1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/signal.h>
8 #include <linux/pagemap.h>
9 #include <linux/writeback.h>
10 #include <linux/blkdev.h>
11 #include <linux/sort.h>
12 #include <linux/rcupdate.h>
13 #include <linux/kthread.h>
14 #include <linux/slab.h>
15 #include <linux/ratelimit.h>
16 #include <linux/percpu_counter.h>
17 #include <linux/lockdep.h>
18 #include <linux/crc32c.h>
20 #include "extent-tree.h"
23 #include "print-tree.h"
27 #include "free-space-cache.h"
28 #include "free-space-tree.h"
31 #include "ref-verify.h"
32 #include "space-info.h"
33 #include "block-rsv.h"
34 #include "delalloc-space.h"
36 #include "rcu-string.h"
38 #include "dev-replace.h"
40 #include "accessors.h"
41 #include "root-tree.h"
42 #include "file-item.h"
44 #include "tree-checker.h"
46 #undef SCRAMBLE_DELAYED_REFS
49 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
50 struct btrfs_delayed_ref_node
*node
, u64 parent
,
51 u64 root_objectid
, u64 owner_objectid
,
52 u64 owner_offset
, int refs_to_drop
,
53 struct btrfs_delayed_extent_op
*extra_op
);
54 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
55 struct extent_buffer
*leaf
,
56 struct btrfs_extent_item
*ei
);
57 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
58 u64 parent
, u64 root_objectid
,
59 u64 flags
, u64 owner
, u64 offset
,
60 struct btrfs_key
*ins
, int ref_mod
);
61 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
62 struct btrfs_delayed_ref_node
*node
,
63 struct btrfs_delayed_extent_op
*extent_op
);
64 static int find_next_key(struct btrfs_path
*path
, int level
,
65 struct btrfs_key
*key
);
67 static int block_group_bits(struct btrfs_block_group
*cache
, u64 bits
)
69 return (cache
->flags
& bits
) == bits
;
72 /* simple helper to search for an existing data extent at a given offset */
73 int btrfs_lookup_data_extent(struct btrfs_fs_info
*fs_info
, u64 start
, u64 len
)
75 struct btrfs_root
*root
= btrfs_extent_root(fs_info
, start
);
78 struct btrfs_path
*path
;
80 path
= btrfs_alloc_path();
86 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
87 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
88 btrfs_free_path(path
);
93 * helper function to lookup reference count and flags of a tree block.
95 * the head node for delayed ref is used to store the sum of all the
96 * reference count modifications queued up in the rbtree. the head
97 * node may also store the extent flags to set. This way you can check
98 * to see what the reference count and extent flags would be if all of
99 * the delayed refs are not processed.
101 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
102 struct btrfs_fs_info
*fs_info
, u64 bytenr
,
103 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
105 struct btrfs_root
*extent_root
;
106 struct btrfs_delayed_ref_head
*head
;
107 struct btrfs_delayed_ref_root
*delayed_refs
;
108 struct btrfs_path
*path
;
109 struct btrfs_extent_item
*ei
;
110 struct extent_buffer
*leaf
;
111 struct btrfs_key key
;
118 * If we don't have skinny metadata, don't bother doing anything
121 if (metadata
&& !btrfs_fs_incompat(fs_info
, SKINNY_METADATA
)) {
122 offset
= fs_info
->nodesize
;
126 path
= btrfs_alloc_path();
131 path
->skip_locking
= 1;
132 path
->search_commit_root
= 1;
136 key
.objectid
= bytenr
;
139 key
.type
= BTRFS_METADATA_ITEM_KEY
;
141 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
143 extent_root
= btrfs_extent_root(fs_info
, bytenr
);
144 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
148 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
149 if (path
->slots
[0]) {
151 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
153 if (key
.objectid
== bytenr
&&
154 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
155 key
.offset
== fs_info
->nodesize
)
161 leaf
= path
->nodes
[0];
162 item_size
= btrfs_item_size(leaf
, path
->slots
[0]);
163 if (item_size
>= sizeof(*ei
)) {
164 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
165 struct btrfs_extent_item
);
166 num_refs
= btrfs_extent_refs(leaf
, ei
);
167 extent_flags
= btrfs_extent_flags(leaf
, ei
);
171 "unexpected extent item size, has %u expect >= %zu",
172 item_size
, sizeof(*ei
));
174 btrfs_abort_transaction(trans
, ret
);
176 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
181 BUG_ON(num_refs
== 0);
191 delayed_refs
= &trans
->transaction
->delayed_refs
;
192 spin_lock(&delayed_refs
->lock
);
193 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
195 if (!mutex_trylock(&head
->mutex
)) {
196 refcount_inc(&head
->refs
);
197 spin_unlock(&delayed_refs
->lock
);
199 btrfs_release_path(path
);
202 * Mutex was contended, block until it's released and try
205 mutex_lock(&head
->mutex
);
206 mutex_unlock(&head
->mutex
);
207 btrfs_put_delayed_ref_head(head
);
210 spin_lock(&head
->lock
);
211 if (head
->extent_op
&& head
->extent_op
->update_flags
)
212 extent_flags
|= head
->extent_op
->flags_to_set
;
214 BUG_ON(num_refs
== 0);
216 num_refs
+= head
->ref_mod
;
217 spin_unlock(&head
->lock
);
218 mutex_unlock(&head
->mutex
);
220 spin_unlock(&delayed_refs
->lock
);
222 WARN_ON(num_refs
== 0);
226 *flags
= extent_flags
;
228 btrfs_free_path(path
);
233 * Back reference rules. Back refs have three main goals:
235 * 1) differentiate between all holders of references to an extent so that
236 * when a reference is dropped we can make sure it was a valid reference
237 * before freeing the extent.
239 * 2) Provide enough information to quickly find the holders of an extent
240 * if we notice a given block is corrupted or bad.
242 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
243 * maintenance. This is actually the same as #2, but with a slightly
244 * different use case.
246 * There are two kinds of back refs. The implicit back refs is optimized
247 * for pointers in non-shared tree blocks. For a given pointer in a block,
248 * back refs of this kind provide information about the block's owner tree
249 * and the pointer's key. These information allow us to find the block by
250 * b-tree searching. The full back refs is for pointers in tree blocks not
251 * referenced by their owner trees. The location of tree block is recorded
252 * in the back refs. Actually the full back refs is generic, and can be
253 * used in all cases the implicit back refs is used. The major shortcoming
254 * of the full back refs is its overhead. Every time a tree block gets
255 * COWed, we have to update back refs entry for all pointers in it.
257 * For a newly allocated tree block, we use implicit back refs for
258 * pointers in it. This means most tree related operations only involve
259 * implicit back refs. For a tree block created in old transaction, the
260 * only way to drop a reference to it is COW it. So we can detect the
261 * event that tree block loses its owner tree's reference and do the
262 * back refs conversion.
264 * When a tree block is COWed through a tree, there are four cases:
266 * The reference count of the block is one and the tree is the block's
267 * owner tree. Nothing to do in this case.
269 * The reference count of the block is one and the tree is not the
270 * block's owner tree. In this case, full back refs is used for pointers
271 * in the block. Remove these full back refs, add implicit back refs for
272 * every pointers in the new block.
274 * The reference count of the block is greater than one and the tree is
275 * the block's owner tree. In this case, implicit back refs is used for
276 * pointers in the block. Add full back refs for every pointers in the
277 * block, increase lower level extents' reference counts. The original
278 * implicit back refs are entailed to the new block.
280 * The reference count of the block is greater than one and the tree is
281 * not the block's owner tree. Add implicit back refs for every pointer in
282 * the new block, increase lower level extents' reference count.
284 * Back Reference Key composing:
286 * The key objectid corresponds to the first byte in the extent,
287 * The key type is used to differentiate between types of back refs.
288 * There are different meanings of the key offset for different types
291 * File extents can be referenced by:
293 * - multiple snapshots, subvolumes, or different generations in one subvol
294 * - different files inside a single subvolume
295 * - different offsets inside a file (bookend extents in file.c)
297 * The extent ref structure for the implicit back refs has fields for:
299 * - Objectid of the subvolume root
300 * - objectid of the file holding the reference
301 * - original offset in the file
302 * - how many bookend extents
304 * The key offset for the implicit back refs is hash of the first
307 * The extent ref structure for the full back refs has field for:
309 * - number of pointers in the tree leaf
311 * The key offset for the implicit back refs is the first byte of
314 * When a file extent is allocated, The implicit back refs is used.
315 * the fields are filled in:
317 * (root_key.objectid, inode objectid, offset in file, 1)
319 * When a file extent is removed file truncation, we find the
320 * corresponding implicit back refs and check the following fields:
322 * (btrfs_header_owner(leaf), inode objectid, offset in file)
324 * Btree extents can be referenced by:
326 * - Different subvolumes
328 * Both the implicit back refs and the full back refs for tree blocks
329 * only consist of key. The key offset for the implicit back refs is
330 * objectid of block's owner tree. The key offset for the full back refs
331 * is the first byte of parent block.
333 * When implicit back refs is used, information about the lowest key and
334 * level of the tree block are required. These information are stored in
335 * tree block info structure.
339 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
340 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
341 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
343 int btrfs_get_extent_inline_ref_type(const struct extent_buffer
*eb
,
344 struct btrfs_extent_inline_ref
*iref
,
345 enum btrfs_inline_ref_type is_data
)
347 int type
= btrfs_extent_inline_ref_type(eb
, iref
);
348 u64 offset
= btrfs_extent_inline_ref_offset(eb
, iref
);
350 if (type
== BTRFS_TREE_BLOCK_REF_KEY
||
351 type
== BTRFS_SHARED_BLOCK_REF_KEY
||
352 type
== BTRFS_SHARED_DATA_REF_KEY
||
353 type
== BTRFS_EXTENT_DATA_REF_KEY
) {
354 if (is_data
== BTRFS_REF_TYPE_BLOCK
) {
355 if (type
== BTRFS_TREE_BLOCK_REF_KEY
)
357 if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
360 * Every shared one has parent tree block,
361 * which must be aligned to sector size.
364 IS_ALIGNED(offset
, eb
->fs_info
->sectorsize
))
367 } else if (is_data
== BTRFS_REF_TYPE_DATA
) {
368 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
370 if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
373 * Every shared one has parent tree block,
374 * which must be aligned to sector size.
377 IS_ALIGNED(offset
, eb
->fs_info
->sectorsize
))
381 ASSERT(is_data
== BTRFS_REF_TYPE_ANY
);
387 btrfs_print_leaf(eb
);
388 btrfs_err(eb
->fs_info
,
389 "eb %llu iref 0x%lx invalid extent inline ref type %d",
390 eb
->start
, (unsigned long)iref
, type
);
392 return BTRFS_REF_TYPE_INVALID
;
395 u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
397 u32 high_crc
= ~(u32
)0;
398 u32 low_crc
= ~(u32
)0;
401 lenum
= cpu_to_le64(root_objectid
);
402 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
403 lenum
= cpu_to_le64(owner
);
404 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
405 lenum
= cpu_to_le64(offset
);
406 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
408 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
411 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
412 struct btrfs_extent_data_ref
*ref
)
414 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
415 btrfs_extent_data_ref_objectid(leaf
, ref
),
416 btrfs_extent_data_ref_offset(leaf
, ref
));
419 static int match_extent_data_ref(struct extent_buffer
*leaf
,
420 struct btrfs_extent_data_ref
*ref
,
421 u64 root_objectid
, u64 owner
, u64 offset
)
423 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
424 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
425 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
430 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
431 struct btrfs_path
*path
,
432 u64 bytenr
, u64 parent
,
434 u64 owner
, u64 offset
)
436 struct btrfs_root
*root
= btrfs_extent_root(trans
->fs_info
, bytenr
);
437 struct btrfs_key key
;
438 struct btrfs_extent_data_ref
*ref
;
439 struct extent_buffer
*leaf
;
445 key
.objectid
= bytenr
;
447 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
450 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
451 key
.offset
= hash_extent_data_ref(root_objectid
,
456 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
468 leaf
= path
->nodes
[0];
469 nritems
= btrfs_header_nritems(leaf
);
471 if (path
->slots
[0] >= nritems
) {
472 ret
= btrfs_next_leaf(root
, path
);
478 leaf
= path
->nodes
[0];
479 nritems
= btrfs_header_nritems(leaf
);
483 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
484 if (key
.objectid
!= bytenr
||
485 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
488 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
489 struct btrfs_extent_data_ref
);
491 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
494 btrfs_release_path(path
);
506 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
507 struct btrfs_path
*path
,
508 u64 bytenr
, u64 parent
,
509 u64 root_objectid
, u64 owner
,
510 u64 offset
, int refs_to_add
)
512 struct btrfs_root
*root
= btrfs_extent_root(trans
->fs_info
, bytenr
);
513 struct btrfs_key key
;
514 struct extent_buffer
*leaf
;
519 key
.objectid
= bytenr
;
521 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
523 size
= sizeof(struct btrfs_shared_data_ref
);
525 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
526 key
.offset
= hash_extent_data_ref(root_objectid
,
528 size
= sizeof(struct btrfs_extent_data_ref
);
531 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
532 if (ret
&& ret
!= -EEXIST
)
535 leaf
= path
->nodes
[0];
537 struct btrfs_shared_data_ref
*ref
;
538 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
539 struct btrfs_shared_data_ref
);
541 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
543 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
544 num_refs
+= refs_to_add
;
545 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
548 struct btrfs_extent_data_ref
*ref
;
549 while (ret
== -EEXIST
) {
550 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
551 struct btrfs_extent_data_ref
);
552 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
555 btrfs_release_path(path
);
557 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
559 if (ret
&& ret
!= -EEXIST
)
562 leaf
= path
->nodes
[0];
564 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
565 struct btrfs_extent_data_ref
);
567 btrfs_set_extent_data_ref_root(leaf
, ref
,
569 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
570 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
571 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
573 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
574 num_refs
+= refs_to_add
;
575 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
578 btrfs_mark_buffer_dirty(leaf
);
581 btrfs_release_path(path
);
585 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
586 struct btrfs_root
*root
,
587 struct btrfs_path
*path
,
590 struct btrfs_key key
;
591 struct btrfs_extent_data_ref
*ref1
= NULL
;
592 struct btrfs_shared_data_ref
*ref2
= NULL
;
593 struct extent_buffer
*leaf
;
597 leaf
= path
->nodes
[0];
598 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
600 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
601 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
602 struct btrfs_extent_data_ref
);
603 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
604 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
605 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
606 struct btrfs_shared_data_ref
);
607 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
609 btrfs_err(trans
->fs_info
,
610 "unrecognized backref key (%llu %u %llu)",
611 key
.objectid
, key
.type
, key
.offset
);
612 btrfs_abort_transaction(trans
, -EUCLEAN
);
616 BUG_ON(num_refs
< refs_to_drop
);
617 num_refs
-= refs_to_drop
;
620 ret
= btrfs_del_item(trans
, root
, path
);
622 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
623 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
624 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
625 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
626 btrfs_mark_buffer_dirty(leaf
);
631 static noinline u32
extent_data_ref_count(struct btrfs_path
*path
,
632 struct btrfs_extent_inline_ref
*iref
)
634 struct btrfs_key key
;
635 struct extent_buffer
*leaf
;
636 struct btrfs_extent_data_ref
*ref1
;
637 struct btrfs_shared_data_ref
*ref2
;
641 leaf
= path
->nodes
[0];
642 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
646 * If type is invalid, we should have bailed out earlier than
649 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_DATA
);
650 ASSERT(type
!= BTRFS_REF_TYPE_INVALID
);
651 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
652 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
653 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
655 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
656 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
658 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
659 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
660 struct btrfs_extent_data_ref
);
661 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
662 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
663 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
664 struct btrfs_shared_data_ref
);
665 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
672 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
673 struct btrfs_path
*path
,
674 u64 bytenr
, u64 parent
,
677 struct btrfs_root
*root
= btrfs_extent_root(trans
->fs_info
, bytenr
);
678 struct btrfs_key key
;
681 key
.objectid
= bytenr
;
683 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
686 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
687 key
.offset
= root_objectid
;
690 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
696 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
697 struct btrfs_path
*path
,
698 u64 bytenr
, u64 parent
,
701 struct btrfs_root
*root
= btrfs_extent_root(trans
->fs_info
, bytenr
);
702 struct btrfs_key key
;
705 key
.objectid
= bytenr
;
707 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
710 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
711 key
.offset
= root_objectid
;
714 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
715 btrfs_release_path(path
);
719 static inline int extent_ref_type(u64 parent
, u64 owner
)
722 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
724 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
726 type
= BTRFS_TREE_BLOCK_REF_KEY
;
729 type
= BTRFS_SHARED_DATA_REF_KEY
;
731 type
= BTRFS_EXTENT_DATA_REF_KEY
;
736 static int find_next_key(struct btrfs_path
*path
, int level
,
737 struct btrfs_key
*key
)
740 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
741 if (!path
->nodes
[level
])
743 if (path
->slots
[level
] + 1 >=
744 btrfs_header_nritems(path
->nodes
[level
]))
747 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
748 path
->slots
[level
] + 1);
750 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
751 path
->slots
[level
] + 1);
758 * look for inline back ref. if back ref is found, *ref_ret is set
759 * to the address of inline back ref, and 0 is returned.
761 * if back ref isn't found, *ref_ret is set to the address where it
762 * should be inserted, and -ENOENT is returned.
764 * if insert is true and there are too many inline back refs, the path
765 * points to the extent item, and -EAGAIN is returned.
767 * NOTE: inline back refs are ordered in the same way that back ref
768 * items in the tree are ordered.
770 static noinline_for_stack
771 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
772 struct btrfs_path
*path
,
773 struct btrfs_extent_inline_ref
**ref_ret
,
774 u64 bytenr
, u64 num_bytes
,
775 u64 parent
, u64 root_objectid
,
776 u64 owner
, u64 offset
, int insert
)
778 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
779 struct btrfs_root
*root
= btrfs_extent_root(fs_info
, bytenr
);
780 struct btrfs_key key
;
781 struct extent_buffer
*leaf
;
782 struct btrfs_extent_item
*ei
;
783 struct btrfs_extent_inline_ref
*iref
;
793 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
796 key
.objectid
= bytenr
;
797 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
798 key
.offset
= num_bytes
;
800 want
= extent_ref_type(parent
, owner
);
802 extra_size
= btrfs_extent_inline_ref_size(want
);
803 path
->search_for_extension
= 1;
804 path
->keep_locks
= 1;
809 * Owner is our level, so we can just add one to get the level for the
810 * block we are interested in.
812 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
813 key
.type
= BTRFS_METADATA_ITEM_KEY
;
818 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
825 * We may be a newly converted file system which still has the old fat
826 * extent entries for metadata, so try and see if we have one of those.
828 if (ret
> 0 && skinny_metadata
) {
829 skinny_metadata
= false;
830 if (path
->slots
[0]) {
832 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
834 if (key
.objectid
== bytenr
&&
835 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
836 key
.offset
== num_bytes
)
840 key
.objectid
= bytenr
;
841 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
842 key
.offset
= num_bytes
;
843 btrfs_release_path(path
);
848 if (ret
&& !insert
) {
851 } else if (WARN_ON(ret
)) {
852 btrfs_print_leaf(path
->nodes
[0]);
854 "extent item not found for insert, bytenr %llu num_bytes %llu parent %llu root_objectid %llu owner %llu offset %llu",
855 bytenr
, num_bytes
, parent
, root_objectid
, owner
,
861 leaf
= path
->nodes
[0];
862 item_size
= btrfs_item_size(leaf
, path
->slots
[0]);
863 if (unlikely(item_size
< sizeof(*ei
))) {
866 "unexpected extent item size, has %llu expect >= %zu",
867 item_size
, sizeof(*ei
));
868 btrfs_abort_transaction(trans
, err
);
872 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
873 flags
= btrfs_extent_flags(leaf
, ei
);
875 ptr
= (unsigned long)(ei
+ 1);
876 end
= (unsigned long)ei
+ item_size
;
878 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
879 ptr
+= sizeof(struct btrfs_tree_block_info
);
883 if (owner
>= BTRFS_FIRST_FREE_OBJECTID
)
884 needed
= BTRFS_REF_TYPE_DATA
;
886 needed
= BTRFS_REF_TYPE_BLOCK
;
893 btrfs_print_leaf(path
->nodes
[0]);
895 "overrun extent record at slot %d while looking for inline extent for root %llu owner %llu offset %llu parent %llu",
896 path
->slots
[0], root_objectid
, owner
, offset
, parent
);
900 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
901 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, needed
);
902 if (type
== BTRFS_REF_TYPE_INVALID
) {
910 ptr
+= btrfs_extent_inline_ref_size(type
);
914 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
915 struct btrfs_extent_data_ref
*dref
;
916 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
917 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
922 if (hash_extent_data_ref_item(leaf
, dref
) <
923 hash_extent_data_ref(root_objectid
, owner
, offset
))
927 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
929 if (parent
== ref_offset
) {
933 if (ref_offset
< parent
)
936 if (root_objectid
== ref_offset
) {
940 if (ref_offset
< root_objectid
)
944 ptr
+= btrfs_extent_inline_ref_size(type
);
946 if (err
== -ENOENT
&& insert
) {
947 if (item_size
+ extra_size
>=
948 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
953 * To add new inline back ref, we have to make sure
954 * there is no corresponding back ref item.
