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