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_bio_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_bio *bbio = NULL;
1320 * Avoid races with device replace and make sure our bbio has devices
1321 * associated to its stripes that don't go away while we are discarding.
1323 btrfs_bio_counter_inc_blocked(fs_info);
1325 struct btrfs_bio_stripe *stripe;
1328 num_bytes = end - cur;
1329 /* Tell the block device(s) that the sectors can be discarded */
1330 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur,
1331 &num_bytes, &bbio, 0);
1333 * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or
1334 * -EOPNOTSUPP. For any such error, @num_bytes is not updated,
1335 * thus we can't continue anyway.
1340 stripe = bbio->stripes;
1341 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1343 struct btrfs_device *device = stripe->dev;
1345 if (!device->bdev) {
1346 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1350 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
1353 ret = do_discard_extent(stripe, &bytes);
1355 discarded_bytes += bytes;
1356 } else if (ret != -EOPNOTSUPP) {
1358 * Logic errors or -ENOMEM, or -EIO, but
1359 * unlikely to happen.
1361 * And since there are two loops, explicitly
1362 * go to out to avoid confusion.
1364 btrfs_put_bbio(bbio);
1369 * Just in case we get back EOPNOTSUPP for some reason,
1370 * just ignore the return value so we don't screw up
1371 * people calling discard_extent.
1375 btrfs_put_bbio(bbio);
1379 btrfs_bio_counter_dec(fs_info);
1382 *actual_bytes = discarded_bytes;
1385 if (ret == -EOPNOTSUPP)
1390 /* Can return -ENOMEM */
1391 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1392 struct btrfs_ref *generic_ref)
1394 struct btrfs_fs_info *fs_info = trans->fs_info;
1397 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1398 generic_ref->action);
1399 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1400 generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID);
1402 if (generic_ref->type == BTRFS_REF_METADATA)
1403 ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL);
1405 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0);
1407 btrfs_ref_tree_mod(fs_info, generic_ref);
1413 * __btrfs_inc_extent_ref - insert backreference for a given extent
1415 * The counterpart is in __btrfs_free_extent(), with examples and more details
1418 * @trans: Handle of transaction
1420 * @node: The delayed ref node used to get the bytenr/length for
1421 * extent whose references are incremented.
1423 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1424 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1425 * bytenr of the parent block. Since new extents are always
1426 * created with indirect references, this will only be the case
1427 * when relocating a shared extent. In that case, root_objectid
1428 * will be BTRFS_TREE_RELOC_OBJECTID. Otherwise, parent must
1431 * @root_objectid: The id of the root where this modification has originated,
1432 * this can be either one of the well-known metadata trees or
1433 * the subvolume id which references this extent.
1435 * @owner: For data extents it is the inode number of the owning file.
1436 * For metadata extents this parameter holds the level in the
1437 * tree of the extent.
1439 * @offset: For metadata extents the offset is ignored and is currently
1440 * always passed as 0. For data extents it is the fileoffset
1441 * this extent belongs to.
1443 * @refs_to_add Number of references to add
1445 * @extent_op Pointer to a structure, holding information necessary when
1446 * updating a tree block's flags
1449 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1450 struct btrfs_delayed_ref_node *node,
1451 u64 parent, u64 root_objectid,
1452 u64 owner, u64 offset, int refs_to_add,
1453 struct btrfs_delayed_extent_op *extent_op)
1455 struct btrfs_path *path;
1456 struct extent_buffer *leaf;
1457 struct btrfs_extent_item *item;
1458 struct btrfs_key key;
1459 u64 bytenr = node->bytenr;
1460 u64 num_bytes = node->num_bytes;
1464 path = btrfs_alloc_path();
1468 /* this will setup the path even if it fails to insert the back ref */
1469 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1470 parent, root_objectid, owner,
1471 offset, refs_to_add, extent_op);
1472 if ((ret < 0 && ret != -EAGAIN) || !ret)
1476 * Ok we had -EAGAIN which means we didn't have space to insert and
1477 * inline extent ref, so just update the reference count and add a
1480 leaf = path->nodes[0];
1481 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1482 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1483 refs = btrfs_extent_refs(leaf, item);
1484 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1486 __run_delayed_extent_op(extent_op, leaf, item);
1488 btrfs_mark_buffer_dirty(leaf);
1489 btrfs_release_path(path);
1491 /* now insert the actual backref */
1492 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1493 BUG_ON(refs_to_add != 1);
1494 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1497 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1498 root_objectid, owner, offset,
1502 btrfs_abort_transaction(trans, ret);
1504 btrfs_free_path(path);
1508 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1509 struct btrfs_delayed_ref_node *node,
1510 struct btrfs_delayed_extent_op *extent_op,
1511 int insert_reserved)
1514 struct btrfs_delayed_data_ref *ref;
1515 struct btrfs_key ins;
1520 ins.objectid = node->bytenr;
1521 ins.offset = node->num_bytes;
1522 ins.type = BTRFS_EXTENT_ITEM_KEY;
1524 ref = btrfs_delayed_node_to_data_ref(node);
1525 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
1527 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1528 parent = ref->parent;
1529 ref_root = ref->root;
1531 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1533 flags |= extent_op->flags_to_set;
1534 ret = alloc_reserved_file_extent(trans, parent, ref_root,
1535 flags, ref->objectid,
1538 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1539 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1540 ref->objectid, ref->offset,
1541 node->ref_mod, extent_op);
1542 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1543 ret = __btrfs_free_extent(trans, node, parent,
1544 ref_root, ref->objectid,
1545 ref->offset, node->ref_mod,
1553 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1554 struct extent_buffer *leaf,
1555 struct btrfs_extent_item *ei)
1557 u64 flags = btrfs_extent_flags(leaf, ei);
1558 if (extent_op->update_flags) {
1559 flags |= extent_op->flags_to_set;
1560 btrfs_set_extent_flags(leaf, ei, flags);
1563 if (extent_op->update_key) {
1564 struct btrfs_tree_block_info *bi;
1565 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1566 bi = (struct btrfs_tree_block_info *)(ei + 1);
1567 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1571 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1572 struct btrfs_delayed_ref_head *head,
1573 struct btrfs_delayed_extent_op *extent_op)
1575 struct btrfs_fs_info *fs_info = trans->fs_info;
1576 struct btrfs_key key;
1577 struct btrfs_path *path;
1578 struct btrfs_extent_item *ei;
1579 struct extent_buffer *leaf;
1583 int metadata = !extent_op->is_data;
1585 if (TRANS_ABORTED(trans))
1588 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1591 path = btrfs_alloc_path();
1595 key.objectid = head->bytenr;
1598 key.type = BTRFS_METADATA_ITEM_KEY;
1599 key.offset = extent_op->level;
1601 key.type = BTRFS_EXTENT_ITEM_KEY;
1602 key.offset = head->num_bytes;
1606 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
1613 if (path->slots[0] > 0) {
1615 btrfs_item_key_to_cpu(path->nodes[0], &key,
1617 if (key.objectid == head->bytenr &&
1618 key.type == BTRFS_EXTENT_ITEM_KEY &&
1619 key.offset == head->num_bytes)
1623 btrfs_release_path(path);
1626 key.objectid = head->bytenr;
1627 key.offset = head->num_bytes;
1628 key.type = BTRFS_EXTENT_ITEM_KEY;
1637 leaf = path->nodes[0];
1638 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1640 if (unlikely(item_size < sizeof(*ei))) {
1642 btrfs_print_v0_err(fs_info);
1643 btrfs_abort_transaction(trans, err);
1647 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1648 __run_delayed_extent_op(extent_op, leaf, ei);
1650 btrfs_mark_buffer_dirty(leaf);
1652 btrfs_free_path(path);
1656 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1657 struct btrfs_delayed_ref_node *node,
1658 struct btrfs_delayed_extent_op *extent_op,
1659 int insert_reserved)
1662 struct btrfs_delayed_tree_ref *ref;
1666 ref = btrfs_delayed_node_to_tree_ref(node);
1667 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1669 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1670 parent = ref->parent;
1671 ref_root = ref->root;
1673 if (node->ref_mod != 1) {
1674 btrfs_err(trans->fs_info,
1675 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
1676 node->bytenr, node->ref_mod, node->action, ref_root,
1680 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1681 BUG_ON(!extent_op || !extent_op->update_flags);
1682 ret = alloc_reserved_tree_block(trans, node, extent_op);
1683 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1684 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1685 ref->level, 0, 1, extent_op);
1686 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1687 ret = __btrfs_free_extent(trans, node, parent, ref_root,
1688 ref->level, 0, 1, extent_op);
1695 /* helper function to actually process a single delayed ref entry */
1696 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1697 struct btrfs_delayed_ref_node *node,
1698 struct btrfs_delayed_extent_op *extent_op,
1699 int insert_reserved)
1703 if (TRANS_ABORTED(trans)) {
1704 if (insert_reserved)
1705 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1709 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1710 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1711 ret = run_delayed_tree_ref(trans, node, extent_op,
1713 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1714 node->type == BTRFS_SHARED_DATA_REF_KEY)
1715 ret = run_delayed_data_ref(trans, node, extent_op,
1719 if (ret && insert_reserved)
1720 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1724 static inline struct btrfs_delayed_ref_node *
1725 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1727 struct btrfs_delayed_ref_node *ref;
1729 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1733 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1734 * This is to prevent a ref count from going down to zero, which deletes
1735 * the extent item from the extent tree, when there still are references
1736 * to add, which would fail because they would not find the extent item.
1738 if (!list_empty(&head->ref_add_list))
1739 return list_first_entry(&head->ref_add_list,
1740 struct btrfs_delayed_ref_node, add_list);
1742 ref = rb_entry(rb_first_cached(&head->ref_tree),
1743 struct btrfs_delayed_ref_node, ref_node);
1744 ASSERT(list_empty(&ref->add_list));
1748 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1749 struct btrfs_delayed_ref_head *head)
1751 spin_lock(&delayed_refs->lock);
1752 head->processing = 0;
1753 delayed_refs->num_heads_ready++;
1754 spin_unlock(&delayed_refs->lock);
1755 btrfs_delayed_ref_unlock(head);
1758 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1759 struct btrfs_delayed_ref_head *head)
1761 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1766 if (head->must_insert_reserved) {
1767 head->extent_op = NULL;
1768 btrfs_free_delayed_extent_op(extent_op);
1774 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1775 struct btrfs_delayed_ref_head *head)
1777 struct btrfs_delayed_extent_op *extent_op;
1780 extent_op = cleanup_extent_op(head);
1783 head->extent_op = NULL;
1784 spin_unlock(&head->lock);
1785 ret = run_delayed_extent_op(trans, head, extent_op);
1786 btrfs_free_delayed_extent_op(extent_op);
1787 return ret ? ret : 1;
1790 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1791 struct btrfs_delayed_ref_root *delayed_refs,
1792 struct btrfs_delayed_ref_head *head)
1794 int nr_items = 1; /* Dropping this ref head update. */
1797 * We had csum deletions accounted for in our delayed refs rsv, we need
1798 * to drop the csum leaves for this update from our delayed_refs_rsv.
1800 if (head->total_ref_mod < 0 && head->is_data) {
1801 spin_lock(&delayed_refs->lock);
1802 delayed_refs->pending_csums -= head->num_bytes;
1803 spin_unlock(&delayed_refs->lock);
1804 nr_items += btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes);
1807 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
1810 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1811 struct btrfs_delayed_ref_head *head)
1814 struct btrfs_fs_info *fs_info = trans->fs_info;
1815 struct btrfs_delayed_ref_root *delayed_refs;
1818 delayed_refs = &trans->transaction->delayed_refs;
1820 ret = run_and_cleanup_extent_op(trans, head);
1822 unselect_delayed_ref_head(delayed_refs, head);
1823 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1830 * Need to drop our head ref lock and re-acquire the delayed ref lock
1831 * and then re-check to make sure nobody got added.
1833 spin_unlock(&head->lock);
1834 spin_lock(&delayed_refs->lock);
1835 spin_lock(&head->lock);
1836 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1837 spin_unlock(&head->lock);
1838 spin_unlock(&delayed_refs->lock);
1841 btrfs_delete_ref_head(delayed_refs, head);
1842 spin_unlock(&head->lock);
1843 spin_unlock(&delayed_refs->lock);
1845 if (head->must_insert_reserved) {
1846 btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
1847 if (head->is_data) {
1848 ret = btrfs_del_csums(trans, fs_info->csum_root,
1849 head->bytenr, head->num_bytes);
1853 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1855 trace_run_delayed_ref_head(fs_info, head, 0);
1856 btrfs_delayed_ref_unlock(head);
1857 btrfs_put_delayed_ref_head(head);
1861 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1862 struct btrfs_trans_handle *trans)
1864 struct btrfs_delayed_ref_root *delayed_refs =
1865 &trans->transaction->delayed_refs;
1866 struct btrfs_delayed_ref_head *head = NULL;
1869 spin_lock(&delayed_refs->lock);
1870 head = btrfs_select_ref_head(delayed_refs);
1872 spin_unlock(&delayed_refs->lock);
1877 * Grab the lock that says we are going to process all the refs for
1880 ret = btrfs_delayed_ref_lock(delayed_refs, head);
1881 spin_unlock(&delayed_refs->lock);
1884 * We may have dropped the spin lock to get the head mutex lock, and
1885 * that might have given someone else time to free the head. If that's
1886 * true, it has been removed from our list and we can move on.
