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 | */ | |
48 | struct target_io { | |
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 AK |
56 | if (bio && bio->bi_private) |
57 | return &((struct target_io *)bio->bi_private)->info; | |
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 |
1da177e4 LT |
71 | |
72 | struct mapped_device { | |
2ca3310e AK |
73 | struct rw_semaphore io_lock; |
74 | struct semaphore suspend_lock; | |
1da177e4 LT |
75 | rwlock_t map_lock; |
76 | atomic_t holders; | |
5c6bd75d | 77 | atomic_t open_count; |
1da177e4 LT |
78 | |
79 | unsigned long flags; | |
80 | ||
81 | request_queue_t *queue; | |
82 | struct gendisk *disk; | |
7e51f257 | 83 | char name[16]; |
1da177e4 LT |
84 | |
85 | void *interface_ptr; | |
86 | ||
87 | /* | |
88 | * A list of ios that arrived while we were suspended. | |
89 | */ | |
90 | atomic_t pending; | |
91 | wait_queue_head_t wait; | |
92 | struct bio_list deferred; | |
93 | ||
94 | /* | |
95 | * The current mapping. | |
96 | */ | |
97 | struct dm_table *map; | |
98 | ||
99 | /* | |
100 | * io objects are allocated from here. | |
101 | */ | |
102 | mempool_t *io_pool; | |
103 | mempool_t *tio_pool; | |
104 | ||
105 | /* | |
106 | * Event handling. | |
107 | */ | |
108 | atomic_t event_nr; | |
109 | wait_queue_head_t eventq; | |
110 | ||
111 | /* | |
112 | * freeze/thaw support require holding onto a super block | |
113 | */ | |
114 | struct super_block *frozen_sb; | |
e39e2e95 | 115 | struct block_device *suspended_bdev; |
3ac51e74 DW |
116 | |
117 | /* forced geometry settings */ | |
118 | struct hd_geometry geometry; | |
1da177e4 LT |
119 | }; |
120 | ||
121 | #define MIN_IOS 256 | |
122 | static kmem_cache_t *_io_cache; | |
123 | static kmem_cache_t *_tio_cache; | |
124 | ||
125 | static struct bio_set *dm_set; | |
126 | ||
127 | static int __init local_init(void) | |
128 | { | |
129 | int r; | |
130 | ||
131 | dm_set = bioset_create(16, 16, 4); | |
132 | if (!dm_set) | |
133 | return -ENOMEM; | |
134 | ||
135 | /* allocate a slab for the dm_ios */ | |
136 | _io_cache = kmem_cache_create("dm_io", | |
137 | sizeof(struct dm_io), 0, 0, NULL, NULL); | |
138 | if (!_io_cache) | |
139 | return -ENOMEM; | |
140 | ||
141 | /* allocate a slab for the target ios */ | |
142 | _tio_cache = kmem_cache_create("dm_tio", sizeof(struct target_io), | |
143 | 0, 0, NULL, NULL); | |
144 | if (!_tio_cache) { | |
145 | kmem_cache_destroy(_io_cache); | |
146 | return -ENOMEM; | |
147 | } | |
148 | ||
149 | _major = major; | |
150 | r = register_blkdev(_major, _name); | |
151 | if (r < 0) { | |
152 | kmem_cache_destroy(_tio_cache); | |
153 | kmem_cache_destroy(_io_cache); | |
154 | return r; | |
155 | } | |
156 | ||
157 | if (!_major) | |
158 | _major = r; | |
159 | ||
160 | return 0; | |
161 | } | |
162 | ||
163 | static void local_exit(void) | |
164 | { | |
165 | kmem_cache_destroy(_tio_cache); | |
166 | kmem_cache_destroy(_io_cache); | |
167 | ||
168 | bioset_free(dm_set); | |
169 | ||
170 | if (unregister_blkdev(_major, _name) < 0) | |
890fbae2 | 171 | DMERR("unregister_blkdev failed"); |
1da177e4 LT |
172 | |
173 | _major = 0; | |
174 | ||
175 | DMINFO("cleaned up"); | |
176 | } | |
177 | ||
178 | int (*_inits[])(void) __initdata = { | |
179 | local_init, | |
180 | dm_target_init, | |
181 | dm_linear_init, | |
182 | dm_stripe_init, | |
183 | dm_interface_init, | |
184 | }; | |
185 | ||
186 | void (*_exits[])(void) = { | |
187 | local_exit, | |
188 | dm_target_exit, | |
189 | dm_linear_exit, | |
190 | dm_stripe_exit, | |
191 | dm_interface_exit, | |
192 | }; | |
193 | ||
194 | static int __init dm_init(void) | |
195 | { | |
196 | const int count = ARRAY_SIZE(_inits); | |
197 | ||
198 | int r, i; | |
199 | ||
200 | for (i = 0; i < count; i++) { | |
201 | r = _inits[i](); | |
202 | if (r) | |
203 | goto bad; | |
204 | } | |
205 | ||
206 | return 0; | |
207 | ||
208 | bad: | |
209 | while (i--) | |
210 | _exits[i](); | |
211 | ||
212 | return r; | |
213 | } | |
214 | ||
215 | static void __exit dm_exit(void) | |
216 | { | |
217 | int i = ARRAY_SIZE(_exits); | |
218 | ||
219 | while (i--) | |
220 | _exits[i](); | |
221 | } | |
222 | ||
223 | /* | |
224 | * Block device functions | |
225 | */ | |
226 | static int dm_blk_open(struct inode *inode, struct file *file) | |
227 | { | |
228 | struct mapped_device *md; | |
229 | ||
fba9f90e JM |
230 | spin_lock(&_minor_lock); |
231 | ||
1da177e4 | 232 | md = inode->i_bdev->bd_disk->private_data; |
fba9f90e JM |
233 | if (!md) |
234 | goto out; | |
235 | ||
5c6bd75d AK |
236 | if (test_bit(DMF_FREEING, &md->flags) || |
237 | test_bit(DMF_DELETING, &md->flags)) { | |
fba9f90e JM |
238 | md = NULL; |
239 | goto out; | |
240 | } | |
241 | ||
1da177e4 | 242 | dm_get(md); |
5c6bd75d | 243 | atomic_inc(&md->open_count); |
fba9f90e JM |
244 | |
245 | out: | |
246 | spin_unlock(&_minor_lock); | |
247 | ||
248 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
249 | } |
250 | ||
251 | static int dm_blk_close(struct inode *inode, struct file *file) | |
252 | { | |
253 | struct mapped_device *md; | |
254 | ||
255 | md = inode->i_bdev->bd_disk->private_data; | |
5c6bd75d | 256 | atomic_dec(&md->open_count); |
1da177e4 LT |
257 | dm_put(md); |
258 | return 0; | |
259 | } | |
260 | ||
5c6bd75d AK |
261 | int dm_open_count(struct mapped_device *md) |
262 | { | |
