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 | 73 | |
304f3f6a MB |
74 | /* |
75 | * Work processed by per-device workqueue. | |
76 | */ | |
77 | struct dm_wq_req { | |
78 | enum { | |
79 | DM_WQ_FLUSH_ALL, | |
80 | DM_WQ_FLUSH_DEFERRED, | |
81 | } type; | |
82 | struct work_struct work; | |
83 | struct mapped_device *md; | |
84 | void *context; | |
85 | }; | |
86 | ||
1da177e4 | 87 | struct mapped_device { |
2ca3310e | 88 | struct rw_semaphore io_lock; |
e61290a4 | 89 | struct mutex suspend_lock; |
2e93ccc1 | 90 | spinlock_t pushback_lock; |
1da177e4 LT |
91 | rwlock_t map_lock; |
92 | atomic_t holders; | |
5c6bd75d | 93 | atomic_t open_count; |
1da177e4 LT |
94 | |
95 | unsigned long flags; | |
96 | ||
165125e1 | 97 | struct request_queue *queue; |
1da177e4 | 98 | struct gendisk *disk; |
7e51f257 | 99 | char name[16]; |
1da177e4 LT |
100 | |
101 | void *interface_ptr; | |
102 | ||
103 | /* | |
104 | * A list of ios that arrived while we were suspended. | |
105 | */ | |
106 | atomic_t pending; | |
107 | wait_queue_head_t wait; | |
74859364 | 108 | struct bio_list deferred; |
2e93ccc1 | 109 | struct bio_list pushback; |
1da177e4 | 110 | |
304f3f6a MB |
111 | /* |
112 | * Processing queue (flush/barriers) | |
113 | */ | |
114 | struct workqueue_struct *wq; | |
115 | ||
1da177e4 LT |
116 | /* |
117 | * The current mapping. | |
118 | */ | |
119 | struct dm_table *map; | |
120 | ||
121 | /* | |
122 | * io objects are allocated from here. | |
123 | */ | |
124 | mempool_t *io_pool; | |
125 | mempool_t *tio_pool; | |
126 | ||
9faf400f SB |
127 | struct bio_set *bs; |
128 | ||
1da177e4 LT |
129 | /* |
130 | * Event handling. | |
131 | */ | |
132 | atomic_t event_nr; | |
133 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
134 | atomic_t uevent_seq; |
135 | struct list_head uevent_list; | |
136 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
137 | |
138 | /* | |
139 | * freeze/thaw support require holding onto a super block | |
140 | */ | |
141 | struct super_block *frozen_sb; | |
e39e2e95 | 142 | struct block_device *suspended_bdev; |
3ac51e74 DW |
143 | |
144 | /* forced geometry settings */ | |
145 | struct hd_geometry geometry; | |
1da177e4 LT |
146 | }; |
147 | ||
148 | #define MIN_IOS 256 | |
e18b890b CL |
149 | static struct kmem_cache *_io_cache; |
150 | static struct kmem_cache *_tio_cache; | |
1da177e4 | 151 | |
1da177e4 LT |
152 | static int __init local_init(void) |
153 | { | |
154 | int r; | |
155 | ||
1da177e4 | 156 | /* allocate a slab for the dm_ios */ |
028867ac | 157 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 LT |
158 | if (!_io_cache) |
159 | return -ENOMEM; | |
160 | ||
161 | /* allocate a slab for the target ios */ | |
028867ac | 162 | _tio_cache = KMEM_CACHE(dm_target_io, 0); |
1da177e4 LT |
163 | if (!_tio_cache) { |
164 | kmem_cache_destroy(_io_cache); | |
165 | return -ENOMEM; | |
166 | } | |
167 | ||
51e5b2bd MA |
168 | r = dm_uevent_init(); |
169 | if (r) { | |
170 | kmem_cache_destroy(_tio_cache); | |
171 | kmem_cache_destroy(_io_cache); | |
172 | return r; | |
173 | } | |
174 | ||
1da177e4 LT |
175 | _major = major; |
176 | r = register_blkdev(_major, _name); | |
177 | if (r < 0) { | |
178 | kmem_cache_destroy(_tio_cache); | |
179 | kmem_cache_destroy(_io_cache); | |
51e5b2bd | 180 | dm_uevent_exit(); |
1da177e4 LT |
181 | return r; |
182 | } | |
183 | ||
184 | if (!_major) | |
185 | _major = r; | |
186 | ||
187 | return 0; | |
188 | } | |
189 | ||
190 | static void local_exit(void) | |
191 | { | |
192 | kmem_cache_destroy(_tio_cache); | |
193 | kmem_cache_destroy(_io_cache); | |
00d59405 | 194 | unregister_blkdev(_major, _name); |
51e5b2bd | 195 | dm_uevent_exit(); |
1da177e4 LT |
196 | |
197 | _major = 0; | |
198 | ||
199 | DMINFO("cleaned up"); | |
200 | } | |
201 | ||
b9249e55 | 202 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
203 | local_init, |
204 | dm_target_init, | |
205 | dm_linear_init, | |
206 | dm_stripe_init, | |
207 | dm_interface_init, | |
208 | }; | |
209 | ||
b9249e55 | 210 | static void (*_exits[])(void) = { |
1da177e4 LT |
211 | local_exit, |
212 | dm_target_exit, | |
213 | dm_linear_exit, | |
214 | dm_stripe_exit, | |
215 | dm_interface_exit, | |
216 | }; | |
217 | ||
218 | static int __init dm_init(void) | |
219 | { | |
220 | const int count = ARRAY_SIZE(_inits); | |
221 | ||
222 | int r, i; | |
223 | ||
224 | for (i = 0; i < count; i++) { | |
225 | r = _inits[i](); | |
226 | if (r) | |
227 | goto bad; | |
228 | } | |
229 | ||
230 | return 0; | |
231 | ||
232 | bad: | |
233 | while (i--) | |
234 | _exits[i](); | |
235 | ||
236 | return r; | |
237 | } | |
238 | ||
239 | static void __exit dm_exit(void) | |
240 | { | |
241 | int i = ARRAY_SIZE(_exits); | |
242 | ||
243 | while (i--) | |
244 | _exits[i](); | |
245 | } | |
246 | ||
247 | /* | |
248 | * Block device functions | |
249 | */ | |
250 | static int dm_blk_open(struct inode *inode, struct file *file) | |
251 | { | |
252 | struct mapped_device *md; | |
253 | ||
fba9f90e JM |
254 | spin_lock(&_minor_lock); |
255 | ||
1da177e4 | 256 | md = inode->i_bdev->bd_disk->private_data; |
fba9f90e JM |
257 | if (!md) |
258 | goto out; | |
259 | ||
5c6bd75d AK |
260 | if (test_bit(DMF_FREEING, &md->flags) || |
261 | test_bit(DMF_DELETING, &md->flags)) { | |
fba9f90e JM |
262 | md = NULL; |
263 | goto out; | |
264 | } | |
265 | ||
1da177e4 | 266 | dm_get(md); |
5c6bd75d | 267 | atomic_inc(&md->open_count); |
fba9f90e JM |
268 | |
269 | out: | |
270 | spin_unlock(&_minor_lock); | |
271 | ||
272 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
273 | } |
274 | ||
275 | static int dm_blk_close(struct inode *inode, struct file *file) | |
276 | { | |
277 | struct mapped_device *md; | |
278 | ||
279 | md = inode->i_bdev->bd_disk->private_data; | |
5c6bd75d | 280 | atomic_dec(&md->open_count); |
1da177e4 LT |
281 | dm_put(md); |
282 | return 0; | |
283 | } | |
284 | ||
5c6bd75d AK |
285 | int dm_open_count(struct mapped_device *md) |
286 | { | |
287 | return atomic_read(&md->open_count); | |
288 | } | |
289 | ||
290 | /* | |
291 | * Guarantees nothing is using the device before it's deleted. | |
292 | */ | |
293 | int dm_lock_for_deletion(struct mapped_device *md) | |
294 | { | |
295 | int r = 0; | |
296 | ||
297 | spin_lock(&_minor_lock); | |
298 | ||
299 | if (dm_open_count(md)) | |
300 | r = -EBUSY; | |
301 | else | |
302 | set_bit(DMF_DELETING, &md->flags); | |
303 | ||
304 | spin_unlock(&_minor_lock); | |
305 | ||
306 | return r; | |
307 | } | |
308 | ||
3ac51e74 DW |
