2 * Copyright (C) 2013 FUJITSU LIMITED. 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.
19 #include "kerncompat.h"
20 #include "androidcompat.h"
23 #include <stdio_ext.h>
25 #include <sys/types.h>
29 #include <uuid/uuid.h>
33 #include "radix-tree.h"
35 #include "extent-cache.h"
38 #include "transaction.h"
44 struct recover_control
{
52 u64 chunk_root_generation
;
54 struct btrfs_fs_devices
*fs_devices
;
56 struct cache_tree chunk
;
57 struct block_group_tree bg
;
58 struct device_extent_tree devext
;
59 struct cache_tree eb_cache
;
61 struct list_head good_chunks
;
62 struct list_head bad_chunks
;
63 struct list_head rebuild_chunks
;
64 struct list_head unrepaired_chunks
;
65 pthread_mutex_t rc_lock
;
68 struct extent_record
{
69 struct cache_extent cache
;
71 u8 csum
[BTRFS_CSUM_SIZE
];
72 struct btrfs_device
*devices
[BTRFS_MAX_MIRRORS
];
73 u64 offsets
[BTRFS_MAX_MIRRORS
];
78 struct recover_control
*rc
;
79 struct btrfs_device
*dev
;
84 static struct extent_record
*btrfs_new_extent_record(struct extent_buffer
*eb
)
86 struct extent_record
*rec
;
88 rec
= calloc(1, sizeof(*rec
));
90 fprintf(stderr
, "Fail to allocate memory for extent record.\n");
94 rec
->cache
.start
= btrfs_header_bytenr(eb
);
95 rec
->cache
.size
= eb
->len
;
96 rec
->generation
= btrfs_header_generation(eb
);
97 read_extent_buffer(eb
, rec
->csum
, (unsigned long)btrfs_header_csum(eb
),
102 static int process_extent_buffer(struct cache_tree
*eb_cache
,
103 struct extent_buffer
*eb
,
104 struct btrfs_device
*device
, u64 offset
)
106 struct extent_record
*rec
;
107 struct extent_record
*exist
;
108 struct cache_extent
*cache
;
111 rec
= btrfs_new_extent_record(eb
);
112 if (!rec
->cache
.size
)
115 cache
= lookup_cache_extent(eb_cache
,
119 exist
= container_of(cache
, struct extent_record
, cache
);
121 if (exist
->generation
> rec
->generation
)
123 if (exist
->generation
== rec
->generation
) {
124 if (exist
->cache
.start
!= rec
->cache
.start
||
125 exist
->cache
.size
!= rec
->cache
.size
||
126 memcmp(exist
->csum
, rec
->csum
, BTRFS_CSUM_SIZE
)) {
129 BUG_ON(exist
->nmirrors
>= BTRFS_MAX_MIRRORS
);
130 exist
->devices
[exist
->nmirrors
] = device
;
131 exist
->offsets
[exist
->nmirrors
] = offset
;
136 remove_cache_extent(eb_cache
, cache
);
141 rec
->devices
[0] = device
;
142 rec
->offsets
[0] = offset
;
144 ret
= insert_cache_extent(eb_cache
, &rec
->cache
);
153 static void free_extent_record(struct cache_extent
*cache
)
155 struct extent_record
*er
;
157 er
= container_of(cache
, struct extent_record
, cache
);
161 FREE_EXTENT_CACHE_BASED_TREE(extent_record
, free_extent_record
);
163 static struct btrfs_chunk
*create_chunk_item(struct chunk_record
*record
)
165 struct btrfs_chunk
*ret
;
166 struct btrfs_stripe
*chunk_stripe
;
169 if (!record
|| record
->num_stripes
== 0)
171 ret
= malloc(btrfs_chunk_item_size(record
->num_stripes
));
174 btrfs_set_stack_chunk_length(ret
, record
->length
);
175 btrfs_set_stack_chunk_owner(ret
, record
->owner
);
176 btrfs_set_stack_chunk_stripe_len(ret
, record
->stripe_len
);
177 btrfs_set_stack_chunk_type(ret
, record
->type_flags
);
178 btrfs_set_stack_chunk_io_align(ret
, record
->io_align
);
179 btrfs_set_stack_chunk_io_width(ret
, record
->io_width
);
180 btrfs_set_stack_chunk_sector_size(ret
, record
->sector_size
);
181 btrfs_set_stack_chunk_num_stripes(ret
, record
->num_stripes
);
182 btrfs_set_stack_chunk_sub_stripes(ret
, record
->sub_stripes
);
183 for (i
= 0, chunk_stripe
= &ret
->stripe
; i
< record
->num_stripes
;
184 i
++, chunk_stripe
++) {
185 btrfs_set_stack_stripe_devid(chunk_stripe
,
186 record
->stripes
[i
].devid
);
187 btrfs_set_stack_stripe_offset(chunk_stripe
,
188 record
->stripes
[i
].offset
);
189 memcpy(chunk_stripe
->dev_uuid
, record
->stripes
[i
].dev_uuid
,
195 static void init_recover_control(struct recover_control
*rc
, int verbose
,
198 memset(rc
, 0, sizeof(struct recover_control
));
199 cache_tree_init(&rc
->chunk
);
200 cache_tree_init(&rc
->eb_cache
);
201 block_group_tree_init(&rc
->bg
);
202 device_extent_tree_init(&rc
->devext
);
204 INIT_LIST_HEAD(&rc
->good_chunks
);
205 INIT_LIST_HEAD(&rc
->bad_chunks
);
206 INIT_LIST_HEAD(&rc
->rebuild_chunks
);
207 INIT_LIST_HEAD(&rc
->unrepaired_chunks
);
209 rc
->verbose
= verbose
;
211 pthread_mutex_init(&rc
->rc_lock
, NULL
);
214 static void free_recover_control(struct recover_control
*rc
)
216 free_block_group_tree(&rc
->bg
);
217 free_chunk_cache_tree(&rc
->chunk
);
218 free_device_extent_tree(&rc
->devext
);
219 free_extent_record_tree(&rc
->eb_cache
);
220 pthread_mutex_destroy(&rc
->rc_lock
);
223 static int process_block_group_item(struct block_group_tree
*bg_cache
,
224 struct extent_buffer
*leaf
,
225 struct btrfs_key
*key
, int slot
)
227 struct block_group_record
*rec
;
228 struct block_group_record
*exist
;
229 struct cache_extent
*cache
;
232 rec
= btrfs_new_block_group_record(leaf
, key
, slot
);
233 if (!rec
->cache
.size
)
236 cache
= lookup_cache_extent(&bg_cache
->tree
,
240 exist
= container_of(cache
, struct block_group_record
, cache
);
242 /*check the generation and replace if needed*/
243 if (exist
->generation
> rec
->generation
)
245 if (exist
->generation
== rec
->generation
) {
246 int offset
= offsetof(struct block_group_record
,
249 * According to the current kernel code, the following
250 * case is impossible, or there is something wrong in
253 if (memcmp(((void *)exist
) + offset
,
254 ((void *)rec
) + offset
,
255 sizeof(*rec
) - offset
))
259 remove_cache_extent(&bg_cache
->tree
, cache
);
260 list_del_init(&exist
->list
);
263 * We must do search again to avoid the following cache.
