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_root *root,
694 struct btrfs_key *key,
695 struct btrfs_chunk *chunk, int item_size)
697 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
698 struct btrfs_disk_key disk_key;
702 array_size = btrfs_super_sys_array_size(super_copy);
703 if (array_size + item_size + sizeof(disk_key)
704 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
707 ptr = super_copy->sys_chunk_array + array_size;
708 btrfs_cpu_key_to_disk(&disk_key, key);
709 memcpy(ptr, &disk_key, sizeof(disk_key));
710 ptr += sizeof(disk_key);
711 memcpy(ptr, chunk, item_size);
712 item_size += sizeof(disk_key);
713 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
717 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
720 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
722 else if (type & BTRFS_BLOCK_GROUP_RAID10)
723 return calc_size * (num_stripes / sub_stripes);
724 else if (type & BTRFS_BLOCK_GROUP_RAID5)
725 return calc_size * (num_stripes - 1);
726 else if (type & BTRFS_BLOCK_GROUP_RAID6)
727 return calc_size * (num_stripes - 2);
729 return calc_size * num_stripes;
733 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
735 /* TODO, add a way to store the preferred stripe size */
736 return BTRFS_STRIPE_LEN;
740 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
742 * It is not equal to "device->total_bytes - device->bytes_used".
743 * We do not allocate any chunk in 1M at beginning of device, and not
744 * allowed to allocate any chunk before alloc_start if it is specified.
745 * So search holes from max(1M, alloc_start) to device->total_bytes.
747 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
748 struct btrfs_device *device,
751 struct btrfs_path *path;
752 struct btrfs_root *root = device->dev_root;
753 struct btrfs_key key;
754 struct btrfs_dev_extent *dev_extent = NULL;
755 struct extent_buffer *l;
756 u64 search_start = root->fs_info->alloc_start;
757 u64 search_end = device->total_bytes;
763 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
765 path = btrfs_alloc_path();
769 key.objectid = device->devid;
770 key.offset = root->fs_info->alloc_start;
771 key.type = BTRFS_DEV_EXTENT_KEY;
774 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
777 ret = btrfs_previous_item(root, path, 0, key.type);
783 slot = path->slots[0];
784 if (slot >= btrfs_header_nritems(l)) {
785 ret = btrfs_next_leaf(root, path);
792 btrfs_item_key_to_cpu(l, &key, slot);
794 if (key.objectid < device->devid)
796 if (key.objectid > device->devid)
798 if (key.type != BTRFS_DEV_EXTENT_KEY)
800 if (key.offset > search_end)
802 if (key.offset > search_start)
803 free_bytes += key.offset - search_start;
805 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
806 extent_end = key.offset + btrfs_dev_extent_length(l,
808 if (extent_end > search_start)
809 search_start = extent_end;
810 if (search_start > search_end)
817 if (search_start < search_end)
818 free_bytes += search_end - search_start;
820 *avail_bytes = free_bytes;
823 btrfs_free_path(path);
827 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
828 - sizeof(struct btrfs_item) \
829 - sizeof(struct btrfs_chunk)) \
830 / sizeof(struct btrfs_stripe) + 1)
832 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
833 - 2 * sizeof(struct btrfs_disk_key) \
834 - 2 * sizeof(struct btrfs_chunk)) \
835 / sizeof(struct btrfs_stripe) + 1)
837 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
838 struct btrfs_root *extent_root, u64 *start,
839 u64 *num_bytes, u64 type)
842 struct btrfs_fs_info *info = extent_root->fs_info;
843 struct btrfs_root *chunk_root = info->chunk_root;
844 struct btrfs_stripe *stripes;
845 struct btrfs_device *device = NULL;
846 struct btrfs_chunk *chunk;
847 struct list_head private_devs;
848 struct list_head *dev_list = &info->fs_devices->devices;
849 struct list_head *cur;
850 struct map_lookup *map;
851 int min_stripe_size = SZ_1M;
852 u64 calc_size = SZ_8M;
854 u64 max_chunk_size = 4 * calc_size;
865 int stripe_len = BTRFS_STRIPE_LEN;
866 struct btrfs_key key;
869 if (list_empty(dev_list)) {
873 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
874 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
875 BTRFS_BLOCK_GROUP_RAID10 |
876 BTRFS_BLOCK_GROUP_DUP)) {
877 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
879 max_chunk_size = calc_size * 2;
880 min_stripe_size = SZ_1M;
881 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
882 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
884 max_chunk_size = 10 * calc_size;
885 min_stripe_size = SZ_64M;
886 max_stripes = BTRFS_MAX_DEVS(chunk_root);
887 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
889 max_chunk_size = 4 * calc_size;
890 min_stripe_size = SZ_32M;
891 max_stripes = BTRFS_MAX_DEVS(chunk_root);
894 if (type & BTRFS_BLOCK_GROUP_RAID1) {
895 num_stripes = min_t(u64, 2,
896 btrfs_super_num_devices(info->super_copy));
901 if (type & BTRFS_BLOCK_GROUP_DUP) {
