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"
33 struct btrfs_device *dev;
37 static inline int nr_parity_stripes(struct map_lookup *map)
39 if (map->type & BTRFS_BLOCK_GROUP_RAID5)
41 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
47 static inline int nr_data_stripes(struct map_lookup *map)
49 return map->num_stripes - nr_parity_stripes(map);
52 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
54 static LIST_HEAD(fs_uuids);
56 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
59 struct btrfs_device *dev;
60 struct list_head *cur;
62 list_for_each(cur, head) {
63 dev = list_entry(cur, struct btrfs_device, dev_list);
64 if (dev->devid == devid &&
65 !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
72 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
74 struct list_head *cur;
75 struct btrfs_fs_devices *fs_devices;
77 list_for_each(cur, &fs_uuids) {
78 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
79 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
85 static int device_list_add(const char *path,
86 struct btrfs_super_block *disk_super,
87 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
89 struct btrfs_device *device;
90 struct btrfs_fs_devices *fs_devices;
91 u64 found_transid = btrfs_super_generation(disk_super);
93 fs_devices = find_fsid(disk_super->fsid);
95 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
98 INIT_LIST_HEAD(&fs_devices->devices);
99 list_add(&fs_devices->list, &fs_uuids);
100 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
101 fs_devices->latest_devid = devid;
102 fs_devices->latest_trans = found_transid;
103 fs_devices->lowest_devid = (u64)-1;
106 device = __find_device(&fs_devices->devices, devid,
107 disk_super->dev_item.uuid);
110 device = kzalloc(sizeof(*device), GFP_NOFS);
112 /* we can safely leave the fs_devices entry around */
116 device->devid = devid;
117 device->generation = found_transid;
118 memcpy(device->uuid, disk_super->dev_item.uuid,
120 device->name = kstrdup(path, GFP_NOFS);
125 device->label = kstrdup(disk_super->label, GFP_NOFS);
126 if (!device->label) {
131 device->total_devs = btrfs_super_num_devices(disk_super);
132 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
133 device->total_bytes =
134 btrfs_stack_device_total_bytes(&disk_super->dev_item);
136 btrfs_stack_device_bytes_used(&disk_super->dev_item);
137 list_add(&device->dev_list, &fs_devices->devices);
138 device->fs_devices = fs_devices;
139 } else if (!device->name || strcmp(device->name, path)) {
140 char *name = strdup(path);
148 if (found_transid > fs_devices->latest_trans) {
149 fs_devices->latest_devid = devid;
150 fs_devices->latest_trans = found_transid;
152 if (fs_devices->lowest_devid > devid) {
153 fs_devices->lowest_devid = devid;
155 *fs_devices_ret = fs_devices;
159 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
161 struct btrfs_fs_devices *seed_devices;
162 struct btrfs_device *device;
168 while (!list_empty(&fs_devices->devices)) {
169 device = list_entry(fs_devices->devices.next,
170 struct btrfs_device, dev_list);
171 if (device->fd != -1) {
172 if (fsync(device->fd) == -1) {
173 warning("fsync on device %llu failed: %s",
174 device->devid, strerror(errno));
177 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
178 fprintf(stderr, "Warning, could not drop caches\n");
182 device->writeable = 0;
183 list_del(&device->dev_list);
184 /* free the memory */
190 seed_devices = fs_devices->seed;
191 fs_devices->seed = NULL;
193 struct btrfs_fs_devices *orig;
196 fs_devices = seed_devices;
197 list_del(&orig->list);
201 list_del(&fs_devices->list);
208 void btrfs_close_all_devices(void)
210 struct btrfs_fs_devices *fs_devices;
212 while (!list_empty(&fs_uuids)) {
213 fs_devices = list_entry(fs_uuids.next, struct btrfs_fs_devices,
215 btrfs_close_devices(fs_devices);
219 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
222 struct list_head *head = &fs_devices->devices;
223 struct list_head *cur;
224 struct btrfs_device *device;
227 list_for_each(cur, head) {
228 device = list_entry(cur, struct btrfs_device, dev_list);
230 printk("no name for device %llu, skip it now\n", device->devid);
234 fd = open(device->name, flags);
237 error("cannot open device '%s': %s", device->name,
242 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
243 fprintf(stderr, "Warning, could not drop caches\n");
245 if (device->devid == fs_devices->latest_devid)
246 fs_devices->latest_bdev = fd;
247 if (device->devid == fs_devices->lowest_devid)
248 fs_devices->lowest_bdev = fd;
251 device->writeable = 1;
255 btrfs_close_devices(fs_devices);
259 int btrfs_scan_one_device(int fd, const char *path,
260 struct btrfs_fs_devices **fs_devices_ret,
261 u64 *total_devs, u64 super_offset, unsigned sbflags)
263 struct btrfs_super_block *disk_super;
264 char buf[BTRFS_SUPER_INFO_SIZE];
268 disk_super = (struct btrfs_super_block *)buf;
269 ret = btrfs_read_dev_super(fd, disk_super, super_offset, sbflags);
272 devid = btrfs_stack_device_id(&disk_super->dev_item);
273 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
276 *total_devs = btrfs_super_num_devices(disk_super);
278 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
284 * find_free_dev_extent_start - find free space in the specified device
285 * @device: the device which we search the free space in
286 * @num_bytes: the size of the free space that we need
287 * @search_start: the position from which to begin the search
288 * @start: store the start of the free space.
289 * @len: the size of the free space. that we find, or the size
290 * of the max free space if we don't find suitable free space
292 * this uses a pretty simple search, the expectation is that it is
293 * called very infrequently and that a given device has a small number
296 * @start is used to store the start of the free space if we find. But if we
297 * don't find suitable free space, it will be used to store the start position
298 * of the max free space.
300 * @len is used to store the size of the free space that we find.
