xref: /linux/drivers/scsi/scsi_lib.c (revision 04eeb606a8383b306f4bc6991da8231b5f3924b0)
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/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/init.h>
20 #include <linux/pci.h>
21 #include <linux/delay.h>
22 #include <linux/hardirq.h>
23 #include <linux/scatterlist.h>
24 #include <linux/blk-mq.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 
34 #include <trace/events/scsi.h>
35 
36 #include "scsi_priv.h"
37 #include "scsi_logging.h"
38 
39 
40 #define SG_MEMPOOL_NR		ARRAY_SIZE(scsi_sg_pools)
41 #define SG_MEMPOOL_SIZE		2
42 
43 struct scsi_host_sg_pool {
44 	size_t		size;
45 	char		*name;
46 	struct kmem_cache	*slab;
47 	mempool_t	*pool;
48 };
49 
50 #define SP(x) { x, "sgpool-" __stringify(x) }
51 #if (SCSI_MAX_SG_SEGMENTS < 32)
52 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
53 #endif
54 static struct scsi_host_sg_pool scsi_sg_pools[] = {
55 	SP(8),
56 	SP(16),
57 #if (SCSI_MAX_SG_SEGMENTS > 32)
58 	SP(32),
59 #if (SCSI_MAX_SG_SEGMENTS > 64)
60 	SP(64),
61 #if (SCSI_MAX_SG_SEGMENTS > 128)
62 	SP(128),
63 #if (SCSI_MAX_SG_SEGMENTS > 256)
64 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
65 #endif
66 #endif
67 #endif
68 #endif
69 	SP(SCSI_MAX_SG_SEGMENTS)
70 };
71 #undef SP
72 
73 struct kmem_cache *scsi_sdb_cache;
74 
75 /*
76  * When to reinvoke queueing after a resource shortage. It's 3 msecs to
77  * not change behaviour from the previous unplug mechanism, experimentation
78  * may prove this needs changing.
79  */
80 #define SCSI_QUEUE_DELAY	3
81 
82 static void
83 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
84 {
85 	struct Scsi_Host *host = cmd->device->host;
86 	struct scsi_device *device = cmd->device;
87 	struct scsi_target *starget = scsi_target(device);
88 
89 	/*
90 	 * Set the appropriate busy bit for the device/host.
91 	 *
92 	 * If the host/device isn't busy, assume that something actually
93 	 * completed, and that we should be able to queue a command now.
94 	 *
95 	 * Note that the prior mid-layer assumption that any host could
96 	 * always queue at least one command is now broken.  The mid-layer
97 	 * will implement a user specifiable stall (see
98 	 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
99 	 * if a command is requeued with no other commands outstanding
100 	 * either for the device or for the host.
101 	 */
102 	switch (reason) {
103 	case SCSI_MLQUEUE_HOST_BUSY:
104 		atomic_set(&host->host_blocked, host->max_host_blocked);
105 		break;
106 	case SCSI_MLQUEUE_DEVICE_BUSY:
107 	case SCSI_MLQUEUE_EH_RETRY:
108 		atomic_set(&device->device_blocked,
109 			   device->max_device_blocked);
110 		break;
111 	case SCSI_MLQUEUE_TARGET_BUSY:
112 		atomic_set(&starget->target_blocked,
113 			   starget->max_target_blocked);
114 		break;
115 	}
116 }
117 
118 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
119 {
120 	struct scsi_device *sdev = cmd->device;
121 	struct request_queue *q = cmd->request->q;
122 
123 	blk_mq_requeue_request(cmd->request);
124 	blk_mq_kick_requeue_list(q);
125 	put_device(&sdev->sdev_gendev);
126 }
127 
128 /**
129  * __scsi_queue_insert - private queue insertion
130  * @cmd: The SCSI command being requeued
131  * @reason:  The reason for the requeue
132  * @unbusy: Whether the queue should be unbusied
133  *
134  * This is a private queue insertion.  The public interface
135  * scsi_queue_insert() always assumes the queue should be unbusied
136  * because it's always called before the completion.  This function is
137  * for a requeue after completion, which should only occur in this
138  * file.
139  */
140 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
141 {
142 	struct scsi_device *device = cmd->device;
143 	struct request_queue *q = device->request_queue;
144 	unsigned long flags;
145 
146 	SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
147 		"Inserting command %p into mlqueue\n", cmd));
148 
149 	scsi_set_blocked(cmd, reason);
150 
151 	/*
152 	 * Decrement the counters, since these commands are no longer
153 	 * active on the host/device.
154 	 */
155 	if (unbusy)
156 		scsi_device_unbusy(device);
157 
158 	/*
159 	 * Requeue this command.  It will go before all other commands
160 	 * that are already in the queue. Schedule requeue work under
161 	 * lock such that the kblockd_schedule_work() call happens
162 	 * before blk_cleanup_queue() finishes.
163 	 */
164 	cmd->result = 0;
165 	if (q->mq_ops) {
166 		scsi_mq_requeue_cmd(cmd);
167 		return;
168 	}
169 	spin_lock_irqsave(q->queue_lock, flags);
170 	blk_requeue_request(q, cmd->request);
171 	kblockd_schedule_work(&device->requeue_work);
172 	spin_unlock_irqrestore(q->queue_lock, flags);
173 }
174 
175 /*
176  * Function:    scsi_queue_insert()
177  *
178  * Purpose:     Insert a command in the midlevel queue.
179  *
180  * Arguments:   cmd    - command that we are adding to queue.
181  *              reason - why we are inserting command to queue.
182  *
183  * Lock status: Assumed that lock is not held upon entry.
184  *
185  * Returns:     Nothing.
186  *
187  * Notes:       We do this for one of two cases.  Either the host is busy
188  *              and it cannot accept any more commands for the time being,
189  *              or the device returned QUEUE_FULL and can accept no more
190  *              commands.
191  * Notes:       This could be called either from an interrupt context or a
192  *              normal process context.
193  */
194 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
195 {
196 	__scsi_queue_insert(cmd, reason, 1);
197 }
198 /**
199  * scsi_execute - insert request and wait for the result
200  * @sdev:	scsi device
201  * @cmd:	scsi command
202  * @data_direction: data direction
203  * @buffer:	data buffer
204  * @bufflen:	len of buffer
205  * @sense:	optional sense buffer
206  * @timeout:	request timeout in seconds
207  * @retries:	number of times to retry request
208  * @flags:	or into request flags;
209  * @resid:	optional residual length
210  *
211  * returns the req->errors value which is the scsi_cmnd result
212  * field.
213  */
214 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
215 		 int data_direction, void *buffer, unsigned bufflen,
216 		 unsigned char *sense, int timeout, int retries, u64 flags,
217 		 int *resid)
218 {
219 	struct request *req;
220 	int write = (data_direction == DMA_TO_DEVICE);
221 	int ret = DRIVER_ERROR << 24;
222 
223 	req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
224 	if (IS_ERR(req))
225 		return ret;
226 	blk_rq_set_block_pc(req);
227 
228 	if (bufflen &&	blk_rq_map_kern(sdev->request_queue, req,
229 					buffer, bufflen, __GFP_WAIT))
230 		goto out;
231 
232 	req->cmd_len = COMMAND_SIZE(cmd[0]);
233 	memcpy(req->cmd, cmd, req->cmd_len);
234 	req->sense = sense;
235 	req->sense_len = 0;
236 	req->retries = retries;
237 	req->timeout = timeout;
238 	req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
239 
240 	/*
241 	 * head injection *required* here otherwise quiesce won't work
242 	 */
243 	blk_execute_rq(req->q, NULL, req, 1);
244 
245 	/*
246 	 * Some devices (USB mass-storage in particular) may transfer
247 	 * garbage data together with a residue indicating that the data
248 	 * is invalid.  Prevent the garbage from being misinterpreted
249 	 * and prevent security leaks by zeroing out the excess data.
250 	 */
251 	if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
252 		memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
253 
254 	if (resid)
255 		*resid = req->resid_len;
256 	ret = req->errors;
257  out:
258 	blk_put_request(req);
259 
260 	return ret;
261 }
262 EXPORT_SYMBOL(scsi_execute);
263 
264 int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
265 		     int data_direction, void *buffer, unsigned bufflen,
266 		     struct scsi_sense_hdr *sshdr, int timeout, int retries,
267 		     int *resid, u64 flags)
268 {
269 	char *sense = NULL;
270 	int result;
271 
272 	if (sshdr) {
273 		sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
274 		if (!sense)
275 			return DRIVER_ERROR << 24;
276 	}
277 	result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
278 			      sense, timeout, retries, flags, resid);
279 	if (sshdr)
280 		scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
281 
282 	kfree(sense);
283 	return result;
284 }
285 EXPORT_SYMBOL(scsi_execute_req_flags);
286 
287 /*
288  * Function:    scsi_init_cmd_errh()
289  *
290  * Purpose:     Initialize cmd fields related to error handling.
291  *
292  * Arguments:   cmd	- command that is ready to be queued.
293  *
294  * Notes:       This function has the job of initializing a number of
295  *              fields related to error handling.   Typically this will
296  *              be called once for each command, as required.
297  */
298 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
299 {
300 	cmd->serial_number = 0;
301 	scsi_set_resid(cmd, 0);
302 	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
303 	if (cmd->cmd_len == 0)
304 		cmd->cmd_len = scsi_command_size(cmd->cmnd);
305 }
306 
307 void scsi_device_unbusy(struct scsi_device *sdev)
308 {
309 	struct Scsi_Host *shost = sdev->host;
310 	struct scsi_target *starget = scsi_target(sdev);
311 	unsigned long flags;
312 
313 	atomic_dec(&shost->host_busy);
314 	if (starget->can_queue > 0)
315 		atomic_dec(&starget->target_busy);
316 
317 	if (unlikely(scsi_host_in_recovery(shost) &&
318 		     (shost->host_failed || shost->host_eh_scheduled))) {
319 		spin_lock_irqsave(shost->host_lock, flags);
320 		scsi_eh_wakeup(shost);
321 		spin_unlock_irqrestore(shost->host_lock, flags);
322 	}
323 
324 	atomic_dec(&sdev->device_busy);
325 }
326 
327 static void scsi_kick_queue(struct request_queue *q)
328 {
329 	if (q->mq_ops)
330 		blk_mq_start_hw_queues(q);
331 	else
332 		blk_run_queue(q);
333 }
334 
335 /*
336  * Called for single_lun devices on IO completion. Clear starget_sdev_user,
337  * and call blk_run_queue for all the scsi_devices on the target -
338  * including current_sdev first.
339  *
340  * Called with *no* scsi locks held.
341  */
342 static void scsi_single_lun_run(struct scsi_device *current_sdev)
343 {
344 	struct Scsi_Host *shost = current_sdev->host;
345 	struct scsi_device *sdev, *tmp;
346 	struct scsi_target *starget = scsi_target(current_sdev);
347 	unsigned long flags;
348 
349 	spin_lock_irqsave(shost->host_lock, flags);
350 	starget->starget_sdev_user = NULL;
351 	spin_unlock_irqrestore(shost->host_lock, flags);
352 
353 	/*
354 	 * Call blk_run_queue for all LUNs on the target, starting with
355 	 * current_sdev. We race with others (to set starget_sdev_user),
356 	 * but in most cases, we will be first. Ideally, each LU on the
357 	 * target would get some limited time or requests on the target.
358 	 */
359 	scsi_kick_queue(current_sdev->request_queue);
360 
361 	spin_lock_irqsave(shost->host_lock, flags);
362 	if (starget->starget_sdev_user)
363 		goto out;
364 	list_for_each_entry_safe(sdev, tmp, &starget->devices,
365 			same_target_siblings) {
366 		if (sdev == current_sdev)
367 			continue;
368 		if (scsi_device_get(sdev))
369 			continue;
370 
371 		spin_unlock_irqrestore(shost->host_lock, flags);
372 		scsi_kick_queue(sdev->request_queue);
373 		spin_lock_irqsave(shost->host_lock, flags);
374 
375 		scsi_device_put(sdev);
376 	}
377  out:
378 	spin_unlock_irqrestore(shost->host_lock, flags);
379 }
380 
381 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
382 {
383 	if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
384 		return true;
385 	if (atomic_read(&sdev->device_blocked) > 0)
386 		return true;
387 	return false;
388 }
389 
390 static inline bool scsi_target_is_busy(struct scsi_target *starget)
391 {
392 	if (starget->can_queue > 0) {
393 		if (atomic_read(&starget->target_busy) >= starget->can_queue)
394 			return true;
395 		if (atomic_read(&starget->target_blocked) > 0)
396 			return true;
397 	}
398 	return false;
399 }
400 
401 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
402 {
403 	if (shost->can_queue > 0 &&
404 	    atomic_read(&shost->host_busy) >= shost->can_queue)
405 		return true;
406 	if (atomic_read(&shost->host_blocked) > 0)
407 		return true;
408 	if (shost->host_self_blocked)
409 		return true;
410 	return false;
411 }
412 
413 static void scsi_starved_list_run(struct Scsi_Host *shost)
414 {
415 	LIST_HEAD(starved_list);
416 	struct scsi_device *sdev;
417 	unsigned long flags;
418 
419 	spin_lock_irqsave(shost->host_lock, flags);
420 	list_splice_init(&shost->starved_list, &starved_list);
421 
422 	while (!list_empty(&starved_list)) {
423 		struct request_queue *slq;
424 
425 		/*
426 		 * As long as shost is accepting commands and we have
427 		 * starved queues, call blk_run_queue. scsi_request_fn
428 		 * drops the queue_lock and can add us back to the
429 		 * starved_list.
