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"
36 #include "rcu-string.h"
38 #include "dev-replace.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
44 struct btrfs_delayed_ref_node *node, u64 parent,
45 u64 root_objectid, u64 owner_objectid,
46 u64 owner_offset, int refs_to_drop,
47 struct btrfs_delayed_extent_op *extra_op);
48 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
49 struct extent_buffer *leaf,
50 struct btrfs_extent_item *ei);
51 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
52 u64 parent, u64 root_objectid,
53 u64 flags, u64 owner, u64 offset,
54 struct btrfs_key *ins, int ref_mod);
55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
56 struct btrfs_delayed_ref_node *node,
57 struct btrfs_delayed_extent_op *extent_op);
58 static int find_next_key(struct btrfs_path *path, int level,
59 struct btrfs_key *key);
61 static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
63 return (cache->flags & bits) == bits;
66 int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info,
67 u64 start, u64 num_bytes)
69 u64 end = start + num_bytes - 1;
70 set_extent_bits(&fs_info->excluded_extents, start, end,
75 void btrfs_free_excluded_extents(struct btrfs_block_group *cache)
77 struct btrfs_fs_info *fs_info = cache->fs_info;
81 end = start + cache->length - 1;
83 clear_extent_bits(&fs_info->excluded_extents, start, end,
87 /* simple helper to search for an existing data extent at a given offset */
88 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
92 struct btrfs_path *path;
94 path = btrfs_alloc_path();
100 key.type = BTRFS_EXTENT_ITEM_KEY;
101 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
102 btrfs_free_path(path);
107 * helper function to lookup reference count and flags of a tree block.
109 * the head node for delayed ref is used to store the sum of all the
110 * reference count modifications queued up in the rbtree. the head
111 * node may also store the extent flags to set. This way you can check
112 * to see what the reference count and extent flags would be if all of
113 * the delayed refs are not processed.
115 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
116 struct btrfs_fs_info *fs_info, u64 bytenr,
117 u64 offset, int metadata, u64 *refs, u64 *flags)
119 struct btrfs_delayed_ref_head *head;
120 struct btrfs_delayed_ref_root *delayed_refs;
121 struct btrfs_path *path;
122 struct btrfs_extent_item *ei;
123 struct extent_buffer *leaf;
124 struct btrfs_key key;
131 * If we don't have skinny metadata, don't bother doing anything
134 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
135 offset = fs_info->nodesize;
139 path = btrfs_alloc_path();
144 path->skip_locking = 1;
145 path->search_commit_root = 1;
149 key.objectid = bytenr;
152 key.type = BTRFS_METADATA_ITEM_KEY;
154 key.type = BTRFS_EXTENT_ITEM_KEY;
156 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
160 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
161 if (path->slots[0]) {
163 btrfs_item_key_to_cpu(path->nodes[0], &key,
165 if (key.objectid == bytenr &&
166 key.type == BTRFS_EXTENT_ITEM_KEY &&
167 key.offset == fs_info->nodesize)
173 leaf = path->nodes[0];
174 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
175 if (item_size >= sizeof(*ei)) {
176 ei = btrfs_item_ptr(leaf, path->slots[0],
177 struct btrfs_extent_item);
178 num_refs = btrfs_extent_refs(leaf, ei);
179 extent_flags = btrfs_extent_flags(leaf, ei);
182 btrfs_print_v0_err(fs_info);
184 btrfs_abort_transaction(trans, ret);
186 btrfs_handle_fs_error(fs_info, ret, NULL);
191 BUG_ON(num_refs == 0);
201 delayed_refs = &trans->transaction->delayed_refs;
202 spin_lock(&delayed_refs->lock);
203 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
205 if (!mutex_trylock(&head->mutex)) {
206 refcount_inc(&head->refs);
207 spin_unlock(&delayed_refs->lock);
209 btrfs_release_path(path);
212 * Mutex was contended, block until it's released and try
215 mutex_lock(&head->mutex);
216 mutex_unlock(&head->mutex);
217 btrfs_put_delayed_ref_head(head);
220 spin_lock(&head->lock);
221 if (head->extent_op && head->extent_op->update_flags)
222 extent_flags |= head->extent_op->flags_to_set;
224 BUG_ON(num_refs == 0);
226 num_refs += head->ref_mod;
227 spin_unlock(&head->lock);
228 mutex_unlock(&head->mutex);
230 spin_unlock(&delayed_refs->lock);
232 WARN_ON(num_refs == 0);
236 *flags = extent_flags;
238 btrfs_free_path(path);
243 * Back reference rules. Back refs have three main goals:
245 * 1) differentiate between all holders of references to an extent so that
246 * when a reference is dropped we can make sure it was a valid reference
247 * before freeing the extent.
249 * 2) Provide enough information to quickly find the holders of an extent
250 * if we notice a given block is corrupted or bad.
252 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
253 * maintenance. This is actually the same as #2, but with a slightly
254 * different use case.
256 * There are two kinds of back refs. The implicit back refs is optimized
257 * for pointers in non-shared tree blocks. For a given pointer in a block,
258 * back refs of this kind provide information about the block's owner tree
259 * and the pointer's key. These information allow us to find the block by
260 * b-tree searching. The full back refs is for pointers in tree blocks not
261 * referenced by their owner trees. The location of tree block is recorded
262 * in the back refs. Actually the full back refs is generic, and can be
263 * used in all cases the implicit back refs is used. The major shortcoming
264 * of the full back refs is its overhead. Every time a tree block gets
265 * COWed, we have to update back refs entry for all pointers in it.
267 * For a newly allocated tree block, we use implicit back refs for
268 * pointers in it. This means most tree related operations only involve
269 * implicit back refs. For a tree block created in old transaction, the
270 * only way to drop a reference to it is COW it. So we can detect the
271 * event that tree block loses its owner tree's reference and do the
272 * back refs conversion.
274 * When a tree block is COWed through a tree, there are four cases:
276 * The reference count of the block is one and the tree is the block's
277 * owner tree. Nothing to do in this case.
279 * The reference count of the block is one and the tree is not the
280 * block's owner tree. In this case, full back refs is used for pointers
281 * in the block. Remove these full back refs, add implicit back refs for
282 * every pointers in the new block.
284 * The reference count of the block is greater than one and the tree is
285 * the block's owner tree. In this case, implicit back refs is used for
286 * pointers in the block. Add full back refs for every pointers in the
287 * block, increase lower level extents' reference counts. The original
288 * implicit back refs are entailed to the new block.
290 * The reference count of the block is greater than one and the tree is
291 * not the block's owner tree. Add implicit back refs for every pointer in
292 * the new block, increase lower level extents' reference count.
294 * Back Reference Key composing:
296 * The key objectid corresponds to the first byte in the extent,
297 * The key type is used to differentiate between types of back refs.
298 * There are different meanings of the key offset for different types
301 * File extents can be referenced by:
303 * - multiple snapshots, subvolumes, or different generations in one subvol
304 * - different files inside a single subvolume
305 * - different offsets inside a file (bookend extents in file.c)
307 * The extent ref structure for the implicit back refs has fields for:
309 * - Objectid of the subvolume root
310 * - objectid of the file holding the reference
311 * - original offset in the file
312 * - how many bookend extents
314 * The key offset for the implicit back refs is hash of the first
317 * The extent ref structure for the full back refs has field for:
319 * - number of pointers in the tree leaf
321 * The key offset for the implicit back refs is the first byte of
324 * When a file extent is allocated, The implicit back refs is used.
325 * the fields are filled in:
327 * (root_key.objectid, inode objectid, offset in file, 1)
329 * When a file extent is removed file truncation, we find the
330 * corresponding implicit back refs and check the following fields:
332 * (btrfs_header_owner(leaf), inode objectid, offset in file)
334 * Btree extents can be referenced by:
336 * - Different subvolumes
338 * Both the implicit back refs and the full back refs for tree blocks
339 * only consist of key. The key offset for the implicit back refs is
340 * objectid of block's owner tree. The key offset for the full back refs
341 * is the first byte of parent block.
343 * When implicit back refs is used, information about the lowest key and
344 * level of the tree block are required. These information are stored in
345 * tree block info structure.
349 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
350 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
351 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
353 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
354 struct btrfs_extent_inline_ref *iref,
355 enum btrfs_inline_ref_type is_data)
357 int type = btrfs_extent_inline_ref_type(eb, iref);
358 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
360 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
361 type == BTRFS_SHARED_BLOCK_REF_KEY ||
362 type == BTRFS_SHARED_DATA_REF_KEY ||
363 type == BTRFS_EXTENT_DATA_REF_KEY) {
364 if (is_data == BTRFS_REF_TYPE_BLOCK) {
365 if (type == BTRFS_TREE_BLOCK_REF_KEY)
367 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
370 * Every shared one has parent tree block,
371 * which must be aligned to sector size.
374 IS_ALIGNED(offset, eb->fs_info->sectorsize))
377 } else if (is_data == BTRFS_REF_TYPE_DATA) {
378 if (type == BTRFS_EXTENT_DATA_REF_KEY)
380 if (type == BTRFS_SHARED_DATA_REF_KEY) {
383 * Every shared one has parent tree block,
384 * which must be aligned to sector size.
387 IS_ALIGNED(offset, eb->fs_info->sectorsize))
391 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
396 btrfs_print_leaf((struct extent_buffer *)eb);
397 btrfs_err(eb->fs_info,
398 "eb %llu iref 0x%lx invalid extent inline ref type %d",
399 eb->start, (unsigned long)iref, type);
402 return BTRFS_REF_TYPE_INVALID;
405 u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
407 u32 high_crc = ~(u32)0;
408 u32 low_crc = ~(u32)0;
411 lenum = cpu_to_le64(root_objectid);
412 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
413 lenum = cpu_to_le64(owner);
414 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
415 lenum = cpu_to_le64(offset);
416 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
418 return ((u64)high_crc << 31) ^ (u64)low_crc;
421 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
422 struct btrfs_extent_data_ref *ref)
424 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
425 btrfs_extent_data_ref_objectid(leaf, ref),
426 btrfs_extent_data_ref_offset(leaf, ref));
429 static int match_extent_data_ref(struct extent_buffer *leaf,
430 struct btrfs_extent_data_ref *ref,
431 u64 root_objectid, u64 owner, u64 offset)
433 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
434 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
435 btrfs_extent_data_ref_offset(leaf, ref) != offset)
440 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
441 struct btrfs_path *path,
442 u64 bytenr, u64 parent,
444 u64 owner, u64 offset)
446 struct btrfs_root *root = trans->fs_info->extent_root;
447 struct btrfs_key key;
448 struct btrfs_extent_data_ref *ref;
449 struct extent_buffer *leaf;
455 key.objectid = bytenr;
457 key.type = BTRFS_SHARED_DATA_REF_KEY;
460 key.type = BTRFS_EXTENT_DATA_REF_KEY;
461 key.offset = hash_extent_data_ref(root_objectid,
466 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
478 leaf = path->nodes[0];
479 nritems = btrfs_header_nritems(leaf);
481 if (path->slots[0] >= nritems) {
482 ret = btrfs_next_leaf(root, path);
488 leaf = path->nodes[0];
489 nritems = btrfs_header_nritems(leaf);
493 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
494 if (key.objectid != bytenr ||
495 key.type != BTRFS_EXTENT_DATA_REF_KEY)
498 ref = btrfs_item_ptr(leaf, path->slots[0],
499 struct btrfs_extent_data_ref);
501 if (match_extent_data_ref(leaf, ref, root_objectid,
504 btrfs_release_path(path);
516 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
517 struct btrfs_path *path,
518 u64 bytenr, u64 parent,
519 u64 root_objectid, u64 owner,
520 u64 offset, int refs_to_add)
522 struct btrfs_root *root = trans->fs_info->extent_root;
523 struct btrfs_key key;
524 struct extent_buffer *leaf;
529 key.objectid = bytenr;
531 key.type = BTRFS_SHARED_DATA_REF_KEY;
533 size = sizeof(struct btrfs_shared_data_ref);
535 key.type = BTRFS_EXTENT_DATA_REF_KEY;
536 key.offset = hash_extent_data_ref(root_objectid,
538 size = sizeof(struct btrfs_extent_data_ref);
541 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
542 if (ret && ret != -EEXIST)
545 leaf = path->nodes[0];
547 struct btrfs_shared_data_ref *ref;
548 ref = btrfs_item_ptr(leaf, path->slots[0],
549 struct btrfs_shared_data_ref);
551 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
553 num_refs = btrfs_shared_data_ref_count(leaf, ref);
554 num_refs += refs_to_add;
555 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
558 struct btrfs_extent_data_ref *ref;
559 while (ret == -EEXIST) {
560 ref = btrfs_item_ptr(leaf, path->slots[0],
561 struct btrfs_extent_data_ref);
562 if (match_extent_data_ref(leaf, ref, root_objectid,
565 btrfs_release_path(path);
567 ret = btrfs_insert_empty_item(trans, root, path, &key,
569 if (ret && ret != -EEXIST)
572 leaf = path->nodes[0];
574 ref = btrfs_item_ptr(leaf, path->slots[0],
575 struct btrfs_extent_data_ref);
577 btrfs_set_extent_data_ref_root(leaf, ref,
579 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
580 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
581 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
583 num_refs = btrfs_extent_data_ref_count(leaf, ref);
584 num_refs += refs_to_add;
585 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
588 btrfs_mark_buffer_dirty(leaf);
591 btrfs_release_path(path);
595 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
596 struct btrfs_path *path,
597 int refs_to_drop, int *last_ref)
599 struct btrfs_key key;
600 struct btrfs_extent_data_ref *ref1 = NULL;
601 struct btrfs_shared_data_ref *ref2 = NULL;
602 struct extent_buffer *leaf;
606 leaf = path->nodes[0];
607 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
609 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
610 ref1 = btrfs_item_ptr(leaf, path->slots[0],
611 struct btrfs_extent_data_ref);
612 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
613 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
614 ref2 = btrfs_item_ptr(leaf, path->slots[0],
615 struct btrfs_shared_data_ref);
616 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
617 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
618 btrfs_print_v0_err(trans->fs_info);
619 btrfs_abort_transaction(trans, -EINVAL);
625 BUG_ON(num_refs < refs_to_drop);
626 num_refs -= refs_to_drop;
629 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
632 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
633 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
634 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
635 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
636 btrfs_mark_buffer_dirty(leaf);
641 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
642 struct btrfs_extent_inline_ref *iref)
644 struct btrfs_key key;
645 struct extent_buffer *leaf;
646 struct btrfs_extent_data_ref *ref1;
647 struct btrfs_shared_data_ref *ref2;
651 leaf = path->nodes[0];
652 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
654 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
657 * If type is invalid, we should have bailed out earlier than
660 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
661 ASSERT(type != BTRFS_REF_TYPE_INVALID);
662 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
663 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
664 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
666 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
667 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
669 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
670 ref1 = btrfs_item_ptr(leaf, path->slots[0],
671 struct btrfs_extent_data_ref);
672 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
673 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
674 ref2 = btrfs_item_ptr(leaf, path->slots[0],
675 struct btrfs_shared_data_ref);
676 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
683 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
684 struct btrfs_path *path,
685 u64 bytenr, u64 parent,
688 struct btrfs_root *root = trans->fs_info->extent_root;
689 struct btrfs_key key;
692 key.objectid = bytenr;
694 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
697 key.type = BTRFS_TREE_BLOCK_REF_KEY;
698 key.offset = root_objectid;
701 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
707 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
708 struct btrfs_path *path,
709 u64 bytenr, u64 parent,
712 struct btrfs_key key;
715 key.objectid = bytenr;
717 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
720 key.type = BTRFS_TREE_BLOCK_REF_KEY;
721 key.offset = root_objectid;
724 ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root,
726 btrfs_release_path(path);
730 static inline int extent_ref_type(u64 parent, u64 owner)
733 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
735 type = BTRFS_SHARED_BLOCK_REF_KEY;
737 type = BTRFS_TREE_BLOCK_REF_KEY;
740 type = BTRFS_SHARED_DATA_REF_KEY;
742 type = BTRFS_EXTENT_DATA_REF_KEY;
747 static int find_next_key(struct btrfs_path *path, int level,
748 struct btrfs_key *key)
751 for (; level < BTRFS_MAX_LEVEL; level++) {
752 if (!path->nodes[level])
754 if (path->slots[level] + 1 >=
755 btrfs_header_nritems(path->nodes[level]))
758 btrfs_item_key_to_cpu(path->nodes[level], key,
759 path->slots[level] + 1);
761 btrfs_node_key_to_cpu(path->nodes[level], key,
762 path->slots[level] + 1);
769 * look for inline back ref. if back ref is found, *ref_ret is set
770 * to the address of inline back ref, and 0 is returned.
772 * if back ref isn't found, *ref_ret is set to the address where it
773 * should be inserted, and -ENOENT is returned.
775 * if insert is true and there are too many inline back refs, the path
776 * points to the extent item, and -EAGAIN is returned.
778 * NOTE: inline back refs are ordered in the same way that back ref
779 * items in the tree are ordered.
781 static noinline_for_stack
782 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
783 struct btrfs_path *path,
784 struct btrfs_extent_inline_ref **ref_ret,
785 u64 bytenr, u64 num_bytes,
786 u64 parent, u64 root_objectid,
787 u64 owner, u64 offset, int insert)
789 struct btrfs_fs_info *fs_info = trans->fs_info;
790 struct btrfs_root *root = fs_info->extent_root;
791 struct btrfs_key key;
792 struct extent_buffer *leaf;
793 struct btrfs_extent_item *ei;
794 struct btrfs_extent_inline_ref *iref;
804 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
807 key.objectid = bytenr;
808 key.type = BTRFS_EXTENT_ITEM_KEY;
809 key.offset = num_bytes;
811 want = extent_ref_type(parent, owner);
813 extra_size = btrfs_extent_inline_ref_size(want);
814 path->search_for_extension = 1;
815 path->keep_locks = 1;
820 * Owner is our level, so we can just add one to get the level for the
821 * block we are interested in.
