2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
9 #include "dm-bio-list.h"
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>
21 static const char *_name = DM_NAME;
23 static unsigned int major = 0;
24 static unsigned int _major = 0;
27 * One of these is allocated per bio.
30 struct mapped_device *md;
37 * One of these is allocated per target within a bio. Hopefully
38 * this will be simplified out one day.
46 union map_info *dm_get_mapinfo(struct bio *bio)
48 if (bio && bio->bi_private)
49 return &((struct target_io *)bio->bi_private)->info;
54 * Bits for the md->flags field.
56 #define DMF_BLOCK_IO 0
57 #define DMF_SUSPENDED 1
58 #define DMF_FS_LOCKED 2
60 struct mapped_device {
61 struct rw_semaphore lock;
67 request_queue_t *queue;
73 * A list of ios that arrived while we were suspended.
76 wait_queue_head_t wait;
77 struct bio_list deferred;
80 * The current mapping.
85 * io objects are allocated from here.
94 wait_queue_head_t eventq;
97 * freeze/thaw support require holding onto a super block
99 struct super_block *frozen_sb;
100 struct block_device *frozen_bdev;
104 static kmem_cache_t *_io_cache;
105 static kmem_cache_t *_tio_cache;
107 static struct bio_set *dm_set;
109 static int __init local_init(void)
113 dm_set = bioset_create(16, 16, 4);
117 /* allocate a slab for the dm_ios */
118 _io_cache = kmem_cache_create("dm_io",
119 sizeof(struct dm_io), 0, 0, NULL, NULL);
123 /* allocate a slab for the target ios */
124 _tio_cache = kmem_cache_create("dm_tio", sizeof(struct target_io),
127 kmem_cache_destroy(_io_cache);
132 r = register_blkdev(_major, _name);
134 kmem_cache_destroy(_tio_cache);
135 kmem_cache_destroy(_io_cache);
145 static void local_exit(void)
147 kmem_cache_destroy(_tio_cache);
148 kmem_cache_destroy(_io_cache);
152 if (unregister_blkdev(_major, _name) < 0)
153 DMERR("devfs_unregister_blkdev failed");
157 DMINFO("cleaned up");
160 int (*_inits[])(void) __initdata = {
168 void (*_exits[])(void) = {
176 static int __init dm_init(void)
178 const int count = ARRAY_SIZE(_inits);
182 for (i = 0; i < count; i++) {
197 static void __exit dm_exit(void)
199 int i = ARRAY_SIZE(_exits);
206 * Block device functions
208 static int dm_blk_open(struct inode *inode, struct file *file)
210 struct mapped_device *md;
212 md = inode->i_bdev->bd_disk->private_data;
217 static int dm_blk_close(struct inode *inode, struct file *file)
219 struct mapped_device *md;
221 md = inode->i_bdev->bd_disk->private_data;
226 static inline struct dm_io *alloc_io(struct mapped_device *md)
228 return mempool_alloc(md->io_pool, GFP_NOIO);
231 static inline void free_io(struct mapped_device *md, struct dm_io *io)
233 mempool_free(io, md->io_pool);
236 static inline struct target_io *alloc_tio(struct mapped_device *md)
238 return mempool_alloc(md->tio_pool, GFP_NOIO);
241 static inline void free_tio(struct mapped_device *md, struct target_io *tio)
243 mempool_free(tio, md->tio_pool);
247 * Add the bio to the list of deferred io.
249 static int queue_io(struct mapped_device *md, struct bio *bio)
251 down_write(&md->lock);
253 if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
258 bio_list_add(&md->deferred, bio);
261 return 0; /* deferred successfully */
265 * Everyone (including functions in this file), should use this
266 * function to access the md->map field, and make sure they call
267 * dm_table_put() when finished.
269 struct dm_table *dm_get_table(struct mapped_device *md)
273 read_lock(&md->map_lock);
277 read_unlock(&md->map_lock);
282 /*-----------------------------------------------------------------
284 * A more elegant soln is in the works that uses the queue
285 * merge fn, unfortunately there are a couple of changes to
286 * the block layer that I want to make for this. So in the
287 * interests of getting something for people to use I give
288 * you this clearly demarcated crap.
289 *---------------------------------------------------------------*/
292 * Decrements the number of outstanding ios that a bio has been
293 * cloned into, completing the original io if necc.
