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);
74 fs_devices->open_devices--;
76 list_del(&dev->dev_list);
84 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
87 struct btrfs_device *dev;
88 struct list_head *cur;
90 list_for_each(cur, head) {
91 dev = list_entry(cur, struct btrfs_device, dev_list);
92 if (dev->devid == devid &&
93 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
100 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
102 struct list_head *cur;
103 struct btrfs_fs_devices *fs_devices;
105 list_for_each(cur, &fs_uuids) {
106 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
107 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
113 static int device_list_add(const char *path,
114 struct btrfs_super_block *disk_super,
115 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
117 struct btrfs_device *device;
118 struct btrfs_fs_devices *fs_devices;
119 u64 found_transid = btrfs_super_generation(disk_super);
121 fs_devices = find_fsid(disk_super->fsid);
123 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
126 INIT_LIST_HEAD(&fs_devices->devices);
127 INIT_LIST_HEAD(&fs_devices->alloc_list);
128 list_add(&fs_devices->list, &fs_uuids);
129 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
130 fs_devices->latest_devid = devid;
131 fs_devices->latest_trans = found_transid;
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 *fs_devices_ret = fs_devices;
166 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
168 struct list_head *head = &fs_devices->devices;
169 struct list_head *cur;
170 struct btrfs_device *device;
172 mutex_lock(&uuid_mutex);
174 list_for_each(cur, head) {
175 device = list_entry(cur, struct btrfs_device, dev_list);
176 if (!device->in_fs_metadata) {
178 close_bdev_excl(device->bdev);
179 fs_devices->open_devices--;
181 list_del(&device->dev_list);
182 list_del(&device->dev_alloc_list);
183 fs_devices->num_devices--;
189 mutex_unlock(&uuid_mutex);
193 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
195 struct list_head *head = &fs_devices->devices;
196 struct list_head *cur;
197 struct btrfs_device *device;
199 mutex_lock(&uuid_mutex);
200 list_for_each(cur, head) {
201 device = list_entry(cur, struct btrfs_device, dev_list);
203 close_bdev_excl(device->bdev);
204 fs_devices->open_devices--;
207 device->in_fs_metadata = 0;
209 fs_devices->mounted = 0;
210 mutex_unlock(&uuid_mutex);
214 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
215 int flags, void *holder)
217 struct block_device *bdev;
218 struct list_head *head = &fs_devices->devices;
219 struct list_head *cur;
220 struct btrfs_device *device;
221 struct block_device *latest_bdev = NULL;
222 struct buffer_head *bh;
223 struct btrfs_super_block *disk_super;
224 u64 latest_devid = 0;
225 u64 latest_transid = 0;
230 mutex_lock(&uuid_mutex);
231 if (fs_devices->mounted)
234 list_for_each(cur, head) {
235 device = list_entry(cur, struct btrfs_device, dev_list);
242 bdev = open_bdev_excl(device->name, flags, holder);
245 printk("open %s failed\n", device->name);
248 set_blocksize(bdev, 4096);
250 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
254 disk_super = (struct btrfs_super_block *)bh->b_data;
255 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
256 sizeof(disk_super->magic)))
259 devid = le64_to_cpu(disk_super->dev_item.devid);
260 if (devid != device->devid)
263 transid = btrfs_super_generation(disk_super);
264 if (!latest_transid || transid > latest_transid) {
265 latest_devid = devid;
266 latest_transid = transid;
271 device->in_fs_metadata = 0;
272 fs_devices->open_devices++;
278 close_bdev_excl(bdev);
282 if (fs_devices->open_devices == 0) {
286 fs_devices->mounted = 1;
287 fs_devices->latest_bdev = latest_bdev;
288 fs_devices->latest_devid = latest_devid;
289 fs_devices->latest_trans = latest_transid;
291 mutex_unlock(&uuid_mutex);
295 int btrfs_scan_one_device(const char *path, int flags, void *holder,
296 struct btrfs_fs_devices **fs_devices_ret)
298 struct btrfs_super_block *disk_super;
299 struct block_device *bdev;
300 struct buffer_head *bh;
305 mutex_lock(&uuid_mutex);
307 bdev = open_bdev_excl(path, flags, holder);
314 ret = set_blocksize(bdev, 4096);
317 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
322 disk_super = (struct btrfs_super_block *)bh->b_data;
323 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
324 sizeof(disk_super->magic))) {
328 devid = le64_to_cpu(disk_super->dev_item.devid);
329 transid = btrfs_super_generation(disk_super);
330 if (disk_super->label[0])
331 printk("device label %s ", disk_super->label);
333 /* FIXME, make a readl uuid parser */
334 printk("device fsid %llx-%llx ",
335 *(unsigned long long *)disk_super->fsid,
336 *(unsigned long long *)(disk_super->fsid + 8));
338 printk("devid %Lu transid %Lu %s\n", devid, transid, path);
339 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
344 close_bdev_excl(bdev);
346 mutex_unlock(&uuid_mutex);
351 * this uses a pretty simple search, the expectation is that it is
352 * called very infrequently and that a given device has a small number
355 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
356 struct btrfs_device *device,
357 struct btrfs_path *path,
358 u64 num_bytes, u64 *start)
360 struct btrfs_key key;
361 struct btrfs_root *root = device->dev_root;
362 struct btrfs_dev_extent *dev_extent = NULL;
365 u64 search_start = 0;
366 u64 search_end = device->total_bytes;
370 struct extent_buffer *l;
375 /* FIXME use last free of some kind */
377 /* we don't want to overwrite the superblock on the drive,
378 * so we make sure to start at an offset of at least 1MB
380 search_start = max((u64)1024 * 1024, search_start);
382 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
383 search_start = max(root->fs_info->alloc_start, search_start);
385 key.objectid = device->devid;
386 key.offset = search_start;
387 key.type = BTRFS_DEV_EXTENT_KEY;
388 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
391 ret = btrfs_previous_item(root, path, 0, key.type);
395 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
398 slot = path->slots[0];
399 if (slot >= btrfs_header_nritems(l)) {
400 ret = btrfs_next_leaf(root, path);
407 if (search_start >= search_end) {
411 *start = search_start;
415 *start = last_byte > search_start ?
