xref: /openbmc/linux/drivers/scsi/scsi_lib.c (revision 2cf1c348)
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 /*
79  * When to reinvoke queueing after a resource shortage. It's 3 msecs to
80  * not change behaviour from the previous unplug mechanism, experimentation
81  * may prove this needs changing.
82  */
83 #define SCSI_QUEUE_DELAY	3
84 
85 static void
86 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
87 {
88 	struct Scsi_Host *host = cmd->device->host;
89 	struct scsi_device *device = cmd->device;
90 	struct scsi_target *starget = scsi_target(device);
91 
92 	/*
93 	 * Set the appropriate busy bit for the device/host.
94 	 *
95 	 * If the host/device isn't busy, assume that something actually
96 	 * completed, and that we should be able to queue a command now.
97 	 *
98 	 * Note that the prior mid-layer assumption that any host could
99 	 * always queue at least one command is now broken.  The mid-layer
100 	 * will implement a user specifiable stall (see
101 	 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
102 	 * if a command is requeued with no other commands outstanding
103 	 * either for the device or for the host.
104 	 */
105 	switch (reason) {
106 	case SCSI_MLQUEUE_HOST_BUSY:
107 		atomic_set(&host->host_blocked, host->max_host_blocked);
108 		break;
109 	case SCSI_MLQUEUE_DEVICE_BUSY:
110 	case SCSI_MLQUEUE_EH_RETRY:
111 		atomic_set(&device->device_blocked,
112 			   device->max_device_blocked);
113 		break;
114 	case SCSI_MLQUEUE_TARGET_BUSY:
115 		atomic_set(&starget->target_blocked,
116 			   starget->max_target_blocked);
117 		break;
118 	}
119 }
120 
121 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
122 {
123 	struct request *rq = scsi_cmd_to_rq(cmd);
124 
125 	if (rq->rq_flags & RQF_DONTPREP) {
126 		rq->rq_flags &= ~RQF_DONTPREP;
127 		scsi_mq_uninit_cmd(cmd);
128 	} else {
129 		WARN_ON_ONCE(true);
130 	}
131 	blk_mq_requeue_request(rq, true);
132 }
133 
134 /**
135  * __scsi_queue_insert - private queue insertion
136  * @cmd: The SCSI command being requeued
137  * @reason:  The reason for the requeue
138  * @unbusy: Whether the queue should be unbusied
139  *
140  * This is a private queue insertion.  The public interface
141  * scsi_queue_insert() always assumes the queue should be unbusied
142  * because it's always called before the completion.  This function is
143  * for a requeue after completion, which should only occur in this
144  * file.
145  */
146 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
147 {
148 	struct scsi_device *device = cmd->device;
149 
150 	SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
151 		"Inserting command %p into mlqueue\n", cmd));
152 
153 	scsi_set_blocked(cmd, reason);
154 
155 	/*
156 	 * Decrement the counters, since these commands are no longer
157 	 * active on the host/device.
158 	 */
159 	if (unbusy)
160 		scsi_device_unbusy(device, cmd);
161 
162 	/*
163 	 * Requeue this command.  It will go before all other commands
164 	 * that are already in the queue. Schedule requeue work under
165 	 * lock such that the kblockd_schedule_work() call happens
166 	 * before blk_cleanup_queue() finishes.
167 	 */
168 	cmd->result = 0;
169 
170 	blk_mq_requeue_request(scsi_cmd_to_rq(cmd), true);
171 }
172 
173 /**
174  * scsi_queue_insert - Reinsert a command in the queue.
175  * @cmd:    command that we are adding to queue.
176  * @reason: why we are inserting command to queue.
177  *
178  * We do this for one of two cases. Either the host is busy and it cannot accept
179  * any more commands for the time being, or the device returned QUEUE_FULL and
180  * can accept no more commands.
181  *
182  * Context: This could be called either from an interrupt context or a normal
183  * process context.
184  */
185 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
186 {
187 	__scsi_queue_insert(cmd, reason, true);
188 }
189 
190 
191 /**
192  * __scsi_execute - insert request and wait for the result
193  * @sdev:	scsi device
194  * @cmd:	scsi command
195  * @data_direction: data direction
196  * @buffer:	data buffer
197  * @bufflen:	len of buffer
198  * @sense:	optional sense buffer
199  * @sshdr:	optional decoded sense header
200  * @timeout:	request timeout in HZ
201  * @retries:	number of times to retry request
202  * @flags:	flags for ->cmd_flags
203  * @rq_flags:	flags for ->rq_flags
204  * @resid:	optional residual length
205  *
206  * Returns the scsi_cmnd result field if a command was executed, or a negative
207  * Linux error code if we didn't get that far.
208  */
209 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
210 		 int data_direction, void *buffer, unsigned bufflen,
211 		 unsigned char *sense, struct scsi_sense_hdr *sshdr,
212 		 int timeout, int retries, u64 flags, req_flags_t rq_flags,
213 		 int *resid)
214 {
215 	struct request *req;
216 	struct scsi_request *rq;
217 	int ret;
218 
219 	req = scsi_alloc_request(sdev->request_queue,
220 			data_direction == DMA_TO_DEVICE ?
221 			REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
222 			rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0);
223 	if (IS_ERR(req))
224 		return PTR_ERR(req);
225 
226 	rq = scsi_req(req);
227 
228 	if (bufflen) {
229 		ret = blk_rq_map_kern(sdev->request_queue, req,
230 				      buffer, bufflen, GFP_NOIO);
231 		if (ret)
232 			goto out;
233 	}
234 	rq->cmd_len = COMMAND_SIZE(cmd[0]);
235 	memcpy(rq->cmd, cmd, rq->cmd_len);
236 	rq->retries = retries;
237 	req->timeout = timeout;
238 	req->cmd_flags |= flags;
239 	req->rq_flags |= rq_flags | RQF_QUIET;
240 
241 	/*
242 	 * head injection *required* here otherwise quiesce won't work
243 	 */
244 	blk_execute_rq(req, true);
245 
246 	/*
247 	 * Some devices (USB mass-storage in particular) may transfer
248 	 * garbage data together with a residue indicating that the data
249 	 * is invalid.  Prevent the garbage from being misinterpreted
250 	 * and prevent security leaks by zeroing out the excess data.
251 	 */
252 	if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
253 		memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
254 
255 	if (resid)
256 		*resid = rq->resid_len;
257 	if (sense && rq->sense_len)
258 		memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
259 	if (sshdr)
260 		scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
261 	ret = rq->result;
262  out:
263 	blk_mq_free_request(req);
264 
265 	return ret;
266 }
267 EXPORT_SYMBOL(__scsi_execute);
268 
269 /*
270  * Wake up the error handler if necessary. Avoid as follows that the error
271  * handler is not woken up if host in-flight requests number ==
272  * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
273  * with an RCU read lock in this function to ensure that this function in
274  * its entirety either finishes before scsi_eh_scmd_add() increases the
275  * host_failed counter or that it notices the shost state change made by
276  * scsi_eh_scmd_add().
277  */
278 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
279 {
280 	unsigned long flags;
281 
282 	rcu_read_lock();
283 	__clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
284 	if (unlikely(scsi_host_in_recovery(shost))) {
285 		spin_lock_irqsave(shost->host_lock, flags);
286 		if (shost->host_failed || shost->host_eh_scheduled)
287 			scsi_eh_wakeup(shost);
288 		spin_unlock_irqrestore(shost->host_lock, flags);
289 	}
290 	rcu_read_unlock();
291 }
292 
293 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
294 {
295 	struct Scsi_Host *shost = sdev->host;
296 	struct scsi_target *starget = scsi_target(sdev);
297 
298 	scsi_dec_host_busy(shost, cmd);
299 
300 	if (starget->can_queue > 0)
301 		atomic_dec(&starget->target_busy);
302 
303 	sbitmap_put(&sdev->budget_map, cmd->budget_token);
304 	cmd->budget_token = -1;
305 }
306 
307 static void scsi_kick_queue(struct request_queue *q)
308 {
309 	blk_mq_run_hw_queues(q, false);
310 }
311 
312 /*
313  * Called for single_lun devices on IO completion. Clear starget_sdev_user,
314  * and call blk_run_queue for all the scsi_devices on the target -
315  * including current_sdev first.
316  *
317  * Called with *no* scsi locks held.
318  */
319 static void scsi_single_lun_run(struct scsi_device *current_sdev)
320 {
321 	struct Scsi_Host *shost = current_sdev->host;
322 	struct scsi_device *sdev, *tmp;
323 	struct scsi_target *starget = scsi_target(current_sdev);
324 	unsigned long flags;
325 
326 	spin_lock_irqsave(shost->host_lock, flags);
327 	starget->starget_sdev_user = NULL;
328 	spin_unlock_irqrestore(shost->host_lock, flags);
329 
330 	/*
331 	 * Call blk_run_queue for all LUNs on the target, starting with
332 	 * current_sdev. We race with others (to set starget_sdev_user),
333 	 * but in most cases, we will be first. Ideally, each LU on the
334 	 * target would get some limited time or requests on the target.
335 	 */
336 	scsi_kick_queue(current_sdev->request_queue);
337 
338 	spin_lock_irqsave(shost->host_lock, flags);
339 	if (starget->starget_sdev_user)
340 		goto out;
341 	list_for_each_entry_safe(sdev, tmp, &starget->devices,
342 			same_target_siblings) {
343 		if (sdev == current_sdev)
344 			continue;
345 		if (scsi_device_get(sdev))
346 			continue;
347 
348 		spin_unlock_irqrestore(shost->host_lock, flags);
349 		scsi_kick_queue(sdev->request_queue);
350 		spin_lock_irqsave(shost->host_lock, flags);
351 
352 		scsi_device_put(sdev);
353 	}
354  out:
355 	spin_unlock_irqrestore(shost->host_lock, flags);
356 }
357 
358 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
359 {
360 	if (scsi_device_busy(sdev) >= sdev->queue_depth)
361 		return true;
362 	if (atomic_read(&sdev->device_blocked) > 0)
363 		return true;
364 	return false;
365 }
366 
367 static inline bool scsi_target_is_busy(struct scsi_target *starget)
368 {
369 	if (starget->can_queue > 0) {
370 		if (atomic_read(&starget->target_busy) >= starget->can_queue)
371 			return true;
372 		if (atomic_read(&starget->target_blocked) > 0)
373 			return true;
374 	}
375 	return false;
376 }
377 
378 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
379 {
380 	if (atomic_read(&shost->host_blocked) > 0)
381 		return true;
382 	if (shost->host_self_blocked)
383 		return true;
384 	return false;
385 }
386 
387 static void scsi_starved_list_run(struct Scsi_Host *shost)
388 {
389 	LIST_HEAD(starved_list);
390 	struct scsi_device *sdev;
391 	unsigned long flags;
392 
393 	spin_lock_irqsave(shost->host_lock, flags);
394 	list_splice_init(&shost->starved_list, &starved_list);
395 
396 	while (!list_empty(&starved_list)) {
397 		struct request_queue *slq;
398 
399 		/*
400 		 * As long as shost is accepting commands and we have
401 		 * starved queues, call blk_run_queue. scsi_request_fn
402 		 * drops the queue_lock and can add us back to the
403 		 * starved_list.
404 		 *
405 		 * host_lock protects the starved_list and starved_entry.
406 		 * scsi_request_fn must get the host_lock before checking
407 		 * or modifying starved_list or starved_entry.
408 		 */
409 		if (scsi_host_is_busy(shost))
410 			break;
411 
412 		sdev = list_entry(starved_list.next,
413 				  struct scsi_device, starved_entry);
414 		list_del_init(&sdev->starved_entry);
415 		if (scsi_target_is_busy(scsi_target(sdev))) {
416 			list_move_tail(&sdev->starved_entry,
417 				       &shost->starved_list);
418 			continue;
419 		}
420 
421 		/*
422 		 * Once we drop the host lock, a racing scsi_remove_device()
423 		 * call may remove the sdev from the starved list and destroy
424 		 * it and the queue.  Mitigate by taking a reference to the
425 		 * queue and never touching the sdev again after we drop the
426 		 * host lock.  Note: if __scsi_remove_device() invokes
427 		 * blk_cleanup_queue() before the queue is run from this
428 		 * function then blk_run_queue() will return immediately since
429 		 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
430 		 */
431 		slq = sdev->request_queue;
432 		if (!blk_get_queue(slq))
433 			continue;
434 		spin_unlock_irqrestore(shost->host_lock, flags);
435 
436 		scsi_kick_queue(slq);
437 		blk_put_queue(slq);
438 
439 		spin_lock_irqsave(shost->host_lock, flags);
440 	}
441 	/* put any unprocessed entries back */
442 	list_splice(&starved_list, &shost->starved_list);
443 	spin_unlock_irqrestore(shost->host_lock, flags);
444 }
445 
446 /**
447  * scsi_run_queue - Select a proper request queue to serve next.
448  * @q:  last request's queue
449  *
450  * The previous command was completely finished, start a new one if possible.
