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: %s",
185 device->devid, strerror(errno));
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': %s", device->name,
250 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
251 fprintf(stderr, "Warning, could not drop caches\n");
253 if (device->devid == fs_devices->latest_devid)
254 fs_devices->latest_bdev = fd;
255 if (device->devid == fs_devices->lowest_devid)
256 fs_devices->lowest_bdev = fd;
259 device->writeable = 1;
263 btrfs_close_devices(fs_devices);
267 int btrfs_scan_one_device(int fd, const char *path,
268 struct btrfs_fs_devices **fs_devices_ret,
269 u64 *total_devs, u64 super_offset, unsigned sbflags)
271 struct btrfs_super_block *disk_super;
272 char buf[BTRFS_SUPER_INFO_SIZE];
276 disk_super = (struct btrfs_super_block *)buf;
277 ret = btrfs_read_dev_super(fd, disk_super, super_offset, sbflags);
280 devid = btrfs_stack_device_id(&disk_super->dev_item);
281 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
284 *total_devs = btrfs_super_num_devices(disk_super);
286 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
292 * find_free_dev_extent_start - find free space in the specified device
293 * @device: the device which we search the free space in
294 * @num_bytes: the size of the free space that we need
295 * @search_start: the position from which to begin the search
296 * @start: store the start of the free space.
297 * @len: the size of the free space. that we find, or the size
298 * of the max free space if we don't find suitable free space
300 * this uses a pretty simple search, the expectation is that it is
301 * called very infrequently and that a given device has a small number
304 * @start is used to store the start of the free space if we find. But if we
305 * don't find suitable free space, it will be used to store the start position
306 * of the max free space.
308 * @len is used to store the size of the free space that we find.
309 * But if we don't find suitable free space, it is used to store the size of
310 * the max free space.
312 static int find_free_dev_extent_start(struct btrfs_trans_handle *trans,
313 struct btrfs_device *device, u64 num_bytes,
314 u64 search_start, u64 *start, u64 *len)
316 struct btrfs_key key;
317 struct btrfs_root *root = device->dev_root;
318 struct btrfs_dev_extent *dev_extent;
319 struct btrfs_path *path;
324 u64 search_end = device->total_bytes;
327 struct extent_buffer *l;
328 u64 min_search_start;
331 * We don't want to overwrite the superblock on the drive nor any area
332 * used by the boot loader (grub for example), so we make sure to start
333 * at an offset of at least 1MB.
335 min_search_start = max(root->fs_info->alloc_start, (u64)SZ_1M);
336 search_start = max(search_start, min_search_start);
338 path = btrfs_alloc_path();
342 max_hole_start = search_start;
345 if (search_start >= search_end) {
352 key.objectid = device->devid;
353 key.offset = search_start;
354 key.type = BTRFS_DEV_EXTENT_KEY;
356 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
360 ret = btrfs_previous_item(root, path, key.objectid, key.type);
367 slot = path->slots[0];
368 if (slot >= btrfs_header_nritems(l)) {
369 ret = btrfs_next_leaf(root, path);
377 btrfs_item_key_to_cpu(l, &key, slot);
379 if (key.objectid < device->devid)
382 if (key.objectid > device->devid)
385 if (key.type != BTRFS_DEV_EXTENT_KEY)
388 if (key.offset > search_start) {
389 hole_size = key.offset - search_start;
392 * Have to check before we set max_hole_start, otherwise
393 * we could end up sending back this offset anyway.
395 if (hole_size > max_hole_size) {
396 max_hole_start = search_start;
397 max_hole_size = hole_size;
401 * If this free space is greater than which we need,
402 * it must be the max free space that we have found
403 * until now, so max_hole_start must point to the start
404 * of this free space and the length of this free space
405 * is stored in max_hole_size. Thus, we return
406 * max_hole_start and max_hole_size and go back to the
409 if (hole_size >= num_bytes) {
415 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
416 extent_end = key.offset + btrfs_dev_extent_length(l,
418 if (extent_end > search_start)
419 search_start = extent_end;
426 * At this point, search_start should be the end of
427 * allocated dev extents, and when shrinking the device,
428 * search_end may be smaller than search_start.
430 if (search_end > search_start) {
431 hole_size = search_end - search_start;
433 if (hole_size > max_hole_size) {
434 max_hole_start = search_start;
435 max_hole_size = hole_size;
440 if (max_hole_size < num_bytes)
446 btrfs_free_path(path);
447 *start = max_hole_start;
449 *len = max_hole_size;
453 int find_free_dev_extent(struct btrfs_trans_handle *trans,
454 struct btrfs_device *device, u64 num_bytes,
457 /* FIXME use last free of some kind */
458 return find_free_dev_extent_start(trans, device,
459 num_bytes, 0, start, NULL);
462 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
463 struct btrfs_device *device,
464 u64 chunk_tree, u64 chunk_objectid,
466 u64 num_bytes, u64 *start, int convert)
469 struct btrfs_path *path;
470 struct btrfs_root *root = device->dev_root;
471 struct btrfs_dev_extent *extent;
472 struct extent_buffer *leaf;
473 struct btrfs_key key;
475 path = btrfs_alloc_path();
480 * For convert case, just skip search free dev_extent, as caller
481 * is responsible to make sure it's free.