1889 head = ERR_PTR(-EAGAIN);
1894 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1895 struct btrfs_delayed_ref_head *locked_ref,
1896 unsigned long *run_refs)
1898 struct btrfs_fs_info *fs_info = trans->fs_info;
1899 struct btrfs_delayed_ref_root *delayed_refs;
1900 struct btrfs_delayed_extent_op *extent_op;
1901 struct btrfs_delayed_ref_node *ref;
1902 int must_insert_reserved = 0;
1905 delayed_refs = &trans->transaction->delayed_refs;
1907 lockdep_assert_held(&locked_ref->mutex);
1908 lockdep_assert_held(&locked_ref->lock);
1910 while ((ref = select_delayed_ref(locked_ref))) {
1912 btrfs_check_delayed_seq(fs_info, ref->seq)) {
1913 spin_unlock(&locked_ref->lock);
1914 unselect_delayed_ref_head(delayed_refs, locked_ref);
1920 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1921 RB_CLEAR_NODE(&ref->ref_node);
1922 if (!list_empty(&ref->add_list))
1923 list_del(&ref->add_list);
1925 * When we play the delayed ref, also correct the ref_mod on
1928 switch (ref->action) {
1929 case BTRFS_ADD_DELAYED_REF:
1930 case BTRFS_ADD_DELAYED_EXTENT:
1931 locked_ref->ref_mod -= ref->ref_mod;
1933 case BTRFS_DROP_DELAYED_REF:
1934 locked_ref->ref_mod += ref->ref_mod;
1939 atomic_dec(&delayed_refs->num_entries);
1942 * Record the must_insert_reserved flag before we drop the
1945 must_insert_reserved = locked_ref->must_insert_reserved;
1946 locked_ref->must_insert_reserved = 0;
1948 extent_op = locked_ref->extent_op;
1949 locked_ref->extent_op = NULL;
1950 spin_unlock(&locked_ref->lock);
1952 ret = run_one_delayed_ref(trans, ref, extent_op,
1953 must_insert_reserved);
1955 btrfs_free_delayed_extent_op(extent_op);
1957 unselect_delayed_ref_head(delayed_refs, locked_ref);
1958 btrfs_put_delayed_ref(ref);
1959 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
1964 btrfs_put_delayed_ref(ref);
1967 spin_lock(&locked_ref->lock);
1968 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
1975 * Returns 0 on success or if called with an already aborted transaction.
1976 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
1978 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1981 struct btrfs_fs_info *fs_info = trans->fs_info;
1982 struct btrfs_delayed_ref_root *delayed_refs;
1983 struct btrfs_delayed_ref_head *locked_ref = NULL;
1984 ktime_t start = ktime_get();
1986 unsigned long count = 0;
1987 unsigned long actual_count = 0;
1989 delayed_refs = &trans->transaction->delayed_refs;
1992 locked_ref = btrfs_obtain_ref_head(trans);
1993 if (IS_ERR_OR_NULL(locked_ref)) {
1994 if (PTR_ERR(locked_ref) == -EAGAIN) {
2003 * We need to try and merge add/drops of the same ref since we
2004 * can run into issues with relocate dropping the implicit ref
2005 * and then it being added back again before the drop can
2006 * finish. If we merged anything we need to re-loop so we can
2008 * Or we can get node references of the same type that weren't
2009 * merged when created due to bumps in the tree mod seq, and
2010 * we need to merge them to prevent adding an inline extent
2011 * backref before dropping it (triggering a BUG_ON at
2012 * insert_inline_extent_backref()).
2014 spin_lock(&locked_ref->lock);
2015 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2017 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2019 if (ret < 0 && ret != -EAGAIN) {
2021 * Error, btrfs_run_delayed_refs_for_head already
2022 * unlocked everything so just bail out
2027 * Success, perform the usual cleanup of a processed
2030 ret = cleanup_ref_head(trans, locked_ref);
2032 /* We dropped our lock, we need to loop. */
2041 * Either success case or btrfs_run_delayed_refs_for_head
2042 * returned -EAGAIN, meaning we need to select another head
2047 } while ((nr != -1 && count < nr) || locked_ref);
2050 * We don't want to include ref heads since we can have empty ref heads
2051 * and those will drastically skew our runtime down since we just do
2052 * accounting, no actual extent tree updates.
2054 if (actual_count > 0) {
2055 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2059 * We weigh the current average higher than our current runtime
2060 * to avoid large swings in the average.
2062 spin_lock(&delayed_refs->lock);
2063 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2064 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2065 spin_unlock(&delayed_refs->lock);
2070 #ifdef SCRAMBLE_DELAYED_REFS
2072 * Normally delayed refs get processed in ascending bytenr order. This
2073 * correlates in most cases to the order added. To expose dependencies on this
2074 * order, we start to process the tree in the middle instead of the beginning
2076 static u64 find_middle(struct rb_root *root)
2078 struct rb_node *n = root->rb_node;
2079 struct btrfs_delayed_ref_node *entry;
2082 u64 first = 0, last = 0;
2086 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2087 first = entry->bytenr;
2091 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2092 last = entry->bytenr;
2097 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2098 WARN_ON(!entry->in_tree);
2100 middle = entry->bytenr;
2114 * this starts processing the delayed reference count updates and
2115 * extent insertions we have queued up so far. count can be
2116 * 0, which means to process everything in the tree at the start
2117 * of the run (but not newly added entries), or it can be some target
2118 * number you'd like to process.
2120 * Returns 0 on success or if called with an aborted transaction
2121 * Returns <0 on error and aborts the transaction
2123 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2124 unsigned long count)
2126 struct btrfs_fs_info *fs_info = trans->fs_info;
2127 struct rb_node *node;
2128 struct btrfs_delayed_ref_root *delayed_refs;
2129 struct btrfs_delayed_ref_head *head;
2131 int run_all = count == (unsigned long)-1;
2133 /* We'll clean this up in btrfs_cleanup_transaction */
2134 if (TRANS_ABORTED(trans))
2137 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2140 delayed_refs = &trans->transaction->delayed_refs;
2142 count = delayed_refs->num_heads_ready;
2145 #ifdef SCRAMBLE_DELAYED_REFS
2146 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2148 ret = __btrfs_run_delayed_refs(trans, count);
2150 btrfs_abort_transaction(trans, ret);
2155 btrfs_create_pending_block_groups(trans);
2157 spin_lock(&delayed_refs->lock);
2158 node = rb_first_cached(&delayed_refs->href_root);
2160 spin_unlock(&delayed_refs->lock);
2163 head = rb_entry(node, struct btrfs_delayed_ref_head,
2165 refcount_inc(&head->refs);
2166 spin_unlock(&delayed_refs->lock);
2168 /* Mutex was contended, block until it's released and retry. */
2169 mutex_lock(&head->mutex);
2170 mutex_unlock(&head->mutex);
2172 btrfs_put_delayed_ref_head(head);
2180 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2181 struct extent_buffer *eb, u64 flags,
2182 int level, int is_data)
2184 struct btrfs_delayed_extent_op *extent_op;
2187 extent_op = btrfs_alloc_delayed_extent_op();
2191 extent_op->flags_to_set = flags;
2192 extent_op->update_flags = true;
2193 extent_op->update_key = false;
2194 extent_op->is_data = is_data ? true : false;
2195 extent_op->level = level;
2197 ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op);
2199 btrfs_free_delayed_extent_op(extent_op);
2203 static noinline int check_delayed_ref(struct btrfs_root *root,
2204 struct btrfs_path *path,
2205 u64 objectid, u64 offset, u64 bytenr)
2207 struct btrfs_delayed_ref_head *head;
2208 struct btrfs_delayed_ref_node *ref;
2209 struct btrfs_delayed_data_ref *data_ref;
2210 struct btrfs_delayed_ref_root *delayed_refs;
2211 struct btrfs_transaction *cur_trans;
2212 struct rb_node *node;
2215 spin_lock(&root->fs_info->trans_lock);
2216 cur_trans = root->fs_info->running_transaction;
2218 refcount_inc(&cur_trans->use_count);
2219 spin_unlock(&root->fs_info->trans_lock);
2223 delayed_refs = &cur_trans->delayed_refs;
2224 spin_lock(&delayed_refs->lock);
2225 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2227 spin_unlock(&delayed_refs->lock);
2228 btrfs_put_transaction(cur_trans);
2232 if (!mutex_trylock(&head->mutex)) {
2233 refcount_inc(&head->refs);
2234 spin_unlock(&delayed_refs->lock);
2236 btrfs_release_path(path);
2239 * Mutex was contended, block until it's released and let
2242 mutex_lock(&head->mutex);
2243 mutex_unlock(&head->mutex);
2244 btrfs_put_delayed_ref_head(head);
2245 btrfs_put_transaction(cur_trans);
2248 spin_unlock(&delayed_refs->lock);
2250 spin_lock(&head->lock);
2252 * XXX: We should replace this with a proper search function in the
2255 for (node = rb_first_cached(&head->ref_tree); node;
2256 node = rb_next(node)) {
2257 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2258 /* If it's a shared ref we know a cross reference exists */
2259 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2264 data_ref = btrfs_delayed_node_to_data_ref(ref);
2267 * If our ref doesn't match the one we're currently looking at
2268 * then we have a cross reference.
2270 if (data_ref->root != root->root_key.objectid ||
2271 data_ref->objectid != objectid ||
2272 data_ref->offset != offset) {
2277 spin_unlock(&head->lock);
2278 mutex_unlock(&head->mutex);
2279 btrfs_put_transaction(cur_trans);
2283 static noinline int check_committed_ref(struct btrfs_root *root,
2284 struct btrfs_path *path,
2285 u64 objectid, u64 offset, u64 bytenr,
2288 struct btrfs_fs_info *fs_info = root->fs_info;
2289 struct btrfs_root *extent_root = fs_info->extent_root;
2290 struct extent_buffer *leaf;
2291 struct btrfs_extent_data_ref *ref;
2292 struct btrfs_extent_inline_ref *iref;
2293 struct btrfs_extent_item *ei;
2294 struct btrfs_key key;
2299 key.objectid = bytenr;
2300 key.offset = (u64)-1;
2301 key.type = BTRFS_EXTENT_ITEM_KEY;
2303 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2306 BUG_ON(ret == 0); /* Corruption */
2309 if (path->slots[0] == 0)
2313 leaf = path->nodes[0];
2314 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2316 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2320 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2321 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2323 /* If extent item has more than 1 inline ref then it's shared */
2324 if (item_size != sizeof(*ei) +
2325 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2329 * If extent created before last snapshot => it's shared unless the
2330 * snapshot has been deleted. Use the heuristic if strict is false.
2333 (btrfs_extent_generation(leaf, ei) <=
2334 btrfs_root_last_snapshot(&root->root_item)))
2337 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2339 /* If this extent has SHARED_DATA_REF then it's shared */
2340 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2341 if (type != BTRFS_EXTENT_DATA_REF_KEY)
2344 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2345 if (btrfs_extent_refs(leaf, ei) !=
2346 btrfs_extent_data_ref_count(leaf, ref) ||
2347 btrfs_extent_data_ref_root(leaf, ref) !=
2348 root->root_key.objectid ||
2349 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2350 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2358 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2359 u64 bytenr, bool strict)
2361 struct btrfs_path *path;
2364 path = btrfs_alloc_path();
2369 ret = check_committed_ref(root, path, objectid,
2370 offset, bytenr, strict);
2371 if (ret && ret != -ENOENT)
2374 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2375 } while (ret == -EAGAIN);
2378 btrfs_free_path(path);
2379 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2384 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2385 struct btrfs_root *root,
2386 struct extent_buffer *buf,
2387 int full_backref, int inc)
2389 struct btrfs_fs_info *fs_info = root->fs_info;
2395 struct btrfs_key key;
2396 struct btrfs_file_extent_item *fi;
2397 struct btrfs_ref generic_ref = { 0 };
2398 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2404 if (btrfs_is_testing(fs_info))
2407 ref_root = btrfs_header_owner(buf);
2408 nritems = btrfs_header_nritems(buf);
2409 level = btrfs_header_level(buf);
2411 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
2415 parent = buf->start;
2419 action = BTRFS_ADD_DELAYED_REF;
2421 action = BTRFS_DROP_DELAYED_REF;
2423 for (i = 0; i < nritems; i++) {
2425 btrfs_item_key_to_cpu(buf, &key, i);
2426 if (key.type != BTRFS_EXTENT_DATA_KEY)
2428 fi = btrfs_item_ptr(buf, i,
2429 struct btrfs_file_extent_item);
2430 if (btrfs_file_extent_type(buf, fi) ==
2431 BTRFS_FILE_EXTENT_INLINE)
2433 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2437 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2438 key.offset -= btrfs_file_extent_offset(buf, fi);
2439 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2441 generic_ref.real_root = root->root_key.objectid;
2442 btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2444 generic_ref.skip_qgroup = for_reloc;
2446 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2448 ret = btrfs_free_extent(trans, &generic_ref);
2452 bytenr = btrfs_node_blockptr(buf, i);
2453 num_bytes = fs_info->nodesize;
2454 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2456 generic_ref.real_root = root->root_key.objectid;
2457 btrfs_init_tree_ref(&generic_ref, level - 1, ref_root);
2458 generic_ref.skip_qgroup = for_reloc;
2460 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2462 ret = btrfs_free_extent(trans, &generic_ref);
2472 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2473 struct extent_buffer *buf, int full_backref)
2475 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2478 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2479 struct extent_buffer *buf, int full_backref)
2481 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2484 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2486 struct btrfs_fs_info *fs_info = root->fs_info;
2491 flags = BTRFS_BLOCK_GROUP_DATA;
2492 else if (root == fs_info->chunk_root)
2493 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2495 flags = BTRFS_BLOCK_GROUP_METADATA;
2497 ret = btrfs_get_alloc_profile(fs_info, flags);
2501 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
2503 struct btrfs_block_group *cache;
2506 spin_lock(&fs_info->block_group_cache_lock);
2507 bytenr = fs_info->first_logical_byte;
2508 spin_unlock(&fs_info->block_group_cache_lock);
2510 if (bytenr < (u64)-1)
2513 cache = btrfs_lookup_first_block_group(fs_info, search_start);
2517 bytenr = cache->start;
2518 btrfs_put_block_group(cache);
2523 static int pin_down_extent(struct btrfs_trans_handle *trans,
2524 struct btrfs_block_group *cache,
2525 u64 bytenr, u64 num_bytes, int reserved)
2527 struct btrfs_fs_info *fs_info = cache->fs_info;
2529 spin_lock(&cache->space_info->lock);
2530 spin_lock(&cache->lock);
2531 cache->pinned += num_bytes;
2532 btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2535 cache->reserved -= num_bytes;
2536 cache->space_info->bytes_reserved -= num_bytes;
2538 spin_unlock(&cache->lock);
2539 spin_unlock(&cache->space_info->lock);
2541 set_extent_dirty(&trans->transaction->pinned_extents, bytenr,
2542 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
2546 int btrfs_pin_extent(struct btrfs_trans_handle *trans,
2547 u64 bytenr, u64 num_bytes, int reserved)
2549 struct btrfs_block_group *cache;
2551 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2552 BUG_ON(!cache); /* Logic error */
2554 pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
2556 btrfs_put_block_group(cache);
2561 * this function must be called within transaction
2563 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
2564 u64 bytenr, u64 num_bytes)
2566 struct btrfs_block_group *cache;
2569 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2574 * pull in the free space cache (if any) so that our pin
2575 * removes the free space from the cache. We have load_only set
2576 * to one because the slow code to read in the free extents does check
2577 * the pinned extents.