416 last_byte : search_start;
417 if (search_end <= *start) {
423 btrfs_item_key_to_cpu(l, &key, slot);
425 if (key.objectid < device->devid)
428 if (key.objectid > device->devid)
431 if (key.offset >= search_start && key.offset > last_byte &&
433 if (last_byte < search_start)
434 last_byte = search_start;
435 hole_size = key.offset - last_byte;
436 if (key.offset > last_byte &&
437 hole_size >= num_bytes) {
442 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
447 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
448 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
454 /* we have to make sure we didn't find an extent that has already
455 * been allocated by the map tree or the original allocation
457 btrfs_release_path(root, path);
458 BUG_ON(*start < search_start);
460 if (*start + num_bytes > search_end) {
464 /* check for pending inserts here */
468 btrfs_release_path(root, path);
472 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
473 struct btrfs_device *device,
477 struct btrfs_path *path;
478 struct btrfs_root *root = device->dev_root;
479 struct btrfs_key key;
480 struct btrfs_key found_key;
481 struct extent_buffer *leaf = NULL;
482 struct btrfs_dev_extent *extent = NULL;
484 path = btrfs_alloc_path();
488 key.objectid = device->devid;
490 key.type = BTRFS_DEV_EXTENT_KEY;
492 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
494 ret = btrfs_previous_item(root, path, key.objectid,
495 BTRFS_DEV_EXTENT_KEY);
497 leaf = path->nodes[0];
498 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
499 extent = btrfs_item_ptr(leaf, path->slots[0],
500 struct btrfs_dev_extent);
501 BUG_ON(found_key.offset > start || found_key.offset +
502 btrfs_dev_extent_length(leaf, extent) < start);
504 } else if (ret == 0) {
505 leaf = path->nodes[0];
506 extent = btrfs_item_ptr(leaf, path->slots[0],
507 struct btrfs_dev_extent);
511 if (device->bytes_used > 0)
512 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
513 ret = btrfs_del_item(trans, root, path);
516 btrfs_free_path(path);
520 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
521 struct btrfs_device *device,
522 u64 chunk_tree, u64 chunk_objectid,
524 u64 num_bytes, u64 *start)
527 struct btrfs_path *path;
528 struct btrfs_root *root = device->dev_root;
529 struct btrfs_dev_extent *extent;
530 struct extent_buffer *leaf;
531 struct btrfs_key key;
533 WARN_ON(!device->in_fs_metadata);
534 path = btrfs_alloc_path();
538 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
543 key.objectid = device->devid;
545 key.type = BTRFS_DEV_EXTENT_KEY;
546 ret = btrfs_insert_empty_item(trans, root, path, &key,
550 leaf = path->nodes[0];
551 extent = btrfs_item_ptr(leaf, path->slots[0],
552 struct btrfs_dev_extent);
553 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
554 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
555 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
557 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
558 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
561 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
562 btrfs_mark_buffer_dirty(leaf);
564 btrfs_free_path(path);
568 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
570 struct btrfs_path *path;
572 struct btrfs_key key;
573 struct btrfs_chunk *chunk;
574 struct btrfs_key found_key;
576 path = btrfs_alloc_path();
579 key.objectid = objectid;
580 key.offset = (u64)-1;
581 key.type = BTRFS_CHUNK_ITEM_KEY;
583 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
589 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
593 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
595 if (found_key.objectid != objectid)
598 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
600 *offset = found_key.offset +
601 btrfs_chunk_length(path->nodes[0], chunk);
606 btrfs_free_path(path);
610 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
614 struct btrfs_key key;
615 struct btrfs_key found_key;
617 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
618 key.type = BTRFS_DEV_ITEM_KEY;
619 key.offset = (u64)-1;
621 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
627 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
632 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
634 *objectid = found_key.offset + 1;
638 btrfs_release_path(root, path);
643 * the device information is stored in the chunk root
644 * the btrfs_device struct should be fully filled in
646 int btrfs_add_device(struct btrfs_trans_handle *trans,
647 struct btrfs_root *root,
648 struct btrfs_device *device)
651 struct btrfs_path *path;
652 struct btrfs_dev_item *dev_item;
653 struct extent_buffer *leaf;
654 struct btrfs_key key;
658 root = root->fs_info->chunk_root;
660 path = btrfs_alloc_path();
664 ret = find_next_devid(root, path, &free_devid);
668 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
669 key.type = BTRFS_DEV_ITEM_KEY;
670 key.offset = free_devid;
672 ret = btrfs_insert_empty_item(trans, root, path, &key,
677 leaf = path->nodes[0];
678 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
680 device->devid = free_devid;
681 btrfs_set_device_id(leaf, dev_item, device->devid);
682 btrfs_set_device_type(leaf, dev_item, device->type);
683 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
684 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
685 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
686 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
687 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
688 btrfs_set_device_group(leaf, dev_item, 0);
689 btrfs_set_device_seek_speed(leaf, dev_item, 0);
690 btrfs_set_device_bandwidth(leaf, dev_item, 0);
692 ptr = (unsigned long)btrfs_device_uuid(dev_item);
693 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
694 btrfs_mark_buffer_dirty(leaf);
698 btrfs_free_path(path);
702 static int btrfs_rm_dev_item(struct btrfs_root *root,
703 struct btrfs_device *device)
706 struct btrfs_path *path;