484 ret = find_free_dev_extent(trans, device, num_bytes,
490 key.objectid = device->devid;
492 key.type = BTRFS_DEV_EXTENT_KEY;
493 ret = btrfs_insert_empty_item(trans, root, path, &key,
497 leaf = path->nodes[0];
498 extent = btrfs_item_ptr(leaf, path->slots[0],
499 struct btrfs_dev_extent);
500 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
501 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
502 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
504 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
505 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
508 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
509 btrfs_mark_buffer_dirty(leaf);
511 btrfs_free_path(path);
515 static int find_next_chunk(struct btrfs_fs_info *fs_info, u64 *offset)
517 struct btrfs_root *root = fs_info->chunk_root;
518 struct btrfs_path *path;
520 struct btrfs_key key;
521 struct btrfs_chunk *chunk;
522 struct btrfs_key found_key;
524 path = btrfs_alloc_path();
528 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
529 key.offset = (u64)-1;
530 key.type = BTRFS_CHUNK_ITEM_KEY;
532 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
538 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
542 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
544 if (found_key.objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID)
547 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
549 *offset = found_key.offset +
550 btrfs_chunk_length(path->nodes[0], chunk);
555 btrfs_free_path(path);
559 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
563 struct btrfs_key key;
564 struct btrfs_key found_key;
566 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
567 key.type = BTRFS_DEV_ITEM_KEY;
568 key.offset = (u64)-1;
570 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
576 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
581 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
583 *objectid = found_key.offset + 1;
587 btrfs_release_path(path);
592 * the device information is stored in the chunk root
593 * the btrfs_device struct should be fully filled in
595 int btrfs_add_device(struct btrfs_trans_handle *trans,
596 struct btrfs_fs_info *fs_info,
597 struct btrfs_device *device)
600 struct btrfs_path *path;
601 struct btrfs_dev_item *dev_item;
602 struct extent_buffer *leaf;
603 struct btrfs_key key;
604 struct btrfs_root *root = fs_info->chunk_root;
608 path = btrfs_alloc_path();
612 ret = find_next_devid(root, path, &free_devid);
616 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
617 key.type = BTRFS_DEV_ITEM_KEY;
618 key.offset = free_devid;
620 ret = btrfs_insert_empty_item(trans, root, path, &key,
625 leaf = path->nodes[0];
626 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
628 device->devid = free_devid;
629 btrfs_set_device_id(leaf, dev_item, device->devid);
630 btrfs_set_device_generation(leaf, dev_item, 0);
631 btrfs_set_device_type(leaf, dev_item, device->type);
632 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
633 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
634 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
635 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
636 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
637 btrfs_set_device_group(leaf, dev_item, 0);
638 btrfs_set_device_seek_speed(leaf, dev_item, 0);
639 btrfs_set_device_bandwidth(leaf, dev_item, 0);
640 btrfs_set_device_start_offset(leaf, dev_item, 0);
642 ptr = (unsigned long)btrfs_device_uuid(dev_item);
643 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
644 ptr = (unsigned long)btrfs_device_fsid(dev_item);
645 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
646 btrfs_mark_buffer_dirty(leaf);
650 btrfs_free_path(path);
654 int btrfs_update_device(struct btrfs_trans_handle *trans,
655 struct btrfs_device *device)
658 struct btrfs_path *path;
659 struct btrfs_root *root;
660 struct btrfs_dev_item *dev_item;
661 struct extent_buffer *leaf;
662 struct btrfs_key key;
664 root = device->dev_root->fs_info->chunk_root;
666 path = btrfs_alloc_path();
670 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
671 key.type = BTRFS_DEV_ITEM_KEY;
672 key.offset = device->devid;
674 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
683 leaf = path->nodes[0];
684 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
686 btrfs_set_device_id(leaf, dev_item, device->devid);
687 btrfs_set_device_type(leaf, dev_item, device->type);
688 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
689 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
690 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
691 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
692 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
693 btrfs_mark_buffer_dirty(leaf);
696 btrfs_free_path(path);
700 int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
701 struct btrfs_chunk *chunk, int item_size)
703 struct btrfs_super_block *super_copy = fs_info->super_copy;
704 struct btrfs_disk_key disk_key;
708 array_size = btrfs_super_sys_array_size(super_copy);
709 if (array_size + item_size + sizeof(disk_key)
710 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
713 ptr = super_copy->sys_chunk_array + array_size;
714 btrfs_cpu_key_to_disk(&disk_key, key);
715 memcpy(ptr, &disk_key, sizeof(disk_key));
716 ptr += sizeof(disk_key);
717 memcpy(ptr, chunk, item_size);
718 item_size += sizeof(disk_key);
719 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
723 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
726 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
728 else if (type & BTRFS_BLOCK_GROUP_RAID10)
729 return calc_size * (num_stripes / sub_stripes);
730 else if (type & BTRFS_BLOCK_GROUP_RAID5)
731 return calc_size * (num_stripes - 1);
732 else if (type & BTRFS_BLOCK_GROUP_RAID6)
733 return calc_size * (num_stripes - 2);
735 return calc_size * num_stripes;
739 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
741 /* TODO, add a way to store the preferred stripe size */
742 return BTRFS_STRIPE_LEN;
746 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
748 * It is not equal to "device->total_bytes - device->bytes_used".
749 * We do not allocate any chunk in 1M at beginning of device, and not
750 * allowed to allocate any chunk before alloc_start if it is specified.
751 * So search holes from max(1M, alloc_start) to device->total_bytes.
753 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
754 struct btrfs_device *device,
757 struct btrfs_path *path;
758 struct btrfs_root *root = device->dev_root;
759 struct btrfs_key key;
760 struct btrfs_dev_extent *dev_extent = NULL;
761 struct extent_buffer *l;
762 u64 search_start = root->fs_info->alloc_start;
763 u64 search_end = device->total_bytes;
769 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
771 path = btrfs_alloc_path();
775 key.objectid = device->devid;
776 key.offset = root->fs_info->alloc_start;
777 key.type = BTRFS_DEV_EXTENT_KEY;
780 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
783 ret = btrfs_previous_item(root, path, 0, key.type);
789 slot = path->slots[0];
790 if (slot >= btrfs_header_nritems(l)) {
791 ret = btrfs_next_leaf(root, path);
798 btrfs_item_key_to_cpu(l, &key, slot);
800 if (key.objectid < device->devid)
