blob: 3ed2e53b9eb6aa71051e7aafc3ecbbee132b732c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07003 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
5 * This file is released under the GPL.
6 */
7
8#include "dm.h"
9#include "dm-bio-list.h"
10
11#include <linux/init.h>
12#include <linux/module.h>
Arjan van de Ven48c9c272006-03-27 01:18:20 -080013#include <linux/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/moduleparam.h>
15#include <linux/blkpg.h>
16#include <linux/bio.h>
17#include <linux/buffer_head.h>
18#include <linux/mempool.h>
19#include <linux/slab.h>
20#include <linux/idr.h>
Darrick J. Wong3ac51e72006-03-27 01:17:54 -080021#include <linux/hdreg.h>
Jens Axboe2056a782006-03-23 20:00:26 +010022#include <linux/blktrace_api.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
Alasdair G Kergon72d94862006-06-26 00:27:35 -070024#define DM_MSG_PREFIX "core"
25
Linus Torvalds1da177e2005-04-16 15:20:36 -070026static const char *_name = DM_NAME;
27
28static unsigned int major = 0;
29static unsigned int _major = 0;
30
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -070031static DEFINE_SPINLOCK(_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070032/*
33 * One of these is allocated per bio.
34 */
35struct dm_io {
36 struct mapped_device *md;
37 int error;
38 struct bio *bio;
39 atomic_t io_count;
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -080040 unsigned long start_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041};
42
43/*
44 * One of these is allocated per target within a bio. Hopefully
45 * this will be simplified out one day.
46 */
47struct target_io {
48 struct dm_io *io;
49 struct dm_target *ti;
50 union map_info info;
51};
52
53union map_info *dm_get_mapinfo(struct bio *bio)
54{
Alasdair G Kergon17b2f662006-06-26 00:27:33 -070055 if (bio && bio->bi_private)
56 return &((struct target_io *)bio->bi_private)->info;
57 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058}
59
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -070060#define MINOR_ALLOCED ((void *)-1)
61
Linus Torvalds1da177e2005-04-16 15:20:36 -070062/*
63 * Bits for the md->flags field.
64 */
65#define DMF_BLOCK_IO 0
66#define DMF_SUSPENDED 1
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -080067#define DMF_FROZEN 2
Jeff Mahoneyfba9f902006-06-26 00:27:23 -070068#define DMF_FREEING 3
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -070069#define DMF_DELETING 4
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
71struct mapped_device {
Alasdair G Kergon2ca33102005-07-28 21:16:00 -070072 struct rw_semaphore io_lock;
73 struct semaphore suspend_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 rwlock_t map_lock;
75 atomic_t holders;
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -070076 atomic_t open_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
78 unsigned long flags;
79
80 request_queue_t *queue;
81 struct gendisk *disk;
Mike Anderson7e51f252006-03-27 01:17:52 -080082 char name[16];
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
84 void *interface_ptr;
85
86 /*
87 * A list of ios that arrived while we were suspended.
88 */
89 atomic_t pending;
90 wait_queue_head_t wait;
91 struct bio_list deferred;
92
93 /*
94 * The current mapping.
95 */
96 struct dm_table *map;
97
98 /*
99 * io objects are allocated from here.
100 */
101 mempool_t *io_pool;
102 mempool_t *tio_pool;
103
104 /*
105 * Event handling.
106 */
107 atomic_t event_nr;
108 wait_queue_head_t eventq;
109
110 /*
111 * freeze/thaw support require holding onto a super block
112 */
113 struct super_block *frozen_sb;
Alasdair G Kergone39e2e92006-01-06 00:20:05 -0800114 struct block_device *suspended_bdev;
Darrick J. Wong3ac51e72006-03-27 01:17:54 -0800115
116 /* forced geometry settings */
117 struct hd_geometry geometry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118};
119
120#define MIN_IOS 256
121static kmem_cache_t *_io_cache;
122static kmem_cache_t *_tio_cache;
123
124static struct bio_set *dm_set;
125
126static int __init local_init(void)
127{
128 int r;
129
130 dm_set = bioset_create(16, 16, 4);
131 if (!dm_set)
132 return -ENOMEM;
133
134 /* allocate a slab for the dm_ios */
135 _io_cache = kmem_cache_create("dm_io",
136 sizeof(struct dm_io), 0, 0, NULL, NULL);
137 if (!_io_cache)
138 return -ENOMEM;
139
140 /* allocate a slab for the target ios */
141 _tio_cache = kmem_cache_create("dm_tio", sizeof(struct target_io),
142 0, 0, NULL, NULL);
143 if (!_tio_cache) {
144 kmem_cache_destroy(_io_cache);
145 return -ENOMEM;
146 }
147
148 _major = major;
149 r = register_blkdev(_major, _name);
150 if (r < 0) {
151 kmem_cache_destroy(_tio_cache);
152 kmem_cache_destroy(_io_cache);
153 return r;
154 }
155
156 if (!_major)
157 _major = r;
158
159 return 0;
160}
161
162static void local_exit(void)
163{
164 kmem_cache_destroy(_tio_cache);
165 kmem_cache_destroy(_io_cache);
166
167 bioset_free(dm_set);
168
169 if (unregister_blkdev(_major, _name) < 0)
170 DMERR("devfs_unregister_blkdev failed");
171
172 _major = 0;
173
174 DMINFO("cleaned up");
175}
176
177int (*_inits[])(void) __initdata = {
178 local_init,
179 dm_target_init,
180 dm_linear_init,
181 dm_stripe_init,
182 dm_interface_init,
183};
184
185void (*_exits[])(void) = {
186 local_exit,
187 dm_target_exit,
