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>
22 #include "print-tree.h"
26 #include "free-space-cache.h"
27 #include "free-space-tree.h"
30 #include "ref-verify.h"
31 #include "space-info.h"
32 #include "block-rsv.h"
33 #include "delalloc-space.h"
34 #include "block-group.h"
37 #undef SCRAMBLE_DELAYED_REFS
40 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
41 struct btrfs_delayed_ref_node *node, u64 parent,
42 u64 root_objectid, u64 owner_objectid,
43 u64 owner_offset, int refs_to_drop,
44 struct btrfs_delayed_extent_op *extra_op);
45 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
46 struct extent_buffer *leaf,
47 struct btrfs_extent_item *ei);
48 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
49 u64 parent, u64 root_objectid,
50 u64 flags, u64 owner, u64 offset,
51 struct btrfs_key *ins, int ref_mod);
52 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
53 struct btrfs_delayed_ref_node *node,
54 struct btrfs_delayed_extent_op *extent_op);
55 static int find_next_key(struct btrfs_path *path, int level,
56 struct btrfs_key *key);
58 static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
60 return (cache->flags & bits) == bits;
63 int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info,
64 u64 start, u64 num_bytes)
66 u64 end = start + num_bytes - 1;
67 set_extent_bits(&fs_info->freed_extents[0],
68 start, end, EXTENT_UPTODATE);
69 set_extent_bits(&fs_info->freed_extents[1],
70 start, end, EXTENT_UPTODATE);
74 void btrfs_free_excluded_extents(struct btrfs_block_group *cache)
76 struct btrfs_fs_info *fs_info = cache->fs_info;
80 end = start + cache->length - 1;
82 clear_extent_bits(&fs_info->freed_extents[0],
83 start, end, EXTENT_UPTODATE);
84 clear_extent_bits(&fs_info->freed_extents[1],
85 start, end, EXTENT_UPTODATE);
88 static u64 generic_ref_to_space_flags(struct btrfs_ref *ref)
90 if (ref->type == BTRFS_REF_METADATA) {
91 if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID)
92 return BTRFS_BLOCK_GROUP_SYSTEM;
94 return BTRFS_BLOCK_GROUP_METADATA;
96 return BTRFS_BLOCK_GROUP_DATA;
99 static void add_pinned_bytes(struct btrfs_fs_info *fs_info,
100 struct btrfs_ref *ref)
102 struct btrfs_space_info *space_info;
103 u64 flags = generic_ref_to_space_flags(ref);
105 space_info = btrfs_find_space_info(fs_info, flags);
107 percpu_counter_add_batch(&space_info->total_bytes_pinned, ref->len,
108 BTRFS_TOTAL_BYTES_PINNED_BATCH);
111 static void sub_pinned_bytes(struct btrfs_fs_info *fs_info,
112 struct btrfs_ref *ref)
114 struct btrfs_space_info *space_info;
115 u64 flags = generic_ref_to_space_flags(ref);
117 space_info = btrfs_find_space_info(fs_info, flags);
119 percpu_counter_add_batch(&space_info->total_bytes_pinned, -ref->len,
120 BTRFS_TOTAL_BYTES_PINNED_BATCH);
123 /* simple helper to search for an existing data extent at a given offset */
124 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
127 struct btrfs_key key;
128 struct btrfs_path *path;
130 path = btrfs_alloc_path();
134 key.objectid = start;
136 key.type = BTRFS_EXTENT_ITEM_KEY;
137 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
138 btrfs_free_path(path);
143 * helper function to lookup reference count and flags of a tree block.
145 * the head node for delayed ref is used to store the sum of all the
146 * reference count modifications queued up in the rbtree. the head
147 * node may also store the extent flags to set. This way you can check
148 * to see what the reference count and extent flags would be if all of
149 * the delayed refs are not processed.
151 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
152 struct btrfs_fs_info *fs_info, u64 bytenr,
153 u64 offset, int metadata, u64 *refs, u64 *flags)
155 struct btrfs_delayed_ref_head *head;
156 struct btrfs_delayed_ref_root *delayed_refs;
157 struct btrfs_path *path;
158 struct btrfs_extent_item *ei;
159 struct extent_buffer *leaf;
160 struct btrfs_key key;
167 * If we don't have skinny metadata, don't bother doing anything
170 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
171 offset = fs_info->nodesize;
175 path = btrfs_alloc_path();
180 path->skip_locking = 1;
181 path->search_commit_root = 1;
185 key.objectid = bytenr;
188 key.type = BTRFS_METADATA_ITEM_KEY;
190 key.type = BTRFS_EXTENT_ITEM_KEY;
192 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
196 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
197 if (path->slots[0]) {
199 btrfs_item_key_to_cpu(path->nodes[0], &key,
201 if (key.objectid == bytenr &&
202 key.type == BTRFS_EXTENT_ITEM_KEY &&
203 key.offset == fs_info->nodesize)
209 leaf = path->nodes[0];
210 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
211 if (item_size >= sizeof(*ei)) {
212 ei = btrfs_item_ptr(leaf, path->slots[0],
213 struct btrfs_extent_item);
214 num_refs = btrfs_extent_refs(leaf, ei);
215 extent_flags = btrfs_extent_flags(leaf, ei);
218 btrfs_print_v0_err(fs_info);
220 btrfs_abort_transaction(trans, ret);
222 btrfs_handle_fs_error(fs_info, ret, NULL);
227 BUG_ON(num_refs == 0);
237 delayed_refs = &trans->transaction->delayed_refs;
238 spin_lock(&delayed_refs->lock);
239 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
241 if (!mutex_trylock(&head->mutex)) {
242 refcount_inc(&head->refs);
243 spin_unlock(&delayed_refs->lock);
245 btrfs_release_path(path);
248 * Mutex was contended, block until it's released and try
251 mutex_lock(&head->mutex);
252 mutex_unlock(&head->mutex);
253 btrfs_put_delayed_ref_head(head);
256 spin_lock(&head->lock);
257 if (head->extent_op && head->extent_op->update_flags)
258 extent_flags |= head->extent_op->flags_to_set;
260 BUG_ON(num_refs == 0);
262 num_refs += head->ref_mod;
263 spin_unlock(&head->lock);
264 mutex_unlock(&head->mutex);
266 spin_unlock(&delayed_refs->lock);
268 WARN_ON(num_refs == 0);
272 *flags = extent_flags;
274 btrfs_free_path(path);
279 * Back reference rules. Back refs have three main goals:
281 * 1) differentiate between all holders of references to an extent so that
282 * when a reference is dropped we can make sure it was a valid reference
283 * before freeing the extent.
285 * 2) Provide enough information to quickly find the holders of an extent
286 * if we notice a given block is corrupted or bad.
288 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
289 * maintenance. This is actually the same as #2, but with a slightly
290 * different use case.
292 * There are two kinds of back refs. The implicit back refs is optimized
293 * for pointers in non-shared tree blocks. For a given pointer in a block,
294 * back refs of this kind provide information about the block's owner tree
295 * and the pointer's key. These information allow us to find the block by
296 * b-tree searching. The full back refs is for pointers in tree blocks not
297 * referenced by their owner trees. The location of tree block is recorded
298 * in the back refs. Actually the full back refs is generic, and can be
299 * used in all cases the implicit back refs is used. The major shortcoming
300 * of the full back refs is its overhead. Every time a tree block gets
301 * COWed, we have to update back refs entry for all pointers in it.
303 * For a newly allocated tree block, we use implicit back refs for
304 * pointers in it. This means most tree related operations only involve
305 * implicit back refs. For a tree block created in old transaction, the
306 * only way to drop a reference to it is COW it. So we can detect the
307 * event that tree block loses its owner tree's reference and do the
308 * back refs conversion.
310 * When a tree block is COWed through a tree, there are four cases:
312 * The reference count of the block is one and the tree is the block's
313 * owner tree. Nothing to do in this case.
315 * The reference count of the block is one and the tree is not the
316 * block's owner tree. In this case, full back refs is used for pointers
317 * in the block. Remove these full back refs, add implicit back refs for
318 * every pointers in the new block.
320 * The reference count of the block is greater than one and the tree is
321 * the block's owner tree. In this case, implicit back refs is used for
322 * pointers in the block. Add full back refs for every pointers in the
323 * block, increase lower level extents' reference counts. The original
324 * implicit back refs are entailed to the new block.
326 * The reference count of the block is greater than one and the tree is
327 * not the block's owner tree. Add implicit back refs for every pointer in
328 * the new block, increase lower level extents' reference count.
330 * Back Reference Key composing:
332 * The key objectid corresponds to the first byte in the extent,
333 * The key type is used to differentiate between types of back refs.
334 * There are different meanings of the key offset for different types
337 * File extents can be referenced by:
339 * - multiple snapshots, subvolumes, or different generations in one subvol
340 * - different files inside a single subvolume
341 * - different offsets inside a file (bookend extents in file.c)
343 * The extent ref structure for the implicit back refs has fields for:
345 * - Objectid of the subvolume root
346 * - objectid of the file holding the reference
347 * - original offset in the file
348 * - how many bookend extents
350 * The key offset for the implicit back refs is hash of the first
353 * The extent ref structure for the full back refs has field for:
355 * - number of pointers in the tree leaf
357 * The key offset for the implicit back refs is the first byte of
360 * When a file extent is allocated, The implicit back refs is used.
361 * the fields are filled in:
363 * (root_key.objectid, inode objectid, offset in file, 1)
365 * When a file extent is removed file truncation, we find the
366 * corresponding implicit back refs and check the following fields:
368 * (btrfs_header_owner(leaf), inode objectid, offset in file)
370 * Btree extents can be referenced by:
372 * - Different subvolumes
374 * Both the implicit back refs and the full back refs for tree blocks
375 * only consist of key. The key offset for the implicit back refs is
376 * objectid of block's owner tree. The key offset for the full back refs
377 * is the first byte of parent block.
379 * When implicit back refs is used, information about the lowest key and
380 * level of the tree block are required. These information are stored in
381 * tree block info structure.
385 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
386 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
387 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
389 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
390 struct btrfs_extent_inline_ref *iref,
391 enum btrfs_inline_ref_type is_data)
393 int type = btrfs_extent_inline_ref_type(eb, iref);
394 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
396 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
397 type == BTRFS_SHARED_BLOCK_REF_KEY ||
398 type == BTRFS_SHARED_DATA_REF_KEY ||
399 type == BTRFS_EXTENT_DATA_REF_KEY) {
400 if (is_data == BTRFS_REF_TYPE_BLOCK) {
401 if (type == BTRFS_TREE_BLOCK_REF_KEY)
403 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
406 * Every shared one has parent tree
407 * block, which must be aligned to
411 IS_ALIGNED(offset, eb->fs_info->nodesize))
414 } else if (is_data == BTRFS_REF_TYPE_DATA) {
415 if (type == BTRFS_EXTENT_DATA_REF_KEY)
417 if (type == BTRFS_SHARED_DATA_REF_KEY) {
420 * Every shared one has parent tree
421 * block, which must be aligned to
425 IS_ALIGNED(offset, eb->fs_info->nodesize))
429 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
434 btrfs_print_leaf((struct extent_buffer *)eb);
435 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
439 return BTRFS_REF_TYPE_INVALID;
442 u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
444 u32 high_crc = ~(u32)0;
445 u32 low_crc = ~(u32)0;
448 lenum = cpu_to_le64(root_objectid);
449 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
450 lenum = cpu_to_le64(owner);
451 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
452 lenum = cpu_to_le64(offset);
453 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
455 return ((u64)high_crc << 31) ^ (u64)low_crc;
458 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
459 struct btrfs_extent_data_ref *ref)
461 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
462 btrfs_extent_data_ref_objectid(leaf, ref),
463 btrfs_extent_data_ref_offset(leaf, ref));
466 static int match_extent_data_ref(struct extent_buffer *leaf,
467 struct btrfs_extent_data_ref *ref,
468 u64 root_objectid, u64 owner, u64 offset)
470 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
471 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
472 btrfs_extent_data_ref_offset(leaf, ref) != offset)
477 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
478 struct btrfs_path *path,
479 u64 bytenr, u64 parent,
481 u64 owner, u64 offset)
483 struct btrfs_root *root = trans->fs_info->extent_root;
484 struct btrfs_key key;
485 struct btrfs_extent_data_ref *ref;
486 struct extent_buffer *leaf;
492 key.objectid = bytenr;
494 key.type = BTRFS_SHARED_DATA_REF_KEY;
497 key.type = BTRFS_EXTENT_DATA_REF_KEY;
498 key.offset = hash_extent_data_ref(root_objectid,
503 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
515 leaf = path->nodes[0];
516 nritems = btrfs_header_nritems(leaf);
518 if (path->slots[0] >= nritems) {
519 ret = btrfs_next_leaf(root, path);
525 leaf = path->nodes[0];
526 nritems = btrfs_header_nritems(leaf);
530 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
531 if (key.objectid != bytenr ||
532 key.type != BTRFS_EXTENT_DATA_REF_KEY)
535 ref = btrfs_item_ptr(leaf, path->slots[0],
536 struct btrfs_extent_data_ref);
538 if (match_extent_data_ref(leaf, ref, root_objectid,
541 btrfs_release_path(path);
553 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
554 struct btrfs_path *path,
555 u64 bytenr, u64 parent,
556 u64 root_objectid, u64 owner,
557 u64 offset, int refs_to_add)
559 struct btrfs_root *root = trans->fs_info->extent_root;
560 struct btrfs_key key;
561 struct extent_buffer *leaf;
566 key.objectid = bytenr;
568 key.type = BTRFS_SHARED_DATA_REF_KEY;
570 size = sizeof(struct btrfs_shared_data_ref);
572 key.type = BTRFS_EXTENT_DATA_REF_KEY;
573 key.offset = hash_extent_data_ref(root_objectid,
575 size = sizeof(struct btrfs_extent_data_ref);
578 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
579 if (ret && ret != -EEXIST)
582 leaf = path->nodes[0];
584 struct btrfs_shared_data_ref *ref;
585 ref = btrfs_item_ptr(leaf, path->slots[0],
586 struct btrfs_shared_data_ref);
588 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
590 num_refs = btrfs_shared_data_ref_count(leaf, ref);
591 num_refs += refs_to_add;
592 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
595 struct btrfs_extent_data_ref *ref;
596 while (ret == -EEXIST) {
597 ref = btrfs_item_ptr(leaf, path->slots[0],
598 struct btrfs_extent_data_ref);
599 if (match_extent_data_ref(leaf, ref, root_objectid,
602 btrfs_release_path(path);
604 ret = btrfs_insert_empty_item(trans, root, path, &key,
606 if (ret && ret != -EEXIST)
609 leaf = path->nodes[0];
611 ref = btrfs_item_ptr(leaf, path->slots[0],
612 struct btrfs_extent_data_ref);
614 btrfs_set_extent_data_ref_root(leaf, ref,
616 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
617 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
618 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
620 num_refs = btrfs_extent_data_ref_count(leaf, ref);
621 num_refs += refs_to_add;
622 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
625 btrfs_mark_buffer_dirty(leaf);
628 btrfs_release_path(path);
632 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
633 struct btrfs_path *path,
634 int refs_to_drop, int *last_ref)
636 struct btrfs_key key;
637 struct btrfs_extent_data_ref *ref1 = NULL;
638 struct btrfs_shared_data_ref *ref2 = NULL;
639 struct extent_buffer *leaf;
643 leaf = path->nodes[0];
644 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
646 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
647 ref1 = btrfs_item_ptr(leaf, path->slots[0],
648 struct btrfs_extent_data_ref);
649 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
650 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
651 ref2 = btrfs_item_ptr(leaf, path->slots[0],
652 struct btrfs_shared_data_ref);
653 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
654 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
655 btrfs_print_v0_err(trans->fs_info);
656 btrfs_abort_transaction(trans, -EINVAL);
662 BUG_ON(num_refs < refs_to_drop);
663 num_refs -= refs_to_drop;
666 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
669 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
670 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
671 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
672 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
673 btrfs_mark_buffer_dirty(leaf);
678 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
679 struct btrfs_extent_inline_ref *iref)
681 struct btrfs_key key;
682 struct extent_buffer *leaf;
683 struct btrfs_extent_data_ref *ref1;
684 struct btrfs_shared_data_ref *ref2;
688 leaf = path->nodes[0];
689 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
691 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
694 * If type is invalid, we should have bailed out earlier than
697 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
698 ASSERT(type != BTRFS_REF_TYPE_INVALID);
699 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
700 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
701 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
703 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
704 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
706 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
707 ref1 = btrfs_item_ptr(leaf, path->slots[0],
708 struct btrfs_extent_data_ref);
709 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
710 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
711 ref2 = btrfs_item_ptr(leaf, path->slots[0],
712 struct btrfs_shared_data_ref);
713 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
720 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
721 struct btrfs_path *path,
722 u64 bytenr, u64 parent,
725 struct btrfs_root *root = trans->fs_info->extent_root;
726 struct btrfs_key key;
729 key.objectid = bytenr;
731 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
734 key.type = BTRFS_TREE_BLOCK_REF_KEY;
735 key.offset = root_objectid;
738 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
744 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
745 struct btrfs_path *path,
746 u64 bytenr, u64 parent,
749 struct btrfs_key key;
752 key.objectid = bytenr;
754 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
757 key.type = BTRFS_TREE_BLOCK_REF_KEY;
758 key.offset = root_objectid;
761 ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root,
763 btrfs_release_path(path);
767 static inline int extent_ref_type(u64 parent, u64 owner)
770 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
772 type = BTRFS_SHARED_BLOCK_REF_KEY;
774 type = BTRFS_TREE_BLOCK_REF_KEY;
777 type = BTRFS_SHARED_DATA_REF_KEY;
779 type = BTRFS_EXTENT_DATA_REF_KEY;
784 static int find_next_key(struct btrfs_path *path, int level,
785 struct btrfs_key *key)
788 for (; level < BTRFS_MAX_LEVEL; level++) {
789 if (!path->nodes[level])
791 if (path->slots[level] + 1 >=
792 btrfs_header_nritems(path->nodes[level]))
795 btrfs_item_key_to_cpu(path->nodes[level], key,
796 path->slots[level] + 1);
798 btrfs_node_key_to_cpu(path->nodes[level], key,
799 path->slots[level] + 1);
806 * look for inline back ref. if back ref is found, *ref_ret is set
807 * to the address of inline back ref, and 0 is returned.