2579 btrfs_cache_block_group(cache, 1);
2581 * Make sure we wait until the cache is completely built in case it is
2582 * missing or is invalid and therefore needs to be rebuilt.
2584 ret = btrfs_wait_block_group_cache_done(cache);
2588 pin_down_extent(trans, cache, bytenr, num_bytes, 0);
2590 /* remove us from the free space cache (if we're there at all) */
2591 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
2593 btrfs_put_block_group(cache);
2597 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2598 u64 start, u64 num_bytes)
2601 struct btrfs_block_group *block_group;
2603 block_group = btrfs_lookup_block_group(fs_info, start);
2607 btrfs_cache_block_group(block_group, 1);
2609 * Make sure we wait until the cache is completely built in case it is
2610 * missing or is invalid and therefore needs to be rebuilt.
2612 ret = btrfs_wait_block_group_cache_done(block_group);
2616 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2618 btrfs_put_block_group(block_group);
2622 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2624 struct btrfs_fs_info *fs_info = eb->fs_info;
2625 struct btrfs_file_extent_item *item;
2626 struct btrfs_key key;
2631 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2634 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2635 btrfs_item_key_to_cpu(eb, &key, i);
2636 if (key.type != BTRFS_EXTENT_DATA_KEY)
2638 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2639 found_type = btrfs_file_extent_type(eb, item);
2640 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2642 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2644 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2645 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2646 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2655 btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2657 atomic_inc(&bg->reservations);
2661 * Returns the free cluster for the given space info and sets empty_cluster to
2662 * what it should be based on the mount options.
2664 static struct btrfs_free_cluster *
2665 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2666 struct btrfs_space_info *space_info, u64 *empty_cluster)
2668 struct btrfs_free_cluster *ret = NULL;
2671 if (btrfs_mixed_space_info(space_info))
2674 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2675 ret = &fs_info->meta_alloc_cluster;
2676 if (btrfs_test_opt(fs_info, SSD))
2677 *empty_cluster = SZ_2M;
2679 *empty_cluster = SZ_64K;
2680 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2681 btrfs_test_opt(fs_info, SSD_SPREAD)) {
2682 *empty_cluster = SZ_2M;
2683 ret = &fs_info->data_alloc_cluster;
2689 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2691 const bool return_free_space)
2693 struct btrfs_block_group *cache = NULL;
2694 struct btrfs_space_info *space_info;
2695 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2696 struct btrfs_free_cluster *cluster = NULL;
2698 u64 total_unpinned = 0;
2699 u64 empty_cluster = 0;
2702 while (start <= end) {
2705 start >= cache->start + cache->length) {
2707 btrfs_put_block_group(cache);
2709 cache = btrfs_lookup_block_group(fs_info, start);
2710 BUG_ON(!cache); /* Logic error */
2712 cluster = fetch_cluster_info(fs_info,
2715 empty_cluster <<= 1;
2718 len = cache->start + cache->length - start;
2719 len = min(len, end + 1 - start);
2721 down_read(&fs_info->commit_root_sem);
2722 if (start < cache->last_byte_to_unpin && return_free_space) {
2723 u64 add_len = min(len, cache->last_byte_to_unpin - start);
2725 btrfs_add_free_space(cache, start, add_len);
2727 up_read(&fs_info->commit_root_sem);
2730 total_unpinned += len;
2731 space_info = cache->space_info;
2734 * If this space cluster has been marked as fragmented and we've
2735 * unpinned enough in this block group to potentially allow a
2736 * cluster to be created inside of it go ahead and clear the
2739 if (cluster && cluster->fragmented &&
2740 total_unpinned > empty_cluster) {
2741 spin_lock(&cluster->lock);
2742 cluster->fragmented = 0;
2743 spin_unlock(&cluster->lock);
2746 spin_lock(&space_info->lock);
2747 spin_lock(&cache->lock);
2748 cache->pinned -= len;
2749 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2750 space_info->max_extent_size = 0;
2752 space_info->bytes_readonly += len;
2754 } else if (btrfs_is_zoned(fs_info)) {
2755 /* Need reset before reusing in a zoned block group */
2756 space_info->bytes_zone_unusable += len;
2759 spin_unlock(&cache->lock);
2760 if (!readonly && return_free_space &&
2761 global_rsv->space_info == space_info) {
2764 spin_lock(&global_rsv->lock);
2765 if (!global_rsv->full) {
2766 to_add = min(len, global_rsv->size -
2767 global_rsv->reserved);
2768 global_rsv->reserved += to_add;
2769 btrfs_space_info_update_bytes_may_use(fs_info,
2770 space_info, to_add);
2771 if (global_rsv->reserved >= global_rsv->size)
2772 global_rsv->full = 1;
2775 spin_unlock(&global_rsv->lock);
2777 /* Add to any tickets we may have */
2778 if (!readonly && return_free_space && len)
2779 btrfs_try_granting_tickets(fs_info, space_info);
2780 spin_unlock(&space_info->lock);
2784 btrfs_put_block_group(cache);
2788 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2790 struct btrfs_fs_info *fs_info = trans->fs_info;
2791 struct btrfs_block_group *block_group, *tmp;
2792 struct list_head *deleted_bgs;
2793 struct extent_io_tree *unpin;
2798 unpin = &trans->transaction->pinned_extents;
2800 while (!TRANS_ABORTED(trans)) {
2801 struct extent_state *cached_state = NULL;
2803 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2804 ret = find_first_extent_bit(unpin, 0, &start, &end,
2805 EXTENT_DIRTY, &cached_state);
2807 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2811 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2812 ret = btrfs_discard_extent(fs_info, start,
2813 end + 1 - start, NULL);
2815 clear_extent_dirty(unpin, start, end, &cached_state);
2816 unpin_extent_range(fs_info, start, end, true);
2817 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2818 free_extent_state(cached_state);
2822 if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2823 btrfs_discard_calc_delay(&fs_info->discard_ctl);
2824 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2828 * Transaction is finished. We don't need the lock anymore. We
2829 * do need to clean up the block groups in case of a transaction
2832 deleted_bgs = &trans->transaction->deleted_bgs;
2833 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2837 if (!TRANS_ABORTED(trans))
2838 ret = btrfs_discard_extent(fs_info,
2840 block_group->length,
2843 list_del_init(&block_group->bg_list);
2844 btrfs_unfreeze_block_group(block_group);
2845 btrfs_put_block_group(block_group);
2848 const char *errstr = btrfs_decode_error(ret);
2850 "discard failed while removing blockgroup: errno=%d %s",
2859 * Drop one or more refs of @node.
2861 * 1. Locate the extent refs.
2862 * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
2863 * Locate it, then reduce the refs number or remove the ref line completely.
2865 * 2. Update the refs count in EXTENT/METADATA_ITEM
2867 * Inline backref case:
2869 * in extent tree we have:
2871 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2872 * refs 2 gen 6 flags DATA
2873 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2874 * extent data backref root FS_TREE objectid 257 offset 0 count 1
2876 * This function gets called with:
2878 * node->bytenr = 13631488
2879 * node->num_bytes = 1048576
2880 * root_objectid = FS_TREE
2881 * owner_objectid = 257
2885 * Then we should get some like:
2887 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2888 * refs 1 gen 6 flags DATA
2889 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2891 * Keyed backref case:
2893 * in extent tree we have:
2895 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2896 * refs 754 gen 6 flags DATA
2898 * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
2899 * extent data backref root FS_TREE objectid 866 offset 0 count 1
2901 * This function get called with:
2903 * node->bytenr = 13631488
2904 * node->num_bytes = 1048576
2905 * root_objectid = FS_TREE
2906 * owner_objectid = 866
2910 * Then we should get some like:
2912 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2913 * refs 753 gen 6 flags DATA
2915 * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
2917 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2918 struct btrfs_delayed_ref_node *node, u64 parent,
2919 u64 root_objectid, u64 owner_objectid,
2920 u64 owner_offset, int refs_to_drop,
2921 struct btrfs_delayed_extent_op *extent_op)
2923 struct btrfs_fs_info *info = trans->fs_info;
2924 struct btrfs_key key;
2925 struct btrfs_path *path;
2926 struct btrfs_root *extent_root = info->extent_root;
2927 struct extent_buffer *leaf;
2928 struct btrfs_extent_item *ei;
2929 struct btrfs_extent_inline_ref *iref;
2932 int extent_slot = 0;
2933 int found_extent = 0;
2937 u64 bytenr = node->bytenr;
2938 u64 num_bytes = node->num_bytes;
2940 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
2942 path = btrfs_alloc_path();
2946 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2948 if (!is_data && refs_to_drop != 1) {
2950 "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
2951 node->bytenr, refs_to_drop);
2953 btrfs_abort_transaction(trans, ret);
2958 skinny_metadata = false;
2960 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
2961 parent, root_objectid, owner_objectid,
2965 * Either the inline backref or the SHARED_DATA_REF/
2966 * SHARED_BLOCK_REF is found
2968 * Here is a quick path to locate EXTENT/METADATA_ITEM.
2969 * It's possible the EXTENT/METADATA_ITEM is near current slot.
2971 extent_slot = path->slots[0];
2972 while (extent_slot >= 0) {
2973 btrfs_item_key_to_cpu(path->nodes[0], &key,
2975 if (key.objectid != bytenr)
2977 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2978 key.offset == num_bytes) {
2982 if (key.type == BTRFS_METADATA_ITEM_KEY &&
2983 key.offset == owner_objectid) {
2988 /* Quick path didn't find the EXTEMT/METADATA_ITEM */
2989 if (path->slots[0] - extent_slot > 5)
2994 if (!found_extent) {
2997 "invalid iref, no EXTENT/METADATA_ITEM found but has inline extent ref");
2998 btrfs_abort_transaction(trans, -EUCLEAN);
3001 /* Must be SHARED_* item, remove the backref first */
3002 ret = remove_extent_backref(trans, path, NULL,
3004 is_data, &last_ref);
3006 btrfs_abort_transaction(trans, ret);
3009 btrfs_release_path(path);
3011 /* Slow path to locate EXTENT/METADATA_ITEM */
3012 key.objectid = bytenr;
3013 key.type = BTRFS_EXTENT_ITEM_KEY;
3014 key.offset = num_bytes;
3016 if (!is_data && skinny_metadata) {
3017 key.type = BTRFS_METADATA_ITEM_KEY;
3018 key.offset = owner_objectid;
3021 ret = btrfs_search_slot(trans, extent_root,
3023 if (ret > 0 && skinny_metadata && path->slots[0]) {
3025 * Couldn't find our skinny metadata item,
3026 * see if we have ye olde extent item.
3029 btrfs_item_key_to_cpu(path->nodes[0], &key,
3031 if (key.objectid == bytenr &&
3032 key.type == BTRFS_EXTENT_ITEM_KEY &&
3033 key.offset == num_bytes)
3037 if (ret > 0 && skinny_metadata) {
3038 skinny_metadata = false;
3039 key.objectid = bytenr;
3040 key.type = BTRFS_EXTENT_ITEM_KEY;
3041 key.offset = num_bytes;
3042 btrfs_release_path(path);
3043 ret = btrfs_search_slot(trans, extent_root,
3049 "umm, got %d back from search, was looking for %llu",
3052 btrfs_print_leaf(path->nodes[0]);
3055 btrfs_abort_transaction(trans, ret);
3058 extent_slot = path->slots[0];
3060 } else if (WARN_ON(ret == -ENOENT)) {
3061 btrfs_print_leaf(path->nodes[0]);
3063 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
3064 bytenr, parent, root_objectid, owner_objectid,
3066 btrfs_abort_transaction(trans, ret);
3069 btrfs_abort_transaction(trans, ret);
3073 leaf = path->nodes[0];
3074 item_size = btrfs_item_size_nr(leaf, extent_slot);
3075 if (unlikely(item_size < sizeof(*ei))) {
3077 btrfs_print_v0_err(info);
3078 btrfs_abort_transaction(trans, ret);
3081 ei = btrfs_item_ptr(leaf, extent_slot,
3082 struct btrfs_extent_item);
3083 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3084 key.type == BTRFS_EXTENT_ITEM_KEY) {
3085 struct btrfs_tree_block_info *bi;
3086 if (item_size < sizeof(*ei) + sizeof(*bi)) {
3088 "invalid extent item size for key (%llu, %u, %llu) owner %llu, has %u expect >= %zu",
3089 key.objectid, key.type, key.offset,
3090 owner_objectid, item_size,
3091 sizeof(*ei) + sizeof(*bi));
3092 btrfs_abort_transaction(trans, -EUCLEAN);
3095 bi = (struct btrfs_tree_block_info *)(ei + 1);
3096 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3099 refs = btrfs_extent_refs(leaf, ei);
3100 if (refs < refs_to_drop) {
3102 "trying to drop %d refs but we only have %llu for bytenr %llu",
3103 refs_to_drop, refs, bytenr);
3104 btrfs_abort_transaction(trans, -EUCLEAN);
3107 refs -= refs_to_drop;
3111 __run_delayed_extent_op(extent_op, leaf, ei);
3113 * In the case of inline back ref, reference count will
3114 * be updated by remove_extent_backref
3117 if (!found_extent) {
3119 "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found");
3120 btrfs_abort_transaction(trans, -EUCLEAN);
3124 btrfs_set_extent_refs(leaf, ei, refs);
3125 btrfs_mark_buffer_dirty(leaf);
3128 ret = remove_extent_backref(trans, path, iref,
3129 refs_to_drop, is_data,
3132 btrfs_abort_transaction(trans, ret);
3137 /* In this branch refs == 1 */
3139 if (is_data && refs_to_drop !=
3140 extent_data_ref_count(path, iref)) {
3142 "invalid refs_to_drop, current refs %u refs_to_drop %u",
3143 extent_data_ref_count(path, iref),
3145 btrfs_abort_transaction(trans, -EUCLEAN);
3149 if (path->slots[0] != extent_slot) {
3151 "invalid iref, extent item key (%llu %u %llu) doesn't have wanted iref",
3152 key.objectid, key.type,
3154 btrfs_abort_transaction(trans, -EUCLEAN);
3159 * No inline ref, we must be at SHARED_* item,
3160 * And it's single ref, it must be:
3161 * | extent_slot ||extent_slot + 1|
3162 * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3164 if (path->slots[0] != extent_slot + 1) {
3166 "invalid SHARED_* item, previous item is not EXTENT/METADATA_ITEM");
3167 btrfs_abort_transaction(trans, -EUCLEAN);
3170 path->slots[0] = extent_slot;
3176 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3179 btrfs_abort_transaction(trans, ret);
3182 btrfs_release_path(path);
3185 ret = btrfs_del_csums(trans, info->csum_root, bytenr,
3188 btrfs_abort_transaction(trans, ret);
3193 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
3195 btrfs_abort_transaction(trans, ret);
3199 ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0);
3201 btrfs_abort_transaction(trans, ret);
3205 btrfs_release_path(path);
3208 btrfs_free_path(path);
3212 * Leaf dump can take up a lot of log buffer, so we only do full leaf
3213 * dump for debug build.