188 dm_linear_exit,
189 dm_stripe_exit,
190 dm_interface_exit,
191};
192
193static int __init dm_init(void)
194{
195 const int count = ARRAY_SIZE(_inits);
196
197 int r, i;
198
199 for (i = 0; i < count; i++) {
200 r = _inits[i]();
201 if (r)
202 goto bad;
203 }
204
205 return 0;
206
207 bad:
208 while (i--)
209 _exits[i]();
210
211 return r;
212}
213
214static void __exit dm_exit(void)
215{
216 int i = ARRAY_SIZE(_exits);
217
218 while (i--)
219 _exits[i]();
220}
221
222/*
223 * Block device functions
224 */
225static int dm_blk_open(struct inode *inode, struct file *file)
226{
227 struct mapped_device *md;
228
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700229 spin_lock(&_minor_lock);
230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 md = inode->i_bdev->bd_disk->private_data;
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700232 if (!md)
233 goto out;
234
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700235 if (test_bit(DMF_FREEING, &md->flags) ||
236 test_bit(DMF_DELETING, &md->flags)) {
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700237 md = NULL;
238 goto out;
239 }
240
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 dm_get(md);
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700242 atomic_inc(&md->open_count);
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700243
244out:
245 spin_unlock(&_minor_lock);
246
247 return md ? 0 : -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248}
249
250static int dm_blk_close(struct inode *inode, struct file *file)
251{
252 struct mapped_device *md;
253
254 md = inode->i_bdev->bd_disk->private_data;
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700255 atomic_dec(&md->open_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 dm_put(md);
257 return 0;
258}
259
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700260int dm_open_count(struct mapped_device *md)
261{
262 return atomic_read(&md->open_count);
263}
264
265/*
266 * Guarantees nothing is using the device before it's deleted.
267 */
268int dm_lock_for_deletion(struct mapped_device *md)
269{
270 int r = 0;
271
272 spin_lock(&_minor_lock);
273
274 if (dm_open_count(md))
275 r = -EBUSY;
276 else
277 set_bit(DMF_DELETING, &md->flags);
278
279 spin_unlock(&_minor_lock);
280
281 return r;
282}
283
Darrick J. Wong3ac51e72006-03-27 01:17:54 -0800284static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
285{
286 struct mapped_device *md = bdev->bd_disk->private_data;
287
288 return dm_get_geometry(md, geo);
289}
290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291static inline struct dm_io *alloc_io(struct mapped_device *md)
292{
293 return mempool_alloc(md->io_pool, GFP_NOIO);
294}
295
296static inline void free_io(struct mapped_device *md, struct dm_io *io)
297{
298 mempool_free(io, md->io_pool);
299}
300
301static inline struct target_io *alloc_tio(struct mapped_device *md)
302{
303 return mempool_alloc(md->tio_pool, GFP_NOIO);
304}
305
306static inline void free_tio(struct mapped_device *md, struct target_io *tio)
307{
308 mempool_free(tio, md->tio_pool);
309}
310
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800311static void start_io_acct(struct dm_io *io)
312{
313 struct mapped_device *md = io->md;
314
315 io->start_time = jiffies;
316
317 preempt_disable();
318 disk_round_stats(dm_disk(md));
319 preempt_enable();
320 dm_disk(md)->in_flight = atomic_inc_return(&md->pending);
321}
322
323static int end_io_acct(struct dm_io *io)
324{
325 struct mapped_device *md = io->md;
326 struct bio *bio = io->bio;
327 unsigned long duration = jiffies - io->start_time;
328 int pending;
329 int rw = bio_data_dir(bio);
330
331 preempt_disable();
332 disk_round_stats(dm_disk(md));
333 preempt_enable();
334 dm_disk(md)->in_flight = pending = atomic_dec_return(&md->pending);
335
336 disk_stat_add(dm_disk(md), ticks[rw], duration);
337
338 return !pending;
339}
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341/*
342 * Add the bio to the list of deferred io.
343 */
344static int queue_io(struct mapped_device *md, struct bio *bio)
345{
Alasdair G Kergon2ca33102005-07-28 21:16:00 -0700346 down_write(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
348 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
Alasdair G Kergon2ca33102005-07-28 21:16:00 -0700349 up_write(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 return 1;
351 }
352
353 bio_list_add(&md->deferred, bio);
354
Alasdair G Kergon2ca33102005-07-28 21:16:00 -0700355 up_write(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 return 0; /* deferred successfully */
357}
358
359/*
360 * Everyone (including functions in this file), should use this
361 * function to access the md->map field, and make sure they call
362 * dm_table_put() when finished.
363 */
364struct dm_table *dm_get_table(struct mapped_device *md)
365{
366 struct dm_table *t;
367
368 read_lock(&md->map_lock);
369 t = md->map;
370 if (t)
371 dm_table_get(t);
372 read_unlock(&md->map_lock);
373
374 return t;
375}
376
Darrick J. Wong3ac51e72006-03-27 01:17:54 -0800377/*
378 * Get the geometry associated with a dm device
379 */
380int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
381{
382 *geo = md->geometry;
383
384 return 0;
385}
386
387/*
388 * Set the geometry of a device.
