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 BUG_ON(refs_to_add
!= 1);
1519 ret
= insert_tree_block_ref(trans
, path
, bytenr
, parent
,
1522 ret
= insert_extent_data_ref(trans
, path
, bytenr
, parent
,
1523 root_objectid
, owner
, offset
,
1527 btrfs_abort_transaction(trans
, ret
);
1529 btrfs_free_path(path
);
1533 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1534 struct btrfs_delayed_ref_node
*node
,
1535 struct btrfs_delayed_extent_op
*extent_op
,
1536 bool insert_reserved
)
1539 struct btrfs_delayed_data_ref
*ref
;
1540 struct btrfs_key ins
;
1545 ins
.objectid
= node
->bytenr
;
1546 ins
.offset
= node
->num_bytes
;
1547 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1549 ref
= btrfs_delayed_node_to_data_ref(node
);
1550 trace_run_delayed_data_ref(trans
->fs_info
, node
, ref
, node
->action
);
1552 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1553 parent
= ref
->parent
;
1554 ref_root
= ref
->root
;
1556 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1558 flags
|= extent_op
->flags_to_set
;
1559 ret
= alloc_reserved_file_extent(trans
, parent
, ref_root
,
1560 flags
, ref
->objectid
,
1563 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1564 ret
= __btrfs_inc_extent_ref(trans
, node
, parent
, ref_root
,
1565 ref
->objectid
, ref
->offset
,
1566 node
->ref_mod
, extent_op
);
1567 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1568 ret
= __btrfs_free_extent(trans
, node
, parent
,
1569 ref_root
, ref
->objectid
,
1570 ref
->offset
, node
->ref_mod
,
1578 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1579 struct extent_buffer
*leaf
,
1580 struct btrfs_extent_item
*ei
)
1582 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1583 if (extent_op
->update_flags
) {
1584 flags
|= extent_op
->flags_to_set
;
1585 btrfs_set_extent_flags(leaf
, ei
, flags
);
1588 if (extent_op
->update_key
) {
1589 struct btrfs_tree_block_info
*bi
;
1590 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1591 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1592 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1596 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1597 struct btrfs_delayed_ref_head
*head
,
1598 struct btrfs_delayed_extent_op
*extent_op
)
1600 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1601 struct btrfs_root
*root
;
1602 struct btrfs_key key
;
1603 struct btrfs_path
*path
;
1604 struct btrfs_extent_item
*ei
;
1605 struct extent_buffer
*leaf
;
1611 if (TRANS_ABORTED(trans
))
1614 if (!btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
1617 path
= btrfs_alloc_path();
1621 key
.objectid
= head
->bytenr
;
1624 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1625 key
.offset
= extent_op
->level
;
1627 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1628 key
.offset
= head
->num_bytes
;
1631 root
= btrfs_extent_root(fs_info
, key
.objectid
);
1633 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1640 if (path
->slots
[0] > 0) {
1642 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1644 if (key
.objectid
== head
->bytenr
&&
1645 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1646 key
.offset
== head
->num_bytes
)
1650 btrfs_release_path(path
);
1653 key
.objectid
= head
->bytenr
;
1654 key
.offset
= head
->num_bytes
;
1655 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1664 leaf
= path
->nodes
[0];
1665 item_size
= btrfs_item_size(leaf
, path
->slots
[0]);
1667 if (unlikely(item_size
< sizeof(*ei
))) {
1670 "unexpected extent item size, has %u expect >= %zu",
1671 item_size
, sizeof(*ei
));
1672 btrfs_abort_transaction(trans
, err
);
1676 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1677 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1679 btrfs_mark_buffer_dirty(leaf
);
1681 btrfs_free_path(path
);
1685 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
1686 struct btrfs_delayed_ref_node
*node
,
1687 struct btrfs_delayed_extent_op
*extent_op
,
1688 bool insert_reserved
)
1691 struct btrfs_delayed_tree_ref
*ref
;
1695 ref
= btrfs_delayed_node_to_tree_ref(node
);
1696 trace_run_delayed_tree_ref(trans
->fs_info
, node
, ref
, node
->action
);
1698 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1699 parent
= ref
->parent
;
1700 ref_root
= ref
->root
;
1702 if (unlikely(node
->ref_mod
!= 1)) {
1703 btrfs_err(trans
->fs_info
,
1704 "btree block %llu has %d references rather than 1: action %d ref_root %llu parent %llu",
1705 node
->bytenr
, node
->ref_mod
, node
->action
, ref_root
,
1709 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1710 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
1711 ret
= alloc_reserved_tree_block(trans
, node
, extent_op
);
1712 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1713 ret
= __btrfs_inc_extent_ref(trans
, node
, parent
, ref_root
,
1714 ref
->level
, 0, 1, extent_op
);
1715 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1716 ret
= __btrfs_free_extent(trans
, node
, parent
, ref_root
,
1717 ref
->level
, 0, 1, extent_op
);
1724 /* helper function to actually process a single delayed ref entry */
1725 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
1726 struct btrfs_delayed_ref_node
*node
,
1727 struct btrfs_delayed_extent_op
*extent_op
,
1728 bool insert_reserved
)
1732 if (TRANS_ABORTED(trans
)) {
1733 if (insert_reserved
)
1734 btrfs_pin_extent(trans
, node
->bytenr
, node
->num_bytes
, 1);
1738 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
1739 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1740 ret
= run_delayed_tree_ref(trans
, node
, extent_op
,
1742 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
1743 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1744 ret
= run_delayed_data_ref(trans
, node
, extent_op
,
1748 if (ret
&& insert_reserved
)
1749 btrfs_pin_extent(trans
, node
->bytenr
, node
->num_bytes
, 1);
1751 btrfs_err(trans
->fs_info
,
1752 "failed to run delayed ref for logical %llu num_bytes %llu type %u action %u ref_mod %d: %d",
1753 node
->bytenr
, node
->num_bytes
, node
->type
,
1754 node
->action
, node
->ref_mod
, ret
);
1758 static inline struct btrfs_delayed_ref_node
*
1759 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
1761 struct btrfs_delayed_ref_node
*ref
;
1763 if (RB_EMPTY_ROOT(&head
->ref_tree
.rb_root
))
1767 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1768 * This is to prevent a ref count from going down to zero, which deletes
1769 * the extent item from the extent tree, when there still are references
1770 * to add, which would fail because they would not find the extent item.
1772 if (!list_empty(&head
->ref_add_list
))
1773 return list_first_entry(&head
->ref_add_list
,
1774 struct btrfs_delayed_ref_node
, add_list
);
1776 ref
= rb_entry(rb_first_cached(&head
->ref_tree
),
1777 struct btrfs_delayed_ref_node
, ref_node
);
1778 ASSERT(list_empty(&ref
->add_list
));
1782 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root
*delayed_refs
,
1783 struct btrfs_delayed_ref_head
*head
)
1785 spin_lock(&delayed_refs
->lock
);
1786 head
->processing
= false;
1787 delayed_refs
->num_heads_ready
++;
1788 spin_unlock(&delayed_refs
->lock
);
1789 btrfs_delayed_ref_unlock(head
);
1792 static struct btrfs_delayed_extent_op
*cleanup_extent_op(
1793 struct btrfs_delayed_ref_head
*head
)
1795 struct btrfs_delayed_extent_op
*extent_op
= head
->extent_op
;
1800 if (head
->must_insert_reserved
) {
1801 head
->extent_op
= NULL
;
1802 btrfs_free_delayed_extent_op(extent_op
);
1808 static int run_and_cleanup_extent_op(struct btrfs_trans_handle
*trans
,
1809 struct btrfs_delayed_ref_head
*head
)
1811 struct btrfs_delayed_extent_op
*extent_op
;
1814 extent_op
= cleanup_extent_op(head
);
1817 head
->extent_op
= NULL
;
1818 spin_unlock(&head
->lock
);
1819 ret
= run_delayed_extent_op(trans
, head
, extent_op
);
1820 btrfs_free_delayed_extent_op(extent_op
);
1821 return ret
? ret
: 1;
1824 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info
*fs_info
,
1825 struct btrfs_delayed_ref_root
*delayed_refs
,
1826 struct btrfs_delayed_ref_head
*head
)
1828 int nr_items
= 1; /* Dropping this ref head update. */
1831 * We had csum deletions accounted for in our delayed refs rsv, we need
1832 * to drop the csum leaves for this update from our delayed_refs_rsv.
1834 if (head
->total_ref_mod
< 0 && head
->is_data
) {
1835 spin_lock(&delayed_refs
->lock
);
1836 delayed_refs
->pending_csums
-= head
->num_bytes
;
1837 spin_unlock(&delayed_refs
->lock
);
1838 nr_items
+= btrfs_csum_bytes_to_leaves(fs_info
, head
->num_bytes
);
1841 btrfs_delayed_refs_rsv_release(fs_info
, nr_items
);
1844 static int cleanup_ref_head(struct btrfs_trans_handle
*trans
,
1845 struct btrfs_delayed_ref_head
*head
)
1848 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1849 struct btrfs_delayed_ref_root
*delayed_refs
;
1852 delayed_refs
= &trans
->transaction
->delayed_refs
;
1854 ret
= run_and_cleanup_extent_op(trans
, head
);
1856 unselect_delayed_ref_head(delayed_refs
, head
);
1857 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
1864 * Need to drop our head ref lock and re-acquire the delayed ref lock
1865 * and then re-check to make sure nobody got added.
1867 spin_unlock(&head
->lock
);
1868 spin_lock(&delayed_refs
->lock
);
1869 spin_lock(&head
->lock
);
1870 if (!RB_EMPTY_ROOT(&head
->ref_tree
.rb_root
) || head
->extent_op
) {
1871 spin_unlock(&head
->lock
);
1872 spin_unlock(&delayed_refs
->lock
);
1875 btrfs_delete_ref_head(delayed_refs
, head
);
1876 spin_unlock(&head
->lock
);
1877 spin_unlock(&delayed_refs
->lock
);
1879 if (head
->must_insert_reserved
) {
1880 btrfs_pin_extent(trans
, head
->bytenr
, head
->num_bytes
, 1);
1881 if (head
->is_data
) {
1882 struct btrfs_root
*csum_root
;
1884 csum_root
= btrfs_csum_root(fs_info
, head
->bytenr
);
1885 ret
= btrfs_del_csums(trans
, csum_root
, head
->bytenr
,
1890 btrfs_cleanup_ref_head_accounting(fs_info
, delayed_refs
, head
);
1892 trace_run_delayed_ref_head(fs_info
, head
, 0);
1893 btrfs_delayed_ref_unlock(head
);
1894 btrfs_put_delayed_ref_head(head
);
1898 static struct btrfs_delayed_ref_head
*btrfs_obtain_ref_head(
1899 struct btrfs_trans_handle
*trans
)
1901 struct btrfs_delayed_ref_root
*delayed_refs
=
1902 &trans
->transaction
->delayed_refs
;
1903 struct btrfs_delayed_ref_head
*head
= NULL
;
1906 spin_lock(&delayed_refs
->lock
);
1907 head
= btrfs_select_ref_head(delayed_refs
);
1909 spin_unlock(&delayed_refs
->lock
);
1914 * Grab the lock that says we are going to process all the refs for
1917 ret
= btrfs_delayed_ref_lock(delayed_refs
, head
);
1918 spin_unlock(&delayed_refs
->lock
);
1921 * We may have dropped the spin lock to get the head mutex lock, and
1922 * that might have given someone else time to free the head. If that's
1923 * true, it has been removed from our list and we can move on.
1926 head
= ERR_PTR(-EAGAIN
);
1931 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle
*trans
,
1932 struct btrfs_delayed_ref_head
*locked_ref
)
1934 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1935 struct btrfs_delayed_ref_root
*delayed_refs
;
1936 struct btrfs_delayed_extent_op
*extent_op
;
1937 struct btrfs_delayed_ref_node
*ref
;
1938 bool must_insert_reserved
;
1941 delayed_refs
= &trans
->transaction
->delayed_refs
;
1943 lockdep_assert_held(&locked_ref
->mutex
);
1944 lockdep_assert_held(&locked_ref
->lock
);
1946 while ((ref
= select_delayed_ref(locked_ref
))) {
1948 btrfs_check_delayed_seq(fs_info
, ref
->seq
)) {
1949 spin_unlock(&locked_ref
->lock
);
1950 unselect_delayed_ref_head(delayed_refs
, locked_ref
);
1954 rb_erase_cached(&ref
->ref_node
, &locked_ref
->ref_tree
);
1955 RB_CLEAR_NODE(&ref
->ref_node
);
1956 if (!list_empty(&ref
->add_list
))
1957 list_del(&ref
->add_list
);
1959 * When we play the delayed ref, also correct the ref_mod on
1962 switch (ref
->action
) {
1963 case BTRFS_ADD_DELAYED_REF
:
1964 case BTRFS_ADD_DELAYED_EXTENT
:
1965 locked_ref
->ref_mod
-= ref
->ref_mod
;
1967 case BTRFS_DROP_DELAYED_REF
:
1968 locked_ref
->ref_mod
+= ref
->ref_mod
;
1973 atomic_dec(&delayed_refs
->num_entries
);
1976 * Record the must_insert_reserved flag before we drop the
1979 must_insert_reserved
= locked_ref
->must_insert_reserved
;
1980 locked_ref
->must_insert_reserved
= false;
1982 extent_op
= locked_ref
->extent_op
;
1983 locked_ref
->extent_op
= NULL
;
1984 spin_unlock(&locked_ref
->lock
);
1986 ret
= run_one_delayed_ref(trans
, ref
, extent_op
,
1987 must_insert_reserved
);
1989 btrfs_free_delayed_extent_op(extent_op
);
1991 unselect_delayed_ref_head(delayed_refs
, locked_ref
);
1992 btrfs_put_delayed_ref(ref
);
1996 btrfs_put_delayed_ref(ref
);
1999 spin_lock(&locked_ref
->lock
);
2000 btrfs_merge_delayed_refs(fs_info
, delayed_refs
, locked_ref
);
2007 * Returns 0 on success or if called with an already aborted transaction.
2008 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2010 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2013 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2014 struct btrfs_delayed_ref_root
*delayed_refs
;
2015 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2017 unsigned long count
= 0;
2019 delayed_refs
= &trans
->transaction
->delayed_refs
;
2022 locked_ref
= btrfs_obtain_ref_head(trans
);
2023 if (IS_ERR_OR_NULL(locked_ref
)) {
2024 if (PTR_ERR(locked_ref
) == -EAGAIN
) {
2033 * We need to try and merge add/drops of the same ref since we
2034 * can run into issues with relocate dropping the implicit ref
2035 * and then it being added back again before the drop can
2036 * finish. If we merged anything we need to re-loop so we can
2038 * Or we can get node references of the same type that weren't
2039 * merged when created due to bumps in the tree mod seq, and
2040 * we need to merge them to prevent adding an inline extent
2041 * backref before dropping it (triggering a BUG_ON at
2042 * insert_inline_extent_backref()).
2044 spin_lock(&locked_ref
->lock
);
2045 btrfs_merge_delayed_refs(fs_info
, delayed_refs
, locked_ref
);
2047 ret
= btrfs_run_delayed_refs_for_head(trans
, locked_ref
);
2048 if (ret
< 0 && ret
!= -EAGAIN
) {
2050 * Error, btrfs_run_delayed_refs_for_head already
2051 * unlocked everything so just bail out
2056 * Success, perform the usual cleanup of a processed
2059 ret
= cleanup_ref_head(trans
, locked_ref
);
2061 /* We dropped our lock, we need to loop. */
2070 * Either success case or btrfs_run_delayed_refs_for_head
2071 * returned -EAGAIN, meaning we need to select another head
2076 } while ((nr
!= -1 && count
< nr
) || locked_ref
);
2081 #ifdef SCRAMBLE_DELAYED_REFS
2083 * Normally delayed refs get processed in ascending bytenr order. This
2084 * correlates in most cases to the order added. To expose dependencies on this
2085 * order, we start to process the tree in the middle instead of the beginning
2087 static u64
find_middle(struct rb_root
*root
)
2089 struct rb_node
*n
= root
->rb_node
;
2090 struct btrfs_delayed_ref_node
*entry
;
2093 u64 first
= 0, last
= 0;
2097 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2098 first
= entry
->bytenr
;
2102 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2103 last
= entry
->bytenr
;
2108 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2109 WARN_ON(!entry
->in_tree
);
2111 middle
= entry
->bytenr
;
2125 * this starts processing the delayed reference count updates and
2126 * extent insertions we have queued up so far. count can be
2127 * 0, which means to process everything in the tree at the start
2128 * of the run (but not newly added entries), or it can be some target
2129 * number you'd like to process.
