]>
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 | ||
6712ecf8 | 487 | bio_endio(io->bio, io->error); |
2e93ccc1 | 488 | } |
2056a782 | 489 | |
1da177e4 LT |
490 | free_io(io->md, io); |
491 | } | |
492 | } | |
493 | ||
6712ecf8 | 494 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
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 | ||
1da177e4 LT |
501 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
502 | error = -EIO; | |
503 | ||
504 | if (endio) { | |
505 | r = endio(tio->ti, bio, error, &tio->info); | |
2e93ccc1 KU |
506 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
507 | /* | |
508 | * error and requeue request are handled | |
509 | * in dec_pending(). | |
510 | */ | |
1da177e4 | 511 | error = r; |
45cbcd79 KU |
512 | else if (r == DM_ENDIO_INCOMPLETE) |
513 | /* The target will handle the io */ | |
6712ecf8 | 514 | return; |
45cbcd79 KU |
515 | else if (r) { |
516 | DMWARN("unimplemented target endio return value: %d", r); | |
517 | BUG(); | |
518 | } | |
1da177e4 LT |
519 | } |
520 | ||
9faf400f SB |
521 | dec_pending(tio->io, error); |
522 | ||
523 | /* | |
524 | * Store md for cleanup instead of tio which is about to get freed. | |
525 | */ | |
526 | bio->bi_private = md->bs; | |
527 | ||
1da177e4 | 528 | bio_put(bio); |
9faf400f | 529 | free_tio(md, tio); |
1da177e4 LT |
530 | } |
531 | ||
532 | static sector_t max_io_len(struct mapped_device *md, | |
533 | sector_t sector, struct dm_target *ti) | |
534 | { | |
535 | sector_t offset = sector - ti->begin; | |
536 | sector_t len = ti->len - offset; | |
537 | ||
538 | /* | |
539 | * Does the target need to split even further ? | |
540 | */ | |
541 | if (ti->split_io) { | |
542 | sector_t boundary; | |
543 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) | |
544 | - offset; | |
545 | if (len > boundary) | |
546 | len = boundary; | |
547 | } | |
548 | ||
549 | return len; | |
550 | } | |
551 | ||
552 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
028867ac | 553 | struct dm_target_io *tio) |
1da177e4 LT |
554 | { |
555 | int r; | |
2056a782 | 556 | sector_t sector; |
9faf400f | 557 | struct mapped_device *md; |
1da177e4 LT |
558 | |
559 | /* | |
560 | * Sanity checks. | |
561 | */ | |
562 | BUG_ON(!clone->bi_size); | |
563 | ||
564 | clone->bi_end_io = clone_endio; | |
565 | clone->bi_private = tio; | |
566 | ||
567 | /* | |
568 | * Map the clone. If r == 0 we don't need to do | |
569 | * anything, the target has assumed ownership of | |
570 | * this io. | |
571 | */ | |
572 | atomic_inc(&tio->io->io_count); | |
2056a782 | 573 | sector = clone->bi_sector; |
1da177e4 | 574 | r = ti->type->map(ti, clone, &tio->info); |
45cbcd79 | 575 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 576 | /* the bio has been remapped so dispatch it */ |
2056a782 | 577 | |
17b2f66f | 578 | blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone, |
c7149d6b AB |
579 | tio->io->bio->bi_bdev->bd_dev, |
580 | clone->bi_sector, sector); | |
2056a782 | 581 | |
1da177e4 | 582 | generic_make_request(clone); |
2e93ccc1 KU |
583 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
584 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
585 | md = tio->io->md; |
586 | dec_pending(tio->io, r); | |
587 | /* | |
588 | * Store bio_set for cleanup. | |
589 | */ | |
590 | clone->bi_private = md->bs; | |
1da177e4 | 591 | bio_put(clone); |
9faf400f | 592 | free_tio(md, tio); |
45cbcd79 KU |
593 | } else if (r) { |
594 | DMWARN("unimplemented target map return value: %d", r); | |
595 | BUG(); | |
1da177e4 LT |
596 | } |
597 | } | |
598 | ||
599 | struct clone_info { | |
600 | struct mapped_device *md; | |
601 | struct dm_table *map; | |
602 | struct bio *bio; | |
603 | struct dm_io *io; | |
604 | sector_t sector; | |
605 | sector_t sector_count; | |
606 | unsigned short idx; | |
607 | }; | |
608 | ||
3676347a PO |
609 | static void dm_bio_destructor(struct bio *bio) |
610 | { | |
9faf400f SB |
611 | struct bio_set *bs = bio->bi_private; |
612 | ||
613 | bio_free(bio, bs); | |
3676347a PO |
614 | } |
615 | ||
1da177e4 LT |
616 | /* |
617 | * Creates a little bio that is just does part of a bvec. | |
618 | */ | |
619 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
620 | unsigned short idx, unsigned int offset, | |
9faf400f | 621 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
622 | { |
623 | struct bio *clone; | |
624 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
625 | ||
9faf400f | 626 | clone = bio_alloc_bioset(GFP_NOIO, 1, bs); |