430 		 *
431 		 * host_lock protects the starved_list and starved_entry.
432 		 * scsi_request_fn must get the host_lock before checking
433 		 * or modifying starved_list or starved_entry.
434 		 */
435 		if (scsi_host_is_busy(shost))
436 			break;
437 
438 		sdev = list_entry(starved_list.next,
439 				  struct scsi_device, starved_entry);
440 		list_del_init(&sdev->starved_entry);
441 		if (scsi_target_is_busy(scsi_target(sdev))) {
442 			list_move_tail(&sdev->starved_entry,
443 				       &shost->starved_list);
444 			continue;
445 		}
446 
447 		/*
448 		 * Once we drop the host lock, a racing scsi_remove_device()
449 		 * call may remove the sdev from the starved list and destroy
450 		 * it and the queue.  Mitigate by taking a reference to the
451 		 * queue and never touching the sdev again after we drop the
452 		 * host lock.  Note: if __scsi_remove_device() invokes
453 		 * blk_cleanup_queue() before the queue is run from this
454 		 * function then blk_run_queue() will return immediately since
455 		 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
456 		 */
457 		slq = sdev->request_queue;
458 		if (!blk_get_queue(slq))
459 			continue;
460 		spin_unlock_irqrestore(shost->host_lock, flags);
461 
462 		scsi_kick_queue(slq);
463 		blk_put_queue(slq);
464 
465 		spin_lock_irqsave(shost->host_lock, flags);
466 	}
467 	/* put any unprocessed entries back */
468 	list_splice(&starved_list, &shost->starved_list);
469 	spin_unlock_irqrestore(shost->host_lock, flags);
470 }
471 
472 /*
473  * Function:   scsi_run_queue()
474  *
475  * Purpose:    Select a proper request queue to serve next
476  *
477  * Arguments:  q       - last request's queue
478  *
479  * Returns:     Nothing
480  *
481  * Notes:      The previous command was completely finished, start
482  *             a new one if possible.
483  */
484 static void scsi_run_queue(struct request_queue *q)
485 {
486 	struct scsi_device *sdev = q->queuedata;
487 
488 	if (scsi_target(sdev)->single_lun)
489 		scsi_single_lun_run(sdev);
490 	if (!list_empty(&sdev->host->starved_list))
491 		scsi_starved_list_run(sdev->host);
492 
493 	if (q->mq_ops)
494 		blk_mq_start_stopped_hw_queues(q, false);
495 	else
496 		blk_run_queue(q);
497 }
498 
499 void scsi_requeue_run_queue(struct work_struct *work)
500 {
501 	struct scsi_device *sdev;
502 	struct request_queue *q;
503 
504 	sdev = container_of(work, struct scsi_device, requeue_work);
505 	q = sdev->request_queue;
506 	scsi_run_queue(q);
507 }
508 
509 /*
510  * Function:	scsi_requeue_command()
511  *
512  * Purpose:	Handle post-processing of completed commands.
513  *
514  * Arguments:	q	- queue to operate on
515  *		cmd	- command that may need to be requeued.
516  *
517  * Returns:	Nothing
518  *
519  * Notes:	After command completion, there may be blocks left
520  *		over which weren't finished by the previous command
521  *		this can be for a number of reasons - the main one is
522  *		I/O errors in the middle of the request, in which case
523  *		we need to request the blocks that come after the bad
524  *		sector.
525  * Notes:	Upon return, cmd is a stale pointer.
526  */
527 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
528 {
529 	struct scsi_device *sdev = cmd->device;
530 	struct request *req = cmd->request;
531 	unsigned long flags;
532 
533 	spin_lock_irqsave(q->queue_lock, flags);
534 	blk_unprep_request(req);
535 	req->special = NULL;
536 	scsi_put_command(cmd);
537 	blk_requeue_request(q, req);
538 	spin_unlock_irqrestore(q->queue_lock, flags);
539 
540 	scsi_run_queue(q);
541 
542 	put_device(&sdev->sdev_gendev);
543 }
544 
545 void scsi_next_command(struct scsi_cmnd *cmd)
546 {
547 	struct scsi_device *sdev = cmd->device;
548 	struct request_queue *q = sdev->request_queue;
549 
550 	scsi_put_command(cmd);
551 	scsi_run_queue(q);
552 
553 	put_device(&sdev->sdev_gendev);
554 }
555 
556 void scsi_run_host_queues(struct Scsi_Host *shost)
557 {
558 	struct scsi_device *sdev;
559 
560 	shost_for_each_device(sdev, shost)
561 		scsi_run_queue(sdev->request_queue);
562 }
563 
564 static inline unsigned int scsi_sgtable_index(unsigned short nents)
565 {
566 	unsigned int index;
567 
568 	BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
569 
570 	if (nents <= 8)
571 		index = 0;
572 	else
573 		index = get_count_order(nents) - 3;
574 
575 	return index;
576 }
577 
578 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
579 {
580 	struct scsi_host_sg_pool *sgp;
581 
582 	sgp = scsi_sg_pools + scsi_sgtable_index(nents);
583 	mempool_free(sgl, sgp->pool);
584 }
585 
586 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
587 {
588 	struct scsi_host_sg_pool *sgp;
589 
590 	sgp = scsi_sg_pools + scsi_sgtable_index(nents);
591 	return mempool_alloc(sgp->pool, gfp_mask);
592 }
593 
594 static void scsi_free_sgtable(struct scsi_data_buffer *sdb, bool mq)
595 {
596 	if (mq && sdb->table.nents <= SCSI_MAX_SG_SEGMENTS)
597 		return;
598 	__sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, mq, scsi_sg_free);
599 }
600 
601 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
602 			      gfp_t gfp_mask, bool mq)
603 {
604 	struct scatterlist *first_chunk = NULL;
605 	int ret;
606 
607 	BUG_ON(!nents);
608 
609 	if (mq) {
610 		if (nents <= SCSI_MAX_SG_SEGMENTS) {
611 			sdb->table.nents = nents;
612 			sg_init_table(sdb->table.sgl, sdb->table.nents);
613 			return 0;
614 		}
615 		first_chunk = sdb->table.sgl;
616 	}
617 
618 	ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
619 			       first_chunk, gfp_mask, scsi_sg_alloc);
620 	if (unlikely(ret))
621 		scsi_free_sgtable(sdb, mq);
622 	return ret;
623 }
624 
625 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
626 {
627 	if (cmd->request->cmd_type == REQ_TYPE_FS) {
628 		struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
629 
630 		if (drv->uninit_command)
631 			drv->uninit_command(cmd);
632 	}
633 }
634 
635 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
636 {
637 	if (cmd->sdb.table.nents)
638 		scsi_free_sgtable(&cmd->sdb, true);
639 	if (cmd->request->next_rq && cmd->request->next_rq->special)
640 		scsi_free_sgtable(cmd->request->next_rq->special, true);
641 	if (scsi_prot_sg_count(cmd))
642 		scsi_free_sgtable(cmd->prot_sdb, true);
643 }
644 
645 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
646 {
647 	struct scsi_device *sdev = cmd->device;
648 	struct Scsi_Host *shost = sdev->host;
649 	unsigned long flags;
650 
651 	scsi_mq_free_sgtables(cmd);
652 	scsi_uninit_cmd(cmd);
653 
654 	if (shost->use_cmd_list) {
655 		BUG_ON(list_empty(&cmd->list));
656 		spin_lock_irqsave(&sdev->list_lock, flags);
657 		list_del_init(&cmd->list);
658 		spin_unlock_irqrestore(&sdev->list_lock, flags);
659 	}
660 }
661 
662 /*
663  * Function:    scsi_release_buffers()
664  *
665  * Purpose:     Free resources allocate for a scsi_command.
666  *
667  * Arguments:   cmd	- command that we are bailing.
668  *
669  * Lock status: Assumed that no lock is held upon entry.
670  *
671  * Returns:     Nothing
672  *
673  * Notes:       In the event that an upper level driver rejects a
674  *		command, we must release resources allocated during
675  *		the __init_io() function.  Primarily this would involve
676  *		the scatter-gather table.
677  */
678 static void scsi_release_buffers(struct scsi_cmnd *cmd)
679 {
680 	if (cmd->sdb.table.nents)
681 		scsi_free_sgtable(&cmd->sdb, false);
682 
683 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
684 
685 	if (scsi_prot_sg_count(cmd))
686 		scsi_free_sgtable(cmd->prot_sdb, false);
687 }
688 
689 static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
690 {
691 	struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
692 
693 	scsi_free_sgtable(bidi_sdb, false);
694 	kmem_cache_free(scsi_sdb_cache, bidi_sdb);
695 	cmd->request->next_rq->special = NULL;
696 }
697 
698 static bool scsi_end_request(struct request *req, int error,
699 		unsigned int bytes, unsigned int bidi_bytes)
700 {
701 	struct scsi_cmnd *cmd = req->special;
702 	struct scsi_device *sdev = cmd->device;
703 	struct request_queue *q = sdev->request_queue;
704 
705 	if (blk_update_request(req, error, bytes))
706 		return true;
707 
708 	/* Bidi request must be completed as a whole */
709 	if (unlikely(bidi_bytes) &&
710 	    blk_update_request(req->next_rq, error, bidi_bytes))
711 		return true;
712 
713 	if (blk_queue_add_random(q))
714 		add_disk_randomness(req->rq_disk);
715 
716 	if (req->mq_ctx) {
717 		/*
718 		 * In the MQ case the command gets freed by __blk_mq_end_request,
719 		 * so we have to do all cleanup that depends on it earlier.
720 		 *
721 		 * We also can't kick the queues from irq context, so we
722 		 * will have to defer it to a workqueue.
723 		 */
724 		scsi_mq_uninit_cmd(cmd);
725 
726 		__blk_mq_end_request(req, error);
727 
728 		if (scsi_target(sdev)->single_lun ||
729 		    !list_empty(&sdev->host->starved_list))
730 			kblockd_schedule_work(&sdev->requeue_work);
731 		else
732 			blk_mq_start_stopped_hw_queues(q, true);
733 
734 		put_device(&sdev->sdev_gendev);
735 	} else {
736 		unsigned long flags;
737 
738 		if (bidi_bytes)
739 			scsi_release_bidi_buffers(cmd);
740 
741 		spin_lock_irqsave(q->queue_lock, flags);
742 		blk_finish_request(req, error);
743 		spin_unlock_irqrestore(q->queue_lock, flags);
744 
745 		scsi_release_buffers(cmd);
746 		scsi_next_command(cmd);
747 	}
748 
749 	return false;
750 }
751 
752 /**
753  * __scsi_error_from_host_byte - translate SCSI error code into errno
754  * @cmd:	SCSI command (unused)
755  * @result:	scsi error code
756  *
757  * Translate SCSI error code into standard UNIX errno.
758  * Return values:
759  * -ENOLINK	temporary transport failure
760  * -EREMOTEIO	permanent target failure, do not retry
761  * -EBADE	permanent nexus failure, retry on other path
762  * -ENOSPC	No write space available
763  * -ENODATA	Medium error
764  * -EIO		unspecified I/O error
765  */
766 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
767 {
768 	int error = 0;
769 
770 	switch(host_byte(result)) {
771 	case DID_TRANSPORT_FAILFAST:
772 		error = -ENOLINK;
773 		break;
774 	case DID_TARGET_FAILURE:
775 		set_host_byte(cmd, DID_OK);
776 		error = -EREMOTEIO;
777 		break;
778 	case DID_NEXUS_FAILURE:
779 		set_host_byte(cmd, DID_OK);
780 		error = -EBADE;
781 		break;
782 	case DID_ALLOC_FAILURE:
783 		set_host_byte(cmd, DID_OK);
784 		error = -ENOSPC;
785 		break;
786 	case DID_MEDIUM_ERROR:
787 		set_host_byte(cmd, DID_OK);
788 		error = -ENODATA;
789 		break;
790 	default:
791 		error = -EIO;
792 		break;
793 	}
794 
795 	return error;
796 }
797 
798 /*
799  * Function:    scsi_io_completion()
800  *
801  * Purpose:     Completion processing for block device I/O requests.
