xref: /linux/drivers/scsi/scsi_lib.c (revision f2d14bc4e437b8ed21e6890ae047a6ec47c030d9)
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 static void scsi_kick_queue(struct request_queue *q)
304 {
305 	blk_mq_run_hw_queues(q, false);
306 }
307 
308 /*
309  * Kick the queue of SCSI device @sdev if @sdev != current_sdev. Called with
310  * interrupts disabled.
311  */
312 static void scsi_kick_sdev_queue(struct scsi_device *sdev, void *data)
313 {
314 	struct scsi_device *current_sdev = data;
315 
316 	if (sdev != current_sdev)
317 		blk_mq_run_hw_queues(sdev->request_queue, true);
318 }
319 
320 /*
321  * Called for single_lun devices on IO completion. Clear starget_sdev_user,
322  * and call blk_run_queue for all the scsi_devices on the target -
323  * including current_sdev first.
324  *
325  * Called with *no* scsi locks held.
326  */
327 static void scsi_single_lun_run(struct scsi_device *current_sdev)
328 {
329 	struct Scsi_Host *shost = current_sdev->host;
330 	struct scsi_target *starget = scsi_target(current_sdev);
331 	unsigned long flags;
332 
333 	spin_lock_irqsave(shost->host_lock, flags);
334 	starget->starget_sdev_user = NULL;
335 	spin_unlock_irqrestore(shost->host_lock, flags);
336 
337 	/*
338 	 * Call blk_run_queue for all LUNs on the target, starting with
339 	 * current_sdev. We race with others (to set starget_sdev_user),
340 	 * but in most cases, we will be first. Ideally, each LU on the
341 	 * target would get some limited time or requests on the target.
342 	 */
343 	scsi_kick_queue(current_sdev->request_queue);
344 
345 	spin_lock_irqsave(shost->host_lock, flags);
346 	if (!starget->starget_sdev_user)
347 		__starget_for_each_device(starget, current_sdev,
348 					  scsi_kick_sdev_queue);
349 	spin_unlock_irqrestore(shost->host_lock, flags);
350 }
351 
352 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
353 {
354 	if (scsi_device_busy(sdev) >= sdev->queue_depth)
355 		return true;
356 	if (atomic_read(&sdev->device_blocked) > 0)
357 		return true;
358 	return false;
359 }
360 
361 static inline bool scsi_target_is_busy(struct scsi_target *starget)
362 {
363 	if (starget->can_queue > 0) {
364 		if (atomic_read(&starget->target_busy) >= starget->can_queue)
365 			return true;
366 		if (atomic_read(&starget->target_blocked) > 0)
367 			return true;
368 	}
369 	return false;
370 }
371 
372 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
373 {
374 	if (atomic_read(&shost->host_blocked) > 0)
375 		return true;
376 	if (shost->host_self_blocked)
377 		return true;
378 	return false;
379 }
380 
381 static void scsi_starved_list_run(struct Scsi_Host *shost)
382 {
383 	LIST_HEAD(starved_list);
384 	struct scsi_device *sdev;
385 	unsigned long flags;
386 
387 	spin_lock_irqsave(shost->host_lock, flags);
388 	list_splice_init(&shost->starved_list, &starved_list);
389 
390 	while (!list_empty(&starved_list)) {
391 		struct request_queue *slq;
392 
393 		/*
394 		 * As long as shost is accepting commands and we have
395 		 * starved queues, call blk_run_queue. scsi_request_fn
396 		 * drops the queue_lock and can add us back to the
397 		 * starved_list.
398 		 *
399 		 * host_lock protects the starved_list and starved_entry.
400 		 * scsi_request_fn must get the host_lock before checking
401 		 * or modifying starved_list or starved_entry.
402 		 */
403 		if (scsi_host_is_busy(shost))
404 			break;
405 
406 		sdev = list_entry(starved_list.next,
407 				  struct scsi_device, starved_entry);
408 		list_del_init(&sdev->starved_entry);
409 		if (scsi_target_is_busy(scsi_target(sdev))) {
410 			list_move_tail(&sdev->starved_entry,
411 				       &shost->starved_list);
412 			continue;
413 		}
414 
415 		/*
416 		 * Once we drop the host lock, a racing scsi_remove_device()
417 		 * call may remove the sdev from the starved list and destroy
418 		 * it and the queue.  Mitigate by taking a reference to the
419 		 * queue and never touching the sdev again after we drop the
420 		 * host lock.  Note: if __scsi_remove_device() invokes
421 		 * blk_mq_destroy_queue() before the queue is run from this
422 		 * function then blk_run_queue() will return immediately since
423 		 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
424 		 */
425 		slq = sdev->request_queue;
426 		if (!blk_get_queue(slq))
427 			continue;
428 		spin_unlock_irqrestore(shost->host_lock, flags);
429 
430 		scsi_kick_queue(slq);
431 		blk_put_queue(slq);
432 
433 		spin_lock_irqsave(shost->host_lock, flags);
434 	}
435 	/* put any unprocessed entries back */
436 	list_splice(&starved_list, &shost->starved_list);
437 	spin_unlock_irqrestore(shost->host_lock, flags);
438 }
439 
440 /**
441  * scsi_run_queue - Select a proper request queue to serve next.
442  * @q:  last request's queue
443  *
444  * The previous command was completely finished, start a new one if possible.
445  */
446 static void scsi_run_queue(struct request_queue *q)
447 {
448 	struct scsi_device *sdev = q->queuedata;
449 
450 	if (scsi_target(sdev)->single_lun)
451 		scsi_single_lun_run(sdev);
452 	if (!list_empty(&sdev->host->starved_list))
453 		scsi_starved_list_run(sdev->host);
454 
455 	blk_mq_kick_requeue_list(q);
456 	blk_mq_run_hw_queues(q, false);
457 }
458 
459 void scsi_requeue_run_queue(struct work_struct *work)
460 {
461 	struct scsi_device *sdev;
462 	struct request_queue *q;
463 
464 	sdev = container_of(work, struct scsi_device, requeue_work);
465 	q = sdev->request_queue;
466 	scsi_run_queue(q);
467 }
468 
469 void scsi_run_host_queues(struct Scsi_Host *shost)
470 {
471 	struct scsi_device *sdev;
472 
473 	shost_for_each_device(sdev, shost)
474 		scsi_run_queue(sdev->request_queue);
475 }
476 
477 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
478 {
479 	if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
480 		struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
481 
482 		if (drv->uninit_command)
483 			drv->uninit_command(cmd);
484 	}
485 }
486 
487 void scsi_free_sgtables(struct scsi_cmnd *cmd)
488 {
489 	if (cmd->sdb.table.nents)
490 		sg_free_table_chained(&cmd->sdb.table,
491 				SCSI_INLINE_SG_CNT);
492 	if (scsi_prot_sg_count(cmd))
493 		sg_free_table_chained(&cmd->prot_sdb->table,
494 				SCSI_INLINE_PROT_SG_CNT);
495 }
496 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
497 
498 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
499 {
500 	scsi_free_sgtables(cmd);
501 	scsi_uninit_cmd(cmd);
502 }
503 
504 static void scsi_run_queue_async(struct scsi_device *sdev)
505 {
506 	if (scsi_host_in_recovery(sdev->host))
507 		return;
508 
509 	if (scsi_target(sdev)->single_lun ||
510 	    !list_empty(&sdev->host->starved_list)) {
511 		kblockd_schedule_work(&sdev->requeue_work);
512 	} else {
513 		/*
514 		 * smp_mb() present in sbitmap_queue_clear() or implied in
515 		 * .end_io is for ordering writing .device_busy in
516 		 * scsi_device_unbusy() and reading sdev->restarts.
517 		 */
518 		int old = atomic_read(&sdev->restarts);
519 
520 		/*
521 		 * ->restarts has to be kept as non-zero if new budget
522 		 *  contention occurs.
523 		 *
524 		 *  No need to run queue when either another re-run
525 		 *  queue wins in updating ->restarts or a new budget
526 		 *  contention occurs.
527 		 */
528 		if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
529 			blk_mq_run_hw_queues(sdev->request_queue, true);
530 	}
531 }
532 
533 /* Returns false when no more bytes to process, true if there are more */
534 static bool scsi_end_request(struct request *req, blk_status_t error,
535 		unsigned int bytes)
536 {
537 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
538 	struct scsi_device *sdev = cmd->device;
539 	struct request_queue *q = sdev->request_queue;
540 
541 	if (blk_update_request(req, error, bytes))
542 		return true;
543 
544 	// XXX:
545 	if (blk_queue_add_random(q))
546 		add_disk_randomness(req->q->disk);
547 
548 	if (!blk_rq_is_passthrough(req)) {
549 		WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
550 		cmd->flags &= ~SCMD_INITIALIZED;
551 	}
552 
553 	/*
554 	 * Calling rcu_barrier() is not necessary here because the
555 	 * SCSI error handler guarantees that the function called by
556 	 * call_rcu() has been called before scsi_end_request() is
557 	 * called.
558 	 */
559 	destroy_rcu_head(&cmd->rcu);
560 
561 	/*
562 	 * In the MQ case the command gets freed by __blk_mq_end_request,
563 	 * so we have to do all cleanup that depends on it earlier.
564 	 *
565 	 * We also can't kick the queues from irq context, so we
566 	 * will have to defer it to a workqueue.
567 	 */
568 	scsi_mq_uninit_cmd(cmd);
569 
570 	/*
571 	 * queue is still alive, so grab the ref for preventing it
572 	 * from being cleaned up during running queue.
573 	 */
574 	percpu_ref_get(&q->q_usage_counter);
575 
576 	__blk_mq_end_request(req, error);
577 
578 	scsi_run_queue_async(sdev);
579 
580 	percpu_ref_put(&q->q_usage_counter);
581 	return false;
582 }
583 
584 /**
585  * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
586  * @result:	scsi error code
587  *
588  * Translate a SCSI result code into a blk_status_t value.
589  */
590 static blk_status_t scsi_result_to_blk_status(int result)
591 {
592 	/*
593 	 * Check the scsi-ml byte first in case we converted a host or status
594 	 * byte.
595 	 */
596 	switch (scsi_ml_byte(result)) {
597 	case SCSIML_STAT_OK:
598 		break;
599 	case SCSIML_STAT_RESV_CONFLICT:
600 		return BLK_STS_RESV_CONFLICT;
601 	case SCSIML_STAT_NOSPC:
602 		return BLK_STS_NOSPC;
603 	case SCSIML_STAT_MED_ERROR:
604 		return BLK_STS_MEDIUM;
605 	case SCSIML_STAT_TGT_FAILURE:
606 		return BLK_STS_TARGET;
607 	case SCSIML_STAT_DL_TIMEOUT:
608 		return BLK_STS_DURATION_LIMIT;
609 	}
610 
611 	switch (host_byte(result)) {
612 	case DID_OK:
613 		if (scsi_status_is_good(result))
614 			return BLK_STS_OK;
615 		return BLK_STS_IOERR;
616 	case DID_TRANSPORT_FAILFAST:
617 	case DID_TRANSPORT_MARGINAL:
618 		return BLK_STS_TRANSPORT;
619 	default:
620 		return BLK_STS_IOERR;
621 	}
622 }
623 
624 /**
625  * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
626  * @rq: request to examine
627  *
628  * Description:
629  *     A request could be merge of IOs which require different failure
630  *     handling.  This function determines the number of bytes which
631  *     can be failed from the beginning of the request without
632  *     crossing into area which need to be retried further.
633  *
634  * Return:
635  *     The number of bytes to fail.
636  */
637 static unsigned int scsi_rq_err_bytes(const struct request *rq)
638 {
639 	blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
640 	unsigned int bytes = 0;
641 	struct bio *bio;
642 
643 	if (!(rq->rq_flags & RQF_MIXED_MERGE))
644 		return blk_rq_bytes(rq);
645 
646 	/*
647 	 * Currently the only 'mixing' which can happen is between
648 	 * different fastfail types.  We can safely fail portions
649 	 * which have all the failfast bits that the first one has -
650 	 * the ones which are at least as eager to fail as the first
651 	 * one.
652 	 */
653 	for (bio = rq->bio; bio; bio = bio->bi_next) {
654 		if ((bio->bi_opf & ff) != ff)
655 			break;
656 		bytes += bio->bi_iter.bi_size;
657 	}
658 
659 	/* this could lead to infinite loop */
660 	BUG_ON(blk_rq_bytes(rq) && !bytes);
661 	return bytes;
662 }
663 
664 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
665 {
666 	struct request *req = scsi_cmd_to_rq(cmd);
667 	unsigned long wait_for;
668 
669 	if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
670 		return false;
671 
672 	wait_for = (cmd->allowed + 1) * req->timeout;
673 	if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
674 		scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
675 			    wait_for/HZ);
676 		return true;
677 	}
678 	return false;
679 }
680 
681 /*
682  * When ALUA transition state is returned, reprep the cmd to
683  * use the ALUA handler's transition timeout. Delay the reprep
684  * 1 sec to avoid aggressive retries of the target in that
685  * state.
