]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
3 | * Copyright (C) 2004 Red Hat, Inc. All rights reserved. | |
4 | * | |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
9 | #include "dm-bio-list.h" | |
10 | ||
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/moduleparam.h> | |
14 | #include <linux/blkpg.h> | |
15 | #include <linux/bio.h> | |
16 | #include <linux/buffer_head.h> | |
17 | #include <linux/mempool.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/idr.h> | |
20 | ||
21 | static const char *_name = DM_NAME; | |
22 | ||
23 | static unsigned int major = 0; | |
24 | static unsigned int _major = 0; | |
25 | ||
26 | /* | |
27 | * One of these is allocated per bio. | |
28 | */ | |
29 | struct dm_io { | |
30 | struct mapped_device *md; | |
31 | int error; | |
32 | struct bio *bio; | |
33 | atomic_t io_count; | |
34 | }; | |
35 | ||
36 | /* | |
37 | * One of these is allocated per target within a bio. Hopefully | |
38 | * this will be simplified out one day. | |
39 | */ | |
40 | struct target_io { | |
41 | struct dm_io *io; | |
42 | struct dm_target *ti; | |
43 | union map_info info; | |
44 | }; | |
45 | ||
46 | union map_info *dm_get_mapinfo(struct bio *bio) | |
47 | { | |
48 | if (bio && bio->bi_private) | |
49 | return &((struct target_io *)bio->bi_private)->info; | |
50 | return NULL; | |
51 | } | |
52 | ||
53 | /* | |
54 | * Bits for the md->flags field. | |
55 | */ | |
56 | #define DMF_BLOCK_IO 0 | |
57 | #define DMF_SUSPENDED 1 | |
58 | #define DMF_FS_LOCKED 2 | |
59 | ||
60 | struct mapped_device { | |
61 | struct rw_semaphore lock; | |
62 | rwlock_t map_lock; | |
63 | atomic_t holders; | |
64 | ||
65 | unsigned long flags; | |
66 | ||
67 | request_queue_t *queue; | |
68 | struct gendisk *disk; | |
69 | ||
70 | void *interface_ptr; | |
71 | ||
72 | /* | |
73 | * A list of ios that arrived while we were suspended. | |
74 | */ | |
75 | atomic_t pending; | |
76 | wait_queue_head_t wait; | |
77 | struct bio_list deferred; | |
78 | ||
79 | /* | |
80 | * The current mapping. | |
81 | */ | |
82 | struct dm_table *map; | |
83 | ||
84 | /* | |
85 | * io objects are allocated from here. | |
86 | */ | |
87 | mempool_t *io_pool; | |
88 | mempool_t *tio_pool; | |
89 | ||
90 | /* | |
91 | * Event handling. | |
92 | */ | |
93 | atomic_t event_nr; | |
94 | wait_queue_head_t eventq; | |
95 | ||
96 | /* | |
97 | * freeze/thaw support require holding onto a super block | |
98 | */ | |
99 | struct super_block *frozen_sb; | |
d1782a3b | 100 | struct block_device *frozen_bdev; |
1da177e4 LT |
101 | }; |
102 | ||
103 | #define MIN_IOS 256 | |
104 | static kmem_cache_t *_io_cache; | |
105 | static kmem_cache_t *_tio_cache; | |
106 | ||
107 | static struct bio_set *dm_set; | |
108 | ||
109 | static int __init local_init(void) | |
110 | { | |
111 | int r; | |
112 | ||
113 | dm_set = bioset_create(16, 16, 4); | |
114 | if (!dm_set) | |
115 | return -ENOMEM; | |
116 | ||
117 | /* allocate a slab for the dm_ios */ | |
118 | _io_cache = kmem_cache_create("dm_io", | |
119 | sizeof(struct dm_io), 0, 0, NULL, NULL); | |
120 | if (!_io_cache) | |
121 | return -ENOMEM; | |
122 | ||
123 | /* allocate a slab for the target ios */ | |
124 | _tio_cache = kmem_cache_create("dm_tio", sizeof(struct target_io), | |
125 | 0, 0, NULL, NULL); | |
126 | if (!_tio_cache) { | |
127 | kmem_cache_destroy(_io_cache); | |
128 | return -ENOMEM; | |
129 | } | |
130 | ||
131 | _major = major; | |
132 | r = register_blkdev(_major, _name); | |
133 | if (r < 0) { | |
134 | kmem_cache_destroy(_tio_cache); | |
135 | kmem_cache_destroy(_io_cache); | |
136 | return r; | |
137 | } | |
138 | ||
139 | if (!_major) | |
140 | _major = r; | |
141 | ||
142 | return 0; | |
143 | } | |
144 | ||
145 | static void local_exit(void) | |
146 | { | |
147 | kmem_cache_destroy(_tio_cache); | |
148 | kmem_cache_destroy(_io_cache); | |
149 | ||
150 | bioset_free(dm_set); | |
151 | ||
152 | if (unregister_blkdev(_major, _name) < 0) | |
153 | DMERR("devfs_unregister_blkdev failed"); | |
154 | ||
155 | _major = 0; | |
156 | ||
157 | DMINFO("cleaned up"); | |
158 | } | |
159 | ||
160 | int (*_inits[])(void) __initdata = { | |
161 | local_init, | |
162 | dm_target_init, | |
163 | dm_linear_init, | |
164 | dm_stripe_init, | |
165 | dm_interface_init, | |
166 | }; | |
167 | ||
168 | void (*_exits[])(void) = { | |
