1 /*
2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3  * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4  *
5  * This file is released under the GPL.
6  */
7 
8 #include "dm-core.h"
9 #include "dm-rq.h"
10 #include "dm-uevent.h"
11 #include "dm-ima.h"
12 
13 #include <linux/init.h>
14 #include <linux/module.h>
15 #include <linux/mutex.h>
16 #include <linux/sched/mm.h>
17 #include <linux/sched/signal.h>
18 #include <linux/blkpg.h>
19 #include <linux/bio.h>
20 #include <linux/mempool.h>
21 #include <linux/dax.h>
22 #include <linux/slab.h>
23 #include <linux/idr.h>
24 #include <linux/uio.h>
25 #include <linux/hdreg.h>
26 #include <linux/delay.h>
27 #include <linux/wait.h>
28 #include <linux/pr.h>
29 #include <linux/refcount.h>
30 #include <linux/part_stat.h>
31 #include <linux/blk-crypto.h>
32 #include <linux/blk-crypto-profile.h>
33 
34 #define DM_MSG_PREFIX "core"
35 
36 /*
37  * Cookies are numeric values sent with CHANGE and REMOVE
38  * uevents while resuming, removing or renaming the device.
39  */
40 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
41 #define DM_COOKIE_LENGTH 24
42 
43 /*
44  * For REQ_POLLED fs bio, this flag is set if we link mapped underlying
45  * dm_io into one list, and reuse bio->bi_private as the list head. Before
46  * ending this fs bio, we will recover its ->bi_private.
47  */
48 #define REQ_DM_POLL_LIST	REQ_DRV
49 
50 static const char *_name = DM_NAME;
51 
52 static unsigned int major = 0;
53 static unsigned int _major = 0;
54 
55 static DEFINE_IDR(_minor_idr);
56 
57 static DEFINE_SPINLOCK(_minor_lock);
58 
59 static void do_deferred_remove(struct work_struct *w);
60 
61 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
62 
63 static struct workqueue_struct *deferred_remove_workqueue;
64 
65 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
66 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
67 
dm_issue_global_event(void)68 void dm_issue_global_event(void)
69 {
70 	atomic_inc(&dm_global_event_nr);
71 	wake_up(&dm_global_eventq);
72 }
73 
74 DEFINE_STATIC_KEY_FALSE(stats_enabled);
75 DEFINE_STATIC_KEY_FALSE(swap_bios_enabled);
76 DEFINE_STATIC_KEY_FALSE(zoned_enabled);
77 
78 /*
79  * One of these is allocated (on-stack) per original bio.
80  */
81 struct clone_info {
82 	struct dm_table *map;
83 	struct bio *bio;
84 	struct dm_io *io;
85 	sector_t sector;
86 	unsigned sector_count;
87 	bool is_abnormal_io:1;
88 	bool submit_as_polled:1;
89 };
90 
clone_to_tio(struct bio * clone)91 static inline struct dm_target_io *clone_to_tio(struct bio *clone)
92 {
93 	return container_of(clone, struct dm_target_io, clone);
94 }
95 
dm_per_bio_data(struct bio * bio,size_t data_size)96 void *dm_per_bio_data(struct bio *bio, size_t data_size)
97 {
98 	if (!dm_tio_flagged(clone_to_tio(bio), DM_TIO_INSIDE_DM_IO))
99 		return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
100 	return (char *)bio - DM_IO_BIO_OFFSET - data_size;
101 }
102 EXPORT_SYMBOL_GPL(dm_per_bio_data);
103 
dm_bio_from_per_bio_data(void * data,size_t data_size)104 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
105 {
106 	struct dm_io *io = (struct dm_io *)((char *)data + data_size);
107 	if (io->magic == DM_IO_MAGIC)
108 		return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
109 	BUG_ON(io->magic != DM_TIO_MAGIC);
110 	return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
111 }
112 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
113 
dm_bio_get_target_bio_nr(const struct bio * bio)114 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
115 {
116 	return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
117 }
118 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
119 
120 #define MINOR_ALLOCED ((void *)-1)
121 
122 #define DM_NUMA_NODE NUMA_NO_NODE
123 static int dm_numa_node = DM_NUMA_NODE;
124 
125 #define DEFAULT_SWAP_BIOS	(8 * 1048576 / PAGE_SIZE)
126 static int swap_bios = DEFAULT_SWAP_BIOS;
get_swap_bios(void)127 static int get_swap_bios(void)
128 {
129 	int latch = READ_ONCE(swap_bios);
130 	if (unlikely(latch <= 0))
131 		latch = DEFAULT_SWAP_BIOS;
132 	return latch;
133 }
134 
135 struct table_device {
136 	struct list_head list;
137 	refcount_t count;
138 	struct dm_dev dm_dev;
139 };
140 
141 /*
142  * Bio-based DM's mempools' reserved IOs set by the user.
143  */
144 #define RESERVED_BIO_BASED_IOS		16
145 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
146 
__dm_get_module_param_int(int * module_param,int min,int max)147 static int __dm_get_module_param_int(int *module_param, int min, int max)
148 {
149 	int param = READ_ONCE(*module_param);
150 	int modified_param = 0;
151 	bool modified = true;
152 
153 	if (param < min)
154 		modified_param = min;
155 	else if (param > max)
156 		modified_param = max;
157 	else
158 		modified = false;
159 
160 	if (modified) {
161 		(void)cmpxchg(module_param, param, modified_param);
162 		param = modified_param;
163 	}
164 
165 	return param;
166 }
167 
__dm_get_module_param(unsigned * module_param,unsigned def,unsigned max)168 unsigned __dm_get_module_param(unsigned *module_param,
169 			       unsigned def, unsigned max)
170 {
171 	unsigned param = READ_ONCE(*module_param);
172 	unsigned modified_param = 0;
173 
174 	if (!param)
175 		modified_param = def;
176 	else if (param > max)
177 		modified_param = max;
178 
179 	if (modified_param) {
180 		(void)cmpxchg(module_param, param, modified_param);
181 		param = modified_param;
182 	}
183 
184 	return param;
185 }
186 
dm_get_reserved_bio_based_ios(void)187 unsigned dm_get_reserved_bio_based_ios(void)
188 {
189 	return __dm_get_module_param(&reserved_bio_based_ios,
190 				     RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
191 }
192 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
193 
dm_get_numa_node(void)194 static unsigned dm_get_numa_node(void)
195 {
196 	return __dm_get_module_param_int(&dm_numa_node,
197 					 DM_NUMA_NODE, num_online_nodes() - 1);
198 }
199 
local_init(void)200 static int __init local_init(void)
201 {
202 	int r;
203 
204 	r = dm_uevent_init();
205 	if (r)
206 		return r;
207 
208 	deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
209 	if (!deferred_remove_workqueue) {
210 		r = -ENOMEM;
211 		goto out_uevent_exit;
212 	}
213 
214 	_major = major;
215 	r = register_blkdev(_major, _name);
216 	if (r < 0)
217 		goto out_free_workqueue;
218 
219 	if (!_major)
220 		_major = r;
221 
222 	return 0;
223 
224 out_free_workqueue:
225 	destroy_workqueue(deferred_remove_workqueue);
226 out_uevent_exit:
227 	dm_uevent_exit();
228 
229 	return r;
230 }
231 
local_exit(void)232 static void local_exit(void)
233 {
234 	flush_scheduled_work();
235 	destroy_workqueue(deferred_remove_workqueue);
236 
237 	unregister_blkdev(_major, _name);
238 	dm_uevent_exit();
239 
240 	_major = 0;
241 
242 	DMINFO("cleaned up");
243 }
244 
245 static int (*_inits[])(void) __initdata = {
246 	local_init,
247 	dm_target_init,
248 	dm_linear_init,
249 	dm_stripe_init,
250 	dm_io_init,
251 	dm_kcopyd_init,
252 	dm_interface_init,
253 	dm_statistics_init,
254 };
255 
256 static void (*_exits[])(void) = {
257 	local_exit,
258 	dm_target_exit,
259 	dm_linear_exit,
260 	dm_stripe_exit,
261 	dm_io_exit,
262 	dm_kcopyd_exit,
263 	dm_interface_exit,
264 	dm_statistics_exit,
265 };
266 
dm_init(void)267 static int __init dm_init(void)
268 {
269 	const int count = ARRAY_SIZE(_inits);
270 	int r, i;
271 
272 #if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
273 	DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
274 	       " Duplicate IMA measurements will not be recorded in the IMA log.");
275 #endif
276 
277 	for (i = 0; i < count; i++) {
278 		r = _inits[i]();
279 		if (r)
280 			goto bad;
281 	}
282 
283 	return 0;
284 bad:
285 	while (i--)
286 		_exits[i]();
287 
288 	return r;
289 }
290 
dm_exit(void)291 static void __exit dm_exit(void)
292 {
293 	int i = ARRAY_SIZE(_exits);
294 
295 	while (i--)
296 		_exits[i]();
297 
298 	/*
299 	 * Should be empty by this point.
300 	 */
301 	idr_destroy(&_minor_idr);
302 }
303 
304 /*
305  * Block device functions
306  */
dm_deleting_md(struct mapped_device * md)307 int dm_deleting_md(struct mapped_device *md)
308 {
309 	return test_bit(DMF_DELETING, &md->flags);
310 }
311 
dm_blk_open(struct block_device * bdev,fmode_t mode)312 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
313 {
314 	struct mapped_device *md;
315 
316 	spin_lock(&_minor_lock);
317 
318 	md = bdev->bd_disk->private_data;
319 	if (!md)
320 		goto out;
321 
322 	if (test_bit(DMF_FREEING, &md->flags) ||
323 	    dm_deleting_md(md)) {
324 		md = NULL;
325 		goto out;
326 	}
327 
328 	dm_get(md);
329 	atomic_inc(&md->open_count);
330 out:
331 	spin_unlock(&_minor_lock);
332 
333 	return md ? 0 : -ENXIO;
334 }
335 
dm_blk_close(struct gendisk * disk,fmode_t mode)336 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
337 {
338 	struct mapped_device *md;
339 
340 	spin_lock(&_minor_lock);
341 
342 	md = disk->private_data;
343 	if (WARN_ON(!md))
344 		goto out;
345 
346 	if (atomic_dec_and_test(&md->open_count) &&
347 	    (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
348 		queue_work(deferred_remove_workqueue, &deferred_remove_work);
349 
350 	dm_put(md);
351 out:
352 	spin_unlock(&_minor_lock);
353 }
354 
dm_open_count(struct mapped_device * md)355 int dm_open_count(struct mapped_device *md)
356 {
357 	return atomic_read(&md->open_count);
358 }
359 
360 /*
361  * Guarantees nothing is using the device before it's deleted.
362  */
dm_lock_for_deletion(struct mapped_device * md,bool mark_deferred,bool only_deferred)363 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
364 {
365 	int r = 0;
366 
367 	spin_lock(&_minor_lock);
368 
369 	if (dm_open_count(md)) {
370 		r = -EBUSY;
371 		if (mark_deferred)
372 			set_bit(DMF_DEFERRED_REMOVE, &md->flags);
373 	} else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
374 		r = -EEXIST;
375 	else
376 		set_bit(DMF_DELETING, &md->flags);
377 
378 	spin_unlock(&_minor_lock);
379 
380 	return r;
381 }
382 
dm_cancel_deferred_remove(struct mapped_device * md)383 int dm_cancel_deferred_remove(struct mapped_device *md)
384 {
385 	int r = 0;
386 
387 	spin_lock(&_minor_lock);
388 
389 	if (test_bit(DMF_DELETING, &md->flags))
390 		r = -EBUSY;
391 	else
392 		clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
393 
394 	spin_unlock(&_minor_lock);
395 
396 	return r;
397 }
398 
do_deferred_remove(struct work_struct * w)399 static void do_deferred_remove(struct work_struct *w)
400 {
401 	dm_deferred_remove();
402 }
403 
dm_blk_getgeo(struct block_device * bdev,struct hd_geometry * geo)404 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
405 {
406 	struct mapped_device *md = bdev->bd_disk->private_data;
407 
408 	return dm_get_geometry(md, geo);
409 }
410 
dm_prepare_ioctl(struct mapped_device * md,int * srcu_idx,struct block_device ** bdev)411 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
412 			    struct block_device **bdev)
413 {
414 	struct dm_target *ti;
415 	struct dm_table *map;
416 	int r;
417 
418 retry:
419 	r = -ENOTTY;
420 	map = dm_get_live_table(md, srcu_idx);
421 	if (!map || !dm_table_get_size(map))
422 		return r;
423 
424 	/* We only support devices that have a single target */
425 	if (map->num_targets != 1)
426 		return r;
427 
428 	ti = dm_table_get_target(map, 0);
429 	if (!ti->type->prepare_ioctl)
430 		return r;
431 
432 	if (dm_suspended_md(md))
433 		return -EAGAIN;
434 
435 	r = ti->type->prepare_ioctl(ti, bdev);
436 	if (r == -ENOTCONN && !fatal_signal_pending(current)) {
437 		dm_put_live_table(md, *srcu_idx);
438 		msleep(10);
439 		goto retry;
440 	}
441 
442 	return r;
443 }
444 
dm_unprepare_ioctl(struct mapped_device * md,int srcu_idx)445 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
446 {
447 	dm_put_live_table(md, srcu_idx);
448 }
449 
dm_blk_ioctl(struct block_device * bdev,fmode_t mode,unsigned int cmd,unsigned long arg)450 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
451 			unsigned int cmd, unsigned long arg)
452 {
453 	struct mapped_device *md = bdev->bd_disk->private_data;
454 	int r, srcu_idx;
455 
456 	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
457 	if (r < 0)
458 		goto out;
459 
460 	if (r > 0) {
461 		/*
462 		 * Target determined this ioctl is being issued against a
463 		 * subset of the parent bdev; require extra privileges.
464 		 */
465 		if (!capable(CAP_SYS_RAWIO)) {
466 			DMDEBUG_LIMIT(
467 	"%s: sending ioctl %x to DM device without required privilege.",
468 				current->comm, cmd);
469 			r = -ENOIOCTLCMD;
470 			goto out;
471 		}
472 	}
473 
474 	if (!bdev->bd_disk->fops->ioctl)
475 		r = -ENOTTY;
476 	else
477 		r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
478 out:
479 	dm_unprepare_ioctl(md, srcu_idx);
480 	return r;
481 }
482 
dm_start_time_ns_from_clone(struct bio * bio)483 u64 dm_start_time_ns_from_clone(struct bio *bio)
484 {
485 	return jiffies_to_nsecs(clone_to_tio(bio)->io->start_time);
486 }
487 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
488 
bio_is_flush_with_data(struct bio * bio)489 static bool bio_is_flush_with_data(struct bio *bio)
490 {
491 	return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size);
492 }
493 
dm_io_acct(struct dm_io * io,bool end)494 static void dm_io_acct(struct dm_io *io, bool end)
495 {
496 	struct dm_stats_aux *stats_aux = &io->stats_aux;
497 	unsigned long start_time = io->start_time;
498 	struct mapped_device *md = io->md;
499 	struct bio *bio = io->orig_bio;
500 	unsigned int sectors;
501 
502 	/*
503 	 * If REQ_PREFLUSH set, don't account payload, it will be
504 	 * submitted (and accounted) after this flush completes.
