]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. | |
4 | * | |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
9 | #include "dm-bio-list.h" | |
10 | #include "dm-uevent.h" | |
11 | ||
12 | #include <linux/init.h> | |
13 | #include <linux/module.h> | |
14 | #include <linux/mutex.h> | |
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> | |
22 | #include <linux/hdreg.h> | |
23 | #include <linux/blktrace_api.h> | |
24 | #include <linux/smp_lock.h> | |
25 | ||
26 | #define DM_MSG_PREFIX "core" | |
27 | ||
28 | static const char *_name = DM_NAME; | |
29 | ||
30 | static unsigned int major = 0; | |
31 | static unsigned int _major = 0; | |
32 | ||
33 | static DEFINE_SPINLOCK(_minor_lock); | |
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; | |
42 | unsigned long start_time; | |
43 | }; | |
44 | ||
45 | /* | |
46 | * One of these is allocated per target within a bio. Hopefully | |
47 | * this will be simplified out one day. | |
48 | */ | |
49 | struct dm_target_io { | |
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 | { | |
57 | if (bio && bio->bi_private) | |
58 | return &((struct dm_target_io *)bio->bi_private)->info; | |
59 | return NULL; | |
60 | } | |
61 | ||
62 | #define MINOR_ALLOCED ((void *)-1) | |
63 | ||
64 | /* | |
65 | * Bits for the md->flags field. | |
66 | */ | |
67 | #define DMF_BLOCK_IO 0 | |
68 | #define DMF_SUSPENDED 1 | |
69 | #define DMF_FROZEN 2 | |
70 | #define DMF_FREEING 3 | |
71 | #define DMF_DELETING 4 | |
72 | #define DMF_NOFLUSH_SUSPENDING 5 | |
73 | ||
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 | ||
87 | struct mapped_device { | |
88 | struct rw_semaphore io_lock; | |
89 | struct mutex suspend_lock; | |
90 | spinlock_t pushback_lock; | |
91 | rwlock_t map_lock; | |
92 | atomic_t holders; | |
93 | atomic_t open_count; | |
94 | ||
95 | unsigned long flags; | |
96 | ||
97 | struct request_queue *queue; | |
98 | struct gendisk *disk; | |
99 | char name[16]; | |
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; | |
108 | struct bio_list deferred; | |
109 | struct bio_list pushback; | |
110 | ||
111 | /* | |
112 | * Processing queue (flush/barriers) | |
113 | */ | |
114 | struct workqueue_struct *wq; | |
115 | ||
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 | ||
127 | struct bio_set *bs; | |
128 | ||
129 | /* | |
130 | * Event handling. | |
131 | */ | |
132 | atomic_t event_nr; | |
133 | wait_queue_head_t eventq; | |
134 | atomic_t uevent_seq; | |
135 | struct list_head uevent_list; | |
136 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
137 | ||
138 | /* | |
139 | * freeze/thaw support require holding onto a super block | |
140 | */ | |
141 | struct super_block *frozen_sb; | |
142 | struct block_device *suspended_bdev; | |
143 | ||
144 | /* forced geometry settings */ | |
145 | struct hd_geometry geometry; | |
146 | }; | |
147 | ||
148 | #define MIN_IOS 256 | |
149 | static struct kmem_cache *_io_cache; | |
150 | static struct kmem_cache *_tio_cache; | |
151 | ||
152 | static int __init local_init(void) | |
153 | { | |
154 | int r; | |
155 | ||
156 | /* allocate a slab for the dm_ios */ | |
157 | _io_cache = KMEM_CACHE(dm_io, 0); | |
158 | if (!_io_cache) | |
159 | return -ENOMEM; | |
160 | ||
161 | /* allocate a slab for the target ios */ | |
162 | _tio_cache = KMEM_CACHE(dm_target_io, 0); | |
163 | if (!_tio_cache) { | |
164 | kmem_cache_destroy(_io_cache); | |
165 | return -ENOMEM; | |
166 | } | |
167 | ||
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 | ||
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); | |
180 | dm_uevent_exit(); | |
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); | |
194 | unregister_blkdev(_major, _name); | |
195 | dm_uevent_exit(); | |
196 | ||
197 | _major = 0; | |
198 | ||
199 | DMINFO("cleaned up"); | |
200 | } | |
201 | ||
202 | static int (*_inits[])(void) __initdata = { | |
203 | local_init, | |
204 | dm_target_init, | |
205 | dm_linear_init, | |
206 | dm_stripe_init, | |
207 | dm_kcopyd_init, | |
208 | dm_interface_init, | |
209 | }; | |
210 | ||
211 | static void (*_exits[])(void) = { | |
212 | local_exit, | |
213 | dm_target_exit, | |
214 | dm_linear_exit, | |
215 | dm_stripe_exit, | |
216 | dm_kcopyd_exit, | |
217 | dm_interface_exit, | |
218 | }; | |
219 | ||
220 | static int __init dm_init(void) | |
221 | { | |
222 | const int count = ARRAY_SIZE(_inits); | |
223 | ||
224 | int r, i; | |
225 | ||
226 | for (i = 0; i < count; i++) { | |
227 | r = _inits[i](); | |
228 | if (r) | |
229 | goto bad; | |
230 | } | |
231 | ||
232 | return 0; | |
233 | ||
234 | bad: | |
235 | while (i--) | |
236 | _exits[i](); | |
237 | ||
238 | return r; | |
239 | } | |
240 | ||
241 | static void __exit dm_exit(void) | |
242 | { | |
243 | int i = ARRAY_SIZE(_exits); | |
244 | ||
245 | while (i--) | |
246 | _exits[i](); | |
247 | } | |
248 | ||
249 | /* | |
250 | * Block device functions | |
251 | */ | |
252 | static int dm_blk_open(struct inode *inode, struct file *file) | |
253 | { | |
254 | struct mapped_device *md; | |
255 | ||
256 | spin_lock(&_minor_lock); | |
257 | ||
258 | md = inode->i_bdev->bd_disk->private_data; | |
259 | if (!md) | |
260 | goto out; | |
261 | ||
262 | if (test_bit(DMF_FREEING, &md->flags) || | |
263 | test_bit(DMF_DELETING, &md->flags)) { | |
264 | md = NULL; | |
265 | goto out; | |
266 | } | |
267 | ||
268 | dm_get(md); | |
269 | atomic_inc(&md->open_count); | |
270 | ||
271 | out: | |
272 | spin_unlock(&_minor_lock); | |
273 | ||
274 | return md ? 0 : -ENXIO; | |
275 | } | |
276 | ||
277 | static int dm_blk_close(struct inode *inode, struct file *file) | |
278 | { | |
279 | struct mapped_device *md; | |
280 | ||
281 | md = inode->i_bdev->bd_disk->private_data; | |
282 | atomic_dec(&md->open_count); | |
283 | dm_put(md); | |
284 | return 0; | |
285 | } | |
286 | ||
287 | int dm_open_count(struct mapped_device *md) | |
288 | { | |
289 | return atomic_read(&md->open_count); | |