2131 * Returns 0 on success or if called with an aborted transaction
2132 * Returns <0 on error and aborts the transaction
2134 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2135 unsigned long count
)
2137 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2138 struct rb_node
*node
;
2139 struct btrfs_delayed_ref_root
*delayed_refs
;
2140 struct btrfs_delayed_ref_head
*head
;
2142 int run_all
= count
== (unsigned long)-1;
2144 /* We'll clean this up in btrfs_cleanup_transaction */
2145 if (TRANS_ABORTED(trans
))
2148 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE
, &fs_info
->flags
))
2151 delayed_refs
= &trans
->transaction
->delayed_refs
;
2153 count
= delayed_refs
->num_heads_ready
;
2156 #ifdef SCRAMBLE_DELAYED_REFS
2157 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2159 ret
= __btrfs_run_delayed_refs(trans
, count
);
2161 btrfs_abort_transaction(trans
, ret
);
2166 btrfs_create_pending_block_groups(trans
);
2168 spin_lock(&delayed_refs
->lock
);
2169 node
= rb_first_cached(&delayed_refs
->href_root
);
2171 spin_unlock(&delayed_refs
->lock
);
2174 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2176 refcount_inc(&head
->refs
);
2177 spin_unlock(&delayed_refs
->lock
);
2179 /* Mutex was contended, block until it's released and retry. */
2180 mutex_lock(&head
->mutex
);
2181 mutex_unlock(&head
->mutex
);
2183 btrfs_put_delayed_ref_head(head
);
2191 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2192 struct extent_buffer
*eb
, u64 flags
)
2194 struct btrfs_delayed_extent_op
*extent_op
;
2195 int level
= btrfs_header_level(eb
);
2198 extent_op
= btrfs_alloc_delayed_extent_op();
2202 extent_op
->flags_to_set
= flags
;
2203 extent_op
->update_flags
= true;
2204 extent_op
->update_key
= false;
2205 extent_op
->level
= level
;
2207 ret
= btrfs_add_delayed_extent_op(trans
, eb
->start
, eb
->len
, extent_op
);
2209 btrfs_free_delayed_extent_op(extent_op
);
2213 static noinline
int check_delayed_ref(struct btrfs_root
*root
,
2214 struct btrfs_path
*path
,
2215 u64 objectid
, u64 offset
, u64 bytenr
)
2217 struct btrfs_delayed_ref_head
*head
;
2218 struct btrfs_delayed_ref_node
*ref
;
2219 struct btrfs_delayed_data_ref
*data_ref
;
2220 struct btrfs_delayed_ref_root
*delayed_refs
;
2221 struct btrfs_transaction
*cur_trans
;
2222 struct rb_node
*node
;
2225 spin_lock(&root
->fs_info
->trans_lock
);
2226 cur_trans
= root
->fs_info
->running_transaction
;
2228 refcount_inc(&cur_trans
->use_count
);
2229 spin_unlock(&root
->fs_info
->trans_lock
);
2233 delayed_refs
= &cur_trans
->delayed_refs
;
2234 spin_lock(&delayed_refs
->lock
);
2235 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
2237 spin_unlock(&delayed_refs
->lock
);
2238 btrfs_put_transaction(cur_trans
);
2242 if (!mutex_trylock(&head
->mutex
)) {
2244 spin_unlock(&delayed_refs
->lock
);
2245 btrfs_put_transaction(cur_trans
);
2249 refcount_inc(&head
->refs
);
2250 spin_unlock(&delayed_refs
->lock
);
2252 btrfs_release_path(path
);
2255 * Mutex was contended, block until it's released and let
2258 mutex_lock(&head
->mutex
);
2259 mutex_unlock(&head
->mutex
);
2260 btrfs_put_delayed_ref_head(head
);
2261 btrfs_put_transaction(cur_trans
);
2264 spin_unlock(&delayed_refs
->lock
);
2266 spin_lock(&head
->lock
);
2268 * XXX: We should replace this with a proper search function in the
2271 for (node
= rb_first_cached(&head
->ref_tree
); node
;
2272 node
= rb_next(node
)) {
2273 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, ref_node
);
2274 /* If it's a shared ref we know a cross reference exists */
2275 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
2280 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2283 * If our ref doesn't match the one we're currently looking at
2284 * then we have a cross reference.
2286 if (data_ref
->root
!= root
->root_key
.objectid
||
2287 data_ref
->objectid
!= objectid
||
2288 data_ref
->offset
!= offset
) {
2293 spin_unlock(&head
->lock
);
2294 mutex_unlock(&head
->mutex
);
2295 btrfs_put_transaction(cur_trans
);
2299 static noinline
int check_committed_ref(struct btrfs_root
*root
,
2300 struct btrfs_path
*path
,
2301 u64 objectid
, u64 offset
, u64 bytenr
,
2304 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2305 struct btrfs_root
*extent_root
= btrfs_extent_root(fs_info
, bytenr
);
2306 struct extent_buffer
*leaf
;
2307 struct btrfs_extent_data_ref
*ref
;
2308 struct btrfs_extent_inline_ref
*iref
;
2309 struct btrfs_extent_item
*ei
;
2310 struct btrfs_key key
;
2315 key
.objectid
= bytenr
;
2316 key
.offset
= (u64
)-1;
2317 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2319 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2322 BUG_ON(ret
== 0); /* Corruption */
2325 if (path
->slots
[0] == 0)
2329 leaf
= path
->nodes
[0];
2330 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2332 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2336 item_size
= btrfs_item_size(leaf
, path
->slots
[0]);
2337 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2339 /* If extent item has more than 1 inline ref then it's shared */
2340 if (item_size
!= sizeof(*ei
) +
2341 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2345 * If extent created before last snapshot => it's shared unless the
2346 * snapshot has been deleted. Use the heuristic if strict is false.
2349 (btrfs_extent_generation(leaf
, ei
) <=
2350 btrfs_root_last_snapshot(&root
->root_item
)))
2353 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2355 /* If this extent has SHARED_DATA_REF then it's shared */
2356 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_DATA
);
2357 if (type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2360 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2361 if (btrfs_extent_refs(leaf
, ei
) !=
2362 btrfs_extent_data_ref_count(leaf
, ref
) ||
2363 btrfs_extent_data_ref_root(leaf
, ref
) !=
2364 root
->root_key
.objectid
||
2365 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2366 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2374 int btrfs_cross_ref_exist(struct btrfs_root
*root
, u64 objectid
, u64 offset
,
2375 u64 bytenr
, bool strict
, struct btrfs_path
*path
)
2380 ret
= check_committed_ref(root
, path
, objectid
,
2381 offset
, bytenr
, strict
);
2382 if (ret
&& ret
!= -ENOENT
)
2385 ret
= check_delayed_ref(root
, path
, objectid
, offset
, bytenr
);
2386 } while (ret
== -EAGAIN
);
2389 btrfs_release_path(path
);
2390 if (btrfs_is_data_reloc_root(root
))
2395 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2396 struct btrfs_root
*root
,
2397 struct extent_buffer
*buf
,
2398 int full_backref
, int inc
)
2400 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2406 struct btrfs_key key
;
2407 struct btrfs_file_extent_item
*fi
;
2408 struct btrfs_ref generic_ref
= { 0 };
2409 bool for_reloc
= btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
);
2415 if (btrfs_is_testing(fs_info
))
2418 ref_root
= btrfs_header_owner(buf
);
2419 nritems
= btrfs_header_nritems(buf
);
2420 level
= btrfs_header_level(buf
);
2422 if (!test_bit(BTRFS_ROOT_SHAREABLE
, &root
->state
) && level
== 0)
2426 parent
= buf
->start
;
2430 action
= BTRFS_ADD_DELAYED_REF
;
2432 action
= BTRFS_DROP_DELAYED_REF
;
2434 for (i
= 0; i
< nritems
; i
++) {
2436 btrfs_item_key_to_cpu(buf
, &key
, i
);
2437 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2439 fi
= btrfs_item_ptr(buf
, i
,
2440 struct btrfs_file_extent_item
);
2441 if (btrfs_file_extent_type(buf
, fi
) ==
2442 BTRFS_FILE_EXTENT_INLINE
)
2444 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2448 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2449 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2450 btrfs_init_generic_ref(&generic_ref
, action
, bytenr
,
2452 btrfs_init_data_ref(&generic_ref
, ref_root
, key
.objectid
,
2453 key
.offset
, root
->root_key
.objectid
,
2456 ret
= btrfs_inc_extent_ref(trans
, &generic_ref
);
2458 ret
= btrfs_free_extent(trans
, &generic_ref
);
2462 bytenr
= btrfs_node_blockptr(buf
, i
);
2463 num_bytes
= fs_info
->nodesize
;
2464 btrfs_init_generic_ref(&generic_ref
, action
, bytenr
,
2466 btrfs_init_tree_ref(&generic_ref
, level
- 1, ref_root
,
2467 root
->root_key
.objectid
, for_reloc
);
2469 ret
= btrfs_inc_extent_ref(trans
, &generic_ref
);
2471 ret
= btrfs_free_extent(trans
, &generic_ref
);
2481 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2482 struct extent_buffer
*buf
, int full_backref
)
2484 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2487 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2488 struct extent_buffer
*buf
, int full_backref
)
2490 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2493 static u64
get_alloc_profile_by_root(struct btrfs_root
*root
, int data
)
2495 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2500 flags
= BTRFS_BLOCK_GROUP_DATA
;
2501 else if (root
== fs_info
->chunk_root
)
2502 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
2504 flags
= BTRFS_BLOCK_GROUP_METADATA
;
2506 ret
= btrfs_get_alloc_profile(fs_info
, flags
);
2510 static u64
first_logical_byte(struct btrfs_fs_info
*fs_info
)
2512 struct rb_node
*leftmost
;
2515 read_lock(&fs_info
->block_group_cache_lock
);
2516 /* Get the block group with the lowest logical start address. */
2517 leftmost
= rb_first_cached(&fs_info
->block_group_cache_tree
);
2519 struct btrfs_block_group
*bg
;
2521 bg
= rb_entry(leftmost
, struct btrfs_block_group
, cache_node
);
2524 read_unlock(&fs_info
->block_group_cache_lock
);
2529 static int pin_down_extent(struct btrfs_trans_handle
*trans
,
2530 struct btrfs_block_group
*cache
,
2531 u64 bytenr
, u64 num_bytes
, int reserved
)
2533 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
2535 spin_lock(&cache
->space_info
->lock
);
2536 spin_lock(&cache
->lock
);
2537 cache
->pinned
+= num_bytes
;
2538 btrfs_space_info_update_bytes_pinned(fs_info
, cache
->space_info
,
2541 cache
->reserved
-= num_bytes
;
2542 cache
->space_info
->bytes_reserved
-= num_bytes
;
2544 spin_unlock(&cache
->lock
);
2545 spin_unlock(&cache
->space_info
->lock
);
2547 set_extent_bit(&trans
->transaction
->pinned_extents
, bytenr
,
2548 bytenr
+ num_bytes
- 1, EXTENT_DIRTY
, NULL
);
2552 int btrfs_pin_extent(struct btrfs_trans_handle
*trans
,
2553 u64 bytenr
, u64 num_bytes
, int reserved
)
2555 struct btrfs_block_group
*cache
;
2557 cache
= btrfs_lookup_block_group(trans
->fs_info
, bytenr
);
2558 BUG_ON(!cache
); /* Logic error */
2560 pin_down_extent(trans
, cache
, bytenr
, num_bytes
, reserved
);
2562 btrfs_put_block_group(cache
);
2567 * this function must be called within transaction
2569 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle
*trans
,
2570 u64 bytenr
, u64 num_bytes
)
2572 struct btrfs_block_group
*cache
;
2575 cache
= btrfs_lookup_block_group(trans
->fs_info
, bytenr
);
2580 * Fully cache the free space first so that our pin removes the free space
2583 ret
= btrfs_cache_block_group(cache
, true);
2587 pin_down_extent(trans
, cache
, bytenr
, num_bytes
, 0);
2589 /* remove us from the free space cache (if we're there at all) */
2590 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
2592 btrfs_put_block_group(cache
);
2596 static int __exclude_logged_extent(struct btrfs_fs_info
*fs_info
,
2597 u64 start
, u64 num_bytes
)
2600 struct btrfs_block_group
*block_group
;
2602 block_group
= btrfs_lookup_block_group(fs_info
, start
);
2606 ret
= btrfs_cache_block_group(block_group
, true);
2610 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
2612 btrfs_put_block_group(block_group
);
2616 int btrfs_exclude_logged_extents(struct extent_buffer
*eb
)
2618 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
2619 struct btrfs_file_extent_item
*item
;
2620 struct btrfs_key key
;
2625 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
))
2628 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
2629 btrfs_item_key_to_cpu(eb
, &key
, i
);
2630 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2632 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
2633 found_type
= btrfs_file_extent_type(eb
, item
);
2634 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
2636 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
2638 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
2639 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
2640 ret
= __exclude_logged_extent(fs_info
, key
.objectid
, key
.offset
);
2649 btrfs_inc_block_group_reservations(struct btrfs_block_group
*bg
)
2651 atomic_inc(&bg
->reservations
);
2655 * Returns the free cluster for the given space info and sets empty_cluster to
2656 * what it should be based on the mount options.
