xref: /openbmc/linux/drivers/scsi/scsi_lib.c (revision 9a8f3203)
1 /*
2  * Copyright (C) 1999 Eric Youngdale
3  * Copyright (C) 2014 Christoph Hellwig
4  *
5  *  SCSI queueing library.
6  *      Initial versions: Eric Youngdale (eric@andante.org).
7  *                        Based upon conversations with large numbers
8  *                        of people at Linux Expo.
9  */
10 
11 #include <linux/bio.h>
12 #include <linux/bitops.h>
13 #include <linux/blkdev.h>
14 #include <linux/completion.h>
15 #include <linux/kernel.h>
16 #include <linux/export.h>
17 #include <linux/init.h>
18 #include <linux/pci.h>
19 #include <linux/delay.h>
20 #include <linux/hardirq.h>
21 #include <linux/scatterlist.h>
22 #include <linux/blk-mq.h>
23 #include <linux/ratelimit.h>
24 #include <asm/unaligned.h>
25 
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_driver.h>
31 #include <scsi/scsi_eh.h>
32 #include <scsi/scsi_host.h>
33 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
34 #include <scsi/scsi_dh.h>
35 
36 #include <trace/events/scsi.h>
37 
38 #include "scsi_debugfs.h"
39 #include "scsi_priv.h"
40 #include "scsi_logging.h"
41 
42 static struct kmem_cache *scsi_sdb_cache;
43 static struct kmem_cache *scsi_sense_cache;
44 static struct kmem_cache *scsi_sense_isadma_cache;
45 static DEFINE_MUTEX(scsi_sense_cache_mutex);
46 
47 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
48 
49 static inline struct kmem_cache *
50 scsi_select_sense_cache(bool unchecked_isa_dma)
51 {
52 	return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
53 }
54 
55 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
56 				   unsigned char *sense_buffer)
57 {
58 	kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
59 			sense_buffer);
60 }
61 
62 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
63 	gfp_t gfp_mask, int numa_node)
64 {
65 	return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
66 				     gfp_mask, numa_node);
67 }
68 
69 int scsi_init_sense_cache(struct Scsi_Host *shost)
70 {
71 	struct kmem_cache *cache;
72 	int ret = 0;
73 
74 	cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
75 	if (cache)
76 		return 0;
77 
78 	mutex_lock(&scsi_sense_cache_mutex);
79 	if (shost->unchecked_isa_dma) {
80 		scsi_sense_isadma_cache =
81 			kmem_cache_create("scsi_sense_cache(DMA)",
82 				SCSI_SENSE_BUFFERSIZE, 0,
83 				SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
84 		if (!scsi_sense_isadma_cache)
85 			ret = -ENOMEM;
86 	} else {
87 		scsi_sense_cache =
88 			kmem_cache_create_usercopy("scsi_sense_cache",
89 				SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
90 				0, SCSI_SENSE_BUFFERSIZE, NULL);
91 		if (!scsi_sense_cache)
92 			ret = -ENOMEM;
93 	}
94 
95 	mutex_unlock(&scsi_sense_cache_mutex);
96 	return ret;
97 }
98 
99 /*
100  * When to reinvoke queueing after a resource shortage. It's 3 msecs to
101  * not change behaviour from the previous unplug mechanism, experimentation
102  * may prove this needs changing.
103  */
104 #define SCSI_QUEUE_DELAY	3
105 
106 static void
107 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
108 {
109 	struct Scsi_Host *host = cmd->device->host;
110 	struct scsi_device *device = cmd->device;
111 	struct scsi_target *starget = scsi_target(device);
112 
113 	/*
114 	 * Set the appropriate busy bit for the device/host.
115 	 *
116 	 * If the host/device isn't busy, assume that something actually
117 	 * completed, and that we should be able to queue a command now.
118 	 *
119 	 * Note that the prior mid-layer assumption that any host could
120 	 * always queue at least one command is now broken.  The mid-layer
121 	 * will implement a user specifiable stall (see
122 	 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
123 	 * if a command is requeued with no other commands outstanding
124 	 * either for the device or for the host.
125 	 */
126 	switch (reason) {
127 	case SCSI_MLQUEUE_HOST_BUSY:
128 		atomic_set(&host->host_blocked, host->max_host_blocked);
129 		break;
130 	case SCSI_MLQUEUE_DEVICE_BUSY:
131 	case SCSI_MLQUEUE_EH_RETRY:
132 		atomic_set(&device->device_blocked,
133 			   device->max_device_blocked);
134 		break;
135 	case SCSI_MLQUEUE_TARGET_BUSY:
136 		atomic_set(&starget->target_blocked,
137 			   starget->max_target_blocked);
138 		break;
139 	}
140 }
141 
142 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
143 {
144 	struct scsi_device *sdev = cmd->device;
145 
146 	if (cmd->request->rq_flags & RQF_DONTPREP) {
147 		cmd->request->rq_flags &= ~RQF_DONTPREP;
148 		scsi_mq_uninit_cmd(cmd);
149 	} else {
150 		WARN_ON_ONCE(true);
151 	}
152 	blk_mq_requeue_request(cmd->request, true);
153 	put_device(&sdev->sdev_gendev);
154 }
155 
156 /**
157  * __scsi_queue_insert - private queue insertion
158  * @cmd: The SCSI command being requeued
159  * @reason:  The reason for the requeue
160  * @unbusy: Whether the queue should be unbusied
161  *
162  * This is a private queue insertion.  The public interface
163  * scsi_queue_insert() always assumes the queue should be unbusied
164  * because it's always called before the completion.  This function is
165  * for a requeue after completion, which should only occur in this
166  * file.
167  */
168 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
169 {
170 	struct scsi_device *device = cmd->device;
171 
172 	SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
173 		"Inserting command %p into mlqueue\n", cmd));
174 
175 	scsi_set_blocked(cmd, reason);
176 
177 	/*
178 	 * Decrement the counters, since these commands are no longer
179 	 * active on the host/device.
180 	 */
181 	if (unbusy)
182 		scsi_device_unbusy(device);
183 
184 	/*
185 	 * Requeue this command.  It will go before all other commands
186 	 * that are already in the queue. Schedule requeue work under
187 	 * lock such that the kblockd_schedule_work() call happens
188 	 * before blk_cleanup_queue() finishes.
189 	 */
190 	cmd->result = 0;
191 
192 	/*
193 	 * Before a SCSI command is dispatched,
194 	 * get_device(&sdev->sdev_gendev) is called and the host,
195 	 * target and device busy counters are increased. Since
196 	 * requeuing a request causes these actions to be repeated and
197 	 * since scsi_device_unbusy() has already been called,
198 	 * put_device(&device->sdev_gendev) must still be called. Call
199 	 * put_device() after blk_mq_requeue_request() to avoid that
200 	 * removal of the SCSI device can start before requeueing has
201 	 * happened.
202 	 */
203 	blk_mq_requeue_request(cmd->request, true);
204 	put_device(&device->sdev_gendev);
205 }
206 
207 /*
208  * Function:    scsi_queue_insert()
209  *
210  * Purpose:     Insert a command in the midlevel queue.
211  *
212  * Arguments:   cmd    - command that we are adding to queue.
213  *              reason - why we are inserting command to queue.
214  *
215  * Lock status: Assumed that lock is not held upon entry.
216  *
217  * Returns:     Nothing.
218  *
219  * Notes:       We do this for one of two cases.  Either the host is busy
220  *              and it cannot accept any more commands for the time being,
221  *              or the device returned QUEUE_FULL and can accept no more
222  *              commands.
223  * Notes:       This could be called either from an interrupt context or a
224  *              normal process context.
225  */
226 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
227 {
228 	__scsi_queue_insert(cmd, reason, true);
229 }
230 
231 
232 /**
233  * __scsi_execute - insert request and wait for the result
234  * @sdev:	scsi device
235  * @cmd:	scsi command
236  * @data_direction: data direction
237  * @buffer:	data buffer
238  * @bufflen:	len of buffer
239  * @sense:	optional sense buffer
240  * @sshdr:	optional decoded sense header
241  * @timeout:	request timeout in seconds
242  * @retries:	number of times to retry request
243  * @flags:	flags for ->cmd_flags
244  * @rq_flags:	flags for ->rq_flags
245  * @resid:	optional residual length
246  *
247  * Returns the scsi_cmnd result field if a command was executed, or a negative
248  * Linux error code if we didn't get that far.
249  */
250 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
251 		 int data_direction, void *buffer, unsigned bufflen,
252 		 unsigned char *sense, struct scsi_sense_hdr *sshdr,
253 		 int timeout, int retries, u64 flags, req_flags_t rq_flags,
254 		 int *resid)
255 {
256 	struct request *req;
257 	struct scsi_request *rq;
258 	int ret = DRIVER_ERROR << 24;
259 
260 	req = blk_get_request(sdev->request_queue,
261 			data_direction == DMA_TO_DEVICE ?
262 			REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
263 	if (IS_ERR(req))
264 		return ret;
265 	rq = scsi_req(req);
266 
267 	if (bufflen &&	blk_rq_map_kern(sdev->request_queue, req,
268 					buffer, bufflen, GFP_NOIO))
269 		goto out;
270 
271 	rq->cmd_len = COMMAND_SIZE(cmd[0]);
272 	memcpy(rq->cmd, cmd, rq->cmd_len);
273 	rq->retries = retries;
274 	req->timeout = timeout;
275 	req->cmd_flags |= flags;
276 	req->rq_flags |= rq_flags | RQF_QUIET;
277 
278 	/*
279 	 * head injection *required* here otherwise quiesce won't work
280 	 */
281 	blk_execute_rq(req->q, NULL, req, 1);
282 
283 	/*
284 	 * Some devices (USB mass-storage in particular) may transfer
285 	 * garbage data together with a residue indicating that the data
286 	 * is invalid.  Prevent the garbage from being misinterpreted
287 	 * and prevent security leaks by zeroing out the excess data.
288 	 */
289 	if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
290 		memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
291 
292 	if (resid)
293 		*resid = rq->resid_len;
294 	if (sense && rq->sense_len)
295 		memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
296 	if (sshdr)
297 		scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
298 	ret = rq->result;
299  out:
300 	blk_put_request(req);
301 
302 	return ret;
303 }
304 EXPORT_SYMBOL(__scsi_execute);
305 
306 /*
307  * Function:    scsi_init_cmd_errh()
308  *
309  * Purpose:     Initialize cmd fields related to error handling.
310  *
311  * Arguments:   cmd	- command that is ready to be queued.
312  *
313  * Notes:       This function has the job of initializing a number of
314  *              fields related to error handling.   Typically this will
315  *              be called once for each command, as required.
316  */
317 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
318 {
319 	scsi_set_resid(cmd, 0);
320 	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
321 	if (cmd->cmd_len == 0)
322 		cmd->cmd_len = scsi_command_size(cmd->cmnd);
323 }
324 
325 /*
326  * Decrement the host_busy counter and wake up the error handler if necessary.
327  * Avoid as follows that the error handler is not woken up if shost->host_busy
328  * == shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
329  * with an RCU read lock in this function to ensure that this function in its
330  * entirety either finishes before scsi_eh_scmd_add() increases the
331  * host_failed counter or that it notices the shost state change made by
332  * scsi_eh_scmd_add().
333  */
334 static void scsi_dec_host_busy(struct Scsi_Host *shost)
335 {
336 	unsigned long flags;
337 
338 	rcu_read_lock();
339 	atomic_dec(&shost->host_busy);
340 	if (unlikely(scsi_host_in_recovery(shost))) {
341 		spin_lock_irqsave(shost->host_lock, flags);
342 		if (shost->host_failed || shost->host_eh_scheduled)
343 			scsi_eh_wakeup(shost);
344 		spin_unlock_irqrestore(shost->host_lock, flags);
345 	}
346 	rcu_read_unlock();
347 }
348 
349 void scsi_device_unbusy(struct scsi_device *sdev)
350 {
351 	struct Scsi_Host *shost = sdev->host;
352 	struct scsi_target *starget = scsi_target(sdev);
353 
354 	scsi_dec_host_busy(shost);
355 
356 	if (starget->can_queue > 0)
357 		atomic_dec(&starget->target_busy);
358 
359 	atomic_dec(&sdev->device_busy);
360 }
361 
362 static void scsi_kick_queue(struct request_queue *q)
363 {
364 	blk_mq_run_hw_queues(q, false);
365 }
366 
367 /*
368  * Called for single_lun devices on IO completion. Clear starget_sdev_user,
369  * and call blk_run_queue for all the scsi_devices on the target -
370  * including current_sdev first.