451  */
452 static void scsi_run_queue(struct request_queue *q)
453 {
454 	struct scsi_device *sdev = q->queuedata;
455 
456 	if (scsi_target(sdev)->single_lun)
457 		scsi_single_lun_run(sdev);
458 	if (!list_empty(&sdev->host->starved_list))
459 		scsi_starved_list_run(sdev->host);
460 
461 	blk_mq_run_hw_queues(q, false);
462 }
463 
464 void scsi_requeue_run_queue(struct work_struct *work)
465 {
466 	struct scsi_device *sdev;
467 	struct request_queue *q;
468 
469 	sdev = container_of(work, struct scsi_device, requeue_work);
470 	q = sdev->request_queue;
471 	scsi_run_queue(q);
472 }
473 
474 void scsi_run_host_queues(struct Scsi_Host *shost)
475 {
476 	struct scsi_device *sdev;
477 
478 	shost_for_each_device(sdev, shost)
479 		scsi_run_queue(sdev->request_queue);
480 }
481 
482 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
483 {
484 	if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
485 		struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
486 
487 		if (drv->uninit_command)
488 			drv->uninit_command(cmd);
489 	}
490 }
491 
492 void scsi_free_sgtables(struct scsi_cmnd *cmd)
493 {
494 	if (cmd->sdb.table.nents)
495 		sg_free_table_chained(&cmd->sdb.table,
496 				SCSI_INLINE_SG_CNT);
497 	if (scsi_prot_sg_count(cmd))
498 		sg_free_table_chained(&cmd->prot_sdb->table,
499 				SCSI_INLINE_PROT_SG_CNT);
500 }
501 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
502 
503 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
504 {
505 	scsi_free_sgtables(cmd);
506 	scsi_uninit_cmd(cmd);
507 }
508 
509 static void scsi_run_queue_async(struct scsi_device *sdev)
510 {
511 	if (scsi_target(sdev)->single_lun ||
512 	    !list_empty(&sdev->host->starved_list)) {
513 		kblockd_schedule_work(&sdev->requeue_work);
514 	} else {
515 		/*
516 		 * smp_mb() present in sbitmap_queue_clear() or implied in
517 		 * .end_io is for ordering writing .device_busy in
518 		 * scsi_device_unbusy() and reading sdev->restarts.
519 		 */
520 		int old = atomic_read(&sdev->restarts);
521 
522 		/*
523 		 * ->restarts has to be kept as non-zero if new budget
524 		 *  contention occurs.
525 		 *
526 		 *  No need to run queue when either another re-run
527 		 *  queue wins in updating ->restarts or a new budget
528 		 *  contention occurs.
529 		 */
530 		if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
531 			blk_mq_run_hw_queues(sdev->request_queue, true);
532 	}
533 }
534 
535 /* Returns false when no more bytes to process, true if there are more */
536 static bool scsi_end_request(struct request *req, blk_status_t error,
537 		unsigned int bytes)
538 {
539 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
540 	struct scsi_device *sdev = cmd->device;
541 	struct request_queue *q = sdev->request_queue;
542 
543 	if (blk_update_request(req, error, bytes))
544 		return true;
545 
546 	// XXX:
547 	if (blk_queue_add_random(q))
548 		add_disk_randomness(req->q->disk);
549 
550 	if (!blk_rq_is_passthrough(req)) {
551 		WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
552 		cmd->flags &= ~SCMD_INITIALIZED;
553 	}
554 
555 	/*
556 	 * Calling rcu_barrier() is not necessary here because the
557 	 * SCSI error handler guarantees that the function called by
558 	 * call_rcu() has been called before scsi_end_request() is
559 	 * called.
560 	 */
561 	destroy_rcu_head(&cmd->rcu);
562 
563 	/*
564 	 * In the MQ case the command gets freed by __blk_mq_end_request,
565 	 * so we have to do all cleanup that depends on it earlier.
566 	 *
567 	 * We also can't kick the queues from irq context, so we
568 	 * will have to defer it to a workqueue.
569 	 */
570 	scsi_mq_uninit_cmd(cmd);
571 
572 	/*
573 	 * queue is still alive, so grab the ref for preventing it
574 	 * from being cleaned up during running queue.
575 	 */
576 	percpu_ref_get(&q->q_usage_counter);
577 
578 	__blk_mq_end_request(req, error);
579 
580 	scsi_run_queue_async(sdev);
581 
582 	percpu_ref_put(&q->q_usage_counter);
583 	return false;
584 }
585 
586 /**
587  * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
588  * @cmd:	SCSI command
589  * @result:	scsi error code
590  *
591  * Translate a SCSI result code into a blk_status_t value. May reset the host
592  * byte of @cmd->result.
593  */
594 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
595 {
596 	switch (host_byte(result)) {
597 	case DID_OK:
598 		if (scsi_status_is_good(result))
599 			return BLK_STS_OK;
600 		return BLK_STS_IOERR;
601 	case DID_TRANSPORT_FAILFAST:
602 	case DID_TRANSPORT_MARGINAL:
603 		return BLK_STS_TRANSPORT;
604 	case DID_TARGET_FAILURE:
605 		set_host_byte(cmd, DID_OK);
606 		return BLK_STS_TARGET;
607 	case DID_NEXUS_FAILURE:
608 		set_host_byte(cmd, DID_OK);
609 		return BLK_STS_NEXUS;
610 	case DID_ALLOC_FAILURE:
611 		set_host_byte(cmd, DID_OK);
612 		return BLK_STS_NOSPC;
613 	case DID_MEDIUM_ERROR:
614 		set_host_byte(cmd, DID_OK);
615 		return BLK_STS_MEDIUM;
616 	default:
617 		return BLK_STS_IOERR;
618 	}
619 }
620 
621 /**
622  * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
623  * @rq: request to examine
624  *
625  * Description:
626  *     A request could be merge of IOs which require different failure
627  *     handling.  This function determines the number of bytes which
628  *     can be failed from the beginning of the request without
629  *     crossing into area which need to be retried further.
630  *
631  * Return:
632  *     The number of bytes to fail.
633  */
634 static unsigned int scsi_rq_err_bytes(const struct request *rq)
635 {
636 	unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
637 	unsigned int bytes = 0;
638 	struct bio *bio;
639 
640 	if (!(rq->rq_flags & RQF_MIXED_MERGE))
641 		return blk_rq_bytes(rq);
642 
643 	/*
644 	 * Currently the only 'mixing' which can happen is between
645 	 * different fastfail types.  We can safely fail portions
646 	 * which have all the failfast bits that the first one has -
647 	 * the ones which are at least as eager to fail as the first
648 	 * one.
649 	 */
650 	for (bio = rq->bio; bio; bio = bio->bi_next) {
651 		if ((bio->bi_opf & ff) != ff)
652 			break;
653 		bytes += bio->bi_iter.bi_size;
654 	}
655 
656 	/* this could lead to infinite loop */
657 	BUG_ON(blk_rq_bytes(rq) && !bytes);
658 	return bytes;
659 }
660 
661 /* Helper for scsi_io_completion() when "reprep" action required. */
662 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
663 				      struct request_queue *q)
664 {
665 	/* A new command will be prepared and issued. */
666 	scsi_mq_requeue_cmd(cmd);
667 }
668 
669 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
670 {
671 	struct request *req = scsi_cmd_to_rq(cmd);
672 	unsigned long wait_for;
673 
674 	if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
675 		return false;
676 
677 	wait_for = (cmd->allowed + 1) * req->timeout;
678 	if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
679 		scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
680 			    wait_for/HZ);
681 		return true;
682 	}
683 	return false;
684 }
685 
686 /* Helper for scsi_io_completion() when special action required. */
687 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
688 {
689 	struct request_queue *q = cmd->device->request_queue;
690 	struct request *req = scsi_cmd_to_rq(cmd);
691 	int level = 0;
692 	enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
693 	      ACTION_DELAYED_RETRY} action;
694 	struct scsi_sense_hdr sshdr;
695 	bool sense_valid;
696 	bool sense_current = true;      /* false implies "deferred sense" */
697 	blk_status_t blk_stat;
698 
699 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
700 	if (sense_valid)
701 		sense_current = !scsi_sense_is_deferred(&sshdr);
702 
703 	blk_stat = scsi_result_to_blk_status(cmd, result);
704 
705 	if (host_byte(result) == DID_RESET) {
706 		/* Third party bus reset or reset for error recovery
707 		 * reasons.  Just retry the command and see what
708 		 * happens.
709 		 */
710 		action = ACTION_RETRY;
711 	} else if (sense_valid && sense_current) {
712 		switch (sshdr.sense_key) {
713 		case UNIT_ATTENTION:
714 			if (cmd->device->removable) {
715 				/* Detected disc change.  Set a bit
716 				 * and quietly refuse further access.
717 				 */
718 				cmd->device->changed = 1;
719 				action = ACTION_FAIL;
720 			} else {
721 				/* Must have been a power glitch, or a
722 				 * bus reset.  Could not have been a
723 				 * media change, so we just retry the
724 				 * command and see what happens.
725 				 */
726 				action = ACTION_RETRY;
727 			}
728 			break;
729 		case ILLEGAL_REQUEST:
730 			/* If we had an ILLEGAL REQUEST returned, then
731 			 * we may have performed an unsupported
732 			 * command.  The only thing this should be
733 			 * would be a ten byte read where only a six
734 			 * byte read was supported.  Also, on a system
735 			 * where READ CAPACITY failed, we may have
736 			 * read past the end of the disk.
737 			 */
738 			if ((cmd->device->use_10_for_rw &&
739 			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
740 			    (cmd->cmnd[0] == READ_10 ||
741 			     cmd->cmnd[0] == WRITE_10)) {
742 				/* This will issue a new 6-byte command. */
743 				cmd->device->use_10_for_rw = 0;
744 				action = ACTION_REPREP;
745 			} else if (sshdr.asc == 0x10) /* DIX */ {
746 				action = ACTION_FAIL;
747 				blk_stat = BLK_STS_PROTECTION;
748 			/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
749 			} else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
750 				action = ACTION_FAIL;
751 				blk_stat = BLK_STS_TARGET;
752 			} else
753 				action = ACTION_FAIL;
754 			break;
755 		case ABORTED_COMMAND:
756 			action = ACTION_FAIL;
757 			if (sshdr.asc == 0x10) /* DIF */
758 				blk_stat = BLK_STS_PROTECTION;
759 			break;
760 		case NOT_READY:
761 			/* If the device is in the process of becoming
762 			 * ready, or has a temporary blockage, retry.
763 			 */
764 			if (sshdr.asc == 0x04) {
765 				switch (sshdr.ascq) {
766 				case 0x01: /* becoming ready */
767 				case 0x04: /* format in progress */
768 				case 0x05: /* rebuild in progress */
769 				case 0x06: /* recalculation in progress */
770 				case 0x07: /* operation in progress */
771 				case 0x08: /* Long write in progress */
772 				case 0x09: /* self test in progress */
773 				case 0x11: /* notify (enable spinup) required */
774 				case 0x14: /* space allocation in progress */
775 				case 0x1a: /* start stop unit in progress */
776 				case 0x1b: /* sanitize in progress */
777 				case 0x1d: /* configuration in progress */
778 				case 0x24: /* depopulation in progress */
779 					action = ACTION_DELAYED_RETRY;
780 					break;
781 				case 0x0a: /* ALUA state transition */
782 					blk_stat = BLK_STS_AGAIN;
783 					fallthrough;
784 				default:
785 					action = ACTION_FAIL;
786 					break;
787 				}
788 			} else
789 				action = ACTION_FAIL;
790 			break;
791 		case VOLUME_OVERFLOW:
792 			/* See SSC3rXX or current. */
793 			action = ACTION_FAIL;
794 			break;
795 		case DATA_PROTECT:
796 			action = ACTION_FAIL;
797 			if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
798 			    (sshdr.asc == 0x55 &&
799 			     (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
800 				/* Insufficient zone resources */
801 				blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
802 			}
803 			break;
804 		default:
805 			action = ACTION_FAIL;
806 			break;
807 		}
808 	} else
809 		action = ACTION_FAIL;
810 
811 	if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
812 		action = ACTION_FAIL;
813 
814 	switch (action) {
815 	case ACTION_FAIL:
816 		/* Give up and fail the remainder of the request */
817 		if (!(req->rq_flags & RQF_QUIET)) {
818 			static DEFINE_RATELIMIT_STATE(_rs,
819 					DEFAULT_RATELIMIT_INTERVAL,
820 					DEFAULT_RATELIMIT_BURST);
821 
822 			if (unlikely(scsi_logging_level))
823 				level =
824 				     SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
825 						    SCSI_LOG_MLCOMPLETE_BITS);
826 
827 			/*
828 			 * if logging is enabled the failure will be printed
829 			 * in scsi_log_completion(), so avoid duplicate messages
830 			 */
831 			if (!level && __ratelimit(&_rs)) {
832 				scsi_print_result(cmd, NULL, FAILED);
833 				if (sense_valid)
834 					scsi_print_sense(cmd);
835 				scsi_print_command(cmd);
836 			}
837 		}
838 		if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
839 			return;
840 		fallthrough;
841 	case ACTION_REPREP:
842 		scsi_io_completion_reprep(cmd, q);
843 		break;
844 	case ACTION_RETRY:
845 		/* Retry the same command immediately */
846 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
847 		break;
848 	case ACTION_DELAYED_RETRY:
849 		/* Retry the same command after a delay */
850 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
851 		break;
852 	}
853 }
854 
855 /*
856  * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
857  * new result that may suppress further error checking. Also modifies
858  * *blk_statp in some cases.
