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;
62 list_for_each_entry(dev, head, 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 btrfs_fs_devices *fs_devices;
75 list_for_each_entry(fs_devices, &fs_uuids, list) {
76 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
82 static int device_list_add(const char *path,
83 struct btrfs_super_block *disk_super,
84 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
86 struct btrfs_device *device;
87 struct btrfs_fs_devices *fs_devices;
88 u64 found_transid = btrfs_super_generation(disk_super);
90 fs_devices = find_fsid(disk_super->fsid);
92 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
95 INIT_LIST_HEAD(&fs_devices->devices);
96 list_add(&fs_devices->list, &fs_uuids);
97 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
98 fs_devices->latest_devid = devid;
99 fs_devices->latest_trans = found_transid;
100 fs_devices->lowest_devid = (u64)-1;
103 device = __find_device(&fs_devices->devices, devid,
104 disk_super->dev_item.uuid);
107 device = kzalloc(sizeof(*device), GFP_NOFS);
109 /* we can safely leave the fs_devices entry around */
113 device->devid = devid;
114 device->generation = found_transid;
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 if (!device->label) {
128 device->total_devs = btrfs_super_num_devices(disk_super);
129 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
130 device->total_bytes =
131 btrfs_stack_device_total_bytes(&disk_super->dev_item);
133 btrfs_stack_device_bytes_used(&disk_super->dev_item);
134 list_add(&device->dev_list, &fs_devices->devices);
135 device->fs_devices = fs_devices;
136 } else if (!device->name || strcmp(device->name, path)) {
140 * The existing device has newer generation, so this one could
141 * be a stale one, don't add it.
143 if (found_transid < device->generation) {
145 "adding device %s gen %llu but found an existing device %s gen %llu",
146 path, found_transid, device->name,
159 if (found_transid > fs_devices->latest_trans) {
160 fs_devices->latest_devid = devid;
161 fs_devices->latest_trans = found_transid;
163 if (fs_devices->lowest_devid > devid) {
164 fs_devices->lowest_devid = devid;
166 *fs_devices_ret = fs_devices;
170 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
172 struct btrfs_fs_devices *seed_devices;
173 struct btrfs_device *device;
179 while (!list_empty(&fs_devices->devices)) {
180 device = list_entry(fs_devices->devices.next,
181 struct btrfs_device, dev_list);
182 if (device->fd != -1) {
183 if (fsync(device->fd) == -1) {
184 warning("fsync on device %llu failed: %m",
188 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
189 fprintf(stderr, "Warning, could not drop caches\n");
193 device->writeable = 0;
194 list_del(&device->dev_list);
195 /* free the memory */
201 seed_devices = fs_devices->seed;
202 fs_devices->seed = NULL;
204 struct btrfs_fs_devices *orig;
207 fs_devices = seed_devices;
208 list_del(&orig->list);
212 list_del(&fs_devices->list);
219 void btrfs_close_all_devices(void)
221 struct btrfs_fs_devices *fs_devices;
223 while (!list_empty(&fs_uuids)) {
224 fs_devices = list_entry(fs_uuids.next, struct btrfs_fs_devices,
226 btrfs_close_devices(fs_devices);
230 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
233 struct btrfs_device *device;
236 list_for_each_entry(device, &fs_devices->devices, dev_list) {
238 printk("no name for device %llu, skip it now\n", device->devid);
242 fd = open(device->name, flags);
245 error("cannot open device '%s': %m", device->name);
249 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
250 fprintf(stderr, "Warning, could not drop caches\n");
252 if (device->devid == fs_devices->latest_devid)
253 fs_devices->latest_bdev = fd;
254 if (device->devid == fs_devices->lowest_devid)
255 fs_devices->lowest_bdev = fd;
258 device->writeable = 1;
262 btrfs_close_devices(fs_devices);
266 int btrfs_scan_one_device(int fd, const char *path,
267 struct btrfs_fs_devices **fs_devices_ret,
268 u64 *total_devs, u64 super_offset, unsigned sbflags)
270 struct btrfs_super_block *disk_super;
271 char buf[BTRFS_SUPER_INFO_SIZE];
275 disk_super = (struct btrfs_super_block *)buf;
276 ret = btrfs_read_dev_super(fd, disk_super, super_offset, sbflags);
279 devid = btrfs_stack_device_id(&disk_super->dev_item);
280 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
283 *total_devs = btrfs_super_num_devices(disk_super);
285 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
291 * find_free_dev_extent_start - find free space in the specified device
292 * @device: the device which we search the free space in
293 * @num_bytes: the size of the free space that we need
294 * @search_start: the position from which to begin the search
295 * @start: store the start of the free space.
296 * @len: the size of the free space. that we find, or the size
297 * of the max free space if we don't find suitable free space
299 * this uses a pretty simple search, the expectation is that it is
300 * called very infrequently and that a given device has a small number
303 * @start is used to store the start of the free space if we find. But if we
304 * don't find suitable free space, it will be used to store the start position
305 * of the max free space.
307 * @len is used to store the size of the free space that we find.
308 * But if we don't find suitable free space, it is used to store the size of
309 * the max free space.
311 static int find_free_dev_extent_start(struct btrfs_device *device,
312 u64 num_bytes, u64 search_start,
313 u64 *start, u64 *len)
315 struct btrfs_key key;
316 struct btrfs_root *root = device->dev_root;
317 struct btrfs_dev_extent *dev_extent;
318 struct btrfs_path *path;
323 u64 search_end = device->total_bytes;
326 struct extent_buffer *l;
327 u64 min_search_start;
330 * We don't want to overwrite the superblock on the drive nor any area
331 * used by the boot loader (grub for example), so we make sure to start
332 * at an offset of at least 1MB.
334 min_search_start = max(root->fs_info->alloc_start, (u64)SZ_1M);
335 search_start = max(search_start, min_search_start);
337 path = btrfs_alloc_path();
341 max_hole_start = search_start;
344 if (search_start >= search_end) {
351 key.objectid = device->devid;
352 key.offset = search_start;
353 key.type = BTRFS_DEV_EXTENT_KEY;
355 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
359 ret = btrfs_previous_item(root, path, key.objectid, key.type);
366 slot = path->slots[0];
367 if (slot >= btrfs_header_nritems(l)) {
368 ret = btrfs_next_leaf(root, path);
376 btrfs_item_key_to_cpu(l, &key, slot);
378 if (key.objectid < device->devid)
381 if (key.objectid > device->devid)
384 if (key.type != BTRFS_DEV_EXTENT_KEY)
387 if (key.offset > search_start) {
388 hole_size = key.offset - search_start;
391 * Have to check before we set max_hole_start, otherwise
392 * we could end up sending back this offset anyway.