169 | local_exit, | |
170 | dm_target_exit, | |
171 | dm_linear_exit, | |
172 | dm_stripe_exit, | |
173 | dm_interface_exit, | |
174 | }; | |
175 | ||
176 | static int __init dm_init(void) | |
177 | { | |
178 | const int count = ARRAY_SIZE(_inits); | |
179 | ||
180 | int r, i; | |
181 | ||
182 | for (i = 0; i < count; i++) { | |
183 | r = _inits[i](); | |
184 | if (r) | |
185 | goto bad; | |
186 | } | |
187 | ||
188 | return 0; | |
189 | ||
190 | bad: | |
191 | while (i--) | |
192 | _exits[i](); | |
193 | ||
194 | return r; | |
195 | } | |
196 | ||
197 | static void __exit dm_exit(void) | |
198 | { | |
199 | int i = ARRAY_SIZE(_exits); | |
200 | ||
201 | while (i--) | |
202 | _exits[i](); | |
203 | } | |
204 | ||
205 | /* | |
206 | * Block device functions | |
207 | */ | |
208 | static int dm_blk_open(struct inode *inode, struct file *file) | |
209 | { | |
210 | struct mapped_device *md; | |
211 | ||
212 | md = inode->i_bdev->bd_disk->private_data; | |
213 | dm_get(md); | |
214 | return 0; | |
215 | } | |
216 | ||
217 | static int dm_blk_close(struct inode *inode, struct file *file) | |
218 | { | |
219 | struct mapped_device *md; | |
220 | ||
221 | md = inode->i_bdev->bd_disk->private_data; | |
222 | dm_put(md); | |
223 | return 0; | |
224 | } | |
225 | ||
226 | static inline struct dm_io *alloc_io(struct mapped_device *md) | |
227 | { | |
228 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
229 | } | |
230 | ||
231 | static inline void free_io(struct mapped_device *md, struct dm_io *io) | |
232 | { | |
233 | mempool_free(io, md->io_pool); | |
234 | } | |
235 | ||
236 | static inline struct target_io *alloc_tio(struct mapped_device *md) | |
237 | { | |
238 | return mempool_alloc(md->tio_pool, GFP_NOIO); | |
239 | } | |
240 | ||
241 | static inline void free_tio(struct mapped_device *md, struct target_io *tio) | |
242 | { | |
243 | mempool_free(tio, md->tio_pool); | |
244 | } | |
245 | ||
246 | /* | |
247 | * Add the bio to the list of deferred io. | |
248 | */ | |
249 | static int queue_io(struct mapped_device *md, struct bio *bio) | |
250 | { | |
251 | down_write(&md->lock); | |
252 | ||
253 | if (!test_bit(DMF_BLOCK_IO, &md->flags)) { | |
254 | up_write(&md->lock); | |
255 | return 1; | |
256 | } | |
257 | ||
258 | bio_list_add(&md->deferred, bio); | |
259 | ||
260 | up_write(&md->lock); | |
261 | return 0; /* deferred successfully */ | |
262 | } | |
263 | ||
264 | /* | |
265 | * Everyone (including functions in this file), should use this | |
266 | * function to access the md->map field, and make sure they call | |
267 | * dm_table_put() when finished. | |
268 | */ | |
269 | struct dm_table *dm_get_table(struct mapped_device *md) | |
270 | { | |
271 | struct dm_table *t; | |
272 | ||
273 | read_lock(&md->map_lock); | |
274 | t = md->map; | |
275 | if (t) | |
276 | dm_table_get(t); | |
277 | read_unlock(&md->map_lock); | |
278 | ||
279 | return t; | |
280 | } | |
281 | ||
282 | /*----------------------------------------------------------------- | |
283 | * CRUD START: | |
284 | * A more elegant soln is in the works that uses the queue | |
285 | * merge fn, unfortunately there are a couple of changes to | |
286 | * the block layer that I want to make for this. So in the | |
287 | * interests of getting something for people to use I give | |
288 | * you this clearly demarcated crap. | |
289 | *---------------------------------------------------------------*/ | |
290 | ||
291 | /* | |
292 | * Decrements the number of outstanding ios that a bio has been | |
293 | * cloned into, completing the original io if necc. | |
294 | */ | |
295 | static inline void dec_pending(struct dm_io *io, int error) | |
296 | { | |
297 | if (error) | |
298 | io->error = error; | |
299 | ||
300 | if (atomic_dec_and_test(&io->io_count)) { | |
301 | if (atomic_dec_and_test(&io->md->pending)) | |
302 | /* nudge anyone waiting on suspend queue */ | |
303 | wake_up(&io->md->wait); | |
304 | ||
305 | bio_endio(io->bio, io->bio->bi_size, io->error); | |
306 | free_io(io->md, io); | |
307 | } | |
308 | } | |
309 | ||
310 | static int clone_endio(struct bio *bio, unsigned int done, int error) | |
311 | { | |
312 | int r = 0; | |
313 | struct target_io *tio = bio->bi_private; | |
314 | struct dm_io *io = tio->io; | |
315 | dm_endio_fn endio = tio->ti->type->end_io; | |
