1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/signal.h>
8 #include <linux/pagemap.h>
9 #include <linux/writeback.h>
10 #include <linux/blkdev.h>
11 #include <linux/sort.h>
12 #include <linux/rcupdate.h>
13 #include <linux/kthread.h>
14 #include <linux/slab.h>
15 #include <linux/ratelimit.h>
16 #include <linux/percpu_counter.h>
17 #include <linux/lockdep.h>
18 #include <linux/crc32c.h>
20 #include "extent-tree.h"
23 #include "print-tree.h"
27 #include "free-space-cache.h"
28 #include "free-space-tree.h"
31 #include "ref-verify.h"
32 #include "space-info.h"
33 #include "block-rsv.h"
34 #include "delalloc-space.h"
36 #include "rcu-string.h"
38 #include "dev-replace.h"
40 #include "accessors.h"
41 #include "root-tree.h"
42 #include "file-item.h"
44 #include "tree-checker.h"
46 #undef SCRAMBLE_DELAYED_REFS
49 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
50 struct btrfs_delayed_ref_node *node, u64 parent,
51 u64 root_objectid, u64 owner_objectid,
52 u64 owner_offset, int refs_to_drop,
53 struct btrfs_delayed_extent_op *extra_op);
54 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
55 struct extent_buffer *leaf,
56 struct btrfs_extent_item *ei);
57 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
58 u64 parent, u64 root_objectid,
59 u64 flags, u64 owner, u64 offset,
60 struct btrfs_key *ins, int ref_mod);
61 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
62 struct btrfs_delayed_ref_node *node,
63 struct btrfs_delayed_extent_op *extent_op);
64 static int find_next_key(struct btrfs_path *path, int level,
65 struct btrfs_key *key);
67 static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
69 return (cache->flags & bits) == bits;
72 /* simple helper to search for an existing data extent at a given offset */
73 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
75 struct btrfs_root *root = btrfs_extent_root(fs_info, start);
78 struct btrfs_path *path;
80 path = btrfs_alloc_path();
86 key.type = BTRFS_EXTENT_ITEM_KEY;
87 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
88 btrfs_free_path(path);
93 * helper function to lookup reference count and flags of a tree block.
95 * the head node for delayed ref is used to store the sum of all the
96 * reference count modifications queued up in the rbtree. the head
97 * node may also store the extent flags to set. This way you can check
98 * to see what the reference count and extent flags would be if all of
99 * the delayed refs are not processed.
101 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
102 struct btrfs_fs_info *fs_info, u64 bytenr,
103 u64 offset, int metadata, u64 *refs, u64 *flags)
105 struct btrfs_root *extent_root;
106 struct btrfs_delayed_ref_head *head;
107 struct btrfs_delayed_ref_root *delayed_refs;
108 struct btrfs_path *path;
109 struct btrfs_extent_item *ei;
110 struct extent_buffer *leaf;
111 struct btrfs_key key;
118 * If we don't have skinny metadata, don't bother doing anything
121 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
122 offset = fs_info->nodesize;
126 path = btrfs_alloc_path();
131 path->skip_locking = 1;
132 path->search_commit_root = 1;
136 key.objectid = bytenr;
139 key.type = BTRFS_METADATA_ITEM_KEY;
141 key.type = BTRFS_EXTENT_ITEM_KEY;
143 extent_root = btrfs_extent_root(fs_info, bytenr);
144 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
148 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
149 if (path->slots[0]) {
151 btrfs_item_key_to_cpu(path->nodes[0], &key,
153 if (key.objectid == bytenr &&
154 key.type == BTRFS_EXTENT_ITEM_KEY &&
155 key.offset == fs_info->nodesize)
161 leaf = path->nodes[0];
162 item_size = btrfs_item_size(leaf, path->slots[0]);
163 if (item_size >= sizeof(*ei)) {
164 ei = btrfs_item_ptr(leaf, path->slots[0],
165 struct btrfs_extent_item);
166 num_refs = btrfs_extent_refs(leaf, ei);
167 extent_flags = btrfs_extent_flags(leaf, ei);
171 "unexpected extent item size, has %u expect >= %zu",
172 item_size, sizeof(*ei));
174 btrfs_abort_transaction(trans, ret);
176 btrfs_handle_fs_error(fs_info, ret, NULL);
181 BUG_ON(num_refs == 0);
191 delayed_refs = &trans->transaction->delayed_refs;
192 spin_lock(&delayed_refs->lock);
193 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
195 if (!mutex_trylock(&head->mutex)) {
196 refcount_inc(&head->refs);
197 spin_unlock(&delayed_refs->lock);
199 btrfs_release_path(path);
202 * Mutex was contended, block until it's released and try
205 mutex_lock(&head->mutex);
206 mutex_unlock(&head->mutex);
207 btrfs_put_delayed_ref_head(head);
210 spin_lock(&head->lock);
211 if (head->extent_op && head->extent_op->update_flags)
212 extent_flags |= head->extent_op->flags_to_set;
214 BUG_ON(num_refs == 0);
216 num_refs += head->ref_mod;
217 spin_unlock(&head->lock);
218 mutex_unlock(&head->mutex);
220 spin_unlock(&delayed_refs->lock);
222 WARN_ON(num_refs == 0);
226 *flags = extent_flags;
228 btrfs_free_path(path);
233 * Back reference rules. Back refs have three main goals:
235 * 1) differentiate between all holders of references to an extent so that
236 * when a reference is dropped we can make sure it was a valid reference
237 * before freeing the extent.
239 * 2) Provide enough information to quickly find the holders of an extent
240 * if we notice a given block is corrupted or bad.
242 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
243 * maintenance. This is actually the same as #2, but with a slightly
244 * different use case.
246 * There are two kinds of back refs. The implicit back refs is optimized
247 * for pointers in non-shared tree blocks. For a given pointer in a block,
248 * back refs of this kind provide information about the block's owner tree
249 * and the pointer's key. These information allow us to find the block by
250 * b-tree searching. The full back refs is for pointers in tree blocks not
251 * referenced by their owner trees. The location of tree block is recorded
252 * in the back refs. Actually the full back refs is generic, and can be
253 * used in all cases the implicit back refs is used. The major shortcoming
254 * of the full back refs is its overhead. Every time a tree block gets
255 * COWed, we have to update back refs entry for all pointers in it.
257 * For a newly allocated tree block, we use implicit back refs for
258 * pointers in it. This means most tree related operations only involve
259 * implicit back refs. For a tree block created in old transaction, the
260 * only way to drop a reference to it is COW it. So we can detect the
261 * event that tree block loses its owner tree's reference and do the
262 * back refs conversion.
264 * When a tree block is COWed through a tree, there are four cases:
266 * The reference count of the block is one and the tree is the block's
267 * owner tree. Nothing to do in this case.
269 * The reference count of the block is one and the tree is not the
270 * block's owner tree. In this case, full back refs is used for pointers
271 * in the block. Remove these full back refs, add implicit back refs for
272 * every pointers in the new block.
274 * The reference count of the block is greater than one and the tree is
275 * the block's owner tree. In this case, implicit back refs is used for
276 * pointers in the block. Add full back refs for every pointers in the
277 * block, increase lower level extents' reference counts. The original
278 * implicit back refs are entailed to the new block.
280 * The reference count of the block is greater than one and the tree is
281 * not the block's owner tree. Add implicit back refs for every pointer in
282 * the new block, increase lower level extents' reference count.
284 * Back Reference Key composing:
286 * The key objectid corresponds to the first byte in the extent,
287 * The key type is used to differentiate between types of back refs.
288 * There are different meanings of the key offset for different types
291 * File extents can be referenced by:
293 * - multiple snapshots, subvolumes, or different generations in one subvol
294 * - different files inside a single subvolume
295 * - different offsets inside a file (bookend extents in file.c)
297 * The extent ref structure for the implicit back refs has fields for:
299 * - Objectid of the subvolume root
300 * - objectid of the file holding the reference
301 * - original offset in the file
302 * - how many bookend extents
304 * The key offset for the implicit back refs is hash of the first
307 * The extent ref structure for the full back refs has field for:
309 * - number of pointers in the tree leaf
311 * The key offset for the implicit back refs is the first byte of
314 * When a file extent is allocated, The implicit back refs is used.
315 * the fields are filled in:
317 * (root_key.objectid, inode objectid, offset in file, 1)
319 * When a file extent is removed file truncation, we find the
320 * corresponding implicit back refs and check the following fields:
322 * (btrfs_header_owner(leaf), inode objectid, offset in file)
324 * Btree extents can be referenced by:
326 * - Different subvolumes
328 * Both the implicit back refs and the full back refs for tree blocks
329 * only consist of key. The key offset for the implicit back refs is
330 * objectid of block's owner tree. The key offset for the full back refs
331 * is the first byte of parent block.
333 * When implicit back refs is used, information about the lowest key and
334 * level of the tree block are required. These information are stored in
335 * tree block info structure.
339 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
340 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
341 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
343 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
344 struct btrfs_extent_inline_ref *iref,
345 enum btrfs_inline_ref_type is_data)
347 int type = btrfs_extent_inline_ref_type(eb, iref);
348 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
350 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
351 type == BTRFS_SHARED_BLOCK_REF_KEY ||
352 type == BTRFS_SHARED_DATA_REF_KEY ||
353 type == BTRFS_EXTENT_DATA_REF_KEY) {
354 if (is_data == BTRFS_REF_TYPE_BLOCK) {
355 if (type == BTRFS_TREE_BLOCK_REF_KEY)
357 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
360 * Every shared one has parent tree block,
361 * which must be aligned to sector size.
364 IS_ALIGNED(offset, eb->fs_info->sectorsize))
367 } else if (is_data == BTRFS_REF_TYPE_DATA) {
368 if (type == BTRFS_EXTENT_DATA_REF_KEY)
370 if (type == BTRFS_SHARED_DATA_REF_KEY) {
373 * Every shared one has parent tree block,
374 * which must be aligned to sector size.
377 IS_ALIGNED(offset, eb->fs_info->sectorsize))
381 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
387 btrfs_print_leaf(eb);
388 btrfs_err(eb->fs_info,
389 "eb %llu iref 0x%lx invalid extent inline ref type %d",
390 eb->start, (unsigned long)iref, type);
392 return BTRFS_REF_TYPE_INVALID;
395 u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
397 u32 high_crc = ~(u32)0;
398 u32 low_crc = ~(u32)0;
401 lenum = cpu_to_le64(root_objectid);
402 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
403 lenum = cpu_to_le64(owner);
404 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
405 lenum = cpu_to_le64(offset);
406 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
408 return ((u64)high_crc << 31) ^ (u64)low_crc;
411 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
412 struct btrfs_extent_data_ref *ref)
414 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
415 btrfs_extent_data_ref_objectid(leaf, ref),
416 btrfs_extent_data_ref_offset(leaf, ref));
419 static int match_extent_data_ref(struct extent_buffer *leaf,
420 struct btrfs_extent_data_ref *ref,
421 u64 root_objectid, u64 owner, u64 offset)
423 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
424 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
425 btrfs_extent_data_ref_offset(leaf, ref) != offset)
430 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
431 struct btrfs_path *path,
432 u64 bytenr, u64 parent,
434 u64 owner, u64 offset)
436 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
437 struct btrfs_key key;
438 struct btrfs_extent_data_ref *ref;
439 struct extent_buffer *leaf;
445 key.objectid = bytenr;
447 key.type = BTRFS_SHARED_DATA_REF_KEY;
450 key.type = BTRFS_EXTENT_DATA_REF_KEY;
451 key.offset = hash_extent_data_ref(root_objectid,
456 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
468 leaf = path->nodes[0];
469 nritems = btrfs_header_nritems(leaf);
471 if (path->slots[0] >= nritems) {
472 ret = btrfs_next_leaf(root, path);
478 leaf = path->nodes[0];
479 nritems = btrfs_header_nritems(leaf);
483 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
484 if (key.objectid != bytenr ||
485 key.type != BTRFS_EXTENT_DATA_REF_KEY)
488 ref = btrfs_item_ptr(leaf, path->slots[0],
489 struct btrfs_extent_data_ref);
491 if (match_extent_data_ref(leaf, ref, root_objectid,
494 btrfs_release_path(path);
506 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
507 struct btrfs_path *path,
508 u64 bytenr, u64 parent,
509 u64 root_objectid, u64 owner,
510 u64 offset, int refs_to_add)
512 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
513 struct btrfs_key key;
514 struct extent_buffer *leaf;
519 key.objectid = bytenr;
521 key.type = BTRFS_SHARED_DATA_REF_KEY;
523 size = sizeof(struct btrfs_shared_data_ref);
525 key.type = BTRFS_EXTENT_DATA_REF_KEY;
526 key.offset = hash_extent_data_ref(root_objectid,
528 size = sizeof(struct btrfs_extent_data_ref);
531 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
532 if (ret && ret != -EEXIST)
535 leaf = path->nodes[0];
537 struct btrfs_shared_data_ref *ref;
538 ref = btrfs_item_ptr(leaf, path->slots[0],
539 struct btrfs_shared_data_ref);
541 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
543 num_refs = btrfs_shared_data_ref_count(leaf, ref);
544 num_refs += refs_to_add;
545 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
548 struct btrfs_extent_data_ref *ref;
549 while (ret == -EEXIST) {
550 ref = btrfs_item_ptr(leaf, path->slots[0],
551 struct btrfs_extent_data_ref);
552 if (match_extent_data_ref(leaf, ref, root_objectid,
555 btrfs_release_path(path);
557 ret = btrfs_insert_empty_item(trans, root, path, &key,
559 if (ret && ret != -EEXIST)
562 leaf = path->nodes[0];
564 ref = btrfs_item_ptr(leaf, path->slots[0],
565 struct btrfs_extent_data_ref);
567 btrfs_set_extent_data_ref_root(leaf, ref,
569 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
570 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
571 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
573 num_refs = btrfs_extent_data_ref_count(leaf, ref);
574 num_refs += refs_to_add;
575 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
578 btrfs_mark_buffer_dirty(trans, leaf);
581 btrfs_release_path(path);
585 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
586 struct btrfs_root *root,
587 struct btrfs_path *path,
590 struct btrfs_key key;
591 struct btrfs_extent_data_ref *ref1 = NULL;
592 struct btrfs_shared_data_ref *ref2 = NULL;
593 struct extent_buffer *leaf;
597 leaf = path->nodes[0];
598 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
600 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
601 ref1 = btrfs_item_ptr(leaf, path->slots[0],
602 struct btrfs_extent_data_ref);
603 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
604 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
605 ref2 = btrfs_item_ptr(leaf, path->slots[0],
606 struct btrfs_shared_data_ref);
607 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
609 btrfs_err(trans->fs_info,
610 "unrecognized backref key (%llu %u %llu)",
611 key.objectid, key.type, key.offset);
612 btrfs_abort_transaction(trans, -EUCLEAN);
616 BUG_ON(num_refs < refs_to_drop);
617 num_refs -= refs_to_drop;
620 ret = btrfs_del_item(trans, root, path);
622 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
623 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
624 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
625 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
626 btrfs_mark_buffer_dirty(trans, leaf);
631 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
632 struct btrfs_extent_inline_ref *iref)
634 struct btrfs_key key;
635 struct extent_buffer *leaf;
636 struct btrfs_extent_data_ref *ref1;
637 struct btrfs_shared_data_ref *ref2;
641 leaf = path->nodes[0];
642 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
646 * If type is invalid, we should have bailed out earlier than
649 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
650 ASSERT(type != BTRFS_REF_TYPE_INVALID);
651 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
652 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
653 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
655 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
656 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
658 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
659 ref1 = btrfs_item_ptr(leaf, path->slots[0],
660 struct btrfs_extent_data_ref);
661 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
662 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
663 ref2 = btrfs_item_ptr(leaf, path->slots[0],
664 struct btrfs_shared_data_ref);
665 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
672 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
673 struct btrfs_path *path,
674 u64 bytenr, u64 parent,
677 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
678 struct btrfs_key key;
681 key.objectid = bytenr;
683 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
686 key.type = BTRFS_TREE_BLOCK_REF_KEY;
687 key.offset = root_objectid;
690 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
696 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
697 struct btrfs_path *path,
698 u64 bytenr, u64 parent,
701 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
702 struct btrfs_key key;
705 key.objectid = bytenr;
707 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
710 key.type = BTRFS_TREE_BLOCK_REF_KEY;
711 key.offset = root_objectid;
714 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
715 btrfs_release_path(path);
719 static inline int extent_ref_type(u64 parent, u64 owner)
722 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
724 type = BTRFS_SHARED_BLOCK_REF_KEY;
726 type = BTRFS_TREE_BLOCK_REF_KEY;
729 type = BTRFS_SHARED_DATA_REF_KEY;
731 type = BTRFS_EXTENT_DATA_REF_KEY;
736 static int find_next_key(struct btrfs_path *path, int level,
737 struct btrfs_key *key)
740 for (; level < BTRFS_MAX_LEVEL; level++) {
741 if (!path->nodes[level])
743 if (path->slots[level] + 1 >=
744 btrfs_header_nritems(path->nodes[level]))
747 btrfs_item_key_to_cpu(path->nodes[level], key,
748 path->slots[level] + 1);
750 btrfs_node_key_to_cpu(path->nodes[level], key,
751 path->slots[level] + 1);
758 * look for inline back ref. if back ref is found, *ref_ret is set
759 * to the address of inline back ref, and 0 is returned.
761 * if back ref isn't found, *ref_ret is set to the address where it
762 * should be inserted, and -ENOENT is returned.
764 * if insert is true and there are too many inline back refs, the path
765 * points to the extent item, and -EAGAIN is returned.
767 * NOTE: inline back refs are ordered in the same way that back ref
768 * items in the tree are ordered.
770 static noinline_for_stack
771 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
772 struct btrfs_path *path,
773 struct btrfs_extent_inline_ref **ref_ret,
774 u64 bytenr, u64 num_bytes,
775 u64 parent, u64 root_objectid,
776 u64 owner, u64 offset, int insert)
778 struct btrfs_fs_info *fs_info = trans->fs_info;
779 struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr);
780 struct btrfs_key key;
781 struct extent_buffer *leaf;
782 struct btrfs_extent_item *ei;
783 struct btrfs_extent_inline_ref *iref;
793 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
796 key.objectid = bytenr;
797 key.type = BTRFS_EXTENT_ITEM_KEY;
798 key.offset = num_bytes;
800 want = extent_ref_type(parent, owner);
802 extra_size = btrfs_extent_inline_ref_size(want);
803 path->search_for_extension = 1;
804 path->keep_locks = 1;
809 * Owner is our level, so we can just add one to get the level for the
810 * block we are interested in.
