]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. | |
4 | * | |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
9 | #include "dm-bio-list.h" | |
10 | #include "dm-uevent.h" | |
11 | ||
12 | #include <linux/init.h> | |
13 | #include <linux/module.h> | |
14 | #include <linux/mutex.h> | |
15 | #include <linux/moduleparam.h> | |
16 | #include <linux/blkpg.h> | |
17 | #include <linux/bio.h> | |
18 | #include <linux/buffer_head.h> | |
19 | #include <linux/mempool.h> | |
20 | #include <linux/slab.h> | |
21 | #include <linux/idr.h> | |
22 | #include <linux/hdreg.h> | |
23 | #include <linux/blktrace_api.h> | |
24 | #include <trace/block.h> | |
25 | ||
26 | #define DM_MSG_PREFIX "core" | |
27 | ||
28 | static const char *_name = DM_NAME; | |
29 | ||
30 | static unsigned int major = 0; | |
31 | static unsigned int _major = 0; | |
32 | ||
33 | static DEFINE_SPINLOCK(_minor_lock); | |
34 | /* | |
35 | * For bio-based dm. | |
36 | * One of these is allocated per bio. | |
37 | */ | |
38 | struct dm_io { | |
39 | struct mapped_device *md; | |
40 | int error; | |
41 | atomic_t io_count; | |
42 | struct bio *bio; | |
43 | unsigned long start_time; | |
44 | }; | |
45 | ||
46 | /* | |
47 | * For bio-based dm. | |
48 | * One of these is allocated per target within a bio. Hopefully | |
49 | * this will be simplified out one day. | |
50 | */ | |
51 | struct dm_target_io { | |
52 | struct dm_io *io; | |
53 | struct dm_target *ti; | |
54 | union map_info info; | |
55 | }; | |
56 | ||
57 | DEFINE_TRACE(block_bio_complete); | |
58 | ||
59 | /* | |
60 | * For request-based dm. | |
61 | * One of these is allocated per request. | |
62 | */ | |
63 | struct dm_rq_target_io { | |
64 | struct mapped_device *md; | |
65 | struct dm_target *ti; | |
66 | struct request *orig, clone; | |
67 | int error; | |
68 | union map_info info; | |
69 | }; | |
70 | ||
71 | /* | |
72 | * For request-based dm. | |
73 | * One of these is allocated per bio. | |
74 | */ | |
75 | struct dm_rq_clone_bio_info { | |
76 | struct bio *orig; | |
77 | struct request *rq; | |
78 | }; | |
79 | ||
80 | union map_info *dm_get_mapinfo(struct bio *bio) | |
81 | { | |
82 | if (bio && bio->bi_private) | |
83 | return &((struct dm_target_io *)bio->bi_private)->info; | |
84 | return NULL; | |
85 | } | |
86 | ||
87 | #define MINOR_ALLOCED ((void *)-1) | |
88 | ||
89 | /* | |
90 | * Bits for the md->flags field. | |
91 | */ | |
92 | #define DMF_BLOCK_IO 0 | |
93 | #define DMF_SUSPENDED 1 | |
94 | #define DMF_FROZEN 2 | |
95 | #define DMF_FREEING 3 | |
96 | #define DMF_DELETING 4 | |
97 | #define DMF_NOFLUSH_SUSPENDING 5 | |
98 | ||
99 | /* | |
100 | * Work processed by per-device workqueue. | |
101 | */ | |
102 | struct dm_wq_req { | |
103 | enum { | |
104 | DM_WQ_FLUSH_DEFERRED, | |
105 | } type; | |
106 | struct work_struct work; | |
107 | struct mapped_device *md; | |
108 | void *context; | |
109 | }; | |
110 | ||
111 | struct mapped_device { | |
112 | struct rw_semaphore io_lock; | |
113 | struct mutex suspend_lock; | |
114 | spinlock_t pushback_lock; | |
115 | rwlock_t map_lock; | |
116 | atomic_t holders; | |
117 | atomic_t open_count; | |
118 | ||
119 | unsigned long flags; | |
120 | ||
121 | struct request_queue *queue; | |
122 | struct gendisk *disk; | |
123 | char name[16]; | |
124 | ||
125 | void *interface_ptr; | |
126 | ||
127 | /* | |
128 | * A list of ios that arrived while we were suspended. | |
129 | */ | |
130 | atomic_t pending; | |
131 | wait_queue_head_t wait; | |
132 | struct bio_list deferred; | |
133 | struct bio_list pushback; | |
134 | ||
135 | /* | |
136 | * Processing queue (flush/barriers) | |
137 | */ | |
138 | struct workqueue_struct *wq; | |
139 | ||
140 | /* | |
141 | * The current mapping. | |
142 | */ | |
143 | struct dm_table *map; | |
144 | ||
145 | /* | |
146 | * io objects are allocated from here. | |
147 | */ | |
148 | mempool_t *io_pool; | |
149 | mempool_t *tio_pool; | |
150 | ||
151 | struct bio_set *bs; | |
152 | ||
153 | /* | |
154 | * Event handling. | |
155 | */ | |
156 | atomic_t event_nr; | |
157 | wait_queue_head_t eventq; | |
158 | atomic_t uevent_seq; | |
159 | struct list_head uevent_list; | |
160 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
161 | ||
162 | /* | |
163 | * freeze/thaw support require holding onto a super block | |
164 | */ | |
165 | struct super_block *frozen_sb; | |
166 | struct block_device *suspended_bdev; | |
167 | ||
168 | /* forced geometry settings */ | |
169 | struct hd_geometry geometry; | |
170 | }; | |
171 | ||
172 | #define MIN_IOS 256 | |
173 | static struct kmem_cache *_io_cache; | |
174 | static struct kmem_cache *_tio_cache; | |
175 | static struct kmem_cache *_rq_tio_cache; | |
176 | static struct kmem_cache *_rq_bio_info_cache; | |
177 | ||
178 | static int __init local_init(void) | |
179 | { | |
180 | int r = -ENOMEM; | |
181 | ||
182 | /* allocate a slab for the dm_ios */ | |
183 | _io_cache = KMEM_CACHE(dm_io, 0); | |
184 | if (!_io_cache) | |
185 | return r; | |
186 | ||
187 | /* allocate a slab for the target ios */ | |
188 | _tio_cache = KMEM_CACHE(dm_target_io, 0); | |
189 | if (!_tio_cache) | |
190 | goto out_free_io_cache; | |
191 | ||
192 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); | |
193 | if (!_rq_tio_cache) | |
194 | goto out_free_tio_cache; | |
195 | ||
196 | _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0); | |
197 | if (!_rq_bio_info_cache) | |
198 | goto out_free_rq_tio_cache; | |
199 | ||
200 | r = dm_uevent_init(); | |
201 | if (r) | |
202 | goto out_free_rq_bio_info_cache; | |
203 | ||
204 | _major = major; | |
205 | r = register_blkdev(_major, _name); | |
206 | if (r < 0) | |
207 | goto out_uevent_exit; | |
208 | ||
209 | if (!_major) | |
210 | _major = r; | |
211 | ||
212 | return 0; | |
213 | ||
214 | out_uevent_exit: | |
215 | dm_uevent_exit(); | |
216 | out_free_rq_bio_info_cache: | |
217 | kmem_cache_destroy(_rq_bio_info_cache); | |
218 | out_free_rq_tio_cache: | |
219 | kmem_cache_destroy(_rq_tio_cache); | |
220 | out_free_tio_cache: | |
221 | kmem_cache_destroy(_tio_cache); | |
222 | out_free_io_cache: | |
223 | kmem_cache_destroy(_io_cache); | |
224 | ||
225 | return r; | |
226 | } | |
227 | ||
228 | static void local_exit(void) | |
229 | { | |
230 | kmem_cache_destroy(_rq_bio_info_cache); | |
231 | kmem_cache_destroy(_rq_tio_cache); | |
232 | kmem_cache_destroy(_tio_cache); | |
