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_fs_info *fs_info,
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;
596 struct btrfs_root *root = fs_info->chunk_root;
600 path = btrfs_alloc_path();
604 ret = find_next_devid(root, path, &free_devid);
608 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
609 key.type = BTRFS_DEV_ITEM_KEY;
610 key.offset = free_devid;
612 ret = btrfs_insert_empty_item(trans, root, path, &key,
617 leaf = path->nodes[0];
618 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
620 device->devid = free_devid;
621 btrfs_set_device_id(leaf, dev_item, device->devid);
622 btrfs_set_device_generation(leaf, dev_item, 0);
623 btrfs_set_device_type(leaf, dev_item, device->type);
624 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
625 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
626 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
627 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
628 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
629 btrfs_set_device_group(leaf, dev_item, 0);
630 btrfs_set_device_seek_speed(leaf, dev_item, 0);
631 btrfs_set_device_bandwidth(leaf, dev_item, 0);
632 btrfs_set_device_start_offset(leaf, dev_item, 0);
634 ptr = (unsigned long)btrfs_device_uuid(dev_item);
635 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
636 ptr = (unsigned long)btrfs_device_fsid(dev_item);
637 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
638 btrfs_mark_buffer_dirty(leaf);
642 btrfs_free_path(path);
646 int btrfs_update_device(struct btrfs_trans_handle *trans,
647 struct btrfs_device *device)
650 struct btrfs_path *path;
651 struct btrfs_root *root;
652 struct btrfs_dev_item *dev_item;
653 struct extent_buffer *leaf;
654 struct btrfs_key key;
656 root = device->dev_root->fs_info->chunk_root;
658 path = btrfs_alloc_path();
662 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
663 key.type = BTRFS_DEV_ITEM_KEY;
664 key.offset = device->devid;
666 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
675 leaf = path->nodes[0];
676 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
678 btrfs_set_device_id(leaf, dev_item, device->devid);
679 btrfs_set_device_type(leaf, dev_item, device->type);
680 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
681 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
682 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
683 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
684 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
685 btrfs_mark_buffer_dirty(leaf);
688 btrfs_free_path(path);
692 int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
693 struct btrfs_chunk *chunk, int item_size)
695 struct btrfs_super_block *super_copy = fs_info->super_copy;
696 struct btrfs_disk_key disk_key;
700 array_size = btrfs_super_sys_array_size(super_copy);
701 if (array_size + item_size + sizeof(disk_key)
702 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
705 ptr = super_copy->sys_chunk_array + array_size;
706 btrfs_cpu_key_to_disk(&disk_key, key);
707 memcpy(ptr, &disk_key, sizeof(disk_key));
708 ptr += sizeof(disk_key);
709 memcpy(ptr, chunk, item_size);
710 item_size += sizeof(disk_key);
711 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
715 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
718 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
720 else if (type & BTRFS_BLOCK_GROUP_RAID10)
721 return calc_size * (num_stripes / sub_stripes);
722 else if (type & BTRFS_BLOCK_GROUP_RAID5)
723 return calc_size * (num_stripes - 1);
724 else if (type & BTRFS_BLOCK_GROUP_RAID6)
725 return calc_size * (num_stripes - 2);
727 return calc_size * num_stripes;
731 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
733 /* TODO, add a way to store the preferred stripe size */
734 return BTRFS_STRIPE_LEN;
738 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
740 * It is not equal to "device->total_bytes - device->bytes_used".
741 * We do not allocate any chunk in 1M at beginning of device, and not
742 * allowed to allocate any chunk before alloc_start if it is specified.
743 * So search holes from max(1M, alloc_start) to device->total_bytes.