505 	 */
506 	if (bio_is_flush_with_data(bio))
507 		sectors = 0;
508 	else if (likely(!(dm_io_flagged(io, DM_IO_WAS_SPLIT))))
509 		sectors = bio_sectors(bio);
510 	else
511 		sectors = io->sectors;
512 
513 	if (!end)
514 		bdev_start_io_acct(bio->bi_bdev, sectors, bio_op(bio),
515 				   start_time);
516 	else
517 		bdev_end_io_acct(bio->bi_bdev, bio_op(bio), start_time);
518 
519 	if (static_branch_unlikely(&stats_enabled) &&
520 	    unlikely(dm_stats_used(&md->stats))) {
521 		sector_t sector;
522 
523 		if (likely(!dm_io_flagged(io, DM_IO_WAS_SPLIT)))
524 			sector = bio->bi_iter.bi_sector;
525 		else
526 			sector = bio_end_sector(bio) - io->sector_offset;
527 
528 		dm_stats_account_io(&md->stats, bio_data_dir(bio),
529 				    sector, sectors,
530 				    end, start_time, stats_aux);
531 	}
532 }
533 
__dm_start_io_acct(struct dm_io * io)534 static void __dm_start_io_acct(struct dm_io *io)
535 {
536 	dm_io_acct(io, false);
537 }
538 
dm_start_io_acct(struct dm_io * io,struct bio * clone)539 static void dm_start_io_acct(struct dm_io *io, struct bio *clone)
540 {
541 	/*
542 	 * Ensure IO accounting is only ever started once.
543 	 */
544 	if (dm_io_flagged(io, DM_IO_ACCOUNTED))
545 		return;
546 
547 	/* Expect no possibility for race unless DM_TIO_IS_DUPLICATE_BIO. */
548 	if (!clone || likely(dm_tio_is_normal(clone_to_tio(clone)))) {
549 		dm_io_set_flag(io, DM_IO_ACCOUNTED);
550 	} else {
551 		unsigned long flags;
552 		/* Can afford locking given DM_TIO_IS_DUPLICATE_BIO */
553 		spin_lock_irqsave(&io->lock, flags);
554 		if (dm_io_flagged(io, DM_IO_ACCOUNTED)) {
555 			spin_unlock_irqrestore(&io->lock, flags);
556 			return;
557 		}
558 		dm_io_set_flag(io, DM_IO_ACCOUNTED);
559 		spin_unlock_irqrestore(&io->lock, flags);
560 	}
561 
562 	__dm_start_io_acct(io);
563 }
564 
dm_end_io_acct(struct dm_io * io)565 static void dm_end_io_acct(struct dm_io *io)
566 {
567 	dm_io_acct(io, true);
568 }
569 
alloc_io(struct mapped_device * md,struct bio * bio)570 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
571 {
572 	struct dm_io *io;
573 	struct dm_target_io *tio;
574 	struct bio *clone;
575 
576 	clone = bio_alloc_clone(NULL, bio, GFP_NOIO, &md->mempools->io_bs);
577 	tio = clone_to_tio(clone);
578 	tio->flags = 0;
579 	dm_tio_set_flag(tio, DM_TIO_INSIDE_DM_IO);
580 	tio->io = NULL;
581 
582 	io = container_of(tio, struct dm_io, tio);
583 	io->magic = DM_IO_MAGIC;
584 	io->status = BLK_STS_OK;
585 
586 	/* one ref is for submission, the other is for completion */
587 	atomic_set(&io->io_count, 2);
588 	this_cpu_inc(*md->pending_io);
589 	io->orig_bio = bio;
590 	io->md = md;
591 	spin_lock_init(&io->lock);
592 	io->start_time = jiffies;
593 	io->flags = 0;
594 
595 	if (static_branch_unlikely(&stats_enabled))
596 		dm_stats_record_start(&md->stats, &io->stats_aux);
597 
598 	return io;
599 }
600 
free_io(struct dm_io * io)601 static void free_io(struct dm_io *io)
602 {
603 	bio_put(&io->tio.clone);
604 }
605 
alloc_tio(struct clone_info * ci,struct dm_target * ti,unsigned target_bio_nr,unsigned * len,gfp_t gfp_mask)606 static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti,
607 			     unsigned target_bio_nr, unsigned *len, gfp_t gfp_mask)
608 {
609 	struct mapped_device *md = ci->io->md;
610 	struct dm_target_io *tio;
611 	struct bio *clone;
612 
613 	if (!ci->io->tio.io) {
614 		/* the dm_target_io embedded in ci->io is available */
615 		tio = &ci->io->tio;
616 		/* alloc_io() already initialized embedded clone */
617 		clone = &tio->clone;
618 	} else {
619 		clone = bio_alloc_clone(NULL, ci->bio, gfp_mask,
620 					&md->mempools->bs);
621 		if (!clone)
622 			return NULL;
623 
624 		/* REQ_DM_POLL_LIST shouldn't be inherited */
625 		clone->bi_opf &= ~REQ_DM_POLL_LIST;
626 
627 		tio = clone_to_tio(clone);
628 		tio->flags = 0; /* also clears DM_TIO_INSIDE_DM_IO */
629 	}
630 
631 	tio->magic = DM_TIO_MAGIC;
632 	tio->io = ci->io;
633 	tio->ti = ti;
634 	tio->target_bio_nr = target_bio_nr;
635 	tio->len_ptr = len;
636 	tio->old_sector = 0;
637 
638 	/* Set default bdev, but target must bio_set_dev() before issuing IO */
639 	clone->bi_bdev = md->disk->part0;
640 	if (unlikely(ti->needs_bio_set_dev))
641 		bio_set_dev(clone, md->disk->part0);
642 
643 	if (len) {
644 		clone->bi_iter.bi_size = to_bytes(*len);
645 		if (bio_integrity(clone))
646 			bio_integrity_trim(clone);
647 	}
648 
649 	return clone;
650 }
651 
free_tio(struct bio * clone)652 static void free_tio(struct bio *clone)
653 {
654 	if (dm_tio_flagged(clone_to_tio(clone), DM_TIO_INSIDE_DM_IO))
655 		return;
656 	bio_put(clone);
657 }
658 
659 /*
660  * Add the bio to the list of deferred io.
661  */
queue_io(struct mapped_device * md,struct bio * bio)662 static void queue_io(struct mapped_device *md, struct bio *bio)
663 {
664 	unsigned long flags;
665 
666 	spin_lock_irqsave(&md->deferred_lock, flags);
667 	bio_list_add(&md->deferred, bio);
668 	spin_unlock_irqrestore(&md->deferred_lock, flags);
669 	queue_work(md->wq, &md->work);
670 }
671 
672 /*
673  * Everyone (including functions in this file), should use this
674  * function to access the md->map field, and make sure they call
675  * dm_put_live_table() when finished.
676  */
dm_get_live_table(struct mapped_device * md,int * srcu_idx)677 struct dm_table *dm_get_live_table(struct mapped_device *md,
678 				   int *srcu_idx) __acquires(md->io_barrier)
679 {
680 	*srcu_idx = srcu_read_lock(&md->io_barrier);
681 
682 	return srcu_dereference(md->map, &md->io_barrier);
683 }
684 
dm_put_live_table(struct mapped_device * md,int srcu_idx)685 void dm_put_live_table(struct mapped_device *md,
686 		       int srcu_idx) __releases(md->io_barrier)
687 {
688 	srcu_read_unlock(&md->io_barrier, srcu_idx);
689 }
690 
dm_sync_table(struct mapped_device * md)691 void dm_sync_table(struct mapped_device *md)
692 {
693 	synchronize_srcu(&md->io_barrier);
694 	synchronize_rcu_expedited();
695 }
696 
697 /*
698  * A fast alternative to dm_get_live_table/dm_put_live_table.
699  * The caller must not block between these two functions.
700  */
dm_get_live_table_fast(struct mapped_device * md)701 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
702 {
703 	rcu_read_lock();
704 	return rcu_dereference(md->map);
705 }
706 
dm_put_live_table_fast(struct mapped_device * md)707 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
708 {
709 	rcu_read_unlock();
710 }
711 
dm_get_live_table_bio(struct mapped_device * md,int * srcu_idx,blk_opf_t bio_opf)712 static inline struct dm_table *dm_get_live_table_bio(struct mapped_device *md,
713 					int *srcu_idx, blk_opf_t bio_opf)
714 {
715 	if (bio_opf & REQ_NOWAIT)
716 		return dm_get_live_table_fast(md);
717 	else
718 		return dm_get_live_table(md, srcu_idx);
719 }
720 
dm_put_live_table_bio(struct mapped_device * md,int srcu_idx,blk_opf_t bio_opf)721 static inline void dm_put_live_table_bio(struct mapped_device *md, int srcu_idx,
722 					 blk_opf_t bio_opf)
723 {
724 	if (bio_opf & REQ_NOWAIT)
725 		dm_put_live_table_fast(md);
726 	else
727 		dm_put_live_table(md, srcu_idx);
728 }
729 
730 static char *_dm_claim_ptr = "I belong to device-mapper";
731 
732 /*
733  * Open a table device so we can use it as a map destination.
734  */
open_table_device(struct mapped_device * md,dev_t dev,fmode_t mode)735 static struct table_device *open_table_device(struct mapped_device *md,
736 		dev_t dev, fmode_t mode)
737 {
738 	struct table_device *td;
739 	struct block_device *bdev;
740 	u64 part_off;
741 	int r;
742 
743 	td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
744 	if (!td)
745 		return ERR_PTR(-ENOMEM);
746 	refcount_set(&td->count, 1);
747 
748 	bdev = blkdev_get_by_dev(dev, mode | FMODE_EXCL, _dm_claim_ptr);
749 	if (IS_ERR(bdev)) {
750 		r = PTR_ERR(bdev);
751 		goto out_free_td;
752 	}
753 
754 	/*
755 	 * We can be called before the dm disk is added.  In that case we can't
756 	 * register the holder relation here.  It will be done once add_disk was
757 	 * called.
758 	 */
759 	if (md->disk->slave_dir) {
760 		r = bd_link_disk_holder(bdev, md->disk);
761 		if (r)
762 			goto out_blkdev_put;
763 	}
764 
765 	td->dm_dev.mode = mode;
766 	td->dm_dev.bdev = bdev;
767 	td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev, &part_off, NULL, NULL);
768 	format_dev_t(td->dm_dev.name, dev);
769 	list_add(&td->list, &md->table_devices);
770 	return td;
771 
772 out_blkdev_put:
773 	blkdev_put(bdev, mode | FMODE_EXCL);
774 out_free_td:
775 	kfree(td);
776 	return ERR_PTR(r);
777 }
778 
779 /*
780  * Close a table device that we've been using.
781  */
close_table_device(struct table_device * td,struct mapped_device * md)782 static void close_table_device(struct table_device *td, struct mapped_device *md)
783 {
784 	if (md->disk->slave_dir)
785 		bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
786 	blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
787 	put_dax(td->dm_dev.dax_dev);
788 	list_del(&td->list);
789 	kfree(td);
790 }
791 
find_table_device(struct list_head * l,dev_t dev,fmode_t mode)792 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
793 					      fmode_t mode)
794 {
795 	struct table_device *td;
796 
797 	list_for_each_entry(td, l, list)
798 		if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
799 			return td;
800 
801 	return NULL;
802 }
803 
dm_get_table_device(struct mapped_device * md,dev_t dev,fmode_t mode,struct dm_dev ** result)804 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
805 			struct dm_dev **result)
806 {
807 	struct table_device *td;
808 
809 	mutex_lock(&md->table_devices_lock);
810 	td = find_table_device(&md->table_devices, dev, mode);
811 	if (!td) {
812 		td = open_table_device(md, dev, mode);
813 		if (IS_ERR(td)) {
814 			mutex_unlock(&md->table_devices_lock);
815 			return PTR_ERR(td);
816 		}
817 	} else {
818 		refcount_inc(&td->count);
819 	}
820 	mutex_unlock(&md->table_devices_lock);
821 
822 	*result = &td->dm_dev;
823 	return 0;
824 }
825 
dm_put_table_device(struct mapped_device * md,struct dm_dev * d)826 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
827 {
828 	struct table_device *td = container_of(d, struct table_device, dm_dev);
829 
830 	mutex_lock(&md->table_devices_lock);
831 	if (refcount_dec_and_test(&td->count))
832 		close_table_device(td, md);
833 	mutex_unlock(&md->table_devices_lock);
834 }
835 
free_table_devices(struct list_head * devices)836 static void free_table_devices(struct list_head *devices)
837 {
838 	struct list_head *tmp, *next;
839 
840 	list_for_each_safe(tmp, next, devices) {
841 		struct table_device *td = list_entry(tmp, struct table_device, list);
842 
843 		DMWARN("dm_destroy: %s still exists with %d references",
844 		       td->dm_dev.name, refcount_read(&td->count));
845 		kfree(td);
846 	}
847 }
848 
849 /*
850  * Get the geometry associated with a dm device
851  */
dm_get_geometry(struct mapped_device * md,struct hd_geometry * geo)852 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
853 {
854 	*geo = md->geometry;
855 
856 	return 0;
857 }
858 
859 /*
860  * Set the geometry of a device.
861  */
dm_set_geometry(struct mapped_device * md,struct hd_geometry * geo)862 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
863 {
864 	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
865 
866 	if (geo->start > sz) {
867 		DMERR("Start sector is beyond the geometry limits.");
868 		return -EINVAL;
869 	}
870 
871 	md->geometry = *geo;
872 
873 	return 0;
874 }
875 
__noflush_suspending(struct mapped_device * md)876 static int __noflush_suspending(struct mapped_device *md)
877 {
878 	return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
879 }
880 
dm_requeue_add_io(struct dm_io * io,bool first_stage)881 static void dm_requeue_add_io(struct dm_io *io, bool first_stage)
882 {
883 	struct mapped_device *md = io->md;
884 
885 	if (first_stage) {
886 		struct dm_io *next = md->requeue_list;
887 
888 		md->requeue_list = io;
889 		io->next = next;
890 	} else {
891 		bio_list_add_head(&md->deferred, io->orig_bio);
892 	}
893 }
894 
dm_kick_requeue(struct mapped_device * md,bool first_stage)895 static void dm_kick_requeue(struct mapped_device *md, bool first_stage)
896 {
897 	if (first_stage)
898 		queue_work(md->wq, &md->requeue_work);
899 	else
900 		queue_work(md->wq, &md->work);
901 }
902 
903 /*
904  * Return true if the dm_io's original bio is requeued.