389 */
390int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
391{
392 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
393
394 if (geo->start > sz) {
395 DMWARN("Start sector is beyond the geometry limits.");
396 return -EINVAL;
397 }
398
399 md->geometry = *geo;
400
401 return 0;
402}
403
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404/*-----------------------------------------------------------------
405 * CRUD START:
406 * A more elegant soln is in the works that uses the queue
407 * merge fn, unfortunately there are a couple of changes to
408 * the block layer that I want to make for this. So in the
409 * interests of getting something for people to use I give
410 * you this clearly demarcated crap.
411 *---------------------------------------------------------------*/
412
413/*
414 * Decrements the number of outstanding ios that a bio has been
415 * cloned into, completing the original io if necc.
416 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800417static void dec_pending(struct dm_io *io, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418{
419 if (error)
420 io->error = error;
421
422 if (atomic_dec_and_test(&io->io_count)) {
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800423 if (end_io_acct(io))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 /* nudge anyone waiting on suspend queue */
425 wake_up(&io->md->wait);
426
Jens Axboe2056a782006-03-23 20:00:26 +0100427 blk_add_trace_bio(io->md->queue, io->bio, BLK_TA_COMPLETE);
428
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 bio_endio(io->bio, io->bio->bi_size, io->error);
430 free_io(io->md, io);
431 }
432}
433
434static int clone_endio(struct bio *bio, unsigned int done, int error)
435{
436 int r = 0;
437 struct target_io *tio = bio->bi_private;
438 struct dm_io *io = tio->io;
439 dm_endio_fn endio = tio->ti->type->end_io;
440
441 if (bio->bi_size)
442 return 1;
443
444 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
445 error = -EIO;
446
447 if (endio) {
448 r = endio(tio->ti, bio, error, &tio->info);
449 if (r < 0)
450 error = r;
451
452 else if (r > 0)
453 /* the target wants another shot at the io */
454 return 1;
455 }
456
457 free_tio(io->md, tio);
458 dec_pending(io, error);
459 bio_put(bio);
460 return r;
461}
462
463static sector_t max_io_len(struct mapped_device *md,
464 sector_t sector, struct dm_target *ti)
465{
466 sector_t offset = sector - ti->begin;
467 sector_t len = ti->len - offset;
468
469 /*
470 * Does the target need to split even further ?
471 */
472 if (ti->split_io) {
473 sector_t boundary;
474 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
475 - offset;
476 if (len > boundary)
477 len = boundary;
478 }
479
480 return len;
481}
482
483static void __map_bio(struct dm_target *ti, struct bio *clone,
484 struct target_io *tio)
485{
486 int r;
Jens Axboe2056a782006-03-23 20:00:26 +0100487 sector_t sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
489 /*
490 * Sanity checks.
491 */
492 BUG_ON(!clone->bi_size);
493
494 clone->bi_end_io = clone_endio;
495 clone->bi_private = tio;
496
497 /*
498 * Map the clone. If r == 0 we don't need to do
499 * anything, the target has assumed ownership of
500 * this io.
501 */
502 atomic_inc(&tio->io->io_count);
Jens Axboe2056a782006-03-23 20:00:26 +0100503 sector = clone->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 r = ti->type->map(ti, clone, &tio->info);
Jens Axboe2056a782006-03-23 20:00:26 +0100505 if (r > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 /* the bio has been remapped so dispatch it */
Jens Axboe2056a782006-03-23 20:00:26 +0100507
Alasdair G Kergon17b2f662006-06-26 00:27:33 -0700508 blk_add_trace_remap(bdev_get_queue(clone->bi_bdev), clone,
509 tio->io->bio->bi_bdev->bd_dev, sector,
Jens Axboe2056a782006-03-23 20:00:26 +0100510 clone->bi_sector);
511
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 generic_make_request(clone);
Jens Axboe2056a782006-03-23 20:00:26 +0100513 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
515 else if (r < 0) {
516 /* error the io and bail out */
517 struct dm_io *io = tio->io;
518 free_tio(tio->io->md, tio);
Alasdair G Kergonf6a80ea2005-07-12 15:53:01 -0700519 dec_pending(io, r);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 bio_put(clone);
521 }
522}
523
524struct clone_info {
525 struct mapped_device *md;
526 struct dm_table *map;
527 struct bio *bio;
528 struct dm_io *io;
529 sector_t sector;
530 sector_t sector_count;
531 unsigned short idx;
532};
533
Peter Osterlund36763472005-09-06 15:16:42 -0700534static void dm_bio_destructor(struct bio *bio)
535{
536 bio_free(bio, dm_set);
537}
538
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539/*
540 * Creates a little bio that is just does part of a bvec.
541 */
542static struct bio *split_bvec(struct bio *bio, sector_t sector,
543 unsigned short idx, unsigned int offset,
544 unsigned int len)
545{
546 struct bio *clone;
547 struct bio_vec *bv = bio->bi_io_vec + idx;
548
549 clone = bio_alloc_bioset(GFP_NOIO, 1, dm_set);
Peter Osterlund36763472005-09-06 15:16:42 -0700550 clone->bi_destructor = dm_bio_destructor;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 *clone->bi_io_vec = *bv;
552
553 clone->bi_sector = sector;
554 clone->bi_bdev = bio->bi_bdev;
555 clone->bi_rw = bio->bi_rw;
556 clone->bi_vcnt = 1;
557 clone->bi_size = to_bytes(len);
558 clone->bi_io_vec->bv_offset = offset;
559 clone->bi_io_vec->bv_len = clone->bi_size;
560
561 return clone;
562}
563
564/*
565 * Creates a bio that consists of range of complete bvecs.
