1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/signal.h>
8 #include <linux/pagemap.h>
9 #include <linux/writeback.h>
10 #include <linux/blkdev.h>
11 #include <linux/sort.h>
12 #include <linux/rcupdate.h>
13 #include <linux/kthread.h>
14 #include <linux/slab.h>
15 #include <linux/ratelimit.h>
16 #include <linux/percpu_counter.h>
17 #include <linux/lockdep.h>
18 #include <linux/crc32c.h>
20 #include "extent-tree.h"
21 #include "transaction.h"
23 #include "print-tree.h"
27 #include "free-space-cache.h"
28 #include "free-space-tree.h"
30 #include "ref-verify.h"
31 #include "space-info.h"
32 #include "block-rsv.h"
35 #include "dev-replace.h"
37 #include "accessors.h"
38 #include "root-tree.h"
39 #include "file-item.h"
41 #include "tree-checker.h"
42 #include "raid-stripe-tree.h"
44 #undef SCRAMBLE_DELAYED_REFS
47 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
48 struct btrfs_delayed_ref_head *href,
49 struct btrfs_delayed_ref_node *node, u64 parent,
50 u64 root_objectid, u64 owner_objectid,
52 struct btrfs_delayed_extent_op *extra_op);
53 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
54 struct extent_buffer *leaf,
55 struct btrfs_extent_item *ei);
56 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
57 u64 parent, u64 root_objectid,
58 u64 flags, u64 owner, u64 offset,
59 struct btrfs_key *ins, int ref_mod, u64 oref_root);
60 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
61 struct btrfs_delayed_ref_node *node,
62 struct btrfs_delayed_extent_op *extent_op);
63 static int find_next_key(struct btrfs_path *path, int level,
64 struct btrfs_key *key);
66 static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
68 return (cache->flags & bits) == bits;
71 /* simple helper to search for an existing data extent at a given offset */
72 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
74 struct btrfs_root *root = btrfs_extent_root(fs_info, start);
77 struct btrfs_path *path;
79 path = btrfs_alloc_path();
85 key.type = BTRFS_EXTENT_ITEM_KEY;
86 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
87 btrfs_free_path(path);
92 * helper function to lookup reference count and flags of a tree block.
94 * the head node for delayed ref is used to store the sum of all the
95 * reference count modifications queued up in the rbtree. the head
96 * node may also store the extent flags to set. This way you can check
97 * to see what the reference count and extent flags would be if all of
98 * the delayed refs are not processed.
100 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
101 struct btrfs_fs_info *fs_info, u64 bytenr,
102 u64 offset, int metadata, u64 *refs, u64 *flags,
105 struct btrfs_root *extent_root;
106 struct btrfs_delayed_ref_head *head;
107 struct btrfs_delayed_ref_root *delayed_refs;
108 struct btrfs_path *path;
109 struct btrfs_extent_item *ei;
110 struct extent_buffer *leaf;
111 struct btrfs_key key;
119 * If we don't have skinny metadata, don't bother doing anything
122 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
123 offset = fs_info->nodesize;
127 path = btrfs_alloc_path();
132 path->skip_locking = 1;
133 path->search_commit_root = 1;
137 key.objectid = bytenr;
140 key.type = BTRFS_METADATA_ITEM_KEY;
142 key.type = BTRFS_EXTENT_ITEM_KEY;
144 extent_root = btrfs_extent_root(fs_info, bytenr);
145 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
149 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
150 if (path->slots[0]) {
152 btrfs_item_key_to_cpu(path->nodes[0], &key,
154 if (key.objectid == bytenr &&
155 key.type == BTRFS_EXTENT_ITEM_KEY &&
156 key.offset == fs_info->nodesize)
162 leaf = path->nodes[0];
163 item_size = btrfs_item_size(leaf, path->slots[0]);
164 if (item_size >= sizeof(*ei)) {
165 ei = btrfs_item_ptr(leaf, path->slots[0],
166 struct btrfs_extent_item);
167 num_refs = btrfs_extent_refs(leaf, ei);
168 extent_flags = btrfs_extent_flags(leaf, ei);
169 owner = btrfs_get_extent_owner_root(fs_info, leaf,
174 "unexpected extent item size, has %u expect >= %zu",
175 item_size, sizeof(*ei));
177 btrfs_abort_transaction(trans, ret);
179 btrfs_handle_fs_error(fs_info, ret, NULL);
184 BUG_ON(num_refs == 0);
194 delayed_refs = &trans->transaction->delayed_refs;
195 spin_lock(&delayed_refs->lock);
196 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
198 if (!mutex_trylock(&head->mutex)) {
199 refcount_inc(&head->refs);
200 spin_unlock(&delayed_refs->lock);
202 btrfs_release_path(path);
205 * Mutex was contended, block until it's released and try
208 mutex_lock(&head->mutex);
209 mutex_unlock(&head->mutex);
210 btrfs_put_delayed_ref_head(head);
213 spin_lock(&head->lock);
214 if (head->extent_op && head->extent_op->update_flags)
215 extent_flags |= head->extent_op->flags_to_set;
217 BUG_ON(num_refs == 0);
219 num_refs += head->ref_mod;
220 spin_unlock(&head->lock);
221 mutex_unlock(&head->mutex);
223 spin_unlock(&delayed_refs->lock);
225 WARN_ON(num_refs == 0);
229 *flags = extent_flags;
231 *owning_root = owner;
233 btrfs_free_path(path);
238 * Back reference rules. Back refs have three main goals:
240 * 1) differentiate between all holders of references to an extent so that
241 * when a reference is dropped we can make sure it was a valid reference
242 * before freeing the extent.
244 * 2) Provide enough information to quickly find the holders of an extent
245 * if we notice a given block is corrupted or bad.
247 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
248 * maintenance. This is actually the same as #2, but with a slightly
249 * different use case.
251 * There are two kinds of back refs. The implicit back refs is optimized
252 * for pointers in non-shared tree blocks. For a given pointer in a block,
253 * back refs of this kind provide information about the block's owner tree
254 * and the pointer's key. These information allow us to find the block by
255 * b-tree searching. The full back refs is for pointers in tree blocks not
256 * referenced by their owner trees. The location of tree block is recorded
257 * in the back refs. Actually the full back refs is generic, and can be
258 * used in all cases the implicit back refs is used. The major shortcoming
259 * of the full back refs is its overhead. Every time a tree block gets
260 * COWed, we have to update back refs entry for all pointers in it.
262 * For a newly allocated tree block, we use implicit back refs for
263 * pointers in it. This means most tree related operations only involve
264 * implicit back refs. For a tree block created in old transaction, the
265 * only way to drop a reference to it is COW it. So we can detect the
266 * event that tree block loses its owner tree's reference and do the
267 * back refs conversion.
269 * When a tree block is COWed through a tree, there are four cases:
271 * The reference count of the block is one and the tree is the block's
272 * owner tree. Nothing to do in this case.
274 * The reference count of the block is one and the tree is not the
275 * block's owner tree. In this case, full back refs is used for pointers
276 * in the block. Remove these full back refs, add implicit back refs for
277 * every pointers in the new block.
279 * The reference count of the block is greater than one and the tree is
280 * the block's owner tree. In this case, implicit back refs is used for
281 * pointers in the block. Add full back refs for every pointers in the
282 * block, increase lower level extents' reference counts. The original
283 * implicit back refs are entailed to the new block.
285 * The reference count of the block is greater than one and the tree is
286 * not the block's owner tree. Add implicit back refs for every pointer in
287 * the new block, increase lower level extents' reference count.
289 * Back Reference Key composing:
291 * The key objectid corresponds to the first byte in the extent,
292 * The key type is used to differentiate between types of back refs.
293 * There are different meanings of the key offset for different types
296 * File extents can be referenced by:
298 * - multiple snapshots, subvolumes, or different generations in one subvol
299 * - different files inside a single subvolume
300 * - different offsets inside a file (bookend extents in file.c)
302 * The extent ref structure for the implicit back refs has fields for:
304 * - Objectid of the subvolume root
305 * - objectid of the file holding the reference
306 * - original offset in the file
307 * - how many bookend extents
309 * The key offset for the implicit back refs is hash of the first
312 * The extent ref structure for the full back refs has field for:
314 * - number of pointers in the tree leaf
316 * The key offset for the implicit back refs is the first byte of
319 * When a file extent is allocated, The implicit back refs is used.
320 * the fields are filled in:
322 * (root_key.objectid, inode objectid, offset in file, 1)
324 * When a file extent is removed file truncation, we find the
325 * corresponding implicit back refs and check the following fields:
327 * (btrfs_header_owner(leaf), inode objectid, offset in file)
329 * Btree extents can be referenced by:
331 * - Different subvolumes
333 * Both the implicit back refs and the full back refs for tree blocks
334 * only consist of key. The key offset for the implicit back refs is
335 * objectid of block's owner tree. The key offset for the full back refs
336 * is the first byte of parent block.
338 * When implicit back refs is used, information about the lowest key and
339 * level of the tree block are required. These information are stored in
340 * tree block info structure.
344 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
345 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
346 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
348 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
349 struct btrfs_extent_inline_ref *iref,
350 enum btrfs_inline_ref_type is_data)
352 struct btrfs_fs_info *fs_info = eb->fs_info;
353 int type = btrfs_extent_inline_ref_type(eb, iref);
354 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
356 if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
357 ASSERT(btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
361 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
362 type == BTRFS_SHARED_BLOCK_REF_KEY ||
363 type == BTRFS_SHARED_DATA_REF_KEY ||
364 type == BTRFS_EXTENT_DATA_REF_KEY) {
365 if (is_data == BTRFS_REF_TYPE_BLOCK) {
366 if (type == BTRFS_TREE_BLOCK_REF_KEY)
368 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
371 * Every shared one has parent tree block,
372 * which must be aligned to sector size.
374 if (offset && IS_ALIGNED(offset, 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, fs_info->sectorsize))
391 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
397 btrfs_print_leaf(eb);
399 "eb %llu iref 0x%lx invalid extent inline ref type %d",
400 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 = crc32c(high_crc, &lenum, sizeof(lenum));
413 lenum = cpu_to_le64(owner);
414 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
415 lenum = cpu_to_le64(offset);
416 low_crc = 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 = btrfs_extent_root(trans->fs_info, bytenr);
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 = btrfs_extent_root(trans->fs_info, bytenr);
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(trans, leaf);
591 btrfs_release_path(path);
595 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
596 struct btrfs_root *root,
597 struct btrfs_path *path,
600 struct btrfs_key key;
601 struct btrfs_extent_data_ref *ref1 = NULL;
602 struct btrfs_shared_data_ref *ref2 = NULL;
603 struct extent_buffer *leaf;
607 leaf = path->nodes[0];
608 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
610 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
611 ref1 = btrfs_item_ptr(leaf, path->slots[0],
612 struct btrfs_extent_data_ref);
613 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
614 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
615 ref2 = btrfs_item_ptr(leaf, path->slots[0],
616 struct btrfs_shared_data_ref);
617 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
619 btrfs_err(trans->fs_info,
620 "unrecognized backref key (%llu %u %llu)",
621 key.objectid, key.type, key.offset);
622 btrfs_abort_transaction(trans, -EUCLEAN);
626 BUG_ON(num_refs < refs_to_drop);
627 num_refs -= refs_to_drop;
630 ret = btrfs_del_item(trans, 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(trans, 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]);
656 * If type is invalid, we should have bailed out earlier than
659 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
660 ASSERT(type != BTRFS_REF_TYPE_INVALID);
661 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
662 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
663 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
665 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
666 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
668 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
669 ref1 = btrfs_item_ptr(leaf, path->slots[0],
670 struct btrfs_extent_data_ref);
671 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
672 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
673 ref2 = btrfs_item_ptr(leaf, path->slots[0],
674 struct btrfs_shared_data_ref);
675 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
682 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
683 struct btrfs_path *path,
684 u64 bytenr, u64 parent,
687 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
688 struct btrfs_key key;
691 key.objectid = bytenr;
693 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
696 key.type = BTRFS_TREE_BLOCK_REF_KEY;
697 key.offset = root_objectid;
700 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
706 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
707 struct btrfs_path *path,
708 u64 bytenr, u64 parent,
711 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
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, root, path, &key, 0);
725 btrfs_release_path(path);
729 static inline int extent_ref_type(u64 parent, u64 owner)
732 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
734 type = BTRFS_SHARED_BLOCK_REF_KEY;
736 type = BTRFS_TREE_BLOCK_REF_KEY;
739 type = BTRFS_SHARED_DATA_REF_KEY;
741 type = BTRFS_EXTENT_DATA_REF_KEY;
746 static int find_next_key(struct btrfs_path *path, int level,
747 struct btrfs_key *key)
750 for (; level < BTRFS_MAX_LEVEL; level++) {
751 if (!path->nodes[level])
753 if (path->slots[level] + 1 >=
754 btrfs_header_nritems(path->nodes[level]))
757 btrfs_item_key_to_cpu(path->nodes[level], key,
758 path->slots[level] + 1);
760 btrfs_node_key_to_cpu(path->nodes[level], key,
761 path->slots[level] + 1);
768 * look for inline back ref. if back ref is found, *ref_ret is set
769 * to the address of inline back ref, and 0 is returned.
771 * if back ref isn't found, *ref_ret is set to the address where it
772 * should be inserted, and -ENOENT is returned.
774 * if insert is true and there are too many inline back refs, the path
775 * points to the extent item, and -EAGAIN is returned.
777 * NOTE: inline back refs are ordered in the same way that back ref
778 * items in the tree are ordered.
780 static noinline_for_stack
781 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
782 struct btrfs_path *path,
783 struct btrfs_extent_inline_ref **ref_ret,
784 u64 bytenr, u64 num_bytes,
785 u64 parent, u64 root_objectid,
786 u64 owner, u64 offset, int insert)
788 struct btrfs_fs_info *fs_info = trans->fs_info;
789 struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr);
790 struct btrfs_key key;
791 struct extent_buffer *leaf;
792 struct btrfs_extent_item *ei;
793 struct btrfs_extent_inline_ref *iref;
802 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
805 key.objectid = bytenr;
806 key.type = BTRFS_EXTENT_ITEM_KEY;
807 key.offset = num_bytes;
809 want = extent_ref_type(parent, owner);
811 extra_size = btrfs_extent_inline_ref_size(want);
812 path->search_for_extension = 1;
813 path->keep_locks = 1;
818 * Owner is our level, so we can just add one to get the level for the
819 * block we are interested in.
821 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
822 key.type = BTRFS_METADATA_ITEM_KEY;
827 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
832 * We may be a newly converted file system which still has the old fat
833 * extent entries for metadata, so try and see if we have one of those.
835 if (ret > 0 && skinny_metadata) {
836 skinny_metadata = false;
837 if (path->slots[0]) {
839 btrfs_item_key_to_cpu(path->nodes[0], &key,
841 if (key.objectid == bytenr &&
842 key.type == BTRFS_EXTENT_ITEM_KEY &&
843 key.offset == num_bytes)
847 key.objectid = bytenr;
848 key.type = BTRFS_EXTENT_ITEM_KEY;
849 key.offset = num_bytes;
850 btrfs_release_path(path);
855 if (ret && !insert) {
858 } else if (WARN_ON(ret)) {
859 btrfs_print_leaf(path->nodes[0]);
861 "extent item not found for insert, bytenr %llu num_bytes %llu parent %llu root_objectid %llu owner %llu offset %llu",
862 bytenr, num_bytes, parent, root_objectid, owner,
868 leaf = path->nodes[0];
869 item_size = btrfs_item_size(leaf, path->slots[0]);
870 if (unlikely(item_size < sizeof(*ei))) {
873 "unexpected extent item size, has %llu expect >= %zu",
874 item_size, sizeof(*ei));
875 btrfs_abort_transaction(trans, ret);
879 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
880 flags = btrfs_extent_flags(leaf, ei);
882 ptr = (unsigned long)(ei + 1);
883 end = (unsigned long)ei + item_size;
885 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
886 ptr += sizeof(struct btrfs_tree_block_info);
890 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
891 needed = BTRFS_REF_TYPE_DATA;
893 needed = BTRFS_REF_TYPE_BLOCK;
897 iref = (struct btrfs_extent_inline_ref *)ptr;
898 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
899 if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
900 ASSERT(btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
901 ptr += btrfs_extent_inline_ref_size(type);
904 if (type == BTRFS_REF_TYPE_INVALID) {
912 ptr += btrfs_extent_inline_ref_size(type);
916 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
917 struct btrfs_extent_data_ref *dref;
918 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
919 if (match_extent_data_ref(leaf, dref, root_objectid,
924 if (hash_extent_data_ref_item(leaf, dref) <
925 hash_extent_data_ref(root_objectid, owner, offset))
929 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
931 if (parent == ref_offset) {
935 if (ref_offset < parent)
938 if (root_objectid == ref_offset) {
942 if (ref_offset < root_objectid)
946 ptr += btrfs_extent_inline_ref_size(type);
949 if (unlikely(ptr > end)) {
951 btrfs_print_leaf(path->nodes[0]);
953 "overrun extent record at slot %d while looking for inline extent for root %llu owner %llu offset %llu parent %llu",
954 path->slots[0], root_objectid, owner, offset, parent);
958 if (ret == -ENOENT && insert) {
959 if (item_size + extra_size >=
960 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
965 * To add new inline back ref, we have to make sure
966 * there is no corresponding back ref item.