316 | ||
317 | if (bio->bi_size) | |
318 | return 1; | |
319 | ||
320 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) | |
321 | error = -EIO; | |
322 | ||
323 | if (endio) { | |
324 | r = endio(tio->ti, bio, error, &tio->info); | |
325 | if (r < 0) | |
326 | error = r; | |
327 | ||
328 | else if (r > 0) | |
329 | /* the target wants another shot at the io */ | |
330 | return 1; | |
331 | } | |
332 | ||
333 | free_tio(io->md, tio); | |
334 | dec_pending(io, error); | |
335 | bio_put(bio); | |
336 | return r; | |
337 | } | |
338 | ||
339 | static sector_t max_io_len(struct mapped_device *md, | |
340 | sector_t sector, struct dm_target *ti) | |
341 | { | |
342 | sector_t offset = sector - ti->begin; | |
343 | sector_t len = ti->len - offset; | |
344 | ||
345 | /* | |
346 | * Does the target need to split even further ? | |
347 | */ | |
348 | if (ti->split_io) { | |
349 | sector_t boundary; | |
350 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) | |
351 | - offset; | |
352 | if (len > boundary) | |
353 | len = boundary; | |
354 | } | |
355 | ||
356 | return len; | |
357 | } | |
358 | ||
359 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
360 | struct target_io *tio) | |
361 | { | |
362 | int r; | |
363 | ||
364 | /* | |
365 | * Sanity checks. | |
366 | */ | |
367 | BUG_ON(!clone->bi_size); | |
368 | ||
369 | clone->bi_end_io = clone_endio; | |
370 | clone->bi_private = tio; | |
371 | ||
372 | /* | |
373 | * Map the clone. If r == 0 we don't need to do | |
374 | * anything, the target has assumed ownership of | |
375 | * this io. | |
376 | */ | |
377 | atomic_inc(&tio->io->io_count); | |
378 | r = ti->type->map(ti, clone, &tio->info); | |
379 | if (r > 0) | |
380 | /* the bio has been remapped so dispatch it */ | |
381 | generic_make_request(clone); | |
382 | ||
383 | else if (r < 0) { | |
384 | /* error the io and bail out */ | |
385 | struct dm_io *io = tio->io; | |
386 | free_tio(tio->io->md, tio); | |
387 | dec_pending(io, -EIO); | |
388 | bio_put(clone); | |
389 | } | |
390 | } | |
391 | ||
392 | struct clone_info { | |
393 | struct mapped_device *md; | |
394 | struct dm_table *map; | |
395 | struct bio *bio; | |
396 | struct dm_io *io; | |
397 | sector_t sector; | |
398 | sector_t sector_count; | |
399 | unsigned short idx; | |
400 | }; | |
401 | ||
402 | /* | |
403 | * Creates a little bio that is just does part of a bvec. | |
404 | */ | |
405 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
406 | unsigned short idx, unsigned int offset, | |
407 | unsigned int len) | |
408 | { | |
409 | struct bio *clone; | |
410 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
411 | ||
412 | clone = bio_alloc_bioset(GFP_NOIO, 1, dm_set); | |
413 | *clone->bi_io_vec = *bv; | |
414 | ||
415 | clone->bi_sector = sector; | |
416 | clone->bi_bdev = bio->bi_bdev; | |
417 | clone->bi_rw = bio->bi_rw; | |
418 | clone->bi_vcnt = 1; | |
419 | clone->bi_size = to_bytes(len); | |
420 | clone->bi_io_vec->bv_offset = offset; | |
421 | clone->bi_io_vec->bv_len = clone->bi_size; | |
422 | ||
423 | return clone; | |
424 | } | |
425 | ||
426 | /* | |
427 | * Creates a bio that consists of range of complete bvecs. | |
428 | */ | |
429 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
430 | unsigned short idx, unsigned short bv_count, | |
431 | unsigned int len) | |
432 | { | |
433 | struct bio *clone; | |
434 | ||
435 | clone = bio_clone(bio, GFP_NOIO); | |
436 | clone->bi_sector = sector; | |
437 | clone->bi_idx = idx; | |
438 | clone->bi_vcnt = idx + bv_count; | |
439 | clone->bi_size = to_bytes(len); | |
440 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
441 | ||
442 | return clone; | |
443 | } | |
444 | ||
445 | static void __clone_and_map(struct clone_info *ci) | |
446 | { | |
447 | struct bio *clone, *bio = ci->bio; | |
448 | struct dm_target *ti = dm_table_find_target(ci->map, ci->sector); | |
449 | sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti); | |
450 | struct target_io *tio; | |
451 | ||
452 | /* | |
453 | * Allocate a target io object. | |
454 | */ | |
455 | tio = alloc_tio(ci->md); | |
456 | tio->io = ci->io; | |
457 | tio->ti = ti; | |
458 | memset(&tio->info, 0, sizeof(tio->info)); | |
459 | ||
460 | if (ci->sector_count <= max) { | |
461 | /* | |
462 | * Optimise for the simple case where we can do all of | |