233 | kmem_cache_destroy(_io_cache); | |
234 | unregister_blkdev(_major, _name); | |
235 | dm_uevent_exit(); | |
236 | ||
237 | _major = 0; | |
238 | ||
239 | DMINFO("cleaned up"); | |
240 | } | |
241 | ||
242 | static int (*_inits[])(void) __initdata = { | |
243 | local_init, | |
244 | dm_target_init, | |
245 | dm_linear_init, | |
246 | dm_stripe_init, | |
247 | dm_kcopyd_init, | |
248 | dm_interface_init, | |
249 | }; | |
250 | ||
251 | static void (*_exits[])(void) = { | |
252 | local_exit, | |
253 | dm_target_exit, | |
254 | dm_linear_exit, | |
255 | dm_stripe_exit, | |
256 | dm_kcopyd_exit, | |
257 | dm_interface_exit, | |
258 | }; | |
259 | ||
260 | static int __init dm_init(void) | |
261 | { | |
262 | const int count = ARRAY_SIZE(_inits); | |
263 | ||
264 | int r, i; | |
265 | ||
266 | for (i = 0; i < count; i++) { | |
267 | r = _inits[i](); | |
268 | if (r) | |
269 | goto bad; | |
270 | } | |
271 | ||
272 | return 0; | |
273 | ||
274 | bad: | |
275 | while (i--) | |
276 | _exits[i](); | |
277 | ||
278 | return r; | |
279 | } | |
280 | ||
281 | static void __exit dm_exit(void) | |
282 | { | |
283 | int i = ARRAY_SIZE(_exits); | |
284 | ||
285 | while (i--) | |
286 | _exits[i](); | |
287 | } | |
288 | ||
289 | /* | |
290 | * Block device functions | |
291 | */ | |
292 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) | |
293 | { | |
294 | struct mapped_device *md; | |
295 | ||
296 | spin_lock(&_minor_lock); | |
297 | ||
298 | md = bdev->bd_disk->private_data; | |
299 | if (!md) | |
300 | goto out; | |
301 | ||
302 | if (test_bit(DMF_FREEING, &md->flags) || | |
303 | test_bit(DMF_DELETING, &md->flags)) { | |
304 | md = NULL; | |
305 | goto out; | |
306 | } | |
307 | ||
308 | dm_get(md); | |
309 | atomic_inc(&md->open_count); | |
310 | ||
311 | out: | |
312 | spin_unlock(&_minor_lock); | |
313 | ||
314 | return md ? 0 : -ENXIO; | |
315 | } | |
316 | ||
317 | static int dm_blk_close(struct gendisk *disk, fmode_t mode) | |
318 | { | |
319 | struct mapped_device *md = disk->private_data; | |
320 | atomic_dec(&md->open_count); | |
321 | dm_put(md); | |
322 | return 0; | |
323 | } | |
324 | ||
325 | int dm_open_count(struct mapped_device *md) | |
326 | { | |
327 | return atomic_read(&md->open_count); | |
328 | } | |
329 | ||
330 | /* | |
331 | * Guarantees nothing is using the device before it's deleted. | |
332 | */ | |
333 | int dm_lock_for_deletion(struct mapped_device *md) | |
334 | { | |
335 | int r = 0; | |
336 | ||
337 | spin_lock(&_minor_lock); | |
338 | ||
339 | if (dm_open_count(md)) | |
340 | r = -EBUSY; | |
341 | else | |
342 | set_bit(DMF_DELETING, &md->flags); | |
343 | ||
344 | spin_unlock(&_minor_lock); | |
345 | ||
346 | return r; | |
347 | } | |
348 | ||
349 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) | |
350 | { | |
351 | struct mapped_device *md = bdev->bd_disk->private_data; | |
352 | ||
353 | return dm_get_geometry(md, geo); | |
354 | } | |
355 | ||
356 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, | |
357 | unsigned int cmd, unsigned long arg) | |
358 | { | |
359 | struct mapped_device *md = bdev->bd_disk->private_data; | |
360 | struct dm_table *map = dm_get_table(md); | |
361 | struct dm_target *tgt; | |
362 | int r = -ENOTTY; | |
363 | ||
364 | if (!map || !dm_table_get_size(map)) | |
365 | goto out; | |
366 | ||
367 | /* We only support devices that have a single target */ | |
368 | if (dm_table_get_num_targets(map) != 1) | |
369 | goto out; | |
370 | ||
371 | tgt = dm_table_get_target(map, 0); | |
372 | ||
373 | if (dm_suspended(md)) { | |
374 | r = -EAGAIN; | |
375 | goto out; | |
376 | } | |
377 | ||
378 | if (tgt->type->ioctl) | |
379 | r = tgt->type->ioctl(tgt, cmd, arg); | |
380 | ||
381 | out: | |
382 | dm_table_put(map); | |
383 | ||
384 | return r; | |
385 | } | |
386 | ||
387 | static struct dm_io *alloc_io(struct mapped_device *md) | |
388 | { | |
389 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
390 | } | |
391 | ||
392 | static void free_io(struct mapped_device *md, struct dm_io *io) | |
393 | { | |
394 | mempool_free(io, md->io_pool); | |
395 | } | |
396 | ||
397 | static struct dm_target_io *alloc_tio(struct mapped_device *md) | |
398 | { | |
399 | return mempool_alloc(md->tio_pool, GFP_NOIO); | |
400 | } | |
401 | ||
402 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) | |
403 | { | |
404 | mempool_free(tio, md->tio_pool); | |
405 | } | |
406 | ||
407 | static void start_io_acct(struct dm_io *io) | |
408 | { | |
409 | struct mapped_device *md = io->md; | |
410 | int cpu; | |
411 | ||
412 | io->start_time = jiffies; | |
413 | ||
414 | cpu = part_stat_lock(); | |
415 | part_round_stats(cpu, &dm_disk(md)->part0); | |
416 | part_stat_unlock(); | |
417 | dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending); | |
418 | } | |
419 | ||
420 | static void end_io_acct(struct dm_io *io) | |
421 | { | |
422 | struct mapped_device *md = io->md; | |
423 | struct bio *bio = io->bio; | |
424 | unsigned long duration = jiffies - io->start_time; | |
425 | int pending, cpu; | |
426 | int rw = bio_data_dir(bio); | |
427 | ||
428 | cpu = part_stat_lock(); | |
429 | part_round_stats(cpu, &dm_disk(md)->part0); | |
430 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
431 | part_stat_unlock(); | |
432 | ||
433 | dm_disk(md)->part0.in_flight = pending = | |
434 | atomic_dec_return(&md->pending); | |
435 | ||
436 | /* nudge anyone waiting on suspend queue */ | |
437 | if (!pending) | |
438 | wake_up(&md->wait); | |
439 | } | |
440 | ||
441 | /* | |
442 | * Add the bio to the list of deferred io. | |
443 | */ | |
444 | static int queue_io(struct mapped_device *md, struct bio *bio) | |
445 | { | |
446 | down_write(&md->io_lock); | |
447 | ||
448 | if (!test_bit(DMF_BLOCK_IO, &md->flags)) { | |
449 | up_write(&md->io_lock); | |
450 | return 1; | |
451 | } | |
452 | ||
453 | bio_list_add(&md->deferred, bio); | |
454 | ||
455 | up_write(&md->io_lock); | |
456 | return 0; /* deferred successfully */ | |
457 | } | |
458 | ||
459 | /* | |
460 | * Everyone (including functions in this file), should use this | |
461 | * function to access the md->map field, and make sure they call | |
462 | * dm_table_put() when finished. | |
463 | */ | |
464 | struct dm_table *dm_get_table(struct mapped_device *md) | |
465 | { | |
466 | struct dm_table *t; | |
467 | ||
468 | read_lock(&md->map_lock); | |
