Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
2b06cfff | 3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
9 | #include "dm-bio-list.h" | |
10 | ||
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
48c9c27b | 13 | #include <linux/mutex.h> |
1da177e4 LT |
14 | #include <linux/moduleparam.h> |
15 | #include <linux/blkpg.h> | |
16 | #include <linux/bio.h> | |
17 | #include <linux/buffer_head.h> | |
18 | #include <linux/mempool.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/idr.h> | |
3ac51e74 | 21 | #include <linux/hdreg.h> |
2056a782 | 22 | #include <linux/blktrace_api.h> |
aa129a22 | 23 | #include <linux/smp_lock.h> |
1da177e4 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "core" |
26 | ||
1da177e4 LT |
27 | static const char *_name = DM_NAME; |
28 | ||
29 | static unsigned int major = 0; | |
30 | static unsigned int _major = 0; | |
31 | ||
f32c10b0 | 32 | static DEFINE_SPINLOCK(_minor_lock); |
1da177e4 LT |
33 | /* |
34 | * One of these is allocated per bio. | |
35 | */ | |
36 | struct dm_io { | |
37 | struct mapped_device *md; | |
38 | int error; | |
39 | struct bio *bio; | |
40 | atomic_t io_count; | |
3eaf840e | 41 | unsigned long start_time; |
1da177e4 LT |
42 | }; |
43 | ||
44 | /* | |
45 | * One of these is allocated per target within a bio. Hopefully | |
46 | * this will be simplified out one day. | |
47 | */ | |
028867ac | 48 | struct dm_target_io { |
1da177e4 LT |
49 | struct dm_io *io; |
50 | struct dm_target *ti; | |
51 | union map_info info; | |
52 | }; | |
53 | ||
54 | union map_info *dm_get_mapinfo(struct bio *bio) | |
55 | { | |
17b2f66f | 56 | if (bio && bio->bi_private) |
028867ac | 57 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 58 | return NULL; |
1da177e4 LT |
59 | } |
60 | ||
ba61fdd1 JM |
61 | #define MINOR_ALLOCED ((void *)-1) |
62 | ||
1da177e4 LT |
63 | /* |
64 | * Bits for the md->flags field. | |
65 | */ | |
66 | #define DMF_BLOCK_IO 0 | |
67 | #define DMF_SUSPENDED 1 | |
aa8d7c2f | 68 | #define DMF_FROZEN 2 |
fba9f90e | 69 | #define DMF_FREEING 3 |
5c6bd75d | 70 | #define DMF_DELETING 4 |
2e93ccc1 | 71 | #define DMF_NOFLUSH_SUSPENDING 5 |
1da177e4 LT |
72 | |
73 | struct mapped_device { | |
2ca3310e AK |
74 | struct rw_semaphore io_lock; |
75 | struct semaphore suspend_lock; | |
2e93ccc1 | 76 | spinlock_t pushback_lock; |
1da177e4 LT |
77 | rwlock_t map_lock; |
78 | atomic_t holders; | |
5c6bd75d | 79 | atomic_t open_count; |
1da177e4 LT |
80 | |
81 | unsigned long flags; | |
82 | ||
165125e1 | 83 | struct request_queue *queue; |
1da177e4 | 84 | struct gendisk *disk; |
7e51f257 | 85 | char name[16]; |
1da177e4 LT |
86 | |
87 | void *interface_ptr; | |
88 | ||
89 | /* | |
90 | * A list of ios that arrived while we were suspended. | |
91 | */ | |
92 | atomic_t pending; | |
93 | wait_queue_head_t wait; | |
74859364 | 94 | struct bio_list deferred; |
2e93ccc1 | 95 | struct bio_list pushback; |
1da177e4 LT |
96 | |
97 | /* | |
98 | * The current mapping. | |
99 | */ | |
100 | struct dm_table *map; | |
101 | ||
102 | /* | |
103 | * io objects are allocated from here. | |
104 | */ | |
105 | mempool_t *io_pool; | |
106 | mempool_t *tio_pool; | |
107 | ||
9faf400f SB |
108 | struct bio_set *bs; |
109 | ||
1da177e4 LT |
110 | /* |
111 | * Event handling. | |
112 | */ | |
113 | atomic_t event_nr; | |
114 | wait_queue_head_t eventq; | |
115 | ||
116 | /* | |
117 | * freeze/thaw support require holding onto a super block | |
118 | */ | |
119 | struct super_block *frozen_sb; | |
e39e2e95 | 120 | struct block_device *suspended_bdev; |
3ac51e74 DW |
121 | |
122 | /* forced geometry settings */ | |
123 | struct hd_geometry geometry; | |
1da177e4 LT |
124 | }; |
125 | ||
126 | #define MIN_IOS 256 | |
e18b890b CL |
127 | static struct kmem_cache *_io_cache; |
128 | static struct kmem_cache *_tio_cache; | |
1da177e4 | 129 | |
1da177e4 LT |
130 | static int __init local_init(void) |
131 | { | |
132 | int r; | |
133 | ||
1da177e4 | 134 | /* allocate a slab for the dm_ios */ |
028867ac | 135 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 LT |
136 | if (!_io_cache) |
137 | return -ENOMEM; | |
138 | ||
139 | /* allocate a slab for the target ios */ | |
028867ac | 140 | _tio_cache = KMEM_CACHE(dm_target_io, 0); |
1da177e4 LT |
141 | if (!_tio_cache) { |
142 | kmem_cache_destroy(_io_cache); | |
143 | return -ENOMEM; | |
144 | } | |
145 | ||
146 | _major = major; | |
147 | r = register_blkdev(_major, _name); | |
148 | if (r < 0) { | |
149 | kmem_cache_destroy(_tio_cache); | |
150 | kmem_cache_destroy(_io_cache); | |
151 | return r; | |
152 | } | |
153 | ||
154 | if (!_major) | |
155 | _major = r; | |
156 | ||
157 | return 0; | |
158 | } | |
159 | ||
160 | static void local_exit(void) | |
161 | { | |
162 | kmem_cache_destroy(_tio_cache); | |
163 | kmem_cache_destroy(_io_cache); | |
00d59405 | 164 | unregister_blkdev(_major, _name); |
1da177e4 LT |
165 | |
166 | _major = 0; | |
167 | ||
168 | DMINFO("cleaned up"); | |
169 | } | |
170 | ||
171 | int (*_inits[])(void) __initdata = { | |
172 | local_init, | |
173 | dm_target_init, | |
174 | dm_linear_init, | |
175 | dm_stripe_init, | |
176 | dm_interface_init, | |
177 | }; | |
178 | ||
179 | void (*_exits[])(void) = { | |
180 | local_exit, | |
181 | dm_target_exit, | |
182 | dm_linear_exit, | |
183 | dm_stripe_exit, | |
184 | dm_interface_exit, | |
185 | }; | |
186 | ||
187 | static int __init dm_init(void) | |
188 | { | |
189 | const int count = ARRAY_SIZE(_inits); | |
190 | ||
191 | int r, i; | |
192 | ||
193 | for (i = 0; i < count; i++) { | |
194 | r = _inits[i](); | |
195 | if (r) | |
196 | goto bad; | |
197 | } | |
198 | ||
199 | return 0; | |
200 | ||
201 | bad: | |
202 | while (i--) | |
203 | _exits[i](); | |
204 | ||
205 | return r; | |
206 | } | |
207 | ||
208 | static void __exit dm_exit(void) | |
209 | { | |
210 | int i = ARRAY_SIZE(_exits); | |
211 | ||
212 | while (i--) | |
213 | _exits[i](); | |
214 | } | |
215 | ||
216 | /* | |
217 | * Block device functions | |
218 | */ | |
219 | static int dm_blk_open(struct inode *inode, struct file *file) | |
220 | { | |
221 | struct mapped_device *md; | |
222 | ||
fba9f90e JM |
223 | spin_lock(&_minor_lock); |
224 | ||
1da177e4 | 225 | md = inode->i_bdev->bd_disk->private_data; |
fba9f90e JM |
226 | if (!md) |
227 | goto out; | |