394 if (hole_size > max_hole_size) {
395 max_hole_start = search_start;
396 max_hole_size = hole_size;
400 * If this free space is greater than which we need,
401 * it must be the max free space that we have found
402 * until now, so max_hole_start must point to the start
403 * of this free space and the length of this free space
404 * is stored in max_hole_size. Thus, we return
405 * max_hole_start and max_hole_size and go back to the
408 if (hole_size >= num_bytes) {
414 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
415 extent_end = key.offset + btrfs_dev_extent_length(l,
417 if (extent_end > search_start)
418 search_start = extent_end;
425 * At this point, search_start should be the end of
426 * allocated dev extents, and when shrinking the device,
427 * search_end may be smaller than search_start.
429 if (search_end > search_start) {
430 hole_size = search_end - search_start;
432 if (hole_size > max_hole_size) {
433 max_hole_start = search_start;
434 max_hole_size = hole_size;
439 if (max_hole_size < num_bytes)
445 btrfs_free_path(path);
446 *start = max_hole_start;
448 *len = max_hole_size;
452 static int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
455 /* FIXME use last free of some kind */
456 return find_free_dev_extent_start(device, num_bytes, 0, start, NULL);
459 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
460 struct btrfs_device *device,
461 u64 chunk_offset, u64 num_bytes, u64 *start,
465 struct btrfs_path *path;
466 struct btrfs_root *root = device->dev_root;
467 struct btrfs_dev_extent *extent;
468 struct extent_buffer *leaf;
469 struct btrfs_key key;
471 path = btrfs_alloc_path();
476 * For convert case, just skip search free dev_extent, as caller
477 * is responsible to make sure it's free.
480 ret = find_free_dev_extent(device, num_bytes, start);
485 key.objectid = device->devid;
487 key.type = BTRFS_DEV_EXTENT_KEY;
488 ret = btrfs_insert_empty_item(trans, root, path, &key,
492 leaf = path->nodes[0];
493 extent = btrfs_item_ptr(leaf, path->slots[0],
494 struct btrfs_dev_extent);
495 btrfs_set_dev_extent_chunk_tree(leaf, extent, BTRFS_CHUNK_TREE_OBJECTID);
496 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
497 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
498 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
500 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
501 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
504 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
505 btrfs_mark_buffer_dirty(leaf);
507 btrfs_free_path(path);
511 static int find_next_chunk(struct btrfs_fs_info *fs_info, u64 *offset)
513 struct btrfs_root *root = fs_info->chunk_root;
514 struct btrfs_path *path;
516 struct btrfs_key key;
517 struct btrfs_chunk *chunk;
518 struct btrfs_key found_key;
520 path = btrfs_alloc_path();
524 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
525 key.offset = (u64)-1;
526 key.type = BTRFS_CHUNK_ITEM_KEY;
528 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
534 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
538 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
540 if (found_key.objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID)
543 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
545 *offset = found_key.offset +
546 btrfs_chunk_length(path->nodes[0], chunk);
551 btrfs_free_path(path);
555 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
559 struct btrfs_key key;
560 struct btrfs_key found_key;
562 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
563 key.type = BTRFS_DEV_ITEM_KEY;
564 key.offset = (u64)-1;
566 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
572 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
577 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
579 *objectid = found_key.offset + 1;
583 btrfs_release_path(path);
588 * the device information is stored in the chunk root
589 * the btrfs_device struct should be fully filled in
591 int btrfs_add_device(struct btrfs_trans_handle *trans,
592 struct btrfs_fs_info *fs_info,
593 struct btrfs_device *device)
596 struct btrfs_path *path;
597 struct btrfs_dev_item *dev_item;
598 struct extent_buffer *leaf;
599 struct btrfs_key key;
600 struct btrfs_root *root = fs_info->chunk_root;
604 path = btrfs_alloc_path();
608 ret = find_next_devid(root, path, &free_devid);
612 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
613 key.type = BTRFS_DEV_ITEM_KEY;
614 key.offset = free_devid;
616 ret = btrfs_insert_empty_item(trans, root, path, &key,
621 leaf = path->nodes[0];
622 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
624 device->devid = free_devid;
625 btrfs_set_device_id(leaf, dev_item, device->devid);
626 btrfs_set_device_generation(leaf, dev_item, 0);
627 btrfs_set_device_type(leaf, dev_item, device->type);
628 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
629 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
630 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
631 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
632 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
633 btrfs_set_device_group(leaf, dev_item, 0);
634 btrfs_set_device_seek_speed(leaf, dev_item, 0);
635 btrfs_set_device_bandwidth(leaf, dev_item, 0);
636 btrfs_set_device_start_offset(leaf, dev_item, 0);
638 ptr = (unsigned long)btrfs_device_uuid(dev_item);
639 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
640 ptr = (unsigned long)btrfs_device_fsid(dev_item);
641 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
642 btrfs_mark_buffer_dirty(leaf);
646 btrfs_free_path(path);
650 int btrfs_update_device(struct btrfs_trans_handle *trans,
651 struct btrfs_device *device)
654 struct btrfs_path *path;
655 struct btrfs_root *root;
656 struct btrfs_dev_item *dev_item;
657 struct extent_buffer *leaf;
658 struct btrfs_key key;
660 root = device->dev_root->fs_info->chunk_root;
662 path = btrfs_alloc_path();
666 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
667 key.type = BTRFS_DEV_ITEM_KEY;
668 key.offset = device->devid;
670 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
679 leaf = path->nodes[0];
680 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
682 btrfs_set_device_id(leaf, dev_item, device->devid);
683 btrfs_set_device_type(leaf, dev_item, device->type);
684 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
685 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
686 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
687 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
688 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
689 btrfs_mark_buffer_dirty(leaf);
692 btrfs_free_path(path);
696 int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
697 struct btrfs_chunk *chunk, int item_size)
699 struct btrfs_super_block *super_copy = fs_info->super_copy;
700 struct btrfs_disk_key disk_key;
704 array_size = btrfs_super_sys_array_size(super_copy);
705 if (array_size + item_size + sizeof(disk_key)
706 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
709 ptr = super_copy->sys_chunk_array + array_size;
710 btrfs_cpu_key_to_disk(&disk_key, key);
711 memcpy(ptr, &disk_key, sizeof(disk_key));
712 ptr += sizeof(disk_key);
713 memcpy(ptr, chunk, item_size);
714 item_size += sizeof(disk_key);
715 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
719 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
722 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
724 else if (type & BTRFS_BLOCK_GROUP_RAID10)
725 return calc_size * (num_stripes / sub_stripes);
726 else if (type & BTRFS_BLOCK_GROUP_RAID5)
727 return calc_size * (num_stripes - 1);
728 else if (type & BTRFS_BLOCK_GROUP_RAID6)
729 return calc_size * (num_stripes - 2);
731 return calc_size * num_stripes;
735 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
737 /* TODO, add a way to store the preferred stripe size */
738 return BTRFS_STRIPE_LEN;
742 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
744 * It is not equal to "device->total_bytes - device->bytes_used".
