1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Copyright (C) 2008 Oracle. All rights reserved.
6 #ifndef BTRFS_DELAYED_REF_H
7 #define BTRFS_DELAYED_REF_H
9 #include <linux/refcount.h>
11 /* these are the possible values of struct btrfs_delayed_ref_node->action */
12 enum btrfs_delayed_ref_action {
13 /* Add one backref to the tree */
14 BTRFS_ADD_DELAYED_REF = 1,
15 /* Delete one backref from the tree */
16 BTRFS_DROP_DELAYED_REF,
17 /* Record a full extent allocation */
18 BTRFS_ADD_DELAYED_EXTENT,
19 /* Not changing ref count on head ref */
20 BTRFS_UPDATE_DELAYED_HEAD,
23 struct btrfs_delayed_ref_node {
24 struct rb_node ref_node;
26 * If action is BTRFS_ADD_DELAYED_REF, also link this node to
27 * ref_head->ref_add_list, then we do not need to iterate the
28 * whole ref_head->ref_list to find BTRFS_ADD_DELAYED_REF nodes.
30 struct list_head add_list;
32 /* the starting bytenr of the extent */
35 /* the size of the extent */
38 /* seq number to keep track of insertion order */
41 /* ref count on this data structure */
45 * how many refs is this entry adding or deleting. For
46 * head refs, this may be a negative number because it is keeping
47 * track of the total mods done to the reference count.
48 * For individual refs, this will always be a positive number
50 * It may be more than one, since it is possible for a single
51 * parent to have more than one ref on an extent
55 unsigned int action:8;
59 struct btrfs_delayed_extent_op {
60 struct btrfs_disk_key key;
68 * the head refs are used to hold a lock on a given extent, which allows us
69 * to make sure that only one process is running the delayed refs
70 * at a time for a single extent. They also store the sum of all the
71 * reference count modifications we've queued up.
73 struct btrfs_delayed_ref_head {
77 * For insertion into struct btrfs_delayed_ref_root::href_root.
78 * Keep it in the same cache line as 'bytenr' for more efficient
79 * searches in the rbtree.
81 struct rb_node href_node;
83 * the mutex is held while running the refs, and it is also
84 * held when checking the sum of reference modifications.
90 /* Protects 'ref_tree' and 'ref_add_list'. */
92 struct rb_root_cached ref_tree;
93 /* accumulate add BTRFS_ADD_DELAYED_REF nodes to this ref_add_list. */
94 struct list_head ref_add_list;
96 struct btrfs_delayed_extent_op *extent_op;
99 * This is used to track the final ref_mod from all the refs associated
100 * with this head ref, this is not adjusted as delayed refs are run,
101 * this is meant to track if we need to do the csum accounting or not.
106 * This is the current outstanding mod references for this bytenr. This
107 * is used with lookup_extent_info to get an accurate reference count
108 * for a bytenr, so it is adjusted as delayed refs are run so that any
109 * on disk reference count + ref_mod is accurate.
114 * The root that triggered the allocation when must_insert_reserved is
120 * Track reserved bytes when setting must_insert_reserved. On success
121 * or cleanup, we will need to free the reservation.
126 * when a new extent is allocated, it is just reserved in memory
127 * The actual extent isn't inserted into the extent allocation tree
128 * until the delayed ref is processed. must_insert_reserved is
129 * used to flag a delayed ref so the accounting can be updated
130 * when a full insert is done.
132 * It is possible the extent will be freed before it is ever
133 * inserted into the extent allocation tree. In this case
134 * we need to update the in ram accounting to properly reflect
135 * the free has happened.