859  */
860 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
861 					blk_status_t *blk_statp)
862 {
863 	bool sense_valid;
864 	bool sense_current = true;	/* false implies "deferred sense" */
865 	struct request *req = scsi_cmd_to_rq(cmd);
866 	struct scsi_sense_hdr sshdr;
867 
868 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
869 	if (sense_valid)
870 		sense_current = !scsi_sense_is_deferred(&sshdr);
871 
872 	if (blk_rq_is_passthrough(req)) {
873 		if (sense_valid) {
874 			/*
875 			 * SG_IO wants current and deferred errors
876 			 */
877 			scsi_req(req)->sense_len =
878 				min(8 + cmd->sense_buffer[7],
879 				    SCSI_SENSE_BUFFERSIZE);
880 		}
881 		if (sense_current)
882 			*blk_statp = scsi_result_to_blk_status(cmd, result);
883 	} else if (blk_rq_bytes(req) == 0 && sense_current) {
884 		/*
885 		 * Flush commands do not transfers any data, and thus cannot use
886 		 * good_bytes != blk_rq_bytes(req) as the signal for an error.
887 		 * This sets *blk_statp explicitly for the problem case.
888 		 */
889 		*blk_statp = scsi_result_to_blk_status(cmd, result);
890 	}
891 	/*
892 	 * Recovered errors need reporting, but they're always treated as
893 	 * success, so fiddle the result code here.  For passthrough requests
894 	 * we already took a copy of the original into sreq->result which
895 	 * is what gets returned to the user
896 	 */
897 	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
898 		bool do_print = true;
899 		/*
900 		 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
901 		 * skip print since caller wants ATA registers. Only occurs
902 		 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
903 		 */
904 		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
905 			do_print = false;
906 		else if (req->rq_flags & RQF_QUIET)
907 			do_print = false;
908 		if (do_print)
909 			scsi_print_sense(cmd);
910 		result = 0;
911 		/* for passthrough, *blk_statp may be set */
912 		*blk_statp = BLK_STS_OK;
913 	}
914 	/*
915 	 * Another corner case: the SCSI status byte is non-zero but 'good'.
916 	 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
917 	 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
918 	 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
919 	 * intermediate statuses (both obsolete in SAM-4) as good.
920 	 */
921 	if ((result & 0xff) && scsi_status_is_good(result)) {
922 		result = 0;
923 		*blk_statp = BLK_STS_OK;
924 	}
925 	return result;
926 }
927 
928 /**
929  * scsi_io_completion - Completion processing for SCSI commands.
930  * @cmd:	command that is finished.
931  * @good_bytes:	number of processed bytes.
932  *
933  * We will finish off the specified number of sectors. If we are done, the
934  * command block will be released and the queue function will be goosed. If we
935  * are not done then we have to figure out what to do next:
936  *
937  *   a) We can call scsi_io_completion_reprep().  The request will be
938  *	unprepared and put back on the queue.  Then a new command will
939  *	be created for it.  This should be used if we made forward
940  *	progress, or if we want to switch from READ(10) to READ(6) for
941  *	example.
942  *
943  *   b) We can call scsi_io_completion_action().  The request will be
944  *	put back on the queue and retried using the same command as
945  *	before, possibly after a delay.
946  *
947  *   c) We can call scsi_end_request() with blk_stat other than
948  *	BLK_STS_OK, to fail the remainder of the request.
949  */
950 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
951 {
952 	int result = cmd->result;
953 	struct request_queue *q = cmd->device->request_queue;
954 	struct request *req = scsi_cmd_to_rq(cmd);
955 	blk_status_t blk_stat = BLK_STS_OK;
956 
957 	if (unlikely(result))	/* a nz result may or may not be an error */
958 		result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
959 
960 	if (unlikely(blk_rq_is_passthrough(req))) {
961 		/*
962 		 * scsi_result_to_blk_status may have reset the host_byte
963 		 */
964 		scsi_req(req)->result = cmd->result;
965 	}
966 
967 	/*
968 	 * Next deal with any sectors which we were able to correctly
969 	 * handle.
970 	 */
971 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
972 		"%u sectors total, %d bytes done.\n",
973 		blk_rq_sectors(req), good_bytes));
974 
975 	/*
976 	 * Failed, zero length commands always need to drop down
977 	 * to retry code. Fast path should return in this block.
978 	 */
979 	if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
980 		if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
981 			return; /* no bytes remaining */
982 	}
983 
984 	/* Kill remainder if no retries. */
985 	if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
986 		if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
987 			WARN_ONCE(true,
988 			    "Bytes remaining after failed, no-retry command");
989 		return;
990 	}
991 
992 	/*
993 	 * If there had been no error, but we have leftover bytes in the
994 	 * request just queue the command up again.
995 	 */
996 	if (likely(result == 0))
997 		scsi_io_completion_reprep(cmd, q);
998 	else
999 		scsi_io_completion_action(cmd, result);
1000 }
1001 
1002 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1003 		struct request *rq)
1004 {
1005 	return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1006 	       !op_is_write(req_op(rq)) &&
1007 	       sdev->host->hostt->dma_need_drain(rq);
1008 }
1009 
1010 /**
1011  * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1012  * @cmd: SCSI command data structure to initialize.
1013  *
1014  * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1015  * for @cmd.
1016  *
1017  * Returns:
1018  * * BLK_STS_OK       - on success
1019  * * BLK_STS_RESOURCE - if the failure is retryable
1020  * * BLK_STS_IOERR    - if the failure is fatal
1021  */
1022 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1023 {
1024 	struct scsi_device *sdev = cmd->device;
1025 	struct request *rq = scsi_cmd_to_rq(cmd);
1026 	unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1027 	struct scatterlist *last_sg = NULL;
1028 	blk_status_t ret;
1029 	bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1030 	int count;
1031 
1032 	if (WARN_ON_ONCE(!nr_segs))
1033 		return BLK_STS_IOERR;
1034 
1035 	/*
1036 	 * Make sure there is space for the drain.  The driver must adjust
1037 	 * max_hw_segments to be prepared for this.
1038 	 */
1039 	if (need_drain)
1040 		nr_segs++;
1041 
1042 	/*
1043 	 * If sg table allocation fails, requeue request later.
1044 	 */
1045 	if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1046 			cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1047 		return BLK_STS_RESOURCE;
1048 
1049 	/*
1050 	 * Next, walk the list, and fill in the addresses and sizes of
1051 	 * each segment.
1052 	 */
1053 	count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1054 
1055 	if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1056 		unsigned int pad_len =
1057 			(rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1058 
1059 		last_sg->length += pad_len;
1060 		cmd->extra_len += pad_len;
1061 	}
1062 
1063 	if (need_drain) {
1064 		sg_unmark_end(last_sg);
1065 		last_sg = sg_next(last_sg);
1066 		sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1067 		sg_mark_end(last_sg);
1068 
1069 		cmd->extra_len += sdev->dma_drain_len;
1070 		count++;
1071 	}
1072 
1073 	BUG_ON(count > cmd->sdb.table.nents);
1074 	cmd->sdb.table.nents = count;
1075 	cmd->sdb.length = blk_rq_payload_bytes(rq);
1076 
1077 	if (blk_integrity_rq(rq)) {
1078 		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1079 		int ivecs;
1080 
1081 		if (WARN_ON_ONCE(!prot_sdb)) {
1082 			/*
1083 			 * This can happen if someone (e.g. multipath)
1084 			 * queues a command to a device on an adapter
1085 			 * that does not support DIX.
1086 			 */
1087 			ret = BLK_STS_IOERR;
1088 			goto out_free_sgtables;
1089 		}
1090 
1091 		ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1092 
1093 		if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1094 				prot_sdb->table.sgl,
1095 				SCSI_INLINE_PROT_SG_CNT)) {
1096 			ret = BLK_STS_RESOURCE;
1097 			goto out_free_sgtables;
1098 		}
1099 
1100 		count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1101 						prot_sdb->table.sgl);
1102 		BUG_ON(count > ivecs);
1103 		BUG_ON(count > queue_max_integrity_segments(rq->q));
1104 
1105 		cmd->prot_sdb = prot_sdb;
1106 		cmd->prot_sdb->table.nents = count;
1107 	}
1108 
1109 	return BLK_STS_OK;
1110 out_free_sgtables:
1111 	scsi_free_sgtables(cmd);
1112 	return ret;
1113 }
1114 EXPORT_SYMBOL(scsi_alloc_sgtables);
1115 
1116 /**
1117  * scsi_initialize_rq - initialize struct scsi_cmnd partially
1118  * @rq: Request associated with the SCSI command to be initialized.
1119  *
1120  * This function initializes the members of struct scsi_cmnd that must be
1121  * initialized before request processing starts and that won't be
1122  * reinitialized if a SCSI command is requeued.
1123  */
1124 static void scsi_initialize_rq(struct request *rq)
1125 {
1126 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1127 	struct scsi_request *req = &cmd->req;
1128 
1129 	memset(req->__cmd, 0, sizeof(req->__cmd));
1130 	req->cmd = req->__cmd;
1131 	req->cmd_len = BLK_MAX_CDB;
1132 	req->sense_len = 0;
1133 
1134 	init_rcu_head(&cmd->rcu);
1135 	cmd->jiffies_at_alloc = jiffies;
1136 	cmd->retries = 0;
1137 }
1138 
1139 struct request *scsi_alloc_request(struct request_queue *q,
1140 		unsigned int op, blk_mq_req_flags_t flags)
1141 {
1142 	struct request *rq;
1143 
1144 	rq = blk_mq_alloc_request(q, op, flags);
1145 	if (!IS_ERR(rq))
1146 		scsi_initialize_rq(rq);
1147 	return rq;
1148 }
1149 EXPORT_SYMBOL_GPL(scsi_alloc_request);
1150 
1151 /*
1152  * Only called when the request isn't completed by SCSI, and not freed by
1153  * SCSI
1154  */
1155 static void scsi_cleanup_rq(struct request *rq)
1156 {
1157 	if (rq->rq_flags & RQF_DONTPREP) {
1158 		scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1159 		rq->rq_flags &= ~RQF_DONTPREP;
1160 	}
1161 }
1162 
1163 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1164 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1165 {
1166 	void *buf = cmd->sense_buffer;
1167 	void *prot = cmd->prot_sdb;
1168 	struct request *rq = scsi_cmd_to_rq(cmd);
1169 	unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1170 	unsigned long jiffies_at_alloc;
1171 	int retries, to_clear;
1172 	bool in_flight;
1173 	int budget_token = cmd->budget_token;
1174 
1175 	if (!blk_rq_is_passthrough(rq) && !(flags & SCMD_INITIALIZED)) {
1176 		flags |= SCMD_INITIALIZED;
1177 		scsi_initialize_rq(rq);
1178 	}
1179 
1180 	jiffies_at_alloc = cmd->jiffies_at_alloc;
1181 	retries = cmd->retries;
1182 	in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1183 	/*
1184 	 * Zero out the cmd, except for the embedded scsi_request. Only clear
1185 	 * the driver-private command data if the LLD does not supply a
1186 	 * function to initialize that data.
1187 	 */
1188 	to_clear = sizeof(*cmd) - sizeof(cmd->req);
1189 	if (!dev->host->hostt->init_cmd_priv)
1190 		to_clear += dev->host->hostt->cmd_size;
1191 	memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
1192 
1193 	cmd->device = dev;
1194 	cmd->sense_buffer = buf;
1195 	cmd->prot_sdb = prot;
1196 	cmd->flags = flags;
1197 	INIT_LIST_HEAD(&cmd->eh_entry);
1198 	INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1199 	cmd->jiffies_at_alloc = jiffies_at_alloc;
1200 	cmd->retries = retries;
1201 	if (in_flight)
1202 		__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1203 	cmd->budget_token = budget_token;
1204 
1205 }
1206 
1207 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1208 		struct request *req)
1209 {
1210 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1211 
1212 	/*
1213 	 * Passthrough requests may transfer data, in which case they must
1214 	 * a bio attached to them.  Or they might contain a SCSI command
1215 	 * that does not transfer data, in which case they may optionally
1216 	 * submit a request without an attached bio.
1217 	 */
1218 	if (req->bio) {
1219 		blk_status_t ret = scsi_alloc_sgtables(cmd);
1220 		if (unlikely(ret != BLK_STS_OK))
1221 			return ret;
1222 	} else {
1223 		BUG_ON(blk_rq_bytes(req));
1224 
1225 		memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1226 	}
1227 
1228 	cmd->cmd_len = scsi_req(req)->cmd_len;
1229 	cmd->cmnd = scsi_req(req)->cmd;
1230 	cmd->transfersize = blk_rq_bytes(req);
1231 	cmd->allowed = scsi_req(req)->retries;
1232 	return BLK_STS_OK;
1233 }
1234 
1235 static blk_status_t
1236 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1237 {
1238 	switch (sdev->sdev_state) {
1239 	case SDEV_CREATED:
1240 		return BLK_STS_OK;
1241 	case SDEV_OFFLINE:
1242 	case SDEV_TRANSPORT_OFFLINE:
1243 		/*
1244 		 * If the device is offline we refuse to process any
1245 		 * commands.  The device must be brought online
1246 		 * before trying any recovery commands.
1247 		 */
1248 		if (!sdev->offline_already) {
1249 			sdev->offline_already = true;
1250 			sdev_printk(KERN_ERR, sdev,
1251 				    "rejecting I/O to offline device\n");
1252 		}
1253 		return BLK_STS_IOERR;
1254 	case SDEV_DEL:
1255 		/*
1256 		 * If the device is fully deleted, we refuse to
1257 		 * process any commands as well.
1258 		 */
1259 		sdev_printk(KERN_ERR, sdev,
1260 			    "rejecting I/O to dead device\n");
1261 		return BLK_STS_IOERR;
1262 	case SDEV_BLOCK:
1263 	case SDEV_CREATED_BLOCK:
1264 		return BLK_STS_RESOURCE;
1265 	case SDEV_QUIESCE:
1266 		/*
1267 		 * If the device is blocked we only accept power management
1268 		 * commands.
