blob: d9f2a4ebd5db86d8637d567ce626903bb37c0747 [file] [log] [blame]
David Sterba9888c342018-04-03 19:16:55 +02001/* SPDX-License-Identifier: GPL-2.0 */
Chris Mason56bec292009-03-13 10:10:06 -04002/*
3 * Copyright (C) 2008 Oracle. All rights reserved.
Chris Mason56bec292009-03-13 10:10:06 -04004 */
David Sterba9888c342018-04-03 19:16:55 +02005
6#ifndef BTRFS_DELAYED_REF_H
7#define BTRFS_DELAYED_REF_H
Chris Mason56bec292009-03-13 10:10:06 -04008
Elena Reshetova6df8cdf2017-03-03 10:55:15 +02009#include <linux/refcount.h>
10
Wang Sheng-Hui44a075b2012-09-21 08:21:01 +080011/* these are the possible values of struct btrfs_delayed_ref_node->action */
Chris Mason56bec292009-03-13 10:10:06 -040012#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 */
Chris Mason1a81af42009-03-25 09:55:11 -040015#define BTRFS_UPDATE_DELAYED_HEAD 4 /* not changing ref count on head ref */
Chris Mason56bec292009-03-13 10:10:06 -040016
17struct btrfs_delayed_ref_node {
Josef Bacik0e0adbc2017-10-19 14:16:00 -040018 struct rb_node ref_node;
Wang Xiaoguang1d57ee92016-10-26 18:07:33 +080019 /*
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;
Qu Wenruoc6fc2452015-03-30 17:03:00 +080025
Chris Mason56bec292009-03-13 10:10:06 -040026 /* the starting bytenr of the extent */
27 u64 bytenr;
28
Chris Mason56bec292009-03-13 10:10:06 -040029 /* the size of the extent */
30 u64 num_bytes;
31
Arne Jansen00f04b82011-09-14 12:37:00 +020032 /* seq number to keep track of insertion order */
33 u64 seq;
34
Chris Mason56bec292009-03-13 10:10:06 -040035 /* ref count on this data structure */
Elena Reshetova6df8cdf2017-03-03 10:55:15 +020036 refcount_t refs;
Chris Mason56bec292009-03-13 10:10:06 -040037
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
Yan Zheng5d4f98a2009-06-10 10:45:14 -040049 unsigned int action:8;
50 unsigned int type:8;
Chris Mason56bec292009-03-13 10:10:06 -040051 /* is this node still in the rbtree? */
Yan Zheng5d4f98a2009-06-10 10:45:14 -040052 unsigned int is_head:1;
Chris Mason56bec292009-03-13 10:10:06 -040053 unsigned int in_tree:1;
54};
55
Yan Zheng5d4f98a2009-06-10 10:45:14 -040056struct btrfs_delayed_extent_op {
57 struct btrfs_disk_key key;
David Sterba35b3ad52015-11-30 16:51:29 +010058 u8 level;
59 bool update_key;
60 bool update_flags;
61 bool is_data;
Yan Zheng5d4f98a2009-06-10 10:45:14 -040062 u64 flags_to_set;
Yan Zheng5d4f98a2009-06-10 10:45:14 -040063};
64
Chris Mason56bec292009-03-13 10:10:06 -040065/*
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 {
Josef Bacikd2788502017-09-29 15:43:57 -040072 u64 bytenr;
73 u64 num_bytes;
74 refcount_t refs;
Chris Mason56bec292009-03-13 10:10:06 -040075 /*
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
Josef Bacikd7df2c72014-01-23 09:21:38 -050081 spinlock_t lock;
Josef Bacik0e0adbc2017-10-19 14:16:00 -040082 struct rb_root ref_tree;
Wang Xiaoguang1d57ee92016-10-26 18:07:33 +080083 /* accumulate add BTRFS_ADD_DELAYED_REF nodes to this ref_add_list. */
84 struct list_head ref_add_list;
Chris Masonc3e69d52009-03-13 10:17:05 -040085
Liu Boc46effa2013-10-14 12:59:45 +080086 struct rb_node href_node;
87
Yan Zheng5d4f98a2009-06-10 10:45:14 -040088 struct btrfs_delayed_extent_op *extent_op;
Josef Bacik12621332015-02-03 07:50:16 -080089
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
Chris Mason56bec292009-03-13 10:10:06 -040097 /*
Josef Bacikd2788502017-09-29 15:43:57 -040098 * 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 /*
Qu Wenruof64d5ca2015-09-08 17:08:36 +0800106 * For qgroup reserved space freeing.
