Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Copyright (C) 2008 Oracle. All rights reserved.
4 */
5
6#ifndef BTRFS_DELAYED_REF_H
7#define BTRFS_DELAYED_REF_H
8
9#include <linux/refcount.h>
10
11/* these are the possible values of struct btrfs_delayed_ref_node->action */
12#define BTRFS_ADD_DELAYED_REF 1 /* add one backref to the tree */
13#define BTRFS_DROP_DELAYED_REF 2 /* delete one backref from the tree */
14#define BTRFS_ADD_DELAYED_EXTENT 3 /* record a full extent allocation */
15#define BTRFS_UPDATE_DELAYED_HEAD 4 /* not changing ref count on head ref */
16
17struct btrfs_delayed_ref_node {
18 struct rb_node ref_node;
19 /*
20 * If action is BTRFS_ADD_DELAYED_REF, also link this node to
21 * ref_head->ref_add_list, then we do not need to iterate the
22 * whole ref_head->ref_list to find BTRFS_ADD_DELAYED_REF nodes.
23 */
24 struct list_head add_list;
25
26 /* the starting bytenr of the extent */
27 u64 bytenr;
28
29 /* the size of the extent */
30 u64 num_bytes;
31
32 /* seq number to keep track of insertion order */
33 u64 seq;
34
35 /* ref count on this data structure */
36 refcount_t refs;
37
38 /*
39 * how many refs is this entry adding or deleting. For
40 * head refs, this may be a negative number because it is keeping
41 * track of the total mods done to the reference count.
42 * For individual refs, this will always be a positive number
43 *
44 * It may be more than one, since it is possible for a single
45 * parent to have more than one ref on an extent
46 */
47 int ref_mod;
48
49 unsigned int action:8;
50 unsigned int type:8;
51 /* is this node still in the rbtree? */
52 unsigned int is_head:1;
53 unsigned int in_tree:1;
54};
55
56struct btrfs_delayed_extent_op {
57 struct btrfs_disk_key key;
58 u8 level;
59 bool update_key;
60 bool update_flags;
61 u64 flags_to_set;
62};
63
64/*
65 * the head refs are used to hold a lock on a given extent, which allows us
66 * to make sure that only one process is running the delayed refs
67 * at a time for a single extent. They also store the sum of all the
68 * reference count modifications we've queued up.
69 */
70struct btrfs_delayed_ref_head {
71 u64 bytenr;
72 u64 num_bytes;
73 refcount_t refs;
74 /*
75 * the mutex is held while running the refs, and it is also
76 * held when checking the sum of reference modifications.
77 */
78 struct mutex mutex;
79
80 spinlock_t lock;
81 struct rb_root_cached ref_tree;
82 /* accumulate add BTRFS_ADD_DELAYED_REF nodes to this ref_add_list. */
83 struct list_head ref_add_list;
84
85 struct rb_node href_node;
86
87 struct btrfs_delayed_extent_op *extent_op;
88
89 /*
90 * This is used to track the final ref_mod from all the refs associated
91 * with this head ref, this is not adjusted as delayed refs are run,
92 * this is meant to track if we need to do the csum accounting or not.
93 */
94 int total_ref_mod;
95
96 /*
97 * This is the current outstanding mod references for this bytenr. This
98 * is used with lookup_extent_info to get an accurate reference count
99 * for a bytenr, so it is adjusted as delayed refs are run so that any
100 * on disk reference count + ref_mod is accurate.
101 */
102 int ref_mod;
103
104 /*
105 * when a new extent is allocated, it is just reserved in memory
106 * The actual extent isn't inserted into the extent allocation tree
107 * until the delayed ref is processed. must_insert_reserved is
108 * used to flag a delayed ref so the accounting can be updated
109 * when a full insert is done.
110 *
111 * It is possible the extent will be freed before it is ever
112 * inserted into the extent allocation tree. In this case
113 * we need to update the in ram accounting to properly reflect
114 * the free has happened.
