2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <sys/types.h>
22 #include <uuid/uuid.h>
27 #include "transaction.h"
28 #include "print-tree.h"
31 #include "kernel-lib/raid56.h"
34 struct btrfs_device *dev;
38 static inline int nr_parity_stripes(struct map_lookup *map)
40 if (map->type & BTRFS_BLOCK_GROUP_RAID5)
42 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
48 static inline int nr_data_stripes(struct map_lookup *map)
50 return map->num_stripes - nr_parity_stripes(map);
53 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
55 static LIST_HEAD(fs_uuids);
57 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
60 struct btrfs_device *dev;
61 struct list_head *cur;
63 list_for_each(cur, head) {
64 dev = list_entry(cur, struct btrfs_device, dev_list);
65 if (dev->devid == devid &&
66 !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
73 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
75 struct list_head *cur;
76 struct btrfs_fs_devices *fs_devices;
78 list_for_each(cur, &fs_uuids) {
79 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
80 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
86 static int device_list_add(const char *path,
87 struct btrfs_super_block *disk_super,
88 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
90 struct btrfs_device *device;
91 struct btrfs_fs_devices *fs_devices;
92 u64 found_transid = btrfs_super_generation(disk_super);
94 fs_devices = find_fsid(disk_super->fsid);
96 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
99 INIT_LIST_HEAD(&fs_devices->devices);
100 list_add(&fs_devices->list, &fs_uuids);
101 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
102 fs_devices->latest_devid = devid;
103 fs_devices->latest_trans = found_transid;
104 fs_devices->lowest_devid = (u64)-1;
107 device = __find_device(&fs_devices->devices, devid,
108 disk_super->dev_item.uuid);
111 device = kzalloc(sizeof(*device), GFP_NOFS);
113 /* we can safely leave the fs_devices entry around */
117 device->devid = devid;
118 device->generation = found_transid;
119 memcpy(device->uuid, disk_super->dev_item.uuid,
121 device->name = kstrdup(path, GFP_NOFS);
126 device->label = kstrdup(disk_super->label, GFP_NOFS);
127 if (!device->label) {
132 device->total_devs = btrfs_super_num_devices(disk_super);
133 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
134 device->total_bytes =
135 btrfs_stack_device_total_bytes(&disk_super->dev_item);
137 btrfs_stack_device_bytes_used(&disk_super->dev_item);
138 list_add(&device->dev_list, &fs_devices->devices);
139 device->fs_devices = fs_devices;
140 } else if (!device->name || strcmp(device->name, path)) {
141 char *name = strdup(path);
149 if (found_transid > fs_devices->latest_trans) {
150 fs_devices->latest_devid = devid;
151 fs_devices->latest_trans = found_transid;
153 if (fs_devices->lowest_devid > devid) {
154 fs_devices->lowest_devid = devid;
156 *fs_devices_ret = fs_devices;
160 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
162 struct btrfs_fs_devices *seed_devices;
163 struct btrfs_device *device;
169 while (!list_empty(&fs_devices->devices)) {
170 device = list_entry(fs_devices->devices.next,
171 struct btrfs_device, dev_list);
172 if (device->fd != -1) {
173 if (fsync(device->fd) == -1) {
174 warning("fsync on device %llu failed: %s",
175 device->devid, strerror(errno));
178 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
179 fprintf(stderr, "Warning, could not drop caches\n");
183 device->writeable = 0;
184 list_del(&device->dev_list);
185 /* free the memory */
191 seed_devices = fs_devices->seed;
192 fs_devices->seed = NULL;
194 struct btrfs_fs_devices *orig;
197 fs_devices = seed_devices;
198 list_del(&orig->list);
202 list_del(&fs_devices->list);
209 void btrfs_close_all_devices(void)
211 struct btrfs_fs_devices *fs_devices;
213 while (!list_empty(&fs_uuids)) {
214 fs_devices = list_entry(fs_uuids.next, struct btrfs_fs_devices,
216 btrfs_close_devices(fs_devices);
220 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
223 struct list_head *head = &fs_devices->devices;
224 struct list_head *cur;
225 struct btrfs_device *device;
228 list_for_each(cur, head) {
229 device = list_entry(cur, struct btrfs_device, dev_list);
231 printk("no name for device %llu, skip it now\n", device->devid);
235 fd = open(device->name, flags);
238 error("cannot open device '%s': %s", device->name,
243 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
244 fprintf(stderr, "Warning, could not drop caches\n");
246 if (device->devid == fs_devices->latest_devid)
247 fs_devices->latest_bdev = fd;
248 if (device->devid == fs_devices->lowest_devid)
249 fs_devices->lowest_bdev = fd;
252 device->writeable = 1;
256 btrfs_close_devices(fs_devices);
260 int btrfs_scan_one_device(int fd, const char *path,
261 struct btrfs_fs_devices **fs_devices_ret,
262 u64 *total_devs, u64 super_offset, unsigned sbflags)
264 struct btrfs_super_block *disk_super;
265 char buf[BTRFS_SUPER_INFO_SIZE];
269 disk_super = (struct btrfs_super_block *)buf;
270 ret = btrfs_read_dev_super(fd, disk_super, super_offset, sbflags);
273 devid = btrfs_stack_device_id(&disk_super->dev_item);
274 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
277 *total_devs = btrfs_super_num_devices(disk_super);
279 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
285 * find_free_dev_extent_start - find free space in the specified device
286 * @device: the device which we search the free space in
287 * @num_bytes: the size of the free space that we need
288 * @search_start: the position from which to begin the search
289 * @start: store the start of the free space.
290 * @len: the size of the free space. that we find, or the size
291 * of the max free space if we don't find suitable free space
293 * this uses a pretty simple search, the expectation is that it is
294 * called very infrequently and that a given device has a small number
297 * @start is used to store the start of the free space if we find. But if we
298 * don't find suitable free space, it will be used to store the start position
299 * of the max free space.
