]>
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" | |
51e5b2bd | 10 | #include "dm-uevent.h" |
1da177e4 LT |
11 | |
12 | #include <linux/init.h> | |
13 | #include <linux/module.h> | |
48c9c27b | 14 | #include <linux/mutex.h> |
1da177e4 LT |
15 | #include <linux/moduleparam.h> |
16 | #include <linux/blkpg.h> | |
17 | #include <linux/bio.h> | |
18 | #include <linux/buffer_head.h> | |
19 | #include <linux/mempool.h> | |
20 | #include <linux/slab.h> | |
21 | #include <linux/idr.h> | |
3ac51e74 | 22 | #include <linux/hdreg.h> |
2056a782 | 23 | #include <linux/blktrace_api.h> |
aa129a22 | 24 | #include <linux/smp_lock.h> |
1da177e4 | 25 | |
72d94861 AK |
26 | #define DM_MSG_PREFIX "core" |
27 | ||
1da177e4 LT |
28 | static const char *_name = DM_NAME; |
29 | ||
30 | static unsigned int major = 0; | |
31 | static unsigned int _major = 0; | |
32 | ||
f32c10b0 | 33 | static DEFINE_SPINLOCK(_minor_lock); |
1da177e4 LT |
34 | /* |
35 | * One of these is allocated per bio. | |
36 | */ | |
37 | struct dm_io { | |
38 | struct mapped_device *md; | |
39 | int error; | |
40 | struct bio *bio; | |
41 | atomic_t io_count; | |
3eaf840e | 42 | unsigned long start_time; |
1da177e4 LT |
43 | }; |
44 | ||
45 | /* | |
46 | * One of these is allocated per target within a bio. Hopefully | |
47 | * this will be simplified out one day. | |
48 | */ | |
028867ac | 49 | struct dm_target_io { |
1da177e4 LT |
50 | struct dm_io *io; |
51 | struct dm_target *ti; | |
52 | union map_info info; | |
53 | }; | |
54 | ||
55 | union map_info *dm_get_mapinfo(struct bio *bio) | |
56 | { | |
17b2f66f | 57 | if (bio && bio->bi_private) |
028867ac | 58 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 59 | return NULL; |
1da177e4 LT |
60 | } |
61 | ||
ba61fdd1 JM |
62 | #define MINOR_ALLOCED ((void *)-1) |
63 | ||
1da177e4 LT |
64 | /* |
65 | * Bits for the md->flags field. | |
66 | */ | |
67 | #define DMF_BLOCK_IO 0 | |
68 | #define DMF_SUSPENDED 1 | |
aa8d7c2f | 69 | #define DMF_FROZEN 2 |
fba9f90e | 70 | #define DMF_FREEING 3 |
5c6bd75d | 71 | #define DMF_DELETING 4 |
2e93ccc1 | 72 | #define DMF_NOFLUSH_SUSPENDING 5 |
1da177e4 LT |
73 | |
74 | struct mapped_device { | |
2ca3310e AK |
75 | struct rw_semaphore io_lock; |
76 | struct semaphore suspend_lock; | |
2e93ccc1 | 77 | spinlock_t pushback_lock; |
1da177e4 LT |
78 | rwlock_t map_lock; |
79 | atomic_t holders; | |
5c6bd75d | 80 | atomic_t open_count; |
1da177e4 LT |
81 | |
82 | unsigned long flags; | |
83 | ||
165125e1 | 84 | struct request_queue *queue; |
1da177e4 | 85 | struct gendisk *disk; |
7e51f257 | 86 | char name[16]; |
1da177e4 LT |
87 | |
88 | void *interface_ptr; | |
89 | ||
90 | /* | |
91 | * A list of ios that arrived while we were suspended. | |
92 | */ | |
93 | atomic_t pending; | |
94 | wait_queue_head_t wait; | |
74859364 | 95 | struct bio_list deferred; |
2e93ccc1 | 96 | struct bio_list pushback; |
1da177e4 LT |
97 | |
98 | /* | |
99 | * The current mapping. | |
100 | */ | |
101 | struct dm_table *map; | |
102 | ||
103 | /* | |
104 | * io objects are allocated from here. | |
105 | */ | |
106 | mempool_t *io_pool; | |
107 | mempool_t *tio_pool; | |
108 | ||
9faf400f SB |
109 | struct bio_set *bs; |
110 | ||
1da177e4 LT |
111 | /* |
112 | * Event handling. | |
113 | */ | |
114 | atomic_t event_nr; | |
115 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
116 | atomic_t uevent_seq; |
117 | struct list_head uevent_list; | |
118 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
119 | |
120 | /* | |
121 | * freeze/thaw support require holding onto a super block | |
122 | */ | |
123 | struct super_block *frozen_sb; | |
e39e2e95 | 124 | struct block_device *suspended_bdev; |
3ac51e74 DW |
125 | |
126 | /* forced geometry settings */ | |
127 | struct hd_geometry geometry; | |
1da177e4 LT |
128 | }; |
129 | ||
130 | #define MIN_IOS 256 | |
e18b890b CL |
131 | static struct kmem_cache *_io_cache; |
132 | static struct kmem_cache *_tio_cache; | |
1da177e4 | 133 | |
1da177e4 LT |
134 | static int __init local_init(void) |
135 | { | |
136 | int r; | |
137 | ||
1da177e4 | 138 | /* allocate a slab for the dm_ios */ |
028867ac | 139 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 LT |
140 | if (!_io_cache) |
141 | return -ENOMEM; | |
142 | ||
143 | /* allocate a slab for the target ios */ | |
028867ac | 144 | _tio_cache = KMEM_CACHE(dm_target_io, 0); |
1da177e4 LT |
145 | if (!_tio_cache) { |
146 | kmem_cache_destroy(_io_cache); | |
147 | return -ENOMEM; | |
148 | } | |
149 | ||
51e5b2bd MA |
150 | r = dm_uevent_init(); |
151 | if (r) { | |
152 | kmem_cache_destroy(_tio_cache); | |
153 | kmem_cache_destroy(_io_cache); | |
154 | return r; | |
155 | } | |
156 | ||
1da177e4 LT |
157 | _major = major; |
158 | r = register_blkdev(_major, _name); | |
159 | if (r < 0) { | |
160 | kmem_cache_destroy(_tio_cache); | |
161 | kmem_cache_destroy(_io_cache); | |
51e5b2bd | 162 | dm_uevent_exit(); |
1da177e4 LT |
163 | return r; |
164 | } | |
165 | ||
166 | if (!_major) | |
167 | _major = r; | |
168 | ||
169 | return 0; | |
170 | } | |
171 | ||
172 | static void local_exit(void) | |
173 | { | |
174 | kmem_cache_destroy(_tio_cache); | |
175 | kmem_cache_destroy(_io_cache); | |
00d59405 | 176 | unregister_blkdev(_major, _name); |
51e5b2bd | 177 | dm_uevent_exit(); |
1da177e4 LT |
178 | |
179 | _major = 0; | |
180 | ||
181 | DMINFO("cleaned up"); | |
182 | } | |
183 | ||
184 | int (*_inits[])(void) __initdata = { | |
185 | local_init, | |
186 | dm_target_init, | |
187 | dm_linear_init, | |
188 | dm_stripe_init, | |
189 | dm_interface_init, | |
190 | }; | |
191 | ||
192 | void (*_exits[])(void) = { | |
193 | local_exit, | |
194 | dm_target_exit, | |
195 | dm_linear_exit, | |
196 | dm_stripe_exit, | |
197 | dm_interface_exit, | |
198 | }; | |
199 | ||
200 | static int __init dm_init(void) | |
201 | { | |
202 | const int count = ARRAY_SIZE(_inits); | |
203 | ||
204 | int r, i; | |
205 | ||
206 | for (i = 0; i < count; i++) { | |
207 | r = _inits[i](); | |
208 | if (r) | |
209 | goto bad; | |
210 | } | |
211 | ||
212 | return 0; | |
213 | ||
214 | bad: | |
215 | while (i--) | |
216 | _exits[i](); | |
217 | ||
218 | return r; | |
219 | } | |
220 | ||
221 | static void __exit dm_exit(void) | |
222 | { | |
223 | int i = ARRAY_SIZE(_exits); | |
224 | ||
225 | while (i--) | |
226 | _exits[i](); | |
227 | } | |
228 | ||
229 | /* | |
230 | * Block device functions | |
231 | */ | |
232 | static int dm_blk_open(struct inode *inode, struct file *file) | |