967 * For simplicity, we just do not add new inline back
968 * ref if there is any kind of item for this block
970 if (find_next_key(path, 0, &key) == 0 &&
971 key.objectid == bytenr &&
972 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
977 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
980 path->keep_locks = 0;
981 path->search_for_extension = 0;
982 btrfs_unlock_up_safe(path, 1);
988 * helper to add new inline back ref
990 static noinline_for_stack
991 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
992 struct btrfs_path *path,
993 struct btrfs_extent_inline_ref *iref,
994 u64 parent, u64 root_objectid,
995 u64 owner, u64 offset, int refs_to_add,
996 struct btrfs_delayed_extent_op *extent_op)
998 struct extent_buffer *leaf;
999 struct btrfs_extent_item *ei;
1002 unsigned long item_offset;
1007 leaf = path->nodes[0];
1008 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1009 item_offset = (unsigned long)iref - (unsigned long)ei;
1011 type = extent_ref_type(parent, owner);
1012 size = btrfs_extent_inline_ref_size(type);
1014 btrfs_extend_item(trans, path, size);
1016 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1017 refs = btrfs_extent_refs(leaf, ei);
1018 refs += refs_to_add;
1019 btrfs_set_extent_refs(leaf, ei, refs);
1021 __run_delayed_extent_op(extent_op, leaf, ei);
1023 ptr = (unsigned long)ei + item_offset;
1024 end = (unsigned long)ei + btrfs_item_size(leaf, path->slots[0]);
1025 if (ptr < end - size)
1026 memmove_extent_buffer(leaf, ptr + size, ptr,
1029 iref = (struct btrfs_extent_inline_ref *)ptr;
1030 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1031 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1032 struct btrfs_extent_data_ref *dref;
1033 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1034 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1035 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1036 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1037 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1038 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1039 struct btrfs_shared_data_ref *sref;
1040 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1041 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1042 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1043 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1044 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1046 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1048 btrfs_mark_buffer_dirty(trans, leaf);
1051 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1052 struct btrfs_path *path,
1053 struct btrfs_extent_inline_ref **ref_ret,
1054 u64 bytenr, u64 num_bytes, u64 parent,
1055 u64 root_objectid, u64 owner, u64 offset)
1059 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1060 num_bytes, parent, root_objectid,
1065 btrfs_release_path(path);
1068 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1069 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1072 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1073 root_objectid, owner, offset);
1079 * helper to update/remove inline back ref
1081 static noinline_for_stack int update_inline_extent_backref(
1082 struct btrfs_trans_handle *trans,
1083 struct btrfs_path *path,
1084 struct btrfs_extent_inline_ref *iref,
1086 struct btrfs_delayed_extent_op *extent_op)
1088 struct extent_buffer *leaf = path->nodes[0];
1089 struct btrfs_fs_info *fs_info = leaf->fs_info;
1090 struct btrfs_extent_item *ei;
1091 struct btrfs_extent_data_ref *dref = NULL;
1092 struct btrfs_shared_data_ref *sref = NULL;
1100 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1101 refs = btrfs_extent_refs(leaf, ei);
1102 if (unlikely(refs_to_mod < 0 && refs + refs_to_mod <= 0)) {
1103 struct btrfs_key key;
1106 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1107 if (key.type == BTRFS_METADATA_ITEM_KEY)
1108 extent_size = fs_info->nodesize;
1110 extent_size = key.offset;
1111 btrfs_print_leaf(leaf);
1113 "invalid refs_to_mod for extent %llu num_bytes %u, has %d expect >= -%llu",
1114 key.objectid, extent_size, refs_to_mod, refs);
1117 refs += refs_to_mod;
1118 btrfs_set_extent_refs(leaf, ei, refs);
1120 __run_delayed_extent_op(extent_op, leaf, ei);
1122 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1124 * Function btrfs_get_extent_inline_ref_type() has already printed
1127 if (unlikely(type == BTRFS_REF_TYPE_INVALID))
1130 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1131 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1132 refs = btrfs_extent_data_ref_count(leaf, dref);
1133 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1134 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1135 refs = btrfs_shared_data_ref_count(leaf, sref);
1139 * For tree blocks we can only drop one ref for it, and tree
1140 * blocks should not have refs > 1.
1142 * Furthermore if we're inserting a new inline backref, we
1143 * won't reach this path either. That would be
1144 * setup_inline_extent_backref().
1146 if (unlikely(refs_to_mod != -1)) {
1147 struct btrfs_key key;
1149 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1151 btrfs_print_leaf(leaf);
1153 "invalid refs_to_mod for tree block %llu, has %d expect -1",
1154 key.objectid, refs_to_mod);
1159 if (unlikely(refs_to_mod < 0 && refs < -refs_to_mod)) {
1160 struct btrfs_key key;
1163 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1164 if (key.type == BTRFS_METADATA_ITEM_KEY)
1165 extent_size = fs_info->nodesize;
1167 extent_size = key.offset;
1168 btrfs_print_leaf(leaf);
1170 "invalid refs_to_mod for backref entry, iref %lu extent %llu num_bytes %u, has %d expect >= -%llu",
1171 (unsigned long)iref, key.objectid, extent_size,
1175 refs += refs_to_mod;
1178 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1179 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1181 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1183 size = btrfs_extent_inline_ref_size(type);
1184 item_size = btrfs_item_size(leaf, path->slots[0]);
1185 ptr = (unsigned long)iref;
1186 end = (unsigned long)ei + item_size;
1187 if (ptr + size < end)
1188 memmove_extent_buffer(leaf, ptr, ptr + size,
1191 btrfs_truncate_item(trans, path, item_size, 1);
1193 btrfs_mark_buffer_dirty(trans, leaf);
1197 static noinline_for_stack
1198 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1199 struct btrfs_path *path,
1200 u64 bytenr, u64 num_bytes, u64 parent,
1201 u64 root_objectid, u64 owner,
1202 u64 offset, int refs_to_add,
1203 struct btrfs_delayed_extent_op *extent_op)
1205 struct btrfs_extent_inline_ref *iref;
1208 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1209 num_bytes, parent, root_objectid,
1213 * We're adding refs to a tree block we already own, this
1214 * should not happen at all.
1216 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1217 btrfs_print_leaf(path->nodes[0]);
1218 btrfs_crit(trans->fs_info,
1219 "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu slot %u",
1220 bytenr, num_bytes, root_objectid, path->slots[0]);
1223 ret = update_inline_extent_backref(trans, path, iref,
1224 refs_to_add, extent_op);
1225 } else if (ret == -ENOENT) {
1226 setup_inline_extent_backref(trans, path, iref, parent,
1227 root_objectid, owner, offset,
1228 refs_to_add, extent_op);
1234 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1235 struct btrfs_root *root,
1236 struct btrfs_path *path,
1237 struct btrfs_extent_inline_ref *iref,
1238 int refs_to_drop, int is_data)
1242 BUG_ON(!is_data && refs_to_drop != 1);
1244 ret = update_inline_extent_backref(trans, path, iref,
1245 -refs_to_drop, NULL);
1247 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1249 ret = btrfs_del_item(trans, root, path);
1253 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1254 u64 *discarded_bytes)
1257 u64 bytes_left, end;
1258 u64 aligned_start = ALIGN(start, 1 << SECTOR_SHIFT);
1260 /* Adjust the range to be aligned to 512B sectors if necessary. */
1261 if (start != aligned_start) {
1262 len -= aligned_start - start;
1263 len = round_down(len, 1 << SECTOR_SHIFT);
1264 start = aligned_start;
1267 *discarded_bytes = 0;
1275 /* Skip any superblocks on this device. */
1276 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1277 u64 sb_start = btrfs_sb_offset(j);
1278 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1279 u64 size = sb_start - start;
1281 if (!in_range(sb_start, start, bytes_left) &&
1282 !in_range(sb_end, start, bytes_left) &&
1283 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1287 * Superblock spans beginning of range. Adjust start and
1290 if (sb_start <= start) {
1291 start += sb_end - start;
1296 bytes_left = end - start;
1301 ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1302 size >> SECTOR_SHIFT,
1305 *discarded_bytes += size;
1306 else if (ret != -EOPNOTSUPP)
1315 bytes_left = end - start;
1319 ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1320 bytes_left >> SECTOR_SHIFT,
1323 *discarded_bytes += bytes_left;
1328 static int do_discard_extent(struct btrfs_discard_stripe *stripe, u64 *bytes)
1330 struct btrfs_device *dev = stripe->dev;
1331 struct btrfs_fs_info *fs_info = dev->fs_info;
1332 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1333 u64 phys = stripe->physical;
1334 u64 len = stripe->length;
1338 /* Zone reset on a zoned filesystem */
1339 if (btrfs_can_zone_reset(dev, phys, len)) {
1342 ret = btrfs_reset_device_zone(dev, phys, len, &discarded);
1346 if (!btrfs_dev_replace_is_ongoing(dev_replace) ||
1347 dev != dev_replace->srcdev)
1350 src_disc = discarded;
1352 /* Send to replace target as well */
1353 ret = btrfs_reset_device_zone(dev_replace->tgtdev, phys, len,
1355 discarded += src_disc;
1356 } else if (bdev_max_discard_sectors(stripe->dev->bdev)) {
1357 ret = btrfs_issue_discard(dev->bdev, phys, len, &discarded);
1368 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1369 u64 num_bytes, u64 *actual_bytes)
1372 u64 discarded_bytes = 0;
1373 u64 end = bytenr + num_bytes;
1377 * Avoid races with device replace and make sure the devices in the
1378 * stripes don't go away while we are discarding.
1380 btrfs_bio_counter_inc_blocked(fs_info);
1382 struct btrfs_discard_stripe *stripes;
1383 unsigned int num_stripes;
1386 num_bytes = end - cur;
1387 stripes = btrfs_map_discard(fs_info, cur, &num_bytes, &num_stripes);
1388 if (IS_ERR(stripes)) {
1389 ret = PTR_ERR(stripes);
1390 if (ret == -EOPNOTSUPP)
1395 for (i = 0; i < num_stripes; i++) {
1396 struct btrfs_discard_stripe *stripe = stripes + i;
1399 if (!stripe->dev->bdev) {
1400 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1404 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
1405 &stripe->dev->dev_state))
1408 ret = do_discard_extent(stripe, &bytes);
1411 * Keep going if discard is not supported by the
1414 if (ret != -EOPNOTSUPP)
1418 discarded_bytes += bytes;
1426 btrfs_bio_counter_dec(fs_info);
1428 *actual_bytes = discarded_bytes;
1432 /* Can return -ENOMEM */
1433 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1434 struct btrfs_ref *generic_ref)
1436 struct btrfs_fs_info *fs_info = trans->fs_info;
1439 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1440 generic_ref->action);
1441 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1442 generic_ref->tree_ref.ref_root == BTRFS_TREE_LOG_OBJECTID);
1444 if (generic_ref->type == BTRFS_REF_METADATA)
1445 ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL);
1447 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0);
1449 btrfs_ref_tree_mod(fs_info, generic_ref);
1455 * Insert backreference for a given extent.
1457 * The counterpart is in __btrfs_free_extent(), with examples and more details
1460 * @trans: Handle of transaction
1462 * @node: The delayed ref node used to get the bytenr/length for
1463 * extent whose references are incremented.
1465 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1466 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1467 * bytenr of the parent block. Since new extents are always
1468 * created with indirect references, this will only be the case
1469 * when relocating a shared extent. In that case, root_objectid
1470 * will be BTRFS_TREE_RELOC_OBJECTID. Otherwise, parent must
1473 * @root_objectid: The id of the root where this modification has originated,
1474 * this can be either one of the well-known metadata trees or
1475 * the subvolume id which references this extent.
1477 * @owner: For data extents it is the inode number of the owning file.
1478 * For metadata extents this parameter holds the level in the
1479 * tree of the extent.
1481 * @offset: For metadata extents the offset is ignored and is currently
1482 * always passed as 0. For data extents it is the fileoffset
1483 * this extent belongs to.
1485 * @extent_op Pointer to a structure, holding information necessary when
1486 * updating a tree block's flags
1489 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1490 struct btrfs_delayed_ref_node *node,
1491 u64 parent, u64 root_objectid,
1492 u64 owner, u64 offset,
1493 struct btrfs_delayed_extent_op *extent_op)
1495 struct btrfs_path *path;
1496 struct extent_buffer *leaf;
1497 struct btrfs_extent_item *item;
1498 struct btrfs_key key;
1499 u64 bytenr = node->bytenr;
1500 u64 num_bytes = node->num_bytes;
1502 int refs_to_add = node->ref_mod;
1505 path = btrfs_alloc_path();
1509 /* this will setup the path even if it fails to insert the back ref */
1510 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1511 parent, root_objectid, owner,
1512 offset, refs_to_add, extent_op);
1513 if ((ret < 0 && ret != -EAGAIN) || !ret)
1517 * Ok we had -EAGAIN which means we didn't have space to insert and
1518 * inline extent ref, so just update the reference count and add a
1521 leaf = path->nodes[0];
1522 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1523 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1524 refs = btrfs_extent_refs(leaf, item);
1525 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1527 __run_delayed_extent_op(extent_op, leaf, item);
1529 btrfs_mark_buffer_dirty(trans, leaf);
1530 btrfs_release_path(path);
1532 /* now insert the actual backref */
1533 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1534 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1537 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1538 root_objectid, owner, offset,
1542 btrfs_abort_transaction(trans, ret);
1544 btrfs_free_path(path);
1548 static void free_head_ref_squota_rsv(struct btrfs_fs_info *fs_info,
1549 struct btrfs_delayed_ref_head *href)
1551 u64 root = href->owning_root;
1554 * Don't check must_insert_reserved, as this is called from contexts
1555 * where it has already been unset.
1557 if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_SIMPLE ||
1558 !href->is_data || !is_fstree(root))
1561 btrfs_qgroup_free_refroot(fs_info, root, href->reserved_bytes,
1562 BTRFS_QGROUP_RSV_DATA);
1565 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1566 struct btrfs_delayed_ref_head *href,
1567 struct btrfs_delayed_ref_node *node,
1568 struct btrfs_delayed_extent_op *extent_op,
1569 bool insert_reserved)
1572 struct btrfs_delayed_data_ref *ref;
1576 ref = btrfs_delayed_node_to_data_ref(node);
1577 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
1579 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1580 parent = ref->parent;
1582 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1583 struct btrfs_key key;
1584 struct btrfs_squota_delta delta = {
1585 .root = href->owning_root,
1586 .num_bytes = node->num_bytes,
1589 .generation = trans->transid,
1593 flags |= extent_op->flags_to_set;
1595 key.objectid = node->bytenr;
1596 key.type = BTRFS_EXTENT_ITEM_KEY;
1597 key.offset = node->num_bytes;
1599 ret = alloc_reserved_file_extent(trans, parent, ref->root,
1600 flags, ref->objectid,
1602 node->ref_mod, href->owning_root);
1603 free_head_ref_squota_rsv(trans->fs_info, href);
1605 ret = btrfs_record_squota_delta(trans->fs_info, &delta);
1606 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1607 ret = __btrfs_inc_extent_ref(trans, node, parent, ref->root,
1608 ref->objectid, ref->offset,
1610 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1611 ret = __btrfs_free_extent(trans, href, node, parent,
1612 ref->root, ref->objectid,
1613 ref->offset, extent_op);
1620 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1621 struct extent_buffer *leaf,
1622 struct btrfs_extent_item *ei)
1624 u64 flags = btrfs_extent_flags(leaf, ei);
1625 if (extent_op->update_flags) {
1626 flags |= extent_op->flags_to_set;
1627 btrfs_set_extent_flags(leaf, ei, flags);
1630 if (extent_op->update_key) {
1631 struct btrfs_tree_block_info *bi;
1632 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1633 bi = (struct btrfs_tree_block_info *)(ei + 1);
1634 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1638 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1639 struct btrfs_delayed_ref_head *head,
1640 struct btrfs_delayed_extent_op *extent_op)
1642 struct btrfs_fs_info *fs_info = trans->fs_info;
1643 struct btrfs_root *root;
1644 struct btrfs_key key;
1645 struct btrfs_path *path;
1646 struct btrfs_extent_item *ei;
1647 struct extent_buffer *leaf;
1652 if (TRANS_ABORTED(trans))
1655 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1658 path = btrfs_alloc_path();
1662 key.objectid = head->bytenr;
1665 key.type = BTRFS_METADATA_ITEM_KEY;
1666 key.offset = extent_op->level;
1668 key.type = BTRFS_EXTENT_ITEM_KEY;
1669 key.offset = head->num_bytes;
1672 root = btrfs_extent_root(fs_info, key.objectid);
1674 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1677 } else if (ret > 0) {
1679 if (path->slots[0] > 0) {
1681 btrfs_item_key_to_cpu(path->nodes[0], &key,
1683 if (key.objectid == head->bytenr &&
1684 key.type == BTRFS_EXTENT_ITEM_KEY &&
1685 key.offset == head->num_bytes)
1689 btrfs_release_path(path);
1692 key.objectid = head->bytenr;
1693 key.offset = head->num_bytes;
1694 key.type = BTRFS_EXTENT_ITEM_KEY;
1700 "missing extent item for extent %llu num_bytes %llu level %d",
1701 head->bytenr, head->num_bytes, extent_op->level);
1706 leaf = path->nodes[0];
1707 item_size = btrfs_item_size(leaf, path->slots[0]);
1709 if (unlikely(item_size < sizeof(*ei))) {
1712 "unexpected extent item size, has %u expect >= %zu",
1713 item_size, sizeof(*ei));
1714 btrfs_abort_transaction(trans, ret);
1718 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1719 __run_delayed_extent_op(extent_op, leaf, ei);
1721 btrfs_mark_buffer_dirty(trans, leaf);
1723 btrfs_free_path(path);
1727 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1728 struct btrfs_delayed_ref_head *href,
1729 struct btrfs_delayed_ref_node *node,
1730 struct btrfs_delayed_extent_op *extent_op,
1731 bool insert_reserved)
1734 struct btrfs_fs_info *fs_info = trans->fs_info;
1735 struct btrfs_delayed_tree_ref *ref;
1739 ref = btrfs_delayed_node_to_tree_ref(node);
1740 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1742 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1743 parent = ref->parent;
1744 ref_root = ref->root;
1746 if (unlikely(node->ref_mod != 1)) {
1747 btrfs_err(trans->fs_info,
1748 "btree block %llu has %d references rather than 1: action %d ref_root %llu parent %llu",
1749 node->bytenr, node->ref_mod, node->action, ref_root,
1753 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1754 struct btrfs_squota_delta delta = {
1755 .root = href->owning_root,
1756 .num_bytes = fs_info->nodesize,
1759 .generation = trans->transid,
1762 BUG_ON(!extent_op || !extent_op->update_flags);
1763 ret = alloc_reserved_tree_block(trans, node, extent_op);
1765 btrfs_record_squota_delta(fs_info, &delta);
1766 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1767 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1768 ref->level, 0, extent_op);
1769 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1770 ret = __btrfs_free_extent(trans, href, node, parent, ref_root,
1771 ref->level, 0, extent_op);
1778 /* helper function to actually process a single delayed ref entry */
1779 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1780 struct btrfs_delayed_ref_head *href,
1781 struct btrfs_delayed_ref_node *node,
1782 struct btrfs_delayed_extent_op *extent_op,
1783 bool insert_reserved)
1787 if (TRANS_ABORTED(trans)) {
1788 if (insert_reserved) {
1789 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1790 free_head_ref_squota_rsv(trans->fs_info, href);
1795 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1796 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1797 ret = run_delayed_tree_ref(trans, href, node, extent_op,
1799 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1800 node->type == BTRFS_SHARED_DATA_REF_KEY)
1801 ret = run_delayed_data_ref(trans, href, node, extent_op,
1803 else if (node->type == BTRFS_EXTENT_OWNER_REF_KEY)
1807 if (ret && insert_reserved)
1808 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1810 btrfs_err(trans->fs_info,
1811 "failed to run delayed ref for logical %llu num_bytes %llu type %u action %u ref_mod %d: %d",
1812 node->bytenr, node->num_bytes, node->type,
1813 node->action, node->ref_mod, ret);
1817 static inline struct btrfs_delayed_ref_node *
1818 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1820 struct btrfs_delayed_ref_node *ref;
1822 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1826 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1827 * This is to prevent a ref count from going down to zero, which deletes
1828 * the extent item from the extent tree, when there still are references
1829 * to add, which would fail because they would not find the extent item.