812 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
813 key.type = BTRFS_METADATA_ITEM_KEY;
818 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
825 * We may be a newly converted file system which still has the old fat
826 * extent entries for metadata, so try and see if we have one of those.
828 if (ret > 0 && skinny_metadata) {
829 skinny_metadata = false;
830 if (path->slots[0]) {
832 btrfs_item_key_to_cpu(path->nodes[0], &key,
834 if (key.objectid == bytenr &&
835 key.type == BTRFS_EXTENT_ITEM_KEY &&
836 key.offset == num_bytes)
840 key.objectid = bytenr;
841 key.type = BTRFS_EXTENT_ITEM_KEY;
842 key.offset = num_bytes;
843 btrfs_release_path(path);
848 if (ret && !insert) {
851 } else if (WARN_ON(ret)) {
852 btrfs_print_leaf(path->nodes[0]);
854 "extent item not found for insert, bytenr %llu num_bytes %llu parent %llu root_objectid %llu owner %llu offset %llu",
855 bytenr, num_bytes, parent, root_objectid, owner,
861 leaf = path->nodes[0];
862 item_size = btrfs_item_size(leaf, path->slots[0]);
863 if (unlikely(item_size < sizeof(*ei))) {
866 "unexpected extent item size, has %llu expect >= %zu",
867 item_size, sizeof(*ei));
868 btrfs_abort_transaction(trans, err);
872 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
873 flags = btrfs_extent_flags(leaf, ei);
875 ptr = (unsigned long)(ei + 1);
876 end = (unsigned long)ei + item_size;
878 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
879 ptr += sizeof(struct btrfs_tree_block_info);
883 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
884 needed = BTRFS_REF_TYPE_DATA;
886 needed = BTRFS_REF_TYPE_BLOCK;
893 btrfs_print_leaf(path->nodes[0]);
895 "overrun extent record at slot %d while looking for inline extent for root %llu owner %llu offset %llu parent %llu",
896 path->slots[0], root_objectid, owner, offset, parent);
900 iref = (struct btrfs_extent_inline_ref *)ptr;
901 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
902 if (type == BTRFS_REF_TYPE_INVALID) {
910 ptr += btrfs_extent_inline_ref_size(type);
914 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
915 struct btrfs_extent_data_ref *dref;
916 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
917 if (match_extent_data_ref(leaf, dref, root_objectid,
922 if (hash_extent_data_ref_item(leaf, dref) <
923 hash_extent_data_ref(root_objectid, owner, offset))
927 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
929 if (parent == ref_offset) {
933 if (ref_offset < parent)
936 if (root_objectid == ref_offset) {
940 if (ref_offset < root_objectid)
944 ptr += btrfs_extent_inline_ref_size(type);
946 if (err == -ENOENT && insert) {
947 if (item_size + extra_size >=
948 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
953 * To add new inline back ref, we have to make sure
954 * there is no corresponding back ref item.
955 * For simplicity, we just do not add new inline back
956 * ref if there is any kind of item for this block
958 if (find_next_key(path, 0, &key) == 0 &&
959 key.objectid == bytenr &&
960 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
965 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
968 path->keep_locks = 0;
969 path->search_for_extension = 0;
970 btrfs_unlock_up_safe(path, 1);
976 * helper to add new inline back ref
978 static noinline_for_stack
979 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
980 struct btrfs_path *path,
981 struct btrfs_extent_inline_ref *iref,
982 u64 parent, u64 root_objectid,
983 u64 owner, u64 offset, int refs_to_add,
984 struct btrfs_delayed_extent_op *extent_op)
986 struct extent_buffer *leaf;
987 struct btrfs_extent_item *ei;
990 unsigned long item_offset;
995 leaf = path->nodes[0];
996 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
997 item_offset = (unsigned long)iref - (unsigned long)ei;
999 type = extent_ref_type(parent, owner);
1000 size = btrfs_extent_inline_ref_size(type);
1002 btrfs_extend_item(trans, path, size);
1004 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1005 refs = btrfs_extent_refs(leaf, ei);
1006 refs += refs_to_add;
1007 btrfs_set_extent_refs(leaf, ei, refs);
1009 __run_delayed_extent_op(extent_op, leaf, ei);
1011 ptr = (unsigned long)ei + item_offset;
1012 end = (unsigned long)ei + btrfs_item_size(leaf, path->slots[0]);
1013 if (ptr < end - size)
1014 memmove_extent_buffer(leaf, ptr + size, ptr,
1017 iref = (struct btrfs_extent_inline_ref *)ptr;
1018 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1019 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1020 struct btrfs_extent_data_ref *dref;
1021 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1022 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1023 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1024 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1025 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1026 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1027 struct btrfs_shared_data_ref *sref;
1028 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1029 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1030 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1031 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1032 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1034 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1036 btrfs_mark_buffer_dirty(trans, leaf);
1039 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1040 struct btrfs_path *path,
1041 struct btrfs_extent_inline_ref **ref_ret,
1042 u64 bytenr, u64 num_bytes, u64 parent,
1043 u64 root_objectid, u64 owner, u64 offset)
1047 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1048 num_bytes, parent, root_objectid,
1053 btrfs_release_path(path);
1056 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1057 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1060 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1061 root_objectid, owner, offset);
1067 * helper to update/remove inline back ref
1069 static noinline_for_stack int update_inline_extent_backref(
1070 struct btrfs_trans_handle *trans,
1071 struct btrfs_path *path,
1072 struct btrfs_extent_inline_ref *iref,
1074 struct btrfs_delayed_extent_op *extent_op)
1076 struct extent_buffer *leaf = path->nodes[0];
1077 struct btrfs_fs_info *fs_info = leaf->fs_info;
1078 struct btrfs_extent_item *ei;
1079 struct btrfs_extent_data_ref *dref = NULL;
1080 struct btrfs_shared_data_ref *sref = NULL;
1088 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1089 refs = btrfs_extent_refs(leaf, ei);
1090 if (unlikely(refs_to_mod < 0 && refs + refs_to_mod <= 0)) {
1091 struct btrfs_key key;
1094 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1095 if (key.type == BTRFS_METADATA_ITEM_KEY)
1096 extent_size = fs_info->nodesize;
1098 extent_size = key.offset;
1099 btrfs_print_leaf(leaf);
1101 "invalid refs_to_mod for extent %llu num_bytes %u, has %d expect >= -%llu",
1102 key.objectid, extent_size, refs_to_mod, refs);
1105 refs += refs_to_mod;
1106 btrfs_set_extent_refs(leaf, ei, refs);
1108 __run_delayed_extent_op(extent_op, leaf, ei);
1110 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1112 * Function btrfs_get_extent_inline_ref_type() has already printed
1115 if (unlikely(type == BTRFS_REF_TYPE_INVALID))
1118 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1119 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1120 refs = btrfs_extent_data_ref_count(leaf, dref);
1121 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1122 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1123 refs = btrfs_shared_data_ref_count(leaf, sref);
1127 * For tree blocks we can only drop one ref for it, and tree
1128 * blocks should not have refs > 1.
1130 * Furthermore if we're inserting a new inline backref, we
1131 * won't reach this path either. That would be
1132 * setup_inline_extent_backref().
1134 if (unlikely(refs_to_mod != -1)) {
1135 struct btrfs_key key;
1137 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1139 btrfs_print_leaf(leaf);
1141 "invalid refs_to_mod for tree block %llu, has %d expect -1",
1142 key.objectid, refs_to_mod);
1147 if (unlikely(refs_to_mod < 0 && refs < -refs_to_mod)) {
1148 struct btrfs_key key;
1151 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1152 if (key.type == BTRFS_METADATA_ITEM_KEY)
1153 extent_size = fs_info->nodesize;
1155 extent_size = key.offset;
1156 btrfs_print_leaf(leaf);
1158 "invalid refs_to_mod for backref entry, iref %lu extent %llu num_bytes %u, has %d expect >= -%llu",
1159 (unsigned long)iref, key.objectid, extent_size,
1163 refs += refs_to_mod;
1166 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1167 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1169 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1171 size = btrfs_extent_inline_ref_size(type);
1172 item_size = btrfs_item_size(leaf, path->slots[0]);
1173 ptr = (unsigned long)iref;
1174 end = (unsigned long)ei + item_size;
1175 if (ptr + size < end)
1176 memmove_extent_buffer(leaf, ptr, ptr + size,
1179 btrfs_truncate_item(trans, path, item_size, 1);
1181 btrfs_mark_buffer_dirty(trans, leaf);
1185 static noinline_for_stack
1186 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1187 struct btrfs_path *path,
1188 u64 bytenr, u64 num_bytes, u64 parent,
1189 u64 root_objectid, u64 owner,
1190 u64 offset, int refs_to_add,
1191 struct btrfs_delayed_extent_op *extent_op)
1193 struct btrfs_extent_inline_ref *iref;
1196 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1197 num_bytes, parent, root_objectid,
1201 * We're adding refs to a tree block we already own, this
1202 * should not happen at all.
1204 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1205 btrfs_print_leaf(path->nodes[0]);
1206 btrfs_crit(trans->fs_info,
1207 "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu slot %u",
1208 bytenr, num_bytes, root_objectid, path->slots[0]);
1211 ret = update_inline_extent_backref(trans, path, iref,
1212 refs_to_add, extent_op);
1213 } else if (ret == -ENOENT) {
1214 setup_inline_extent_backref(trans, path, iref, parent,
1215 root_objectid, owner, offset,
1216 refs_to_add, extent_op);
1222 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1223 struct btrfs_root *root,
1224 struct btrfs_path *path,
1225 struct btrfs_extent_inline_ref *iref,
1226 int refs_to_drop, int is_data)
1230 BUG_ON(!is_data && refs_to_drop != 1);
1232 ret = update_inline_extent_backref(trans, path, iref,
1233 -refs_to_drop, NULL);
1235 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1237 ret = btrfs_del_item(trans, root, path);
1241 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1242 u64 *discarded_bytes)
1245 u64 bytes_left, end;
1246 u64 aligned_start = ALIGN(start, 1 << SECTOR_SHIFT);
1248 if (WARN_ON(start != aligned_start)) {
1249 len -= aligned_start - start;
1250 len = round_down(len, 1 << SECTOR_SHIFT);
1251 start = aligned_start;
1254 *discarded_bytes = 0;
1262 /* Skip any superblocks on this device. */
1263 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1264 u64 sb_start = btrfs_sb_offset(j);
1265 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1266 u64 size = sb_start - start;
1268 if (!in_range(sb_start, start, bytes_left) &&
1269 !in_range(sb_end, start, bytes_left) &&
1270 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1274 * Superblock spans beginning of range. Adjust start and
1277 if (sb_start <= start) {
1278 start += sb_end - start;
1283 bytes_left = end - start;
1288 ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1289 size >> SECTOR_SHIFT,
1292 *discarded_bytes += size;
1293 else if (ret != -EOPNOTSUPP)
1302 bytes_left = end - start;
1306 ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1307 bytes_left >> SECTOR_SHIFT,
1310 *discarded_bytes += bytes_left;
1315 static int do_discard_extent(struct btrfs_discard_stripe *stripe, u64 *bytes)
1317 struct btrfs_device *dev = stripe->dev;
1318 struct btrfs_fs_info *fs_info = dev->fs_info;
1319 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1320 u64 phys = stripe->physical;
1321 u64 len = stripe->length;
1325 /* Zone reset on a zoned filesystem */
1326 if (btrfs_can_zone_reset(dev, phys, len)) {
1329 ret = btrfs_reset_device_zone(dev, phys, len, &discarded);
1333 if (!btrfs_dev_replace_is_ongoing(dev_replace) ||
1334 dev != dev_replace->srcdev)
1337 src_disc = discarded;
1339 /* Send to replace target as well */
1340 ret = btrfs_reset_device_zone(dev_replace->tgtdev, phys, len,
1342 discarded += src_disc;
1343 } else if (bdev_max_discard_sectors(stripe->dev->bdev)) {
1344 ret = btrfs_issue_discard(dev->bdev, phys, len, &discarded);
1355 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1356 u64 num_bytes, u64 *actual_bytes)
1359 u64 discarded_bytes = 0;
1360 u64 end = bytenr + num_bytes;
1364 * Avoid races with device replace and make sure the devices in the
1365 * stripes don't go away while we are discarding.
1367 btrfs_bio_counter_inc_blocked(fs_info);
1369 struct btrfs_discard_stripe *stripes;
1370 unsigned int num_stripes;
1373 num_bytes = end - cur;
1374 stripes = btrfs_map_discard(fs_info, cur, &num_bytes, &num_stripes);
1375 if (IS_ERR(stripes)) {
1376 ret = PTR_ERR(stripes);
1377 if (ret == -EOPNOTSUPP)
1382 for (i = 0; i < num_stripes; i++) {
1383 struct btrfs_discard_stripe *stripe = stripes + i;
1386 if (!stripe->dev->bdev) {
1387 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1391 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
1392 &stripe->dev->dev_state))
1395 ret = do_discard_extent(stripe, &bytes);
1398 * Keep going if discard is not supported by the
1401 if (ret != -EOPNOTSUPP)
1405 discarded_bytes += bytes;
1413 btrfs_bio_counter_dec(fs_info);
1415 *actual_bytes = discarded_bytes;
1419 /* Can return -ENOMEM */
1420 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1421 struct btrfs_ref *generic_ref)
1423 struct btrfs_fs_info *fs_info = trans->fs_info;
1426 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1427 generic_ref->action);
1428 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1429 generic_ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID);
1431 if (generic_ref->type == BTRFS_REF_METADATA)
1432 ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL);
1434 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0);
1436 btrfs_ref_tree_mod(fs_info, generic_ref);
1442 * __btrfs_inc_extent_ref - insert backreference for a given extent
1444 * The counterpart is in __btrfs_free_extent(), with examples and more details
1447 * @trans: Handle of transaction
1449 * @node: The delayed ref node used to get the bytenr/length for
1450 * extent whose references are incremented.
1452 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1453 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1454 * bytenr of the parent block. Since new extents are always
1455 * created with indirect references, this will only be the case
1456 * when relocating a shared extent. In that case, root_objectid
1457 * will be BTRFS_TREE_RELOC_OBJECTID. Otherwise, parent must
1460 * @root_objectid: The id of the root where this modification has originated,
1461 * this can be either one of the well-known metadata trees or
1462 * the subvolume id which references this extent.
1464 * @owner: For data extents it is the inode number of the owning file.
1465 * For metadata extents this parameter holds the level in the
1466 * tree of the extent.
1468 * @offset: For metadata extents the offset is ignored and is currently
1469 * always passed as 0. For data extents it is the fileoffset
1470 * this extent belongs to.
1472 * @refs_to_add Number of references to add
1474 * @extent_op Pointer to a structure, holding information necessary when
1475 * updating a tree block's flags
1478 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1479 struct btrfs_delayed_ref_node *node,
1480 u64 parent, u64 root_objectid,
1481 u64 owner, u64 offset, int refs_to_add,
1482 struct btrfs_delayed_extent_op *extent_op)
1484 struct btrfs_path *path;
1485 struct extent_buffer *leaf;
1486 struct btrfs_extent_item *item;
1487 struct btrfs_key key;
1488 u64 bytenr = node->bytenr;
1489 u64 num_bytes = node->num_bytes;
1493 path = btrfs_alloc_path();
1497 /* this will setup the path even if it fails to insert the back ref */
1498 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1499 parent, root_objectid, owner,
1500 offset, refs_to_add, extent_op);
1501 if ((ret < 0 && ret != -EAGAIN) || !ret)
1505 * Ok we had -EAGAIN which means we didn't have space to insert and
1506 * inline extent ref, so just update the reference count and add a
1509 leaf = path->nodes[0];
1510 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1511 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1512 refs = btrfs_extent_refs(leaf, item);
1513 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1515 __run_delayed_extent_op(extent_op, leaf, item);
1517 btrfs_mark_buffer_dirty(trans, leaf);
1518 btrfs_release_path(path);
1520 /* now insert the actual backref */
1521 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1522 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1525 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1526 root_objectid, owner, offset,
1530 btrfs_abort_transaction(trans, ret);
1532 btrfs_free_path(path);
1536 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1537 struct btrfs_delayed_ref_node *node,
1538 struct btrfs_delayed_extent_op *extent_op,
1539 bool insert_reserved)
1542 struct btrfs_delayed_data_ref *ref;
1543 struct btrfs_key ins;
1548 ins.objectid = node->bytenr;
1549 ins.offset = node->num_bytes;
1550 ins.type = BTRFS_EXTENT_ITEM_KEY;
1552 ref = btrfs_delayed_node_to_data_ref(node);
1553 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
1555 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1556 parent = ref->parent;
1557 ref_root = ref->root;
1559 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1561 flags |= extent_op->flags_to_set;
1562 ret = alloc_reserved_file_extent(trans, parent, ref_root,
1563 flags, ref->objectid,
1566 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1567 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1568 ref->objectid, ref->offset,
1569 node->ref_mod, extent_op);
1570 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1571 ret = __btrfs_free_extent(trans, node, parent,
1572 ref_root, ref->objectid,
1573 ref->offset, node->ref_mod,
1581 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1582 struct extent_buffer *leaf,
1583 struct btrfs_extent_item *ei)
1585 u64 flags = btrfs_extent_flags(leaf, ei);
1586 if (extent_op->update_flags) {
1587 flags |= extent_op->flags_to_set;
1588 btrfs_set_extent_flags(leaf, ei, flags);
1591 if (extent_op->update_key) {
1592 struct btrfs_tree_block_info *bi;
1593 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1594 bi = (struct btrfs_tree_block_info *)(ei + 1);
1595 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1599 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1600 struct btrfs_delayed_ref_head *head,
1601 struct btrfs_delayed_extent_op *extent_op)
1603 struct btrfs_fs_info *fs_info = trans->fs_info;
1604 struct btrfs_root *root;
1605 struct btrfs_key key;
1606 struct btrfs_path *path;
1607 struct btrfs_extent_item *ei;
1608 struct extent_buffer *leaf;
1614 if (TRANS_ABORTED(trans))
1617 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1620 path = btrfs_alloc_path();
1624 key.objectid = head->bytenr;
1627 key.type = BTRFS_METADATA_ITEM_KEY;
1628 key.offset = extent_op->level;
1630 key.type = BTRFS_EXTENT_ITEM_KEY;
1631 key.offset = head->num_bytes;
1634 root = btrfs_extent_root(fs_info, key.objectid);
1636 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1643 if (path->slots[0] > 0) {
1645 btrfs_item_key_to_cpu(path->nodes[0], &key,
1647 if (key.objectid == head->bytenr &&
1648 key.type == BTRFS_EXTENT_ITEM_KEY &&
1649 key.offset == head->num_bytes)
1653 btrfs_release_path(path);
1656 key.objectid = head->bytenr;
1657 key.offset = head->num_bytes;
1658 key.type = BTRFS_EXTENT_ITEM_KEY;
1664 "missing extent item for extent %llu num_bytes %llu level %d",
1665 head->bytenr, head->num_bytes, extent_op->level);
1670 leaf = path->nodes[0];
1671 item_size = btrfs_item_size(leaf, path->slots[0]);
1673 if (unlikely(item_size < sizeof(*ei))) {
1676 "unexpected extent item size, has %u expect >= %zu",
1677 item_size, sizeof(*ei));
1678 btrfs_abort_transaction(trans, err);
1682 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1683 __run_delayed_extent_op(extent_op, leaf, ei);
1685 btrfs_mark_buffer_dirty(trans, leaf);
1687 btrfs_free_path(path);
1691 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1692 struct btrfs_delayed_ref_node *node,
1693 struct btrfs_delayed_extent_op *extent_op,
1694 bool insert_reserved)
1697 struct btrfs_delayed_tree_ref *ref;
1701 ref = btrfs_delayed_node_to_tree_ref(node);
1702 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1704 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1705 parent = ref->parent;
1706 ref_root = ref->root;
1708 if (unlikely(node->ref_mod != 1)) {
1709 btrfs_err(trans->fs_info,
1710 "btree block %llu has %d references rather than 1: action %d ref_root %llu parent %llu",
1711 node->bytenr, node->ref_mod, node->action, ref_root,
1715 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1716 BUG_ON(!extent_op || !extent_op->update_flags);
1717 ret = alloc_reserved_tree_block(trans, node, extent_op);
1718 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1719 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1720 ref->level, 0, 1, extent_op);
1721 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1722 ret = __btrfs_free_extent(trans, node, parent, ref_root,
1723 ref->level, 0, 1, extent_op);
1730 /* helper function to actually process a single delayed ref entry */
1731 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1732 struct btrfs_delayed_ref_node *node,
1733 struct btrfs_delayed_extent_op *extent_op,
1734 bool insert_reserved)
1738 if (TRANS_ABORTED(trans)) {
1739 if (insert_reserved)
1740 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1744 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1745 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1746 ret = run_delayed_tree_ref(trans, node, extent_op,
1748 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1749 node->type == BTRFS_SHARED_DATA_REF_KEY)
1750 ret = run_delayed_data_ref(trans, node, extent_op,
1754 if (ret && insert_reserved)
1755 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1757 btrfs_err(trans->fs_info,
1758 "failed to run delayed ref for logical %llu num_bytes %llu type %u action %u ref_mod %d: %d",
1759 node->bytenr, node->num_bytes, node->type,
1760 node->action, node->ref_mod, ret);
1764 static inline struct btrfs_delayed_ref_node *
1765 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1767 struct btrfs_delayed_ref_node *ref;
1769 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1773 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1774 * This is to prevent a ref count from going down to zero, which deletes
1775 * the extent item from the extent tree, when there still are references
1776 * to add, which would fail because they would not find the extent item.