301 * @len is used to store the size of the free space that we find.
302 * But if we don't find suitable free space, it is used to store the size of
303 * the max free space.
305 static int find_free_dev_extent_start(struct btrfs_trans_handle *trans,
306 struct btrfs_device *device, u64 num_bytes,
307 u64 search_start, u64 *start, u64 *len)
309 struct btrfs_key key;
310 struct btrfs_root *root = device->dev_root;
311 struct btrfs_dev_extent *dev_extent;
312 struct btrfs_path *path;
317 u64 search_end = device->total_bytes;
320 struct extent_buffer *l;
321 u64 min_search_start;
324 * We don't want to overwrite the superblock on the drive nor any area
325 * used by the boot loader (grub for example), so we make sure to start
326 * at an offset of at least 1MB.
328 min_search_start = max(root->fs_info->alloc_start, (u64)SZ_1M);
329 search_start = max(search_start, min_search_start);
331 path = btrfs_alloc_path();
335 max_hole_start = search_start;
338 if (search_start >= search_end) {
345 key.objectid = device->devid;
346 key.offset = search_start;
347 key.type = BTRFS_DEV_EXTENT_KEY;
349 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
353 ret = btrfs_previous_item(root, path, key.objectid, key.type);
360 slot = path->slots[0];
361 if (slot >= btrfs_header_nritems(l)) {
362 ret = btrfs_next_leaf(root, path);
370 btrfs_item_key_to_cpu(l, &key, slot);
372 if (key.objectid < device->devid)
375 if (key.objectid > device->devid)
378 if (key.type != BTRFS_DEV_EXTENT_KEY)
381 if (key.offset > search_start) {
382 hole_size = key.offset - search_start;
385 * Have to check before we set max_hole_start, otherwise
386 * we could end up sending back this offset anyway.
388 if (hole_size > max_hole_size) {
389 max_hole_start = search_start;
390 max_hole_size = hole_size;
394 * If this free space is greater than which we need,
395 * it must be the max free space that we have found
396 * until now, so max_hole_start must point to the start
397 * of this free space and the length of this free space
398 * is stored in max_hole_size. Thus, we return
399 * max_hole_start and max_hole_size and go back to the
402 if (hole_size >= num_bytes) {
408 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
409 extent_end = key.offset + btrfs_dev_extent_length(l,
411 if (extent_end > search_start)
412 search_start = extent_end;
419 * At this point, search_start should be the end of
420 * allocated dev extents, and when shrinking the device,
421 * search_end may be smaller than search_start.
423 if (search_end > search_start) {
424 hole_size = search_end - search_start;
426 if (hole_size > max_hole_size) {
427 max_hole_start = search_start;
428 max_hole_size = hole_size;
433 if (max_hole_size < num_bytes)
439 btrfs_free_path(path);
440 *start = max_hole_start;
442 *len = max_hole_size;
446 int find_free_dev_extent(struct btrfs_trans_handle *trans,
447 struct btrfs_device *device, u64 num_bytes,
450 /* FIXME use last free of some kind */
451 return find_free_dev_extent_start(trans, device,
452 num_bytes, 0, start, NULL);
455 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
456 struct btrfs_device *device,
457 u64 chunk_tree, u64 chunk_objectid,
459 u64 num_bytes, u64 *start, int convert)
462 struct btrfs_path *path;
463 struct btrfs_root *root = device->dev_root;
464 struct btrfs_dev_extent *extent;
465 struct extent_buffer *leaf;
466 struct btrfs_key key;
468 path = btrfs_alloc_path();
473 * For convert case, just skip search free dev_extent, as caller
474 * is responsible to make sure it's free.
477 ret = find_free_dev_extent(trans, device, num_bytes,
483 key.objectid = device->devid;
485 key.type = BTRFS_DEV_EXTENT_KEY;
486 ret = btrfs_insert_empty_item(trans, root, path, &key,
490 leaf = path->nodes[0];
491 extent = btrfs_item_ptr(leaf, path->slots[0],
492 struct btrfs_dev_extent);
493 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
494 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
495 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
497 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
498 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
501 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
502 btrfs_mark_buffer_dirty(leaf);
504 btrfs_free_path(path);
508 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
510 struct btrfs_path *path;
512 struct btrfs_key key;
513 struct btrfs_chunk *chunk;
514 struct btrfs_key found_key;
516 path = btrfs_alloc_path();
520 key.objectid = objectid;
521 key.offset = (u64)-1;
522 key.type = BTRFS_CHUNK_ITEM_KEY;
524 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
530 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
534 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
536 if (found_key.objectid != objectid)
539 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
541 *offset = found_key.offset +
542 btrfs_chunk_length(path->nodes[0], chunk);
547 btrfs_free_path(path);
551 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
555 struct btrfs_key key;
556 struct btrfs_key found_key;
558 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
559 key.type = BTRFS_DEV_ITEM_KEY;
560 key.offset = (u64)-1;
562 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
568 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
573 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
575 *objectid = found_key.offset + 1;
579 btrfs_release_path(path);
584 * the device information is stored in the chunk root
585 * the btrfs_device struct should be fully filled in
587 int btrfs_add_device(struct btrfs_trans_handle *trans,
588 struct btrfs_root *root,
589 struct btrfs_device *device)
592 struct btrfs_path *path;
593 struct btrfs_dev_item *dev_item;
594 struct extent_buffer *leaf;
595 struct btrfs_key key;
599 root = root->fs_info->chunk_root;
601 path = btrfs_alloc_path();
605 ret = find_next_devid(root, path, &free_devid);
609 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
610 key.type = BTRFS_DEV_ITEM_KEY;
611 key.offset = free_devid;
613 ret = btrfs_insert_empty_item(trans, root, path, &key,
618 leaf = path->nodes[0];
619 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
621 device->devid = free_devid;
622 btrfs_set_device_id(leaf, dev_item, device->devid);
623 btrfs_set_device_generation(leaf, dev_item, 0);
624 btrfs_set_device_type(leaf, dev_item, device->type);
625 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
626 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
627 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