905  * io->status is updated with error if requeue disallowed.
906  */
dm_handle_requeue(struct dm_io * io,bool first_stage)907 static bool dm_handle_requeue(struct dm_io *io, bool first_stage)
908 {
909 	struct bio *bio = io->orig_bio;
910 	bool handle_requeue = (io->status == BLK_STS_DM_REQUEUE);
911 	bool handle_polled_eagain = ((io->status == BLK_STS_AGAIN) &&
912 				     (bio->bi_opf & REQ_POLLED));
913 	struct mapped_device *md = io->md;
914 	bool requeued = false;
915 
916 	if (handle_requeue || handle_polled_eagain) {
917 		unsigned long flags;
918 
919 		if (bio->bi_opf & REQ_POLLED) {
920 			/*
921 			 * Upper layer won't help us poll split bio
922 			 * (io->orig_bio may only reflect a subset of the
923 			 * pre-split original) so clear REQ_POLLED.
924 			 */
925 			bio_clear_polled(bio);
926 		}
927 
928 		/*
929 		 * Target requested pushing back the I/O or
930 		 * polled IO hit BLK_STS_AGAIN.
931 		 */
932 		spin_lock_irqsave(&md->deferred_lock, flags);
933 		if ((__noflush_suspending(md) &&
934 		     !WARN_ON_ONCE(dm_is_zone_write(md, bio))) ||
935 		    handle_polled_eagain || first_stage) {
936 			dm_requeue_add_io(io, first_stage);
937 			requeued = true;
938 		} else {
939 			/*
940 			 * noflush suspend was interrupted or this is
941 			 * a write to a zoned target.
942 			 */
943 			io->status = BLK_STS_IOERR;
944 		}
945 		spin_unlock_irqrestore(&md->deferred_lock, flags);
946 	}
947 
948 	if (requeued)
949 		dm_kick_requeue(md, first_stage);
950 
951 	return requeued;
952 }
953 
__dm_io_complete(struct dm_io * io,bool first_stage)954 static void __dm_io_complete(struct dm_io *io, bool first_stage)
955 {
956 	struct bio *bio = io->orig_bio;
957 	struct mapped_device *md = io->md;
958 	blk_status_t io_error;
959 	bool requeued;
960 
961 	requeued = dm_handle_requeue(io, first_stage);
962 	if (requeued && first_stage)
963 		return;
964 
965 	io_error = io->status;
966 	if (dm_io_flagged(io, DM_IO_ACCOUNTED))
967 		dm_end_io_acct(io);
968 	else if (!io_error) {
969 		/*
970 		 * Must handle target that DM_MAPIO_SUBMITTED only to
971 		 * then bio_endio() rather than dm_submit_bio_remap()
972 		 */
973 		__dm_start_io_acct(io);
974 		dm_end_io_acct(io);
975 	}
976 	free_io(io);
977 	smp_wmb();
978 	this_cpu_dec(*md->pending_io);
979 
980 	/* nudge anyone waiting on suspend queue */
981 	if (unlikely(wq_has_sleeper(&md->wait)))
982 		wake_up(&md->wait);
983 
984 	/* Return early if the original bio was requeued */
985 	if (requeued)
986 		return;
987 
988 	if (bio_is_flush_with_data(bio)) {
989 		/*
990 		 * Preflush done for flush with data, reissue
991 		 * without REQ_PREFLUSH.
992 		 */
993 		bio->bi_opf &= ~REQ_PREFLUSH;
994 		queue_io(md, bio);
995 	} else {
996 		/* done with normal IO or empty flush */
997 		if (io_error)
998 			bio->bi_status = io_error;
999 		bio_endio(bio);
1000 	}
1001 }
1002 
dm_wq_requeue_work(struct work_struct * work)1003 static void dm_wq_requeue_work(struct work_struct *work)
1004 {
1005 	struct mapped_device *md = container_of(work, struct mapped_device,
1006 						requeue_work);
1007 	unsigned long flags;
1008 	struct dm_io *io;
1009 
1010 	/* reuse deferred lock to simplify dm_handle_requeue */
1011 	spin_lock_irqsave(&md->deferred_lock, flags);
1012 	io = md->requeue_list;
1013 	md->requeue_list = NULL;
1014 	spin_unlock_irqrestore(&md->deferred_lock, flags);
1015 
1016 	while (io) {
1017 		struct dm_io *next = io->next;
1018 
1019 		dm_io_rewind(io, &md->disk->bio_split);
1020 
1021 		io->next = NULL;
1022 		__dm_io_complete(io, false);
1023 		io = next;
1024 	}
1025 }
1026 
1027 /*
1028  * Two staged requeue:
1029  *
1030  * 1) io->orig_bio points to the real original bio, and the part mapped to
1031  *    this io must be requeued, instead of other parts of the original bio.
1032  *
1033  * 2) io->orig_bio points to new cloned bio which matches the requeued dm_io.
1034  */
dm_io_complete(struct dm_io * io)1035 static void dm_io_complete(struct dm_io *io)
1036 {
1037 	bool first_requeue;
1038 
1039 	/*
1040 	 * Only dm_io that has been split needs two stage requeue, otherwise
1041 	 * we may run into long bio clone chain during suspend and OOM could
1042 	 * be triggered.
1043 	 *
1044 	 * Also flush data dm_io won't be marked as DM_IO_WAS_SPLIT, so they
1045 	 * also aren't handled via the first stage requeue.
1046 	 */
1047 	if (dm_io_flagged(io, DM_IO_WAS_SPLIT))
1048 		first_requeue = true;
1049 	else
1050 		first_requeue = false;
1051 
1052 	__dm_io_complete(io, first_requeue);
1053 }
1054 
1055 /*
1056  * Decrements the number of outstanding ios that a bio has been
1057  * cloned into, completing the original io if necc.
1058  */
__dm_io_dec_pending(struct dm_io * io)1059 static inline void __dm_io_dec_pending(struct dm_io *io)
1060 {
1061 	if (atomic_dec_and_test(&io->io_count))
1062 		dm_io_complete(io);
1063 }
1064 
dm_io_set_error(struct dm_io * io,blk_status_t error)1065 static void dm_io_set_error(struct dm_io *io, blk_status_t error)
1066 {
1067 	unsigned long flags;
1068 
1069 	/* Push-back supersedes any I/O errors */
1070 	spin_lock_irqsave(&io->lock, flags);
1071 	if (!(io->status == BLK_STS_DM_REQUEUE &&
1072 	      __noflush_suspending(io->md))) {
1073 		io->status = error;
1074 	}
1075 	spin_unlock_irqrestore(&io->lock, flags);
1076 }
1077 
dm_io_dec_pending(struct dm_io * io,blk_status_t error)1078 static void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
1079 {
1080 	if (unlikely(error))
1081 		dm_io_set_error(io, error);
1082 
1083 	__dm_io_dec_pending(io);
1084 }
1085 
disable_discard(struct mapped_device * md)1086 void disable_discard(struct mapped_device *md)
1087 {
1088 	struct queue_limits *limits = dm_get_queue_limits(md);
1089 
1090 	/* device doesn't really support DISCARD, disable it */
1091 	limits->max_discard_sectors = 0;
1092 }
1093 
disable_write_zeroes(struct mapped_device * md)1094 void disable_write_zeroes(struct mapped_device *md)
1095 {
1096 	struct queue_limits *limits = dm_get_queue_limits(md);
1097 
1098 	/* device doesn't really support WRITE ZEROES, disable it */
1099 	limits->max_write_zeroes_sectors = 0;
1100 }
1101 
swap_bios_limit(struct dm_target * ti,struct bio * bio)1102 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
1103 {
1104 	return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
1105 }
1106 
clone_endio(struct bio * bio)1107 static void clone_endio(struct bio *bio)
1108 {
1109 	blk_status_t error = bio->bi_status;
1110 	struct dm_target_io *tio = clone_to_tio(bio);
1111 	struct dm_target *ti = tio->ti;
1112 	dm_endio_fn endio = ti->type->end_io;
1113 	struct dm_io *io = tio->io;
1114 	struct mapped_device *md = io->md;
1115 
1116 	if (unlikely(error == BLK_STS_TARGET)) {
1117 		if (bio_op(bio) == REQ_OP_DISCARD &&
1118 		    !bdev_max_discard_sectors(bio->bi_bdev))
1119 			disable_discard(md);
1120 		else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1121 			 !bdev_write_zeroes_sectors(bio->bi_bdev))
1122 			disable_write_zeroes(md);
1123 	}
1124 
1125 	if (static_branch_unlikely(&zoned_enabled) &&
1126 	    unlikely(bdev_is_zoned(bio->bi_bdev)))
1127 		dm_zone_endio(io, bio);
1128 
1129 	if (endio) {
1130 		int r = endio(ti, bio, &error);
1131 		switch (r) {
1132 		case DM_ENDIO_REQUEUE:
1133 			if (static_branch_unlikely(&zoned_enabled)) {
1134 				/*
1135 				 * Requeuing writes to a sequential zone of a zoned
1136 				 * target will break the sequential write pattern:
1137 				 * fail such IO.
1138 				 */
1139 				if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
1140 					error = BLK_STS_IOERR;
1141 				else
1142 					error = BLK_STS_DM_REQUEUE;
1143 			} else
1144 				error = BLK_STS_DM_REQUEUE;
1145 			fallthrough;
1146 		case DM_ENDIO_DONE:
1147 			break;
1148 		case DM_ENDIO_INCOMPLETE:
1149 			/* The target will handle the io */
1150 			return;
1151 		default:
1152 			DMCRIT("unimplemented target endio return value: %d", r);
1153 			BUG();
1154 		}
1155 	}
1156 
1157 	if (static_branch_unlikely(&swap_bios_enabled) &&
1158 	    unlikely(swap_bios_limit(ti, bio)))
1159 		up(&md->swap_bios_semaphore);
1160 
1161 	free_tio(bio);
1162 	dm_io_dec_pending(io, error);
1163 }
1164 
1165 /*
1166  * Return maximum size of I/O possible at the supplied sector up to the current
1167  * target boundary.
1168  */
max_io_len_target_boundary(struct dm_target * ti,sector_t target_offset)1169 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1170 						  sector_t target_offset)
1171 {
1172 	return ti->len - target_offset;
1173 }
1174 
max_io_len(struct dm_target * ti,sector_t sector)1175 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
1176 {
1177 	sector_t target_offset = dm_target_offset(ti, sector);
1178 	sector_t len = max_io_len_target_boundary(ti, target_offset);
1179 
1180 	/*
1181 	 * Does the target need to split IO even further?
1182 	 * - varied (per target) IO splitting is a tenet of DM; this
1183 	 *   explains why stacked chunk_sectors based splitting via
1184 	 *   bio_split_to_limits() isn't possible here.
1185 	 */
1186 	if (!ti->max_io_len)
1187 		return len;
1188 	return min_t(sector_t, len,
1189 		min(queue_max_sectors(ti->table->md->queue),
1190 		    blk_chunk_sectors_left(target_offset, ti->max_io_len)));
1191 }
1192 
dm_set_target_max_io_len(struct dm_target * ti,sector_t len)1193 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1194 {
1195 	if (len > UINT_MAX) {
1196 		DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1197 		      (unsigned long long)len, UINT_MAX);
1198 		ti->error = "Maximum size of target IO is too large";
1199 		return -EINVAL;
1200 	}
1201 
1202 	ti->max_io_len = (uint32_t) len;
1203 
1204 	return 0;
1205 }
1206 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1207 
dm_dax_get_live_target(struct mapped_device * md,sector_t sector,int * srcu_idx)1208 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1209 						sector_t sector, int *srcu_idx)
1210 	__acquires(md->io_barrier)
1211 {
1212 	struct dm_table *map;
1213 	struct dm_target *ti;
1214 
1215 	map = dm_get_live_table(md, srcu_idx);
1216 	if (!map)
1217 		return NULL;
1218 
1219 	ti = dm_table_find_target(map, sector);
1220 	if (!ti)
1221 		return NULL;
1222 
1223 	return ti;
1224 }
1225 
dm_dax_direct_access(struct dax_device * dax_dev,pgoff_t pgoff,long nr_pages,enum dax_access_mode mode,void ** kaddr,pfn_t * pfn)1226 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1227 		long nr_pages, enum dax_access_mode mode, void **kaddr,
1228 		pfn_t *pfn)
1229 {
1230 	struct mapped_device *md = dax_get_private(dax_dev);
1231 	sector_t sector = pgoff * PAGE_SECTORS;
1232 	struct dm_target *ti;
1233 	long len, ret = -EIO;
1234 	int srcu_idx;
1235 
1236 	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1237 
1238 	if (!ti)
1239 		goto out;
1240 	if (!ti->type->direct_access)
1241 		goto out;
1242 	len = max_io_len(ti, sector) / PAGE_SECTORS;
1243 	if (len < 1)
1244 		goto out;
1245 	nr_pages = min(len, nr_pages);
1246 	ret = ti->type->direct_access(ti, pgoff, nr_pages, mode, kaddr, pfn);
1247 
1248  out:
1249 	dm_put_live_table(md, srcu_idx);
1250 
1251 	return ret;
1252 }
1253 
dm_dax_zero_page_range(struct dax_device * dax_dev,pgoff_t pgoff,size_t nr_pages)1254 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1255 				  size_t nr_pages)
1256 {
1257 	struct mapped_device *md = dax_get_private(dax_dev);
1258 	sector_t sector = pgoff * PAGE_SECTORS;
1259 	struct dm_target *ti;
1260 	int ret = -EIO;
1261 	int srcu_idx;
1262 
1263 	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1264 
1265 	if (!ti)
1266 		goto out;
1267 	if (WARN_ON(!ti->type->dax_zero_page_range)) {
1268 		/*
1269 		 * ->zero_page_range() is mandatory dax operation. If we are
1270 		 *  here, something is wrong.
1271 		 */
1272 		goto out;
1273 	}
1274 	ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1275  out:
1276 	dm_put_live_table(md, srcu_idx);
1277 
1278 	return ret;
1279 }
1280 
dm_dax_recovery_write(struct dax_device * dax_dev,pgoff_t pgoff,void * addr,size_t bytes,struct iov_iter * i)1281 static size_t dm_dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
1282 		void *addr, size_t bytes, struct iov_iter *i)
1283 {
1284 	struct mapped_device *md = dax_get_private(dax_dev);
1285 	sector_t sector = pgoff * PAGE_SECTORS;
1286 	struct dm_target *ti;
1287 	int srcu_idx;
1288 	long ret = 0;
1289 
1290 	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1291 	if (!ti || !ti->type->dax_recovery_write)
1292 		goto out;
1293 
1294 	ret = ti->type->dax_recovery_write(ti, pgoff, addr, bytes, i);
1295 out:
1296 	dm_put_live_table(md, srcu_idx);
1297 	return ret;
1298 }
1299 
1300 /*
1301  * A target may call dm_accept_partial_bio only from the map routine.  It is
1302  * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1303  * operations, REQ_OP_ZONE_APPEND (zone append writes) and any bio serviced by
1304  * __send_duplicate_bios().