469 | t = md->map; | |
470 | if (t) | |
471 | dm_table_get(t); | |
472 | read_unlock(&md->map_lock); | |
473 | ||
474 | return t; | |
475 | } | |
476 | ||
477 | /* | |
478 | * Get the geometry associated with a dm device | |
479 | */ | |
480 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
481 | { | |
482 | *geo = md->geometry; | |
483 | ||
484 | return 0; | |
485 | } | |
486 | ||
487 | /* | |
488 | * Set the geometry of a device. | |
489 | */ | |
490 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
491 | { | |
492 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
493 | ||
494 | if (geo->start > sz) { | |
495 | DMWARN("Start sector is beyond the geometry limits."); | |
496 | return -EINVAL; | |
497 | } | |
498 | ||
499 | md->geometry = *geo; | |
500 | ||
501 | return 0; | |
502 | } | |
503 | ||
504 | /*----------------------------------------------------------------- | |
505 | * CRUD START: | |
506 | * A more elegant soln is in the works that uses the queue | |
507 | * merge fn, unfortunately there are a couple of changes to | |
508 | * the block layer that I want to make for this. So in the | |
509 | * interests of getting something for people to use I give | |
510 | * you this clearly demarcated crap. | |
511 | *---------------------------------------------------------------*/ | |
512 | ||
513 | static int __noflush_suspending(struct mapped_device *md) | |
514 | { | |
515 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
516 | } | |
517 | ||
518 | /* | |
519 | * Decrements the number of outstanding ios that a bio has been | |
520 | * cloned into, completing the original io if necc. | |
521 | */ | |
522 | static void dec_pending(struct dm_io *io, int error) | |
523 | { | |
524 | unsigned long flags; | |
525 | ||
526 | /* Push-back supersedes any I/O errors */ | |
527 | if (error && !(io->error > 0 && __noflush_suspending(io->md))) | |
528 | io->error = error; | |
529 | ||
530 | if (atomic_dec_and_test(&io->io_count)) { | |
531 | if (io->error == DM_ENDIO_REQUEUE) { | |
532 | /* | |
533 | * Target requested pushing back the I/O. | |
534 | * This must be handled before the sleeper on | |
535 | * suspend queue merges the pushback list. | |
536 | */ | |
537 | spin_lock_irqsave(&io->md->pushback_lock, flags); | |
538 | if (__noflush_suspending(io->md)) | |
539 | bio_list_add(&io->md->pushback, io->bio); | |
540 | else | |
541 | /* noflush suspend was interrupted. */ | |
542 | io->error = -EIO; | |
543 | spin_unlock_irqrestore(&io->md->pushback_lock, flags); | |
544 | } | |
545 | ||
546 | end_io_acct(io); | |
547 | ||
548 | if (io->error != DM_ENDIO_REQUEUE) { | |
549 | trace_block_bio_complete(io->md->queue, io->bio); | |
550 | ||
551 | bio_endio(io->bio, io->error); | |
552 | } | |
553 | ||
554 | free_io(io->md, io); | |
555 | } | |
556 | } | |
557 | ||
558 | static void clone_endio(struct bio *bio, int error) | |
559 | { | |
560 | int r = 0; | |
561 | struct dm_target_io *tio = bio->bi_private; | |
562 | struct mapped_device *md = tio->io->md; | |
563 | dm_endio_fn endio = tio->ti->type->end_io; | |
564 | ||
565 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) | |
566 | error = -EIO; | |
567 | ||
568 | if (endio) { | |
569 | r = endio(tio->ti, bio, error, &tio->info); | |
570 | if (r < 0 || r == DM_ENDIO_REQUEUE) | |
571 | /* | |
572 | * error and requeue request are handled | |
573 | * in dec_pending(). | |
574 | */ | |
575 | error = r; | |
576 | else if (r == DM_ENDIO_INCOMPLETE) | |
577 | /* The target will handle the io */ | |
578 | return; | |
579 | else if (r) { | |
580 | DMWARN("unimplemented target endio return value: %d", r); | |
581 | BUG(); | |
582 | } | |
583 | } | |
584 | ||
585 | dec_pending(tio->io, error); | |
586 | ||
587 | /* | |
588 | * Store md for cleanup instead of tio which is about to get freed. | |
589 | */ | |
590 | bio->bi_private = md->bs; | |
591 | ||
592 | bio_put(bio); | |
593 | free_tio(md, tio); | |
594 | } | |
595 | ||
596 | static sector_t max_io_len(struct mapped_device *md, | |
597 | sector_t sector, struct dm_target *ti) | |
598 | { | |
599 | sector_t offset = sector - ti->begin; | |
600 | sector_t len = ti->len - offset; | |
601 | ||
602 | /* | |
603 | * Does the target need to split even further ? | |
604 | */ | |
605 | if (ti->split_io) { | |
606 | sector_t boundary; | |
607 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) | |
608 | - offset; | |
609 | if (len > boundary) | |
610 | len = boundary; | |
611 | } | |
612 | ||
613 | return len; | |
614 | } | |
615 | ||
616 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
617 | struct dm_target_io *tio) | |
618 | { | |
619 | int r; | |
620 | sector_t sector; | |
621 | struct mapped_device *md; | |
622 | ||
623 | /* | |
624 | * Sanity checks. | |
625 | */ | |
626 | BUG_ON(!clone->bi_size); | |
627 | ||
628 | clone->bi_end_io = clone_endio; | |
629 | clone->bi_private = tio; | |
630 | ||
631 | /* | |
632 | * Map the clone. If r == 0 we don't need to do | |
633 | * anything, the target has assumed ownership of | |
634 | * this io. | |
635 | */ | |
636 | atomic_inc(&tio->io->io_count); | |
637 | sector = clone->bi_sector; | |
638 | r = ti->type->map(ti, clone, &tio->info); | |
639 | if (r == DM_MAPIO_REMAPPED) { | |
640 | /* the bio has been remapped so dispatch it */ | |
641 | ||
642 | trace_block_remap(bdev_get_queue(clone->bi_bdev), clone, | |
643 | tio->io->bio->bi_bdev->bd_dev, | |
644 | clone->bi_sector, sector); | |
645 | ||
646 | generic_make_request(clone); | |
647 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { | |
648 | /* error the io and bail out, or requeue it if needed */ | |
649 | md = tio->io->md; | |
650 | dec_pending(tio->io, r); | |
651 | /* | |
652 | * Store bio_set for cleanup. | |
653 | */ | |
654 | clone->bi_private = md->bs; | |
655 | bio_put(clone); | |
656 | free_tio(md, tio); | |
657 | } else if (r) { | |
658 | DMWARN("unimplemented target map return value: %d", r); | |
659 | BUG(); | |
660 | } | |
661 | } | |
662 | ||
663 | struct clone_info { | |
664 | struct mapped_device *md; | |
665 | struct dm_table *map; | |
666 | struct bio *bio; | |
667 | struct dm_io *io; | |
668 | sector_t sector; | |
669 | sector_t sector_count; | |
670 | unsigned short idx; | |
671 | }; | |
672 | ||
673 | static void dm_bio_destructor(struct bio *bio) | |
674 | { | |
675 | struct bio_set *bs = bio->bi_private; | |
676 | ||