1269 		 */
1270 		if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1271 			return BLK_STS_RESOURCE;
1272 		return BLK_STS_OK;
1273 	default:
1274 		/*
1275 		 * For any other not fully online state we only allow
1276 		 * power management commands.
1277 		 */
1278 		if (req && !(req->rq_flags & RQF_PM))
1279 			return BLK_STS_IOERR;
1280 		return BLK_STS_OK;
1281 	}
1282 }
1283 
1284 /*
1285  * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1286  * and return the token else return -1.
1287  */
1288 static inline int scsi_dev_queue_ready(struct request_queue *q,
1289 				  struct scsi_device *sdev)
1290 {
1291 	int token;
1292 
1293 	token = sbitmap_get(&sdev->budget_map);
1294 	if (atomic_read(&sdev->device_blocked)) {
1295 		if (token < 0)
1296 			goto out;
1297 
1298 		if (scsi_device_busy(sdev) > 1)
1299 			goto out_dec;
1300 
1301 		/*
1302 		 * unblock after device_blocked iterates to zero
1303 		 */
1304 		if (atomic_dec_return(&sdev->device_blocked) > 0)
1305 			goto out_dec;
1306 		SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1307 				   "unblocking device at zero depth\n"));
1308 	}
1309 
1310 	return token;
1311 out_dec:
1312 	if (token >= 0)
1313 		sbitmap_put(&sdev->budget_map, token);
1314 out:
1315 	return -1;
1316 }
1317 
1318 /*
1319  * scsi_target_queue_ready: checks if there we can send commands to target
1320  * @sdev: scsi device on starget to check.
1321  */
1322 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1323 					   struct scsi_device *sdev)
1324 {
1325 	struct scsi_target *starget = scsi_target(sdev);
1326 	unsigned int busy;
1327 
1328 	if (starget->single_lun) {
1329 		spin_lock_irq(shost->host_lock);
1330 		if (starget->starget_sdev_user &&
1331 		    starget->starget_sdev_user != sdev) {
1332 			spin_unlock_irq(shost->host_lock);
1333 			return 0;
1334 		}
1335 		starget->starget_sdev_user = sdev;
1336 		spin_unlock_irq(shost->host_lock);
1337 	}
1338 
1339 	if (starget->can_queue <= 0)
1340 		return 1;
1341 
1342 	busy = atomic_inc_return(&starget->target_busy) - 1;
1343 	if (atomic_read(&starget->target_blocked) > 0) {
1344 		if (busy)
1345 			goto starved;
1346 
1347 		/*
1348 		 * unblock after target_blocked iterates to zero
1349 		 */
1350 		if (atomic_dec_return(&starget->target_blocked) > 0)
1351 			goto out_dec;
1352 
1353 		SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1354 				 "unblocking target at zero depth\n"));
1355 	}
1356 
1357 	if (busy >= starget->can_queue)
1358 		goto starved;
1359 
1360 	return 1;
1361 
1362 starved:
1363 	spin_lock_irq(shost->host_lock);
1364 	list_move_tail(&sdev->starved_entry, &shost->starved_list);
1365 	spin_unlock_irq(shost->host_lock);
1366 out_dec:
1367 	if (starget->can_queue > 0)
1368 		atomic_dec(&starget->target_busy);
1369 	return 0;
1370 }
1371 
1372 /*
1373  * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1374  * return 0. We must end up running the queue again whenever 0 is
1375  * returned, else IO can hang.
1376  */
1377 static inline int scsi_host_queue_ready(struct request_queue *q,
1378 				   struct Scsi_Host *shost,
1379 				   struct scsi_device *sdev,
1380 				   struct scsi_cmnd *cmd)
1381 {
1382 	if (scsi_host_in_recovery(shost))
1383 		return 0;
1384 
1385 	if (atomic_read(&shost->host_blocked) > 0) {
1386 		if (scsi_host_busy(shost) > 0)
1387 			goto starved;
1388 
1389 		/*
1390 		 * unblock after host_blocked iterates to zero
1391 		 */
1392 		if (atomic_dec_return(&shost->host_blocked) > 0)
1393 			goto out_dec;
1394 
1395 		SCSI_LOG_MLQUEUE(3,
1396 			shost_printk(KERN_INFO, shost,
1397 				     "unblocking host at zero depth\n"));
1398 	}
1399 
1400 	if (shost->host_self_blocked)
1401 		goto starved;
1402 
1403 	/* We're OK to process the command, so we can't be starved */
1404 	if (!list_empty(&sdev->starved_entry)) {
1405 		spin_lock_irq(shost->host_lock);
1406 		if (!list_empty(&sdev->starved_entry))
1407 			list_del_init(&sdev->starved_entry);
1408 		spin_unlock_irq(shost->host_lock);
1409 	}
1410 
1411 	__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1412 
1413 	return 1;
1414 
1415 starved:
1416 	spin_lock_irq(shost->host_lock);
1417 	if (list_empty(&sdev->starved_entry))
1418 		list_add_tail(&sdev->starved_entry, &shost->starved_list);
1419 	spin_unlock_irq(shost->host_lock);
1420 out_dec:
1421 	scsi_dec_host_busy(shost, cmd);
1422 	return 0;
1423 }
1424 
1425 /*
1426  * Busy state exporting function for request stacking drivers.
1427  *
1428  * For efficiency, no lock is taken to check the busy state of
1429  * shost/starget/sdev, since the returned value is not guaranteed and
1430  * may be changed after request stacking drivers call the function,
1431  * regardless of taking lock or not.
1432  *
1433  * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1434  * needs to return 'not busy'. Otherwise, request stacking drivers
1435  * may hold requests forever.
1436  */
1437 static bool scsi_mq_lld_busy(struct request_queue *q)
1438 {
1439 	struct scsi_device *sdev = q->queuedata;
1440 	struct Scsi_Host *shost;
1441 
1442 	if (blk_queue_dying(q))
1443 		return false;
1444 
1445 	shost = sdev->host;
1446 
1447 	/*
1448 	 * Ignore host/starget busy state.
1449 	 * Since block layer does not have a concept of fairness across
1450 	 * multiple queues, congestion of host/starget needs to be handled
1451 	 * in SCSI layer.
1452 	 */
1453 	if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1454 		return true;
1455 
1456 	return false;
1457 }
1458 
1459 /*
1460  * Block layer request completion callback. May be called from interrupt
1461  * context.
1462  */
1463 static void scsi_complete(struct request *rq)
1464 {
1465 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1466 	enum scsi_disposition disposition;
1467 
1468 	INIT_LIST_HEAD(&cmd->eh_entry);
1469 
1470 	atomic_inc(&cmd->device->iodone_cnt);
1471 	if (cmd->result)
1472 		atomic_inc(&cmd->device->ioerr_cnt);
1473 
1474 	disposition = scsi_decide_disposition(cmd);
1475 	if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1476 		disposition = SUCCESS;
1477 
1478 	scsi_log_completion(cmd, disposition);
1479 
1480 	switch (disposition) {
1481 	case SUCCESS:
1482 		scsi_finish_command(cmd);
1483 		break;
1484 	case NEEDS_RETRY:
1485 		scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1486 		break;
1487 	case ADD_TO_MLQUEUE:
1488 		scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1489 		break;
1490 	default:
1491 		scsi_eh_scmd_add(cmd);
1492 		break;
1493 	}
1494 }
1495 
1496 /**
1497  * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1498  * @cmd: command block we are dispatching.
1499  *
1500  * Return: nonzero return request was rejected and device's queue needs to be
1501  * plugged.
1502  */
1503 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1504 {
1505 	struct Scsi_Host *host = cmd->device->host;
1506 	int rtn = 0;
1507 
1508 	atomic_inc(&cmd->device->iorequest_cnt);
1509 
1510 	/* check if the device is still usable */
1511 	if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1512 		/* in SDEV_DEL we error all commands. DID_NO_CONNECT
1513 		 * returns an immediate error upwards, and signals
1514 		 * that the device is no longer present */
1515 		cmd->result = DID_NO_CONNECT << 16;
1516 		goto done;
1517 	}
1518 
1519 	/* Check to see if the scsi lld made this device blocked. */
1520 	if (unlikely(scsi_device_blocked(cmd->device))) {
1521 		/*
1522 		 * in blocked state, the command is just put back on
1523 		 * the device queue.  The suspend state has already
1524 		 * blocked the queue so future requests should not
1525 		 * occur until the device transitions out of the
1526 		 * suspend state.
1527 		 */
1528 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1529 			"queuecommand : device blocked\n"));
1530 		return SCSI_MLQUEUE_DEVICE_BUSY;
1531 	}
1532 
1533 	/* Store the LUN value in cmnd, if needed. */
1534 	if (cmd->device->lun_in_cdb)
1535 		cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1536 			       (cmd->device->lun << 5 & 0xe0);
1537 
1538 	scsi_log_send(cmd);
1539 
1540 	/*
1541 	 * Before we queue this command, check if the command
1542 	 * length exceeds what the host adapter can handle.
1543 	 */
1544 	if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1545 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1546 			       "queuecommand : command too long. "
1547 			       "cdb_size=%d host->max_cmd_len=%d\n",
1548 			       cmd->cmd_len, cmd->device->host->max_cmd_len));
1549 		cmd->result = (DID_ABORT << 16);
1550 		goto done;
1551 	}
1552 
1553 	if (unlikely(host->shost_state == SHOST_DEL)) {
1554 		cmd->result = (DID_NO_CONNECT << 16);
1555 		goto done;
1556 
1557 	}
1558 
1559 	trace_scsi_dispatch_cmd_start(cmd);
1560 	rtn = host->hostt->queuecommand(host, cmd);
1561 	if (rtn) {
1562 		trace_scsi_dispatch_cmd_error(cmd, rtn);
1563 		if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1564 		    rtn != SCSI_MLQUEUE_TARGET_BUSY)
1565 			rtn = SCSI_MLQUEUE_HOST_BUSY;
1566 
1567 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1568 			"queuecommand : request rejected\n"));
1569 	}
1570 
1571 	return rtn;
1572  done:
1573 	scsi_done(cmd);
1574 	return 0;
1575 }
1576 
1577 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1578 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1579 {
1580 	return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1581 		sizeof(struct scatterlist);
1582 }
1583 
1584 static blk_status_t scsi_prepare_cmd(struct request *req)
1585 {
1586 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1587 	struct scsi_device *sdev = req->q->queuedata;
1588 	struct Scsi_Host *shost = sdev->host;
1589 	struct scatterlist *sg;
1590 
1591 	scsi_init_command(sdev, cmd);
1592 
1593 	cmd->prot_op = SCSI_PROT_NORMAL;
1594 	if (blk_rq_bytes(req))
1595 		cmd->sc_data_direction = rq_dma_dir(req);
1596 	else
1597 		cmd->sc_data_direction = DMA_NONE;
1598 
1599 	sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1600 	cmd->sdb.table.sgl = sg;
1601 
1602 	if (scsi_host_get_prot(shost)) {
1603 		memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1604 
1605 		cmd->prot_sdb->table.sgl =
1606 			(struct scatterlist *)(cmd->prot_sdb + 1);
1607 	}
1608 
1609 	/*
1610 	 * Special handling for passthrough commands, which don't go to the ULP
1611 	 * at all:
1612 	 */
1613 	if (blk_rq_is_passthrough(req))
1614 		return scsi_setup_scsi_cmnd(sdev, req);
1615 
1616 	if (sdev->handler && sdev->handler->prep_fn) {
1617 		blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1618 
1619 		if (ret != BLK_STS_OK)
1620 			return ret;
1621 	}
1622 
1623 	cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1624 	memset(cmd->cmnd, 0, BLK_MAX_CDB);
1625 	return scsi_cmd_to_driver(cmd)->init_command(cmd);
1626 }
1627 
1628 void scsi_done(struct scsi_cmnd *cmd)
1629 {
1630 	switch (cmd->submitter) {
1631 	case SUBMITTED_BY_BLOCK_LAYER:
1632 		break;
1633 	case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1634 		return scsi_eh_done(cmd);
1635 	case SUBMITTED_BY_SCSI_RESET_IOCTL:
1636 		return;
1637 	}
1638 
1639 	if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1640 		return;
1641 	if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1642 		return;
1643 	trace_scsi_dispatch_cmd_done(cmd);
1644 	blk_mq_complete_request(scsi_cmd_to_rq(cmd));
1645 }
1646 EXPORT_SYMBOL(scsi_done);
1647 
1648 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1649 {
1650 	struct scsi_device *sdev = q->queuedata;
1651 
1652 	sbitmap_put(&sdev->budget_map, budget_token);
1653 }
1654 
1655 static int scsi_mq_get_budget(struct request_queue *q)
1656 {
1657 	struct scsi_device *sdev = q->queuedata;
1658 	int token = scsi_dev_queue_ready(q, sdev);
1659 
1660 	if (token >= 0)
1661 		return token;
1662 
1663 	atomic_inc(&sdev->restarts);
1664 
1665 	/*
1666 	 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1667 	 * .restarts must be incremented before .device_busy is read because the
1668 	 * code in scsi_run_queue_async() depends on the order of these operations.
1669 	 */
1670 	smp_mb__after_atomic();
1671 
1672 	/*
1673 	 * If all in-flight requests originated from this LUN are completed
1674 	 * before reading .device_busy, sdev->device_busy will be observed as
1675 	 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1676 	 * soon. Otherwise, completion of one of these requests will observe
1677 	 * the .restarts flag, and the request queue will be run for handling
1678 	 * this request, see scsi_end_request().