566 */
567static struct bio *clone_bio(struct bio *bio, sector_t sector,
568 unsigned short idx, unsigned short bv_count,
569 unsigned int len)
570{
571 struct bio *clone;
572
573 clone = bio_clone(bio, GFP_NOIO);
574 clone->bi_sector = sector;
575 clone->bi_idx = idx;
576 clone->bi_vcnt = idx + bv_count;
577 clone->bi_size = to_bytes(len);
578 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
579
580 return clone;
581}
582
583static void __clone_and_map(struct clone_info *ci)
584{
585 struct bio *clone, *bio = ci->bio;
586 struct dm_target *ti = dm_table_find_target(ci->map, ci->sector);
587 sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti);
588 struct target_io *tio;
589
590 /*
591 * Allocate a target io object.
592 */
593 tio = alloc_tio(ci->md);
594 tio->io = ci->io;
595 tio->ti = ti;
596 memset(&tio->info, 0, sizeof(tio->info));
597
598 if (ci->sector_count <= max) {
599 /*
600 * Optimise for the simple case where we can do all of
601 * the remaining io with a single clone.
602 */
603 clone = clone_bio(bio, ci->sector, ci->idx,
604 bio->bi_vcnt - ci->idx, ci->sector_count);
605 __map_bio(ti, clone, tio);
606 ci->sector_count = 0;
607
608 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
609 /*
610 * There are some bvecs that don't span targets.
611 * Do as many of these as possible.
612 */
613 int i;
614 sector_t remaining = max;
615 sector_t bv_len;
616
617 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
618 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
619
620 if (bv_len > remaining)
621 break;
622
623 remaining -= bv_len;
624 len += bv_len;
625 }
626
627 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len);
628 __map_bio(ti, clone, tio);
629
630 ci->sector += len;
631 ci->sector_count -= len;
632 ci->idx = i;
633
634 } else {
635 /*
Alasdair G Kergond2044a92006-03-22 00:07:42 -0800636 * Handle a bvec that must be split between two or more targets.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 */
638 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
Alasdair G Kergond2044a92006-03-22 00:07:42 -0800639 sector_t remaining = to_sector(bv->bv_len);
640 unsigned int offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
Alasdair G Kergond2044a92006-03-22 00:07:42 -0800642 do {
643 if (offset) {
644 ti = dm_table_find_target(ci->map, ci->sector);
645 max = max_io_len(ci->md, ci->sector, ti);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
Alasdair G Kergond2044a92006-03-22 00:07:42 -0800647 tio = alloc_tio(ci->md);
648 tio->io = ci->io;
649 tio->ti = ti;
650 memset(&tio->info, 0, sizeof(tio->info));
651 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Alasdair G Kergond2044a92006-03-22 00:07:42 -0800653 len = min(remaining, max);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
Alasdair G Kergond2044a92006-03-22 00:07:42 -0800655 clone = split_bvec(bio, ci->sector, ci->idx,
656 bv->bv_offset + offset, len);
657
658 __map_bio(ti, clone, tio);
659
660 ci->sector += len;
661 ci->sector_count -= len;
662 offset += to_bytes(len);
663 } while (remaining -= len);
664
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 ci->idx++;
666 }
667}
668
669/*
670 * Split the bio into several clones.
671 */
672static void __split_bio(struct mapped_device *md, struct bio *bio)
673{
674 struct clone_info ci;
675
676 ci.map = dm_get_table(md);
677 if (!ci.map) {
678 bio_io_error(bio, bio->bi_size);
679 return;
680 }
681
682 ci.md = md;
683 ci.bio = bio;
684 ci.io = alloc_io(md);
685 ci.io->error = 0;
686 atomic_set(&ci.io->io_count, 1);
687 ci.io->bio = bio;
688 ci.io->md = md;
689 ci.sector = bio->bi_sector;
690 ci.sector_count = bio_sectors(bio);
691 ci.idx = bio->bi_idx;
692
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800693 start_io_acct(ci.io);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 while (ci.sector_count)
695 __clone_and_map(&ci);
696
697 /* drop the extra reference count */
698 dec_pending(ci.io, 0);
699 dm_table_put(ci.map);
700}
701/*-----------------------------------------------------------------
702 * CRUD END
703 *---------------------------------------------------------------*/
704
705/*
706 * The request function that just remaps the bio built up by
707 * dm_merge_bvec.
708 */
709static int dm_request(request_queue_t *q, struct bio *bio)
710{
711 int r;
Kevin Corry12f03a42006-02-01 03:04:52 -0800712 int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 struct mapped_device *md = q->queuedata;
714
Alasdair G Kergon2ca33102005-07-28 21:16:00 -0700715 down_read(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
Kevin Corry12f03a42006-02-01 03:04:52 -0800717 disk_stat_inc(dm_disk(md), ios[rw]);
718 disk_stat_add(dm_disk(md), sectors[rw], bio_sectors(bio));
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 /*
721 * If we're suspended we have to queue
722 * this io for later.