263 | return atomic_read(&md->open_count); | |
264 | } | |
265 | ||
266 | /* | |
267 | * Guarantees nothing is using the device before it's deleted. | |
268 | */ | |
269 | int dm_lock_for_deletion(struct mapped_device *md) | |
270 | { | |
271 | int r = 0; | |
272 | ||
273 | spin_lock(&_minor_lock); | |
274 | ||
275 | if (dm_open_count(md)) | |
276 | r = -EBUSY; | |
277 | else | |
278 | set_bit(DMF_DELETING, &md->flags); | |
279 | ||
280 | spin_unlock(&_minor_lock); | |
281 | ||
282 | return r; | |
283 | } | |
284 | ||
3ac51e74 DW |
285 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
286 | { | |
287 | struct mapped_device *md = bdev->bd_disk->private_data; | |
288 | ||
289 | return dm_get_geometry(md, geo); | |
290 | } | |
291 | ||
aa129a22 MB |
292 | static int dm_blk_ioctl(struct inode *inode, struct file *file, |
293 | unsigned int cmd, unsigned long arg) | |
294 | { | |
295 | struct mapped_device *md; | |
296 | struct dm_table *map; | |
297 | struct dm_target *tgt; | |
298 | int r = -ENOTTY; | |
299 | ||
300 | /* We don't really need this lock, but we do need 'inode'. */ | |
301 | unlock_kernel(); | |
302 | ||
303 | md = inode->i_bdev->bd_disk->private_data; | |
304 | ||
305 | map = dm_get_table(md); | |
306 | ||
307 | if (!map || !dm_table_get_size(map)) | |
308 | goto out; | |
309 | ||
310 | /* We only support devices that have a single target */ | |
311 | if (dm_table_get_num_targets(map) != 1) | |
312 | goto out; | |
313 | ||
314 | tgt = dm_table_get_target(map, 0); | |
315 | ||
316 | if (dm_suspended(md)) { | |
317 | r = -EAGAIN; | |
318 | goto out; | |
319 | } | |
320 | ||
321 | if (tgt->type->ioctl) | |
322 | r = tgt->type->ioctl(tgt, inode, file, cmd, arg); | |
323 | ||
324 | out: | |
325 | dm_table_put(map); | |
326 | ||
327 | lock_kernel(); | |
328 | return r; | |
329 | } | |
330 | ||
1da177e4 LT |
331 | static inline struct dm_io *alloc_io(struct mapped_device *md) |
332 | { | |
333 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
334 | } | |
335 | ||
336 | static inline void free_io(struct mapped_device *md, struct dm_io *io) | |
337 | { | |
338 | mempool_free(io, md->io_pool); | |
339 | } | |
340 | ||
341 | static inline struct target_io *alloc_tio(struct mapped_device *md) | |
342 | { | |
343 | return mempool_alloc(md->tio_pool, GFP_NOIO); | |
344 | } | |
345 | ||
346 | static inline void free_tio(struct mapped_device *md, struct target_io *tio) | |
347 | { | |
348 | mempool_free(tio, md->tio_pool); | |
349 | } | |
350 | ||
3eaf840e JNN |
351 | static void start_io_acct(struct dm_io *io) |
352 | { | |
353 | struct mapped_device *md = io->md; | |
354 | ||
355 | io->start_time = jiffies; | |
356 | ||
357 | preempt_disable(); | |
358 | disk_round_stats(dm_disk(md)); | |
359 | preempt_enable(); | |
360 | dm_disk(md)->in_flight = atomic_inc_return(&md->pending); | |
361 | } | |
362 | ||
363 | static int end_io_acct(struct dm_io *io) | |
364 | { | |
365 | struct mapped_device *md = io->md; | |
366 | struct bio *bio = io->bio; | |
367 | unsigned long duration = jiffies - io->start_time; | |
368 | int pending; | |
369 | int rw = bio_data_dir(bio); | |
370 | ||
371 | preempt_disable(); | |
372 | disk_round_stats(dm_disk(md)); | |
373 | preempt_enable(); | |
374 | dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending); | |
375 | ||
376 | disk_stat_add(dm_disk(md), ticks[rw], duration); | |
377 | ||
378 | return !pending; | |
379 | } | |
380 | ||
1da177e4 LT |
381 | /* |
382 | * Add the bio to the list of deferred io. | |
383 | */ | |
384 | static int queue_io(struct mapped_device *md, struct bio *bio) | |
385 | { | |
2ca3310e | 386 | down_write(&md->io_lock); |
1da177e4 LT |
387 | |
388 | if (!test_bit(DMF_BLOCK_IO, &md->flags)) { | |
2ca3310e | 389 | up_write(&md->io_lock); |
1da177e4 LT |
390 | return 1; |
391 | } | |
392 | ||
393 | bio_list_add(&md->deferred, bio); | |
394 | ||
2ca3310e | 395 | up_write(&md->io_lock); |
1da177e4 LT |
396 | return 0; /* deferred successfully */ |
397 | } | |
398 | ||
399 | /* | |
400 | * Everyone (including functions in this file), should use this | |
401 | * function to access the md->map field, and make sure they call | |
402 | * dm_table_put() when finished. | |
403 | */ | |
404 | struct dm_table *dm_get_table(struct mapped_device *md) | |
405 | { | |
406 | struct dm_table *t; | |
407 | ||
408 | read_lock(&md->map_lock); | |
409 | t = md->map; | |
410 | if (t) | |
411 | dm_table_get(t); | |
412 | read_unlock(&md->map_lock); | |
413 | ||
414 | return t; | |
415 | } | |
416 | ||
3ac51e74 DW |
417 | /* |
418 | * Get the geometry associated with a dm device | |
419 | */ | |
420 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
421 | { | |
422 | *geo = md->geometry; | |
423 | ||
424 | return 0; | |
425 | } | |
426 | ||
427 | /* | |
428 | * Set the geometry of a device. | |
429 | */ | |
430 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
431 | { | |
432 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
433 | ||
434 | if (geo->start > sz) { | |
435 | DMWARN("Start sector is beyond the geometry limits."); | |
436 | return -EINVAL; | |
437 | } | |
438 | ||
439 | md->geometry = *geo; | |
440 | ||
441 | return 0; | |
442 | } | |
443 | ||
1da177e4 LT |
444 | /*----------------------------------------------------------------- |
445 | * CRUD START: | |
446 | * A more elegant soln is in the works that uses the queue | |