802  *
803  * Arguments:   cmd   - command that is finished.
804  *
805  * Lock status: Assumed that no lock is held upon entry.
806  *
807  * Returns:     Nothing
808  *
809  * Notes:       We will finish off the specified number of sectors.  If we
810  *		are done, the command block will be released and the queue
811  *		function will be goosed.  If we are not done then we have to
812  *		figure out what to do next:
813  *
814  *		a) We can call scsi_requeue_command().  The request
815  *		   will be unprepared and put back on the queue.  Then
816  *		   a new command will be created for it.  This should
817  *		   be used if we made forward progress, or if we want
818  *		   to switch from READ(10) to READ(6) for example.
819  *
820  *		b) We can call __scsi_queue_insert().  The request will
821  *		   be put back on the queue and retried using the same
822  *		   command as before, possibly after a delay.
823  *
824  *		c) We can call scsi_end_request() with -EIO to fail
825  *		   the remainder of the request.
826  */
827 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
828 {
829 	int result = cmd->result;
830 	struct request_queue *q = cmd->device->request_queue;
831 	struct request *req = cmd->request;
832 	int error = 0;
833 	struct scsi_sense_hdr sshdr;
834 	int sense_valid = 0;
835 	int sense_deferred = 0;
836 	enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
837 	      ACTION_DELAYED_RETRY} action;
838 	unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
839 
840 	if (result) {
841 		sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
842 		if (sense_valid)
843 			sense_deferred = scsi_sense_is_deferred(&sshdr);
844 	}
845 
846 	if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
847 		if (result) {
848 			if (sense_valid && req->sense) {
849 				/*
850 				 * SG_IO wants current and deferred errors
851 				 */
852 				int len = 8 + cmd->sense_buffer[7];
853 
854 				if (len > SCSI_SENSE_BUFFERSIZE)
855 					len = SCSI_SENSE_BUFFERSIZE;
856 				memcpy(req->sense, cmd->sense_buffer,  len);
857 				req->sense_len = len;
858 			}
859 			if (!sense_deferred)
860 				error = __scsi_error_from_host_byte(cmd, result);
861 		}
862 		/*
863 		 * __scsi_error_from_host_byte may have reset the host_byte
864 		 */
865 		req->errors = cmd->result;
866 
867 		req->resid_len = scsi_get_resid(cmd);
868 
869 		if (scsi_bidi_cmnd(cmd)) {
870 			/*
871 			 * Bidi commands Must be complete as a whole,
872 			 * both sides at once.
873 			 */
874 			req->next_rq->resid_len = scsi_in(cmd)->resid;
875 			if (scsi_end_request(req, 0, blk_rq_bytes(req),
876 					blk_rq_bytes(req->next_rq)))
877 				BUG();
878 			return;
879 		}
880 	} else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
881 		/*
882 		 * Certain non BLOCK_PC requests are commands that don't
883 		 * actually transfer anything (FLUSH), so cannot use
884 		 * good_bytes != blk_rq_bytes(req) as the signal for an error.
885 		 * This sets the error explicitly for the problem case.
886 		 */
887 		error = __scsi_error_from_host_byte(cmd, result);
888 	}
889 
890 	/* no bidi support for !REQ_TYPE_BLOCK_PC yet */
891 	BUG_ON(blk_bidi_rq(req));
892 
893 	/*
894 	 * Next deal with any sectors which we were able to correctly
895 	 * handle.
896 	 */
897 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
898 		"%u sectors total, %d bytes done.\n",
899 		blk_rq_sectors(req), good_bytes));
900 
901 	/*
902 	 * Recovered errors need reporting, but they're always treated
903 	 * as success, so fiddle the result code here.  For BLOCK_PC
904 	 * we already took a copy of the original into rq->errors which
905 	 * is what gets returned to the user
906 	 */
907 	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
908 		/* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
909 		 * print since caller wants ATA registers. Only occurs on
910 		 * SCSI ATA PASS_THROUGH commands when CK_COND=1
911 		 */
912 		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
913 			;
914 		else if (!(req->cmd_flags & REQ_QUIET))
915 			scsi_print_sense("", cmd);
916 		result = 0;
917 		/* BLOCK_PC may have set error */
918 		error = 0;
919 	}
920 
921 	/*
922 	 * If we finished all bytes in the request we are done now.
923 	 */
924 	if (!scsi_end_request(req, error, good_bytes, 0))
925 		return;
926 
927 	/*
928 	 * Kill remainder if no retrys.
929 	 */
930 	if (error && scsi_noretry_cmd(cmd)) {
931 		if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
932 			BUG();
933 		return;
934 	}
935 
936 	/*
937 	 * If there had been no error, but we have leftover bytes in the
938 	 * requeues just queue the command up again.
939 	 */
940 	if (result == 0)
941 		goto requeue;
942 
943 	error = __scsi_error_from_host_byte(cmd, result);
944 
945 	if (host_byte(result) == DID_RESET) {
946 		/* Third party bus reset or reset for error recovery
947 		 * reasons.  Just retry the command and see what
948 		 * happens.
949 		 */
950 		action = ACTION_RETRY;
951 	} else if (sense_valid && !sense_deferred) {
952 		switch (sshdr.sense_key) {
953 		case UNIT_ATTENTION:
954 			if (cmd->device->removable) {
955 				/* Detected disc change.  Set a bit
956 				 * and quietly refuse further access.
957 				 */
958 				cmd->device->changed = 1;
959 				action = ACTION_FAIL;
960 			} else {
961 				/* Must have been a power glitch, or a
962 				 * bus reset.  Could not have been a
963 				 * media change, so we just retry the
964 				 * command and see what happens.
965 				 */
966 				action = ACTION_RETRY;
967 			}
968 			break;
969 		case ILLEGAL_REQUEST:
970 			/* If we had an ILLEGAL REQUEST returned, then
971 			 * we may have performed an unsupported
972 			 * command.  The only thing this should be
973 			 * would be a ten byte read where only a six
974 			 * byte read was supported.  Also, on a system
975 			 * where READ CAPACITY failed, we may have
976 			 * read past the end of the disk.
977 			 */
978 			if ((cmd->device->use_10_for_rw &&
979 			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
980 			    (cmd->cmnd[0] == READ_10 ||
981 			     cmd->cmnd[0] == WRITE_10)) {
982 				/* This will issue a new 6-byte command. */
983 				cmd->device->use_10_for_rw = 0;
984 				action = ACTION_REPREP;
985 			} else if (sshdr.asc == 0x10) /* DIX */ {
986 				action = ACTION_FAIL;
987 				error = -EILSEQ;
988 			/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
989 			} else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
990 				action = ACTION_FAIL;
991 				error = -EREMOTEIO;
992 			} else
993 				action = ACTION_FAIL;
994 			break;
995 		case ABORTED_COMMAND:
996 			action = ACTION_FAIL;
997 			if (sshdr.asc == 0x10) /* DIF */
998 				error = -EILSEQ;
999 			break;
1000 		case NOT_READY:
1001 			/* If the device is in the process of becoming
1002 			 * ready, or has a temporary blockage, retry.
1003 			 */
1004 			if (sshdr.asc == 0x04) {
1005 				switch (sshdr.ascq) {
1006 				case 0x01: /* becoming ready */
1007 				case 0x04: /* format in progress */
1008 				case 0x05: /* rebuild in progress */
1009 				case 0x06: /* recalculation in progress */
1010 				case 0x07: /* operation in progress */
1011 				case 0x08: /* Long write in progress */
1012 				case 0x09: /* self test in progress */
1013 				case 0x14: /* space allocation in progress */
1014 					action = ACTION_DELAYED_RETRY;
1015 					break;
1016 				default:
1017 					action = ACTION_FAIL;
1018 					break;
1019 				}
1020 			} else
1021 				action = ACTION_FAIL;
1022 			break;
1023 		case VOLUME_OVERFLOW:
1024 			/* See SSC3rXX or current. */
1025 			action = ACTION_FAIL;
1026 			break;
1027 		default:
1028 			action = ACTION_FAIL;
1029 			break;
1030 		}
1031 	} else
1032 		action = ACTION_FAIL;
1033 
1034 	if (action != ACTION_FAIL &&
1035 	    time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
1036 		action = ACTION_FAIL;
1037 
1038 	switch (action) {
1039 	case ACTION_FAIL:
1040 		/* Give up and fail the remainder of the request */
1041 		if (!(req->cmd_flags & REQ_QUIET)) {
1042 			scsi_print_result(cmd);
1043 			if (driver_byte(result) & DRIVER_SENSE)
1044 				scsi_print_sense("", cmd);
1045 			scsi_print_command(cmd);
1046 		}
1047 		if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
1048 			return;
1049 		/*FALLTHRU*/
1050 	case ACTION_REPREP:
1051 	requeue:
1052 		/* Unprep the request and put it back at the head of the queue.
1053 		 * A new command will be prepared and issued.
1054 		 */
1055 		if (q->mq_ops) {
1056 			cmd->request->cmd_flags &= ~REQ_DONTPREP;
1057 			scsi_mq_uninit_cmd(cmd);
1058 			scsi_mq_requeue_cmd(cmd);
1059 		} else {
1060 			scsi_release_buffers(cmd);
1061 			scsi_requeue_command(q, cmd);
1062 		}
1063 		break;
1064 	case ACTION_RETRY:
1065 		/* Retry the same command immediately */
1066 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1067 		break;
1068 	case ACTION_DELAYED_RETRY:
1069 		/* Retry the same command after a delay */
1070 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1071 		break;
1072 	}
1073 }
1074 
1075 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
1076 			     gfp_t gfp_mask)
1077 {
1078 	int count;
1079 
1080 	/*
1081 	 * If sg table allocation fails, requeue request later.
1082 	 */
1083 	if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1084 					gfp_mask, req->mq_ctx != NULL)))
1085 		return BLKPREP_DEFER;
1086 
1087 	/*
1088 	 * Next, walk the list, and fill in the addresses and sizes of
1089 	 * each segment.
1090 	 */
1091 	count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1092 	BUG_ON(count > sdb->table.nents);
1093 	sdb->table.nents = count;
1094 	sdb->length = blk_rq_bytes(req);
1095 	return BLKPREP_OK;
1096 }
1097 
1098 /*
1099  * Function:    scsi_init_io()
1100  *
1101  * Purpose:     SCSI I/O initialize function.
1102  *
1103  * Arguments:   cmd   - Command descriptor we wish to initialize
1104  *
1105  * Returns:     0 on success
1106  *		BLKPREP_DEFER if the failure is retryable
1107  *		BLKPREP_KILL if the failure is fatal
1108  */
1109 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1110 {
1111 	struct scsi_device *sdev = cmd->device;
1112 	struct request *rq = cmd->request;
1113 	bool is_mq = (rq->mq_ctx != NULL);
1114 	int error;
1115 
1116 	BUG_ON(!rq->nr_phys_segments);
1117 
1118 	error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
1119 	if (error)
1120 		goto err_exit;
1121 
1122 	if (blk_bidi_rq(rq)) {
1123 		if (!rq->q->mq_ops) {
1124 			struct scsi_data_buffer *bidi_sdb =
1125 				kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1126 			if (!bidi_sdb) {
1127 				error = BLKPREP_DEFER;
1128 				goto err_exit;
1129 			}
1130 
1131 			rq->next_rq->special = bidi_sdb;
1132 		}
1133 
1134 		error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special,
1135 					  GFP_ATOMIC);
1136 		if (error)
1137 			goto err_exit;
1138 	}
1139 
1140 	if (blk_integrity_rq(rq)) {
1141 		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1142 		int ivecs, count;
1143 
1144 		BUG_ON(prot_sdb == NULL);
1145 		ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1146 
1147 		if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask, is_mq)) {
1148 			error = BLKPREP_DEFER;
1149 			goto err_exit;
1150 		}
1151 
1152 		count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1153 						prot_sdb->table.sgl);
1154 		BUG_ON(unlikely(count > ivecs));
1155 		BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1156 
1157 		cmd->prot_sdb = prot_sdb;
1158 		cmd->prot_sdb->table.nents = count;
1159 	}
1160 
1161 	return BLKPREP_OK;
1162 err_exit:
1163 	if (is_mq) {
1164 		scsi_mq_free_sgtables(cmd);
1165 	} else {
1166 		scsi_release_buffers(cmd);
1167 		cmd->request->special = NULL;
1168 		scsi_put_command(cmd);
1169 		put_device(&sdev->sdev_gendev);
1170 	}
1171 	return error;
1172 }
1173 EXPORT_SYMBOL(scsi_init_io);
1174 
1175 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1176 		struct request *req)
1177 {
1178 	struct scsi_cmnd *cmd;
1179 
1180 	if (!req->special) {
1181 		/* Bail if we can't get a reference to the device */
1182 		if (!get_device(&sdev->sdev_gendev))
1183 			return NULL;
1184 
1185 		cmd = scsi_get_command(sdev, GFP_ATOMIC);
1186 		if (unlikely(!cmd)) {
1187 			put_device(&sdev->sdev_gendev);
1188 			return NULL;
1189 		}
1190 		req->special = cmd;
1191 	} else {
1192 		cmd = req->special;
1193 	}
1194 
1195 	/* pull a tag out of the request if we have one */
1196 	cmd->tag = req->tag;
1197 	cmd->request = req;
1198 
1199 	cmd->cmnd = req->cmd;
1200 	cmd->prot_op = SCSI_PROT_NORMAL;
1201 
1202 	return cmd;
1203 }
1204 
1205 static int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1206 {
1207 	struct scsi_cmnd *cmd = req->special;
1208 
1209 	/*
1210 	 * BLOCK_PC requests may transfer data, in which case they must
1211 	 * a bio attached to them.  Or they might contain a SCSI command
1212 	 * that does not transfer data, in which case they may optionally
1213 	 * submit a request without an attached bio.