264 * /--old bg 1--//--old bg 2--/
270 ret
= insert_block_group_record(bg_cache
, rec
);
279 static int process_chunk_item(struct cache_tree
*chunk_cache
,
280 struct extent_buffer
*leaf
, struct btrfs_key
*key
,
283 struct chunk_record
*rec
;
284 struct chunk_record
*exist
;
285 struct cache_extent
*cache
;
288 rec
= btrfs_new_chunk_record(leaf
, key
, slot
);
289 if (!rec
->cache
.size
)
292 cache
= lookup_cache_extent(chunk_cache
, rec
->offset
, rec
->length
);
294 exist
= container_of(cache
, struct chunk_record
, cache
);
296 if (exist
->generation
> rec
->generation
)
298 if (exist
->generation
== rec
->generation
) {
299 int num_stripes
= rec
->num_stripes
;
300 int rec_size
= btrfs_chunk_record_size(num_stripes
);
301 int offset
= offsetof(struct chunk_record
, generation
);
303 if (exist
->num_stripes
!= rec
->num_stripes
||
304 memcmp(((void *)exist
) + offset
,
305 ((void *)rec
) + offset
,
310 remove_cache_extent(chunk_cache
, cache
);
314 ret
= insert_cache_extent(chunk_cache
, &rec
->cache
);
323 static int process_device_extent_item(struct device_extent_tree
*devext_cache
,
324 struct extent_buffer
*leaf
,
325 struct btrfs_key
*key
, int slot
)
327 struct device_extent_record
*rec
;
328 struct device_extent_record
*exist
;
329 struct cache_extent
*cache
;
332 rec
= btrfs_new_device_extent_record(leaf
, key
, slot
);
333 if (!rec
->cache
.size
)
336 cache
= lookup_cache_extent2(&devext_cache
->tree
,
341 exist
= container_of(cache
, struct device_extent_record
, cache
);
342 if (exist
->generation
> rec
->generation
)
344 if (exist
->generation
== rec
->generation
) {
345 int offset
= offsetof(struct device_extent_record
,
347 if (memcmp(((void *)exist
) + offset
,
348 ((void *)rec
) + offset
,
349 sizeof(*rec
) - offset
))
353 remove_cache_extent(&devext_cache
->tree
, cache
);
354 list_del_init(&exist
->chunk_list
);
355 list_del_init(&exist
->device_list
);
360 ret
= insert_device_extent_record(devext_cache
, rec
);
369 static void print_block_group_info(struct block_group_record
*rec
, char *prefix
)
372 printf("%s", prefix
);
373 printf("Block Group: start = %llu, len = %llu, flag = %llx\n",
374 rec
->objectid
, rec
->offset
, rec
->flags
);
377 static void print_block_group_tree(struct block_group_tree
*tree
)
379 struct cache_extent
*cache
;
380 struct block_group_record
*rec
;
382 printf("All Block Groups:\n");
383 for (cache
= first_cache_extent(&tree
->tree
); cache
;
384 cache
= next_cache_extent(cache
)) {
385 rec
= container_of(cache
, struct block_group_record
, cache
);
386 print_block_group_info(rec
, "\t");
391 static void print_stripe_info(struct stripe
*data
, char *prefix1
, char *prefix2
,
395 printf("%s", prefix1
);
397 printf("%s", prefix2
);
398 printf("[%2d] Stripe: devid = %llu, offset = %llu\n",
399 index
, data
->devid
, data
->offset
);
402 static void print_chunk_self_info(struct chunk_record
*rec
, char *prefix
)
407 printf("%s", prefix
);
408 printf("Chunk: start = %llu, len = %llu, type = %llx, num_stripes = %u\n",
409 rec
->offset
, rec
->length
, rec
->type_flags
, rec
->num_stripes
);
411 printf("%s", prefix
);
412 printf(" Stripes list:\n");
413 for (i
= 0; i
< rec
->num_stripes
; i
++)
414 print_stripe_info(&rec
->stripes
[i
], prefix
, " ", i
);
417 static void print_chunk_tree(struct cache_tree
*tree
)
419 struct cache_extent
*n
;
420 struct chunk_record
*entry
;
422 printf("All Chunks:\n");
423 for (n
= first_cache_extent(tree
); n
;
424 n
= next_cache_extent(n
)) {
425 entry
= container_of(n
, struct chunk_record
, cache
);
426 print_chunk_self_info(entry
, "\t");
431 static void print_device_extent_info(struct device_extent_record
*rec
,
435 printf("%s", prefix
);
436 printf("Device extent: devid = %llu, start = %llu, len = %llu, chunk offset = %llu\n",
437 rec
->objectid
, rec
->offset
, rec
->length
, rec
->chunk_offset
);
440 static void print_device_extent_tree(struct device_extent_tree
*tree
)
442 struct cache_extent
*n
;
443 struct device_extent_record
*entry
;
445 printf("All Device Extents:\n");
446 for (n
= first_cache_extent(&tree
->tree
); n
;
447 n
= next_cache_extent(n
)) {
448 entry
= container_of(n
, struct device_extent_record
, cache
);
449 print_device_extent_info(entry
, "\t");
454 static void print_device_info(struct btrfs_device
*device
, char *prefix
)
457 printf("%s", prefix
);
458 printf("Device: id = %llu, name = %s\n",
459 device
->devid
, device
->name
);
462 static void print_all_devices(struct list_head
*devices
)
464 struct btrfs_device
*dev
;
466 printf("All Devices:\n");
467 list_for_each_entry(dev
, devices
, dev_list
)
468 print_device_info(dev
, "\t");
472 static void print_scan_result(struct recover_control
*rc
)
477 printf("DEVICE SCAN RESULT:\n");
478 printf("Filesystem Information:\n");
479 printf("\tsectorsize: %d\n", rc
->sectorsize
);
480 printf("\tnodesize: %d\n", rc
->nodesize
);
481 printf("\ttree root generation: %llu\n", rc
->generation
);
482 printf("\tchunk root generation: %llu\n", rc
->chunk_root_generation
);
485 print_all_devices(&rc
->fs_devices
->devices
);
486 print_block_group_tree(&rc
->bg
);
487 print_chunk_tree(&rc
->chunk
);
488 print_device_extent_tree(&rc
->devext
);
491 static void print_chunk_info(struct chunk_record
*chunk
, char *prefix
)
493 struct device_extent_record
*devext
;
496 print_chunk_self_info(chunk
, prefix
);
498 printf("%s", prefix
);
500 print_block_group_info(chunk
->bg_rec
, " ");
502 printf(" No block group.\n");
504 printf("%s", prefix
);
505 if (list_empty(&chunk
->dextents
)) {
506 printf(" No device extent.\n");
508 printf(" Device extent list:\n");
510 list_for_each_entry(devext
, &chunk
->dextents
, chunk_list
) {
512 printf("%s", prefix
);
513 printf("%s[%2d]", " ", i
);
514 print_device_extent_info(devext
, NULL
);
520 static void print_check_result(struct recover_control
*rc
)
522 struct chunk_record
*chunk
;
523 struct block_group_record
*bg
;
524 struct device_extent_record
*devext
;
532 printf("CHECK RESULT:\n");
533 printf("Recoverable Chunks:\n");
534 list_for_each_entry(chunk
, &rc
->good_chunks
, list
) {
535 print_chunk_info(chunk
, " ");
539 list_for_each_entry(chunk
, &rc
->rebuild_chunks
, list
) {
540 print_chunk_info(chunk
, " ");
544 list_for_each_entry(chunk
, &rc
->unrepaired_chunks
, list
) {
545 print_chunk_info(chunk
, " ");
549 printf("Unrecoverable Chunks:\n");
550 list_for_each_entry(chunk
, &rc
->bad_chunks
, list
) {
551 print_chunk_info(chunk
, " ");
556 printf("Total Chunks:\t\t%d\n", total
);
557 printf(" Recoverable:\t\t%d\n", good
);
558 printf(" Unrecoverable:\t%d\n", bad
);
561 printf("Orphan Block Groups:\n");
562 list_for_each_entry(bg
, &rc
->bg
.block_groups
, list
)
563 print_block_group_info(bg
, " ");
566 printf("Orphan Device Extents:\n");
567 list_for_each_entry(devext
, &rc
->devext
.no_chunk_orphans
, chunk_list
)
568 print_device_extent_info(devext
, " ");
572 static int check_chunk_by_metadata(struct recover_control
*rc
,
573 struct btrfs_root
*root
,
574 struct chunk_record
*chunk
, int bg_only
)
579 struct btrfs_path path
;
580 struct btrfs_key key
;
581 struct btrfs_root
*dev_root
;
582 struct stripe
*stripe
;
583 struct btrfs_dev_extent
*dev_extent
;
584 struct btrfs_block_group_item
*bg_ptr
;
585 struct extent_buffer
*l
;
587 btrfs_init_path(&path
);
592 dev_root
= root
->fs_info
->dev_root
;
593 for (i
= 0; i
< chunk
->num_stripes
; i
++) {
594 stripe
= &chunk
->stripes
[i
];
596 key
.objectid
= stripe
->devid
;
597 key
.offset
= stripe
->offset
;
598 key
.type
= BTRFS_DEV_EXTENT_KEY
;
600 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, &path
, 0, 0);
602 fprintf(stderr
, "Search device extent failed(%d)\n",
604 btrfs_release_path(&path
);
606 } else if (ret
> 0) {
609 "No device extent[%llu, %llu]\n",
610 stripe
->devid
, stripe
->offset
);
611 btrfs_release_path(&path
);
615 slot
= path
.slots
[0];
616 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
618 btrfs_dev_extent_chunk_offset(l
, dev_extent
)) {
621 "Device tree mismatch with chunks dev_extent[%llu, %llu], chunk[%llu, %llu]\n",
622 btrfs_dev_extent_chunk_offset(l
,
