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 <asm/div64.h>
24 #include "extent_map.h"
26 #include "transaction.h"
27 #include "print-tree.h"
38 struct btrfs_bio_stripe stripes[];
41 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
42 (sizeof(struct btrfs_bio_stripe) * (n)))
44 static DEFINE_MUTEX(uuid_mutex);
45 static LIST_HEAD(fs_uuids);
47 int btrfs_cleanup_fs_uuids(void)
49 struct btrfs_fs_devices *fs_devices;
50 struct list_head *uuid_cur;
51 struct list_head *devices_cur;
52 struct btrfs_device *dev;
54 list_for_each(uuid_cur, &fs_uuids) {
55 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
57 while(!list_empty(&fs_devices->devices)) {
58 devices_cur = fs_devices->devices.next;
59 dev = list_entry(devices_cur, struct btrfs_device,
62 close_bdev_excl(dev->bdev);
64 list_del(&dev->dev_list);
71 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
74 struct btrfs_device *dev;
75 struct list_head *cur;
77 list_for_each(cur, head) {
78 dev = list_entry(cur, struct btrfs_device, dev_list);
79 if (dev->devid == devid &&
80 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
87 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
89 struct list_head *cur;
90 struct btrfs_fs_devices *fs_devices;
92 list_for_each(cur, &fs_uuids) {
93 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
94 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
100 static int device_list_add(const char *path,
101 struct btrfs_super_block *disk_super,
102 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
104 struct btrfs_device *device;
105 struct btrfs_fs_devices *fs_devices;
106 u64 found_transid = btrfs_super_generation(disk_super);
108 fs_devices = find_fsid(disk_super->fsid);
110 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
113 INIT_LIST_HEAD(&fs_devices->devices);
114 INIT_LIST_HEAD(&fs_devices->alloc_list);
115 list_add(&fs_devices->list, &fs_uuids);
116 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
117 fs_devices->latest_devid = devid;
118 fs_devices->latest_trans = found_transid;
119 fs_devices->lowest_devid = (u64)-1;
120 fs_devices->num_devices = 0;
123 device = __find_device(&fs_devices->devices, devid,
124 disk_super->dev_item.uuid);
127 device = kzalloc(sizeof(*device), GFP_NOFS);
129 /* we can safely leave the fs_devices entry around */
132 device->devid = devid;
133 memcpy(device->uuid, disk_super->dev_item.uuid,
135 device->barriers = 1;
136 spin_lock_init(&device->io_lock);
137 device->name = kstrdup(path, GFP_NOFS);
142 list_add(&device->dev_list, &fs_devices->devices);
143 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
144 fs_devices->num_devices++;
147 if (found_transid > fs_devices->latest_trans) {
148 fs_devices->latest_devid = devid;
149 fs_devices->latest_trans = found_transid;
151 if (fs_devices->lowest_devid > devid) {
152 fs_devices->lowest_devid = devid;
154 *fs_devices_ret = fs_devices;
158 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
160 struct list_head *head = &fs_devices->devices;
161 struct list_head *cur;
162 struct btrfs_device *device;
164 mutex_lock(&uuid_mutex);
165 list_for_each(cur, head) {
166 device = list_entry(cur, struct btrfs_device, dev_list);
168 close_bdev_excl(device->bdev);
172 mutex_unlock(&uuid_mutex);
176 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
177 int flags, void *holder)
179 struct block_device *bdev;
180 struct list_head *head = &fs_devices->devices;
181 struct list_head *cur;
182 struct btrfs_device *device;
185 mutex_lock(&uuid_mutex);
186 list_for_each(cur, head) {
187 device = list_entry(cur, struct btrfs_device, dev_list);
188 bdev = open_bdev_excl(device->name, flags, holder);
191 printk("open %s failed\n", device->name);
195 if (device->devid == fs_devices->latest_devid)
196 fs_devices->latest_bdev = bdev;
197 if (device->devid == fs_devices->lowest_devid) {
198 fs_devices->lowest_bdev = bdev;
202 mutex_unlock(&uuid_mutex);
205 mutex_unlock(&uuid_mutex);
206 btrfs_close_devices(fs_devices);
210 int btrfs_scan_one_device(const char *path, int flags, void *holder,
211 struct btrfs_fs_devices **fs_devices_ret)
213 struct btrfs_super_block *disk_super;
214 struct block_device *bdev;
215 struct buffer_head *bh;
220 mutex_lock(&uuid_mutex);
222 bdev = open_bdev_excl(path, flags, holder);
229 ret = set_blocksize(bdev, 4096);
232 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
237 disk_super = (struct btrfs_super_block *)bh->b_data;
238 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
239 sizeof(disk_super->magic))) {
243 devid = le64_to_cpu(disk_super->dev_item.devid);
244 transid = btrfs_super_generation(disk_super);
245 if (disk_super->label[0])
246 printk("device label %s ", disk_super->label);
248 /* FIXME, make a readl uuid parser */
249 printk("device fsid %llx-%llx ",
250 *(unsigned long long *)disk_super->fsid,
251 *(unsigned long long *)(disk_super->fsid + 8));
253 printk("devid %Lu transid %Lu %s\n", devid, transid, path);
254 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
259 close_bdev_excl(bdev);
261 mutex_unlock(&uuid_mutex);
266 * this uses a pretty simple search, the expectation is that it is
267 * called very infrequently and that a given device has a small number
270 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