809 * if back ref isn't found, *ref_ret is set to the address where it
810 * should be inserted, and -ENOENT is returned.
812 * if insert is true and there are too many inline back refs, the path
813 * points to the extent item, and -EAGAIN is returned.
815 * NOTE: inline back refs are ordered in the same way that back ref
816 * items in the tree are ordered.
818 static noinline_for_stack
819 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
820 struct btrfs_path *path,
821 struct btrfs_extent_inline_ref **ref_ret,
822 u64 bytenr, u64 num_bytes,
823 u64 parent, u64 root_objectid,
824 u64 owner, u64 offset, int insert)
826 struct btrfs_fs_info *fs_info = trans->fs_info;
827 struct btrfs_root *root = fs_info->extent_root;
828 struct btrfs_key key;
829 struct extent_buffer *leaf;
830 struct btrfs_extent_item *ei;
831 struct btrfs_extent_inline_ref *iref;
841 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
844 key.objectid = bytenr;
845 key.type = BTRFS_EXTENT_ITEM_KEY;
846 key.offset = num_bytes;
848 want = extent_ref_type(parent, owner);
850 extra_size = btrfs_extent_inline_ref_size(want);
851 path->keep_locks = 1;
856 * Owner is our level, so we can just add one to get the level for the
857 * block we are interested in.
859 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
860 key.type = BTRFS_METADATA_ITEM_KEY;
865 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
872 * We may be a newly converted file system which still has the old fat
873 * extent entries for metadata, so try and see if we have one of those.
875 if (ret > 0 && skinny_metadata) {
876 skinny_metadata = false;
877 if (path->slots[0]) {
879 btrfs_item_key_to_cpu(path->nodes[0], &key,
881 if (key.objectid == bytenr &&
882 key.type == BTRFS_EXTENT_ITEM_KEY &&
883 key.offset == num_bytes)
887 key.objectid = bytenr;
888 key.type = BTRFS_EXTENT_ITEM_KEY;
889 key.offset = num_bytes;
890 btrfs_release_path(path);
895 if (ret && !insert) {
898 } else if (WARN_ON(ret)) {
903 leaf = path->nodes[0];
904 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
905 if (unlikely(item_size < sizeof(*ei))) {
907 btrfs_print_v0_err(fs_info);
908 btrfs_abort_transaction(trans, err);
912 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
913 flags = btrfs_extent_flags(leaf, ei);
915 ptr = (unsigned long)(ei + 1);
916 end = (unsigned long)ei + item_size;
918 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
919 ptr += sizeof(struct btrfs_tree_block_info);
923 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
924 needed = BTRFS_REF_TYPE_DATA;
926 needed = BTRFS_REF_TYPE_BLOCK;
934 iref = (struct btrfs_extent_inline_ref *)ptr;
935 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
936 if (type == BTRFS_REF_TYPE_INVALID) {
944 ptr += btrfs_extent_inline_ref_size(type);
948 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
949 struct btrfs_extent_data_ref *dref;
950 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
951 if (match_extent_data_ref(leaf, dref, root_objectid,
956 if (hash_extent_data_ref_item(leaf, dref) <
957 hash_extent_data_ref(root_objectid, owner, offset))
961 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
963 if (parent == ref_offset) {
967 if (ref_offset < parent)
970 if (root_objectid == ref_offset) {
974 if (ref_offset < root_objectid)
978 ptr += btrfs_extent_inline_ref_size(type);
980 if (err == -ENOENT && insert) {
981 if (item_size + extra_size >=
982 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
987 * To add new inline back ref, we have to make sure
988 * there is no corresponding back ref item.
989 * For simplicity, we just do not add new inline back
990 * ref if there is any kind of item for this block
992 if (find_next_key(path, 0, &key) == 0 &&
993 key.objectid == bytenr &&
994 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
999 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1002 path->keep_locks = 0;
1003 btrfs_unlock_up_safe(path, 1);
1009 * helper to add new inline back ref
1011 static noinline_for_stack
1012 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1013 struct btrfs_path *path,
1014 struct btrfs_extent_inline_ref *iref,
1015 u64 parent, u64 root_objectid,
1016 u64 owner, u64 offset, int refs_to_add,
1017 struct btrfs_delayed_extent_op *extent_op)
1019 struct extent_buffer *leaf;
1020 struct btrfs_extent_item *ei;
1023 unsigned long item_offset;
1028 leaf = path->nodes[0];
1029 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1030 item_offset = (unsigned long)iref - (unsigned long)ei;
1032 type = extent_ref_type(parent, owner);
1033 size = btrfs_extent_inline_ref_size(type);
1035 btrfs_extend_item(path, size);
1037 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1038 refs = btrfs_extent_refs(leaf, ei);
1039 refs += refs_to_add;
1040 btrfs_set_extent_refs(leaf, ei, refs);
1042 __run_delayed_extent_op(extent_op, leaf, ei);
1044 ptr = (unsigned long)ei + item_offset;
1045 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1046 if (ptr < end - size)
1047 memmove_extent_buffer(leaf, ptr + size, ptr,
1050 iref = (struct btrfs_extent_inline_ref *)ptr;
1051 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1052 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1053 struct btrfs_extent_data_ref *dref;
1054 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1055 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1056 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1057 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1058 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1059 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1060 struct btrfs_shared_data_ref *sref;
1061 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1062 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1063 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1064 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1065 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1067 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1069 btrfs_mark_buffer_dirty(leaf);
1072 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1073 struct btrfs_path *path,
1074 struct btrfs_extent_inline_ref **ref_ret,
1075 u64 bytenr, u64 num_bytes, u64 parent,
1076 u64 root_objectid, u64 owner, u64 offset)
1080 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1081 num_bytes, parent, root_objectid,
1086 btrfs_release_path(path);
1089 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1090 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1093 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1094 root_objectid, owner, offset);
1100 * helper to update/remove inline back ref
1102 static noinline_for_stack
1103 void update_inline_extent_backref(struct btrfs_path *path,
1104 struct btrfs_extent_inline_ref *iref,
1106 struct btrfs_delayed_extent_op *extent_op,
1109 struct extent_buffer *leaf = path->nodes[0];
1110 struct btrfs_extent_item *ei;
1111 struct btrfs_extent_data_ref *dref = NULL;
1112 struct btrfs_shared_data_ref *sref = NULL;
1120 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1121 refs = btrfs_extent_refs(leaf, ei);
1122 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1123 refs += refs_to_mod;
1124 btrfs_set_extent_refs(leaf, ei, refs);
1126 __run_delayed_extent_op(extent_op, leaf, ei);
1129 * If type is invalid, we should have bailed out after
1130 * lookup_inline_extent_backref().
1132 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1133 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1135 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1136 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1137 refs = btrfs_extent_data_ref_count(leaf, dref);
1138 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1139 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1140 refs = btrfs_shared_data_ref_count(leaf, sref);
1143 BUG_ON(refs_to_mod != -1);
1146 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1147 refs += refs_to_mod;
1150 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1151 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1153 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1156 size = btrfs_extent_inline_ref_size(type);
1157 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1158 ptr = (unsigned long)iref;
1159 end = (unsigned long)ei + item_size;
1160 if (ptr + size < end)
1161 memmove_extent_buffer(leaf, ptr, ptr + size,
1164 btrfs_truncate_item(path, item_size, 1);
1166 btrfs_mark_buffer_dirty(leaf);
1169 static noinline_for_stack
1170 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1171 struct btrfs_path *path,
1172 u64 bytenr, u64 num_bytes, u64 parent,
1173 u64 root_objectid, u64 owner,
1174 u64 offset, int refs_to_add,
1175 struct btrfs_delayed_extent_op *extent_op)
1177 struct btrfs_extent_inline_ref *iref;
1180 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1181 num_bytes, parent, root_objectid,
1184 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1185 update_inline_extent_backref(path, iref, refs_to_add,
1187 } else if (ret == -ENOENT) {
1188 setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1189 root_objectid, owner, offset,
1190 refs_to_add, extent_op);
1196 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1197 struct btrfs_path *path,
1198 u64 bytenr, u64 parent, u64 root_objectid,
1199 u64 owner, u64 offset, int refs_to_add)
1202 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1203 BUG_ON(refs_to_add != 1);
1204 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1207 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1208 root_objectid, owner, offset,
1214 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1215 struct btrfs_path *path,
1216 struct btrfs_extent_inline_ref *iref,
1217 int refs_to_drop, int is_data, int *last_ref)
1221 BUG_ON(!is_data && refs_to_drop != 1);
1223 update_inline_extent_backref(path, iref, -refs_to_drop, NULL,
1225 } else if (is_data) {
1226 ret = remove_extent_data_ref(trans, path, refs_to_drop,
1230 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1235 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1236 u64 *discarded_bytes)
1239 u64 bytes_left, end;
1240 u64 aligned_start = ALIGN(start, 1 << 9);
1242 if (WARN_ON(start != aligned_start)) {
1243 len -= aligned_start - start;
1244 len = round_down(len, 1 << 9);
1245 start = aligned_start;
1248 *discarded_bytes = 0;
1256 /* Skip any superblocks on this device. */
1257 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1258 u64 sb_start = btrfs_sb_offset(j);
1259 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1260 u64 size = sb_start - start;
1262 if (!in_range(sb_start, start, bytes_left) &&
1263 !in_range(sb_end, start, bytes_left) &&
1264 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1268 * Superblock spans beginning of range. Adjust start and
1271 if (sb_start <= start) {
1272 start += sb_end - start;
1277 bytes_left = end - start;
1282 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1285 *discarded_bytes += size;
1286 else if (ret != -EOPNOTSUPP)
1295 bytes_left = end - start;
1299 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1302 *discarded_bytes += bytes_left;
1307 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1308 u64 num_bytes, u64 *actual_bytes)
1311 u64 discarded_bytes = 0;
1312 u64 end = bytenr + num_bytes;
1314 struct btrfs_bio *bbio = NULL;
1318 * Avoid races with device replace and make sure our bbio has devices
1319 * associated to its stripes that don't go away while we are discarding.
1321 btrfs_bio_counter_inc_blocked(fs_info);
1323 struct btrfs_bio_stripe *stripe;
1326 num_bytes = end - cur;
1327 /* Tell the block device(s) that the sectors can be discarded */
1328 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur,
1329 &num_bytes, &bbio, 0);
1331 * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or
1332 * -EOPNOTSUPP. For any such error, @num_bytes is not updated,
1333 * thus we can't continue anyway.
1338 stripe = bbio->stripes;
1339 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1341 struct request_queue *req_q;
1343 if (!stripe->dev->bdev) {
1344 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1347 req_q = bdev_get_queue(stripe->dev->bdev);
1348 if (!blk_queue_discard(req_q))
1351 ret = btrfs_issue_discard(stripe->dev->bdev,
1356 discarded_bytes += bytes;
1357 } else if (ret != -EOPNOTSUPP) {
1359 * Logic errors or -ENOMEM, or -EIO, but
1360 * unlikely to happen.
1362 * And since there are two loops, explicitly
1363 * go to out to avoid confusion.
1365 btrfs_put_bbio(bbio);
1370 * Just in case we get back EOPNOTSUPP for some reason,
1371 * just ignore the return value so we don't screw up
1372 * people calling discard_extent.
1376 btrfs_put_bbio(bbio);
1380 btrfs_bio_counter_dec(fs_info);
1383 *actual_bytes = discarded_bytes;
1386 if (ret == -EOPNOTSUPP)
1391 /* Can return -ENOMEM */
1392 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1393 struct btrfs_ref *generic_ref)
1395 struct btrfs_fs_info *fs_info = trans->fs_info;
1396 int old_ref_mod, new_ref_mod;
1399 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1400 generic_ref->action);
1401 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1402 generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID);
1404 if (generic_ref->type == BTRFS_REF_METADATA)
1405 ret = btrfs_add_delayed_tree_ref(trans, generic_ref,
1406 NULL, &old_ref_mod, &new_ref_mod);
1408 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0,
1409 &old_ref_mod, &new_ref_mod);
1411 btrfs_ref_tree_mod(fs_info, generic_ref);
1413 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
1414 sub_pinned_bytes(fs_info, generic_ref);
1420 * __btrfs_inc_extent_ref - insert backreference for a given extent
1422 * @trans: Handle of transaction
1424 * @node: The delayed ref node used to get the bytenr/length for
1425 * extent whose references are incremented.
1427 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1428 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1429 * bytenr of the parent block. Since new extents are always
1430 * created with indirect references, this will only be the case
1431 * when relocating a shared extent. In that case, root_objectid
1432 * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must
1435 * @root_objectid: The id of the root where this modification has originated,
1436 * this can be either one of the well-known metadata trees or
1437 * the subvolume id which references this extent.
1439 * @owner: For data extents it is the inode number of the owning file.
1440 * For metadata extents this parameter holds the level in the
1441 * tree of the extent.
1443 * @offset: For metadata extents the offset is ignored and is currently
1444 * always passed as 0. For data extents it is the fileoffset
1445 * this extent belongs to.