107 *
108 * ref_root and reserved will be recorded after
109 * BTRFS_ADD_DELAYED_EXTENT is called.
110 * And will be used to free reserved qgroup space at
111 * run_delayed_refs() time.
112 */
113 u64 qgroup_ref_root;
114 u64 qgroup_reserved;
115
116 /*
Chris Mason56bec292009-03-13 10:10:06 -0400117 * when a new extent is allocated, it is just reserved in memory
118 * The actual extent isn't inserted into the extent allocation tree
119 * until the delayed ref is processed. must_insert_reserved is
120 * used to flag a delayed ref so the accounting can be updated
121 * when a full insert is done.
122 *
123 * It is possible the extent will be freed before it is ever
124 * inserted into the extent allocation tree. In this case
125 * we need to update the in ram accounting to properly reflect
126 * the free has happened.
127 */
128 unsigned int must_insert_reserved:1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400129 unsigned int is_data:1;
Nikolay Borisov5e388e92018-04-18 09:41:54 +0300130 unsigned int is_system:1;
Josef Bacikd7df2c72014-01-23 09:21:38 -0500131 unsigned int processing:1;
Chris Mason56bec292009-03-13 10:10:06 -0400132};
133
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400134struct btrfs_delayed_tree_ref {
Chris Mason56bec292009-03-13 10:10:06 -0400135 struct btrfs_delayed_ref_node node;
Arne Janseneebe0632011-09-14 14:01:24 +0200136 u64 root;
137 u64 parent;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400138 int level;
139};
Chris Mason56bec292009-03-13 10:10:06 -0400140
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400141struct btrfs_delayed_data_ref {
142 struct btrfs_delayed_ref_node node;
Arne Janseneebe0632011-09-14 14:01:24 +0200143 u64 root;
144 u64 parent;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400145 u64 objectid;
146 u64 offset;
Chris Mason56bec292009-03-13 10:10:06 -0400147};
148
149struct btrfs_delayed_ref_root {
Liu Boc46effa2013-10-14 12:59:45 +0800150 /* head ref rbtree */
151 struct rb_root href_root;
152
Qu Wenruo3368d002015-04-16 14:34:17 +0800153 /* dirty extent records */
154 struct rb_root dirty_extent_root;
155
Chris Mason56bec292009-03-13 10:10:06 -0400156 /* this spin lock protects the rbtree and the entries inside */
157 spinlock_t lock;
158
159 /* how many delayed ref updates we've queued, used by the
160 * throttling code
161 */
Josef Bacikd7df2c72014-01-23 09:21:38 -0500162 atomic_t num_entries;
Chris Mason56bec292009-03-13 10:10:06 -0400163
Chris Masonc3e69d52009-03-13 10:17:05 -0400164 /* total number of head nodes in tree */
165 unsigned long num_heads;
166
167 /* total number of head nodes ready for processing */
168 unsigned long num_heads_ready;
169
Josef Bacik12621332015-02-03 07:50:16 -0800170 u64 pending_csums;
171
Chris Mason56bec292009-03-13 10:10:06 -0400172 /*
173 * set when the tree is flushing before a transaction commit,
174 * used by the throttling code to decide if new updates need
175 * to be run right away
176 */
177 int flushing;
Chris Masonc3e69d52009-03-13 10:17:05 -0400178
179 u64 run_delayed_start;
Qu Wenruo9086db82015-04-20 09:53:50 +0800180
181 /*
182 * To make qgroup to skip given root.