955 * For simplicity, we just do not add new inline back
956 * ref if there is any kind of item for this block
958 if (find_next_key(path
, 0, &key
) == 0 &&
959 key
.objectid
== bytenr
&&
960 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
965 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
968 path
->keep_locks
= 0;
969 path
->search_for_extension
= 0;
970 btrfs_unlock_up_safe(path
, 1);
976 * helper to add new inline back ref
978 static noinline_for_stack
979 void setup_inline_extent_backref(struct btrfs_fs_info
*fs_info
,
980 struct btrfs_path
*path
,
981 struct btrfs_extent_inline_ref
*iref
,
982 u64 parent
, u64 root_objectid
,
983 u64 owner
, u64 offset
, int refs_to_add
,
984 struct btrfs_delayed_extent_op
*extent_op
)
986 struct extent_buffer
*leaf
;
987 struct btrfs_extent_item
*ei
;
990 unsigned long item_offset
;
995 leaf
= path
->nodes
[0];
996 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
997 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
999 type
= extent_ref_type(parent
, owner
);
1000 size
= btrfs_extent_inline_ref_size(type
);
1002 btrfs_extend_item(path
, size
);
1004 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1005 refs
= btrfs_extent_refs(leaf
, ei
);
1006 refs
+= refs_to_add
;
1007 btrfs_set_extent_refs(leaf
, ei
, refs
);
1009 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1011 ptr
= (unsigned long)ei
+ item_offset
;
1012 end
= (unsigned long)ei
+ btrfs_item_size(leaf
, path
->slots
[0]);
1013 if (ptr
< end
- size
)
1014 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1017 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1018 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1019 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1020 struct btrfs_extent_data_ref
*dref
;
1021 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1022 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1023 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1024 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1025 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1026 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1027 struct btrfs_shared_data_ref
*sref
;
1028 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1029 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1030 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1031 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1032 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1034 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1036 btrfs_mark_buffer_dirty(leaf
);
1039 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1040 struct btrfs_path
*path
,
1041 struct btrfs_extent_inline_ref
**ref_ret
,
1042 u64 bytenr
, u64 num_bytes
, u64 parent
,
1043 u64 root_objectid
, u64 owner
, u64 offset
)
1047 ret
= lookup_inline_extent_backref(trans
, path
, ref_ret
, bytenr
,
1048 num_bytes
, parent
, root_objectid
,
1053 btrfs_release_path(path
);
1056 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1057 ret
= lookup_tree_block_ref(trans
, path
, bytenr
, parent
,
1060 ret
= lookup_extent_data_ref(trans
, path
, bytenr
, parent
,
1061 root_objectid
, owner
, offset
);
1067 * helper to update/remove inline back ref
1069 static noinline_for_stack
int update_inline_extent_backref(struct btrfs_path
*path
,
1070 struct btrfs_extent_inline_ref
*iref
,
1072 struct btrfs_delayed_extent_op
*extent_op
)
1074 struct extent_buffer
*leaf
= path
->nodes
[0];
1075 struct btrfs_fs_info
*fs_info
= leaf
->fs_info
;
1076 struct btrfs_extent_item
*ei
;
1077 struct btrfs_extent_data_ref
*dref
= NULL
;
1078 struct btrfs_shared_data_ref
*sref
= NULL
;
1086 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1087 refs
= btrfs_extent_refs(leaf
, ei
);
1088 if (unlikely(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0)) {
1089 struct btrfs_key key
;
1092 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1093 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
1094 extent_size
= fs_info
->nodesize
;
1096 extent_size
= key
.offset
;
1097 btrfs_print_leaf(leaf
);
1099 "invalid refs_to_mod for extent %llu num_bytes %u, has %d expect >= -%llu",
1100 key
.objectid
, extent_size
, refs_to_mod
, refs
);
1103 refs
+= refs_to_mod
;
1104 btrfs_set_extent_refs(leaf
, ei
, refs
);
1106 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1108 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_ANY
);
1110 * Function btrfs_get_extent_inline_ref_type() has already printed
1113 if (unlikely(type
== BTRFS_REF_TYPE_INVALID
))
1116 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1117 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1118 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1119 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1120 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1121 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1125 * For tree blocks we can only drop one ref for it, and tree
1126 * blocks should not have refs > 1.
1128 * Furthermore if we're inserting a new inline backref, we
1129 * won't reach this path either. That would be
1130 * setup_inline_extent_backref().
1132 if (unlikely(refs_to_mod
!= -1)) {
1133 struct btrfs_key key
;
1135 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1137 btrfs_print_leaf(leaf
);
1139 "invalid refs_to_mod for tree block %llu, has %d expect -1",
1140 key
.objectid
, refs_to_mod
);
1145 if (unlikely(refs_to_mod
< 0 && refs
< -refs_to_mod
)) {
1146 struct btrfs_key key
;
1149 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1150 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
1151 extent_size
= fs_info
->nodesize
;
1153 extent_size
= key
.offset
;
1154 btrfs_print_leaf(leaf
);
1156 "invalid refs_to_mod for backref entry, iref %lu extent %llu num_bytes %u, has %d expect >= -%llu",
1157 (unsigned long)iref
, key
.objectid
, extent_size
,
1161 refs
+= refs_to_mod
;
1164 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1165 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1167 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1169 size
= btrfs_extent_inline_ref_size(type
);
1170 item_size
= btrfs_item_size(leaf
, path
->slots
[0]);
1171 ptr
= (unsigned long)iref
;
1172 end
= (unsigned long)ei
+ item_size
;
1173 if (ptr
+ size
< end
)
1174 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1177 btrfs_truncate_item(path
, item_size
, 1);
1179 btrfs_mark_buffer_dirty(leaf
);
1183 static noinline_for_stack
1184 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1185 struct btrfs_path
*path
,
1186 u64 bytenr
, u64 num_bytes
, u64 parent
,
1187 u64 root_objectid
, u64 owner
,
1188 u64 offset
, int refs_to_add
,
1189 struct btrfs_delayed_extent_op
*extent_op
)
1191 struct btrfs_extent_inline_ref
*iref
;
1194 ret
= lookup_inline_extent_backref(trans
, path
, &iref
, bytenr
,
1195 num_bytes
, parent
, root_objectid
,
1199 * We're adding refs to a tree block we already own, this
1200 * should not happen at all.
1202 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1203 btrfs_print_leaf(path
->nodes
[0]);
1204 btrfs_crit(trans
->fs_info
,
1205 "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu slot %u",
1206 bytenr
, num_bytes
, root_objectid
, path
->slots
[0]);
1209 ret
= update_inline_extent_backref(path
, iref
, refs_to_add
, extent_op
);
1210 } else if (ret
== -ENOENT
) {
1211 setup_inline_extent_backref(trans
->fs_info
, path
, iref
, parent
,
1212 root_objectid
, owner
, offset
,
1213 refs_to_add
, extent_op
);
1219 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1220 struct btrfs_root
*root
,
1221 struct btrfs_path
*path
,
1222 struct btrfs_extent_inline_ref
*iref
,
1223 int refs_to_drop
, int is_data
)
1227 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1229 ret
= update_inline_extent_backref(path
, iref
, -refs_to_drop
, NULL
);
1231 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1233 ret
= btrfs_del_item(trans
, root
, path
);
1237 static int btrfs_issue_discard(struct block_device
*bdev
, u64 start
, u64 len
,
1238 u64
*discarded_bytes
)
1241 u64 bytes_left
, end
;
1242 u64 aligned_start
= ALIGN(start
, 1 << SECTOR_SHIFT
);
1244 if (WARN_ON(start
!= aligned_start
)) {
1245 len
-= aligned_start
- start
;
1246 len
= round_down(len
, 1 << SECTOR_SHIFT
);
1247 start
= aligned_start
;
1250 *discarded_bytes
= 0;
1258 /* Skip any superblocks on this device. */
1259 for (j
= 0; j
< BTRFS_SUPER_MIRROR_MAX
; j
++) {
1260 u64 sb_start
= btrfs_sb_offset(j
);
1261 u64 sb_end
= sb_start
+ BTRFS_SUPER_INFO_SIZE
;
1262 u64 size
= sb_start
- start
;
1264 if (!in_range(sb_start
, start
, bytes_left
) &&
1265 !in_range(sb_end
, start
, bytes_left
) &&
1266 !in_range(start
, sb_start
, BTRFS_SUPER_INFO_SIZE
))
1270 * Superblock spans beginning of range. Adjust start and
1273 if (sb_start
<= start
) {
1274 start
+= sb_end
- start
;
1279 bytes_left
= end
- start
;
1284 ret
= blkdev_issue_discard(bdev
, start
>> SECTOR_SHIFT
,
1285 size
>> SECTOR_SHIFT
,
1288 *discarded_bytes
+= size
;
1289 else if (ret
!= -EOPNOTSUPP
)
1298 bytes_left
= end
- start
;
1302 ret
= blkdev_issue_discard(bdev
, start
>> SECTOR_SHIFT
,
1303 bytes_left
>> SECTOR_SHIFT
,
1306 *discarded_bytes
+= bytes_left
;
1311 static int do_discard_extent(struct btrfs_discard_stripe
*stripe
, u64
*bytes
)
1313 struct btrfs_device
*dev
= stripe
->dev
;
1314 struct btrfs_fs_info
*fs_info
= dev
->fs_info
;
1315 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
1316 u64 phys
= stripe
->physical
;
1317 u64 len
= stripe
->length
;
1321 /* Zone reset on a zoned filesystem */
1322 if (btrfs_can_zone_reset(dev
, phys
, len
)) {
1325 ret
= btrfs_reset_device_zone(dev
, phys
, len
, &discarded
);
1329 if (!btrfs_dev_replace_is_ongoing(dev_replace
) ||
1330 dev
!= dev_replace
->srcdev
)
1333 src_disc
= discarded
;
1335 /* Send to replace target as well */
1336 ret
= btrfs_reset_device_zone(dev_replace
->tgtdev
, phys
, len
,
1338 discarded
+= src_disc
;
1339 } else if (bdev_max_discard_sectors(stripe
->dev
->bdev
)) {
1340 ret
= btrfs_issue_discard(dev
->bdev
, phys
, len
, &discarded
);
1351 int btrfs_discard_extent(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1352 u64 num_bytes
, u64
*actual_bytes
)
1355 u64 discarded_bytes
= 0;
1356 u64 end
= bytenr
+ num_bytes
;
1360 * Avoid races with device replace and make sure the devices in the
1361 * stripes don't go away while we are discarding.
1363 btrfs_bio_counter_inc_blocked(fs_info
);
1365 struct btrfs_discard_stripe
*stripes
;
1366 unsigned int num_stripes
;
1369 num_bytes
= end
- cur
;
1370 stripes
= btrfs_map_discard(fs_info
, cur
, &num_bytes
, &num_stripes
);
1371 if (IS_ERR(stripes
)) {
1372 ret
= PTR_ERR(stripes
);
1373 if (ret
== -EOPNOTSUPP
)
1378 for (i
= 0; i
< num_stripes
; i
++) {
1379 struct btrfs_discard_stripe
*stripe
= stripes
+ i
;
1382 if (!stripe
->dev
->bdev
) {
1383 ASSERT(btrfs_test_opt(fs_info
, DEGRADED
));
1387 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
,
1388 &stripe
->dev
->dev_state
))
1391 ret
= do_discard_extent(stripe
, &bytes
);
1394 * Keep going if discard is not supported by the
1397 if (ret
!= -EOPNOTSUPP
)
1401 discarded_bytes
+= bytes
;
1409 btrfs_bio_counter_dec(fs_info
);
1411 *actual_bytes
= discarded_bytes
;
1415 /* Can return -ENOMEM */
1416 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1417 struct btrfs_ref
*generic_ref
)
1419 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1422 ASSERT(generic_ref
->type
!= BTRFS_REF_NOT_SET
&&
1423 generic_ref
->action
);
1424 BUG_ON(generic_ref
->type
== BTRFS_REF_METADATA
&&
1425 generic_ref
->tree_ref
.owning_root
== BTRFS_TREE_LOG_OBJECTID
);
1427 if (generic_ref
->type
== BTRFS_REF_METADATA
)
1428 ret
= btrfs_add_delayed_tree_ref(trans
, generic_ref
, NULL
);
1430 ret
= btrfs_add_delayed_data_ref(trans
, generic_ref
, 0);
1432 btrfs_ref_tree_mod(fs_info
, generic_ref
);
1438 * __btrfs_inc_extent_ref - insert backreference for a given extent
1440 * The counterpart is in __btrfs_free_extent(), with examples and more details
1443 * @trans: Handle of transaction
1445 * @node: The delayed ref node used to get the bytenr/length for
1446 * extent whose references are incremented.
1448 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1449 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1450 * bytenr of the parent block. Since new extents are always
1451 * created with indirect references, this will only be the case
1452 * when relocating a shared extent. In that case, root_objectid
1453 * will be BTRFS_TREE_RELOC_OBJECTID. Otherwise, parent must
1456 * @root_objectid: The id of the root where this modification has originated,
1457 * this can be either one of the well-known metadata trees or
1458 * the subvolume id which references this extent.
1460 * @owner: For data extents it is the inode number of the owning file.
1461 * For metadata extents this parameter holds the level in the
1462 * tree of the extent.
1464 * @offset: For metadata extents the offset is ignored and is currently
1465 * always passed as 0. For data extents it is the fileoffset
1466 * this extent belongs to.
1468 * @refs_to_add Number of references to add
1470 * @extent_op Pointer to a structure, holding information necessary when
1471 * updating a tree block's flags
1474 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1475 struct btrfs_delayed_ref_node
*node
,
1476 u64 parent
, u64 root_objectid
,
1477 u64 owner
, u64 offset
, int refs_to_add
,
1478 struct btrfs_delayed_extent_op
*extent_op
)
1480 struct btrfs_path
*path
;
1481 struct extent_buffer
*leaf
;
1482 struct btrfs_extent_item
*item
;
1483 struct btrfs_key key
;
1484 u64 bytenr
= node
->bytenr
;
1485 u64 num_bytes
= node
->num_bytes
;
1489 path
= btrfs_alloc_path();
1493 /* this will setup the path even if it fails to insert the back ref */
1494 ret
= insert_inline_extent_backref(trans
, path
, bytenr
, num_bytes
,
1495 parent
, root_objectid
, owner
,
1496 offset
, refs_to_add
, extent_op
);
1497 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
1501 * Ok we had -EAGAIN which means we didn't have space to insert and
1502 * inline extent ref, so just update the reference count and add a
1505 leaf
= path
->nodes
[0];
1506 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1507 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1508 refs
= btrfs_extent_refs(leaf
, item
);
1509 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1511 __run_delayed_extent_op(extent_op
, leaf
, item
);
1513 btrfs_mark_buffer_dirty(leaf
);
1514 btrfs_release_path(path
);
1516 /* now insert the actual backref */
1517 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1518 ret
= insert_tree_block_ref(trans
, path
, bytenr
, parent
,
1521 ret
= insert_extent_data_ref(trans
, path
, bytenr
, parent
,
1522 root_objectid
, owner
, offset
,
1526 btrfs_abort_transaction(trans
, ret
);
1528 btrfs_free_path(path
);
1532 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1533 struct btrfs_delayed_ref_node
*node
,
1534 struct btrfs_delayed_extent_op
*extent_op
,
1535 bool insert_reserved
)
1538 struct btrfs_delayed_data_ref
*ref
;
1539 struct btrfs_key ins
;
1544 ins
.objectid
= node
->bytenr
;
1545 ins
.offset
= node
->num_bytes
;
1546 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1548 ref
= btrfs_delayed_node_to_data_ref(node
);
1549 trace_run_delayed_data_ref(trans
->fs_info
, node
, ref
, node
->action
);
1551 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1552 parent
= ref
->parent
;
1553 ref_root
= ref
->root
;
1555 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1557 flags
|= extent_op
->flags_to_set
;
1558 ret
= alloc_reserved_file_extent(trans
, parent
, ref_root
,
1559 flags
, ref
->objectid
,
1562 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1563 ret
= __btrfs_inc_extent_ref(trans
, node
, parent
, ref_root
,
1564 ref
->objectid
, ref
->offset
,
1565 node
->ref_mod
, extent_op
);
1566 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1567 ret
= __btrfs_free_extent(trans
, node
, parent
,
1568 ref_root
, ref
->objectid
,
1569 ref
->offset
, node
->ref_mod
,
1577 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1578 struct extent_buffer
*leaf
,
1579 struct btrfs_extent_item
*ei
)
1581 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1582 if (extent_op
->update_flags
) {
1583 flags
|= extent_op
->flags_to_set
;
1584 btrfs_set_extent_flags(leaf
, ei
, flags
);
1587 if (extent_op
->update_key
) {
1588 struct btrfs_tree_block_info
*bi
;
1589 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1590 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1591 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1595 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1596 struct btrfs_delayed_ref_head
*head
,
1597 struct btrfs_delayed_extent_op
*extent_op
)
1599 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1600 struct btrfs_root
*root
;
1601 struct btrfs_key key
;
1602 struct btrfs_path
*path
;
1603 struct btrfs_extent_item
*ei
;
1604 struct extent_buffer
*leaf
;
1610 if (TRANS_ABORTED(trans
))
1613 if (!btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
1616 path
= btrfs_alloc_path();
1620 key
.objectid
= head
->bytenr
;
1623 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1624 key
.offset
= extent_op
->level
;
1626 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1627 key
.offset
= head
->num_bytes
;
1630 root
= btrfs_extent_root(fs_info
, key
.objectid
);
1632 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1639 if (path
->slots
[0] > 0) {
1641 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1643 if (key
.objectid
== head
->bytenr
&&
1644 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1645 key
.offset
== head
->num_bytes
)
1649 btrfs_release_path(path
);
1652 key
.objectid
= head
->bytenr
;
1653 key
.offset
= head
->num_bytes
;
1654 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1660 "missing extent item for extent %llu num_bytes %llu level %d",
1661 head
->bytenr
, head
->num_bytes
, extent_op
->level
);
1666 leaf
= path
->nodes
[0];
1667 item_size
= btrfs_item_size(leaf
, path
->slots
[0]);
1669 if (unlikely(item_size
< sizeof(*ei
))) {
1672 "unexpected extent item size, has %u expect >= %zu",
1673 item_size
, sizeof(*ei
));
1674 btrfs_abort_transaction(trans
, err
);
1678 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1679 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1681 btrfs_mark_buffer_dirty(leaf
);
1683 btrfs_free_path(path
);
1687 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
1688 struct btrfs_delayed_ref_node
*node
,
1689 struct btrfs_delayed_extent_op
*extent_op
,
1690 bool insert_reserved
)
1693 struct btrfs_delayed_tree_ref
*ref
;
1697 ref
= btrfs_delayed_node_to_tree_ref(node
);
1698 trace_run_delayed_tree_ref(trans
->fs_info
, node
, ref
, node
->action
);
1700 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1701 parent
= ref
->parent
;
1702 ref_root
= ref
->root
;
1704 if (unlikely(node
->ref_mod
!= 1)) {
1705 btrfs_err(trans
->fs_info
,
1706 "btree block %llu has %d references rather than 1: action %d ref_root %llu parent %llu",
1707 node
->bytenr
, node
->ref_mod
, node
->action
, ref_root
,
1711 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1712 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
1713 ret
= alloc_reserved_tree_block(trans
, node
, extent_op
);
1714 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1715 ret
= __btrfs_inc_extent_ref(trans
, node
, parent
, ref_root
,
1716 ref
->level
, 0, 1, extent_op
);
1717 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1718 ret
= __btrfs_free_extent(trans
, node
, parent
, ref_root
,
1719 ref
->level
, 0, 1, extent_op
);
1726 /* helper function to actually process a single delayed ref entry */
1727 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
1728 struct btrfs_delayed_ref_node
*node
,
1729 struct btrfs_delayed_extent_op
*extent_op
,
1730 bool insert_reserved
)
1734 if (TRANS_ABORTED(trans
)) {
1735 if (insert_reserved
)
1736 btrfs_pin_extent(trans
, node
->bytenr
, node
->num_bytes
, 1);
1740 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
1741 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1742 ret
= run_delayed_tree_ref(trans
, node
, extent_op
,
1744 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
1745 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1746 ret
= run_delayed_data_ref(trans
, node
, extent_op
,
1750 if (ret
&& insert_reserved
)
1751 btrfs_pin_extent(trans
, node
->bytenr
, node
->num_bytes
, 1);
1753 btrfs_err(trans
->fs_info
,
1754 "failed to run delayed ref for logical %llu num_bytes %llu type %u action %u ref_mod %d: %d",
1755 node
->bytenr
, node
->num_bytes
, node
->type
,
1756 node
->action
, node
->ref_mod
, ret
);
1760 static inline struct btrfs_delayed_ref_node
*
1761 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
1763 struct btrfs_delayed_ref_node
*ref
;
1765 if (RB_EMPTY_ROOT(&head
->ref_tree
.rb_root
))
1769 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1770 * This is to prevent a ref count from going down to zero, which deletes
1771 * the extent item from the extent tree, when there still are references
1772 * to add, which would fail because they would not find the extent item.