1214 	 */
1215 	if (req->bio) {
1216 		int ret = scsi_init_io(cmd, GFP_ATOMIC);
1217 		if (unlikely(ret))
1218 			return ret;
1219 	} else {
1220 		BUG_ON(blk_rq_bytes(req));
1221 
1222 		memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1223 	}
1224 
1225 	cmd->cmd_len = req->cmd_len;
1226 	cmd->transfersize = blk_rq_bytes(req);
1227 	cmd->allowed = req->retries;
1228 	return BLKPREP_OK;
1229 }
1230 
1231 /*
1232  * Setup a REQ_TYPE_FS command.  These are simple request from filesystems
1233  * that still need to be translated to SCSI CDBs from the ULD.
1234  */
1235 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1236 {
1237 	struct scsi_cmnd *cmd = req->special;
1238 
1239 	if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1240 			 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1241 		int ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1242 		if (ret != BLKPREP_OK)
1243 			return ret;
1244 	}
1245 
1246 	memset(cmd->cmnd, 0, BLK_MAX_CDB);
1247 	return scsi_cmd_to_driver(cmd)->init_command(cmd);
1248 }
1249 
1250 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1251 {
1252 	struct scsi_cmnd *cmd = req->special;
1253 
1254 	if (!blk_rq_bytes(req))
1255 		cmd->sc_data_direction = DMA_NONE;
1256 	else if (rq_data_dir(req) == WRITE)
1257 		cmd->sc_data_direction = DMA_TO_DEVICE;
1258 	else
1259 		cmd->sc_data_direction = DMA_FROM_DEVICE;
1260 
1261 	switch (req->cmd_type) {
1262 	case REQ_TYPE_FS:
1263 		return scsi_setup_fs_cmnd(sdev, req);
1264 	case REQ_TYPE_BLOCK_PC:
1265 		return scsi_setup_blk_pc_cmnd(sdev, req);
1266 	default:
1267 		return BLKPREP_KILL;
1268 	}
1269 }
1270 
1271 static int
1272 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1273 {
1274 	int ret = BLKPREP_OK;
1275 
1276 	/*
1277 	 * If the device is not in running state we will reject some
1278 	 * or all commands.
1279 	 */
1280 	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1281 		switch (sdev->sdev_state) {
1282 		case SDEV_OFFLINE:
1283 		case SDEV_TRANSPORT_OFFLINE:
1284 			/*
1285 			 * If the device is offline we refuse to process any
1286 			 * commands.  The device must be brought online
1287 			 * before trying any recovery commands.
1288 			 */
1289 			sdev_printk(KERN_ERR, sdev,
1290 				    "rejecting I/O to offline device\n");
1291 			ret = BLKPREP_KILL;
1292 			break;
1293 		case SDEV_DEL:
1294 			/*
1295 			 * If the device is fully deleted, we refuse to
1296 			 * process any commands as well.
1297 			 */
1298 			sdev_printk(KERN_ERR, sdev,
1299 				    "rejecting I/O to dead device\n");
1300 			ret = BLKPREP_KILL;
1301 			break;
1302 		case SDEV_QUIESCE:
1303 		case SDEV_BLOCK:
1304 		case SDEV_CREATED_BLOCK:
1305 			/*
1306 			 * If the devices is blocked we defer normal commands.
1307 			 */
1308 			if (!(req->cmd_flags & REQ_PREEMPT))
1309 				ret = BLKPREP_DEFER;
1310 			break;
1311 		default:
1312 			/*
1313 			 * For any other not fully online state we only allow
1314 			 * special commands.  In particular any user initiated
1315 			 * command is not allowed.
1316 			 */
1317 			if (!(req->cmd_flags & REQ_PREEMPT))
1318 				ret = BLKPREP_KILL;
1319 			break;
1320 		}
1321 	}
1322 	return ret;
1323 }
1324 
1325 static int
1326 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1327 {
1328 	struct scsi_device *sdev = q->queuedata;
1329 
1330 	switch (ret) {
1331 	case BLKPREP_KILL:
1332 		req->errors = DID_NO_CONNECT << 16;
1333 		/* release the command and kill it */
1334 		if (req->special) {
1335 			struct scsi_cmnd *cmd = req->special;
1336 			scsi_release_buffers(cmd);
1337 			scsi_put_command(cmd);
1338 			put_device(&sdev->sdev_gendev);
1339 			req->special = NULL;
1340 		}
1341 		break;
1342 	case BLKPREP_DEFER:
1343 		/*
1344 		 * If we defer, the blk_peek_request() returns NULL, but the
1345 		 * queue must be restarted, so we schedule a callback to happen
1346 		 * shortly.
1347 		 */
1348 		if (atomic_read(&sdev->device_busy) == 0)
1349 			blk_delay_queue(q, SCSI_QUEUE_DELAY);
1350 		break;
1351 	default:
1352 		req->cmd_flags |= REQ_DONTPREP;
1353 	}
1354 
1355 	return ret;
1356 }
1357 
1358 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1359 {
1360 	struct scsi_device *sdev = q->queuedata;
1361 	struct scsi_cmnd *cmd;
1362 	int ret;
1363 
1364 	ret = scsi_prep_state_check(sdev, req);
1365 	if (ret != BLKPREP_OK)
1366 		goto out;
1367 
1368 	cmd = scsi_get_cmd_from_req(sdev, req);
1369 	if (unlikely(!cmd)) {
1370 		ret = BLKPREP_DEFER;
1371 		goto out;
1372 	}
1373 
1374 	ret = scsi_setup_cmnd(sdev, req);
1375 out:
1376 	return scsi_prep_return(q, req, ret);
1377 }
1378 
1379 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1380 {
1381 	scsi_uninit_cmd(req->special);
1382 }
1383 
1384 /*
1385  * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1386  * return 0.
1387  *
1388  * Called with the queue_lock held.
1389  */
1390 static inline int scsi_dev_queue_ready(struct request_queue *q,
1391 				  struct scsi_device *sdev)
1392 {
1393 	unsigned int busy;
1394 
1395 	busy = atomic_inc_return(&sdev->device_busy) - 1;
1396 	if (atomic_read(&sdev->device_blocked)) {
1397 		if (busy)
1398 			goto out_dec;
1399 
1400 		/*
1401 		 * unblock after device_blocked iterates to zero
1402 		 */
1403 		if (atomic_dec_return(&sdev->device_blocked) > 0) {
1404 			/*
1405 			 * For the MQ case we take care of this in the caller.
1406 			 */
1407 			if (!q->mq_ops)
1408 				blk_delay_queue(q, SCSI_QUEUE_DELAY);
1409 			goto out_dec;
1410 		}
1411 		SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1412 				   "unblocking device at zero depth\n"));
1413 	}
1414 
1415 	if (busy >= sdev->queue_depth)
1416 		goto out_dec;
1417 
1418 	return 1;
1419 out_dec:
1420 	atomic_dec(&sdev->device_busy);
1421 	return 0;
1422 }
1423 
1424 /*
1425  * scsi_target_queue_ready: checks if there we can send commands to target
1426  * @sdev: scsi device on starget to check.
1427  */
1428 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1429 					   struct scsi_device *sdev)
1430 {
1431 	struct scsi_target *starget = scsi_target(sdev);
1432 	unsigned int busy;
1433 
1434 	if (starget->single_lun) {
1435 		spin_lock_irq(shost->host_lock);
1436 		if (starget->starget_sdev_user &&
1437 		    starget->starget_sdev_user != sdev) {
1438 			spin_unlock_irq(shost->host_lock);
1439 			return 0;
1440 		}
1441 		starget->starget_sdev_user = sdev;
1442 		spin_unlock_irq(shost->host_lock);
1443 	}
1444 
1445 	if (starget->can_queue <= 0)
1446 		return 1;
1447 
1448 	busy = atomic_inc_return(&starget->target_busy) - 1;
1449 	if (atomic_read(&starget->target_blocked) > 0) {
1450 		if (busy)
1451 			goto starved;
1452 
1453 		/*
1454 		 * unblock after target_blocked iterates to zero
1455 		 */
1456 		if (atomic_dec_return(&starget->target_blocked) > 0)
1457 			goto out_dec;
1458 
1459 		SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1460 				 "unblocking target at zero depth\n"));
1461 	}
1462 
1463 	if (busy >= starget->can_queue)
1464 		goto starved;
1465 
1466 	return 1;
1467 
1468 starved:
1469 	spin_lock_irq(shost->host_lock);
1470 	list_move_tail(&sdev->starved_entry, &shost->starved_list);
1471 	spin_unlock_irq(shost->host_lock);
1472 out_dec:
1473 	if (starget->can_queue > 0)
1474 		atomic_dec(&starget->target_busy);
1475 	return 0;
1476 }
1477 
1478 /*
1479  * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1480  * return 0. We must end up running the queue again whenever 0 is
1481  * returned, else IO can hang.
1482  */
1483 static inline int scsi_host_queue_ready(struct request_queue *q,
1484 				   struct Scsi_Host *shost,
1485 				   struct scsi_device *sdev)
1486 {
1487 	unsigned int busy;
1488 
1489 	if (scsi_host_in_recovery(shost))
1490 		return 0;
1491 
1492 	busy = atomic_inc_return(&shost->host_busy) - 1;
1493 	if (atomic_read(&shost->host_blocked) > 0) {
1494 		if (busy)
1495 			goto starved;
1496 
1497 		/*
1498 		 * unblock after host_blocked iterates to zero
1499 		 */
1500 		if (atomic_dec_return(&shost->host_blocked) > 0)
1501 			goto out_dec;
1502 
1503 		SCSI_LOG_MLQUEUE(3,
1504 			shost_printk(KERN_INFO, shost,
1505 				     "unblocking host at zero depth\n"));
1506 	}
1507 
1508 	if (shost->can_queue > 0 && busy >= shost->can_queue)
1509 		goto starved;
1510 	if (shost->host_self_blocked)
1511 		goto starved;
1512 
1513 	/* We're OK to process the command, so we can't be starved */
1514 	if (!list_empty(&sdev->starved_entry)) {
1515 		spin_lock_irq(shost->host_lock);
1516 		if (!list_empty(&sdev->starved_entry))
1517 			list_del_init(&sdev->starved_entry);
1518 		spin_unlock_irq(shost->host_lock);
1519 	}
1520 
1521 	return 1;
1522 
1523 starved:
1524 	spin_lock_irq(shost->host_lock);
1525 	if (list_empty(&sdev->starved_entry))
1526 		list_add_tail(&sdev->starved_entry, &shost->starved_list);
1527 	spin_unlock_irq(shost->host_lock);
1528 out_dec:
1529 	atomic_dec(&shost->host_busy);
1530 	return 0;
1531 }
1532 
1533 /*
1534  * Busy state exporting function for request stacking drivers.
1535  *
1536  * For efficiency, no lock is taken to check the busy state of
1537  * shost/starget/sdev, since the returned value is not guaranteed and
1538  * may be changed after request stacking drivers call the function,
1539  * regardless of taking lock or not.
1540  *
1541  * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1542  * needs to return 'not busy'. Otherwise, request stacking drivers
1543  * may hold requests forever.
