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 
17 struct 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 
56 struct 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  */
71 struct 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 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 	 * 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 	/*
117 	 * 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;
129 	unsigned int is_data:1;
130 	unsigned int is_system:1;
131 	unsigned int processing:1;
132 };
133 
134 struct btrfs_delayed_tree_ref {
135 	struct btrfs_delayed_ref_node node;
136 	u64 root;
137 	u64 parent;
138 	int level;
139 };
140 
141 struct btrfs_delayed_data_ref {
142 	struct btrfs_delayed_ref_node node;
143 	u64 root;
144 	u64 parent;
145 	u64 objectid;
146 	u64 offset;
147 };
148 
149 struct btrfs_delayed_ref_root {
150 	/* head ref rbtree */
151 	struct rb_root href_root;
152 
153 	/* dirty extent records */
154 	struct rb_root dirty_extent_root;
155 
156 	/* 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 	 */
162 	atomic_t num_entries;
163 
164 	/* 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 
170 	u64 pending_csums;
171 
172 	/*
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;
178 
179 	u64 run_delayed_start;
180 
181 	/*
182 	 * To make qgroup to skip given root.
183 	 * This is for snapshot, as btrfs_qgroup_inherit() will manually
184 	 * 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;
188 };
189 
190 extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
191 extern struct kmem_cache *btrfs_delayed_tree_ref_cachep;
192 extern struct kmem_cache *btrfs_delayed_data_ref_cachep;
193 extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
194 
195 int __init btrfs_delayed_ref_init(void);
196 void __cold btrfs_delayed_ref_exit(void);
197 
198 static inline struct btrfs_delayed_extent_op *
btrfs_alloc_delayed_extent_op(void)199 btrfs_alloc_delayed_extent_op(void)
200 {
201 	return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
202 }
203 
204 static inline void
btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op * op)205 btrfs_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 
btrfs_put_delayed_ref(struct btrfs_delayed_ref_node * ref)211 static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
212 {
213 	WARN_ON(refcount_read(&ref->refs) == 0);
214 	if (refcount_dec_and_test(&ref->refs)) {
215 		WARN_ON(ref->in_tree);
216 		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;
225 		default:
226 			BUG();
227 		}
228 	}
229 }
230 
btrfs_put_delayed_ref_head(struct btrfs_delayed_ref_head * head)231 static 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 
237 int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
238 			       u64 bytenr, u64 num_bytes, u64 parent,
239 			       u64 ref_root, int level, int action,
240 			       struct btrfs_delayed_extent_op *extent_op,
241 			       int *old_ref_mod, int *new_ref_mod);
242 int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
243 			       u64 bytenr, u64 num_bytes,
244 			       u64 parent, u64 ref_root,
245 			       u64 owner, u64 offset, u64 reserved, int action,
246 			       int *old_ref_mod, int *new_ref_mod);
247 int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
248 				struct btrfs_trans_handle *trans,
249 				u64 bytenr, u64 num_bytes,
250 				struct btrfs_delayed_extent_op *extent_op);
251 void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
252 			      struct btrfs_delayed_ref_root *delayed_refs,
253 			      struct btrfs_delayed_ref_head *head);
254 
255 struct btrfs_delayed_ref_head *
256 btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
257 			    u64 bytenr);
258 int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
259 			   struct btrfs_delayed_ref_head *head);
btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head * head)260 static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
261 {
262 	mutex_unlock(&head->mutex);
263 }
264 
265 
266 struct btrfs_delayed_ref_head *
267 btrfs_select_ref_head(struct btrfs_trans_handle *trans);
268 
269 int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq);
270 
271 /*
272  * helper functions to cast a node into its container
273  */
274 static inline struct btrfs_delayed_tree_ref *
btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node * node)275 btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
276 {
277 	return container_of(node, struct btrfs_delayed_tree_ref, node);
278 }
279 
280 static inline struct btrfs_delayed_data_ref *
btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node * node)281 btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
282 {
283 	return container_of(node, struct btrfs_delayed_data_ref, node);
284 }
285 
286 #endif
287