xref: /openbmc/linux/fs/btrfs/delayed-ref.h (revision cfbb9be8)
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 #include <linux/refcount.h>
22 
23 /* these are the possible values of struct btrfs_delayed_ref_node->action */
24 #define BTRFS_ADD_DELAYED_REF    1 /* add one backref to the tree */
25 #define BTRFS_DROP_DELAYED_REF   2 /* delete one backref from the tree */
26 #define BTRFS_ADD_DELAYED_EXTENT 3 /* record a full extent allocation */
27 #define BTRFS_UPDATE_DELAYED_HEAD 4 /* not changing ref count on head ref */
28 
29 struct btrfs_delayed_ref_node {
30 	struct rb_node ref_node;
31 	/*
32 	 * If action is BTRFS_ADD_DELAYED_REF, also link this node to
33 	 * ref_head->ref_add_list, then we do not need to iterate the
34 	 * whole ref_head->ref_list to find BTRFS_ADD_DELAYED_REF nodes.
35 	 */
36 	struct list_head add_list;
37 
38 	/* the starting bytenr of the extent */
39 	u64 bytenr;
40 
41 	/* the size of the extent */
42 	u64 num_bytes;
43 
44 	/* seq number to keep track of insertion order */
45 	u64 seq;
46 
47 	/* ref count on this data structure */
48 	refcount_t refs;
49 
50 	/*
51 	 * how many refs is this entry adding or deleting.  For
52 	 * head refs, this may be a negative number because it is keeping
53 	 * track of the total mods done to the reference count.
54 	 * For individual refs, this will always be a positive number
55 	 *
56 	 * It may be more than one, since it is possible for a single
57 	 * parent to have more than one ref on an extent
58 	 */
59 	int ref_mod;
60 
61 	unsigned int action:8;
62 	unsigned int type:8;
63 	/* is this node still in the rbtree? */
64 	unsigned int is_head:1;
65 	unsigned int in_tree:1;
66 };
67 
68 struct btrfs_delayed_extent_op {
69 	struct btrfs_disk_key key;
70 	u8 level;
71 	bool update_key;
72 	bool update_flags;
73 	bool is_data;
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  */
83 struct btrfs_delayed_ref_head {
84 	u64 bytenr;
85 	u64 num_bytes;
86 	refcount_t refs;
87 	/*
88 	 * the mutex is held while running the refs, and it is also
89 	 * held when checking the sum of reference modifications.
90 	 */
91 	struct mutex mutex;
92 
93 	spinlock_t lock;
94 	struct rb_root ref_tree;
95 	/* accumulate add BTRFS_ADD_DELAYED_REF nodes to this ref_add_list. */
96 	struct list_head ref_add_list;
97 
98 	struct rb_node href_node;
99 
100 	struct btrfs_delayed_extent_op *extent_op;
101 
102 	/*
103 	 * This is used to track the final ref_mod from all the refs associated
104 	 * with this head ref, this is not adjusted as delayed refs are run,
105 	 * this is meant to track if we need to do the csum accounting or not.
106 	 */
107 	int total_ref_mod;
108 
109 	/*
110 	 * This is the current outstanding mod references for this bytenr.  This
111 	 * is used with lookup_extent_info to get an accurate reference count
112 	 * for a bytenr, so it is adjusted as delayed refs are run so that any
113 	 * on disk reference count + ref_mod is accurate.
114 	 */
115 	int ref_mod;
116 
117 	/*
118 	 * For qgroup reserved space freeing.
119 	 *
120 	 * ref_root and reserved will be recorded after
121 	 * BTRFS_ADD_DELAYED_EXTENT is called.
122 	 * And will be used to free reserved qgroup space at
123 	 * run_delayed_refs() time.
124 	 */
125 	u64 qgroup_ref_root;
126 	u64 qgroup_reserved;
127 
128 	/*
129 	 * when a new extent is allocated, it is just reserved in memory
130 	 * The actual extent isn't inserted into the extent allocation tree
131 	 * until the delayed ref is processed.  must_insert_reserved is
132 	 * used to flag a delayed ref so the accounting can be updated
133 	 * when a full insert is done.
134 	 *
135 	 * It is possible the extent will be freed before it is ever
136 	 * inserted into the extent allocation tree.  In this case
137 	 * we need to update the in ram accounting to properly reflect
138 	 * the free has happened.
