xref: /openbmc/linux/drivers/scsi/scsi_lib.c (revision 8795a739)
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/ratelimit.h>
25 #include <asm/unaligned.h>
26 
27 #include <scsi/scsi.h>
28 #include <scsi/scsi_cmnd.h>
29 #include <scsi/scsi_dbg.h>
30 #include <scsi/scsi_device.h>
31 #include <scsi/scsi_driver.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_host.h>
34 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
35 #include <scsi/scsi_dh.h>
36 
37 #include <trace/events/scsi.h>
38 
39 #include "scsi_debugfs.h"
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
42 
43 /*
44  * Size of integrity metadata is usually small, 1 inline sg should
45  * cover normal cases.
46  */
47 #ifdef CONFIG_ARCH_NO_SG_CHAIN
48 #define  SCSI_INLINE_PROT_SG_CNT  0
49 #define  SCSI_INLINE_SG_CNT  0
50 #else
51 #define  SCSI_INLINE_PROT_SG_CNT  1
52 #define  SCSI_INLINE_SG_CNT  2
53 #endif
54 
55 static struct kmem_cache *scsi_sdb_cache;
56 static struct kmem_cache *scsi_sense_cache;
57 static struct kmem_cache *scsi_sense_isadma_cache;
58 static DEFINE_MUTEX(scsi_sense_cache_mutex);
59 
60 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
61 
62 static inline struct kmem_cache *
63 scsi_select_sense_cache(bool unchecked_isa_dma)
64 {
65 	return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
66 }
67 
68 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
69 				   unsigned char *sense_buffer)
70 {
71 	kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
72 			sense_buffer);
73 }
74 
75 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
76 	gfp_t gfp_mask, int numa_node)
77 {
78 	return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
79 				     gfp_mask, numa_node);
80 }
81 
82 int scsi_init_sense_cache(struct Scsi_Host *shost)
83 {
84 	struct kmem_cache *cache;
85 	int ret = 0;
86 
87 	mutex_lock(&scsi_sense_cache_mutex);
88 	cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
89 	if (cache)
90 		goto exit;
91 
92 	if (shost->unchecked_isa_dma) {
93 		scsi_sense_isadma_cache =
94 			kmem_cache_create("scsi_sense_cache(DMA)",
95 				SCSI_SENSE_BUFFERSIZE, 0,
96 				SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
97 		if (!scsi_sense_isadma_cache)
98 			ret = -ENOMEM;
99 	} else {
100 		scsi_sense_cache =
101 			kmem_cache_create_usercopy("scsi_sense_cache",
102 				SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
103 				0, SCSI_SENSE_BUFFERSIZE, NULL);
104 		if (!scsi_sense_cache)
105 			ret = -ENOMEM;
106 	}
107  exit:
108 	mutex_unlock(&scsi_sense_cache_mutex);
109 	return ret;
110 }
111 
112 /*
113  * When to reinvoke queueing after a resource shortage. It's 3 msecs to
114  * not change behaviour from the previous unplug mechanism, experimentation
115  * may prove this needs changing.
116  */
117 #define SCSI_QUEUE_DELAY	3
118 
119 static void
120 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
121 {
122 	struct Scsi_Host *host = cmd->device->host;
123 	struct scsi_device *device = cmd->device;
124 	struct scsi_target *starget = scsi_target(device);
125 
126 	/*
127 	 * Set the appropriate busy bit for the device/host.
128 	 *
129 	 * If the host/device isn't busy, assume that something actually
130 	 * completed, and that we should be able to queue a command now.
131 	 *
132 	 * Note that the prior mid-layer assumption that any host could
133 	 * always queue at least one command is now broken.  The mid-layer
134 	 * will implement a user specifiable stall (see
135 	 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
136 	 * if a command is requeued with no other commands outstanding
137 	 * either for the device or for the host.
138 	 */
139 	switch (reason) {
140 	case SCSI_MLQUEUE_HOST_BUSY:
141 		atomic_set(&host->host_blocked, host->max_host_blocked);
142 		break;
143 	case SCSI_MLQUEUE_DEVICE_BUSY:
144 	case SCSI_MLQUEUE_EH_RETRY:
145 		atomic_set(&device->device_blocked,
146 			   device->max_device_blocked);
147 		break;
148 	case SCSI_MLQUEUE_TARGET_BUSY:
149 		atomic_set(&starget->target_blocked,
150 			   starget->max_target_blocked);
151 		break;
152 	}
153 }
154 
155 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
156 {
157 	if (cmd->request->rq_flags & RQF_DONTPREP) {
158 		cmd->request->rq_flags &= ~RQF_DONTPREP;
159 		scsi_mq_uninit_cmd(cmd);
160 	} else {
161 		WARN_ON_ONCE(true);
162 	}
163 	blk_mq_requeue_request(cmd->request, true);
164 }
165 
166 /**
167  * __scsi_queue_insert - private queue insertion
168  * @cmd: The SCSI command being requeued
169  * @reason:  The reason for the requeue
170  * @unbusy: Whether the queue should be unbusied
171  *
172  * This is a private queue insertion.  The public interface
173  * scsi_queue_insert() always assumes the queue should be unbusied
174  * because it's always called before the completion.  This function is
175  * for a requeue after completion, which should only occur in this
176  * file.
177  */
178 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
179 {
180 	struct scsi_device *device = cmd->device;
181 
182 	SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
183 		"Inserting command %p into mlqueue\n", cmd));
184 
185 	scsi_set_blocked(cmd, reason);
186 
187 	/*
188 	 * Decrement the counters, since these commands are no longer
189 	 * active on the host/device.
190 	 */
191 	if (unbusy)
192 		scsi_device_unbusy(device);
193 
194 	/*
195 	 * Requeue this command.  It will go before all other commands
196 	 * that are already in the queue. Schedule requeue work under
197 	 * lock such that the kblockd_schedule_work() call happens
198 	 * before blk_cleanup_queue() finishes.
199 	 */
200 	cmd->result = 0;
201 
202 	blk_mq_requeue_request(cmd->request, true);
203 }
204 
205 /*
206  * Function:    scsi_queue_insert()
207  *
208  * Purpose:     Insert a command in the midlevel queue.
209  *
210  * Arguments:   cmd    - command that we are adding to queue.
211  *              reason - why we are inserting command to queue.
212  *
213  * Lock status: Assumed that lock is not held upon entry.
214  *
215  * Returns:     Nothing.
216  *
217  * Notes:       We do this for one of two cases.  Either the host is busy
218  *              and it cannot accept any more commands for the time being,
219  *              or the device returned QUEUE_FULL and can accept no more
220  *              commands.
221  * Notes:       This could be called either from an interrupt context or a
222  *              normal process context.
223  */
224 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
225 {
226 	__scsi_queue_insert(cmd, reason, true);
227 }
228 
229 
230 /**
231  * __scsi_execute - insert request and wait for the result
232  * @sdev:	scsi device
233  * @cmd:	scsi command
234  * @data_direction: data direction
235  * @buffer:	data buffer
236  * @bufflen:	len of buffer
237  * @sense:	optional sense buffer
238  * @sshdr:	optional decoded sense header
239  * @timeout:	request timeout in seconds
240  * @retries:	number of times to retry request
241  * @flags:	flags for ->cmd_flags
242  * @rq_flags:	flags for ->rq_flags
243  * @resid:	optional residual length
244  *
245  * Returns the scsi_cmnd result field if a command was executed, or a negative
246  * Linux error code if we didn't get that far.
247  */
248 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
249 		 int data_direction, void *buffer, unsigned bufflen,
250 		 unsigned char *sense, struct scsi_sense_hdr *sshdr,
251 		 int timeout, int retries, u64 flags, req_flags_t rq_flags,
252 		 int *resid)
253 {
254 	struct request *req;
255 	struct scsi_request *rq;
256 	int ret = DRIVER_ERROR << 24;
257 
258 	req = blk_get_request(sdev->request_queue,
259 			data_direction == DMA_TO_DEVICE ?
260 			REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
261 	if (IS_ERR(req))
262 		return ret;
263 	rq = scsi_req(req);
264 
265 	if (bufflen &&	blk_rq_map_kern(sdev->request_queue, req,
266 					buffer, bufflen, GFP_NOIO))
267 		goto out;
268 
269 	rq->cmd_len = COMMAND_SIZE(cmd[0]);
270 	memcpy(rq->cmd, cmd, rq->cmd_len);
271 	rq->retries = retries;
272 	req->timeout = timeout;
273 	req->cmd_flags |= flags;
274 	req->rq_flags |= rq_flags | RQF_QUIET;
275 
276 	/*
277 	 * head injection *required* here otherwise quiesce won't work
278 	 */
279 	blk_execute_rq(req->q, NULL, req, 1);
280 
281 	/*
282 	 * Some devices (USB mass-storage in particular) may transfer
283 	 * garbage data together with a residue indicating that the data
284 	 * is invalid.  Prevent the garbage from being misinterpreted
285 	 * and prevent security leaks by zeroing out the excess data.
286 	 */
287 	if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
288 		memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
289 
290 	if (resid)
291 		*resid = rq->resid_len;
292 	if (sense && rq->sense_len)
293 		memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
294 	if (sshdr)
295 		scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
296 	ret = rq->result;
297  out:
298 	blk_put_request(req);
299 
300 	return ret;
301 }
302 EXPORT_SYMBOL(__scsi_execute);
303 
304 /*
305  * Function:    scsi_init_cmd_errh()
306  *
307  * Purpose:     Initialize cmd fields related to error handling.
308  *
309  * Arguments:   cmd	- command that is ready to be queued.
310  *
311  * Notes:       This function has the job of initializing a number of
312  *              fields related to error handling.   Typically this will
313  *              be called once for each command, as required.
314  */
315 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
316 {
317 	scsi_set_resid(cmd, 0);
318 	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
319 	if (cmd->cmd_len == 0)
320 		cmd->cmd_len = scsi_command_size(cmd->cmnd);
321 }
322 
323 /*
324  * Decrement the host_busy counter and wake up the error handler if necessary.
325  * Avoid as follows that the error handler is not woken up if shost->host_busy
326  * == shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
327  * with an RCU read lock in this function to ensure that this function in its
328  * entirety either finishes before scsi_eh_scmd_add() increases the
329  * host_failed counter or that it notices the shost state change made by
330  * scsi_eh_scmd_add().
331  */
332 static void scsi_dec_host_busy(struct Scsi_Host *shost)
333 {
334 	unsigned long flags;
335 
336 	rcu_read_lock();
337 	atomic_dec(&shost->host_busy);
338 	if (unlikely(scsi_host_in_recovery(shost))) {
339 		spin_lock_irqsave(shost->host_lock, flags);
340 		if (shost->host_failed || shost->host_eh_scheduled)
341 			scsi_eh_wakeup(shost);
342 		spin_unlock_irqrestore(shost->host_lock, flags);
343 	}
344 	rcu_read_unlock();
345 }
346 
347 void scsi_device_unbusy(struct scsi_device *sdev)
348 {
349 	struct Scsi_Host *shost = sdev->host;
350 	struct scsi_target *starget = scsi_target(sdev);
351 
352 	scsi_dec_host_busy(shost);
353 
354 	if (starget->can_queue > 0)
355 		atomic_dec(&starget->target_busy);
356 
357 	atomic_dec(&sdev->device_busy);
358 }
359 
360 static void scsi_kick_queue(struct request_queue *q)
361 {
362 	blk_mq_run_hw_queues(q, false);
363 }
364 
365 /*
366  * Called for single_lun devices on IO completion. Clear starget_sdev_user,
367  * and call blk_run_queue for all the scsi_devices on the target -
368  * including current_sdev first.
369  *
370  * Called with *no* scsi locks held.
371  */
372 static void scsi_single_lun_run(struct scsi_device *current_sdev)
373 {
374 	struct Scsi_Host *shost = current_sdev->host;
375 	struct scsi_device *sdev, *tmp;
376 	struct scsi_target *starget = scsi_target(current_sdev);
377 	unsigned long flags;
378 
379 	spin_lock_irqsave(shost->host_lock, flags);
380 	starget->starget_sdev_user = NULL;
381 	spin_unlock_irqrestore(shost->host_lock, flags);
382 
383 	/*
384 	 * Call blk_run_queue for all LUNs on the target, starting with
385 	 * current_sdev. We race with others (to set starget_sdev_user),
386 	 * but in most cases, we will be first. Ideally, each LU on the
387 	 * target would get some limited time or requests on the target.