309 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
310 | { | |
311 | struct mapped_device *md = bdev->bd_disk->private_data; | |
312 | ||
313 | return dm_get_geometry(md, geo); | |
314 | } | |
315 | ||
aa129a22 MB |
316 | static int dm_blk_ioctl(struct inode *inode, struct file *file, |
317 | unsigned int cmd, unsigned long arg) | |
318 | { | |
319 | struct mapped_device *md; | |
320 | struct dm_table *map; | |
321 | struct dm_target *tgt; | |
322 | int r = -ENOTTY; | |
323 | ||
324 | /* We don't really need this lock, but we do need 'inode'. */ | |
325 | unlock_kernel(); | |
326 | ||
327 | md = inode->i_bdev->bd_disk->private_data; | |
328 | ||
329 | map = dm_get_table(md); | |
330 | ||
331 | if (!map || !dm_table_get_size(map)) | |
332 | goto out; | |
333 | ||
334 | /* We only support devices that have a single target */ | |
335 | if (dm_table_get_num_targets(map) != 1) | |
336 | goto out; | |
337 | ||
338 | tgt = dm_table_get_target(map, 0); | |
339 | ||
340 | if (dm_suspended(md)) { | |
341 | r = -EAGAIN; | |
342 | goto out; | |
343 | } | |
344 | ||
345 | if (tgt->type->ioctl) | |
346 | r = tgt->type->ioctl(tgt, inode, file, cmd, arg); | |
347 | ||
348 | out: | |
349 | dm_table_put(map); | |
350 | ||
351 | lock_kernel(); | |
352 | return r; | |
353 | } | |
354 | ||
028867ac | 355 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
356 | { |
357 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
358 | } | |
359 | ||
028867ac | 360 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
361 | { |
362 | mempool_free(io, md->io_pool); | |
363 | } | |
364 | ||
028867ac | 365 | static struct dm_target_io *alloc_tio(struct mapped_device *md) |
1da177e4 LT |
366 | { |
367 | return mempool_alloc(md->tio_pool, GFP_NOIO); | |
368 | } | |
369 | ||
028867ac | 370 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 LT |
371 | { |
372 | mempool_free(tio, md->tio_pool); | |
373 | } | |
374 | ||
3eaf840e JNN |
375 | static void start_io_acct(struct dm_io *io) |
376 | { | |
377 | struct mapped_device *md = io->md; | |
378 | ||
379 | io->start_time = jiffies; | |
380 | ||
381 | preempt_disable(); | |
382 | disk_round_stats(dm_disk(md)); | |
383 | preempt_enable(); | |
384 | dm_disk(md)->in_flight = atomic_inc_return(&md->pending); | |
385 | } | |
386 | ||
387 | static int end_io_acct(struct dm_io *io) | |
388 | { | |
389 | struct mapped_device *md = io->md; | |
390 | struct bio *bio = io->bio; | |
391 | unsigned long duration = jiffies - io->start_time; | |
392 | int pending; | |
393 | int rw = bio_data_dir(bio); | |
394 | ||
395 | preempt_disable(); | |
396 | disk_round_stats(dm_disk(md)); | |
397 | preempt_enable(); | |
398 | dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending); | |
399 | ||
400 | disk_stat_add(dm_disk(md), ticks[rw], duration); | |
401 | ||
402 | return !pending; | |
403 | } | |
404 | ||
1da177e4 LT |
405 | /* |
406 | * Add the bio to the list of deferred io. | |
407 | */ | |
408 | static int queue_io(struct mapped_device *md, struct bio *bio) | |
409 | { | |
2ca3310e | 410 | down_write(&md->io_lock); |
1da177e4 LT |
411 | |
412 | if (!test_bit(DMF_BLOCK_IO, &md->flags)) { | |
2ca3310e | 413 | up_write(&md->io_lock); |
1da177e4 LT |
414 | return 1; |
415 | } | |
416 | ||
417 | bio_list_add(&md->deferred, bio); | |
418 | ||
2ca3310e | 419 | up_write(&md->io_lock); |
1da177e4 LT |
420 | return 0; /* deferred successfully */ |
421 | } | |
422 | ||
423 | /* | |
424 | * Everyone (including functions in this file), should use this | |
425 | * function to access the md->map field, and make sure they call | |
426 | * dm_table_put() when finished. | |
427 | */ | |
428 | struct dm_table *dm_get_table(struct mapped_device *md) | |
429 | { | |
430 | struct dm_table *t; | |
431 | ||
432 | read_lock(&md->map_lock); | |
433 | t = md->map; | |
434 | if (t) | |
435 | dm_table_get(t); | |
436 | read_unlock(&md->map_lock); | |
437 | ||
438 | return t; | |
439 | } | |
440 | ||
3ac51e74 DW |
441 | /* |
442 | * Get the geometry associated with a dm device | |
443 | */ | |
444 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
445 | { | |
446 | *geo = md->geometry; | |
447 | ||
448 | return 0; | |
449 | } | |
450 | ||
451 | /* | |
452 | * Set the geometry of a device. | |
453 | */ | |
454 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
455 | { | |
456 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
457 | ||
458 | if (geo->start > sz) { | |
459 | DMWARN("Start sector is beyond the geometry limits."); | |
460 | return -EINVAL; | |
461 | } | |
462 | ||
463 | md->geometry = *geo; | |
464 | ||
465 | return 0; | |
466 | } | |
467 | ||
1da177e4 LT |
468 | /*----------------------------------------------------------------- |
469 | * CRUD START: | |
470 | * A more elegant soln is in the works that uses the queue | |
471 | * merge fn, unfortunately there are a couple of changes to | |
472 | * the block layer that I want to make for this. So in the | |
473 | * interests of getting something for people to use I give | |
474 | * you this clearly demarcated crap. | |
475 | *---------------------------------------------------------------*/ | |
476 | ||
2e93ccc1 KU |
477 | static int __noflush_suspending(struct mapped_device *md) |
478 | { | |
479 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
480 | } | |
481 | ||
1da177e4 LT |
482 | /* |
483 | * Decrements the number of outstanding ios that a bio has been | |
484 | * cloned into, completing the original io if necc. | |
485 | */ | |
858119e1 | 486 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 487 | { |
2e93ccc1 KU |
488 | unsigned long flags; |
489 | ||
490 | /* Push-back supersedes any I/O errors */ | |
491 | if (error && !(io->error > 0 && __noflush_suspending(io->md))) | |
1da177e4 LT |
492 | io->error = error; |
493 | ||
494 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
495 | if (io->error == DM_ENDIO_REQUEUE) { |
496 | /* | |
497 | * Target requested pushing back the I/O. | |
498 | * This must be handled before the sleeper on | |
499 | * suspend queue merges the pushback list. | |
500 | */ | |
501 | spin_lock_irqsave(&io->md->pushback_lock, flags); | |
502 | if (__noflush_suspending(io->md)) | |
503 | bio_list_add(&io->md->pushback, io->bio); | |
504 | else | |
505 | /* noflush suspend was interrupted. */ | |
506 | io->error = -EIO; | |
507 | spin_unlock_irqrestore(&io->md->pushback_lock, flags); | |
508 | } | |
509 | ||
3eaf840e | 510 | if (end_io_acct(io)) |
1da177e4 LT |
