2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/buffer_head.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <asm/div64.h>
25 #include "extent_map.h"
27 #include "transaction.h"
28 #include "print-tree.h"
39 struct btrfs_bio_stripe stripes[];
42 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
43 (sizeof(struct btrfs_bio_stripe) * (n)))
45 static DEFINE_MUTEX(uuid_mutex);
46 static LIST_HEAD(fs_uuids);
48 void btrfs_lock_volumes(void)
50 mutex_lock(&uuid_mutex);
53 void btrfs_unlock_volumes(void)
55 mutex_unlock(&uuid_mutex);
58 int btrfs_cleanup_fs_uuids(void)
60 struct btrfs_fs_devices *fs_devices;
61 struct list_head *uuid_cur;
62 struct list_head *devices_cur;
63 struct btrfs_device *dev;
65 list_for_each(uuid_cur, &fs_uuids) {
66 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
68 while(!list_empty(&fs_devices->devices)) {
69 devices_cur = fs_devices->devices.next;
70 dev = list_entry(devices_cur, struct btrfs_device,
73 close_bdev_excl(dev->bdev);
75 list_del(&dev->dev_list);
82 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
85 struct btrfs_device *dev;
86 struct list_head *cur;
88 list_for_each(cur, head) {
89 dev = list_entry(cur, struct btrfs_device, dev_list);
90 if (dev->devid == devid &&
91 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
98 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
100 struct list_head *cur;
101 struct btrfs_fs_devices *fs_devices;
103 list_for_each(cur, &fs_uuids) {
104 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
105 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
111 static int device_list_add(const char *path,
112 struct btrfs_super_block *disk_super,
113 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
115 struct btrfs_device *device;
116 struct btrfs_fs_devices *fs_devices;
117 u64 found_transid = btrfs_super_generation(disk_super);
119 fs_devices = find_fsid(disk_super->fsid);
121 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
124 INIT_LIST_HEAD(&fs_devices->devices);
125 INIT_LIST_HEAD(&fs_devices->alloc_list);
126 list_add(&fs_devices->list, &fs_uuids);
127 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
128 fs_devices->latest_devid = devid;
129 fs_devices->latest_trans = found_transid;
130 fs_devices->lowest_devid = (u64)-1;
131 fs_devices->num_devices = 0;
134 device = __find_device(&fs_devices->devices, devid,
135 disk_super->dev_item.uuid);
138 device = kzalloc(sizeof(*device), GFP_NOFS);
140 /* we can safely leave the fs_devices entry around */
143 device->devid = devid;
144 memcpy(device->uuid, disk_super->dev_item.uuid,
146 device->barriers = 1;
147 spin_lock_init(&device->io_lock);
148 device->name = kstrdup(path, GFP_NOFS);
153 list_add(&device->dev_list, &fs_devices->devices);
154 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
155 fs_devices->num_devices++;
158 if (found_transid > fs_devices->latest_trans) {
159 fs_devices->latest_devid = devid;
160 fs_devices->latest_trans = found_transid;
162 if (fs_devices->lowest_devid > devid) {
163 fs_devices->lowest_devid = devid;
165 *fs_devices_ret = fs_devices;
169 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
171 struct list_head *head = &fs_devices->devices;
172 struct list_head *cur;
173 struct btrfs_device *device;
175 mutex_lock(&uuid_mutex);
176 list_for_each(cur, head) {
177 device = list_entry(cur, struct btrfs_device, dev_list);
179 close_bdev_excl(device->bdev);
183 mutex_unlock(&uuid_mutex);
187 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
188 int flags, void *holder)
190 struct block_device *bdev;
191 struct list_head *head = &fs_devices->devices;
192 struct list_head *cur;
193 struct btrfs_device *device;
196 mutex_lock(&uuid_mutex);
197 list_for_each(cur, head) {
198 device = list_entry(cur, struct btrfs_device, dev_list);
199 bdev = open_bdev_excl(device->name, flags, holder);
202 printk("open %s failed\n", device->name);
206 set_blocksize(bdev, 4096);
207 if (device->devid == fs_devices->latest_devid)
208 fs_devices->latest_bdev = bdev;
209 if (device->devid == fs_devices->lowest_devid) {
210 fs_devices->lowest_bdev = bdev;
215 mutex_unlock(&uuid_mutex);
218 mutex_unlock(&uuid_mutex);
219 btrfs_close_devices(fs_devices);
223 int btrfs_scan_one_device(const char *path, int flags, void *holder,
224 struct btrfs_fs_devices **fs_devices_ret)
226 struct btrfs_super_block *disk_super;
227 struct block_device *bdev;
228 struct buffer_head *bh;
233 mutex_lock(&uuid_mutex);
235 bdev = open_bdev_excl(path, flags, holder);
242 ret = set_blocksize(bdev, 4096);
245 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
250 disk_super = (struct btrfs_super_block *)bh->b_data;
251 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
252 sizeof(disk_super->magic))) {
256 devid = le64_to_cpu(disk_super->dev_item.devid);
257 transid = btrfs_super_generation(disk_super);
258 if (disk_super->label[0])
259 printk("device label %s ", disk_super->label);
261 /* FIXME, make a readl uuid parser */
262 printk("device fsid %llx-%llx ",
263 *(unsigned long long *)disk_super->fsid,
264 *(unsigned long long *)(disk_super->fsid + 8));
266 printk("devid %Lu transid %Lu %s\n", devid, transid, path);
267 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
272 close_bdev_excl(bdev);
274 mutex_unlock(&uuid_mutex);
279 * this uses a pretty simple search, the expectation is that it is
280 * called very infrequently and that a given device has a small number
283 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
284 struct btrfs_device *device,
285 struct btrfs_path *path,
286 u64 num_bytes, u64 *start)
288 struct btrfs_key key;
289 struct btrfs_root *root = device->dev_root;
290 struct btrfs_dev_extent *dev_extent = NULL;
293 u64 search_start = 0;
294 u64 search_end = device->total_bytes;
298 struct extent_buffer *l;
303 /* FIXME use last free of some kind */
305 /* we don't want to overwrite the superblock on the drive,
306 * so we make sure to start at an offset of at least 1MB
308 search_start = max((u64)1024 * 1024, search_start);
310 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
311 search_start = max(root->fs_info->alloc_start, search_start);
313 key.objectid = device->devid;
314 key.offset = search_start;
315 key.type = BTRFS_DEV_EXTENT_KEY;
316 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
319 ret = btrfs_previous_item(root, path, 0, key.type);
323 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
326 slot = path->slots[0];
327 if (slot >= btrfs_header_nritems(l)) {
328 ret = btrfs_next_leaf(root, path);
335 if (search_start >= search_end) {
339 *start = search_start;
343 *start = last_byte > search_start ?
