xref: /linux/fs/btrfs/delayed-ref.h (revision 93df8a1ed6231727c5db94a80b1a6bd5ee67cec3)
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 
27 /*
28  * XXX: Qu: I really hate the design that ref_head and tree/data ref shares the
29  * same ref_node structure.
30  * Ref_head is in a higher logic level than tree/data ref, and duplicated
31  * bytenr/num_bytes in ref_node is really a waste or memory, they should be
32  * referred from ref_head.
33  * This gets more disgusting after we use list to store tree/data ref in
34  * ref_head. Must clean this mess up later.
35  */
36 struct btrfs_delayed_ref_node {
37 	/*
38 	 * ref_head use rb tree, stored in ref_root->href.
39 	 * indexed by bytenr
40 	 */
41 	struct rb_node rb_node;
42 
43 	/*data/tree ref use list, stored in ref_head->ref_list. */
44 	struct list_head list;
45 
46 	/* the starting bytenr of the extent */
47 	u64 bytenr;
48 
49 	/* the size of the extent */
50 	u64 num_bytes;
51 
52 	/* seq number to keep track of insertion order */
53 	u64 seq;
54 
55 	/* ref count on this data structure */
56 	atomic_t refs;
57 
58 	/*
59 	 * how many refs is this entry adding or deleting.  For
60 	 * head refs, this may be a negative number because it is keeping
61 	 * track of the total mods done to the reference count.
62 	 * For individual refs, this will always be a positive number
63 	 *
64 	 * It may be more than one, since it is possible for a single
65 	 * parent to have more than one ref on an extent
66 	 */
67 	int ref_mod;
68 
69 	unsigned int action:8;
70 	unsigned int type:8;
71 	unsigned int no_quota:1;
72 	/* is this node still in the rbtree? */
73 	unsigned int is_head:1;
74 	unsigned int in_tree:1;
75 };
76 
77 struct btrfs_delayed_extent_op {
78 	struct btrfs_disk_key key;
79 	u64 flags_to_set;
80 	int level;
81 	unsigned int update_key:1;
82 	unsigned int update_flags:1;
83 	unsigned int is_data:1;
84 };
85 
86 /*
87  * the head refs are used to hold a lock on a given extent, which allows us
88  * to make sure that only one process is running the delayed refs
89  * at a time for a single extent.  They also store the sum of all the
90  * reference count modifications we've queued up.
91  */
92 struct btrfs_delayed_ref_head {
93 	struct btrfs_delayed_ref_node node;
94 
95 	/*
96 	 * the mutex is held while running the refs, and it is also
97 	 * held when checking the sum of reference modifications.
98 	 */
99 	struct mutex mutex;
100 
101 	spinlock_t lock;
102 	struct list_head ref_list;
103 
104 	struct rb_node href_node;
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 	 * when a new extent is allocated, it is just reserved in memory
117 	 * The actual extent isn't inserted into the extent allocation tree
118 	 * until the delayed ref is processed.  must_insert_reserved is
119 	 * used to flag a delayed ref so the accounting can be updated
120 	 * when a full insert is done.
121 	 *
122 	 * It is possible the extent will be freed before it is ever
123 	 * inserted into the extent allocation tree.  In this case
124 	 * we need to update the in ram accounting to properly reflect
125 	 * the free has happened.
126 	 */
127 	unsigned int must_insert_reserved:1;
128 	unsigned int is_data:1;
129 	unsigned int processing:1;
130 };
131 
132 struct btrfs_delayed_tree_ref {
133 	struct btrfs_delayed_ref_node node;
134 	u64 root;
135 	u64 parent;
136 	int level;
137 };
138 
139 struct btrfs_delayed_data_ref {
140 	struct btrfs_delayed_ref_node node;
141 	u64 root;
142 	u64 parent;
143 	u64 objectid;
144 	u64 offset;
145 };
146 
147 struct btrfs_delayed_ref_root {
148 	/* head ref rbtree */
149 	struct rb_root href_root;
150 
151 	/* dirty extent records */
152 	struct rb_root dirty_extent_root;
153 
154 	/* this spin lock protects the rbtree and the entries inside */
155 	spinlock_t lock;
156 
157 	/* how many delayed ref updates we've queued, used by the
158 	 * throttling code
159 	 */
160 	atomic_t num_entries;
161 
162 	/* total number of head nodes in tree */
163 	unsigned long num_heads;
164 
165 	/* total number of head nodes ready for processing */
166 	unsigned long num_heads_ready;
167 
168 	u64 pending_csums;
169 
170 	/*
171 	 * set when the tree is flushing before a transaction commit,
172 	 * used by the throttling code to decide if new updates need
173 	 * to be run right away
174 	 */
175 	int flushing;
176 
177 	u64 run_delayed_start;
178 
179 	/*
180 	 * To make qgroup to skip given root.