3215 if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) {
3216 btrfs_crit(info, "path->slots[0]=%d extent_slot=%d",
3217 path->slots[0], extent_slot);
3218 btrfs_print_leaf(path->nodes[0]);
3221 btrfs_free_path(path);
3226 * when we free an block, it is possible (and likely) that we free the last
3227 * delayed ref for that extent as well. This searches the delayed ref tree for
3228 * a given extent, and if there are no other delayed refs to be processed, it
3229 * removes it from the tree.
3231 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3234 struct btrfs_delayed_ref_head *head;
3235 struct btrfs_delayed_ref_root *delayed_refs;
3238 delayed_refs = &trans->transaction->delayed_refs;
3239 spin_lock(&delayed_refs->lock);
3240 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3242 goto out_delayed_unlock;
3244 spin_lock(&head->lock);
3245 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3248 if (cleanup_extent_op(head) != NULL)
3252 * waiting for the lock here would deadlock. If someone else has it
3253 * locked they are already in the process of dropping it anyway
3255 if (!mutex_trylock(&head->mutex))
3258 btrfs_delete_ref_head(delayed_refs, head);
3259 head->processing = 0;
3261 spin_unlock(&head->lock);
3262 spin_unlock(&delayed_refs->lock);
3264 BUG_ON(head->extent_op);
3265 if (head->must_insert_reserved)
3268 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3269 mutex_unlock(&head->mutex);
3270 btrfs_put_delayed_ref_head(head);
3273 spin_unlock(&head->lock);
3276 spin_unlock(&delayed_refs->lock);
3280 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3281 struct btrfs_root *root,
3282 struct extent_buffer *buf,
3283 u64 parent, int last_ref)
3285 struct btrfs_fs_info *fs_info = root->fs_info;
3286 struct btrfs_ref generic_ref = { 0 };
3289 btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3290 buf->start, buf->len, parent);
3291 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3292 root->root_key.objectid);
3294 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
3295 btrfs_ref_tree_mod(fs_info, &generic_ref);
3296 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL);
3297 BUG_ON(ret); /* -ENOMEM */
3300 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
3301 struct btrfs_block_group *cache;
3302 bool must_pin = false;
3304 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
3305 ret = check_ref_cleanup(trans, buf->start);
3307 btrfs_redirty_list_add(trans->transaction, buf);
3312 cache = btrfs_lookup_block_group(fs_info, buf->start);
3314 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3315 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3316 btrfs_put_block_group(cache);
3321 * If this is a leaf and there are tree mod log users, we may
3322 * have recorded mod log operations that point to this leaf.
3323 * So we must make sure no one reuses this leaf's extent before
3324 * mod log operations are applied to a node, otherwise after
3325 * rewinding a node using the mod log operations we get an
3326 * inconsistent btree, as the leaf's extent may now be used as
3327 * a node or leaf for another different btree.
3328 * We are safe from races here because at this point no other
3329 * node or root points to this extent buffer, so if after this
3330 * check a new tree mod log user joins, it will not be able to
3331 * find a node pointing to this leaf and record operations that
3332 * point to this leaf.
3334 if (btrfs_header_level(buf) == 0 &&
3335 test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags))
3338 if (must_pin || btrfs_is_zoned(fs_info)) {
3339 btrfs_redirty_list_add(trans->transaction, buf);
3340 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3341 btrfs_put_block_group(cache);
3345 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3347 btrfs_add_free_space(cache, buf->start, buf->len);
3348 btrfs_free_reserved_bytes(cache, buf->len, 0);
3349 btrfs_put_block_group(cache);
3350 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3355 * Deleting the buffer, clear the corrupt flag since it doesn't
3358 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3362 /* Can return -ENOMEM */
3363 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3365 struct btrfs_fs_info *fs_info = trans->fs_info;
3368 if (btrfs_is_testing(fs_info))
3372 * tree log blocks never actually go into the extent allocation
3373 * tree, just update pinning info and exit early.
3375 if ((ref->type == BTRFS_REF_METADATA &&
3376 ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
3377 (ref->type == BTRFS_REF_DATA &&
3378 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) {
3379 /* unlocks the pinned mutex */
3380 btrfs_pin_extent(trans, ref->bytenr, ref->len, 1);
3382 } else if (ref->type == BTRFS_REF_METADATA) {
3383 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL);
3385 ret = btrfs_add_delayed_data_ref(trans, ref, 0);
3388 if (!((ref->type == BTRFS_REF_METADATA &&
3389 ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
3390 (ref->type == BTRFS_REF_DATA &&
3391 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)))
3392 btrfs_ref_tree_mod(fs_info, ref);
3397 enum btrfs_loop_type {
3398 LOOP_CACHING_NOWAIT,
3405 btrfs_lock_block_group(struct btrfs_block_group *cache,
3409 down_read(&cache->data_rwsem);
3412 static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3415 btrfs_get_block_group(cache);
3417 down_read(&cache->data_rwsem);
3420 static struct btrfs_block_group *btrfs_lock_cluster(
3421 struct btrfs_block_group *block_group,
3422 struct btrfs_free_cluster *cluster,
3424 __acquires(&cluster->refill_lock)
3426 struct btrfs_block_group *used_bg = NULL;
3428 spin_lock(&cluster->refill_lock);
3430 used_bg = cluster->block_group;
3434 if (used_bg == block_group)
3437 btrfs_get_block_group(used_bg);
3442 if (down_read_trylock(&used_bg->data_rwsem))
3445 spin_unlock(&cluster->refill_lock);
3447 /* We should only have one-level nested. */
3448 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3450 spin_lock(&cluster->refill_lock);
3451 if (used_bg == cluster->block_group)
3454 up_read(&used_bg->data_rwsem);
3455 btrfs_put_block_group(used_bg);
3460 btrfs_release_block_group(struct btrfs_block_group *cache,
3464 up_read(&cache->data_rwsem);
3465 btrfs_put_block_group(cache);
3468 enum btrfs_extent_allocation_policy {
3469 BTRFS_EXTENT_ALLOC_CLUSTERED,
3470 BTRFS_EXTENT_ALLOC_ZONED,
3474 * Structure used internally for find_free_extent() function. Wraps needed
3477 struct find_free_extent_ctl {
3478 /* Basic allocation info */
3484 /* Where to start the search inside the bg */
3487 /* For clustered allocation */
3489 struct btrfs_free_cluster *last_ptr;
3492 bool have_caching_bg;
3493 bool orig_have_caching_bg;
3495 /* Allocation is called for tree-log */
3498 /* RAID index, converted from flags */
3502 * Current loop number, check find_free_extent_update_loop() for details
3507 * Whether we're refilling a cluster, if true we need to re-search
3508 * current block group but don't try to refill the cluster again.
3510 bool retry_clustered;
3513 * Whether we're updating free space cache, if true we need to re-search
3514 * current block group but don't try updating free space cache again.
3516 bool retry_unclustered;
3518 /* If current block group is cached */
3521 /* Max contiguous hole found */
3522 u64 max_extent_size;
3524 /* Total free space from free space cache, not always contiguous */
3525 u64 total_free_space;
3530 /* Hint where to start looking for an empty space */
3533 /* Allocation policy */
3534 enum btrfs_extent_allocation_policy policy;
3539 * Helper function for find_free_extent().
3541 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3542 * Return -EAGAIN to inform caller that we need to re-search this block group
3543 * Return >0 to inform caller that we find nothing
3544 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3546 static int find_free_extent_clustered(struct btrfs_block_group *bg,
3547 struct find_free_extent_ctl *ffe_ctl,
3548 struct btrfs_block_group **cluster_bg_ret)
3550 struct btrfs_block_group *cluster_bg;
3551 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3552 u64 aligned_cluster;
3556 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3558 goto refill_cluster;
3559 if (cluster_bg != bg && (cluster_bg->ro ||
3560 !block_group_bits(cluster_bg, ffe_ctl->flags)))
3561 goto release_cluster;
3563 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3564 ffe_ctl->num_bytes, cluster_bg->start,
3565 &ffe_ctl->max_extent_size);
3567 /* We have a block, we're done */
3568 spin_unlock(&last_ptr->refill_lock);
3569 trace_btrfs_reserve_extent_cluster(cluster_bg,
3570 ffe_ctl->search_start, ffe_ctl->num_bytes);
3571 *cluster_bg_ret = cluster_bg;
3572 ffe_ctl->found_offset = offset;
3575 WARN_ON(last_ptr->block_group != cluster_bg);
3579 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3580 * lets just skip it and let the allocator find whatever block it can
3581 * find. If we reach this point, we will have tried the cluster
3582 * allocator plenty of times and not have found anything, so we are
3583 * likely way too fragmented for the clustering stuff to find anything.
3585 * However, if the cluster is taken from the current block group,
3586 * release the cluster first, so that we stand a better chance of
3587 * succeeding in the unclustered allocation.
3589 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3590 spin_unlock(&last_ptr->refill_lock);
3591 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3595 /* This cluster didn't work out, free it and start over */
3596 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3598 if (cluster_bg != bg)
3599 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3602 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3603 spin_unlock(&last_ptr->refill_lock);
3607 aligned_cluster = max_t(u64,
3608 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3609 bg->full_stripe_len);
3610 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3611 ffe_ctl->num_bytes, aligned_cluster);
3613 /* Now pull our allocation out of this cluster */
3614 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3615 ffe_ctl->num_bytes, ffe_ctl->search_start,
3616 &ffe_ctl->max_extent_size);
3618 /* We found one, proceed */
3619 spin_unlock(&last_ptr->refill_lock);
3620 trace_btrfs_reserve_extent_cluster(bg,
3621 ffe_ctl->search_start,
3622 ffe_ctl->num_bytes);
3623 ffe_ctl->found_offset = offset;
3626 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
3627 !ffe_ctl->retry_clustered) {
3628 spin_unlock(&last_ptr->refill_lock);
3630 ffe_ctl->retry_clustered = true;
3631 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3632 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
3636 * At this point we either didn't find a cluster or we weren't able to
3637 * allocate a block from our cluster. Free the cluster we've been
3638 * trying to use, and go to the next block group.
3640 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3641 spin_unlock(&last_ptr->refill_lock);
3646 * Return >0 to inform caller that we find nothing
3647 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3648 * Return -EAGAIN to inform caller that we need to re-search this block group
3650 static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3651 struct find_free_extent_ctl *ffe_ctl)
3653 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3657 * We are doing an unclustered allocation, set the fragmented flag so
3658 * we don't bother trying to setup a cluster again until we get more
3661 if (unlikely(last_ptr)) {
3662 spin_lock(&last_ptr->lock);
3663 last_ptr->fragmented = 1;
3664 spin_unlock(&last_ptr->lock);
3666 if (ffe_ctl->cached) {
3667 struct btrfs_free_space_ctl *free_space_ctl;
3669 free_space_ctl = bg->free_space_ctl;
3670 spin_lock(&free_space_ctl->tree_lock);
3671 if (free_space_ctl->free_space <
3672 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3673 ffe_ctl->empty_size) {
3674 ffe_ctl->total_free_space = max_t(u64,
3675 ffe_ctl->total_free_space,
3676 free_space_ctl->free_space);
3677 spin_unlock(&free_space_ctl->tree_lock);
3680 spin_unlock(&free_space_ctl->tree_lock);
3683 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3684 ffe_ctl->num_bytes, ffe_ctl->empty_size,
3685 &ffe_ctl->max_extent_size);
3688 * If we didn't find a chunk, and we haven't failed on this block group
3689 * before, and this block group is in the middle of caching and we are
3690 * ok with waiting, then go ahead and wait for progress to be made, and
3691 * set @retry_unclustered to true.
3693 * If @retry_unclustered is true then we've already waited on this
3694 * block group once and should move on to the next block group.
