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 #define _XOPEN_SOURCE 600
22 #include <sys/types.h>
24 #include <uuid/uuid.h>
29 #include "transaction.h"
30 #include "print-tree.h"
34 struct btrfs_device *dev;
39 struct cache_extent ce;
47 struct btrfs_bio_stripe stripes[];
50 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
51 (sizeof(struct btrfs_bio_stripe) * (n)))
53 static LIST_HEAD(fs_uuids);
55 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
58 struct btrfs_device *dev;
59 struct list_head *cur;
61 list_for_each(cur, head) {
62 dev = list_entry(cur, struct btrfs_device, dev_list);
63 if (dev->devid == devid &&
64 !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
71 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
73 struct list_head *cur;
74 struct btrfs_fs_devices *fs_devices;
76 list_for_each(cur, &fs_uuids) {
77 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
78 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
84 static int device_list_add(const char *path,
85 struct btrfs_super_block *disk_super,
86 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
88 struct btrfs_device *device;
89 struct btrfs_fs_devices *fs_devices;
90 u64 found_transid = btrfs_super_generation(disk_super);
92 fs_devices = find_fsid(disk_super->fsid);
94 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
97 INIT_LIST_HEAD(&fs_devices->devices);
98 list_add(&fs_devices->list, &fs_uuids);
99 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
100 fs_devices->latest_devid = devid;
101 fs_devices->latest_trans = found_transid;
102 fs_devices->lowest_devid = (u64)-1;
105 device = __find_device(&fs_devices->devices, devid,
106 disk_super->dev_item.uuid);
109 device = kzalloc(sizeof(*device), GFP_NOFS);
111 /* we can safely leave the fs_devices entry around */
114 device->devid = devid;
115 memcpy(device->uuid, disk_super->dev_item.uuid,
117 device->name = kstrdup(path, GFP_NOFS);
122 device->label = kstrdup(disk_super->label, GFP_NOFS);
123 device->total_devs = btrfs_super_num_devices(disk_super);
124 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
125 device->total_bytes =
126 btrfs_stack_device_total_bytes(&disk_super->dev_item);
128 btrfs_stack_device_bytes_used(&disk_super->dev_item);
129 list_add(&device->dev_list, &fs_devices->devices);
130 device->fs_devices = fs_devices;
133 if (found_transid > fs_devices->latest_trans) {
134 fs_devices->latest_devid = devid;
135 fs_devices->latest_trans = found_transid;
137 if (fs_devices->lowest_devid > devid) {
138 fs_devices->lowest_devid = devid;
140 *fs_devices_ret = fs_devices;
144 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
146 struct btrfs_fs_devices *seed_devices;
147 struct list_head *cur;
148 struct btrfs_device *device;
150 list_for_each(cur, &fs_devices->devices) {
151 device = list_entry(cur, struct btrfs_device, dev_list);
154 device->writeable = 0;
157 seed_devices = fs_devices->seed;
158 fs_devices->seed = NULL;
160 fs_devices = seed_devices;
167 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
170 struct list_head *head = &fs_devices->devices;
171 struct list_head *cur;
172 struct btrfs_device *device;
175 list_for_each(cur, head) {
176 device = list_entry(cur, struct btrfs_device, dev_list);
178 fd = open(device->name, flags);
184 if (device->devid == fs_devices->latest_devid)
185 fs_devices->latest_bdev = fd;
186 if (device->devid == fs_devices->lowest_devid)
187 fs_devices->lowest_bdev = fd;
190 device->writeable = 1;
194 btrfs_close_devices(fs_devices);
198 int btrfs_scan_one_device(int fd, const char *path,
199 struct btrfs_fs_devices **fs_devices_ret,
200 u64 *total_devs, u64 super_offset)
202 struct btrfs_super_block *disk_super;
213 ret = pread(fd, buf, 4096, super_offset);
218 disk_super = (struct btrfs_super_block *)buf;
219 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
220 sizeof(disk_super->magic))) {
224 devid = le64_to_cpu(disk_super->dev_item.devid);
225 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
228 *total_devs = btrfs_super_num_devices(disk_super);
229 uuid_unparse(disk_super->fsid, uuidbuf);
231 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
240 * this uses a pretty simple search, the expectation is that it is
241 * called very infrequently and that a given device has a small number
244 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
245 struct btrfs_device *device,
246 struct btrfs_path *path,
247 u64 num_bytes, u64 *start)
249 struct btrfs_key key;
250 struct btrfs_root *root = device->dev_root;
251 struct btrfs_dev_extent *dev_extent = NULL;
254 u64 search_start = 0;
255 u64 search_end = device->total_bytes;
259 struct extent_buffer *l;
264 /* FIXME use last free of some kind */
266 /* we don't want to overwrite the superblock on the drive,
267 * so we make sure to start at an offset of at least 1MB
269 search_start = max((u64)1024 * 1024, search_start);
271 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
272 search_start = max(root->fs_info->alloc_start, search_start);
274 key.objectid = device->devid;
275 key.offset = search_start;
276 key.type = BTRFS_DEV_EXTENT_KEY;
277 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
280 ret = btrfs_previous_item(root, path, 0, key.type);
284 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
287 slot = path->slots[0];
288 if (slot >= btrfs_header_nritems(l)) {
289 ret = btrfs_next_leaf(root, path);
296 if (search_start >= search_end) {
300 *start = search_start;
304 *start = last_byte > search_start ?