344 last_byte : search_start;
345 if (search_end <= *start) {
351 btrfs_item_key_to_cpu(l, &key, slot);
353 if (key.objectid < device->devid)
356 if (key.objectid > device->devid)
359 if (key.offset >= search_start && key.offset > last_byte &&
361 if (last_byte < search_start)
362 last_byte = search_start;
363 hole_size = key.offset - last_byte;
364 if (key.offset > last_byte &&
365 hole_size >= num_bytes) {
370 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
375 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
376 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
382 /* we have to make sure we didn't find an extent that has already
383 * been allocated by the map tree or the original allocation
385 btrfs_release_path(root, path);
386 BUG_ON(*start < search_start);
388 if (*start + num_bytes > search_end) {
392 /* check for pending inserts here */
396 btrfs_release_path(root, path);
400 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
401 struct btrfs_device *device,
405 struct btrfs_path *path;
406 struct btrfs_root *root = device->dev_root;
407 struct btrfs_key key;
408 struct btrfs_key found_key;
409 struct extent_buffer *leaf = NULL;
410 struct btrfs_dev_extent *extent = NULL;
412 path = btrfs_alloc_path();
416 key.objectid = device->devid;
418 key.type = BTRFS_DEV_EXTENT_KEY;
420 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
422 ret = btrfs_previous_item(root, path, key.objectid,
423 BTRFS_DEV_EXTENT_KEY);
425 leaf = path->nodes[0];
426 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
427 extent = btrfs_item_ptr(leaf, path->slots[0],
428 struct btrfs_dev_extent);
429 BUG_ON(found_key.offset > start || found_key.offset +
430 btrfs_dev_extent_length(leaf, extent) < start);
432 } else if (ret == 0) {
433 leaf = path->nodes[0];
434 extent = btrfs_item_ptr(leaf, path->slots[0],
435 struct btrfs_dev_extent);
439 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
440 ret = btrfs_del_item(trans, root, path);
443 btrfs_free_path(path);
447 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
448 struct btrfs_device *device,
449 u64 chunk_tree, u64 chunk_objectid,
451 u64 num_bytes, u64 *start)
454 struct btrfs_path *path;
455 struct btrfs_root *root = device->dev_root;
456 struct btrfs_dev_extent *extent;
457 struct extent_buffer *leaf;
458 struct btrfs_key key;
460 path = btrfs_alloc_path();
464 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
469 key.objectid = device->devid;
471 key.type = BTRFS_DEV_EXTENT_KEY;
472 ret = btrfs_insert_empty_item(trans, root, path, &key,
476 leaf = path->nodes[0];
477 extent = btrfs_item_ptr(leaf, path->slots[0],
478 struct btrfs_dev_extent);
479 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
480 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
481 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
483 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
484 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
487 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
488 btrfs_mark_buffer_dirty(leaf);
490 btrfs_free_path(path);
494 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
496 struct btrfs_path *path;
498 struct btrfs_key key;
499 struct btrfs_chunk *chunk;
500 struct btrfs_key found_key;
502 path = btrfs_alloc_path();
505 key.objectid = objectid;
506 key.offset = (u64)-1;
507 key.type = BTRFS_CHUNK_ITEM_KEY;
509 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
515 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
519 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
521 if (found_key.objectid != objectid)
524 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
526 *offset = found_key.offset +
527 btrfs_chunk_length(path->nodes[0], chunk);
532 btrfs_free_path(path);
536 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
540 struct btrfs_key key;
541 struct btrfs_key found_key;
543 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
544 key.type = BTRFS_DEV_ITEM_KEY;
545 key.offset = (u64)-1;
547 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
553 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
558 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
560 *objectid = found_key.offset + 1;
564 btrfs_release_path(root, path);
569 * the device information is stored in the chunk root
570 * the btrfs_device struct should be fully filled in
572 int btrfs_add_device(struct btrfs_trans_handle *trans,
573 struct btrfs_root *root,
574 struct btrfs_device *device)
577 struct btrfs_path *path;
578 struct btrfs_dev_item *dev_item;
579 struct extent_buffer *leaf;
580 struct btrfs_key key;
584 root = root->fs_info->chunk_root;
586 path = btrfs_alloc_path();
590 ret = find_next_devid(root, path, &free_devid);
594 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
595 key.type = BTRFS_DEV_ITEM_KEY;
596 key.offset = free_devid;
598 ret = btrfs_insert_empty_item(trans, root, path, &key,
603 leaf = path->nodes[0];
604 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
606 device->devid = free_devid;
607 btrfs_set_device_id(leaf, dev_item, device->devid);
608 btrfs_set_device_type(leaf, dev_item, device->type);
609 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
610 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
611 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
612 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
613 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
614 btrfs_set_device_group(leaf, dev_item, 0);
615 btrfs_set_device_seek_speed(leaf, dev_item, 0);
616 btrfs_set_device_bandwidth(leaf, dev_item, 0);
618 ptr = (unsigned long)btrfs_device_uuid(dev_item);
619 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
620 btrfs_mark_buffer_dirty(leaf);