511 | /* nudge anyone waiting on suspend queue */ |
512 | wake_up(&io->md->wait); | |
513 | ||
2e93ccc1 KU |
514 | if (io->error != DM_ENDIO_REQUEUE) { |
515 | blk_add_trace_bio(io->md->queue, io->bio, | |
516 | BLK_TA_COMPLETE); | |
517 | ||
6712ecf8 | 518 | bio_endio(io->bio, io->error); |
2e93ccc1 | 519 | } |
2056a782 | 520 | |
1da177e4 LT |
521 | free_io(io->md, io); |
522 | } | |
523 | } | |
524 | ||
6712ecf8 | 525 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
526 | { |
527 | int r = 0; | |
028867ac | 528 | struct dm_target_io *tio = bio->bi_private; |
9faf400f | 529 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
530 | dm_endio_fn endio = tio->ti->type->end_io; |
531 | ||
1da177e4 LT |
532 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
533 | error = -EIO; | |
534 | ||
535 | if (endio) { | |
536 | r = endio(tio->ti, bio, error, &tio->info); | |
2e93ccc1 KU |
537 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
538 | /* | |
539 | * error and requeue request are handled | |
540 | * in dec_pending(). | |
541 | */ | |
1da177e4 | 542 | error = r; |
45cbcd79 KU |
543 | else if (r == DM_ENDIO_INCOMPLETE) |
544 | /* The target will handle the io */ | |
6712ecf8 | 545 | return; |
45cbcd79 KU |
546 | else if (r) { |
547 | DMWARN("unimplemented target endio return value: %d", r); | |
548 | BUG(); | |
549 | } | |
1da177e4 LT |
550 | } |
551 | ||
9faf400f SB |
552 | dec_pending(tio->io, error); |
553 | ||
554 | /* | |
555 | * Store md for cleanup instead of tio which is about to get freed. | |
556 | */ | |
557 | bio->bi_private = md->bs; | |
558 | ||
1da177e4 | 559 | bio_put(bio); |
9faf400f | 560 | free_tio(md, tio); |
1da177e4 LT |
561 | } |
562 | ||
563 | static sector_t max_io_len(struct mapped_device *md, | |
564 | sector_t sector, struct dm_target *ti) | |
565 | { | |
566 | sector_t offset = sector - ti->begin; | |
567 | sector_t len = ti->len - offset; | |
568 | ||
569 | /* | |
570 | * Does the target need to split even further ? | |
571 | */ | |
572 | if (ti->split_io) { | |
573 | sector_t boundary; | |
574 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) | |
575 | - offset; | |
576 | if (len > boundary) | |
577 | len = boundary; | |
578 | } | |
579 | ||
580 | return len; | |
581 | } | |
582 | ||
583 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
028867ac | 584 | struct dm_target_io *tio) |
1da177e4 LT |
585 | { |
586 | int r; | |
2056a782 | 587 | sector_t sector; |
9faf400f | 588 | struct mapped_device *md; |
1da177e4 LT |
589 | |
590 | /* | |
591 | * Sanity checks. | |
592 | */ | |
593 | BUG_ON(!clone->bi_size); | |
594 | ||
595 | clone->bi_end_io = clone_endio; | |
596 | clone->bi_private = tio; | |
597 | ||
598 | /* | |
599 | * Map the clone. If r == 0 we don't need to do | |
600 | * anything, the target has assumed ownership of | |
601 | * this io. | |
602 | */ | |
603 | atomic_inc(&tio->io->io_count); | |
2056a782 | 604 | sector = clone->bi_sector; |
1da177e4 | 605 | r = ti->type->map(ti, clone, &tio->info); |
45cbcd79 | 606 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 607 | /* the bio has been remapped so dispatch it */ |
2056a782 | 608 | |
17b2f66f | 609 | blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone, |
c7149d6b AB |
610 | tio->io->bio->bi_bdev->bd_dev, |
611 | clone->bi_sector, sector); | |
2056a782 | 612 | |
1da177e4 | 613 | generic_make_request(clone); |
2e93ccc1 KU |
614 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
615 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
616 | md = tio->io->md; |
617 | dec_pending(tio->io, r); | |
618 | /* | |
619 | * Store bio_set for cleanup. | |
620 | */ | |
621 | clone->bi_private = md->bs; | |
1da177e4 | 622 | bio_put(clone); |
9faf400f | 623 | free_tio(md, tio); |
45cbcd79 KU |
624 | } else if (r) { |
625 | DMWARN("unimplemented target map return value: %d", r); | |
626 | BUG(); | |
1da177e4 LT |
627 | } |
628 | } | |
629 | ||
630 | struct clone_info { | |
631 | struct mapped_device *md; | |
632 | struct dm_table *map; | |
633 | struct bio *bio; | |
634 | struct dm_io *io; | |
635 | sector_t sector; | |
636 | sector_t sector_count; | |
637 | unsigned short idx; | |
638 | }; | |
639 | ||
3676347a PO |
640 | static void dm_bio_destructor(struct bio *bio) |
641 | { | |
9faf400f SB |
642 | struct bio_set *bs = bio->bi_private; |
643 | ||
644 | bio_free(bio, bs); | |
3676347a PO |
645 | } |
646 | ||
1da177e4 LT |
647 | /* |
648 | * Creates a little bio that is just does part of a bvec. | |
649 | */ | |
650 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
651 | unsigned short idx, unsigned int offset, | |
9faf400f | 652 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
653 | { |
654 | struct bio *clone; | |
655 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
656 | ||
9faf400f | 657 | clone = bio_alloc_bioset(GFP_NOIO, 1, bs); |
3676347a | 658 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
659 | *clone->bi_io_vec = *bv; |
660 | ||
661 | clone->bi_sector = sector; | |
662 | clone->bi_bdev = bio->bi_bdev; | |
663 | clone->bi_rw = bio->bi_rw; | |
664 | clone->bi_vcnt = 1; | |
665 | clone->bi_size = to_bytes(len); | |
666 | clone->bi_io_vec->bv_offset = offset; | |
667 | clone->bi_io_vec->bv_len = clone->bi_size; | |
668 | ||
669 | return clone; | |
670 | } | |
671 | ||
672 | /* | |
673 | * Creates a bio that consists of range of complete bvecs. | |
674 | */ | |
675 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
676 | unsigned short idx, unsigned short bv_count, | |
9faf400f | 677 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
678 | { |
679 | struct bio *clone; | |
680 | ||
9faf400f SB |
681 | clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); |
682 | __bio_clone(clone, bio); | |
683 | clone->bi_destructor = dm_bio_destructor; | |
1da177e4 LT |
684 | clone->bi_sector = sector; |
685 | clone->bi_idx = idx; | |
686 | clone->bi_vcnt = idx + bv_count; | |
687 | clone->bi_size = to_bytes(len); | |
688 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
689 | ||
690 | return clone; | |
691 | } | |
692 | ||
512875bd | 693 | static int __clone_and_map(struct clone_info *ci) |
1da177e4 LT |
694 | { |
695 | struct bio *clone, *bio = ci->bio; | |
512875bd JN |
696 | struct dm_target *ti; |
697 | sector_t len = 0, max; | |
028867ac | 698 | struct dm_target_io *tio; |
1da177e4 | 699 | |
512875bd JN |
700 | ti = dm_table_find_target(ci->map, ci->sector); |
701 | if (!dm_target_is_valid(ti)) | |
702 | return -EIO; | |
703 | ||
704 | max = max_io_len(ci->md, ci->sector, ti); | |