388 	 */
389 	scsi_kick_queue(current_sdev->request_queue);
390 
391 	spin_lock_irqsave(shost->host_lock, flags);
392 	if (starget->starget_sdev_user)
393 		goto out;
394 	list_for_each_entry_safe(sdev, tmp, &starget->devices,
395 			same_target_siblings) {
396 		if (sdev == current_sdev)
397 			continue;
398 		if (scsi_device_get(sdev))
399 			continue;
400 
401 		spin_unlock_irqrestore(shost->host_lock, flags);
402 		scsi_kick_queue(sdev->request_queue);
403 		spin_lock_irqsave(shost->host_lock, flags);
404 
405 		scsi_device_put(sdev);
406 	}
407  out:
408 	spin_unlock_irqrestore(shost->host_lock, flags);
409 }
410 
411 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
412 {
413 	if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
414 		return true;
415 	if (atomic_read(&sdev->device_blocked) > 0)
416 		return true;
417 	return false;
418 }
419 
420 static inline bool scsi_target_is_busy(struct scsi_target *starget)
421 {
422 	if (starget->can_queue > 0) {
423 		if (atomic_read(&starget->target_busy) >= starget->can_queue)
424 			return true;
425 		if (atomic_read(&starget->target_blocked) > 0)
426 			return true;
427 	}
428 	return false;
429 }
430 
431 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
432 {
433 	if (shost->can_queue > 0 &&
434 	    atomic_read(&shost->host_busy) >= shost->can_queue)
435 		return true;
436 	if (atomic_read(&shost->host_blocked) > 0)
437 		return true;
438 	if (shost->host_self_blocked)
439 		return true;
440 	return false;
441 }
442 
443 static void scsi_starved_list_run(struct Scsi_Host *shost)
444 {
445 	LIST_HEAD(starved_list);
446 	struct scsi_device *sdev;
447 	unsigned long flags;
448 
449 	spin_lock_irqsave(shost->host_lock, flags);
450 	list_splice_init(&shost->starved_list, &starved_list);
451 
452 	while (!list_empty(&starved_list)) {
453 		struct request_queue *slq;
454 
455 		/*
456 		 * As long as shost is accepting commands and we have
457 		 * starved queues, call blk_run_queue. scsi_request_fn
458 		 * drops the queue_lock and can add us back to the
459 		 * starved_list.
460 		 *
461 		 * host_lock protects the starved_list and starved_entry.
462 		 * scsi_request_fn must get the host_lock before checking
463 		 * or modifying starved_list or starved_entry.
464 		 */
465 		if (scsi_host_is_busy(shost))
466 			break;
467 
468 		sdev = list_entry(starved_list.next,
469 				  struct scsi_device, starved_entry);
470 		list_del_init(&sdev->starved_entry);
471 		if (scsi_target_is_busy(scsi_target(sdev))) {
472 			list_move_tail(&sdev->starved_entry,
473 				       &shost->starved_list);
474 			continue;
475 		}
476 
477 		/*
478 		 * Once we drop the host lock, a racing scsi_remove_device()
479 		 * call may remove the sdev from the starved list and destroy
480 		 * it and the queue.  Mitigate by taking a reference to the
481 		 * queue and never touching the sdev again after we drop the
482 		 * host lock.  Note: if __scsi_remove_device() invokes
483 		 * blk_cleanup_queue() before the queue is run from this
484 		 * function then blk_run_queue() will return immediately since
485 		 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
486 		 */
487 		slq = sdev->request_queue;
488 		if (!blk_get_queue(slq))
489 			continue;
490 		spin_unlock_irqrestore(shost->host_lock, flags);
491 
492 		scsi_kick_queue(slq);
493 		blk_put_queue(slq);
494 
495 		spin_lock_irqsave(shost->host_lock, flags);
496 	}
497 	/* put any unprocessed entries back */
498 	list_splice(&starved_list, &shost->starved_list);
499 	spin_unlock_irqrestore(shost->host_lock, flags);
500 }
501 
502 /*
503  * Function:   scsi_run_queue()
504  *
505  * Purpose:    Select a proper request queue to serve next
506  *
507  * Arguments:  q       - last request's queue
508  *
509  * Returns:     Nothing
510  *
511  * Notes:      The previous command was completely finished, start
512  *             a new one if possible.
513  */
514 static void scsi_run_queue(struct request_queue *q)
515 {
516 	struct scsi_device *sdev = q->queuedata;
517 
518 	if (scsi_target(sdev)->single_lun)
519 		scsi_single_lun_run(sdev);
520 	if (!list_empty(&sdev->host->starved_list))
521 		scsi_starved_list_run(sdev->host);
522 
523 	blk_mq_run_hw_queues(q, false);
524 }
525 
526 void scsi_requeue_run_queue(struct work_struct *work)
527 {
528 	struct scsi_device *sdev;
529 	struct request_queue *q;
530 
531 	sdev = container_of(work, struct scsi_device, requeue_work);
532 	q = sdev->request_queue;
533 	scsi_run_queue(q);
534 }
535 
536 void scsi_run_host_queues(struct Scsi_Host *shost)
537 {
538 	struct scsi_device *sdev;
539 
540 	shost_for_each_device(sdev, shost)
541 		scsi_run_queue(sdev->request_queue);
542 }
543 
544 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
545 {
546 	if (!blk_rq_is_passthrough(cmd->request)) {
547 		struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
548 
549 		if (drv->uninit_command)
550 			drv->uninit_command(cmd);
551 	}
552 }
553 
554 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
555 {
556 	if (cmd->sdb.table.nents)
557 		sg_free_table_chained(&cmd->sdb.table,
558 				SCSI_INLINE_SG_CNT);
559 	if (scsi_prot_sg_count(cmd))
560 		sg_free_table_chained(&cmd->prot_sdb->table,
561 				SCSI_INLINE_PROT_SG_CNT);
562 }
563 
564 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
565 {
566 	scsi_mq_free_sgtables(cmd);
567 	scsi_uninit_cmd(cmd);
568 	scsi_del_cmd_from_list(cmd);
569 }
570 
571 /* Returns false when no more bytes to process, true if there are more */
572 static bool scsi_end_request(struct request *req, blk_status_t error,
573 		unsigned int bytes)
574 {
575 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
576 	struct scsi_device *sdev = cmd->device;
577 	struct request_queue *q = sdev->request_queue;
578 
579 	if (blk_update_request(req, error, bytes))
580 		return true;
581 
582 	if (blk_queue_add_random(q))
583 		add_disk_randomness(req->rq_disk);
584 
585 	if (!blk_rq_is_scsi(req)) {
586 		WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
587 		cmd->flags &= ~SCMD_INITIALIZED;
588 	}
589 
590 	/*
591 	 * Calling rcu_barrier() is not necessary here because the
592 	 * SCSI error handler guarantees that the function called by
593 	 * call_rcu() has been called before scsi_end_request() is
594 	 * called.
595 	 */
596 	destroy_rcu_head(&cmd->rcu);
597 
598 	/*
599 	 * In the MQ case the command gets freed by __blk_mq_end_request,
600 	 * so we have to do all cleanup that depends on it earlier.
601 	 *
602 	 * We also can't kick the queues from irq context, so we
603 	 * will have to defer it to a workqueue.
604 	 */
605 	scsi_mq_uninit_cmd(cmd);
606 
607 	/*
608 	 * queue is still alive, so grab the ref for preventing it
609 	 * from being cleaned up during running queue.
610 	 */
611 	percpu_ref_get(&q->q_usage_counter);
612 
613 	__blk_mq_end_request(req, error);
614 
615 	if (scsi_target(sdev)->single_lun ||
616 	    !list_empty(&sdev->host->starved_list))
617 		kblockd_schedule_work(&sdev->requeue_work);
618 	else
619 		blk_mq_run_hw_queues(q, true);
620 
621 	percpu_ref_put(&q->q_usage_counter);
622 	return false;
623 }
624 
625 /**
626  * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
627  * @cmd:	SCSI command
628  * @result:	scsi error code
629  *
630  * Translate a SCSI result code into a blk_status_t value. May reset the host
631  * byte of @cmd->result.
632  */
633 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
634 {
635 	switch (host_byte(result)) {
636 	case DID_OK:
637 		/*
638 		 * Also check the other bytes than the status byte in result
639 		 * to handle the case when a SCSI LLD sets result to
640 		 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
641 		 */
642 		if (scsi_status_is_good(result) && (result & ~0xff) == 0)
643 			return BLK_STS_OK;
644 		return BLK_STS_IOERR;
645 	case DID_TRANSPORT_FAILFAST:
646 		return BLK_STS_TRANSPORT;
647 	case DID_TARGET_FAILURE:
648 		set_host_byte(cmd, DID_OK);
649 		return BLK_STS_TARGET;
650 	case DID_NEXUS_FAILURE:
651 		set_host_byte(cmd, DID_OK);
652 		return BLK_STS_NEXUS;
653 	case DID_ALLOC_FAILURE:
654 		set_host_byte(cmd, DID_OK);
655 		return BLK_STS_NOSPC;
656 	case DID_MEDIUM_ERROR:
657 		set_host_byte(cmd, DID_OK);
658 		return BLK_STS_MEDIUM;
659 	default:
660 		return BLK_STS_IOERR;
661 	}
662 }
663 
664 /* Helper for scsi_io_completion() when "reprep" action required. */
665 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
666 				      struct request_queue *q)
667 {
668 	/* A new command will be prepared and issued. */
669 	scsi_mq_requeue_cmd(cmd);
670 }
671 
672 /* Helper for scsi_io_completion() when special action required. */
673 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
674 {
675 	struct request_queue *q = cmd->device->request_queue;
676 	struct request *req = cmd->request;
677 	int level = 0;
678 	enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
679 	      ACTION_DELAYED_RETRY} action;
680 	unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
681 	struct scsi_sense_hdr sshdr;
682 	bool sense_valid;
683 	bool sense_current = true;      /* false implies "deferred sense" */
684 	blk_status_t blk_stat;
685 
686 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
687 	if (sense_valid)
688 		sense_current = !scsi_sense_is_deferred(&sshdr);
689 
690 	blk_stat = scsi_result_to_blk_status(cmd, result);
691 
692 	if (host_byte(result) == DID_RESET) {
693 		/* Third party bus reset or reset for error recovery
694 		 * reasons.  Just retry the command and see what
695 		 * happens.
696 		 */
697 		action = ACTION_RETRY;
698 	} else if (sense_valid && sense_current) {
699 		switch (sshdr.sense_key) {
700 		case UNIT_ATTENTION:
701 			if (cmd->device->removable) {
702 				/* Detected disc change.  Set a bit
703 				 * and quietly refuse further access.
704 				 */
705 				cmd->device->changed = 1;
706 				action = ACTION_FAIL;
707 			} else {
708 				/* Must have been a power glitch, or a
709 				 * bus reset.  Could not have been a
710 				 * media change, so we just retry the
711 				 * command and see what happens.
712 				 */
713 				action = ACTION_RETRY;
714 			}
715 			break;
716 		case ILLEGAL_REQUEST:
717 			/* If we had an ILLEGAL REQUEST returned, then
718 			 * we may have performed an unsupported
719 			 * command.  The only thing this should be
720 			 * would be a ten byte read where only a six
721 			 * byte read was supported.  Also, on a system
722 			 * where READ CAPACITY failed, we may have
723 			 * read past the end of the disk.
724 			 */
725 			if ((cmd->device->use_10_for_rw &&
726 			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
727 			    (cmd->cmnd[0] == READ_10 ||
728 			     cmd->cmnd[0] == WRITE_10)) {
729 				/* This will issue a new 6-byte command. */
730 				cmd->device->use_10_for_rw = 0;
731 				action = ACTION_REPREP;
732 			} else if (sshdr.asc == 0x10) /* DIX */ {
733 				action = ACTION_FAIL;
734 				blk_stat = BLK_STS_PROTECTION;
735 			/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
736 			} else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
737 				action = ACTION_FAIL;
738 				blk_stat = BLK_STS_TARGET;
739 			} else
740 				action = ACTION_FAIL;
741 			break;
742 		case ABORTED_COMMAND:
743 			action = ACTION_FAIL;
744 			if (sshdr.asc == 0x10) /* DIF */
745 				blk_stat = BLK_STS_PROTECTION;
746 			break;
747 		case NOT_READY:
748 			/* If the device is in the process of becoming
749 			 * ready, or has a temporary blockage, retry.