139 	 */
140 	unsigned int must_insert_reserved:1;
141 	unsigned int is_data:1;
142 	unsigned int processing:1;
143 };
144 
145 struct btrfs_delayed_tree_ref {
146 	struct btrfs_delayed_ref_node node;
147 	u64 root;
148 	u64 parent;
149 	int level;
150 };
151 
152 struct btrfs_delayed_data_ref {
153 	struct btrfs_delayed_ref_node node;
154 	u64 root;
155 	u64 parent;
156 	u64 objectid;
157 	u64 offset;
158 };
159 
160 struct btrfs_delayed_ref_root {
161 	/* head ref rbtree */
162 	struct rb_root href_root;
163 
164 	/* dirty extent records */
165 	struct rb_root dirty_extent_root;
166 
167 	/* this spin lock protects the rbtree and the entries inside */
168 	spinlock_t lock;
169 
170 	/* how many delayed ref updates we've queued, used by the
171 	 * throttling code
172 	 */
173 	atomic_t num_entries;
174 
175 	/* total number of head nodes in tree */
176 	unsigned long num_heads;
177 
178 	/* total number of head nodes ready for processing */
179 	unsigned long num_heads_ready;
180 
181 	u64 pending_csums;
182 
183 	/*
184 	 * set when the tree is flushing before a transaction commit,
185 	 * used by the throttling code to decide if new updates need
186 	 * to be run right away
187 	 */
188 	int flushing;
189 
190 	u64 run_delayed_start;
191 
192 	/*
193 	 * To make qgroup to skip given root.
194 	 * This is for snapshot, as btrfs_qgroup_inherit() will manually
195 	 * modify counters for snapshot and its source, so we should skip
196 	 * the snapshot in new_root/old_roots or it will get calculated twice
197 	 */
198 	u64 qgroup_to_skip;
199 };
200 
201 extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
202 extern struct kmem_cache *btrfs_delayed_tree_ref_cachep;
203 extern struct kmem_cache *btrfs_delayed_data_ref_cachep;
204 extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
205 
206 int __init btrfs_delayed_ref_init(void);
207 void btrfs_delayed_ref_exit(void);
208 
209 static inline struct btrfs_delayed_extent_op *
210 btrfs_alloc_delayed_extent_op(void)
211 {
212 	return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
213 }
214 
215 static inline void
216 btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
217 {
218 	if (op)
219 		kmem_cache_free(btrfs_delayed_extent_op_cachep, op);
220 }
221 
222 static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
223 {
224 	WARN_ON(refcount_read(&ref->refs) == 0);
225 	if (refcount_dec_and_test(&ref->refs)) {
226 		WARN_ON(ref->in_tree);
227 		switch (ref->type) {
228 		case BTRFS_TREE_BLOCK_REF_KEY:
229 		case BTRFS_SHARED_BLOCK_REF_KEY:
230 			kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
231 			break;
232 		case BTRFS_EXTENT_DATA_REF_KEY:
233 		case BTRFS_SHARED_DATA_REF_KEY:
234 			kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
235 			break;
236 		default:
237 			BUG();
238 		}
239 	}
240 }
241 
242 static inline void btrfs_put_delayed_ref_head(struct btrfs_delayed_ref_head *head)
243 {
244 	if (refcount_dec_and_test(&head->refs))
245 		kmem_cache_free(btrfs_delayed_ref_head_cachep, head);
246 }
247 
248 int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
249 			       struct btrfs_trans_handle *trans,
250 			       u64 bytenr, u64 num_bytes, u64 parent,
251 			       u64 ref_root, int level, int action,
252 			       struct btrfs_delayed_extent_op *extent_op,
253 			       int *old_ref_mod, int *new_ref_mod);
254 int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info,
255 			       struct btrfs_trans_handle *trans,
256 			       u64 bytenr, u64 num_bytes,
257 			       u64 parent, u64 ref_root,
258 			       u64 owner, u64 offset, u64 reserved, int action,
259 			       int *old_ref_mod, int *new_ref_mod);
260 int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
261 				struct btrfs_trans_handle *trans,
262 				u64 bytenr, u64 num_bytes,
263 				struct btrfs_delayed_extent_op *extent_op);
264 void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
265 			      struct btrfs_fs_info *fs_info,
266 			      struct btrfs_delayed_ref_root *delayed_refs,
267 			      struct btrfs_delayed_ref_head *head);
268 
269 struct btrfs_delayed_ref_head *
270 btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
271 			    u64 bytenr);
272 int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
273 			   struct btrfs_delayed_ref_head *head);
274 static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
275 {
276 	mutex_unlock(&head->mutex);
277 }
278 
279 
280 struct btrfs_delayed_ref_head *
281 btrfs_select_ref_head(struct btrfs_trans_handle *trans);
282 
283 int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info,
284 			    struct btrfs_delayed_ref_root *delayed_refs,
285 			    u64 seq);
286 
287 /*
288  * helper functions to cast a node into its container
289  */
290 static inline struct btrfs_delayed_tree_ref *
291 btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
292 {
293 	return container_of(node, struct btrfs_delayed_tree_ref, node);
294 }
295 
296 static inline struct btrfs_delayed_data_ref *
297 btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
298 {
299 	return container_of(node, struct btrfs_delayed_data_ref, node);
300 }
301 #endif
302