686  */
687 #define ALUA_TRANSITION_REPREP_DELAY	1000
688 
689 /* Helper for scsi_io_completion() when special action required. */
690 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
691 {
692 	struct request *req = scsi_cmd_to_rq(cmd);
693 	int level = 0;
694 	enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
695 	      ACTION_RETRY, ACTION_DELAYED_RETRY} action;
696 	struct scsi_sense_hdr sshdr;
697 	bool sense_valid;
698 	bool sense_current = true;      /* false implies "deferred sense" */
699 	blk_status_t blk_stat;
700 
701 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
702 	if (sense_valid)
703 		sense_current = !scsi_sense_is_deferred(&sshdr);
704 
705 	blk_stat = scsi_result_to_blk_status(result);
706 
707 	if (host_byte(result) == DID_RESET) {
708 		/* Third party bus reset or reset for error recovery
709 		 * reasons.  Just retry the command and see what
710 		 * happens.
711 		 */
712 		action = ACTION_RETRY;
713 	} else if (sense_valid && sense_current) {
714 		switch (sshdr.sense_key) {
715 		case UNIT_ATTENTION:
716 			if (cmd->device->removable) {
717 				/* Detected disc change.  Set a bit
718 				 * and quietly refuse further access.
719 				 */
720 				cmd->device->changed = 1;
721 				action = ACTION_FAIL;
722 			} else {
723 				/* Must have been a power glitch, or a
724 				 * bus reset.  Could not have been a
725 				 * media change, so we just retry the
726 				 * command and see what happens.
727 				 */
728 				action = ACTION_RETRY;
729 			}
730 			break;
731 		case ILLEGAL_REQUEST:
732 			/* If we had an ILLEGAL REQUEST returned, then
733 			 * we may have performed an unsupported
734 			 * command.  The only thing this should be
735 			 * would be a ten byte read where only a six
736 			 * byte read was supported.  Also, on a system
737 			 * where READ CAPACITY failed, we may have
738 			 * read past the end of the disk.
739 			 */
740 			if ((cmd->device->use_10_for_rw &&
741 			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
742 			    (cmd->cmnd[0] == READ_10 ||
743 			     cmd->cmnd[0] == WRITE_10)) {
744 				/* This will issue a new 6-byte command. */
745 				cmd->device->use_10_for_rw = 0;
746 				action = ACTION_REPREP;
747 			} else if (sshdr.asc == 0x10) /* DIX */ {
748 				action = ACTION_FAIL;
749 				blk_stat = BLK_STS_PROTECTION;
750 			/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
751 			} else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
752 				action = ACTION_FAIL;
753 				blk_stat = BLK_STS_TARGET;
754 			} else
755 				action = ACTION_FAIL;
756 			break;
757 		case ABORTED_COMMAND:
758 			action = ACTION_FAIL;
759 			if (sshdr.asc == 0x10) /* DIF */
760 				blk_stat = BLK_STS_PROTECTION;
761 			break;
762 		case NOT_READY:
763 			/* If the device is in the process of becoming
764 			 * ready, or has a temporary blockage, retry.
765 			 */
766 			if (sshdr.asc == 0x04) {
767 				switch (sshdr.ascq) {
768 				case 0x01: /* becoming ready */
769 				case 0x04: /* format in progress */
770 				case 0x05: /* rebuild in progress */
771 				case 0x06: /* recalculation in progress */
772 				case 0x07: /* operation in progress */
773 				case 0x08: /* Long write in progress */
774 				case 0x09: /* self test in progress */
775 				case 0x11: /* notify (enable spinup) required */
776 				case 0x14: /* space allocation in progress */
777 				case 0x1a: /* start stop unit in progress */
778 				case 0x1b: /* sanitize in progress */
779 				case 0x1d: /* configuration in progress */
780 				case 0x24: /* depopulation in progress */
781 					action = ACTION_DELAYED_RETRY;
782 					break;
783 				case 0x0a: /* ALUA state transition */
784 					action = ACTION_DELAYED_REPREP;
785 					break;
786 				default:
787 					action = ACTION_FAIL;
788 					break;
789 				}
790 			} else
791 				action = ACTION_FAIL;
792 			break;
793 		case VOLUME_OVERFLOW:
794 			/* See SSC3rXX or current. */
795 			action = ACTION_FAIL;
796 			break;
797 		case DATA_PROTECT:
798 			action = ACTION_FAIL;
799 			if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
800 			    (sshdr.asc == 0x55 &&
801 			     (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
802 				/* Insufficient zone resources */
803 				blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
804 			}
805 			break;
806 		case COMPLETED:
807 			fallthrough;
808 		default:
809 			action = ACTION_FAIL;
810 			break;
811 		}
812 	} else
813 		action = ACTION_FAIL;
814 
815 	if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
816 		action = ACTION_FAIL;
817 
818 	switch (action) {
819 	case ACTION_FAIL:
820 		/* Give up and fail the remainder of the request */
821 		if (!(req->rq_flags & RQF_QUIET)) {
822 			static DEFINE_RATELIMIT_STATE(_rs,
823 					DEFAULT_RATELIMIT_INTERVAL,
824 					DEFAULT_RATELIMIT_BURST);
825 
826 			if (unlikely(scsi_logging_level))
827 				level =
828 				     SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
829 						    SCSI_LOG_MLCOMPLETE_BITS);
830 
831 			/*
832 			 * if logging is enabled the failure will be printed
833 			 * in scsi_log_completion(), so avoid duplicate messages
834 			 */
835 			if (!level && __ratelimit(&_rs)) {
836 				scsi_print_result(cmd, NULL, FAILED);
837 				if (sense_valid)
838 					scsi_print_sense(cmd);
839 				scsi_print_command(cmd);
840 			}
841 		}
842 		if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
843 			return;
844 		fallthrough;
845 	case ACTION_REPREP:
846 		scsi_mq_requeue_cmd(cmd, 0);
847 		break;
848 	case ACTION_DELAYED_REPREP:
849 		scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
850 		break;
851 	case ACTION_RETRY:
852 		/* Retry the same command immediately */
853 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
854 		break;
855 	case ACTION_DELAYED_RETRY:
856 		/* Retry the same command after a delay */
857 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
858 		break;
859 	}
860 }
861 
862 /*
863  * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
864  * new result that may suppress further error checking. Also modifies
865  * *blk_statp in some cases.
866  */
867 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
868 					blk_status_t *blk_statp)
869 {
870 	bool sense_valid;
871 	bool sense_current = true;	/* false implies "deferred sense" */
872 	struct request *req = scsi_cmd_to_rq(cmd);
873 	struct scsi_sense_hdr sshdr;
874 
875 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
876 	if (sense_valid)
877 		sense_current = !scsi_sense_is_deferred(&sshdr);
878 
879 	if (blk_rq_is_passthrough(req)) {
880 		if (sense_valid) {
881 			/*
882 			 * SG_IO wants current and deferred errors
883 			 */
884 			cmd->sense_len = min(8 + cmd->sense_buffer[7],
885 					     SCSI_SENSE_BUFFERSIZE);
886 		}
887 		if (sense_current)
888 			*blk_statp = scsi_result_to_blk_status(result);
889 	} else if (blk_rq_bytes(req) == 0 && sense_current) {
890 		/*
891 		 * Flush commands do not transfers any data, and thus cannot use
892 		 * good_bytes != blk_rq_bytes(req) as the signal for an error.
893 		 * This sets *blk_statp explicitly for the problem case.
894 		 */
895 		*blk_statp = scsi_result_to_blk_status(result);
896 	}
897 	/*
898 	 * Recovered errors need reporting, but they're always treated as
899 	 * success, so fiddle the result code here.  For passthrough requests
900 	 * we already took a copy of the original into sreq->result which
901 	 * is what gets returned to the user
902 	 */
903 	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
904 		bool do_print = true;
905 		/*
906 		 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
907 		 * skip print since caller wants ATA registers. Only occurs
908 		 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
909 		 */
910 		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
911 			do_print = false;
912 		else if (req->rq_flags & RQF_QUIET)
913 			do_print = false;
914 		if (do_print)
915 			scsi_print_sense(cmd);
916 		result = 0;
917 		/* for passthrough, *blk_statp may be set */
918 		*blk_statp = BLK_STS_OK;
919 	}
920 	/*
921 	 * Another corner case: the SCSI status byte is non-zero but 'good'.
922 	 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
923 	 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
924 	 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
925 	 * intermediate statuses (both obsolete in SAM-4) as good.
926 	 */
927 	if ((result & 0xff) && scsi_status_is_good(result)) {
928 		result = 0;
929 		*blk_statp = BLK_STS_OK;
930 	}
931 	return result;
932 }
933 
934 /**
935  * scsi_io_completion - Completion processing for SCSI commands.
936  * @cmd:	command that is finished.
937  * @good_bytes:	number of processed bytes.
938  *
939  * We will finish off the specified number of sectors. If we are done, the
940  * command block will be released and the queue function will be goosed. If we
941  * are not done then we have to figure out what to do next:
942  *
943  *   a) We can call scsi_mq_requeue_cmd().  The request will be
944  *	unprepared and put back on the queue.  Then a new command will
945  *	be created for it.  This should be used if we made forward
946  *	progress, or if we want to switch from READ(10) to READ(6) for
947  *	example.
948  *
949  *   b) We can call scsi_io_completion_action().  The request will be
950  *	put back on the queue and retried using the same command as
951  *	before, possibly after a delay.
952  *
953  *   c) We can call scsi_end_request() with blk_stat other than
954  *	BLK_STS_OK, to fail the remainder of the request.
955  */
956 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
957 {
958 	int result = cmd->result;
959 	struct request *req = scsi_cmd_to_rq(cmd);
960 	blk_status_t blk_stat = BLK_STS_OK;
961 
962 	if (unlikely(result))	/* a nz result may or may not be an error */
963 		result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
964 
965 	/*
966 	 * Next deal with any sectors which we were able to correctly
967 	 * handle.
968 	 */
969 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
970 		"%u sectors total, %d bytes done.\n",
971 		blk_rq_sectors(req), good_bytes));
972 
973 	/*
974 	 * Failed, zero length commands always need to drop down
975 	 * to retry code. Fast path should return in this block.
976 	 */
977 	if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
978 		if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
979 			return; /* no bytes remaining */
980 	}
981 
982 	/* Kill remainder if no retries. */
983 	if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
984 		if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
985 			WARN_ONCE(true,
986 			    "Bytes remaining after failed, no-retry command");
987 		return;
988 	}
989 
990 	/*
991 	 * If there had been no error, but we have leftover bytes in the
992 	 * request just queue the command up again.
993 	 */
994 	if (likely(result == 0))
995 		scsi_mq_requeue_cmd(cmd, 0);
996 	else
997 		scsi_io_completion_action(cmd, result);
998 }
999 
1000 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1001 		struct request *rq)
1002 {
1003 	return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1004 	       !op_is_write(req_op(rq)) &&
1005 	       sdev->host->hostt->dma_need_drain(rq);
1006 }
1007 
1008 /**
1009  * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1010  * @cmd: SCSI command data structure to initialize.
1011  *
1012  * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1013  * for @cmd.
1014  *
1015  * Returns:
1016  * * BLK_STS_OK       - on success
1017  * * BLK_STS_RESOURCE - if the failure is retryable
1018  * * BLK_STS_IOERR    - if the failure is fatal
1019  */
1020 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1021 {
1022 	struct scsi_device *sdev = cmd->device;
1023 	struct request *rq = scsi_cmd_to_rq(cmd);
1024 	unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1025 	struct scatterlist *last_sg = NULL;
1026 	blk_status_t ret;
1027 	bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1028 	int count;
1029 
1030 	if (WARN_ON_ONCE(!nr_segs))
1031 		return BLK_STS_IOERR;
1032 
1033 	/*
1034 	 * Make sure there is space for the drain.  The driver must adjust
1035 	 * max_hw_segments to be prepared for this.
1036 	 */
1037 	if (need_drain)
1038 		nr_segs++;
1039 
1040 	/*
1041 	 * If sg table allocation fails, requeue request later.
1042 	 */
1043 	if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1044 			cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1045 		return BLK_STS_RESOURCE;
1046 
1047 	/*
1048 	 * Next, walk the list, and fill in the addresses and sizes of
1049 	 * each segment.