233 | { | |
234 | struct mapped_device *md; | |
235 | ||
fba9f90e JM |
236 | spin_lock(&_minor_lock); |
237 | ||
1da177e4 | 238 | md = inode->i_bdev->bd_disk->private_data; |
fba9f90e JM |
239 | if (!md) |
240 | goto out; | |
241 | ||
5c6bd75d AK |
242 | if (test_bit(DMF_FREEING, &md->flags) || |
243 | test_bit(DMF_DELETING, &md->flags)) { | |
fba9f90e JM |
244 | md = NULL; |
245 | goto out; | |
246 | } | |
247 | ||
1da177e4 | 248 | dm_get(md); |
5c6bd75d | 249 | atomic_inc(&md->open_count); |
fba9f90e JM |
250 | |
251 | out: | |
252 | spin_unlock(&_minor_lock); | |
253 | ||
254 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
255 | } |
256 | ||
257 | static int dm_blk_close(struct inode *inode, struct file *file) | |
258 | { | |
259 | struct mapped_device *md; | |
260 | ||
261 | md = inode->i_bdev->bd_disk->private_data; | |
5c6bd75d | 262 | atomic_dec(&md->open_count); |
1da177e4 LT |
263 | dm_put(md); |
264 | return 0; | |
265 | } | |
266 | ||
5c6bd75d AK |
267 | int dm_open_count(struct mapped_device *md) |
268 | { | |
269 | return atomic_read(&md->open_count); | |
270 | } | |
271 | ||
272 | /* | |
273 | * Guarantees nothing is using the device before it's deleted. | |
274 | */ | |
275 | int dm_lock_for_deletion(struct mapped_device *md) | |
276 | { | |
277 | int r = 0; | |
278 | ||
279 | spin_lock(&_minor_lock); | |
280 | ||
281 | if (dm_open_count(md)) | |
282 | r = -EBUSY; | |
283 | else | |
284 | set_bit(DMF_DELETING, &md->flags); | |
285 | ||
286 | spin_unlock(&_minor_lock); | |
287 | ||
288 | return r; | |
289 | } | |
290 | ||
3ac51e74 DW |
291 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
292 | { | |
293 | struct mapped_device *md = bdev->bd_disk->private_data; | |
294 | ||
295 | return dm_get_geometry(md, geo); | |
296 | } | |
297 | ||
aa129a22 MB |
298 | static int dm_blk_ioctl(struct inode *inode, struct file *file, |
299 | unsigned int cmd, unsigned long arg) | |
300 | { | |
301 | struct mapped_device *md; | |
302 | struct dm_table *map; | |
303 | struct dm_target *tgt; | |
304 | int r = -ENOTTY; | |
305 | ||
306 | /* We don't really need this lock, but we do need 'inode'. */ | |
307 | unlock_kernel(); | |
308 | ||
309 | md = inode->i_bdev->bd_disk->private_data; | |
310 | ||
311 | map = dm_get_table(md); | |
312 | ||
313 | if (!map || !dm_table_get_size(map)) | |
314 | goto out; | |
315 | ||
316 | /* We only support devices that have a single target */ | |
317 | if (dm_table_get_num_targets(map) != 1) | |
318 | goto out; | |
319 | ||
320 | tgt = dm_table_get_target(map, 0); | |
321 | ||
322 | if (dm_suspended(md)) { | |
323 | r = -EAGAIN; | |
324 | goto out; | |
325 | } | |
326 | ||
327 | if (tgt->type->ioctl) | |
328 | r = tgt->type->ioctl(tgt, inode, file, cmd, arg); | |
329 | ||
330 | out: | |
331 | dm_table_put(map); | |
332 | ||
333 | lock_kernel(); | |
334 | return r; | |
335 | } | |
336 | ||
028867ac | 337 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
338 | { |
339 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
340 | } | |
341 | ||
028867ac | 342 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
343 | { |
344 | mempool_free(io, md->io_pool); | |
345 | } | |
346 | ||
028867ac | 347 | static struct dm_target_io *alloc_tio(struct mapped_device *md) |
1da177e4 LT |
348 | { |
349 | return mempool_alloc(md->tio_pool, GFP_NOIO); | |
350 | } | |
351 | ||
028867ac | 352 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 LT |
353 | { |
354 | mempool_free(tio, md->tio_pool); | |
355 | } | |
356 | ||
3eaf840e JNN |
357 | static void start_io_acct(struct dm_io *io) |
358 | { | |
359 | struct mapped_device *md = io->md; | |
360 | ||
361 | io->start_time = jiffies; | |
362 | ||
363 | preempt_disable(); | |
364 | disk_round_stats(dm_disk(md)); | |
365 | preempt_enable(); | |
366 | dm_disk(md)->in_flight = atomic_inc_return(&md->pending); | |
367 | } | |
368 | ||
369 | static int end_io_acct(struct dm_io *io) | |
370 | { | |
371 | struct mapped_device *md = io->md; | |
372 | struct bio *bio = io->bio; | |
373 | unsigned long duration = jiffies - io->start_time; | |
374 | int pending; | |
375 | int rw = bio_data_dir(bio); | |
376 | ||
377 | preempt_disable(); | |
378 | disk_round_stats(dm_disk(md)); | |
379 | preempt_enable(); | |
380 | dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending); | |
381 | ||
382 | disk_stat_add(dm_disk(md), ticks[rw], duration); | |
383 | ||
384 | return !pending; | |
385 | } | |
386 | ||
1da177e4 LT |
387 | /* |
388 | * Add the bio to the list of deferred io. | |
389 | */ | |
390 | static int queue_io(struct mapped_device *md, struct bio *bio) | |
391 | { | |
2ca3310e | 392 | down_write(&md->io_lock); |
1da177e4 LT |
393 | |
394 | if (!test_bit(DMF_BLOCK_IO, &md->flags)) { | |
2ca3310e | 395 | up_write(&md->io_lock); |
1da177e4 LT |
396 | return 1; |
397 | } | |
398 | ||
399 | bio_list_add(&md->deferred, bio); | |
400 | ||
2ca3310e | 401 | up_write(&md->io_lock); |
1da177e4 LT |
402 | return 0; /* deferred successfully */ |
403 | } | |
404 | ||
405 | /* | |
406 | * Everyone (including functions in this file), should use this | |
407 | * function to access the md->map field, and make sure they call | |
408 | * dm_table_put() when finished. | |
409 | */ | |
410 | struct dm_table *dm_get_table(struct mapped_device *md) | |
411 | { | |
412 | struct dm_table *t; | |
413 | ||
414 | read_lock(&md->map_lock); | |
415 | t = md->map; | |
416 | if (t) | |
417 | dm_table_get(t); | |
418 | read_unlock(&md->map_lock); | |
419 | ||
420 | return t; | |
421 | } | |
422 | ||
3ac51e74 DW |
423 | /* |
424 | * Get the geometry associated with a dm device | |
425 | */ | |
426 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
427 | { | |
428 | *geo = md->geometry; | |
429 | ||
430 | return 0; | |
431 | } | |
432 | ||
433 | /* | |
434 | * Set the geometry of a device. | |
435 | */ | |
436 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
437 | { | |
438 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
439 | ||
440 | if (geo->start > sz) { | |
441 | DMWARN("Start sector is beyond the geometry limits."); | |
442 | return -EINVAL; | |
443 | } | |
444 | ||
445 | md->geometry = *geo; | |
446 | ||
447 | return 0; | |
448 | } | |
449 | ||
1da177e4 LT |
450 | /*----------------------------------------------------------------- |
451 | * CRUD START: | |
452 | * A more elegant soln is in the works that uses the queue | |
453 | * merge fn, unfortunately there are a couple of changes to | |