115 */
116 unsigned int must_insert_reserved:1;
117 unsigned int is_data:1;
118 unsigned int is_system:1;
119 unsigned int processing:1;
120};
121
122struct btrfs_delayed_tree_ref {
123 struct btrfs_delayed_ref_node node;
124 u64 root;
125 u64 parent;
126 int level;
127};
128
129struct btrfs_delayed_data_ref {
130 struct btrfs_delayed_ref_node node;
131 u64 root;
132 u64 parent;
133 u64 objectid;
134 u64 offset;
135};
136
137enum btrfs_delayed_ref_flags {
138 /* Indicate that we are flushing delayed refs for the commit */
139 BTRFS_DELAYED_REFS_FLUSHING,
140};
141
142struct btrfs_delayed_ref_root {
143 /* head ref rbtree */
144 struct rb_root_cached href_root;
145
146 /* dirty extent records */
147 struct rb_root dirty_extent_root;
148
149 /* this spin lock protects the rbtree and the entries inside */
150 spinlock_t lock;
151
152 /* how many delayed ref updates we've queued, used by the
153 * throttling code
154 */
155 atomic_t num_entries;
156
157 /* total number of head nodes in tree */
158 unsigned long num_heads;
159
160 /* total number of head nodes ready for processing */
161 unsigned long num_heads_ready;
162
163 u64 pending_csums;
164
165 unsigned long flags;
166
167 u64 run_delayed_start;
168
169 /*
170 * To make qgroup to skip given root.
171 * This is for snapshot, as btrfs_qgroup_inherit() will manually
172 * modify counters for snapshot and its source, so we should skip
173 * the snapshot in new_root/old_roots or it will get calculated twice
174 */
175 u64 qgroup_to_skip;
176};
177
178enum btrfs_ref_type {
179 BTRFS_REF_NOT_SET,
180 BTRFS_REF_DATA,
181 BTRFS_REF_METADATA,
182 BTRFS_REF_LAST,
183};
184
185struct btrfs_data_ref {
186 /* For EXTENT_DATA_REF */
187
188 /* Original root this data extent belongs to */
189 u64 owning_root;
190
191 /* Inode which refers to this data extent */
192 u64 ino;
193
194 /*
195 * file_offset - extent_offset
196 *
197 * file_offset is the key.offset of the EXTENT_DATA key.
198 * extent_offset is btrfs_file_extent_offset() of the EXTENT_DATA data.
199 */
200 u64 offset;
201};
202
203struct btrfs_tree_ref {
204 /*
205 * Level of this tree block
206 *
207 * Shared for skinny (TREE_BLOCK_REF) and normal tree ref.
208 */
209 int level;
210
211 /*
212 * Root which owns this tree block.
213 *
214 * For TREE_BLOCK_REF (skinny metadata, either inline or keyed)
215 */
216 u64 owning_root;
217
218 /* For non-skinny metadata, no special member needed */
219};
220
221struct btrfs_ref {
222 enum btrfs_ref_type type;
223 int action;
224
225 /*
226 * Whether this extent should go through qgroup record.
227 *
228 * Normally false, but for certain cases like delayed subtree scan,
229 * setting this flag can hugely reduce qgroup overhead.