1050 	 */
1051 	count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1052 
1053 	if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1054 		unsigned int pad_len =
1055 			(rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1056 
1057 		last_sg->length += pad_len;
1058 		cmd->extra_len += pad_len;
1059 	}
1060 
1061 	if (need_drain) {
1062 		sg_unmark_end(last_sg);
1063 		last_sg = sg_next(last_sg);
1064 		sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1065 		sg_mark_end(last_sg);
1066 
1067 		cmd->extra_len += sdev->dma_drain_len;
1068 		count++;
1069 	}
1070 
1071 	BUG_ON(count > cmd->sdb.table.nents);
1072 	cmd->sdb.table.nents = count;
1073 	cmd->sdb.length = blk_rq_payload_bytes(rq);
1074 
1075 	if (blk_integrity_rq(rq)) {
1076 		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1077 		int ivecs;
1078 
1079 		if (WARN_ON_ONCE(!prot_sdb)) {
1080 			/*
1081 			 * This can happen if someone (e.g. multipath)
1082 			 * queues a command to a device on an adapter
1083 			 * that does not support DIX.
1084 			 */
1085 			ret = BLK_STS_IOERR;
1086 			goto out_free_sgtables;
1087 		}
1088 
1089 		ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1090 
1091 		if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1092 				prot_sdb->table.sgl,
1093 				SCSI_INLINE_PROT_SG_CNT)) {
1094 			ret = BLK_STS_RESOURCE;
1095 			goto out_free_sgtables;
1096 		}
1097 
1098 		count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1099 						prot_sdb->table.sgl);
1100 		BUG_ON(count > ivecs);
1101 		BUG_ON(count > queue_max_integrity_segments(rq->q));
1102 
1103 		cmd->prot_sdb = prot_sdb;
1104 		cmd->prot_sdb->table.nents = count;
1105 	}
1106 
1107 	return BLK_STS_OK;
1108 out_free_sgtables:
1109 	scsi_free_sgtables(cmd);
1110 	return ret;
1111 }
1112 EXPORT_SYMBOL(scsi_alloc_sgtables);
1113 
1114 /**
1115  * scsi_initialize_rq - initialize struct scsi_cmnd partially
1116  * @rq: Request associated with the SCSI command to be initialized.
1117  *
1118  * This function initializes the members of struct scsi_cmnd that must be
1119  * initialized before request processing starts and that won't be
1120  * reinitialized if a SCSI command is requeued.
1121  */
1122 static void scsi_initialize_rq(struct request *rq)
1123 {
1124 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1125 
1126 	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1127 	cmd->cmd_len = MAX_COMMAND_SIZE;
1128 	cmd->sense_len = 0;
1129 	init_rcu_head(&cmd->rcu);
1130 	cmd->jiffies_at_alloc = jiffies;
1131 	cmd->retries = 0;
1132 }
1133 
1134 struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1135 				   blk_mq_req_flags_t flags)
1136 {
1137 	struct request *rq;
1138 
1139 	rq = blk_mq_alloc_request(q, opf, flags);
1140 	if (!IS_ERR(rq))
1141 		scsi_initialize_rq(rq);
1142 	return rq;
1143 }
1144 EXPORT_SYMBOL_GPL(scsi_alloc_request);
1145 
1146 /*
1147  * Only called when the request isn't completed by SCSI, and not freed by
1148  * SCSI
1149  */
1150 static void scsi_cleanup_rq(struct request *rq)
1151 {
1152 	if (rq->rq_flags & RQF_DONTPREP) {
1153 		scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1154 		rq->rq_flags &= ~RQF_DONTPREP;
1155 	}
1156 }
1157 
1158 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1159 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1160 {
1161 	struct request *rq = scsi_cmd_to_rq(cmd);
1162 
1163 	if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1164 		cmd->flags |= SCMD_INITIALIZED;
1165 		scsi_initialize_rq(rq);
1166 	}
1167 
1168 	cmd->device = dev;
1169 	INIT_LIST_HEAD(&cmd->eh_entry);
1170 	INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1171 }
1172 
1173 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1174 		struct request *req)
1175 {
1176 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1177 
1178 	/*
1179 	 * Passthrough requests may transfer data, in which case they must
1180 	 * a bio attached to them.  Or they might contain a SCSI command
1181 	 * that does not transfer data, in which case they may optionally
1182 	 * submit a request without an attached bio.
1183 	 */
1184 	if (req->bio) {
1185 		blk_status_t ret = scsi_alloc_sgtables(cmd);
1186 		if (unlikely(ret != BLK_STS_OK))
1187 			return ret;
1188 	} else {
1189 		BUG_ON(blk_rq_bytes(req));
1190 
1191 		memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1192 	}
1193 
1194 	cmd->transfersize = blk_rq_bytes(req);
1195 	return BLK_STS_OK;
1196 }
1197 
1198 static blk_status_t
1199 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1200 {
1201 	switch (sdev->sdev_state) {
1202 	case SDEV_CREATED:
1203 		return BLK_STS_OK;
1204 	case SDEV_OFFLINE:
1205 	case SDEV_TRANSPORT_OFFLINE:
1206 		/*
1207 		 * If the device is offline we refuse to process any
1208 		 * commands.  The device must be brought online
1209 		 * before trying any recovery commands.
1210 		 */
1211 		if (!sdev->offline_already) {
1212 			sdev->offline_already = true;
1213 			sdev_printk(KERN_ERR, sdev,
1214 				    "rejecting I/O to offline device\n");
1215 		}
1216 		return BLK_STS_IOERR;
1217 	case SDEV_DEL:
1218 		/*
1219 		 * If the device is fully deleted, we refuse to
1220 		 * process any commands as well.
1221 		 */
1222 		sdev_printk(KERN_ERR, sdev,
1223 			    "rejecting I/O to dead device\n");
1224 		return BLK_STS_IOERR;
1225 	case SDEV_BLOCK:
1226 	case SDEV_CREATED_BLOCK:
1227 		return BLK_STS_RESOURCE;
1228 	case SDEV_QUIESCE:
1229 		/*
1230 		 * If the device is blocked we only accept power management
1231 		 * commands.
1232 		 */
1233 		if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1234 			return BLK_STS_RESOURCE;
1235 		return BLK_STS_OK;
1236 	default:
1237 		/*
1238 		 * For any other not fully online state we only allow
1239 		 * power management commands.
1240 		 */
1241 		if (req && !(req->rq_flags & RQF_PM))
1242 			return BLK_STS_OFFLINE;
1243 		return BLK_STS_OK;
1244 	}
1245 }
1246 
1247 /*
1248  * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1249  * and return the token else return -1.
1250  */
1251 static inline int scsi_dev_queue_ready(struct request_queue *q,
1252 				  struct scsi_device *sdev)
1253 {
1254 	int token;
1255 
1256 	token = sbitmap_get(&sdev->budget_map);
1257 	if (atomic_read(&sdev->device_blocked)) {
1258 		if (token < 0)
1259 			goto out;
1260 
1261 		if (scsi_device_busy(sdev) > 1)
1262 			goto out_dec;
1263 
1264 		/*
1265 		 * unblock after device_blocked iterates to zero
1266 		 */
1267 		if (atomic_dec_return(&sdev->device_blocked) > 0)
1268 			goto out_dec;
1269 		SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1270 				   "unblocking device at zero depth\n"));
1271 	}
1272 
1273 	return token;
1274 out_dec:
1275 	if (token >= 0)
1276 		sbitmap_put(&sdev->budget_map, token);
1277 out:
1278 	return -1;
1279 }
1280 
1281 /*
1282  * scsi_target_queue_ready: checks if there we can send commands to target
1283  * @sdev: scsi device on starget to check.
1284  */
1285 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1286 					   struct scsi_device *sdev)
1287 {
1288 	struct scsi_target *starget = scsi_target(sdev);
1289 	unsigned int busy;
1290 
1291 	if (starget->single_lun) {
1292 		spin_lock_irq(shost->host_lock);
1293 		if (starget->starget_sdev_user &&
1294 		    starget->starget_sdev_user != sdev) {
1295 			spin_unlock_irq(shost->host_lock);
1296 			return 0;
1297 		}
1298 		starget->starget_sdev_user = sdev;
1299 		spin_unlock_irq(shost->host_lock);
1300 	}
1301 
1302 	if (starget->can_queue <= 0)
1303 		return 1;
1304 
1305 	busy = atomic_inc_return(&starget->target_busy) - 1;
1306 	if (atomic_read(&starget->target_blocked) > 0) {
1307 		if (busy)
1308 			goto starved;
1309 
1310 		/*
1311 		 * unblock after target_blocked iterates to zero
1312 		 */
1313 		if (atomic_dec_return(&starget->target_blocked) > 0)
1314 			goto out_dec;
1315 
1316 		SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1317 				 "unblocking target at zero depth\n"));
1318 	}
1319 
1320 	if (busy >= starget->can_queue)
1321 		goto starved;
1322 
1323 	return 1;
1324 
1325 starved:
1326 	spin_lock_irq(shost->host_lock);
1327 	list_move_tail(&sdev->starved_entry, &shost->starved_list);
1328 	spin_unlock_irq(shost->host_lock);
1329 out_dec:
1330 	if (starget->can_queue > 0)
1331 		atomic_dec(&starget->target_busy);
1332 	return 0;
1333 }
1334 
1335 /*
1336  * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1337  * return 0. We must end up running the queue again whenever 0 is
1338  * returned, else IO can hang.
1339  */
1340 static inline int scsi_host_queue_ready(struct request_queue *q,
1341 				   struct Scsi_Host *shost,
1342 				   struct scsi_device *sdev,
1343 				   struct scsi_cmnd *cmd)
1344 {
1345 	if (atomic_read(&shost->host_blocked) > 0) {
1346 		if (scsi_host_busy(shost) > 0)
1347 			goto starved;
1348 
1349 		/*
1350 		 * unblock after host_blocked iterates to zero
1351 		 */
1352 		if (atomic_dec_return(&shost->host_blocked) > 0)
1353 			goto out_dec;
1354 
1355 		SCSI_LOG_MLQUEUE(3,
1356 			shost_printk(KERN_INFO, shost,
1357 				     "unblocking host at zero depth\n"));
1358 	}
1359 
1360 	if (shost->host_self_blocked)
1361 		goto starved;
1362 
1363 	/* We're OK to process the command, so we can't be starved */
1364 	if (!list_empty(&sdev->starved_entry)) {
1365 		spin_lock_irq(shost->host_lock);
1366 		if (!list_empty(&sdev->starved_entry))
1367 			list_del_init(&sdev->starved_entry);
1368 		spin_unlock_irq(shost->host_lock);
1369 	}
1370 
1371 	__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1372 
1373 	return 1;
1374 
1375 starved:
1376 	spin_lock_irq(shost->host_lock);
1377 	if (list_empty(&sdev->starved_entry))
1378 		list_add_tail(&sdev->starved_entry, &shost->starved_list);
1379 	spin_unlock_irq(shost->host_lock);
1380 out_dec:
1381 	scsi_dec_host_busy(shost, cmd);
1382 	return 0;
1383 }
1384 
1385 /*
1386  * Busy state exporting function for request stacking drivers.
1387  *
1388  * For efficiency, no lock is taken to check the busy state of
1389  * shost/starget/sdev, since the returned value is not guaranteed and
1390  * may be changed after request stacking drivers call the function,
1391  * regardless of taking lock or not.
1392  *
1393  * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1394  * needs to return 'not busy'. Otherwise, request stacking drivers
1395  * may hold requests forever.
1396  */
1397 static bool scsi_mq_lld_busy(struct request_queue *q)
1398 {
1399 	struct scsi_device *sdev = q->queuedata;
1400 	struct Scsi_Host *shost;
1401 
1402 	if (blk_queue_dying(q))
1403 		return false;
1404 
1405 	shost = sdev->host;
1406 
1407 	/*
1408 	 * Ignore host/starget busy state.
1409 	 * Since block layer does not have a concept of fairness across
1410 	 * multiple queues, congestion of host/starget needs to be handled
1411 	 * in SCSI layer.
1412 	 */
1413 	if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1414 		return true;
1415 
1416 	return false;
1417 }
1418 
1419 /*
1420  * Block layer request completion callback. May be called from interrupt
1421  * context.
1422  */
1423 static void scsi_complete(struct request *rq)
1424 {
1425 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1426 	enum scsi_disposition disposition;
1427 
1428 	INIT_LIST_HEAD(&cmd->eh_entry);
1429 
1430 	atomic_inc(&cmd->device->iodone_cnt);
1431 	if (cmd->result)
1432 		atomic_inc(&cmd->device->ioerr_cnt);
1433 
1434 	disposition = scsi_decide_disposition(cmd);
1435 	if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1436 		disposition = SUCCESS;
1437 
1438 	scsi_log_completion(cmd, disposition);
1439 
1440 	switch (disposition) {
1441 	case SUCCESS:
1442 		scsi_finish_command(cmd);
1443 		break;
1444 	case NEEDS_RETRY:
1445 		scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1446 		break;
1447 	case ADD_TO_MLQUEUE:
1448 		scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1449 		break;
1450 	default:
1451 		scsi_eh_scmd_add(cmd);
1452 		break;
1453 	}
1454 }
1455 
1456 /**
1457  * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1458  * @cmd: command block we are dispatching.