750 			 */
751 			if (sshdr.asc == 0x04) {
752 				switch (sshdr.ascq) {
753 				case 0x01: /* becoming ready */
754 				case 0x04: /* format in progress */
755 				case 0x05: /* rebuild in progress */
756 				case 0x06: /* recalculation in progress */
757 				case 0x07: /* operation in progress */
758 				case 0x08: /* Long write in progress */
759 				case 0x09: /* self test in progress */
760 				case 0x14: /* space allocation in progress */
761 				case 0x1a: /* start stop unit in progress */
762 				case 0x1b: /* sanitize in progress */
763 				case 0x1d: /* configuration in progress */
764 				case 0x24: /* depopulation in progress */
765 					action = ACTION_DELAYED_RETRY;
766 					break;
767 				default:
768 					action = ACTION_FAIL;
769 					break;
770 				}
771 			} else
772 				action = ACTION_FAIL;
773 			break;
774 		case VOLUME_OVERFLOW:
775 			/* See SSC3rXX or current. */
776 			action = ACTION_FAIL;
777 			break;
778 		default:
779 			action = ACTION_FAIL;
780 			break;
781 		}
782 	} else
783 		action = ACTION_FAIL;
784 
785 	if (action != ACTION_FAIL &&
786 	    time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
787 		action = ACTION_FAIL;
788 
789 	switch (action) {
790 	case ACTION_FAIL:
791 		/* Give up and fail the remainder of the request */
792 		if (!(req->rq_flags & RQF_QUIET)) {
793 			static DEFINE_RATELIMIT_STATE(_rs,
794 					DEFAULT_RATELIMIT_INTERVAL,
795 					DEFAULT_RATELIMIT_BURST);
796 
797 			if (unlikely(scsi_logging_level))
798 				level =
799 				     SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
800 						    SCSI_LOG_MLCOMPLETE_BITS);
801 
802 			/*
803 			 * if logging is enabled the failure will be printed
804 			 * in scsi_log_completion(), so avoid duplicate messages
805 			 */
806 			if (!level && __ratelimit(&_rs)) {
807 				scsi_print_result(cmd, NULL, FAILED);
808 				if (driver_byte(result) == DRIVER_SENSE)
809 					scsi_print_sense(cmd);
810 				scsi_print_command(cmd);
811 			}
812 		}
813 		if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
814 			return;
815 		/*FALLTHRU*/
816 	case ACTION_REPREP:
817 		scsi_io_completion_reprep(cmd, q);
818 		break;
819 	case ACTION_RETRY:
820 		/* Retry the same command immediately */
821 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
822 		break;
823 	case ACTION_DELAYED_RETRY:
824 		/* Retry the same command after a delay */
825 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
826 		break;
827 	}
828 }
829 
830 /*
831  * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
832  * new result that may suppress further error checking. Also modifies
833  * *blk_statp in some cases.
834  */
835 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
836 					blk_status_t *blk_statp)
837 {
838 	bool sense_valid;
839 	bool sense_current = true;	/* false implies "deferred sense" */
840 	struct request *req = cmd->request;
841 	struct scsi_sense_hdr sshdr;
842 
843 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
844 	if (sense_valid)
845 		sense_current = !scsi_sense_is_deferred(&sshdr);
846 
847 	if (blk_rq_is_passthrough(req)) {
848 		if (sense_valid) {
849 			/*
850 			 * SG_IO wants current and deferred errors
851 			 */
852 			scsi_req(req)->sense_len =
853 				min(8 + cmd->sense_buffer[7],
854 				    SCSI_SENSE_BUFFERSIZE);
855 		}
856 		if (sense_current)
857 			*blk_statp = scsi_result_to_blk_status(cmd, result);
858 	} else if (blk_rq_bytes(req) == 0 && sense_current) {
859 		/*
860 		 * Flush commands do not transfers any data, and thus cannot use
861 		 * good_bytes != blk_rq_bytes(req) as the signal for an error.
862 		 * This sets *blk_statp explicitly for the problem case.
863 		 */
864 		*blk_statp = scsi_result_to_blk_status(cmd, result);
865 	}
866 	/*
867 	 * Recovered errors need reporting, but they're always treated as
868 	 * success, so fiddle the result code here.  For passthrough requests
869 	 * we already took a copy of the original into sreq->result which
870 	 * is what gets returned to the user
871 	 */
872 	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
873 		bool do_print = true;
874 		/*
875 		 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
876 		 * skip print since caller wants ATA registers. Only occurs
877 		 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
878 		 */
879 		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
880 			do_print = false;
881 		else if (req->rq_flags & RQF_QUIET)
882 			do_print = false;
883 		if (do_print)
884 			scsi_print_sense(cmd);
885 		result = 0;
886 		/* for passthrough, *blk_statp may be set */
887 		*blk_statp = BLK_STS_OK;
888 	}
889 	/*
890 	 * Another corner case: the SCSI status byte is non-zero but 'good'.
891 	 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
892 	 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
893 	 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
894 	 * intermediate statuses (both obsolete in SAM-4) as good.
895 	 */
896 	if (status_byte(result) && scsi_status_is_good(result)) {
897 		result = 0;
898 		*blk_statp = BLK_STS_OK;
899 	}
900 	return result;
901 }
902 
903 /*
904  * Function:    scsi_io_completion()
905  *
906  * Purpose:     Completion processing for block device I/O requests.
907  *
908  * Arguments:   cmd   - command that is finished.
909  *
910  * Lock status: Assumed that no lock is held upon entry.
911  *
912  * Returns:     Nothing
913  *
914  * Notes:       We will finish off the specified number of sectors.  If we
915  *		are done, the command block will be released and the queue
916  *		function will be goosed.  If we are not done then we have to
917  *		figure out what to do next:
918  *
919  *		a) We can call scsi_requeue_command().  The request
920  *		   will be unprepared and put back on the queue.  Then
921  *		   a new command will be created for it.  This should
922  *		   be used if we made forward progress, or if we want
923  *		   to switch from READ(10) to READ(6) for example.
924  *
925  *		b) We can call __scsi_queue_insert().  The request will
926  *		   be put back on the queue and retried using the same
927  *		   command as before, possibly after a delay.
928  *
929  *		c) We can call scsi_end_request() with blk_stat other than
930  *		   BLK_STS_OK, to fail the remainder of the request.
931  */
932 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
933 {
934 	int result = cmd->result;
935 	struct request_queue *q = cmd->device->request_queue;
936 	struct request *req = cmd->request;
937 	blk_status_t blk_stat = BLK_STS_OK;
938 
939 	if (unlikely(result))	/* a nz result may or may not be an error */
940 		result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
941 
942 	if (unlikely(blk_rq_is_passthrough(req))) {
943 		/*
944 		 * scsi_result_to_blk_status may have reset the host_byte
945 		 */
946 		scsi_req(req)->result = cmd->result;
947 	}
948 
949 	/*
950 	 * Next deal with any sectors which we were able to correctly
951 	 * handle.
952 	 */
953 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
954 		"%u sectors total, %d bytes done.\n",
955 		blk_rq_sectors(req), good_bytes));
956 
957 	/*
958 	 * Next deal with any sectors which we were able to correctly
959 	 * handle. Failed, zero length commands always need to drop down
960 	 * to retry code. Fast path should return in this block.
961 	 */
962 	if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
963 		if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
964 			return; /* no bytes remaining */
965 	}
966 
967 	/* Kill remainder if no retries. */
968 	if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
969 		if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
970 			WARN_ONCE(true,
971 			    "Bytes remaining after failed, no-retry command");
972 		return;
973 	}
974 
975 	/*
976 	 * If there had been no error, but we have leftover bytes in the
977 	 * requeues just queue the command up again.
978 	 */
979 	if (likely(result == 0))
980 		scsi_io_completion_reprep(cmd, q);
981 	else
982 		scsi_io_completion_action(cmd, result);
983 }
984 
985 static blk_status_t scsi_init_sgtable(struct request *req,
986 		struct scsi_data_buffer *sdb)
987 {
988 	int count;
989 
990 	/*
991 	 * If sg table allocation fails, requeue request later.
992 	 */
993 	if (unlikely(sg_alloc_table_chained(&sdb->table,
994 			blk_rq_nr_phys_segments(req), sdb->table.sgl,
995 			SCSI_INLINE_SG_CNT)))
996 		return BLK_STS_RESOURCE;
997 
998 	/*
999 	 * Next, walk the list, and fill in the addresses and sizes of
1000 	 * each segment.
1001 	 */
1002 	count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1003 	BUG_ON(count > sdb->table.nents);
1004 	sdb->table.nents = count;
1005 	sdb->length = blk_rq_payload_bytes(req);
1006 	return BLK_STS_OK;
1007 }
1008 
1009 /*
1010  * Function:    scsi_init_io()
1011  *
1012  * Purpose:     SCSI I/O initialize function.
1013  *
1014  * Arguments:   cmd   - Command descriptor we wish to initialize
1015  *
1016  * Returns:     BLK_STS_OK on success
1017  *		BLK_STS_RESOURCE if the failure is retryable
1018  *		BLK_STS_IOERR if the failure is fatal
1019  */
1020 blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
1021 {
1022 	struct request *rq = cmd->request;
1023 	blk_status_t ret;
1024 
1025 	if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1026 		return BLK_STS_IOERR;
1027 
1028 	ret = scsi_init_sgtable(rq, &cmd->sdb);
1029 	if (ret)
1030 		return ret;
1031 
1032 	if (blk_integrity_rq(rq)) {
1033 		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1034 		int ivecs, count;
1035 
1036 		if (WARN_ON_ONCE(!prot_sdb)) {
1037 			/*
1038 			 * This can happen if someone (e.g. multipath)
1039 			 * queues a command to a device on an adapter
1040 			 * that does not support DIX.
1041 			 */
1042 			ret = BLK_STS_IOERR;
1043 			goto out_free_sgtables;
1044 		}
1045 
1046 		ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1047 
1048 		if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1049 				prot_sdb->table.sgl,
1050 				SCSI_INLINE_PROT_SG_CNT)) {
1051 			ret = BLK_STS_RESOURCE;
1052 			goto out_free_sgtables;
1053 		}
1054 
1055 		count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1056 						prot_sdb->table.sgl);
1057 		BUG_ON(count > ivecs);
1058 		BUG_ON(count > queue_max_integrity_segments(rq->q));
1059 
1060 		cmd->prot_sdb = prot_sdb;
1061 		cmd->prot_sdb->table.nents = count;
1062 	}
1063 
1064 	return BLK_STS_OK;
1065 out_free_sgtables:
1066 	scsi_mq_free_sgtables(cmd);
1067 	return ret;
1068 }
1069 EXPORT_SYMBOL(scsi_init_io);
1070 
1071 /**
1072  * scsi_initialize_rq - initialize struct scsi_cmnd partially
1073  * @rq: Request associated with the SCSI command to be initialized.
1074  *
1075  * This function initializes the members of struct scsi_cmnd that must be
1076  * initialized before request processing starts and that won't be
1077  * reinitialized if a SCSI command is requeued.
1078  *
1079  * Called from inside blk_get_request() for pass-through requests and from
1080  * inside scsi_init_command() for filesystem requests.
1081  */
1082 static void scsi_initialize_rq(struct request *rq)
1083 {
1084 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1085 
1086 	scsi_req_init(&cmd->req);
1087 	init_rcu_head(&cmd->rcu);
1088 	cmd->jiffies_at_alloc = jiffies;
1089 	cmd->retries = 0;
1090 }
1091 
1092 /*
1093  * Only called when the request isn't completed by SCSI, and not freed by
1094  * SCSI
1095  */
1096 static void scsi_cleanup_rq(struct request *rq)
1097 {
1098 	if (rq->rq_flags & RQF_DONTPREP) {
1099 		scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1100 		rq->rq_flags &= ~RQF_DONTPREP;
1101 	}
1102 }
1103 
1104 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1105 void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1106 {
1107 	struct scsi_device *sdev = cmd->device;
1108 	struct Scsi_Host *shost = sdev->host;
1109 	unsigned long flags;
1110 
1111 	if (shost->use_cmd_list) {
1112 		spin_lock_irqsave(&sdev->list_lock, flags);
1113 		list_add_tail(&cmd->list, &sdev->cmd_list);
1114 		spin_unlock_irqrestore(&sdev->list_lock, flags);
1115 	}
1116 }
1117 
1118 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1119 void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1120 {
1121 	struct scsi_device *sdev = cmd->device;
1122 	struct Scsi_Host *shost = sdev->host;
1123 	unsigned long flags;
1124 
1125 	if (shost->use_cmd_list) {
1126 		spin_lock_irqsave(&sdev->list_lock, flags);
1127 		BUG_ON(list_empty(&cmd->list));
1128 		list_del_init(&cmd->list);
1129 		spin_unlock_irqrestore(&sdev->list_lock, flags);
1130 	}
1131 }
1132 
1133 /* Called after a request has been started. */
1134 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1135 {
1136 	void *buf = cmd->sense_buffer;
1137 	void *prot = cmd->prot_sdb;
1138 	struct request *rq = blk_mq_rq_from_pdu(cmd);
1139 	unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1140 	unsigned long jiffies_at_alloc;
1141 	int retries;
1142 
1143 	if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1144 		flags |= SCMD_INITIALIZED;
1145 		scsi_initialize_rq(rq);
1146 	}
1147 
1148 	jiffies_at_alloc = cmd->jiffies_at_alloc;
1149 	retries = cmd->retries;
1150 	/* zero out the cmd, except for the embedded scsi_request */
1151 	memset((char *)cmd + sizeof(cmd->req), 0,
1152 		sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1153 
1154 	cmd->device = dev;
1155 	cmd->sense_buffer = buf;
1156 	cmd->prot_sdb = prot;
1157 	cmd->flags = flags;
1158 	INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1159 	cmd->jiffies_at_alloc = jiffies_at_alloc;
1160 	cmd->retries = retries;
1161 
1162 	scsi_add_cmd_to_list(cmd);
1163 }
1164 
1165 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1166 		struct request *req)
1167 {
1168 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1169 
1170 	/*
1171 	 * Passthrough requests may transfer data, in which case they must
1172 	 * a bio attached to them.  Or they might contain a SCSI command
1173 	 * that does not transfer data, in which case they may optionally
1174 	 * submit a request without an attached bio.