454 | * the block layer that I want to make for this. So in the | |
455 | * interests of getting something for people to use I give | |
456 | * you this clearly demarcated crap. | |
457 | *---------------------------------------------------------------*/ | |
458 | ||
2e93ccc1 KU |
459 | static int __noflush_suspending(struct mapped_device *md) |
460 | { | |
461 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
462 | } | |
463 | ||
1da177e4 LT |
464 | /* |
465 | * Decrements the number of outstanding ios that a bio has been | |
466 | * cloned into, completing the original io if necc. | |
467 | */ | |
858119e1 | 468 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 469 | { |
2e93ccc1 KU |
470 | unsigned long flags; |
471 | ||
472 | /* Push-back supersedes any I/O errors */ | |
473 | if (error && !(io->error > 0 && __noflush_suspending(io->md))) | |
1da177e4 LT |
474 | io->error = error; |
475 | ||
476 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
477 | if (io->error == DM_ENDIO_REQUEUE) { |
478 | /* | |
479 | * Target requested pushing back the I/O. | |
480 | * This must be handled before the sleeper on | |
481 | * suspend queue merges the pushback list. | |
482 | */ | |
483 | spin_lock_irqsave(&io->md->pushback_lock, flags); | |
484 | if (__noflush_suspending(io->md)) | |
485 | bio_list_add(&io->md->pushback, io->bio); | |
486 | else | |
487 | /* noflush suspend was interrupted. */ | |
488 | io->error = -EIO; | |
489 | spin_unlock_irqrestore(&io->md->pushback_lock, flags); | |
490 | } | |
491 | ||
3eaf840e | 492 | if (end_io_acct(io)) |
1da177e4 LT |
493 | /* nudge anyone waiting on suspend queue */ |
494 | wake_up(&io->md->wait); | |
495 | ||
2e93ccc1 KU |
496 | if (io->error != DM_ENDIO_REQUEUE) { |
497 | blk_add_trace_bio(io->md->queue, io->bio, | |
498 | BLK_TA_COMPLETE); | |
499 | ||
6712ecf8 | 500 | bio_endio(io->bio, io->error); |
2e93ccc1 | 501 | } |
2056a782 | 502 | |
1da177e4 LT |
503 | free_io(io->md, io); |
504 | } | |
505 | } | |
506 | ||
6712ecf8 | 507 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
508 | { |
509 | int r = 0; | |
028867ac | 510 | struct dm_target_io *tio = bio->bi_private; |
9faf400f | 511 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
512 | dm_endio_fn endio = tio->ti->type->end_io; |
513 | ||
1da177e4 LT |
514 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
515 | error = -EIO; | |
516 | ||
517 | if (endio) { | |
518 | r = endio(tio->ti, bio, error, &tio->info); | |
2e93ccc1 KU |
519 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
520 | /* | |
521 | * error and requeue request are handled | |
522 | * in dec_pending(). | |
523 | */ | |
1da177e4 | 524 | error = r; |
45cbcd79 KU |
525 | else if (r == DM_ENDIO_INCOMPLETE) |
526 | /* The target will handle the io */ | |
6712ecf8 | 527 | return; |
45cbcd79 KU |
528 | else if (r) { |
529 | DMWARN("unimplemented target endio return value: %d", r); | |
530 | BUG(); | |
531 | } | |
1da177e4 LT |
532 | } |
533 | ||
9faf400f SB |
534 | dec_pending(tio->io, error); |
535 | ||
536 | /* | |
537 | * Store md for cleanup instead of tio which is about to get freed. | |
538 | */ | |
539 | bio->bi_private = md->bs; | |
540 | ||
1da177e4 | 541 | bio_put(bio); |
9faf400f | 542 | free_tio(md, tio); |
1da177e4 LT |
543 | } |
544 | ||
545 | static sector_t max_io_len(struct mapped_device *md, | |
546 | sector_t sector, struct dm_target *ti) | |
547 | { | |
548 | sector_t offset = sector - ti->begin; | |
549 | sector_t len = ti->len - offset; | |
550 | ||
551 | /* | |
552 | * Does the target need to split even further ? | |
553 | */ | |
554 | if (ti->split_io) { | |
555 | sector_t boundary; | |
556 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) | |
557 | - offset; | |
558 | if (len > boundary) | |
559 | len = boundary; | |
560 | } | |
561 | ||
562 | return len; | |
563 | } | |
564 | ||
565 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
028867ac | 566 | struct dm_target_io *tio) |
1da177e4 LT |
567 | { |
568 | int r; | |
2056a782 | 569 | sector_t sector; |
9faf400f | 570 | struct mapped_device *md; |
1da177e4 LT |
571 | |
572 | /* | |
573 | * Sanity checks. | |
574 | */ | |
575 | BUG_ON(!clone->bi_size); | |
576 | ||
577 | clone->bi_end_io = clone_endio; | |
578 | clone->bi_private = tio; | |
579 | ||
580 | /* | |
581 | * Map the clone. If r == 0 we don't need to do | |
582 | * anything, the target has assumed ownership of | |
583 | * this io. | |
584 | */ | |
585 | atomic_inc(&tio->io->io_count); | |
2056a782 | 586 | sector = clone->bi_sector; |
1da177e4 | 587 | r = ti->type->map(ti, clone, &tio->info); |
45cbcd79 | 588 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 589 | /* the bio has been remapped so dispatch it */ |
2056a782 | 590 | |
17b2f66f | 591 | blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone, |
c7149d6b AB |
592 | tio->io->bio->bi_bdev->bd_dev, |
593 | clone->bi_sector, sector); | |
2056a782 | 594 | |
1da177e4 | 595 | generic_make_request(clone); |
2e93ccc1 KU |
596 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
597 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
598 | md = tio->io->md; |
599 | dec_pending(tio->io, r); | |
600 | /* | |
601 | * Store bio_set for cleanup. | |
602 | */ | |
603 | clone->bi_private = md->bs; | |
1da177e4 | 604 | bio_put(clone); |
9faf400f | 605 | free_tio(md, tio); |
45cbcd79 KU |
606 | } else if (r) { |
607 | DMWARN("unimplemented target map return value: %d", r); | |
608 | BUG(); | |
1da177e4 LT |
609 | } |
610 | } | |
611 | ||
612 | struct clone_info { | |
613 | struct mapped_device *md; | |
614 | struct dm_table *map; | |
615 | struct bio *bio; | |
616 | struct dm_io *io; | |
617 | sector_t sector; | |
618 | sector_t sector_count; | |
619 | unsigned short idx; | |
620 | }; | |
621 | ||
3676347a PO |
622 | static void dm_bio_destructor(struct bio *bio) |
623 | { | |
9faf400f SB |
624 | struct bio_set *bs = bio->bi_private; |
625 | ||
626 | bio_free(bio, bs); | |
3676347a PO |
627 | } |
628 | ||
1da177e4 LT |
629 | /* |
630 | * Creates a little bio that is just does part of a bvec. | |
631 | */ | |
632 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
633 | unsigned short idx, unsigned int offset, | |
9faf400f | 634 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
635 | { |
636 | struct bio *clone; | |
637 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
638 | ||
9faf400f | 639 | clone = bio_alloc_bioset(GFP_NOIO, 1, bs); |
