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.
20 #include <sys/types.h>
22 #include <uuid/uuid.h>
27 #include "transaction.h"
28 #include "print-tree.h"
31 #include "kernel-lib/raid56.h"
34 struct btrfs_device *dev;
38 static inline int nr_parity_stripes(struct map_lookup *map)
40 if (map->type & BTRFS_BLOCK_GROUP_RAID5)
42 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
48 static inline int nr_data_stripes(struct map_lookup *map)
50 return map->num_stripes - nr_parity_stripes(map);
53 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
55 static LIST_HEAD(fs_uuids);
57 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
60 struct btrfs_device *dev;
61 struct list_head *cur;
63 list_for_each(cur, head) {
64 dev = list_entry(cur, struct btrfs_device, dev_list);
65 if (dev->devid == devid &&
66 !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
73 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
75 struct list_head *cur;
76 struct btrfs_fs_devices *fs_devices;
78 list_for_each(cur, &fs_uuids) {
79 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
80 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
86 static int device_list_add(const char *path,
87 struct btrfs_super_block *disk_super,
88 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
90 struct btrfs_device *device;
91 struct btrfs_fs_devices *fs_devices;
92 u64 found_transid = btrfs_super_generation(disk_super);
94 fs_devices = find_fsid(disk_super->fsid);
96 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
99 INIT_LIST_HEAD(&fs_devices->devices);
100 list_add(&fs_devices->list, &fs_uuids);
101 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
102 fs_devices->latest_devid = devid;
103 fs_devices->latest_trans = found_transid;
104 fs_devices->lowest_devid = (u64)-1;
107 device = __find_device(&fs_devices->devices, devid,
108 disk_super->dev_item.uuid);
111 device = kzalloc(sizeof(*device), GFP_NOFS);
113 /* we can safely leave the fs_devices entry around */
117 device->devid = devid;
118 device->generation = found_transid;
119 memcpy(device->uuid, disk_super->dev_item.uuid,
121 device->name = kstrdup(path, GFP_NOFS);
126 device->label = kstrdup(disk_super->label, GFP_NOFS);
127 if (!device->label) {
132 device->total_devs = btrfs_super_num_devices(disk_super);
133 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
134 device->total_bytes =
135 btrfs_stack_device_total_bytes(&disk_super->dev_item);
137 btrfs_stack_device_bytes_used(&disk_super->dev_item);
138 list_add(&device->dev_list, &fs_devices->devices);
139 device->fs_devices = fs_devices;
140 } else if (!device->name || strcmp(device->name, path)) {
144 * The existing device has newer generation, so this one could
145 * be a stale one, don't add it.
147 if (found_transid < device->generation) {
149 "adding device %s gen %llu but found an existing device %s gen %llu",
150 path, found_transid, device->name,
163 if (found_transid > fs_devices->latest_trans) {
164 fs_devices->latest_devid = devid;
165 fs_devices->latest_trans = found_transid;
167 if (fs_devices->lowest_devid > devid) {
168 fs_devices->lowest_devid = devid;
170 *fs_devices_ret = fs_devices;
174 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
176 struct btrfs_fs_devices *seed_devices;
177 struct btrfs_device *device;
183 while (!list_empty(&fs_devices->devices)) {
184 device = list_entry(fs_devices->devices.next,
185 struct btrfs_device, dev_list);
186 if (device->fd != -1) {
187 if (fsync(device->fd) == -1) {
188 warning("fsync on device %llu failed: %s",
189 device->devid, strerror(errno));
192 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
193 fprintf(stderr, "Warning, could not drop caches\n");
197 device->writeable = 0;
198 list_del(&device->dev_list);
199 /* free the memory */
205 seed_devices = fs_devices->seed;
206 fs_devices->seed = NULL;
208 struct btrfs_fs_devices *orig;
211 fs_devices = seed_devices;
212 list_del(&orig->list);
216 list_del(&fs_devices->list);
223 void btrfs_close_all_devices(void)
225 struct btrfs_fs_devices *fs_devices;
227 while (!list_empty(&fs_uuids)) {
228 fs_devices = list_entry(fs_uuids.next, struct btrfs_fs_devices,
230 btrfs_close_devices(fs_devices);
234 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
237 struct list_head *head = &fs_devices->devices;
238 struct list_head *cur;
239 struct btrfs_device *device;
242 list_for_each(cur, head) {
243 device = list_entry(cur, struct btrfs_device, dev_list);
245 printk("no name for device %llu, skip it now\n", device->devid);
249 fd = open(device->name, flags);
252 error("cannot open device '%s': %s", device->name,
257 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
258 fprintf(stderr, "Warning, could not drop caches\n");
260 if (device->devid == fs_devices->latest_devid)
261 fs_devices->latest_bdev = fd;
262 if (device->devid == fs_devices->lowest_devid)
263 fs_devices->lowest_bdev = fd;
266 device->writeable = 1;
270 btrfs_close_devices(fs_devices);
274 int btrfs_scan_one_device(int fd, const char *path,
275 struct btrfs_fs_devices **fs_devices_ret,
276 u64 *total_devs, u64 super_offset, unsigned sbflags)
278 struct btrfs_super_block *disk_super;
279 char buf[BTRFS_SUPER_INFO_SIZE];
283 disk_super = (struct btrfs_super_block *)buf;
284 ret = btrfs_read_dev_super(fd, disk_super, super_offset, sbflags);
287 devid = btrfs_stack_device_id(&disk_super->dev_item);
288 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
291 *total_devs = btrfs_super_num_devices(disk_super);
293 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
299 * find_free_dev_extent_start - find free space in the specified device
300 * @device: the device which we search the free space in
301 * @num_bytes: the size of the free space that we need
302 * @search_start: the position from which to begin the search
303 * @start: store the start of the free space.
304 * @len: the size of the free space. that we find, or the size
305 * of the max free space if we don't find suitable free space
307 * this uses a pretty simple search, the expectation is that it is
308 * called very infrequently and that a given device has a small number
311 * @start is used to store the start of the free space if we find. But if we
312 * don't find suitable free space, it will be used to store the start position
313 * of the max free space.
315 * @len is used to store the size of the free space that we find.
316 * But if we don't find suitable free space, it is used to store the size of
317 * the max free space.