745 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
746 struct btrfs_device *device,
749 struct btrfs_path *path;
750 struct btrfs_root *root = device->dev_root;
751 struct btrfs_key key;
752 struct btrfs_dev_extent *dev_extent = NULL;
753 struct extent_buffer *l;
754 u64 search_start = root->fs_info->alloc_start;
755 u64 search_end = device->total_bytes;
761 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
763 path = btrfs_alloc_path();
767 key.objectid = device->devid;
768 key.offset = root->fs_info->alloc_start;
769 key.type = BTRFS_DEV_EXTENT_KEY;
772 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
775 ret = btrfs_previous_item(root, path, 0, key.type);
781 slot = path->slots[0];
782 if (slot >= btrfs_header_nritems(l)) {
783 ret = btrfs_next_leaf(root, path);
790 btrfs_item_key_to_cpu(l, &key, slot);
792 if (key.objectid < device->devid)
794 if (key.objectid > device->devid)
796 if (key.type != BTRFS_DEV_EXTENT_KEY)
798 if (key.offset > search_end)
800 if (key.offset > search_start)
801 free_bytes += key.offset - search_start;
803 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
804 extent_end = key.offset + btrfs_dev_extent_length(l,
806 if (extent_end > search_start)
807 search_start = extent_end;
808 if (search_start > search_end)
815 if (search_start < search_end)
816 free_bytes += search_end - search_start;
818 *avail_bytes = free_bytes;
821 btrfs_free_path(path);
825 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
826 - sizeof(struct btrfs_item) \
827 - sizeof(struct btrfs_chunk)) \
828 / sizeof(struct btrfs_stripe) + 1)
830 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
831 - 2 * sizeof(struct btrfs_disk_key) \
832 - 2 * sizeof(struct btrfs_chunk)) \
833 / sizeof(struct btrfs_stripe) + 1)
835 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
836 struct btrfs_fs_info *info, u64 *start,
837 u64 *num_bytes, u64 type)
840 struct btrfs_root *extent_root = info->extent_root;
841 struct btrfs_root *chunk_root = info->chunk_root;
842 struct btrfs_stripe *stripes;
843 struct btrfs_device *device = NULL;
844 struct btrfs_chunk *chunk;
845 struct list_head private_devs;
846 struct list_head *dev_list = &info->fs_devices->devices;
847 struct list_head *cur;
848 struct map_lookup *map;
849 int min_stripe_size = SZ_1M;
850 u64 calc_size = SZ_8M;
852 u64 max_chunk_size = 4 * calc_size;
863 int stripe_len = BTRFS_STRIPE_LEN;
864 struct btrfs_key key;
867 if (list_empty(dev_list)) {
871 if (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
872 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
874 max_chunk_size = calc_size * 2;
875 min_stripe_size = SZ_1M;
876 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
877 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
879 max_chunk_size = 10 * calc_size;
880 min_stripe_size = SZ_64M;
881 max_stripes = BTRFS_MAX_DEVS(chunk_root);
882 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
884 max_chunk_size = 4 * calc_size;
885 min_stripe_size = SZ_32M;
886 max_stripes = BTRFS_MAX_DEVS(chunk_root);
889 if (type & BTRFS_BLOCK_GROUP_RAID1) {
890 num_stripes = min_t(u64, 2,
891 btrfs_super_num_devices(info->super_copy));
896 if (type & BTRFS_BLOCK_GROUP_DUP) {
900 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
901 num_stripes = btrfs_super_num_devices(info->super_copy);
902 if (num_stripes > max_stripes)
903 num_stripes = max_stripes;
906 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
907 num_stripes = btrfs_super_num_devices(info->super_copy);
908 if (num_stripes > max_stripes)
909 num_stripes = max_stripes;
912 num_stripes &= ~(u32)1;
916 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
917 num_stripes = btrfs_super_num_devices(info->super_copy);
918 if (num_stripes > max_stripes)
919 num_stripes = max_stripes;
923 stripe_len = find_raid56_stripe_len(num_stripes - 1,
924 btrfs_super_stripesize(info->super_copy));
926 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
927 num_stripes = btrfs_super_num_devices(info->super_copy);
928 if (num_stripes > max_stripes)
929 num_stripes = max_stripes;
933 stripe_len = find_raid56_stripe_len(num_stripes - 2,
934 btrfs_super_stripesize(info->super_copy));
937 /* we don't want a chunk larger than 10% of the FS */
938 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
939 max_chunk_size = min(percent_max, max_chunk_size);
942 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
944 calc_size = max_chunk_size;
945 calc_size /= num_stripes;
946 calc_size /= stripe_len;
947 calc_size *= stripe_len;
949 /* we don't want tiny stripes */
950 calc_size = max_t(u64, calc_size, min_stripe_size);
952 calc_size /= stripe_len;
953 calc_size *= stripe_len;
954 INIT_LIST_HEAD(&private_devs);
955 cur = dev_list->next;
958 if (type & BTRFS_BLOCK_GROUP_DUP)
959 min_free = calc_size * 2;
961 min_free = calc_size;
963 /* build a private list of devices we will allocate from */
964 while(index < num_stripes) {
965 device = list_entry(cur, struct btrfs_device, dev_list);
966 ret = btrfs_device_avail_bytes(trans, device, &avail);
970 if (avail >= min_free) {
971 list_move_tail(&device->dev_list, &private_devs);
973 if (type & BTRFS_BLOCK_GROUP_DUP)
975 } else if (avail > max_avail)
980 if (index < num_stripes) {
981 list_splice(&private_devs, dev_list);
982 if (index >= min_stripes) {
984 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
985 num_stripes /= sub_stripes;
986 num_stripes *= sub_stripes;
991 if (!looped && max_avail > 0) {
993 calc_size = max_avail;
998 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1002 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1003 key.type = BTRFS_CHUNK_ITEM_KEY;
1004 key.offset = offset;
1006 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1010 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1016 stripes = &chunk->stripe;
1017 *num_bytes = chunk_bytes_by_type(type, calc_size,
1018 num_stripes, sub_stripes);
1020 while(index < num_stripes) {
1021 struct btrfs_stripe *stripe;
1022 BUG_ON(list_empty(&private_devs));
1023 cur = private_devs.next;
1024 device = list_entry(cur, struct btrfs_device, dev_list);
1026 /* loop over this device again if we're doing a dup group */
1027 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1028 (index == num_stripes - 1))
1029 list_move_tail(&device->dev_list, dev_list);