290 | } | |
291 | ||
292 | /* | |
293 | * Guarantees nothing is using the device before it's deleted. | |
294 | */ | |
295 | int dm_lock_for_deletion(struct mapped_device *md) | |
296 | { | |
297 | int r = 0; | |
298 | ||
299 | spin_lock(&_minor_lock); | |
300 | ||
301 | if (dm_open_count(md)) | |
302 | r = -EBUSY; | |
303 | else | |
304 | set_bit(DMF_DELETING, &md->flags); | |
305 | ||
306 | spin_unlock(&_minor_lock); | |
307 | ||
308 | return r; | |
309 | } | |
310 | ||
311 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) | |
312 | { | |
313 | struct mapped_device *md = bdev->bd_disk->private_data; | |
314 | ||
315 | return dm_get_geometry(md, geo); | |
316 | } | |
317 | ||
318 | static int dm_blk_ioctl(struct inode *inode, struct file *file, | |
319 | unsigned int cmd, unsigned long arg) | |
320 | { | |
321 | struct mapped_device *md; | |
322 | struct dm_table *map; | |
323 | struct dm_target *tgt; | |
324 | int r = -ENOTTY; | |
325 | ||
326 | /* We don't really need this lock, but we do need 'inode'. */ | |
327 | unlock_kernel(); | |
328 | ||
329 | md = inode->i_bdev->bd_disk->private_data; | |
330 | ||
331 | map = dm_get_table(md); | |
332 | ||
333 | if (!map || !dm_table_get_size(map)) | |
334 | goto out; | |
335 | ||
336 | /* We only support devices that have a single target */ | |
337 | if (dm_table_get_num_targets(map) != 1) | |
338 | goto out; | |
339 | ||
340 | tgt = dm_table_get_target(map, 0); | |
341 | ||
342 | if (dm_suspended(md)) { | |
343 | r = -EAGAIN; | |
344 | goto out; | |
345 | } | |
346 | ||
347 | if (tgt->type->ioctl) | |
348 | r = tgt->type->ioctl(tgt, inode, file, cmd, arg); | |
349 | ||
350 | out: | |
351 | dm_table_put(map); | |
352 | ||
353 | lock_kernel(); | |
354 | return r; | |
355 | } | |
356 | ||
357 | static struct dm_io *alloc_io(struct mapped_device *md) | |
358 | { | |
359 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
360 | } | |
361 | ||
362 | static void free_io(struct mapped_device *md, struct dm_io *io) | |
363 | { | |
364 | mempool_free(io, md->io_pool); | |
365 | } | |
366 | ||
367 | static struct dm_target_io *alloc_tio(struct mapped_device *md) | |
368 | { | |
369 | return mempool_alloc(md->tio_pool, GFP_NOIO); | |
370 | } | |
371 | ||
372 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) | |
373 | { | |
374 | mempool_free(tio, md->tio_pool); | |
375 | } | |
376 | ||
377 | static void start_io_acct(struct dm_io *io) | |
378 | { | |
379 | struct mapped_device *md = io->md; | |
380 | ||
381 | io->start_time = jiffies; | |
382 | ||
383 | preempt_disable(); | |
384 | disk_round_stats(dm_disk(md)); | |
385 | preempt_enable(); | |
386 | dm_disk(md)->in_flight = atomic_inc_return(&md->pending); | |
387 | } | |
388 | ||
389 | static int end_io_acct(struct dm_io *io) | |
390 | { | |
391 | struct mapped_device *md = io->md; | |
392 | struct bio *bio = io->bio; | |
393 | unsigned long duration = jiffies - io->start_time; | |
394 | int pending; | |
395 | int rw = bio_data_dir(bio); | |
396 | ||
397 | preempt_disable(); | |
398 | disk_round_stats(dm_disk(md)); | |
399 | preempt_enable(); | |
400 | dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending); | |
401 | ||
402 | disk_stat_add(dm_disk(md), ticks[rw], duration); | |
403 | ||
404 | return !pending; | |
405 | } | |
406 | ||
407 | /* | |
408 | * Add the bio to the list of deferred io. | |
409 | */ | |
410 | static int queue_io(struct mapped_device *md, struct bio *bio) | |
411 | { | |
412 | down_write(&md->io_lock); | |
413 | ||
414 | if (!test_bit(DMF_BLOCK_IO, &md->flags)) { | |
415 | up_write(&md->io_lock); | |
416 | return 1; | |
417 | } | |
418 | ||
419 | bio_list_add(&md->deferred, bio); | |
420 | ||
421 | up_write(&md->io_lock); | |
422 | return 0; /* deferred successfully */ | |
423 | } | |
424 | ||
425 | /* | |
426 | * Everyone (including functions in this file), should use this | |
427 | * function to access the md->map field, and make sure they call | |
428 | * dm_table_put() when finished. | |
429 | */ | |
430 | struct dm_table *dm_get_table(struct mapped_device *md) | |
431 | { | |
432 | struct dm_table *t; | |
433 | ||
434 | read_lock(&md->map_lock); | |
435 | t = md->map; | |
436 | if (t) | |
437 | dm_table_get(t); | |
438 | read_unlock(&md->map_lock); | |
439 | ||
440 | return t; | |
441 | } | |
442 | ||
443 | /* | |
444 | * Get the geometry associated with a dm device | |
445 | */ | |
446 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
447 | { | |
448 | *geo = md->geometry; | |
449 | ||
450 | return 0; | |
451 | } | |
452 | ||
453 | /* | |
454 | * Set the geometry of a device. | |
455 | */ | |
456 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
457 | { | |
458 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
459 | ||
460 | if (geo->start > sz) { | |
461 | DMWARN("Start sector is beyond the geometry limits."); | |
462 | return -EINVAL; | |
463 | } | |
464 | ||
465 | md->geometry = *geo; | |
466 | ||
467 | return 0; | |
468 | } | |
469 | ||
470 | /*----------------------------------------------------------------- | |
471 | * CRUD START: | |
472 | * A more elegant soln is in the works that uses the queue | |
473 | * merge fn, unfortunately there are a couple of changes to | |
474 | * the block layer that I want to make for this. So in the | |
475 | * interests of getting something for people to use I give | |
476 | * you this clearly demarcated crap. | |
477 | *---------------------------------------------------------------*/ | |
478 | ||
479 | static int __noflush_suspending(struct mapped_device *md) | |
480 | { | |
481 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
482 | } | |
483 | ||
484 | /* | |
485 | * Decrements the number of outstanding ios that a bio has been | |
486 | * cloned into, completing the original io if necc. | |
487 | */ | |
488 | static void dec_pending(struct dm_io *io, int error) | |
489 | { | |