3676347a | 640 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
641 | *clone->bi_io_vec = *bv; |
642 | ||
643 | clone->bi_sector = sector; | |
644 | clone->bi_bdev = bio->bi_bdev; | |
645 | clone->bi_rw = bio->bi_rw; | |
646 | clone->bi_vcnt = 1; | |
647 | clone->bi_size = to_bytes(len); | |
648 | clone->bi_io_vec->bv_offset = offset; | |
649 | clone->bi_io_vec->bv_len = clone->bi_size; | |
650 | ||
651 | return clone; | |
652 | } | |
653 | ||
654 | /* | |
655 | * Creates a bio that consists of range of complete bvecs. | |
656 | */ | |
657 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
658 | unsigned short idx, unsigned short bv_count, | |
9faf400f | 659 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
660 | { |
661 | struct bio *clone; | |
662 | ||
9faf400f SB |
663 | clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); |
664 | __bio_clone(clone, bio); | |
665 | clone->bi_destructor = dm_bio_destructor; | |
1da177e4 LT |
666 | clone->bi_sector = sector; |
667 | clone->bi_idx = idx; | |
668 | clone->bi_vcnt = idx + bv_count; | |
669 | clone->bi_size = to_bytes(len); | |
670 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
671 | ||
672 | return clone; | |
673 | } | |
674 | ||
512875bd | 675 | static int __clone_and_map(struct clone_info *ci) |
1da177e4 LT |
676 | { |
677 | struct bio *clone, *bio = ci->bio; | |
512875bd JN |
678 | struct dm_target *ti; |
679 | sector_t len = 0, max; | |
028867ac | 680 | struct dm_target_io *tio; |
1da177e4 | 681 | |
512875bd JN |
682 | ti = dm_table_find_target(ci->map, ci->sector); |
683 | if (!dm_target_is_valid(ti)) | |
684 | return -EIO; | |
685 | ||
686 | max = max_io_len(ci->md, ci->sector, ti); | |
687 | ||
1da177e4 LT |
688 | /* |
689 | * Allocate a target io object. | |
690 | */ | |
691 | tio = alloc_tio(ci->md); | |
692 | tio->io = ci->io; | |
693 | tio->ti = ti; | |
694 | memset(&tio->info, 0, sizeof(tio->info)); | |
695 | ||
696 | if (ci->sector_count <= max) { | |
697 | /* | |
698 | * Optimise for the simple case where we can do all of | |
699 | * the remaining io with a single clone. | |
700 | */ | |
701 | clone = clone_bio(bio, ci->sector, ci->idx, | |
9faf400f SB |
702 | bio->bi_vcnt - ci->idx, ci->sector_count, |
703 | ci->md->bs); | |
1da177e4 LT |
704 | __map_bio(ti, clone, tio); |
705 | ci->sector_count = 0; | |
706 | ||
707 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
708 | /* | |
709 | * There are some bvecs that don't span targets. | |
710 | * Do as many of these as possible. | |
711 | */ | |
712 | int i; | |
713 | sector_t remaining = max; | |
714 | sector_t bv_len; | |
715 | ||
716 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
717 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
718 | ||
719 | if (bv_len > remaining) | |
720 | break; | |
721 | ||
722 | remaining -= bv_len; | |
723 | len += bv_len; | |
724 | } | |
725 | ||
9faf400f SB |
726 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, |
727 | ci->md->bs); | |
1da177e4 LT |
728 | __map_bio(ti, clone, tio); |
729 | ||
730 | ci->sector += len; | |
731 | ci->sector_count -= len; | |
732 | ci->idx = i; | |
733 | ||
734 | } else { | |
735 | /* | |
d2044a94 | 736 | * Handle a bvec that must be split between two or more targets. |
1da177e4 LT |
737 | */ |
738 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
d2044a94 AK |
739 | sector_t remaining = to_sector(bv->bv_len); |
740 | unsigned int offset = 0; | |
1da177e4 | 741 | |
d2044a94 AK |
742 | do { |
743 | if (offset) { | |
744 | ti = dm_table_find_target(ci->map, ci->sector); | |
512875bd JN |
745 | if (!dm_target_is_valid(ti)) |
746 | return -EIO; | |
747 | ||
d2044a94 | 748 | max = max_io_len(ci->md, ci->sector, ti); |
1da177e4 | 749 | |
d2044a94 AK |
750 | tio = alloc_tio(ci->md); |
751 | tio->io = ci->io; | |
752 | tio->ti = ti; | |
753 | memset(&tio->info, 0, sizeof(tio->info)); | |
754 | } | |
755 | ||
756 | len = min(remaining, max); | |
757 | ||
758 | clone = split_bvec(bio, ci->sector, ci->idx, | |
9faf400f SB |
759 | bv->bv_offset + offset, len, |
760 | ci->md->bs); | |
d2044a94 AK |
761 | |
762 | __map_bio(ti, clone, tio); | |
763 | ||
764 | ci->sector += len; | |
765 | ci->sector_count -= len; | |
766 | offset += to_bytes(len); | |
767 | } while (remaining -= len); | |
1da177e4 | 768 | |
1da177e4 LT |
769 | ci->idx++; |
770 | } | |
512875bd JN |
771 | |
772 | return 0; | |
1da177e4 LT |
773 | } |
774 | ||
775 | /* | |
776 | * Split the bio into several clones. | |
777 | */ | |
9e4e5f87 | 778 | static int __split_bio(struct mapped_device *md, struct bio *bio) |
1da177e4 LT |
779 | { |
780 | struct clone_info ci; | |
512875bd | 781 | int error = 0; |
1da177e4 LT |
782 | |
783 | ci.map = dm_get_table(md); | |
9e4e5f87 MB |
784 | if (unlikely(!ci.map)) |
785 | return -EIO; | |
1da177e4 LT |
786 | |
787 | ci.md = md; | |
788 | ci.bio = bio; | |
789 | ci.io = alloc_io(md); | |
790 | ci.io->error = 0; | |
791 | atomic_set(&ci.io->io_count, 1); | |
792 | ci.io->bio = bio; | |
793 | ci.io->md = md; | |
794 | ci.sector = bio->bi_sector; | |
795 | ci.sector_count = bio_sectors(bio); | |
796 | ci.idx = bio->bi_idx; | |
797 | ||
3eaf840e | 798 | start_io_acct(ci.io); |
512875bd JN |
799 | while (ci.sector_count && !error) |
800 | error = __clone_and_map(&ci); | |
1da177e4 LT |
801 | |
802 | /* drop the extra reference count */ | |
512875bd | 803 | dec_pending(ci.io, error); |
1da177e4 | 804 | dm_table_put(ci.map); |
9e4e5f87 MB |
805 | |
806 | return 0; | |
1da177e4 LT |
807 | } |
808 | /*----------------------------------------------------------------- | |
809 | * CRUD END | |
810 | *---------------------------------------------------------------*/ | |
811 | ||
812 | /* | |
813 | * The request function that just remaps the bio built up by | |
814 | * dm_merge_bvec. | |
815 | */ | |
165125e1 | 816 | static int dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 817 | { |
9e4e5f87 | 818 | int r = -EIO; |
12f03a49 | 819 | int rw = bio_data_dir(bio); |
1da177e4 LT |
820 | struct mapped_device *md = q->queuedata; |
821 | ||
07a83c47 SB |
822 | /* |
823 | * There is no use in forwarding any barrier request since we can't | |
824 | * guarantee it is (or can be) handled by the targets correctly. | |
825 | */ | |
826 | if (unlikely(bio_barrier(bio))) { | |
6712ecf8 | 827 | bio_endio(bio, -EOPNOTSUPP); |
07a83c47 SB |
828 | return 0; |
829 | } | |
830 | ||
2ca3310e | 831 | down_read(&md->io_lock); |
1da177e4 | 832 | |
12f03a49 KC |
833 | disk_stat_inc(dm_disk(md), ios[rw]); |