3676347a | 627 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
628 | *clone->bi_io_vec = *bv; |
629 | ||
630 | clone->bi_sector = sector; | |
631 | clone->bi_bdev = bio->bi_bdev; | |
632 | clone->bi_rw = bio->bi_rw; | |
633 | clone->bi_vcnt = 1; | |
634 | clone->bi_size = to_bytes(len); | |
635 | clone->bi_io_vec->bv_offset = offset; | |
636 | clone->bi_io_vec->bv_len = clone->bi_size; | |
637 | ||
638 | return clone; | |
639 | } | |
640 | ||
641 | /* | |
642 | * Creates a bio that consists of range of complete bvecs. | |
643 | */ | |
644 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
645 | unsigned short idx, unsigned short bv_count, | |
9faf400f | 646 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
647 | { |
648 | struct bio *clone; | |
649 | ||
9faf400f SB |
650 | clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); |
651 | __bio_clone(clone, bio); | |
652 | clone->bi_destructor = dm_bio_destructor; | |
1da177e4 LT |
653 | clone->bi_sector = sector; |
654 | clone->bi_idx = idx; | |
655 | clone->bi_vcnt = idx + bv_count; | |
656 | clone->bi_size = to_bytes(len); | |
657 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
658 | ||
659 | return clone; | |
660 | } | |
661 | ||
662 | static void __clone_and_map(struct clone_info *ci) | |
663 | { | |
664 | struct bio *clone, *bio = ci->bio; | |
665 | struct dm_target *ti = dm_table_find_target(ci->map, ci->sector); | |
666 | sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti); | |
028867ac | 667 | struct dm_target_io *tio; |
1da177e4 LT |
668 | |
669 | /* | |
670 | * Allocate a target io object. | |
671 | */ | |
672 | tio = alloc_tio(ci->md); | |
673 | tio->io = ci->io; | |
674 | tio->ti = ti; | |
675 | memset(&tio->info, 0, sizeof(tio->info)); | |
676 | ||
677 | if (ci->sector_count <= max) { | |
678 | /* | |
679 | * Optimise for the simple case where we can do all of | |
680 | * the remaining io with a single clone. | |
681 | */ | |
682 | clone = clone_bio(bio, ci->sector, ci->idx, | |
9faf400f SB |
683 | bio->bi_vcnt - ci->idx, ci->sector_count, |
684 | ci->md->bs); | |
1da177e4 LT |
685 | __map_bio(ti, clone, tio); |
686 | ci->sector_count = 0; | |
687 | ||
688 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
689 | /* | |
690 | * There are some bvecs that don't span targets. | |
691 | * Do as many of these as possible. | |
692 | */ | |
693 | int i; | |
694 | sector_t remaining = max; | |
695 | sector_t bv_len; | |
696 | ||
697 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
698 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
699 | ||
700 | if (bv_len > remaining) | |
701 | break; | |
702 | ||
703 | remaining -= bv_len; | |
704 | len += bv_len; | |
705 | } | |
706 | ||
9faf400f SB |
707 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, |
708 | ci->md->bs); | |
1da177e4 LT |
709 | __map_bio(ti, clone, tio); |
710 | ||
711 | ci->sector += len; | |
712 | ci->sector_count -= len; | |
713 | ci->idx = i; | |
714 | ||
715 | } else { | |
716 | /* | |
d2044a94 | 717 | * Handle a bvec that must be split between two or more targets. |
1da177e4 LT |
718 | */ |
719 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
d2044a94 AK |
720 | sector_t remaining = to_sector(bv->bv_len); |
721 | unsigned int offset = 0; | |
1da177e4 | 722 | |
d2044a94 AK |
723 | do { |
724 | if (offset) { | |
725 | ti = dm_table_find_target(ci->map, ci->sector); | |
726 | max = max_io_len(ci->md, ci->sector, ti); | |
1da177e4 | 727 | |
d2044a94 AK |
728 | tio = alloc_tio(ci->md); |
729 | tio->io = ci->io; | |
730 | tio->ti = ti; | |
731 | memset(&tio->info, 0, sizeof(tio->info)); | |
732 | } | |
733 | ||
734 | len = min(remaining, max); | |
735 | ||
736 | clone = split_bvec(bio, ci->sector, ci->idx, | |
9faf400f SB |
737 | bv->bv_offset + offset, len, |
738 | ci->md->bs); | |
d2044a94 AK |
739 | |
740 | __map_bio(ti, clone, tio); | |
741 | ||
742 | ci->sector += len; | |
743 | ci->sector_count -= len; | |
744 | offset += to_bytes(len); | |
745 | } while (remaining -= len); | |
1da177e4 | 746 | |
1da177e4 LT |
747 | ci->idx++; |
748 | } | |
749 | } | |
750 | ||
751 | /* | |
752 | * Split the bio into several clones. | |
753 | */ | |
754 | static void __split_bio(struct mapped_device *md, struct bio *bio) | |
755 | { | |
756 | struct clone_info ci; | |
757 | ||
758 | ci.map = dm_get_table(md); | |
759 | if (!ci.map) { | |
6712ecf8 | 760 | bio_io_error(bio); |
1da177e4 LT |
761 | return; |
762 | } | |