490 | unsigned long flags; | |
491 | ||
492 | /* Push-back supersedes any I/O errors */ | |
493 | if (error && !(io->error > 0 && __noflush_suspending(io->md))) | |
494 | io->error = error; | |
495 | ||
496 | if (atomic_dec_and_test(&io->io_count)) { | |
497 | if (io->error == DM_ENDIO_REQUEUE) { | |
498 | /* | |
499 | * Target requested pushing back the I/O. | |
500 | * This must be handled before the sleeper on | |
501 | * suspend queue merges the pushback list. | |
502 | */ | |
503 | spin_lock_irqsave(&io->md->pushback_lock, flags); | |
504 | if (__noflush_suspending(io->md)) | |
505 | bio_list_add(&io->md->pushback, io->bio); | |
506 | else | |
507 | /* noflush suspend was interrupted. */ | |
508 | io->error = -EIO; | |
509 | spin_unlock_irqrestore(&io->md->pushback_lock, flags); | |
510 | } | |
511 | ||
512 | if (end_io_acct(io)) | |
513 | /* nudge anyone waiting on suspend queue */ | |
514 | wake_up(&io->md->wait); | |
515 | ||
516 | if (io->error != DM_ENDIO_REQUEUE) { | |
517 | blk_add_trace_bio(io->md->queue, io->bio, | |
518 | BLK_TA_COMPLETE); | |
519 | ||
520 | bio_endio(io->bio, io->error); | |
521 | } | |
522 | ||
523 | free_io(io->md, io); | |
524 | } | |
525 | } | |
526 | ||
527 | static void clone_endio(struct bio *bio, int error) | |
528 | { | |
529 | int r = 0; | |
530 | struct dm_target_io *tio = bio->bi_private; | |
531 | struct mapped_device *md = tio->io->md; | |
532 | dm_endio_fn endio = tio->ti->type->end_io; | |
533 | ||
534 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) | |
535 | error = -EIO; | |
536 | ||
537 | if (endio) { | |
538 | r = endio(tio->ti, bio, error, &tio->info); | |
539 | if (r < 0 || r == DM_ENDIO_REQUEUE) | |
540 | /* | |
541 | * error and requeue request are handled | |
542 | * in dec_pending(). | |
543 | */ | |
544 | error = r; | |
545 | else if (r == DM_ENDIO_INCOMPLETE) | |
546 | /* The target will handle the io */ | |
547 | return; | |
548 | else if (r) { | |
549 | DMWARN("unimplemented target endio return value: %d", r); | |
550 | BUG(); | |
551 | } | |
552 | } | |
553 | ||
554 | dec_pending(tio->io, error); | |
555 | ||
556 | /* | |
557 | * Store md for cleanup instead of tio which is about to get freed. | |
558 | */ | |
559 | bio->bi_private = md->bs; | |
560 | ||
561 | bio_put(bio); | |
562 | free_tio(md, tio); | |
563 | } | |
564 | ||
565 | static sector_t max_io_len(struct mapped_device *md, | |
566 | sector_t sector, struct dm_target *ti) | |
567 | { | |
568 | sector_t offset = sector - ti->begin; | |
569 | sector_t len = ti->len - offset; | |
570 | ||
571 | /* | |
572 | * Does the target need to split even further ? | |
573 | */ | |
574 | if (ti->split_io) { | |
575 | sector_t boundary; | |
576 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) | |
577 | - offset; | |
578 | if (len > boundary) | |
579 | len = boundary; | |
580 | } | |
581 | ||
582 | return len; | |
583 | } | |
584 | ||
585 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
586 | struct dm_target_io *tio) | |
587 | { | |
588 | int r; | |
589 | sector_t sector; | |
590 | struct mapped_device *md; | |
591 | ||
592 | /* | |
593 | * Sanity checks. | |
594 | */ | |
595 | BUG_ON(!clone->bi_size); | |
596 | ||
597 | clone->bi_end_io = clone_endio; | |
598 | clone->bi_private = tio; | |
599 | ||
600 | /* | |
601 | * Map the clone. If r == 0 we don't need to do | |
602 | * anything, the target has assumed ownership of | |
603 | * this io. | |
604 | */ | |
605 | atomic_inc(&tio->io->io_count); | |
606 | sector = clone->bi_sector; | |
607 | r = ti->type->map(ti, clone, &tio->info); | |
608 | if (r == DM_MAPIO_REMAPPED) { | |
609 | /* the bio has been remapped so dispatch it */ | |
610 | ||
611 | blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone, | |
612 | tio->io->bio->bi_bdev->bd_dev, | |
613 | clone->bi_sector, sector); | |
614 | ||
615 | generic_make_request(clone); | |
616 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { | |
617 | /* error the io and bail out, or requeue it if needed */ | |
618 | md = tio->io->md; | |
619 | dec_pending(tio->io, r); | |
620 | /* | |
621 | * Store bio_set for cleanup. | |
622 | */ | |
623 | clone->bi_private = md->bs; | |
624 | bio_put(clone); | |
625 | free_tio(md, tio); | |
626 | } else if (r) { | |
627 | DMWARN("unimplemented target map return value: %d", r); | |
628 | BUG(); | |
629 | } | |
630 | } | |
631 | ||
632 | struct clone_info { | |
633 | struct mapped_device *md; | |
634 | struct dm_table *map; | |
635 | struct bio *bio; | |
636 | struct dm_io *io; | |
637 | sector_t sector; | |
638 | sector_t sector_count; | |
639 | unsigned short idx; | |
640 | }; | |
641 | ||
642 | static void dm_bio_destructor(struct bio *bio) | |
643 | { | |
644 | struct bio_set *bs = bio->bi_private; | |
645 | ||
646 | bio_free(bio, bs); | |
647 | } | |
648 | ||
649 | /* | |
650 | * Creates a little bio that is just does part of a bvec. | |
651 | */ | |
652 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
653 | unsigned short idx, unsigned int offset, | |
654 | unsigned int len, struct bio_set *bs) | |
655 | { | |
656 | struct bio *clone; | |
657 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
658 | ||
659 | clone = bio_alloc_bioset(GFP_NOIO, 1, bs); | |
660 | clone->bi_destructor = dm_bio_destructor; | |
661 | *clone->bi_io_vec = *bv; | |
662 | ||
663 | clone->bi_sector = sector; | |
664 | clone->bi_bdev = bio->bi_bdev; | |
665 | clone->bi_rw = bio->bi_rw; | |
666 | clone->bi_vcnt = 1; | |
667 | clone->bi_size = to_bytes(len); | |
668 | clone->bi_io_vec->bv_offset = offset; | |
669 | clone->bi_io_vec->bv_len = clone->bi_size; | |
670 | ||
671 | return clone; | |
672 | } | |
673 | ||
674 | /* | |
675 | * Creates a bio that consists of range of complete bvecs. | |
676 | */ | |
677 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
678 | unsigned short idx, unsigned short bv_count, | |
679 | unsigned int len, struct bio_set *bs) | |