1778 if (!list_empty(&head->ref_add_list))
1779 return list_first_entry(&head->ref_add_list,
1780 struct btrfs_delayed_ref_node, add_list);
1782 ref = rb_entry(rb_first_cached(&head->ref_tree),
1783 struct btrfs_delayed_ref_node, ref_node);
1784 ASSERT(list_empty(&ref->add_list));
1788 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1789 struct btrfs_delayed_ref_head *head)
1791 spin_lock(&delayed_refs->lock);
1792 head->processing = false;
1793 delayed_refs->num_heads_ready++;
1794 spin_unlock(&delayed_refs->lock);
1795 btrfs_delayed_ref_unlock(head);
1798 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1799 struct btrfs_delayed_ref_head *head)
1801 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1806 if (head->must_insert_reserved) {
1807 head->extent_op = NULL;
1808 btrfs_free_delayed_extent_op(extent_op);
1814 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1815 struct btrfs_delayed_ref_head *head)
1817 struct btrfs_delayed_extent_op *extent_op;
1820 extent_op = cleanup_extent_op(head);
1823 head->extent_op = NULL;
1824 spin_unlock(&head->lock);
1825 ret = run_delayed_extent_op(trans, head, extent_op);
1826 btrfs_free_delayed_extent_op(extent_op);
1827 return ret ? ret : 1;
1830 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1831 struct btrfs_delayed_ref_root *delayed_refs,
1832 struct btrfs_delayed_ref_head *head)
1834 int nr_items = 1; /* Dropping this ref head update. */
1837 * We had csum deletions accounted for in our delayed refs rsv, we need
1838 * to drop the csum leaves for this update from our delayed_refs_rsv.
1840 if (head->total_ref_mod < 0 && head->is_data) {
1841 spin_lock(&delayed_refs->lock);
1842 delayed_refs->pending_csums -= head->num_bytes;
1843 spin_unlock(&delayed_refs->lock);
1844 nr_items += btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes);
1847 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
1850 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1851 struct btrfs_delayed_ref_head *head)
1854 struct btrfs_fs_info *fs_info = trans->fs_info;
1855 struct btrfs_delayed_ref_root *delayed_refs;
1858 delayed_refs = &trans->transaction->delayed_refs;
1860 ret = run_and_cleanup_extent_op(trans, head);
1862 unselect_delayed_ref_head(delayed_refs, head);
1863 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1870 * Need to drop our head ref lock and re-acquire the delayed ref lock
1871 * and then re-check to make sure nobody got added.
1873 spin_unlock(&head->lock);
1874 spin_lock(&delayed_refs->lock);
1875 spin_lock(&head->lock);
1876 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1877 spin_unlock(&head->lock);
1878 spin_unlock(&delayed_refs->lock);
1881 btrfs_delete_ref_head(delayed_refs, head);
1882 spin_unlock(&head->lock);
1883 spin_unlock(&delayed_refs->lock);
1885 if (head->must_insert_reserved) {
1886 btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
1887 if (head->is_data) {
1888 struct btrfs_root *csum_root;
1890 csum_root = btrfs_csum_root(fs_info, head->bytenr);
1891 ret = btrfs_del_csums(trans, csum_root, head->bytenr,
1896 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1898 trace_run_delayed_ref_head(fs_info, head, 0);
1899 btrfs_delayed_ref_unlock(head);
1900 btrfs_put_delayed_ref_head(head);
1904 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1905 struct btrfs_trans_handle *trans)
1907 struct btrfs_delayed_ref_root *delayed_refs =
1908 &trans->transaction->delayed_refs;
1909 struct btrfs_delayed_ref_head *head = NULL;
1912 spin_lock(&delayed_refs->lock);
1913 head = btrfs_select_ref_head(delayed_refs);
1915 spin_unlock(&delayed_refs->lock);
1920 * Grab the lock that says we are going to process all the refs for
1923 ret = btrfs_delayed_ref_lock(delayed_refs, head);
1924 spin_unlock(&delayed_refs->lock);
1927 * We may have dropped the spin lock to get the head mutex lock, and
1928 * that might have given someone else time to free the head. If that's
1929 * true, it has been removed from our list and we can move on.
1932 head = ERR_PTR(-EAGAIN);
1937 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1938 struct btrfs_delayed_ref_head *locked_ref)
1940 struct btrfs_fs_info *fs_info = trans->fs_info;
1941 struct btrfs_delayed_ref_root *delayed_refs;
1942 struct btrfs_delayed_extent_op *extent_op;
1943 struct btrfs_delayed_ref_node *ref;
1944 bool must_insert_reserved;
1947 delayed_refs = &trans->transaction->delayed_refs;
1949 lockdep_assert_held(&locked_ref->mutex);
1950 lockdep_assert_held(&locked_ref->lock);
1952 while ((ref = select_delayed_ref(locked_ref))) {
1954 btrfs_check_delayed_seq(fs_info, ref->seq)) {
1955 spin_unlock(&locked_ref->lock);
1956 unselect_delayed_ref_head(delayed_refs, locked_ref);
1960 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1961 RB_CLEAR_NODE(&ref->ref_node);
1962 if (!list_empty(&ref->add_list))
1963 list_del(&ref->add_list);
1965 * When we play the delayed ref, also correct the ref_mod on
1968 switch (ref->action) {
1969 case BTRFS_ADD_DELAYED_REF:
1970 case BTRFS_ADD_DELAYED_EXTENT:
1971 locked_ref->ref_mod -= ref->ref_mod;
1973 case BTRFS_DROP_DELAYED_REF:
1974 locked_ref->ref_mod += ref->ref_mod;
1979 atomic_dec(&delayed_refs->num_entries);
1982 * Record the must_insert_reserved flag before we drop the
1985 must_insert_reserved = locked_ref->must_insert_reserved;
1986 locked_ref->must_insert_reserved = false;
1988 extent_op = locked_ref->extent_op;
1989 locked_ref->extent_op = NULL;
1990 spin_unlock(&locked_ref->lock);
1992 ret = run_one_delayed_ref(trans, ref, extent_op,
1993 must_insert_reserved);
1995 btrfs_free_delayed_extent_op(extent_op);
1997 unselect_delayed_ref_head(delayed_refs, locked_ref);
1998 btrfs_put_delayed_ref(ref);
2002 btrfs_put_delayed_ref(ref);
2005 spin_lock(&locked_ref->lock);
2006 btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
2013 * Returns 0 on success or if called with an already aborted transaction.
2014 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2016 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2019 struct btrfs_fs_info *fs_info = trans->fs_info;
2020 struct btrfs_delayed_ref_root *delayed_refs;
2021 struct btrfs_delayed_ref_head *locked_ref = NULL;
2023 unsigned long count = 0;
2025 delayed_refs = &trans->transaction->delayed_refs;
2028 locked_ref = btrfs_obtain_ref_head(trans);
2029 if (IS_ERR_OR_NULL(locked_ref)) {
2030 if (PTR_ERR(locked_ref) == -EAGAIN) {
2039 * We need to try and merge add/drops of the same ref since we
2040 * can run into issues with relocate dropping the implicit ref
2041 * and then it being added back again before the drop can
2042 * finish. If we merged anything we need to re-loop so we can
2044 * Or we can get node references of the same type that weren't
2045 * merged when created due to bumps in the tree mod seq, and
2046 * we need to merge them to prevent adding an inline extent
2047 * backref before dropping it (triggering a BUG_ON at
2048 * insert_inline_extent_backref()).
2050 spin_lock(&locked_ref->lock);
2051 btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
2053 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref);
2054 if (ret < 0 && ret != -EAGAIN) {
2056 * Error, btrfs_run_delayed_refs_for_head already
2057 * unlocked everything so just bail out
2062 * Success, perform the usual cleanup of a processed
2065 ret = cleanup_ref_head(trans, locked_ref);
2067 /* We dropped our lock, we need to loop. */
2076 * Either success case or btrfs_run_delayed_refs_for_head
2077 * returned -EAGAIN, meaning we need to select another head
2082 } while ((nr != -1 && count < nr) || locked_ref);
2087 #ifdef SCRAMBLE_DELAYED_REFS
2089 * Normally delayed refs get processed in ascending bytenr order. This
2090 * correlates in most cases to the order added. To expose dependencies on this
2091 * order, we start to process the tree in the middle instead of the beginning
2093 static u64 find_middle(struct rb_root *root)
2095 struct rb_node *n = root->rb_node;
2096 struct btrfs_delayed_ref_node *entry;
2099 u64 first = 0, last = 0;
2103 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2104 first = entry->bytenr;
2108 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2109 last = entry->bytenr;
2114 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2115 WARN_ON(!entry->in_tree);
2117 middle = entry->bytenr;
2131 * this starts processing the delayed reference count updates and
2132 * extent insertions we have queued up so far. count can be
2133 * 0, which means to process everything in the tree at the start
2134 * of the run (but not newly added entries), or it can be some target
2135 * number you'd like to process.
2137 * Returns 0 on success or if called with an aborted transaction
2138 * Returns <0 on error and aborts the transaction
2140 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2141 unsigned long count)
2143 struct btrfs_fs_info *fs_info = trans->fs_info;
2144 struct rb_node *node;
2145 struct btrfs_delayed_ref_root *delayed_refs;
2146 struct btrfs_delayed_ref_head *head;
2148 int run_all = count == (unsigned long)-1;
2150 /* We'll clean this up in btrfs_cleanup_transaction */
2151 if (TRANS_ABORTED(trans))
2154 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2157 delayed_refs = &trans->transaction->delayed_refs;
2159 count = delayed_refs->num_heads_ready;
2162 #ifdef SCRAMBLE_DELAYED_REFS
2163 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2165 ret = __btrfs_run_delayed_refs(trans, count);
2167 btrfs_abort_transaction(trans, ret);
2172 btrfs_create_pending_block_groups(trans);
2174 spin_lock(&delayed_refs->lock);
2175 node = rb_first_cached(&delayed_refs->href_root);
2177 spin_unlock(&delayed_refs->lock);
2180 head = rb_entry(node, struct btrfs_delayed_ref_head,
2182 refcount_inc(&head->refs);
2183 spin_unlock(&delayed_refs->lock);
2185 /* Mutex was contended, block until it's released and retry. */
2186 mutex_lock(&head->mutex);
2187 mutex_unlock(&head->mutex);
2189 btrfs_put_delayed_ref_head(head);
2197 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2198 struct extent_buffer *eb, u64 flags)
2200 struct btrfs_delayed_extent_op *extent_op;
2201 int level = btrfs_header_level(eb);
2204 extent_op = btrfs_alloc_delayed_extent_op();
2208 extent_op->flags_to_set = flags;
2209 extent_op->update_flags = true;
2210 extent_op->update_key = false;
2211 extent_op->level = level;
2213 ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op);
2215 btrfs_free_delayed_extent_op(extent_op);
2219 static noinline int check_delayed_ref(struct btrfs_root *root,
2220 struct btrfs_path *path,
2221 u64 objectid, u64 offset, u64 bytenr)
2223 struct btrfs_delayed_ref_head *head;
2224 struct btrfs_delayed_ref_node *ref;
2225 struct btrfs_delayed_data_ref *data_ref;
2226 struct btrfs_delayed_ref_root *delayed_refs;
2227 struct btrfs_transaction *cur_trans;
2228 struct rb_node *node;
2231 spin_lock(&root->fs_info->trans_lock);
2232 cur_trans = root->fs_info->running_transaction;
2234 refcount_inc(&cur_trans->use_count);
2235 spin_unlock(&root->fs_info->trans_lock);
2239 delayed_refs = &cur_trans->delayed_refs;
2240 spin_lock(&delayed_refs->lock);
2241 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2243 spin_unlock(&delayed_refs->lock);
2244 btrfs_put_transaction(cur_trans);
2248 if (!mutex_trylock(&head->mutex)) {
2250 spin_unlock(&delayed_refs->lock);
2251 btrfs_put_transaction(cur_trans);
2255 refcount_inc(&head->refs);
2256 spin_unlock(&delayed_refs->lock);
2258 btrfs_release_path(path);
2261 * Mutex was contended, block until it's released and let
2264 mutex_lock(&head->mutex);
2265 mutex_unlock(&head->mutex);
2266 btrfs_put_delayed_ref_head(head);
2267 btrfs_put_transaction(cur_trans);
2270 spin_unlock(&delayed_refs->lock);
2272 spin_lock(&head->lock);
2274 * XXX: We should replace this with a proper search function in the
2277 for (node = rb_first_cached(&head->ref_tree); node;
2278 node = rb_next(node)) {
2279 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2280 /* If it's a shared ref we know a cross reference exists */
2281 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2286 data_ref = btrfs_delayed_node_to_data_ref(ref);
2289 * If our ref doesn't match the one we're currently looking at
2290 * then we have a cross reference.
2292 if (data_ref->root != root->root_key.objectid ||
2293 data_ref->objectid != objectid ||
2294 data_ref->offset != offset) {
2299 spin_unlock(&head->lock);
2300 mutex_unlock(&head->mutex);
2301 btrfs_put_transaction(cur_trans);
2305 static noinline int check_committed_ref(struct btrfs_root *root,
2306 struct btrfs_path *path,
2307 u64 objectid, u64 offset, u64 bytenr,
2310 struct btrfs_fs_info *fs_info = root->fs_info;
2311 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2312 struct extent_buffer *leaf;
2313 struct btrfs_extent_data_ref *ref;
2314 struct btrfs_extent_inline_ref *iref;
2315 struct btrfs_extent_item *ei;
2316 struct btrfs_key key;
2321 key.objectid = bytenr;
2322 key.offset = (u64)-1;
2323 key.type = BTRFS_EXTENT_ITEM_KEY;
2325 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2328 BUG_ON(ret == 0); /* Corruption */
2331 if (path->slots[0] == 0)
2335 leaf = path->nodes[0];
2336 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2338 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2342 item_size = btrfs_item_size(leaf, path->slots[0]);
2343 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2345 /* If extent item has more than 1 inline ref then it's shared */
2346 if (item_size != sizeof(*ei) +
2347 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2351 * If extent created before last snapshot => it's shared unless the
2352 * snapshot has been deleted. Use the heuristic if strict is false.