628 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
629 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
630 btrfs_set_device_group(leaf, dev_item, 0);
631 btrfs_set_device_seek_speed(leaf, dev_item, 0);
632 btrfs_set_device_bandwidth(leaf, dev_item, 0);
633 btrfs_set_device_start_offset(leaf, dev_item, 0);
635 ptr = (unsigned long)btrfs_device_uuid(dev_item);
636 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
637 ptr = (unsigned long)btrfs_device_fsid(dev_item);
638 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
639 btrfs_mark_buffer_dirty(leaf);
643 btrfs_free_path(path);
647 int btrfs_update_device(struct btrfs_trans_handle *trans,
648 struct btrfs_device *device)
651 struct btrfs_path *path;
652 struct btrfs_root *root;
653 struct btrfs_dev_item *dev_item;
654 struct extent_buffer *leaf;
655 struct btrfs_key key;
657 root = device->dev_root->fs_info->chunk_root;
659 path = btrfs_alloc_path();
663 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
664 key.type = BTRFS_DEV_ITEM_KEY;
665 key.offset = device->devid;
667 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
676 leaf = path->nodes[0];
677 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
679 btrfs_set_device_id(leaf, dev_item, device->devid);
680 btrfs_set_device_type(leaf, dev_item, device->type);
681 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
682 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
683 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
684 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
685 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
686 btrfs_mark_buffer_dirty(leaf);
689 btrfs_free_path(path);
693 int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
694 struct btrfs_chunk *chunk, int item_size)
696 struct btrfs_super_block *super_copy = fs_info->super_copy;
697 struct btrfs_disk_key disk_key;
701 array_size = btrfs_super_sys_array_size(super_copy);
702 if (array_size + item_size + sizeof(disk_key)
703 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
706 ptr = super_copy->sys_chunk_array + array_size;
707 btrfs_cpu_key_to_disk(&disk_key, key);
708 memcpy(ptr, &disk_key, sizeof(disk_key));
709 ptr += sizeof(disk_key);
710 memcpy(ptr, chunk, item_size);
711 item_size += sizeof(disk_key);
712 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
716 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
719 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
721 else if (type & BTRFS_BLOCK_GROUP_RAID10)
722 return calc_size * (num_stripes / sub_stripes);
723 else if (type & BTRFS_BLOCK_GROUP_RAID5)
724 return calc_size * (num_stripes - 1);
725 else if (type & BTRFS_BLOCK_GROUP_RAID6)
726 return calc_size * (num_stripes - 2);
728 return calc_size * num_stripes;
732 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
734 /* TODO, add a way to store the preferred stripe size */
735 return BTRFS_STRIPE_LEN;
739 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
741 * It is not equal to "device->total_bytes - device->bytes_used".
742 * We do not allocate any chunk in 1M at beginning of device, and not
743 * allowed to allocate any chunk before alloc_start if it is specified.
744 * So search holes from max(1M, alloc_start) to device->total_bytes.
746 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
747 struct btrfs_device *device,
750 struct btrfs_path *path;
751 struct btrfs_root *root = device->dev_root;
752 struct btrfs_key key;
753 struct btrfs_dev_extent *dev_extent = NULL;
754 struct extent_buffer *l;
755 u64 search_start = root->fs_info->alloc_start;
756 u64 search_end = device->total_bytes;
762 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
764 path = btrfs_alloc_path();
768 key.objectid = device->devid;
769 key.offset = root->fs_info->alloc_start;
770 key.type = BTRFS_DEV_EXTENT_KEY;
773 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
776 ret = btrfs_previous_item(root, path, 0, key.type);
782 slot = path->slots[0];
783 if (slot >= btrfs_header_nritems(l)) {
784 ret = btrfs_next_leaf(root, path);
791 btrfs_item_key_to_cpu(l, &key, slot);
793 if (key.objectid < device->devid)
795 if (key.objectid > device->devid)
797 if (key.type != BTRFS_DEV_EXTENT_KEY)
799 if (key.offset > search_end)
801 if (key.offset > search_start)
802 free_bytes += key.offset - search_start;
804 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
805 extent_end = key.offset + btrfs_dev_extent_length(l,
807 if (extent_end > search_start)
808 search_start = extent_end;
809 if (search_start > search_end)
816 if (search_start < search_end)
817 free_bytes += search_end - search_start;
819 *avail_bytes = free_bytes;
822 btrfs_free_path(path);
826 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
827 - sizeof(struct btrfs_item) \
828 - sizeof(struct btrfs_chunk)) \
829 / sizeof(struct btrfs_stripe) + 1)
831 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
832 - 2 * sizeof(struct btrfs_disk_key) \
833 - 2 * sizeof(struct btrfs_chunk)) \
834 / sizeof(struct btrfs_stripe) + 1)
836 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
837 struct btrfs_root *extent_root, u64 *start,
838 u64 *num_bytes, u64 type)
841 struct btrfs_fs_info *info = extent_root->fs_info;
842 struct btrfs_root *chunk_root = info->chunk_root;
843 struct btrfs_stripe *stripes;
844 struct btrfs_device *device = NULL;
845 struct btrfs_chunk *chunk;
846 struct list_head private_devs;
847 struct list_head *dev_list = &info->fs_devices->devices;
848 struct list_head *cur;
849 struct map_lookup *map;
850 int min_stripe_size = SZ_1M;
851 u64 calc_size = SZ_8M;
853 u64 max_chunk_size = 4 * calc_size;
864 int stripe_len = BTRFS_STRIPE_LEN;
865 struct btrfs_key key;
868 if (list_empty(dev_list)) {
872 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
873 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
874 BTRFS_BLOCK_GROUP_RAID10 |
875 BTRFS_BLOCK_GROUP_DUP)) {
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(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1006 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1007 key.type = BTRFS_CHUNK_ITEM_KEY;
1008 key.offset = offset;
1010 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1014 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1020 stripes = &chunk->stripe;
1021 *num_bytes = chunk_bytes_by_type(type, calc_size,
1022 num_stripes, sub_stripes);
1024 while(index < num_stripes) {
1025 struct btrfs_stripe *stripe;
1026 BUG_ON(list_empty(&private_devs));