1774 if (!list_empty(&head
->ref_add_list
))
1775 return list_first_entry(&head
->ref_add_list
,
1776 struct btrfs_delayed_ref_node
, add_list
);
1778 ref
= rb_entry(rb_first_cached(&head
->ref_tree
),
1779 struct btrfs_delayed_ref_node
, ref_node
);
1780 ASSERT(list_empty(&ref
->add_list
));
1784 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root
*delayed_refs
,
1785 struct btrfs_delayed_ref_head
*head
)
1787 spin_lock(&delayed_refs
->lock
);
1788 head
->processing
= false;
1789 delayed_refs
->num_heads_ready
++;
1790 spin_unlock(&delayed_refs
->lock
);
1791 btrfs_delayed_ref_unlock(head
);
1794 static struct btrfs_delayed_extent_op
*cleanup_extent_op(
1795 struct btrfs_delayed_ref_head
*head
)
1797 struct btrfs_delayed_extent_op
*extent_op
= head
->extent_op
;
1802 if (head
->must_insert_reserved
) {
1803 head
->extent_op
= NULL
;
1804 btrfs_free_delayed_extent_op(extent_op
);
1810 static int run_and_cleanup_extent_op(struct btrfs_trans_handle
*trans
,
1811 struct btrfs_delayed_ref_head
*head
)
1813 struct btrfs_delayed_extent_op
*extent_op
;
1816 extent_op
= cleanup_extent_op(head
);
1819 head
->extent_op
= NULL
;
1820 spin_unlock(&head
->lock
);
1821 ret
= run_delayed_extent_op(trans
, head
, extent_op
);
1822 btrfs_free_delayed_extent_op(extent_op
);
1823 return ret
? ret
: 1;
1826 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info
*fs_info
,
1827 struct btrfs_delayed_ref_root
*delayed_refs
,
1828 struct btrfs_delayed_ref_head
*head
)
1830 int nr_items
= 1; /* Dropping this ref head update. */
1833 * We had csum deletions accounted for in our delayed refs rsv, we need
1834 * to drop the csum leaves for this update from our delayed_refs_rsv.
1836 if (head
->total_ref_mod
< 0 && head
->is_data
) {
1837 spin_lock(&delayed_refs
->lock
);
1838 delayed_refs
->pending_csums
-= head
->num_bytes
;
1839 spin_unlock(&delayed_refs
->lock
);
1840 nr_items
+= btrfs_csum_bytes_to_leaves(fs_info
, head
->num_bytes
);
1843 btrfs_delayed_refs_rsv_release(fs_info
, nr_items
);
1846 static int cleanup_ref_head(struct btrfs_trans_handle
*trans
,
1847 struct btrfs_delayed_ref_head
*head
)
1850 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1851 struct btrfs_delayed_ref_root
*delayed_refs
;
1854 delayed_refs
= &trans
->transaction
->delayed_refs
;
1856 ret
= run_and_cleanup_extent_op(trans
, head
);
1858 unselect_delayed_ref_head(delayed_refs
, head
);
1859 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
1866 * Need to drop our head ref lock and re-acquire the delayed ref lock
1867 * and then re-check to make sure nobody got added.
1869 spin_unlock(&head
->lock
);
1870 spin_lock(&delayed_refs
->lock
);
1871 spin_lock(&head
->lock
);
1872 if (!RB_EMPTY_ROOT(&head
->ref_tree
.rb_root
) || head
->extent_op
) {
1873 spin_unlock(&head
->lock
);
1874 spin_unlock(&delayed_refs
->lock
);
1877 btrfs_delete_ref_head(delayed_refs
, head
);
1878 spin_unlock(&head
->lock
);
1879 spin_unlock(&delayed_refs
->lock
);
1881 if (head
->must_insert_reserved
) {
1882 btrfs_pin_extent(trans
, head
->bytenr
, head
->num_bytes
, 1);
1883 if (head
->is_data
) {
1884 struct btrfs_root
*csum_root
;
1886 csum_root
= btrfs_csum_root(fs_info
, head
->bytenr
);
1887 ret
= btrfs_del_csums(trans
, csum_root
, head
->bytenr
,
1892 btrfs_cleanup_ref_head_accounting(fs_info
, delayed_refs
, head
);
1894 trace_run_delayed_ref_head(fs_info
, head
, 0);
1895 btrfs_delayed_ref_unlock(head
);
1896 btrfs_put_delayed_ref_head(head
);
1900 static struct btrfs_delayed_ref_head
*btrfs_obtain_ref_head(
1901 struct btrfs_trans_handle
*trans
)
1903 struct btrfs_delayed_ref_root
*delayed_refs
=
1904 &trans
->transaction
->delayed_refs
;
1905 struct btrfs_delayed_ref_head
*head
= NULL
;
1908 spin_lock(&delayed_refs
->lock
);
1909 head
= btrfs_select_ref_head(delayed_refs
);
1911 spin_unlock(&delayed_refs
->lock
);
1916 * Grab the lock that says we are going to process all the refs for
1919 ret
= btrfs_delayed_ref_lock(delayed_refs
, head
);
1920 spin_unlock(&delayed_refs
->lock
);
1923 * We may have dropped the spin lock to get the head mutex lock, and
1924 * that might have given someone else time to free the head. If that's
1925 * true, it has been removed from our list and we can move on.
1928 head
= ERR_PTR(-EAGAIN
);
1933 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle
*trans
,
1934 struct btrfs_delayed_ref_head
*locked_ref
)
1936 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1937 struct btrfs_delayed_ref_root
*delayed_refs
;
1938 struct btrfs_delayed_extent_op
*extent_op
;
1939 struct btrfs_delayed_ref_node
*ref
;
1940 bool must_insert_reserved
;
1943 delayed_refs
= &trans
->transaction
->delayed_refs
;
1945 lockdep_assert_held(&locked_ref
->mutex
);
1946 lockdep_assert_held(&locked_ref
->lock
);
1948 while ((ref
= select_delayed_ref(locked_ref
))) {
1950 btrfs_check_delayed_seq(fs_info
, ref
->seq
)) {
1951 spin_unlock(&locked_ref
->lock
);
1952 unselect_delayed_ref_head(delayed_refs
, locked_ref
);
1956 rb_erase_cached(&ref
->ref_node
, &locked_ref
->ref_tree
);
1957 RB_CLEAR_NODE(&ref
->ref_node
);
1958 if (!list_empty(&ref
->add_list
))
1959 list_del(&ref
->add_list
);
1961 * When we play the delayed ref, also correct the ref_mod on
1964 switch (ref
->action
) {
1965 case BTRFS_ADD_DELAYED_REF
:
1966 case BTRFS_ADD_DELAYED_EXTENT
:
1967 locked_ref
->ref_mod
-= ref
->ref_mod
;
1969 case BTRFS_DROP_DELAYED_REF
:
1970 locked_ref
->ref_mod
+= ref
->ref_mod
;
1975 atomic_dec(&delayed_refs
->num_entries
);
1978 * Record the must_insert_reserved flag before we drop the
1981 must_insert_reserved
= locked_ref
->must_insert_reserved
;
1982 locked_ref
->must_insert_reserved
= false;
1984 extent_op
= locked_ref
->extent_op
;
1985 locked_ref
->extent_op
= NULL
;
1986 spin_unlock(&locked_ref
->lock
);
1988 ret
= run_one_delayed_ref(trans
, ref
, extent_op
,
1989 must_insert_reserved
);
1991 btrfs_free_delayed_extent_op(extent_op
);
1993 unselect_delayed_ref_head(delayed_refs
, locked_ref
);
1994 btrfs_put_delayed_ref(ref
);
1998 btrfs_put_delayed_ref(ref
);
2001 spin_lock(&locked_ref
->lock
);
2002 btrfs_merge_delayed_refs(fs_info
, delayed_refs
, locked_ref
);
2009 * Returns 0 on success or if called with an already aborted transaction.
2010 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2012 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2015 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2016 struct btrfs_delayed_ref_root
*delayed_refs
;
2017 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2019 unsigned long count
= 0;
2021 delayed_refs
= &trans
->transaction
->delayed_refs
;
2024 locked_ref
= btrfs_obtain_ref_head(trans
);
2025 if (IS_ERR_OR_NULL(locked_ref
)) {
2026 if (PTR_ERR(locked_ref
) == -EAGAIN
) {
2035 * We need to try and merge add/drops of the same ref since we
2036 * can run into issues with relocate dropping the implicit ref
2037 * and then it being added back again before the drop can
2038 * finish. If we merged anything we need to re-loop so we can
2040 * Or we can get node references of the same type that weren't
2041 * merged when created due to bumps in the tree mod seq, and
2042 * we need to merge them to prevent adding an inline extent
2043 * backref before dropping it (triggering a BUG_ON at
2044 * insert_inline_extent_backref()).
2046 spin_lock(&locked_ref
->lock
);
2047 btrfs_merge_delayed_refs(fs_info
, delayed_refs
, locked_ref
);
2049 ret
= btrfs_run_delayed_refs_for_head(trans
, locked_ref
);
2050 if (ret
< 0 && ret
!= -EAGAIN
) {
2052 * Error, btrfs_run_delayed_refs_for_head already
2053 * unlocked everything so just bail out
2058 * Success, perform the usual cleanup of a processed
2061 ret
= cleanup_ref_head(trans
, locked_ref
);
2063 /* We dropped our lock, we need to loop. */
2072 * Either success case or btrfs_run_delayed_refs_for_head
2073 * returned -EAGAIN, meaning we need to select another head
2078 } while ((nr
!= -1 && count
< nr
) || locked_ref
);
2083 #ifdef SCRAMBLE_DELAYED_REFS
2085 * Normally delayed refs get processed in ascending bytenr order. This
2086 * correlates in most cases to the order added. To expose dependencies on this
2087 * order, we start to process the tree in the middle instead of the beginning
2089 static u64
find_middle(struct rb_root
*root
)
2091 struct rb_node
*n
= root
->rb_node
;
2092 struct btrfs_delayed_ref_node
*entry
;
2095 u64 first
= 0, last
= 0;
2099 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2100 first
= entry
->bytenr
;
2104 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2105 last
= entry
->bytenr
;
2110 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2111 WARN_ON(!entry
->in_tree
);
2113 middle
= entry
->bytenr
;
2127 * this starts processing the delayed reference count updates and
2128 * extent insertions we have queued up so far. count can be
2129 * 0, which means to process everything in the tree at the start
2130 * of the run (but not newly added entries), or it can be some target
2131 * number you'd like to process.
2133 * Returns 0 on success or if called with an aborted transaction
2134 * Returns <0 on error and aborts the transaction
2136 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2137 unsigned long count
)
2139 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2140 struct rb_node
*node
;
2141 struct btrfs_delayed_ref_root
*delayed_refs
;
2142 struct btrfs_delayed_ref_head
*head
;
2144 int run_all
= count
== (unsigned long)-1;
2146 /* We'll clean this up in btrfs_cleanup_transaction */
2147 if (TRANS_ABORTED(trans
))
2150 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE
, &fs_info
->flags
))
2153 delayed_refs
= &trans
->transaction
->delayed_refs
;
2155 count
= delayed_refs
->num_heads_ready
;
2158 #ifdef SCRAMBLE_DELAYED_REFS
2159 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2161 ret
= __btrfs_run_delayed_refs(trans
, count
);
2163 btrfs_abort_transaction(trans
, ret
);
2168 btrfs_create_pending_block_groups(trans
);
2170 spin_lock(&delayed_refs
->lock
);
2171 node
= rb_first_cached(&delayed_refs
->href_root
);
2173 spin_unlock(&delayed_refs
->lock
);
2176 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2178 refcount_inc(&head
->refs
);
2179 spin_unlock(&delayed_refs
->lock
);
2181 /* Mutex was contended, block until it's released and retry. */
2182 mutex_lock(&head
->mutex
);
2183 mutex_unlock(&head
->mutex
);
2185 btrfs_put_delayed_ref_head(head
);
2193 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2194 struct extent_buffer
*eb
, u64 flags
)
2196 struct btrfs_delayed_extent_op
*extent_op
;
2197 int level
= btrfs_header_level(eb
);
2200 extent_op
= btrfs_alloc_delayed_extent_op();
2204 extent_op
->flags_to_set
= flags
;
2205 extent_op
->update_flags
= true;
2206 extent_op
->update_key
= false;
2207 extent_op
->level
= level
;
2209 ret
= btrfs_add_delayed_extent_op(trans
, eb
->start
, eb
->len
, extent_op
);
2211 btrfs_free_delayed_extent_op(extent_op
);
2215 static noinline
int check_delayed_ref(struct btrfs_root
*root
,
2216 struct btrfs_path
*path
,
2217 u64 objectid
, u64 offset
, u64 bytenr
)
2219 struct btrfs_delayed_ref_head
*head
;
2220 struct btrfs_delayed_ref_node
*ref
;
2221 struct btrfs_delayed_data_ref
*data_ref
;
2222 struct btrfs_delayed_ref_root
*delayed_refs
;
2223 struct btrfs_transaction
*cur_trans
;
2224 struct rb_node
*node
;
2227 spin_lock(&root
->fs_info
->trans_lock
);
2228 cur_trans
= root
->fs_info
->running_transaction
;
2230 refcount_inc(&cur_trans
->use_count
);
2231 spin_unlock(&root
->fs_info
->trans_lock
);
2235 delayed_refs
= &cur_trans
->delayed_refs
;
2236 spin_lock(&delayed_refs
->lock
);
2237 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
2239 spin_unlock(&delayed_refs
->lock
);
2240 btrfs_put_transaction(cur_trans
);
2244 if (!mutex_trylock(&head
->mutex
)) {
2246 spin_unlock(&delayed_refs
->lock
);
2247 btrfs_put_transaction(cur_trans
);
2251 refcount_inc(&head
->refs
);
2252 spin_unlock(&delayed_refs
->lock
);
2254 btrfs_release_path(path
);
2257 * Mutex was contended, block until it's released and let
2260 mutex_lock(&head
->mutex
);
2261 mutex_unlock(&head
->mutex
);
2262 btrfs_put_delayed_ref_head(head
);
2263 btrfs_put_transaction(cur_trans
);
2266 spin_unlock(&delayed_refs
->lock
);
2268 spin_lock(&head
->lock
);
2270 * XXX: We should replace this with a proper search function in the
2273 for (node
= rb_first_cached(&head
->ref_tree
); node
;
2274 node
= rb_next(node
)) {
2275 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, ref_node
);
2276 /* If it's a shared ref we know a cross reference exists */
2277 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
2282 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2285 * If our ref doesn't match the one we're currently looking at
2286 * then we have a cross reference.
2288 if (data_ref
->root
!= root
->root_key
.objectid
||
2289 data_ref
->objectid
!= objectid
||
2290 data_ref
->offset
!= offset
) {
2295 spin_unlock(&head
->lock
);
2296 mutex_unlock(&head
->mutex
);
2297 btrfs_put_transaction(cur_trans
);
2301 static noinline
int check_committed_ref(struct btrfs_root
*root
,
2302 struct btrfs_path
*path
,
2303 u64 objectid
, u64 offset
, u64 bytenr
,
2306 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2307 struct btrfs_root
*extent_root
= btrfs_extent_root(fs_info
, bytenr
);
2308 struct extent_buffer
*leaf
;
2309 struct btrfs_extent_data_ref
*ref
;
2310 struct btrfs_extent_inline_ref
*iref
;
2311 struct btrfs_extent_item
*ei
;
2312 struct btrfs_key key
;
2317 key
.objectid
= bytenr
;
2318 key
.offset
= (u64
)-1;
2319 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2321 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2324 BUG_ON(ret
== 0); /* Corruption */
2327 if (path
->slots
[0] == 0)
2331 leaf
= path
->nodes
[0];
2332 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2334 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2338 item_size
= btrfs_item_size(leaf
, path
->slots
[0]);
2339 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2341 /* If extent item has more than 1 inline ref then it's shared */
2342 if (item_size
!= sizeof(*ei
) +
2343 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2347 * If extent created before last snapshot => it's shared unless the
2348 * snapshot has been deleted. Use the heuristic if strict is false.
2351 (btrfs_extent_generation(leaf
, ei
) <=
2352 btrfs_root_last_snapshot(&root
->root_item
)))
2355 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2357 /* If this extent has SHARED_DATA_REF then it's shared */
2358 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_DATA
);
2359 if (type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2362 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2363 if (btrfs_extent_refs(leaf
, ei
) !=
2364 btrfs_extent_data_ref_count(leaf
, ref
) ||
2365 btrfs_extent_data_ref_root(leaf
, ref
) !=
2366 root
->root_key
.objectid
||
2367 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2368 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2376 int btrfs_cross_ref_exist(struct btrfs_root
*root
, u64 objectid
, u64 offset
,
2377 u64 bytenr
, bool strict
, struct btrfs_path
*path
)
2382 ret
= check_committed_ref(root
, path
, objectid
,
2383 offset
, bytenr
, strict
);
2384 if (ret
&& ret
!= -ENOENT
)
2387 ret
= check_delayed_ref(root
, path
, objectid
, offset
, bytenr
);
2388 } while (ret
== -EAGAIN
);
2391 btrfs_release_path(path
);
2392 if (btrfs_is_data_reloc_root(root
))
2397 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2398 struct btrfs_root
*root
,
2399 struct extent_buffer
*buf
,
2400 int full_backref
, int inc
)
2402 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2408 struct btrfs_key key
;
2409 struct btrfs_file_extent_item
*fi
;
2410 struct btrfs_ref generic_ref
= { 0 };
2411 bool for_reloc
= btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
);
2417 if (btrfs_is_testing(fs_info
))
2420 ref_root
= btrfs_header_owner(buf
);
2421 nritems
= btrfs_header_nritems(buf
);
2422 level
= btrfs_header_level(buf
);
2424 if (!test_bit(BTRFS_ROOT_SHAREABLE
, &root
->state
) && level
== 0)
2428 parent
= buf
->start
;
2432 action
= BTRFS_ADD_DELAYED_REF
;
2434 action
= BTRFS_DROP_DELAYED_REF
;
2436 for (i
= 0; i
< nritems
; i
++) {
2438 btrfs_item_key_to_cpu(buf
, &key
, i
);
2439 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2441 fi
= btrfs_item_ptr(buf
, i
,
2442 struct btrfs_file_extent_item
);
2443 if (btrfs_file_extent_type(buf
, fi
) ==
2444 BTRFS_FILE_EXTENT_INLINE
)
2446 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2450 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2451 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2452 btrfs_init_generic_ref(&generic_ref
, action
, bytenr
,
2454 btrfs_init_data_ref(&generic_ref
, ref_root
, key
.objectid
,
2455 key
.offset
, root
->root_key
.objectid
,
2458 ret
= btrfs_inc_extent_ref(trans
, &generic_ref
);
2460 ret
= btrfs_free_extent(trans
, &generic_ref
);
2464 bytenr
= btrfs_node_blockptr(buf
, i
);
2465 num_bytes
= fs_info
->nodesize
;
2466 btrfs_init_generic_ref(&generic_ref
, action
, bytenr
,
2468 btrfs_init_tree_ref(&generic_ref
, level
- 1, ref_root
,
2469 root
->root_key
.objectid
, for_reloc
);
2471 ret
= btrfs_inc_extent_ref(trans
, &generic_ref
);
2473 ret
= btrfs_free_extent(trans
, &generic_ref
);
2483 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2484 struct extent_buffer
*buf
, int full_backref
)
2486 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2489 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2490 struct extent_buffer
*buf
, int full_backref
)
2492 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2495 static u64
get_alloc_profile_by_root(struct btrfs_root
*root
, int data
)
2497 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2502 flags
= BTRFS_BLOCK_GROUP_DATA
;
2503 else if (root
== fs_info
->chunk_root
)
2504 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
2506 flags
= BTRFS_BLOCK_GROUP_METADATA
;
2508 ret
= btrfs_get_alloc_profile(fs_info
, flags
);
2512 static u64
first_logical_byte(struct btrfs_fs_info
*fs_info
)
2514 struct rb_node
*leftmost
;
2517 read_lock(&fs_info
->block_group_cache_lock
);
2518 /* Get the block group with the lowest logical start address. */
2519 leftmost
= rb_first_cached(&fs_info
->block_group_cache_tree
);
2521 struct btrfs_block_group
*bg
;
2523 bg
= rb_entry(leftmost
, struct btrfs_block_group
, cache_node
);
2526 read_unlock(&fs_info
->block_group_cache_lock
);
2531 static int pin_down_extent(struct btrfs_trans_handle
*trans
,
2532 struct btrfs_block_group
*cache
,
2533 u64 bytenr
, u64 num_bytes
, int reserved
)
2535 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
2537 spin_lock(&cache
->space_info
->lock
);
2538 spin_lock(&cache
->lock
);
2539 cache
->pinned
+= num_bytes
;
2540 btrfs_space_info_update_bytes_pinned(fs_info
, cache
->space_info
,
2543 cache
->reserved
-= num_bytes
;
2544 cache
->space_info
->bytes_reserved
-= num_bytes
;
2546 spin_unlock(&cache
->lock
);
2547 spin_unlock(&cache
->space_info
->lock
);
2549 set_extent_bit(&trans
->transaction
->pinned_extents
, bytenr
,
2550 bytenr
+ num_bytes
- 1, EXTENT_DIRTY
, NULL
);
2554 int btrfs_pin_extent(struct btrfs_trans_handle
*trans
,
2555 u64 bytenr
, u64 num_bytes
, int reserved
)
2557 struct btrfs_block_group
*cache
;
2559 cache
= btrfs_lookup_block_group(trans
->fs_info
, bytenr
);
2560 BUG_ON(!cache
); /* Logic error */
2562 pin_down_extent(trans
, cache
, bytenr
, num_bytes
, reserved
);
2564 btrfs_put_block_group(cache
);
2569 * this function must be called within transaction
2571 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle
*trans
,
2572 u64 bytenr
, u64 num_bytes
)
2574 struct btrfs_block_group
*cache
;
2577 cache
= btrfs_lookup_block_group(trans
->fs_info
, bytenr
);
2582 * Fully cache the free space first so that our pin removes the free space
2585 ret
= btrfs_cache_block_group(cache
, true);
2589 pin_down_extent(trans
, cache
, bytenr
, num_bytes
, 0);
2591 /* remove us from the free space cache (if we're there at all) */
2592 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
2594 btrfs_put_block_group(cache
);
2598 static int __exclude_logged_extent(struct btrfs_fs_info
*fs_info
,
2599 u64 start
, u64 num_bytes
)
2602 struct btrfs_block_group
*block_group
;
2604 block_group
= btrfs_lookup_block_group(fs_info
, start
);
2608 ret
= btrfs_cache_block_group(block_group
, true);
2612 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
2614 btrfs_put_block_group(block_group
);
2618 int btrfs_exclude_logged_extents(struct extent_buffer
*eb
)
2620 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
2621 struct btrfs_file_extent_item
*item
;
2622 struct btrfs_key key
;
2627 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
))
2630 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
2631 btrfs_item_key_to_cpu(eb
, &key
, i
);
2632 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2634 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
2635 found_type
= btrfs_file_extent_type(eb
, item
);
2636 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
2638 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
2640 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
2641 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
2642 ret
= __exclude_logged_extent(fs_info
, key
.objectid
, key
.offset
);
2651 btrfs_inc_block_group_reservations(struct btrfs_block_group
*bg
)
2653 atomic_inc(&bg
->reservations
);
2657 * Returns the free cluster for the given space info and sets empty_cluster to
2658 * what it should be based on the mount options.