624 btrfs_free_path(path);
628 static int btrfs_rm_dev_item(struct btrfs_root *root,
629 struct btrfs_device *device)
632 struct btrfs_path *path;
633 struct block_device *bdev = device->bdev;
634 struct btrfs_device *next_dev;
635 struct btrfs_key key;
637 struct btrfs_fs_devices *fs_devices;
638 struct btrfs_trans_handle *trans;
640 root = root->fs_info->chunk_root;
642 path = btrfs_alloc_path();
646 trans = btrfs_start_transaction(root, 1);
647 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
648 key.type = BTRFS_DEV_ITEM_KEY;
649 key.offset = device->devid;
651 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
660 ret = btrfs_del_item(trans, root, path);
665 * at this point, the device is zero sized. We want to
666 * remove it from the devices list and zero out the old super
668 list_del_init(&device->dev_list);
669 list_del_init(&device->dev_alloc_list);
670 fs_devices = root->fs_info->fs_devices;
672 next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
674 if (bdev == fs_devices->lowest_bdev)
675 fs_devices->lowest_bdev = next_dev->bdev;
676 if (bdev == root->fs_info->sb->s_bdev)
677 root->fs_info->sb->s_bdev = next_dev->bdev;
678 if (bdev == fs_devices->latest_bdev)
679 fs_devices->latest_bdev = next_dev->bdev;
681 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
682 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
683 total_bytes - device->total_bytes);
685 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
686 btrfs_set_super_num_devices(&root->fs_info->super_copy,
689 btrfs_free_path(path);
690 btrfs_commit_transaction(trans, root);
694 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
696 struct btrfs_device *device;
697 struct block_device *bdev;
698 struct buffer_head *bh;
699 struct btrfs_super_block *disk_super;
704 mutex_lock(&root->fs_info->fs_mutex);
705 mutex_lock(&uuid_mutex);
707 all_avail = root->fs_info->avail_data_alloc_bits |
708 root->fs_info->avail_system_alloc_bits |
709 root->fs_info->avail_metadata_alloc_bits;
711 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
712 root->fs_info->fs_devices->num_devices <= 4) {
713 printk("btrfs: unable to go below four devices on raid10\n");
718 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
719 root->fs_info->fs_devices->num_devices <= 2) {
720 printk("btrfs: unable to go below two devices on raid1\n");
725 bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
731 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
736 disk_super = (struct btrfs_super_block *)bh->b_data;
737 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
738 sizeof(disk_super->magic))) {
742 if (memcmp(disk_super->fsid, root->fs_info->fsid, BTRFS_FSID_SIZE)) {
746 devid = le64_to_cpu(disk_super->dev_item.devid);
747 device = btrfs_find_device(root, devid, NULL);
753 root->fs_info->fs_devices->num_devices--;
755 ret = btrfs_shrink_device(device, 0);
760 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
764 /* make sure this device isn't detected as part of the FS anymore */
765 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
766 set_buffer_dirty(bh);
767 sync_dirty_buffer(bh);
771 /* one close for the device struct or super_block */
772 close_bdev_excl(device->bdev);
774 /* one close for us */
775 close_bdev_excl(device->bdev);
785 close_bdev_excl(bdev);
787 mutex_unlock(&uuid_mutex);
788 mutex_unlock(&root->fs_info->fs_mutex);
792 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
794 struct btrfs_trans_handle *trans;
795 struct btrfs_device *device;
796 struct block_device *bdev;
797 struct list_head *cur;
798 struct list_head *devices;
803 bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
807 mutex_lock(&root->fs_info->fs_mutex);
808 trans = btrfs_start_transaction(root, 1);
809 devices = &root->fs_info->fs_devices->devices;
810 list_for_each(cur, devices) {
811 device = list_entry(cur, struct btrfs_device, dev_list);
812 if (device->bdev == bdev) {
818 device = kzalloc(sizeof(*device), GFP_NOFS);
820 /* we can safely leave the fs_devices entry around */
825 device->barriers = 1;
826 generate_random_uuid(device->uuid);
827 spin_lock_init(&device->io_lock);
828 device->name = kstrdup(device_path, GFP_NOFS);
833 device->io_width = root->sectorsize;
834 device->io_align = root->sectorsize;
835 device->sector_size = root->sectorsize;
836 device->total_bytes = i_size_read(bdev->bd_inode);
837 device->dev_root = root->fs_info->dev_root;
840 ret = btrfs_add_device(trans, root, device);
844 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
845 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
846 total_bytes + device->total_bytes);
848 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
849 btrfs_set_super_num_devices(&root->fs_info->super_copy,
852 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
853 list_add(&device->dev_alloc_list,
854 &root->fs_info->fs_devices->alloc_list);
855 root->fs_info->fs_devices->num_devices++;
857 btrfs_end_transaction(trans, root);
858 mutex_unlock(&root->fs_info->fs_mutex);
862 close_bdev_excl(bdev);
866 int btrfs_update_device(struct btrfs_trans_handle *trans,
867 struct btrfs_device *device)
870 struct btrfs_path *path;
871 struct btrfs_root *root;
872 struct btrfs_dev_item *dev_item;
873 struct extent_buffer *leaf;
874 struct btrfs_key key;
876 root = device->dev_root->fs_info->chunk_root;
878 path = btrfs_alloc_path();
882 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
883 key.type = BTRFS_DEV_ITEM_KEY;
884 key.offset = device->devid;
886 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
895 leaf = path->nodes[0];
896 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
898 btrfs_set_device_id(leaf, dev_item, device->devid);
899 btrfs_set_device_type(leaf, dev_item, device->type);