1027 cur = private_devs.next;
1028 device = list_entry(cur, struct btrfs_device, dev_list);
1030 /* loop over this device again if we're doing a dup group */
1031 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1032 (index == num_stripes - 1))
1033 list_move_tail(&device->dev_list, dev_list);
1035 ret = btrfs_alloc_dev_extent(trans, device,
1036 info->chunk_root->root_key.objectid,
1037 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1038 calc_size, &dev_offset, 0);
1041 device->bytes_used += calc_size;
1042 ret = btrfs_update_device(trans, device);
1045 map->stripes[index].dev = device;
1046 map->stripes[index].physical = dev_offset;
1047 stripe = stripes + index;
1048 btrfs_set_stack_stripe_devid(stripe, device->devid);
1049 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1050 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1053 BUG_ON(!list_empty(&private_devs));
1055 /* key was set above */
1056 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1057 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1058 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1059 btrfs_set_stack_chunk_type(chunk, type);
1060 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1061 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1062 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1063 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1064 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1065 map->sector_size = info->sectorsize;
1066 map->stripe_len = stripe_len;
1067 map->io_align = stripe_len;
1068 map->io_width = stripe_len;
1070 map->num_stripes = num_stripes;
1071 map->sub_stripes = sub_stripes;
1073 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1074 btrfs_chunk_item_size(num_stripes));
1076 *start = key.offset;;
1078 map->ce.start = key.offset;
1079 map->ce.size = *num_bytes;
1081 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));
1095 * Alloc a DATA chunk with SINGLE profile.
1097 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1098 * (btrfs logical bytenr == on-disk bytenr)
1099 * For that case, caller must make sure the chunk and dev_extent are not
1102 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1103 struct btrfs_root *extent_root, u64 *start,
1104 u64 num_bytes, u64 type, int convert)
1107 struct btrfs_fs_info *info = extent_root->fs_info;
1108 struct btrfs_root *chunk_root = info->chunk_root;
1109 struct btrfs_stripe *stripes;
1110 struct btrfs_device *device = NULL;
1111 struct btrfs_chunk *chunk;
1112 struct list_head *dev_list = &info->fs_devices->devices;
1113 struct list_head *cur;
1114 struct map_lookup *map;
1115 u64 calc_size = SZ_8M;
1116 int num_stripes = 1;
1117 int sub_stripes = 0;
1120 int stripe_len = BTRFS_STRIPE_LEN;
1121 struct btrfs_key key;
1123 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1124 key.type = BTRFS_CHUNK_ITEM_KEY;
1126 if (*start != round_down(*start, info->sectorsize)) {
1127 error("DATA chunk start not sectorsize aligned: %llu",
1128 (unsigned long long)*start);
1131 key.offset = *start;
1132 dev_offset = *start;
1136 ret = find_next_chunk(chunk_root,
1137 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1144 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1148 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1154 stripes = &chunk->stripe;
1155 calc_size = num_bytes;
1158 cur = dev_list->next;
1159 device = list_entry(cur, struct btrfs_device, dev_list);
1161 while (index < num_stripes) {
1162 struct btrfs_stripe *stripe;
1164 ret = btrfs_alloc_dev_extent(trans, device,
1165 info->chunk_root->root_key.objectid,
1166 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1167 calc_size, &dev_offset, convert);
1170 device->bytes_used += calc_size;
1171 ret = btrfs_update_device(trans, device);
1174 map->stripes[index].dev = device;
1175 map->stripes[index].physical = dev_offset;
1176 stripe = stripes + index;
1177 btrfs_set_stack_stripe_devid(stripe, device->devid);
1178 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1179 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1183 /* key was set above */
1184 btrfs_set_stack_chunk_length(chunk, num_bytes);
1185 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1186 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1187 btrfs_set_stack_chunk_type(chunk, type);
1188 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1189 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1190 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1191 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1192 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1193 map->sector_size = info->sectorsize;
1194 map->stripe_len = stripe_len;
1195 map->io_align = stripe_len;
1196 map->io_width = stripe_len;
1198 map->num_stripes = num_stripes;
1199 map->sub_stripes = sub_stripes;
1201 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1202 btrfs_chunk_item_size(num_stripes));
1205 *start = key.offset;
1207 map->ce.start = key.offset;
1208 map->ce.size = num_bytes;
1210 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1217 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
1219 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1220 struct cache_extent *ce;
1221 struct map_lookup *map;
1224 ce = search_cache_extent(&map_tree->cache_tree, logical);
1226 fprintf(stderr, "No mapping for %llu-%llu\n",
1227 (unsigned long long)logical,
1228 (unsigned long long)logical+len);
1231 if (ce->start > logical || ce->start + ce->size < logical) {
1232 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1233 "%llu-%llu\n", (unsigned long long)logical,
1234 (unsigned long long)logical+len,
1235 (unsigned long long)ce->start,
1236 (unsigned long long)ce->start + ce->size);
1239 map = container_of(ce, struct map_lookup, ce);
1241 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1242 ret = map->num_stripes;
1243 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1244 ret = map->sub_stripes;
1245 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1247 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1254 int btrfs_next_bg(struct btrfs_fs_info *fs_info, u64 *logical,
1255 u64 *size, u64 type)
1257 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1258 struct cache_extent *ce;
1259 struct map_lookup *map;
1262 ce = search_cache_extent(&map_tree->cache_tree, cur);
1266 * only jump to next bg if our cur is not 0
1267 * As the initial logical for btrfs_next_bg() is 0, and
1268 * if we jump to next bg, we skipped a valid bg.