228 | ||
5c6bd75d AK |
229 | if (test_bit(DMF_FREEING, &md->flags) || |
230 | test_bit(DMF_DELETING, &md->flags)) { | |
fba9f90e JM |
231 | md = NULL; |
232 | goto out; | |
233 | } | |
234 | ||
1da177e4 | 235 | dm_get(md); |
5c6bd75d | 236 | atomic_inc(&md->open_count); |
fba9f90e JM |
237 | |
238 | out: | |
239 | spin_unlock(&_minor_lock); | |
240 | ||
241 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
242 | } |
243 | ||
244 | static int dm_blk_close(struct inode *inode, struct file *file) | |
245 | { | |
246 | struct mapped_device *md; | |
247 | ||
248 | md = inode->i_bdev->bd_disk->private_data; | |
5c6bd75d | 249 | atomic_dec(&md->open_count); |
1da177e4 LT |
250 | dm_put(md); |
251 | return 0; | |
252 | } | |
253 | ||
5c6bd75d AK |
254 | int dm_open_count(struct mapped_device *md) |
255 | { | |
256 | return atomic_read(&md->open_count); | |
257 | } | |
258 | ||
259 | /* | |
260 | * Guarantees nothing is using the device before it's deleted. | |
261 | */ | |
262 | int dm_lock_for_deletion(struct mapped_device *md) | |
263 | { | |
264 | int r = 0; | |
265 | ||
266 | spin_lock(&_minor_lock); | |
267 | ||
268 | if (dm_open_count(md)) | |
269 | r = -EBUSY; | |
270 | else | |
271 | set_bit(DMF_DELETING, &md->flags); | |
272 | ||
273 | spin_unlock(&_minor_lock); | |
274 | ||
275 | return r; | |
276 | } | |
277 | ||
3ac51e74 DW |
278 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
279 | { | |
280 | struct mapped_device *md = bdev->bd_disk->private_data; | |
281 | ||
282 | return dm_get_geometry(md, geo); | |
283 | } | |
284 | ||
aa129a22 MB |
285 | static int dm_blk_ioctl(struct inode *inode, struct file *file, |
286 | unsigned int cmd, unsigned long arg) | |
287 | { | |
288 | struct mapped_device *md; | |
289 | struct dm_table *map; | |
290 | struct dm_target *tgt; | |
291 | int r = -ENOTTY; | |
292 | ||
293 | /* We don't really need this lock, but we do need 'inode'. */ | |
294 | unlock_kernel(); | |
295 | ||
296 | md = inode->i_bdev->bd_disk->private_data; | |
297 | ||
298 | map = dm_get_table(md); | |
299 | ||
300 | if (!map || !dm_table_get_size(map)) | |
301 | goto out; | |
302 | ||
303 | /* We only support devices that have a single target */ | |
304 | if (dm_table_get_num_targets(map) != 1) | |
305 | goto out; | |
306 | ||
307 | tgt = dm_table_get_target(map, 0); | |
308 | ||
309 | if (dm_suspended(md)) { | |
310 | r = -EAGAIN; | |
311 | goto out; | |
312 | } | |
313 | ||
314 | if (tgt->type->ioctl) | |
315 | r = tgt->type->ioctl(tgt, inode, file, cmd, arg); | |
316 | ||
317 | out: | |
318 | dm_table_put(map); | |
319 | ||
320 | lock_kernel(); | |
321 | return r; | |
322 | } | |
323 | ||
028867ac | 324 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
325 | { |
326 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
327 | } | |
328 | ||
028867ac | 329 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
330 | { |
331 | mempool_free(io, md->io_pool); | |
332 | } | |
333 | ||
028867ac | 334 | static struct dm_target_io *alloc_tio(struct mapped_device *md) |
1da177e4 LT |
335 | { |
336 | return mempool_alloc(md->tio_pool, GFP_NOIO); | |
337 | } | |
338 | ||
028867ac | 339 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 LT |
340 | { |
341 | mempool_free(tio, md->tio_pool); | |
342 | } | |
343 | ||
3eaf840e JNN |
344 | static void start_io_acct(struct dm_io *io) |
345 | { | |
346 | struct mapped_device *md = io->md; | |
347 | ||
348 | io->start_time = jiffies; | |
349 | ||
350 | preempt_disable(); | |
351 | disk_round_stats(dm_disk(md)); | |
352 | preempt_enable(); | |
353 | dm_disk(md)->in_flight = atomic_inc_return(&md->pending); | |
354 | } | |
355 | ||
356 | static int end_io_acct(struct dm_io *io) | |
357 | { | |
358 | struct mapped_device *md = io->md; | |
359 | struct bio *bio = io->bio; | |
360 | unsigned long duration = jiffies - io->start_time; | |
361 | int pending; | |
362 | int rw = bio_data_dir(bio); | |
363 | ||
364 | preempt_disable(); | |
365 | disk_round_stats(dm_disk(md)); | |
366 | preempt_enable(); | |
367 | dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending); | |
368 | ||
369 | disk_stat_add(dm_disk(md), ticks[rw], duration); | |
370 | ||
371 | return !pending; | |
372 | } | |
373 | ||
1da177e4 LT |
374 | /* |
375 | * Add the bio to the list of deferred io. | |
376 | */ | |
377 | static int queue_io(struct mapped_device *md, struct bio *bio) | |
378 | { | |
2ca3310e | 379 | down_write(&md->io_lock); |
1da177e4 LT |
380 | |
381 | if (!test_bit(DMF_BLOCK_IO, &md->flags)) { | |
2ca3310e | 382 | up_write(&md->io_lock); |
1da177e4 LT |
383 | return 1; |
384 | } | |
385 | ||
386 | bio_list_add(&md->deferred, bio); | |
387 | ||
2ca3310e | 388 | up_write(&md->io_lock); |
1da177e4 LT |
389 | return 0; /* deferred successfully */ |
390 | } | |
391 | ||
392 | /* | |
393 | * Everyone (including functions in this file), should use this | |
394 | * function to access the md->map field, and make sure they call | |
395 | * dm_table_put() when finished. | |
396 | */ | |
397 | struct dm_table *dm_get_table(struct mapped_device *md) | |
398 | { | |
399 | struct dm_table *t; | |
400 | ||
401 | read_lock(&md->map_lock); | |
402 | t = md->map; | |
403 | if (t) | |
404 | dm_table_get(t); | |
405 | read_unlock(&md->map_lock); | |
406 | ||
407 | return t; | |
408 | } | |
409 | ||
3ac51e74 DW |
410 | /* |
411 | * Get the geometry associated with a dm device | |
412 | */ | |
413 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
414 | { | |
415 | *geo = md->geometry; | |
416 | ||
417 | return 0; | |
418 | } | |
419 | ||
420 | /* | |
421 | * Set the geometry of a device. | |
422 | */ | |
423 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
424 | { | |
425 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
426 | ||
427 | if (geo->start > sz) { | |
428 | DMWARN("Start sector is beyond the geometry limits."); | |
429 | return -EINVAL; | |
430 | } | |
431 | ||
432 | md->geometry = *geo; | |
433 | ||
434 | return 0; | |
435 | } | |
436 | ||
1da177e4 LT |
437 | /*----------------------------------------------------------------- |
438 | * CRUD START: | |
439 | * A more elegant soln is in the works that uses the queue | |
440 | * merge fn, unfortunately there are a couple of changes to | |
441 | * the block layer that I want to make for this. So in the | |
