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;
165 while (!list_empty(&fs_devices->devices)) {
166 device = list_entry(fs_devices->devices.next,
167 struct btrfs_device, dev_list);
168 if (device->fd != -1) {
170 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
171 fprintf(stderr, "Warning, could not drop caches\n");
175 device->writeable = 0;
176 list_del(&device->dev_list);
177 /* free the memory */
183 seed_devices = fs_devices->seed;
184 fs_devices->seed = NULL;
186 struct btrfs_fs_devices *orig;
189 fs_devices = seed_devices;
190 list_del(&orig->list);
194 list_del(&fs_devices->list);
201 void btrfs_close_all_devices(void)
203 struct btrfs_fs_devices *fs_devices;
205 while (!list_empty(&fs_uuids)) {
206 fs_devices = list_entry(fs_uuids.next, struct btrfs_fs_devices,
208 btrfs_close_devices(fs_devices);
212 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
215 struct list_head *head = &fs_devices->devices;
216 struct list_head *cur;
217 struct btrfs_device *device;
220 list_for_each(cur, head) {
221 device = list_entry(cur, struct btrfs_device, dev_list);
223 printk("no name for device %llu, skip it now\n", device->devid);
227 fd = open(device->name, flags);
233 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
234 fprintf(stderr, "Warning, could not drop caches\n");
236 if (device->devid == fs_devices->latest_devid)
237 fs_devices->latest_bdev = fd;
238 if (device->devid == fs_devices->lowest_devid)
239 fs_devices->lowest_bdev = fd;
242 device->writeable = 1;
246 btrfs_close_devices(fs_devices);
250 int btrfs_scan_one_device(int fd, const char *path,
251 struct btrfs_fs_devices **fs_devices_ret,
252 u64 *total_devs, u64 super_offset, int super_recover)
254 struct btrfs_super_block *disk_super;
255 char buf[BTRFS_SUPER_INFO_SIZE];
259 disk_super = (struct btrfs_super_block *)buf;
260 ret = btrfs_read_dev_super(fd, disk_super, super_offset, super_recover);
263 devid = btrfs_stack_device_id(&disk_super->dev_item);
264 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
267 *total_devs = btrfs_super_num_devices(disk_super);
269 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
275 * this uses a pretty simple search, the expectation is that it is
276 * called very infrequently and that a given device has a small number
279 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
280 struct btrfs_device *device,
281 struct btrfs_path *path,
282 u64 num_bytes, u64 *start)
284 struct btrfs_key key;
285 struct btrfs_root *root = device->dev_root;
286 struct btrfs_dev_extent *dev_extent = NULL;
289 u64 search_start = root->fs_info->alloc_start;
290 u64 search_end = device->total_bytes;
294 struct extent_buffer *l;
299 /* FIXME use last free of some kind */
301 /* we don't want to overwrite the superblock on the drive,
302 * so we make sure to start at an offset of at least 1MB
304 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
306 if (search_start >= search_end) {
311 key.objectid = device->devid;
312 key.offset = search_start;
313 key.type = BTRFS_DEV_EXTENT_KEY;
314 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
317 ret = btrfs_previous_item(root, path, 0, key.type);
321 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
324 slot = path->slots[0];
325 if (slot >= btrfs_header_nritems(l)) {
326 ret = btrfs_next_leaf(root, path);
333 if (search_start >= search_end) {
337 *start = search_start;
341 *start = last_byte > search_start ?
342 last_byte : search_start;
343 if (search_end <= *start) {
349 btrfs_item_key_to_cpu(l, &key, slot);
351 if (key.objectid < device->devid)
354 if (key.objectid > device->devid)
357 if (key.offset >= search_start && key.offset > last_byte &&
359 if (last_byte < search_start)
360 last_byte = search_start;
361 hole_size = key.offset - last_byte;
362 if (key.offset > last_byte &&
363 hole_size >= num_bytes) {
368 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
373 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
374 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
380 /* we have to make sure we didn't find an extent that has already
381 * been allocated by the map tree or the original allocation
383 btrfs_release_path(path);
384 BUG_ON(*start < search_start);
386 if (*start + num_bytes > search_end) {
390 /* check for pending inserts here */
394 btrfs_release_path(path);
398 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
399 struct btrfs_device *device,
400 u64 chunk_tree, u64 chunk_objectid,
402 u64 num_bytes, u64 *start, int convert)
405 struct btrfs_path *path;
406 struct btrfs_root *root = device->dev_root;
407 struct btrfs_dev_extent *extent;
408 struct extent_buffer *leaf;
409 struct btrfs_key key;
411 path = btrfs_alloc_path();
416 * For convert case, just skip search free dev_extent, as caller
417 * is responsible to make sure it's free.