823 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
824 key.type = BTRFS_METADATA_ITEM_KEY;
829 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
836 * We may be a newly converted file system which still has the old fat
837 * extent entries for metadata, so try and see if we have one of those.
839 if (ret > 0 && skinny_metadata) {
840 skinny_metadata = false;
841 if (path->slots[0]) {
843 btrfs_item_key_to_cpu(path->nodes[0], &key,
845 if (key.objectid == bytenr &&
846 key.type == BTRFS_EXTENT_ITEM_KEY &&
847 key.offset == num_bytes)
851 key.objectid = bytenr;
852 key.type = BTRFS_EXTENT_ITEM_KEY;
853 key.offset = num_bytes;
854 btrfs_release_path(path);
859 if (ret && !insert) {
862 } else if (WARN_ON(ret)) {
867 leaf = path->nodes[0];
868 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
869 if (unlikely(item_size < sizeof(*ei))) {
871 btrfs_print_v0_err(fs_info);
872 btrfs_abort_transaction(trans, err);
876 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
877 flags = btrfs_extent_flags(leaf, ei);
879 ptr = (unsigned long)(ei + 1);
880 end = (unsigned long)ei + item_size;
882 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
883 ptr += sizeof(struct btrfs_tree_block_info);
887 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
888 needed = BTRFS_REF_TYPE_DATA;
890 needed = BTRFS_REF_TYPE_BLOCK;
898 iref = (struct btrfs_extent_inline_ref *)ptr;
899 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
900 if (type == BTRFS_REF_TYPE_INVALID) {
908 ptr += btrfs_extent_inline_ref_size(type);
912 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
913 struct btrfs_extent_data_ref *dref;
914 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
915 if (match_extent_data_ref(leaf, dref, root_objectid,
920 if (hash_extent_data_ref_item(leaf, dref) <
921 hash_extent_data_ref(root_objectid, owner, offset))
925 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
927 if (parent == ref_offset) {
931 if (ref_offset < parent)
934 if (root_objectid == ref_offset) {
938 if (ref_offset < root_objectid)
942 ptr += btrfs_extent_inline_ref_size(type);
944 if (err == -ENOENT && insert) {
945 if (item_size + extra_size >=
946 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
951 * To add new inline back ref, we have to make sure
952 * there is no corresponding back ref item.
953 * For simplicity, we just do not add new inline back
954 * ref if there is any kind of item for this block
956 if (find_next_key(path, 0, &key) == 0 &&
957 key.objectid == bytenr &&
958 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
963 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
966 path->keep_locks = 0;
967 path->search_for_extension = 0;
968 btrfs_unlock_up_safe(path, 1);
974 * helper to add new inline back ref
976 static noinline_for_stack
977 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
978 struct btrfs_path *path,
979 struct btrfs_extent_inline_ref *iref,
980 u64 parent, u64 root_objectid,
981 u64 owner, u64 offset, int refs_to_add,
982 struct btrfs_delayed_extent_op *extent_op)
984 struct extent_buffer *leaf;
985 struct btrfs_extent_item *ei;
988 unsigned long item_offset;
993 leaf = path->nodes[0];
994 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
995 item_offset = (unsigned long)iref - (unsigned long)ei;
997 type = extent_ref_type(parent, owner);
998 size = btrfs_extent_inline_ref_size(type);
1000 btrfs_extend_item(path, size);
1002 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1003 refs = btrfs_extent_refs(leaf, ei);
1004 refs += refs_to_add;
1005 btrfs_set_extent_refs(leaf, ei, refs);
1007 __run_delayed_extent_op(extent_op, leaf, ei);
1009 ptr = (unsigned long)ei + item_offset;
1010 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1011 if (ptr < end - size)
1012 memmove_extent_buffer(leaf, ptr + size, ptr,
1015 iref = (struct btrfs_extent_inline_ref *)ptr;
1016 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1017 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1018 struct btrfs_extent_data_ref *dref;
1019 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1020 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1021 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1022 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1023 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1024 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1025 struct btrfs_shared_data_ref *sref;
1026 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1027 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1028 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1029 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1030 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1032 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1034 btrfs_mark_buffer_dirty(leaf);
1037 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1038 struct btrfs_path *path,
1039 struct btrfs_extent_inline_ref **ref_ret,
1040 u64 bytenr, u64 num_bytes, u64 parent,
1041 u64 root_objectid, u64 owner, u64 offset)
1045 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1046 num_bytes, parent, root_objectid,
1051 btrfs_release_path(path);
1054 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1055 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1058 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1059 root_objectid, owner, offset);
1065 * helper to update/remove inline back ref
1067 static noinline_for_stack
1068 void update_inline_extent_backref(struct btrfs_path *path,
1069 struct btrfs_extent_inline_ref *iref,
1071 struct btrfs_delayed_extent_op *extent_op,
1074 struct extent_buffer *leaf = path->nodes[0];
1075 struct btrfs_extent_item *ei;
1076 struct btrfs_extent_data_ref *dref = NULL;
1077 struct btrfs_shared_data_ref *sref = NULL;
1085 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1086 refs = btrfs_extent_refs(leaf, ei);
1087 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1088 refs += refs_to_mod;
1089 btrfs_set_extent_refs(leaf, ei, refs);
1091 __run_delayed_extent_op(extent_op, leaf, ei);
1094 * If type is invalid, we should have bailed out after
1095 * lookup_inline_extent_backref().
1097 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1098 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1100 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1101 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1102 refs = btrfs_extent_data_ref_count(leaf, dref);
1103 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1104 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1105 refs = btrfs_shared_data_ref_count(leaf, sref);
1108 BUG_ON(refs_to_mod != -1);
1111 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1112 refs += refs_to_mod;
1115 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1116 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1118 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1121 size = btrfs_extent_inline_ref_size(type);
1122 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1123 ptr = (unsigned long)iref;
1124 end = (unsigned long)ei + item_size;
1125 if (ptr + size < end)
1126 memmove_extent_buffer(leaf, ptr, ptr + size,
1129 btrfs_truncate_item(path, item_size, 1);
1131 btrfs_mark_buffer_dirty(leaf);
1134 static noinline_for_stack
1135 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1136 struct btrfs_path *path,
1137 u64 bytenr, u64 num_bytes, u64 parent,
1138 u64 root_objectid, u64 owner,
1139 u64 offset, int refs_to_add,
1140 struct btrfs_delayed_extent_op *extent_op)
1142 struct btrfs_extent_inline_ref *iref;
1145 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1146 num_bytes, parent, root_objectid,
1150 * We're adding refs to a tree block we already own, this
1151 * should not happen at all.
1153 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1154 btrfs_crit(trans->fs_info,
1155 "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu",
1156 bytenr, num_bytes, root_objectid);
1157 if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) {
1159 btrfs_crit(trans->fs_info,
1160 "path->slots[0]=%d path->nodes[0]:", path->slots[0]);
1161 btrfs_print_leaf(path->nodes[0]);
1165 update_inline_extent_backref(path, iref, refs_to_add,
1167 } else if (ret == -ENOENT) {
1168 setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1169 root_objectid, owner, offset,
1170 refs_to_add, extent_op);
1176 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1177 struct btrfs_path *path,
1178 struct btrfs_extent_inline_ref *iref,
1179 int refs_to_drop, int is_data, int *last_ref)
1183 BUG_ON(!is_data && refs_to_drop != 1);
1185 update_inline_extent_backref(path, iref, -refs_to_drop, NULL,
1187 } else if (is_data) {
1188 ret = remove_extent_data_ref(trans, path, refs_to_drop,
1192 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1197 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1198 u64 *discarded_bytes)
1201 u64 bytes_left, end;
1202 u64 aligned_start = ALIGN(start, 1 << 9);
1204 if (WARN_ON(start != aligned_start)) {
1205 len -= aligned_start - start;
1206 len = round_down(len, 1 << 9);
1207 start = aligned_start;
1210 *discarded_bytes = 0;
1218 /* Skip any superblocks on this device. */
1219 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1220 u64 sb_start = btrfs_sb_offset(j);
1221 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1222 u64 size = sb_start - start;
1224 if (!in_range(sb_start, start, bytes_left) &&
1225 !in_range(sb_end, start, bytes_left) &&
1226 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1230 * Superblock spans beginning of range. Adjust start and
1233 if (sb_start <= start) {
1234 start += sb_end - start;
1239 bytes_left = end - start;
1244 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1247 *discarded_bytes += size;
1248 else if (ret != -EOPNOTSUPP)
1257 bytes_left = end - start;
1261 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1264 *discarded_bytes += bytes_left;
1269 static int do_discard_extent(struct btrfs_io_stripe *stripe, u64 *bytes)
1271 struct btrfs_device *dev = stripe->dev;
1272 struct btrfs_fs_info *fs_info = dev->fs_info;
1273 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1274 u64 phys = stripe->physical;
1275 u64 len = stripe->length;
1279 /* Zone reset on a zoned filesystem */
1280 if (btrfs_can_zone_reset(dev, phys, len)) {
1283 ret = btrfs_reset_device_zone(dev, phys, len, &discarded);
1287 if (!btrfs_dev_replace_is_ongoing(dev_replace) ||
1288 dev != dev_replace->srcdev)
1291 src_disc = discarded;
1293 /* Send to replace target as well */
1294 ret = btrfs_reset_device_zone(dev_replace->tgtdev, phys, len,
1296 discarded += src_disc;
1297 } else if (blk_queue_discard(bdev_get_queue(stripe->dev->bdev))) {
1298 ret = btrfs_issue_discard(dev->bdev, phys, len, &discarded);
1309 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1310 u64 num_bytes, u64 *actual_bytes)
1313 u64 discarded_bytes = 0;
1314 u64 end = bytenr + num_bytes;
1316 struct btrfs_io_context *bioc = NULL;
1319 * Avoid races with device replace and make sure our bioc has devices
1320 * associated to its stripes that don't go away while we are discarding.
1322 btrfs_bio_counter_inc_blocked(fs_info);
1324 struct btrfs_io_stripe *stripe;
1327 num_bytes = end - cur;
1328 /* Tell the block device(s) that the sectors can be discarded */
1329 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur,
1330 &num_bytes, &bioc, 0);
1332 * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or
1333 * -EOPNOTSUPP. For any such error, @num_bytes is not updated,
1334 * thus we can't continue anyway.
1339 stripe = bioc->stripes;
1340 for (i = 0; i < bioc->num_stripes; i++, stripe++) {
1342 struct btrfs_device *device = stripe->dev;
1344 if (!device->bdev) {
1345 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1349 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
1352 ret = do_discard_extent(stripe, &bytes);
1354 discarded_bytes += bytes;
1355 } else if (ret != -EOPNOTSUPP) {
1357 * Logic errors or -ENOMEM, or -EIO, but
1358 * unlikely to happen.
1360 * And since there are two loops, explicitly
1361 * go to out to avoid confusion.
1363 btrfs_put_bioc(bioc);
1368 * Just in case we get back EOPNOTSUPP for some reason,
1369 * just ignore the return value so we don't screw up
1370 * people calling discard_extent.
1374 btrfs_put_bioc(bioc);
1378 btrfs_bio_counter_dec(fs_info);
1381 *actual_bytes = discarded_bytes;
1384 if (ret == -EOPNOTSUPP)
1389 /* Can return -ENOMEM */
1390 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1391 struct btrfs_ref *generic_ref)
1393 struct btrfs_fs_info *fs_info = trans->fs_info;
1396 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1397 generic_ref->action);
1398 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1399 generic_ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID);
1401 if (generic_ref->type == BTRFS_REF_METADATA)
1402 ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL);
1404 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0);
1406 btrfs_ref_tree_mod(fs_info, generic_ref);
1412 * __btrfs_inc_extent_ref - insert backreference for a given extent
1414 * The counterpart is in __btrfs_free_extent(), with examples and more details
1417 * @trans: Handle of transaction
1419 * @node: The delayed ref node used to get the bytenr/length for
1420 * extent whose references are incremented.
1422 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1423 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1424 * bytenr of the parent block. Since new extents are always
1425 * created with indirect references, this will only be the case
1426 * when relocating a shared extent. In that case, root_objectid
1427 * will be BTRFS_TREE_RELOC_OBJECTID. Otherwise, parent must
1430 * @root_objectid: The id of the root where this modification has originated,
1431 * this can be either one of the well-known metadata trees or
1432 * the subvolume id which references this extent.
1434 * @owner: For data extents it is the inode number of the owning file.
1435 * For metadata extents this parameter holds the level in the
1436 * tree of the extent.
1438 * @offset: For metadata extents the offset is ignored and is currently
1439 * always passed as 0. For data extents it is the fileoffset
1440 * this extent belongs to.
1442 * @refs_to_add Number of references to add
1444 * @extent_op Pointer to a structure, holding information necessary when
1445 * updating a tree block's flags
1448 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1449 struct btrfs_delayed_ref_node *node,
1450 u64 parent, u64 root_objectid,
1451 u64 owner, u64 offset, int refs_to_add,
1452 struct btrfs_delayed_extent_op *extent_op)
1454 struct btrfs_path *path;
1455 struct extent_buffer *leaf;
1456 struct btrfs_extent_item *item;
1457 struct btrfs_key key;
1458 u64 bytenr = node->bytenr;
1459 u64 num_bytes = node->num_bytes;
1463 path = btrfs_alloc_path();
1467 /* this will setup the path even if it fails to insert the back ref */
1468 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1469 parent, root_objectid, owner,
1470 offset, refs_to_add, extent_op);
1471 if ((ret < 0 && ret != -EAGAIN) || !ret)
1475 * Ok we had -EAGAIN which means we didn't have space to insert and
1476 * inline extent ref, so just update the reference count and add a
1479 leaf = path->nodes[0];
1480 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1481 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1482 refs = btrfs_extent_refs(leaf, item);
1483 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1485 __run_delayed_extent_op(extent_op, leaf, item);
1487 btrfs_mark_buffer_dirty(leaf);
1488 btrfs_release_path(path);
1490 /* now insert the actual backref */
1491 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1492 BUG_ON(refs_to_add != 1);
1493 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1496 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1497 root_objectid, owner, offset,
1501 btrfs_abort_transaction(trans, ret);
1503 btrfs_free_path(path);
1507 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1508 struct btrfs_delayed_ref_node *node,
1509 struct btrfs_delayed_extent_op *extent_op,
1510 int insert_reserved)
1513 struct btrfs_delayed_data_ref *ref;
1514 struct btrfs_key ins;
1519 ins.objectid = node->bytenr;
1520 ins.offset = node->num_bytes;
1521 ins.type = BTRFS_EXTENT_ITEM_KEY;
1523 ref = btrfs_delayed_node_to_data_ref(node);
1524 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
1526 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1527 parent = ref->parent;
1528 ref_root = ref->root;
1530 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1532 flags |= extent_op->flags_to_set;
1533 ret = alloc_reserved_file_extent(trans, parent, ref_root,
1534 flags, ref->objectid,
1537 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1538 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1539 ref->objectid, ref->offset,
1540 node->ref_mod, extent_op);
1541 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1542 ret = __btrfs_free_extent(trans, node, parent,
1543 ref_root, ref->objectid,
1544 ref->offset, node->ref_mod,
1552 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1553 struct extent_buffer *leaf,
1554 struct btrfs_extent_item *ei)
1556 u64 flags = btrfs_extent_flags(leaf, ei);
1557 if (extent_op->update_flags) {
1558 flags |= extent_op->flags_to_set;
1559 btrfs_set_extent_flags(leaf, ei, flags);
1562 if (extent_op->update_key) {
1563 struct btrfs_tree_block_info *bi;
1564 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1565 bi = (struct btrfs_tree_block_info *)(ei + 1);
1566 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1570 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1571 struct btrfs_delayed_ref_head *head,
1572 struct btrfs_delayed_extent_op *extent_op)
1574 struct btrfs_fs_info *fs_info = trans->fs_info;
1575 struct btrfs_key key;
1576 struct btrfs_path *path;
1577 struct btrfs_extent_item *ei;
1578 struct extent_buffer *leaf;
1582 int metadata = !extent_op->is_data;
1584 if (TRANS_ABORTED(trans))
1587 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1590 path = btrfs_alloc_path();
1594 key.objectid = head->bytenr;
1597 key.type = BTRFS_METADATA_ITEM_KEY;
1598 key.offset = extent_op->level;
1600 key.type = BTRFS_EXTENT_ITEM_KEY;
1601 key.offset = head->num_bytes;
1605 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
1612 if (path->slots[0] > 0) {
1614 btrfs_item_key_to_cpu(path->nodes[0], &key,
1616 if (key.objectid == head->bytenr &&
1617 key.type == BTRFS_EXTENT_ITEM_KEY &&
1618 key.offset == head->num_bytes)
1622 btrfs_release_path(path);
1625 key.objectid = head->bytenr;
1626 key.offset = head->num_bytes;
1627 key.type = BTRFS_EXTENT_ITEM_KEY;
1636 leaf = path->nodes[0];
1637 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1639 if (unlikely(item_size < sizeof(*ei))) {
1641 btrfs_print_v0_err(fs_info);
1642 btrfs_abort_transaction(trans, err);
1646 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1647 __run_delayed_extent_op(extent_op, leaf, ei);
1649 btrfs_mark_buffer_dirty(leaf);
1651 btrfs_free_path(path);
1655 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1656 struct btrfs_delayed_ref_node *node,
1657 struct btrfs_delayed_extent_op *extent_op,
1658 int insert_reserved)
1661 struct btrfs_delayed_tree_ref *ref;
1665 ref = btrfs_delayed_node_to_tree_ref(node);
1666 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1668 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1669 parent = ref->parent;
1670 ref_root = ref->root;
1672 if (node->ref_mod != 1) {
1673 btrfs_err(trans->fs_info,
1674 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
1675 node->bytenr, node->ref_mod, node->action, ref_root,
1679 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1680 BUG_ON(!extent_op || !extent_op->update_flags);
1681 ret = alloc_reserved_tree_block(trans, node, extent_op);
1682 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1683 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1684 ref->level, 0, 1, extent_op);
1685 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1686 ret = __btrfs_free_extent(trans, node, parent, ref_root,
1687 ref->level, 0, 1, extent_op);
1694 /* helper function to actually process a single delayed ref entry */
1695 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1696 struct btrfs_delayed_ref_node *node,
1697 struct btrfs_delayed_extent_op *extent_op,
1698 int insert_reserved)
1702 if (TRANS_ABORTED(trans)) {
1703 if (insert_reserved)
1704 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1708 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1709 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1710 ret = run_delayed_tree_ref(trans, node, extent_op,
1712 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1713 node->type == BTRFS_SHARED_DATA_REF_KEY)
1714 ret = run_delayed_data_ref(trans, node, extent_op,
1718 if (ret && insert_reserved)
1719 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1723 static inline struct btrfs_delayed_ref_node *
1724 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1726 struct btrfs_delayed_ref_node *ref;
1728 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1732 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1733 * This is to prevent a ref count from going down to zero, which deletes
1734 * the extent item from the extent tree, when there still are references
1735 * to add, which would fail because they would not find the extent item.
