]>
Commit | Line | Data |
---|---|---|
9888c340 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
fcebe456 JB |
2 | /* |
3 | * Copyright (C) 2014 Facebook. All rights reserved. | |
fcebe456 JB |
4 | */ |
5 | ||
9888c340 DS |
6 | #ifndef BTRFS_QGROUP_H |
7 | #define BTRFS_QGROUP_H | |
fcebe456 | 8 | |
3368d001 QW |
9 | #include "ulist.h" |
10 | #include "delayed-ref.h" | |
11 | ||
1d2beaa9 QW |
12 | /* |
13 | * Btrfs qgroup overview | |
14 | * | |
15 | * Btrfs qgroup splits into 3 main part: | |
16 | * 1) Reserve | |
17 | * Reserve metadata/data space for incoming operations | |
18 | * Affect how qgroup limit works | |
19 | * | |
20 | * 2) Trace | |
21 | * Tell btrfs qgroup to trace dirty extents. | |
22 | * | |
23 | * Dirty extents including: | |
24 | * - Newly allocated extents | |
25 | * - Extents going to be deleted (in this trans) | |
26 | * - Extents whose owner is going to be modified | |
27 | * | |
28 | * This is the main part affects whether qgroup numbers will stay | |
29 | * consistent. | |
30 | * Btrfs qgroup can trace clean extents and won't cause any problem, | |
31 | * but it will consume extra CPU time, it should be avoided if possible. | |
32 | * | |
33 | * 3) Account | |
34 | * Btrfs qgroup will updates its numbers, based on dirty extents traced | |
35 | * in previous step. | |
36 | * | |
37 | * Normally at qgroup rescan and transaction commit time. | |
38 | */ | |
39 | ||
3368d001 QW |
40 | /* |
41 | * Record a dirty extent, and info qgroup to update quota on it | |
42 | * TODO: Use kmem cache to alloc it. | |
43 | */ | |
44 | struct btrfs_qgroup_extent_record { | |
45 | struct rb_node node; | |
46 | u64 bytenr; | |
47 | u64 num_bytes; | |
48 | struct ulist *old_roots; | |
49 | }; | |
50 | ||
733e03a0 QW |
51 | /* |
52 | * Qgroup reservation types: | |
53 | * | |
54 | * DATA: | |
55 | * space reserved for data | |
56 | * | |
57 | * META_PERTRANS: | |
58 | * Space reserved for metadata (per-transaction) | |
59 | * Due to the fact that qgroup data is only updated at transaction commit | |
60 | * time, reserved space for metadata must be kept until transaction | |
61 | * commits. | |
62 | * Any metadata reserved that are used in btrfs_start_transaction() should | |
63 | * be of this type. | |
64 | * | |
65 | * META_PREALLOC: | |
66 | * There are cases where metadata space is reserved before starting | |
67 | * transaction, and then btrfs_join_transaction() to get a trans handle. | |
68 | * Any metadata reserved for such usage should be of this type. | |
69 | * And after join_transaction() part (or all) of such reservation should | |
70 | * be converted into META_PERTRANS. | |
71 | */ | |
d4e5c920 QW |
72 | enum btrfs_qgroup_rsv_type { |
73 | BTRFS_QGROUP_RSV_DATA = 0, | |
733e03a0 QW |
74 | BTRFS_QGROUP_RSV_META_PERTRANS, |
75 | BTRFS_QGROUP_RSV_META_PREALLOC, | |
d4e5c920 QW |
76 | BTRFS_QGROUP_RSV_LAST, |
77 | }; | |
78 | ||
79 | /* | |
80 | * Represents how many bytes we have reserved for this qgroup. | |
81 | * | |
82 | * Each type should have different reservation behavior. | |
83 | * E.g, data follows its io_tree flag modification, while | |
84 | * *currently* meta is just reserve-and-clear during transcation. | |
85 | * | |
86 | * TODO: Add new type for reservation which can survive transaction commit. | |
87 | * Currect metadata reservation behavior is not suitable for such case. | |
88 | */ | |
89 | struct btrfs_qgroup_rsv { | |
90 | u64 values[BTRFS_QGROUP_RSV_LAST]; | |
91 | }; | |
92 | ||
3159fe7b QW |