707 struct block_device *bdev = device->bdev;
708 struct btrfs_device *next_dev;
709 struct btrfs_key key;
711 struct btrfs_fs_devices *fs_devices;
712 struct btrfs_trans_handle *trans;
714 root = root->fs_info->chunk_root;
716 path = btrfs_alloc_path();
720 trans = btrfs_start_transaction(root, 1);
721 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
722 key.type = BTRFS_DEV_ITEM_KEY;
723 key.offset = device->devid;
725 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
734 ret = btrfs_del_item(trans, root, path);
739 * at this point, the device is zero sized. We want to
740 * remove it from the devices list and zero out the old super
742 list_del_init(&device->dev_list);
743 list_del_init(&device->dev_alloc_list);
744 fs_devices = root->fs_info->fs_devices;
746 next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
748 if (bdev == root->fs_info->sb->s_bdev)
749 root->fs_info->sb->s_bdev = next_dev->bdev;
750 if (bdev == fs_devices->latest_bdev)
751 fs_devices->latest_bdev = next_dev->bdev;
753 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
754 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
755 total_bytes - device->total_bytes);
757 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
758 btrfs_set_super_num_devices(&root->fs_info->super_copy,
761 btrfs_free_path(path);
762 btrfs_commit_transaction(trans, root);
766 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
768 struct btrfs_device *device;
769 struct block_device *bdev;
770 struct buffer_head *bh = NULL;
771 struct btrfs_super_block *disk_super;
776 mutex_lock(&root->fs_info->fs_mutex);
777 mutex_lock(&uuid_mutex);
779 all_avail = root->fs_info->avail_data_alloc_bits |
780 root->fs_info->avail_system_alloc_bits |
781 root->fs_info->avail_metadata_alloc_bits;
783 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
784 btrfs_super_num_devices(&root->fs_info->super_copy) <= 4) {
785 printk("btrfs: unable to go below four devices on raid10\n");
790 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
791 btrfs_super_num_devices(&root->fs_info->super_copy) <= 2) {
792 printk("btrfs: unable to go below two devices on raid1\n");
797 if (strcmp(device_path, "missing") == 0) {
798 struct list_head *cur;
799 struct list_head *devices;
800 struct btrfs_device *tmp;
803 devices = &root->fs_info->fs_devices->devices;
804 list_for_each(cur, devices) {
805 tmp = list_entry(cur, struct btrfs_device, dev_list);
806 if (tmp->in_fs_metadata && !tmp->bdev) {
815 printk("btrfs: no missing devices found to remove\n");
820 bdev = open_bdev_excl(device_path, 0,
821 root->fs_info->bdev_holder);
827 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
832 disk_super = (struct btrfs_super_block *)bh->b_data;
833 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
834 sizeof(disk_super->magic))) {
838 if (memcmp(disk_super->fsid, root->fs_info->fsid,
843 devid = le64_to_cpu(disk_super->dev_item.devid);
844 device = btrfs_find_device(root, devid, NULL);
851 root->fs_info->fs_devices->num_devices--;
853 ret = btrfs_shrink_device(device, 0);
858 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
863 /* make sure this device isn't detected as part of
866 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
867 set_buffer_dirty(bh);
868 sync_dirty_buffer(bh);
874 /* one close for the device struct or super_block */
875 close_bdev_excl(device->bdev);
876 root->fs_info->fs_devices->open_devices--;
879 /* one close for us */
880 close_bdev_excl(bdev);
891 close_bdev_excl(bdev);
893 mutex_unlock(&uuid_mutex);
894 mutex_unlock(&root->fs_info->fs_mutex);
898 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
900 struct btrfs_trans_handle *trans;
901 struct btrfs_device *device;
902 struct block_device *bdev;
903 struct list_head *cur;
904 struct list_head *devices;
909 bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
913 mutex_lock(&root->fs_info->fs_mutex);
914 trans = btrfs_start_transaction(root, 1);
915 devices = &root->fs_info->fs_devices->devices;
916 list_for_each(cur, devices) {
917 device = list_entry(cur, struct btrfs_device, dev_list);
918 if (device->bdev == bdev) {
924 device = kzalloc(sizeof(*device), GFP_NOFS);
926 /* we can safely leave the fs_devices entry around */
931 device->barriers = 1;
932 generate_random_uuid(device->uuid);
933 spin_lock_init(&device->io_lock);
934 device->name = kstrdup(device_path, GFP_NOFS);
939 device->io_width = root->sectorsize;
940 device->io_align = root->sectorsize;
941 device->sector_size = root->sectorsize;
942 device->total_bytes = i_size_read(bdev->bd_inode);
943 device->dev_root = root->fs_info->dev_root;
945 device->in_fs_metadata = 1;
947 ret = btrfs_add_device(trans, root, device);
951 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
952 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
953 total_bytes + device->total_bytes);
955 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
956 btrfs_set_super_num_devices(&root->fs_info->super_copy,
959 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
960 list_add(&device->dev_alloc_list,
961 &root->fs_info->fs_devices->alloc_list);
962 root->fs_info->fs_devices->num_devices++;
963 root->fs_info->fs_devices->open_devices++;
965 btrfs_end_transaction(trans, root);
966 mutex_unlock(&root->fs_info->fs_mutex);
970 close_bdev_excl(bdev);
974 int btrfs_update_device(struct btrfs_trans_handle *trans,
975 struct btrfs_device *device)
978 struct btrfs_path *path;
979 struct btrfs_root *root;
980 struct btrfs_dev_item *dev_item;
981 struct extent_buffer *leaf;
982 struct btrfs_key key;
984 root = device->dev_root->fs_info->chunk_root;
986 path = btrfs_alloc_path();
990 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
991 key.type = BTRFS_DEV_ITEM_KEY;
992 key.offset = device->devid;
994 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1003 leaf = path->nodes[0];
1004 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1006 btrfs_set_device_id(leaf, dev_item, device->devid);