1737 if (!list_empty(&head->ref_add_list))
1738 return list_first_entry(&head->ref_add_list,
1739 struct btrfs_delayed_ref_node, add_list);
1741 ref = rb_entry(rb_first_cached(&head->ref_tree),
1742 struct btrfs_delayed_ref_node, ref_node);
1743 ASSERT(list_empty(&ref->add_list));
1747 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1748 struct btrfs_delayed_ref_head *head)
1750 spin_lock(&delayed_refs->lock);
1751 head->processing = 0;
1752 delayed_refs->num_heads_ready++;
1753 spin_unlock(&delayed_refs->lock);
1754 btrfs_delayed_ref_unlock(head);
1757 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1758 struct btrfs_delayed_ref_head *head)
1760 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1765 if (head->must_insert_reserved) {
1766 head->extent_op = NULL;
1767 btrfs_free_delayed_extent_op(extent_op);
1773 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1774 struct btrfs_delayed_ref_head *head)
1776 struct btrfs_delayed_extent_op *extent_op;
1779 extent_op = cleanup_extent_op(head);
1782 head->extent_op = NULL;
1783 spin_unlock(&head->lock);
1784 ret = run_delayed_extent_op(trans, head, extent_op);
1785 btrfs_free_delayed_extent_op(extent_op);
1786 return ret ? ret : 1;
1789 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1790 struct btrfs_delayed_ref_root *delayed_refs,
1791 struct btrfs_delayed_ref_head *head)
1793 int nr_items = 1; /* Dropping this ref head update. */
1796 * We had csum deletions accounted for in our delayed refs rsv, we need
1797 * to drop the csum leaves for this update from our delayed_refs_rsv.
1799 if (head->total_ref_mod < 0 && head->is_data) {
1800 spin_lock(&delayed_refs->lock);
1801 delayed_refs->pending_csums -= head->num_bytes;
1802 spin_unlock(&delayed_refs->lock);
1803 nr_items += btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes);
1806 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
1809 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1810 struct btrfs_delayed_ref_head *head)
1813 struct btrfs_fs_info *fs_info = trans->fs_info;
1814 struct btrfs_delayed_ref_root *delayed_refs;
1817 delayed_refs = &trans->transaction->delayed_refs;
1819 ret = run_and_cleanup_extent_op(trans, head);
1821 unselect_delayed_ref_head(delayed_refs, head);
1822 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1829 * Need to drop our head ref lock and re-acquire the delayed ref lock
1830 * and then re-check to make sure nobody got added.
1832 spin_unlock(&head->lock);
1833 spin_lock(&delayed_refs->lock);
1834 spin_lock(&head->lock);
1835 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1836 spin_unlock(&head->lock);
1837 spin_unlock(&delayed_refs->lock);
1840 btrfs_delete_ref_head(delayed_refs, head);
1841 spin_unlock(&head->lock);
1842 spin_unlock(&delayed_refs->lock);
1844 if (head->must_insert_reserved) {
1845 btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
1846 if (head->is_data) {
1847 ret = btrfs_del_csums(trans, fs_info->csum_root,
1848 head->bytenr, head->num_bytes);
1852 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1854 trace_run_delayed_ref_head(fs_info, head, 0);
1855 btrfs_delayed_ref_unlock(head);
1856 btrfs_put_delayed_ref_head(head);
1860 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1861 struct btrfs_trans_handle *trans)
1863 struct btrfs_delayed_ref_root *delayed_refs =
1864 &trans->transaction->delayed_refs;
1865 struct btrfs_delayed_ref_head *head = NULL;
1868 spin_lock(&delayed_refs->lock);
1869 head = btrfs_select_ref_head(delayed_refs);
1871 spin_unlock(&delayed_refs->lock);
1876 * Grab the lock that says we are going to process all the refs for
1879 ret = btrfs_delayed_ref_lock(delayed_refs, head);
1880 spin_unlock(&delayed_refs->lock);
1883 * We may have dropped the spin lock to get the head mutex lock, and
1884 * that might have given someone else time to free the head. If that's
1885 * true, it has been removed from our list and we can move on.
1888 head = ERR_PTR(-EAGAIN);
1893 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1894 struct btrfs_delayed_ref_head *locked_ref,
1895 unsigned long *run_refs)
1897 struct btrfs_fs_info *fs_info = trans->fs_info;
1898 struct btrfs_delayed_ref_root *delayed_refs;
1899 struct btrfs_delayed_extent_op *extent_op;
1900 struct btrfs_delayed_ref_node *ref;
1901 int must_insert_reserved = 0;
1904 delayed_refs = &trans->transaction->delayed_refs;
1906 lockdep_assert_held(&locked_ref->mutex);
1907 lockdep_assert_held(&locked_ref->lock);
1909 while ((ref = select_delayed_ref(locked_ref))) {
1911 btrfs_check_delayed_seq(fs_info, ref->seq)) {
1912 spin_unlock(&locked_ref->lock);
1913 unselect_delayed_ref_head(delayed_refs, locked_ref);
1919 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1920 RB_CLEAR_NODE(&ref->ref_node);
1921 if (!list_empty(&ref->add_list))
1922 list_del(&ref->add_list);
1924 * When we play the delayed ref, also correct the ref_mod on
1927 switch (ref->action) {
1928 case BTRFS_ADD_DELAYED_REF:
1929 case BTRFS_ADD_DELAYED_EXTENT:
1930 locked_ref->ref_mod -= ref->ref_mod;
1932 case BTRFS_DROP_DELAYED_REF:
1933 locked_ref->ref_mod += ref->ref_mod;
1938 atomic_dec(&delayed_refs->num_entries);
1941 * Record the must_insert_reserved flag before we drop the
1944 must_insert_reserved = locked_ref->must_insert_reserved;
1945 locked_ref->must_insert_reserved = 0;
1947 extent_op = locked_ref->extent_op;
1948 locked_ref->extent_op = NULL;
1949 spin_unlock(&locked_ref->lock);
1951 ret = run_one_delayed_ref(trans, ref, extent_op,
1952 must_insert_reserved);
1954 btrfs_free_delayed_extent_op(extent_op);
1956 unselect_delayed_ref_head(delayed_refs, locked_ref);
1957 btrfs_put_delayed_ref(ref);
1958 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
1963 btrfs_put_delayed_ref(ref);
1966 spin_lock(&locked_ref->lock);
1967 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
1974 * Returns 0 on success or if called with an already aborted transaction.
1975 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
1977 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1980 struct btrfs_fs_info *fs_info = trans->fs_info;
1981 struct btrfs_delayed_ref_root *delayed_refs;
1982 struct btrfs_delayed_ref_head *locked_ref = NULL;
1983 ktime_t start = ktime_get();
1985 unsigned long count = 0;
1986 unsigned long actual_count = 0;
1988 delayed_refs = &trans->transaction->delayed_refs;
1991 locked_ref = btrfs_obtain_ref_head(trans);
1992 if (IS_ERR_OR_NULL(locked_ref)) {
1993 if (PTR_ERR(locked_ref) == -EAGAIN) {
2002 * We need to try and merge add/drops of the same ref since we
2003 * can run into issues with relocate dropping the implicit ref
2004 * and then it being added back again before the drop can
2005 * finish. If we merged anything we need to re-loop so we can
2007 * Or we can get node references of the same type that weren't
2008 * merged when created due to bumps in the tree mod seq, and
2009 * we need to merge them to prevent adding an inline extent
2010 * backref before dropping it (triggering a BUG_ON at
2011 * insert_inline_extent_backref()).
2013 spin_lock(&locked_ref->lock);
2014 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2016 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2018 if (ret < 0 && ret != -EAGAIN) {
2020 * Error, btrfs_run_delayed_refs_for_head already
2021 * unlocked everything so just bail out
2026 * Success, perform the usual cleanup of a processed
2029 ret = cleanup_ref_head(trans, locked_ref);
2031 /* We dropped our lock, we need to loop. */
2040 * Either success case or btrfs_run_delayed_refs_for_head
2041 * returned -EAGAIN, meaning we need to select another head
2046 } while ((nr != -1 && count < nr) || locked_ref);
2049 * We don't want to include ref heads since we can have empty ref heads
2050 * and those will drastically skew our runtime down since we just do
2051 * accounting, no actual extent tree updates.
2053 if (actual_count > 0) {
2054 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2058 * We weigh the current average higher than our current runtime
2059 * to avoid large swings in the average.
2061 spin_lock(&delayed_refs->lock);
2062 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2063 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2064 spin_unlock(&delayed_refs->lock);
2069 #ifdef SCRAMBLE_DELAYED_REFS
2071 * Normally delayed refs get processed in ascending bytenr order. This
2072 * correlates in most cases to the order added. To expose dependencies on this
2073 * order, we start to process the tree in the middle instead of the beginning
2075 static u64 find_middle(struct rb_root *root)
2077 struct rb_node *n = root->rb_node;
2078 struct btrfs_delayed_ref_node *entry;
2081 u64 first = 0, last = 0;
2085 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2086 first = entry->bytenr;
2090 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2091 last = entry->bytenr;
2096 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2097 WARN_ON(!entry->in_tree);
2099 middle = entry->bytenr;
2113 * this starts processing the delayed reference count updates and
2114 * extent insertions we have queued up so far. count can be
2115 * 0, which means to process everything in the tree at the start
2116 * of the run (but not newly added entries), or it can be some target
2117 * number you'd like to process.
2119 * Returns 0 on success or if called with an aborted transaction
2120 * Returns <0 on error and aborts the transaction
2122 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2123 unsigned long count)
2125 struct btrfs_fs_info *fs_info = trans->fs_info;
2126 struct rb_node *node;
2127 struct btrfs_delayed_ref_root *delayed_refs;
2128 struct btrfs_delayed_ref_head *head;
2130 int run_all = count == (unsigned long)-1;
2132 /* We'll clean this up in btrfs_cleanup_transaction */
2133 if (TRANS_ABORTED(trans))
2136 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2139 delayed_refs = &trans->transaction->delayed_refs;
2141 count = delayed_refs->num_heads_ready;
2144 #ifdef SCRAMBLE_DELAYED_REFS
2145 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2147 ret = __btrfs_run_delayed_refs(trans, count);
2149 btrfs_abort_transaction(trans, ret);
2154 btrfs_create_pending_block_groups(trans);
2156 spin_lock(&delayed_refs->lock);
2157 node = rb_first_cached(&delayed_refs->href_root);
2159 spin_unlock(&delayed_refs->lock);
2162 head = rb_entry(node, struct btrfs_delayed_ref_head,
2164 refcount_inc(&head->refs);
2165 spin_unlock(&delayed_refs->lock);
2167 /* Mutex was contended, block until it's released and retry. */
2168 mutex_lock(&head->mutex);
2169 mutex_unlock(&head->mutex);
2171 btrfs_put_delayed_ref_head(head);
2179 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2180 struct extent_buffer *eb, u64 flags,
2181 int level, int is_data)
2183 struct btrfs_delayed_extent_op *extent_op;
2186 extent_op = btrfs_alloc_delayed_extent_op();
2190 extent_op->flags_to_set = flags;
2191 extent_op->update_flags = true;
2192 extent_op->update_key = false;
2193 extent_op->is_data = is_data ? true : false;
2194 extent_op->level = level;
2196 ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op);
2198 btrfs_free_delayed_extent_op(extent_op);
2202 static noinline int check_delayed_ref(struct btrfs_root *root,
2203 struct btrfs_path *path,
2204 u64 objectid, u64 offset, u64 bytenr)
2206 struct btrfs_delayed_ref_head *head;
2207 struct btrfs_delayed_ref_node *ref;
2208 struct btrfs_delayed_data_ref *data_ref;
2209 struct btrfs_delayed_ref_root *delayed_refs;
2210 struct btrfs_transaction *cur_trans;
2211 struct rb_node *node;
2214 spin_lock(&root->fs_info->trans_lock);
2215 cur_trans = root->fs_info->running_transaction;
2217 refcount_inc(&cur_trans->use_count);
2218 spin_unlock(&root->fs_info->trans_lock);
2222 delayed_refs = &cur_trans->delayed_refs;
2223 spin_lock(&delayed_refs->lock);
2224 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2226 spin_unlock(&delayed_refs->lock);
2227 btrfs_put_transaction(cur_trans);
2231 if (!mutex_trylock(&head->mutex)) {
2232 refcount_inc(&head->refs);
2233 spin_unlock(&delayed_refs->lock);
2235 btrfs_release_path(path);
2238 * Mutex was contended, block until it's released and let
2241 mutex_lock(&head->mutex);
2242 mutex_unlock(&head->mutex);
2243 btrfs_put_delayed_ref_head(head);
2244 btrfs_put_transaction(cur_trans);
2247 spin_unlock(&delayed_refs->lock);
2249 spin_lock(&head->lock);
2251 * XXX: We should replace this with a proper search function in the
2254 for (node = rb_first_cached(&head->ref_tree); node;
2255 node = rb_next(node)) {
2256 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2257 /* If it's a shared ref we know a cross reference exists */
2258 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2263 data_ref = btrfs_delayed_node_to_data_ref(ref);
2266 * If our ref doesn't match the one we're currently looking at
2267 * then we have a cross reference.
2269 if (data_ref->root != root->root_key.objectid ||
2270 data_ref->objectid != objectid ||
2271 data_ref->offset != offset) {
2276 spin_unlock(&head->lock);
2277 mutex_unlock(&head->mutex);
2278 btrfs_put_transaction(cur_trans);
2282 static noinline int check_committed_ref(struct btrfs_root *root,
2283 struct btrfs_path *path,
2284 u64 objectid, u64 offset, u64 bytenr,
2287 struct btrfs_fs_info *fs_info = root->fs_info;
2288 struct btrfs_root *extent_root = fs_info->extent_root;
2289 struct extent_buffer *leaf;
2290 struct btrfs_extent_data_ref *ref;
2291 struct btrfs_extent_inline_ref *iref;
2292 struct btrfs_extent_item *ei;
2293 struct btrfs_key key;
2298 key.objectid = bytenr;
2299 key.offset = (u64)-1;
2300 key.type = BTRFS_EXTENT_ITEM_KEY;
2302 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2305 BUG_ON(ret == 0); /* Corruption */
2308 if (path->slots[0] == 0)
2312 leaf = path->nodes[0];
2313 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2315 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2319 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2320 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2322 /* If extent item has more than 1 inline ref then it's shared */
2323 if (item_size != sizeof(*ei) +
2324 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2328 * If extent created before last snapshot => it's shared unless the
2329 * snapshot has been deleted. Use the heuristic if strict is false.