1544  */
1545 static int scsi_lld_busy(struct request_queue *q)
1546 {
1547 	struct scsi_device *sdev = q->queuedata;
1548 	struct Scsi_Host *shost;
1549 
1550 	if (blk_queue_dying(q))
1551 		return 0;
1552 
1553 	shost = sdev->host;
1554 
1555 	/*
1556 	 * Ignore host/starget busy state.
1557 	 * Since block layer does not have a concept of fairness across
1558 	 * multiple queues, congestion of host/starget needs to be handled
1559 	 * in SCSI layer.
1560 	 */
1561 	if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1562 		return 1;
1563 
1564 	return 0;
1565 }
1566 
1567 /*
1568  * Kill a request for a dead device
1569  */
1570 static void scsi_kill_request(struct request *req, struct request_queue *q)
1571 {
1572 	struct scsi_cmnd *cmd = req->special;
1573 	struct scsi_device *sdev;
1574 	struct scsi_target *starget;
1575 	struct Scsi_Host *shost;
1576 
1577 	blk_start_request(req);
1578 
1579 	scmd_printk(KERN_INFO, cmd, "killing request\n");
1580 
1581 	sdev = cmd->device;
1582 	starget = scsi_target(sdev);
1583 	shost = sdev->host;
1584 	scsi_init_cmd_errh(cmd);
1585 	cmd->result = DID_NO_CONNECT << 16;
1586 	atomic_inc(&cmd->device->iorequest_cnt);
1587 
1588 	/*
1589 	 * SCSI request completion path will do scsi_device_unbusy(),
1590 	 * bump busy counts.  To bump the counters, we need to dance
1591 	 * with the locks as normal issue path does.
1592 	 */
1593 	atomic_inc(&sdev->device_busy);
1594 	atomic_inc(&shost->host_busy);
1595 	if (starget->can_queue > 0)
1596 		atomic_inc(&starget->target_busy);
1597 
1598 	blk_complete_request(req);
1599 }
1600 
1601 static void scsi_softirq_done(struct request *rq)
1602 {
1603 	struct scsi_cmnd *cmd = rq->special;
1604 	unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1605 	int disposition;
1606 
1607 	INIT_LIST_HEAD(&cmd->eh_entry);
1608 
1609 	atomic_inc(&cmd->device->iodone_cnt);
1610 	if (cmd->result)
1611 		atomic_inc(&cmd->device->ioerr_cnt);
1612 
1613 	disposition = scsi_decide_disposition(cmd);
1614 	if (disposition != SUCCESS &&
1615 	    time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1616 		sdev_printk(KERN_ERR, cmd->device,
1617 			    "timing out command, waited %lus\n",
1618 			    wait_for/HZ);
1619 		disposition = SUCCESS;
1620 	}
1621 
1622 	scsi_log_completion(cmd, disposition);
1623 
1624 	switch (disposition) {
1625 		case SUCCESS:
1626 			scsi_finish_command(cmd);
1627 			break;
1628 		case NEEDS_RETRY:
1629 			scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1630 			break;
1631 		case ADD_TO_MLQUEUE:
1632 			scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1633 			break;
1634 		default:
1635 			if (!scsi_eh_scmd_add(cmd, 0))
1636 				scsi_finish_command(cmd);
1637 	}
1638 }
1639 
1640 /**
1641  * scsi_done - Invoke completion on finished SCSI command.
1642  * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1643  * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1644  *
1645  * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1646  * which regains ownership of the SCSI command (de facto) from a LLDD, and
1647  * calls blk_complete_request() for further processing.
1648  *
1649  * This function is interrupt context safe.
1650  */
1651 static void scsi_done(struct scsi_cmnd *cmd)
1652 {
1653 	trace_scsi_dispatch_cmd_done(cmd);
1654 	blk_complete_request(cmd->request);
1655 }
1656 
1657 /*
1658  * Function:    scsi_request_fn()
1659  *
1660  * Purpose:     Main strategy routine for SCSI.
1661  *
1662  * Arguments:   q       - Pointer to actual queue.
1663  *
1664  * Returns:     Nothing
1665  *
1666  * Lock status: IO request lock assumed to be held when called.
1667  */
1668 static void scsi_request_fn(struct request_queue *q)
1669 	__releases(q->queue_lock)
1670 	__acquires(q->queue_lock)
1671 {
1672 	struct scsi_device *sdev = q->queuedata;
1673 	struct Scsi_Host *shost;
1674 	struct scsi_cmnd *cmd;
1675 	struct request *req;
1676 
1677 	/*
1678 	 * To start with, we keep looping until the queue is empty, or until
1679 	 * the host is no longer able to accept any more requests.
1680 	 */
1681 	shost = sdev->host;
1682 	for (;;) {
1683 		int rtn;
1684 		/*
1685 		 * get next queueable request.  We do this early to make sure
1686 		 * that the request is fully prepared even if we cannot
1687 		 * accept it.
1688 		 */
1689 		req = blk_peek_request(q);
1690 		if (!req)
1691 			break;
1692 
1693 		if (unlikely(!scsi_device_online(sdev))) {
1694 			sdev_printk(KERN_ERR, sdev,
1695 				    "rejecting I/O to offline device\n");
1696 			scsi_kill_request(req, q);
1697 			continue;
1698 		}
1699 
1700 		if (!scsi_dev_queue_ready(q, sdev))
1701 			break;
1702 
1703 		/*
1704 		 * Remove the request from the request list.
1705 		 */
1706 		if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1707 			blk_start_request(req);
1708 
1709 		spin_unlock_irq(q->queue_lock);
1710 		cmd = req->special;
1711 		if (unlikely(cmd == NULL)) {
1712 			printk(KERN_CRIT "impossible request in %s.\n"
1713 					 "please mail a stack trace to "
1714 					 "linux-scsi@vger.kernel.org\n",
1715 					 __func__);
1716 			blk_dump_rq_flags(req, "foo");
1717 			BUG();
1718 		}
1719 
1720 		/*
1721 		 * We hit this when the driver is using a host wide
1722 		 * tag map. For device level tag maps the queue_depth check
1723 		 * in the device ready fn would prevent us from trying
1724 		 * to allocate a tag. Since the map is a shared host resource
1725 		 * we add the dev to the starved list so it eventually gets
1726 		 * a run when a tag is freed.
1727 		 */
1728 		if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1729 			spin_lock_irq(shost->host_lock);
1730 			if (list_empty(&sdev->starved_entry))
1731 				list_add_tail(&sdev->starved_entry,
1732 					      &shost->starved_list);
1733 			spin_unlock_irq(shost->host_lock);
1734 			goto not_ready;
1735 		}
1736 
1737 		if (!scsi_target_queue_ready(shost, sdev))
1738 			goto not_ready;
1739 
1740 		if (!scsi_host_queue_ready(q, shost, sdev))
1741 			goto host_not_ready;
1742 
1743 		/*
1744 		 * Finally, initialize any error handling parameters, and set up
1745 		 * the timers for timeouts.
1746 		 */
1747 		scsi_init_cmd_errh(cmd);
1748 
1749 		/*
1750 		 * Dispatch the command to the low-level driver.
1751 		 */
1752 		cmd->scsi_done = scsi_done;
1753 		rtn = scsi_dispatch_cmd(cmd);
1754 		if (rtn) {
1755 			scsi_queue_insert(cmd, rtn);
1756 			spin_lock_irq(q->queue_lock);
1757 			goto out_delay;
1758 		}
1759 		spin_lock_irq(q->queue_lock);
1760 	}
1761 
1762 	return;
1763 
1764  host_not_ready:
1765 	if (scsi_target(sdev)->can_queue > 0)
1766 		atomic_dec(&scsi_target(sdev)->target_busy);
1767  not_ready:
1768 	/*
1769 	 * lock q, handle tag, requeue req, and decrement device_busy. We
1770 	 * must return with queue_lock held.
1771 	 *
1772 	 * Decrementing device_busy without checking it is OK, as all such
1773 	 * cases (host limits or settings) should run the queue at some
1774 	 * later time.
1775 	 */
1776 	spin_lock_irq(q->queue_lock);
1777 	blk_requeue_request(q, req);
1778 	atomic_dec(&sdev->device_busy);
1779 out_delay:
1780 	if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
1781 		blk_delay_queue(q, SCSI_QUEUE_DELAY);
1782 }
1783 
1784 static inline int prep_to_mq(int ret)
1785 {
1786 	switch (ret) {
1787 	case BLKPREP_OK:
1788 		return 0;
1789 	case BLKPREP_DEFER:
1790 		return BLK_MQ_RQ_QUEUE_BUSY;
1791 	default:
1792 		return BLK_MQ_RQ_QUEUE_ERROR;
1793 	}
1794 }
1795 
1796 static int scsi_mq_prep_fn(struct request *req)
1797 {
1798 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1799 	struct scsi_device *sdev = req->q->queuedata;
1800 	struct Scsi_Host *shost = sdev->host;
1801 	unsigned char *sense_buf = cmd->sense_buffer;
1802 	struct scatterlist *sg;
1803 
1804 	memset(cmd, 0, sizeof(struct scsi_cmnd));
1805 
1806 	req->special = cmd;
1807 
1808 	cmd->request = req;
1809 	cmd->device = sdev;
1810 	cmd->sense_buffer = sense_buf;
1811 
1812 	cmd->tag = req->tag;
1813 
1814 	cmd->cmnd = req->cmd;
1815 	cmd->prot_op = SCSI_PROT_NORMAL;
1816 
1817 	INIT_LIST_HEAD(&cmd->list);
1818 	INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1819 	cmd->jiffies_at_alloc = jiffies;
1820 
1821 	if (shost->use_cmd_list) {
1822 		spin_lock_irq(&sdev->list_lock);
1823 		list_add_tail(&cmd->list, &sdev->cmd_list);
1824 		spin_unlock_irq(&sdev->list_lock);
1825 	}
1826 
1827 	sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1828 	cmd->sdb.table.sgl = sg;
1829 
1830 	if (scsi_host_get_prot(shost)) {
1831 		cmd->prot_sdb = (void *)sg +
1832 			shost->sg_tablesize * sizeof(struct scatterlist);
1833 		memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1834 
1835 		cmd->prot_sdb->table.sgl =
1836 			(struct scatterlist *)(cmd->prot_sdb + 1);
1837 	}
1838 
1839 	if (blk_bidi_rq(req)) {
1840 		struct request *next_rq = req->next_rq;
1841 		struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1842 
1843 		memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1844 		bidi_sdb->table.sgl =
1845 			(struct scatterlist *)(bidi_sdb + 1);
1846 
1847 		next_rq->special = bidi_sdb;
1848 	}
1849 
1850 	blk_mq_start_request(req);
1851 
1852 	return scsi_setup_cmnd(sdev, req);
1853 }
1854 
1855 static void scsi_mq_done(struct scsi_cmnd *cmd)
1856 {
1857 	trace_scsi_dispatch_cmd_done(cmd);
1858 	blk_mq_complete_request(cmd->request);
1859 }
1860 
1861 static int scsi_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *req,
1862 		bool last)
1863 {
1864 	struct request_queue *q = req->q;
1865 	struct scsi_device *sdev = q->queuedata;
1866 	struct Scsi_Host *shost = sdev->host;
1867 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1868 	int ret;
1869 	int reason;
1870 
1871 	ret = prep_to_mq(scsi_prep_state_check(sdev, req));
1872 	if (ret)
1873 		goto out;
1874 
1875 	ret = BLK_MQ_RQ_QUEUE_BUSY;
1876 	if (!get_device(&sdev->sdev_gendev))
1877 		goto out;
1878 
1879 	if (!scsi_dev_queue_ready(q, sdev))
1880 		goto out_put_device;
1881 	if (!scsi_target_queue_ready(shost, sdev))
1882 		goto out_dec_device_busy;
1883 	if (!scsi_host_queue_ready(q, shost, sdev))
1884 		goto out_dec_target_busy;
1885 
1886 
1887 	if (!(req->cmd_flags & REQ_DONTPREP)) {
1888 		ret = prep_to_mq(scsi_mq_prep_fn(req));
1889 		if (ret)
1890 			goto out_dec_host_busy;
1891 		req->cmd_flags |= REQ_DONTPREP;
1892 	} else {
1893 		blk_mq_start_request(req);
1894 	}
1895 
1896 	scsi_init_cmd_errh(cmd);
1897 	cmd->scsi_done = scsi_mq_done;
1898 
1899 	reason = scsi_dispatch_cmd(cmd);
1900 	if (reason) {
1901 		scsi_set_blocked(cmd, reason);
1902 		ret = BLK_MQ_RQ_QUEUE_BUSY;
1903 		goto out_dec_host_busy;
1904 	}
1905 
1906 	return BLK_MQ_RQ_QUEUE_OK;
1907 
1908 out_dec_host_busy:
1909 	atomic_dec(&shost->host_busy);
1910 out_dec_target_busy:
1911 	if (scsi_target(sdev)->can_queue > 0)
1912 		atomic_dec(&scsi_target(sdev)->target_busy);
1913 out_dec_device_busy:
1914 	atomic_dec(&sdev->device_busy);
1915 out_put_device:
1916 	put_device(&sdev->sdev_gendev);
1917 out:
1918 	switch (ret) {
1919 	case BLK_MQ_RQ_QUEUE_BUSY:
1920 		blk_mq_stop_hw_queue(hctx);
1921 		if (atomic_read(&sdev->device_busy) == 0 &&
1922 		    !scsi_device_blocked(sdev))
1923 			blk_mq_delay_queue(hctx, SCSI_QUEUE_DELAY);
1924 		break;
1925 	case BLK_MQ_RQ_QUEUE_ERROR:
1926 		/*
1927 		 * Make sure to release all allocated ressources when
1928 		 * we hit an error, as we will never see this command
1929 		 * again.