802 if (key.objectid > device->devid)
804 if (key.type != BTRFS_DEV_EXTENT_KEY)
806 if (key.offset > search_end)
808 if (key.offset > search_start)
809 free_bytes += key.offset - search_start;
811 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
812 extent_end = key.offset + btrfs_dev_extent_length(l,
814 if (extent_end > search_start)
815 search_start = extent_end;
816 if (search_start > search_end)
823 if (search_start < search_end)
824 free_bytes += search_end - search_start;
826 *avail_bytes = free_bytes;
829 btrfs_free_path(path);
833 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
834 - sizeof(struct btrfs_item) \
835 - sizeof(struct btrfs_chunk)) \
836 / sizeof(struct btrfs_stripe) + 1)
838 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
839 - 2 * sizeof(struct btrfs_disk_key) \
840 - 2 * sizeof(struct btrfs_chunk)) \
841 / sizeof(struct btrfs_stripe) + 1)
843 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
844 struct btrfs_fs_info *info, u64 *start,
845 u64 *num_bytes, u64 type)
848 struct btrfs_root *extent_root = info->extent_root;
849 struct btrfs_root *chunk_root = info->chunk_root;
850 struct btrfs_stripe *stripes;
851 struct btrfs_device *device = NULL;
852 struct btrfs_chunk *chunk;
853 struct list_head private_devs;
854 struct list_head *dev_list = &info->fs_devices->devices;
855 struct list_head *cur;
856 struct map_lookup *map;
857 int min_stripe_size = SZ_1M;
858 u64 calc_size = SZ_8M;
860 u64 max_chunk_size = 4 * calc_size;
871 int stripe_len = BTRFS_STRIPE_LEN;
872 struct btrfs_key key;
875 if (list_empty(dev_list)) {
879 if (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
880 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
882 max_chunk_size = calc_size * 2;
883 min_stripe_size = SZ_1M;
884 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
885 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
887 max_chunk_size = 10 * calc_size;
888 min_stripe_size = SZ_64M;
889 max_stripes = BTRFS_MAX_DEVS(chunk_root);
890 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
892 max_chunk_size = 4 * calc_size;
893 min_stripe_size = SZ_32M;
894 max_stripes = BTRFS_MAX_DEVS(chunk_root);
897 if (type & BTRFS_BLOCK_GROUP_RAID1) {
898 num_stripes = min_t(u64, 2,
899 btrfs_super_num_devices(info->super_copy));
904 if (type & BTRFS_BLOCK_GROUP_DUP) {
908 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
909 num_stripes = btrfs_super_num_devices(info->super_copy);
910 if (num_stripes > max_stripes)
911 num_stripes = max_stripes;
914 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
915 num_stripes = btrfs_super_num_devices(info->super_copy);
916 if (num_stripes > max_stripes)
917 num_stripes = max_stripes;
920 num_stripes &= ~(u32)1;
924 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
925 num_stripes = btrfs_super_num_devices(info->super_copy);
926 if (num_stripes > max_stripes)
927 num_stripes = max_stripes;
931 stripe_len = find_raid56_stripe_len(num_stripes - 1,
932 btrfs_super_stripesize(info->super_copy));
934 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
935 num_stripes = btrfs_super_num_devices(info->super_copy);
936 if (num_stripes > max_stripes)
937 num_stripes = max_stripes;
941 stripe_len = find_raid56_stripe_len(num_stripes - 2,
942 btrfs_super_stripesize(info->super_copy));
945 /* we don't want a chunk larger than 10% of the FS */
946 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
947 max_chunk_size = min(percent_max, max_chunk_size);
950 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
952 calc_size = max_chunk_size;
953 calc_size /= num_stripes;
954 calc_size /= stripe_len;
955 calc_size *= stripe_len;
957 /* we don't want tiny stripes */
958 calc_size = max_t(u64, calc_size, min_stripe_size);
960 calc_size /= stripe_len;
961 calc_size *= stripe_len;
962 INIT_LIST_HEAD(&private_devs);
963 cur = dev_list->next;
966 if (type & BTRFS_BLOCK_GROUP_DUP)
967 min_free = calc_size * 2;
969 min_free = calc_size;
971 /* build a private list of devices we will allocate from */
972 while(index < num_stripes) {
973 device = list_entry(cur, struct btrfs_device, dev_list);
974 ret = btrfs_device_avail_bytes(trans, device, &avail);
978 if (avail >= min_free) {
979 list_move_tail(&device->dev_list, &private_devs);
981 if (type & BTRFS_BLOCK_GROUP_DUP)
983 } else if (avail > max_avail)
988 if (index < num_stripes) {
989 list_splice(&private_devs, dev_list);
990 if (index >= min_stripes) {
992 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
993 num_stripes /= sub_stripes;
994 num_stripes *= sub_stripes;
999 if (!looped && max_avail > 0) {
1001 calc_size = max_avail;
1006 ret = find_next_chunk(info, &offset);
1009 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1010 key.type = BTRFS_CHUNK_ITEM_KEY;
1011 key.offset = offset;
1013 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1017 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1023 stripes = &chunk->stripe;
1024 *num_bytes = chunk_bytes_by_type(type, calc_size,
1025 num_stripes, sub_stripes);
1027 while(index < num_stripes) {
1028 struct btrfs_stripe *stripe;
1029 BUG_ON(list_empty(&private_devs));
1030 cur = private_devs.next;
1031 device = list_entry(cur, struct btrfs_device, dev_list);
1033 /* loop over this device again if we're doing a dup group */
1034 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1035 (index == num_stripes - 1))
1036 list_move_tail(&device->dev_list, dev_list);
1038 ret = btrfs_alloc_dev_extent(trans, device,
1039 info->chunk_root->root_key.objectid,
1040 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1041 calc_size, &dev_offset, 0);
1045 device->bytes_used += calc_size;
1046 ret = btrfs_update_device(trans, device);
1050 map->stripes[index].dev = device;
1051 map->stripes[index].physical = dev_offset;
1052 stripe = stripes + index;
1053 btrfs_set_stack_stripe_devid(stripe, device->devid);
1054 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1055 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1058 BUG_ON(!list_empty(&private_devs));
1060 /* key was set above */
1061 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1062 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1063 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1064 btrfs_set_stack_chunk_type(chunk, type);
1065 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1066 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1067 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1068 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1069 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1070 map->sector_size = info->sectorsize;
1071 map->stripe_len = stripe_len;
1072 map->io_align = stripe_len;
1073 map->io_width = stripe_len;
1075 map->num_stripes = num_stripes;
1076 map->sub_stripes = sub_stripes;
1078 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1079 btrfs_chunk_item_size(num_stripes));
1081 *start = key.offset;;
1083 map->ce.start = key.offset;
1084 map->ce.size = *num_bytes;
1086 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1090 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1091 ret = btrfs_add_system_chunk(info, &key,
1092 chunk, btrfs_chunk_item_size(num_stripes));
1108 * Alloc a DATA chunk with SINGLE profile.