301 * But if we don't find suitable free space, it is used to store the size of
302 * the max free space.
304 static int find_free_dev_extent_start(struct btrfs_trans_handle *trans,
305 struct btrfs_device *device, u64 num_bytes,
306 u64 search_start, u64 *start, u64 *len)
308 struct btrfs_key key;
309 struct btrfs_root *root = device->dev_root;
310 struct btrfs_dev_extent *dev_extent;
311 struct btrfs_path *path;
316 u64 search_end = device->total_bytes;
319 struct extent_buffer *l;
320 u64 min_search_start;
323 * We don't want to overwrite the superblock on the drive nor any area
324 * used by the boot loader (grub for example), so we make sure to start
325 * at an offset of at least 1MB.
327 min_search_start = max(root->fs_info->alloc_start, (u64)SZ_1M);
328 search_start = max(search_start, min_search_start);
330 path = btrfs_alloc_path();
334 max_hole_start = search_start;
337 if (search_start >= search_end) {
344 key.objectid = device->devid;
345 key.offset = search_start;
346 key.type = BTRFS_DEV_EXTENT_KEY;
348 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
352 ret = btrfs_previous_item(root, path, key.objectid, key.type);
359 slot = path->slots[0];
360 if (slot >= btrfs_header_nritems(l)) {
361 ret = btrfs_next_leaf(root, path);
369 btrfs_item_key_to_cpu(l, &key, slot);
371 if (key.objectid < device->devid)
374 if (key.objectid > device->devid)
377 if (key.type != BTRFS_DEV_EXTENT_KEY)
380 if (key.offset > search_start) {
381 hole_size = key.offset - search_start;
384 * Have to check before we set max_hole_start, otherwise
385 * we could end up sending back this offset anyway.
387 if (hole_size > max_hole_size) {
388 max_hole_start = search_start;
389 max_hole_size = hole_size;
393 * If this free space is greater than which we need,
394 * it must be the max free space that we have found
395 * until now, so max_hole_start must point to the start
396 * of this free space and the length of this free space
397 * is stored in max_hole_size. Thus, we return
398 * max_hole_start and max_hole_size and go back to the
401 if (hole_size >= num_bytes) {
407 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
408 extent_end = key.offset + btrfs_dev_extent_length(l,
410 if (extent_end > search_start)
411 search_start = extent_end;
418 * At this point, search_start should be the end of
419 * allocated dev extents, and when shrinking the device,
420 * search_end may be smaller than search_start.
422 if (search_end > search_start) {
423 hole_size = search_end - search_start;
425 if (hole_size > max_hole_size) {
426 max_hole_start = search_start;
427 max_hole_size = hole_size;
432 if (max_hole_size < num_bytes)
438 btrfs_free_path(path);
439 *start = max_hole_start;
441 *len = max_hole_size;
445 int find_free_dev_extent(struct btrfs_trans_handle *trans,
446 struct btrfs_device *device, u64 num_bytes,
449 /* FIXME use last free of some kind */
450 return find_free_dev_extent_start(trans, device,
451 num_bytes, 0, start, NULL);
454 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
455 struct btrfs_device *device,
456 u64 chunk_tree, u64 chunk_objectid,
458 u64 num_bytes, u64 *start, int convert)
461 struct btrfs_path *path;
462 struct btrfs_root *root = device->dev_root;
463 struct btrfs_dev_extent *extent;
464 struct extent_buffer *leaf;
465 struct btrfs_key key;
467 path = btrfs_alloc_path();
472 * For convert case, just skip search free dev_extent, as caller
473 * is responsible to make sure it's free.
476 ret = find_free_dev_extent(trans, device, num_bytes,
482 key.objectid = device->devid;
484 key.type = BTRFS_DEV_EXTENT_KEY;
485 ret = btrfs_insert_empty_item(trans, root, path, &key,
489 leaf = path->nodes[0];
490 extent = btrfs_item_ptr(leaf, path->slots[0],
491 struct btrfs_dev_extent);
492 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
493 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
494 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
496 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
497 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
500 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
501 btrfs_mark_buffer_dirty(leaf);
503 btrfs_free_path(path);
507 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
509 struct btrfs_path *path;
511 struct btrfs_key key;
512 struct btrfs_chunk *chunk;
513 struct btrfs_key found_key;
515 path = btrfs_alloc_path();
519 key.objectid = objectid;
520 key.offset = (u64)-1;
521 key.type = BTRFS_CHUNK_ITEM_KEY;
523 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
529 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
533 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
535 if (found_key.objectid != objectid)
538 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
540 *offset = found_key.offset +
541 btrfs_chunk_length(path->nodes[0], chunk);
546 btrfs_free_path(path);
550 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
554 struct btrfs_key key;
555 struct btrfs_key found_key;
557 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
558 key.type = BTRFS_DEV_ITEM_KEY;
559 key.offset = (u64)-1;
561 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
567 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
572 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
574 *objectid = found_key.offset + 1;
578 btrfs_release_path(path);
583 * the device information is stored in the chunk root
584 * the btrfs_device struct should be fully filled in
586 int btrfs_add_device(struct btrfs_trans_handle *trans,
587 struct btrfs_root *root,
588 struct btrfs_device *device)
591 struct btrfs_path *path;
592 struct btrfs_dev_item *dev_item;
593 struct extent_buffer *leaf;
594 struct btrfs_key key;
598 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, root->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_root *root,
693 struct btrfs_key *key,
694 struct btrfs_chunk *chunk, int item_size)
696 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
697 struct btrfs_disk_key disk_key;
701 array_size = btrfs_super_sys_array_size(super_copy);
702 if (array_size + item_size + sizeof(disk_key)
703 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
706 ptr = super_copy->sys_chunk_array + array_size;
707 btrfs_cpu_key_to_disk(&disk_key, key);
708 memcpy(ptr, &disk_key, sizeof(disk_key));
709 ptr += sizeof(disk_key);
710 memcpy(ptr, chunk, item_size);
711 item_size += sizeof(disk_key);
712 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
716 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
719 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
721 else if (type & BTRFS_BLOCK_GROUP_RAID10)
722 return calc_size * (num_stripes / sub_stripes);
723 else if (type & BTRFS_BLOCK_GROUP_RAID5)
724 return calc_size * (num_stripes - 1);
725 else if (type & BTRFS_BLOCK_GROUP_RAID6)
726 return calc_size * (num_stripes - 2);
728 return calc_size * num_stripes;
732 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
734 /* TODO, add a way to store the preferred stripe size */
735 return BTRFS_STRIPE_LEN;
739 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
741 * It is not equal to "device->total_bytes - device->bytes_used".