420 ret = find_free_dev_extent(trans, device, path, num_bytes,
426 key.objectid = device->devid;
428 key.type = BTRFS_DEV_EXTENT_KEY;
429 ret = btrfs_insert_empty_item(trans, root, path, &key,
433 leaf = path->nodes[0];
434 extent = btrfs_item_ptr(leaf, path->slots[0],
435 struct btrfs_dev_extent);
436 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
437 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
438 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
440 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
441 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
444 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
445 btrfs_mark_buffer_dirty(leaf);
447 btrfs_free_path(path);
451 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
453 struct btrfs_path *path;
455 struct btrfs_key key;
456 struct btrfs_chunk *chunk;
457 struct btrfs_key found_key;
459 path = btrfs_alloc_path();
462 key.objectid = objectid;
463 key.offset = (u64)-1;
464 key.type = BTRFS_CHUNK_ITEM_KEY;
466 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
472 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
476 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
478 if (found_key.objectid != objectid)
481 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
483 *offset = found_key.offset +
484 btrfs_chunk_length(path->nodes[0], chunk);
489 btrfs_free_path(path);
493 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
497 struct btrfs_key key;
498 struct btrfs_key found_key;
500 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
501 key.type = BTRFS_DEV_ITEM_KEY;
502 key.offset = (u64)-1;
504 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
510 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
515 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
517 *objectid = found_key.offset + 1;
521 btrfs_release_path(path);
526 * the device information is stored in the chunk root
527 * the btrfs_device struct should be fully filled in
529 int btrfs_add_device(struct btrfs_trans_handle *trans,
530 struct btrfs_root *root,
531 struct btrfs_device *device)
534 struct btrfs_path *path;
535 struct btrfs_dev_item *dev_item;
536 struct extent_buffer *leaf;
537 struct btrfs_key key;
541 root = root->fs_info->chunk_root;
543 path = btrfs_alloc_path();
547 ret = find_next_devid(root, path, &free_devid);
551 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
552 key.type = BTRFS_DEV_ITEM_KEY;
553 key.offset = free_devid;
555 ret = btrfs_insert_empty_item(trans, root, path, &key,
560 leaf = path->nodes[0];
561 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
563 device->devid = free_devid;
564 btrfs_set_device_id(leaf, dev_item, device->devid);
565 btrfs_set_device_generation(leaf, dev_item, 0);
566 btrfs_set_device_type(leaf, dev_item, device->type);
567 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
568 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
569 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
570 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
571 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
572 btrfs_set_device_group(leaf, dev_item, 0);
573 btrfs_set_device_seek_speed(leaf, dev_item, 0);
574 btrfs_set_device_bandwidth(leaf, dev_item, 0);
575 btrfs_set_device_start_offset(leaf, dev_item, 0);
577 ptr = (unsigned long)btrfs_device_uuid(dev_item);
578 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
579 ptr = (unsigned long)btrfs_device_fsid(dev_item);
580 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
581 btrfs_mark_buffer_dirty(leaf);
585 btrfs_free_path(path);
589 int btrfs_update_device(struct btrfs_trans_handle *trans,
590 struct btrfs_device *device)
593 struct btrfs_path *path;
594 struct btrfs_root *root;
595 struct btrfs_dev_item *dev_item;
596 struct extent_buffer *leaf;
597 struct btrfs_key key;
599 root = device->dev_root->fs_info->chunk_root;
601 path = btrfs_alloc_path();
605 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
606 key.type = BTRFS_DEV_ITEM_KEY;
607 key.offset = device->devid;
609 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
618 leaf = path->nodes[0];
619 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
621 btrfs_set_device_id(leaf, dev_item, device->devid);
622 btrfs_set_device_type(leaf, dev_item, device->type);
623 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
624 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
625 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
626 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
627 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
628 btrfs_mark_buffer_dirty(leaf);
631 btrfs_free_path(path);
635 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
636 struct btrfs_root *root,
637 struct btrfs_key *key,
638 struct btrfs_chunk *chunk, int item_size)
640 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
641 struct btrfs_disk_key disk_key;
645 array_size = btrfs_super_sys_array_size(super_copy);
646 if (array_size + item_size + sizeof(disk_key)
647 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
650 ptr = super_copy->sys_chunk_array + array_size;
651 btrfs_cpu_key_to_disk(&disk_key, key);
652 memcpy(ptr, &disk_key, sizeof(disk_key));
653 ptr += sizeof(disk_key);
654 memcpy(ptr, chunk, item_size);
655 item_size += sizeof(disk_key);
656 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
660 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
663 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
665 else if (type & BTRFS_BLOCK_GROUP_RAID10)
666 return calc_size * (num_stripes / sub_stripes);
667 else if (type & BTRFS_BLOCK_GROUP_RAID5)
668 return calc_size * (num_stripes - 1);
669 else if (type & BTRFS_BLOCK_GROUP_RAID6)
670 return calc_size * (num_stripes - 2);
672 return calc_size * num_stripes;
676 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
678 /* TODO, add a way to store the preferred stripe size */
679 return BTRFS_STRIPE_LEN;
683 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
685 * It is not equal to "device->total_bytes - device->bytes_used".
686 * We do not allocate any chunk in 1M at beginning of device, and not
687 * allowed to allocate any chunk before alloc_start if it is specified.
688 * So search holes from max(1M, alloc_start) to device->total_bytes.