319 static int find_free_dev_extent_start(struct btrfs_trans_handle *trans,
320 struct btrfs_device *device, u64 num_bytes,
321 u64 search_start, u64 *start, u64 *len)
323 struct btrfs_key key;
324 struct btrfs_root *root = device->dev_root;
325 struct btrfs_dev_extent *dev_extent;
326 struct btrfs_path *path;
331 u64 search_end = device->total_bytes;
334 struct extent_buffer *l;
335 u64 min_search_start;
338 * We don't want to overwrite the superblock on the drive nor any area
339 * used by the boot loader (grub for example), so we make sure to start
340 * at an offset of at least 1MB.
342 min_search_start = max(root->fs_info->alloc_start, (u64)SZ_1M);
343 search_start = max(search_start, min_search_start);
345 path = btrfs_alloc_path();
349 max_hole_start = search_start;
352 if (search_start >= search_end) {
359 key.objectid = device->devid;
360 key.offset = search_start;
361 key.type = BTRFS_DEV_EXTENT_KEY;
363 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
367 ret = btrfs_previous_item(root, path, key.objectid, key.type);
374 slot = path->slots[0];
375 if (slot >= btrfs_header_nritems(l)) {
376 ret = btrfs_next_leaf(root, path);
384 btrfs_item_key_to_cpu(l, &key, slot);
386 if (key.objectid < device->devid)
389 if (key.objectid > device->devid)
392 if (key.type != BTRFS_DEV_EXTENT_KEY)
395 if (key.offset > search_start) {
396 hole_size = key.offset - search_start;
399 * Have to check before we set max_hole_start, otherwise
400 * we could end up sending back this offset anyway.
402 if (hole_size > max_hole_size) {
403 max_hole_start = search_start;
404 max_hole_size = hole_size;
408 * If this free space is greater than which we need,
409 * it must be the max free space that we have found
410 * until now, so max_hole_start must point to the start
411 * of this free space and the length of this free space
412 * is stored in max_hole_size. Thus, we return
413 * max_hole_start and max_hole_size and go back to the
416 if (hole_size >= num_bytes) {
422 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
423 extent_end = key.offset + btrfs_dev_extent_length(l,
425 if (extent_end > search_start)
426 search_start = extent_end;
433 * At this point, search_start should be the end of
434 * allocated dev extents, and when shrinking the device,
435 * search_end may be smaller than search_start.
437 if (search_end > search_start) {
438 hole_size = search_end - search_start;
440 if (hole_size > max_hole_size) {
441 max_hole_start = search_start;
442 max_hole_size = hole_size;
447 if (max_hole_size < num_bytes)
453 btrfs_free_path(path);
454 *start = max_hole_start;
456 *len = max_hole_size;
460 int find_free_dev_extent(struct btrfs_trans_handle *trans,
461 struct btrfs_device *device, u64 num_bytes,
464 /* FIXME use last free of some kind */
465 return find_free_dev_extent_start(trans, device,
466 num_bytes, 0, start, NULL);
469 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
470 struct btrfs_device *device,
471 u64 chunk_tree, u64 chunk_objectid,
473 u64 num_bytes, u64 *start, int convert)
476 struct btrfs_path *path;
477 struct btrfs_root *root = device->dev_root;
478 struct btrfs_dev_extent *extent;
479 struct extent_buffer *leaf;
480 struct btrfs_key key;
482 path = btrfs_alloc_path();
487 * For convert case, just skip search free dev_extent, as caller
488 * is responsible to make sure it's free.
491 ret = find_free_dev_extent(trans, device, num_bytes,
497 key.objectid = device->devid;
499 key.type = BTRFS_DEV_EXTENT_KEY;
500 ret = btrfs_insert_empty_item(trans, root, path, &key,
504 leaf = path->nodes[0];
505 extent = btrfs_item_ptr(leaf, path->slots[0],
506 struct btrfs_dev_extent);
507 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
508 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
509 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
511 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
512 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
515 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
516 btrfs_mark_buffer_dirty(leaf);
518 btrfs_free_path(path);
522 static int find_next_chunk(struct btrfs_fs_info *fs_info, u64 *offset)
524 struct btrfs_root *root = fs_info->chunk_root;
525 struct btrfs_path *path;
527 struct btrfs_key key;
528 struct btrfs_chunk *chunk;
529 struct btrfs_key found_key;
531 path = btrfs_alloc_path();
535 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
536 key.offset = (u64)-1;
537 key.type = BTRFS_CHUNK_ITEM_KEY;
539 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
545 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
549 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
551 if (found_key.objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID)
554 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
556 *offset = found_key.offset +
557 btrfs_chunk_length(path->nodes[0], chunk);
562 btrfs_free_path(path);
566 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
570 struct btrfs_key key;
571 struct btrfs_key found_key;
573 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
574 key.type = BTRFS_DEV_ITEM_KEY;
575 key.offset = (u64)-1;
577 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
583 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
588 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
590 *objectid = found_key.offset + 1;
594 btrfs_release_path(path);
599 * the device information is stored in the chunk root
600 * the btrfs_device struct should be fully filled in
602 int btrfs_add_device(struct btrfs_trans_handle *trans,
603 struct btrfs_fs_info *fs_info,
604 struct btrfs_device *device)
607 struct btrfs_path *path;
608 struct btrfs_dev_item *dev_item;
609 struct extent_buffer *leaf;
610 struct btrfs_key key;
611 struct btrfs_root *root = fs_info->chunk_root;
615 path = btrfs_alloc_path();
619 ret = find_next_devid(root, path, &free_devid);
623 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
624 key.type = BTRFS_DEV_ITEM_KEY;
625 key.offset = free_devid;
627 ret = btrfs_insert_empty_item(trans, root, path, &key,
632 leaf = path->nodes[0];
633 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
635 device->devid = free_devid;
636 btrfs_set_device_id(leaf, dev_item, device->devid);
637 btrfs_set_device_generation(leaf, dev_item, 0);
638 btrfs_set_device_type(leaf, dev_item, device->type);
639 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
640 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
641 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
642 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
643 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
644 btrfs_set_device_group(leaf, dev_item, 0);
645 btrfs_set_device_seek_speed(leaf, dev_item, 0);
646 btrfs_set_device_bandwidth(leaf, dev_item, 0);
647 btrfs_set_device_start_offset(leaf, dev_item, 0);
649 ptr = (unsigned long)btrfs_device_uuid(dev_item);
650 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
651 ptr = (unsigned long)btrfs_device_fsid(dev_item);
652 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
653 btrfs_mark_buffer_dirty(leaf);
657 btrfs_free_path(path);
661 int btrfs_update_device(struct btrfs_trans_handle *trans,
662 struct btrfs_device *device)
665 struct btrfs_path *path;
666 struct btrfs_root *root;
667 struct btrfs_dev_item *dev_item;
668 struct extent_buffer *leaf;
669 struct btrfs_key key;
671 root = device->dev_root->fs_info->chunk_root;
673 path = btrfs_alloc_path();