723 */
724 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
Alasdair G Kergon2ca33102005-07-28 21:16:00 -0700725 up_read(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
727 if (bio_rw(bio) == READA) {
728 bio_io_error(bio, bio->bi_size);
729 return 0;
730 }
731
732 r = queue_io(md, bio);
733 if (r < 0) {
734 bio_io_error(bio, bio->bi_size);
735 return 0;
736
737 } else if (r == 0)
738 return 0; /* deferred successfully */
739
740 /*
741 * We're in a while loop, because someone could suspend
742 * before we get to the following read lock.
743 */
Alasdair G Kergon2ca33102005-07-28 21:16:00 -0700744 down_read(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 }
746
747 __split_bio(md, bio);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -0700748 up_read(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 return 0;
750}
751
752static int dm_flush_all(request_queue_t *q, struct gendisk *disk,
753 sector_t *error_sector)
754{
755 struct mapped_device *md = q->queuedata;
756 struct dm_table *map = dm_get_table(md);
757 int ret = -ENXIO;
758
759 if (map) {
Alasdair G Kergoncf222b32005-07-28 21:15:57 -0700760 ret = dm_table_flush_all(map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 dm_table_put(map);
762 }
763
764 return ret;
765}
766
767static void dm_unplug_all(request_queue_t *q)
768{
769 struct mapped_device *md = q->queuedata;
770 struct dm_table *map = dm_get_table(md);
771
772 if (map) {
773 dm_table_unplug_all(map);
774 dm_table_put(map);
775 }
776}
777
778static int dm_any_congested(void *congested_data, int bdi_bits)
779{
780 int r;
781 struct mapped_device *md = (struct mapped_device *) congested_data;
782 struct dm_table *map = dm_get_table(md);
783
784 if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
785 r = bdi_bits;
786 else
787 r = dm_table_any_congested(map, bdi_bits);
788
789 dm_table_put(map);
790 return r;
791}
792
793/*-----------------------------------------------------------------
794 * An IDR is used to keep track of allocated minor numbers.
795 *---------------------------------------------------------------*/
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796static DEFINE_IDR(_minor_idr);
797
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -0700798static void free_minor(int minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799{
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -0700800 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 idr_remove(&_minor_idr, minor);
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -0700802 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803}
804
805/*
806 * See if the device with a specific minor # is free.
807 */
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -0700808static int specific_minor(struct mapped_device *md, int minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809{
810 int r, m;
811
812 if (minor >= (1 << MINORBITS))
813 return -EINVAL;
814
Jeff Mahoney62f75c22006-06-26 00:27:21 -0700815 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
816 if (!r)
817 return -ENOMEM;
818
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -0700819 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821 if (idr_find(&_minor_idr, minor)) {
822 r = -EBUSY;
823 goto out;
824 }
825
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -0700826 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
Jeff Mahoney62f75c22006-06-26 00:27:21 -0700827 if (r)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
830 if (m != minor) {
831 idr_remove(&_minor_idr, m);
832 r = -EBUSY;
833 goto out;
834 }
835
836out:
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -0700837 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 return r;
839}
840
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -0700841static int next_free_minor(struct mapped_device *md, int *minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842{
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -0700843 int r, m;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
Jeff Mahoney62f75c22006-06-26 00:27:21 -0700846 if (!r)
847 return -ENOMEM;
848
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -0700849 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -0700851 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 if (r) {
853 goto out;
854 }
855
856 if (m >= (1 << MINORBITS)) {
857 idr_remove(&_minor_idr, m);
858 r = -ENOSPC;
859 goto out;
860 }
861
862 *minor = m;
863
864out:
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -0700865 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 return r;
867}
868
869static struct block_device_operations dm_blk_dops;
870
871/*
872 * Allocate and initialise a blank device with a given minor.