271 struct btrfs_device *device,
272 struct btrfs_path *path,
273 u64 num_bytes, u64 *start)
275 struct btrfs_key key;
276 struct btrfs_root *root = device->dev_root;
277 struct btrfs_dev_extent *dev_extent = NULL;
280 u64 search_start = 0;
281 u64 search_end = device->total_bytes;
285 struct extent_buffer *l;
290 /* FIXME use last free of some kind */
292 /* we don't want to overwrite the superblock on the drive,
293 * so we make sure to start at an offset of at least 1MB
295 search_start = max((u64)1024 * 1024, search_start);
297 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
298 search_start = max(root->fs_info->alloc_start, search_start);
300 key.objectid = device->devid;
301 key.offset = search_start;
302 key.type = BTRFS_DEV_EXTENT_KEY;
303 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
306 ret = btrfs_previous_item(root, path, 0, key.type);
310 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
313 slot = path->slots[0];
314 if (slot >= btrfs_header_nritems(l)) {
315 ret = btrfs_next_leaf(root, path);
322 if (search_start >= search_end) {
326 *start = search_start;
330 *start = last_byte > search_start ?
331 last_byte : search_start;
332 if (search_end <= *start) {
338 btrfs_item_key_to_cpu(l, &key, slot);
340 if (key.objectid < device->devid)
343 if (key.objectid > device->devid)
346 if (key.offset >= search_start && key.offset > last_byte &&
348 if (last_byte < search_start)
349 last_byte = search_start;
350 hole_size = key.offset - last_byte;
351 if (key.offset > last_byte &&
352 hole_size >= num_bytes) {
357 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
362 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
363 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
369 /* we have to make sure we didn't find an extent that has already
370 * been allocated by the map tree or the original allocation
372 btrfs_release_path(root, path);
373 BUG_ON(*start < search_start);
375 if (*start + num_bytes > search_end) {
379 /* check for pending inserts here */
383 btrfs_release_path(root, path);
387 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
388 struct btrfs_device *device,
392 struct btrfs_path *path;
393 struct btrfs_root *root = device->dev_root;
394 struct btrfs_key key;
396 path = btrfs_alloc_path();
400 key.objectid = device->devid;
402 key.type = BTRFS_DEV_EXTENT_KEY;
404 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
407 ret = btrfs_del_item(trans, root, path);
410 btrfs_free_path(path);
414 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
415 struct btrfs_device *device,
416 u64 chunk_tree, u64 chunk_objectid,
418 u64 num_bytes, u64 *start)
421 struct btrfs_path *path;
422 struct btrfs_root *root = device->dev_root;
423 struct btrfs_dev_extent *extent;
424 struct extent_buffer *leaf;
425 struct btrfs_key key;
427 path = btrfs_alloc_path();
431 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
436 key.objectid = device->devid;
438 key.type = BTRFS_DEV_EXTENT_KEY;
439 ret = btrfs_insert_empty_item(trans, root, path, &key,
443 leaf = path->nodes[0];
444 extent = btrfs_item_ptr(leaf, path->slots[0],
445 struct btrfs_dev_extent);
446 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
447 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
448 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
450 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
451 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
454 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
455 btrfs_mark_buffer_dirty(leaf);
457 btrfs_free_path(path);
461 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
463 struct btrfs_path *path;
465 struct btrfs_key key;
466 struct btrfs_chunk *chunk;
467 struct btrfs_key found_key;
469 path = btrfs_alloc_path();
472 key.objectid = objectid;
473 key.offset = (u64)-1;
474 key.type = BTRFS_CHUNK_ITEM_KEY;
476 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
482 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
486 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
488 if (found_key.objectid != objectid)
491 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
493 *offset = found_key.offset +
494 btrfs_chunk_length(path->nodes[0], chunk);
499 btrfs_free_path(path);
503 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
507 struct btrfs_key key;
508 struct btrfs_key found_key;
510 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
511 key.type = BTRFS_DEV_ITEM_KEY;
512 key.offset = (u64)-1;
514 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
520 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
525 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
527 *objectid = found_key.offset + 1;
531 btrfs_release_path(root, path);
536 * the device information is stored in the chunk root
537 * the btrfs_device struct should be fully filled in
539 int btrfs_add_device(struct btrfs_trans_handle *trans,
540 struct btrfs_root *root,
541 struct btrfs_device *device)
544 struct btrfs_path *path;
545 struct btrfs_dev_item *dev_item;
546 struct extent_buffer *leaf;
547 struct btrfs_key key;
551 root = root->fs_info->chunk_root;
553 path = btrfs_alloc_path();
557 ret = find_next_devid(root, path, &free_devid);
561 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