705 | ||
1da177e4 LT |
706 | /* |
707 | * Allocate a target io object. | |
708 | */ | |
709 | tio = alloc_tio(ci->md); | |
710 | tio->io = ci->io; | |
711 | tio->ti = ti; | |
712 | memset(&tio->info, 0, sizeof(tio->info)); | |
713 | ||
714 | if (ci->sector_count <= max) { | |
715 | /* | |
716 | * Optimise for the simple case where we can do all of | |
717 | * the remaining io with a single clone. | |
718 | */ | |
719 | clone = clone_bio(bio, ci->sector, ci->idx, | |
9faf400f SB |
720 | bio->bi_vcnt - ci->idx, ci->sector_count, |
721 | ci->md->bs); | |
1da177e4 LT |
722 | __map_bio(ti, clone, tio); |
723 | ci->sector_count = 0; | |
724 | ||
725 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
726 | /* | |
727 | * There are some bvecs that don't span targets. | |
728 | * Do as many of these as possible. | |
729 | */ | |
730 | int i; | |
731 | sector_t remaining = max; | |
732 | sector_t bv_len; | |
733 | ||
734 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
735 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
736 | ||
737 | if (bv_len > remaining) | |
738 | break; | |
739 | ||
740 | remaining -= bv_len; | |
741 | len += bv_len; | |
742 | } | |
743 | ||
9faf400f SB |
744 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, |
745 | ci->md->bs); | |
1da177e4 LT |
746 | __map_bio(ti, clone, tio); |
747 | ||
748 | ci->sector += len; | |
749 | ci->sector_count -= len; | |
750 | ci->idx = i; | |
751 | ||
752 | } else { | |
753 | /* | |
d2044a94 | 754 | * Handle a bvec that must be split between two or more targets. |
1da177e4 LT |
755 | */ |
756 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
d2044a94 AK |
757 | sector_t remaining = to_sector(bv->bv_len); |
758 | unsigned int offset = 0; | |
1da177e4 | 759 | |
d2044a94 AK |
760 | do { |
761 | if (offset) { | |
762 | ti = dm_table_find_target(ci->map, ci->sector); | |
512875bd JN |
763 | if (!dm_target_is_valid(ti)) |
764 | return -EIO; | |
765 | ||
d2044a94 | 766 | max = max_io_len(ci->md, ci->sector, ti); |
1da177e4 | 767 | |
d2044a94 AK |
768 | tio = alloc_tio(ci->md); |
769 | tio->io = ci->io; | |
770 | tio->ti = ti; | |
771 | memset(&tio->info, 0, sizeof(tio->info)); | |
772 | } | |
773 | ||
774 | len = min(remaining, max); | |
775 | ||
776 | clone = split_bvec(bio, ci->sector, ci->idx, | |
9faf400f SB |
777 | bv->bv_offset + offset, len, |
778 | ci->md->bs); | |
d2044a94 AK |
779 | |
780 | __map_bio(ti, clone, tio); | |
781 | ||
782 | ci->sector += len; | |
783 | ci->sector_count -= len; | |
784 | offset += to_bytes(len); | |
785 | } while (remaining -= len); | |
1da177e4 | 786 | |
1da177e4 LT |
787 | ci->idx++; |
788 | } | |
512875bd JN |
789 | |
790 | return 0; | |
1da177e4 LT |
791 | } |
792 | ||
793 | /* | |
794 | * Split the bio into several clones. | |
795 | */ | |
9e4e5f87 | 796 | static int __split_bio(struct mapped_device *md, struct bio *bio) |
1da177e4 LT |
797 | { |
798 | struct clone_info ci; | |
512875bd | 799 | int error = 0; |
1da177e4 LT |
800 | |
801 | ci.map = dm_get_table(md); | |
9e4e5f87 MB |
802 | if (unlikely(!ci.map)) |
803 | return -EIO; | |
1da177e4 LT |
804 | |
805 | ci.md = md; | |
806 | ci.bio = bio; | |
807 | ci.io = alloc_io(md); | |
808 | ci.io->error = 0; | |
809 | atomic_set(&ci.io->io_count, 1); | |
810 | ci.io->bio = bio; | |
811 | ci.io->md = md; | |
812 | ci.sector = bio->bi_sector; | |
813 | ci.sector_count = bio_sectors(bio); | |
814 | ci.idx = bio->bi_idx; | |
815 | ||
3eaf840e | 816 | start_io_acct(ci.io); |
512875bd JN |
817 | while (ci.sector_count && !error) |
818 | error = __clone_and_map(&ci); | |
1da177e4 LT |
819 | |
820 | /* drop the extra reference count */ | |
512875bd | 821 | dec_pending(ci.io, error); |
1da177e4 | 822 | dm_table_put(ci.map); |
9e4e5f87 MB |
823 | |
824 | return 0; | |
1da177e4 LT |
825 | } |
826 | /*----------------------------------------------------------------- | |
827 | * CRUD END | |
828 | *---------------------------------------------------------------*/ | |
829 | ||
830 | /* | |
831 | * The request function that just remaps the bio built up by | |
832 | * dm_merge_bvec. | |
833 | */ | |
165125e1 | 834 | static int dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 835 | { |
9e4e5f87 | 836 | int r = -EIO; |
12f03a49 | 837 | int rw = bio_data_dir(bio); |
1da177e4 LT |
838 | struct mapped_device *md = q->queuedata; |
839 | ||
07a83c47 SB |
840 | /* |
841 | * There is no use in forwarding any barrier request since we can't | |
842 | * guarantee it is (or can be) handled by the targets correctly. | |
843 | */ | |
844 | if (unlikely(bio_barrier(bio))) { | |
6712ecf8 | 845 | bio_endio(bio, -EOPNOTSUPP); |
07a83c47 SB |
846 | return 0; |
847 | } | |
848 | ||
2ca3310e | 849 | down_read(&md->io_lock); |
1da177e4 | 850 | |
12f03a49 KC |
851 | disk_stat_inc(dm_disk(md), ios[rw]); |
852 | disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio)); | |
853 | ||
1da177e4 LT |
854 | /* |
855 | * If we're suspended we have to queue | |
856 | * this io for later. | |
857 | */ | |
858 | while (test_bit(DMF_BLOCK_IO, &md->flags)) { | |
2ca3310e | 859 | up_read(&md->io_lock); |
1da177e4 | 860 | |
9e4e5f87 MB |
861 | if (bio_rw(bio) != READA) |
862 | r = queue_io(md, bio); | |
1da177e4 | 863 | |
9e4e5f87 MB |
864 | if (r <= 0) |
865 | goto out_req; | |
1da177e4 LT |
866 | |
867 | /* | |
868 | * We're in a while loop, because someone could suspend | |
869 | * before we get to the following read lock. | |
870 | */ | |
2ca3310e | 871 | down_read(&md->io_lock); |
1da177e4 LT |
872 | } |
873 | ||
9e4e5f87 | 874 | r = __split_bio(md, bio); |
2ca3310e | 875 | up_read(&md->io_lock); |
9e4e5f87 MB |
876 | |
877 | out_req: | |
878 | if (r < 0) | |
879 | bio_io_error(bio); | |
880 | ||
1da177e4 LT |
881 | return 0; |
882 | } | |
883 | ||
165125e1 | 884 | static void dm_unplug_all(struct request_queue *q) |
1da177e4 LT |
885 | { |
886 | struct mapped_device *md = q->queuedata; | |
887 | struct dm_table *map = dm_get_table(md); | |
888 | ||
889 | if (map) { | |
890 | dm_table_unplug_all(map); | |
891 | dm_table_put(map); | |
892 | } | |
893 | } | |
894 | ||
895 | static int dm_any_congested(void *congested_data, int bdi_bits) | |
896 | { | |
897 | int r; | |
898 | struct mapped_device *md = (struct mapped_device *) congested_data; | |
899 | struct dm_table *map = dm_get_table(md); | |
900 | ||
901 | if (!map || test_bit(DMF_BLOCK_IO, &md->flags)) | |
902 | r = bdi_bits; | |
903 | else | |
904 | r = dm_table_any_congested(map, bdi_bits); | |
905 | ||