1110 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1111 * (btrfs logical bytenr == on-disk bytenr)
1112 * For that case, caller must make sure the chunk and dev_extent are not
1115 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1116 struct btrfs_fs_info *info, u64 *start,
1117 u64 num_bytes, u64 type, int convert)
1120 struct btrfs_root *extent_root = info->extent_root;
1121 struct btrfs_root *chunk_root = info->chunk_root;
1122 struct btrfs_stripe *stripes;
1123 struct btrfs_device *device = NULL;
1124 struct btrfs_chunk *chunk;
1125 struct list_head *dev_list = &info->fs_devices->devices;
1126 struct list_head *cur;
1127 struct map_lookup *map;
1128 u64 calc_size = SZ_8M;
1129 int num_stripes = 1;
1130 int sub_stripes = 0;
1133 int stripe_len = BTRFS_STRIPE_LEN;
1134 struct btrfs_key key;
1136 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1137 key.type = BTRFS_CHUNK_ITEM_KEY;
1139 if (*start != round_down(*start, info->sectorsize)) {
1140 error("DATA chunk start not sectorsize aligned: %llu",
1141 (unsigned long long)*start);
1144 key.offset = *start;
1145 dev_offset = *start;
1149 ret = find_next_chunk(info, &tmp);
1155 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1159 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1165 stripes = &chunk->stripe;
1166 calc_size = num_bytes;
1169 cur = dev_list->next;
1170 device = list_entry(cur, struct btrfs_device, dev_list);
1172 while (index < num_stripes) {
1173 struct btrfs_stripe *stripe;
1175 ret = btrfs_alloc_dev_extent(trans, device,
1176 info->chunk_root->root_key.objectid,
1177 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1178 calc_size, &dev_offset, convert);
1181 device->bytes_used += calc_size;
1182 ret = btrfs_update_device(trans, device);
1185 map->stripes[index].dev = device;
1186 map->stripes[index].physical = dev_offset;
1187 stripe = stripes + index;
1188 btrfs_set_stack_stripe_devid(stripe, device->devid);
1189 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1190 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1194 /* key was set above */
1195 btrfs_set_stack_chunk_length(chunk, num_bytes);
1196 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1197 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1198 btrfs_set_stack_chunk_type(chunk, type);
1199 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1200 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1201 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1202 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1203 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1204 map->sector_size = info->sectorsize;
1205 map->stripe_len = stripe_len;
1206 map->io_align = stripe_len;
1207 map->io_width = stripe_len;
1209 map->num_stripes = num_stripes;
1210 map->sub_stripes = sub_stripes;
1212 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1213 btrfs_chunk_item_size(num_stripes));
1216 *start = key.offset;
1218 map->ce.start = key.offset;
1219 map->ce.size = num_bytes;
1221 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1228 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
1230 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1231 struct cache_extent *ce;
1232 struct map_lookup *map;
1235 ce = search_cache_extent(&map_tree->cache_tree, logical);
1237 fprintf(stderr, "No mapping for %llu-%llu\n",
1238 (unsigned long long)logical,
1239 (unsigned long long)logical+len);
1242 if (ce->start > logical || ce->start + ce->size < logical) {
1243 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1244 "%llu-%llu\n", (unsigned long long)logical,
1245 (unsigned long long)logical+len,
1246 (unsigned long long)ce->start,
1247 (unsigned long long)ce->start + ce->size);
1250 map = container_of(ce, struct map_lookup, ce);
1252 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1253 ret = map->num_stripes;
1254 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1255 ret = map->sub_stripes;
1256 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1258 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1265 int btrfs_next_bg(struct btrfs_fs_info *fs_info, u64 *logical,
1266 u64 *size, u64 type)
1268 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1269 struct cache_extent *ce;
1270 struct map_lookup *map;
1273 ce = search_cache_extent(&map_tree->cache_tree, cur);
1277 * only jump to next bg if our cur is not 0
1278 * As the initial logical for btrfs_next_bg() is 0, and
1279 * if we jump to next bg, we skipped a valid bg.
1282 ce = next_cache_extent(ce);
1288 map = container_of(ce, struct map_lookup, ce);
1289 if (map->type & type) {
1290 *logical = ce->start;
1295 ce = next_cache_extent(ce);
1301 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
1302 u64 chunk_start, u64 physical, u64 devid,
1303 u64 **logical, int *naddrs, int *stripe_len)
1305 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1306 struct cache_extent *ce;
1307 struct map_lookup *map;
1315 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1317 map = container_of(ce, struct map_lookup, ce);
1320 rmap_len = map->stripe_len;
1321 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1322 length = ce->size / (map->num_stripes / map->sub_stripes);
1323 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1324 length = ce->size / map->num_stripes;
1325 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1326 BTRFS_BLOCK_GROUP_RAID6)) {
1327 length = ce->size / nr_data_stripes(map);
1328 rmap_len = map->stripe_len * nr_data_stripes(map);
1331 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1333 for (i = 0; i < map->num_stripes; i++) {
1334 if (devid && map->stripes[i].dev->devid != devid)
1336 if (map->stripes[i].physical > physical ||
1337 map->stripes[i].physical + length <= physical)
1340 stripe_nr = (physical - map->stripes[i].physical) /
1343 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1344 stripe_nr = (stripe_nr * map->num_stripes + i) /
1346 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1347 stripe_nr = stripe_nr * map->num_stripes + i;
1348 } /* else if RAID[56], multiply by nr_data_stripes().