677 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
678 key.type = BTRFS_DEV_ITEM_KEY;
679 key.offset = device->devid;
681 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
690 leaf = path->nodes[0];
691 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
693 btrfs_set_device_id(leaf, dev_item, device->devid);
694 btrfs_set_device_type(leaf, dev_item, device->type);
695 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
696 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
697 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
698 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
699 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
700 btrfs_mark_buffer_dirty(leaf);
703 btrfs_free_path(path);
707 int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
708 struct btrfs_chunk *chunk, int item_size)
710 struct btrfs_super_block *super_copy = fs_info->super_copy;
711 struct btrfs_disk_key disk_key;
715 array_size = btrfs_super_sys_array_size(super_copy);
716 if (array_size + item_size + sizeof(disk_key)
717 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
720 ptr = super_copy->sys_chunk_array + array_size;
721 btrfs_cpu_key_to_disk(&disk_key, key);
722 memcpy(ptr, &disk_key, sizeof(disk_key));
723 ptr += sizeof(disk_key);
724 memcpy(ptr, chunk, item_size);
725 item_size += sizeof(disk_key);
726 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
730 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
733 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
735 else if (type & BTRFS_BLOCK_GROUP_RAID10)
736 return calc_size * (num_stripes / sub_stripes);
737 else if (type & BTRFS_BLOCK_GROUP_RAID5)
738 return calc_size * (num_stripes - 1);
739 else if (type & BTRFS_BLOCK_GROUP_RAID6)
740 return calc_size * (num_stripes - 2);
742 return calc_size * num_stripes;
746 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
748 /* TODO, add a way to store the preferred stripe size */
749 return BTRFS_STRIPE_LEN;
753 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
755 * It is not equal to "device->total_bytes - device->bytes_used".
756 * We do not allocate any chunk in 1M at beginning of device, and not
757 * allowed to allocate any chunk before alloc_start if it is specified.
758 * So search holes from max(1M, alloc_start) to device->total_bytes.
760 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
761 struct btrfs_device *device,
764 struct btrfs_path *path;
765 struct btrfs_root *root = device->dev_root;
766 struct btrfs_key key;
767 struct btrfs_dev_extent *dev_extent = NULL;
768 struct extent_buffer *l;
769 u64 search_start = root->fs_info->alloc_start;
770 u64 search_end = device->total_bytes;
776 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
778 path = btrfs_alloc_path();
782 key.objectid = device->devid;
783 key.offset = root->fs_info->alloc_start;
784 key.type = BTRFS_DEV_EXTENT_KEY;
787 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
790 ret = btrfs_previous_item(root, path, 0, key.type);
796 slot = path->slots[0];
797 if (slot >= btrfs_header_nritems(l)) {
798 ret = btrfs_next_leaf(root, path);
805 btrfs_item_key_to_cpu(l, &key, slot);
807 if (key.objectid < device->devid)
809 if (key.objectid > device->devid)
811 if (key.type != BTRFS_DEV_EXTENT_KEY)
813 if (key.offset > search_end)
815 if (key.offset > search_start)
816 free_bytes += key.offset - search_start;
818 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
819 extent_end = key.offset + btrfs_dev_extent_length(l,
821 if (extent_end > search_start)
822 search_start = extent_end;
823 if (search_start > search_end)
830 if (search_start < search_end)
831 free_bytes += search_end - search_start;
833 *avail_bytes = free_bytes;
836 btrfs_free_path(path);
840 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
841 - sizeof(struct btrfs_item) \
842 - sizeof(struct btrfs_chunk)) \
843 / sizeof(struct btrfs_stripe) + 1)
845 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
846 - 2 * sizeof(struct btrfs_disk_key) \
847 - 2 * sizeof(struct btrfs_chunk)) \
848 / sizeof(struct btrfs_stripe) + 1)
850 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
851 struct btrfs_fs_info *info, u64 *start,
852 u64 *num_bytes, u64 type)
855 struct btrfs_root *extent_root = info->extent_root;
856 struct btrfs_root *chunk_root = info->chunk_root;
857 struct btrfs_stripe *stripes;
858 struct btrfs_device *device = NULL;
859 struct btrfs_chunk *chunk;
860 struct list_head private_devs;
861 struct list_head *dev_list = &info->fs_devices->devices;
862 struct list_head *cur;
863 struct map_lookup *map;
864 int min_stripe_size = SZ_1M;
865 u64 calc_size = SZ_8M;
867 u64 max_chunk_size = 4 * calc_size;
878 int stripe_len = BTRFS_STRIPE_LEN;
879 struct btrfs_key key;
882 if (list_empty(dev_list)) {
886 if (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
887 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
889 max_chunk_size = calc_size * 2;
890 min_stripe_size = SZ_1M;
891 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
892 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
894 max_chunk_size = 10 * calc_size;
895 min_stripe_size = SZ_64M;
896 max_stripes = BTRFS_MAX_DEVS(chunk_root);
897 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
899 max_chunk_size = 4 * calc_size;
900 min_stripe_size = SZ_32M;
901 max_stripes = BTRFS_MAX_DEVS(chunk_root);
904 if (type & BTRFS_BLOCK_GROUP_RAID1) {
905 num_stripes = min_t(u64, 2,
906 btrfs_super_num_devices(info->super_copy));
911 if (type & BTRFS_BLOCK_GROUP_DUP) {
915 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
916 num_stripes = btrfs_super_num_devices(info->super_copy);
917 if (num_stripes > max_stripes)
918 num_stripes = max_stripes;
921 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
922 num_stripes = btrfs_super_num_devices(info->super_copy);
923 if (num_stripes > max_stripes)
924 num_stripes = max_stripes;
927 num_stripes &= ~(u32)1;
931 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
932 num_stripes = btrfs_super_num_devices(info->super_copy);
933 if (num_stripes > max_stripes)
934 num_stripes = max_stripes;
938 stripe_len = find_raid56_stripe_len(num_stripes - 1,
939 btrfs_super_stripesize(info->super_copy));
941 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
942 num_stripes = btrfs_super_num_devices(info->super_copy);
943 if (num_stripes > max_stripes)
944 num_stripes = max_stripes;
948 stripe_len = find_raid56_stripe_len(num_stripes - 2,
949 btrfs_super_stripesize(info->super_copy));
952 /* we don't want a chunk larger than 10% of the FS */
953 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
954 max_chunk_size = min(percent_max, max_chunk_size);
957 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
959 calc_size = max_chunk_size;
960 calc_size /= num_stripes;
961 calc_size /= stripe_len;
962 calc_size *= stripe_len;
964 /* we don't want tiny stripes */
965 calc_size = max_t(u64, calc_size, min_stripe_size);
967 calc_size /= stripe_len;
968 calc_size *= stripe_len;
969 INIT_LIST_HEAD(&private_devs);
970 cur = dev_list->next;
973 if (type & BTRFS_BLOCK_GROUP_DUP)
974 min_free = calc_size * 2;
976 min_free = calc_size;
978 /* build a private list of devices we will allocate from */
979 while(index < num_stripes) {
980 device = list_entry(cur, struct btrfs_device, dev_list);
981 ret = btrfs_device_avail_bytes(trans, device, &avail);
985 if (avail >= min_free) {
986 list_move_tail(&device->dev_list, &private_devs);