906 | dm_table_put(map); | |
907 | return r; | |
908 | } | |
909 | ||
910 | /*----------------------------------------------------------------- | |
911 | * An IDR is used to keep track of allocated minor numbers. | |
912 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
913 | static DEFINE_IDR(_minor_idr); |
914 | ||
2b06cfff | 915 | static void free_minor(int minor) |
1da177e4 | 916 | { |
f32c10b0 | 917 | spin_lock(&_minor_lock); |
1da177e4 | 918 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 919 | spin_unlock(&_minor_lock); |
1da177e4 LT |
920 | } |
921 | ||
922 | /* | |
923 | * See if the device with a specific minor # is free. | |
924 | */ | |
2b06cfff | 925 | static int specific_minor(struct mapped_device *md, int minor) |
1da177e4 LT |
926 | { |
927 | int r, m; | |
928 | ||
929 | if (minor >= (1 << MINORBITS)) | |
930 | return -EINVAL; | |
931 | ||
62f75c2f JM |
932 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
933 | if (!r) | |
934 | return -ENOMEM; | |
935 | ||
f32c10b0 | 936 | spin_lock(&_minor_lock); |
1da177e4 LT |
937 | |
938 | if (idr_find(&_minor_idr, minor)) { | |
939 | r = -EBUSY; | |
940 | goto out; | |
941 | } | |
942 | ||
ba61fdd1 | 943 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); |
62f75c2f | 944 | if (r) |
1da177e4 | 945 | goto out; |
1da177e4 LT |
946 | |
947 | if (m != minor) { | |
948 | idr_remove(&_minor_idr, m); | |
949 | r = -EBUSY; | |
950 | goto out; | |
951 | } | |
952 | ||
953 | out: | |
f32c10b0 | 954 | spin_unlock(&_minor_lock); |
1da177e4 LT |
955 | return r; |
956 | } | |
957 | ||
2b06cfff | 958 | static int next_free_minor(struct mapped_device *md, int *minor) |
1da177e4 | 959 | { |
2b06cfff | 960 | int r, m; |
1da177e4 | 961 | |
1da177e4 | 962 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
62f75c2f JM |
963 | if (!r) |
964 | return -ENOMEM; | |
965 | ||
f32c10b0 | 966 | spin_lock(&_minor_lock); |
1da177e4 | 967 | |
ba61fdd1 | 968 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); |
1da177e4 LT |
969 | if (r) { |
970 | goto out; | |
971 | } | |
972 | ||
973 | if (m >= (1 << MINORBITS)) { | |
974 | idr_remove(&_minor_idr, m); | |
975 | r = -ENOSPC; | |
976 | goto out; | |
977 | } | |
978 | ||
979 | *minor = m; | |
980 | ||
981 | out: | |
f32c10b0 | 982 | spin_unlock(&_minor_lock); |
1da177e4 LT |
983 | return r; |
984 | } | |
985 | ||
986 | static struct block_device_operations dm_blk_dops; | |
987 | ||
988 | /* | |
989 | * Allocate and initialise a blank device with a given minor. | |
990 | */ | |
2b06cfff | 991 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
992 | { |
993 | int r; | |
994 | struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL); | |
ba61fdd1 | 995 | void *old_md; |
1da177e4 LT |
996 | |
997 | if (!md) { | |
998 | DMWARN("unable to allocate device, out of memory."); | |
999 | return NULL; | |
1000 | } | |
1001 | ||
10da4f79 | 1002 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1003 | goto bad_module_get; |
10da4f79 | 1004 | |
1da177e4 | 1005 | /* get a minor number for the dev */ |
2b06cfff AK |
1006 | if (minor == DM_ANY_MINOR) |
1007 | r = next_free_minor(md, &minor); | |
1008 | else | |
1009 | r = specific_minor(md, minor); | |
1da177e4 | 1010 | if (r < 0) |
6ed7ade8 | 1011 | goto bad_minor; |
1da177e4 LT |
1012 | |
1013 | memset(md, 0, sizeof(*md)); | |
2ca3310e | 1014 | init_rwsem(&md->io_lock); |
e61290a4 | 1015 | mutex_init(&md->suspend_lock); |
2e93ccc1 | 1016 | spin_lock_init(&md->pushback_lock); |
1da177e4 LT |
1017 | rwlock_init(&md->map_lock); |
1018 | atomic_set(&md->holders, 1); | |
5c6bd75d | 1019 | atomic_set(&md->open_count, 0); |
1da177e4 | 1020 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1021 | atomic_set(&md->uevent_seq, 0); |
1022 | INIT_LIST_HEAD(&md->uevent_list); | |
1023 | spin_lock_init(&md->uevent_lock); | |
1da177e4 LT |
1024 | |
1025 | md->queue = blk_alloc_queue(GFP_KERNEL); | |
1026 | if (!md->queue) | |
6ed7ade8 | 1027 | goto bad_queue; |
1da177e4 LT |
1028 | |
1029 | md->queue->queuedata = md; | |
1030 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1031 | md->queue->backing_dev_info.congested_data = md; | |
1032 | blk_queue_make_request(md->queue, dm_request); | |
daef265f | 1033 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
1da177e4 | 1034 | md->queue->unplug_fn = dm_unplug_all; |
1da177e4 | 1035 | |
93d2341c | 1036 | md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache); |
74859364 | 1037 | if (!md->io_pool) |
6ed7ade8 | 1038 | goto bad_io_pool; |
1da177e4 | 1039 | |
93d2341c | 1040 | md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache); |
1da177e4 | 1041 | if (!md->tio_pool) |
6ed7ade8 | 1042 | goto bad_tio_pool; |
1da177e4 | 1043 | |
5972511b | 1044 | md->bs = bioset_create(16, 16); |
9faf400f SB |
1045 | if (!md->bs) |
1046 | goto bad_no_bioset; | |
1047 | ||
1da177e4 LT |
1048 | md->disk = alloc_disk(1); |
1049 | if (!md->disk) | |
6ed7ade8 | 1050 | goto bad_disk; |
1da177e4 | 1051 | |
f0b04115 JM |
1052 | atomic_set(&md->pending, 0); |
1053 | init_waitqueue_head(&md->wait); | |
1054 | init_waitqueue_head(&md->eventq); | |
1055 | ||
1da177e4 LT |
1056 | md->disk->major = _major; |
1057 | md->disk->first_minor = minor; | |
1058 | md->disk->fops = &dm_blk_dops; | |
1059 | md->disk->queue = md->queue; | |
1060 | md->disk->private_data = md; | |
1061 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1062 | add_disk(md->disk); | |
7e51f257 | 1063 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1064 | |
304f3f6a MB |
1065 | md->wq = create_singlethread_workqueue("kdmflush"); |
1066 | if (!md->wq) | |
1067 | goto bad_thread; | |
1068 | ||
ba61fdd1 | 1069 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1070 | spin_lock(&_minor_lock); |
ba61fdd1 | 1071 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1072 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1073 | |
1074 | BUG_ON(old_md != MINOR_ALLOCED); | |
1075 | ||
1da177e4 LT |
1076 | return md; |
1077 | ||
304f3f6a MB |
1078 | bad_thread: |
1079 | put_disk(md->disk); | |
6ed7ade8 | 1080 | bad_disk: |
9faf400f | 1081 | bioset_free(md->bs); |
6ed7ade8 | 1082 | bad_no_bioset: |
1da177e4 | 1083 | mempool_destroy(md->tio_pool); |
6ed7ade8 | 1084 | bad_tio_pool: |
1da177e4 | 1085 | mempool_destroy(md->io_pool); |
6ed7ade8 | 1086 | bad_io_pool: |
1312f40e | 1087 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 1088 | bad_queue: |
1da177e4 | 1089 | free_minor(minor); |
6ed7ade8 | 1090 | bad_minor: |
10da4f79 | 1091 | module_put(THIS_MODULE); |