834 | disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio)); | |
835 | ||
1da177e4 LT |
836 | /* |
837 | * If we're suspended we have to queue | |
838 | * this io for later. | |
839 | */ | |
840 | while (test_bit(DMF_BLOCK_IO, &md->flags)) { | |
2ca3310e | 841 | up_read(&md->io_lock); |
1da177e4 | 842 | |
9e4e5f87 MB |
843 | if (bio_rw(bio) != READA) |
844 | r = queue_io(md, bio); | |
1da177e4 | 845 | |
9e4e5f87 MB |
846 | if (r <= 0) |
847 | goto out_req; | |
1da177e4 LT |
848 | |
849 | /* | |
850 | * We're in a while loop, because someone could suspend | |
851 | * before we get to the following read lock. | |
852 | */ | |
2ca3310e | 853 | down_read(&md->io_lock); |
1da177e4 LT |
854 | } |
855 | ||
9e4e5f87 | 856 | r = __split_bio(md, bio); |
2ca3310e | 857 | up_read(&md->io_lock); |
9e4e5f87 MB |
858 | |
859 | out_req: | |
860 | if (r < 0) | |
861 | bio_io_error(bio); | |
862 | ||
1da177e4 LT |
863 | return 0; |
864 | } | |
865 | ||
165125e1 | 866 | static void dm_unplug_all(struct request_queue *q) |
1da177e4 LT |
867 | { |
868 | struct mapped_device *md = q->queuedata; | |
869 | struct dm_table *map = dm_get_table(md); | |
870 | ||
871 | if (map) { | |
872 | dm_table_unplug_all(map); | |
873 | dm_table_put(map); | |
874 | } | |
875 | } | |
876 | ||
877 | static int dm_any_congested(void *congested_data, int bdi_bits) | |
878 | { | |
879 | int r; | |
880 | struct mapped_device *md = (struct mapped_device *) congested_data; | |
881 | struct dm_table *map = dm_get_table(md); | |
882 | ||
883 | if (!map || test_bit(DMF_BLOCK_IO, &md->flags)) | |
884 | r = bdi_bits; | |
885 | else | |
886 | r = dm_table_any_congested(map, bdi_bits); | |
887 | ||
888 | dm_table_put(map); | |
889 | return r; | |
890 | } | |
891 | ||
892 | /*----------------------------------------------------------------- | |
893 | * An IDR is used to keep track of allocated minor numbers. | |
894 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
895 | static DEFINE_IDR(_minor_idr); |
896 | ||
2b06cfff | 897 | static void free_minor(int minor) |
1da177e4 | 898 | { |
f32c10b0 | 899 | spin_lock(&_minor_lock); |
1da177e4 | 900 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 901 | spin_unlock(&_minor_lock); |
1da177e4 LT |
902 | } |
903 | ||
904 | /* | |
905 | * See if the device with a specific minor # is free. | |
906 | */ | |
2b06cfff | 907 | static int specific_minor(struct mapped_device *md, int minor) |
1da177e4 LT |
908 | { |
909 | int r, m; | |
910 | ||
911 | if (minor >= (1 << MINORBITS)) | |
912 | return -EINVAL; | |
913 | ||
62f75c2f JM |
914 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
915 | if (!r) | |
916 | return -ENOMEM; | |
917 | ||
f32c10b0 | 918 | spin_lock(&_minor_lock); |
1da177e4 LT |
919 | |
920 | if (idr_find(&_minor_idr, minor)) { | |
921 | r = -EBUSY; | |
922 | goto out; | |
923 | } | |
924 | ||
ba61fdd1 | 925 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); |
62f75c2f | 926 | if (r) |
1da177e4 | 927 | goto out; |
1da177e4 LT |
928 | |
929 | if (m != minor) { | |
930 | idr_remove(&_minor_idr, m); | |
931 | r = -EBUSY; | |
932 | goto out; | |
933 | } | |
934 | ||
935 | out: | |
f32c10b0 | 936 | spin_unlock(&_minor_lock); |
1da177e4 LT |
937 | return r; |
938 | } | |
939 | ||
2b06cfff | 940 | static int next_free_minor(struct mapped_device *md, int *minor) |
1da177e4 | 941 | { |
2b06cfff | 942 | int r, m; |
1da177e4 | 943 | |
1da177e4 | 944 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
62f75c2f JM |
945 | if (!r) |
946 | return -ENOMEM; | |
947 | ||
f32c10b0 | 948 | spin_lock(&_minor_lock); |
1da177e4 | 949 | |
ba61fdd1 | 950 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); |
1da177e4 LT |
951 | if (r) { |
952 | goto out; | |
953 | } | |
954 | ||
955 | if (m >= (1 << MINORBITS)) { | |
956 | idr_remove(&_minor_idr, m); | |
957 | r = -ENOSPC; | |
958 | goto out; | |
959 | } | |
960 | ||
961 | *minor = m; | |
962 | ||
963 | out: | |
f32c10b0 | 964 | spin_unlock(&_minor_lock); |
1da177e4 LT |
965 | return r; |
966 | } | |
967 | ||
968 | static struct block_device_operations dm_blk_dops; | |
969 | ||
970 | /* | |
971 | * Allocate and initialise a blank device with a given minor. | |
972 | */ | |
2b06cfff | 973 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
974 | { |
975 | int r; | |
976 | struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL); | |
ba61fdd1 | 977 | void *old_md; |
1da177e4 LT |
978 | |
979 | if (!md) { | |
980 | DMWARN("unable to allocate device, out of memory."); | |
981 | return NULL; | |
982 | } | |
983 | ||
10da4f79 JM |
984 | if (!try_module_get(THIS_MODULE)) |
985 | goto bad0; | |
986 | ||
1da177e4 | 987 | /* get a minor number for the dev */ |
2b06cfff AK |
988 | if (minor == DM_ANY_MINOR) |
989 | r = next_free_minor(md, &minor); | |
990 | else | |
991 | r = specific_minor(md, minor); | |
1da177e4 LT |
992 | if (r < 0) |
993 | goto bad1; | |
994 | ||
995 | memset(md, 0, sizeof(*md)); | |
2ca3310e AK |
996 | init_rwsem(&md->io_lock); |
997 | init_MUTEX(&md->suspend_lock); | |
2e93ccc1 | 998 | spin_lock_init(&md->pushback_lock); |
1da177e4 LT |
999 | rwlock_init(&md->map_lock); |
1000 | atomic_set(&md->holders, 1); | |
5c6bd75d | 1001 | atomic_set(&md->open_count, 0); |
1da177e4 | 1002 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1003 | atomic_set(&md->uevent_seq, 0); |
1004 | INIT_LIST_HEAD(&md->uevent_list); | |
1005 | spin_lock_init(&md->uevent_lock); | |
1da177e4 LT |
1006 | |
1007 | md->queue = blk_alloc_queue(GFP_KERNEL); | |
1008 | if (!md->queue) | |
e3f6ac61 | 1009 | goto bad1_free_minor; |
1da177e4 LT |
1010 | |
1011 | md->queue->queuedata = md; | |
1012 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1013 | md->queue->backing_dev_info.congested_data = md; | |
1014 | blk_queue_make_request(md->queue, dm_request); | |
daef265f | 1015 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
1da177e4 | 1016 | md->queue->unplug_fn = dm_unplug_all; |
1da177e4 | 1017 | |
93d2341c | 1018 | md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache); |
74859364 KU |
1019 | if (!md->io_pool) |
1020 | goto bad2; | |
1da177e4 | 1021 | |
93d2341c | 1022 | md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache); |
1da177e4 LT |
1023 | if (!md->tio_pool) |
1024 | goto bad3; | |
1025 | ||
5972511b | 1026 | md->bs = bioset_create(16, 16); |
9faf400f SB |
1027 | if (!md->bs) |
1028 | goto bad_no_bioset; | |
1029 | ||
1da177e4 LT |
1030 | md->disk = alloc_disk(1); |
1031 | if (!md->disk) | |