1930 		 */
1931 		if (req->cmd_flags & REQ_DONTPREP)
1932 			scsi_mq_uninit_cmd(cmd);
1933 		break;
1934 	default:
1935 		break;
1936 	}
1937 	return ret;
1938 }
1939 
1940 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1941 		bool reserved)
1942 {
1943 	if (reserved)
1944 		return BLK_EH_RESET_TIMER;
1945 	return scsi_times_out(req);
1946 }
1947 
1948 static int scsi_init_request(void *data, struct request *rq,
1949 		unsigned int hctx_idx, unsigned int request_idx,
1950 		unsigned int numa_node)
1951 {
1952 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1953 
1954 	cmd->sense_buffer = kzalloc_node(SCSI_SENSE_BUFFERSIZE, GFP_KERNEL,
1955 			numa_node);
1956 	if (!cmd->sense_buffer)
1957 		return -ENOMEM;
1958 	return 0;
1959 }
1960 
1961 static void scsi_exit_request(void *data, struct request *rq,
1962 		unsigned int hctx_idx, unsigned int request_idx)
1963 {
1964 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1965 
1966 	kfree(cmd->sense_buffer);
1967 }
1968 
1969 static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1970 {
1971 	struct device *host_dev;
1972 	u64 bounce_limit = 0xffffffff;
1973 
1974 	if (shost->unchecked_isa_dma)
1975 		return BLK_BOUNCE_ISA;
1976 	/*
1977 	 * Platforms with virtual-DMA translation
1978 	 * hardware have no practical limit.
1979 	 */
1980 	if (!PCI_DMA_BUS_IS_PHYS)
1981 		return BLK_BOUNCE_ANY;
1982 
1983 	host_dev = scsi_get_device(shost);
1984 	if (host_dev && host_dev->dma_mask)
1985 		bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
1986 
1987 	return bounce_limit;
1988 }
1989 
1990 static void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1991 {
1992 	struct device *dev = shost->dma_dev;
1993 
1994 	/*
1995 	 * this limit is imposed by hardware restrictions
1996 	 */
1997 	blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1998 					SCSI_MAX_SG_CHAIN_SEGMENTS));
1999 
2000 	if (scsi_host_prot_dma(shost)) {
2001 		shost->sg_prot_tablesize =
2002 			min_not_zero(shost->sg_prot_tablesize,
2003 				     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2004 		BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2005 		blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2006 	}
2007 
2008 	blk_queue_max_hw_sectors(q, shost->max_sectors);
2009 	blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2010 	blk_queue_segment_boundary(q, shost->dma_boundary);
2011 	dma_set_seg_boundary(dev, shost->dma_boundary);
2012 
2013 	blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2014 
2015 	if (!shost->use_clustering)
2016 		q->limits.cluster = 0;
2017 
2018 	/*
2019 	 * set a reasonable default alignment on word boundaries: the
2020 	 * host and device may alter it using
2021 	 * blk_queue_update_dma_alignment() later.
2022 	 */
2023 	blk_queue_dma_alignment(q, 0x03);
2024 }
2025 
2026 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
2027 					 request_fn_proc *request_fn)
2028 {
2029 	struct request_queue *q;
2030 
2031 	q = blk_init_queue(request_fn, NULL);
2032 	if (!q)
2033 		return NULL;
2034 	__scsi_init_queue(shost, q);
2035 	return q;
2036 }
2037 EXPORT_SYMBOL(__scsi_alloc_queue);
2038 
2039 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
2040 {
2041 	struct request_queue *q;
2042 
2043 	q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
2044 	if (!q)
2045 		return NULL;
2046 
2047 	blk_queue_prep_rq(q, scsi_prep_fn);
2048 	blk_queue_unprep_rq(q, scsi_unprep_fn);
2049 	blk_queue_softirq_done(q, scsi_softirq_done);
2050 	blk_queue_rq_timed_out(q, scsi_times_out);
2051 	blk_queue_lld_busy(q, scsi_lld_busy);
2052 	return q;
2053 }
2054 
2055 static struct blk_mq_ops scsi_mq_ops = {
2056 	.map_queue	= blk_mq_map_queue,
2057 	.queue_rq	= scsi_queue_rq,
2058 	.complete	= scsi_softirq_done,
2059 	.timeout	= scsi_timeout,
2060 	.init_request	= scsi_init_request,
2061 	.exit_request	= scsi_exit_request,
2062 };
2063 
2064 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2065 {
2066 	sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2067 	if (IS_ERR(sdev->request_queue))
2068 		return NULL;
2069 
2070 	sdev->request_queue->queuedata = sdev;
2071 	__scsi_init_queue(sdev->host, sdev->request_queue);
2072 	return sdev->request_queue;
2073 }
2074 
2075 int scsi_mq_setup_tags(struct Scsi_Host *shost)
2076 {
2077 	unsigned int cmd_size, sgl_size, tbl_size;
2078 
2079 	tbl_size = shost->sg_tablesize;
2080 	if (tbl_size > SCSI_MAX_SG_SEGMENTS)
2081 		tbl_size = SCSI_MAX_SG_SEGMENTS;
2082 	sgl_size = tbl_size * sizeof(struct scatterlist);
2083 	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2084 	if (scsi_host_get_prot(shost))
2085 		cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2086 
2087 	memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2088 	shost->tag_set.ops = &scsi_mq_ops;
2089 	shost->tag_set.nr_hw_queues = 1;
2090 	shost->tag_set.queue_depth = shost->can_queue;
2091 	shost->tag_set.cmd_size = cmd_size;
2092 	shost->tag_set.numa_node = NUMA_NO_NODE;
2093 	shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2094 	shost->tag_set.driver_data = shost;
2095 
2096 	return blk_mq_alloc_tag_set(&shost->tag_set);
2097 }
2098 
2099 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2100 {
2101 	blk_mq_free_tag_set(&shost->tag_set);
2102 }
2103 
2104 /*
2105  * Function:    scsi_block_requests()
2106  *
2107  * Purpose:     Utility function used by low-level drivers to prevent further
2108  *		commands from being queued to the device.
2109  *
2110  * Arguments:   shost       - Host in question
2111  *
2112  * Returns:     Nothing
2113  *
2114  * Lock status: No locks are assumed held.
2115  *
2116  * Notes:       There is no timer nor any other means by which the requests
2117  *		get unblocked other than the low-level driver calling
2118  *		scsi_unblock_requests().
2119  */
2120 void scsi_block_requests(struct Scsi_Host *shost)
2121 {
2122 	shost->host_self_blocked = 1;
2123 }
2124 EXPORT_SYMBOL(scsi_block_requests);
2125 
2126 /*
2127  * Function:    scsi_unblock_requests()
2128  *
2129  * Purpose:     Utility function used by low-level drivers to allow further
2130  *		commands from being queued to the device.
2131  *
2132  * Arguments:   shost       - Host in question
2133  *
2134  * Returns:     Nothing
2135  *
2136  * Lock status: No locks are assumed held.
2137  *
2138  * Notes:       There is no timer nor any other means by which the requests
2139  *		get unblocked other than the low-level driver calling
2140  *		scsi_unblock_requests().
2141  *
2142  *		This is done as an API function so that changes to the
2143  *		internals of the scsi mid-layer won't require wholesale
2144  *		changes to drivers that use this feature.
2145  */
2146 void scsi_unblock_requests(struct Scsi_Host *shost)
2147 {
2148 	shost->host_self_blocked = 0;
2149 	scsi_run_host_queues(shost);
2150 }
2151 EXPORT_SYMBOL(scsi_unblock_requests);
2152 
2153 int __init scsi_init_queue(void)
2154 {
2155 	int i;
2156 
2157 	scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2158 					   sizeof(struct scsi_data_buffer),
2159 					   0, 0, NULL);
2160 	if (!scsi_sdb_cache) {
2161 		printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2162 		return -ENOMEM;
2163 	}
2164 
2165 	for (i = 0; i < SG_MEMPOOL_NR; i++) {
2166 		struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2167 		int size = sgp->size * sizeof(struct scatterlist);
2168 
2169 		sgp->slab = kmem_cache_create(sgp->name, size, 0,
2170 				SLAB_HWCACHE_ALIGN, NULL);
2171 		if (!sgp->slab) {
2172 			printk(KERN_ERR "SCSI: can't init sg slab %s\n",
2173 					sgp->name);
2174 			goto cleanup_sdb;
2175 		}
2176 
2177 		sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
2178 						     sgp->slab);
2179 		if (!sgp->pool) {
2180 			printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
2181 					sgp->name);
2182 			goto cleanup_sdb;
2183 		}
2184 	}
2185 
2186 	return 0;
2187 
2188 cleanup_sdb:
2189 	for (i = 0; i < SG_MEMPOOL_NR; i++) {
2190 		struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2191 		if (sgp->pool)
2192 			mempool_destroy(sgp->pool);
2193 		if (sgp->slab)
2194 			kmem_cache_destroy(sgp->slab);
2195 	}
2196 	kmem_cache_destroy(scsi_sdb_cache);
2197 
2198 	return -ENOMEM;
2199 }
2200 
2201 void scsi_exit_queue(void)
2202 {
2203 	int i;
2204 
2205 	kmem_cache_destroy(scsi_sdb_cache);
2206 
2207 	for (i = 0; i < SG_MEMPOOL_NR; i++) {
2208 		struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2209 		mempool_destroy(sgp->pool);
2210 		kmem_cache_destroy(sgp->slab);
2211 	}
2212 }
2213 
2214 /**
2215  *	scsi_mode_select - issue a mode select
2216  *	@sdev:	SCSI device to be queried
2217  *	@pf:	Page format bit (1 == standard, 0 == vendor specific)
2218  *	@sp:	Save page bit (0 == don't save, 1 == save)
2219  *	@modepage: mode page being requested
2220  *	@buffer: request buffer (may not be smaller than eight bytes)
2221  *	@len:	length of request buffer.
2222  *	@timeout: command timeout
2223  *	@retries: number of retries before failing
2224  *	@data: returns a structure abstracting the mode header data
2225  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2226  *		must be SCSI_SENSE_BUFFERSIZE big.
2227  *
2228  *	Returns zero if successful; negative error number or scsi
2229  *	status on error
2230  *
2231  */
2232 int
2233 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2234 		 unsigned char *buffer, int len, int timeout, int retries,
2235 		 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2236 {
2237 	unsigned char cmd[10];
2238 	unsigned char *real_buffer;
2239 	int ret;
2240 
2241 	memset(cmd, 0, sizeof(cmd));
2242 	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2243 
2244 	if (sdev->use_10_for_ms) {
2245 		if (len > 65535)
2246 			return -EINVAL;
2247 		real_buffer = kmalloc(8 + len, GFP_KERNEL);
2248 		if (!real_buffer)
2249 			return -ENOMEM;
2250 		memcpy(real_buffer + 8, buffer, len);
2251 		len += 8;
2252 		real_buffer[0] = 0;
2253 		real_buffer[1] = 0;
2254 		real_buffer[2] = data->medium_type;
2255 		real_buffer[3] = data->device_specific;
2256 		real_buffer[4] = data->longlba ? 0x01 : 0;
2257 		real_buffer[5] = 0;
2258 		real_buffer[6] = data->block_descriptor_length >> 8;
2259 		real_buffer[7] = data->block_descriptor_length;
2260 
2261 		cmd[0] = MODE_SELECT_10;
2262 		cmd[7] = len >> 8;
2263 		cmd[8] = len;
2264 	} else {
2265 		if (len > 255 || data->block_descriptor_length > 255 ||
2266 		    data->longlba)
2267 			return -EINVAL;
2268 
2269 		real_buffer = kmalloc(4 + len, GFP_KERNEL);
2270 		if (!real_buffer)
2271 			return -ENOMEM;
2272 		memcpy(real_buffer + 4, buffer, len);
2273 		len += 4;
2274 		real_buffer[0] = 0;
2275 		real_buffer[1] = data->medium_type;
2276 		real_buffer[2] = data->device_specific;
2277 		real_buffer[3] = data->block_descriptor_length;
2278 
2279 
2280 		cmd[0] = MODE_SELECT;
2281 		cmd[4] = len;
2282 	}
2283 
2284 	ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2285 			       sshdr, timeout, retries, NULL);
2286 	kfree(real_buffer);
2287 	return ret;
2288 }
2289 EXPORT_SYMBOL_GPL(scsi_mode_select);
2290 
2291 /**
2292  *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2293  *	@sdev:	SCSI device to be queried
2294  *	@dbd:	set if mode sense will allow block descriptors to be returned
2295  *	@modepage: mode page being requested
2296  *	@buffer: request buffer (may not be smaller than eight bytes)
2297  *	@len:	length of request buffer.