2332 (btrfs_extent_generation(leaf, ei) <=
2333 btrfs_root_last_snapshot(&root->root_item)))
2336 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2338 /* If this extent has SHARED_DATA_REF then it's shared */
2339 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2340 if (type != BTRFS_EXTENT_DATA_REF_KEY)
2343 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2344 if (btrfs_extent_refs(leaf, ei) !=
2345 btrfs_extent_data_ref_count(leaf, ref) ||
2346 btrfs_extent_data_ref_root(leaf, ref) !=
2347 root->root_key.objectid ||
2348 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2349 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2357 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2358 u64 bytenr, bool strict)
2360 struct btrfs_path *path;
2363 path = btrfs_alloc_path();
2368 ret = check_committed_ref(root, path, objectid,
2369 offset, bytenr, strict);
2370 if (ret && ret != -ENOENT)
2373 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2374 } while (ret == -EAGAIN);
2377 btrfs_free_path(path);
2378 if (btrfs_is_data_reloc_root(root))
2383 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2384 struct btrfs_root *root,
2385 struct extent_buffer *buf,
2386 int full_backref, int inc)
2388 struct btrfs_fs_info *fs_info = root->fs_info;
2394 struct btrfs_key key;
2395 struct btrfs_file_extent_item *fi;
2396 struct btrfs_ref generic_ref = { 0 };
2397 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2403 if (btrfs_is_testing(fs_info))
2406 ref_root = btrfs_header_owner(buf);
2407 nritems = btrfs_header_nritems(buf);
2408 level = btrfs_header_level(buf);
2410 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
2414 parent = buf->start;
2418 action = BTRFS_ADD_DELAYED_REF;
2420 action = BTRFS_DROP_DELAYED_REF;
2422 for (i = 0; i < nritems; i++) {
2424 btrfs_item_key_to_cpu(buf, &key, i);
2425 if (key.type != BTRFS_EXTENT_DATA_KEY)
2427 fi = btrfs_item_ptr(buf, i,
2428 struct btrfs_file_extent_item);
2429 if (btrfs_file_extent_type(buf, fi) ==
2430 BTRFS_FILE_EXTENT_INLINE)
2432 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2436 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2437 key.offset -= btrfs_file_extent_offset(buf, fi);
2438 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2440 generic_ref.real_root = root->root_key.objectid;
2441 btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2443 generic_ref.skip_qgroup = for_reloc;
2445 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2447 ret = btrfs_free_extent(trans, &generic_ref);
2451 bytenr = btrfs_node_blockptr(buf, i);
2452 num_bytes = fs_info->nodesize;
2453 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2455 generic_ref.real_root = root->root_key.objectid;
2456 btrfs_init_tree_ref(&generic_ref, level - 1, ref_root);
2457 generic_ref.skip_qgroup = for_reloc;
2459 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2461 ret = btrfs_free_extent(trans, &generic_ref);
2471 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2472 struct extent_buffer *buf, int full_backref)
2474 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2477 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2478 struct extent_buffer *buf, int full_backref)
2480 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2483 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2485 struct btrfs_fs_info *fs_info = root->fs_info;
2490 flags = BTRFS_BLOCK_GROUP_DATA;
2491 else if (root == fs_info->chunk_root)
2492 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2494 flags = BTRFS_BLOCK_GROUP_METADATA;
2496 ret = btrfs_get_alloc_profile(fs_info, flags);
2500 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
2502 struct btrfs_block_group *cache;
2505 spin_lock(&fs_info->block_group_cache_lock);
2506 bytenr = fs_info->first_logical_byte;
2507 spin_unlock(&fs_info->block_group_cache_lock);
2509 if (bytenr < (u64)-1)
2512 cache = btrfs_lookup_first_block_group(fs_info, search_start);
2516 bytenr = cache->start;
2517 btrfs_put_block_group(cache);
2522 static int pin_down_extent(struct btrfs_trans_handle *trans,
2523 struct btrfs_block_group *cache,
2524 u64 bytenr, u64 num_bytes, int reserved)
2526 struct btrfs_fs_info *fs_info = cache->fs_info;
2528 spin_lock(&cache->space_info->lock);
2529 spin_lock(&cache->lock);
2530 cache->pinned += num_bytes;
2531 btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2534 cache->reserved -= num_bytes;
2535 cache->space_info->bytes_reserved -= num_bytes;
2537 spin_unlock(&cache->lock);
2538 spin_unlock(&cache->space_info->lock);
2540 set_extent_dirty(&trans->transaction->pinned_extents, bytenr,
2541 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
2545 int btrfs_pin_extent(struct btrfs_trans_handle *trans,
2546 u64 bytenr, u64 num_bytes, int reserved)
2548 struct btrfs_block_group *cache;
2550 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2551 BUG_ON(!cache); /* Logic error */
2553 pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
2555 btrfs_put_block_group(cache);
2560 * this function must be called within transaction
2562 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
2563 u64 bytenr, u64 num_bytes)
2565 struct btrfs_block_group *cache;
2568 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2573 * pull in the free space cache (if any) so that our pin
2574 * removes the free space from the cache. We have load_only set
2575 * to one because the slow code to read in the free extents does check
2576 * the pinned extents.
2578 btrfs_cache_block_group(cache, 1);
2580 * Make sure we wait until the cache is completely built in case it is
2581 * missing or is invalid and therefore needs to be rebuilt.
2583 ret = btrfs_wait_block_group_cache_done(cache);
2587 pin_down_extent(trans, cache, bytenr, num_bytes, 0);
2589 /* remove us from the free space cache (if we're there at all) */
2590 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
2592 btrfs_put_block_group(cache);
2596 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2597 u64 start, u64 num_bytes)
2600 struct btrfs_block_group *block_group;
2602 block_group = btrfs_lookup_block_group(fs_info, start);
2606 btrfs_cache_block_group(block_group, 1);
2608 * Make sure we wait until the cache is completely built in case it is
2609 * missing or is invalid and therefore needs to be rebuilt.
2611 ret = btrfs_wait_block_group_cache_done(block_group);
2615 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2617 btrfs_put_block_group(block_group);
2621 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2623 struct btrfs_fs_info *fs_info = eb->fs_info;
2624 struct btrfs_file_extent_item *item;
2625 struct btrfs_key key;
2630 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2633 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2634 btrfs_item_key_to_cpu(eb, &key, i);
2635 if (key.type != BTRFS_EXTENT_DATA_KEY)
2637 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2638 found_type = btrfs_file_extent_type(eb, item);
2639 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2641 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2643 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2644 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2645 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2654 btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2656 atomic_inc(&bg->reservations);
2660 * Returns the free cluster for the given space info and sets empty_cluster to
2661 * what it should be based on the mount options.
2663 static struct btrfs_free_cluster *
2664 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2665 struct btrfs_space_info *space_info, u64 *empty_cluster)
2667 struct btrfs_free_cluster *ret = NULL;
2670 if (btrfs_mixed_space_info(space_info))
2673 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2674 ret = &fs_info->meta_alloc_cluster;
2675 if (btrfs_test_opt(fs_info, SSD))
2676 *empty_cluster = SZ_2M;
2678 *empty_cluster = SZ_64K;
2679 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2680 btrfs_test_opt(fs_info, SSD_SPREAD)) {
2681 *empty_cluster = SZ_2M;
2682 ret = &fs_info->data_alloc_cluster;
2688 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2690 const bool return_free_space)
2692 struct btrfs_block_group *cache = NULL;
2693 struct btrfs_space_info *space_info;
2694 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2695 struct btrfs_free_cluster *cluster = NULL;
2697 u64 total_unpinned = 0;
2698 u64 empty_cluster = 0;
2701 while (start <= end) {
2704 start >= cache->start + cache->length) {
2706 btrfs_put_block_group(cache);
2708 cache = btrfs_lookup_block_group(fs_info, start);
2709 BUG_ON(!cache); /* Logic error */
2711 cluster = fetch_cluster_info(fs_info,
2714 empty_cluster <<= 1;
2717 len = cache->start + cache->length - start;
2718 len = min(len, end + 1 - start);
2720 down_read(&fs_info->commit_root_sem);
2721 if (start < cache->last_byte_to_unpin && return_free_space) {
2722 u64 add_len = min(len, cache->last_byte_to_unpin - start);
2724 btrfs_add_free_space(cache, start, add_len);
2726 up_read(&fs_info->commit_root_sem);
2729 total_unpinned += len;
2730 space_info = cache->space_info;
2733 * If this space cluster has been marked as fragmented and we've
2734 * unpinned enough in this block group to potentially allow a
2735 * cluster to be created inside of it go ahead and clear the
2738 if (cluster && cluster->fragmented &&
2739 total_unpinned > empty_cluster) {
2740 spin_lock(&cluster->lock);
2741 cluster->fragmented = 0;
2742 spin_unlock(&cluster->lock);
2745 spin_lock(&space_info->lock);
2746 spin_lock(&cache->lock);
2747 cache->pinned -= len;
2748 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2749 space_info->max_extent_size = 0;
2751 space_info->bytes_readonly += len;
2753 } else if (btrfs_is_zoned(fs_info)) {
2754 /* Need reset before reusing in a zoned block group */
2755 space_info->bytes_zone_unusable += len;
2758 spin_unlock(&cache->lock);
2759 if (!readonly && return_free_space &&
2760 global_rsv->space_info == space_info) {
2763 spin_lock(&global_rsv->lock);
2764 if (!global_rsv->full) {
2765 to_add = min(len, global_rsv->size -
2766 global_rsv->reserved);
2767 global_rsv->reserved += to_add;
2768 btrfs_space_info_update_bytes_may_use(fs_info,
2769 space_info, to_add);
2770 if (global_rsv->reserved >= global_rsv->size)
2771 global_rsv->full = 1;
2774 spin_unlock(&global_rsv->lock);
2776 /* Add to any tickets we may have */
2777 if (!readonly && return_free_space && len)
2778 btrfs_try_granting_tickets(fs_info, space_info);
2779 spin_unlock(&space_info->lock);
2783 btrfs_put_block_group(cache);
2787 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2789 struct btrfs_fs_info *fs_info = trans->fs_info;
2790 struct btrfs_block_group *block_group, *tmp;
2791 struct list_head *deleted_bgs;
2792 struct extent_io_tree *unpin;
2797 unpin = &trans->transaction->pinned_extents;
2799 while (!TRANS_ABORTED(trans)) {
2800 struct extent_state *cached_state = NULL;
2802 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2803 ret = find_first_extent_bit(unpin, 0, &start, &end,
2804 EXTENT_DIRTY, &cached_state);
2806 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2810 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2811 ret = btrfs_discard_extent(fs_info, start,
2812 end + 1 - start, NULL);
2814 clear_extent_dirty(unpin, start, end, &cached_state);
2815 unpin_extent_range(fs_info, start, end, true);
2816 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2817 free_extent_state(cached_state);
2821 if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2822 btrfs_discard_calc_delay(&fs_info->discard_ctl);
2823 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2827 * Transaction is finished. We don't need the lock anymore. We
2828 * do need to clean up the block groups in case of a transaction
2831 deleted_bgs = &trans->transaction->deleted_bgs;
2832 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2836 if (!TRANS_ABORTED(trans))
2837 ret = btrfs_discard_extent(fs_info,
2839 block_group->length,
2842 list_del_init(&block_group->bg_list);
2843 btrfs_unfreeze_block_group(block_group);
2844 btrfs_put_block_group(block_group);
2847 const char *errstr = btrfs_decode_error(ret);
2849 "discard failed while removing blockgroup: errno=%d %s",
2858 * Drop one or more refs of @node.
2860 * 1. Locate the extent refs.
2861 * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
2862 * Locate it, then reduce the refs number or remove the ref line completely.
2864 * 2. Update the refs count in EXTENT/METADATA_ITEM
2866 * Inline backref case:
2868 * in extent tree we have:
2870 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2871 * refs 2 gen 6 flags DATA
2872 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2873 * extent data backref root FS_TREE objectid 257 offset 0 count 1
2875 * This function gets called with:
2877 * node->bytenr = 13631488
2878 * node->num_bytes = 1048576
2879 * root_objectid = FS_TREE
2880 * owner_objectid = 257
2884 * Then we should get some like:
2886 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2887 * refs 1 gen 6 flags DATA
2888 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2890 * Keyed backref case:
2892 * in extent tree we have:
2894 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2895 * refs 754 gen 6 flags DATA
2897 * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
2898 * extent data backref root FS_TREE objectid 866 offset 0 count 1
2900 * This function get called with:
2902 * node->bytenr = 13631488
2903 * node->num_bytes = 1048576
2904 * root_objectid = FS_TREE
2905 * owner_objectid = 866
2909 * Then we should get some like:
2911 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2912 * refs 753 gen 6 flags DATA
2914 * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
2916 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2917 struct btrfs_delayed_ref_node *node, u64 parent,
2918 u64 root_objectid, u64 owner_objectid,
2919 u64 owner_offset, int refs_to_drop,
2920 struct btrfs_delayed_extent_op *extent_op)
2922 struct btrfs_fs_info *info = trans->fs_info;
2923 struct btrfs_key key;
2924 struct btrfs_path *path;
2925 struct btrfs_root *extent_root = info->extent_root;
2926 struct extent_buffer *leaf;
2927 struct btrfs_extent_item *ei;
2928 struct btrfs_extent_inline_ref *iref;
2931 int extent_slot = 0;
2932 int found_extent = 0;
2936 u64 bytenr = node->bytenr;
2937 u64 num_bytes = node->num_bytes;
2939 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
2941 path = btrfs_alloc_path();
2945 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2947 if (!is_data && refs_to_drop != 1) {
2949 "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
2950 node->bytenr, refs_to_drop);
2952 btrfs_abort_transaction(trans, ret);
2957 skinny_metadata = false;
2959 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
2960 parent, root_objectid, owner_objectid,
2964 * Either the inline backref or the SHARED_DATA_REF/
2965 * SHARED_BLOCK_REF is found
2967 * Here is a quick path to locate EXTENT/METADATA_ITEM.
2968 * It's possible the EXTENT/METADATA_ITEM is near current slot.
2970 extent_slot = path->slots[0];
2971 while (extent_slot >= 0) {
2972 btrfs_item_key_to_cpu(path->nodes[0], &key,
2974 if (key.objectid != bytenr)
2976 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2977 key.offset == num_bytes) {
2981 if (key.type == BTRFS_METADATA_ITEM_KEY &&
2982 key.offset == owner_objectid) {
2987 /* Quick path didn't find the EXTEMT/METADATA_ITEM */
2988 if (path->slots[0] - extent_slot > 5)
2993 if (!found_extent) {
2996 "invalid iref, no EXTENT/METADATA_ITEM found but has inline extent ref");
2997 btrfs_abort_transaction(trans, -EUCLEAN);
3000 /* Must be SHARED_* item, remove the backref first */
3001 ret = remove_extent_backref(trans, path, NULL,
3003 is_data, &last_ref);
3005 btrfs_abort_transaction(trans, ret);
3008 btrfs_release_path(path);
3010 /* Slow path to locate EXTENT/METADATA_ITEM */
3011 key.objectid = bytenr;
3012 key.type = BTRFS_EXTENT_ITEM_KEY;
3013 key.offset = num_bytes;
3015 if (!is_data && skinny_metadata) {
3016 key.type = BTRFS_METADATA_ITEM_KEY;
3017 key.offset = owner_objectid;
3020 ret = btrfs_search_slot(trans, extent_root,
3022 if (ret > 0 && skinny_metadata && path->slots[0]) {
3024 * Couldn't find our skinny metadata item,
3025 * see if we have ye olde extent item.
3028 btrfs_item_key_to_cpu(path->nodes[0], &key,
3030 if (key.objectid == bytenr &&
3031 key.type == BTRFS_EXTENT_ITEM_KEY &&
3032 key.offset == num_bytes)
3036 if (ret > 0 && skinny_metadata) {
3037 skinny_metadata = false;
3038 key.objectid = bytenr;
3039 key.type = BTRFS_EXTENT_ITEM_KEY;
3040 key.offset = num_bytes;
3041 btrfs_release_path(path);
3042 ret = btrfs_search_slot(trans, extent_root,
3048 "umm, got %d back from search, was looking for %llu",
3051 btrfs_print_leaf(path->nodes[0]);
3054 btrfs_abort_transaction(trans, ret);
3057 extent_slot = path->slots[0];
3059 } else if (WARN_ON(ret == -ENOENT)) {
3060 btrfs_print_leaf(path->nodes[0]);
3062 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
3063 bytenr, parent, root_objectid, owner_objectid,
3065 btrfs_abort_transaction(trans, ret);
3068 btrfs_abort_transaction(trans, ret);
3072 leaf = path->nodes[0];
3073 item_size = btrfs_item_size_nr(leaf, extent_slot);
3074 if (unlikely(item_size < sizeof(*ei))) {
3076 btrfs_print_v0_err(info);
3077 btrfs_abort_transaction(trans, ret);
3080 ei = btrfs_item_ptr(leaf, extent_slot,
3081 struct btrfs_extent_item);
3082 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3083 key.type == BTRFS_EXTENT_ITEM_KEY) {
3084 struct btrfs_tree_block_info *bi;
3085 if (item_size < sizeof(*ei) + sizeof(*bi)) {
3087 "invalid extent item size for key (%llu, %u, %llu) owner %llu, has %u expect >= %zu",
3088 key.objectid, key.type, key.offset,
3089 owner_objectid, item_size,
3090 sizeof(*ei) + sizeof(*bi));
3091 btrfs_abort_transaction(trans, -EUCLEAN);
3094 bi = (struct btrfs_tree_block_info *)(ei + 1);
3095 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3098 refs = btrfs_extent_refs(leaf, ei);
3099 if (refs < refs_to_drop) {
3101 "trying to drop %d refs but we only have %llu for bytenr %llu",
3102 refs_to_drop, refs, bytenr);
3103 btrfs_abort_transaction(trans, -EUCLEAN);
3106 refs -= refs_to_drop;
3110 __run_delayed_extent_op(extent_op, leaf, ei);
3112 * In the case of inline back ref, reference count will
3113 * be updated by remove_extent_backref
3116 if (!found_extent) {
3118 "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found");
3119 btrfs_abort_transaction(trans, -EUCLEAN);
3123 btrfs_set_extent_refs(leaf, ei, refs);
3124 btrfs_mark_buffer_dirty(leaf);
3127 ret = remove_extent_backref(trans, path, iref,
3128 refs_to_drop, is_data,
3131 btrfs_abort_transaction(trans, ret);
3136 /* In this branch refs == 1 */
3138 if (is_data && refs_to_drop !=
3139 extent_data_ref_count(path, iref)) {
3141 "invalid refs_to_drop, current refs %u refs_to_drop %u",
3142 extent_data_ref_count(path, iref),
3144 btrfs_abort_transaction(trans, -EUCLEAN);
3148 if (path->slots[0] != extent_slot) {
3150 "invalid iref, extent item key (%llu %u %llu) doesn't have wanted iref",
3151 key.objectid, key.type,
3153 btrfs_abort_transaction(trans, -EUCLEAN);
3158 * No inline ref, we must be at SHARED_* item,
3159 * And it's single ref, it must be:
3160 * | extent_slot ||extent_slot + 1|
3161 * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3163 if (path->slots[0] != extent_slot + 1) {
3165 "invalid SHARED_* item, previous item is not EXTENT/METADATA_ITEM");
3166 btrfs_abort_transaction(trans, -EUCLEAN);
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);
3211 * Leaf dump can take up a lot of log buffer, so we only do full leaf
3212 * dump for debug build.