763 | ||
764 | ci.md = md; | |
765 | ci.bio = bio; | |
766 | ci.io = alloc_io(md); | |
767 | ci.io->error = 0; | |
768 | atomic_set(&ci.io->io_count, 1); | |
769 | ci.io->bio = bio; | |
770 | ci.io->md = md; | |
771 | ci.sector = bio->bi_sector; | |
772 | ci.sector_count = bio_sectors(bio); | |
773 | ci.idx = bio->bi_idx; | |
774 | ||
3eaf840e | 775 | start_io_acct(ci.io); |
1da177e4 LT |
776 | while (ci.sector_count) |
777 | __clone_and_map(&ci); | |
778 | ||
779 | /* drop the extra reference count */ | |
780 | dec_pending(ci.io, 0); | |
781 | dm_table_put(ci.map); | |
782 | } | |
783 | /*----------------------------------------------------------------- | |
784 | * CRUD END | |
785 | *---------------------------------------------------------------*/ | |
786 | ||
787 | /* | |
788 | * The request function that just remaps the bio built up by | |
789 | * dm_merge_bvec. | |
790 | */ | |
165125e1 | 791 | static int dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 LT |
792 | { |
793 | int r; | |
12f03a49 | 794 | int rw = bio_data_dir(bio); |
1da177e4 LT |
795 | struct mapped_device *md = q->queuedata; |
796 | ||
07a83c47 SB |
797 | /* |
798 | * There is no use in forwarding any barrier request since we can't | |
799 | * guarantee it is (or can be) handled by the targets correctly. | |
800 | */ | |
801 | if (unlikely(bio_barrier(bio))) { | |
6712ecf8 | 802 | bio_endio(bio, -EOPNOTSUPP); |
07a83c47 SB |
803 | return 0; |
804 | } | |
805 | ||
2ca3310e | 806 | down_read(&md->io_lock); |
1da177e4 | 807 | |
12f03a49 KC |
808 | disk_stat_inc(dm_disk(md), ios[rw]); |
809 | disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio)); | |
810 | ||
1da177e4 LT |
811 | /* |
812 | * If we're suspended we have to queue | |
813 | * this io for later. | |
814 | */ | |
815 | while (test_bit(DMF_BLOCK_IO, &md->flags)) { | |
2ca3310e | 816 | up_read(&md->io_lock); |
1da177e4 LT |
817 | |
818 | if (bio_rw(bio) == READA) { | |
6712ecf8 | 819 | bio_io_error(bio); |
1da177e4 LT |
820 | return 0; |
821 | } | |
822 | ||
823 | r = queue_io(md, bio); | |
824 | if (r < 0) { | |
6712ecf8 | 825 | bio_io_error(bio); |
1da177e4 LT |
826 | return 0; |
827 | ||
828 | } else if (r == 0) | |
829 | return 0; /* deferred successfully */ | |
830 | ||
831 | /* | |
832 | * We're in a while loop, because someone could suspend | |
833 | * before we get to the following read lock. | |
834 | */ | |
2ca3310e | 835 | down_read(&md->io_lock); |
1da177e4 LT |
836 | } |
837 | ||
838 | __split_bio(md, bio); | |
2ca3310e | 839 | up_read(&md->io_lock); |
1da177e4 LT |
840 | return 0; |
841 | } | |
842 | ||
165125e1 | 843 | static int dm_flush_all(struct request_queue *q, struct gendisk *disk, |
1da177e4 LT |
844 | sector_t *error_sector) |
845 | { | |
846 | struct mapped_device *md = q->queuedata; | |
847 | struct dm_table *map = dm_get_table(md); | |
848 | int ret = -ENXIO; | |
849 | ||
850 | if (map) { | |
cf222b37 | 851 | ret = dm_table_flush_all(map); |
1da177e4 LT |
852 | dm_table_put(map); |
853 | } | |
854 | ||
855 | return ret; | |
856 | } | |
857 | ||
165125e1 | 858 | static void dm_unplug_all(struct request_queue *q) |
1da177e4 LT |
859 | { |
860 | struct mapped_device *md = q->queuedata; | |
861 | struct dm_table *map = dm_get_table(md); | |
862 | ||
863 | if (map) { | |
864 | dm_table_unplug_all(map); | |
865 | dm_table_put(map); | |
866 | } | |
867 | } | |
868 | ||
869 | static int dm_any_congested(void *congested_data, int bdi_bits) | |
870 | { | |
871 | int r; | |
872 | struct mapped_device *md = (struct mapped_device *) congested_data; | |
873 | struct dm_table *map = dm_get_table(md); | |
874 | ||
875 | if (!map || test_bit(DMF_BLOCK_IO, &md->flags)) | |
876 | r = bdi_bits; | |
877 | else | |
878 | r = dm_table_any_congested(map, bdi_bits); | |
879 | ||
880 | dm_table_put(map); | |
881 | return r; | |
882 | } | |
883 | ||
884 | /*----------------------------------------------------------------- | |
885 | * An IDR is used to keep track of allocated minor numbers. | |
886 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
887 | static DEFINE_IDR(_minor_idr); |
888 | ||
2b06cfff | 889 | static void free_minor(int minor) |
1da177e4 | 890 | { |
f32c10b0 | 891 | spin_lock(&_minor_lock); |
1da177e4 | 892 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 893 | spin_unlock(&_minor_lock); |
1da177e4 LT |
894 | } |
895 | ||
896 | /* | |
897 | * See if the device with a specific minor # is free. | |
898 | */ | |
