WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2639 
2640 	if (sdev->quiesced_by == current)
2641 		return 0;
2642 
2643 	blk_set_pm_only(q);
2644 
2645 	blk_mq_freeze_queue(q);
2646 	/*
2647 	 * Ensure that the effect of blk_set_pm_only() will be visible
2648 	 * for percpu_ref_tryget() callers that occur after the queue
2649 	 * unfreeze even if the queue was already frozen before this function
2650 	 * was called. See also https://lwn.net/Articles/573497/.
2651 	 */
2652 	synchronize_rcu();
2653 	blk_mq_unfreeze_queue(q);
2654 
2655 	mutex_lock(&sdev->state_mutex);
2656 	err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2657 	if (err == 0)
2658 		sdev->quiesced_by = current;
2659 	else
2660 		blk_clear_pm_only(q);
2661 	mutex_unlock(&sdev->state_mutex);
2662 
2663 	return err;
2664 }
2665 EXPORT_SYMBOL(scsi_device_quiesce);
2666 
2667 /**
2668  *	scsi_device_resume - Restart user issued commands to a quiesced device.
2669  *	@sdev:	scsi device to resume.
2670  *
2671  *	Moves the device from quiesced back to running and restarts the
2672  *	queues.
2673  *
2674  *	Must be called with user context, may sleep.
2675  */
2676 void scsi_device_resume(struct scsi_device *sdev)
2677 {
2678 	/* check if the device state was mutated prior to resume, and if
2679 	 * so assume the state is being managed elsewhere (for example
2680 	 * device deleted during suspend)
2681 	 */
2682 	mutex_lock(&sdev->state_mutex);
2683 	if (sdev->sdev_state == SDEV_QUIESCE)
2684 		scsi_device_set_state(sdev, SDEV_RUNNING);
2685 	if (sdev->quiesced_by) {
2686 		sdev->quiesced_by = NULL;
2687 		blk_clear_pm_only(sdev->request_queue);
2688 	}
2689 	mutex_unlock(&sdev->state_mutex);
2690 }
2691 EXPORT_SYMBOL(scsi_device_resume);
2692 
2693 static void
2694 device_quiesce_fn(struct scsi_device *sdev, void *data)
2695 {
2696 	scsi_device_quiesce(sdev);
2697 }
2698 
2699 void
2700 scsi_target_quiesce(struct scsi_target *starget)
2701 {
2702 	starget_for_each_device(starget, NULL, device_quiesce_fn);
2703 }
2704 EXPORT_SYMBOL(scsi_target_quiesce);
2705 
2706 static void
2707 device_resume_fn(struct scsi_device *sdev, void *data)
2708 {
2709 	scsi_device_resume(sdev);
2710 }
2711 
2712 void
2713 scsi_target_resume(struct scsi_target *starget)
2714 {
2715 	starget_for_each_device(starget, NULL, device_resume_fn);
2716 }
2717 EXPORT_SYMBOL(scsi_target_resume);
2718 
2719 static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2720 {
2721 	if (scsi_device_set_state(sdev, SDEV_BLOCK))
2722 		return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2723 
2724 	return 0;
2725 }
2726 
2727 void scsi_start_queue(struct scsi_device *sdev)
2728 {
2729 	if (cmpxchg(&sdev->queue_stopped, 1, 0))
2730 		blk_mq_unquiesce_queue(sdev->request_queue);
2731 }
2732 
2733 static void scsi_stop_queue(struct scsi_device *sdev)
2734 {
2735 	/*
2736 	 * The atomic variable of ->queue_stopped covers that
2737 	 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2738 	 *
2739 	 * The caller needs to wait until quiesce is done.
2740 	 */
2741 	if (!cmpxchg(&sdev->queue_stopped, 0, 1))
2742 		blk_mq_quiesce_queue_nowait(sdev->request_queue);
2743 }
2744 
2745 /**
2746  * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2747  * @sdev: device to block
2748  *
2749  * Pause SCSI command processing on the specified device. Does not sleep.