1271 ce = next_cache_extent(ce);
1277 map = container_of(ce, struct map_lookup, ce);
1278 if (map->type & type) {
1279 *logical = ce->start;
1288 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
1289 u64 chunk_start, u64 physical, u64 devid,
1290 u64 **logical, int *naddrs, int *stripe_len)
1292 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1293 struct cache_extent *ce;
1294 struct map_lookup *map;
1302 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1304 map = container_of(ce, struct map_lookup, ce);
1307 rmap_len = map->stripe_len;
1308 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1309 length = ce->size / (map->num_stripes / map->sub_stripes);
1310 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1311 length = ce->size / map->num_stripes;
1312 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1313 BTRFS_BLOCK_GROUP_RAID6)) {
1314 length = ce->size / nr_data_stripes(map);
1315 rmap_len = map->stripe_len * nr_data_stripes(map);
1318 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1320 for (i = 0; i < map->num_stripes; i++) {
1321 if (devid && map->stripes[i].dev->devid != devid)
1323 if (map->stripes[i].physical > physical ||
1324 map->stripes[i].physical + length <= physical)
1327 stripe_nr = (physical - map->stripes[i].physical) /
1330 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1331 stripe_nr = (stripe_nr * map->num_stripes + i) /
1333 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1334 stripe_nr = stripe_nr * map->num_stripes + i;
1335 } /* else if RAID[56], multiply by nr_data_stripes().
1336 * Alternatively, just use rmap_len below instead of
1337 * map->stripe_len */
1339 bytenr = ce->start + stripe_nr * rmap_len;
1340 for (j = 0; j < nr; j++) {
1341 if (buf[j] == bytenr)
1350 *stripe_len = rmap_len;
1355 static inline int parity_smaller(u64 a, u64 b)
1360 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1361 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1363 struct btrfs_bio_stripe s;
1370 for (i = 0; i < bbio->num_stripes - 1; i++) {
1371 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1372 s = bbio->stripes[i];
1374 bbio->stripes[i] = bbio->stripes[i+1];
1375 raid_map[i] = raid_map[i+1];
1376 bbio->stripes[i+1] = s;
1384 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1385 u64 logical, u64 *length,
1386 struct btrfs_multi_bio **multi_ret, int mirror_num,
1389 return __btrfs_map_block(fs_info, rw, logical, length, NULL,
1390 multi_ret, mirror_num, raid_map_ret);
1393 int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1394 u64 logical, u64 *length, u64 *type,
1395 struct btrfs_multi_bio **multi_ret, int mirror_num,
1398 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1399 struct cache_extent *ce;
1400 struct map_lookup *map;
1404 u64 *raid_map = NULL;
1405 int stripes_allocated = 8;
1406 int stripes_required = 1;
1409 struct btrfs_multi_bio *multi = NULL;
1411 if (multi_ret && rw == READ) {
1412 stripes_allocated = 1;
1415 ce = search_cache_extent(&map_tree->cache_tree, logical);
1421 if (ce->start > logical) {
1423 *length = ce->start - logical;
1428 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1433 map = container_of(ce, struct map_lookup, ce);
1434 offset = logical - ce->start;
1437 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1438 BTRFS_BLOCK_GROUP_DUP)) {
1439 stripes_required = map->num_stripes;
1440 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1441 stripes_required = map->sub_stripes;
1444 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1445 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1446 /* RAID[56] write or recovery. Return all stripes */
1447 stripes_required = map->num_stripes;
1449 /* Only allocate the map if we've already got a large enough multi_ret */
1450 if (stripes_allocated >= stripes_required) {
1451 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1459 /* if our multi bio struct is too small, back off and try again */
1460 if (multi_ret && stripes_allocated < stripes_required) {
1461 stripes_allocated = stripes_required;
1468 * stripe_nr counts the total number of stripes we have to stride
1469 * to get to this block
1471 stripe_nr = stripe_nr / map->stripe_len;
1473 stripe_offset = stripe_nr * map->stripe_len;
1474 BUG_ON(offset < stripe_offset);
1476 /* stripe_offset is the offset of this block in its stripe*/
1477 stripe_offset = offset - stripe_offset;
1479 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1480 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1481 BTRFS_BLOCK_GROUP_RAID10 |
1482 BTRFS_BLOCK_GROUP_DUP)) {
1483 /* we limit the length of each bio to what fits in a stripe */
1484 *length = min_t(u64, ce->size - offset,
1485 map->stripe_len - stripe_offset);
1487 *length = ce->size - offset;
1493 multi->num_stripes = 1;
1495 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1497 multi->num_stripes = map->num_stripes;
1498 else if (mirror_num)
1499 stripe_index = mirror_num - 1;
1501 stripe_index = stripe_nr % map->num_stripes;
1502 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1503 int factor = map->num_stripes / map->sub_stripes;
1505 stripe_index = stripe_nr % factor;
1506 stripe_index *= map->sub_stripes;
1509 multi->num_stripes = map->sub_stripes;
1510 else if (mirror_num)
1511 stripe_index += mirror_num - 1;
1513 stripe_nr = stripe_nr / factor;
1514 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1516 multi->num_stripes = map->num_stripes;
1517 else if (mirror_num)
1518 stripe_index = mirror_num - 1;
1519 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1520 BTRFS_BLOCK_GROUP_RAID6)) {
1525 u64 raid56_full_stripe_start;
1526 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1529 * align the start of our data stripe in the logical
1532 raid56_full_stripe_start = offset / full_stripe_len;
1533 raid56_full_stripe_start *= full_stripe_len;
1535 /* get the data stripe number */
1536 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1537 stripe_nr = stripe_nr / nr_data_stripes(map);
1539 /* Work out the disk rotation on this stripe-set */
1540 rot = stripe_nr % map->num_stripes;
1542 /* Fill in the logical address of each stripe */
1543 tmp = stripe_nr * nr_data_stripes(map);
1545 for (i = 0; i < nr_data_stripes(map); i++)
1546 raid_map[(i+rot) % map->num_stripes] =
1547 ce->start + (tmp + i) * map->stripe_len;
1549 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1550 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1551 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1553 *length = map->stripe_len;
1556 multi->num_stripes = map->num_stripes;
1558 stripe_index = stripe_nr % nr_data_stripes(map);
1559 stripe_nr = stripe_nr / nr_data_stripes(map);
1562 * Mirror #0 or #1 means the original data block.