3214 if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) {
3215 btrfs_crit(info, "path->slots[0]=%d extent_slot=%d",
3216 path->slots[0], extent_slot);
3217 btrfs_print_leaf(path->nodes[0]);
3220 btrfs_free_path(path);
3225 * when we free an block, it is possible (and likely) that we free the last
3226 * delayed ref for that extent as well. This searches the delayed ref tree for
3227 * a given extent, and if there are no other delayed refs to be processed, it
3228 * removes it from the tree.
3230 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3233 struct btrfs_delayed_ref_head *head;
3234 struct btrfs_delayed_ref_root *delayed_refs;
3237 delayed_refs = &trans->transaction->delayed_refs;
3238 spin_lock(&delayed_refs->lock);
3239 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3241 goto out_delayed_unlock;
3243 spin_lock(&head->lock);
3244 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3247 if (cleanup_extent_op(head) != NULL)
3251 * waiting for the lock here would deadlock. If someone else has it
3252 * locked they are already in the process of dropping it anyway
3254 if (!mutex_trylock(&head->mutex))
3257 btrfs_delete_ref_head(delayed_refs, head);
3258 head->processing = 0;
3260 spin_unlock(&head->lock);
3261 spin_unlock(&delayed_refs->lock);
3263 BUG_ON(head->extent_op);
3264 if (head->must_insert_reserved)
3267 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3268 mutex_unlock(&head->mutex);
3269 btrfs_put_delayed_ref_head(head);
3272 spin_unlock(&head->lock);
3275 spin_unlock(&delayed_refs->lock);
3279 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3280 struct btrfs_root *root,
3281 struct extent_buffer *buf,
3282 u64 parent, int last_ref)
3284 struct btrfs_fs_info *fs_info = root->fs_info;
3285 struct btrfs_ref generic_ref = { 0 };
3288 btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3289 buf->start, buf->len, parent);
3290 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3291 root->root_key.objectid);
3293 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
3294 btrfs_ref_tree_mod(fs_info, &generic_ref);
3295 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL);
3296 BUG_ON(ret); /* -ENOMEM */
3299 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
3300 struct btrfs_block_group *cache;
3301 bool must_pin = false;
3303 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
3304 ret = check_ref_cleanup(trans, buf->start);
3306 btrfs_redirty_list_add(trans->transaction, buf);
3311 cache = btrfs_lookup_block_group(fs_info, buf->start);
3313 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3314 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3315 btrfs_put_block_group(cache);
3320 * If this is a leaf and there are tree mod log users, we may
3321 * have recorded mod log operations that point to this leaf.
3322 * So we must make sure no one reuses this leaf's extent before
3323 * mod log operations are applied to a node, otherwise after
3324 * rewinding a node using the mod log operations we get an
3325 * inconsistent btree, as the leaf's extent may now be used as
3326 * a node or leaf for another different btree.
3327 * We are safe from races here because at this point no other
3328 * node or root points to this extent buffer, so if after this
3329 * check a new tree mod log user joins, it will not be able to
3330 * find a node pointing to this leaf and record operations that
3331 * point to this leaf.
3333 if (btrfs_header_level(buf) == 0 &&
3334 test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags))
3337 if (must_pin || btrfs_is_zoned(fs_info)) {
3338 btrfs_redirty_list_add(trans->transaction, buf);
3339 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3340 btrfs_put_block_group(cache);
3344 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3346 btrfs_add_free_space(cache, buf->start, buf->len);
3347 btrfs_free_reserved_bytes(cache, buf->len, 0);
3348 btrfs_put_block_group(cache);
3349 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3354 * Deleting the buffer, clear the corrupt flag since it doesn't
3357 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3361 /* Can return -ENOMEM */
3362 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3364 struct btrfs_fs_info *fs_info = trans->fs_info;
3367 if (btrfs_is_testing(fs_info))
3371 * tree log blocks never actually go into the extent allocation
3372 * tree, just update pinning info and exit early.
3374 if ((ref->type == BTRFS_REF_METADATA &&
3375 ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3376 (ref->type == BTRFS_REF_DATA &&
3377 ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)) {
3378 /* unlocks the pinned mutex */
3379 btrfs_pin_extent(trans, ref->bytenr, ref->len, 1);
3381 } else if (ref->type == BTRFS_REF_METADATA) {
3382 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL);
3384 ret = btrfs_add_delayed_data_ref(trans, ref, 0);
3387 if (!((ref->type == BTRFS_REF_METADATA &&
3388 ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3389 (ref->type == BTRFS_REF_DATA &&
3390 ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)))
3391 btrfs_ref_tree_mod(fs_info, ref);
3396 enum btrfs_loop_type {
3397 LOOP_CACHING_NOWAIT,
3404 btrfs_lock_block_group(struct btrfs_block_group *cache,
3408 down_read(&cache->data_rwsem);
3411 static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3414 btrfs_get_block_group(cache);
3416 down_read(&cache->data_rwsem);
3419 static struct btrfs_block_group *btrfs_lock_cluster(
3420 struct btrfs_block_group *block_group,
3421 struct btrfs_free_cluster *cluster,
3423 __acquires(&cluster->refill_lock)
3425 struct btrfs_block_group *used_bg = NULL;
3427 spin_lock(&cluster->refill_lock);
3429 used_bg = cluster->block_group;
3433 if (used_bg == block_group)
3436 btrfs_get_block_group(used_bg);
3441 if (down_read_trylock(&used_bg->data_rwsem))
3444 spin_unlock(&cluster->refill_lock);
3446 /* We should only have one-level nested. */
3447 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3449 spin_lock(&cluster->refill_lock);
3450 if (used_bg == cluster->block_group)
3453 up_read(&used_bg->data_rwsem);
3454 btrfs_put_block_group(used_bg);
3459 btrfs_release_block_group(struct btrfs_block_group *cache,
3463 up_read(&cache->data_rwsem);
3464 btrfs_put_block_group(cache);
3467 enum btrfs_extent_allocation_policy {
3468 BTRFS_EXTENT_ALLOC_CLUSTERED,
3469 BTRFS_EXTENT_ALLOC_ZONED,
3473 * Structure used internally for find_free_extent() function. Wraps needed
3476 struct find_free_extent_ctl {
3477 /* Basic allocation info */
3485 /* Where to start the search inside the bg */
3488 /* For clustered allocation */
3490 struct btrfs_free_cluster *last_ptr;
3493 bool have_caching_bg;
3494 bool orig_have_caching_bg;
3496 /* Allocation is called for tree-log */
3499 /* Allocation is called for data relocation */
3500 bool for_data_reloc;
3502 /* RAID index, converted from flags */
3506 * Current loop number, check find_free_extent_update_loop() for details
3511 * Whether we're refilling a cluster, if true we need to re-search
3512 * current block group but don't try to refill the cluster again.
3514 bool retry_clustered;
3517 * Whether we're updating free space cache, if true we need to re-search
3518 * current block group but don't try updating free space cache again.
3520 bool retry_unclustered;
3522 /* If current block group is cached */
3525 /* Max contiguous hole found */
3526 u64 max_extent_size;
3528 /* Total free space from free space cache, not always contiguous */
3529 u64 total_free_space;
3534 /* Hint where to start looking for an empty space */
3537 /* Allocation policy */
3538 enum btrfs_extent_allocation_policy policy;
3543 * Helper function for find_free_extent().
3545 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3546 * Return -EAGAIN to inform caller that we need to re-search this block group
3547 * Return >0 to inform caller that we find nothing
3548 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3550 static int find_free_extent_clustered(struct btrfs_block_group *bg,
3551 struct find_free_extent_ctl *ffe_ctl,
3552 struct btrfs_block_group **cluster_bg_ret)
3554 struct btrfs_block_group *cluster_bg;
3555 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3556 u64 aligned_cluster;
3560 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3562 goto refill_cluster;
3563 if (cluster_bg != bg && (cluster_bg->ro ||
3564 !block_group_bits(cluster_bg, ffe_ctl->flags)))
3565 goto release_cluster;
3567 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3568 ffe_ctl->num_bytes, cluster_bg->start,
3569 &ffe_ctl->max_extent_size);
3571 /* We have a block, we're done */
3572 spin_unlock(&last_ptr->refill_lock);
3573 trace_btrfs_reserve_extent_cluster(cluster_bg,
3574 ffe_ctl->search_start, ffe_ctl->num_bytes);
3575 *cluster_bg_ret = cluster_bg;
3576 ffe_ctl->found_offset = offset;
3579 WARN_ON(last_ptr->block_group != cluster_bg);
3583 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3584 * lets just skip it and let the allocator find whatever block it can
3585 * find. If we reach this point, we will have tried the cluster
3586 * allocator plenty of times and not have found anything, so we are
3587 * likely way too fragmented for the clustering stuff to find anything.
3589 * However, if the cluster is taken from the current block group,
3590 * release the cluster first, so that we stand a better chance of
3591 * succeeding in the unclustered allocation.
3593 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3594 spin_unlock(&last_ptr->refill_lock);
3595 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3599 /* This cluster didn't work out, free it and start over */
3600 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3602 if (cluster_bg != bg)
3603 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3606 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3607 spin_unlock(&last_ptr->refill_lock);
3611 aligned_cluster = max_t(u64,
3612 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3613 bg->full_stripe_len);
3614 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3615 ffe_ctl->num_bytes, aligned_cluster);
3617 /* Now pull our allocation out of this cluster */
3618 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3619 ffe_ctl->num_bytes, ffe_ctl->search_start,
3620 &ffe_ctl->max_extent_size);
3622 /* We found one, proceed */
3623 spin_unlock(&last_ptr->refill_lock);
3624 trace_btrfs_reserve_extent_cluster(bg,
3625 ffe_ctl->search_start,
3626 ffe_ctl->num_bytes);
3627 ffe_ctl->found_offset = offset;
3630 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
3631 !ffe_ctl->retry_clustered) {
3632 spin_unlock(&last_ptr->refill_lock);
3634 ffe_ctl->retry_clustered = true;
3635 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3636 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
3640 * At this point we either didn't find a cluster or we weren't able to
3641 * allocate a block from our cluster. Free the cluster we've been
3642 * trying to use, and go to the next block group.
3644 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3645 spin_unlock(&last_ptr->refill_lock);
3650 * Return >0 to inform caller that we find nothing
3651 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3652 * Return -EAGAIN to inform caller that we need to re-search this block group
3654 static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3655 struct find_free_extent_ctl *ffe_ctl)
3657 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3661 * We are doing an unclustered allocation, set the fragmented flag so
3662 * we don't bother trying to setup a cluster again until we get more
3665 if (unlikely(last_ptr)) {
3666 spin_lock(&last_ptr->lock);
3667 last_ptr->fragmented = 1;
3668 spin_unlock(&last_ptr->lock);
3670 if (ffe_ctl->cached) {
3671 struct btrfs_free_space_ctl *free_space_ctl;
3673 free_space_ctl = bg->free_space_ctl;
3674 spin_lock(&free_space_ctl->tree_lock);
3675 if (free_space_ctl->free_space <
3676 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3677 ffe_ctl->empty_size) {
3678 ffe_ctl->total_free_space = max_t(u64,
3679 ffe_ctl->total_free_space,
3680 free_space_ctl->free_space);
3681 spin_unlock(&free_space_ctl->tree_lock);
3684 spin_unlock(&free_space_ctl->tree_lock);
3687 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3688 ffe_ctl->num_bytes, ffe_ctl->empty_size,
3689 &ffe_ctl->max_extent_size);
3692 * If we didn't find a chunk, and we haven't failed on this block group
3693 * before, and this block group is in the middle of caching and we are
3694 * ok with waiting, then go ahead and wait for progress to be made, and
3695 * set @retry_unclustered to true.
3697 * If @retry_unclustered is true then we've already waited on this
3698 * block group once and should move on to the next block group.
3700 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
3701 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
3702 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3703 ffe_ctl->empty_size);
3704 ffe_ctl->retry_unclustered = true;
3706 } else if (!offset) {
3709 ffe_ctl->found_offset = offset;
3713 static int do_allocation_clustered(struct btrfs_block_group *block_group,
3714 struct find_free_extent_ctl *ffe_ctl,
3715 struct btrfs_block_group **bg_ret)
3719 /* We want to try and use the cluster allocator, so lets look there */
3720 if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
3721 ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
3722 if (ret >= 0 || ret == -EAGAIN)
3724 /* ret == -ENOENT case falls through */
3727 return find_free_extent_unclustered(block_group, ffe_ctl);
3731 * Tree-log block group locking
3732 * ============================
3734 * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3735 * indicates the starting address of a block group, which is reserved only
3736 * for tree-log metadata.
3743 * fs_info::treelog_bg_lock
3747 * Simple allocator for sequential-only block group. It only allows sequential
3748 * allocation. No need to play with trees. This function also reserves the
3749 * bytes as in btrfs_add_reserved_bytes.
3751 static int do_allocation_zoned(struct btrfs_block_group *block_group,
3752 struct find_free_extent_ctl *ffe_ctl,
3753 struct btrfs_block_group **bg_ret)
3755 struct btrfs_fs_info *fs_info = block_group->fs_info;
3756 struct btrfs_space_info *space_info = block_group->space_info;
3757 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3758 u64 start = block_group->start;
3759 u64 num_bytes = ffe_ctl->num_bytes;
3761 u64 bytenr = block_group->start;
3763 u64 data_reloc_bytenr;
3767 ASSERT(btrfs_is_zoned(block_group->fs_info));
3770 * Do not allow non-tree-log blocks in the dedicated tree-log block
3771 * group, and vice versa.
3773 spin_lock(&fs_info->treelog_bg_lock);
3774 log_bytenr = fs_info->treelog_bg;
3775 if (log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) ||
3776 (!ffe_ctl->for_treelog && bytenr == log_bytenr)))
3778 spin_unlock(&fs_info->treelog_bg_lock);
3783 * Do not allow non-relocation blocks in the dedicated relocation block
3784 * group, and vice versa.
3786 spin_lock(&fs_info->relocation_bg_lock);
3787 data_reloc_bytenr = fs_info->data_reloc_bg;
3788 if (data_reloc_bytenr &&
3789 ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) ||
3790 (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr)))
3792 spin_unlock(&fs_info->relocation_bg_lock);
3795 /* Check RO and no space case before trying to activate it */
3796 spin_lock(&block_group->lock);
3797 if (block_group->ro ||
3798 block_group->alloc_offset == block_group->zone_capacity) {
3799 spin_unlock(&block_group->lock);
3802 spin_unlock(&block_group->lock);
3804 if (!btrfs_zone_activate(block_group))
3807 spin_lock(&space_info->lock);
3808 spin_lock(&block_group->lock);
3809 spin_lock(&fs_info->treelog_bg_lock);
3810 spin_lock(&fs_info->relocation_bg_lock);
3812 ASSERT(!ffe_ctl->for_treelog ||
3813 block_group->start == fs_info->treelog_bg ||
3814 fs_info->treelog_bg == 0);
3815 ASSERT(!ffe_ctl->for_data_reloc ||
3816 block_group->start == fs_info->data_reloc_bg ||
3817 fs_info->data_reloc_bg == 0);
3819 if (block_group->ro) {
3825 * Do not allow currently using block group to be tree-log dedicated
3828 if (ffe_ctl->for_treelog && !fs_info->treelog_bg &&
3829 (block_group->used || block_group->reserved)) {
3835 * Do not allow currently used block group to be the data relocation
3836 * dedicated block group.
3838 if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg &&
3839 (block_group->used || block_group->reserved)) {
3844 WARN_ON_ONCE(block_group->alloc_offset > block_group->zone_capacity);
3845 avail = block_group->zone_capacity - block_group->alloc_offset;
3846 if (avail < num_bytes) {
3847 if (ffe_ctl->max_extent_size < avail) {
3849 * With sequential allocator, free space is always
3852 ffe_ctl->max_extent_size = avail;
3853 ffe_ctl->total_free_space = avail;
3859 if (ffe_ctl->for_treelog && !fs_info->treelog_bg)
3860 fs_info->treelog_bg = block_group->start;
3862 if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg)
3863 fs_info->data_reloc_bg = block_group->start;
3865 ffe_ctl->found_offset = start + block_group->alloc_offset;
3866 block_group->alloc_offset += num_bytes;
3867 spin_lock(&ctl->tree_lock);
3868 ctl->free_space -= num_bytes;
3869 spin_unlock(&ctl->tree_lock);
3872 * We do not check if found_offset is aligned to stripesize. The
3873 * address is anyway rewritten when using zone append writing.
3876 ffe_ctl->search_start = ffe_ctl->found_offset;
3879 if (ret && ffe_ctl->for_treelog)
3880 fs_info->treelog_bg = 0;
3881 if (ret && ffe_ctl->for_data_reloc)
3882 fs_info->data_reloc_bg = 0;
3883 spin_unlock(&fs_info->relocation_bg_lock);
3884 spin_unlock(&fs_info->treelog_bg_lock);
3885 spin_unlock(&block_group->lock);
3886 spin_unlock(&space_info->lock);
3890 static int do_allocation(struct btrfs_block_group *block_group,
3891 struct find_free_extent_ctl *ffe_ctl,
3892 struct btrfs_block_group **bg_ret)
3894 switch (ffe_ctl->policy) {
3895 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3896 return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
3897 case BTRFS_EXTENT_ALLOC_ZONED:
3898 return do_allocation_zoned(block_group, ffe_ctl, bg_ret);
3904 static void release_block_group(struct btrfs_block_group *block_group,
3905 struct find_free_extent_ctl *ffe_ctl,
3908 switch (ffe_ctl->policy) {
3909 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3910 ffe_ctl->retry_clustered = false;
3911 ffe_ctl->retry_unclustered = false;
3913 case BTRFS_EXTENT_ALLOC_ZONED:
3920 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
3922 btrfs_release_block_group(block_group, delalloc);
3925 static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
3926 struct btrfs_key *ins)
3928 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3930 if (!ffe_ctl->use_cluster && last_ptr) {
3931 spin_lock(&last_ptr->lock);
3932 last_ptr->window_start = ins->objectid;
3933 spin_unlock(&last_ptr->lock);
3937 static void found_extent(struct find_free_extent_ctl *ffe_ctl,
3938 struct btrfs_key *ins)
3940 switch (ffe_ctl->policy) {
3941 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3942 found_extent_clustered(ffe_ctl, ins);
3944 case BTRFS_EXTENT_ALLOC_ZONED:
3952 static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl)
3954 switch (ffe_ctl->policy) {
3955 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3957 * If we can't allocate a new chunk we've already looped through
3958 * at least once, move on to the NO_EMPTY_SIZE case.