905 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
906 num_stripes = btrfs_super_num_devices(info->super_copy);
907 if (num_stripes > max_stripes)
908 num_stripes = max_stripes;
911 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
912 num_stripes = btrfs_super_num_devices(info->super_copy);
913 if (num_stripes > max_stripes)
914 num_stripes = max_stripes;
917 num_stripes &= ~(u32)1;
921 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
922 num_stripes = btrfs_super_num_devices(info->super_copy);
923 if (num_stripes > max_stripes)
924 num_stripes = max_stripes;
928 stripe_len = find_raid56_stripe_len(num_stripes - 1,
929 btrfs_super_stripesize(info->super_copy));
931 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
932 num_stripes = btrfs_super_num_devices(info->super_copy);
933 if (num_stripes > max_stripes)
934 num_stripes = max_stripes;
938 stripe_len = find_raid56_stripe_len(num_stripes - 2,
939 btrfs_super_stripesize(info->super_copy));
942 /* we don't want a chunk larger than 10% of the FS */
943 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
944 max_chunk_size = min(percent_max, max_chunk_size);
947 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
949 calc_size = max_chunk_size;
950 calc_size /= num_stripes;
951 calc_size /= stripe_len;
952 calc_size *= stripe_len;
954 /* we don't want tiny stripes */
955 calc_size = max_t(u64, calc_size, min_stripe_size);
957 calc_size /= stripe_len;
958 calc_size *= stripe_len;
959 INIT_LIST_HEAD(&private_devs);
960 cur = dev_list->next;
963 if (type & BTRFS_BLOCK_GROUP_DUP)
964 min_free = calc_size * 2;
966 min_free = calc_size;
968 /* build a private list of devices we will allocate from */
969 while(index < num_stripes) {
970 device = list_entry(cur, struct btrfs_device, dev_list);
971 ret = btrfs_device_avail_bytes(trans, device, &avail);
975 if (avail >= min_free) {
976 list_move_tail(&device->dev_list, &private_devs);
978 if (type & BTRFS_BLOCK_GROUP_DUP)
980 } else if (avail > max_avail)
985 if (index < num_stripes) {
986 list_splice(&private_devs, dev_list);
987 if (index >= min_stripes) {
989 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
990 num_stripes /= sub_stripes;
991 num_stripes *= sub_stripes;
996 if (!looped && max_avail > 0) {
998 calc_size = max_avail;
1003 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1007 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1008 key.type = BTRFS_CHUNK_ITEM_KEY;
1009 key.offset = offset;
1011 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1015 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1021 stripes = &chunk->stripe;
1022 *num_bytes = chunk_bytes_by_type(type, calc_size,
1023 num_stripes, sub_stripes);
1025 while(index < num_stripes) {
1026 struct btrfs_stripe *stripe;
1027 BUG_ON(list_empty(&private_devs));
1028 cur = private_devs.next;
1029 device = list_entry(cur, struct btrfs_device, dev_list);
1031 /* loop over this device again if we're doing a dup group */
1032 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1033 (index == num_stripes - 1))
1034 list_move_tail(&device->dev_list, dev_list);
1036 ret = btrfs_alloc_dev_extent(trans, device,
1037 info->chunk_root->root_key.objectid,
1038 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1039 calc_size, &dev_offset, 0);
1042 device->bytes_used += calc_size;
1043 ret = btrfs_update_device(trans, device);
1046 map->stripes[index].dev = device;
1047 map->stripes[index].physical = dev_offset;
1048 stripe = stripes + index;
1049 btrfs_set_stack_stripe_devid(stripe, device->devid);
1050 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1051 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1054 BUG_ON(!list_empty(&private_devs));
1056 /* key was set above */
1057 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1058 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1059 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1060 btrfs_set_stack_chunk_type(chunk, type);
1061 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1062 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1063 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1064 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1065 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1066 map->sector_size = info->sectorsize;
1067 map->stripe_len = stripe_len;
1068 map->io_align = stripe_len;
1069 map->io_width = stripe_len;
1071 map->num_stripes = num_stripes;
1072 map->sub_stripes = sub_stripes;
1074 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1075 btrfs_chunk_item_size(num_stripes));
1077 *start = key.offset;;
1079 map->ce.start = key.offset;
1080 map->ce.size = *num_bytes;
1082 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1085 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1086 ret = btrfs_add_system_chunk(chunk_root, &key,
1087 chunk, btrfs_chunk_item_size(num_stripes));
1096 * Alloc a DATA chunk with SINGLE profile.