6ed7ade8 | 1092 | bad_module_get: |
1da177e4 LT |
1093 | kfree(md); |
1094 | return NULL; | |
1095 | } | |
1096 | ||
ae9da83f JN |
1097 | static void unlock_fs(struct mapped_device *md); |
1098 | ||
1da177e4 LT |
1099 | static void free_dev(struct mapped_device *md) |
1100 | { | |
2b06cfff | 1101 | int minor = md->disk->first_minor; |
63d94e48 | 1102 | |
d9dde59b | 1103 | if (md->suspended_bdev) { |
ae9da83f | 1104 | unlock_fs(md); |
d9dde59b JN |
1105 | bdput(md->suspended_bdev); |
1106 | } | |
304f3f6a | 1107 | destroy_workqueue(md->wq); |
1da177e4 LT |
1108 | mempool_destroy(md->tio_pool); |
1109 | mempool_destroy(md->io_pool); | |
9faf400f | 1110 | bioset_free(md->bs); |
1da177e4 | 1111 | del_gendisk(md->disk); |
63d94e48 | 1112 | free_minor(minor); |
fba9f90e JM |
1113 | |
1114 | spin_lock(&_minor_lock); | |
1115 | md->disk->private_data = NULL; | |
1116 | spin_unlock(&_minor_lock); | |
1117 | ||
1da177e4 | 1118 | put_disk(md->disk); |
1312f40e | 1119 | blk_cleanup_queue(md->queue); |
10da4f79 | 1120 | module_put(THIS_MODULE); |
1da177e4 LT |
1121 | kfree(md); |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * Bind a table to the device. | |
1126 | */ | |
1127 | static void event_callback(void *context) | |
1128 | { | |
7a8c3d3b MA |
1129 | unsigned long flags; |
1130 | LIST_HEAD(uevents); | |
1da177e4 LT |
1131 | struct mapped_device *md = (struct mapped_device *) context; |
1132 | ||
7a8c3d3b MA |
1133 | spin_lock_irqsave(&md->uevent_lock, flags); |
1134 | list_splice_init(&md->uevent_list, &uevents); | |
1135 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1136 | ||
edfaa7c3 | 1137 | dm_send_uevents(&uevents, &md->disk->dev.kobj); |
7a8c3d3b | 1138 | |
1da177e4 LT |
1139 | atomic_inc(&md->event_nr); |
1140 | wake_up(&md->eventq); | |
1141 | } | |
1142 | ||
4e90188b | 1143 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1144 | { |
4e90188b | 1145 | set_capacity(md->disk, size); |
1da177e4 | 1146 | |
1b1dcc1b | 1147 | mutex_lock(&md->suspended_bdev->bd_inode->i_mutex); |
e39e2e95 | 1148 | i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1b1dcc1b | 1149 | mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex); |
1da177e4 LT |
1150 | } |
1151 | ||
1152 | static int __bind(struct mapped_device *md, struct dm_table *t) | |
1153 | { | |
165125e1 | 1154 | struct request_queue *q = md->queue; |
1da177e4 LT |
1155 | sector_t size; |
1156 | ||
1157 | size = dm_table_get_size(t); | |
3ac51e74 DW |
1158 | |
1159 | /* | |
1160 | * Wipe any geometry if the size of the table changed. | |
1161 | */ | |
1162 | if (size != get_capacity(md->disk)) | |
1163 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
1164 | ||
bfa152fa JN |
1165 | if (md->suspended_bdev) |
1166 | __set_size(md, size); | |
1da177e4 LT |
1167 | if (size == 0) |
1168 | return 0; | |
1169 | ||
2ca3310e AK |
1170 | dm_table_get(t); |
1171 | dm_table_event_callback(t, event_callback, md); | |
1172 | ||
1da177e4 LT |
1173 | write_lock(&md->map_lock); |
1174 | md->map = t; | |
2ca3310e | 1175 | dm_table_set_restrictions(t, q); |
1da177e4 LT |
1176 | write_unlock(&md->map_lock); |
1177 | ||
1da177e4 LT |
1178 | return 0; |
1179 | } | |
1180 | ||
1181 | static void __unbind(struct mapped_device *md) | |
1182 | { | |
1183 | struct dm_table *map = md->map; | |
1184 | ||
1185 | if (!map) | |
1186 | return; | |
1187 | ||
1188 | dm_table_event_callback(map, NULL, NULL); | |
1189 | write_lock(&md->map_lock); | |
1190 | md->map = NULL; | |
1191 | write_unlock(&md->map_lock); | |
1192 | dm_table_put(map); | |
1193 | } | |
1194 | ||
1195 | /* | |
1196 | * Constructor for a new device. | |
1197 | */ | |
2b06cfff | 1198 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
1199 | { |
1200 | struct mapped_device *md; | |
1201 | ||
2b06cfff | 1202 | md = alloc_dev(minor); |
1da177e4 LT |
1203 | if (!md) |
1204 | return -ENXIO; | |
1205 | ||
1206 | *result = md; | |
1207 | return 0; | |
1208 | } | |
1209 | ||
637842cf | 1210 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
1211 | { |
1212 | struct mapped_device *md; | |
1da177e4 LT |
1213 | unsigned minor = MINOR(dev); |
1214 | ||
1215 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
1216 | return NULL; | |
1217 | ||
f32c10b0 | 1218 | spin_lock(&_minor_lock); |
1da177e4 LT |
1219 | |
1220 | md = idr_find(&_minor_idr, minor); | |
fba9f90e JM |
1221 | if (md && (md == MINOR_ALLOCED || |
1222 | (dm_disk(md)->first_minor != minor) || | |
17b2f66f | 1223 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 1224 | md = NULL; |
fba9f90e JM |
1225 | goto out; |
1226 | } | |
1da177e4 | 1227 | |
fba9f90e | 1228 | out: |
f32c10b0 | 1229 | spin_unlock(&_minor_lock); |
1da177e4 | 1230 | |
637842cf DT |
1231 | return md; |
1232 | } | |
1233 | ||
d229a958 DT |
1234 | struct mapped_device *dm_get_md(dev_t dev) |
1235 | { | |
1236 | struct mapped_device *md = dm_find_md(dev); | |
1237 | ||
1238 | if (md) | |
1239 | dm_get(md); | |
1240 | ||
1241 | return md; | |
1242 | } | |
1243 | ||
9ade92a9 | 1244 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 1245 | { |
9ade92a9 | 1246 | return md->interface_ptr; |
1da177e4 LT |
1247 | } |
1248 | ||
1249 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
1250 | { | |
1251 | md->interface_ptr = ptr; | |
1252 | } | |
1253 | ||
1254 | void dm_get(struct mapped_device *md) | |
1255 | { | |
1256 | atomic_inc(&md->holders); | |
1257 | } | |
1258 | ||
72d94861 AK |
1259 | const char *dm_device_name(struct mapped_device *md) |
1260 | { | |
1261 | return md->name; | |
1262 | } | |
1263 | EXPORT_SYMBOL_GPL(dm_device_name); | |
1264 | ||
1da177e4 LT |
1265 | void dm_put(struct mapped_device *md) |
1266 | { | |
1134e5ae | 1267 | struct dm_table *map; |
1da177e4 | 1268 | |
fba9f90e JM |
1269 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1270 | ||
f32c10b0 | 1271 | if (atomic_dec_and_lock(&md->holders, &_minor_lock)) { |
1134e5ae | 1272 | map = dm_get_table(md); |
ba61fdd1 | 1273 | idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor); |
fba9f90e | 1274 | set_bit(DMF_FREEING, &md->flags); |
f32c10b0 | 1275 | spin_unlock(&_minor_lock); |
cf222b37 | 1276 | if (!dm_suspended(md)) { |
1da177e4 LT |
1277 | dm_table_presuspend_targets(map); |
1278 | dm_table_postsuspend_targets(map); | |
1279 | } | |
1280 | __unbind(md); | |
1134e5ae | 1281 | dm_table_put(map); |
1da177e4 LT |
1282 | free_dev(md); |
1283 | } | |
1da177e4 | 1284 | } |
79eb885c | 1285 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 1286 | |
46125c1c MB |