Nicholas D Steeves01327612016-05-19 21:18:45 -0400183 * This is for snapshot, as btrfs_qgroup_inherit() will manually
Qu Wenruo9086db82015-04-20 09:53:50 +0800184 * modify counters for snapshot and its source, so we should skip
185 * the snapshot in new_root/old_roots or it will get calculated twice
186 */
187 u64 qgroup_to_skip;
Chris Mason56bec292009-03-13 10:10:06 -0400188};
189
Miao Xie78a61842012-11-21 02:21:28 +0000190extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
191extern struct kmem_cache *btrfs_delayed_tree_ref_cachep;
192extern struct kmem_cache *btrfs_delayed_data_ref_cachep;
193extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
194
Liu Bof5c29bd2017-11-02 17:21:50 -0600195int __init btrfs_delayed_ref_init(void);
David Sterbae67c7182018-02-19 17:24:18 +0100196void __cold btrfs_delayed_ref_exit(void);
Miao Xie78a61842012-11-21 02:21:28 +0000197
198static inline struct btrfs_delayed_extent_op *
199btrfs_alloc_delayed_extent_op(void)
200{
201 return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
202}
203
204static inline void
205btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
206{
207 if (op)
208 kmem_cache_free(btrfs_delayed_extent_op_cachep, op);
209}
210
Chris Mason56bec292009-03-13 10:10:06 -0400211static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
212{
Elena Reshetova6df8cdf2017-03-03 10:55:15 +0200213 WARN_ON(refcount_read(&ref->refs) == 0);
214 if (refcount_dec_and_test(&ref->refs)) {
Chris Mason56bec292009-03-13 10:10:06 -0400215 WARN_ON(ref->in_tree);
Miao Xie78a61842012-11-21 02:21:28 +0000216 switch (ref->type) {
217 case BTRFS_TREE_BLOCK_REF_KEY:
218 case BTRFS_SHARED_BLOCK_REF_KEY:
219 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
220 break;
221 case BTRFS_EXTENT_DATA_REF_KEY:
222 case BTRFS_SHARED_DATA_REF_KEY:
223 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
224 break;
Miao Xie78a61842012-11-21 02:21:28 +0000225 default:
226 BUG();
227 }
Chris Mason56bec292009-03-13 10:10:06 -0400228 }
229}
230
Josef Bacikd2788502017-09-29 15:43:57 -0400231static inline void btrfs_put_delayed_ref_head(struct btrfs_delayed_ref_head *head)
232{
233 if (refcount_dec_and_test(&head->refs))
234 kmem_cache_free(btrfs_delayed_ref_head_cachep, head);
235}
236
Nikolay Borisov44e1c472018-06-20 15:48:53 +0300237int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400238 u64 bytenr, u64 num_bytes, u64 parent,
239 u64 ref_root, int level, int action,
Omar Sandoval7be07912017-06-06 16:45:30 -0700240 struct btrfs_delayed_extent_op *extent_op,
241 int *old_ref_mod, int *new_ref_mod);
Nikolay Borisov88a979c2018-06-20 15:48:54 +0300242int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400243 u64 bytenr, u64 num_bytes,
244 u64 parent, u64 ref_root,
Omar Sandoval7be07912017-06-06 16:45:30 -0700245 u64 owner, u64 offset, u64 reserved, int action,
246 int *old_ref_mod, int *new_ref_mod);
Arne Jansen66d7e7f2011-09-12 15:26:38 +0200247int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
248 struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400249 u64 bytenr, u64 num_bytes,
250 struct btrfs_delayed_extent_op *extent_op);
Josef Bacikae1e2062012-08-07 16:00:32 -0400251void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
Josef Bacikae1e2062012-08-07 16:00:32 -0400252 struct btrfs_delayed_ref_root *delayed_refs,
253 struct btrfs_delayed_ref_head *head);
Chris Mason56bec292009-03-13 10:10:06 -0400254
Chris Mason1887be62009-03-13 10:11:24 -0400255struct btrfs_delayed_ref_head *
Liu Bof72ad18e2017-01-30 12:24:37 -0800256btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
257 u64 bytenr);
Chris Masonc3e69d52009-03-13 10:17:05 -0400258int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
259 struct btrfs_delayed_ref_head *head);
Miao Xie093486c2012-12-19 08:10:10 +0000260static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
261{
262 mutex_unlock(&head->mutex);
263}
264
Josef Bacikd7df2c72014-01-23 09:21:38 -0500265
266struct btrfs_delayed_ref_head *
267btrfs_select_ref_head(struct btrfs_trans_handle *trans);
Arne Jansen00f04b82011-09-14 12:37:00 +0200268
Nikolay Borisov41d0bd32018-04-04 15:57:42 +0300269int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq);
Arne Jansen00f04b82011-09-14 12:37:00 +0200270
271/*
Chris Mason56bec292009-03-13 10:10:06 -0400272 * helper functions to cast a node into its container
273 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400274static inline struct btrfs_delayed_tree_ref *
275btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
Chris Mason56bec292009-03-13 10:10:06 -0400276{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400277 return container_of(node, struct btrfs_delayed_tree_ref, node);
278}
Chris Mason56bec292009-03-13 10:10:06 -0400279
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400280static inline struct btrfs_delayed_data_ref *
281btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
282{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400283 return container_of(node, struct btrfs_delayed_data_ref, node);
Chris Mason56bec292009-03-13 10:10:06 -0400284}
David Sterba9888c342018-04-03 19:16:55 +0200285
Chris Mason56bec292009-03-13 10:10:06 -0400286#endif