1447 * @refs_to_add Number of references to add
1449 * @extent_op Pointer to a structure, holding information necessary when
1450 * updating a tree block's flags
1453 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1454 struct btrfs_delayed_ref_node *node,
1455 u64 parent, u64 root_objectid,
1456 u64 owner, u64 offset, int refs_to_add,
1457 struct btrfs_delayed_extent_op *extent_op)
1459 struct btrfs_path *path;
1460 struct extent_buffer *leaf;
1461 struct btrfs_extent_item *item;
1462 struct btrfs_key key;
1463 u64 bytenr = node->bytenr;
1464 u64 num_bytes = node->num_bytes;
1468 path = btrfs_alloc_path();
1472 path->reada = READA_FORWARD;
1473 path->leave_spinning = 1;
1474 /* this will setup the path even if it fails to insert the back ref */
1475 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1476 parent, root_objectid, owner,
1477 offset, refs_to_add, extent_op);
1478 if ((ret < 0 && ret != -EAGAIN) || !ret)
1482 * Ok we had -EAGAIN which means we didn't have space to insert and
1483 * inline extent ref, so just update the reference count and add a
1486 leaf = path->nodes[0];
1487 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1488 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1489 refs = btrfs_extent_refs(leaf, item);
1490 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1492 __run_delayed_extent_op(extent_op, leaf, item);
1494 btrfs_mark_buffer_dirty(leaf);
1495 btrfs_release_path(path);
1497 path->reada = READA_FORWARD;
1498 path->leave_spinning = 1;
1499 /* now insert the actual backref */
1500 ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid,
1501 owner, offset, refs_to_add);
1503 btrfs_abort_transaction(trans, ret);
1505 btrfs_free_path(path);
1509 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1510 struct btrfs_delayed_ref_node *node,
1511 struct btrfs_delayed_extent_op *extent_op,
1512 int insert_reserved)
1515 struct btrfs_delayed_data_ref *ref;
1516 struct btrfs_key ins;
1521 ins.objectid = node->bytenr;
1522 ins.offset = node->num_bytes;
1523 ins.type = BTRFS_EXTENT_ITEM_KEY;
1525 ref = btrfs_delayed_node_to_data_ref(node);
1526 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
1528 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1529 parent = ref->parent;
1530 ref_root = ref->root;
1532 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1534 flags |= extent_op->flags_to_set;
1535 ret = alloc_reserved_file_extent(trans, parent, ref_root,
1536 flags, ref->objectid,
1539 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1540 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1541 ref->objectid, ref->offset,
1542 node->ref_mod, extent_op);
1543 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1544 ret = __btrfs_free_extent(trans, node, parent,
1545 ref_root, ref->objectid,
1546 ref->offset, node->ref_mod,
1554 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1555 struct extent_buffer *leaf,
1556 struct btrfs_extent_item *ei)
1558 u64 flags = btrfs_extent_flags(leaf, ei);
1559 if (extent_op->update_flags) {
1560 flags |= extent_op->flags_to_set;
1561 btrfs_set_extent_flags(leaf, ei, flags);
1564 if (extent_op->update_key) {
1565 struct btrfs_tree_block_info *bi;
1566 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1567 bi = (struct btrfs_tree_block_info *)(ei + 1);
1568 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1572 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1573 struct btrfs_delayed_ref_head *head,
1574 struct btrfs_delayed_extent_op *extent_op)
1576 struct btrfs_fs_info *fs_info = trans->fs_info;
1577 struct btrfs_key key;
1578 struct btrfs_path *path;
1579 struct btrfs_extent_item *ei;
1580 struct extent_buffer *leaf;
1584 int metadata = !extent_op->is_data;
1589 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1592 path = btrfs_alloc_path();
1596 key.objectid = head->bytenr;
1599 key.type = BTRFS_METADATA_ITEM_KEY;
1600 key.offset = extent_op->level;
1602 key.type = BTRFS_EXTENT_ITEM_KEY;
1603 key.offset = head->num_bytes;
1607 path->reada = READA_FORWARD;
1608 path->leave_spinning = 1;
1609 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
1616 if (path->slots[0] > 0) {
1618 btrfs_item_key_to_cpu(path->nodes[0], &key,
1620 if (key.objectid == head->bytenr &&
1621 key.type == BTRFS_EXTENT_ITEM_KEY &&
1622 key.offset == head->num_bytes)
1626 btrfs_release_path(path);
1629 key.objectid = head->bytenr;
1630 key.offset = head->num_bytes;
1631 key.type = BTRFS_EXTENT_ITEM_KEY;
1640 leaf = path->nodes[0];
1641 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1643 if (unlikely(item_size < sizeof(*ei))) {
1645 btrfs_print_v0_err(fs_info);
1646 btrfs_abort_transaction(trans, err);
1650 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1651 __run_delayed_extent_op(extent_op, leaf, ei);
1653 btrfs_mark_buffer_dirty(leaf);
1655 btrfs_free_path(path);
1659 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1660 struct btrfs_delayed_ref_node *node,
1661 struct btrfs_delayed_extent_op *extent_op,
1662 int insert_reserved)
1665 struct btrfs_delayed_tree_ref *ref;
1669 ref = btrfs_delayed_node_to_tree_ref(node);
1670 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1672 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1673 parent = ref->parent;
1674 ref_root = ref->root;
1676 if (node->ref_mod != 1) {
1677 btrfs_err(trans->fs_info,
1678 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
1679 node->bytenr, node->ref_mod, node->action, ref_root,
1683 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1684 BUG_ON(!extent_op || !extent_op->update_flags);
1685 ret = alloc_reserved_tree_block(trans, node, extent_op);
1686 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1687 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1688 ref->level, 0, 1, extent_op);
1689 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1690 ret = __btrfs_free_extent(trans, node, parent, ref_root,
1691 ref->level, 0, 1, extent_op);
1698 /* helper function to actually process a single delayed ref entry */
1699 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1700 struct btrfs_delayed_ref_node *node,
1701 struct btrfs_delayed_extent_op *extent_op,
1702 int insert_reserved)
1706 if (trans->aborted) {
1707 if (insert_reserved)
1708 btrfs_pin_extent(trans->fs_info, node->bytenr,
1709 node->num_bytes, 1);
1713 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1714 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1715 ret = run_delayed_tree_ref(trans, node, extent_op,
1717 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1718 node->type == BTRFS_SHARED_DATA_REF_KEY)
1719 ret = run_delayed_data_ref(trans, node, extent_op,
1723 if (ret && insert_reserved)
1724 btrfs_pin_extent(trans->fs_info, node->bytenr,
1725 node->num_bytes, 1);
1729 static inline struct btrfs_delayed_ref_node *
1730 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1732 struct btrfs_delayed_ref_node *ref;
1734 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1738 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1739 * This is to prevent a ref count from going down to zero, which deletes
1740 * the extent item from the extent tree, when there still are references
1741 * to add, which would fail because they would not find the extent item.
1743 if (!list_empty(&head->ref_add_list))
1744 return list_first_entry(&head->ref_add_list,
1745 struct btrfs_delayed_ref_node, add_list);
1747 ref = rb_entry(rb_first_cached(&head->ref_tree),
1748 struct btrfs_delayed_ref_node, ref_node);
1749 ASSERT(list_empty(&ref->add_list));
1753 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1754 struct btrfs_delayed_ref_head *head)
1756 spin_lock(&delayed_refs->lock);
1757 head->processing = 0;
1758 delayed_refs->num_heads_ready++;
1759 spin_unlock(&delayed_refs->lock);
1760 btrfs_delayed_ref_unlock(head);
1763 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1764 struct btrfs_delayed_ref_head *head)
1766 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1771 if (head->must_insert_reserved) {
1772 head->extent_op = NULL;
1773 btrfs_free_delayed_extent_op(extent_op);
1779 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1780 struct btrfs_delayed_ref_head *head)
1782 struct btrfs_delayed_extent_op *extent_op;
1785 extent_op = cleanup_extent_op(head);
1788 head->extent_op = NULL;
1789 spin_unlock(&head->lock);
1790 ret = run_delayed_extent_op(trans, head, extent_op);
1791 btrfs_free_delayed_extent_op(extent_op);
1792 return ret ? ret : 1;
1795 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1796 struct btrfs_delayed_ref_root *delayed_refs,
1797 struct btrfs_delayed_ref_head *head)
1799 int nr_items = 1; /* Dropping this ref head update. */
1801 if (head->total_ref_mod < 0) {
1802 struct btrfs_space_info *space_info;
1806 flags = BTRFS_BLOCK_GROUP_DATA;
1807 else if (head->is_system)
1808 flags = BTRFS_BLOCK_GROUP_SYSTEM;
1810 flags = BTRFS_BLOCK_GROUP_METADATA;
1811 space_info = btrfs_find_space_info(fs_info, flags);
1813 percpu_counter_add_batch(&space_info->total_bytes_pinned,
1815 BTRFS_TOTAL_BYTES_PINNED_BATCH);
1818 * We had csum deletions accounted for in our delayed refs rsv,
1819 * we need to drop the csum leaves for this update from our
1822 if (head->is_data) {
1823 spin_lock(&delayed_refs->lock);
1824 delayed_refs->pending_csums -= head->num_bytes;
1825 spin_unlock(&delayed_refs->lock);
1826 nr_items += btrfs_csum_bytes_to_leaves(fs_info,
1831 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
1834 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1835 struct btrfs_delayed_ref_head *head)
1838 struct btrfs_fs_info *fs_info = trans->fs_info;
1839 struct btrfs_delayed_ref_root *delayed_refs;
1842 delayed_refs = &trans->transaction->delayed_refs;
1844 ret = run_and_cleanup_extent_op(trans, head);
1846 unselect_delayed_ref_head(delayed_refs, head);
1847 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1854 * Need to drop our head ref lock and re-acquire the delayed ref lock
1855 * and then re-check to make sure nobody got added.
1857 spin_unlock(&head->lock);
1858 spin_lock(&delayed_refs->lock);
1859 spin_lock(&head->lock);
1860 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1861 spin_unlock(&head->lock);
1862 spin_unlock(&delayed_refs->lock);
1865 btrfs_delete_ref_head(delayed_refs, head);
1866 spin_unlock(&head->lock);
1867 spin_unlock(&delayed_refs->lock);
1869 if (head->must_insert_reserved) {
1870 btrfs_pin_extent(fs_info, head->bytenr,
1871 head->num_bytes, 1);
1872 if (head->is_data) {
1873 ret = btrfs_del_csums(trans, fs_info->csum_root,
1874 head->bytenr, head->num_bytes);
1878 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1880 trace_run_delayed_ref_head(fs_info, head, 0);
1881 btrfs_delayed_ref_unlock(head);
1882 btrfs_put_delayed_ref_head(head);
1886 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1887 struct btrfs_trans_handle *trans)
1889 struct btrfs_delayed_ref_root *delayed_refs =
1890 &trans->transaction->delayed_refs;
1891 struct btrfs_delayed_ref_head *head = NULL;
1894 spin_lock(&delayed_refs->lock);
1895 head = btrfs_select_ref_head(delayed_refs);
1897 spin_unlock(&delayed_refs->lock);
1902 * Grab the lock that says we are going to process all the refs for
1905 ret = btrfs_delayed_ref_lock(delayed_refs, head);
1906 spin_unlock(&delayed_refs->lock);
1909 * We may have dropped the spin lock to get the head mutex lock, and
1910 * that might have given someone else time to free the head. If that's
1911 * true, it has been removed from our list and we can move on.
1914 head = ERR_PTR(-EAGAIN);
1919 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1920 struct btrfs_delayed_ref_head *locked_ref,
1921 unsigned long *run_refs)
1923 struct btrfs_fs_info *fs_info = trans->fs_info;
1924 struct btrfs_delayed_ref_root *delayed_refs;
1925 struct btrfs_delayed_extent_op *extent_op;
1926 struct btrfs_delayed_ref_node *ref;
1927 int must_insert_reserved = 0;
1930 delayed_refs = &trans->transaction->delayed_refs;
1932 lockdep_assert_held(&locked_ref->mutex);
1933 lockdep_assert_held(&locked_ref->lock);
1935 while ((ref = select_delayed_ref(locked_ref))) {
1937 btrfs_check_delayed_seq(fs_info, ref->seq)) {
1938 spin_unlock(&locked_ref->lock);
1939 unselect_delayed_ref_head(delayed_refs, locked_ref);
1945 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1946 RB_CLEAR_NODE(&ref->ref_node);
1947 if (!list_empty(&ref->add_list))
1948 list_del(&ref->add_list);
1950 * When we play the delayed ref, also correct the ref_mod on
1953 switch (ref->action) {
1954 case BTRFS_ADD_DELAYED_REF:
1955 case BTRFS_ADD_DELAYED_EXTENT:
1956 locked_ref->ref_mod -= ref->ref_mod;
1958 case BTRFS_DROP_DELAYED_REF:
1959 locked_ref->ref_mod += ref->ref_mod;
1964 atomic_dec(&delayed_refs->num_entries);
1967 * Record the must_insert_reserved flag before we drop the
1970 must_insert_reserved = locked_ref->must_insert_reserved;
1971 locked_ref->must_insert_reserved = 0;
1973 extent_op = locked_ref->extent_op;
1974 locked_ref->extent_op = NULL;
1975 spin_unlock(&locked_ref->lock);
1977 ret = run_one_delayed_ref(trans, ref, extent_op,
1978 must_insert_reserved);
1980 btrfs_free_delayed_extent_op(extent_op);
1982 unselect_delayed_ref_head(delayed_refs, locked_ref);
1983 btrfs_put_delayed_ref(ref);
1984 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
1989 btrfs_put_delayed_ref(ref);
1992 spin_lock(&locked_ref->lock);
1993 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2000 * Returns 0 on success or if called with an already aborted transaction.
2001 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2003 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2006 struct btrfs_fs_info *fs_info = trans->fs_info;
2007 struct btrfs_delayed_ref_root *delayed_refs;
2008 struct btrfs_delayed_ref_head *locked_ref = NULL;
2009 ktime_t start = ktime_get();
2011 unsigned long count = 0;
2012 unsigned long actual_count = 0;
2014 delayed_refs = &trans->transaction->delayed_refs;
2017 locked_ref = btrfs_obtain_ref_head(trans);
2018 if (IS_ERR_OR_NULL(locked_ref)) {
2019 if (PTR_ERR(locked_ref) == -EAGAIN) {
2028 * We need to try and merge add/drops of the same ref since we
2029 * can run into issues with relocate dropping the implicit ref
2030 * and then it being added back again before the drop can
2031 * finish. If we merged anything we need to re-loop so we can
2033 * Or we can get node references of the same type that weren't
2034 * merged when created due to bumps in the tree mod seq, and
2035 * we need to merge them to prevent adding an inline extent
2036 * backref before dropping it (triggering a BUG_ON at
2037 * insert_inline_extent_backref()).
2039 spin_lock(&locked_ref->lock);
2040 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2042 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2044 if (ret < 0 && ret != -EAGAIN) {
2046 * Error, btrfs_run_delayed_refs_for_head already
2047 * unlocked everything so just bail out
2052 * Success, perform the usual cleanup of a processed
2055 ret = cleanup_ref_head(trans, locked_ref);
2057 /* We dropped our lock, we need to loop. */
2066 * Either success case or btrfs_run_delayed_refs_for_head
2067 * returned -EAGAIN, meaning we need to select another head
2072 } while ((nr != -1 && count < nr) || locked_ref);
2075 * We don't want to include ref heads since we can have empty ref heads
2076 * and those will drastically skew our runtime down since we just do
2077 * accounting, no actual extent tree updates.
2079 if (actual_count > 0) {
2080 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2084 * We weigh the current average higher than our current runtime
2085 * to avoid large swings in the average.
2087 spin_lock(&delayed_refs->lock);
2088 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2089 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2090 spin_unlock(&delayed_refs->lock);
2095 #ifdef SCRAMBLE_DELAYED_REFS
2097 * Normally delayed refs get processed in ascending bytenr order. This
2098 * correlates in most cases to the order added. To expose dependencies on this
2099 * order, we start to process the tree in the middle instead of the beginning
2101 static u64 find_middle(struct rb_root *root)
2103 struct rb_node *n = root->rb_node;
2104 struct btrfs_delayed_ref_node *entry;
2107 u64 first = 0, last = 0;
2111 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2112 first = entry->bytenr;
2116 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2117 last = entry->bytenr;
2122 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2123 WARN_ON(!entry->in_tree);
2125 middle = entry->bytenr;
2138 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2142 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2143 sizeof(struct btrfs_extent_inline_ref));
2144 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2145 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2148 * We don't ever fill up leaves all the way so multiply by 2 just to be
2149 * closer to what we're really going to want to use.
2151 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2155 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2156 * would require to store the csums for that many bytes.
2158 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2161 u64 num_csums_per_leaf;
2164 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2165 num_csums_per_leaf = div64_u64(csum_size,
2166 (u64)btrfs_super_csum_size(fs_info->super_copy));
2167 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2168 num_csums += num_csums_per_leaf - 1;
2169 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2174 * this starts processing the delayed reference count updates and
2175 * extent insertions we have queued up so far. count can be
2176 * 0, which means to process everything in the tree at the start
2177 * of the run (but not newly added entries), or it can be some target
2178 * number you'd like to process.
