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_node->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 int level;
64 unsigned int update_key:1;
65 unsigned int update_flags:1;
66 unsigned int is_data:1;
67};
68
69/*
70 * the head refs are used to hold a lock on a given extent, which allows us
71 * to make sure that only one process is running the delayed refs
72 * at a time for a single extent. They also store the sum of all the
73 * reference count modifications we've queued up.
74 */
75struct btrfs_delayed_ref_head {
76 struct btrfs_delayed_ref_node node;
77
78 /*
79 * the mutex is held while running the refs, and it is also
80 * held when checking the sum of reference modifications.
81 */
82 struct mutex mutex;
83
84 spinlock_t lock;
85 struct rb_root ref_root;
86
87 struct rb_node href_node;
88
89 struct btrfs_delayed_extent_op *extent_op;
90 /*
91 * when a new extent is allocated, it is just reserved in memory
92 * The actual extent isn't inserted into the extent allocation tree
93 * until the delayed ref is processed. must_insert_reserved is
94 * used to flag a delayed ref so the accounting can be updated
95 * when a full insert is done.
96 *
97 * It is possible the extent will be freed before it is ever
98 * inserted into the extent allocation tree. In this case
99 * we need to update the in ram accounting to properly reflect
100 * the free has happened.
101 */
102 unsigned int must_insert_reserved:1;
103 unsigned int is_data:1;
104 unsigned int processing:1;
105};
106
107struct btrfs_delayed_tree_ref {
108 struct btrfs_delayed_ref_node node;
109 u64 root;
110 u64 parent;
111 int level;
112};
113
114struct btrfs_delayed_data_ref {
115 struct btrfs_delayed_ref_node node;
116 u64 root;
117 u64 parent;
118 u64 objectid;
119 u64 offset;
120};
121
122struct btrfs_delayed_ref_root {
123 /* head ref rbtree */
124 struct rb_root href_root;
125
126 /* this spin lock protects the rbtree and the entries inside */
127 spinlock_t lock;
128
129 /* how many delayed ref updates we've queued, used by the
130 * throttling code
131 */
132 atomic_t num_entries;
133
134 /* total number of head nodes in tree */
135 unsigned long num_heads;
136
137 /* total number of head nodes ready for processing */
138 unsigned long num_heads_ready;
139
140 /*
141 * set when the tree is flushing before a transaction commit,
142 * used by the throttling code to decide if new updates need
143 * to be run right away
144 */
145 int flushing;
146
147 u64 run_delayed_start;
148};
149
150extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
151extern struct kmem_cache *btrfs_delayed_tree_ref_cachep;
152extern struct kmem_cache *btrfs_delayed_data_ref_cachep;
153extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
154
155int btrfs_delayed_ref_init(void);
156void btrfs_delayed_ref_exit(void);
157
158static inline struct btrfs_delayed_extent_op *
159btrfs_alloc_delayed_extent_op(void)
160{
161 return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
162}
163
164static inline void
165btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
166{
167 if (op)
168 kmem_cache_free(btrfs_delayed_extent_op_cachep, op);
169}
170
171static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
172{
173 WARN_ON(atomic_read(&ref->refs) == 0);
174 if (atomic_dec_and_test(&ref->refs)) {
175 WARN_ON(ref->in_tree);
176 switch (ref->type) {
177 case BTRFS_TREE_BLOCK_REF_KEY:
178 case BTRFS_SHARED_BLOCK_REF_KEY:
179 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
180 break;
181 case BTRFS_EXTENT_DATA_REF_KEY:
182 case BTRFS_SHARED_DATA_REF_KEY:
183 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
184 break;
185 case 0:
186 kmem_cache_free(btrfs_delayed_ref_head_cachep, ref);
187 break;
188 default:
189 BUG();
190 }
191 }
192}
193
194int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
195 struct btrfs_trans_handle *trans,
196 u64 bytenr, u64 num_bytes, u64 parent,
197 u64 ref_root, int level, int action,
198 struct btrfs_delayed_extent_op *extent_op,
199 int for_cow);
200int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info,
201 struct btrfs_trans_handle *trans,
202 u64 bytenr, u64 num_bytes,
203 u64 parent, u64 ref_root,
204 u64 owner, u64 offset, int action,
205 struct btrfs_delayed_extent_op *extent_op,
206 int for_cow);
207int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
208 struct btrfs_trans_handle *trans,
209 u64 bytenr, u64 num_bytes,
210 struct btrfs_delayed_extent_op *extent_op);
211void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
212 struct btrfs_fs_info *fs_info,
213 struct btrfs_delayed_ref_root *delayed_refs,
214 struct btrfs_delayed_ref_head *head);
215
216struct btrfs_delayed_ref_head *
217btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr);
218int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
219 struct btrfs_delayed_ref_head *head);
220static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
221{
222 mutex_unlock(&head->mutex);
223}
224
225
226struct btrfs_delayed_ref_head *
227btrfs_select_ref_head(struct btrfs_trans_handle *trans);
228
229int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info,
230 struct btrfs_delayed_ref_root *delayed_refs,
231 u64 seq);
232
233/*
234 * delayed refs with a ref_seq > 0 must be held back during backref walking.
