Loading...
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
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 bool is_data;
62 u64 flags_to_set;
63};
64
65/*
66 * the head refs are used to hold a lock on a given extent, which allows us
67 * to make sure that only one process is running the delayed refs
68 * at a time for a single extent. They also store the sum of all the
69 * reference count modifications we've queued up.
70 */
71struct btrfs_delayed_ref_head {
72 u64 bytenr;
73 u64 num_bytes;
74 refcount_t refs;
75 /*
76 * the mutex is held while running the refs, and it is also
77 * held when checking the sum of reference modifications.
78 */
79 struct mutex mutex;
80
81 spinlock_t lock;
82 struct rb_root_cached ref_tree;
83 /* accumulate add BTRFS_ADD_DELAYED_REF nodes to this ref_add_list. */
84 struct list_head ref_add_list;
85
86 struct rb_node href_node;
87
88 struct btrfs_delayed_extent_op *extent_op;
89
90 /*
91 * This is used to track the final ref_mod from all the refs associated
92 * with this head ref, this is not adjusted as delayed refs are run,
93 * this is meant to track if we need to do the csum accounting or not.
94 */
95 int total_ref_mod;
96
97 /*
98 * This is the current outstanding mod references for this bytenr. This
99 * is used with lookup_extent_info to get an accurate reference count
100 * for a bytenr, so it is adjusted as delayed refs are run so that any
101 * on disk reference count + ref_mod is accurate.
102 */
103 int ref_mod;
104
105 /*
106 * when a new extent is allocated, it is just reserved in memory
107 * The actual extent isn't inserted into the extent allocation tree
108 * until the delayed ref is processed. must_insert_reserved is
109 * used to flag a delayed ref so the accounting can be updated
110 * when a full insert is done.
111 *
112 * It is possible the extent will be freed before it is ever
113 * inserted into the extent allocation tree. In this case
114 * we need to update the in ram accounting to properly reflect
115 * the free has happened.
116 */
117 unsigned int must_insert_reserved:1;
118 unsigned int is_data:1;
119 unsigned int is_system:1;
120 unsigned int processing:1;
121};
122
123struct btrfs_delayed_tree_ref {
124 struct btrfs_delayed_ref_node node;
125 u64 root;
126 u64 parent;
127 int level;
128};
129
130struct btrfs_delayed_data_ref {
131 struct btrfs_delayed_ref_node node;
132 u64 root;
133 u64 parent;
134 u64 objectid;
135 u64 offset;
136};
137
138enum btrfs_delayed_ref_flags {
139 /* Indicate that we are flushing delayed refs for the commit */
140 BTRFS_DELAYED_REFS_FLUSHING,
141};
142
143struct btrfs_delayed_ref_root {
144 /* head ref rbtree */
145 struct rb_root_cached href_root;
146
147 /* dirty extent records */
148 struct rb_root dirty_extent_root;
149
150 /* this spin lock protects the rbtree and the entries inside */
151 spinlock_t lock;
152
153 /* how many delayed ref updates we've queued, used by the
154 * throttling code
155 */
156 atomic_t num_entries;
157
158 /* total number of head nodes in tree */
159 unsigned long num_heads;
160
161 /* total number of head nodes ready for processing */
162 unsigned long num_heads_ready;
163
164 u64 pending_csums;
165
166 unsigned long flags;
167
168 u64 run_delayed_start;
169
170 /*
171 * To make qgroup to skip given root.
172 * This is for snapshot, as btrfs_qgroup_inherit() will manually
173 * modify counters for snapshot and its source, so we should skip
174 * the snapshot in new_root/old_roots or it will get calculated twice
175 */
176 u64 qgroup_to_skip;
177};
178
179enum btrfs_ref_type {
180 BTRFS_REF_NOT_SET,
181 BTRFS_REF_DATA,
182 BTRFS_REF_METADATA,
183 BTRFS_REF_LAST,
184};
185
186struct btrfs_data_ref {
187 /* For EXTENT_DATA_REF */
188
189 /* Root which refers to this data extent */
190 u64 ref_root;
191
192 /* Inode which refers to this data extent */
193 u64 ino;
194
195 /*
196 * file_offset - extent_offset
197 *
198 * file_offset is the key.offset of the EXTENT_DATA key.
199 * extent_offset is btrfs_file_extent_offset() of the EXTENT_DATA data.
200 */
201 u64 offset;
202};
203
204struct btrfs_tree_ref {
205 /*
206 * Level of this tree block
207 *
208 * Shared for skinny (TREE_BLOCK_REF) and normal tree ref.
209 */
210 int level;
211
212 /*
213 * Root which refers to this tree block.
214 *
215 * For TREE_BLOCK_REF (skinny metadata, either inline or keyed)
216 */
217 u64 root;
218
219 /* For non-skinny metadata, no special member needed */
220};
221
222struct btrfs_ref {
223 enum btrfs_ref_type type;
224 int action;
225
226 /*
227 * Whether this extent should go through qgroup record.
228 *
229 * Normally false, but for certain cases like delayed subtree scan,
230 * setting this flag can hugely reduce qgroup overhead.
231 */
232 bool skip_qgroup;
233
234 /*
235 * Optional. For which root is this modification.
236 * Mostly used for qgroup optimization.
237 *
238 * When unset, data/tree ref init code will populate it.