1831 if (!list_empty(&head->ref_add_list))
1832 return list_first_entry(&head->ref_add_list,
1833 struct btrfs_delayed_ref_node, add_list);
1835 ref = rb_entry(rb_first_cached(&head->ref_tree),
1836 struct btrfs_delayed_ref_node, ref_node);
1837 ASSERT(list_empty(&ref->add_list));
1841 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1842 struct btrfs_delayed_ref_head *head)
1844 spin_lock(&delayed_refs->lock);
1845 head->processing = false;
1846 delayed_refs->num_heads_ready++;
1847 spin_unlock(&delayed_refs->lock);
1848 btrfs_delayed_ref_unlock(head);
1851 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1852 struct btrfs_delayed_ref_head *head)
1854 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1859 if (head->must_insert_reserved) {
1860 head->extent_op = NULL;
1861 btrfs_free_delayed_extent_op(extent_op);
1867 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1868 struct btrfs_delayed_ref_head *head)
1870 struct btrfs_delayed_extent_op *extent_op;
1873 extent_op = cleanup_extent_op(head);
1876 head->extent_op = NULL;
1877 spin_unlock(&head->lock);
1878 ret = run_delayed_extent_op(trans, head, extent_op);
1879 btrfs_free_delayed_extent_op(extent_op);
1880 return ret ? ret : 1;
1883 u64 btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1884 struct btrfs_delayed_ref_root *delayed_refs,
1885 struct btrfs_delayed_ref_head *head)
1890 * We had csum deletions accounted for in our delayed refs rsv, we need
1891 * to drop the csum leaves for this update from our delayed_refs_rsv.
1893 if (head->total_ref_mod < 0 && head->is_data) {
1896 spin_lock(&delayed_refs->lock);
1897 delayed_refs->pending_csums -= head->num_bytes;
1898 spin_unlock(&delayed_refs->lock);
1899 nr_csums = btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes);
1901 btrfs_delayed_refs_rsv_release(fs_info, 0, nr_csums);
1903 ret = btrfs_calc_delayed_ref_csum_bytes(fs_info, nr_csums);
1905 /* must_insert_reserved can be set only if we didn't run the head ref. */
1906 if (head->must_insert_reserved)
1907 free_head_ref_squota_rsv(fs_info, head);
1912 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1913 struct btrfs_delayed_ref_head *head,
1914 u64 *bytes_released)
1917 struct btrfs_fs_info *fs_info = trans->fs_info;
1918 struct btrfs_delayed_ref_root *delayed_refs;
1921 delayed_refs = &trans->transaction->delayed_refs;
1923 ret = run_and_cleanup_extent_op(trans, head);
1925 unselect_delayed_ref_head(delayed_refs, head);
1926 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1933 * Need to drop our head ref lock and re-acquire the delayed ref lock
1934 * and then re-check to make sure nobody got added.
1936 spin_unlock(&head->lock);
1937 spin_lock(&delayed_refs->lock);
1938 spin_lock(&head->lock);
1939 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1940 spin_unlock(&head->lock);
1941 spin_unlock(&delayed_refs->lock);
1944 btrfs_delete_ref_head(delayed_refs, head);
1945 spin_unlock(&head->lock);
1946 spin_unlock(&delayed_refs->lock);
1948 if (head->must_insert_reserved) {
1949 btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
1950 if (head->is_data) {
1951 struct btrfs_root *csum_root;
1953 csum_root = btrfs_csum_root(fs_info, head->bytenr);
1954 ret = btrfs_del_csums(trans, csum_root, head->bytenr,
1959 *bytes_released += btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1961 trace_run_delayed_ref_head(fs_info, head, 0);
1962 btrfs_delayed_ref_unlock(head);
1963 btrfs_put_delayed_ref_head(head);
1967 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1968 struct btrfs_trans_handle *trans)
1970 struct btrfs_delayed_ref_root *delayed_refs =
1971 &trans->transaction->delayed_refs;
1972 struct btrfs_delayed_ref_head *head = NULL;
1975 spin_lock(&delayed_refs->lock);
1976 head = btrfs_select_ref_head(delayed_refs);
1978 spin_unlock(&delayed_refs->lock);
1983 * Grab the lock that says we are going to process all the refs for
1986 ret = btrfs_delayed_ref_lock(delayed_refs, head);
1987 spin_unlock(&delayed_refs->lock);
1990 * We may have dropped the spin lock to get the head mutex lock, and
1991 * that might have given someone else time to free the head. If that's
1992 * true, it has been removed from our list and we can move on.
1995 head = ERR_PTR(-EAGAIN);
2000 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
2001 struct btrfs_delayed_ref_head *locked_ref,
2002 u64 *bytes_released)
2004 struct btrfs_fs_info *fs_info = trans->fs_info;
2005 struct btrfs_delayed_ref_root *delayed_refs;
2006 struct btrfs_delayed_extent_op *extent_op;
2007 struct btrfs_delayed_ref_node *ref;
2008 bool must_insert_reserved;
2011 delayed_refs = &trans->transaction->delayed_refs;
2013 lockdep_assert_held(&locked_ref->mutex);
2014 lockdep_assert_held(&locked_ref->lock);
2016 while ((ref = select_delayed_ref(locked_ref))) {
2018 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2019 spin_unlock(&locked_ref->lock);
2020 unselect_delayed_ref_head(delayed_refs, locked_ref);
2024 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
2025 RB_CLEAR_NODE(&ref->ref_node);
2026 if (!list_empty(&ref->add_list))
2027 list_del(&ref->add_list);
2029 * When we play the delayed ref, also correct the ref_mod on
2032 switch (ref->action) {
2033 case BTRFS_ADD_DELAYED_REF:
2034 case BTRFS_ADD_DELAYED_EXTENT:
2035 locked_ref->ref_mod -= ref->ref_mod;
2037 case BTRFS_DROP_DELAYED_REF:
2038 locked_ref->ref_mod += ref->ref_mod;
2043 atomic_dec(&delayed_refs->num_entries);
2046 * Record the must_insert_reserved flag before we drop the
2049 must_insert_reserved = locked_ref->must_insert_reserved;
2051 * Unsetting this on the head ref relinquishes ownership of
2052 * the rsv_bytes, so it is critical that every possible code
2053 * path from here forward frees all reserves including qgroup
2056 locked_ref->must_insert_reserved = false;
2058 extent_op = locked_ref->extent_op;
2059 locked_ref->extent_op = NULL;
2060 spin_unlock(&locked_ref->lock);
2062 ret = run_one_delayed_ref(trans, locked_ref, ref, extent_op,
2063 must_insert_reserved);
2064 btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
2065 *bytes_released += btrfs_calc_delayed_ref_bytes(fs_info, 1);
2067 btrfs_free_delayed_extent_op(extent_op);
2069 unselect_delayed_ref_head(delayed_refs, locked_ref);
2070 btrfs_put_delayed_ref(ref);
2074 btrfs_put_delayed_ref(ref);
2077 spin_lock(&locked_ref->lock);
2078 btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
2085 * Returns 0 on success or if called with an already aborted transaction.
2086 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2088 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2091 struct btrfs_fs_info *fs_info = trans->fs_info;
2092 struct btrfs_delayed_ref_root *delayed_refs;
2093 struct btrfs_delayed_ref_head *locked_ref = NULL;
2095 unsigned long count = 0;
2096 unsigned long max_count = 0;
2097 u64 bytes_processed = 0;
2099 delayed_refs = &trans->transaction->delayed_refs;
2100 if (min_bytes == 0) {
2101 max_count = delayed_refs->num_heads_ready;
2102 min_bytes = U64_MAX;
2107 locked_ref = btrfs_obtain_ref_head(trans);
2108 if (IS_ERR_OR_NULL(locked_ref)) {
2109 if (PTR_ERR(locked_ref) == -EAGAIN) {
2118 * We need to try and merge add/drops of the same ref since we
2119 * can run into issues with relocate dropping the implicit ref
2120 * and then it being added back again before the drop can
2121 * finish. If we merged anything we need to re-loop so we can
2123 * Or we can get node references of the same type that weren't
2124 * merged when created due to bumps in the tree mod seq, and
2125 * we need to merge them to prevent adding an inline extent
2126 * backref before dropping it (triggering a BUG_ON at
2127 * insert_inline_extent_backref()).
2129 spin_lock(&locked_ref->lock);
2130 btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
2132 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, &bytes_processed);
2133 if (ret < 0 && ret != -EAGAIN) {
2135 * Error, btrfs_run_delayed_refs_for_head already
2136 * unlocked everything so just bail out
2141 * Success, perform the usual cleanup of a processed
2144 ret = cleanup_ref_head(trans, locked_ref, &bytes_processed);
2146 /* We dropped our lock, we need to loop. */
2155 * Either success case or btrfs_run_delayed_refs_for_head
2156 * returned -EAGAIN, meaning we need to select another head
2161 } while ((min_bytes != U64_MAX && bytes_processed < min_bytes) ||
2162 (max_count > 0 && count < max_count) ||
2168 #ifdef SCRAMBLE_DELAYED_REFS
2170 * Normally delayed refs get processed in ascending bytenr order. This
2171 * correlates in most cases to the order added. To expose dependencies on this
2172 * order, we start to process the tree in the middle instead of the beginning
2174 static u64 find_middle(struct rb_root *root)
2176 struct rb_node *n = root->rb_node;
2177 struct btrfs_delayed_ref_node *entry;
2180 u64 first = 0, last = 0;
2184 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2185 first = entry->bytenr;
2189 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2190 last = entry->bytenr;
2195 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2196 WARN_ON(!entry->in_tree);
2198 middle = entry->bytenr;
2212 * Start processing the delayed reference count updates and extent insertions
2213 * we have queued up so far.
2215 * @trans: Transaction handle.
2216 * @min_bytes: How many bytes of delayed references to process. After this
2217 * many bytes we stop processing delayed references if there are
2218 * any more. If 0 it means to run all existing delayed references,
2219 * but not new ones added after running all existing ones.
2220 * Use (u64)-1 (U64_MAX) to run all existing delayed references
2221 * plus any new ones that are added.
2223 * Returns 0 on success or if called with an aborted transaction
2224 * Returns <0 on error and aborts the transaction
2226 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, u64 min_bytes)
2228 struct btrfs_fs_info *fs_info = trans->fs_info;
2229 struct btrfs_delayed_ref_root *delayed_refs;
2232 /* We'll clean this up in btrfs_cleanup_transaction */
2233 if (TRANS_ABORTED(trans))
2236 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2239 delayed_refs = &trans->transaction->delayed_refs;
2241 #ifdef SCRAMBLE_DELAYED_REFS
2242 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2244 ret = __btrfs_run_delayed_refs(trans, min_bytes);
2246 btrfs_abort_transaction(trans, ret);
2250 if (min_bytes == U64_MAX) {
2251 btrfs_create_pending_block_groups(trans);
2253 spin_lock(&delayed_refs->lock);
2254 if (RB_EMPTY_ROOT(&delayed_refs->href_root.rb_root)) {
2255 spin_unlock(&delayed_refs->lock);
2258 spin_unlock(&delayed_refs->lock);
2267 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2268 struct extent_buffer *eb, u64 flags)
2270 struct btrfs_delayed_extent_op *extent_op;
2271 int level = btrfs_header_level(eb);
2274 extent_op = btrfs_alloc_delayed_extent_op();
2278 extent_op->flags_to_set = flags;
2279 extent_op->update_flags = true;
2280 extent_op->update_key = false;
2281 extent_op->level = level;
2283 ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op);
2285 btrfs_free_delayed_extent_op(extent_op);
2289 static noinline int check_delayed_ref(struct btrfs_root *root,
2290 struct btrfs_path *path,
2291 u64 objectid, u64 offset, u64 bytenr)
2293 struct btrfs_delayed_ref_head *head;
2294 struct btrfs_delayed_ref_node *ref;
2295 struct btrfs_delayed_data_ref *data_ref;
2296 struct btrfs_delayed_ref_root *delayed_refs;
2297 struct btrfs_transaction *cur_trans;
2298 struct rb_node *node;
2301 spin_lock(&root->fs_info->trans_lock);
2302 cur_trans = root->fs_info->running_transaction;
2304 refcount_inc(&cur_trans->use_count);
2305 spin_unlock(&root->fs_info->trans_lock);
2309 delayed_refs = &cur_trans->delayed_refs;
2310 spin_lock(&delayed_refs->lock);
2311 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2313 spin_unlock(&delayed_refs->lock);
2314 btrfs_put_transaction(cur_trans);
2318 if (!mutex_trylock(&head->mutex)) {
2320 spin_unlock(&delayed_refs->lock);
2321 btrfs_put_transaction(cur_trans);
2325 refcount_inc(&head->refs);
2326 spin_unlock(&delayed_refs->lock);
2328 btrfs_release_path(path);
2331 * Mutex was contended, block until it's released and let
2334 mutex_lock(&head->mutex);
2335 mutex_unlock(&head->mutex);
2336 btrfs_put_delayed_ref_head(head);
2337 btrfs_put_transaction(cur_trans);
2340 spin_unlock(&delayed_refs->lock);
2342 spin_lock(&head->lock);
2344 * XXX: We should replace this with a proper search function in the
2347 for (node = rb_first_cached(&head->ref_tree); node;
2348 node = rb_next(node)) {
2349 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2350 /* If it's a shared ref we know a cross reference exists */
2351 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2356 data_ref = btrfs_delayed_node_to_data_ref(ref);
2359 * If our ref doesn't match the one we're currently looking at
2360 * then we have a cross reference.
2362 if (data_ref->root != root->root_key.objectid ||
2363 data_ref->objectid != objectid ||
2364 data_ref->offset != offset) {
2369 spin_unlock(&head->lock);
2370 mutex_unlock(&head->mutex);
2371 btrfs_put_transaction(cur_trans);
2375 static noinline int check_committed_ref(struct btrfs_root *root,
2376 struct btrfs_path *path,
2377 u64 objectid, u64 offset, u64 bytenr,
2380 struct btrfs_fs_info *fs_info = root->fs_info;
2381 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2382 struct extent_buffer *leaf;
2383 struct btrfs_extent_data_ref *ref;
2384 struct btrfs_extent_inline_ref *iref;
2385 struct btrfs_extent_item *ei;
2386 struct btrfs_key key;
2392 key.objectid = bytenr;
2393 key.offset = (u64)-1;
2394 key.type = BTRFS_EXTENT_ITEM_KEY;
2396 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2401 * Key with offset -1 found, there would have to exist an extent
2402 * item with such offset, but this is out of the valid range.
2409 if (path->slots[0] == 0)
2413 leaf = path->nodes[0];
2414 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2416 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2420 item_size = btrfs_item_size(leaf, path->slots[0]);
2421 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2422 expected_size = sizeof(*ei) + btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY);
2424 /* No inline refs; we need to bail before checking for owner ref. */
2425 if (item_size == sizeof(*ei))
2428 /* Check for an owner ref; skip over it to the real inline refs. */
2429 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2430 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2431 if (btrfs_fs_incompat(fs_info, SIMPLE_QUOTA) && type == BTRFS_EXTENT_OWNER_REF_KEY) {
2432 expected_size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY);
2433 iref = (struct btrfs_extent_inline_ref *)(iref + 1);
2436 /* If extent item has more than 1 inline ref then it's shared */
2437 if (item_size != expected_size)
2441 * If extent created before last snapshot => it's shared unless the
2442 * snapshot has been deleted. Use the heuristic if strict is false.
2445 (btrfs_extent_generation(leaf, ei) <=
2446 btrfs_root_last_snapshot(&root->root_item)))
2449 /* If this extent has SHARED_DATA_REF then it's shared */
2450 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2451 if (type != BTRFS_EXTENT_DATA_REF_KEY)
2454 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2455 if (btrfs_extent_refs(leaf, ei) !=
2456 btrfs_extent_data_ref_count(leaf, ref) ||
2457 btrfs_extent_data_ref_root(leaf, ref) !=
2458 root->root_key.objectid ||
2459 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2460 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2468 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2469 u64 bytenr, bool strict, struct btrfs_path *path)
2474 ret = check_committed_ref(root, path, objectid,
2475 offset, bytenr, strict);
2476 if (ret && ret != -ENOENT)
2479 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2480 } while (ret == -EAGAIN);
2483 btrfs_release_path(path);
2484 if (btrfs_is_data_reloc_root(root))
2489 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2490 struct btrfs_root *root,
2491 struct extent_buffer *buf,
2492 int full_backref, int inc)
2494 struct btrfs_fs_info *fs_info = root->fs_info;
2500 struct btrfs_key key;
2501 struct btrfs_file_extent_item *fi;
2502 struct btrfs_ref generic_ref = { 0 };
2503 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2509 if (btrfs_is_testing(fs_info))
2512 ref_root = btrfs_header_owner(buf);
2513 nritems = btrfs_header_nritems(buf);
2514 level = btrfs_header_level(buf);
2516 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
2520 parent = buf->start;
2524 action = BTRFS_ADD_DELAYED_REF;
2526 action = BTRFS_DROP_DELAYED_REF;
2528 for (i = 0; i < nritems; i++) {
2530 btrfs_item_key_to_cpu(buf, &key, i);
2531 if (key.type != BTRFS_EXTENT_DATA_KEY)
2533 fi = btrfs_item_ptr(buf, i,
2534 struct btrfs_file_extent_item);
2535 if (btrfs_file_extent_type(buf, fi) ==
2536 BTRFS_FILE_EXTENT_INLINE)
2538 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2542 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2543 key.offset -= btrfs_file_extent_offset(buf, fi);
2544 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2545 num_bytes, parent, ref_root);
2546 btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2547 key.offset, root->root_key.objectid,
2550 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2552 ret = btrfs_free_extent(trans, &generic_ref);
2556 bytenr = btrfs_node_blockptr(buf, i);
2557 num_bytes = fs_info->nodesize;
2558 /* We don't know the owning_root, use 0. */
2559 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2560 num_bytes, parent, 0);
2561 btrfs_init_tree_ref(&generic_ref, level - 1, ref_root,
2562 root->root_key.objectid, for_reloc);
2564 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2566 ret = btrfs_free_extent(trans, &generic_ref);
2576 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2577 struct extent_buffer *buf, int full_backref)
2579 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2582 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2583 struct extent_buffer *buf, int full_backref)
2585 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2588 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2590 struct btrfs_fs_info *fs_info = root->fs_info;
2595 flags = BTRFS_BLOCK_GROUP_DATA;
2596 else if (root == fs_info->chunk_root)
2597 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2599 flags = BTRFS_BLOCK_GROUP_METADATA;
2601 ret = btrfs_get_alloc_profile(fs_info, flags);
2605 static u64 first_logical_byte(struct btrfs_fs_info *fs_info)
2607 struct rb_node *leftmost;
2610 read_lock(&fs_info->block_group_cache_lock);
2611 /* Get the block group with the lowest logical start address. */
2612 leftmost = rb_first_cached(&fs_info->block_group_cache_tree);
2614 struct btrfs_block_group *bg;
2616 bg = rb_entry(leftmost, struct btrfs_block_group, cache_node);
2619 read_unlock(&fs_info->block_group_cache_lock);
2624 static int pin_down_extent(struct btrfs_trans_handle *trans,
2625 struct btrfs_block_group *cache,
2626 u64 bytenr, u64 num_bytes, int reserved)
2628 struct btrfs_fs_info *fs_info = cache->fs_info;
2630 spin_lock(&cache->space_info->lock);
2631 spin_lock(&cache->lock);
2632 cache->pinned += num_bytes;
2633 btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2636 cache->reserved -= num_bytes;
2637 cache->space_info->bytes_reserved -= num_bytes;
2639 spin_unlock(&cache->lock);
2640 spin_unlock(&cache->space_info->lock);
2642 set_extent_bit(&trans->transaction->pinned_extents, bytenr,
2643 bytenr + num_bytes - 1, EXTENT_DIRTY, NULL);
2647 int btrfs_pin_extent(struct btrfs_trans_handle *trans,
2648 u64 bytenr, u64 num_bytes, int reserved)
2650 struct btrfs_block_group *cache;
2652 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2653 BUG_ON(!cache); /* Logic error */
2655 pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
2657 btrfs_put_block_group(cache);
2661 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
2662 const struct extent_buffer *eb)
2664 struct btrfs_block_group *cache;
2667 cache = btrfs_lookup_block_group(trans->fs_info, eb->start);
2672 * Fully cache the free space first so that our pin removes the free space
2675 ret = btrfs_cache_block_group(cache, true);
2679 pin_down_extent(trans, cache, eb->start, eb->len, 0);
2681 /* remove us from the free space cache (if we're there at all) */
2682 ret = btrfs_remove_free_space(cache, eb->start, eb->len);
2684 btrfs_put_block_group(cache);
2688 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2689 u64 start, u64 num_bytes)
2692 struct btrfs_block_group *block_group;
2694 block_group = btrfs_lookup_block_group(fs_info, start);
2698 ret = btrfs_cache_block_group(block_group, true);
2702 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2704 btrfs_put_block_group(block_group);
2708 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2710 struct btrfs_fs_info *fs_info = eb->fs_info;
2711 struct btrfs_file_extent_item *item;
2712 struct btrfs_key key;
2717 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2720 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2721 btrfs_item_key_to_cpu(eb, &key, i);
2722 if (key.type != BTRFS_EXTENT_DATA_KEY)
2724 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2725 found_type = btrfs_file_extent_type(eb, item);
2726 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2728 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2730 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2731 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2732 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2741 btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2743 atomic_inc(&bg->reservations);
2747 * Returns the free cluster for the given space info and sets empty_cluster to
2748 * what it should be based on the mount options.