1349 * Alternatively, just use rmap_len below instead of
1350 * map->stripe_len */
1352 bytenr = ce->start + stripe_nr * rmap_len;
1353 for (j = 0; j < nr; j++) {
1354 if (buf[j] == bytenr)
1363 *stripe_len = rmap_len;
1368 static inline int parity_smaller(u64 a, u64 b)
1373 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1374 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1376 struct btrfs_bio_stripe s;
1383 for (i = 0; i < bbio->num_stripes - 1; i++) {
1384 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1385 s = bbio->stripes[i];
1387 bbio->stripes[i] = bbio->stripes[i+1];
1388 raid_map[i] = raid_map[i+1];
1389 bbio->stripes[i+1] = s;
1397 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1398 u64 logical, u64 *length,
1399 struct btrfs_multi_bio **multi_ret, int mirror_num,
1402 return __btrfs_map_block(fs_info, rw, logical, length, NULL,
1403 multi_ret, mirror_num, raid_map_ret);
1406 int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1407 u64 logical, u64 *length, u64 *type,
1408 struct btrfs_multi_bio **multi_ret, int mirror_num,
1411 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1412 struct cache_extent *ce;
1413 struct map_lookup *map;
1417 u64 *raid_map = NULL;
1418 int stripes_allocated = 8;
1419 int stripes_required = 1;
1422 struct btrfs_multi_bio *multi = NULL;
1424 if (multi_ret && rw == READ) {
1425 stripes_allocated = 1;
1428 ce = search_cache_extent(&map_tree->cache_tree, logical);
1434 if (ce->start > logical) {
1436 *length = ce->start - logical;
1441 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1446 map = container_of(ce, struct map_lookup, ce);
1447 offset = logical - ce->start;
1450 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1451 BTRFS_BLOCK_GROUP_DUP)) {
1452 stripes_required = map->num_stripes;
1453 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1454 stripes_required = map->sub_stripes;
1457 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1458 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1459 /* RAID[56] write or recovery. Return all stripes */
1460 stripes_required = map->num_stripes;
1462 /* Only allocate the map if we've already got a large enough multi_ret */
1463 if (stripes_allocated >= stripes_required) {
1464 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1472 /* if our multi bio struct is too small, back off and try again */
1473 if (multi_ret && stripes_allocated < stripes_required) {
1474 stripes_allocated = stripes_required;
1481 * stripe_nr counts the total number of stripes we have to stride
1482 * to get to this block
1484 stripe_nr = stripe_nr / map->stripe_len;
1486 stripe_offset = stripe_nr * map->stripe_len;
1487 BUG_ON(offset < stripe_offset);
1489 /* stripe_offset is the offset of this block in its stripe*/
1490 stripe_offset = offset - stripe_offset;
1492 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1493 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1494 BTRFS_BLOCK_GROUP_RAID10 |
1495 BTRFS_BLOCK_GROUP_DUP)) {
1496 /* we limit the length of each bio to what fits in a stripe */
1497 *length = min_t(u64, ce->size - offset,
1498 map->stripe_len - stripe_offset);
1500 *length = ce->size - offset;
1506 multi->num_stripes = 1;
1508 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1510 multi->num_stripes = map->num_stripes;
1511 else if (mirror_num)
1512 stripe_index = mirror_num - 1;
1514 stripe_index = stripe_nr % map->num_stripes;
1515 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1516 int factor = map->num_stripes / map->sub_stripes;
1518 stripe_index = stripe_nr % factor;
1519 stripe_index *= map->sub_stripes;
1522 multi->num_stripes = map->sub_stripes;
1523 else if (mirror_num)
1524 stripe_index += mirror_num - 1;
1526 stripe_nr = stripe_nr / factor;
1527 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1529 multi->num_stripes = map->num_stripes;
1530 else if (mirror_num)
1531 stripe_index = mirror_num - 1;
1532 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1533 BTRFS_BLOCK_GROUP_RAID6)) {
1538 u64 raid56_full_stripe_start;
1539 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1542 * align the start of our data stripe in the logical
1545 raid56_full_stripe_start = offset / full_stripe_len;
1546 raid56_full_stripe_start *= full_stripe_len;
1548 /* get the data stripe number */
1549 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1550 stripe_nr = stripe_nr / nr_data_stripes(map);
1552 /* Work out the disk rotation on this stripe-set */
1553 rot = stripe_nr % map->num_stripes;
1555 /* Fill in the logical address of each stripe */
1556 tmp = stripe_nr * nr_data_stripes(map);
1558 for (i = 0; i < nr_data_stripes(map); i++)
1559 raid_map[(i+rot) % map->num_stripes] =
1560 ce->start + (tmp + i) * map->stripe_len;
1562 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1563 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1564 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1566 *length = map->stripe_len;
1569 multi->num_stripes = map->num_stripes;
1571 stripe_index = stripe_nr % nr_data_stripes(map);
1572 stripe_nr = stripe_nr / nr_data_stripes(map);
1575 * Mirror #0 or #1 means the original data block.
1576 * Mirror #2 is RAID5 parity block.
1577 * Mirror #3 is RAID6 Q block.
1580 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1582 /* We distribute the parity blocks across stripes */
1583 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1587 * after this do_div call, stripe_nr is the number of stripes
1588 * on this device we have to walk to find the data, and
1589 * stripe_index is the number of our device in the stripe array
1591 stripe_index = stripe_nr % map->num_stripes;
1592 stripe_nr = stripe_nr / map->num_stripes;
1594 BUG_ON(stripe_index >= map->num_stripes);
1596 for (i = 0; i < multi->num_stripes; i++) {
1597 multi->stripes[i].physical =
1598 map->stripes[stripe_index].physical + stripe_offset +
1599 stripe_nr * map->stripe_len;
1600 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1609 sort_parity_stripes(multi, raid_map);
1610 *raid_map_ret = raid_map;
1616 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
1619 struct btrfs_device *device;
1620 struct btrfs_fs_devices *cur_devices;
1622 cur_devices = fs_info->fs_devices;
1623 while (cur_devices) {
1625 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1626 fs_info->ignore_fsid_mismatch)) {
1627 device = __find_device(&cur_devices->devices,
1632 cur_devices = cur_devices->seed;
1637 struct btrfs_device *
1638 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1639 u64 devid, int instance)
1641 struct list_head *head = &fs_devices->devices;
1642 struct btrfs_device *dev;
1645 list_for_each_entry(dev, head, dev_list) {
1646 if (dev->devid == devid && num_found++ == instance)
1652 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
1654 struct cache_extent *ce;
1655 struct map_lookup *map;
1656 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1661 * During chunk recovering, we may fail to find block group's
1662 * corresponding chunk, we will rebuild it later
1664 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1665 if (!fs_info->is_chunk_recover)
1670 map = container_of(ce, struct map_lookup, ce);
1671 for (i = 0; i < map->num_stripes; i++) {
1672 if (!map->stripes[i].dev->writeable) {
1681 static struct btrfs_device *fill_missing_device(u64 devid)
1683 struct btrfs_device *device;
1685 device = kzalloc(sizeof(*device), GFP_NOFS);
1686 device->devid = devid;
1692 * slot == -1: SYSTEM chunk
1693 * return -EIO on error, otherwise return 0
1695 int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
1696 struct extent_buffer *leaf,
1697 struct btrfs_chunk *chunk,
1698 int slot, u64 logical)
1705 u32 chunk_ondisk_size;
1706 u32 sectorsize = fs_info->sectorsize;
1708 length = btrfs_chunk_length(leaf, chunk);
1709 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1710 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1711 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1712 type = btrfs_chunk_type(leaf, chunk);
1715 * These valid checks may be insufficient to cover every corner cases.