3960 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
3962 case BTRFS_EXTENT_ALLOC_ZONED:
3971 * Return >0 means caller needs to re-search for free extent
3972 * Return 0 means we have the needed free extent.
3973 * Return <0 means we failed to locate any free extent.
3975 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
3976 struct btrfs_key *ins,
3977 struct find_free_extent_ctl *ffe_ctl,
3980 struct btrfs_root *root = fs_info->extent_root;
3983 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
3984 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
3985 ffe_ctl->orig_have_caching_bg = true;
3987 if (ins->objectid) {
3988 found_extent(ffe_ctl, ins);
3992 if (ffe_ctl->max_extent_size >= ffe_ctl->min_alloc_size &&
3993 !btrfs_can_activate_zone(fs_info->fs_devices, ffe_ctl->index)) {
3995 * If we have enough free space left in an already active block
3996 * group and we can't activate any other zone now, retry the
3997 * active ones with a smaller allocation size. Returning early
3998 * from here will tell btrfs_reserve_extent() to haven the
4004 if (ffe_ctl->loop >= LOOP_CACHING_WAIT && ffe_ctl->have_caching_bg)
4008 if (ffe_ctl->index < BTRFS_NR_RAID_TYPES)
4012 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4013 * caching kthreads as we move along
4014 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4015 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4016 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4019 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
4021 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
4023 * We want to skip the LOOP_CACHING_WAIT step if we
4024 * don't have any uncached bgs and we've already done a
4025 * full search through.
4027 if (ffe_ctl->orig_have_caching_bg || !full_search)
4028 ffe_ctl->loop = LOOP_CACHING_WAIT;
4030 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
4035 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
4036 struct btrfs_trans_handle *trans;
4039 trans = current->journal_info;
4043 trans = btrfs_join_transaction(root);
4045 if (IS_ERR(trans)) {
4046 ret = PTR_ERR(trans);
4050 ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
4053 /* Do not bail out on ENOSPC since we can do more. */
4055 ret = chunk_allocation_failed(ffe_ctl);
4057 btrfs_abort_transaction(trans, ret);
4061 btrfs_end_transaction(trans);
4066 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
4067 if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
4071 * Don't loop again if we already have no empty_size and
4074 if (ffe_ctl->empty_size == 0 &&
4075 ffe_ctl->empty_cluster == 0)
4077 ffe_ctl->empty_size = 0;
4078 ffe_ctl->empty_cluster = 0;
4085 static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
4086 struct find_free_extent_ctl *ffe_ctl,
4087 struct btrfs_space_info *space_info,
4088 struct btrfs_key *ins)
4091 * If our free space is heavily fragmented we may not be able to make
4092 * big contiguous allocations, so instead of doing the expensive search
4093 * for free space, simply return ENOSPC with our max_extent_size so we
4094 * can go ahead and search for a more manageable chunk.
4096 * If our max_extent_size is large enough for our allocation simply
4097 * disable clustering since we will likely not be able to find enough
4098 * space to create a cluster and induce latency trying.
4100 if (space_info->max_extent_size) {
4101 spin_lock(&space_info->lock);
4102 if (space_info->max_extent_size &&
4103 ffe_ctl->num_bytes > space_info->max_extent_size) {
4104 ins->offset = space_info->max_extent_size;
4105 spin_unlock(&space_info->lock);
4107 } else if (space_info->max_extent_size) {
4108 ffe_ctl->use_cluster = false;
4110 spin_unlock(&space_info->lock);
4113 ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
4114 &ffe_ctl->empty_cluster);
4115 if (ffe_ctl->last_ptr) {
4116 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4118 spin_lock(&last_ptr->lock);
4119 if (last_ptr->block_group)
4120 ffe_ctl->hint_byte = last_ptr->window_start;
4121 if (last_ptr->fragmented) {
4123 * We still set window_start so we can keep track of the
4124 * last place we found an allocation to try and save
4127 ffe_ctl->hint_byte = last_ptr->window_start;
4128 ffe_ctl->use_cluster = false;
4130 spin_unlock(&last_ptr->lock);
4136 static int prepare_allocation(struct btrfs_fs_info *fs_info,
4137 struct find_free_extent_ctl *ffe_ctl,
4138 struct btrfs_space_info *space_info,
4139 struct btrfs_key *ins)
4141 switch (ffe_ctl->policy) {
4142 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4143 return prepare_allocation_clustered(fs_info, ffe_ctl,
4145 case BTRFS_EXTENT_ALLOC_ZONED:
4146 if (ffe_ctl->for_treelog) {
4147 spin_lock(&fs_info->treelog_bg_lock);
4148 if (fs_info->treelog_bg)
4149 ffe_ctl->hint_byte = fs_info->treelog_bg;
4150 spin_unlock(&fs_info->treelog_bg_lock);
4152 if (ffe_ctl->for_data_reloc) {
4153 spin_lock(&fs_info->relocation_bg_lock);
4154 if (fs_info->data_reloc_bg)
4155 ffe_ctl->hint_byte = fs_info->data_reloc_bg;
4156 spin_unlock(&fs_info->relocation_bg_lock);
4165 * walks the btree of allocated extents and find a hole of a given size.
4166 * The key ins is changed to record the hole:
4167 * ins->objectid == start position
4168 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4169 * ins->offset == the size of the hole.
4170 * Any available blocks before search_start are skipped.
4172 * If there is no suitable free space, we will record the max size of
4173 * the free space extent currently.
4175 * The overall logic and call chain:
4177 * find_free_extent()
4178 * |- Iterate through all block groups
4179 * | |- Get a valid block group
4180 * | |- Try to do clustered allocation in that block group
4181 * | |- Try to do unclustered allocation in that block group
4182 * | |- Check if the result is valid
4183 * | | |- If valid, then exit
4184 * | |- Jump to next block group
4186 * |- Push harder to find free extents
4187 * |- If not found, re-iterate all block groups
4189 static noinline int find_free_extent(struct btrfs_root *root,
4190 struct btrfs_key *ins,
4191 struct find_free_extent_ctl *ffe_ctl)
4193 struct btrfs_fs_info *fs_info = root->fs_info;
4195 int cache_block_group_error = 0;
4196 struct btrfs_block_group *block_group = NULL;
4197 struct btrfs_space_info *space_info;
4198 bool full_search = false;
4200 WARN_ON(ffe_ctl->num_bytes < fs_info->sectorsize);
4202 ffe_ctl->search_start = 0;
4203 /* For clustered allocation */
4204 ffe_ctl->empty_cluster = 0;
4205 ffe_ctl->last_ptr = NULL;
4206 ffe_ctl->use_cluster = true;
4207 ffe_ctl->have_caching_bg = false;
4208 ffe_ctl->orig_have_caching_bg = false;
4209 ffe_ctl->index = btrfs_bg_flags_to_raid_index(ffe_ctl->flags);
4211 /* For clustered allocation */
4212 ffe_ctl->retry_clustered = false;
4213 ffe_ctl->retry_unclustered = false;
4214 ffe_ctl->cached = 0;
4215 ffe_ctl->max_extent_size = 0;
4216 ffe_ctl->total_free_space = 0;
4217 ffe_ctl->found_offset = 0;
4218 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
4220 if (btrfs_is_zoned(fs_info))
4221 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_ZONED;
4223 ins->type = BTRFS_EXTENT_ITEM_KEY;
4227 trace_find_free_extent(root, ffe_ctl->num_bytes, ffe_ctl->empty_size,
4230 space_info = btrfs_find_space_info(fs_info, ffe_ctl->flags);
4232 btrfs_err(fs_info, "No space info for %llu", ffe_ctl->flags);
4236 ret = prepare_allocation(fs_info, ffe_ctl, space_info, ins);
4240 ffe_ctl->search_start = max(ffe_ctl->search_start,
4241 first_logical_byte(fs_info, 0));
4242 ffe_ctl->search_start = max(ffe_ctl->search_start, ffe_ctl->hint_byte);
4243 if (ffe_ctl->search_start == ffe_ctl->hint_byte) {
4244 block_group = btrfs_lookup_block_group(fs_info,
4245 ffe_ctl->search_start);
4247 * we don't want to use the block group if it doesn't match our
4248 * allocation bits, or if its not cached.
4250 * However if we are re-searching with an ideal block group
4251 * picked out then we don't care that the block group is cached.
4253 if (block_group && block_group_bits(block_group, ffe_ctl->flags) &&
4254 block_group->cached != BTRFS_CACHE_NO) {
4255 down_read(&space_info->groups_sem);
4256 if (list_empty(&block_group->list) ||
4259 * someone is removing this block group,
4260 * we can't jump into the have_block_group
4261 * target because our list pointers are not
4264 btrfs_put_block_group(block_group);
4265 up_read(&space_info->groups_sem);
4267 ffe_ctl->index = btrfs_bg_flags_to_raid_index(
4268 block_group->flags);
4269 btrfs_lock_block_group(block_group,
4271 goto have_block_group;
4273 } else if (block_group) {
4274 btrfs_put_block_group(block_group);
4278 ffe_ctl->have_caching_bg = false;
4279 if (ffe_ctl->index == btrfs_bg_flags_to_raid_index(ffe_ctl->flags) ||
4280 ffe_ctl->index == 0)
4282 down_read(&space_info->groups_sem);
4283 list_for_each_entry(block_group,
4284 &space_info->block_groups[ffe_ctl->index], list) {
4285 struct btrfs_block_group *bg_ret;
4287 /* If the block group is read-only, we can skip it entirely. */
4288 if (unlikely(block_group->ro)) {
4289 if (ffe_ctl->for_treelog)
4290 btrfs_clear_treelog_bg(block_group);
4291 if (ffe_ctl->for_data_reloc)
4292 btrfs_clear_data_reloc_bg(block_group);
4296 btrfs_grab_block_group(block_group, ffe_ctl->delalloc);
4297 ffe_ctl->search_start = block_group->start;
4300 * this can happen if we end up cycling through all the
4301 * raid types, but we want to make sure we only allocate
4302 * for the proper type.
4304 if (!block_group_bits(block_group, ffe_ctl->flags)) {
4305 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4306 BTRFS_BLOCK_GROUP_RAID1_MASK |
4307 BTRFS_BLOCK_GROUP_RAID56_MASK |
4308 BTRFS_BLOCK_GROUP_RAID10;
4311 * if they asked for extra copies and this block group
4312 * doesn't provide them, bail. This does allow us to
4313 * fill raid0 from raid1.
4315 if ((ffe_ctl->flags & extra) && !(block_group->flags & extra))
4319 * This block group has different flags than we want.
4320 * It's possible that we have MIXED_GROUP flag but no
4321 * block group is mixed. Just skip such block group.
4323 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4328 ffe_ctl->cached = btrfs_block_group_done(block_group);
4329 if (unlikely(!ffe_ctl->cached)) {
4330 ffe_ctl->have_caching_bg = true;
4331 ret = btrfs_cache_block_group(block_group, 0);
4334 * If we get ENOMEM here or something else we want to
4335 * try other block groups, because it may not be fatal.
4336 * However if we can't find anything else we need to
4337 * save our return here so that we return the actual
4338 * error that caused problems, not ENOSPC.
4341 if (!cache_block_group_error)
4342 cache_block_group_error = ret;
4349 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
4353 ret = do_allocation(block_group, ffe_ctl, &bg_ret);
4355 if (bg_ret && bg_ret != block_group) {
4356 btrfs_release_block_group(block_group,
4358 block_group = bg_ret;
4360 } else if (ret == -EAGAIN) {
4361 goto have_block_group;
4362 } else if (ret > 0) {
4367 ffe_ctl->search_start = round_up(ffe_ctl->found_offset,
4368 fs_info->stripesize);
4370 /* move on to the next group */
4371 if (ffe_ctl->search_start + ffe_ctl->num_bytes >
4372 block_group->start + block_group->length) {
4373 btrfs_add_free_space_unused(block_group,
4374 ffe_ctl->found_offset,
4375 ffe_ctl->num_bytes);
4379 if (ffe_ctl->found_offset < ffe_ctl->search_start)
4380 btrfs_add_free_space_unused(block_group,
4381 ffe_ctl->found_offset,
4382 ffe_ctl->search_start - ffe_ctl->found_offset);
4384 ret = btrfs_add_reserved_bytes(block_group, ffe_ctl->ram_bytes,
4387 if (ret == -EAGAIN) {
4388 btrfs_add_free_space_unused(block_group,
4389 ffe_ctl->found_offset,
4390 ffe_ctl->num_bytes);
4393 btrfs_inc_block_group_reservations(block_group);
4395 /* we are all good, lets return */
4396 ins->objectid = ffe_ctl->search_start;
4397 ins->offset = ffe_ctl->num_bytes;
4399 trace_btrfs_reserve_extent(block_group, ffe_ctl->search_start,
4400 ffe_ctl->num_bytes);
4401 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4404 release_block_group(block_group, ffe_ctl, ffe_ctl->delalloc);
4407 up_read(&space_info->groups_sem);
4409 ret = find_free_extent_update_loop(fs_info, ins, ffe_ctl, full_search);
4413 if (ret == -ENOSPC && !cache_block_group_error) {
4415 * Use ffe_ctl->total_free_space as fallback if we can't find
4416 * any contiguous hole.
4418 if (!ffe_ctl->max_extent_size)
4419 ffe_ctl->max_extent_size = ffe_ctl->total_free_space;
4420 spin_lock(&space_info->lock);
4421 space_info->max_extent_size = ffe_ctl->max_extent_size;
4422 spin_unlock(&space_info->lock);
4423 ins->offset = ffe_ctl->max_extent_size;
4424 } else if (ret == -ENOSPC) {
4425 ret = cache_block_group_error;
4431 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4432 * hole that is at least as big as @num_bytes.
4434 * @root - The root that will contain this extent
4436 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4437 * is used for accounting purposes. This value differs
4438 * from @num_bytes only in the case of compressed extents.
4440 * @num_bytes - Number of bytes to allocate on-disk.
4442 * @min_alloc_size - Indicates the minimum amount of space that the
4443 * allocator should try to satisfy. In some cases
4444 * @num_bytes may be larger than what is required and if
4445 * the filesystem is fragmented then allocation fails.
4446 * However, the presence of @min_alloc_size gives a
4447 * chance to try and satisfy the smaller allocation.
4449 * @empty_size - A hint that you plan on doing more COW. This is the
4450 * size in bytes the allocator should try to find free
4451 * next to the block it returns. This is just a hint and
4452 * may be ignored by the allocator.
4454 * @hint_byte - Hint to the allocator to start searching above the byte
4455 * address passed. It might be ignored.
4457 * @ins - This key is modified to record the found hole. It will
4458 * have the following values:
4459 * ins->objectid == start position
4460 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4461 * ins->offset == the size of the hole.
4463 * @is_data - Boolean flag indicating whether an extent is
4464 * allocated for data (true) or metadata (false)
4466 * @delalloc - Boolean flag indicating whether this allocation is for
4467 * delalloc or not. If 'true' data_rwsem of block groups
4468 * is going to be acquired.
4471 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4472 * case -ENOSPC is returned then @ins->offset will contain the size of the
4473 * largest available hole the allocator managed to find.