677 | bio_free(bio, bs); | |
678 | } | |
679 | ||
680 | /* | |
681 | * Creates a little bio that is just does part of a bvec. | |
682 | */ | |
683 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
684 | unsigned short idx, unsigned int offset, | |
685 | unsigned int len, struct bio_set *bs) | |
686 | { | |
687 | struct bio *clone; | |
688 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
689 | ||
690 | clone = bio_alloc_bioset(GFP_NOIO, 1, bs); | |
691 | clone->bi_destructor = dm_bio_destructor; | |
692 | *clone->bi_io_vec = *bv; | |
693 | ||
694 | clone->bi_sector = sector; | |
695 | clone->bi_bdev = bio->bi_bdev; | |
696 | clone->bi_rw = bio->bi_rw; | |
697 | clone->bi_vcnt = 1; | |
698 | clone->bi_size = to_bytes(len); | |
699 | clone->bi_io_vec->bv_offset = offset; | |
700 | clone->bi_io_vec->bv_len = clone->bi_size; | |
701 | clone->bi_flags |= 1 << BIO_CLONED; | |
702 | ||
703 | return clone; | |
704 | } | |
705 | ||
706 | /* | |
707 | * Creates a bio that consists of range of complete bvecs. | |
708 | */ | |
709 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
710 | unsigned short idx, unsigned short bv_count, | |
711 | unsigned int len, struct bio_set *bs) | |
712 | { | |
713 | struct bio *clone; | |
714 | ||
715 | clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); | |
716 | __bio_clone(clone, bio); | |
717 | clone->bi_destructor = dm_bio_destructor; | |
718 | clone->bi_sector = sector; | |
719 | clone->bi_idx = idx; | |
720 | clone->bi_vcnt = idx + bv_count; | |
721 | clone->bi_size = to_bytes(len); | |
722 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
723 | ||
724 | return clone; | |
725 | } | |
726 | ||
727 | static int __clone_and_map(struct clone_info *ci) | |
728 | { | |
729 | struct bio *clone, *bio = ci->bio; | |
730 | struct dm_target *ti; | |
731 | sector_t len = 0, max; | |
732 | struct dm_target_io *tio; | |
733 | ||
734 | ti = dm_table_find_target(ci->map, ci->sector); | |
735 | if (!dm_target_is_valid(ti)) | |
736 | return -EIO; | |
737 | ||
738 | max = max_io_len(ci->md, ci->sector, ti); | |
739 | ||
740 | /* | |
741 | * Allocate a target io object. | |
742 | */ | |
743 | tio = alloc_tio(ci->md); | |
744 | tio->io = ci->io; | |
745 | tio->ti = ti; | |
746 | memset(&tio->info, 0, sizeof(tio->info)); | |
747 | ||
748 | if (ci->sector_count <= max) { | |
749 | /* | |
750 | * Optimise for the simple case where we can do all of | |
751 | * the remaining io with a single clone. | |
752 | */ | |
753 | clone = clone_bio(bio, ci->sector, ci->idx, | |
754 | bio->bi_vcnt - ci->idx, ci->sector_count, | |
755 | ci->md->bs); | |
756 | __map_bio(ti, clone, tio); | |
757 | ci->sector_count = 0; | |
758 | ||
759 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
760 | /* | |
761 | * There are some bvecs that don't span targets. | |
762 | * Do as many of these as possible. | |
763 | */ | |
764 | int i; | |
765 | sector_t remaining = max; | |
766 | sector_t bv_len; | |
767 | ||
768 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
769 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
770 | ||
771 | if (bv_len > remaining) | |
772 | break; | |
773 | ||
774 | remaining -= bv_len; | |
775 | len += bv_len; | |
776 | } | |
777 | ||
778 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, | |
779 | ci->md->bs); | |
780 | __map_bio(ti, clone, tio); | |
781 | ||
782 | ci->sector += len; | |
783 | ci->sector_count -= len; | |
784 | ci->idx = i; | |
785 | ||
786 | } else { | |
787 | /* | |
788 | * Handle a bvec that must be split between two or more targets. | |
789 | */ | |
790 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
791 | sector_t remaining = to_sector(bv->bv_len); | |
792 | unsigned int offset = 0; | |
793 | ||
794 | do { | |
795 | if (offset) { | |
796 | ti = dm_table_find_target(ci->map, ci->sector); | |
797 | if (!dm_target_is_valid(ti)) | |
798 | return -EIO; | |
799 | ||
800 | max = max_io_len(ci->md, ci->sector, ti); | |
801 | ||
802 | tio = alloc_tio(ci->md); | |
803 | tio->io = ci->io; | |
804 | tio->ti = ti; | |
805 | memset(&tio->info, 0, sizeof(tio->info)); | |
806 | } | |
807 | ||
808 | len = min(remaining, max); | |
809 | ||
810 | clone = split_bvec(bio, ci->sector, ci->idx, | |
811 | bv->bv_offset + offset, len, | |
812 | ci->md->bs); | |
813 | ||
814 | __map_bio(ti, clone, tio); | |
815 | ||
816 | ci->sector += len; | |
817 | ci->sector_count -= len; | |
818 | offset += to_bytes(len); | |
819 | } while (remaining -= len); | |
820 | ||
821 | ci->idx++; | |
822 | } | |
823 | ||
824 | return 0; | |
825 | } | |
826 | ||
827 | /* | |
828 | * Split the bio into several clones. | |
829 | */ | |
830 | static int __split_bio(struct mapped_device *md, struct bio *bio) | |
831 | { | |
832 | struct clone_info ci; | |
833 | int error = 0; | |
834 | ||
835 | ci.map = dm_get_table(md); | |
836 | if (unlikely(!ci.map)) | |
837 | return -EIO; | |
838 | if (unlikely(bio_barrier(bio) && !dm_table_barrier_ok(ci.map))) { | |
839 | dm_table_put(ci.map); | |
840 | bio_endio(bio, -EOPNOTSUPP); | |
841 | return 0; | |
842 | } | |
843 | ci.md = md; | |
844 | ci.bio = bio; | |
845 | ci.io = alloc_io(md); | |
846 | ci.io->error = 0; | |
847 | atomic_set(&ci.io->io_count, 1); | |
848 | ci.io->bio = bio; | |
849 | ci.io->md = md; | |
850 | ci.sector = bio->bi_sector; | |
851 | ci.sector_count = bio_sectors(bio); | |
852 | ci.idx = bio->bi_idx; | |
853 | ||
854 | start_io_acct(ci.io); | |
855 | while (ci.sector_count && !error) | |
856 | error = __clone_and_map(&ci); | |
857 | ||
858 | /* drop the extra reference count */ | |
859 | dec_pending(ci.io, error); | |
860 | dm_table_put(ci.map); | |
861 | ||
862 | return 0; | |
863 | } | |
864 | /*----------------------------------------------------------------- | |
865 | * CRUD END | |
866 | *---------------------------------------------------------------*/ | |
867 | ||
868 | static int dm_merge_bvec(struct request_queue *q, | |
869 | struct bvec_merge_data *bvm, | |
870 | struct bio_vec *biovec) | |
871 | { | |
872 | struct mapped_device *md = q->queuedata; | |
873 | struct dm_table *map = dm_get_table(md); | |
874 | struct dm_target *ti; | |
875 | sector_t max_sectors; | |
876 | int max_size = 0; | |
877 | ||
878 | if (unlikely(!map)) | |
879 | goto out; | |
880 | ||
881 | ti = dm_table_find_target(map, bvm->bi_sector); | |
882 | if (!dm_target_is_valid(ti)) | |