1717 if (!IS_ALIGNED(logical, sectorsize)) {
1718 error("invalid chunk logical %llu", logical);
1721 if (btrfs_chunk_sector_size(leaf, chunk) != sectorsize) {
1722 error("invalid chunk sectorsize %llu",
1723 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1726 if (!length || !IS_ALIGNED(length, sectorsize)) {
1727 error("invalid chunk length %llu", length);
1730 if (stripe_len != BTRFS_STRIPE_LEN) {
1731 error("invalid chunk stripe length: %llu", stripe_len);
1734 /* Check on chunk item type */
1735 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1736 error("invalid chunk type %llu", type);
1739 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1740 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1741 error("unrecognized chunk type: %llu",
1742 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1743 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1746 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1747 error("missing chunk type flag: %llu", type);
1750 if (!(is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) ||
1751 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0)) {
1752 error("conflicting chunk type detected: %llu", type);
1755 if ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
1756 !is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1757 error("conflicting chunk profile detected: %llu", type);
1761 chunk_ondisk_size = btrfs_chunk_item_size(num_stripes);
1763 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1764 * it can't exceed the system chunk array size
1765 * For normal chunk, it should match its chunk item size.
1767 if (num_stripes < 1 ||
1768 (slot == -1 && chunk_ondisk_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1769 (slot >= 0 && chunk_ondisk_size > btrfs_item_size_nr(leaf, slot))) {
1770 error("invalid num_stripes: %u", num_stripes);
1774 * Device number check against profile
1776 if ((type & BTRFS_BLOCK_GROUP_RAID10 && (sub_stripes != 2 ||
1777 !IS_ALIGNED(num_stripes, sub_stripes))) ||
1778 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1779 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1780 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1781 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1782 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1783 num_stripes != 1)) {
1784 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1785 num_stripes, sub_stripes,
1786 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1794 * Slot is used to verify the chunk item is valid
1796 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1798 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
1799 struct extent_buffer *leaf,
1800 struct btrfs_chunk *chunk, int slot)
1802 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1803 struct map_lookup *map;
1804 struct cache_extent *ce;
1808 u8 uuid[BTRFS_UUID_SIZE];
1813 logical = key->offset;
1814 length = btrfs_chunk_length(leaf, chunk);
1815 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1816 /* Validation check */
1817 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, slot, logical);
1819 error("%s checksums match, but it has an invalid chunk, %s",
1820 (slot == -1) ? "Superblock" : "Metadata",
1821 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1825 ce = search_cache_extent(&map_tree->cache_tree, logical);
1827 /* already mapped? */
1828 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1832 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1836 map->ce.start = logical;
1837 map->ce.size = length;
1838 map->num_stripes = num_stripes;
1839 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1840 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1841 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1842 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1843 map->type = btrfs_chunk_type(leaf, chunk);
1844 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1846 for (i = 0; i < num_stripes; i++) {
1847 map->stripes[i].physical =
1848 btrfs_stripe_offset_nr(leaf, chunk, i);
1849 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1850 read_extent_buffer(leaf, uuid, (unsigned long)
1851 btrfs_stripe_dev_uuid_nr(chunk, i),
1853 map->stripes[i].dev = btrfs_find_device(fs_info, devid, uuid,
1855 if (!map->stripes[i].dev) {
1856 map->stripes[i].dev = fill_missing_device(devid);
1857 printf("warning, device %llu is missing\n",
1858 (unsigned long long)devid);
1859 list_add(&map->stripes[i].dev->dev_list,
1860 &fs_info->fs_devices->devices);
1864 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1870 static int fill_device_from_item(struct extent_buffer *leaf,
1871 struct btrfs_dev_item *dev_item,
1872 struct btrfs_device *device)
1876 device->devid = btrfs_device_id(leaf, dev_item);
1877 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1878 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1879 device->type = btrfs_device_type(leaf, dev_item);
1880 device->io_align = btrfs_device_io_align(leaf, dev_item);
1881 device->io_width = btrfs_device_io_width(leaf, dev_item);
1882 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1884 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1885 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1890 static int open_seed_devices(struct btrfs_fs_info *fs_info, u8 *fsid)
1892 struct btrfs_fs_devices *fs_devices;
1895 fs_devices = fs_info->fs_devices->seed;
1896 while (fs_devices) {
1897 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1901 fs_devices = fs_devices->seed;
1904 fs_devices = find_fsid(fsid);
1906 /* missing all seed devices */
1907 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1912 INIT_LIST_HEAD(&fs_devices->devices);
1913 list_add(&fs_devices->list, &fs_uuids);
1914 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1917 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1921 fs_devices->seed = fs_info->fs_devices->seed;
1922 fs_info->fs_devices->seed = fs_devices;
1927 static int read_one_dev(struct btrfs_fs_info *fs_info,
1928 struct extent_buffer *leaf,
1929 struct btrfs_dev_item *dev_item)
1931 struct btrfs_device *device;
1934 u8 fs_uuid[BTRFS_UUID_SIZE];
1935 u8 dev_uuid[BTRFS_UUID_SIZE];
1937 devid = btrfs_device_id(leaf, dev_item);
1938 read_extent_buffer(leaf, dev_uuid,
1939 (unsigned long)btrfs_device_uuid(dev_item),
1941 read_extent_buffer(leaf, fs_uuid,
1942 (unsigned long)btrfs_device_fsid(dev_item),
1945 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
1946 ret = open_seed_devices(fs_info, fs_uuid);
1951 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
1953 device = kzalloc(sizeof(*device), GFP_NOFS);
1957 list_add(&device->dev_list,
1958 &fs_info->fs_devices->devices);
1961 fill_device_from_item(leaf, dev_item, device);
1962 device->dev_root = fs_info->dev_root;
1966 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
1968 struct btrfs_super_block *super_copy = fs_info->super_copy;
1969 struct extent_buffer *sb;
1970 struct btrfs_disk_key *disk_key;