624 btrfs_dev_extent_length(l
, dev_extent
),
625 chunk
->offset
, chunk
->length
);
626 btrfs_release_path(&path
);
629 btrfs_release_path(&path
);
633 key
.objectid
= chunk
->offset
;
634 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
635 key
.offset
= chunk
->length
;
637 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, &path
,
640 fprintf(stderr
, "Search block group failed(%d)\n", ret
);
641 btrfs_release_path(&path
);
643 } else if (ret
> 0) {
645 fprintf(stderr
, "No block group[%llu, %llu]\n",
646 key
.objectid
, key
.offset
);
647 btrfs_release_path(&path
);
652 slot
= path
.slots
[0];
653 bg_ptr
= btrfs_item_ptr(l
, slot
, struct btrfs_block_group_item
);
654 if (chunk
->type_flags
!= btrfs_disk_block_group_flags(l
, bg_ptr
)) {
657 "Chunk[%llu, %llu]'s type(%llu) is different with Block Group's type(%llu)\n",
658 chunk
->offset
, chunk
->length
, chunk
->type_flags
,
659 btrfs_disk_block_group_flags(l
, bg_ptr
));
660 btrfs_release_path(&path
);
663 btrfs_release_path(&path
);
667 static int check_all_chunks_by_metadata(struct recover_control
*rc
,
668 struct btrfs_root
*root
)
670 struct chunk_record
*chunk
;
671 struct chunk_record
*next
;
672 LIST_HEAD(orphan_chunks
);
676 list_for_each_entry_safe(chunk
, next
, &rc
->good_chunks
, list
) {
677 err
= check_chunk_by_metadata(rc
, root
, chunk
, 0);
680 list_move_tail(&chunk
->list
, &orphan_chunks
);
681 else if (err
&& !ret
)
686 list_for_each_entry_safe(chunk
, next
, &rc
->unrepaired_chunks
, list
) {
687 err
= check_chunk_by_metadata(rc
, root
, chunk
, 1);
689 list_move_tail(&chunk
->list
, &orphan_chunks
);
690 else if (err
&& !ret
)
694 list_for_each_entry(chunk
, &rc
->bad_chunks
, list
) {
695 err
= check_chunk_by_metadata(rc
, root
, chunk
, 1);
696 if (err
!= -ENOENT
&& !ret
)
697 ret
= err
? err
: -EINVAL
;
699 list_splice(&orphan_chunks
, &rc
->bad_chunks
);
703 static int extract_metadata_record(struct recover_control
*rc
,
704 struct extent_buffer
*leaf
)
706 struct btrfs_key key
;
711 nritems
= btrfs_header_nritems(leaf
);
712 for (i
= 0; i
< nritems
; i
++) {
713 btrfs_item_key_to_cpu(leaf
, &key
, i
);
715 case BTRFS_BLOCK_GROUP_ITEM_KEY
:
716 pthread_mutex_lock(&rc
->rc_lock
);
717 ret
= process_block_group_item(&rc
->bg
, leaf
, &key
, i
);
718 pthread_mutex_unlock(&rc
->rc_lock
);
720 case BTRFS_CHUNK_ITEM_KEY
:
721 pthread_mutex_lock(&rc
->rc_lock
);
722 ret
= process_chunk_item(&rc
->chunk
, leaf
, &key
, i
);
723 pthread_mutex_unlock(&rc
->rc_lock
);
725 case BTRFS_DEV_EXTENT_KEY
:
726 pthread_mutex_lock(&rc
->rc_lock
);
727 ret
= process_device_extent_item(&rc
->devext
, leaf
,
729 pthread_mutex_unlock(&rc
->rc_lock
);
738 static inline int is_super_block_address(u64 offset
)
742 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
743 if (offset
== btrfs_sb_offset(i
))
749 static int scan_one_device(void *dev_scan_struct
)
751 struct extent_buffer
*buf
;
754 struct device_scan
*dev_scan
= (struct device_scan
*)dev_scan_struct
;
755 struct recover_control
*rc
= dev_scan
->rc
;
756 struct btrfs_device
*device
= dev_scan
->dev
;
757 int fd
= dev_scan
->fd
;
760 ret
= pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS
, &oldtype
);
764 buf
= malloc(sizeof(*buf
) + rc
->nodesize
);
767 buf
->len
= rc
->nodesize
;
771 dev_scan
->bytenr
= bytenr
;
773 if (is_super_block_address(bytenr
))
774 bytenr
+= rc
->sectorsize
;
776 if (pread64(fd
, buf
->data
, rc
->nodesize
, bytenr
) <
780 if (memcmp_extent_buffer(buf
, rc
->fs_devices
->fsid
,
783 bytenr
+= rc
->sectorsize
;
787 if (verify_tree_block_csum_silent(buf
, rc
->csum_size
)) {
788 bytenr
+= rc
->sectorsize
;
792 pthread_mutex_lock(&rc
->rc_lock
);
793 ret
= process_extent_buffer(&rc
->eb_cache
, buf
, device
, bytenr
);
794 pthread_mutex_unlock(&rc
->rc_lock
);
798 if (btrfs_header_level(buf
) != 0)
801 switch (btrfs_header_owner(buf
)) {
802 case BTRFS_EXTENT_TREE_OBJECTID
:
803 case BTRFS_DEV_TREE_OBJECTID
:
804 /* different tree use different generation */
805 if (btrfs_header_generation(buf
) > rc
->generation
)
807 ret
= extract_metadata_record(rc
, buf
);
811 case BTRFS_CHUNK_TREE_OBJECTID
:
812 if (btrfs_header_generation(buf
) >
813 rc
->chunk_root_generation
)
815 ret
= extract_metadata_record(rc
, buf
);
821 bytenr
+= rc
->nodesize
;
829 static int scan_devices(struct recover_control
*rc
)
833 struct btrfs_device
*dev
;
834 struct device_scan
*dev_scans
;
842 list_for_each_entry(dev
, &rc
->fs_devices
->devices
, dev_list
)
844 dev_scans
= (struct device_scan
*)malloc(sizeof(struct device_scan
)
848 t_scans
= (pthread_t
*)malloc(sizeof(pthread_t
) * devnr
);
853 t_rets
= (long *)malloc(sizeof(long) * devnr
);
860 list_for_each_entry(dev
, &rc
->fs_devices
->devices
, dev_list
) {
861 fd
= open(dev
->name
, O_RDONLY
);
863 fprintf(stderr
, "Failed to open device %s\n",
868 dev_scans
[devidx
].rc
= rc
;
869 dev_scans
[devidx
].dev
= dev
;
870 dev_scans
[devidx
].fd
= fd
;
871 dev_scans
[devidx
].bytenr
= -1;
875 for (i
= 0; i
< devidx
; i
++) {
876 ret
= pthread_create(&t_scans
[i
], NULL
,
877 (void *)scan_one_device
,
878 (void *)&dev_scans
[i
]);
882 dev_scans
[i
].bytenr
= 0;
887 for (i
= 0; i
< devidx
; i
++) {
888 if (dev_scans
[i
].bytenr
== -1)
890 ret
= pthread_tryjoin_np(t_scans
[i
],
891 (void **)&t_rets
[i
]);
896 if (ret
|| t_rets
[i
]) {
900 dev_scans
[i
].bytenr
= -1;
903 printf("\rScanning: ");
904 for (i
= 0; i
< devidx
; i
++) {
905 if (dev_scans
[i
].bytenr
== -1)
906 printf("%sDONE in dev%d",
909 printf("%s%llu in dev%d",
910 i
? ", " : "", dev_scans
[i
].bytenr
, i
);
912 /* clear chars if exist in tail */
914 printf("\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b");
925 for (i
= 0; i
< devidx
; i
++) {
926 if (dev_scans
[i
].bytenr
== -1)
928 pthread_cancel(t_scans
[i
]);
937 static int build_device_map_by_chunk_record(struct btrfs_root
*root
,
938 struct chunk_record
*chunk
)
943 u8 uuid
[BTRFS_UUID_SIZE
];
945 struct btrfs_mapping_tree
*map_tree
;
946 struct map_lookup
*map
;
947 struct stripe
*stripe
;
949 map_tree
= &root
->fs_info
->mapping_tree
;
950 num_stripes
= chunk
->num_stripes
;
951 map
= malloc(btrfs_map_lookup_size(num_stripes
));
954 map
->ce
.start
= chunk
->offset
;
955 map
->ce
.size
= chunk
->length
;
956 map
->num_stripes
= num_stripes
;
957 map
->io_width
= chunk
->io_width
;
958 map
->io_align
= chunk
->io_align
;
959 map
->sector_size
= chunk
->sector_size
;
960 map
->stripe_len
= chunk
->stripe_len
;
961 map
->type
= chunk
->type_flags
;
962 map
->sub_stripes
= chunk
->sub_stripes
;
964 for (i
= 0, stripe
= chunk
->stripes
; i
< num_stripes
; i
++, stripe
++) {
965 devid
= stripe
->devid
;
966 memcpy(uuid
, stripe
->dev_uuid
, BTRFS_UUID_SIZE
);
967 map
->stripes
[i
].physical
= stripe
->offset
;
968 map
->stripes
[i
].dev
= btrfs_find_device(root
, devid
,
970 if (!map
->stripes
[i
].dev
) {
976 ret
= insert_cache_extent(&map_tree
->cache_tree
, &map
->ce
);
980 static int build_device_maps_by_chunk_records(struct recover_control
*rc
,
981 struct btrfs_root
*root
)
984 struct chunk_record
*chunk
;
986 list_for_each_entry(chunk
, &rc
->good_chunks
, list
) {
987 ret
= build_device_map_by_chunk_record(root
, chunk
);
991 list_for_each_entry(chunk
, &rc
->rebuild_chunks
, list
) {
992 ret
= build_device_map_by_chunk_record(root
, chunk
);
999 static int block_group_remove_all_extent_items(struct btrfs_trans_handle
*trans
,
1000 struct btrfs_root
*root
,
1001 struct block_group_record
*bg
)
1003 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1004 struct btrfs_key key
;
1005 struct btrfs_path path
;
1006 struct extent_buffer
*leaf
;
1007 u64 start
= bg
->objectid
;
1008 u64 end
= bg
->objectid
+ bg
->offset
;
1015 btrfs_init_path(&path
);
1016 root
= root
->fs_info
->extent_root
;
1018 key
.objectid
= start
;
1020 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1022 ret
= btrfs_search_slot(trans
, root
, &key
, &path
, -1, 1);
1028 leaf
= path
.nodes
[0];
1029 nitems
= btrfs_header_nritems(leaf
);
1031 /* The tree is empty. */
1036 if (path
.slots
[0] >= nitems
) {
1037 ret
= btrfs_next_leaf(root
, &path
);
1044 leaf