742 * We do not allocate any chunk in 1M at beginning of device, and not
743 * allowed to allocate any chunk before alloc_start if it is specified.
744 * So search holes from max(1M, alloc_start) to device->total_bytes.
746 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
747 struct btrfs_device *device,
750 struct btrfs_path *path;
751 struct btrfs_root *root = device->dev_root;
752 struct btrfs_key key;
753 struct btrfs_dev_extent *dev_extent = NULL;
754 struct extent_buffer *l;
755 u64 search_start = root->fs_info->alloc_start;
756 u64 search_end = device->total_bytes;
762 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
764 path = btrfs_alloc_path();
768 key.objectid = device->devid;
769 key.offset = root->fs_info->alloc_start;
770 key.type = BTRFS_DEV_EXTENT_KEY;
773 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
776 ret = btrfs_previous_item(root, path, 0, key.type);
782 slot = path->slots[0];
783 if (slot >= btrfs_header_nritems(l)) {
784 ret = btrfs_next_leaf(root, path);
791 btrfs_item_key_to_cpu(l, &key, slot);
793 if (key.objectid < device->devid)
795 if (key.objectid > device->devid)
797 if (key.type != BTRFS_DEV_EXTENT_KEY)
799 if (key.offset > search_end)
801 if (key.offset > search_start)
802 free_bytes += key.offset - search_start;
804 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
805 extent_end = key.offset + btrfs_dev_extent_length(l,
807 if (extent_end > search_start)
808 search_start = extent_end;
809 if (search_start > search_end)
816 if (search_start < search_end)
817 free_bytes += search_end - search_start;
819 *avail_bytes = free_bytes;
822 btrfs_free_path(path);
826 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
827 - sizeof(struct btrfs_item) \
828 - sizeof(struct btrfs_chunk)) \
829 / sizeof(struct btrfs_stripe) + 1)
831 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
832 - 2 * sizeof(struct btrfs_disk_key) \
833 - 2 * sizeof(struct btrfs_chunk)) \
834 / sizeof(struct btrfs_stripe) + 1)
836 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
837 struct btrfs_root *extent_root, u64 *start,
838 u64 *num_bytes, u64 type)
841 struct btrfs_fs_info *info = extent_root->fs_info;
842 struct btrfs_root *chunk_root = info->chunk_root;
843 struct btrfs_stripe *stripes;
844 struct btrfs_device *device = NULL;
845 struct btrfs_chunk *chunk;
846 struct list_head private_devs;
847 struct list_head *dev_list = &info->fs_devices->devices;
848 struct list_head *cur;
849 struct map_lookup *map;
850 int min_stripe_size = SZ_1M;
851 u64 calc_size = SZ_8M;
853 u64 max_chunk_size = 4 * calc_size;
864 int stripe_len = BTRFS_STRIPE_LEN;
865 struct btrfs_key key;
868 if (list_empty(dev_list)) {
872 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
873 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
874 BTRFS_BLOCK_GROUP_RAID10 |
875 BTRFS_BLOCK_GROUP_DUP)) {
876 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
878 max_chunk_size = calc_size * 2;
879 min_stripe_size = SZ_1M;
880 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
881 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
883 max_chunk_size = 10 * calc_size;
884 min_stripe_size = SZ_64M;
885 max_stripes = BTRFS_MAX_DEVS(chunk_root);
886 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
888 max_chunk_size = 4 * calc_size;
889 min_stripe_size = SZ_32M;
890 max_stripes = BTRFS_MAX_DEVS(chunk_root);
893 if (type & BTRFS_BLOCK_GROUP_RAID1) {
894 num_stripes = min_t(u64, 2,
895 btrfs_super_num_devices(info->super_copy));
900 if (type & BTRFS_BLOCK_GROUP_DUP) {
904 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
905 num_stripes = btrfs_super_num_devices(info->super_copy);
906 if (num_stripes > max_stripes)
907 num_stripes = max_stripes;
910 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
911 num_stripes = btrfs_super_num_devices(info->super_copy);
912 if (num_stripes > max_stripes)
913 num_stripes = max_stripes;
916 num_stripes &= ~(u32)1;
920 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
921 num_stripes = btrfs_super_num_devices(info->super_copy);
922 if (num_stripes > max_stripes)
923 num_stripes = max_stripes;
927 stripe_len = find_raid56_stripe_len(num_stripes - 1,
928 btrfs_super_stripesize(info->super_copy));
930 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
931 num_stripes = btrfs_super_num_devices(info->super_copy);
932 if (num_stripes > max_stripes)
933 num_stripes = max_stripes;
937 stripe_len = find_raid56_stripe_len(num_stripes - 2,
938 btrfs_super_stripesize(info->super_copy));
941 /* we don't want a chunk larger than 10% of the FS */
942 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
943 max_chunk_size = min(percent_max, max_chunk_size);
946 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
948 calc_size = max_chunk_size;
949 calc_size /= num_stripes;
950 calc_size /= stripe_len;
951 calc_size *= stripe_len;
953 /* we don't want tiny stripes */
954 calc_size = max_t(u64, calc_size, min_stripe_size);
956 calc_size /= stripe_len;
957 calc_size *= stripe_len;
958 INIT_LIST_HEAD(&private_devs);
959 cur = dev_list->next;
962 if (type & BTRFS_BLOCK_GROUP_DUP)
963 min_free = calc_size * 2;
965 min_free = calc_size;
967 /* build a private list of devices we will allocate from */
968 while(index < num_stripes) {
969 device = list_entry(cur, struct btrfs_device, dev_list);
970 ret = btrfs_device_avail_bytes(trans, device, &avail);
974 if (avail >= min_free) {
975 list_move_tail(&device->dev_list, &private_devs);
977 if (type & BTRFS_BLOCK_GROUP_DUP)
979 } else if (avail > max_avail)
984 if (index < num_stripes) {
985 list_splice(&private_devs, dev_list);
986 if (index >= min_stripes) {
988 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
989 num_stripes /= sub_stripes;
990 num_stripes *= sub_stripes;
995 if (!looped && max_avail > 0) {
997 calc_size = max_avail;
1002 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1006 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1007 key.type = BTRFS_CHUNK_ITEM_KEY;
1008 key.offset = offset;
1010 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1014 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1020 stripes = &chunk->stripe;
1021 *num_bytes = chunk_bytes_by_type(type, calc_size,
1022 num_stripes, sub_stripes);
1024 while(index < num_stripes) {
1025 struct btrfs_stripe *stripe;
1026 BUG_ON(list_empty(&private_devs));
1027 cur = private_devs.next;
1028 device = list_entry(cur, struct btrfs_device, dev_list);