442 | * interests of getting something for people to use I give | |
443 | * you this clearly demarcated crap. | |
444 | *---------------------------------------------------------------*/ | |
445 | ||
2e93ccc1 KU |
446 | static int __noflush_suspending(struct mapped_device *md) |
447 | { | |
448 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
449 | } | |
450 | ||
1da177e4 LT |
451 | /* |
452 | * Decrements the number of outstanding ios that a bio has been | |
453 | * cloned into, completing the original io if necc. | |
454 | */ | |
858119e1 | 455 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 456 | { |
2e93ccc1 KU |
457 | unsigned long flags; |
458 | ||
459 | /* Push-back supersedes any I/O errors */ | |
460 | if (error && !(io->error > 0 && __noflush_suspending(io->md))) | |
1da177e4 LT |
461 | io->error = error; |
462 | ||
463 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
464 | if (io->error == DM_ENDIO_REQUEUE) { |
465 | /* | |
466 | * Target requested pushing back the I/O. | |
467 | * This must be handled before the sleeper on | |
468 | * suspend queue merges the pushback list. | |
469 | */ | |
470 | spin_lock_irqsave(&io->md->pushback_lock, flags); | |
471 | if (__noflush_suspending(io->md)) | |
472 | bio_list_add(&io->md->pushback, io->bio); | |
473 | else | |
474 | /* noflush suspend was interrupted. */ | |
475 | io->error = -EIO; | |
476 | spin_unlock_irqrestore(&io->md->pushback_lock, flags); | |
477 | } | |
478 | ||
3eaf840e | 479 | if (end_io_acct(io)) |
1da177e4 LT |
480 | /* nudge anyone waiting on suspend queue */ |
481 | wake_up(&io->md->wait); | |
482 | ||
2e93ccc1 KU |
483 | if (io->error != DM_ENDIO_REQUEUE) { |
484 | blk_add_trace_bio(io->md->queue, io->bio, | |
485 | BLK_TA_COMPLETE); | |
486 | ||
487 | bio_endio(io->bio, io->bio->bi_size, io->error); | |
488 | } | |
2056a782 | 489 | |
1da177e4 LT |
490 | free_io(io->md, io); |
491 | } | |
492 | } | |
493 | ||
494 | static int clone_endio(struct bio *bio, unsigned int done, int error) | |
495 | { | |
496 | int r = 0; | |
028867ac | 497 | struct dm_target_io *tio = bio->bi_private; |
9faf400f | 498 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
499 | dm_endio_fn endio = tio->ti->type->end_io; |
500 | ||
501 | if (bio->bi_size) | |
502 | return 1; | |
503 | ||
504 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) | |
505 | error = -EIO; | |
506 | ||
507 | if (endio) { | |
508 | r = endio(tio->ti, bio, error, &tio->info); | |
2e93ccc1 KU |
509 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
510 | /* | |
511 | * error and requeue request are handled | |
512 | * in dec_pending(). | |
513 | */ | |
1da177e4 | 514 | error = r; |
45cbcd79 KU |
515 | else if (r == DM_ENDIO_INCOMPLETE) |
516 | /* The target will handle the io */ | |
1da177e4 | 517 | return 1; |
45cbcd79 KU |
518 | else if (r) { |
519 | DMWARN("unimplemented target endio return value: %d", r); | |
520 | BUG(); | |
521 | } | |
1da177e4 LT |
522 | } |
523 | ||
9faf400f SB |
524 | dec_pending(tio->io, error); |
525 | ||
526 | /* | |
527 | * Store md for cleanup instead of tio which is about to get freed. | |
528 | */ | |
529 | bio->bi_private = md->bs; | |
530 | ||
1da177e4 | 531 | bio_put(bio); |
9faf400f | 532 | free_tio(md, tio); |
1da177e4 LT |
533 | return r; |
534 | } | |
535 | ||
536 | static sector_t max_io_len(struct mapped_device *md, | |
537 | sector_t sector, struct dm_target *ti) | |
538 | { | |
539 | sector_t offset = sector - ti->begin; | |
540 | sector_t len = ti->len - offset; | |
541 | ||
542 | /* | |
543 | * Does the target need to split even further ? | |
544 | */ | |
545 | if (ti->split_io) { | |
546 | sector_t boundary; | |
547 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) | |
548 | - offset; | |
549 | if (len > boundary) | |
550 | len = boundary; | |
551 | } | |
552 | ||
553 | return len; | |
554 | } | |
555 | ||
556 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
028867ac | 557 | struct dm_target_io *tio) |
1da177e4 LT |
558 | { |
559 | int r; | |
2056a782 | 560 | sector_t sector; |
9faf400f | 561 | struct mapped_device *md; |
1da177e4 LT |
562 | |
563 | /* | |
564 | * Sanity checks. | |
565 | */ | |
566 | BUG_ON(!clone->bi_size); | |
567 | ||
568 | clone->bi_end_io = clone_endio; | |
569 | clone->bi_private = tio; | |
570 | ||
571 | /* | |
572 | * Map the clone. If r == 0 we don't need to do | |
573 | * anything, the target has assumed ownership of | |
574 | * this io. | |
575 | */ | |
576 | atomic_inc(&tio->io->io_count); | |
2056a782 | 577 | sector = clone->bi_sector; |
1da177e4 | 578 | r = ti->type->map(ti, clone, &tio->info); |
45cbcd79 | 579 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 580 | /* the bio has been remapped so dispatch it */ |
2056a782 | 581 | |
17b2f66f | 582 | blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone, |
c7149d6b AB |
583 | tio->io->bio->bi_bdev->bd_dev, |
584 | clone->bi_sector, sector); | |
2056a782 | 585 | |
1da177e4 | 586 | generic_make_request(clone); |
2e93ccc1 KU |
587 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
588 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
589 | md = tio->io->md; |
590 | dec_pending(tio->io, r); | |
591 | /* | |
592 | * Store bio_set for cleanup. | |
593 | */ | |
594 | clone->bi_private = md->bs; | |
1da177e4 | 595 | bio_put(clone); |
9faf400f | 596 | free_tio(md, tio); |
45cbcd79 KU |
597 | } else if (r) { |
598 | DMWARN("unimplemented target map return value: %d", r); | |
599 | BUG(); | |
1da177e4 LT |
600 | } |
601 | } | |
602 | ||
603 | struct clone_info { | |
604 | struct mapped_device *md; | |
605 | struct dm_table *map; | |
606 | struct bio *bio; | |
607 | struct dm_io *io; | |
608 | sector_t sector; | |
609 | sector_t sector_count; | |
610 | unsigned short idx; | |
611 | }; | |
612 | ||
3676347a PO |
613 | static void dm_bio_destructor(struct bio *bio) |
614 | { | |
9faf400f SB |
615 | struct bio_set *bs = bio->bi_private; |
616 | ||
617 | bio_free(bio, bs); | |
3676347a PO |
618 | } |
619 | ||
1da177e4 LT |
620 | /* |
621 | * Creates a little bio that is just does part of a bvec. | |
622 | */ | |
623 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
624 | unsigned short idx, unsigned int offset, | |
9faf400f | 625 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