4475 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4476 u64 num_bytes, u64 min_alloc_size,
4477 u64 empty_size, u64 hint_byte,
4478 struct btrfs_key *ins, int is_data, int delalloc)
4480 struct btrfs_fs_info *fs_info = root->fs_info;
4481 struct find_free_extent_ctl ffe_ctl = {};
4482 bool final_tried = num_bytes == min_alloc_size;
4485 bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4486 bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data);
4488 flags = get_alloc_profile_by_root(root, is_data);
4490 WARN_ON(num_bytes < fs_info->sectorsize);
4492 ffe_ctl.ram_bytes = ram_bytes;
4493 ffe_ctl.num_bytes = num_bytes;
4494 ffe_ctl.min_alloc_size = min_alloc_size;
4495 ffe_ctl.empty_size = empty_size;
4496 ffe_ctl.flags = flags;
4497 ffe_ctl.delalloc = delalloc;
4498 ffe_ctl.hint_byte = hint_byte;
4499 ffe_ctl.for_treelog = for_treelog;
4500 ffe_ctl.for_data_reloc = for_data_reloc;
4502 ret = find_free_extent(root, ins, &ffe_ctl);
4503 if (!ret && !is_data) {
4504 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4505 } else if (ret == -ENOSPC) {
4506 if (!final_tried && ins->offset) {
4507 num_bytes = min(num_bytes >> 1, ins->offset);
4508 num_bytes = round_down(num_bytes,
4509 fs_info->sectorsize);
4510 num_bytes = max(num_bytes, min_alloc_size);
4511 ram_bytes = num_bytes;
4512 if (num_bytes == min_alloc_size)
4515 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4516 struct btrfs_space_info *sinfo;
4518 sinfo = btrfs_find_space_info(fs_info, flags);
4520 "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d",
4521 flags, num_bytes, for_treelog, for_data_reloc);
4523 btrfs_dump_space_info(fs_info, sinfo,
4531 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4532 u64 start, u64 len, int delalloc)
4534 struct btrfs_block_group *cache;
4536 cache = btrfs_lookup_block_group(fs_info, start);
4538 btrfs_err(fs_info, "Unable to find block group for %llu",
4543 btrfs_add_free_space(cache, start, len);
4544 btrfs_free_reserved_bytes(cache, len, delalloc);
4545 trace_btrfs_reserved_extent_free(fs_info, start, len);
4547 btrfs_put_block_group(cache);
4551 int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start,
4554 struct btrfs_block_group *cache;
4557 cache = btrfs_lookup_block_group(trans->fs_info, start);
4559 btrfs_err(trans->fs_info, "unable to find block group for %llu",
4564 ret = pin_down_extent(trans, cache, start, len, 1);
4565 btrfs_put_block_group(cache);
4569 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4570 u64 parent, u64 root_objectid,
4571 u64 flags, u64 owner, u64 offset,
4572 struct btrfs_key *ins, int ref_mod)
4574 struct btrfs_fs_info *fs_info = trans->fs_info;
4576 struct btrfs_extent_item *extent_item;
4577 struct btrfs_extent_inline_ref *iref;
4578 struct btrfs_path *path;
4579 struct extent_buffer *leaf;
4584 type = BTRFS_SHARED_DATA_REF_KEY;
4586 type = BTRFS_EXTENT_DATA_REF_KEY;
4588 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4590 path = btrfs_alloc_path();
4594 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4597 btrfs_free_path(path);
4601 leaf = path->nodes[0];
4602 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4603 struct btrfs_extent_item);
4604 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4605 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4606 btrfs_set_extent_flags(leaf, extent_item,
4607 flags | BTRFS_EXTENT_FLAG_DATA);
4609 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4610 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4612 struct btrfs_shared_data_ref *ref;
4613 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4614 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4615 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4617 struct btrfs_extent_data_ref *ref;
4618 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4619 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4620 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4621 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4622 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4625 btrfs_mark_buffer_dirty(path->nodes[0]);
4626 btrfs_free_path(path);
4628 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
4632 ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1);
4633 if (ret) { /* -ENOENT, logic error */
4634 btrfs_err(fs_info, "update block group failed for %llu %llu",
4635 ins->objectid, ins->offset);
4638 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
4642 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4643 struct btrfs_delayed_ref_node *node,
4644 struct btrfs_delayed_extent_op *extent_op)
4646 struct btrfs_fs_info *fs_info = trans->fs_info;
4648 struct btrfs_extent_item *extent_item;
4649 struct btrfs_key extent_key;
4650 struct btrfs_tree_block_info *block_info;
4651 struct btrfs_extent_inline_ref *iref;
4652 struct btrfs_path *path;
4653 struct extent_buffer *leaf;
4654 struct btrfs_delayed_tree_ref *ref;
4655 u32 size = sizeof(*extent_item) + sizeof(*iref);
4657 u64 flags = extent_op->flags_to_set;
4658 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4660 ref = btrfs_delayed_node_to_tree_ref(node);
4662 extent_key.objectid = node->bytenr;
4663 if (skinny_metadata) {
4664 extent_key.offset = ref->level;
4665 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4666 num_bytes = fs_info->nodesize;
4668 extent_key.offset = node->num_bytes;
4669 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4670 size += sizeof(*block_info);
4671 num_bytes = node->num_bytes;
4674 path = btrfs_alloc_path();
4678 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4681 btrfs_free_path(path);
4685 leaf = path->nodes[0];
4686 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4687 struct btrfs_extent_item);
4688 btrfs_set_extent_refs(leaf, extent_item, 1);
4689 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4690 btrfs_set_extent_flags(leaf, extent_item,
4691 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4693 if (skinny_metadata) {
4694 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4696 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4697 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4698 btrfs_set_tree_block_level(leaf, block_info, ref->level);
4699 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4702 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4703 btrfs_set_extent_inline_ref_type(leaf, iref,
4704 BTRFS_SHARED_BLOCK_REF_KEY);
4705 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4707 btrfs_set_extent_inline_ref_type(leaf, iref,
4708 BTRFS_TREE_BLOCK_REF_KEY);
4709 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4712 btrfs_mark_buffer_dirty(leaf);
4713 btrfs_free_path(path);
4715 ret = remove_from_free_space_tree(trans, extent_key.objectid,
4720 ret = btrfs_update_block_group(trans, extent_key.objectid,
4721 fs_info->nodesize, 1);
4722 if (ret) { /* -ENOENT, logic error */
4723 btrfs_err(fs_info, "update block group failed for %llu %llu",
4724 extent_key.objectid, extent_key.offset);
4728 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
4733 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4734 struct btrfs_root *root, u64 owner,
4735 u64 offset, u64 ram_bytes,
4736 struct btrfs_key *ins)
4738 struct btrfs_ref generic_ref = { 0 };
4740 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4742 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4743 ins->objectid, ins->offset, 0);
4744 btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset);
4745 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4747 return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes);
4751 * this is used by the tree logging recovery code. It records that
4752 * an extent has been allocated and makes sure to clear the free
4753 * space cache bits as well
4755 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4756 u64 root_objectid, u64 owner, u64 offset,
4757 struct btrfs_key *ins)
4759 struct btrfs_fs_info *fs_info = trans->fs_info;
4761 struct btrfs_block_group *block_group;
4762 struct btrfs_space_info *space_info;
4765 * Mixed block groups will exclude before processing the log so we only
4766 * need to do the exclude dance if this fs isn't mixed.
4768 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4769 ret = __exclude_logged_extent(fs_info, ins->objectid,
4775 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4779 space_info = block_group->space_info;
4780 spin_lock(&space_info->lock);
4781 spin_lock(&block_group->lock);
4782 space_info->bytes_reserved += ins->offset;
4783 block_group->reserved += ins->offset;
4784 spin_unlock(&block_group->lock);
4785 spin_unlock(&space_info->lock);
4787 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4790 btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
4791 btrfs_put_block_group(block_group);
4795 static struct extent_buffer *
4796 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4797 u64 bytenr, int level, u64 owner,
4798 enum btrfs_lock_nesting nest)
4800 struct btrfs_fs_info *fs_info = root->fs_info;
4801 struct extent_buffer *buf;
4803 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level);
4808 * Extra safety check in case the extent tree is corrupted and extent
4809 * allocator chooses to use a tree block which is already used and
4812 if (buf->lock_owner == current->pid) {
4813 btrfs_err_rl(fs_info,
4814 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4815 buf->start, btrfs_header_owner(buf), current->pid);
4816 free_extent_buffer(buf);
4817 return ERR_PTR(-EUCLEAN);
4821 * This needs to stay, because we could allocate a freed block from an
4822 * old tree into a new tree, so we need to make sure this new block is
4823 * set to the appropriate level and owner.
4825 btrfs_set_buffer_lockdep_class(owner, buf, level);
4826 __btrfs_tree_lock(buf, nest);
4827 btrfs_clean_tree_block(buf);
4828 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
4829 clear_bit(EXTENT_BUFFER_NO_CHECK, &buf->bflags);
4831 set_extent_buffer_uptodate(buf);
4833 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
4834 btrfs_set_header_level(buf, level);
4835 btrfs_set_header_bytenr(buf, buf->start);
4836 btrfs_set_header_generation(buf, trans->transid);
4837 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
4838 btrfs_set_header_owner(buf, owner);
4839 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
4840 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
4841 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4842 buf->log_index = root->log_transid % 2;
4844 * we allow two log transactions at a time, use different
4845 * EXTENT bit to differentiate dirty pages.
4847 if (buf->log_index == 0)
4848 set_extent_dirty(&root->dirty_log_pages, buf->start,
4849 buf->start + buf->len - 1, GFP_NOFS);
4851 set_extent_new(&root->dirty_log_pages, buf->start,
4852 buf->start + buf->len - 1);
4854 buf->log_index = -1;
4855 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4856 buf->start + buf->len - 1, GFP_NOFS);
4858 /* this returns a buffer locked for blocking */
4863 * finds a free extent and does all the dirty work required for allocation
4864 * returns the tree buffer or an ERR_PTR on error.
4866 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
4867 struct btrfs_root *root,
4868 u64 parent, u64 root_objectid,
4869 const struct btrfs_disk_key *key,
4870 int level, u64 hint,
4872 enum btrfs_lock_nesting nest)
4874 struct btrfs_fs_info *fs_info = root->fs_info;
4875 struct btrfs_key ins;
4876 struct btrfs_block_rsv *block_rsv;
4877 struct extent_buffer *buf;
4878 struct btrfs_delayed_extent_op *extent_op;
4879 struct btrfs_ref generic_ref = { 0 };
4882 u32 blocksize = fs_info->nodesize;
4883 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4885 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4886 if (btrfs_is_testing(fs_info)) {
4887 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
4888 level, root_objectid, nest);
4890 root->alloc_bytenr += blocksize;
4895 block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
4896 if (IS_ERR(block_rsv))
4897 return ERR_CAST(block_rsv);
4899 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
4900 empty_size, hint, &ins, 0, 0);
4904 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
4905 root_objectid, nest);
4908 goto out_free_reserved;
4911 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4913 parent = ins.objectid;
4914 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4918 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4919 extent_op = btrfs_alloc_delayed_extent_op();
4925 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4927 memset(&extent_op->key, 0, sizeof(extent_op->key));
4928 extent_op->flags_to_set = flags;
4929 extent_op->update_key = skinny_metadata ? false : true;
4930 extent_op->update_flags = true;
4931 extent_op->is_data = false;
4932 extent_op->level = level;
4934 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4935 ins.objectid, ins.offset, parent);
4936 generic_ref.real_root = root->root_key.objectid;
4937 btrfs_init_tree_ref(&generic_ref, level, root_objectid);
4938 btrfs_ref_tree_mod(fs_info, &generic_ref);
4939 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op);
4941 goto out_free_delayed;
4946 btrfs_free_delayed_extent_op(extent_op);
4948 btrfs_tree_unlock(buf);
4949 free_extent_buffer(buf);
4951 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
4953 btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
4954 return ERR_PTR(ret);
4957 struct walk_control {
4958 u64 refs[BTRFS_MAX_LEVEL];
4959 u64 flags[BTRFS_MAX_LEVEL];
4960 struct btrfs_key update_progress;
4961 struct btrfs_key drop_progress;
4973 #define DROP_REFERENCE 1
4974 #define UPDATE_BACKREF 2
4976 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4977 struct btrfs_root *root,
4978 struct walk_control *wc,
4979 struct btrfs_path *path)
4981 struct btrfs_fs_info *fs_info = root->fs_info;
4987 struct btrfs_key key;
4988 struct extent_buffer *eb;
4993 if (path->slots[wc->level] < wc->reada_slot) {
4994 wc->reada_count = wc->reada_count * 2 / 3;
4995 wc->reada_count = max(wc->reada_count, 2);
4997 wc->reada_count = wc->reada_count * 3 / 2;
4998 wc->reada_count = min_t(int, wc->reada_count,
4999 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
5002 eb = path->nodes[wc->level];
5003 nritems = btrfs_header_nritems(eb);
5005 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5006 if (nread >= wc->reada_count)
5010 bytenr = btrfs_node_blockptr(eb, slot);
5011 generation = btrfs_node_ptr_generation(eb, slot);
5013 if (slot == path->slots[wc->level])
5016 if (wc->stage == UPDATE_BACKREF &&
5017 generation <= root->root_key.offset)
5020 /* We don't lock the tree block, it's OK to be racy here */
5021 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
5022 wc->level - 1, 1, &refs,
5024 /* We don't care about errors in readahead. */
5029 if (wc->stage == DROP_REFERENCE) {
5033 if (wc->level == 1 &&
5034 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5036 if (!wc->update_ref ||
5037 generation <= root->root_key.offset)
5039 btrfs_node_key_to_cpu(eb, &key, slot);
5040 ret = btrfs_comp_cpu_keys(&key,
5041 &wc->update_progress);
5045 if (wc->level == 1 &&
5046 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5050 btrfs_readahead_node_child(eb, slot);
5053 wc->reada_slot = slot;
5057 * helper to process tree block while walking down the tree.
5059 * when wc->stage == UPDATE_BACKREF, this function updates
5060 * back refs for pointers in the block.
5062 * NOTE: return value 1 means we should stop walking down.
5064 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5065 struct btrfs_root *root,
5066 struct btrfs_path *path,
5067 struct walk_control *wc, int lookup_info)
5069 struct btrfs_fs_info *fs_info = root->fs_info;
5070 int level = wc->level;
5071 struct extent_buffer *eb = path->nodes[level];
5072 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5075 if (wc->stage == UPDATE_BACKREF &&
5076 btrfs_header_owner(eb) != root->root_key.objectid)
5080 * when reference count of tree block is 1, it won't increase
5081 * again. once full backref flag is set, we never clear it.
5084 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5085 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5086 BUG_ON(!path->locks[level]);
5087 ret = btrfs_lookup_extent_info(trans, fs_info,
5088 eb->start, level, 1,
5091 BUG_ON(ret == -ENOMEM);
5094 BUG_ON(wc->refs[level] == 0);
5097 if (wc->stage == DROP_REFERENCE) {
5098 if (wc->refs[level] > 1)
5101 if (path->locks[level] && !wc->keep_locks) {
5102 btrfs_tree_unlock_rw(eb, path->locks[level]);
5103 path->locks[level] = 0;
5108 /* wc->stage == UPDATE_BACKREF */
5109 if (!(wc->flags[level] & flag)) {
5110 BUG_ON(!path->locks[level]);
5111 ret = btrfs_inc_ref(trans, root, eb, 1);
5112 BUG_ON(ret); /* -ENOMEM */
5113 ret = btrfs_dec_ref(trans, root, eb, 0);
5114 BUG_ON(ret); /* -ENOMEM */
5115 ret = btrfs_set_disk_extent_flags(trans, eb, flag,
5116 btrfs_header_level(eb), 0);
5117 BUG_ON(ret); /* -ENOMEM */
5118 wc->flags[level] |= flag;
5122 * the block is shared by multiple trees, so it's not good to
5123 * keep the tree lock
5125 if (path->locks[level] && level > 0) {
5126 btrfs_tree_unlock_rw(eb, path->locks[level]);
5127 path->locks[level] = 0;
5133 * This is used to verify a ref exists for this root to deal with a bug where we
5134 * would have a drop_progress key that hadn't been updated properly.
5136 static int check_ref_exists(struct btrfs_trans_handle *trans,
5137 struct btrfs_root *root, u64 bytenr, u64 parent,
5140 struct btrfs_path *path;
5141 struct btrfs_extent_inline_ref *iref;
5144 path = btrfs_alloc_path();
5148 ret = lookup_extent_backref(trans, path, &iref, bytenr,
5149 root->fs_info->nodesize, parent,
5150 root->root_key.objectid, level, 0);
5151 btrfs_free_path(path);
5160 * helper to process tree block pointer.
5162 * when wc->stage == DROP_REFERENCE, this function checks
5163 * reference count of the block pointed to. if the block
5164 * is shared and we need update back refs for the subtree
5165 * rooted at the block, this function changes wc->stage to
5166 * UPDATE_BACKREF. if the block is shared and there is no
5167 * need to update back, this function drops the reference
5170 * NOTE: return value 1 means we should stop walking down.
5172 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5173 struct btrfs_root *root,
5174 struct btrfs_path *path,
5175 struct walk_control *wc, int *lookup_info)
5177 struct btrfs_fs_info *fs_info = root->fs_info;
5181 struct btrfs_key key;
5182 struct btrfs_key first_key;
5183 struct btrfs_ref ref = { 0 };
5184 struct extent_buffer *next;
5185 int level = wc->level;
5188 bool need_account = false;
5190 generation = btrfs_node_ptr_generation(path->nodes[level],
5191 path->slots[level]);
5193 * if the lower level block was created before the snapshot
5194 * was created, we know there is no need to update back refs
5197 if (wc->stage == UPDATE_BACKREF &&
5198 generation <= root->root_key.offset) {
5203 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5204 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
5205 path->slots[level]);
5207 next = find_extent_buffer(fs_info, bytenr);
5209 next = btrfs_find_create_tree_block(fs_info, bytenr,
5210 root->root_key.objectid, level - 1);
5212 return PTR_ERR(next);
5215 btrfs_tree_lock(next);
5217 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
5218 &wc->refs[level - 1],
5219 &wc->flags[level - 1]);
5223 if (unlikely(wc->refs[level - 1] == 0)) {
5224 btrfs_err(fs_info, "Missing references.");
5230 if (wc->stage == DROP_REFERENCE) {
5231 if (wc->refs[level - 1] > 1) {
5232 need_account = true;
5234 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5237 if (!wc->update_ref ||
5238 generation <= root->root_key.offset)
5241 btrfs_node_key_to_cpu(path->nodes[level], &key,
5242 path->slots[level]);
5243 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5247 wc->stage = UPDATE_BACKREF;
5248 wc->shared_level = level - 1;
5252 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5256 if (!btrfs_buffer_uptodate(next, generation, 0)) {
5257 btrfs_tree_unlock(next);
5258 free_extent_buffer(next);
5264 if (reada && level == 1)
5265 reada_walk_down(trans, root, wc, path);
5266 next = read_tree_block(fs_info, bytenr, root->root_key.objectid,
5267 generation, level - 1, &first_key);
5269 return PTR_ERR(next);
5270 } else if (!extent_buffer_uptodate(next)) {
5271 free_extent_buffer(next);
5274 btrfs_tree_lock(next);
5278 ASSERT(level == btrfs_header_level(next));
5279 if (level != btrfs_header_level(next)) {
5280 btrfs_err(root->fs_info, "mismatched level");
5284 path->nodes[level] = next;
5285 path->slots[level] = 0;
5286 path->locks[level] = BTRFS_WRITE_LOCK;
5292 wc->refs[level - 1] = 0;
5293 wc->flags[level - 1] = 0;
5294 if (wc->stage == DROP_REFERENCE) {
5295 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5296 parent = path->nodes[level]->start;
5298 ASSERT(root->root_key.objectid ==
5299 btrfs_header_owner(path->nodes[level]));
5300 if (root->root_key.objectid !=
5301 btrfs_header_owner(path->nodes[level])) {
5302 btrfs_err(root->fs_info,
5303 "mismatched block owner");
5311 * If we had a drop_progress we need to verify the refs are set
5312 * as expected. If we find our ref then we know that from here
5313 * on out everything should be correct, and we can clear the
5316 if (wc->restarted) {
5317 ret = check_ref_exists(trans, root, bytenr, parent,
5328 * Reloc tree doesn't contribute to qgroup numbers, and we have
5329 * already accounted them at merge time (replace_path),
5330 * thus we could skip expensive subtree trace here.