= path
.nodes
[0];
1045 btrfs_item_key_to_cpu(leaf
, &key
, 0);
1046 if (key
.objectid
>= end
)
1048 btrfs_release_path(&path
);
1054 for (i
= path
.slots
[0]; i
< nitems
; i
++) {
1055 btrfs_item_key_to_cpu(leaf
, &key
, i
);
1056 if (key
.objectid
>= end
)
1059 if (key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1069 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
1070 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
1071 old_val
= btrfs_super_bytes_used(fs_info
->super_copy
);
1072 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
1073 old_val
+= root
->nodesize
;
1075 old_val
+= key
.offset
;
1076 btrfs_set_super_bytes_used(fs_info
->super_copy
,
1082 ret
= btrfs_del_items(trans
, root
, &path
, del_s
, del_nr
);
1087 if (key
.objectid
< end
) {
1088 if (key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1089 key
.objectid
+= root
->sectorsize
;
1090 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1093 btrfs_release_path(&path
);
1097 btrfs_release_path(&path
);
1101 static int block_group_free_all_extent(struct btrfs_root
*root
,
1102 struct block_group_record
*bg
)
1104 struct btrfs_block_group_cache
*cache
;
1105 struct btrfs_fs_info
*info
;
1109 info
= root
->fs_info
;
1110 cache
= btrfs_lookup_block_group(info
, bg
->objectid
);
1114 start
= cache
->key
.objectid
;
1115 end
= start
+ cache
->key
.offset
- 1;
1117 set_extent_bits(&info
->block_group_cache
, start
, end
,
1119 set_extent_dirty(&info
->free_space_cache
, start
, end
);
1121 btrfs_set_block_group_used(&cache
->item
, 0);
1126 static int remove_chunk_extent_item(struct btrfs_trans_handle
*trans
,
1127 struct recover_control
*rc
,
1128 struct btrfs_root
*root
)
1130 struct chunk_record
*chunk
;
1133 list_for_each_entry(chunk
, &rc
->good_chunks
, list
) {
1134 if (!(chunk
->type_flags
& BTRFS_BLOCK_GROUP_SYSTEM
))
1136 ret
= block_group_remove_all_extent_items(trans
, root
,
1141 ret
= block_group_free_all_extent(root
, chunk
->bg_rec
);
1148 static int __rebuild_chunk_root(struct btrfs_trans_handle
*trans
,
1149 struct recover_control
*rc
,
1150 struct btrfs_root
*root
)
1153 struct btrfs_device
*dev
;
1154 struct extent_buffer
*cow
;
1155 struct btrfs_disk_key disk_key
;
1158 list_for_each_entry(dev
, &rc
->fs_devices
->devices
, dev_list
) {
1159 if (min_devid
> dev
->devid
)
1160 min_devid
= dev
->devid
;
1162 btrfs_set_disk_key_objectid(&disk_key
, BTRFS_DEV_ITEMS_OBJECTID
);
1163 btrfs_set_disk_key_type(&disk_key
, BTRFS_DEV_ITEM_KEY
);
1164 btrfs_set_disk_key_offset(&disk_key
, min_devid
);
1166 cow
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
,
1167 BTRFS_CHUNK_TREE_OBJECTID
,
1168 &disk_key
, 0, 0, 0);
1169 btrfs_set_header_bytenr(cow
, cow
->start
);
1170 btrfs_set_header_generation(cow
, trans
->transid
);
1171 btrfs_set_header_nritems(cow
, 0);
1172 btrfs_set_header_level(cow
, 0);
1173 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
1174 btrfs_set_header_owner(cow
, BTRFS_CHUNK_TREE_OBJECTID
);
1175 write_extent_buffer(cow
, root
->fs_info
->fsid
,
1176 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
1178 write_extent_buffer(cow
, root
->fs_info
->chunk_tree_uuid
,
1179 btrfs_header_chunk_tree_uuid(cow
),
1183 btrfs_mark_buffer_dirty(cow
);
1188 static int __rebuild_device_items(struct btrfs_trans_handle
*trans
,
1189 struct recover_control
*rc
,
1190 struct btrfs_root
*root
)
1192 struct btrfs_device
*dev
;
1193 struct btrfs_key key
;
1194 struct btrfs_dev_item dev_item_tmp
;
1195 struct btrfs_dev_item
*dev_item
= &dev_item_tmp
;
1198 list_for_each_entry(dev
, &rc
->fs_devices
->devices
, dev_list
) {
1199 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1200 key
.type
= BTRFS_DEV_ITEM_KEY
;
1201 key
.offset
= dev
->devid
;
1203 btrfs_set_stack_device_generation(dev_item
, 0);
1204 btrfs_set_stack_device_type(dev_item
, dev
->type
);
1205 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
1206 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
1207 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
1208 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
1209 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
1210 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
1211 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
1212 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
1214 ret
= btrfs_insert_item(trans
, root
, &key
,
1215 dev_item
, sizeof(*dev_item
));
1221 static int __insert_chunk_item(struct btrfs_trans_handle
*trans
,
1222 struct chunk_record
*chunk_rec
,
1223 struct btrfs_root
*chunk_root
)
1225 struct btrfs_key key
;
1226 struct btrfs_chunk
*chunk
= NULL
;
1229 chunk
= create_chunk_item(chunk_rec
);
1232 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
1233 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1234 key
.offset
= chunk_rec
->offset
;
1236 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
,
1237 btrfs_chunk_item_size(chunk_rec
->num_stripes
));
1242 static int __rebuild_chunk_items(struct btrfs_trans_handle
*trans
,
1243 struct recover_control
*rc
,
1244 struct btrfs_root
*root
)
1246 struct btrfs_root
*chunk_root
;
1247 struct chunk_record
*chunk_rec
;
1250 chunk_root
= root
->fs_info
->chunk_root
;
1252 list_for_each_entry(chunk_rec
, &rc
->good_chunks
, list
) {
1253 ret
= __insert_chunk_item(trans
, chunk_rec
, chunk_root
);
1257 list_for_each_entry(chunk_rec
, &rc
->rebuild_chunks
, list
) {
1258 ret
= __insert_chunk_item(trans
, chunk_rec
, chunk_root
);
1265 static int rebuild_chunk_tree(struct btrfs_trans_handle
*trans
,
1266 struct recover_control
*rc
,
1267 struct btrfs_root
*root
)
1271 root
= root
->fs_info
->chunk_root
;
1273 ret
= __rebuild_chunk_root(trans
, rc
, root
);
1277 ret
= __rebuild_device_items(trans
, rc
, root
);
1281 ret
= __rebuild_chunk_items(trans
, rc
, root
);
1286 static int rebuild_sys_array(struct recover_control
*rc
,
1287 struct btrfs_root
*root
)
1289 struct btrfs_chunk
*chunk
;
1290 struct btrfs_key key
;
1291 struct chunk_record
*chunk_rec
;
1295 btrfs_set_super_sys_array_size(root
->fs_info
->super_copy
, 0);
1297 list_for_each_entry(chunk_rec
, &rc
->good_chunks
, list
) {
1298 if (!(chunk_rec
->type_flags
& BTRFS_BLOCK_GROUP_SYSTEM
))
1301 num_stripes
= chunk_rec
->num_stripes
;
1302 chunk
= create_chunk_item(chunk_rec
);
1308 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
1309 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1310 key
.offset
= chunk_rec
->offset
;
1312 ret
= btrfs_add_system_chunk(root
, &key
, chunk
,
1313 btrfs_chunk_item_size(num_stripes
));
1322 static int calculate_bg_used(struct btrfs_root
*extent_root
,
1323 struct chunk_record
*chunk_rec
,
1324 struct btrfs_path
*path
,
1327 struct extent_buffer
*node
;
1328 struct btrfs_key found_key
;
1334 node
= path
->nodes
[0];
1335 slot
= path
->slots
[0];
1336 btrfs_item_key_to_cpu(node
, &found_key
, slot
);
1337 if (found_key
.objectid
>= chunk_rec
->offset
+ chunk_rec
->length
)
1339 if (found_key
.type
!= BTRFS_METADATA_ITEM_KEY
&&
1340 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
1342 if (found_key
.type
== BTRFS_METADATA_ITEM_KEY
)
1343 used_ret
+= extent_root
->nodesize
;
1345 used_ret
+= found_key
.offset
;
1347 if (slot
+ 1 < btrfs_header_nritems(node
)) {
1350 ret
= btrfs_next_leaf(extent_root
, path
);
1364 static int __insert_block_group(struct btrfs_trans_handle
*trans
,
1365 struct chunk_record
*chunk_rec
,
1366 struct btrfs_root
*extent_root
,
1369 struct btrfs_block_group_item bg_item
;
1370 struct btrfs_key key
;
1373 btrfs_set_block_group_used(&bg_item
, used
);
1374 btrfs_set_block_group_chunk_objectid(&bg_item
, used
);
1375 btrfs_set_block_group_flags(&bg_item
, chunk_rec
->type_flags
);
1376 key
.objectid
= chunk_rec
->offset
;
1377 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
1378 key
.offset
= chunk_rec
->length
;
1380 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &bg_item
,
1386 * Search through the extent tree to rebuild the 'used' member of the block
1388 * However, since block group and extent item shares the extent tree,
1389 * the extent item may also missing.