447 | * merge fn, unfortunately there are a couple of changes to | |
448 | * the block layer that I want to make for this. So in the | |
449 | * interests of getting something for people to use I give | |
450 | * you this clearly demarcated crap. | |
451 | *---------------------------------------------------------------*/ | |
452 | ||
453 | /* | |
454 | * Decrements the number of outstanding ios that a bio has been | |
455 | * cloned into, completing the original io if necc. | |
456 | */ | |
858119e1 | 457 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 LT |
458 | { |
459 | if (error) | |
460 | io->error = error; | |
461 | ||
462 | if (atomic_dec_and_test(&io->io_count)) { | |
3eaf840e | 463 | if (end_io_acct(io)) |
1da177e4 LT |
464 | /* nudge anyone waiting on suspend queue */ |
465 | wake_up(&io->md->wait); | |
466 | ||
2056a782 JA |
467 | blk_add_trace_bio(io->md->queue, io->bio, BLK_TA_COMPLETE); |
468 | ||
1da177e4 LT |
469 | bio_endio(io->bio, io->bio->bi_size, io->error); |
470 | free_io(io->md, io); | |
471 | } | |
472 | } | |
473 | ||
474 | static int clone_endio(struct bio *bio, unsigned int done, int error) | |
475 | { | |
476 | int r = 0; | |
477 | struct target_io *tio = bio->bi_private; | |
478 | struct dm_io *io = tio->io; | |
479 | dm_endio_fn endio = tio->ti->type->end_io; | |
480 | ||
481 | if (bio->bi_size) | |
482 | return 1; | |
483 | ||
484 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) | |
485 | error = -EIO; | |
486 | ||
487 | if (endio) { | |
488 | r = endio(tio->ti, bio, error, &tio->info); | |
489 | if (r < 0) | |
490 | error = r; | |
491 | ||
492 | else if (r > 0) | |
493 | /* the target wants another shot at the io */ | |
494 | return 1; | |
495 | } | |
496 | ||
497 | free_tio(io->md, tio); | |
498 | dec_pending(io, error); | |
499 | bio_put(bio); | |
500 | return r; | |
501 | } | |
502 | ||
503 | static sector_t max_io_len(struct mapped_device *md, | |
504 | sector_t sector, struct dm_target *ti) | |
505 | { | |
506 | sector_t offset = sector - ti->begin; | |
507 | sector_t len = ti->len - offset; | |
508 | ||
509 | /* | |
510 | * Does the target need to split even further ? | |
511 | */ | |
512 | if (ti->split_io) { | |
513 | sector_t boundary; | |
514 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) | |
515 | - offset; | |
516 | if (len > boundary) | |
517 | len = boundary; | |
518 | } | |
519 | ||
520 | return len; | |
521 | } | |
522 | ||
523 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
524 | struct target_io *tio) | |
525 | { | |
526 | int r; | |
2056a782 | 527 | sector_t sector; |
1da177e4 LT |
528 | |
529 | /* | |
530 | * Sanity checks. | |
531 | */ | |
532 | BUG_ON(!clone->bi_size); | |
533 | ||
534 | clone->bi_end_io = clone_endio; | |
535 | clone->bi_private = tio; | |
536 | ||
537 | /* | |
538 | * Map the clone. If r == 0 we don't need to do | |
539 | * anything, the target has assumed ownership of | |
540 | * this io. | |
541 | */ | |
542 | atomic_inc(&tio->io->io_count); | |
2056a782 | 543 | sector = clone->bi_sector; |
1da177e4 | 544 | r = ti->type->map(ti, clone, &tio->info); |
2056a782 | 545 | if (r > 0) { |
1da177e4 | 546 | /* the bio has been remapped so dispatch it */ |
2056a782 | 547 | |
17b2f66f AK |
548 | blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone, |
549 | tio->io->bio->bi_bdev->bd_dev, sector, | |
2056a782 JA |
550 | clone->bi_sector); |
551 | ||
1da177e4 | 552 | generic_make_request(clone); |
2056a782 | 553 | } |
1da177e4 LT |
554 | |
555 | else if (r < 0) { | |
556 | /* error the io and bail out */ | |
557 | struct dm_io *io = tio->io; | |
558 | free_tio(tio->io->md, tio); | |
f6a80ea8 | 559 | dec_pending(io, r); |
1da177e4 LT |
560 | bio_put(clone); |
561 | } | |
562 | } | |
563 | ||
564 | struct clone_info { | |
565 | struct mapped_device *md; | |
566 | struct dm_table *map; | |
567 | struct bio *bio; | |
568 | struct dm_io *io; | |
569 | sector_t sector; | |
570 | sector_t sector_count; | |
571 | unsigned short idx; | |
572 | }; | |
573 | ||
3676347a PO |
574 | static void dm_bio_destructor(struct bio *bio) |
575 | { | |
576 | bio_free(bio, dm_set); | |
577 | } | |
578 | ||
1da177e4 LT |
579 | /* |
580 | * Creates a little bio that is just does part of a bvec. | |
581 | */ | |
582 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
583 | unsigned short idx, unsigned int offset, | |
584 | unsigned int len) | |
585 | { | |
586 | struct bio *clone; | |
587 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
588 | ||
589 | clone = bio_alloc_bioset(GFP_NOIO, 1, dm_set); | |
3676347a | 590 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
591 | *clone->bi_io_vec = *bv; |
592 | ||
593 | clone->bi_sector = sector; | |
594 | clone->bi_bdev = bio->bi_bdev; | |
595 | clone->bi_rw = bio->bi_rw; | |
596 | clone->bi_vcnt = 1; | |
597 | clone->bi_size = to_bytes(len); | |
598 | clone->bi_io_vec->bv_offset = offset; | |
599 | clone->bi_io_vec->bv_len = clone->bi_size; | |
600 | ||
601 | return clone; | |
602 | } | |
603 | ||
604 | /* | |
605 | * Creates a bio that consists of range of complete bvecs. | |
606 | */ | |
607 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
608 | unsigned short idx, unsigned short bv_count, | |
609 | unsigned int len) | |
610 | { | |
611 | struct bio *clone; | |
612 | ||
613 | clone = bio_clone(bio, GFP_NOIO); | |
