xref: /openbmc/linux/fs/btrfs/qgroup.h (revision e2c75e76)
1 /*
2  * Copyright (C) 2014 Facebook.  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 
19 #ifndef __BTRFS_QGROUP__
20 #define __BTRFS_QGROUP__
21 
22 #include "ulist.h"
23 #include "delayed-ref.h"
24 
25 /*
26  * Btrfs qgroup overview
27  *
28  * Btrfs qgroup splits into 3 main part:
29  * 1) Reserve
30  *    Reserve metadata/data space for incoming operations
31  *    Affect how qgroup limit works
32  *
33  * 2) Trace
34  *    Tell btrfs qgroup to trace dirty extents.
35  *
36  *    Dirty extents including:
37  *    - Newly allocated extents
38  *    - Extents going to be deleted (in this trans)
39  *    - Extents whose owner is going to be modified
40  *
41  *    This is the main part affects whether qgroup numbers will stay
42  *    consistent.
43  *    Btrfs qgroup can trace clean extents and won't cause any problem,
44  *    but it will consume extra CPU time, it should be avoided if possible.
45  *
46  * 3) Account
47  *    Btrfs qgroup will updates its numbers, based on dirty extents traced
48  *    in previous step.
49  *
50  *    Normally at qgroup rescan and transaction commit time.
51  */
52 
53 /*
54  * Record a dirty extent, and info qgroup to update quota on it
55  * TODO: Use kmem cache to alloc it.
56  */
57 struct btrfs_qgroup_extent_record {
58 	struct rb_node node;
59 	u64 bytenr;
60 	u64 num_bytes;
61 	struct ulist *old_roots;
62 };
63 
64 /*
65  * one struct for each qgroup, organized in fs_info->qgroup_tree.
66  */
67 struct btrfs_qgroup {
68 	u64 qgroupid;
69 
70 	/*
71 	 * state
72 	 */
73 	u64 rfer;	/* referenced */
74 	u64 rfer_cmpr;	/* referenced compressed */
75 	u64 excl;	/* exclusive */
76 	u64 excl_cmpr;	/* exclusive compressed */
77 
78 	/*
79 	 * limits
80 	 */
81 	u64 lim_flags;	/* which limits are set */
82 	u64 max_rfer;
83 	u64 max_excl;
84 	u64 rsv_rfer;
85 	u64 rsv_excl;
86 
87 	/*
88 	 * reservation tracking
89 	 */
90 	u64 reserved;
91 
92 	/*
93 	 * lists
94 	 */
95 	struct list_head groups;  /* groups this group is member of */
96 	struct list_head members; /* groups that are members of this group */
97 	struct list_head dirty;   /* dirty groups */
98 	struct rb_node node;	  /* tree of qgroups */
99 
100 	/*
101 	 * temp variables for accounting operations
102 	 * Refer to qgroup_shared_accounting() for details.
103 	 */
104 	u64 old_refcnt;
105 	u64 new_refcnt;
106 };
107 
108 /*
109  * For qgroup event trace points only
110  */
111 #define QGROUP_RESERVE		(1<<0)
112 #define QGROUP_RELEASE		(1<<1)
113 #define QGROUP_FREE		(1<<2)
114 
115 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
116 		       struct btrfs_fs_info *fs_info);
117 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
118 			struct btrfs_fs_info *fs_info);
119 int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
120 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
121 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
122 				     bool interruptible);
123 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
124 			      struct btrfs_fs_info *fs_info, u64 src, u64 dst);
125 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
126 			      struct btrfs_fs_info *fs_info, u64 src, u64 dst);
127 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
128 			struct btrfs_fs_info *fs_info, u64 qgroupid);
129 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
130 			      struct btrfs_fs_info *fs_info, u64 qgroupid);
131 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
132 		       struct btrfs_fs_info *fs_info, u64 qgroupid,
133 		       struct btrfs_qgroup_limit *limit);
134 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info);
135 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info);
136 struct btrfs_delayed_extent_op;
137 
138 /*
139  * Inform qgroup to trace one dirty extent, its info is recorded in @record.
140  * So qgroup can account it at transaction committing time.
141  *
142  * No lock version, caller must acquire delayed ref lock and allocated memory,
143  * then call btrfs_qgroup_trace_extent_post() after exiting lock context.