1305  *
1306  * dm_accept_partial_bio informs the dm that the target only wants to process
1307  * additional n_sectors sectors of the bio and the rest of the data should be
1308  * sent in a next bio.
1309  *
1310  * A diagram that explains the arithmetics:
1311  * +--------------------+---------------+-------+
1312  * |         1          |       2       |   3   |
1313  * +--------------------+---------------+-------+
1314  *
1315  * <-------------- *tio->len_ptr --------------->
1316  *                      <----- bio_sectors ----->
1317  *                      <-- n_sectors -->
1318  *
1319  * Region 1 was already iterated over with bio_advance or similar function.
1320  *	(it may be empty if the target doesn't use bio_advance)
1321  * Region 2 is the remaining bio size that the target wants to process.
1322  *	(it may be empty if region 1 is non-empty, although there is no reason
1323  *	 to make it empty)
1324  * The target requires that region 3 is to be sent in the next bio.
1325  *
1326  * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1327  * the partially processed part (the sum of regions 1+2) must be the same for all
1328  * copies of the bio.
1329  */
dm_accept_partial_bio(struct bio * bio,unsigned n_sectors)1330 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1331 {
1332 	struct dm_target_io *tio = clone_to_tio(bio);
1333 	struct dm_io *io = tio->io;
1334 	unsigned bio_sectors = bio_sectors(bio);
1335 
1336 	BUG_ON(dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
1337 	BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1338 	BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1339 	BUG_ON(bio_sectors > *tio->len_ptr);
1340 	BUG_ON(n_sectors > bio_sectors);
1341 
1342 	*tio->len_ptr -= bio_sectors - n_sectors;
1343 	bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1344 
1345 	/*
1346 	 * __split_and_process_bio() may have already saved mapped part
1347 	 * for accounting but it is being reduced so update accordingly.
1348 	 */
1349 	dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1350 	io->sectors = n_sectors;
1351 	io->sector_offset = bio_sectors(io->orig_bio);
1352 }
1353 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1354 
1355 /*
1356  * @clone: clone bio that DM core passed to target's .map function
1357  * @tgt_clone: clone of @clone bio that target needs submitted
1358  *
1359  * Targets should use this interface to submit bios they take
1360  * ownership of when returning DM_MAPIO_SUBMITTED.
1361  *
1362  * Target should also enable ti->accounts_remapped_io
1363  */
dm_submit_bio_remap(struct bio * clone,struct bio * tgt_clone)1364 void dm_submit_bio_remap(struct bio *clone, struct bio *tgt_clone)
1365 {
1366 	struct dm_target_io *tio = clone_to_tio(clone);
1367 	struct dm_io *io = tio->io;
1368 
1369 	/* establish bio that will get submitted */
1370 	if (!tgt_clone)
1371 		tgt_clone = clone;
1372 
1373 	/*
1374 	 * Account io->origin_bio to DM dev on behalf of target
1375 	 * that took ownership of IO with DM_MAPIO_SUBMITTED.
1376 	 */
1377 	dm_start_io_acct(io, clone);
1378 
1379 	trace_block_bio_remap(tgt_clone, disk_devt(io->md->disk),
1380 			      tio->old_sector);
1381 	submit_bio_noacct(tgt_clone);
1382 }
1383 EXPORT_SYMBOL_GPL(dm_submit_bio_remap);
1384 
__set_swap_bios_limit(struct mapped_device * md,int latch)1385 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1386 {
1387 	mutex_lock(&md->swap_bios_lock);
1388 	while (latch < md->swap_bios) {
1389 		cond_resched();
1390 		down(&md->swap_bios_semaphore);
1391 		md->swap_bios--;
1392 	}
1393 	while (latch > md->swap_bios) {
1394 		cond_resched();
1395 		up(&md->swap_bios_semaphore);
1396 		md->swap_bios++;
1397 	}
1398 	mutex_unlock(&md->swap_bios_lock);
1399 }
1400 
__map_bio(struct bio * clone)1401 static void __map_bio(struct bio *clone)
1402 {
1403 	struct dm_target_io *tio = clone_to_tio(clone);
1404 	struct dm_target *ti = tio->ti;
1405 	struct dm_io *io = tio->io;
1406 	struct mapped_device *md = io->md;
1407 	int r;
1408 
1409 	clone->bi_end_io = clone_endio;
1410 
1411 	/*
1412 	 * Map the clone.
1413 	 */
1414 	tio->old_sector = clone->bi_iter.bi_sector;
1415 
1416 	if (static_branch_unlikely(&swap_bios_enabled) &&
1417 	    unlikely(swap_bios_limit(ti, clone))) {
1418 		int latch = get_swap_bios();
1419 		if (unlikely(latch != md->swap_bios))
1420 			__set_swap_bios_limit(md, latch);
1421 		down(&md->swap_bios_semaphore);
1422 	}
1423 
1424 	if (static_branch_unlikely(&zoned_enabled)) {
1425 		/*
1426 		 * Check if the IO needs a special mapping due to zone append
1427 		 * emulation on zoned target. In this case, dm_zone_map_bio()
1428 		 * calls the target map operation.
1429 		 */
1430 		if (unlikely(dm_emulate_zone_append(md)))
1431 			r = dm_zone_map_bio(tio);
1432 		else
1433 			r = ti->type->map(ti, clone);
1434 	} else
1435 		r = ti->type->map(ti, clone);
1436 
1437 	switch (r) {
1438 	case DM_MAPIO_SUBMITTED:
1439 		/* target has assumed ownership of this io */
1440 		if (!ti->accounts_remapped_io)
1441 			dm_start_io_acct(io, clone);
1442 		break;
1443 	case DM_MAPIO_REMAPPED:
1444 		dm_submit_bio_remap(clone, NULL);
1445 		break;
1446 	case DM_MAPIO_KILL:
1447 	case DM_MAPIO_REQUEUE:
1448 		if (static_branch_unlikely(&swap_bios_enabled) &&
1449 		    unlikely(swap_bios_limit(ti, clone)))
1450 			up(&md->swap_bios_semaphore);
1451 		free_tio(clone);
1452 		if (r == DM_MAPIO_KILL)
1453 			dm_io_dec_pending(io, BLK_STS_IOERR);
1454 		else
1455 			dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1456 		break;
1457 	default:
1458 		DMCRIT("unimplemented target map return value: %d", r);
1459 		BUG();
1460 	}
1461 }
1462 
setup_split_accounting(struct clone_info * ci,unsigned len)1463 static void setup_split_accounting(struct clone_info *ci, unsigned len)
1464 {
1465 	struct dm_io *io = ci->io;
1466 
1467 	if (ci->sector_count > len) {
1468 		/*
1469 		 * Split needed, save the mapped part for accounting.
1470 		 * NOTE: dm_accept_partial_bio() will update accordingly.
1471 		 */
1472 		dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1473 		io->sectors = len;
1474 		io->sector_offset = bio_sectors(ci->bio);
1475 	}
1476 }
1477 
alloc_multiple_bios(struct bio_list * blist,struct clone_info * ci,struct dm_target * ti,unsigned num_bios)1478 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1479 				struct dm_target *ti, unsigned num_bios)
1480 {
1481 	struct bio *bio;
1482 	int try;
1483 
1484 	for (try = 0; try < 2; try++) {
1485 		int bio_nr;
1486 
1487 		if (try)
1488 			mutex_lock(&ci->io->md->table_devices_lock);
1489 		for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1490 			bio = alloc_tio(ci, ti, bio_nr, NULL,
1491 					try ? GFP_NOIO : GFP_NOWAIT);
1492 			if (!bio)
1493 				break;
1494 
1495 			bio_list_add(blist, bio);
1496 		}
1497 		if (try)
1498 			mutex_unlock(&ci->io->md->table_devices_lock);
1499 		if (bio_nr == num_bios)
1500 			return;
1501 
1502 		while ((bio = bio_list_pop(blist)))
1503 			free_tio(bio);
1504 	}
1505 }
1506 
__send_duplicate_bios(struct clone_info * ci,struct dm_target * ti,unsigned int num_bios,unsigned * len)1507 static int __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1508 				 unsigned int num_bios, unsigned *len)
1509 {
1510 	struct bio_list blist = BIO_EMPTY_LIST;
1511 	struct bio *clone;
1512 	unsigned int ret = 0;
1513 
1514 	switch (num_bios) {
1515 	case 0:
1516 		break;
1517 	case 1:
1518 		if (len)
1519 			setup_split_accounting(ci, *len);
1520 		clone = alloc_tio(ci, ti, 0, len, GFP_NOIO);
1521 		__map_bio(clone);
1522 		ret = 1;
1523 		break;
1524 	default:
1525 		/* dm_accept_partial_bio() is not supported with shared tio->len_ptr */
1526 		alloc_multiple_bios(&blist, ci, ti, num_bios);
1527 		while ((clone = bio_list_pop(&blist))) {
1528 			dm_tio_set_flag(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO);
1529 			__map_bio(clone);
1530 			ret += 1;
1531 		}
1532 		break;
1533 	}
1534 
1535 	return ret;
1536 }
1537 
__send_empty_flush(struct clone_info * ci)1538 static void __send_empty_flush(struct clone_info *ci)
1539 {
1540 	struct dm_table *t = ci->map;
1541 	struct bio flush_bio;
1542 
1543 	/*
1544 	 * Use an on-stack bio for this, it's safe since we don't
1545 	 * need to reference it after submit. It's just used as
1546 	 * the basis for the clone(s).
1547 	 */
1548 	bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0,
1549 		 REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC);
1550 
1551 	ci->bio = &flush_bio;
1552 	ci->sector_count = 0;
1553 	ci->io->tio.clone.bi_iter.bi_size = 0;
1554 
1555 	for (unsigned int i = 0; i < t->num_targets; i++) {
1556 		unsigned int bios;
1557 		struct dm_target *ti = dm_table_get_target(t, i);
1558 
1559 		atomic_add(ti->num_flush_bios, &ci->io->io_count);
1560 		bios = __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1561 		atomic_sub(ti->num_flush_bios - bios, &ci->io->io_count);
1562 	}
1563 
1564 	/*
1565 	 * alloc_io() takes one extra reference for submission, so the
1566 	 * reference won't reach 0 without the following subtraction
1567 	 */
1568 	atomic_sub(1, &ci->io->io_count);
1569 
1570 	bio_uninit(ci->bio);
1571 }
1572 
__send_changing_extent_only(struct clone_info * ci,struct dm_target * ti,unsigned num_bios)1573 static void __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1574 					unsigned num_bios)
1575 {
1576 	unsigned len;
1577 	unsigned int bios;
1578 
1579 	len = min_t(sector_t, ci->sector_count,
1580 		    max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1581 
1582 	atomic_add(num_bios, &ci->io->io_count);
1583 	bios = __send_duplicate_bios(ci, ti, num_bios, &len);
1584 	/*
1585 	 * alloc_io() takes one extra reference for submission, so the
1586 	 * reference won't reach 0 without the following (+1) subtraction
1587 	 */
1588 	atomic_sub(num_bios - bios + 1, &ci->io->io_count);
1589 
1590 	ci->sector += len;
1591 	ci->sector_count -= len;
1592 }
1593 
is_abnormal_io(struct bio * bio)1594 static bool is_abnormal_io(struct bio *bio)
1595 {
1596 	enum req_op op = bio_op(bio);
1597 
1598 	if (op != REQ_OP_READ && op != REQ_OP_WRITE && op != REQ_OP_FLUSH) {
1599 		switch (op) {
1600 		case REQ_OP_DISCARD:
1601 		case REQ_OP_SECURE_ERASE:
1602 		case REQ_OP_WRITE_ZEROES:
1603 			return true;
1604 		default:
1605 			break;
1606 		}
1607 	}
1608 
1609 	return false;
1610 }
1611 
__process_abnormal_io(struct clone_info * ci,struct dm_target * ti)1612 static blk_status_t __process_abnormal_io(struct clone_info *ci,
1613 					  struct dm_target *ti)
1614 {
1615 	unsigned num_bios = 0;
1616 
1617 	switch (bio_op(ci->bio)) {
1618 	case REQ_OP_DISCARD:
1619 		num_bios = ti->num_discard_bios;
1620 		break;
1621 	case REQ_OP_SECURE_ERASE:
1622 		num_bios = ti->num_secure_erase_bios;
1623 		break;
1624 	case REQ_OP_WRITE_ZEROES:
1625 		num_bios = ti->num_write_zeroes_bios;
1626 		break;
1627 	default:
1628 		break;
1629 	}
1630 
1631 	/*
1632 	 * Even though the device advertised support for this type of
1633 	 * request, that does not mean every target supports it, and
1634 	 * reconfiguration might also have changed that since the
1635 	 * check was performed.
1636 	 */
1637 	if (unlikely(!num_bios))
1638 		return BLK_STS_NOTSUPP;
1639 
1640 	__send_changing_extent_only(ci, ti, num_bios);
1641 	return BLK_STS_OK;
1642 }
1643 
1644 /*
1645  * Reuse ->bi_private as dm_io list head for storing all dm_io instances
1646  * associated with this bio, and this bio's bi_private needs to be
1647  * stored in dm_io->data before the reuse.
1648  *
1649  * bio->bi_private is owned by fs or upper layer, so block layer won't
1650  * touch it after splitting. Meantime it won't be changed by anyone after
1651  * bio is submitted. So this reuse is safe.
1652  */
dm_poll_list_head(struct bio * bio)1653 static inline struct dm_io **dm_poll_list_head(struct bio *bio)
1654 {
1655 	return (struct dm_io **)&bio->bi_private;
1656 }
1657 
dm_queue_poll_io(struct bio * bio,struct dm_io * io)1658 static void dm_queue_poll_io(struct bio *bio, struct dm_io *io)
1659 {
1660 	struct dm_io **head = dm_poll_list_head(bio);
1661 
1662 	if (!(bio->bi_opf & REQ_DM_POLL_LIST)) {
1663 		bio->bi_opf |= REQ_DM_POLL_LIST;
1664 		/*
1665 		 * Save .bi_private into dm_io, so that we can reuse
1666 		 * .bi_private as dm_io list head for storing dm_io list
1667 		 */
1668 		io->data = bio->bi_private;
1669 
1670 		/* tell block layer to poll for completion */
1671 		bio->bi_cookie = ~BLK_QC_T_NONE;
1672 
1673 		io->next = NULL;
1674 	} else {
1675 		/*
1676 		 * bio recursed due to split, reuse original poll list,
1677 		 * and save bio->bi_private too.
1678 		 */
1679 		io->data = (*head)->data;
1680 		io->next = *head;
1681 	}
1682 
1683 	*head = io;
1684 }
1685 
1686 /*
1687  * Select the correct strategy for processing a non-flush bio.