93 | /* |
94 | * one struct for each qgroup, organized in fs_info->qgroup_tree. | |
95 | */ | |
96 | struct btrfs_qgroup { | |
97 | u64 qgroupid; | |
98 | ||
99 | /* | |
100 | * state | |
101 | */ | |
102 | u64 rfer; /* referenced */ | |
103 | u64 rfer_cmpr; /* referenced compressed */ | |
104 | u64 excl; /* exclusive */ | |
105 | u64 excl_cmpr; /* exclusive compressed */ | |
106 | ||
107 | /* | |
108 | * limits | |
109 | */ | |
110 | u64 lim_flags; /* which limits are set */ | |
111 | u64 max_rfer; | |
112 | u64 max_excl; | |
113 | u64 rsv_rfer; | |
114 | u64 rsv_excl; | |
115 | ||
116 | /* | |
117 | * reservation tracking | |
118 | */ | |
d4e5c920 | 119 | struct btrfs_qgroup_rsv rsv; |
3159fe7b QW |
120 | |
121 | /* | |
122 | * lists | |
123 | */ | |
124 | struct list_head groups; /* groups this group is member of */ | |
125 | struct list_head members; /* groups that are members of this group */ | |
126 | struct list_head dirty; /* dirty groups */ | |
127 | struct rb_node node; /* tree of qgroups */ | |
128 | ||
129 | /* | |
130 | * temp variables for accounting operations | |
131 | * Refer to qgroup_shared_accounting() for details. | |
132 | */ | |
133 | u64 old_refcnt; | |
134 | u64 new_refcnt; | |
135 | }; | |
136 | ||
81fb6f77 QW |
137 | /* |
138 | * For qgroup event trace points only | |
139 | */ | |
140 | #define QGROUP_RESERVE (1<<0) | |
141 | #define QGROUP_RELEASE (1<<1) | |
142 | #define QGROUP_FREE (1<<2) | |
143 | ||
340f1aa2 NB |
144 | int btrfs_quota_enable(struct btrfs_fs_info *fs_info); |
145 | int btrfs_quota_disable(struct btrfs_fs_info *fs_info); | |
fcebe456 JB |
146 | int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info); |
147 | void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info); | |
d06f23d6 JM |
148 | int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info, |
149 | bool interruptible); | |
9f8a6ce6 LF |
150 | int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src, |
151 | u64 dst); | |
fcebe456 JB |
152 | int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, |
153 | struct btrfs_fs_info *fs_info, u64 src, u64 dst); | |
154 | int btrfs_create_qgroup(struct btrfs_trans_handle *trans, | |
4087cf24 | 155 | struct btrfs_fs_info *fs_info, u64 qgroupid); |
fcebe456 JB |
156 | int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, |
157 | struct btrfs_fs_info *fs_info, u64 qgroupid); | |
158 | int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, | |
159 | struct btrfs_fs_info *fs_info, u64 qgroupid, | |
160 | struct btrfs_qgroup_limit *limit); | |
161 | int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info); | |
162 | void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info); | |
163 | struct btrfs_delayed_extent_op; | |
d1b8b94a | 164 | |
cb93b52c | 165 | /* |
50b3e040 | 166 | * Inform qgroup to trace one dirty extent, its info is recorded in @record. |
fb235dc0 | 167 | * So qgroup can account it at transaction committing time. |
cb93b52c | 168 | * |
fb235dc0 QW |
169 | * No lock version, caller must acquire delayed ref lock and allocated memory, |
170 | * then call btrfs_qgroup_trace_extent_post() after exiting lock context. | |
cb93b52c QW |
171 | * |
172 | * Return 0 for success insert | |
173 | * Return >0 for existing record, caller can free @record safely. | |
174 | * Error is not possible | |
175 | */ | |
50b3e040 | 176 | int btrfs_qgroup_trace_extent_nolock( |
cb93b52c QW |
177 | struct btrfs_fs_info *fs_info, |
178 | struct btrfs_delayed_ref_root *delayed_refs, | |