1175 	 */
1176 	if (req->bio) {
1177 		blk_status_t ret = scsi_init_io(cmd);
1178 		if (unlikely(ret != BLK_STS_OK))
1179 			return ret;
1180 	} else {
1181 		BUG_ON(blk_rq_bytes(req));
1182 
1183 		memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1184 	}
1185 
1186 	cmd->cmd_len = scsi_req(req)->cmd_len;
1187 	cmd->cmnd = scsi_req(req)->cmd;
1188 	cmd->transfersize = blk_rq_bytes(req);
1189 	cmd->allowed = scsi_req(req)->retries;
1190 	return BLK_STS_OK;
1191 }
1192 
1193 /*
1194  * Setup a normal block command.  These are simple request from filesystems
1195  * that still need to be translated to SCSI CDBs from the ULD.
1196  */
1197 static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1198 		struct request *req)
1199 {
1200 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1201 
1202 	if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1203 		blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1204 		if (ret != BLK_STS_OK)
1205 			return ret;
1206 	}
1207 
1208 	cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1209 	memset(cmd->cmnd, 0, BLK_MAX_CDB);
1210 	return scsi_cmd_to_driver(cmd)->init_command(cmd);
1211 }
1212 
1213 static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1214 		struct request *req)
1215 {
1216 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1217 
1218 	if (!blk_rq_bytes(req))
1219 		cmd->sc_data_direction = DMA_NONE;
1220 	else if (rq_data_dir(req) == WRITE)
1221 		cmd->sc_data_direction = DMA_TO_DEVICE;
1222 	else
1223 		cmd->sc_data_direction = DMA_FROM_DEVICE;
1224 
1225 	if (blk_rq_is_scsi(req))
1226 		return scsi_setup_scsi_cmnd(sdev, req);
1227 	else
1228 		return scsi_setup_fs_cmnd(sdev, req);
1229 }
1230 
1231 static blk_status_t
1232 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1233 {
1234 	switch (sdev->sdev_state) {
1235 	case SDEV_OFFLINE:
1236 	case SDEV_TRANSPORT_OFFLINE:
1237 		/*
1238 		 * If the device is offline we refuse to process any
1239 		 * commands.  The device must be brought online
1240 		 * before trying any recovery commands.
1241 		 */
1242 		sdev_printk(KERN_ERR, sdev,
1243 			    "rejecting I/O to offline device\n");
1244 		return BLK_STS_IOERR;
1245 	case SDEV_DEL:
1246 		/*
1247 		 * If the device is fully deleted, we refuse to
1248 		 * process any commands as well.
1249 		 */
1250 		sdev_printk(KERN_ERR, sdev,
1251 			    "rejecting I/O to dead device\n");
1252 		return BLK_STS_IOERR;
1253 	case SDEV_BLOCK:
1254 	case SDEV_CREATED_BLOCK:
1255 		return BLK_STS_RESOURCE;
1256 	case SDEV_QUIESCE:
1257 		/*
1258 		 * If the devices is blocked we defer normal commands.
1259 		 */
1260 		if (req && !(req->rq_flags & RQF_PREEMPT))
1261 			return BLK_STS_RESOURCE;
1262 		return BLK_STS_OK;
1263 	default:
1264 		/*
1265 		 * For any other not fully online state we only allow
1266 		 * special commands.  In particular any user initiated
1267 		 * command is not allowed.
1268 		 */
1269 		if (req && !(req->rq_flags & RQF_PREEMPT))
1270 			return BLK_STS_IOERR;
1271 		return BLK_STS_OK;
1272 	}
1273 }
1274 
1275 /*
1276  * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1277  * return 0.
1278  *
1279  * Called with the queue_lock held.
1280  */
1281 static inline int scsi_dev_queue_ready(struct request_queue *q,
1282 				  struct scsi_device *sdev)
1283 {
1284 	unsigned int busy;
1285 
1286 	busy = atomic_inc_return(&sdev->device_busy) - 1;
1287 	if (atomic_read(&sdev->device_blocked)) {
1288 		if (busy)
1289 			goto out_dec;
1290 
1291 		/*
1292 		 * unblock after device_blocked iterates to zero
1293 		 */
1294 		if (atomic_dec_return(&sdev->device_blocked) > 0)
1295 			goto out_dec;
1296 		SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1297 				   "unblocking device at zero depth\n"));
1298 	}
1299 
1300 	if (busy >= sdev->queue_depth)
1301 		goto out_dec;
1302 
1303 	return 1;
1304 out_dec:
1305 	atomic_dec(&sdev->device_busy);
1306 	return 0;
1307 }
1308 
1309 /*
1310  * scsi_target_queue_ready: checks if there we can send commands to target
1311  * @sdev: scsi device on starget to check.
1312  */
1313 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1314 					   struct scsi_device *sdev)
1315 {
1316 	struct scsi_target *starget = scsi_target(sdev);
1317 	unsigned int busy;
1318 
1319 	if (starget->single_lun) {
1320 		spin_lock_irq(shost->host_lock);
1321 		if (starget->starget_sdev_user &&
1322 		    starget->starget_sdev_user != sdev) {
1323 			spin_unlock_irq(shost->host_lock);
1324 			return 0;
1325 		}
1326 		starget->starget_sdev_user = sdev;
1327 		spin_unlock_irq(shost->host_lock);
1328 	}
1329 
1330 	if (starget->can_queue <= 0)
1331 		return 1;
1332 
1333 	busy = atomic_inc_return(&starget->target_busy) - 1;
1334 	if (atomic_read(&starget->target_blocked) > 0) {
1335 		if (busy)
1336 			goto starved;
1337 
1338 		/*
1339 		 * unblock after target_blocked iterates to zero
1340 		 */
1341 		if (atomic_dec_return(&starget->target_blocked) > 0)
1342 			goto out_dec;
1343 
1344 		SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1345 				 "unblocking target at zero depth\n"));
1346 	}
1347 
1348 	if (busy >= starget->can_queue)
1349 		goto starved;
1350 
1351 	return 1;
1352 
1353 starved:
1354 	spin_lock_irq(shost->host_lock);
1355 	list_move_tail(&sdev->starved_entry, &shost->starved_list);
1356 	spin_unlock_irq(shost->host_lock);
1357 out_dec:
1358 	if (starget->can_queue > 0)
1359 		atomic_dec(&starget->target_busy);
1360 	return 0;
1361 }
1362 
1363 /*
1364  * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1365  * return 0. We must end up running the queue again whenever 0 is
1366  * returned, else IO can hang.
1367  */
1368 static inline int scsi_host_queue_ready(struct request_queue *q,
1369 				   struct Scsi_Host *shost,
1370 				   struct scsi_device *sdev)
1371 {
1372 	unsigned int busy;
1373 
1374 	if (scsi_host_in_recovery(shost))
1375 		return 0;
1376 
1377 	busy = atomic_inc_return(&shost->host_busy) - 1;
1378 	if (atomic_read(&shost->host_blocked) > 0) {
1379 		if (busy)
1380 			goto starved;
1381 
1382 		/*
1383 		 * unblock after host_blocked iterates to zero
1384 		 */
1385 		if (atomic_dec_return(&shost->host_blocked) > 0)
1386 			goto out_dec;
1387 
1388 		SCSI_LOG_MLQUEUE(3,
1389 			shost_printk(KERN_INFO, shost,
1390 				     "unblocking host at zero depth\n"));
1391 	}
1392 
1393 	if (shost->can_queue > 0 && busy >= shost->can_queue)
1394 		goto starved;
1395 	if (shost->host_self_blocked)
1396 		goto starved;
1397 
1398 	/* We're OK to process the command, so we can't be starved */
1399 	if (!list_empty(&sdev->starved_entry)) {
1400 		spin_lock_irq(shost->host_lock);
1401 		if (!list_empty(&sdev->starved_entry))
1402 			list_del_init(&sdev->starved_entry);
1403 		spin_unlock_irq(shost->host_lock);
1404 	}
1405 
1406 	return 1;
1407 
1408 starved:
1409 	spin_lock_irq(shost->host_lock);
1410 	if (list_empty(&sdev->starved_entry))
1411 		list_add_tail(&sdev->starved_entry, &shost->starved_list);
1412 	spin_unlock_irq(shost->host_lock);
1413 out_dec:
1414 	scsi_dec_host_busy(shost);
1415 	return 0;
1416 }
1417 
1418 /*
1419  * Busy state exporting function for request stacking drivers.
1420  *
1421  * For efficiency, no lock is taken to check the busy state of
1422  * shost/starget/sdev, since the returned value is not guaranteed and
1423  * may be changed after request stacking drivers call the function,
1424  * regardless of taking lock or not.
1425  *
1426  * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1427  * needs to return 'not busy'. Otherwise, request stacking drivers
1428  * may hold requests forever.
1429  */
1430 static bool scsi_mq_lld_busy(struct request_queue *q)
1431 {
1432 	struct scsi_device *sdev = q->queuedata;
1433 	struct Scsi_Host *shost;
1434 
1435 	if (blk_queue_dying(q))
1436 		return false;
1437 
1438 	shost = sdev->host;
1439 
1440 	/*
1441 	 * Ignore host/starget busy state.
1442 	 * Since block layer does not have a concept of fairness across
1443 	 * multiple queues, congestion of host/starget needs to be handled
1444 	 * in SCSI layer.
1445 	 */
1446 	if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1447 		return true;
1448 
1449 	return false;
1450 }
1451 
1452 static void scsi_softirq_done(struct request *rq)
1453 {
1454 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1455 	unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1456 	int disposition;
1457 
1458 	INIT_LIST_HEAD(&cmd->eh_entry);
1459 
1460 	atomic_inc(&cmd->device->iodone_cnt);
1461 	if (cmd->result)
1462 		atomic_inc(&cmd->device->ioerr_cnt);
1463 
1464 	disposition = scsi_decide_disposition(cmd);
1465 	if (disposition != SUCCESS &&
1466 	    time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1467 		scmd_printk(KERN_ERR, cmd,
1468 			    "timing out command, waited %lus\n",
1469 			    wait_for/HZ);
1470 		disposition = SUCCESS;
1471 	}
1472 
1473 	scsi_log_completion(cmd, disposition);
1474 
1475 	switch (disposition) {
1476 		case SUCCESS:
1477 			scsi_finish_command(cmd);
1478 			break;
1479 		case NEEDS_RETRY:
1480 			scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1481 			break;
1482 		case ADD_TO_MLQUEUE:
1483 			scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1484 			break;
1485 		default:
1486 			scsi_eh_scmd_add(cmd);
1487 			break;
1488 	}
1489 }
1490 
1491 /**
1492  * scsi_dispatch_command - Dispatch a command to the low-level driver.
1493  * @cmd: command block we are dispatching.
1494  *
1495  * Return: nonzero return request was rejected and device's queue needs to be
1496  * plugged.
1497  */
1498 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1499 {
1500 	struct Scsi_Host *host = cmd->device->host;
1501 	int rtn = 0;
1502 
1503 	atomic_inc(&cmd->device->iorequest_cnt);
1504 
1505 	/* check if the device is still usable */
1506 	if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1507 		/* in SDEV_DEL we error all commands. DID_NO_CONNECT
1508 		 * returns an immediate error upwards, and signals
1509 		 * that the device is no longer present */
1510 		cmd->result = DID_NO_CONNECT << 16;
1511 		goto done;
1512 	}
1513 
1514 	/* Check to see if the scsi lld made this device blocked. */
1515 	if (unlikely(scsi_device_blocked(cmd->device))) {
1516 		/*
1517 		 * in blocked state, the command is just put back on
1518 		 * the device queue.  The suspend state has already
1519 		 * blocked the queue so future requests should not
1520 		 * occur until the device transitions out of the
1521 		 * suspend state.
1522 		 */
1523 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1524 			"queuecommand : device blocked\n"));
1525 		return SCSI_MLQUEUE_DEVICE_BUSY;
1526 	}
1527 
1528 	/* Store the LUN value in cmnd, if needed. */
1529 	if (cmd->device->lun_in_cdb)
1530 		cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1531 			       (cmd->device->lun << 5 & 0xe0);
1532 
1533 	scsi_log_send(cmd);
1534 
1535 	/*
1536 	 * Before we queue this command, check if the command
1537 	 * length exceeds what the host adapter can handle.