900 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
901 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
902 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
903 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
904 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
905 btrfs_mark_buffer_dirty(leaf);
908 btrfs_free_path(path);
912 int btrfs_grow_device(struct btrfs_trans_handle *trans,
913 struct btrfs_device *device, u64 new_size)
915 struct btrfs_super_block *super_copy =
916 &device->dev_root->fs_info->super_copy;
917 u64 old_total = btrfs_super_total_bytes(super_copy);
918 u64 diff = new_size - device->total_bytes;
920 btrfs_set_super_total_bytes(super_copy, old_total + diff);
921 return btrfs_update_device(trans, device);
924 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
925 struct btrfs_root *root,
926 u64 chunk_tree, u64 chunk_objectid,
930 struct btrfs_path *path;
931 struct btrfs_key key;
933 root = root->fs_info->chunk_root;
934 path = btrfs_alloc_path();
938 key.objectid = chunk_objectid;
939 key.offset = chunk_offset;
940 key.type = BTRFS_CHUNK_ITEM_KEY;
942 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
945 ret = btrfs_del_item(trans, root, path);
948 btrfs_free_path(path);
952 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
955 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
956 struct btrfs_disk_key *disk_key;
957 struct btrfs_chunk *chunk;
964 struct btrfs_key key;
966 array_size = btrfs_super_sys_array_size(super_copy);
968 ptr = super_copy->sys_chunk_array;
971 while (cur < array_size) {
972 disk_key = (struct btrfs_disk_key *)ptr;
973 btrfs_disk_key_to_cpu(&key, disk_key);
975 len = sizeof(*disk_key);
977 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
978 chunk = (struct btrfs_chunk *)(ptr + len);
979 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
980 len += btrfs_chunk_item_size(num_stripes);
985 if (key.objectid == chunk_objectid &&
986 key.offset == chunk_offset) {
987 memmove(ptr, ptr + len, array_size - (cur + len));
989 btrfs_set_super_sys_array_size(super_copy, array_size);
999 int btrfs_relocate_chunk(struct btrfs_root *root,
1000 u64 chunk_tree, u64 chunk_objectid,
1003 struct extent_map_tree *em_tree;
1004 struct btrfs_root *extent_root;
1005 struct btrfs_trans_handle *trans;
1006 struct extent_map *em;
1007 struct map_lookup *map;
1011 root = root->fs_info->chunk_root;
1012 extent_root = root->fs_info->extent_root;
1013 em_tree = &root->fs_info->mapping_tree.map_tree;
1015 /* step one, relocate all the extents inside this chunk */
1016 ret = btrfs_shrink_extent_tree(extent_root, chunk_offset);
1019 trans = btrfs_start_transaction(root, 1);
1023 * step two, delete the device extents and the
1024 * chunk tree entries
1026 spin_lock(&em_tree->lock);
1027 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1028 spin_unlock(&em_tree->lock);
1030 BUG_ON(em->start > chunk_offset ||
1031 em->start + em->len < chunk_offset);
1032 map = (struct map_lookup *)em->bdev;
1034 for (i = 0; i < map->num_stripes; i++) {
1035 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1036 map->stripes[i].physical);
1039 ret = btrfs_update_device(trans, map->stripes[i].dev);
1042 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1047 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1048 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1052 spin_lock(&em_tree->lock);
1053 remove_extent_mapping(em_tree, em);
1057 /* once for the tree */
1058 free_extent_map(em);
1059 spin_unlock(&em_tree->lock);
1062 free_extent_map(em);
1064 btrfs_end_transaction(trans, root);
1068 static u64 div_factor(u64 num, int factor)
1078 int btrfs_balance(struct btrfs_root *dev_root)
1081 struct list_head *cur;
1082 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1083 struct btrfs_device *device;
1086 struct btrfs_path *path;
1087 struct btrfs_key key;
1088 struct btrfs_chunk *chunk;
1089 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1090 struct btrfs_trans_handle *trans;
1091 struct btrfs_key found_key;
1094 dev_root = dev_root->fs_info->dev_root;
1096 mutex_lock(&dev_root->fs_info->fs_mutex);
1097 /* step one make some room on all the devices */
1098 list_for_each(cur, devices) {
1099 device = list_entry(cur, struct btrfs_device, dev_list);
1100 old_size = device->total_bytes;
1101 size_to_free = div_factor(old_size, 1);
1102 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1103 if (device->total_bytes - device->bytes_used > size_to_free)
1106 ret = btrfs_shrink_device(device, old_size - size_to_free);
1109 trans = btrfs_start_transaction(dev_root, 1);
1112 ret = btrfs_grow_device(trans, device, old_size);
1115 btrfs_end_transaction(trans, dev_root);
1118 /* step two, relocate all the chunks */
1119 path = btrfs_alloc_path();
1122 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1123 key.offset = (u64)-1;
1124 key.type = BTRFS_CHUNK_ITEM_KEY;
1127 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1132 * this shouldn't happen, it means the last relocate
1138 ret = btrfs_previous_item(chunk_root, path, 0,
1139 BTRFS_CHUNK_ITEM_KEY);
1143 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1145 if (found_key.objectid != key.objectid)
1147 chunk = btrfs_item_ptr(path->nodes[0],
1149 struct btrfs_chunk);
1150 key.offset = found_key.offset;
1151 /* chunk zero is special */
1152 if (key.offset == 0)
1155 ret = btrfs_relocate_chunk(chunk_root,
1156 chunk_root->root_key.objectid,
1160 btrfs_release_path(chunk_root, path);
1164 btrfs_free_path(path);
1165 mutex_unlock(&dev_root->fs_info->fs_mutex);
1170 * shrinking a device means finding all of the device extents past
1171 * the new size, and then following the back refs to the chunks.