745 * We do not allocate any chunk in 1M at beginning of device, and not
746 * allowed to allocate any chunk before alloc_start if it is specified.
747 * So search holes from max(1M, alloc_start) to device->total_bytes.
749 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
750 struct btrfs_device *device,
753 struct btrfs_path *path;
754 struct btrfs_root *root = device->dev_root;
755 struct btrfs_key key;
756 struct btrfs_dev_extent *dev_extent = NULL;
757 struct extent_buffer *l;
758 u64 search_start = root->fs_info->alloc_start;
759 u64 search_end = device->total_bytes;
765 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
767 path = btrfs_alloc_path();
771 key.objectid = device->devid;
772 key.offset = root->fs_info->alloc_start;
773 key.type = BTRFS_DEV_EXTENT_KEY;
776 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
779 ret = btrfs_previous_item(root, path, 0, key.type);
785 slot = path->slots[0];
786 if (slot >= btrfs_header_nritems(l)) {
787 ret = btrfs_next_leaf(root, path);
794 btrfs_item_key_to_cpu(l, &key, slot);
796 if (key.objectid < device->devid)
798 if (key.objectid > device->devid)
800 if (key.type != BTRFS_DEV_EXTENT_KEY)
802 if (key.offset > search_end)
804 if (key.offset > search_start)
805 free_bytes += key.offset - search_start;
807 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
808 extent_end = key.offset + btrfs_dev_extent_length(l,
810 if (extent_end > search_start)
811 search_start = extent_end;
812 if (search_start > search_end)
819 if (search_start < search_end)
820 free_bytes += search_end - search_start;
822 *avail_bytes = free_bytes;
825 btrfs_free_path(path);
829 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r->fs_info) \
830 - sizeof(struct btrfs_item) \
831 - sizeof(struct btrfs_chunk)) \
832 / sizeof(struct btrfs_stripe) + 1)
834 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
835 - 2 * sizeof(struct btrfs_disk_key) \
836 - 2 * sizeof(struct btrfs_chunk)) \
837 / sizeof(struct btrfs_stripe) + 1)
839 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
840 struct btrfs_fs_info *info, u64 *start,
841 u64 *num_bytes, u64 type)
844 struct btrfs_root *extent_root = info->extent_root;
845 struct btrfs_root *chunk_root = info->chunk_root;
846 struct btrfs_stripe *stripes;
847 struct btrfs_device *device = NULL;
848 struct btrfs_chunk *chunk;
849 struct list_head private_devs;
850 struct list_head *dev_list = &info->fs_devices->devices;
851 struct list_head *cur;
852 struct map_lookup *map;
853 int min_stripe_size = SZ_1M;
854 u64 calc_size = SZ_8M;
856 u64 max_chunk_size = 4 * calc_size;
867 int stripe_len = BTRFS_STRIPE_LEN;
868 struct btrfs_key key;
871 if (list_empty(dev_list)) {
875 if (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
876 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
878 max_chunk_size = calc_size * 2;
879 min_stripe_size = SZ_1M;
880 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
881 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
883 max_chunk_size = 10 * calc_size;
884 min_stripe_size = SZ_64M;
885 max_stripes = BTRFS_MAX_DEVS(chunk_root);
886 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
888 max_chunk_size = 4 * calc_size;
889 min_stripe_size = SZ_32M;
890 max_stripes = BTRFS_MAX_DEVS(chunk_root);
893 if (type & BTRFS_BLOCK_GROUP_RAID1) {
894 num_stripes = min_t(u64, 2,
895 btrfs_super_num_devices(info->super_copy));
900 if (type & BTRFS_BLOCK_GROUP_DUP) {
904 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
905 num_stripes = btrfs_super_num_devices(info->super_copy);
906 if (num_stripes > max_stripes)
907 num_stripes = max_stripes;
910 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
911 num_stripes = btrfs_super_num_devices(info->super_copy);
912 if (num_stripes > max_stripes)
913 num_stripes = max_stripes;
916 num_stripes &= ~(u32)1;
920 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
921 num_stripes = btrfs_super_num_devices(info->super_copy);
922 if (num_stripes > max_stripes)
923 num_stripes = max_stripes;
927 stripe_len = find_raid56_stripe_len(num_stripes - 1,
928 btrfs_super_stripesize(info->super_copy));
930 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
931 num_stripes = btrfs_super_num_devices(info->super_copy);
932 if (num_stripes > max_stripes)
933 num_stripes = max_stripes;
937 stripe_len = find_raid56_stripe_len(num_stripes - 2,
938 btrfs_super_stripesize(info->super_copy));
941 /* we don't want a chunk larger than 10% of the FS */
942 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
943 max_chunk_size = min(percent_max, max_chunk_size);
946 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
948 calc_size = max_chunk_size;
949 calc_size /= num_stripes;
950 calc_size /= stripe_len;
951 calc_size *= stripe_len;
953 /* we don't want tiny stripes */
954 calc_size = max_t(u64, calc_size, min_stripe_size);
956 calc_size /= stripe_len;
957 calc_size *= stripe_len;
958 INIT_LIST_HEAD(&private_devs);
959 cur = dev_list->next;
962 if (type & BTRFS_BLOCK_GROUP_DUP)
963 min_free = calc_size * 2;
965 min_free = calc_size;
967 /* build a private list of devices we will allocate from */
968 while(index < num_stripes) {
969 device = list_entry(cur, struct btrfs_device, dev_list);
970 ret = btrfs_device_avail_bytes(trans, device, &avail);
974 if (avail >= min_free) {
975 list_move_tail(&device->dev_list, &private_devs);
977 if (type & BTRFS_BLOCK_GROUP_DUP)
979 } else if (avail > max_avail)
984 if (index < num_stripes) {
985 list_splice(&private_devs, dev_list);
986 if (index >= min_stripes) {
988 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
989 num_stripes /= sub_stripes;
990 num_stripes *= sub_stripes;
995 if (!looped && max_avail > 0) {
997 calc_size = max_avail;
1002 ret = find_next_chunk(info, &offset);
1005 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1006 key.type = BTRFS_CHUNK_ITEM_KEY;
1007 key.offset = offset;
1009 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1013 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1019 stripes = &chunk->stripe;
1020 *num_bytes = chunk_bytes_by_type(type, calc_size,
1021 num_stripes, sub_stripes);
1023 while(index < num_stripes) {
1024 struct btrfs_stripe *stripe;
1025 BUG_ON(list_empty(&private_devs));
1026 cur = private_devs.next;
1027 device = list_entry(cur, struct btrfs_device, dev_list);
1029 /* loop over this device again if we're doing a dup group */
1030 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1031 (index == num_stripes - 1))
1032 list_move_tail(&device->dev_list, dev_list);
1034 ret = btrfs_alloc_dev_extent(trans, device, key.offset,
1035 calc_size, &dev_offset, 0);
1039 device->bytes_used += calc_size;
1040 ret = btrfs_update_device(trans, device);
1044 map->stripes[index].dev = device;
1045 map->stripes[index].physical = dev_offset;
1046 stripe = stripes + index;
1047 btrfs_set_stack_stripe_devid(stripe, device->devid);
1048 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1049 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1052 BUG_ON(!list_empty(&private_devs));
1054 /* key was set above */
1055 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1056 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1057 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1058 btrfs_set_stack_chunk_type(chunk, type);
1059 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1060 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1061 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1062 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1063 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1064 map->sector_size = info->sectorsize;
1065 map->stripe_len = stripe_len;
1066 map->io_align = stripe_len;
1067 map->io_width = stripe_len;
1069 map->num_stripes = num_stripes;
1070 map->sub_stripes = sub_stripes;
1072 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1073 btrfs_chunk_item_size(num_stripes));
1075 *start = key.offset;;
1077 map->ce.start = key.offset;
1078 map->ce.size = *num_bytes;
1080 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1084 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1085 ret = btrfs_add_system_chunk(info, &key,
1086 chunk, btrfs_chunk_item_size(num_stripes));
1102 * Alloc a DATA chunk with SINGLE profile.