2355 (btrfs_extent_generation(leaf, ei) <=
2356 btrfs_root_last_snapshot(&root->root_item)))
2359 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2361 /* If this extent has SHARED_DATA_REF then it's shared */
2362 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2363 if (type != BTRFS_EXTENT_DATA_REF_KEY)
2366 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2367 if (btrfs_extent_refs(leaf, ei) !=
2368 btrfs_extent_data_ref_count(leaf, ref) ||
2369 btrfs_extent_data_ref_root(leaf, ref) !=
2370 root->root_key.objectid ||
2371 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2372 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2380 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2381 u64 bytenr, bool strict, struct btrfs_path *path)
2386 ret = check_committed_ref(root, path, objectid,
2387 offset, bytenr, strict);
2388 if (ret && ret != -ENOENT)
2391 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2392 } while (ret == -EAGAIN);
2395 btrfs_release_path(path);
2396 if (btrfs_is_data_reloc_root(root))
2401 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2402 struct btrfs_root *root,
2403 struct extent_buffer *buf,
2404 int full_backref, int inc)
2406 struct btrfs_fs_info *fs_info = root->fs_info;
2412 struct btrfs_key key;
2413 struct btrfs_file_extent_item *fi;
2414 struct btrfs_ref generic_ref = { 0 };
2415 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2421 if (btrfs_is_testing(fs_info))
2424 ref_root = btrfs_header_owner(buf);
2425 nritems = btrfs_header_nritems(buf);
2426 level = btrfs_header_level(buf);
2428 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
2432 parent = buf->start;
2436 action = BTRFS_ADD_DELAYED_REF;
2438 action = BTRFS_DROP_DELAYED_REF;
2440 for (i = 0; i < nritems; i++) {
2442 btrfs_item_key_to_cpu(buf, &key, i);
2443 if (key.type != BTRFS_EXTENT_DATA_KEY)
2445 fi = btrfs_item_ptr(buf, i,
2446 struct btrfs_file_extent_item);
2447 if (btrfs_file_extent_type(buf, fi) ==
2448 BTRFS_FILE_EXTENT_INLINE)
2450 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2454 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2455 key.offset -= btrfs_file_extent_offset(buf, fi);
2456 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2458 btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2459 key.offset, root->root_key.objectid,
2462 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2464 ret = btrfs_free_extent(trans, &generic_ref);
2468 bytenr = btrfs_node_blockptr(buf, i);
2469 num_bytes = fs_info->nodesize;
2470 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2472 btrfs_init_tree_ref(&generic_ref, level - 1, ref_root,
2473 root->root_key.objectid, for_reloc);
2475 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2477 ret = btrfs_free_extent(trans, &generic_ref);
2487 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2488 struct extent_buffer *buf, int full_backref)
2490 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2493 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2494 struct extent_buffer *buf, int full_backref)
2496 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2499 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2501 struct btrfs_fs_info *fs_info = root->fs_info;
2506 flags = BTRFS_BLOCK_GROUP_DATA;
2507 else if (root == fs_info->chunk_root)
2508 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2510 flags = BTRFS_BLOCK_GROUP_METADATA;
2512 ret = btrfs_get_alloc_profile(fs_info, flags);
2516 static u64 first_logical_byte(struct btrfs_fs_info *fs_info)
2518 struct rb_node *leftmost;
2521 read_lock(&fs_info->block_group_cache_lock);
2522 /* Get the block group with the lowest logical start address. */
2523 leftmost = rb_first_cached(&fs_info->block_group_cache_tree);
2525 struct btrfs_block_group *bg;
2527 bg = rb_entry(leftmost, struct btrfs_block_group, cache_node);
2530 read_unlock(&fs_info->block_group_cache_lock);
2535 static int pin_down_extent(struct btrfs_trans_handle *trans,
2536 struct btrfs_block_group *cache,
2537 u64 bytenr, u64 num_bytes, int reserved)
2539 struct btrfs_fs_info *fs_info = cache->fs_info;
2541 spin_lock(&cache->space_info->lock);
2542 spin_lock(&cache->lock);
2543 cache->pinned += num_bytes;
2544 btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2547 cache->reserved -= num_bytes;
2548 cache->space_info->bytes_reserved -= num_bytes;
2550 spin_unlock(&cache->lock);
2551 spin_unlock(&cache->space_info->lock);
2553 set_extent_bit(&trans->transaction->pinned_extents, bytenr,
2554 bytenr + num_bytes - 1, EXTENT_DIRTY, NULL);
2558 int btrfs_pin_extent(struct btrfs_trans_handle *trans,
2559 u64 bytenr, u64 num_bytes, int reserved)
2561 struct btrfs_block_group *cache;
2563 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2564 BUG_ON(!cache); /* Logic error */
2566 pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
2568 btrfs_put_block_group(cache);
2573 * this function must be called within transaction
2575 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
2576 u64 bytenr, u64 num_bytes)
2578 struct btrfs_block_group *cache;
2581 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2586 * Fully cache the free space first so that our pin removes the free space
2589 ret = btrfs_cache_block_group(cache, true);
2593 pin_down_extent(trans, cache, bytenr, num_bytes, 0);
2595 /* remove us from the free space cache (if we're there at all) */
2596 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
2598 btrfs_put_block_group(cache);
2602 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2603 u64 start, u64 num_bytes)
2606 struct btrfs_block_group *block_group;
2608 block_group = btrfs_lookup_block_group(fs_info, start);
2612 ret = btrfs_cache_block_group(block_group, true);
2616 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2618 btrfs_put_block_group(block_group);
2622 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2624 struct btrfs_fs_info *fs_info = eb->fs_info;
2625 struct btrfs_file_extent_item *item;
2626 struct btrfs_key key;
2631 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2634 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2635 btrfs_item_key_to_cpu(eb, &key, i);
2636 if (key.type != BTRFS_EXTENT_DATA_KEY)
2638 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2639 found_type = btrfs_file_extent_type(eb, item);
2640 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2642 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2644 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2645 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2646 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2655 btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2657 atomic_inc(&bg->reservations);
2661 * Returns the free cluster for the given space info and sets empty_cluster to
2662 * what it should be based on the mount options.
2664 static struct btrfs_free_cluster *
2665 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2666 struct btrfs_space_info *space_info, u64 *empty_cluster)
2668 struct btrfs_free_cluster *ret = NULL;
2671 if (btrfs_mixed_space_info(space_info))
2674 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2675 ret = &fs_info->meta_alloc_cluster;
2676 if (btrfs_test_opt(fs_info, SSD))
2677 *empty_cluster = SZ_2M;
2679 *empty_cluster = SZ_64K;
2680 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2681 btrfs_test_opt(fs_info, SSD_SPREAD)) {
2682 *empty_cluster = SZ_2M;
2683 ret = &fs_info->data_alloc_cluster;
2689 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2691 const bool return_free_space)
2693 struct btrfs_block_group *cache = NULL;
2694 struct btrfs_space_info *space_info;
2695 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2696 struct btrfs_free_cluster *cluster = NULL;
2698 u64 total_unpinned = 0;
2699 u64 empty_cluster = 0;
2702 while (start <= end) {
2705 start >= cache->start + cache->length) {
2707 btrfs_put_block_group(cache);
2709 cache = btrfs_lookup_block_group(fs_info, start);
2710 BUG_ON(!cache); /* Logic error */
2712 cluster = fetch_cluster_info(fs_info,
2715 empty_cluster <<= 1;
2718 len = cache->start + cache->length - start;
2719 len = min(len, end + 1 - start);
2721 if (return_free_space)
2722 btrfs_add_free_space(cache, start, len);
2725 total_unpinned += len;
2726 space_info = cache->space_info;
2729 * If this space cluster has been marked as fragmented and we've
2730 * unpinned enough in this block group to potentially allow a
2731 * cluster to be created inside of it go ahead and clear the
2734 if (cluster && cluster->fragmented &&
2735 total_unpinned > empty_cluster) {
2736 spin_lock(&cluster->lock);
2737 cluster->fragmented = 0;
2738 spin_unlock(&cluster->lock);
2741 spin_lock(&space_info->lock);
2742 spin_lock(&cache->lock);
2743 cache->pinned -= len;
2744 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2745 space_info->max_extent_size = 0;
2747 space_info->bytes_readonly += len;
2749 } else if (btrfs_is_zoned(fs_info)) {
2750 /* Need reset before reusing in a zoned block group */
2751 space_info->bytes_zone_unusable += len;
2754 spin_unlock(&cache->lock);
2755 if (!readonly && return_free_space &&
2756 global_rsv->space_info == space_info) {
2757 spin_lock(&global_rsv->lock);
2758 if (!global_rsv->full) {
2759 u64 to_add = min(len, global_rsv->size -
2760 global_rsv->reserved);
2762 global_rsv->reserved += to_add;
2763 btrfs_space_info_update_bytes_may_use(fs_info,
2764 space_info, to_add);
2765 if (global_rsv->reserved >= global_rsv->size)
2766 global_rsv->full = 1;
2769 spin_unlock(&global_rsv->lock);
2771 /* Add to any tickets we may have */
2772 if (!readonly && return_free_space && len)
2773 btrfs_try_granting_tickets(fs_info, space_info);
2774 spin_unlock(&space_info->lock);
2778 btrfs_put_block_group(cache);
2782 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2784 struct btrfs_fs_info *fs_info = trans->fs_info;
2785 struct btrfs_block_group *block_group, *tmp;
2786 struct list_head *deleted_bgs;
2787 struct extent_io_tree *unpin;
2792 unpin = &trans->transaction->pinned_extents;
2794 while (!TRANS_ABORTED(trans)) {
2795 struct extent_state *cached_state = NULL;
2797 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2798 if (!find_first_extent_bit(unpin, 0, &start, &end,
2799 EXTENT_DIRTY, &cached_state)) {
2800 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2804 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2805 ret = btrfs_discard_extent(fs_info, start,
2806 end + 1 - start, NULL);
2808 clear_extent_dirty(unpin, start, end, &cached_state);
2809 unpin_extent_range(fs_info, start, end, true);
2810 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2811 free_extent_state(cached_state);
2815 if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2816 btrfs_discard_calc_delay(&fs_info->discard_ctl);
2817 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2821 * Transaction is finished. We don't need the lock anymore. We
2822 * do need to clean up the block groups in case of a transaction
2825 deleted_bgs = &trans->transaction->deleted_bgs;
2826 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2830 if (!TRANS_ABORTED(trans))
2831 ret = btrfs_discard_extent(fs_info,
2833 block_group->length,
2836 list_del_init(&block_group->bg_list);
2837 btrfs_unfreeze_block_group(block_group);
2838 btrfs_put_block_group(block_group);
2841 const char *errstr = btrfs_decode_error(ret);
2843 "discard failed while removing blockgroup: errno=%d %s",
2851 static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
2852 u64 bytenr, u64 num_bytes, bool is_data)
2857 struct btrfs_root *csum_root;
2859 csum_root = btrfs_csum_root(trans->fs_info, bytenr);
2860 ret = btrfs_del_csums(trans, csum_root, bytenr, num_bytes);
2862 btrfs_abort_transaction(trans, ret);
2867 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
2869 btrfs_abort_transaction(trans, ret);
2873 ret = btrfs_update_block_group(trans, bytenr, num_bytes, false);
2875 btrfs_abort_transaction(trans, ret);
2880 #define abort_and_dump(trans, path, fmt, args...) \
2882 btrfs_abort_transaction(trans, -EUCLEAN); \
2883 btrfs_print_leaf(path->nodes[0]); \
2884 btrfs_crit(trans->fs_info, fmt, ##args); \
2888 * Drop one or more refs of @node.
2890 * 1. Locate the extent refs.
2891 * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
2892 * Locate it, then reduce the refs number or remove the ref line completely.
2894 * 2. Update the refs count in EXTENT/METADATA_ITEM
2896 * Inline backref case:
2898 * in extent tree we have:
2900 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2901 * refs 2 gen 6 flags DATA
2902 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2903 * extent data backref root FS_TREE objectid 257 offset 0 count 1
2905 * This function gets called with:
2907 * node->bytenr = 13631488
2908 * node->num_bytes = 1048576
2909 * root_objectid = FS_TREE
2910 * owner_objectid = 257
2914 * Then we should get some like:
2916 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2917 * refs 1 gen 6 flags DATA
2918 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2920 * Keyed backref case:
2922 * in extent tree we have:
2924 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2925 * refs 754 gen 6 flags DATA
2927 * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
2928 * extent data backref root FS_TREE objectid 866 offset 0 count 1
2930 * This function get called with:
2932 * node->bytenr = 13631488
2933 * node->num_bytes = 1048576
2934 * root_objectid = FS_TREE
2935 * owner_objectid = 866
2939 * Then we should get some like:
2941 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2942 * refs 753 gen 6 flags DATA
2944 * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
2946 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2947 struct btrfs_delayed_ref_node *node, u64 parent,
2948 u64 root_objectid, u64 owner_objectid,
2949 u64 owner_offset, int refs_to_drop,
2950 struct btrfs_delayed_extent_op *extent_op)
2952 struct btrfs_fs_info *info = trans->fs_info;
2953 struct btrfs_key key;
2954 struct btrfs_path *path;
2955 struct btrfs_root *extent_root;
2956 struct extent_buffer *leaf;
2957 struct btrfs_extent_item *ei;
2958 struct btrfs_extent_inline_ref *iref;
2961 int extent_slot = 0;
2962 int found_extent = 0;
2966 u64 bytenr = node->bytenr;
2967 u64 num_bytes = node->num_bytes;
2968 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
2970 extent_root = btrfs_extent_root(info, bytenr);
2971 ASSERT(extent_root);
2973 path = btrfs_alloc_path();
2977 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2979 if (!is_data && refs_to_drop != 1) {
2981 "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
2982 node->bytenr, refs_to_drop);
2984 btrfs_abort_transaction(trans, ret);
2989 skinny_metadata = false;
2991 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
2992 parent, root_objectid, owner_objectid,
2996 * Either the inline backref or the SHARED_DATA_REF/
2997 * SHARED_BLOCK_REF is found
2999 * Here is a quick path to locate EXTENT/METADATA_ITEM.
3000 * It's possible the EXTENT/METADATA_ITEM is near current slot.
3002 extent_slot = path->slots[0];
3003 while (extent_slot >= 0) {
3004 btrfs_item_key_to_cpu(path->nodes[0], &key,
3006 if (key.objectid != bytenr)
3008 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3009 key.offset == num_bytes) {
3013 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3014 key.offset == owner_objectid) {
3019 /* Quick path didn't find the EXTEMT/METADATA_ITEM */
3020 if (path->slots[0] - extent_slot > 5)
3025 if (!found_extent) {
3027 abort_and_dump(trans, path,
3028 "invalid iref slot %u, no EXTENT/METADATA_ITEM found but has inline extent ref",
3033 /* Must be SHARED_* item, remove the backref first */
3034 ret = remove_extent_backref(trans, extent_root, path,
3035 NULL, refs_to_drop, is_data);
3037 btrfs_abort_transaction(trans, ret);
3040 btrfs_release_path(path);
3042 /* Slow path to locate EXTENT/METADATA_ITEM */
3043 key.objectid = bytenr;
3044 key.type = BTRFS_EXTENT_ITEM_KEY;
3045 key.offset = num_bytes;
3047 if (!is_data && skinny_metadata) {
3048 key.type = BTRFS_METADATA_ITEM_KEY;
3049 key.offset = owner_objectid;
3052 ret = btrfs_search_slot(trans, extent_root,
3054 if (ret > 0 && skinny_metadata && path->slots[0]) {
3056 * Couldn't find our skinny metadata item,
3057 * see if we have ye olde extent item.
3060 btrfs_item_key_to_cpu(path->nodes[0], &key,
3062 if (key.objectid == bytenr &&
3063 key.type == BTRFS_EXTENT_ITEM_KEY &&
3064 key.offset == num_bytes)
3068 if (ret > 0 && skinny_metadata) {
3069 skinny_metadata = false;
3070 key.objectid = bytenr;
3071 key.type = BTRFS_EXTENT_ITEM_KEY;
3072 key.offset = num_bytes;
3073 btrfs_release_path(path);
3074 ret = btrfs_search_slot(trans, extent_root,
3080 btrfs_print_leaf(path->nodes[0]);
3082 "umm, got %d back from search, was looking for %llu, slot %d",
3083 ret, bytenr, path->slots[0]);
3086 btrfs_abort_transaction(trans, ret);
3089 extent_slot = path->slots[0];
3091 } else if (WARN_ON(ret == -ENOENT)) {
3092 abort_and_dump(trans, path,
3093 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu slot %d",
3094 bytenr, parent, root_objectid, owner_objectid,
3095 owner_offset, path->slots[0]);
3098 btrfs_abort_transaction(trans, ret);
3102 leaf = path->nodes[0];
3103 item_size = btrfs_item_size(leaf, extent_slot);
3104 if (unlikely(item_size < sizeof(*ei))) {
3106 btrfs_err(trans->fs_info,
3107 "unexpected extent item size, has %u expect >= %zu",
3108 item_size, sizeof(*ei));
3109 btrfs_abort_transaction(trans, ret);
3112 ei = btrfs_item_ptr(leaf, extent_slot,
3113 struct btrfs_extent_item);
3114 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3115 key.type == BTRFS_EXTENT_ITEM_KEY) {
3116 struct btrfs_tree_block_info *bi;
3118 if (item_size < sizeof(*ei) + sizeof(*bi)) {
3119 abort_and_dump(trans, path,
3120 "invalid extent item size for key (%llu, %u, %llu) slot %u owner %llu, has %u expect >= %zu",
3121 key.objectid, key.type, key.offset,
3122 path->slots[0], owner_objectid, item_size,
3123 sizeof(*ei) + sizeof(*bi));
3127 bi = (struct btrfs_tree_block_info *)(ei + 1);
3128 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3131 refs = btrfs_extent_refs(leaf, ei);
3132 if (refs < refs_to_drop) {
3133 abort_and_dump(trans, path,
3134 "trying to drop %d refs but we only have %llu for bytenr %llu slot %u",
3135 refs_to_drop, refs, bytenr, path->slots[0]);
3139 refs -= refs_to_drop;
3143 __run_delayed_extent_op(extent_op, leaf, ei);
3145 * In the case of inline back ref, reference count will
3146 * be updated by remove_extent_backref
3149 if (!found_extent) {
3150 abort_and_dump(trans, path,
3151 "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found, slot %u",
3157 btrfs_set_extent_refs(leaf, ei, refs);
3158 btrfs_mark_buffer_dirty(trans, leaf);
3161 ret = remove_extent_backref(trans, extent_root, path,
3162 iref, refs_to_drop, is_data);
3164 btrfs_abort_transaction(trans, ret);
3169 /* In this branch refs == 1 */
3171 if (is_data && refs_to_drop !=
3172 extent_data_ref_count(path, iref)) {
3173 abort_and_dump(trans, path,
3174 "invalid refs_to_drop, current refs %u refs_to_drop %u slot %u",
3175 extent_data_ref_count(path, iref),
3176 refs_to_drop, path->slots[0]);
3181 if (path->slots[0] != extent_slot) {
3182 abort_and_dump(trans, path,
3183 "invalid iref, extent item key (%llu %u %llu) slot %u doesn't have wanted iref",
3184 key.objectid, key.type,
3185 key.offset, path->slots[0]);
3191 * No inline ref, we must be at SHARED_* item,
3192 * And it's single ref, it must be:
3193 * | extent_slot ||extent_slot + 1|
3194 * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3196 if (path->slots[0] != extent_slot + 1) {
3197 abort_and_dump(trans, path,
3198 "invalid SHARED_* item slot %u, previous item is not EXTENT/METADATA_ITEM",
3203 path->slots[0] = extent_slot;
3208 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3211 btrfs_abort_transaction(trans, ret);
3214 btrfs_release_path(path);
3216 ret = do_free_extent_accounting(trans, bytenr, num_bytes, is_data);
3218 btrfs_release_path(path);
3221 btrfs_free_path(path);
3226 * when we free an block, it is possible (and likely) that we free the last
3227 * delayed ref for that extent as well. This searches the delayed ref tree for
3228 * a given extent, and if there are no other delayed refs to be processed, it
3229 * removes it from the tree.