1007 btrfs_set_device_type(leaf, dev_item, device->type);
1008 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1009 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1010 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1011 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1012 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1013 btrfs_mark_buffer_dirty(leaf);
1016 btrfs_free_path(path);
1020 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1021 struct btrfs_device *device, u64 new_size)
1023 struct btrfs_super_block *super_copy =
1024 &device->dev_root->fs_info->super_copy;
1025 u64 old_total = btrfs_super_total_bytes(super_copy);
1026 u64 diff = new_size - device->total_bytes;
1028 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1029 return btrfs_update_device(trans, device);
1032 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1033 struct btrfs_root *root,
1034 u64 chunk_tree, u64 chunk_objectid,
1038 struct btrfs_path *path;
1039 struct btrfs_key key;
1041 root = root->fs_info->chunk_root;
1042 path = btrfs_alloc_path();
1046 key.objectid = chunk_objectid;
1047 key.offset = chunk_offset;
1048 key.type = BTRFS_CHUNK_ITEM_KEY;
1050 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1053 ret = btrfs_del_item(trans, root, path);
1056 btrfs_free_path(path);
1060 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1063 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1064 struct btrfs_disk_key *disk_key;
1065 struct btrfs_chunk *chunk;
1072 struct btrfs_key key;
1074 array_size = btrfs_super_sys_array_size(super_copy);
1076 ptr = super_copy->sys_chunk_array;
1079 while (cur < array_size) {
1080 disk_key = (struct btrfs_disk_key *)ptr;
1081 btrfs_disk_key_to_cpu(&key, disk_key);
1083 len = sizeof(*disk_key);
1085 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1086 chunk = (struct btrfs_chunk *)(ptr + len);
1087 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1088 len += btrfs_chunk_item_size(num_stripes);
1093 if (key.objectid == chunk_objectid &&
1094 key.offset == chunk_offset) {
1095 memmove(ptr, ptr + len, array_size - (cur + len));
1097 btrfs_set_super_sys_array_size(super_copy, array_size);
1107 int btrfs_relocate_chunk(struct btrfs_root *root,
1108 u64 chunk_tree, u64 chunk_objectid,
1111 struct extent_map_tree *em_tree;
1112 struct btrfs_root *extent_root;
1113 struct btrfs_trans_handle *trans;
1114 struct extent_map *em;
1115 struct map_lookup *map;
1119 printk("btrfs relocating chunk %llu\n",
1120 (unsigned long long)chunk_offset);
1121 root = root->fs_info->chunk_root;
1122 extent_root = root->fs_info->extent_root;
1123 em_tree = &root->fs_info->mapping_tree.map_tree;
1125 /* step one, relocate all the extents inside this chunk */
1126 ret = btrfs_shrink_extent_tree(extent_root, chunk_offset);
1129 trans = btrfs_start_transaction(root, 1);
1133 * step two, delete the device extents and the
1134 * chunk tree entries
1136 spin_lock(&em_tree->lock);
1137 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1138 spin_unlock(&em_tree->lock);
1140 BUG_ON(em->start > chunk_offset ||
1141 em->start + em->len < chunk_offset);
1142 map = (struct map_lookup *)em->bdev;
1144 for (i = 0; i < map->num_stripes; i++) {
1145 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1146 map->stripes[i].physical);
1149 if (map->stripes[i].dev) {
1150 ret = btrfs_update_device(trans, map->stripes[i].dev);
1154 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1159 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1160 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1164 spin_lock(&em_tree->lock);
1165 remove_extent_mapping(em_tree, em);
1169 /* once for the tree */
1170 free_extent_map(em);
1171 spin_unlock(&em_tree->lock);
1174 free_extent_map(em);
1176 btrfs_end_transaction(trans, root);
1180 static u64 div_factor(u64 num, int factor)
1190 int btrfs_balance(struct btrfs_root *dev_root)
1193 struct list_head *cur;
1194 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1195 struct btrfs_device *device;
1198 struct btrfs_path *path;
1199 struct btrfs_key key;
1200 struct btrfs_chunk *chunk;
1201 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1202 struct btrfs_trans_handle *trans;
1203 struct btrfs_key found_key;
1206 dev_root = dev_root->fs_info->dev_root;
1208 mutex_lock(&dev_root->fs_info->fs_mutex);
1209 /* step one make some room on all the devices */
1210 list_for_each(cur, devices) {
1211 device = list_entry(cur, struct btrfs_device, dev_list);
1212 old_size = device->total_bytes;
1213 size_to_free = div_factor(old_size, 1);
1214 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1215 if (device->total_bytes - device->bytes_used > size_to_free)
1218 ret = btrfs_shrink_device(device, old_size - size_to_free);
1221 trans = btrfs_start_transaction(dev_root, 1);
1224 ret = btrfs_grow_device(trans, device, old_size);
1227 btrfs_end_transaction(trans, dev_root);
1230 /* step two, relocate all the chunks */
1231 path = btrfs_alloc_path();
1234 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1235 key.offset = (u64)-1;
1236 key.type = BTRFS_CHUNK_ITEM_KEY;
1239 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1244 * this shouldn't happen, it means the last relocate
1250 ret = btrfs_previous_item(chunk_root, path, 0,
1251 BTRFS_CHUNK_ITEM_KEY);
1255 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1257 if (found_key.objectid != key.objectid)
1259 chunk = btrfs_item_ptr(path->nodes[0],
1261 struct btrfs_chunk);
1262 key.offset = found_key.offset;
1263 /* chunk zero is special */
1264 if (key.offset == 0)
1267 ret = btrfs_relocate_chunk(chunk_root,
1268 chunk_root->root_key.objectid,
1272 btrfs_release_path(chunk_root, path);
1276 btrfs_free_path(path);
1277 mutex_unlock(&dev_root->fs_info->fs_mutex);
1282 * shrinking a device means finding all of the device extents past
1283 * the new size, and then following the back refs to the chunks.