1098 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1099 * (btrfs logical bytenr == on-disk bytenr)
1100 * For that case, caller must make sure the chunk and dev_extent are not
1103 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1104 struct btrfs_root *extent_root, u64 *start,
1105 u64 num_bytes, u64 type, int convert)
1108 struct btrfs_fs_info *info = extent_root->fs_info;
1109 struct btrfs_root *chunk_root = info->chunk_root;
1110 struct btrfs_stripe *stripes;
1111 struct btrfs_device *device = NULL;
1112 struct btrfs_chunk *chunk;
1113 struct list_head *dev_list = &info->fs_devices->devices;
1114 struct list_head *cur;
1115 struct map_lookup *map;
1116 u64 calc_size = SZ_8M;
1117 int num_stripes = 1;
1118 int sub_stripes = 0;
1121 int stripe_len = BTRFS_STRIPE_LEN;
1122 struct btrfs_key key;
1124 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1125 key.type = BTRFS_CHUNK_ITEM_KEY;
1127 if (*start != round_down(*start, info->sectorsize)) {
1128 error("DATA chunk start not sectorsize aligned: %llu",
1129 (unsigned long long)*start);
1132 key.offset = *start;
1133 dev_offset = *start;
1137 ret = find_next_chunk(chunk_root,
1138 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1145 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1149 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1155 stripes = &chunk->stripe;
1156 calc_size = num_bytes;
1159 cur = dev_list->next;
1160 device = list_entry(cur, struct btrfs_device, dev_list);
1162 while (index < num_stripes) {
1163 struct btrfs_stripe *stripe;
1165 ret = btrfs_alloc_dev_extent(trans, device,
1166 info->chunk_root->root_key.objectid,
1167 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1168 calc_size, &dev_offset, convert);
1171 device->bytes_used += calc_size;
1172 ret = btrfs_update_device(trans, device);
1175 map->stripes[index].dev = device;
1176 map->stripes[index].physical = dev_offset;
1177 stripe = stripes + index;
1178 btrfs_set_stack_stripe_devid(stripe, device->devid);
1179 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1180 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1184 /* key was set above */
1185 btrfs_set_stack_chunk_length(chunk, num_bytes);
1186 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1187 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1188 btrfs_set_stack_chunk_type(chunk, type);
1189 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1190 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1191 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1192 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1193 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1194 map->sector_size = info->sectorsize;
1195 map->stripe_len = stripe_len;
1196 map->io_align = stripe_len;
1197 map->io_width = stripe_len;
1199 map->num_stripes = num_stripes;
1200 map->sub_stripes = sub_stripes;
1202 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1203 btrfs_chunk_item_size(num_stripes));
1206 *start = key.offset;
1208 map->ce.start = key.offset;
1209 map->ce.size = num_bytes;
1211 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1218 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
1220 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1221 struct cache_extent *ce;
1222 struct map_lookup *map;
1225 ce = search_cache_extent(&map_tree->cache_tree, logical);
1227 fprintf(stderr, "No mapping for %llu-%llu\n",
1228 (unsigned long long)logical,
1229 (unsigned long long)logical+len);
1232 if (ce->start > logical || ce->start + ce->size < logical) {
1233 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1234 "%llu-%llu\n", (unsigned long long)logical,
1235 (unsigned long long)logical+len,
1236 (unsigned long long)ce->start,
1237 (unsigned long long)ce->start + ce->size);
1240 map = container_of(ce, struct map_lookup, ce);
1242 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1243 ret = map->num_stripes;
1244 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1245 ret = map->sub_stripes;
1246 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1248 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1255 int btrfs_next_bg(struct btrfs_fs_info *fs_info, u64 *logical,
1256 u64 *size, u64 type)
1258 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1259 struct cache_extent *ce;
1260 struct map_lookup *map;
1263 ce = search_cache_extent(&map_tree->cache_tree, cur);
1267 * only jump to next bg if our cur is not 0
1268 * As the initial logical for btrfs_next_bg() is 0, and
1269 * if we jump to next bg, we skipped a valid bg.