1679 	 */
1680 	if (unlikely(scsi_device_busy(sdev) == 0 &&
1681 				!scsi_device_blocked(sdev)))
1682 		blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1683 	return -1;
1684 }
1685 
1686 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1687 {
1688 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1689 
1690 	cmd->budget_token = token;
1691 }
1692 
1693 static int scsi_mq_get_rq_budget_token(struct request *req)
1694 {
1695 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1696 
1697 	return cmd->budget_token;
1698 }
1699 
1700 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1701 			 const struct blk_mq_queue_data *bd)
1702 {
1703 	struct request *req = bd->rq;
1704 	struct request_queue *q = req->q;
1705 	struct scsi_device *sdev = q->queuedata;
1706 	struct Scsi_Host *shost = sdev->host;
1707 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1708 	blk_status_t ret;
1709 	int reason;
1710 
1711 	WARN_ON_ONCE(cmd->budget_token < 0);
1712 
1713 	/*
1714 	 * If the device is not in running state we will reject some or all
1715 	 * commands.
1716 	 */
1717 	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1718 		ret = scsi_device_state_check(sdev, req);
1719 		if (ret != BLK_STS_OK)
1720 			goto out_put_budget;
1721 	}
1722 
1723 	ret = BLK_STS_RESOURCE;
1724 	if (!scsi_target_queue_ready(shost, sdev))
1725 		goto out_put_budget;
1726 	if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1727 		goto out_dec_target_busy;
1728 
1729 	if (!(req->rq_flags & RQF_DONTPREP)) {
1730 		ret = scsi_prepare_cmd(req);
1731 		if (ret != BLK_STS_OK)
1732 			goto out_dec_host_busy;
1733 		req->rq_flags |= RQF_DONTPREP;
1734 	} else {
1735 		clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1736 	}
1737 
1738 	cmd->flags &= SCMD_PRESERVED_FLAGS;
1739 	if (sdev->simple_tags)
1740 		cmd->flags |= SCMD_TAGGED;
1741 	if (bd->last)
1742 		cmd->flags |= SCMD_LAST;
1743 
1744 	scsi_set_resid(cmd, 0);
1745 	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1746 	cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1747 
1748 	blk_mq_start_request(req);
1749 	reason = scsi_dispatch_cmd(cmd);
1750 	if (reason) {
1751 		scsi_set_blocked(cmd, reason);
1752 		ret = BLK_STS_RESOURCE;
1753 		goto out_dec_host_busy;
1754 	}
1755 
1756 	return BLK_STS_OK;
1757 
1758 out_dec_host_busy:
1759 	scsi_dec_host_busy(shost, cmd);
1760 out_dec_target_busy:
1761 	if (scsi_target(sdev)->can_queue > 0)
1762 		atomic_dec(&scsi_target(sdev)->target_busy);
1763 out_put_budget:
1764 	scsi_mq_put_budget(q, cmd->budget_token);
1765 	cmd->budget_token = -1;
1766 	switch (ret) {
1767 	case BLK_STS_OK:
1768 		break;
1769 	case BLK_STS_RESOURCE:
1770 	case BLK_STS_ZONE_RESOURCE:
1771 		if (scsi_device_blocked(sdev))
1772 			ret = BLK_STS_DEV_RESOURCE;
1773 		break;
1774 	case BLK_STS_AGAIN:
1775 		scsi_req(req)->result = DID_BUS_BUSY << 16;
1776 		if (req->rq_flags & RQF_DONTPREP)
1777 			scsi_mq_uninit_cmd(cmd);
1778 		break;
1779 	default:
1780 		if (unlikely(!scsi_device_online(sdev)))
1781 			scsi_req(req)->result = DID_NO_CONNECT << 16;
1782 		else
1783 			scsi_req(req)->result = DID_ERROR << 16;
1784 		/*
1785 		 * Make sure to release all allocated resources when
1786 		 * we hit an error, as we will never see this command
1787 		 * again.
1788 		 */
1789 		if (req->rq_flags & RQF_DONTPREP)
1790 			scsi_mq_uninit_cmd(cmd);
1791 		scsi_run_queue_async(sdev);
1792 		break;
1793 	}
1794 	return ret;
1795 }
1796 
1797 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1798 		bool reserved)
1799 {
1800 	if (reserved)
1801 		return BLK_EH_RESET_TIMER;
1802 	return scsi_times_out(req);
1803 }
1804 
1805 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1806 				unsigned int hctx_idx, unsigned int numa_node)
1807 {
1808 	struct Scsi_Host *shost = set->driver_data;
1809 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1810 	struct scatterlist *sg;
1811 	int ret = 0;
1812 
1813 	cmd->sense_buffer =
1814 		kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1815 	if (!cmd->sense_buffer)
1816 		return -ENOMEM;
1817 	cmd->req.sense = cmd->sense_buffer;
1818 
1819 	if (scsi_host_get_prot(shost)) {
1820 		sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1821 			shost->hostt->cmd_size;
1822 		cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1823 	}
1824 
1825 	if (shost->hostt->init_cmd_priv) {
1826 		ret = shost->hostt->init_cmd_priv(shost, cmd);
1827 		if (ret < 0)
1828 			kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1829 	}
1830 
1831 	return ret;
1832 }
1833 
1834 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1835 				 unsigned int hctx_idx)
1836 {
1837 	struct Scsi_Host *shost = set->driver_data;
1838 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1839 
1840 	if (shost->hostt->exit_cmd_priv)
1841 		shost->hostt->exit_cmd_priv(shost, cmd);
1842 	kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1843 }
1844 
1845 
1846 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1847 {
1848 	struct Scsi_Host *shost = hctx->driver_data;
1849 
1850 	if (shost->hostt->mq_poll)
1851 		return shost->hostt->mq_poll(shost, hctx->queue_num);
1852 
1853 	return 0;
1854 }
1855 
1856 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1857 			  unsigned int hctx_idx)
1858 {
1859 	struct Scsi_Host *shost = data;
1860 
1861 	hctx->driver_data = shost;
1862 	return 0;
1863 }
1864 
1865 static int scsi_map_queues(struct blk_mq_tag_set *set)
1866 {
1867 	struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1868 
1869 	if (shost->hostt->map_queues)
1870 		return shost->hostt->map_queues(shost);
1871 	return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1872 }
1873 
1874 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1875 {
1876 	struct device *dev = shost->dma_dev;
1877 
1878 	/*
1879 	 * this limit is imposed by hardware restrictions
1880 	 */
1881 	blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1882 					SG_MAX_SEGMENTS));
1883 
1884 	if (scsi_host_prot_dma(shost)) {
1885 		shost->sg_prot_tablesize =
1886 			min_not_zero(shost->sg_prot_tablesize,
1887 				     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1888 		BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1889 		blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1890 	}
1891 
1892 	if (dev->dma_mask) {
1893 		shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1894 				dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1895 	}
1896 	blk_queue_max_hw_sectors(q, shost->max_sectors);
1897 	blk_queue_segment_boundary(q, shost->dma_boundary);
1898 	dma_set_seg_boundary(dev, shost->dma_boundary);
1899 
1900 	blk_queue_max_segment_size(q, shost->max_segment_size);
1901 	blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1902 	dma_set_max_seg_size(dev, queue_max_segment_size(q));
1903 
1904 	/*
1905 	 * Set a reasonable default alignment:  The larger of 32-byte (dword),
1906 	 * which is a common minimum for HBAs, and the minimum DMA alignment,
1907 	 * which is set by the platform.
1908 	 *
1909 	 * Devices that require a bigger alignment can increase it later.
1910 	 */
1911 	blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1912 }
1913 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1914 
1915 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1916 	.get_budget	= scsi_mq_get_budget,
1917 	.put_budget	= scsi_mq_put_budget,
1918 	.queue_rq	= scsi_queue_rq,
1919 	.complete	= scsi_complete,
1920 	.timeout	= scsi_timeout,
1921 #ifdef CONFIG_BLK_DEBUG_FS
1922 	.show_rq	= scsi_show_rq,
1923 #endif
1924 	.init_request	= scsi_mq_init_request,
1925 	.exit_request	= scsi_mq_exit_request,
1926 	.cleanup_rq	= scsi_cleanup_rq,
1927 	.busy		= scsi_mq_lld_busy,
1928 	.map_queues	= scsi_map_queues,
1929 	.init_hctx	= scsi_init_hctx,
1930 	.poll		= scsi_mq_poll,
1931 	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
1932 	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
1933 };
1934 
1935 
1936 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1937 {
1938 	struct Scsi_Host *shost = hctx->driver_data;
1939 
1940 	shost->hostt->commit_rqs(shost, hctx->queue_num);
1941 }
1942 
1943 static const struct blk_mq_ops scsi_mq_ops = {
1944 	.get_budget	= scsi_mq_get_budget,
1945 	.put_budget	= scsi_mq_put_budget,
1946 	.queue_rq	= scsi_queue_rq,
1947 	.commit_rqs	= scsi_commit_rqs,
1948 	.complete	= scsi_complete,
1949 	.timeout	= scsi_timeout,
1950 #ifdef CONFIG_BLK_DEBUG_FS
1951 	.show_rq	= scsi_show_rq,
1952 #endif
1953 	.init_request	= scsi_mq_init_request,
1954 	.exit_request	= scsi_mq_exit_request,
1955 	.cleanup_rq	= scsi_cleanup_rq,
1956 	.busy		= scsi_mq_lld_busy,
1957 	.map_queues	= scsi_map_queues,
1958 	.init_hctx	= scsi_init_hctx,
1959 	.poll		= scsi_mq_poll,
1960 	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
1961 	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
1962 };
1963 
1964 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1965 {
1966 	unsigned int cmd_size, sgl_size;
1967 	struct blk_mq_tag_set *tag_set = &shost->tag_set;
1968 
1969 	sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1970 				scsi_mq_inline_sgl_size(shost));
1971 	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1972 	if (scsi_host_get_prot(shost))
1973 		cmd_size += sizeof(struct scsi_data_buffer) +
1974 			sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1975 
1976 	memset(tag_set, 0, sizeof(*tag_set));
1977 	if (shost->hostt->commit_rqs)
1978 		tag_set->ops = &scsi_mq_ops;
1979 	else
1980 		tag_set->ops = &scsi_mq_ops_no_commit;
1981 	tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1982 	tag_set->nr_maps = shost->nr_maps ? : 1;
1983 	tag_set->queue_depth = shost->can_queue;
1984 	tag_set->cmd_size = cmd_size;
1985 	tag_set->numa_node = NUMA_NO_NODE;
1986 	tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1987 	tag_set->flags |=
1988 		BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1989 	tag_set->driver_data = shost;
1990 	if (shost->host_tagset)
1991 		tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1992 
1993 	return blk_mq_alloc_tag_set(tag_set);
1994 }
1995 
1996 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1997 {
1998 	blk_mq_free_tag_set(&shost->tag_set);
1999 }
2000 
2001 /**
2002  * scsi_device_from_queue - return sdev associated with a request_queue
2003  * @q: The request queue to return the sdev from
2004  *
2005  * Return the sdev associated with a request queue or NULL if the
2006  * request_queue does not reference a SCSI device.
2007  */
2008 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2009 {
2010 	struct scsi_device *sdev = NULL;
2011 
2012 	if (q->mq_ops == &scsi_mq_ops_no_commit ||
2013 	    q->mq_ops == &scsi_mq_ops)
2014 		sdev = q->queuedata;
2015 	if (!sdev || !get_device(&sdev->sdev_gendev))
2016 		sdev = NULL;
2017 
2018 	return sdev;
2019 }
2020 /*
2021  * pktcdvd should have been integrated into the SCSI layers, but for historical
2022  * reasons like the old IDE driver it isn't.  This export allows it to safely
2023  * probe if a given device is a SCSI one and only attach to that.
2024  */
2025 #ifdef CONFIG_CDROM_PKTCDVD_MODULE
2026 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2027 #endif
2028 
2029 /**
2030  * scsi_block_requests - Utility function used by low-level drivers to prevent
2031  * further commands from being queued to the device.
2032  * @shost:  host in question
2033  *
2034  * There is no timer nor any other means by which the requests get unblocked
2035  * other than the low-level driver calling scsi_unblock_requests().
2036  */
2037 void scsi_block_requests(struct Scsi_Host *shost)
2038 {
2039 	shost->host_self_blocked = 1;
2040 }
2041 EXPORT_SYMBOL(scsi_block_requests);
2042 
2043 /**
2044  * scsi_unblock_requests - Utility function used by low-level drivers to allow
2045  * further commands to be queued to the device.
2046  * @shost:  host in question
2047  *
2048  * There is no timer nor any other means by which the requests get unblocked
2049  * other than the low-level driver calling scsi_unblock_requests(). This is done
2050  * as an API function so that changes to the internals of the scsi mid-layer
2051  * won't require wholesale changes to drivers that use this feature.
2052  */
2053 void scsi_unblock_requests(struct Scsi_Host *shost)
2054 {
2055 	shost->host_self_blocked = 0;
2056 	scsi_run_host_queues(shost);
2057 }
2058 EXPORT_SYMBOL(scsi_unblock_requests);
2059 
2060 void scsi_exit_queue(void)
2061 {
2062 	kmem_cache_destroy(scsi_sense_cache);
2063 }
2064 
2065 /**
2066  *	scsi_mode_select - issue a mode select
2067  *	@sdev:	SCSI device to be queried
2068  *	@pf:	Page format bit (1 == standard, 0 == vendor specific)
2069  *	@sp:	Save page bit (0 == don't save, 1 == save)
2070  *	@buffer: request buffer (may not be smaller than eight bytes)
2071  *	@len:	length of request buffer.