1031 ret = btrfs_alloc_dev_extent(trans, device,
1032 info->chunk_root->root_key.objectid,
1033 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1034 calc_size, &dev_offset, 0);
1037 device->bytes_used += calc_size;
1038 ret = btrfs_update_device(trans, device);
1041 map->stripes[index].dev = device;
1042 map->stripes[index].physical = dev_offset;
1043 stripe = stripes + index;
1044 btrfs_set_stack_stripe_devid(stripe, device->devid);
1045 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1046 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1049 BUG_ON(!list_empty(&private_devs));
1051 /* key was set above */
1052 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1053 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1054 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1055 btrfs_set_stack_chunk_type(chunk, type);
1056 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1057 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1058 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1059 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1060 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1061 map->sector_size = info->sectorsize;
1062 map->stripe_len = stripe_len;
1063 map->io_align = stripe_len;
1064 map->io_width = stripe_len;
1066 map->num_stripes = num_stripes;
1067 map->sub_stripes = sub_stripes;
1069 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1070 btrfs_chunk_item_size(num_stripes));
1072 *start = key.offset;;
1074 map->ce.start = key.offset;
1075 map->ce.size = *num_bytes;
1077 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1080 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1081 ret = btrfs_add_system_chunk(info, &key,
1082 chunk, btrfs_chunk_item_size(num_stripes));
1091 * Alloc a DATA chunk with SINGLE profile.
1093 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1094 * (btrfs logical bytenr == on-disk bytenr)
1095 * For that case, caller must make sure the chunk and dev_extent are not
1098 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1099 struct btrfs_fs_info *info, u64 *start,
1100 u64 num_bytes, u64 type, int convert)
1103 struct btrfs_root *extent_root = info->extent_root;
1104 struct btrfs_root *chunk_root = info->chunk_root;
1105 struct btrfs_stripe *stripes;
1106 struct btrfs_device *device = NULL;
1107 struct btrfs_chunk *chunk;
1108 struct list_head *dev_list = &info->fs_devices->devices;
1109 struct list_head *cur;
1110 struct map_lookup *map;
1111 u64 calc_size = SZ_8M;
1112 int num_stripes = 1;
1113 int sub_stripes = 0;
1116 int stripe_len = BTRFS_STRIPE_LEN;
1117 struct btrfs_key key;
1119 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1120 key.type = BTRFS_CHUNK_ITEM_KEY;
1122 if (*start != round_down(*start, info->sectorsize)) {
1123 error("DATA chunk start not sectorsize aligned: %llu",
1124 (unsigned long long)*start);
1127 key.offset = *start;
1128 dev_offset = *start;
1132 ret = find_next_chunk(chunk_root,
1133 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1140 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1144 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1150 stripes = &chunk->stripe;
1151 calc_size = num_bytes;
1154 cur = dev_list->next;
1155 device = list_entry(cur, struct btrfs_device, dev_list);
1157 while (index < num_stripes) {
1158 struct btrfs_stripe *stripe;
1160 ret = btrfs_alloc_dev_extent(trans, device,
1161 info->chunk_root->root_key.objectid,
1162 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1163 calc_size, &dev_offset, convert);
1166 device->bytes_used += calc_size;
1167 ret = btrfs_update_device(trans, device);
1170 map->stripes[index].dev = device;
1171 map->stripes[index].physical = dev_offset;
1172 stripe = stripes + index;
1173 btrfs_set_stack_stripe_devid(stripe, device->devid);
1174 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1175 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1179 /* key was set above */
1180 btrfs_set_stack_chunk_length(chunk, num_bytes);
1181 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1182 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1183 btrfs_set_stack_chunk_type(chunk, type);
1184 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1185 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1186 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1187 btrfs_set_stack_chunk_sector_size(chunk, info->sectorsize);
1188 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1189 map->sector_size = info->sectorsize;
1190 map->stripe_len = stripe_len;
1191 map->io_align = stripe_len;
1192 map->io_width = stripe_len;
1194 map->num_stripes = num_stripes;
1195 map->sub_stripes = sub_stripes;
1197 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1198 btrfs_chunk_item_size(num_stripes));
1201 *start = key.offset;
1203 map->ce.start = key.offset;
1204 map->ce.size = num_bytes;
1206 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1213 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
1215 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1216 struct cache_extent *ce;
1217 struct map_lookup *map;
1220 ce = search_cache_extent(&map_tree->cache_tree, logical);
1222 fprintf(stderr, "No mapping for %llu-%llu\n",
1223 (unsigned long long)logical,
1224 (unsigned long long)logical+len);
1227 if (ce->start > logical || ce->start + ce->size < logical) {
1228 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1229 "%llu-%llu\n", (unsigned long long)logical,
1230 (unsigned long long)logical+len,
1231 (unsigned long long)ce->start,
1232 (unsigned long long)ce->start + ce->size);
1235 map = container_of(ce, struct map_lookup, ce);
1237 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1238 ret = map->num_stripes;
1239 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1240 ret = map->sub_stripes;
1241 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1243 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1250 int btrfs_next_bg(struct btrfs_fs_info *fs_info, u64 *logical,
1251 u64 *size, u64 type)
1253 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1254 struct cache_extent *ce;
1255 struct map_lookup *map;
1258 ce = search_cache_extent(&map_tree->cache_tree, cur);
1262 * only jump to next bg if our cur is not 0
1263 * As the initial logical for btrfs_next_bg() is 0, and
1264 * if we jump to next bg, we skipped a valid bg.