690 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
691 struct btrfs_device *device,
694 struct btrfs_path *path;
695 struct btrfs_root *root = device->dev_root;
696 struct btrfs_key key;
697 struct btrfs_dev_extent *dev_extent = NULL;
698 struct extent_buffer *l;
699 u64 search_start = root->fs_info->alloc_start;
700 u64 search_end = device->total_bytes;
706 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
708 path = btrfs_alloc_path();
712 key.objectid = device->devid;
713 key.offset = root->fs_info->alloc_start;
714 key.type = BTRFS_DEV_EXTENT_KEY;
717 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
720 ret = btrfs_previous_item(root, path, 0, key.type);
726 slot = path->slots[0];
727 if (slot >= btrfs_header_nritems(l)) {
728 ret = btrfs_next_leaf(root, path);
735 btrfs_item_key_to_cpu(l, &key, slot);
737 if (key.objectid < device->devid)
739 if (key.objectid > device->devid)
741 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
743 if (key.offset > search_end)
745 if (key.offset > search_start)
746 free_bytes += key.offset - search_start;
748 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
749 extent_end = key.offset + btrfs_dev_extent_length(l,
751 if (extent_end > search_start)
752 search_start = extent_end;
753 if (search_start > search_end)
760 if (search_start < search_end)
761 free_bytes += search_end - search_start;
763 *avail_bytes = free_bytes;
766 btrfs_free_path(path);
770 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
771 - sizeof(struct btrfs_item) \
772 - sizeof(struct btrfs_chunk)) \
773 / sizeof(struct btrfs_stripe) + 1)
775 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
776 - 2 * sizeof(struct btrfs_disk_key) \
777 - 2 * sizeof(struct btrfs_chunk)) \
778 / sizeof(struct btrfs_stripe) + 1)
780 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
781 struct btrfs_root *extent_root, u64 *start,
782 u64 *num_bytes, u64 type)
785 struct btrfs_fs_info *info = extent_root->fs_info;
786 struct btrfs_root *chunk_root = info->chunk_root;
787 struct btrfs_stripe *stripes;
788 struct btrfs_device *device = NULL;
789 struct btrfs_chunk *chunk;
790 struct list_head private_devs;
791 struct list_head *dev_list = &info->fs_devices->devices;
792 struct list_head *cur;
793 struct map_lookup *map;
794 int min_stripe_size = 1 * 1024 * 1024;
795 u64 calc_size = 8 * 1024 * 1024;
797 u64 max_chunk_size = 4 * calc_size;
808 int stripe_len = BTRFS_STRIPE_LEN;
809 struct btrfs_key key;
812 if (list_empty(dev_list)) {
816 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
817 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
818 BTRFS_BLOCK_GROUP_RAID10 |
819 BTRFS_BLOCK_GROUP_DUP)) {
820 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
821 calc_size = 8 * 1024 * 1024;
822 max_chunk_size = calc_size * 2;
823 min_stripe_size = 1 * 1024 * 1024;
824 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
825 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
826 calc_size = 1024 * 1024 * 1024;
827 max_chunk_size = 10 * calc_size;
828 min_stripe_size = 64 * 1024 * 1024;
829 max_stripes = BTRFS_MAX_DEVS(chunk_root);
830 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
831 calc_size = 1024 * 1024 * 1024;
832 max_chunk_size = 4 * calc_size;
833 min_stripe_size = 32 * 1024 * 1024;
834 max_stripes = BTRFS_MAX_DEVS(chunk_root);
837 if (type & BTRFS_BLOCK_GROUP_RAID1) {
838 num_stripes = min_t(u64, 2,
839 btrfs_super_num_devices(info->super_copy));
844 if (type & BTRFS_BLOCK_GROUP_DUP) {
848 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
849 num_stripes = btrfs_super_num_devices(info->super_copy);
850 if (num_stripes > max_stripes)
851 num_stripes = max_stripes;
854 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
855 num_stripes = btrfs_super_num_devices(info->super_copy);
856 if (num_stripes > max_stripes)
857 num_stripes = max_stripes;
860 num_stripes &= ~(u32)1;
864 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
865 num_stripes = btrfs_super_num_devices(info->super_copy);
866 if (num_stripes > max_stripes)
867 num_stripes = max_stripes;
871 stripe_len = find_raid56_stripe_len(num_stripes - 1,
872 btrfs_super_stripesize(info->super_copy));
874 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
875 num_stripes = btrfs_super_num_devices(info->super_copy);
876 if (num_stripes > max_stripes)
877 num_stripes = max_stripes;
881 stripe_len = find_raid56_stripe_len(num_stripes - 2,
882 btrfs_super_stripesize(info->super_copy));
885 /* we don't want a chunk larger than 10% of the FS */
886 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
887 max_chunk_size = min(percent_max, max_chunk_size);
890 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
892 calc_size = max_chunk_size;
893 calc_size /= num_stripes;
894 calc_size /= stripe_len;
895 calc_size *= stripe_len;
897 /* we don't want tiny stripes */
898 calc_size = max_t(u64, calc_size, min_stripe_size);
900 calc_size /= stripe_len;
901 calc_size *= stripe_len;
902 INIT_LIST_HEAD(&private_devs);
903 cur = dev_list->next;
906 if (type & BTRFS_BLOCK_GROUP_DUP)
907 min_free = calc_size * 2;
909 min_free = calc_size;
911 /* build a private list of devices we will allocate from */
912 while(index < num_stripes) {
913 device = list_entry(cur, struct btrfs_device, dev_list);
914 ret = btrfs_device_avail_bytes(trans, device, &avail);
918 if (avail >= min_free) {
919 list_move_tail(&device->dev_list, &private_devs);
921 if (type & BTRFS_BLOCK_GROUP_DUP)
923 } else if (avail > max_avail)
928 if (index < num_stripes) {
929 list_splice(&private_devs, dev_list);
930 if (index >= min_stripes) {
932 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
933 num_stripes /= sub_stripes;
934 num_stripes *= sub_stripes;
939 if (!looped && max_avail > 0) {
941 calc_size = max_avail;
946 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
950 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
951 key.type = BTRFS_CHUNK_ITEM_KEY;
954 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
958 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
964 stripes = &chunk->stripe;
965 *num_bytes = chunk_bytes_by_type(type, calc_size,
966 num_stripes, sub_stripes);
968 while(index < num_stripes) {
969 struct btrfs_stripe *stripe;
970 BUG_ON(list_empty(&private_devs));
971 cur = private_devs.next;
972 device = list_entry(cur, struct btrfs_device, dev_list);