1104 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1105 * (btrfs logical bytenr == on-disk bytenr)
1106 * For that case, caller must make sure the chunk and dev_extent are not
1109 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1110 struct btrfs_fs_info *info, u64 *start,
1111 u64 num_bytes, u64 type, int convert)
1114 struct btrfs_root *extent_root = info->extent_root;
1115 struct btrfs_root *chunk_root = info->chunk_root;
1116 struct btrfs_stripe *stripes;
1117 struct btrfs_device *device = NULL;
1118 struct btrfs_chunk *chunk;
1119 struct list_head *dev_list = &info->fs_devices->devices;
1120 struct list_head *cur;
1121 struct map_lookup *map;
1122 u64 calc_size = SZ_8M;
1123 int num_stripes = 1;
1124 int sub_stripes = 0;
1127 int stripe_len = BTRFS_STRIPE_LEN;
1128 struct btrfs_key key;
1130 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1131 key.type = BTRFS_CHUNK_ITEM_KEY;
1133 if (*start != round_down(*start, info->sectorsize)) {
1134 error("DATA chunk start not sectorsize aligned: %llu",
1135 (unsigned long long)*start);
1138 key.offset = *start;
1139 dev_offset = *start;
1143 ret = find_next_chunk(info, &tmp);
1149 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1153 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1159 stripes = &chunk->stripe;
1160 calc_size = num_bytes;
1163 cur = dev_list->next;
1164 device = list_entry(cur, struct btrfs_device, dev_list);
1166 while (index < num_stripes) {
1167 struct btrfs_stripe *stripe;
1169 ret = btrfs_alloc_dev_extent(trans, device, key.offset,
1170 calc_size, &dev_offset, convert);
1173 device->bytes_used += calc_size;
1174 ret = btrfs_update_device(trans, device);
1177 map->stripes[index].dev = device;
1178 map->stripes[index].physical = dev_offset;
1179 stripe = stripes + index;
1180 btrfs_set_stack_stripe_devid(stripe, device->devid);
1181 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1182 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1186 /* key was set above */
1187 btrfs_set_stack_chunk_length(chunk, num_bytes);
1188 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1189 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1190 btrfs_set_stack_chunk_type(chunk, type);
1191 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1192 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1193 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1194 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1195 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1196 map->sector_size = info->sectorsize;
1197 map->stripe_len = stripe_len;
1198 map->io_align = stripe_len;
1199 map->io_width = stripe_len;
1201 map->num_stripes = num_stripes;
1202 map->sub_stripes = sub_stripes;
1204 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1205 btrfs_chunk_item_size(num_stripes));
1208 *start = key.offset;
1210 map->ce.start = key.offset;
1211 map->ce.size = num_bytes;
1213 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1220 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
1222 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1223 struct cache_extent *ce;
1224 struct map_lookup *map;
1227 ce = search_cache_extent(&map_tree->cache_tree, logical);
1229 fprintf(stderr, "No mapping for %llu-%llu\n",
1230 (unsigned long long)logical,
1231 (unsigned long long)logical+len);
1234 if (ce->start > logical || ce->start + ce->size < logical) {
1235 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1236 "%llu-%llu\n", (unsigned long long)logical,
1237 (unsigned long long)logical+len,
1238 (unsigned long long)ce->start,
1239 (unsigned long long)ce->start + ce->size);
1242 map = container_of(ce, struct map_lookup, ce);
1244 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1245 ret = map->num_stripes;
1246 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1247 ret = map->sub_stripes;
1248 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1250 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1257 int btrfs_next_bg(struct btrfs_fs_info *fs_info, u64 *logical,
1258 u64 *size, u64 type)
1260 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1261 struct cache_extent *ce;
1262 struct map_lookup *map;
1265 ce = search_cache_extent(&map_tree->cache_tree, cur);
1269 * only jump to next bg if our cur is not 0
1270 * As the initial logical for btrfs_next_bg() is 0, and
1271 * if we jump to next bg, we skipped a valid bg.
1274 ce = next_cache_extent(ce);
1280 map = container_of(ce, struct map_lookup, ce);
1281 if (map->type & type) {
1282 *logical = ce->start;
1287 ce = next_cache_extent(ce);
1293 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
1294 u64 chunk_start, u64 physical, u64 devid,
1295 u64 **logical, int *naddrs, int *stripe_len)
1297 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1298 struct cache_extent *ce;
1299 struct map_lookup *map;
1307 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1309 map = container_of(ce, struct map_lookup, ce);
1312 rmap_len = map->stripe_len;
1313 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1314 length = ce->size / (map->num_stripes / map->sub_stripes);
1315 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1316 length = ce->size / map->num_stripes;
1317 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1318 BTRFS_BLOCK_GROUP_RAID6)) {
1319 length = ce->size / nr_data_stripes(map);
1320 rmap_len = map->stripe_len * nr_data_stripes(map);
1323 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1325 for (i = 0; i < map->num_stripes; i++) {
1326 if (devid && map->stripes[i].dev->devid != devid)
1328 if (map->stripes[i].physical > physical ||
1329 map->stripes[i].physical + length <= physical)
1332 stripe_nr = (physical - map->stripes[i].physical) /
1335 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1336 stripe_nr = (stripe_nr * map->num_stripes + i) /
1338 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1339 stripe_nr = stripe_nr * map->num_stripes + i;
1340 } /* else if RAID[56], multiply by nr_data_stripes().