305 last_byte : search_start;
306 if (search_end <= *start) {
312 btrfs_item_key_to_cpu(l, &key, slot);
314 if (key.objectid < device->devid)
317 if (key.objectid > device->devid)
320 if (key.offset >= search_start && key.offset > last_byte &&
322 if (last_byte < search_start)
323 last_byte = search_start;
324 hole_size = key.offset - last_byte;
325 if (key.offset > last_byte &&
326 hole_size >= num_bytes) {
331 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
336 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
337 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
343 /* we have to make sure we didn't find an extent that has already
344 * been allocated by the map tree or the original allocation
346 btrfs_release_path(root, path);
347 BUG_ON(*start < search_start);
349 if (*start + num_bytes > search_end) {
353 /* check for pending inserts here */
357 btrfs_release_path(root, path);
361 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
362 struct btrfs_device *device,
363 u64 chunk_tree, u64 chunk_objectid,
365 u64 num_bytes, u64 *start)
368 struct btrfs_path *path;
369 struct btrfs_root *root = device->dev_root;
370 struct btrfs_dev_extent *extent;
371 struct extent_buffer *leaf;
372 struct btrfs_key key;
374 path = btrfs_alloc_path();
378 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
383 key.objectid = device->devid;
385 key.type = BTRFS_DEV_EXTENT_KEY;
386 ret = btrfs_insert_empty_item(trans, root, path, &key,
390 leaf = path->nodes[0];
391 extent = btrfs_item_ptr(leaf, path->slots[0],
392 struct btrfs_dev_extent);
393 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
394 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
395 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
397 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
398 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
401 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
402 btrfs_mark_buffer_dirty(leaf);
404 btrfs_free_path(path);
408 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
410 struct btrfs_path *path;
412 struct btrfs_key key;
413 struct btrfs_chunk *chunk;
414 struct btrfs_key found_key;
416 path = btrfs_alloc_path();
419 key.objectid = objectid;
420 key.offset = (u64)-1;
421 key.type = BTRFS_CHUNK_ITEM_KEY;
423 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
429 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
433 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
435 if (found_key.objectid != objectid)
438 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
440 *offset = found_key.offset +
441 btrfs_chunk_length(path->nodes[0], chunk);
446 btrfs_free_path(path);
450 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
454 struct btrfs_key key;
455 struct btrfs_key found_key;
457 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
458 key.type = BTRFS_DEV_ITEM_KEY;
459 key.offset = (u64)-1;
461 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
467 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
472 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
474 *objectid = found_key.offset + 1;
478 btrfs_release_path(root, path);
483 * the device information is stored in the chunk root
484 * the btrfs_device struct should be fully filled in
486 int btrfs_add_device(struct btrfs_trans_handle *trans,
487 struct btrfs_root *root,
488 struct btrfs_device *device)
491 struct btrfs_path *path;
492 struct btrfs_dev_item *dev_item;
493 struct extent_buffer *leaf;
494 struct btrfs_key key;
498 root = root->fs_info->chunk_root;
500 path = btrfs_alloc_path();
504 ret = find_next_devid(root, path, &free_devid);
508 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
509 key.type = BTRFS_DEV_ITEM_KEY;
510 key.offset = free_devid;
512 ret = btrfs_insert_empty_item(trans, root, path, &key,
517 leaf = path->nodes[0];
518 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
520 device->devid = free_devid;
521 btrfs_set_device_id(leaf, dev_item, device->devid);
522 btrfs_set_device_generation(leaf, dev_item, 0);
523 btrfs_set_device_type(leaf, dev_item, device->type);
524 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
525 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
526 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
527 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