988 if (type & BTRFS_BLOCK_GROUP_DUP)
990 } else if (avail > max_avail)
995 if (index < num_stripes) {
996 list_splice(&private_devs, dev_list);
997 if (index >= min_stripes) {
999 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1000 num_stripes /= sub_stripes;
1001 num_stripes *= sub_stripes;
1006 if (!looped && max_avail > 0) {
1008 calc_size = max_avail;
1013 ret = find_next_chunk(info, &offset);
1016 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1017 key.type = BTRFS_CHUNK_ITEM_KEY;
1018 key.offset = offset;
1020 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1024 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1030 stripes = &chunk->stripe;
1031 *num_bytes = chunk_bytes_by_type(type, calc_size,
1032 num_stripes, sub_stripes);
1034 while(index < num_stripes) {
1035 struct btrfs_stripe *stripe;
1036 BUG_ON(list_empty(&private_devs));
1037 cur = private_devs.next;
1038 device = list_entry(cur, struct btrfs_device, dev_list);
1040 /* loop over this device again if we're doing a dup group */
1041 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1042 (index == num_stripes - 1))
1043 list_move_tail(&device->dev_list, dev_list);
1045 ret = btrfs_alloc_dev_extent(trans, device,
1046 info->chunk_root->root_key.objectid,
1047 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1048 calc_size, &dev_offset, 0);
1052 device->bytes_used += calc_size;
1053 ret = btrfs_update_device(trans, device);
1057 map->stripes[index].dev = device;
1058 map->stripes[index].physical = dev_offset;
1059 stripe = stripes + index;
1060 btrfs_set_stack_stripe_devid(stripe, device->devid);
1061 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1062 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1065 BUG_ON(!list_empty(&private_devs));
1067 /* key was set above */
1068 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1069 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1070 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1071 btrfs_set_stack_chunk_type(chunk, type);
1072 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1073 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1074 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1075 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1076 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1077 map->sector_size = info->sectorsize;
1078 map->stripe_len = stripe_len;
1079 map->io_align = stripe_len;
1080 map->io_width = stripe_len;
1082 map->num_stripes = num_stripes;
1083 map->sub_stripes = sub_stripes;
1085 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1086 btrfs_chunk_item_size(num_stripes));
1088 *start = key.offset;;
1090 map->ce.start = key.offset;
1091 map->ce.size = *num_bytes;
1093 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1097 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1098 ret = btrfs_add_system_chunk(info, &key,
1099 chunk, btrfs_chunk_item_size(num_stripes));
1115 * Alloc a DATA chunk with SINGLE profile.
1117 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1118 * (btrfs logical bytenr == on-disk bytenr)
1119 * For that case, caller must make sure the chunk and dev_extent are not
1122 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1123 struct btrfs_fs_info *info, u64 *start,
1124 u64 num_bytes, u64 type, int convert)
1127 struct btrfs_root *extent_root = info->extent_root;
1128 struct btrfs_root *chunk_root = info->chunk_root;
1129 struct btrfs_stripe *stripes;
1130 struct btrfs_device *device = NULL;
1131 struct btrfs_chunk *chunk;
1132 struct list_head *dev_list = &info->fs_devices->devices;
1133 struct list_head *cur;
1134 struct map_lookup *map;
1135 u64 calc_size = SZ_8M;
1136 int num_stripes = 1;
1137 int sub_stripes = 0;
1140 int stripe_len = BTRFS_STRIPE_LEN;
1141 struct btrfs_key key;
1143 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1144 key.type = BTRFS_CHUNK_ITEM_KEY;
1146 if (*start != round_down(*start, info->sectorsize)) {
1147 error("DATA chunk start not sectorsize aligned: %llu",
1148 (unsigned long long)*start);
1151 key.offset = *start;
1152 dev_offset = *start;
1156 ret = find_next_chunk(info, &tmp);
1162 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1166 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1172 stripes = &chunk->stripe;
1173 calc_size = num_bytes;
1176 cur = dev_list->next;
1177 device = list_entry(cur, struct btrfs_device, dev_list);
1179 while (index < num_stripes) {
1180 struct btrfs_stripe *stripe;
1182 ret = btrfs_alloc_dev_extent(trans, device,
1183 info->chunk_root->root_key.objectid,
1184 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1185 calc_size, &dev_offset, convert);
1188 device->bytes_used += calc_size;
1189 ret = btrfs_update_device(trans, device);
1192 map->stripes[index].dev = device;
1193 map->stripes[index].physical = dev_offset;
1194 stripe = stripes + index;
1195 btrfs_set_stack_stripe_devid(stripe, device->devid);
1196 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1197 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1201 /* key was set above */
1202 btrfs_set_stack_chunk_length(chunk, num_bytes);
1203 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1204 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1205 btrfs_set_stack_chunk_type(chunk, type);
1206 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1207 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1208 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1209 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1210 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1211 map->sector_size = info->sectorsize;
1212 map->stripe_len = stripe_len;
1213 map->io_align = stripe_len;
1214 map->io_width = stripe_len;
1216 map->num_stripes = num_stripes;
1217 map->sub_stripes = sub_stripes;
1219 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1220 btrfs_chunk_item_size(num_stripes));
1223 *start = key.offset;
1225 map->ce.start = key.offset;
1226 map->ce.size = num_bytes;
1228 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1235 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
1237 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1238 struct cache_extent *ce;
1239 struct map_lookup *map;
1242 ce = search_cache_extent(&map_tree->cache_tree, logical);
1244 fprintf(stderr, "No mapping for %llu-%llu\n",
1245 (unsigned long long)logical,
1246 (unsigned long long)logical+len);
1249 if (ce->start > logical || ce->start + ce->size < logical) {
1250 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1251 "%llu-%llu\n", (unsigned long long)logical,
1252 (unsigned long long)logical+len,
1253 (unsigned long long)ce->start,
1254 (unsigned long long)ce->start + ce->size);
1257 map = container_of(ce, struct map_lookup, ce);
1259 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1260 ret = map->num_stripes;
1261 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1262 ret = map->sub_stripes;
1263 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1265 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1272 int btrfs_next_bg(struct btrfs_fs_info *fs_info, u64 *logical,
1273 u64 *size, u64 type)
1275 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1276 struct cache_extent *ce;
1277 struct map_lookup *map;
1280 ce = search_cache_extent(&map_tree->cache_tree, cur);
1284 * only jump to next bg if our cur is not 0
1285 * As the initial logical for btrfs_next_bg() is 0, and
1286 * if we jump to next bg, we skipped a valid bg.