614 | clone->bi_sector = sector; | |
615 | clone->bi_idx = idx; | |
616 | clone->bi_vcnt = idx + bv_count; | |
617 | clone->bi_size = to_bytes(len); | |
618 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
619 | ||
620 | return clone; | |
621 | } | |
622 | ||
623 | static void __clone_and_map(struct clone_info *ci) | |
624 | { | |
625 | struct bio *clone, *bio = ci->bio; | |
626 | struct dm_target *ti = dm_table_find_target(ci->map, ci->sector); | |
627 | sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti); | |
628 | struct target_io *tio; | |
629 | ||
630 | /* | |
631 | * Allocate a target io object. | |
632 | */ | |
633 | tio = alloc_tio(ci->md); | |
634 | tio->io = ci->io; | |
635 | tio->ti = ti; | |
636 | memset(&tio->info, 0, sizeof(tio->info)); | |
637 | ||
638 | if (ci->sector_count <= max) { | |
639 | /* | |
640 | * Optimise for the simple case where we can do all of | |
641 | * the remaining io with a single clone. | |
642 | */ | |
643 | clone = clone_bio(bio, ci->sector, ci->idx, | |
644 | bio->bi_vcnt - ci->idx, ci->sector_count); | |
645 | __map_bio(ti, clone, tio); | |
646 | ci->sector_count = 0; | |
647 | ||
648 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
649 | /* | |
650 | * There are some bvecs that don't span targets. | |
651 | * Do as many of these as possible. | |
652 | */ | |
653 | int i; | |
654 | sector_t remaining = max; | |
655 | sector_t bv_len; | |
656 | ||
657 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
658 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
659 | ||
660 | if (bv_len > remaining) | |
661 | break; | |
662 | ||
663 | remaining -= bv_len; | |
664 | len += bv_len; | |
665 | } | |
666 | ||
667 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len); | |
668 | __map_bio(ti, clone, tio); | |
669 | ||
670 | ci->sector += len; | |
671 | ci->sector_count -= len; | |
672 | ci->idx = i; | |
673 | ||
674 | } else { | |
675 | /* | |
d2044a94 | 676 | * Handle a bvec that must be split between two or more targets. |
1da177e4 LT |
677 | */ |
678 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
d2044a94 AK |
679 | sector_t remaining = to_sector(bv->bv_len); |
680 | unsigned int offset = 0; | |
1da177e4 | 681 | |
d2044a94 AK |
682 | do { |
683 | if (offset) { | |
684 | ti = dm_table_find_target(ci->map, ci->sector); | |
685 | max = max_io_len(ci->md, ci->sector, ti); | |
1da177e4 | 686 | |
d2044a94 AK |
687 | tio = alloc_tio(ci->md); |
688 | tio->io = ci->io; | |
689 | tio->ti = ti; | |
690 | memset(&tio->info, 0, sizeof(tio->info)); | |
691 | } | |
692 | ||
693 | len = min(remaining, max); | |
694 | ||
695 | clone = split_bvec(bio, ci->sector, ci->idx, | |
696 | bv->bv_offset + offset, len); | |
697 | ||
698 | __map_bio(ti, clone, tio); | |
699 | ||
700 | ci->sector += len; | |
701 | ci->sector_count -= len; | |
702 | offset += to_bytes(len); | |
703 | } while (remaining -= len); | |
1da177e4 | 704 | |
1da177e4 LT |
705 | ci->idx++; |
706 | } | |
707 | } | |
708 | ||
709 | /* | |
710 | * Split the bio into several clones. | |
711 | */ | |
712 | static void __split_bio(struct mapped_device *md, struct bio *bio) | |
713 | { | |
714 | struct clone_info ci; | |
715 | ||
716 | ci.map = dm_get_table(md); | |
717 | if (!ci.map) { | |
718 | bio_io_error(bio, bio->bi_size); | |
719 | return; | |
720 | } | |
721 | ||
722 | ci.md = md; | |
723 | ci.bio = bio; | |
724 | ci.io = alloc_io(md); | |
725 | ci.io->error = 0; | |
726 | atomic_set(&ci.io->io_count, 1); | |
727 | ci.io->bio = bio; | |
728 | ci.io->md = md; | |
729 | ci.sector = bio->bi_sector; | |
730 | ci.sector_count = bio_sectors(bio); | |
731 | ci.idx = bio->bi_idx; | |
732 | ||
3eaf840e | 733 | start_io_acct(ci.io); |
1da177e4 LT |
734 | while (ci.sector_count) |
735 | __clone_and_map(&ci); | |
736 | ||
737 | /* drop the extra reference count */ | |
738 | dec_pending(ci.io, 0); | |
739 | dm_table_put(ci.map); | |
740 | } | |
741 | /*----------------------------------------------------------------- | |
742 | * CRUD END | |
743 | *---------------------------------------------------------------*/ | |
744 | ||
745 | /* | |
746 | * The request function that just remaps the bio built up by | |
747 | * dm_merge_bvec. | |
748 | */ | |
749 | static int dm_request(request_queue_t *q, struct bio *bio) | |
750 | { | |
751 | int r; | |
12f03a49 | 752 | int rw = bio_data_dir(bio); |
1da177e4 LT |
753 | struct mapped_device *md = q->queuedata; |
754 | ||
2ca3310e | 755 | down_read(&md->io_lock); |
1da177e4 | 756 | |
12f03a49 KC |
757 | disk_stat_inc(dm_disk(md), ios[rw]); |
758 | disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio)); | |
759 | ||
1da177e4 LT |
760 | /* |
761 | * If we're suspended we have to queue | |
762 | * this io for later. | |
763 | */ | |
764 | while (test_bit(DMF_BLOCK_IO, &md->flags)) { | |
2ca3310e | 765 | up_read(&md->io_lock); |
1da177e4 LT |
766 | |
767 | if (bio_rw(bio) == READA) { | |
768 | bio_io_error(bio, bio->bi_size); | |
769 | return 0; | |
770 | } | |
771 | ||
772 | r = queue_io(md, bio); | |
773 | if (r < 0) { | |
774 | bio_io_error(bio, bio->bi_size); | |
775 | return 0; | |
776 | ||
777 | } else if (r == 0) | |
778 | return 0; /* deferred successfully */ | |
779 | ||
780 | /* | |
781 | * We're in a while loop, because someone could suspend | |
782 | * before we get to the following read lock. | |
783 | */ | |