3231 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3234 struct btrfs_delayed_ref_head *head;
3235 struct btrfs_delayed_ref_root *delayed_refs;
3238 delayed_refs = &trans->transaction->delayed_refs;
3239 spin_lock(&delayed_refs->lock);
3240 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3242 goto out_delayed_unlock;
3244 spin_lock(&head->lock);
3245 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3248 if (cleanup_extent_op(head) != NULL)
3252 * waiting for the lock here would deadlock. If someone else has it
3253 * locked they are already in the process of dropping it anyway
3255 if (!mutex_trylock(&head->mutex))
3258 btrfs_delete_ref_head(delayed_refs, head);
3259 head->processing = false;
3261 spin_unlock(&head->lock);
3262 spin_unlock(&delayed_refs->lock);
3264 BUG_ON(head->extent_op);
3265 if (head->must_insert_reserved)
3268 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3269 mutex_unlock(&head->mutex);
3270 btrfs_put_delayed_ref_head(head);
3273 spin_unlock(&head->lock);
3276 spin_unlock(&delayed_refs->lock);
3280 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3282 struct extent_buffer *buf,
3283 u64 parent, int last_ref)
3285 struct btrfs_fs_info *fs_info = trans->fs_info;
3286 struct btrfs_ref generic_ref = { 0 };
3289 btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3290 buf->start, buf->len, parent);
3291 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3294 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3295 btrfs_ref_tree_mod(fs_info, &generic_ref);
3296 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL);
3297 BUG_ON(ret); /* -ENOMEM */
3300 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
3301 struct btrfs_block_group *cache;
3302 bool must_pin = false;
3304 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3305 ret = check_ref_cleanup(trans, buf->start);
3307 btrfs_redirty_list_add(trans->transaction, buf);
3312 cache = btrfs_lookup_block_group(fs_info, buf->start);
3314 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3315 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3316 btrfs_put_block_group(cache);
3321 * If there are tree mod log users we may have recorded mod log
3322 * operations for this node. If we re-allocate this node we
3323 * could replay operations on this node that happened when it
3324 * existed in a completely different root. For example if it
3325 * was part of root A, then was reallocated to root B, and we
3326 * are doing a btrfs_old_search_slot(root b), we could replay
3327 * operations that happened when the block was part of root A,
3328 * giving us an inconsistent view of the btree.
3330 * We are safe from races here because at this point no other
3331 * node or root points to this extent buffer, so if after this
3332 * check a new tree mod log user joins we will not have an
3333 * existing log of operations on this node that we have to
3336 if (test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags))
3339 if (must_pin || btrfs_is_zoned(fs_info)) {
3340 btrfs_redirty_list_add(trans->transaction, buf);
3341 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3342 btrfs_put_block_group(cache);
3346 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3348 btrfs_add_free_space(cache, buf->start, buf->len);
3349 btrfs_free_reserved_bytes(cache, buf->len, 0);
3350 btrfs_put_block_group(cache);
3351 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3356 * Deleting the buffer, clear the corrupt flag since it doesn't
3359 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3363 /* Can return -ENOMEM */
3364 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3366 struct btrfs_fs_info *fs_info = trans->fs_info;
3369 if (btrfs_is_testing(fs_info))
3373 * tree log blocks never actually go into the extent allocation
3374 * tree, just update pinning info and exit early.
3376 if ((ref->type == BTRFS_REF_METADATA &&
3377 ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3378 (ref->type == BTRFS_REF_DATA &&
3379 ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)) {
3380 /* unlocks the pinned mutex */
3381 btrfs_pin_extent(trans, ref->bytenr, ref->len, 1);
3383 } else if (ref->type == BTRFS_REF_METADATA) {
3384 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL);
3386 ret = btrfs_add_delayed_data_ref(trans, ref, 0);
3389 if (!((ref->type == BTRFS_REF_METADATA &&
3390 ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3391 (ref->type == BTRFS_REF_DATA &&
3392 ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)))
3393 btrfs_ref_tree_mod(fs_info, ref);
3398 enum btrfs_loop_type {
3400 * Start caching block groups but do not wait for progress or for them
3403 LOOP_CACHING_NOWAIT,
3406 * Wait for the block group free_space >= the space we're waiting for if
3407 * the block group isn't cached.
3412 * Allow allocations to happen from block groups that do not yet have a
3413 * size classification.
3415 LOOP_UNSET_SIZE_CLASS,
3418 * Allocate a chunk and then retry the allocation.
3423 * Ignore the size class restrictions for this allocation.
3425 LOOP_WRONG_SIZE_CLASS,
3428 * Ignore the empty size, only try to allocate the number of bytes
3429 * needed for this allocation.
3435 btrfs_lock_block_group(struct btrfs_block_group *cache,
3439 down_read(&cache->data_rwsem);
3442 static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3445 btrfs_get_block_group(cache);
3447 down_read(&cache->data_rwsem);
3450 static struct btrfs_block_group *btrfs_lock_cluster(
3451 struct btrfs_block_group *block_group,
3452 struct btrfs_free_cluster *cluster,
3454 __acquires(&cluster->refill_lock)
3456 struct btrfs_block_group *used_bg = NULL;
3458 spin_lock(&cluster->refill_lock);
3460 used_bg = cluster->block_group;
3464 if (used_bg == block_group)
3467 btrfs_get_block_group(used_bg);
3472 if (down_read_trylock(&used_bg->data_rwsem))
3475 spin_unlock(&cluster->refill_lock);
3477 /* We should only have one-level nested. */
3478 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3480 spin_lock(&cluster->refill_lock);
3481 if (used_bg == cluster->block_group)
3484 up_read(&used_bg->data_rwsem);
3485 btrfs_put_block_group(used_bg);
3490 btrfs_release_block_group(struct btrfs_block_group *cache,
3494 up_read(&cache->data_rwsem);
3495 btrfs_put_block_group(cache);
3499 * Helper function for find_free_extent().
3501 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3502 * Return >0 to inform caller that we find nothing
3503 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3505 static int find_free_extent_clustered(struct btrfs_block_group *bg,
3506 struct find_free_extent_ctl *ffe_ctl,
3507 struct btrfs_block_group **cluster_bg_ret)
3509 struct btrfs_block_group *cluster_bg;
3510 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3511 u64 aligned_cluster;
3515 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3517 goto refill_cluster;
3518 if (cluster_bg != bg && (cluster_bg->ro ||
3519 !block_group_bits(cluster_bg, ffe_ctl->flags)))
3520 goto release_cluster;
3522 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3523 ffe_ctl->num_bytes, cluster_bg->start,
3524 &ffe_ctl->max_extent_size);
3526 /* We have a block, we're done */
3527 spin_unlock(&last_ptr->refill_lock);
3528 trace_btrfs_reserve_extent_cluster(cluster_bg, ffe_ctl);
3529 *cluster_bg_ret = cluster_bg;
3530 ffe_ctl->found_offset = offset;
3533 WARN_ON(last_ptr->block_group != cluster_bg);
3537 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3538 * lets just skip it and let the allocator find whatever block it can
3539 * find. If we reach this point, we will have tried the cluster
3540 * allocator plenty of times and not have found anything, so we are
3541 * likely way too fragmented for the clustering stuff to find anything.
3543 * However, if the cluster is taken from the current block group,
3544 * release the cluster first, so that we stand a better chance of
3545 * succeeding in the unclustered allocation.
3547 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3548 spin_unlock(&last_ptr->refill_lock);
3549 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3553 /* This cluster didn't work out, free it and start over */
3554 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3556 if (cluster_bg != bg)
3557 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3560 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3561 spin_unlock(&last_ptr->refill_lock);
3565 aligned_cluster = max_t(u64,
3566 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3567 bg->full_stripe_len);
3568 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3569 ffe_ctl->num_bytes, aligned_cluster);
3571 /* Now pull our allocation out of this cluster */
3572 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3573 ffe_ctl->num_bytes, ffe_ctl->search_start,
3574 &ffe_ctl->max_extent_size);
3576 /* We found one, proceed */
3577 spin_unlock(&last_ptr->refill_lock);
3578 ffe_ctl->found_offset = offset;
3579 trace_btrfs_reserve_extent_cluster(bg, ffe_ctl);
3584 * At this point we either didn't find a cluster or we weren't able to
3585 * allocate a block from our cluster. Free the cluster we've been
3586 * trying to use, and go to the next block group.
3588 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3589 spin_unlock(&last_ptr->refill_lock);
3594 * Return >0 to inform caller that we find nothing
3595 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3597 static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3598 struct find_free_extent_ctl *ffe_ctl)
3600 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3604 * We are doing an unclustered allocation, set the fragmented flag so
3605 * we don't bother trying to setup a cluster again until we get more
3608 if (unlikely(last_ptr)) {
3609 spin_lock(&last_ptr->lock);
3610 last_ptr->fragmented = 1;
3611 spin_unlock(&last_ptr->lock);
3613 if (ffe_ctl->cached) {
3614 struct btrfs_free_space_ctl *free_space_ctl;
3616 free_space_ctl = bg->free_space_ctl;
3617 spin_lock(&free_space_ctl->tree_lock);
3618 if (free_space_ctl->free_space <
3619 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3620 ffe_ctl->empty_size) {
3621 ffe_ctl->total_free_space = max_t(u64,
3622 ffe_ctl->total_free_space,
3623 free_space_ctl->free_space);
3624 spin_unlock(&free_space_ctl->tree_lock);
3627 spin_unlock(&free_space_ctl->tree_lock);
3630 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3631 ffe_ctl->num_bytes, ffe_ctl->empty_size,
3632 &ffe_ctl->max_extent_size);
3635 ffe_ctl->found_offset = offset;
3639 static int do_allocation_clustered(struct btrfs_block_group *block_group,
3640 struct find_free_extent_ctl *ffe_ctl,
3641 struct btrfs_block_group **bg_ret)
3645 /* We want to try and use the cluster allocator, so lets look there */
3646 if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
3647 ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
3650 /* ret == -ENOENT case falls through */
3653 return find_free_extent_unclustered(block_group, ffe_ctl);
3657 * Tree-log block group locking
3658 * ============================
3660 * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3661 * indicates the starting address of a block group, which is reserved only
3662 * for tree-log metadata.
3669 * fs_info::treelog_bg_lock
3673 * Simple allocator for sequential-only block group. It only allows sequential
3674 * allocation. No need to play with trees. This function also reserves the
3675 * bytes as in btrfs_add_reserved_bytes.
3677 static int do_allocation_zoned(struct btrfs_block_group *block_group,
3678 struct find_free_extent_ctl *ffe_ctl,
3679 struct btrfs_block_group **bg_ret)
3681 struct btrfs_fs_info *fs_info = block_group->fs_info;
3682 struct btrfs_space_info *space_info = block_group->space_info;
3683 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3684 u64 start = block_group->start;
3685 u64 num_bytes = ffe_ctl->num_bytes;
3687 u64 bytenr = block_group->start;
3689 u64 data_reloc_bytenr;
3693 ASSERT(btrfs_is_zoned(block_group->fs_info));
3696 * Do not allow non-tree-log blocks in the dedicated tree-log block
3697 * group, and vice versa.
3699 spin_lock(&fs_info->treelog_bg_lock);
3700 log_bytenr = fs_info->treelog_bg;
3701 if (log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) ||
3702 (!ffe_ctl->for_treelog && bytenr == log_bytenr)))
3704 spin_unlock(&fs_info->treelog_bg_lock);
3709 * Do not allow non-relocation blocks in the dedicated relocation block
3710 * group, and vice versa.
3712 spin_lock(&fs_info->relocation_bg_lock);
3713 data_reloc_bytenr = fs_info->data_reloc_bg;
3714 if (data_reloc_bytenr &&
3715 ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) ||
3716 (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr)))
3718 spin_unlock(&fs_info->relocation_bg_lock);
3722 /* Check RO and no space case before trying to activate it */
3723 spin_lock(&block_group->lock);
3724 if (block_group->ro || btrfs_zoned_bg_is_full(block_group)) {
3727 * May need to clear fs_info->{treelog,data_reloc}_bg.
3728 * Return the error after taking the locks.
3731 spin_unlock(&block_group->lock);
3733 /* Metadata block group is activated at write time. */
3734 if (!ret && (block_group->flags & BTRFS_BLOCK_GROUP_DATA) &&
3735 !btrfs_zone_activate(block_group)) {
3738 * May need to clear fs_info->{treelog,data_reloc}_bg.
3739 * Return the error after taking the locks.
3743 spin_lock(&space_info->lock);
3744 spin_lock(&block_group->lock);
3745 spin_lock(&fs_info->treelog_bg_lock);
3746 spin_lock(&fs_info->relocation_bg_lock);
3751 ASSERT(!ffe_ctl->for_treelog ||
3752 block_group->start == fs_info->treelog_bg ||
3753 fs_info->treelog_bg == 0);
3754 ASSERT(!ffe_ctl->for_data_reloc ||
3755 block_group->start == fs_info->data_reloc_bg ||
3756 fs_info->data_reloc_bg == 0);
3758 if (block_group->ro ||
3759 (!ffe_ctl->for_data_reloc &&
3760 test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags))) {
3766 * Do not allow currently using block group to be tree-log dedicated
3769 if (ffe_ctl->for_treelog && !fs_info->treelog_bg &&
3770 (block_group->used || block_group->reserved)) {
3776 * Do not allow currently used block group to be the data relocation
3777 * dedicated block group.
3779 if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg &&
3780 (block_group->used || block_group->reserved)) {
3785 WARN_ON_ONCE(block_group->alloc_offset > block_group->zone_capacity);
3786 avail = block_group->zone_capacity - block_group->alloc_offset;
3787 if (avail < num_bytes) {
3788 if (ffe_ctl->max_extent_size < avail) {
3790 * With sequential allocator, free space is always
3793 ffe_ctl->max_extent_size = avail;
3794 ffe_ctl->total_free_space = avail;
3800 if (ffe_ctl->for_treelog && !fs_info->treelog_bg)
3801 fs_info->treelog_bg = block_group->start;
3803 if (ffe_ctl->for_data_reloc) {
3804 if (!fs_info->data_reloc_bg)
3805 fs_info->data_reloc_bg = block_group->start;
3807 * Do not allow allocations from this block group, unless it is
3808 * for data relocation. Compared to increasing the ->ro, setting
3809 * the ->zoned_data_reloc_ongoing flag still allows nocow
3810 * writers to come in. See btrfs_inc_nocow_writers().
3812 * We need to disable an allocation to avoid an allocation of
3813 * regular (non-relocation data) extent. With mix of relocation
3814 * extents and regular extents, we can dispatch WRITE commands
3815 * (for relocation extents) and ZONE APPEND commands (for
3816 * regular extents) at the same time to the same zone, which
3817 * easily break the write pointer.
3819 * Also, this flag avoids this block group to be zone finished.
3821 set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags);
3824 ffe_ctl->found_offset = start + block_group->alloc_offset;
3825 block_group->alloc_offset += num_bytes;
3826 spin_lock(&ctl->tree_lock);
3827 ctl->free_space -= num_bytes;
3828 spin_unlock(&ctl->tree_lock);
3831 * We do not check if found_offset is aligned to stripesize. The
3832 * address is anyway rewritten when using zone append writing.
3835 ffe_ctl->search_start = ffe_ctl->found_offset;
3838 if (ret && ffe_ctl->for_treelog)
3839 fs_info->treelog_bg = 0;
3840 if (ret && ffe_ctl->for_data_reloc)
3841 fs_info->data_reloc_bg = 0;
3842 spin_unlock(&fs_info->relocation_bg_lock);
3843 spin_unlock(&fs_info->treelog_bg_lock);
3844 spin_unlock(&block_group->lock);
3845 spin_unlock(&space_info->lock);
3849 static int do_allocation(struct btrfs_block_group *block_group,
3850 struct find_free_extent_ctl *ffe_ctl,
3851 struct btrfs_block_group **bg_ret)
3853 switch (ffe_ctl->policy) {
3854 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3855 return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
3856 case BTRFS_EXTENT_ALLOC_ZONED:
3857 return do_allocation_zoned(block_group, ffe_ctl, bg_ret);
3863 static void release_block_group(struct btrfs_block_group *block_group,
3864 struct find_free_extent_ctl *ffe_ctl,
3867 switch (ffe_ctl->policy) {
3868 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3869 ffe_ctl->retry_uncached = false;
3871 case BTRFS_EXTENT_ALLOC_ZONED:
3878 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
3880 btrfs_release_block_group(block_group, delalloc);
3883 static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
3884 struct btrfs_key *ins)
3886 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3888 if (!ffe_ctl->use_cluster && last_ptr) {
3889 spin_lock(&last_ptr->lock);
3890 last_ptr->window_start = ins->objectid;
3891 spin_unlock(&last_ptr->lock);
3895 static void found_extent(struct find_free_extent_ctl *ffe_ctl,
3896 struct btrfs_key *ins)
3898 switch (ffe_ctl->policy) {
3899 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3900 found_extent_clustered(ffe_ctl, ins);
3902 case BTRFS_EXTENT_ALLOC_ZONED:
3910 static int can_allocate_chunk_zoned(struct btrfs_fs_info *fs_info,
3911 struct find_free_extent_ctl *ffe_ctl)
3913 /* Block group's activeness is not a requirement for METADATA block groups. */
3914 if (!(ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA))
3917 /* If we can activate new zone, just allocate a chunk and use it */
3918 if (btrfs_can_activate_zone(fs_info->fs_devices, ffe_ctl->flags))
3922 * We already reached the max active zones. Try to finish one block
3923 * group to make a room for a new block group. This is only possible
3924 * for a data block group because btrfs_zone_finish() may need to wait
3925 * for a running transaction which can cause a deadlock for metadata
3928 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
3929 int ret = btrfs_zone_finish_one_bg(fs_info);
3938 * If we have enough free space left in an already active block group
3939 * and we can't activate any other zone now, do not allow allocating a
3940 * new chunk and let find_free_extent() retry with a smaller size.