230 */
231 bool skip_qgroup;
232
233#ifdef CONFIG_BTRFS_FS_REF_VERIFY
234 /* Through which root is this modification. */
235 u64 real_root;
236#endif
237 u64 bytenr;
238 u64 len;
239
240 /* Bytenr of the parent tree block */
241 u64 parent;
242 union {
243 struct btrfs_data_ref data_ref;
244 struct btrfs_tree_ref tree_ref;
245 };
246};
247
248extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
249extern struct kmem_cache *btrfs_delayed_tree_ref_cachep;
250extern struct kmem_cache *btrfs_delayed_data_ref_cachep;
251extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
252
253int __init btrfs_delayed_ref_init(void);
254void __cold btrfs_delayed_ref_exit(void);
255
256static inline void btrfs_init_generic_ref(struct btrfs_ref *generic_ref,
257 int action, u64 bytenr, u64 len, u64 parent)
258{
259 generic_ref->action = action;
260 generic_ref->bytenr = bytenr;
261 generic_ref->len = len;
262 generic_ref->parent = parent;
263}
264
265static inline void btrfs_init_tree_ref(struct btrfs_ref *generic_ref,
266 int level, u64 root, u64 mod_root, bool skip_qgroup)
267{
268#ifdef CONFIG_BTRFS_FS_REF_VERIFY
269 /* If @real_root not set, use @root as fallback */
270 generic_ref->real_root = mod_root ?: root;
271#endif
272 generic_ref->tree_ref.level = level;
273 generic_ref->tree_ref.owning_root = root;
274 generic_ref->type = BTRFS_REF_METADATA;
275 if (skip_qgroup || !(is_fstree(root) &&
276 (!mod_root || is_fstree(mod_root))))
277 generic_ref->skip_qgroup = true;
278 else
279 generic_ref->skip_qgroup = false;
280
281}
282
283static inline void btrfs_init_data_ref(struct btrfs_ref *generic_ref,
284 u64 ref_root, u64 ino, u64 offset, u64 mod_root,
285 bool skip_qgroup)
286{
287#ifdef CONFIG_BTRFS_FS_REF_VERIFY
288 /* If @real_root not set, use @root as fallback */
289 generic_ref->real_root = mod_root ?: ref_root;
290#endif
291 generic_ref->data_ref.owning_root = ref_root;
292 generic_ref->data_ref.ino = ino;
293 generic_ref->data_ref.offset = offset;
294 generic_ref->type = BTRFS_REF_DATA;
295 if (skip_qgroup || !(is_fstree(ref_root) &&
296 (!mod_root || is_fstree(mod_root))))
297 generic_ref->skip_qgroup = true;
298 else
299 generic_ref->skip_qgroup = false;
300}
301
302static inline struct btrfs_delayed_extent_op *
303btrfs_alloc_delayed_extent_op(void)
304{
305 return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
306}
307
308static inline void
309btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
310{
311 if (op)
312 kmem_cache_free(btrfs_delayed_extent_op_cachep, op);
313}
314
315static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
316{
317 WARN_ON(refcount_read(&ref->refs) == 0);
318 if (refcount_dec_and_test(&ref->refs)) {
319 WARN_ON(ref->in_tree);
320 switch (ref->type) {
321 case BTRFS_TREE_BLOCK_REF_KEY:
322 case BTRFS_SHARED_BLOCK_REF_KEY:
323 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
324 break;
325 case BTRFS_EXTENT_DATA_REF_KEY:
326 case BTRFS_SHARED_DATA_REF_KEY:
327 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
328 break;
329 default:
330 BUG();
331 }
332 }
333}
334
335static inline u64 btrfs_ref_head_to_space_flags(
336 struct btrfs_delayed_ref_head *head_ref)
337{
338 if (head_ref->is_data)
339 return BTRFS_BLOCK_GROUP_DATA;
340 else if (head_ref->is_system)
341 return BTRFS_BLOCK_GROUP_SYSTEM;
342 return BTRFS_BLOCK_GROUP_METADATA;
343}
344
345static inline void btrfs_put_delayed_ref_head(struct btrfs_delayed_ref_head *head)
346{
347 if (refcount_dec_and_test(&head->refs))
348 kmem_cache_free(btrfs_delayed_ref_head_cachep, head);
349}
350
351int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
352 struct btrfs_ref *generic_ref,
353 struct btrfs_delayed_extent_op *extent_op);
354int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
355 struct btrfs_ref *generic_ref,
356 u64 reserved);
357int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
358 u64 bytenr, u64 num_bytes,
359 struct btrfs_delayed_extent_op *extent_op);
360void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
361 struct btrfs_delayed_ref_root *delayed_refs,
362 struct btrfs_delayed_ref_head *head);
363
364struct btrfs_delayed_ref_head *
365btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
366 u64 bytenr);
367int btrfs_delayed_ref_lock(struct btrfs_delayed_ref_root *delayed_refs,
368 struct btrfs_delayed_ref_head *head);
369static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
370{
371 mutex_unlock(&head->mutex);
372}
373void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
374 struct btrfs_delayed_ref_head *head);
375
376struct btrfs_delayed_ref_head *btrfs_select_ref_head(
377 struct btrfs_delayed_ref_root *delayed_refs);
378
379int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq);
380
381void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr);