2072  *	@timeout: command timeout
2073  *	@retries: number of retries before failing
2074  *	@data: returns a structure abstracting the mode header data
2075  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2076  *		must be SCSI_SENSE_BUFFERSIZE big.
2077  *
2078  *	Returns zero if successful; negative error number or scsi
2079  *	status on error
2080  *
2081  */
2082 int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2083 		     unsigned char *buffer, int len, int timeout, int retries,
2084 		     struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2085 {
2086 	unsigned char cmd[10];
2087 	unsigned char *real_buffer;
2088 	int ret;
2089 
2090 	memset(cmd, 0, sizeof(cmd));
2091 	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2092 
2093 	/*
2094 	 * Use MODE SELECT(10) if the device asked for it or if the mode page
2095 	 * and the mode select header cannot fit within the maximumm 255 bytes
2096 	 * of the MODE SELECT(6) command.
2097 	 */
2098 	if (sdev->use_10_for_ms ||
2099 	    len + 4 > 255 ||
2100 	    data->block_descriptor_length > 255) {
2101 		if (len > 65535 - 8)
2102 			return -EINVAL;
2103 		real_buffer = kmalloc(8 + len, GFP_KERNEL);
2104 		if (!real_buffer)
2105 			return -ENOMEM;
2106 		memcpy(real_buffer + 8, buffer, len);
2107 		len += 8;
2108 		real_buffer[0] = 0;
2109 		real_buffer[1] = 0;
2110 		real_buffer[2] = data->medium_type;
2111 		real_buffer[3] = data->device_specific;
2112 		real_buffer[4] = data->longlba ? 0x01 : 0;
2113 		real_buffer[5] = 0;
2114 		put_unaligned_be16(data->block_descriptor_length,
2115 				   &real_buffer[6]);
2116 
2117 		cmd[0] = MODE_SELECT_10;
2118 		put_unaligned_be16(len, &cmd[7]);
2119 	} else {
2120 		if (data->longlba)
2121 			return -EINVAL;
2122 
2123 		real_buffer = kmalloc(4 + len, GFP_KERNEL);
2124 		if (!real_buffer)
2125 			return -ENOMEM;
2126 		memcpy(real_buffer + 4, buffer, len);
2127 		len += 4;
2128 		real_buffer[0] = 0;
2129 		real_buffer[1] = data->medium_type;
2130 		real_buffer[2] = data->device_specific;
2131 		real_buffer[3] = data->block_descriptor_length;
2132 
2133 		cmd[0] = MODE_SELECT;
2134 		cmd[4] = len;
2135 	}
2136 
2137 	ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2138 			       sshdr, timeout, retries, NULL);
2139 	kfree(real_buffer);
2140 	return ret;
2141 }
2142 EXPORT_SYMBOL_GPL(scsi_mode_select);
2143 
2144 /**
2145  *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2146  *	@sdev:	SCSI device to be queried
2147  *	@dbd:	set to prevent mode sense from returning block descriptors
2148  *	@modepage: mode page being requested
2149  *	@buffer: request buffer (may not be smaller than eight bytes)
2150  *	@len:	length of request buffer.
2151  *	@timeout: command timeout
2152  *	@retries: number of retries before failing
2153  *	@data: returns a structure abstracting the mode header data
2154  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2155  *		must be SCSI_SENSE_BUFFERSIZE big.
2156  *
2157  *	Returns zero if successful, or a negative error number on failure
2158  */
2159 int
2160 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2161 		  unsigned char *buffer, int len, int timeout, int retries,
2162 		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2163 {
2164 	unsigned char cmd[12];
2165 	int use_10_for_ms;
2166 	int header_length;
2167 	int result, retry_count = retries;
2168 	struct scsi_sense_hdr my_sshdr;
2169 
2170 	memset(data, 0, sizeof(*data));
2171 	memset(&cmd[0], 0, 12);
2172 
2173 	dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2174 	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */
2175 	cmd[2] = modepage;
2176 
2177 	/* caller might not be interested in sense, but we need it */
2178 	if (!sshdr)
2179 		sshdr = &my_sshdr;
2180 
2181  retry:
2182 	use_10_for_ms = sdev->use_10_for_ms || len > 255;
2183 
2184 	if (use_10_for_ms) {
2185 		if (len < 8 || len > 65535)
2186 			return -EINVAL;
2187 
2188 		cmd[0] = MODE_SENSE_10;
2189 		put_unaligned_be16(len, &cmd[7]);
2190 		header_length = 8;
2191 	} else {
2192 		if (len < 4)
2193 			return -EINVAL;
2194 
2195 		cmd[0] = MODE_SENSE;
2196 		cmd[4] = len;
2197 		header_length = 4;
2198 	}
2199 
2200 	memset(buffer, 0, len);
2201 
2202 	result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2203 				  sshdr, timeout, retries, NULL);
2204 	if (result < 0)
2205 		return result;
2206 
2207 	/* This code looks awful: what it's doing is making sure an
2208 	 * ILLEGAL REQUEST sense return identifies the actual command
2209 	 * byte as the problem.  MODE_SENSE commands can return
2210 	 * ILLEGAL REQUEST if the code page isn't supported */
2211 
2212 	if (!scsi_status_is_good(result)) {
2213 		if (scsi_sense_valid(sshdr)) {
2214 			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2215 			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2216 				/*
2217 				 * Invalid command operation code: retry using
2218 				 * MODE SENSE(6) if this was a MODE SENSE(10)
2219 				 * request, except if the request mode page is
2220 				 * too large for MODE SENSE single byte
2221 				 * allocation length field.
2222 				 */
2223 				if (use_10_for_ms) {
2224 					if (len > 255)
2225 						return -EIO;
2226 					sdev->use_10_for_ms = 0;
2227 					goto retry;
2228 				}
2229 			}
2230 			if (scsi_status_is_check_condition(result) &&
2231 			    sshdr->sense_key == UNIT_ATTENTION &&
2232 			    retry_count) {
2233 				retry_count--;
2234 				goto retry;
2235 			}
2236 		}
2237 		return -EIO;
2238 	}
2239 	if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2240 		     (modepage == 6 || modepage == 8))) {
2241 		/* Initio breakage? */
2242 		header_length = 0;
2243 		data->length = 13;
2244 		data->medium_type = 0;
2245 		data->device_specific = 0;
2246 		data->longlba = 0;
2247 		data->block_descriptor_length = 0;
2248 	} else if (use_10_for_ms) {
2249 		data->length = get_unaligned_be16(&buffer[0]) + 2;
2250 		data->medium_type = buffer[2];
2251 		data->device_specific = buffer[3];
2252 		data->longlba = buffer[4] & 0x01;
2253 		data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2254 	} else {
2255 		data->length = buffer[0] + 1;
2256 		data->medium_type = buffer[1];
2257 		data->device_specific = buffer[2];
2258 		data->block_descriptor_length = buffer[3];
2259 	}
2260 	data->header_length = header_length;
2261 
2262 	return 0;
2263 }
2264 EXPORT_SYMBOL(scsi_mode_sense);
2265 
2266 /**
2267  *	scsi_test_unit_ready - test if unit is ready
2268  *	@sdev:	scsi device to change the state of.
2269  *	@timeout: command timeout
2270  *	@retries: number of retries before failing
2271  *	@sshdr: outpout pointer for decoded sense information.
2272  *
2273  *	Returns zero if unsuccessful or an error if TUR failed.  For
2274  *	removable media, UNIT_ATTENTION sets ->changed flag.
2275  **/
2276 int
2277 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2278 		     struct scsi_sense_hdr *sshdr)
2279 {
2280 	char cmd[] = {
2281 		TEST_UNIT_READY, 0, 0, 0, 0, 0,
2282 	};
2283 	int result;
2284 
2285 	/* try to eat the UNIT_ATTENTION if there are enough retries */
2286 	do {
2287 		result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2288 					  timeout, 1, NULL);
2289 		if (sdev->removable && scsi_sense_valid(sshdr) &&
2290 		    sshdr->sense_key == UNIT_ATTENTION)
2291 			sdev->changed = 1;
2292 	} while (scsi_sense_valid(sshdr) &&
2293 		 sshdr->sense_key == UNIT_ATTENTION && --retries);
2294 
2295 	return result;
2296 }
2297 EXPORT_SYMBOL(scsi_test_unit_ready);
2298 
2299 /**
2300  *	scsi_device_set_state - Take the given device through the device state model.
2301  *	@sdev:	scsi device to change the state of.
2302  *	@state:	state to change to.
2303  *
2304  *	Returns zero if successful or an error if the requested
2305  *	transition is illegal.
2306  */
2307 int
2308 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2309 {
2310 	enum scsi_device_state oldstate = sdev->sdev_state;
2311 
2312 	if (state == oldstate)
2313 		return 0;
2314 
2315 	switch (state) {
2316 	case SDEV_CREATED:
2317 		switch (oldstate) {
2318 		case SDEV_CREATED_BLOCK:
2319 			break;
2320 		default:
2321 			goto illegal;
2322 		}
2323 		break;
2324 
2325 	case SDEV_RUNNING:
2326 		switch (oldstate) {
2327 		case SDEV_CREATED:
2328 		case SDEV_OFFLINE:
2329 		case SDEV_TRANSPORT_OFFLINE:
2330 		case SDEV_QUIESCE:
2331 		case SDEV_BLOCK:
2332 			break;
2333 		default:
2334 			goto illegal;
2335 		}
2336 		break;
2337 
2338 	case SDEV_QUIESCE:
2339 		switch (oldstate) {
2340 		case SDEV_RUNNING:
2341 		case SDEV_OFFLINE:
2342 		case SDEV_TRANSPORT_OFFLINE:
2343 			break;
2344 		default:
2345 			goto illegal;
2346 		}
2347 		break;
2348 
2349 	case SDEV_OFFLINE:
2350 	case SDEV_TRANSPORT_OFFLINE:
2351 		switch (oldstate) {
2352 		case SDEV_CREATED:
2353 		case SDEV_RUNNING:
2354 		case SDEV_QUIESCE:
2355 		case SDEV_BLOCK:
2356 			break;
2357 		default:
2358 			goto illegal;
2359 		}
2360 		break;
2361 
2362 	case SDEV_BLOCK:
2363 		switch (oldstate) {
2364 		case SDEV_RUNNING:
2365 		case SDEV_CREATED_BLOCK:
2366 		case SDEV_QUIESCE:
2367 		case SDEV_OFFLINE:
2368 			break;
2369 		default:
2370 			goto illegal;
2371 		}
2372 		break;
2373 
2374 	case SDEV_CREATED_BLOCK:
2375 		switch (oldstate) {
2376 		case SDEV_CREATED:
2377 			break;
2378 		default:
2379 			goto illegal;
2380 		}
2381 		break;
2382 
2383 	case SDEV_CANCEL:
2384 		switch (oldstate) {
2385 		case SDEV_CREATED:
2386 		case SDEV_RUNNING:
2387 		case SDEV_QUIESCE:
2388 		case SDEV_OFFLINE:
2389 		case SDEV_TRANSPORT_OFFLINE:
2390 			break;
2391 		default:
2392 			goto illegal;
2393 		}
2394 		break;
2395 
2396 	case SDEV_DEL:
2397 		switch (oldstate) {
2398 		case SDEV_CREATED:
2399 		case SDEV_RUNNING:
2400 		case SDEV_OFFLINE:
2401 		case SDEV_TRANSPORT_OFFLINE:
2402 		case SDEV_CANCEL:
2403 		case SDEV_BLOCK:
2404 		case SDEV_CREATED_BLOCK:
2405 			break;
2406 		default:
2407 			goto illegal;
2408 		}
2409 		break;
2410 
2411 	}
2412 	sdev->offline_already = false;
2413 	sdev->sdev_state = state;
2414 	return 0;
2415 
2416  illegal:
2417 	SCSI_LOG_ERROR_RECOVERY(1,
2418 				sdev_printk(KERN_ERR, sdev,
2419 					    "Illegal state transition %s->%s",
2420 					    scsi_device_state_name(oldstate),
2421 					    scsi_device_state_name(state))
2422 				);
2423 	return -EINVAL;
2424 }
2425 EXPORT_SYMBOL(scsi_device_set_state);
2426 
2427 /**
2428  *	scsi_evt_emit - emit a single SCSI device uevent
2429  *	@sdev: associated SCSI device
2430  *	@evt: event to emit
2431  *
2432  *	Send a single uevent (scsi_event) to the associated scsi_device.
2433  */
2434 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2435 {
2436 	int idx = 0;
2437 	char *envp[3];
2438 
2439 	switch (evt->evt_type) {
2440 	case SDEV_EVT_MEDIA_CHANGE:
2441 		envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2442 		break;
2443 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2444 		scsi_rescan_device(&sdev->sdev_gendev);
2445 		envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2446 		break;
2447 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2448 		envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2449 		break;
2450 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2451 	       envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2452 		break;
2453 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2454 		envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2455 		break;
2456 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2457 		envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2458 		break;
2459 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2460 		envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2461 		break;
2462 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2463 		envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2464 		break;
2465 	default:
2466 		/* do nothing */
2467 		break;
2468 	}
2469 
2470 	envp[idx++] = NULL;
2471 
2472 	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2473 }
2474 
2475 /**
2476  *	scsi_evt_thread - send a uevent for each scsi event
2477  *	@work: work struct for scsi_device
2478  *
2479  *	Dispatch queued events to their associated scsi_device kobjects
2480  *	as uevents.