974 /* loop over this device again if we're doing a dup group */
975 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
976 (index == num_stripes - 1))
977 list_move_tail(&device->dev_list, dev_list);
979 ret = btrfs_alloc_dev_extent(trans, device,
980 info->chunk_root->root_key.objectid,
981 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
982 calc_size, &dev_offset, 0);
985 device->bytes_used += calc_size;
986 ret = btrfs_update_device(trans, device);
989 map->stripes[index].dev = device;
990 map->stripes[index].physical = dev_offset;
991 stripe = stripes + index;
992 btrfs_set_stack_stripe_devid(stripe, device->devid);
993 btrfs_set_stack_stripe_offset(stripe, dev_offset);
994 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
997 BUG_ON(!list_empty(&private_devs));
999 /* key was set above */
1000 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1001 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1002 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1003 btrfs_set_stack_chunk_type(chunk, type);
1004 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1005 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1006 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1007 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1008 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1009 map->sector_size = extent_root->sectorsize;
1010 map->stripe_len = stripe_len;
1011 map->io_align = stripe_len;
1012 map->io_width = stripe_len;
1014 map->num_stripes = num_stripes;
1015 map->sub_stripes = sub_stripes;
1017 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1018 btrfs_chunk_item_size(num_stripes));
1020 *start = key.offset;;
1022 map->ce.start = key.offset;
1023 map->ce.size = *num_bytes;
1025 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1028 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1029 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1030 chunk, btrfs_chunk_item_size(num_stripes));
1039 * Alloc a DATA chunk with SINGLE profile.
1041 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1042 * (btrfs logical bytenr == on-disk bytenr)
1043 * For that case, caller must make sure the chunk and dev_extent are not
1046 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1047 struct btrfs_root *extent_root, u64 *start,
1048 u64 num_bytes, u64 type, int convert)
1051 struct btrfs_fs_info *info = extent_root->fs_info;
1052 struct btrfs_root *chunk_root = info->chunk_root;
1053 struct btrfs_stripe *stripes;
1054 struct btrfs_device *device = NULL;
1055 struct btrfs_chunk *chunk;
1056 struct list_head *dev_list = &info->fs_devices->devices;
1057 struct list_head *cur;
1058 struct map_lookup *map;
1059 u64 calc_size = 8 * 1024 * 1024;
1060 int num_stripes = 1;
1061 int sub_stripes = 0;
1064 int stripe_len = BTRFS_STRIPE_LEN;
1065 struct btrfs_key key;
1067 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1068 key.type = BTRFS_CHUNK_ITEM_KEY;
1070 BUG_ON(*start != round_down(*start, extent_root->sectorsize));
1071 key.offset = *start;
1072 dev_offset = *start;
1074 ret = find_next_chunk(chunk_root,
1075 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1081 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1085 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1091 stripes = &chunk->stripe;
1092 calc_size = num_bytes;
1095 cur = dev_list->next;
1096 device = list_entry(cur, struct btrfs_device, dev_list);
1098 while (index < num_stripes) {
1099 struct btrfs_stripe *stripe;
1101 ret = btrfs_alloc_dev_extent(trans, device,
1102 info->chunk_root->root_key.objectid,
1103 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1104 calc_size, &dev_offset, convert);
1107 device->bytes_used += calc_size;
1108 ret = btrfs_update_device(trans, device);
1111 map->stripes[index].dev = device;
1112 map->stripes[index].physical = dev_offset;
1113 stripe = stripes + index;
1114 btrfs_set_stack_stripe_devid(stripe, device->devid);
1115 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1116 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1120 /* key was set above */
1121 btrfs_set_stack_chunk_length(chunk, num_bytes);
1122 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1123 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1124 btrfs_set_stack_chunk_type(chunk, type);
1125 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1126 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1127 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1128 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1129 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1130 map->sector_size = extent_root->sectorsize;
1131 map->stripe_len = stripe_len;
1132 map->io_align = stripe_len;
1133 map->io_width = stripe_len;
1135 map->num_stripes = num_stripes;
1136 map->sub_stripes = sub_stripes;
1138 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1139 btrfs_chunk_item_size(num_stripes));
1142 *start = key.offset;
1144 map->ce.start = key.offset;
1145 map->ce.size = num_bytes;
1147 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1154 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1156 struct cache_extent *ce;
1157 struct map_lookup *map;
1160 ce = search_cache_extent(&map_tree->cache_tree, logical);
1162 fprintf(stderr, "No mapping for %llu-%llu\n",
1163 (unsigned long long)logical,
1164 (unsigned long long)logical+len);
1167 if (ce->start > logical || ce->start + ce->size < logical) {
1168 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1169 "%llu-%llu\n", (unsigned long long)logical,
1170 (unsigned long long)logical+len,
1171 (unsigned long long)ce->start,
1172 (unsigned long long)ce->start + ce->size);
1175 map = container_of(ce, struct map_lookup, ce);
1177 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1178 ret = map->num_stripes;
1179 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1180 ret = map->sub_stripes;
1181 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1183 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1190 int btrfs_next_bg(struct btrfs_mapping_tree *map_tree, u64 *logical,
1191 u64 *size, u64 type)
1193 struct cache_extent *ce;
1194 struct map_lookup *map;
1197 ce = search_cache_extent(&map_tree->cache_tree, cur);
1201 * only jump to next bg if our cur is not 0
1202 * As the initial logical for btrfs_next_bg() is 0, and
1203 * if we jump to next bg, we skipped a valid bg.