1688  */
__split_and_process_bio(struct clone_info * ci)1689 static blk_status_t __split_and_process_bio(struct clone_info *ci)
1690 {
1691 	struct bio *clone;
1692 	struct dm_target *ti;
1693 	unsigned len;
1694 
1695 	ti = dm_table_find_target(ci->map, ci->sector);
1696 	if (unlikely(!ti))
1697 		return BLK_STS_IOERR;
1698 
1699 	if (unlikely((ci->bio->bi_opf & REQ_NOWAIT) != 0) &&
1700 	    unlikely(!dm_target_supports_nowait(ti->type)))
1701 		return BLK_STS_NOTSUPP;
1702 
1703 	if (unlikely(ci->is_abnormal_io))
1704 		return __process_abnormal_io(ci, ti);
1705 
1706 	/*
1707 	 * Only support bio polling for normal IO, and the target io is
1708 	 * exactly inside the dm_io instance (verified in dm_poll_dm_io)
1709 	 */
1710 	ci->submit_as_polled = !!(ci->bio->bi_opf & REQ_POLLED);
1711 
1712 	len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1713 	setup_split_accounting(ci, len);
1714 	clone = alloc_tio(ci, ti, 0, &len, GFP_NOIO);
1715 	__map_bio(clone);
1716 
1717 	ci->sector += len;
1718 	ci->sector_count -= len;
1719 
1720 	return BLK_STS_OK;
1721 }
1722 
init_clone_info(struct clone_info * ci,struct mapped_device * md,struct dm_table * map,struct bio * bio,bool is_abnormal)1723 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1724 			    struct dm_table *map, struct bio *bio, bool is_abnormal)
1725 {
1726 	ci->map = map;
1727 	ci->io = alloc_io(md, bio);
1728 	ci->bio = bio;
1729 	ci->is_abnormal_io = is_abnormal;
1730 	ci->submit_as_polled = false;
1731 	ci->sector = bio->bi_iter.bi_sector;
1732 	ci->sector_count = bio_sectors(bio);
1733 
1734 	/* Shouldn't happen but sector_count was being set to 0 so... */
1735 	if (static_branch_unlikely(&zoned_enabled) &&
1736 	    WARN_ON_ONCE(op_is_zone_mgmt(bio_op(bio)) && ci->sector_count))
1737 		ci->sector_count = 0;
1738 }
1739 
1740 /*
1741  * Entry point to split a bio into clones and submit them to the targets.
1742  */
dm_split_and_process_bio(struct mapped_device * md,struct dm_table * map,struct bio * bio)1743 static void dm_split_and_process_bio(struct mapped_device *md,
1744 				     struct dm_table *map, struct bio *bio)
1745 {
1746 	struct clone_info ci;
1747 	struct dm_io *io;
1748 	blk_status_t error = BLK_STS_OK;
1749 	bool is_abnormal;
1750 
1751 	is_abnormal = is_abnormal_io(bio);
1752 	if (unlikely(is_abnormal)) {
1753 		/*
1754 		 * Use bio_split_to_limits() for abnormal IO (e.g. discard, etc)
1755 		 * otherwise associated queue_limits won't be imposed.
1756 		 */
1757 		bio = bio_split_to_limits(bio);
1758 		if (!bio)
1759 			return;
1760 	}
1761 
1762 	init_clone_info(&ci, md, map, bio, is_abnormal);
1763 	io = ci.io;
1764 
1765 	if (bio->bi_opf & REQ_PREFLUSH) {
1766 		__send_empty_flush(&ci);
1767 		/* dm_io_complete submits any data associated with flush */
1768 		goto out;
1769 	}
1770 
1771 	error = __split_and_process_bio(&ci);
1772 	if (error || !ci.sector_count)
1773 		goto out;
1774 	/*
1775 	 * Remainder must be passed to submit_bio_noacct() so it gets handled
1776 	 * *after* bios already submitted have been completely processed.
1777 	 */
1778 	bio_trim(bio, io->sectors, ci.sector_count);
1779 	trace_block_split(bio, bio->bi_iter.bi_sector);
1780 	bio_inc_remaining(bio);
1781 	submit_bio_noacct(bio);
1782 out:
1783 	/*
1784 	 * Drop the extra reference count for non-POLLED bio, and hold one
1785 	 * reference for POLLED bio, which will be released in dm_poll_bio
1786 	 *
1787 	 * Add every dm_io instance into the dm_io list head which is stored
1788 	 * in bio->bi_private, so that dm_poll_bio can poll them all.
1789 	 */
1790 	if (error || !ci.submit_as_polled) {
1791 		/*
1792 		 * In case of submission failure, the extra reference for
1793 		 * submitting io isn't consumed yet
1794 		 */
1795 		if (error)
1796 			atomic_dec(&io->io_count);
1797 		dm_io_dec_pending(io, error);
1798 	} else
1799 		dm_queue_poll_io(bio, io);
1800 }
1801 
dm_submit_bio(struct bio * bio)1802 static void dm_submit_bio(struct bio *bio)
1803 {
1804 	struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1805 	int srcu_idx;
1806 	struct dm_table *map;
1807 	blk_opf_t bio_opf = bio->bi_opf;
1808 
1809 	map = dm_get_live_table_bio(md, &srcu_idx, bio_opf);
1810 
1811 	/* If suspended, or map not yet available, queue this IO for later */
1812 	if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1813 	    unlikely(!map)) {
1814 		if (bio->bi_opf & REQ_NOWAIT)
1815 			bio_wouldblock_error(bio);
1816 		else if (bio->bi_opf & REQ_RAHEAD)
1817 			bio_io_error(bio);
1818 		else
1819 			queue_io(md, bio);
1820 		goto out;
1821 	}
1822 
1823 	dm_split_and_process_bio(md, map, bio);
1824 out:
1825 	dm_put_live_table_bio(md, srcu_idx, bio_opf);
1826 }
1827 
dm_poll_dm_io(struct dm_io * io,struct io_comp_batch * iob,unsigned int flags)1828 static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob,
1829 			  unsigned int flags)
1830 {
1831 	WARN_ON_ONCE(!dm_tio_is_normal(&io->tio));
1832 
1833 	/* don't poll if the mapped io is done */
1834 	if (atomic_read(&io->io_count) > 1)
1835 		bio_poll(&io->tio.clone, iob, flags);
1836 
1837 	/* bio_poll holds the last reference */
1838 	return atomic_read(&io->io_count) == 1;
1839 }
1840 
dm_poll_bio(struct bio * bio,struct io_comp_batch * iob,unsigned int flags)1841 static int dm_poll_bio(struct bio *bio, struct io_comp_batch *iob,
1842 		       unsigned int flags)
1843 {
1844 	struct dm_io **head = dm_poll_list_head(bio);
1845 	struct dm_io *list = *head;
1846 	struct dm_io *tmp = NULL;
1847 	struct dm_io *curr, *next;
1848 
1849 	/* Only poll normal bio which was marked as REQ_DM_POLL_LIST */
1850 	if (!(bio->bi_opf & REQ_DM_POLL_LIST))
1851 		return 0;
1852 
1853 	WARN_ON_ONCE(!list);
1854 
1855 	/*
1856 	 * Restore .bi_private before possibly completing dm_io.
1857 	 *
1858 	 * bio_poll() is only possible once @bio has been completely
1859 	 * submitted via submit_bio_noacct()'s depth-first submission.
1860 	 * So there is no dm_queue_poll_io() race associated with
1861 	 * clearing REQ_DM_POLL_LIST here.
1862 	 */
1863 	bio->bi_opf &= ~REQ_DM_POLL_LIST;
1864 	bio->bi_private = list->data;
1865 
1866 	for (curr = list, next = curr->next; curr; curr = next, next =
1867 			curr ? curr->next : NULL) {
1868 		if (dm_poll_dm_io(curr, iob, flags)) {
1869 			/*
1870 			 * clone_endio() has already occurred, so no
1871 			 * error handling is needed here.
1872 			 */
1873 			__dm_io_dec_pending(curr);
1874 		} else {
1875 			curr->next = tmp;
1876 			tmp = curr;
1877 		}
1878 	}
1879 
1880 	/* Not done? */
1881 	if (tmp) {
1882 		bio->bi_opf |= REQ_DM_POLL_LIST;
1883 		/* Reset bio->bi_private to dm_io list head */
1884 		*head = tmp;
1885 		return 0;
1886 	}
1887 	return 1;
1888 }
1889 
1890 /*-----------------------------------------------------------------
1891  * An IDR is used to keep track of allocated minor numbers.
1892  *---------------------------------------------------------------*/
free_minor(int minor)1893 static void free_minor(int minor)
1894 {
1895 	spin_lock(&_minor_lock);
1896 	idr_remove(&_minor_idr, minor);
1897 	spin_unlock(&_minor_lock);
1898 }
1899 
1900 /*
1901  * See if the device with a specific minor # is free.
1902  */
specific_minor(int minor)1903 static int specific_minor(int minor)
1904 {
1905 	int r;
1906 
1907 	if (minor >= (1 << MINORBITS))
1908 		return -EINVAL;
1909 
1910 	idr_preload(GFP_KERNEL);
1911 	spin_lock(&_minor_lock);
1912 
1913 	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1914 
1915 	spin_unlock(&_minor_lock);
1916 	idr_preload_end();
1917 	if (r < 0)
1918 		return r == -ENOSPC ? -EBUSY : r;
1919 	return 0;
1920 }
1921 
next_free_minor(int * minor)1922 static int next_free_minor(int *minor)
1923 {
1924 	int r;
1925 
1926 	idr_preload(GFP_KERNEL);
1927 	spin_lock(&_minor_lock);
1928 
1929 	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1930 
1931 	spin_unlock(&_minor_lock);
1932 	idr_preload_end();
1933 	if (r < 0)
1934 		return r;
1935 	*minor = r;
1936 	return 0;
1937 }
1938 
1939 static const struct block_device_operations dm_blk_dops;
1940 static const struct block_device_operations dm_rq_blk_dops;
1941 static const struct dax_operations dm_dax_ops;
1942 
1943 static void dm_wq_work(struct work_struct *work);
1944 
1945 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
dm_queue_destroy_crypto_profile(struct request_queue * q)1946 static void dm_queue_destroy_crypto_profile(struct request_queue *q)
1947 {
1948 	dm_destroy_crypto_profile(q->crypto_profile);
1949 }
1950 
1951 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1952 
dm_queue_destroy_crypto_profile(struct request_queue * q)1953 static inline void dm_queue_destroy_crypto_profile(struct request_queue *q)
1954 {
1955 }
1956 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1957 
cleanup_mapped_device(struct mapped_device * md)1958 static void cleanup_mapped_device(struct mapped_device *md)
1959 {
1960 	if (md->wq)
1961 		destroy_workqueue(md->wq);
1962 	dm_free_md_mempools(md->mempools);
1963 
1964 	if (md->dax_dev) {
1965 		dax_remove_host(md->disk);
1966 		kill_dax(md->dax_dev);
1967 		put_dax(md->dax_dev);
1968 		md->dax_dev = NULL;
1969 	}
1970 
1971 	dm_cleanup_zoned_dev(md);
1972 	if (md->disk) {
1973 		spin_lock(&_minor_lock);
1974 		md->disk->private_data = NULL;
1975 		spin_unlock(&_minor_lock);
1976 		if (dm_get_md_type(md) != DM_TYPE_NONE) {
1977 			struct table_device *td;
1978 
1979 			dm_sysfs_exit(md);
1980 			list_for_each_entry(td, &md->table_devices, list) {
1981 				bd_unlink_disk_holder(td->dm_dev.bdev,
1982 						      md->disk);
1983 			}
1984 
1985 			/*
1986 			 * Hold lock to make sure del_gendisk() won't concurrent
1987 			 * with open/close_table_device().
1988 			 */
1989 			mutex_lock(&md->table_devices_lock);
1990 			del_gendisk(md->disk);
1991 			mutex_unlock(&md->table_devices_lock);
1992 		}
1993 		dm_queue_destroy_crypto_profile(md->queue);
1994 		put_disk(md->disk);
1995 	}
1996 
1997 	if (md->pending_io) {
1998 		free_percpu(md->pending_io);
1999 		md->pending_io = NULL;
2000 	}
2001 
2002 	cleanup_srcu_struct(&md->io_barrier);
2003 
2004 	mutex_destroy(&md->suspend_lock);
2005 	mutex_destroy(&md->type_lock);
2006 	mutex_destroy(&md->table_devices_lock);
2007 	mutex_destroy(&md->swap_bios_lock);
2008 
2009 	dm_mq_cleanup_mapped_device(md);
2010 }
2011 
2012 /*
2013  * Allocate and initialise a blank device with a given minor.
2014  */
alloc_dev(int minor)2015 static struct mapped_device *alloc_dev(int minor)
2016 {
2017 	int r, numa_node_id = dm_get_numa_node();
2018 	struct mapped_device *md;
2019 	void *old_md;
2020 
2021 	md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
2022 	if (!md) {
2023 		DMERR("unable to allocate device, out of memory.");
2024 		return NULL;
2025 	}
2026 
2027 	if (!try_module_get(THIS_MODULE))
2028 		goto bad_module_get;
2029 
2030 	/* get a minor number for the dev */
2031 	if (minor == DM_ANY_MINOR)
2032 		r = next_free_minor(&minor);
2033 	else
2034 		r = specific_minor(minor);
2035 	if (r < 0)
2036 		goto bad_minor;
2037 
2038 	r = init_srcu_struct(&md->io_barrier);
2039 	if (r < 0)
2040 		goto bad_io_barrier;
2041 
2042 	md->numa_node_id = numa_node_id;
2043 	md->init_tio_pdu = false;
2044 	md->type = DM_TYPE_NONE;
2045 	mutex_init(&md->suspend_lock);
2046 	mutex_init(&md->type_lock);
2047 	mutex_init(&md->table_devices_lock);
2048 	spin_lock_init(&md->deferred_lock);
2049 	atomic_set(&md->holders, 1);
2050 	atomic_set(&md->open_count, 0);
2051 	atomic_set(&md->event_nr, 0);
2052 	atomic_set(&md->uevent_seq, 0);
2053 	INIT_LIST_HEAD(&md->uevent_list);
2054 	INIT_LIST_HEAD(&md->table_devices);
2055 	spin_lock_init(&md->uevent_lock);
2056 
2057 	/*
2058 	 * default to bio-based until DM table is loaded and md->type
2059 	 * established. If request-based table is loaded: blk-mq will
2060 	 * override accordingly.