2298  *	@timeout: command timeout
2299  *	@retries: number of retries before failing
2300  *	@data: returns a structure abstracting the mode header data
2301  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2302  *		must be SCSI_SENSE_BUFFERSIZE big.
2303  *
2304  *	Returns zero if unsuccessful, or the header offset (either 4
2305  *	or 8 depending on whether a six or ten byte command was
2306  *	issued) if successful.
2307  */
2308 int
2309 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2310 		  unsigned char *buffer, int len, int timeout, int retries,
2311 		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2312 {
2313 	unsigned char cmd[12];
2314 	int use_10_for_ms;
2315 	int header_length;
2316 	int result;
2317 	struct scsi_sense_hdr my_sshdr;
2318 
2319 	memset(data, 0, sizeof(*data));
2320 	memset(&cmd[0], 0, 12);
2321 	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */
2322 	cmd[2] = modepage;
2323 
2324 	/* caller might not be interested in sense, but we need it */
2325 	if (!sshdr)
2326 		sshdr = &my_sshdr;
2327 
2328  retry:
2329 	use_10_for_ms = sdev->use_10_for_ms;
2330 
2331 	if (use_10_for_ms) {
2332 		if (len < 8)
2333 			len = 8;
2334 
2335 		cmd[0] = MODE_SENSE_10;
2336 		cmd[8] = len;
2337 		header_length = 8;
2338 	} else {
2339 		if (len < 4)
2340 			len = 4;
2341 
2342 		cmd[0] = MODE_SENSE;
2343 		cmd[4] = len;
2344 		header_length = 4;
2345 	}
2346 
2347 	memset(buffer, 0, len);
2348 
2349 	result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2350 				  sshdr, timeout, retries, NULL);
2351 
2352 	/* This code looks awful: what it's doing is making sure an
2353 	 * ILLEGAL REQUEST sense return identifies the actual command
2354 	 * byte as the problem.  MODE_SENSE commands can return
2355 	 * ILLEGAL REQUEST if the code page isn't supported */
2356 
2357 	if (use_10_for_ms && !scsi_status_is_good(result) &&
2358 	    (driver_byte(result) & DRIVER_SENSE)) {
2359 		if (scsi_sense_valid(sshdr)) {
2360 			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2361 			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2362 				/*
2363 				 * Invalid command operation code
2364 				 */
2365 				sdev->use_10_for_ms = 0;
2366 				goto retry;
2367 			}
2368 		}
2369 	}
2370 
2371 	if(scsi_status_is_good(result)) {
2372 		if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2373 			     (modepage == 6 || modepage == 8))) {
2374 			/* Initio breakage? */
2375 			header_length = 0;
2376 			data->length = 13;
2377 			data->medium_type = 0;
2378 			data->device_specific = 0;
2379 			data->longlba = 0;
2380 			data->block_descriptor_length = 0;
2381 		} else if(use_10_for_ms) {
2382 			data->length = buffer[0]*256 + buffer[1] + 2;
2383 			data->medium_type = buffer[2];
2384 			data->device_specific = buffer[3];
2385 			data->longlba = buffer[4] & 0x01;
2386 			data->block_descriptor_length = buffer[6]*256
2387 				+ buffer[7];
2388 		} else {
2389 			data->length = buffer[0] + 1;
2390 			data->medium_type = buffer[1];
2391 			data->device_specific = buffer[2];
2392 			data->block_descriptor_length = buffer[3];
2393 		}
2394 		data->header_length = header_length;
2395 	}
2396 
2397 	return result;
2398 }
2399 EXPORT_SYMBOL(scsi_mode_sense);
2400 
2401 /**
2402  *	scsi_test_unit_ready - test if unit is ready
2403  *	@sdev:	scsi device to change the state of.
2404  *	@timeout: command timeout
2405  *	@retries: number of retries before failing
2406  *	@sshdr_external: Optional pointer to struct scsi_sense_hdr for
2407  *		returning sense. Make sure that this is cleared before passing
2408  *		in.
2409  *
2410  *	Returns zero if unsuccessful or an error if TUR failed.  For
2411  *	removable media, UNIT_ATTENTION sets ->changed flag.
2412  **/
2413 int
2414 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2415 		     struct scsi_sense_hdr *sshdr_external)
2416 {
2417 	char cmd[] = {
2418 		TEST_UNIT_READY, 0, 0, 0, 0, 0,
2419 	};
2420 	struct scsi_sense_hdr *sshdr;
2421 	int result;
2422 
2423 	if (!sshdr_external)
2424 		sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2425 	else
2426 		sshdr = sshdr_external;
2427 
2428 	/* try to eat the UNIT_ATTENTION if there are enough retries */
2429 	do {
2430 		result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2431 					  timeout, retries, NULL);
2432 		if (sdev->removable && scsi_sense_valid(sshdr) &&
2433 		    sshdr->sense_key == UNIT_ATTENTION)
2434 			sdev->changed = 1;
2435 	} while (scsi_sense_valid(sshdr) &&
2436 		 sshdr->sense_key == UNIT_ATTENTION && --retries);
2437 
2438 	if (!sshdr_external)
2439 		kfree(sshdr);
2440 	return result;
2441 }
2442 EXPORT_SYMBOL(scsi_test_unit_ready);
2443 
2444 /**
2445  *	scsi_device_set_state - Take the given device through the device state model.
2446  *	@sdev:	scsi device to change the state of.
2447  *	@state:	state to change to.
2448  *
2449  *	Returns zero if unsuccessful or an error if the requested
2450  *	transition is illegal.
2451  */
2452 int
2453 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2454 {
2455 	enum scsi_device_state oldstate = sdev->sdev_state;
2456 
2457 	if (state == oldstate)
2458 		return 0;
2459 
2460 	switch (state) {
2461 	case SDEV_CREATED:
2462 		switch (oldstate) {
2463 		case SDEV_CREATED_BLOCK:
2464 			break;
2465 		default:
2466 			goto illegal;
2467 		}
2468 		break;
2469 
2470 	case SDEV_RUNNING:
2471 		switch (oldstate) {
2472 		case SDEV_CREATED:
2473 		case SDEV_OFFLINE:
2474 		case SDEV_TRANSPORT_OFFLINE:
2475 		case SDEV_QUIESCE:
2476 		case SDEV_BLOCK:
2477 			break;
2478 		default:
2479 			goto illegal;
2480 		}
2481 		break;
2482 
2483 	case SDEV_QUIESCE:
2484 		switch (oldstate) {
2485 		case SDEV_RUNNING:
2486 		case SDEV_OFFLINE:
2487 		case SDEV_TRANSPORT_OFFLINE:
2488 			break;
2489 		default:
2490 			goto illegal;
2491 		}
2492 		break;
2493 
2494 	case SDEV_OFFLINE:
2495 	case SDEV_TRANSPORT_OFFLINE:
2496 		switch (oldstate) {
2497 		case SDEV_CREATED:
2498 		case SDEV_RUNNING:
2499 		case SDEV_QUIESCE:
2500 		case SDEV_BLOCK:
2501 			break;
2502 		default:
2503 			goto illegal;
2504 		}
2505 		break;
2506 
2507 	case SDEV_BLOCK:
2508 		switch (oldstate) {
2509 		case SDEV_RUNNING:
2510 		case SDEV_CREATED_BLOCK:
2511 			break;
2512 		default:
2513 			goto illegal;
2514 		}
2515 		break;
2516 
2517 	case SDEV_CREATED_BLOCK:
2518 		switch (oldstate) {
2519 		case SDEV_CREATED:
2520 			break;
2521 		default:
2522 			goto illegal;
2523 		}
2524 		break;
2525 
2526 	case SDEV_CANCEL:
2527 		switch (oldstate) {
2528 		case SDEV_CREATED:
2529 		case SDEV_RUNNING:
2530 		case SDEV_QUIESCE:
2531 		case SDEV_OFFLINE:
2532 		case SDEV_TRANSPORT_OFFLINE:
2533 		case SDEV_BLOCK:
2534 			break;
2535 		default:
2536 			goto illegal;
2537 		}
2538 		break;
2539 
2540 	case SDEV_DEL:
2541 		switch (oldstate) {
2542 		case SDEV_CREATED:
2543 		case SDEV_RUNNING:
2544 		case SDEV_OFFLINE:
2545 		case SDEV_TRANSPORT_OFFLINE:
2546 		case SDEV_CANCEL:
2547 		case SDEV_CREATED_BLOCK:
2548 			break;
2549 		default:
2550 			goto illegal;
2551 		}
2552 		break;
2553 
2554 	}
2555 	sdev->sdev_state = state;
2556 	return 0;
2557 
2558  illegal:
2559 	SCSI_LOG_ERROR_RECOVERY(1,
2560 				sdev_printk(KERN_ERR, sdev,
2561 					    "Illegal state transition %s->%s",
2562 					    scsi_device_state_name(oldstate),
2563 					    scsi_device_state_name(state))
2564 				);
2565 	return -EINVAL;
2566 }
2567 EXPORT_SYMBOL(scsi_device_set_state);
2568 
2569 /**
2570  * 	sdev_evt_emit - emit a single SCSI device uevent
2571  *	@sdev: associated SCSI device
2572  *	@evt: event to emit
2573  *
2574  *	Send a single uevent (scsi_event) to the associated scsi_device.
2575  */
2576 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2577 {
2578 	int idx = 0;
2579 	char *envp[3];
2580 
2581 	switch (evt->evt_type) {
2582 	case SDEV_EVT_MEDIA_CHANGE:
2583 		envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2584 		break;
2585 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2586 		envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2587 		break;
2588 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2589 		envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2590 		break;
2591 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2592 	       envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2593 		break;
2594 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2595 		envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2596 		break;
2597 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2598 		envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2599 		break;
2600 	default:
2601 		/* do nothing */
2602 		break;
2603 	}
2604 
2605 	envp[idx++] = NULL;
2606 
2607 	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2608 }
2609 
2610 /**
2611  * 	sdev_evt_thread - send a uevent for each scsi event
2612  *	@work: work struct for scsi_device
2613  *
2614  *	Dispatch queued events to their associated scsi_device kobjects
2615  *	as uevents.
2616  */
2617 void scsi_evt_thread(struct work_struct *work)
2618 {
2619 	struct scsi_device *sdev;
2620 	enum scsi_device_event evt_type;
2621 	LIST_HEAD(event_list);
2622 
2623 	sdev = container_of(work, struct scsi_device, event_work);
2624 
2625 	for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2626 		if (test_and_clear_bit(evt_type, sdev->pending_events))
2627 			sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2628 
2629 	while (1) {
2630 		struct scsi_event *evt;
2631 		struct list_head *this, *tmp;
2632 		unsigned long flags;
2633 
2634 		spin_lock_irqsave(&sdev->list_lock, flags);
2635 		list_splice_init(&sdev->event_list, &event_list);
2636 		spin_unlock_irqrestore(&sdev->list_lock, flags);
2637 
2638 		if (list_empty(&event_list))
2639 			break;
2640 
2641 		list_for_each_safe(this, tmp, &event_list) {
2642 			evt = list_entry(this, struct scsi_event, node);
2643 			list_del(&evt->node);
2644 			scsi_evt_emit(sdev, evt);
2645 			kfree(evt);
2646 		}
2647 	}
2648 }
2649 
2650 /**
2651  * 	sdev_evt_send - send asserted event to uevent thread
2652  *	@sdev: scsi_device event occurred on
2653  *	@evt: event to send
2654  *
2655  *	Assert scsi device event asynchronously.