1390 * In that case, we fill the 'used' with the length of the block group to
1391 * ensure no write into the block group.
1392 * Btrfsck will hate it but we will inform user to call '--init-extent-tree'
1393 * if possible, or just salvage as much data as possible from the fs.
1395 static int rebuild_block_group(struct btrfs_trans_handle
*trans
,
1396 struct recover_control
*rc
,
1397 struct btrfs_root
*root
)
1399 struct chunk_record
*chunk_rec
;
1400 struct btrfs_key search_key
;
1401 struct btrfs_path path
;
1405 if (list_empty(&rc
->rebuild_chunks
))
1408 btrfs_init_path(&path
);
1409 list_for_each_entry(chunk_rec
, &rc
->rebuild_chunks
, list
) {
1410 search_key
.objectid
= chunk_rec
->offset
;
1411 search_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1412 search_key
.offset
= 0;
1413 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
,
1414 &search_key
, &path
, 0, 0);
1417 ret
= calculate_bg_used(root
->fs_info
->extent_root
,
1418 chunk_rec
, &path
, &used
);
1420 * Extent tree is damaged, better to rebuild the whole extent
1421 * tree. Currently, change the used to chunk's len to prevent
1422 * write/block reserve happening in that block group.
1426 "Fail to search extent tree for block group: [%llu,%llu]\n",
1428 chunk_rec
->offset
+ chunk_rec
->length
);
1430 "Mark the block group full to prevent block rsv problems\n");
1431 used
= chunk_rec
->length
;
1433 btrfs_release_path(&path
);
1434 ret
= __insert_block_group(trans
, chunk_rec
,
1435 root
->fs_info
->extent_root
,
1441 btrfs_release_path(&path
);
1445 static struct btrfs_root
*
1446 open_ctree_with_broken_chunk(struct recover_control
*rc
)
1448 struct btrfs_fs_info
*fs_info
;
1449 struct btrfs_super_block
*disk_super
;
1450 struct extent_buffer
*eb
;
1457 fs_info
= btrfs_new_fs_info(1, BTRFS_SUPER_INFO_OFFSET
);
1459 fprintf(stderr
, "Failed to allocate memory for fs_info\n");
1460 return ERR_PTR(-ENOMEM
);
1462 fs_info
->is_chunk_recover
= 1;
1464 fs_info
->fs_devices
= rc
->fs_devices
;
1465 ret
= btrfs_open_devices(fs_info
->fs_devices
, O_RDWR
);
1469 disk_super
= fs_info
->super_copy
;
1470 ret
= btrfs_read_dev_super(fs_info
->fs_devices
->latest_bdev
,
1471 disk_super
, fs_info
->super_bytenr
,
1474 fprintf(stderr
, "No valid btrfs found\n");
1478 memcpy(fs_info
->fsid
, &disk_super
->fsid
, BTRFS_FSID_SIZE
);
1480 ret
= btrfs_check_fs_compatibility(disk_super
, OPEN_CTREE_WRITES
);
1484 nodesize
= btrfs_super_nodesize(disk_super
);
1485 leafsize
= btrfs_super_leafsize(disk_super
);
1486 sectorsize
= btrfs_super_sectorsize(disk_super
);
1487 stripesize
= btrfs_super_stripesize(disk_super
);
1489 btrfs_setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
1490 fs_info
->chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
1492 ret
= build_device_maps_by_chunk_records(rc
, fs_info
->chunk_root
);
1496 ret
= btrfs_setup_all_roots(fs_info
, 0, 0);
1500 eb
= fs_info
->tree_root
->node
;
1501 read_extent_buffer(eb
, fs_info
->chunk_tree_uuid
,
1502 btrfs_header_chunk_tree_uuid(eb
),
1505 return fs_info
->fs_root
;
1507 btrfs_release_all_roots(fs_info
);
1509 btrfs_cleanup_all_caches(fs_info
);
1511 btrfs_close_devices(fs_info
->fs_devices
);
1513 btrfs_free_fs_info(fs_info
);
1514 return ERR_PTR(ret
);
1517 static int recover_prepare(struct recover_control
*rc
, char *path
)
1521 struct btrfs_super_block
*sb
;
1522 char buf
[BTRFS_SUPER_INFO_SIZE
];
1523 struct btrfs_fs_devices
*fs_devices
;
1526 fd
= open(path
, O_RDONLY
);
1528 fprintf(stderr
, "open %s\n error.\n", path
);
1532 sb
= (struct btrfs_super_block
*)buf
;
1533 ret
= btrfs_read_dev_super(fd
, sb
, BTRFS_SUPER_INFO_OFFSET
,
1536 fprintf(stderr
, "read super block error\n");
1540 rc
->sectorsize
= btrfs_super_sectorsize(sb
);
1541 rc
->nodesize
= btrfs_super_nodesize(sb
);
1542 rc
->generation
= btrfs_super_generation(sb
);
1543 rc
->chunk_root_generation
= btrfs_super_chunk_root_generation(sb
);
1544 rc
->csum_size
= btrfs_super_csum_size(sb
);
1546 /* if seed, the result of scanning below will be partial */
1547 if (btrfs_super_flags(sb
) & BTRFS_SUPER_FLAG_SEEDING
) {
1548 fprintf(stderr
, "this device is seed device\n");
1553 ret
= btrfs_scan_fs_devices(fd
, path
, &fs_devices
, 0, SBREAD_RECOVER
, 0);
1557 rc
->fs_devices
= fs_devices
;
1560 print_all_devices(&rc
->fs_devices
->devices
);
1567 static int btrfs_get_device_extents(u64 chunk_object
,
1568 struct list_head
*orphan_devexts
,
1569 struct list_head
*ret_list
)
1571 struct device_extent_record
*devext
;
1572 struct device_extent_record
*next
;
1575 list_for_each_entry_safe(devext
, next
, orphan_devexts
, chunk_list
) {
1576 if (devext
->chunk_offset
== chunk_object
) {
1577 list_move_tail(&devext
->chunk_list
, ret_list
);
1584 static int calc_num_stripes(u64 type
)
1586 if (type
& (BTRFS_BLOCK_GROUP_RAID0
|
1587 BTRFS_BLOCK_GROUP_RAID10
|
1588 BTRFS_BLOCK_GROUP_RAID5
|
1589 BTRFS_BLOCK_GROUP_RAID6
))
1591 else if (type
& (BTRFS_BLOCK_GROUP_RAID1
|
1592 BTRFS_BLOCK_GROUP_DUP
))
1598 static inline int calc_sub_nstripes(u64 type
)
1600 if (type
& BTRFS_BLOCK_GROUP_RAID10
)
1606 static int btrfs_verify_device_extents(struct block_group_record
*bg
,
1607 struct list_head
*devexts
, int ndevexts
)
1609 struct device_extent_record
*devext
;
1611 int expected_num_stripes
;
1613 expected_num_stripes
= calc_num_stripes(bg
->flags
);
1614 if (expected_num_stripes