1341 * Alternatively, just use rmap_len below instead of
1342 * map->stripe_len */
1344 bytenr = ce->start + stripe_nr * rmap_len;
1345 for (j = 0; j < nr; j++) {
1346 if (buf[j] == bytenr)
1355 *stripe_len = rmap_len;
1360 static inline int parity_smaller(u64 a, u64 b)
1365 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1366 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1368 struct btrfs_bio_stripe s;
1375 for (i = 0; i < bbio->num_stripes - 1; i++) {
1376 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1377 s = bbio->stripes[i];
1379 bbio->stripes[i] = bbio->stripes[i+1];
1380 raid_map[i] = raid_map[i+1];
1381 bbio->stripes[i+1] = s;
1389 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1390 u64 logical, u64 *length,
1391 struct btrfs_multi_bio **multi_ret, int mirror_num,
1394 return __btrfs_map_block(fs_info, rw, logical, length, NULL,
1395 multi_ret, mirror_num, raid_map_ret);
1398 int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1399 u64 logical, u64 *length, u64 *type,
1400 struct btrfs_multi_bio **multi_ret, int mirror_num,
1403 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1404 struct cache_extent *ce;
1405 struct map_lookup *map;
1409 u64 *raid_map = NULL;
1410 int stripes_allocated = 8;
1411 int stripes_required = 1;
1414 struct btrfs_multi_bio *multi = NULL;
1416 if (multi_ret && rw == READ) {
1417 stripes_allocated = 1;
1420 ce = search_cache_extent(&map_tree->cache_tree, logical);
1426 if (ce->start > logical) {
1428 *length = ce->start - logical;
1433 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1438 map = container_of(ce, struct map_lookup, ce);
1439 offset = logical - ce->start;
1442 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1443 BTRFS_BLOCK_GROUP_DUP)) {
1444 stripes_required = map->num_stripes;
1445 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1446 stripes_required = map->sub_stripes;
1449 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1450 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1451 /* RAID[56] write or recovery. Return all stripes */
1452 stripes_required = map->num_stripes;
1454 /* Only allocate the map if we've already got a large enough multi_ret */
1455 if (stripes_allocated >= stripes_required) {
1456 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1464 /* if our multi bio struct is too small, back off and try again */
1465 if (multi_ret && stripes_allocated < stripes_required) {
1466 stripes_allocated = stripes_required;
1473 * stripe_nr counts the total number of stripes we have to stride
1474 * to get to this block
1476 stripe_nr = stripe_nr / map->stripe_len;
1478 stripe_offset = stripe_nr * map->stripe_len;
1479 BUG_ON(offset < stripe_offset);
1481 /* stripe_offset is the offset of this block in its stripe*/
1482 stripe_offset = offset - stripe_offset;
1484 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1485 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1486 BTRFS_BLOCK_GROUP_RAID10 |
1487 BTRFS_BLOCK_GROUP_DUP)) {
1488 /* we limit the length of each bio to what fits in a stripe */
1489 *length = min_t(u64, ce->size - offset,
1490 map->stripe_len - stripe_offset);
1492 *length = ce->size - offset;
1498 multi->num_stripes = 1;
1500 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1502 multi->num_stripes = map->num_stripes;
1503 else if (mirror_num)
1504 stripe_index = mirror_num - 1;
1506 stripe_index = stripe_nr % map->num_stripes;
1507 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1508 int factor = map->num_stripes / map->sub_stripes;
1510 stripe_index = stripe_nr % factor;
1511 stripe_index *= map->sub_stripes;
1514 multi->num_stripes = map->sub_stripes;
1515 else if (mirror_num)
1516 stripe_index += mirror_num - 1;
1518 stripe_nr = stripe_nr / factor;
1519 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1521 multi->num_stripes = map->num_stripes;
1522 else if (mirror_num)
1523 stripe_index = mirror_num - 1;
1524 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1525 BTRFS_BLOCK_GROUP_RAID6)) {
1530 u64 raid56_full_stripe_start;
1531 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1534 * align the start of our data stripe in the logical
1537 raid56_full_stripe_start = offset / full_stripe_len;
1538 raid56_full_stripe_start *= full_stripe_len;
1540 /* get the data stripe number */
1541 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1542 stripe_nr = stripe_nr / nr_data_stripes(map);
1544 /* Work out the disk rotation on this stripe-set */
1545 rot = stripe_nr % map->num_stripes;
1547 /* Fill in the logical address of each stripe */
1548 tmp = stripe_nr * nr_data_stripes(map);
1550 for (i = 0; i < nr_data_stripes(map); i++)
1551 raid_map[(i+rot) % map->num_stripes] =
1552 ce->start + (tmp + i) * map->stripe_len;
1554 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1555 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1556 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1558 *length = map->stripe_len;
1561 multi->num_stripes = map->num_stripes;
1563 stripe_index = stripe_nr % nr_data_stripes(map);
1564 stripe_nr = stripe_nr / nr_data_stripes(map);
1567 * Mirror #0 or #1 means the original data block.
1568 * Mirror #2 is RAID5 parity block.
1569 * Mirror #3 is RAID6 Q block.
1572 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1574 /* We distribute the parity blocks across stripes */
1575 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1579 * after this do_div call, stripe_nr is the number of stripes
1580 * on this device we have to walk to find the data, and
1581 * stripe_index is the number of our device in the stripe array
1583 stripe_index = stripe_nr % map->num_stripes;
1584 stripe_nr = stripe_nr / map->num_stripes;
1586 BUG_ON(stripe_index >= map->num_stripes);
1588 for (i = 0; i < multi->num_stripes; i++) {
1589 multi->stripes[i].physical =
1590 map->stripes[stripe_index].physical + stripe_offset +
1591 stripe_nr * map->stripe_len;
1592 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1601 sort_parity_stripes(multi, raid_map);
1602 *raid_map_ret = raid_map;
1608 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
1611 struct btrfs_device *device;
1612 struct btrfs_fs_devices *cur_devices;
1614 cur_devices = fs_info->fs_devices;
1615 while (cur_devices) {
1617 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1618 fs_info->ignore_fsid_mismatch)) {
1619 device = __find_device(&cur_devices->devices,
1624 cur_devices = cur_devices->seed;
1629 struct btrfs_device *
1630 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1631 u64 devid, int instance)
1633 struct list_head *head = &fs_devices->devices;
1634 struct btrfs_device *dev;
1637 list_for_each_entry(dev, head, dev_list) {
1638 if (dev->devid == devid && num_found++ == instance)
1644 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
1646 struct cache_extent *ce;
1647 struct map_lookup *map;
1648 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1653 * During chunk recovering, we may fail to find block group's
1654 * corresponding chunk, we will rebuild it later
1656 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1657 if (!fs_info->is_chunk_recover)
1662 map = container_of(ce, struct map_lookup, ce);
1663 for (i = 0; i < map->num_stripes; i++) {
1664 if (!map->stripes[i].dev->writeable) {
1673 static struct btrfs_device *fill_missing_device(u64 devid)
1675 struct btrfs_device *device;
1677 device = kzalloc(sizeof(*device), GFP_NOFS);
1678 device->devid = devid;
1684 * slot == -1: SYSTEM chunk
1685 * return -EIO on error, otherwise return 0
1687 int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
1688 struct extent_buffer *leaf,
1689 struct btrfs_chunk *chunk,
1690 int slot, u64 logical)
1697 u32 chunk_ondisk_size;
1698 u32 sectorsize = fs_info->sectorsize;
1700 length = btrfs_chunk_length(leaf, chunk);
1701 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1702 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1703 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1704 type = btrfs_chunk_type(leaf, chunk);
1707 * These valid checks may be insufficient to cover every corner cases.