371  *
372  * Called with *no* scsi locks held.
373  */
374 static void scsi_single_lun_run(struct scsi_device *current_sdev)
375 {
376 	struct Scsi_Host *shost = current_sdev->host;
377 	struct scsi_device *sdev, *tmp;
378 	struct scsi_target *starget = scsi_target(current_sdev);
379 	unsigned long flags;
380 
381 	spin_lock_irqsave(shost->host_lock, flags);
382 	starget->starget_sdev_user = NULL;
383 	spin_unlock_irqrestore(shost->host_lock, flags);
384 
385 	/*
386 	 * Call blk_run_queue for all LUNs on the target, starting with
387 	 * current_sdev. We race with others (to set starget_sdev_user),
388 	 * but in most cases, we will be first. Ideally, each LU on the
389 	 * target would get some limited time or requests on the target.
390 	 */
391 	scsi_kick_queue(current_sdev->request_queue);
392 
393 	spin_lock_irqsave(shost->host_lock, flags);
394 	if (starget->starget_sdev_user)
395 		goto out;
396 	list_for_each_entry_safe(sdev, tmp, &starget->devices,
397 			same_target_siblings) {
398 		if (sdev == current_sdev)
399 			continue;
400 		if (scsi_device_get(sdev))
401 			continue;
402 
403 		spin_unlock_irqrestore(shost->host_lock, flags);
404 		scsi_kick_queue(sdev->request_queue);
405 		spin_lock_irqsave(shost->host_lock, flags);
406 
407 		scsi_device_put(sdev);
408 	}
409  out:
410 	spin_unlock_irqrestore(shost->host_lock, flags);
411 }
412 
413 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
414 {
415 	if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
416 		return true;
417 	if (atomic_read(&sdev->device_blocked) > 0)
418 		return true;
419 	return false;
420 }
421 
422 static inline bool scsi_target_is_busy(struct scsi_target *starget)
423 {
424 	if (starget->can_queue > 0) {
425 		if (atomic_read(&starget->target_busy) >= starget->can_queue)
426 			return true;
427 		if (atomic_read(&starget->target_blocked) > 0)
428 			return true;
429 	}
430 	return false;
431 }
432 
433 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
434 {
435 	if (shost->can_queue > 0 &&
436 	    atomic_read(&shost->host_busy) >= shost->can_queue)
437 		return true;
438 	if (atomic_read(&shost->host_blocked) > 0)
439 		return true;
440 	if (shost->host_self_blocked)
441 		return true;
442 	return false;
443 }
444 
445 static void scsi_starved_list_run(struct Scsi_Host *shost)
446 {
447 	LIST_HEAD(starved_list);
448 	struct scsi_device *sdev;
449 	unsigned long flags;
450 
451 	spin_lock_irqsave(shost->host_lock, flags);
452 	list_splice_init(&shost->starved_list, &starved_list);
453 
454 	while (!list_empty(&starved_list)) {
455 		struct request_queue *slq;
456 
457 		/*
458 		 * As long as shost is accepting commands and we have
459 		 * starved queues, call blk_run_queue. scsi_request_fn
460 		 * drops the queue_lock and can add us back to the
461 		 * starved_list.
462 		 *
463 		 * host_lock protects the starved_list and starved_entry.
464 		 * scsi_request_fn must get the host_lock before checking
465 		 * or modifying starved_list or starved_entry.
466 		 */
467 		if (scsi_host_is_busy(shost))
468 			break;
469 
470 		sdev = list_entry(starved_list.next,
471 				  struct scsi_device, starved_entry);
472 		list_del_init(&sdev->starved_entry);
473 		if (scsi_target_is_busy(scsi_target(sdev))) {
474 			list_move_tail(&sdev->starved_entry,
475 				       &shost->starved_list);
476 			continue;
477 		}
478 
479 		/*
480 		 * Once we drop the host lock, a racing scsi_remove_device()
481 		 * call may remove the sdev from the starved list and destroy
482 		 * it and the queue.  Mitigate by taking a reference to the
483 		 * queue and never touching the sdev again after we drop the
484 		 * host lock.  Note: if __scsi_remove_device() invokes
485 		 * blk_cleanup_queue() before the queue is run from this
486 		 * function then blk_run_queue() will return immediately since
487 		 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
488 		 */
489 		slq = sdev->request_queue;
490 		if (!blk_get_queue(slq))
491 			continue;
492 		spin_unlock_irqrestore(shost->host_lock, flags);
493 
494 		scsi_kick_queue(slq);
495 		blk_put_queue(slq);
496 
497 		spin_lock_irqsave(shost->host_lock, flags);
498 	}
499 	/* put any unprocessed entries back */
500 	list_splice(&starved_list, &shost->starved_list);
501 	spin_unlock_irqrestore(shost->host_lock, flags);
502 }
503 
504 /*
505  * Function:   scsi_run_queue()
506  *
507  * Purpose:    Select a proper request queue to serve next
508  *
509  * Arguments:  q       - last request's queue
510  *
511  * Returns:     Nothing
512  *
513  * Notes:      The previous command was completely finished, start
514  *             a new one if possible.
515  */
516 static void scsi_run_queue(struct request_queue *q)
517 {
518 	struct scsi_device *sdev = q->queuedata;
519 
520 	if (scsi_target(sdev)->single_lun)
521 		scsi_single_lun_run(sdev);
522 	if (!list_empty(&sdev->host->starved_list))
523 		scsi_starved_list_run(sdev->host);
524 
525 	blk_mq_run_hw_queues(q, false);
526 }
527 
528 void scsi_requeue_run_queue(struct work_struct *work)
529 {
530 	struct scsi_device *sdev;
531 	struct request_queue *q;
532 
533 	sdev = container_of(work, struct scsi_device, requeue_work);
534 	q = sdev->request_queue;
535 	scsi_run_queue(q);
536 }
537 
538 void scsi_run_host_queues(struct Scsi_Host *shost)
539 {
540 	struct scsi_device *sdev;
541 
542 	shost_for_each_device(sdev, shost)
543 		scsi_run_queue(sdev->request_queue);
544 }
545 
546 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
547 {
548 	if (!blk_rq_is_passthrough(cmd->request)) {
549 		struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
550 
551 		if (drv->uninit_command)
552 			drv->uninit_command(cmd);
553 	}
554 }
555 
556 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
557 {
558 	if (cmd->sdb.table.nents)
559 		sg_free_table_chained(&cmd->sdb.table, true);
560 	if (scsi_prot_sg_count(cmd))
561 		sg_free_table_chained(&cmd->prot_sdb->table, true);
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 	put_device(&sdev->sdev_gendev);
623 	return false;
624 }
625 
626 /**
627  * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
628  * @cmd:	SCSI command
629  * @result:	scsi error code
630  *
631  * Translate a SCSI result code into a blk_status_t value. May reset the host
632  * byte of @cmd->result.
633  */
634 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
635 {
636 	switch (host_byte(result)) {
637 	case DID_OK:
638 		/*
639 		 * Also check the other bytes than the status byte in result
640 		 * to handle the case when a SCSI LLD sets result to
641 		 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
642 		 */
643 		if (scsi_status_is_good(result) && (result & ~0xff) == 0)
644 			return BLK_STS_OK;
645 		return BLK_STS_IOERR;
646 	case DID_TRANSPORT_FAILFAST:
647 		return BLK_STS_TRANSPORT;
648 	case DID_TARGET_FAILURE:
649 		set_host_byte(cmd, DID_OK);
650 		return BLK_STS_TARGET;
651 	case DID_NEXUS_FAILURE:
652 		set_host_byte(cmd, DID_OK);
653 		return BLK_STS_NEXUS;
654 	case DID_ALLOC_FAILURE:
655 		set_host_byte(cmd, DID_OK);
656 		return BLK_STS_NOSPC;
657 	case DID_MEDIUM_ERROR:
658 		set_host_byte(cmd, DID_OK);
659 		return BLK_STS_MEDIUM;
660 	default:
661 		return BLK_STS_IOERR;
662 	}
663 }
664 
665 /* Helper for scsi_io_completion() when "reprep" action required. */
666 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
667 				      struct request_queue *q)
668 {
669 	/* A new command will be prepared and issued. */
670 	scsi_mq_requeue_cmd(cmd);
671 }
672 
673 /* Helper for scsi_io_completion() when special action required. */
674 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
675 {
676 	struct request_queue *q = cmd->device->request_queue;
677 	struct request *req = cmd->request;
678 	int level = 0;
679 	enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
680 	      ACTION_DELAYED_RETRY} action;
681 	unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
682 	struct scsi_sense_hdr sshdr;
683 	bool sense_valid;
684 	bool sense_current = true;      /* false implies "deferred sense" */
685 	blk_status_t blk_stat;
686 
687 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
688 	if (sense_valid)
689 		sense_current = !scsi_sense_is_deferred(&sshdr);
690 
691 	blk_stat = scsi_result_to_blk_status(cmd, result);
692 
693 	if (host_byte(result) == DID_RESET) {
694 		/* Third party bus reset or reset for error recovery
695 		 * reasons.  Just retry the command and see what
696 		 * happens.
697 		 */
698 		action = ACTION_RETRY;
699 	} else if (sense_valid && sense_current) {
700 		switch (sshdr.sense_key) {
701 		case UNIT_ATTENTION:
702 			if (cmd->device->removable) {
703 				/* Detected disc change.  Set a bit
704 				 * and quietly refuse further access.
705 				 */
706 				cmd->device->changed = 1;
707 				action = ACTION_FAIL;
708 			} else {
709 				/* Must have been a power glitch, or a
710 				 * bus reset.  Could not have been a
711 				 * media change, so we just retry the
712 				 * command and see what happens.
713 				 */
714 				action = ACTION_RETRY;
715 			}
716 			break;
717 		case ILLEGAL_REQUEST:
718 			/* If we had an ILLEGAL REQUEST returned, then
719 			 * we may have performed an unsupported
720 			 * command.  The only thing this should be
721 			 * would be a ten byte read where only a six
722 			 * byte read was supported.  Also, on a system
723 			 * where READ CAPACITY failed, we may have
724 			 * read past the end of the disk.
725 			 */
726 			if ((cmd->device->use_10_for_rw &&
727 			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
728 			    (cmd->cmnd[0] == READ_10 ||
729 			     cmd->cmnd[0] == WRITE_10)) {
730 				/* This will issue a new 6-byte command. */
731 				cmd->device->use_10_for_rw = 0;
732 				action = ACTION_REPREP;
733 			} else if (sshdr.asc == 0x10) /* DIX */ {
734 				action = ACTION_FAIL;
735 				blk_stat = BLK_STS_PROTECTION;
736 			/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
737 			} else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
738 				action = ACTION_FAIL;
739 				blk_stat = BLK_STS_TARGET;
740 			} else
741 				action = ACTION_FAIL;
742 			break;
743 		case ABORTED_COMMAND:
744 			action = ACTION_FAIL;
745 			if (sshdr.asc == 0x10) /* DIF */
746 				blk_stat = BLK_STS_PROTECTION;
747 			break;
748 		case NOT_READY:
749 			/* If the device is in the process of becoming
750 			 * ready, or has a temporary blockage, retry.