3696 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
3697 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
3698 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3699 ffe_ctl->empty_size);
3700 ffe_ctl->retry_unclustered = true;
3702 } else if (!offset) {
3705 ffe_ctl->found_offset = offset;
3709 static int do_allocation_clustered(struct btrfs_block_group *block_group,
3710 struct find_free_extent_ctl *ffe_ctl,
3711 struct btrfs_block_group **bg_ret)
3715 /* We want to try and use the cluster allocator, so lets look there */
3716 if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
3717 ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
3718 if (ret >= 0 || ret == -EAGAIN)
3720 /* ret == -ENOENT case falls through */
3723 return find_free_extent_unclustered(block_group, ffe_ctl);
3727 * Tree-log block group locking
3728 * ============================
3730 * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3731 * indicates the starting address of a block group, which is reserved only
3732 * for tree-log metadata.
3739 * fs_info::treelog_bg_lock
3743 * Simple allocator for sequential-only block group. It only allows sequential
3744 * allocation. No need to play with trees. This function also reserves the
3745 * bytes as in btrfs_add_reserved_bytes.
3747 static int do_allocation_zoned(struct btrfs_block_group *block_group,
3748 struct find_free_extent_ctl *ffe_ctl,
3749 struct btrfs_block_group **bg_ret)
3751 struct btrfs_fs_info *fs_info = block_group->fs_info;
3752 struct btrfs_space_info *space_info = block_group->space_info;
3753 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3754 u64 start = block_group->start;
3755 u64 num_bytes = ffe_ctl->num_bytes;
3757 u64 bytenr = block_group->start;
3762 ASSERT(btrfs_is_zoned(block_group->fs_info));
3765 * Do not allow non-tree-log blocks in the dedicated tree-log block
3766 * group, and vice versa.
3768 spin_lock(&fs_info->treelog_bg_lock);
3769 log_bytenr = fs_info->treelog_bg;
3770 skip = log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) ||
3771 (!ffe_ctl->for_treelog && bytenr == log_bytenr));
3772 spin_unlock(&fs_info->treelog_bg_lock);
3776 spin_lock(&space_info->lock);
3777 spin_lock(&block_group->lock);
3778 spin_lock(&fs_info->treelog_bg_lock);
3780 ASSERT(!ffe_ctl->for_treelog ||
3781 block_group->start == fs_info->treelog_bg ||
3782 fs_info->treelog_bg == 0);
3784 if (block_group->ro) {
3790 * Do not allow currently using block group to be tree-log dedicated
3793 if (ffe_ctl->for_treelog && !fs_info->treelog_bg &&
3794 (block_group->used || block_group->reserved)) {
3799 avail = block_group->length - block_group->alloc_offset;
3800 if (avail < num_bytes) {
3801 if (ffe_ctl->max_extent_size < avail) {
3803 * With sequential allocator, free space is always
3806 ffe_ctl->max_extent_size = avail;
3807 ffe_ctl->total_free_space = avail;
3813 if (ffe_ctl->for_treelog && !fs_info->treelog_bg)
3814 fs_info->treelog_bg = block_group->start;
3816 ffe_ctl->found_offset = start + block_group->alloc_offset;
3817 block_group->alloc_offset += num_bytes;
3818 spin_lock(&ctl->tree_lock);
3819 ctl->free_space -= num_bytes;
3820 spin_unlock(&ctl->tree_lock);
3823 * We do not check if found_offset is aligned to stripesize. The
3824 * address is anyway rewritten when using zone append writing.
3827 ffe_ctl->search_start = ffe_ctl->found_offset;
3830 if (ret && ffe_ctl->for_treelog)
3831 fs_info->treelog_bg = 0;
3832 spin_unlock(&fs_info->treelog_bg_lock);
3833 spin_unlock(&block_group->lock);
3834 spin_unlock(&space_info->lock);
3838 static int do_allocation(struct btrfs_block_group *block_group,
3839 struct find_free_extent_ctl *ffe_ctl,
3840 struct btrfs_block_group **bg_ret)
3842 switch (ffe_ctl->policy) {
3843 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3844 return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
3845 case BTRFS_EXTENT_ALLOC_ZONED:
3846 return do_allocation_zoned(block_group, ffe_ctl, bg_ret);
3852 static void release_block_group(struct btrfs_block_group *block_group,
3853 struct find_free_extent_ctl *ffe_ctl,
3856 switch (ffe_ctl->policy) {
3857 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3858 ffe_ctl->retry_clustered = false;
3859 ffe_ctl->retry_unclustered = false;
3861 case BTRFS_EXTENT_ALLOC_ZONED:
3868 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
3870 btrfs_release_block_group(block_group, delalloc);
3873 static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
3874 struct btrfs_key *ins)
3876 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3878 if (!ffe_ctl->use_cluster && last_ptr) {
3879 spin_lock(&last_ptr->lock);
3880 last_ptr->window_start = ins->objectid;
3881 spin_unlock(&last_ptr->lock);
3885 static void found_extent(struct find_free_extent_ctl *ffe_ctl,
3886 struct btrfs_key *ins)
3888 switch (ffe_ctl->policy) {
3889 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3890 found_extent_clustered(ffe_ctl, ins);
3892 case BTRFS_EXTENT_ALLOC_ZONED:
3900 static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl)
3902 switch (ffe_ctl->policy) {
3903 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3905 * If we can't allocate a new chunk we've already looped through
3906 * at least once, move on to the NO_EMPTY_SIZE case.
3908 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
3910 case BTRFS_EXTENT_ALLOC_ZONED:
3919 * Return >0 means caller needs to re-search for free extent
3920 * Return 0 means we have the needed free extent.
3921 * Return <0 means we failed to locate any free extent.
3923 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
3924 struct btrfs_key *ins,
3925 struct find_free_extent_ctl *ffe_ctl,
3928 struct btrfs_root *root = fs_info->extent_root;
3931 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
3932 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
3933 ffe_ctl->orig_have_caching_bg = true;
3935 if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
3936 ffe_ctl->have_caching_bg)
3939 if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
3942 if (ins->objectid) {
3943 found_extent(ffe_ctl, ins);
3948 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
3949 * caching kthreads as we move along
3950 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
3951 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
3952 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
3955 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
3957 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
3959 * We want to skip the LOOP_CACHING_WAIT step if we
3960 * don't have any uncached bgs and we've already done a
3961 * full search through.
3963 if (ffe_ctl->orig_have_caching_bg || !full_search)
3964 ffe_ctl->loop = LOOP_CACHING_WAIT;
3966 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
3971 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
3972 struct btrfs_trans_handle *trans;
3975 trans = current->journal_info;
3979 trans = btrfs_join_transaction(root);
3981 if (IS_ERR(trans)) {
3982 ret = PTR_ERR(trans);
3986 ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
3989 /* Do not bail out on ENOSPC since we can do more. */
3991 ret = chunk_allocation_failed(ffe_ctl);
3993 btrfs_abort_transaction(trans, ret);
3997 btrfs_end_transaction(trans);
4002 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
4003 if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
4007 * Don't loop again if we already have no empty_size and
4010 if (ffe_ctl->empty_size == 0 &&
4011 ffe_ctl->empty_cluster == 0)
4013 ffe_ctl->empty_size = 0;
4014 ffe_ctl->empty_cluster = 0;
4021 static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
4022 struct find_free_extent_ctl *ffe_ctl,
4023 struct btrfs_space_info *space_info,
4024 struct btrfs_key *ins)
4027 * If our free space is heavily fragmented we may not be able to make
4028 * big contiguous allocations, so instead of doing the expensive search
4029 * for free space, simply return ENOSPC with our max_extent_size so we
4030 * can go ahead and search for a more manageable chunk.
4032 * If our max_extent_size is large enough for our allocation simply
4033 * disable clustering since we will likely not be able to find enough
4034 * space to create a cluster and induce latency trying.
4036 if (space_info->max_extent_size) {
4037 spin_lock(&space_info->lock);
4038 if (space_info->max_extent_size &&
4039 ffe_ctl->num_bytes > space_info->max_extent_size) {
4040 ins->offset = space_info->max_extent_size;
4041 spin_unlock(&space_info->lock);
4043 } else if (space_info->max_extent_size) {
4044 ffe_ctl->use_cluster = false;
4046 spin_unlock(&space_info->lock);
4049 ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
4050 &ffe_ctl->empty_cluster);
4051 if (ffe_ctl->last_ptr) {
4052 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4054 spin_lock(&last_ptr->lock);
4055 if (last_ptr->block_group)
4056 ffe_ctl->hint_byte = last_ptr->window_start;
4057 if (last_ptr->fragmented) {
4059 * We still set window_start so we can keep track of the
4060 * last place we found an allocation to try and save
4063 ffe_ctl->hint_byte = last_ptr->window_start;
4064 ffe_ctl->use_cluster = false;
4066 spin_unlock(&last_ptr->lock);
4072 static int prepare_allocation(struct btrfs_fs_info *fs_info,
4073 struct find_free_extent_ctl *ffe_ctl,
4074 struct btrfs_space_info *space_info,
4075 struct btrfs_key *ins)
4077 switch (ffe_ctl->policy) {
4078 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4079 return prepare_allocation_clustered(fs_info, ffe_ctl,
4081 case BTRFS_EXTENT_ALLOC_ZONED:
4082 if (ffe_ctl->for_treelog) {
4083 spin_lock(&fs_info->treelog_bg_lock);
4084 if (fs_info->treelog_bg)
4085 ffe_ctl->hint_byte = fs_info->treelog_bg;
4086 spin_unlock(&fs_info->treelog_bg_lock);
4095 * walks the btree of allocated extents and find a hole of a given size.
4096 * The key ins is changed to record the hole:
4097 * ins->objectid == start position
4098 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4099 * ins->offset == the size of the hole.
4100 * Any available blocks before search_start are skipped.
4102 * If there is no suitable free space, we will record the max size of
4103 * the free space extent currently.
4105 * The overall logic and call chain:
4107 * find_free_extent()
4108 * |- Iterate through all block groups
4109 * | |- Get a valid block group
4110 * | |- Try to do clustered allocation in that block group
4111 * | |- Try to do unclustered allocation in that block group
4112 * | |- Check if the result is valid
4113 * | | |- If valid, then exit
4114 * | |- Jump to next block group
4116 * |- Push harder to find free extents
4117 * |- If not found, re-iterate all block groups
4119 static noinline int find_free_extent(struct btrfs_root *root,
4120 u64 ram_bytes, u64 num_bytes, u64 empty_size,
4121 u64 hint_byte_orig, struct btrfs_key *ins,
4122 u64 flags, int delalloc)
4124 struct btrfs_fs_info *fs_info = root->fs_info;
4126 int cache_block_group_error = 0;
4127 struct btrfs_block_group *block_group = NULL;
4128 struct find_free_extent_ctl ffe_ctl = {0};
4129 struct btrfs_space_info *space_info;
4130 bool full_search = false;
4131 bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4133 WARN_ON(num_bytes < fs_info->sectorsize);
4135 ffe_ctl.num_bytes = num_bytes;
4136 ffe_ctl.empty_size = empty_size;
4137 ffe_ctl.flags = flags;
4138 ffe_ctl.search_start = 0;
4139 ffe_ctl.delalloc = delalloc;
4140 ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
4141 ffe_ctl.have_caching_bg = false;
4142 ffe_ctl.orig_have_caching_bg = false;
4143 ffe_ctl.found_offset = 0;
4144 ffe_ctl.hint_byte = hint_byte_orig;
4145 ffe_ctl.for_treelog = for_treelog;
4146 ffe_ctl.policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
4148 /* For clustered allocation */
4149 ffe_ctl.retry_clustered = false;
4150 ffe_ctl.retry_unclustered = false;
4151 ffe_ctl.last_ptr = NULL;
4152 ffe_ctl.use_cluster = true;
4154 if (btrfs_is_zoned(fs_info))
4155 ffe_ctl.policy = BTRFS_EXTENT_ALLOC_ZONED;
4157 ins->type = BTRFS_EXTENT_ITEM_KEY;
4161 trace_find_free_extent(root, num_bytes, empty_size, flags);
4163 space_info = btrfs_find_space_info(fs_info, flags);
4165 btrfs_err(fs_info, "No space info for %llu", flags);
4169 ret = prepare_allocation(fs_info, &ffe_ctl, space_info, ins);
4173 ffe_ctl.search_start = max(ffe_ctl.search_start,
4174 first_logical_byte(fs_info, 0));
4175 ffe_ctl.search_start = max(ffe_ctl.search_start, ffe_ctl.hint_byte);
4176 if (ffe_ctl.search_start == ffe_ctl.hint_byte) {
4177 block_group = btrfs_lookup_block_group(fs_info,
4178 ffe_ctl.search_start);
4180 * we don't want to use the block group if it doesn't match our
4181 * allocation bits, or if its not cached.
4183 * However if we are re-searching with an ideal block group
4184 * picked out then we don't care that the block group is cached.
4186 if (block_group && block_group_bits(block_group, flags) &&
4187 block_group->cached != BTRFS_CACHE_NO) {
4188 down_read(&space_info->groups_sem);
4189 if (list_empty(&block_group->list) ||
4192 * someone is removing this block group,
4193 * we can't jump into the have_block_group
4194 * target because our list pointers are not
4197 btrfs_put_block_group(block_group);
4198 up_read(&space_info->groups_sem);
4200 ffe_ctl.index = btrfs_bg_flags_to_raid_index(
4201 block_group->flags);
4202 btrfs_lock_block_group(block_group, delalloc);
4203 goto have_block_group;
4205 } else if (block_group) {
4206 btrfs_put_block_group(block_group);
4210 ffe_ctl.have_caching_bg = false;
4211 if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
4214 down_read(&space_info->groups_sem);
4215 list_for_each_entry(block_group,
4216 &space_info->block_groups[ffe_ctl.index], list) {
4217 struct btrfs_block_group *bg_ret;
4219 /* If the block group is read-only, we can skip it entirely. */
4220 if (unlikely(block_group->ro)) {
4222 btrfs_clear_treelog_bg(block_group);
4226 btrfs_grab_block_group(block_group, delalloc);
4227 ffe_ctl.search_start = block_group->start;
4230 * this can happen if we end up cycling through all the
4231 * raid types, but we want to make sure we only allocate
4232 * for the proper type.