626 | { |
627 | struct bio *clone; | |
628 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
629 | ||
9faf400f | 630 | clone = bio_alloc_bioset(GFP_NOIO, 1, bs); |
3676347a | 631 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
632 | *clone->bi_io_vec = *bv; |
633 | ||
634 | clone->bi_sector = sector; | |
635 | clone->bi_bdev = bio->bi_bdev; | |
636 | clone->bi_rw = bio->bi_rw; | |
637 | clone->bi_vcnt = 1; | |
638 | clone->bi_size = to_bytes(len); | |
639 | clone->bi_io_vec->bv_offset = offset; | |
640 | clone->bi_io_vec->bv_len = clone->bi_size; | |
641 | ||
642 | return clone; | |
643 | } | |
644 | ||
645 | /* | |
646 | * Creates a bio that consists of range of complete bvecs. | |
647 | */ | |
648 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
649 | unsigned short idx, unsigned short bv_count, | |
9faf400f | 650 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
651 | { |
652 | struct bio *clone; | |
653 | ||
9faf400f SB |
654 | clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); |
655 | __bio_clone(clone, bio); | |
656 | clone->bi_destructor = dm_bio_destructor; | |
1da177e4 LT |
657 | clone->bi_sector = sector; |
658 | clone->bi_idx = idx; | |
659 | clone->bi_vcnt = idx + bv_count; | |
660 | clone->bi_size = to_bytes(len); | |
661 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
662 | ||
663 | return clone; | |
664 | } | |
665 | ||
666 | static void __clone_and_map(struct clone_info *ci) | |
667 | { | |
668 | struct bio *clone, *bio = ci->bio; | |
669 | struct dm_target *ti = dm_table_find_target(ci->map, ci->sector); | |
670 | sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti); | |
028867ac | 671 | struct dm_target_io *tio; |
1da177e4 LT |
672 | |
673 | /* | |
674 | * Allocate a target io object. | |
675 | */ | |
676 | tio = alloc_tio(ci->md); | |
677 | tio->io = ci->io; | |
678 | tio->ti = ti; | |
679 | memset(&tio->info, 0, sizeof(tio->info)); | |
680 | ||
681 | if (ci->sector_count <= max) { | |
682 | /* | |
683 | * Optimise for the simple case where we can do all of | |
684 | * the remaining io with a single clone. | |
685 | */ | |
686 | clone = clone_bio(bio, ci->sector, ci->idx, | |
9faf400f SB |
687 | bio->bi_vcnt - ci->idx, ci->sector_count, |
688 | ci->md->bs); | |
1da177e4 LT |
689 | __map_bio(ti, clone, tio); |
690 | ci->sector_count = 0; | |
691 | ||
692 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
693 | /* | |
694 | * There are some bvecs that don't span targets. | |
695 | * Do as many of these as possible. | |
696 | */ | |
697 | int i; | |
698 | sector_t remaining = max; | |
699 | sector_t bv_len; | |
700 | ||
701 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
702 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
703 | ||
704 | if (bv_len > remaining) | |
705 | break; | |
706 | ||
707 | remaining -= bv_len; | |
708 | len += bv_len; | |
709 | } | |
710 | ||
9faf400f SB |
711 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, |
712 | ci->md->bs); | |
1da177e4 LT |
713 | __map_bio(ti, clone, tio); |
714 | ||
715 | ci->sector += len; | |
716 | ci->sector_count -= len; | |
717 | ci->idx = i; | |
718 | ||
719 | } else { | |
720 | /* | |
d2044a94 | 721 | * Handle a bvec that must be split between two or more targets. |
1da177e4 LT |
722 | */ |
723 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
d2044a94 AK |
724 | sector_t remaining = to_sector(bv->bv_len); |
725 | unsigned int offset = 0; | |
1da177e4 | 726 | |
d2044a94 AK |
727 | do { |
728 | if (offset) { | |
729 | ti = dm_table_find_target(ci->map, ci->sector); | |
730 | max = max_io_len(ci->md, ci->sector, ti); | |
1da177e4 | 731 | |
d2044a94 AK |
732 | tio = alloc_tio(ci->md); |
733 | tio->io = ci->io; | |
734 | tio->ti = ti; | |
735 | memset(&tio->info, 0, sizeof(tio->info)); | |
736 | } | |
737 | ||
738 | len = min(remaining, max); | |
739 | ||
740 | clone = split_bvec(bio, ci->sector, ci->idx, | |
9faf400f SB |
741 | bv->bv_offset + offset, len, |
742 | ci->md->bs); | |
d2044a94 AK |
743 | |
744 | __map_bio(ti, clone, tio); | |
745 | ||
746 | ci->sector += len; | |
747 | ci->sector_count -= len; | |
748 | offset += to_bytes(len); | |
749 | } while (remaining -= len); | |
1da177e4 | 750 | |
1da177e4 LT |
751 | ci->idx++; |
752 | } | |
753 | } | |
754 | ||
755 | /* | |
756 | * Split the bio into several clones. | |
757 | */ | |
758 | static void __split_bio(struct mapped_device *md, struct bio *bio) | |
759 | { | |
760 | struct clone_info ci; | |
761 | ||
762 | ci.map = dm_get_table(md); | |
763 | if (!ci.map) { | |
764 | bio_io_error(bio, bio->bi_size); | |
765 | return; | |
766 | } | |
767 | ||
768 | ci.md = md; | |
769 | ci.bio = bio; | |
770 | ci.io = alloc_io(md); | |
771 | ci.io->error = 0; | |
772 | atomic_set(&ci.io->io_count, 1); | |
773 | ci.io->bio = bio; | |
774 | ci.io->md = md; | |
775 | ci.sector = bio->bi_sector; | |
776 | ci.sector_count = bio_sectors(bio); | |
777 | ci.idx = bio->bi_idx; | |
778 | ||
3eaf840e | 779 | start_io_acct(ci.io); |
1da177e4 LT |
780 | while (ci.sector_count) |
781 | __clone_and_map(&ci); | |
782 | ||
783 | /* drop the extra reference count */ | |
784 | dec_pending(ci.io, 0); | |
785 | dm_table_put(ci.map); | |
786 | } | |
787 | /*----------------------------------------------------------------- | |
788 | * CRUD END | |
789 | *---------------------------------------------------------------*/ | |
790 | ||
791 | /* | |
792 | * The request function that just remaps the bio built up by | |
793 | * dm_merge_bvec. | |
794 | */ | |
165125e1 | 795 | static int dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 LT |
796 | { |
797 | int r; | |
12f03a49 | 798 | int rw = bio_data_dir(bio); |
1da177e4 LT |
799 | struct mapped_device *md = q->queuedata; |
800 | ||
07a83c47 SB |
801 | /* |
802 | * There is no use in forwarding any barrier request since we can't | |
803 | * guarantee it is (or can be) handled by the targets correctly. | |
804 | */ | |
805 | if (unlikely(bio_barrier(bio))) { | |
806 | bio_endio(bio, bio->bi_size, -EOPNOTSUPP); | |
807 | return 0; | |
808 | } | |
809 | ||
2ca3310e | 810 | down_read(&md->io_lock); |
1da177e4 | 811 | |
12f03a49 KC |
812 | disk_stat_inc(dm_disk(md), ios[rw]); |
813 | disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio)); | |
814 | ||
1da177e4 LT |