1272 ce = next_cache_extent(ce);
1278 map = container_of(ce, struct map_lookup, ce);
1279 if (map->type & type) {
1280 *logical = ce->start;
1289 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
1290 u64 chunk_start, u64 physical, u64 devid,
1291 u64 **logical, int *naddrs, int *stripe_len)
1293 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1294 struct cache_extent *ce;
1295 struct map_lookup *map;
1303 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1305 map = container_of(ce, struct map_lookup, ce);
1308 rmap_len = map->stripe_len;
1309 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1310 length = ce->size / (map->num_stripes / map->sub_stripes);
1311 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1312 length = ce->size / map->num_stripes;
1313 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1314 BTRFS_BLOCK_GROUP_RAID6)) {
1315 length = ce->size / nr_data_stripes(map);
1316 rmap_len = map->stripe_len * nr_data_stripes(map);
1319 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1321 for (i = 0; i < map->num_stripes; i++) {
1322 if (devid && map->stripes[i].dev->devid != devid)
1324 if (map->stripes[i].physical > physical ||
1325 map->stripes[i].physical + length <= physical)
1328 stripe_nr = (physical - map->stripes[i].physical) /
1331 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1332 stripe_nr = (stripe_nr * map->num_stripes + i) /
1334 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1335 stripe_nr = stripe_nr * map->num_stripes + i;
1336 } /* else if RAID[56], multiply by nr_data_stripes().
1337 * Alternatively, just use rmap_len below instead of
1338 * map->stripe_len */
1340 bytenr = ce->start + stripe_nr * rmap_len;
1341 for (j = 0; j < nr; j++) {
1342 if (buf[j] == bytenr)
1351 *stripe_len = rmap_len;
1356 static inline int parity_smaller(u64 a, u64 b)
1361 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1362 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1364 struct btrfs_bio_stripe s;
1371 for (i = 0; i < bbio->num_stripes - 1; i++) {
1372 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1373 s = bbio->stripes[i];
1375 bbio->stripes[i] = bbio->stripes[i+1];
1376 raid_map[i] = raid_map[i+1];
1377 bbio->stripes[i+1] = s;
1385 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1386 u64 logical, u64 *length,
1387 struct btrfs_multi_bio **multi_ret, int mirror_num,
1390 return __btrfs_map_block(fs_info, rw, logical, length, NULL,
1391 multi_ret, mirror_num, raid_map_ret);
1394 int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1395 u64 logical, u64 *length, u64 *type,
1396 struct btrfs_multi_bio **multi_ret, int mirror_num,
1399 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1400 struct cache_extent *ce;
1401 struct map_lookup *map;
1405 u64 *raid_map = NULL;
1406 int stripes_allocated = 8;
1407 int stripes_required = 1;
1410 struct btrfs_multi_bio *multi = NULL;
1412 if (multi_ret && rw == READ) {
1413 stripes_allocated = 1;
1416 ce = search_cache_extent(&map_tree->cache_tree, logical);
1422 if (ce->start > logical) {
1424 *length = ce->start - logical;
1429 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1434 map = container_of(ce, struct map_lookup, ce);
1435 offset = logical - ce->start;
1438 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1439 BTRFS_BLOCK_GROUP_DUP)) {
1440 stripes_required = map->num_stripes;
1441 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1442 stripes_required = map->sub_stripes;
1445 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1446 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1447 /* RAID[56] write or recovery. Return all stripes */
1448 stripes_required = map->num_stripes;
1450 /* Only allocate the map if we've already got a large enough multi_ret */
1451 if (stripes_allocated >= stripes_required) {
1452 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1460 /* if our multi bio struct is too small, back off and try again */
1461 if (multi_ret && stripes_allocated < stripes_required) {
1462 stripes_allocated = stripes_required;
1469 * stripe_nr counts the total number of stripes we have to stride
1470 * to get to this block
1472 stripe_nr = stripe_nr / map->stripe_len;
1474 stripe_offset = stripe_nr * map->stripe_len;
1475 BUG_ON(offset < stripe_offset);
1477 /* stripe_offset is the offset of this block in its stripe*/
1478 stripe_offset = offset - stripe_offset;
1480 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1481 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1482 BTRFS_BLOCK_GROUP_RAID10 |
1483 BTRFS_BLOCK_GROUP_DUP)) {
1484 /* we limit the length of each bio to what fits in a stripe */
1485 *length = min_t(u64, ce->size - offset,
1486 map->stripe_len - stripe_offset);
1488 *length = ce->size - offset;
1494 multi->num_stripes = 1;
1496 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1498 multi->num_stripes = map->num_stripes;
1499 else if (mirror_num)
1500 stripe_index = mirror_num - 1;
1502 stripe_index = stripe_nr % map->num_stripes;
1503 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1504 int factor = map->num_stripes / map->sub_stripes;
1506 stripe_index = stripe_nr % factor;
1507 stripe_index *= map->sub_stripes;
1510 multi->num_stripes = map->sub_stripes;
1511 else if (mirror_num)
1512 stripe_index += mirror_num - 1;
1514 stripe_nr = stripe_nr / factor;
1515 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1517 multi->num_stripes = map->num_stripes;
1518 else if (mirror_num)
1519 stripe_index = mirror_num - 1;
1520 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1521 BTRFS_BLOCK_GROUP_RAID6)) {
1526 u64 raid56_full_stripe_start;
1527 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1530 * align the start of our data stripe in the logical
1533 raid56_full_stripe_start = offset / full_stripe_len;
1534 raid56_full_stripe_start *= full_stripe_len;
1536 /* get the data stripe number */
1537 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1538 stripe_nr = stripe_nr / nr_data_stripes(map);
1540 /* Work out the disk rotation on this stripe-set */
1541 rot = stripe_nr % map->num_stripes;
1543 /* Fill in the logical address of each stripe */
1544 tmp = stripe_nr * nr_data_stripes(map);
1546 for (i = 0; i < nr_data_stripes(map); i++)
1547 raid_map[(i+rot) % map->num_stripes] =
1548 ce->start + (tmp + i) * map->stripe_len;
1550 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1551 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1552 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1554 *length = map->stripe_len;
1557 multi->num_stripes = map->num_stripes;
1559 stripe_index = stripe_nr % nr_data_stripes(map);
1560 stripe_nr = stripe_nr / nr_data_stripes(map);
1563 * Mirror #0 or #1 means the original data block.