1459  *
1460  * Return: nonzero return request was rejected and device's queue needs to be
1461  * plugged.
1462  */
1463 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1464 {
1465 	struct Scsi_Host *host = cmd->device->host;
1466 	int rtn = 0;
1467 
1468 	atomic_inc(&cmd->device->iorequest_cnt);
1469 
1470 	/* check if the device is still usable */
1471 	if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1472 		/* in SDEV_DEL we error all commands. DID_NO_CONNECT
1473 		 * returns an immediate error upwards, and signals
1474 		 * that the device is no longer present */
1475 		cmd->result = DID_NO_CONNECT << 16;
1476 		goto done;
1477 	}
1478 
1479 	/* Check to see if the scsi lld made this device blocked. */
1480 	if (unlikely(scsi_device_blocked(cmd->device))) {
1481 		/*
1482 		 * in blocked state, the command is just put back on
1483 		 * the device queue.  The suspend state has already
1484 		 * blocked the queue so future requests should not
1485 		 * occur until the device transitions out of the
1486 		 * suspend state.
1487 		 */
1488 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1489 			"queuecommand : device blocked\n"));
1490 		atomic_dec(&cmd->device->iorequest_cnt);
1491 		return SCSI_MLQUEUE_DEVICE_BUSY;
1492 	}
1493 
1494 	/* Store the LUN value in cmnd, if needed. */
1495 	if (cmd->device->lun_in_cdb)
1496 		cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1497 			       (cmd->device->lun << 5 & 0xe0);
1498 
1499 	scsi_log_send(cmd);
1500 
1501 	/*
1502 	 * Before we queue this command, check if the command
1503 	 * length exceeds what the host adapter can handle.
1504 	 */
1505 	if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1506 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1507 			       "queuecommand : command too long. "
1508 			       "cdb_size=%d host->max_cmd_len=%d\n",
1509 			       cmd->cmd_len, cmd->device->host->max_cmd_len));
1510 		cmd->result = (DID_ABORT << 16);
1511 		goto done;
1512 	}
1513 
1514 	if (unlikely(host->shost_state == SHOST_DEL)) {
1515 		cmd->result = (DID_NO_CONNECT << 16);
1516 		goto done;
1517 
1518 	}
1519 
1520 	trace_scsi_dispatch_cmd_start(cmd);
1521 	rtn = host->hostt->queuecommand(host, cmd);
1522 	if (rtn) {
1523 		atomic_dec(&cmd->device->iorequest_cnt);
1524 		trace_scsi_dispatch_cmd_error(cmd, rtn);
1525 		if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1526 		    rtn != SCSI_MLQUEUE_TARGET_BUSY)
1527 			rtn = SCSI_MLQUEUE_HOST_BUSY;
1528 
1529 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1530 			"queuecommand : request rejected\n"));
1531 	}
1532 
1533 	return rtn;
1534  done:
1535 	scsi_done(cmd);
1536 	return 0;
1537 }
1538 
1539 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1540 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1541 {
1542 	return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1543 		sizeof(struct scatterlist);
1544 }
1545 
1546 static blk_status_t scsi_prepare_cmd(struct request *req)
1547 {
1548 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1549 	struct scsi_device *sdev = req->q->queuedata;
1550 	struct Scsi_Host *shost = sdev->host;
1551 	bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1552 	struct scatterlist *sg;
1553 
1554 	scsi_init_command(sdev, cmd);
1555 
1556 	cmd->eh_eflags = 0;
1557 	cmd->prot_type = 0;
1558 	cmd->prot_flags = 0;
1559 	cmd->submitter = 0;
1560 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1561 	cmd->underflow = 0;
1562 	cmd->transfersize = 0;
1563 	cmd->host_scribble = NULL;
1564 	cmd->result = 0;
1565 	cmd->extra_len = 0;
1566 	cmd->state = 0;
1567 	if (in_flight)
1568 		__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1569 
1570 	/*
1571 	 * Only clear the driver-private command data if the LLD does not supply
1572 	 * a function to initialize that data.
1573 	 */
1574 	if (!shost->hostt->init_cmd_priv)
1575 		memset(cmd + 1, 0, shost->hostt->cmd_size);
1576 
1577 	cmd->prot_op = SCSI_PROT_NORMAL;
1578 	if (blk_rq_bytes(req))
1579 		cmd->sc_data_direction = rq_dma_dir(req);
1580 	else
1581 		cmd->sc_data_direction = DMA_NONE;
1582 
1583 	sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1584 	cmd->sdb.table.sgl = sg;
1585 
1586 	if (scsi_host_get_prot(shost)) {
1587 		memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1588 
1589 		cmd->prot_sdb->table.sgl =
1590 			(struct scatterlist *)(cmd->prot_sdb + 1);
1591 	}
1592 
1593 	/*
1594 	 * Special handling for passthrough commands, which don't go to the ULP
1595 	 * at all:
1596 	 */
1597 	if (blk_rq_is_passthrough(req))
1598 		return scsi_setup_scsi_cmnd(sdev, req);
1599 
1600 	if (sdev->handler && sdev->handler->prep_fn) {
1601 		blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1602 
1603 		if (ret != BLK_STS_OK)
1604 			return ret;
1605 	}
1606 
1607 	/* Usually overridden by the ULP */
1608 	cmd->allowed = 0;
1609 	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1610 	return scsi_cmd_to_driver(cmd)->init_command(cmd);
1611 }
1612 
1613 static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1614 {
1615 	struct request *req = scsi_cmd_to_rq(cmd);
1616 
1617 	switch (cmd->submitter) {
1618 	case SUBMITTED_BY_BLOCK_LAYER:
1619 		break;
1620 	case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1621 		return scsi_eh_done(cmd);
1622 	case SUBMITTED_BY_SCSI_RESET_IOCTL:
1623 		return;
1624 	}
1625 
1626 	if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1627 		return;
1628 	if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1629 		return;
1630 	trace_scsi_dispatch_cmd_done(cmd);
1631 
1632 	if (complete_directly)
1633 		blk_mq_complete_request_direct(req, scsi_complete);
1634 	else
1635 		blk_mq_complete_request(req);
1636 }
1637 
1638 void scsi_done(struct scsi_cmnd *cmd)
1639 {
1640 	scsi_done_internal(cmd, false);
1641 }
1642 EXPORT_SYMBOL(scsi_done);
1643 
1644 void scsi_done_direct(struct scsi_cmnd *cmd)
1645 {
1646 	scsi_done_internal(cmd, true);
1647 }
1648 EXPORT_SYMBOL(scsi_done_direct);
1649 
1650 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1651 {
1652 	struct scsi_device *sdev = q->queuedata;
1653 
1654 	sbitmap_put(&sdev->budget_map, budget_token);
1655 }
1656 
1657 /*
1658  * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1659  * not change behaviour from the previous unplug mechanism, experimentation
1660  * may prove this needs changing.
1661  */
1662 #define SCSI_QUEUE_DELAY 3
1663 
1664 static int scsi_mq_get_budget(struct request_queue *q)
1665 {
1666 	struct scsi_device *sdev = q->queuedata;
1667 	int token = scsi_dev_queue_ready(q, sdev);
1668 
1669 	if (token >= 0)
1670 		return token;
1671 
1672 	atomic_inc(&sdev->restarts);
1673 
1674 	/*
1675 	 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1676 	 * .restarts must be incremented before .device_busy is read because the
1677 	 * code in scsi_run_queue_async() depends on the order of these operations.
1678 	 */
1679 	smp_mb__after_atomic();
1680 
1681 	/*
1682 	 * If all in-flight requests originated from this LUN are completed
1683 	 * before reading .device_busy, sdev->device_busy will be observed as
1684 	 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1685 	 * soon. Otherwise, completion of one of these requests will observe
1686 	 * the .restarts flag, and the request queue will be run for handling
1687 	 * this request, see scsi_end_request().
1688 	 */
1689 	if (unlikely(scsi_device_busy(sdev) == 0 &&
1690 				!scsi_device_blocked(sdev)))
1691 		blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1692 	return -1;
1693 }
1694 
1695 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1696 {
1697 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1698 
1699 	cmd->budget_token = token;
1700 }
1701 
1702 static int scsi_mq_get_rq_budget_token(struct request *req)
1703 {
1704 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1705 
1706 	return cmd->budget_token;
1707 }
1708 
1709 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1710 			 const struct blk_mq_queue_data *bd)
1711 {
1712 	struct request *req = bd->rq;
1713 	struct request_queue *q = req->q;
1714 	struct scsi_device *sdev = q->queuedata;
1715 	struct Scsi_Host *shost = sdev->host;
1716 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1717 	blk_status_t ret;
1718 	int reason;
1719 
1720 	WARN_ON_ONCE(cmd->budget_token < 0);
1721 
1722 	/*
1723 	 * If the device is not in running state we will reject some or all
1724 	 * commands.
1725 	 */
1726 	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1727 		ret = scsi_device_state_check(sdev, req);
1728 		if (ret != BLK_STS_OK)
1729 			goto out_put_budget;
1730 	}
1731 
1732 	ret = BLK_STS_RESOURCE;
1733 	if (!scsi_target_queue_ready(shost, sdev))
1734 		goto out_put_budget;
1735 	if (unlikely(scsi_host_in_recovery(shost))) {
1736 		if (cmd->flags & SCMD_FAIL_IF_RECOVERING)
1737 			ret = BLK_STS_OFFLINE;
1738 		goto out_dec_target_busy;
1739 	}
1740 	if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1741 		goto out_dec_target_busy;
1742 
1743 	if (!(req->rq_flags & RQF_DONTPREP)) {
1744 		ret = scsi_prepare_cmd(req);
1745 		if (ret != BLK_STS_OK)
1746 			goto out_dec_host_busy;
1747 		req->rq_flags |= RQF_DONTPREP;
1748 	} else {
1749 		clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1750 	}
1751 
1752 	cmd->flags &= SCMD_PRESERVED_FLAGS;
1753 	if (sdev->simple_tags)
1754 		cmd->flags |= SCMD_TAGGED;
1755 	if (bd->last)
1756 		cmd->flags |= SCMD_LAST;
1757 
1758 	scsi_set_resid(cmd, 0);
1759 	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1760 	cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1761 
1762 	blk_mq_start_request(req);
1763 	reason = scsi_dispatch_cmd(cmd);
1764 	if (reason) {
1765 		scsi_set_blocked(cmd, reason);
1766 		ret = BLK_STS_RESOURCE;
1767 		goto out_dec_host_busy;
1768 	}
1769 
1770 	return BLK_STS_OK;
1771 
1772 out_dec_host_busy:
1773 	scsi_dec_host_busy(shost, cmd);
1774 out_dec_target_busy:
1775 	if (scsi_target(sdev)->can_queue > 0)
1776 		atomic_dec(&scsi_target(sdev)->target_busy);
1777 out_put_budget:
1778 	scsi_mq_put_budget(q, cmd->budget_token);
1779 	cmd->budget_token = -1;
1780 	switch (ret) {
1781 	case BLK_STS_OK:
1782 		break;
1783 	case BLK_STS_RESOURCE:
1784 	case BLK_STS_ZONE_RESOURCE:
1785 		if (scsi_device_blocked(sdev))
1786 			ret = BLK_STS_DEV_RESOURCE;
1787 		break;
1788 	case BLK_STS_AGAIN:
1789 		cmd->result = DID_BUS_BUSY << 16;
1790 		if (req->rq_flags & RQF_DONTPREP)
1791 			scsi_mq_uninit_cmd(cmd);
1792 		break;
1793 	default:
1794 		if (unlikely(!scsi_device_online(sdev)))
1795 			cmd->result = DID_NO_CONNECT << 16;
1796 		else
1797 			cmd->result = DID_ERROR << 16;
1798 		/*
1799 		 * Make sure to release all allocated resources when
1800 		 * we hit an error, as we will never see this command
1801 		 * again.