2750  *
2751  * Returns zero if successful or a negative error code upon failure.
2752  *
2753  * Notes:
2754  * This routine transitions the device to the SDEV_BLOCK state (which must be
2755  * a legal transition). When the device is in this state, command processing
2756  * is paused until the device leaves the SDEV_BLOCK state. See also
2757  * scsi_internal_device_unblock_nowait().
2758  */
2759 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2760 {
2761 	int ret = __scsi_internal_device_block_nowait(sdev);
2762 
2763 	/*
2764 	 * The device has transitioned to SDEV_BLOCK.  Stop the
2765 	 * block layer from calling the midlayer with this device's
2766 	 * request queue.
2767 	 */
2768 	if (!ret)
2769 		scsi_stop_queue(sdev);
2770 	return ret;
2771 }
2772 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2773 
2774 /**
2775  * scsi_device_block - try to transition to the SDEV_BLOCK state
2776  * @sdev: device to block
2777  * @data: dummy argument, ignored
2778  *
2779  * Pause SCSI command processing on the specified device. Callers must wait
2780  * until all ongoing scsi_queue_rq() calls have finished after this function
2781  * returns.
2782  *
2783  * Note:
2784  * This routine transitions the device to the SDEV_BLOCK state (which must be
2785  * a legal transition). When the device is in this state, command processing
2786  * is paused until the device leaves the SDEV_BLOCK state. See also
2787  * scsi_internal_device_unblock().
2788  */
2789 static void scsi_device_block(struct scsi_device *sdev, void *data)
2790 {
2791 	int err;
2792 	enum scsi_device_state state;
2793 
2794 	mutex_lock(&sdev->state_mutex);
2795 	err = __scsi_internal_device_block_nowait(sdev);
2796 	state = sdev->sdev_state;
2797 	if (err == 0)
2798 		/*
2799 		 * scsi_stop_queue() must be called with the state_mutex
2800 		 * held. Otherwise a simultaneous scsi_start_queue() call
2801 		 * might unquiesce the queue before we quiesce it.
2802 		 */
2803 		scsi_stop_queue(sdev);
2804 
2805 	mutex_unlock(&sdev->state_mutex);
2806 
2807 	WARN_ONCE(err, "%s: failed to block %s in state %d\n",
2808 		  __func__, dev_name(&sdev->sdev_gendev), state);
2809 }
2810 
2811 /**
2812  * scsi_internal_device_unblock_nowait - resume a device after a block request
2813  * @sdev:	device to resume
2814  * @new_state:	state to set the device to after unblocking
2815  *
2816  * Restart the device queue for a previously suspended SCSI device. Does not
2817  * sleep.
2818  *
2819  * Returns zero if successful or a negative error code upon failure.
2820  *
2821  * Notes:
2822  * This routine transitions the device to the SDEV_RUNNING state or to one of
2823  * the offline states (which must be a legal transition) allowing the midlayer
2824  * to goose the queue for this device.
2825  */
2826 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2827 					enum scsi_device_state new_state)
2828 {
2829 	switch (new_state) {
2830 	case SDEV_RUNNING:
2831 	case SDEV_TRANSPORT_OFFLINE:
2832 		break;
2833 	default:
2834 		return -EINVAL;
2835 	}
2836 
2837 	/*
2838 	 * Try to transition the scsi device to SDEV_RUNNING or one of the
2839 	 * offlined states and goose the device queue if successful.
2840 	 */
2841 	switch (sdev->sdev_state) {
2842 	case SDEV_BLOCK:
2843 	case SDEV_TRANSPORT_OFFLINE:
2844 		sdev->sdev_state = new_state;
2845 		break;
2846 	case SDEV_CREATED_BLOCK:
2847 		if (new_state == SDEV_TRANSPORT_OFFLINE ||
2848 		    new_state == SDEV_OFFLINE)
2849 			sdev->sdev_state = new_state;
2850 		else
2851 			sdev->sdev_state = SDEV_CREATED;
2852 		break;
2853 	case SDEV_CANCEL:
2854 	case SDEV_OFFLINE:
2855 		break;
2856 	default:
2857 		return -EINVAL;
2858 	}
2859 	scsi_start_queue(sdev);
2860 
2861 	return 0;
2862 }
2863 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2864 
2865 /**
2866  * scsi_internal_device_unblock - resume a device after a block request
2867  * @sdev:	device to resume
2868  * @new_state:	state to set the device to after unblocking
2869  *
2870  * Restart the device queue for a previously suspended SCSI device. May sleep.