2b06cfff | 899 | static int specific_minor(struct mapped_device *md, int minor) |
1da177e4 LT |
900 | { |
901 | int r, m; | |
902 | ||
903 | if (minor >= (1 << MINORBITS)) | |
904 | return -EINVAL; | |
905 | ||
62f75c2f JM |
906 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
907 | if (!r) | |
908 | return -ENOMEM; | |
909 | ||
f32c10b0 | 910 | spin_lock(&_minor_lock); |
1da177e4 LT |
911 | |
912 | if (idr_find(&_minor_idr, minor)) { | |
913 | r = -EBUSY; | |
914 | goto out; | |
915 | } | |
916 | ||
ba61fdd1 | 917 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); |
62f75c2f | 918 | if (r) |
1da177e4 | 919 | goto out; |
1da177e4 LT |
920 | |
921 | if (m != minor) { | |
922 | idr_remove(&_minor_idr, m); | |
923 | r = -EBUSY; | |
924 | goto out; | |
925 | } | |
926 | ||
927 | out: | |
f32c10b0 | 928 | spin_unlock(&_minor_lock); |
1da177e4 LT |
929 | return r; |
930 | } | |
931 | ||
2b06cfff | 932 | static int next_free_minor(struct mapped_device *md, int *minor) |
1da177e4 | 933 | { |
2b06cfff | 934 | int r, m; |
1da177e4 | 935 | |
1da177e4 | 936 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
62f75c2f JM |
937 | if (!r) |
938 | return -ENOMEM; | |
939 | ||
f32c10b0 | 940 | spin_lock(&_minor_lock); |
1da177e4 | 941 | |
ba61fdd1 | 942 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); |
1da177e4 LT |
943 | if (r) { |
944 | goto out; | |
945 | } | |
946 | ||
947 | if (m >= (1 << MINORBITS)) { | |
948 | idr_remove(&_minor_idr, m); | |
949 | r = -ENOSPC; | |
950 | goto out; | |
951 | } | |
952 | ||
953 | *minor = m; | |
954 | ||
955 | out: | |
f32c10b0 | 956 | spin_unlock(&_minor_lock); |
1da177e4 LT |
957 | return r; |
958 | } | |
959 | ||
960 | static struct block_device_operations dm_blk_dops; | |
961 | ||
962 | /* | |
963 | * Allocate and initialise a blank device with a given minor. | |
964 | */ | |
2b06cfff | 965 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
966 | { |
967 | int r; | |
968 | struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL); | |
ba61fdd1 | 969 | void *old_md; |
1da177e4 LT |
970 | |
971 | if (!md) { | |
972 | DMWARN("unable to allocate device, out of memory."); | |
973 | return NULL; | |
974 | } | |
975 | ||
10da4f79 JM |
976 | if (!try_module_get(THIS_MODULE)) |
977 | goto bad0; | |
978 | ||
1da177e4 | 979 | /* get a minor number for the dev */ |
2b06cfff AK |
980 | if (minor == DM_ANY_MINOR) |
981 | r = next_free_minor(md, &minor); | |
982 | else | |
983 | r = specific_minor(md, minor); | |
1da177e4 LT |
984 | if (r < 0) |
985 | goto bad1; | |
986 | ||
987 | memset(md, 0, sizeof(*md)); | |
2ca3310e AK |
988 | init_rwsem(&md->io_lock); |
989 | init_MUTEX(&md->suspend_lock); | |
2e93ccc1 | 990 | spin_lock_init(&md->pushback_lock); |
1da177e4 LT |
991 | rwlock_init(&md->map_lock); |
992 | atomic_set(&md->holders, 1); | |
5c6bd75d | 993 | atomic_set(&md->open_count, 0); |
1da177e4 LT |
994 | atomic_set(&md->event_nr, 0); |
995 | ||
996 | md->queue = blk_alloc_queue(GFP_KERNEL); | |
997 | if (!md->queue) | |
e3f6ac61 | 998 | goto bad1_free_minor; |
1da177e4 LT |
999 | |
1000 | md->queue->queuedata = md; | |
1001 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1002 | md->queue->backing_dev_info.congested_data = md; | |
1003 | blk_queue_make_request(md->queue, dm_request); | |
daef265f | 1004 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
1da177e4 LT |
1005 | md->queue->unplug_fn = dm_unplug_all; |
1006 | md->queue->issue_flush_fn = dm_flush_all; | |
1007 | ||
93d2341c | 1008 | md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache); |
74859364 KU |
1009 | if (!md->io_pool) |
1010 | goto bad2; | |
1da177e4 | 1011 | |
93d2341c | 1012 | md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache); |
1da177e4 LT |
1013 | if (!md->tio_pool) |
1014 | goto bad3; | |
1015 | ||
5972511b | 1016 | md->bs = bioset_create(16, 16); |
9faf400f SB |
1017 | if (!md->bs) |
1018 | goto bad_no_bioset; | |
1019 | ||
1da177e4 LT |
1020 | md->disk = alloc_disk(1); |
1021 | if (!md->disk) | |
1022 | goto bad4; | |
1023 | ||
f0b04115 JM |
1024 | atomic_set(&md->pending, 0); |
1025 | init_waitqueue_head(&md->wait); | |
1026 | init_waitqueue_head(&md->eventq); | |
1027 | ||
1da177e4 LT |
1028 | md->disk->major = _major; |
1029 | md->disk->first_minor = minor; | |
1030 | md->disk->fops = &dm_blk_dops; | |
1031 | md->disk->queue = md->queue; | |
1032 | md->disk->private_data = md; | |
1033 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1034 | add_disk(md->disk); | |