&& expected_num_stripes
!= ndevexts
)
1617 if (check_num_stripes(bg
->flags
, ndevexts
) < 0)
1620 stripe_length
= calc_stripe_length(bg
->flags
, bg
->offset
, ndevexts
);
1621 list_for_each_entry(devext
, devexts
, chunk_list
) {
1622 if (devext
->length
!= stripe_length
)
1628 static int btrfs_rebuild_unordered_chunk_stripes(struct recover_control
*rc
,
1629 struct chunk_record
*chunk
)
1631 struct device_extent_record
*devext
;
1632 struct btrfs_device
*device
;
1635 devext
= list_first_entry(&chunk
->dextents
, struct device_extent_record
,
1637 for (i
= 0; i
< chunk
->num_stripes
; i
++) {
1638 chunk
->stripes
[i
].devid
= devext
->objectid
;
1639 chunk
->stripes
[i
].offset
= devext
->offset
;
1640 device
= btrfs_find_device_by_devid(rc
->fs_devices
,
1645 BUG_ON(btrfs_find_device_by_devid(rc
->fs_devices
,
1648 memcpy(chunk
->stripes
[i
].dev_uuid
, device
->uuid
,
1650 devext
= list_next_entry(devext
, chunk_list
);
1655 static int btrfs_calc_stripe_index(struct chunk_record
*chunk
, u64 logical
)
1657 u64 offset
= logical
- chunk
->offset
;
1659 int nr_data_stripes
;
1662 stripe_nr
= offset
/ chunk
->stripe_len
;
1663 if (chunk
->type_flags
& BTRFS_BLOCK_GROUP_RAID0
) {
1664 index
= stripe_nr
% chunk
->num_stripes
;
1665 } else if (chunk
->type_flags
& BTRFS_BLOCK_GROUP_RAID10
) {
1666 index
= stripe_nr
% (chunk
->num_stripes
/ chunk
->sub_stripes
);
1667 index
*= chunk
->sub_stripes
;
1668 } else if (chunk
->type_flags
& BTRFS_BLOCK_GROUP_RAID5
) {
1669 nr_data_stripes
= chunk
->num_stripes
- 1;
1670 index
= stripe_nr
% nr_data_stripes
;
1671 stripe_nr
/= nr_data_stripes
;
1672 index
= (index
+ stripe_nr
) % chunk
->num_stripes
;
1673 } else if (chunk
->type_flags
& BTRFS_BLOCK_GROUP_RAID6
) {
1674 nr_data_stripes
= chunk
->num_stripes
- 2;
1675 index
= stripe_nr
% nr_data_stripes
;
1676 stripe_nr
/= nr_data_stripes
;
1677 index
= (index
+ stripe_nr
) % chunk
->num_stripes
;
1684 /* calc the logical offset which is the start of the next stripe. */
1685 static inline u64
btrfs_next_stripe_logical_offset(struct chunk_record
*chunk
,
1688 u64 offset
= logical
- chunk
->offset
;
1690 offset
/= chunk
->stripe_len
;
1691 offset
*= chunk
->stripe_len
;
1692 offset
+= chunk
->stripe_len
;
1694 return offset
+ chunk
->offset
;
1697 static int is_extent_record_in_device_extent(struct extent_record
*er
,
1698 struct device_extent_record
*dext
,
1703 for (i
= 0; i
< er
->nmirrors
; i
++) {
1704 if (er
->devices
[i
]->devid
== dext
->objectid
&&
1705 er
->offsets
[i
] >= dext
->offset
&&
1706 er
->offsets
[i
] < dext
->offset
+ dext
->length
) {
1715 btrfs_rebuild_ordered_meta_chunk_stripes(struct recover_control
*rc
,
1716 struct chunk_record
*chunk
)
1718 u64 start
= chunk
->offset
;
1719 u64 end
= chunk
->offset
+ chunk
->length
;
1720 struct cache_extent
*cache
;
1721 struct extent_record
*er
;
1722 struct device_extent_record
*devext
;
1723 struct device_extent_record
*next
;
1724 struct btrfs_device
*device
;
1730 cache
= lookup_cache_extent(&rc
->eb_cache
,
1731 start
, chunk
->length
);
1733 /* No used space, we can reorder the stripes freely. */
1734 ret
= btrfs_rebuild_unordered_chunk_stripes(rc
, chunk
);
1738 list_splice_init(&chunk
->dextents
, &devexts
);
1740 er
= container_of(cache
, struct extent_record
, cache
);
1741 index
= btrfs_calc_stripe_index(chunk
, er
->cache
.start
);
1742 BUG_ON(index
== -1);
1743 if (chunk
->stripes
[index
].devid
)
1745 list_for_each_entry_safe(devext
, next
, &devexts
, chunk_list
) {
1746 if (is_extent_record_in_device_extent(er
, devext
, &mirror
)) {
1747 chunk
->stripes
[index
].devid
= devext
->objectid
;
1748 chunk
->stripes
[index
].offset
= devext
->offset
;
1749 memcpy(chunk
->stripes
[index
].dev_uuid
,
1750 er
->devices
[mirror
]->uuid
,
1753 list_move(&devext
->chunk_list
, &chunk
->dextents
);
1757 start
= btrfs_next_stripe_logical_offset(chunk
, er
->cache
.start
);
1759 goto no_extent_record
;
1761 cache
= lookup_cache_extent(&rc
->eb_cache
, start
, end
- start
);
1765 if (list_empty(&devexts
))
1768 if (chunk
->type_flags
& (BTRFS_BLOCK_GROUP_RAID5
|
1769 BTRFS_BLOCK_GROUP_RAID6
)) {
1770 /* Fixme: try to recover the order by the parity block. */
1771 list_splice_tail(&devexts
, &chunk
->dextents
);
1775 /* There is no data on the lost stripes, we can reorder them freely. */
1776 for (index
= 0; index
< chunk
->num_stripes
; index
++) {
1777 if (chunk
->stripes
[index
].devid
)
1780 devext
= list_first_entry(&devexts
,
1781 struct device_extent_record
,
1783 list_move(&devext
->chunk_list
, &chunk
->dextents
);
1785 chunk
->stripes
[index
].devid
= devext
->objectid
;
1786 chunk
->stripes
[index
].offset
= devext
->offset
;
1787 device
= btrfs_find_device_by_devid(rc
->fs_devices
,
1791 list_splice_tail(&devexts
, &chunk
->dextents
);
1794 BUG_ON(btrfs_find_device_by_devid(rc
->fs_devices
,
1797 memcpy(chunk
->stripes
[index
].dev_uuid
, device
->uuid
,
1803 #define BTRFS_ORDERED_RAID (BTRFS_BLOCK_GROUP_RAID0 | \
1804 BTRFS_BLOCK_GROUP_RAID10 | \
1805 BTRFS_BLOCK_GROUP_RAID5 | \
1806 BTRFS_BLOCK_GROUP_RAID6)
1808 static int btrfs_rebuild_chunk_stripes(struct recover_control
*rc
,
1809 struct chunk_record
*chunk
)
1814 * All the data in the system metadata chunk will be dropped,
1815 * so we need not guarantee that the data is right or not, that
1816 * is we can reorder the stripes in the system metadata chunk.