680 | { | |
681 | struct bio *clone; | |
682 | ||
683 | clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); | |
684 | __bio_clone(clone, bio); | |
685 | clone->bi_destructor = dm_bio_destructor; | |
686 | clone->bi_sector = sector; | |
687 | clone->bi_idx = idx; | |
688 | clone->bi_vcnt = idx + bv_count; | |
689 | clone->bi_size = to_bytes(len); | |
690 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
691 | ||
692 | return clone; | |
693 | } | |
694 | ||
695 | static int __clone_and_map(struct clone_info *ci) | |
696 | { | |
697 | struct bio *clone, *bio = ci->bio; | |
698 | struct dm_target *ti; | |
699 | sector_t len = 0, max; | |
700 | struct dm_target_io *tio; | |
701 | ||
702 | ti = dm_table_find_target(ci->map, ci->sector); | |
703 | if (!dm_target_is_valid(ti)) | |
704 | return -EIO; | |
705 | ||
706 | max = max_io_len(ci->md, ci->sector, ti); | |
707 | ||
708 | /* | |
709 | * Allocate a target io object. | |
710 | */ | |
711 | tio = alloc_tio(ci->md); | |
712 | tio->io = ci->io; | |
713 | tio->ti = ti; | |
714 | memset(&tio->info, 0, sizeof(tio->info)); | |
715 | ||
716 | if (ci->sector_count <= max) { | |
717 | /* | |
718 | * Optimise for the simple case where we can do all of | |
719 | * the remaining io with a single clone. | |
720 | */ | |
721 | clone = clone_bio(bio, ci->sector, ci->idx, | |
722 | bio->bi_vcnt - ci->idx, ci->sector_count, | |
723 | ci->md->bs); | |
724 | __map_bio(ti, clone, tio); | |
725 | ci->sector_count = 0; | |
726 | ||
727 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
728 | /* | |
729 | * There are some bvecs that don't span targets. | |
730 | * Do as many of these as possible. | |
731 | */ | |
732 | int i; | |
733 | sector_t remaining = max; | |
734 | sector_t bv_len; | |
735 | ||
736 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
737 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
738 | ||
739 | if (bv_len > remaining) | |
740 | break; | |
741 | ||
742 | remaining -= bv_len; | |
743 | len += bv_len; | |
744 | } | |
745 | ||
746 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, | |
747 | ci->md->bs); | |
748 | __map_bio(ti, clone, tio); | |
749 | ||
750 | ci->sector += len; | |
751 | ci->sector_count -= len; | |
752 | ci->idx = i; | |
753 | ||
754 | } else { | |
755 | /* | |
756 | * Handle a bvec that must be split between two or more targets. | |
757 | */ | |
758 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
759 | sector_t remaining = to_sector(bv->bv_len); | |
760 | unsigned int offset = 0; | |
761 | ||
762 | do { | |
763 | if (offset) { | |
764 | ti = dm_table_find_target(ci->map, ci->sector); | |
765 | if (!dm_target_is_valid(ti)) | |
766 | return -EIO; | |
767 | ||
768 | max = max_io_len(ci->md, ci->sector, ti); | |
769 | ||
770 | tio = alloc_tio(ci->md); | |
771 | tio->io = ci->io; | |
772 | tio->ti = ti; | |
773 | memset(&tio->info, 0, sizeof(tio->info)); | |
774 | } | |
775 | ||
776 | len = min(remaining, max); | |
777 | ||
778 | clone = split_bvec(bio, ci->sector, ci->idx, | |
779 | bv->bv_offset + offset, len, | |
780 | ci->md->bs); | |
781 | ||
782 | __map_bio(ti, clone, tio); | |
783 | ||
784 | ci->sector += len; | |
785 | ci->sector_count -= len; | |
786 | offset += to_bytes(len); | |
787 | } while (remaining -= len); | |
788 | ||
789 | ci->idx++; | |
790 | } | |
791 | ||
792 | return 0; | |
793 | } | |
794 | ||
795 | /* | |
796 | * Split the bio into several clones. | |
797 | */ | |
798 | static int __split_bio(struct mapped_device *md, struct bio *bio) | |
799 | { | |
800 | struct clone_info ci; | |
801 | int error = 0; | |
802 | ||
803 | ci.map = dm_get_table(md); | |
804 | if (unlikely(!ci.map)) | |
805 | return -EIO; | |
806 | ||
807 | ci.md = md; | |
808 | ci.bio = bio; | |
809 | ci.io = alloc_io(md); | |
810 | ci.io->error = 0; | |
811 | atomic_set(&ci.io->io_count, 1); | |
812 | ci.io->bio = bio; | |
813 | ci.io->md = md; | |
814 | ci.sector = bio->bi_sector; | |
815 | ci.sector_count = bio_sectors(bio); | |
816 | ci.idx = bio->bi_idx; | |
817 | ||
818 | start_io_acct(ci.io); | |
819 | while (ci.sector_count && !error) | |
820 | error = __clone_and_map(&ci); | |
821 | ||
822 | /* drop the extra reference count */ | |
823 | dec_pending(ci.io, error); | |
824 | dm_table_put(ci.map); | |
825 | ||
826 | return 0; | |
827 | } | |
828 | /*----------------------------------------------------------------- | |
829 | * CRUD END | |
830 | *---------------------------------------------------------------*/ | |
831 | ||
832 | /* | |
833 | * The request function that just remaps the bio built up by | |
834 | * dm_merge_bvec. | |
835 | */ | |
836 | static int dm_request(struct request_queue *q, struct bio *bio) | |
837 | { | |
838 | int r = -EIO; | |
839 | int rw = bio_data_dir(bio); | |
840 | struct mapped_device *md = q->queuedata; | |
841 | ||
842 | /* | |
843 | * There is no use in forwarding any barrier request since we can't | |
844 | * guarantee it is (or can be) handled by the targets correctly. | |
845 | */ | |
846 | if (unlikely(bio_barrier(bio))) { | |
847 | bio_endio(bio, -EOPNOTSUPP); | |
848 | return 0; | |
849 | } | |
850 | ||
851 | down_read(&md->io_lock); | |
852 | ||
853 | disk_stat_inc(dm_disk(md), ios[rw]); | |
854 | disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio)); | |
855 | ||
856 | /* | |
857 | * If we're suspended we have to queue | |
858 | * this io for later. | |
859 | */ | |
860 | while (test_bit(DMF_BLOCK_IO, &md->flags)) { | |
861 | up_read(&md->io_lock); | |
862 | ||
863 | if (bio_rw(bio) != READA) | |
864 | r = queue_io(md, bio); | |
865 | ||
866 | if (r <= 0) | |
867 | goto out_req; | |
868 | ||
869 | /* | |
870 | * We're in a while loop, because someone could suspend | |
871 | * before we get to the following read lock. | |
872 | */ | |
873 | down_read(&md->io_lock); | |
874 | } | |
875 | ||
876 | r = __split_bio(md, bio); | |
877 | up_read(&md->io_lock); | |
