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 /* ref count on this data structure */
37 atomic_t refs;
38
39 /*
40 * how many refs is this entry adding or deleting. For
41 * head refs, this may be a negative number because it is keeping
42 * track of the total mods done to the reference count.
43 * For individual refs, this will always be a positive number
44 *
45 * It may be more than one, since it is possible for a single
46 * parent to have more than one ref on an extent
47 */
48 int ref_mod;
49
50 unsigned int action:8;
51 unsigned int type:8;
52 /* is this node still in the rbtree? */
53 unsigned int is_head:1;
54 unsigned int in_tree:1;
55};
56
57struct btrfs_delayed_extent_op {
58 struct btrfs_disk_key key;
59 u64 flags_to_set;
60 unsigned int update_key:1;
61 unsigned int update_flags:1;
62 unsigned int is_data:1;
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 struct btrfs_delayed_ref_node node;
73
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 struct list_head cluster;
81
82 struct btrfs_delayed_extent_op *extent_op;
83 /*
84 * when a new extent is allocated, it is just reserved in memory
85 * The actual extent isn't inserted into the extent allocation tree
86 * until the delayed ref is processed. must_insert_reserved is
87 * used to flag a delayed ref so the accounting can be updated
88 * when a full insert is done.
89 *
90 * It is possible the extent will be freed before it is ever
91 * inserted into the extent allocation tree. In this case
92 * we need to update the in ram accounting to properly reflect
93 * the free has happened.
94 */
95 unsigned int must_insert_reserved:1;
96 unsigned int is_data:1;
97};
98
99struct btrfs_delayed_tree_ref {
100 struct btrfs_delayed_ref_node node;
101 union {
102 u64 root;
103 u64 parent;
104 };
105 int level;
106};
107
108struct btrfs_delayed_data_ref {
109 struct btrfs_delayed_ref_node node;
110 union {
111 u64 root;
112 u64 parent;
113 };
114 u64 objectid;
115 u64 offset;
116};
117
118struct btrfs_delayed_ref_root {
119 struct rb_root root;
120
121 /* this spin lock protects the rbtree and the entries inside */
122 spinlock_t lock;
123
124 /* how many delayed ref updates we've queued, used by the
125 * throttling code
126 */
127 unsigned long num_entries;
128
129 /* total number of head nodes in tree */
130 unsigned long num_heads;
131
132 /* total number of head nodes ready for processing */
133 unsigned long num_heads_ready;
134
135 /*
136 * set when the tree is flushing before a transaction commit,
137 * used by the throttling code to decide if new updates need
138 * to be run right away
139 */
140 int flushing;
141
142 u64 run_delayed_start;
143};
144
145static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
146{
147 WARN_ON(atomic_read(&ref->refs) == 0);
148 if (atomic_dec_and_test(&ref->refs)) {
149 WARN_ON(ref->in_tree);
150 kfree(ref);
151 }
152}
153
154int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
155 u64 bytenr, u64 num_bytes, u64 parent,
156 u64 ref_root, int level, int action,
157 struct btrfs_delayed_extent_op *extent_op);
158int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
159 u64 bytenr, u64 num_bytes,
160 u64 parent, u64 ref_root,
161 u64 owner, u64 offset, int action,
162 struct btrfs_delayed_extent_op *extent_op);
163int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
164 u64 bytenr, u64 num_bytes,
165 struct btrfs_delayed_extent_op *extent_op);
166
167struct btrfs_delayed_ref_head *
168btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr);
169int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
170 struct btrfs_delayed_ref_head *head);
171int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans,
172 struct list_head *cluster, u64 search_start);
173/*
174 * a node might live in a head or a regular ref, this lets you
175 * test for the proper type to use.
176 */
177static int btrfs_delayed_ref_is_head(struct btrfs_delayed_ref_node *node)
178{
179 return node->is_head;
180}
181
182/*
183 * helper functions to cast a node into its container
184 */
185static inline struct btrfs_delayed_tree_ref *
186btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
187{
188 WARN_ON(btrfs_delayed_ref_is_head(node));
189 return container_of(node, struct btrfs_delayed_tree_ref, node);
190}
191
192static inline struct btrfs_delayed_data_ref *
193btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
194{
195 WARN_ON(btrfs_delayed_ref_is_head(node));
196 return container_of(node, struct btrfs_delayed_data_ref, node);
197}
198
199static inline struct btrfs_delayed_ref_head *
200btrfs_delayed_node_to_head(struct btrfs_delayed_ref_node *node)
201{
202 WARN_ON(!btrfs_delayed_ref_is_head(node));
203 return container_of(node, struct btrfs_delayed_ref_head, node);
204}
205#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_data_ref {
34 /* For EXTENT_DATA_REF */
35
36 /* Inode which refers to this data extent */
37 u64 objectid;
38
39 /*
40 * file_offset - extent_offset
41 *
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.