1563 * Mirror #2 is RAID5 parity block.
1564 * Mirror #3 is RAID6 Q block.
1567 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1569 /* We distribute the parity blocks across stripes */
1570 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1574 * after this do_div call, stripe_nr is the number of stripes
1575 * on this device we have to walk to find the data, and
1576 * stripe_index is the number of our device in the stripe array
1578 stripe_index = stripe_nr % map->num_stripes;
1579 stripe_nr = stripe_nr / map->num_stripes;
1581 BUG_ON(stripe_index >= map->num_stripes);
1583 for (i = 0; i < multi->num_stripes; i++) {
1584 multi->stripes[i].physical =
1585 map->stripes[stripe_index].physical + stripe_offset +
1586 stripe_nr * map->stripe_len;
1587 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1596 sort_parity_stripes(multi, raid_map);
1597 *raid_map_ret = raid_map;
1603 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
1606 struct btrfs_device *device;
1607 struct btrfs_fs_devices *cur_devices;
1609 cur_devices = fs_info->fs_devices;
1610 while (cur_devices) {
1612 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1613 fs_info->ignore_fsid_mismatch)) {
1614 device = __find_device(&cur_devices->devices,
1619 cur_devices = cur_devices->seed;
1624 struct btrfs_device *
1625 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1626 u64 devid, int instance)
1628 struct list_head *head = &fs_devices->devices;
1629 struct btrfs_device *dev;
1632 list_for_each_entry(dev, head, dev_list) {
1633 if (dev->devid == devid && num_found++ == instance)
1639 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1641 struct cache_extent *ce;
1642 struct map_lookup *map;
1643 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1648 * During chunk recovering, we may fail to find block group's
1649 * corresponding chunk, we will rebuild it later
1651 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1652 if (!root->fs_info->is_chunk_recover)
1657 map = container_of(ce, struct map_lookup, ce);
1658 for (i = 0; i < map->num_stripes; i++) {
1659 if (!map->stripes[i].dev->writeable) {
1668 static struct btrfs_device *fill_missing_device(u64 devid)
1670 struct btrfs_device *device;
1672 device = kzalloc(sizeof(*device), GFP_NOFS);
1673 device->devid = devid;
1679 * slot == -1: SYSTEM chunk
1680 * return -EIO on error, otherwise return 0
1682 int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
1683 struct extent_buffer *leaf,
1684 struct btrfs_chunk *chunk,
1685 int slot, u64 logical)
1692 u32 chunk_ondisk_size;
1693 u32 sectorsize = fs_info->sectorsize;
1695 length = btrfs_chunk_length(leaf, chunk);
1696 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1697 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1698 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1699 type = btrfs_chunk_type(leaf, chunk);
1702 * These valid checks may be insufficient to cover every corner cases.
1704 if (!IS_ALIGNED(logical, sectorsize)) {
1705 error("invalid chunk logical %llu", logical);
1708 if (btrfs_chunk_sector_size(leaf, chunk) != sectorsize) {
1709 error("invalid chunk sectorsize %llu",
1710 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1713 if (!length || !IS_ALIGNED(length, sectorsize)) {
1714 error("invalid chunk length %llu", length);
1717 if (stripe_len != BTRFS_STRIPE_LEN) {
1718 error("invalid chunk stripe length: %llu", stripe_len);
1721 /* Check on chunk item type */
1722 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1723 error("invalid chunk type %llu", type);
1726 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1727 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1728 error("unrecognized chunk type: %llu",
1729 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1730 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1733 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1734 error("missing chunk type flag: %llu", type);
1737 if (!(is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) ||
1738 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0)) {
1739 error("conflicting chunk type detected: %llu", type);
1742 if ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
1743 !is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1744 error("conflicting chunk profile detected: %llu", type);
1748 chunk_ondisk_size = btrfs_chunk_item_size(num_stripes);
1750 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1751 * it can't exceed the system chunk array size
1752 * For normal chunk, it should match its chunk item size.