751 			 */
752 			if (sshdr.asc == 0x04) {
753 				switch (sshdr.ascq) {
754 				case 0x01: /* becoming ready */
755 				case 0x04: /* format in progress */
756 				case 0x05: /* rebuild in progress */
757 				case 0x06: /* recalculation in progress */
758 				case 0x07: /* operation in progress */
759 				case 0x08: /* Long write in progress */
760 				case 0x09: /* self test in progress */
761 				case 0x14: /* space allocation in progress */
762 				case 0x1a: /* start stop unit in progress */
763 				case 0x1b: /* sanitize in progress */
764 				case 0x1d: /* configuration in progress */
765 				case 0x24: /* depopulation in progress */
766 					action = ACTION_DELAYED_RETRY;
767 					break;
768 				default:
769 					action = ACTION_FAIL;
770 					break;
771 				}
772 			} else
773 				action = ACTION_FAIL;
774 			break;
775 		case VOLUME_OVERFLOW:
776 			/* See SSC3rXX or current. */
777 			action = ACTION_FAIL;
778 			break;
779 		default:
780 			action = ACTION_FAIL;
781 			break;
782 		}
783 	} else
784 		action = ACTION_FAIL;
785 
786 	if (action != ACTION_FAIL &&
787 	    time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
788 		action = ACTION_FAIL;
789 
790 	switch (action) {
791 	case ACTION_FAIL:
792 		/* Give up and fail the remainder of the request */
793 		if (!(req->rq_flags & RQF_QUIET)) {
794 			static DEFINE_RATELIMIT_STATE(_rs,
795 					DEFAULT_RATELIMIT_INTERVAL,
796 					DEFAULT_RATELIMIT_BURST);
797 
798 			if (unlikely(scsi_logging_level))
799 				level =
800 				     SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
801 						    SCSI_LOG_MLCOMPLETE_BITS);
802 
803 			/*
804 			 * if logging is enabled the failure will be printed
805 			 * in scsi_log_completion(), so avoid duplicate messages
806 			 */
807 			if (!level && __ratelimit(&_rs)) {
808 				scsi_print_result(cmd, NULL, FAILED);
809 				if (driver_byte(result) == DRIVER_SENSE)
810 					scsi_print_sense(cmd);
811 				scsi_print_command(cmd);
812 			}
813 		}
814 		if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
815 			return;
816 		/*FALLTHRU*/
817 	case ACTION_REPREP:
818 		scsi_io_completion_reprep(cmd, q);
819 		break;
820 	case ACTION_RETRY:
821 		/* Retry the same command immediately */
822 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
823 		break;
824 	case ACTION_DELAYED_RETRY:
825 		/* Retry the same command after a delay */
826 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
827 		break;
828 	}
829 }
830 
831 /*
832  * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
833  * new result that may suppress further error checking. Also modifies
834  * *blk_statp in some cases.
835  */
836 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
837 					blk_status_t *blk_statp)
838 {
839 	bool sense_valid;
840 	bool sense_current = true;	/* false implies "deferred sense" */
841 	struct request *req = cmd->request;
842 	struct scsi_sense_hdr sshdr;
843 
844 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
845 	if (sense_valid)
846 		sense_current = !scsi_sense_is_deferred(&sshdr);
847 
848 	if (blk_rq_is_passthrough(req)) {
849 		if (sense_valid) {
850 			/*
851 			 * SG_IO wants current and deferred errors
852 			 */
853 			scsi_req(req)->sense_len =
854 				min(8 + cmd->sense_buffer[7],
855 				    SCSI_SENSE_BUFFERSIZE);
856 		}
857 		if (sense_current)
858 			*blk_statp = scsi_result_to_blk_status(cmd, result);
859 	} else if (blk_rq_bytes(req) == 0 && sense_current) {
860 		/*
861 		 * Flush commands do not transfers any data, and thus cannot use
862 		 * good_bytes != blk_rq_bytes(req) as the signal for an error.
863 		 * This sets *blk_statp explicitly for the problem case.
864 		 */
865 		*blk_statp = scsi_result_to_blk_status(cmd, result);
866 	}
867 	/*
868 	 * Recovered errors need reporting, but they're always treated as
869 	 * success, so fiddle the result code here.  For passthrough requests
870 	 * we already took a copy of the original into sreq->result which
871 	 * is what gets returned to the user
872 	 */
873 	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
874 		bool do_print = true;
875 		/*
876 		 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
877 		 * skip print since caller wants ATA registers. Only occurs
878 		 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
879 		 */
880 		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
881 			do_print = false;
882 		else if (req->rq_flags & RQF_QUIET)
883 			do_print = false;
884 		if (do_print)
885 			scsi_print_sense(cmd);
886 		result = 0;
887 		/* for passthrough, *blk_statp may be set */
888 		*blk_statp = BLK_STS_OK;
889 	}
890 	/*
891 	 * Another corner case: the SCSI status byte is non-zero but 'good'.
892 	 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
893 	 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
894 	 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
895 	 * intermediate statuses (both obsolete in SAM-4) as good.
896 	 */
897 	if (status_byte(result) && scsi_status_is_good(result)) {
898 		result = 0;
899 		*blk_statp = BLK_STS_OK;
900 	}
901 	return result;
902 }
903 
904 /*
905  * Function:    scsi_io_completion()
906  *
907  * Purpose:     Completion processing for block device I/O requests.
908  *
909  * Arguments:   cmd   - command that is finished.
910  *
911  * Lock status: Assumed that no lock is held upon entry.
912  *
913  * Returns:     Nothing
914  *
915  * Notes:       We will finish off the specified number of sectors.  If we
916  *		are done, the command block will be released and the queue
917  *		function will be goosed.  If we are not done then we have to
918  *		figure out what to do next:
919  *
920  *		a) We can call scsi_requeue_command().  The request
921  *		   will be unprepared and put back on the queue.  Then
922  *		   a new command will be created for it.  This should
923  *		   be used if we made forward progress, or if we want
924  *		   to switch from READ(10) to READ(6) for example.
925  *
926  *		b) We can call __scsi_queue_insert().  The request will
927  *		   be put back on the queue and retried using the same
928  *		   command as before, possibly after a delay.
929  *
930  *		c) We can call scsi_end_request() with blk_stat other than
931  *		   BLK_STS_OK, to fail the remainder of the request.
932  */
933 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
934 {
935 	int result = cmd->result;
936 	struct request_queue *q = cmd->device->request_queue;
937 	struct request *req = cmd->request;
938 	blk_status_t blk_stat = BLK_STS_OK;
939 
940 	if (unlikely(result))	/* a nz result may or may not be an error */
941 		result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
942 
943 	if (unlikely(blk_rq_is_passthrough(req))) {
944 		/*
945 		 * scsi_result_to_blk_status may have reset the host_byte
946 		 */
947 		scsi_req(req)->result = cmd->result;
948 	}
949 
950 	/*
951 	 * Next deal with any sectors which we were able to correctly
952 	 * handle.
953 	 */
954 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
955 		"%u sectors total, %d bytes done.\n",
956 		blk_rq_sectors(req), good_bytes));
957 
958 	/*
959 	 * Next deal with any sectors which we were able to correctly
960 	 * handle. Failed, zero length commands always need to drop down
961 	 * to retry code. Fast path should return in this block.
962 	 */
963 	if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
964 		if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
965 			return; /* no bytes remaining */
966 	}
967 
968 	/* Kill remainder if no retries. */
969 	if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
970 		if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
971 			WARN_ONCE(true,
972 			    "Bytes remaining after failed, no-retry command");
973 		return;
974 	}
975 
976 	/*
977 	 * If there had been no error, but we have leftover bytes in the
978 	 * requeues just queue the command up again.
979 	 */
980 	if (likely(result == 0))
981 		scsi_io_completion_reprep(cmd, q);
982 	else
983 		scsi_io_completion_action(cmd, result);
984 }
985 
986 static blk_status_t scsi_init_sgtable(struct request *req,
987 		struct scsi_data_buffer *sdb)
988 {
989 	int count;
990 
991 	/*
992 	 * If sg table allocation fails, requeue request later.
993 	 */
994 	if (unlikely(sg_alloc_table_chained(&sdb->table,
995 			blk_rq_nr_phys_segments(req), sdb->table.sgl)))
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 			ret = BLK_STS_RESOURCE;
1051 			goto out_free_sgtables;
1052 		}
1053 
1054 		count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1055 						prot_sdb->table.sgl);
1056 		BUG_ON(count > ivecs);
1057 		BUG_ON(count > queue_max_integrity_segments(rq->q));
1058 
1059 		cmd->prot_sdb = prot_sdb;
1060 		cmd->prot_sdb->table.nents = count;
1061 	}
1062 
1063 	return BLK_STS_OK;
1064 out_free_sgtables:
1065 	scsi_mq_free_sgtables(cmd);
1066 	return ret;
1067 }
1068 EXPORT_SYMBOL(scsi_init_io);
1069 
1070 /**
1071  * scsi_initialize_rq - initialize struct scsi_cmnd partially
1072  * @rq: Request associated with the SCSI command to be initialized.
1073  *
1074  * This function initializes the members of struct scsi_cmnd that must be
1075  * initialized before request processing starts and that won't be
1076  * reinitialized if a SCSI command is requeued.
1077  *
1078  * Called from inside blk_get_request() for pass-through requests and from
1079  * inside scsi_init_command() for filesystem requests.
1080  */
1081 static void scsi_initialize_rq(struct request *rq)
1082 {
1083 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1084 
1085 	scsi_req_init(&cmd->req);
1086 	init_rcu_head(&cmd->rcu);
1087 	cmd->jiffies_at_alloc = jiffies;
1088 	cmd->retries = 0;
1089 }
1090 
1091 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1092 void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1093 {
1094 	struct scsi_device *sdev = cmd->device;
1095 	struct Scsi_Host *shost = sdev->host;
1096 	unsigned long flags;
1097 
1098 	if (shost->use_cmd_list) {
1099 		spin_lock_irqsave(&sdev->list_lock, flags);
1100 		list_add_tail(&cmd->list, &sdev->cmd_list);
1101 		spin_unlock_irqrestore(&sdev->list_lock, flags);
1102 	}
1103 }
1104 
1105 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1106 void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1107 {
1108 	struct scsi_device *sdev = cmd->device;
1109 	struct Scsi_Host *shost = sdev->host;
1110 	unsigned long flags;
1111 
1112 	if (shost->use_cmd_list) {
1113 		spin_lock_irqsave(&sdev->list_lock, flags);
1114 		BUG_ON(list_empty(&cmd->list));
1115 		list_del_init(&cmd->list);
1116 		spin_unlock_irqrestore(&sdev->list_lock, flags);
1117 	}
1118 }
1119 
1120 /* Called after a request has been started. */
1121 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1122 {
1123 	void *buf = cmd->sense_buffer;
1124 	void *prot = cmd->prot_sdb;
1125 	struct request *rq = blk_mq_rq_from_pdu(cmd);
1126 	unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1127 	unsigned long jiffies_at_alloc;
1128 	int retries;
1129 
1130 	if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1131 		flags |= SCMD_INITIALIZED;
1132 		scsi_initialize_rq(rq);
1133 	}
1134 
1135 	jiffies_at_alloc = cmd->jiffies_at_alloc;
1136 	retries = cmd->retries;
1137 	/* zero out the cmd, except for the embedded scsi_request */
1138 	memset((char *)cmd + sizeof(cmd->req), 0,
1139 		sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1140 
1141 	cmd->device = dev;
1142 	cmd->sense_buffer = buf;
1143 	cmd->prot_sdb = prot;
1144 	cmd->flags = flags;
1145 	INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1146 	cmd->jiffies_at_alloc = jiffies_at_alloc;
1147 	cmd->retries = retries;
1148 
1149 	scsi_add_cmd_to_list(cmd);
1150 }
1151 
1152 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1153 		struct request *req)
1154 {
1155 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1156 
1157 	/*
1158 	 * Passthrough requests may transfer data, in which case they must
1159 	 * a bio attached to them.  Or they might contain a SCSI command
1160 	 * that does not transfer data, in which case they may optionally
1161 	 * submit a request without an attached bio.
1162 	 */
1163 	if (req->bio) {
1164 		blk_status_t ret = scsi_init_io(cmd);
1165 		if (unlikely(ret != BLK_STS_OK))
1166 			return ret;
1167 	} else {
1168 		BUG_ON(blk_rq_bytes(req));
1169 
1170 		memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1171 	}
1172 
1173 	cmd->cmd_len = scsi_req(req)->cmd_len;
1174 	cmd->cmnd = scsi_req(req)->cmd;
1175 	cmd->transfersize = blk_rq_bytes(req);
1176 	cmd->allowed = scsi_req(req)->retries;
1177 	return BLK_STS_OK;
1178 }
1179 
1180 /*
1181  * Setup a normal block command.  These are simple request from filesystems
1182  * that still need to be translated to SCSI CDBs from the ULD.