883 | goto out_table; | |
884 | ||
885 | /* | |
886 | * Find maximum amount of I/O that won't need splitting | |
887 | */ | |
888 | max_sectors = min(max_io_len(md, bvm->bi_sector, ti), | |
889 | (sector_t) BIO_MAX_SECTORS); | |
890 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
891 | if (max_size < 0) | |
892 | max_size = 0; | |
893 | ||
894 | /* | |
895 | * merge_bvec_fn() returns number of bytes | |
896 | * it can accept at this offset | |
897 | * max is precomputed maximal io size | |
898 | */ | |
899 | if (max_size && ti->type->merge) | |
900 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
901 | ||
902 | out_table: | |
903 | dm_table_put(map); | |
904 | ||
905 | out: | |
906 | /* | |
907 | * Always allow an entire first page | |
908 | */ | |
909 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
910 | max_size = biovec->bv_len; | |
911 | ||
912 | return max_size; | |
913 | } | |
914 | ||
915 | /* | |
916 | * The request function that just remaps the bio built up by | |
917 | * dm_merge_bvec. | |
918 | */ | |
919 | static int dm_request(struct request_queue *q, struct bio *bio) | |
920 | { | |
921 | int r = -EIO; | |
922 | int rw = bio_data_dir(bio); | |
923 | struct mapped_device *md = q->queuedata; | |
924 | int cpu; | |
925 | ||
926 | down_read(&md->io_lock); | |
927 | ||
928 | cpu = part_stat_lock(); | |
929 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
930 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
931 | part_stat_unlock(); | |
932 | ||
933 | /* | |
934 | * If we're suspended we have to queue | |
935 | * this io for later. | |
936 | */ | |
937 | while (test_bit(DMF_BLOCK_IO, &md->flags)) { | |
938 | up_read(&md->io_lock); | |
939 | ||
940 | if (bio_rw(bio) != READA) | |
941 | r = queue_io(md, bio); | |
942 | ||
943 | if (r <= 0) | |
944 | goto out_req; | |
945 | ||
946 | /* | |
947 | * We're in a while loop, because someone could suspend | |
948 | * before we get to the following read lock. | |
949 | */ | |
950 | down_read(&md->io_lock); | |
951 | } | |
952 | ||
953 | r = __split_bio(md, bio); | |
954 | up_read(&md->io_lock); | |
955 | ||
956 | out_req: | |
957 | if (r < 0) | |
958 | bio_io_error(bio); | |
959 | ||
960 | return 0; | |
961 | } | |
962 | ||
963 | static void dm_unplug_all(struct request_queue *q) | |
964 | { | |
965 | struct mapped_device *md = q->queuedata; | |
966 | struct dm_table *map = dm_get_table(md); | |
967 | ||
968 | if (map) { | |
969 | dm_table_unplug_all(map); | |
970 | dm_table_put(map); | |
971 | } | |
972 | } | |
973 | ||
974 | static int dm_any_congested(void *congested_data, int bdi_bits) | |
975 | { | |
976 | int r = bdi_bits; | |
977 | struct mapped_device *md = congested_data; | |
978 | struct dm_table *map; | |
979 | ||
980 | if (!test_bit(DMF_BLOCK_IO, &md->flags)) { | |
981 | map = dm_get_table(md); | |
982 | if (map) { | |
983 | r = dm_table_any_congested(map, bdi_bits); | |
984 | dm_table_put(map); | |
985 | } | |
986 | } | |
987 | ||
988 | return r; | |
989 | } | |
990 | ||
991 | /*----------------------------------------------------------------- | |
992 | * An IDR is used to keep track of allocated minor numbers. | |
993 | *---------------------------------------------------------------*/ | |
994 | static DEFINE_IDR(_minor_idr); | |
995 | ||
996 | static void free_minor(int minor) | |
997 | { | |
998 | spin_lock(&_minor_lock); | |
999 | idr_remove(&_minor_idr, minor); | |
1000 | spin_unlock(&_minor_lock); | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * See if the device with a specific minor # is free. | |
1005 | */ | |
1006 | static int specific_minor(int minor) | |
1007 | { | |
1008 | int r, m; | |
1009 | ||
1010 | if (minor >= (1 << MINORBITS)) | |
1011 | return -EINVAL; | |
1012 | ||
1013 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); | |
1014 | if (!r) | |
1015 | return -ENOMEM; | |
1016 | ||
1017 | spin_lock(&_minor_lock); | |
1018 | ||
1019 | if (idr_find(&_minor_idr, minor)) { | |
1020 | r = -EBUSY; | |
1021 | goto out; | |
1022 | } | |
1023 | ||
1024 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); | |
1025 | if (r) | |
1026 | goto out; | |
1027 | ||
1028 | if (m != minor) { | |
1029 | idr_remove(&_minor_idr, m); | |
1030 | r = -EBUSY; | |
1031 | goto out; | |
1032 | } | |
1033 | ||
1034 | out: | |
1035 | spin_unlock(&_minor_lock); | |
1036 | return r; | |
1037 | } | |
1038 | ||
1039 | static int next_free_minor(int *minor) | |
1040 | { | |
1041 | int r, m; | |
1042 | ||
1043 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); | |
1044 | if (!r) | |
1045 | return -ENOMEM; | |
1046 | ||
1047 | spin_lock(&_minor_lock); | |
1048 | ||
1049 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); | |
1050 | if (r) | |
1051 | goto out; | |
1052 | ||
1053 | if (m >= (1 << MINORBITS)) { | |
1054 | idr_remove(&_minor_idr, m); | |
1055 | r = -ENOSPC; | |
1056 | goto out; | |
1057 | } | |
1058 | ||
1059 | *minor = m; | |
1060 | ||
1061 | out: | |
1062 | spin_unlock(&_minor_lock); | |
1063 | return r; | |
1064 | } | |
1065 | ||
1066 | static struct block_device_operations dm_blk_dops; | |
1067 | ||
1068 | /* | |
1069 | * Allocate and initialise a blank device with a given minor. | |
1070 | */ | |
1071 | static struct mapped_device *alloc_dev(int minor) | |
1072 | { | |
1073 | int r; | |
1074 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); | |
1075 | void *old_md; | |
1076 | ||
1077 | if (!md) { | |
1078 | DMWARN("unable to allocate device, out of memory."); | |
1079 | return NULL; | |
1080 | } | |
1081 | ||
1082 | if (!try_module_get(THIS_MODULE)) | |
1083 | goto bad_module_get; | |
1084 | ||
1085 | /* get a minor number for the dev */ | |
1086 | if (minor == DM_ANY_MINOR) | |
1087 | r = next_free_minor(&minor); | |
1088 | else | |
1089 | r = specific_minor(minor); | |
1090 | if (r < 0) | |
1091 | goto bad_minor; | |
1092 | ||
1093 | init_rwsem(&md->io_lock); | |
1094 | mutex_init(&md->suspend_lock); | |
1095 | spin_lock_init(&md->pushback_lock); | |
1096 | rwlock_init(&md->map_lock); | |
1097 | atomic_set(&md->holders, 1); | |
1098 | atomic_set(&md->open_count, 0); | |
1099 | atomic_set(&md->event_nr, 0); | |
1100 | atomic_set(&md->uevent_seq, 0); | |
1101 | INIT_LIST_HEAD(&md->uevent_list); | |
1102 | spin_lock_init(&md->uevent_lock); | |
1103 | ||
1104 | md->queue = blk_alloc_queue(GFP_KERNEL); | |
1105 | if (!md->queue) | |
