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/types.h>
10 #include <linux/refcount.h>
11 #include <linux/list.h>
12 #include <linux/rbtree.h>
13 #include <linux/mutex.h>
14 #include <linux/spinlock.h>
15 #include <linux/slab.h>
16 #include <uapi/linux/btrfs_tree.h>
18 struct btrfs_trans_handle;
21 /* these are the possible values of struct btrfs_delayed_ref_node->action */
22 enum btrfs_delayed_ref_action {
23 /* Add one backref to the tree */
24 BTRFS_ADD_DELAYED_REF = 1,
25 /* Delete one backref from the tree */
26 BTRFS_DROP_DELAYED_REF,
27 /* Record a full extent allocation */
28 BTRFS_ADD_DELAYED_EXTENT,
29 /* Not changing ref count on head ref */
30 BTRFS_UPDATE_DELAYED_HEAD,
33 struct btrfs_data_ref {
34 /* For EXTENT_DATA_REF */
36 /* Inode which refers to this data extent */
40 * file_offset - extent_offset
42 * file_offset is the key.offset of the EXTENT_DATA key.
43 * extent_offset is btrfs_file_extent_offset() of the EXTENT_DATA data.
48 struct btrfs_tree_ref {
50 * Level of this tree block.
52 * Shared for skinny (TREE_BLOCK_REF) and normal tree ref.
56 /* For non-skinny metadata, no special member needed */
59 struct btrfs_delayed_ref_node {
60 struct rb_node ref_node;
62 * If action is BTRFS_ADD_DELAYED_REF, also link this node to
63 * ref_head->ref_add_list, then we do not need to iterate the
64 * refs rbtree in the corresponding delayed ref head
65 * (struct btrfs_delayed_ref_head::ref_tree).
67 struct list_head add_list;
69 /* the starting bytenr of the extent */
72 /* the size of the extent */
75 /* seq number to keep track of insertion order */
78 /* The ref_root for this ref */
82 * The parent for this ref, if this isn't set the ref_root is the
87 /* ref count on this data structure */
91 * how many refs is this entry adding or deleting. For
92 * head refs, this may be a negative number because it is keeping
93 * track of the total mods done to the reference count.
94 * For individual refs, this will always be a positive number
96 * It may be more than one, since it is possible for a single
97 * parent to have more than one ref on an extent
101 unsigned int action:8;
105 struct btrfs_tree_ref tree_ref;
106 struct btrfs_data_ref data_ref;
110 struct btrfs_delayed_extent_op {
111 struct btrfs_disk_key key;
118 * the head refs are used to hold a lock on a given extent, which allows us
119 * to make sure that only one process is running the delayed refs
120 * at a time for a single extent. They also store the sum of all the
121 * reference count modifications we've queued up.
123 struct btrfs_delayed_ref_head {
127 * the mutex is held while running the refs, and it is also
128 * held when checking the sum of reference modifications.
134 /* Protects 'ref_tree' and 'ref_add_list'. */
136 struct rb_root_cached ref_tree;
137 /* accumulate add BTRFS_ADD_DELAYED_REF nodes to this ref_add_list. */
138 struct list_head ref_add_list;
140 struct btrfs_delayed_extent_op *extent_op;
143 * This is used to track the final ref_mod from all the refs associated
144 * with this head ref, this is not adjusted as delayed refs are run,
145 * this is meant to track if we need to do the csum accounting or not.
150 * This is the current outstanding mod references for this bytenr. This
151 * is used with lookup_extent_info to get an accurate reference count
152 * for a bytenr, so it is adjusted as delayed refs are run so that any
153 * on disk reference count + ref_mod is accurate.
158 * The root that triggered the allocation when must_insert_reserved is
164 * Track reserved bytes when setting must_insert_reserved. On success
165 * or cleanup, we will need to free the reservation.