1183  */
1184 static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1185 		struct request *req)
1186 {
1187 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1188 
1189 	if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1190 		blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1191 		if (ret != BLK_STS_OK)
1192 			return ret;
1193 	}
1194 
1195 	cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1196 	memset(cmd->cmnd, 0, BLK_MAX_CDB);
1197 	return scsi_cmd_to_driver(cmd)->init_command(cmd);
1198 }
1199 
1200 static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1201 		struct request *req)
1202 {
1203 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1204 
1205 	if (!blk_rq_bytes(req))
1206 		cmd->sc_data_direction = DMA_NONE;
1207 	else if (rq_data_dir(req) == WRITE)
1208 		cmd->sc_data_direction = DMA_TO_DEVICE;
1209 	else
1210 		cmd->sc_data_direction = DMA_FROM_DEVICE;
1211 
1212 	if (blk_rq_is_scsi(req))
1213 		return scsi_setup_scsi_cmnd(sdev, req);
1214 	else
1215 		return scsi_setup_fs_cmnd(sdev, req);
1216 }
1217 
1218 static blk_status_t
1219 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1220 {
1221 	switch (sdev->sdev_state) {
1222 	case SDEV_OFFLINE:
1223 	case SDEV_TRANSPORT_OFFLINE:
1224 		/*
1225 		 * If the device is offline we refuse to process any
1226 		 * commands.  The device must be brought online
1227 		 * before trying any recovery commands.
1228 		 */
1229 		sdev_printk(KERN_ERR, sdev,
1230 			    "rejecting I/O to offline device\n");
1231 		return BLK_STS_IOERR;
1232 	case SDEV_DEL:
1233 		/*
1234 		 * If the device is fully deleted, we refuse to
1235 		 * process any commands as well.
1236 		 */
1237 		sdev_printk(KERN_ERR, sdev,
1238 			    "rejecting I/O to dead device\n");
1239 		return BLK_STS_IOERR;
1240 	case SDEV_BLOCK:
1241 	case SDEV_CREATED_BLOCK:
1242 		return BLK_STS_RESOURCE;
1243 	case SDEV_QUIESCE:
1244 		/*
1245 		 * If the devices is blocked we defer normal commands.
1246 		 */
1247 		if (req && !(req->rq_flags & RQF_PREEMPT))
1248 			return BLK_STS_RESOURCE;
1249 		return BLK_STS_OK;
1250 	default:
1251 		/*
1252 		 * For any other not fully online state we only allow
1253 		 * special commands.  In particular any user initiated
1254 		 * command is not allowed.
1255 		 */
1256 		if (req && !(req->rq_flags & RQF_PREEMPT))
1257 			return BLK_STS_IOERR;
1258 		return BLK_STS_OK;
1259 	}
1260 }
1261 
1262 /*
1263  * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1264  * return 0.
1265  *
1266  * Called with the queue_lock held.
1267  */
1268 static inline int scsi_dev_queue_ready(struct request_queue *q,
1269 				  struct scsi_device *sdev)
1270 {
1271 	unsigned int busy;
1272 
1273 	busy = atomic_inc_return(&sdev->device_busy) - 1;
1274 	if (atomic_read(&sdev->device_blocked)) {
1275 		if (busy)
1276 			goto out_dec;
1277 
1278 		/*
1279 		 * unblock after device_blocked iterates to zero
1280 		 */
1281 		if (atomic_dec_return(&sdev->device_blocked) > 0)
1282 			goto out_dec;
1283 		SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1284 				   "unblocking device at zero depth\n"));
1285 	}
1286 
1287 	if (busy >= sdev->queue_depth)
1288 		goto out_dec;
1289 
1290 	return 1;
1291 out_dec:
1292 	atomic_dec(&sdev->device_busy);
1293 	return 0;
1294 }
1295 
1296 /*
1297  * scsi_target_queue_ready: checks if there we can send commands to target
1298  * @sdev: scsi device on starget to check.
1299  */
1300 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1301 					   struct scsi_device *sdev)
1302 {
1303 	struct scsi_target *starget = scsi_target(sdev);
1304 	unsigned int busy;
1305 
1306 	if (starget->single_lun) {
1307 		spin_lock_irq(shost->host_lock);
1308 		if (starget->starget_sdev_user &&
1309 		    starget->starget_sdev_user != sdev) {
1310 			spin_unlock_irq(shost->host_lock);
1311 			return 0;
1312 		}
1313 		starget->starget_sdev_user = sdev;
1314 		spin_unlock_irq(shost->host_lock);
1315 	}
1316 
1317 	if (starget->can_queue <= 0)
1318 		return 1;
1319 
1320 	busy = atomic_inc_return(&starget->target_busy) - 1;
1321 	if (atomic_read(&starget->target_blocked) > 0) {
1322 		if (busy)
1323 			goto starved;
1324 
1325 		/*
1326 		 * unblock after target_blocked iterates to zero
1327 		 */
1328 		if (atomic_dec_return(&starget->target_blocked) > 0)
1329 			goto out_dec;
1330 
1331 		SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1332 				 "unblocking target at zero depth\n"));
1333 	}
1334 
1335 	if (busy >= starget->can_queue)
1336 		goto starved;
1337 
1338 	return 1;
1339 
1340 starved:
1341 	spin_lock_irq(shost->host_lock);
1342 	list_move_tail(&sdev->starved_entry, &shost->starved_list);
1343 	spin_unlock_irq(shost->host_lock);
1344 out_dec:
1345 	if (starget->can_queue > 0)
1346 		atomic_dec(&starget->target_busy);
1347 	return 0;
1348 }
1349 
1350 /*
1351  * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1352  * return 0. We must end up running the queue again whenever 0 is
1353  * returned, else IO can hang.
1354  */
1355 static inline int scsi_host_queue_ready(struct request_queue *q,
1356 				   struct Scsi_Host *shost,
1357 				   struct scsi_device *sdev)
1358 {
1359 	unsigned int busy;
1360 
1361 	if (scsi_host_in_recovery(shost))
1362 		return 0;
1363 
1364 	busy = atomic_inc_return(&shost->host_busy) - 1;
1365 	if (atomic_read(&shost->host_blocked) > 0) {
1366 		if (busy)
1367 			goto starved;
1368 
1369 		/*
1370 		 * unblock after host_blocked iterates to zero
1371 		 */
1372 		if (atomic_dec_return(&shost->host_blocked) > 0)
1373 			goto out_dec;
1374 
1375 		SCSI_LOG_MLQUEUE(3,
1376 			shost_printk(KERN_INFO, shost,
1377 				     "unblocking host at zero depth\n"));
1378 	}
1379 
1380 	if (shost->can_queue > 0 && busy >= shost->can_queue)
1381 		goto starved;
1382 	if (shost->host_self_blocked)
1383 		goto starved;
1384 
1385 	/* We're OK to process the command, so we can't be starved */
1386 	if (!list_empty(&sdev->starved_entry)) {
1387 		spin_lock_irq(shost->host_lock);
1388 		if (!list_empty(&sdev->starved_entry))
1389 			list_del_init(&sdev->starved_entry);
1390 		spin_unlock_irq(shost->host_lock);
1391 	}
1392 
1393 	return 1;
1394 
1395 starved:
1396 	spin_lock_irq(shost->host_lock);
1397 	if (list_empty(&sdev->starved_entry))
1398 		list_add_tail(&sdev->starved_entry, &shost->starved_list);
1399 	spin_unlock_irq(shost->host_lock);
1400 out_dec:
1401 	scsi_dec_host_busy(shost);
1402 	return 0;
1403 }
1404 
1405 /*
1406  * Busy state exporting function for request stacking drivers.
1407  *
1408  * For efficiency, no lock is taken to check the busy state of
1409  * shost/starget/sdev, since the returned value is not guaranteed and
1410  * may be changed after request stacking drivers call the function,
1411  * regardless of taking lock or not.
1412  *
1413  * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1414  * needs to return 'not busy'. Otherwise, request stacking drivers
1415  * may hold requests forever.
1416  */
1417 static bool scsi_mq_lld_busy(struct request_queue *q)
1418 {
1419 	struct scsi_device *sdev = q->queuedata;
1420 	struct Scsi_Host *shost;
1421 
1422 	if (blk_queue_dying(q))
1423 		return false;
1424 
1425 	shost = sdev->host;
1426 
1427 	/*
1428 	 * Ignore host/starget busy state.
1429 	 * Since block layer does not have a concept of fairness across
1430 	 * multiple queues, congestion of host/starget needs to be handled
1431 	 * in SCSI layer.
1432 	 */
1433 	if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1434 		return true;
1435 
1436 	return false;
1437 }
1438 
1439 static void scsi_softirq_done(struct request *rq)
1440 {
1441 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1442 	unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1443 	int disposition;
1444 
1445 	INIT_LIST_HEAD(&cmd->eh_entry);
1446 
1447 	atomic_inc(&cmd->device->iodone_cnt);
1448 	if (cmd->result)
1449 		atomic_inc(&cmd->device->ioerr_cnt);
1450 
1451 	disposition = scsi_decide_disposition(cmd);
1452 	if (disposition != SUCCESS &&
1453 	    time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1454 		sdev_printk(KERN_ERR, cmd->device,
1455 			    "timing out command, waited %lus\n",
1456 			    wait_for/HZ);
1457 		disposition = SUCCESS;
1458 	}
1459 
1460 	scsi_log_completion(cmd, disposition);
1461 
1462 	switch (disposition) {
1463 		case SUCCESS:
1464 			scsi_finish_command(cmd);
1465 			break;
1466 		case NEEDS_RETRY:
1467 			scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1468 			break;
1469 		case ADD_TO_MLQUEUE:
1470 			scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1471 			break;
1472 		default:
1473 			scsi_eh_scmd_add(cmd);
1474 			break;
1475 	}
1476 }
1477 
1478 /**
1479  * scsi_dispatch_command - Dispatch a command to the low-level driver.
1480  * @cmd: command block we are dispatching.
1481  *
1482  * Return: nonzero return request was rejected and device's queue needs to be
1483  * plugged.
1484  */
1485 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1486 {
1487 	struct Scsi_Host *host = cmd->device->host;
1488 	int rtn = 0;
1489 
1490 	atomic_inc(&cmd->device->iorequest_cnt);
1491 
1492 	/* check if the device is still usable */
1493 	if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1494 		/* in SDEV_DEL we error all commands. DID_NO_CONNECT
1495 		 * returns an immediate error upwards, and signals
1496 		 * that the device is no longer present */
1497 		cmd->result = DID_NO_CONNECT << 16;
1498 		goto done;
1499 	}
1500 
1501 	/* Check to see if the scsi lld made this device blocked. */
1502 	if (unlikely(scsi_device_blocked(cmd->device))) {
1503 		/*
1504 		 * in blocked state, the command is just put back on
1505 		 * the device queue.  The suspend state has already
1506 		 * blocked the queue so future requests should not
1507 		 * occur until the device transitions out of the
1508 		 * suspend state.
1509 		 */
1510 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1511 			"queuecommand : device blocked\n"));
1512 		return SCSI_MLQUEUE_DEVICE_BUSY;
1513 	}
1514 
1515 	/* Store the LUN value in cmnd, if needed. */
1516 	if (cmd->device->lun_in_cdb)
1517 		cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1518 			       (cmd->device->lun << 5 & 0xe0);
1519 
1520 	scsi_log_send(cmd);
1521 
1522 	/*
1523 	 * Before we queue this command, check if the command
1524 	 * length exceeds what the host adapter can handle.
1525 	 */
1526 	if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1527 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1528 			       "queuecommand : command too long. "
1529 			       "cdb_size=%d host->max_cmd_len=%d\n",
1530 			       cmd->cmd_len, cmd->device->host->max_cmd_len));
1531 		cmd->result = (DID_ABORT << 16);
1532 		goto done;
1533 	}
1534 
1535 	if (unlikely(host->shost_state == SHOST_DEL)) {
1536 		cmd->result = (DID_NO_CONNECT << 16);
1537 		goto done;
1538 
1539 	}
1540 
1541 	trace_scsi_dispatch_cmd_start(cmd);
1542 	rtn = host->hostt->queuecommand(host, cmd);
1543 	if (rtn) {
1544 		trace_scsi_dispatch_cmd_error(cmd, rtn);
1545 		if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1546 		    rtn != SCSI_MLQUEUE_TARGET_BUSY)
1547 			rtn = SCSI_MLQUEUE_HOST_BUSY;
1548 
1549 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1550 			"queuecommand : request rejected\n"));
1551 	}
1552 
1553 	return rtn;
1554  done:
1555 	cmd->scsi_done(cmd);
1556 	return 0;
1557 }
1558 
1559 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1560 static unsigned int scsi_mq_sgl_size(struct Scsi_Host *shost)
1561 {
1562 	return min_t(unsigned int, shost->sg_tablesize, SG_CHUNK_SIZE) *
1563 		sizeof(struct scatterlist);
1564 }
1565 
1566 static blk_status_t scsi_mq_prep_fn(struct request *req)
1567 {
1568 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1569 	struct scsi_device *sdev = req->q->queuedata;
1570 	struct Scsi_Host *shost = sdev->host;
1571 	struct scatterlist *sg;
1572 
1573 	scsi_init_command(sdev, cmd);
1574 
1575 	cmd->request = req;
1576 	cmd->tag = req->tag;
1577 	cmd->prot_op = SCSI_PROT_NORMAL;
1578 
1579 	sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1580 	cmd->sdb.table.sgl = sg;
1581 
1582 	if (scsi_host_get_prot(shost)) {
1583 		memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1584 
1585 		cmd->prot_sdb->table.sgl =
1586 			(struct scatterlist *)(cmd->prot_sdb + 1);
1587 	}
1588 
1589 	blk_mq_start_request(req);
1590 
1591 	return scsi_setup_cmnd(sdev, req);
1592 }
1593 
1594 static void scsi_mq_done(struct scsi_cmnd *cmd)
1595 {
1596 	if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1597 		return;
1598 	trace_scsi_dispatch_cmd_done(cmd);
1599 
1600 	/*
1601 	 * If the block layer didn't complete the request due to a timeout
1602 	 * injection, scsi must clear its internal completed state so that the
1603 	 * timeout handler will see it needs to escalate its own error
1604 	 * recovery.