2481  */
2482 void scsi_evt_thread(struct work_struct *work)
2483 {
2484 	struct scsi_device *sdev;
2485 	enum scsi_device_event evt_type;
2486 	LIST_HEAD(event_list);
2487 
2488 	sdev = container_of(work, struct scsi_device, event_work);
2489 
2490 	for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2491 		if (test_and_clear_bit(evt_type, sdev->pending_events))
2492 			sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2493 
2494 	while (1) {
2495 		struct scsi_event *evt;
2496 		struct list_head *this, *tmp;
2497 		unsigned long flags;
2498 
2499 		spin_lock_irqsave(&sdev->list_lock, flags);
2500 		list_splice_init(&sdev->event_list, &event_list);
2501 		spin_unlock_irqrestore(&sdev->list_lock, flags);
2502 
2503 		if (list_empty(&event_list))
2504 			break;
2505 
2506 		list_for_each_safe(this, tmp, &event_list) {
2507 			evt = list_entry(this, struct scsi_event, node);
2508 			list_del(&evt->node);
2509 			scsi_evt_emit(sdev, evt);
2510 			kfree(evt);
2511 		}
2512 	}
2513 }
2514 
2515 /**
2516  * 	sdev_evt_send - send asserted event to uevent thread
2517  *	@sdev: scsi_device event occurred on
2518  *	@evt: event to send
2519  *
2520  *	Assert scsi device event asynchronously.
2521  */
2522 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2523 {
2524 	unsigned long flags;
2525 
2526 #if 0
2527 	/* FIXME: currently this check eliminates all media change events
2528 	 * for polled devices.  Need to update to discriminate between AN
2529 	 * and polled events */
2530 	if (!test_bit(evt->evt_type, sdev->supported_events)) {
2531 		kfree(evt);
2532 		return;
2533 	}
2534 #endif
2535 
2536 	spin_lock_irqsave(&sdev->list_lock, flags);
2537 	list_add_tail(&evt->node, &sdev->event_list);
2538 	schedule_work(&sdev->event_work);
2539 	spin_unlock_irqrestore(&sdev->list_lock, flags);
2540 }
2541 EXPORT_SYMBOL_GPL(sdev_evt_send);
2542 
2543 /**
2544  * 	sdev_evt_alloc - allocate a new scsi event
2545  *	@evt_type: type of event to allocate
2546  *	@gfpflags: GFP flags for allocation
2547  *
2548  *	Allocates and returns a new scsi_event.
2549  */
2550 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2551 				  gfp_t gfpflags)
2552 {
2553 	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2554 	if (!evt)
2555 		return NULL;
2556 
2557 	evt->evt_type = evt_type;
2558 	INIT_LIST_HEAD(&evt->node);
2559 
2560 	/* evt_type-specific initialization, if any */
2561 	switch (evt_type) {
2562 	case SDEV_EVT_MEDIA_CHANGE:
2563 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2564 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2565 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2566 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2567 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2568 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2569 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2570 	default:
2571 		/* do nothing */
2572 		break;
2573 	}
2574 
2575 	return evt;
2576 }
2577 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2578 
2579 /**
2580  * 	sdev_evt_send_simple - send asserted event to uevent thread
2581  *	@sdev: scsi_device event occurred on
2582  *	@evt_type: type of event to send
2583  *	@gfpflags: GFP flags for allocation
2584  *
2585  *	Assert scsi device event asynchronously, given an event type.
2586  */
2587 void sdev_evt_send_simple(struct scsi_device *sdev,
2588 			  enum scsi_device_event evt_type, gfp_t gfpflags)
2589 {
2590 	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2591 	if (!evt) {
2592 		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2593 			    evt_type);
2594 		return;
2595 	}
2596 
2597 	sdev_evt_send(sdev, evt);
2598 }
2599 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2600 
2601 /**
2602  *	scsi_device_quiesce - Block all commands except power management.
2603  *	@sdev:	scsi device to quiesce.
2604  *
2605  *	This works by trying to transition to the SDEV_QUIESCE state
2606  *	(which must be a legal transition).  When the device is in this
2607  *	state, only power management requests will be accepted, all others will
2608  *	be deferred.
2609  *
2610  *	Must be called with user context, may sleep.
2611  *
2612  *	Returns zero if unsuccessful or an error if not.
2613  */
2614 int
2615 scsi_device_quiesce(struct scsi_device *sdev)
2616 {
2617 	struct request_queue *q = sdev->request_queue;
2618 	int err;
2619 
2620 	/*
2621 	 * It is allowed to call scsi_device_quiesce() multiple times from
2622 	 * the same context but concurrent scsi_device_quiesce() calls are
2623 	 * not allowed.
2624 	 */
2625 	WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2626 
2627 	if (sdev->quiesced_by == current)
2628 		return 0;
2629 
2630 	blk_set_pm_only(q);
2631 
2632 	blk_mq_freeze_queue(q);
2633 	/*
2634 	 * Ensure that the effect of blk_set_pm_only() will be visible
2635 	 * for percpu_ref_tryget() callers that occur after the queue
2636 	 * unfreeze even if the queue was already frozen before this function
2637 	 * was called. See also https://lwn.net/Articles/573497/.
2638 	 */
2639 	synchronize_rcu();
2640 	blk_mq_unfreeze_queue(q);
2641 
2642 	mutex_lock(&sdev->state_mutex);
2643 	err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2644 	if (err == 0)
2645 		sdev->quiesced_by = current;
2646 	else
2647 		blk_clear_pm_only(q);
2648 	mutex_unlock(&sdev->state_mutex);
2649 
2650 	return err;
2651 }
2652 EXPORT_SYMBOL(scsi_device_quiesce);
2653 
2654 /**
2655  *	scsi_device_resume - Restart user issued commands to a quiesced device.
2656  *	@sdev:	scsi device to resume.
2657  *
2658  *	Moves the device from quiesced back to running and restarts the
2659  *	queues.
2660  *
2661  *	Must be called with user context, may sleep.
2662  */
2663 void scsi_device_resume(struct scsi_device *sdev)
2664 {
2665 	/* check if the device state was mutated prior to resume, and if
2666 	 * so assume the state is being managed elsewhere (for example
2667 	 * device deleted during suspend)
2668 	 */
2669 	mutex_lock(&sdev->state_mutex);
2670 	if (sdev->sdev_state == SDEV_QUIESCE)
2671 		scsi_device_set_state(sdev, SDEV_RUNNING);
2672 	if (sdev->quiesced_by) {
2673 		sdev->quiesced_by = NULL;
2674 		blk_clear_pm_only(sdev->request_queue);
2675 	}
2676 	mutex_unlock(&sdev->state_mutex);
2677 }
2678 EXPORT_SYMBOL(scsi_device_resume);
2679 
2680 static void
2681 device_quiesce_fn(struct scsi_device *sdev, void *data)
2682 {
2683 	scsi_device_quiesce(sdev);
2684 }
2685 
2686 void
2687 scsi_target_quiesce(struct scsi_target *starget)
2688 {
2689 	starget_for_each_device(starget, NULL, device_quiesce_fn);
2690 }
2691 EXPORT_SYMBOL(scsi_target_quiesce);
2692 
2693 static void
2694 device_resume_fn(struct scsi_device *sdev, void *data)
2695 {
2696 	scsi_device_resume(sdev);
2697 }
2698 
2699 void
2700 scsi_target_resume(struct scsi_target *starget)
2701 {
2702 	starget_for_each_device(starget, NULL, device_resume_fn);
2703 }
2704 EXPORT_SYMBOL(scsi_target_resume);
2705 
2706 static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2707 {
2708 	if (scsi_device_set_state(sdev, SDEV_BLOCK))
2709 		return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2710 
2711 	return 0;
2712 }
2713 
2714 void scsi_start_queue(struct scsi_device *sdev)
2715 {
2716 	if (cmpxchg(&sdev->queue_stopped, 1, 0))
2717 		blk_mq_unquiesce_queue(sdev->request_queue);
2718 }
2719 
2720 static void scsi_stop_queue(struct scsi_device *sdev, bool nowait)
2721 {
2722 	/*
2723 	 * The atomic variable of ->queue_stopped covers that
2724 	 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2725 	 *
2726 	 * However, we still need to wait until quiesce is done
2727 	 * in case that queue has been stopped.
2728 	 */
2729 	if (!cmpxchg(&sdev->queue_stopped, 0, 1)) {
2730 		if (nowait)
2731 			blk_mq_quiesce_queue_nowait(sdev->request_queue);
2732 		else
2733 			blk_mq_quiesce_queue(sdev->request_queue);
2734 	} else {
2735 		if (!nowait)
2736 			blk_mq_wait_quiesce_done(sdev->request_queue);
2737 	}
2738 }
2739 
2740 /**
2741  * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2742  * @sdev: device to block
2743  *
2744  * Pause SCSI command processing on the specified device. Does not sleep.
2745  *
2746  * Returns zero if successful or a negative error code upon failure.
2747  *
2748  * Notes:
2749  * This routine transitions the device to the SDEV_BLOCK state (which must be
2750  * a legal transition). When the device is in this state, command processing
2751  * is paused until the device leaves the SDEV_BLOCK state. See also
2752  * scsi_internal_device_unblock_nowait().
2753  */
2754 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2755 {
2756 	int ret = __scsi_internal_device_block_nowait(sdev);
2757 
2758 	/*
2759 	 * The device has transitioned to SDEV_BLOCK.  Stop the
2760 	 * block layer from calling the midlayer with this device's
2761 	 * request queue.
2762 	 */
2763 	if (!ret)
2764 		scsi_stop_queue(sdev, true);
2765 	return ret;
2766 }
2767 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2768 
2769 /**
2770  * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2771  * @sdev: device to block
2772  *
2773  * Pause SCSI command processing on the specified device and wait until all
2774  * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2775  *
2776  * Returns zero if successful or a negative error code upon failure.
2777  *
2778  * Note:
2779  * This routine transitions the device to the SDEV_BLOCK state (which must be
2780  * a legal transition). When the device is in this state, command processing
2781  * is paused until the device leaves the SDEV_BLOCK state. See also
2782  * scsi_internal_device_unblock().
2783  */
2784 static int scsi_internal_device_block(struct scsi_device *sdev)
2785 {
2786 	int err;
2787 
2788 	mutex_lock(&sdev->state_mutex);
2789 	err = __scsi_internal_device_block_nowait(sdev);
2790 	if (err == 0)
2791 		scsi_stop_queue(sdev, false);
2792 	mutex_unlock(&sdev->state_mutex);
2793 
2794 	return err;
2795 }
2796 
2797 /**
2798  * scsi_internal_device_unblock_nowait - resume a device after a block request
2799  * @sdev:	device to resume
2800  * @new_state:	state to set the device to after unblocking
2801  *
2802  * Restart the device queue for a previously suspended SCSI device. Does not
2803  * sleep.
2804  *
2805  * Returns zero if successful or a negative error code upon failure.
2806  *
2807  * Notes:
2808  * This routine transitions the device to the SDEV_RUNNING state or to one of
2809  * the offline states (which must be a legal transition) allowing the midlayer
2810  * to goose the queue for this device.
2811  */
2812 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2813 					enum scsi_device_state new_state)
2814 {
2815 	switch (new_state) {
2816 	case SDEV_RUNNING:
2817 	case SDEV_TRANSPORT_OFFLINE:
2818 		break;
2819 	default:
2820 		return -EINVAL;
2821 	}
2822 
2823 	/*
2824 	 * Try to transition the scsi device to SDEV_RUNNING or one of the
2825 	 * offlined states and goose the device queue if successful.
2826 	 */
2827 	switch (sdev->sdev_state) {
2828 	case SDEV_BLOCK:
2829 	case SDEV_TRANSPORT_OFFLINE:
2830 		sdev->sdev_state = new_state;
2831 		break;
2832 	case SDEV_CREATED_BLOCK:
2833 		if (new_state == SDEV_TRANSPORT_OFFLINE ||
2834 		    new_state == SDEV_OFFLINE)
2835 			sdev->sdev_state = new_state;
2836 		else
2837 			sdev->sdev_state = SDEV_CREATED;
2838 		break;
2839 	case SDEV_CANCEL:
2840 	case SDEV_OFFLINE:
2841 		break;
2842 	default:
2843 		return -EINVAL;
2844 	}
2845 	scsi_start_queue(sdev);
2846 
2847 	return 0;
2848 }
2849 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2850 
2851 /**
2852  * scsi_internal_device_unblock - resume a device after a block request
2853  * @sdev:	device to resume
2854  * @new_state:	state to set the device to after unblocking
2855  *
2856  * Restart the device queue for a previously suspended SCSI device. May sleep.
2857  *
2858  * Returns zero if successful or a negative error code upon failure.
2859  *
2860  * Notes:
2861  * This routine transitions the device to the SDEV_RUNNING state or to one of
2862  * the offline states (which must be a legal transition) allowing the midlayer
2863  * to goose the queue for this device.