1032 | goto bad4; | |
1033 | ||
f0b04115 JM |
1034 | atomic_set(&md->pending, 0); |
1035 | init_waitqueue_head(&md->wait); | |
1036 | init_waitqueue_head(&md->eventq); | |
1037 | ||
1da177e4 LT |
1038 | md->disk->major = _major; |
1039 | md->disk->first_minor = minor; | |
1040 | md->disk->fops = &dm_blk_dops; | |
1041 | md->disk->queue = md->queue; | |
1042 | md->disk->private_data = md; | |
1043 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1044 | add_disk(md->disk); | |
7e51f257 | 1045 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1046 | |
ba61fdd1 | 1047 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1048 | spin_lock(&_minor_lock); |
ba61fdd1 | 1049 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1050 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1051 | |
1052 | BUG_ON(old_md != MINOR_ALLOCED); | |
1053 | ||
1da177e4 LT |
1054 | return md; |
1055 | ||
1056 | bad4: | |
9faf400f SB |
1057 | bioset_free(md->bs); |
1058 | bad_no_bioset: | |
1da177e4 LT |
1059 | mempool_destroy(md->tio_pool); |
1060 | bad3: | |
1061 | mempool_destroy(md->io_pool); | |
1062 | bad2: | |
1312f40e | 1063 | blk_cleanup_queue(md->queue); |
e3f6ac61 | 1064 | bad1_free_minor: |
1da177e4 LT |
1065 | free_minor(minor); |
1066 | bad1: | |
10da4f79 JM |
1067 | module_put(THIS_MODULE); |
1068 | bad0: | |
1da177e4 LT |
1069 | kfree(md); |
1070 | return NULL; | |
1071 | } | |
1072 | ||
ae9da83f JN |
1073 | static void unlock_fs(struct mapped_device *md); |
1074 | ||
1da177e4 LT |
1075 | static void free_dev(struct mapped_device *md) |
1076 | { | |
2b06cfff | 1077 | int minor = md->disk->first_minor; |
63d94e48 | 1078 | |
d9dde59b | 1079 | if (md->suspended_bdev) { |
ae9da83f | 1080 | unlock_fs(md); |
d9dde59b JN |
1081 | bdput(md->suspended_bdev); |
1082 | } | |
1da177e4 LT |
1083 | mempool_destroy(md->tio_pool); |
1084 | mempool_destroy(md->io_pool); | |
9faf400f | 1085 | bioset_free(md->bs); |
1da177e4 | 1086 | del_gendisk(md->disk); |
63d94e48 | 1087 | free_minor(minor); |
fba9f90e JM |
1088 | |
1089 | spin_lock(&_minor_lock); | |
1090 | md->disk->private_data = NULL; | |
1091 | spin_unlock(&_minor_lock); | |
1092 | ||
1da177e4 | 1093 | put_disk(md->disk); |
1312f40e | 1094 | blk_cleanup_queue(md->queue); |
10da4f79 | 1095 | module_put(THIS_MODULE); |
1da177e4 LT |
1096 | kfree(md); |
1097 | } | |
1098 | ||
1099 | /* | |
1100 | * Bind a table to the device. | |
1101 | */ | |
1102 | static void event_callback(void *context) | |
1103 | { | |
7a8c3d3b MA |
1104 | unsigned long flags; |
1105 | LIST_HEAD(uevents); | |
1da177e4 LT |
1106 | struct mapped_device *md = (struct mapped_device *) context; |
1107 | ||
7a8c3d3b MA |
1108 | spin_lock_irqsave(&md->uevent_lock, flags); |
1109 | list_splice_init(&md->uevent_list, &uevents); | |
1110 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1111 | ||
1112 | dm_send_uevents(&uevents, &md->disk->kobj); | |
1113 | ||
1da177e4 LT |
1114 | atomic_inc(&md->event_nr); |
1115 | wake_up(&md->eventq); | |
1116 | } | |
1117 | ||
4e90188b | 1118 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1119 | { |
4e90188b | 1120 | set_capacity(md->disk, size); |
1da177e4 | 1121 | |
1b1dcc1b | 1122 | mutex_lock(&md->suspended_bdev->bd_inode->i_mutex); |
e39e2e95 | 1123 | i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1b1dcc1b | 1124 | mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex); |
1da177e4 LT |
1125 | } |
1126 | ||
1127 | static int __bind(struct mapped_device *md, struct dm_table *t) | |
1128 | { | |
165125e1 | 1129 | struct request_queue *q = md->queue; |
1da177e4 LT |
1130 | sector_t size; |
1131 | ||
1132 | size = dm_table_get_size(t); | |
3ac51e74 DW |
1133 | |
1134 | /* | |
1135 | * Wipe any geometry if the size of the table changed. | |
1136 | */ | |
1137 | if (size != get_capacity(md->disk)) | |
1138 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
1139 | ||
bfa152fa JN |
1140 | if (md->suspended_bdev) |
1141 | __set_size(md, size); | |
1da177e4 LT |
1142 | if (size == 0) |
1143 | return 0; | |
1144 | ||
2ca3310e AK |
1145 | dm_table_get(t); |
1146 | dm_table_event_callback(t, event_callback, md); | |
1147 | ||
1da177e4 LT |
1148 | write_lock(&md->map_lock); |
1149 | md->map = t; | |
2ca3310e | 1150 | dm_table_set_restrictions(t, q); |
1da177e4 LT |
1151 | write_unlock(&md->map_lock); |
1152 | ||
1da177e4 LT |
1153 | return 0; |
1154 | } | |
1155 | ||
1156 | static void __unbind(struct mapped_device *md) | |
1157 | { | |
1158 | struct dm_table *map = md->map; | |
1159 | ||
1160 | if (!map) | |
1161 | return; | |
1162 | ||
1163 | dm_table_event_callback(map, NULL, NULL); | |
1164 | write_lock(&md->map_lock); | |
1165 | md->map = NULL; | |
1166 | write_unlock(&md->map_lock); | |
1167 | dm_table_put(map); | |
1168 | } | |
1169 | ||
1170 | /* | |
1171 | * Constructor for a new device. | |
1172 | */ | |
2b06cfff | 1173 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
1174 | { |
1175 | struct mapped_device *md; | |
1176 | ||
2b06cfff | 1177 | md = alloc_dev(minor); |
1da177e4 LT |
1178 | if (!md) |
1179 | return -ENXIO; | |
1180 | ||
1181 | *result = md; | |
1182 | return 0; | |
1183 | } | |
1184 | ||
637842cf | 1185 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
1186 | { |
1187 | struct mapped_device *md; | |
1da177e4 LT |
1188 | unsigned minor = MINOR(dev); |
1189 | ||
1190 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
1191 | return NULL; | |
1192 | ||
f32c10b0 | 1193 | spin_lock(&_minor_lock); |
1da177e4 LT |
1194 | |
1195 | md = idr_find(&_minor_idr, minor); | |
fba9f90e JM |
1196 | if (md && (md == MINOR_ALLOCED || |
1197 | (dm_disk(md)->first_minor != minor) || | |
17b2f66f | 1198 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 1199 | md = NULL; |
fba9f90e JM |
1200 | goto out; |
1201 | } | |
1da177e4 | 1202 | |
fba9f90e | 1203 | out: |
f32c10b0 | 1204 | spin_unlock(&_minor_lock); |
1da177e4 | 1205 | |
637842cf DT |
1206 | return md; |
1207 | } | |
1208 | ||
d229a958 DT |
1209 | struct mapped_device *dm_get_md(dev_t dev) |
1210 | { | |
1211 | struct mapped_device *md = dm_find_md(dev); | |
1212 | ||
1213 | if (md) | |
1214 | dm_get(md); | |
1215 | ||
1216 | return md; | |
1217 | } | |
1218 | ||
9ade92a9 | 1219 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 1220 | { |
9ade92a9 | 1221 | return md->interface_ptr; |
1da177e4 LT |
1222 | } |
1223 | ||
1224 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
1225 | { | |
1226 | md->interface_ptr = ptr; | |