463 | * the remaining io with a single clone. | |
464 | */ | |
465 | clone = clone_bio(bio, ci->sector, ci->idx, | |
466 | bio->bi_vcnt - ci->idx, ci->sector_count); | |
467 | __map_bio(ti, clone, tio); | |
468 | ci->sector_count = 0; | |
469 | ||
470 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
471 | /* | |
472 | * There are some bvecs that don't span targets. | |
473 | * Do as many of these as possible. | |
474 | */ | |
475 | int i; | |
476 | sector_t remaining = max; | |
477 | sector_t bv_len; | |
478 | ||
479 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
480 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
481 | ||
482 | if (bv_len > remaining) | |
483 | break; | |
484 | ||
485 | remaining -= bv_len; | |
486 | len += bv_len; | |
487 | } | |
488 | ||
489 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len); | |
490 | __map_bio(ti, clone, tio); | |
491 | ||
492 | ci->sector += len; | |
493 | ci->sector_count -= len; | |
494 | ci->idx = i; | |
495 | ||
496 | } else { | |
497 | /* | |
498 | * Create two copy bios to deal with io that has | |
499 | * been split across a target. | |
500 | */ | |
501 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
502 | ||
503 | clone = split_bvec(bio, ci->sector, ci->idx, | |
504 | bv->bv_offset, max); | |
505 | __map_bio(ti, clone, tio); | |
506 | ||
507 | ci->sector += max; | |
508 | ci->sector_count -= max; | |
509 | ti = dm_table_find_target(ci->map, ci->sector); | |
510 | ||
511 | len = to_sector(bv->bv_len) - max; | |
512 | clone = split_bvec(bio, ci->sector, ci->idx, | |
513 | bv->bv_offset + to_bytes(max), len); | |
514 | tio = alloc_tio(ci->md); | |
515 | tio->io = ci->io; | |
516 | tio->ti = ti; | |
517 | memset(&tio->info, 0, sizeof(tio->info)); | |
518 | __map_bio(ti, clone, tio); | |
519 | ||
520 | ci->sector += len; | |
521 | ci->sector_count -= len; | |
522 | ci->idx++; | |
523 | } | |
524 | } | |
525 | ||
526 | /* | |
527 | * Split the bio into several clones. | |
528 | */ | |
529 | static void __split_bio(struct mapped_device *md, struct bio *bio) | |
530 | { | |
531 | struct clone_info ci; | |
532 | ||
533 | ci.map = dm_get_table(md); | |
534 | if (!ci.map) { | |
535 | bio_io_error(bio, bio->bi_size); | |
536 | return; | |
537 | } | |
538 | ||
539 | ci.md = md; | |
540 | ci.bio = bio; | |
541 | ci.io = alloc_io(md); | |
542 | ci.io->error = 0; | |
543 | atomic_set(&ci.io->io_count, 1); | |
544 | ci.io->bio = bio; | |
545 | ci.io->md = md; | |
546 | ci.sector = bio->bi_sector; | |
547 | ci.sector_count = bio_sectors(bio); | |
548 | ci.idx = bio->bi_idx; | |
549 | ||
550 | atomic_inc(&md->pending); | |
551 | while (ci.sector_count) | |
552 | __clone_and_map(&ci); | |
553 | ||
554 | /* drop the extra reference count */ | |
555 | dec_pending(ci.io, 0); | |
556 | dm_table_put(ci.map); | |
557 | } | |
558 | /*----------------------------------------------------------------- | |
559 | * CRUD END | |
560 | *---------------------------------------------------------------*/ | |
561 | ||
562 | /* | |
563 | * The request function that just remaps the bio built up by | |
564 | * dm_merge_bvec. | |
565 | */ | |
566 | static int dm_request(request_queue_t *q, struct bio *bio) | |
567 | { | |
568 | int r; | |
569 | struct mapped_device *md = q->queuedata; | |
570 | ||
571 | down_read(&md->lock); | |
572 | ||
573 | /* | |
574 | * If we're suspended we have to queue | |
575 | * this io for later. | |
576 | */ | |
577 | while (test_bit(DMF_BLOCK_IO, &md->flags)) { | |
578 | up_read(&md->lock); | |
579 | ||
580 | if (bio_rw(bio) == READA) { | |
581 | bio_io_error(bio, bio->bi_size); | |
582 | return 0; | |
583 | } | |
584 | ||
585 | r = queue_io(md, bio); | |
586 | if (r < 0) { | |
587 | bio_io_error(bio, bio->bi_size); | |
588 | return 0; | |
589 | ||
590 | } else if (r == 0) | |
591 | return 0; /* deferred successfully */ | |
592 | ||
593 | /* | |
594 | * We're in a while loop, because someone could suspend | |
595 | * before we get to the following read lock. | |
596 | */ | |
597 | down_read(&md->lock); | |
598 | } | |
599 | ||
600 | __split_bio(md, bio); | |
601 | up_read(&md->lock); | |
602 | return 0; | |
603 | } | |
604 | ||
605 | static int dm_flush_all(request_queue_t *q, struct gendisk *disk, | |
606 | sector_t *error_sector) | |
607 | { | |