2180 * Returns 0 on success or if called with an aborted transaction
2181 * Returns <0 on error and aborts the transaction
2183 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2184 unsigned long count)
2186 struct btrfs_fs_info *fs_info = trans->fs_info;
2187 struct rb_node *node;
2188 struct btrfs_delayed_ref_root *delayed_refs;
2189 struct btrfs_delayed_ref_head *head;
2191 int run_all = count == (unsigned long)-1;
2193 /* We'll clean this up in btrfs_cleanup_transaction */
2197 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2200 delayed_refs = &trans->transaction->delayed_refs;
2202 count = atomic_read(&delayed_refs->num_entries) * 2;
2205 #ifdef SCRAMBLE_DELAYED_REFS
2206 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2208 ret = __btrfs_run_delayed_refs(trans, count);
2210 btrfs_abort_transaction(trans, ret);
2215 btrfs_create_pending_block_groups(trans);
2217 spin_lock(&delayed_refs->lock);
2218 node = rb_first_cached(&delayed_refs->href_root);
2220 spin_unlock(&delayed_refs->lock);
2223 head = rb_entry(node, struct btrfs_delayed_ref_head,
2225 refcount_inc(&head->refs);
2226 spin_unlock(&delayed_refs->lock);
2228 /* Mutex was contended, block until it's released and retry. */
2229 mutex_lock(&head->mutex);
2230 mutex_unlock(&head->mutex);
2232 btrfs_put_delayed_ref_head(head);
2240 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2241 u64 bytenr, u64 num_bytes, u64 flags,
2242 int level, int is_data)
2244 struct btrfs_delayed_extent_op *extent_op;
2247 extent_op = btrfs_alloc_delayed_extent_op();
2251 extent_op->flags_to_set = flags;
2252 extent_op->update_flags = true;
2253 extent_op->update_key = false;
2254 extent_op->is_data = is_data ? true : false;
2255 extent_op->level = level;
2257 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2259 btrfs_free_delayed_extent_op(extent_op);
2263 static noinline int check_delayed_ref(struct btrfs_root *root,
2264 struct btrfs_path *path,
2265 u64 objectid, u64 offset, u64 bytenr)
2267 struct btrfs_delayed_ref_head *head;
2268 struct btrfs_delayed_ref_node *ref;
2269 struct btrfs_delayed_data_ref *data_ref;
2270 struct btrfs_delayed_ref_root *delayed_refs;
2271 struct btrfs_transaction *cur_trans;
2272 struct rb_node *node;
2275 spin_lock(&root->fs_info->trans_lock);
2276 cur_trans = root->fs_info->running_transaction;
2278 refcount_inc(&cur_trans->use_count);
2279 spin_unlock(&root->fs_info->trans_lock);
2283 delayed_refs = &cur_trans->delayed_refs;
2284 spin_lock(&delayed_refs->lock);
2285 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2287 spin_unlock(&delayed_refs->lock);
2288 btrfs_put_transaction(cur_trans);
2292 if (!mutex_trylock(&head->mutex)) {
2293 refcount_inc(&head->refs);
2294 spin_unlock(&delayed_refs->lock);
2296 btrfs_release_path(path);
2299 * Mutex was contended, block until it's released and let
2302 mutex_lock(&head->mutex);
2303 mutex_unlock(&head->mutex);
2304 btrfs_put_delayed_ref_head(head);
2305 btrfs_put_transaction(cur_trans);
2308 spin_unlock(&delayed_refs->lock);
2310 spin_lock(&head->lock);
2312 * XXX: We should replace this with a proper search function in the
2315 for (node = rb_first_cached(&head->ref_tree); node;
2316 node = rb_next(node)) {
2317 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2318 /* If it's a shared ref we know a cross reference exists */
2319 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2324 data_ref = btrfs_delayed_node_to_data_ref(ref);
2327 * If our ref doesn't match the one we're currently looking at
2328 * then we have a cross reference.
2330 if (data_ref->root != root->root_key.objectid ||
2331 data_ref->objectid != objectid ||
2332 data_ref->offset != offset) {
2337 spin_unlock(&head->lock);
2338 mutex_unlock(&head->mutex);
2339 btrfs_put_transaction(cur_trans);
2343 static noinline int check_committed_ref(struct btrfs_root *root,
2344 struct btrfs_path *path,
2345 u64 objectid, u64 offset, u64 bytenr)
2347 struct btrfs_fs_info *fs_info = root->fs_info;
2348 struct btrfs_root *extent_root = fs_info->extent_root;
2349 struct extent_buffer *leaf;
2350 struct btrfs_extent_data_ref *ref;
2351 struct btrfs_extent_inline_ref *iref;
2352 struct btrfs_extent_item *ei;
2353 struct btrfs_key key;
2358 key.objectid = bytenr;
2359 key.offset = (u64)-1;
2360 key.type = BTRFS_EXTENT_ITEM_KEY;
2362 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2365 BUG_ON(ret == 0); /* Corruption */
2368 if (path->slots[0] == 0)
2372 leaf = path->nodes[0];
2373 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2375 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2379 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2380 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2382 /* If extent item has more than 1 inline ref then it's shared */
2383 if (item_size != sizeof(*ei) +
2384 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2387 /* If extent created before last snapshot => it's definitely shared */
2388 if (btrfs_extent_generation(leaf, ei) <=
2389 btrfs_root_last_snapshot(&root->root_item))
2392 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2394 /* If this extent has SHARED_DATA_REF then it's shared */
2395 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2396 if (type != BTRFS_EXTENT_DATA_REF_KEY)
2399 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2400 if (btrfs_extent_refs(leaf, ei) !=
2401 btrfs_extent_data_ref_count(leaf, ref) ||
2402 btrfs_extent_data_ref_root(leaf, ref) !=
2403 root->root_key.objectid ||
2404 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2405 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2413 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2416 struct btrfs_path *path;
2419 path = btrfs_alloc_path();
2424 ret = check_committed_ref(root, path, objectid,
2426 if (ret && ret != -ENOENT)
2429 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2430 } while (ret == -EAGAIN);
2433 btrfs_free_path(path);
2434 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2439 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2440 struct btrfs_root *root,
2441 struct extent_buffer *buf,
2442 int full_backref, int inc)
2444 struct btrfs_fs_info *fs_info = root->fs_info;
2450 struct btrfs_key key;
2451 struct btrfs_file_extent_item *fi;
2452 struct btrfs_ref generic_ref = { 0 };
2453 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2459 if (btrfs_is_testing(fs_info))
2462 ref_root = btrfs_header_owner(buf);
2463 nritems = btrfs_header_nritems(buf);
2464 level = btrfs_header_level(buf);
2466 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
2470 parent = buf->start;
2474 action = BTRFS_ADD_DELAYED_REF;
2476 action = BTRFS_DROP_DELAYED_REF;
2478 for (i = 0; i < nritems; i++) {
2480 btrfs_item_key_to_cpu(buf, &key, i);
2481 if (key.type != BTRFS_EXTENT_DATA_KEY)
2483 fi = btrfs_item_ptr(buf, i,
2484 struct btrfs_file_extent_item);
2485 if (btrfs_file_extent_type(buf, fi) ==
2486 BTRFS_FILE_EXTENT_INLINE)
2488 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2492 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2493 key.offset -= btrfs_file_extent_offset(buf, fi);
2494 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2496 generic_ref.real_root = root->root_key.objectid;
2497 btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2499 generic_ref.skip_qgroup = for_reloc;
2501 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2503 ret = btrfs_free_extent(trans, &generic_ref);
2507 bytenr = btrfs_node_blockptr(buf, i);
2508 num_bytes = fs_info->nodesize;
2509 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2511 generic_ref.real_root = root->root_key.objectid;
2512 btrfs_init_tree_ref(&generic_ref, level - 1, ref_root);
2513 generic_ref.skip_qgroup = for_reloc;
2515 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2517 ret = btrfs_free_extent(trans, &generic_ref);
2527 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2528 struct extent_buffer *buf, int full_backref)
2530 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2533 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2534 struct extent_buffer *buf, int full_backref)
2536 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2539 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
2541 struct btrfs_block_group *block_group;
2544 block_group = btrfs_lookup_block_group(fs_info, bytenr);
2545 if (!block_group || block_group->ro)
2548 btrfs_put_block_group(block_group);
2552 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2554 struct btrfs_fs_info *fs_info = root->fs_info;
2559 flags = BTRFS_BLOCK_GROUP_DATA;
2560 else if (root == fs_info->chunk_root)
2561 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2563 flags = BTRFS_BLOCK_GROUP_METADATA;
2565 ret = btrfs_get_alloc_profile(fs_info, flags);
2569 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
2571 struct btrfs_block_group *cache;
2574 spin_lock(&fs_info->block_group_cache_lock);
2575 bytenr = fs_info->first_logical_byte;
2576 spin_unlock(&fs_info->block_group_cache_lock);
2578 if (bytenr < (u64)-1)
2581 cache = btrfs_lookup_first_block_group(fs_info, search_start);
2585 bytenr = cache->start;
2586 btrfs_put_block_group(cache);
2591 static int pin_down_extent(struct btrfs_block_group *cache,
2592 u64 bytenr, u64 num_bytes, int reserved)
2594 struct btrfs_fs_info *fs_info = cache->fs_info;
2596 spin_lock(&cache->space_info->lock);
2597 spin_lock(&cache->lock);
2598 cache->pinned += num_bytes;
2599 btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2602 cache->reserved -= num_bytes;
2603 cache->space_info->bytes_reserved -= num_bytes;
2605 spin_unlock(&cache->lock);
2606 spin_unlock(&cache->space_info->lock);
2608 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
2609 num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
2610 set_extent_dirty(fs_info->pinned_extents, bytenr,
2611 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
2615 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
2616 u64 bytenr, u64 num_bytes, int reserved)
2618 struct btrfs_block_group *cache;
2620 ASSERT(fs_info->running_transaction);
2622 cache = btrfs_lookup_block_group(fs_info, bytenr);
2623 BUG_ON(!cache); /* Logic error */
2625 pin_down_extent(cache, bytenr, num_bytes, reserved);
2627 btrfs_put_block_group(cache);
2632 * this function must be called within transaction
2634 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
2635 u64 bytenr, u64 num_bytes)
2637 struct btrfs_block_group *cache;
2640 cache = btrfs_lookup_block_group(fs_info, bytenr);
2645 * pull in the free space cache (if any) so that our pin
2646 * removes the free space from the cache. We have load_only set
2647 * to one because the slow code to read in the free extents does check
2648 * the pinned extents.
2650 btrfs_cache_block_group(cache, 1);
2652 pin_down_extent(cache, bytenr, num_bytes, 0);
2654 /* remove us from the free space cache (if we're there at all) */
2655 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
2656 btrfs_put_block_group(cache);
2660 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2661 u64 start, u64 num_bytes)
2664 struct btrfs_block_group *block_group;
2665 struct btrfs_caching_control *caching_ctl;
2667 block_group = btrfs_lookup_block_group(fs_info, start);
2671 btrfs_cache_block_group(block_group, 0);
2672 caching_ctl = btrfs_get_caching_control(block_group);
2676 BUG_ON(!btrfs_block_group_done(block_group));
2677 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2679 mutex_lock(&caching_ctl->mutex);
2681 if (start >= caching_ctl->progress) {
2682 ret = btrfs_add_excluded_extent(fs_info, start,
2684 } else if (start + num_bytes <= caching_ctl->progress) {
2685 ret = btrfs_remove_free_space(block_group,
2688 num_bytes = caching_ctl->progress - start;
2689 ret = btrfs_remove_free_space(block_group,
2694 num_bytes = (start + num_bytes) -
2695 caching_ctl->progress;
2696 start = caching_ctl->progress;
2697 ret = btrfs_add_excluded_extent(fs_info, start,
2701 mutex_unlock(&caching_ctl->mutex);
2702 btrfs_put_caching_control(caching_ctl);
2704 btrfs_put_block_group(block_group);
2708 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2710 struct btrfs_fs_info *fs_info = eb->fs_info;
2711 struct btrfs_file_extent_item *item;
2712 struct btrfs_key key;
2717 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2720 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2721 btrfs_item_key_to_cpu(eb, &key, i);
2722 if (key.type != BTRFS_EXTENT_DATA_KEY)
2724 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2725 found_type = btrfs_file_extent_type(eb, item);
2726 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2728 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2730 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2731 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2732 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2741 btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2743 atomic_inc(&bg->reservations);
2746 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
2748 struct btrfs_caching_control *next;
2749 struct btrfs_caching_control *caching_ctl;
2750 struct btrfs_block_group *cache;
2752 down_write(&fs_info->commit_root_sem);
2754 list_for_each_entry_safe(caching_ctl, next,
2755 &fs_info->caching_block_groups, list) {
2756 cache = caching_ctl->block_group;
2757 if (btrfs_block_group_done(cache)) {
2758 cache->last_byte_to_unpin = (u64)-1;
2759 list_del_init(&caching_ctl->list);
2760 btrfs_put_caching_control(caching_ctl);
2762 cache->last_byte_to_unpin = caching_ctl->progress;
2766 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
2767 fs_info->pinned_extents = &fs_info->freed_extents[1];
2769 fs_info->pinned_extents = &fs_info->freed_extents[0];
2771 up_write(&fs_info->commit_root_sem);
2773 btrfs_update_global_block_rsv(fs_info);
2777 * Returns the free cluster for the given space info and sets empty_cluster to
2778 * what it should be based on the mount options.
2780 static struct btrfs_free_cluster *
2781 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2782 struct btrfs_space_info *space_info, u64 *empty_cluster)
2784 struct btrfs_free_cluster *ret = NULL;
2787 if (btrfs_mixed_space_info(space_info))
2790 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2791 ret = &fs_info->meta_alloc_cluster;
2792 if (btrfs_test_opt(fs_info, SSD))
2793 *empty_cluster = SZ_2M;
2795 *empty_cluster = SZ_64K;
2796 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2797 btrfs_test_opt(fs_info, SSD_SPREAD)) {
2798 *empty_cluster = SZ_2M;
2799 ret = &fs_info->data_alloc_cluster;
2805 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2807 const bool return_free_space)
2809 struct btrfs_block_group *cache = NULL;
2810 struct btrfs_space_info *space_info;
2811 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2812 struct btrfs_free_cluster *cluster = NULL;
2814 u64 total_unpinned = 0;
2815 u64 empty_cluster = 0;
2818 while (start <= end) {
2821 start >= cache->start + cache->length) {
2823 btrfs_put_block_group(cache);
2825 cache = btrfs_lookup_block_group(fs_info, start);
2826 BUG_ON(!cache); /* Logic error */
2828 cluster = fetch_cluster_info(fs_info,
2831 empty_cluster <<= 1;
2834 len = cache->start + cache->length - start;
2835 len = min(len, end + 1 - start);
2837 if (start < cache->last_byte_to_unpin) {
2838 len = min(len, cache->last_byte_to_unpin - start);
2839 if (return_free_space)
2840 btrfs_add_free_space(cache, start, len);
2844 total_unpinned += len;
2845 space_info = cache->space_info;
2848 * If this space cluster has been marked as fragmented and we've
2849 * unpinned enough in this block group to potentially allow a
2850 * cluster to be created inside of it go ahead and clear the
2853 if (cluster && cluster->fragmented &&
2854 total_unpinned > empty_cluster) {
2855 spin_lock(&cluster->lock);
2856 cluster->fragmented = 0;
2857 spin_unlock(&cluster->lock);
2860 spin_lock(&space_info->lock);
2861 spin_lock(&cache->lock);
2862 cache->pinned -= len;
2863 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2864 space_info->max_extent_size = 0;
2865 percpu_counter_add_batch(&space_info->total_bytes_pinned,
2866 -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
2868 space_info->bytes_readonly += len;
2871 spin_unlock(&cache->lock);
2872 if (!readonly && return_free_space &&
2873 global_rsv->space_info == space_info) {
2876 spin_lock(&global_rsv->lock);
2877 if (!global_rsv->full) {
2878 to_add = min(len, global_rsv->size -
2879 global_rsv->reserved);
2880 global_rsv->reserved += to_add;
2881 btrfs_space_info_update_bytes_may_use(fs_info,
2882 space_info, to_add);
2883 if (global_rsv->reserved >= global_rsv->size)
2884 global_rsv->full = 1;
2887 spin_unlock(&global_rsv->lock);
2888 /* Add to any tickets we may have */
2890 btrfs_try_granting_tickets(fs_info,
2893 spin_unlock(&space_info->lock);
2897 btrfs_put_block_group(cache);
2901 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2903 struct btrfs_fs_info *fs_info = trans->fs_info;
2904 struct btrfs_block_group *block_group, *tmp;
2905 struct list_head *deleted_bgs;
2906 struct extent_io_tree *unpin;
2911 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
2912 unpin = &fs_info->freed_extents[1];
2914 unpin = &fs_info->freed_extents[0];
2916 while (!trans->aborted) {
2917 struct extent_state *cached_state = NULL;
2919 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2920 ret = find_first_extent_bit(unpin, 0, &start, &end,
2921 EXTENT_DIRTY, &cached_state);
2923 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2927 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2928 ret = btrfs_discard_extent(fs_info, start,
2929 end + 1 - start, NULL);
2931 clear_extent_dirty(unpin, start, end, &cached_state);
2932 unpin_extent_range(fs_info, start, end, true);
2933 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2934 free_extent_state(cached_state);
2938 if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2939 btrfs_discard_calc_delay(&fs_info->discard_ctl);
2940 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2944 * Transaction is finished. We don't need the lock anymore. We
2945 * do need to clean up the block groups in case of a transaction
2948 deleted_bgs = &trans->transaction->deleted_bgs;
2949 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2953 if (!trans->aborted)
2954 ret = btrfs_discard_extent(fs_info,
2956 block_group->length,
2959 list_del_init(&block_group->bg_list);
2960 btrfs_put_block_group_trimming(block_group);
2961 btrfs_put_block_group(block_group);
2964 const char *errstr = btrfs_decode_error(ret);
2966 "discard failed while removing blockgroup: errno=%d %s",
2974 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2975 struct btrfs_delayed_ref_node *node, u64 parent,
2976 u64 root_objectid, u64 owner_objectid,
2977 u64 owner_offset, int refs_to_drop,
2978 struct btrfs_delayed_extent_op *extent_op)
2980 struct btrfs_fs_info *info = trans->fs_info;
2981 struct btrfs_key key;
2982 struct btrfs_path *path;
2983 struct btrfs_root *extent_root = info->extent_root;
2984 struct extent_buffer *leaf;
2985 struct btrfs_extent_item *ei;
2986 struct btrfs_extent_inline_ref *iref;
2989 int extent_slot = 0;
2990 int found_extent = 0;
2994 u64 bytenr = node->bytenr;
2995 u64 num_bytes = node->num_bytes;
2997 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
2999 path = btrfs_alloc_path();
3003 path->reada = READA_FORWARD;
3004 path->leave_spinning = 1;
3006 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3007 BUG_ON(!is_data && refs_to_drop != 1);
3010 skinny_metadata = false;
3012 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
3013 parent, root_objectid, owner_objectid,
3016 extent_slot = path->slots[0];
3017 while (extent_slot >= 0) {
3018 btrfs_item_key_to_cpu(path->nodes[0], &key,
3020 if (key.objectid != bytenr)
3022 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3023 key.offset == num_bytes) {
3027 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3028 key.offset == owner_objectid) {
3032 if (path->slots[0] - extent_slot > 5)
3037 if (!found_extent) {
3039 ret = remove_extent_backref(trans, path, NULL,
3041 is_data, &last_ref);
3043 btrfs_abort_transaction(trans, ret);
3046 btrfs_release_path(path);
3047 path->leave_spinning = 1;
3049 key.objectid = bytenr;
3050 key.type = BTRFS_EXTENT_ITEM_KEY;
3051 key.offset = num_bytes;
3053 if (!is_data && skinny_metadata) {
3054 key.type = BTRFS_METADATA_ITEM_KEY;
3055 key.offset = owner_objectid;
3058 ret = btrfs_search_slot(trans, extent_root,
3060 if (ret > 0 && skinny_metadata && path->slots[0]) {
3062 * Couldn't find our skinny metadata item,
3063 * see if we have ye olde extent item.