7e51f257 | 1035 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1036 | |
ba61fdd1 | 1037 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1038 | spin_lock(&_minor_lock); |
ba61fdd1 | 1039 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1040 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1041 | |
1042 | BUG_ON(old_md != MINOR_ALLOCED); | |
1043 | ||
1da177e4 LT |
1044 | return md; |
1045 | ||
1046 | bad4: | |
9faf400f SB |
1047 | bioset_free(md->bs); |
1048 | bad_no_bioset: | |
1da177e4 LT |
1049 | mempool_destroy(md->tio_pool); |
1050 | bad3: | |
1051 | mempool_destroy(md->io_pool); | |
1052 | bad2: | |
1312f40e | 1053 | blk_cleanup_queue(md->queue); |
e3f6ac61 | 1054 | bad1_free_minor: |
1da177e4 LT |
1055 | free_minor(minor); |
1056 | bad1: | |
10da4f79 JM |
1057 | module_put(THIS_MODULE); |
1058 | bad0: | |
1da177e4 LT |
1059 | kfree(md); |
1060 | return NULL; | |
1061 | } | |
1062 | ||
1063 | static void free_dev(struct mapped_device *md) | |
1064 | { | |
2b06cfff | 1065 | int minor = md->disk->first_minor; |
63d94e48 | 1066 | |
d9dde59b JN |
1067 | if (md->suspended_bdev) { |
1068 | thaw_bdev(md->suspended_bdev, NULL); | |
1069 | bdput(md->suspended_bdev); | |
1070 | } | |
1da177e4 LT |
1071 | mempool_destroy(md->tio_pool); |
1072 | mempool_destroy(md->io_pool); | |
9faf400f | 1073 | bioset_free(md->bs); |
1da177e4 | 1074 | del_gendisk(md->disk); |
63d94e48 | 1075 | free_minor(minor); |
fba9f90e JM |
1076 | |
1077 | spin_lock(&_minor_lock); | |
1078 | md->disk->private_data = NULL; | |
1079 | spin_unlock(&_minor_lock); | |
1080 | ||
1da177e4 | 1081 | put_disk(md->disk); |
1312f40e | 1082 | blk_cleanup_queue(md->queue); |
10da4f79 | 1083 | module_put(THIS_MODULE); |
1da177e4 LT |
1084 | kfree(md); |
1085 | } | |
1086 | ||
1087 | /* | |
1088 | * Bind a table to the device. | |
1089 | */ | |
1090 | static void event_callback(void *context) | |
1091 | { | |
1092 | struct mapped_device *md = (struct mapped_device *) context; | |
1093 | ||
1094 | atomic_inc(&md->event_nr); | |
1095 | wake_up(&md->eventq); | |
1096 | } | |
1097 | ||
4e90188b | 1098 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1099 | { |
4e90188b | 1100 | set_capacity(md->disk, size); |
1da177e4 | 1101 | |
1b1dcc1b | 1102 | mutex_lock(&md->suspended_bdev->bd_inode->i_mutex); |
e39e2e95 | 1103 | i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1b1dcc1b | 1104 | mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex); |
1da177e4 LT |
1105 | } |
1106 | ||
1107 | static int __bind(struct mapped_device *md, struct dm_table *t) | |
1108 | { | |
165125e1 | 1109 | struct request_queue *q = md->queue; |
1da177e4 LT |
1110 | sector_t size; |
1111 | ||
1112 | size = dm_table_get_size(t); | |
3ac51e74 DW |
1113 | |
1114 | /* | |
1115 | * Wipe any geometry if the size of the table changed. | |
1116 | */ | |
1117 | if (size != get_capacity(md->disk)) | |
1118 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
1119 | ||
bfa152fa JN |
1120 | if (md->suspended_bdev) |
1121 | __set_size(md, size); | |
1da177e4 LT |
1122 | if (size == 0) |
1123 | return 0; | |
1124 | ||
2ca3310e AK |
1125 | dm_table_get(t); |
1126 | dm_table_event_callback(t, event_callback, md); | |
1127 | ||
1da177e4 LT |
1128 | write_lock(&md->map_lock); |
1129 | md->map = t; | |
2ca3310e | 1130 | dm_table_set_restrictions(t, q); |
1da177e4 LT |
1131 | write_unlock(&md->map_lock); |
1132 | ||
1da177e4 LT |
1133 | return 0; |
1134 | } | |
1135 | ||
1136 | static void __unbind(struct mapped_device *md) | |
1137 | { | |
1138 | struct dm_table *map = md->map; | |
1139 | ||
1140 | if (!map) | |
1141 | return; | |
1142 | ||
1143 | dm_table_event_callback(map, NULL, NULL); | |
1144 | write_lock(&md->map_lock); | |
1145 | md->map = NULL; | |
1146 | write_unlock(&md->map_lock); | |
1147 | dm_table_put(map); | |
1148 | } | |
1149 | ||
1150 | /* | |
1151 | * Constructor for a new device. | |
1152 | */ | |
2b06cfff | 1153 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
1154 | { |
1155 | struct mapped_device *md; | |
1156 | ||
2b06cfff | 1157 | md = alloc_dev(minor); |
1da177e4 LT |
1158 | if (!md) |
1159 | return -ENXIO; | |
1160 | ||
1161 | *result = md; | |
1162 | return 0; | |
1163 | } | |
1164 | ||
637842cf | 1165 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
1166 | { |
1167 | struct mapped_device *md; | |
1da177e4 LT |
1168 | unsigned minor = MINOR(dev); |
1169 | ||
1170 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
1171 | return NULL; | |