2750 static struct btrfs_free_cluster *
2751 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2752 struct btrfs_space_info *space_info, u64 *empty_cluster)
2754 struct btrfs_free_cluster *ret = NULL;
2757 if (btrfs_mixed_space_info(space_info))
2760 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2761 ret = &fs_info->meta_alloc_cluster;
2762 if (btrfs_test_opt(fs_info, SSD))
2763 *empty_cluster = SZ_2M;
2765 *empty_cluster = SZ_64K;
2766 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2767 btrfs_test_opt(fs_info, SSD_SPREAD)) {
2768 *empty_cluster = SZ_2M;
2769 ret = &fs_info->data_alloc_cluster;
2775 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2777 const bool return_free_space)
2779 struct btrfs_block_group *cache = NULL;
2780 struct btrfs_space_info *space_info;
2781 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2782 struct btrfs_free_cluster *cluster = NULL;
2784 u64 total_unpinned = 0;
2785 u64 empty_cluster = 0;
2789 while (start <= end) {
2792 start >= cache->start + cache->length) {
2794 btrfs_put_block_group(cache);
2796 cache = btrfs_lookup_block_group(fs_info, start);
2797 if (cache == NULL) {
2798 /* Logic error, something removed the block group. */
2803 cluster = fetch_cluster_info(fs_info,
2806 empty_cluster <<= 1;
2809 len = cache->start + cache->length - start;
2810 len = min(len, end + 1 - start);
2812 if (return_free_space)
2813 btrfs_add_free_space(cache, start, len);
2816 total_unpinned += len;
2817 space_info = cache->space_info;
2820 * If this space cluster has been marked as fragmented and we've
2821 * unpinned enough in this block group to potentially allow a
2822 * cluster to be created inside of it go ahead and clear the
2825 if (cluster && cluster->fragmented &&
2826 total_unpinned > empty_cluster) {
2827 spin_lock(&cluster->lock);
2828 cluster->fragmented = 0;
2829 spin_unlock(&cluster->lock);
2832 spin_lock(&space_info->lock);
2833 spin_lock(&cache->lock);
2834 cache->pinned -= len;
2835 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2836 space_info->max_extent_size = 0;
2838 space_info->bytes_readonly += len;
2840 } else if (btrfs_is_zoned(fs_info)) {
2841 /* Need reset before reusing in a zoned block group */
2842 space_info->bytes_zone_unusable += len;
2845 spin_unlock(&cache->lock);
2846 if (!readonly && return_free_space &&
2847 global_rsv->space_info == space_info) {
2848 spin_lock(&global_rsv->lock);
2849 if (!global_rsv->full) {
2850 u64 to_add = min(len, global_rsv->size -
2851 global_rsv->reserved);
2853 global_rsv->reserved += to_add;
2854 btrfs_space_info_update_bytes_may_use(fs_info,
2855 space_info, to_add);
2856 if (global_rsv->reserved >= global_rsv->size)
2857 global_rsv->full = 1;
2860 spin_unlock(&global_rsv->lock);
2862 /* Add to any tickets we may have */
2863 if (!readonly && return_free_space && len)
2864 btrfs_try_granting_tickets(fs_info, space_info);
2865 spin_unlock(&space_info->lock);
2869 btrfs_put_block_group(cache);
2874 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2876 struct btrfs_fs_info *fs_info = trans->fs_info;
2877 struct btrfs_block_group *block_group, *tmp;
2878 struct list_head *deleted_bgs;
2879 struct extent_io_tree *unpin;
2884 unpin = &trans->transaction->pinned_extents;
2886 while (!TRANS_ABORTED(trans)) {
2887 struct extent_state *cached_state = NULL;
2889 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2890 if (!find_first_extent_bit(unpin, 0, &start, &end,
2891 EXTENT_DIRTY, &cached_state)) {
2892 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2896 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2897 ret = btrfs_discard_extent(fs_info, start,
2898 end + 1 - start, NULL);
2900 clear_extent_dirty(unpin, start, end, &cached_state);
2901 ret = unpin_extent_range(fs_info, start, end, true);
2903 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2904 free_extent_state(cached_state);
2908 if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2909 btrfs_discard_calc_delay(&fs_info->discard_ctl);
2910 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2914 * Transaction is finished. We don't need the lock anymore. We
2915 * do need to clean up the block groups in case of a transaction
2918 deleted_bgs = &trans->transaction->deleted_bgs;
2919 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2923 if (!TRANS_ABORTED(trans))
2924 ret = btrfs_discard_extent(fs_info,
2926 block_group->length,
2929 list_del_init(&block_group->bg_list);
2930 btrfs_unfreeze_block_group(block_group);
2931 btrfs_put_block_group(block_group);
2934 const char *errstr = btrfs_decode_error(ret);
2936 "discard failed while removing blockgroup: errno=%d %s",
2945 * Parse an extent item's inline extents looking for a simple quotas owner ref.
2947 * @fs_info: the btrfs_fs_info for this mount
2948 * @leaf: a leaf in the extent tree containing the extent item
2949 * @slot: the slot in the leaf where the extent item is found
2951 * Returns the objectid of the root that originally allocated the extent item
2952 * if the inline owner ref is expected and present, otherwise 0.
2954 * If an extent item has an owner ref item, it will be the first inline ref
2955 * item. Therefore the logic is to check whether there are any inline ref
2956 * items, then check the type of the first one.
2958 u64 btrfs_get_extent_owner_root(struct btrfs_fs_info *fs_info,
2959 struct extent_buffer *leaf, int slot)
2961 struct btrfs_extent_item *ei;
2962 struct btrfs_extent_inline_ref *iref;
2963 struct btrfs_extent_owner_ref *oref;
2968 if (!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA))
2971 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
2972 ptr = (unsigned long)(ei + 1);
2973 end = (unsigned long)ei + btrfs_item_size(leaf, slot);
2975 /* No inline ref items of any kind, can't check type. */
2979 iref = (struct btrfs_extent_inline_ref *)ptr;
2980 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
2982 /* We found an owner ref, get the root out of it. */
2983 if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
2984 oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
2985 return btrfs_extent_owner_ref_root_id(leaf, oref);
2988 /* We have inline refs, but not an owner ref. */
2992 static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
2993 u64 bytenr, struct btrfs_squota_delta *delta)
2996 u64 num_bytes = delta->num_bytes;
2998 if (delta->is_data) {
2999 struct btrfs_root *csum_root;
3001 csum_root = btrfs_csum_root(trans->fs_info, bytenr);
3002 ret = btrfs_del_csums(trans, csum_root, bytenr, num_bytes);
3004 btrfs_abort_transaction(trans, ret);
3008 ret = btrfs_delete_raid_extent(trans, bytenr, num_bytes);
3010 btrfs_abort_transaction(trans, ret);
3015 ret = btrfs_record_squota_delta(trans->fs_info, delta);
3017 btrfs_abort_transaction(trans, ret);
3021 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
3023 btrfs_abort_transaction(trans, ret);
3027 ret = btrfs_update_block_group(trans, bytenr, num_bytes, false);
3029 btrfs_abort_transaction(trans, ret);
3034 #define abort_and_dump(trans, path, fmt, args...) \
3036 btrfs_abort_transaction(trans, -EUCLEAN); \
3037 btrfs_print_leaf(path->nodes[0]); \
3038 btrfs_crit(trans->fs_info, fmt, ##args); \
3042 * Drop one or more refs of @node.
3044 * 1. Locate the extent refs.
3045 * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
3046 * Locate it, then reduce the refs number or remove the ref line completely.
3048 * 2. Update the refs count in EXTENT/METADATA_ITEM
3050 * Inline backref case:
3052 * in extent tree we have:
3054 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
3055 * refs 2 gen 6 flags DATA
3056 * extent data backref root FS_TREE objectid 258 offset 0 count 1
3057 * extent data backref root FS_TREE objectid 257 offset 0 count 1
3059 * This function gets called with:
3061 * node->bytenr = 13631488
3062 * node->num_bytes = 1048576
3063 * root_objectid = FS_TREE
3064 * owner_objectid = 257
3068 * Then we should get some like:
3070 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
3071 * refs 1 gen 6 flags DATA
3072 * extent data backref root FS_TREE objectid 258 offset 0 count 1
3074 * Keyed backref case:
3076 * in extent tree we have:
3078 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
3079 * refs 754 gen 6 flags DATA
3081 * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
3082 * extent data backref root FS_TREE objectid 866 offset 0 count 1
3084 * This function get called with:
3086 * node->bytenr = 13631488
3087 * node->num_bytes = 1048576
3088 * root_objectid = FS_TREE
3089 * owner_objectid = 866
3093 * Then we should get some like:
3095 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
3096 * refs 753 gen 6 flags DATA
3098 * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
3100 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3101 struct btrfs_delayed_ref_head *href,
3102 struct btrfs_delayed_ref_node *node, u64 parent,
3103 u64 root_objectid, u64 owner_objectid,
3105 struct btrfs_delayed_extent_op *extent_op)
3107 struct btrfs_fs_info *info = trans->fs_info;
3108 struct btrfs_key key;
3109 struct btrfs_path *path;
3110 struct btrfs_root *extent_root;
3111 struct extent_buffer *leaf;
3112 struct btrfs_extent_item *ei;
3113 struct btrfs_extent_inline_ref *iref;
3116 int extent_slot = 0;
3117 int found_extent = 0;
3119 int refs_to_drop = node->ref_mod;
3122 u64 bytenr = node->bytenr;
3123 u64 num_bytes = node->num_bytes;
3124 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
3125 u64 delayed_ref_root = href->owning_root;
3127 extent_root = btrfs_extent_root(info, bytenr);
3128 ASSERT(extent_root);
3130 path = btrfs_alloc_path();
3134 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3136 if (!is_data && refs_to_drop != 1) {
3138 "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
3139 node->bytenr, refs_to_drop);
3141 btrfs_abort_transaction(trans, ret);
3146 skinny_metadata = false;
3148 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
3149 parent, root_objectid, owner_objectid,
3153 * Either the inline backref or the SHARED_DATA_REF/
3154 * SHARED_BLOCK_REF is found
3156 * Here is a quick path to locate EXTENT/METADATA_ITEM.
3157 * It's possible the EXTENT/METADATA_ITEM is near current slot.
3159 extent_slot = path->slots[0];
3160 while (extent_slot >= 0) {
3161 btrfs_item_key_to_cpu(path->nodes[0], &key,
3163 if (key.objectid != bytenr)
3165 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3166 key.offset == num_bytes) {
3170 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3171 key.offset == owner_objectid) {
3176 /* Quick path didn't find the EXTEMT/METADATA_ITEM */
3177 if (path->slots[0] - extent_slot > 5)
3182 if (!found_extent) {
3184 abort_and_dump(trans, path,
3185 "invalid iref slot %u, no EXTENT/METADATA_ITEM found but has inline extent ref",
3190 /* Must be SHARED_* item, remove the backref first */
3191 ret = remove_extent_backref(trans, extent_root, path,
3192 NULL, refs_to_drop, is_data);
3194 btrfs_abort_transaction(trans, ret);
3197 btrfs_release_path(path);
3199 /* Slow path to locate EXTENT/METADATA_ITEM */
3200 key.objectid = bytenr;
3201 key.type = BTRFS_EXTENT_ITEM_KEY;
3202 key.offset = num_bytes;
3204 if (!is_data && skinny_metadata) {
3205 key.type = BTRFS_METADATA_ITEM_KEY;
3206 key.offset = owner_objectid;
3209 ret = btrfs_search_slot(trans, extent_root,
3211 if (ret > 0 && skinny_metadata && path->slots[0]) {
3213 * Couldn't find our skinny metadata item,
3214 * see if we have ye olde extent item.
3217 btrfs_item_key_to_cpu(path->nodes[0], &key,
3219 if (key.objectid == bytenr &&
3220 key.type == BTRFS_EXTENT_ITEM_KEY &&
3221 key.offset == num_bytes)
3225 if (ret > 0 && skinny_metadata) {
3226 skinny_metadata = false;
3227 key.objectid = bytenr;
3228 key.type = BTRFS_EXTENT_ITEM_KEY;
3229 key.offset = num_bytes;
3230 btrfs_release_path(path);
3231 ret = btrfs_search_slot(trans, extent_root,
3237 btrfs_print_leaf(path->nodes[0]);
3239 "umm, got %d back from search, was looking for %llu, slot %d",
3240 ret, bytenr, path->slots[0]);
3243 btrfs_abort_transaction(trans, ret);
3246 extent_slot = path->slots[0];
3248 } else if (WARN_ON(ret == -ENOENT)) {
3249 abort_and_dump(trans, path,
3250 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu slot %d",
3251 bytenr, parent, root_objectid, owner_objectid,
3252 owner_offset, path->slots[0]);
3255 btrfs_abort_transaction(trans, ret);
3259 leaf = path->nodes[0];
3260 item_size = btrfs_item_size(leaf, extent_slot);
3261 if (unlikely(item_size < sizeof(*ei))) {
3263 btrfs_err(trans->fs_info,
3264 "unexpected extent item size, has %u expect >= %zu",
3265 item_size, sizeof(*ei));
3266 btrfs_abort_transaction(trans, ret);
3269 ei = btrfs_item_ptr(leaf, extent_slot,
3270 struct btrfs_extent_item);
3271 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3272 key.type == BTRFS_EXTENT_ITEM_KEY) {
3273 struct btrfs_tree_block_info *bi;
3275 if (item_size < sizeof(*ei) + sizeof(*bi)) {
3276 abort_and_dump(trans, path,
3277 "invalid extent item size for key (%llu, %u, %llu) slot %u owner %llu, has %u expect >= %zu",
3278 key.objectid, key.type, key.offset,
3279 path->slots[0], owner_objectid, item_size,
3280 sizeof(*ei) + sizeof(*bi));
3284 bi = (struct btrfs_tree_block_info *)(ei + 1);
3285 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3288 refs = btrfs_extent_refs(leaf, ei);
3289 if (refs < refs_to_drop) {
3290 abort_and_dump(trans, path,
3291 "trying to drop %d refs but we only have %llu for bytenr %llu slot %u",
3292 refs_to_drop, refs, bytenr, path->slots[0]);
3296 refs -= refs_to_drop;
3300 __run_delayed_extent_op(extent_op, leaf, ei);
3302 * In the case of inline back ref, reference count will
3303 * be updated by remove_extent_backref
3306 if (!found_extent) {
3307 abort_and_dump(trans, path,
3308 "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found, slot %u",
3314 btrfs_set_extent_refs(leaf, ei, refs);
3315 btrfs_mark_buffer_dirty(trans, leaf);
3318 ret = remove_extent_backref(trans, extent_root, path,
3319 iref, refs_to_drop, is_data);
3321 btrfs_abort_transaction(trans, ret);
3326 struct btrfs_squota_delta delta = {
3327 .root = delayed_ref_root,
3328 .num_bytes = num_bytes,
3331 .generation = btrfs_extent_generation(leaf, ei),
3334 /* In this branch refs == 1 */
3336 if (is_data && refs_to_drop !=
3337 extent_data_ref_count(path, iref)) {
3338 abort_and_dump(trans, path,
3339 "invalid refs_to_drop, current refs %u refs_to_drop %u slot %u",
3340 extent_data_ref_count(path, iref),
3341 refs_to_drop, path->slots[0]);
3346 if (path->slots[0] != extent_slot) {
3347 abort_and_dump(trans, path,
3348 "invalid iref, extent item key (%llu %u %llu) slot %u doesn't have wanted iref",
3349 key.objectid, key.type,
3350 key.offset, path->slots[0]);
3356 * No inline ref, we must be at SHARED_* item,
3357 * And it's single ref, it must be:
3358 * | extent_slot ||extent_slot + 1|
3359 * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3361 if (path->slots[0] != extent_slot + 1) {
3362 abort_and_dump(trans, path,
3363 "invalid SHARED_* item slot %u, previous item is not EXTENT/METADATA_ITEM",
3368 path->slots[0] = extent_slot;
3373 * We can't infer the data owner from the delayed ref, so we need
3374 * to try to get it from the owning ref item.
3376 * If it is not present, then that extent was not written under
3377 * simple quotas mode, so we don't need to account for its deletion.
3380 delta.root = btrfs_get_extent_owner_root(trans->fs_info,
3383 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3386 btrfs_abort_transaction(trans, ret);
3389 btrfs_release_path(path);
3391 ret = do_free_extent_accounting(trans, bytenr, &delta);
3393 btrfs_release_path(path);
3396 btrfs_free_path(path);
3401 * when we free an block, it is possible (and likely) that we free the last
3402 * delayed ref for that extent as well. This searches the delayed ref tree for
3403 * a given extent, and if there are no other delayed refs to be processed, it
3404 * removes it from the tree.