1802 		 */
1803 		if (req->rq_flags & RQF_DONTPREP)
1804 			scsi_mq_uninit_cmd(cmd);
1805 		scsi_run_queue_async(sdev);
1806 		break;
1807 	}
1808 	return ret;
1809 }
1810 
1811 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1812 				unsigned int hctx_idx, unsigned int numa_node)
1813 {
1814 	struct Scsi_Host *shost = set->driver_data;
1815 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1816 	struct scatterlist *sg;
1817 	int ret = 0;
1818 
1819 	cmd->sense_buffer =
1820 		kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1821 	if (!cmd->sense_buffer)
1822 		return -ENOMEM;
1823 
1824 	if (scsi_host_get_prot(shost)) {
1825 		sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1826 			shost->hostt->cmd_size;
1827 		cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1828 	}
1829 
1830 	if (shost->hostt->init_cmd_priv) {
1831 		ret = shost->hostt->init_cmd_priv(shost, cmd);
1832 		if (ret < 0)
1833 			kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1834 	}
1835 
1836 	return ret;
1837 }
1838 
1839 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1840 				 unsigned int hctx_idx)
1841 {
1842 	struct Scsi_Host *shost = set->driver_data;
1843 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1844 
1845 	if (shost->hostt->exit_cmd_priv)
1846 		shost->hostt->exit_cmd_priv(shost, cmd);
1847 	kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1848 }
1849 
1850 
1851 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1852 {
1853 	struct Scsi_Host *shost = hctx->driver_data;
1854 
1855 	if (shost->hostt->mq_poll)
1856 		return shost->hostt->mq_poll(shost, hctx->queue_num);
1857 
1858 	return 0;
1859 }
1860 
1861 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1862 			  unsigned int hctx_idx)
1863 {
1864 	struct Scsi_Host *shost = data;
1865 
1866 	hctx->driver_data = shost;
1867 	return 0;
1868 }
1869 
1870 static void scsi_map_queues(struct blk_mq_tag_set *set)
1871 {
1872 	struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1873 
1874 	if (shost->hostt->map_queues)
1875 		return shost->hostt->map_queues(shost);
1876 	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1877 }
1878 
1879 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1880 {
1881 	struct device *dev = shost->dma_dev;
1882 
1883 	/*
1884 	 * this limit is imposed by hardware restrictions
1885 	 */
1886 	blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1887 					SG_MAX_SEGMENTS));
1888 
1889 	if (scsi_host_prot_dma(shost)) {
1890 		shost->sg_prot_tablesize =
1891 			min_not_zero(shost->sg_prot_tablesize,
1892 				     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1893 		BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1894 		blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1895 	}
1896 
1897 	blk_queue_max_hw_sectors(q, shost->max_sectors);
1898 	blk_queue_segment_boundary(q, shost->dma_boundary);
1899 	dma_set_seg_boundary(dev, shost->dma_boundary);
1900 
1901 	blk_queue_max_segment_size(q, shost->max_segment_size);
1902 	blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1903 	dma_set_max_seg_size(dev, queue_max_segment_size(q));
1904 
1905 	/*
1906 	 * Set a reasonable default alignment:  The larger of 32-byte (dword),
1907 	 * which is a common minimum for HBAs, and the minimum DMA alignment,
1908 	 * which is set by the platform.
1909 	 *
1910 	 * Devices that require a bigger alignment can increase it later.
1911 	 */
1912 	blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1913 }
1914 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1915 
1916 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1917 	.get_budget	= scsi_mq_get_budget,
1918 	.put_budget	= scsi_mq_put_budget,
1919 	.queue_rq	= scsi_queue_rq,
1920 	.complete	= scsi_complete,
1921 	.timeout	= scsi_timeout,
1922 #ifdef CONFIG_BLK_DEBUG_FS
1923 	.show_rq	= scsi_show_rq,
1924 #endif
1925 	.init_request	= scsi_mq_init_request,
1926 	.exit_request	= scsi_mq_exit_request,
1927 	.cleanup_rq	= scsi_cleanup_rq,
1928 	.busy		= scsi_mq_lld_busy,
1929 	.map_queues	= scsi_map_queues,
1930 	.init_hctx	= scsi_init_hctx,
1931 	.poll		= scsi_mq_poll,
1932 	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
1933 	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
1934 };
1935 
1936 
1937 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1938 {
1939 	struct Scsi_Host *shost = hctx->driver_data;
1940 
1941 	shost->hostt->commit_rqs(shost, hctx->queue_num);
1942 }
1943 
1944 static const struct blk_mq_ops scsi_mq_ops = {
1945 	.get_budget	= scsi_mq_get_budget,
1946 	.put_budget	= scsi_mq_put_budget,
1947 	.queue_rq	= scsi_queue_rq,
1948 	.commit_rqs	= scsi_commit_rqs,
1949 	.complete	= scsi_complete,
1950 	.timeout	= scsi_timeout,
1951 #ifdef CONFIG_BLK_DEBUG_FS
1952 	.show_rq	= scsi_show_rq,
1953 #endif
1954 	.init_request	= scsi_mq_init_request,
1955 	.exit_request	= scsi_mq_exit_request,
1956 	.cleanup_rq	= scsi_cleanup_rq,
1957 	.busy		= scsi_mq_lld_busy,
1958 	.map_queues	= scsi_map_queues,
1959 	.init_hctx	= scsi_init_hctx,
1960 	.poll		= scsi_mq_poll,
1961 	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
1962 	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
1963 };
1964 
1965 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1966 {
1967 	unsigned int cmd_size, sgl_size;
1968 	struct blk_mq_tag_set *tag_set = &shost->tag_set;
1969 
1970 	sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1971 				scsi_mq_inline_sgl_size(shost));
1972 	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1973 	if (scsi_host_get_prot(shost))
1974 		cmd_size += sizeof(struct scsi_data_buffer) +
1975 			sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1976 
1977 	memset(tag_set, 0, sizeof(*tag_set));
1978 	if (shost->hostt->commit_rqs)
1979 		tag_set->ops = &scsi_mq_ops;
1980 	else
1981 		tag_set->ops = &scsi_mq_ops_no_commit;
1982 	tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1983 	tag_set->nr_maps = shost->nr_maps ? : 1;
1984 	tag_set->queue_depth = shost->can_queue;
1985 	tag_set->cmd_size = cmd_size;
1986 	tag_set->numa_node = dev_to_node(shost->dma_dev);
1987 	tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1988 	tag_set->flags |=
1989 		BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1990 	if (shost->queuecommand_may_block)
1991 		tag_set->flags |= BLK_MQ_F_BLOCKING;
1992 	tag_set->driver_data = shost;
1993 	if (shost->host_tagset)
1994 		tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1995 
1996 	return blk_mq_alloc_tag_set(tag_set);
1997 }
1998 
1999 void scsi_mq_free_tags(struct kref *kref)
2000 {
2001 	struct Scsi_Host *shost = container_of(kref, typeof(*shost),
2002 					       tagset_refcnt);
2003 
2004 	blk_mq_free_tag_set(&shost->tag_set);
2005 	complete(&shost->tagset_freed);
2006 }
2007 
2008 /**
2009  * scsi_device_from_queue - return sdev associated with a request_queue
2010  * @q: The request queue to return the sdev from
2011  *
2012  * Return the sdev associated with a request queue or NULL if the
2013  * request_queue does not reference a SCSI device.
2014  */
2015 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2016 {
2017 	struct scsi_device *sdev = NULL;
2018 
2019 	if (q->mq_ops == &scsi_mq_ops_no_commit ||
2020 	    q->mq_ops == &scsi_mq_ops)
2021 		sdev = q->queuedata;
2022 	if (!sdev || !get_device(&sdev->sdev_gendev))
2023 		sdev = NULL;
2024 
2025 	return sdev;
2026 }
2027 /*
2028  * pktcdvd should have been integrated into the SCSI layers, but for historical
2029  * reasons like the old IDE driver it isn't.  This export allows it to safely
2030  * probe if a given device is a SCSI one and only attach to that.
2031  */
2032 #ifdef CONFIG_CDROM_PKTCDVD_MODULE
2033 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2034 #endif
2035 
2036 /**
2037  * scsi_block_requests - Utility function used by low-level drivers to prevent
2038  * further commands from being queued to the device.
2039  * @shost:  host in question
2040  *
2041  * There is no timer nor any other means by which the requests get unblocked
2042  * other than the low-level driver calling scsi_unblock_requests().
2043  */
2044 void scsi_block_requests(struct Scsi_Host *shost)
2045 {
2046 	shost->host_self_blocked = 1;
2047 }
2048 EXPORT_SYMBOL(scsi_block_requests);
2049 
2050 /**
2051  * scsi_unblock_requests - Utility function used by low-level drivers to allow
2052  * further commands to be queued to the device.
2053  * @shost:  host in question
2054  *
2055  * There is no timer nor any other means by which the requests get unblocked
2056  * other than the low-level driver calling scsi_unblock_requests(). This is done
2057  * as an API function so that changes to the internals of the scsi mid-layer
2058  * won't require wholesale changes to drivers that use this feature.
2059  */
2060 void scsi_unblock_requests(struct Scsi_Host *shost)
2061 {
2062 	shost->host_self_blocked = 0;
2063 	scsi_run_host_queues(shost);
2064 }
2065 EXPORT_SYMBOL(scsi_unblock_requests);
2066 
2067 void scsi_exit_queue(void)
2068 {
2069 	kmem_cache_destroy(scsi_sense_cache);
2070 }
2071 
2072 /**
2073  *	scsi_mode_select - issue a mode select
2074  *	@sdev:	SCSI device to be queried
2075  *	@pf:	Page format bit (1 == standard, 0 == vendor specific)
2076  *	@sp:	Save page bit (0 == don't save, 1 == save)
2077  *	@buffer: request buffer (may not be smaller than eight bytes)
2078  *	@len:	length of request buffer.
2079  *	@timeout: command timeout
2080  *	@retries: number of retries before failing
2081  *	@data: returns a structure abstracting the mode header data
2082  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2083  *		must be SCSI_SENSE_BUFFERSIZE big.
2084  *
2085  *	Returns zero if successful; negative error number or scsi
2086  *	status on error
2087  *
2088  */
2089 int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2090 		     unsigned char *buffer, int len, int timeout, int retries,
2091 		     struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2092 {
2093 	unsigned char cmd[10];
2094 	unsigned char *real_buffer;
2095 	const struct scsi_exec_args exec_args = {
2096 		.sshdr = sshdr,
2097 	};
2098 	int ret;
2099 
2100 	memset(cmd, 0, sizeof(cmd));
2101 	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2102 
2103 	/*
2104 	 * Use MODE SELECT(10) if the device asked for it or if the mode page
2105 	 * and the mode select header cannot fit within the maximumm 255 bytes
2106 	 * of the MODE SELECT(6) command.
2107 	 */
2108 	if (sdev->use_10_for_ms ||
2109 	    len + 4 > 255 ||
2110 	    data->block_descriptor_length > 255) {
2111 		if (len > 65535 - 8)
2112 			return -EINVAL;
2113 		real_buffer = kmalloc(8 + len, GFP_KERNEL);
2114 		if (!real_buffer)
2115 			return -ENOMEM;
2116 		memcpy(real_buffer + 8, buffer, len);
2117 		len += 8;
2118 		real_buffer[0] = 0;
2119 		real_buffer[1] = 0;
2120 		real_buffer[2] = data->medium_type;
2121 		real_buffer[3] = data->device_specific;
2122 		real_buffer[4] = data->longlba ? 0x01 : 0;
2123 		real_buffer[5] = 0;
2124 		put_unaligned_be16(data->block_descriptor_length,
2125 				   &real_buffer[6]);
2126 
2127 		cmd[0] = MODE_SELECT_10;
2128 		put_unaligned_be16(len, &cmd[7]);
2129 	} else {
2130 		if (data->longlba)
2131 			return -EINVAL;
2132 
2133 		real_buffer = kmalloc(4 + len, GFP_KERNEL);
2134 		if (!real_buffer)
2135 			return -ENOMEM;
2136 		memcpy(real_buffer + 4, buffer, len);
2137 		len += 4;
2138 		real_buffer[0] = 0;
2139 		real_buffer[1] = data->medium_type;
2140 		real_buffer[2] = data->device_specific;
2141 		real_buffer[3] = data->block_descriptor_length;
2142 
2143 		cmd[0] = MODE_SELECT;
2144 		cmd[4] = len;
2145 	}
2146 
2147 	ret = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, real_buffer, len,
2148 			       timeout, retries, &exec_args);
2149 	kfree(real_buffer);
2150 	return ret;
2151 }
2152 EXPORT_SYMBOL_GPL(scsi_mode_select);
2153 
2154 /**
2155  *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2156  *	@sdev:	SCSI device to be queried
2157  *	@dbd:	set to prevent mode sense from returning block descriptors
2158  *	@modepage: mode page being requested
2159  *	@subpage: sub-page of the mode page being requested
2160  *	@buffer: request buffer (may not be smaller than eight bytes)
2161  *	@len:	length of request buffer.
2162  *	@timeout: command timeout
2163  *	@retries: number of retries before failing
2164  *	@data: returns a structure abstracting the mode header data
2165  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2166  *		must be SCSI_SENSE_BUFFERSIZE big.