295 static inline void dec_pending(struct dm_io *io, int error)
300 if (atomic_dec_and_test(&io->io_count)) {
301 if (atomic_dec_and_test(&io->md->pending))
302 /* nudge anyone waiting on suspend queue */
303 wake_up(&io->md->wait);
305 bio_endio(io->bio, io->bio->bi_size, io->error);
310 static int clone_endio(struct bio *bio, unsigned int done, int error)
313 struct target_io *tio = bio->bi_private;
314 struct dm_io *io = tio->io;
315 dm_endio_fn endio = tio->ti->type->end_io;
320 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
324 r = endio(tio->ti, bio, error, &tio->info);
329 /* the target wants another shot at the io */
333 free_tio(io->md, tio);
334 dec_pending(io, error);
339 static sector_t max_io_len(struct mapped_device *md,
340 sector_t sector, struct dm_target *ti)
342 sector_t offset = sector - ti->begin;
343 sector_t len = ti->len - offset;
346 * Does the target need to split even further ?
350 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
359 static void __map_bio(struct dm_target *ti, struct bio *clone,
360 struct target_io *tio)
367 BUG_ON(!clone->bi_size);
369 clone->bi_end_io = clone_endio;
370 clone->bi_private = tio;
373 * Map the clone. If r == 0 we don't need to do
374 * anything, the target has assumed ownership of
377 atomic_inc(&tio->io->io_count);
378 r = ti->type->map(ti, clone, &tio->info);
380 /* the bio has been remapped so dispatch it */
381 generic_make_request(clone);
384 /* error the io and bail out */
385 struct dm_io *io = tio->io;
386 free_tio(tio->io->md, tio);
393 struct mapped_device *md;
394 struct dm_table *map;
398 sector_t sector_count;
403 * Creates a little bio that is just does part of a bvec.
405 static struct bio *split_bvec(struct bio *bio, sector_t sector,
406 unsigned short idx, unsigned int offset,
410 struct bio_vec *bv = bio->bi_io_vec + idx;
412 clone = bio_alloc_bioset(GFP_NOIO, 1, dm_set);
413 *clone->bi_io_vec = *bv;
415 clone->bi_sector = sector;
416 clone->bi_bdev = bio->bi_bdev;
417 clone->bi_rw = bio->bi_rw;
419 clone->bi_size = to_bytes(len);
420 clone->bi_io_vec->bv_offset = offset;
421 clone->bi_io_vec->bv_len = clone->bi_size;
427 * Creates a bio that consists of range of complete bvecs.
429 static struct bio *clone_bio(struct bio *bio, sector_t sector,
430 unsigned short idx, unsigned short bv_count,
435 clone = bio_clone(bio, GFP_NOIO);
436 clone->bi_sector = sector;
438 clone->bi_vcnt = idx + bv_count;
439 clone->bi_size = to_bytes(len);
440 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
445 static void __clone_and_map(struct clone_info *ci)
447 struct bio *clone, *bio = ci->bio;
448 struct dm_target *ti = dm_table_find_target(ci->map, ci->sector);
449 sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti);
450 struct target_io *tio;
453 * Allocate a target io object.
455 tio = alloc_tio(ci->md);
458 memset(&tio->info, 0, sizeof(tio->info));
460 if (ci->sector_count <= max) {
462 * Optimise for the simple case where we can do all of
463 * the remaining io with a single clone.
465 clone = clone_bio(bio, ci->sector, ci->idx,
466 bio->bi_vcnt - ci->idx, ci->sector_count);
467 __map_bio(ti, clone, tio);
468 ci->sector_count = 0;
470 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
472 * There are some bvecs that don't span targets.
473 * Do as many of these as possible.
476 sector_t remaining = max;
479 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
480 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
482 if (bv_len > remaining)
489 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len);
490 __map_bio(ti, clone, tio);
493 ci->sector_count -= len;
498 * Create two copy bios to deal with io that has
499 * been split across a target.
501 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
503 clone = split_bvec(bio, ci->sector, ci->idx,
505 __map_bio(ti, clone, tio);
508 ci->sector_count -= max;
509 ti = dm_table_find_target(ci->map, ci->sector);
511 len = to_sector(bv->bv_len) - max;
512 clone = split_bvec(bio, ci->sector, ci->idx,
513 bv->bv_offset + to_bytes(max), len);
514 tio = alloc_tio(ci->md);
517 memset(&tio->info, 0, sizeof(tio->info));
518 __map_bio(ti, clone, tio);
521 ci->sector_count -= len;
527 * Split the bio into several clones.