1106 | goto bad_queue; | |
1107 | ||
1108 | md->queue->queuedata = md; | |
1109 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1110 | md->queue->backing_dev_info.congested_data = md; | |
1111 | blk_queue_make_request(md->queue, dm_request); | |
1112 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
1113 | md->queue->unplug_fn = dm_unplug_all; | |
1114 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); | |
1115 | ||
1116 | md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache); | |
1117 | if (!md->io_pool) | |
1118 | goto bad_io_pool; | |
1119 | ||
1120 | md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache); | |
1121 | if (!md->tio_pool) | |
1122 | goto bad_tio_pool; | |
1123 | ||
1124 | md->bs = bioset_create(16, 0); | |
1125 | if (!md->bs) | |
1126 | goto bad_no_bioset; | |
1127 | ||
1128 | md->disk = alloc_disk(1); | |
1129 | if (!md->disk) | |
1130 | goto bad_disk; | |
1131 | ||
1132 | atomic_set(&md->pending, 0); | |
1133 | init_waitqueue_head(&md->wait); | |
1134 | init_waitqueue_head(&md->eventq); | |
1135 | ||
1136 | md->disk->major = _major; | |
1137 | md->disk->first_minor = minor; | |
1138 | md->disk->fops = &dm_blk_dops; | |
1139 | md->disk->queue = md->queue; | |
1140 | md->disk->private_data = md; | |
1141 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1142 | add_disk(md->disk); | |
1143 | format_dev_t(md->name, MKDEV(_major, minor)); | |
1144 | ||
1145 | md->wq = create_singlethread_workqueue("kdmflush"); | |
1146 | if (!md->wq) | |
1147 | goto bad_thread; | |
1148 | ||
1149 | /* Populate the mapping, nobody knows we exist yet */ | |
1150 | spin_lock(&_minor_lock); | |
1151 | old_md = idr_replace(&_minor_idr, md, minor); | |
1152 | spin_unlock(&_minor_lock); | |
1153 | ||
1154 | BUG_ON(old_md != MINOR_ALLOCED); | |
1155 | ||
1156 | return md; | |
1157 | ||
1158 | bad_thread: | |
1159 | put_disk(md->disk); | |
1160 | bad_disk: | |
1161 | bioset_free(md->bs); | |
1162 | bad_no_bioset: | |
1163 | mempool_destroy(md->tio_pool); | |
1164 | bad_tio_pool: | |
1165 | mempool_destroy(md->io_pool); | |
1166 | bad_io_pool: | |
1167 | blk_cleanup_queue(md->queue); | |
1168 | bad_queue: | |
1169 | free_minor(minor); | |
1170 | bad_minor: | |
1171 | module_put(THIS_MODULE); | |
1172 | bad_module_get: | |
1173 | kfree(md); | |
1174 | return NULL; | |
1175 | } | |
1176 | ||
1177 | static void unlock_fs(struct mapped_device *md); | |
1178 | ||
1179 | static void free_dev(struct mapped_device *md) | |
1180 | { | |
1181 | int minor = MINOR(disk_devt(md->disk)); | |
1182 | ||
1183 | if (md->suspended_bdev) { | |
1184 | unlock_fs(md); | |
1185 | bdput(md->suspended_bdev); | |
1186 | } | |
1187 | destroy_workqueue(md->wq); | |
1188 | mempool_destroy(md->tio_pool); | |
1189 | mempool_destroy(md->io_pool); | |
1190 | bioset_free(md->bs); | |
1191 | del_gendisk(md->disk); | |
1192 | free_minor(minor); | |
1193 | ||
1194 | spin_lock(&_minor_lock); | |
1195 | md->disk->private_data = NULL; | |
1196 | spin_unlock(&_minor_lock); | |
1197 | ||
1198 | put_disk(md->disk); | |
1199 | blk_cleanup_queue(md->queue); | |
1200 | module_put(THIS_MODULE); | |
1201 | kfree(md); | |
1202 | } | |
1203 | ||
1204 | /* | |
1205 | * Bind a table to the device. | |
1206 | */ | |
1207 | static void event_callback(void *context) | |
1208 | { | |
1209 | unsigned long flags; | |
1210 | LIST_HEAD(uevents); | |
1211 | struct mapped_device *md = (struct mapped_device *) context; | |
1212 | ||
1213 | spin_lock_irqsave(&md->uevent_lock, flags); | |
1214 | list_splice_init(&md->uevent_list, &uevents); | |
1215 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1216 | ||
1217 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); | |
1218 | ||
1219 | atomic_inc(&md->event_nr); | |
1220 | wake_up(&md->eventq); | |
1221 | } | |
1222 | ||
1223 | static void __set_size(struct mapped_device *md, sector_t size) | |
1224 | { | |
1225 | set_capacity(md->disk, size); | |
1226 | ||
1227 | mutex_lock(&md->suspended_bdev->bd_inode->i_mutex); | |
1228 | i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); | |
1229 | mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex); | |
1230 | } | |
1231 | ||
1232 | static int __bind(struct mapped_device *md, struct dm_table *t) | |
1233 | { | |
1234 | struct request_queue *q = md->queue; | |
1235 | sector_t size; | |
1236 | ||
1237 | size = dm_table_get_size(t); | |
1238 | ||
1239 | /* | |
1240 | * Wipe any geometry if the size of the table changed. | |
1241 | */ | |
1242 | if (size != get_capacity(md->disk)) | |
1243 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
1244 | ||
1245 | if (md->suspended_bdev) | |
1246 | __set_size(md, size); | |
1247 | ||
1248 | if (!size) { | |
1249 | dm_table_destroy(t); | |
1250 | return 0; | |
1251 | } | |
1252 | ||
1253 | dm_table_event_callback(t, event_callback, md); | |
1254 | ||
1255 | write_lock(&md->map_lock); | |
1256 | md->map = t; | |
1257 | dm_table_set_restrictions(t, q); | |
1258 | write_unlock(&md->map_lock); | |
1259 | ||
1260 | return 0; | |
1261 | } | |
1262 | ||
1263 | static void __unbind(struct mapped_device *md) | |
1264 | { | |
1265 | struct dm_table *map = md->map; | |
1266 | ||
1267 | if (!map) | |
1268 | return; | |
1269 | ||
1270 | dm_table_event_callback(map, NULL, NULL); | |
1271 | write_lock(&md->map_lock); | |
1272 | md->map = NULL; | |
1273 | write_unlock(&md->map_lock); | |
1274 | dm_table_destroy(map); | |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * Constructor for a new device. | |
1279 | */ | |
1280 | int dm_create(int minor, struct mapped_device **result) | |
1281 | { | |
1282 | struct mapped_device *md; | |
1283 | ||
1284 | md = alloc_dev(minor); | |
1285 | if (!md) | |
1286 | return -ENXIO; | |
1287 | ||
1288 | *result = md; | |
1289 | return 0; | |
1290 | } | |
1291 | ||
1292 | static struct mapped_device *dm_find_md(dev_t dev) | |
1293 | { | |
1294 | struct mapped_device *md; | |
1295 | unsigned minor = MINOR(dev); | |
1296 | ||
1297 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
1298 | return NULL; | |
1299 | ||
1300 | spin_lock(&_minor_lock); | |
1301 | ||
1302 | md = idr_find(&_minor_idr, minor); | |
1303 | if (md && (md == MINOR_ALLOCED || | |
1304 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
1305 | test_bit(DMF_FREEING, &md->flags))) { | |
1306 | md = NULL; | |
1307 | goto out; | |
1308 | } | |
1309 | ||
1310 | out: | |