235 * this only applies to items in one of the fs-trees. for_cow items never need
236 * to be held back, so they won't get a ref_seq number.
237 */
238static inline int need_ref_seq(int for_cow, u64 rootid)
239{
240 if (for_cow)
241 return 0;
242
243 if (rootid == BTRFS_FS_TREE_OBJECTID)
244 return 1;
245
246 if ((s64)rootid >= (s64)BTRFS_FIRST_FREE_OBJECTID)
247 return 1;
248
249 return 0;
250}
251
252/*
253 * a node might live in a head or a regular ref, this lets you
254 * test for the proper type to use.
255 */
256static int btrfs_delayed_ref_is_head(struct btrfs_delayed_ref_node *node)
257{
258 return node->is_head;
259}
260
261/*
262 * helper functions to cast a node into its container
263 */
264static inline struct btrfs_delayed_tree_ref *
265btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
266{
267 WARN_ON(btrfs_delayed_ref_is_head(node));
268 return container_of(node, struct btrfs_delayed_tree_ref, node);
269}
270
271static inline struct btrfs_delayed_data_ref *
272btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
273{
274 WARN_ON(btrfs_delayed_ref_is_head(node));
275 return container_of(node, struct btrfs_delayed_data_ref, node);
276}
277
278static inline struct btrfs_delayed_ref_head *
279btrfs_delayed_node_to_head(struct btrfs_delayed_ref_node *node)
280{
281 WARN_ON(!btrfs_delayed_ref_is_head(node));
282 return container_of(node, struct btrfs_delayed_ref_head, node);
283}
284#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/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>
17
18struct btrfs_trans_handle;
19struct btrfs_fs_info;
20
21/* these are the possible values of struct btrfs_delayed_ref_node->action */
22enum 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,
31} __packed;
32
33struct btrfs_delayed_ref_node {
34 struct rb_node ref_node;
35 /*
36 * If action is BTRFS_ADD_DELAYED_REF, also link this node to
37 * ref_head->ref_add_list, then we do not need to iterate the
38 * whole ref_head->ref_list to find BTRFS_ADD_DELAYED_REF nodes.
39 */
40 struct list_head add_list;
41
42 /* the starting bytenr of the extent */
43 u64 bytenr;
44
45 /* the size of the extent */
46 u64 num_bytes;
47
48 /* seq number to keep track of insertion order */
49 u64 seq;
50
51 /* ref count on this data structure */
52 refcount_t refs;
53
54 /*
55 * how many refs is this entry adding or deleting. For
56 * head refs, this may be a negative number because it is keeping
57 * track of the total mods done to the reference count.
58 * For individual refs, this will always be a positive number
59 *
60 * It may be more than one, since it is possible for a single
61 * parent to have more than one ref on an extent
62 */
63 int ref_mod;
64
65 unsigned int action:8;
66 unsigned int type:8;
67};
68
69struct btrfs_delayed_extent_op {
70 struct btrfs_disk_key key;
71 u8 level;
72 bool update_key;
73 bool update_flags;
74 u64 flags_to_set;
75};
76
77/*
78 * the head refs are used to hold a lock on a given extent, which allows us
79 * to make sure that only one process is running the delayed refs
80 * at a time for a single extent. They also store the sum of all the
81 * reference count modifications we've queued up.
82 */
83struct btrfs_delayed_ref_head {
84 u64 bytenr;
85 u64 num_bytes;
86 /*
87 * For insertion into struct btrfs_delayed_ref_root::href_root.
88 * Keep it in the same cache line as 'bytenr' for more efficient
89 * searches in the rbtree.
90 */
91 struct rb_node href_node;
92 /*
93 * the mutex is held while running the refs, and it is also
94 * held when checking the sum of reference modifications.
95 */
96 struct mutex mutex;
97
98 refcount_t refs;
99
100 /* Protects 'ref_tree' and 'ref_add_list'. */
101 spinlock_t lock;
102 struct rb_root_cached ref_tree;
103 /* accumulate add BTRFS_ADD_DELAYED_REF nodes to this ref_add_list. */
104 struct list_head ref_add_list;
105
106 struct btrfs_delayed_extent_op *extent_op;
107
108 /*
109 * This is used to track the final ref_mod from all the refs associated
110 * with this head ref, this is not adjusted as delayed refs are run,
111 * this is meant to track if we need to do the csum accounting or not.