1605 	 */
1606 	if (unlikely(!blk_mq_complete_request(cmd->request)))
1607 		clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1608 }
1609 
1610 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
1611 {
1612 	struct request_queue *q = hctx->queue;
1613 	struct scsi_device *sdev = q->queuedata;
1614 
1615 	atomic_dec(&sdev->device_busy);
1616 	put_device(&sdev->sdev_gendev);
1617 }
1618 
1619 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
1620 {
1621 	struct request_queue *q = hctx->queue;
1622 	struct scsi_device *sdev = q->queuedata;
1623 
1624 	if (!get_device(&sdev->sdev_gendev))
1625 		goto out;
1626 	if (!scsi_dev_queue_ready(q, sdev))
1627 		goto out_put_device;
1628 
1629 	return true;
1630 
1631 out_put_device:
1632 	put_device(&sdev->sdev_gendev);
1633 out:
1634 	if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
1635 		blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
1636 	return false;
1637 }
1638 
1639 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1640 			 const struct blk_mq_queue_data *bd)
1641 {
1642 	struct request *req = bd->rq;
1643 	struct request_queue *q = req->q;
1644 	struct scsi_device *sdev = q->queuedata;
1645 	struct Scsi_Host *shost = sdev->host;
1646 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1647 	blk_status_t ret;
1648 	int reason;
1649 
1650 	/*
1651 	 * If the device is not in running state we will reject some or all
1652 	 * commands.
1653 	 */
1654 	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1655 		ret = scsi_prep_state_check(sdev, req);
1656 		if (ret != BLK_STS_OK)
1657 			goto out_put_budget;
1658 	}
1659 
1660 	ret = BLK_STS_RESOURCE;
1661 	if (!scsi_target_queue_ready(shost, sdev))
1662 		goto out_put_budget;
1663 	if (!scsi_host_queue_ready(q, shost, sdev))
1664 		goto out_dec_target_busy;
1665 
1666 	if (!(req->rq_flags & RQF_DONTPREP)) {
1667 		ret = scsi_mq_prep_fn(req);
1668 		if (ret != BLK_STS_OK)
1669 			goto out_dec_host_busy;
1670 		req->rq_flags |= RQF_DONTPREP;
1671 	} else {
1672 		clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1673 		blk_mq_start_request(req);
1674 	}
1675 
1676 	if (sdev->simple_tags)
1677 		cmd->flags |= SCMD_TAGGED;
1678 	else
1679 		cmd->flags &= ~SCMD_TAGGED;
1680 
1681 	scsi_init_cmd_errh(cmd);
1682 	cmd->scsi_done = scsi_mq_done;
1683 
1684 	reason = scsi_dispatch_cmd(cmd);
1685 	if (reason) {
1686 		scsi_set_blocked(cmd, reason);
1687 		ret = BLK_STS_RESOURCE;
1688 		goto out_dec_host_busy;
1689 	}
1690 
1691 	return BLK_STS_OK;
1692 
1693 out_dec_host_busy:
1694 	scsi_dec_host_busy(shost);
1695 out_dec_target_busy:
1696 	if (scsi_target(sdev)->can_queue > 0)
1697 		atomic_dec(&scsi_target(sdev)->target_busy);
1698 out_put_budget:
1699 	scsi_mq_put_budget(hctx);
1700 	switch (ret) {
1701 	case BLK_STS_OK:
1702 		break;
1703 	case BLK_STS_RESOURCE:
1704 		if (atomic_read(&sdev->device_busy) ||
1705 		    scsi_device_blocked(sdev))
1706 			ret = BLK_STS_DEV_RESOURCE;
1707 		break;
1708 	default:
1709 		/*
1710 		 * Make sure to release all allocated ressources when
1711 		 * we hit an error, as we will never see this command
1712 		 * again.
1713 		 */
1714 		if (req->rq_flags & RQF_DONTPREP)
1715 			scsi_mq_uninit_cmd(cmd);
1716 		break;
1717 	}
1718 	return ret;
1719 }
1720 
1721 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1722 		bool reserved)
1723 {
1724 	if (reserved)
1725 		return BLK_EH_RESET_TIMER;
1726 	return scsi_times_out(req);
1727 }
1728 
1729 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1730 				unsigned int hctx_idx, unsigned int numa_node)
1731 {
1732 	struct Scsi_Host *shost = set->driver_data;
1733 	const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1734 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1735 	struct scatterlist *sg;
1736 
1737 	if (unchecked_isa_dma)
1738 		cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1739 	cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1740 						    GFP_KERNEL, numa_node);
1741 	if (!cmd->sense_buffer)
1742 		return -ENOMEM;
1743 	cmd->req.sense = cmd->sense_buffer;
1744 
1745 	if (scsi_host_get_prot(shost)) {
1746 		sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1747 			shost->hostt->cmd_size;
1748 		cmd->prot_sdb = (void *)sg + scsi_mq_sgl_size(shost);
1749 	}
1750 
1751 	return 0;
1752 }
1753 
1754 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1755 				 unsigned int hctx_idx)
1756 {
1757 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1758 
1759 	scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1760 			       cmd->sense_buffer);
1761 }
1762 
1763 static int scsi_map_queues(struct blk_mq_tag_set *set)
1764 {
1765 	struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1766 
1767 	if (shost->hostt->map_queues)
1768 		return shost->hostt->map_queues(shost);
1769 	return blk_mq_map_queues(&set->map[0]);
1770 }
1771 
1772 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1773 {
1774 	struct device *dev = shost->dma_dev;
1775 
1776 	/*
1777 	 * this limit is imposed by hardware restrictions
1778 	 */
1779 	blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1780 					SG_MAX_SEGMENTS));
1781 
1782 	if (scsi_host_prot_dma(shost)) {
1783 		shost->sg_prot_tablesize =
1784 			min_not_zero(shost->sg_prot_tablesize,
1785 				     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1786 		BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1787 		blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1788 	}
1789 
1790 	blk_queue_max_hw_sectors(q, shost->max_sectors);
1791 	if (shost->unchecked_isa_dma)
1792 		blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1793 	blk_queue_segment_boundary(q, shost->dma_boundary);
1794 	dma_set_seg_boundary(dev, shost->dma_boundary);
1795 
1796 	blk_queue_max_segment_size(q, shost->max_segment_size);
1797 	dma_set_max_seg_size(dev, shost->max_segment_size);
1798 
1799 	/*
1800 	 * Set a reasonable default alignment:  The larger of 32-byte (dword),
1801 	 * which is a common minimum for HBAs, and the minimum DMA alignment,
1802 	 * which is set by the platform.
1803 	 *
1804 	 * Devices that require a bigger alignment can increase it later.
1805 	 */
1806 	blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1807 }
1808 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1809 
1810 static const struct blk_mq_ops scsi_mq_ops = {
1811 	.get_budget	= scsi_mq_get_budget,
1812 	.put_budget	= scsi_mq_put_budget,
1813 	.queue_rq	= scsi_queue_rq,
1814 	.complete	= scsi_softirq_done,
1815 	.timeout	= scsi_timeout,
1816 #ifdef CONFIG_BLK_DEBUG_FS
1817 	.show_rq	= scsi_show_rq,
1818 #endif
1819 	.init_request	= scsi_mq_init_request,
1820 	.exit_request	= scsi_mq_exit_request,
1821 	.initialize_rq_fn = scsi_initialize_rq,
1822 	.busy		= scsi_mq_lld_busy,
1823 	.map_queues	= scsi_map_queues,
1824 };
1825 
1826 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1827 {
1828 	sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1829 	if (IS_ERR(sdev->request_queue))
1830 		return NULL;
1831 
1832 	sdev->request_queue->queuedata = sdev;
1833 	__scsi_init_queue(sdev->host, sdev->request_queue);
1834 	blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1835 	return sdev->request_queue;
1836 }
1837 
1838 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1839 {
1840 	unsigned int cmd_size, sgl_size;
1841 
1842 	sgl_size = scsi_mq_sgl_size(shost);
1843 	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1844 	if (scsi_host_get_prot(shost))
1845 		cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
1846 
1847 	memset(&shost->tag_set, 0, sizeof(shost->tag_set));
1848 	shost->tag_set.ops = &scsi_mq_ops;
1849 	shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
1850 	shost->tag_set.queue_depth = shost->can_queue;
1851 	shost->tag_set.cmd_size = cmd_size;
1852 	shost->tag_set.numa_node = NUMA_NO_NODE;
1853 	shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1854 	shost->tag_set.flags |=
1855 		BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1856 	shost->tag_set.driver_data = shost;
1857 
1858 	return blk_mq_alloc_tag_set(&shost->tag_set);
1859 }
1860 
1861 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1862 {
1863 	blk_mq_free_tag_set(&shost->tag_set);
1864 }
1865 
1866 /**
1867  * scsi_device_from_queue - return sdev associated with a request_queue
1868  * @q: The request queue to return the sdev from
1869  *
1870  * Return the sdev associated with a request queue or NULL if the
1871  * request_queue does not reference a SCSI device.
1872  */
1873 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1874 {
1875 	struct scsi_device *sdev = NULL;
1876 
1877 	if (q->mq_ops == &scsi_mq_ops)
1878 		sdev = q->queuedata;
1879 	if (!sdev || !get_device(&sdev->sdev_gendev))
1880 		sdev = NULL;
1881 
1882 	return sdev;
1883 }
1884 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1885 
1886 /*
1887  * Function:    scsi_block_requests()
1888  *
1889  * Purpose:     Utility function used by low-level drivers to prevent further
1890  *		commands from being queued to the device.
1891  *
1892  * Arguments:   shost       - Host in question
1893  *
1894  * Returns:     Nothing
1895  *
1896  * Lock status: No locks are assumed held.
1897  *
1898  * Notes:       There is no timer nor any other means by which the requests
1899  *		get unblocked other than the low-level driver calling
1900  *		scsi_unblock_requests().
1901  */
1902 void scsi_block_requests(struct Scsi_Host *shost)
1903 {
1904 	shost->host_self_blocked = 1;
1905 }
1906 EXPORT_SYMBOL(scsi_block_requests);
1907 
1908 /*
1909  * Function:    scsi_unblock_requests()
1910  *
1911  * Purpose:     Utility function used by low-level drivers to allow further
1912  *		commands from being queued to the device.
1913  *
1914  * Arguments:   shost       - Host in question
1915  *
1916  * Returns:     Nothing
1917  *
1918  * Lock status: No locks are assumed held.
1919  *
1920  * Notes:       There is no timer nor any other means by which the requests
1921  *		get unblocked other than the low-level driver calling
1922  *		scsi_unblock_requests().
1923  *
1924  *		This is done as an API function so that changes to the
1925  *		internals of the scsi mid-layer won't require wholesale
1926  *		changes to drivers that use this feature.