3942 if (ffe_ctl->max_extent_size >= ffe_ctl->min_alloc_size)
3946 * Even min_alloc_size is not left in any block groups. Since we cannot
3947 * activate a new block group, allocating it may not help. Let's tell a
3948 * caller to try again and hope it progress something by writing some
3949 * parts of the region. That is only possible for data block groups,
3950 * where a part of the region can be written.
3952 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA)
3956 * We cannot activate a new block group and no enough space left in any
3957 * block groups. So, allocating a new block group may not help. But,
3958 * there is nothing to do anyway, so let's go with it.
3963 static int can_allocate_chunk(struct btrfs_fs_info *fs_info,
3964 struct find_free_extent_ctl *ffe_ctl)
3966 switch (ffe_ctl->policy) {
3967 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3969 case BTRFS_EXTENT_ALLOC_ZONED:
3970 return can_allocate_chunk_zoned(fs_info, ffe_ctl);
3977 * Return >0 means caller needs to re-search for free extent
3978 * Return 0 means we have the needed free extent.
3979 * Return <0 means we failed to locate any free extent.
3981 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
3982 struct btrfs_key *ins,
3983 struct find_free_extent_ctl *ffe_ctl,
3986 struct btrfs_root *root = fs_info->chunk_root;
3989 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
3990 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
3991 ffe_ctl->orig_have_caching_bg = true;
3993 if (ins->objectid) {
3994 found_extent(ffe_ctl, ins);
3998 if (ffe_ctl->loop >= LOOP_CACHING_WAIT && ffe_ctl->have_caching_bg)
4002 if (ffe_ctl->index < BTRFS_NR_RAID_TYPES)
4005 /* See the comments for btrfs_loop_type for an explanation of the phases. */
4006 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
4009 * We want to skip the LOOP_CACHING_WAIT step if we don't have
4010 * any uncached bgs and we've already done a full search
4013 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT &&
4014 (!ffe_ctl->orig_have_caching_bg && full_search))
4018 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
4019 struct btrfs_trans_handle *trans;
4022 /* Check if allocation policy allows to create a new chunk */
4023 ret = can_allocate_chunk(fs_info, ffe_ctl);
4027 trans = current->journal_info;
4031 trans = btrfs_join_transaction(root);
4033 if (IS_ERR(trans)) {
4034 ret = PTR_ERR(trans);
4038 ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
4039 CHUNK_ALLOC_FORCE_FOR_EXTENT);
4041 /* Do not bail out on ENOSPC since we can do more. */
4042 if (ret == -ENOSPC) {
4047 btrfs_abort_transaction(trans, ret);
4051 btrfs_end_transaction(trans);
4056 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
4057 if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
4061 * Don't loop again if we already have no empty_size and
4064 if (ffe_ctl->empty_size == 0 &&
4065 ffe_ctl->empty_cluster == 0)
4067 ffe_ctl->empty_size = 0;
4068 ffe_ctl->empty_cluster = 0;
4075 static bool find_free_extent_check_size_class(struct find_free_extent_ctl *ffe_ctl,
4076 struct btrfs_block_group *bg)
4078 if (ffe_ctl->policy == BTRFS_EXTENT_ALLOC_ZONED)
4080 if (!btrfs_block_group_should_use_size_class(bg))
4082 if (ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS)
4084 if (ffe_ctl->loop >= LOOP_UNSET_SIZE_CLASS &&
4085 bg->size_class == BTRFS_BG_SZ_NONE)
4087 return ffe_ctl->size_class == bg->size_class;
4090 static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
4091 struct find_free_extent_ctl *ffe_ctl,
4092 struct btrfs_space_info *space_info,
4093 struct btrfs_key *ins)
4096 * If our free space is heavily fragmented we may not be able to make
4097 * big contiguous allocations, so instead of doing the expensive search
4098 * for free space, simply return ENOSPC with our max_extent_size so we
4099 * can go ahead and search for a more manageable chunk.
4101 * If our max_extent_size is large enough for our allocation simply
4102 * disable clustering since we will likely not be able to find enough
4103 * space to create a cluster and induce latency trying.
4105 if (space_info->max_extent_size) {
4106 spin_lock(&space_info->lock);
4107 if (space_info->max_extent_size &&
4108 ffe_ctl->num_bytes > space_info->max_extent_size) {
4109 ins->offset = space_info->max_extent_size;
4110 spin_unlock(&space_info->lock);
4112 } else if (space_info->max_extent_size) {
4113 ffe_ctl->use_cluster = false;
4115 spin_unlock(&space_info->lock);
4118 ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
4119 &ffe_ctl->empty_cluster);
4120 if (ffe_ctl->last_ptr) {
4121 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4123 spin_lock(&last_ptr->lock);
4124 if (last_ptr->block_group)
4125 ffe_ctl->hint_byte = last_ptr->window_start;
4126 if (last_ptr->fragmented) {
4128 * We still set window_start so we can keep track of the
4129 * last place we found an allocation to try and save
4132 ffe_ctl->hint_byte = last_ptr->window_start;
4133 ffe_ctl->use_cluster = false;
4135 spin_unlock(&last_ptr->lock);
4141 static int prepare_allocation(struct btrfs_fs_info *fs_info,
4142 struct find_free_extent_ctl *ffe_ctl,
4143 struct btrfs_space_info *space_info,
4144 struct btrfs_key *ins)
4146 switch (ffe_ctl->policy) {
4147 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4148 return prepare_allocation_clustered(fs_info, ffe_ctl,
4150 case BTRFS_EXTENT_ALLOC_ZONED:
4151 if (ffe_ctl->for_treelog) {
4152 spin_lock(&fs_info->treelog_bg_lock);
4153 if (fs_info->treelog_bg)
4154 ffe_ctl->hint_byte = fs_info->treelog_bg;
4155 spin_unlock(&fs_info->treelog_bg_lock);
4157 if (ffe_ctl->for_data_reloc) {
4158 spin_lock(&fs_info->relocation_bg_lock);
4159 if (fs_info->data_reloc_bg)
4160 ffe_ctl->hint_byte = fs_info->data_reloc_bg;
4161 spin_unlock(&fs_info->relocation_bg_lock);
4170 * walks the btree of allocated extents and find a hole of a given size.
4171 * The key ins is changed to record the hole:
4172 * ins->objectid == start position
4173 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4174 * ins->offset == the size of the hole.
4175 * Any available blocks before search_start are skipped.
4177 * If there is no suitable free space, we will record the max size of
4178 * the free space extent currently.
4180 * The overall logic and call chain:
4182 * find_free_extent()
4183 * |- Iterate through all block groups
4184 * | |- Get a valid block group
4185 * | |- Try to do clustered allocation in that block group
4186 * | |- Try to do unclustered allocation in that block group
4187 * | |- Check if the result is valid
4188 * | | |- If valid, then exit
4189 * | |- Jump to next block group
4191 * |- Push harder to find free extents
4192 * |- If not found, re-iterate all block groups
4194 static noinline int find_free_extent(struct btrfs_root *root,
4195 struct btrfs_key *ins,
4196 struct find_free_extent_ctl *ffe_ctl)
4198 struct btrfs_fs_info *fs_info = root->fs_info;
4200 int cache_block_group_error = 0;
4201 struct btrfs_block_group *block_group = NULL;
4202 struct btrfs_space_info *space_info;
4203 bool full_search = false;
4205 WARN_ON(ffe_ctl->num_bytes < fs_info->sectorsize);
4207 ffe_ctl->search_start = 0;
4208 /* For clustered allocation */
4209 ffe_ctl->empty_cluster = 0;
4210 ffe_ctl->last_ptr = NULL;
4211 ffe_ctl->use_cluster = true;
4212 ffe_ctl->have_caching_bg = false;
4213 ffe_ctl->orig_have_caching_bg = false;
4214 ffe_ctl->index = btrfs_bg_flags_to_raid_index(ffe_ctl->flags);
4216 ffe_ctl->retry_uncached = false;
4217 ffe_ctl->cached = 0;
4218 ffe_ctl->max_extent_size = 0;
4219 ffe_ctl->total_free_space = 0;
4220 ffe_ctl->found_offset = 0;
4221 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
4222 ffe_ctl->size_class = btrfs_calc_block_group_size_class(ffe_ctl->num_bytes);
4224 if (btrfs_is_zoned(fs_info))
4225 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_ZONED;
4227 ins->type = BTRFS_EXTENT_ITEM_KEY;
4231 trace_find_free_extent(root, ffe_ctl);
4233 space_info = btrfs_find_space_info(fs_info, ffe_ctl->flags);
4235 btrfs_err(fs_info, "No space info for %llu", ffe_ctl->flags);
4239 ret = prepare_allocation(fs_info, ffe_ctl, space_info, ins);
4243 ffe_ctl->search_start = max(ffe_ctl->search_start,
4244 first_logical_byte(fs_info));
4245 ffe_ctl->search_start = max(ffe_ctl->search_start, ffe_ctl->hint_byte);
4246 if (ffe_ctl->search_start == ffe_ctl->hint_byte) {
4247 block_group = btrfs_lookup_block_group(fs_info,
4248 ffe_ctl->search_start);
4250 * we don't want to use the block group if it doesn't match our
4251 * allocation bits, or if its not cached.
4253 * However if we are re-searching with an ideal block group
4254 * picked out then we don't care that the block group is cached.
4256 if (block_group && block_group_bits(block_group, ffe_ctl->flags) &&
4257 block_group->cached != BTRFS_CACHE_NO) {
4258 down_read(&space_info->groups_sem);
4259 if (list_empty(&block_group->list) ||
4262 * someone is removing this block group,
4263 * we can't jump into the have_block_group
4264 * target because our list pointers are not
4267 btrfs_put_block_group(block_group);
4268 up_read(&space_info->groups_sem);
4270 ffe_ctl->index = btrfs_bg_flags_to_raid_index(
4271 block_group->flags);
4272 btrfs_lock_block_group(block_group,
4274 ffe_ctl->hinted = true;
4275 goto have_block_group;
4277 } else if (block_group) {
4278 btrfs_put_block_group(block_group);
4282 trace_find_free_extent_search_loop(root, ffe_ctl);
4283 ffe_ctl->have_caching_bg = false;
4284 if (ffe_ctl->index == btrfs_bg_flags_to_raid_index(ffe_ctl->flags) ||
4285 ffe_ctl->index == 0)
4287 down_read(&space_info->groups_sem);
4288 list_for_each_entry(block_group,
4289 &space_info->block_groups[ffe_ctl->index], list) {
4290 struct btrfs_block_group *bg_ret;
4292 ffe_ctl->hinted = false;
4293 /* If the block group is read-only, we can skip it entirely. */
4294 if (unlikely(block_group->ro)) {
4295 if (ffe_ctl->for_treelog)
4296 btrfs_clear_treelog_bg(block_group);
4297 if (ffe_ctl->for_data_reloc)
4298 btrfs_clear_data_reloc_bg(block_group);
4302 btrfs_grab_block_group(block_group, ffe_ctl->delalloc);
4303 ffe_ctl->search_start = block_group->start;
4306 * this can happen if we end up cycling through all the
4307 * raid types, but we want to make sure we only allocate
4308 * for the proper type.
4310 if (!block_group_bits(block_group, ffe_ctl->flags)) {
4311 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4312 BTRFS_BLOCK_GROUP_RAID1_MASK |
4313 BTRFS_BLOCK_GROUP_RAID56_MASK |
4314 BTRFS_BLOCK_GROUP_RAID10;
4317 * if they asked for extra copies and this block group
4318 * doesn't provide them, bail. This does allow us to
4319 * fill raid0 from raid1.
4321 if ((ffe_ctl->flags & extra) && !(block_group->flags & extra))
4325 * This block group has different flags than we want.
4326 * It's possible that we have MIXED_GROUP flag but no
4327 * block group is mixed. Just skip such block group.
4329 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4334 trace_find_free_extent_have_block_group(root, ffe_ctl, block_group);
4335 ffe_ctl->cached = btrfs_block_group_done(block_group);
4336 if (unlikely(!ffe_ctl->cached)) {
4337 ffe_ctl->have_caching_bg = true;
4338 ret = btrfs_cache_block_group(block_group, false);
4341 * If we get ENOMEM here or something else we want to
4342 * try other block groups, because it may not be fatal.
4343 * However if we can't find anything else we need to
4344 * save our return here so that we return the actual
4345 * error that caused problems, not ENOSPC.
4348 if (!cache_block_group_error)
4349 cache_block_group_error = ret;
4356 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) {
4357 if (!cache_block_group_error)
4358 cache_block_group_error = -EIO;
4362 if (!find_free_extent_check_size_class(ffe_ctl, block_group))
4366 ret = do_allocation(block_group, ffe_ctl, &bg_ret);
4370 if (bg_ret && bg_ret != block_group) {
4371 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4372 block_group = bg_ret;
4376 ffe_ctl->search_start = round_up(ffe_ctl->found_offset,
4377 fs_info->stripesize);
4379 /* move on to the next group */
4380 if (ffe_ctl->search_start + ffe_ctl->num_bytes >
4381 block_group->start + block_group->length) {
4382 btrfs_add_free_space_unused(block_group,
4383 ffe_ctl->found_offset,
4384 ffe_ctl->num_bytes);
4388 if (ffe_ctl->found_offset < ffe_ctl->search_start)
4389 btrfs_add_free_space_unused(block_group,
4390 ffe_ctl->found_offset,
4391 ffe_ctl->search_start - ffe_ctl->found_offset);
4393 ret = btrfs_add_reserved_bytes(block_group, ffe_ctl->ram_bytes,
4396 ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS);
4397 if (ret == -EAGAIN) {
4398 btrfs_add_free_space_unused(block_group,
4399 ffe_ctl->found_offset,
4400 ffe_ctl->num_bytes);
4403 btrfs_inc_block_group_reservations(block_group);
4405 /* we are all good, lets return */
4406 ins->objectid = ffe_ctl->search_start;
4407 ins->offset = ffe_ctl->num_bytes;
4409 trace_btrfs_reserve_extent(block_group, ffe_ctl);
4410 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4413 if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
4414 !ffe_ctl->retry_uncached) {
4415 ffe_ctl->retry_uncached = true;
4416 btrfs_wait_block_group_cache_progress(block_group,
4417 ffe_ctl->num_bytes +
4418 ffe_ctl->empty_cluster +
4419 ffe_ctl->empty_size);
4420 goto have_block_group;
4422 release_block_group(block_group, ffe_ctl, ffe_ctl->delalloc);
4425 up_read(&space_info->groups_sem);
4427 ret = find_free_extent_update_loop(fs_info, ins, ffe_ctl, full_search);
4431 if (ret == -ENOSPC && !cache_block_group_error) {
4433 * Use ffe_ctl->total_free_space as fallback if we can't find
4434 * any contiguous hole.
4436 if (!ffe_ctl->max_extent_size)
4437 ffe_ctl->max_extent_size = ffe_ctl->total_free_space;
4438 spin_lock(&space_info->lock);
4439 space_info->max_extent_size = ffe_ctl->max_extent_size;
4440 spin_unlock(&space_info->lock);
4441 ins->offset = ffe_ctl->max_extent_size;
4442 } else if (ret == -ENOSPC) {
4443 ret = cache_block_group_error;
4449 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4450 * hole that is at least as big as @num_bytes.
4452 * @root - The root that will contain this extent
4454 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4455 * is used for accounting purposes. This value differs
4456 * from @num_bytes only in the case of compressed extents.
4458 * @num_bytes - Number of bytes to allocate on-disk.
4460 * @min_alloc_size - Indicates the minimum amount of space that the
4461 * allocator should try to satisfy. In some cases
4462 * @num_bytes may be larger than what is required and if
4463 * the filesystem is fragmented then allocation fails.
4464 * However, the presence of @min_alloc_size gives a
4465 * chance to try and satisfy the smaller allocation.
4467 * @empty_size - A hint that you plan on doing more COW. This is the
4468 * size in bytes the allocator should try to find free
4469 * next to the block it returns. This is just a hint and
4470 * may be ignored by the allocator.
4472 * @hint_byte - Hint to the allocator to start searching above the byte
4473 * address passed. It might be ignored.
4475 * @ins - This key is modified to record the found hole. It will
4476 * have the following values:
4477 * ins->objectid == start position
4478 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4479 * ins->offset == the size of the hole.
4481 * @is_data - Boolean flag indicating whether an extent is
4482 * allocated for data (true) or metadata (false)
4484 * @delalloc - Boolean flag indicating whether this allocation is for
4485 * delalloc or not. If 'true' data_rwsem of block groups
4486 * is going to be acquired.
4489 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4490 * case -ENOSPC is returned then @ins->offset will contain the size of the
4491 * largest available hole the allocator managed to find.