1030 /* loop over this device again if we're doing a dup group */
1031 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1032 (index == num_stripes - 1))
1033 list_move_tail(&device->dev_list, dev_list);
1035 ret = btrfs_alloc_dev_extent(trans, device,
1036 info->chunk_root->root_key.objectid,
1037 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1038 calc_size, &dev_offset, 0);
1041 device->bytes_used += calc_size;
1042 ret = btrfs_update_device(trans, device);
1045 map->stripes[index].dev = device;
1046 map->stripes[index].physical = dev_offset;
1047 stripe = stripes + index;
1048 btrfs_set_stack_stripe_devid(stripe, device->devid);
1049 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1050 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1053 BUG_ON(!list_empty(&private_devs));
1055 /* key was set above */
1056 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1057 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1058 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1059 btrfs_set_stack_chunk_type(chunk, type);
1060 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1061 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1062 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1063 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1064 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1065 map->sector_size = extent_root->sectorsize;
1066 map->stripe_len = stripe_len;
1067 map->io_align = stripe_len;
1068 map->io_width = stripe_len;
1070 map->num_stripes = num_stripes;
1071 map->sub_stripes = sub_stripes;
1073 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1074 btrfs_chunk_item_size(num_stripes));
1076 *start = key.offset;;
1078 map->ce.start = key.offset;
1079 map->ce.size = *num_bytes;
1081 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1084 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1085 ret = btrfs_add_system_chunk(chunk_root, &key,
1086 chunk, btrfs_chunk_item_size(num_stripes));
1095 * Alloc a DATA chunk with SINGLE profile.
1097 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1098 * (btrfs logical bytenr == on-disk bytenr)
1099 * For that case, caller must make sure the chunk and dev_extent are not
1102 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1103 struct btrfs_root *extent_root, u64 *start,
1104 u64 num_bytes, u64 type, int convert)
1107 struct btrfs_fs_info *info = extent_root->fs_info;
1108 struct btrfs_root *chunk_root = info->chunk_root;
1109 struct btrfs_stripe *stripes;
1110 struct btrfs_device *device = NULL;
1111 struct btrfs_chunk *chunk;
1112 struct list_head *dev_list = &info->fs_devices->devices;
1113 struct list_head *cur;
1114 struct map_lookup *map;
1115 u64 calc_size = SZ_8M;
1116 int num_stripes = 1;
1117 int sub_stripes = 0;
1120 int stripe_len = BTRFS_STRIPE_LEN;
1121 struct btrfs_key key;
1123 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1124 key.type = BTRFS_CHUNK_ITEM_KEY;
1126 if (*start != round_down(*start, extent_root->sectorsize)) {
1127 error("DATA chunk start not sectorsize aligned: %llu",
1128 (unsigned long long)*start);
1131 key.offset = *start;
1132 dev_offset = *start;
1136 ret = find_next_chunk(chunk_root,
1137 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1144 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1148 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1154 stripes = &chunk->stripe;
1155 calc_size = num_bytes;
1158 cur = dev_list->next;
1159 device = list_entry(cur, struct btrfs_device, dev_list);
1161 while (index < num_stripes) {
1162 struct btrfs_stripe *stripe;
1164 ret = btrfs_alloc_dev_extent(trans, device,
1165 info->chunk_root->root_key.objectid,
1166 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1167 calc_size, &dev_offset, convert);
1170 device->bytes_used += calc_size;
1171 ret = btrfs_update_device(trans, device);
1174 map->stripes[index].dev = device;
1175 map->stripes[index].physical = dev_offset;
1176 stripe = stripes + index;
1177 btrfs_set_stack_stripe_devid(stripe, device->devid);
1178 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1179 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1183 /* key was set above */
1184 btrfs_set_stack_chunk_length(chunk, num_bytes);
1185 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1186 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1187 btrfs_set_stack_chunk_type(chunk, type);
1188 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1189 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1190 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1191 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1192 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1193 map->sector_size = extent_root->sectorsize;
1194 map->stripe_len = stripe_len;
1195 map->io_align = stripe_len;
1196 map->io_width = stripe_len;
1198 map->num_stripes = num_stripes;
1199 map->sub_stripes = sub_stripes;
1201 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1202 btrfs_chunk_item_size(num_stripes));
1205 *start = key.offset;
1207 map->ce.start = key.offset;
1208 map->ce.size = num_bytes;
1210 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1217 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1219 struct cache_extent *ce;
1220 struct map_lookup *map;
1223 ce = search_cache_extent(&map_tree->cache_tree, logical);
1225 fprintf(stderr, "No mapping for %llu-%llu\n",
1226 (unsigned long long)logical,
1227 (unsigned long long)logical+len);
1230 if (ce->start > logical || ce->start + ce->size < logical) {
1231 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1232 "%llu-%llu\n", (unsigned long long)logical,
1233 (unsigned long long)logical+len,
1234 (unsigned long long)ce->start,
1235 (unsigned long long)ce->start + ce->size);
1238 map = container_of(ce, struct map_lookup, ce);
1240 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1241 ret = map->num_stripes;
1242 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1243 ret = map->sub_stripes;
1244 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1246 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1253 int btrfs_next_bg(struct btrfs_mapping_tree *map_tree, u64 *logical,
1254 u64 *size, u64 type)
1256 struct cache_extent *ce;
1257 struct map_lookup *map;
1260 ce = search_cache_extent(&map_tree->cache_tree, cur);
1264 * only jump to next bg if our cur is not 0
1265 * As the initial logical for btrfs_next_bg() is 0, and
1266 * if we jump to next bg, we skipped a valid bg.