1206 ce = next_cache_extent(ce);
1212 map = container_of(ce, struct map_lookup, ce);
1213 if (map->type & type) {
1214 *logical = ce->start;
1223 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1224 u64 chunk_start, u64 physical, u64 devid,
1225 u64 **logical, int *naddrs, int *stripe_len)
1227 struct cache_extent *ce;
1228 struct map_lookup *map;
1236 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1238 map = container_of(ce, struct map_lookup, ce);
1241 rmap_len = map->stripe_len;
1242 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1243 length = ce->size / (map->num_stripes / map->sub_stripes);
1244 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1245 length = ce->size / map->num_stripes;
1246 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1247 BTRFS_BLOCK_GROUP_RAID6)) {
1248 length = ce->size / nr_data_stripes(map);
1249 rmap_len = map->stripe_len * nr_data_stripes(map);
1252 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1254 for (i = 0; i < map->num_stripes; i++) {
1255 if (devid && map->stripes[i].dev->devid != devid)
1257 if (map->stripes[i].physical > physical ||
1258 map->stripes[i].physical + length <= physical)
1261 stripe_nr = (physical - map->stripes[i].physical) /
1264 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1265 stripe_nr = (stripe_nr * map->num_stripes + i) /
1267 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1268 stripe_nr = stripe_nr * map->num_stripes + i;
1269 } /* else if RAID[56], multiply by nr_data_stripes().
1270 * Alternatively, just use rmap_len below instead of
1271 * map->stripe_len */
1273 bytenr = ce->start + stripe_nr * rmap_len;
1274 for (j = 0; j < nr; j++) {
1275 if (buf[j] == bytenr)
1284 *stripe_len = rmap_len;
1289 static inline int parity_smaller(u64 a, u64 b)
1294 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1295 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1297 struct btrfs_bio_stripe s;
1304 for (i = 0; i < bbio->num_stripes - 1; i++) {
1305 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1306 s = bbio->stripes[i];
1308 bbio->stripes[i] = bbio->stripes[i+1];
1309 raid_map[i] = raid_map[i+1];
1310 bbio->stripes[i+1] = s;
1318 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1319 u64 logical, u64 *length,
1320 struct btrfs_multi_bio **multi_ret, int mirror_num,
1323 return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1324 multi_ret, mirror_num, raid_map_ret);
1327 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1328 u64 logical, u64 *length, u64 *type,
1329 struct btrfs_multi_bio **multi_ret, int mirror_num,
1332 struct cache_extent *ce;
1333 struct map_lookup *map;
1337 u64 *raid_map = NULL;
1338 int stripes_allocated = 8;
1339 int stripes_required = 1;
1342 struct btrfs_multi_bio *multi = NULL;
1344 if (multi_ret && rw == READ) {
1345 stripes_allocated = 1;
1348 ce = search_cache_extent(&map_tree->cache_tree, logical);
1354 if (ce->start > logical) {
1356 *length = ce->start - logical;
1361 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1366 map = container_of(ce, struct map_lookup, ce);
1367 offset = logical - ce->start;
1370 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1371 BTRFS_BLOCK_GROUP_DUP)) {
1372 stripes_required = map->num_stripes;
1373 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1374 stripes_required = map->sub_stripes;
1377 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1378 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1379 /* RAID[56] write or recovery. Return all stripes */
1380 stripes_required = map->num_stripes;
1382 /* Only allocate the map if we've already got a large enough multi_ret */
1383 if (stripes_allocated >= stripes_required) {
1384 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1392 /* if our multi bio struct is too small, back off and try again */
1393 if (multi_ret && stripes_allocated < stripes_required) {
1394 stripes_allocated = stripes_required;
1401 * stripe_nr counts the total number of stripes we have to stride
1402 * to get to this block
1404 stripe_nr = stripe_nr / map->stripe_len;
1406 stripe_offset = stripe_nr * map->stripe_len;
1407 BUG_ON(offset < stripe_offset);
1409 /* stripe_offset is the offset of this block in its stripe*/
1410 stripe_offset = offset - stripe_offset;
1412 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1413 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1414 BTRFS_BLOCK_GROUP_RAID10 |
1415 BTRFS_BLOCK_GROUP_DUP)) {
1416 /* we limit the length of each bio to what fits in a stripe */
1417 *length = min_t(u64, ce->size - offset,
1418 map->stripe_len - stripe_offset);
1420 *length = ce->size - offset;
1426 multi->num_stripes = 1;
1428 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1430 multi->num_stripes = map->num_stripes;
1431 else if (mirror_num)
1432 stripe_index = mirror_num - 1;
1434 stripe_index = stripe_nr % map->num_stripes;
1435 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1436 int factor = map->num_stripes / map->sub_stripes;
1438 stripe_index = stripe_nr % factor;
1439 stripe_index *= map->sub_stripes;
1442 multi->num_stripes = map->sub_stripes;
1443 else if (mirror_num)
1444 stripe_index += mirror_num - 1;
1446 stripe_nr = stripe_nr / factor;
1447 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1449 multi->num_stripes = map->num_stripes;
1450 else if (mirror_num)
1451 stripe_index = mirror_num - 1;
1452 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1453 BTRFS_BLOCK_GROUP_RAID6)) {
1458 u64 raid56_full_stripe_start;
1459 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1462 * align the start of our data stripe in the logical
1465 raid56_full_stripe_start = offset / full_stripe_len;
1466 raid56_full_stripe_start *= full_stripe_len;
1468 /* get the data stripe number */
1469 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1470 stripe_nr = stripe_nr / nr_data_stripes(map);
1472 /* Work out the disk rotation on this stripe-set */
1473 rot = stripe_nr % map->num_stripes;
1475 /* Fill in the logical address of each stripe */
1476 tmp = stripe_nr * nr_data_stripes(map);
1478 for (i = 0; i < nr_data_stripes(map); i++)
1479 raid_map[(i+rot) % map->num_stripes] =
1480 ce->start + (tmp + i) * map->stripe_len;
1482 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1483 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1484 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1486 *length = map->stripe_len;
1489 multi->num_stripes = map->num_stripes;
1491 stripe_index = stripe_nr % nr_data_stripes(map);
1492 stripe_nr = stripe_nr / nr_data_stripes(map);
1495 * Mirror #0 or #1 means the original data block.