4234 if (!block_group_bits(block_group, flags)) {
4235 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4236 BTRFS_BLOCK_GROUP_RAID1_MASK |
4237 BTRFS_BLOCK_GROUP_RAID56_MASK |
4238 BTRFS_BLOCK_GROUP_RAID10;
4241 * if they asked for extra copies and this block group
4242 * doesn't provide them, bail. This does allow us to
4243 * fill raid0 from raid1.
4245 if ((flags & extra) && !(block_group->flags & extra))
4249 * This block group has different flags than we want.
4250 * It's possible that we have MIXED_GROUP flag but no
4251 * block group is mixed. Just skip such block group.
4253 btrfs_release_block_group(block_group, delalloc);
4258 ffe_ctl.cached = btrfs_block_group_done(block_group);
4259 if (unlikely(!ffe_ctl.cached)) {
4260 ffe_ctl.have_caching_bg = true;
4261 ret = btrfs_cache_block_group(block_group, 0);
4264 * If we get ENOMEM here or something else we want to
4265 * try other block groups, because it may not be fatal.
4266 * However if we can't find anything else we need to
4267 * save our return here so that we return the actual
4268 * error that caused problems, not ENOSPC.
4271 if (!cache_block_group_error)
4272 cache_block_group_error = ret;
4279 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
4283 ret = do_allocation(block_group, &ffe_ctl, &bg_ret);
4285 if (bg_ret && bg_ret != block_group) {
4286 btrfs_release_block_group(block_group, delalloc);
4287 block_group = bg_ret;
4289 } else if (ret == -EAGAIN) {
4290 goto have_block_group;
4291 } else if (ret > 0) {
4296 ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
4297 fs_info->stripesize);
4299 /* move on to the next group */
4300 if (ffe_ctl.search_start + num_bytes >
4301 block_group->start + block_group->length) {
4302 btrfs_add_free_space_unused(block_group,
4303 ffe_ctl.found_offset, num_bytes);
4307 if (ffe_ctl.found_offset < ffe_ctl.search_start)
4308 btrfs_add_free_space_unused(block_group,
4309 ffe_ctl.found_offset,
4310 ffe_ctl.search_start - ffe_ctl.found_offset);
4312 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
4313 num_bytes, delalloc);
4314 if (ret == -EAGAIN) {
4315 btrfs_add_free_space_unused(block_group,
4316 ffe_ctl.found_offset, num_bytes);
4319 btrfs_inc_block_group_reservations(block_group);
4321 /* we are all good, lets return */
4322 ins->objectid = ffe_ctl.search_start;
4323 ins->offset = num_bytes;
4325 trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
4327 btrfs_release_block_group(block_group, delalloc);
4330 release_block_group(block_group, &ffe_ctl, delalloc);
4333 up_read(&space_info->groups_sem);
4335 ret = find_free_extent_update_loop(fs_info, ins, &ffe_ctl, full_search);
4339 if (ret == -ENOSPC && !cache_block_group_error) {
4341 * Use ffe_ctl->total_free_space as fallback if we can't find
4342 * any contiguous hole.
4344 if (!ffe_ctl.max_extent_size)
4345 ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
4346 spin_lock(&space_info->lock);
4347 space_info->max_extent_size = ffe_ctl.max_extent_size;
4348 spin_unlock(&space_info->lock);
4349 ins->offset = ffe_ctl.max_extent_size;
4350 } else if (ret == -ENOSPC) {
4351 ret = cache_block_group_error;
4357 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4358 * hole that is at least as big as @num_bytes.
4360 * @root - The root that will contain this extent
4362 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4363 * is used for accounting purposes. This value differs
4364 * from @num_bytes only in the case of compressed extents.
4366 * @num_bytes - Number of bytes to allocate on-disk.
4368 * @min_alloc_size - Indicates the minimum amount of space that the
4369 * allocator should try to satisfy. In some cases
4370 * @num_bytes may be larger than what is required and if
4371 * the filesystem is fragmented then allocation fails.
4372 * However, the presence of @min_alloc_size gives a
4373 * chance to try and satisfy the smaller allocation.
4375 * @empty_size - A hint that you plan on doing more COW. This is the
4376 * size in bytes the allocator should try to find free
4377 * next to the block it returns. This is just a hint and
4378 * may be ignored by the allocator.
4380 * @hint_byte - Hint to the allocator to start searching above the byte
4381 * address passed. It might be ignored.
4383 * @ins - This key is modified to record the found hole. It will
4384 * have the following values:
4385 * ins->objectid == start position
4386 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4387 * ins->offset == the size of the hole.
4389 * @is_data - Boolean flag indicating whether an extent is
4390 * allocated for data (true) or metadata (false)
4392 * @delalloc - Boolean flag indicating whether this allocation is for
4393 * delalloc or not. If 'true' data_rwsem of block groups
4394 * is going to be acquired.
4397 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4398 * case -ENOSPC is returned then @ins->offset will contain the size of the
4399 * largest available hole the allocator managed to find.
4401 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4402 u64 num_bytes, u64 min_alloc_size,
4403 u64 empty_size, u64 hint_byte,
4404 struct btrfs_key *ins, int is_data, int delalloc)
4406 struct btrfs_fs_info *fs_info = root->fs_info;
4407 bool final_tried = num_bytes == min_alloc_size;
4410 bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4412 flags = get_alloc_profile_by_root(root, is_data);
4414 WARN_ON(num_bytes < fs_info->sectorsize);
4415 ret = find_free_extent(root, ram_bytes, num_bytes, empty_size,
4416 hint_byte, ins, flags, delalloc);
4417 if (!ret && !is_data) {
4418 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4419 } else if (ret == -ENOSPC) {
4420 if (!final_tried && ins->offset) {
4421 num_bytes = min(num_bytes >> 1, ins->offset);
4422 num_bytes = round_down(num_bytes,
4423 fs_info->sectorsize);
4424 num_bytes = max(num_bytes, min_alloc_size);
4425 ram_bytes = num_bytes;
4426 if (num_bytes == min_alloc_size)
4429 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4430 struct btrfs_space_info *sinfo;
4432 sinfo = btrfs_find_space_info(fs_info, flags);
4434 "allocation failed flags %llu, wanted %llu tree-log %d",
4435 flags, num_bytes, for_treelog);
4437 btrfs_dump_space_info(fs_info, sinfo,
4445 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4446 u64 start, u64 len, int delalloc)
4448 struct btrfs_block_group *cache;
4450 cache = btrfs_lookup_block_group(fs_info, start);
4452 btrfs_err(fs_info, "Unable to find block group for %llu",
4457 btrfs_add_free_space(cache, start, len);
4458 btrfs_free_reserved_bytes(cache, len, delalloc);
4459 trace_btrfs_reserved_extent_free(fs_info, start, len);
4461 btrfs_put_block_group(cache);
4465 int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start,
4468 struct btrfs_block_group *cache;
4471 cache = btrfs_lookup_block_group(trans->fs_info, start);
4473 btrfs_err(trans->fs_info, "unable to find block group for %llu",
4478 ret = pin_down_extent(trans, cache, start, len, 1);
4479 btrfs_put_block_group(cache);
4483 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4484 u64 parent, u64 root_objectid,
4485 u64 flags, u64 owner, u64 offset,
4486 struct btrfs_key *ins, int ref_mod)
4488 struct btrfs_fs_info *fs_info = trans->fs_info;
4490 struct btrfs_extent_item *extent_item;
4491 struct btrfs_extent_inline_ref *iref;
4492 struct btrfs_path *path;
4493 struct extent_buffer *leaf;
4498 type = BTRFS_SHARED_DATA_REF_KEY;
4500 type = BTRFS_EXTENT_DATA_REF_KEY;
4502 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4504 path = btrfs_alloc_path();
4508 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4511 btrfs_free_path(path);
4515 leaf = path->nodes[0];
4516 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4517 struct btrfs_extent_item);
4518 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4519 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4520 btrfs_set_extent_flags(leaf, extent_item,
4521 flags | BTRFS_EXTENT_FLAG_DATA);
4523 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4524 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4526 struct btrfs_shared_data_ref *ref;
4527 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4528 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4529 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4531 struct btrfs_extent_data_ref *ref;
4532 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4533 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4534 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4535 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4536 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4539 btrfs_mark_buffer_dirty(path->nodes[0]);
4540 btrfs_free_path(path);
4542 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
4546 ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1);
4547 if (ret) { /* -ENOENT, logic error */
4548 btrfs_err(fs_info, "update block group failed for %llu %llu",
4549 ins->objectid, ins->offset);
4552 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
4556 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4557 struct btrfs_delayed_ref_node *node,
4558 struct btrfs_delayed_extent_op *extent_op)
4560 struct btrfs_fs_info *fs_info = trans->fs_info;
4562 struct btrfs_extent_item *extent_item;
4563 struct btrfs_key extent_key;
4564 struct btrfs_tree_block_info *block_info;
4565 struct btrfs_extent_inline_ref *iref;
4566 struct btrfs_path *path;
4567 struct extent_buffer *leaf;
4568 struct btrfs_delayed_tree_ref *ref;
4569 u32 size = sizeof(*extent_item) + sizeof(*iref);
4571 u64 flags = extent_op->flags_to_set;
4572 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4574 ref = btrfs_delayed_node_to_tree_ref(node);
4576 extent_key.objectid = node->bytenr;
4577 if (skinny_metadata) {
4578 extent_key.offset = ref->level;
4579 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4580 num_bytes = fs_info->nodesize;
4582 extent_key.offset = node->num_bytes;
4583 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4584 size += sizeof(*block_info);
4585 num_bytes = node->num_bytes;
4588 path = btrfs_alloc_path();
4592 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4595 btrfs_free_path(path);
4599 leaf = path->nodes[0];
4600 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4601 struct btrfs_extent_item);
4602 btrfs_set_extent_refs(leaf, extent_item, 1);
4603 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4604 btrfs_set_extent_flags(leaf, extent_item,
4605 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4607 if (skinny_metadata) {
4608 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4610 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4611 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4612 btrfs_set_tree_block_level(leaf, block_info, ref->level);
4613 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4616 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4617 btrfs_set_extent_inline_ref_type(leaf, iref,
4618 BTRFS_SHARED_BLOCK_REF_KEY);
4619 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4621 btrfs_set_extent_inline_ref_type(leaf, iref,
4622 BTRFS_TREE_BLOCK_REF_KEY);
4623 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4626 btrfs_mark_buffer_dirty(leaf);
4627 btrfs_free_path(path);
4629 ret = remove_from_free_space_tree(trans, extent_key.objectid,
4634 ret = btrfs_update_block_group(trans, extent_key.objectid,
4635 fs_info->nodesize, 1);
4636 if (ret) { /* -ENOENT, logic error */
4637 btrfs_err(fs_info, "update block group failed for %llu %llu",
4638 extent_key.objectid, extent_key.offset);
4642 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
4647 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4648 struct btrfs_root *root, u64 owner,
4649 u64 offset, u64 ram_bytes,
4650 struct btrfs_key *ins)
4652 struct btrfs_ref generic_ref = { 0 };
4654 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4656 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4657 ins->objectid, ins->offset, 0);
4658 btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset);
4659 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4661 return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes);
4665 * this is used by the tree logging recovery code. It records that
4666 * an extent has been allocated and makes sure to clear the free
4667 * space cache bits as well
4669 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4670 u64 root_objectid, u64 owner, u64 offset,
4671 struct btrfs_key *ins)
4673 struct btrfs_fs_info *fs_info = trans->fs_info;
4675 struct btrfs_block_group *block_group;
4676 struct btrfs_space_info *space_info;
4679 * Mixed block groups will exclude before processing the log so we only
4680 * need to do the exclude dance if this fs isn't mixed.
4682 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4683 ret = __exclude_logged_extent(fs_info, ins->objectid,
4689 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4693 space_info = block_group->space_info;
4694 spin_lock(&space_info->lock);
4695 spin_lock(&block_group->lock);
4696 space_info->bytes_reserved += ins->offset;
4697 block_group->reserved += ins->offset;
4698 spin_unlock(&block_group->lock);
4699 spin_unlock(&space_info->lock);
4701 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4704 btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
4705 btrfs_put_block_group(block_group);
4709 static struct extent_buffer *
4710 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4711 u64 bytenr, int level, u64 owner,
4712 enum btrfs_lock_nesting nest)
4714 struct btrfs_fs_info *fs_info = root->fs_info;
4715 struct extent_buffer *buf;
4717 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level);
4722 * Extra safety check in case the extent tree is corrupted and extent
4723 * allocator chooses to use a tree block which is already used and
4726 if (buf->lock_owner == current->pid) {
4727 btrfs_err_rl(fs_info,
4728 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4729 buf->start, btrfs_header_owner(buf), current->pid);
4730 free_extent_buffer(buf);
4731 return ERR_PTR(-EUCLEAN);
4735 * This needs to stay, because we could allocate a freed block from an
4736 * old tree into a new tree, so we need to make sure this new block is
4737 * set to the appropriate level and owner.
4739 btrfs_set_buffer_lockdep_class(owner, buf, level);
4740 __btrfs_tree_lock(buf, nest);
4741 btrfs_clean_tree_block(buf);
4742 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
4743 clear_bit(EXTENT_BUFFER_NO_CHECK, &buf->bflags);
4745 set_extent_buffer_uptodate(buf);
4747 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
4748 btrfs_set_header_level(buf, level);
4749 btrfs_set_header_bytenr(buf, buf->start);
4750 btrfs_set_header_generation(buf, trans->transid);
4751 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
4752 btrfs_set_header_owner(buf, owner);
4753 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
4754 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
4755 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4756 buf->log_index = root->log_transid % 2;
4758 * we allow two log transactions at a time, use different
4759 * EXTENT bit to differentiate dirty pages.