2658 static struct btrfs_free_cluster
*
2659 fetch_cluster_info(struct btrfs_fs_info
*fs_info
,
2660 struct btrfs_space_info
*space_info
, u64
*empty_cluster
)
2662 struct btrfs_free_cluster
*ret
= NULL
;
2665 if (btrfs_mixed_space_info(space_info
))
2668 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
2669 ret
= &fs_info
->meta_alloc_cluster
;
2670 if (btrfs_test_opt(fs_info
, SSD
))
2671 *empty_cluster
= SZ_2M
;
2673 *empty_cluster
= SZ_64K
;
2674 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) &&
2675 btrfs_test_opt(fs_info
, SSD_SPREAD
)) {
2676 *empty_cluster
= SZ_2M
;
2677 ret
= &fs_info
->data_alloc_cluster
;
2683 static int unpin_extent_range(struct btrfs_fs_info
*fs_info
,
2685 const bool return_free_space
)
2687 struct btrfs_block_group
*cache
= NULL
;
2688 struct btrfs_space_info
*space_info
;
2689 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
2690 struct btrfs_free_cluster
*cluster
= NULL
;
2692 u64 total_unpinned
= 0;
2693 u64 empty_cluster
= 0;
2696 while (start
<= end
) {
2699 start
>= cache
->start
+ cache
->length
) {
2701 btrfs_put_block_group(cache
);
2703 cache
= btrfs_lookup_block_group(fs_info
, start
);
2704 BUG_ON(!cache
); /* Logic error */
2706 cluster
= fetch_cluster_info(fs_info
,
2709 empty_cluster
<<= 1;
2712 len
= cache
->start
+ cache
->length
- start
;
2713 len
= min(len
, end
+ 1 - start
);
2715 if (return_free_space
)
2716 btrfs_add_free_space(cache
, start
, len
);
2719 total_unpinned
+= len
;
2720 space_info
= cache
->space_info
;
2723 * If this space cluster has been marked as fragmented and we've
2724 * unpinned enough in this block group to potentially allow a
2725 * cluster to be created inside of it go ahead and clear the
2728 if (cluster
&& cluster
->fragmented
&&
2729 total_unpinned
> empty_cluster
) {
2730 spin_lock(&cluster
->lock
);
2731 cluster
->fragmented
= 0;
2732 spin_unlock(&cluster
->lock
);
2735 spin_lock(&space_info
->lock
);
2736 spin_lock(&cache
->lock
);
2737 cache
->pinned
-= len
;
2738 btrfs_space_info_update_bytes_pinned(fs_info
, space_info
, -len
);
2739 space_info
->max_extent_size
= 0;
2741 space_info
->bytes_readonly
+= len
;
2743 } else if (btrfs_is_zoned(fs_info
)) {
2744 /* Need reset before reusing in a zoned block group */
2745 space_info
->bytes_zone_unusable
+= len
;
2748 spin_unlock(&cache
->lock
);
2749 if (!readonly
&& return_free_space
&&
2750 global_rsv
->space_info
== space_info
) {
2751 spin_lock(&global_rsv
->lock
);
2752 if (!global_rsv
->full
) {
2753 u64 to_add
= min(len
, global_rsv
->size
-
2754 global_rsv
->reserved
);
2756 global_rsv
->reserved
+= to_add
;
2757 btrfs_space_info_update_bytes_may_use(fs_info
,
2758 space_info
, to_add
);
2759 if (global_rsv
->reserved
>= global_rsv
->size
)
2760 global_rsv
->full
= 1;
2763 spin_unlock(&global_rsv
->lock
);
2765 /* Add to any tickets we may have */
2766 if (!readonly
&& return_free_space
&& len
)
2767 btrfs_try_granting_tickets(fs_info
, space_info
);
2768 spin_unlock(&space_info
->lock
);
2772 btrfs_put_block_group(cache
);
2776 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
)
2778 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2779 struct btrfs_block_group
*block_group
, *tmp
;
2780 struct list_head
*deleted_bgs
;
2781 struct extent_io_tree
*unpin
;
2786 unpin
= &trans
->transaction
->pinned_extents
;
2788 while (!TRANS_ABORTED(trans
)) {
2789 struct extent_state
*cached_state
= NULL
;
2791 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
2792 if (!find_first_extent_bit(unpin
, 0, &start
, &end
,
2793 EXTENT_DIRTY
, &cached_state
)) {
2794 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
2798 if (btrfs_test_opt(fs_info
, DISCARD_SYNC
))
2799 ret
= btrfs_discard_extent(fs_info
, start
,
2800 end
+ 1 - start
, NULL
);
2802 clear_extent_dirty(unpin
, start
, end
, &cached_state
);
2803 unpin_extent_range(fs_info
, start
, end
, true);
2804 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
2805 free_extent_state(cached_state
);
2809 if (btrfs_test_opt(fs_info
, DISCARD_ASYNC
)) {
2810 btrfs_discard_calc_delay(&fs_info
->discard_ctl
);
2811 btrfs_discard_schedule_work(&fs_info
->discard_ctl
, true);
2815 * Transaction is finished. We don't need the lock anymore. We
2816 * do need to clean up the block groups in case of a transaction
2819 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
2820 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
2824 if (!TRANS_ABORTED(trans
))
2825 ret
= btrfs_discard_extent(fs_info
,
2827 block_group
->length
,
2830 list_del_init(&block_group
->bg_list
);
2831 btrfs_unfreeze_block_group(block_group
);
2832 btrfs_put_block_group(block_group
);
2835 const char *errstr
= btrfs_decode_error(ret
);
2837 "discard failed while removing blockgroup: errno=%d %s",
2845 static int do_free_extent_accounting(struct btrfs_trans_handle
*trans
,
2846 u64 bytenr
, u64 num_bytes
, bool is_data
)
2851 struct btrfs_root
*csum_root
;
2853 csum_root
= btrfs_csum_root(trans
->fs_info
, bytenr
);
2854 ret
= btrfs_del_csums(trans
, csum_root
, bytenr
, num_bytes
);
2856 btrfs_abort_transaction(trans
, ret
);
2861 ret
= add_to_free_space_tree(trans
, bytenr
, num_bytes
);
2863 btrfs_abort_transaction(trans
, ret
);
2867 ret
= btrfs_update_block_group(trans
, bytenr
, num_bytes
, false);
2869 btrfs_abort_transaction(trans
, ret
);
2874 #define abort_and_dump(trans, path, fmt, args...) \
2876 btrfs_abort_transaction(trans, -EUCLEAN); \
2877 btrfs_print_leaf(path->nodes[0]); \
2878 btrfs_crit(trans->fs_info, fmt, ##args); \
2882 * Drop one or more refs of @node.
2884 * 1. Locate the extent refs.
2885 * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
2886 * Locate it, then reduce the refs number or remove the ref line completely.
2888 * 2. Update the refs count in EXTENT/METADATA_ITEM
2890 * Inline backref case:
2892 * in extent tree we have:
2894 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2895 * refs 2 gen 6 flags DATA
2896 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2897 * extent data backref root FS_TREE objectid 257 offset 0 count 1
2899 * This function gets called with:
2901 * node->bytenr = 13631488
2902 * node->num_bytes = 1048576
2903 * root_objectid = FS_TREE
2904 * owner_objectid = 257
2908 * Then we should get some like:
2910 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2911 * refs 1 gen 6 flags DATA
2912 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2914 * Keyed backref case:
2916 * in extent tree we have:
2918 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2919 * refs 754 gen 6 flags DATA
2921 * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
2922 * extent data backref root FS_TREE objectid 866 offset 0 count 1
2924 * This function get called with:
2926 * node->bytenr = 13631488
2927 * node->num_bytes = 1048576
2928 * root_objectid = FS_TREE
2929 * owner_objectid = 866
2933 * Then we should get some like:
2935 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2936 * refs 753 gen 6 flags DATA
2938 * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
2940 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
2941 struct btrfs_delayed_ref_node
*node
, u64 parent
,
2942 u64 root_objectid
, u64 owner_objectid
,
2943 u64 owner_offset
, int refs_to_drop
,
2944 struct btrfs_delayed_extent_op
*extent_op
)
2946 struct btrfs_fs_info
*info
= trans
->fs_info
;
2947 struct btrfs_key key
;
2948 struct btrfs_path
*path
;
2949 struct btrfs_root
*extent_root
;
2950 struct extent_buffer
*leaf
;
2951 struct btrfs_extent_item
*ei
;
2952 struct btrfs_extent_inline_ref
*iref
;
2955 int extent_slot
= 0;
2956 int found_extent
= 0;
2960 u64 bytenr
= node
->bytenr
;
2961 u64 num_bytes
= node
->num_bytes
;
2962 bool skinny_metadata
= btrfs_fs_incompat(info
, SKINNY_METADATA
);
2964 extent_root
= btrfs_extent_root(info
, bytenr
);
2965 ASSERT(extent_root
);
2967 path
= btrfs_alloc_path();
2971 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
2973 if (!is_data
&& refs_to_drop
!= 1) {
2975 "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
2976 node
->bytenr
, refs_to_drop
);
2978 btrfs_abort_transaction(trans
, ret
);
2983 skinny_metadata
= false;
2985 ret
= lookup_extent_backref(trans
, path
, &iref
, bytenr
, num_bytes
,
2986 parent
, root_objectid
, owner_objectid
,
2990 * Either the inline backref or the SHARED_DATA_REF/
2991 * SHARED_BLOCK_REF is found
2993 * Here is a quick path to locate EXTENT/METADATA_ITEM.
2994 * It's possible the EXTENT/METADATA_ITEM is near current slot.
2996 extent_slot
= path
->slots
[0];
2997 while (extent_slot
>= 0) {
2998 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
3000 if (key
.objectid
!= bytenr
)
3002 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
3003 key
.offset
== num_bytes
) {
3007 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
3008 key
.offset
== owner_objectid
) {
3013 /* Quick path didn't find the EXTEMT/METADATA_ITEM */
3014 if (path
->slots
[0] - extent_slot
> 5)
3019 if (!found_extent
) {
3021 abort_and_dump(trans
, path
,
3022 "invalid iref slot %u, no EXTENT/METADATA_ITEM found but has inline extent ref",
3027 /* Must be SHARED_* item, remove the backref first */
3028 ret
= remove_extent_backref(trans
, extent_root
, path
,
3029 NULL
, refs_to_drop
, is_data
);
3031 btrfs_abort_transaction(trans
, ret
);
3034 btrfs_release_path(path
);
3036 /* Slow path to locate EXTENT/METADATA_ITEM */
3037 key
.objectid
= bytenr
;
3038 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3039 key
.offset
= num_bytes
;
3041 if (!is_data
&& skinny_metadata
) {
3042 key
.type
= BTRFS_METADATA_ITEM_KEY
;
3043 key
.offset
= owner_objectid
;
3046 ret
= btrfs_search_slot(trans
, extent_root
,
3048 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
3050 * Couldn't find our skinny metadata item,
3051 * see if we have ye olde extent item.
3054 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
3056 if (key
.objectid
== bytenr
&&
3057 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
3058 key
.offset
== num_bytes
)
3062 if (ret
> 0 && skinny_metadata
) {
3063 skinny_metadata
= false;
3064 key
.objectid
= bytenr
;
3065 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3066 key
.offset
= num_bytes
;
3067 btrfs_release_path(path
);
3068 ret
= btrfs_search_slot(trans
, extent_root
,
3074 btrfs_print_leaf(path
->nodes
[0]);
3076 "umm, got %d back from search, was looking for %llu, slot %d",
3077 ret
, bytenr
, path
->slots
[0]);
3080 btrfs_abort_transaction(trans
, ret
);
3083 extent_slot
= path
->slots
[0];
3085 } else if (WARN_ON(ret
== -ENOENT
)) {
3086 abort_and_dump(trans
, path
,
3087 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu slot %d",
3088 bytenr
, parent
, root_objectid
, owner_objectid
,
3089 owner_offset
, path
->slots
[0]);
3092 btrfs_abort_transaction(trans
, ret
);
3096 leaf
= path
->nodes
[0];
3097 item_size
= btrfs_item_size(leaf
, extent_slot
);
3098 if (unlikely(item_size
< sizeof(*ei
))) {
3100 btrfs_err(trans
->fs_info
,
3101 "unexpected extent item size, has %u expect >= %zu",
3102 item_size
, sizeof(*ei
));
3103 btrfs_abort_transaction(trans
, ret
);
3106 ei
= btrfs_item_ptr(leaf
, extent_slot
,
3107 struct btrfs_extent_item
);
3108 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
3109 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
3110 struct btrfs_tree_block_info
*bi
;
3112 if (item_size
< sizeof(*ei
) + sizeof(*bi
)) {
3113 abort_and_dump(trans
, path
,
3114 "invalid extent item size for key (%llu, %u, %llu) slot %u owner %llu, has %u expect >= %zu",
3115 key
.objectid
, key
.type
, key
.offset
,
3116 path
->slots
[0], owner_objectid
, item_size
,
3117 sizeof(*ei
) + sizeof(*bi
));
3121 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
3122 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
3125 refs
= btrfs_extent_refs(leaf
, ei
);
3126 if (refs
< refs_to_drop
) {
3127 abort_and_dump(trans
, path
,
3128 "trying to drop %d refs but we only have %llu for bytenr %llu slot %u",
3129 refs_to_drop
, refs
, bytenr
, path
->slots
[0]);
3133 refs
-= refs_to_drop
;
3137 __run_delayed_extent_op(extent_op
, leaf
, ei
);
3139 * In the case of inline back ref, reference count will
3140 * be updated by remove_extent_backref
3143 if (!found_extent
) {
3144 abort_and_dump(trans
, path
,
3145 "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found, slot %u",
3151 btrfs_set_extent_refs(leaf
, ei
, refs
);
3152 btrfs_mark_buffer_dirty(leaf
);
3155 ret
= remove_extent_backref(trans
, extent_root
, path
,
3156 iref
, refs_to_drop
, is_data
);
3158 btrfs_abort_transaction(trans
, ret
);
3163 /* In this branch refs == 1 */
3165 if (is_data
&& refs_to_drop
!=
3166 extent_data_ref_count(path
, iref
)) {
3167 abort_and_dump(trans
, path
,
3168 "invalid refs_to_drop, current refs %u refs_to_drop %u slot %u",
3169 extent_data_ref_count(path
, iref
),
3170 refs_to_drop
, path
->slots
[0]);
3175 if (path
->slots
[0] != extent_slot
) {
3176 abort_and_dump(trans
, path
,
3177 "invalid iref, extent item key (%llu %u %llu) slot %u doesn't have wanted iref",
3178 key
.objectid
, key
.type
,
3179 key
.offset
, path
->slots
[0]);
3185 * No inline ref, we must be at SHARED_* item,
3186 * And it's single ref, it must be:
3187 * | extent_slot ||extent_slot + 1|
3188 * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3190 if (path
->slots
[0] != extent_slot
+ 1) {
3191 abort_and_dump(trans
, path
,
3192 "invalid SHARED_* item slot %u, previous item is not EXTENT/METADATA_ITEM",
3197 path
->slots
[0] = extent_slot
;
3202 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
3205 btrfs_abort_transaction(trans
, ret
);
3208 btrfs_release_path(path
);
3210 ret
= do_free_extent_accounting(trans
, bytenr
, num_bytes
, is_data
);
3212 btrfs_release_path(path
);
3215 btrfs_free_path(path
);
3220 * when we free an block, it is possible (and likely) that we free the last
3221 * delayed ref for that extent as well. This searches the delayed ref tree for
3222 * a given extent, and if there are no other delayed refs to be processed, it
3223 * removes it from the tree.
3225 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
3228 struct btrfs_delayed_ref_head
*head
;
3229 struct btrfs_delayed_ref_root
*delayed_refs
;
3232 delayed_refs
= &trans
->transaction
->delayed_refs
;
3233 spin_lock(&delayed_refs
->lock
);
3234 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
3236 goto out_delayed_unlock
;
3238 spin_lock(&head
->lock
);
3239 if (!RB_EMPTY_ROOT(&head
->ref_tree
.rb_root
))
3242 if (cleanup_extent_op(head
) != NULL
)
3246 * waiting for the lock here would deadlock. If someone else has it
3247 * locked they are already in the process of dropping it anyway
3249 if (!mutex_trylock(&head
->mutex
))
3252 btrfs_delete_ref_head(delayed_refs
, head
);
3253 head
->processing
= false;
3255 spin_unlock(&head
->lock
);
3256 spin_unlock(&delayed_refs
->lock
);
3258 BUG_ON(head
->extent_op
);
3259 if (head
->must_insert_reserved
)
3262 btrfs_cleanup_ref_head_accounting(trans
->fs_info
, delayed_refs
, head
);
3263 mutex_unlock(&head
->mutex
);
3264 btrfs_put_delayed_ref_head(head
);
3267 spin_unlock(&head
->lock
);
3270 spin_unlock(&delayed_refs
->lock
);
3274 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
3276 struct extent_buffer
*buf
,
3277 u64 parent
, int last_ref
)
3279 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3280 struct btrfs_ref generic_ref
= { 0 };
3283 btrfs_init_generic_ref(&generic_ref
, BTRFS_DROP_DELAYED_REF
,
3284 buf
->start
, buf
->len
, parent
);
3285 btrfs_init_tree_ref(&generic_ref
, btrfs_header_level(buf
),
3288 if (root_id
!= BTRFS_TREE_LOG_OBJECTID
) {
3289 btrfs_ref_tree_mod(fs_info
, &generic_ref
);
3290 ret
= btrfs_add_delayed_tree_ref(trans
, &generic_ref
, NULL
);
3291 BUG_ON(ret
); /* -ENOMEM */
3294 if (last_ref
&& btrfs_header_generation(buf
) == trans
->transid
) {
3295 struct btrfs_block_group
*cache
;
3296 bool must_pin
= false;
3298 if (root_id
!= BTRFS_TREE_LOG_OBJECTID
) {
3299 ret
= check_ref_cleanup(trans
, buf
->start
);
3301 btrfs_redirty_list_add(trans
->transaction
, buf
);
3306 cache
= btrfs_lookup_block_group(fs_info
, buf
->start
);
3308 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
3309 pin_down_extent(trans
, cache
, buf
->start
, buf
->len
, 1);
3310 btrfs_put_block_group(cache
);
3315 * If there are tree mod log users we may have recorded mod log
3316 * operations for this node. If we re-allocate this node we
3317 * could replay operations on this node that happened when it
3318 * existed in a completely different root. For example if it
3319 * was part of root A, then was reallocated to root B, and we
3320 * are doing a btrfs_old_search_slot(root b), we could replay
3321 * operations that happened when the block was part of root A,
3322 * giving us an inconsistent view of the btree.
3324 * We are safe from races here because at this point no other
3325 * node or root points to this extent buffer, so if after this
3326 * check a new tree mod log user joins we will not have an
3327 * existing log of operations on this node that we have to
3330 if (test_bit(BTRFS_FS_TREE_MOD_LOG_USERS
, &fs_info
->flags
))
3333 if (must_pin
|| btrfs_is_zoned(fs_info
)) {
3334 btrfs_redirty_list_add(trans
->transaction
, buf
);
3335 pin_down_extent(trans
, cache
, buf
->start
, buf
->len
, 1);
3336 btrfs_put_block_group(cache
);
3340 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
3342 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
3343 btrfs_free_reserved_bytes(cache
, buf
->len
, 0);
3344 btrfs_put_block_group(cache
);
3345 trace_btrfs_reserved_extent_free(fs_info
, buf
->start
, buf
->len
);
3350 * Deleting the buffer, clear the corrupt flag since it doesn't
3353 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
3357 /* Can return -ENOMEM */
3358 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_ref
*ref
)
3360 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3363 if (btrfs_is_testing(fs_info
))
3367 * tree log blocks never actually go into the extent allocation
3368 * tree, just update pinning info and exit early.