1311 | spin_unlock(&_minor_lock); | |
1312 | ||
1313 | return md; | |
1314 | } | |
1315 | ||
1316 | struct mapped_device *dm_get_md(dev_t dev) | |
1317 | { | |
1318 | struct mapped_device *md = dm_find_md(dev); | |
1319 | ||
1320 | if (md) | |
1321 | dm_get(md); | |
1322 | ||
1323 | return md; | |
1324 | } | |
1325 | ||
1326 | void *dm_get_mdptr(struct mapped_device *md) | |
1327 | { | |
1328 | return md->interface_ptr; | |
1329 | } | |
1330 | ||
1331 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
1332 | { | |
1333 | md->interface_ptr = ptr; | |
1334 | } | |
1335 | ||
1336 | void dm_get(struct mapped_device *md) | |
1337 | { | |
1338 | atomic_inc(&md->holders); | |
1339 | } | |
1340 | ||
1341 | const char *dm_device_name(struct mapped_device *md) | |
1342 | { | |
1343 | return md->name; | |
1344 | } | |
1345 | EXPORT_SYMBOL_GPL(dm_device_name); | |
1346 | ||
1347 | void dm_put(struct mapped_device *md) | |
1348 | { | |
1349 | struct dm_table *map; | |
1350 | ||
1351 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); | |
1352 | ||
1353 | if (atomic_dec_and_lock(&md->holders, &_minor_lock)) { | |
1354 | map = dm_get_table(md); | |
1355 | idr_replace(&_minor_idr, MINOR_ALLOCED, | |
1356 | MINOR(disk_devt(dm_disk(md)))); | |
1357 | set_bit(DMF_FREEING, &md->flags); | |
1358 | spin_unlock(&_minor_lock); | |
1359 | if (!dm_suspended(md)) { | |
1360 | dm_table_presuspend_targets(map); | |
1361 | dm_table_postsuspend_targets(map); | |
1362 | } | |
1363 | dm_table_put(map); | |
1364 | __unbind(md); | |
1365 | free_dev(md); | |
1366 | } | |
1367 | } | |
1368 | EXPORT_SYMBOL_GPL(dm_put); | |
1369 | ||
1370 | static int dm_wait_for_completion(struct mapped_device *md) | |
1371 | { | |
1372 | int r = 0; | |
1373 | ||
1374 | while (1) { | |
1375 | set_current_state(TASK_INTERRUPTIBLE); | |
1376 | ||
1377 | smp_mb(); | |
1378 | if (!atomic_read(&md->pending)) | |
1379 | break; | |
1380 | ||
1381 | if (signal_pending(current)) { | |
1382 | r = -EINTR; | |
1383 | break; | |
1384 | } | |
1385 | ||
1386 | io_schedule(); | |
1387 | } | |
1388 | set_current_state(TASK_RUNNING); | |
1389 | ||
1390 | return r; | |
1391 | } | |
1392 | ||
1393 | /* | |
1394 | * Process the deferred bios | |
1395 | */ | |
1396 | static void __flush_deferred_io(struct mapped_device *md) | |
1397 | { | |
1398 | struct bio *c; | |
1399 | ||
1400 | while ((c = bio_list_pop(&md->deferred))) { | |
1401 | if (__split_bio(md, c)) | |
1402 | bio_io_error(c); | |
1403 | } | |
1404 | ||
1405 | clear_bit(DMF_BLOCK_IO, &md->flags); | |
1406 | } | |
1407 | ||
1408 | static void __merge_pushback_list(struct mapped_device *md) | |
1409 | { | |
1410 | unsigned long flags; | |
1411 | ||
1412 | spin_lock_irqsave(&md->pushback_lock, flags); | |
1413 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1414 | bio_list_merge_head(&md->deferred, &md->pushback); | |
1415 | bio_list_init(&md->pushback); | |
1416 | spin_unlock_irqrestore(&md->pushback_lock, flags); | |
1417 | } | |
1418 | ||
1419 | static void dm_wq_work(struct work_struct *work) | |
1420 | { | |
1421 | struct dm_wq_req *req = container_of(work, struct dm_wq_req, work); | |
1422 | struct mapped_device *md = req->md; | |
1423 | ||
1424 | down_write(&md->io_lock); | |
1425 | switch (req->type) { | |
1426 | case DM_WQ_FLUSH_DEFERRED: | |
1427 | __flush_deferred_io(md); | |
1428 | break; | |
1429 | default: | |
1430 | DMERR("dm_wq_work: unrecognised work type %d", req->type); | |
1431 | BUG(); | |
1432 | } | |
1433 | up_write(&md->io_lock); | |
1434 | } | |
1435 | ||
1436 | static void dm_wq_queue(struct mapped_device *md, int type, void *context, | |
1437 | struct dm_wq_req *req) | |
1438 | { | |
1439 | req->type = type; | |
1440 | req->md = md; | |
1441 | req->context = context; | |
1442 | INIT_WORK(&req->work, dm_wq_work); | |
1443 | queue_work(md->wq, &req->work); | |
1444 | } | |
1445 | ||
1446 | static void dm_queue_flush(struct mapped_device *md, int type, void *context) | |
1447 | { | |
1448 | struct dm_wq_req req; | |
1449 | ||
1450 | dm_wq_queue(md, type, context, &req); | |
1451 | flush_workqueue(md->wq); | |
1452 | } | |
1453 | ||
1454 | /* | |
1455 | * Swap in a new table (destroying old one). | |
1456 | */ | |
1457 | int dm_swap_table(struct mapped_device *md, struct dm_table *table) | |
1458 | { | |
1459 | int r = -EINVAL; | |
1460 | ||
1461 | mutex_lock(&md->suspend_lock); | |
1462 | ||
1463 | /* device must be suspended */ | |
1464 | if (!dm_suspended(md)) | |
1465 | goto out; | |
1466 | ||
1467 | /* without bdev, the device size cannot be changed */ | |
1468 | if (!md->suspended_bdev) | |
1469 | if (get_capacity(md->disk) != dm_table_get_size(table)) | |
1470 | goto out; | |
1471 | ||
1472 | __unbind(md); | |
1473 | r = __bind(md, table); | |
1474 | ||
1475 | out: | |
1476 | mutex_unlock(&md->suspend_lock); | |
1477 | return r; | |
1478 | } | |
1479 | ||
1480 | /* | |
1481 | * Functions to lock and unlock any filesystem running on the | |
1482 | * device. | |
1483 | */ | |
1484 | static int lock_fs(struct mapped_device *md) | |
1485 | { | |
1486 | int r; | |
1487 | ||
1488 | WARN_ON(md->frozen_sb); | |
1489 | ||
1490 | md->frozen_sb = freeze_bdev(md->suspended_bdev); | |
1491 | if (IS_ERR(md->frozen_sb)) { | |
1492 | r = PTR_ERR(md->frozen_sb); | |
1493 | md->frozen_sb = NULL; | |
1494 | return r; | |
1495 | } | |
1496 | ||
1497 | set_bit(DMF_FROZEN, &md->flags); | |
1498 | ||
1499 | /* don't bdput right now, we don't want the bdev | |
1500 | * to go away while it is locked. | |
1501 | */ | |
1502 | return 0; | |
1503 | } | |
1504 | ||
1505 | static void unlock_fs(struct mapped_device *md) | |
1506 | { | |
1507 | if (!test_bit(DMF_FROZEN, &md->flags)) | |
1508 | return; | |
1509 | ||
1510 | thaw_bdev(md->suspended_bdev, md->frozen_sb); | |
1511 | md->frozen_sb = NULL; | |
1512 | clear_bit(DMF_FROZEN, &md->flags); | |
1513 | } | |
1514 | ||
1515 | /* | |
1516 | * We need to be able to change a mapping table under a mounted | |
1517 | * filesystem. For example we might want to move some data in | |
1518 | * the background. Before the table can be swapped with | |
1519 | * dm_bind_table, dm_suspend must be called to flush any in | |
1520 | * flight bios and ensure that any further io gets deferred. | |
1521 | */ | |
1522 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
1523 | { | |
1524 | struct dm_table *map = NULL; | |