112 */
113 int total_ref_mod;
114
115 /*
116 * This is the current outstanding mod references for this bytenr. This
117 * is used with lookup_extent_info to get an accurate reference count
118 * for a bytenr, so it is adjusted as delayed refs are run so that any
119 * on disk reference count + ref_mod is accurate.
120 */
121 int ref_mod;
122
123 /*
124 * The root that triggered the allocation when must_insert_reserved is
125 * set to true.
126 */
127 u64 owning_root;
128
129 /*
130 * Track reserved bytes when setting must_insert_reserved. On success
131 * or cleanup, we will need to free the reservation.
132 */
133 u64 reserved_bytes;
134
135 /*
136 * when a new extent is allocated, it is just reserved in memory
137 * The actual extent isn't inserted into the extent allocation tree
138 * until the delayed ref is processed. must_insert_reserved is
139 * used to flag a delayed ref so the accounting can be updated
140 * when a full insert is done.
141 *
142 * It is possible the extent will be freed before it is ever
143 * inserted into the extent allocation tree. In this case
144 * we need to update the in ram accounting to properly reflect
145 * the free has happened.
146 */
147 bool must_insert_reserved;
148
149 bool is_data;
150 bool is_system;
151 bool processing;
152};
153
154struct btrfs_delayed_tree_ref {
155 struct btrfs_delayed_ref_node node;
156 u64 root;
157 u64 parent;
158 int level;
159};
160
161struct btrfs_delayed_data_ref {
162 struct btrfs_delayed_ref_node node;
163 u64 root;
164 u64 parent;
165 u64 objectid;
166 u64 offset;
167};
168
169enum btrfs_delayed_ref_flags {
170 /* Indicate that we are flushing delayed refs for the commit */
171 BTRFS_DELAYED_REFS_FLUSHING,
172};
173
174struct btrfs_delayed_ref_root {
175 /* head ref rbtree */
176 struct rb_root_cached href_root;
177
178 /* dirty extent records */
179 struct rb_root dirty_extent_root;
180
181 /* this spin lock protects the rbtree and the entries inside */
182 spinlock_t lock;
183
184 /* how many delayed ref updates we've queued, used by the
185 * throttling code
186 */
187 atomic_t num_entries;
188
189 /* total number of head nodes in tree */
190 unsigned long num_heads;
191
192 /* total number of head nodes ready for processing */
193 unsigned long num_heads_ready;
194
195 u64 pending_csums;
196
197 unsigned long flags;
198
199 u64 run_delayed_start;
200
201 /*
202 * To make qgroup to skip given root.
203 * This is for snapshot, as btrfs_qgroup_inherit() will manually
204 * modify counters for snapshot and its source, so we should skip
205 * the snapshot in new_root/old_roots or it will get calculated twice
206 */
207 u64 qgroup_to_skip;
208};
209
210enum btrfs_ref_type {
211 BTRFS_REF_NOT_SET,
212 BTRFS_REF_DATA,
213 BTRFS_REF_METADATA,
214 BTRFS_REF_LAST,
215} __packed;
216
217struct btrfs_data_ref {
218 /* For EXTENT_DATA_REF */
219
220 /* Root which owns this data reference. */
221 u64 ref_root;
222
223 /* Inode which refers to this data extent */
224 u64 ino;
225
226 /*
227 * file_offset - extent_offset
228 *
229 * file_offset is the key.offset of the EXTENT_DATA key.
230 * extent_offset is btrfs_file_extent_offset() of the EXTENT_DATA data.
231 */
232 u64 offset;
233};
234
235struct btrfs_tree_ref {
236 /*
237 * Level of this tree block
238 *
239 * Shared for skinny (TREE_BLOCK_REF) and normal tree ref.
240 */
241 int level;
242
243 /*
244 * Root which owns this tree block reference.
245 *
246 * For TREE_BLOCK_REF (skinny metadata, either inline or keyed)
247 */
248 u64 ref_root;
249
250 /* For non-skinny metadata, no special member needed */
251};
252
253struct btrfs_ref {
254 enum btrfs_ref_type type;
255 enum btrfs_delayed_ref_action action;
256
257 /*
258 * Whether this extent should go through qgroup record.
259 *
260 * Normally false, but for certain cases like delayed subtree scan,
261 * setting this flag can hugely reduce qgroup overhead.