4493 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4494 u64 num_bytes, u64 min_alloc_size,
4495 u64 empty_size, u64 hint_byte,
4496 struct btrfs_key *ins, int is_data, int delalloc)
4498 struct btrfs_fs_info *fs_info = root->fs_info;
4499 struct find_free_extent_ctl ffe_ctl = {};
4500 bool final_tried = num_bytes == min_alloc_size;
4503 bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4504 bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data);
4506 flags = get_alloc_profile_by_root(root, is_data);
4508 WARN_ON(num_bytes < fs_info->sectorsize);
4510 ffe_ctl.ram_bytes = ram_bytes;
4511 ffe_ctl.num_bytes = num_bytes;
4512 ffe_ctl.min_alloc_size = min_alloc_size;
4513 ffe_ctl.empty_size = empty_size;
4514 ffe_ctl.flags = flags;
4515 ffe_ctl.delalloc = delalloc;
4516 ffe_ctl.hint_byte = hint_byte;
4517 ffe_ctl.for_treelog = for_treelog;
4518 ffe_ctl.for_data_reloc = for_data_reloc;
4520 ret = find_free_extent(root, ins, &ffe_ctl);
4521 if (!ret && !is_data) {
4522 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4523 } else if (ret == -ENOSPC) {
4524 if (!final_tried && ins->offset) {
4525 num_bytes = min(num_bytes >> 1, ins->offset);
4526 num_bytes = round_down(num_bytes,
4527 fs_info->sectorsize);
4528 num_bytes = max(num_bytes, min_alloc_size);
4529 ram_bytes = num_bytes;
4530 if (num_bytes == min_alloc_size)
4533 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4534 struct btrfs_space_info *sinfo;
4536 sinfo = btrfs_find_space_info(fs_info, flags);
4538 "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d",
4539 flags, num_bytes, for_treelog, for_data_reloc);
4541 btrfs_dump_space_info(fs_info, sinfo,
4549 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4550 u64 start, u64 len, int delalloc)
4552 struct btrfs_block_group *cache;
4554 cache = btrfs_lookup_block_group(fs_info, start);
4556 btrfs_err(fs_info, "Unable to find block group for %llu",
4561 btrfs_add_free_space(cache, start, len);
4562 btrfs_free_reserved_bytes(cache, len, delalloc);
4563 trace_btrfs_reserved_extent_free(fs_info, start, len);
4565 btrfs_put_block_group(cache);
4569 int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start,
4572 struct btrfs_block_group *cache;
4575 cache = btrfs_lookup_block_group(trans->fs_info, start);
4577 btrfs_err(trans->fs_info, "unable to find block group for %llu",
4582 ret = pin_down_extent(trans, cache, start, len, 1);
4583 btrfs_put_block_group(cache);
4587 static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr,
4590 struct btrfs_fs_info *fs_info = trans->fs_info;
4593 ret = remove_from_free_space_tree(trans, bytenr, num_bytes);
4597 ret = btrfs_update_block_group(trans, bytenr, num_bytes, true);
4600 btrfs_err(fs_info, "update block group failed for %llu %llu",
4605 trace_btrfs_reserved_extent_alloc(fs_info, bytenr, num_bytes);
4609 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4610 u64 parent, u64 root_objectid,
4611 u64 flags, u64 owner, u64 offset,
4612 struct btrfs_key *ins, int ref_mod)
4614 struct btrfs_fs_info *fs_info = trans->fs_info;
4615 struct btrfs_root *extent_root;
4617 struct btrfs_extent_item *extent_item;
4618 struct btrfs_extent_inline_ref *iref;
4619 struct btrfs_path *path;
4620 struct extent_buffer *leaf;
4625 type = BTRFS_SHARED_DATA_REF_KEY;
4627 type = BTRFS_EXTENT_DATA_REF_KEY;
4629 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4631 path = btrfs_alloc_path();
4635 extent_root = btrfs_extent_root(fs_info, ins->objectid);
4636 ret = btrfs_insert_empty_item(trans, extent_root, path, ins, size);
4638 btrfs_free_path(path);
4642 leaf = path->nodes[0];
4643 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4644 struct btrfs_extent_item);
4645 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4646 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4647 btrfs_set_extent_flags(leaf, extent_item,
4648 flags | BTRFS_EXTENT_FLAG_DATA);
4650 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4651 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4653 struct btrfs_shared_data_ref *ref;
4654 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4655 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4656 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4658 struct btrfs_extent_data_ref *ref;
4659 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4660 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4661 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4662 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4663 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4666 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
4667 btrfs_free_path(path);
4669 return alloc_reserved_extent(trans, ins->objectid, ins->offset);
4672 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4673 struct btrfs_delayed_ref_node *node,
4674 struct btrfs_delayed_extent_op *extent_op)
4676 struct btrfs_fs_info *fs_info = trans->fs_info;
4677 struct btrfs_root *extent_root;
4679 struct btrfs_extent_item *extent_item;
4680 struct btrfs_key extent_key;
4681 struct btrfs_tree_block_info *block_info;
4682 struct btrfs_extent_inline_ref *iref;
4683 struct btrfs_path *path;
4684 struct extent_buffer *leaf;
4685 struct btrfs_delayed_tree_ref *ref;
4686 u32 size = sizeof(*extent_item) + sizeof(*iref);
4687 u64 flags = extent_op->flags_to_set;
4688 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4690 ref = btrfs_delayed_node_to_tree_ref(node);
4692 extent_key.objectid = node->bytenr;
4693 if (skinny_metadata) {
4694 extent_key.offset = ref->level;
4695 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4697 extent_key.offset = node->num_bytes;
4698 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4699 size += sizeof(*block_info);
4702 path = btrfs_alloc_path();
4706 extent_root = btrfs_extent_root(fs_info, extent_key.objectid);
4707 ret = btrfs_insert_empty_item(trans, extent_root, path, &extent_key,
4710 btrfs_free_path(path);
4714 leaf = path->nodes[0];
4715 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4716 struct btrfs_extent_item);
4717 btrfs_set_extent_refs(leaf, extent_item, 1);
4718 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4719 btrfs_set_extent_flags(leaf, extent_item,
4720 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4722 if (skinny_metadata) {
4723 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4725 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4726 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4727 btrfs_set_tree_block_level(leaf, block_info, ref->level);
4728 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4731 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4732 btrfs_set_extent_inline_ref_type(leaf, iref,
4733 BTRFS_SHARED_BLOCK_REF_KEY);
4734 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4736 btrfs_set_extent_inline_ref_type(leaf, iref,
4737 BTRFS_TREE_BLOCK_REF_KEY);
4738 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4741 btrfs_mark_buffer_dirty(trans, leaf);
4742 btrfs_free_path(path);
4744 return alloc_reserved_extent(trans, node->bytenr, fs_info->nodesize);
4747 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4748 struct btrfs_root *root, u64 owner,
4749 u64 offset, u64 ram_bytes,
4750 struct btrfs_key *ins)
4752 struct btrfs_ref generic_ref = { 0 };
4754 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4756 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4757 ins->objectid, ins->offset, 0);
4758 btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner,
4760 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4762 return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes);
4766 * this is used by the tree logging recovery code. It records that
4767 * an extent has been allocated and makes sure to clear the free
4768 * space cache bits as well
4770 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4771 u64 root_objectid, u64 owner, u64 offset,
4772 struct btrfs_key *ins)
4774 struct btrfs_fs_info *fs_info = trans->fs_info;
4776 struct btrfs_block_group *block_group;
4777 struct btrfs_space_info *space_info;
4780 * Mixed block groups will exclude before processing the log so we only
4781 * need to do the exclude dance if this fs isn't mixed.
4783 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4784 ret = __exclude_logged_extent(fs_info, ins->objectid,
4790 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4794 space_info = block_group->space_info;
4795 spin_lock(&space_info->lock);
4796 spin_lock(&block_group->lock);
4797 space_info->bytes_reserved += ins->offset;
4798 block_group->reserved += ins->offset;
4799 spin_unlock(&block_group->lock);
4800 spin_unlock(&space_info->lock);
4802 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4805 btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
4806 btrfs_put_block_group(block_group);
4810 static struct extent_buffer *
4811 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4812 u64 bytenr, int level, u64 owner,
4813 enum btrfs_lock_nesting nest)
4815 struct btrfs_fs_info *fs_info = root->fs_info;
4816 struct extent_buffer *buf;
4817 u64 lockdep_owner = owner;
4819 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level);
4824 * Extra safety check in case the extent tree is corrupted and extent
4825 * allocator chooses to use a tree block which is already used and
4828 if (buf->lock_owner == current->pid) {
4829 btrfs_err_rl(fs_info,
4830 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4831 buf->start, btrfs_header_owner(buf), current->pid);
4832 free_extent_buffer(buf);
4833 return ERR_PTR(-EUCLEAN);
4837 * The reloc trees are just snapshots, so we need them to appear to be
4838 * just like any other fs tree WRT lockdep.
4840 * The exception however is in replace_path() in relocation, where we
4841 * hold the lock on the original fs root and then search for the reloc
4842 * root. At that point we need to make sure any reloc root buffers are
4843 * set to the BTRFS_TREE_RELOC_OBJECTID lockdep class in order to make
4846 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID &&
4847 !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state))
4848 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
4850 /* btrfs_clear_buffer_dirty() accesses generation field. */
4851 btrfs_set_header_generation(buf, trans->transid);
4854 * This needs to stay, because we could allocate a freed block from an
4855 * old tree into a new tree, so we need to make sure this new block is
4856 * set to the appropriate level and owner.
4858 btrfs_set_buffer_lockdep_class(lockdep_owner, buf, level);
4860 __btrfs_tree_lock(buf, nest);
4861 btrfs_clear_buffer_dirty(trans, buf);
4862 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
4863 clear_bit(EXTENT_BUFFER_NO_CHECK, &buf->bflags);
4865 set_extent_buffer_uptodate(buf);
4867 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
4868 btrfs_set_header_level(buf, level);
4869 btrfs_set_header_bytenr(buf, buf->start);
4870 btrfs_set_header_generation(buf, trans->transid);
4871 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
4872 btrfs_set_header_owner(buf, owner);
4873 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
4874 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
4875 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4876 buf->log_index = root->log_transid % 2;
4878 * we allow two log transactions at a time, use different
4879 * EXTENT bit to differentiate dirty pages.
4881 if (buf->log_index == 0)
4882 set_extent_bit(&root->dirty_log_pages, buf->start,
4883 buf->start + buf->len - 1,
4884 EXTENT_DIRTY, NULL);
4886 set_extent_bit(&root->dirty_log_pages, buf->start,
4887 buf->start + buf->len - 1,
4890 buf->log_index = -1;
4891 set_extent_bit(&trans->transaction->dirty_pages, buf->start,
4892 buf->start + buf->len - 1, EXTENT_DIRTY, NULL);
4894 /* this returns a buffer locked for blocking */
4899 * finds a free extent and does all the dirty work required for allocation
4900 * returns the tree buffer or an ERR_PTR on error.
4902 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
4903 struct btrfs_root *root,
4904 u64 parent, u64 root_objectid,
4905 const struct btrfs_disk_key *key,
4906 int level, u64 hint,
4908 enum btrfs_lock_nesting nest)
4910 struct btrfs_fs_info *fs_info = root->fs_info;
4911 struct btrfs_key ins;
4912 struct btrfs_block_rsv *block_rsv;
4913 struct extent_buffer *buf;
4914 struct btrfs_delayed_extent_op *extent_op;
4915 struct btrfs_ref generic_ref = { 0 };
4918 u32 blocksize = fs_info->nodesize;
4919 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4921 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4922 if (btrfs_is_testing(fs_info)) {
4923 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
4924 level, root_objectid, nest);
4926 root->alloc_bytenr += blocksize;
4931 block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
4932 if (IS_ERR(block_rsv))
4933 return ERR_CAST(block_rsv);
4935 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
4936 empty_size, hint, &ins, 0, 0);
4940 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
4941 root_objectid, nest);
4944 goto out_free_reserved;
4947 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4949 parent = ins.objectid;
4950 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4954 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4955 extent_op = btrfs_alloc_delayed_extent_op();
4961 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4963 memset(&extent_op->key, 0, sizeof(extent_op->key));
4964 extent_op->flags_to_set = flags;
4965 extent_op->update_key = skinny_metadata ? false : true;
4966 extent_op->update_flags = true;
4967 extent_op->level = level;
4969 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4970 ins.objectid, ins.offset, parent);
4971 btrfs_init_tree_ref(&generic_ref, level, root_objectid,
4972 root->root_key.objectid, false);
4973 btrfs_ref_tree_mod(fs_info, &generic_ref);
4974 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op);
4976 goto out_free_delayed;
4981 btrfs_free_delayed_extent_op(extent_op);
4983 btrfs_tree_unlock(buf);
4984 free_extent_buffer(buf);
4986 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
4988 btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
4989 return ERR_PTR(ret);
4992 struct walk_control {
4993 u64 refs[BTRFS_MAX_LEVEL];
4994 u64 flags[BTRFS_MAX_LEVEL];
4995 struct btrfs_key update_progress;
4996 struct btrfs_key drop_progress;
5008 #define DROP_REFERENCE 1
5009 #define UPDATE_BACKREF 2
5011 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5012 struct btrfs_root *root,
5013 struct walk_control *wc,
5014 struct btrfs_path *path)
5016 struct btrfs_fs_info *fs_info = root->fs_info;
5022 struct btrfs_key key;
5023 struct extent_buffer *eb;
5028 if (path->slots[wc->level] < wc->reada_slot) {
5029 wc->reada_count = wc->reada_count * 2 / 3;
5030 wc->reada_count = max(wc->reada_count, 2);
5032 wc->reada_count = wc->reada_count * 3 / 2;
5033 wc->reada_count = min_t(int, wc->reada_count,
5034 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
5037 eb = path->nodes[wc->level];
5038 nritems = btrfs_header_nritems(eb);
5040 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5041 if (nread >= wc->reada_count)
5045 bytenr = btrfs_node_blockptr(eb, slot);
5046 generation = btrfs_node_ptr_generation(eb, slot);
5048 if (slot == path->slots[wc->level])
5051 if (wc->stage == UPDATE_BACKREF &&
5052 generation <= root->root_key.offset)
5055 /* We don't lock the tree block, it's OK to be racy here */
5056 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
5057 wc->level - 1, 1, &refs,
5059 /* We don't care about errors in readahead. */
5064 if (wc->stage == DROP_REFERENCE) {
5068 if (wc->level == 1 &&
5069 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5071 if (!wc->update_ref ||
5072 generation <= root->root_key.offset)
5074 btrfs_node_key_to_cpu(eb, &key, slot);
5075 ret = btrfs_comp_cpu_keys(&key,
5076 &wc->update_progress);
5080 if (wc->level == 1 &&
5081 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5085 btrfs_readahead_node_child(eb, slot);
5088 wc->reada_slot = slot;
5092 * helper to process tree block while walking down the tree.
5094 * when wc->stage == UPDATE_BACKREF, this function updates
5095 * back refs for pointers in the block.
5097 * NOTE: return value 1 means we should stop walking down.
5099 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5100 struct btrfs_root *root,
5101 struct btrfs_path *path,
5102 struct walk_control *wc, int lookup_info)
5104 struct btrfs_fs_info *fs_info = root->fs_info;
5105 int level = wc->level;
5106 struct extent_buffer *eb = path->nodes[level];
5107 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5110 if (wc->stage == UPDATE_BACKREF &&
5111 btrfs_header_owner(eb) != root->root_key.objectid)
5115 * when reference count of tree block is 1, it won't increase
5116 * again. once full backref flag is set, we never clear it.
5119 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5120 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5121 BUG_ON(!path->locks[level]);
5122 ret = btrfs_lookup_extent_info(trans, fs_info,
5123 eb->start, level, 1,
5126 BUG_ON(ret == -ENOMEM);
5129 BUG_ON(wc->refs[level] == 0);
5132 if (wc->stage == DROP_REFERENCE) {
5133 if (wc->refs[level] > 1)
5136 if (path->locks[level] && !wc->keep_locks) {
5137 btrfs_tree_unlock_rw(eb, path->locks[level]);
5138 path->locks[level] = 0;
5143 /* wc->stage == UPDATE_BACKREF */
5144 if (!(wc->flags[level] & flag)) {
5145 BUG_ON(!path->locks[level]);
5146 ret = btrfs_inc_ref(trans, root, eb, 1);
5147 BUG_ON(ret); /* -ENOMEM */
5148 ret = btrfs_dec_ref(trans, root, eb, 0);
5149 BUG_ON(ret); /* -ENOMEM */
5150 ret = btrfs_set_disk_extent_flags(trans, eb, flag);
5151 BUG_ON(ret); /* -ENOMEM */
5152 wc->flags[level] |= flag;
5156 * the block is shared by multiple trees, so it's not good to
5157 * keep the tree lock
5159 if (path->locks[level] && level > 0) {
5160 btrfs_tree_unlock_rw(eb, path->locks[level]);
5161 path->locks[level] = 0;
5167 * This is used to verify a ref exists for this root to deal with a bug where we
5168 * would have a drop_progress key that hadn't been updated properly.
5170 static int check_ref_exists(struct btrfs_trans_handle *trans,
5171 struct btrfs_root *root, u64 bytenr, u64 parent,
5174 struct btrfs_path *path;
5175 struct btrfs_extent_inline_ref *iref;
5178 path = btrfs_alloc_path();
5182 ret = lookup_extent_backref(trans, path, &iref, bytenr,
5183 root->fs_info->nodesize, parent,
5184 root->root_key.objectid, level, 0);
5185 btrfs_free_path(path);
5194 * helper to process tree block pointer.
5196 * when wc->stage == DROP_REFERENCE, this function checks
5197 * reference count of the block pointed to. if the block
5198 * is shared and we need update back refs for the subtree
5199 * rooted at the block, this function changes wc->stage to
5200 * UPDATE_BACKREF. if the block is shared and there is no
5201 * need to update back, this function drops the reference
5204 * NOTE: return value 1 means we should stop walking down.