878 | ||
879 | out_req: | |
880 | if (r < 0) | |
881 | bio_io_error(bio); | |
882 | ||
883 | return 0; | |
884 | } | |
885 | ||
886 | static void dm_unplug_all(struct request_queue *q) | |
887 | { | |
888 | struct mapped_device *md = q->queuedata; | |
889 | struct dm_table *map = dm_get_table(md); | |
890 | ||
891 | if (map) { | |
892 | dm_table_unplug_all(map); | |
893 | dm_table_put(map); | |
894 | } | |
895 | } | |
896 | ||
897 | static int dm_any_congested(void *congested_data, int bdi_bits) | |
898 | { | |
899 | int r; | |
900 | struct mapped_device *md = (struct mapped_device *) congested_data; | |
901 | struct dm_table *map = dm_get_table(md); | |
902 | ||
903 | if (!map || test_bit(DMF_BLOCK_IO, &md->flags)) | |
904 | r = bdi_bits; | |
905 | else | |
906 | r = dm_table_any_congested(map, bdi_bits); | |
907 | ||
908 | dm_table_put(map); | |
909 | return r; | |
910 | } | |
911 | ||
912 | /*----------------------------------------------------------------- | |
913 | * An IDR is used to keep track of allocated minor numbers. | |
914 | *---------------------------------------------------------------*/ | |
915 | static DEFINE_IDR(_minor_idr); | |
916 | ||
917 | static void free_minor(int minor) | |
918 | { | |
919 | spin_lock(&_minor_lock); | |
920 | idr_remove(&_minor_idr, minor); | |
921 | spin_unlock(&_minor_lock); | |
922 | } | |
923 | ||
924 | /* | |
925 | * See if the device with a specific minor # is free. | |
926 | */ | |
927 | static int specific_minor(struct mapped_device *md, int minor) | |
928 | { | |
929 | int r, m; | |
930 | ||
931 | if (minor >= (1 << MINORBITS)) | |
932 | return -EINVAL; | |
933 | ||
934 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); | |
935 | if (!r) | |
936 | return -ENOMEM; | |
937 | ||
938 | spin_lock(&_minor_lock); | |
939 | ||
940 | if (idr_find(&_minor_idr, minor)) { | |
941 | r = -EBUSY; | |
942 | goto out; | |
943 | } | |
944 | ||
945 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); | |
946 | if (r) | |
947 | goto out; | |
948 | ||
949 | if (m != minor) { | |
950 | idr_remove(&_minor_idr, m); | |
951 | r = -EBUSY; | |
952 | goto out; | |
953 | } | |
954 | ||
955 | out: | |
956 | spin_unlock(&_minor_lock); | |
957 | return r; | |
958 | } | |
959 | ||
960 | static int next_free_minor(struct mapped_device *md, int *minor) | |
961 | { | |
962 | int r, m; | |
963 | ||
964 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); | |
965 | if (!r) | |
966 | return -ENOMEM; | |
967 | ||
968 | spin_lock(&_minor_lock); | |
969 | ||
970 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); | |
971 | if (r) { | |
972 | goto out; | |
973 | } | |
974 | ||
975 | if (m >= (1 << MINORBITS)) { | |
976 | idr_remove(&_minor_idr, m); | |
977 | r = -ENOSPC; | |
978 | goto out; | |
979 | } | |
980 | ||
981 | *minor = m; | |
982 | ||
983 | out: | |
984 | spin_unlock(&_minor_lock); | |
985 | return r; | |
986 | } | |
987 | ||
988 | static struct block_device_operations dm_blk_dops; | |
989 | ||
990 | /* | |
991 | * Allocate and initialise a blank device with a given minor. | |
992 | */ | |
993 | static struct mapped_device *alloc_dev(int minor) | |
994 | { | |
995 | int r; | |
996 | struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL); | |
997 | void *old_md; | |
998 | ||
999 | if (!md) { | |
1000 | DMWARN("unable to allocate device, out of memory."); | |
1001 | return NULL; | |
1002 | } | |
1003 | ||
1004 | if (!try_module_get(THIS_MODULE)) | |
1005 | goto bad_module_get; | |
1006 | ||
1007 | /* get a minor number for the dev */ | |
1008 | if (minor == DM_ANY_MINOR) | |
1009 | r = next_free_minor(md, &minor); | |
1010 | else | |
1011 | r = specific_minor(md, minor); | |
1012 | if (r < 0) | |
1013 | goto bad_minor; | |
1014 | ||
1015 | memset(md, 0, sizeof(*md)); | |
1016 | init_rwsem(&md->io_lock); | |
1017 | mutex_init(&md->suspend_lock); | |
1018 | spin_lock_init(&md->pushback_lock); | |
1019 | rwlock_init(&md->map_lock); | |
1020 | atomic_set(&md->holders, 1); | |
1021 | atomic_set(&md->open_count, 0); | |
1022 | atomic_set(&md->event_nr, 0); | |
1023 | atomic_set(&md->uevent_seq, 0); | |
1024 | INIT_LIST_HEAD(&md->uevent_list); | |
1025 | spin_lock_init(&md->uevent_lock); | |
1026 | ||
1027 | md->queue = blk_alloc_queue(GFP_KERNEL); | |
1028 | if (!md->queue) | |
1029 | goto bad_queue; | |
1030 | ||
1031 | md->queue->queuedata = md; | |
1032 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1033 | md->queue->backing_dev_info.congested_data = md; | |
1034 | blk_queue_make_request(md->queue, dm_request); | |
1035 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
1036 | md->queue->unplug_fn = dm_unplug_all; | |
1037 | ||
1038 | md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache); | |
1039 | if (!md->io_pool) | |
1040 | goto bad_io_pool; | |
1041 | ||
1042 | md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache); | |
1043 | if (!md->tio_pool) | |
1044 | goto bad_tio_pool; | |
1045 | ||
1046 | md->bs = bioset_create(16, 16); | |
1047 | if (!md->bs) | |
1048 | goto bad_no_bioset; | |
1049 | ||
1050 | md->disk = alloc_disk(1); | |
1051 | if (!md->disk) | |
1052 | goto bad_disk; | |
1053 | ||
1054 | atomic_set(&md->pending, 0); | |
1055 | init_waitqueue_head(&md->wait); | |
1056 | init_waitqueue_head(&md->eventq); | |
1057 | ||
1058 | md->disk->major = _major; | |
1059 | md->disk->first_minor = minor; | |
1060 | md->disk->fops = &dm_blk_dops; | |
1061 | md->disk->queue = md->queue; | |
1062 | md->disk->private_data = md; | |
1063 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1064 | add_disk(md->disk); | |
1065 | format_dev_t(md->name, MKDEV(_major, minor)); | |
1066 | ||
1067 | md->wq = create_singlethread_workqueue("kdmflush"); | |
1068 | if (!md->wq) | |
1069 | goto bad_thread; | |
1070 | ||
1071 | /* Populate the mapping, nobody knows we exist yet */ | |
1072 | spin_lock(&_minor_lock); | |
1073 | old_md = idr_replace(&_minor_idr, md, minor); | |