873 */
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -0700874static struct mapped_device *alloc_dev(int minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875{
876 int r;
877 struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -0700878 void *old_md;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
880 if (!md) {
881 DMWARN("unable to allocate device, out of memory.");
882 return NULL;
883 }
884
Jeff Mahoney10da4f72006-06-26 00:27:25 -0700885 if (!try_module_get(THIS_MODULE))
886 goto bad0;
887
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 /* get a minor number for the dev */
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -0700889 if (minor == DM_ANY_MINOR)
890 r = next_free_minor(md, &minor);
891 else
892 r = specific_minor(md, minor);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if (r < 0)
894 goto bad1;
895
896 memset(md, 0, sizeof(*md));
Alasdair G Kergon2ca33102005-07-28 21:16:00 -0700897 init_rwsem(&md->io_lock);
898 init_MUTEX(&md->suspend_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 rwlock_init(&md->map_lock);
900 atomic_set(&md->holders, 1);
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700901 atomic_set(&md->open_count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 atomic_set(&md->event_nr, 0);
903
904 md->queue = blk_alloc_queue(GFP_KERNEL);
905 if (!md->queue)
906 goto bad1;
907
908 md->queue->queuedata = md;
909 md->queue->backing_dev_info.congested_fn = dm_any_congested;
910 md->queue->backing_dev_info.congested_data = md;
911 blk_queue_make_request(md->queue, dm_request);
Jens Axboedaef2652006-01-10 10:48:02 +0100912 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 md->queue->unplug_fn = dm_unplug_all;
914 md->queue->issue_flush_fn = dm_flush_all;
915
Matthew Dobson93d23412006-03-26 01:37:50 -0800916 md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 if (!md->io_pool)
918 goto bad2;
919
Matthew Dobson93d23412006-03-26 01:37:50 -0800920 md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 if (!md->tio_pool)
922 goto bad3;
923
924 md->disk = alloc_disk(1);
925 if (!md->disk)
926 goto bad4;
927
Jeff Mahoneyf0b04112006-06-26 00:27:25 -0700928 atomic_set(&md->pending, 0);
929 init_waitqueue_head(&md->wait);
930 init_waitqueue_head(&md->eventq);
931
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 md->disk->major = _major;
933 md->disk->first_minor = minor;
934 md->disk->fops = &dm_blk_dops;
935 md->disk->queue = md->queue;
936 md->disk->private_data = md;
937 sprintf(md->disk->disk_name, "dm-%d", minor);
938 add_disk(md->disk);
Mike Anderson7e51f252006-03-27 01:17:52 -0800939 format_dev_t(md->name, MKDEV(_major, minor));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -0700941 /* Populate the mapping, nobody knows we exist yet */
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -0700942 spin_lock(&_minor_lock);
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -0700943 old_md = idr_replace(&_minor_idr, md, minor);
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -0700944 spin_unlock(&_minor_lock);
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -0700945
946 BUG_ON(old_md != MINOR_ALLOCED);
947
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 return md;
949
950 bad4:
951 mempool_destroy(md->tio_pool);
952 bad3:
953 mempool_destroy(md->io_pool);
954 bad2:
Al Viro1312f402006-03-12 11:02:03 -0500955 blk_cleanup_queue(md->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 free_minor(minor);
957 bad1:
Jeff Mahoney10da4f72006-06-26 00:27:25 -0700958 module_put(THIS_MODULE);
959 bad0:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 kfree(md);
961 return NULL;
962}
963
964static void free_dev(struct mapped_device *md)
965{
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -0700966 int minor = md->disk->first_minor;
Jun'ichi Nomura63d94e42006-02-24 13:04:25 -0800967
Jun'ichi Nomurad9dde592006-02-24 13:04:24 -0800968 if (md->suspended_bdev) {
969 thaw_bdev(md->suspended_bdev, NULL);
970 bdput(md->suspended_bdev);
971 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 mempool_destroy(md->tio_pool);
973 mempool_destroy(md->io_pool);
974 del_gendisk(md->disk);
Jun'ichi Nomura63d94e42006-02-24 13:04:25 -0800975 free_minor(minor);
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700976
977 spin_lock(&_minor_lock);
978 md->disk->private_data = NULL;
979 spin_unlock(&_minor_lock);
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 put_disk(md->disk);
Al Viro1312f402006-03-12 11:02:03 -0500982 blk_cleanup_queue(md->queue);
Jeff Mahoney10da4f72006-06-26 00:27:25 -0700983 module_put(THIS_MODULE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 kfree(md);
985}
986
987/*
988 * Bind a table to the device.
989 */
990static void event_callback(void *context)
991{
992 struct mapped_device *md = (struct mapped_device *) context;
993
994 atomic_inc(&md->event_nr);
995 wake_up(&md->eventq);
996}
997
Alasdair G Kergon4e90188be2005-07-28 21:15:59 -0700998static void __set_size(struct mapped_device *md, sector_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999{
Alasdair G Kergon4e90188be2005-07-28 21:15:59 -07001000 set_capacity(md->disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001002 mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08001003 i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001004 mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005}
1006
1007static int __bind(struct mapped_device *md, struct dm_table *t)
1008{
1009 request_queue_t *q = md->queue;
1010 sector_t size;
1011
1012 size = dm_table_get_size(t);
Darrick J. Wong3ac51e72006-03-27 01:17:54 -08001013
1014 /*
1015 * Wipe any geometry if the size of the table changed.
1016 */
1017 if (size != get_capacity(md->disk))
1018 memset(&md->geometry, 0, sizeof(md->geometry));
1019
Alasdair G Kergon4e90188be2005-07-28 21:15:59 -07001020 __set_size(md, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 if (size == 0)
1022 return 0;
1023
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 dm_table_get(t);
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001025 dm_table_event_callback(t, event_callback, md);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001026
1027 write_lock(&md->map_lock);
1028 md->map = t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 dm_table_set_restrictions(t, q);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001030 write_unlock(&md->map_lock);
1031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 return 0;
1033}
1034
1035static void __unbind(struct mapped_device *md)
1036{
1037 struct dm_table *map = md->map;
1038
1039 if (!map)
1040 return;
1041
1042 dm_table_event_callback(map, NULL, NULL);
1043 write_lock(&md->map_lock);
1044 md->map = NULL;
1045 write_unlock(&md->map_lock);
1046 dm_table_put(map);
1047}
1048
1049/*
1050 * Constructor for a new device.