2871  *
2872  * Returns zero if successful or a negative error code upon failure.
2873  *
2874  * Notes:
2875  * This routine transitions the device to the SDEV_RUNNING state or to one of
2876  * the offline states (which must be a legal transition) allowing the midlayer
2877  * to goose the queue for this device.
2878  */
2879 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2880 					enum scsi_device_state new_state)
2881 {
2882 	int ret;
2883 
2884 	mutex_lock(&sdev->state_mutex);
2885 	ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2886 	mutex_unlock(&sdev->state_mutex);
2887 
2888 	return ret;
2889 }
2890 
2891 static int
2892 target_block(struct device *dev, void *data)
2893 {
2894 	if (scsi_is_target_device(dev))
2895 		starget_for_each_device(to_scsi_target(dev), NULL,
2896 					scsi_device_block);
2897 	return 0;
2898 }
2899 
2900 /**
2901  * scsi_block_targets - transition all SCSI child devices to SDEV_BLOCK state
2902  * @dev: a parent device of one or more scsi_target devices
2903  * @shost: the Scsi_Host to which this device belongs
2904  *
2905  * Iterate over all children of @dev, which should be scsi_target devices,
2906  * and switch all subordinate scsi devices to SDEV_BLOCK state. Wait for
2907  * ongoing scsi_queue_rq() calls to finish. May sleep.
2908  *
2909  * Note:
2910  * @dev must not itself be a scsi_target device.
2911  */
2912 void
2913 scsi_block_targets(struct Scsi_Host *shost, struct device *dev)
2914 {
2915 	WARN_ON_ONCE(scsi_is_target_device(dev));
2916 	device_for_each_child(dev, NULL, target_block);
2917 	blk_mq_wait_quiesce_done(&shost->tag_set);
2918 }
2919 EXPORT_SYMBOL_GPL(scsi_block_targets);
2920 
2921 static void
2922 device_unblock(struct scsi_device *sdev, void *data)
2923 {
2924 	scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2925 }
2926 
2927 static int
2928 target_unblock(struct device *dev, void *data)
2929 {
2930 	if (scsi_is_target_device(dev))
2931 		starget_for_each_device(to_scsi_target(dev), data,
2932 					device_unblock);
2933 	return 0;
2934 }
2935 
2936 void
2937 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2938 {
2939 	if (scsi_is_target_device(dev))
2940 		starget_for_each_device(to_scsi_target(dev), &new_state,
2941 					device_unblock);
2942 	else
2943 		device_for_each_child(dev, &new_state, target_unblock);
2944 }
2945 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2946 
2947 /**
2948  * scsi_host_block - Try to transition all logical units to the SDEV_BLOCK state
2949  * @shost: device to block
2950  *
2951  * Pause SCSI command processing for all logical units associated with the SCSI
2952  * host and wait until pending scsi_queue_rq() calls have finished.
2953  *
2954  * Returns zero if successful or a negative error code upon failure.
2955  */
2956 int
2957 scsi_host_block(struct Scsi_Host *shost)
2958 {
2959 	struct scsi_device *sdev;
2960 	int ret;
2961 
2962 	/*
2963 	 * Call scsi_internal_device_block_nowait so we can avoid
2964 	 * calling synchronize_rcu() for each LUN.