1227 | } | |
1228 | ||
1229 | void dm_get(struct mapped_device *md) | |
1230 | { | |
1231 | atomic_inc(&md->holders); | |
1232 | } | |
1233 | ||
72d94861 AK |
1234 | const char *dm_device_name(struct mapped_device *md) |
1235 | { | |
1236 | return md->name; | |
1237 | } | |
1238 | EXPORT_SYMBOL_GPL(dm_device_name); | |
1239 | ||
1da177e4 LT |
1240 | void dm_put(struct mapped_device *md) |
1241 | { | |
1134e5ae | 1242 | struct dm_table *map; |
1da177e4 | 1243 | |
fba9f90e JM |
1244 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1245 | ||
f32c10b0 | 1246 | if (atomic_dec_and_lock(&md->holders, &_minor_lock)) { |
1134e5ae | 1247 | map = dm_get_table(md); |
ba61fdd1 | 1248 | idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor); |
fba9f90e | 1249 | set_bit(DMF_FREEING, &md->flags); |
f32c10b0 | 1250 | spin_unlock(&_minor_lock); |
cf222b37 | 1251 | if (!dm_suspended(md)) { |
1da177e4 LT |
1252 | dm_table_presuspend_targets(map); |
1253 | dm_table_postsuspend_targets(map); | |
1254 | } | |
1255 | __unbind(md); | |
1134e5ae | 1256 | dm_table_put(map); |
1da177e4 LT |
1257 | free_dev(md); |
1258 | } | |
1da177e4 | 1259 | } |
79eb885c | 1260 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 LT |
1261 | |
1262 | /* | |
1263 | * Process the deferred bios | |
1264 | */ | |
1265 | static void __flush_deferred_io(struct mapped_device *md, struct bio *c) | |
1266 | { | |
1267 | struct bio *n; | |
1268 | ||
1269 | while (c) { | |
1270 | n = c->bi_next; | |
1271 | c->bi_next = NULL; | |
9e4e5f87 MB |
1272 | if (__split_bio(md, c)) |
1273 | bio_io_error(c); | |
1da177e4 LT |
1274 | c = n; |
1275 | } | |
1276 | } | |
1277 | ||
1278 | /* | |
1279 | * Swap in a new table (destroying old one). | |
1280 | */ | |
1281 | int dm_swap_table(struct mapped_device *md, struct dm_table *table) | |
1282 | { | |
93c534ae | 1283 | int r = -EINVAL; |
1da177e4 | 1284 | |
2ca3310e | 1285 | down(&md->suspend_lock); |
1da177e4 LT |
1286 | |
1287 | /* device must be suspended */ | |
cf222b37 | 1288 | if (!dm_suspended(md)) |
93c534ae | 1289 | goto out; |
1da177e4 | 1290 | |
bfa152fa JN |
1291 | /* without bdev, the device size cannot be changed */ |
1292 | if (!md->suspended_bdev) | |
1293 | if (get_capacity(md->disk) != dm_table_get_size(table)) | |
1294 | goto out; | |
1295 | ||
1da177e4 LT |
1296 | __unbind(md); |
1297 | r = __bind(md, table); | |
1da177e4 | 1298 | |
93c534ae | 1299 | out: |
2ca3310e | 1300 | up(&md->suspend_lock); |
93c534ae | 1301 | return r; |
1da177e4 LT |
1302 | } |
1303 | ||
1304 | /* | |
1305 | * Functions to lock and unlock any filesystem running on the | |
1306 | * device. | |
1307 | */ | |
2ca3310e | 1308 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 1309 | { |
e39e2e95 | 1310 | int r; |
1da177e4 LT |
1311 | |
1312 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 1313 | |
e39e2e95 | 1314 | md->frozen_sb = freeze_bdev(md->suspended_bdev); |
dfbe03f6 | 1315 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 1316 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
1317 | md->frozen_sb = NULL; |
1318 | return r; | |
dfbe03f6 AK |
1319 | } |
1320 | ||
aa8d7c2f AK |
1321 | set_bit(DMF_FROZEN, &md->flags); |
1322 | ||
1da177e4 | 1323 | /* don't bdput right now, we don't want the bdev |
e39e2e95 | 1324 | * to go away while it is locked. |
1da177e4 LT |
1325 | */ |
1326 | return 0; | |
1327 | } | |
1328 | ||
2ca3310e | 1329 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 1330 | { |
aa8d7c2f AK |
1331 | if (!test_bit(DMF_FROZEN, &md->flags)) |
1332 | return; | |
1333 | ||
e39e2e95 | 1334 | thaw_bdev(md->suspended_bdev, md->frozen_sb); |
1da177e4 | 1335 | md->frozen_sb = NULL; |
aa8d7c2f | 1336 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
1337 | } |
1338 | ||
1339 | /* | |
1340 | * We need to be able to change a mapping table under a mounted | |
1341 | * filesystem. For example we might want to move some data in | |
1342 | * the background. Before the table can be swapped with | |
1343 | * dm_bind_table, dm_suspend must be called to flush any in | |
1344 | * flight bios and ensure that any further io gets deferred. | |
1345 | */ | |
a3d77d35 | 1346 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 1347 | { |
2ca3310e | 1348 | struct dm_table *map = NULL; |
2e93ccc1 | 1349 | unsigned long flags; |
1da177e4 | 1350 | DECLARE_WAITQUEUE(wait, current); |
1ecac7fd | 1351 | struct bio *def; |
cf222b37 | 1352 | int r = -EINVAL; |
a3d77d35 | 1353 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 1354 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 1355 | |
2ca3310e AK |
1356 | down(&md->suspend_lock); |
1357 | ||
1358 | if (dm_suspended(md)) | |
d287483d | 1359 | goto out_unlock; |
1da177e4 LT |
1360 | |
1361 | map = dm_get_table(md); | |
1da177e4 | 1362 | |
2e93ccc1 KU |
1363 | /* |
1364 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
1365 | * This flag is cleared before dm_suspend returns. | |
1366 | */ | |
1367 | if (noflush) | |
1368 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1369 | ||
cf222b37 AK |
1370 | /* This does not get reverted if there's an error later. */ |
1371 | dm_table_presuspend_targets(map); | |
1372 | ||
bfa152fa JN |
1373 | /* bdget() can stall if the pending I/Os are not flushed */ |
1374 | if (!noflush) { | |
1375 | md->suspended_bdev = bdget_disk(md->disk, 0); | |
1376 | if (!md->suspended_bdev) { | |
1377 | DMWARN("bdget failed in dm_suspend"); | |
1378 | r = -ENOMEM; | |
1379 | goto flush_and_out; | |
1380 | } | |
e39e2e95 AK |
1381 | } |
1382 | ||
2e93ccc1 KU |
1383 | /* |
1384 | * Flush I/O to the device. | |
1385 | * noflush supersedes do_lockfs, because lock_fs() needs to flush I/Os. | |
1386 | */ | |
1387 | if (do_lockfs && !noflush) { | |
aa8d7c2f AK |
1388 | r = lock_fs(md); |
1389 | if (r) | |
1390 | goto out; | |
1391 | } | |
1da177e4 LT |
1392 | |
1393 | /* | |
354e0071 | 1394 | * First we set the BLOCK_IO flag so no more ios will be mapped. |
1da177e4 | 1395 | */ |
2ca3310e AK |
1396 | down_write(&md->io_lock); |
1397 | set_bit(DMF_BLOCK_IO, &md->flags); | |
1da177e4 | 1398 | |
1da177e4 | 1399 | add_wait_queue(&md->wait, &wait); |
2ca3310e | 1400 | up_write(&md->io_lock); |
1da177e4 LT |
1401 | |
1402 | /* unplug */ | |
2ca3310e | 1403 | if (map) |
1da177e4 | 1404 | dm_table_unplug_all(map); |
1da177e4 LT |
1405 | |
1406 | /* | |
1407 | * Then we wait for the already mapped ios to | |
1408 | * complete. | |
1409 | */ | |
1410 | while (1) { | |