1269 ce = next_cache_extent(ce);
1275 map = container_of(ce, struct map_lookup, ce);
1276 if (map->type & type) {
1277 *logical = ce->start;
1286 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1287 u64 chunk_start, u64 physical, u64 devid,
1288 u64 **logical, int *naddrs, int *stripe_len)
1290 struct cache_extent *ce;
1291 struct map_lookup *map;
1299 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1301 map = container_of(ce, struct map_lookup, ce);
1304 rmap_len = map->stripe_len;
1305 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1306 length = ce->size / (map->num_stripes / map->sub_stripes);
1307 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1308 length = ce->size / map->num_stripes;
1309 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1310 BTRFS_BLOCK_GROUP_RAID6)) {
1311 length = ce->size / nr_data_stripes(map);
1312 rmap_len = map->stripe_len * nr_data_stripes(map);
1315 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1317 for (i = 0; i < map->num_stripes; i++) {
1318 if (devid && map->stripes[i].dev->devid != devid)
1320 if (map->stripes[i].physical > physical ||
1321 map->stripes[i].physical + length <= physical)
1324 stripe_nr = (physical - map->stripes[i].physical) /
1327 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1328 stripe_nr = (stripe_nr * map->num_stripes + i) /
1330 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1331 stripe_nr = stripe_nr * map->num_stripes + i;
1332 } /* else if RAID[56], multiply by nr_data_stripes().
1333 * Alternatively, just use rmap_len below instead of
1334 * map->stripe_len */
1336 bytenr = ce->start + stripe_nr * rmap_len;
1337 for (j = 0; j < nr; j++) {
1338 if (buf[j] == bytenr)
1347 *stripe_len = rmap_len;
1352 static inline int parity_smaller(u64 a, u64 b)
1357 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1358 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1360 struct btrfs_bio_stripe s;
1367 for (i = 0; i < bbio->num_stripes - 1; i++) {
1368 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1369 s = bbio->stripes[i];
1371 bbio->stripes[i] = bbio->stripes[i+1];
1372 raid_map[i] = raid_map[i+1];
1373 bbio->stripes[i+1] = s;
1381 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1382 u64 logical, u64 *length,
1383 struct btrfs_multi_bio **multi_ret, int mirror_num,
1386 return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1387 multi_ret, mirror_num, raid_map_ret);
1390 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1391 u64 logical, u64 *length, u64 *type,
1392 struct btrfs_multi_bio **multi_ret, int mirror_num,
1395 struct cache_extent *ce;
1396 struct map_lookup *map;
1400 u64 *raid_map = NULL;
1401 int stripes_allocated = 8;
1402 int stripes_required = 1;
1405 struct btrfs_multi_bio *multi = NULL;
1407 if (multi_ret && rw == READ) {
1408 stripes_allocated = 1;
1411 ce = search_cache_extent(&map_tree->cache_tree, logical);
1417 if (ce->start > logical) {
1419 *length = ce->start - logical;
1424 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1429 map = container_of(ce, struct map_lookup, ce);
1430 offset = logical - ce->start;
1433 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1434 BTRFS_BLOCK_GROUP_DUP)) {
1435 stripes_required = map->num_stripes;
1436 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1437 stripes_required = map->sub_stripes;
1440 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1441 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1442 /* RAID[56] write or recovery. Return all stripes */
1443 stripes_required = map->num_stripes;
1445 /* Only allocate the map if we've already got a large enough multi_ret */
1446 if (stripes_allocated >= stripes_required) {
1447 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1455 /* if our multi bio struct is too small, back off and try again */
1456 if (multi_ret && stripes_allocated < stripes_required) {
1457 stripes_allocated = stripes_required;
1464 * stripe_nr counts the total number of stripes we have to stride
1465 * to get to this block
1467 stripe_nr = stripe_nr / map->stripe_len;
1469 stripe_offset = stripe_nr * map->stripe_len;
1470 BUG_ON(offset < stripe_offset);
1472 /* stripe_offset is the offset of this block in its stripe*/
1473 stripe_offset = offset - stripe_offset;
1475 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1476 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1477 BTRFS_BLOCK_GROUP_RAID10 |
1478 BTRFS_BLOCK_GROUP_DUP)) {
1479 /* we limit the length of each bio to what fits in a stripe */
1480 *length = min_t(u64, ce->size - offset,
1481 map->stripe_len - stripe_offset);
1483 *length = ce->size - offset;
1489 multi->num_stripes = 1;
1491 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1493 multi->num_stripes = map->num_stripes;
1494 else if (mirror_num)
1495 stripe_index = mirror_num - 1;
1497 stripe_index = stripe_nr % map->num_stripes;
1498 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1499 int factor = map->num_stripes / map->sub_stripes;
1501 stripe_index = stripe_nr % factor;
1502 stripe_index *= map->sub_stripes;
1505 multi->num_stripes = map->sub_stripes;
1506 else if (mirror_num)
1507 stripe_index += mirror_num - 1;
1509 stripe_nr = stripe_nr / factor;
1510 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1512 multi->num_stripes = map->num_stripes;