262 */
263 bool skip_qgroup;
264
265#ifdef CONFIG_BTRFS_FS_REF_VERIFY
266 /* Through which root is this modification. */
267 u64 real_root;
268#endif
269 u64 bytenr;
270 u64 len;
271 u64 owning_root;
272
273 /* Bytenr of the parent tree block */
274 u64 parent;
275 union {
276 struct btrfs_data_ref data_ref;
277 struct btrfs_tree_ref tree_ref;
278 };
279};
280
281extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
282extern struct kmem_cache *btrfs_delayed_tree_ref_cachep;
283extern struct kmem_cache *btrfs_delayed_data_ref_cachep;
284extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
285
286int __init btrfs_delayed_ref_init(void);
287void __cold btrfs_delayed_ref_exit(void);
288
289static inline u64 btrfs_calc_delayed_ref_bytes(const struct btrfs_fs_info *fs_info,
290 int num_delayed_refs)
291{
292 u64 num_bytes;
293
294 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_delayed_refs);
295
296 /*
297 * We have to check the mount option here because we could be enabling
298 * the free space tree for the first time and don't have the compat_ro
299 * option set yet.
300 *
301 * We need extra reservations if we have the free space tree because
302 * we'll have to modify that tree as well.
303 */
304 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE))
305 num_bytes *= 2;
306
307 return num_bytes;
308}
309
310static inline u64 btrfs_calc_delayed_ref_csum_bytes(const struct btrfs_fs_info *fs_info,
311 int num_csum_items)
312{
313 /*
314 * Deleting csum items does not result in new nodes/leaves and does not
315 * require changing the free space tree, only the csum tree, so this is
316 * all we need.
317 */
318 return btrfs_calc_metadata_size(fs_info, num_csum_items);
319}
320
321void btrfs_init_generic_ref(struct btrfs_ref *generic_ref, int action, u64 bytenr,
322 u64 len, u64 parent, u64 owning_root);
323void btrfs_init_tree_ref(struct btrfs_ref *generic_ref, int level, u64 root,
324 u64 mod_root, bool skip_qgroup);
325void btrfs_init_data_ref(struct btrfs_ref *generic_ref, u64 ref_root, u64 ino,
326 u64 offset, u64 mod_root, bool skip_qgroup);
327
328static inline struct btrfs_delayed_extent_op *
329btrfs_alloc_delayed_extent_op(void)
330{
331 return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
332}
333
334static inline void
335btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
336{
337 if (op)
338 kmem_cache_free(btrfs_delayed_extent_op_cachep, op);
339}
340
341void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref);
342
343static inline u64 btrfs_ref_head_to_space_flags(
344 struct btrfs_delayed_ref_head *head_ref)
345{
346 if (head_ref->is_data)
347 return BTRFS_BLOCK_GROUP_DATA;
348 else if (head_ref->is_system)
349 return BTRFS_BLOCK_GROUP_SYSTEM;
350 return BTRFS_BLOCK_GROUP_METADATA;
351}
352
353static inline void btrfs_put_delayed_ref_head(struct btrfs_delayed_ref_head *head)
354{
355 if (refcount_dec_and_test(&head->refs))
356 kmem_cache_free(btrfs_delayed_ref_head_cachep, head);
357}
358
359int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
360 struct btrfs_ref *generic_ref,
361 struct btrfs_delayed_extent_op *extent_op);
362int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
363 struct btrfs_ref *generic_ref,
364 u64 reserved);
365int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
366 u64 bytenr, u64 num_bytes,
367 struct btrfs_delayed_extent_op *extent_op);
368void btrfs_merge_delayed_refs(struct btrfs_fs_info *fs_info,
369 struct btrfs_delayed_ref_root *delayed_refs,
370 struct btrfs_delayed_ref_head *head);
371
372struct btrfs_delayed_ref_head *
373btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
374 u64 bytenr);
375int btrfs_delayed_ref_lock(struct btrfs_delayed_ref_root *delayed_refs,
376 struct btrfs_delayed_ref_head *head);
377static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
378{
379 mutex_unlock(&head->mutex);
380}
381void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
382 struct btrfs_delayed_ref_head *head);
383
384struct btrfs_delayed_ref_head *btrfs_select_ref_head(
385 struct btrfs_delayed_ref_root *delayed_refs);
386
387int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq);
388
389void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr_refs, int nr_csums);
390void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans);
391void btrfs_inc_delayed_refs_rsv_bg_inserts(struct btrfs_fs_info *fs_info);
392void btrfs_dec_delayed_refs_rsv_bg_inserts(struct btrfs_fs_info *fs_info);
393void btrfs_inc_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
394void btrfs_dec_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
395int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
396 enum btrfs_reserve_flush_enum flush);
397void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
398 u64 num_bytes);
399bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info);
400
401/*
402 * helper functions to cast a node into its container
403 */
404static inline struct btrfs_delayed_tree_ref *
405btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
406{
407 return container_of(node, struct btrfs_delayed_tree_ref, node);
408}
409
410static inline struct btrfs_delayed_data_ref *
411btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
412{
413 return container_of(node, struct btrfs_delayed_data_ref, node);
414}
415
416#endif