529 static void __split_bio(struct mapped_device *md, struct bio *bio)
531 struct clone_info ci;
533 ci.map = dm_get_table(md);
535 bio_io_error(bio, bio->bi_size);
541 ci.io = alloc_io(md);
543 atomic_set(&ci.io->io_count, 1);
546 ci.sector = bio->bi_sector;
547 ci.sector_count = bio_sectors(bio);
548 ci.idx = bio->bi_idx;
550 atomic_inc(&md->pending);
551 while (ci.sector_count)
552 __clone_and_map(&ci);
554 /* drop the extra reference count */
555 dec_pending(ci.io, 0);
556 dm_table_put(ci.map);
558 /*-----------------------------------------------------------------
560 *---------------------------------------------------------------*/
563 * The request function that just remaps the bio built up by
566 static int dm_request(request_queue_t *q, struct bio *bio)
569 struct mapped_device *md = q->queuedata;
571 down_read(&md->lock);
574 * If we're suspended we have to queue
577 while (test_bit(DMF_BLOCK_IO, &md->flags)) {
580 if (bio_rw(bio) == READA) {
581 bio_io_error(bio, bio->bi_size);
585 r = queue_io(md, bio);
587 bio_io_error(bio, bio->bi_size);
591 return 0; /* deferred successfully */
594 * We're in a while loop, because someone could suspend
595 * before we get to the following read lock.
597 down_read(&md->lock);
600 __split_bio(md, bio);
605 static int dm_flush_all(request_queue_t *q, struct gendisk *disk,
606 sector_t *error_sector)
608 struct mapped_device *md = q->queuedata;
609 struct dm_table *map = dm_get_table(md);
613 ret = dm_table_flush_all(md->map);
620 static void dm_unplug_all(request_queue_t *q)
622 struct mapped_device *md = q->queuedata;
623 struct dm_table *map = dm_get_table(md);
626 dm_table_unplug_all(map);
631 static int dm_any_congested(void *congested_data, int bdi_bits)
634 struct mapped_device *md = (struct mapped_device *) congested_data;
635 struct dm_table *map = dm_get_table(md);
637 if (!map || test_bit(DMF_BLOCK_IO, &md->flags))
640 r = dm_table_any_congested(map, bdi_bits);
646 /*-----------------------------------------------------------------
647 * An IDR is used to keep track of allocated minor numbers.
648 *---------------------------------------------------------------*/
649 static DECLARE_MUTEX(_minor_lock);
650 static DEFINE_IDR(_minor_idr);
652 static void free_minor(unsigned int minor)
655 idr_remove(&_minor_idr, minor);
660 * See if the device with a specific minor # is free.
662 static int specific_minor(struct mapped_device *md, unsigned int minor)
666 if (minor >= (1 << MINORBITS))
671 if (idr_find(&_minor_idr, minor)) {
676 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
682 r = idr_get_new_above(&_minor_idr, md, minor, &m);
688 idr_remove(&_minor_idr, m);
698 static int next_free_minor(struct mapped_device *md, unsigned int *minor)
705 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
711 r = idr_get_new(&_minor_idr, md, &m);
716 if (m >= (1 << MINORBITS)) {
717 idr_remove(&_minor_idr, m);
729 static struct block_device_operations dm_blk_dops;
732 * Allocate and initialise a blank device with a given minor.
734 static struct mapped_device *alloc_dev(unsigned int minor, int persistent)
737 struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL);
740 DMWARN("unable to allocate device, out of memory.");
744 /* get a minor number for the dev */
745 r = persistent ? specific_minor(md, minor) : next_free_minor(md, &minor);
749 memset(md, 0, sizeof(*md));
750 init_rwsem(&md->lock);
751 rwlock_init(&md->map_lock);
752 atomic_set(&md->holders, 1);
753 atomic_set(&md->event_nr, 0);
755 md->queue = blk_alloc_queue(GFP_KERNEL);
759 md->queue->queuedata = md;
760 md->queue->backing_dev_info.congested_fn = dm_any_congested;
761 md->queue->backing_dev_info.congested_data = md;
762 blk_queue_make_request(md->queue, dm_request);
763 md->queue->unplug_fn = dm_unplug_all;
764 md->queue->issue_flush_fn = dm_flush_all;
766 md->io_pool = mempool_create(MIN_IOS, mempool_alloc_slab,
767 mempool_free_slab, _io_cache);
771 md->tio_pool = mempool_create(MIN_IOS, mempool_alloc_slab,
772 mempool_free_slab, _tio_cache);
776 md->disk = alloc_disk(1);
780 md->disk->major = _major;
781 md->disk->first_minor = minor;
782 md->disk->fops = &dm_blk_dops;
783 md->disk->queue = md->queue;
784 md->disk->private_data = md;
785 sprintf(md->disk->disk_name, "dm-%d", minor);
788 atomic_set(&md->pending, 0);
789 init_waitqueue_head(&md->wait);
790 init_waitqueue_head(&md->eventq);
795 mempool_destroy(md->tio_pool);
797 mempool_destroy(md->io_pool);
799 blk_put_queue(md->queue);
806 static void free_dev(struct mapped_device *md)
808 free_minor(md->disk->first_minor);
809 mempool_destroy(md->tio_pool);
810 mempool_destroy(md->io_pool);
811 del_gendisk(md->disk);
813 blk_put_queue(md->queue);
818 * Bind a table to the device.