3406 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3409 struct btrfs_delayed_ref_head *head;
3410 struct btrfs_delayed_ref_root *delayed_refs;
3413 delayed_refs = &trans->transaction->delayed_refs;
3414 spin_lock(&delayed_refs->lock);
3415 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3417 goto out_delayed_unlock;
3419 spin_lock(&head->lock);
3420 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3423 if (cleanup_extent_op(head) != NULL)
3427 * waiting for the lock here would deadlock. If someone else has it
3428 * locked they are already in the process of dropping it anyway
3430 if (!mutex_trylock(&head->mutex))
3433 btrfs_delete_ref_head(delayed_refs, head);
3434 head->processing = false;
3436 spin_unlock(&head->lock);
3437 spin_unlock(&delayed_refs->lock);
3439 BUG_ON(head->extent_op);
3440 if (head->must_insert_reserved)
3443 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3444 mutex_unlock(&head->mutex);
3445 btrfs_put_delayed_ref_head(head);
3448 spin_unlock(&head->lock);
3451 spin_unlock(&delayed_refs->lock);
3455 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3457 struct extent_buffer *buf,
3458 u64 parent, int last_ref)
3460 struct btrfs_fs_info *fs_info = trans->fs_info;
3461 struct btrfs_block_group *bg;
3464 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3465 struct btrfs_ref generic_ref = { 0 };
3468 * Assert that the extent buffer is not cleared due to
3469 * EXTENT_BUFFER_ZONED_ZEROOUT. Please refer
3470 * btrfs_clear_buffer_dirty() and btree_csum_one_bio() for
3473 ASSERT(btrfs_header_bytenr(buf) != 0);
3475 btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3476 buf->start, buf->len, parent,
3477 btrfs_header_owner(buf));
3478 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3480 btrfs_ref_tree_mod(fs_info, &generic_ref);
3481 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL);
3482 BUG_ON(ret); /* -ENOMEM */
3488 if (btrfs_header_generation(buf) != trans->transid)
3491 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3492 ret = check_ref_cleanup(trans, buf->start);
3497 bg = btrfs_lookup_block_group(fs_info, buf->start);
3499 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3500 pin_down_extent(trans, bg, buf->start, buf->len, 1);
3501 btrfs_put_block_group(bg);
3506 * If there are tree mod log users we may have recorded mod log
3507 * operations for this node. If we re-allocate this node we
3508 * could replay operations on this node that happened when it
3509 * existed in a completely different root. For example if it
3510 * was part of root A, then was reallocated to root B, and we
3511 * are doing a btrfs_old_search_slot(root b), we could replay
3512 * operations that happened when the block was part of root A,
3513 * giving us an inconsistent view of the btree.
3515 * We are safe from races here because at this point no other
3516 * node or root points to this extent buffer, so if after this
3517 * check a new tree mod log user joins we will not have an
3518 * existing log of operations on this node that we have to
3522 if (test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags)
3523 || btrfs_is_zoned(fs_info)) {
3524 pin_down_extent(trans, bg, buf->start, buf->len, 1);
3525 btrfs_put_block_group(bg);
3529 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3531 btrfs_add_free_space(bg, buf->start, buf->len);
3532 btrfs_free_reserved_bytes(bg, buf->len, 0);
3533 btrfs_put_block_group(bg);
3534 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3539 * Deleting the buffer, clear the corrupt flag since it doesn't
3542 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3545 /* Can return -ENOMEM */
3546 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3548 struct btrfs_fs_info *fs_info = trans->fs_info;
3551 if (btrfs_is_testing(fs_info))
3555 * tree log blocks never actually go into the extent allocation
3556 * tree, just update pinning info and exit early.
3558 if ((ref->type == BTRFS_REF_METADATA &&
3559 ref->tree_ref.ref_root == BTRFS_TREE_LOG_OBJECTID) ||
3560 (ref->type == BTRFS_REF_DATA &&
3561 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) {
3562 btrfs_pin_extent(trans, ref->bytenr, ref->len, 1);
3564 } else if (ref->type == BTRFS_REF_METADATA) {
3565 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL);
3567 ret = btrfs_add_delayed_data_ref(trans, ref, 0);
3570 if (!((ref->type == BTRFS_REF_METADATA &&
3571 ref->tree_ref.ref_root == BTRFS_TREE_LOG_OBJECTID) ||
3572 (ref->type == BTRFS_REF_DATA &&
3573 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)))
3574 btrfs_ref_tree_mod(fs_info, ref);
3579 enum btrfs_loop_type {
3581 * Start caching block groups but do not wait for progress or for them
3584 LOOP_CACHING_NOWAIT,
3587 * Wait for the block group free_space >= the space we're waiting for if
3588 * the block group isn't cached.
3593 * Allow allocations to happen from block groups that do not yet have a
3594 * size classification.
3596 LOOP_UNSET_SIZE_CLASS,
3599 * Allocate a chunk and then retry the allocation.
3604 * Ignore the size class restrictions for this allocation.
3606 LOOP_WRONG_SIZE_CLASS,
3609 * Ignore the empty size, only try to allocate the number of bytes
3610 * needed for this allocation.
3616 btrfs_lock_block_group(struct btrfs_block_group *cache,
3620 down_read(&cache->data_rwsem);
3623 static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3626 btrfs_get_block_group(cache);
3628 down_read(&cache->data_rwsem);
3631 static struct btrfs_block_group *btrfs_lock_cluster(
3632 struct btrfs_block_group *block_group,
3633 struct btrfs_free_cluster *cluster,
3635 __acquires(&cluster->refill_lock)
3637 struct btrfs_block_group *used_bg = NULL;
3639 spin_lock(&cluster->refill_lock);
3641 used_bg = cluster->block_group;
3645 if (used_bg == block_group)
3648 btrfs_get_block_group(used_bg);
3653 if (down_read_trylock(&used_bg->data_rwsem))
3656 spin_unlock(&cluster->refill_lock);
3658 /* We should only have one-level nested. */
3659 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3661 spin_lock(&cluster->refill_lock);
3662 if (used_bg == cluster->block_group)
3665 up_read(&used_bg->data_rwsem);
3666 btrfs_put_block_group(used_bg);
3671 btrfs_release_block_group(struct btrfs_block_group *cache,
3675 up_read(&cache->data_rwsem);
3676 btrfs_put_block_group(cache);
3680 * Helper function for find_free_extent().
3682 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3683 * Return >0 to inform caller that we find nothing
3684 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3686 static int find_free_extent_clustered(struct btrfs_block_group *bg,
3687 struct find_free_extent_ctl *ffe_ctl,
3688 struct btrfs_block_group **cluster_bg_ret)
3690 struct btrfs_block_group *cluster_bg;
3691 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3692 u64 aligned_cluster;
3696 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3698 goto refill_cluster;
3699 if (cluster_bg != bg && (cluster_bg->ro ||
3700 !block_group_bits(cluster_bg, ffe_ctl->flags)))
3701 goto release_cluster;
3703 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3704 ffe_ctl->num_bytes, cluster_bg->start,
3705 &ffe_ctl->max_extent_size);
3707 /* We have a block, we're done */
3708 spin_unlock(&last_ptr->refill_lock);
3709 trace_btrfs_reserve_extent_cluster(cluster_bg, ffe_ctl);
3710 *cluster_bg_ret = cluster_bg;
3711 ffe_ctl->found_offset = offset;
3714 WARN_ON(last_ptr->block_group != cluster_bg);
3718 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3719 * lets just skip it and let the allocator find whatever block it can
3720 * find. If we reach this point, we will have tried the cluster
3721 * allocator plenty of times and not have found anything, so we are
3722 * likely way too fragmented for the clustering stuff to find anything.
3724 * However, if the cluster is taken from the current block group,
3725 * release the cluster first, so that we stand a better chance of
3726 * succeeding in the unclustered allocation.
3728 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3729 spin_unlock(&last_ptr->refill_lock);
3730 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3734 /* This cluster didn't work out, free it and start over */
3735 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3737 if (cluster_bg != bg)
3738 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3741 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3742 spin_unlock(&last_ptr->refill_lock);
3746 aligned_cluster = max_t(u64,
3747 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3748 bg->full_stripe_len);
3749 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3750 ffe_ctl->num_bytes, aligned_cluster);
3752 /* Now pull our allocation out of this cluster */
3753 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3754 ffe_ctl->num_bytes, ffe_ctl->search_start,
3755 &ffe_ctl->max_extent_size);
3757 /* We found one, proceed */
3758 spin_unlock(&last_ptr->refill_lock);
3759 ffe_ctl->found_offset = offset;
3760 trace_btrfs_reserve_extent_cluster(bg, ffe_ctl);
3765 * At this point we either didn't find a cluster or we weren't able to
3766 * allocate a block from our cluster. Free the cluster we've been
3767 * trying to use, and go to the next block group.
3769 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3770 spin_unlock(&last_ptr->refill_lock);
3775 * Return >0 to inform caller that we find nothing
3776 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3778 static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3779 struct find_free_extent_ctl *ffe_ctl)
3781 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3785 * We are doing an unclustered allocation, set the fragmented flag so
3786 * we don't bother trying to setup a cluster again until we get more
3789 if (unlikely(last_ptr)) {
3790 spin_lock(&last_ptr->lock);
3791 last_ptr->fragmented = 1;
3792 spin_unlock(&last_ptr->lock);
3794 if (ffe_ctl->cached) {
3795 struct btrfs_free_space_ctl *free_space_ctl;
3797 free_space_ctl = bg->free_space_ctl;
3798 spin_lock(&free_space_ctl->tree_lock);
3799 if (free_space_ctl->free_space <
3800 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3801 ffe_ctl->empty_size) {
3802 ffe_ctl->total_free_space = max_t(u64,
3803 ffe_ctl->total_free_space,
3804 free_space_ctl->free_space);
3805 spin_unlock(&free_space_ctl->tree_lock);
3808 spin_unlock(&free_space_ctl->tree_lock);
3811 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3812 ffe_ctl->num_bytes, ffe_ctl->empty_size,
3813 &ffe_ctl->max_extent_size);
3816 ffe_ctl->found_offset = offset;
3820 static int do_allocation_clustered(struct btrfs_block_group *block_group,
3821 struct find_free_extent_ctl *ffe_ctl,
3822 struct btrfs_block_group **bg_ret)
3826 /* We want to try and use the cluster allocator, so lets look there */
3827 if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
3828 ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
3831 /* ret == -ENOENT case falls through */
3834 return find_free_extent_unclustered(block_group, ffe_ctl);
3838 * Tree-log block group locking
3839 * ============================
3841 * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3842 * indicates the starting address of a block group, which is reserved only
3843 * for tree-log metadata.
3850 * fs_info::treelog_bg_lock
3854 * Simple allocator for sequential-only block group. It only allows sequential
3855 * allocation. No need to play with trees. This function also reserves the
3856 * bytes as in btrfs_add_reserved_bytes.
3858 static int do_allocation_zoned(struct btrfs_block_group *block_group,
3859 struct find_free_extent_ctl *ffe_ctl,
3860 struct btrfs_block_group **bg_ret)
3862 struct btrfs_fs_info *fs_info = block_group->fs_info;
3863 struct btrfs_space_info *space_info = block_group->space_info;
3864 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3865 u64 start = block_group->start;
3866 u64 num_bytes = ffe_ctl->num_bytes;
3868 u64 bytenr = block_group->start;
3870 u64 data_reloc_bytenr;
3874 ASSERT(btrfs_is_zoned(block_group->fs_info));
3877 * Do not allow non-tree-log blocks in the dedicated tree-log block
3878 * group, and vice versa.
3880 spin_lock(&fs_info->treelog_bg_lock);
3881 log_bytenr = fs_info->treelog_bg;
3882 if (log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) ||
3883 (!ffe_ctl->for_treelog && bytenr == log_bytenr)))
3885 spin_unlock(&fs_info->treelog_bg_lock);
3890 * Do not allow non-relocation blocks in the dedicated relocation block
3891 * group, and vice versa.
3893 spin_lock(&fs_info->relocation_bg_lock);
3894 data_reloc_bytenr = fs_info->data_reloc_bg;
3895 if (data_reloc_bytenr &&
3896 ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) ||
3897 (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr)))
3899 spin_unlock(&fs_info->relocation_bg_lock);
3903 /* Check RO and no space case before trying to activate it */
3904 spin_lock(&block_group->lock);
3905 if (block_group->ro || btrfs_zoned_bg_is_full(block_group)) {
3908 * May need to clear fs_info->{treelog,data_reloc}_bg.
3909 * Return the error after taking the locks.
3912 spin_unlock(&block_group->lock);
3914 /* Metadata block group is activated at write time. */
3915 if (!ret && (block_group->flags & BTRFS_BLOCK_GROUP_DATA) &&
3916 !btrfs_zone_activate(block_group)) {
3919 * May need to clear fs_info->{treelog,data_reloc}_bg.
3920 * Return the error after taking the locks.
3924 spin_lock(&space_info->lock);
3925 spin_lock(&block_group->lock);
3926 spin_lock(&fs_info->treelog_bg_lock);
3927 spin_lock(&fs_info->relocation_bg_lock);
3932 ASSERT(!ffe_ctl->for_treelog ||
3933 block_group->start == fs_info->treelog_bg ||
3934 fs_info->treelog_bg == 0);
3935 ASSERT(!ffe_ctl->for_data_reloc ||
3936 block_group->start == fs_info->data_reloc_bg ||
3937 fs_info->data_reloc_bg == 0);
3939 if (block_group->ro ||
3940 (!ffe_ctl->for_data_reloc &&
3941 test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags))) {
3947 * Do not allow currently using block group to be tree-log dedicated
3950 if (ffe_ctl->for_treelog && !fs_info->treelog_bg &&
3951 (block_group->used || block_group->reserved)) {
3957 * Do not allow currently used block group to be the data relocation
3958 * dedicated block group.
3960 if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg &&
3961 (block_group->used || block_group->reserved)) {
3966 WARN_ON_ONCE(block_group->alloc_offset > block_group->zone_capacity);
3967 avail = block_group->zone_capacity - block_group->alloc_offset;
3968 if (avail < num_bytes) {
3969 if (ffe_ctl->max_extent_size < avail) {
3971 * With sequential allocator, free space is always
3974 ffe_ctl->max_extent_size = avail;
3975 ffe_ctl->total_free_space = avail;
3981 if (ffe_ctl->for_treelog && !fs_info->treelog_bg)
3982 fs_info->treelog_bg = block_group->start;
3984 if (ffe_ctl->for_data_reloc) {
3985 if (!fs_info->data_reloc_bg)
3986 fs_info->data_reloc_bg = block_group->start;
3988 * Do not allow allocations from this block group, unless it is
3989 * for data relocation. Compared to increasing the ->ro, setting
3990 * the ->zoned_data_reloc_ongoing flag still allows nocow
3991 * writers to come in. See btrfs_inc_nocow_writers().
3993 * We need to disable an allocation to avoid an allocation of
3994 * regular (non-relocation data) extent. With mix of relocation
3995 * extents and regular extents, we can dispatch WRITE commands
3996 * (for relocation extents) and ZONE APPEND commands (for
3997 * regular extents) at the same time to the same zone, which
3998 * easily break the write pointer.
4000 * Also, this flag avoids this block group to be zone finished.
4002 set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags);
4005 ffe_ctl->found_offset = start + block_group->alloc_offset;
4006 block_group->alloc_offset += num_bytes;
4007 spin_lock(&ctl->tree_lock);
4008 ctl->free_space -= num_bytes;
4009 spin_unlock(&ctl->tree_lock);
4012 * We do not check if found_offset is aligned to stripesize. The
4013 * address is anyway rewritten when using zone append writing.
4016 ffe_ctl->search_start = ffe_ctl->found_offset;
4019 if (ret && ffe_ctl->for_treelog)
4020 fs_info->treelog_bg = 0;
4021 if (ret && ffe_ctl->for_data_reloc)
4022 fs_info->data_reloc_bg = 0;
4023 spin_unlock(&fs_info->relocation_bg_lock);
4024 spin_unlock(&fs_info->treelog_bg_lock);
4025 spin_unlock(&block_group->lock);
4026 spin_unlock(&space_info->lock);
4030 static int do_allocation(struct btrfs_block_group *block_group,
4031 struct find_free_extent_ctl *ffe_ctl,
4032 struct btrfs_block_group **bg_ret)
4034 switch (ffe_ctl->policy) {
4035 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4036 return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
4037 case BTRFS_EXTENT_ALLOC_ZONED:
4038 return do_allocation_zoned(block_group, ffe_ctl, bg_ret);
4044 static void release_block_group(struct btrfs_block_group *block_group,
4045 struct find_free_extent_ctl *ffe_ctl,
4048 switch (ffe_ctl->policy) {
4049 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4050 ffe_ctl->retry_uncached = false;
4052 case BTRFS_EXTENT_ALLOC_ZONED:
4059 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
4061 btrfs_release_block_group(block_group, delalloc);
4064 static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
4065 struct btrfs_key *ins)
4067 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4069 if (!ffe_ctl->use_cluster && last_ptr) {
4070 spin_lock(&last_ptr->lock);
4071 last_ptr->window_start = ins->objectid;
4072 spin_unlock(&last_ptr->lock);
4076 static void found_extent(struct find_free_extent_ctl *ffe_ctl,
4077 struct btrfs_key *ins)
4079 switch (ffe_ctl->policy) {
4080 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4081 found_extent_clustered(ffe_ctl, ins);
4083 case BTRFS_EXTENT_ALLOC_ZONED:
4091 static int can_allocate_chunk_zoned(struct btrfs_fs_info *fs_info,
4092 struct find_free_extent_ctl *ffe_ctl)
4094 /* Block group's activeness is not a requirement for METADATA block groups. */
4095 if (!(ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA))
4098 /* If we can activate new zone, just allocate a chunk and use it */
4099 if (btrfs_can_activate_zone(fs_info->fs_devices, ffe_ctl->flags))
4103 * We already reached the max active zones. Try to finish one block
4104 * group to make a room for a new block group. This is only possible
4105 * for a data block group because btrfs_zone_finish() may need to wait
4106 * for a running transaction which can cause a deadlock for metadata
4109 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
4110 int ret = btrfs_zone_finish_one_bg(fs_info);
4119 * If we have enough free space left in an already active block group
4120 * and we can't activate any other zone now, do not allow allocating a
4121 * new chunk and let find_free_extent() retry with a smaller size.
4123 if (ffe_ctl->max_extent_size >= ffe_ctl->min_alloc_size)
4127 * Even min_alloc_size is not left in any block groups. Since we cannot
4128 * activate a new block group, allocating it may not help. Let's tell a
4129 * caller to try again and hope it progress something by writing some
4130 * parts of the region. That is only possible for data block groups,
4131 * where a part of the region can be written.
4133 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA)
4137 * We cannot activate a new block group and no enough space left in any
4138 * block groups. So, allocating a new block group may not help. But,
4139 * there is nothing to do anyway, so let's go with it.