181 	 * This is for snapshot, as btrfs_qgroup_inherit() will manully
182 	 * modify counters for snapshot and its source, so we should skip
183 	 * the snapshot in new_root/old_roots or it will get calculated twice
184 	 */
185 	u64 qgroup_to_skip;
186 };
187 
188 extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
189 extern struct kmem_cache *btrfs_delayed_tree_ref_cachep;
190 extern struct kmem_cache *btrfs_delayed_data_ref_cachep;
191 extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
192 
193 int btrfs_delayed_ref_init(void);
194 void btrfs_delayed_ref_exit(void);
195 
196 static inline struct btrfs_delayed_extent_op *
197 btrfs_alloc_delayed_extent_op(void)
198 {
199 	return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
200 }
201 
202 static inline void
203 btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
204 {
205 	if (op)
206 		kmem_cache_free(btrfs_delayed_extent_op_cachep, op);
207 }
208 
209 static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
210 {
211 	WARN_ON(atomic_read(&ref->refs) == 0);
212 	if (atomic_dec_and_test(&ref->refs)) {
213 		WARN_ON(ref->in_tree);
214 		switch (ref->type) {
215 		case BTRFS_TREE_BLOCK_REF_KEY:
216 		case BTRFS_SHARED_BLOCK_REF_KEY:
217 			kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
218 			break;
219 		case BTRFS_EXTENT_DATA_REF_KEY:
220 		case BTRFS_SHARED_DATA_REF_KEY:
221 			kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
222 			break;
223 		case 0:
224 			kmem_cache_free(btrfs_delayed_ref_head_cachep, ref);
225 			break;
226 		default:
227 			BUG();
228 		}
229 	}
230 }
231 
232 int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
233 			       struct btrfs_trans_handle *trans,
234 			       u64 bytenr, u64 num_bytes, u64 parent,
235 			       u64 ref_root, int level, int action,
236 			       struct btrfs_delayed_extent_op *extent_op,
237 			       int no_quota);
238 int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info,
239 			       struct btrfs_trans_handle *trans,
240 			       u64 bytenr, u64 num_bytes,
241 			       u64 parent, u64 ref_root,
242 			       u64 owner, u64 offset, int action,
243 			       struct btrfs_delayed_extent_op *extent_op,
244 			       int no_quota);
245 int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
246 				struct btrfs_trans_handle *trans,
247 				u64 bytenr, u64 num_bytes,
248 				struct btrfs_delayed_extent_op *extent_op);
249 void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
250 			      struct btrfs_fs_info *fs_info,
251 			      struct btrfs_delayed_ref_root *delayed_refs,
252 			      struct btrfs_delayed_ref_head *head);
253 
254 struct btrfs_delayed_ref_head *
255 btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr);
256 int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
257 			   struct btrfs_delayed_ref_head *head);
258 static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
259 {
260 	mutex_unlock(&head->mutex);
261 }
262 
263 
264 struct btrfs_delayed_ref_head *
265 btrfs_select_ref_head(struct btrfs_trans_handle *trans);
266 
267 int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info,
268 			    struct btrfs_delayed_ref_root *delayed_refs,
269 			    u64 seq);
270 
271 /*
272  * a node might live in a head or a regular ref, this lets you
273  * test for the proper type to use.
274  */
275 static int btrfs_delayed_ref_is_head(struct btrfs_delayed_ref_node *node)
276 {
277 	return node->is_head;
278 }
279 
280 /*
281  * helper functions to cast a node into its container
282  */
283 static inline struct btrfs_delayed_tree_ref *
284 btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
285 {
286 	WARN_ON(btrfs_delayed_ref_is_head(node));
287 	return container_of(node, struct btrfs_delayed_tree_ref, node);
288 }
289 
290 static inline struct btrfs_delayed_data_ref *
291 btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
292 {
293 	WARN_ON(btrfs_delayed_ref_is_head(node));
294 	return container_of(node, struct btrfs_delayed_data_ref, node);
295 }
296 
297 static inline struct btrfs_delayed_ref_head *
298 btrfs_delayed_node_to_head(struct btrfs_delayed_ref_node *node)
299 {
300 	WARN_ON(!btrfs_delayed_ref_is_head(node));
301 	return container_of(node, struct btrfs_delayed_ref_head, node);
302 }
303 #endif
304