1709 if (!IS_ALIGNED(logical, sectorsize)) {
1710 error("invalid chunk logical %llu", logical);
1713 if (btrfs_chunk_sector_size(leaf, chunk) != sectorsize) {
1714 error("invalid chunk sectorsize %llu",
1715 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1718 if (!length || !IS_ALIGNED(length, sectorsize)) {
1719 error("invalid chunk length %llu", length);
1722 if (stripe_len != BTRFS_STRIPE_LEN) {
1723 error("invalid chunk stripe length: %llu", stripe_len);
1726 /* Check on chunk item type */
1727 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1728 error("invalid chunk type %llu", type);
1731 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1732 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1733 error("unrecognized chunk type: %llu",
1734 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1735 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1738 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1739 error("missing chunk type flag: %llu", type);
1742 if (!(is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) ||
1743 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0)) {
1744 error("conflicting chunk type detected: %llu", type);
1747 if ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
1748 !is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1749 error("conflicting chunk profile detected: %llu", type);
1753 chunk_ondisk_size = btrfs_chunk_item_size(num_stripes);
1755 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1756 * it can't exceed the system chunk array size
1757 * For normal chunk, it should match its chunk item size.
1759 if (num_stripes < 1 ||
1760 (slot == -1 && chunk_ondisk_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1761 (slot >= 0 && chunk_ondisk_size > btrfs_item_size_nr(leaf, slot))) {
1762 error("invalid num_stripes: %u", num_stripes);
1766 * Device number check against profile
1768 if ((type & BTRFS_BLOCK_GROUP_RAID10 && (sub_stripes != 2 ||
1769 !IS_ALIGNED(num_stripes, sub_stripes))) ||
1770 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1771 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1772 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1773 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1774 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1775 num_stripes != 1)) {
1776 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1777 num_stripes, sub_stripes,
1778 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1786 * Slot is used to verify the chunk item is valid
1788 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1790 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
1791 struct extent_buffer *leaf,
1792 struct btrfs_chunk *chunk, int slot)
1794 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1795 struct map_lookup *map;
1796 struct cache_extent *ce;
1800 u8 uuid[BTRFS_UUID_SIZE];
1805 logical = key->offset;
1806 length = btrfs_chunk_length(leaf, chunk);
1807 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1808 /* Validation check */
1809 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, slot, logical);
1811 error("%s checksums match, but it has an invalid chunk, %s",
1812 (slot == -1) ? "Superblock" : "Metadata",
1813 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1817 ce = search_cache_extent(&map_tree->cache_tree, logical);
1819 /* already mapped? */
1820 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1824 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1828 map->ce.start = logical;
1829 map->ce.size = length;
1830 map->num_stripes = num_stripes;
1831 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1832 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1833 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1834 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1835 map->type = btrfs_chunk_type(leaf, chunk);
1836 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1838 for (i = 0; i < num_stripes; i++) {
1839 map->stripes[i].physical =
1840 btrfs_stripe_offset_nr(leaf, chunk, i);
1841 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1842 read_extent_buffer(leaf, uuid, (unsigned long)
1843 btrfs_stripe_dev_uuid_nr(chunk, i),
1845 map->stripes[i].dev = btrfs_find_device(fs_info, devid, uuid,
1847 if (!map->stripes[i].dev) {
1848 map->stripes[i].dev = fill_missing_device(devid);
1849 printf("warning, device %llu is missing\n",
1850 (unsigned long long)devid);
1851 list_add(&map->stripes[i].dev->dev_list,
1852 &fs_info->fs_devices->devices);
1856 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1862 static int fill_device_from_item(struct extent_buffer *leaf,
1863 struct btrfs_dev_item *dev_item,
1864 struct btrfs_device *device)
1868 device->devid = btrfs_device_id(leaf, dev_item);
1869 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1870 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1871 device->type = btrfs_device_type(leaf, dev_item);
1872 device->io_align = btrfs_device_io_align(leaf, dev_item);
1873 device->io_width = btrfs_device_io_width(leaf, dev_item);
1874 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1876 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1877 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1882 static int open_seed_devices(struct btrfs_fs_info *fs_info, u8 *fsid)
1884 struct btrfs_fs_devices *fs_devices;
1887 fs_devices = fs_info->fs_devices->seed;
1888 while (fs_devices) {
1889 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1893 fs_devices = fs_devices->seed;
1896 fs_devices = find_fsid(fsid);
1898 /* missing all seed devices */
1899 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1904 INIT_LIST_HEAD(&fs_devices->devices);
1905 list_add(&fs_devices->list, &fs_uuids);
1906 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1909 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1913 fs_devices->seed = fs_info->fs_devices->seed;
1914 fs_info->fs_devices->seed = fs_devices;
1919 static int read_one_dev(struct btrfs_fs_info *fs_info,
1920 struct extent_buffer *leaf,
1921 struct btrfs_dev_item *dev_item)
1923 struct btrfs_device *device;
1926 u8 fs_uuid[BTRFS_UUID_SIZE];
1927 u8 dev_uuid[BTRFS_UUID_SIZE];
1929 devid = btrfs_device_id(leaf, dev_item);
1930 read_extent_buffer(leaf, dev_uuid,
1931 (unsigned long)btrfs_device_uuid(dev_item),
1933 read_extent_buffer(leaf, fs_uuid,
1934 (unsigned long)btrfs_device_fsid(dev_item),
1937 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
1938 ret = open_seed_devices(fs_info, fs_uuid);
1943 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
1945 device = kzalloc(sizeof(*device), GFP_NOFS);
1949 list_add(&device->dev_list,
1950 &fs_info->fs_devices->devices);
1953 fill_device_from_item(leaf, dev_item, device);
1954 device->dev_root = fs_info->dev_root;
1958 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
1960 struct btrfs_super_block *super_copy = fs_info->super_copy;
1961 struct extent_buffer *sb;
1962 struct btrfs_disk_key *disk_key;