5206 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5207 struct btrfs_root *root,
5208 struct btrfs_path *path,
5209 struct walk_control *wc, int *lookup_info)
5211 struct btrfs_fs_info *fs_info = root->fs_info;
5215 struct btrfs_tree_parent_check check = { 0 };
5216 struct btrfs_key key;
5217 struct btrfs_ref ref = { 0 };
5218 struct extent_buffer *next;
5219 int level = wc->level;
5222 bool need_account = false;
5224 generation = btrfs_node_ptr_generation(path->nodes[level],
5225 path->slots[level]);
5227 * if the lower level block was created before the snapshot
5228 * was created, we know there is no need to update back refs
5231 if (wc->stage == UPDATE_BACKREF &&
5232 generation <= root->root_key.offset) {
5237 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5239 check.level = level - 1;
5240 check.transid = generation;
5241 check.owner_root = root->root_key.objectid;
5242 check.has_first_key = true;
5243 btrfs_node_key_to_cpu(path->nodes[level], &check.first_key,
5244 path->slots[level]);
5246 next = find_extent_buffer(fs_info, bytenr);
5248 next = btrfs_find_create_tree_block(fs_info, bytenr,
5249 root->root_key.objectid, level - 1);
5251 return PTR_ERR(next);
5254 btrfs_tree_lock(next);
5256 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
5257 &wc->refs[level - 1],
5258 &wc->flags[level - 1]);
5262 if (unlikely(wc->refs[level - 1] == 0)) {
5263 btrfs_err(fs_info, "Missing references.");
5269 if (wc->stage == DROP_REFERENCE) {
5270 if (wc->refs[level - 1] > 1) {
5271 need_account = true;
5273 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5276 if (!wc->update_ref ||
5277 generation <= root->root_key.offset)
5280 btrfs_node_key_to_cpu(path->nodes[level], &key,
5281 path->slots[level]);
5282 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5286 wc->stage = UPDATE_BACKREF;
5287 wc->shared_level = level - 1;
5291 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5295 if (!btrfs_buffer_uptodate(next, generation, 0)) {
5296 btrfs_tree_unlock(next);
5297 free_extent_buffer(next);
5303 if (reada && level == 1)
5304 reada_walk_down(trans, root, wc, path);
5305 next = read_tree_block(fs_info, bytenr, &check);
5307 return PTR_ERR(next);
5308 } else if (!extent_buffer_uptodate(next)) {
5309 free_extent_buffer(next);
5312 btrfs_tree_lock(next);
5316 ASSERT(level == btrfs_header_level(next));
5317 if (level != btrfs_header_level(next)) {
5318 btrfs_err(root->fs_info, "mismatched level");
5322 path->nodes[level] = next;
5323 path->slots[level] = 0;
5324 path->locks[level] = BTRFS_WRITE_LOCK;
5330 wc->refs[level - 1] = 0;
5331 wc->flags[level - 1] = 0;
5332 if (wc->stage == DROP_REFERENCE) {
5333 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5334 parent = path->nodes[level]->start;
5336 ASSERT(root->root_key.objectid ==
5337 btrfs_header_owner(path->nodes[level]));
5338 if (root->root_key.objectid !=
5339 btrfs_header_owner(path->nodes[level])) {
5340 btrfs_err(root->fs_info,
5341 "mismatched block owner");
5349 * If we had a drop_progress we need to verify the refs are set
5350 * as expected. If we find our ref then we know that from here
5351 * on out everything should be correct, and we can clear the
5354 if (wc->restarted) {
5355 ret = check_ref_exists(trans, root, bytenr, parent,
5366 * Reloc tree doesn't contribute to qgroup numbers, and we have
5367 * already accounted them at merge time (replace_path),
5368 * thus we could skip expensive subtree trace here.
5370 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
5372 ret = btrfs_qgroup_trace_subtree(trans, next,
5373 generation, level - 1);
5375 btrfs_err_rl(fs_info,
5376 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5382 * We need to update the next key in our walk control so we can
5383 * update the drop_progress key accordingly. We don't care if
5384 * find_next_key doesn't find a key because that means we're at
5385 * the end and are going to clean up now.
5387 wc->drop_level = level;
5388 find_next_key(path, level, &wc->drop_progress);
5390 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
5391 fs_info->nodesize, parent);
5392 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid,
5394 ret = btrfs_free_extent(trans, &ref);
5403 btrfs_tree_unlock(next);
5404 free_extent_buffer(next);
5410 * helper to process tree block while walking up the tree.
5412 * when wc->stage == DROP_REFERENCE, this function drops
5413 * reference count on the block.
5415 * when wc->stage == UPDATE_BACKREF, this function changes
5416 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5417 * to UPDATE_BACKREF previously while processing the block.
5419 * NOTE: return value 1 means we should stop walking up.
5421 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5422 struct btrfs_root *root,
5423 struct btrfs_path *path,
5424 struct walk_control *wc)
5426 struct btrfs_fs_info *fs_info = root->fs_info;
5428 int level = wc->level;
5429 struct extent_buffer *eb = path->nodes[level];
5432 if (wc->stage == UPDATE_BACKREF) {
5433 BUG_ON(wc->shared_level < level);
5434 if (level < wc->shared_level)
5437 ret = find_next_key(path, level + 1, &wc->update_progress);
5441 wc->stage = DROP_REFERENCE;
5442 wc->shared_level = -1;
5443 path->slots[level] = 0;
5446 * check reference count again if the block isn't locked.
5447 * we should start walking down the tree again if reference
5450 if (!path->locks[level]) {
5452 btrfs_tree_lock(eb);
5453 path->locks[level] = BTRFS_WRITE_LOCK;
5455 ret = btrfs_lookup_extent_info(trans, fs_info,
5456 eb->start, level, 1,
5460 btrfs_tree_unlock_rw(eb, path->locks[level]);
5461 path->locks[level] = 0;
5464 BUG_ON(wc->refs[level] == 0);
5465 if (wc->refs[level] == 1) {
5466 btrfs_tree_unlock_rw(eb, path->locks[level]);
5467 path->locks[level] = 0;
5473 /* wc->stage == DROP_REFERENCE */
5474 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5476 if (wc->refs[level] == 1) {
5478 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5479 ret = btrfs_dec_ref(trans, root, eb, 1);
5481 ret = btrfs_dec_ref(trans, root, eb, 0);
5482 BUG_ON(ret); /* -ENOMEM */
5483 if (is_fstree(root->root_key.objectid)) {
5484 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5486 btrfs_err_rl(fs_info,
5487 "error %d accounting leaf items, quota is out of sync, rescan required",
5492 /* Make block locked assertion in btrfs_clear_buffer_dirty happy. */
5493 if (!path->locks[level]) {
5494 btrfs_tree_lock(eb);
5495 path->locks[level] = BTRFS_WRITE_LOCK;
5497 btrfs_clear_buffer_dirty(trans, eb);
5500 if (eb == root->node) {
5501 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5503 else if (root->root_key.objectid != btrfs_header_owner(eb))
5504 goto owner_mismatch;
5506 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5507 parent = path->nodes[level + 1]->start;
5508 else if (root->root_key.objectid !=
5509 btrfs_header_owner(path->nodes[level + 1]))
5510 goto owner_mismatch;
5513 btrfs_free_tree_block(trans, btrfs_root_id(root), eb, parent,
5514 wc->refs[level] == 1);
5516 wc->refs[level] = 0;
5517 wc->flags[level] = 0;
5521 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5522 btrfs_header_owner(eb), root->root_key.objectid);
5526 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5527 struct btrfs_root *root,
5528 struct btrfs_path *path,
5529 struct walk_control *wc)
5531 int level = wc->level;
5532 int lookup_info = 1;
5535 while (level >= 0) {
5536 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5543 if (path->slots[level] >=
5544 btrfs_header_nritems(path->nodes[level]))
5547 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5549 path->slots[level]++;
5555 return (ret == 1) ? 0 : ret;
5558 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5559 struct btrfs_root *root,
5560 struct btrfs_path *path,
5561 struct walk_control *wc, int max_level)
5563 int level = wc->level;
5566 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5567 while (level < max_level && path->nodes[level]) {
5569 if (path->slots[level] + 1 <
5570 btrfs_header_nritems(path->nodes[level])) {
5571 path->slots[level]++;
5574 ret = walk_up_proc(trans, root, path, wc);
5580 if (path->locks[level]) {
5581 btrfs_tree_unlock_rw(path->nodes[level],
5582 path->locks[level]);
5583 path->locks[level] = 0;
5585 free_extent_buffer(path->nodes[level]);
5586 path->nodes[level] = NULL;
5594 * drop a subvolume tree.
5596 * this function traverses the tree freeing any blocks that only
5597 * referenced by the tree.
5599 * when a shared tree block is found. this function decreases its
5600 * reference count by one. if update_ref is true, this function
5601 * also make sure backrefs for the shared block and all lower level
5602 * blocks are properly updated.
5604 * If called with for_reloc == 0, may exit early with -EAGAIN
5606 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
5608 const bool is_reloc_root = (root->root_key.objectid ==
5609 BTRFS_TREE_RELOC_OBJECTID);
5610 struct btrfs_fs_info *fs_info = root->fs_info;
5611 struct btrfs_path *path;
5612 struct btrfs_trans_handle *trans;
5613 struct btrfs_root *tree_root = fs_info->tree_root;
5614 struct btrfs_root_item *root_item = &root->root_item;
5615 struct walk_control *wc;
5616 struct btrfs_key key;
5620 bool root_dropped = false;
5621 bool unfinished_drop = false;
5623 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5625 path = btrfs_alloc_path();
5631 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5633 btrfs_free_path(path);
5639 * Use join to avoid potential EINTR from transaction start. See
5640 * wait_reserve_ticket and the whole reservation callchain.
5643 trans = btrfs_join_transaction(tree_root);
5645 trans = btrfs_start_transaction(tree_root, 0);
5646 if (IS_ERR(trans)) {
5647 err = PTR_ERR(trans);
5651 err = btrfs_run_delayed_items(trans);
5656 * This will help us catch people modifying the fs tree while we're
5657 * dropping it. It is unsafe to mess with the fs tree while it's being
5658 * dropped as we unlock the root node and parent nodes as we walk down
5659 * the tree, assuming nothing will change. If something does change
5660 * then we'll have stale information and drop references to blocks we've
5663 set_bit(BTRFS_ROOT_DELETING, &root->state);
5664 unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
5666 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5667 level = btrfs_header_level(root->node);
5668 path->nodes[level] = btrfs_lock_root_node(root);
5669 path->slots[level] = 0;
5670 path->locks[level] = BTRFS_WRITE_LOCK;
5671 memset(&wc->update_progress, 0,
5672 sizeof(wc->update_progress));
5674 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5675 memcpy(&wc->update_progress, &key,
5676 sizeof(wc->update_progress));
5678 level = btrfs_root_drop_level(root_item);
5680 path->lowest_level = level;
5681 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5682 path->lowest_level = 0;
5690 * unlock our path, this is safe because only this
5691 * function is allowed to delete this snapshot
5693 btrfs_unlock_up_safe(path, 0);
5695 level = btrfs_header_level(root->node);
5697 btrfs_tree_lock(path->nodes[level]);
5698 path->locks[level] = BTRFS_WRITE_LOCK;
5700 ret = btrfs_lookup_extent_info(trans, fs_info,
5701 path->nodes[level]->start,
5702 level, 1, &wc->refs[level],
5708 BUG_ON(wc->refs[level] == 0);
5710 if (level == btrfs_root_drop_level(root_item))
5713 btrfs_tree_unlock(path->nodes[level]);
5714 path->locks[level] = 0;
5715 WARN_ON(wc->refs[level] != 1);
5720 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5722 wc->shared_level = -1;
5723 wc->stage = DROP_REFERENCE;
5724 wc->update_ref = update_ref;
5726 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5730 ret = walk_down_tree(trans, root, path, wc);
5732 btrfs_abort_transaction(trans, ret);
5737 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5739 btrfs_abort_transaction(trans, ret);
5745 BUG_ON(wc->stage != DROP_REFERENCE);
5749 if (wc->stage == DROP_REFERENCE) {
5750 wc->drop_level = wc->level;
5751 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5753 path->slots[wc->drop_level]);
5755 btrfs_cpu_key_to_disk(&root_item->drop_progress,
5756 &wc->drop_progress);
5757 btrfs_set_root_drop_level(root_item, wc->drop_level);
5759 BUG_ON(wc->level == 0);
5760 if (btrfs_should_end_transaction(trans) ||
5761 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5762 ret = btrfs_update_root(trans, tree_root,
5766 btrfs_abort_transaction(trans, ret);
5772 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
5774 btrfs_end_transaction_throttle(trans);
5775 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5776 btrfs_debug(fs_info,
5777 "drop snapshot early exit");
5783 * Use join to avoid potential EINTR from transaction
5784 * start. See wait_reserve_ticket and the whole
5785 * reservation callchain.
5788 trans = btrfs_join_transaction(tree_root);
5790 trans = btrfs_start_transaction(tree_root, 0);
5791 if (IS_ERR(trans)) {
5792 err = PTR_ERR(trans);
5797 btrfs_release_path(path);
5801 ret = btrfs_del_root(trans, &root->root_key);
5803 btrfs_abort_transaction(trans, ret);
5808 if (!is_reloc_root) {
5809 ret = btrfs_find_root(tree_root, &root->root_key, path,
5812 btrfs_abort_transaction(trans, ret);
5815 } else if (ret > 0) {
5816 /* if we fail to delete the orphan item this time
5817 * around, it'll get picked up the next time.
5819 * The most common failure here is just -ENOENT.
5821 btrfs_del_orphan_item(trans, tree_root,
5822 root->root_key.objectid);
5827 * This subvolume is going to be completely dropped, and won't be
5828 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
5829 * commit transaction time. So free it here manually.
5831 btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
5832 btrfs_qgroup_free_meta_all_pertrans(root);
5834 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state))
5835 btrfs_add_dropped_root(trans, root);
5837 btrfs_put_root(root);
5838 root_dropped = true;
5841 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
5843 btrfs_end_transaction_throttle(trans);
5846 btrfs_free_path(path);
5849 * We were an unfinished drop root, check to see if there are any
5850 * pending, and if not clear and wake up any waiters.
5852 if (!err && unfinished_drop)
5853 btrfs_maybe_wake_unfinished_drop(fs_info);
5856 * So if we need to stop dropping the snapshot for whatever reason we
5857 * need to make sure to add it back to the dead root list so that we
5858 * keep trying to do the work later. This also cleans up roots if we
5859 * don't have it in the radix (like when we recover after a power fail
5860 * or unmount) so we don't leak memory.
5862 if (!for_reloc && !root_dropped)
5863 btrfs_add_dead_root(root);
5868 * drop subtree rooted at tree block 'node'.
5870 * NOTE: this function will unlock and release tree block 'node'
5871 * only used by relocation code
5873 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5874 struct btrfs_root *root,
5875 struct extent_buffer *node,
5876 struct extent_buffer *parent)
5878 struct btrfs_fs_info *fs_info = root->fs_info;
5879 struct btrfs_path *path;
5880 struct walk_control *wc;
5886 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5888 path = btrfs_alloc_path();
5892 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5894 btrfs_free_path(path);
5898 btrfs_assert_tree_write_locked(parent);
5899 parent_level = btrfs_header_level(parent);
5900 atomic_inc(&parent->refs);
5901 path->nodes[parent_level] = parent;
5902 path->slots[parent_level] = btrfs_header_nritems(parent);
5904 btrfs_assert_tree_write_locked(node);
5905 level = btrfs_header_level(node);
5906 path->nodes[level] = node;
5907 path->slots[level] = 0;
5908 path->locks[level] = BTRFS_WRITE_LOCK;
5910 wc->refs[parent_level] = 1;
5911 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5913 wc->shared_level = -1;
5914 wc->stage = DROP_REFERENCE;
5917 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5920 wret = walk_down_tree(trans, root, path, wc);
5926 wret = walk_up_tree(trans, root, path, wc, parent_level);
5934 btrfs_free_path(path);
5938 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
5941 return unpin_extent_range(fs_info, start, end, false);
5945 * It used to be that old block groups would be left around forever.
5946 * Iterating over them would be enough to trim unused space. Since we
5947 * now automatically remove them, we also need to iterate over unallocated
5950 * We don't want a transaction for this since the discard may take a
5951 * substantial amount of time. We don't require that a transaction be
5952 * running, but we do need to take a running transaction into account
5953 * to ensure that we're not discarding chunks that were released or
5954 * allocated in the current transaction.
5956 * Holding the chunks lock will prevent other threads from allocating
5957 * or releasing chunks, but it won't prevent a running transaction
5958 * from committing and releasing the memory that the pending chunks
5959 * list head uses. For that, we need to take a reference to the
5960 * transaction and hold the commit root sem. We only need to hold
5961 * it while performing the free space search since we have already
5962 * held back allocations.
5964 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
5966 u64 start = BTRFS_DEVICE_RANGE_RESERVED, len = 0, end = 0;
5971 /* Discard not supported = nothing to do. */
5972 if (!bdev_max_discard_sectors(device->bdev))
5975 /* Not writable = nothing to do. */
5976 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5979 /* No free space = nothing to do. */
5980 if (device->total_bytes <= device->bytes_used)
5986 struct btrfs_fs_info *fs_info = device->fs_info;
5989 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
5993 find_first_clear_extent_bit(&device->alloc_state, start,
5995 CHUNK_TRIMMED | CHUNK_ALLOCATED);
5997 /* Check if there are any CHUNK_* bits left */
5998 if (start > device->total_bytes) {
5999 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6000 btrfs_warn_in_rcu(fs_info,
6001 "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
6002 start, end - start + 1,
6003 btrfs_dev_name(device),
6004 device->total_bytes);
6005 mutex_unlock(&fs_info->chunk_mutex);
6010 /* Ensure we skip the reserved space on each device. */
6011 start = max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
6014 * If find_first_clear_extent_bit find a range that spans the
6015 * end of the device it will set end to -1, in this case it's up
6016 * to the caller to trim the value to the size of the device.
6018 end = min(end, device->total_bytes - 1);
6020 len = end - start + 1;
6022 /* We didn't find any extents */
6024 mutex_unlock(&fs_info->chunk_mutex);
6029 ret = btrfs_issue_discard(device->bdev, start, len,
6032 set_extent_bit(&device->alloc_state, start,
6033 start + bytes - 1, CHUNK_TRIMMED, NULL);
6034 mutex_unlock(&fs_info->chunk_mutex);
6042 if (fatal_signal_pending(current)) {
6054 * Trim the whole filesystem by:
6055 * 1) trimming the free space in each block group
6056 * 2) trimming the unallocated space on each device
6058 * This will also continue trimming even if a block group or device encounters
6059 * an error. The return value will be the last error, or 0 if nothing bad
6062 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
6064 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6065 struct btrfs_block_group *cache = NULL;
6066 struct btrfs_device *device;
6068 u64 range_end = U64_MAX;
6078 if (range->start == U64_MAX)
6082 * Check range overflow if range->len is set.
6083 * The default range->len is U64_MAX.
6085 if (range->len != U64_MAX &&
6086 check_add_overflow(range->start, range->len, &range_end))
6089 cache = btrfs_lookup_first_block_group(fs_info, range->start);
6090 for (; cache; cache = btrfs_next_block_group(cache)) {
6091 if (cache->start >= range_end) {
6092 btrfs_put_block_group(cache);
6096 start = max(range->start, cache->start);
6097 end = min(range_end, cache->start + cache->length);
6099 if (end - start >= range->minlen) {
6100 if (!btrfs_block_group_done(cache)) {
6101 ret = btrfs_cache_block_group(cache, true);
6108 ret = btrfs_trim_block_group(cache,
6114 trimmed += group_trimmed;
6125 "failed to trim %llu block group(s), last error %d",
6128 mutex_lock(&fs_devices->device_list_mutex);
6129 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6130 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
6133 ret = btrfs_trim_free_extents(device, &group_trimmed);
6140 trimmed += group_trimmed;
6142 mutex_unlock(&fs_devices->device_list_mutex);
6146 "failed to trim %llu device(s), last error %d",
6147 dev_failed, dev_ret);
6148 range->len = trimmed;