xref: /linux/fs/btrfs/qgroup.h (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
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  * For qgroup event trace points only
66  */
67 #define QGROUP_RESERVE		(1<<0)
68 #define QGROUP_RELEASE		(1<<1)
69 #define QGROUP_FREE		(1<<2)
70 
71 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
72 		       struct btrfs_fs_info *fs_info);
73 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
74 			struct btrfs_fs_info *fs_info);
75 int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
76 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
77 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
78 				     bool interruptible);
79 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
80 			      struct btrfs_fs_info *fs_info, u64 src, u64 dst);
81 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
82 			      struct btrfs_fs_info *fs_info, u64 src, u64 dst);
83 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
84 			struct btrfs_fs_info *fs_info, u64 qgroupid);
85 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
86 			      struct btrfs_fs_info *fs_info, u64 qgroupid);
87 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
88 		       struct btrfs_fs_info *fs_info, u64 qgroupid,
89 		       struct btrfs_qgroup_limit *limit);
90 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info);
91 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info);
92 struct btrfs_delayed_extent_op;
93 int btrfs_qgroup_prepare_account_extents(struct btrfs_trans_handle *trans,
94 					 struct btrfs_fs_info *fs_info);
95 /*
96  * Inform qgroup to trace one dirty extent, its info is recorded in @record.
97  * So qgroup can account it at transaction committing time.
98  *
99  * No lock version, caller must acquire delayed ref lock and allocated memory,
100  * then call btrfs_qgroup_trace_extent_post() after exiting lock context.
101  *
102  * Return 0 for success insert
103  * Return >0 for existing record, caller can free @record safely.
104  * Error is not possible
105  */
106 int btrfs_qgroup_trace_extent_nolock(
107 		struct btrfs_fs_info *fs_info,
108 		struct btrfs_delayed_ref_root *delayed_refs,
109 		struct btrfs_qgroup_extent_record *record);
110 
111 /*
112  * Post handler after qgroup_trace_extent_nolock().
113  *
114  * NOTE: Current qgroup does the expensive backref walk at transaction
115  * committing time with TRANS_STATE_COMMIT_DOING, this blocks incoming
116  * new transaction.
117  * This is designed to allow btrfs_find_all_roots() to get correct new_roots
118  * result.
119  *
120  * However for old_roots there is no need to do backref walk at that time,
121  * since we search commit roots to walk backref and result will always be
122  * correct.
123  *
124  * Due to the nature of no lock version, we can't do backref there.
125  * So we must call btrfs_qgroup_trace_extent_post() after exiting
126  * spinlock context.
127  *
128  * TODO: If we can fix and prove btrfs_find_all_roots() can get correct result
129  * using current root, then we can move all expensive backref walk out of
130  * transaction committing, but not now as qgroup accounting will be wrong again.
131  */
132 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
133 				   struct btrfs_qgroup_extent_record *qrecord);
134 
135 /*
136  * Inform qgroup to trace one dirty extent, specified by @bytenr and
137  * @num_bytes.
138  * So qgroup can account it at commit trans time.
139  *
140  * Better encapsulated version, with memory allocation and backref walk for
141  * commit roots.
142  * So this can sleep.
143  *
144  * Return 0 if the operation is done.
145  * Return <0 for error, like memory allocation failure or invalid parameter
146  * (NULL trans)
147  */
148 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
149 		struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
150 		gfp_t gfp_flag);
151 
152 /*
153  * Inform qgroup to trace all leaf items of data
154  *
155  * Return 0 for success
156  * Return <0 for error(ENOMEM)
157  */
158 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
159 				  struct btrfs_fs_info *fs_info,
160 				  struct extent_buffer *eb);
161 /*
162  * Inform qgroup to trace a whole subtree, including all its child tree
163  * blocks and data.
164  * The root tree block is specified by @root_eb.
165  *
166  * Normally used by relocation(tree block swap) and subvolume deletion.
167  *
168  * Return 0 for success
169  * Return <0 for error(ENOMEM or tree search error)
170  */
171 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
172 			       struct btrfs_root *root,
173 			       struct extent_buffer *root_eb,
174 			       u64 root_gen, int root_level);
175 int
176 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
177 			    struct btrfs_fs_info *fs_info,
178 			    u64 bytenr, u64 num_bytes,
179 			    struct ulist *old_roots, struct ulist *new_roots);
180 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
181 				 struct btrfs_fs_info *fs_info);
182 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
183 		      struct btrfs_fs_info *fs_info);
184 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
185 			 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
186 			 struct btrfs_qgroup_inherit *inherit);
187 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
188 			       u64 ref_root, u64 num_bytes);
189 /*
190  * TODO: Add proper trace point for it, as btrfs_qgroup_free() is
191  * called by everywhere, can't provide good trace for delayed ref case.
192  */
193 static inline void btrfs_qgroup_free_delayed_ref(struct btrfs_fs_info *fs_info,
194 						 u64 ref_root, u64 num_bytes)
195 {
196 	btrfs_qgroup_free_refroot(fs_info, ref_root, num_bytes);
197 	trace_btrfs_qgroup_free_delayed_ref(fs_info, ref_root, num_bytes);
198 }
199 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans);
200 
201 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
202 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
203 			       u64 rfer, u64 excl);
204 #endif
205 
206 /* New io_tree based accurate qgroup reserve API */
207 int btrfs_qgroup_reserve_data(struct inode *inode, u64 start, u64 len);
208 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len);
209 int btrfs_qgroup_free_data(struct inode *inode, u64 start, u64 len);
210 
211 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
212 			      bool enforce);
213 void btrfs_qgroup_free_meta_all(struct btrfs_root *root);
214 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes);
215 void btrfs_qgroup_check_reserved_leak(struct inode *inode);
216 #endif /* __BTRFS_QGROUP__ */
217