1051 */
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001052int dm_create(int minor, struct mapped_device **result)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053{
1054 struct mapped_device *md;
1055
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001056 md = alloc_dev(minor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 if (!md)
1058 return -ENXIO;
1059
1060 *result = md;
1061 return 0;
1062}
1063
David Teigland637842c2006-01-06 00:20:00 -08001064static struct mapped_device *dm_find_md(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065{
1066 struct mapped_device *md;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 unsigned minor = MINOR(dev);
1068
1069 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1070 return NULL;
1071
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001072 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
1074 md = idr_find(&_minor_idr, minor);
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07001075 if (md && (md == MINOR_ALLOCED ||
1076 (dm_disk(md)->first_minor != minor) ||
Alasdair G Kergon17b2f662006-06-26 00:27:33 -07001077 test_bit(DMF_FREEING, &md->flags))) {
David Teigland637842c2006-01-06 00:20:00 -08001078 md = NULL;
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07001079 goto out;
1080 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07001082out:
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001083 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
David Teigland637842c2006-01-06 00:20:00 -08001085 return md;
1086}
1087
David Teiglandd229a952006-01-06 00:20:01 -08001088struct mapped_device *dm_get_md(dev_t dev)
1089{
1090 struct mapped_device *md = dm_find_md(dev);
1091
1092 if (md)
1093 dm_get(md);
1094
1095 return md;
1096}
1097
Alasdair G Kergon9ade92a2006-03-27 01:17:53 -08001098void *dm_get_mdptr(struct mapped_device *md)
David Teigland637842c2006-01-06 00:20:00 -08001099{
Alasdair G Kergon9ade92a2006-03-27 01:17:53 -08001100 return md->interface_ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101}
1102
1103void dm_set_mdptr(struct mapped_device *md, void *ptr)
1104{
1105 md->interface_ptr = ptr;
1106}
1107
1108void dm_get(struct mapped_device *md)
1109{
1110 atomic_inc(&md->holders);
1111}
1112
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001113const char *dm_device_name(struct mapped_device *md)
1114{
1115 return md->name;
1116}
1117EXPORT_SYMBOL_GPL(dm_device_name);
1118
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119void dm_put(struct mapped_device *md)
1120{
Mike Anderson1134e5a2006-03-27 01:17:54 -08001121 struct dm_table *map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07001123 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1124
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001125 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
Mike Anderson1134e5a2006-03-27 01:17:54 -08001126 map = dm_get_table(md);
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001127 idr_replace(&_minor_idr, MINOR_ALLOCED, dm_disk(md)->first_minor);
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07001128 set_bit(DMF_FREEING, &md->flags);
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001129 spin_unlock(&_minor_lock);
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001130 if (!dm_suspended(md)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 dm_table_presuspend_targets(map);
1132 dm_table_postsuspend_targets(map);
1133 }
1134 __unbind(md);
Mike Anderson1134e5a2006-03-27 01:17:54 -08001135 dm_table_put(map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 free_dev(md);
1137 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138}
1139
1140/*
1141 * Process the deferred bios
1142 */
1143static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
1144{
1145 struct bio *n;
1146
1147 while (c) {
1148 n = c->bi_next;
1149 c->bi_next = NULL;
1150 __split_bio(md, c);
1151 c = n;
1152 }
1153}
1154
1155/*
1156 * Swap in a new table (destroying old one).
1157 */
1158int dm_swap_table(struct mapped_device *md, struct dm_table *table)
1159{
Alasdair G Kergon93c534a2005-07-12 15:53:05 -07001160 int r = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001162 down(&md->suspend_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
1164 /* device must be suspended */
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001165 if (!dm_suspended(md))
Alasdair G Kergon93c534a2005-07-12 15:53:05 -07001166 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167
1168 __unbind(md);
1169 r = __bind(md, table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
Alasdair G Kergon93c534a2005-07-12 15:53:05 -07001171out:
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001172 up(&md->suspend_lock);
Alasdair G Kergon93c534a2005-07-12 15:53:05 -07001173 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174}
1175
1176/*
1177 * Functions to lock and unlock any filesystem running on the
1178 * device.
1179 */
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001180static int lock_fs(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181{
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08001182 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
1184 WARN_ON(md->frozen_sb);
Alasdair G Kergondfbe03f2005-05-05 16:16:04 -07001185
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08001186 md->frozen_sb = freeze_bdev(md->suspended_bdev);
Alasdair G Kergondfbe03f2005-05-05 16:16:04 -07001187 if (IS_ERR(md->frozen_sb)) {
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001188 r = PTR_ERR(md->frozen_sb);
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08001189 md->frozen_sb = NULL;
1190 return r;
Alasdair G Kergondfbe03f2005-05-05 16:16:04 -07001191 }
1192
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08001193 set_bit(DMF_FROZEN, &md->flags);
1194
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 /* don't bdput right now, we don't want the bdev
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08001196 * to go away while it is locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 */
1198 return 0;
1199}
1200
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001201static void unlock_fs(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202{
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08001203 if (!test_bit(DMF_FROZEN, &md->flags))
1204 return;
1205
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08001206 thaw_bdev(md->suspended_bdev, md->frozen_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 md->frozen_sb = NULL;
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08001208 clear_bit(DMF_FROZEN, &md->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209}
1210
1211/*
1212 * We need to be able to change a mapping table under a mounted
1213 * filesystem. For example we might want to move some data in
1214 * the background. Before the table can be swapped with
1215 * dm_bind_table, dm_suspend must be called to flush any in
1216 * flight bios and ensure that any further io gets deferred.