5332 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
5334 ret = btrfs_qgroup_trace_subtree(trans, next,
5335 generation, level - 1);
5337 btrfs_err_rl(fs_info,
5338 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5344 * We need to update the next key in our walk control so we can
5345 * update the drop_progress key accordingly. We don't care if
5346 * find_next_key doesn't find a key because that means we're at
5347 * the end and are going to clean up now.
5349 wc->drop_level = level;
5350 find_next_key(path, level, &wc->drop_progress);
5352 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
5353 fs_info->nodesize, parent);
5354 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid);
5355 ret = btrfs_free_extent(trans, &ref);
5364 btrfs_tree_unlock(next);
5365 free_extent_buffer(next);
5371 * helper to process tree block while walking up the tree.
5373 * when wc->stage == DROP_REFERENCE, this function drops
5374 * reference count on the block.
5376 * when wc->stage == UPDATE_BACKREF, this function changes
5377 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5378 * to UPDATE_BACKREF previously while processing the block.
5380 * NOTE: return value 1 means we should stop walking up.
5382 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5383 struct btrfs_root *root,
5384 struct btrfs_path *path,
5385 struct walk_control *wc)
5387 struct btrfs_fs_info *fs_info = root->fs_info;
5389 int level = wc->level;
5390 struct extent_buffer *eb = path->nodes[level];
5393 if (wc->stage == UPDATE_BACKREF) {
5394 BUG_ON(wc->shared_level < level);
5395 if (level < wc->shared_level)
5398 ret = find_next_key(path, level + 1, &wc->update_progress);
5402 wc->stage = DROP_REFERENCE;
5403 wc->shared_level = -1;
5404 path->slots[level] = 0;
5407 * check reference count again if the block isn't locked.
5408 * we should start walking down the tree again if reference
5411 if (!path->locks[level]) {
5413 btrfs_tree_lock(eb);
5414 path->locks[level] = BTRFS_WRITE_LOCK;
5416 ret = btrfs_lookup_extent_info(trans, fs_info,
5417 eb->start, level, 1,
5421 btrfs_tree_unlock_rw(eb, path->locks[level]);
5422 path->locks[level] = 0;
5425 BUG_ON(wc->refs[level] == 0);
5426 if (wc->refs[level] == 1) {
5427 btrfs_tree_unlock_rw(eb, path->locks[level]);
5428 path->locks[level] = 0;
5434 /* wc->stage == DROP_REFERENCE */
5435 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5437 if (wc->refs[level] == 1) {
5439 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5440 ret = btrfs_dec_ref(trans, root, eb, 1);
5442 ret = btrfs_dec_ref(trans, root, eb, 0);
5443 BUG_ON(ret); /* -ENOMEM */
5444 if (is_fstree(root->root_key.objectid)) {
5445 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5447 btrfs_err_rl(fs_info,
5448 "error %d accounting leaf items, quota is out of sync, rescan required",
5453 /* make block locked assertion in btrfs_clean_tree_block happy */
5454 if (!path->locks[level] &&
5455 btrfs_header_generation(eb) == trans->transid) {
5456 btrfs_tree_lock(eb);
5457 path->locks[level] = BTRFS_WRITE_LOCK;
5459 btrfs_clean_tree_block(eb);
5462 if (eb == root->node) {
5463 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5465 else if (root->root_key.objectid != btrfs_header_owner(eb))
5466 goto owner_mismatch;
5468 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5469 parent = path->nodes[level + 1]->start;
5470 else if (root->root_key.objectid !=
5471 btrfs_header_owner(path->nodes[level + 1]))
5472 goto owner_mismatch;
5475 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
5477 wc->refs[level] = 0;
5478 wc->flags[level] = 0;
5482 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5483 btrfs_header_owner(eb), root->root_key.objectid);
5487 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5488 struct btrfs_root *root,
5489 struct btrfs_path *path,
5490 struct walk_control *wc)
5492 int level = wc->level;
5493 int lookup_info = 1;
5496 while (level >= 0) {
5497 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5504 if (path->slots[level] >=
5505 btrfs_header_nritems(path->nodes[level]))
5508 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5510 path->slots[level]++;
5519 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5520 struct btrfs_root *root,
5521 struct btrfs_path *path,
5522 struct walk_control *wc, int max_level)
5524 int level = wc->level;
5527 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5528 while (level < max_level && path->nodes[level]) {
5530 if (path->slots[level] + 1 <
5531 btrfs_header_nritems(path->nodes[level])) {
5532 path->slots[level]++;
5535 ret = walk_up_proc(trans, root, path, wc);
5541 if (path->locks[level]) {
5542 btrfs_tree_unlock_rw(path->nodes[level],
5543 path->locks[level]);
5544 path->locks[level] = 0;
5546 free_extent_buffer(path->nodes[level]);
5547 path->nodes[level] = NULL;
5555 * drop a subvolume tree.
5557 * this function traverses the tree freeing any blocks that only
5558 * referenced by the tree.
5560 * when a shared tree block is found. this function decreases its
5561 * reference count by one. if update_ref is true, this function
5562 * also make sure backrefs for the shared block and all lower level
5563 * blocks are properly updated.
5565 * If called with for_reloc == 0, may exit early with -EAGAIN
5567 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
5569 struct btrfs_fs_info *fs_info = root->fs_info;
5570 struct btrfs_path *path;
5571 struct btrfs_trans_handle *trans;
5572 struct btrfs_root *tree_root = fs_info->tree_root;
5573 struct btrfs_root_item *root_item = &root->root_item;
5574 struct walk_control *wc;
5575 struct btrfs_key key;
5579 bool root_dropped = false;
5581 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5583 path = btrfs_alloc_path();
5589 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5591 btrfs_free_path(path);
5597 * Use join to avoid potential EINTR from transaction start. See
5598 * wait_reserve_ticket and the whole reservation callchain.
5601 trans = btrfs_join_transaction(tree_root);
5603 trans = btrfs_start_transaction(tree_root, 0);
5604 if (IS_ERR(trans)) {
5605 err = PTR_ERR(trans);
5609 err = btrfs_run_delayed_items(trans);
5614 * This will help us catch people modifying the fs tree while we're
5615 * dropping it. It is unsafe to mess with the fs tree while it's being
5616 * dropped as we unlock the root node and parent nodes as we walk down
5617 * the tree, assuming nothing will change. If something does change
5618 * then we'll have stale information and drop references to blocks we've
5621 set_bit(BTRFS_ROOT_DELETING, &root->state);
5622 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5623 level = btrfs_header_level(root->node);
5624 path->nodes[level] = btrfs_lock_root_node(root);
5625 path->slots[level] = 0;
5626 path->locks[level] = BTRFS_WRITE_LOCK;
5627 memset(&wc->update_progress, 0,
5628 sizeof(wc->update_progress));
5630 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5631 memcpy(&wc->update_progress, &key,
5632 sizeof(wc->update_progress));
5634 level = btrfs_root_drop_level(root_item);
5636 path->lowest_level = level;
5637 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5638 path->lowest_level = 0;
5646 * unlock our path, this is safe because only this
5647 * function is allowed to delete this snapshot
5649 btrfs_unlock_up_safe(path, 0);
5651 level = btrfs_header_level(root->node);
5653 btrfs_tree_lock(path->nodes[level]);
5654 path->locks[level] = BTRFS_WRITE_LOCK;
5656 ret = btrfs_lookup_extent_info(trans, fs_info,
5657 path->nodes[level]->start,
5658 level, 1, &wc->refs[level],
5664 BUG_ON(wc->refs[level] == 0);
5666 if (level == btrfs_root_drop_level(root_item))
5669 btrfs_tree_unlock(path->nodes[level]);
5670 path->locks[level] = 0;
5671 WARN_ON(wc->refs[level] != 1);
5676 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5678 wc->shared_level = -1;
5679 wc->stage = DROP_REFERENCE;
5680 wc->update_ref = update_ref;
5682 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5686 ret = walk_down_tree(trans, root, path, wc);
5692 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5699 BUG_ON(wc->stage != DROP_REFERENCE);
5703 if (wc->stage == DROP_REFERENCE) {
5704 wc->drop_level = wc->level;
5705 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5707 path->slots[wc->drop_level]);
5709 btrfs_cpu_key_to_disk(&root_item->drop_progress,
5710 &wc->drop_progress);
5711 btrfs_set_root_drop_level(root_item, wc->drop_level);
5713 BUG_ON(wc->level == 0);
5714 if (btrfs_should_end_transaction(trans) ||
5715 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5716 ret = btrfs_update_root(trans, tree_root,
5720 btrfs_abort_transaction(trans, ret);
5725 btrfs_end_transaction_throttle(trans);
5726 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5727 btrfs_debug(fs_info,
5728 "drop snapshot early exit");
5734 * Use join to avoid potential EINTR from transaction
5735 * start. See wait_reserve_ticket and the whole
5736 * reservation callchain.
5739 trans = btrfs_join_transaction(tree_root);
5741 trans = btrfs_start_transaction(tree_root, 0);
5742 if (IS_ERR(trans)) {
5743 err = PTR_ERR(trans);
5748 btrfs_release_path(path);
5752 ret = btrfs_del_root(trans, &root->root_key);
5754 btrfs_abort_transaction(trans, ret);
5759 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5760 ret = btrfs_find_root(tree_root, &root->root_key, path,
5763 btrfs_abort_transaction(trans, ret);
5766 } else if (ret > 0) {
5767 /* if we fail to delete the orphan item this time
5768 * around, it'll get picked up the next time.
5770 * The most common failure here is just -ENOENT.
5772 btrfs_del_orphan_item(trans, tree_root,
5773 root->root_key.objectid);
5778 * This subvolume is going to be completely dropped, and won't be
5779 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
5780 * commit transaction time. So free it here manually.
5782 btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
5783 btrfs_qgroup_free_meta_all_pertrans(root);
5785 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state))
5786 btrfs_add_dropped_root(trans, root);
5788 btrfs_put_root(root);
5789 root_dropped = true;
5791 btrfs_end_transaction_throttle(trans);
5794 btrfs_free_path(path);
5797 * So if we need to stop dropping the snapshot for whatever reason we
5798 * need to make sure to add it back to the dead root list so that we
5799 * keep trying to do the work later. This also cleans up roots if we
5800 * don't have it in the radix (like when we recover after a power fail
5801 * or unmount) so we don't leak memory.
5803 if (!for_reloc && !root_dropped)
5804 btrfs_add_dead_root(root);
5809 * drop subtree rooted at tree block 'node'.
5811 * NOTE: this function will unlock and release tree block 'node'
5812 * only used by relocation code
5814 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5815 struct btrfs_root *root,
5816 struct extent_buffer *node,
5817 struct extent_buffer *parent)
5819 struct btrfs_fs_info *fs_info = root->fs_info;
5820 struct btrfs_path *path;
5821 struct walk_control *wc;
5827 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5829 path = btrfs_alloc_path();
5833 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5835 btrfs_free_path(path);
5839 btrfs_assert_tree_write_locked(parent);
5840 parent_level = btrfs_header_level(parent);
5841 atomic_inc(&parent->refs);
5842 path->nodes[parent_level] = parent;
5843 path->slots[parent_level] = btrfs_header_nritems(parent);
5845 btrfs_assert_tree_write_locked(node);
5846 level = btrfs_header_level(node);
5847 path->nodes[level] = node;
5848 path->slots[level] = 0;
5849 path->locks[level] = BTRFS_WRITE_LOCK;
5851 wc->refs[parent_level] = 1;
5852 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5854 wc->shared_level = -1;
5855 wc->stage = DROP_REFERENCE;
5858 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5861 wret = walk_down_tree(trans, root, path, wc);
5867 wret = walk_up_tree(trans, root, path, wc, parent_level);
5875 btrfs_free_path(path);
5880 * helper to account the unused space of all the readonly block group in the
5881 * space_info. takes mirrors into account.
5883 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
5885 struct btrfs_block_group *block_group;
5889 /* It's df, we don't care if it's racy */
5890 if (list_empty(&sinfo->ro_bgs))
5893 spin_lock(&sinfo->lock);
5894 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
5895 spin_lock(&block_group->lock);
5897 if (!block_group->ro) {
5898 spin_unlock(&block_group->lock);
5902 factor = btrfs_bg_type_to_factor(block_group->flags);
5903 free_bytes += (block_group->length -
5904 block_group->used) * factor;
5906 spin_unlock(&block_group->lock);
5908 spin_unlock(&sinfo->lock);
5913 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
5916 return unpin_extent_range(fs_info, start, end, false);
5920 * It used to be that old block groups would be left around forever.
5921 * Iterating over them would be enough to trim unused space. Since we
5922 * now automatically remove them, we also need to iterate over unallocated
5925 * We don't want a transaction for this since the discard may take a
5926 * substantial amount of time. We don't require that a transaction be
5927 * running, but we do need to take a running transaction into account
5928 * to ensure that we're not discarding chunks that were released or
5929 * allocated in the current transaction.
5931 * Holding the chunks lock will prevent other threads from allocating
5932 * or releasing chunks, but it won't prevent a running transaction
5933 * from committing and releasing the memory that the pending chunks
5934 * list head uses. For that, we need to take a reference to the
5935 * transaction and hold the commit root sem. We only need to hold
5936 * it while performing the free space search since we have already
5937 * held back allocations.
5939 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
5941 u64 start = SZ_1M, len = 0, end = 0;
5946 /* Discard not supported = nothing to do. */
5947 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
5950 /* Not writable = nothing to do. */
5951 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5954 /* No free space = nothing to do. */
5955 if (device->total_bytes <= device->bytes_used)
5961 struct btrfs_fs_info *fs_info = device->fs_info;
5964 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
5968 find_first_clear_extent_bit(&device->alloc_state, start,
5970 CHUNK_TRIMMED | CHUNK_ALLOCATED);
5972 /* Check if there are any CHUNK_* bits left */
5973 if (start > device->total_bytes) {
5974 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
5975 btrfs_warn_in_rcu(fs_info,
5976 "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
5977 start, end - start + 1,
5978 rcu_str_deref(device->name),
5979 device->total_bytes);
5980 mutex_unlock(&fs_info->chunk_mutex);
5985 /* Ensure we skip the reserved area in the first 1M */
5986 start = max_t(u64, start, SZ_1M);
5989 * If find_first_clear_extent_bit find a range that spans the
5990 * end of the device it will set end to -1, in this case it's up
5991 * to the caller to trim the value to the size of the device.
5993 end = min(end, device->total_bytes - 1);
5995 len = end - start + 1;
5997 /* We didn't find any extents */
5999 mutex_unlock(&fs_info->chunk_mutex);
6004 ret = btrfs_issue_discard(device->bdev, start, len,
6007 set_extent_bits(&device->alloc_state, start,
6010 mutex_unlock(&fs_info->chunk_mutex);
6018 if (fatal_signal_pending(current)) {
6030 * Trim the whole filesystem by:
6031 * 1) trimming the free space in each block group
6032 * 2) trimming the unallocated space on each device
6034 * This will also continue trimming even if a block group or device encounters
6035 * an error. The return value will be the last error, or 0 if nothing bad
6038 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
6040 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6041 struct btrfs_block_group *cache = NULL;
6042 struct btrfs_device *device;
6044 u64 range_end = U64_MAX;
6055 * Check range overflow if range->len is set.
6056 * The default range->len is U64_MAX.
6058 if (range->len != U64_MAX &&
6059 check_add_overflow(range->start, range->len, &range_end))
6062 cache = btrfs_lookup_first_block_group(fs_info, range->start);
6063 for (; cache; cache = btrfs_next_block_group(cache)) {
6064 if (cache->start >= range_end) {
6065 btrfs_put_block_group(cache);
6069 start = max(range->start, cache->start);
6070 end = min(range_end, cache->start + cache->length);
6072 if (end - start >= range->minlen) {
6073 if (!btrfs_block_group_done(cache)) {
6074 ret = btrfs_cache_block_group(cache, 0);
6080 ret = btrfs_wait_block_group_cache_done(cache);
6087 ret = btrfs_trim_block_group(cache,
6093 trimmed += group_trimmed;
6104 "failed to trim %llu block group(s), last error %d",
6107 mutex_lock(&fs_devices->device_list_mutex);
6108 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6109 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
6112 ret = btrfs_trim_free_extents(device, &group_trimmed);
6119 trimmed += group_trimmed;
6121 mutex_unlock(&fs_devices->device_list_mutex);
6125 "failed to trim %llu device(s), last error %d",
6126 dev_failed, dev_ret);
6127 range->len = trimmed;