144  *
145  * Return 0 for success insert
146  * Return >0 for existing record, caller can free @record safely.
147  * Error is not possible
148  */
149 int btrfs_qgroup_trace_extent_nolock(
150 		struct btrfs_fs_info *fs_info,
151 		struct btrfs_delayed_ref_root *delayed_refs,
152 		struct btrfs_qgroup_extent_record *record);
153 
154 /*
155  * Post handler after qgroup_trace_extent_nolock().
156  *
157  * NOTE: Current qgroup does the expensive backref walk at transaction
158  * committing time with TRANS_STATE_COMMIT_DOING, this blocks incoming
159  * new transaction.
160  * This is designed to allow btrfs_find_all_roots() to get correct new_roots
161  * result.
162  *
163  * However for old_roots there is no need to do backref walk at that time,
164  * since we search commit roots to walk backref and result will always be
165  * correct.
166  *
167  * Due to the nature of no lock version, we can't do backref there.
168  * So we must call btrfs_qgroup_trace_extent_post() after exiting
169  * spinlock context.
170  *
171  * TODO: If we can fix and prove btrfs_find_all_roots() can get correct result
172  * using current root, then we can move all expensive backref walk out of
173  * transaction committing, but not now as qgroup accounting will be wrong again.
174  */
175 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
176 				   struct btrfs_qgroup_extent_record *qrecord);
177 
178 /*
179  * Inform qgroup to trace one dirty extent, specified by @bytenr and
180  * @num_bytes.
181  * So qgroup can account it at commit trans time.
182  *
183  * Better encapsulated version, with memory allocation and backref walk for
184  * commit roots.
185  * So this can sleep.
186  *
187  * Return 0 if the operation is done.
188  * Return <0 for error, like memory allocation failure or invalid parameter
189  * (NULL trans)
190  */
191 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
192 		struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
193 		gfp_t gfp_flag);
194 
195 /*
196  * Inform qgroup to trace all leaf items of data
197  *
198  * Return 0 for success
199  * Return <0 for error(ENOMEM)
200  */
201 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
202 				  struct btrfs_fs_info *fs_info,
203 				  struct extent_buffer *eb);
204 /*
205  * Inform qgroup to trace a whole subtree, including all its child tree
206  * blocks and data.
207  * The root tree block is specified by @root_eb.
208  *
209  * Normally used by relocation(tree block swap) and subvolume deletion.
210  *
211  * Return 0 for success
212  * Return <0 for error(ENOMEM or tree search error)
213  */
214 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
215 			       struct btrfs_root *root,
216 			       struct extent_buffer *root_eb,
217 			       u64 root_gen, int root_level);
218 int
219 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
220 			    struct btrfs_fs_info *fs_info,
221 			    u64 bytenr, u64 num_bytes,
222 			    struct ulist *old_roots, struct ulist *new_roots);
223 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
224 				 struct btrfs_fs_info *fs_info);
225 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
226 		      struct btrfs_fs_info *fs_info);
227 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
228 			 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
229 			 struct btrfs_qgroup_inherit *inherit);
230 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
231 			       u64 ref_root, u64 num_bytes);
232 static inline void btrfs_qgroup_free_delayed_ref(struct btrfs_fs_info *fs_info,
233 						 u64 ref_root, u64 num_bytes)
234 {
235 	trace_btrfs_qgroup_free_delayed_ref(fs_info, ref_root, num_bytes);
236 	btrfs_qgroup_free_refroot(fs_info, ref_root, num_bytes);
237 }
238 
239 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
240 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
241 			       u64 rfer, u64 excl);
242 #endif
243 
244 /* New io_tree based accurate qgroup reserve API */
245 int btrfs_qgroup_reserve_data(struct inode *inode,
246 			struct extent_changeset **reserved, u64 start, u64 len);
247 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len);
248 int btrfs_qgroup_free_data(struct inode *inode,
249 			struct extent_changeset *reserved, u64 start, u64 len);
250 
251 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
252 			      bool enforce);
253 void btrfs_qgroup_free_meta_all(struct btrfs_root *root);
254 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes);
255 void btrfs_qgroup_check_reserved_leak(struct inode *inode);
256 #endif /* __BTRFS_QGROUP__ */
257