1267 ce = next_cache_extent(ce);
1273 map = container_of(ce, struct map_lookup, ce);
1274 if (map->type & type) {
1275 *logical = ce->start;
1280 ce = next_cache_extent(ce);
1286 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
1287 u64 chunk_start, u64 physical, u64 devid,
1288 u64 **logical, int *naddrs, int *stripe_len)
1290 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1291 struct cache_extent *ce;
1292 struct map_lookup *map;
1300 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1302 map = container_of(ce, struct map_lookup, ce);
1305 rmap_len = map->stripe_len;
1306 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1307 length = ce->size / (map->num_stripes / map->sub_stripes);
1308 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1309 length = ce->size / map->num_stripes;
1310 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1311 BTRFS_BLOCK_GROUP_RAID6)) {
1312 length = ce->size / nr_data_stripes(map);
1313 rmap_len = map->stripe_len * nr_data_stripes(map);
1316 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1318 for (i = 0; i < map->num_stripes; i++) {
1319 if (devid && map->stripes[i].dev->devid != devid)
1321 if (map->stripes[i].physical > physical ||
1322 map->stripes[i].physical + length <= physical)
1325 stripe_nr = (physical - map->stripes[i].physical) /
1328 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1329 stripe_nr = (stripe_nr * map->num_stripes + i) /
1331 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1332 stripe_nr = stripe_nr * map->num_stripes + i;
1333 } /* else if RAID[56], multiply by nr_data_stripes().
1334 * Alternatively, just use rmap_len below instead of
1335 * map->stripe_len */
1337 bytenr = ce->start + stripe_nr * rmap_len;
1338 for (j = 0; j < nr; j++) {
1339 if (buf[j] == bytenr)
1348 *stripe_len = rmap_len;
1353 static inline int parity_smaller(u64 a, u64 b)
1358 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1359 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1361 struct btrfs_bio_stripe s;
1368 for (i = 0; i < bbio->num_stripes - 1; i++) {
1369 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1370 s = bbio->stripes[i];
1372 bbio->stripes[i] = bbio->stripes[i+1];
1373 raid_map[i] = raid_map[i+1];
1374 bbio->stripes[i+1] = s;
1382 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1383 u64 logical, u64 *length,
1384 struct btrfs_multi_bio **multi_ret, int mirror_num,
1387 return __btrfs_map_block(fs_info, rw, logical, length, NULL,
1388 multi_ret, mirror_num, raid_map_ret);
1391 int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
1392 u64 logical, u64 *length, u64 *type,
1393 struct btrfs_multi_bio **multi_ret, int mirror_num,
1396 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1397 struct cache_extent *ce;
1398 struct map_lookup *map;
1402 u64 *raid_map = NULL;
1403 int stripes_allocated = 8;
1404 int stripes_required = 1;
1407 struct btrfs_multi_bio *multi = NULL;
1409 if (multi_ret && rw == READ) {
1410 stripes_allocated = 1;
1413 ce = search_cache_extent(&map_tree->cache_tree, logical);
1419 if (ce->start > logical) {
1421 *length = ce->start - logical;
1426 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1431 map = container_of(ce, struct map_lookup, ce);
1432 offset = logical - ce->start;
1435 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1436 BTRFS_BLOCK_GROUP_DUP)) {
1437 stripes_required = map->num_stripes;
1438 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1439 stripes_required = map->sub_stripes;
1442 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1443 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1444 /* RAID[56] write or recovery. Return all stripes */
1445 stripes_required = map->num_stripes;
1447 /* Only allocate the map if we've already got a large enough multi_ret */
1448 if (stripes_allocated >= stripes_required) {
1449 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1457 /* if our multi bio struct is too small, back off and try again */
1458 if (multi_ret && stripes_allocated < stripes_required) {
1459 stripes_allocated = stripes_required;
1466 * stripe_nr counts the total number of stripes we have to stride
1467 * to get to this block
1469 stripe_nr = stripe_nr / map->stripe_len;
1471 stripe_offset = stripe_nr * map->stripe_len;
1472 BUG_ON(offset < stripe_offset);
1474 /* stripe_offset is the offset of this block in its stripe*/
1475 stripe_offset = offset - stripe_offset;
1477 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1478 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1479 BTRFS_BLOCK_GROUP_RAID10 |
1480 BTRFS_BLOCK_GROUP_DUP)) {
1481 /* we limit the length of each bio to what fits in a stripe */
1482 *length = min_t(u64, ce->size - offset,
1483 map->stripe_len - stripe_offset);
1485 *length = ce->size - offset;
1491 multi->num_stripes = 1;