815 | /* |
816 | * If we're suspended we have to queue | |
817 | * this io for later. | |
818 | */ | |
819 | while (test_bit(DMF_BLOCK_IO, &md->flags)) { | |
2ca3310e | 820 | up_read(&md->io_lock); |
1da177e4 LT |
821 | |
822 | if (bio_rw(bio) == READA) { | |
823 | bio_io_error(bio, bio->bi_size); | |
824 | return 0; | |
825 | } | |
826 | ||
827 | r = queue_io(md, bio); | |
828 | if (r < 0) { | |
829 | bio_io_error(bio, bio->bi_size); | |
830 | return 0; | |
831 | ||
832 | } else if (r == 0) | |
833 | return 0; /* deferred successfully */ | |
834 | ||
835 | /* | |
836 | * We're in a while loop, because someone could suspend | |
837 | * before we get to the following read lock. | |
838 | */ | |
2ca3310e | 839 | down_read(&md->io_lock); |
1da177e4 LT |
840 | } |
841 | ||
842 | __split_bio(md, bio); | |
2ca3310e | 843 | up_read(&md->io_lock); |
1da177e4 LT |
844 | return 0; |
845 | } | |
846 | ||
165125e1 | 847 | static int dm_flush_all(struct request_queue *q, struct gendisk *disk, |
1da177e4 LT |
848 | sector_t *error_sector) |
849 | { | |
850 | struct mapped_device *md = q->queuedata; | |
851 | struct dm_table *map = dm_get_table(md); | |
852 | int ret = -ENXIO; | |
853 | ||
854 | if (map) { | |
cf222b37 | 855 | ret = dm_table_flush_all(map); |
1da177e4 LT |
856 | dm_table_put(map); |
857 | } | |
858 | ||
859 | return ret; | |
860 | } | |
861 | ||
165125e1 | 862 | static void dm_unplug_all(struct request_queue *q) |
1da177e4 LT |
863 | { |
864 | struct mapped_device *md = q->queuedata; | |
865 | struct dm_table *map = dm_get_table(md); | |
866 | ||
867 | if (map) { | |
868 | dm_table_unplug_all(map); | |
869 | dm_table_put(map); | |
870 | } | |
871 | } | |
872 | ||
873 | static int dm_any_congested(void *congested_data, int bdi_bits) | |
874 | { | |
875 | int r; | |
876 | struct mapped_device *md = (struct mapped_device *) congested_data; | |
877 | struct dm_table *map = dm_get_table(md); | |
878 | ||
879 | if (!map || test_bit(DMF_BLOCK_IO, &md->flags)) | |
880 | r = bdi_bits; | |
881 | else | |
882 | r = dm_table_any_congested(map, bdi_bits); | |
883 | ||
884 | dm_table_put(map); | |
885 | return r; | |
886 | } | |
887 | ||
888 | /*----------------------------------------------------------------- | |
889 | * An IDR is used to keep track of allocated minor numbers. | |
890 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
891 | static DEFINE_IDR(_minor_idr); |
892 | ||
2b06cfff | 893 | static void free_minor(int minor) |
1da177e4 | 894 | { |
f32c10b0 | 895 | spin_lock(&_minor_lock); |
1da177e4 | 896 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 897 | spin_unlock(&_minor_lock); |
1da177e4 LT |
898 | } |
899 | ||
900 | /* | |
901 | * See if the device with a specific minor # is free. | |
902 | */ | |
2b06cfff | 903 | static int specific_minor(struct mapped_device *md, int minor) |
1da177e4 LT |
904 | { |
905 | int r, m; | |
906 | ||
907 | if (minor >= (1 << MINORBITS)) | |
908 | return -EINVAL; | |
909 | ||
62f75c2f JM |
910 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
911 | if (!r) | |
912 | return -ENOMEM; | |
913 | ||
f32c10b0 | 914 | spin_lock(&_minor_lock); |
1da177e4 LT |
915 | |
916 | if (idr_find(&_minor_idr, minor)) { | |
917 | r = -EBUSY; | |
918 | goto out; | |
919 | } | |
920 | ||
ba61fdd1 | 921 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); |
62f75c2f | 922 | if (r) |
1da177e4 | 923 | goto out; |
1da177e4 LT |
924 | |
925 | if (m != minor) { | |
926 | idr_remove(&_minor_idr, m); | |
927 | r = -EBUSY; | |
928 | goto out; | |
929 | } | |
930 | ||
931 | out: | |
f32c10b0 | 932 | spin_unlock(&_minor_lock); |
1da177e4 LT |
933 | return r; |
934 | } | |
935 | ||
2b06cfff | 936 | static int next_free_minor(struct mapped_device *md, int *minor) |
1da177e4 | 937 | { |
2b06cfff | 938 | int r, m; |
1da177e4 | 939 | |
1da177e4 | 940 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
62f75c2f JM |
941 | if (!r) |
942 | return -ENOMEM; | |
943 | ||
f32c10b0 | 944 | spin_lock(&_minor_lock); |
1da177e4 | 945 | |
ba61fdd1 | 946 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); |
1da177e4 LT |
947 | if (r) { |
948 | goto out; | |
949 | } | |
950 | ||
951 | if (m >= (1 << MINORBITS)) { | |
952 | idr_remove(&_minor_idr, m); | |
953 | r = -ENOSPC; | |
954 | goto out; | |
955 | } | |
956 | ||
957 | *minor = m; | |
958 | ||
959 | out: | |
f32c10b0 | 960 | spin_unlock(&_minor_lock); |
1da177e4 LT |
961 | return r; |
962 | } | |
963 | ||
964 | static struct block_device_operations dm_blk_dops; | |
965 | ||
966 | /* | |
967 | * Allocate and initialise a blank device with a given minor. | |
968 | */ | |
2b06cfff | 969 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
970 | { |
971 | int r; | |
972 | struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL); | |
ba61fdd1 | 973 | void *old_md; |
1da177e4 LT |
974 | |
975 | if (!md) { | |
976 | DMWARN("unable to allocate device, out of memory."); | |
977 | return NULL; | |
978 | } | |
979 | ||
10da4f79 JM |
980 | if (!try_module_get(THIS_MODULE)) |
981 | goto bad0; | |
982 | ||
1da177e4 | 983 | /* get a minor number for the dev */ |
2b06cfff AK |
984 | if (minor == DM_ANY_MINOR) |
985 | r = next_free_minor(md, &minor); | |
986 | else | |
987 | r = specific_minor(md, minor); | |
1da177e4 LT |
988 | if (r < 0) |
989 | goto bad1; | |
990 | ||
991 | memset(md, 0, sizeof(*md)); | |
2ca3310e AK |
992 | init_rwsem(&md->io_lock); |
993 | init_MUTEX(&md->suspend_lock); | |
2e93ccc1 | 994 | spin_lock_init(&md->pushback_lock); |
1da177e4 LT |
995 | rwlock_init(&md->map_lock); |
996 | atomic_set(&md->holders, 1); | |
5c6bd75d | 997 | atomic_set(&md->open_count, 0); |
1da177e4 LT |
998 | atomic_set(&md->event_nr, 0); |
999 | ||
1000 | md->queue = blk_alloc_queue(GFP_KERNEL); | |
1001 | if (!md->queue) | |
e3f6ac61 | 1002 | goto bad1_free_minor; |
1da177e4 LT |
1003 | |
1004 | md->queue->queuedata = md; | |
1005 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1006 | md->queue->backing_dev_info.congested_data = md; | |
1007 | blk_queue_make_request(md->queue, dm_request); | |
daef265f | 1008 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
1da177e4 LT |
1009 | md->queue->unplug_fn = dm_unplug_all; |
1010 | md->queue->issue_flush_fn = dm_flush_all; | |
1011 | ||
93d2341c | 1012 | md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache); |