528 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
529 btrfs_set_device_group(leaf, dev_item, 0);
530 btrfs_set_device_seek_speed(leaf, dev_item, 0);
531 btrfs_set_device_bandwidth(leaf, dev_item, 0);
533 ptr = (unsigned long)btrfs_device_uuid(dev_item);
534 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
535 ptr = (unsigned long)btrfs_device_fsid(dev_item);
536 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
537 btrfs_mark_buffer_dirty(leaf);
541 btrfs_free_path(path);
545 int btrfs_update_device(struct btrfs_trans_handle *trans,
546 struct btrfs_device *device)
549 struct btrfs_path *path;
550 struct btrfs_root *root;
551 struct btrfs_dev_item *dev_item;
552 struct extent_buffer *leaf;
553 struct btrfs_key key;
555 root = device->dev_root->fs_info->chunk_root;
557 path = btrfs_alloc_path();
561 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
562 key.type = BTRFS_DEV_ITEM_KEY;
563 key.offset = device->devid;
565 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
574 leaf = path->nodes[0];
575 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
577 btrfs_set_device_id(leaf, dev_item, device->devid);
578 btrfs_set_device_type(leaf, dev_item, device->type);
579 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
580 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
581 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
582 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
583 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
584 btrfs_mark_buffer_dirty(leaf);
587 btrfs_free_path(path);
591 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
592 struct btrfs_root *root,
593 struct btrfs_key *key,
594 struct btrfs_chunk *chunk, int item_size)
596 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
597 struct btrfs_disk_key disk_key;
601 array_size = btrfs_super_sys_array_size(super_copy);
602 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
605 ptr = super_copy->sys_chunk_array + array_size;
606 btrfs_cpu_key_to_disk(&disk_key, key);
607 memcpy(ptr, &disk_key, sizeof(disk_key));
608 ptr += sizeof(disk_key);
609 memcpy(ptr, chunk, item_size);
610 item_size += sizeof(disk_key);
611 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
615 static u64 div_factor(u64 num, int factor)
623 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
626 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
628 else if (type & BTRFS_BLOCK_GROUP_RAID10)
629 return calc_size * (num_stripes / sub_stripes);
631 return calc_size * num_stripes;
635 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
636 struct btrfs_root *extent_root, u64 *start,
637 u64 *num_bytes, u64 type)
640 struct btrfs_fs_info *info = extent_root->fs_info;
641 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
642 struct btrfs_stripe *stripes;
643 struct btrfs_device *device = NULL;
644 struct btrfs_chunk *chunk;
645 struct list_head private_devs;
646 struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices;
647 struct list_head *cur;
648 struct map_lookup *map;
649 int min_stripe_size = 1 * 1024 * 1024;
651 u64 calc_size = 8 * 1024 * 1024;
653 u64 max_chunk_size = 4 * calc_size;
663 int stripe_len = 64 * 1024;
664 struct btrfs_key key;
666 if (list_empty(dev_list)) {
670 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
671 BTRFS_BLOCK_GROUP_RAID10 |
672 BTRFS_BLOCK_GROUP_DUP)) {
673 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
674 calc_size = 8 * 1024 * 1024;
675 max_chunk_size = calc_size * 2;
676 min_stripe_size = 1 * 1024 * 1024;
677 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
678 calc_size = 1024 * 1024 * 1024;
679 max_chunk_size = 10 * calc_size;
680 min_stripe_size = 64 * 1024 * 1024;
681 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
682 calc_size = 1024 * 1024 * 1024;
683 max_chunk_size = 4 * calc_size;
684 min_stripe_size = 32 * 1024 * 1024;
687 if (type & BTRFS_BLOCK_GROUP_RAID1) {
688 num_stripes = min_t(u64, 2,
689 btrfs_super_num_devices(&info->super_copy));
694 if (type & BTRFS_BLOCK_GROUP_DUP) {
698 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
699 num_stripes = btrfs_super_num_devices(&info->super_copy);
702 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
703 num_stripes = btrfs_super_num_devices(&info->super_copy);
