1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Copyright (C) 2014 Facebook. All rights reserved.
4 */
5
6 #ifndef BTRFS_QGROUP_H
7 #define BTRFS_QGROUP_H
8
9 #include <linux/types.h>
10 #include <linux/spinlock.h>
11 #include <linux/rbtree.h>
12 #include <linux/kobject.h>
13 #include <linux/list.h>
14 #include <uapi/linux/btrfs_tree.h>
15
16 struct extent_buffer;
17 struct extent_changeset;
18 struct btrfs_delayed_extent_op;
19 struct btrfs_fs_info;
20 struct btrfs_root;
21 struct btrfs_ioctl_quota_ctl_args;
22 struct btrfs_trans_handle;
23 struct btrfs_delayed_ref_root;
24 struct btrfs_inode;
25
26 /*
27 * Btrfs qgroup overview
28 *
29 * Btrfs qgroup splits into 3 main part:
30 * 1) Reserve
31 * Reserve metadata/data space for incoming operations
32 * Affect how qgroup limit works
33 *
34 * 2) Trace
35 * Tell btrfs qgroup to trace dirty extents.
36 *
37 * Dirty extents including:
38 * - Newly allocated extents
39 * - Extents going to be deleted (in this trans)
40 * - Extents whose owner is going to be modified
41 *
42 * This is the main part affects whether qgroup numbers will stay
43 * consistent.
44 * Btrfs qgroup can trace clean extents and won't cause any problem,
45 * but it will consume extra CPU time, it should be avoided if possible.
46 *
47 * 3) Account
48 * Btrfs qgroup will updates its numbers, based on dirty extents traced
49 * in previous step.
50 *
51 * Normally at qgroup rescan and transaction commit time.
52 */
53
54 /*
55 * Special performance optimization for balance.
56 *
57 * For balance, we need to swap subtree of subvolume and reloc trees.
58 * In theory, we need to trace all subtree blocks of both subvolume and reloc
59 * trees, since their owner has changed during such swap.
60 *
61 * However since balance has ensured that both subtrees are containing the
62 * same contents and have the same tree structures, such swap won't cause
63 * qgroup number change.
64 *
65 * But there is a race window between subtree swap and transaction commit,
66 * during that window, if we increase/decrease tree level or merge/split tree
67 * blocks, we still need to trace the original subtrees.
68 *
69 * So for balance, we use a delayed subtree tracing, whose workflow is:
70 *
71 * 1) Record the subtree root block get swapped.
72 *
73 * During subtree swap:
74 * O = Old tree blocks
75 * N = New tree blocks
76 * reloc tree subvolume tree X
77 * Root Root
78 * / \ / \
79 * NA OB OA OB
80 * / | | \ / | | \
81 * NC ND OE OF OC OD OE OF
82 *
83 * In this case, NA and OA are going to be swapped, record (NA, OA) into
84 * subvolume tree X.
85 *
86 * 2) After subtree swap.
87 * reloc tree subvolume tree X
88 * Root Root
89 * / \ / \
90 * OA OB NA OB
91 * / | | \ / | | \
92 * OC OD OE OF NC ND OE OF
93 *
94 * 3a) COW happens for OB
95 * If we are going to COW tree block OB, we check OB's bytenr against
96 * tree X's swapped_blocks structure.
97 * If it doesn't fit any, nothing will happen.
98 *
99 * 3b) COW happens for NA
100 * Check NA's bytenr against tree X's swapped_blocks, and get a hit.
101 * Then we do subtree scan on both subtrees OA and NA.
102 * Resulting 6 tree blocks to be scanned (OA, OC, OD, NA, NC, ND).
103 *
104 * Then no matter what we do to subvolume tree X, qgroup numbers will
105 * still be correct.
106 * Then NA's record gets removed from X's swapped_blocks.
107 *
108 * 4) Transaction commit
109 * Any record in X's swapped_blocks gets removed, since there is no
110 * modification to the swapped subtrees, no need to trigger heavy qgroup
111 * subtree rescan for them.
112 */
113
114 /*
115 * These flags share the flags field of the btrfs_qgroup_status_item with the
116 * persisted flags defined in btrfs_tree.h.
117 *
118 * To minimize the chance of collision with new persisted status flags, these
119 * count backwards from the MSB.
120 */
121 #define BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN (1ULL << 63)
122 #define BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING (1ULL << 62)
123
124 #define BTRFS_QGROUP_DROP_SUBTREE_THRES_DEFAULT (3)
125
126 /*
127 * Record a dirty extent, and info qgroup to update quota on it
128 */
129 struct btrfs_qgroup_extent_record {
130 u64 bytenr;
131 u64 num_bytes;
132
133 /*
134 * For qgroup reserved data space freeing.
135 *
136 * @data_rsv_refroot and @data_rsv will be recorded after
137 * BTRFS_ADD_DELAYED_EXTENT is called.
138 * And will be used to free reserved qgroup space at
139 * transaction commit time.
140 */
141 u32 data_rsv; /* reserved data space needs to be freed */
142 u64 data_rsv_refroot; /* which root the reserved data belongs to */
143 struct ulist *old_roots;
144 };
145
146 struct btrfs_qgroup_swapped_block {
147 struct rb_node node;
148
149 int level;
150 bool trace_leaf;
151
152 /* bytenr/generation of the tree block in subvolume tree after swap */
153 u64 subvol_bytenr;
154 u64 subvol_generation;
155
156 /* bytenr/generation of the tree block in reloc tree after swap */
157 u64 reloc_bytenr;
158 u64 reloc_generation;
159
160 u64 last_snapshot;
161 struct btrfs_key first_key;
162 };
163
164 /*
165 * Qgroup reservation types:
166 *
167 * DATA:
168 * space reserved for data
169 *
170 * META_PERTRANS:
171 * Space reserved for metadata (per-transaction)
172 * Due to the fact that qgroup data is only updated at transaction commit
173 * time, reserved space for metadata must be kept until transaction
174 * commits.