1564 * Mirror #2 is RAID5 parity block.
1565 * Mirror #3 is RAID6 Q block.
1568 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1570 /* We distribute the parity blocks across stripes */
1571 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1575 * after this do_div call, stripe_nr is the number of stripes
1576 * on this device we have to walk to find the data, and
1577 * stripe_index is the number of our device in the stripe array
1579 stripe_index = stripe_nr % map->num_stripes;
1580 stripe_nr = stripe_nr / map->num_stripes;
1582 BUG_ON(stripe_index >= map->num_stripes);
1584 for (i = 0; i < multi->num_stripes; i++) {
1585 multi->stripes[i].physical =
1586 map->stripes[stripe_index].physical + stripe_offset +
1587 stripe_nr * map->stripe_len;
1588 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1597 sort_parity_stripes(multi, raid_map);
1598 *raid_map_ret = raid_map;
1604 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1607 struct btrfs_device *device;
1608 struct btrfs_fs_devices *cur_devices;
1610 cur_devices = root->fs_info->fs_devices;
1611 while (cur_devices) {
1613 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1614 root->fs_info->ignore_fsid_mismatch)) {
1615 device = __find_device(&cur_devices->devices,
1620 cur_devices = cur_devices->seed;
1625 struct btrfs_device *
1626 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1627 u64 devid, int instance)
1629 struct list_head *head = &fs_devices->devices;
1630 struct btrfs_device *dev;
1633 list_for_each_entry(dev, head, dev_list) {
1634 if (dev->devid == devid && num_found++ == instance)
1640 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1642 struct cache_extent *ce;
1643 struct map_lookup *map;
1644 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1649 * During chunk recovering, we may fail to find block group's
1650 * corresponding chunk, we will rebuild it later
1652 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1653 if (!root->fs_info->is_chunk_recover)
1658 map = container_of(ce, struct map_lookup, ce);
1659 for (i = 0; i < map->num_stripes; i++) {
1660 if (!map->stripes[i].dev->writeable) {
1669 static struct btrfs_device *fill_missing_device(u64 devid)
1671 struct btrfs_device *device;
1673 device = kzalloc(sizeof(*device), GFP_NOFS);
1674 device->devid = devid;
1680 * slot == -1: SYSTEM chunk
1681 * return -EIO on error, otherwise return 0
1683 int btrfs_check_chunk_valid(struct btrfs_root *root,
1684 struct extent_buffer *leaf,
1685 struct btrfs_chunk *chunk,
1686 int slot, u64 logical)
1693 u32 chunk_ondisk_size;
1694 u32 sectorsize = root->fs_info->sectorsize;
1696 length = btrfs_chunk_length(leaf, chunk);
1697 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1698 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1699 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1700 type = btrfs_chunk_type(leaf, chunk);
1703 * These valid checks may be insufficient to cover every corner cases.