706 num_stripes &= ~(u32)1;
711 /* we don't want a chunk larger than 10% of the FS */
712 percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
713 max_chunk_size = min(percent_max, max_chunk_size);
716 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
718 calc_size = max_chunk_size;
719 calc_size /= num_stripes;
720 calc_size /= stripe_len;
721 calc_size *= stripe_len;
723 /* we don't want tiny stripes */
724 calc_size = max_t(u64, calc_size, min_stripe_size);
726 calc_size /= stripe_len;
727 calc_size *= stripe_len;
728 INIT_LIST_HEAD(&private_devs);
729 cur = dev_list->next;
732 if (type & BTRFS_BLOCK_GROUP_DUP)
733 min_free = calc_size * 2;
735 min_free = calc_size;
737 /* build a private list of devices we will allocate from */
738 while(index < num_stripes) {
739 device = list_entry(cur, struct btrfs_device, dev_list);
740 avail = device->total_bytes - device->bytes_used;
742 if (avail >= min_free) {
743 list_move_tail(&device->dev_list, &private_devs);
745 if (type & BTRFS_BLOCK_GROUP_DUP)
747 } else if (avail > max_avail)
752 if (index < num_stripes) {
753 list_splice(&private_devs, dev_list);
754 if (index >= min_stripes) {
756 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
757 num_stripes /= sub_stripes;
758 num_stripes *= sub_stripes;
763 if (!looped && max_avail > 0) {
765 calc_size = max_avail;
770 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
771 key.type = BTRFS_CHUNK_ITEM_KEY;
772 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
777 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
781 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
787 stripes = &chunk->stripe;
788 *num_bytes = chunk_bytes_by_type(type, calc_size,
789 num_stripes, sub_stripes);
791 while(index < num_stripes) {
792 struct btrfs_stripe *stripe;
793 BUG_ON(list_empty(&private_devs));
794 cur = private_devs.next;
795 device = list_entry(cur, struct btrfs_device, dev_list);
797 /* loop over this device again if we're doing a dup group */
798 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
799 (index == num_stripes - 1))
800 list_move_tail(&device->dev_list, dev_list);
802 ret = btrfs_alloc_dev_extent(trans, device,
803 info->chunk_root->root_key.objectid,
804 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
805 calc_size, &dev_offset);
808 device->bytes_used += calc_size;
809 ret = btrfs_update_device(trans, device);
812 map->stripes[index].dev = device;
813 map->stripes[index].physical = dev_offset;
814 stripe = stripes + index;
815 btrfs_set_stack_stripe_devid(stripe, device->devid);
816 btrfs_set_stack_stripe_offset(stripe, dev_offset);
817 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
818 physical = dev_offset;
821 BUG_ON(!list_empty(&private_devs));
823 /* key was set above */
824 btrfs_set_stack_chunk_length(chunk, *num_bytes);
825 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
826 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
827 btrfs_set_stack_chunk_type(chunk, type);
828 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
829 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
830 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
831 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
832 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
833 map->sector_size = extent_root->sectorsize;
834 map->stripe_len = stripe_len;
835 map->io_align = stripe_len;
836 map->io_width = stripe_len;
838 map->num_stripes = num_stripes;
839 map->sub_stripes = sub_stripes;
841 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
842 btrfs_chunk_item_size(num_stripes));
844 *start = key.offset;;
846 map->ce.start = key.offset;
847 map->ce.size = *num_bytes;
849 ret = insert_existing_cache_extent(
850 &extent_root->fs_info->mapping_tree.cache_tree,
854 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
855 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
856 chunk, btrfs_chunk_item_size(num_stripes));
864 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
866 cache_tree_init(&tree->cache_tree);
869 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
871 struct cache_extent *ce;
872 struct map_lookup *map;
876 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
878 BUG_ON(ce->start > logical || ce->start + ce->size < logical);
879 map = container_of(ce, struct map_lookup, ce);