137 bool must_insert_reserved;
144 struct btrfs_delayed_tree_ref {
145 struct btrfs_delayed_ref_node node;
151 struct btrfs_delayed_data_ref {
152 struct btrfs_delayed_ref_node node;
159 enum btrfs_delayed_ref_flags {
160 /* Indicate that we are flushing delayed refs for the commit */
161 BTRFS_DELAYED_REFS_FLUSHING,
164 struct btrfs_delayed_ref_root {
165 /* head ref rbtree */
166 struct rb_root_cached href_root;
168 /* dirty extent records */
169 struct rb_root dirty_extent_root;
171 /* this spin lock protects the rbtree and the entries inside */
174 /* how many delayed ref updates we've queued, used by the
177 atomic_t num_entries;
179 /* total number of head nodes in tree */
180 unsigned long num_heads;
182 /* total number of head nodes ready for processing */
183 unsigned long num_heads_ready;
189 u64 run_delayed_start;
192 * To make qgroup to skip given root.
193 * This is for snapshot, as btrfs_qgroup_inherit() will manually
194 * modify counters for snapshot and its source, so we should skip
195 * the snapshot in new_root/old_roots or it will get calculated twice
200 enum btrfs_ref_type {
207 struct btrfs_data_ref {
208 /* For EXTENT_DATA_REF */
210 /* Root which owns this data reference. */
213 /* Inode which refers to this data extent */
217 * file_offset - extent_offset
219 * file_offset is the key.offset of the EXTENT_DATA key.
220 * extent_offset is btrfs_file_extent_offset() of the EXTENT_DATA data.
225 struct btrfs_tree_ref {
227 * Level of this tree block
229 * Shared for skinny (TREE_BLOCK_REF) and normal tree ref.
234 * Root which owns this tree block reference.
236 * For TREE_BLOCK_REF (skinny metadata, either inline or keyed)
240 /* For non-skinny metadata, no special member needed */
244 enum btrfs_ref_type type;
245 enum btrfs_delayed_ref_action action;
248 * Whether this extent should go through qgroup record.
250 * Normally false, but for certain cases like delayed subtree scan,
251 * setting this flag can hugely reduce qgroup overhead.
255 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
256 /* Through which root is this modification. */
263 /* Bytenr of the parent tree block */
266 struct btrfs_data_ref data_ref;
267 struct btrfs_tree_ref tree_ref;
271 extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
272 extern struct kmem_cache *btrfs_delayed_tree_ref_cachep;
273 extern struct kmem_cache *btrfs_delayed_data_ref_cachep;
274 extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
276 int __init btrfs_delayed_ref_init(void);
277 void __cold btrfs_delayed_ref_exit(void);
279 static inline u64 btrfs_calc_delayed_ref_bytes(const struct btrfs_fs_info *fs_info,
280 int num_delayed_refs)
284 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_delayed_refs);
287 * We have to check the mount option here because we could be enabling
288 * the free space tree for the first time and don't have the compat_ro
291 * We need extra reservations if we have the free space tree because
292 * we'll have to modify that tree as well.
294 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE))
300 static inline u64 btrfs_calc_delayed_ref_csum_bytes(const struct btrfs_fs_info *fs_info,
304 * Deleting csum items does not result in new nodes/leaves and does not
305 * require changing the free space tree, only the csum tree, so this is
308 return btrfs_calc_metadata_size(fs_info, num_csum_items);
311 static inline void btrfs_init_generic_ref(struct btrfs_ref *generic_ref,
312 int action, u64 bytenr, u64 len,
313 u64 parent, u64 owning_root)
315 generic_ref->action = action;
316 generic_ref->bytenr = bytenr;
317 generic_ref->len = len;
318 generic_ref->parent = parent;
319 generic_ref->owning_root = owning_root;
322 static inline void btrfs_init_tree_ref(struct btrfs_ref *generic_ref, int level,
323 u64 root, u64 mod_root, bool skip_qgroup)
325 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
326 /* If @real_root not set, use @root as fallback */
327 generic_ref->real_root = mod_root ?: root;
329 generic_ref->tree_ref.level = level;
330 generic_ref->tree_ref.ref_root = root;
331 generic_ref->type = BTRFS_REF_METADATA;