4761 if (buf->log_index == 0)
4762 set_extent_dirty(&root->dirty_log_pages, buf->start,
4763 buf->start + buf->len - 1, GFP_NOFS);
4765 set_extent_new(&root->dirty_log_pages, buf->start,
4766 buf->start + buf->len - 1);
4768 buf->log_index = -1;
4769 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4770 buf->start + buf->len - 1, GFP_NOFS);
4772 /* this returns a buffer locked for blocking */
4777 * finds a free extent and does all the dirty work required for allocation
4778 * returns the tree buffer or an ERR_PTR on error.
4780 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
4781 struct btrfs_root *root,
4782 u64 parent, u64 root_objectid,
4783 const struct btrfs_disk_key *key,
4784 int level, u64 hint,
4786 enum btrfs_lock_nesting nest)
4788 struct btrfs_fs_info *fs_info = root->fs_info;
4789 struct btrfs_key ins;
4790 struct btrfs_block_rsv *block_rsv;
4791 struct extent_buffer *buf;
4792 struct btrfs_delayed_extent_op *extent_op;
4793 struct btrfs_ref generic_ref = { 0 };
4796 u32 blocksize = fs_info->nodesize;
4797 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4799 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4800 if (btrfs_is_testing(fs_info)) {
4801 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
4802 level, root_objectid, nest);
4804 root->alloc_bytenr += blocksize;
4809 block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
4810 if (IS_ERR(block_rsv))
4811 return ERR_CAST(block_rsv);
4813 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
4814 empty_size, hint, &ins, 0, 0);
4818 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
4819 root_objectid, nest);
4822 goto out_free_reserved;
4825 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4827 parent = ins.objectid;
4828 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4832 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4833 extent_op = btrfs_alloc_delayed_extent_op();
4839 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4841 memset(&extent_op->key, 0, sizeof(extent_op->key));
4842 extent_op->flags_to_set = flags;
4843 extent_op->update_key = skinny_metadata ? false : true;
4844 extent_op->update_flags = true;
4845 extent_op->is_data = false;
4846 extent_op->level = level;
4848 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4849 ins.objectid, ins.offset, parent);
4850 generic_ref.real_root = root->root_key.objectid;
4851 btrfs_init_tree_ref(&generic_ref, level, root_objectid);
4852 btrfs_ref_tree_mod(fs_info, &generic_ref);
4853 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op);
4855 goto out_free_delayed;
4860 btrfs_free_delayed_extent_op(extent_op);
4862 btrfs_tree_unlock(buf);
4863 free_extent_buffer(buf);
4865 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
4867 btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
4868 return ERR_PTR(ret);
4871 struct walk_control {
4872 u64 refs[BTRFS_MAX_LEVEL];
4873 u64 flags[BTRFS_MAX_LEVEL];
4874 struct btrfs_key update_progress;
4875 struct btrfs_key drop_progress;
4887 #define DROP_REFERENCE 1
4888 #define UPDATE_BACKREF 2
4890 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4891 struct btrfs_root *root,
4892 struct walk_control *wc,
4893 struct btrfs_path *path)
4895 struct btrfs_fs_info *fs_info = root->fs_info;
4901 struct btrfs_key key;
4902 struct extent_buffer *eb;
4907 if (path->slots[wc->level] < wc->reada_slot) {
4908 wc->reada_count = wc->reada_count * 2 / 3;
4909 wc->reada_count = max(wc->reada_count, 2);
4911 wc->reada_count = wc->reada_count * 3 / 2;
4912 wc->reada_count = min_t(int, wc->reada_count,
4913 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
4916 eb = path->nodes[wc->level];
4917 nritems = btrfs_header_nritems(eb);
4919 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4920 if (nread >= wc->reada_count)
4924 bytenr = btrfs_node_blockptr(eb, slot);
4925 generation = btrfs_node_ptr_generation(eb, slot);
4927 if (slot == path->slots[wc->level])
4930 if (wc->stage == UPDATE_BACKREF &&
4931 generation <= root->root_key.offset)
4934 /* We don't lock the tree block, it's OK to be racy here */
4935 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
4936 wc->level - 1, 1, &refs,
4938 /* We don't care about errors in readahead. */
4943 if (wc->stage == DROP_REFERENCE) {
4947 if (wc->level == 1 &&
4948 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4950 if (!wc->update_ref ||
4951 generation <= root->root_key.offset)
4953 btrfs_node_key_to_cpu(eb, &key, slot);
4954 ret = btrfs_comp_cpu_keys(&key,
4955 &wc->update_progress);
4959 if (wc->level == 1 &&
4960 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4964 btrfs_readahead_node_child(eb, slot);
4967 wc->reada_slot = slot;
4971 * helper to process tree block while walking down the tree.
4973 * when wc->stage == UPDATE_BACKREF, this function updates
4974 * back refs for pointers in the block.
4976 * NOTE: return value 1 means we should stop walking down.
4978 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4979 struct btrfs_root *root,
4980 struct btrfs_path *path,
4981 struct walk_control *wc, int lookup_info)
4983 struct btrfs_fs_info *fs_info = root->fs_info;
4984 int level = wc->level;
4985 struct extent_buffer *eb = path->nodes[level];
4986 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4989 if (wc->stage == UPDATE_BACKREF &&
4990 btrfs_header_owner(eb) != root->root_key.objectid)
4994 * when reference count of tree block is 1, it won't increase
4995 * again. once full backref flag is set, we never clear it.
4998 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4999 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5000 BUG_ON(!path->locks[level]);
5001 ret = btrfs_lookup_extent_info(trans, fs_info,
5002 eb->start, level, 1,
5005 BUG_ON(ret == -ENOMEM);
5008 BUG_ON(wc->refs[level] == 0);
5011 if (wc->stage == DROP_REFERENCE) {
5012 if (wc->refs[level] > 1)
5015 if (path->locks[level] && !wc->keep_locks) {
5016 btrfs_tree_unlock_rw(eb, path->locks[level]);
5017 path->locks[level] = 0;
5022 /* wc->stage == UPDATE_BACKREF */
5023 if (!(wc->flags[level] & flag)) {
5024 BUG_ON(!path->locks[level]);
5025 ret = btrfs_inc_ref(trans, root, eb, 1);
5026 BUG_ON(ret); /* -ENOMEM */
5027 ret = btrfs_dec_ref(trans, root, eb, 0);
5028 BUG_ON(ret); /* -ENOMEM */
5029 ret = btrfs_set_disk_extent_flags(trans, eb, flag,
5030 btrfs_header_level(eb), 0);
5031 BUG_ON(ret); /* -ENOMEM */
5032 wc->flags[level] |= flag;
5036 * the block is shared by multiple trees, so it's not good to
5037 * keep the tree lock
5039 if (path->locks[level] && level > 0) {
5040 btrfs_tree_unlock_rw(eb, path->locks[level]);
5041 path->locks[level] = 0;
5047 * This is used to verify a ref exists for this root to deal with a bug where we
5048 * would have a drop_progress key that hadn't been updated properly.
5050 static int check_ref_exists(struct btrfs_trans_handle *trans,
5051 struct btrfs_root *root, u64 bytenr, u64 parent,
5054 struct btrfs_path *path;
5055 struct btrfs_extent_inline_ref *iref;
5058 path = btrfs_alloc_path();
5062 ret = lookup_extent_backref(trans, path, &iref, bytenr,
5063 root->fs_info->nodesize, parent,
5064 root->root_key.objectid, level, 0);
5065 btrfs_free_path(path);
5074 * helper to process tree block pointer.
5076 * when wc->stage == DROP_REFERENCE, this function checks
5077 * reference count of the block pointed to. if the block
5078 * is shared and we need update back refs for the subtree
5079 * rooted at the block, this function changes wc->stage to
5080 * UPDATE_BACKREF. if the block is shared and there is no
5081 * need to update back, this function drops the reference
5084 * NOTE: return value 1 means we should stop walking down.
5086 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5087 struct btrfs_root *root,
5088 struct btrfs_path *path,
5089 struct walk_control *wc, int *lookup_info)
5091 struct btrfs_fs_info *fs_info = root->fs_info;
5095 struct btrfs_key key;
5096 struct btrfs_key first_key;
5097 struct btrfs_ref ref = { 0 };
5098 struct extent_buffer *next;
5099 int level = wc->level;
5102 bool need_account = false;
5104 generation = btrfs_node_ptr_generation(path->nodes[level],
5105 path->slots[level]);
5107 * if the lower level block was created before the snapshot
5108 * was created, we know there is no need to update back refs
5111 if (wc->stage == UPDATE_BACKREF &&
5112 generation <= root->root_key.offset) {
5117 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5118 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
5119 path->slots[level]);
5121 next = find_extent_buffer(fs_info, bytenr);
5123 next = btrfs_find_create_tree_block(fs_info, bytenr,
5124 root->root_key.objectid, level - 1);
5126 return PTR_ERR(next);
5129 btrfs_tree_lock(next);
5131 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
5132 &wc->refs[level - 1],
5133 &wc->flags[level - 1]);
5137 if (unlikely(wc->refs[level - 1] == 0)) {
5138 btrfs_err(fs_info, "Missing references.");
5144 if (wc->stage == DROP_REFERENCE) {
5145 if (wc->refs[level - 1] > 1) {
5146 need_account = true;
5148 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5151 if (!wc->update_ref ||
5152 generation <= root->root_key.offset)
5155 btrfs_node_key_to_cpu(path->nodes[level], &key,
5156 path->slots[level]);
5157 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5161 wc->stage = UPDATE_BACKREF;
5162 wc->shared_level = level - 1;
5166 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5170 if (!btrfs_buffer_uptodate(next, generation, 0)) {
5171 btrfs_tree_unlock(next);
5172 free_extent_buffer(next);
5178 if (reada && level == 1)
5179 reada_walk_down(trans, root, wc, path);
5180 next = read_tree_block(fs_info, bytenr, root->root_key.objectid,
5181 generation, level - 1, &first_key);
5183 return PTR_ERR(next);
5184 } else if (!extent_buffer_uptodate(next)) {
5185 free_extent_buffer(next);
5188 btrfs_tree_lock(next);
5192 ASSERT(level == btrfs_header_level(next));
5193 if (level != btrfs_header_level(next)) {
5194 btrfs_err(root->fs_info, "mismatched level");
5198 path->nodes[level] = next;
5199 path->slots[level] = 0;
5200 path->locks[level] = BTRFS_WRITE_LOCK;
5206 wc->refs[level - 1] = 0;
5207 wc->flags[level - 1] = 0;
5208 if (wc->stage == DROP_REFERENCE) {
5209 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5210 parent = path->nodes[level]->start;
5212 ASSERT(root->root_key.objectid ==
5213 btrfs_header_owner(path->nodes[level]));
5214 if (root->root_key.objectid !=
5215 btrfs_header_owner(path->nodes[level])) {
5216 btrfs_err(root->fs_info,
5217 "mismatched block owner");
5225 * If we had a drop_progress we need to verify the refs are set
5226 * as expected. If we find our ref then we know that from here
5227 * on out everything should be correct, and we can clear the
5230 if (wc->restarted) {
5231 ret = check_ref_exists(trans, root, bytenr, parent,
5242 * Reloc tree doesn't contribute to qgroup numbers, and we have
5243 * already accounted them at merge time (replace_path),
5244 * thus we could skip expensive subtree trace here.
5246 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
5248 ret = btrfs_qgroup_trace_subtree(trans, next,
5249 generation, level - 1);
5251 btrfs_err_rl(fs_info,
5252 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5258 * We need to update the next key in our walk control so we can
5259 * update the drop_progress key accordingly. We don't care if
5260 * find_next_key doesn't find a key because that means we're at
5261 * the end and are going to clean up now.
5263 wc->drop_level = level;
5264 find_next_key(path, level, &wc->drop_progress);
5266 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
5267 fs_info->nodesize, parent);
5268 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid);
5269 ret = btrfs_free_extent(trans, &ref);
5278 btrfs_tree_unlock(next);
5279 free_extent_buffer(next);
5285 * helper to process tree block while walking up the tree.
5287 * when wc->stage == DROP_REFERENCE, this function drops
5288 * reference count on the block.
5290 * when wc->stage == UPDATE_BACKREF, this function changes
5291 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5292 * to UPDATE_BACKREF previously while processing the block.
5294 * NOTE: return value 1 means we should stop walking up.
5296 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5297 struct btrfs_root *root,
5298 struct btrfs_path *path,
5299 struct walk_control *wc)
5301 struct btrfs_fs_info *fs_info = root->fs_info;
5303 int level = wc->level;
5304 struct extent_buffer *eb = path->nodes[level];
5307 if (wc->stage == UPDATE_BACKREF) {
5308 BUG_ON(wc->shared_level < level);
5309 if (level < wc->shared_level)
5312 ret = find_next_key(path, level + 1, &wc->update_progress);
5316 wc->stage = DROP_REFERENCE;
5317 wc->shared_level = -1;
5318 path->slots[level] = 0;
5321 * check reference count again if the block isn't locked.