881 offset = logical - ce->start;
882 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
883 ret = map->num_stripes;
884 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
885 ret = map->sub_stripes;
891 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
892 u64 chunk_start, u64 physical, u64 devid,
893 u64 **logical, int *naddrs, int *stripe_len)
895 struct cache_extent *ce;
896 struct map_lookup *map;
903 ce = find_first_cache_extent(&map_tree->cache_tree, chunk_start);
904 BUG_ON(!ce || ce->start != chunk_start);
905 map = container_of(ce, struct map_lookup, ce);
908 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
909 length = ce->size / (map->num_stripes / map->sub_stripes);
910 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
911 length = ce->size / map->num_stripes;
913 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
915 for (i = 0; i < map->num_stripes; i++) {
916 if (devid && map->stripes[i].dev->devid != devid)
918 if (map->stripes[i].physical > physical ||
919 map->stripes[i].physical + length <= physical)
922 stripe_nr = (physical - map->stripes[i].physical) /
925 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
926 stripe_nr = (stripe_nr * map->num_stripes + i) /
928 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
929 stripe_nr = stripe_nr * map->num_stripes + i;
931 bytenr = chunk_start + stripe_nr * map->stripe_len;
932 for (j = 0; j < nr; j++) {
933 if (buf[j] == bytenr)
942 *stripe_len = map->stripe_len;
947 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
948 u64 logical, u64 *length,
949 struct btrfs_multi_bio **multi_ret, int mirror_num)
951 struct cache_extent *ce;
952 struct map_lookup *map;
956 int stripes_allocated = 8;
957 int stripes_required = 1;
960 struct btrfs_multi_bio *multi = NULL;
962 if (multi_ret && rw == READ) {
963 stripes_allocated = 1;
967 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
973 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
975 BUG_ON(ce->start > logical || ce->start + ce->size < logical);
976 map = container_of(ce, struct map_lookup, ce);
977 offset = logical - ce->start;
980 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
981 BTRFS_BLOCK_GROUP_DUP)) {
982 stripes_required = map->num_stripes;
983 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
984 stripes_required = map->sub_stripes;
987 /* if our multi bio struct is too small, back off and try again */
988 if (multi_ret && rw == WRITE &&
989 stripes_allocated < stripes_required) {
990 stripes_allocated = map->num_stripes;
996 * stripe_nr counts the total number of stripes we have to stride
997 * to get to this block
999 stripe_nr = stripe_nr / map->stripe_len;
1001 stripe_offset = stripe_nr * map->stripe_len;
1002 BUG_ON(offset < stripe_offset);
1004 /* stripe_offset is the offset of this block in its stripe*/
1005 stripe_offset = offset - stripe_offset;
1007 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1008 BTRFS_BLOCK_GROUP_RAID10 |
1009 BTRFS_BLOCK_GROUP_DUP)) {
1010 /* we limit the length of each bio to what fits in a stripe */
1011 *length = min_t(u64, ce->size - offset,
1012 map->stripe_len - stripe_offset);
1014 *length = ce->size - offset;
1020 multi->num_stripes = 1;
1022 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1024 multi->num_stripes = map->num_stripes;
1025 else if (mirror_num)
1026 stripe_index = mirror_num - 1;
1028 stripe_index = stripe_nr % map->num_stripes;
1029 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1030 int factor = map->num_stripes / map->sub_stripes;
1032 stripe_index = stripe_nr % factor;
1033 stripe_index *= map->sub_stripes;
1036 multi->num_stripes = map->sub_stripes;
1037 else if (mirror_num)
1038 stripe_index += mirror_num - 1;
1040 stripe_index = stripe_nr % map->sub_stripes;
1042 stripe_nr = stripe_nr / factor;
1043 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1045 multi->num_stripes = map->num_stripes;
1046 else if (mirror_num)
1047 stripe_index = mirror_num - 1;
1050 * after this do_div call, stripe_nr is the number of stripes
1051 * on this device we have to walk to find the data, and
1052 * stripe_index is the number of our device in the stripe array
1054 stripe_index = stripe_nr % map->num_stripes;
1055 stripe_nr = stripe_nr / map->num_stripes;