1818 if ((chunk
->type_flags
& BTRFS_BLOCK_GROUP_METADATA
) &&
1819 (chunk
->type_flags
& BTRFS_ORDERED_RAID
))
1820 ret
=btrfs_rebuild_ordered_meta_chunk_stripes(rc
, chunk
);
1821 else if ((chunk
->type_flags
& BTRFS_BLOCK_GROUP_DATA
) &&
1822 (chunk
->type_flags
& BTRFS_ORDERED_RAID
))
1823 ret
= 1; /* Be handled after the fs is opened. */
1825 ret
= btrfs_rebuild_unordered_chunk_stripes(rc
, chunk
);
1830 static int next_csum(struct btrfs_root
*root
,
1831 struct extent_buffer
**leaf
,
1832 struct btrfs_path
*path
,
1837 struct btrfs_key
*key
)
1840 struct btrfs_root
*csum_root
= root
->fs_info
->csum_root
;
1841 struct btrfs_csum_item
*csum_item
;
1842 u32 blocksize
= root
->sectorsize
;
1843 u16 csum_size
= btrfs_super_csum_size(root
->fs_info
->super_copy
);
1844 int csums_in_item
= btrfs_item_size_nr(*leaf
, *slot
) / csum_size
;
1846 if (*csum_offset
>= csums_in_item
) {
1849 if (*slot
>= btrfs_header_nritems(*leaf
)) {
1850 ret
= btrfs_next_leaf(csum_root
, path
);
1855 *leaf
= path
->nodes
[0];
1856 *slot
= path
->slots
[0];
1858 btrfs_item_key_to_cpu(*leaf
, key
, *slot
);
1861 if (key
->offset
+ (*csum_offset
) * blocksize
>= end
)
1863 csum_item
= btrfs_item_ptr(*leaf
, *slot
, struct btrfs_csum_item
);
1864 csum_item
= (struct btrfs_csum_item
*)((unsigned char *)csum_item
1865 + (*csum_offset
) * csum_size
);
1866 read_extent_buffer(*leaf
, tree_csum
,
1867 (unsigned long)csum_item
, csum_size
);
1871 static u64
calc_data_offset(struct btrfs_key
*key
,
1872 struct chunk_record
*chunk
,
1878 int logical_stripe_nr
;
1880 int nr_data_stripes
;
1882 data_offset
= key
->offset
+ csum_offset
* blocksize
- chunk
->offset
;
1883 nr_data_stripes
= chunk
->num_stripes
;
1885 if (chunk
->type_flags
& BTRFS_BLOCK_GROUP_RAID5
)
1886 nr_data_stripes
-= 1;
1887 else if (chunk
->type_flags
& BTRFS_BLOCK_GROUP_RAID6
)
1888 nr_data_stripes
-= 2;
1890 logical_stripe_nr
= data_offset
/ chunk
->stripe_len
;
1891 dev_stripe_nr
= logical_stripe_nr
/ nr_data_stripes
;
1893 data_offset
-= logical_stripe_nr
* chunk
->stripe_len
;
1894 data_offset
+= dev_stripe_nr
* chunk
->stripe_len
;
1896 return dev_offset
+ data_offset
;
1899 static int check_one_csum(int fd
, u64 start
, u32 len
, u32 tree_csum
)
1903 u32 csum_result
= ~(u32
)0;
1908 ret
= pread64(fd
, data
, len
, start
);
1909 if (ret
< 0 || ret
!= len
) {
1914 csum_result
= btrfs_csum_data(data
, csum_result
, len
);
1915 btrfs_csum_final(csum_result
, (u8
*)&csum_result
);
1916 if (csum_result
!= tree_csum
)
1923 static u64
item_end_offset(struct btrfs_root
*root
, struct btrfs_key
*key
,
1924 struct extent_buffer
*leaf
, int slot
) {
1925 u32 blocksize
= root
->sectorsize
;
1926 u16 csum_size
= btrfs_super_csum_size(root
->fs_info
->super_copy
);
1928 u64 offset
= btrfs_item_size_nr(leaf
, slot
);
1929 offset
/= csum_size
;
1930 offset
*= blocksize
;
1931 offset
+= key
->offset
;
1936 static int insert_stripe(struct list_head
*devexts
,
1937 struct recover_control
*rc
,
1938 struct chunk_record
*chunk
,
1940 struct device_extent_record
*devext
;
1941 struct btrfs_device
*dev
;
1943 devext
= list_entry(devexts
->next
, struct device_extent_record
,
1945 dev
= btrfs_find_device_by_devid(rc
->fs_devices
, devext
->objectid
,
1949 if (btrfs_find_device_by_devid(rc
->fs_devices
, devext
->objectid
, 1)) {
1950 error("unexpected: found another device with id %llu",
1951 (unsigned long long)devext
->objectid
);
1955 chunk
->stripes
[index
].devid
= devext
->objectid
;
1956 chunk
->stripes
[index
].offset
= devext
->offset
;
1957 memcpy(chunk
->stripes
[index
].dev_uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
1959 list_move(&devext
->chunk_list
, &chunk
->dextents
);
1964 static inline int count_devext_records(struct list_head
*record_list
)
1966 int num_of_records
= 0;
1967 struct device_extent_record
*devext
;
1969 list_for_each_entry(devext
, record_list
, chunk_list
)
1972 return num_of_records
;
1975 static int fill_chunk_up(struct chunk_record
*chunk
, struct list_head
*devexts
,
1976 struct recover_control
*rc
)
1981 for (i
= 0; i
< chunk
->num_stripes
; i
++) {
1982 if (!chunk
->stripes
[i
].devid
) {
1983 ret
= insert_stripe(devexts
, rc
, chunk
, i
);
1992 #define EQUAL_STRIPE (1 << 0)
1994 static int rebuild_raid_data_chunk_stripes(struct recover_control
*rc
,
1995 struct btrfs_root
*root
,
1996 struct chunk_record
*chunk
,
2002 struct btrfs_path path
;
2003 struct btrfs_key prev_key
;
2004 struct btrfs_key key
;
2005 struct btrfs_root
*csum_root
;
2006 struct extent_buffer
*leaf
;
2007 struct device_extent_record
*devext
;
2008 struct device_extent_record
*next
;
2009 struct btrfs_device
*dev
;
2010 u64 start
= chunk
->offset
;
2011 u64 end
= start
+ chunk
->stripe_len
;
2012 u64 chunk_end
= chunk
->offset
+ chunk
->length
;
2013 u64 csum_offset
= 0;
2015 u32 blocksize
= root
->sectorsize
;
2018 int num_unordered
= 0;
2019 LIST_HEAD(unordered
);
2020 LIST_HEAD(candidates
);
2022 csum_root
= root
->fs_info
->csum_root
;
2023 btrfs_init_path(&path
);
2024 list_splice_init(&chunk
->dextents
, &candidates
);
2026 if (list_is_last(candidates
.next
, &candidates
))
2029 key
.objectid
= BTRFS_EXTENT_CSUM_OBJECTID
;
2030 key
.type
= BTRFS_EXTENT_CSUM_KEY
;
2033 ret
= btrfs_search_slot(NULL
, csum_root
, &key
, &path
, 0, 0);
2035 fprintf(stderr
, "Search csum failed(%d)\n", ret
);
2038 leaf
= path
.nodes
[0];
2039 slot
= path
.slots
[0];
2041 if (slot
>= btrfs_header_nritems(leaf
)) {
2042 ret
= btrfs_next_leaf(csum_root
, &path
);
2045 "Walk tree failed(%d)\n", ret
);
2047 } else if (ret
> 0) {
2048 slot
= btrfs_header_nritems(leaf
) - 1;
2049 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2050 if (item_end_offset(root
, &key
, leaf
, slot
)
2052 csum_offset
= start
- key
.offset
;
2053 csum_offset
/= blocksize
;
2058 leaf
= path
.nodes
[0];
2059 slot
= path
.slots
[0];
2061 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2062 ret
= btrfs_previous_item(csum_root
, &path
, 0,
2063 BTRFS_EXTENT_CSUM_KEY
);
2067 if (key
.offset
>= end
)
2072 leaf
= path
.nodes
[0];
2073 slot
= path
.slots
[0];
2075 btrfs_item_key_to_cpu(leaf
, &prev_key
, slot
);
2076 if (item_end_offset(root
, &prev_key
, leaf
, slot
) > start
) {
2077 csum_offset
= start
- prev_key
.offset
;
2078 csum_offset
/= blocksize
;
2079 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2081 if (key
.offset
>= end
)
2085 if (key
.offset
+ csum_offset
* blocksize
> chunk_end
)
2089 ret
= next_csum(root
, &leaf
, &path
, &slot
, &csum_offset
, &tree_csum
,
2092 fprintf(stderr
, "Fetch csum failed\n");
2094 } else if (ret
== 1) {
2095 if (!(*flags
& EQUAL_STRIPE
))
2096 *flags
|= EQUAL_STRIPE
;
2098 } else if (ret
== 2)
2101 list_for_each_entry_safe(devext
, next
, &candidates
, chunk_list
) {
2102 data_offset
= calc_data_offset(&key
, chunk
, devext
->offset
,
2103 csum_offset
, blocksize
);
2104 dev
= btrfs_find_device_by_devid(rc
->fs_devices
,
2105 devext
->objectid
, 0);
2110 BUG_ON(btrfs_find_device_by_devid(rc
->fs_devices
,
2111 devext
->objectid
, 1));
2113 ret
= check_one_csum(dev
->fd
, data_offset
, blocksize
,
2118 list_move(&devext
->chunk_list
, &unordered
);