1074 | spin_unlock(&_minor_lock); | |
1075 | ||
1076 | BUG_ON(old_md != MINOR_ALLOCED); | |
1077 | ||
1078 | return md; | |
1079 | ||
1080 | bad_thread: | |
1081 | put_disk(md->disk); | |
1082 | bad_disk: | |
1083 | bioset_free(md->bs); | |
1084 | bad_no_bioset: | |
1085 | mempool_destroy(md->tio_pool); | |
1086 | bad_tio_pool: | |
1087 | mempool_destroy(md->io_pool); | |
1088 | bad_io_pool: | |
1089 | blk_cleanup_queue(md->queue); | |
1090 | bad_queue: | |
1091 | free_minor(minor); | |
1092 | bad_minor: | |
1093 | module_put(THIS_MODULE); | |
1094 | bad_module_get: | |
1095 | kfree(md); | |
1096 | return NULL; | |
1097 | } | |
1098 | ||
1099 | static void unlock_fs(struct mapped_device *md); | |
1100 | ||
1101 | static void free_dev(struct mapped_device *md) | |
1102 | { | |
1103 | int minor = md->disk->first_minor; | |
1104 | ||
1105 | if (md->suspended_bdev) { | |
1106 | unlock_fs(md); | |
1107 | bdput(md->suspended_bdev); | |
1108 | } | |
1109 | destroy_workqueue(md->wq); | |
1110 | mempool_destroy(md->tio_pool); | |
1111 | mempool_destroy(md->io_pool); | |
1112 | bioset_free(md->bs); | |
1113 | del_gendisk(md->disk); | |
1114 | free_minor(minor); | |
1115 | ||
1116 | spin_lock(&_minor_lock); | |
1117 | md->disk->private_data = NULL; | |
1118 | spin_unlock(&_minor_lock); | |
1119 | ||
1120 | put_disk(md->disk); | |
1121 | blk_cleanup_queue(md->queue); | |
1122 | module_put(THIS_MODULE); | |
1123 | kfree(md); | |
1124 | } | |
1125 | ||
1126 | /* | |
1127 | * Bind a table to the device. | |
1128 | */ | |
1129 | static void event_callback(void *context) | |
1130 | { | |
1131 | unsigned long flags; | |
1132 | LIST_HEAD(uevents); | |
1133 | struct mapped_device *md = (struct mapped_device *) context; | |
1134 | ||
1135 | spin_lock_irqsave(&md->uevent_lock, flags); | |
1136 | list_splice_init(&md->uevent_list, &uevents); | |
1137 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1138 | ||
1139 | dm_send_uevents(&uevents, &md->disk->dev.kobj); | |
1140 | ||
1141 | atomic_inc(&md->event_nr); | |
1142 | wake_up(&md->eventq); | |
1143 | } | |
1144 | ||
1145 | static void __set_size(struct mapped_device *md, sector_t size) | |
1146 | { | |
1147 | set_capacity(md->disk, size); | |
1148 | ||
1149 | mutex_lock(&md->suspended_bdev->bd_inode->i_mutex); | |
1150 | i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); | |
1151 | mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex); | |
1152 | } | |
1153 | ||
1154 | static int __bind(struct mapped_device *md, struct dm_table *t) | |
1155 | { | |
1156 | struct request_queue *q = md->queue; | |
1157 | sector_t size; | |
1158 | ||
1159 | size = dm_table_get_size(t); | |
1160 | ||
1161 | /* | |
1162 | * Wipe any geometry if the size of the table changed. | |
1163 | */ | |
1164 | if (size != get_capacity(md->disk)) | |
1165 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
1166 | ||
1167 | if (md->suspended_bdev) | |
1168 | __set_size(md, size); | |
1169 | if (size == 0) | |
1170 | return 0; | |
1171 | ||
1172 | dm_table_get(t); | |
1173 | dm_table_event_callback(t, event_callback, md); | |
1174 | ||
1175 | write_lock(&md->map_lock); | |
1176 | md->map = t; | |
1177 | dm_table_set_restrictions(t, q); | |
1178 | write_unlock(&md->map_lock); | |
1179 | ||
1180 | return 0; | |
1181 | } | |
1182 | ||
1183 | static void __unbind(struct mapped_device *md) | |
1184 | { | |
1185 | struct dm_table *map = md->map; | |
1186 | ||
1187 | if (!map) | |
1188 | return; | |
1189 | ||
1190 | dm_table_event_callback(map, NULL, NULL); | |
1191 | write_lock(&md->map_lock); | |
1192 | md->map = NULL; | |
1193 | write_unlock(&md->map_lock); | |
1194 | dm_table_put(map); | |
1195 | } | |
1196 | ||
1197 | /* | |
1198 | * Constructor for a new device. | |
1199 | */ | |
1200 | int dm_create(int minor, struct mapped_device **result) | |
1201 | { | |
1202 | struct mapped_device *md; | |
1203 | ||
1204 | md = alloc_dev(minor); | |
1205 | if (!md) | |
1206 | return -ENXIO; | |
1207 | ||
1208 | *result = md; | |
1209 | return 0; | |
1210 | } | |
1211 | ||
1212 | static struct mapped_device *dm_find_md(dev_t dev) | |
1213 | { | |
1214 | struct mapped_device *md; | |
1215 | unsigned minor = MINOR(dev); | |
1216 | ||
1217 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
1218 | return NULL; | |
1219 | ||
1220 | spin_lock(&_minor_lock); | |
1221 | ||
1222 | md = idr_find(&_minor_idr, minor); | |
1223 | if (md && (md == MINOR_ALLOCED || | |
1224 | (dm_disk(md)->first_minor != minor) || | |
1225 | test_bit(DMF_FREEING, &md->flags))) { | |
1226 | md = NULL; | |
1227 | goto out; | |
1228 | } | |
1229 | ||
1230 | out: | |
1231 | spin_unlock(&_minor_lock); | |
1232 | ||
1233 | return md; | |
1234 | } | |
1235 | ||
1236 | struct mapped_device *dm_get_md(dev_t dev) | |
1237 | { | |
1238 | struct mapped_device *md = dm_find_md(dev); | |
1239 | ||
1240 | if (md) | |
1241 | dm_get(md); | |
1242 | ||
1243 | return md; | |
1244 | } | |
1245 | ||
1246 | void *dm_get_mdptr(struct mapped_device *md) | |
1247 | { | |
1248 | return md->interface_ptr; | |
1249 | } | |
1250 | ||
1251 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
1252 | { | |
1253 | md->interface_ptr = ptr; | |
1254 | } | |
1255 | ||
1256 | void dm_get(struct mapped_device *md) | |
1257 | { | |
1258 | atomic_inc(&md->holders); | |
1259 | } | |
1260 | ||
1261 | const char *dm_device_name(struct mapped_device *md) | |
1262 | { | |
1263 | return md->name; | |
1264 | } | |
1265 | EXPORT_SYMBOL_GPL(dm_device_name); | |
1266 | ||
1267 | void dm_put(struct mapped_device *md) | |
1268 | { | |
1269 | struct dm_table *map; | |
1270 | ||
1271 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); | |
1272 | ||
1273 | if (atomic_dec_and_lock(&md->holders, &_minor_lock)) { | |
1274 | map = dm_get_table(md); | |
1275 | idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor); | |
1276 | set_bit(DMF_FREEING, &md->flags); | |