1284 * The chunk relocation code actually frees the device extent
1286 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1288 struct btrfs_trans_handle *trans;
1289 struct btrfs_root *root = device->dev_root;
1290 struct btrfs_dev_extent *dev_extent = NULL;
1291 struct btrfs_path *path;
1298 struct extent_buffer *l;
1299 struct btrfs_key key;
1300 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1301 u64 old_total = btrfs_super_total_bytes(super_copy);
1302 u64 diff = device->total_bytes - new_size;
1305 path = btrfs_alloc_path();
1309 trans = btrfs_start_transaction(root, 1);
1317 device->total_bytes = new_size;
1318 ret = btrfs_update_device(trans, device);
1320 btrfs_end_transaction(trans, root);
1323 WARN_ON(diff > old_total);
1324 btrfs_set_super_total_bytes(super_copy, old_total - diff);
1325 btrfs_end_transaction(trans, root);
1327 key.objectid = device->devid;
1328 key.offset = (u64)-1;
1329 key.type = BTRFS_DEV_EXTENT_KEY;
1332 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1336 ret = btrfs_previous_item(root, path, 0, key.type);
1345 slot = path->slots[0];
1346 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1348 if (key.objectid != device->devid)
1351 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1352 length = btrfs_dev_extent_length(l, dev_extent);
1354 if (key.offset + length <= new_size)
1357 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1358 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1359 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1360 btrfs_release_path(root, path);
1362 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1369 btrfs_free_path(path);
1373 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1374 struct btrfs_root *root,
1375 struct btrfs_key *key,
1376 struct btrfs_chunk *chunk, int item_size)
1378 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1379 struct btrfs_disk_key disk_key;
1383 array_size = btrfs_super_sys_array_size(super_copy);
1384 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1387 ptr = super_copy->sys_chunk_array + array_size;
1388 btrfs_cpu_key_to_disk(&disk_key, key);
1389 memcpy(ptr, &disk_key, sizeof(disk_key));
1390 ptr += sizeof(disk_key);
1391 memcpy(ptr, chunk, item_size);
1392 item_size += sizeof(disk_key);
1393 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1397 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
1400 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1402 else if (type & BTRFS_BLOCK_GROUP_RAID10)
1403 return calc_size * (num_stripes / sub_stripes);
1405 return calc_size * num_stripes;
1409 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1410 struct btrfs_root *extent_root, u64 *start,
1411 u64 *num_bytes, u64 type)
1414 struct btrfs_fs_info *info = extent_root->fs_info;
1415 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1416 struct btrfs_path *path;
1417 struct btrfs_stripe *stripes;
1418 struct btrfs_device *device = NULL;
1419 struct btrfs_chunk *chunk;
1420 struct list_head private_devs;
1421 struct list_head *dev_list;
1422 struct list_head *cur;
1423 struct extent_map_tree *em_tree;
1424 struct map_lookup *map;
1425 struct extent_map *em;
1426 int min_stripe_size = 1 * 1024 * 1024;
1428 u64 calc_size = 1024 * 1024 * 1024;
1429 u64 max_chunk_size = calc_size;
1434 int num_stripes = 1;
1435 int min_stripes = 1;
1436 int sub_stripes = 0;
1440 int stripe_len = 64 * 1024;
1441 struct btrfs_key key;
1443 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1444 (type & BTRFS_BLOCK_GROUP_DUP)) {
1446 type &= ~BTRFS_BLOCK_GROUP_DUP;
1448 dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1449 if (list_empty(dev_list))
1452 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1453 num_stripes = btrfs_super_num_devices(&info->super_copy);
1456 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1460 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1461 num_stripes = min_t(u64, 2,
1462 btrfs_super_num_devices(&info->super_copy));
1463 if (num_stripes < 2)
1467 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1468 num_stripes = btrfs_super_num_devices(&info->super_copy);
1469 if (num_stripes < 4)
1471 num_stripes &= ~(u32)1;
1476 if (type & BTRFS_BLOCK_GROUP_DATA) {
1477 max_chunk_size = 10 * calc_size;
1478 min_stripe_size = 64 * 1024 * 1024;
1479 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1480 max_chunk_size = 4 * calc_size;
1481 min_stripe_size = 32 * 1024 * 1024;
1482 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1483 calc_size = 8 * 1024 * 1024;
1484 max_chunk_size = calc_size * 2;
1485 min_stripe_size = 1 * 1024 * 1024;
1488 path = btrfs_alloc_path();
1492 /* we don't want a chunk larger than 10% of the FS */
1493 percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1494 max_chunk_size = min(percent_max, max_chunk_size);
1497 if (calc_size * num_stripes > max_chunk_size) {
1498 calc_size = max_chunk_size;
1499 do_div(calc_size, num_stripes);
1500 do_div(calc_size, stripe_len);
1501 calc_size *= stripe_len;
1503 /* we don't want tiny stripes */
1504 calc_size = max_t(u64, min_stripe_size, calc_size);
1506 do_div(calc_size, stripe_len);
1507 calc_size *= stripe_len;
1509 INIT_LIST_HEAD(&private_devs);
1510 cur = dev_list->next;
1513 if (type & BTRFS_BLOCK_GROUP_DUP)
1514 min_free = calc_size * 2;
1516 min_free = calc_size;
1518 /* we add 1MB because we never use the first 1MB of the device */
1519 min_free += 1024 * 1024;
1521 /* build a private list of devices we will allocate from */
1522 while(index < num_stripes) {
1523 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1525 if (device->total_bytes > device->bytes_used)
1526 avail = device->total_bytes - device->bytes_used;
1531 if (device->in_fs_metadata && avail >= min_free) {
1532 u64 ignored_start = 0;
1533 ret = find_free_dev_extent(trans, device, path,
1537 list_move_tail(&device->dev_alloc_list,
1540 if (type & BTRFS_BLOCK_GROUP_DUP)
1543 } else if (device->in_fs_metadata && avail > max_avail)
1545 if (cur == dev_list)
1548 if (index < num_stripes) {
1549 list_splice(&private_devs, dev_list);
1550 if (index >= min_stripes) {