562 key.type = BTRFS_DEV_ITEM_KEY;
563 key.offset = free_devid;
565 ret = btrfs_insert_empty_item(trans, root, path, &key,
570 leaf = path->nodes[0];
571 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
573 device->devid = free_devid;
574 btrfs_set_device_id(leaf, dev_item, device->devid);
575 btrfs_set_device_type(leaf, dev_item, device->type);
576 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
577 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
578 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
579 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
580 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
581 btrfs_set_device_group(leaf, dev_item, 0);
582 btrfs_set_device_seek_speed(leaf, dev_item, 0);
583 btrfs_set_device_bandwidth(leaf, dev_item, 0);
585 ptr = (unsigned long)btrfs_device_uuid(dev_item);
586 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
587 btrfs_mark_buffer_dirty(leaf);
591 btrfs_free_path(path);
595 int btrfs_update_device(struct btrfs_trans_handle *trans,
596 struct btrfs_device *device)
599 struct btrfs_path *path;
600 struct btrfs_root *root;
601 struct btrfs_dev_item *dev_item;
602 struct extent_buffer *leaf;
603 struct btrfs_key key;
605 root = device->dev_root->fs_info->chunk_root;
607 path = btrfs_alloc_path();
611 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
612 key.type = BTRFS_DEV_ITEM_KEY;
613 key.offset = device->devid;
615 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
624 leaf = path->nodes[0];
625 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
627 btrfs_set_device_id(leaf, dev_item, device->devid);
628 btrfs_set_device_type(leaf, dev_item, device->type);
629 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
630 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
631 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
632 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
633 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
634 btrfs_mark_buffer_dirty(leaf);
637 btrfs_free_path(path);
641 int btrfs_grow_device(struct btrfs_trans_handle *trans,
642 struct btrfs_device *device, u64 new_size)
644 struct btrfs_super_block *super_copy =
645 &device->dev_root->fs_info->super_copy;
646 u64 old_total = btrfs_super_total_bytes(super_copy);
647 u64 diff = new_size - device->total_bytes;
649 btrfs_set_super_total_bytes(super_copy, old_total + diff);
650 return btrfs_update_device(trans, device);
653 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
654 struct btrfs_root *root,
655 u64 chunk_tree, u64 chunk_objectid,
659 struct btrfs_path *path;
660 struct btrfs_key key;
662 root = root->fs_info->chunk_root;
663 path = btrfs_alloc_path();
667 key.objectid = chunk_objectid;
668 key.offset = chunk_offset;
669 key.type = BTRFS_CHUNK_ITEM_KEY;
671 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
674 ret = btrfs_del_item(trans, root, path);
677 btrfs_free_path(path);
681 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
684 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
685 struct btrfs_disk_key *disk_key;
686 struct btrfs_chunk *chunk;
693 struct btrfs_key key;
695 array_size = btrfs_super_sys_array_size(super_copy);
697 ptr = super_copy->sys_chunk_array;
700 while (cur < array_size) {
701 disk_key = (struct btrfs_disk_key *)ptr;
702 btrfs_disk_key_to_cpu(&key, disk_key);
704 len = sizeof(*disk_key);
706 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
707 chunk = (struct btrfs_chunk *)(ptr + len);
708 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
709 len += btrfs_chunk_item_size(num_stripes);
714 if (key.objectid == chunk_objectid &&
715 key.offset == chunk_offset) {
716 memmove(ptr, ptr + len, array_size - (cur + len));
718 btrfs_set_super_sys_array_size(super_copy, array_size);
728 int btrfs_relocate_chunk(struct btrfs_root *root,
729 u64 chunk_tree, u64 chunk_objectid,
732 struct extent_map_tree *em_tree;
733 struct btrfs_root *extent_root;
734 struct btrfs_trans_handle *trans;
735 struct extent_map *em;
736 struct map_lookup *map;
740 root = root->fs_info->chunk_root;
741 extent_root = root->fs_info->extent_root;
742 em_tree = &root->fs_info->mapping_tree.map_tree;
744 /* step one, relocate all the extents inside this chunk */
745 ret = btrfs_shrink_extent_tree(extent_root, chunk_offset);
748 trans = btrfs_start_transaction(root, 1);
752 * step two, delete the device extents and the
755 spin_lock(&em_tree->lock);
756 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
757 spin_unlock(&em_tree->lock);
759 BUG_ON(em->start > chunk_offset || em->start + em->len < chunk_offset);
760 map = (struct map_lookup *)em->bdev;
762 for (i = 0; i < map->num_stripes; i++) {
763 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
764 map->stripes[i].physical);
767 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
772 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
773 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
780 spin_lock(&em_tree->lock);
781 remove_extent_mapping(em_tree, em);
785 /* once for the tree */
787 spin_unlock(&em_tree->lock);
793 btrfs_end_transaction(trans, root);
798 * shrinking a device means finding all of the device extents past
799 * the new size, and then following the back refs to the chunks.