2061 	 */
2062 	md->disk = blk_alloc_disk(md->numa_node_id);
2063 	if (!md->disk)
2064 		goto bad;
2065 	md->queue = md->disk->queue;
2066 
2067 	init_waitqueue_head(&md->wait);
2068 	INIT_WORK(&md->work, dm_wq_work);
2069 	INIT_WORK(&md->requeue_work, dm_wq_requeue_work);
2070 	init_waitqueue_head(&md->eventq);
2071 	init_completion(&md->kobj_holder.completion);
2072 
2073 	md->requeue_list = NULL;
2074 	md->swap_bios = get_swap_bios();
2075 	sema_init(&md->swap_bios_semaphore, md->swap_bios);
2076 	mutex_init(&md->swap_bios_lock);
2077 
2078 	md->disk->major = _major;
2079 	md->disk->first_minor = minor;
2080 	md->disk->minors = 1;
2081 	md->disk->flags |= GENHD_FL_NO_PART;
2082 	md->disk->fops = &dm_blk_dops;
2083 	md->disk->private_data = md;
2084 	sprintf(md->disk->disk_name, "dm-%d", minor);
2085 
2086 	if (IS_ENABLED(CONFIG_FS_DAX)) {
2087 		md->dax_dev = alloc_dax(md, &dm_dax_ops);
2088 		if (IS_ERR(md->dax_dev)) {
2089 			md->dax_dev = NULL;
2090 			goto bad;
2091 		}
2092 		set_dax_nocache(md->dax_dev);
2093 		set_dax_nomc(md->dax_dev);
2094 		if (dax_add_host(md->dax_dev, md->disk))
2095 			goto bad;
2096 	}
2097 
2098 	format_dev_t(md->name, MKDEV(_major, minor));
2099 
2100 	md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name);
2101 	if (!md->wq)
2102 		goto bad;
2103 
2104 	md->pending_io = alloc_percpu(unsigned long);
2105 	if (!md->pending_io)
2106 		goto bad;
2107 
2108 	dm_stats_init(&md->stats);
2109 
2110 	/* Populate the mapping, nobody knows we exist yet */
2111 	spin_lock(&_minor_lock);
2112 	old_md = idr_replace(&_minor_idr, md, minor);
2113 	spin_unlock(&_minor_lock);
2114 
2115 	BUG_ON(old_md != MINOR_ALLOCED);
2116 
2117 	return md;
2118 
2119 bad:
2120 	cleanup_mapped_device(md);
2121 bad_io_barrier:
2122 	free_minor(minor);
2123 bad_minor:
2124 	module_put(THIS_MODULE);
2125 bad_module_get:
2126 	kvfree(md);
2127 	return NULL;
2128 }
2129 
2130 static void unlock_fs(struct mapped_device *md);
2131 
free_dev(struct mapped_device * md)2132 static void free_dev(struct mapped_device *md)
2133 {
2134 	int minor = MINOR(disk_devt(md->disk));
2135 
2136 	unlock_fs(md);
2137 
2138 	cleanup_mapped_device(md);
2139 
2140 	free_table_devices(&md->table_devices);
2141 	dm_stats_cleanup(&md->stats);
2142 	free_minor(minor);
2143 
2144 	module_put(THIS_MODULE);
2145 	kvfree(md);
2146 }
2147 
2148 /*
2149  * Bind a table to the device.
2150  */
event_callback(void * context)2151 static void event_callback(void *context)
2152 {
2153 	unsigned long flags;
2154 	LIST_HEAD(uevents);
2155 	struct mapped_device *md = (struct mapped_device *) context;
2156 
2157 	spin_lock_irqsave(&md->uevent_lock, flags);
2158 	list_splice_init(&md->uevent_list, &uevents);
2159 	spin_unlock_irqrestore(&md->uevent_lock, flags);
2160 
2161 	dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2162 
2163 	atomic_inc(&md->event_nr);
2164 	wake_up(&md->eventq);
2165 	dm_issue_global_event();
2166 }
2167 
2168 /*
2169  * Returns old map, which caller must destroy.
2170  */
__bind(struct mapped_device * md,struct dm_table * t,struct queue_limits * limits)2171 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2172 			       struct queue_limits *limits)
2173 {
2174 	struct dm_table *old_map;
2175 	sector_t size;
2176 	int ret;
2177 
2178 	lockdep_assert_held(&md->suspend_lock);
2179 
2180 	size = dm_table_get_size(t);
2181 
2182 	/*
2183 	 * Wipe any geometry if the size of the table changed.
2184 	 */
2185 	if (size != dm_get_size(md))
2186 		memset(&md->geometry, 0, sizeof(md->geometry));
2187 
2188 	if (!get_capacity(md->disk))
2189 		set_capacity(md->disk, size);
2190 	else
2191 		set_capacity_and_notify(md->disk, size);
2192 
2193 	dm_table_event_callback(t, event_callback, md);
2194 
2195 	if (dm_table_request_based(t)) {
2196 		/*
2197 		 * Leverage the fact that request-based DM targets are
2198 		 * immutable singletons - used to optimize dm_mq_queue_rq.
2199 		 */
2200 		md->immutable_target = dm_table_get_immutable_target(t);
2201 
2202 		/*
2203 		 * There is no need to reload with request-based dm because the
2204 		 * size of front_pad doesn't change.
2205 		 *
2206 		 * Note for future: If you are to reload bioset, prep-ed
2207 		 * requests in the queue may refer to bio from the old bioset,
2208 		 * so you must walk through the queue to unprep.
2209 		 */
2210 		if (!md->mempools) {
2211 			md->mempools = t->mempools;
2212 			t->mempools = NULL;
2213 		}
2214 	} else {
2215 		/*
2216 		 * The md may already have mempools that need changing.
2217 		 * If so, reload bioset because front_pad may have changed
2218 		 * because a different table was loaded.
2219 		 */
2220 		dm_free_md_mempools(md->mempools);
2221 		md->mempools = t->mempools;
2222 		t->mempools = NULL;
2223 	}
2224 
2225 	ret = dm_table_set_restrictions(t, md->queue, limits);
2226 	if (ret) {
2227 		old_map = ERR_PTR(ret);
2228 		goto out;
2229 	}
2230 
2231 	old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2232 	rcu_assign_pointer(md->map, (void *)t);
2233 	md->immutable_target_type = dm_table_get_immutable_target_type(t);
2234 
2235 	if (old_map)
2236 		dm_sync_table(md);
2237 out:
2238 	return old_map;
2239 }
2240 
2241 /*
2242  * Returns unbound table for the caller to free.
2243  */
__unbind(struct mapped_device * md)2244 static struct dm_table *__unbind(struct mapped_device *md)
2245 {
2246 	struct dm_table *map = rcu_dereference_protected(md->map, 1);
2247 
2248 	if (!map)
2249 		return NULL;
2250 
2251 	dm_table_event_callback(map, NULL, NULL);
2252 	RCU_INIT_POINTER(md->map, NULL);
2253 	dm_sync_table(md);
2254 
2255 	return map;
2256 }
2257 
2258 /*
2259  * Constructor for a new device.
2260  */
dm_create(int minor,struct mapped_device ** result)2261 int dm_create(int minor, struct mapped_device **result)
2262 {
2263 	struct mapped_device *md;
2264 
2265 	md = alloc_dev(minor);
2266 	if (!md)
2267 		return -ENXIO;
2268 
2269 	dm_ima_reset_data(md);
2270 
2271 	*result = md;
2272 	return 0;
2273 }
2274 
2275 /*
2276  * Functions to manage md->type.
2277  * All are required to hold md->type_lock.
2278  */
dm_lock_md_type(struct mapped_device * md)2279 void dm_lock_md_type(struct mapped_device *md)
2280 {
2281 	mutex_lock(&md->type_lock);
2282 }
2283 
dm_unlock_md_type(struct mapped_device * md)2284 void dm_unlock_md_type(struct mapped_device *md)
2285 {
2286 	mutex_unlock(&md->type_lock);
2287 }
2288 
dm_set_md_type(struct mapped_device * md,enum dm_queue_mode type)2289 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2290 {
2291 	BUG_ON(!mutex_is_locked(&md->type_lock));
2292 	md->type = type;
2293 }
2294 
dm_get_md_type(struct mapped_device * md)2295 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2296 {
2297 	return md->type;
2298 }
2299 
dm_get_immutable_target_type(struct mapped_device * md)2300 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2301 {
2302 	return md->immutable_target_type;
2303 }
2304 
2305 /*
2306  * The queue_limits are only valid as long as you have a reference
2307  * count on 'md'.
2308  */
dm_get_queue_limits(struct mapped_device * md)2309 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2310 {
2311 	BUG_ON(!atomic_read(&md->holders));
2312 	return &md->queue->limits;
2313 }
2314 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2315 
2316 /*
2317  * Setup the DM device's queue based on md's type
2318  */
dm_setup_md_queue(struct mapped_device * md,struct dm_table * t)2319 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2320 {
2321 	enum dm_queue_mode type = dm_table_get_type(t);
2322 	struct queue_limits limits;
2323 	struct table_device *td;
2324 	int r;
2325 
2326 	switch (type) {
2327 	case DM_TYPE_REQUEST_BASED:
2328 		md->disk->fops = &dm_rq_blk_dops;
2329 		r = dm_mq_init_request_queue(md, t);
2330 		if (r) {
2331 			DMERR("Cannot initialize queue for request-based dm mapped device");
2332 			return r;
2333 		}
2334 		break;
2335 	case DM_TYPE_BIO_BASED:
2336 	case DM_TYPE_DAX_BIO_BASED:
2337 		break;
2338 	case DM_TYPE_NONE:
2339 		WARN_ON_ONCE(true);
2340 		break;
2341 	}
2342 
2343 	r = dm_calculate_queue_limits(t, &limits);
2344 	if (r) {
2345 		DMERR("Cannot calculate initial queue limits");
2346 		return r;
2347 	}
2348 	r = dm_table_set_restrictions(t, md->queue, &limits);
2349 	if (r)
2350 		return r;
2351 
2352 	/*
2353 	 * Hold lock to make sure add_disk() and del_gendisk() won't concurrent
2354 	 * with open_table_device() and close_table_device().
2355 	 */
2356 	mutex_lock(&md->table_devices_lock);
2357 	r = add_disk(md->disk);
2358 	mutex_unlock(&md->table_devices_lock);
2359 	if (r)
2360 		return r;
2361 
2362 	/*
2363 	 * Register the holder relationship for devices added before the disk
2364 	 * was live.
2365 	 */
2366 	list_for_each_entry(td, &md->table_devices, list) {
2367 		r = bd_link_disk_holder(td->dm_dev.bdev, md->disk);
2368 		if (r)
2369 			goto out_undo_holders;
2370 	}
2371 
2372 	r = dm_sysfs_init(md);
2373 	if (r)
2374 		goto out_undo_holders;
2375 
2376 	md->type = type;
2377 	return 0;
2378 
2379 out_undo_holders:
2380 	list_for_each_entry_continue_reverse(td, &md->table_devices, list)
2381 		bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
2382 	mutex_lock(&md->table_devices_lock);
2383 	del_gendisk(md->disk);
2384 	mutex_unlock(&md->table_devices_lock);
2385 	return r;
2386 }
2387 
dm_get_md(dev_t dev)2388 struct mapped_device *dm_get_md(dev_t dev)
2389 {
2390 	struct mapped_device *md;
2391 	unsigned minor = MINOR(dev);
2392 
2393 	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2394 		return NULL;
2395 
2396 	spin_lock(&_minor_lock);
2397 
2398 	md = idr_find(&_minor_idr, minor);
2399 	if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2400 	    test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2401 		md = NULL;
2402 		goto out;
2403 	}
2404 	dm_get(md);
2405 out:
2406 	spin_unlock(&_minor_lock);
2407 
2408 	return md;
2409 }
2410 EXPORT_SYMBOL_GPL(dm_get_md);
2411 
dm_get_mdptr(struct mapped_device * md)2412 void *dm_get_mdptr(struct mapped_device *md)
2413 {
2414 	return md->interface_ptr;
2415 }
2416 
dm_set_mdptr(struct mapped_device * md,void * ptr)2417 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2418 {
2419 	md->interface_ptr = ptr;
2420 }
2421 
dm_get(struct mapped_device * md)2422 void dm_get(struct mapped_device *md)
2423 {
2424 	atomic_inc(&md->holders);
2425 	BUG_ON(test_bit(DMF_FREEING, &md->flags));
2426 }
2427 
dm_hold(struct mapped_device * md)2428 int dm_hold(struct mapped_device *md)
2429 {
2430 	spin_lock(&_minor_lock);
2431 	if (test_bit(DMF_FREEING, &md->flags)) {
2432 		spin_unlock(&_minor_lock);
2433 		return -EBUSY;
2434 	}
2435 	dm_get(md);
2436 	spin_unlock(&_minor_lock);
2437 	return 0;
2438 }
2439 EXPORT_SYMBOL_GPL(dm_hold);
2440 
dm_device_name(struct mapped_device * md)2441 const char *dm_device_name(struct mapped_device *md)
2442 {
2443 	return md->name;
2444 }
2445 EXPORT_SYMBOL_GPL(dm_device_name);
2446 
__dm_destroy(struct mapped_device * md,bool wait)2447 static void __dm_destroy(struct mapped_device *md, bool wait)
2448 {
2449 	struct dm_table *map;
2450 	int srcu_idx;
2451 
2452 	might_sleep();
2453 
2454 	spin_lock(&_minor_lock);
2455 	idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2456 	set_bit(DMF_FREEING, &md->flags);
2457 	spin_unlock(&_minor_lock);
2458 
2459 	blk_mark_disk_dead(md->disk);
2460 
2461 	/*
2462 	 * Take suspend_lock so that presuspend and postsuspend methods
2463 	 * do not race with internal suspend.
2464 	 */
2465 	mutex_lock(&md->suspend_lock);
2466 	map = dm_get_live_table(md, &srcu_idx);
2467 	if (!dm_suspended_md(md)) {
2468 		dm_table_presuspend_targets(map);
2469 		set_bit(DMF_SUSPENDED, &md->flags);
2470 		set_bit(DMF_POST_SUSPENDING, &md->flags);
2471 		dm_table_postsuspend_targets(map);
2472 	}
2473 	/* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2474 	dm_put_live_table(md, srcu_idx);
2475 	mutex_unlock(&md->suspend_lock);
2476 
2477 	/*
2478 	 * Rare, but there may be I/O requests still going to complete,
2479 	 * for example.  Wait for all references to disappear.
2480 	 * No one should increment the reference count of the mapped_device,
2481 	 * after the mapped_device state becomes DMF_FREEING.