2121 if (list_empty(&candidates
)) {
2122 num_unordered
= count_devext_records(&unordered
);
2123 if (chunk
->type_flags
& BTRFS_BLOCK_GROUP_RAID6
2124 && num_unordered
== 2) {
2125 btrfs_release_path(&path
);
2126 ret
= fill_chunk_up(chunk
, &unordered
, rc
);
2133 if (list_is_last(candidates
.next
, &candidates
)) {
2134 index
= btrfs_calc_stripe_index(chunk
,
2135 key
.offset
+ csum_offset
* blocksize
);
2136 BUG_ON(index
== -1);
2137 if (chunk
->stripes
[index
].devid
)
2139 ret
= insert_stripe(&candidates
, rc
, chunk
, index
);
2147 start
= btrfs_next_stripe_logical_offset(chunk
, start
);
2148 end
= min(start
+ chunk
->stripe_len
, chunk_end
);
2149 list_splice_init(&unordered
, &candidates
);
2150 btrfs_release_path(&path
);
2152 if (end
< chunk_end
)
2156 list_splice_init(&candidates
, &unordered
);
2157 num_unordered
= count_devext_records(&unordered
);
2158 if (num_unordered
== 1) {
2159 for (i
= 0; i
< chunk
->num_stripes
; i
++) {
2160 if (!chunk
->stripes
[i
].devid
) {
2165 ret
= insert_stripe(&unordered
, rc
, chunk
, index
);
2169 if ((num_unordered
== 2 && chunk
->type_flags
2170 & BTRFS_BLOCK_GROUP_RAID5
)
2171 || (num_unordered
== 3 && chunk
->type_flags
2172 & BTRFS_BLOCK_GROUP_RAID6
)) {
2173 ret
= fill_chunk_up(chunk
, &unordered
, rc
);
2177 ret
= !!ret
|| (list_empty(&unordered
) ? 0 : 1);
2178 list_splice_init(&candidates
, &chunk
->dextents
);
2179 list_splice_init(&unordered
, &chunk
->dextents
);
2180 btrfs_release_path(&path
);
2185 static int btrfs_rebuild_ordered_data_chunk_stripes(struct recover_control
*rc
,
2186 struct btrfs_root
*root
)
2188 struct chunk_record
*chunk
;
2189 struct chunk_record
*next
;
2194 list_for_each_entry_safe(chunk
, next
, &rc
->unrepaired_chunks
, list
) {
2195 if ((chunk
->type_flags
& BTRFS_BLOCK_GROUP_DATA
)
2196 && (chunk
->type_flags
& BTRFS_ORDERED_RAID
)) {
2198 err
= rebuild_raid_data_chunk_stripes(rc
, root
, chunk
,
2201 list_move(&chunk
->list
, &rc
->bad_chunks
);
2202 if (flags
& EQUAL_STRIPE
)
2204 "Failure: too many equal stripes in chunk[%llu %llu]\n",
2205 chunk
->offset
, chunk
->length
);
2209 list_move(&chunk
->list
, &rc
->good_chunks
);
2215 static int btrfs_recover_chunks(struct recover_control
*rc
)
2217 struct chunk_record
*chunk
;
2218 struct block_group_record
*bg
;
2219 struct block_group_record
*next
;
2220 LIST_HEAD(new_chunks
);
2225 /* create the chunk by block group */
2226 list_for_each_entry_safe(bg
, next
, &rc
->bg
.block_groups
, list
) {
2227 nstripes
= btrfs_get_device_extents(bg
->objectid
,
2228 &rc
->devext
.no_chunk_orphans
,
2230 chunk
= calloc(1, btrfs_chunk_record_size(nstripes
));
2233 INIT_LIST_HEAD(&chunk
->dextents
);
2235 chunk
->cache
.start
= bg
->objectid
;
2236 chunk
->cache
.size
= bg
->offset
;
2237 chunk
->objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2238 chunk
->type
= BTRFS_CHUNK_ITEM_KEY
;
2239 chunk
->offset
= bg
->objectid
;
2240 chunk
->generation
= bg
->generation
;
2241 chunk
->length
= bg
->offset
;
2242 chunk
->owner
= BTRFS_CHUNK_TREE_OBJECTID
;
2243 chunk
->stripe_len
= BTRFS_STRIPE_LEN
;
2244 chunk
->type_flags
= bg
->flags
;
2245 chunk
->io_width
= BTRFS_STRIPE_LEN
;
2246 chunk
->io_align
= BTRFS_STRIPE_LEN
;
2247 chunk
->sector_size
= rc
->sectorsize
;
2248 chunk
->sub_stripes
= calc_sub_nstripes(bg
->flags
);
2250 ret
= insert_cache_extent(&rc
->chunk
, &chunk
->cache
);
2251 if (ret
== -EEXIST
) {
2252 error("duplicate entry in cache start %llu size %llu",
2253 (unsigned long long)chunk
->cache
.start
,
2254 (unsigned long long)chunk
->cache
.size
);
2260 list_del_init(&bg
->list
);
2262 list_add_tail(&chunk
->list
, &rc
->bad_chunks
);
2266 list_splice_init(&devexts
, &chunk
->dextents
);
2268 ret
= btrfs_verify_device_extents(bg
, &devexts
, nstripes
);
2270 list_add_tail(&chunk
->list
, &rc
->bad_chunks
);
2274 chunk
->num_stripes
= nstripes
;
2275 ret
= btrfs_rebuild_chunk_stripes(rc
, chunk
);
2277 list_add_tail(&chunk
->list
, &rc
->unrepaired_chunks
);
2279 list_add_tail(&chunk
->list
, &rc
->bad_chunks
);
2281 list_add_tail(&chunk
->list
, &rc
->good_chunks
);
2284 * Don't worry about the lost orphan device extents, they don't
2285 * have its chunk and block group, they must be the old ones that
2291 static inline int is_chunk_overlap(struct chunk_record
*chunk1
,
2292 struct chunk_record
*chunk2
)
2294 if (chunk1
->offset
>= chunk2
->offset
+ chunk2
->length
||
2295 chunk1
->offset
+ chunk1
->length
<= chunk2
->offset
)
2300 /* Move invalid(overlap with good chunks) rebuild chunks to bad chunk list */
2301 static void validate_rebuild_chunks(struct recover_control
*rc
)
2303 struct chunk_record
*good
;
2304 struct chunk_record
*rebuild
;
2305 struct chunk_record
*tmp
;
2307 list_for_each_entry_safe(rebuild
, tmp
, &rc
->rebuild_chunks
, list
) {
2308 list_for_each_entry(good
, &rc
->good_chunks
, list
) {
2309 if (is_chunk_overlap(rebuild
, good
)) {
2310 list_move_tail(&rebuild
->list
,
2319 * Return 0 when successful, < 0 on error and > 0 if aborted by user
2321 int btrfs_recover_chunk_tree(char *path
, int verbose
, int yes
)
2324 struct btrfs_root
*root
= NULL
;
2325 struct btrfs_trans_handle
*trans
;
2326 struct recover_control rc
;
2328 init_recover_control(&rc
, verbose
, yes
);
2330 ret
= recover_prepare(&rc
, path
);
2332 fprintf(stderr
, "recover prepare error\n");
2336 ret
= scan_devices(&rc
);
2338 fprintf(stderr
, "scan chunk headers error\n");
2342 if (cache_tree_empty(&rc
.chunk
) &&
2343 cache_tree_empty(&rc
.bg
.tree
) &&
2344 cache_tree_empty(&rc
.devext
.tree
)) {
2345 fprintf(stderr
, "no recoverable chunk\n");
2349 print_scan_result(&rc
);
2351 ret
= check_chunks(&rc
.chunk
, &rc
.bg
, &rc
.devext
, &rc
.good_chunks
,
2352 &rc
.bad_chunks
, &rc
.rebuild_chunks
, 1);
2354 if (!list_empty(&rc
.bg
.block_groups
) ||
2355 !list_empty(&rc
.devext
.no_chunk_orphans
)) {
2356 ret
= btrfs_recover_chunks(&rc
);
2361 print_check_result(&rc
);
2362 printf("Check chunks successfully with no orphans\n");
2365 validate_rebuild_chunks(&rc
);
2366 print_check_result(&rc
);
2368 root
= open_ctree_with_broken_chunk(&rc
);
2370 fprintf(stderr
, "open with broken chunk error\n");
2371 ret
= PTR_ERR(root
);
2375 ret
= check_all_chunks_by_metadata(&rc
, root
);
2377 fprintf(stderr
, "The chunks in memory can not match the metadata of the fs. Repair failed.\n");
2378 goto fail_close_ctree
;
2381 ret
= btrfs_rebuild_ordered_data_chunk_stripes(&rc
, root
);
2383 fprintf(stderr
, "Failed to rebuild ordered chunk stripes.\n");
2384 goto fail_close_ctree
;
2388 ret
= ask_user("We are going to rebuild the chunk tree on disk, it might destroy the old metadata on the disk, Are you sure?");
2391 goto fail_close_ctree
;
2395 trans
= btrfs_start_transaction(root
, 1);
2396 ret
= remove_chunk_extent_item(trans
, &rc
, root
);
2399 ret
= rebuild_chunk_tree(trans
, &rc
, root
);
2402 ret
= rebuild_sys_array(&rc
, root
);
2405 ret
= rebuild_block_group(trans
, &rc
, root
);
2407 printf("Fail to rebuild block groups.\n");
2408 printf("Recommend to run 'btrfs check --init-extent-tree <dev>' after recovery\n");
2411 btrfs_commit_transaction(trans
, root
);
2415 free_recover_control(&rc
);