3370 if ((ref
->type
== BTRFS_REF_METADATA
&&
3371 ref
->tree_ref
.owning_root
== BTRFS_TREE_LOG_OBJECTID
) ||
3372 (ref
->type
== BTRFS_REF_DATA
&&
3373 ref
->data_ref
.owning_root
== BTRFS_TREE_LOG_OBJECTID
)) {
3374 /* unlocks the pinned mutex */
3375 btrfs_pin_extent(trans
, ref
->bytenr
, ref
->len
, 1);
3377 } else if (ref
->type
== BTRFS_REF_METADATA
) {
3378 ret
= btrfs_add_delayed_tree_ref(trans
, ref
, NULL
);
3380 ret
= btrfs_add_delayed_data_ref(trans
, ref
, 0);
3383 if (!((ref
->type
== BTRFS_REF_METADATA
&&
3384 ref
->tree_ref
.owning_root
== BTRFS_TREE_LOG_OBJECTID
) ||
3385 (ref
->type
== BTRFS_REF_DATA
&&
3386 ref
->data_ref
.owning_root
== BTRFS_TREE_LOG_OBJECTID
)))
3387 btrfs_ref_tree_mod(fs_info
, ref
);
3392 enum btrfs_loop_type
{
3394 * Start caching block groups but do not wait for progress or for them
3397 LOOP_CACHING_NOWAIT
,
3400 * Wait for the block group free_space >= the space we're waiting for if
3401 * the block group isn't cached.
3406 * Allow allocations to happen from block groups that do not yet have a
3407 * size classification.
3409 LOOP_UNSET_SIZE_CLASS
,
3412 * Allocate a chunk and then retry the allocation.
3417 * Ignore the size class restrictions for this allocation.
3419 LOOP_WRONG_SIZE_CLASS
,
3422 * Ignore the empty size, only try to allocate the number of bytes
3423 * needed for this allocation.
3429 btrfs_lock_block_group(struct btrfs_block_group
*cache
,
3433 down_read(&cache
->data_rwsem
);
3436 static inline void btrfs_grab_block_group(struct btrfs_block_group
*cache
,
3439 btrfs_get_block_group(cache
);
3441 down_read(&cache
->data_rwsem
);
3444 static struct btrfs_block_group
*btrfs_lock_cluster(
3445 struct btrfs_block_group
*block_group
,
3446 struct btrfs_free_cluster
*cluster
,
3448 __acquires(&cluster
->refill_lock
)
3450 struct btrfs_block_group
*used_bg
= NULL
;
3452 spin_lock(&cluster
->refill_lock
);
3454 used_bg
= cluster
->block_group
;
3458 if (used_bg
== block_group
)
3461 btrfs_get_block_group(used_bg
);
3466 if (down_read_trylock(&used_bg
->data_rwsem
))
3469 spin_unlock(&cluster
->refill_lock
);
3471 /* We should only have one-level nested. */
3472 down_read_nested(&used_bg
->data_rwsem
, SINGLE_DEPTH_NESTING
);
3474 spin_lock(&cluster
->refill_lock
);
3475 if (used_bg
== cluster
->block_group
)
3478 up_read(&used_bg
->data_rwsem
);
3479 btrfs_put_block_group(used_bg
);
3484 btrfs_release_block_group(struct btrfs_block_group
*cache
,
3488 up_read(&cache
->data_rwsem
);
3489 btrfs_put_block_group(cache
);
3493 * Helper function for find_free_extent().
3495 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3496 * Return >0 to inform caller that we find nothing
3497 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3499 static int find_free_extent_clustered(struct btrfs_block_group
*bg
,
3500 struct find_free_extent_ctl
*ffe_ctl
,
3501 struct btrfs_block_group
**cluster_bg_ret
)
3503 struct btrfs_block_group
*cluster_bg
;
3504 struct btrfs_free_cluster
*last_ptr
= ffe_ctl
->last_ptr
;
3505 u64 aligned_cluster
;
3509 cluster_bg
= btrfs_lock_cluster(bg
, last_ptr
, ffe_ctl
->delalloc
);
3511 goto refill_cluster
;
3512 if (cluster_bg
!= bg
&& (cluster_bg
->ro
||
3513 !block_group_bits(cluster_bg
, ffe_ctl
->flags
)))
3514 goto release_cluster
;
3516 offset
= btrfs_alloc_from_cluster(cluster_bg
, last_ptr
,
3517 ffe_ctl
->num_bytes
, cluster_bg
->start
,
3518 &ffe_ctl
->max_extent_size
);
3520 /* We have a block, we're done */
3521 spin_unlock(&last_ptr
->refill_lock
);
3522 trace_btrfs_reserve_extent_cluster(cluster_bg
, ffe_ctl
);
3523 *cluster_bg_ret
= cluster_bg
;
3524 ffe_ctl
->found_offset
= offset
;
3527 WARN_ON(last_ptr
->block_group
!= cluster_bg
);
3531 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3532 * lets just skip it and let the allocator find whatever block it can
3533 * find. If we reach this point, we will have tried the cluster
3534 * allocator plenty of times and not have found anything, so we are
3535 * likely way too fragmented for the clustering stuff to find anything.
3537 * However, if the cluster is taken from the current block group,
3538 * release the cluster first, so that we stand a better chance of
3539 * succeeding in the unclustered allocation.
3541 if (ffe_ctl
->loop
>= LOOP_NO_EMPTY_SIZE
&& cluster_bg
!= bg
) {
3542 spin_unlock(&last_ptr
->refill_lock
);
3543 btrfs_release_block_group(cluster_bg
, ffe_ctl
->delalloc
);
3547 /* This cluster didn't work out, free it and start over */
3548 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
3550 if (cluster_bg
!= bg
)
3551 btrfs_release_block_group(cluster_bg
, ffe_ctl
->delalloc
);
3554 if (ffe_ctl
->loop
>= LOOP_NO_EMPTY_SIZE
) {
3555 spin_unlock(&last_ptr
->refill_lock
);
3559 aligned_cluster
= max_t(u64
,
3560 ffe_ctl
->empty_cluster
+ ffe_ctl
->empty_size
,
3561 bg
->full_stripe_len
);
3562 ret
= btrfs_find_space_cluster(bg
, last_ptr
, ffe_ctl
->search_start
,
3563 ffe_ctl
->num_bytes
, aligned_cluster
);
3565 /* Now pull our allocation out of this cluster */
3566 offset
= btrfs_alloc_from_cluster(bg
, last_ptr
,
3567 ffe_ctl
->num_bytes
, ffe_ctl
->search_start
,
3568 &ffe_ctl
->max_extent_size
);
3570 /* We found one, proceed */
3571 spin_unlock(&last_ptr
->refill_lock
);
3572 ffe_ctl
->found_offset
= offset
;
3573 trace_btrfs_reserve_extent_cluster(bg
, ffe_ctl
);
3578 * At this point we either didn't find a cluster or we weren't able to
3579 * allocate a block from our cluster. Free the cluster we've been
3580 * trying to use, and go to the next block group.
3582 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
3583 spin_unlock(&last_ptr
->refill_lock
);
3588 * Return >0 to inform caller that we find nothing
3589 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3591 static int find_free_extent_unclustered(struct btrfs_block_group
*bg
,
3592 struct find_free_extent_ctl
*ffe_ctl
)
3594 struct btrfs_free_cluster
*last_ptr
= ffe_ctl
->last_ptr
;
3598 * We are doing an unclustered allocation, set the fragmented flag so
3599 * we don't bother trying to setup a cluster again until we get more
3602 if (unlikely(last_ptr
)) {
3603 spin_lock(&last_ptr
->lock
);
3604 last_ptr
->fragmented
= 1;
3605 spin_unlock(&last_ptr
->lock
);
3607 if (ffe_ctl
->cached
) {
3608 struct btrfs_free_space_ctl
*free_space_ctl
;
3610 free_space_ctl
= bg
->free_space_ctl
;
3611 spin_lock(&free_space_ctl
->tree_lock
);
3612 if (free_space_ctl
->free_space
<
3613 ffe_ctl
->num_bytes
+ ffe_ctl
->empty_cluster
+
3614 ffe_ctl
->empty_size
) {
3615 ffe_ctl
->total_free_space
= max_t(u64
,
3616 ffe_ctl
->total_free_space
,
3617 free_space_ctl
->free_space
);
3618 spin_unlock(&free_space_ctl
->tree_lock
);
3621 spin_unlock(&free_space_ctl
->tree_lock
);
3624 offset
= btrfs_find_space_for_alloc(bg
, ffe_ctl
->search_start
,
3625 ffe_ctl
->num_bytes
, ffe_ctl
->empty_size
,
3626 &ffe_ctl
->max_extent_size
);
3629 ffe_ctl
->found_offset
= offset
;
3633 static int do_allocation_clustered(struct btrfs_block_group
*block_group
,
3634 struct find_free_extent_ctl
*ffe_ctl
,
3635 struct btrfs_block_group
**bg_ret
)
3639 /* We want to try and use the cluster allocator, so lets look there */
3640 if (ffe_ctl
->last_ptr
&& ffe_ctl
->use_cluster
) {
3641 ret
= find_free_extent_clustered(block_group
, ffe_ctl
, bg_ret
);
3644 /* ret == -ENOENT case falls through */
3647 return find_free_extent_unclustered(block_group
, ffe_ctl
);
3651 * Tree-log block group locking
3652 * ============================
3654 * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3655 * indicates the starting address of a block group, which is reserved only
3656 * for tree-log metadata.
3663 * fs_info::treelog_bg_lock
3667 * Simple allocator for sequential-only block group. It only allows sequential
3668 * allocation. No need to play with trees. This function also reserves the
3669 * bytes as in btrfs_add_reserved_bytes.
3671 static int do_allocation_zoned(struct btrfs_block_group
*block_group
,
3672 struct find_free_extent_ctl
*ffe_ctl
,
3673 struct btrfs_block_group
**bg_ret
)
3675 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
3676 struct btrfs_space_info
*space_info
= block_group
->space_info
;
3677 struct btrfs_free_space_ctl
*ctl
= block_group
->free_space_ctl
;
3678 u64 start
= block_group
->start
;
3679 u64 num_bytes
= ffe_ctl
->num_bytes
;
3681 u64 bytenr
= block_group
->start
;
3683 u64 data_reloc_bytenr
;
3687 ASSERT(btrfs_is_zoned(block_group
->fs_info
));
3690 * Do not allow non-tree-log blocks in the dedicated tree-log block
3691 * group, and vice versa.
3693 spin_lock(&fs_info
->treelog_bg_lock
);
3694 log_bytenr
= fs_info
->treelog_bg
;
3695 if (log_bytenr
&& ((ffe_ctl
->for_treelog
&& bytenr
!= log_bytenr
) ||
3696 (!ffe_ctl
->for_treelog
&& bytenr
== log_bytenr
)))
3698 spin_unlock(&fs_info
->treelog_bg_lock
);
3703 * Do not allow non-relocation blocks in the dedicated relocation block
3704 * group, and vice versa.
3706 spin_lock(&fs_info
->relocation_bg_lock
);
3707 data_reloc_bytenr
= fs_info
->data_reloc_bg
;
3708 if (data_reloc_bytenr
&&
3709 ((ffe_ctl
->for_data_reloc
&& bytenr
!= data_reloc_bytenr
) ||
3710 (!ffe_ctl
->for_data_reloc
&& bytenr
== data_reloc_bytenr
)))
3712 spin_unlock(&fs_info
->relocation_bg_lock
);
3716 /* Check RO and no space case before trying to activate it */
3717 spin_lock(&block_group
->lock
);
3718 if (block_group
->ro
|| btrfs_zoned_bg_is_full(block_group
)) {
3721 * May need to clear fs_info->{treelog,data_reloc}_bg.
3722 * Return the error after taking the locks.
3725 spin_unlock(&block_group
->lock
);
3727 /* Metadata block group is activated at write time. */
3728 if (!ret
&& (block_group
->flags
& BTRFS_BLOCK_GROUP_DATA
) &&
3729 !btrfs_zone_activate(block_group
)) {
3732 * May need to clear fs_info->{treelog,data_reloc}_bg.
3733 * Return the error after taking the locks.
3737 spin_lock(&space_info
->lock
);
3738 spin_lock(&block_group
->lock
);
3739 spin_lock(&fs_info
->treelog_bg_lock
);
3740 spin_lock(&fs_info
->relocation_bg_lock
);
3745 ASSERT(!ffe_ctl
->for_treelog
||
3746 block_group
->start
== fs_info
->treelog_bg
||
3747 fs_info
->treelog_bg
== 0);
3748 ASSERT(!ffe_ctl
->for_data_reloc
||
3749 block_group
->start
== fs_info
->data_reloc_bg
||
3750 fs_info
->data_reloc_bg
== 0);
3752 if (block_group
->ro
||
3753 (!ffe_ctl
->for_data_reloc
&&
3754 test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC
, &block_group
->runtime_flags
))) {
3760 * Do not allow currently using block group to be tree-log dedicated
3763 if (ffe_ctl
->for_treelog
&& !fs_info
->treelog_bg
&&
3764 (block_group
->used
|| block_group
->reserved
)) {
3770 * Do not allow currently used block group to be the data relocation
3771 * dedicated block group.
3773 if (ffe_ctl
->for_data_reloc
&& !fs_info
->data_reloc_bg
&&
3774 (block_group
->used
|| block_group
->reserved
)) {
3779 WARN_ON_ONCE(block_group
->alloc_offset
> block_group
->zone_capacity
);
3780 avail
= block_group
->zone_capacity
- block_group
->alloc_offset
;
3781 if (avail
< num_bytes
) {
3782 if (ffe_ctl
->max_extent_size
< avail
) {
3784 * With sequential allocator, free space is always
3787 ffe_ctl
->max_extent_size
= avail
;
3788 ffe_ctl
->total_free_space
= avail
;
3794 if (ffe_ctl
->for_treelog
&& !fs_info
->treelog_bg
)
3795 fs_info
->treelog_bg
= block_group
->start
;
3797 if (ffe_ctl
->for_data_reloc
) {
3798 if (!fs_info
->data_reloc_bg
)
3799 fs_info
->data_reloc_bg
= block_group
->start
;
3801 * Do not allow allocations from this block group, unless it is
3802 * for data relocation. Compared to increasing the ->ro, setting
3803 * the ->zoned_data_reloc_ongoing flag still allows nocow
3804 * writers to come in. See btrfs_inc_nocow_writers().
3806 * We need to disable an allocation to avoid an allocation of
3807 * regular (non-relocation data) extent. With mix of relocation
3808 * extents and regular extents, we can dispatch WRITE commands
3809 * (for relocation extents) and ZONE APPEND commands (for
3810 * regular extents) at the same time to the same zone, which
3811 * easily break the write pointer.
3813 * Also, this flag avoids this block group to be zone finished.
3815 set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC
, &block_group
->runtime_flags
);
3818 ffe_ctl
->found_offset
= start
+ block_group
->alloc_offset
;
3819 block_group
->alloc_offset
+= num_bytes
;
3820 spin_lock(&ctl
->tree_lock
);
3821 ctl
->free_space
-= num_bytes
;
3822 spin_unlock(&ctl
->tree_lock
);
3825 * We do not check if found_offset is aligned to stripesize. The
3826 * address is anyway rewritten when using zone append writing.