1493 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1495 multi->num_stripes = map->num_stripes;
1496 else if (mirror_num)
1497 stripe_index = mirror_num - 1;
1499 stripe_index = stripe_nr % map->num_stripes;
1500 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1501 int factor = map->num_stripes / map->sub_stripes;
1503 stripe_index = stripe_nr % factor;
1504 stripe_index *= map->sub_stripes;
1507 multi->num_stripes = map->sub_stripes;
1508 else if (mirror_num)
1509 stripe_index += mirror_num - 1;
1511 stripe_nr = stripe_nr / factor;
1512 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1514 multi->num_stripes = map->num_stripes;
1515 else if (mirror_num)
1516 stripe_index = mirror_num - 1;
1517 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1518 BTRFS_BLOCK_GROUP_RAID6)) {
1523 u64 raid56_full_stripe_start;
1524 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1527 * align the start of our data stripe in the logical
1530 raid56_full_stripe_start = offset / full_stripe_len;
1531 raid56_full_stripe_start *= full_stripe_len;
1533 /* get the data stripe number */
1534 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1535 stripe_nr = stripe_nr / nr_data_stripes(map);
1537 /* Work out the disk rotation on this stripe-set */
1538 rot = stripe_nr % map->num_stripes;
1540 /* Fill in the logical address of each stripe */
1541 tmp = stripe_nr * nr_data_stripes(map);
1543 for (i = 0; i < nr_data_stripes(map); i++)
1544 raid_map[(i+rot) % map->num_stripes] =
1545 ce->start + (tmp + i) * map->stripe_len;
1547 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1548 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1549 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1551 *length = map->stripe_len;
1554 multi->num_stripes = map->num_stripes;
1556 stripe_index = stripe_nr % nr_data_stripes(map);
1557 stripe_nr = stripe_nr / nr_data_stripes(map);
1560 * Mirror #0 or #1 means the original data block.
1561 * Mirror #2 is RAID5 parity block.
1562 * Mirror #3 is RAID6 Q block.
1565 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1567 /* We distribute the parity blocks across stripes */
1568 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1572 * after this do_div call, stripe_nr is the number of stripes
1573 * on this device we have to walk to find the data, and
1574 * stripe_index is the number of our device in the stripe array
1576 stripe_index = stripe_nr % map->num_stripes;
1577 stripe_nr = stripe_nr / map->num_stripes;
1579 BUG_ON(stripe_index >= map->num_stripes);
1581 for (i = 0; i < multi->num_stripes; i++) {
1582 multi->stripes[i].physical =
1583 map->stripes[stripe_index].physical + stripe_offset +
1584 stripe_nr * map->stripe_len;
1585 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1594 sort_parity_stripes(multi, raid_map);
1595 *raid_map_ret = raid_map;
1601 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
1604 struct btrfs_device *device;
1605 struct btrfs_fs_devices *cur_devices;
1607 cur_devices = fs_info->fs_devices;
1608 while (cur_devices) {
1610 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1611 fs_info->ignore_fsid_mismatch)) {
1612 device = __find_device(&cur_devices->devices,
1617 cur_devices = cur_devices->seed;
1622 struct btrfs_device *
1623 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1624 u64 devid, int instance)
1626 struct list_head *head = &fs_devices->devices;
1627 struct btrfs_device *dev;
1630 list_for_each_entry(dev, head, dev_list) {
1631 if (dev->devid == devid && num_found++ == instance)
1637 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
1639 struct cache_extent *ce;
1640 struct map_lookup *map;
1641 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1646 * During chunk recovering, we may fail to find block group's
1647 * corresponding chunk, we will rebuild it later
1649 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1650 if (!fs_info->is_chunk_recover)
1655 map = container_of(ce, struct map_lookup, ce);
1656 for (i = 0; i < map->num_stripes; i++) {
1657 if (!map->stripes[i].dev->writeable) {
1666 static struct btrfs_device *fill_missing_device(u64 devid)
1668 struct btrfs_device *device;
1670 device = kzalloc(sizeof(*device), GFP_NOFS);
1671 device->devid = devid;
1677 * slot == -1: SYSTEM chunk
1678 * return -EIO on error, otherwise return 0
1680 int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
1681 struct extent_buffer *leaf,
1682 struct btrfs_chunk *chunk,
1683 int slot, u64 logical)
1690 u32 chunk_ondisk_size;
1691 u32 sectorsize = fs_info->sectorsize;
1693 length = btrfs_chunk_length(leaf, chunk);
1694 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1695 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1696 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1697 type = btrfs_chunk_type(leaf, chunk);
1700 * These valid checks may be insufficient to cover every corner cases.