5322 * we should start walking down the tree again if reference
5325 if (!path->locks[level]) {
5327 btrfs_tree_lock(eb);
5328 path->locks[level] = BTRFS_WRITE_LOCK;
5330 ret = btrfs_lookup_extent_info(trans, fs_info,
5331 eb->start, level, 1,
5335 btrfs_tree_unlock_rw(eb, path->locks[level]);
5336 path->locks[level] = 0;
5339 BUG_ON(wc->refs[level] == 0);
5340 if (wc->refs[level] == 1) {
5341 btrfs_tree_unlock_rw(eb, path->locks[level]);
5342 path->locks[level] = 0;
5348 /* wc->stage == DROP_REFERENCE */
5349 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5351 if (wc->refs[level] == 1) {
5353 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5354 ret = btrfs_dec_ref(trans, root, eb, 1);
5356 ret = btrfs_dec_ref(trans, root, eb, 0);
5357 BUG_ON(ret); /* -ENOMEM */
5358 if (is_fstree(root->root_key.objectid)) {
5359 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5361 btrfs_err_rl(fs_info,
5362 "error %d accounting leaf items, quota is out of sync, rescan required",
5367 /* make block locked assertion in btrfs_clean_tree_block happy */
5368 if (!path->locks[level] &&
5369 btrfs_header_generation(eb) == trans->transid) {
5370 btrfs_tree_lock(eb);
5371 path->locks[level] = BTRFS_WRITE_LOCK;
5373 btrfs_clean_tree_block(eb);
5376 if (eb == root->node) {
5377 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5379 else if (root->root_key.objectid != btrfs_header_owner(eb))
5380 goto owner_mismatch;
5382 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5383 parent = path->nodes[level + 1]->start;
5384 else if (root->root_key.objectid !=
5385 btrfs_header_owner(path->nodes[level + 1]))
5386 goto owner_mismatch;
5389 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
5391 wc->refs[level] = 0;
5392 wc->flags[level] = 0;
5396 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5397 btrfs_header_owner(eb), root->root_key.objectid);
5401 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5402 struct btrfs_root *root,
5403 struct btrfs_path *path,
5404 struct walk_control *wc)
5406 int level = wc->level;
5407 int lookup_info = 1;
5410 while (level >= 0) {
5411 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5418 if (path->slots[level] >=
5419 btrfs_header_nritems(path->nodes[level]))
5422 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5424 path->slots[level]++;
5433 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5434 struct btrfs_root *root,
5435 struct btrfs_path *path,
5436 struct walk_control *wc, int max_level)
5438 int level = wc->level;
5441 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5442 while (level < max_level && path->nodes[level]) {
5444 if (path->slots[level] + 1 <
5445 btrfs_header_nritems(path->nodes[level])) {
5446 path->slots[level]++;
5449 ret = walk_up_proc(trans, root, path, wc);
5455 if (path->locks[level]) {
5456 btrfs_tree_unlock_rw(path->nodes[level],
5457 path->locks[level]);
5458 path->locks[level] = 0;
5460 free_extent_buffer(path->nodes[level]);
5461 path->nodes[level] = NULL;
5469 * drop a subvolume tree.
5471 * this function traverses the tree freeing any blocks that only
5472 * referenced by the tree.
5474 * when a shared tree block is found. this function decreases its
5475 * reference count by one. if update_ref is true, this function
5476 * also make sure backrefs for the shared block and all lower level
5477 * blocks are properly updated.
5479 * If called with for_reloc == 0, may exit early with -EAGAIN
5481 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
5483 struct btrfs_fs_info *fs_info = root->fs_info;
5484 struct btrfs_path *path;
5485 struct btrfs_trans_handle *trans;
5486 struct btrfs_root *tree_root = fs_info->tree_root;
5487 struct btrfs_root_item *root_item = &root->root_item;
5488 struct walk_control *wc;
5489 struct btrfs_key key;
5493 bool root_dropped = false;
5495 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5497 path = btrfs_alloc_path();
5503 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5505 btrfs_free_path(path);
5511 * Use join to avoid potential EINTR from transaction start. See
5512 * wait_reserve_ticket and the whole reservation callchain.
5515 trans = btrfs_join_transaction(tree_root);
5517 trans = btrfs_start_transaction(tree_root, 0);
5518 if (IS_ERR(trans)) {
5519 err = PTR_ERR(trans);
5523 err = btrfs_run_delayed_items(trans);
5528 * This will help us catch people modifying the fs tree while we're
5529 * dropping it. It is unsafe to mess with the fs tree while it's being
5530 * dropped as we unlock the root node and parent nodes as we walk down
5531 * the tree, assuming nothing will change. If something does change
5532 * then we'll have stale information and drop references to blocks we've
5535 set_bit(BTRFS_ROOT_DELETING, &root->state);
5536 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5537 level = btrfs_header_level(root->node);
5538 path->nodes[level] = btrfs_lock_root_node(root);
5539 path->slots[level] = 0;
5540 path->locks[level] = BTRFS_WRITE_LOCK;
5541 memset(&wc->update_progress, 0,
5542 sizeof(wc->update_progress));
5544 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5545 memcpy(&wc->update_progress, &key,
5546 sizeof(wc->update_progress));
5548 level = btrfs_root_drop_level(root_item);
5550 path->lowest_level = level;
5551 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5552 path->lowest_level = 0;
5560 * unlock our path, this is safe because only this
5561 * function is allowed to delete this snapshot
5563 btrfs_unlock_up_safe(path, 0);
5565 level = btrfs_header_level(root->node);
5567 btrfs_tree_lock(path->nodes[level]);
5568 path->locks[level] = BTRFS_WRITE_LOCK;
5570 ret = btrfs_lookup_extent_info(trans, fs_info,
5571 path->nodes[level]->start,
5572 level, 1, &wc->refs[level],
5578 BUG_ON(wc->refs[level] == 0);
5580 if (level == btrfs_root_drop_level(root_item))
5583 btrfs_tree_unlock(path->nodes[level]);
5584 path->locks[level] = 0;
5585 WARN_ON(wc->refs[level] != 1);
5590 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5592 wc->shared_level = -1;
5593 wc->stage = DROP_REFERENCE;
5594 wc->update_ref = update_ref;
5596 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5600 ret = walk_down_tree(trans, root, path, wc);
5606 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5613 BUG_ON(wc->stage != DROP_REFERENCE);
5617 if (wc->stage == DROP_REFERENCE) {
5618 wc->drop_level = wc->level;
5619 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5621 path->slots[wc->drop_level]);
5623 btrfs_cpu_key_to_disk(&root_item->drop_progress,
5624 &wc->drop_progress);
5625 btrfs_set_root_drop_level(root_item, wc->drop_level);
5627 BUG_ON(wc->level == 0);
5628 if (btrfs_should_end_transaction(trans) ||
5629 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5630 ret = btrfs_update_root(trans, tree_root,
5634 btrfs_abort_transaction(trans, ret);
5639 btrfs_end_transaction_throttle(trans);
5640 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5641 btrfs_debug(fs_info,
5642 "drop snapshot early exit");
5648 * Use join to avoid potential EINTR from transaction
5649 * start. See wait_reserve_ticket and the whole
5650 * reservation callchain.
5653 trans = btrfs_join_transaction(tree_root);
5655 trans = btrfs_start_transaction(tree_root, 0);
5656 if (IS_ERR(trans)) {
5657 err = PTR_ERR(trans);
5662 btrfs_release_path(path);
5666 ret = btrfs_del_root(trans, &root->root_key);
5668 btrfs_abort_transaction(trans, ret);
5673 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5674 ret = btrfs_find_root(tree_root, &root->root_key, path,
5677 btrfs_abort_transaction(trans, ret);
5680 } else if (ret > 0) {
5681 /* if we fail to delete the orphan item this time
5682 * around, it'll get picked up the next time.
5684 * The most common failure here is just -ENOENT.
5686 btrfs_del_orphan_item(trans, tree_root,
5687 root->root_key.objectid);
5692 * This subvolume is going to be completely dropped, and won't be
5693 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
5694 * commit transaction time. So free it here manually.
5696 btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
5697 btrfs_qgroup_free_meta_all_pertrans(root);
5699 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state))
5700 btrfs_add_dropped_root(trans, root);
5702 btrfs_put_root(root);
5703 root_dropped = true;
5705 btrfs_end_transaction_throttle(trans);
5708 btrfs_free_path(path);
5711 * So if we need to stop dropping the snapshot for whatever reason we
5712 * need to make sure to add it back to the dead root list so that we
5713 * keep trying to do the work later. This also cleans up roots if we
5714 * don't have it in the radix (like when we recover after a power fail
5715 * or unmount) so we don't leak memory.
5717 if (!for_reloc && !root_dropped)
5718 btrfs_add_dead_root(root);
5723 * drop subtree rooted at tree block 'node'.
5725 * NOTE: this function will unlock and release tree block 'node'
5726 * only used by relocation code
5728 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5729 struct btrfs_root *root,
5730 struct extent_buffer *node,
5731 struct extent_buffer *parent)
5733 struct btrfs_fs_info *fs_info = root->fs_info;
5734 struct btrfs_path *path;
5735 struct walk_control *wc;
5741 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5743 path = btrfs_alloc_path();
5747 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5749 btrfs_free_path(path);
5753 btrfs_assert_tree_locked(parent);
5754 parent_level = btrfs_header_level(parent);
5755 atomic_inc(&parent->refs);
5756 path->nodes[parent_level] = parent;
5757 path->slots[parent_level] = btrfs_header_nritems(parent);
5759 btrfs_assert_tree_locked(node);
5760 level = btrfs_header_level(node);
5761 path->nodes[level] = node;
5762 path->slots[level] = 0;
5763 path->locks[level] = BTRFS_WRITE_LOCK;
5765 wc->refs[parent_level] = 1;
5766 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5768 wc->shared_level = -1;
5769 wc->stage = DROP_REFERENCE;
5772 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5775 wret = walk_down_tree(trans, root, path, wc);
5781 wret = walk_up_tree(trans, root, path, wc, parent_level);
5789 btrfs_free_path(path);
5794 * helper to account the unused space of all the readonly block group in the
5795 * space_info. takes mirrors into account.
5797 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
5799 struct btrfs_block_group *block_group;
5803 /* It's df, we don't care if it's racy */
5804 if (list_empty(&sinfo->ro_bgs))
5807 spin_lock(&sinfo->lock);
5808 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
5809 spin_lock(&block_group->lock);
5811 if (!block_group->ro) {
5812 spin_unlock(&block_group->lock);
5816 factor = btrfs_bg_type_to_factor(block_group->flags);
5817 free_bytes += (block_group->length -
5818 block_group->used) * factor;
5820 spin_unlock(&block_group->lock);
5822 spin_unlock(&sinfo->lock);
5827 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
5830 return unpin_extent_range(fs_info, start, end, false);
5834 * It used to be that old block groups would be left around forever.
5835 * Iterating over them would be enough to trim unused space. Since we
5836 * now automatically remove them, we also need to iterate over unallocated
5839 * We don't want a transaction for this since the discard may take a
5840 * substantial amount of time. We don't require that a transaction be
5841 * running, but we do need to take a running transaction into account
5842 * to ensure that we're not discarding chunks that were released or
5843 * allocated in the current transaction.
5845 * Holding the chunks lock will prevent other threads from allocating
5846 * or releasing chunks, but it won't prevent a running transaction
5847 * from committing and releasing the memory that the pending chunks
5848 * list head uses. For that, we need to take a reference to the
5849 * transaction and hold the commit root sem. We only need to hold
5850 * it while performing the free space search since we have already
5851 * held back allocations.
5853 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
5855 u64 start = SZ_1M, len = 0, end = 0;
5860 /* Discard not supported = nothing to do. */
5861 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
5864 /* Not writable = nothing to do. */
5865 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5868 /* No free space = nothing to do. */
5869 if (device->total_bytes <= device->bytes_used)
5875 struct btrfs_fs_info *fs_info = device->fs_info;
5878 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
5882 find_first_clear_extent_bit(&device->alloc_state, start,
5884 CHUNK_TRIMMED | CHUNK_ALLOCATED);
5886 /* Check if there are any CHUNK_* bits left */
5887 if (start > device->total_bytes) {
5888 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
5889 btrfs_warn_in_rcu(fs_info,
5890 "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
5891 start, end - start + 1,
5892 rcu_str_deref(device->name),
5893 device->total_bytes);
5894 mutex_unlock(&fs_info->chunk_mutex);
5899 /* Ensure we skip the reserved area in the first 1M */
5900 start = max_t(u64, start, SZ_1M);
5903 * If find_first_clear_extent_bit find a range that spans the
5904 * end of the device it will set end to -1, in this case it's up
5905 * to the caller to trim the value to the size of the device.
5907 end = min(end, device->total_bytes - 1);
5909 len = end - start + 1;
5911 /* We didn't find any extents */
5913 mutex_unlock(&fs_info->chunk_mutex);
5918 ret = btrfs_issue_discard(device->bdev, start, len,
5921 set_extent_bits(&device->alloc_state, start,
5924 mutex_unlock(&fs_info->chunk_mutex);
5932 if (fatal_signal_pending(current)) {
5944 * Trim the whole filesystem by:
5945 * 1) trimming the free space in each block group
5946 * 2) trimming the unallocated space on each device
5948 * This will also continue trimming even if a block group or device encounters
5949 * an error. The return value will be the last error, or 0 if nothing bad
5952 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
5954 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5955 struct btrfs_block_group *cache = NULL;
5956 struct btrfs_device *device;
5958 u64 range_end = U64_MAX;
5969 * Check range overflow if range->len is set.
5970 * The default range->len is U64_MAX.
5972 if (range->len != U64_MAX &&
5973 check_add_overflow(range->start, range->len, &range_end))
5976 cache = btrfs_lookup_first_block_group(fs_info, range->start);
5977 for (; cache; cache = btrfs_next_block_group(cache)) {
5978 if (cache->start >= range_end) {
5979 btrfs_put_block_group(cache);
5983 start = max(range->start, cache->start);
5984 end = min(range_end, cache->start + cache->length);
5986 if (end - start >= range->minlen) {
5987 if (!btrfs_block_group_done(cache)) {
5988 ret = btrfs_cache_block_group(cache, 0);
5994 ret = btrfs_wait_block_group_cache_done(cache);
6001 ret = btrfs_trim_block_group(cache,
6007 trimmed += group_trimmed;
6018 "failed to trim %llu block group(s), last error %d",
6021 mutex_lock(&fs_devices->device_list_mutex);
6022 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6023 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
6026 ret = btrfs_trim_free_extents(device, &group_trimmed);
6033 trimmed += group_trimmed;
6035 mutex_unlock(&fs_devices->device_list_mutex);
6039 "failed to trim %llu device(s), last error %d",
6040 dev_failed, dev_ret);
6041 range->len = trimmed;