1927  */
1928 void scsi_unblock_requests(struct Scsi_Host *shost)
1929 {
1930 	shost->host_self_blocked = 0;
1931 	scsi_run_host_queues(shost);
1932 }
1933 EXPORT_SYMBOL(scsi_unblock_requests);
1934 
1935 int __init scsi_init_queue(void)
1936 {
1937 	scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1938 					   sizeof(struct scsi_data_buffer),
1939 					   0, 0, NULL);
1940 	if (!scsi_sdb_cache) {
1941 		printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1942 		return -ENOMEM;
1943 	}
1944 
1945 	return 0;
1946 }
1947 
1948 void scsi_exit_queue(void)
1949 {
1950 	kmem_cache_destroy(scsi_sense_cache);
1951 	kmem_cache_destroy(scsi_sense_isadma_cache);
1952 	kmem_cache_destroy(scsi_sdb_cache);
1953 }
1954 
1955 /**
1956  *	scsi_mode_select - issue a mode select
1957  *	@sdev:	SCSI device to be queried
1958  *	@pf:	Page format bit (1 == standard, 0 == vendor specific)
1959  *	@sp:	Save page bit (0 == don't save, 1 == save)
1960  *	@modepage: mode page being requested
1961  *	@buffer: request buffer (may not be smaller than eight bytes)
1962  *	@len:	length of request buffer.
1963  *	@timeout: command timeout
1964  *	@retries: number of retries before failing
1965  *	@data: returns a structure abstracting the mode header data
1966  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
1967  *		must be SCSI_SENSE_BUFFERSIZE big.
1968  *
1969  *	Returns zero if successful; negative error number or scsi
1970  *	status on error
1971  *
1972  */
1973 int
1974 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1975 		 unsigned char *buffer, int len, int timeout, int retries,
1976 		 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1977 {
1978 	unsigned char cmd[10];
1979 	unsigned char *real_buffer;
1980 	int ret;
1981 
1982 	memset(cmd, 0, sizeof(cmd));
1983 	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1984 
1985 	if (sdev->use_10_for_ms) {
1986 		if (len > 65535)
1987 			return -EINVAL;
1988 		real_buffer = kmalloc(8 + len, GFP_KERNEL);
1989 		if (!real_buffer)
1990 			return -ENOMEM;
1991 		memcpy(real_buffer + 8, buffer, len);
1992 		len += 8;
1993 		real_buffer[0] = 0;
1994 		real_buffer[1] = 0;
1995 		real_buffer[2] = data->medium_type;
1996 		real_buffer[3] = data->device_specific;
1997 		real_buffer[4] = data->longlba ? 0x01 : 0;
1998 		real_buffer[5] = 0;
1999 		real_buffer[6] = data->block_descriptor_length >> 8;
2000 		real_buffer[7] = data->block_descriptor_length;
2001 
2002 		cmd[0] = MODE_SELECT_10;
2003 		cmd[7] = len >> 8;
2004 		cmd[8] = len;
2005 	} else {
2006 		if (len > 255 || data->block_descriptor_length > 255 ||
2007 		    data->longlba)
2008 			return -EINVAL;
2009 
2010 		real_buffer = kmalloc(4 + len, GFP_KERNEL);
2011 		if (!real_buffer)
2012 			return -ENOMEM;
2013 		memcpy(real_buffer + 4, buffer, len);
2014 		len += 4;
2015 		real_buffer[0] = 0;
2016 		real_buffer[1] = data->medium_type;
2017 		real_buffer[2] = data->device_specific;
2018 		real_buffer[3] = data->block_descriptor_length;
2019 
2020 
2021 		cmd[0] = MODE_SELECT;
2022 		cmd[4] = len;
2023 	}
2024 
2025 	ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2026 			       sshdr, timeout, retries, NULL);
2027 	kfree(real_buffer);
2028 	return ret;
2029 }
2030 EXPORT_SYMBOL_GPL(scsi_mode_select);
2031 
2032 /**
2033  *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2034  *	@sdev:	SCSI device to be queried
2035  *	@dbd:	set if mode sense will allow block descriptors to be returned
2036  *	@modepage: mode page being requested
2037  *	@buffer: request buffer (may not be smaller than eight bytes)
2038  *	@len:	length of request buffer.
2039  *	@timeout: command timeout
2040  *	@retries: number of retries before failing
2041  *	@data: returns a structure abstracting the mode header data
2042  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2043  *		must be SCSI_SENSE_BUFFERSIZE big.
2044  *
2045  *	Returns zero if unsuccessful, or the header offset (either 4
2046  *	or 8 depending on whether a six or ten byte command was
2047  *	issued) if successful.
2048  */
2049 int
2050 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2051 		  unsigned char *buffer, int len, int timeout, int retries,
2052 		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2053 {
2054 	unsigned char cmd[12];
2055 	int use_10_for_ms;
2056 	int header_length;
2057 	int result, retry_count = retries;
2058 	struct scsi_sense_hdr my_sshdr;
2059 
2060 	memset(data, 0, sizeof(*data));
2061 	memset(&cmd[0], 0, 12);
2062 	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */
2063 	cmd[2] = modepage;
2064 
2065 	/* caller might not be interested in sense, but we need it */
2066 	if (!sshdr)
2067 		sshdr = &my_sshdr;
2068 
2069  retry:
2070 	use_10_for_ms = sdev->use_10_for_ms;
2071 
2072 	if (use_10_for_ms) {
2073 		if (len < 8)
2074 			len = 8;
2075 
2076 		cmd[0] = MODE_SENSE_10;
2077 		cmd[8] = len;
2078 		header_length = 8;
2079 	} else {
2080 		if (len < 4)
2081 			len = 4;
2082 
2083 		cmd[0] = MODE_SENSE;
2084 		cmd[4] = len;
2085 		header_length = 4;
2086 	}
2087 
2088 	memset(buffer, 0, len);
2089 
2090 	result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2091 				  sshdr, timeout, retries, NULL);
2092 
2093 	/* This code looks awful: what it's doing is making sure an
2094 	 * ILLEGAL REQUEST sense return identifies the actual command
2095 	 * byte as the problem.  MODE_SENSE commands can return
2096 	 * ILLEGAL REQUEST if the code page isn't supported */
2097 
2098 	if (use_10_for_ms && !scsi_status_is_good(result) &&
2099 	    driver_byte(result) == DRIVER_SENSE) {
2100 		if (scsi_sense_valid(sshdr)) {
2101 			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2102 			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2103 				/*
2104 				 * Invalid command operation code
2105 				 */
2106 				sdev->use_10_for_ms = 0;
2107 				goto retry;
2108 			}
2109 		}
2110 	}
2111 
2112 	if(scsi_status_is_good(result)) {
2113 		if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2114 			     (modepage == 6 || modepage == 8))) {
2115 			/* Initio breakage? */
2116 			header_length = 0;
2117 			data->length = 13;
2118 			data->medium_type = 0;
2119 			data->device_specific = 0;
2120 			data->longlba = 0;
2121 			data->block_descriptor_length = 0;
2122 		} else if(use_10_for_ms) {
2123 			data->length = buffer[0]*256 + buffer[1] + 2;
2124 			data->medium_type = buffer[2];
2125 			data->device_specific = buffer[3];
2126 			data->longlba = buffer[4] & 0x01;
2127 			data->block_descriptor_length = buffer[6]*256
2128 				+ buffer[7];
2129 		} else {
2130 			data->length = buffer[0] + 1;
2131 			data->medium_type = buffer[1];
2132 			data->device_specific = buffer[2];
2133 			data->block_descriptor_length = buffer[3];
2134 		}
2135 		data->header_length = header_length;
2136 	} else if ((status_byte(result) == CHECK_CONDITION) &&
2137 		   scsi_sense_valid(sshdr) &&
2138 		   sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2139 		retry_count--;
2140 		goto retry;
2141 	}
2142 
2143 	return result;
2144 }
2145 EXPORT_SYMBOL(scsi_mode_sense);
2146 
2147 /**
2148  *	scsi_test_unit_ready - test if unit is ready
2149  *	@sdev:	scsi device to change the state of.
2150  *	@timeout: command timeout
2151  *	@retries: number of retries before failing
2152  *	@sshdr: outpout pointer for decoded sense information.
2153  *
2154  *	Returns zero if unsuccessful or an error if TUR failed.  For
2155  *	removable media, UNIT_ATTENTION sets ->changed flag.
2156  **/
2157 int
2158 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2159 		     struct scsi_sense_hdr *sshdr)
2160 {
2161 	char cmd[] = {
2162 		TEST_UNIT_READY, 0, 0, 0, 0, 0,
2163 	};
2164 	int result;
2165 
2166 	/* try to eat the UNIT_ATTENTION if there are enough retries */
2167 	do {
2168 		result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2169 					  timeout, 1, NULL);
2170 		if (sdev->removable && scsi_sense_valid(sshdr) &&
2171 		    sshdr->sense_key == UNIT_ATTENTION)
2172 			sdev->changed = 1;
2173 	} while (scsi_sense_valid(sshdr) &&
2174 		 sshdr->sense_key == UNIT_ATTENTION && --retries);
2175 
2176 	return result;
2177 }
2178 EXPORT_SYMBOL(scsi_test_unit_ready);
2179 
2180 /**
2181  *	scsi_device_set_state - Take the given device through the device state model.
2182  *	@sdev:	scsi device to change the state of.
2183  *	@state:	state to change to.
2184  *
2185  *	Returns zero if successful or an error if the requested
2186  *	transition is illegal.
2187  */
2188 int
2189 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2190 {
2191 	enum scsi_device_state oldstate = sdev->sdev_state;
2192 
2193 	if (state == oldstate)
2194 		return 0;
2195 
2196 	switch (state) {
2197 	case SDEV_CREATED:
2198 		switch (oldstate) {
2199 		case SDEV_CREATED_BLOCK:
2200 			break;
2201 		default:
2202 			goto illegal;
2203 		}
2204 		break;
2205 
2206 	case SDEV_RUNNING:
2207 		switch (oldstate) {
2208 		case SDEV_CREATED:
2209 		case SDEV_OFFLINE:
2210 		case SDEV_TRANSPORT_OFFLINE:
2211 		case SDEV_QUIESCE:
2212 		case SDEV_BLOCK:
2213 			break;
2214 		default:
2215 			goto illegal;
2216 		}
2217 		break;
2218 
2219 	case SDEV_QUIESCE:
2220 		switch (oldstate) {
2221 		case SDEV_RUNNING:
2222 		case SDEV_OFFLINE:
2223 		case SDEV_TRANSPORT_OFFLINE:
2224 			break;
2225 		default:
2226 			goto illegal;
2227 		}
2228 		break;
2229 
2230 	case SDEV_OFFLINE:
2231 	case SDEV_TRANSPORT_OFFLINE:
2232 		switch (oldstate) {
2233 		case SDEV_CREATED:
2234 		case SDEV_RUNNING:
2235 		case SDEV_QUIESCE:
2236 		case SDEV_BLOCK:
2237 			break;
2238 		default:
2239 			goto illegal;
2240 		}
2241 		break;
2242 
2243 	case SDEV_BLOCK:
2244 		switch (oldstate) {
2245 		case SDEV_RUNNING:
2246 		case SDEV_CREATED_BLOCK:
2247 		case SDEV_OFFLINE:
2248 			break;
2249 		default:
2250 			goto illegal;
2251 		}
2252 		break;
2253 
2254 	case SDEV_CREATED_BLOCK:
2255 		switch (oldstate) {
2256 		case SDEV_CREATED:
2257 			break;
2258 		default:
2259 			goto illegal;
2260 		}
2261 		break;
2262 
2263 	case SDEV_CANCEL:
2264 		switch (oldstate) {
2265 		case SDEV_CREATED:
2266 		case SDEV_RUNNING:
2267 		case SDEV_QUIESCE:
2268 		case SDEV_OFFLINE:
2269 		case SDEV_TRANSPORT_OFFLINE:
2270 			break;
2271 		default:
2272 			goto illegal;
2273 		}
2274 		break;
2275 
2276 	case SDEV_DEL:
2277 		switch (oldstate) {
2278 		case SDEV_CREATED:
2279 		case SDEV_RUNNING:
2280 		case SDEV_OFFLINE:
2281 		case SDEV_TRANSPORT_OFFLINE:
2282 		case SDEV_CANCEL:
2283 		case SDEV_BLOCK:
2284 		case SDEV_CREATED_BLOCK:
2285 			break;
2286 		default:
2287 			goto illegal;
2288 		}
2289 		break;
2290 
2291 	}
2292 	sdev->sdev_state = state;
2293 	return 0;
2294 
2295  illegal:
2296 	SCSI_LOG_ERROR_RECOVERY(1,
2297 				sdev_printk(KERN_ERR, sdev,
2298 					    "Illegal state transition %s->%s",
2299 					    scsi_device_state_name(oldstate),
2300 					    scsi_device_state_name(state))
2301 				);
2302 	return -EINVAL;
2303 }
2304 EXPORT_SYMBOL(scsi_device_set_state);
2305 
2306 /**
2307  * 	sdev_evt_emit - emit a single SCSI device uevent
2308  *	@sdev: associated SCSI device
2309  *	@evt: event to emit
2310  *
2311  *	Send a single uevent (scsi_event) to the associated scsi_device.