44 */
45 u64 offset;
46};
47
48struct btrfs_tree_ref {
49 /*
50 * Level of this tree block.
51 *
52 * Shared for skinny (TREE_BLOCK_REF) and normal tree ref.
53 */
54 int level;
55
56 /* For non-skinny metadata, no special member needed */
57};
58
59struct btrfs_delayed_ref_node {
60 struct rb_node ref_node;
61 /*
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).
66 */
67 struct list_head add_list;
68
69 /* the starting bytenr of the extent */
70 u64 bytenr;
71
72 /* the size of the extent */
73 u64 num_bytes;
74
75 /* seq number to keep track of insertion order */
76 u64 seq;
77
78 /* The ref_root for this ref */
79 u64 ref_root;
80
81 /*
82 * The parent for this ref, if this isn't set the ref_root is the
83 * reference owner.
84 */
85 u64 parent;
86
87 /* ref count on this data structure */
88 refcount_t refs;
89
90 /*
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
95 *
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
98 */
99 int ref_mod;
100
101 unsigned int action:8;
102 unsigned int type:8;
103
104 union {
105 struct btrfs_tree_ref tree_ref;
106 struct btrfs_data_ref data_ref;
107 };
108};
109
110struct btrfs_delayed_extent_op {
111 struct btrfs_disk_key key;
112 bool update_key;
113 bool update_flags;
114 u64 flags_to_set;
115};
116
117/*
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.
122 */
123struct btrfs_delayed_ref_head {
124 u64 bytenr;
125 u64 num_bytes;
126 /*
127 * the mutex is held while running the refs, and it is also
128 * held when checking the sum of reference modifications.
129 */
130 struct mutex mutex;
131
132 refcount_t refs;
133
134 /* Protects 'ref_tree' and 'ref_add_list'. */
135 spinlock_t lock;
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;
139
140 struct btrfs_delayed_extent_op *extent_op;
141
142 /*
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.
146 */
147 int total_ref_mod;
148
149 /*
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.
154 */
155 int ref_mod;
156
157 /*
158 * The root that triggered the allocation when must_insert_reserved is
159 * set to true.
160 */
161 u64 owning_root;
162
163 /*
164 * Track reserved bytes when setting must_insert_reserved. On success
165 * or cleanup, we will need to free the reservation.
166 */
167 u64 reserved_bytes;
168
169 /* Tree block level, for metadata only. */
170 u8 level;
171
172 /*
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.
178 *
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.
183 */
184 bool must_insert_reserved;
185
186 bool is_data;
187 bool is_system;
188 bool processing;
189 /*
190 * Indicate if it's currently in the data structure that tracks head
191 * refs (struct btrfs_delayed_ref_root::head_refs).
192 */
193 bool tracked;
194};
195
196enum btrfs_delayed_ref_flags {
197 /* Indicate that we are flushing delayed refs for the commit */
198 BTRFS_DELAYED_REFS_FLUSHING,
199};
200
201struct btrfs_delayed_ref_root {
202 /*
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.
211 */
212 struct xarray head_refs;
213
214 /*
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.
218 */
219 struct xarray dirty_extents;
220
221 /*
222 * Protects the xarray head_refs, its entries and the following fields:
223 * num_heads, num_heads_ready, pending_csums and run_delayed_start.
224 */
225 spinlock_t lock;
226
227 /* Total number of head refs, protected by the spinlock 'lock'. */
228 unsigned long num_heads;
229
230 /*
231 * Total number of head refs ready for processing, protected by the
232 * spinlock 'lock'.
233 */
234 unsigned long num_heads_ready;
235
236 /*
237 * Track space reserved for deleting csums of data extents.
238 * Protected by the spinlock 'lock'.
239 */
240 u64 pending_csums;
241
242 unsigned long flags;
243
244 /*
245 * Track from which bytenr to start searching ref heads.
246 * Protected by the spinlock 'lock'.
247 */
248 u64 run_delayed_start;
249
250 /*
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
255 */
256 u64 qgroup_to_skip;
257};
258
259enum btrfs_ref_type {
260 BTRFS_REF_NOT_SET,
261 BTRFS_REF_DATA,
262 BTRFS_REF_METADATA,
263 BTRFS_REF_LAST,
264} __packed;
265
266struct btrfs_ref {
267 enum btrfs_ref_type type;
268 enum btrfs_delayed_ref_action action;
269
270 /*
271 * Whether this extent should go through qgroup record.
272 *
273 * Normally false, but for certain cases like delayed subtree scan,
274 * setting this flag can hugely reduce qgroup overhead.
275 */
276 bool skip_qgroup;
277
278#ifdef CONFIG_BTRFS_FS_REF_VERIFY
279 /* Through which root is this modification. */
280 u64 real_root;
281#endif
282 u64 bytenr;
283 u64 num_bytes;
284 u64 owning_root;
285
286 /*
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.