1289 ce = next_cache_extent(ce);
1295 map = container_of(ce, struct map_lookup, ce);
1296 if (map->type & type) {
1297 *logical = ce->start;
1302 ce = next_cache_extent(ce);
1308 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
1309 u64 chunk_start, u64 physical, u64 devid,
1310 u64 **logical, int *naddrs, int *stripe_len)
1312 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1313 struct cache_extent *ce;
1314 struct map_lookup *map;
1322 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1324 map = container_of(ce, struct map_lookup, ce);
1327 rmap_len = map->stripe_len;
1328 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1329 length = ce->size / (map->num_stripes / map->sub_stripes);
1330 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1331 length = ce->size / map->num_stripes;
1332 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1333 BTRFS_BLOCK_GROUP_RAID6)) {
1334 length = ce->size / nr_data_stripes(map);
1335 rmap_len = map->stripe_len * nr_data_stripes(map);
1338 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1340 for (i = 0; i < map->num_stripes; i++) {
1341 if (devid && map->stripes[i].dev->devid != devid)
1343 if (map->stripes[i].physical > physical ||
1344 map->stripes[i].physical + length <= physical)
1347 stripe_nr = (physical - map->stripes[i].physical) /
1350 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1351 stripe_nr = (stripe_nr * map->num_stripes + i) /
1353 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1354 stripe_nr = stripe_nr * map->num_stripes + i;
1355 } /* else if RAID[56], multiply by nr_data_stripes().
1356 * Alternatively, just use rmap_len below instead of
1357 * map->stripe_len */
1359 bytenr = ce->start + stripe_nr * rmap_len;
1360 for (j = 0; j < nr; j++) {
1361 if (buf[j] == bytenr)
1370 *stripe_len = rmap_len;
1375 static inline int parity_smaller(u64 a, u64 b)
1380 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1381 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1383 struct btrfs_bio_stripe s;
1390 for (i = 0; i < bbio->num_stripes - 1; i++) {
1391 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1392 s = bbio->stripes[i];
1394 bbio->stripes[i] = bbio->stripes[i+1];
1395 raid_map[i] = raid_map[i+1];
1396 bbio->stripes[i+1] = s;
1404 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1405 u64 logical, u64 *length,
1406 struct btrfs_multi_bio **multi_ret, int mirror_num,
1409 return __btrfs_map_block(fs_info, rw, logical, length, NULL,
1410 multi_ret, mirror_num, raid_map_ret);
1413 int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1414 u64 logical, u64 *length, u64 *type,
1415 struct btrfs_multi_bio **multi_ret, int mirror_num,
1418 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1419 struct cache_extent *ce;
1420 struct map_lookup *map;
1424 u64 *raid_map = NULL;
1425 int stripes_allocated = 8;
1426 int stripes_required = 1;
1429 struct btrfs_multi_bio *multi = NULL;
1431 if (multi_ret && rw == READ) {
1432 stripes_allocated = 1;
1435 ce = search_cache_extent(&map_tree->cache_tree, logical);
1441 if (ce->start > logical) {
1443 *length = ce->start - logical;
1448 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1453 map = container_of(ce, struct map_lookup, ce);
1454 offset = logical - ce->start;
1457 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1458 BTRFS_BLOCK_GROUP_DUP)) {
1459 stripes_required = map->num_stripes;
1460 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1461 stripes_required = map->sub_stripes;
1464 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1465 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1466 /* RAID[56] write or recovery. Return all stripes */
1467 stripes_required = map->num_stripes;
1469 /* Only allocate the map if we've already got a large enough multi_ret */
1470 if (stripes_allocated >= stripes_required) {
1471 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1479 /* if our multi bio struct is too small, back off and try again */
1480 if (multi_ret && stripes_allocated < stripes_required) {
1481 stripes_allocated = stripes_required;
1488 * stripe_nr counts the total number of stripes we have to stride
1489 * to get to this block
1491 stripe_nr = stripe_nr / map->stripe_len;
1493 stripe_offset = stripe_nr * map->stripe_len;
1494 BUG_ON(offset < stripe_offset);
1496 /* stripe_offset is the offset of this block in its stripe*/
1497 stripe_offset = offset - stripe_offset;
1499 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1500 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1501 BTRFS_BLOCK_GROUP_RAID10 |
1502 BTRFS_BLOCK_GROUP_DUP)) {
1503 /* we limit the length of each bio to what fits in a stripe */
1504 *length = min_t(u64, ce->size - offset,
1505 map->stripe_len - stripe_offset);
1507 *length = ce->size - offset;
1513 multi->num_stripes = 1;
1515 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1517 multi->num_stripes = map->num_stripes;
1518 else if (mirror_num)
1519 stripe_index = mirror_num - 1;
1521 stripe_index = stripe_nr % map->num_stripes;
1522 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1523 int factor = map->num_stripes / map->sub_stripes;
1525 stripe_index = stripe_nr % factor;
1526 stripe_index *= map->sub_stripes;
1529 multi->num_stripes = map->sub_stripes;
1530 else if (mirror_num)
1531 stripe_index += mirror_num - 1;
1533 stripe_nr = stripe_nr / factor;
1534 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1536 multi->num_stripes = map->num_stripes;
1537 else if (mirror_num)
1538 stripe_index = mirror_num - 1;
1539 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1540 BTRFS_BLOCK_GROUP_RAID6)) {
1545 u64 raid56_full_stripe_start;
1546 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1549 * align the start of our data stripe in the logical
1552 raid56_full_stripe_start = offset / full_stripe_len;
1553 raid56_full_stripe_start *= full_stripe_len;
1555 /* get the data stripe number */
1556 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1557 stripe_nr = stripe_nr / nr_data_stripes(map);
1559 /* Work out the disk rotation on this stripe-set */
1560 rot = stripe_nr % map->num_stripes;
1562 /* Fill in the logical address of each stripe */
1563 tmp = stripe_nr * nr_data_stripes(map);
1565 for (i = 0; i < nr_data_stripes(map); i++)
1566 raid_map[(i+rot) % map->num_stripes] =
1567 ce->start + (tmp + i) * map->stripe_len;
1569 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1570 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1571 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1573 *length = map->stripe_len;
1576 multi->num_stripes = map->num_stripes;
1578 stripe_index = stripe_nr % nr_data_stripes(map);
1579 stripe_nr = stripe_nr / nr_data_stripes(map);
1582 * Mirror #0 or #1 means the original data block.