1513 else if (mirror_num)
1514 stripe_index = mirror_num - 1;
1515 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1516 BTRFS_BLOCK_GROUP_RAID6)) {
1521 u64 raid56_full_stripe_start;
1522 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1525 * align the start of our data stripe in the logical
1528 raid56_full_stripe_start = offset / full_stripe_len;
1529 raid56_full_stripe_start *= full_stripe_len;
1531 /* get the data stripe number */
1532 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1533 stripe_nr = stripe_nr / nr_data_stripes(map);
1535 /* Work out the disk rotation on this stripe-set */
1536 rot = stripe_nr % map->num_stripes;
1538 /* Fill in the logical address of each stripe */
1539 tmp = stripe_nr * nr_data_stripes(map);
1541 for (i = 0; i < nr_data_stripes(map); i++)
1542 raid_map[(i+rot) % map->num_stripes] =
1543 ce->start + (tmp + i) * map->stripe_len;
1545 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1546 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1547 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1549 *length = map->stripe_len;
1552 multi->num_stripes = map->num_stripes;
1554 stripe_index = stripe_nr % nr_data_stripes(map);
1555 stripe_nr = stripe_nr / nr_data_stripes(map);
1558 * Mirror #0 or #1 means the original data block.
1559 * Mirror #2 is RAID5 parity block.
1560 * Mirror #3 is RAID6 Q block.
1563 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1565 /* We distribute the parity blocks across stripes */
1566 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1570 * after this do_div call, stripe_nr is the number of stripes
1571 * on this device we have to walk to find the data, and
1572 * stripe_index is the number of our device in the stripe array
1574 stripe_index = stripe_nr % map->num_stripes;
1575 stripe_nr = stripe_nr / map->num_stripes;
1577 BUG_ON(stripe_index >= map->num_stripes);
1579 for (i = 0; i < multi->num_stripes; i++) {
1580 multi->stripes[i].physical =
1581 map->stripes[stripe_index].physical + stripe_offset +
1582 stripe_nr * map->stripe_len;
1583 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1592 sort_parity_stripes(multi, raid_map);
1593 *raid_map_ret = raid_map;
1599 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1602 struct btrfs_device *device;
1603 struct btrfs_fs_devices *cur_devices;
1605 cur_devices = root->fs_info->fs_devices;
1606 while (cur_devices) {
1608 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1609 root->fs_info->ignore_fsid_mismatch)) {
1610 device = __find_device(&cur_devices->devices,
1615 cur_devices = cur_devices->seed;
1620 struct btrfs_device *
1621 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1622 u64 devid, int instance)
1624 struct list_head *head = &fs_devices->devices;
1625 struct btrfs_device *dev;
1628 list_for_each_entry(dev, head, dev_list) {
1629 if (dev->devid == devid && num_found++ == instance)
1635 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1637 struct cache_extent *ce;
1638 struct map_lookup *map;
1639 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1644 * During chunk recovering, we may fail to find block group's
1645 * corresponding chunk, we will rebuild it later
1647 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1648 if (!root->fs_info->is_chunk_recover)
1653 map = container_of(ce, struct map_lookup, ce);
1654 for (i = 0; i < map->num_stripes; i++) {
1655 if (!map->stripes[i].dev->writeable) {
1664 static struct btrfs_device *fill_missing_device(u64 devid)
1666 struct btrfs_device *device;
1668 device = kzalloc(sizeof(*device), GFP_NOFS);
1669 device->devid = devid;
1675 * slot == -1: SYSTEM chunk
1676 * return -EIO on error, otherwise return 0
1678 int btrfs_check_chunk_valid(struct btrfs_root *root,
1679 struct extent_buffer *leaf,
1680 struct btrfs_chunk *chunk,
1681 int slot, u64 logical)
1689 length = btrfs_chunk_length(leaf, chunk);
1690 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1691 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1692 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1693 type = btrfs_chunk_type(leaf, chunk);
1696 * These valid checks may be insufficient to cover every corner cases.
1698 if (!IS_ALIGNED(logical, root->sectorsize)) {
1699 error("invalid chunk logical %llu", logical);
1702 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
1703 error("invalid chunk sectorsize %llu",
1704 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1707 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
1708 error("invalid chunk length %llu", length);
1711 if (stripe_len != BTRFS_STRIPE_LEN) {
1712 error("invalid chunk stripe length: %llu", stripe_len);
1715 /* Check on chunk item type */
1716 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1717 error("invalid chunk type %llu", type);
1720 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1721 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1722 error("unrecognized chunk type: %llu",
1723 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1724 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1728 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1729 * it can't exceed the system chunk array size
1730 * For normal chunk, it should match its chunk item size.