2167  *
2168  *	Returns zero if successful, or a negative error number on failure
2169  */
2170 int
2171 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, int subpage,
2172 		  unsigned char *buffer, int len, int timeout, int retries,
2173 		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2174 {
2175 	unsigned char cmd[12];
2176 	int use_10_for_ms;
2177 	int header_length;
2178 	int result, retry_count = retries;
2179 	struct scsi_sense_hdr my_sshdr;
2180 	const struct scsi_exec_args exec_args = {
2181 		/* caller might not be interested in sense, but we need it */
2182 		.sshdr = sshdr ? : &my_sshdr,
2183 	};
2184 
2185 	memset(data, 0, sizeof(*data));
2186 	memset(&cmd[0], 0, 12);
2187 
2188 	dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2189 	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */
2190 	cmd[2] = modepage;
2191 	cmd[3] = subpage;
2192 
2193 	sshdr = exec_args.sshdr;
2194 
2195  retry:
2196 	use_10_for_ms = sdev->use_10_for_ms || len > 255;
2197 
2198 	if (use_10_for_ms) {
2199 		if (len < 8 || len > 65535)
2200 			return -EINVAL;
2201 
2202 		cmd[0] = MODE_SENSE_10;
2203 		put_unaligned_be16(len, &cmd[7]);
2204 		header_length = 8;
2205 	} else {
2206 		if (len < 4)
2207 			return -EINVAL;
2208 
2209 		cmd[0] = MODE_SENSE;
2210 		cmd[4] = len;
2211 		header_length = 4;
2212 	}
2213 
2214 	memset(buffer, 0, len);
2215 
2216 	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, buffer, len,
2217 				  timeout, retries, &exec_args);
2218 	if (result < 0)
2219 		return result;
2220 
2221 	/* This code looks awful: what it's doing is making sure an
2222 	 * ILLEGAL REQUEST sense return identifies the actual command
2223 	 * byte as the problem.  MODE_SENSE commands can return
2224 	 * ILLEGAL REQUEST if the code page isn't supported */
2225 
2226 	if (!scsi_status_is_good(result)) {
2227 		if (scsi_sense_valid(sshdr)) {
2228 			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2229 			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2230 				/*
2231 				 * Invalid command operation code: retry using
2232 				 * MODE SENSE(6) if this was a MODE SENSE(10)
2233 				 * request, except if the request mode page is
2234 				 * too large for MODE SENSE single byte
2235 				 * allocation length field.
2236 				 */
2237 				if (use_10_for_ms) {
2238 					if (len > 255)
2239 						return -EIO;
2240 					sdev->use_10_for_ms = 0;
2241 					goto retry;
2242 				}
2243 			}
2244 			if (scsi_status_is_check_condition(result) &&
2245 			    sshdr->sense_key == UNIT_ATTENTION &&
2246 			    retry_count) {
2247 				retry_count--;
2248 				goto retry;
2249 			}
2250 		}
2251 		return -EIO;
2252 	}
2253 	if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2254 		     (modepage == 6 || modepage == 8))) {
2255 		/* Initio breakage? */
2256 		header_length = 0;
2257 		data->length = 13;
2258 		data->medium_type = 0;
2259 		data->device_specific = 0;
2260 		data->longlba = 0;
2261 		data->block_descriptor_length = 0;
2262 	} else if (use_10_for_ms) {
2263 		data->length = get_unaligned_be16(&buffer[0]) + 2;
2264 		data->medium_type = buffer[2];
2265 		data->device_specific = buffer[3];
2266 		data->longlba = buffer[4] & 0x01;
2267 		data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2268 	} else {
2269 		data->length = buffer[0] + 1;
2270 		data->medium_type = buffer[1];
2271 		data->device_specific = buffer[2];
2272 		data->block_descriptor_length = buffer[3];
2273 	}
2274 	data->header_length = header_length;
2275 
2276 	return 0;
2277 }
2278 EXPORT_SYMBOL(scsi_mode_sense);
2279 
2280 /**
2281  *	scsi_test_unit_ready - test if unit is ready
2282  *	@sdev:	scsi device to change the state of.
2283  *	@timeout: command timeout
2284  *	@retries: number of retries before failing
2285  *	@sshdr: outpout pointer for decoded sense information.
2286  *
2287  *	Returns zero if unsuccessful or an error if TUR failed.  For
2288  *	removable media, UNIT_ATTENTION sets ->changed flag.
2289  **/
2290 int
2291 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2292 		     struct scsi_sense_hdr *sshdr)
2293 {
2294 	char cmd[] = {
2295 		TEST_UNIT_READY, 0, 0, 0, 0, 0,
2296 	};
2297 	const struct scsi_exec_args exec_args = {
2298 		.sshdr = sshdr,
2299 	};
2300 	int result;
2301 
2302 	/* try to eat the UNIT_ATTENTION if there are enough retries */
2303 	do {
2304 		result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, NULL, 0,
2305 					  timeout, 1, &exec_args);
2306 		if (sdev->removable && scsi_sense_valid(sshdr) &&
2307 		    sshdr->sense_key == UNIT_ATTENTION)
2308 			sdev->changed = 1;
2309 	} while (scsi_sense_valid(sshdr) &&
2310 		 sshdr->sense_key == UNIT_ATTENTION && --retries);
2311 
2312 	return result;
2313 }
2314 EXPORT_SYMBOL(scsi_test_unit_ready);
2315 
2316 /**
2317  *	scsi_device_set_state - Take the given device through the device state model.
2318  *	@sdev:	scsi device to change the state of.
2319  *	@state:	state to change to.
2320  *
2321  *	Returns zero if successful or an error if the requested
2322  *	transition is illegal.
2323  */
2324 int
2325 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2326 {
2327 	enum scsi_device_state oldstate = sdev->sdev_state;
2328 
2329 	if (state == oldstate)
2330 		return 0;
2331 
2332 	switch (state) {
2333 	case SDEV_CREATED:
2334 		switch (oldstate) {
2335 		case SDEV_CREATED_BLOCK:
2336 			break;
2337 		default:
2338 			goto illegal;
2339 		}
2340 		break;
2341 
2342 	case SDEV_RUNNING:
2343 		switch (oldstate) {
2344 		case SDEV_CREATED:
2345 		case SDEV_OFFLINE:
2346 		case SDEV_TRANSPORT_OFFLINE:
2347 		case SDEV_QUIESCE:
2348 		case SDEV_BLOCK:
2349 			break;
2350 		default:
2351 			goto illegal;
2352 		}
2353 		break;
2354 
2355 	case SDEV_QUIESCE:
2356 		switch (oldstate) {
2357 		case SDEV_RUNNING:
2358 		case SDEV_OFFLINE:
2359 		case SDEV_TRANSPORT_OFFLINE:
2360 			break;
2361 		default:
2362 			goto illegal;
2363 		}
2364 		break;
2365 
2366 	case SDEV_OFFLINE:
2367 	case SDEV_TRANSPORT_OFFLINE:
2368 		switch (oldstate) {
2369 		case SDEV_CREATED:
2370 		case SDEV_RUNNING:
2371 		case SDEV_QUIESCE:
2372 		case SDEV_BLOCK:
2373 			break;
2374 		default:
2375 			goto illegal;
2376 		}
2377 		break;
2378 
2379 	case SDEV_BLOCK:
2380 		switch (oldstate) {
2381 		case SDEV_RUNNING:
2382 		case SDEV_CREATED_BLOCK:
2383 		case SDEV_QUIESCE:
2384 		case SDEV_OFFLINE:
2385 			break;
2386 		default:
2387 			goto illegal;
2388 		}
2389 		break;
2390 
2391 	case SDEV_CREATED_BLOCK:
2392 		switch (oldstate) {
2393 		case SDEV_CREATED:
2394 			break;
2395 		default:
2396 			goto illegal;
2397 		}
2398 		break;
2399 
2400 	case SDEV_CANCEL:
2401 		switch (oldstate) {
2402 		case SDEV_CREATED:
2403 		case SDEV_RUNNING:
2404 		case SDEV_QUIESCE:
2405 		case SDEV_OFFLINE:
2406 		case SDEV_TRANSPORT_OFFLINE:
2407 			break;
2408 		default:
2409 			goto illegal;
2410 		}
2411 		break;
2412 
2413 	case SDEV_DEL:
2414 		switch (oldstate) {
2415 		case SDEV_CREATED:
2416 		case SDEV_RUNNING:
2417 		case SDEV_OFFLINE:
2418 		case SDEV_TRANSPORT_OFFLINE:
2419 		case SDEV_CANCEL:
2420 		case SDEV_BLOCK:
2421 		case SDEV_CREATED_BLOCK:
2422 			break;
2423 		default:
2424 			goto illegal;
2425 		}
2426 		break;
2427 
2428 	}
2429 	sdev->offline_already = false;
2430 	sdev->sdev_state = state;
2431 	return 0;
2432 
2433  illegal:
2434 	SCSI_LOG_ERROR_RECOVERY(1,
2435 				sdev_printk(KERN_ERR, sdev,
2436 					    "Illegal state transition %s->%s",
2437 					    scsi_device_state_name(oldstate),
2438 					    scsi_device_state_name(state))
2439 				);
2440 	return -EINVAL;
2441 }
2442 EXPORT_SYMBOL(scsi_device_set_state);
2443 
2444 /**
2445  *	scsi_evt_emit - emit a single SCSI device uevent
2446  *	@sdev: associated SCSI device
2447  *	@evt: event to emit
2448  *
2449  *	Send a single uevent (scsi_event) to the associated scsi_device.
2450  */
2451 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2452 {
2453 	int idx = 0;
2454 	char *envp[3];
2455 
2456 	switch (evt->evt_type) {
2457 	case SDEV_EVT_MEDIA_CHANGE:
2458 		envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2459 		break;
2460 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2461 		scsi_rescan_device(&sdev->sdev_gendev);
2462 		envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2463 		break;
2464 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2465 		envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2466 		break;
2467 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2468 	       envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2469 		break;
2470 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2471 		envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2472 		break;
2473 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2474 		envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2475 		break;
2476 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2477 		envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2478 		break;
2479 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2480 		envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2481 		break;
2482 	default:
2483 		/* do nothing */
2484 		break;
2485 	}
2486 
2487 	envp[idx++] = NULL;
2488 
2489 	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2490 }
2491 
2492 /**
2493  *	scsi_evt_thread - send a uevent for each scsi event
2494  *	@work: work struct for scsi_device
2495  *
2496  *	Dispatch queued events to their associated scsi_device kobjects
2497  *	as uevents.
2498  */
2499 void scsi_evt_thread(struct work_struct *work)
2500 {
2501 	struct scsi_device *sdev;
2502 	enum scsi_device_event evt_type;
2503 	LIST_HEAD(event_list);
2504 
2505 	sdev = container_of(work, struct scsi_device, event_work);
2506 
2507 	for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2508 		if (test_and_clear_bit(evt_type, sdev->pending_events))
2509 			sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2510 
2511 	while (1) {
2512 		struct scsi_event *evt;
2513 		struct list_head *this, *tmp;
2514 		unsigned long flags;
2515 
2516 		spin_lock_irqsave(&sdev->list_lock, flags);
2517 		list_splice_init(&sdev->event_list, &event_list);
2518 		spin_unlock_irqrestore(&sdev->list_lock, flags);
2519 
2520 		if (list_empty(&event_list))
2521 			break;
2522 
2523 		list_for_each_safe(this, tmp, &event_list) {
2524 			evt = list_entry(this, struct scsi_event, node);
2525 			list_del(&evt->node);
2526 			scsi_evt_emit(sdev, evt);
2527 			kfree(evt);
2528 		}
2529 	}
2530 }
2531 
2532 /**
2533  * 	sdev_evt_send - send asserted event to uevent thread
2534  *	@sdev: scsi_device event occurred on
2535  *	@evt: event to send
2536  *
2537  *	Assert scsi device event asynchronously.
2538  */
2539 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2540 {
2541 	unsigned long flags;
2542 
2543 #if 0
2544 	/* FIXME: currently this check eliminates all media change events
2545 	 * for polled devices.  Need to update to discriminate between AN
2546 	 * and polled events */
2547 	if (!test_bit(evt->evt_type, sdev->supported_events)) {
2548 		kfree(evt);
2549 		return;
2550 	}
2551 #endif
2552 
2553 	spin_lock_irqsave(&sdev->list_lock, flags);
2554 	list_add_tail(&evt->node, &sdev->event_list);
2555 	schedule_work(&sdev->event_work);
2556 	spin_unlock_irqrestore(&sdev->list_lock, flags);
2557 }
2558 EXPORT_SYMBOL_GPL(sdev_evt_send);
2559 
2560 /**
2561  * 	sdev_evt_alloc - allocate a new scsi event
2562  *	@evt_type: type of event to allocate
2563  *	@gfpflags: GFP flags for allocation
2564  *
2565  *	Allocates and returns a new scsi_event.