800 * The chunk relocation code actually frees the device extent
802 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
804 struct btrfs_trans_handle *trans;
805 struct btrfs_root *root = device->dev_root;
806 struct btrfs_dev_extent *dev_extent = NULL;
807 struct btrfs_path *path;
814 struct extent_buffer *l;
815 struct btrfs_key key;
816 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
817 u64 old_total = btrfs_super_total_bytes(super_copy);
818 u64 diff = device->total_bytes - new_size;
821 path = btrfs_alloc_path();
825 trans = btrfs_start_transaction(root, 1);
833 device->total_bytes = new_size;
834 ret = btrfs_update_device(trans, device);
836 btrfs_end_transaction(trans, root);
839 WARN_ON(diff > old_total);
840 btrfs_set_super_total_bytes(super_copy, old_total - diff);
841 btrfs_end_transaction(trans, root);
843 key.objectid = device->devid;
844 key.offset = (u64)-1;
845 key.type = BTRFS_DEV_EXTENT_KEY;
848 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
852 ret = btrfs_previous_item(root, path, 0, key.type);
861 slot = path->slots[0];
862 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
864 if (key.objectid != device->devid)
867 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
868 length = btrfs_dev_extent_length(l, dev_extent);
870 if (key.offset + length <= new_size)
873 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
874 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
875 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
876 btrfs_release_path(root, path);
878 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
885 btrfs_free_path(path);
889 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
890 struct btrfs_root *root,
891 struct btrfs_key *key,
892 struct btrfs_chunk *chunk, int item_size)
894 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
895 struct btrfs_disk_key disk_key;
899 array_size = btrfs_super_sys_array_size(super_copy);
900 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
903 ptr = super_copy->sys_chunk_array + array_size;
904 btrfs_cpu_key_to_disk(&disk_key, key);
905 memcpy(ptr, &disk_key, sizeof(disk_key));
906 ptr += sizeof(disk_key);
907 memcpy(ptr, chunk, item_size);
908 item_size += sizeof(disk_key);
909 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
913 static u64 div_factor(u64 num, int factor)
922 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
925 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
927 else if (type & BTRFS_BLOCK_GROUP_RAID10)
928 return calc_size * (num_stripes / sub_stripes);
930 return calc_size * num_stripes;
934 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
935 struct btrfs_root *extent_root, u64 *start,
936 u64 *num_bytes, u64 type)
939 struct btrfs_fs_info *info = extent_root->fs_info;
940 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
941 struct btrfs_path *path;
942 struct btrfs_stripe *stripes;
943 struct btrfs_device *device = NULL;
944 struct btrfs_chunk *chunk;
945 struct list_head private_devs;
946 struct list_head *dev_list;
947 struct list_head *cur;
948 struct extent_map_tree *em_tree;
949 struct map_lookup *map;
950 struct extent_map *em;
951 int min_stripe_size = 1 * 1024 * 1024;
953 u64 calc_size = 1024 * 1024 * 1024;
954 u64 max_chunk_size = calc_size;
965 int stripe_len = 64 * 1024;
966 struct btrfs_key key;
968 dev_list = &extent_root->fs_info->fs_devices->alloc_list;
969 if (list_empty(dev_list))
972 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
973 num_stripes = btrfs_super_num_devices(&info->super_copy);
976 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
980 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
981 num_stripes = min_t(u64, 2,
982 btrfs_super_num_devices(&info->super_copy));
987 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
988 num_stripes = btrfs_super_num_devices(&info->super_copy);
991 num_stripes &= ~(u32)1;
996 if (type & BTRFS_BLOCK_GROUP_DATA) {
997 max_chunk_size = 10 * calc_size;
998 min_stripe_size = 64 * 1024 * 1024;
999 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1000 max_chunk_size = 4 * calc_size;
1001 min_stripe_size = 32 * 1024 * 1024;
1002 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1003 calc_size = 8 * 1024 * 1024;
1004 max_chunk_size = calc_size * 2;
1005 min_stripe_size = 1 * 1024 * 1024;
1008 path = btrfs_alloc_path();
1012 /* we don't want a chunk larger than 10% of the FS */
1013 percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1014 max_chunk_size = min(percent_max, max_chunk_size);