820 static void event_callback(void *context)
822 struct mapped_device *md = (struct mapped_device *) context;
824 atomic_inc(&md->event_nr);
825 wake_up(&md->eventq);
828 static void __set_size(struct gendisk *disk, sector_t size)
830 struct block_device *bdev;
832 set_capacity(disk, size);
833 bdev = bdget_disk(disk, 0);
835 down(&bdev->bd_inode->i_sem);
836 i_size_write(bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
837 up(&bdev->bd_inode->i_sem);
842 static int __bind(struct mapped_device *md, struct dm_table *t)
844 request_queue_t *q = md->queue;
847 size = dm_table_get_size(t);
848 __set_size(md->disk, size);
852 write_lock(&md->map_lock);
854 write_unlock(&md->map_lock);
857 dm_table_event_callback(md->map, event_callback, md);
858 dm_table_set_restrictions(t, q);
862 static void __unbind(struct mapped_device *md)
864 struct dm_table *map = md->map;
869 dm_table_event_callback(map, NULL, NULL);
870 write_lock(&md->map_lock);
872 write_unlock(&md->map_lock);
877 * Constructor for a new device.
879 static int create_aux(unsigned int minor, int persistent,
880 struct mapped_device **result)
882 struct mapped_device *md;
884 md = alloc_dev(minor, persistent);
892 int dm_create(struct mapped_device **result)
894 return create_aux(0, 0, result);
897 int dm_create_with_minor(unsigned int minor, struct mapped_device **result)
899 return create_aux(minor, 1, result);
902 void *dm_get_mdptr(dev_t dev)
904 struct mapped_device *md;
906 unsigned minor = MINOR(dev);
908 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
913 md = idr_find(&_minor_idr, minor);
915 if (md && (dm_disk(md)->first_minor == minor))
916 mdptr = md->interface_ptr;
923 void dm_set_mdptr(struct mapped_device *md, void *ptr)
925 md->interface_ptr = ptr;
928 void dm_get(struct mapped_device *md)
930 atomic_inc(&md->holders);
933 void dm_put(struct mapped_device *md)
935 struct dm_table *map = dm_get_table(md);
937 if (atomic_dec_and_test(&md->holders)) {
938 if (!test_bit(DMF_SUSPENDED, &md->flags) && map) {
939 dm_table_presuspend_targets(map);
940 dm_table_postsuspend_targets(map);
950 * Process the deferred bios
952 static void __flush_deferred_io(struct mapped_device *md, struct bio *c)
965 * Swap in a new table (destroying old one).
967 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
971 down_write(&md->lock);
973 /* device must be suspended */
974 if (!test_bit(DMF_SUSPENDED, &md->flags))
978 r = __bind(md, table);
986 * Functions to lock and unlock any filesystem running on the
989 static int __lock_fs(struct mapped_device *md)
993 if (test_and_set_bit(DMF_FS_LOCKED, &md->flags))
996 md->frozen_bdev = bdget_disk(md->disk, 0);
997 if (!md->frozen_bdev) {
998 DMWARN("bdget failed in __lock_fs");
1002 WARN_ON(md->frozen_sb);
1004 md->frozen_sb = freeze_bdev(md->frozen_bdev);
1005 if (IS_ERR(md->frozen_sb)) {
1006 error = PTR_ERR(md->frozen_sb);
1010 /* don't bdput right now, we don't want the bdev
1011 * to go away while it is locked. We'll bdput
1017 bdput(md->frozen_bdev);
1018 md->frozen_sb = NULL;
1019 md->frozen_bdev = NULL;
1021 clear_bit(DMF_FS_LOCKED, &md->flags);
1025 static void __unlock_fs(struct mapped_device *md)
1027 if (!test_and_clear_bit(DMF_FS_LOCKED, &md->flags))
1030 thaw_bdev(md->frozen_bdev, md->frozen_sb);
1031 bdput(md->frozen_bdev);
1033 md->frozen_sb = NULL;
1034 md->frozen_bdev = NULL;
1038 * We need to be able to change a mapping table under a mounted
1039 * filesystem. For example we might want to move some data in
1040 * the background. Before the table can be swapped with
1041 * dm_bind_table, dm_suspend must be called to flush any in
1042 * flight bios and ensure that any further io gets deferred.