2ca3310e | 784 | down_read(&md->io_lock); |
1da177e4 LT |
785 | } |
786 | ||
787 | __split_bio(md, bio); | |
2ca3310e | 788 | up_read(&md->io_lock); |
1da177e4 LT |
789 | return 0; |
790 | } | |
791 | ||
792 | static int dm_flush_all(request_queue_t *q, struct gendisk *disk, | |
793 | sector_t *error_sector) | |
794 | { | |
795 | struct mapped_device *md = q->queuedata; | |
796 | struct dm_table *map = dm_get_table(md); | |
797 | int ret = -ENXIO; | |
798 | ||
799 | if (map) { | |
cf222b37 | 800 | ret = dm_table_flush_all(map); |
1da177e4 LT |
801 | dm_table_put(map); |
802 | } | |
803 | ||
804 | return ret; | |
805 | } | |
806 | ||
807 | static void dm_unplug_all(request_queue_t *q) | |
808 | { | |
809 | struct mapped_device *md = q->queuedata; | |
810 | struct dm_table *map = dm_get_table(md); | |
811 | ||
812 | if (map) { | |
813 | dm_table_unplug_all(map); | |
814 | dm_table_put(map); | |
815 | } | |
816 | } | |
817 | ||
818 | static int dm_any_congested(void *congested_data, int bdi_bits) | |
819 | { | |
820 | int r; | |
821 | struct mapped_device *md = (struct mapped_device *) congested_data; | |
822 | struct dm_table *map = dm_get_table(md); | |
823 | ||
824 | if (!map || test_bit(DMF_BLOCK_IO, &md->flags)) | |
825 | r = bdi_bits; | |
826 | else | |
827 | r = dm_table_any_congested(map, bdi_bits); | |
828 | ||
829 | dm_table_put(map); | |
830 | return r; | |
831 | } | |
832 | ||
833 | /*----------------------------------------------------------------- | |
834 | * An IDR is used to keep track of allocated minor numbers. | |
835 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
836 | static DEFINE_IDR(_minor_idr); |
837 | ||
2b06cfff | 838 | static void free_minor(int minor) |
1da177e4 | 839 | { |
f32c10b0 | 840 | spin_lock(&_minor_lock); |
1da177e4 | 841 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 842 | spin_unlock(&_minor_lock); |
1da177e4 LT |
843 | } |
844 | ||
845 | /* | |
846 | * See if the device with a specific minor # is free. | |
847 | */ | |
2b06cfff | 848 | static int specific_minor(struct mapped_device *md, int minor) |
1da177e4 LT |
849 | { |
850 | int r, m; | |
851 | ||
852 | if (minor >= (1 << MINORBITS)) | |
853 | return -EINVAL; | |
854 | ||
62f75c2f JM |
855 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
856 | if (!r) | |
857 | return -ENOMEM; | |
858 | ||
f32c10b0 | 859 | spin_lock(&_minor_lock); |
1da177e4 LT |
860 | |
861 | if (idr_find(&_minor_idr, minor)) { | |
862 | r = -EBUSY; | |
863 | goto out; | |
864 | } | |
865 | ||
ba61fdd1 | 866 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); |
62f75c2f | 867 | if (r) |
1da177e4 | 868 | goto out; |
1da177e4 LT |
869 | |
870 | if (m != minor) { | |
871 | idr_remove(&_minor_idr, m); | |
872 | r = -EBUSY; | |
873 | goto out; | |
874 | } | |
875 | ||
876 | out: | |
f32c10b0 | 877 | spin_unlock(&_minor_lock); |
1da177e4 LT |
878 | return r; |
879 | } | |
880 | ||
2b06cfff | 881 | static int next_free_minor(struct mapped_device *md, int *minor) |
1da177e4 | 882 | { |
2b06cfff | 883 | int r, m; |
1da177e4 | 884 | |
1da177e4 | 885 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
62f75c2f JM |
886 | if (!r) |
887 | return -ENOMEM; | |
888 | ||
f32c10b0 | 889 | spin_lock(&_minor_lock); |
1da177e4 | 890 | |
ba61fdd1 | 891 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); |
1da177e4 LT |
892 | if (r) { |
893 | goto out; | |
894 | } | |
895 | ||
896 | if (m >= (1 << MINORBITS)) { | |
897 | idr_remove(&_minor_idr, m); | |
898 | r = -ENOSPC; | |
899 | goto out; | |
900 | } | |
901 | ||
902 | *minor = m; | |
903 | ||
904 | out: | |
f32c10b0 | 905 | spin_unlock(&_minor_lock); |
1da177e4 LT |
906 | return r; |
907 | } | |
908 | ||
909 | static struct block_device_operations dm_blk_dops; | |
910 | ||
911 | /* | |
912 | * Allocate and initialise a blank device with a given minor. | |
913 | */ | |
2b06cfff | 914 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
915 | { |
916 | int r; | |
917 | struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL); | |
ba61fdd1 | 918 | void *old_md; |
1da177e4 LT |
919 | |
920 | if (!md) { | |
921 | DMWARN("unable to allocate device, out of memory."); | |
922 | return NULL; | |
923 | } | |
924 | ||
10da4f79 JM |
925 | if (!try_module_get(THIS_MODULE)) |
926 | goto bad0; | |
927 | ||
1da177e4 | 928 | /* get a minor number for the dev */ |
2b06cfff AK |
929 | if (minor == DM_ANY_MINOR) |
930 | r = next_free_minor(md, &minor); | |
931 | else | |
932 | r = specific_minor(md, minor); | |
1da177e4 LT |
933 | if (r < 0) |
934 | goto bad1; | |
935 | ||
936 | memset(md, 0, sizeof(*md)); | |
2ca3310e AK |
937 | init_rwsem(&md->io_lock); |
938 | init_MUTEX(&md->suspend_lock); | |
1da177e4 LT |
939 | rwlock_init(&md->map_lock); |
940 | atomic_set(&md->holders, 1); | |
5c6bd75d | 941 | atomic_set(&md->open_count, 0); |
1da177e4 LT |
942 | atomic_set(&md->event_nr, 0); |
943 | ||
944 | md->queue = blk_alloc_queue(GFP_KERNEL); | |
945 | if (!md->queue) | |
e3f6ac61 | 946 | goto bad1_free_minor; |
1da177e4 LT |
947 | |
948 | md->queue->queuedata = md; | |
949 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
950 | md->queue->backing_dev_info.congested_data = md; | |
951 | blk_queue_make_request(md->queue, dm_request); | |
daef265f | 952 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
1da177e4 LT |
953 | md->queue->unplug_fn = dm_unplug_all; |