1017 if (calc_size * num_stripes > max_chunk_size) {
1018 calc_size = max_chunk_size;
1019 do_div(calc_size, num_stripes);
1020 do_div(calc_size, stripe_len);
1021 calc_size *= stripe_len;
1023 /* we don't want tiny stripes */
1024 calc_size = max_t(u64, min_stripe_size, calc_size);
1026 do_div(calc_size, stripe_len);
1027 calc_size *= stripe_len;
1029 INIT_LIST_HEAD(&private_devs);
1030 cur = dev_list->next;
1033 if (type & BTRFS_BLOCK_GROUP_DUP)
1034 min_free = calc_size * 2;
1036 min_free = calc_size;
1038 /* we add 1MB because we never use the first 1MB of the device */
1039 min_free += 1024 * 1024;
1041 /* build a private list of devices we will allocate from */
1042 while(index < num_stripes) {
1043 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1045 avail = device->total_bytes - device->bytes_used;
1048 if (avail >= min_free) {
1049 u64 ignored_start = 0;
1050 ret = find_free_dev_extent(trans, device, path,
1054 list_move_tail(&device->dev_alloc_list,
1057 if (type & BTRFS_BLOCK_GROUP_DUP)
1060 } else if (avail > max_avail)
1062 if (cur == dev_list)
1065 if (index < num_stripes) {
1066 list_splice(&private_devs, dev_list);
1067 if (index >= min_stripes) {
1068 num_stripes = index;
1069 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1070 num_stripes /= sub_stripes;
1071 num_stripes *= sub_stripes;
1076 if (!looped && max_avail > 0) {
1078 calc_size = max_avail;
1081 btrfs_free_path(path);
1084 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1085 key.type = BTRFS_CHUNK_ITEM_KEY;
1086 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1089 btrfs_free_path(path);
1093 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1095 btrfs_free_path(path);
1099 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1102 btrfs_free_path(path);
1105 btrfs_free_path(path);
1108 stripes = &chunk->stripe;
1109 *num_bytes = chunk_bytes_by_type(type, calc_size,
1110 num_stripes, sub_stripes);
1113 printk("new chunk type %Lu start %Lu size %Lu\n", type, key.offset, *num_bytes);
1114 while(index < num_stripes) {
1115 struct btrfs_stripe *stripe;
1116 BUG_ON(list_empty(&private_devs));
1117 cur = private_devs.next;
1118 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1120 /* loop over this device again if we're doing a dup group */
1121 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1122 (index == num_stripes - 1))
1123 list_move_tail(&device->dev_alloc_list, dev_list);
1125 ret = btrfs_alloc_dev_extent(trans, device,
1126 info->chunk_root->root_key.objectid,
1127 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1128 calc_size, &dev_offset);
1130 printk("alloc chunk start %Lu size %Lu from dev %Lu type %Lu\n", key.offset, calc_size, device->devid, type);
1131 device->bytes_used += calc_size;
1132 ret = btrfs_update_device(trans, device);
1135 map->stripes[index].dev = device;
1136 map->stripes[index].physical = dev_offset;
1137 stripe = stripes + index;
1138 btrfs_set_stack_stripe_devid(stripe, device->devid);
1139 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1140 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1141 physical = dev_offset;
1144 BUG_ON(!list_empty(&private_devs));
1146 /* key was set above */
1147 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1148 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1149 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1150 btrfs_set_stack_chunk_type(chunk, type);
1151 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1152 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1153 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1154 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1155 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1156 map->sector_size = extent_root->sectorsize;
1157 map->stripe_len = stripe_len;
1158 map->io_align = stripe_len;
1159 map->io_width = stripe_len;
1161 map->num_stripes = num_stripes;
1162 map->sub_stripes = sub_stripes;
1164 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1165 btrfs_chunk_item_size(num_stripes));
1167 *start = key.offset;;
1169 em = alloc_extent_map(GFP_NOFS);
1172 em->bdev = (struct block_device *)map;
1173 em->start = key.offset;
1174 em->len = *num_bytes;
1175 em->block_start = 0;
1177 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1178 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1179 chunk, btrfs_chunk_item_size(num_stripes));