1172 * The chunk relocation code actually frees the device extent
1174 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1176 struct btrfs_trans_handle *trans;
1177 struct btrfs_root *root = device->dev_root;
1178 struct btrfs_dev_extent *dev_extent = NULL;
1179 struct btrfs_path *path;
1186 struct extent_buffer *l;
1187 struct btrfs_key key;
1188 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1189 u64 old_total = btrfs_super_total_bytes(super_copy);
1190 u64 diff = device->total_bytes - new_size;
1193 path = btrfs_alloc_path();
1197 trans = btrfs_start_transaction(root, 1);
1205 device->total_bytes = new_size;
1206 ret = btrfs_update_device(trans, device);
1208 btrfs_end_transaction(trans, root);
1211 WARN_ON(diff > old_total);
1212 btrfs_set_super_total_bytes(super_copy, old_total - diff);
1213 btrfs_end_transaction(trans, root);
1215 key.objectid = device->devid;
1216 key.offset = (u64)-1;
1217 key.type = BTRFS_DEV_EXTENT_KEY;
1220 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1224 ret = btrfs_previous_item(root, path, 0, key.type);
1233 slot = path->slots[0];
1234 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1236 if (key.objectid != device->devid)
1239 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1240 length = btrfs_dev_extent_length(l, dev_extent);
1242 if (key.offset + length <= new_size)
1245 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1246 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1247 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1248 btrfs_release_path(root, path);
1250 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1257 btrfs_free_path(path);
1261 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root,
1263 struct btrfs_key *key,
1264 struct btrfs_chunk *chunk, int item_size)
1266 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1267 struct btrfs_disk_key disk_key;
1271 array_size = btrfs_super_sys_array_size(super_copy);
1272 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1275 ptr = super_copy->sys_chunk_array + array_size;
1276 btrfs_cpu_key_to_disk(&disk_key, key);
1277 memcpy(ptr, &disk_key, sizeof(disk_key));
1278 ptr += sizeof(disk_key);
1279 memcpy(ptr, chunk, item_size);
1280 item_size += sizeof(disk_key);
1281 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1285 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
1288 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1290 else if (type & BTRFS_BLOCK_GROUP_RAID10)
1291 return calc_size * (num_stripes / sub_stripes);
1293 return calc_size * num_stripes;
1297 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1298 struct btrfs_root *extent_root, u64 *start,
1299 u64 *num_bytes, u64 type)
1302 struct btrfs_fs_info *info = extent_root->fs_info;
1303 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1304 struct btrfs_path *path;
1305 struct btrfs_stripe *stripes;
1306 struct btrfs_device *device = NULL;
1307 struct btrfs_chunk *chunk;
1308 struct list_head private_devs;
1309 struct list_head *dev_list;
1310 struct list_head *cur;
1311 struct extent_map_tree *em_tree;
1312 struct map_lookup *map;
1313 struct extent_map *em;
1314 int min_stripe_size = 1 * 1024 * 1024;
1316 u64 calc_size = 1024 * 1024 * 1024;
1317 u64 max_chunk_size = calc_size;
1322 int num_stripes = 1;
1323 int min_stripes = 1;
1324 int sub_stripes = 0;
1328 int stripe_len = 64 * 1024;
1329 struct btrfs_key key;
1331 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1332 (type & BTRFS_BLOCK_GROUP_DUP)) {
1334 type &= ~BTRFS_BLOCK_GROUP_DUP;
1336 dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1337 if (list_empty(dev_list))
1340 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1341 num_stripes = btrfs_super_num_devices(&info->super_copy);
1344 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1348 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1349 num_stripes = min_t(u64, 2,
1350 btrfs_super_num_devices(&info->super_copy));
1351 if (num_stripes < 2)
1355 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1356 num_stripes = btrfs_super_num_devices(&info->super_copy);
1357 if (num_stripes < 4)
1359 num_stripes &= ~(u32)1;
1364 if (type & BTRFS_BLOCK_GROUP_DATA) {
1365 max_chunk_size = 10 * calc_size;
1366 min_stripe_size = 64 * 1024 * 1024;
1367 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1368 max_chunk_size = 4 * calc_size;
1369 min_stripe_size = 32 * 1024 * 1024;
1370 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1371 calc_size = 8 * 1024 * 1024;
1372 max_chunk_size = calc_size * 2;
1373 min_stripe_size = 1 * 1024 * 1024;
1376 path = btrfs_alloc_path();
1380 /* we don't want a chunk larger than 10% of the FS */
1381 percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1382 max_chunk_size = min(percent_max, max_chunk_size);
1385 if (calc_size * num_stripes > max_chunk_size) {
1386 calc_size = max_chunk_size;
1387 do_div(calc_size, num_stripes);
1388 do_div(calc_size, stripe_len);
1389 calc_size *= stripe_len;
1391 /* we don't want tiny stripes */
1392 calc_size = max_t(u64, min_stripe_size, calc_size);
1394 do_div(calc_size, stripe_len);
1395 calc_size *= stripe_len;
1397 INIT_LIST_HEAD(&private_devs);
1398 cur = dev_list->next;
1401 if (type & BTRFS_BLOCK_GROUP_DUP)
1402 min_free = calc_size * 2;
1404 min_free = calc_size;
1406 /* we add 1MB because we never use the first 1MB of the device */
1407 min_free += 1024 * 1024;
1409 /* build a private list of devices we will allocate from */
1410 while(index < num_stripes) {
1411 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1413 avail = device->total_bytes - device->bytes_used;
1416 if (avail >= min_free) {
1417 u64 ignored_start = 0;
1418 ret = find_free_dev_extent(trans, device, path,