1172 | ||
f32c10b0 | 1173 | spin_lock(&_minor_lock); |
1da177e4 LT |
1174 | |
1175 | md = idr_find(&_minor_idr, minor); | |
fba9f90e JM |
1176 | if (md && (md == MINOR_ALLOCED || |
1177 | (dm_disk(md)->first_minor != minor) || | |
17b2f66f | 1178 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 1179 | md = NULL; |
fba9f90e JM |
1180 | goto out; |
1181 | } | |
1da177e4 | 1182 | |
fba9f90e | 1183 | out: |
f32c10b0 | 1184 | spin_unlock(&_minor_lock); |
1da177e4 | 1185 | |
637842cf DT |
1186 | return md; |
1187 | } | |
1188 | ||
d229a958 DT |
1189 | struct mapped_device *dm_get_md(dev_t dev) |
1190 | { | |
1191 | struct mapped_device *md = dm_find_md(dev); | |
1192 | ||
1193 | if (md) | |
1194 | dm_get(md); | |
1195 | ||
1196 | return md; | |
1197 | } | |
1198 | ||
9ade92a9 | 1199 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 1200 | { |
9ade92a9 | 1201 | return md->interface_ptr; |
1da177e4 LT |
1202 | } |
1203 | ||
1204 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
1205 | { | |
1206 | md->interface_ptr = ptr; | |
1207 | } | |
1208 | ||
1209 | void dm_get(struct mapped_device *md) | |
1210 | { | |
1211 | atomic_inc(&md->holders); | |
1212 | } | |
1213 | ||
72d94861 AK |
1214 | const char *dm_device_name(struct mapped_device *md) |
1215 | { | |
1216 | return md->name; | |
1217 | } | |
1218 | EXPORT_SYMBOL_GPL(dm_device_name); | |
1219 | ||
1da177e4 LT |
1220 | void dm_put(struct mapped_device *md) |
1221 | { | |
1134e5ae | 1222 | struct dm_table *map; |
1da177e4 | 1223 | |
fba9f90e JM |
1224 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1225 | ||
f32c10b0 | 1226 | if (atomic_dec_and_lock(&md->holders, &_minor_lock)) { |
1134e5ae | 1227 | map = dm_get_table(md); |
ba61fdd1 | 1228 | idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor); |
fba9f90e | 1229 | set_bit(DMF_FREEING, &md->flags); |
f32c10b0 | 1230 | spin_unlock(&_minor_lock); |
cf222b37 | 1231 | if (!dm_suspended(md)) { |
1da177e4 LT |
1232 | dm_table_presuspend_targets(map); |
1233 | dm_table_postsuspend_targets(map); | |
1234 | } | |
1235 | __unbind(md); | |
1134e5ae | 1236 | dm_table_put(map); |
1da177e4 LT |
1237 | free_dev(md); |
1238 | } | |
1da177e4 | 1239 | } |
79eb885c | 1240 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 LT |
1241 | |
1242 | /* | |
1243 | * Process the deferred bios | |
1244 | */ | |
1245 | static void __flush_deferred_io(struct mapped_device *md, struct bio *c) | |
1246 | { | |
1247 | struct bio *n; | |
1248 | ||
1249 | while (c) { | |
1250 | n = c->bi_next; | |
1251 | c->bi_next = NULL; | |
1252 | __split_bio(md, c); | |
1253 | c = n; | |
1254 | } | |
1255 | } | |
1256 | ||
1257 | /* | |
1258 | * Swap in a new table (destroying old one). | |
1259 | */ | |
1260 | int dm_swap_table(struct mapped_device *md, struct dm_table *table) | |
1261 | { | |
93c534ae | 1262 | int r = -EINVAL; |
1da177e4 | 1263 | |
2ca3310e | 1264 | down(&md->suspend_lock); |
1da177e4 LT |
1265 | |
1266 | /* device must be suspended */ | |
cf222b37 | 1267 | if (!dm_suspended(md)) |
93c534ae | 1268 | goto out; |
1da177e4 | 1269 | |
bfa152fa JN |
1270 | /* without bdev, the device size cannot be changed */ |
1271 | if (!md->suspended_bdev) | |
1272 | if (get_capacity(md->disk) != dm_table_get_size(table)) | |
1273 | goto out; | |
1274 | ||
1da177e4 LT |
1275 | __unbind(md); |
1276 | r = __bind(md, table); | |
1da177e4 | 1277 | |
93c534ae | 1278 | out: |
2ca3310e | 1279 | up(&md->suspend_lock); |
93c534ae | 1280 | return r; |
1da177e4 LT |
1281 | } |
1282 | ||
1283 | /* | |
1284 | * Functions to lock and unlock any filesystem running on the | |
1285 | * device. | |
1286 | */ | |
2ca3310e | 1287 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 1288 | { |
e39e2e95 | 1289 | int r; |
1da177e4 LT |
1290 | |
1291 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 1292 | |
e39e2e95 | 1293 | md->frozen_sb = freeze_bdev(md->suspended_bdev); |
dfbe03f6 | 1294 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 1295 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
1296 | md->frozen_sb = NULL; |
1297 | return r; | |
dfbe03f6 AK |
1298 | } |
1299 | ||
aa8d7c2f AK |
1300 | set_bit(DMF_FROZEN, &md->flags); |
1301 | ||
1da177e4 | 1302 | /* don't bdput right now, we don't want the bdev |
e39e2e95 | 1303 | * to go away while it is locked. |
1da177e4 LT |
1304 | */ |
1305 | return 0; | |
1306 | } | |
1307 | ||
2ca3310e | 1308 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 1309 | { |
aa8d7c2f AK |