2660 static struct btrfs_free_cluster
*
2661 fetch_cluster_info(struct btrfs_fs_info
*fs_info
,
2662 struct btrfs_space_info
*space_info
, u64
*empty_cluster
)
2664 struct btrfs_free_cluster
*ret
= NULL
;
2667 if (btrfs_mixed_space_info(space_info
))
2670 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
2671 ret
= &fs_info
->meta_alloc_cluster
;
2672 if (btrfs_test_opt(fs_info
, SSD
))
2673 *empty_cluster
= SZ_2M
;
2675 *empty_cluster
= SZ_64K
;
2676 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) &&
2677 btrfs_test_opt(fs_info
, SSD_SPREAD
)) {
2678 *empty_cluster
= SZ_2M
;
2679 ret
= &fs_info
->data_alloc_cluster
;
2685 static int unpin_extent_range(struct btrfs_fs_info
*fs_info
,
2687 const bool return_free_space
)
2689 struct btrfs_block_group
*cache
= NULL
;
2690 struct btrfs_space_info
*space_info
;
2691 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
2692 struct btrfs_free_cluster
*cluster
= NULL
;
2694 u64 total_unpinned
= 0;
2695 u64 empty_cluster
= 0;
2698 while (start
<= end
) {
2701 start
>= cache
->start
+ cache
->length
) {
2703 btrfs_put_block_group(cache
);
2705 cache
= btrfs_lookup_block_group(fs_info
, start
);
2706 BUG_ON(!cache
); /* Logic error */
2708 cluster
= fetch_cluster_info(fs_info
,
2711 empty_cluster
<<= 1;
2714 len
= cache
->start
+ cache
->length
- start
;
2715 len
= min(len
, end
+ 1 - start
);
2717 if (return_free_space
)
2718 btrfs_add_free_space(cache
, start
, len
);
2721 total_unpinned
+= len
;
2722 space_info
= cache
->space_info
;
2725 * If this space cluster has been marked as fragmented and we've
2726 * unpinned enough in this block group to potentially allow a
2727 * cluster to be created inside of it go ahead and clear the
2730 if (cluster
&& cluster
->fragmented
&&
2731 total_unpinned
> empty_cluster
) {
2732 spin_lock(&cluster
->lock
);
2733 cluster
->fragmented
= 0;
2734 spin_unlock(&cluster
->lock
);
2737 spin_lock(&space_info
->lock
);
2738 spin_lock(&cache
->lock
);
2739 cache
->pinned
-= len
;
2740 btrfs_space_info_update_bytes_pinned(fs_info
, space_info
, -len
);
2741 space_info
->max_extent_size
= 0;
2743 space_info
->bytes_readonly
+= len
;
2745 } else if (btrfs_is_zoned(fs_info
)) {
2746 /* Need reset before reusing in a zoned block group */
2747 space_info
->bytes_zone_unusable
+= len
;
2750 spin_unlock(&cache
->lock
);
2751 if (!readonly
&& return_free_space
&&
2752 global_rsv
->space_info
== space_info
) {
2753 spin_lock(&global_rsv
->lock
);
2754 if (!global_rsv
->full
) {
2755 u64 to_add
= min(len
, global_rsv
->size
-
2756 global_rsv
->reserved
);
2758 global_rsv
->reserved
+= to_add
;
2759 btrfs_space_info_update_bytes_may_use(fs_info
,
2760 space_info
, to_add
);
2761 if (global_rsv
->reserved
>= global_rsv
->size
)
2762 global_rsv
->full
= 1;
2765 spin_unlock(&global_rsv
->lock
);
2767 /* Add to any tickets we may have */
2768 if (!readonly
&& return_free_space
&& len
)
2769 btrfs_try_granting_tickets(fs_info
, space_info
);
2770 spin_unlock(&space_info
->lock
);
2774 btrfs_put_block_group(cache
);
2778 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
)
2780 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2781 struct btrfs_block_group
*block_group
, *tmp
;
2782 struct list_head
*deleted_bgs
;
2783 struct extent_io_tree
*unpin
;
2788 unpin
= &trans
->transaction
->pinned_extents
;
2790 while (!TRANS_ABORTED(trans
)) {
2791 struct extent_state
*cached_state
= NULL
;
2793 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
2794 if (!find_first_extent_bit(unpin
, 0, &start
, &end
,
2795 EXTENT_DIRTY
, &cached_state
)) {
2796 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
2800 if (btrfs_test_opt(fs_info
, DISCARD_SYNC
))
2801 ret
= btrfs_discard_extent(fs_info
, start
,
2802 end
+ 1 - start
, NULL
);
2804 clear_extent_dirty(unpin
, start
, end
, &cached_state
);
2805 unpin_extent_range(fs_info
, start
, end
, true);
2806 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
2807 free_extent_state(cached_state
);
2811 if (btrfs_test_opt(fs_info
, DISCARD_ASYNC
)) {
2812 btrfs_discard_calc_delay(&fs_info
->discard_ctl
);
2813 btrfs_discard_schedule_work(&fs_info
->discard_ctl
, true);
2817 * Transaction is finished. We don't need the lock anymore. We
2818 * do need to clean up the block groups in case of a transaction
2821 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
2822 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
2826 if (!TRANS_ABORTED(trans
))
2827 ret
= btrfs_discard_extent(fs_info
,
2829 block_group
->length
,
2832 list_del_init(&block_group
->bg_list
);
2833 btrfs_unfreeze_block_group(block_group
);
2834 btrfs_put_block_group(block_group
);
2837 const char *errstr
= btrfs_decode_error(ret
);
2839 "discard failed while removing blockgroup: errno=%d %s",
2847 static int do_free_extent_accounting(struct btrfs_trans_handle
*trans
,
2848 u64 bytenr
, u64 num_bytes
, bool is_data
)
2853 struct btrfs_root
*csum_root
;
2855 csum_root
= btrfs_csum_root(trans
->fs_info
, bytenr
);
2856 ret
= btrfs_del_csums(trans
, csum_root
, bytenr
, num_bytes
);
2858 btrfs_abort_transaction(trans
, ret
);
2863 ret
= add_to_free_space_tree(trans
, bytenr
, num_bytes
);
2865 btrfs_abort_transaction(trans
, ret
);
2869 ret
= btrfs_update_block_group(trans
, bytenr
, num_bytes
, false);
2871 btrfs_abort_transaction(trans
, ret
);
2876 #define abort_and_dump(trans, path, fmt, args...) \
2878 btrfs_abort_transaction(trans, -EUCLEAN); \
2879 btrfs_print_leaf(path->nodes[0]); \
2880 btrfs_crit(trans->fs_info, fmt, ##args); \
2884 * Drop one or more refs of @node.
2886 * 1. Locate the extent refs.
2887 * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
2888 * Locate it, then reduce the refs number or remove the ref line completely.
2890 * 2. Update the refs count in EXTENT/METADATA_ITEM
2892 * Inline backref case:
2894 * in extent tree we have:
2896 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2897 * refs 2 gen 6 flags DATA
2898 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2899 * extent data backref root FS_TREE objectid 257 offset 0 count 1
2901 * This function gets called with:
2903 * node->bytenr = 13631488
2904 * node->num_bytes = 1048576
2905 * root_objectid = FS_TREE
2906 * owner_objectid = 257
2910 * Then we should get some like:
2912 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2913 * refs 1 gen 6 flags DATA
2914 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2916 * Keyed backref case:
2918 * in extent tree we have:
2920 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2921 * refs 754 gen 6 flags DATA
2923 * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
2924 * extent data backref root FS_TREE objectid 866 offset 0 count 1
2926 * This function get called with:
2928 * node->bytenr = 13631488
2929 * node->num_bytes = 1048576
2930 * root_objectid = FS_TREE
2931 * owner_objectid = 866
2935 * Then we should get some like:
2937 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2938 * refs 753 gen 6 flags DATA
2940 * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
2942 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
2943 struct btrfs_delayed_ref_node
*node
, u64 parent
,
2944 u64 root_objectid
, u64 owner_objectid
,
2945 u64 owner_offset
, int refs_to_drop
,
2946 struct btrfs_delayed_extent_op
*extent_op
)
2948 struct btrfs_fs_info
*info
= trans
->fs_info
;
2949 struct btrfs_key key
;
2950 struct btrfs_path
*path
;
2951 struct btrfs_root
*extent_root
;
2952 struct extent_buffer
*leaf
;
2953 struct btrfs_extent_item
*ei
;
2954 struct btrfs_extent_inline_ref
*iref
;
2957 int extent_slot
= 0;
2958 int found_extent
= 0;
2962 u64 bytenr
= node
->bytenr
;
2963 u64 num_bytes
= node
->num_bytes
;
2964 bool skinny_metadata
= btrfs_fs_incompat(info
, SKINNY_METADATA
);
2966 extent_root
= btrfs_extent_root(info
, bytenr
);
2967 ASSERT(extent_root
);
2969 path
= btrfs_alloc_path();
2973 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
2975 if (!is_data
&& refs_to_drop
!= 1) {
2977 "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
2978 node
->bytenr
, refs_to_drop
);
2980 btrfs_abort_transaction(trans
, ret
);
2985 skinny_metadata
= false;
2987 ret
= lookup_extent_backref(trans
, path
, &iref
, bytenr
, num_bytes
,
2988 parent
, root_objectid
, owner_objectid
,
2992 * Either the inline backref or the SHARED_DATA_REF/
2993 * SHARED_BLOCK_REF is found
2995 * Here is a quick path to locate EXTENT/METADATA_ITEM.
2996 * It's possible the EXTENT/METADATA_ITEM is near current slot.
2998 extent_slot
= path
->slots
[0];
2999 while (extent_slot
>= 0) {
3000 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
3002 if (key
.objectid
!= bytenr
)
3004 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
3005 key
.offset
== num_bytes
) {
3009 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
3010 key
.offset
== owner_objectid
) {
3015 /* Quick path didn't find the EXTEMT/METADATA_ITEM */
3016 if (path
->slots
[0] - extent_slot
> 5)
3021 if (!found_extent
) {
3023 abort_and_dump(trans
, path
,
3024 "invalid iref slot %u, no EXTENT/METADATA_ITEM found but has inline extent ref",
3029 /* Must be SHARED_* item, remove the backref first */
3030 ret
= remove_extent_backref(trans
, extent_root
, path
,
3031 NULL
, refs_to_drop
, is_data
);
3033 btrfs_abort_transaction(trans
, ret
);
3036 btrfs_release_path(path
);
3038 /* Slow path to locate EXTENT/METADATA_ITEM */
3039 key
.objectid
= bytenr
;
3040 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3041 key
.offset
= num_bytes
;
3043 if (!is_data
&& skinny_metadata
) {
3044 key
.type
= BTRFS_METADATA_ITEM_KEY
;
3045 key
.offset
= owner_objectid
;
3048 ret
= btrfs_search_slot(trans
, extent_root
,
3050 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
3052 * Couldn't find our skinny metadata item,
3053 * see if we have ye olde extent item.
3056 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
3058 if (key
.objectid
== bytenr
&&
3059 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
3060 key
.offset
== num_bytes
)
3064 if (ret
> 0 && skinny_metadata
) {
3065 skinny_metadata
= false;
3066 key
.objectid
= bytenr
;
3067 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3068 key
.offset
= num_bytes
;
3069 btrfs_release_path(path
);
3070 ret
= btrfs_search_slot(trans
, extent_root
,
3076 btrfs_print_leaf(path
->nodes
[0]);
3078 "umm, got %d back from search, was looking for %llu, slot %d",
3079 ret
, bytenr
, path
->slots
[0]);
3082 btrfs_abort_transaction(trans
, ret
);
3085 extent_slot
= path
->slots
[0];
3087 } else if (WARN_ON(ret
== -ENOENT
)) {
3088 abort_and_dump(trans
, path
,
3089 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu slot %d",
3090 bytenr
, parent
, root_objectid
, owner_objectid
,
3091 owner_offset
, path
->slots
[0]);
3094 btrfs_abort_transaction(trans
, ret
);
3098 leaf
= path
->nodes
[0];
3099 item_size
= btrfs_item_size(leaf
, extent_slot
);
3100 if (unlikely(item_size
< sizeof(*ei
))) {
3102 btrfs_err(trans
->fs_info
,
3103 "unexpected extent item size, has %u expect >= %zu",
3104 item_size
, sizeof(*ei
));
3105 btrfs_abort_transaction(trans
, ret
);
3108 ei
= btrfs_item_ptr(leaf
, extent_slot
,
3109 struct btrfs_extent_item
);
3110 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
3111 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
3112 struct btrfs_tree_block_info
*bi
;
3114 if (item_size
< sizeof(*ei
) + sizeof(*bi
)) {
3115 abort_and_dump(trans
, path
,
3116 "invalid extent item size for key (%llu, %u, %llu) slot %u owner %llu, has %u expect >= %zu",
3117 key
.objectid
, key
.type
, key
.offset
,
3118 path
->slots
[0], owner_objectid
, item_size
,
3119 sizeof(*ei
) + sizeof(*bi
));
3123 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
3124 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
3127 refs
= btrfs_extent_refs(leaf
, ei
);
3128 if (refs
< refs_to_drop
) {
3129 abort_and_dump(trans
, path
,
3130 "trying to drop %d refs but we only have %llu for bytenr %llu slot %u",
3131 refs_to_drop
, refs
, bytenr
, path
->slots
[0]);
3135 refs
-= refs_to_drop
;
3139 __run_delayed_extent_op(extent_op
, leaf
, ei
);
3141 * In the case of inline back ref, reference count will
3142 * be updated by remove_extent_backref
3145 if (!found_extent
) {
3146 abort_and_dump(trans
, path
,
3147 "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found, slot %u",
3153 btrfs_set_extent_refs(leaf
, ei
, refs
);
3154 btrfs_mark_buffer_dirty(leaf
);
3157 ret
= remove_extent_backref(trans
, extent_root
, path
,
3158 iref
, refs_to_drop
, is_data
);
3160 btrfs_abort_transaction(trans
, ret
);
3165 /* In this branch refs == 1 */
3167 if (is_data
&& refs_to_drop
!=
3168 extent_data_ref_count(path
, iref
)) {
3169 abort_and_dump(trans
, path
,
3170 "invalid refs_to_drop, current refs %u refs_to_drop %u slot %u",
3171 extent_data_ref_count(path
, iref
),
3172 refs_to_drop
, path
->slots
[0]);
3177 if (path
->slots
[0] != extent_slot
) {
3178 abort_and_dump(trans
, path
,
3179 "invalid iref, extent item key (%llu %u %llu) slot %u doesn't have wanted iref",
3180 key
.objectid
, key
.type
,
3181 key
.offset
, path
->slots
[0]);
3187 * No inline ref, we must be at SHARED_* item,
3188 * And it's single ref, it must be:
3189 * | extent_slot ||extent_slot + 1|
3190 * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3192 if (path
->slots
[0] != extent_slot
+ 1) {
3193 abort_and_dump(trans
, path
,
3194 "invalid SHARED_* item slot %u, previous item is not EXTENT/METADATA_ITEM",
3199 path
->slots
[0] = extent_slot
;
3204 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
3207 btrfs_abort_transaction(trans
, ret
);
3210 btrfs_release_path(path
);
3212 ret
= do_free_extent_accounting(trans
, bytenr
, num_bytes
, is_data
);
3214 btrfs_release_path(path
);
3217 btrfs_free_path(path
);
3222 * when we free an block, it is possible (and likely) that we free the last
3223 * delayed ref for that extent as well. This searches the delayed ref tree for
3224 * a given extent, and if there are no other delayed refs to be processed, it
3225 * removes it from the tree.
3227 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
3230 struct btrfs_delayed_ref_head
*head
;
3231 struct btrfs_delayed_ref_root
*delayed_refs
;
3234 delayed_refs
= &trans
->transaction
->delayed_refs
;
3235 spin_lock(&delayed_refs
->lock
);
3236 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
3238 goto out_delayed_unlock
;
3240 spin_lock(&head
->lock
);
3241 if (!RB_EMPTY_ROOT(&head
->ref_tree
.rb_root
))
3244 if (cleanup_extent_op(head
) != NULL
)
3248 * waiting for the lock here would deadlock. If someone else has it
3249 * locked they are already in the process of dropping it anyway
3251 if (!mutex_trylock(&head
->mutex
))
3254 btrfs_delete_ref_head(delayed_refs
, head
);
3255 head
->processing
= false;
3257 spin_unlock(&head
->lock
);
3258 spin_unlock(&delayed_refs
->lock
);
3260 BUG_ON(head
->extent_op
);
3261 if (head
->must_insert_reserved
)
3264 btrfs_cleanup_ref_head_accounting(trans
->fs_info
, delayed_refs
, head
);
3265 mutex_unlock(&head
->mutex
);
3266 btrfs_put_delayed_ref_head(head
);
3269 spin_unlock(&head
->lock
);
3272 spin_unlock(&delayed_refs
->lock
);
3276 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
3278 struct extent_buffer
*buf
,
3279 u64 parent
, int last_ref
)
3281 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3282 struct btrfs_ref generic_ref
= { 0 };
3285 btrfs_init_generic_ref(&generic_ref
, BTRFS_DROP_DELAYED_REF
,
3286 buf
->start
, buf
->len
, parent
);
3287 btrfs_init_tree_ref(&generic_ref
, btrfs_header_level(buf
),
3290 if (root_id
!= BTRFS_TREE_LOG_OBJECTID
) {
3291 btrfs_ref_tree_mod(fs_info
, &generic_ref
);
3292 ret
= btrfs_add_delayed_tree_ref(trans
, &generic_ref
, NULL
);
3293 BUG_ON(ret
); /* -ENOMEM */
3296 if (last_ref
&& btrfs_header_generation(buf
) == trans
->transid
) {
3297 struct btrfs_block_group
*cache
;
3298 bool must_pin
= false;
3300 if (root_id
!= BTRFS_TREE_LOG_OBJECTID
) {
3301 ret
= check_ref_cleanup(trans
, buf
->start
);
3303 btrfs_redirty_list_add(trans
->transaction
, buf
);
3308 cache
= btrfs_lookup_block_group(fs_info
, buf
->start
);
3310 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
3311 pin_down_extent(trans
, cache
, buf
->start
, buf
->len
, 1);
3312 btrfs_put_block_group(cache
);
3317 * If there are tree mod log users we may have recorded mod log
3318 * operations for this node. If we re-allocate this node we
3319 * could replay operations on this node that happened when it
3320 * existed in a completely different root. For example if it
3321 * was part of root A, then was reallocated to root B, and we
3322 * are doing a btrfs_old_search_slot(root b), we could replay
3323 * operations that happened when the block was part of root A,
3324 * giving us an inconsistent view of the btree.