1583 * Mirror #2 is RAID5 parity block.
1584 * Mirror #3 is RAID6 Q block.
1587 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1589 /* We distribute the parity blocks across stripes */
1590 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1594 * after this do_div call, stripe_nr is the number of stripes
1595 * on this device we have to walk to find the data, and
1596 * stripe_index is the number of our device in the stripe array
1598 stripe_index = stripe_nr % map->num_stripes;
1599 stripe_nr = stripe_nr / map->num_stripes;
1601 BUG_ON(stripe_index >= map->num_stripes);
1603 for (i = 0; i < multi->num_stripes; i++) {
1604 multi->stripes[i].physical =
1605 map->stripes[stripe_index].physical + stripe_offset +
1606 stripe_nr * map->stripe_len;
1607 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1616 sort_parity_stripes(multi, raid_map);
1617 *raid_map_ret = raid_map;
1623 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
1626 struct btrfs_device *device;
1627 struct btrfs_fs_devices *cur_devices;
1629 cur_devices = fs_info->fs_devices;
1630 while (cur_devices) {
1632 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1633 fs_info->ignore_fsid_mismatch)) {
1634 device = __find_device(&cur_devices->devices,
1639 cur_devices = cur_devices->seed;
1644 struct btrfs_device *
1645 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1646 u64 devid, int instance)
1648 struct list_head *head = &fs_devices->devices;
1649 struct btrfs_device *dev;
1652 list_for_each_entry(dev, head, dev_list) {
1653 if (dev->devid == devid && num_found++ == instance)
1659 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
1661 struct cache_extent *ce;
1662 struct map_lookup *map;
1663 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1668 * During chunk recovering, we may fail to find block group's
1669 * corresponding chunk, we will rebuild it later
1671 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1672 if (!fs_info->is_chunk_recover)
1677 map = container_of(ce, struct map_lookup, ce);
1678 for (i = 0; i < map->num_stripes; i++) {
1679 if (!map->stripes[i].dev->writeable) {
1688 static struct btrfs_device *fill_missing_device(u64 devid)
1690 struct btrfs_device *device;
1692 device = kzalloc(sizeof(*device), GFP_NOFS);
1693 device->devid = devid;
1699 * slot == -1: SYSTEM chunk
1700 * return -EIO on error, otherwise return 0
1702 int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
1703 struct extent_buffer *leaf,
1704 struct btrfs_chunk *chunk,
1705 int slot, u64 logical)
1712 u32 chunk_ondisk_size;
1713 u32 sectorsize = fs_info->sectorsize;
1715 length = btrfs_chunk_length(leaf, chunk);
1716 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1717 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1718 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1719 type = btrfs_chunk_type(leaf, chunk);
1722 * These valid checks may be insufficient to cover every corner cases.
1724 if (!IS_ALIGNED(logical, sectorsize)) {
1725 error("invalid chunk logical %llu", logical);
1728 if (btrfs_chunk_sector_size(leaf, chunk) != sectorsize) {
1729 error("invalid chunk sectorsize %llu",
1730 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1733 if (!length || !IS_ALIGNED(length, sectorsize)) {
1734 error("invalid chunk length %llu", length);
1737 if (stripe_len != BTRFS_STRIPE_LEN) {
1738 error("invalid chunk stripe length: %llu", stripe_len);
1741 /* Check on chunk item type */
1742 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1743 error("invalid chunk type %llu", type);
1746 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1747 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1748 error("unrecognized chunk type: %llu",
1749 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1750 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1753 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1754 error("missing chunk type flag: %llu", type);
1757 if (!(is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) ||
1758 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0)) {
1759 error("conflicting chunk type detected: %llu", type);
1762 if ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
1763 !is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1764 error("conflicting chunk profile detected: %llu", type);
1768 chunk_ondisk_size = btrfs_chunk_item_size(num_stripes);
1770 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1771 * it can't exceed the system chunk array size
1772 * For normal chunk, it should match its chunk item size.