1963 struct btrfs_chunk *chunk;
1965 unsigned long sb_array_offset;
1971 struct btrfs_key key;
1973 if (fs_info->nodesize < BTRFS_SUPER_INFO_SIZE) {
1974 printf("ERROR: nodesize %u too small to read superblock\n",
1978 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
1981 btrfs_set_buffer_uptodate(sb);
1982 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1983 array_size = btrfs_super_sys_array_size(super_copy);
1985 array_ptr = super_copy->sys_chunk_array;
1986 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1989 while (cur_offset < array_size) {
1990 disk_key = (struct btrfs_disk_key *)array_ptr;
1991 len = sizeof(*disk_key);
1992 if (cur_offset + len > array_size)
1993 goto out_short_read;
1995 btrfs_disk_key_to_cpu(&key, disk_key);
1998 sb_array_offset += len;
2001 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2002 chunk = (struct btrfs_chunk *)sb_array_offset;
2004 * At least one btrfs_chunk with one stripe must be
2005 * present, exact stripe count check comes afterwards
2007 len = btrfs_chunk_item_size(1);
2008 if (cur_offset + len > array_size)
2009 goto out_short_read;
2011 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2014 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
2015 num_stripes, cur_offset);
2020 len = btrfs_chunk_item_size(num_stripes);
2021 if (cur_offset + len > array_size)
2022 goto out_short_read;
2024 ret = read_one_chunk(fs_info, &key, sb, chunk, -1);
2029 "ERROR: unexpected item type %u in sys_array at offset %u\n",
2030 (u32)key.type, cur_offset);
2035 sb_array_offset += len;
2038 free_extent_buffer(sb);
2042 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
2044 free_extent_buffer(sb);
2048 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
2050 struct btrfs_path *path;
2051 struct extent_buffer *leaf;
2052 struct btrfs_key key;
2053 struct btrfs_key found_key;
2054 struct btrfs_root *root = fs_info->chunk_root;
2058 path = btrfs_alloc_path();
2063 * Read all device items, and then all the chunk items. All
2064 * device items are found before any chunk item (their object id
2065 * is smaller than the lowest possible object id for a chunk
2066 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
2068 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2071 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2075 leaf = path->nodes[0];
2076 slot = path->slots[0];
2077 if (slot >= btrfs_header_nritems(leaf)) {
2078 ret = btrfs_next_leaf(root, path);
2085 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2086 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2087 struct btrfs_dev_item *dev_item;
2088 dev_item = btrfs_item_ptr(leaf, slot,
2089 struct btrfs_dev_item);
2090 ret = read_one_dev(fs_info, leaf, dev_item);
2092 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2093 struct btrfs_chunk *chunk;
2094 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2095 ret = read_one_chunk(fs_info, &found_key, leaf, chunk,
2104 btrfs_free_path(path);
2108 struct list_head *btrfs_scanned_uuids(void)
2113 static int rmw_eb(struct btrfs_fs_info *info,
2114 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2117 unsigned long orig_off = 0;
2118 unsigned long dest_off = 0;
2119 unsigned long copy_len = eb->len;
2121 ret = read_whole_eb(info, eb, 0);
2125 if (eb->start + eb->len <= orig_eb->start ||
2126 eb->start >= orig_eb->start + orig_eb->len)
2129 * | ----- orig_eb ------- |
2130 * | ----- stripe ------- |
2131 * | ----- orig_eb ------- |
2132 * | ----- orig_eb ------- |
2134 if (eb->start > orig_eb->start)
2135 orig_off = eb->start - orig_eb->start;
2136 if (orig_eb->start > eb->start)
2137 dest_off = orig_eb->start - eb->start;
2139 if (copy_len > orig_eb->len - orig_off)
2140 copy_len = orig_eb->len - orig_off;
2141 if (copy_len > eb->len - dest_off)
2142 copy_len = eb->len - dest_off;
2144 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2148 static int split_eb_for_raid56(struct btrfs_fs_info *info,
2149 struct extent_buffer *orig_eb,
2150 struct extent_buffer **ebs,
2151 u64 stripe_len, u64 *raid_map,
2154 struct extent_buffer **tmp_ebs;
2155 u64 start = orig_eb->start;
2160 tmp_ebs = calloc(num_stripes, sizeof(*tmp_ebs));
2164 /* Alloc memory in a row for data stripes */
2165 for (i = 0; i < num_stripes; i++) {
2166 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2169 tmp_ebs[i] = calloc(1, sizeof(**tmp_ebs) + stripe_len);
2176 for (i = 0; i < num_stripes; i++) {
2177 struct extent_buffer *eb = tmp_ebs[i];
2179 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2182 eb->start = raid_map[i];
2183 eb->len = stripe_len;
2187 eb->dev_bytenr = (u64)-1;
2189 this_eb_start = raid_map[i];
2191 if (start > this_eb_start ||
2192 start + orig_eb->len < this_eb_start + stripe_len) {
2193 ret = rmw_eb(info, eb, orig_eb);
2197 memcpy(eb->data, orig_eb->data + eb->start - start,
2205 for (i = 0; i < num_stripes; i++)
2211 int write_raid56_with_parity(struct btrfs_fs_info *info,
2212 struct extent_buffer *eb,
2213 struct btrfs_multi_bio *multi,
2214 u64 stripe_len, u64 *raid_map)
2216 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2219 int alloc_size = eb->len;
2222 ebs = malloc(sizeof(*ebs) * multi->num_stripes);
2223 pointers = malloc(sizeof(*pointers) * multi->num_stripes);
2224 if (!ebs || !pointers) {
2230 if (stripe_len > alloc_size)
2231 alloc_size = stripe_len;
2233 ret = split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2234 multi->num_stripes);
2238 for (i = 0; i < multi->num_stripes; i++) {
2239 struct extent_buffer *new_eb;
2240 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2241 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2242 ebs[i]->fd = multi->stripes[i].dev->fd;
2243 multi->stripes[i].dev->total_ios++;
2244 if (ebs[i]->start != raid_map[i]) {
2246 goto out_free_split;
2250 new_eb = malloc(sizeof(*eb) + alloc_size);
2253 goto out_free_split;
2255 new_eb->dev_bytenr = multi->stripes[i].physical;
2256 new_eb->fd = multi->stripes[i].dev->fd;
2257 multi->stripes[i].dev->total_ios++;
2258 new_eb->len = stripe_len;
2260 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2262 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2266 ebs[multi->num_stripes - 2] = p_eb;
2267 ebs[multi->num_stripes - 1] = q_eb;
2269 for (i = 0; i < multi->num_stripes; i++)
2270 pointers[i] = ebs[i]->data;
2272 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2274 ebs[multi->num_stripes - 1] = p_eb;
2275 for (i = 0; i < multi->num_stripes; i++)
2276 pointers[i] = ebs[i]->data;
2277 ret = raid5_gen_result(multi->num_stripes, stripe_len,
2278 multi->num_stripes - 1, pointers);
2280 goto out_free_split;
2283 for (i = 0; i < multi->num_stripes; i++) {
2284 ret = write_extent_to_disk(ebs[i]);
2286 goto out_free_split;
2290 for (i = 0; i < multi->num_stripes; i++) {
2302 * Get stripe length from chunk item and its stripe items
2304 * Caller should only call this function after validating the chunk item
2305 * by using btrfs_check_chunk_valid().