2312  */
2313 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2314 {
2315 	int idx = 0;
2316 	char *envp[3];
2317 
2318 	switch (evt->evt_type) {
2319 	case SDEV_EVT_MEDIA_CHANGE:
2320 		envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2321 		break;
2322 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2323 		scsi_rescan_device(&sdev->sdev_gendev);
2324 		envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2325 		break;
2326 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2327 		envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2328 		break;
2329 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2330 	       envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2331 		break;
2332 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2333 		envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2334 		break;
2335 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2336 		envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2337 		break;
2338 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2339 		envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2340 		break;
2341 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2342 		envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2343 		break;
2344 	default:
2345 		/* do nothing */
2346 		break;
2347 	}
2348 
2349 	envp[idx++] = NULL;
2350 
2351 	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2352 }
2353 
2354 /**
2355  * 	sdev_evt_thread - send a uevent for each scsi event
2356  *	@work: work struct for scsi_device
2357  *
2358  *	Dispatch queued events to their associated scsi_device kobjects
2359  *	as uevents.
2360  */
2361 void scsi_evt_thread(struct work_struct *work)
2362 {
2363 	struct scsi_device *sdev;
2364 	enum scsi_device_event evt_type;
2365 	LIST_HEAD(event_list);
2366 
2367 	sdev = container_of(work, struct scsi_device, event_work);
2368 
2369 	for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2370 		if (test_and_clear_bit(evt_type, sdev->pending_events))
2371 			sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2372 
2373 	while (1) {
2374 		struct scsi_event *evt;
2375 		struct list_head *this, *tmp;
2376 		unsigned long flags;
2377 
2378 		spin_lock_irqsave(&sdev->list_lock, flags);
2379 		list_splice_init(&sdev->event_list, &event_list);
2380 		spin_unlock_irqrestore(&sdev->list_lock, flags);
2381 
2382 		if (list_empty(&event_list))
2383 			break;
2384 
2385 		list_for_each_safe(this, tmp, &event_list) {
2386 			evt = list_entry(this, struct scsi_event, node);
2387 			list_del(&evt->node);
2388 			scsi_evt_emit(sdev, evt);
2389 			kfree(evt);
2390 		}
2391 	}
2392 }
2393 
2394 /**
2395  * 	sdev_evt_send - send asserted event to uevent thread
2396  *	@sdev: scsi_device event occurred on
2397  *	@evt: event to send
2398  *
2399  *	Assert scsi device event asynchronously.
2400  */
2401 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2402 {
2403 	unsigned long flags;
2404 
2405 #if 0
2406 	/* FIXME: currently this check eliminates all media change events
2407 	 * for polled devices.  Need to update to discriminate between AN
2408 	 * and polled events */
2409 	if (!test_bit(evt->evt_type, sdev->supported_events)) {
2410 		kfree(evt);
2411 		return;
2412 	}
2413 #endif
2414 
2415 	spin_lock_irqsave(&sdev->list_lock, flags);
2416 	list_add_tail(&evt->node, &sdev->event_list);
2417 	schedule_work(&sdev->event_work);
2418 	spin_unlock_irqrestore(&sdev->list_lock, flags);
2419 }
2420 EXPORT_SYMBOL_GPL(sdev_evt_send);
2421 
2422 /**
2423  * 	sdev_evt_alloc - allocate a new scsi event
2424  *	@evt_type: type of event to allocate
2425  *	@gfpflags: GFP flags for allocation
2426  *
2427  *	Allocates and returns a new scsi_event.
2428  */
2429 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2430 				  gfp_t gfpflags)
2431 {
2432 	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2433 	if (!evt)
2434 		return NULL;
2435 
2436 	evt->evt_type = evt_type;
2437 	INIT_LIST_HEAD(&evt->node);
2438 
2439 	/* evt_type-specific initialization, if any */
2440 	switch (evt_type) {
2441 	case SDEV_EVT_MEDIA_CHANGE:
2442 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2443 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2444 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2445 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2446 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2447 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2448 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2449 	default:
2450 		/* do nothing */
2451 		break;
2452 	}
2453 
2454 	return evt;
2455 }
2456 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2457 
2458 /**
2459  * 	sdev_evt_send_simple - send asserted event to uevent thread
2460  *	@sdev: scsi_device event occurred on
2461  *	@evt_type: type of event to send
2462  *	@gfpflags: GFP flags for allocation
2463  *
2464  *	Assert scsi device event asynchronously, given an event type.
2465  */
2466 void sdev_evt_send_simple(struct scsi_device *sdev,
2467 			  enum scsi_device_event evt_type, gfp_t gfpflags)
2468 {
2469 	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2470 	if (!evt) {
2471 		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2472 			    evt_type);
2473 		return;
2474 	}
2475 
2476 	sdev_evt_send(sdev, evt);
2477 }
2478 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2479 
2480 /**
2481  *	scsi_device_quiesce - Block user issued commands.
2482  *	@sdev:	scsi device to quiesce.
2483  *
2484  *	This works by trying to transition to the SDEV_QUIESCE state
2485  *	(which must be a legal transition).  When the device is in this
2486  *	state, only special requests will be accepted, all others will
2487  *	be deferred.  Since special requests may also be requeued requests,
2488  *	a successful return doesn't guarantee the device will be
2489  *	totally quiescent.
2490  *
2491  *	Must be called with user context, may sleep.
2492  *
2493  *	Returns zero if unsuccessful or an error if not.
2494  */
2495 int
2496 scsi_device_quiesce(struct scsi_device *sdev)
2497 {
2498 	struct request_queue *q = sdev->request_queue;
2499 	int err;
2500 
2501 	/*
2502 	 * It is allowed to call scsi_device_quiesce() multiple times from
2503 	 * the same context but concurrent scsi_device_quiesce() calls are
2504 	 * not allowed.
2505 	 */
2506 	WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2507 
2508 	if (sdev->quiesced_by == current)
2509 		return 0;
2510 
2511 	blk_set_pm_only(q);
2512 
2513 	blk_mq_freeze_queue(q);
2514 	/*
2515 	 * Ensure that the effect of blk_set_pm_only() will be visible
2516 	 * for percpu_ref_tryget() callers that occur after the queue
2517 	 * unfreeze even if the queue was already frozen before this function
2518 	 * was called. See also https://lwn.net/Articles/573497/.
2519 	 */
2520 	synchronize_rcu();
2521 	blk_mq_unfreeze_queue(q);
2522 
2523 	mutex_lock(&sdev->state_mutex);
2524 	err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2525 	if (err == 0)
2526 		sdev->quiesced_by = current;
2527 	else
2528 		blk_clear_pm_only(q);
2529 	mutex_unlock(&sdev->state_mutex);
2530 
2531 	return err;
2532 }
2533 EXPORT_SYMBOL(scsi_device_quiesce);
2534 
2535 /**
2536  *	scsi_device_resume - Restart user issued commands to a quiesced device.
2537  *	@sdev:	scsi device to resume.
2538  *
2539  *	Moves the device from quiesced back to running and restarts the
2540  *	queues.
2541  *
2542  *	Must be called with user context, may sleep.
2543  */
2544 void scsi_device_resume(struct scsi_device *sdev)
2545 {
2546 	/* check if the device state was mutated prior to resume, and if
2547 	 * so assume the state is being managed elsewhere (for example
2548 	 * device deleted during suspend)
2549 	 */
2550 	mutex_lock(&sdev->state_mutex);
2551 	if (sdev->quiesced_by) {
2552 		sdev->quiesced_by = NULL;
2553 		blk_clear_pm_only(sdev->request_queue);
2554 	}
2555 	if (sdev->sdev_state == SDEV_QUIESCE)
2556 		scsi_device_set_state(sdev, SDEV_RUNNING);
2557 	mutex_unlock(&sdev->state_mutex);
2558 }
2559 EXPORT_SYMBOL(scsi_device_resume);
2560 
2561 static void
2562 device_quiesce_fn(struct scsi_device *sdev, void *data)
2563 {
2564 	scsi_device_quiesce(sdev);
2565 }
2566 
2567 void
2568 scsi_target_quiesce(struct scsi_target *starget)
2569 {
2570 	starget_for_each_device(starget, NULL, device_quiesce_fn);
2571 }
2572 EXPORT_SYMBOL(scsi_target_quiesce);
2573 
2574 static void
2575 device_resume_fn(struct scsi_device *sdev, void *data)
2576 {
2577 	scsi_device_resume(sdev);
2578 }
2579 
2580 void
2581 scsi_target_resume(struct scsi_target *starget)
2582 {
2583 	starget_for_each_device(starget, NULL, device_resume_fn);
2584 }
2585 EXPORT_SYMBOL(scsi_target_resume);
2586 
2587 /**
2588  * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2589  * @sdev: device to block
2590  *
2591  * Pause SCSI command processing on the specified device. Does not sleep.
2592  *
2593  * Returns zero if successful or a negative error code upon failure.
2594  *
2595  * Notes:
2596  * This routine transitions the device to the SDEV_BLOCK state (which must be
2597  * a legal transition). When the device is in this state, command processing
2598  * is paused until the device leaves the SDEV_BLOCK state. See also
2599  * scsi_internal_device_unblock_nowait().
2600  */
2601 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2602 {
2603 	struct request_queue *q = sdev->request_queue;
2604 	int err = 0;
2605 
2606 	err = scsi_device_set_state(sdev, SDEV_BLOCK);
2607 	if (err) {
2608 		err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2609 
2610 		if (err)
2611 			return err;
2612 	}
2613 
2614 	/*
2615 	 * The device has transitioned to SDEV_BLOCK.  Stop the
2616 	 * block layer from calling the midlayer with this device's
2617 	 * request queue.
2618 	 */
2619 	blk_mq_quiesce_queue_nowait(q);
2620 	return 0;
2621 }
2622 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2623 
2624 /**
2625  * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2626  * @sdev: device to block
2627  *
2628  * Pause SCSI command processing on the specified device and wait until all
2629  * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2630  *
2631  * Returns zero if successful or a negative error code upon failure.
2632  *
2633  * Note:
2634  * This routine transitions the device to the SDEV_BLOCK state (which must be
2635  * a legal transition). When the device is in this state, command processing
2636  * is paused until the device leaves the SDEV_BLOCK state. See also
2637  * scsi_internal_device_unblock().
2638  *
2639  * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
2640  * scsi_internal_device_block() has blocked a SCSI device and also
2641  * remove the rport mutex lock and unlock calls from srp_queuecommand().
2642  */
2643 static int scsi_internal_device_block(struct scsi_device *sdev)
2644 {
2645 	struct request_queue *q = sdev->request_queue;
2646 	int err;
2647 
2648 	mutex_lock(&sdev->state_mutex);
2649 	err = scsi_internal_device_block_nowait(sdev);
2650 	if (err == 0)
2651 		blk_mq_quiesce_queue(q);
2652 	mutex_unlock(&sdev->state_mutex);
2653 
2654 	return err;
2655 }
2656 
2657 void scsi_start_queue(struct scsi_device *sdev)
2658 {
2659 	struct request_queue *q = sdev->request_queue;
2660 
2661 	blk_mq_unquiesce_queue(q);
2662 }
2663 
2664 /**
2665  * scsi_internal_device_unblock_nowait - resume a device after a block request
2666  * @sdev:	device to resume
2667  * @new_state:	state to set the device to after unblocking
2668  *
2669  * Restart the device queue for a previously suspended SCSI device. Does not
2670  * sleep.
2671  *
2672  * Returns zero if successful or a negative error code upon failure.
2673  *
2674  * Notes:
2675  * This routine transitions the device to the SDEV_RUNNING state or to one of
2676  * the offline states (which must be a legal transition) allowing the midlayer
2677  * to goose the queue for this device.
2678  */
2679 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2680 					enum scsi_device_state new_state)
2681 {
2682 	/*
2683 	 * Try to transition the scsi device to SDEV_RUNNING or one of the
2684 	 * offlined states and goose the device queue if successful.