179 | struct btrfs_qgroup_extent_record *record); | |
180 | ||
fb235dc0 QW |
181 | /* |
182 | * Post handler after qgroup_trace_extent_nolock(). | |
183 | * | |
184 | * NOTE: Current qgroup does the expensive backref walk at transaction | |
185 | * committing time with TRANS_STATE_COMMIT_DOING, this blocks incoming | |
186 | * new transaction. | |
187 | * This is designed to allow btrfs_find_all_roots() to get correct new_roots | |
188 | * result. | |
189 | * | |
190 | * However for old_roots there is no need to do backref walk at that time, | |
191 | * since we search commit roots to walk backref and result will always be | |
192 | * correct. | |
193 | * | |
194 | * Due to the nature of no lock version, we can't do backref there. | |
195 | * So we must call btrfs_qgroup_trace_extent_post() after exiting | |
196 | * spinlock context. | |
197 | * | |
198 | * TODO: If we can fix and prove btrfs_find_all_roots() can get correct result | |
199 | * using current root, then we can move all expensive backref walk out of | |
200 | * transaction committing, but not now as qgroup accounting will be wrong again. | |
201 | */ | |
202 | int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info, | |
203 | struct btrfs_qgroup_extent_record *qrecord); | |
204 | ||
cb93b52c | 205 | /* |
50b3e040 QW |
206 | * Inform qgroup to trace one dirty extent, specified by @bytenr and |
207 | * @num_bytes. | |
208 | * So qgroup can account it at commit trans time. | |
cb93b52c | 209 | * |
fb235dc0 QW |
210 | * Better encapsulated version, with memory allocation and backref walk for |
211 | * commit roots. | |
212 | * So this can sleep. | |
cb93b52c QW |
213 | * |
214 | * Return 0 if the operation is done. | |
215 | * Return <0 for error, like memory allocation failure or invalid parameter | |
216 | * (NULL trans) | |
217 | */ | |
50b3e040 | 218 | int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, |
cb93b52c QW |
219 | struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes, |
220 | gfp_t gfp_flag); | |
221 | ||
33d1f05c QW |
222 | /* |
223 | * Inform qgroup to trace all leaf items of data | |
224 | * | |
225 | * Return 0 for success | |
226 | * Return <0 for error(ENOMEM) | |
227 | */ | |
228 | int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans, | |
2ff7e61e | 229 | struct btrfs_fs_info *fs_info, |
33d1f05c QW |
230 | struct extent_buffer *eb); |
231 | /* | |
232 | * Inform qgroup to trace a whole subtree, including all its child tree | |
233 | * blocks and data. | |
234 | * The root tree block is specified by @root_eb. | |
235 | * | |
236 | * Normally used by relocation(tree block swap) and subvolume deletion. | |
237 | * | |
238 | * Return 0 for success | |
239 | * Return <0 for error(ENOMEM or tree search error) | |
240 | */ | |
241 | int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans, | |
242 | struct btrfs_root *root, | |
243 | struct extent_buffer *root_eb, | |
244 | u64 root_gen, int root_level); | |
442244c9 QW |
245 | int |
246 | btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, | |
247 | struct btrfs_fs_info *fs_info, | |
248 | u64 bytenr, u64 num_bytes, | |
249 | struct ulist *old_roots, struct ulist *new_roots); | |
460fb20a | 250 | int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans); |
fcebe456 JB |
251 | int btrfs_run_qgroups(struct btrfs_trans_handle *trans, |
252 | struct btrfs_fs_info *fs_info); | |
253 | int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, | |