1277 | spin_unlock(&_minor_lock); | |
1278 | if (!dm_suspended(md)) { | |
1279 | dm_table_presuspend_targets(map); | |
1280 | dm_table_postsuspend_targets(map); | |
1281 | } | |
1282 | __unbind(md); | |
1283 | dm_table_put(map); | |
1284 | free_dev(md); | |
1285 | } | |
1286 | } | |
1287 | EXPORT_SYMBOL_GPL(dm_put); | |
1288 | ||
1289 | static int dm_wait_for_completion(struct mapped_device *md) | |
1290 | { | |
1291 | int r = 0; | |
1292 | ||
1293 | while (1) { | |
1294 | set_current_state(TASK_INTERRUPTIBLE); | |
1295 | ||
1296 | smp_mb(); | |
1297 | if (!atomic_read(&md->pending)) | |
1298 | break; | |
1299 | ||
1300 | if (signal_pending(current)) { | |
1301 | r = -EINTR; | |
1302 | break; | |
1303 | } | |
1304 | ||
1305 | io_schedule(); | |
1306 | } | |
1307 | set_current_state(TASK_RUNNING); | |
1308 | ||
1309 | return r; | |
1310 | } | |
1311 | ||
1312 | /* | |
1313 | * Process the deferred bios | |
1314 | */ | |
1315 | static void __flush_deferred_io(struct mapped_device *md) | |
1316 | { | |
1317 | struct bio *c; | |
1318 | ||
1319 | while ((c = bio_list_pop(&md->deferred))) { | |
1320 | if (__split_bio(md, c)) | |
1321 | bio_io_error(c); | |
1322 | } | |
1323 | ||
1324 | clear_bit(DMF_BLOCK_IO, &md->flags); | |
1325 | } | |
1326 | ||
1327 | static void __merge_pushback_list(struct mapped_device *md) | |
1328 | { | |
1329 | unsigned long flags; | |
1330 | ||
1331 | spin_lock_irqsave(&md->pushback_lock, flags); | |
1332 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1333 | bio_list_merge_head(&md->deferred, &md->pushback); | |
1334 | bio_list_init(&md->pushback); | |
1335 | spin_unlock_irqrestore(&md->pushback_lock, flags); | |
1336 | } | |
1337 | ||
1338 | static void dm_wq_work(struct work_struct *work) | |
1339 | { | |
1340 | struct dm_wq_req *req = container_of(work, struct dm_wq_req, work); | |
1341 | struct mapped_device *md = req->md; | |
1342 | ||
1343 | down_write(&md->io_lock); | |
1344 | switch (req->type) { | |
1345 | case DM_WQ_FLUSH_ALL: | |
1346 | __merge_pushback_list(md); | |
1347 | /* pass through */ | |
1348 | case DM_WQ_FLUSH_DEFERRED: | |
1349 | __flush_deferred_io(md); | |
1350 | break; | |
1351 | default: | |
1352 | DMERR("dm_wq_work: unrecognised work type %d", req->type); | |
1353 | BUG(); | |
1354 | } | |
1355 | up_write(&md->io_lock); | |
1356 | } | |
1357 | ||
1358 | static void dm_wq_queue(struct mapped_device *md, int type, void *context, | |
1359 | struct dm_wq_req *req) | |
1360 | { | |
1361 | req->type = type; | |
1362 | req->md = md; | |
1363 | req->context = context; | |
1364 | INIT_WORK(&req->work, dm_wq_work); | |
1365 | queue_work(md->wq, &req->work); | |
1366 | } | |
1367 | ||
1368 | static void dm_queue_flush(struct mapped_device *md, int type, void *context) | |
1369 | { | |
1370 | struct dm_wq_req req; | |
1371 | ||
1372 | dm_wq_queue(md, type, context, &req); | |
1373 | flush_workqueue(md->wq); | |
1374 | } | |
1375 | ||
1376 | /* | |
1377 | * Swap in a new table (destroying old one). | |
1378 | */ | |
1379 | int dm_swap_table(struct mapped_device *md, struct dm_table *table) | |
1380 | { | |
1381 | int r = -EINVAL; | |
1382 | ||
1383 | mutex_lock(&md->suspend_lock); | |
1384 | ||
1385 | /* device must be suspended */ | |
1386 | if (!dm_suspended(md)) | |
1387 | goto out; | |
1388 | ||
1389 | /* without bdev, the device size cannot be changed */ | |
1390 | if (!md->suspended_bdev) | |
1391 | if (get_capacity(md->disk) != dm_table_get_size(table)) | |
1392 | goto out; | |
1393 | ||
1394 | __unbind(md); | |
1395 | r = __bind(md, table); | |
1396 | ||
1397 | out: | |
1398 | mutex_unlock(&md->suspend_lock); | |
1399 | return r; | |
1400 | } | |
1401 | ||
1402 | /* | |
1403 | * Functions to lock and unlock any filesystem running on the | |
1404 | * device. | |
1405 | */ | |
1406 | static int lock_fs(struct mapped_device *md) | |
1407 | { | |
1408 | int r; | |
1409 | ||
1410 | WARN_ON(md->frozen_sb); | |
1411 | ||
1412 | md->frozen_sb = freeze_bdev(md->suspended_bdev); | |
1413 | if (IS_ERR(md->frozen_sb)) { | |
1414 | r = PTR_ERR(md->frozen_sb); | |
1415 | md->frozen_sb = NULL; | |
1416 | return r; | |
1417 | } | |
1418 | ||
1419 | set_bit(DMF_FROZEN, &md->flags); | |
1420 | ||
1421 | /* don't bdput right now, we don't want the bdev | |
1422 | * to go away while it is locked. | |
1423 | */ | |
1424 | return 0; | |
1425 | } | |
1426 | ||
1427 | static void unlock_fs(struct mapped_device *md) | |
1428 | { | |
1429 | if (!test_bit(DMF_FROZEN, &md->flags)) | |
1430 | return; | |
1431 | ||
1432 | thaw_bdev(md->suspended_bdev, md->frozen_sb); | |
1433 | md->frozen_sb = NULL; | |
1434 | clear_bit(DMF_FROZEN, &md->flags); | |
1435 | } | |
1436 | ||
1437 | /* | |
1438 | * We need to be able to change a mapping table under a mounted | |
1439 | * filesystem. For example we might want to move some data in | |
1440 | * the background. Before the table can be swapped with | |
1441 | * dm_bind_table, dm_suspend must be called to flush any in | |
1442 | * flight bios and ensure that any further io gets deferred. | |
1443 | */ | |
1444 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
1445 | { | |
1446 | struct dm_table *map = NULL; | |
1447 | DECLARE_WAITQUEUE(wait, current); | |
1448 | int r = 0; | |
1449 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; | |
1450 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; | |
1451 | ||
1452 | mutex_lock(&md->suspend_lock); | |
1453 | ||
1454 | if (dm_suspended(md)) { | |
1455 | r = -EINVAL; | |
1456 | goto out_unlock; | |
1457 | } | |
1458 | ||
1459 | map = dm_get_table(md); | |
1460 | ||
1461 | /* | |
1462 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
1463 | * This flag is cleared before dm_suspend returns. | |
1464 | */ | |
1465 | if (noflush) | |
1466 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
1467 | ||
1468 | /* This does not get reverted if there's an error later. */ | |