3066 btrfs_item_key_to_cpu(path->nodes[0], &key,
3068 if (key.objectid == bytenr &&
3069 key.type == BTRFS_EXTENT_ITEM_KEY &&
3070 key.offset == num_bytes)
3074 if (ret > 0 && skinny_metadata) {
3075 skinny_metadata = false;
3076 key.objectid = bytenr;
3077 key.type = BTRFS_EXTENT_ITEM_KEY;
3078 key.offset = num_bytes;
3079 btrfs_release_path(path);
3080 ret = btrfs_search_slot(trans, extent_root,
3086 "umm, got %d back from search, was looking for %llu",
3089 btrfs_print_leaf(path->nodes[0]);
3092 btrfs_abort_transaction(trans, ret);
3095 extent_slot = path->slots[0];
3097 } else if (WARN_ON(ret == -ENOENT)) {
3098 btrfs_print_leaf(path->nodes[0]);
3100 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
3101 bytenr, parent, root_objectid, owner_objectid,
3103 btrfs_abort_transaction(trans, ret);
3106 btrfs_abort_transaction(trans, ret);
3110 leaf = path->nodes[0];
3111 item_size = btrfs_item_size_nr(leaf, extent_slot);
3112 if (unlikely(item_size < sizeof(*ei))) {
3114 btrfs_print_v0_err(info);
3115 btrfs_abort_transaction(trans, ret);
3118 ei = btrfs_item_ptr(leaf, extent_slot,
3119 struct btrfs_extent_item);
3120 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3121 key.type == BTRFS_EXTENT_ITEM_KEY) {
3122 struct btrfs_tree_block_info *bi;
3123 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3124 bi = (struct btrfs_tree_block_info *)(ei + 1);
3125 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3128 refs = btrfs_extent_refs(leaf, ei);
3129 if (refs < refs_to_drop) {
3131 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
3132 refs_to_drop, refs, bytenr);
3134 btrfs_abort_transaction(trans, ret);
3137 refs -= refs_to_drop;
3141 __run_delayed_extent_op(extent_op, leaf, ei);
3143 * In the case of inline back ref, reference count will
3144 * be updated by remove_extent_backref
3147 BUG_ON(!found_extent);
3149 btrfs_set_extent_refs(leaf, ei, refs);
3150 btrfs_mark_buffer_dirty(leaf);
3153 ret = remove_extent_backref(trans, path, iref,
3154 refs_to_drop, is_data,
3157 btrfs_abort_transaction(trans, ret);
3163 BUG_ON(is_data && refs_to_drop !=
3164 extent_data_ref_count(path, iref));
3166 BUG_ON(path->slots[0] != extent_slot);
3168 BUG_ON(path->slots[0] != extent_slot + 1);
3169 path->slots[0] = extent_slot;
3175 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3178 btrfs_abort_transaction(trans, ret);
3181 btrfs_release_path(path);
3184 ret = btrfs_del_csums(trans, info->csum_root, bytenr,
3187 btrfs_abort_transaction(trans, ret);
3192 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
3194 btrfs_abort_transaction(trans, ret);
3198 ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0);
3200 btrfs_abort_transaction(trans, ret);
3204 btrfs_release_path(path);
3207 btrfs_free_path(path);
3212 * when we free an block, it is possible (and likely) that we free the last
3213 * delayed ref for that extent as well. This searches the delayed ref tree for
3214 * a given extent, and if there are no other delayed refs to be processed, it
3215 * removes it from the tree.
3217 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3220 struct btrfs_delayed_ref_head *head;
3221 struct btrfs_delayed_ref_root *delayed_refs;
3224 delayed_refs = &trans->transaction->delayed_refs;
3225 spin_lock(&delayed_refs->lock);
3226 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3228 goto out_delayed_unlock;
3230 spin_lock(&head->lock);
3231 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3234 if (cleanup_extent_op(head) != NULL)
3238 * waiting for the lock here would deadlock. If someone else has it
3239 * locked they are already in the process of dropping it anyway
3241 if (!mutex_trylock(&head->mutex))
3244 btrfs_delete_ref_head(delayed_refs, head);
3245 head->processing = 0;
3247 spin_unlock(&head->lock);
3248 spin_unlock(&delayed_refs->lock);
3250 BUG_ON(head->extent_op);
3251 if (head->must_insert_reserved)
3254 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3255 mutex_unlock(&head->mutex);
3256 btrfs_put_delayed_ref_head(head);
3259 spin_unlock(&head->lock);
3262 spin_unlock(&delayed_refs->lock);
3266 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3267 struct btrfs_root *root,
3268 struct extent_buffer *buf,
3269 u64 parent, int last_ref)
3271 struct btrfs_fs_info *fs_info = root->fs_info;
3272 struct btrfs_ref generic_ref = { 0 };
3276 btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3277 buf->start, buf->len, parent);
3278 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3279 root->root_key.objectid);
3281 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
3282 int old_ref_mod, new_ref_mod;
3284 btrfs_ref_tree_mod(fs_info, &generic_ref);
3285 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL,
3286 &old_ref_mod, &new_ref_mod);
3287 BUG_ON(ret); /* -ENOMEM */
3288 pin = old_ref_mod >= 0 && new_ref_mod < 0;
3291 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
3292 struct btrfs_block_group *cache;
3294 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
3295 ret = check_ref_cleanup(trans, buf->start);
3301 cache = btrfs_lookup_block_group(fs_info, buf->start);
3303 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3304 pin_down_extent(cache, buf->start, buf->len, 1);
3305 btrfs_put_block_group(cache);
3309 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3311 btrfs_add_free_space(cache, buf->start, buf->len);
3312 btrfs_free_reserved_bytes(cache, buf->len, 0);
3313 btrfs_put_block_group(cache);
3314 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3318 add_pinned_bytes(fs_info, &generic_ref);
3322 * Deleting the buffer, clear the corrupt flag since it doesn't
3325 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3329 /* Can return -ENOMEM */
3330 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3332 struct btrfs_fs_info *fs_info = trans->fs_info;
3333 int old_ref_mod, new_ref_mod;
3336 if (btrfs_is_testing(fs_info))
3340 * tree log blocks never actually go into the extent allocation
3341 * tree, just update pinning info and exit early.
3343 if ((ref->type == BTRFS_REF_METADATA &&
3344 ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
3345 (ref->type == BTRFS_REF_DATA &&
3346 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) {
3347 /* unlocks the pinned mutex */
3348 btrfs_pin_extent(fs_info, ref->bytenr, ref->len, 1);
3349 old_ref_mod = new_ref_mod = 0;
3351 } else if (ref->type == BTRFS_REF_METADATA) {
3352 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL,
3353 &old_ref_mod, &new_ref_mod);
3355 ret = btrfs_add_delayed_data_ref(trans, ref, 0,
3356 &old_ref_mod, &new_ref_mod);
3359 if (!((ref->type == BTRFS_REF_METADATA &&
3360 ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
3361 (ref->type == BTRFS_REF_DATA &&
3362 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)))
3363 btrfs_ref_tree_mod(fs_info, ref);
3365 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
3366 add_pinned_bytes(fs_info, ref);
3371 enum btrfs_loop_type {
3372 LOOP_CACHING_NOWAIT,
3379 btrfs_lock_block_group(struct btrfs_block_group *cache,
3383 down_read(&cache->data_rwsem);
3386 static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3389 btrfs_get_block_group(cache);
3391 down_read(&cache->data_rwsem);
3394 static struct btrfs_block_group *btrfs_lock_cluster(
3395 struct btrfs_block_group *block_group,
3396 struct btrfs_free_cluster *cluster,
3399 struct btrfs_block_group *used_bg = NULL;
3401 spin_lock(&cluster->refill_lock);
3403 used_bg = cluster->block_group;
3407 if (used_bg == block_group)
3410 btrfs_get_block_group(used_bg);
3415 if (down_read_trylock(&used_bg->data_rwsem))
3418 spin_unlock(&cluster->refill_lock);
3420 /* We should only have one-level nested. */
3421 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3423 spin_lock(&cluster->refill_lock);
3424 if (used_bg == cluster->block_group)
3427 up_read(&used_bg->data_rwsem);
3428 btrfs_put_block_group(used_bg);
3433 btrfs_release_block_group(struct btrfs_block_group *cache,
3437 up_read(&cache->data_rwsem);
3438 btrfs_put_block_group(cache);
3442 * Structure used internally for find_free_extent() function. Wraps needed
3445 struct find_free_extent_ctl {
3446 /* Basic allocation info */
3452 /* Where to start the search inside the bg */
3455 /* For clustered allocation */
3458 bool have_caching_bg;
3459 bool orig_have_caching_bg;
3461 /* RAID index, converted from flags */
3465 * Current loop number, check find_free_extent_update_loop() for details
3470 * Whether we're refilling a cluster, if true we need to re-search
3471 * current block group but don't try to refill the cluster again.
3473 bool retry_clustered;
3476 * Whether we're updating free space cache, if true we need to re-search
3477 * current block group but don't try updating free space cache again.
3479 bool retry_unclustered;
3481 /* If current block group is cached */
3484 /* Max contiguous hole found */
3485 u64 max_extent_size;
3487 /* Total free space from free space cache, not always contiguous */
3488 u64 total_free_space;
3496 * Helper function for find_free_extent().
3498 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3499 * Return -EAGAIN to inform caller that we need to re-search this block group
3500 * Return >0 to inform caller that we find nothing
3501 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3503 static int find_free_extent_clustered(struct btrfs_block_group *bg,
3504 struct btrfs_free_cluster *last_ptr,
3505 struct find_free_extent_ctl *ffe_ctl,
3506 struct btrfs_block_group **cluster_bg_ret)
3508 struct btrfs_block_group *cluster_bg;
3509 u64 aligned_cluster;
3513 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3515 goto refill_cluster;
3516 if (cluster_bg != bg && (cluster_bg->ro ||
3517 !block_group_bits(cluster_bg, ffe_ctl->flags)))
3518 goto release_cluster;
3520 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3521 ffe_ctl->num_bytes, cluster_bg->start,
3522 &ffe_ctl->max_extent_size);
3524 /* We have a block, we're done */
3525 spin_unlock(&last_ptr->refill_lock);
3526 trace_btrfs_reserve_extent_cluster(cluster_bg,
3527 ffe_ctl->search_start, ffe_ctl->num_bytes);
3528 *cluster_bg_ret = cluster_bg;
3529 ffe_ctl->found_offset = offset;
3532 WARN_ON(last_ptr->block_group != cluster_bg);
3536 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3537 * lets just skip it and let the allocator find whatever block it can
3538 * find. If we reach this point, we will have tried the cluster
3539 * allocator plenty of times and not have found anything, so we are
3540 * likely way too fragmented for the clustering stuff to find anything.
3542 * However, if the cluster is taken from the current block group,
3543 * release the cluster first, so that we stand a better chance of
3544 * succeeding in the unclustered allocation.
3546 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3547 spin_unlock(&last_ptr->refill_lock);
3548 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3552 /* This cluster didn't work out, free it and start over */
3553 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3555 if (cluster_bg != bg)
3556 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3559 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3560 spin_unlock(&last_ptr->refill_lock);
3564 aligned_cluster = max_t(u64,
3565 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3566 bg->full_stripe_len);
3567 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3568 ffe_ctl->num_bytes, aligned_cluster);
3570 /* Now pull our allocation out of this cluster */
3571 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3572 ffe_ctl->num_bytes, ffe_ctl->search_start,
3573 &ffe_ctl->max_extent_size);
3575 /* We found one, proceed */
3576 spin_unlock(&last_ptr->refill_lock);
3577 trace_btrfs_reserve_extent_cluster(bg,
3578 ffe_ctl->search_start,
3579 ffe_ctl->num_bytes);
3580 ffe_ctl->found_offset = offset;
3583 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
3584 !ffe_ctl->retry_clustered) {
3585 spin_unlock(&last_ptr->refill_lock);
3587 ffe_ctl->retry_clustered = true;
3588 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3589 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
3593 * At this point we either didn't find a cluster or we weren't able to
3594 * allocate a block from our cluster. Free the cluster we've been
3595 * trying to use, and go to the next block group.
3597 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3598 spin_unlock(&last_ptr->refill_lock);
3603 * Return >0 to inform caller that we find nothing
3604 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3605 * Return -EAGAIN to inform caller that we need to re-search this block group
3607 static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3608 struct btrfs_free_cluster *last_ptr,
3609 struct find_free_extent_ctl *ffe_ctl)
3614 * We are doing an unclustered allocation, set the fragmented flag so
3615 * we don't bother trying to setup a cluster again until we get more
3618 if (unlikely(last_ptr)) {
3619 spin_lock(&last_ptr->lock);
3620 last_ptr->fragmented = 1;
3621 spin_unlock(&last_ptr->lock);
3623 if (ffe_ctl->cached) {
3624 struct btrfs_free_space_ctl *free_space_ctl;
3626 free_space_ctl = bg->free_space_ctl;
3627 spin_lock(&free_space_ctl->tree_lock);
3628 if (free_space_ctl->free_space <
3629 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3630 ffe_ctl->empty_size) {
3631 ffe_ctl->total_free_space = max_t(u64,
3632 ffe_ctl->total_free_space,
3633 free_space_ctl->free_space);
3634 spin_unlock(&free_space_ctl->tree_lock);
3637 spin_unlock(&free_space_ctl->tree_lock);
3640 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3641 ffe_ctl->num_bytes, ffe_ctl->empty_size,
3642 &ffe_ctl->max_extent_size);
3645 * If we didn't find a chunk, and we haven't failed on this block group
3646 * before, and this block group is in the middle of caching and we are
3647 * ok with waiting, then go ahead and wait for progress to be made, and
3648 * set @retry_unclustered to true.