2307 u64 btrfs_stripe_length(struct btrfs_fs_info *fs_info,
2308 struct extent_buffer *leaf,
2309 struct btrfs_chunk *chunk)
2313 u32 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2314 u64 profile = btrfs_chunk_type(leaf, chunk) &
2315 BTRFS_BLOCK_GROUP_PROFILE_MASK;
2317 chunk_len = btrfs_chunk_length(leaf, chunk);
2320 case 0: /* Single profile */
2321 case BTRFS_BLOCK_GROUP_RAID1:
2322 case BTRFS_BLOCK_GROUP_DUP:
2323 stripe_len = chunk_len;
2325 case BTRFS_BLOCK_GROUP_RAID0:
2326 stripe_len = chunk_len / num_stripes;
2328 case BTRFS_BLOCK_GROUP_RAID5:
2329 stripe_len = chunk_len / (num_stripes - 1);
2331 case BTRFS_BLOCK_GROUP_RAID6:
2332 stripe_len = chunk_len / (num_stripes - 2);
2334 case BTRFS_BLOCK_GROUP_RAID10:
2335 stripe_len = chunk_len / (num_stripes /
2336 btrfs_chunk_sub_stripes(leaf, chunk));
2339 /* Invalid chunk profile found */
2346 * Return 0 if size of @device is already good
2347 * Return >0 if size of @device is not aligned but fixed without problems
2348 * Return <0 if something wrong happened when aligning the size of @device
2350 int btrfs_fix_device_size(struct btrfs_fs_info *fs_info,
2351 struct btrfs_device *device)
2353 struct btrfs_trans_handle *trans;
2354 struct btrfs_key key;
2355 struct btrfs_path path;
2356 struct btrfs_root *chunk_root = fs_info->chunk_root;
2357 struct btrfs_dev_item *di;
2358 u64 old_bytes = device->total_bytes;
2361 if (IS_ALIGNED(old_bytes, fs_info->sectorsize))
2364 /* Align the in-memory total_bytes first, and use it as correct size */
2365 device->total_bytes = round_down(device->total_bytes,
2366 fs_info->sectorsize);
2368 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2369 key.type = BTRFS_DEV_ITEM_KEY;
2370 key.offset = device->devid;
2372 trans = btrfs_start_transaction(chunk_root, 1);
2373 if (IS_ERR(trans)) {
2374 ret = PTR_ERR(trans);
2375 error("error starting transaction: %d (%s)",
2376 ret, strerror(-ret));
2380 btrfs_init_path(&path);
2381 ret = btrfs_search_slot(trans, chunk_root, &key, &path, 0, 1);
2383 error("failed to find DEV_ITEM for devid %llu", device->devid);
2388 error("failed to search chunk root: %d (%s)",
2389 ret, strerror(-ret));
2392 di = btrfs_item_ptr(path.nodes[0], path.slots[0], struct btrfs_dev_item);
2393 btrfs_set_device_total_bytes(path.nodes[0], di, device->total_bytes);
2394 btrfs_mark_buffer_dirty(path.nodes[0]);
2395 ret = btrfs_commit_transaction(trans, chunk_root);
2397 error("failed to commit current transaction: %d (%s)",
2398 ret, strerror(-ret));
2399 btrfs_release_path(&path);
2402 btrfs_release_path(&path);
2403 printf("Fixed device size for devid %llu, old size: %llu new size: %llu\n",
2404 device->devid, old_bytes, device->total_bytes);
2408 /* We haven't modified anything, it's OK to commit current trans */
2409 btrfs_commit_transaction(trans, chunk_root);
2410 btrfs_release_path(&path);
2415 * Return 0 if super block total_bytes matches all devices' total_bytes
2416 * Return >0 if super block total_bytes mismatch but fixed without problem
2417 * Return <0 if we failed to fix super block total_bytes
2419 int btrfs_fix_super_size(struct btrfs_fs_info *fs_info)
2421 struct btrfs_trans_handle *trans;
2422 struct btrfs_device *device;
2423 struct list_head *dev_list = &fs_info->fs_devices->devices;
2424 u64 total_bytes = 0;
2425 u64 old_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2428 list_for_each_entry(device, dev_list, dev_list) {
2430 * Caller should ensure this function is called after aligning
2431 * all devices' total_bytes.
2433 if (!IS_ALIGNED(device->total_bytes, fs_info->sectorsize)) {
2434 error("device %llu total_bytes %llu not aligned to %u",
2435 device->devid, device->total_bytes,
2436 fs_info->sectorsize);
2439 total_bytes += device->total_bytes;
2442 if (total_bytes == old_bytes)
2445 btrfs_set_super_total_bytes(fs_info->super_copy, total_bytes);
2447 /* Commit transaction to update all super blocks */
2448 trans = btrfs_start_transaction(fs_info->tree_root, 1);
2449 if (IS_ERR(trans)) {
2450 ret = PTR_ERR(trans);
2451 error("error starting transaction: %d (%s)",
2452 ret, strerror(-ret));
2455 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
2457 error("failed to commit current transaction: %d (%s)",
2458 ret, strerror(-ret));
2461 printf("Fixed super total bytes, old size: %llu new size: %llu\n",
2462 old_bytes, total_bytes);
2467 * Return 0 if all devices and super block sizes are good
2468 * Return >0 if any device/super size problem was found, but fixed
2469 * Return <0 if something wrong happened during fixing
2471 int btrfs_fix_device_and_super_size(struct btrfs_fs_info *fs_info)
2473 struct btrfs_device *device;
2474 struct list_head *dev_list = &fs_info->fs_devices->devices;
2475 bool have_bad_value = false;
2478 /* Seed device is not supported yet */
2479 if (fs_info->fs_devices->seed) {
2480 error("fixing device size with seed device is not supported yet");
2484 /* All devices must be set up before repairing */
2485 if (list_empty(dev_list)) {
2486 error("no device found");
2489 list_for_each_entry(device, dev_list, dev_list) {
2490 if (device->fd == -1 || !device->writeable) {
2491 error("devid %llu is missing or not writeable",
2494 "fixing device size needs all device(s) to be present and writeable");
2499 /* Repair total_bytes of each device */
2500 list_for_each_entry(device, dev_list, dev_list) {
2501 ret = btrfs_fix_device_size(fs_info, device);
2505 have_bad_value = true;
2508 /* Repair super total_byte */
2509 ret = btrfs_fix_super_size(fs_info);
2511 have_bad_value = true;
2512 if (have_bad_value) {
2514 "Fixed unaligned/mismatched total_bytes for super block and device items\n");
2517 printf("No device size related problem found\n");