1496 * Mirror #2 is RAID5 parity block.
1497 * Mirror #3 is RAID6 Q block.
1500 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1502 /* We distribute the parity blocks across stripes */
1503 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1507 * after this do_div call, stripe_nr is the number of stripes
1508 * on this device we have to walk to find the data, and
1509 * stripe_index is the number of our device in the stripe array
1511 stripe_index = stripe_nr % map->num_stripes;
1512 stripe_nr = stripe_nr / map->num_stripes;
1514 BUG_ON(stripe_index >= map->num_stripes);
1516 for (i = 0; i < multi->num_stripes; i++) {
1517 multi->stripes[i].physical =
1518 map->stripes[stripe_index].physical + stripe_offset +
1519 stripe_nr * map->stripe_len;
1520 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1529 sort_parity_stripes(multi, raid_map);
1530 *raid_map_ret = raid_map;
1536 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1539 struct btrfs_device *device;
1540 struct btrfs_fs_devices *cur_devices;
1542 cur_devices = root->fs_info->fs_devices;
1543 while (cur_devices) {
1545 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1546 root->fs_info->ignore_fsid_mismatch)) {
1547 device = __find_device(&cur_devices->devices,
1552 cur_devices = cur_devices->seed;
1557 struct btrfs_device *
1558 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1559 u64 devid, int instance)
1561 struct list_head *head = &fs_devices->devices;
1562 struct btrfs_device *dev;
1565 list_for_each_entry(dev, head, dev_list) {
1566 if (dev->devid == devid && num_found++ == instance)
1572 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1574 struct cache_extent *ce;
1575 struct map_lookup *map;
1576 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1581 * During chunk recovering, we may fail to find block group's
1582 * corresponding chunk, we will rebuild it later
1584 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1585 if (!root->fs_info->is_chunk_recover)
1590 map = container_of(ce, struct map_lookup, ce);
1591 for (i = 0; i < map->num_stripes; i++) {
1592 if (!map->stripes[i].dev->writeable) {
1601 static struct btrfs_device *fill_missing_device(u64 devid)
1603 struct btrfs_device *device;
1605 device = kzalloc(sizeof(*device), GFP_NOFS);
1606 device->devid = devid;
1612 * slot == -1: SYSTEM chunk
1613 * return -EIO on error, otherwise return 0
1615 static int btrfs_check_chunk_valid(struct btrfs_root *root,
1616 struct extent_buffer *leaf,
1617 struct btrfs_chunk *chunk,
1618 int slot, u64 logical)
1626 length = btrfs_chunk_length(leaf, chunk);
1627 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1628 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1629 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1630 type = btrfs_chunk_type(leaf, chunk);
1633 * These valid checks may be insufficient to cover every corner cases.
1635 if (!IS_ALIGNED(logical, root->sectorsize)) {
1636 error("invalid chunk logical %llu", logical);
1639 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
1640 error("invalid chunk sectorsize %llu",
1641 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1644 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
1645 error("invalid chunk length %llu", length);
1648 if (stripe_len != BTRFS_STRIPE_LEN) {
1649 error("invalid chunk stripe length: %llu", stripe_len);
1652 /* Check on chunk item type */
1653 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1654 error("invalid chunk type %llu", type);
1657 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1658 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1659 error("unrecognized chunk type: %llu",
1660 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1661 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1665 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1666 * it can't exceed the system chunk array size
1667 * For normal chunk, it should match its chunk item size.