2482 	 */
2483 	if (wait)
2484 		while (atomic_read(&md->holders))
2485 			msleep(1);
2486 	else if (atomic_read(&md->holders))
2487 		DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2488 		       dm_device_name(md), atomic_read(&md->holders));
2489 
2490 	dm_table_destroy(__unbind(md));
2491 	free_dev(md);
2492 }
2493 
dm_destroy(struct mapped_device * md)2494 void dm_destroy(struct mapped_device *md)
2495 {
2496 	__dm_destroy(md, true);
2497 }
2498 
dm_destroy_immediate(struct mapped_device * md)2499 void dm_destroy_immediate(struct mapped_device *md)
2500 {
2501 	__dm_destroy(md, false);
2502 }
2503 
dm_put(struct mapped_device * md)2504 void dm_put(struct mapped_device *md)
2505 {
2506 	atomic_dec(&md->holders);
2507 }
2508 EXPORT_SYMBOL_GPL(dm_put);
2509 
dm_in_flight_bios(struct mapped_device * md)2510 static bool dm_in_flight_bios(struct mapped_device *md)
2511 {
2512 	int cpu;
2513 	unsigned long sum = 0;
2514 
2515 	for_each_possible_cpu(cpu)
2516 		sum += *per_cpu_ptr(md->pending_io, cpu);
2517 
2518 	return sum != 0;
2519 }
2520 
dm_wait_for_bios_completion(struct mapped_device * md,unsigned int task_state)2521 static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2522 {
2523 	int r = 0;
2524 	DEFINE_WAIT(wait);
2525 
2526 	while (true) {
2527 		prepare_to_wait(&md->wait, &wait, task_state);
2528 
2529 		if (!dm_in_flight_bios(md))
2530 			break;
2531 
2532 		if (signal_pending_state(task_state, current)) {
2533 			r = -EINTR;
2534 			break;
2535 		}
2536 
2537 		io_schedule();
2538 	}
2539 	finish_wait(&md->wait, &wait);
2540 
2541 	smp_rmb();
2542 
2543 	return r;
2544 }
2545 
dm_wait_for_completion(struct mapped_device * md,unsigned int task_state)2546 static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2547 {
2548 	int r = 0;
2549 
2550 	if (!queue_is_mq(md->queue))
2551 		return dm_wait_for_bios_completion(md, task_state);
2552 
2553 	while (true) {
2554 		if (!blk_mq_queue_inflight(md->queue))
2555 			break;
2556 
2557 		if (signal_pending_state(task_state, current)) {
2558 			r = -EINTR;
2559 			break;
2560 		}
2561 
2562 		msleep(5);
2563 	}
2564 
2565 	return r;
2566 }
2567 
2568 /*
2569  * Process the deferred bios
2570  */
dm_wq_work(struct work_struct * work)2571 static void dm_wq_work(struct work_struct *work)
2572 {
2573 	struct mapped_device *md = container_of(work, struct mapped_device, work);
2574 	struct bio *bio;
2575 
2576 	while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2577 		spin_lock_irq(&md->deferred_lock);
2578 		bio = bio_list_pop(&md->deferred);
2579 		spin_unlock_irq(&md->deferred_lock);
2580 
2581 		if (!bio)
2582 			break;
2583 
2584 		submit_bio_noacct(bio);
2585 	}
2586 }
2587 
dm_queue_flush(struct mapped_device * md)2588 static void dm_queue_flush(struct mapped_device *md)
2589 {
2590 	clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2591 	smp_mb__after_atomic();
2592 	queue_work(md->wq, &md->work);
2593 }
2594 
2595 /*
2596  * Swap in a new table, returning the old one for the caller to destroy.
2597  */
dm_swap_table(struct mapped_device * md,struct dm_table * table)2598 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2599 {
2600 	struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2601 	struct queue_limits limits;
2602 	int r;
2603 
2604 	mutex_lock(&md->suspend_lock);
2605 
2606 	/* device must be suspended */
2607 	if (!dm_suspended_md(md))
2608 		goto out;
2609 
2610 	/*
2611 	 * If the new table has no data devices, retain the existing limits.
2612 	 * This helps multipath with queue_if_no_path if all paths disappear,
2613 	 * then new I/O is queued based on these limits, and then some paths
2614 	 * reappear.
2615 	 */
2616 	if (dm_table_has_no_data_devices(table)) {
2617 		live_map = dm_get_live_table_fast(md);
2618 		if (live_map)
2619 			limits = md->queue->limits;
2620 		dm_put_live_table_fast(md);
2621 	}
2622 
2623 	if (!live_map) {
2624 		r = dm_calculate_queue_limits(table, &limits);
2625 		if (r) {
2626 			map = ERR_PTR(r);
2627 			goto out;
2628 		}
2629 	}
2630 
2631 	map = __bind(md, table, &limits);
2632 	dm_issue_global_event();
2633 
2634 out:
2635 	mutex_unlock(&md->suspend_lock);
2636 	return map;
2637 }
2638 
2639 /*
2640  * Functions to lock and unlock any filesystem running on the
2641  * device.
2642  */
lock_fs(struct mapped_device * md)2643 static int lock_fs(struct mapped_device *md)
2644 {
2645 	int r;
2646 
2647 	WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2648 
2649 	r = freeze_bdev(md->disk->part0);
2650 	if (!r)
2651 		set_bit(DMF_FROZEN, &md->flags);
2652 	return r;
2653 }
2654 
unlock_fs(struct mapped_device * md)2655 static void unlock_fs(struct mapped_device *md)
2656 {
2657 	if (!test_bit(DMF_FROZEN, &md->flags))
2658 		return;
2659 	thaw_bdev(md->disk->part0);
2660 	clear_bit(DMF_FROZEN, &md->flags);
2661 }
2662 
2663 /*
2664  * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2665  * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2666  * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2667  *
2668  * If __dm_suspend returns 0, the device is completely quiescent
2669  * now. There is no request-processing activity. All new requests
2670  * are being added to md->deferred list.
2671  */
__dm_suspend(struct mapped_device * md,struct dm_table * map,unsigned suspend_flags,unsigned int task_state,int dmf_suspended_flag)2672 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2673 			unsigned suspend_flags, unsigned int task_state,
2674 			int dmf_suspended_flag)
2675 {
2676 	bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2677 	bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2678 	int r;
2679 
2680 	lockdep_assert_held(&md->suspend_lock);
2681 
2682 	/*
2683 	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2684 	 * This flag is cleared before dm_suspend returns.
2685 	 */
2686 	if (noflush)
2687 		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2688 	else
2689 		DMDEBUG("%s: suspending with flush", dm_device_name(md));
2690 
2691 	/*
2692 	 * This gets reverted if there's an error later and the targets
2693 	 * provide the .presuspend_undo hook.
2694 	 */
2695 	dm_table_presuspend_targets(map);
2696 
2697 	/*
2698 	 * Flush I/O to the device.
2699 	 * Any I/O submitted after lock_fs() may not be flushed.
2700 	 * noflush takes precedence over do_lockfs.
2701 	 * (lock_fs() flushes I/Os and waits for them to complete.)
2702 	 */
2703 	if (!noflush && do_lockfs) {
2704 		r = lock_fs(md);
2705 		if (r) {
2706 			dm_table_presuspend_undo_targets(map);
2707 			return r;
2708 		}
2709 	}
2710 
2711 	/*
2712 	 * Here we must make sure that no processes are submitting requests
2713 	 * to target drivers i.e. no one may be executing
2714 	 * dm_split_and_process_bio from dm_submit_bio.
2715 	 *
2716 	 * To get all processes out of dm_split_and_process_bio in dm_submit_bio,
2717 	 * we take the write lock. To prevent any process from reentering
2718 	 * dm_split_and_process_bio from dm_submit_bio and quiesce the thread
2719 	 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2720 	 * flush_workqueue(md->wq).
2721 	 */
2722 	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2723 	if (map)
2724 		synchronize_srcu(&md->io_barrier);
2725 
2726 	/*
2727 	 * Stop md->queue before flushing md->wq in case request-based
2728 	 * dm defers requests to md->wq from md->queue.
2729 	 */
2730 	if (dm_request_based(md))
2731 		dm_stop_queue(md->queue);
2732 
2733 	flush_workqueue(md->wq);
2734 
2735 	/*
2736 	 * At this point no more requests are entering target request routines.
2737 	 * We call dm_wait_for_completion to wait for all existing requests
2738 	 * to finish.
2739 	 */
2740 	r = dm_wait_for_completion(md, task_state);
2741 	if (!r)
2742 		set_bit(dmf_suspended_flag, &md->flags);
2743 
2744 	if (noflush)
2745 		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2746 	if (map)
2747 		synchronize_srcu(&md->io_barrier);
2748 
2749 	/* were we interrupted ? */
2750 	if (r < 0) {
2751 		dm_queue_flush(md);
2752 
2753 		if (dm_request_based(md))
2754 			dm_start_queue(md->queue);
2755 
2756 		unlock_fs(md);
2757 		dm_table_presuspend_undo_targets(map);
2758 		/* pushback list is already flushed, so skip flush */
2759 	}
2760 
2761 	return r;
2762 }
2763 
2764 /*
2765  * We need to be able to change a mapping table under a mounted
2766  * filesystem.  For example we might want to move some data in
2767  * the background.  Before the table can be swapped with
2768  * dm_bind_table, dm_suspend must be called to flush any in
2769  * flight bios and ensure that any further io gets deferred.
2770  */
2771 /*
2772  * Suspend mechanism in request-based dm.
2773  *
2774  * 1. Flush all I/Os by lock_fs() if needed.
2775  * 2. Stop dispatching any I/O by stopping the request_queue.
2776  * 3. Wait for all in-flight I/Os to be completed or requeued.
2777  *
2778  * To abort suspend, start the request_queue.
2779  */
dm_suspend(struct mapped_device * md,unsigned suspend_flags)2780 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2781 {
2782 	struct dm_table *map = NULL;
2783 	int r = 0;
2784 
2785 retry:
2786 	mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2787 
2788 	if (dm_suspended_md(md)) {
2789 		r = -EINVAL;
2790 		goto out_unlock;
2791 	}
2792 
2793 	if (dm_suspended_internally_md(md)) {
2794 		/* already internally suspended, wait for internal resume */
2795 		mutex_unlock(&md->suspend_lock);
2796 		r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2797 		if (r)
2798 			return r;
2799 		goto retry;
2800 	}
2801 
2802 	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2803 
2804 	r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2805 	if (r)
2806 		goto out_unlock;
2807 
2808 	set_bit(DMF_POST_SUSPENDING, &md->flags);
2809 	dm_table_postsuspend_targets(map);
2810 	clear_bit(DMF_POST_SUSPENDING, &md->flags);
2811 
2812 out_unlock:
2813 	mutex_unlock(&md->suspend_lock);
2814 	return r;
2815 }
2816 
__dm_resume(struct mapped_device * md,struct dm_table * map)2817 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2818 {
2819 	if (map) {
2820 		int r = dm_table_resume_targets(map);
2821 		if (r)
2822 			return r;
2823 	}
2824 
2825 	dm_queue_flush(md);
2826 
2827 	/*
2828 	 * Flushing deferred I/Os must be done after targets are resumed
2829 	 * so that mapping of targets can work correctly.
2830 	 * Request-based dm is queueing the deferred I/Os in its request_queue.
2831 	 */
2832 	if (dm_request_based(md))
2833 		dm_start_queue(md->queue);
2834 
2835 	unlock_fs(md);
2836 
2837 	return 0;
2838 }
2839 
dm_resume(struct mapped_device * md)2840 int dm_resume(struct mapped_device *md)
2841 {
2842 	int r;
2843 	struct dm_table *map = NULL;
2844 
2845 retry:
2846 	r = -EINVAL;
2847 	mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2848 
2849 	if (!dm_suspended_md(md))
2850 		goto out;
2851 
2852 	if (dm_suspended_internally_md(md)) {
2853 		/* already internally suspended, wait for internal resume */
2854 		mutex_unlock(&md->suspend_lock);
2855 		r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2856 		if (r)
2857 			return r;
2858 		goto retry;
2859 	}
2860 
2861 	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2862 	if (!map || !dm_table_get_size(map))
2863 		goto out;
2864 
2865 	r = __dm_resume(md, map);
2866 	if (r)
2867 		goto out;
2868 
2869 	clear_bit(DMF_SUSPENDED, &md->flags);
2870 out:
2871 	mutex_unlock(&md->suspend_lock);
2872 
2873 	return r;
2874 }
2875 
2876 /*
2877  * Internal suspend/resume works like userspace-driven suspend. It waits
2878  * until all bios finish and prevents issuing new bios to the target drivers.
2879  * It may be used only from the kernel.
2880  */
2881 
__dm_internal_suspend(struct mapped_device * md,unsigned suspend_flags)2882 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2883 {
2884 	struct dm_table *map = NULL;
2885 
2886 	lockdep_assert_held(&md->suspend_lock);
2887 
2888 	if (md->internal_suspend_count++)
2889 		return; /* nested internal suspend */
2890 
2891 	if (dm_suspended_md(md)) {
2892 		set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2893 		return; /* nest suspend */
2894 	}
2895 
2896 	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2897 
2898 	/*
2899 	 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2900 	 * supported.  Properly supporting a TASK_INTERRUPTIBLE internal suspend
2901 	 * would require changing .presuspend to return an error -- avoid this
2902 	 * until there is a need for more elaborate variants of internal suspend.
2903 	 */
2904 	(void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2905 			    DMF_SUSPENDED_INTERNALLY);
2906 
2907 	set_bit(DMF_POST_SUSPENDING, &md->flags);
2908 	dm_table_postsuspend_targets(map);
2909 	clear_bit(DMF_POST_SUSPENDING, &md->flags);
2910 }
2911 
__dm_internal_resume(struct mapped_device * md)2912 static void __dm_internal_resume(struct mapped_device *md)
2913 {
2914 	BUG_ON(!md->internal_suspend_count);
2915 
2916 	if (--md->internal_suspend_count)
2917 		return; /* resume from nested internal suspend */
2918 
2919 	if (dm_suspended_md(md))
2920 		goto done; /* resume from nested suspend */
2921 
2922 	/*
2923 	 * NOTE: existing callers don't need to call dm_table_resume_targets
2924 	 * (which may fail -- so best to avoid it for now by passing NULL map)
2925 	 */
2926 	(void) __dm_resume(md, NULL);
2927 
2928 done:
2929 	clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2930 	smp_mb__after_atomic();
2931 	wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2932 }
2933 
dm_internal_suspend_noflush(struct mapped_device * md)2934 void dm_internal_suspend_noflush(struct mapped_device *md)
2935 {
2936 	mutex_lock(&md->suspend_lock);
2937 	__dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2938 	mutex_unlock(&md->suspend_lock);
2939 }
2940 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2941 
dm_internal_resume(struct mapped_device * md)2942 void dm_internal_resume(struct mapped_device *md)
2943 {
2944 	mutex_lock(&md->suspend_lock);
2945 	__dm_internal_resume(md);
2946 	mutex_unlock(&md->suspend_lock);
2947 }
2948 EXPORT_SYMBOL_GPL(dm_internal_resume);
2949 
2950 /*
2951  * Fast variants of internal suspend/resume hold md->suspend_lock,
2952  * which prevents interaction with userspace-driven suspend.