239 * In certain cases, we're modifying reference for a different root.
240 * E.g. COW fs tree blocks for balance.
241 * In that case, tree_ref::root will be fs tree, but we're doing this
242 * for reloc tree, then we should set @real_root to reloc tree.
243 */
244 u64 real_root;
245 u64 bytenr;
246 u64 len;
247
248 /* Bytenr of the parent tree block */
249 u64 parent;
250 union {
251 struct btrfs_data_ref data_ref;
252 struct btrfs_tree_ref tree_ref;
253 };
254};
255
256extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
257extern struct kmem_cache *btrfs_delayed_tree_ref_cachep;
258extern struct kmem_cache *btrfs_delayed_data_ref_cachep;
259extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
260
261int __init btrfs_delayed_ref_init(void);
262void __cold btrfs_delayed_ref_exit(void);
263
264static inline void btrfs_init_generic_ref(struct btrfs_ref *generic_ref,
265 int action, u64 bytenr, u64 len, u64 parent)
266{
267 generic_ref->action = action;
268 generic_ref->bytenr = bytenr;
269 generic_ref->len = len;
270 generic_ref->parent = parent;
271}
272
273static inline void btrfs_init_tree_ref(struct btrfs_ref *generic_ref,
274 int level, u64 root)
275{
276 /* If @real_root not set, use @root as fallback */
277 if (!generic_ref->real_root)
278 generic_ref->real_root = root;
279 generic_ref->tree_ref.level = level;
280 generic_ref->tree_ref.root = root;
281 generic_ref->type = BTRFS_REF_METADATA;
282}
283
284static inline void btrfs_init_data_ref(struct btrfs_ref *generic_ref,
285 u64 ref_root, u64 ino, u64 offset)
286{
287 /* If @real_root not set, use @root as fallback */
288 if (!generic_ref->real_root)
289 generic_ref->real_root = ref_root;
290 generic_ref->data_ref.ref_root = ref_root;
291 generic_ref->data_ref.ino = ino;
292 generic_ref->data_ref.offset = offset;
293 generic_ref->type = BTRFS_REF_DATA;
294}
295
296static inline struct btrfs_delayed_extent_op *
297btrfs_alloc_delayed_extent_op(void)
298{
299 return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
300}
301
302static inline void
303btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
304{
305 if (op)
306 kmem_cache_free(btrfs_delayed_extent_op_cachep, op);
307}
308
309static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
310{
311 WARN_ON(refcount_read(&ref->refs) == 0);
312 if (refcount_dec_and_test(&ref->refs)) {
313 WARN_ON(ref->in_tree);
314 switch (ref->type) {
315 case BTRFS_TREE_BLOCK_REF_KEY:
316 case BTRFS_SHARED_BLOCK_REF_KEY:
317 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
318 break;
319 case BTRFS_EXTENT_DATA_REF_KEY:
320 case BTRFS_SHARED_DATA_REF_KEY:
321 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
322 break;
323 default:
324 BUG();
325 }
326 }
327}
328
329static inline u64 btrfs_ref_head_to_space_flags(
330 struct btrfs_delayed_ref_head *head_ref)
331{
332 if (head_ref->is_data)
333 return BTRFS_BLOCK_GROUP_DATA;
334 else if (head_ref->is_system)
335 return BTRFS_BLOCK_GROUP_SYSTEM;
336 return BTRFS_BLOCK_GROUP_METADATA;
337}
338
339static inline void btrfs_put_delayed_ref_head(struct btrfs_delayed_ref_head *head)
340{
341 if (refcount_dec_and_test(&head->refs))
342 kmem_cache_free(btrfs_delayed_ref_head_cachep, head);
343}
344
345int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
346 struct btrfs_ref *generic_ref,
347 struct btrfs_delayed_extent_op *extent_op);
348int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
349 struct btrfs_ref *generic_ref,
350 u64 reserved);
351int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
352 u64 bytenr, u64 num_bytes,
353 struct btrfs_delayed_extent_op *extent_op);
354void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
355 struct btrfs_delayed_ref_root *delayed_refs,
356 struct btrfs_delayed_ref_head *head);
357
358struct btrfs_delayed_ref_head *
359btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
360 u64 bytenr);
361int btrfs_delayed_ref_lock(struct btrfs_delayed_ref_root *delayed_refs,
362 struct btrfs_delayed_ref_head *head);
363static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
364{
365 mutex_unlock(&head->mutex);
366}
367void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
368 struct btrfs_delayed_ref_head *head);
369
370struct btrfs_delayed_ref_head *btrfs_select_ref_head(
371 struct btrfs_delayed_ref_root *delayed_refs);
372
373int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq);
374
375void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr);
376void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans);
377int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
378 enum btrfs_reserve_flush_enum flush);
379void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
380 struct btrfs_block_rsv *src,
381 u64 num_bytes);
382int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans);
383bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info);
384
385/*
386 * helper functions to cast a node into its container
387 */
388static inline struct btrfs_delayed_tree_ref *
389btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
390{
391 return container_of(node, struct btrfs_delayed_tree_ref, node);
392}
393
394static inline struct btrfs_delayed_data_ref *
395btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
396{
397 return container_of(node, struct btrfs_delayed_data_ref, node);
398}
399
400#endif