1538 	 */
1539 	if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1540 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1541 			       "queuecommand : command too long. "
1542 			       "cdb_size=%d host->max_cmd_len=%d\n",
1543 			       cmd->cmd_len, cmd->device->host->max_cmd_len));
1544 		cmd->result = (DID_ABORT << 16);
1545 		goto done;
1546 	}
1547 
1548 	if (unlikely(host->shost_state == SHOST_DEL)) {
1549 		cmd->result = (DID_NO_CONNECT << 16);
1550 		goto done;
1551 
1552 	}
1553 
1554 	trace_scsi_dispatch_cmd_start(cmd);
1555 	rtn = host->hostt->queuecommand(host, cmd);
1556 	if (rtn) {
1557 		trace_scsi_dispatch_cmd_error(cmd, rtn);
1558 		if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1559 		    rtn != SCSI_MLQUEUE_TARGET_BUSY)
1560 			rtn = SCSI_MLQUEUE_HOST_BUSY;
1561 
1562 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1563 			"queuecommand : request rejected\n"));
1564 	}
1565 
1566 	return rtn;
1567  done:
1568 	cmd->scsi_done(cmd);
1569 	return 0;
1570 }
1571 
1572 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1573 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1574 {
1575 	return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1576 		sizeof(struct scatterlist);
1577 }
1578 
1579 static blk_status_t scsi_mq_prep_fn(struct request *req)
1580 {
1581 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1582 	struct scsi_device *sdev = req->q->queuedata;
1583 	struct Scsi_Host *shost = sdev->host;
1584 	struct scatterlist *sg;
1585 
1586 	scsi_init_command(sdev, cmd);
1587 
1588 	cmd->request = req;
1589 	cmd->tag = req->tag;
1590 	cmd->prot_op = SCSI_PROT_NORMAL;
1591 
1592 	sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1593 	cmd->sdb.table.sgl = sg;
1594 
1595 	if (scsi_host_get_prot(shost)) {
1596 		memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1597 
1598 		cmd->prot_sdb->table.sgl =
1599 			(struct scatterlist *)(cmd->prot_sdb + 1);
1600 	}
1601 
1602 	blk_mq_start_request(req);
1603 
1604 	return scsi_setup_cmnd(sdev, req);
1605 }
1606 
1607 static void scsi_mq_done(struct scsi_cmnd *cmd)
1608 {
1609 	if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1610 		return;
1611 	trace_scsi_dispatch_cmd_done(cmd);
1612 
1613 	/*
1614 	 * If the block layer didn't complete the request due to a timeout
1615 	 * injection, scsi must clear its internal completed state so that the
1616 	 * timeout handler will see it needs to escalate its own error
1617 	 * recovery.
1618 	 */
1619 	if (unlikely(!blk_mq_complete_request(cmd->request)))
1620 		clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1621 }
1622 
1623 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
1624 {
1625 	struct request_queue *q = hctx->queue;
1626 	struct scsi_device *sdev = q->queuedata;
1627 
1628 	atomic_dec(&sdev->device_busy);
1629 }
1630 
1631 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
1632 {
1633 	struct request_queue *q = hctx->queue;
1634 	struct scsi_device *sdev = q->queuedata;
1635 
1636 	if (scsi_dev_queue_ready(q, sdev))
1637 		return true;
1638 
1639 	if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
1640 		blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
1641 	return false;
1642 }
1643 
1644 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1645 			 const struct blk_mq_queue_data *bd)
1646 {
1647 	struct request *req = bd->rq;
1648 	struct request_queue *q = req->q;
1649 	struct scsi_device *sdev = q->queuedata;
1650 	struct Scsi_Host *shost = sdev->host;
1651 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1652 	blk_status_t ret;
1653 	int reason;
1654 
1655 	/*
1656 	 * If the device is not in running state we will reject some or all
1657 	 * commands.
1658 	 */
1659 	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1660 		ret = scsi_prep_state_check(sdev, req);
1661 		if (ret != BLK_STS_OK)
1662 			goto out_put_budget;
1663 	}
1664 
1665 	ret = BLK_STS_RESOURCE;
1666 	if (!scsi_target_queue_ready(shost, sdev))
1667 		goto out_put_budget;
1668 	if (!scsi_host_queue_ready(q, shost, sdev))
1669 		goto out_dec_target_busy;
1670 
1671 	if (!(req->rq_flags & RQF_DONTPREP)) {
1672 		ret = scsi_mq_prep_fn(req);
1673 		if (ret != BLK_STS_OK)
1674 			goto out_dec_host_busy;
1675 		req->rq_flags |= RQF_DONTPREP;
1676 	} else {
1677 		clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1678 		blk_mq_start_request(req);
1679 	}
1680 
1681 	cmd->flags &= SCMD_PRESERVED_FLAGS;
1682 	if (sdev->simple_tags)
1683 		cmd->flags |= SCMD_TAGGED;
1684 	if (bd->last)
1685 		cmd->flags |= SCMD_LAST;
1686 
1687 	scsi_init_cmd_errh(cmd);
1688 	cmd->scsi_done = scsi_mq_done;
1689 
1690 	reason = scsi_dispatch_cmd(cmd);
1691 	if (reason) {
1692 		scsi_set_blocked(cmd, reason);
1693 		ret = BLK_STS_RESOURCE;
1694 		goto out_dec_host_busy;
1695 	}
1696 
1697 	return BLK_STS_OK;
1698 
1699 out_dec_host_busy:
1700 	scsi_dec_host_busy(shost);
1701 out_dec_target_busy:
1702 	if (scsi_target(sdev)->can_queue > 0)
1703 		atomic_dec(&scsi_target(sdev)->target_busy);
1704 out_put_budget:
1705 	scsi_mq_put_budget(hctx);
1706 	switch (ret) {
1707 	case BLK_STS_OK:
1708 		break;
1709 	case BLK_STS_RESOURCE:
1710 		if (atomic_read(&sdev->device_busy) ||
1711 		    scsi_device_blocked(sdev))
1712 			ret = BLK_STS_DEV_RESOURCE;
1713 		break;
1714 	default:
1715 		if (unlikely(!scsi_device_online(sdev)))
1716 			scsi_req(req)->result = DID_NO_CONNECT << 16;
1717 		else
1718 			scsi_req(req)->result = DID_ERROR << 16;
1719 		/*
1720 		 * Make sure to release all allocated resources when
1721 		 * we hit an error, as we will never see this command
1722 		 * again.
1723 		 */
1724 		if (req->rq_flags & RQF_DONTPREP)
1725 			scsi_mq_uninit_cmd(cmd);
1726 		break;
1727 	}
1728 	return ret;
1729 }
1730 
1731 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1732 		bool reserved)
1733 {
1734 	if (reserved)
1735 		return BLK_EH_RESET_TIMER;
1736 	return scsi_times_out(req);
1737 }
1738 
1739 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1740 				unsigned int hctx_idx, unsigned int numa_node)
1741 {
1742 	struct Scsi_Host *shost = set->driver_data;
1743 	const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1744 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1745 	struct scatterlist *sg;
1746 
1747 	if (unchecked_isa_dma)
1748 		cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1749 	cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1750 						    GFP_KERNEL, numa_node);
1751 	if (!cmd->sense_buffer)
1752 		return -ENOMEM;
1753 	cmd->req.sense = cmd->sense_buffer;
1754 
1755 	if (scsi_host_get_prot(shost)) {
1756 		sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1757 			shost->hostt->cmd_size;
1758 		cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1759 	}
1760 
1761 	return 0;
1762 }
1763 
1764 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1765 				 unsigned int hctx_idx)
1766 {
1767 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1768 
1769 	scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1770 			       cmd->sense_buffer);
1771 }
1772 
1773 static int scsi_map_queues(struct blk_mq_tag_set *set)
1774 {
1775 	struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1776 
1777 	if (shost->hostt->map_queues)
1778 		return shost->hostt->map_queues(shost);
1779 	return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1780 }
1781 
1782 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1783 {
1784 	struct device *dev = shost->dma_dev;
1785 
1786 	/*
1787 	 * this limit is imposed by hardware restrictions
1788 	 */
1789 	blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1790 					SG_MAX_SEGMENTS));
1791 
1792 	if (scsi_host_prot_dma(shost)) {
1793 		shost->sg_prot_tablesize =
1794 			min_not_zero(shost->sg_prot_tablesize,
1795 				     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1796 		BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1797 		blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1798 	}
1799 
1800 	if (dev->dma_mask) {
1801 		shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1802 				dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1803 	}
1804 	blk_queue_max_hw_sectors(q, shost->max_sectors);
1805 	if (shost->unchecked_isa_dma)
1806 		blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1807 	blk_queue_segment_boundary(q, shost->dma_boundary);
1808 	dma_set_seg_boundary(dev, shost->dma_boundary);
1809 
1810 	blk_queue_max_segment_size(q, shost->max_segment_size);
1811 	blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1812 	dma_set_max_seg_size(dev, queue_max_segment_size(q));
1813 
1814 	/*
1815 	 * Set a reasonable default alignment:  The larger of 32-byte (dword),
1816 	 * which is a common minimum for HBAs, and the minimum DMA alignment,
1817 	 * which is set by the platform.
1818 	 *
1819 	 * Devices that require a bigger alignment can increase it later.
1820 	 */
1821 	blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1822 }
1823 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1824 
1825 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1826 	.get_budget	= scsi_mq_get_budget,
1827 	.put_budget	= scsi_mq_put_budget,
1828 	.queue_rq	= scsi_queue_rq,
1829 	.complete	= scsi_softirq_done,
1830 	.timeout	= scsi_timeout,
1831 #ifdef CONFIG_BLK_DEBUG_FS
1832 	.show_rq	= scsi_show_rq,
1833 #endif
1834 	.init_request	= scsi_mq_init_request,
1835 	.exit_request	= scsi_mq_exit_request,
1836 	.initialize_rq_fn = scsi_initialize_rq,
1837 	.busy		= scsi_mq_lld_busy,
1838 	.map_queues	= scsi_map_queues,
1839 };
1840 
1841 
1842 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1843 {
1844 	struct request_queue *q = hctx->queue;
1845 	struct scsi_device *sdev = q->queuedata;
1846 	struct Scsi_Host *shost = sdev->host;
1847 
1848 	shost->hostt->commit_rqs(shost, hctx->queue_num);
1849 }
1850 
1851 static const struct blk_mq_ops scsi_mq_ops = {
1852 	.get_budget	= scsi_mq_get_budget,
1853 	.put_budget	= scsi_mq_put_budget,
1854 	.queue_rq	= scsi_queue_rq,
1855 	.commit_rqs	= scsi_commit_rqs,
1856 	.complete	= scsi_softirq_done,
1857 	.timeout	= scsi_timeout,
1858 #ifdef CONFIG_BLK_DEBUG_FS
1859 	.show_rq	= scsi_show_rq,
1860 #endif
1861 	.init_request	= scsi_mq_init_request,
1862 	.exit_request	= scsi_mq_exit_request,
1863 	.initialize_rq_fn = scsi_initialize_rq,
1864 	.cleanup_rq	= scsi_cleanup_rq,
1865 	.busy		= scsi_mq_lld_busy,
1866 	.map_queues	= scsi_map_queues,
1867 };
1868 
1869 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1870 {
1871 	sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1872 	if (IS_ERR(sdev->request_queue))
1873 		return NULL;
1874 
1875 	sdev->request_queue->queuedata = sdev;
1876 	__scsi_init_queue(sdev->host, sdev->request_queue);
1877 	blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1878 	return sdev->request_queue;
1879 }
1880 
1881 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1882 {
1883 	unsigned int cmd_size, sgl_size;
1884 
1885 	sgl_size = scsi_mq_inline_sgl_size(shost);
1886 	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1887 	if (scsi_host_get_prot(shost))
1888 		cmd_size += sizeof(struct scsi_data_buffer) +
1889 			sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1890 
1891 	memset(&shost->tag_set, 0, sizeof(shost->tag_set));
1892 	if (shost->hostt->commit_rqs)
1893 		shost->tag_set.ops = &scsi_mq_ops;
1894 	else
1895 		shost->tag_set.ops = &scsi_mq_ops_no_commit;
1896 	shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
1897 	shost->tag_set.queue_depth = shost->can_queue;
1898 	shost->tag_set.cmd_size = cmd_size;
1899 	shost->tag_set.numa_node = NUMA_NO_NODE;
1900 	shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1901 	shost->tag_set.flags |=
1902 		BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1903 	shost->tag_set.driver_data = shost;
1904 
1905 	return blk_mq_alloc_tag_set(&shost->tag_set);
1906 }
1907 
1908 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1909 {
1910 	blk_mq_free_tag_set(&shost->tag_set);
1911 }
1912 
1913 /**
1914  * scsi_device_from_queue - return sdev associated with a request_queue
1915  * @q: The request queue to return the sdev from
1916  *
1917  * Return the sdev associated with a request queue or NULL if the
1918  * request_queue does not reference a SCSI device.
1919  */
1920 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1921 {
1922 	struct scsi_device *sdev = NULL;
1923 
1924 	if (q->mq_ops == &scsi_mq_ops)
1925 		sdev = q->queuedata;
1926 	if (!sdev || !get_device(&sdev->sdev_gendev))
1927 		sdev = NULL;
1928 
1929 	return sdev;
1930 }
1931 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1932 
1933 /*
1934  * Function:    scsi_block_requests()
1935  *
1936  * Purpose:     Utility function used by low-level drivers to prevent further
1937  *		commands from being queued to the device.