2953  */
2954 
dm_internal_suspend_fast(struct mapped_device * md)2955 void dm_internal_suspend_fast(struct mapped_device *md)
2956 {
2957 	mutex_lock(&md->suspend_lock);
2958 	if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2959 		return;
2960 
2961 	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2962 	synchronize_srcu(&md->io_barrier);
2963 	flush_workqueue(md->wq);
2964 	dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2965 }
2966 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2967 
dm_internal_resume_fast(struct mapped_device * md)2968 void dm_internal_resume_fast(struct mapped_device *md)
2969 {
2970 	if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2971 		goto done;
2972 
2973 	dm_queue_flush(md);
2974 
2975 done:
2976 	mutex_unlock(&md->suspend_lock);
2977 }
2978 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2979 
2980 /*-----------------------------------------------------------------
2981  * Event notification.
2982  *---------------------------------------------------------------*/
dm_kobject_uevent(struct mapped_device * md,enum kobject_action action,unsigned cookie)2983 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2984 		       unsigned cookie)
2985 {
2986 	int r;
2987 	unsigned noio_flag;
2988 	char udev_cookie[DM_COOKIE_LENGTH];
2989 	char *envp[] = { udev_cookie, NULL };
2990 
2991 	noio_flag = memalloc_noio_save();
2992 
2993 	if (!cookie)
2994 		r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2995 	else {
2996 		snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2997 			 DM_COOKIE_ENV_VAR_NAME, cookie);
2998 		r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2999 				       action, envp);
3000 	}
3001 
3002 	memalloc_noio_restore(noio_flag);
3003 
3004 	return r;
3005 }
3006 
dm_next_uevent_seq(struct mapped_device * md)3007 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3008 {
3009 	return atomic_add_return(1, &md->uevent_seq);
3010 }
3011 
dm_get_event_nr(struct mapped_device * md)3012 uint32_t dm_get_event_nr(struct mapped_device *md)
3013 {
3014 	return atomic_read(&md->event_nr);
3015 }
3016 
dm_wait_event(struct mapped_device * md,int event_nr)3017 int dm_wait_event(struct mapped_device *md, int event_nr)
3018 {
3019 	return wait_event_interruptible(md->eventq,
3020 			(event_nr != atomic_read(&md->event_nr)));
3021 }
3022 
dm_uevent_add(struct mapped_device * md,struct list_head * elist)3023 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3024 {
3025 	unsigned long flags;
3026 
3027 	spin_lock_irqsave(&md->uevent_lock, flags);
3028 	list_add(elist, &md->uevent_list);
3029 	spin_unlock_irqrestore(&md->uevent_lock, flags);
3030 }
3031 
3032 /*
3033  * The gendisk is only valid as long as you have a reference
3034  * count on 'md'.
3035  */
dm_disk(struct mapped_device * md)3036 struct gendisk *dm_disk(struct mapped_device *md)
3037 {
3038 	return md->disk;
3039 }
3040 EXPORT_SYMBOL_GPL(dm_disk);
3041 
dm_kobject(struct mapped_device * md)3042 struct kobject *dm_kobject(struct mapped_device *md)
3043 {
3044 	return &md->kobj_holder.kobj;
3045 }
3046 
dm_get_from_kobject(struct kobject * kobj)3047 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3048 {
3049 	struct mapped_device *md;
3050 
3051 	md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3052 
3053 	spin_lock(&_minor_lock);
3054 	if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
3055 		md = NULL;
3056 		goto out;
3057 	}
3058 	dm_get(md);
3059 out:
3060 	spin_unlock(&_minor_lock);
3061 
3062 	return md;
3063 }
3064 
dm_suspended_md(struct mapped_device * md)3065 int dm_suspended_md(struct mapped_device *md)
3066 {
3067 	return test_bit(DMF_SUSPENDED, &md->flags);
3068 }
3069 
dm_post_suspending_md(struct mapped_device * md)3070 static int dm_post_suspending_md(struct mapped_device *md)
3071 {
3072 	return test_bit(DMF_POST_SUSPENDING, &md->flags);
3073 }
3074 
dm_suspended_internally_md(struct mapped_device * md)3075 int dm_suspended_internally_md(struct mapped_device *md)
3076 {
3077 	return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3078 }
3079 
dm_test_deferred_remove_flag(struct mapped_device * md)3080 int dm_test_deferred_remove_flag(struct mapped_device *md)
3081 {
3082 	return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3083 }
3084 
dm_suspended(struct dm_target * ti)3085 int dm_suspended(struct dm_target *ti)
3086 {
3087 	return dm_suspended_md(ti->table->md);
3088 }
3089 EXPORT_SYMBOL_GPL(dm_suspended);
3090 
dm_post_suspending(struct dm_target * ti)3091 int dm_post_suspending(struct dm_target *ti)
3092 {
3093 	return dm_post_suspending_md(ti->table->md);
3094 }
3095 EXPORT_SYMBOL_GPL(dm_post_suspending);
3096 
dm_noflush_suspending(struct dm_target * ti)3097 int dm_noflush_suspending(struct dm_target *ti)
3098 {
3099 	return __noflush_suspending(ti->table->md);
3100 }
3101 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3102 
dm_free_md_mempools(struct dm_md_mempools * pools)3103 void dm_free_md_mempools(struct dm_md_mempools *pools)
3104 {
3105 	if (!pools)
3106 		return;
3107 
3108 	bioset_exit(&pools->bs);
3109 	bioset_exit(&pools->io_bs);
3110 
3111 	kfree(pools);
3112 }
3113 
3114 struct dm_pr {
3115 	u64	old_key;
3116 	u64	new_key;
3117 	u32	flags;
3118 	bool	abort;
3119 	bool	fail_early;
3120 	int	ret;
3121 	enum pr_type type;
3122 };
3123 
dm_call_pr(struct block_device * bdev,iterate_devices_callout_fn fn,struct dm_pr * pr)3124 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3125 		      struct dm_pr *pr)
3126 {
3127 	struct mapped_device *md = bdev->bd_disk->private_data;
3128 	struct dm_table *table;
3129 	struct dm_target *ti;
3130 	int ret = -ENOTTY, srcu_idx;
3131 
3132 	table = dm_get_live_table(md, &srcu_idx);
3133 	if (!table || !dm_table_get_size(table))
3134 		goto out;
3135 
3136 	/* We only support devices that have a single target */
3137 	if (table->num_targets != 1)
3138 		goto out;
3139 	ti = dm_table_get_target(table, 0);
3140 
3141 	if (dm_suspended_md(md)) {
3142 		ret = -EAGAIN;
3143 		goto out;
3144 	}
3145 
3146 	ret = -EINVAL;
3147 	if (!ti->type->iterate_devices)
3148 		goto out;
3149 
3150 	ti->type->iterate_devices(ti, fn, pr);
3151 	ret = 0;
3152 out:
3153 	dm_put_live_table(md, srcu_idx);
3154 	return ret;
3155 }
3156 
3157 /*
3158  * For register / unregister we need to manually call out to every path.
3159  */
__dm_pr_register(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)3160 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3161 			    sector_t start, sector_t len, void *data)
3162 {
3163 	struct dm_pr *pr = data;
3164 	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3165 	int ret;
3166 
3167 	if (!ops || !ops->pr_register) {
3168 		pr->ret = -EOPNOTSUPP;
3169 		return -1;
3170 	}
3171 
3172 	ret = ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3173 	if (!ret)
3174 		return 0;
3175 
3176 	if (!pr->ret)
3177 		pr->ret = ret;
3178 
3179 	if (pr->fail_early)
3180 		return -1;
3181 
3182 	return 0;
3183 }
3184 
dm_pr_register(struct block_device * bdev,u64 old_key,u64 new_key,u32 flags)3185 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3186 			  u32 flags)
3187 {
3188 	struct dm_pr pr = {
3189 		.old_key	= old_key,
3190 		.new_key	= new_key,
3191 		.flags		= flags,
3192 		.fail_early	= true,
3193 		.ret		= 0,
3194 	};
3195 	int ret;
3196 
3197 	ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3198 	if (ret) {
3199 		/* Didn't even get to register a path */
3200 		return ret;
3201 	}
3202 
3203 	if (!pr.ret)
3204 		return 0;
3205 	ret = pr.ret;
3206 
3207 	if (!new_key)
3208 		return ret;
3209 
3210 	/* unregister all paths if we failed to register any path */
3211 	pr.old_key = new_key;
3212 	pr.new_key = 0;
3213 	pr.flags = 0;
3214 	pr.fail_early = false;
3215 	(void) dm_call_pr(bdev, __dm_pr_register, &pr);
3216 	return ret;
3217 }
3218 
3219 
__dm_pr_reserve(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)3220 static int __dm_pr_reserve(struct dm_target *ti, struct dm_dev *dev,
3221 			   sector_t start, sector_t len, void *data)
3222 {
3223 	struct dm_pr *pr = data;
3224 	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3225 
3226 	if (!ops || !ops->pr_reserve) {
3227 		pr->ret = -EOPNOTSUPP;
3228 		return -1;
3229 	}
3230 
3231 	pr->ret = ops->pr_reserve(dev->bdev, pr->old_key, pr->type, pr->flags);
3232 	if (!pr->ret)
3233 		return -1;
3234 
3235 	return 0;
3236 }
3237 
dm_pr_reserve(struct block_device * bdev,u64 key,enum pr_type type,u32 flags)3238 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3239 			 u32 flags)
3240 {
3241 	struct dm_pr pr = {
3242 		.old_key	= key,
3243 		.flags		= flags,
3244 		.type		= type,
3245 		.fail_early	= false,
3246 		.ret		= 0,
3247 	};
3248 	int ret;
3249 
3250 	ret = dm_call_pr(bdev, __dm_pr_reserve, &pr);
3251 	if (ret)
3252 		return ret;
3253 
3254 	return pr.ret;
3255 }
3256 
3257 /*
3258  * If there is a non-All Registrants type of reservation, the release must be
3259  * sent down the holding path. For the cases where there is no reservation or
3260  * the path is not the holder the device will also return success, so we must
3261  * try each path to make sure we got the correct path.
3262  */
__dm_pr_release(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)3263 static int __dm_pr_release(struct dm_target *ti, struct dm_dev *dev,
3264 			   sector_t start, sector_t len, void *data)
3265 {
3266 	struct dm_pr *pr = data;
3267 	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3268 
3269 	if (!ops || !ops->pr_release) {
3270 		pr->ret = -EOPNOTSUPP;
3271 		return -1;
3272 	}
3273 
3274 	pr->ret = ops->pr_release(dev->bdev, pr->old_key, pr->type);
3275 	if (pr->ret)
3276 		return -1;
3277 
3278 	return 0;
3279 }
3280 
dm_pr_release(struct block_device * bdev,u64 key,enum pr_type type)3281 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3282 {
3283 	struct dm_pr pr = {
3284 		.old_key	= key,
3285 		.type		= type,
3286 		.fail_early	= false,
3287 	};
3288 	int ret;
3289 
3290 	ret = dm_call_pr(bdev, __dm_pr_release, &pr);
3291 	if (ret)
3292 		return ret;
3293 
3294 	return pr.ret;
3295 }
3296 
__dm_pr_preempt(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)3297 static int __dm_pr_preempt(struct dm_target *ti, struct dm_dev *dev,
3298 			   sector_t start, sector_t len, void *data)
3299 {
3300 	struct dm_pr *pr = data;
3301 	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3302 
3303 	if (!ops || !ops->pr_preempt) {
3304 		pr->ret = -EOPNOTSUPP;
3305 		return -1;
3306 	}
3307 
3308 	pr->ret = ops->pr_preempt(dev->bdev, pr->old_key, pr->new_key, pr->type,
3309 				  pr->abort);
3310 	if (!pr->ret)
3311 		return -1;
3312 
3313 	return 0;
3314 }
3315 
dm_pr_preempt(struct block_device * bdev,u64 old_key,u64 new_key,enum pr_type type,bool abort)3316 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3317 			 enum pr_type type, bool abort)
3318 {
3319 	struct dm_pr pr = {
3320 		.new_key	= new_key,
3321 		.old_key	= old_key,
3322 		.type		= type,
3323 		.fail_early	= false,
3324 	};
3325 	int ret;
3326 
3327 	ret = dm_call_pr(bdev, __dm_pr_preempt, &pr);
3328 	if (ret)
3329 		return ret;
3330 
3331 	return pr.ret;
3332 }
3333 
dm_pr_clear(struct block_device * bdev,u64 key)3334 static int dm_pr_clear(struct block_device *bdev, u64 key)
3335 {
3336 	struct mapped_device *md = bdev->bd_disk->private_data;
3337 	const struct pr_ops *ops;
3338 	int r, srcu_idx;
3339 
3340 	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3341 	if (r < 0)
3342 		goto out;
3343 
3344 	ops = bdev->bd_disk->fops->pr_ops;
3345 	if (ops && ops->pr_clear)
3346 		r = ops->pr_clear(bdev, key);
3347 	else
3348 		r = -EOPNOTSUPP;
3349 out:
3350 	dm_unprepare_ioctl(md, srcu_idx);
3351 	return r;
3352 }
3353 
3354 static const struct pr_ops dm_pr_ops = {
3355 	.pr_register	= dm_pr_register,
3356 	.pr_reserve	= dm_pr_reserve,
3357 	.pr_release	= dm_pr_release,
3358 	.pr_preempt	= dm_pr_preempt,
3359 	.pr_clear	= dm_pr_clear,
3360 };
3361 
3362 static const struct block_device_operations dm_blk_dops = {
3363 	.submit_bio = dm_submit_bio,
3364 	.poll_bio = dm_poll_bio,
3365 	.open = dm_blk_open,
3366 	.release = dm_blk_close,
3367 	.ioctl = dm_blk_ioctl,
3368 	.getgeo = dm_blk_getgeo,
3369 	.report_zones = dm_blk_report_zones,
3370 	.pr_ops = &dm_pr_ops,
3371 	.owner = THIS_MODULE
3372 };
3373 
3374 static const struct block_device_operations dm_rq_blk_dops = {
3375 	.open = dm_blk_open,
3376 	.release = dm_blk_close,
3377 	.ioctl = dm_blk_ioctl,
3378 	.getgeo = dm_blk_getgeo,
3379 	.pr_ops = &dm_pr_ops,
3380 	.owner = THIS_MODULE
3381 };
3382 
3383 static const struct dax_operations dm_dax_ops = {
3384 	.direct_access = dm_dax_direct_access,
3385 	.zero_page_range = dm_dax_zero_page_range,
3386 	.recovery_write = dm_dax_recovery_write,
3387 };
3388 
3389 /*
3390  * module hooks
3391  */
3392 module_init(dm_init);
3393 module_exit(dm_exit);
3394 
3395 module_param(major, uint, 0);
3396 MODULE_PARM_DESC(major, "The major number of the device mapper");
3397 
3398 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3399 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3400 
3401 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3402 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3403 
3404 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3405 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3406 
3407 MODULE_DESCRIPTION(DM_NAME " driver");
3408 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3409 MODULE_LICENSE("GPL");
3410