1732 if (num_stripes < 1 ||
1733 (slot == -1 && sizeof(struct btrfs_stripe) * num_stripes >
1734 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1735 (slot >= 0 && sizeof(struct btrfs_stripe) * (num_stripes - 1) >
1736 btrfs_item_size_nr(leaf, slot))) {
1737 error("invalid num_stripes: %u", num_stripes);
1741 * Device number check against profile
1743 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes == 0) ||
1744 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1745 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1746 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1747 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1748 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1749 num_stripes != 1)) {
1750 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1751 num_stripes, sub_stripes,
1752 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1760 * Slot is used to verify the chunk item is valid
1762 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1764 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1765 struct extent_buffer *leaf,
1766 struct btrfs_chunk *chunk, int slot)
1768 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1769 struct map_lookup *map;
1770 struct cache_extent *ce;
1774 u8 uuid[BTRFS_UUID_SIZE];
1779 logical = key->offset;
1780 length = btrfs_chunk_length(leaf, chunk);
1781 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1782 /* Validation check */
1783 ret = btrfs_check_chunk_valid(root, leaf, chunk, slot, logical);
1785 error("%s checksums match, but it has an invalid chunk, %s",
1786 (slot == -1) ? "Superblock" : "Metadata",
1787 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1791 ce = search_cache_extent(&map_tree->cache_tree, logical);
1793 /* already mapped? */
1794 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1798 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1802 map->ce.start = logical;
1803 map->ce.size = length;
1804 map->num_stripes = num_stripes;
1805 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1806 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1807 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1808 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1809 map->type = btrfs_chunk_type(leaf, chunk);
1810 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1812 for (i = 0; i < num_stripes; i++) {
1813 map->stripes[i].physical =
1814 btrfs_stripe_offset_nr(leaf, chunk, i);
1815 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1816 read_extent_buffer(leaf, uuid, (unsigned long)
1817 btrfs_stripe_dev_uuid_nr(chunk, i),
1819 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1821 if (!map->stripes[i].dev) {
1822 map->stripes[i].dev = fill_missing_device(devid);
1823 printf("warning, device %llu is missing\n",
1824 (unsigned long long)devid);
1825 list_add(&map->stripes[i].dev->dev_list,
1826 &root->fs_info->fs_devices->devices);
1830 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1836 static int fill_device_from_item(struct extent_buffer *leaf,
1837 struct btrfs_dev_item *dev_item,
1838 struct btrfs_device *device)
1842 device->devid = btrfs_device_id(leaf, dev_item);
1843 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1844 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1845 device->type = btrfs_device_type(leaf, dev_item);
1846 device->io_align = btrfs_device_io_align(leaf, dev_item);
1847 device->io_width = btrfs_device_io_width(leaf, dev_item);
1848 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1850 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1851 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1856 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1858 struct btrfs_fs_devices *fs_devices;
1861 fs_devices = root->fs_info->fs_devices->seed;
1862 while (fs_devices) {
1863 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1867 fs_devices = fs_devices->seed;
1870 fs_devices = find_fsid(fsid);
1872 /* missing all seed devices */
1873 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1878 INIT_LIST_HEAD(&fs_devices->devices);
1879 list_add(&fs_devices->list, &fs_uuids);
1880 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1883 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1887 fs_devices->seed = root->fs_info->fs_devices->seed;
1888 root->fs_info->fs_devices->seed = fs_devices;
1893 static int read_one_dev(struct btrfs_root *root,
1894 struct extent_buffer *leaf,
1895 struct btrfs_dev_item *dev_item)
1897 struct btrfs_device *device;
1900 u8 fs_uuid[BTRFS_UUID_SIZE];
1901 u8 dev_uuid[BTRFS_UUID_SIZE];
1903 devid = btrfs_device_id(leaf, dev_item);
1904 read_extent_buffer(leaf, dev_uuid,
1905 (unsigned long)btrfs_device_uuid(dev_item),
1907 read_extent_buffer(leaf, fs_uuid,
1908 (unsigned long)btrfs_device_fsid(dev_item),
1911 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1912 ret = open_seed_devices(root, fs_uuid);
1917 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1919 device = kzalloc(sizeof(*device), GFP_NOFS);
1923 list_add(&device->dev_list,
1924 &root->fs_info->fs_devices->devices);
1927 fill_device_from_item(leaf, dev_item, device);
1928 device->dev_root = root->fs_info->dev_root;
1932 int btrfs_read_sys_array(struct btrfs_root *root)
1934 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1935 struct extent_buffer *sb;
1936 struct btrfs_disk_key *disk_key;
1937 struct btrfs_chunk *chunk;
1939 unsigned long sb_array_offset;
1945 struct btrfs_key key;
1947 sb = btrfs_find_create_tree_block(root->fs_info,
1948 BTRFS_SUPER_INFO_OFFSET,
1949 BTRFS_SUPER_INFO_SIZE);
1952 btrfs_set_buffer_uptodate(sb);
1953 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1954 array_size = btrfs_super_sys_array_size(super_copy);
1956 array_ptr = super_copy->sys_chunk_array;
1957 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1960 while (cur_offset < array_size) {
1961 disk_key = (struct btrfs_disk_key *)array_ptr;
1962 len = sizeof(*disk_key);
1963 if (cur_offset + len > array_size)
1964 goto out_short_read;