2566  */
2567 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2568 				  gfp_t gfpflags)
2569 {
2570 	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2571 	if (!evt)
2572 		return NULL;
2573 
2574 	evt->evt_type = evt_type;
2575 	INIT_LIST_HEAD(&evt->node);
2576 
2577 	/* evt_type-specific initialization, if any */
2578 	switch (evt_type) {
2579 	case SDEV_EVT_MEDIA_CHANGE:
2580 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2581 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2582 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2583 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2584 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2585 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2586 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2587 	default:
2588 		/* do nothing */
2589 		break;
2590 	}
2591 
2592 	return evt;
2593 }
2594 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2595 
2596 /**
2597  * 	sdev_evt_send_simple - send asserted event to uevent thread
2598  *	@sdev: scsi_device event occurred on
2599  *	@evt_type: type of event to send
2600  *	@gfpflags: GFP flags for allocation
2601  *
2602  *	Assert scsi device event asynchronously, given an event type.
2603  */
2604 void sdev_evt_send_simple(struct scsi_device *sdev,
2605 			  enum scsi_device_event evt_type, gfp_t gfpflags)
2606 {
2607 	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2608 	if (!evt) {
2609 		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2610 			    evt_type);
2611 		return;
2612 	}
2613 
2614 	sdev_evt_send(sdev, evt);
2615 }
2616 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2617 
2618 /**
2619  *	scsi_device_quiesce - Block all commands except power management.
2620  *	@sdev:	scsi device to quiesce.
2621  *
2622  *	This works by trying to transition to the SDEV_QUIESCE state
2623  *	(which must be a legal transition).  When the device is in this
2624  *	state, only power management requests will be accepted, all others will
2625  *	be deferred.
2626  *
2627  *	Must be called with user context, may sleep.
2628  *
2629  *	Returns zero if unsuccessful or an error if not.
2630  */
2631 int
2632 scsi_device_quiesce(struct scsi_device *sdev)
2633 {
2634 	struct request_queue *q = sdev->request_queue;
2635 	int err;
2636 
2637 	/*
2638 	 * It is allowed to call scsi_device_quiesce() multiple times from
2639 	 * the same context but concurrent scsi_device_quiesce() calls are
2640 	 * not allowed.
2641 	 */
2642 	WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2643 
2644 	if (sdev->quiesced_by == current)
2645 		return 0;
2646 
2647 	blk_set_pm_only(q);
2648 
2649 	blk_mq_freeze_queue(q);
2650 	/*
2651 	 * Ensure that the effect of blk_set_pm_only() will be visible
2652 	 * for percpu_ref_tryget() callers that occur after the queue
2653 	 * unfreeze even if the queue was already frozen before this function
2654 	 * was called. See also https://lwn.net/Articles/573497/.
2655 	 */
2656 	synchronize_rcu();
2657 	blk_mq_unfreeze_queue(q);
2658 
2659 	mutex_lock(&sdev->state_mutex);
2660 	err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2661 	if (err == 0)
2662 		sdev->quiesced_by = current;
2663 	else
2664 		blk_clear_pm_only(q);
2665 	mutex_unlock(&sdev->state_mutex);
2666 
2667 	return err;
2668 }
2669 EXPORT_SYMBOL(scsi_device_quiesce);
2670 
2671 /**
2672  *	scsi_device_resume - Restart user issued commands to a quiesced device.
2673  *	@sdev:	scsi device to resume.
2674  *
2675  *	Moves the device from quiesced back to running and restarts the
2676  *	queues.
2677  *
2678  *	Must be called with user context, may sleep.
2679  */
2680 void scsi_device_resume(struct scsi_device *sdev)
2681 {
2682 	/* check if the device state was mutated prior to resume, and if
2683 	 * so assume the state is being managed elsewhere (for example
2684 	 * device deleted during suspend)
2685 	 */
2686 	mutex_lock(&sdev->state_mutex);
2687 	if (sdev->sdev_state == SDEV_QUIESCE)
2688 		scsi_device_set_state(sdev, SDEV_RUNNING);
2689 	if (sdev->quiesced_by) {
2690 		sdev->quiesced_by = NULL;
2691 		blk_clear_pm_only(sdev->request_queue);
2692 	}
2693 	mutex_unlock(&sdev->state_mutex);
2694 }
2695 EXPORT_SYMBOL(scsi_device_resume);
2696 
2697 static void
2698 device_quiesce_fn(struct scsi_device *sdev, void *data)
2699 {
2700 	scsi_device_quiesce(sdev);
2701 }
2702 
2703 void
2704 scsi_target_quiesce(struct scsi_target *starget)
2705 {
2706 	starget_for_each_device(starget, NULL, device_quiesce_fn);
2707 }
2708 EXPORT_SYMBOL(scsi_target_quiesce);
2709 
2710 static void
2711 device_resume_fn(struct scsi_device *sdev, void *data)
2712 {
2713 	scsi_device_resume(sdev);
2714 }
2715 
2716 void
2717 scsi_target_resume(struct scsi_target *starget)
2718 {
2719 	starget_for_each_device(starget, NULL, device_resume_fn);
2720 }
2721 EXPORT_SYMBOL(scsi_target_resume);
2722 
2723 static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2724 {
2725 	if (scsi_device_set_state(sdev, SDEV_BLOCK))
2726 		return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2727 
2728 	return 0;
2729 }
2730 
2731 void scsi_start_queue(struct scsi_device *sdev)
2732 {
2733 	if (cmpxchg(&sdev->queue_stopped, 1, 0))
2734 		blk_mq_unquiesce_queue(sdev->request_queue);
2735 }
2736 
2737 static void scsi_stop_queue(struct scsi_device *sdev)
2738 {
2739 	/*
2740 	 * The atomic variable of ->queue_stopped covers that
2741 	 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2742 	 *
2743 	 * The caller needs to wait until quiesce is done.
2744 	 */
2745 	if (!cmpxchg(&sdev->queue_stopped, 0, 1))
2746 		blk_mq_quiesce_queue_nowait(sdev->request_queue);
2747 }
2748 
2749 /**
2750  * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2751  * @sdev: device to block
2752  *
2753  * Pause SCSI command processing on the specified device. Does not sleep.
2754  *
2755  * Returns zero if successful or a negative error code upon failure.
2756  *
2757  * Notes:
2758  * This routine transitions the device to the SDEV_BLOCK state (which must be
2759  * a legal transition). When the device is in this state, command processing
2760  * is paused until the device leaves the SDEV_BLOCK state. See also
2761  * scsi_internal_device_unblock_nowait().
2762  */
2763 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2764 {
2765 	int ret = __scsi_internal_device_block_nowait(sdev);
2766 
2767 	/*
2768 	 * The device has transitioned to SDEV_BLOCK.  Stop the
2769 	 * block layer from calling the midlayer with this device's
2770 	 * request queue.
2771 	 */
2772 	if (!ret)
2773 		scsi_stop_queue(sdev);
2774 	return ret;
2775 }
2776 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2777 
2778 /**
2779  * scsi_device_block - try to transition to the SDEV_BLOCK state
2780  * @sdev: device to block
2781  * @data: dummy argument, ignored
2782  *
2783  * Pause SCSI command processing on the specified device. Callers must wait
2784  * until all ongoing scsi_queue_rq() calls have finished after this function
2785  * returns.
2786  *
2787  * Note:
2788  * This routine transitions the device to the SDEV_BLOCK state (which must be
2789  * a legal transition). When the device is in this state, command processing
2790  * is paused until the device leaves the SDEV_BLOCK state. See also
2791  * scsi_internal_device_unblock().
2792  */
2793 static void scsi_device_block(struct scsi_device *sdev, void *data)
2794 {
2795 	int err;
2796 	enum scsi_device_state state;
2797 
2798 	mutex_lock(&sdev->state_mutex);
2799 	err = __scsi_internal_device_block_nowait(sdev);
2800 	state = sdev->sdev_state;
2801 	if (err == 0)
2802 		/*
2803 		 * scsi_stop_queue() must be called with the state_mutex
2804 		 * held. Otherwise a simultaneous scsi_start_queue() call
2805 		 * might unquiesce the queue before we quiesce it.
2806 		 */
2807 		scsi_stop_queue(sdev);
2808 
2809 	mutex_unlock(&sdev->state_mutex);
2810 
2811 	WARN_ONCE(err, "%s: failed to block %s in state %d\n",
2812 		  __func__, dev_name(&sdev->sdev_gendev), state);
2813 }
2814 
2815 /**
2816  * scsi_internal_device_unblock_nowait - resume a device after a block request
2817  * @sdev:	device to resume
2818  * @new_state:	state to set the device to after unblocking
2819  *
2820  * Restart the device queue for a previously suspended SCSI device. Does not
2821  * sleep.
2822  *
2823  * Returns zero if successful or a negative error code upon failure.
2824  *
2825  * Notes:
2826  * This routine transitions the device to the SDEV_RUNNING state or to one of
2827  * the offline states (which must be a legal transition) allowing the midlayer
2828  * to goose the queue for this device.
2829  */
2830 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2831 					enum scsi_device_state new_state)
2832 {
2833 	switch (new_state) {
2834 	case SDEV_RUNNING:
2835 	case SDEV_TRANSPORT_OFFLINE:
2836 		break;
2837 	default:
2838 		return -EINVAL;
2839 	}
2840 
2841 	/*
2842 	 * Try to transition the scsi device to SDEV_RUNNING or one of the
2843 	 * offlined states and goose the device queue if successful.
2844 	 */
2845 	switch (sdev->sdev_state) {
2846 	case SDEV_BLOCK:
2847 	case SDEV_TRANSPORT_OFFLINE:
2848 		sdev->sdev_state = new_state;
2849 		break;
2850 	case SDEV_CREATED_BLOCK:
2851 		if (new_state == SDEV_TRANSPORT_OFFLINE ||
2852 		    new_state == SDEV_OFFLINE)
2853 			sdev->sdev_state = new_state;
2854 		else
2855 			sdev->sdev_state = SDEV_CREATED;
2856 		break;
2857 	case SDEV_CANCEL:
2858 	case SDEV_OFFLINE:
2859 		break;
2860 	default:
2861 		return -EINVAL;
2862 	}
2863 	scsi_start_queue(sdev);
2864 
2865 	return 0;
2866 }
2867 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2868 
2869 /**
2870  * scsi_internal_device_unblock - resume a device after a block request
2871  * @sdev:	device to resume
2872  * @new_state:	state to set the device to after unblocking
2873  *
2874  * Restart the device queue for a previously suspended SCSI device. May sleep.
2875  *
2876  * Returns zero if successful or a negative error code upon failure.
2877  *
2878  * Notes:
2879  * This routine transitions the device to the SDEV_RUNNING state or to one of
2880  * the offline states (which must be a legal transition) allowing the midlayer
2881  * to goose the queue for this device.
2882  */
2883 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2884 					enum scsi_device_state new_state)
2885 {
2886 	int ret;
2887 
2888 	mutex_lock(&sdev->state_mutex);
2889 	ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2890 	mutex_unlock(&sdev->state_mutex);
2891 
2892 	return ret;
2893 }
2894 
2895 static int
2896 target_block(struct device *dev, void *data)
2897 {
2898 	if (scsi_is_target_device(dev))
2899 		starget_for_each_device(to_scsi_target(dev), NULL,
2900 					scsi_device_block);
2901 	return 0;
2902 }
2903 
2904 /**
2905  * scsi_block_targets - transition all SCSI child devices to SDEV_BLOCK state
2906  * @dev: a parent device of one or more scsi_target devices
2907  * @shost: the Scsi_Host to which this device belongs
2908  *
2909  * Iterate over all children of @dev, which should be scsi_target devices,
2910  * and switch all subordinate scsi devices to SDEV_BLOCK state. Wait for
2911  * ongoing scsi_queue_rq() calls to finish. May sleep.
2912  *
2913  * Note:
2914  * @dev must not itself be a scsi_target device.
2915  */
2916 void
2917 scsi_block_targets(struct Scsi_Host *shost, struct device *dev)
2918 {
2919 	WARN_ON_ONCE(scsi_is_target_device(dev));
2920 	device_for_each_child(dev, NULL, target_block);
2921 	blk_mq_wait_quiesce_done(&shost->tag_set);
2922 }
2923 EXPORT_SYMBOL_GPL(scsi_block_targets);
2924 
2925 static void
2926 device_unblock(struct scsi_device *sdev, void *data)
2927 {
2928 	scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2929 }
2930 
2931 static int
2932 target_unblock(struct device *dev, void *data)
2933 {
2934 	if (scsi_is_target_device(dev))
2935 		starget_for_each_device(to_scsi_target(dev), data,
2936 					device_unblock);
2937 	return 0;
2938 }
2939 
2940 void
2941 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2942 {
2943 	if (scsi_is_target_device(dev))
2944 		starget_for_each_device(to_scsi_target(dev), &new_state,
2945 					device_unblock);
2946 	else
2947 		device_for_each_child(dev, &new_state, target_unblock);
2948 }
2949 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2950 
2951 /**
2952  * scsi_host_block - Try to transition all logical units to the SDEV_BLOCK state
2953  * @shost: device to block
2954  *
2955  * Pause SCSI command processing for all logical units associated with the SCSI
2956  * host and wait until pending scsi_queue_rq() calls have finished.