4144 static int can_allocate_chunk(struct btrfs_fs_info *fs_info,
4145 struct find_free_extent_ctl *ffe_ctl)
4147 switch (ffe_ctl->policy) {
4148 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4150 case BTRFS_EXTENT_ALLOC_ZONED:
4151 return can_allocate_chunk_zoned(fs_info, ffe_ctl);
4158 * Return >0 means caller needs to re-search for free extent
4159 * Return 0 means we have the needed free extent.
4160 * Return <0 means we failed to locate any free extent.
4162 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
4163 struct btrfs_key *ins,
4164 struct find_free_extent_ctl *ffe_ctl,
4167 struct btrfs_root *root = fs_info->chunk_root;
4170 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
4171 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
4172 ffe_ctl->orig_have_caching_bg = true;
4174 if (ins->objectid) {
4175 found_extent(ffe_ctl, ins);
4179 if (ffe_ctl->loop >= LOOP_CACHING_WAIT && ffe_ctl->have_caching_bg)
4183 if (ffe_ctl->index < BTRFS_NR_RAID_TYPES)
4186 /* See the comments for btrfs_loop_type for an explanation of the phases. */
4187 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
4190 * We want to skip the LOOP_CACHING_WAIT step if we don't have
4191 * any uncached bgs and we've already done a full search
4194 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT &&
4195 (!ffe_ctl->orig_have_caching_bg && full_search))
4199 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
4200 struct btrfs_trans_handle *trans;
4203 /* Check if allocation policy allows to create a new chunk */
4204 ret = can_allocate_chunk(fs_info, ffe_ctl);
4208 trans = current->journal_info;
4212 trans = btrfs_join_transaction(root);
4214 if (IS_ERR(trans)) {
4215 ret = PTR_ERR(trans);
4219 ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
4220 CHUNK_ALLOC_FORCE_FOR_EXTENT);
4222 /* Do not bail out on ENOSPC since we can do more. */
4223 if (ret == -ENOSPC) {
4228 btrfs_abort_transaction(trans, ret);
4232 btrfs_end_transaction(trans);
4237 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
4238 if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
4242 * Don't loop again if we already have no empty_size and
4245 if (ffe_ctl->empty_size == 0 &&
4246 ffe_ctl->empty_cluster == 0)
4248 ffe_ctl->empty_size = 0;
4249 ffe_ctl->empty_cluster = 0;
4256 static bool find_free_extent_check_size_class(struct find_free_extent_ctl *ffe_ctl,
4257 struct btrfs_block_group *bg)
4259 if (ffe_ctl->policy == BTRFS_EXTENT_ALLOC_ZONED)
4261 if (!btrfs_block_group_should_use_size_class(bg))
4263 if (ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS)
4265 if (ffe_ctl->loop >= LOOP_UNSET_SIZE_CLASS &&
4266 bg->size_class == BTRFS_BG_SZ_NONE)
4268 return ffe_ctl->size_class == bg->size_class;
4271 static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
4272 struct find_free_extent_ctl *ffe_ctl,
4273 struct btrfs_space_info *space_info,
4274 struct btrfs_key *ins)
4277 * If our free space is heavily fragmented we may not be able to make
4278 * big contiguous allocations, so instead of doing the expensive search
4279 * for free space, simply return ENOSPC with our max_extent_size so we
4280 * can go ahead and search for a more manageable chunk.
4282 * If our max_extent_size is large enough for our allocation simply
4283 * disable clustering since we will likely not be able to find enough
4284 * space to create a cluster and induce latency trying.
4286 if (space_info->max_extent_size) {
4287 spin_lock(&space_info->lock);
4288 if (space_info->max_extent_size &&
4289 ffe_ctl->num_bytes > space_info->max_extent_size) {
4290 ins->offset = space_info->max_extent_size;
4291 spin_unlock(&space_info->lock);
4293 } else if (space_info->max_extent_size) {
4294 ffe_ctl->use_cluster = false;
4296 spin_unlock(&space_info->lock);
4299 ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
4300 &ffe_ctl->empty_cluster);
4301 if (ffe_ctl->last_ptr) {
4302 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4304 spin_lock(&last_ptr->lock);
4305 if (last_ptr->block_group)
4306 ffe_ctl->hint_byte = last_ptr->window_start;
4307 if (last_ptr->fragmented) {
4309 * We still set window_start so we can keep track of the
4310 * last place we found an allocation to try and save
4313 ffe_ctl->hint_byte = last_ptr->window_start;
4314 ffe_ctl->use_cluster = false;
4316 spin_unlock(&last_ptr->lock);
4322 static int prepare_allocation_zoned(struct btrfs_fs_info *fs_info,
4323 struct find_free_extent_ctl *ffe_ctl)
4325 if (ffe_ctl->for_treelog) {
4326 spin_lock(&fs_info->treelog_bg_lock);
4327 if (fs_info->treelog_bg)
4328 ffe_ctl->hint_byte = fs_info->treelog_bg;
4329 spin_unlock(&fs_info->treelog_bg_lock);
4330 } else if (ffe_ctl->for_data_reloc) {
4331 spin_lock(&fs_info->relocation_bg_lock);
4332 if (fs_info->data_reloc_bg)
4333 ffe_ctl->hint_byte = fs_info->data_reloc_bg;
4334 spin_unlock(&fs_info->relocation_bg_lock);
4335 } else if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
4336 struct btrfs_block_group *block_group;
4338 spin_lock(&fs_info->zone_active_bgs_lock);
4339 list_for_each_entry(block_group, &fs_info->zone_active_bgs, active_bg_list) {
4341 * No lock is OK here because avail is monotinically
4342 * decreasing, and this is just a hint.
4344 u64 avail = block_group->zone_capacity - block_group->alloc_offset;
4346 if (block_group_bits(block_group, ffe_ctl->flags) &&
4347 avail >= ffe_ctl->num_bytes) {
4348 ffe_ctl->hint_byte = block_group->start;
4352 spin_unlock(&fs_info->zone_active_bgs_lock);
4358 static int prepare_allocation(struct btrfs_fs_info *fs_info,
4359 struct find_free_extent_ctl *ffe_ctl,
4360 struct btrfs_space_info *space_info,
4361 struct btrfs_key *ins)
4363 switch (ffe_ctl->policy) {
4364 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4365 return prepare_allocation_clustered(fs_info, ffe_ctl,
4367 case BTRFS_EXTENT_ALLOC_ZONED:
4368 return prepare_allocation_zoned(fs_info, ffe_ctl);
4375 * walks the btree of allocated extents and find a hole of a given size.
4376 * The key ins is changed to record the hole:
4377 * ins->objectid == start position
4378 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4379 * ins->offset == the size of the hole.
4380 * Any available blocks before search_start are skipped.
4382 * If there is no suitable free space, we will record the max size of
4383 * the free space extent currently.
4385 * The overall logic and call chain:
4387 * find_free_extent()
4388 * |- Iterate through all block groups
4389 * | |- Get a valid block group
4390 * | |- Try to do clustered allocation in that block group
4391 * | |- Try to do unclustered allocation in that block group
4392 * | |- Check if the result is valid
4393 * | | |- If valid, then exit
4394 * | |- Jump to next block group
4396 * |- Push harder to find free extents
4397 * |- If not found, re-iterate all block groups
4399 static noinline int find_free_extent(struct btrfs_root *root,
4400 struct btrfs_key *ins,
4401 struct find_free_extent_ctl *ffe_ctl)
4403 struct btrfs_fs_info *fs_info = root->fs_info;
4405 int cache_block_group_error = 0;
4406 struct btrfs_block_group *block_group = NULL;
4407 struct btrfs_space_info *space_info;
4408 bool full_search = false;
4410 WARN_ON(ffe_ctl->num_bytes < fs_info->sectorsize);
4412 ffe_ctl->search_start = 0;
4413 /* For clustered allocation */
4414 ffe_ctl->empty_cluster = 0;
4415 ffe_ctl->last_ptr = NULL;
4416 ffe_ctl->use_cluster = true;
4417 ffe_ctl->have_caching_bg = false;
4418 ffe_ctl->orig_have_caching_bg = false;
4419 ffe_ctl->index = btrfs_bg_flags_to_raid_index(ffe_ctl->flags);
4421 ffe_ctl->retry_uncached = false;
4422 ffe_ctl->cached = 0;
4423 ffe_ctl->max_extent_size = 0;
4424 ffe_ctl->total_free_space = 0;
4425 ffe_ctl->found_offset = 0;
4426 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
4427 ffe_ctl->size_class = btrfs_calc_block_group_size_class(ffe_ctl->num_bytes);
4429 if (btrfs_is_zoned(fs_info))
4430 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_ZONED;
4432 ins->type = BTRFS_EXTENT_ITEM_KEY;
4436 trace_find_free_extent(root, ffe_ctl);
4438 space_info = btrfs_find_space_info(fs_info, ffe_ctl->flags);
4440 btrfs_err(fs_info, "No space info for %llu", ffe_ctl->flags);
4444 ret = prepare_allocation(fs_info, ffe_ctl, space_info, ins);
4448 ffe_ctl->search_start = max(ffe_ctl->search_start,
4449 first_logical_byte(fs_info));
4450 ffe_ctl->search_start = max(ffe_ctl->search_start, ffe_ctl->hint_byte);
4451 if (ffe_ctl->search_start == ffe_ctl->hint_byte) {
4452 block_group = btrfs_lookup_block_group(fs_info,
4453 ffe_ctl->search_start);
4455 * we don't want to use the block group if it doesn't match our
4456 * allocation bits, or if its not cached.
4458 * However if we are re-searching with an ideal block group
4459 * picked out then we don't care that the block group is cached.
4461 if (block_group && block_group_bits(block_group, ffe_ctl->flags) &&
4462 block_group->cached != BTRFS_CACHE_NO) {
4463 down_read(&space_info->groups_sem);
4464 if (list_empty(&block_group->list) ||
4467 * someone is removing this block group,
4468 * we can't jump into the have_block_group
4469 * target because our list pointers are not
4472 btrfs_put_block_group(block_group);
4473 up_read(&space_info->groups_sem);
4475 ffe_ctl->index = btrfs_bg_flags_to_raid_index(
4476 block_group->flags);
4477 btrfs_lock_block_group(block_group,
4479 ffe_ctl->hinted = true;
4480 goto have_block_group;
4482 } else if (block_group) {
4483 btrfs_put_block_group(block_group);
4487 trace_find_free_extent_search_loop(root, ffe_ctl);
4488 ffe_ctl->have_caching_bg = false;
4489 if (ffe_ctl->index == btrfs_bg_flags_to_raid_index(ffe_ctl->flags) ||
4490 ffe_ctl->index == 0)
4492 down_read(&space_info->groups_sem);
4493 list_for_each_entry(block_group,
4494 &space_info->block_groups[ffe_ctl->index], list) {
4495 struct btrfs_block_group *bg_ret;
4497 ffe_ctl->hinted = false;
4498 /* If the block group is read-only, we can skip it entirely. */
4499 if (unlikely(block_group->ro)) {
4500 if (ffe_ctl->for_treelog)
4501 btrfs_clear_treelog_bg(block_group);
4502 if (ffe_ctl->for_data_reloc)
4503 btrfs_clear_data_reloc_bg(block_group);
4507 btrfs_grab_block_group(block_group, ffe_ctl->delalloc);
4508 ffe_ctl->search_start = block_group->start;
4511 * this can happen if we end up cycling through all the
4512 * raid types, but we want to make sure we only allocate
4513 * for the proper type.
4515 if (!block_group_bits(block_group, ffe_ctl->flags)) {
4516 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4517 BTRFS_BLOCK_GROUP_RAID1_MASK |
4518 BTRFS_BLOCK_GROUP_RAID56_MASK |
4519 BTRFS_BLOCK_GROUP_RAID10;
4522 * if they asked for extra copies and this block group
4523 * doesn't provide them, bail. This does allow us to
4524 * fill raid0 from raid1.
4526 if ((ffe_ctl->flags & extra) && !(block_group->flags & extra))
4530 * This block group has different flags than we want.
4531 * It's possible that we have MIXED_GROUP flag but no
4532 * block group is mixed. Just skip such block group.
4534 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4539 trace_find_free_extent_have_block_group(root, ffe_ctl, block_group);
4540 ffe_ctl->cached = btrfs_block_group_done(block_group);
4541 if (unlikely(!ffe_ctl->cached)) {
4542 ffe_ctl->have_caching_bg = true;
4543 ret = btrfs_cache_block_group(block_group, false);
4546 * If we get ENOMEM here or something else we want to
4547 * try other block groups, because it may not be fatal.
4548 * However if we can't find anything else we need to
4549 * save our return here so that we return the actual
4550 * error that caused problems, not ENOSPC.
4553 if (!cache_block_group_error)
4554 cache_block_group_error = ret;
4561 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) {
4562 if (!cache_block_group_error)
4563 cache_block_group_error = -EIO;
4567 if (!find_free_extent_check_size_class(ffe_ctl, block_group))
4571 ret = do_allocation(block_group, ffe_ctl, &bg_ret);
4575 if (bg_ret && bg_ret != block_group) {
4576 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4577 block_group = bg_ret;
4581 ffe_ctl->search_start = round_up(ffe_ctl->found_offset,
4582 fs_info->stripesize);
4584 /* move on to the next group */
4585 if (ffe_ctl->search_start + ffe_ctl->num_bytes >
4586 block_group->start + block_group->length) {
4587 btrfs_add_free_space_unused(block_group,
4588 ffe_ctl->found_offset,
4589 ffe_ctl->num_bytes);
4593 if (ffe_ctl->found_offset < ffe_ctl->search_start)
4594 btrfs_add_free_space_unused(block_group,
4595 ffe_ctl->found_offset,
4596 ffe_ctl->search_start - ffe_ctl->found_offset);
4598 ret = btrfs_add_reserved_bytes(block_group, ffe_ctl->ram_bytes,
4601 ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS);
4602 if (ret == -EAGAIN) {
4603 btrfs_add_free_space_unused(block_group,
4604 ffe_ctl->found_offset,
4605 ffe_ctl->num_bytes);
4608 btrfs_inc_block_group_reservations(block_group);
4610 /* we are all good, lets return */
4611 ins->objectid = ffe_ctl->search_start;
4612 ins->offset = ffe_ctl->num_bytes;
4614 trace_btrfs_reserve_extent(block_group, ffe_ctl);
4615 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4618 if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
4619 !ffe_ctl->retry_uncached) {
4620 ffe_ctl->retry_uncached = true;
4621 btrfs_wait_block_group_cache_progress(block_group,
4622 ffe_ctl->num_bytes +
4623 ffe_ctl->empty_cluster +
4624 ffe_ctl->empty_size);
4625 goto have_block_group;
4627 release_block_group(block_group, ffe_ctl, ffe_ctl->delalloc);
4630 up_read(&space_info->groups_sem);
4632 ret = find_free_extent_update_loop(fs_info, ins, ffe_ctl, full_search);
4636 if (ret == -ENOSPC && !cache_block_group_error) {
4638 * Use ffe_ctl->total_free_space as fallback if we can't find
4639 * any contiguous hole.
4641 if (!ffe_ctl->max_extent_size)
4642 ffe_ctl->max_extent_size = ffe_ctl->total_free_space;
4643 spin_lock(&space_info->lock);
4644 space_info->max_extent_size = ffe_ctl->max_extent_size;
4645 spin_unlock(&space_info->lock);
4646 ins->offset = ffe_ctl->max_extent_size;
4647 } else if (ret == -ENOSPC) {
4648 ret = cache_block_group_error;
4654 * Entry point to the extent allocator. Tries to find a hole that is at least
4655 * as big as @num_bytes.
4657 * @root - The root that will contain this extent
4659 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4660 * is used for accounting purposes. This value differs
4661 * from @num_bytes only in the case of compressed extents.
4663 * @num_bytes - Number of bytes to allocate on-disk.
4665 * @min_alloc_size - Indicates the minimum amount of space that the
4666 * allocator should try to satisfy. In some cases
4667 * @num_bytes may be larger than what is required and if
4668 * the filesystem is fragmented then allocation fails.
4669 * However, the presence of @min_alloc_size gives a
4670 * chance to try and satisfy the smaller allocation.
4672 * @empty_size - A hint that you plan on doing more COW. This is the
4673 * size in bytes the allocator should try to find free
4674 * next to the block it returns. This is just a hint and
4675 * may be ignored by the allocator.
4677 * @hint_byte - Hint to the allocator to start searching above the byte
4678 * address passed. It might be ignored.
4680 * @ins - This key is modified to record the found hole. It will
4681 * have the following values:
4682 * ins->objectid == start position
4683 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4684 * ins->offset == the size of the hole.
4686 * @is_data - Boolean flag indicating whether an extent is
4687 * allocated for data (true) or metadata (false)
4689 * @delalloc - Boolean flag indicating whether this allocation is for
4690 * delalloc or not. If 'true' data_rwsem of block groups
4691 * is going to be acquired.
4694 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4695 * case -ENOSPC is returned then @ins->offset will contain the size of the
4696 * largest available hole the allocator managed to find.