608 | struct mapped_device *md = q->queuedata; | |
609 | struct dm_table *map = dm_get_table(md); | |
610 | int ret = -ENXIO; | |
611 | ||
612 | if (map) { | |
613 | ret = dm_table_flush_all(md->map); | |
614 | dm_table_put(map); | |
615 | } | |
616 | ||
617 | return ret; | |
618 | } | |
619 | ||
620 | static void dm_unplug_all(request_queue_t *q) | |
621 | { | |
622 | struct mapped_device *md = q->queuedata; | |
623 | struct dm_table *map = dm_get_table(md); | |
624 | ||
625 | if (map) { | |
626 | dm_table_unplug_all(map); | |
627 | dm_table_put(map); | |
628 | } | |
629 | } | |
630 | ||
631 | static int dm_any_congested(void *congested_data, int bdi_bits) | |
632 | { | |
633 | int r; | |
634 | struct mapped_device *md = (struct mapped_device *) congested_data; | |
635 | struct dm_table *map = dm_get_table(md); | |
636 | ||
637 | if (!map || test_bit(DMF_BLOCK_IO, &md->flags)) | |
638 | r = bdi_bits; | |
639 | else | |
640 | r = dm_table_any_congested(map, bdi_bits); | |
641 | ||
642 | dm_table_put(map); | |
643 | return r; | |
644 | } | |
645 | ||
646 | /*----------------------------------------------------------------- | |
647 | * An IDR is used to keep track of allocated minor numbers. | |
648 | *---------------------------------------------------------------*/ | |
649 | static DECLARE_MUTEX(_minor_lock); | |
650 | static DEFINE_IDR(_minor_idr); | |
651 | ||
652 | static void free_minor(unsigned int minor) | |
653 | { | |
654 | down(&_minor_lock); | |
655 | idr_remove(&_minor_idr, minor); | |
656 | up(&_minor_lock); | |
657 | } | |
658 | ||
659 | /* | |
660 | * See if the device with a specific minor # is free. | |
661 | */ | |
662 | static int specific_minor(struct mapped_device *md, unsigned int minor) | |
663 | { | |
664 | int r, m; | |
665 | ||
666 | if (minor >= (1 << MINORBITS)) | |
667 | return -EINVAL; | |
668 | ||
669 | down(&_minor_lock); | |
670 | ||
671 | if (idr_find(&_minor_idr, minor)) { | |
672 | r = -EBUSY; | |
673 | goto out; | |
674 | } | |
675 | ||
676 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); | |
677 | if (!r) { | |
678 | r = -ENOMEM; | |
679 | goto out; | |
680 | } | |
681 | ||
682 | r = idr_get_new_above(&_minor_idr, md, minor, &m); | |
683 | if (r) { | |
684 | goto out; | |
685 | } | |
686 | ||
687 | if (m != minor) { | |
688 | idr_remove(&_minor_idr, m); | |
689 | r = -EBUSY; | |
690 | goto out; | |
691 | } | |
692 | ||
693 | out: | |
694 | up(&_minor_lock); | |
695 | return r; | |
696 | } | |
697 | ||
698 | static int next_free_minor(struct mapped_device *md, unsigned int *minor) | |
699 | { | |
700 | int r; | |
701 | unsigned int m; | |
702 | ||
703 | down(&_minor_lock); | |
704 | ||
705 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); | |
706 | if (!r) { | |
707 | r = -ENOMEM; | |
708 | goto out; | |
709 | } | |
710 | ||
711 | r = idr_get_new(&_minor_idr, md, &m); | |
712 | if (r) { | |
713 | goto out; | |
714 | } | |
715 | ||
716 | if (m >= (1 << MINORBITS)) { | |
717 | idr_remove(&_minor_idr, m); | |
718 | r = -ENOSPC; | |
719 | goto out; | |
720 | } | |
721 | ||
722 | *minor = m; | |
723 | ||
724 | out: | |
725 | up(&_minor_lock); | |
726 | return r; | |
727 | } | |
728 | ||
729 | static struct block_device_operations dm_blk_dops; | |
730 | ||
731 | /* | |
732 | * Allocate and initialise a blank device with a given minor. | |
733 | */ | |
734 | static struct mapped_device *alloc_dev(unsigned int minor, int persistent) | |
735 | { | |
736 | int r; | |
737 | struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL); | |
738 | ||
739 | if (!md) { | |
740 | DMWARN("unable to allocate device, out of memory."); | |
741 | return NULL; | |
742 | } | |
743 | ||
744 | /* get a minor number for the dev */ | |
745 | r = persistent ? specific_minor(md, minor) : next_free_minor(md, &minor); | |
746 | if (r < 0) | |
747 | goto bad1; | |
748 | ||
749 | memset(md, 0, sizeof(*md)); | |
750 | init_rwsem(&md->lock); | |
751 | rwlock_init(&md->map_lock); | |
752 | atomic_set(&md->holders, 1); | |
753 | atomic_set(&md->event_nr, 0); | |
754 | ||
755 | md->queue = blk_alloc_queue(GFP_KERNEL); | |
756 | if (!md->queue) | |
757 | goto bad1; | |
758 | ||
759 | md->queue->queuedata = md; | |
760 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
761 | md->queue->backing_dev_info.congested_data = md; | |