1551 num_stripes = index;
1552 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1553 num_stripes /= sub_stripes;
1554 num_stripes *= sub_stripes;
1559 if (!looped && max_avail > 0) {
1561 calc_size = max_avail;
1564 btrfs_free_path(path);
1567 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1568 key.type = BTRFS_CHUNK_ITEM_KEY;
1569 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1572 btrfs_free_path(path);
1576 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1578 btrfs_free_path(path);
1582 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1585 btrfs_free_path(path);
1588 btrfs_free_path(path);
1591 stripes = &chunk->stripe;
1592 *num_bytes = chunk_bytes_by_type(type, calc_size,
1593 num_stripes, sub_stripes);
1596 while(index < num_stripes) {
1597 struct btrfs_stripe *stripe;
1598 BUG_ON(list_empty(&private_devs));
1599 cur = private_devs.next;
1600 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1602 /* loop over this device again if we're doing a dup group */
1603 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1604 (index == num_stripes - 1))
1605 list_move_tail(&device->dev_alloc_list, dev_list);
1607 ret = btrfs_alloc_dev_extent(trans, device,
1608 info->chunk_root->root_key.objectid,
1609 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1610 calc_size, &dev_offset);
1612 device->bytes_used += calc_size;
1613 ret = btrfs_update_device(trans, device);
1616 map->stripes[index].dev = device;
1617 map->stripes[index].physical = dev_offset;
1618 stripe = stripes + index;
1619 btrfs_set_stack_stripe_devid(stripe, device->devid);
1620 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1621 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1622 physical = dev_offset;
1625 BUG_ON(!list_empty(&private_devs));
1627 /* key was set above */
1628 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1629 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1630 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1631 btrfs_set_stack_chunk_type(chunk, type);
1632 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1633 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1634 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1635 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1636 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1637 map->sector_size = extent_root->sectorsize;
1638 map->stripe_len = stripe_len;
1639 map->io_align = stripe_len;
1640 map->io_width = stripe_len;
1642 map->num_stripes = num_stripes;
1643 map->sub_stripes = sub_stripes;
1645 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1646 btrfs_chunk_item_size(num_stripes));
1648 *start = key.offset;;
1650 em = alloc_extent_map(GFP_NOFS);
1653 em->bdev = (struct block_device *)map;
1654 em->start = key.offset;
1655 em->len = *num_bytes;
1656 em->block_start = 0;
1658 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1659 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1660 chunk, btrfs_chunk_item_size(num_stripes));
1665 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1666 spin_lock(&em_tree->lock);
1667 ret = add_extent_mapping(em_tree, em);
1668 spin_unlock(&em_tree->lock);
1670 free_extent_map(em);
1674 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1676 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1679 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1681 struct extent_map *em;
1684 spin_lock(&tree->map_tree.lock);
1685 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1687 remove_extent_mapping(&tree->map_tree, em);
1688 spin_unlock(&tree->map_tree.lock);
1693 free_extent_map(em);
1694 /* once for the tree */
1695 free_extent_map(em);
1699 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1701 struct extent_map *em;
1702 struct map_lookup *map;
1703 struct extent_map_tree *em_tree = &map_tree->map_tree;
1706 spin_lock(&em_tree->lock);
1707 em = lookup_extent_mapping(em_tree, logical, len);
1708 spin_unlock(&em_tree->lock);
1711 BUG_ON(em->start > logical || em->start + em->len < logical);
1712 map = (struct map_lookup *)em->bdev;
1713 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1714 ret = map->num_stripes;
1715 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1716 ret = map->sub_stripes;
1719 free_extent_map(em);
1723 static int find_live_mirror(struct map_lookup *map, int first, int num,
1727 if (map->stripes[optimal].dev->bdev)
1729 for (i = first; i < first + num; i++) {
1730 if (map->stripes[i].dev->bdev)
1733 /* we couldn't find one that doesn't fail. Just return something
1734 * and the io error handling code will clean up eventually
1739 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1740 u64 logical, u64 *length,
1741 struct btrfs_multi_bio **multi_ret,
1742 int mirror_num, struct page *unplug_page)
1744 struct extent_map *em;
1745 struct map_lookup *map;
1746 struct extent_map_tree *em_tree = &map_tree->map_tree;
1750 int stripes_allocated = 8;
1751 int stripes_required = 1;
1756 struct btrfs_multi_bio *multi = NULL;
1758 if (multi_ret && !(rw & (1 << BIO_RW))) {
1759 stripes_allocated = 1;
1763 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1768 atomic_set(&multi->error, 0);
1771 spin_lock(&em_tree->lock);
1772 em = lookup_extent_mapping(em_tree, logical, *length);
1773 spin_unlock(&em_tree->lock);
1775 if (!em && unplug_page)
1779 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1783 BUG_ON(em->start > logical || em->start + em->len < logical);
1784 map = (struct map_lookup *)em->bdev;
1785 offset = logical - em->start;
1787 if (mirror_num > map->num_stripes)
1790 /* if our multi bio struct is too small, back off and try again */
1791 if (rw & (1 << BIO_RW)) {
1792 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1793 BTRFS_BLOCK_GROUP_DUP)) {
1794 stripes_required = map->num_stripes;
1796 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1797 stripes_required = map->sub_stripes;
1801 if (multi_ret && rw == WRITE &&
1802 stripes_allocated < stripes_required) {
1803 stripes_allocated = map->num_stripes;
1804 free_extent_map(em);