1310 | if (!test_bit(DMF_FROZEN, &md->flags)) |
1311 | return; | |
1312 | ||
e39e2e95 | 1313 | thaw_bdev(md->suspended_bdev, md->frozen_sb); |
1da177e4 | 1314 | md->frozen_sb = NULL; |
aa8d7c2f | 1315 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
1316 | } |
1317 | ||
1318 | /* | |
1319 | * We need to be able to change a mapping table under a mounted | |
1320 | * filesystem. For example we might want to move some data in | |
1321 | * the background. Before the table can be swapped with | |
1322 | * dm_bind_table, dm_suspend must be called to flush any in | |
1323 | * flight bios and ensure that any further io gets deferred. | |
1324 | */ | |
a3d77d35 | 1325 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 1326 | { |
2ca3310e | 1327 | struct dm_table *map = NULL; |
2e93ccc1 | 1328 | unsigned long flags; |
1da177e4 | 1329 | DECLARE_WAITQUEUE(wait, current); |
1ecac7fd | 1330 | struct bio *def; |
cf222b37 | 1331 | int r = -EINVAL; |
a3d77d35 | 1332 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 1333 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 1334 | |
2ca3310e AK |
1335 | down(&md->suspend_lock); |
1336 | ||
1337 | if (dm_suspended(md)) | |
d287483d | 1338 | goto out_unlock; |
1da177e4 LT |
1339 | |
1340 | map = dm_get_table(md); | |
1da177e4 | 1341 | |
2e93ccc1 KU |
1342 | /* |
1343 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
1344 | * This flag is cleared before dm_suspend returns. | |
1345 | */ | |
1346 | if (noflush) | |
1347 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1348 | ||
cf222b37 AK |
1349 | /* This does not get reverted if there's an error later. */ |
1350 | dm_table_presuspend_targets(map); | |
1351 | ||
bfa152fa JN |
1352 | /* bdget() can stall if the pending I/Os are not flushed */ |
1353 | if (!noflush) { | |
1354 | md->suspended_bdev = bdget_disk(md->disk, 0); | |
1355 | if (!md->suspended_bdev) { | |
1356 | DMWARN("bdget failed in dm_suspend"); | |
1357 | r = -ENOMEM; | |
1358 | goto flush_and_out; | |
1359 | } | |
e39e2e95 AK |
1360 | } |
1361 | ||
2e93ccc1 KU |
1362 | /* |
1363 | * Flush I/O to the device. | |
1364 | * noflush supersedes do_lockfs, because lock_fs() needs to flush I/Os. | |
1365 | */ | |
1366 | if (do_lockfs && !noflush) { | |
aa8d7c2f AK |
1367 | r = lock_fs(md); |
1368 | if (r) | |
1369 | goto out; | |
1370 | } | |
1da177e4 LT |
1371 | |
1372 | /* | |
354e0071 | 1373 | * First we set the BLOCK_IO flag so no more ios will be mapped. |
1da177e4 | 1374 | */ |
2ca3310e AK |
1375 | down_write(&md->io_lock); |
1376 | set_bit(DMF_BLOCK_IO, &md->flags); | |
1da177e4 | 1377 | |
1da177e4 | 1378 | add_wait_queue(&md->wait, &wait); |
2ca3310e | 1379 | up_write(&md->io_lock); |
1da177e4 LT |
1380 | |
1381 | /* unplug */ | |
2ca3310e | 1382 | if (map) |
1da177e4 | 1383 | dm_table_unplug_all(map); |
1da177e4 LT |
1384 | |
1385 | /* | |
1386 | * Then we wait for the already mapped ios to | |
1387 | * complete. | |
1388 | */ | |
1389 | while (1) { | |
1390 | set_current_state(TASK_INTERRUPTIBLE); | |
1391 | ||
1392 | if (!atomic_read(&md->pending) || signal_pending(current)) | |
1393 | break; | |
1394 | ||
1395 | io_schedule(); | |
1396 | } | |
1397 | set_current_state(TASK_RUNNING); | |
1398 | ||
2ca3310e | 1399 | down_write(&md->io_lock); |
1da177e4 LT |
1400 | remove_wait_queue(&md->wait, &wait); |
1401 | ||
2e93ccc1 KU |
1402 | if (noflush) { |
1403 | spin_lock_irqsave(&md->pushback_lock, flags); | |
1404 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1405 | bio_list_merge_head(&md->deferred, &md->pushback); | |
1406 | bio_list_init(&md->pushback); | |
1407 | spin_unlock_irqrestore(&md->pushback_lock, flags); | |
1408 | } | |
1409 | ||
1da177e4 | 1410 | /* were we interrupted ? */ |
cf222b37 | 1411 | r = -EINTR; |
2ca3310e | 1412 | if (atomic_read(&md->pending)) { |
1ecac7fd JN |
1413 | clear_bit(DMF_BLOCK_IO, &md->flags); |
1414 | def = bio_list_get(&md->deferred); | |
1415 | __flush_deferred_io(md, def); | |
2ca3310e AK |
1416 | up_write(&md->io_lock); |
1417 | unlock_fs(md); | |
2e93ccc1 | 1418 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e AK |
1419 | } |
1420 | up_write(&md->io_lock); | |
1da177e4 | 1421 | |
cf222b37 | 1422 | dm_table_postsuspend_targets(map); |
1da177e4 | 1423 | |
2ca3310e | 1424 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 1425 | |
2ca3310e | 1426 | r = 0; |
b84b0287 | 1427 | |
2e93ccc1 KU |
1428 | flush_and_out: |
1429 | if (r && noflush) { | |
1430 | /* | |