762 | blk_queue_make_request(md->queue, dm_request); | |
763 | md->queue->unplug_fn = dm_unplug_all; | |
764 | md->queue->issue_flush_fn = dm_flush_all; | |
765 | ||
766 | md->io_pool = mempool_create(MIN_IOS, mempool_alloc_slab, | |
767 | mempool_free_slab, _io_cache); | |
768 | if (!md->io_pool) | |
769 | goto bad2; | |
770 | ||
771 | md->tio_pool = mempool_create(MIN_IOS, mempool_alloc_slab, | |
772 | mempool_free_slab, _tio_cache); | |
773 | if (!md->tio_pool) | |
774 | goto bad3; | |
775 | ||
776 | md->disk = alloc_disk(1); | |
777 | if (!md->disk) | |
778 | goto bad4; | |
779 | ||
780 | md->disk->major = _major; | |
781 | md->disk->first_minor = minor; | |
782 | md->disk->fops = &dm_blk_dops; | |
783 | md->disk->queue = md->queue; | |
784 | md->disk->private_data = md; | |
785 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
786 | add_disk(md->disk); | |
787 | ||
788 | atomic_set(&md->pending, 0); | |
789 | init_waitqueue_head(&md->wait); | |
790 | init_waitqueue_head(&md->eventq); | |
791 | ||
792 | return md; | |
793 | ||
794 | bad4: | |
795 | mempool_destroy(md->tio_pool); | |
796 | bad3: | |
797 | mempool_destroy(md->io_pool); | |
798 | bad2: | |
799 | blk_put_queue(md->queue); | |
800 | free_minor(minor); | |
801 | bad1: | |
802 | kfree(md); | |
803 | return NULL; | |
804 | } | |
805 | ||
806 | static void free_dev(struct mapped_device *md) | |
807 | { | |
808 | free_minor(md->disk->first_minor); | |
809 | mempool_destroy(md->tio_pool); | |
810 | mempool_destroy(md->io_pool); | |
811 | del_gendisk(md->disk); | |
812 | put_disk(md->disk); | |
813 | blk_put_queue(md->queue); | |
814 | kfree(md); | |
815 | } | |
816 | ||
817 | /* | |
818 | * Bind a table to the device. | |
819 | */ | |
820 | static void event_callback(void *context) | |
821 | { | |
822 | struct mapped_device *md = (struct mapped_device *) context; | |
823 | ||
824 | atomic_inc(&md->event_nr); | |
825 | wake_up(&md->eventq); | |
826 | } | |
827 | ||
828 | static void __set_size(struct gendisk *disk, sector_t size) | |
829 | { | |
830 | struct block_device *bdev; | |
831 | ||
832 | set_capacity(disk, size); | |
833 | bdev = bdget_disk(disk, 0); | |
834 | if (bdev) { | |
835 | down(&bdev->bd_inode->i_sem); | |
836 | i_size_write(bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); | |
837 | up(&bdev->bd_inode->i_sem); | |
838 | bdput(bdev); | |
839 | } | |
840 | } | |
841 | ||
842 | static int __bind(struct mapped_device *md, struct dm_table *t) | |
843 | { | |
844 | request_queue_t *q = md->queue; | |
845 | sector_t size; | |
846 | ||
847 | size = dm_table_get_size(t); | |
848 | __set_size(md->disk, size); | |
849 | if (size == 0) | |
850 | return 0; | |
851 | ||
852 | write_lock(&md->map_lock); | |
853 | md->map = t; | |
854 | write_unlock(&md->map_lock); | |
855 | ||
856 | dm_table_get(t); | |
857 | dm_table_event_callback(md->map, event_callback, md); | |
858 | dm_table_set_restrictions(t, q); | |
859 | return 0; | |
860 | } | |
861 | ||
862 | static void __unbind(struct mapped_device *md) | |
863 | { | |
864 | struct dm_table *map = md->map; | |
865 | ||
866 | if (!map) | |
867 | return; | |
868 | ||
869 | dm_table_event_callback(map, NULL, NULL); | |
870 | write_lock(&md->map_lock); | |
871 | md->map = NULL; | |
872 | write_unlock(&md->map_lock); | |
873 | dm_table_put(map); | |
874 | } | |
875 | ||
876 | /* | |
877 | * Constructor for a new device. | |
878 | */ | |
879 | static int create_aux(unsigned int minor, int persistent, | |
880 | struct mapped_device **result) | |
881 | { | |
882 | struct mapped_device *md; | |
883 | ||
884 | md = alloc_dev(minor, persistent); | |
885 | if (!md) | |
886 | return -ENXIO; | |
887 | ||
888 | *result = md; | |
889 | return 0; | |
890 | } | |
891 | ||
892 | int dm_create(struct mapped_device **result) | |
893 | { | |
894 | return create_aux(0, 0, result); | |
895 | } | |
896 | ||
897 | int dm_create_with_minor(unsigned int minor, struct mapped_device **result) | |
898 | { | |
899 | return create_aux(minor, 1, result); | |
900 | } | |
901 | ||
902 | void *dm_get_mdptr(dev_t dev) | |
903 | { | |
904 | struct mapped_device *md; | |
905 | void *mdptr = NULL; | |
906 | unsigned minor = MINOR(dev); | |
907 | ||
908 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
909 | return NULL; | |
910 | ||
911 | down(&_minor_lock); | |
912 | ||