1422 list_move_tail(&device->dev_alloc_list,
1425 if (type & BTRFS_BLOCK_GROUP_DUP)
1428 } else if (avail > max_avail)
1430 if (cur == dev_list)
1433 if (index < num_stripes) {
1434 list_splice(&private_devs, dev_list);
1435 if (index >= min_stripes) {
1436 num_stripes = index;
1437 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1438 num_stripes /= sub_stripes;
1439 num_stripes *= sub_stripes;
1444 if (!looped && max_avail > 0) {
1446 calc_size = max_avail;
1449 btrfs_free_path(path);
1452 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1453 key.type = BTRFS_CHUNK_ITEM_KEY;
1454 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1457 btrfs_free_path(path);
1461 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1463 btrfs_free_path(path);
1467 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1470 btrfs_free_path(path);
1473 btrfs_free_path(path);
1476 stripes = &chunk->stripe;
1477 *num_bytes = chunk_bytes_by_type(type, calc_size,
1478 num_stripes, sub_stripes);
1481 while(index < num_stripes) {
1482 struct btrfs_stripe *stripe;
1483 BUG_ON(list_empty(&private_devs));
1484 cur = private_devs.next;
1485 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1487 /* loop over this device again if we're doing a dup group */
1488 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1489 (index == num_stripes - 1))
1490 list_move_tail(&device->dev_alloc_list, dev_list);
1492 ret = btrfs_alloc_dev_extent(trans, device,
1493 info->chunk_root->root_key.objectid,
1494 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1495 calc_size, &dev_offset);
1497 device->bytes_used += calc_size;
1498 ret = btrfs_update_device(trans, device);
1501 map->stripes[index].dev = device;
1502 map->stripes[index].physical = dev_offset;
1503 stripe = stripes + index;
1504 btrfs_set_stack_stripe_devid(stripe, device->devid);
1505 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1506 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1507 physical = dev_offset;
1510 BUG_ON(!list_empty(&private_devs));
1512 /* key was set above */
1513 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1514 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1515 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1516 btrfs_set_stack_chunk_type(chunk, type);
1517 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1518 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1519 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1520 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1521 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1522 map->sector_size = extent_root->sectorsize;
1523 map->stripe_len = stripe_len;
1524 map->io_align = stripe_len;
1525 map->io_width = stripe_len;
1527 map->num_stripes = num_stripes;
1528 map->sub_stripes = sub_stripes;
1530 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1531 btrfs_chunk_item_size(num_stripes));
1533 *start = key.offset;;
1535 em = alloc_extent_map(GFP_NOFS);
1538 em->bdev = (struct block_device *)map;
1539 em->start = key.offset;
1540 em->len = *num_bytes;
1541 em->block_start = 0;
1543 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1544 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1545 chunk, btrfs_chunk_item_size(num_stripes));
1550 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1551 spin_lock(&em_tree->lock);
1552 ret = add_extent_mapping(em_tree, em);
1553 spin_unlock(&em_tree->lock);
1555 free_extent_map(em);
1559 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1561 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1564 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1566 struct extent_map *em;
1569 spin_lock(&tree->map_tree.lock);
1570 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1572 remove_extent_mapping(&tree->map_tree, em);
1573 spin_unlock(&tree->map_tree.lock);
1578 free_extent_map(em);
1579 /* once for the tree */
1580 free_extent_map(em);
1584 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1586 struct extent_map *em;
1587 struct map_lookup *map;
1588 struct extent_map_tree *em_tree = &map_tree->map_tree;
1591 spin_lock(&em_tree->lock);
1592 em = lookup_extent_mapping(em_tree, logical, len);
1593 spin_unlock(&em_tree->lock);
1596 BUG_ON(em->start > logical || em->start + em->len < logical);
1597 map = (struct map_lookup *)em->bdev;
1598 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1599 ret = map->num_stripes;
1600 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1601 ret = map->sub_stripes;
1604 free_extent_map(em);
1608 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1609 u64 logical, u64 *length,
1610 struct btrfs_multi_bio **multi_ret,
1611 int mirror_num, struct page *unplug_page)
1613 struct extent_map *em;
1614 struct map_lookup *map;
1615 struct extent_map_tree *em_tree = &map_tree->map_tree;
1619 int stripes_allocated = 8;
1620 int stripes_required = 1;
1625 struct btrfs_multi_bio *multi = NULL;
1627 if (multi_ret && !(rw & (1 << BIO_RW))) {
1628 stripes_allocated = 1;
1632 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1637 atomic_set(&multi->error, 0);
1640 spin_lock(&em_tree->lock);
1641 em = lookup_extent_mapping(em_tree, logical, *length);
1642 spin_unlock(&em_tree->lock);
1644 if (!em && unplug_page)
1648 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1652 BUG_ON(em->start > logical || em->start + em->len < logical);
1653 map = (struct map_lookup *)em->bdev;
1654 offset = logical - em->start;
1656 if (mirror_num > map->num_stripes)
1659 /* if our multi bio struct is too small, back off and try again */
1660 if (rw & (1 << BIO_RW)) {
1661 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1662 BTRFS_BLOCK_GROUP_DUP)) {