954 | md->queue->issue_flush_fn = dm_flush_all; | |
955 | ||
93d2341c | 956 | md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache); |
1da177e4 LT |
957 | if (!md->io_pool) |
958 | goto bad2; | |
959 | ||
93d2341c | 960 | md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache); |
1da177e4 LT |
961 | if (!md->tio_pool) |
962 | goto bad3; | |
963 | ||
964 | md->disk = alloc_disk(1); | |
965 | if (!md->disk) | |
966 | goto bad4; | |
967 | ||
f0b04115 JM |
968 | atomic_set(&md->pending, 0); |
969 | init_waitqueue_head(&md->wait); | |
970 | init_waitqueue_head(&md->eventq); | |
971 | ||
1da177e4 LT |
972 | md->disk->major = _major; |
973 | md->disk->first_minor = minor; | |
974 | md->disk->fops = &dm_blk_dops; | |
975 | md->disk->queue = md->queue; | |
976 | md->disk->private_data = md; | |
977 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
978 | add_disk(md->disk); | |
7e51f257 | 979 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 980 | |
ba61fdd1 | 981 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 982 | spin_lock(&_minor_lock); |
ba61fdd1 | 983 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 984 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
985 | |
986 | BUG_ON(old_md != MINOR_ALLOCED); | |
987 | ||
1da177e4 LT |
988 | return md; |
989 | ||
990 | bad4: | |
991 | mempool_destroy(md->tio_pool); | |
992 | bad3: | |
993 | mempool_destroy(md->io_pool); | |
994 | bad2: | |
1312f40e | 995 | blk_cleanup_queue(md->queue); |
e3f6ac61 | 996 | bad1_free_minor: |
1da177e4 LT |
997 | free_minor(minor); |
998 | bad1: | |
10da4f79 JM |
999 | module_put(THIS_MODULE); |
1000 | bad0: | |
1da177e4 LT |
1001 | kfree(md); |
1002 | return NULL; | |
1003 | } | |
1004 | ||
1005 | static void free_dev(struct mapped_device *md) | |
1006 | { | |
2b06cfff | 1007 | int minor = md->disk->first_minor; |
63d94e48 | 1008 | |
d9dde59b JN |
1009 | if (md->suspended_bdev) { |
1010 | thaw_bdev(md->suspended_bdev, NULL); | |
1011 | bdput(md->suspended_bdev); | |
1012 | } | |
1da177e4 LT |
1013 | mempool_destroy(md->tio_pool); |
1014 | mempool_destroy(md->io_pool); | |
1015 | del_gendisk(md->disk); | |
63d94e48 | 1016 | free_minor(minor); |
fba9f90e JM |
1017 | |
1018 | spin_lock(&_minor_lock); | |
1019 | md->disk->private_data = NULL; | |
1020 | spin_unlock(&_minor_lock); | |
1021 | ||
1da177e4 | 1022 | put_disk(md->disk); |
1312f40e | 1023 | blk_cleanup_queue(md->queue); |
10da4f79 | 1024 | module_put(THIS_MODULE); |
1da177e4 LT |
1025 | kfree(md); |
1026 | } | |
1027 | ||
1028 | /* | |
1029 | * Bind a table to the device. | |
1030 | */ | |
1031 | static void event_callback(void *context) | |
1032 | { | |
1033 | struct mapped_device *md = (struct mapped_device *) context; | |
1034 | ||
1035 | atomic_inc(&md->event_nr); | |
1036 | wake_up(&md->eventq); | |
1037 | } | |
1038 | ||
4e90188b | 1039 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1040 | { |
4e90188b | 1041 | set_capacity(md->disk, size); |
1da177e4 | 1042 | |
1b1dcc1b | 1043 | mutex_lock(&md->suspended_bdev->bd_inode->i_mutex); |
e39e2e95 | 1044 | i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1b1dcc1b | 1045 | mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex); |
1da177e4 LT |
1046 | } |
1047 | ||
1048 | static int __bind(struct mapped_device *md, struct dm_table *t) | |
1049 | { | |
1050 | request_queue_t *q = md->queue; | |
1051 | sector_t size; | |
1052 | ||
1053 | size = dm_table_get_size(t); | |
3ac51e74 DW |
1054 | |
1055 | /* | |
1056 | * Wipe any geometry if the size of the table changed. | |
1057 | */ | |
1058 | if (size != get_capacity(md->disk)) | |
1059 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
1060 | ||
4e90188b | 1061 | __set_size(md, size); |
1da177e4 LT |
1062 | if (size == 0) |
1063 | return 0; | |
1064 | ||
2ca3310e AK |
1065 | dm_table_get(t); |
1066 | dm_table_event_callback(t, event_callback, md); | |
1067 | ||
1da177e4 LT |
1068 | write_lock(&md->map_lock); |
1069 | md->map = t; | |
2ca3310e | 1070 | dm_table_set_restrictions(t, q); |
1da177e4 LT |
1071 | write_unlock(&md->map_lock); |
1072 | ||
1da177e4 LT |
1073 | return 0; |
1074 | } | |
1075 | ||
1076 | static void __unbind(struct mapped_device *md) | |
1077 | { | |
1078 | struct dm_table *map = md->map; | |
1079 | ||
1080 | if (!map) | |
1081 | return; | |
1082 | ||
1083 | dm_table_event_callback(map, NULL, NULL); | |
1084 | write_lock(&md->map_lock); | |
1085 | md->map = NULL; | |
1086 | write_unlock(&md->map_lock); | |
1087 | dm_table_put(map); | |
1088 | } | |
1089 | ||
1090 | /* | |
1091 | * Constructor for a new device. | |
1092 | */ | |
2b06cfff | 1093 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
1094 | { |
1095 | struct mapped_device *md; | |
1096 | ||
2b06cfff | 1097 | md = alloc_dev(minor); |
1da177e4 LT |
1098 | if (!md) |
1099 | return -ENXIO; | |
1100 | ||
1101 | *result = md; | |
1102 | return 0; | |
1103 | } | |
1104 | ||
637842cf | 1105 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
1106 | { |
1107 | struct mapped_device *md; | |
1da177e4 LT |
1108 | unsigned minor = MINOR(dev); |
1109 | ||
1110 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
1111 | return NULL; | |
1112 | ||
f32c10b0 | 1113 | spin_lock(&_minor_lock); |
1da177e4 LT |
1114 | |
1115 | md = idr_find(&_minor_idr, minor); | |
fba9f90e JM |
1116 | if (md && (md == MINOR_ALLOCED || |