3326 * We are safe from races here because at this point no other
3327 * node or root points to this extent buffer, so if after this
3328 * check a new tree mod log user joins we will not have an
3329 * existing log of operations on this node that we have to
3332 if (test_bit(BTRFS_FS_TREE_MOD_LOG_USERS
, &fs_info
->flags
))
3335 if (must_pin
|| btrfs_is_zoned(fs_info
)) {
3336 btrfs_redirty_list_add(trans
->transaction
, buf
);
3337 pin_down_extent(trans
, cache
, buf
->start
, buf
->len
, 1);
3338 btrfs_put_block_group(cache
);
3342 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
3344 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
3345 btrfs_free_reserved_bytes(cache
, buf
->len
, 0);
3346 btrfs_put_block_group(cache
);
3347 trace_btrfs_reserved_extent_free(fs_info
, buf
->start
, buf
->len
);
3352 * Deleting the buffer, clear the corrupt flag since it doesn't
3355 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
3359 /* Can return -ENOMEM */
3360 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_ref
*ref
)
3362 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3365 if (btrfs_is_testing(fs_info
))
3369 * tree log blocks never actually go into the extent allocation
3370 * tree, just update pinning info and exit early.
3372 if ((ref
->type
== BTRFS_REF_METADATA
&&
3373 ref
->tree_ref
.owning_root
== BTRFS_TREE_LOG_OBJECTID
) ||
3374 (ref
->type
== BTRFS_REF_DATA
&&
3375 ref
->data_ref
.owning_root
== BTRFS_TREE_LOG_OBJECTID
)) {
3376 /* unlocks the pinned mutex */
3377 btrfs_pin_extent(trans
, ref
->bytenr
, ref
->len
, 1);
3379 } else if (ref
->type
== BTRFS_REF_METADATA
) {
3380 ret
= btrfs_add_delayed_tree_ref(trans
, ref
, NULL
);
3382 ret
= btrfs_add_delayed_data_ref(trans
, ref
, 0);
3385 if (!((ref
->type
== BTRFS_REF_METADATA
&&
3386 ref
->tree_ref
.owning_root
== BTRFS_TREE_LOG_OBJECTID
) ||
3387 (ref
->type
== BTRFS_REF_DATA
&&
3388 ref
->data_ref
.owning_root
== BTRFS_TREE_LOG_OBJECTID
)))
3389 btrfs_ref_tree_mod(fs_info
, ref
);
3394 enum btrfs_loop_type
{
3396 * Start caching block groups but do not wait for progress or for them
3399 LOOP_CACHING_NOWAIT
,
3402 * Wait for the block group free_space >= the space we're waiting for if
3403 * the block group isn't cached.
3408 * Allow allocations to happen from block groups that do not yet have a
3409 * size classification.
3411 LOOP_UNSET_SIZE_CLASS
,
3414 * Allocate a chunk and then retry the allocation.
3419 * Ignore the size class restrictions for this allocation.
3421 LOOP_WRONG_SIZE_CLASS
,
3424 * Ignore the empty size, only try to allocate the number of bytes
3425 * needed for this allocation.
3431 btrfs_lock_block_group(struct btrfs_block_group
*cache
,
3435 down_read(&cache
->data_rwsem
);
3438 static inline void btrfs_grab_block_group(struct btrfs_block_group
*cache
,
3441 btrfs_get_block_group(cache
);
3443 down_read(&cache
->data_rwsem
);
3446 static struct btrfs_block_group
*btrfs_lock_cluster(
3447 struct btrfs_block_group
*block_group
,
3448 struct btrfs_free_cluster
*cluster
,
3450 __acquires(&cluster
->refill_lock
)
3452 struct btrfs_block_group
*used_bg
= NULL
;
3454 spin_lock(&cluster
->refill_lock
);
3456 used_bg
= cluster
->block_group
;
3460 if (used_bg
== block_group
)
3463 btrfs_get_block_group(used_bg
);
3468 if (down_read_trylock(&used_bg
->data_rwsem
))
3471 spin_unlock(&cluster
->refill_lock
);
3473 /* We should only have one-level nested. */
3474 down_read_nested(&used_bg
->data_rwsem
, SINGLE_DEPTH_NESTING
);
3476 spin_lock(&cluster
->refill_lock
);
3477 if (used_bg
== cluster
->block_group
)
3480 up_read(&used_bg
->data_rwsem
);
3481 btrfs_put_block_group(used_bg
);
3486 btrfs_release_block_group(struct btrfs_block_group
*cache
,
3490 up_read(&cache
->data_rwsem
);
3491 btrfs_put_block_group(cache
);
3495 * Helper function for find_free_extent().
3497 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3498 * Return >0 to inform caller that we find nothing
3499 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3501 static int find_free_extent_clustered(struct btrfs_block_group
*bg
,
3502 struct find_free_extent_ctl
*ffe_ctl
,
3503 struct btrfs_block_group
**cluster_bg_ret
)
3505 struct btrfs_block_group
*cluster_bg
;
3506 struct btrfs_free_cluster
*last_ptr
= ffe_ctl
->last_ptr
;
3507 u64 aligned_cluster
;
3511 cluster_bg
= btrfs_lock_cluster(bg
, last_ptr
, ffe_ctl
->delalloc
);
3513 goto refill_cluster
;
3514 if (cluster_bg
!= bg
&& (cluster_bg
->ro
||
3515 !block_group_bits(cluster_bg
, ffe_ctl
->flags
)))
3516 goto release_cluster
;
3518 offset
= btrfs_alloc_from_cluster(cluster_bg
, last_ptr
,
3519 ffe_ctl
->num_bytes
, cluster_bg
->start
,
3520 &ffe_ctl
->max_extent_size
);
3522 /* We have a block, we're done */
3523 spin_unlock(&last_ptr
->refill_lock
);
3524 trace_btrfs_reserve_extent_cluster(cluster_bg
, ffe_ctl
);
3525 *cluster_bg_ret
= cluster_bg
;
3526 ffe_ctl
->found_offset
= offset
;
3529 WARN_ON(last_ptr
->block_group
!= cluster_bg
);
3533 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3534 * lets just skip it and let the allocator find whatever block it can
3535 * find. If we reach this point, we will have tried the cluster
3536 * allocator plenty of times and not have found anything, so we are
3537 * likely way too fragmented for the clustering stuff to find anything.
3539 * However, if the cluster is taken from the current block group,
3540 * release the cluster first, so that we stand a better chance of
3541 * succeeding in the unclustered allocation.
3543 if (ffe_ctl
->loop
>= LOOP_NO_EMPTY_SIZE
&& cluster_bg
!= bg
) {
3544 spin_unlock(&last_ptr
->refill_lock
);
3545 btrfs_release_block_group(cluster_bg
, ffe_ctl
->delalloc
);
3549 /* This cluster didn't work out, free it and start over */
3550 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
3552 if (cluster_bg
!= bg
)
3553 btrfs_release_block_group(cluster_bg
, ffe_ctl
->delalloc
);
3556 if (ffe_ctl
->loop
>= LOOP_NO_EMPTY_SIZE
) {
3557 spin_unlock(&last_ptr
->refill_lock
);
3561 aligned_cluster
= max_t(u64
,
3562 ffe_ctl
->empty_cluster
+ ffe_ctl
->empty_size
,
3563 bg
->full_stripe_len
);
3564 ret
= btrfs_find_space_cluster(bg
, last_ptr
, ffe_ctl
->search_start
,
3565 ffe_ctl
->num_bytes
, aligned_cluster
);
3567 /* Now pull our allocation out of this cluster */
3568 offset
= btrfs_alloc_from_cluster(bg
, last_ptr
,
3569 ffe_ctl
->num_bytes
, ffe_ctl
->search_start
,
3570 &ffe_ctl
->max_extent_size
);
3572 /* We found one, proceed */
3573 spin_unlock(&last_ptr
->refill_lock
);
3574 ffe_ctl
->found_offset
= offset
;
3575 trace_btrfs_reserve_extent_cluster(bg
, ffe_ctl
);
3580 * At this point we either didn't find a cluster or we weren't able to
3581 * allocate a block from our cluster. Free the cluster we've been
3582 * trying to use, and go to the next block group.
3584 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
3585 spin_unlock(&last_ptr
->refill_lock
);
3590 * Return >0 to inform caller that we find nothing
3591 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3593 static int find_free_extent_unclustered(struct btrfs_block_group
*bg
,
3594 struct find_free_extent_ctl
*ffe_ctl
)
3596 struct btrfs_free_cluster
*last_ptr
= ffe_ctl
->last_ptr
;
3600 * We are doing an unclustered allocation, set the fragmented flag so
3601 * we don't bother trying to setup a cluster again until we get more
3604 if (unlikely(last_ptr
)) {
3605 spin_lock(&last_ptr
->lock
);
3606 last_ptr
->fragmented
= 1;
3607 spin_unlock(&last_ptr
->lock
);
3609 if (ffe_ctl
->cached
) {
3610 struct btrfs_free_space_ctl
*free_space_ctl
;
3612 free_space_ctl
= bg
->free_space_ctl
;
3613 spin_lock(&free_space_ctl
->tree_lock
);
3614 if (free_space_ctl
->free_space
<
3615 ffe_ctl
->num_bytes
+ ffe_ctl
->empty_cluster
+
3616 ffe_ctl
->empty_size
) {
3617 ffe_ctl
->total_free_space
= max_t(u64
,
3618 ffe_ctl
->total_free_space
,
3619 free_space_ctl
->free_space
);
3620 spin_unlock(&free_space_ctl
->tree_lock
);
3623 spin_unlock(&free_space_ctl
->tree_lock
);
3626 offset
= btrfs_find_space_for_alloc(bg
, ffe_ctl
->search_start
,
3627 ffe_ctl
->num_bytes
, ffe_ctl
->empty_size
,
3628 &ffe_ctl
->max_extent_size
);
3631 ffe_ctl
->found_offset
= offset
;
3635 static int do_allocation_clustered(struct btrfs_block_group
*block_group
,
3636 struct find_free_extent_ctl
*ffe_ctl
,
3637 struct btrfs_block_group
**bg_ret
)
3641 /* We want to try and use the cluster allocator, so lets look there */
3642 if (ffe_ctl
->last_ptr
&& ffe_ctl
->use_cluster
) {
3643 ret
= find_free_extent_clustered(block_group
, ffe_ctl
, bg_ret
);
3646 /* ret == -ENOENT case falls through */
3649 return find_free_extent_unclustered(block_group
, ffe_ctl
);
3653 * Tree-log block group locking
3654 * ============================
3656 * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3657 * indicates the starting address of a block group, which is reserved only
3658 * for tree-log metadata.
3665 * fs_info::treelog_bg_lock
3669 * Simple allocator for sequential-only block group. It only allows sequential
3670 * allocation. No need to play with trees. This function also reserves the
3671 * bytes as in btrfs_add_reserved_bytes.
3673 static int do_allocation_zoned(struct btrfs_block_group
*block_group
,
3674 struct find_free_extent_ctl
*ffe_ctl
,
3675 struct btrfs_block_group
**bg_ret
)
3677 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
3678 struct btrfs_space_info
*space_info
= block_group
->space_info
;
3679 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
3680 u64 start
= block_group
->start
;
3681 u64 num_bytes
= ffe_ctl
->num_bytes
;
3683 u64 bytenr
= block_group
->start
;
3685 u64 data_reloc_bytenr
;
3689 ASSERT(btrfs_is_zoned(block_group
->fs_info
));
3692 * Do not allow non-tree-log blocks in the dedicated tree-log block
3693 * group, and vice versa.
3695 spin_lock(&fs_info
->treelog_bg_lock
);
3696 log_bytenr
= fs_info
->treelog_bg
;
3697 if (log_bytenr
&& ((ffe_ctl
->for_treelog
&& bytenr
!= log_bytenr
) ||
3698 (!ffe_ctl
->for_treelog
&& bytenr
== log_bytenr
)))
3700 spin_unlock(&fs_info
->treelog_bg_lock
);
3705 * Do not allow non-relocation blocks in the dedicated relocation block
3706 * group, and vice versa.
3708 spin_lock(&fs_info
->relocation_bg_lock
);
3709 data_reloc_bytenr
= fs_info
->data_reloc_bg
;
3710 if (data_reloc_bytenr
&&
3711 ((ffe_ctl
->for_data_reloc
&& bytenr
!= data_reloc_bytenr
) ||
3712 (!ffe_ctl
->for_data_reloc
&& bytenr
== data_reloc_bytenr
)))
3714 spin_unlock(&fs_info
->relocation_bg_lock
);
3718 /* Check RO and no space case before trying to activate it */
3719 spin_lock(&block_group
->lock
);
3720 if (block_group
->ro
|| btrfs_zoned_bg_is_full(block_group
)) {
3723 * May need to clear fs_info->{treelog,data_reloc}_bg.
3724 * Return the error after taking the locks.
3727 spin_unlock(&block_group
->lock
);
3729 /* Metadata block group is activated at write time. */
3730 if (!ret
&& (block_group
->flags
& BTRFS_BLOCK_GROUP_DATA
) &&
3731 !btrfs_zone_activate(block_group
)) {
3734 * May need to clear fs_info->{treelog,data_reloc}_bg.
3735 * Return the error after taking the locks.
3739 spin_lock(&space_info
->lock
);
3740 spin_lock(&block_group
->lock
);
3741 spin_lock(&fs_info
->treelog_bg_lock
);
3742 spin_lock(&fs_info
->relocation_bg_lock
);
3747 ASSERT(!ffe_ctl
->for_treelog
||
3748 block_group
->start
== fs_info
->treelog_bg
||
3749 fs_info
->treelog_bg
== 0);
3750 ASSERT(!ffe_ctl
->for_data_reloc
||
3751 block_group
->start
== fs_info
->data_reloc_bg
||
3752 fs_info
->data_reloc_bg
== 0);
3754 if (block_group
->ro
||
3755 (!ffe_ctl
->for_data_reloc
&&
3756 test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC
, &block_group
->runtime_flags
))) {
3762 * Do not allow currently using block group to be tree-log dedicated
3765 if (ffe_ctl
->for_treelog
&& !fs_info
->treelog_bg
&&
3766 (block_group
->used
|| block_group
->reserved
)) {
3772 * Do not allow currently used block group to be the data relocation
3773 * dedicated block group.
3775 if (ffe_ctl
->for_data_reloc
&& !fs_info
->data_reloc_bg
&&
3776 (block_group
->used
|| block_group
->reserved
)) {
3781 WARN_ON_ONCE(block_group
->alloc_offset
> block_group
->zone_capacity
);
3782 avail
= block_group
->zone_capacity
- block_group
->alloc_offset
;
3783 if (avail
< num_bytes
) {
3784 if (ffe_ctl
->max_extent_size
< avail
) {
3786 * With sequential allocator, free space is always
3789 ffe_ctl
->max_extent_size
= avail
;
3790 ffe_ctl
->total_free_space
= avail
;
3796 if (ffe_ctl
->for_treelog
&& !fs_info
->treelog_bg
)
3797 fs_info
->treelog_bg
= block_group
->start
;
3799 if (ffe_ctl
->for_data_reloc
) {
3800 if (!fs_info
->data_reloc_bg
)
3801 fs_info
->data_reloc_bg
= block_group
->start
;
3803 * Do not allow allocations from this block group, unless it is
3804 * for data relocation. Compared to increasing the ->ro, setting
3805 * the ->zoned_data_reloc_ongoing flag still allows nocow
3806 * writers to come in. See btrfs_inc_nocow_writers().
3808 * We need to disable an allocation to avoid an allocation of
3809 * regular (non-relocation data) extent. With mix of relocation
3810 * extents and regular extents, we can dispatch WRITE commands
3811 * (for relocation extents) and ZONE APPEND commands (for
3812 * regular extents) at the same time to the same zone, which
3813 * easily break the write pointer.
3815 * Also, this flag avoids this block group to be zone finished.
3817 set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC
, &block_group
->runtime_flags
);
3820 ffe_ctl
->found_offset
= start
+ block_group
->alloc_offset
;
3821 block_group
->alloc_offset
+= num_bytes
;
3822 spin_lock(&ctl
->tree_lock
);
3823 ctl
->free_space
-= num_bytes
;
3824 spin_unlock(&ctl
->tree_lock
);
3827 * We do not check if found_offset is aligned to stripesize. The
3828 * address is anyway rewritten when using zone append writing.
3831 ffe_ctl
->search_start
= ffe_ctl
->found_offset
;
3834 if (ret
&& ffe_ctl
->for_treelog
)
3835 fs_info
->treelog_bg
= 0;
3836 if (ret
&& ffe_ctl
->for_data_reloc
)
3837 fs_info
->data_reloc_bg
= 0;
3838 spin_unlock(&fs_info
->relocation_bg_lock
);
3839 spin_unlock(&fs_info
->treelog_bg_lock
);
3840 spin_unlock(&block_group
->lock
);
3841 spin_unlock(&space_info
->lock
);
3845 static int do_allocation(struct btrfs_block_group
*block_group
,
3846 struct find_free_extent_ctl
*ffe_ctl
,
3847 struct btrfs_block_group
**bg_ret
)
3849 switch (ffe_ctl
->policy
) {
3850 case BTRFS_EXTENT_ALLOC_CLUSTERED
:
3851 return do_allocation_clustered(block_group
, ffe_ctl
, bg_ret
);
3852 case BTRFS_EXTENT_ALLOC_ZONED
:
3853 return do_allocation_zoned(block_group
, ffe_ctl
, bg_ret
);
3859 static void release_block_group(struct btrfs_block_group
*block_group
,
3860 struct find_free_extent_ctl
*ffe_ctl
,
3863 switch (ffe_ctl
->policy
) {
3864 case BTRFS_EXTENT_ALLOC_CLUSTERED
:
3865 ffe_ctl
->retry_uncached
= false;
3867 case BTRFS_EXTENT_ALLOC_ZONED
:
3874 BUG_ON(btrfs_bg_flags_to_raid_index(block_group
->flags
) !=
3876 btrfs_release_block_group(block_group
, delalloc
);
3879 static void found_extent_clustered(struct find_free_extent_ctl
*ffe_ctl
,
3880 struct btrfs_key
*ins
)
3882 struct btrfs_free_cluster
*last_ptr
= ffe_ctl
->last_ptr
;
3884 if (!ffe_ctl
->use_cluster
&& last_ptr
) {
3885 spin_lock(&last_ptr
->lock
);
3886 last_ptr
->window_start
= ins
->objectid
;
3887 spin_unlock(&last_ptr
->lock
);
3891 static void found_extent(struct find_free_extent_ctl
*ffe_ctl
,
3892 struct btrfs_key
*ins
)
3894 switch (ffe_ctl
->policy
) {
3895 case BTRFS_EXTENT_ALLOC_CLUSTERED
:
3896 found_extent_clustered(ffe_ctl
, ins
);
3898 case BTRFS_EXTENT_ALLOC_ZONED
:
3906 static int can_allocate_chunk_zoned(struct btrfs_fs_info
*fs_info
,
3907 struct find_free_extent_ctl
*ffe_ctl
)
3909 /* Block group's activeness is not a requirement for METADATA block groups. */
3910 if (!(ffe_ctl
->flags
& BTRFS_BLOCK_GROUP_DATA
))
3913 /* If we can activate new zone, just allocate a chunk and use it */
3914 if (btrfs_can_activate_zone(fs_info
->fs_devices
, ffe_ctl
->flags
))
3918 * We already reached the max active zones. Try to finish one block
3919 * group to make a room for a new block group. This is only possible
3920 * for a data block group because btrfs_zone_finish() may need to wait
3921 * for a running transaction which can cause a deadlock for metadata
3924 if (ffe_ctl
->flags
& BTRFS_BLOCK_GROUP_DATA
) {
3925 int ret
= btrfs_zone_finish_one_bg(fs_info
);
3934 * If we have enough free space left in an already active block group
3935 * and we can't activate any other zone now, do not allow allocating a
3936 * new chunk and let find_free_extent() retry with a smaller size.