1705 if (!IS_ALIGNED(logical, sectorsize)) {
1706 error("invalid chunk logical %llu", logical);
1709 if (btrfs_chunk_sector_size(leaf, chunk) != sectorsize) {
1710 error("invalid chunk sectorsize %llu",
1711 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1714 if (!length || !IS_ALIGNED(length, sectorsize)) {
1715 error("invalid chunk length %llu", length);
1718 if (stripe_len != BTRFS_STRIPE_LEN) {
1719 error("invalid chunk stripe length: %llu", stripe_len);
1722 /* Check on chunk item type */
1723 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1724 error("invalid chunk type %llu", type);
1727 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1728 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1729 error("unrecognized chunk type: %llu",
1730 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1731 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1734 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1735 error("missing chunk type flag: %llu", type);
1738 if (!(is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) ||
1739 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0)) {
1740 error("conflicting chunk type detected: %llu", type);
1743 if ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
1744 !is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1745 error("conflicting chunk profile detected: %llu", type);
1749 chunk_ondisk_size = btrfs_chunk_item_size(num_stripes);
1751 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1752 * it can't exceed the system chunk array size
1753 * For normal chunk, it should match its chunk item size.
1755 if (num_stripes < 1 ||
1756 (slot == -1 && chunk_ondisk_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1757 (slot >= 0 && chunk_ondisk_size > btrfs_item_size_nr(leaf, slot))) {
1758 error("invalid num_stripes: %u", num_stripes);
1762 * Device number check against profile
1764 if ((type & BTRFS_BLOCK_GROUP_RAID10 && (sub_stripes != 2 ||
1765 !IS_ALIGNED(num_stripes, sub_stripes))) ||
1766 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1767 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1768 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1769 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1770 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1771 num_stripes != 1)) {
1772 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1773 num_stripes, sub_stripes,
1774 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1782 * Slot is used to verify the chunk item is valid
1784 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1786 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1787 struct extent_buffer *leaf,
1788 struct btrfs_chunk *chunk, int slot)
1790 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1791 struct map_lookup *map;
1792 struct cache_extent *ce;
1796 u8 uuid[BTRFS_UUID_SIZE];
1801 logical = key->offset;
1802 length = btrfs_chunk_length(leaf, chunk);
1803 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1804 /* Validation check */
1805 ret = btrfs_check_chunk_valid(root, leaf, chunk, slot, logical);
1807 error("%s checksums match, but it has an invalid chunk, %s",
1808 (slot == -1) ? "Superblock" : "Metadata",
1809 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1813 ce = search_cache_extent(&map_tree->cache_tree, logical);
1815 /* already mapped? */
1816 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1820 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1824 map->ce.start = logical;
1825 map->ce.size = length;
1826 map->num_stripes = num_stripes;
1827 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1828 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1829 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1830 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1831 map->type = btrfs_chunk_type(leaf, chunk);
1832 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1834 for (i = 0; i < num_stripes; i++) {
1835 map->stripes[i].physical =
1836 btrfs_stripe_offset_nr(leaf, chunk, i);
1837 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1838 read_extent_buffer(leaf, uuid, (unsigned long)
1839 btrfs_stripe_dev_uuid_nr(chunk, i),
1841 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1843 if (!map->stripes[i].dev) {
1844 map->stripes[i].dev = fill_missing_device(devid);
1845 printf("warning, device %llu is missing\n",
1846 (unsigned long long)devid);
1847 list_add(&map->stripes[i].dev->dev_list,
1848 &root->fs_info->fs_devices->devices);
1852 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1858 static int fill_device_from_item(struct extent_buffer *leaf,
1859 struct btrfs_dev_item *dev_item,
1860 struct btrfs_device *device)
1864 device->devid = btrfs_device_id(leaf, dev_item);
1865 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1866 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1867 device->type = btrfs_device_type(leaf, dev_item);