1938  *
1939  * Arguments:   shost       - Host in question
1940  *
1941  * Returns:     Nothing
1942  *
1943  * Lock status: No locks are assumed held.
1944  *
1945  * Notes:       There is no timer nor any other means by which the requests
1946  *		get unblocked other than the low-level driver calling
1947  *		scsi_unblock_requests().
1948  */
1949 void scsi_block_requests(struct Scsi_Host *shost)
1950 {
1951 	shost->host_self_blocked = 1;
1952 }
1953 EXPORT_SYMBOL(scsi_block_requests);
1954 
1955 /*
1956  * Function:    scsi_unblock_requests()
1957  *
1958  * Purpose:     Utility function used by low-level drivers to allow further
1959  *		commands from being queued to the device.
1960  *
1961  * Arguments:   shost       - Host in question
1962  *
1963  * Returns:     Nothing
1964  *
1965  * Lock status: No locks are assumed held.
1966  *
1967  * Notes:       There is no timer nor any other means by which the requests
1968  *		get unblocked other than the low-level driver calling
1969  *		scsi_unblock_requests().
1970  *
1971  *		This is done as an API function so that changes to the
1972  *		internals of the scsi mid-layer won't require wholesale
1973  *		changes to drivers that use this feature.
1974  */
1975 void scsi_unblock_requests(struct Scsi_Host *shost)
1976 {
1977 	shost->host_self_blocked = 0;
1978 	scsi_run_host_queues(shost);
1979 }
1980 EXPORT_SYMBOL(scsi_unblock_requests);
1981 
1982 int __init scsi_init_queue(void)
1983 {
1984 	scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1985 					   sizeof(struct scsi_data_buffer),
1986 					   0, 0, NULL);
1987 	if (!scsi_sdb_cache) {
1988 		printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1989 		return -ENOMEM;
1990 	}
1991 
1992 	return 0;
1993 }
1994 
1995 void scsi_exit_queue(void)
1996 {
1997 	kmem_cache_destroy(scsi_sense_cache);
1998 	kmem_cache_destroy(scsi_sense_isadma_cache);
1999 	kmem_cache_destroy(scsi_sdb_cache);
2000 }
2001 
2002 /**
2003  *	scsi_mode_select - issue a mode select
2004  *	@sdev:	SCSI device to be queried
2005  *	@pf:	Page format bit (1 == standard, 0 == vendor specific)
2006  *	@sp:	Save page bit (0 == don't save, 1 == save)
2007  *	@modepage: mode page being requested
2008  *	@buffer: request buffer (may not be smaller than eight bytes)
2009  *	@len:	length of request buffer.
2010  *	@timeout: command timeout
2011  *	@retries: number of retries before failing
2012  *	@data: returns a structure abstracting the mode header data
2013  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2014  *		must be SCSI_SENSE_BUFFERSIZE big.
2015  *
2016  *	Returns zero if successful; negative error number or scsi
2017  *	status on error
2018  *
2019  */
2020 int
2021 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2022 		 unsigned char *buffer, int len, int timeout, int retries,
2023 		 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2024 {
2025 	unsigned char cmd[10];
2026 	unsigned char *real_buffer;
2027 	int ret;
2028 
2029 	memset(cmd, 0, sizeof(cmd));
2030 	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2031 
2032 	if (sdev->use_10_for_ms) {
2033 		if (len > 65535)
2034 			return -EINVAL;
2035 		real_buffer = kmalloc(8 + len, GFP_KERNEL);
2036 		if (!real_buffer)
2037 			return -ENOMEM;
2038 		memcpy(real_buffer + 8, buffer, len);
2039 		len += 8;
2040 		real_buffer[0] = 0;
2041 		real_buffer[1] = 0;
2042 		real_buffer[2] = data->medium_type;
2043 		real_buffer[3] = data->device_specific;
2044 		real_buffer[4] = data->longlba ? 0x01 : 0;
2045 		real_buffer[5] = 0;
2046 		real_buffer[6] = data->block_descriptor_length >> 8;
2047 		real_buffer[7] = data->block_descriptor_length;
2048 
2049 		cmd[0] = MODE_SELECT_10;
2050 		cmd[7] = len >> 8;
2051 		cmd[8] = len;
2052 	} else {
2053 		if (len > 255 || data->block_descriptor_length > 255 ||
2054 		    data->longlba)
2055 			return -EINVAL;
2056 
2057 		real_buffer = kmalloc(4 + len, GFP_KERNEL);
2058 		if (!real_buffer)
2059 			return -ENOMEM;
2060 		memcpy(real_buffer + 4, buffer, len);
2061 		len += 4;
2062 		real_buffer[0] = 0;
2063 		real_buffer[1] = data->medium_type;
2064 		real_buffer[2] = data->device_specific;
2065 		real_buffer[3] = data->block_descriptor_length;
2066 
2067 
2068 		cmd[0] = MODE_SELECT;
2069 		cmd[4] = len;
2070 	}
2071 
2072 	ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2073 			       sshdr, timeout, retries, NULL);
2074 	kfree(real_buffer);
2075 	return ret;
2076 }
2077 EXPORT_SYMBOL_GPL(scsi_mode_select);
2078 
2079 /**
2080  *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2081  *	@sdev:	SCSI device to be queried
2082  *	@dbd:	set if mode sense will allow block descriptors to be returned
2083  *	@modepage: mode page being requested
2084  *	@buffer: request buffer (may not be smaller than eight bytes)
2085  *	@len:	length of request buffer.
2086  *	@timeout: command timeout
2087  *	@retries: number of retries before failing
2088  *	@data: returns a structure abstracting the mode header data
2089  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2090  *		must be SCSI_SENSE_BUFFERSIZE big.
2091  *
2092  *	Returns zero if unsuccessful, or the header offset (either 4
2093  *	or 8 depending on whether a six or ten byte command was
2094  *	issued) if successful.
2095  */
2096 int
2097 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2098 		  unsigned char *buffer, int len, int timeout, int retries,
2099 		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2100 {
2101 	unsigned char cmd[12];
2102 	int use_10_for_ms;
2103 	int header_length;
2104 	int result, retry_count = retries;
2105 	struct scsi_sense_hdr my_sshdr;
2106 
2107 	memset(data, 0, sizeof(*data));
2108 	memset(&cmd[0], 0, 12);
2109 	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */
2110 	cmd[2] = modepage;
2111 
2112 	/* caller might not be interested in sense, but we need it */
2113 	if (!sshdr)
2114 		sshdr = &my_sshdr;
2115 
2116  retry:
2117 	use_10_for_ms = sdev->use_10_for_ms;
2118 
2119 	if (use_10_for_ms) {
2120 		if (len < 8)
2121 			len = 8;
2122 
2123 		cmd[0] = MODE_SENSE_10;
2124 		cmd[8] = len;
2125 		header_length = 8;
2126 	} else {
2127 		if (len < 4)
2128 			len = 4;
2129 
2130 		cmd[0] = MODE_SENSE;
2131 		cmd[4] = len;
2132 		header_length = 4;
2133 	}
2134 
2135 	memset(buffer, 0, len);
2136 
2137 	result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2138 				  sshdr, timeout, retries, NULL);
2139 
2140 	/* This code looks awful: what it's doing is making sure an
2141 	 * ILLEGAL REQUEST sense return identifies the actual command
2142 	 * byte as the problem.  MODE_SENSE commands can return
2143 	 * ILLEGAL REQUEST if the code page isn't supported */
2144 
2145 	if (use_10_for_ms && !scsi_status_is_good(result) &&
2146 	    driver_byte(result) == DRIVER_SENSE) {
2147 		if (scsi_sense_valid(sshdr)) {
2148 			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2149 			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2150 				/*
2151 				 * Invalid command operation code
2152 				 */
2153 				sdev->use_10_for_ms = 0;
2154 				goto retry;
2155 			}
2156 		}
2157 	}
2158 
2159 	if(scsi_status_is_good(result)) {
2160 		if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2161 			     (modepage == 6 || modepage == 8))) {
2162 			/* Initio breakage? */
2163 			header_length = 0;
2164 			data->length = 13;
2165 			data->medium_type = 0;
2166 			data->device_specific = 0;
2167 			data->longlba = 0;
2168 			data->block_descriptor_length = 0;
2169 		} else if(use_10_for_ms) {
2170 			data->length = buffer[0]*256 + buffer[1] + 2;
2171 			data->medium_type = buffer[2];
2172 			data->device_specific = buffer[3];
2173 			data->longlba = buffer[4] & 0x01;
2174 			data->block_descriptor_length = buffer[6]*256
2175 				+ buffer[7];
2176 		} else {
2177 			data->length = buffer[0] + 1;
2178 			data->medium_type = buffer[1];
2179 			data->device_specific = buffer[2];
2180 			data->block_descriptor_length = buffer[3];
2181 		}
2182 		data->header_length = header_length;
2183 	} else if ((status_byte(result) == CHECK_CONDITION) &&
2184 		   scsi_sense_valid(sshdr) &&
2185 		   sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2186 		retry_count--;
2187 		goto retry;
2188 	}
2189 
2190 	return result;
2191 }
2192 EXPORT_SYMBOL(scsi_mode_sense);
2193 
2194 /**
2195  *	scsi_test_unit_ready - test if unit is ready
2196  *	@sdev:	scsi device to change the state of.
2197  *	@timeout: command timeout
2198  *	@retries: number of retries before failing
2199  *	@sshdr: outpout pointer for decoded sense information.
2200  *
2201  *	Returns zero if unsuccessful or an error if TUR failed.  For
2202  *	removable media, UNIT_ATTENTION sets ->changed flag.
2203  **/
2204 int
2205 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2206 		     struct scsi_sense_hdr *sshdr)
2207 {
2208 	char cmd[] = {
2209 		TEST_UNIT_READY, 0, 0, 0, 0, 0,
2210 	};
2211 	int result;
2212 
2213 	/* try to eat the UNIT_ATTENTION if there are enough retries */
2214 	do {
2215 		result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2216 					  timeout, 1, NULL);
2217 		if (sdev->removable && scsi_sense_valid(sshdr) &&
2218 		    sshdr->sense_key == UNIT_ATTENTION)
2219 			sdev->changed = 1;
2220 	} while (scsi_sense_valid(sshdr) &&
2221 		 sshdr->sense_key == UNIT_ATTENTION && --retries);
2222 
2223 	return result;
2224 }
2225 EXPORT_SYMBOL(scsi_test_unit_ready);
2226 
2227 /**
2228  *	scsi_device_set_state - Take the given device through the device state model.
2229  *	@sdev:	scsi device to change the state of.
2230  *	@state:	state to change to.
2231  *
2232  *	Returns zero if successful or an error if the requested
2233  *	transition is illegal.
2234  */
2235 int
2236 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2237 {
2238 	enum scsi_device_state oldstate = sdev->sdev_state;
2239 
2240 	if (state == oldstate)
2241 		return 0;
2242 
2243 	switch (state) {
2244 	case SDEV_CREATED:
2245 		switch (oldstate) {
2246 		case SDEV_CREATED_BLOCK:
2247 			break;
2248 		default:
2249 			goto illegal;
2250 		}
2251 		break;
2252 
2253 	case SDEV_RUNNING:
2254 		switch (oldstate) {
2255 		case SDEV_CREATED:
2256 		case SDEV_OFFLINE:
2257 		case SDEV_TRANSPORT_OFFLINE:
2258 		case SDEV_QUIESCE:
2259 		case SDEV_BLOCK:
2260 			break;
2261 		default:
2262 			goto illegal;
2263 		}
2264 		break;
2265 
2266 	case SDEV_QUIESCE:
2267 		switch (oldstate) {
2268 		case SDEV_RUNNING:
2269 		case SDEV_OFFLINE:
2270 		case SDEV_TRANSPORT_OFFLINE:
2271 			break;
2272 		default:
2273 			goto illegal;
2274 		}
2275 		break;
2276 
2277 	case SDEV_OFFLINE:
2278 	case SDEV_TRANSPORT_OFFLINE:
2279 		switch (oldstate) {
2280 		case SDEV_CREATED:
2281 		case SDEV_RUNNING:
2282 		case SDEV_QUIESCE:
2283 		case SDEV_BLOCK:
2284 			break;
2285 		default:
2286 			goto illegal;
2287 		}
2288 		break;
2289 
2290 	case SDEV_BLOCK:
2291 		switch (oldstate) {
2292 		case SDEV_RUNNING:
2293 		case SDEV_CREATED_BLOCK:
2294 		case SDEV_OFFLINE:
2295 			break;
2296 		default:
2297 			goto illegal;
2298 		}
2299 		break;
2300 
2301 	case SDEV_CREATED_BLOCK:
2302 		switch (oldstate) {
2303 		case SDEV_CREATED:
2304 			break;
2305 		default:
2306 			goto illegal;
2307 		}
2308 		break;
2309 
2310 	case SDEV_CANCEL:
2311 		switch (oldstate) {
2312 		case SDEV_CREATED:
2313 		case SDEV_RUNNING:
2314 		case SDEV_QUIESCE:
2315 		case SDEV_OFFLINE:
2316 		case SDEV_TRANSPORT_OFFLINE:
2317 			break;
2318 		default:
2319 			goto illegal;
2320 		}
2321 		break;
2322 
2323 	case SDEV_DEL:
2324 		switch (oldstate) {
2325 		case SDEV_CREATED:
2326 		case SDEV_RUNNING:
2327 		case SDEV_OFFLINE:
2328 		case SDEV_TRANSPORT_OFFLINE:
2329 		case SDEV_CANCEL:
2330 		case SDEV_BLOCK:
2331 		case SDEV_CREATED_BLOCK:
2332 			break;
2333 		default:
2334 			goto illegal;
2335 		}
2336 		break;
2337 
2338 	}
2339 	sdev->sdev_state = state;
2340 	return 0;
2341 
2342  illegal:
2343 	SCSI_LOG_ERROR_RECOVERY(1,
2344 				sdev_printk(KERN_ERR, sdev,
2345 					    "Illegal state transition %s->%s",
2346 					    scsi_device_state_name(oldstate),
2347 					    scsi_device_state_name(state))
2348 				);
2349 	return -EINVAL;
2350 }
2351 EXPORT_SYMBOL(scsi_device_set_state);
2352 
2353 /**
2354  * 	sdev_evt_emit - emit a single SCSI device uevent
2355  *	@sdev: associated SCSI device
2356  *	@evt: event to emit
2357  *
2358  *	Send a single uevent (scsi_event) to the associated scsi_device.