1663 stripes_required = map->num_stripes;
1665 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1666 stripes_required = map->sub_stripes;
1670 if (multi_ret && rw == WRITE &&
1671 stripes_allocated < stripes_required) {
1672 stripes_allocated = map->num_stripes;
1673 free_extent_map(em);
1679 * stripe_nr counts the total number of stripes we have to stride
1680 * to get to this block
1682 do_div(stripe_nr, map->stripe_len);
1684 stripe_offset = stripe_nr * map->stripe_len;
1685 BUG_ON(offset < stripe_offset);
1687 /* stripe_offset is the offset of this block in its stripe*/
1688 stripe_offset = offset - stripe_offset;
1690 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1691 BTRFS_BLOCK_GROUP_RAID10 |
1692 BTRFS_BLOCK_GROUP_DUP)) {
1693 /* we limit the length of each bio to what fits in a stripe */
1694 *length = min_t(u64, em->len - offset,
1695 map->stripe_len - stripe_offset);
1697 *length = em->len - offset;
1700 if (!multi_ret && !unplug_page)
1705 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1706 if (unplug_page || (rw & (1 << BIO_RW)))
1707 num_stripes = map->num_stripes;
1708 else if (mirror_num)
1709 stripe_index = mirror_num - 1;
1711 stripe_index = current->pid % map->num_stripes;
1713 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1714 if (rw & (1 << BIO_RW))
1715 num_stripes = map->num_stripes;
1716 else if (mirror_num)
1717 stripe_index = mirror_num - 1;
1719 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1720 int factor = map->num_stripes / map->sub_stripes;
1722 stripe_index = do_div(stripe_nr, factor);
1723 stripe_index *= map->sub_stripes;
1725 if (unplug_page || (rw & (1 << BIO_RW)))
1726 num_stripes = map->sub_stripes;
1727 else if (mirror_num)
1728 stripe_index += mirror_num - 1;
1730 stripe_index += current->pid % map->sub_stripes;
1733 * after this do_div call, stripe_nr is the number of stripes
1734 * on this device we have to walk to find the data, and
1735 * stripe_index is the number of our device in the stripe array
1737 stripe_index = do_div(stripe_nr, map->num_stripes);
1739 BUG_ON(stripe_index >= map->num_stripes);
1741 for (i = 0; i < num_stripes; i++) {
1743 struct btrfs_device *device;
1744 struct backing_dev_info *bdi;
1746 device = map->stripes[stripe_index].dev;
1747 bdi = blk_get_backing_dev_info(device->bdev);
1748 if (bdi->unplug_io_fn) {
1749 bdi->unplug_io_fn(bdi, unplug_page);
1752 multi->stripes[i].physical =
1753 map->stripes[stripe_index].physical +
1754 stripe_offset + stripe_nr * map->stripe_len;
1755 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1761 multi->num_stripes = num_stripes;
1762 multi->max_errors = max_errors;
1765 free_extent_map(em);
1769 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1770 u64 logical, u64 *length,
1771 struct btrfs_multi_bio **multi_ret, int mirror_num)
1773 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
1777 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
1778 u64 logical, struct page *page)
1780 u64 length = PAGE_CACHE_SIZE;
1781 return __btrfs_map_block(map_tree, READ, logical, &length,
1786 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1787 static void end_bio_multi_stripe(struct bio *bio, int err)
1789 static int end_bio_multi_stripe(struct bio *bio,
1790 unsigned int bytes_done, int err)
1793 struct btrfs_multi_bio *multi = bio->bi_private;
1795 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1800 atomic_inc(&multi->error);
1802 if (atomic_dec_and_test(&multi->stripes_pending)) {
1803 bio->bi_private = multi->private;
1804 bio->bi_end_io = multi->end_io;
1806 /* only send an error to the higher layers if it is
1807 * beyond the tolerance of the multi-bio
1809 if (atomic_read(&multi->error) > multi->max_errors)
1815 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1816 bio_endio(bio, bio->bi_size, err);
1818 bio_endio(bio, err);
1823 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1828 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
1831 struct btrfs_mapping_tree *map_tree;
1832 struct btrfs_device *dev;
1833 struct bio *first_bio = bio;
1834 u64 logical = bio->bi_sector << 9;
1837 struct btrfs_multi_bio *multi = NULL;
1842 length = bio->bi_size;
1843 map_tree = &root->fs_info->mapping_tree;
1844 map_length = length;
1846 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
1850 total_devs = multi->num_stripes;
1851 if (map_length < length) {
1852 printk("mapping failed logical %Lu bio len %Lu "
1853 "len %Lu\n", logical, length, map_length);
1856 multi->end_io = first_bio->bi_end_io;
1857 multi->private = first_bio->bi_private;
1858 atomic_set(&multi->stripes_pending, multi->num_stripes);
1860 while(dev_nr < total_devs) {
1861 if (total_devs > 1) {
1862 if (dev_nr < total_devs - 1) {
1863 bio = bio_clone(first_bio, GFP_NOFS);
1868 bio->bi_private = multi;
1869 bio->bi_end_io = end_bio_multi_stripe;
1871 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
1872 dev = multi->stripes[dev_nr].dev;
1874 bio->bi_bdev = dev->bdev;
1875 spin_lock(&dev->io_lock);
1877 spin_unlock(&dev->io_lock);
1878 submit_bio(rw, bio);
1881 if (total_devs == 1)
1886 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1889 struct list_head *head = &root->fs_info->fs_devices->devices;
1891 return __find_device(head, devid, uuid);
1894 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1895 struct extent_buffer *leaf,
1896 struct btrfs_chunk *chunk)
1898 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1899 struct map_lookup *map;
1900 struct extent_map *em;
1904 u8 uuid[BTRFS_UUID_SIZE];
1909 logical = key->offset;
1910 length = btrfs_chunk_length(leaf, chunk);
1912 spin_lock(&map_tree->map_tree.lock);