1044 int dm_suspend(struct mapped_device *md)
1046 struct dm_table *map;
1047 DECLARE_WAITQUEUE(wait, current);
1048 int error = -EINVAL;
1050 /* Flush I/O to the device. */
1051 down_read(&md->lock);
1052 if (test_bit(DMF_BLOCK_IO, &md->flags))
1053 goto out_read_unlock;
1055 map = dm_get_table(md);
1057 /* This does not get reverted if there's an error later. */
1058 dm_table_presuspend_targets(map);
1060 error = __lock_fs(md);
1063 goto out_read_unlock;
1069 * First we set the BLOCK_IO flag so no more ios will be mapped.
1071 * If the flag is already set we know another thread is trying to
1072 * suspend as well, so we leave the fs locked for this thread.
1075 down_write(&md->lock);
1076 if (test_and_set_bit(DMF_BLOCK_IO, &md->flags)) {
1079 goto out_write_unlock;
1082 add_wait_queue(&md->wait, &wait);
1083 up_write(&md->lock);
1087 dm_table_unplug_all(map);
1092 * Then we wait for the already mapped ios to
1096 set_current_state(TASK_INTERRUPTIBLE);
1098 if (!atomic_read(&md->pending) || signal_pending(current))
1103 set_current_state(TASK_RUNNING);
1105 down_write(&md->lock);
1106 remove_wait_queue(&md->wait, &wait);
1108 /* were we interrupted ? */
1110 if (atomic_read(&md->pending))
1113 set_bit(DMF_SUSPENDED, &md->flags);
1115 map = dm_get_table(md);
1117 dm_table_postsuspend_targets(map);
1119 up_write(&md->lock);
1125 clear_bit(DMF_BLOCK_IO, &md->flags);
1127 up_write(&md->lock);
1135 int dm_resume(struct mapped_device *md)
1138 struct dm_table *map = dm_get_table(md);
1140 down_write(&md->lock);
1142 !test_bit(DMF_SUSPENDED, &md->flags) ||
1143 !dm_table_get_size(map)) {
1144 up_write(&md->lock);
1149 dm_table_resume_targets(map);
1150 clear_bit(DMF_SUSPENDED, &md->flags);
1151 clear_bit(DMF_BLOCK_IO, &md->flags);
1153 def = bio_list_get(&md->deferred);
1154 __flush_deferred_io(md, def);
1155 up_write(&md->lock);
1157 dm_table_unplug_all(map);
1163 /*-----------------------------------------------------------------
1164 * Event notification.
1165 *---------------------------------------------------------------*/
1166 uint32_t dm_get_event_nr(struct mapped_device *md)
1168 return atomic_read(&md->event_nr);
1171 int dm_wait_event(struct mapped_device *md, int event_nr)
1173 return wait_event_interruptible(md->eventq,
1174 (event_nr != atomic_read(&md->event_nr)));
1178 * The gendisk is only valid as long as you have a reference
1181 struct gendisk *dm_disk(struct mapped_device *md)
1186 int dm_suspended(struct mapped_device *md)
1188 return test_bit(DMF_SUSPENDED, &md->flags);
1191 static struct block_device_operations dm_blk_dops = {
1192 .open = dm_blk_open,
1193 .release = dm_blk_close,
1194 .owner = THIS_MODULE
1197 EXPORT_SYMBOL(dm_get_mapinfo);
1202 module_init(dm_init);
1203 module_exit(dm_exit);
1205 module_param(major, uint, 0);
1206 MODULE_PARM_DESC(major, "The major number of the device mapper");
1207 MODULE_DESCRIPTION(DM_NAME " driver");
1208 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1209 MODULE_LICENSE("GPL");