382void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans);
383int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
384 enum btrfs_reserve_flush_enum flush);
385void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
386 struct btrfs_block_rsv *src,
387 u64 num_bytes);
388int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans);
389bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info);
390
391/*
392 * helper functions to cast a node into its container
393 */
394static inline struct btrfs_delayed_tree_ref *
395btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
396{
397 return container_of(node, struct btrfs_delayed_tree_ref, node);
398}
399
400static inline struct btrfs_delayed_data_ref *
401btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
402{
403 return container_of(node, struct btrfs_delayed_data_ref, node);
404}
405
406#endif
1/*
2 * Copyright (C) 2008 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18#ifndef __DELAYED_REF__
19#define __DELAYED_REF__
20
21/* these are the possible values of struct btrfs_delayed_ref->action */
22#define BTRFS_ADD_DELAYED_REF 1 /* add one backref to the tree */
23#define BTRFS_DROP_DELAYED_REF 2 /* delete one backref from the tree */
24#define BTRFS_ADD_DELAYED_EXTENT 3 /* record a full extent allocation */
25#define BTRFS_UPDATE_DELAYED_HEAD 4 /* not changing ref count on head ref */
26
27struct btrfs_delayed_ref_node {
28 struct rb_node rb_node;
29
30 /* the starting bytenr of the extent */
31 u64 bytenr;
32
33 /* the size of the extent */
34 u64 num_bytes;
35
36 /* seq number to keep track of insertion order */
37 u64 seq;
38
39 /* ref count on this data structure */
40 atomic_t refs;
41
42 /*
43 * how many refs is this entry adding or deleting. For
44 * head refs, this may be a negative number because it is keeping
45 * track of the total mods done to the reference count.
46 * For individual refs, this will always be a positive number
47 *
48 * It may be more than one, since it is possible for a single
49 * parent to have more than one ref on an extent
50 */
51 int ref_mod;
52
53 unsigned int action:8;
54 unsigned int type:8;
55 /* is this node still in the rbtree? */
56 unsigned int is_head:1;
57 unsigned int in_tree:1;
58};
59
60struct btrfs_delayed_extent_op {
61 struct btrfs_disk_key key;
62 u64 flags_to_set;
63 unsigned int update_key:1;
64 unsigned int update_flags:1;
65 unsigned int is_data:1;
66};
67
68/*
69 * the head refs are used to hold a lock on a given extent, which allows us
70 * to make sure that only one process is running the delayed refs
71 * at a time for a single extent. They also store the sum of all the
72 * reference count modifications we've queued up.
73 */
74struct btrfs_delayed_ref_head {
75 struct btrfs_delayed_ref_node node;
76
77 /*
78 * the mutex is held while running the refs, and it is also
79 * held when checking the sum of reference modifications.
80 */
81 struct mutex mutex;
82
83 struct list_head cluster;
84
85 struct btrfs_delayed_extent_op *extent_op;
86 /*
87 * when a new extent is allocated, it is just reserved in memory
88 * The actual extent isn't inserted into the extent allocation tree
89 * until the delayed ref is processed. must_insert_reserved is
90 * used to flag a delayed ref so the accounting can be updated
91 * when a full insert is done.
92 *
93 * It is possible the extent will be freed before it is ever
94 * inserted into the extent allocation tree. In this case
95 * we need to update the in ram accounting to properly reflect
96 * the free has happened.
97 */
98 unsigned int must_insert_reserved:1;
99 unsigned int is_data:1;
100};
101
102struct btrfs_delayed_tree_ref {
103 struct btrfs_delayed_ref_node node;
104 u64 root;
105 u64 parent;
106 int level;
107};
108
109struct btrfs_delayed_data_ref {
110 struct btrfs_delayed_ref_node node;
111 u64 root;
112 u64 parent;
113 u64 objectid;
114 u64 offset;
115};
116
117struct btrfs_delayed_ref_root {
118 struct rb_root root;
119
120 /* this spin lock protects the rbtree and the entries inside */
121 spinlock_t lock;
122
123 /* how many delayed ref updates we've queued, used by the
124 * throttling code
125 */
126 unsigned long num_entries;
127
128 /* total number of head nodes in tree */
129 unsigned long num_heads;
130
131 /* total number of head nodes ready for processing */
132 unsigned long num_heads_ready;
133
134 /*
135 * set when the tree is flushing before a transaction commit,
136 * used by the throttling code to decide if new updates need
137 * to be run right away
138 */
139 int flushing;
140
141 u64 run_delayed_start;
142
143 /*
144 * seq number of delayed refs. We need to know if a backref was being
145 * added before the currently processed ref or afterwards.