3829 ffe_ctl
->search_start
= ffe_ctl
->found_offset
;
3832 if (ret
&& ffe_ctl
->for_treelog
)
3833 fs_info
->treelog_bg
= 0;
3834 if (ret
&& ffe_ctl
->for_data_reloc
)
3835 fs_info
->data_reloc_bg
= 0;
3836 spin_unlock(&fs_info
->relocation_bg_lock
);
3837 spin_unlock(&fs_info
->treelog_bg_lock
);
3838 spin_unlock(&block_group
->lock
);
3839 spin_unlock(&space_info
->lock
);
3843 static int do_allocation(struct btrfs_block_group
*block_group
,
3844 struct find_free_extent_ctl
*ffe_ctl
,
3845 struct btrfs_block_group
**bg_ret
)
3847 switch (ffe_ctl
->policy
) {
3848 case BTRFS_EXTENT_ALLOC_CLUSTERED
:
3849 return do_allocation_clustered(block_group
, ffe_ctl
, bg_ret
);
3850 case BTRFS_EXTENT_ALLOC_ZONED
:
3851 return do_allocation_zoned(block_group
, ffe_ctl
, bg_ret
);
3857 static void release_block_group(struct btrfs_block_group
*block_group
,
3858 struct find_free_extent_ctl
*ffe_ctl
,
3861 switch (ffe_ctl
->policy
) {
3862 case BTRFS_EXTENT_ALLOC_CLUSTERED
:
3863 ffe_ctl
->retry_uncached
= false;
3865 case BTRFS_EXTENT_ALLOC_ZONED
:
3872 BUG_ON(btrfs_bg_flags_to_raid_index(block_group
->flags
) !=
3874 btrfs_release_block_group(block_group
, delalloc
);
3877 static void found_extent_clustered(struct find_free_extent_ctl
*ffe_ctl
,
3878 struct btrfs_key
*ins
)
3880 struct btrfs_free_cluster
*last_ptr
= ffe_ctl
->last_ptr
;
3882 if (!ffe_ctl
->use_cluster
&& last_ptr
) {
3883 spin_lock(&last_ptr
->lock
);
3884 last_ptr
->window_start
= ins
->objectid
;
3885 spin_unlock(&last_ptr
->lock
);
3889 static void found_extent(struct find_free_extent_ctl
*ffe_ctl
,
3890 struct btrfs_key
*ins
)
3892 switch (ffe_ctl
->policy
) {
3893 case BTRFS_EXTENT_ALLOC_CLUSTERED
:
3894 found_extent_clustered(ffe_ctl
, ins
);
3896 case BTRFS_EXTENT_ALLOC_ZONED
:
3904 static int can_allocate_chunk_zoned(struct btrfs_fs_info
*fs_info
,
3905 struct find_free_extent_ctl
*ffe_ctl
)
3907 /* Block group's activeness is not a requirement for METADATA block groups. */
3908 if (!(ffe_ctl
->flags
& BTRFS_BLOCK_GROUP_DATA
))
3911 /* If we can activate new zone, just allocate a chunk and use it */
3912 if (btrfs_can_activate_zone(fs_info
->fs_devices
, ffe_ctl
->flags
))
3916 * We already reached the max active zones. Try to finish one block
3917 * group to make a room for a new block group. This is only possible
3918 * for a data block group because btrfs_zone_finish() may need to wait
3919 * for a running transaction which can cause a deadlock for metadata
3922 if (ffe_ctl
->flags
& BTRFS_BLOCK_GROUP_DATA
) {
3923 int ret
= btrfs_zone_finish_one_bg(fs_info
);
3932 * If we have enough free space left in an already active block group
3933 * and we can't activate any other zone now, do not allow allocating a
3934 * new chunk and let find_free_extent() retry with a smaller size.
3936 if (ffe_ctl
->max_extent_size
>= ffe_ctl
->min_alloc_size
)
3940 * Even min_alloc_size is not left in any block groups. Since we cannot
3941 * activate a new block group, allocating it may not help. Let's tell a
3942 * caller to try again and hope it progress something by writing some
3943 * parts of the region. That is only possible for data block groups,
3944 * where a part of the region can be written.
3946 if (ffe_ctl
->flags
& BTRFS_BLOCK_GROUP_DATA
)
3950 * We cannot activate a new block group and no enough space left in any
3951 * block groups. So, allocating a new block group may not help. But,
3952 * there is nothing to do anyway, so let's go with it.
3957 static int can_allocate_chunk(struct btrfs_fs_info
*fs_info
,
3958 struct find_free_extent_ctl
*ffe_ctl
)
3960 switch (ffe_ctl
->policy
) {
3961 case BTRFS_EXTENT_ALLOC_CLUSTERED
:
3963 case BTRFS_EXTENT_ALLOC_ZONED
:
3964 return can_allocate_chunk_zoned(fs_info
, ffe_ctl
);
3971 * Return >0 means caller needs to re-search for free extent
3972 * Return 0 means we have the needed free extent.
3973 * Return <0 means we failed to locate any free extent.
3975 static int find_free_extent_update_loop(struct btrfs_fs_info
*fs_info
,
3976 struct btrfs_key
*ins
,
3977 struct find_free_extent_ctl
*ffe_ctl
,
3980 struct btrfs_root
*root
= fs_info
->chunk_root
;
3983 if ((ffe_ctl
->loop
== LOOP_CACHING_NOWAIT
) &&
3984 ffe_ctl
->have_caching_bg
&& !ffe_ctl
->orig_have_caching_bg
)
3985 ffe_ctl
->orig_have_caching_bg
= true;
3987 if (ins
->objectid
) {
3988 found_extent(ffe_ctl
, ins
);
3992 if (ffe_ctl
->loop
>= LOOP_CACHING_WAIT
&& ffe_ctl
->have_caching_bg
)
3996 if (ffe_ctl
->index
< BTRFS_NR_RAID_TYPES
)
3999 /* See the comments for btrfs_loop_type for an explanation of the phases. */
4000 if (ffe_ctl
->loop
< LOOP_NO_EMPTY_SIZE
) {
4003 * We want to skip the LOOP_CACHING_WAIT step if we don't have
4004 * any uncached bgs and we've already done a full search
4007 if (ffe_ctl
->loop
== LOOP_CACHING_NOWAIT
&&
4008 (!ffe_ctl
->orig_have_caching_bg
&& full_search
))
4012 if (ffe_ctl
->loop
== LOOP_ALLOC_CHUNK
) {
4013 struct btrfs_trans_handle
*trans
;
4016 /* Check if allocation policy allows to create a new chunk */
4017 ret
= can_allocate_chunk(fs_info
, ffe_ctl
);
4021 trans
= current
->journal_info
;
4025 trans
= btrfs_join_transaction(root
);
4027 if (IS_ERR(trans
)) {
4028 ret
= PTR_ERR(trans
);
4032 ret
= btrfs_chunk_alloc(trans
, ffe_ctl
->flags
,
4033 CHUNK_ALLOC_FORCE_FOR_EXTENT
);
4035 /* Do not bail out on ENOSPC since we can do more. */
4036 if (ret
== -ENOSPC
) {
4041 btrfs_abort_transaction(trans
, ret
);
4045 btrfs_end_transaction(trans
);
4050 if (ffe_ctl
->loop
== LOOP_NO_EMPTY_SIZE
) {
4051 if (ffe_ctl
->policy
!= BTRFS_EXTENT_ALLOC_CLUSTERED
)
4055 * Don't loop again if we already have no empty_size and
4058 if (ffe_ctl
->empty_size
== 0 &&
4059 ffe_ctl
->empty_cluster
== 0)
4061 ffe_ctl
->empty_size
= 0;
4062 ffe_ctl
->empty_cluster
= 0;
4069 static bool find_free_extent_check_size_class(struct find_free_extent_ctl
*ffe_ctl
,
4070 struct btrfs_block_group
*bg
)
4072 if (ffe_ctl
->policy
== BTRFS_EXTENT_ALLOC_ZONED
)
4074 if (!btrfs_block_group_should_use_size_class(bg
))
4076 if (ffe_ctl
->loop
>= LOOP_WRONG_SIZE_CLASS
)
4078 if (ffe_ctl
->loop
>= LOOP_UNSET_SIZE_CLASS
&&
4079 bg
->size_class
== BTRFS_BG_SZ_NONE
)
4081 return ffe_ctl
->size_class
== bg
->size_class
;
4084 static int prepare_allocation_clustered(struct btrfs_fs_info
*fs_info
,
4085 struct find_free_extent_ctl
*ffe_ctl
,
4086 struct btrfs_space_info
*space_info
,
4087 struct btrfs_key
*ins
)
4090 * If our free space is heavily fragmented we may not be able to make
4091 * big contiguous allocations, so instead of doing the expensive search
4092 * for free space, simply return ENOSPC with our max_extent_size so we
4093 * can go ahead and search for a more manageable chunk.
4095 * If our max_extent_size is large enough for our allocation simply
4096 * disable clustering since we will likely not be able to find enough
4097 * space to create a cluster and induce latency trying.
4099 if (space_info
->max_extent_size
) {
4100 spin_lock(&space_info
->lock
);
4101 if (space_info
->max_extent_size
&&
4102 ffe_ctl
->num_bytes
> space_info
->max_extent_size
) {
4103 ins
->offset
= space_info
->max_extent_size
;
4104 spin_unlock(&space_info
->lock
);
4106 } else if (space_info
->max_extent_size
) {
4107 ffe_ctl
->use_cluster
= false;
4109 spin_unlock(&space_info
->lock
);
4112 ffe_ctl
->last_ptr
= fetch_cluster_info(fs_info
, space_info
,
4113 &ffe_ctl
->empty_cluster
);
4114 if (ffe_ctl
->last_ptr
) {
4115 struct btrfs_free_cluster
*last_ptr
= ffe_ctl
->last_ptr
;
4117 spin_lock(&last_ptr
->lock
);
4118 if (last_ptr
->block_group
)
4119 ffe_ctl
->hint_byte
= last_ptr
->window_start
;
4120 if (last_ptr
->fragmented
) {
4122 * We still set window_start so we can keep track of the
4123 * last place we found an allocation to try and save
4126 ffe_ctl
->hint_byte
= last_ptr
->window_start
;
4127 ffe_ctl
->use_cluster
= false;
4129 spin_unlock(&last_ptr
->lock
);
4135 static int prepare_allocation(struct btrfs_fs_info
*fs_info
,
4136 struct find_free_extent_ctl
*ffe_ctl
,
4137 struct btrfs_space_info
*space_info
,
4138 struct btrfs_key
*ins
)
4140 switch (ffe_ctl
->policy
) {
4141 case BTRFS_EXTENT_ALLOC_CLUSTERED
:
4142 return prepare_allocation_clustered(fs_info
, ffe_ctl
,
4144 case BTRFS_EXTENT_ALLOC_ZONED
:
4145 if (ffe_ctl
->for_treelog
) {
4146 spin_lock(&fs_info
->treelog_bg_lock
);
4147 if (fs_info
->treelog_bg
)
4148 ffe_ctl
->hint_byte
= fs_info
->treelog_bg
;
4149 spin_unlock(&fs_info
->treelog_bg_lock
);
4151 if (ffe_ctl
->for_data_reloc
) {
4152 spin_lock(&fs_info
->relocation_bg_lock
);
4153 if (fs_info
->data_reloc_bg
)
4154 ffe_ctl
->hint_byte
= fs_info
->data_reloc_bg
;
4155 spin_unlock(&fs_info
->relocation_bg_lock
);
4164 * walks the btree of allocated extents and find a hole of a given size.
4165 * The key ins is changed to record the hole:
4166 * ins->objectid == start position
4167 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4168 * ins->offset == the size of the hole.
4169 * Any available blocks before search_start are skipped.
4171 * If there is no suitable free space, we will record the max size of
4172 * the free space extent currently.
4174 * The overall logic and call chain:
4176 * find_free_extent()
4177 * |- Iterate through all block groups
4178 * | |- Get a valid block group
4179 * | |- Try to do clustered allocation in that block group
4180 * | |- Try to do unclustered allocation in that block group
4181 * | |- Check if the result is valid
4182 * | | |- If valid, then exit
4183 * | |- Jump to next block group
4185 * |- Push harder to find free extents
4186 * |- If not found, re-iterate all block groups
4188 static noinline
int find_free_extent(struct btrfs_root
*root
,
4189 struct btrfs_key
*ins
,
4190 struct find_free_extent_ctl
*ffe_ctl
)
4192 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4194 int cache_block_group_error
= 0;
4195 struct btrfs_block_group
*block_group
= NULL
;
4196 struct btrfs_space_info
*space_info
;
4197 bool full_search
= false;
4199 WARN_ON(ffe_ctl
->num_bytes
< fs_info
->sectorsize
);
4201 ffe_ctl
->search_start
= 0;
4202 /* For clustered allocation */
4203 ffe_ctl
->empty_cluster
= 0;
4204 ffe_ctl
->last_ptr
= NULL
;
4205 ffe_ctl
->use_cluster
= true;
4206 ffe_ctl
->have_caching_bg
= false;
4207 ffe_ctl
->orig_have_caching_bg
= false;
4208 ffe_ctl
->index
= btrfs_bg_flags_to_raid_index(ffe_ctl
->flags
);
4210 ffe_ctl
->retry_uncached
= false;
4211 ffe_ctl
->cached
= 0;
4212 ffe_ctl
->max_extent_size
= 0;
4213 ffe_ctl
->total_free_space
= 0;
4214 ffe_ctl
->found_offset
= 0;
4215 ffe_ctl
->policy
= BTRFS_EXTENT_ALLOC_CLUSTERED
;
4216 ffe_ctl
->size_class
= btrfs_calc_block_group_size_class(ffe_ctl
->num_bytes
);
4218 if (btrfs_is_zoned(fs_info
))
4219 ffe_ctl
->policy
= BTRFS_EXTENT_ALLOC_ZONED
;
4221 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
4225 trace_find_free_extent(root
, ffe_ctl
);
4227 space_info
= btrfs_find_space_info(fs_info
, ffe_ctl
->flags
);
4229 btrfs_err(fs_info
, "No space info for %llu", ffe_ctl
->flags
);
4233 ret
= prepare_allocation(fs_info
, ffe_ctl
, space_info
, ins
);
4237 ffe_ctl
->search_start
= max(ffe_ctl
->search_start
,
4238 first_logical_byte(fs_info
));
4239 ffe_ctl
->search_start
= max(ffe_ctl
->search_start
, ffe_ctl
->hint_byte
);
4240 if (ffe_ctl
->search_start
== ffe_ctl
->hint_byte
) {
4241 block_group
= btrfs_lookup_block_group(fs_info
,
4242 ffe_ctl
->search_start
);
4244 * we don't want to use the block group if it doesn't match our
4245 * allocation bits, or if its not cached.
4247 * However if we are re-searching with an ideal block group
4248 * picked out then we don't care that the block group is cached.
4250 if (block_group
&& block_group_bits(block_group
, ffe_ctl
->flags
) &&
4251 block_group
->cached
!= BTRFS_CACHE_NO
) {
4252 down_read(&space_info
->groups_sem
);
4253 if (list_empty(&block_group
->list
) ||
4256 * someone is removing this block group,
4257 * we can't jump into the have_block_group
4258 * target because our list pointers are not
4261 btrfs_put_block_group(block_group
);
4262 up_read(&space_info
->groups_sem
);
4264 ffe_ctl
->index
= btrfs_bg_flags_to_raid_index(
4265 block_group
->flags
);
4266 btrfs_lock_block_group(block_group
,
4268 ffe_ctl
->hinted
= true;
4269 goto have_block_group
;
4271 } else if (block_group
) {
4272 btrfs_put_block_group(block_group
);
4276 trace_find_free_extent_search_loop(root
, ffe_ctl
);
4277 ffe_ctl
->have_caching_bg
= false;
4278 if (ffe_ctl
->index
== btrfs_bg_flags_to_raid_index(ffe_ctl
->flags
) ||
4279 ffe_ctl
->index
== 0)
4281 down_read(&space_info
->groups_sem
);
4282 list_for_each_entry(block_group
,
4283 &space_info
->block_groups
[ffe_ctl
->index
], list
) {
4284 struct btrfs_block_group
*bg_ret
;
4286 ffe_ctl
->hinted
= false;
4287 /* If the block group is read-only, we can skip it entirely. */
4288 if (unlikely(block_group
->ro
)) {
4289 if (ffe_ctl
->for_treelog
)
4290 btrfs_clear_treelog_bg(block_group
);
4291 if (ffe_ctl
->for_data_reloc
)
4292 btrfs_clear_data_reloc_bg(block_group
);
4296 btrfs_grab_block_group(block_group
, ffe_ctl
->delalloc
);
4297 ffe_ctl
->search_start
= block_group
->start
;
4300 * this can happen if we end up cycling through all the
4301 * raid types, but we want to make sure we only allocate
4302 * for the proper type.