1057 BUG_ON(stripe_index >= map->num_stripes);
1059 for (i = 0; i < multi->num_stripes; i++) {
1060 multi->stripes[i].physical =
1061 map->stripes[stripe_index].physical + stripe_offset +
1062 stripe_nr * map->stripe_len;
1063 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1071 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1074 struct btrfs_device *device;
1075 struct btrfs_fs_devices *cur_devices;
1077 cur_devices = root->fs_info->fs_devices;
1078 while (cur_devices) {
1080 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1081 device = __find_device(&cur_devices->devices,
1086 cur_devices = cur_devices->seed;
1091 int btrfs_bootstrap_super_map(struct btrfs_mapping_tree *map_tree,
1092 struct btrfs_fs_devices *fs_devices)
1094 struct map_lookup *map;
1095 u64 logical = BTRFS_SUPER_INFO_OFFSET;
1096 u64 length = BTRFS_SUPER_INFO_SIZE;
1097 int num_stripes = 0;
1098 int sub_stripes = 0;
1101 struct list_head *cur;
1103 list_for_each(cur, &fs_devices->devices) {
1106 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1110 map->ce.start = logical;
1111 map->ce.size = length;
1112 map->num_stripes = num_stripes;
1113 map->sub_stripes = sub_stripes;
1114 map->io_width = length;
1115 map->io_align = length;
1116 map->sector_size = length;
1117 map->stripe_len = length;
1118 map->type = BTRFS_BLOCK_GROUP_RAID1;
1121 list_for_each(cur, &fs_devices->devices) {
1122 struct btrfs_device *device = list_entry(cur,
1123 struct btrfs_device,
1125 map->stripes[i].physical = logical;
1126 map->stripes[i].dev = device;
1129 ret = insert_existing_cache_extent(&map_tree->cache_tree, &map->ce);
1130 if (ret == -EEXIST) {
1131 struct cache_extent *old;
1132 struct map_lookup *old_map;
1133 old = find_cache_extent(&map_tree->cache_tree, logical, length);
1134 old_map = container_of(old, struct map_lookup, ce);
1135 remove_cache_extent(&map_tree->cache_tree, old);
1137 ret = insert_existing_cache_extent(&map_tree->cache_tree,
1144 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1146 struct cache_extent *ce;
1147 struct map_lookup *map;
1148 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1152 ce = find_first_cache_extent(&map_tree->cache_tree, chunk_offset);
1155 map = container_of(ce, struct map_lookup, ce);
1156 for (i = 0; i < map->num_stripes; i++) {
1157 if (!map->stripes[i].dev->writeable) {
1166 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1167 struct extent_buffer *leaf,
1168 struct btrfs_chunk *chunk)
1170 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1171 struct map_lookup *map;
1172 struct cache_extent *ce;
1176 u8 uuid[BTRFS_UUID_SIZE];
1181 logical = key->offset;
1182 length = btrfs_chunk_length(leaf, chunk);
1184 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
1186 /* already mapped? */
1187 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1191 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1192 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1196 map->ce.start = logical;
1197 map->ce.size = length;
1198 map->num_stripes = num_stripes;
1199 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1200 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1201 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1202 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1203 map->type = btrfs_chunk_type(leaf, chunk);
1204 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1206 for (i = 0; i < num_stripes; i++) {
1207 map->stripes[i].physical =
1208 btrfs_stripe_offset_nr(leaf, chunk, i);
1209 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1210 read_extent_buffer(leaf, uuid, (unsigned long)
1211 btrfs_stripe_dev_uuid_nr(chunk, i),
1213 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1215 if (!map->stripes[i].dev) {
1221 ret = insert_existing_cache_extent(&map_tree->cache_tree, &map->ce);
1227 static int fill_device_from_item(struct extent_buffer *leaf,
1228 struct btrfs_dev_item *dev_item,
1229 struct btrfs_device *device)
1233 device->devid = btrfs_device_id(leaf, dev_item);
1234 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1235 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1236 device->type = btrfs_device_type(leaf, dev_item);
1237 device->io_align = btrfs_device_io_align(leaf, dev_item);
1238 device->io_width = btrfs_device_io_width(leaf, dev_item);