1184 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1185 spin_lock(&em_tree->lock);
1186 ret = add_extent_mapping(em_tree, em);
1187 spin_unlock(&em_tree->lock);
1189 free_extent_map(em);
1193 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1195 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1198 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1200 struct extent_map *em;
1203 spin_lock(&tree->map_tree.lock);
1204 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1206 remove_extent_mapping(&tree->map_tree, em);
1207 spin_unlock(&tree->map_tree.lock);
1212 free_extent_map(em);
1213 /* once for the tree */
1214 free_extent_map(em);
1218 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1220 struct extent_map *em;
1221 struct map_lookup *map;
1222 struct extent_map_tree *em_tree = &map_tree->map_tree;
1225 spin_lock(&em_tree->lock);
1226 em = lookup_extent_mapping(em_tree, logical, len);
1227 spin_unlock(&em_tree->lock);
1230 BUG_ON(em->start > logical || em->start + em->len < logical);
1231 map = (struct map_lookup *)em->bdev;
1232 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1233 ret = map->num_stripes;
1234 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1235 ret = map->sub_stripes;
1238 free_extent_map(em);
1242 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1243 u64 logical, u64 *length,
1244 struct btrfs_multi_bio **multi_ret,
1245 int mirror_num, struct page *unplug_page)
1247 struct extent_map *em;
1248 struct map_lookup *map;
1249 struct extent_map_tree *em_tree = &map_tree->map_tree;
1253 int stripes_allocated = 8;
1254 int stripes_required = 1;
1258 struct btrfs_multi_bio *multi = NULL;
1260 if (multi_ret && !(rw & (1 << BIO_RW))) {
1261 stripes_allocated = 1;
1265 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1271 spin_lock(&em_tree->lock);
1272 em = lookup_extent_mapping(em_tree, logical, *length);
1273 spin_unlock(&em_tree->lock);
1275 if (!em && unplug_page)
1279 printk("unable to find logical %Lu\n", logical);
1283 BUG_ON(em->start > logical || em->start + em->len < logical);
1284 map = (struct map_lookup *)em->bdev;
1285 offset = logical - em->start;
1287 if (mirror_num > map->num_stripes)
1290 /* if our multi bio struct is too small, back off and try again */
1291 if (rw & (1 << BIO_RW)) {
1292 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1293 BTRFS_BLOCK_GROUP_DUP)) {
1294 stripes_required = map->num_stripes;
1295 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1296 stripes_required = map->sub_stripes;
1299 if (multi_ret && rw == WRITE &&
1300 stripes_allocated < stripes_required) {
1301 stripes_allocated = map->num_stripes;
1302 free_extent_map(em);
1308 * stripe_nr counts the total number of stripes we have to stride
1309 * to get to this block
1311 do_div(stripe_nr, map->stripe_len);
1313 stripe_offset = stripe_nr * map->stripe_len;
1314 BUG_ON(offset < stripe_offset);
1316 /* stripe_offset is the offset of this block in its stripe*/
1317 stripe_offset = offset - stripe_offset;
1319 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1320 BTRFS_BLOCK_GROUP_RAID10 |
1321 BTRFS_BLOCK_GROUP_DUP)) {
1322 /* we limit the length of each bio to what fits in a stripe */
1323 *length = min_t(u64, em->len - offset,
1324 map->stripe_len - stripe_offset);
1326 *length = em->len - offset;
1329 if (!multi_ret && !unplug_page)
1334 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1335 if (unplug_page || (rw & (1 << BIO_RW)))
1336 num_stripes = map->num_stripes;
1337 else if (mirror_num) {
1338 stripe_index = mirror_num - 1;
1340 u64 orig_stripe_nr = stripe_nr;
1341 stripe_index = do_div(orig_stripe_nr, num_stripes);
1343 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1344 if (rw & (1 << BIO_RW))
1345 num_stripes = map->num_stripes;
1346 else if (mirror_num)
1347 stripe_index = mirror_num - 1;
1348 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1349 int factor = map->num_stripes / map->sub_stripes;
1351 stripe_index = do_div(stripe_nr, factor);
1352 stripe_index *= map->sub_stripes;
1354 if (unplug_page || (rw & (1 << BIO_RW)))
1355 num_stripes = map->sub_stripes;
1356 else if (mirror_num)
1357 stripe_index += mirror_num - 1;
1359 u64 orig_stripe_nr = stripe_nr;
1360 stripe_index += do_div(orig_stripe_nr,
1365 * after this do_div call, stripe_nr is the number of stripes
1366 * on this device we have to walk to find the data, and
1367 * stripe_index is the number of our device in the stripe array
1369 stripe_index = do_div(stripe_nr, map->num_stripes);