1754 if (num_stripes < 1 ||
1755 (slot == -1 && chunk_ondisk_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1756 (slot >= 0 && chunk_ondisk_size > btrfs_item_size_nr(leaf, slot))) {
1757 error("invalid num_stripes: %u", num_stripes);
1761 * Device number check against profile
1763 if ((type & BTRFS_BLOCK_GROUP_RAID10 && (sub_stripes != 2 ||
1764 !IS_ALIGNED(num_stripes, sub_stripes))) ||
1765 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1766 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1767 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1768 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1769 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1770 num_stripes != 1)) {
1771 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1772 num_stripes, sub_stripes,
1773 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1781 * Slot is used to verify the chunk item is valid
1783 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1785 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
1786 struct extent_buffer *leaf,
1787 struct btrfs_chunk *chunk, int slot)
1789 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1790 struct map_lookup *map;
1791 struct cache_extent *ce;
1795 u8 uuid[BTRFS_UUID_SIZE];
1800 logical = key->offset;
1801 length = btrfs_chunk_length(leaf, chunk);
1802 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1803 /* Validation check */
1804 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, slot, logical);
1806 error("%s checksums match, but it has an invalid chunk, %s",
1807 (slot == -1) ? "Superblock" : "Metadata",
1808 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1812 ce = search_cache_extent(&map_tree->cache_tree, logical);
1814 /* already mapped? */
1815 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1819 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1823 map->ce.start = logical;
1824 map->ce.size = length;
1825 map->num_stripes = num_stripes;
1826 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1827 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1828 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1829 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1830 map->type = btrfs_chunk_type(leaf, chunk);
1831 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1833 for (i = 0; i < num_stripes; i++) {
1834 map->stripes[i].physical =
1835 btrfs_stripe_offset_nr(leaf, chunk, i);
1836 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1837 read_extent_buffer(leaf, uuid, (unsigned long)
1838 btrfs_stripe_dev_uuid_nr(chunk, i),
1840 map->stripes[i].dev = btrfs_find_device(fs_info, devid, uuid,
1842 if (!map->stripes[i].dev) {
1843 map->stripes[i].dev = fill_missing_device(devid);
1844 printf("warning, device %llu is missing\n",
1845 (unsigned long long)devid);
1846 list_add(&map->stripes[i].dev->dev_list,
1847 &fs_info->fs_devices->devices);
1851 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1857 static int fill_device_from_item(struct extent_buffer *leaf,
1858 struct btrfs_dev_item *dev_item,
1859 struct btrfs_device *device)
1863 device->devid = btrfs_device_id(leaf, dev_item);
1864 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1865 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1866 device->type = btrfs_device_type(leaf, dev_item);
1867 device->io_align = btrfs_device_io_align(leaf, dev_item);
1868 device->io_width = btrfs_device_io_width(leaf, dev_item);
1869 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1871 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1872 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1877 static int open_seed_devices(struct btrfs_fs_info *fs_info, u8 *fsid)
1879 struct btrfs_fs_devices *fs_devices;
1882 fs_devices = fs_info->fs_devices->seed;
1883 while (fs_devices) {
1884 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1888 fs_devices = fs_devices->seed;
1891 fs_devices = find_fsid(fsid);
1893 /* missing all seed devices */
1894 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1899 INIT_LIST_HEAD(&fs_devices->devices);
1900 list_add(&fs_devices->list, &fs_uuids);
1901 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1904 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1908 fs_devices->seed = fs_info->fs_devices->seed;
1909 fs_info->fs_devices->seed = fs_devices;
1914 static int read_one_dev(struct btrfs_fs_info *fs_info,
1915 struct extent_buffer *leaf,
1916 struct btrfs_dev_item *dev_item)
1918 struct btrfs_device *device;
1921 u8 fs_uuid[BTRFS_UUID_SIZE];
1922 u8 dev_uuid[BTRFS_UUID_SIZE];
1924 devid = btrfs_device_id(leaf, dev_item);
1925 read_extent_buffer(leaf, dev_uuid,
1926 (unsigned long)btrfs_device_uuid(dev_item),
1928 read_extent_buffer(leaf, fs_uuid,
1929 (unsigned long)btrfs_device_fsid(dev_item),
1932 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
1933 ret = open_seed_devices(fs_info, fs_uuid);
1938 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
1940 device = kzalloc(sizeof(*device), GFP_NOFS);
1944 list_add(&device->dev_list,
1945 &fs_info->fs_devices->devices);
1948 fill_device_from_item(leaf, dev_item, device);
1949 device->dev_root = fs_info->dev_root;
1953 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
1955 struct btrfs_super_block *super_copy = fs_info->super_copy;
1956 struct extent_buffer *sb;
1957 struct btrfs_disk_key *disk_key;
1958 struct btrfs_chunk *chunk;
1960 unsigned long sb_array_offset;
1966 struct btrfs_key key;
1968 sb = btrfs_find_create_tree_block(fs_info,
1969 BTRFS_SUPER_INFO_OFFSET,
1970 BTRFS_SUPER_INFO_SIZE);
1973 btrfs_set_buffer_uptodate(sb);
1974 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1975 array_size = btrfs_super_sys_array_size(super_copy);
1977 array_ptr = super_copy->sys_chunk_array;
1978 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1981 while (cur_offset < array_size) {
1982 disk_key = (struct btrfs_disk_key *)array_ptr;
1983 len = sizeof(*disk_key);
1984 if (cur_offset + len > array_size)
1985 goto out_short_read;
1987 btrfs_disk_key_to_cpu(&key, disk_key);
1990 sb_array_offset += len;
1993 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1994 chunk = (struct btrfs_chunk *)sb_array_offset;
1996 * At least one btrfs_chunk with one stripe must be
1997 * present, exact stripe count check comes afterwards
1999 len = btrfs_chunk_item_size(1);
2000 if (cur_offset + len > array_size)
2001 goto out_short_read;
2003 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2006 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
2007 num_stripes, cur_offset);