4304 if (!block_group_bits(block_group
, ffe_ctl
->flags
)) {
4305 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
4306 BTRFS_BLOCK_GROUP_RAID1_MASK
|
4307 BTRFS_BLOCK_GROUP_RAID56_MASK
|
4308 BTRFS_BLOCK_GROUP_RAID10
;
4311 * if they asked for extra copies and this block group
4312 * doesn't provide them, bail. This does allow us to
4313 * fill raid0 from raid1.
4315 if ((ffe_ctl
->flags
& extra
) && !(block_group
->flags
& extra
))
4319 * This block group has different flags than we want.
4320 * It's possible that we have MIXED_GROUP flag but no
4321 * block group is mixed. Just skip such block group.
4323 btrfs_release_block_group(block_group
, ffe_ctl
->delalloc
);
4328 trace_find_free_extent_have_block_group(root
, ffe_ctl
, block_group
);
4329 ffe_ctl
->cached
= btrfs_block_group_done(block_group
);
4330 if (unlikely(!ffe_ctl
->cached
)) {
4331 ffe_ctl
->have_caching_bg
= true;
4332 ret
= btrfs_cache_block_group(block_group
, false);
4335 * If we get ENOMEM here or something else we want to
4336 * try other block groups, because it may not be fatal.
4337 * However if we can't find anything else we need to
4338 * save our return here so that we return the actual
4339 * error that caused problems, not ENOSPC.
4342 if (!cache_block_group_error
)
4343 cache_block_group_error
= ret
;
4350 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
)) {
4351 if (!cache_block_group_error
)
4352 cache_block_group_error
= -EIO
;
4356 if (!find_free_extent_check_size_class(ffe_ctl
, block_group
))
4360 ret
= do_allocation(block_group
, ffe_ctl
, &bg_ret
);
4364 if (bg_ret
&& bg_ret
!= block_group
) {
4365 btrfs_release_block_group(block_group
, ffe_ctl
->delalloc
);
4366 block_group
= bg_ret
;
4370 ffe_ctl
->search_start
= round_up(ffe_ctl
->found_offset
,
4371 fs_info
->stripesize
);
4373 /* move on to the next group */
4374 if (ffe_ctl
->search_start
+ ffe_ctl
->num_bytes
>
4375 block_group
->start
+ block_group
->length
) {
4376 btrfs_add_free_space_unused(block_group
,
4377 ffe_ctl
->found_offset
,
4378 ffe_ctl
->num_bytes
);
4382 if (ffe_ctl
->found_offset
< ffe_ctl
->search_start
)
4383 btrfs_add_free_space_unused(block_group
,
4384 ffe_ctl
->found_offset
,
4385 ffe_ctl
->search_start
- ffe_ctl
->found_offset
);
4387 ret
= btrfs_add_reserved_bytes(block_group
, ffe_ctl
->ram_bytes
,
4390 ffe_ctl
->loop
>= LOOP_WRONG_SIZE_CLASS
);
4391 if (ret
== -EAGAIN
) {
4392 btrfs_add_free_space_unused(block_group
,
4393 ffe_ctl
->found_offset
,
4394 ffe_ctl
->num_bytes
);
4397 btrfs_inc_block_group_reservations(block_group
);
4399 /* we are all good, lets return */
4400 ins
->objectid
= ffe_ctl
->search_start
;
4401 ins
->offset
= ffe_ctl
->num_bytes
;
4403 trace_btrfs_reserve_extent(block_group
, ffe_ctl
);
4404 btrfs_release_block_group(block_group
, ffe_ctl
->delalloc
);
4407 if (!ffe_ctl
->cached
&& ffe_ctl
->loop
> LOOP_CACHING_NOWAIT
&&
4408 !ffe_ctl
->retry_uncached
) {
4409 ffe_ctl
->retry_uncached
= true;
4410 btrfs_wait_block_group_cache_progress(block_group
,
4411 ffe_ctl
->num_bytes
+
4412 ffe_ctl
->empty_cluster
+
4413 ffe_ctl
->empty_size
);
4414 goto have_block_group
;
4416 release_block_group(block_group
, ffe_ctl
, ffe_ctl
->delalloc
);
4419 up_read(&space_info
->groups_sem
);
4421 ret
= find_free_extent_update_loop(fs_info
, ins
, ffe_ctl
, full_search
);
4425 if (ret
== -ENOSPC
&& !cache_block_group_error
) {
4427 * Use ffe_ctl->total_free_space as fallback if we can't find
4428 * any contiguous hole.
4430 if (!ffe_ctl
->max_extent_size
)
4431 ffe_ctl
->max_extent_size
= ffe_ctl
->total_free_space
;
4432 spin_lock(&space_info
->lock
);
4433 space_info
->max_extent_size
= ffe_ctl
->max_extent_size
;
4434 spin_unlock(&space_info
->lock
);
4435 ins
->offset
= ffe_ctl
->max_extent_size
;
4436 } else if (ret
== -ENOSPC
) {
4437 ret
= cache_block_group_error
;
4443 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4444 * hole that is at least as big as @num_bytes.
4446 * @root - The root that will contain this extent
4448 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4449 * is used for accounting purposes. This value differs
4450 * from @num_bytes only in the case of compressed extents.
4452 * @num_bytes - Number of bytes to allocate on-disk.
4454 * @min_alloc_size - Indicates the minimum amount of space that the
4455 * allocator should try to satisfy. In some cases
4456 * @num_bytes may be larger than what is required and if
4457 * the filesystem is fragmented then allocation fails.
4458 * However, the presence of @min_alloc_size gives a
4459 * chance to try and satisfy the smaller allocation.
4461 * @empty_size - A hint that you plan on doing more COW. This is the
4462 * size in bytes the allocator should try to find free
4463 * next to the block it returns. This is just a hint and
4464 * may be ignored by the allocator.
4466 * @hint_byte - Hint to the allocator to start searching above the byte
4467 * address passed. It might be ignored.
4469 * @ins - This key is modified to record the found hole. It will
4470 * have the following values:
4471 * ins->objectid == start position
4472 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4473 * ins->offset == the size of the hole.
4475 * @is_data - Boolean flag indicating whether an extent is
4476 * allocated for data (true) or metadata (false)
4478 * @delalloc - Boolean flag indicating whether this allocation is for
4479 * delalloc or not. If 'true' data_rwsem of block groups
4480 * is going to be acquired.
4483 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4484 * case -ENOSPC is returned then @ins->offset will contain the size of the
4485 * largest available hole the allocator managed to find.
4487 int btrfs_reserve_extent(struct btrfs_root
*root
, u64 ram_bytes
,
4488 u64 num_bytes
, u64 min_alloc_size
,
4489 u64 empty_size
, u64 hint_byte
,
4490 struct btrfs_key
*ins
, int is_data
, int delalloc
)
4492 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4493 struct find_free_extent_ctl ffe_ctl
= {};
4494 bool final_tried
= num_bytes
== min_alloc_size
;
4497 bool for_treelog
= (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
);
4498 bool for_data_reloc
= (btrfs_is_data_reloc_root(root
) && is_data
);
4500 flags
= get_alloc_profile_by_root(root
, is_data
);
4502 WARN_ON(num_bytes
< fs_info
->sectorsize
);
4504 ffe_ctl
.ram_bytes
= ram_bytes
;
4505 ffe_ctl
.num_bytes
= num_bytes
;
4506 ffe_ctl
.min_alloc_size
= min_alloc_size
;
4507 ffe_ctl
.empty_size
= empty_size
;
4508 ffe_ctl
.flags
= flags
;
4509 ffe_ctl
.delalloc
= delalloc
;
4510 ffe_ctl
.hint_byte
= hint_byte
;
4511 ffe_ctl
.for_treelog
= for_treelog
;
4512 ffe_ctl
.for_data_reloc
= for_data_reloc
;
4514 ret
= find_free_extent(root
, ins
, &ffe_ctl
);
4515 if (!ret
&& !is_data
) {
4516 btrfs_dec_block_group_reservations(fs_info
, ins
->objectid
);
4517 } else if (ret
== -ENOSPC
) {
4518 if (!final_tried
&& ins
->offset
) {
4519 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
4520 num_bytes
= round_down(num_bytes
,
4521 fs_info
->sectorsize
);
4522 num_bytes
= max(num_bytes
, min_alloc_size
);
4523 ram_bytes
= num_bytes
;
4524 if (num_bytes
== min_alloc_size
)
4527 } else if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
4528 struct btrfs_space_info
*sinfo
;
4530 sinfo
= btrfs_find_space_info(fs_info
, flags
);
4532 "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d",
4533 flags
, num_bytes
, for_treelog
, for_data_reloc
);
4535 btrfs_dump_space_info(fs_info
, sinfo
,
4543 int btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
4544 u64 start
, u64 len
, int delalloc
)
4546 struct btrfs_block_group
*cache
;
4548 cache
= btrfs_lookup_block_group(fs_info
, start
);
4550 btrfs_err(fs_info
, "Unable to find block group for %llu",
4555 btrfs_add_free_space(cache
, start
, len
);
4556 btrfs_free_reserved_bytes(cache
, len
, delalloc
);
4557 trace_btrfs_reserved_extent_free(fs_info
, start
, len
);
4559 btrfs_put_block_group(cache
);
4563 int btrfs_pin_reserved_extent(struct btrfs_trans_handle
*trans
, u64 start
,
4566 struct btrfs_block_group
*cache
;
4569 cache
= btrfs_lookup_block_group(trans
->fs_info
, start
);
4571 btrfs_err(trans
->fs_info
, "unable to find block group for %llu",
4576 ret
= pin_down_extent(trans
, cache
, start
, len
, 1);
4577 btrfs_put_block_group(cache
);
4581 static int alloc_reserved_extent(struct btrfs_trans_handle
*trans
, u64 bytenr
,
4584 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4587 ret
= remove_from_free_space_tree(trans
, bytenr
, num_bytes
);
4591 ret
= btrfs_update_block_group(trans
, bytenr
, num_bytes
, true);
4594 btrfs_err(fs_info
, "update block group failed for %llu %llu",
4599 trace_btrfs_reserved_extent_alloc(fs_info
, bytenr
, num_bytes
);
4603 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
4604 u64 parent
, u64 root_objectid
,
4605 u64 flags
, u64 owner
, u64 offset
,
4606 struct btrfs_key
*ins
, int ref_mod
)
4608 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4609 struct btrfs_root
*extent_root
;
4611 struct btrfs_extent_item
*extent_item
;
4612 struct btrfs_extent_inline_ref
*iref
;
4613 struct btrfs_path
*path
;
4614 struct extent_buffer
*leaf
;
4619 type
= BTRFS_SHARED_DATA_REF_KEY
;
4621 type
= BTRFS_EXTENT_DATA_REF_KEY
;
4623 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
4625 path
= btrfs_alloc_path();
4629 extent_root
= btrfs_extent_root(fs_info
, ins
->objectid
);
4630 ret
= btrfs_insert_empty_item(trans
, extent_root
, path
, ins
, size
);
4632 btrfs_free_path(path
);
4636 leaf
= path
->nodes
[0];
4637 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4638 struct btrfs_extent_item
);
4639 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
4640 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
4641 btrfs_set_extent_flags(leaf
, extent_item
,
4642 flags
| BTRFS_EXTENT_FLAG_DATA
);
4644 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
4645 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
4647 struct btrfs_shared_data_ref
*ref
;
4648 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
4649 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
4650 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
4652 struct btrfs_extent_data_ref
*ref
;
4653 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
4654 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
4655 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
4656 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
4657 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
4660 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4661 btrfs_free_path(path
);
4663 return alloc_reserved_extent(trans
, ins
->objectid
, ins
->offset
);
4666 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
4667 struct btrfs_delayed_ref_node
*node
,
4668 struct btrfs_delayed_extent_op
*extent_op
)
4670 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4671 struct btrfs_root
*extent_root
;
4673 struct btrfs_extent_item
*extent_item
;
4674 struct btrfs_key extent_key
;
4675 struct btrfs_tree_block_info
*block_info
;
4676 struct btrfs_extent_inline_ref
*iref
;
4677 struct btrfs_path
*path
;
4678 struct extent_buffer
*leaf
;
4679 struct btrfs_delayed_tree_ref
*ref
;
4680 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
4681 u64 flags
= extent_op
->flags_to_set
;
4682 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
4684 ref
= btrfs_delayed_node_to_tree_ref(node
);
4686 extent_key
.objectid
= node
->bytenr
;
4687 if (skinny_metadata
) {
4688 extent_key
.offset
= ref
->level
;
4689 extent_key
.type
= BTRFS_METADATA_ITEM_KEY
;
4691 extent_key
.offset
= node
->num_bytes
;
4692 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4693 size
+= sizeof(*block_info
);
4696 path
= btrfs_alloc_path();
4700 extent_root
= btrfs_extent_root(fs_info
, extent_key
.objectid
);
4701 ret
= btrfs_insert_empty_item(trans
, extent_root
, path
, &extent_key
,
4704 btrfs_free_path(path
);
4708 leaf
= path
->nodes
[0];
4709 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4710 struct btrfs_extent_item
);
4711 btrfs_set_extent_refs(leaf
, extent_item
, 1);
4712 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
4713 btrfs_set_extent_flags(leaf
, extent_item
,
4714 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
4716 if (skinny_metadata
) {
4717 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
4719 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
4720 btrfs_set_tree_block_key(leaf
, block_info
, &extent_op
->key
);
4721 btrfs_set_tree_block_level(leaf
, block_info
, ref
->level
);
4722 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
4725 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
4726 btrfs_set_extent_inline_ref_type(leaf
, iref
,
4727 BTRFS_SHARED_BLOCK_REF_KEY
);
4728 btrfs_set_extent_inline_ref_offset(leaf
, iref
, ref
->parent
);
4730 btrfs_set_extent_inline_ref_type(leaf
, iref
,
4731 BTRFS_TREE_BLOCK_REF_KEY
);
4732 btrfs_set_extent_inline_ref_offset(leaf
, iref
, ref
->root
);
4735 btrfs_mark_buffer_dirty(leaf
);
4736 btrfs_free_path(path
);
4738 return alloc_reserved_extent(trans
, node
->bytenr
, fs_info
->nodesize
);
4741 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
4742 struct btrfs_root
*root
, u64 owner
,
4743 u64 offset
, u64 ram_bytes
,
4744 struct btrfs_key
*ins
)
4746 struct btrfs_ref generic_ref
= { 0 };
4748 BUG_ON(root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
);
4750 btrfs_init_generic_ref(&generic_ref
, BTRFS_ADD_DELAYED_EXTENT
,
4751 ins
->objectid
, ins
->offset
, 0);
4752 btrfs_init_data_ref(&generic_ref
, root
->root_key
.objectid
, owner
,
4754 btrfs_ref_tree_mod(root
->fs_info
, &generic_ref
);
4756 return btrfs_add_delayed_data_ref(trans
, &generic_ref
, ram_bytes
);
4760 * this is used by the tree logging recovery code. It records that
4761 * an extent has been allocated and makes sure to clear the free
4762 * space cache bits as well
4764 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
4765 u64 root_objectid
, u64 owner
, u64 offset
,
4766 struct btrfs_key
*ins
)
4768 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4770 struct btrfs_block_group
*block_group
;
4771 struct btrfs_space_info
*space_info
;
4774 * Mixed block groups will exclude before processing the log so we only
4775 * need to do the exclude dance if this fs isn't mixed.
4777 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
)) {
4778 ret
= __exclude_logged_extent(fs_info
, ins
->objectid
,
4784 block_group
= btrfs_lookup_block_group(fs_info
, ins
->objectid
);
4788 space_info
= block_group
->space_info
;
4789 spin_lock(&space_info
->lock
);
4790 spin_lock(&block_group
->lock
);
4791 space_info
->bytes_reserved
+= ins
->offset
;
4792 block_group
->reserved
+= ins
->offset
;
4793 spin_unlock(&block_group
->lock
);
4794 spin_unlock(&space_info
->lock
);
4796 ret
= alloc_reserved_file_extent(trans
, 0, root_objectid
, 0, owner
,
4799 btrfs_pin_extent(trans
, ins
->objectid
, ins
->offset
, 1);
4800 btrfs_put_block_group(block_group
);
4804 static struct extent_buffer
*
4805 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4806 u64 bytenr
, int level
, u64 owner
,
4807 enum btrfs_lock_nesting nest
)
4809 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4810 struct extent_buffer
*buf
;
4811 u64 lockdep_owner
= owner
;
4813 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
, owner
, level
);
4818 * Extra safety check in case the extent tree is corrupted and extent
4819 * allocator chooses to use a tree block which is already used and
4822 if (buf
->lock_owner
== current
->pid
) {
4823 btrfs_err_rl(fs_info
,
4824 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4825 buf
->start
, btrfs_header_owner(buf
), current
->pid
);
4826 free_extent_buffer(buf
);
4827 return ERR_PTR(-EUCLEAN
);
4831 * The reloc trees are just snapshots, so we need them to appear to be
4832 * just like any other fs tree WRT lockdep.