1774 if (num_stripes < 1 ||
1775 (slot == -1 && chunk_ondisk_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1776 (slot >= 0 && chunk_ondisk_size > btrfs_item_size_nr(leaf, slot))) {
1777 error("invalid num_stripes: %u", num_stripes);
1781 * Device number check against profile
1783 if ((type & BTRFS_BLOCK_GROUP_RAID10 && (sub_stripes != 2 ||
1784 !IS_ALIGNED(num_stripes, sub_stripes))) ||
1785 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1786 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1787 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1788 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1789 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1790 num_stripes != 1)) {
1791 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1792 num_stripes, sub_stripes,
1793 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1801 * Slot is used to verify the chunk item is valid
1803 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1805 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
1806 struct extent_buffer *leaf,
1807 struct btrfs_chunk *chunk, int slot)
1809 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1810 struct map_lookup *map;
1811 struct cache_extent *ce;
1815 u8 uuid[BTRFS_UUID_SIZE];
1820 logical = key->offset;
1821 length = btrfs_chunk_length(leaf, chunk);
1822 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1823 /* Validation check */
1824 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, slot, logical);
1826 error("%s checksums match, but it has an invalid chunk, %s",
1827 (slot == -1) ? "Superblock" : "Metadata",
1828 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1832 ce = search_cache_extent(&map_tree->cache_tree, logical);
1834 /* already mapped? */
1835 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1839 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1843 map->ce.start = logical;
1844 map->ce.size = length;
1845 map->num_stripes = num_stripes;
1846 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1847 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1848 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1849 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1850 map->type = btrfs_chunk_type(leaf, chunk);
1851 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1853 for (i = 0; i < num_stripes; i++) {
1854 map->stripes[i].physical =
1855 btrfs_stripe_offset_nr(leaf, chunk, i);
1856 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1857 read_extent_buffer(leaf, uuid, (unsigned long)
1858 btrfs_stripe_dev_uuid_nr(chunk, i),
1860 map->stripes[i].dev = btrfs_find_device(fs_info, devid, uuid,
1862 if (!map->stripes[i].dev) {
1863 map->stripes[i].dev = fill_missing_device(devid);
1864 printf("warning, device %llu is missing\n",
1865 (unsigned long long)devid);
1866 list_add(&map->stripes[i].dev->dev_list,
1867 &fs_info->fs_devices->devices);
1871 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1877 static int fill_device_from_item(struct extent_buffer *leaf,
1878 struct btrfs_dev_item *dev_item,
1879 struct btrfs_device *device)
1883 device->devid = btrfs_device_id(leaf, dev_item);
1884 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1885 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1886 device->type = btrfs_device_type(leaf, dev_item);
1887 device->io_align = btrfs_device_io_align(leaf, dev_item);
1888 device->io_width = btrfs_device_io_width(leaf, dev_item);
1889 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1891 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1892 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1897 static int open_seed_devices(struct btrfs_fs_info *fs_info, u8 *fsid)
1899 struct btrfs_fs_devices *fs_devices;
1902 fs_devices = fs_info->fs_devices->seed;
1903 while (fs_devices) {
1904 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1908 fs_devices = fs_devices->seed;
1911 fs_devices = find_fsid(fsid);
1913 /* missing all seed devices */
1914 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1919 INIT_LIST_HEAD(&fs_devices->devices);
1920 list_add(&fs_devices->list, &fs_uuids);
1921 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1924 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1928 fs_devices->seed = fs_info->fs_devices->seed;
1929 fs_info->fs_devices->seed = fs_devices;
1934 static int read_one_dev(struct btrfs_fs_info *fs_info,
1935 struct extent_buffer *leaf,
1936 struct btrfs_dev_item *dev_item)
1938 struct btrfs_device *device;
1941 u8 fs_uuid[BTRFS_UUID_SIZE];
1942 u8 dev_uuid[BTRFS_UUID_SIZE];
1944 devid = btrfs_device_id(leaf, dev_item);
1945 read_extent_buffer(leaf, dev_uuid,
1946 (unsigned long)btrfs_device_uuid(dev_item),
1948 read_extent_buffer(leaf, fs_uuid,
1949 (unsigned long)btrfs_device_fsid(dev_item),
1952 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
1953 ret = open_seed_devices(fs_info, fs_uuid);
1958 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
1960 device = kzalloc(sizeof(*device), GFP_NOFS);
1964 list_add(&device->dev_list,
1965 &fs_info->fs_devices->devices);
1968 fill_device_from_item(leaf, dev_item, device);
1969 device->dev_root = fs_info->dev_root;
1973 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
1975 struct btrfs_super_block *super_copy = fs_info->super_copy;
1976 struct extent_buffer *sb;
1977 struct btrfs_disk_key *disk_key;
1978 struct btrfs_chunk *chunk;
1980 unsigned long sb_array_offset;
1986 struct btrfs_key key;
1988 if (fs_info->nodesize < BTRFS_SUPER_INFO_SIZE) {
1989 printf("ERROR: nodesize %u too small to read superblock\n",
1993 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
1996 btrfs_set_buffer_uptodate(sb);
1997 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1998 array_size = btrfs_super_sys_array_size(super_copy);
2000 array_ptr = super_copy->sys_chunk_array;
2001 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
2004 while (cur_offset < array_size) {
2005 disk_key = (struct btrfs_disk_key *)array_ptr;
2006 len = sizeof(*disk_key);
2007 if (cur_offset + len > array_size)
2008 goto out_short_read;
2010 btrfs_disk_key_to_cpu(&key, disk_key);
2013 sb_array_offset += len;
2016 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2017 chunk = (struct btrfs_chunk *)sb_array_offset;
2019 * At least one btrfs_chunk with one stripe must be
2020 * present, exact stripe count check comes afterwards
2022 len = btrfs_chunk_item_size(1);
2023 if (cur_offset + len > array_size)
2024 goto out_short_read;
2026 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2029 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