1287 | static int dm_wait_for_completion(struct mapped_device *md) |
1288 | { | |
1289 | int r = 0; | |
1290 | ||
1291 | while (1) { | |
1292 | set_current_state(TASK_INTERRUPTIBLE); | |
1293 | ||
1294 | smp_mb(); | |
1295 | if (!atomic_read(&md->pending)) | |
1296 | break; | |
1297 | ||
1298 | if (signal_pending(current)) { | |
1299 | r = -EINTR; | |
1300 | break; | |
1301 | } | |
1302 | ||
1303 | io_schedule(); | |
1304 | } | |
1305 | set_current_state(TASK_RUNNING); | |
1306 | ||
1307 | return r; | |
1308 | } | |
1309 | ||
1da177e4 LT |
1310 | /* |
1311 | * Process the deferred bios | |
1312 | */ | |
6d6f10df | 1313 | static void __flush_deferred_io(struct mapped_device *md) |
1da177e4 | 1314 | { |
6d6f10df | 1315 | struct bio *c; |
1da177e4 | 1316 | |
6d6f10df | 1317 | while ((c = bio_list_pop(&md->deferred))) { |
9e4e5f87 MB |
1318 | if (__split_bio(md, c)) |
1319 | bio_io_error(c); | |
1da177e4 | 1320 | } |
73d410c0 MB |
1321 | |
1322 | clear_bit(DMF_BLOCK_IO, &md->flags); | |
1da177e4 LT |
1323 | } |
1324 | ||
6d6f10df MB |
1325 | static void __merge_pushback_list(struct mapped_device *md) |
1326 | { | |
1327 | unsigned long flags; | |
1328 | ||
1329 | spin_lock_irqsave(&md->pushback_lock, flags); | |
1330 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1331 | bio_list_merge_head(&md->deferred, &md->pushback); | |
1332 | bio_list_init(&md->pushback); | |
1333 | spin_unlock_irqrestore(&md->pushback_lock, flags); | |
1334 | } | |
1335 | ||
304f3f6a MB |
1336 | static void dm_wq_work(struct work_struct *work) |
1337 | { | |
1338 | struct dm_wq_req *req = container_of(work, struct dm_wq_req, work); | |
1339 | struct mapped_device *md = req->md; | |
1340 | ||
1341 | down_write(&md->io_lock); | |
1342 | switch (req->type) { | |
1343 | case DM_WQ_FLUSH_ALL: | |
1344 | __merge_pushback_list(md); | |
1345 | /* pass through */ | |
1346 | case DM_WQ_FLUSH_DEFERRED: | |
1347 | __flush_deferred_io(md); | |
1348 | break; | |
1349 | default: | |
1350 | DMERR("dm_wq_work: unrecognised work type %d", req->type); | |
1351 | BUG(); | |
1352 | } | |
1353 | up_write(&md->io_lock); | |
1354 | } | |
1355 | ||
1356 | static void dm_wq_queue(struct mapped_device *md, int type, void *context, | |
1357 | struct dm_wq_req *req) | |
1358 | { | |
1359 | req->type = type; | |
1360 | req->md = md; | |
1361 | req->context = context; | |
1362 | INIT_WORK(&req->work, dm_wq_work); | |
1363 | queue_work(md->wq, &req->work); | |
1364 | } | |
1365 | ||
1366 | static void dm_queue_flush(struct mapped_device *md, int type, void *context) | |
1367 | { | |
1368 | struct dm_wq_req req; | |
1369 | ||
1370 | dm_wq_queue(md, type, context, &req); | |
1371 | flush_workqueue(md->wq); | |
1372 | } | |
1373 | ||
1da177e4 LT |
1374 | /* |
1375 | * Swap in a new table (destroying old one). | |
1376 | */ | |
1377 | int dm_swap_table(struct mapped_device *md, struct dm_table *table) | |
1378 | { | |
93c534ae | 1379 | int r = -EINVAL; |
1da177e4 | 1380 | |
e61290a4 | 1381 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
1382 | |
1383 | /* device must be suspended */ | |
cf222b37 | 1384 | if (!dm_suspended(md)) |
93c534ae | 1385 | goto out; |
1da177e4 | 1386 | |
bfa152fa JN |
1387 | /* without bdev, the device size cannot be changed */ |
1388 | if (!md->suspended_bdev) | |
1389 | if (get_capacity(md->disk) != dm_table_get_size(table)) | |
1390 | goto out; | |
1391 | ||
1da177e4 LT |
1392 | __unbind(md); |
1393 | r = __bind(md, table); | |
1da177e4 | 1394 | |
93c534ae | 1395 | out: |
e61290a4 | 1396 | mutex_unlock(&md->suspend_lock); |
93c534ae | 1397 | return r; |
1da177e4 LT |
1398 | } |
1399 | ||
1400 | /* | |
1401 | * Functions to lock and unlock any filesystem running on the | |
1402 | * device. | |
1403 | */ | |
2ca3310e | 1404 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 1405 | { |
e39e2e95 | 1406 | int r; |
1da177e4 LT |
1407 | |
1408 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 1409 | |
e39e2e95 | 1410 | md->frozen_sb = freeze_bdev(md->suspended_bdev); |
dfbe03f6 | 1411 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 1412 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
1413 | md->frozen_sb = NULL; |
1414 | return r; | |
dfbe03f6 AK |
1415 | } |
1416 | ||
aa8d7c2f AK |
1417 | set_bit(DMF_FROZEN, &md->flags); |
1418 | ||
1da177e4 | 1419 | /* don't bdput right now, we don't want the bdev |
e39e2e95 | 1420 | * to go away while it is locked. |
1da177e4 LT |
1421 | */ |
1422 | return 0; | |
1423 | } | |
1424 | ||
2ca3310e | 1425 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 1426 | { |
aa8d7c2f AK |
1427 | if (!test_bit(DMF_FROZEN, &md->flags)) |
1428 | return; | |
1429 | ||
e39e2e95 | 1430 | thaw_bdev(md->suspended_bdev, md->frozen_sb); |
1da177e4 | 1431 | md->frozen_sb = NULL; |
aa8d7c2f | 1432 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
1433 | } |
1434 | ||
1435 | /* | |
1436 | * We need to be able to change a mapping table under a mounted | |
1437 | * filesystem. For example we might want to move some data in | |
1438 | * the background. Before the table can be swapped with | |
1439 | * dm_bind_table, dm_suspend must be called to flush any in | |
1440 | * flight bios and ensure that any further io gets deferred. | |
1441 | */ | |
a3d77d35 | 1442 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 1443 | { |
2ca3310e | 1444 | struct dm_table *map = NULL; |
1da177e4 | 1445 | DECLARE_WAITQUEUE(wait, current); |
46125c1c | 1446 | int r = 0; |
a3d77d35 | 1447 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 1448 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 1449 | |
e61290a4 | 1450 | mutex_lock(&md->suspend_lock); |
2ca3310e | 1451 | |
73d410c0 MB |
1452 | if (dm_suspended(md)) { |
1453 | r = -EINVAL; | |
d287483d | 1454 | goto out_unlock; |
73d410c0 | 1455 | } |
1da177e4 LT |
1456 | |
1457 | map = dm_get_table(md); | |
1da177e4 | 1458 | |
2e93ccc1 KU |
1459 | /* |
1460 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
1461 | * This flag is cleared before dm_suspend returns. | |
1462 | */ | |
1463 | if (noflush) | |
1464 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1465 | ||
cf222b37 AK |
1466 | /* This does not get reverted if there's an error later. */ |
1467 | dm_table_presuspend_targets(map); | |
1468 | ||
bfa152fa JN |
1469 | /* bdget() can stall if the pending I/Os are not flushed */ |
1470 | if (!noflush) { | |
1471 | md->suspended_bdev = bdget_disk(md->disk, 0); | |
1472 | if (!md->suspended_bdev) { | |
1473 | DMWARN("bdget failed in dm_suspend"); | |