1117 | (dm_disk(md)->first_minor != minor) || | |
17b2f66f | 1118 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 1119 | md = NULL; |
fba9f90e JM |
1120 | goto out; |
1121 | } | |
1da177e4 | 1122 | |
fba9f90e | 1123 | out: |
f32c10b0 | 1124 | spin_unlock(&_minor_lock); |
1da177e4 | 1125 | |
637842cf DT |
1126 | return md; |
1127 | } | |
1128 | ||
d229a958 DT |
1129 | struct mapped_device *dm_get_md(dev_t dev) |
1130 | { | |
1131 | struct mapped_device *md = dm_find_md(dev); | |
1132 | ||
1133 | if (md) | |
1134 | dm_get(md); | |
1135 | ||
1136 | return md; | |
1137 | } | |
1138 | ||
9ade92a9 | 1139 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 1140 | { |
9ade92a9 | 1141 | return md->interface_ptr; |
1da177e4 LT |
1142 | } |
1143 | ||
1144 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
1145 | { | |
1146 | md->interface_ptr = ptr; | |
1147 | } | |
1148 | ||
1149 | void dm_get(struct mapped_device *md) | |
1150 | { | |
1151 | atomic_inc(&md->holders); | |
1152 | } | |
1153 | ||
72d94861 AK |
1154 | const char *dm_device_name(struct mapped_device *md) |
1155 | { | |
1156 | return md->name; | |
1157 | } | |
1158 | EXPORT_SYMBOL_GPL(dm_device_name); | |
1159 | ||
1da177e4 LT |
1160 | void dm_put(struct mapped_device *md) |
1161 | { | |
1134e5ae | 1162 | struct dm_table *map; |
1da177e4 | 1163 | |
fba9f90e JM |
1164 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1165 | ||
f32c10b0 | 1166 | if (atomic_dec_and_lock(&md->holders, &_minor_lock)) { |
1134e5ae | 1167 | map = dm_get_table(md); |
ba61fdd1 | 1168 | idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor); |
fba9f90e | 1169 | set_bit(DMF_FREEING, &md->flags); |
f32c10b0 | 1170 | spin_unlock(&_minor_lock); |
cf222b37 | 1171 | if (!dm_suspended(md)) { |
1da177e4 LT |
1172 | dm_table_presuspend_targets(map); |
1173 | dm_table_postsuspend_targets(map); | |
1174 | } | |
1175 | __unbind(md); | |
1134e5ae | 1176 | dm_table_put(map); |
1da177e4 LT |
1177 | free_dev(md); |
1178 | } | |
1da177e4 LT |
1179 | } |
1180 | ||
1181 | /* | |
1182 | * Process the deferred bios | |
1183 | */ | |
1184 | static void __flush_deferred_io(struct mapped_device *md, struct bio *c) | |
1185 | { | |
1186 | struct bio *n; | |
1187 | ||
1188 | while (c) { | |
1189 | n = c->bi_next; | |
1190 | c->bi_next = NULL; | |
1191 | __split_bio(md, c); | |
1192 | c = n; | |
1193 | } | |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * Swap in a new table (destroying old one). | |
1198 | */ | |
1199 | int dm_swap_table(struct mapped_device *md, struct dm_table *table) | |
1200 | { | |
93c534ae | 1201 | int r = -EINVAL; |
1da177e4 | 1202 | |
2ca3310e | 1203 | down(&md->suspend_lock); |
1da177e4 LT |
1204 | |
1205 | /* device must be suspended */ | |
cf222b37 | 1206 | if (!dm_suspended(md)) |
93c534ae | 1207 | goto out; |
1da177e4 LT |
1208 | |
1209 | __unbind(md); | |
1210 | r = __bind(md, table); | |
1da177e4 | 1211 | |
93c534ae | 1212 | out: |
2ca3310e | 1213 | up(&md->suspend_lock); |
93c534ae | 1214 | return r; |
1da177e4 LT |
1215 | } |
1216 | ||
1217 | /* | |
1218 | * Functions to lock and unlock any filesystem running on the | |
1219 | * device. | |
1220 | */ | |
2ca3310e | 1221 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 1222 | { |
e39e2e95 | 1223 | int r; |
1da177e4 LT |
1224 | |
1225 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 1226 | |
e39e2e95 | 1227 | md->frozen_sb = freeze_bdev(md->suspended_bdev); |
dfbe03f6 | 1228 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 1229 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
1230 | md->frozen_sb = NULL; |
1231 | return r; | |
dfbe03f6 AK |
1232 | } |
1233 | ||
aa8d7c2f AK |
1234 | set_bit(DMF_FROZEN, &md->flags); |
1235 | ||
1da177e4 | 1236 | /* don't bdput right now, we don't want the bdev |
e39e2e95 | 1237 | * to go away while it is locked. |
1da177e4 LT |
1238 | */ |
1239 | return 0; | |
1240 | } | |
1241 | ||
2ca3310e | 1242 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 1243 | { |
aa8d7c2f AK |
1244 | if (!test_bit(DMF_FROZEN, &md->flags)) |
1245 | return; | |
1246 | ||
e39e2e95 | 1247 | thaw_bdev(md->suspended_bdev, md->frozen_sb); |
1da177e4 | 1248 | md->frozen_sb = NULL; |
aa8d7c2f | 1249 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
1250 | } |
1251 | ||
1252 | /* | |
1253 | * We need to be able to change a mapping table under a mounted | |
1254 | * filesystem. For example we might want to move some data in | |
1255 | * the background. Before the table can be swapped with | |
1256 | * dm_bind_table, dm_suspend must be called to flush any in | |
1257 | * flight bios and ensure that any further io gets deferred. | |
1258 | */ | |
aa8d7c2f | 1259 | int dm_suspend(struct mapped_device *md, int do_lockfs) |
1da177e4 | 1260 | { |
2ca3310e | 1261 | struct dm_table *map = NULL; |
1da177e4 | 1262 | DECLARE_WAITQUEUE(wait, current); |
1ecac7fd | 1263 | struct bio *def; |
cf222b37 | 1264 | int r = -EINVAL; |
1da177e4 | 1265 | |
2ca3310e AK |
1266 | down(&md->suspend_lock); |
1267 | ||
1268 | if (dm_suspended(md)) | |
1269 | goto out; | |
1da177e4 LT |
1270 | |
1271 | map = dm_get_table(md); | |
1da177e4 | 1272 | |
cf222b37 AK |
1273 | /* This does not get reverted if there's an error later. */ |
1274 | dm_table_presuspend_targets(map); | |
1275 | ||
e39e2e95 AK |
1276 | md->suspended_bdev = bdget_disk(md->disk, 0); |