2965 	 */
2966 	shost_for_each_device(sdev, shost) {
2967 		mutex_lock(&sdev->state_mutex);
2968 		ret = scsi_internal_device_block_nowait(sdev);
2969 		mutex_unlock(&sdev->state_mutex);
2970 		if (ret) {
2971 			scsi_device_put(sdev);
2972 			return ret;
2973 		}
2974 	}
2975 
2976 	/* Wait for ongoing scsi_queue_rq() calls to finish. */
2977 	blk_mq_wait_quiesce_done(&shost->tag_set);
2978 
2979 	return 0;
2980 }
2981 EXPORT_SYMBOL_GPL(scsi_host_block);
2982 
2983 int
2984 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2985 {
2986 	struct scsi_device *sdev;
2987 	int ret = 0;
2988 
2989 	shost_for_each_device(sdev, shost) {
2990 		ret = scsi_internal_device_unblock(sdev, new_state);
2991 		if (ret) {
2992 			scsi_device_put(sdev);
2993 			break;
2994 		}
2995 	}
2996 	return ret;
2997 }
2998 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2999 
3000 /**
3001  * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3002  * @sgl:	scatter-gather list
3003  * @sg_count:	number of segments in sg
3004  * @offset:	offset in bytes into sg, on return offset into the mapped area
3005  * @len:	bytes to map, on return number of bytes mapped
3006  *
3007  * Returns virtual address of the start of the mapped page
3008  */
3009 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3010 			  size_t *offset, size_t *len)
3011 {
3012 	int i;
3013 	size_t sg_len = 0, len_complete = 0;
3014 	struct scatterlist *sg;
3015 	struct page *page;
3016 
3017 	WARN_ON(!irqs_disabled());
3018 
3019 	for_each_sg(sgl, sg, sg_count, i) {
3020 		len_complete = sg_len; /* Complete sg-entries */
3021 		sg_len += sg->length;
3022 		if (sg_len > *offset)
3023 			break;
3024 	}
3025 
3026 	if (unlikely(i == sg_count)) {
3027 		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3028 			"elements %d\n",
3029 		       __func__, sg_len, *offset, sg_count);
3030 		WARN_ON(1);
3031 		return NULL;
3032 	}
3033 
3034 	/* Offset starting from the beginning of first page in this sg-entry */
3035 	*offset = *offset - len_complete + sg->offset;
3036 
3037 	/* Assumption: contiguous pages can be accessed as "page + i" */
3038 	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3039 	*offset &= ~PAGE_MASK;
3040 
3041 	/* Bytes in this sg-entry from *offset to the end of the page */
3042 	sg_len = PAGE_SIZE - *offset;
3043 	if (*len > sg_len)
3044 		*len = sg_len;
3045 
3046 	return kmap_atomic(page);
3047 }
3048 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3049 
3050 /**
3051  * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3052  * @virt:	virtual address to be unmapped
3053  */
3054 void scsi_kunmap_atomic_sg(void *virt)
3055 {
3056 	kunmap_atomic(virt);
3057 }
3058 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3059 
3060 void sdev_disable_disk_events(struct scsi_device *sdev)
3061 {
3062 	atomic_inc(&sdev->disk_events_disable_depth);
3063 }
3064 EXPORT_SYMBOL(sdev_disable_disk_events);
3065 
3066 void sdev_enable_disk_events(struct scsi_device *sdev)
3067 {
3068 	if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3069 		return;
3070 	atomic_dec(&sdev->disk_events_disable_depth);
3071 }
3072 EXPORT_SYMBOL(sdev_enable_disk_events);
3073 
3074 static unsigned char designator_prio(const unsigned char *d)
3075 {
3076 	if (d[1] & 0x30)
3077 		/* not associated with LUN */
3078 		return 0;
3079 
3080 	if (d[3] == 0)
3081 		/* invalid length */
3082 		return 0;
3083 
3084 	/*
3085 	 * Order of preference for lun descriptor:
3086 	 * - SCSI name string
3087 	 * - NAA IEEE Registered Extended
3088 	 * - EUI-64 based 16-byte
3089 	 * - EUI-64 based 12-byte
3090 	 * - NAA IEEE Registered
3091 	 * - NAA IEEE Extended
3092 	 * - EUI-64 based 8-byte
3093 	 * - SCSI name string (truncated)
3094 	 * - T10 Vendor ID
3095 	 * as longer descriptors reduce the likelyhood
3096 	 * of identification clashes.