1469 | dm_table_presuspend_targets(map); | |
1470 | ||
1471 | /* bdget() can stall if the pending I/Os are not flushed */ | |
1472 | if (!noflush) { | |
1473 | md->suspended_bdev = bdget_disk(md->disk, 0); | |
1474 | if (!md->suspended_bdev) { | |
1475 | DMWARN("bdget failed in dm_suspend"); | |
1476 | r = -ENOMEM; | |
1477 | goto flush_and_out; | |
1478 | } | |
1479 | ||
1480 | /* | |
1481 | * Flush I/O to the device. noflush supersedes do_lockfs, | |
1482 | * because lock_fs() needs to flush I/Os. | |
1483 | */ | |
1484 | if (do_lockfs) { | |
1485 | r = lock_fs(md); | |
1486 | if (r) | |
1487 | goto out; | |
1488 | } | |
1489 | } | |
1490 | ||
1491 | /* | |
1492 | * First we set the BLOCK_IO flag so no more ios will be mapped. | |
1493 | */ | |
1494 | down_write(&md->io_lock); | |
1495 | set_bit(DMF_BLOCK_IO, &md->flags); | |
1496 | ||
1497 | add_wait_queue(&md->wait, &wait); | |
1498 | up_write(&md->io_lock); | |
1499 | ||
1500 | /* unplug */ | |
1501 | if (map) | |
1502 | dm_table_unplug_all(map); | |
1503 | ||
1504 | /* | |
1505 | * Wait for the already-mapped ios to complete. | |
1506 | */ | |
1507 | r = dm_wait_for_completion(md); | |
1508 | ||
1509 | down_write(&md->io_lock); | |
1510 | remove_wait_queue(&md->wait, &wait); | |
1511 | ||
1512 | if (noflush) | |
1513 | __merge_pushback_list(md); | |
1514 | up_write(&md->io_lock); | |
1515 | ||
1516 | /* were we interrupted ? */ | |
1517 | if (r < 0) { | |
1518 | dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL); | |
1519 | ||
1520 | unlock_fs(md); | |
1521 | goto out; /* pushback list is already flushed, so skip flush */ | |
1522 | } | |
1523 | ||
1524 | dm_table_postsuspend_targets(map); | |
1525 | ||
1526 | set_bit(DMF_SUSPENDED, &md->flags); | |
1527 | ||
1528 | flush_and_out: | |
1529 | if (r && noflush) | |
1530 | /* | |
1531 | * Because there may be already I/Os in the pushback list, | |
1532 | * flush them before return. | |
1533 | */ | |
1534 | dm_queue_flush(md, DM_WQ_FLUSH_ALL, NULL); | |
1535 | ||
1536 | out: | |
1537 | if (r && md->suspended_bdev) { | |
1538 | bdput(md->suspended_bdev); | |
1539 | md->suspended_bdev = NULL; | |
1540 | } | |
1541 | ||
1542 | dm_table_put(map); | |
1543 | ||
1544 | out_unlock: | |
1545 | mutex_unlock(&md->suspend_lock); | |
1546 | return r; | |
1547 | } | |
1548 | ||
1549 | int dm_resume(struct mapped_device *md) | |
1550 | { | |
1551 | int r = -EINVAL; | |
1552 | struct dm_table *map = NULL; | |
1553 | ||
1554 | mutex_lock(&md->suspend_lock); | |
1555 | if (!dm_suspended(md)) | |
1556 | goto out; | |
1557 | ||
1558 | map = dm_get_table(md); | |
1559 | if (!map || !dm_table_get_size(map)) | |
1560 | goto out; | |
1561 | ||
1562 | r = dm_table_resume_targets(map); | |
1563 | if (r) | |
1564 | goto out; | |
1565 | ||
1566 | dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL); | |
1567 | ||
1568 | unlock_fs(md); | |
1569 | ||
1570 | if (md->suspended_bdev) { | |
1571 | bdput(md->suspended_bdev); | |
1572 | md->suspended_bdev = NULL; | |
1573 | } | |
1574 | ||
1575 | clear_bit(DMF_SUSPENDED, &md->flags); | |
1576 | ||
1577 | dm_table_unplug_all(map); | |
1578 | ||
1579 | dm_kobject_uevent(md); | |
1580 | ||
1581 | r = 0; | |
1582 | ||
1583 | out: | |
1584 | dm_table_put(map); | |
1585 | mutex_unlock(&md->suspend_lock); | |
1586 | ||
1587 | return r; | |
1588 | } | |
1589 | ||
1590 | /*----------------------------------------------------------------- | |
1591 | * Event notification. | |
1592 | *---------------------------------------------------------------*/ | |
1593 | void dm_kobject_uevent(struct mapped_device *md) | |
1594 | { | |
1595 | kobject_uevent(&md->disk->dev.kobj, KOBJ_CHANGE); | |
1596 | } | |
1597 | ||
1598 | uint32_t dm_next_uevent_seq(struct mapped_device *md) | |
1599 | { | |
1600 | return atomic_add_return(1, &md->uevent_seq); | |
1601 | } | |
1602 | ||
1603 | uint32_t dm_get_event_nr(struct mapped_device *md) | |
1604 | { | |
1605 | return atomic_read(&md->event_nr); | |
1606 | } | |
1607 | ||
1608 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
1609 | { | |
1610 | return wait_event_interruptible(md->eventq, | |
1611 | (event_nr != atomic_read(&md->event_nr))); | |
1612 | } | |
1613 | ||
1614 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) | |
1615 | { | |
1616 | unsigned long flags; | |
1617 | ||
1618 | spin_lock_irqsave(&md->uevent_lock, flags); | |
1619 | list_add(elist, &md->uevent_list); | |
1620 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1621 | } | |
1622 | ||
1623 | /* | |
1624 | * The gendisk is only valid as long as you have a reference | |
1625 | * count on 'md'. | |
1626 | */ | |
1627 | struct gendisk *dm_disk(struct mapped_device *md) | |
1628 | { | |
1629 | return md->disk; | |
1630 | } | |
1631 | ||
1632 | int dm_suspended(struct mapped_device *md) | |
1633 | { | |
1634 | return test_bit(DMF_SUSPENDED, &md->flags); | |
1635 | } | |
1636 | ||
1637 | int dm_noflush_suspending(struct dm_target *ti) | |
1638 | { | |
1639 | struct mapped_device *md = dm_table_get_md(ti->table); | |
1640 | int r = __noflush_suspending(md); | |
1641 | ||
1642 | dm_put(md); | |
1643 | ||
1644 | return r; | |
1645 | } | |
1646 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
1647 | ||
1648 | static struct block_device_operations dm_blk_dops = { | |
1649 | .open = dm_blk_open, | |
1650 | .release = dm_blk_close, | |
1651 | .ioctl = dm_blk_ioctl, | |
1652 | .getgeo = dm_blk_getgeo, | |
1653 | .owner = THIS_MODULE | |
1654 | }; | |
1655 | ||
1656 | EXPORT_SYMBOL(dm_get_mapinfo); | |
1657 | ||
1658 | /* | |
1659 | * module hooks | |
1660 | */ | |
1661 | module_init(dm_init); | |
1662 | module_exit(dm_exit); | |
1663 | ||
1664 | module_param(major, uint, 0); | |
1665 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
1666 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
1667 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
1668 | MODULE_LICENSE("GPL"); |