2359  */
2360 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2361 {
2362 	int idx = 0;
2363 	char *envp[3];
2364 
2365 	switch (evt->evt_type) {
2366 	case SDEV_EVT_MEDIA_CHANGE:
2367 		envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2368 		break;
2369 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2370 		scsi_rescan_device(&sdev->sdev_gendev);
2371 		envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2372 		break;
2373 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2374 		envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2375 		break;
2376 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2377 	       envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2378 		break;
2379 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2380 		envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2381 		break;
2382 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2383 		envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2384 		break;
2385 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2386 		envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2387 		break;
2388 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2389 		envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2390 		break;
2391 	default:
2392 		/* do nothing */
2393 		break;
2394 	}
2395 
2396 	envp[idx++] = NULL;
2397 
2398 	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2399 }
2400 
2401 /**
2402  * 	sdev_evt_thread - send a uevent for each scsi event
2403  *	@work: work struct for scsi_device
2404  *
2405  *	Dispatch queued events to their associated scsi_device kobjects
2406  *	as uevents.
2407  */
2408 void scsi_evt_thread(struct work_struct *work)
2409 {
2410 	struct scsi_device *sdev;
2411 	enum scsi_device_event evt_type;
2412 	LIST_HEAD(event_list);
2413 
2414 	sdev = container_of(work, struct scsi_device, event_work);
2415 
2416 	for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2417 		if (test_and_clear_bit(evt_type, sdev->pending_events))
2418 			sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2419 
2420 	while (1) {
2421 		struct scsi_event *evt;
2422 		struct list_head *this, *tmp;
2423 		unsigned long flags;
2424 
2425 		spin_lock_irqsave(&sdev->list_lock, flags);
2426 		list_splice_init(&sdev->event_list, &event_list);
2427 		spin_unlock_irqrestore(&sdev->list_lock, flags);
2428 
2429 		if (list_empty(&event_list))
2430 			break;
2431 
2432 		list_for_each_safe(this, tmp, &event_list) {
2433 			evt = list_entry(this, struct scsi_event, node);
2434 			list_del(&evt->node);
2435 			scsi_evt_emit(sdev, evt);
2436 			kfree(evt);
2437 		}
2438 	}
2439 }
2440 
2441 /**
2442  * 	sdev_evt_send - send asserted event to uevent thread
2443  *	@sdev: scsi_device event occurred on
2444  *	@evt: event to send
2445  *
2446  *	Assert scsi device event asynchronously.
2447  */
2448 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2449 {
2450 	unsigned long flags;
2451 
2452 #if 0
2453 	/* FIXME: currently this check eliminates all media change events
2454 	 * for polled devices.  Need to update to discriminate between AN
2455 	 * and polled events */
2456 	if (!test_bit(evt->evt_type, sdev->supported_events)) {
2457 		kfree(evt);
2458 		return;
2459 	}
2460 #endif
2461 
2462 	spin_lock_irqsave(&sdev->list_lock, flags);
2463 	list_add_tail(&evt->node, &sdev->event_list);
2464 	schedule_work(&sdev->event_work);
2465 	spin_unlock_irqrestore(&sdev->list_lock, flags);
2466 }
2467 EXPORT_SYMBOL_GPL(sdev_evt_send);
2468 
2469 /**
2470  * 	sdev_evt_alloc - allocate a new scsi event
2471  *	@evt_type: type of event to allocate
2472  *	@gfpflags: GFP flags for allocation
2473  *
2474  *	Allocates and returns a new scsi_event.
2475  */
2476 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2477 				  gfp_t gfpflags)
2478 {
2479 	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2480 	if (!evt)
2481 		return NULL;
2482 
2483 	evt->evt_type = evt_type;
2484 	INIT_LIST_HEAD(&evt->node);
2485 
2486 	/* evt_type-specific initialization, if any */
2487 	switch (evt_type) {
2488 	case SDEV_EVT_MEDIA_CHANGE:
2489 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2490 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2491 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2492 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2493 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2494 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2495 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2496 	default:
2497 		/* do nothing */
2498 		break;
2499 	}
2500 
2501 	return evt;
2502 }
2503 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2504 
2505 /**
2506  * 	sdev_evt_send_simple - send asserted event to uevent thread
2507  *	@sdev: scsi_device event occurred on
2508  *	@evt_type: type of event to send
2509  *	@gfpflags: GFP flags for allocation
2510  *
2511  *	Assert scsi device event asynchronously, given an event type.
2512  */
2513 void sdev_evt_send_simple(struct scsi_device *sdev,
2514 			  enum scsi_device_event evt_type, gfp_t gfpflags)
2515 {
2516 	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2517 	if (!evt) {
2518 		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2519 			    evt_type);
2520 		return;
2521 	}
2522 
2523 	sdev_evt_send(sdev, evt);
2524 }
2525 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2526 
2527 /**
2528  *	scsi_device_quiesce - Block user issued commands.
2529  *	@sdev:	scsi device to quiesce.
2530  *
2531  *	This works by trying to transition to the SDEV_QUIESCE state
2532  *	(which must be a legal transition).  When the device is in this
2533  *	state, only special requests will be accepted, all others will
2534  *	be deferred.  Since special requests may also be requeued requests,
2535  *	a successful return doesn't guarantee the device will be
2536  *	totally quiescent.
2537  *
2538  *	Must be called with user context, may sleep.
2539  *
2540  *	Returns zero if unsuccessful or an error if not.
2541  */
2542 int
2543 scsi_device_quiesce(struct scsi_device *sdev)
2544 {
2545 	struct request_queue *q = sdev->request_queue;
2546 	int err;
2547 
2548 	/*
2549 	 * It is allowed to call scsi_device_quiesce() multiple times from
2550 	 * the same context but concurrent scsi_device_quiesce() calls are
2551 	 * not allowed.
2552 	 */
2553 	WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2554 
2555 	if (sdev->quiesced_by == current)
2556 		return 0;
2557 
2558 	blk_set_pm_only(q);
2559 
2560 	blk_mq_freeze_queue(q);
2561 	/*
2562 	 * Ensure that the effect of blk_set_pm_only() will be visible
2563 	 * for percpu_ref_tryget() callers that occur after the queue
2564 	 * unfreeze even if the queue was already frozen before this function
2565 	 * was called. See also https://lwn.net/Articles/573497/.
2566 	 */
2567 	synchronize_rcu();
2568 	blk_mq_unfreeze_queue(q);
2569 
2570 	mutex_lock(&sdev->state_mutex);
2571 	err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2572 	if (err == 0)
2573 		sdev->quiesced_by = current;
2574 	else
2575 		blk_clear_pm_only(q);
2576 	mutex_unlock(&sdev->state_mutex);
2577 
2578 	return err;
2579 }
2580 EXPORT_SYMBOL(scsi_device_quiesce);
2581 
2582 /**
2583  *	scsi_device_resume - Restart user issued commands to a quiesced device.
2584  *	@sdev:	scsi device to resume.
2585  *
2586  *	Moves the device from quiesced back to running and restarts the
2587  *	queues.
2588  *
2589  *	Must be called with user context, may sleep.
2590  */
2591 void scsi_device_resume(struct scsi_device *sdev)
2592 {
2593 	/* check if the device state was mutated prior to resume, and if
2594 	 * so assume the state is being managed elsewhere (for example
2595 	 * device deleted during suspend)
2596 	 */
2597 	mutex_lock(&sdev->state_mutex);
2598 	if (sdev->quiesced_by) {
2599 		sdev->quiesced_by = NULL;
2600 		blk_clear_pm_only(sdev->request_queue);
2601 	}
2602 	if (sdev->sdev_state == SDEV_QUIESCE)
2603 		scsi_device_set_state(sdev, SDEV_RUNNING);
2604 	mutex_unlock(&sdev->state_mutex);
2605 }
2606 EXPORT_SYMBOL(scsi_device_resume);
2607 
2608 static void
2609 device_quiesce_fn(struct scsi_device *sdev, void *data)
2610 {
2611 	scsi_device_quiesce(sdev);
2612 }
2613 
2614 void
2615 scsi_target_quiesce(struct scsi_target *starget)
2616 {
2617 	starget_for_each_device(starget, NULL, device_quiesce_fn);
2618 }
2619 EXPORT_SYMBOL(scsi_target_quiesce);
2620 
2621 static void
2622 device_resume_fn(struct scsi_device *sdev, void *data)
2623 {
2624 	scsi_device_resume(sdev);
2625 }
2626 
2627 void
2628 scsi_target_resume(struct scsi_target *starget)
2629 {
2630 	starget_for_each_device(starget, NULL, device_resume_fn);
2631 }
2632 EXPORT_SYMBOL(scsi_target_resume);
2633 
2634 /**
2635  * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2636  * @sdev: device to block
2637  *
2638  * Pause SCSI command processing on the specified device. Does not sleep.
2639  *
2640  * Returns zero if successful or a negative error code upon failure.
2641  *
2642  * Notes:
2643  * This routine transitions the device to the SDEV_BLOCK state (which must be
2644  * a legal transition). When the device is in this state, command processing
2645  * is paused until the device leaves the SDEV_BLOCK state. See also
2646  * scsi_internal_device_unblock_nowait().
2647  */
2648 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2649 {
2650 	struct request_queue *q = sdev->request_queue;
2651 	int err = 0;
2652 
2653 	err = scsi_device_set_state(sdev, SDEV_BLOCK);
2654 	if (err) {
2655 		err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2656 
2657 		if (err)
2658 			return err;
2659 	}
2660 
2661 	/*
2662 	 * The device has transitioned to SDEV_BLOCK.  Stop the
2663 	 * block layer from calling the midlayer with this device's
2664 	 * request queue.
2665 	 */
2666 	blk_mq_quiesce_queue_nowait(q);
2667 	return 0;
2668 }
2669 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2670 
2671 /**
2672  * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2673  * @sdev: device to block
2674  *
2675  * Pause SCSI command processing on the specified device and wait until all
2676  * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2677  *
2678  * Returns zero if successful or a negative error code upon failure.
2679  *
2680  * Note:
2681  * This routine transitions the device to the SDEV_BLOCK state (which must be
2682  * a legal transition). When the device is in this state, command processing
2683  * is paused until the device leaves the SDEV_BLOCK state. See also
2684  * scsi_internal_device_unblock().
2685  */
2686 static int scsi_internal_device_block(struct scsi_device *sdev)
2687 {
2688 	struct request_queue *q = sdev->request_queue;
2689 	int err;
2690 
2691 	mutex_lock(&sdev->state_mutex);
2692 	err = scsi_internal_device_block_nowait(sdev);
2693 	if (err == 0)
2694 		blk_mq_quiesce_queue(q);
2695 	mutex_unlock(&sdev->state_mutex);
2696 
2697 	return err;
2698 }
2699 
2700 void scsi_start_queue(struct scsi_device *sdev)
2701 {
2702 	struct request_queue *q = sdev->request_queue;
2703 
2704 	blk_mq_unquiesce_queue(q);
2705 }
2706 
2707 /**
2708  * scsi_internal_device_unblock_nowait - resume a device after a block request
2709  * @sdev:	device to resume
2710  * @new_state:	state to set the device to after unblocking
2711  *
2712  * Restart the device queue for a previously suspended SCSI device. Does not
2713  * sleep.
2714  *
2715  * Returns zero if successful or a negative error code upon failure.
2716  *
2717  * Notes:
2718  * This routine transitions the device to the SDEV_RUNNING state or to one of
2719  * the offline states (which must be a legal transition) allowing the midlayer
2720  * to goose the queue for this device.