289 */
290 u64 ref_root;
291
292 /* Bytenr of the parent tree block */
293 u64 parent;
294 union {
295 struct btrfs_data_ref data_ref;
296 struct btrfs_tree_ref tree_ref;
297 };
298};
299
300extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
301extern struct kmem_cache *btrfs_delayed_ref_node_cachep;
302extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
303
304int __init btrfs_delayed_ref_init(void);
305void __cold btrfs_delayed_ref_exit(void);
306
307static inline u64 btrfs_calc_delayed_ref_bytes(const struct btrfs_fs_info *fs_info,
308 int num_delayed_refs)
309{
310 u64 num_bytes;
311
312 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_delayed_refs);
313
314 /*
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
317 * option set yet.
318 *
319 * We need extra reservations if we have the free space tree because
320 * we'll have to modify that tree as well.
321 */
322 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE))
323 num_bytes *= 2;
324
325 return num_bytes;
326}
327
328static inline u64 btrfs_calc_delayed_ref_csum_bytes(const struct btrfs_fs_info *fs_info,
329 int num_csum_items)
330{
331 /*
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
334 * all we need.
335 */
336 return btrfs_calc_metadata_size(fs_info, num_csum_items);
337}
338
339void btrfs_init_tree_ref(struct btrfs_ref *generic_ref, int level, u64 mod_root,
340 bool skip_qgroup);
341void btrfs_init_data_ref(struct btrfs_ref *generic_ref, u64 ino, u64 offset,
342 u64 mod_root, bool skip_qgroup);
343
344static inline struct btrfs_delayed_extent_op *
345btrfs_alloc_delayed_extent_op(void)
346{
347 return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
348}
349
350static inline void
351btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
352{
353 if (op)
354 kmem_cache_free(btrfs_delayed_extent_op_cachep, op);
355}
356
357void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref);
358
359static inline u64 btrfs_ref_head_to_space_flags(
360 struct btrfs_delayed_ref_head *head_ref)
361{
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;
367}
368
369static inline void btrfs_put_delayed_ref_head(struct btrfs_delayed_ref_head *head)
370{
371 if (refcount_dec_and_test(&head->refs))
372 kmem_cache_free(btrfs_delayed_ref_head_cachep, head);
373}
374
375int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
376 struct btrfs_ref *generic_ref,
377 struct btrfs_delayed_extent_op *extent_op);
378int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
379 struct btrfs_ref *generic_ref,
380 u64 reserved);
381int 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);
384void 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);
387
388struct btrfs_delayed_ref_head *
389btrfs_find_delayed_ref_head(const struct btrfs_fs_info *fs_info,
390 struct btrfs_delayed_ref_root *delayed_refs,
391 u64 bytenr);
392static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
393{
394 mutex_unlock(&head->mutex);
395}
396void 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);
399
400struct btrfs_delayed_ref_head *btrfs_select_ref_head(
401 const struct btrfs_fs_info *fs_info,
402 struct btrfs_delayed_ref_root *delayed_refs);
403void btrfs_unselect_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
404 struct btrfs_delayed_ref_head *head);
405
406int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq);
407
408void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr_refs, int nr_csums);
409void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans);
410void btrfs_inc_delayed_refs_rsv_bg_inserts(struct btrfs_fs_info *fs_info);
411void btrfs_dec_delayed_refs_rsv_bg_inserts(struct btrfs_fs_info *fs_info);
412void btrfs_inc_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
413void btrfs_dec_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
414int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
415 enum btrfs_reserve_flush_enum flush);
416bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info);
417bool btrfs_find_delayed_tree_ref(struct btrfs_delayed_ref_head *head,
418 u64 root, u64 parent);
419void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans);
420
421static inline u64 btrfs_delayed_ref_owner(struct btrfs_delayed_ref_node *node)
422{
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;
427}
428
429static inline u64 btrfs_delayed_ref_offset(struct btrfs_delayed_ref_node *node)
430{
431 if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
432 node->type == BTRFS_SHARED_DATA_REF_KEY)
433 return node->data_ref.offset;
434 return 0;
435}
436
437static inline u8 btrfs_ref_type(struct btrfs_ref *ref)
438{
439 ASSERT(ref->type == BTRFS_REF_DATA || ref->type == BTRFS_REF_METADATA);
440
441 if (ref->type == BTRFS_REF_DATA) {
442 if (ref->parent)
443 return BTRFS_SHARED_DATA_REF_KEY;
444 else
445 return BTRFS_EXTENT_DATA_REF_KEY;
446 } else {
447 if (ref->parent)
448 return BTRFS_SHARED_BLOCK_REF_KEY;
449 else
450 return BTRFS_TREE_BLOCK_REF_KEY;
451 }
452
453 return 0;
454}
455
456#endif