4834 * The exception however is in replace_path() in relocation, where we
4835 * hold the lock on the original fs root and then search for the reloc
4836 * root. At that point we need to make sure any reloc root buffers are
4837 * set to the BTRFS_TREE_RELOC_OBJECTID lockdep class in order to make
4840 if (lockdep_owner
== BTRFS_TREE_RELOC_OBJECTID
&&
4841 !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS
, &root
->state
))
4842 lockdep_owner
= BTRFS_FS_TREE_OBJECTID
;
4844 /* btrfs_clear_buffer_dirty() accesses generation field. */
4845 btrfs_set_header_generation(buf
, trans
->transid
);
4848 * This needs to stay, because we could allocate a freed block from an
4849 * old tree into a new tree, so we need to make sure this new block is
4850 * set to the appropriate level and owner.
4852 btrfs_set_buffer_lockdep_class(lockdep_owner
, buf
, level
);
4854 __btrfs_tree_lock(buf
, nest
);
4855 btrfs_clear_buffer_dirty(trans
, buf
);
4856 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
4857 clear_bit(EXTENT_BUFFER_NO_CHECK
, &buf
->bflags
);
4859 set_extent_buffer_uptodate(buf
);
4861 memzero_extent_buffer(buf
, 0, sizeof(struct btrfs_header
));
4862 btrfs_set_header_level(buf
, level
);
4863 btrfs_set_header_bytenr(buf
, buf
->start
);
4864 btrfs_set_header_generation(buf
, trans
->transid
);
4865 btrfs_set_header_backref_rev(buf
, BTRFS_MIXED_BACKREF_REV
);
4866 btrfs_set_header_owner(buf
, owner
);
4867 write_extent_buffer_fsid(buf
, fs_info
->fs_devices
->metadata_uuid
);
4868 write_extent_buffer_chunk_tree_uuid(buf
, fs_info
->chunk_tree_uuid
);
4869 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4870 buf
->log_index
= root
->log_transid
% 2;
4872 * we allow two log transactions at a time, use different
4873 * EXTENT bit to differentiate dirty pages.
4875 if (buf
->log_index
== 0)
4876 set_extent_bit(&root
->dirty_log_pages
, buf
->start
,
4877 buf
->start
+ buf
->len
- 1,
4878 EXTENT_DIRTY
, NULL
);
4880 set_extent_bit(&root
->dirty_log_pages
, buf
->start
,
4881 buf
->start
+ buf
->len
- 1,
4884 buf
->log_index
= -1;
4885 set_extent_bit(&trans
->transaction
->dirty_pages
, buf
->start
,
4886 buf
->start
+ buf
->len
- 1, EXTENT_DIRTY
, NULL
);
4888 /* this returns a buffer locked for blocking */
4893 * finds a free extent and does all the dirty work required for allocation
4894 * returns the tree buffer or an ERR_PTR on error.
4896 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
4897 struct btrfs_root
*root
,
4898 u64 parent
, u64 root_objectid
,
4899 const struct btrfs_disk_key
*key
,
4900 int level
, u64 hint
,
4902 enum btrfs_lock_nesting nest
)
4904 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4905 struct btrfs_key ins
;
4906 struct btrfs_block_rsv
*block_rsv
;
4907 struct extent_buffer
*buf
;
4908 struct btrfs_delayed_extent_op
*extent_op
;
4909 struct btrfs_ref generic_ref
= { 0 };
4912 u32 blocksize
= fs_info
->nodesize
;
4913 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
4915 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4916 if (btrfs_is_testing(fs_info
)) {
4917 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
4918 level
, root_objectid
, nest
);
4920 root
->alloc_bytenr
+= blocksize
;
4925 block_rsv
= btrfs_use_block_rsv(trans
, root
, blocksize
);
4926 if (IS_ERR(block_rsv
))
4927 return ERR_CAST(block_rsv
);
4929 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
, blocksize
,
4930 empty_size
, hint
, &ins
, 0, 0);
4934 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
,
4935 root_objectid
, nest
);
4938 goto out_free_reserved
;
4941 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
4943 parent
= ins
.objectid
;
4944 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
4948 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4949 extent_op
= btrfs_alloc_delayed_extent_op();
4955 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
4957 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
4958 extent_op
->flags_to_set
= flags
;
4959 extent_op
->update_key
= skinny_metadata
? false : true;
4960 extent_op
->update_flags
= true;
4961 extent_op
->level
= level
;
4963 btrfs_init_generic_ref(&generic_ref
, BTRFS_ADD_DELAYED_EXTENT
,
4964 ins
.objectid
, ins
.offset
, parent
);
4965 btrfs_init_tree_ref(&generic_ref
, level
, root_objectid
,
4966 root
->root_key
.objectid
, false);
4967 btrfs_ref_tree_mod(fs_info
, &generic_ref
);
4968 ret
= btrfs_add_delayed_tree_ref(trans
, &generic_ref
, extent_op
);
4970 goto out_free_delayed
;
4975 btrfs_free_delayed_extent_op(extent_op
);
4977 btrfs_tree_unlock(buf
);
4978 free_extent_buffer(buf
);
4980 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 0);
4982 btrfs_unuse_block_rsv(fs_info
, block_rsv
, blocksize
);
4983 return ERR_PTR(ret
);
4986 struct walk_control
{
4987 u64 refs
[BTRFS_MAX_LEVEL
];
4988 u64 flags
[BTRFS_MAX_LEVEL
];
4989 struct btrfs_key update_progress
;
4990 struct btrfs_key drop_progress
;
5002 #define DROP_REFERENCE 1
5003 #define UPDATE_BACKREF 2
5005 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
5006 struct btrfs_root
*root
,
5007 struct walk_control
*wc
,
5008 struct btrfs_path
*path
)
5010 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5016 struct btrfs_key key
;
5017 struct extent_buffer
*eb
;
5022 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
5023 wc
->reada_count
= wc
->reada_count
* 2 / 3;
5024 wc
->reada_count
= max(wc
->reada_count
, 2);
5026 wc
->reada_count
= wc
->reada_count
* 3 / 2;
5027 wc
->reada_count
= min_t(int, wc
->reada_count
,
5028 BTRFS_NODEPTRS_PER_BLOCK(fs_info
));
5031 eb
= path
->nodes
[wc
->level
];
5032 nritems
= btrfs_header_nritems(eb
);
5034 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
5035 if (nread
>= wc
->reada_count
)
5039 bytenr
= btrfs_node_blockptr(eb
, slot
);
5040 generation
= btrfs_node_ptr_generation(eb
, slot
);
5042 if (slot
== path
->slots
[wc
->level
])
5045 if (wc
->stage
== UPDATE_BACKREF
&&
5046 generation
<= root
->root_key
.offset
)
5049 /* We don't lock the tree block, it's OK to be racy here */
5050 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
,
5051 wc
->level
- 1, 1, &refs
,
5053 /* We don't care about errors in readahead. */
5058 if (wc
->stage
== DROP_REFERENCE
) {
5062 if (wc
->level
== 1 &&
5063 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5065 if (!wc
->update_ref
||
5066 generation
<= root
->root_key
.offset
)
5068 btrfs_node_key_to_cpu(eb
, &key
, slot
);
5069 ret
= btrfs_comp_cpu_keys(&key
,
5070 &wc
->update_progress
);
5074 if (wc
->level
== 1 &&
5075 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5079 btrfs_readahead_node_child(eb
, slot
);
5082 wc
->reada_slot
= slot
;
5086 * helper to process tree block while walking down the tree.
5088 * when wc->stage == UPDATE_BACKREF, this function updates
5089 * back refs for pointers in the block.
5091 * NOTE: return value 1 means we should stop walking down.
5093 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
5094 struct btrfs_root
*root
,
5095 struct btrfs_path
*path
,
5096 struct walk_control
*wc
, int lookup_info
)
5098 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5099 int level
= wc
->level
;
5100 struct extent_buffer
*eb
= path
->nodes
[level
];
5101 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5104 if (wc
->stage
== UPDATE_BACKREF
&&
5105 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
5109 * when reference count of tree block is 1, it won't increase
5110 * again. once full backref flag is set, we never clear it.
5113 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
5114 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
5115 BUG_ON(!path
->locks
[level
]);
5116 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
5117 eb
->start
, level
, 1,
5120 BUG_ON(ret
== -ENOMEM
);
5123 BUG_ON(wc
->refs
[level
] == 0);
5126 if (wc
->stage
== DROP_REFERENCE
) {
5127 if (wc
->refs
[level
] > 1)
5130 if (path
->locks
[level
] && !wc
->keep_locks
) {
5131 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5132 path
->locks
[level
] = 0;
5137 /* wc->stage == UPDATE_BACKREF */
5138 if (!(wc
->flags
[level
] & flag
)) {
5139 BUG_ON(!path
->locks
[level
]);
5140 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
5141 BUG_ON(ret
); /* -ENOMEM */
5142 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5143 BUG_ON(ret
); /* -ENOMEM */
5144 ret
= btrfs_set_disk_extent_flags(trans
, eb
, flag
);
5145 BUG_ON(ret
); /* -ENOMEM */
5146 wc
->flags
[level
] |= flag
;
5150 * the block is shared by multiple trees, so it's not good to
5151 * keep the tree lock
5153 if (path
->locks
[level
] && level
> 0) {
5154 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5155 path
->locks
[level
] = 0;
5161 * This is used to verify a ref exists for this root to deal with a bug where we
5162 * would have a drop_progress key that hadn't been updated properly.
5164 static int check_ref_exists(struct btrfs_trans_handle
*trans
,
5165 struct btrfs_root
*root
, u64 bytenr
, u64 parent
,
5168 struct btrfs_path
*path
;
5169 struct btrfs_extent_inline_ref
*iref
;
5172 path
= btrfs_alloc_path();
5176 ret
= lookup_extent_backref(trans
, path
, &iref
, bytenr
,
5177 root
->fs_info
->nodesize
, parent
,
5178 root
->root_key
.objectid
, level
, 0);
5179 btrfs_free_path(path
);
5188 * helper to process tree block pointer.
5190 * when wc->stage == DROP_REFERENCE, this function checks
5191 * reference count of the block pointed to. if the block
5192 * is shared and we need update back refs for the subtree
5193 * rooted at the block, this function changes wc->stage to
5194 * UPDATE_BACKREF. if the block is shared and there is no
5195 * need to update back, this function drops the reference
5198 * NOTE: return value 1 means we should stop walking down.
5200 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
5201 struct btrfs_root
*root
,
5202 struct btrfs_path
*path
,
5203 struct walk_control
*wc
, int *lookup_info
)
5205 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5209 struct btrfs_tree_parent_check check
= { 0 };
5210 struct btrfs_key key
;
5211 struct btrfs_ref ref
= { 0 };
5212 struct extent_buffer
*next
;
5213 int level
= wc
->level
;
5216 bool need_account
= false;
5218 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
5219 path
->slots
[level
]);
5221 * if the lower level block was created before the snapshot
5222 * was created, we know there is no need to update back refs
5225 if (wc
->stage
== UPDATE_BACKREF
&&
5226 generation
<= root
->root_key
.offset
) {
5231 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
5233 check
.level
= level
- 1;
5234 check
.transid
= generation
;
5235 check
.owner_root
= root
->root_key
.objectid
;
5236 check
.has_first_key
= true;
5237 btrfs_node_key_to_cpu(path
->nodes
[level
], &check
.first_key
,
5238 path
->slots
[level
]);
5240 next
= find_extent_buffer(fs_info
, bytenr
);
5242 next
= btrfs_find_create_tree_block(fs_info
, bytenr
,
5243 root
->root_key
.objectid
, level
- 1);
5245 return PTR_ERR(next
);
5248 btrfs_tree_lock(next
);
5250 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
, level
- 1, 1,
5251 &wc
->refs
[level
- 1],
5252 &wc
->flags
[level
- 1]);
5256 if (unlikely(wc
->refs
[level
- 1] == 0)) {
5257 btrfs_err(fs_info
, "Missing references.");
5263 if (wc
->stage
== DROP_REFERENCE
) {
5264 if (wc
->refs
[level
- 1] > 1) {
5265 need_account
= true;
5267 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5270 if (!wc
->update_ref
||
5271 generation
<= root
->root_key
.offset
)
5274 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
5275 path
->slots
[level
]);
5276 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
5280 wc
->stage
= UPDATE_BACKREF
;
5281 wc
->shared_level
= level
- 1;
5285 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5289 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
5290 btrfs_tree_unlock(next
);
5291 free_extent_buffer(next
);
5297 if (reada
&& level
== 1)
5298 reada_walk_down(trans
, root
, wc
, path
);
5299 next
= read_tree_block(fs_info
, bytenr
, &check
);
5301 return PTR_ERR(next
);
5302 } else if (!extent_buffer_uptodate(next
)) {
5303 free_extent_buffer(next
);
5306 btrfs_tree_lock(next
);
5310 ASSERT(level
== btrfs_header_level(next
));
5311 if (level
!= btrfs_header_level(next
)) {
5312 btrfs_err(root
->fs_info
, "mismatched level");
5316 path
->nodes
[level
] = next
;
5317 path
->slots
[level
] = 0;
5318 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
5324 wc
->refs
[level
- 1] = 0;
5325 wc
->flags
[level
- 1] = 0;
5326 if (wc
->stage
== DROP_REFERENCE
) {
5327 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
5328 parent
= path
->nodes
[level
]->start
;
5330 ASSERT(root
->root_key
.objectid
==
5331 btrfs_header_owner(path
->nodes
[level
]));
5332 if (root
->root_key
.objectid
!=
5333 btrfs_header_owner(path
->nodes
[level
])) {
5334 btrfs_err(root
->fs_info
,
5335 "mismatched block owner");
5343 * If we had a drop_progress we need to verify the refs are set
5344 * as expected. If we find our ref then we know that from here
5345 * on out everything should be correct, and we can clear the
5348 if (wc
->restarted
) {
5349 ret
= check_ref_exists(trans
, root
, bytenr
, parent
,
5360 * Reloc tree doesn't contribute to qgroup numbers, and we have
5361 * already accounted them at merge time (replace_path),
5362 * thus we could skip expensive subtree trace here.
5364 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
&&
5366 ret
= btrfs_qgroup_trace_subtree(trans
, next
,
5367 generation
, level
- 1);
5369 btrfs_err_rl(fs_info
,
5370 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5376 * We need to update the next key in our walk control so we can
5377 * update the drop_progress key accordingly. We don't care if
5378 * find_next_key doesn't find a key because that means we're at
5379 * the end and are going to clean up now.
5381 wc
->drop_level
= level
;
5382 find_next_key(path
, level
, &wc
->drop_progress
);
5384 btrfs_init_generic_ref(&ref
, BTRFS_DROP_DELAYED_REF
, bytenr
,
5385 fs_info
->nodesize
, parent
);
5386 btrfs_init_tree_ref(&ref
, level
- 1, root
->root_key
.objectid
,
5388 ret
= btrfs_free_extent(trans
, &ref
);
5397 btrfs_tree_unlock(next
);
5398 free_extent_buffer(next
);
5404 * helper to process tree block while walking up the tree.
5406 * when wc->stage == DROP_REFERENCE, this function drops
5407 * reference count on the block.
5409 * when wc->stage == UPDATE_BACKREF, this function changes
5410 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5411 * to UPDATE_BACKREF previously while processing the block.
5413 * NOTE: return value 1 means we should stop walking up.
5415 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
5416 struct btrfs_root
*root
,
5417 struct btrfs_path
*path
,
5418 struct walk_control
*wc
)
5420 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5422 int level
= wc
->level
;
5423 struct extent_buffer
*eb
= path
->nodes
[level
];
5426 if (wc
->stage
== UPDATE_BACKREF
) {
5427 BUG_ON(wc
->shared_level
< level
);
5428 if (level
< wc
->shared_level
)
5431 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
5435 wc
->stage
= DROP_REFERENCE
;
5436 wc
->shared_level
= -1;
5437 path
->slots
[level
] = 0;
5440 * check reference count again if the block isn't locked.