3650 * If @retry_unclustered is true then we've already waited on this
3651 * block group once and should move on to the next block group.
3653 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
3654 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
3655 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3656 ffe_ctl->empty_size);
3657 ffe_ctl->retry_unclustered = true;
3659 } else if (!offset) {
3662 ffe_ctl->found_offset = offset;
3667 * Return >0 means caller needs to re-search for free extent
3668 * Return 0 means we have the needed free extent.
3669 * Return <0 means we failed to locate any free extent.
3671 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
3672 struct btrfs_free_cluster *last_ptr,
3673 struct btrfs_key *ins,
3674 struct find_free_extent_ctl *ffe_ctl,
3675 int full_search, bool use_cluster)
3677 struct btrfs_root *root = fs_info->extent_root;
3680 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
3681 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
3682 ffe_ctl->orig_have_caching_bg = true;
3684 if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
3685 ffe_ctl->have_caching_bg)
3688 if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
3691 if (ins->objectid) {
3692 if (!use_cluster && last_ptr) {
3693 spin_lock(&last_ptr->lock);
3694 last_ptr->window_start = ins->objectid;
3695 spin_unlock(&last_ptr->lock);
3701 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
3702 * caching kthreads as we move along
3703 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
3704 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
3705 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
3708 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
3710 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
3712 * We want to skip the LOOP_CACHING_WAIT step if we
3713 * don't have any uncached bgs and we've already done a
3714 * full search through.
3716 if (ffe_ctl->orig_have_caching_bg || !full_search)
3717 ffe_ctl->loop = LOOP_CACHING_WAIT;
3719 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
3724 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
3725 struct btrfs_trans_handle *trans;
3728 trans = current->journal_info;
3732 trans = btrfs_join_transaction(root);
3734 if (IS_ERR(trans)) {
3735 ret = PTR_ERR(trans);
3739 ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
3743 * If we can't allocate a new chunk we've already looped
3744 * through at least once, move on to the NO_EMPTY_SIZE
3748 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
3750 /* Do not bail out on ENOSPC since we can do more. */
3751 if (ret < 0 && ret != -ENOSPC)
3752 btrfs_abort_transaction(trans, ret);
3756 btrfs_end_transaction(trans);
3761 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
3763 * Don't loop again if we already have no empty_size and
3766 if (ffe_ctl->empty_size == 0 &&
3767 ffe_ctl->empty_cluster == 0)
3769 ffe_ctl->empty_size = 0;
3770 ffe_ctl->empty_cluster = 0;
3778 * walks the btree of allocated extents and find a hole of a given size.
3779 * The key ins is changed to record the hole:
3780 * ins->objectid == start position
3781 * ins->flags = BTRFS_EXTENT_ITEM_KEY
3782 * ins->offset == the size of the hole.
3783 * Any available blocks before search_start are skipped.
3785 * If there is no suitable free space, we will record the max size of
3786 * the free space extent currently.
3788 * The overall logic and call chain:
3790 * find_free_extent()
3791 * |- Iterate through all block groups
3792 * | |- Get a valid block group
3793 * | |- Try to do clustered allocation in that block group
3794 * | |- Try to do unclustered allocation in that block group
3795 * | |- Check if the result is valid
3796 * | | |- If valid, then exit
3797 * | |- Jump to next block group
3799 * |- Push harder to find free extents
3800 * |- If not found, re-iterate all block groups
3802 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
3803 u64 ram_bytes, u64 num_bytes, u64 empty_size,
3804 u64 hint_byte, struct btrfs_key *ins,
3805 u64 flags, int delalloc)
3808 int cache_block_group_error = 0;
3809 struct btrfs_free_cluster *last_ptr = NULL;
3810 struct btrfs_block_group *block_group = NULL;
3811 struct find_free_extent_ctl ffe_ctl = {0};
3812 struct btrfs_space_info *space_info;
3813 bool use_cluster = true;
3814 bool full_search = false;
3816 WARN_ON(num_bytes < fs_info->sectorsize);
3818 ffe_ctl.num_bytes = num_bytes;
3819 ffe_ctl.empty_size = empty_size;
3820 ffe_ctl.flags = flags;
3821 ffe_ctl.search_start = 0;
3822 ffe_ctl.retry_clustered = false;
3823 ffe_ctl.retry_unclustered = false;
3824 ffe_ctl.delalloc = delalloc;
3825 ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
3826 ffe_ctl.have_caching_bg = false;
3827 ffe_ctl.orig_have_caching_bg = false;
3828 ffe_ctl.found_offset = 0;
3830 ins->type = BTRFS_EXTENT_ITEM_KEY;
3834 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
3836 space_info = btrfs_find_space_info(fs_info, flags);
3838 btrfs_err(fs_info, "No space info for %llu", flags);
3843 * If our free space is heavily fragmented we may not be able to make
3844 * big contiguous allocations, so instead of doing the expensive search
3845 * for free space, simply return ENOSPC with our max_extent_size so we
3846 * can go ahead and search for a more manageable chunk.
3848 * If our max_extent_size is large enough for our allocation simply
3849 * disable clustering since we will likely not be able to find enough
3850 * space to create a cluster and induce latency trying.
3852 if (unlikely(space_info->max_extent_size)) {
3853 spin_lock(&space_info->lock);
3854 if (space_info->max_extent_size &&
3855 num_bytes > space_info->max_extent_size) {
3856 ins->offset = space_info->max_extent_size;
3857 spin_unlock(&space_info->lock);
3859 } else if (space_info->max_extent_size) {
3860 use_cluster = false;
3862 spin_unlock(&space_info->lock);
3865 last_ptr = fetch_cluster_info(fs_info, space_info,
3866 &ffe_ctl.empty_cluster);
3868 spin_lock(&last_ptr->lock);
3869 if (last_ptr->block_group)
3870 hint_byte = last_ptr->window_start;
3871 if (last_ptr->fragmented) {
3873 * We still set window_start so we can keep track of the
3874 * last place we found an allocation to try and save
3877 hint_byte = last_ptr->window_start;
3878 use_cluster = false;
3880 spin_unlock(&last_ptr->lock);
3883 ffe_ctl.search_start = max(ffe_ctl.search_start,
3884 first_logical_byte(fs_info, 0));
3885 ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte);
3886 if (ffe_ctl.search_start == hint_byte) {
3887 block_group = btrfs_lookup_block_group(fs_info,
3888 ffe_ctl.search_start);
3890 * we don't want to use the block group if it doesn't match our
3891 * allocation bits, or if its not cached.
3893 * However if we are re-searching with an ideal block group
3894 * picked out then we don't care that the block group is cached.
3896 if (block_group && block_group_bits(block_group, flags) &&
3897 block_group->cached != BTRFS_CACHE_NO) {
3898 down_read(&space_info->groups_sem);
3899 if (list_empty(&block_group->list) ||
3902 * someone is removing this block group,
3903 * we can't jump into the have_block_group
3904 * target because our list pointers are not
3907 btrfs_put_block_group(block_group);
3908 up_read(&space_info->groups_sem);
3910 ffe_ctl.index = btrfs_bg_flags_to_raid_index(
3911 block_group->flags);
3912 btrfs_lock_block_group(block_group, delalloc);
3913 goto have_block_group;
3915 } else if (block_group) {
3916 btrfs_put_block_group(block_group);
3920 ffe_ctl.have_caching_bg = false;
3921 if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
3924 down_read(&space_info->groups_sem);
3925 list_for_each_entry(block_group,
3926 &space_info->block_groups[ffe_ctl.index], list) {
3927 /* If the block group is read-only, we can skip it entirely. */
3928 if (unlikely(block_group->ro))
3931 btrfs_grab_block_group(block_group, delalloc);
3932 ffe_ctl.search_start = block_group->start;
3935 * this can happen if we end up cycling through all the
3936 * raid types, but we want to make sure we only allocate
3937 * for the proper type.
3939 if (!block_group_bits(block_group, flags)) {
3940 u64 extra = BTRFS_BLOCK_GROUP_DUP |
3941 BTRFS_BLOCK_GROUP_RAID1_MASK |
3942 BTRFS_BLOCK_GROUP_RAID56_MASK |
3943 BTRFS_BLOCK_GROUP_RAID10;
3946 * if they asked for extra copies and this block group
3947 * doesn't provide them, bail. This does allow us to
3948 * fill raid0 from raid1.
3950 if ((flags & extra) && !(block_group->flags & extra))
3954 * This block group has different flags than we want.
3955 * It's possible that we have MIXED_GROUP flag but no
3956 * block group is mixed. Just skip such block group.
3958 btrfs_release_block_group(block_group, delalloc);
3963 ffe_ctl.cached = btrfs_block_group_done(block_group);
3964 if (unlikely(!ffe_ctl.cached)) {
3965 ffe_ctl.have_caching_bg = true;
3966 ret = btrfs_cache_block_group(block_group, 0);
3969 * If we get ENOMEM here or something else we want to
3970 * try other block groups, because it may not be fatal.
3971 * However if we can't find anything else we need to
3972 * save our return here so that we return the actual
3973 * error that caused problems, not ENOSPC.
3976 if (!cache_block_group_error)
3977 cache_block_group_error = ret;
3984 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
3988 * Ok we want to try and use the cluster allocator, so
3991 if (last_ptr && use_cluster) {
3992 struct btrfs_block_group *cluster_bg = NULL;
3994 ret = find_free_extent_clustered(block_group, last_ptr,
3995 &ffe_ctl, &cluster_bg);
3998 if (cluster_bg && cluster_bg != block_group) {
3999 btrfs_release_block_group(block_group,
4001 block_group = cluster_bg;
4004 } else if (ret == -EAGAIN) {
4005 goto have_block_group;
4006 } else if (ret > 0) {
4009 /* ret == -ENOENT case falls through */
4012 ret = find_free_extent_unclustered(block_group, last_ptr,
4015 goto have_block_group;
4018 /* ret == 0 case falls through */
4020 ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
4021 fs_info->stripesize);
4023 /* move on to the next group */
4024 if (ffe_ctl.search_start + num_bytes >
4025 block_group->start + block_group->length) {
4026 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
4031 if (ffe_ctl.found_offset < ffe_ctl.search_start)
4032 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
4033 ffe_ctl.search_start - ffe_ctl.found_offset);
4035 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
4036 num_bytes, delalloc);
4037 if (ret == -EAGAIN) {
4038 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
4042 btrfs_inc_block_group_reservations(block_group);
4044 /* we are all good, lets return */
4045 ins->objectid = ffe_ctl.search_start;
4046 ins->offset = num_bytes;
4048 trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
4050 btrfs_release_block_group(block_group, delalloc);
4053 ffe_ctl.retry_clustered = false;
4054 ffe_ctl.retry_unclustered = false;
4055 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
4057 btrfs_release_block_group(block_group, delalloc);
4060 up_read(&space_info->groups_sem);
4062 ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl,
4063 full_search, use_cluster);
4067 if (ret == -ENOSPC && !cache_block_group_error) {
4069 * Use ffe_ctl->total_free_space as fallback if we can't find
4070 * any contiguous hole.
4072 if (!ffe_ctl.max_extent_size)
4073 ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
4074 spin_lock(&space_info->lock);
4075 space_info->max_extent_size = ffe_ctl.max_extent_size;
4076 spin_unlock(&space_info->lock);
4077 ins->offset = ffe_ctl.max_extent_size;
4078 } else if (ret == -ENOSPC) {
4079 ret = cache_block_group_error;
4085 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4086 * hole that is at least as big as @num_bytes.
4088 * @root - The root that will contain this extent
4090 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4091 * is used for accounting purposes. This value differs
4092 * from @num_bytes only in the case of compressed extents.
4094 * @num_bytes - Number of bytes to allocate on-disk.
4096 * @min_alloc_size - Indicates the minimum amount of space that the
4097 * allocator should try to satisfy. In some cases
4098 * @num_bytes may be larger than what is required and if
4099 * the filesystem is fragmented then allocation fails.
4100 * However, the presence of @min_alloc_size gives a
4101 * chance to try and satisfy the smaller allocation.
4103 * @empty_size - A hint that you plan on doing more COW. This is the
4104 * size in bytes the allocator should try to find free
4105 * next to the block it returns. This is just a hint and
4106 * may be ignored by the allocator.
4108 * @hint_byte - Hint to the allocator to start searching above the byte
4109 * address passed. It might be ignored.
4111 * @ins - This key is modified to record the found hole. It will
4112 * have the following values:
4113 * ins->objectid == start position
4114 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4115 * ins->offset == the size of the hole.
4117 * @is_data - Boolean flag indicating whether an extent is
4118 * allocated for data (true) or metadata (false)
4120 * @delalloc - Boolean flag indicating whether this allocation is for
4121 * delalloc or not. If 'true' data_rwsem of block groups
4122 * is going to be acquired.
4125 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4126 * case -ENOSPC is returned then @ins->offset will contain the size of the
4127 * largest available hole the allocator managed to find.
4129 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4130 u64 num_bytes, u64 min_alloc_size,
4131 u64 empty_size, u64 hint_byte,
4132 struct btrfs_key *ins, int is_data, int delalloc)
4134 struct btrfs_fs_info *fs_info = root->fs_info;
4135 bool final_tried = num_bytes == min_alloc_size;
4139 flags = get_alloc_profile_by_root(root, is_data);
4141 WARN_ON(num_bytes < fs_info->sectorsize);
4142 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
4143 hint_byte, ins, flags, delalloc);
4144 if (!ret && !is_data) {
4145 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4146 } else if (ret == -ENOSPC) {
4147 if (!final_tried && ins->offset) {
4148 num_bytes = min(num_bytes >> 1, ins->offset);
4149 num_bytes = round_down(num_bytes,
4150 fs_info->sectorsize);
4151 num_bytes = max(num_bytes, min_alloc_size);
4152 ram_bytes = num_bytes;
4153 if (num_bytes == min_alloc_size)
4156 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4157 struct btrfs_space_info *sinfo;
4159 sinfo = btrfs_find_space_info(fs_info, flags);
4161 "allocation failed flags %llu, wanted %llu",
4164 btrfs_dump_space_info(fs_info, sinfo,
4172 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4173 u64 start, u64 len, int delalloc)
4175 struct btrfs_block_group *cache;
4177 cache = btrfs_lookup_block_group(fs_info, start);
4179 btrfs_err(fs_info, "Unable to find block group for %llu",
4184 btrfs_add_free_space(cache, start, len);
4185 btrfs_free_reserved_bytes(cache, len, delalloc);
4186 trace_btrfs_reserved_extent_free(fs_info, start, len);
4188 btrfs_put_block_group(cache);
4192 int btrfs_pin_reserved_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
4194 struct btrfs_block_group *cache;
4197 cache = btrfs_lookup_block_group(fs_info, start);
4199 btrfs_err(fs_info, "unable to find block group for %llu", start);
4203 ret = pin_down_extent(cache, start, len, 1);
4204 btrfs_put_block_group(cache);
4208 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4209 u64 parent, u64 root_objectid,
4210 u64 flags, u64 owner, u64 offset,
4211 struct btrfs_key *ins, int ref_mod)
4213 struct btrfs_fs_info *fs_info = trans->fs_info;
4215 struct btrfs_extent_item *extent_item;
4216 struct btrfs_extent_inline_ref *iref;
4217 struct btrfs_path *path;
4218 struct extent_buffer *leaf;
4223 type = BTRFS_SHARED_DATA_REF_KEY;
4225 type = BTRFS_EXTENT_DATA_REF_KEY;
4227 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4229 path = btrfs_alloc_path();
4233 path->leave_spinning = 1;
4234 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4237 btrfs_free_path(path);
4241 leaf = path->nodes[0];
4242 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4243 struct btrfs_extent_item);
4244 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4245 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4246 btrfs_set_extent_flags(leaf, extent_item,
4247 flags | BTRFS_EXTENT_FLAG_DATA);
4249 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4250 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4252 struct btrfs_shared_data_ref *ref;
4253 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4254 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4255 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4257 struct btrfs_extent_data_ref *ref;
4258 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4259 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4260 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4261 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4262 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4265 btrfs_mark_buffer_dirty(path->nodes[0]);
4266 btrfs_free_path(path);
4268 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
4272 ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1);
4273 if (ret) { /* -ENOENT, logic error */
4274 btrfs_err(fs_info, "update block group failed for %llu %llu",
4275 ins->objectid, ins->offset);
4278 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
4282 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4283 struct btrfs_delayed_ref_node *node,
4284 struct btrfs_delayed_extent_op *extent_op)
4286 struct btrfs_fs_info *fs_info = trans->fs_info;
4288 struct btrfs_extent_item *extent_item;
4289 struct btrfs_key extent_key;
4290 struct btrfs_tree_block_info *block_info;
4291 struct btrfs_extent_inline_ref *iref;
4292 struct btrfs_path *path;
4293 struct extent_buffer *leaf;
4294 struct btrfs_delayed_tree_ref *ref;
4295 u32 size = sizeof(*extent_item) + sizeof(*iref);
4297 u64 flags = extent_op->flags_to_set;
4298 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4300 ref = btrfs_delayed_node_to_tree_ref(node);
4302 extent_key.objectid = node->bytenr;
4303 if (skinny_metadata) {
4304 extent_key.offset = ref->level;
4305 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4306 num_bytes = fs_info->nodesize;
4308 extent_key.offset = node->num_bytes;
4309 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4310 size += sizeof(*block_info);
4311 num_bytes = node->num_bytes;
4314 path = btrfs_alloc_path();
4318 path->leave_spinning = 1;
4319 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4322 btrfs_free_path(path);
4326 leaf = path->nodes[0];
4327 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4328 struct btrfs_extent_item);
4329 btrfs_set_extent_refs(leaf, extent_item, 1);
4330 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4331 btrfs_set_extent_flags(leaf, extent_item,
4332 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4334 if (skinny_metadata) {
4335 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4337 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4338 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4339 btrfs_set_tree_block_level(leaf, block_info, ref->level);
4340 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4343 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4344 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4345 btrfs_set_extent_inline_ref_type(leaf, iref,
4346 BTRFS_SHARED_BLOCK_REF_KEY);
4347 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4349 btrfs_set_extent_inline_ref_type(leaf, iref,
4350 BTRFS_TREE_BLOCK_REF_KEY);
4351 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4354 btrfs_mark_buffer_dirty(leaf);
4355 btrfs_free_path(path);
4357 ret = remove_from_free_space_tree(trans, extent_key.objectid,
4362 ret = btrfs_update_block_group(trans, extent_key.objectid,
4363 fs_info->nodesize, 1);
4364 if (ret) { /* -ENOENT, logic error */
4365 btrfs_err(fs_info, "update block group failed for %llu %llu",
4366 extent_key.objectid, extent_key.offset);
4370 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
4375 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4376 struct btrfs_root *root, u64 owner,
4377 u64 offset, u64 ram_bytes,
4378 struct btrfs_key *ins)
4380 struct btrfs_ref generic_ref = { 0 };
4383 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4385 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4386 ins->objectid, ins->offset, 0);
4387 btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset);
4388 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4389 ret = btrfs_add_delayed_data_ref(trans, &generic_ref,
4390 ram_bytes, NULL, NULL);
4395 * this is used by the tree logging recovery code. It records that
4396 * an extent has been allocated and makes sure to clear the free
4397 * space cache bits as well
4399 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4400 u64 root_objectid, u64 owner, u64 offset,
4401 struct btrfs_key *ins)
4403 struct btrfs_fs_info *fs_info = trans->fs_info;
4405 struct btrfs_block_group *block_group;
4406 struct btrfs_space_info *space_info;
4409 * Mixed block groups will exclude before processing the log so we only
4410 * need to do the exclude dance if this fs isn't mixed.