2864  */
2865 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2866 					enum scsi_device_state new_state)
2867 {
2868 	int ret;
2869 
2870 	mutex_lock(&sdev->state_mutex);
2871 	ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2872 	mutex_unlock(&sdev->state_mutex);
2873 
2874 	return ret;
2875 }
2876 
2877 static void
2878 device_block(struct scsi_device *sdev, void *data)
2879 {
2880 	int ret;
2881 
2882 	ret = scsi_internal_device_block(sdev);
2883 
2884 	WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2885 		  dev_name(&sdev->sdev_gendev), ret);
2886 }
2887 
2888 static int
2889 target_block(struct device *dev, void *data)
2890 {
2891 	if (scsi_is_target_device(dev))
2892 		starget_for_each_device(to_scsi_target(dev), NULL,
2893 					device_block);
2894 	return 0;
2895 }
2896 
2897 void
2898 scsi_target_block(struct device *dev)
2899 {
2900 	if (scsi_is_target_device(dev))
2901 		starget_for_each_device(to_scsi_target(dev), NULL,
2902 					device_block);
2903 	else
2904 		device_for_each_child(dev, NULL, target_block);
2905 }
2906 EXPORT_SYMBOL_GPL(scsi_target_block);
2907 
2908 static void
2909 device_unblock(struct scsi_device *sdev, void *data)
2910 {
2911 	scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2912 }
2913 
2914 static int
2915 target_unblock(struct device *dev, void *data)
2916 {
2917 	if (scsi_is_target_device(dev))
2918 		starget_for_each_device(to_scsi_target(dev), data,
2919 					device_unblock);
2920 	return 0;
2921 }
2922 
2923 void
2924 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2925 {
2926 	if (scsi_is_target_device(dev))
2927 		starget_for_each_device(to_scsi_target(dev), &new_state,
2928 					device_unblock);
2929 	else
2930 		device_for_each_child(dev, &new_state, target_unblock);
2931 }
2932 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2933 
2934 int
2935 scsi_host_block(struct Scsi_Host *shost)
2936 {
2937 	struct scsi_device *sdev;
2938 	int ret = 0;
2939 
2940 	/*
2941 	 * Call scsi_internal_device_block_nowait so we can avoid
2942 	 * calling synchronize_rcu() for each LUN.
2943 	 */
2944 	shost_for_each_device(sdev, shost) {
2945 		mutex_lock(&sdev->state_mutex);
2946 		ret = scsi_internal_device_block_nowait(sdev);
2947 		mutex_unlock(&sdev->state_mutex);
2948 		if (ret) {
2949 			scsi_device_put(sdev);
2950 			break;
2951 		}
2952 	}
2953 
2954 	/*
2955 	 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2956 	 * calling synchronize_rcu() once is enough.
2957 	 */
2958 	WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2959 
2960 	if (!ret)
2961 		synchronize_rcu();
2962 
2963 	return ret;
2964 }
2965 EXPORT_SYMBOL_GPL(scsi_host_block);
2966 
2967 int
2968 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2969 {
2970 	struct scsi_device *sdev;
2971 	int ret = 0;
2972 
2973 	shost_for_each_device(sdev, shost) {
2974 		ret = scsi_internal_device_unblock(sdev, new_state);
2975 		if (ret) {
2976 			scsi_device_put(sdev);
2977 			break;
2978 		}
2979 	}
2980 	return ret;
2981 }
2982 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2983 
2984 /**
2985  * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2986  * @sgl:	scatter-gather list
2987  * @sg_count:	number of segments in sg
2988  * @offset:	offset in bytes into sg, on return offset into the mapped area
2989  * @len:	bytes to map, on return number of bytes mapped
2990  *
2991  * Returns virtual address of the start of the mapped page
2992  */
2993 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2994 			  size_t *offset, size_t *len)
2995 {
2996 	int i;
2997 	size_t sg_len = 0, len_complete = 0;
2998 	struct scatterlist *sg;
2999 	struct page *page;
3000 
3001 	WARN_ON(!irqs_disabled());
3002 
3003 	for_each_sg(sgl, sg, sg_count, i) {
3004 		len_complete = sg_len; /* Complete sg-entries */
3005 		sg_len += sg->length;
3006 		if (sg_len > *offset)
3007 			break;
3008 	}
3009 
3010 	if (unlikely(i == sg_count)) {
3011 		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3012 			"elements %d\n",
3013 		       __func__, sg_len, *offset, sg_count);
3014 		WARN_ON(1);
3015 		return NULL;
3016 	}
3017 
3018 	/* Offset starting from the beginning of first page in this sg-entry */
3019 	*offset = *offset - len_complete + sg->offset;
3020 
3021 	/* Assumption: contiguous pages can be accessed as "page + i" */
3022 	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3023 	*offset &= ~PAGE_MASK;
3024 
3025 	/* Bytes in this sg-entry from *offset to the end of the page */
3026 	sg_len = PAGE_SIZE - *offset;
3027 	if (*len > sg_len)
3028 		*len = sg_len;
3029 
3030 	return kmap_atomic(page);
3031 }
3032 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3033 
3034 /**
3035  * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3036  * @virt:	virtual address to be unmapped
3037  */
3038 void scsi_kunmap_atomic_sg(void *virt)
3039 {
3040 	kunmap_atomic(virt);
3041 }
3042 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3043 
3044 void sdev_disable_disk_events(struct scsi_device *sdev)
3045 {
3046 	atomic_inc(&sdev->disk_events_disable_depth);
3047 }
3048 EXPORT_SYMBOL(sdev_disable_disk_events);
3049 
3050 void sdev_enable_disk_events(struct scsi_device *sdev)
3051 {
3052 	if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3053 		return;
3054 	atomic_dec(&sdev->disk_events_disable_depth);
3055 }
3056 EXPORT_SYMBOL(sdev_enable_disk_events);
3057 
3058 static unsigned char designator_prio(const unsigned char *d)
3059 {
3060 	if (d[1] & 0x30)
3061 		/* not associated with LUN */
3062 		return 0;
3063 
3064 	if (d[3] == 0)
3065 		/* invalid length */
3066 		return 0;
3067 
3068 	/*
3069 	 * Order of preference for lun descriptor:
3070 	 * - SCSI name string
3071 	 * - NAA IEEE Registered Extended
3072 	 * - EUI-64 based 16-byte
3073 	 * - EUI-64 based 12-byte
3074 	 * - NAA IEEE Registered
3075 	 * - NAA IEEE Extended
3076 	 * - EUI-64 based 8-byte
3077 	 * - SCSI name string (truncated)
3078 	 * - T10 Vendor ID
3079 	 * as longer descriptors reduce the likelyhood
3080 	 * of identification clashes.
3081 	 */
3082 
3083 	switch (d[1] & 0xf) {
3084 	case 8:
3085 		/* SCSI name string, variable-length UTF-8 */
3086 		return 9;
3087 	case 3:
3088 		switch (d[4] >> 4) {
3089 		case 6:
3090 			/* NAA registered extended */
3091 			return 8;
3092 		case 5:
3093 			/* NAA registered */
3094 			return 5;
3095 		case 4:
3096 			/* NAA extended */
3097 			return 4;
3098 		case 3:
3099 			/* NAA locally assigned */
3100 			return 1;
3101 		default:
3102 			break;
3103 		}
3104 		break;
3105 	case 2:
3106 		switch (d[3]) {
3107 		case 16:
3108 			/* EUI64-based, 16 byte */
3109 			return 7;
3110 		case 12:
3111 			/* EUI64-based, 12 byte */
3112 			return 6;
3113 		case 8:
3114 			/* EUI64-based, 8 byte */
3115 			return 3;
3116 		default:
3117 			break;
3118 		}
3119 		break;
3120 	case 1:
3121 		/* T10 vendor ID */
3122 		return 1;
3123 	default:
3124 		break;
3125 	}
3126 
3127 	return 0;
3128 }
3129 
3130 /**
3131  * scsi_vpd_lun_id - return a unique device identification
3132  * @sdev: SCSI device
3133  * @id:   buffer for the identification
3134  * @id_len:  length of the buffer
3135  *
3136  * Copies a unique device identification into @id based
3137  * on the information in the VPD page 0x83 of the device.
3138  * The string will be formatted as a SCSI name string.
3139  *
3140  * Returns the length of the identification or error on failure.
3141  * If the identifier is longer than the supplied buffer the actual
3142  * identifier length is returned and the buffer is not zero-padded.
3143  */
3144 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3145 {
3146 	u8 cur_id_prio = 0;
3147 	u8 cur_id_size = 0;
3148 	const unsigned char *d, *cur_id_str;
3149 	const struct scsi_vpd *vpd_pg83;
3150 	int id_size = -EINVAL;
3151 
3152 	rcu_read_lock();
3153 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3154 	if (!vpd_pg83) {
3155 		rcu_read_unlock();
3156 		return -ENXIO;
3157 	}
3158 
3159 	/* The id string must be at least 20 bytes + terminating NULL byte */
3160 	if (id_len < 21) {
3161 		rcu_read_unlock();
3162 		return -EINVAL;
3163 	}
3164 
3165 	memset(id, 0, id_len);
3166 	for (d = vpd_pg83->data + 4;
3167 	     d < vpd_pg83->data + vpd_pg83->len;
3168 	     d += d[3] + 4) {
3169 		u8 prio = designator_prio(d);
3170 
3171 		if (prio == 0 || cur_id_prio > prio)
3172 			continue;
3173 
3174 		switch (d[1] & 0xf) {
3175 		case 0x1:
3176 			/* T10 Vendor ID */
3177 			if (cur_id_size > d[3])
3178 				break;
3179 			cur_id_prio = prio;
3180 			cur_id_size = d[3];
3181 			if (cur_id_size + 4 > id_len)
3182 				cur_id_size = id_len - 4;
3183 			cur_id_str = d + 4;
3184 			id_size = snprintf(id, id_len, "t10.%*pE",
3185 					   cur_id_size, cur_id_str);
3186 			break;
3187 		case 0x2:
3188 			/* EUI-64 */
3189 			cur_id_prio = prio;
3190 			cur_id_size = d[3];
3191 			cur_id_str = d + 4;
3192 			switch (cur_id_size) {
3193 			case 8:
3194 				id_size = snprintf(id, id_len,
3195 						   "eui.%8phN",
3196 						   cur_id_str);
3197 				break;
3198 			case 12:
3199 				id_size = snprintf(id, id_len,
3200 						   "eui.%12phN",
3201 						   cur_id_str);
3202 				break;
3203 			case 16:
3204 				id_size = snprintf(id, id_len,
3205 						   "eui.%16phN",
3206 						   cur_id_str);
3207 				break;
3208 			default:
3209 				break;
3210 			}
3211 			break;
3212 		case 0x3:
3213 			/* NAA */
3214 			cur_id_prio = prio;
3215 			cur_id_size = d[3];
3216 			cur_id_str = d + 4;
3217 			switch (cur_id_size) {
3218 			case 8:
3219 				id_size = snprintf(id, id_len,
3220 						   "naa.%8phN",
3221 						   cur_id_str);
3222 				break;
3223 			case 16:
3224 				id_size = snprintf(id, id_len,
3225 						   "naa.%16phN",
3226 						   cur_id_str);
3227 				break;
3228 			default:
3229 				break;
3230 			}
3231 			break;
3232 		case 0x8:
3233 			/* SCSI name string */
3234 			if (cur_id_size > d[3])
3235 				break;
3236 			/* Prefer others for truncated descriptor */
3237 			if (d[3] > id_len) {
3238 				prio = 2;
3239 				if (cur_id_prio > prio)
3240 					break;
3241 			}
3242 			cur_id_prio = prio;
3243 			cur_id_size = id_size = d[3];
3244 			cur_id_str = d + 4;
3245 			if (cur_id_size >= id_len)
3246 				cur_id_size = id_len - 1;
3247 			memcpy(id, cur_id_str, cur_id_size);
3248 			break;
3249 		default:
3250 			break;
3251 		}
3252 	}
3253 	rcu_read_unlock();
3254 
3255 	return id_size;
3256 }
3257 EXPORT_SYMBOL(scsi_vpd_lun_id);
3258 
3259 /*
3260  * scsi_vpd_tpg_id - return a target port group identifier
3261  * @sdev: SCSI device
3262  *
3263  * Returns the Target Port Group identifier from the information
3264  * froom VPD page 0x83 of the device.
3265  *
3266  * Returns the identifier or error on failure.
3267  */
3268 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3269 {
3270 	const unsigned char *d;
3271 	const struct scsi_vpd *vpd_pg83;
3272 	int group_id = -EAGAIN, rel_port = -1;
3273 
3274 	rcu_read_lock();
3275 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3276 	if (!vpd_pg83) {
3277 		rcu_read_unlock();
3278 		return -ENXIO;
3279 	}
3280 
3281 	d = vpd_pg83->data + 4;
3282 	while (d < vpd_pg83->data + vpd_pg83->len) {
3283 		switch (d[1] & 0xf) {
3284 		case 0x4:
3285 			/* Relative target port */
3286 			rel_port = get_unaligned_be16(&d[6]);
3287 			break;
3288 		case 0x5:
3289 			/* Target port group */
3290 			group_id = get_unaligned_be16(&d[6]);
3291 			break;
3292 		default:
3293 			break;
3294 		}
3295 		d += d[3] + 4;
3296 	}
3297 	rcu_read_unlock();
3298 
3299 	if (group_id >= 0 && rel_id && rel_port != -1)
3300 		*rel_id = rel_port;
3301 
3302 	return group_id;
3303 }
3304 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3305 
3306 /**
3307  * scsi_build_sense - build sense data for a command
3308  * @scmd:	scsi command for which the sense should be formatted
3309  * @desc:	Sense format (non-zero == descriptor format,
3310  *              0 == fixed format)
3311  * @key:	Sense key
3312  * @asc:	Additional sense code
3313  * @ascq:	Additional sense code qualifier
3314  *
3315  **/
3316 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3317 {
3318 	scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3319 	scmd->result = SAM_STAT_CHECK_CONDITION;
3320 }
3321 EXPORT_SYMBOL_GPL(scsi_build_sense);
3322