254 | struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid, | |
255 | struct btrfs_qgroup_inherit *inherit); | |
297d750b | 256 | void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info, |
d4e5c920 QW |
257 | u64 ref_root, u64 num_bytes, |
258 | enum btrfs_qgroup_rsv_type type); | |
297d750b QW |
259 | static inline void btrfs_qgroup_free_delayed_ref(struct btrfs_fs_info *fs_info, |
260 | u64 ref_root, u64 num_bytes) | |
261 | { | |
bc074524 | 262 | trace_btrfs_qgroup_free_delayed_ref(fs_info, ref_root, num_bytes); |
d4e5c920 QW |
263 | btrfs_qgroup_free_refroot(fs_info, ref_root, num_bytes, |
264 | BTRFS_QGROUP_RSV_DATA); | |
297d750b | 265 | } |
fcebe456 JB |
266 | |
267 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS | |
268 | int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid, | |
269 | u64 rfer, u64 excl); | |
270 | #endif | |
271 | ||
52472553 | 272 | /* New io_tree based accurate qgroup reserve API */ |
364ecf36 QW |
273 | int btrfs_qgroup_reserve_data(struct inode *inode, |
274 | struct extent_changeset **reserved, u64 start, u64 len); | |
f695fdce | 275 | int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len); |
bc42bda2 QW |
276 | int btrfs_qgroup_free_data(struct inode *inode, |
277 | struct extent_changeset *reserved, u64 start, u64 len); | |
55eeaf05 | 278 | |
733e03a0 QW |
279 | int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes, |
280 | enum btrfs_qgroup_rsv_type type, bool enforce); | |
281 | /* Reserve metadata space for pertrans and prealloc type */ | |
282 | static inline int btrfs_qgroup_reserve_meta_pertrans(struct btrfs_root *root, | |
283 | int num_bytes, bool enforce) | |
284 | { | |
285 | return __btrfs_qgroup_reserve_meta(root, num_bytes, | |
286 | BTRFS_QGROUP_RSV_META_PERTRANS, enforce); | |
287 | } | |
288 | static inline int btrfs_qgroup_reserve_meta_prealloc(struct btrfs_root *root, | |
289 | int num_bytes, bool enforce) | |
290 | { | |
291 | return __btrfs_qgroup_reserve_meta(root, num_bytes, | |
292 | BTRFS_QGROUP_RSV_META_PREALLOC, enforce); | |
293 | } | |
294 | ||
295 | void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes, | |
296 | enum btrfs_qgroup_rsv_type type); | |
297 | ||
298 | /* Free per-transaction meta reservation for error handling */ | |
299 | static inline void btrfs_qgroup_free_meta_pertrans(struct btrfs_root *root, | |
300 | int num_bytes) | |
301 | { | |
302 | __btrfs_qgroup_free_meta(root, num_bytes, | |
303 | BTRFS_QGROUP_RSV_META_PERTRANS); | |
304 | } | |
305 | ||
306 | /* Pre-allocated meta reservation can be freed at need */ | |
307 | static inline void btrfs_qgroup_free_meta_prealloc(struct btrfs_root *root, | |
308 | int num_bytes) | |
309 | { | |
310 | __btrfs_qgroup_free_meta(root, num_bytes, | |
311 | BTRFS_QGROUP_RSV_META_PREALLOC); | |
312 | } | |
313 | ||
314 | /* | |
315 | * Per-transaction meta reservation should be all freed at transaction commit | |
316 | * time | |
317 | */ | |
318 | void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root); | |
319 | ||
64cfaef6 QW |
320 | /* |
321 | * Convert @num_bytes of META_PREALLOCATED reservation to META_PERTRANS. | |
322 | * | |
323 | * This is called when preallocated meta reservation needs to be used. | |
324 | * Normally after btrfs_join_transaction() call. | |
325 | */ | |
326 | void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes); | |
327 | ||
56fa9d07 | 328 | void btrfs_qgroup_check_reserved_leak(struct inode *inode); |
9888c340 DS |
329 | |
330 | #endif |