1868 device->io_align = btrfs_device_io_align(leaf, dev_item);
1869 device->io_width = btrfs_device_io_width(leaf, dev_item);
1870 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1872 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1873 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1878 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1880 struct btrfs_fs_devices *fs_devices;
1883 fs_devices = root->fs_info->fs_devices->seed;
1884 while (fs_devices) {
1885 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1889 fs_devices = fs_devices->seed;
1892 fs_devices = find_fsid(fsid);
1894 /* missing all seed devices */
1895 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1900 INIT_LIST_HEAD(&fs_devices->devices);
1901 list_add(&fs_devices->list, &fs_uuids);
1902 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1905 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1909 fs_devices->seed = root->fs_info->fs_devices->seed;
1910 root->fs_info->fs_devices->seed = fs_devices;
1915 static int read_one_dev(struct btrfs_root *root,
1916 struct extent_buffer *leaf,
1917 struct btrfs_dev_item *dev_item)
1919 struct btrfs_device *device;
1922 u8 fs_uuid[BTRFS_UUID_SIZE];
1923 u8 dev_uuid[BTRFS_UUID_SIZE];
1925 devid = btrfs_device_id(leaf, dev_item);
1926 read_extent_buffer(leaf, dev_uuid,
1927 (unsigned long)btrfs_device_uuid(dev_item),
1929 read_extent_buffer(leaf, fs_uuid,
1930 (unsigned long)btrfs_device_fsid(dev_item),
1933 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1934 ret = open_seed_devices(root, fs_uuid);
1939 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1941 device = kzalloc(sizeof(*device), GFP_NOFS);
1945 list_add(&device->dev_list,
1946 &root->fs_info->fs_devices->devices);
1949 fill_device_from_item(leaf, dev_item, device);
1950 device->dev_root = root->fs_info->dev_root;
1954 int btrfs_read_sys_array(struct btrfs_root *root)
1956 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1957 struct extent_buffer *sb;
1958 struct btrfs_disk_key *disk_key;
1959 struct btrfs_chunk *chunk;
1961 unsigned long sb_array_offset;
1967 struct btrfs_key key;
1969 sb = btrfs_find_create_tree_block(root->fs_info,
1970 BTRFS_SUPER_INFO_OFFSET,
1971 BTRFS_SUPER_INFO_SIZE);
1974 btrfs_set_buffer_uptodate(sb);
1975 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1976 array_size = btrfs_super_sys_array_size(super_copy);
1978 array_ptr = super_copy->sys_chunk_array;
1979 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1982 while (cur_offset < array_size) {
1983 disk_key = (struct btrfs_disk_key *)array_ptr;
1984 len = sizeof(*disk_key);
1985 if (cur_offset + len > array_size)
1986 goto out_short_read;
1988 btrfs_disk_key_to_cpu(&key, disk_key);
1991 sb_array_offset += len;
1994 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1995 chunk = (struct btrfs_chunk *)sb_array_offset;
1997 * At least one btrfs_chunk with one stripe must be
1998 * present, exact stripe count check comes afterwards
2000 len = btrfs_chunk_item_size(1);
2001 if (cur_offset + len > array_size)
2002 goto out_short_read;
2004 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2007 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
2008 num_stripes, cur_offset);
2013 len = btrfs_chunk_item_size(num_stripes);
2014 if (cur_offset + len > array_size)
2015 goto out_short_read;
2017 ret = read_one_chunk(root, &key, sb, chunk, -1);
2022 "ERROR: unexpected item type %u in sys_array at offset %u\n",
2023 (u32)key.type, cur_offset);
2028 sb_array_offset += len;
2031 free_extent_buffer(sb);
2035 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
2037 free_extent_buffer(sb);
2041 int btrfs_read_chunk_tree(struct btrfs_root *root)
2043 struct btrfs_path *path;
2044 struct extent_buffer *leaf;
2045 struct btrfs_key key;
2046 struct btrfs_key found_key;
2050 root = root->fs_info->chunk_root;
2052 path = btrfs_alloc_path();
2057 * Read all device items, and then all the chunk items. All
2058 * device items are found before any chunk item (their object id
2059 * is smaller than the lowest possible object id for a chunk
2060 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
2062 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2065 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2069 leaf = path->nodes[0];
2070 slot = path->slots[0];
2071 if (slot >= btrfs_header_nritems(leaf)) {
2072 ret = btrfs_next_leaf(root, path);
2079 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2080 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2081 struct btrfs_dev_item *dev_item;
2082 dev_item = btrfs_item_ptr(leaf, slot,
2083 struct btrfs_dev_item);
2084 ret = read_one_dev(root, leaf, dev_item);
2086 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2087 struct btrfs_chunk *chunk;
2088 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2089 ret = read_one_chunk(root, &found_key, leaf, chunk,