2012 len = btrfs_chunk_item_size(num_stripes);
2013 if (cur_offset + len > array_size)
2014 goto out_short_read;
2016 ret = read_one_chunk(fs_info, &key, sb, chunk, -1);
2021 "ERROR: unexpected item type %u in sys_array at offset %u\n",
2022 (u32)key.type, cur_offset);
2027 sb_array_offset += len;
2030 free_extent_buffer(sb);
2034 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
2036 free_extent_buffer(sb);
2040 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
2042 struct btrfs_path *path;
2043 struct extent_buffer *leaf;
2044 struct btrfs_key key;
2045 struct btrfs_key found_key;
2046 struct btrfs_root *root = fs_info->chunk_root;
2050 path = btrfs_alloc_path();
2055 * Read all device items, and then all the chunk items. All
2056 * device items are found before any chunk item (their object id
2057 * is smaller than the lowest possible object id for a chunk
2058 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
2060 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2063 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2067 leaf = path->nodes[0];
2068 slot = path->slots[0];
2069 if (slot >= btrfs_header_nritems(leaf)) {
2070 ret = btrfs_next_leaf(root, path);
2077 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2078 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2079 struct btrfs_dev_item *dev_item;
2080 dev_item = btrfs_item_ptr(leaf, slot,
2081 struct btrfs_dev_item);
2082 ret = read_one_dev(fs_info, leaf, dev_item);
2084 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2085 struct btrfs_chunk *chunk;
2086 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2087 ret = read_one_chunk(fs_info, &found_key, leaf, chunk,
2096 btrfs_free_path(path);
2100 struct list_head *btrfs_scanned_uuids(void)
2105 static int rmw_eb(struct btrfs_fs_info *info,
2106 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2109 unsigned long orig_off = 0;
2110 unsigned long dest_off = 0;
2111 unsigned long copy_len = eb->len;
2113 ret = read_whole_eb(info, eb, 0);
2117 if (eb->start + eb->len <= orig_eb->start ||
2118 eb->start >= orig_eb->start + orig_eb->len)
2121 * | ----- orig_eb ------- |
2122 * | ----- stripe ------- |
2123 * | ----- orig_eb ------- |
2124 * | ----- orig_eb ------- |
2126 if (eb->start > orig_eb->start)
2127 orig_off = eb->start - orig_eb->start;
2128 if (orig_eb->start > eb->start)
2129 dest_off = orig_eb->start - eb->start;
2131 if (copy_len > orig_eb->len - orig_off)
2132 copy_len = orig_eb->len - orig_off;
2133 if (copy_len > eb->len - dest_off)
2134 copy_len = eb->len - dest_off;
2136 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2140 static int split_eb_for_raid56(struct btrfs_fs_info *info,
2141 struct extent_buffer *orig_eb,
2142 struct extent_buffer **ebs,
2143 u64 stripe_len, u64 *raid_map,
2146 struct extent_buffer **tmp_ebs;
2147 u64 start = orig_eb->start;
2152 tmp_ebs = calloc(num_stripes, sizeof(*tmp_ebs));
2156 /* Alloc memory in a row for data stripes */
2157 for (i = 0; i < num_stripes; i++) {
2158 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2161 tmp_ebs[i] = calloc(1, sizeof(**tmp_ebs) + stripe_len);
2168 for (i = 0; i < num_stripes; i++) {
2169 struct extent_buffer *eb = tmp_ebs[i];
2171 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2174 eb->start = raid_map[i];
2175 eb->len = stripe_len;
2179 eb->dev_bytenr = (u64)-1;
2181 this_eb_start = raid_map[i];
2183 if (start > this_eb_start ||
2184 start + orig_eb->len < this_eb_start + stripe_len) {
2185 ret = rmw_eb(info, eb, orig_eb);
2189 memcpy(eb->data, orig_eb->data + eb->start - start,
2197 for (i = 0; i < num_stripes; i++)
2203 int write_raid56_with_parity(struct btrfs_fs_info *info,
2204 struct extent_buffer *eb,
2205 struct btrfs_multi_bio *multi,
2206 u64 stripe_len, u64 *raid_map)
2208 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2211 int alloc_size = eb->len;
2214 ebs = malloc(sizeof(*ebs) * multi->num_stripes);
2215 pointers = malloc(sizeof(*pointers) * multi->num_stripes);
2216 if (!ebs || !pointers) {
2222 if (stripe_len > alloc_size)
2223 alloc_size = stripe_len;
2225 ret = split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2226 multi->num_stripes);
2230 for (i = 0; i < multi->num_stripes; i++) {
2231 struct extent_buffer *new_eb;
2232 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2233 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2234 ebs[i]->fd = multi->stripes[i].dev->fd;
2235 multi->stripes[i].dev->total_ios++;
2236 if (ebs[i]->start != raid_map[i]) {
2238 goto out_free_split;
2242 new_eb = malloc(sizeof(*eb) + alloc_size);
2245 goto out_free_split;
2247 new_eb->dev_bytenr = multi->stripes[i].physical;
2248 new_eb->fd = multi->stripes[i].dev->fd;
2249 multi->stripes[i].dev->total_ios++;
2250 new_eb->len = stripe_len;
2252 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2254 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2258 ebs[multi->num_stripes - 2] = p_eb;
2259 ebs[multi->num_stripes - 1] = q_eb;
2261 for (i = 0; i < multi->num_stripes; i++)
2262 pointers[i] = ebs[i]->data;
2264 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2266 ebs[multi->num_stripes - 1] = p_eb;
2267 for (i = 0; i < multi->num_stripes; i++)
2268 pointers[i] = ebs[i]->data;
2269 ret = raid5_gen_result(multi->num_stripes, stripe_len,
2270 multi->num_stripes - 1, pointers);
2272 goto out_free_split;
2275 for (i = 0; i < multi->num_stripes; i++) {
2276 ret = write_extent_to_disk(ebs[i]);
2278 goto out_free_split;
2282 for (i = 0; i < multi->num_stripes; i++) {
2294 * Get stripe length from chunk item and its stripe items
2296 * Caller should only call this function after validating the chunk item
2297 * by using btrfs_check_chunk_valid().
2299 u64 btrfs_stripe_length(struct btrfs_fs_info *fs_info,
2300 struct extent_buffer *leaf,
2301 struct btrfs_chunk *chunk)
2305 u32 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2306 u64 profile = btrfs_chunk_type(leaf, chunk) &
2307 BTRFS_BLOCK_GROUP_PROFILE_MASK;
2309 chunk_len = btrfs_chunk_length(leaf, chunk);
2312 case 0: /* Single profile */
2313 case BTRFS_BLOCK_GROUP_RAID1:
2314 case BTRFS_BLOCK_GROUP_DUP:
2315 stripe_len = chunk_len;
2317 case BTRFS_BLOCK_GROUP_RAID0:
2318 stripe_len = chunk_len / num_stripes;
2320 case BTRFS_BLOCK_GROUP_RAID5:
2321 stripe_len = chunk_len / (num_stripes - 1);
2323 case BTRFS_BLOCK_GROUP_RAID6:
2324 stripe_len = chunk_len / (num_stripes - 2);
2326 case BTRFS_BLOCK_GROUP_RAID10:
2327 stripe_len = chunk_len / (num_stripes /
2328 btrfs_chunk_sub_stripes(leaf, chunk));
2331 /* Invalid chunk profile found */