3938 if (ffe_ctl
->max_extent_size
>= ffe_ctl
->min_alloc_size
)
3942 * Even min_alloc_size is not left in any block groups. Since we cannot
3943 * activate a new block group, allocating it may not help. Let's tell a
3944 * caller to try again and hope it progress something by writing some
3945 * parts of the region. That is only possible for data block groups,
3946 * where a part of the region can be written.
3948 if (ffe_ctl
->flags
& BTRFS_BLOCK_GROUP_DATA
)
3952 * We cannot activate a new block group and no enough space left in any
3953 * block groups. So, allocating a new block group may not help. But,
3954 * there is nothing to do anyway, so let's go with it.
3959 static int can_allocate_chunk(struct btrfs_fs_info
*fs_info
,
3960 struct find_free_extent_ctl
*ffe_ctl
)
3962 switch (ffe_ctl
->policy
) {
3963 case BTRFS_EXTENT_ALLOC_CLUSTERED
:
3965 case BTRFS_EXTENT_ALLOC_ZONED
:
3966 return can_allocate_chunk_zoned(fs_info
, ffe_ctl
);
3973 * Return >0 means caller needs to re-search for free extent
3974 * Return 0 means we have the needed free extent.
3975 * Return <0 means we failed to locate any free extent.
3977 static int find_free_extent_update_loop(struct btrfs_fs_info
*fs_info
,
3978 struct btrfs_key
*ins
,
3979 struct find_free_extent_ctl
*ffe_ctl
,
3982 struct btrfs_root
*root
= fs_info
->chunk_root
;
3985 if ((ffe_ctl
->loop
== LOOP_CACHING_NOWAIT
) &&
3986 ffe_ctl
->have_caching_bg
&& !ffe_ctl
->orig_have_caching_bg
)
3987 ffe_ctl
->orig_have_caching_bg
= true;
3989 if (ins
->objectid
) {
3990 found_extent(ffe_ctl
, ins
);
3994 if (ffe_ctl
->loop
>= LOOP_CACHING_WAIT
&& ffe_ctl
->have_caching_bg
)
3998 if (ffe_ctl
->index
< BTRFS_NR_RAID_TYPES
)
4001 /* See the comments for btrfs_loop_type for an explanation of the phases. */
4002 if (ffe_ctl
->loop
< LOOP_NO_EMPTY_SIZE
) {
4005 * We want to skip the LOOP_CACHING_WAIT step if we don't have
4006 * any uncached bgs and we've already done a full search
4009 if (ffe_ctl
->loop
== LOOP_CACHING_NOWAIT
&&
4010 (!ffe_ctl
->orig_have_caching_bg
&& full_search
))
4014 if (ffe_ctl
->loop
== LOOP_ALLOC_CHUNK
) {
4015 struct btrfs_trans_handle
*trans
;
4018 /* Check if allocation policy allows to create a new chunk */
4019 ret
= can_allocate_chunk(fs_info
, ffe_ctl
);
4023 trans
= current
->journal_info
;
4027 trans
= btrfs_join_transaction(root
);
4029 if (IS_ERR(trans
)) {
4030 ret
= PTR_ERR(trans
);
4034 ret
= btrfs_chunk_alloc(trans
, ffe_ctl
->flags
,
4035 CHUNK_ALLOC_FORCE_FOR_EXTENT
);
4037 /* Do not bail out on ENOSPC since we can do more. */
4038 if (ret
== -ENOSPC
) {
4043 btrfs_abort_transaction(trans
, ret
);
4047 btrfs_end_transaction(trans
);
4052 if (ffe_ctl
->loop
== LOOP_NO_EMPTY_SIZE
) {
4053 if (ffe_ctl
->policy
!= BTRFS_EXTENT_ALLOC_CLUSTERED
)
4057 * Don't loop again if we already have no empty_size and
4060 if (ffe_ctl
->empty_size
== 0 &&
4061 ffe_ctl
->empty_cluster
== 0)
4063 ffe_ctl
->empty_size
= 0;
4064 ffe_ctl
->empty_cluster
= 0;
4071 static bool find_free_extent_check_size_class(struct find_free_extent_ctl
*ffe_ctl
,
4072 struct btrfs_block_group
*bg
)
4074 if (ffe_ctl
->policy
== BTRFS_EXTENT_ALLOC_ZONED
)
4076 if (!btrfs_block_group_should_use_size_class(bg
))
4078 if (ffe_ctl
->loop
>= LOOP_WRONG_SIZE_CLASS
)
4080 if (ffe_ctl
->loop
>= LOOP_UNSET_SIZE_CLASS
&&
4081 bg
->size_class
== BTRFS_BG_SZ_NONE
)
4083 return ffe_ctl
->size_class
== bg
->size_class
;
4086 static int prepare_allocation_clustered(struct btrfs_fs_info
*fs_info
,
4087 struct find_free_extent_ctl
*ffe_ctl
,
4088 struct btrfs_space_info
*space_info
,
4089 struct btrfs_key
*ins
)
4092 * If our free space is heavily fragmented we may not be able to make
4093 * big contiguous allocations, so instead of doing the expensive search
4094 * for free space, simply return ENOSPC with our max_extent_size so we
4095 * can go ahead and search for a more manageable chunk.
4097 * If our max_extent_size is large enough for our allocation simply
4098 * disable clustering since we will likely not be able to find enough
4099 * space to create a cluster and induce latency trying.
4101 if (space_info
->max_extent_size
) {
4102 spin_lock(&space_info
->lock
);
4103 if (space_info
->max_extent_size
&&
4104 ffe_ctl
->num_bytes
> space_info
->max_extent_size
) {
4105 ins
->offset
= space_info
->max_extent_size
;
4106 spin_unlock(&space_info
->lock
);
4108 } else if (space_info
->max_extent_size
) {
4109 ffe_ctl
->use_cluster
= false;
4111 spin_unlock(&space_info
->lock
);
4114 ffe_ctl
->last_ptr
= fetch_cluster_info(fs_info
, space_info
,
4115 &ffe_ctl
->empty_cluster
);
4116 if (ffe_ctl
->last_ptr
) {
4117 struct btrfs_free_cluster
*last_ptr
= ffe_ctl
->last_ptr
;
4119 spin_lock(&last_ptr
->lock
);
4120 if (last_ptr
->block_group
)
4121 ffe_ctl
->hint_byte
= last_ptr
->window_start
;
4122 if (last_ptr
->fragmented
) {
4124 * We still set window_start so we can keep track of the
4125 * last place we found an allocation to try and save
4128 ffe_ctl
->hint_byte
= last_ptr
->window_start
;
4129 ffe_ctl
->use_cluster
= false;
4131 spin_unlock(&last_ptr
->lock
);
4137 static int prepare_allocation(struct btrfs_fs_info
*fs_info
,
4138 struct find_free_extent_ctl
*ffe_ctl
,
4139 struct btrfs_space_info
*space_info
,
4140 struct btrfs_key
*ins
)
4142 switch (ffe_ctl
->policy
) {
4143 case BTRFS_EXTENT_ALLOC_CLUSTERED
:
4144 return prepare_allocation_clustered(fs_info
, ffe_ctl
,
4146 case BTRFS_EXTENT_ALLOC_ZONED
:
4147 if (ffe_ctl
->for_treelog
) {
4148 spin_lock(&fs_info
->treelog_bg_lock
);
4149 if (fs_info
->treelog_bg
)
4150 ffe_ctl
->hint_byte
= fs_info
->treelog_bg
;
4151 spin_unlock(&fs_info
->treelog_bg_lock
);
4153 if (ffe_ctl
->for_data_reloc
) {
4154 spin_lock(&fs_info
->relocation_bg_lock
);
4155 if (fs_info
->data_reloc_bg
)
4156 ffe_ctl
->hint_byte
= fs_info
->data_reloc_bg
;
4157 spin_unlock(&fs_info
->relocation_bg_lock
);
4166 * walks the btree of allocated extents and find a hole of a given size.
4167 * The key ins is changed to record the hole:
4168 * ins->objectid == start position
4169 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4170 * ins->offset == the size of the hole.
4171 * Any available blocks before search_start are skipped.
4173 * If there is no suitable free space, we will record the max size of
4174 * the free space extent currently.
4176 * The overall logic and call chain:
4178 * find_free_extent()
4179 * |- Iterate through all block groups
4180 * | |- Get a valid block group
4181 * | |- Try to do clustered allocation in that block group
4182 * | |- Try to do unclustered allocation in that block group
4183 * | |- Check if the result is valid
4184 * | | |- If valid, then exit
4185 * | |- Jump to next block group
4187 * |- Push harder to find free extents
4188 * |- If not found, re-iterate all block groups
4190 static noinline
int find_free_extent(struct btrfs_root
*root
,
4191 struct btrfs_key
*ins
,
4192 struct find_free_extent_ctl
*ffe_ctl
)
4194 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4196 int cache_block_group_error
= 0;
4197 struct btrfs_block_group
*block_group
= NULL
;
4198 struct btrfs_space_info
*space_info
;
4199 bool full_search
= false;
4201 WARN_ON(ffe_ctl
->num_bytes
< fs_info
->sectorsize
);
4203 ffe_ctl
->search_start
= 0;
4204 /* For clustered allocation */
4205 ffe_ctl
->empty_cluster
= 0;
4206 ffe_ctl
->last_ptr
= NULL
;
4207 ffe_ctl
->use_cluster
= true;
4208 ffe_ctl
->have_caching_bg
= false;
4209 ffe_ctl
->orig_have_caching_bg
= false;
4210 ffe_ctl
->index
= btrfs_bg_flags_to_raid_index(ffe_ctl
->flags
);
4212 ffe_ctl
->retry_uncached
= false;
4213 ffe_ctl
->cached
= 0;
4214 ffe_ctl
->max_extent_size
= 0;
4215 ffe_ctl
->total_free_space
= 0;
4216 ffe_ctl
->found_offset
= 0;
4217 ffe_ctl
->policy
= BTRFS_EXTENT_ALLOC_CLUSTERED
;
4218 ffe_ctl
->size_class
= btrfs_calc_block_group_size_class(ffe_ctl
->num_bytes
);
4220 if (btrfs_is_zoned(fs_info
))
4221 ffe_ctl
->policy
= BTRFS_EXTENT_ALLOC_ZONED
;
4223 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
4227 trace_find_free_extent(root
, ffe_ctl
);
4229 space_info
= btrfs_find_space_info(fs_info
, ffe_ctl
->flags
);
4231 btrfs_err(fs_info
, "No space info for %llu", ffe_ctl
->flags
);
4235 ret
= prepare_allocation(fs_info
, ffe_ctl
, space_info
, ins
);
4239 ffe_ctl
->search_start
= max(ffe_ctl
->search_start
,
4240 first_logical_byte(fs_info
));
4241 ffe_ctl
->search_start
= max(ffe_ctl
->search_start
, ffe_ctl
->hint_byte
);
4242 if (ffe_ctl
->search_start
== ffe_ctl
->hint_byte
) {
4243 block_group
= btrfs_lookup_block_group(fs_info
,
4244 ffe_ctl
->search_start
);
4246 * we don't want to use the block group if it doesn't match our
4247 * allocation bits, or if its not cached.
4249 * However if we are re-searching with an ideal block group
4250 * picked out then we don't care that the block group is cached.
4252 if (block_group
&& block_group_bits(block_group
, ffe_ctl
->flags
) &&
4253 block_group
->cached
!= BTRFS_CACHE_NO
) {
4254 down_read(&space_info
->groups_sem
);
4255 if (list_empty(&block_group
->list
) ||
4258 * someone is removing this block group,
4259 * we can't jump into the have_block_group
4260 * target because our list pointers are not
4263 btrfs_put_block_group(block_group
);
4264 up_read(&space_info
->groups_sem
);
4266 ffe_ctl
->index
= btrfs_bg_flags_to_raid_index(
4267 block_group
->flags
);
4268 btrfs_lock_block_group(block_group
,
4270 ffe_ctl
->hinted
= true;
4271 goto have_block_group
;
4273 } else if (block_group
) {
4274 btrfs_put_block_group(block_group
);
4278 trace_find_free_extent_search_loop(root
, ffe_ctl
);
4279 ffe_ctl
->have_caching_bg
= false;
4280 if (ffe_ctl
->index
== btrfs_bg_flags_to_raid_index(ffe_ctl
->flags
) ||
4281 ffe_ctl
->index
== 0)
4283 down_read(&space_info
->groups_sem
);
4284 list_for_each_entry(block_group
,
4285 &space_info
->block_groups
[ffe_ctl
->index
], list
) {
4286 struct btrfs_block_group
*bg_ret
;
4288 ffe_ctl
->hinted
= false;
4289 /* If the block group is read-only, we can skip it entirely. */
4290 if (unlikely(block_group
->ro
)) {
4291 if (ffe_ctl
->for_treelog
)
4292 btrfs_clear_treelog_bg(block_group
);
4293 if (ffe_ctl
->for_data_reloc
)
4294 btrfs_clear_data_reloc_bg(block_group
);
4298 btrfs_grab_block_group(block_group
, ffe_ctl
->delalloc
);
4299 ffe_ctl
->search_start
= block_group
->start
;
4302 * this can happen if we end up cycling through all the
4303 * raid types, but we want to make sure we only allocate
4304 * for the proper type.
4306 if (!block_group_bits(block_group
, ffe_ctl
->flags
)) {
4307 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
4308 BTRFS_BLOCK_GROUP_RAID1_MASK
|
4309 BTRFS_BLOCK_GROUP_RAID56_MASK
|
4310 BTRFS_BLOCK_GROUP_RAID10
;
4313 * if they asked for extra copies and this block group
4314 * doesn't provide them, bail. This does allow us to
4315 * fill raid0 from raid1.
4317 if ((ffe_ctl
->flags
& extra
) && !(block_group
->flags
& extra
))
4321 * This block group has different flags than we want.
4322 * It's possible that we have MIXED_GROUP flag but no
4323 * block group is mixed. Just skip such block group.
4325 btrfs_release_block_group(block_group
, ffe_ctl
->delalloc
);
4330 trace_find_free_extent_have_block_group(root
, ffe_ctl
, block_group
);
4331 ffe_ctl
->cached
= btrfs_block_group_done(block_group
);
4332 if (unlikely(!ffe_ctl
->cached
)) {
4333 ffe_ctl
->have_caching_bg
= true;
4334 ret
= btrfs_cache_block_group(block_group
, false);
4337 * If we get ENOMEM here or something else we want to
4338 * try other block groups, because it may not be fatal.
4339 * However if we can't find anything else we need to
4340 * save our return here so that we return the actual
4341 * error that caused problems, not ENOSPC.
4344 if (!cache_block_group_error
)
4345 cache_block_group_error
= ret
;
4352 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
)) {
4353 if (!cache_block_group_error
)
4354 cache_block_group_error
= -EIO
;
4358 if (!find_free_extent_check_size_class(ffe_ctl
, block_group
))
4362 ret
= do_allocation(block_group
, ffe_ctl
, &bg_ret
);
4366 if (bg_ret
&& bg_ret
!= block_group
) {
4367 btrfs_release_block_group(block_group
, ffe_ctl
->delalloc
);
4368 block_group
= bg_ret
;
4372 ffe_ctl
->search_start
= round_up(ffe_ctl
->found_offset
,
4373 fs_info
->stripesize
);
4375 /* move on to the next group */
4376 if (ffe_ctl
->search_start
+ ffe_ctl
->num_bytes
>
4377 block_group
->start
+ block_group
->length
) {
4378 btrfs_add_free_space_unused(block_group
,
4379 ffe_ctl
->found_offset
,
4380 ffe_ctl
->num_bytes
);
4384 if (ffe_ctl
->found_offset
< ffe_ctl
->search_start
)
4385 btrfs_add_free_space_unused(block_group
,
4386 ffe_ctl
->found_offset
,
4387 ffe_ctl
->search_start
- ffe_ctl
->found_offset
);
4389 ret
= btrfs_add_reserved_bytes(block_group
, ffe_ctl
->ram_bytes
,
4392 ffe_ctl
->loop
>= LOOP_WRONG_SIZE_CLASS
);
4393 if (ret
== -EAGAIN
) {
4394 btrfs_add_free_space_unused(block_group
,
4395 ffe_ctl
->found_offset
,
4396 ffe_ctl
->num_bytes
);
4399 btrfs_inc_block_group_reservations(block_group
);
4401 /* we are all good, lets return */
4402 ins
->objectid
= ffe_ctl
->search_start
;
4403 ins
->offset
= ffe_ctl
->num_bytes
;
4405 trace_btrfs_reserve_extent(block_group
, ffe_ctl
);
4406 btrfs_release_block_group(block_group
, ffe_ctl
->delalloc
);
4409 if (!ffe_ctl
->cached
&& ffe_ctl
->loop
> LOOP_CACHING_NOWAIT
&&
4410 !ffe_ctl
->retry_uncached
) {
4411 ffe_ctl
->retry_uncached
= true;
4412 btrfs_wait_block_group_cache_progress(block_group
,
4413 ffe_ctl
->num_bytes
+
4414 ffe_ctl
->empty_cluster
+
4415 ffe_ctl
->empty_size
);
4416 goto have_block_group
;
4418 release_block_group(block_group
, ffe_ctl
, ffe_ctl
->delalloc
);
4421 up_read(&space_info
->groups_sem
);
4423 ret
= find_free_extent_update_loop(fs_info
, ins
, ffe_ctl
, full_search
);
4427 if (ret
== -ENOSPC
&& !cache_block_group_error
) {
4429 * Use ffe_ctl->total_free_space as fallback if we can't find
4430 * any contiguous hole.
4432 if (!ffe_ctl
->max_extent_size
)
4433 ffe_ctl
->max_extent_size
= ffe_ctl
->total_free_space
;
4434 spin_lock(&space_info
->lock
);
4435 space_info
->max_extent_size
= ffe_ctl
->max_extent_size
;
4436 spin_unlock(&space_info
->lock
);
4437 ins
->offset
= ffe_ctl
->max_extent_size
;
4438 } else if (ret
== -ENOSPC
) {
4439 ret
= cache_block_group_error
;
4445 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4446 * hole that is at least as big as @num_bytes.
4448 * @root - The root that will contain this extent
4450 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4451 * is used for accounting purposes. This value differs
4452 * from @num_bytes only in the case of compressed extents.
4454 * @num_bytes - Number of bytes to allocate on-disk.
4456 * @min_alloc_size - Indicates the minimum amount of space that the
4457 * allocator should try to satisfy. In some cases
4458 * @num_bytes may be larger than what is required and if
4459 * the filesystem is fragmented then allocation fails.
4460 * However, the presence of @min_alloc_size gives a
4461 * chance to try and satisfy the smaller allocation.
4463 * @empty_size - A hint that you plan on doing more COW. This is the
4464 * size in bytes the allocator should try to find free
4465 * next to the block it returns. This is just a hint and
4466 * may be ignored by the allocator.
4468 * @hint_byte - Hint to the allocator to start searching above the byte
4469 * address passed. It might be ignored.
4471 * @ins - This key is modified to record the found hole. It will
4472 * have the following values:
4473 * ins->objectid == start position
4474 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4475 * ins->offset == the size of the hole.
4477 * @is_data - Boolean flag indicating whether an extent is
4478 * allocated for data (true) or metadata (false)
4480 * @delalloc - Boolean flag indicating whether this allocation is for
4481 * delalloc or not. If 'true' data_rwsem of block groups
4482 * is going to be acquired.
4485 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4486 * case -ENOSPC is returned then @ins->offset will contain the size of the
4487 * largest available hole the allocator managed to find.