1669 if (num_stripes < 1 ||
1670 (slot == -1 && sizeof(struct btrfs_stripe) * num_stripes >
1671 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1672 (slot >= 0 && sizeof(struct btrfs_stripe) * (num_stripes - 1) >
1673 btrfs_item_size_nr(leaf, slot))) {
1674 error("invalid num_stripes: %u", num_stripes);
1678 * Device number check against profile
1680 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes == 0) ||
1681 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1682 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1683 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1684 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1685 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1686 num_stripes != 1)) {
1687 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1688 num_stripes, sub_stripes,
1689 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1697 * Slot is used to verify the chunk item is valid
1699 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1701 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1702 struct extent_buffer *leaf,
1703 struct btrfs_chunk *chunk, int slot)
1705 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1706 struct map_lookup *map;
1707 struct cache_extent *ce;
1711 u8 uuid[BTRFS_UUID_SIZE];
1716 logical = key->offset;
1717 length = btrfs_chunk_length(leaf, chunk);
1718 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1719 /* Validation check */
1720 ret = btrfs_check_chunk_valid(root, leaf, chunk, slot, logical);
1722 error("%s checksums match, but it has an invalid chunk, %s",
1723 (slot == -1) ? "Superblock" : "Metadata",
1724 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1728 ce = search_cache_extent(&map_tree->cache_tree, logical);
1730 /* already mapped? */
1731 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1735 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1739 map->ce.start = logical;
1740 map->ce.size = length;
1741 map->num_stripes = num_stripes;
1742 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1743 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1744 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1745 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1746 map->type = btrfs_chunk_type(leaf, chunk);
1747 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1749 for (i = 0; i < num_stripes; i++) {
1750 map->stripes[i].physical =
1751 btrfs_stripe_offset_nr(leaf, chunk, i);
1752 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1753 read_extent_buffer(leaf, uuid, (unsigned long)
1754 btrfs_stripe_dev_uuid_nr(chunk, i),
1756 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1758 if (!map->stripes[i].dev) {
1759 map->stripes[i].dev = fill_missing_device(devid);
1760 printf("warning, device %llu is missing\n",
1761 (unsigned long long)devid);
1765 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1771 static int fill_device_from_item(struct extent_buffer *leaf,
1772 struct btrfs_dev_item *dev_item,
1773 struct btrfs_device *device)
1777 device->devid = btrfs_device_id(leaf, dev_item);
1778 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1779 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1780 device->type = btrfs_device_type(leaf, dev_item);
1781 device->io_align = btrfs_device_io_align(leaf, dev_item);
1782 device->io_width = btrfs_device_io_width(leaf, dev_item);
1783 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1785 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1786 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1791 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1793 struct btrfs_fs_devices *fs_devices;
1796 fs_devices = root->fs_info->fs_devices->seed;
1797 while (fs_devices) {
1798 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1802 fs_devices = fs_devices->seed;
1805 fs_devices = find_fsid(fsid);
1807 /* missing all seed devices */
1808 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1813 INIT_LIST_HEAD(&fs_devices->devices);
1814 list_add(&fs_devices->list, &fs_uuids);
1815 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1818 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1822 fs_devices->seed = root->fs_info->fs_devices->seed;
1823 root->fs_info->fs_devices->seed = fs_devices;
1828 static int read_one_dev(struct btrfs_root *root,
1829 struct extent_buffer *leaf,
1830 struct btrfs_dev_item *dev_item)
1832 struct btrfs_device *device;
1835 u8 fs_uuid[BTRFS_UUID_SIZE];
1836 u8 dev_uuid[BTRFS_UUID_SIZE];
1838 devid = btrfs_device_id(leaf, dev_item);
1839 read_extent_buffer(leaf, dev_uuid,
1840 (unsigned long)btrfs_device_uuid(dev_item),
1842 read_extent_buffer(leaf, fs_uuid,
1843 (unsigned long)btrfs_device_fsid(dev_item),
1846 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1847 ret = open_seed_devices(root, fs_uuid);
1852 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1854 device = kzalloc(sizeof(*device), GFP_NOFS);
1858 list_add(&device->dev_list,
1859 &root->fs_info->fs_devices->devices);
1862 fill_device_from_item(leaf, dev_item, device);
1863 device->dev_root = root->fs_info->dev_root;
1867 int btrfs_read_sys_array(struct btrfs_root *root)
1869 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1870 struct extent_buffer *sb;
1871 struct btrfs_disk_key *disk_key;
1872 struct btrfs_chunk *chunk;
1874 unsigned long sb_array_offset;
1880 struct btrfs_key key;
1882 sb = btrfs_find_create_tree_block(root->fs_info,
1883 BTRFS_SUPER_INFO_OFFSET,
1884 BTRFS_SUPER_INFO_SIZE);
1887 btrfs_set_buffer_uptodate(sb);
1888 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1889 array_size = btrfs_super_sys_array_size(super_copy);
1891 array_ptr = super_copy->sys_chunk_array;
1892 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1895 while (cur_offset < array_size) {