169 /* Tree block level, for metadata only. */
173 * when a new extent is allocated, it is just reserved in memory
174 * The actual extent isn't inserted into the extent allocation tree
175 * until the delayed ref is processed. must_insert_reserved is
176 * used to flag a delayed ref so the accounting can be updated
177 * when a full insert is done.
179 * It is possible the extent will be freed before it is ever
180 * inserted into the extent allocation tree. In this case
181 * we need to update the in ram accounting to properly reflect
182 * the free has happened.
184 bool must_insert_reserved;
190 * Indicate if it's currently in the data structure that tracks head
191 * refs (struct btrfs_delayed_ref_root::head_refs).
196 enum btrfs_delayed_ref_flags {
197 /* Indicate that we are flushing delayed refs for the commit */
198 BTRFS_DELAYED_REFS_FLUSHING,
201 struct btrfs_delayed_ref_root {
203 * Track head references.
204 * The keys correspond to the logical address of the extent ("bytenr")
205 * right shifted by fs_info->sectorsize_bits. This is both to get a more
206 * dense index space (optimizes xarray structure) and because indexes in
207 * xarrays are of "unsigned long" type, meaning they are 32 bits wide on
208 * 32 bits platforms, limiting the extent range to 4G which is too low
209 * and makes it unusable (truncated index values) on 32 bits platforms.
210 * Protected by the spinlock 'lock' defined below.
212 struct xarray head_refs;
215 * Track dirty extent records.
216 * The keys correspond to the logical address of the extent ("bytenr")
217 * right shifted by fs_info->sectorsize_bits, for same reasons as above.
219 struct xarray dirty_extents;
222 * Protects the xarray head_refs, its entries and the following fields:
223 * num_heads, num_heads_ready, pending_csums and run_delayed_start.
227 /* Total number of head refs, protected by the spinlock 'lock'. */
228 unsigned long num_heads;
231 * Total number of head refs ready for processing, protected by the
234 unsigned long num_heads_ready;
237 * Track space reserved for deleting csums of data extents.
238 * Protected by the spinlock 'lock'.
245 * Track from which bytenr to start searching ref heads.
246 * Protected by the spinlock 'lock'.
248 u64 run_delayed_start;
251 * To make qgroup to skip given root.
252 * This is for snapshot, as btrfs_qgroup_inherit() will manually
253 * modify counters for snapshot and its source, so we should skip
254 * the snapshot in new_root/old_roots or it will get calculated twice
259 enum btrfs_ref_type {
267 enum btrfs_ref_type type;
268 enum btrfs_delayed_ref_action action;
271 * Whether this extent should go through qgroup record.
273 * Normally false, but for certain cases like delayed subtree scan,
274 * setting this flag can hugely reduce qgroup overhead.
278 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
279 /* Through which root is this modification. */
287 * The root that owns the reference for this reference, this will be set
288 * or ->parent will be set, depending on what type of reference this is.
292 /* Bytenr of the parent tree block */
295 struct btrfs_data_ref data_ref;
296 struct btrfs_tree_ref tree_ref;
300 extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
301 extern struct kmem_cache *btrfs_delayed_ref_node_cachep;
302 extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
304 int __init btrfs_delayed_ref_init(void);
305 void __cold btrfs_delayed_ref_exit(void);
307 static inline u64 btrfs_calc_delayed_ref_bytes(const struct btrfs_fs_info *fs_info,
308 int num_delayed_refs)
312 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_delayed_refs);
315 * We have to check the mount option here because we could be enabling
316 * the free space tree for the first time and don't have the compat_ro
319 * We need extra reservations if we have the free space tree because
320 * we'll have to modify that tree as well.