4489 int btrfs_reserve_extent(struct btrfs_root
*root
, u64 ram_bytes
,
4490 u64 num_bytes
, u64 min_alloc_size
,
4491 u64 empty_size
, u64 hint_byte
,
4492 struct btrfs_key
*ins
, int is_data
, int delalloc
)
4494 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4495 struct find_free_extent_ctl ffe_ctl
= {};
4496 bool final_tried
= num_bytes
== min_alloc_size
;
4499 bool for_treelog
= (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
);
4500 bool for_data_reloc
= (btrfs_is_data_reloc_root(root
) && is_data
);
4502 flags
= get_alloc_profile_by_root(root
, is_data
);
4504 WARN_ON(num_bytes
< fs_info
->sectorsize
);
4506 ffe_ctl
.ram_bytes
= ram_bytes
;
4507 ffe_ctl
.num_bytes
= num_bytes
;
4508 ffe_ctl
.min_alloc_size
= min_alloc_size
;
4509 ffe_ctl
.empty_size
= empty_size
;
4510 ffe_ctl
.flags
= flags
;
4511 ffe_ctl
.delalloc
= delalloc
;
4512 ffe_ctl
.hint_byte
= hint_byte
;
4513 ffe_ctl
.for_treelog
= for_treelog
;
4514 ffe_ctl
.for_data_reloc
= for_data_reloc
;
4516 ret
= find_free_extent(root
, ins
, &ffe_ctl
);
4517 if (!ret
&& !is_data
) {
4518 btrfs_dec_block_group_reservations(fs_info
, ins
->objectid
);
4519 } else if (ret
== -ENOSPC
) {
4520 if (!final_tried
&& ins
->offset
) {
4521 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
4522 num_bytes
= round_down(num_bytes
,
4523 fs_info
->sectorsize
);
4524 num_bytes
= max(num_bytes
, min_alloc_size
);
4525 ram_bytes
= num_bytes
;
4526 if (num_bytes
== min_alloc_size
)
4529 } else if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
4530 struct btrfs_space_info
*sinfo
;
4532 sinfo
= btrfs_find_space_info(fs_info
, flags
);
4534 "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d",
4535 flags
, num_bytes
, for_treelog
, for_data_reloc
);
4537 btrfs_dump_space_info(fs_info
, sinfo
,
4545 int btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
4546 u64 start
, u64 len
, int delalloc
)
4548 struct btrfs_block_group
*cache
;
4550 cache
= btrfs_lookup_block_group(fs_info
, start
);
4552 btrfs_err(fs_info
, "Unable to find block group for %llu",
4557 btrfs_add_free_space(cache
, start
, len
);
4558 btrfs_free_reserved_bytes(cache
, len
, delalloc
);
4559 trace_btrfs_reserved_extent_free(fs_info
, start
, len
);
4561 btrfs_put_block_group(cache
);
4565 int btrfs_pin_reserved_extent(struct btrfs_trans_handle
*trans
, u64 start
,
4568 struct btrfs_block_group
*cache
;
4571 cache
= btrfs_lookup_block_group(trans
->fs_info
, start
);
4573 btrfs_err(trans
->fs_info
, "unable to find block group for %llu",
4578 ret
= pin_down_extent(trans
, cache
, start
, len
, 1);
4579 btrfs_put_block_group(cache
);
4583 static int alloc_reserved_extent(struct btrfs_trans_handle
*trans
, u64 bytenr
,
4586 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4589 ret
= remove_from_free_space_tree(trans
, bytenr
, num_bytes
);
4593 ret
= btrfs_update_block_group(trans
, bytenr
, num_bytes
, true);
4596 btrfs_err(fs_info
, "update block group failed for %llu %llu",
4601 trace_btrfs_reserved_extent_alloc(fs_info
, bytenr
, num_bytes
);
4605 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
4606 u64 parent
, u64 root_objectid
,
4607 u64 flags
, u64 owner
, u64 offset
,
4608 struct btrfs_key
*ins
, int ref_mod
)
4610 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4611 struct btrfs_root
*extent_root
;
4613 struct btrfs_extent_item
*extent_item
;
4614 struct btrfs_extent_inline_ref
*iref
;
4615 struct btrfs_path
*path
;
4616 struct extent_buffer
*leaf
;
4621 type
= BTRFS_SHARED_DATA_REF_KEY
;
4623 type
= BTRFS_EXTENT_DATA_REF_KEY
;
4625 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
4627 path
= btrfs_alloc_path();
4631 extent_root
= btrfs_extent_root(fs_info
, ins
->objectid
);
4632 ret
= btrfs_insert_empty_item(trans
, extent_root
, path
, ins
, size
);
4634 btrfs_free_path(path
);
4638 leaf
= path
->nodes
[0];
4639 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4640 struct btrfs_extent_item
);
4641 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
4642 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
4643 btrfs_set_extent_flags(leaf
, extent_item
,
4644 flags
| BTRFS_EXTENT_FLAG_DATA
);
4646 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
4647 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
4649 struct btrfs_shared_data_ref
*ref
;
4650 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
4651 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
4652 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
4654 struct btrfs_extent_data_ref
*ref
;
4655 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
4656 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
4657 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
4658 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
4659 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
4662 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4663 btrfs_free_path(path
);
4665 return alloc_reserved_extent(trans
, ins
->objectid
, ins
->offset
);
4668 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
4669 struct btrfs_delayed_ref_node
*node
,
4670 struct btrfs_delayed_extent_op
*extent_op
)
4672 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4673 struct btrfs_root
*extent_root
;
4675 struct btrfs_extent_item
*extent_item
;
4676 struct btrfs_key extent_key
;
4677 struct btrfs_tree_block_info
*block_info
;
4678 struct btrfs_extent_inline_ref
*iref
;
4679 struct btrfs_path
*path
;
4680 struct extent_buffer
*leaf
;
4681 struct btrfs_delayed_tree_ref
*ref
;
4682 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
4683 u64 flags
= extent_op
->flags_to_set
;
4684 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
4686 ref
= btrfs_delayed_node_to_tree_ref(node
);
4688 extent_key
.objectid
= node
->bytenr
;
4689 if (skinny_metadata
) {
4690 extent_key
.offset
= ref
->level
;
4691 extent_key
.type
= BTRFS_METADATA_ITEM_KEY
;
4693 extent_key
.offset
= node
->num_bytes
;
4694 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4695 size
+= sizeof(*block_info
);
4698 path
= btrfs_alloc_path();
4702 extent_root
= btrfs_extent_root(fs_info
, extent_key
.objectid
);
4703 ret
= btrfs_insert_empty_item(trans
, extent_root
, path
, &extent_key
,
4706 btrfs_free_path(path
);
4710 leaf
= path
->nodes
[0];
4711 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4712 struct btrfs_extent_item
);
4713 btrfs_set_extent_refs(leaf
, extent_item
, 1);
4714 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
4715 btrfs_set_extent_flags(leaf
, extent_item
,
4716 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
4718 if (skinny_metadata
) {
4719 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
4721 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
4722 btrfs_set_tree_block_key(leaf
, block_info
, &extent_op
->key
);
4723 btrfs_set_tree_block_level(leaf
, block_info
, ref
->level
);
4724 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
4727 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
4728 btrfs_set_extent_inline_ref_type(leaf
, iref
,
4729 BTRFS_SHARED_BLOCK_REF_KEY
);
4730 btrfs_set_extent_inline_ref_offset(leaf
, iref
, ref
->parent
);
4732 btrfs_set_extent_inline_ref_type(leaf
, iref
,
4733 BTRFS_TREE_BLOCK_REF_KEY
);
4734 btrfs_set_extent_inline_ref_offset(leaf
, iref
, ref
->root
);
4737 btrfs_mark_buffer_dirty(leaf
);
4738 btrfs_free_path(path
);
4740 return alloc_reserved_extent(trans
, node
->bytenr
, fs_info
->nodesize
);
4743 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
4744 struct btrfs_root
*root
, u64 owner
,
4745 u64 offset
, u64 ram_bytes
,
4746 struct btrfs_key
*ins
)
4748 struct btrfs_ref generic_ref
= { 0 };
4750 BUG_ON(root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
);
4752 btrfs_init_generic_ref(&generic_ref
, BTRFS_ADD_DELAYED_EXTENT
,
4753 ins
->objectid
, ins
->offset
, 0);
4754 btrfs_init_data_ref(&generic_ref
, root
->root_key
.objectid
, owner
,
4756 btrfs_ref_tree_mod(root
->fs_info
, &generic_ref
);
4758 return btrfs_add_delayed_data_ref(trans
, &generic_ref
, ram_bytes
);
4762 * this is used by the tree logging recovery code. It records that
4763 * an extent has been allocated and makes sure to clear the free
4764 * space cache bits as well
4766 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
4767 u64 root_objectid
, u64 owner
, u64 offset
,
4768 struct btrfs_key
*ins
)
4770 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4772 struct btrfs_block_group
*block_group
;
4773 struct btrfs_space_info
*space_info
;
4776 * Mixed block groups will exclude before processing the log so we only
4777 * need to do the exclude dance if this fs isn't mixed.
4779 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
)) {
4780 ret
= __exclude_logged_extent(fs_info
, ins
->objectid
,
4786 block_group
= btrfs_lookup_block_group(fs_info
, ins
->objectid
);
4790 space_info
= block_group
->space_info
;
4791 spin_lock(&space_info
->lock
);
4792 spin_lock(&block_group
->lock
);
4793 space_info
->bytes_reserved
+= ins
->offset
;
4794 block_group
->reserved
+= ins
->offset
;
4795 spin_unlock(&block_group
->lock
);
4796 spin_unlock(&space_info
->lock
);
4798 ret
= alloc_reserved_file_extent(trans
, 0, root_objectid
, 0, owner
,
4801 btrfs_pin_extent(trans
, ins
->objectid
, ins
->offset
, 1);
4802 btrfs_put_block_group(block_group
);
4806 static struct extent_buffer
*
4807 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4808 u64 bytenr
, int level
, u64 owner
,
4809 enum btrfs_lock_nesting nest
)
4811 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4812 struct extent_buffer
*buf
;
4813 u64 lockdep_owner
= owner
;
4815 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
, owner
, level
);
4820 * Extra safety check in case the extent tree is corrupted and extent
4821 * allocator chooses to use a tree block which is already used and
4824 if (buf
->lock_owner
== current
->pid
) {
4825 btrfs_err_rl(fs_info
,
4826 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4827 buf
->start
, btrfs_header_owner(buf
), current
->pid
);
4828 free_extent_buffer(buf
);
4829 return ERR_PTR(-EUCLEAN
);
4833 * The reloc trees are just snapshots, so we need them to appear to be
4834 * just like any other fs tree WRT lockdep.
4836 * The exception however is in replace_path() in relocation, where we
4837 * hold the lock on the original fs root and then search for the reloc
4838 * root. At that point we need to make sure any reloc root buffers are
4839 * set to the BTRFS_TREE_RELOC_OBJECTID lockdep class in order to make
4842 if (lockdep_owner
== BTRFS_TREE_RELOC_OBJECTID
&&
4843 !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS
, &root
->state
))
4844 lockdep_owner
= BTRFS_FS_TREE_OBJECTID
;
4846 /* btrfs_clear_buffer_dirty() accesses generation field. */
4847 btrfs_set_header_generation(buf
, trans
->transid
);
4850 * This needs to stay, because we could allocate a freed block from an
4851 * old tree into a new tree, so we need to make sure this new block is
4852 * set to the appropriate level and owner.
4854 btrfs_set_buffer_lockdep_class(lockdep_owner
, buf
, level
);
4856 __btrfs_tree_lock(buf
, nest
);
4857 btrfs_clear_buffer_dirty(trans
, buf
);
4858 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
4859 clear_bit(EXTENT_BUFFER_NO_CHECK
, &buf
->bflags
);
4861 set_extent_buffer_uptodate(buf
);
4863 memzero_extent_buffer(buf
, 0, sizeof(struct btrfs_header
));
4864 btrfs_set_header_level(buf
, level
);
4865 btrfs_set_header_bytenr(buf
, buf
->start
);
4866 btrfs_set_header_generation(buf
, trans
->transid
);
4867 btrfs_set_header_backref_rev(buf
, BTRFS_MIXED_BACKREF_REV
);
4868 btrfs_set_header_owner(buf
, owner
);
4869 write_extent_buffer_fsid(buf
, fs_info
->fs_devices
->metadata_uuid
);
4870 write_extent_buffer_chunk_tree_uuid(buf
, fs_info
->chunk_tree_uuid
);
4871 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4872 buf
->log_index
= root
->log_transid
% 2;
4874 * we allow two log transactions at a time, use different
4875 * EXTENT bit to differentiate dirty pages.
4877 if (buf
->log_index
== 0)
4878 set_extent_bit(&root
->dirty_log_pages
, buf
->start
,
4879 buf
->start
+ buf
->len
- 1,
4880 EXTENT_DIRTY
, NULL
);
4882 set_extent_bit(&root
->dirty_log_pages
, buf
->start
,
4883 buf
->start
+ buf
->len
- 1,
4886 buf
->log_index
= -1;
4887 set_extent_bit(&trans
->transaction
->dirty_pages
, buf
->start
,
4888 buf
->start
+ buf
->len
- 1, EXTENT_DIRTY
, NULL
);
4890 /* this returns a buffer locked for blocking */
4895 * finds a free extent and does all the dirty work required for allocation
4896 * returns the tree buffer or an ERR_PTR on error.
4898 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
4899 struct btrfs_root
*root
,
4900 u64 parent
, u64 root_objectid
,
4901 const struct btrfs_disk_key
*key
,
4902 int level
, u64 hint
,
4904 enum btrfs_lock_nesting nest
)
4906 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4907 struct btrfs_key ins
;
4908 struct btrfs_block_rsv
*block_rsv
;
4909 struct extent_buffer
*buf
;
4910 struct btrfs_delayed_extent_op
*extent_op
;
4911 struct btrfs_ref generic_ref
= { 0 };
4914 u32 blocksize
= fs_info
->nodesize
;
4915 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
4917 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4918 if (btrfs_is_testing(fs_info
)) {
4919 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
4920 level
, root_objectid
, nest
);
4922 root
->alloc_bytenr
+= blocksize
;
4927 block_rsv
= btrfs_use_block_rsv(trans
, root
, blocksize
);
4928 if (IS_ERR(block_rsv
))
4929 return ERR_CAST(block_rsv
);
4931 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
, blocksize
,
4932 empty_size
, hint
, &ins
, 0, 0);
4936 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
,
4937 root_objectid
, nest
);
4940 goto out_free_reserved
;
4943 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
4945 parent
= ins
.objectid
;
4946 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
4950 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4951 extent_op
= btrfs_alloc_delayed_extent_op();
4957 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
4959 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
4960 extent_op
->flags_to_set
= flags
;
4961 extent_op
->update_key
= skinny_metadata
? false : true;
4962 extent_op
->update_flags
= true;
4963 extent_op
->level
= level
;
4965 btrfs_init_generic_ref(&generic_ref
, BTRFS_ADD_DELAYED_EXTENT
,
4966 ins
.objectid
, ins
.offset
, parent
);
4967 btrfs_init_tree_ref(&generic_ref
, level
, root_objectid
,
4968 root
->root_key
.objectid
, false);
4969 btrfs_ref_tree_mod(fs_info
, &generic_ref
);
4970 ret
= btrfs_add_delayed_tree_ref(trans
, &generic_ref
, extent_op
);
4972 goto out_free_delayed
;
4977 btrfs_free_delayed_extent_op(extent_op
);
4979 btrfs_tree_unlock(buf
);
4980 free_extent_buffer(buf
);
4982 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 0);
4984 btrfs_unuse_block_rsv(fs_info
, block_rsv
, blocksize
);
4985 return ERR_PTR(ret
);
4988 struct walk_control
{
4989 u64 refs
[BTRFS_MAX_LEVEL
];
4990 u64 flags
[BTRFS_MAX_LEVEL
];
4991 struct btrfs_key update_progress
;
4992 struct btrfs_key drop_progress
;
5004 #define DROP_REFERENCE 1
5005 #define UPDATE_BACKREF 2
5007 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
5008 struct btrfs_root
*root
,
5009 struct walk_control
*wc
,
5010 struct btrfs_path
*path
)
5012 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5018 struct btrfs_key key
;
5019 struct extent_buffer
*eb
;
5024 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
5025 wc
->reada_count
= wc
->reada_count
* 2 / 3;
5026 wc
->reada_count
= max(wc
->reada_count
, 2);
5028 wc
->reada_count
= wc
->reada_count
* 3 / 2;
5029 wc
->reada_count
= min_t(int, wc
->reada_count
,
5030 BTRFS_NODEPTRS_PER_BLOCK(fs_info
));
5033 eb
= path
->nodes
[wc
->level
];
5034 nritems
= btrfs_header_nritems(eb
);
5036 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
5037 if (nread
>= wc
->reada_count
)
5041 bytenr
= btrfs_node_blockptr(eb
, slot
);
5042 generation
= btrfs_node_ptr_generation(eb
, slot
);
5044 if (slot
== path
->slots
[wc
->level
])
5047 if (wc
->stage
== UPDATE_BACKREF
&&
5048 generation
<= root
->root_key
.offset
)
5051 /* We don't lock the tree block, it's OK to be racy here */
5052 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
,
5053 wc
->level
- 1, 1, &refs
,
5055 /* We don't care about errors in readahead. */
5060 if (wc
->stage
== DROP_REFERENCE
) {
5064 if (wc
->level
== 1 &&
5065 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5067 if (!wc
->update_ref
||
5068 generation
<= root
->root_key
.offset
)
5070 btrfs_node_key_to_cpu(eb
, &key
, slot
);
5071 ret
= btrfs_comp_cpu_keys(&key
,
5072 &wc
->update_progress
);
5076 if (wc
->level
== 1 &&
5077 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5081 btrfs_readahead_node_child(eb
, slot
);
5084 wc
->reada_slot
= slot
;
5088 * helper to process tree block while walking down the tree.
5090 * when wc->stage == UPDATE_BACKREF, this function updates
5091 * back refs for pointers in the block.
5093 * NOTE: return value 1 means we should stop walking down.
5095 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
5096 struct btrfs_root
*root
,
5097 struct btrfs_path
*path
,
5098 struct walk_control
*wc
, int lookup_info
)
5100 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5101 int level
= wc
->level
;
5102 struct extent_buffer
*eb
= path
->nodes
[level
];
5103 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5106 if (wc
->stage
== UPDATE_BACKREF
&&
5107 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
5111 * when reference count of tree block is 1, it won't increase
5112 * again. once full backref flag is set, we never clear it.
5115 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
5116 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
5117 BUG_ON(!path
->locks
[level
]);
5118 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
5119 eb
->start
, level
, 1,
5122 BUG_ON(ret
== -ENOMEM
);
5125 BUG_ON(wc
->refs
[level
] == 0);
5128 if (wc
->stage
== DROP_REFERENCE
) {
5129 if (wc
->refs
[level
] > 1)
5132 if (path
->locks
[level
] && !wc
->keep_locks
) {
5133 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5134 path
->locks
[level
] = 0;
5139 /* wc->stage == UPDATE_BACKREF */
5140 if (!(wc
->flags
[level
] & flag
)) {
5141 BUG_ON(!path
->locks
[level
]);
5142 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
5143 BUG_ON(ret
); /* -ENOMEM */
5144 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5145 BUG_ON(ret
); /* -ENOMEM */
5146 ret
= btrfs_set_disk_extent_flags(trans
, eb
, flag
);
5147 BUG_ON(ret
); /* -ENOMEM */
5148 wc
->flags
[level
] |= flag
;
5152 * the block is shared by multiple trees, so it's not good to
5153 * keep the tree lock
5155 if (path
->locks
[level
] && level
> 0) {
5156 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5157 path
->locks
[level
] = 0;
5163 * This is used to verify a ref exists for this root to deal with a bug where we
5164 * would have a drop_progress key that hadn't been updated properly.
5166 static int check_ref_exists(struct btrfs_trans_handle
*trans
,
5167 struct btrfs_root
*root
, u64 bytenr
, u64 parent
,
5170 struct btrfs_path
*path
;
5171 struct btrfs_extent_inline_ref
*iref
;
5174 path
= btrfs_alloc_path();
5178 ret
= lookup_extent_backref(trans
, path
, &iref
, bytenr
,
5179 root
->fs_info
->nodesize
, parent
,
5180 root
->root_key
.objectid
, level
, 0);
5181 btrfs_free_path(path
);
5190 * helper to process tree block pointer.
5192 * when wc->stage == DROP_REFERENCE, this function checks
5193 * reference count of the block pointed to. if the block
5194 * is shared and we need update back refs for the subtree
5195 * rooted at the block, this function changes wc->stage to
5196 * UPDATE_BACKREF. if the block is shared and there is no
5197 * need to update back, this function drops the reference
5200 * NOTE: return value 1 means we should stop walking down.