1702 if (!IS_ALIGNED(logical, sectorsize)) {
1703 error("invalid chunk logical %llu", logical);
1706 if (btrfs_chunk_sector_size(leaf, chunk) != sectorsize) {
1707 error("invalid chunk sectorsize %llu",
1708 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1711 if (!length || !IS_ALIGNED(length, sectorsize)) {
1712 error("invalid chunk length %llu", length);
1715 if (stripe_len != BTRFS_STRIPE_LEN) {
1716 error("invalid chunk stripe length: %llu", stripe_len);
1719 /* Check on chunk item type */
1720 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1721 error("invalid chunk type %llu", type);
1724 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1725 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1726 error("unrecognized chunk type: %llu",
1727 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1728 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1731 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
1732 error("missing chunk type flag: %llu", type);
1735 if (!(is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) ||
1736 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0)) {
1737 error("conflicting chunk type detected: %llu", type);
1740 if ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
1741 !is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1742 error("conflicting chunk profile detected: %llu", type);
1746 chunk_ondisk_size = btrfs_chunk_item_size(num_stripes);
1748 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1749 * it can't exceed the system chunk array size
1750 * For normal chunk, it should match its chunk item size.
1752 if (num_stripes < 1 ||
1753 (slot == -1 && chunk_ondisk_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1754 (slot >= 0 && chunk_ondisk_size > btrfs_item_size_nr(leaf, slot))) {
1755 error("invalid num_stripes: %u", num_stripes);
1759 * Device number check against profile
1761 if ((type & BTRFS_BLOCK_GROUP_RAID10 && (sub_stripes != 2 ||
1762 !IS_ALIGNED(num_stripes, sub_stripes))) ||
1763 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1764 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1765 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1766 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1767 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1768 num_stripes != 1)) {
1769 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1770 num_stripes, sub_stripes,
1771 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1779 * Slot is used to verify the chunk item is valid
1781 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1783 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
1784 struct extent_buffer *leaf,
1785 struct btrfs_chunk *chunk, int slot)
1787 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
1788 struct map_lookup *map;
1789 struct cache_extent *ce;
1793 u8 uuid[BTRFS_UUID_SIZE];
1798 logical = key->offset;
1799 length = btrfs_chunk_length(leaf, chunk);
1800 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1801 /* Validation check */
1802 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, slot, logical);
1804 error("%s checksums match, but it has an invalid chunk, %s",
1805 (slot == -1) ? "Superblock" : "Metadata",
1806 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1810 ce = search_cache_extent(&map_tree->cache_tree, logical);
1812 /* already mapped? */
1813 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1817 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1821 map->ce.start = logical;
1822 map->ce.size = length;
1823 map->num_stripes = num_stripes;
1824 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1825 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1826 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1827 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1828 map->type = btrfs_chunk_type(leaf, chunk);
1829 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1831 for (i = 0; i < num_stripes; i++) {
1832 map->stripes[i].physical =
1833 btrfs_stripe_offset_nr(leaf, chunk, i);
1834 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1835 read_extent_buffer(leaf, uuid, (unsigned long)
1836 btrfs_stripe_dev_uuid_nr(chunk, i),
1838 map->stripes[i].dev = btrfs_find_device(fs_info, devid, uuid,
1840 if (!map->stripes[i].dev) {
1841 map->stripes[i].dev = fill_missing_device(devid);
1842 printf("warning, device %llu is missing\n",
1843 (unsigned long long)devid);
1844 list_add(&map->stripes[i].dev->dev_list,
1845 &fs_info->fs_devices->devices);
1849 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1855 static int fill_device_from_item(struct extent_buffer *leaf,
1856 struct btrfs_dev_item *dev_item,
1857 struct btrfs_device *device)
1861 device->devid = btrfs_device_id(leaf, dev_item);
1862 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1863 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1864 device->type = btrfs_device_type(leaf, dev_item);