1913 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
1914 spin_unlock(&map_tree->map_tree.lock);
1916 /* already mapped? */
1917 if (em && em->start <= logical && em->start + em->len > logical) {
1918 free_extent_map(em);
1921 free_extent_map(em);
1924 map = kzalloc(sizeof(*map), GFP_NOFS);
1928 em = alloc_extent_map(GFP_NOFS);
1931 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1932 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1934 free_extent_map(em);
1938 em->bdev = (struct block_device *)map;
1939 em->start = logical;
1941 em->block_start = 0;
1943 map->num_stripes = num_stripes;
1944 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1945 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1946 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1947 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1948 map->type = btrfs_chunk_type(leaf, chunk);
1949 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1950 for (i = 0; i < num_stripes; i++) {
1951 map->stripes[i].physical =
1952 btrfs_stripe_offset_nr(leaf, chunk, i);
1953 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1954 read_extent_buffer(leaf, uuid, (unsigned long)
1955 btrfs_stripe_dev_uuid_nr(chunk, i),
1957 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
1958 if (!map->stripes[i].dev) {
1960 free_extent_map(em);
1965 spin_lock(&map_tree->map_tree.lock);
1966 ret = add_extent_mapping(&map_tree->map_tree, em);
1967 spin_unlock(&map_tree->map_tree.lock);
1969 free_extent_map(em);
1974 static int fill_device_from_item(struct extent_buffer *leaf,
1975 struct btrfs_dev_item *dev_item,
1976 struct btrfs_device *device)
1980 device->devid = btrfs_device_id(leaf, dev_item);
1981 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1982 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1983 device->type = btrfs_device_type(leaf, dev_item);
1984 device->io_align = btrfs_device_io_align(leaf, dev_item);
1985 device->io_width = btrfs_device_io_width(leaf, dev_item);
1986 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1988 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1989 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1994 static int read_one_dev(struct btrfs_root *root,
1995 struct extent_buffer *leaf,
1996 struct btrfs_dev_item *dev_item)
1998 struct btrfs_device *device;
2001 u8 dev_uuid[BTRFS_UUID_SIZE];
2003 devid = btrfs_device_id(leaf, dev_item);
2004 read_extent_buffer(leaf, dev_uuid,
2005 (unsigned long)btrfs_device_uuid(dev_item),
2007 device = btrfs_find_device(root, devid, dev_uuid);
2009 printk("warning devid %Lu not found already\n", devid);
2010 device = kzalloc(sizeof(*device), GFP_NOFS);
2013 list_add(&device->dev_list,
2014 &root->fs_info->fs_devices->devices);
2015 list_add(&device->dev_alloc_list,
2016 &root->fs_info->fs_devices->alloc_list);
2017 device->barriers = 1;
2018 spin_lock_init(&device->io_lock);
2021 fill_device_from_item(leaf, dev_item, device);
2022 device->dev_root = root->fs_info->dev_root;
2025 ret = btrfs_open_device(device);
2033 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2035 struct btrfs_dev_item *dev_item;
2037 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2039 return read_one_dev(root, buf, dev_item);
2042 int btrfs_read_sys_array(struct btrfs_root *root)
2044 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2045 struct extent_buffer *sb;
2046 struct btrfs_disk_key *disk_key;
2047 struct btrfs_chunk *chunk;
2049 unsigned long sb_ptr;
2055 struct btrfs_key key;
2057 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2058 BTRFS_SUPER_INFO_SIZE);
2061 btrfs_set_buffer_uptodate(sb);
2062 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2063 array_size = btrfs_super_sys_array_size(super_copy);
2065 ptr = super_copy->sys_chunk_array;
2066 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2069 while (cur < array_size) {
2070 disk_key = (struct btrfs_disk_key *)ptr;
2071 btrfs_disk_key_to_cpu(&key, disk_key);
2073 len = sizeof(*disk_key); ptr += len;
2077 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2078 chunk = (struct btrfs_chunk *)sb_ptr;
2079 ret = read_one_chunk(root, &key, sb, chunk);
2082 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2083 len = btrfs_chunk_item_size(num_stripes);
2092 free_extent_buffer(sb);
2096 int btrfs_read_chunk_tree(struct btrfs_root *root)
2098 struct btrfs_path *path;
2099 struct extent_buffer *leaf;
2100 struct btrfs_key key;
2101 struct btrfs_key found_key;
2105 root = root->fs_info->chunk_root;
2107 path = btrfs_alloc_path();
2111 /* first we search for all of the device items, and then we
2112 * read in all of the chunk items. This way we can create chunk
2113 * mappings that reference all of the devices that are afound
2115 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2119 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2121 leaf = path->nodes[0];
2122 slot = path->slots[0];
2123 if (slot >= btrfs_header_nritems(leaf)) {
2124 ret = btrfs_next_leaf(root, path);
2131 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2132 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2133 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2135 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2136 struct btrfs_dev_item *dev_item;
2137 dev_item = btrfs_item_ptr(leaf, slot,
2138 struct btrfs_dev_item);
2139 ret = read_one_dev(root, leaf, dev_item);
2142 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2143 struct btrfs_chunk *chunk;
2144 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2145 ret = read_one_chunk(root, &found_key, leaf, chunk);
2149 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2151 btrfs_release_path(root, path);
2155 btrfs_free_path(path);