1971 struct btrfs_chunk *chunk;
1973 unsigned long sb_array_offset;
1979 struct btrfs_key key;
1981 if (fs_info->nodesize < BTRFS_SUPER_INFO_SIZE) {
1982 printf("ERROR: nodesize %u too small to read superblock\n",
1986 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
1989 btrfs_set_buffer_uptodate(sb);
1990 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1991 array_size = btrfs_super_sys_array_size(super_copy);
1993 array_ptr = super_copy->sys_chunk_array;
1994 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1997 while (cur_offset < array_size) {
1998 disk_key = (struct btrfs_disk_key *)array_ptr;
1999 len = sizeof(*disk_key);
2000 if (cur_offset + len > array_size)
2001 goto out_short_read;
2003 btrfs_disk_key_to_cpu(&key, disk_key);
2006 sb_array_offset += len;
2009 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2010 chunk = (struct btrfs_chunk *)sb_array_offset;
2012 * At least one btrfs_chunk with one stripe must be
2013 * present, exact stripe count check comes afterwards
2015 len = btrfs_chunk_item_size(1);
2016 if (cur_offset + len > array_size)
2017 goto out_short_read;
2019 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2022 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
2023 num_stripes, cur_offset);
2028 len = btrfs_chunk_item_size(num_stripes);
2029 if (cur_offset + len > array_size)
2030 goto out_short_read;
2032 ret = read_one_chunk(fs_info, &key, sb, chunk, -1);
2037 "ERROR: unexpected item type %u in sys_array at offset %u\n",
2038 (u32)key.type, cur_offset);
2043 sb_array_offset += len;
2046 free_extent_buffer(sb);
2050 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
2052 free_extent_buffer(sb);
2056 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
2058 struct btrfs_path *path;
2059 struct extent_buffer *leaf;
2060 struct btrfs_key key;
2061 struct btrfs_key found_key;
2062 struct btrfs_root *root = fs_info->chunk_root;
2066 path = btrfs_alloc_path();
2071 * Read all device items, and then all the chunk items. All
2072 * device items are found before any chunk item (their object id
2073 * is smaller than the lowest possible object id for a chunk
2074 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
2076 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2079 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2083 leaf = path->nodes[0];
2084 slot = path->slots[0];
2085 if (slot >= btrfs_header_nritems(leaf)) {
2086 ret = btrfs_next_leaf(root, path);
2093 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2094 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2095 struct btrfs_dev_item *dev_item;
2096 dev_item = btrfs_item_ptr(leaf, slot,
2097 struct btrfs_dev_item);
2098 ret = read_one_dev(fs_info, leaf, dev_item);
2100 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2101 struct btrfs_chunk *chunk;
2102 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2103 ret = read_one_chunk(fs_info, &found_key, leaf, chunk,
2112 btrfs_free_path(path);
2116 struct list_head *btrfs_scanned_uuids(void)
2121 static int rmw_eb(struct btrfs_fs_info *info,
2122 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2125 unsigned long orig_off = 0;
2126 unsigned long dest_off = 0;
2127 unsigned long copy_len = eb->len;
2129 ret = read_whole_eb(info, eb, 0);
2133 if (eb->start + eb->len <= orig_eb->start ||
2134 eb->start >= orig_eb->start + orig_eb->len)
2137 * | ----- orig_eb ------- |
2138 * | ----- stripe ------- |
2139 * | ----- orig_eb ------- |
2140 * | ----- orig_eb ------- |
2142 if (eb->start > orig_eb->start)
2143 orig_off = eb->start - orig_eb->start;
2144 if (orig_eb->start > eb->start)
2145 dest_off = orig_eb->start - eb->start;
2147 if (copy_len > orig_eb->len - orig_off)
2148 copy_len = orig_eb->len - orig_off;
2149 if (copy_len > eb->len - dest_off)
2150 copy_len = eb->len - dest_off;
2152 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2156 static int split_eb_for_raid56(struct btrfs_fs_info *info,
2157 struct extent_buffer *orig_eb,
2158 struct extent_buffer **ebs,
2159 u64 stripe_len, u64 *raid_map,
2162 struct extent_buffer **tmp_ebs;
2163 u64 start = orig_eb->start;
2168 tmp_ebs = calloc(num_stripes, sizeof(*tmp_ebs));
2172 /* Alloc memory in a row for data stripes */
2173 for (i = 0; i < num_stripes; i++) {
2174 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2177 tmp_ebs[i] = calloc(1, sizeof(**tmp_ebs) + stripe_len);
2184 for (i = 0; i < num_stripes; i++) {
2185 struct extent_buffer *eb = tmp_ebs[i];
2187 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2190 eb->start = raid_map[i];
2191 eb->len = stripe_len;
2195 eb->dev_bytenr = (u64)-1;
2197 this_eb_start = raid_map[i];
2199 if (start > this_eb_start ||
2200 start + orig_eb->len < this_eb_start + stripe_len) {
2201 ret = rmw_eb(info, eb, orig_eb);
2205 memcpy(eb->data, orig_eb->data + eb->start - start,
2213 for (i = 0; i < num_stripes; i++)
2219 int write_raid56_with_parity(struct btrfs_fs_info *info,
2220 struct extent_buffer *eb,
2221 struct btrfs_multi_bio *multi,
2222 u64 stripe_len, u64 *raid_map)
2224 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2227 int alloc_size = eb->len;
2230 ebs = malloc(sizeof(*ebs) * multi->num_stripes);
2231 pointers = malloc(sizeof(*pointers) * multi->num_stripes);
2232 if (!ebs || !pointers) {
2238 if (stripe_len > alloc_size)
2239 alloc_size = stripe_len;
2241 ret = split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2242 multi->num_stripes);
2246 for (i = 0; i < multi->num_stripes; i++) {
2247 struct extent_buffer *new_eb;
2248 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2249 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2250 ebs[i]->fd = multi->stripes[i].dev->fd;
2251 multi->stripes[i].dev->total_ios++;
2252 if (ebs[i]->start != raid_map[i]) {
2254 goto out_free_split;
2258 new_eb = malloc(sizeof(*eb) + alloc_size);
2261 goto out_free_split;
2263 new_eb->dev_bytenr = multi->stripes[i].physical;
2264 new_eb->fd = multi->stripes[i].dev->fd;
2265 multi->stripes[i].dev->total_ios++;
2266 new_eb->len = stripe_len;
2268 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2270 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2274 ebs[multi->num_stripes - 2] = p_eb;
2275 ebs[multi->num_stripes - 1] = q_eb;
2277 for (i = 0; i < multi->num_stripes; i++)
2278 pointers[i] = ebs[i]->data;
2280 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2282 ebs[multi->num_stripes - 1] = p_eb;
2283 for (i = 0; i < multi->num_stripes; i++)
2284 pointers[i] = ebs[i]->data;
2285 ret = raid5_gen_result(multi->num_stripes, stripe_len,
2286 multi->num_stripes - 1, pointers);
2288 goto out_free_split;
2291 for (i = 0; i < multi->num_stripes; i++) {
2292 ret = write_extent_to_disk(ebs[i]);
2294 goto out_free_split;
2298 for (i = 0; i < multi->num_stripes; i++) {
2310 * Get stripe length from chunk item and its stripe items
2312 * Caller should only call this function after validating the chunk item
2313 * by using btrfs_check_chunk_valid().