1896 disk_key = (struct btrfs_disk_key *)array_ptr;
1897 len = sizeof(*disk_key);
1898 if (cur_offset + len > array_size)
1899 goto out_short_read;
1901 btrfs_disk_key_to_cpu(&key, disk_key);
1904 sb_array_offset += len;
1907 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1908 chunk = (struct btrfs_chunk *)sb_array_offset;
1910 * At least one btrfs_chunk with one stripe must be
1911 * present, exact stripe count check comes afterwards
1913 len = btrfs_chunk_item_size(1);
1914 if (cur_offset + len > array_size)
1915 goto out_short_read;
1917 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1920 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
1921 num_stripes, cur_offset);
1926 len = btrfs_chunk_item_size(num_stripes);
1927 if (cur_offset + len > array_size)
1928 goto out_short_read;
1930 ret = read_one_chunk(root, &key, sb, chunk, -1);
1935 "ERROR: unexpected item type %u in sys_array at offset %u\n",
1936 (u32)key.type, cur_offset);
1941 sb_array_offset += len;
1944 free_extent_buffer(sb);
1948 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
1950 free_extent_buffer(sb);
1954 int btrfs_read_chunk_tree(struct btrfs_root *root)
1956 struct btrfs_path *path;
1957 struct extent_buffer *leaf;
1958 struct btrfs_key key;
1959 struct btrfs_key found_key;
1963 root = root->fs_info->chunk_root;
1965 path = btrfs_alloc_path();
1970 * Read all device items, and then all the chunk items. All
1971 * device items are found before any chunk item (their object id
1972 * is smaller than the lowest possible object id for a chunk
1973 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
1975 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1978 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1982 leaf = path->nodes[0];
1983 slot = path->slots[0];
1984 if (slot >= btrfs_header_nritems(leaf)) {
1985 ret = btrfs_next_leaf(root, path);
1992 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1993 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1994 struct btrfs_dev_item *dev_item;
1995 dev_item = btrfs_item_ptr(leaf, slot,
1996 struct btrfs_dev_item);
1997 ret = read_one_dev(root, leaf, dev_item);
1999 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2000 struct btrfs_chunk *chunk;
2001 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2002 ret = read_one_chunk(root, &found_key, leaf, chunk,
2011 btrfs_free_path(path);
2015 struct list_head *btrfs_scanned_uuids(void)
2020 static int rmw_eb(struct btrfs_fs_info *info,
2021 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2024 unsigned long orig_off = 0;
2025 unsigned long dest_off = 0;
2026 unsigned long copy_len = eb->len;
2028 ret = read_whole_eb(info, eb, 0);
2032 if (eb->start + eb->len <= orig_eb->start ||
2033 eb->start >= orig_eb->start + orig_eb->len)
2036 * | ----- orig_eb ------- |
2037 * | ----- stripe ------- |
2038 * | ----- orig_eb ------- |
2039 * | ----- orig_eb ------- |
2041 if (eb->start > orig_eb->start)
2042 orig_off = eb->start - orig_eb->start;
2043 if (orig_eb->start > eb->start)
2044 dest_off = orig_eb->start - eb->start;
2046 if (copy_len > orig_eb->len - orig_off)
2047 copy_len = orig_eb->len - orig_off;
2048 if (copy_len > eb->len - dest_off)
2049 copy_len = eb->len - dest_off;
2051 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2055 static void split_eb_for_raid56(struct btrfs_fs_info *info,
2056 struct extent_buffer *orig_eb,
2057 struct extent_buffer **ebs,
2058 u64 stripe_len, u64 *raid_map,
2061 struct extent_buffer *eb;
2062 u64 start = orig_eb->start;
2067 for (i = 0; i < num_stripes; i++) {
2068 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2071 eb = calloc(1, sizeof(struct extent_buffer) + stripe_len);
2075 eb->start = raid_map[i];
2076 eb->len = stripe_len;
2080 eb->dev_bytenr = (u64)-1;
2082 this_eb_start = raid_map[i];
2084 if (start > this_eb_start ||
2085 start + orig_eb->len < this_eb_start + stripe_len) {
2086 ret = rmw_eb(info, eb, orig_eb);
2089 memcpy(eb->data, orig_eb->data + eb->start - start, stripe_len);
2095 int write_raid56_with_parity(struct btrfs_fs_info *info,
2096 struct extent_buffer *eb,
2097 struct btrfs_multi_bio *multi,
2098 u64 stripe_len, u64 *raid_map)
2100 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2104 int alloc_size = eb->len;
2106 ebs = kmalloc(sizeof(*ebs) * multi->num_stripes, GFP_NOFS);
2109 if (stripe_len > alloc_size)
2110 alloc_size = stripe_len;
2112 split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2113 multi->num_stripes);
2115 for (i = 0; i < multi->num_stripes; i++) {
2116 struct extent_buffer *new_eb;
2117 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2118 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2119 ebs[i]->fd = multi->stripes[i].dev->fd;
2120 multi->stripes[i].dev->total_ios++;
2121 BUG_ON(ebs[i]->start != raid_map[i]);
2124 new_eb = kmalloc(sizeof(*eb) + alloc_size, GFP_NOFS);
2126 new_eb->dev_bytenr = multi->stripes[i].physical;
2127 new_eb->fd = multi->stripes[i].dev->fd;
2128 multi->stripes[i].dev->total_ios++;
2129 new_eb->len = stripe_len;
2131 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2133 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2139 pointers = kmalloc(sizeof(*pointers) * multi->num_stripes,
2143 ebs[multi->num_stripes - 2] = p_eb;
2144 ebs[multi->num_stripes - 1] = q_eb;
2146 for (i = 0; i < multi->num_stripes; i++)
2147 pointers[i] = ebs[i]->data;
2149 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2152 ebs[multi->num_stripes - 1] = p_eb;
2153 memcpy(p_eb->data, ebs[0]->data, stripe_len);
2154 for (j = 1; j < multi->num_stripes - 1; j++) {
2155 for (i = 0; i < stripe_len; i += sizeof(unsigned long)) {
2156 *(unsigned long *)(p_eb->data + i) ^=
2157 *(unsigned long *)(ebs[j]->data + i);
2162 for (i = 0; i < multi->num_stripes; i++) {
2163 ret = write_extent_to_disk(ebs[i]);