2030 num_stripes, cur_offset);
2035 len = btrfs_chunk_item_size(num_stripes);
2036 if (cur_offset + len > array_size)
2037 goto out_short_read;
2039 ret = read_one_chunk(fs_info, &key, sb, chunk, -1);
2044 "ERROR: unexpected item type %u in sys_array at offset %u\n",
2045 (u32)key.type, cur_offset);
2050 sb_array_offset += len;
2053 free_extent_buffer(sb);
2057 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
2059 free_extent_buffer(sb);
2063 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
2065 struct btrfs_path *path;
2066 struct extent_buffer *leaf;
2067 struct btrfs_key key;
2068 struct btrfs_key found_key;
2069 struct btrfs_root *root = fs_info->chunk_root;
2073 path = btrfs_alloc_path();
2078 * Read all device items, and then all the chunk items. All
2079 * device items are found before any chunk item (their object id
2080 * is smaller than the lowest possible object id for a chunk
2081 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
2083 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2086 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2090 leaf = path->nodes[0];
2091 slot = path->slots[0];
2092 if (slot >= btrfs_header_nritems(leaf)) {
2093 ret = btrfs_next_leaf(root, path);
2100 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2101 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2102 struct btrfs_dev_item *dev_item;
2103 dev_item = btrfs_item_ptr(leaf, slot,
2104 struct btrfs_dev_item);
2105 ret = read_one_dev(fs_info, leaf, dev_item);
2107 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2108 struct btrfs_chunk *chunk;
2109 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2110 ret = read_one_chunk(fs_info, &found_key, leaf, chunk,
2119 btrfs_free_path(path);
2123 struct list_head *btrfs_scanned_uuids(void)
2128 static int rmw_eb(struct btrfs_fs_info *info,
2129 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2132 unsigned long orig_off = 0;
2133 unsigned long dest_off = 0;
2134 unsigned long copy_len = eb->len;
2136 ret = read_whole_eb(info, eb, 0);
2140 if (eb->start + eb->len <= orig_eb->start ||
2141 eb->start >= orig_eb->start + orig_eb->len)
2144 * | ----- orig_eb ------- |
2145 * | ----- stripe ------- |
2146 * | ----- orig_eb ------- |
2147 * | ----- orig_eb ------- |
2149 if (eb->start > orig_eb->start)
2150 orig_off = eb->start - orig_eb->start;
2151 if (orig_eb->start > eb->start)
2152 dest_off = orig_eb->start - eb->start;
2154 if (copy_len > orig_eb->len - orig_off)
2155 copy_len = orig_eb->len - orig_off;
2156 if (copy_len > eb->len - dest_off)
2157 copy_len = eb->len - dest_off;
2159 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2163 static int split_eb_for_raid56(struct btrfs_fs_info *info,
2164 struct extent_buffer *orig_eb,
2165 struct extent_buffer **ebs,
2166 u64 stripe_len, u64 *raid_map,
2169 struct extent_buffer **tmp_ebs;
2170 u64 start = orig_eb->start;
2175 tmp_ebs = calloc(num_stripes, sizeof(*tmp_ebs));
2179 /* Alloc memory in a row for data stripes */
2180 for (i = 0; i < num_stripes; i++) {
2181 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2184 tmp_ebs[i] = calloc(1, sizeof(**tmp_ebs) + stripe_len);
2191 for (i = 0; i < num_stripes; i++) {
2192 struct extent_buffer *eb = tmp_ebs[i];
2194 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2197 eb->start = raid_map[i];
2198 eb->len = stripe_len;
2202 eb->dev_bytenr = (u64)-1;
2204 this_eb_start = raid_map[i];
2206 if (start > this_eb_start ||
2207 start + orig_eb->len < this_eb_start + stripe_len) {
2208 ret = rmw_eb(info, eb, orig_eb);
2212 memcpy(eb->data, orig_eb->data + eb->start - start,
2220 for (i = 0; i < num_stripes; i++)
2226 int write_raid56_with_parity(struct btrfs_fs_info *info,
2227 struct extent_buffer *eb,
2228 struct btrfs_multi_bio *multi,
2229 u64 stripe_len, u64 *raid_map)
2231 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2234 int alloc_size = eb->len;
2237 ebs = malloc(sizeof(*ebs) * multi->num_stripes);
2238 pointers = malloc(sizeof(*pointers) * multi->num_stripes);
2239 if (!ebs || !pointers) {
2245 if (stripe_len > alloc_size)
2246 alloc_size = stripe_len;
2248 ret = split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2249 multi->num_stripes);
2253 for (i = 0; i < multi->num_stripes; i++) {
2254 struct extent_buffer *new_eb;
2255 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2256 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2257 ebs[i]->fd = multi->stripes[i].dev->fd;
2258 multi->stripes[i].dev->total_ios++;
2259 if (ebs[i]->start != raid_map[i]) {
2261 goto out_free_split;
2265 new_eb = malloc(sizeof(*eb) + alloc_size);
2268 goto out_free_split;
2270 new_eb->dev_bytenr = multi->stripes[i].physical;
2271 new_eb->fd = multi->stripes[i].dev->fd;
2272 multi->stripes[i].dev->total_ios++;
2273 new_eb->len = stripe_len;
2275 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2277 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2281 ebs[multi->num_stripes - 2] = p_eb;
2282 ebs[multi->num_stripes - 1] = q_eb;
2284 for (i = 0; i < multi->num_stripes; i++)
2285 pointers[i] = ebs[i]->data;
2287 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2289 ebs[multi->num_stripes - 1] = p_eb;
2290 for (i = 0; i < multi->num_stripes; i++)
2291 pointers[i] = ebs[i]->data;
2292 ret = raid5_gen_result(multi->num_stripes, stripe_len,
2293 multi->num_stripes - 1, pointers);
2295 goto out_free_split;
2298 for (i = 0; i < multi->num_stripes; i++) {
2299 ret = write_extent_to_disk(ebs[i]);
2301 goto out_free_split;
2305 for (i = 0; i < multi->num_stripes; i++) {
2317 * Get stripe length from chunk item and its stripe items
2319 * Caller should only call this function after validating the chunk item
2320 * by using btrfs_check_chunk_valid().
2322 u64 btrfs_stripe_length(struct btrfs_fs_info *fs_info,
2323 struct extent_buffer *leaf,
2324 struct btrfs_chunk *chunk)
2328 u32 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2329 u64 profile = btrfs_chunk_type(leaf, chunk) &
2330 BTRFS_BLOCK_GROUP_PROFILE_MASK;
2332 chunk_len = btrfs_chunk_length(leaf, chunk);
2335 case 0: /* Single profile */
2336 case BTRFS_BLOCK_GROUP_RAID1:
2337 case BTRFS_BLOCK_GROUP_DUP:
2338 stripe_len = chunk_len;
2340 case BTRFS_BLOCK_GROUP_RAID0:
2341 stripe_len = chunk_len / num_stripes;
2343 case BTRFS_BLOCK_GROUP_RAID5:
2344 stripe_len = chunk_len / (num_stripes - 1);
2346 case BTRFS_BLOCK_GROUP_RAID6:
2347 stripe_len = chunk_len / (num_stripes - 2);
2349 case BTRFS_BLOCK_GROUP_RAID10:
2350 stripe_len = chunk_len / (num_stripes /
2351 btrfs_chunk_sub_stripes(leaf, chunk));
2354 /* Invalid chunk profile found */