1371 BUG_ON(stripe_index >= map->num_stripes);
1373 for (i = 0; i < num_stripes; i++) {
1375 struct btrfs_device *device;
1376 struct backing_dev_info *bdi;
1378 device = map->stripes[stripe_index].dev;
1379 bdi = blk_get_backing_dev_info(device->bdev);
1380 if (bdi->unplug_io_fn) {
1381 bdi->unplug_io_fn(bdi, unplug_page);
1384 multi->stripes[i].physical =
1385 map->stripes[stripe_index].physical +
1386 stripe_offset + stripe_nr * map->stripe_len;
1387 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1393 multi->num_stripes = num_stripes;
1396 free_extent_map(em);
1400 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1401 u64 logical, u64 *length,
1402 struct btrfs_multi_bio **multi_ret, int mirror_num)
1404 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
1408 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
1409 u64 logical, struct page *page)
1411 u64 length = PAGE_CACHE_SIZE;
1412 return __btrfs_map_block(map_tree, READ, logical, &length,
1417 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1418 static void end_bio_multi_stripe(struct bio *bio, int err)
1420 static int end_bio_multi_stripe(struct bio *bio,
1421 unsigned int bytes_done, int err)
1424 struct btrfs_multi_bio *multi = bio->bi_private;
1426 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1433 if (atomic_dec_and_test(&multi->stripes_pending)) {
1434 bio->bi_private = multi->private;
1435 bio->bi_end_io = multi->end_io;
1437 if (!err && multi->error)
1441 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1442 bio_endio(bio, bio->bi_size, err);
1444 bio_endio(bio, err);
1449 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1454 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
1457 struct btrfs_mapping_tree *map_tree;
1458 struct btrfs_device *dev;
1459 struct bio *first_bio = bio;
1460 u64 logical = bio->bi_sector << 9;
1463 struct btrfs_multi_bio *multi = NULL;
1468 length = bio->bi_size;
1469 map_tree = &root->fs_info->mapping_tree;
1470 map_length = length;
1472 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
1476 total_devs = multi->num_stripes;
1477 if (map_length < length) {
1478 printk("mapping failed logical %Lu bio len %Lu "
1479 "len %Lu\n", logical, length, map_length);
1482 multi->end_io = first_bio->bi_end_io;
1483 multi->private = first_bio->bi_private;
1484 atomic_set(&multi->stripes_pending, multi->num_stripes);
1486 while(dev_nr < total_devs) {
1487 if (total_devs > 1) {
1488 if (dev_nr < total_devs - 1) {
1489 bio = bio_clone(first_bio, GFP_NOFS);
1494 bio->bi_private = multi;
1495 bio->bi_end_io = end_bio_multi_stripe;
1497 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
1498 dev = multi->stripes[dev_nr].dev;
1500 bio->bi_bdev = dev->bdev;
1501 spin_lock(&dev->io_lock);
1503 spin_unlock(&dev->io_lock);
1504 submit_bio(rw, bio);
1507 if (total_devs == 1)
1512 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1515 struct list_head *head = &root->fs_info->fs_devices->devices;
1517 return __find_device(head, devid, uuid);
1520 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1521 struct extent_buffer *leaf,
1522 struct btrfs_chunk *chunk)
1524 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1525 struct map_lookup *map;
1526 struct extent_map *em;
1530 u8 uuid[BTRFS_UUID_SIZE];
1535 logical = key->offset;
1536 length = btrfs_chunk_length(leaf, chunk);
1537 spin_lock(&map_tree->map_tree.lock);
1538 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
1539 spin_unlock(&map_tree->map_tree.lock);
1541 /* already mapped? */
1542 if (em && em->start <= logical && em->start + em->len > logical) {
1543 free_extent_map(em);
1546 free_extent_map(em);
1549 map = kzalloc(sizeof(*map), GFP_NOFS);
1553 em = alloc_extent_map(GFP_NOFS);
1556 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1557 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1559 free_extent_map(em);
1563 em->bdev = (struct block_device *)map;
1564 em->start = logical;
1566 em->block_start = 0;
1568 map->num_stripes = num_stripes;
1569 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1570 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1571 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1572 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1573 map->type = btrfs_chunk_type(leaf, chunk);
1574 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1575 for (i = 0; i < num_stripes; i++) {
1576 map->stripes[i].physical =
1577 btrfs_stripe_offset_nr(leaf, chunk, i);