74859364 KU |
1013 | if (!md->io_pool) |
1014 | goto bad2; | |
1da177e4 | 1015 | |
93d2341c | 1016 | md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache); |
1da177e4 LT |
1017 | if (!md->tio_pool) |
1018 | goto bad3; | |
1019 | ||
5972511b | 1020 | md->bs = bioset_create(16, 16); |
9faf400f SB |
1021 | if (!md->bs) |
1022 | goto bad_no_bioset; | |
1023 | ||
1da177e4 LT |
1024 | md->disk = alloc_disk(1); |
1025 | if (!md->disk) | |
1026 | goto bad4; | |
1027 | ||
f0b04115 JM |
1028 | atomic_set(&md->pending, 0); |
1029 | init_waitqueue_head(&md->wait); | |
1030 | init_waitqueue_head(&md->eventq); | |
1031 | ||
1da177e4 LT |
1032 | md->disk->major = _major; |
1033 | md->disk->first_minor = minor; | |
1034 | md->disk->fops = &dm_blk_dops; | |
1035 | md->disk->queue = md->queue; | |
1036 | md->disk->private_data = md; | |
1037 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1038 | add_disk(md->disk); | |
7e51f257 | 1039 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1040 | |
ba61fdd1 | 1041 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1042 | spin_lock(&_minor_lock); |
ba61fdd1 | 1043 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1044 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1045 | |
1046 | BUG_ON(old_md != MINOR_ALLOCED); | |
1047 | ||
1da177e4 LT |
1048 | return md; |
1049 | ||
1050 | bad4: | |
9faf400f SB |
1051 | bioset_free(md->bs); |
1052 | bad_no_bioset: | |
1da177e4 LT |
1053 | mempool_destroy(md->tio_pool); |
1054 | bad3: | |
1055 | mempool_destroy(md->io_pool); | |
1056 | bad2: | |
1312f40e | 1057 | blk_cleanup_queue(md->queue); |
e3f6ac61 | 1058 | bad1_free_minor: |
1da177e4 LT |
1059 | free_minor(minor); |
1060 | bad1: | |
10da4f79 JM |
1061 | module_put(THIS_MODULE); |
1062 | bad0: | |
1da177e4 LT |
1063 | kfree(md); |
1064 | return NULL; | |
1065 | } | |
1066 | ||
1067 | static void free_dev(struct mapped_device *md) | |
1068 | { | |
2b06cfff | 1069 | int minor = md->disk->first_minor; |
63d94e48 | 1070 | |
d9dde59b JN |
1071 | if (md->suspended_bdev) { |
1072 | thaw_bdev(md->suspended_bdev, NULL); | |
1073 | bdput(md->suspended_bdev); | |
1074 | } | |
1da177e4 LT |
1075 | mempool_destroy(md->tio_pool); |
1076 | mempool_destroy(md->io_pool); | |
9faf400f | 1077 | bioset_free(md->bs); |
1da177e4 | 1078 | del_gendisk(md->disk); |
63d94e48 | 1079 | free_minor(minor); |
fba9f90e JM |
1080 | |
1081 | spin_lock(&_minor_lock); | |
1082 | md->disk->private_data = NULL; | |
1083 | spin_unlock(&_minor_lock); | |
1084 | ||
1da177e4 | 1085 | put_disk(md->disk); |
1312f40e | 1086 | blk_cleanup_queue(md->queue); |
10da4f79 | 1087 | module_put(THIS_MODULE); |
1da177e4 LT |
1088 | kfree(md); |
1089 | } | |
1090 | ||
1091 | /* | |
1092 | * Bind a table to the device. | |
1093 | */ | |
1094 | static void event_callback(void *context) | |
1095 | { | |
1096 | struct mapped_device *md = (struct mapped_device *) context; | |
1097 | ||
1098 | atomic_inc(&md->event_nr); | |
1099 | wake_up(&md->eventq); | |
1100 | } | |
1101 | ||
4e90188b | 1102 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1103 | { |
4e90188b | 1104 | set_capacity(md->disk, size); |
1da177e4 | 1105 | |
1b1dcc1b | 1106 | mutex_lock(&md->suspended_bdev->bd_inode->i_mutex); |
e39e2e95 | 1107 | i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1b1dcc1b | 1108 | mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex); |
1da177e4 LT |
1109 | } |
1110 | ||
1111 | static int __bind(struct mapped_device *md, struct dm_table *t) | |
1112 | { | |
165125e1 | 1113 | struct request_queue *q = md->queue; |
1da177e4 LT |
1114 | sector_t size; |
1115 | ||
1116 | size = dm_table_get_size(t); | |
3ac51e74 DW |
1117 | |
1118 | /* | |
1119 | * Wipe any geometry if the size of the table changed. | |
1120 | */ | |
1121 | if (size != get_capacity(md->disk)) | |
1122 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
1123 | ||
bfa152fa JN |
1124 | if (md->suspended_bdev) |
1125 | __set_size(md, size); | |
1da177e4 LT |
1126 | if (size == 0) |
1127 | return 0; | |
1128 | ||
2ca3310e AK |
1129 | dm_table_get(t); |
1130 | dm_table_event_callback(t, event_callback, md); | |
1131 | ||
1da177e4 LT |
1132 | write_lock(&md->map_lock); |
1133 | md->map = t; | |
2ca3310e | 1134 | dm_table_set_restrictions(t, q); |
1da177e4 LT |
1135 | write_unlock(&md->map_lock); |
1136 | ||
1da177e4 LT |
1137 | return 0; |
1138 | } | |
1139 | ||
1140 | static void __unbind(struct mapped_device *md) | |
1141 | { | |
1142 | struct dm_table *map = md->map; | |
1143 | ||
1144 | if (!map) | |
1145 | return; | |
1146 | ||
1147 | dm_table_event_callback(map, NULL, NULL); | |
1148 | write_lock(&md->map_lock); | |
1149 | md->map = NULL; | |
1150 | write_unlock(&md->map_lock); | |
1151 | dm_table_put(map); | |
1152 | } | |
1153 | ||
1154 | /* | |
1155 | * Constructor for a new device. | |
1156 | */ | |
2b06cfff | 1157 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
1158 | { |
1159 | struct mapped_device *md; | |
1160 | ||
2b06cfff | 1161 | md = alloc_dev(minor); |
1da177e4 LT |
1162 | if (!md) |
1163 | return -ENXIO; | |
1164 | ||
1165 | *result = md; | |
1166 | return 0; | |
1167 | } | |
1168 | ||
637842cf | 1169 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
1170 | { |
1171 | struct mapped_device *md; | |
1da177e4 LT |
1172 | unsigned minor = MINOR(dev); |
1173 | ||
1174 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
1175 | return NULL; | |
1176 | ||
f32c10b0 | 1177 | spin_lock(&_minor_lock); |
1da177e4 LT |
1178 | |
1179 | md = idr_find(&_minor_idr, minor); | |
fba9f90e JM |
1180 | if (md && (md == MINOR_ALLOCED || |
1181 | (dm_disk(md)->first_minor != minor) || | |
17b2f66f | 1182 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 1183 | md = NULL; |
fba9f90e JM |
1184 | goto out; |
1185 | } | |
1da177e4 | 1186 | |
fba9f90e | 1187 | out: |
f32c10b0 | 1188 | spin_unlock(&_minor_lock); |
1da177e4 | 1189 | |
637842cf DT |
1190 | return md; |
1191 | } | |
1192 | ||
d229a958 DT |
1193 | struct mapped_device *dm_get_md(dev_t dev) |
1194 | { | |
1195 | struct mapped_device *md = dm_find_md(dev); | |
1196 | ||
1197 | if (md) | |
1198 | dm_get(md); | |
1199 | ||
1200 | return md; | |
1201 | } | |
1202 | ||
9ade92a9 | 1203 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 1204 | { |