146 */
147 u64 seq;
148
149 /*
150 * seq_list holds a list of all seq numbers that are currently being
151 * added to the list. While walking backrefs (btrfs_find_all_roots,
152 * qgroups), which might take some time, no newer ref must be processed,
153 * as it might influence the outcome of the walk.
154 */
155 struct list_head seq_head;
156
157 /*
158 * when the only refs we have in the list must not be processed, we want
159 * to wait for more refs to show up or for the end of backref walking.
160 */
161 wait_queue_head_t seq_wait;
162};
163
164static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
165{
166 WARN_ON(atomic_read(&ref->refs) == 0);
167 if (atomic_dec_and_test(&ref->refs)) {
168 WARN_ON(ref->in_tree);
169 kfree(ref);
170 }
171}
172
173int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
174 struct btrfs_trans_handle *trans,
175 u64 bytenr, u64 num_bytes, u64 parent,
176 u64 ref_root, int level, int action,
177 struct btrfs_delayed_extent_op *extent_op,
178 int for_cow);
179int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info,
180 struct btrfs_trans_handle *trans,
181 u64 bytenr, u64 num_bytes,
182 u64 parent, u64 ref_root,
183 u64 owner, u64 offset, int action,
184 struct btrfs_delayed_extent_op *extent_op,
185 int for_cow);
186int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
187 struct btrfs_trans_handle *trans,
188 u64 bytenr, u64 num_bytes,
189 struct btrfs_delayed_extent_op *extent_op);
190
191struct btrfs_delayed_ref_head *
192btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr);
193int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
194 struct btrfs_delayed_ref_head *head);
195int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans,
196 struct list_head *cluster, u64 search_start);
197
198static inline u64 inc_delayed_seq(struct btrfs_delayed_ref_root *delayed_refs)
199{
200 assert_spin_locked(&delayed_refs->lock);
201 ++delayed_refs->seq;
202 return delayed_refs->seq;
203}
204
205static inline void
206btrfs_get_delayed_seq(struct btrfs_delayed_ref_root *delayed_refs,
207 struct seq_list *elem)
208{
209 assert_spin_locked(&delayed_refs->lock);
210 elem->seq = delayed_refs->seq;
211 list_add_tail(&elem->list, &delayed_refs->seq_head);
212}
213
214static inline void
215btrfs_put_delayed_seq(struct btrfs_delayed_ref_root *delayed_refs,
216 struct seq_list *elem)
217{
218 spin_lock(&delayed_refs->lock);
219 list_del(&elem->list);
220 wake_up(&delayed_refs->seq_wait);
221 spin_unlock(&delayed_refs->lock);
222}
223
224int btrfs_check_delayed_seq(struct btrfs_delayed_ref_root *delayed_refs,
225 u64 seq);
226
227/*
228 * a node might live in a head or a regular ref, this lets you
229 * test for the proper type to use.
230 */
231static int btrfs_delayed_ref_is_head(struct btrfs_delayed_ref_node *node)
232{
233 return node->is_head;
234}
235
236/*
237 * helper functions to cast a node into its container
238 */
239static inline struct btrfs_delayed_tree_ref *
240btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
241{
242 WARN_ON(btrfs_delayed_ref_is_head(node));
243 return container_of(node, struct btrfs_delayed_tree_ref, node);
244}
245
246static inline struct btrfs_delayed_data_ref *
247btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
248{
249 WARN_ON(btrfs_delayed_ref_is_head(node));
250 return container_of(node, struct btrfs_delayed_data_ref, node);
251}
252
253static inline struct btrfs_delayed_ref_head *
254btrfs_delayed_node_to_head(struct btrfs_delayed_ref_node *node)
255{
256 WARN_ON(!btrfs_delayed_ref_is_head(node));
257 return container_of(node, struct btrfs_delayed_ref_head, node);
258}
259#endif