322 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE))
328 static inline u64 btrfs_calc_delayed_ref_csum_bytes(const struct btrfs_fs_info *fs_info,
332 * Deleting csum items does not result in new nodes/leaves and does not
333 * require changing the free space tree, only the csum tree, so this is
336 return btrfs_calc_metadata_size(fs_info, num_csum_items);
339 void btrfs_init_tree_ref(struct btrfs_ref *generic_ref, int level, u64 mod_root,
341 void btrfs_init_data_ref(struct btrfs_ref *generic_ref, u64 ino, u64 offset,
342 u64 mod_root, bool skip_qgroup);
344 static inline struct btrfs_delayed_extent_op *
345 btrfs_alloc_delayed_extent_op(void)
347 return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
351 btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
354 kmem_cache_free(btrfs_delayed_extent_op_cachep, op);
357 void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref);
359 static inline u64 btrfs_ref_head_to_space_flags(
360 struct btrfs_delayed_ref_head *head_ref)
362 if (head_ref->is_data)
363 return BTRFS_BLOCK_GROUP_DATA;
364 else if (head_ref->is_system)
365 return BTRFS_BLOCK_GROUP_SYSTEM;
366 return BTRFS_BLOCK_GROUP_METADATA;
369 static inline void btrfs_put_delayed_ref_head(struct btrfs_delayed_ref_head *head)
371 if (refcount_dec_and_test(&head->refs))
372 kmem_cache_free(btrfs_delayed_ref_head_cachep, head);
375 int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
376 struct btrfs_ref *generic_ref,
377 struct btrfs_delayed_extent_op *extent_op);
378 int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
379 struct btrfs_ref *generic_ref,
381 int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
382 u64 bytenr, u64 num_bytes, u8 level,
383 struct btrfs_delayed_extent_op *extent_op);
384 void btrfs_merge_delayed_refs(struct btrfs_fs_info *fs_info,
385 struct btrfs_delayed_ref_root *delayed_refs,
386 struct btrfs_delayed_ref_head *head);
388 struct btrfs_delayed_ref_head *
389 btrfs_find_delayed_ref_head(const struct btrfs_fs_info *fs_info,
390 struct btrfs_delayed_ref_root *delayed_refs,
392 static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
394 mutex_unlock(&head->mutex);
396 void btrfs_delete_ref_head(const struct btrfs_fs_info *fs_info,
397 struct btrfs_delayed_ref_root *delayed_refs,
398 struct btrfs_delayed_ref_head *head);
400 struct btrfs_delayed_ref_head *btrfs_select_ref_head(
401 const struct btrfs_fs_info *fs_info,
402 struct btrfs_delayed_ref_root *delayed_refs);
403 void btrfs_unselect_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
404 struct btrfs_delayed_ref_head *head);
406 int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq);
408 void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr_refs, int nr_csums);
409 void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans);
410 void btrfs_inc_delayed_refs_rsv_bg_inserts(struct btrfs_fs_info *fs_info);
411 void btrfs_dec_delayed_refs_rsv_bg_inserts(struct btrfs_fs_info *fs_info);
412 void btrfs_inc_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
413 void btrfs_dec_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
414 int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
415 enum btrfs_reserve_flush_enum flush);
416 bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info);
417 bool btrfs_find_delayed_tree_ref(struct btrfs_delayed_ref_head *head,
418 u64 root, u64 parent);
419 void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans);
421 static inline u64 btrfs_delayed_ref_owner(struct btrfs_delayed_ref_node *node)
423 if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
424 node->type == BTRFS_SHARED_DATA_REF_KEY)
425 return node->data_ref.objectid;
426 return node->tree_ref.level;
429 static inline u64 btrfs_delayed_ref_offset(struct btrfs_delayed_ref_node *node)
431 if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
432 node->type == BTRFS_SHARED_DATA_REF_KEY)
433 return node->data_ref.offset;
437 static inline u8 btrfs_ref_type(struct btrfs_ref *ref)
439 ASSERT(ref->type == BTRFS_REF_DATA || ref->type == BTRFS_REF_METADATA);
441 if (ref->type == BTRFS_REF_DATA) {
443 return BTRFS_SHARED_DATA_REF_KEY;
445 return BTRFS_EXTENT_DATA_REF_KEY;
448 return BTRFS_SHARED_BLOCK_REF_KEY;
450 return BTRFS_TREE_BLOCK_REF_KEY;