913 | md = idr_find(&_minor_idr, minor); | |
914 | ||
915 | if (md && (dm_disk(md)->first_minor == minor)) | |
916 | mdptr = md->interface_ptr; | |
917 | ||
918 | up(&_minor_lock); | |
919 | ||
920 | return mdptr; | |
921 | } | |
922 | ||
923 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
924 | { | |
925 | md->interface_ptr = ptr; | |
926 | } | |
927 | ||
928 | void dm_get(struct mapped_device *md) | |
929 | { | |
930 | atomic_inc(&md->holders); | |
931 | } | |
932 | ||
933 | void dm_put(struct mapped_device *md) | |
934 | { | |
935 | struct dm_table *map = dm_get_table(md); | |
936 | ||
937 | if (atomic_dec_and_test(&md->holders)) { | |
938 | if (!test_bit(DMF_SUSPENDED, &md->flags) && map) { | |
939 | dm_table_presuspend_targets(map); | |
940 | dm_table_postsuspend_targets(map); | |
941 | } | |
942 | __unbind(md); | |
943 | free_dev(md); | |
944 | } | |
945 | ||
946 | dm_table_put(map); | |
947 | } | |
948 | ||
949 | /* | |
950 | * Process the deferred bios | |
951 | */ | |
952 | static void __flush_deferred_io(struct mapped_device *md, struct bio *c) | |
953 | { | |
954 | struct bio *n; | |
955 | ||
956 | while (c) { | |
957 | n = c->bi_next; | |
958 | c->bi_next = NULL; | |
959 | __split_bio(md, c); | |
960 | c = n; | |
961 | } | |
962 | } | |
963 | ||
964 | /* | |
965 | * Swap in a new table (destroying old one). | |
966 | */ | |
967 | int dm_swap_table(struct mapped_device *md, struct dm_table *table) | |
968 | { | |
969 | int r; | |
970 | ||
971 | down_write(&md->lock); | |
972 | ||
973 | /* device must be suspended */ | |
974 | if (!test_bit(DMF_SUSPENDED, &md->flags)) { | |
975 | up_write(&md->lock); | |
976 | return -EPERM; | |
977 | } | |
978 | ||
979 | __unbind(md); | |
980 | r = __bind(md, table); | |
981 | if (r) | |
982 | return r; | |
983 | ||
984 | up_write(&md->lock); | |
985 | return 0; | |
986 | } | |
987 | ||
988 | /* | |
989 | * Functions to lock and unlock any filesystem running on the | |
990 | * device. | |
991 | */ | |
992 | static int __lock_fs(struct mapped_device *md) | |
993 | { | |
dfbe03f6 AK |
994 | int error = -ENOMEM; |
995 | ||
1da177e4 LT |
996 | if (test_and_set_bit(DMF_FS_LOCKED, &md->flags)) |
997 | return 0; | |
998 | ||
d1782a3b AK |
999 | md->frozen_bdev = bdget_disk(md->disk, 0); |
1000 | if (!md->frozen_bdev) { | |
1da177e4 | 1001 | DMWARN("bdget failed in __lock_fs"); |
dfbe03f6 | 1002 | goto out; |
1da177e4 LT |
1003 | } |
1004 | ||
1005 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 1006 | |
d1782a3b | 1007 | md->frozen_sb = freeze_bdev(md->frozen_bdev); |
dfbe03f6 AK |
1008 | if (IS_ERR(md->frozen_sb)) { |
1009 | error = PTR_ERR(md->frozen_sb); | |
1010 | goto out_bdput; | |
1011 | } | |
1012 | ||
1da177e4 LT |
1013 | /* don't bdput right now, we don't want the bdev |
1014 | * to go away while it is locked. We'll bdput | |
1015 | * in __unlock_fs | |
1016 | */ | |
1017 | return 0; | |
dfbe03f6 AK |
1018 | |
1019 | out_bdput: | |
1020 | bdput(md->frozen_bdev); | |
1021 | md->frozen_sb = NULL; | |
1022 | md->frozen_bdev = NULL; | |
1023 | out: | |
1024 | clear_bit(DMF_FS_LOCKED, &md->flags); | |
1025 | return error; | |
1da177e4 LT |
1026 | } |
1027 | ||
3dcee806 | 1028 | static void __unlock_fs(struct mapped_device *md) |
1da177e4 | 1029 | { |
1da177e4 | 1030 | if (!test_and_clear_bit(DMF_FS_LOCKED, &md->flags)) |
3dcee806 | 1031 | return; |
1da177e4 | 1032 | |
d1782a3b AK |
1033 | thaw_bdev(md->frozen_bdev, md->frozen_sb); |
1034 | bdput(md->frozen_bdev); | |
1da177e4 | 1035 | |
1da177e4 | 1036 | md->frozen_sb = NULL; |
d1782a3b | 1037 | md->frozen_bdev = NULL; |
1da177e4 LT |
1038 | } |
1039 | ||
1040 | /* | |
1041 | * We need to be able to change a mapping table under a mounted | |
1042 | * filesystem. For example we might want to move some data in | |
1043 | * the background. Before the table can be swapped with | |
1044 | * dm_bind_table, dm_suspend must be called to flush any in | |
1045 | * flight bios and ensure that any further io gets deferred. | |
1046 | */ | |
1047 | int dm_suspend(struct mapped_device *md) | |
1048 | { | |
1049 | struct dm_table *map; | |
1050 | DECLARE_WAITQUEUE(wait, current); | |
1051 | ||
1052 | /* Flush I/O to the device. */ | |
1053 | down_read(&md->lock); | |
1054 | if (test_bit(DMF_BLOCK_IO, &md->flags)) { | |
1055 | up_read(&md->lock); | |