9ade92a9 | 1205 | return md->interface_ptr; |
1da177e4 LT |
1206 | } |
1207 | ||
1208 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
1209 | { | |
1210 | md->interface_ptr = ptr; | |
1211 | } | |
1212 | ||
1213 | void dm_get(struct mapped_device *md) | |
1214 | { | |
1215 | atomic_inc(&md->holders); | |
1216 | } | |
1217 | ||
72d94861 AK |
1218 | const char *dm_device_name(struct mapped_device *md) |
1219 | { | |
1220 | return md->name; | |
1221 | } | |
1222 | EXPORT_SYMBOL_GPL(dm_device_name); | |
1223 | ||
1da177e4 LT |
1224 | void dm_put(struct mapped_device *md) |
1225 | { | |
1134e5ae | 1226 | struct dm_table *map; |
1da177e4 | 1227 | |
fba9f90e JM |
1228 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1229 | ||
f32c10b0 | 1230 | if (atomic_dec_and_lock(&md->holders, &_minor_lock)) { |
1134e5ae | 1231 | map = dm_get_table(md); |
ba61fdd1 | 1232 | idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor); |
fba9f90e | 1233 | set_bit(DMF_FREEING, &md->flags); |
f32c10b0 | 1234 | spin_unlock(&_minor_lock); |
cf222b37 | 1235 | if (!dm_suspended(md)) { |
1da177e4 LT |
1236 | dm_table_presuspend_targets(map); |
1237 | dm_table_postsuspend_targets(map); | |
1238 | } | |
1239 | __unbind(md); | |
1134e5ae | 1240 | dm_table_put(map); |
1da177e4 LT |
1241 | free_dev(md); |
1242 | } | |
1da177e4 | 1243 | } |
79eb885c | 1244 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 LT |
1245 | |
1246 | /* | |
1247 | * Process the deferred bios | |
1248 | */ | |
1249 | static void __flush_deferred_io(struct mapped_device *md, struct bio *c) | |
1250 | { | |
1251 | struct bio *n; | |
1252 | ||
1253 | while (c) { | |
1254 | n = c->bi_next; | |
1255 | c->bi_next = NULL; | |
1256 | __split_bio(md, c); | |
1257 | c = n; | |
1258 | } | |
1259 | } | |
1260 | ||
1261 | /* | |
1262 | * Swap in a new table (destroying old one). | |
1263 | */ | |
1264 | int dm_swap_table(struct mapped_device *md, struct dm_table *table) | |
1265 | { | |
93c534ae | 1266 | int r = -EINVAL; |
1da177e4 | 1267 | |
2ca3310e | 1268 | down(&md->suspend_lock); |
1da177e4 LT |
1269 | |
1270 | /* device must be suspended */ | |
cf222b37 | 1271 | if (!dm_suspended(md)) |
93c534ae | 1272 | goto out; |
1da177e4 | 1273 | |
bfa152fa JN |
1274 | /* without bdev, the device size cannot be changed */ |
1275 | if (!md->suspended_bdev) | |
1276 | if (get_capacity(md->disk) != dm_table_get_size(table)) | |
1277 | goto out; | |
1278 | ||
1da177e4 LT |
1279 | __unbind(md); |
1280 | r = __bind(md, table); | |
1da177e4 | 1281 | |
93c534ae | 1282 | out: |
2ca3310e | 1283 | up(&md->suspend_lock); |
93c534ae | 1284 | return r; |
1da177e4 LT |
1285 | } |
1286 | ||
1287 | /* | |
1288 | * Functions to lock and unlock any filesystem running on the | |
1289 | * device. | |
1290 | */ | |
2ca3310e | 1291 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 1292 | { |
e39e2e95 | 1293 | int r; |
1da177e4 LT |
1294 | |
1295 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 1296 | |
e39e2e95 | 1297 | md->frozen_sb = freeze_bdev(md->suspended_bdev); |
dfbe03f6 | 1298 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 1299 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
1300 | md->frozen_sb = NULL; |
1301 | return r; | |
dfbe03f6 AK |
1302 | } |
1303 | ||
aa8d7c2f AK |
1304 | set_bit(DMF_FROZEN, &md->flags); |
1305 | ||
1da177e4 | 1306 | /* don't bdput right now, we don't want the bdev |
e39e2e95 | 1307 | * to go away while it is locked. |
1da177e4 LT |
1308 | */ |
1309 | return 0; | |
1310 | } | |
1311 | ||
2ca3310e | 1312 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 1313 | { |
aa8d7c2f AK |
1314 | if (!test_bit(DMF_FROZEN, &md->flags)) |
1315 | return; | |
1316 | ||
e39e2e95 | 1317 | thaw_bdev(md->suspended_bdev, md->frozen_sb); |
1da177e4 | 1318 | md->frozen_sb = NULL; |
aa8d7c2f | 1319 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
1320 | } |
1321 | ||
1322 | /* | |
1323 | * We need to be able to change a mapping table under a mounted | |
1324 | * filesystem. For example we might want to move some data in | |
1325 | * the background. Before the table can be swapped with | |
1326 | * dm_bind_table, dm_suspend must be called to flush any in | |
1327 | * flight bios and ensure that any further io gets deferred. | |
1328 | */ | |
a3d77d35 | 1329 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 1330 | { |
2ca3310e | 1331 | struct dm_table *map = NULL; |
2e93ccc1 | 1332 | unsigned long flags; |
1da177e4 | 1333 | DECLARE_WAITQUEUE(wait, current); |
1ecac7fd | 1334 | struct bio *def; |
cf222b37 | 1335 | int r = -EINVAL; |
a3d77d35 | 1336 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 1337 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 1338 | |
2ca3310e AK |
1339 | down(&md->suspend_lock); |
1340 | ||
1341 | if (dm_suspended(md)) | |
d287483d | 1342 | goto out_unlock; |
1da177e4 LT |
1343 | |
1344 | map = dm_get_table(md); | |
1da177e4 | 1345 | |
2e93ccc1 KU |
1346 | /* |
1347 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
1348 | * This flag is cleared before dm_suspend returns. | |
1349 | */ | |
1350 | if (noflush) | |
1351 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1352 | ||
cf222b37 AK |
1353 | /* This does not get reverted if there's an error later. */ |
1354 | dm_table_presuspend_targets(map); | |
1355 | ||
bfa152fa JN |
1356 | /* bdget() can stall if the pending I/Os are not flushed */ |
1357 | if (!noflush) { | |
1358 | md->suspended_bdev = bdget_disk(md->disk, 0); | |
1359 | if (!md->suspended_bdev) { | |
1360 | DMWARN("bdget failed in dm_suspend"); | |
1361 | r = -ENOMEM; | |
1362 | goto flush_and_out; | |
1363 | } | |
e39e2e95 AK |
1364 | } |
1365 | ||
2e93ccc1 KU |
1366 | /* |
1367 | * Flush I/O to the device. | |
1368 | * noflush supersedes do_lockfs, because lock_fs() needs to flush I/Os. | |
1369 | */ | |
1370 | if (do_lockfs && !noflush) { | |
aa8d7c2f AK |
1371 | r = lock_fs(md); |
1372 | if (r) | |
1373 | goto out; | |
1374 | } | |
1da177e4 LT |
1375 | |
1376 | /* | |
354e0071 | 1377 | * First we set the BLOCK_IO flag so no more ios will be mapped. |
1da177e4 | 1378 | */ |
2ca3310e AK |
1379 | down_write(&md->io_lock); |
1380 | set_bit(DMF_BLOCK_IO, &md->flags); | |
1da177e4 | 1381 | |
1da177e4 | 1382 | add_wait_queue(&md->wait, &wait); |