4412 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4413 ret = __exclude_logged_extent(fs_info, ins->objectid,
4419 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4423 space_info = block_group->space_info;
4424 spin_lock(&space_info->lock);
4425 spin_lock(&block_group->lock);
4426 space_info->bytes_reserved += ins->offset;
4427 block_group->reserved += ins->offset;
4428 spin_unlock(&block_group->lock);
4429 spin_unlock(&space_info->lock);
4431 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4433 btrfs_put_block_group(block_group);
4437 static struct extent_buffer *
4438 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4439 u64 bytenr, int level, u64 owner)
4441 struct btrfs_fs_info *fs_info = root->fs_info;
4442 struct extent_buffer *buf;
4444 buf = btrfs_find_create_tree_block(fs_info, bytenr);
4449 * Extra safety check in case the extent tree is corrupted and extent
4450 * allocator chooses to use a tree block which is already used and
4453 if (buf->lock_owner == current->pid) {
4454 btrfs_err_rl(fs_info,
4455 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4456 buf->start, btrfs_header_owner(buf), current->pid);
4457 free_extent_buffer(buf);
4458 return ERR_PTR(-EUCLEAN);
4461 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
4462 btrfs_tree_lock(buf);
4463 btrfs_clean_tree_block(buf);
4464 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
4466 btrfs_set_lock_blocking_write(buf);
4467 set_extent_buffer_uptodate(buf);
4469 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
4470 btrfs_set_header_level(buf, level);
4471 btrfs_set_header_bytenr(buf, buf->start);
4472 btrfs_set_header_generation(buf, trans->transid);
4473 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
4474 btrfs_set_header_owner(buf, owner);
4475 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
4476 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
4477 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4478 buf->log_index = root->log_transid % 2;
4480 * we allow two log transactions at a time, use different
4481 * EXTENT bit to differentiate dirty pages.
4483 if (buf->log_index == 0)
4484 set_extent_dirty(&root->dirty_log_pages, buf->start,
4485 buf->start + buf->len - 1, GFP_NOFS);
4487 set_extent_new(&root->dirty_log_pages, buf->start,
4488 buf->start + buf->len - 1);
4490 buf->log_index = -1;
4491 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4492 buf->start + buf->len - 1, GFP_NOFS);
4494 trans->dirty = true;
4495 /* this returns a buffer locked for blocking */
4500 * finds a free extent and does all the dirty work required for allocation
4501 * returns the tree buffer or an ERR_PTR on error.
4503 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
4504 struct btrfs_root *root,
4505 u64 parent, u64 root_objectid,
4506 const struct btrfs_disk_key *key,
4507 int level, u64 hint,
4510 struct btrfs_fs_info *fs_info = root->fs_info;
4511 struct btrfs_key ins;
4512 struct btrfs_block_rsv *block_rsv;
4513 struct extent_buffer *buf;
4514 struct btrfs_delayed_extent_op *extent_op;
4515 struct btrfs_ref generic_ref = { 0 };
4518 u32 blocksize = fs_info->nodesize;
4519 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4521 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4522 if (btrfs_is_testing(fs_info)) {
4523 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
4524 level, root_objectid);
4526 root->alloc_bytenr += blocksize;
4531 block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
4532 if (IS_ERR(block_rsv))
4533 return ERR_CAST(block_rsv);
4535 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
4536 empty_size, hint, &ins, 0, 0);
4540 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
4544 goto out_free_reserved;
4547 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4549 parent = ins.objectid;
4550 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4554 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4555 extent_op = btrfs_alloc_delayed_extent_op();
4561 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4563 memset(&extent_op->key, 0, sizeof(extent_op->key));
4564 extent_op->flags_to_set = flags;
4565 extent_op->update_key = skinny_metadata ? false : true;
4566 extent_op->update_flags = true;
4567 extent_op->is_data = false;
4568 extent_op->level = level;
4570 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4571 ins.objectid, ins.offset, parent);
4572 generic_ref.real_root = root->root_key.objectid;
4573 btrfs_init_tree_ref(&generic_ref, level, root_objectid);
4574 btrfs_ref_tree_mod(fs_info, &generic_ref);
4575 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref,
4576 extent_op, NULL, NULL);
4578 goto out_free_delayed;
4583 btrfs_free_delayed_extent_op(extent_op);
4585 free_extent_buffer(buf);
4587 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
4589 btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
4590 return ERR_PTR(ret);
4593 struct walk_control {
4594 u64 refs[BTRFS_MAX_LEVEL];
4595 u64 flags[BTRFS_MAX_LEVEL];
4596 struct btrfs_key update_progress;
4597 struct btrfs_key drop_progress;
4609 #define DROP_REFERENCE 1
4610 #define UPDATE_BACKREF 2
4612 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4613 struct btrfs_root *root,
4614 struct walk_control *wc,
4615 struct btrfs_path *path)
4617 struct btrfs_fs_info *fs_info = root->fs_info;
4623 struct btrfs_key key;
4624 struct extent_buffer *eb;
4629 if (path->slots[wc->level] < wc->reada_slot) {
4630 wc->reada_count = wc->reada_count * 2 / 3;
4631 wc->reada_count = max(wc->reada_count, 2);
4633 wc->reada_count = wc->reada_count * 3 / 2;
4634 wc->reada_count = min_t(int, wc->reada_count,
4635 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
4638 eb = path->nodes[wc->level];
4639 nritems = btrfs_header_nritems(eb);
4641 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4642 if (nread >= wc->reada_count)
4646 bytenr = btrfs_node_blockptr(eb, slot);
4647 generation = btrfs_node_ptr_generation(eb, slot);
4649 if (slot == path->slots[wc->level])
4652 if (wc->stage == UPDATE_BACKREF &&
4653 generation <= root->root_key.offset)
4656 /* We don't lock the tree block, it's OK to be racy here */
4657 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
4658 wc->level - 1, 1, &refs,
4660 /* We don't care about errors in readahead. */
4665 if (wc->stage == DROP_REFERENCE) {
4669 if (wc->level == 1 &&
4670 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4672 if (!wc->update_ref ||
4673 generation <= root->root_key.offset)
4675 btrfs_node_key_to_cpu(eb, &key, slot);
4676 ret = btrfs_comp_cpu_keys(&key,
4677 &wc->update_progress);
4681 if (wc->level == 1 &&
4682 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4686 readahead_tree_block(fs_info, bytenr);
4689 wc->reada_slot = slot;
4693 * helper to process tree block while walking down the tree.
4695 * when wc->stage == UPDATE_BACKREF, this function updates
4696 * back refs for pointers in the block.
4698 * NOTE: return value 1 means we should stop walking down.
4700 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4701 struct btrfs_root *root,
4702 struct btrfs_path *path,
4703 struct walk_control *wc, int lookup_info)
4705 struct btrfs_fs_info *fs_info = root->fs_info;
4706 int level = wc->level;
4707 struct extent_buffer *eb = path->nodes[level];
4708 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4711 if (wc->stage == UPDATE_BACKREF &&
4712 btrfs_header_owner(eb) != root->root_key.objectid)
4716 * when reference count of tree block is 1, it won't increase
4717 * again. once full backref flag is set, we never clear it.
4720 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4721 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
4722 BUG_ON(!path->locks[level]);
4723 ret = btrfs_lookup_extent_info(trans, fs_info,
4724 eb->start, level, 1,
4727 BUG_ON(ret == -ENOMEM);
4730 BUG_ON(wc->refs[level] == 0);
4733 if (wc->stage == DROP_REFERENCE) {
4734 if (wc->refs[level] > 1)
4737 if (path->locks[level] && !wc->keep_locks) {
4738 btrfs_tree_unlock_rw(eb, path->locks[level]);
4739 path->locks[level] = 0;
4744 /* wc->stage == UPDATE_BACKREF */
4745 if (!(wc->flags[level] & flag)) {
4746 BUG_ON(!path->locks[level]);
4747 ret = btrfs_inc_ref(trans, root, eb, 1);
4748 BUG_ON(ret); /* -ENOMEM */
4749 ret = btrfs_dec_ref(trans, root, eb, 0);
4750 BUG_ON(ret); /* -ENOMEM */
4751 ret = btrfs_set_disk_extent_flags(trans, eb->start,
4753 btrfs_header_level(eb), 0);
4754 BUG_ON(ret); /* -ENOMEM */
4755 wc->flags[level] |= flag;
4759 * the block is shared by multiple trees, so it's not good to
4760 * keep the tree lock
4762 if (path->locks[level] && level > 0) {
4763 btrfs_tree_unlock_rw(eb, path->locks[level]);
4764 path->locks[level] = 0;
4770 * This is used to verify a ref exists for this root to deal with a bug where we
4771 * would have a drop_progress key that hadn't been updated properly.
4773 static int check_ref_exists(struct btrfs_trans_handle *trans,
4774 struct btrfs_root *root, u64 bytenr, u64 parent,
4777 struct btrfs_path *path;
4778 struct btrfs_extent_inline_ref *iref;
4781 path = btrfs_alloc_path();
4785 ret = lookup_extent_backref(trans, path, &iref, bytenr,
4786 root->fs_info->nodesize, parent,
4787 root->root_key.objectid, level, 0);
4788 btrfs_free_path(path);
4797 * helper to process tree block pointer.
4799 * when wc->stage == DROP_REFERENCE, this function checks
4800 * reference count of the block pointed to. if the block
4801 * is shared and we need update back refs for the subtree
4802 * rooted at the block, this function changes wc->stage to
4803 * UPDATE_BACKREF. if the block is shared and there is no
4804 * need to update back, this function drops the reference
4807 * NOTE: return value 1 means we should stop walking down.
4809 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
4810 struct btrfs_root *root,
4811 struct btrfs_path *path,
4812 struct walk_control *wc, int *lookup_info)
4814 struct btrfs_fs_info *fs_info = root->fs_info;
4818 struct btrfs_key key;
4819 struct btrfs_key first_key;
4820 struct btrfs_ref ref = { 0 };
4821 struct extent_buffer *next;
4822 int level = wc->level;
4825 bool need_account = false;
4827 generation = btrfs_node_ptr_generation(path->nodes[level],
4828 path->slots[level]);
4830 * if the lower level block was created before the snapshot
4831 * was created, we know there is no need to update back refs
4834 if (wc->stage == UPDATE_BACKREF &&
4835 generation <= root->root_key.offset) {
4840 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
4841 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
4842 path->slots[level]);
4844 next = find_extent_buffer(fs_info, bytenr);
4846 next = btrfs_find_create_tree_block(fs_info, bytenr);
4848 return PTR_ERR(next);
4850 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
4854 btrfs_tree_lock(next);
4855 btrfs_set_lock_blocking_write(next);
4857 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
4858 &wc->refs[level - 1],
4859 &wc->flags[level - 1]);
4863 if (unlikely(wc->refs[level - 1] == 0)) {
4864 btrfs_err(fs_info, "Missing references.");
4870 if (wc->stage == DROP_REFERENCE) {
4871 if (wc->refs[level - 1] > 1) {
4872 need_account = true;
4874 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4877 if (!wc->update_ref ||
4878 generation <= root->root_key.offset)
4881 btrfs_node_key_to_cpu(path->nodes[level], &key,
4882 path->slots[level]);
4883 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
4887 wc->stage = UPDATE_BACKREF;
4888 wc->shared_level = level - 1;
4892 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4896 if (!btrfs_buffer_uptodate(next, generation, 0)) {
4897 btrfs_tree_unlock(next);
4898 free_extent_buffer(next);
4904 if (reada && level == 1)
4905 reada_walk_down(trans, root, wc, path);
4906 next = read_tree_block(fs_info, bytenr, generation, level - 1,
4909 return PTR_ERR(next);
4910 } else if (!extent_buffer_uptodate(next)) {
4911 free_extent_buffer(next);
4914 btrfs_tree_lock(next);
4915 btrfs_set_lock_blocking_write(next);
4919 ASSERT(level == btrfs_header_level(next));
4920 if (level != btrfs_header_level(next)) {
4921 btrfs_err(root->fs_info, "mismatched level");
4925 path->nodes[level] = next;
4926 path->slots[level] = 0;
4927 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
4933 wc->refs[level - 1] = 0;
4934 wc->flags[level - 1] = 0;
4935 if (wc->stage == DROP_REFERENCE) {
4936 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
4937 parent = path->nodes[level]->start;
4939 ASSERT(root->root_key.objectid ==
4940 btrfs_header_owner(path->nodes[level]));
4941 if (root->root_key.objectid !=
4942 btrfs_header_owner(path->nodes[level])) {
4943 btrfs_err(root->fs_info,
4944 "mismatched block owner");
4952 * If we had a drop_progress we need to verify the refs are set
4953 * as expected. If we find our ref then we know that from here
4954 * on out everything should be correct, and we can clear the
4957 if (wc->restarted) {
4958 ret = check_ref_exists(trans, root, bytenr, parent,
4969 * Reloc tree doesn't contribute to qgroup numbers, and we have
4970 * already accounted them at merge time (replace_path),
4971 * thus we could skip expensive subtree trace here.
4973 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
4975 ret = btrfs_qgroup_trace_subtree(trans, next,
4976 generation, level - 1);
4978 btrfs_err_rl(fs_info,
4979 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
4985 * We need to update the next key in our walk control so we can
4986 * update the drop_progress key accordingly. We don't care if
4987 * find_next_key doesn't find a key because that means we're at
4988 * the end and are going to clean up now.