4698 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4699 u64 num_bytes, u64 min_alloc_size,
4700 u64 empty_size, u64 hint_byte,
4701 struct btrfs_key *ins, int is_data, int delalloc)
4703 struct btrfs_fs_info *fs_info = root->fs_info;
4704 struct find_free_extent_ctl ffe_ctl = {};
4705 bool final_tried = num_bytes == min_alloc_size;
4708 bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4709 bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data);
4711 flags = get_alloc_profile_by_root(root, is_data);
4713 WARN_ON(num_bytes < fs_info->sectorsize);
4715 ffe_ctl.ram_bytes = ram_bytes;
4716 ffe_ctl.num_bytes = num_bytes;
4717 ffe_ctl.min_alloc_size = min_alloc_size;
4718 ffe_ctl.empty_size = empty_size;
4719 ffe_ctl.flags = flags;
4720 ffe_ctl.delalloc = delalloc;
4721 ffe_ctl.hint_byte = hint_byte;
4722 ffe_ctl.for_treelog = for_treelog;
4723 ffe_ctl.for_data_reloc = for_data_reloc;
4725 ret = find_free_extent(root, ins, &ffe_ctl);
4726 if (!ret && !is_data) {
4727 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4728 } else if (ret == -ENOSPC) {
4729 if (!final_tried && ins->offset) {
4730 num_bytes = min(num_bytes >> 1, ins->offset);
4731 num_bytes = round_down(num_bytes,
4732 fs_info->sectorsize);
4733 num_bytes = max(num_bytes, min_alloc_size);
4734 ram_bytes = num_bytes;
4735 if (num_bytes == min_alloc_size)
4738 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4739 struct btrfs_space_info *sinfo;
4741 sinfo = btrfs_find_space_info(fs_info, flags);
4743 "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d",
4744 flags, num_bytes, for_treelog, for_data_reloc);
4746 btrfs_dump_space_info(fs_info, sinfo,
4754 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4755 u64 start, u64 len, int delalloc)
4757 struct btrfs_block_group *cache;
4759 cache = btrfs_lookup_block_group(fs_info, start);
4761 btrfs_err(fs_info, "Unable to find block group for %llu",
4766 btrfs_add_free_space(cache, start, len);
4767 btrfs_free_reserved_bytes(cache, len, delalloc);
4768 trace_btrfs_reserved_extent_free(fs_info, start, len);
4770 btrfs_put_block_group(cache);
4774 int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans,
4775 const struct extent_buffer *eb)
4777 struct btrfs_block_group *cache;
4780 cache = btrfs_lookup_block_group(trans->fs_info, eb->start);
4782 btrfs_err(trans->fs_info, "unable to find block group for %llu",
4787 ret = pin_down_extent(trans, cache, eb->start, eb->len, 1);
4788 btrfs_put_block_group(cache);
4792 static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr,
4795 struct btrfs_fs_info *fs_info = trans->fs_info;
4798 ret = remove_from_free_space_tree(trans, bytenr, num_bytes);
4802 ret = btrfs_update_block_group(trans, bytenr, num_bytes, true);
4805 btrfs_err(fs_info, "update block group failed for %llu %llu",
4810 trace_btrfs_reserved_extent_alloc(fs_info, bytenr, num_bytes);
4814 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4815 u64 parent, u64 root_objectid,
4816 u64 flags, u64 owner, u64 offset,
4817 struct btrfs_key *ins, int ref_mod, u64 oref_root)
4819 struct btrfs_fs_info *fs_info = trans->fs_info;
4820 struct btrfs_root *extent_root;
4822 struct btrfs_extent_item *extent_item;
4823 struct btrfs_extent_owner_ref *oref;
4824 struct btrfs_extent_inline_ref *iref;
4825 struct btrfs_path *path;
4826 struct extent_buffer *leaf;
4829 const bool simple_quota = (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE);
4832 type = BTRFS_SHARED_DATA_REF_KEY;
4834 type = BTRFS_EXTENT_DATA_REF_KEY;
4836 size = sizeof(*extent_item);
4838 size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY);
4839 size += btrfs_extent_inline_ref_size(type);
4841 path = btrfs_alloc_path();
4845 extent_root = btrfs_extent_root(fs_info, ins->objectid);
4846 ret = btrfs_insert_empty_item(trans, extent_root, path, ins, size);
4848 btrfs_free_path(path);
4852 leaf = path->nodes[0];
4853 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4854 struct btrfs_extent_item);
4855 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4856 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4857 btrfs_set_extent_flags(leaf, extent_item,
4858 flags | BTRFS_EXTENT_FLAG_DATA);
4860 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4862 btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_EXTENT_OWNER_REF_KEY);
4863 oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
4864 btrfs_set_extent_owner_ref_root_id(leaf, oref, oref_root);
4865 iref = (struct btrfs_extent_inline_ref *)(oref + 1);
4867 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4870 struct btrfs_shared_data_ref *ref;
4871 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4872 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4873 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4875 struct btrfs_extent_data_ref *ref;
4876 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4877 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4878 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4879 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4880 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4883 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
4884 btrfs_free_path(path);
4886 return alloc_reserved_extent(trans, ins->objectid, ins->offset);
4889 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4890 struct btrfs_delayed_ref_node *node,
4891 struct btrfs_delayed_extent_op *extent_op)
4893 struct btrfs_fs_info *fs_info = trans->fs_info;
4894 struct btrfs_root *extent_root;
4896 struct btrfs_extent_item *extent_item;
4897 struct btrfs_key extent_key;
4898 struct btrfs_tree_block_info *block_info;
4899 struct btrfs_extent_inline_ref *iref;
4900 struct btrfs_path *path;
4901 struct extent_buffer *leaf;
4902 struct btrfs_delayed_tree_ref *ref;
4903 u32 size = sizeof(*extent_item) + sizeof(*iref);
4904 u64 flags = extent_op->flags_to_set;
4905 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4907 ref = btrfs_delayed_node_to_tree_ref(node);
4909 extent_key.objectid = node->bytenr;
4910 if (skinny_metadata) {
4911 extent_key.offset = ref->level;
4912 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4914 extent_key.offset = node->num_bytes;
4915 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4916 size += sizeof(*block_info);
4919 path = btrfs_alloc_path();
4923 extent_root = btrfs_extent_root(fs_info, extent_key.objectid);
4924 ret = btrfs_insert_empty_item(trans, extent_root, path, &extent_key,
4927 btrfs_free_path(path);
4931 leaf = path->nodes[0];
4932 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4933 struct btrfs_extent_item);
4934 btrfs_set_extent_refs(leaf, extent_item, 1);
4935 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4936 btrfs_set_extent_flags(leaf, extent_item,
4937 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4939 if (skinny_metadata) {
4940 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4942 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4943 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4944 btrfs_set_tree_block_level(leaf, block_info, ref->level);
4945 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4948 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4949 btrfs_set_extent_inline_ref_type(leaf, iref,
4950 BTRFS_SHARED_BLOCK_REF_KEY);
4951 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4953 btrfs_set_extent_inline_ref_type(leaf, iref,
4954 BTRFS_TREE_BLOCK_REF_KEY);
4955 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4958 btrfs_mark_buffer_dirty(trans, leaf);
4959 btrfs_free_path(path);
4961 return alloc_reserved_extent(trans, node->bytenr, fs_info->nodesize);
4964 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4965 struct btrfs_root *root, u64 owner,
4966 u64 offset, u64 ram_bytes,
4967 struct btrfs_key *ins)
4969 struct btrfs_ref generic_ref = { 0 };
4970 u64 root_objectid = root->root_key.objectid;
4971 u64 owning_root = root_objectid;
4973 ASSERT(root_objectid != BTRFS_TREE_LOG_OBJECTID);
4975 if (btrfs_is_data_reloc_root(root) && is_fstree(root->relocation_src_root))
4976 owning_root = root->relocation_src_root;
4978 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4979 ins->objectid, ins->offset, 0, owning_root);
4980 btrfs_init_data_ref(&generic_ref, root_objectid, owner,
4982 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4984 return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes);
4988 * this is used by the tree logging recovery code. It records that
4989 * an extent has been allocated and makes sure to clear the free
4990 * space cache bits as well
4992 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4993 u64 root_objectid, u64 owner, u64 offset,
4994 struct btrfs_key *ins)
4996 struct btrfs_fs_info *fs_info = trans->fs_info;
4998 struct btrfs_block_group *block_group;
4999 struct btrfs_space_info *space_info;
5000 struct btrfs_squota_delta delta = {
5001 .root = root_objectid,
5002 .num_bytes = ins->offset,
5003 .generation = trans->transid,
5009 * Mixed block groups will exclude before processing the log so we only
5010 * need to do the exclude dance if this fs isn't mixed.
5012 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
5013 ret = __exclude_logged_extent(fs_info, ins->objectid,
5019 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
5023 space_info = block_group->space_info;
5024 spin_lock(&space_info->lock);
5025 spin_lock(&block_group->lock);
5026 space_info->bytes_reserved += ins->offset;
5027 block_group->reserved += ins->offset;
5028 spin_unlock(&block_group->lock);
5029 spin_unlock(&space_info->lock);
5031 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
5032 offset, ins, 1, root_objectid);
5034 btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
5035 ret = btrfs_record_squota_delta(fs_info, &delta);
5036 btrfs_put_block_group(block_group);
5040 #ifdef CONFIG_BTRFS_DEBUG
5042 * Extra safety check in case the extent tree is corrupted and extent allocator
5043 * chooses to use a tree block which is already used and locked.
5045 static bool check_eb_lock_owner(const struct extent_buffer *eb)
5047 if (eb->lock_owner == current->pid) {
5048 btrfs_err_rl(eb->fs_info,
5049 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
5050 eb->start, btrfs_header_owner(eb), current->pid);
5056 static bool check_eb_lock_owner(struct extent_buffer *eb)
5062 static struct extent_buffer *
5063 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5064 u64 bytenr, int level, u64 owner,
5065 enum btrfs_lock_nesting nest)
5067 struct btrfs_fs_info *fs_info = root->fs_info;
5068 struct extent_buffer *buf;
5069 u64 lockdep_owner = owner;
5071 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level);
5075 if (check_eb_lock_owner(buf)) {
5076 free_extent_buffer(buf);
5077 return ERR_PTR(-EUCLEAN);
5081 * The reloc trees are just snapshots, so we need them to appear to be
5082 * just like any other fs tree WRT lockdep.
5084 * The exception however is in replace_path() in relocation, where we
5085 * hold the lock on the original fs root and then search for the reloc
5086 * root. At that point we need to make sure any reloc root buffers are
5087 * set to the BTRFS_TREE_RELOC_OBJECTID lockdep class in order to make
5090 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID &&
5091 !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state))
5092 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
5094 /* btrfs_clear_buffer_dirty() accesses generation field. */
5095 btrfs_set_header_generation(buf, trans->transid);
5098 * This needs to stay, because we could allocate a freed block from an
5099 * old tree into a new tree, so we need to make sure this new block is
5100 * set to the appropriate level and owner.
5102 btrfs_set_buffer_lockdep_class(lockdep_owner, buf, level);
5104 __btrfs_tree_lock(buf, nest);
5105 btrfs_clear_buffer_dirty(trans, buf);
5106 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
5107 clear_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &buf->bflags);
5109 set_extent_buffer_uptodate(buf);
5111 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
5112 btrfs_set_header_level(buf, level);
5113 btrfs_set_header_bytenr(buf, buf->start);
5114 btrfs_set_header_generation(buf, trans->transid);
5115 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
5116 btrfs_set_header_owner(buf, owner);
5117 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
5118 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
5119 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5120 buf->log_index = root->log_transid % 2;
5122 * we allow two log transactions at a time, use different
5123 * EXTENT bit to differentiate dirty pages.
5125 if (buf->log_index == 0)
5126 set_extent_bit(&root->dirty_log_pages, buf->start,
5127 buf->start + buf->len - 1,
5128 EXTENT_DIRTY, NULL);
5130 set_extent_bit(&root->dirty_log_pages, buf->start,
5131 buf->start + buf->len - 1,
5134 buf->log_index = -1;
5135 set_extent_bit(&trans->transaction->dirty_pages, buf->start,
5136 buf->start + buf->len - 1, EXTENT_DIRTY, NULL);
5138 /* this returns a buffer locked for blocking */
5143 * finds a free extent and does all the dirty work required for allocation
5144 * returns the tree buffer or an ERR_PTR on error.
5146 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
5147 struct btrfs_root *root,
5148 u64 parent, u64 root_objectid,
5149 const struct btrfs_disk_key *key,
5150 int level, u64 hint,
5153 enum btrfs_lock_nesting nest)
5155 struct btrfs_fs_info *fs_info = root->fs_info;
5156 struct btrfs_key ins;
5157 struct btrfs_block_rsv *block_rsv;
5158 struct extent_buffer *buf;
5159 struct btrfs_delayed_extent_op *extent_op;
5160 struct btrfs_ref generic_ref = { 0 };
5163 u32 blocksize = fs_info->nodesize;
5164 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
5167 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5168 if (btrfs_is_testing(fs_info)) {
5169 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
5170 level, root_objectid, nest);
5172 root->alloc_bytenr += blocksize;
5177 block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
5178 if (IS_ERR(block_rsv))
5179 return ERR_CAST(block_rsv);
5181 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
5182 empty_size, hint, &ins, 0, 0);
5186 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
5187 root_objectid, nest);
5190 goto out_free_reserved;
5192 owning_root = btrfs_header_owner(buf);
5194 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5196 parent = ins.objectid;
5197 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5198 owning_root = reloc_src_root;
5202 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5203 extent_op = btrfs_alloc_delayed_extent_op();
5209 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5211 memset(&extent_op->key, 0, sizeof(extent_op->key));
5212 extent_op->flags_to_set = flags;
5213 extent_op->update_key = skinny_metadata ? false : true;
5214 extent_op->update_flags = true;
5215 extent_op->level = level;
5217 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
5218 ins.objectid, ins.offset, parent, owning_root);
5219 btrfs_init_tree_ref(&generic_ref, level, root_objectid,
5220 root->root_key.objectid, false);
5221 btrfs_ref_tree_mod(fs_info, &generic_ref);
5222 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op);
5224 goto out_free_delayed;
5229 btrfs_free_delayed_extent_op(extent_op);
5231 btrfs_tree_unlock(buf);
5232 free_extent_buffer(buf);
5234 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
5236 btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
5237 return ERR_PTR(ret);
5240 struct walk_control {
5241 u64 refs[BTRFS_MAX_LEVEL];
5242 u64 flags[BTRFS_MAX_LEVEL];
5243 struct btrfs_key update_progress;
5244 struct btrfs_key drop_progress;
5256 #define DROP_REFERENCE 1
5257 #define UPDATE_BACKREF 2
5259 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5260 struct btrfs_root *root,
5261 struct walk_control *wc,
5262 struct btrfs_path *path)
5264 struct btrfs_fs_info *fs_info = root->fs_info;
5270 struct btrfs_key key;
5271 struct extent_buffer *eb;
5276 if (path->slots[wc->level] < wc->reada_slot) {
5277 wc->reada_count = wc->reada_count * 2 / 3;
5278 wc->reada_count = max(wc->reada_count, 2);
5280 wc->reada_count = wc->reada_count * 3 / 2;
5281 wc->reada_count = min_t(int, wc->reada_count,
5282 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
5285 eb = path->nodes[wc->level];
5286 nritems = btrfs_header_nritems(eb);
5288 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5289 if (nread >= wc->reada_count)
5293 bytenr = btrfs_node_blockptr(eb, slot);
5294 generation = btrfs_node_ptr_generation(eb, slot);
5296 if (slot == path->slots[wc->level])
5299 if (wc->stage == UPDATE_BACKREF &&
5300 generation <= root->root_key.offset)
5303 /* We don't lock the tree block, it's OK to be racy here */
5304 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
5305 wc->level - 1, 1, &refs,
5307 /* We don't care about errors in readahead. */
5312 if (wc->stage == DROP_REFERENCE) {
5316 if (wc->level == 1 &&
5317 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5319 if (!wc->update_ref ||
5320 generation <= root->root_key.offset)
5322 btrfs_node_key_to_cpu(eb, &key, slot);
5323 ret = btrfs_comp_cpu_keys(&key,
5324 &wc->update_progress);
5328 if (wc->level == 1 &&
5329 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5333 btrfs_readahead_node_child(eb, slot);
5336 wc->reada_slot = slot;
5340 * helper to process tree block while walking down the tree.
5342 * when wc->stage == UPDATE_BACKREF, this function updates
5343 * back refs for pointers in the block.
5345 * NOTE: return value 1 means we should stop walking down.
5347 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5348 struct btrfs_root *root,
5349 struct btrfs_path *path,
5350 struct walk_control *wc, int lookup_info)
5352 struct btrfs_fs_info *fs_info = root->fs_info;
5353 int level = wc->level;
5354 struct extent_buffer *eb = path->nodes[level];
5355 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5358 if (wc->stage == UPDATE_BACKREF &&
5359 btrfs_header_owner(eb) != root->root_key.objectid)
5363 * when reference count of tree block is 1, it won't increase
5364 * again. once full backref flag is set, we never clear it.
5367 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5368 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5369 BUG_ON(!path->locks[level]);
5370 ret = btrfs_lookup_extent_info(trans, fs_info,
5371 eb->start, level, 1,
5375 BUG_ON(ret == -ENOMEM);
5378 BUG_ON(wc->refs[level] == 0);
5381 if (wc->stage == DROP_REFERENCE) {
5382 if (wc->refs[level] > 1)
5385 if (path->locks[level] && !wc->keep_locks) {
5386 btrfs_tree_unlock_rw(eb, path->locks[level]);
5387 path->locks[level] = 0;
5392 /* wc->stage == UPDATE_BACKREF */
5393 if (!(wc->flags[level] & flag)) {
5394 BUG_ON(!path->locks[level]);
5395 ret = btrfs_inc_ref(trans, root, eb, 1);
5396 BUG_ON(ret); /* -ENOMEM */
5397 ret = btrfs_dec_ref(trans, root, eb, 0);
5398 BUG_ON(ret); /* -ENOMEM */
5399 ret = btrfs_set_disk_extent_flags(trans, eb, flag);
5400 BUG_ON(ret); /* -ENOMEM */
5401 wc->flags[level] |= flag;
5405 * the block is shared by multiple trees, so it's not good to
5406 * keep the tree lock
5408 if (path->locks[level] && level > 0) {
5409 btrfs_tree_unlock_rw(eb, path->locks[level]);
5410 path->locks[level] = 0;
5416 * This is used to verify a ref exists for this root to deal with a bug where we
5417 * would have a drop_progress key that hadn't been updated properly.
5419 static int check_ref_exists(struct btrfs_trans_handle *trans,
5420 struct btrfs_root *root, u64 bytenr, u64 parent,
5423 struct btrfs_path *path;
5424 struct btrfs_extent_inline_ref *iref;
5427 path = btrfs_alloc_path();
5431 ret = lookup_extent_backref(trans, path, &iref, bytenr,
5432 root->fs_info->nodesize, parent,
5433 root->root_key.objectid, level, 0);
5434 btrfs_free_path(path);
5443 * helper to process tree block pointer.
5445 * when wc->stage == DROP_REFERENCE, this function checks
5446 * reference count of the block pointed to. if the block
5447 * is shared and we need update back refs for the subtree
5448 * rooted at the block, this function changes wc->stage to
5449 * UPDATE_BACKREF. if the block is shared and there is no
5450 * need to update back, this function drops the reference
5453 * NOTE: return value 1 means we should stop walking down.