1239 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1241 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1242 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1247 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1249 struct btrfs_fs_devices *fs_devices;
1252 fs_devices = root->fs_info->fs_devices->seed;
1253 while (fs_devices) {
1254 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1258 fs_devices = fs_devices->seed;
1261 fs_devices = find_fsid(fsid);
1267 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1271 fs_devices->seed = root->fs_info->fs_devices->seed;
1272 root->fs_info->fs_devices->seed = fs_devices;
1277 static int read_one_dev(struct btrfs_root *root,
1278 struct extent_buffer *leaf,
1279 struct btrfs_dev_item *dev_item)
1281 struct btrfs_device *device;
1284 u8 fs_uuid[BTRFS_UUID_SIZE];
1285 u8 dev_uuid[BTRFS_UUID_SIZE];
1287 devid = btrfs_device_id(leaf, dev_item);
1288 read_extent_buffer(leaf, dev_uuid,
1289 (unsigned long)btrfs_device_uuid(dev_item),
1291 read_extent_buffer(leaf, fs_uuid,
1292 (unsigned long)btrfs_device_fsid(dev_item),
1295 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1296 ret = open_seed_devices(root, fs_uuid);
1301 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1303 printk("warning devid %llu not found already\n",
1304 (unsigned long long)devid);
1305 device = kmalloc(sizeof(*device), GFP_NOFS);
1308 device->total_ios = 0;
1309 list_add(&device->dev_list,
1310 &root->fs_info->fs_devices->devices);
1313 fill_device_from_item(leaf, dev_item, device);
1314 device->dev_root = root->fs_info->dev_root;
1318 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
1320 struct btrfs_dev_item *dev_item;
1322 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1324 return read_one_dev(root, buf, dev_item);
1327 int btrfs_read_sys_array(struct btrfs_root *root)
1329 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1330 struct extent_buffer *sb = root->fs_info->sb_buffer;
1331 struct btrfs_disk_key *disk_key;
1332 struct btrfs_chunk *chunk;
1333 struct btrfs_key key;
1338 unsigned long sb_ptr;
1342 array_size = btrfs_super_sys_array_size(super_copy);
1345 * we do this loop twice, once for the device items and
1346 * once for all of the chunks. This way there are device
1347 * structs filled in for every chunk
1349 ptr = super_copy->sys_chunk_array;
1350 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
1353 while (cur < array_size) {
1354 disk_key = (struct btrfs_disk_key *)ptr;
1355 btrfs_disk_key_to_cpu(&key, disk_key);
1357 len = sizeof(*disk_key);
1362 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1363 chunk = (struct btrfs_chunk *)sb_ptr;
1364 ret = read_one_chunk(root, &key, sb, chunk);
1366 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1367 len = btrfs_chunk_item_size(num_stripes);
1378 int btrfs_read_chunk_tree(struct btrfs_root *root)
1380 struct btrfs_path *path;
1381 struct extent_buffer *leaf;
1382 struct btrfs_key key;
1383 struct btrfs_key found_key;
1387 root = root->fs_info->chunk_root;
1389 path = btrfs_alloc_path();
1393 /* first we search for all of the device items, and then we
1394 * read in all of the chunk items. This way we can create chunk
1395 * mappings that reference all of the devices that are afound
1397 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1401 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1403 leaf = path->nodes[0];
1404 slot = path->slots[0];
1405 if (slot >= btrfs_header_nritems(leaf)) {
1406 ret = btrfs_next_leaf(root, path);
1413 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1414 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1415 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1417 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1418 struct btrfs_dev_item *dev_item;
1419 dev_item = btrfs_item_ptr(leaf, slot,
1420 struct btrfs_dev_item);
1421 ret = read_one_dev(root, leaf, dev_item);
1424 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1425 struct btrfs_chunk *chunk;
1426 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1427 ret = read_one_chunk(root, &found_key, leaf, chunk);
1432 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1434 btrfs_release_path(root, path);
1438 btrfs_free_path(path);
1444 struct list_head *btrfs_scanned_uuids(void)