332 if (skip_qgroup || !(is_fstree(root) &&
333 (!mod_root || is_fstree(mod_root))))
334 generic_ref->skip_qgroup = true;
336 generic_ref->skip_qgroup = false;
340 static inline void btrfs_init_data_ref(struct btrfs_ref *generic_ref,
341 u64 ref_root, u64 ino, u64 offset, u64 mod_root,
344 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
345 /* If @real_root not set, use @root as fallback */
346 generic_ref->real_root = mod_root ?: ref_root;
348 generic_ref->data_ref.ref_root = ref_root;
349 generic_ref->data_ref.ino = ino;
350 generic_ref->data_ref.offset = offset;
351 generic_ref->type = BTRFS_REF_DATA;
352 if (skip_qgroup || !(is_fstree(ref_root) &&
353 (!mod_root || is_fstree(mod_root))))
354 generic_ref->skip_qgroup = true;
356 generic_ref->skip_qgroup = false;
359 static inline struct btrfs_delayed_extent_op *
360 btrfs_alloc_delayed_extent_op(void)
362 return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
366 btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
369 kmem_cache_free(btrfs_delayed_extent_op_cachep, op);
372 static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
374 if (refcount_dec_and_test(&ref->refs)) {
375 WARN_ON(!RB_EMPTY_NODE(&ref->ref_node));
377 case BTRFS_TREE_BLOCK_REF_KEY:
378 case BTRFS_SHARED_BLOCK_REF_KEY:
379 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
381 case BTRFS_EXTENT_DATA_REF_KEY:
382 case BTRFS_SHARED_DATA_REF_KEY:
383 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
391 static inline u64 btrfs_ref_head_to_space_flags(
392 struct btrfs_delayed_ref_head *head_ref)
394 if (head_ref->is_data)
395 return BTRFS_BLOCK_GROUP_DATA;
396 else if (head_ref->is_system)
397 return BTRFS_BLOCK_GROUP_SYSTEM;
398 return BTRFS_BLOCK_GROUP_METADATA;
401 static inline void btrfs_put_delayed_ref_head(struct btrfs_delayed_ref_head *head)
403 if (refcount_dec_and_test(&head->refs))
404 kmem_cache_free(btrfs_delayed_ref_head_cachep, head);
407 int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
408 struct btrfs_ref *generic_ref,
409 struct btrfs_delayed_extent_op *extent_op);
410 int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
411 struct btrfs_ref *generic_ref,
413 int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
414 u64 bytenr, u64 num_bytes,
415 struct btrfs_delayed_extent_op *extent_op);
416 void btrfs_merge_delayed_refs(struct btrfs_fs_info *fs_info,
417 struct btrfs_delayed_ref_root *delayed_refs,
418 struct btrfs_delayed_ref_head *head);
420 struct btrfs_delayed_ref_head *
421 btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
423 int btrfs_delayed_ref_lock(struct btrfs_delayed_ref_root *delayed_refs,
424 struct btrfs_delayed_ref_head *head);
425 static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
427 mutex_unlock(&head->mutex);
429 void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
430 struct btrfs_delayed_ref_head *head);
432 struct btrfs_delayed_ref_head *btrfs_select_ref_head(
433 struct btrfs_delayed_ref_root *delayed_refs);
435 int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq);
437 void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr_refs, int nr_csums);
438 void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans);
439 void btrfs_inc_delayed_refs_rsv_bg_inserts(struct btrfs_fs_info *fs_info);
440 void btrfs_dec_delayed_refs_rsv_bg_inserts(struct btrfs_fs_info *fs_info);
441 void btrfs_inc_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
442 void btrfs_dec_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
443 int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
444 enum btrfs_reserve_flush_enum flush);
445 void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
447 bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info);
450 * helper functions to cast a node into its container
452 static inline struct btrfs_delayed_tree_ref *
453 btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
455 return container_of(node, struct btrfs_delayed_tree_ref, node);
458 static inline struct btrfs_delayed_data_ref *
459 btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
461 return container_of(node, struct btrfs_delayed_data_ref, node);