2721  */
2722 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2723 					enum scsi_device_state new_state)
2724 {
2725 	switch (new_state) {
2726 	case SDEV_RUNNING:
2727 	case SDEV_TRANSPORT_OFFLINE:
2728 		break;
2729 	default:
2730 		return -EINVAL;
2731 	}
2732 
2733 	/*
2734 	 * Try to transition the scsi device to SDEV_RUNNING or one of the
2735 	 * offlined states and goose the device queue if successful.
2736 	 */
2737 	switch (sdev->sdev_state) {
2738 	case SDEV_BLOCK:
2739 	case SDEV_TRANSPORT_OFFLINE:
2740 		sdev->sdev_state = new_state;
2741 		break;
2742 	case SDEV_CREATED_BLOCK:
2743 		if (new_state == SDEV_TRANSPORT_OFFLINE ||
2744 		    new_state == SDEV_OFFLINE)
2745 			sdev->sdev_state = new_state;
2746 		else
2747 			sdev->sdev_state = SDEV_CREATED;
2748 		break;
2749 	case SDEV_CANCEL:
2750 	case SDEV_OFFLINE:
2751 		break;
2752 	default:
2753 		return -EINVAL;
2754 	}
2755 	scsi_start_queue(sdev);
2756 
2757 	return 0;
2758 }
2759 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2760 
2761 /**
2762  * scsi_internal_device_unblock - resume a device after a block request
2763  * @sdev:	device to resume
2764  * @new_state:	state to set the device to after unblocking
2765  *
2766  * Restart the device queue for a previously suspended SCSI device. May sleep.
2767  *
2768  * Returns zero if successful or a negative error code upon failure.
2769  *
2770  * Notes:
2771  * This routine transitions the device to the SDEV_RUNNING state or to one of
2772  * the offline states (which must be a legal transition) allowing the midlayer
2773  * to goose the queue for this device.
2774  */
2775 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2776 					enum scsi_device_state new_state)
2777 {
2778 	int ret;
2779 
2780 	mutex_lock(&sdev->state_mutex);
2781 	ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2782 	mutex_unlock(&sdev->state_mutex);
2783 
2784 	return ret;
2785 }
2786 
2787 static void
2788 device_block(struct scsi_device *sdev, void *data)
2789 {
2790 	int ret;
2791 
2792 	ret = scsi_internal_device_block(sdev);
2793 
2794 	WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2795 		  dev_name(&sdev->sdev_gendev), ret);
2796 }
2797 
2798 static int
2799 target_block(struct device *dev, void *data)
2800 {
2801 	if (scsi_is_target_device(dev))
2802 		starget_for_each_device(to_scsi_target(dev), NULL,
2803 					device_block);
2804 	return 0;
2805 }
2806 
2807 void
2808 scsi_target_block(struct device *dev)
2809 {
2810 	if (scsi_is_target_device(dev))
2811 		starget_for_each_device(to_scsi_target(dev), NULL,
2812 					device_block);
2813 	else
2814 		device_for_each_child(dev, NULL, target_block);
2815 }
2816 EXPORT_SYMBOL_GPL(scsi_target_block);
2817 
2818 static void
2819 device_unblock(struct scsi_device *sdev, void *data)
2820 {
2821 	scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2822 }
2823 
2824 static int
2825 target_unblock(struct device *dev, void *data)
2826 {
2827 	if (scsi_is_target_device(dev))
2828 		starget_for_each_device(to_scsi_target(dev), data,
2829 					device_unblock);
2830 	return 0;
2831 }
2832 
2833 void
2834 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2835 {
2836 	if (scsi_is_target_device(dev))
2837 		starget_for_each_device(to_scsi_target(dev), &new_state,
2838 					device_unblock);
2839 	else
2840 		device_for_each_child(dev, &new_state, target_unblock);
2841 }
2842 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2843 
2844 /**
2845  * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2846  * @sgl:	scatter-gather list
2847  * @sg_count:	number of segments in sg
2848  * @offset:	offset in bytes into sg, on return offset into the mapped area
2849  * @len:	bytes to map, on return number of bytes mapped
2850  *
2851  * Returns virtual address of the start of the mapped page
2852  */
2853 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2854 			  size_t *offset, size_t *len)
2855 {
2856 	int i;
2857 	size_t sg_len = 0, len_complete = 0;
2858 	struct scatterlist *sg;
2859 	struct page *page;
2860 
2861 	WARN_ON(!irqs_disabled());
2862 
2863 	for_each_sg(sgl, sg, sg_count, i) {
2864 		len_complete = sg_len; /* Complete sg-entries */
2865 		sg_len += sg->length;
2866 		if (sg_len > *offset)
2867 			break;
2868 	}
2869 
2870 	if (unlikely(i == sg_count)) {
2871 		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2872 			"elements %d\n",
2873 		       __func__, sg_len, *offset, sg_count);
2874 		WARN_ON(1);
2875 		return NULL;
2876 	}
2877 
2878 	/* Offset starting from the beginning of first page in this sg-entry */
2879 	*offset = *offset - len_complete + sg->offset;
2880 
2881 	/* Assumption: contiguous pages can be accessed as "page + i" */
2882 	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2883 	*offset &= ~PAGE_MASK;
2884 
2885 	/* Bytes in this sg-entry from *offset to the end of the page */
2886 	sg_len = PAGE_SIZE - *offset;
2887 	if (*len > sg_len)
2888 		*len = sg_len;
2889 
2890 	return kmap_atomic(page);
2891 }
2892 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2893 
2894 /**
2895  * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2896  * @virt:	virtual address to be unmapped
2897  */
2898 void scsi_kunmap_atomic_sg(void *virt)
2899 {
2900 	kunmap_atomic(virt);
2901 }
2902 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2903 
2904 void sdev_disable_disk_events(struct scsi_device *sdev)
2905 {
2906 	atomic_inc(&sdev->disk_events_disable_depth);
2907 }
2908 EXPORT_SYMBOL(sdev_disable_disk_events);
2909 
2910 void sdev_enable_disk_events(struct scsi_device *sdev)
2911 {
2912 	if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2913 		return;
2914 	atomic_dec(&sdev->disk_events_disable_depth);
2915 }
2916 EXPORT_SYMBOL(sdev_enable_disk_events);
2917 
2918 /**
2919  * scsi_vpd_lun_id - return a unique device identification
2920  * @sdev: SCSI device
2921  * @id:   buffer for the identification
2922  * @id_len:  length of the buffer
2923  *
2924  * Copies a unique device identification into @id based
2925  * on the information in the VPD page 0x83 of the device.
2926  * The string will be formatted as a SCSI name string.
2927  *
2928  * Returns the length of the identification or error on failure.
2929  * If the identifier is longer than the supplied buffer the actual
2930  * identifier length is returned and the buffer is not zero-padded.
2931  */
2932 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2933 {
2934 	u8 cur_id_type = 0xff;
2935 	u8 cur_id_size = 0;
2936 	const unsigned char *d, *cur_id_str;
2937 	const struct scsi_vpd *vpd_pg83;
2938 	int id_size = -EINVAL;
2939 
2940 	rcu_read_lock();
2941 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2942 	if (!vpd_pg83) {
2943 		rcu_read_unlock();
2944 		return -ENXIO;
2945 	}
2946 
2947 	/*
2948 	 * Look for the correct descriptor.
2949 	 * Order of preference for lun descriptor:
2950 	 * - SCSI name string
2951 	 * - NAA IEEE Registered Extended
2952 	 * - EUI-64 based 16-byte
2953 	 * - EUI-64 based 12-byte
2954 	 * - NAA IEEE Registered
2955 	 * - NAA IEEE Extended
2956 	 * - T10 Vendor ID
2957 	 * as longer descriptors reduce the likelyhood
2958 	 * of identification clashes.
2959 	 */
2960 
2961 	/* The id string must be at least 20 bytes + terminating NULL byte */
2962 	if (id_len < 21) {
2963 		rcu_read_unlock();
2964 		return -EINVAL;
2965 	}
2966 
2967 	memset(id, 0, id_len);
2968 	d = vpd_pg83->data + 4;
2969 	while (d < vpd_pg83->data + vpd_pg83->len) {
2970 		/* Skip designators not referring to the LUN */
2971 		if ((d[1] & 0x30) != 0x00)
2972 			goto next_desig;
2973 
2974 		switch (d[1] & 0xf) {
2975 		case 0x1:
2976 			/* T10 Vendor ID */
2977 			if (cur_id_size > d[3])
2978 				break;
2979 			/* Prefer anything */
2980 			if (cur_id_type > 0x01 && cur_id_type != 0xff)
2981 				break;
2982 			cur_id_size = d[3];
2983 			if (cur_id_size + 4 > id_len)
2984 				cur_id_size = id_len - 4;
2985 			cur_id_str = d + 4;
2986 			cur_id_type = d[1] & 0xf;
2987 			id_size = snprintf(id, id_len, "t10.%*pE",
2988 					   cur_id_size, cur_id_str);
2989 			break;
2990 		case 0x2:
2991 			/* EUI-64 */
2992 			if (cur_id_size > d[3])
2993 				break;
2994 			/* Prefer NAA IEEE Registered Extended */
2995 			if (cur_id_type == 0x3 &&
2996 			    cur_id_size == d[3])
2997 				break;
2998 			cur_id_size = d[3];
2999 			cur_id_str = d + 4;
3000 			cur_id_type = d[1] & 0xf;
3001 			switch (cur_id_size) {
3002 			case 8:
3003 				id_size = snprintf(id, id_len,
3004 						   "eui.%8phN",
3005 						   cur_id_str);
3006 				break;
3007 			case 12:
3008 				id_size = snprintf(id, id_len,
3009 						   "eui.%12phN",
3010 						   cur_id_str);
3011 				break;
3012 			case 16:
3013 				id_size = snprintf(id, id_len,
3014 						   "eui.%16phN",
3015 						   cur_id_str);
3016 				break;
3017 			default:
3018 				cur_id_size = 0;
3019 				break;
3020 			}
3021 			break;
3022 		case 0x3:
3023 			/* NAA */
3024 			if (cur_id_size > d[3])
3025 				break;
3026 			cur_id_size = d[3];
3027 			cur_id_str = d + 4;
3028 			cur_id_type = d[1] & 0xf;
3029 			switch (cur_id_size) {
3030 			case 8:
3031 				id_size = snprintf(id, id_len,
3032 						   "naa.%8phN",
3033 						   cur_id_str);
3034 				break;
3035 			case 16:
3036 				id_size = snprintf(id, id_len,
3037 						   "naa.%16phN",
3038 						   cur_id_str);
3039 				break;
3040 			default:
3041 				cur_id_size = 0;
3042 				break;
3043 			}
3044 			break;
3045 		case 0x8:
3046 			/* SCSI name string */
3047 			if (cur_id_size + 4 > d[3])
3048 				break;
3049 			/* Prefer others for truncated descriptor */
3050 			if (cur_id_size && d[3] > id_len)
3051 				break;
3052 			cur_id_size = id_size = d[3];
3053 			cur_id_str = d + 4;
3054 			cur_id_type = d[1] & 0xf;
3055 			if (cur_id_size >= id_len)
3056 				cur_id_size = id_len - 1;
3057 			memcpy(id, cur_id_str, cur_id_size);
3058 			/* Decrease priority for truncated descriptor */
3059 			if (cur_id_size != id_size)
3060 				cur_id_size = 6;
3061 			break;
3062 		default:
3063 			break;
3064 		}
3065 next_desig:
3066 		d += d[3] + 4;
3067 	}
3068 	rcu_read_unlock();
3069 
3070 	return id_size;
3071 }
3072 EXPORT_SYMBOL(scsi_vpd_lun_id);
3073 
3074 /*
3075  * scsi_vpd_tpg_id - return a target port group identifier
3076  * @sdev: SCSI device
3077  *
3078  * Returns the Target Port Group identifier from the information
3079  * froom VPD page 0x83 of the device.
3080  *
3081  * Returns the identifier or error on failure.
3082  */
3083 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3084 {
3085 	const unsigned char *d;
3086 	const struct scsi_vpd *vpd_pg83;
3087 	int group_id = -EAGAIN, rel_port = -1;
3088 
3089 	rcu_read_lock();
3090 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3091 	if (!vpd_pg83) {
3092 		rcu_read_unlock();
3093 		return -ENXIO;
3094 	}
3095 
3096 	d = vpd_pg83->data + 4;
3097 	while (d < vpd_pg83->data + vpd_pg83->len) {
3098 		switch (d[1] & 0xf) {
3099 		case 0x4:
3100 			/* Relative target port */
3101 			rel_port = get_unaligned_be16(&d[6]);
3102 			break;
3103 		case 0x5:
3104 			/* Target port group */
3105 			group_id = get_unaligned_be16(&d[6]);
3106 			break;
3107 		default:
3108 			break;
3109 		}
3110 		d += d[3] + 4;
3111 	}
3112 	rcu_read_unlock();
3113 
3114 	if (group_id >= 0 && rel_id && rel_port != -1)
3115 		*rel_id = rel_port;
3116 
3117 	return group_id;
3118 }
3119 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3120