1578 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1579 read_extent_buffer(leaf, uuid, (unsigned long)
1580 btrfs_stripe_dev_uuid_nr(chunk, i),
1582 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
1583 if (!map->stripes[i].dev) {
1585 free_extent_map(em);
1590 spin_lock(&map_tree->map_tree.lock);
1591 ret = add_extent_mapping(&map_tree->map_tree, em);
1592 spin_unlock(&map_tree->map_tree.lock);
1594 free_extent_map(em);
1599 static int fill_device_from_item(struct extent_buffer *leaf,
1600 struct btrfs_dev_item *dev_item,
1601 struct btrfs_device *device)
1605 device->devid = btrfs_device_id(leaf, dev_item);
1606 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1607 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1608 device->type = btrfs_device_type(leaf, dev_item);
1609 device->io_align = btrfs_device_io_align(leaf, dev_item);
1610 device->io_width = btrfs_device_io_width(leaf, dev_item);
1611 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1613 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1614 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1619 static int read_one_dev(struct btrfs_root *root,
1620 struct extent_buffer *leaf,
1621 struct btrfs_dev_item *dev_item)
1623 struct btrfs_device *device;
1626 u8 dev_uuid[BTRFS_UUID_SIZE];
1628 devid = btrfs_device_id(leaf, dev_item);
1629 read_extent_buffer(leaf, dev_uuid,
1630 (unsigned long)btrfs_device_uuid(dev_item),
1632 device = btrfs_find_device(root, devid, dev_uuid);
1634 printk("warning devid %Lu not found already\n", devid);
1635 device = kzalloc(sizeof(*device), GFP_NOFS);
1638 list_add(&device->dev_list,
1639 &root->fs_info->fs_devices->devices);
1640 list_add(&device->dev_alloc_list,
1641 &root->fs_info->fs_devices->alloc_list);
1642 device->barriers = 1;
1643 spin_lock_init(&device->io_lock);
1646 fill_device_from_item(leaf, dev_item, device);
1647 device->dev_root = root->fs_info->dev_root;
1650 ret = btrfs_open_device(device);
1658 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
1660 struct btrfs_dev_item *dev_item;
1662 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1664 return read_one_dev(root, buf, dev_item);
1667 int btrfs_read_sys_array(struct btrfs_root *root)
1669 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1670 struct extent_buffer *sb = root->fs_info->sb_buffer;
1671 struct btrfs_disk_key *disk_key;
1672 struct btrfs_chunk *chunk;
1674 unsigned long sb_ptr;
1680 struct btrfs_key key;
1682 array_size = btrfs_super_sys_array_size(super_copy);
1684 ptr = super_copy->sys_chunk_array;
1685 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
1688 while (cur < array_size) {
1689 disk_key = (struct btrfs_disk_key *)ptr;
1690 btrfs_disk_key_to_cpu(&key, disk_key);
1692 len = sizeof(*disk_key);
1697 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1698 chunk = (struct btrfs_chunk *)sb_ptr;
1699 ret = read_one_chunk(root, &key, sb, chunk);
1702 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1703 len = btrfs_chunk_item_size(num_stripes);
1715 int btrfs_read_chunk_tree(struct btrfs_root *root)
1717 struct btrfs_path *path;
1718 struct extent_buffer *leaf;
1719 struct btrfs_key key;
1720 struct btrfs_key found_key;
1724 root = root->fs_info->chunk_root;
1726 path = btrfs_alloc_path();
1730 /* first we search for all of the device items, and then we
1731 * read in all of the chunk items. This way we can create chunk
1732 * mappings that reference all of the devices that are afound
1734 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1738 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1740 leaf = path->nodes[0];
1741 slot = path->slots[0];
1742 if (slot >= btrfs_header_nritems(leaf)) {
1743 ret = btrfs_next_leaf(root, path);
1750 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1751 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1752 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1754 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1755 struct btrfs_dev_item *dev_item;
1756 dev_item = btrfs_item_ptr(leaf, slot,
1757 struct btrfs_dev_item);
1758 ret = read_one_dev(root, leaf, dev_item);
1761 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1762 struct btrfs_chunk *chunk;
1763 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1764 ret = read_one_chunk(root, &found_key, leaf, chunk);
1768 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1770 btrfs_release_path(root, path);
1774 btrfs_free_path(path);