1865 device->io_align = btrfs_device_io_align(leaf, dev_item);
1866 device->io_width = btrfs_device_io_width(leaf, dev_item);
1867 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1869 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1870 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1875 static int open_seed_devices(struct btrfs_fs_info *fs_info, u8 *fsid)
1877 struct btrfs_fs_devices *fs_devices;
1880 fs_devices = fs_info->fs_devices->seed;
1881 while (fs_devices) {
1882 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1886 fs_devices = fs_devices->seed;
1889 fs_devices = find_fsid(fsid);
1891 /* missing all seed devices */
1892 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1897 INIT_LIST_HEAD(&fs_devices->devices);
1898 list_add(&fs_devices->list, &fs_uuids);
1899 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1902 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1906 fs_devices->seed = fs_info->fs_devices->seed;
1907 fs_info->fs_devices->seed = fs_devices;
1912 static int read_one_dev(struct btrfs_fs_info *fs_info,
1913 struct extent_buffer *leaf,
1914 struct btrfs_dev_item *dev_item)
1916 struct btrfs_device *device;
1919 u8 fs_uuid[BTRFS_UUID_SIZE];
1920 u8 dev_uuid[BTRFS_UUID_SIZE];
1922 devid = btrfs_device_id(leaf, dev_item);
1923 read_extent_buffer(leaf, dev_uuid,
1924 (unsigned long)btrfs_device_uuid(dev_item),
1926 read_extent_buffer(leaf, fs_uuid,
1927 (unsigned long)btrfs_device_fsid(dev_item),
1930 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
1931 ret = open_seed_devices(fs_info, fs_uuid);
1936 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
1938 device = kzalloc(sizeof(*device), GFP_NOFS);
1942 list_add(&device->dev_list,
1943 &fs_info->fs_devices->devices);
1946 fill_device_from_item(leaf, dev_item, device);
1947 device->dev_root = fs_info->dev_root;
1951 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
1953 struct btrfs_super_block *super_copy = fs_info->super_copy;
1954 struct extent_buffer *sb;
1955 struct btrfs_disk_key *disk_key;
1956 struct btrfs_chunk *chunk;
1958 unsigned long sb_array_offset;
1964 struct btrfs_key key;
1966 sb = btrfs_find_create_tree_block(fs_info,
1967 BTRFS_SUPER_INFO_OFFSET,
1968 BTRFS_SUPER_INFO_SIZE);
1971 btrfs_set_buffer_uptodate(sb);
1972 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1973 array_size = btrfs_super_sys_array_size(super_copy);
1975 array_ptr = super_copy->sys_chunk_array;
1976 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1979 while (cur_offset < array_size) {
1980 disk_key = (struct btrfs_disk_key *)array_ptr;
1981 len = sizeof(*disk_key);
1982 if (cur_offset + len > array_size)
1983 goto out_short_read;
1985 btrfs_disk_key_to_cpu(&key, disk_key);
1988 sb_array_offset += len;
1991 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1992 chunk = (struct btrfs_chunk *)sb_array_offset;
1994 * At least one btrfs_chunk with one stripe must be
1995 * present, exact stripe count check comes afterwards
1997 len = btrfs_chunk_item_size(1);
1998 if (cur_offset + len > array_size)
1999 goto out_short_read;
2001 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2004 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
2005 num_stripes, cur_offset);
2010 len = btrfs_chunk_item_size(num_stripes);
2011 if (cur_offset + len > array_size)
2012 goto out_short_read;
2014 ret = read_one_chunk(fs_info, &key, sb, chunk, -1);
2019 "ERROR: unexpected item type %u in sys_array at offset %u\n",
2020 (u32)key.type, cur_offset);
2025 sb_array_offset += len;
2028 free_extent_buffer(sb);
2032 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
2034 free_extent_buffer(sb);
2038 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
2040 struct btrfs_path *path;
2041 struct extent_buffer *leaf;
2042 struct btrfs_key key;
2043 struct btrfs_key found_key;
2044 struct btrfs_root *root = fs_info->chunk_root;
2048 path = btrfs_alloc_path();
2053 * Read all device items, and then all the chunk items. All
2054 * device items are found before any chunk item (their object id
2055 * is smaller than the lowest possible object id for a chunk
2056 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
2058 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2061 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2065 leaf = path->nodes[0];
2066 slot = path->slots[0];
2067 if (slot >= btrfs_header_nritems(leaf)) {
2068 ret = btrfs_next_leaf(root, path);
2075 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2076 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2077 struct btrfs_dev_item *dev_item;
2078 dev_item = btrfs_item_ptr(leaf, slot,
2079 struct btrfs_dev_item);
2080 ret = read_one_dev(fs_info, leaf, dev_item);
2082 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2083 struct btrfs_chunk *chunk;
2084 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2085 ret = read_one_chunk(fs_info, &found_key, leaf, chunk,