5202 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
5203 struct btrfs_root
*root
,
5204 struct btrfs_path
*path
,
5205 struct walk_control
*wc
, int *lookup_info
)
5207 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5211 struct btrfs_tree_parent_check check
= { 0 };
5212 struct btrfs_key key
;
5213 struct btrfs_ref ref
= { 0 };
5214 struct extent_buffer
*next
;
5215 int level
= wc
->level
;
5218 bool need_account
= false;
5220 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
5221 path
->slots
[level
]);
5223 * if the lower level block was created before the snapshot
5224 * was created, we know there is no need to update back refs
5227 if (wc
->stage
== UPDATE_BACKREF
&&
5228 generation
<= root
->root_key
.offset
) {
5233 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
5235 check
.level
= level
- 1;
5236 check
.transid
= generation
;
5237 check
.owner_root
= root
->root_key
.objectid
;
5238 check
.has_first_key
= true;
5239 btrfs_node_key_to_cpu(path
->nodes
[level
], &check
.first_key
,
5240 path
->slots
[level
]);
5242 next
= find_extent_buffer(fs_info
, bytenr
);
5244 next
= btrfs_find_create_tree_block(fs_info
, bytenr
,
5245 root
->root_key
.objectid
, level
- 1);
5247 return PTR_ERR(next
);
5250 btrfs_tree_lock(next
);
5252 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
, level
- 1, 1,
5253 &wc
->refs
[level
- 1],
5254 &wc
->flags
[level
- 1]);
5258 if (unlikely(wc
->refs
[level
- 1] == 0)) {
5259 btrfs_err(fs_info
, "Missing references.");
5265 if (wc
->stage
== DROP_REFERENCE
) {
5266 if (wc
->refs
[level
- 1] > 1) {
5267 need_account
= true;
5269 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5272 if (!wc
->update_ref
||
5273 generation
<= root
->root_key
.offset
)
5276 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
5277 path
->slots
[level
]);
5278 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
5282 wc
->stage
= UPDATE_BACKREF
;
5283 wc
->shared_level
= level
- 1;
5287 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5291 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
5292 btrfs_tree_unlock(next
);
5293 free_extent_buffer(next
);
5299 if (reada
&& level
== 1)
5300 reada_walk_down(trans
, root
, wc
, path
);
5301 next
= read_tree_block(fs_info
, bytenr
, &check
);
5303 return PTR_ERR(next
);
5304 } else if (!extent_buffer_uptodate(next
)) {
5305 free_extent_buffer(next
);
5308 btrfs_tree_lock(next
);
5312 ASSERT(level
== btrfs_header_level(next
));
5313 if (level
!= btrfs_header_level(next
)) {
5314 btrfs_err(root
->fs_info
, "mismatched level");
5318 path
->nodes
[level
] = next
;
5319 path
->slots
[level
] = 0;
5320 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
5326 wc
->refs
[level
- 1] = 0;
5327 wc
->flags
[level
- 1] = 0;
5328 if (wc
->stage
== DROP_REFERENCE
) {
5329 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
5330 parent
= path
->nodes
[level
]->start
;
5332 ASSERT(root
->root_key
.objectid
==
5333 btrfs_header_owner(path
->nodes
[level
]));
5334 if (root
->root_key
.objectid
!=
5335 btrfs_header_owner(path
->nodes
[level
])) {
5336 btrfs_err(root
->fs_info
,
5337 "mismatched block owner");
5345 * If we had a drop_progress we need to verify the refs are set
5346 * as expected. If we find our ref then we know that from here
5347 * on out everything should be correct, and we can clear the
5350 if (wc
->restarted
) {
5351 ret
= check_ref_exists(trans
, root
, bytenr
, parent
,
5362 * Reloc tree doesn't contribute to qgroup numbers, and we have
5363 * already accounted them at merge time (replace_path),
5364 * thus we could skip expensive subtree trace here.
5366 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
&&
5368 ret
= btrfs_qgroup_trace_subtree(trans
, next
,
5369 generation
, level
- 1);
5371 btrfs_err_rl(fs_info
,
5372 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5378 * We need to update the next key in our walk control so we can
5379 * update the drop_progress key accordingly. We don't care if
5380 * find_next_key doesn't find a key because that means we're at
5381 * the end and are going to clean up now.
5383 wc
->drop_level
= level
;
5384 find_next_key(path
, level
, &wc
->drop_progress
);
5386 btrfs_init_generic_ref(&ref
, BTRFS_DROP_DELAYED_REF
, bytenr
,
5387 fs_info
->nodesize
, parent
);
5388 btrfs_init_tree_ref(&ref
, level
- 1, root
->root_key
.objectid
,
5390 ret
= btrfs_free_extent(trans
, &ref
);
5399 btrfs_tree_unlock(next
);
5400 free_extent_buffer(next
);
5406 * helper to process tree block while walking up the tree.
5408 * when wc->stage == DROP_REFERENCE, this function drops
5409 * reference count on the block.
5411 * when wc->stage == UPDATE_BACKREF, this function changes
5412 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5413 * to UPDATE_BACKREF previously while processing the block.
5415 * NOTE: return value 1 means we should stop walking up.
5417 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
5418 struct btrfs_root
*root
,
5419 struct btrfs_path
*path
,
5420 struct walk_control
*wc
)
5422 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5424 int level
= wc
->level
;
5425 struct extent_buffer
*eb
= path
->nodes
[level
];
5428 if (wc
->stage
== UPDATE_BACKREF
) {
5429 BUG_ON(wc
->shared_level
< level
);
5430 if (level
< wc
->shared_level
)
5433 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
5437 wc
->stage
= DROP_REFERENCE
;
5438 wc
->shared_level
= -1;
5439 path
->slots
[level
] = 0;
5442 * check reference count again if the block isn't locked.
5443 * we should start walking down the tree again if reference
5446 if (!path
->locks
[level
]) {
5448 btrfs_tree_lock(eb
);
5449 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
5451 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
5452 eb
->start
, level
, 1,
5456 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5457 path
->locks
[level
] = 0;
5460 BUG_ON(wc
->refs
[level
] == 0);
5461 if (wc
->refs
[level
] == 1) {
5462 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5463 path
->locks
[level
] = 0;
5469 /* wc->stage == DROP_REFERENCE */
5470 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
5472 if (wc
->refs
[level
] == 1) {
5474 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5475 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
5477 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5478 BUG_ON(ret
); /* -ENOMEM */
5479 if (is_fstree(root
->root_key
.objectid
)) {
5480 ret
= btrfs_qgroup_trace_leaf_items(trans
, eb
);
5482 btrfs_err_rl(fs_info
,
5483 "error %d accounting leaf items, quota is out of sync, rescan required",
5488 /* Make block locked assertion in btrfs_clear_buffer_dirty happy. */
5489 if (!path
->locks
[level
]) {
5490 btrfs_tree_lock(eb
);
5491 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
5493 btrfs_clear_buffer_dirty(trans
, eb
);
5496 if (eb
== root
->node
) {
5497 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5499 else if (root
->root_key
.objectid
!= btrfs_header_owner(eb
))
5500 goto owner_mismatch
;
5502 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5503 parent
= path
->nodes
[level
+ 1]->start
;
5504 else if (root
->root_key
.objectid
!=
5505 btrfs_header_owner(path
->nodes
[level
+ 1]))
5506 goto owner_mismatch
;
5509 btrfs_free_tree_block(trans
, btrfs_root_id(root
), eb
, parent
,
5510 wc
->refs
[level
] == 1);
5512 wc
->refs
[level
] = 0;
5513 wc
->flags
[level
] = 0;
5517 btrfs_err_rl(fs_info
, "unexpected tree owner, have %llu expect %llu",
5518 btrfs_header_owner(eb
), root
->root_key
.objectid
);
5522 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
5523 struct btrfs_root
*root
,
5524 struct btrfs_path
*path
,
5525 struct walk_control
*wc
)
5527 int level
= wc
->level
;
5528 int lookup_info
= 1;
5531 while (level
>= 0) {
5532 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
5539 if (path
->slots
[level
] >=
5540 btrfs_header_nritems(path
->nodes
[level
]))
5543 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
5545 path
->slots
[level
]++;
5551 return (ret
== 1) ? 0 : ret
;
5554 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
5555 struct btrfs_root
*root
,
5556 struct btrfs_path
*path
,
5557 struct walk_control
*wc
, int max_level
)
5559 int level
= wc
->level
;
5562 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
5563 while (level
< max_level
&& path
->nodes
[level
]) {
5565 if (path
->slots
[level
] + 1 <
5566 btrfs_header_nritems(path
->nodes
[level
])) {
5567 path
->slots
[level
]++;
5570 ret
= walk_up_proc(trans
, root
, path
, wc
);
5576 if (path
->locks
[level
]) {
5577 btrfs_tree_unlock_rw(path
->nodes
[level
],
5578 path
->locks
[level
]);
5579 path
->locks
[level
] = 0;
5581 free_extent_buffer(path
->nodes
[level
]);
5582 path
->nodes
[level
] = NULL
;
5590 * drop a subvolume tree.
5592 * this function traverses the tree freeing any blocks that only
5593 * referenced by the tree.
5595 * when a shared tree block is found. this function decreases its
5596 * reference count by one. if update_ref is true, this function
5597 * also make sure backrefs for the shared block and all lower level
5598 * blocks are properly updated.
5600 * If called with for_reloc == 0, may exit early with -EAGAIN
5602 int btrfs_drop_snapshot(struct btrfs_root
*root
, int update_ref
, int for_reloc
)
5604 const bool is_reloc_root
= (root
->root_key
.objectid
==
5605 BTRFS_TREE_RELOC_OBJECTID
);
5606 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5607 struct btrfs_path
*path
;
5608 struct btrfs_trans_handle
*trans
;
5609 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
5610 struct btrfs_root_item
*root_item
= &root
->root_item
;
5611 struct walk_control
*wc
;
5612 struct btrfs_key key
;
5616 bool root_dropped
= false;
5617 bool unfinished_drop
= false;
5619 btrfs_debug(fs_info
, "Drop subvolume %llu", root
->root_key
.objectid
);
5621 path
= btrfs_alloc_path();
5627 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
5629 btrfs_free_path(path
);
5635 * Use join to avoid potential EINTR from transaction start. See
5636 * wait_reserve_ticket and the whole reservation callchain.
5639 trans
= btrfs_join_transaction(tree_root
);
5641 trans
= btrfs_start_transaction(tree_root
, 0);
5642 if (IS_ERR(trans
)) {
5643 err
= PTR_ERR(trans
);
5647 err
= btrfs_run_delayed_items(trans
);
5652 * This will help us catch people modifying the fs tree while we're
5653 * dropping it. It is unsafe to mess with the fs tree while it's being
5654 * dropped as we unlock the root node and parent nodes as we walk down
5655 * the tree, assuming nothing will change. If something does change
5656 * then we'll have stale information and drop references to blocks we've
5659 set_bit(BTRFS_ROOT_DELETING
, &root
->state
);
5660 unfinished_drop
= test_bit(BTRFS_ROOT_UNFINISHED_DROP
, &root
->state
);
5662 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
5663 level
= btrfs_header_level(root
->node
);
5664 path
->nodes
[level
] = btrfs_lock_root_node(root
);
5665 path
->slots
[level
] = 0;
5666 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
5667 memset(&wc
->update_progress
, 0,
5668 sizeof(wc
->update_progress
));
5670 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
5671 memcpy(&wc
->update_progress
, &key
,
5672 sizeof(wc
->update_progress
));
5674 level
= btrfs_root_drop_level(root_item
);
5676 path
->lowest_level
= level
;
5677 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5678 path
->lowest_level
= 0;
5686 * unlock our path, this is safe because only this
5687 * function is allowed to delete this snapshot
5689 btrfs_unlock_up_safe(path
, 0);
5691 level
= btrfs_header_level(root
->node
);
5693 btrfs_tree_lock(path
->nodes
[level
]);
5694 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
5696 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
5697 path
->nodes
[level
]->start
,
5698 level
, 1, &wc
->refs
[level
],
5704 BUG_ON(wc
->refs
[level
] == 0);
5706 if (level
== btrfs_root_drop_level(root_item
))
5709 btrfs_tree_unlock(path
->nodes
[level
]);
5710 path
->locks
[level
] = 0;
5711 WARN_ON(wc
->refs
[level
] != 1);
5716 wc
->restarted
= test_bit(BTRFS_ROOT_DEAD_TREE
, &root
->state
);
5718 wc
->shared_level
= -1;
5719 wc
->stage
= DROP_REFERENCE
;
5720 wc
->update_ref
= update_ref
;
5722 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
5726 ret
= walk_down_tree(trans
, root
, path
, wc
);
5728 btrfs_abort_transaction(trans
, ret
);
5733 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
5735 btrfs_abort_transaction(trans
, ret
);
5741 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
5745 if (wc
->stage
== DROP_REFERENCE
) {
5746 wc
->drop_level
= wc
->level
;
5747 btrfs_node_key_to_cpu(path
->nodes
[wc
->drop_level
],
5749 path
->slots
[wc
->drop_level
]);
5751 btrfs_cpu_key_to_disk(&root_item
->drop_progress
,
5752 &wc
->drop_progress
);
5753 btrfs_set_root_drop_level(root_item
, wc
->drop_level
);
5755 BUG_ON(wc
->level
== 0);
5756 if (btrfs_should_end_transaction(trans
) ||
5757 (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
))) {
5758 ret
= btrfs_update_root(trans
, tree_root
,
5762 btrfs_abort_transaction(trans
, ret
);
5768 btrfs_set_last_root_drop_gen(fs_info
, trans
->transid
);
5770 btrfs_end_transaction_throttle(trans
);
5771 if (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
)) {
5772 btrfs_debug(fs_info
,
5773 "drop snapshot early exit");
5779 * Use join to avoid potential EINTR from transaction
5780 * start. See wait_reserve_ticket and the whole
5781 * reservation callchain.
5784 trans
= btrfs_join_transaction(tree_root
);
5786 trans
= btrfs_start_transaction(tree_root
, 0);
5787 if (IS_ERR(trans
)) {
5788 err
= PTR_ERR(trans
);
5793 btrfs_release_path(path
);
5797 ret
= btrfs_del_root(trans
, &root
->root_key
);
5799 btrfs_abort_transaction(trans
, ret
);
5804 if (!is_reloc_root
) {
5805 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
5808 btrfs_abort_transaction(trans
, ret
);
5811 } else if (ret
> 0) {
5812 /* if we fail to delete the orphan item this time
5813 * around, it'll get picked up the next time.
5815 * The most common failure here is just -ENOENT.
5817 btrfs_del_orphan_item(trans
, tree_root
,
5818 root
->root_key
.objectid
);
5823 * This subvolume is going to be completely dropped, and won't be
5824 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
5825 * commit transaction time. So free it here manually.
5827 btrfs_qgroup_convert_reserved_meta(root
, INT_MAX
);
5828 btrfs_qgroup_free_meta_all_pertrans(root
);
5830 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
))
5831 btrfs_add_dropped_root(trans
, root
);
5833 btrfs_put_root(root
);
5834 root_dropped
= true;
5837 btrfs_set_last_root_drop_gen(fs_info
, trans
->transid
);
5839 btrfs_end_transaction_throttle(trans
);
5842 btrfs_free_path(path
);
5845 * We were an unfinished drop root, check to see if there are any
5846 * pending, and if not clear and wake up any waiters.
5848 if (!err
&& unfinished_drop
)
5849 btrfs_maybe_wake_unfinished_drop(fs_info
);
5852 * So if we need to stop dropping the snapshot for whatever reason we
5853 * need to make sure to add it back to the dead root list so that we
5854 * keep trying to do the work later. This also cleans up roots if we
5855 * don't have it in the radix (like when we recover after a power fail
5856 * or unmount) so we don't leak memory.
5858 if (!for_reloc
&& !root_dropped
)
5859 btrfs_add_dead_root(root
);
5864 * drop subtree rooted at tree block 'node'.
5866 * NOTE: this function will unlock and release tree block 'node'
5867 * only used by relocation code
5869 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
5870 struct btrfs_root
*root
,
5871 struct extent_buffer
*node
,
5872 struct extent_buffer
*parent
)
5874 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5875 struct btrfs_path
*path
;
5876 struct walk_control
*wc
;
5882 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
5884 path
= btrfs_alloc_path();
5888 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
5890 btrfs_free_path(path
);
5894 btrfs_assert_tree_write_locked(parent
);
5895 parent_level
= btrfs_header_level(parent
);
5896 atomic_inc(&parent
->refs
);
5897 path
->nodes
[parent_level
] = parent
;
5898 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
5900 btrfs_assert_tree_write_locked(node
);
5901 level
= btrfs_header_level(node
);
5902 path
->nodes
[level
] = node
;
5903 path
->slots
[level
] = 0;
5904 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
5906 wc
->refs
[parent_level
] = 1;
5907 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5909 wc
->shared_level
= -1;
5910 wc
->stage
= DROP_REFERENCE
;
5913 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
5916 wret
= walk_down_tree(trans
, root
, path
, wc
);
5922 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
5930 btrfs_free_path(path
);
5934 int btrfs_error_unpin_extent_range(struct btrfs_fs_info
*fs_info
,
5937 return unpin_extent_range(fs_info
, start
, end
, false);
5941 * It used to be that old block groups would be left around forever.
5942 * Iterating over them would be enough to trim unused space. Since we
5943 * now automatically remove them, we also need to iterate over unallocated
5946 * We don't want a transaction for this since the discard may take a
5947 * substantial amount of time. We don't require that a transaction be
5948 * running, but we do need to take a running transaction into account
5949 * to ensure that we're not discarding chunks that were released or
5950 * allocated in the current transaction.
5952 * Holding the chunks lock will prevent other threads from allocating
5953 * or releasing chunks, but it won't prevent a running transaction
5954 * from committing and releasing the memory that the pending chunks
5955 * list head uses. For that, we need to take a reference to the
5956 * transaction and hold the commit root sem. We only need to hold
5957 * it while performing the free space search since we have already
5958 * held back allocations.
5960 static int btrfs_trim_free_extents(struct btrfs_device
*device
, u64
*trimmed
)
5962 u64 start
= BTRFS_DEVICE_RANGE_RESERVED
, len
= 0, end
= 0;
5967 /* Discard not supported = nothing to do. */
5968 if (!bdev_max_discard_sectors(device
->bdev
))
5971 /* Not writable = nothing to do. */
5972 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
))
5975 /* No free space = nothing to do. */
5976 if (device
->total_bytes
<= device
->bytes_used
)
5982 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
5985 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
5989 find_first_clear_extent_bit(&device
->alloc_state
, start
,
5991 CHUNK_TRIMMED
| CHUNK_ALLOCATED
);
5993 /* Check if there are any CHUNK_* bits left */
5994 if (start
> device
->total_bytes
) {
5995 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG
));
5996 btrfs_warn_in_rcu(fs_info
,
5997 "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
5998 start
, end
- start
+ 1,
5999 btrfs_dev_name(device
),
6000 device
->total_bytes
);
6001 mutex_unlock(&fs_info
->chunk_mutex
);
6006 /* Ensure we skip the reserved space on each device. */
6007 start
= max_t(u64
, start
, BTRFS_DEVICE_RANGE_RESERVED
);
6010 * If find_first_clear_extent_bit find a range that spans the
6011 * end of the device it will set end to -1, in this case it's up
6012 * to the caller to trim the value to the size of the device.
6014 end
= min(end
, device
->total_bytes
- 1);
6016 len
= end
- start
+ 1;
6018 /* We didn't find any extents */
6020 mutex_unlock(&fs_info
->chunk_mutex
);
6025 ret
= btrfs_issue_discard(device
->bdev
, start
, len
,
6028 set_extent_bit(&device
->alloc_state
, start
,
6029 start
+ bytes
- 1, CHUNK_TRIMMED
, NULL
);
6030 mutex_unlock(&fs_info
->chunk_mutex
);
6038 if (fatal_signal_pending(current
)) {
6050 * Trim the whole filesystem by:
6051 * 1) trimming the free space in each block group
6052 * 2) trimming the unallocated space on each device
6054 * This will also continue trimming even if a block group or device encounters
6055 * an error. The return value will be the last error, or 0 if nothing bad
6058 int btrfs_trim_fs(struct btrfs_fs_info
*fs_info
, struct fstrim_range
*range
)
6060 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6061 struct btrfs_block_group
*cache
= NULL
;
6062 struct btrfs_device
*device
;
6064 u64 range_end
= U64_MAX
;
6074 if (range
->start
== U64_MAX
)
6078 * Check range overflow if range->len is set.
6079 * The default range->len is U64_MAX.
6081 if (range
->len
!= U64_MAX
&&
6082 check_add_overflow(range
->start
, range
->len
, &range_end
))
6085 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
6086 for (; cache
; cache
= btrfs_next_block_group(cache
)) {
6087 if (cache
->start
>= range_end
) {
6088 btrfs_put_block_group(cache
);
6092 start
= max(range
->start
, cache
->start
);
6093 end
= min(range_end
, cache
->start
+ cache
->length
);
6095 if (end
- start
>= range
->minlen
) {
6096 if (!btrfs_block_group_done(cache
)) {
6097 ret
= btrfs_cache_block_group(cache
, true);
6104 ret
= btrfs_trim_block_group(cache
,
6110 trimmed
+= group_trimmed
;
6121 "failed to trim %llu block group(s), last error %d",
6124 mutex_lock(&fs_devices
->device_list_mutex
);
6125 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6126 if (test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
))
6129 ret
= btrfs_trim_free_extents(device
, &group_trimmed
);
6136 trimmed
+= group_trimmed
;
6138 mutex_unlock(&fs_devices
->device_list_mutex
);
6142 "failed to trim %llu device(s), last error %d",
6143 dev_failed
, dev_ret
);
6144 range
->len
= trimmed
;