1474 | r = -ENOMEM; | |
1475 | goto flush_and_out; | |
1476 | } | |
e39e2e95 | 1477 | |
6d6f10df MB |
1478 | /* |
1479 | * Flush I/O to the device. noflush supersedes do_lockfs, | |
1480 | * because lock_fs() needs to flush I/Os. | |
1481 | */ | |
1482 | if (do_lockfs) { | |
1483 | r = lock_fs(md); | |
1484 | if (r) | |
1485 | goto out; | |
1486 | } | |
aa8d7c2f | 1487 | } |
1da177e4 LT |
1488 | |
1489 | /* | |
354e0071 | 1490 | * First we set the BLOCK_IO flag so no more ios will be mapped. |
1da177e4 | 1491 | */ |
2ca3310e AK |
1492 | down_write(&md->io_lock); |
1493 | set_bit(DMF_BLOCK_IO, &md->flags); | |
1da177e4 | 1494 | |
1da177e4 | 1495 | add_wait_queue(&md->wait, &wait); |
2ca3310e | 1496 | up_write(&md->io_lock); |
1da177e4 LT |
1497 | |
1498 | /* unplug */ | |
2ca3310e | 1499 | if (map) |
1da177e4 | 1500 | dm_table_unplug_all(map); |
1da177e4 LT |
1501 | |
1502 | /* | |
46125c1c | 1503 | * Wait for the already-mapped ios to complete. |
1da177e4 | 1504 | */ |
46125c1c | 1505 | r = dm_wait_for_completion(md); |
1da177e4 | 1506 | |
2ca3310e | 1507 | down_write(&md->io_lock); |
1da177e4 LT |
1508 | remove_wait_queue(&md->wait, &wait); |
1509 | ||
6d6f10df MB |
1510 | if (noflush) |
1511 | __merge_pushback_list(md); | |
94d6351e | 1512 | up_write(&md->io_lock); |
2e93ccc1 | 1513 | |
1da177e4 | 1514 | /* were we interrupted ? */ |
46125c1c | 1515 | if (r < 0) { |
304f3f6a | 1516 | dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL); |
73d410c0 | 1517 | |
2ca3310e | 1518 | unlock_fs(md); |
2e93ccc1 | 1519 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 1520 | } |
1da177e4 | 1521 | |
cf222b37 | 1522 | dm_table_postsuspend_targets(map); |
1da177e4 | 1523 | |
2ca3310e | 1524 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 1525 | |
2e93ccc1 | 1526 | flush_and_out: |
304f3f6a | 1527 | if (r && noflush) |
2e93ccc1 KU |
1528 | /* |
1529 | * Because there may be already I/Os in the pushback list, | |
1530 | * flush them before return. | |
1531 | */ | |
304f3f6a | 1532 | dm_queue_flush(md, DM_WQ_FLUSH_ALL, NULL); |
2e93ccc1 | 1533 | |
2ca3310e | 1534 | out: |
e39e2e95 AK |
1535 | if (r && md->suspended_bdev) { |
1536 | bdput(md->suspended_bdev); | |
1537 | md->suspended_bdev = NULL; | |
1538 | } | |
1539 | ||
2ca3310e | 1540 | dm_table_put(map); |
d287483d AK |
1541 | |
1542 | out_unlock: | |
e61290a4 | 1543 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 1544 | return r; |
1da177e4 LT |
1545 | } |
1546 | ||
1547 | int dm_resume(struct mapped_device *md) | |
1548 | { | |
cf222b37 | 1549 | int r = -EINVAL; |
cf222b37 | 1550 | struct dm_table *map = NULL; |
1da177e4 | 1551 | |
e61290a4 | 1552 | mutex_lock(&md->suspend_lock); |
2ca3310e | 1553 | if (!dm_suspended(md)) |
cf222b37 | 1554 | goto out; |
cf222b37 AK |
1555 | |
1556 | map = dm_get_table(md); | |
2ca3310e | 1557 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 1558 | goto out; |
1da177e4 | 1559 | |
8757b776 MB |
1560 | r = dm_table_resume_targets(map); |
1561 | if (r) | |
1562 | goto out; | |
2ca3310e | 1563 | |
304f3f6a | 1564 | dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL); |
2ca3310e AK |
1565 | |
1566 | unlock_fs(md); | |
1567 | ||
bfa152fa JN |
1568 | if (md->suspended_bdev) { |
1569 | bdput(md->suspended_bdev); | |
1570 | md->suspended_bdev = NULL; | |
1571 | } | |
e39e2e95 | 1572 | |
2ca3310e AK |
1573 | clear_bit(DMF_SUSPENDED, &md->flags); |
1574 | ||
1da177e4 | 1575 | dm_table_unplug_all(map); |
1da177e4 | 1576 | |
69267a30 | 1577 | dm_kobject_uevent(md); |
8560ed6f | 1578 | |
cf222b37 | 1579 | r = 0; |
2ca3310e | 1580 | |
cf222b37 AK |
1581 | out: |
1582 | dm_table_put(map); | |
e61290a4 | 1583 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 1584 | |
cf222b37 | 1585 | return r; |
1da177e4 LT |
1586 | } |
1587 | ||
1588 | /*----------------------------------------------------------------- | |
1589 | * Event notification. | |
1590 | *---------------------------------------------------------------*/ | |
69267a30 AK |
1591 | void dm_kobject_uevent(struct mapped_device *md) |
1592 | { | |
edfaa7c3 | 1593 | kobject_uevent(&md->disk->dev.kobj, KOBJ_CHANGE); |
69267a30 AK |
1594 | } |
1595 | ||
7a8c3d3b MA |
1596 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
1597 | { | |
1598 | return atomic_add_return(1, &md->uevent_seq); | |
1599 | } | |
1600 | ||
1da177e4 LT |
1601 | uint32_t dm_get_event_nr(struct mapped_device *md) |
1602 | { | |
1603 | return atomic_read(&md->event_nr); | |
1604 | } | |
1605 | ||
1606 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
1607 | { | |
1608 | return wait_event_interruptible(md->eventq, | |
1609 | (event_nr != atomic_read(&md->event_nr))); | |
1610 | } | |
1611 | ||
7a8c3d3b MA |
1612 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
1613 | { | |
1614 | unsigned long flags; | |
1615 | ||
1616 | spin_lock_irqsave(&md->uevent_lock, flags); | |
1617 | list_add(elist, &md->uevent_list); | |
1618 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1619 | } | |
1620 | ||
1da177e4 LT |
1621 | /* |
1622 | * The gendisk is only valid as long as you have a reference | |
1623 | * count on 'md'. | |
1624 | */ | |
1625 | struct gendisk *dm_disk(struct mapped_device *md) | |
1626 | { | |
1627 | return md->disk; | |
1628 | } | |
1629 | ||
1630 | int dm_suspended(struct mapped_device *md) | |
1631 | { | |
1632 | return test_bit(DMF_SUSPENDED, &md->flags); | |
1633 | } | |
1634 | ||
2e93ccc1 KU |
1635 | int dm_noflush_suspending(struct dm_target *ti) |
1636 | { | |
1637 | struct mapped_device *md = dm_table_get_md(ti->table); | |
1638 | int r = __noflush_suspending(md); | |
1639 | ||
1640 | dm_put(md); | |
1641 | ||
1642 | return r; | |
1643 | } | |
1644 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
1645 | ||
1da177e4 LT |
1646 | static struct block_device_operations dm_blk_dops = { |
1647 | .open = dm_blk_open, | |
1648 | .release = dm_blk_close, | |
aa129a22 | 1649 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 1650 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
1651 | .owner = THIS_MODULE |
1652 | }; | |
1653 | ||
1654 | EXPORT_SYMBOL(dm_get_mapinfo); | |
1655 | ||
1656 | /* | |
1657 | * module hooks | |
1658 | */ | |
1659 | module_init(dm_init); | |
1660 | module_exit(dm_exit); | |
1661 | ||
1662 | module_param(major, uint, 0); | |
1663 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
1664 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
1665 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
1666 | MODULE_LICENSE("GPL"); |