5455 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5456 struct btrfs_root *root,
5457 struct btrfs_path *path,
5458 struct walk_control *wc, int *lookup_info)
5460 struct btrfs_fs_info *fs_info = root->fs_info;
5465 struct btrfs_tree_parent_check check = { 0 };
5466 struct btrfs_key key;
5467 struct btrfs_ref ref = { 0 };
5468 struct extent_buffer *next;
5469 int level = wc->level;
5472 bool need_account = false;
5474 generation = btrfs_node_ptr_generation(path->nodes[level],
5475 path->slots[level]);
5477 * if the lower level block was created before the snapshot
5478 * was created, we know there is no need to update back refs
5481 if (wc->stage == UPDATE_BACKREF &&
5482 generation <= root->root_key.offset) {
5487 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5489 check.level = level - 1;
5490 check.transid = generation;
5491 check.owner_root = root->root_key.objectid;
5492 check.has_first_key = true;
5493 btrfs_node_key_to_cpu(path->nodes[level], &check.first_key,
5494 path->slots[level]);
5496 next = find_extent_buffer(fs_info, bytenr);
5498 next = btrfs_find_create_tree_block(fs_info, bytenr,
5499 root->root_key.objectid, level - 1);
5501 return PTR_ERR(next);
5504 btrfs_tree_lock(next);
5506 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
5507 &wc->refs[level - 1],
5508 &wc->flags[level - 1],
5513 if (unlikely(wc->refs[level - 1] == 0)) {
5514 btrfs_err(fs_info, "Missing references.");
5520 if (wc->stage == DROP_REFERENCE) {
5521 if (wc->refs[level - 1] > 1) {
5522 need_account = true;
5524 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5527 if (!wc->update_ref ||
5528 generation <= root->root_key.offset)
5531 btrfs_node_key_to_cpu(path->nodes[level], &key,
5532 path->slots[level]);
5533 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5537 wc->stage = UPDATE_BACKREF;
5538 wc->shared_level = level - 1;
5542 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5546 if (!btrfs_buffer_uptodate(next, generation, 0)) {
5547 btrfs_tree_unlock(next);
5548 free_extent_buffer(next);
5554 if (reada && level == 1)
5555 reada_walk_down(trans, root, wc, path);
5556 next = read_tree_block(fs_info, bytenr, &check);
5558 return PTR_ERR(next);
5559 } else if (!extent_buffer_uptodate(next)) {
5560 free_extent_buffer(next);
5563 btrfs_tree_lock(next);
5567 ASSERT(level == btrfs_header_level(next));
5568 if (level != btrfs_header_level(next)) {
5569 btrfs_err(root->fs_info, "mismatched level");
5573 path->nodes[level] = next;
5574 path->slots[level] = 0;
5575 path->locks[level] = BTRFS_WRITE_LOCK;
5581 wc->refs[level - 1] = 0;
5582 wc->flags[level - 1] = 0;
5583 if (wc->stage == DROP_REFERENCE) {
5584 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5585 parent = path->nodes[level]->start;
5587 ASSERT(root->root_key.objectid ==
5588 btrfs_header_owner(path->nodes[level]));
5589 if (root->root_key.objectid !=
5590 btrfs_header_owner(path->nodes[level])) {
5591 btrfs_err(root->fs_info,
5592 "mismatched block owner");
5600 * If we had a drop_progress we need to verify the refs are set
5601 * as expected. If we find our ref then we know that from here
5602 * on out everything should be correct, and we can clear the
5605 if (wc->restarted) {
5606 ret = check_ref_exists(trans, root, bytenr, parent,
5617 * Reloc tree doesn't contribute to qgroup numbers, and we have
5618 * already accounted them at merge time (replace_path),
5619 * thus we could skip expensive subtree trace here.
5621 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
5623 ret = btrfs_qgroup_trace_subtree(trans, next,
5624 generation, level - 1);
5626 btrfs_err_rl(fs_info,
5627 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5633 * We need to update the next key in our walk control so we can
5634 * update the drop_progress key accordingly. We don't care if
5635 * find_next_key doesn't find a key because that means we're at
5636 * the end and are going to clean up now.
5638 wc->drop_level = level;
5639 find_next_key(path, level, &wc->drop_progress);
5641 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
5642 fs_info->nodesize, parent, owner_root);
5643 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid,
5645 ret = btrfs_free_extent(trans, &ref);
5654 btrfs_tree_unlock(next);
5655 free_extent_buffer(next);
5661 * helper to process tree block while walking up the tree.
5663 * when wc->stage == DROP_REFERENCE, this function drops
5664 * reference count on the block.
5666 * when wc->stage == UPDATE_BACKREF, this function changes
5667 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5668 * to UPDATE_BACKREF previously while processing the block.
5670 * NOTE: return value 1 means we should stop walking up.
5672 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5673 struct btrfs_root *root,
5674 struct btrfs_path *path,
5675 struct walk_control *wc)
5677 struct btrfs_fs_info *fs_info = root->fs_info;
5679 int level = wc->level;
5680 struct extent_buffer *eb = path->nodes[level];
5683 if (wc->stage == UPDATE_BACKREF) {
5684 BUG_ON(wc->shared_level < level);
5685 if (level < wc->shared_level)
5688 ret = find_next_key(path, level + 1, &wc->update_progress);
5692 wc->stage = DROP_REFERENCE;
5693 wc->shared_level = -1;
5694 path->slots[level] = 0;
5697 * check reference count again if the block isn't locked.
5698 * we should start walking down the tree again if reference
5701 if (!path->locks[level]) {
5703 btrfs_tree_lock(eb);
5704 path->locks[level] = BTRFS_WRITE_LOCK;
5706 ret = btrfs_lookup_extent_info(trans, fs_info,
5707 eb->start, level, 1,
5712 btrfs_tree_unlock_rw(eb, path->locks[level]);
5713 path->locks[level] = 0;
5716 BUG_ON(wc->refs[level] == 0);
5717 if (wc->refs[level] == 1) {
5718 btrfs_tree_unlock_rw(eb, path->locks[level]);
5719 path->locks[level] = 0;
5725 /* wc->stage == DROP_REFERENCE */
5726 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5728 if (wc->refs[level] == 1) {
5730 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5731 ret = btrfs_dec_ref(trans, root, eb, 1);
5733 ret = btrfs_dec_ref(trans, root, eb, 0);
5734 BUG_ON(ret); /* -ENOMEM */
5735 if (is_fstree(root->root_key.objectid)) {
5736 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5738 btrfs_err_rl(fs_info,
5739 "error %d accounting leaf items, quota is out of sync, rescan required",
5744 /* Make block locked assertion in btrfs_clear_buffer_dirty happy. */
5745 if (!path->locks[level]) {
5746 btrfs_tree_lock(eb);
5747 path->locks[level] = BTRFS_WRITE_LOCK;
5749 btrfs_clear_buffer_dirty(trans, eb);
5752 if (eb == root->node) {
5753 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5755 else if (root->root_key.objectid != btrfs_header_owner(eb))
5756 goto owner_mismatch;
5758 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5759 parent = path->nodes[level + 1]->start;
5760 else if (root->root_key.objectid !=
5761 btrfs_header_owner(path->nodes[level + 1]))
5762 goto owner_mismatch;
5765 btrfs_free_tree_block(trans, btrfs_root_id(root), eb, parent,
5766 wc->refs[level] == 1);
5768 wc->refs[level] = 0;
5769 wc->flags[level] = 0;
5773 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5774 btrfs_header_owner(eb), root->root_key.objectid);
5778 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5779 struct btrfs_root *root,
5780 struct btrfs_path *path,
5781 struct walk_control *wc)
5783 int level = wc->level;
5784 int lookup_info = 1;
5787 while (level >= 0) {
5788 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5795 if (path->slots[level] >=
5796 btrfs_header_nritems(path->nodes[level]))
5799 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5801 path->slots[level]++;
5807 return (ret == 1) ? 0 : ret;
5810 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5811 struct btrfs_root *root,
5812 struct btrfs_path *path,
5813 struct walk_control *wc, int max_level)
5815 int level = wc->level;
5818 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5819 while (level < max_level && path->nodes[level]) {
5821 if (path->slots[level] + 1 <
5822 btrfs_header_nritems(path->nodes[level])) {
5823 path->slots[level]++;
5826 ret = walk_up_proc(trans, root, path, wc);
5832 if (path->locks[level]) {
5833 btrfs_tree_unlock_rw(path->nodes[level],
5834 path->locks[level]);
5835 path->locks[level] = 0;
5837 free_extent_buffer(path->nodes[level]);
5838 path->nodes[level] = NULL;
5846 * drop a subvolume tree.
5848 * this function traverses the tree freeing any blocks that only
5849 * referenced by the tree.
5851 * when a shared tree block is found. this function decreases its
5852 * reference count by one. if update_ref is true, this function
5853 * also make sure backrefs for the shared block and all lower level
5854 * blocks are properly updated.
5856 * If called with for_reloc == 0, may exit early with -EAGAIN
5858 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
5860 const bool is_reloc_root = (root->root_key.objectid ==
5861 BTRFS_TREE_RELOC_OBJECTID);
5862 struct btrfs_fs_info *fs_info = root->fs_info;
5863 struct btrfs_path *path;
5864 struct btrfs_trans_handle *trans;
5865 struct btrfs_root *tree_root = fs_info->tree_root;
5866 struct btrfs_root_item *root_item = &root->root_item;
5867 struct walk_control *wc;
5868 struct btrfs_key key;
5872 bool root_dropped = false;
5873 bool unfinished_drop = false;
5875 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5877 path = btrfs_alloc_path();
5883 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5885 btrfs_free_path(path);
5891 * Use join to avoid potential EINTR from transaction start. See
5892 * wait_reserve_ticket and the whole reservation callchain.
5895 trans = btrfs_join_transaction(tree_root);
5897 trans = btrfs_start_transaction(tree_root, 0);
5898 if (IS_ERR(trans)) {
5899 err = PTR_ERR(trans);
5903 err = btrfs_run_delayed_items(trans);
5908 * This will help us catch people modifying the fs tree while we're
5909 * dropping it. It is unsafe to mess with the fs tree while it's being
5910 * dropped as we unlock the root node and parent nodes as we walk down
5911 * the tree, assuming nothing will change. If something does change
5912 * then we'll have stale information and drop references to blocks we've
5915 set_bit(BTRFS_ROOT_DELETING, &root->state);
5916 unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
5918 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5919 level = btrfs_header_level(root->node);
5920 path->nodes[level] = btrfs_lock_root_node(root);
5921 path->slots[level] = 0;
5922 path->locks[level] = BTRFS_WRITE_LOCK;
5923 memset(&wc->update_progress, 0,
5924 sizeof(wc->update_progress));
5926 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5927 memcpy(&wc->update_progress, &key,
5928 sizeof(wc->update_progress));
5930 level = btrfs_root_drop_level(root_item);
5932 path->lowest_level = level;
5933 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5934 path->lowest_level = 0;
5942 * unlock our path, this is safe because only this
5943 * function is allowed to delete this snapshot
5945 btrfs_unlock_up_safe(path, 0);
5947 level = btrfs_header_level(root->node);
5949 btrfs_tree_lock(path->nodes[level]);
5950 path->locks[level] = BTRFS_WRITE_LOCK;
5952 ret = btrfs_lookup_extent_info(trans, fs_info,
5953 path->nodes[level]->start,
5954 level, 1, &wc->refs[level],
5955 &wc->flags[level], NULL);
5960 BUG_ON(wc->refs[level] == 0);
5962 if (level == btrfs_root_drop_level(root_item))
5965 btrfs_tree_unlock(path->nodes[level]);
5966 path->locks[level] = 0;
5967 WARN_ON(wc->refs[level] != 1);
5972 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5974 wc->shared_level = -1;
5975 wc->stage = DROP_REFERENCE;
5976 wc->update_ref = update_ref;
5978 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5982 ret = walk_down_tree(trans, root, path, wc);
5984 btrfs_abort_transaction(trans, ret);
5989 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5991 btrfs_abort_transaction(trans, ret);
5997 BUG_ON(wc->stage != DROP_REFERENCE);
6001 if (wc->stage == DROP_REFERENCE) {
6002 wc->drop_level = wc->level;
6003 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
6005 path->slots[wc->drop_level]);
6007 btrfs_cpu_key_to_disk(&root_item->drop_progress,
6008 &wc->drop_progress);
6009 btrfs_set_root_drop_level(root_item, wc->drop_level);
6011 BUG_ON(wc->level == 0);
6012 if (btrfs_should_end_transaction(trans) ||
6013 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
6014 ret = btrfs_update_root(trans, tree_root,
6018 btrfs_abort_transaction(trans, ret);
6024 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
6026 btrfs_end_transaction_throttle(trans);
6027 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
6028 btrfs_debug(fs_info,
6029 "drop snapshot early exit");
6035 * Use join to avoid potential EINTR from transaction
6036 * start. See wait_reserve_ticket and the whole
6037 * reservation callchain.
6040 trans = btrfs_join_transaction(tree_root);
6042 trans = btrfs_start_transaction(tree_root, 0);
6043 if (IS_ERR(trans)) {
6044 err = PTR_ERR(trans);
6049 btrfs_release_path(path);
6053 ret = btrfs_del_root(trans, &root->root_key);
6055 btrfs_abort_transaction(trans, ret);
6060 if (!is_reloc_root) {
6061 ret = btrfs_find_root(tree_root, &root->root_key, path,
6064 btrfs_abort_transaction(trans, ret);
6067 } else if (ret > 0) {
6068 /* if we fail to delete the orphan item this time
6069 * around, it'll get picked up the next time.
6071 * The most common failure here is just -ENOENT.
6073 btrfs_del_orphan_item(trans, tree_root,
6074 root->root_key.objectid);
6079 * This subvolume is going to be completely dropped, and won't be
6080 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
6081 * commit transaction time. So free it here manually.
6083 btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
6084 btrfs_qgroup_free_meta_all_pertrans(root);
6086 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state))
6087 btrfs_add_dropped_root(trans, root);
6089 btrfs_put_root(root);
6090 root_dropped = true;
6093 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
6095 btrfs_end_transaction_throttle(trans);
6098 btrfs_free_path(path);
6101 * We were an unfinished drop root, check to see if there are any
6102 * pending, and if not clear and wake up any waiters.
6104 if (!err && unfinished_drop)
6105 btrfs_maybe_wake_unfinished_drop(fs_info);
6108 * So if we need to stop dropping the snapshot for whatever reason we
6109 * need to make sure to add it back to the dead root list so that we
6110 * keep trying to do the work later. This also cleans up roots if we
6111 * don't have it in the radix (like when we recover after a power fail
6112 * or unmount) so we don't leak memory.
6114 if (!for_reloc && !root_dropped)
6115 btrfs_add_dead_root(root);
6120 * drop subtree rooted at tree block 'node'.
6122 * NOTE: this function will unlock and release tree block 'node'
6123 * only used by relocation code
6125 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6126 struct btrfs_root *root,
6127 struct extent_buffer *node,
6128 struct extent_buffer *parent)
6130 struct btrfs_fs_info *fs_info = root->fs_info;
6131 struct btrfs_path *path;
6132 struct walk_control *wc;
6138 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6140 path = btrfs_alloc_path();
6144 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6146 btrfs_free_path(path);
6150 btrfs_assert_tree_write_locked(parent);
6151 parent_level = btrfs_header_level(parent);
6152 atomic_inc(&parent->refs);
6153 path->nodes[parent_level] = parent;
6154 path->slots[parent_level] = btrfs_header_nritems(parent);
6156 btrfs_assert_tree_write_locked(node);
6157 level = btrfs_header_level(node);
6158 path->nodes[level] = node;
6159 path->slots[level] = 0;
6160 path->locks[level] = BTRFS_WRITE_LOCK;
6162 wc->refs[parent_level] = 1;
6163 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6165 wc->shared_level = -1;
6166 wc->stage = DROP_REFERENCE;
6169 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
6172 wret = walk_down_tree(trans, root, path, wc);
6178 wret = walk_up_tree(trans, root, path, wc, parent_level);
6186 btrfs_free_path(path);
6191 * Unpin the extent range in an error context and don't add the space back.
6192 * Errors are not propagated further.
6194 void btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, u64 start, u64 end)
6196 unpin_extent_range(fs_info, start, end, false);
6200 * It used to be that old block groups would be left around forever.
6201 * Iterating over them would be enough to trim unused space. Since we
6202 * now automatically remove them, we also need to iterate over unallocated
6205 * We don't want a transaction for this since the discard may take a
6206 * substantial amount of time. We don't require that a transaction be
6207 * running, but we do need to take a running transaction into account
6208 * to ensure that we're not discarding chunks that were released or
6209 * allocated in the current transaction.
6211 * Holding the chunks lock will prevent other threads from allocating
6212 * or releasing chunks, but it won't prevent a running transaction
6213 * from committing and releasing the memory that the pending chunks
6214 * list head uses. For that, we need to take a reference to the
6215 * transaction and hold the commit root sem. We only need to hold
6216 * it while performing the free space search since we have already
6217 * held back allocations.
6219 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
6221 u64 start = BTRFS_DEVICE_RANGE_RESERVED, len = 0, end = 0;
6226 /* Discard not supported = nothing to do. */
6227 if (!bdev_max_discard_sectors(device->bdev))
6230 /* Not writable = nothing to do. */
6231 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
6234 /* No free space = nothing to do. */
6235 if (device->total_bytes <= device->bytes_used)
6241 struct btrfs_fs_info *fs_info = device->fs_info;
6244 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
6248 find_first_clear_extent_bit(&device->alloc_state, start,
6250 CHUNK_TRIMMED | CHUNK_ALLOCATED);
6252 /* Check if there are any CHUNK_* bits left */
6253 if (start > device->total_bytes) {
6254 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6255 btrfs_warn_in_rcu(fs_info,
6256 "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
6257 start, end - start + 1,
6258 btrfs_dev_name(device),
6259 device->total_bytes);
6260 mutex_unlock(&fs_info->chunk_mutex);
6265 /* Ensure we skip the reserved space on each device. */
6266 start = max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
6269 * If find_first_clear_extent_bit find a range that spans the
6270 * end of the device it will set end to -1, in this case it's up
6271 * to the caller to trim the value to the size of the device.
6273 end = min(end, device->total_bytes - 1);
6275 len = end - start + 1;
6277 /* We didn't find any extents */
6279 mutex_unlock(&fs_info->chunk_mutex);
6284 ret = btrfs_issue_discard(device->bdev, start, len,
6287 set_extent_bit(&device->alloc_state, start,
6288 start + bytes - 1, CHUNK_TRIMMED, NULL);
6289 mutex_unlock(&fs_info->chunk_mutex);
6297 if (fatal_signal_pending(current)) {
6309 * Trim the whole filesystem by:
6310 * 1) trimming the free space in each block group
6311 * 2) trimming the unallocated space on each device
6313 * This will also continue trimming even if a block group or device encounters
6314 * an error. The return value will be the last error, or 0 if nothing bad
6317 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
6319 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6320 struct btrfs_block_group *cache = NULL;
6321 struct btrfs_device *device;
6323 u64 range_end = U64_MAX;
6333 if (range->start == U64_MAX)
6337 * Check range overflow if range->len is set.
6338 * The default range->len is U64_MAX.
6340 if (range->len != U64_MAX &&
6341 check_add_overflow(range->start, range->len, &range_end))
6344 cache = btrfs_lookup_first_block_group(fs_info, range->start);
6345 for (; cache; cache = btrfs_next_block_group(cache)) {
6346 if (cache->start >= range_end) {
6347 btrfs_put_block_group(cache);
6351 start = max(range->start, cache->start);
6352 end = min(range_end, cache->start + cache->length);
6354 if (end - start >= range->minlen) {
6355 if (!btrfs_block_group_done(cache)) {
6356 ret = btrfs_cache_block_group(cache, true);
6363 ret = btrfs_trim_block_group(cache,
6369 trimmed += group_trimmed;
6380 "failed to trim %llu block group(s), last error %d",
6383 mutex_lock(&fs_devices->device_list_mutex);
6384 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6385 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
6388 ret = btrfs_trim_free_extents(device, &group_trimmed);
6395 trimmed += group_trimmed;
6397 mutex_unlock(&fs_devices->device_list_mutex);
6401 "failed to trim %llu device(s), last error %d",
6402 dev_failed, dev_ret);
6403 range->len = trimmed;