1810 * stripe_nr counts the total number of stripes we have to stride
1811 * to get to this block
1813 do_div(stripe_nr, map->stripe_len);
1815 stripe_offset = stripe_nr * map->stripe_len;
1816 BUG_ON(offset < stripe_offset);
1818 /* stripe_offset is the offset of this block in its stripe*/
1819 stripe_offset = offset - stripe_offset;
1821 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1822 BTRFS_BLOCK_GROUP_RAID10 |
1823 BTRFS_BLOCK_GROUP_DUP)) {
1824 /* we limit the length of each bio to what fits in a stripe */
1825 *length = min_t(u64, em->len - offset,
1826 map->stripe_len - stripe_offset);
1828 *length = em->len - offset;
1831 if (!multi_ret && !unplug_page)
1836 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1837 if (unplug_page || (rw & (1 << BIO_RW)))
1838 num_stripes = map->num_stripes;
1839 else if (mirror_num)
1840 stripe_index = mirror_num - 1;
1842 stripe_index = find_live_mirror(map, 0,
1844 current->pid % map->num_stripes);
1847 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1848 if (rw & (1 << BIO_RW))
1849 num_stripes = map->num_stripes;
1850 else if (mirror_num)
1851 stripe_index = mirror_num - 1;
1853 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1854 int factor = map->num_stripes / map->sub_stripes;
1856 stripe_index = do_div(stripe_nr, factor);
1857 stripe_index *= map->sub_stripes;
1859 if (unplug_page || (rw & (1 << BIO_RW)))
1860 num_stripes = map->sub_stripes;
1861 else if (mirror_num)
1862 stripe_index += mirror_num - 1;
1864 stripe_index = find_live_mirror(map, stripe_index,
1865 map->sub_stripes, stripe_index +
1866 current->pid % map->sub_stripes);
1870 * after this do_div call, stripe_nr is the number of stripes
1871 * on this device we have to walk to find the data, and
1872 * stripe_index is the number of our device in the stripe array
1874 stripe_index = do_div(stripe_nr, map->num_stripes);
1876 BUG_ON(stripe_index >= map->num_stripes);
1878 for (i = 0; i < num_stripes; i++) {
1880 struct btrfs_device *device;
1881 struct backing_dev_info *bdi;
1883 device = map->stripes[stripe_index].dev;
1885 bdi = blk_get_backing_dev_info(device->bdev);
1886 if (bdi->unplug_io_fn) {
1887 bdi->unplug_io_fn(bdi, unplug_page);
1891 multi->stripes[i].physical =
1892 map->stripes[stripe_index].physical +
1893 stripe_offset + stripe_nr * map->stripe_len;
1894 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1900 multi->num_stripes = num_stripes;
1901 multi->max_errors = max_errors;
1904 free_extent_map(em);
1908 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1909 u64 logical, u64 *length,
1910 struct btrfs_multi_bio **multi_ret, int mirror_num)
1912 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
1916 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
1917 u64 logical, struct page *page)
1919 u64 length = PAGE_CACHE_SIZE;
1920 return __btrfs_map_block(map_tree, READ, logical, &length,
1925 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1926 static void end_bio_multi_stripe(struct bio *bio, int err)
1928 static int end_bio_multi_stripe(struct bio *bio,
1929 unsigned int bytes_done, int err)
1932 struct btrfs_multi_bio *multi = bio->bi_private;
1934 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1939 atomic_inc(&multi->error);
1941 if (atomic_dec_and_test(&multi->stripes_pending)) {
1942 bio->bi_private = multi->private;
1943 bio->bi_end_io = multi->end_io;
1944 /* only send an error to the higher layers if it is
1945 * beyond the tolerance of the multi-bio
1947 if (atomic_read(&multi->error) > multi->max_errors) {
1951 * this bio is actually up to date, we didn't
1952 * go over the max number of errors
1954 set_bit(BIO_UPTODATE, &bio->bi_flags);
1959 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1960 bio_endio(bio, bio->bi_size, err);
1962 bio_endio(bio, err);
1967 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1972 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
1975 struct btrfs_mapping_tree *map_tree;
1976 struct btrfs_device *dev;
1977 struct bio *first_bio = bio;
1978 u64 logical = bio->bi_sector << 9;
1981 struct btrfs_multi_bio *multi = NULL;
1986 length = bio->bi_size;
1987 map_tree = &root->fs_info->mapping_tree;
1988 map_length = length;
1990 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
1994 total_devs = multi->num_stripes;
1995 if (map_length < length) {
1996 printk("mapping failed logical %Lu bio len %Lu "
1997 "len %Lu\n", logical, length, map_length);
2000 multi->end_io = first_bio->bi_end_io;
2001 multi->private = first_bio->bi_private;
2002 atomic_set(&multi->stripes_pending, multi->num_stripes);
2004 while(dev_nr < total_devs) {
2005 if (total_devs > 1) {
2006 if (dev_nr < total_devs - 1) {
2007 bio = bio_clone(first_bio, GFP_NOFS);
2012 bio->bi_private = multi;
2013 bio->bi_end_io = end_bio_multi_stripe;
2015 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2016 dev = multi->stripes[dev_nr].dev;
2017 if (dev && dev->bdev) {
2018 bio->bi_bdev = dev->bdev;
2019 spin_lock(&dev->io_lock);
2021 spin_unlock(&dev->io_lock);
2022 submit_bio(rw, bio);
2024 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2025 bio->bi_sector = logical >> 9;
2026 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2027 bio_endio(bio, bio->bi_size, -EIO);
2029 bio_endio(bio, -EIO);
2034 if (total_devs == 1)
2039 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2042 struct list_head *head = &root->fs_info->fs_devices->devices;
2044 return __find_device(head, devid, uuid);
2047 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2048 u64 devid, u8 *dev_uuid)
2050 struct btrfs_device *device;
2051 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2053 device = kzalloc(sizeof(*device), GFP_NOFS);
2054 list_add(&device->dev_list,
2055 &fs_devices->devices);
2056 list_add(&device->dev_alloc_list,
2057 &fs_devices->alloc_list);
2058 device->barriers = 1;
2059 device->dev_root = root->fs_info->dev_root;
2060 device->devid = devid;
2061 fs_devices->num_devices++;
2062 spin_lock_init(&device->io_lock);
2063 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2068 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2069 struct extent_buffer *leaf,