1056 | return -EINVAL; | |
1057 | } | |
1058 | ||
1059 | map = dm_get_table(md); | |
1060 | if (map) | |
1061 | dm_table_presuspend_targets(map); | |
1062 | __lock_fs(md); | |
1063 | ||
1064 | up_read(&md->lock); | |
1065 | ||
1066 | /* | |
1067 | * First we set the BLOCK_IO flag so no more ios will be | |
1068 | * mapped. | |
1069 | */ | |
1070 | down_write(&md->lock); | |
1071 | if (test_bit(DMF_BLOCK_IO, &md->flags)) { | |
1072 | /* | |
1073 | * If we get here we know another thread is | |
1074 | * trying to suspend as well, so we leave the fs | |
1075 | * locked for this thread. | |
1076 | */ | |
1077 | up_write(&md->lock); | |
1078 | return -EINVAL; | |
1079 | } | |
1080 | ||
1081 | set_bit(DMF_BLOCK_IO, &md->flags); | |
1082 | add_wait_queue(&md->wait, &wait); | |
1083 | up_write(&md->lock); | |
1084 | ||
1085 | /* unplug */ | |
1086 | if (map) { | |
1087 | dm_table_unplug_all(map); | |
1088 | dm_table_put(map); | |
1089 | } | |
1090 | ||
1091 | /* | |
1092 | * Then we wait for the already mapped ios to | |
1093 | * complete. | |
1094 | */ | |
1095 | while (1) { | |
1096 | set_current_state(TASK_INTERRUPTIBLE); | |
1097 | ||
1098 | if (!atomic_read(&md->pending) || signal_pending(current)) | |
1099 | break; | |
1100 | ||
1101 | io_schedule(); | |
1102 | } | |
1103 | set_current_state(TASK_RUNNING); | |
1104 | ||
1105 | down_write(&md->lock); | |
1106 | remove_wait_queue(&md->wait, &wait); | |
1107 | ||
1108 | /* were we interrupted ? */ | |
1109 | if (atomic_read(&md->pending)) { | |
1110 | __unlock_fs(md); | |
1111 | clear_bit(DMF_BLOCK_IO, &md->flags); | |
1112 | up_write(&md->lock); | |
1113 | return -EINTR; | |
1114 | } | |
1115 | ||
1116 | set_bit(DMF_SUSPENDED, &md->flags); | |
1117 | ||
1118 | map = dm_get_table(md); | |
1119 | if (map) | |
1120 | dm_table_postsuspend_targets(map); | |
1121 | dm_table_put(map); | |
1122 | up_write(&md->lock); | |
1123 | ||
1124 | return 0; | |
1125 | } | |
1126 | ||
1127 | int dm_resume(struct mapped_device *md) | |
1128 | { | |
1129 | struct bio *def; | |
1130 | struct dm_table *map = dm_get_table(md); | |
1131 | ||
1132 | down_write(&md->lock); | |
1133 | if (!map || | |
1134 | !test_bit(DMF_SUSPENDED, &md->flags) || | |
1135 | !dm_table_get_size(map)) { | |
1136 | up_write(&md->lock); | |
1137 | dm_table_put(map); | |
1138 | return -EINVAL; | |
1139 | } | |
1140 | ||
1141 | dm_table_resume_targets(map); | |
1142 | clear_bit(DMF_SUSPENDED, &md->flags); | |
1143 | clear_bit(DMF_BLOCK_IO, &md->flags); | |
1144 | ||
1145 | def = bio_list_get(&md->deferred); | |
1146 | __flush_deferred_io(md, def); | |
1147 | up_write(&md->lock); | |
1148 | __unlock_fs(md); | |
1149 | dm_table_unplug_all(map); | |
1150 | dm_table_put(map); | |
1151 | ||
1152 | return 0; | |
1153 | } | |
1154 | ||
1155 | /*----------------------------------------------------------------- | |
1156 | * Event notification. | |
1157 | *---------------------------------------------------------------*/ | |
1158 | uint32_t dm_get_event_nr(struct mapped_device *md) | |
1159 | { | |
1160 | return atomic_read(&md->event_nr); | |
1161 | } | |
1162 | ||
1163 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
1164 | { | |
1165 | return wait_event_interruptible(md->eventq, | |
1166 | (event_nr != atomic_read(&md->event_nr))); | |
1167 | } | |
1168 | ||
1169 | /* | |
1170 | * The gendisk is only valid as long as you have a reference | |
1171 | * count on 'md'. | |
1172 | */ | |
1173 | struct gendisk *dm_disk(struct mapped_device *md) | |
1174 | { | |
1175 | return md->disk; | |
1176 | } | |
1177 | ||
1178 | int dm_suspended(struct mapped_device *md) | |
1179 | { | |
1180 | return test_bit(DMF_SUSPENDED, &md->flags); | |
1181 | } | |
1182 | ||
1183 | static struct block_device_operations dm_blk_dops = { | |
1184 | .open = dm_blk_open, | |
1185 | .release = dm_blk_close, | |
1186 | .owner = THIS_MODULE | |
1187 | }; | |
1188 | ||
1189 | EXPORT_SYMBOL(dm_get_mapinfo); | |
1190 | ||
1191 | /* | |
1192 | * module hooks | |
1193 | */ | |
1194 | module_init(dm_init); | |
1195 | module_exit(dm_exit); | |
1196 | ||
1197 | module_param(major, uint, 0); | |
1198 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
1199 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
1200 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
1201 | MODULE_LICENSE("GPL"); |