3097 	 */
3098 
3099 	switch (d[1] & 0xf) {
3100 	case 8:
3101 		/* SCSI name string, variable-length UTF-8 */
3102 		return 9;
3103 	case 3:
3104 		switch (d[4] >> 4) {
3105 		case 6:
3106 			/* NAA registered extended */
3107 			return 8;
3108 		case 5:
3109 			/* NAA registered */
3110 			return 5;
3111 		case 4:
3112 			/* NAA extended */
3113 			return 4;
3114 		case 3:
3115 			/* NAA locally assigned */
3116 			return 1;
3117 		default:
3118 			break;
3119 		}
3120 		break;
3121 	case 2:
3122 		switch (d[3]) {
3123 		case 16:
3124 			/* EUI64-based, 16 byte */
3125 			return 7;
3126 		case 12:
3127 			/* EUI64-based, 12 byte */
3128 			return 6;
3129 		case 8:
3130 			/* EUI64-based, 8 byte */
3131 			return 3;
3132 		default:
3133 			break;
3134 		}
3135 		break;
3136 	case 1:
3137 		/* T10 vendor ID */
3138 		return 1;
3139 	default:
3140 		break;
3141 	}
3142 
3143 	return 0;
3144 }
3145 
3146 /**
3147  * scsi_vpd_lun_id - return a unique device identification
3148  * @sdev: SCSI device
3149  * @id:   buffer for the identification
3150  * @id_len:  length of the buffer
3151  *
3152  * Copies a unique device identification into @id based
3153  * on the information in the VPD page 0x83 of the device.
3154  * The string will be formatted as a SCSI name string.
3155  *
3156  * Returns the length of the identification or error on failure.
3157  * If the identifier is longer than the supplied buffer the actual
3158  * identifier length is returned and the buffer is not zero-padded.
3159  */
3160 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3161 {
3162 	u8 cur_id_prio = 0;
3163 	u8 cur_id_size = 0;
3164 	const unsigned char *d, *cur_id_str;
3165 	const struct scsi_vpd *vpd_pg83;
3166 	int id_size = -EINVAL;
3167 
3168 	rcu_read_lock();
3169 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3170 	if (!vpd_pg83) {
3171 		rcu_read_unlock();
3172 		return -ENXIO;
3173 	}
3174 
3175 	/* The id string must be at least 20 bytes + terminating NULL byte */
3176 	if (id_len < 21) {
3177 		rcu_read_unlock();
3178 		return -EINVAL;
3179 	}
3180 
3181 	memset(id, 0, id_len);
3182 	for (d = vpd_pg83->data + 4;
3183 	     d < vpd_pg83->data + vpd_pg83->len;
3184 	     d += d[3] + 4) {
3185 		u8 prio = designator_prio(d);
3186 
3187 		if (prio == 0 || cur_id_prio > prio)
3188 			continue;
3189 
3190 		switch (d[1] & 0xf) {
3191 		case 0x1:
3192 			/* T10 Vendor ID */
3193 			if (cur_id_size > d[3])
3194 				break;
3195 			cur_id_prio = prio;
3196 			cur_id_size = d[3];
3197 			if (cur_id_size + 4 > id_len)
3198 				cur_id_size = id_len - 4;
3199 			cur_id_str = d + 4;
3200 			id_size = snprintf(id, id_len, "t10.%*pE",
3201 					   cur_id_size, cur_id_str);
3202 			break;
3203 		case 0x2:
3204 			/* EUI-64 */
3205 			cur_id_prio = prio;
3206 			cur_id_size = d[3];
3207 			cur_id_str = d + 4;
3208 			switch (cur_id_size) {
3209 			case 8:
3210 				id_size = snprintf(id, id_len,
3211 						   "eui.%8phN",
3212 						   cur_id_str);
3213 				break;
3214 			case 12:
3215 				id_size = snprintf(id, id_len,
3216 						   "eui.%12phN",
3217 						   cur_id_str);
3218 				break;