2957  *
2958  * Returns zero if successful or a negative error code upon failure.
2959  */
2960 int
2961 scsi_host_block(struct Scsi_Host *shost)
2962 {
2963 	struct scsi_device *sdev;
2964 	int ret;
2965 
2966 	/*
2967 	 * Call scsi_internal_device_block_nowait so we can avoid
2968 	 * calling synchronize_rcu() for each LUN.
2969 	 */
2970 	shost_for_each_device(sdev, shost) {
2971 		mutex_lock(&sdev->state_mutex);
2972 		ret = scsi_internal_device_block_nowait(sdev);
2973 		mutex_unlock(&sdev->state_mutex);
2974 		if (ret) {
2975 			scsi_device_put(sdev);
2976 			return ret;
2977 		}
2978 	}
2979 
2980 	/* Wait for ongoing scsi_queue_rq() calls to finish. */
2981 	blk_mq_wait_quiesce_done(&shost->tag_set);
2982 
2983 	return 0;
2984 }
2985 EXPORT_SYMBOL_GPL(scsi_host_block);
2986 
2987 int
2988 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2989 {
2990 	struct scsi_device *sdev;
2991 	int ret = 0;
2992 
2993 	shost_for_each_device(sdev, shost) {
2994 		ret = scsi_internal_device_unblock(sdev, new_state);
2995 		if (ret) {
2996 			scsi_device_put(sdev);
2997 			break;
2998 		}
2999 	}
3000 	return ret;
3001 }
3002 EXPORT_SYMBOL_GPL(scsi_host_unblock);
3003 
3004 /**
3005  * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3006  * @sgl:	scatter-gather list
3007  * @sg_count:	number of segments in sg
3008  * @offset:	offset in bytes into sg, on return offset into the mapped area
3009  * @len:	bytes to map, on return number of bytes mapped
3010  *
3011  * Returns virtual address of the start of the mapped page
3012  */
3013 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3014 			  size_t *offset, size_t *len)
3015 {
3016 	int i;
3017 	size_t sg_len = 0, len_complete = 0;
3018 	struct scatterlist *sg;
3019 	struct page *page;
3020 
3021 	WARN_ON(!irqs_disabled());
3022 
3023 	for_each_sg(sgl, sg, sg_count, i) {
3024 		len_complete = sg_len; /* Complete sg-entries */
3025 		sg_len += sg->length;
3026 		if (sg_len > *offset)
3027 			break;
3028 	}
3029 
3030 	if (unlikely(i == sg_count)) {
3031 		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3032 			"elements %d\n",
3033 		       __func__, sg_len, *offset, sg_count);
3034 		WARN_ON(1);
3035 		return NULL;
3036 	}
3037 
3038 	/* Offset starting from the beginning of first page in this sg-entry */
3039 	*offset = *offset - len_complete + sg->offset;
3040 
3041 	/* Assumption: contiguous pages can be accessed as "page + i" */
3042 	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3043 	*offset &= ~PAGE_MASK;
3044 
3045 	/* Bytes in this sg-entry from *offset to the end of the page */
3046 	sg_len = PAGE_SIZE - *offset;
3047 	if (*len > sg_len)
3048 		*len = sg_len;
3049 
3050 	return kmap_atomic(page);
3051 }
3052 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3053 
3054 /**
3055  * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3056  * @virt:	virtual address to be unmapped
3057  */
3058 void scsi_kunmap_atomic_sg(void *virt)
3059 {
3060 	kunmap_atomic(virt);
3061 }
3062 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3063 
3064 void sdev_disable_disk_events(struct scsi_device *sdev)
3065 {
3066 	atomic_inc(&sdev->disk_events_disable_depth);
3067 }
3068 EXPORT_SYMBOL(sdev_disable_disk_events);
3069 
3070 void sdev_enable_disk_events(struct scsi_device *sdev)
3071 {
3072 	if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3073 		return;
3074 	atomic_dec(&sdev->disk_events_disable_depth);
3075 }
3076 EXPORT_SYMBOL(sdev_enable_disk_events);
3077 
3078 static unsigned char designator_prio(const unsigned char *d)
3079 {
3080 	if (d[1] & 0x30)
3081 		/* not associated with LUN */
3082 		return 0;
3083 
3084 	if (d[3] == 0)
3085 		/* invalid length */
3086 		return 0;
3087 
3088 	/*
3089 	 * Order of preference for lun descriptor:
3090 	 * - SCSI name string
3091 	 * - NAA IEEE Registered Extended
3092 	 * - EUI-64 based 16-byte
3093 	 * - EUI-64 based 12-byte
3094 	 * - NAA IEEE Registered
3095 	 * - NAA IEEE Extended
3096 	 * - EUI-64 based 8-byte
3097 	 * - SCSI name string (truncated)
3098 	 * - T10 Vendor ID
3099 	 * as longer descriptors reduce the likelyhood
3100 	 * of identification clashes.
3101 	 */
3102 
3103 	switch (d[1] & 0xf) {
3104 	case 8:
3105 		/* SCSI name string, variable-length UTF-8 */
3106 		return 9;
3107 	case 3:
3108 		switch (d[4] >> 4) {
3109 		case 6:
3110 			/* NAA registered extended */
3111 			return 8;
3112 		case 5:
3113 			/* NAA registered */
3114 			return 5;
3115 		case 4:
3116 			/* NAA extended */
3117 			return 4;
3118 		case 3:
3119 			/* NAA locally assigned */
3120 			return 1;
3121 		default:
3122 			break;
3123 		}
3124 		break;
3125 	case 2:
3126 		switch (d[3]) {
3127 		case 16:
3128 			/* EUI64-based, 16 byte */
3129 			return 7;
3130 		case 12:
3131 			/* EUI64-based, 12 byte */
3132 			return 6;
3133 		case 8:
3134 			/* EUI64-based, 8 byte */
3135 			return 3;
3136 		default:
3137 			break;
3138 		}
3139 		break;
3140 	case 1:
3141 		/* T10 vendor ID */
3142 		return 1;
3143 	default:
3144 		break;
3145 	}
3146 
3147 	return 0;
3148 }
3149 
3150 /**
3151  * scsi_vpd_lun_id - return a unique device identification
3152  * @sdev: SCSI device
3153  * @id:   buffer for the identification
3154  * @id_len:  length of the buffer
3155  *
3156  * Copies a unique device identification into @id based
3157  * on the information in the VPD page 0x83 of the device.
3158  * The string will be formatted as a SCSI name string.
3159  *
3160  * Returns the length of the identification or error on failure.
3161  * If the identifier is longer than the supplied buffer the actual
3162  * identifier length is returned and the buffer is not zero-padded.
3163  */
3164 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3165 {
3166 	u8 cur_id_prio = 0;
3167 	u8 cur_id_size = 0;
3168 	const unsigned char *d, *cur_id_str;
3169 	const struct scsi_vpd *vpd_pg83;
3170 	int id_size = -EINVAL;
3171 
3172 	rcu_read_lock();
3173 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3174 	if (!vpd_pg83) {
3175 		rcu_read_unlock();
3176 		return -ENXIO;
3177 	}
3178 
3179 	/* The id string must be at least 20 bytes + terminating NULL byte */
3180 	if (id_len < 21) {
3181 		rcu_read_unlock();
3182 		return -EINVAL;
3183 	}
3184 
3185 	memset(id, 0, id_len);
3186 	for (d = vpd_pg83->data + 4;
3187 	     d < vpd_pg83->data + vpd_pg83->len;
3188 	     d += d[3] + 4) {
3189 		u8 prio = designator_prio(d);
3190 
3191 		if (prio == 0 || cur_id_prio > prio)
3192 			continue;
3193 
3194 		switch (d[1] & 0xf) {
3195 		case 0x1:
3196 			/* T10 Vendor ID */
3197 			if (cur_id_size > d[3])
3198 				break;
3199 			cur_id_prio = prio;
3200 			cur_id_size = d[3];
3201 			if (cur_id_size + 4 > id_len)
3202 				cur_id_size = id_len - 4;
3203 			cur_id_str = d + 4;
3204 			id_size = snprintf(id, id_len, "t10.%*pE",
3205 					   cur_id_size, cur_id_str);
3206 			break;
3207 		case 0x2:
3208 			/* EUI-64 */
3209 			cur_id_prio = prio;
3210 			cur_id_size = d[3];
3211 			cur_id_str = d + 4;
3212 			switch (cur_id_size) {
3213 			case 8:
3214 				id_size = snprintf(id, id_len,
3215 						   "eui.%8phN",
3216 						   cur_id_str);
3217 				break;
3218 			case 12:
3219 				id_size = snprintf(id, id_len,
3220 						   "eui.%12phN",
3221 						   cur_id_str);
3222 				break;
3223 			case 16:
3224 				id_size = snprintf(id, id_len,
3225 						   "eui.%16phN",
3226 						   cur_id_str);
3227 				break;
3228 			default:
3229 				break;
3230 			}
3231 			break;
3232 		case 0x3:
3233 			/* NAA */
3234 			cur_id_prio = prio;
3235 			cur_id_size = d[3];
3236 			cur_id_str = d + 4;
3237 			switch (cur_id_size) {
3238 			case 8:
3239 				id_size = snprintf(id, id_len,
3240 						   "naa.%8phN",
3241 						   cur_id_str);
3242 				break;
3243 			case 16:
3244 				id_size = snprintf(id, id_len,
3245 						   "naa.%16phN",
3246 						   cur_id_str);
3247 				break;
3248 			default:
3249 				break;
3250 			}
3251 			break;
3252 		case 0x8:
3253 			/* SCSI name string */
3254 			if (cur_id_size > d[3])
3255 				break;
3256 			/* Prefer others for truncated descriptor */
3257 			if (d[3] > id_len) {
3258 				prio = 2;
3259 				if (cur_id_prio > prio)
3260 					break;
3261 			}
3262 			cur_id_prio = prio;
3263 			cur_id_size = id_size = d[3];
3264 			cur_id_str = d + 4;
3265 			if (cur_id_size >= id_len)
3266 				cur_id_size = id_len - 1;
3267 			memcpy(id, cur_id_str, cur_id_size);
3268 			break;
3269 		default:
3270 			break;
3271 		}
3272 	}
3273 	rcu_read_unlock();
3274 
3275 	return id_size;
3276 }
3277 EXPORT_SYMBOL(scsi_vpd_lun_id);
3278 
3279 /*
3280  * scsi_vpd_tpg_id - return a target port group identifier
3281  * @sdev: SCSI device
3282  *
3283  * Returns the Target Port Group identifier from the information
3284  * froom VPD page 0x83 of the device.
3285  *
3286  * Returns the identifier or error on failure.
3287  */
3288 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3289 {
3290 	const unsigned char *d;
3291 	const struct scsi_vpd *vpd_pg83;
3292 	int group_id = -EAGAIN, rel_port = -1;
3293 
3294 	rcu_read_lock();
3295 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3296 	if (!vpd_pg83) {
3297 		rcu_read_unlock();
3298 		return -ENXIO;
3299 	}
3300 
3301 	d = vpd_pg83->data + 4;
3302 	while (d < vpd_pg83->data + vpd_pg83->len) {
3303 		switch (d[1] & 0xf) {
3304 		case 0x4:
3305 			/* Relative target port */
3306 			rel_port = get_unaligned_be16(&d[6]);
3307 			break;
3308 		case 0x5:
3309 			/* Target port group */
3310 			group_id = get_unaligned_be16(&d[6]);
3311 			break;
3312 		default:
3313 			break;
3314 		}
3315 		d += d[3] + 4;
3316 	}
3317 	rcu_read_unlock();
3318 
3319 	if (group_id >= 0 && rel_id && rel_port != -1)
3320 		*rel_id = rel_port;
3321 
3322 	return group_id;
3323 }
3324 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3325 
3326 /**
3327  * scsi_build_sense - build sense data for a command
3328  * @scmd:	scsi command for which the sense should be formatted
3329  * @desc:	Sense format (non-zero == descriptor format,
3330  *              0 == fixed format)
3331  * @key:	Sense key
3332  * @asc:	Additional sense code
3333  * @ascq:	Additional sense code qualifier
3334  *
3335  **/
3336 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3337 {
3338 	scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3339 	scmd->result = SAM_STAT_CHECK_CONDITION;
3340 }
3341 EXPORT_SYMBOL_GPL(scsi_build_sense);
3342