1217 */
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08001218int dm_suspend(struct mapped_device *md, int do_lockfs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219{
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001220 struct dm_table *map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 DECLARE_WAITQUEUE(wait, current);
Jun'ichi Nomura1ecac7f2006-03-27 01:17:51 -08001222 struct bio *def;
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001223 int r = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001225 down(&md->suspend_lock);
1226
1227 if (dm_suspended(md))
1228 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
1230 map = dm_get_table(md);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001232 /* This does not get reverted if there's an error later. */
1233 dm_table_presuspend_targets(map);
1234
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08001235 md->suspended_bdev = bdget_disk(md->disk, 0);
1236 if (!md->suspended_bdev) {
1237 DMWARN("bdget failed in dm_suspend");
1238 r = -ENOMEM;
1239 goto out;
1240 }
1241
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001242 /* Flush I/O to the device. */
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08001243 if (do_lockfs) {
1244 r = lock_fs(md);
1245 if (r)
1246 goto out;
1247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
1249 /*
Alasdair G Kergon354e0072005-05-05 16:16:05 -07001250 * First we set the BLOCK_IO flag so no more ios will be mapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 */
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001252 down_write(&md->io_lock);
1253 set_bit(DMF_BLOCK_IO, &md->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 add_wait_queue(&md->wait, &wait);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001256 up_write(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257
1258 /* unplug */
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001259 if (map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 dm_table_unplug_all(map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
1262 /*
1263 * Then we wait for the already mapped ios to
1264 * complete.
1265 */
1266 while (1) {
1267 set_current_state(TASK_INTERRUPTIBLE);
1268
1269 if (!atomic_read(&md->pending) || signal_pending(current))
1270 break;
1271
1272 io_schedule();
1273 }
1274 set_current_state(TASK_RUNNING);
1275
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001276 down_write(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 remove_wait_queue(&md->wait, &wait);
1278
1279 /* were we interrupted ? */
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001280 r = -EINTR;
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001281 if (atomic_read(&md->pending)) {
Jun'ichi Nomura1ecac7f2006-03-27 01:17:51 -08001282 clear_bit(DMF_BLOCK_IO, &md->flags);
1283 def = bio_list_get(&md->deferred);
1284 __flush_deferred_io(md, def);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001285 up_write(&md->io_lock);
1286 unlock_fs(md);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001287 goto out;
1288 }
1289 up_write(&md->io_lock);
1290
1291 dm_table_postsuspend_targets(map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
1293 set_bit(DMF_SUSPENDED, &md->flags);
1294
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001295 r = 0;
1296
1297out:
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08001298 if (r && md->suspended_bdev) {
1299 bdput(md->suspended_bdev);
1300 md->suspended_bdev = NULL;
1301 }
1302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 dm_table_put(map);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001304 up(&md->suspend_lock);
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001305 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306}
1307
1308int dm_resume(struct mapped_device *md)
1309{
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001310 int r = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 struct bio *def;
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001312 struct dm_table *map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001314 down(&md->suspend_lock);
1315 if (!dm_suspended(md))
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001316 goto out;
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001317
1318 map = dm_get_table(md);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001319 if (!map || !dm_table_get_size(map))
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001320 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
1322 dm_table_resume_targets(map);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001323
1324 down_write(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 clear_bit(DMF_BLOCK_IO, &md->flags);
1326
1327 def = bio_list_get(&md->deferred);
1328 __flush_deferred_io(md, def);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001329 up_write(&md->io_lock);
1330
1331 unlock_fs(md);
1332
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08001333 bdput(md->suspended_bdev);
1334 md->suspended_bdev = NULL;
1335
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001336 clear_bit(DMF_SUSPENDED, &md->flags);
1337
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 dm_table_unplug_all(map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001340 r = 0;
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001341
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001342out:
1343 dm_table_put(map);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001344 up(&md->suspend_lock);
1345
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001346 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347}
1348
1349/*-----------------------------------------------------------------
1350 * Event notification.
1351 *---------------------------------------------------------------*/
1352uint32_t dm_get_event_nr(struct mapped_device *md)
1353{
1354 return atomic_read(&md->event_nr);
1355}
1356
1357int dm_wait_event(struct mapped_device *md, int event_nr)
1358{
1359 return wait_event_interruptible(md->eventq,
1360 (event_nr != atomic_read(&md->event_nr)));
1361}
1362
1363/*
1364 * The gendisk is only valid as long as you have a reference
1365 * count on 'md'.
1366 */
1367struct gendisk *dm_disk(struct mapped_device *md)
1368{
1369 return md->disk;
1370}
1371
1372int dm_suspended(struct mapped_device *md)
1373{
1374 return test_bit(DMF_SUSPENDED, &md->flags);
1375}
1376
1377static struct block_device_operations dm_blk_dops = {
1378 .open = dm_blk_open,
1379 .release = dm_blk_close,
Darrick J. Wong3ac51e72006-03-27 01:17:54 -08001380 .getgeo = dm_blk_getgeo,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 .owner = THIS_MODULE
1382};
1383
1384EXPORT_SYMBOL(dm_get_mapinfo);
1385
1386/*
1387 * module hooks
1388 */
1389module_init(dm_init);
1390module_exit(dm_exit);
1391
1392module_param(major, uint, 0);
1393MODULE_PARM_DESC(major, "The major number of the device mapper");
1394MODULE_DESCRIPTION(DM_NAME " driver");
1395MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1396MODULE_LICENSE("GPL");