3219 			case 16:
3220 				id_size = snprintf(id, id_len,
3221 						   "eui.%16phN",
3222 						   cur_id_str);
3223 				break;
3224 			default:
3225 				break;
3226 			}
3227 			break;
3228 		case 0x3:
3229 			/* NAA */
3230 			cur_id_prio = prio;
3231 			cur_id_size = d[3];
3232 			cur_id_str = d + 4;
3233 			switch (cur_id_size) {
3234 			case 8:
3235 				id_size = snprintf(id, id_len,
3236 						   "naa.%8phN",
3237 						   cur_id_str);
3238 				break;
3239 			case 16:
3240 				id_size = snprintf(id, id_len,
3241 						   "naa.%16phN",
3242 						   cur_id_str);
3243 				break;
3244 			default:
3245 				break;
3246 			}
3247 			break;
3248 		case 0x8:
3249 			/* SCSI name string */
3250 			if (cur_id_size > d[3])
3251 				break;
3252 			/* Prefer others for truncated descriptor */
3253 			if (d[3] > id_len) {
3254 				prio = 2;
3255 				if (cur_id_prio > prio)
3256 					break;
3257 			}
3258 			cur_id_prio = prio;
3259 			cur_id_size = id_size = d[3];
3260 			cur_id_str = d + 4;
3261 			if (cur_id_size >= id_len)
3262 				cur_id_size = id_len - 1;
3263 			memcpy(id, cur_id_str, cur_id_size);
3264 			break;
3265 		default:
3266 			break;
3267 		}
3268 	}
3269 	rcu_read_unlock();
3270 
3271 	return id_size;
3272 }
3273 EXPORT_SYMBOL(scsi_vpd_lun_id);
3274 
3275 /*
3276  * scsi_vpd_tpg_id - return a target port group identifier
3277  * @sdev: SCSI device
3278  *
3279  * Returns the Target Port Group identifier from the information
3280  * froom VPD page 0x83 of the device.
3281  *
3282  * Returns the identifier or error on failure.
3283  */
3284 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3285 {
3286 	const unsigned char *d;
3287 	const struct scsi_vpd *vpd_pg83;
3288 	int group_id = -EAGAIN, rel_port = -1;
3289 
3290 	rcu_read_lock();
3291 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3292 	if (!vpd_pg83) {
3293 		rcu_read_unlock();
3294 		return -ENXIO;
3295 	}
3296 
3297 	d = vpd_pg83->data + 4;
3298 	while (d < vpd_pg83->data + vpd_pg83->len) {
3299 		switch (d[1] & 0xf) {
3300 		case 0x4:
3301 			/* Relative target port */
3302 			rel_port = get_unaligned_be16(&d[6]);
3303 			break;
3304 		case 0x5:
3305 			/* Target port group */
3306 			group_id = get_unaligned_be16(&d[6]);
3307 			break;
3308 		default:
3309 			break;
3310 		}
3311 		d += d[3] + 4;
3312 	}
3313 	rcu_read_unlock();
3314 
3315 	if (group_id >= 0 && rel_id && rel_port != -1)
3316 		*rel_id = rel_port;
3317 
3318 	return group_id;
3319 }
3320 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3321 
3322 /**
3323  * scsi_build_sense - build sense data for a command
3324  * @scmd:	scsi command for which the sense should be formatted
3325  * @desc:	Sense format (non-zero == descriptor format,
3326  *              0 == fixed format)
3327  * @key:	Sense key
3328  * @asc:	Additional sense code
3329  * @ascq:	Additional sense code qualifier
3330  *
3331  **/
3332 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3333 {
3334 	scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3335 	scmd->result = SAM_STAT_CHECK_CONDITION;
3336 }
3337 EXPORT_SYMBOL_GPL(scsi_build_sense);
3338