2094 btrfs_free_path(path);
2098 struct list_head *btrfs_scanned_uuids(void)
2103 static int rmw_eb(struct btrfs_fs_info *info,
2104 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2107 unsigned long orig_off = 0;
2108 unsigned long dest_off = 0;
2109 unsigned long copy_len = eb->len;
2111 ret = read_whole_eb(info, eb, 0);
2115 if (eb->start + eb->len <= orig_eb->start ||
2116 eb->start >= orig_eb->start + orig_eb->len)
2119 * | ----- orig_eb ------- |
2120 * | ----- stripe ------- |
2121 * | ----- orig_eb ------- |
2122 * | ----- orig_eb ------- |
2124 if (eb->start > orig_eb->start)
2125 orig_off = eb->start - orig_eb->start;
2126 if (orig_eb->start > eb->start)
2127 dest_off = orig_eb->start - eb->start;
2129 if (copy_len > orig_eb->len - orig_off)
2130 copy_len = orig_eb->len - orig_off;
2131 if (copy_len > eb->len - dest_off)
2132 copy_len = eb->len - dest_off;
2134 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2138 static int split_eb_for_raid56(struct btrfs_fs_info *info,
2139 struct extent_buffer *orig_eb,
2140 struct extent_buffer **ebs,
2141 u64 stripe_len, u64 *raid_map,
2144 struct extent_buffer **tmp_ebs;
2145 u64 start = orig_eb->start;
2150 tmp_ebs = calloc(num_stripes, sizeof(*tmp_ebs));
2154 /* Alloc memory in a row for data stripes */
2155 for (i = 0; i < num_stripes; i++) {
2156 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2159 tmp_ebs[i] = calloc(1, sizeof(**tmp_ebs) + stripe_len);
2166 for (i = 0; i < num_stripes; i++) {
2167 struct extent_buffer *eb = tmp_ebs[i];
2169 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2172 eb->start = raid_map[i];
2173 eb->len = stripe_len;
2177 eb->dev_bytenr = (u64)-1;
2179 this_eb_start = raid_map[i];
2181 if (start > this_eb_start ||
2182 start + orig_eb->len < this_eb_start + stripe_len) {
2183 ret = rmw_eb(info, eb, orig_eb);
2187 memcpy(eb->data, orig_eb->data + eb->start - start,
2195 for (i = 0; i < num_stripes; i++)
2201 int write_raid56_with_parity(struct btrfs_fs_info *info,
2202 struct extent_buffer *eb,
2203 struct btrfs_multi_bio *multi,
2204 u64 stripe_len, u64 *raid_map)
2206 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2209 int alloc_size = eb->len;
2212 ebs = malloc(sizeof(*ebs) * multi->num_stripes);
2213 pointers = malloc(sizeof(*pointers) * multi->num_stripes);
2214 if (!ebs || !pointers) {
2220 if (stripe_len > alloc_size)
2221 alloc_size = stripe_len;
2223 ret = split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2224 multi->num_stripes);
2228 for (i = 0; i < multi->num_stripes; i++) {
2229 struct extent_buffer *new_eb;
2230 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2231 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2232 ebs[i]->fd = multi->stripes[i].dev->fd;
2233 multi->stripes[i].dev->total_ios++;
2234 if (ebs[i]->start != raid_map[i]) {
2236 goto out_free_split;
2240 new_eb = malloc(sizeof(*eb) + alloc_size);
2243 goto out_free_split;
2245 new_eb->dev_bytenr = multi->stripes[i].physical;
2246 new_eb->fd = multi->stripes[i].dev->fd;
2247 multi->stripes[i].dev->total_ios++;
2248 new_eb->len = stripe_len;
2250 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2252 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2256 ebs[multi->num_stripes - 2] = p_eb;
2257 ebs[multi->num_stripes - 1] = q_eb;
2259 for (i = 0; i < multi->num_stripes; i++)
2260 pointers[i] = ebs[i]->data;
2262 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2264 ebs[multi->num_stripes - 1] = p_eb;
2265 for (i = 0; i < multi->num_stripes; i++)
2266 pointers[i] = ebs[i]->data;
2267 ret = raid5_gen_result(multi->num_stripes, stripe_len,
2268 multi->num_stripes - 1, pointers);
2270 goto out_free_split;
2273 for (i = 0; i < multi->num_stripes; i++) {
2274 ret = write_extent_to_disk(ebs[i]);
2276 goto out_free_split;
2280 for (i = 0; i < multi->num_stripes; i++) {
2292 * Get stripe length from chunk item and its stripe items
2294 * Caller should only call this function after validating the chunk item
2295 * by using btrfs_check_chunk_valid().
2297 u64 btrfs_stripe_length(struct btrfs_fs_info *fs_info,
2298 struct extent_buffer *leaf,
2299 struct btrfs_chunk *chunk)
2303 u32 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2304 u64 profile = btrfs_chunk_type(leaf, chunk) &
2305 BTRFS_BLOCK_GROUP_PROFILE_MASK;
2307 chunk_len = btrfs_chunk_length(leaf, chunk);
2310 case 0: /* Single profile */
2311 case BTRFS_BLOCK_GROUP_RAID1:
2312 case BTRFS_BLOCK_GROUP_DUP:
2313 stripe_len = chunk_len;
2315 case BTRFS_BLOCK_GROUP_RAID0:
2316 stripe_len = chunk_len / num_stripes;
2318 case BTRFS_BLOCK_GROUP_RAID5:
2319 stripe_len = chunk_len / (num_stripes - 1);
2321 case BTRFS_BLOCK_GROUP_RAID6:
2322 stripe_len = chunk_len / (num_stripes - 2);
2324 case BTRFS_BLOCK_GROUP_RAID10:
2325 stripe_len = chunk_len / (num_stripes /
2326 btrfs_chunk_sub_stripes(leaf, chunk));
2329 /* Invalid chunk profile found */