2315 u64 btrfs_stripe_length(struct btrfs_fs_info *fs_info,
2316 struct extent_buffer *leaf,
2317 struct btrfs_chunk *chunk)
2321 u32 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2322 u64 profile = btrfs_chunk_type(leaf, chunk) &
2323 BTRFS_BLOCK_GROUP_PROFILE_MASK;
2325 chunk_len = btrfs_chunk_length(leaf, chunk);
2328 case 0: /* Single profile */
2329 case BTRFS_BLOCK_GROUP_RAID1:
2330 case BTRFS_BLOCK_GROUP_DUP:
2331 stripe_len = chunk_len;
2333 case BTRFS_BLOCK_GROUP_RAID0:
2334 stripe_len = chunk_len / num_stripes;
2336 case BTRFS_BLOCK_GROUP_RAID5:
2337 stripe_len = chunk_len / (num_stripes - 1);
2339 case BTRFS_BLOCK_GROUP_RAID6:
2340 stripe_len = chunk_len / (num_stripes - 2);
2342 case BTRFS_BLOCK_GROUP_RAID10:
2343 stripe_len = chunk_len / (num_stripes /
2344 btrfs_chunk_sub_stripes(leaf, chunk));
2347 /* Invalid chunk profile found */
2354 * Return 0 if size of @device is already good
2355 * Return >0 if size of @device is not aligned but fixed without problems
2356 * Return <0 if something wrong happened when aligning the size of @device
2358 int btrfs_fix_device_size(struct btrfs_fs_info *fs_info,
2359 struct btrfs_device *device)
2361 struct btrfs_trans_handle *trans;
2362 struct btrfs_key key;
2363 struct btrfs_path path;
2364 struct btrfs_root *chunk_root = fs_info->chunk_root;
2365 struct btrfs_dev_item *di;
2366 u64 old_bytes = device->total_bytes;
2369 if (IS_ALIGNED(old_bytes, fs_info->sectorsize))
2372 /* Align the in-memory total_bytes first, and use it as correct size */
2373 device->total_bytes = round_down(device->total_bytes,
2374 fs_info->sectorsize);
2376 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2377 key.type = BTRFS_DEV_ITEM_KEY;
2378 key.offset = device->devid;
2380 trans = btrfs_start_transaction(chunk_root, 1);
2381 if (IS_ERR(trans)) {
2382 ret = PTR_ERR(trans);
2383 error("error starting transaction: %d (%s)",
2384 ret, strerror(-ret));
2388 btrfs_init_path(&path);
2389 ret = btrfs_search_slot(trans, chunk_root, &key, &path, 0, 1);
2391 error("failed to find DEV_ITEM for devid %llu", device->devid);
2396 error("failed to search chunk root: %d (%s)",
2397 ret, strerror(-ret));
2400 di = btrfs_item_ptr(path.nodes[0], path.slots[0], struct btrfs_dev_item);
2401 btrfs_set_device_total_bytes(path.nodes[0], di, device->total_bytes);
2402 btrfs_mark_buffer_dirty(path.nodes[0]);
2403 ret = btrfs_commit_transaction(trans, chunk_root);
2405 error("failed to commit current transaction: %d (%s)",
2406 ret, strerror(-ret));
2407 btrfs_release_path(&path);
2410 btrfs_release_path(&path);
2411 printf("Fixed device size for devid %llu, old size: %llu new size: %llu\n",
2412 device->devid, old_bytes, device->total_bytes);
2416 /* We haven't modified anything, it's OK to commit current trans */
2417 btrfs_commit_transaction(trans, chunk_root);
2418 btrfs_release_path(&path);
2423 * Return 0 if super block total_bytes matches all devices' total_bytes
2424 * Return >0 if super block total_bytes mismatch but fixed without problem
2425 * Return <0 if we failed to fix super block total_bytes
2427 int btrfs_fix_super_size(struct btrfs_fs_info *fs_info)
2429 struct btrfs_trans_handle *trans;
2430 struct btrfs_device *device;
2431 struct list_head *dev_list = &fs_info->fs_devices->devices;
2432 u64 total_bytes = 0;
2433 u64 old_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2436 list_for_each_entry(device, dev_list, dev_list) {
2438 * Caller should ensure this function is called after aligning
2439 * all devices' total_bytes.
2441 if (!IS_ALIGNED(device->total_bytes, fs_info->sectorsize)) {
2442 error("device %llu total_bytes %llu not aligned to %u",
2443 device->devid, device->total_bytes,
2444 fs_info->sectorsize);
2447 total_bytes += device->total_bytes;
2450 if (total_bytes == old_bytes)
2453 btrfs_set_super_total_bytes(fs_info->super_copy, total_bytes);
2455 /* Commit transaction to update all super blocks */
2456 trans = btrfs_start_transaction(fs_info->tree_root, 1);
2457 if (IS_ERR(trans)) {
2458 ret = PTR_ERR(trans);
2459 error("error starting transaction: %d (%s)",
2460 ret, strerror(-ret));
2463 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
2465 error("failed to commit current transaction: %d (%s)",
2466 ret, strerror(-ret));
2469 printf("Fixed super total bytes, old size: %llu new size: %llu\n",
2470 old_bytes, total_bytes);
2475 * Return 0 if all devices and super block sizes are good
2476 * Return >0 if any device/super size problem was found, but fixed
2477 * Return <0 if something wrong happened during fixing
2479 int btrfs_fix_device_and_super_size(struct btrfs_fs_info *fs_info)
2481 struct btrfs_device *device;
2482 struct list_head *dev_list = &fs_info->fs_devices->devices;
2483 bool have_bad_value = false;
2486 /* Seed device is not supported yet */
2487 if (fs_info->fs_devices->seed) {
2488 error("fixing device size with seed device is not supported yet");
2492 /* All devices must be set up before repairing */
2493 if (list_empty(dev_list)) {
2494 error("no device found");
2497 list_for_each_entry(device, dev_list, dev_list) {
2498 if (device->fd == -1 || !device->writeable) {
2499 error("devid %llu is missing or not writeable",
2502 "fixing device size needs all device(s) to be present and writeable");
2507 /* Repair total_bytes of each device */
2508 list_for_each_entry(device, dev_list, dev_list) {
2509 ret = btrfs_fix_device_size(fs_info, device);
2513 have_bad_value = true;
2516 /* Repair super total_byte */
2517 ret = btrfs_fix_super_size(fs_info);
2519 have_bad_value = true;
2520 if (have_bad_value) {
2522 "Fixed unaligned/mismatched total_bytes for super block and device items\n");
2525 printf("No device size related problem found\n");