2656  */
2657 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2658 {
2659 	unsigned long flags;
2660 
2661 #if 0
2662 	/* FIXME: currently this check eliminates all media change events
2663 	 * for polled devices.  Need to update to discriminate between AN
2664 	 * and polled events */
2665 	if (!test_bit(evt->evt_type, sdev->supported_events)) {
2666 		kfree(evt);
2667 		return;
2668 	}
2669 #endif
2670 
2671 	spin_lock_irqsave(&sdev->list_lock, flags);
2672 	list_add_tail(&evt->node, &sdev->event_list);
2673 	schedule_work(&sdev->event_work);
2674 	spin_unlock_irqrestore(&sdev->list_lock, flags);
2675 }
2676 EXPORT_SYMBOL_GPL(sdev_evt_send);
2677 
2678 /**
2679  * 	sdev_evt_alloc - allocate a new scsi event
2680  *	@evt_type: type of event to allocate
2681  *	@gfpflags: GFP flags for allocation
2682  *
2683  *	Allocates and returns a new scsi_event.
2684  */
2685 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2686 				  gfp_t gfpflags)
2687 {
2688 	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2689 	if (!evt)
2690 		return NULL;
2691 
2692 	evt->evt_type = evt_type;
2693 	INIT_LIST_HEAD(&evt->node);
2694 
2695 	/* evt_type-specific initialization, if any */
2696 	switch (evt_type) {
2697 	case SDEV_EVT_MEDIA_CHANGE:
2698 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2699 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2700 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2701 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2702 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2703 	default:
2704 		/* do nothing */
2705 		break;
2706 	}
2707 
2708 	return evt;
2709 }
2710 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2711 
2712 /**
2713  * 	sdev_evt_send_simple - send asserted event to uevent thread
2714  *	@sdev: scsi_device event occurred on
2715  *	@evt_type: type of event to send
2716  *	@gfpflags: GFP flags for allocation
2717  *
2718  *	Assert scsi device event asynchronously, given an event type.
2719  */
2720 void sdev_evt_send_simple(struct scsi_device *sdev,
2721 			  enum scsi_device_event evt_type, gfp_t gfpflags)
2722 {
2723 	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2724 	if (!evt) {
2725 		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2726 			    evt_type);
2727 		return;
2728 	}
2729 
2730 	sdev_evt_send(sdev, evt);
2731 }
2732 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2733 
2734 /**
2735  *	scsi_device_quiesce - Block user issued commands.
2736  *	@sdev:	scsi device to quiesce.
2737  *
2738  *	This works by trying to transition to the SDEV_QUIESCE state
2739  *	(which must be a legal transition).  When the device is in this
2740  *	state, only special requests will be accepted, all others will
2741  *	be deferred.  Since special requests may also be requeued requests,
2742  *	a successful return doesn't guarantee the device will be
2743  *	totally quiescent.
2744  *
2745  *	Must be called with user context, may sleep.
2746  *
2747  *	Returns zero if unsuccessful or an error if not.
2748  */
2749 int
2750 scsi_device_quiesce(struct scsi_device *sdev)
2751 {
2752 	int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2753 	if (err)
2754 		return err;
2755 
2756 	scsi_run_queue(sdev->request_queue);
2757 	while (atomic_read(&sdev->device_busy)) {
2758 		msleep_interruptible(200);
2759 		scsi_run_queue(sdev->request_queue);
2760 	}
2761 	return 0;
2762 }
2763 EXPORT_SYMBOL(scsi_device_quiesce);
2764 
2765 /**
2766  *	scsi_device_resume - Restart user issued commands to a quiesced device.
2767  *	@sdev:	scsi device to resume.
2768  *
2769  *	Moves the device from quiesced back to running and restarts the
2770  *	queues.
2771  *
2772  *	Must be called with user context, may sleep.
2773  */
2774 void scsi_device_resume(struct scsi_device *sdev)
2775 {
2776 	/* check if the device state was mutated prior to resume, and if
2777 	 * so assume the state is being managed elsewhere (for example
2778 	 * device deleted during suspend)
2779 	 */
2780 	if (sdev->sdev_state != SDEV_QUIESCE ||
2781 	    scsi_device_set_state(sdev, SDEV_RUNNING))
2782 		return;
2783 	scsi_run_queue(sdev->request_queue);
2784 }
2785 EXPORT_SYMBOL(scsi_device_resume);
2786 
2787 static void
2788 device_quiesce_fn(struct scsi_device *sdev, void *data)
2789 {
2790 	scsi_device_quiesce(sdev);
2791 }
2792 
2793 void
2794 scsi_target_quiesce(struct scsi_target *starget)
2795 {
2796 	starget_for_each_device(starget, NULL, device_quiesce_fn);
2797 }
2798 EXPORT_SYMBOL(scsi_target_quiesce);
2799 
2800 static void
2801 device_resume_fn(struct scsi_device *sdev, void *data)
2802 {
2803 	scsi_device_resume(sdev);
2804 }
2805 
2806 void
2807 scsi_target_resume(struct scsi_target *starget)
2808 {
2809 	starget_for_each_device(starget, NULL, device_resume_fn);
2810 }
2811 EXPORT_SYMBOL(scsi_target_resume);
2812 
2813 /**
2814  * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2815  * @sdev:	device to block
2816  *
2817  * Block request made by scsi lld's to temporarily stop all
2818  * scsi commands on the specified device.  Called from interrupt
2819  * or normal process context.
2820  *
2821  * Returns zero if successful or error if not
2822  *
2823  * Notes:
2824  *	This routine transitions the device to the SDEV_BLOCK state
2825  *	(which must be a legal transition).  When the device is in this
2826  *	state, all commands are deferred until the scsi lld reenables
2827  *	the device with scsi_device_unblock or device_block_tmo fires.
2828  */
2829 int
2830 scsi_internal_device_block(struct scsi_device *sdev)
2831 {
2832 	struct request_queue *q = sdev->request_queue;
2833 	unsigned long flags;
2834 	int err = 0;
2835 
2836 	err = scsi_device_set_state(sdev, SDEV_BLOCK);
2837 	if (err) {
2838 		err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2839 
2840 		if (err)
2841 			return err;
2842 	}
2843 
2844 	/*
2845 	 * The device has transitioned to SDEV_BLOCK.  Stop the
2846 	 * block layer from calling the midlayer with this device's
2847 	 * request queue.
2848 	 */
2849 	if (q->mq_ops) {
2850 		blk_mq_stop_hw_queues(q);
2851 	} else {
2852 		spin_lock_irqsave(q->queue_lock, flags);
2853 		blk_stop_queue(q);
2854 		spin_unlock_irqrestore(q->queue_lock, flags);
2855 	}
2856 
2857 	return 0;
2858 }
2859 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2860 
2861 /**
2862  * scsi_internal_device_unblock - resume a device after a block request
2863  * @sdev:	device to resume
2864  * @new_state:	state to set devices to after unblocking
2865  *
2866  * Called by scsi lld's or the midlayer to restart the device queue
2867  * for the previously suspended scsi device.  Called from interrupt or
2868  * normal process context.
2869  *
2870  * Returns zero if successful or error if not.
2871  *
2872  * Notes:
2873  *	This routine transitions the device to the SDEV_RUNNING state
2874  *	or to one of the offline states (which must be a legal transition)
2875  *	allowing the midlayer to goose the queue for this device.
2876  */
2877 int
2878 scsi_internal_device_unblock(struct scsi_device *sdev,
2879 			     enum scsi_device_state new_state)
2880 {
2881 	struct request_queue *q = sdev->request_queue;
2882 	unsigned long flags;
2883 
2884 	/*
2885 	 * Try to transition the scsi device to SDEV_RUNNING or one of the
2886 	 * offlined states and goose the device queue if successful.
2887 	 */
2888 	if ((sdev->sdev_state == SDEV_BLOCK) ||
2889 	    (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2890 		sdev->sdev_state = new_state;
2891 	else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2892 		if (new_state == SDEV_TRANSPORT_OFFLINE ||
2893 		    new_state == SDEV_OFFLINE)
2894 			sdev->sdev_state = new_state;
2895 		else
2896 			sdev->sdev_state = SDEV_CREATED;
2897 	} else if (sdev->sdev_state != SDEV_CANCEL &&
2898 		 sdev->sdev_state != SDEV_OFFLINE)
2899 		return -EINVAL;
2900 
2901 	if (q->mq_ops) {
2902 		blk_mq_start_stopped_hw_queues(q, false);
2903 	} else {
2904 		spin_lock_irqsave(q->queue_lock, flags);
2905 		blk_start_queue(q);
2906 		spin_unlock_irqrestore(q->queue_lock, flags);
2907 	}
2908 
2909 	return 0;
2910 }
2911 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2912 
2913 static void
2914 device_block(struct scsi_device *sdev, void *data)
2915 {
2916 	scsi_internal_device_block(sdev);
2917 }
2918 
2919 static int
2920 target_block(struct device *dev, void *data)
2921 {
2922 	if (scsi_is_target_device(dev))
2923 		starget_for_each_device(to_scsi_target(dev), NULL,
2924 					device_block);
2925 	return 0;
2926 }
2927 
2928 void
2929 scsi_target_block(struct device *dev)
2930 {
2931 	if (scsi_is_target_device(dev))
2932 		starget_for_each_device(to_scsi_target(dev), NULL,
2933 					device_block);
2934 	else
2935 		device_for_each_child(dev, NULL, target_block);
2936 }
2937 EXPORT_SYMBOL_GPL(scsi_target_block);
2938 
2939 static void
2940 device_unblock(struct scsi_device *sdev, void *data)
2941 {
2942 	scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2943 }
2944 
2945 static int
2946 target_unblock(struct device *dev, void *data)
2947 {
2948 	if (scsi_is_target_device(dev))
2949 		starget_for_each_device(to_scsi_target(dev), data,
2950 					device_unblock);
2951 	return 0;
2952 }
2953 
2954 void
2955 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2956 {
2957 	if (scsi_is_target_device(dev))
2958 		starget_for_each_device(to_scsi_target(dev), &new_state,
2959 					device_unblock);
2960 	else
2961 		device_for_each_child(dev, &new_state, target_unblock);
2962 }
2963 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2964 
2965 /**
2966  * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2967  * @sgl:	scatter-gather list
2968  * @sg_count:	number of segments in sg
2969  * @offset:	offset in bytes into sg, on return offset into the mapped area
2970  * @len:	bytes to map, on return number of bytes mapped
2971  *
2972  * Returns virtual address of the start of the mapped page
2973  */
2974 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2975 			  size_t *offset, size_t *len)
2976 {
2977 	int i;
2978 	size_t sg_len = 0, len_complete = 0;
2979 	struct scatterlist *sg;
2980 	struct page *page;
2981 
2982 	WARN_ON(!irqs_disabled());
2983 
2984 	for_each_sg(sgl, sg, sg_count, i) {
2985 		len_complete = sg_len; /* Complete sg-entries */
2986 		sg_len += sg->length;
2987 		if (sg_len > *offset)
2988 			break;
2989 	}
2990 
2991 	if (unlikely(i == sg_count)) {
2992 		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2993 			"elements %d\n",
2994 		       __func__, sg_len, *offset, sg_count);
2995 		WARN_ON(1);
2996 		return NULL;
2997 	}
2998 
2999 	/* Offset starting from the beginning of first page in this sg-entry */
3000 	*offset = *offset - len_complete + sg->offset;
3001 
3002 	/* Assumption: contiguous pages can be accessed as "page + i" */
3003 	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3004 	*offset &= ~PAGE_MASK;
3005 
3006 	/* Bytes in this sg-entry from *offset to the end of the page */
3007 	sg_len = PAGE_SIZE - *offset;
3008 	if (*len > sg_len)
3009 		*len = sg_len;
3010 
3011 	return kmap_atomic(page);
3012 }
3013 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3014 
3015 /**
3016  * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3017  * @virt:	virtual address to be unmapped
3018  */
3019 void scsi_kunmap_atomic_sg(void *virt)
3020 {
3021 	kunmap_atomic(virt);
3022 }
3023 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3024 
3025 void sdev_disable_disk_events(struct scsi_device *sdev)
3026 {
3027 	atomic_inc(&sdev->disk_events_disable_depth);
3028 }
3029 EXPORT_SYMBOL(sdev_disable_disk_events);
3030 
3031 void sdev_enable_disk_events(struct scsi_device *sdev)
3032 {
3033 	if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3034 		return;
3035 	atomic_dec(&sdev->disk_events_disable_depth);
3036 }
3037 EXPORT_SYMBOL(sdev_enable_disk_events);
3038