1277 | if (!md->suspended_bdev) { | |
1278 | DMWARN("bdget failed in dm_suspend"); | |
1279 | r = -ENOMEM; | |
1280 | goto out; | |
1281 | } | |
1282 | ||
cf222b37 | 1283 | /* Flush I/O to the device. */ |
aa8d7c2f AK |
1284 | if (do_lockfs) { |
1285 | r = lock_fs(md); | |
1286 | if (r) | |
1287 | goto out; | |
1288 | } | |
1da177e4 LT |
1289 | |
1290 | /* | |
354e0071 | 1291 | * First we set the BLOCK_IO flag so no more ios will be mapped. |
1da177e4 | 1292 | */ |
2ca3310e AK |
1293 | down_write(&md->io_lock); |
1294 | set_bit(DMF_BLOCK_IO, &md->flags); | |
1da177e4 | 1295 | |
1da177e4 | 1296 | add_wait_queue(&md->wait, &wait); |
2ca3310e | 1297 | up_write(&md->io_lock); |
1da177e4 LT |
1298 | |
1299 | /* unplug */ | |
2ca3310e | 1300 | if (map) |
1da177e4 | 1301 | dm_table_unplug_all(map); |
1da177e4 LT |
1302 | |
1303 | /* | |
1304 | * Then we wait for the already mapped ios to | |
1305 | * complete. | |
1306 | */ | |
1307 | while (1) { | |
1308 | set_current_state(TASK_INTERRUPTIBLE); | |
1309 | ||
1310 | if (!atomic_read(&md->pending) || signal_pending(current)) | |
1311 | break; | |
1312 | ||
1313 | io_schedule(); | |
1314 | } | |
1315 | set_current_state(TASK_RUNNING); | |
1316 | ||
2ca3310e | 1317 | down_write(&md->io_lock); |
1da177e4 LT |
1318 | remove_wait_queue(&md->wait, &wait); |
1319 | ||
1320 | /* were we interrupted ? */ | |
cf222b37 | 1321 | r = -EINTR; |
2ca3310e | 1322 | if (atomic_read(&md->pending)) { |
1ecac7fd JN |
1323 | clear_bit(DMF_BLOCK_IO, &md->flags); |
1324 | def = bio_list_get(&md->deferred); | |
1325 | __flush_deferred_io(md, def); | |
2ca3310e AK |
1326 | up_write(&md->io_lock); |
1327 | unlock_fs(md); | |
2ca3310e AK |
1328 | goto out; |
1329 | } | |
1330 | up_write(&md->io_lock); | |
1da177e4 | 1331 | |
cf222b37 | 1332 | dm_table_postsuspend_targets(map); |
1da177e4 | 1333 | |
2ca3310e | 1334 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 1335 | |
2ca3310e | 1336 | r = 0; |
b84b0287 | 1337 | |
2ca3310e | 1338 | out: |
e39e2e95 AK |
1339 | if (r && md->suspended_bdev) { |
1340 | bdput(md->suspended_bdev); | |
1341 | md->suspended_bdev = NULL; | |
1342 | } | |
1343 | ||
2ca3310e AK |
1344 | dm_table_put(map); |
1345 | up(&md->suspend_lock); | |
cf222b37 | 1346 | return r; |
1da177e4 LT |
1347 | } |
1348 | ||
1349 | int dm_resume(struct mapped_device *md) | |
1350 | { | |
cf222b37 | 1351 | int r = -EINVAL; |
1da177e4 | 1352 | struct bio *def; |
cf222b37 | 1353 | struct dm_table *map = NULL; |
1da177e4 | 1354 | |
2ca3310e AK |
1355 | down(&md->suspend_lock); |
1356 | if (!dm_suspended(md)) | |
cf222b37 | 1357 | goto out; |
cf222b37 AK |
1358 | |
1359 | map = dm_get_table(md); | |
2ca3310e | 1360 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 1361 | goto out; |
1da177e4 LT |
1362 | |
1363 | dm_table_resume_targets(map); | |
2ca3310e AK |
1364 | |
1365 | down_write(&md->io_lock); | |
1da177e4 LT |
1366 | clear_bit(DMF_BLOCK_IO, &md->flags); |
1367 | ||
1368 | def = bio_list_get(&md->deferred); | |
1369 | __flush_deferred_io(md, def); | |
2ca3310e AK |
1370 | up_write(&md->io_lock); |
1371 | ||
1372 | unlock_fs(md); | |
1373 | ||
e39e2e95 AK |
1374 | bdput(md->suspended_bdev); |
1375 | md->suspended_bdev = NULL; | |
1376 | ||
2ca3310e AK |
1377 | clear_bit(DMF_SUSPENDED, &md->flags); |
1378 | ||
1da177e4 | 1379 | dm_table_unplug_all(map); |
1da177e4 | 1380 | |
8560ed6f HR |
1381 | kobject_uevent(&md->disk->kobj, KOBJ_CHANGE); |
1382 | ||
cf222b37 | 1383 | r = 0; |
2ca3310e | 1384 | |
cf222b37 AK |
1385 | out: |
1386 | dm_table_put(map); | |
2ca3310e AK |
1387 | up(&md->suspend_lock); |
1388 | ||
cf222b37 | 1389 | return r; |
1da177e4 LT |
1390 | } |
1391 | ||
1392 | /*----------------------------------------------------------------- | |
1393 | * Event notification. | |
1394 | *---------------------------------------------------------------*/ | |
1395 | uint32_t dm_get_event_nr(struct mapped_device *md) | |
1396 | { | |
1397 | return atomic_read(&md->event_nr); | |
1398 | } | |
1399 | ||
1400 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
1401 | { | |
1402 | return wait_event_interruptible(md->eventq, | |
1403 | (event_nr != atomic_read(&md->event_nr))); | |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * The gendisk is only valid as long as you have a reference | |
1408 | * count on 'md'. | |
1409 | */ | |
1410 | struct gendisk *dm_disk(struct mapped_device *md) | |
1411 | { | |
1412 | return md->disk; | |
1413 | } | |
1414 | ||
1415 | int dm_suspended(struct mapped_device *md) | |
1416 | { | |
1417 | return test_bit(DMF_SUSPENDED, &md->flags); | |
1418 | } | |
1419 | ||
1420 | static struct block_device_operations dm_blk_dops = { | |
1421 | .open = dm_blk_open, | |
1422 | .release = dm_blk_close, | |
aa129a22 | 1423 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 1424 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
1425 | .owner = THIS_MODULE |
1426 | }; | |
1427 | ||
1428 | EXPORT_SYMBOL(dm_get_mapinfo); | |
1429 | ||
1430 | /* | |
1431 | * module hooks | |
1432 | */ | |
1433 | module_init(dm_init); | |
1434 | module_exit(dm_exit); | |
1435 | ||
1436 | module_param(major, uint, 0); | |
1437 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
1438 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
1439 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
1440 | MODULE_LICENSE("GPL"); |