1966 btrfs_disk_key_to_cpu(&key, disk_key);
1969 sb_array_offset += len;
1972 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1973 chunk = (struct btrfs_chunk *)sb_array_offset;
1975 * At least one btrfs_chunk with one stripe must be
1976 * present, exact stripe count check comes afterwards
1978 len = btrfs_chunk_item_size(1);
1979 if (cur_offset + len > array_size)
1980 goto out_short_read;
1982 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1985 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
1986 num_stripes, cur_offset);
1991 len = btrfs_chunk_item_size(num_stripes);
1992 if (cur_offset + len > array_size)
1993 goto out_short_read;
1995 ret = read_one_chunk(root, &key, sb, chunk, -1);
2000 "ERROR: unexpected item type %u in sys_array at offset %u\n",
2001 (u32)key.type, cur_offset);
2006 sb_array_offset += len;
2009 free_extent_buffer(sb);
2013 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
2015 free_extent_buffer(sb);
2019 int btrfs_read_chunk_tree(struct btrfs_root *root)
2021 struct btrfs_path *path;
2022 struct extent_buffer *leaf;
2023 struct btrfs_key key;
2024 struct btrfs_key found_key;
2028 root = root->fs_info->chunk_root;
2030 path = btrfs_alloc_path();
2035 * Read all device items, and then all the chunk items. All
2036 * device items are found before any chunk item (their object id
2037 * is smaller than the lowest possible object id for a chunk
2038 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
2040 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2043 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2047 leaf = path->nodes[0];
2048 slot = path->slots[0];
2049 if (slot >= btrfs_header_nritems(leaf)) {
2050 ret = btrfs_next_leaf(root, path);
2057 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2058 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2059 struct btrfs_dev_item *dev_item;
2060 dev_item = btrfs_item_ptr(leaf, slot,
2061 struct btrfs_dev_item);
2062 ret = read_one_dev(root, leaf, dev_item);
2064 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2065 struct btrfs_chunk *chunk;
2066 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2067 ret = read_one_chunk(root, &found_key, leaf, chunk,
2076 btrfs_free_path(path);
2080 struct list_head *btrfs_scanned_uuids(void)
2085 static int rmw_eb(struct btrfs_fs_info *info,
2086 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2089 unsigned long orig_off = 0;
2090 unsigned long dest_off = 0;
2091 unsigned long copy_len = eb->len;
2093 ret = read_whole_eb(info, eb, 0);
2097 if (eb->start + eb->len <= orig_eb->start ||
2098 eb->start >= orig_eb->start + orig_eb->len)
2101 * | ----- orig_eb ------- |
2102 * | ----- stripe ------- |
2103 * | ----- orig_eb ------- |
2104 * | ----- orig_eb ------- |
2106 if (eb->start > orig_eb->start)
2107 orig_off = eb->start - orig_eb->start;
2108 if (orig_eb->start > eb->start)
2109 dest_off = orig_eb->start - eb->start;
2111 if (copy_len > orig_eb->len - orig_off)
2112 copy_len = orig_eb->len - orig_off;
2113 if (copy_len > eb->len - dest_off)
2114 copy_len = eb->len - dest_off;
2116 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2120 static int split_eb_for_raid56(struct btrfs_fs_info *info,
2121 struct extent_buffer *orig_eb,
2122 struct extent_buffer **ebs,
2123 u64 stripe_len, u64 *raid_map,
2126 struct extent_buffer **tmp_ebs;
2127 u64 start = orig_eb->start;
2132 tmp_ebs = calloc(num_stripes, sizeof(*tmp_ebs));
2136 /* Alloc memory in a row for data stripes */
2137 for (i = 0; i < num_stripes; i++) {
2138 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2141 tmp_ebs[i] = calloc(1, sizeof(**tmp_ebs) + stripe_len);
2148 for (i = 0; i < num_stripes; i++) {
2149 struct extent_buffer *eb = tmp_ebs[i];
2151 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2154 eb->start = raid_map[i];
2155 eb->len = stripe_len;
2159 eb->dev_bytenr = (u64)-1;
2161 this_eb_start = raid_map[i];
2163 if (start > this_eb_start ||
2164 start + orig_eb->len < this_eb_start + stripe_len) {
2165 ret = rmw_eb(info, eb, orig_eb);
2169 memcpy(eb->data, orig_eb->data + eb->start - start,
2177 for (i = 0; i < num_stripes; i++)
2183 int write_raid56_with_parity(struct btrfs_fs_info *info,
2184 struct extent_buffer *eb,
2185 struct btrfs_multi_bio *multi,
2186 u64 stripe_len, u64 *raid_map)
2188 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2191 int alloc_size = eb->len;
2194 ebs = malloc(sizeof(*ebs) * multi->num_stripes);
2195 pointers = malloc(sizeof(*pointers) * multi->num_stripes);
2196 if (!ebs || !pointers) {
2202 if (stripe_len > alloc_size)
2203 alloc_size = stripe_len;
2205 ret = split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2206 multi->num_stripes);
2210 for (i = 0; i < multi->num_stripes; i++) {
2211 struct extent_buffer *new_eb;
2212 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2213 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2214 ebs[i]->fd = multi->stripes[i].dev->fd;
2215 multi->stripes[i].dev->total_ios++;
2216 if (ebs[i]->start != raid_map[i]) {
2218 goto out_free_split;
2222 new_eb = malloc(sizeof(*eb) + alloc_size);
2225 goto out_free_split;
2227 new_eb->dev_bytenr = multi->stripes[i].physical;
2228 new_eb->fd = multi->stripes[i].dev->fd;
2229 multi->stripes[i].dev->total_ios++;
2230 new_eb->len = stripe_len;
2232 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2234 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2238 ebs[multi->num_stripes - 2] = p_eb;
2239 ebs[multi->num_stripes - 1] = q_eb;
2241 for (i = 0; i < multi->num_stripes; i++)
2242 pointers[i] = ebs[i]->data;
2244 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2246 ebs[multi->num_stripes - 1] = p_eb;
2247 for (i = 0; i < multi->num_stripes; i++)
2248 pointers[i] = ebs[i]->data;
2249 ret = raid5_gen_result(multi->num_stripes, stripe_len,
2250 multi->num_stripes - 1, pointers);
2252 goto out_free_split;
2255 for (i = 0; i < multi->num_stripes; i++) {
2256 ret = write_extent_to_disk(ebs[i]);
2258 goto out_free_split;
2262 for (i = 0; i < multi->num_stripes; i++) {