2ca3310e | 1383 | up_write(&md->io_lock); |
1da177e4 LT |
1384 | |
1385 | /* unplug */ | |
2ca3310e | 1386 | if (map) |
1da177e4 | 1387 | dm_table_unplug_all(map); |
1da177e4 LT |
1388 | |
1389 | /* | |
1390 | * Then we wait for the already mapped ios to | |
1391 | * complete. | |
1392 | */ | |
1393 | while (1) { | |
1394 | set_current_state(TASK_INTERRUPTIBLE); | |
1395 | ||
1396 | if (!atomic_read(&md->pending) || signal_pending(current)) | |
1397 | break; | |
1398 | ||
1399 | io_schedule(); | |
1400 | } | |
1401 | set_current_state(TASK_RUNNING); | |
1402 | ||
2ca3310e | 1403 | down_write(&md->io_lock); |
1da177e4 LT |
1404 | remove_wait_queue(&md->wait, &wait); |
1405 | ||
2e93ccc1 KU |
1406 | if (noflush) { |
1407 | spin_lock_irqsave(&md->pushback_lock, flags); | |
1408 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1409 | bio_list_merge_head(&md->deferred, &md->pushback); | |
1410 | bio_list_init(&md->pushback); | |
1411 | spin_unlock_irqrestore(&md->pushback_lock, flags); | |
1412 | } | |
1413 | ||
1da177e4 | 1414 | /* were we interrupted ? */ |
cf222b37 | 1415 | r = -EINTR; |
2ca3310e | 1416 | if (atomic_read(&md->pending)) { |
1ecac7fd JN |
1417 | clear_bit(DMF_BLOCK_IO, &md->flags); |
1418 | def = bio_list_get(&md->deferred); | |
1419 | __flush_deferred_io(md, def); | |
2ca3310e AK |
1420 | up_write(&md->io_lock); |
1421 | unlock_fs(md); | |
2e93ccc1 | 1422 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e AK |
1423 | } |
1424 | up_write(&md->io_lock); | |
1da177e4 | 1425 | |
cf222b37 | 1426 | dm_table_postsuspend_targets(map); |
1da177e4 | 1427 | |
2ca3310e | 1428 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 1429 | |
2ca3310e | 1430 | r = 0; |
b84b0287 | 1431 | |
2e93ccc1 KU |
1432 | flush_and_out: |
1433 | if (r && noflush) { | |
1434 | /* | |
1435 | * Because there may be already I/Os in the pushback list, | |
1436 | * flush them before return. | |
1437 | */ | |
1438 | down_write(&md->io_lock); | |
1439 | ||
1440 | spin_lock_irqsave(&md->pushback_lock, flags); | |
1441 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1442 | bio_list_merge_head(&md->deferred, &md->pushback); | |
1443 | bio_list_init(&md->pushback); | |
1444 | spin_unlock_irqrestore(&md->pushback_lock, flags); | |
1445 | ||
1446 | def = bio_list_get(&md->deferred); | |
1447 | __flush_deferred_io(md, def); | |
1448 | up_write(&md->io_lock); | |
1449 | } | |
1450 | ||
2ca3310e | 1451 | out: |
e39e2e95 AK |
1452 | if (r && md->suspended_bdev) { |
1453 | bdput(md->suspended_bdev); | |
1454 | md->suspended_bdev = NULL; | |
1455 | } | |
1456 | ||
2ca3310e | 1457 | dm_table_put(map); |
d287483d AK |
1458 | |
1459 | out_unlock: | |
2ca3310e | 1460 | up(&md->suspend_lock); |
cf222b37 | 1461 | return r; |
1da177e4 LT |
1462 | } |
1463 | ||
1464 | int dm_resume(struct mapped_device *md) | |
1465 | { | |
cf222b37 | 1466 | int r = -EINVAL; |
1da177e4 | 1467 | struct bio *def; |
cf222b37 | 1468 | struct dm_table *map = NULL; |
1da177e4 | 1469 | |
2ca3310e AK |
1470 | down(&md->suspend_lock); |
1471 | if (!dm_suspended(md)) | |
cf222b37 | 1472 | goto out; |
cf222b37 AK |
1473 | |
1474 | map = dm_get_table(md); | |
2ca3310e | 1475 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 1476 | goto out; |
1da177e4 | 1477 | |
8757b776 MB |
1478 | r = dm_table_resume_targets(map); |
1479 | if (r) | |
1480 | goto out; | |
2ca3310e AK |
1481 | |
1482 | down_write(&md->io_lock); | |
1da177e4 LT |
1483 | clear_bit(DMF_BLOCK_IO, &md->flags); |
1484 | ||
1485 | def = bio_list_get(&md->deferred); | |
1486 | __flush_deferred_io(md, def); | |
2ca3310e AK |
1487 | up_write(&md->io_lock); |
1488 | ||
1489 | unlock_fs(md); | |
1490 | ||
bfa152fa JN |
1491 | if (md->suspended_bdev) { |
1492 | bdput(md->suspended_bdev); | |
1493 | md->suspended_bdev = NULL; | |
1494 | } | |
e39e2e95 | 1495 | |
2ca3310e AK |
1496 | clear_bit(DMF_SUSPENDED, &md->flags); |
1497 | ||
1da177e4 | 1498 | dm_table_unplug_all(map); |
1da177e4 | 1499 | |
8560ed6f HR |
1500 | kobject_uevent(&md->disk->kobj, KOBJ_CHANGE); |
1501 | ||
cf222b37 | 1502 | r = 0; |
2ca3310e | 1503 | |
cf222b37 AK |
1504 | out: |
1505 | dm_table_put(map); | |
2ca3310e AK |
1506 | up(&md->suspend_lock); |
1507 | ||
cf222b37 | 1508 | return r; |
1da177e4 LT |
1509 | } |
1510 | ||
1511 | /*----------------------------------------------------------------- | |
1512 | * Event notification. | |
1513 | *---------------------------------------------------------------*/ | |
1514 | uint32_t dm_get_event_nr(struct mapped_device *md) | |
1515 | { | |
1516 | return atomic_read(&md->event_nr); | |
1517 | } | |
1518 | ||
1519 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
1520 | { | |
1521 | return wait_event_interruptible(md->eventq, | |
1522 | (event_nr != atomic_read(&md->event_nr))); | |
1523 | } | |
1524 | ||
1525 | /* | |
1526 | * The gendisk is only valid as long as you have a reference | |
1527 | * count on 'md'. | |
1528 | */ | |
1529 | struct gendisk *dm_disk(struct mapped_device *md) | |
1530 | { | |
1531 | return md->disk; | |
1532 | } | |
1533 | ||
1534 | int dm_suspended(struct mapped_device *md) | |
1535 | { | |
1536 | return test_bit(DMF_SUSPENDED, &md->flags); | |
1537 | } | |
1538 | ||
2e93ccc1 KU |
1539 | int dm_noflush_suspending(struct dm_target *ti) |
1540 | { | |
1541 | struct mapped_device *md = dm_table_get_md(ti->table); | |
1542 | int r = __noflush_suspending(md); | |
1543 | ||
1544 | dm_put(md); | |
1545 | ||
1546 | return r; | |
1547 | } | |
1548 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
1549 | ||
1da177e4 LT |
1550 | static struct block_device_operations dm_blk_dops = { |
1551 | .open = dm_blk_open, | |
1552 | .release = dm_blk_close, | |
aa129a22 | 1553 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 1554 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
1555 | .owner = THIS_MODULE |
1556 | }; | |
1557 | ||
1558 | EXPORT_SYMBOL(dm_get_mapinfo); | |
1559 | ||
1560 | /* | |
1561 | * module hooks | |
1562 | */ | |
1563 | module_init(dm_init); | |
1564 | module_exit(dm_exit); | |
1565 | ||
1566 | module_param(major, uint, 0); | |
1567 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
1568 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
1569 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
1570 | MODULE_LICENSE("GPL"); |