5441 * we should start walking down the tree again if reference
5444 if (!path
->locks
[level
]) {
5446 btrfs_tree_lock(eb
);
5447 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
5449 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
5450 eb
->start
, level
, 1,
5454 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5455 path
->locks
[level
] = 0;
5458 BUG_ON(wc
->refs
[level
] == 0);
5459 if (wc
->refs
[level
] == 1) {
5460 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5461 path
->locks
[level
] = 0;
5467 /* wc->stage == DROP_REFERENCE */
5468 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
5470 if (wc
->refs
[level
] == 1) {
5472 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5473 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
5475 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5476 BUG_ON(ret
); /* -ENOMEM */
5477 if (is_fstree(root
->root_key
.objectid
)) {
5478 ret
= btrfs_qgroup_trace_leaf_items(trans
, eb
);
5480 btrfs_err_rl(fs_info
,
5481 "error %d accounting leaf items, quota is out of sync, rescan required",
5486 /* Make block locked assertion in btrfs_clear_buffer_dirty happy. */
5487 if (!path
->locks
[level
]) {
5488 btrfs_tree_lock(eb
);
5489 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
5491 btrfs_clear_buffer_dirty(trans
, eb
);
5494 if (eb
== root
->node
) {
5495 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5497 else if (root
->root_key
.objectid
!= btrfs_header_owner(eb
))
5498 goto owner_mismatch
;
5500 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
5501 parent
= path
->nodes
[level
+ 1]->start
;
5502 else if (root
->root_key
.objectid
!=
5503 btrfs_header_owner(path
->nodes
[level
+ 1]))
5504 goto owner_mismatch
;
5507 btrfs_free_tree_block(trans
, btrfs_root_id(root
), eb
, parent
,
5508 wc
->refs
[level
] == 1);
5510 wc
->refs
[level
] = 0;
5511 wc
->flags
[level
] = 0;
5515 btrfs_err_rl(fs_info
, "unexpected tree owner, have %llu expect %llu",
5516 btrfs_header_owner(eb
), root
->root_key
.objectid
);
5520 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
5521 struct btrfs_root
*root
,
5522 struct btrfs_path
*path
,
5523 struct walk_control
*wc
)
5525 int level
= wc
->level
;
5526 int lookup_info
= 1;
5529 while (level
>= 0) {
5530 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
5537 if (path
->slots
[level
] >=
5538 btrfs_header_nritems(path
->nodes
[level
]))
5541 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
5543 path
->slots
[level
]++;
5549 return (ret
== 1) ? 0 : ret
;
5552 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
5553 struct btrfs_root
*root
,
5554 struct btrfs_path
*path
,
5555 struct walk_control
*wc
, int max_level
)
5557 int level
= wc
->level
;
5560 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
5561 while (level
< max_level
&& path
->nodes
[level
]) {
5563 if (path
->slots
[level
] + 1 <
5564 btrfs_header_nritems(path
->nodes
[level
])) {
5565 path
->slots
[level
]++;
5568 ret
= walk_up_proc(trans
, root
, path
, wc
);
5574 if (path
->locks
[level
]) {
5575 btrfs_tree_unlock_rw(path
->nodes
[level
],
5576 path
->locks
[level
]);
5577 path
->locks
[level
] = 0;
5579 free_extent_buffer(path
->nodes
[level
]);
5580 path
->nodes
[level
] = NULL
;
5588 * drop a subvolume tree.
5590 * this function traverses the tree freeing any blocks that only
5591 * referenced by the tree.
5593 * when a shared tree block is found. this function decreases its
5594 * reference count by one. if update_ref is true, this function
5595 * also make sure backrefs for the shared block and all lower level
5596 * blocks are properly updated.
5598 * If called with for_reloc == 0, may exit early with -EAGAIN
5600 int btrfs_drop_snapshot(struct btrfs_root
*root
, int update_ref
, int for_reloc
)
5602 const bool is_reloc_root
= (root
->root_key
.objectid
==
5603 BTRFS_TREE_RELOC_OBJECTID
);
5604 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5605 struct btrfs_path
*path
;
5606 struct btrfs_trans_handle
*trans
;
5607 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
5608 struct btrfs_root_item
*root_item
= &root
->root_item
;
5609 struct walk_control
*wc
;
5610 struct btrfs_key key
;
5614 bool root_dropped
= false;
5615 bool unfinished_drop
= false;
5617 btrfs_debug(fs_info
, "Drop subvolume %llu", root
->root_key
.objectid
);
5619 path
= btrfs_alloc_path();
5625 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
5627 btrfs_free_path(path
);
5633 * Use join to avoid potential EINTR from transaction start. See
5634 * wait_reserve_ticket and the whole reservation callchain.
5637 trans
= btrfs_join_transaction(tree_root
);
5639 trans
= btrfs_start_transaction(tree_root
, 0);
5640 if (IS_ERR(trans
)) {
5641 err
= PTR_ERR(trans
);
5645 err
= btrfs_run_delayed_items(trans
);
5650 * This will help us catch people modifying the fs tree while we're
5651 * dropping it. It is unsafe to mess with the fs tree while it's being
5652 * dropped as we unlock the root node and parent nodes as we walk down
5653 * the tree, assuming nothing will change. If something does change
5654 * then we'll have stale information and drop references to blocks we've
5657 set_bit(BTRFS_ROOT_DELETING
, &root
->state
);
5658 unfinished_drop
= test_bit(BTRFS_ROOT_UNFINISHED_DROP
, &root
->state
);
5660 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
5661 level
= btrfs_header_level(root
->node
);
5662 path
->nodes
[level
] = btrfs_lock_root_node(root
);
5663 path
->slots
[level
] = 0;
5664 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
5665 memset(&wc
->update_progress
, 0,
5666 sizeof(wc
->update_progress
));
5668 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
5669 memcpy(&wc
->update_progress
, &key
,
5670 sizeof(wc
->update_progress
));
5672 level
= btrfs_root_drop_level(root_item
);
5674 path
->lowest_level
= level
;
5675 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5676 path
->lowest_level
= 0;
5684 * unlock our path, this is safe because only this
5685 * function is allowed to delete this snapshot
5687 btrfs_unlock_up_safe(path
, 0);
5689 level
= btrfs_header_level(root
->node
);
5691 btrfs_tree_lock(path
->nodes
[level
]);
5692 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
5694 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
5695 path
->nodes
[level
]->start
,
5696 level
, 1, &wc
->refs
[level
],
5702 BUG_ON(wc
->refs
[level
] == 0);
5704 if (level
== btrfs_root_drop_level(root_item
))
5707 btrfs_tree_unlock(path
->nodes
[level
]);
5708 path
->locks
[level
] = 0;
5709 WARN_ON(wc
->refs
[level
] != 1);
5714 wc
->restarted
= test_bit(BTRFS_ROOT_DEAD_TREE
, &root
->state
);
5716 wc
->shared_level
= -1;
5717 wc
->stage
= DROP_REFERENCE
;
5718 wc
->update_ref
= update_ref
;
5720 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
5724 ret
= walk_down_tree(trans
, root
, path
, wc
);
5726 btrfs_abort_transaction(trans
, ret
);
5731 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
5733 btrfs_abort_transaction(trans
, ret
);
5739 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
5743 if (wc
->stage
== DROP_REFERENCE
) {
5744 wc
->drop_level
= wc
->level
;
5745 btrfs_node_key_to_cpu(path
->nodes
[wc
->drop_level
],
5747 path
->slots
[wc
->drop_level
]);
5749 btrfs_cpu_key_to_disk(&root_item
->drop_progress
,
5750 &wc
->drop_progress
);
5751 btrfs_set_root_drop_level(root_item
, wc
->drop_level
);
5753 BUG_ON(wc
->level
== 0);
5754 if (btrfs_should_end_transaction(trans
) ||
5755 (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
))) {
5756 ret
= btrfs_update_root(trans
, tree_root
,
5760 btrfs_abort_transaction(trans
, ret
);
5766 btrfs_set_last_root_drop_gen(fs_info
, trans
->transid
);
5768 btrfs_end_transaction_throttle(trans
);
5769 if (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
)) {
5770 btrfs_debug(fs_info
,
5771 "drop snapshot early exit");
5777 * Use join to avoid potential EINTR from transaction
5778 * start. See wait_reserve_ticket and the whole
5779 * reservation callchain.
5782 trans
= btrfs_join_transaction(tree_root
);
5784 trans
= btrfs_start_transaction(tree_root
, 0);
5785 if (IS_ERR(trans
)) {
5786 err
= PTR_ERR(trans
);
5791 btrfs_release_path(path
);
5795 ret
= btrfs_del_root(trans
, &root
->root_key
);
5797 btrfs_abort_transaction(trans
, ret
);
5802 if (!is_reloc_root
) {
5803 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
5806 btrfs_abort_transaction(trans
, ret
);
5809 } else if (ret
> 0) {
5810 /* if we fail to delete the orphan item this time
5811 * around, it'll get picked up the next time.
5813 * The most common failure here is just -ENOENT.
5815 btrfs_del_orphan_item(trans
, tree_root
,
5816 root
->root_key
.objectid
);
5821 * This subvolume is going to be completely dropped, and won't be
5822 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
5823 * commit transaction time. So free it here manually.
5825 btrfs_qgroup_convert_reserved_meta(root
, INT_MAX
);
5826 btrfs_qgroup_free_meta_all_pertrans(root
);
5828 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
))
5829 btrfs_add_dropped_root(trans
, root
);
5831 btrfs_put_root(root
);
5832 root_dropped
= true;
5835 btrfs_set_last_root_drop_gen(fs_info
, trans
->transid
);
5837 btrfs_end_transaction_throttle(trans
);
5840 btrfs_free_path(path
);
5843 * We were an unfinished drop root, check to see if there are any
5844 * pending, and if not clear and wake up any waiters.
5846 if (!err
&& unfinished_drop
)
5847 btrfs_maybe_wake_unfinished_drop(fs_info
);
5850 * So if we need to stop dropping the snapshot for whatever reason we
5851 * need to make sure to add it back to the dead root list so that we
5852 * keep trying to do the work later. This also cleans up roots if we
5853 * don't have it in the radix (like when we recover after a power fail
5854 * or unmount) so we don't leak memory.
5856 if (!for_reloc
&& !root_dropped
)
5857 btrfs_add_dead_root(root
);
5862 * drop subtree rooted at tree block 'node'.
5864 * NOTE: this function will unlock and release tree block 'node'
5865 * only used by relocation code
5867 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
5868 struct btrfs_root
*root
,
5869 struct extent_buffer
*node
,
5870 struct extent_buffer
*parent
)
5872 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5873 struct btrfs_path
*path
;
5874 struct walk_control
*wc
;
5880 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
5882 path
= btrfs_alloc_path();
5886 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
5888 btrfs_free_path(path
);
5892 btrfs_assert_tree_write_locked(parent
);
5893 parent_level
= btrfs_header_level(parent
);
5894 atomic_inc(&parent
->refs
);
5895 path
->nodes
[parent_level
] = parent
;
5896 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
5898 btrfs_assert_tree_write_locked(node
);
5899 level
= btrfs_header_level(node
);
5900 path
->nodes
[level
] = node
;
5901 path
->slots
[level
] = 0;
5902 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
5904 wc
->refs
[parent_level
] = 1;
5905 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5907 wc
->shared_level
= -1;
5908 wc
->stage
= DROP_REFERENCE
;
5911 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
5914 wret
= walk_down_tree(trans
, root
, path
, wc
);
5920 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
5928 btrfs_free_path(path
);
5932 int btrfs_error_unpin_extent_range(struct btrfs_fs_info
*fs_info
,
5935 return unpin_extent_range(fs_info
, start
, end
, false);
5939 * It used to be that old block groups would be left around forever.
5940 * Iterating over them would be enough to trim unused space. Since we
5941 * now automatically remove them, we also need to iterate over unallocated
5944 * We don't want a transaction for this since the discard may take a
5945 * substantial amount of time. We don't require that a transaction be
5946 * running, but we do need to take a running transaction into account
5947 * to ensure that we're not discarding chunks that were released or
5948 * allocated in the current transaction.
5950 * Holding the chunks lock will prevent other threads from allocating
5951 * or releasing chunks, but it won't prevent a running transaction
5952 * from committing and releasing the memory that the pending chunks
5953 * list head uses. For that, we need to take a reference to the
5954 * transaction and hold the commit root sem. We only need to hold
5955 * it while performing the free space search since we have already
5956 * held back allocations.
5958 static int btrfs_trim_free_extents(struct btrfs_device
*device
, u64
*trimmed
)
5960 u64 start
= BTRFS_DEVICE_RANGE_RESERVED
, len
= 0, end
= 0;
5965 /* Discard not supported = nothing to do. */
5966 if (!bdev_max_discard_sectors(device
->bdev
))
5969 /* Not writable = nothing to do. */
5970 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
))
5973 /* No free space = nothing to do. */
5974 if (device
->total_bytes
<= device
->bytes_used
)
5980 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
5983 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
5987 find_first_clear_extent_bit(&device
->alloc_state
, start
,
5989 CHUNK_TRIMMED
| CHUNK_ALLOCATED
);
5991 /* Check if there are any CHUNK_* bits left */
5992 if (start
> device
->total_bytes
) {
5993 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG
));
5994 btrfs_warn_in_rcu(fs_info
,
5995 "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
5996 start
, end
- start
+ 1,
5997 btrfs_dev_name(device
),
5998 device
->total_bytes
);
5999 mutex_unlock(&fs_info
->chunk_mutex
);
6004 /* Ensure we skip the reserved space on each device. */
6005 start
= max_t(u64
, start
, BTRFS_DEVICE_RANGE_RESERVED
);
6008 * If find_first_clear_extent_bit find a range that spans the
6009 * end of the device it will set end to -1, in this case it's up
6010 * to the caller to trim the value to the size of the device.
6012 end
= min(end
, device
->total_bytes
- 1);
6014 len
= end
- start
+ 1;
6016 /* We didn't find any extents */
6018 mutex_unlock(&fs_info
->chunk_mutex
);
6023 ret
= btrfs_issue_discard(device
->bdev
, start
, len
,
6026 set_extent_bit(&device
->alloc_state
, start
,
6027 start
+ bytes
- 1, CHUNK_TRIMMED
, NULL
);
6028 mutex_unlock(&fs_info
->chunk_mutex
);
6036 if (fatal_signal_pending(current
)) {
6048 * Trim the whole filesystem by:
6049 * 1) trimming the free space in each block group
6050 * 2) trimming the unallocated space on each device
6052 * This will also continue trimming even if a block group or device encounters
6053 * an error. The return value will be the last error, or 0 if nothing bad
6056 int btrfs_trim_fs(struct btrfs_fs_info
*fs_info
, struct fstrim_range
*range
)
6058 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6059 struct btrfs_block_group
*cache
= NULL
;
6060 struct btrfs_device
*device
;
6062 u64 range_end
= U64_MAX
;
6072 if (range
->start
== U64_MAX
)
6076 * Check range overflow if range->len is set.
6077 * The default range->len is U64_MAX.
6079 if (range
->len
!= U64_MAX
&&
6080 check_add_overflow(range
->start
, range
->len
, &range_end
))
6083 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
6084 for (; cache
; cache
= btrfs_next_block_group(cache
)) {
6085 if (cache
->start
>= range_end
) {
6086 btrfs_put_block_group(cache
);
6090 start
= max(range
->start
, cache
->start
);
6091 end
= min(range_end
, cache
->start
+ cache
->length
);
6093 if (end
- start
>= range
->minlen
) {
6094 if (!btrfs_block_group_done(cache
)) {
6095 ret
= btrfs_cache_block_group(cache
, true);
6102 ret
= btrfs_trim_block_group(cache
,
6108 trimmed
+= group_trimmed
;
6119 "failed to trim %llu block group(s), last error %d",
6122 mutex_lock(&fs_devices
->device_list_mutex
);
6123 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6124 if (test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
))
6127 ret
= btrfs_trim_free_extents(device
, &group_trimmed
);
6134 trimmed
+= group_trimmed
;
6136 mutex_unlock(&fs_devices
->device_list_mutex
);
6140 "failed to trim %llu device(s), last error %d",
6141 dev_failed
, dev_ret
);
6142 range
->len
= trimmed
;