1411 | set_current_state(TASK_INTERRUPTIBLE); | |
1412 | ||
1413 | if (!atomic_read(&md->pending) || signal_pending(current)) | |
1414 | break; | |
1415 | ||
1416 | io_schedule(); | |
1417 | } | |
1418 | set_current_state(TASK_RUNNING); | |
1419 | ||
2ca3310e | 1420 | down_write(&md->io_lock); |
1da177e4 LT |
1421 | remove_wait_queue(&md->wait, &wait); |
1422 | ||
2e93ccc1 KU |
1423 | if (noflush) { |
1424 | spin_lock_irqsave(&md->pushback_lock, flags); | |
1425 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1426 | bio_list_merge_head(&md->deferred, &md->pushback); | |
1427 | bio_list_init(&md->pushback); | |
1428 | spin_unlock_irqrestore(&md->pushback_lock, flags); | |
1429 | } | |
1430 | ||
1da177e4 | 1431 | /* were we interrupted ? */ |
cf222b37 | 1432 | r = -EINTR; |
2ca3310e | 1433 | if (atomic_read(&md->pending)) { |
1ecac7fd JN |
1434 | clear_bit(DMF_BLOCK_IO, &md->flags); |
1435 | def = bio_list_get(&md->deferred); | |
1436 | __flush_deferred_io(md, def); | |
2ca3310e AK |
1437 | up_write(&md->io_lock); |
1438 | unlock_fs(md); | |
2e93ccc1 | 1439 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e AK |
1440 | } |
1441 | up_write(&md->io_lock); | |
1da177e4 | 1442 | |
cf222b37 | 1443 | dm_table_postsuspend_targets(map); |
1da177e4 | 1444 | |
2ca3310e | 1445 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 1446 | |
2ca3310e | 1447 | r = 0; |
b84b0287 | 1448 | |
2e93ccc1 KU |
1449 | flush_and_out: |
1450 | if (r && noflush) { | |
1451 | /* | |
1452 | * Because there may be already I/Os in the pushback list, | |
1453 | * flush them before return. | |
1454 | */ | |
1455 | down_write(&md->io_lock); | |
1456 | ||
1457 | spin_lock_irqsave(&md->pushback_lock, flags); | |
1458 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1459 | bio_list_merge_head(&md->deferred, &md->pushback); | |
1460 | bio_list_init(&md->pushback); | |
1461 | spin_unlock_irqrestore(&md->pushback_lock, flags); | |
1462 | ||
1463 | def = bio_list_get(&md->deferred); | |
1464 | __flush_deferred_io(md, def); | |
1465 | up_write(&md->io_lock); | |
1466 | } | |
1467 | ||
2ca3310e | 1468 | out: |
e39e2e95 AK |
1469 | if (r && md->suspended_bdev) { |
1470 | bdput(md->suspended_bdev); | |
1471 | md->suspended_bdev = NULL; | |
1472 | } | |
1473 | ||
2ca3310e | 1474 | dm_table_put(map); |
d287483d AK |
1475 | |
1476 | out_unlock: | |
2ca3310e | 1477 | up(&md->suspend_lock); |
cf222b37 | 1478 | return r; |
1da177e4 LT |
1479 | } |
1480 | ||
1481 | int dm_resume(struct mapped_device *md) | |
1482 | { | |
cf222b37 | 1483 | int r = -EINVAL; |
1da177e4 | 1484 | struct bio *def; |
cf222b37 | 1485 | struct dm_table *map = NULL; |
1da177e4 | 1486 | |
2ca3310e AK |
1487 | down(&md->suspend_lock); |
1488 | if (!dm_suspended(md)) | |
cf222b37 | 1489 | goto out; |
cf222b37 AK |
1490 | |
1491 | map = dm_get_table(md); | |
2ca3310e | 1492 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 1493 | goto out; |
1da177e4 | 1494 | |
8757b776 MB |
1495 | r = dm_table_resume_targets(map); |
1496 | if (r) | |
1497 | goto out; | |
2ca3310e AK |
1498 | |
1499 | down_write(&md->io_lock); | |
1da177e4 LT |
1500 | clear_bit(DMF_BLOCK_IO, &md->flags); |
1501 | ||
1502 | def = bio_list_get(&md->deferred); | |
1503 | __flush_deferred_io(md, def); | |
2ca3310e AK |
1504 | up_write(&md->io_lock); |
1505 | ||
1506 | unlock_fs(md); | |
1507 | ||
bfa152fa JN |
1508 | if (md->suspended_bdev) { |
1509 | bdput(md->suspended_bdev); | |
1510 | md->suspended_bdev = NULL; | |
1511 | } | |
e39e2e95 | 1512 | |
2ca3310e AK |
1513 | clear_bit(DMF_SUSPENDED, &md->flags); |
1514 | ||
1da177e4 | 1515 | dm_table_unplug_all(map); |
1da177e4 | 1516 | |
8560ed6f HR |
1517 | kobject_uevent(&md->disk->kobj, KOBJ_CHANGE); |
1518 | ||
cf222b37 | 1519 | r = 0; |
2ca3310e | 1520 | |
cf222b37 AK |
1521 | out: |
1522 | dm_table_put(map); | |
2ca3310e AK |
1523 | up(&md->suspend_lock); |
1524 | ||
cf222b37 | 1525 | return r; |
1da177e4 LT |
1526 | } |
1527 | ||
1528 | /*----------------------------------------------------------------- | |
1529 | * Event notification. | |
1530 | *---------------------------------------------------------------*/ | |
7a8c3d3b MA |
1531 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
1532 | { | |
1533 | return atomic_add_return(1, &md->uevent_seq); | |
1534 | } | |
1535 | ||
1da177e4 LT |
1536 | uint32_t dm_get_event_nr(struct mapped_device *md) |
1537 | { | |
1538 | return atomic_read(&md->event_nr); | |
1539 | } | |
1540 | ||
1541 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
1542 | { | |
1543 | return wait_event_interruptible(md->eventq, | |
1544 | (event_nr != atomic_read(&md->event_nr))); | |
1545 | } | |
1546 | ||
7a8c3d3b MA |
1547 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
1548 | { | |
1549 | unsigned long flags; | |
1550 | ||
1551 | spin_lock_irqsave(&md->uevent_lock, flags); | |
1552 | list_add(elist, &md->uevent_list); | |
1553 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1554 | } | |
1555 | ||
1da177e4 LT |
1556 | /* |
1557 | * The gendisk is only valid as long as you have a reference | |
1558 | * count on 'md'. | |
1559 | */ | |
1560 | struct gendisk *dm_disk(struct mapped_device *md) | |
1561 | { | |
1562 | return md->disk; | |
1563 | } | |
1564 | ||
1565 | int dm_suspended(struct mapped_device *md) | |
1566 | { | |
1567 | return test_bit(DMF_SUSPENDED, &md->flags); | |
1568 | } | |
1569 | ||
2e93ccc1 KU |
1570 | int dm_noflush_suspending(struct dm_target *ti) |
1571 | { | |
1572 | struct mapped_device *md = dm_table_get_md(ti->table); | |
1573 | int r = __noflush_suspending(md); | |
1574 | ||
1575 | dm_put(md); | |
1576 | ||
1577 | return r; | |
1578 | } | |
1579 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
1580 | ||
1da177e4 LT |
1581 | static struct block_device_operations dm_blk_dops = { |
1582 | .open = dm_blk_open, | |
1583 | .release = dm_blk_close, | |
aa129a22 | 1584 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 1585 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
1586 | .owner = THIS_MODULE |
1587 | }; | |
1588 | ||
1589 | EXPORT_SYMBOL(dm_get_mapinfo); | |
1590 | ||
1591 | /* | |
1592 | * module hooks | |
1593 | */ | |
1594 | module_init(dm_init); | |
1595 | module_exit(dm_exit); | |
1596 | ||
1597 | module_param(major, uint, 0); | |
1598 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
1599 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
1600 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
1601 | MODULE_LICENSE("GPL"); |