1431 | * Because there may be already I/Os in the pushback list, | |
1432 | * flush them before return. | |
1433 | */ | |
1434 | down_write(&md->io_lock); | |
1435 | ||
1436 | spin_lock_irqsave(&md->pushback_lock, flags); | |
1437 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1438 | bio_list_merge_head(&md->deferred, &md->pushback); | |
1439 | bio_list_init(&md->pushback); | |
1440 | spin_unlock_irqrestore(&md->pushback_lock, flags); | |
1441 | ||
1442 | def = bio_list_get(&md->deferred); | |
1443 | __flush_deferred_io(md, def); | |
1444 | up_write(&md->io_lock); | |
1445 | } | |
1446 | ||
2ca3310e | 1447 | out: |
e39e2e95 AK |
1448 | if (r && md->suspended_bdev) { |
1449 | bdput(md->suspended_bdev); | |
1450 | md->suspended_bdev = NULL; | |
1451 | } | |
1452 | ||
2ca3310e | 1453 | dm_table_put(map); |
d287483d AK |
1454 | |
1455 | out_unlock: | |
2ca3310e | 1456 | up(&md->suspend_lock); |
cf222b37 | 1457 | return r; |
1da177e4 LT |
1458 | } |
1459 | ||
1460 | int dm_resume(struct mapped_device *md) | |
1461 | { | |
cf222b37 | 1462 | int r = -EINVAL; |
1da177e4 | 1463 | struct bio *def; |
cf222b37 | 1464 | struct dm_table *map = NULL; |
1da177e4 | 1465 | |
2ca3310e AK |
1466 | down(&md->suspend_lock); |
1467 | if (!dm_suspended(md)) | |
cf222b37 | 1468 | goto out; |
cf222b37 AK |
1469 | |
1470 | map = dm_get_table(md); | |
2ca3310e | 1471 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 1472 | goto out; |
1da177e4 | 1473 | |
8757b776 MB |
1474 | r = dm_table_resume_targets(map); |
1475 | if (r) | |
1476 | goto out; | |
2ca3310e AK |
1477 | |
1478 | down_write(&md->io_lock); | |
1da177e4 LT |
1479 | clear_bit(DMF_BLOCK_IO, &md->flags); |
1480 | ||
1481 | def = bio_list_get(&md->deferred); | |
1482 | __flush_deferred_io(md, def); | |
2ca3310e AK |
1483 | up_write(&md->io_lock); |
1484 | ||
1485 | unlock_fs(md); | |
1486 | ||
bfa152fa JN |
1487 | if (md->suspended_bdev) { |
1488 | bdput(md->suspended_bdev); | |
1489 | md->suspended_bdev = NULL; | |
1490 | } | |
e39e2e95 | 1491 | |
2ca3310e AK |
1492 | clear_bit(DMF_SUSPENDED, &md->flags); |
1493 | ||
1da177e4 | 1494 | dm_table_unplug_all(map); |
1da177e4 | 1495 | |
8560ed6f HR |
1496 | kobject_uevent(&md->disk->kobj, KOBJ_CHANGE); |
1497 | ||
cf222b37 | 1498 | r = 0; |
2ca3310e | 1499 | |
cf222b37 AK |
1500 | out: |
1501 | dm_table_put(map); | |
2ca3310e AK |
1502 | up(&md->suspend_lock); |
1503 | ||
cf222b37 | 1504 | return r; |
1da177e4 LT |
1505 | } |
1506 | ||
1507 | /*----------------------------------------------------------------- | |
1508 | * Event notification. | |
1509 | *---------------------------------------------------------------*/ | |
1510 | uint32_t dm_get_event_nr(struct mapped_device *md) | |
1511 | { | |
1512 | return atomic_read(&md->event_nr); | |
1513 | } | |
1514 | ||
1515 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
1516 | { | |
1517 | return wait_event_interruptible(md->eventq, | |
1518 | (event_nr != atomic_read(&md->event_nr))); | |
1519 | } | |
1520 | ||
1521 | /* | |
1522 | * The gendisk is only valid as long as you have a reference | |
1523 | * count on 'md'. | |
1524 | */ | |
1525 | struct gendisk *dm_disk(struct mapped_device *md) | |
1526 | { | |
1527 | return md->disk; | |
1528 | } | |
1529 | ||
1530 | int dm_suspended(struct mapped_device *md) | |
1531 | { | |
1532 | return test_bit(DMF_SUSPENDED, &md->flags); | |
1533 | } | |
1534 | ||
2e93ccc1 KU |
1535 | int dm_noflush_suspending(struct dm_target *ti) |
1536 | { | |
1537 | struct mapped_device *md = dm_table_get_md(ti->table); | |
1538 | int r = __noflush_suspending(md); | |
1539 | ||
1540 | dm_put(md); | |
1541 | ||
1542 | return r; | |
1543 | } | |
1544 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
1545 | ||
1da177e4 LT |
1546 | static struct block_device_operations dm_blk_dops = { |
1547 | .open = dm_blk_open, | |
1548 | .release = dm_blk_close, | |
aa129a22 | 1549 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 1550 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
1551 | .owner = THIS_MODULE |
1552 | }; | |
1553 | ||
1554 | EXPORT_SYMBOL(dm_get_mapinfo); | |
1555 | ||
1556 | /* | |
1557 | * module hooks | |
1558 | */ | |
1559 | module_init(dm_init); | |
1560 | module_exit(dm_exit); | |
1561 | ||
1562 | module_param(major, uint, 0); | |
1563 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
1564 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
1565 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
1566 | MODULE_LICENSE("GPL"); |