4990 wc->drop_level = level;
4991 find_next_key(path, level, &wc->drop_progress);
4993 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
4994 fs_info->nodesize, parent);
4995 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid);
4996 ret = btrfs_free_extent(trans, &ref);
5005 btrfs_tree_unlock(next);
5006 free_extent_buffer(next);
5012 * helper to process tree block while walking up the tree.
5014 * when wc->stage == DROP_REFERENCE, this function drops
5015 * reference count on the block.
5017 * when wc->stage == UPDATE_BACKREF, this function changes
5018 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5019 * to UPDATE_BACKREF previously while processing the block.
5021 * NOTE: return value 1 means we should stop walking up.
5023 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5024 struct btrfs_root *root,
5025 struct btrfs_path *path,
5026 struct walk_control *wc)
5028 struct btrfs_fs_info *fs_info = root->fs_info;
5030 int level = wc->level;
5031 struct extent_buffer *eb = path->nodes[level];
5034 if (wc->stage == UPDATE_BACKREF) {
5035 BUG_ON(wc->shared_level < level);
5036 if (level < wc->shared_level)
5039 ret = find_next_key(path, level + 1, &wc->update_progress);
5043 wc->stage = DROP_REFERENCE;
5044 wc->shared_level = -1;
5045 path->slots[level] = 0;
5048 * check reference count again if the block isn't locked.
5049 * we should start walking down the tree again if reference
5052 if (!path->locks[level]) {
5054 btrfs_tree_lock(eb);
5055 btrfs_set_lock_blocking_write(eb);
5056 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5058 ret = btrfs_lookup_extent_info(trans, fs_info,
5059 eb->start, level, 1,
5063 btrfs_tree_unlock_rw(eb, path->locks[level]);
5064 path->locks[level] = 0;
5067 BUG_ON(wc->refs[level] == 0);
5068 if (wc->refs[level] == 1) {
5069 btrfs_tree_unlock_rw(eb, path->locks[level]);
5070 path->locks[level] = 0;
5076 /* wc->stage == DROP_REFERENCE */
5077 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5079 if (wc->refs[level] == 1) {
5081 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5082 ret = btrfs_dec_ref(trans, root, eb, 1);
5084 ret = btrfs_dec_ref(trans, root, eb, 0);
5085 BUG_ON(ret); /* -ENOMEM */
5086 if (is_fstree(root->root_key.objectid)) {
5087 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5089 btrfs_err_rl(fs_info,
5090 "error %d accounting leaf items, quota is out of sync, rescan required",
5095 /* make block locked assertion in btrfs_clean_tree_block happy */
5096 if (!path->locks[level] &&
5097 btrfs_header_generation(eb) == trans->transid) {
5098 btrfs_tree_lock(eb);
5099 btrfs_set_lock_blocking_write(eb);
5100 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5102 btrfs_clean_tree_block(eb);
5105 if (eb == root->node) {
5106 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5108 else if (root->root_key.objectid != btrfs_header_owner(eb))
5109 goto owner_mismatch;
5111 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5112 parent = path->nodes[level + 1]->start;
5113 else if (root->root_key.objectid !=
5114 btrfs_header_owner(path->nodes[level + 1]))
5115 goto owner_mismatch;
5118 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
5120 wc->refs[level] = 0;
5121 wc->flags[level] = 0;
5125 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5126 btrfs_header_owner(eb), root->root_key.objectid);
5130 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5131 struct btrfs_root *root,
5132 struct btrfs_path *path,
5133 struct walk_control *wc)
5135 int level = wc->level;
5136 int lookup_info = 1;
5139 while (level >= 0) {
5140 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5147 if (path->slots[level] >=
5148 btrfs_header_nritems(path->nodes[level]))
5151 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5153 path->slots[level]++;
5162 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5163 struct btrfs_root *root,
5164 struct btrfs_path *path,
5165 struct walk_control *wc, int max_level)
5167 int level = wc->level;
5170 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5171 while (level < max_level && path->nodes[level]) {
5173 if (path->slots[level] + 1 <
5174 btrfs_header_nritems(path->nodes[level])) {
5175 path->slots[level]++;
5178 ret = walk_up_proc(trans, root, path, wc);
5184 if (path->locks[level]) {
5185 btrfs_tree_unlock_rw(path->nodes[level],
5186 path->locks[level]);
5187 path->locks[level] = 0;
5189 free_extent_buffer(path->nodes[level]);
5190 path->nodes[level] = NULL;
5198 * drop a subvolume tree.
5200 * this function traverses the tree freeing any blocks that only
5201 * referenced by the tree.
5203 * when a shared tree block is found. this function decreases its
5204 * reference count by one. if update_ref is true, this function
5205 * also make sure backrefs for the shared block and all lower level
5206 * blocks are properly updated.
5208 * If called with for_reloc == 0, may exit early with -EAGAIN
5210 int btrfs_drop_snapshot(struct btrfs_root *root,
5211 struct btrfs_block_rsv *block_rsv, int update_ref,
5214 struct btrfs_fs_info *fs_info = root->fs_info;
5215 struct btrfs_path *path;
5216 struct btrfs_trans_handle *trans;
5217 struct btrfs_root *tree_root = fs_info->tree_root;
5218 struct btrfs_root_item *root_item = &root->root_item;
5219 struct walk_control *wc;
5220 struct btrfs_key key;
5224 bool root_dropped = false;
5226 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5228 path = btrfs_alloc_path();
5234 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5236 btrfs_free_path(path);
5241 trans = btrfs_start_transaction(tree_root, 0);
5242 if (IS_ERR(trans)) {
5243 err = PTR_ERR(trans);
5247 err = btrfs_run_delayed_items(trans);
5252 trans->block_rsv = block_rsv;
5255 * This will help us catch people modifying the fs tree while we're
5256 * dropping it. It is unsafe to mess with the fs tree while it's being
5257 * dropped as we unlock the root node and parent nodes as we walk down
5258 * the tree, assuming nothing will change. If something does change
5259 * then we'll have stale information and drop references to blocks we've
5262 set_bit(BTRFS_ROOT_DELETING, &root->state);
5263 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5264 level = btrfs_header_level(root->node);
5265 path->nodes[level] = btrfs_lock_root_node(root);
5266 btrfs_set_lock_blocking_write(path->nodes[level]);
5267 path->slots[level] = 0;
5268 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5269 memset(&wc->update_progress, 0,
5270 sizeof(wc->update_progress));
5272 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5273 memcpy(&wc->update_progress, &key,
5274 sizeof(wc->update_progress));
5276 level = root_item->drop_level;
5278 path->lowest_level = level;
5279 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5280 path->lowest_level = 0;
5288 * unlock our path, this is safe because only this
5289 * function is allowed to delete this snapshot
5291 btrfs_unlock_up_safe(path, 0);
5293 level = btrfs_header_level(root->node);
5295 btrfs_tree_lock(path->nodes[level]);
5296 btrfs_set_lock_blocking_write(path->nodes[level]);
5297 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5299 ret = btrfs_lookup_extent_info(trans, fs_info,
5300 path->nodes[level]->start,
5301 level, 1, &wc->refs[level],
5307 BUG_ON(wc->refs[level] == 0);
5309 if (level == root_item->drop_level)
5312 btrfs_tree_unlock(path->nodes[level]);
5313 path->locks[level] = 0;
5314 WARN_ON(wc->refs[level] != 1);
5319 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5321 wc->shared_level = -1;
5322 wc->stage = DROP_REFERENCE;
5323 wc->update_ref = update_ref;
5325 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5329 ret = walk_down_tree(trans, root, path, wc);
5335 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5342 BUG_ON(wc->stage != DROP_REFERENCE);
5346 if (wc->stage == DROP_REFERENCE) {
5347 wc->drop_level = wc->level;
5348 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5350 path->slots[wc->drop_level]);
5352 btrfs_cpu_key_to_disk(&root_item->drop_progress,
5353 &wc->drop_progress);
5354 root_item->drop_level = wc->drop_level;
5356 BUG_ON(wc->level == 0);
5357 if (btrfs_should_end_transaction(trans) ||
5358 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5359 ret = btrfs_update_root(trans, tree_root,
5363 btrfs_abort_transaction(trans, ret);
5368 btrfs_end_transaction_throttle(trans);
5369 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5370 btrfs_debug(fs_info,
5371 "drop snapshot early exit");
5376 trans = btrfs_start_transaction(tree_root, 0);
5377 if (IS_ERR(trans)) {
5378 err = PTR_ERR(trans);
5382 trans->block_rsv = block_rsv;
5385 btrfs_release_path(path);
5389 ret = btrfs_del_root(trans, &root->root_key);
5391 btrfs_abort_transaction(trans, ret);
5396 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5397 ret = btrfs_find_root(tree_root, &root->root_key, path,
5400 btrfs_abort_transaction(trans, ret);
5403 } else if (ret > 0) {
5404 /* if we fail to delete the orphan item this time
5405 * around, it'll get picked up the next time.
5407 * The most common failure here is just -ENOENT.
5409 btrfs_del_orphan_item(trans, tree_root,
5410 root->root_key.objectid);
5414 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
5415 btrfs_add_dropped_root(trans, root);
5417 free_extent_buffer(root->node);
5418 free_extent_buffer(root->commit_root);
5419 btrfs_put_fs_root(root);
5421 root_dropped = true;
5423 btrfs_end_transaction_throttle(trans);
5426 btrfs_free_path(path);
5429 * So if we need to stop dropping the snapshot for whatever reason we
5430 * need to make sure to add it back to the dead root list so that we
5431 * keep trying to do the work later. This also cleans up roots if we
5432 * don't have it in the radix (like when we recover after a power fail
5433 * or unmount) so we don't leak memory.
5435 if (!for_reloc && !root_dropped)
5436 btrfs_add_dead_root(root);
5437 if (err && err != -EAGAIN)
5438 btrfs_handle_fs_error(fs_info, err, NULL);
5443 * drop subtree rooted at tree block 'node'.
5445 * NOTE: this function will unlock and release tree block 'node'
5446 * only used by relocation code
5448 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5449 struct btrfs_root *root,
5450 struct extent_buffer *node,
5451 struct extent_buffer *parent)
5453 struct btrfs_fs_info *fs_info = root->fs_info;
5454 struct btrfs_path *path;
5455 struct walk_control *wc;
5461 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5463 path = btrfs_alloc_path();
5467 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5469 btrfs_free_path(path);
5473 btrfs_assert_tree_locked(parent);
5474 parent_level = btrfs_header_level(parent);
5475 atomic_inc(&parent->refs);
5476 path->nodes[parent_level] = parent;
5477 path->slots[parent_level] = btrfs_header_nritems(parent);
5479 btrfs_assert_tree_locked(node);
5480 level = btrfs_header_level(node);
5481 path->nodes[level] = node;
5482 path->slots[level] = 0;
5483 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5485 wc->refs[parent_level] = 1;
5486 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5488 wc->shared_level = -1;
5489 wc->stage = DROP_REFERENCE;
5492 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5495 wret = walk_down_tree(trans, root, path, wc);
5501 wret = walk_up_tree(trans, root, path, wc, parent_level);
5509 btrfs_free_path(path);
5514 * helper to account the unused space of all the readonly block group in the
5515 * space_info. takes mirrors into account.
5517 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
5519 struct btrfs_block_group *block_group;
5523 /* It's df, we don't care if it's racy */
5524 if (list_empty(&sinfo->ro_bgs))
5527 spin_lock(&sinfo->lock);
5528 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
5529 spin_lock(&block_group->lock);
5531 if (!block_group->ro) {
5532 spin_unlock(&block_group->lock);
5536 factor = btrfs_bg_type_to_factor(block_group->flags);
5537 free_bytes += (block_group->length -
5538 block_group->used) * factor;
5540 spin_unlock(&block_group->lock);
5542 spin_unlock(&sinfo->lock);
5547 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
5550 return unpin_extent_range(fs_info, start, end, false);
5554 * It used to be that old block groups would be left around forever.
5555 * Iterating over them would be enough to trim unused space. Since we
5556 * now automatically remove them, we also need to iterate over unallocated
5559 * We don't want a transaction for this since the discard may take a
5560 * substantial amount of time. We don't require that a transaction be
5561 * running, but we do need to take a running transaction into account
5562 * to ensure that we're not discarding chunks that were released or
5563 * allocated in the current transaction.
5565 * Holding the chunks lock will prevent other threads from allocating
5566 * or releasing chunks, but it won't prevent a running transaction
5567 * from committing and releasing the memory that the pending chunks
5568 * list head uses. For that, we need to take a reference to the
5569 * transaction and hold the commit root sem. We only need to hold
5570 * it while performing the free space search since we have already
5571 * held back allocations.
5573 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
5575 u64 start = SZ_1M, len = 0, end = 0;
5580 /* Discard not supported = nothing to do. */
5581 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
5584 /* Not writable = nothing to do. */
5585 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5588 /* No free space = nothing to do. */
5589 if (device->total_bytes <= device->bytes_used)
5595 struct btrfs_fs_info *fs_info = device->fs_info;
5598 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
5602 find_first_clear_extent_bit(&device->alloc_state, start,
5604 CHUNK_TRIMMED | CHUNK_ALLOCATED);
5606 /* Ensure we skip the reserved area in the first 1M */
5607 start = max_t(u64, start, SZ_1M);
5610 * If find_first_clear_extent_bit find a range that spans the
5611 * end of the device it will set end to -1, in this case it's up
5612 * to the caller to trim the value to the size of the device.
5614 end = min(end, device->total_bytes - 1);
5616 len = end - start + 1;
5618 /* We didn't find any extents */
5620 mutex_unlock(&fs_info->chunk_mutex);
5625 ret = btrfs_issue_discard(device->bdev, start, len,
5628 set_extent_bits(&device->alloc_state, start,
5631 mutex_unlock(&fs_info->chunk_mutex);
5639 if (fatal_signal_pending(current)) {
5651 * Trim the whole filesystem by:
5652 * 1) trimming the free space in each block group
5653 * 2) trimming the unallocated space on each device
5655 * This will also continue trimming even if a block group or device encounters
5656 * an error. The return value will be the last error, or 0 if nothing bad
5659 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
5661 struct btrfs_block_group *cache = NULL;
5662 struct btrfs_device *device;
5663 struct list_head *devices;
5665 u64 range_end = U64_MAX;
5676 * Check range overflow if range->len is set.
5677 * The default range->len is U64_MAX.
5679 if (range->len != U64_MAX &&
5680 check_add_overflow(range->start, range->len, &range_end))
5683 cache = btrfs_lookup_first_block_group(fs_info, range->start);
5684 for (; cache; cache = btrfs_next_block_group(cache)) {
5685 if (cache->start >= range_end) {
5686 btrfs_put_block_group(cache);
5690 start = max(range->start, cache->start);
5691 end = min(range_end, cache->start + cache->length);
5693 if (end - start >= range->minlen) {
5694 if (!btrfs_block_group_done(cache)) {
5695 ret = btrfs_cache_block_group(cache, 0);
5701 ret = btrfs_wait_block_group_cache_done(cache);
5708 ret = btrfs_trim_block_group(cache,
5714 trimmed += group_trimmed;
5725 "failed to trim %llu block group(s), last error %d",
5727 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5728 devices = &fs_info->fs_devices->devices;
5729 list_for_each_entry(device, devices, dev_list) {
5730 ret = btrfs_trim_free_extents(device, &group_trimmed);
5737 trimmed += group_trimmed;
5739 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5743 "failed to trim %llu device(s), last error %d",
5744 dev_failed, dev_ret);
5745 range->len = trimmed;
5752 * btrfs_{start,end}_write_no_snapshotting() are similar to
5753 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
5754 * data into the page cache through nocow before the subvolume is snapshoted,
5755 * but flush the data into disk after the snapshot creation, or to prevent
5756 * operations while snapshotting is ongoing and that cause the snapshot to be
5757 * inconsistent (writes followed by expanding truncates for example).
5759 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
5761 percpu_counter_dec(&root->subv_writers->counter);
5762 cond_wake_up(&root->subv_writers->wait);
5765 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
5767 if (atomic_read(&root->will_be_snapshotted))
5770 percpu_counter_inc(&root->subv_writers->counter);
5772 * Make sure counter is updated before we check for snapshot creation.
5775 if (atomic_read(&root->will_be_snapshotted)) {
5776 btrfs_end_write_no_snapshotting(root);
5782 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
5787 ret = btrfs_start_write_no_snapshotting(root);
5790 wait_var_event(&root->will_be_snapshotted,
5791 !atomic_read(&root->will_be_snapshotted));