2685 	 */
2686 	switch (sdev->sdev_state) {
2687 	case SDEV_BLOCK:
2688 	case SDEV_TRANSPORT_OFFLINE:
2689 		sdev->sdev_state = new_state;
2690 		break;
2691 	case SDEV_CREATED_BLOCK:
2692 		if (new_state == SDEV_TRANSPORT_OFFLINE ||
2693 		    new_state == SDEV_OFFLINE)
2694 			sdev->sdev_state = new_state;
2695 		else
2696 			sdev->sdev_state = SDEV_CREATED;
2697 		break;
2698 	case SDEV_CANCEL:
2699 	case SDEV_OFFLINE:
2700 		break;
2701 	default:
2702 		return -EINVAL;
2703 	}
2704 	scsi_start_queue(sdev);
2705 
2706 	return 0;
2707 }
2708 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2709 
2710 /**
2711  * scsi_internal_device_unblock - resume a device after a block request
2712  * @sdev:	device to resume
2713  * @new_state:	state to set the device to after unblocking
2714  *
2715  * Restart the device queue for a previously suspended SCSI device. May sleep.
2716  *
2717  * Returns zero if successful or a negative error code upon failure.
2718  *
2719  * Notes:
2720  * This routine transitions the device to the SDEV_RUNNING state or to one of
2721  * the offline states (which must be a legal transition) allowing the midlayer
2722  * to goose the queue for this device.
2723  */
2724 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2725 					enum scsi_device_state new_state)
2726 {
2727 	int ret;
2728 
2729 	mutex_lock(&sdev->state_mutex);
2730 	ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2731 	mutex_unlock(&sdev->state_mutex);
2732 
2733 	return ret;
2734 }
2735 
2736 static void
2737 device_block(struct scsi_device *sdev, void *data)
2738 {
2739 	scsi_internal_device_block(sdev);
2740 }
2741 
2742 static int
2743 target_block(struct device *dev, void *data)
2744 {
2745 	if (scsi_is_target_device(dev))
2746 		starget_for_each_device(to_scsi_target(dev), NULL,
2747 					device_block);
2748 	return 0;
2749 }
2750 
2751 void
2752 scsi_target_block(struct device *dev)
2753 {
2754 	if (scsi_is_target_device(dev))
2755 		starget_for_each_device(to_scsi_target(dev), NULL,
2756 					device_block);
2757 	else
2758 		device_for_each_child(dev, NULL, target_block);
2759 }
2760 EXPORT_SYMBOL_GPL(scsi_target_block);
2761 
2762 static void
2763 device_unblock(struct scsi_device *sdev, void *data)
2764 {
2765 	scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2766 }
2767 
2768 static int
2769 target_unblock(struct device *dev, void *data)
2770 {
2771 	if (scsi_is_target_device(dev))
2772 		starget_for_each_device(to_scsi_target(dev), data,
2773 					device_unblock);
2774 	return 0;
2775 }
2776 
2777 void
2778 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2779 {
2780 	if (scsi_is_target_device(dev))
2781 		starget_for_each_device(to_scsi_target(dev), &new_state,
2782 					device_unblock);
2783 	else
2784 		device_for_each_child(dev, &new_state, target_unblock);
2785 }
2786 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2787 
2788 /**
2789  * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2790  * @sgl:	scatter-gather list
2791  * @sg_count:	number of segments in sg
2792  * @offset:	offset in bytes into sg, on return offset into the mapped area
2793  * @len:	bytes to map, on return number of bytes mapped
2794  *
2795  * Returns virtual address of the start of the mapped page
2796  */
2797 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2798 			  size_t *offset, size_t *len)
2799 {
2800 	int i;
2801 	size_t sg_len = 0, len_complete = 0;
2802 	struct scatterlist *sg;
2803 	struct page *page;
2804 
2805 	WARN_ON(!irqs_disabled());
2806 
2807 	for_each_sg(sgl, sg, sg_count, i) {
2808 		len_complete = sg_len; /* Complete sg-entries */
2809 		sg_len += sg->length;
2810 		if (sg_len > *offset)
2811 			break;
2812 	}
2813 
2814 	if (unlikely(i == sg_count)) {
2815 		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2816 			"elements %d\n",
2817 		       __func__, sg_len, *offset, sg_count);
2818 		WARN_ON(1);
2819 		return NULL;
2820 	}
2821 
2822 	/* Offset starting from the beginning of first page in this sg-entry */
2823 	*offset = *offset - len_complete + sg->offset;
2824 
2825 	/* Assumption: contiguous pages can be accessed as "page + i" */
2826 	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2827 	*offset &= ~PAGE_MASK;
2828 
2829 	/* Bytes in this sg-entry from *offset to the end of the page */
2830 	sg_len = PAGE_SIZE - *offset;
2831 	if (*len > sg_len)
2832 		*len = sg_len;
2833 
2834 	return kmap_atomic(page);
2835 }
2836 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2837 
2838 /**
2839  * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2840  * @virt:	virtual address to be unmapped
2841  */
2842 void scsi_kunmap_atomic_sg(void *virt)
2843 {
2844 	kunmap_atomic(virt);
2845 }
2846 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2847 
2848 void sdev_disable_disk_events(struct scsi_device *sdev)
2849 {
2850 	atomic_inc(&sdev->disk_events_disable_depth);
2851 }
2852 EXPORT_SYMBOL(sdev_disable_disk_events);
2853 
2854 void sdev_enable_disk_events(struct scsi_device *sdev)
2855 {
2856 	if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2857 		return;
2858 	atomic_dec(&sdev->disk_events_disable_depth);
2859 }
2860 EXPORT_SYMBOL(sdev_enable_disk_events);
2861 
2862 /**
2863  * scsi_vpd_lun_id - return a unique device identification
2864  * @sdev: SCSI device
2865  * @id:   buffer for the identification
2866  * @id_len:  length of the buffer
2867  *
2868  * Copies a unique device identification into @id based
2869  * on the information in the VPD page 0x83 of the device.
2870  * The string will be formatted as a SCSI name string.
2871  *
2872  * Returns the length of the identification or error on failure.
2873  * If the identifier is longer than the supplied buffer the actual
2874  * identifier length is returned and the buffer is not zero-padded.
2875  */
2876 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2877 {
2878 	u8 cur_id_type = 0xff;
2879 	u8 cur_id_size = 0;
2880 	const unsigned char *d, *cur_id_str;
2881 	const struct scsi_vpd *vpd_pg83;
2882 	int id_size = -EINVAL;
2883 
2884 	rcu_read_lock();
2885 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2886 	if (!vpd_pg83) {
2887 		rcu_read_unlock();
2888 		return -ENXIO;
2889 	}
2890 
2891 	/*
2892 	 * Look for the correct descriptor.
2893 	 * Order of preference for lun descriptor:
2894 	 * - SCSI name string
2895 	 * - NAA IEEE Registered Extended
2896 	 * - EUI-64 based 16-byte
2897 	 * - EUI-64 based 12-byte
2898 	 * - NAA IEEE Registered
2899 	 * - NAA IEEE Extended
2900 	 * - T10 Vendor ID
2901 	 * as longer descriptors reduce the likelyhood
2902 	 * of identification clashes.
2903 	 */
2904 
2905 	/* The id string must be at least 20 bytes + terminating NULL byte */
2906 	if (id_len < 21) {
2907 		rcu_read_unlock();
2908 		return -EINVAL;
2909 	}
2910 
2911 	memset(id, 0, id_len);
2912 	d = vpd_pg83->data + 4;
2913 	while (d < vpd_pg83->data + vpd_pg83->len) {
2914 		/* Skip designators not referring to the LUN */
2915 		if ((d[1] & 0x30) != 0x00)
2916 			goto next_desig;
2917 
2918 		switch (d[1] & 0xf) {
2919 		case 0x1:
2920 			/* T10 Vendor ID */
2921 			if (cur_id_size > d[3])
2922 				break;
2923 			/* Prefer anything */
2924 			if (cur_id_type > 0x01 && cur_id_type != 0xff)
2925 				break;
2926 			cur_id_size = d[3];
2927 			if (cur_id_size + 4 > id_len)
2928 				cur_id_size = id_len - 4;
2929 			cur_id_str = d + 4;
2930 			cur_id_type = d[1] & 0xf;
2931 			id_size = snprintf(id, id_len, "t10.%*pE",
2932 					   cur_id_size, cur_id_str);
2933 			break;
2934 		case 0x2:
2935 			/* EUI-64 */
2936 			if (cur_id_size > d[3])
2937 				break;
2938 			/* Prefer NAA IEEE Registered Extended */
2939 			if (cur_id_type == 0x3 &&
2940 			    cur_id_size == d[3])
2941 				break;
2942 			cur_id_size = d[3];
2943 			cur_id_str = d + 4;
2944 			cur_id_type = d[1] & 0xf;
2945 			switch (cur_id_size) {
2946 			case 8:
2947 				id_size = snprintf(id, id_len,
2948 						   "eui.%8phN",
2949 						   cur_id_str);
2950 				break;
2951 			case 12:
2952 				id_size = snprintf(id, id_len,
2953 						   "eui.%12phN",
2954 						   cur_id_str);
2955 				break;
2956 			case 16:
2957 				id_size = snprintf(id, id_len,
2958 						   "eui.%16phN",
2959 						   cur_id_str);
2960 				break;
2961 			default:
2962 				cur_id_size = 0;
2963 				break;
2964 			}
2965 			break;
2966 		case 0x3:
2967 			/* NAA */
2968 			if (cur_id_size > d[3])
2969 				break;
2970 			cur_id_size = d[3];
2971 			cur_id_str = d + 4;
2972 			cur_id_type = d[1] & 0xf;
2973 			switch (cur_id_size) {
2974 			case 8:
2975 				id_size = snprintf(id, id_len,
2976 						   "naa.%8phN",
2977 						   cur_id_str);
2978 				break;
2979 			case 16:
2980 				id_size = snprintf(id, id_len,
2981 						   "naa.%16phN",
2982 						   cur_id_str);
2983 				break;
2984 			default:
2985 				cur_id_size = 0;
2986 				break;
2987 			}
2988 			break;
2989 		case 0x8:
2990 			/* SCSI name string */
2991 			if (cur_id_size + 4 > d[3])
2992 				break;
2993 			/* Prefer others for truncated descriptor */
2994 			if (cur_id_size && d[3] > id_len)
2995 				break;
2996 			cur_id_size = id_size = d[3];
2997 			cur_id_str = d + 4;
2998 			cur_id_type = d[1] & 0xf;
2999 			if (cur_id_size >= id_len)
3000 				cur_id_size = id_len - 1;
3001 			memcpy(id, cur_id_str, cur_id_size);
3002 			/* Decrease priority for truncated descriptor */
3003 			if (cur_id_size != id_size)
3004 				cur_id_size = 6;
3005 			break;
3006 		default:
3007 			break;
3008 		}
3009 next_desig:
3010 		d += d[3] + 4;
3011 	}
3012 	rcu_read_unlock();
3013 
3014 	return id_size;
3015 }
3016 EXPORT_SYMBOL(scsi_vpd_lun_id);
3017 
3018 /*
3019  * scsi_vpd_tpg_id - return a target port group identifier
3020  * @sdev: SCSI device
3021  *
3022  * Returns the Target Port Group identifier from the information
3023  * froom VPD page 0x83 of the device.
3024  *
3025  * Returns the identifier or error on failure.
3026  */
3027 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3028 {
3029 	const unsigned char *d;
3030 	const struct scsi_vpd *vpd_pg83;
3031 	int group_id = -EAGAIN, rel_port = -1;
3032 
3033 	rcu_read_lock();
3034 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3035 	if (!vpd_pg83) {
3036 		rcu_read_unlock();
3037 		return -ENXIO;
3038 	}
3039 
3040 	d = vpd_pg83->data + 4;
3041 	while (d < vpd_pg83->data + vpd_pg83->len) {
3042 		switch (d[1] & 0xf) {
3043 		case 0x4:
3044 			/* Relative target port */
3045 			rel_port = get_unaligned_be16(&d[6]);
3046 			break;
3047 		case 0x5:
3048 			/* Target port group */
3049 			group_id = get_unaligned_be16(&d[6]);
3050 			break;
3051 		default:
3052 			break;
3053 		}
3054 		d += d[3] + 4;
3055 	}
3056 	rcu_read_unlock();
3057 
3058 	if (group_id >= 0 && rel_id && rel_port != -1)
3059 		*rel_id = rel_port;
3060 
3061 	return group_id;
3062 }
3063 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3064