1525 | DECLARE_WAITQUEUE(wait, current); | |
1526 | int r = 0; | |
1527 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; | |
1528 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; | |
1529 | ||
1530 | mutex_lock(&md->suspend_lock); | |
1531 | ||
1532 | if (dm_suspended(md)) { | |
1533 | r = -EINVAL; | |
1534 | goto out_unlock; | |
1535 | } | |
1536 | ||
1537 | map = dm_get_table(md); | |
1538 | ||
1539 | /* | |
1540 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
1541 | * This flag is cleared before dm_suspend returns. | |
1542 | */ | |
1543 | if (noflush) | |
1544 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1545 | ||
1546 | /* This does not get reverted if there's an error later. */ | |
1547 | dm_table_presuspend_targets(map); | |
1548 | ||
1549 | /* bdget() can stall if the pending I/Os are not flushed */ | |
1550 | if (!noflush) { | |
1551 | md->suspended_bdev = bdget_disk(md->disk, 0); | |
1552 | if (!md->suspended_bdev) { | |
1553 | DMWARN("bdget failed in dm_suspend"); | |
1554 | r = -ENOMEM; | |
1555 | goto out; | |
1556 | } | |
1557 | ||
1558 | /* | |
1559 | * Flush I/O to the device. noflush supersedes do_lockfs, | |
1560 | * because lock_fs() needs to flush I/Os. | |
1561 | */ | |
1562 | if (do_lockfs) { | |
1563 | r = lock_fs(md); | |
1564 | if (r) | |
1565 | goto out; | |
1566 | } | |
1567 | } | |
1568 | ||
1569 | /* | |
1570 | * First we set the BLOCK_IO flag so no more ios will be mapped. | |
1571 | */ | |
1572 | down_write(&md->io_lock); | |
1573 | set_bit(DMF_BLOCK_IO, &md->flags); | |
1574 | ||
1575 | add_wait_queue(&md->wait, &wait); | |
1576 | up_write(&md->io_lock); | |
1577 | ||
1578 | /* unplug */ | |
1579 | if (map) | |
1580 | dm_table_unplug_all(map); | |
1581 | ||
1582 | /* | |
1583 | * Wait for the already-mapped ios to complete. | |
1584 | */ | |
1585 | r = dm_wait_for_completion(md); | |
1586 | ||
1587 | down_write(&md->io_lock); | |
1588 | remove_wait_queue(&md->wait, &wait); | |
1589 | ||
1590 | if (noflush) | |
1591 | __merge_pushback_list(md); | |
1592 | up_write(&md->io_lock); | |
1593 | ||
1594 | /* were we interrupted ? */ | |
1595 | if (r < 0) { | |
1596 | dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL); | |
1597 | ||
1598 | unlock_fs(md); | |
1599 | goto out; /* pushback list is already flushed, so skip flush */ | |
1600 | } | |
1601 | ||
1602 | dm_table_postsuspend_targets(map); | |
1603 | ||
1604 | set_bit(DMF_SUSPENDED, &md->flags); | |
1605 | ||
1606 | out: | |
1607 | if (r && md->suspended_bdev) { | |
1608 | bdput(md->suspended_bdev); | |
1609 | md->suspended_bdev = NULL; | |
1610 | } | |
1611 | ||
1612 | dm_table_put(map); | |
1613 | ||
1614 | out_unlock: | |
1615 | mutex_unlock(&md->suspend_lock); | |
1616 | return r; | |
1617 | } | |
1618 | ||
1619 | int dm_resume(struct mapped_device *md) | |
1620 | { | |
1621 | int r = -EINVAL; | |
1622 | struct dm_table *map = NULL; | |
1623 | ||
1624 | mutex_lock(&md->suspend_lock); | |
1625 | if (!dm_suspended(md)) | |
1626 | goto out; | |
1627 | ||
1628 | map = dm_get_table(md); | |
1629 | if (!map || !dm_table_get_size(map)) | |
1630 | goto out; | |
1631 | ||
1632 | r = dm_table_resume_targets(map); | |
1633 | if (r) | |
1634 | goto out; | |
1635 | ||
1636 | dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL); | |
1637 | ||
1638 | unlock_fs(md); | |
1639 | ||
1640 | if (md->suspended_bdev) { | |
1641 | bdput(md->suspended_bdev); | |
1642 | md->suspended_bdev = NULL; | |
1643 | } | |
1644 | ||
1645 | clear_bit(DMF_SUSPENDED, &md->flags); | |
1646 | ||
1647 | dm_table_unplug_all(map); | |
1648 | ||
1649 | dm_kobject_uevent(md); | |
1650 | ||
1651 | r = 0; | |
1652 | ||
1653 | out: | |
1654 | dm_table_put(map); | |
1655 | mutex_unlock(&md->suspend_lock); | |
1656 | ||
1657 | return r; | |
1658 | } | |
1659 | ||
1660 | /*----------------------------------------------------------------- | |
1661 | * Event notification. | |
1662 | *---------------------------------------------------------------*/ | |
1663 | void dm_kobject_uevent(struct mapped_device *md) | |
1664 | { | |
1665 | kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE); | |
1666 | } | |
1667 | ||
1668 | uint32_t dm_next_uevent_seq(struct mapped_device *md) | |
1669 | { | |
1670 | return atomic_add_return(1, &md->uevent_seq); | |
1671 | } | |
1672 | ||
1673 | uint32_t dm_get_event_nr(struct mapped_device *md) | |
1674 | { | |
1675 | return atomic_read(&md->event_nr); | |
1676 | } | |
1677 | ||
1678 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
1679 | { | |
1680 | return wait_event_interruptible(md->eventq, | |
1681 | (event_nr != atomic_read(&md->event_nr))); | |
1682 | } | |
1683 | ||
1684 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) | |
1685 | { | |
1686 | unsigned long flags; | |
1687 | ||
1688 | spin_lock_irqsave(&md->uevent_lock, flags); | |
1689 | list_add(elist, &md->uevent_list); | |
1690 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1691 | } | |
1692 | ||
1693 | /* | |
1694 | * The gendisk is only valid as long as you have a reference | |
1695 | * count on 'md'. | |
1696 | */ | |
1697 | struct gendisk *dm_disk(struct mapped_device *md) | |
1698 | { | |
1699 | return md->disk; | |
1700 | } | |
1701 | ||
1702 | int dm_suspended(struct mapped_device *md) | |
1703 | { | |
1704 | return test_bit(DMF_SUSPENDED, &md->flags); | |
1705 | } | |
1706 | ||
1707 | int dm_noflush_suspending(struct dm_target *ti) | |
1708 | { | |
1709 | struct mapped_device *md = dm_table_get_md(ti->table); | |
1710 | int r = __noflush_suspending(md); | |
1711 | ||
1712 | dm_put(md); | |
1713 | ||
1714 | return r; | |
1715 | } | |
1716 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
1717 | ||
1718 | static struct block_device_operations dm_blk_dops = { | |
1719 | .open = dm_blk_open, | |
1720 | .release = dm_blk_close, | |
1721 | .ioctl = dm_blk_ioctl, | |
1722 | .getgeo = dm_blk_getgeo, | |
1723 | .owner = THIS_MODULE | |
1724 | }; | |
1725 | ||
1726 | EXPORT_SYMBOL(dm_get_mapinfo); | |
1727 | ||
1728 | /* | |
1729 | * module hooks | |
1730 | */ | |
1731 | module_init(dm_init); | |
1732 | module_exit(dm_exit); | |
1733 | ||
1734 | module_param(major, uint, 0); | |
1735 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
1736 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
1737 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
1738 | MODULE_LICENSE("GPL"); |