2070 struct btrfs_chunk *chunk)
2072 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2073 struct map_lookup *map;
2074 struct extent_map *em;
2078 u8 uuid[BTRFS_UUID_SIZE];
2083 logical = key->offset;
2084 length = btrfs_chunk_length(leaf, chunk);
2086 spin_lock(&map_tree->map_tree.lock);
2087 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2088 spin_unlock(&map_tree->map_tree.lock);
2090 /* already mapped? */
2091 if (em && em->start <= logical && em->start + em->len > logical) {
2092 free_extent_map(em);
2095 free_extent_map(em);
2098 map = kzalloc(sizeof(*map), GFP_NOFS);
2102 em = alloc_extent_map(GFP_NOFS);
2105 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2106 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2108 free_extent_map(em);
2112 em->bdev = (struct block_device *)map;
2113 em->start = logical;
2115 em->block_start = 0;
2117 map->num_stripes = num_stripes;
2118 map->io_width = btrfs_chunk_io_width(leaf, chunk);
2119 map->io_align = btrfs_chunk_io_align(leaf, chunk);
2120 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2121 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2122 map->type = btrfs_chunk_type(leaf, chunk);
2123 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2124 for (i = 0; i < num_stripes; i++) {
2125 map->stripes[i].physical =
2126 btrfs_stripe_offset_nr(leaf, chunk, i);
2127 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2128 read_extent_buffer(leaf, uuid, (unsigned long)
2129 btrfs_stripe_dev_uuid_nr(chunk, i),
2131 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
2133 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2135 free_extent_map(em);
2138 if (!map->stripes[i].dev) {
2139 map->stripes[i].dev =
2140 add_missing_dev(root, devid, uuid);
2141 if (!map->stripes[i].dev) {
2143 free_extent_map(em);
2147 map->stripes[i].dev->in_fs_metadata = 1;
2150 spin_lock(&map_tree->map_tree.lock);
2151 ret = add_extent_mapping(&map_tree->map_tree, em);
2152 spin_unlock(&map_tree->map_tree.lock);
2154 free_extent_map(em);
2159 static int fill_device_from_item(struct extent_buffer *leaf,
2160 struct btrfs_dev_item *dev_item,
2161 struct btrfs_device *device)
2165 device->devid = btrfs_device_id(leaf, dev_item);
2166 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2167 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2168 device->type = btrfs_device_type(leaf, dev_item);
2169 device->io_align = btrfs_device_io_align(leaf, dev_item);
2170 device->io_width = btrfs_device_io_width(leaf, dev_item);
2171 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2173 ptr = (unsigned long)btrfs_device_uuid(dev_item);
2174 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2179 static int read_one_dev(struct btrfs_root *root,
2180 struct extent_buffer *leaf,
2181 struct btrfs_dev_item *dev_item)
2183 struct btrfs_device *device;
2186 u8 dev_uuid[BTRFS_UUID_SIZE];
2188 devid = btrfs_device_id(leaf, dev_item);
2189 read_extent_buffer(leaf, dev_uuid,
2190 (unsigned long)btrfs_device_uuid(dev_item),
2192 device = btrfs_find_device(root, devid, dev_uuid);
2194 printk("warning devid %Lu missing\n", devid);
2195 device = add_missing_dev(root, devid, dev_uuid);
2200 fill_device_from_item(leaf, dev_item, device);
2201 device->dev_root = root->fs_info->dev_root;
2202 device->in_fs_metadata = 1;
2205 ret = btrfs_open_device(device);
2213 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2215 struct btrfs_dev_item *dev_item;
2217 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2219 return read_one_dev(root, buf, dev_item);
2222 int btrfs_read_sys_array(struct btrfs_root *root)
2224 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2225 struct extent_buffer *sb;
2226 struct btrfs_disk_key *disk_key;
2227 struct btrfs_chunk *chunk;
2229 unsigned long sb_ptr;
2235 struct btrfs_key key;
2237 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2238 BTRFS_SUPER_INFO_SIZE);
2241 btrfs_set_buffer_uptodate(sb);
2242 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2243 array_size = btrfs_super_sys_array_size(super_copy);
2245 ptr = super_copy->sys_chunk_array;
2246 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2249 while (cur < array_size) {
2250 disk_key = (struct btrfs_disk_key *)ptr;
2251 btrfs_disk_key_to_cpu(&key, disk_key);
2253 len = sizeof(*disk_key); ptr += len;
2257 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2258 chunk = (struct btrfs_chunk *)sb_ptr;
2259 ret = read_one_chunk(root, &key, sb, chunk);
2262 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2263 len = btrfs_chunk_item_size(num_stripes);
2272 free_extent_buffer(sb);
2276 int btrfs_read_chunk_tree(struct btrfs_root *root)
2278 struct btrfs_path *path;
2279 struct extent_buffer *leaf;
2280 struct btrfs_key key;
2281 struct btrfs_key found_key;
2285 root = root->fs_info->chunk_root;
2287 path = btrfs_alloc_path();
2291 /* first we search for all of the device items, and then we
2292 * read in all of the chunk items. This way we can create chunk
2293 * mappings that reference all of the devices that are afound
2295 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2299 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2301 leaf = path->nodes[0];
2302 slot = path->slots[0];
2303 if (slot >= btrfs_header_nritems(leaf)) {
2304 ret = btrfs_next_leaf(root, path);
2311 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2312 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2313 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2315 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2316 struct btrfs_dev_item *dev_item;
2317 dev_item = btrfs_item_ptr(leaf, slot,
2318 struct btrfs_dev_item);
2319 ret = read_one_dev(root, leaf, dev_item);
2322 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2323 struct btrfs_chunk *chunk;
2324 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2325 ret = read_one_chunk(root, &found_key, leaf, chunk);
2329 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2331 btrfs_release_path(root, path);
2335 btrfs_free_path(path);