2098 btrfs_free_path(path);
2102 struct list_head *btrfs_scanned_uuids(void)
2107 static int rmw_eb(struct btrfs_fs_info *info,
2108 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2111 unsigned long orig_off = 0;
2112 unsigned long dest_off = 0;
2113 unsigned long copy_len = eb->len;
2115 ret = read_whole_eb(info, eb, 0);
2119 if (eb->start + eb->len <= orig_eb->start ||
2120 eb->start >= orig_eb->start + orig_eb->len)
2123 * | ----- orig_eb ------- |
2124 * | ----- stripe ------- |
2125 * | ----- orig_eb ------- |
2126 * | ----- orig_eb ------- |
2128 if (eb->start > orig_eb->start)
2129 orig_off = eb->start - orig_eb->start;
2130 if (orig_eb->start > eb->start)
2131 dest_off = orig_eb->start - eb->start;
2133 if (copy_len > orig_eb->len - orig_off)
2134 copy_len = orig_eb->len - orig_off;
2135 if (copy_len > eb->len - dest_off)
2136 copy_len = eb->len - dest_off;
2138 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2142 static int split_eb_for_raid56(struct btrfs_fs_info *info,
2143 struct extent_buffer *orig_eb,
2144 struct extent_buffer **ebs,
2145 u64 stripe_len, u64 *raid_map,
2148 struct extent_buffer **tmp_ebs;
2149 u64 start = orig_eb->start;
2154 tmp_ebs = calloc(num_stripes, sizeof(*tmp_ebs));
2158 /* Alloc memory in a row for data stripes */
2159 for (i = 0; i < num_stripes; i++) {
2160 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2163 tmp_ebs[i] = calloc(1, sizeof(**tmp_ebs) + stripe_len);
2170 for (i = 0; i < num_stripes; i++) {
2171 struct extent_buffer *eb = tmp_ebs[i];
2173 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2176 eb->start = raid_map[i];
2177 eb->len = stripe_len;
2181 eb->dev_bytenr = (u64)-1;
2183 this_eb_start = raid_map[i];
2185 if (start > this_eb_start ||
2186 start + orig_eb->len < this_eb_start + stripe_len) {
2187 ret = rmw_eb(info, eb, orig_eb);
2191 memcpy(eb->data, orig_eb->data + eb->start - start,
2199 for (i = 0; i < num_stripes; i++)
2205 int write_raid56_with_parity(struct btrfs_fs_info *info,
2206 struct extent_buffer *eb,
2207 struct btrfs_multi_bio *multi,
2208 u64 stripe_len, u64 *raid_map)
2210 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2213 int alloc_size = eb->len;
2216 ebs = malloc(sizeof(*ebs) * multi->num_stripes);
2217 pointers = malloc(sizeof(*pointers) * multi->num_stripes);
2218 if (!ebs || !pointers) {
2224 if (stripe_len > alloc_size)
2225 alloc_size = stripe_len;
2227 ret = split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2228 multi->num_stripes);
2232 for (i = 0; i < multi->num_stripes; i++) {
2233 struct extent_buffer *new_eb;
2234 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2235 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2236 ebs[i]->fd = multi->stripes[i].dev->fd;
2237 multi->stripes[i].dev->total_ios++;
2238 if (ebs[i]->start != raid_map[i]) {
2240 goto out_free_split;
2244 new_eb = malloc(sizeof(*eb) + alloc_size);
2247 goto out_free_split;
2249 new_eb->dev_bytenr = multi->stripes[i].physical;
2250 new_eb->fd = multi->stripes[i].dev->fd;
2251 multi->stripes[i].dev->total_ios++;
2252 new_eb->len = stripe_len;
2254 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2256 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2260 ebs[multi->num_stripes - 2] = p_eb;
2261 ebs[multi->num_stripes - 1] = q_eb;
2263 for (i = 0; i < multi->num_stripes; i++)
2264 pointers[i] = ebs[i]->data;
2266 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2268 ebs[multi->num_stripes - 1] = p_eb;
2269 for (i = 0; i < multi->num_stripes; i++)
2270 pointers[i] = ebs[i]->data;
2271 ret = raid5_gen_result(multi->num_stripes, stripe_len,
2272 multi->num_stripes - 1, pointers);
2274 goto out_free_split;
2277 for (i = 0; i < multi->num_stripes; i++) {
2278 ret = write_extent_to_disk(ebs[i]);
2280 goto out_free_split;
2284 for (i = 0; i < multi->num_stripes; i++) {
2296 * Get stripe length from chunk item and its stripe items
2298 * Caller should only call this function after validating the chunk item
2299 * by using btrfs_check_chunk_valid().
2301 u64 btrfs_stripe_length(struct btrfs_fs_info *fs_info,
2302 struct extent_buffer *leaf,
2303 struct btrfs_chunk *chunk)
2307 u32 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2308 u64 profile = btrfs_chunk_type(leaf, chunk) &
2309 BTRFS_BLOCK_GROUP_PROFILE_MASK;
2311 chunk_len = btrfs_chunk_length(leaf, chunk);
2314 case 0: /* Single profile */
2315 case BTRFS_BLOCK_GROUP_RAID1:
2316 case BTRFS_BLOCK_GROUP_DUP:
2317 stripe_len = chunk_len;
2319 case BTRFS_BLOCK_GROUP_RAID0:
2320 stripe_len = chunk_len / num_stripes;
2322 case BTRFS_BLOCK_GROUP_RAID5:
2323 stripe_len = chunk_len / (num_stripes - 1);
2325 case BTRFS_BLOCK_GROUP_RAID6:
2326 stripe_len = chunk_len / (num_stripes - 2);
2328 case BTRFS_BLOCK_GROUP_RAID10:
2329 stripe_len = chunk_len / (num_stripes /
2330 btrfs_chunk_sub_stripes(leaf, chunk));
2333 /* Invalid chunk profile found */