175 * Any metadata reserved that are used in btrfs_start_transaction() should
176 * be of this type.
177 *
178 * META_PREALLOC:
179 * There are cases where metadata space is reserved before starting
180 * transaction, and then btrfs_join_transaction() to get a trans handle.
181 * Any metadata reserved for such usage should be of this type.
182 * And after join_transaction() part (or all) of such reservation should
183 * be converted into META_PERTRANS.
184 */
185 enum btrfs_qgroup_rsv_type {
186 BTRFS_QGROUP_RSV_DATA,
187 BTRFS_QGROUP_RSV_META_PERTRANS,
188 BTRFS_QGROUP_RSV_META_PREALLOC,
189 BTRFS_QGROUP_RSV_LAST,
190 };
191
192 /*
193 * Represents how many bytes we have reserved for this qgroup.
194 *
195 * Each type should have different reservation behavior.
196 * E.g, data follows its io_tree flag modification, while
197 * *currently* meta is just reserve-and-clear during transaction.
198 *
199 * TODO: Add new type for reservation which can survive transaction commit.
200 * Current metadata reservation behavior is not suitable for such case.
201 */
202 struct btrfs_qgroup_rsv {
203 u64 values[BTRFS_QGROUP_RSV_LAST];
204 };
205
206 /*
207 * one struct for each qgroup, organized in fs_info->qgroup_tree.
208 */
209 struct btrfs_qgroup {
210 u64 qgroupid;
211
212 /*
213 * state
214 */
215 u64 rfer; /* referenced */
216 u64 rfer_cmpr; /* referenced compressed */
217 u64 excl; /* exclusive */
218 u64 excl_cmpr; /* exclusive compressed */
219
220 /*
221 * limits
222 */
223 u64 lim_flags; /* which limits are set */
224 u64 max_rfer;
225 u64 max_excl;
226 u64 rsv_rfer;
227 u64 rsv_excl;
228
229 /*
230 * reservation tracking
231 */
232 struct btrfs_qgroup_rsv rsv;
233
234 /*
235 * lists
236 */
237 struct list_head groups; /* groups this group is member of */
238 struct list_head members; /* groups that are members of this group */
239 struct list_head dirty; /* dirty groups */
240
241 /*
242 * For qgroup iteration usage.
243 *
244 * The iteration list should always be empty until qgroup_iterator_add()
245 * is called. And should be reset to empty after the iteration is
246 * finished.
247 */
248 struct list_head iterator;
249
250 /*
251 * For nested iterator usage.
252 *
253 * Here we support at most one level of nested iterator calls like:
254 *
255 * LIST_HEAD(all_qgroups);
256 * {
257 * LIST_HEAD(local_qgroups);
258 * qgroup_iterator_add(local_qgroups, qg);
259 * qgroup_iterator_nested_add(all_qgroups, qg);
260 * do_some_work(local_qgroups);
261 * qgroup_iterator_clean(local_qgroups);
262 * }
263 * do_some_work(all_qgroups);
264 * qgroup_iterator_nested_clean(all_qgroups);
265 */
266 struct list_head nested_iterator;
267 struct rb_node node; /* tree of qgroups */
268
269 /*
270 * temp variables for accounting operations
271 * Refer to qgroup_shared_accounting() for details.
272 */
273 u64 old_refcnt;
274 u64 new_refcnt;
275
276 /*
277 * Sysfs kobjectid
278 */
279 struct kobject kobj;
280 };
281
282 /* Glue structure to represent the relations between qgroups. */
283 struct btrfs_qgroup_list {
284 struct list_head next_group;
285 struct list_head next_member;
286 struct btrfs_qgroup *group;
287 struct btrfs_qgroup *member;
288 };
289
290 struct btrfs_squota_delta {
291 /* The fstree root this delta counts against. */
292 u64 root;
293 /* The number of bytes in the extent being counted. */
294 u64 num_bytes;
295 /* The generation the extent was created in. */
296 u64 generation;
297 /* Whether we are using or freeing the extent. */
298 bool is_inc;
299 /* Whether the extent is data or metadata. */
300 bool is_data;
301 };
302
btrfs_qgroup_subvolid(u64 qgroupid)303 static inline u64 btrfs_qgroup_subvolid(u64 qgroupid)
304 {
305 return (qgroupid & ((1ULL << BTRFS_QGROUP_LEVEL_SHIFT) - 1));
306 }
307
308 /*
309 * For qgroup event trace points only
310 */
311 enum {
312 ENUM_BIT(QGROUP_RESERVE),
313 ENUM_BIT(QGROUP_RELEASE),
314 ENUM_BIT(QGROUP_FREE),
315 };
316
317 enum btrfs_qgroup_mode {
318 BTRFS_QGROUP_MODE_DISABLED,
319 BTRFS_QGROUP_MODE_FULL,
320 BTRFS_QGROUP_MODE_SIMPLE
321 };
322
323 enum btrfs_qgroup_mode btrfs_qgroup_mode(const struct btrfs_fs_info *fs_info);
324 bool btrfs_qgroup_enabled(const struct btrfs_fs_info *fs_info);
325 bool btrfs_qgroup_full_accounting(const struct btrfs_fs_info *fs_info);
326 int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
327 struct btrfs_ioctl_quota_ctl_args *quota_ctl_args);
328 int btrfs_quota_disable(struct btrfs_fs_info *fs_info);
329 int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
330 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
331 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
332 bool interruptible);
333 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src, u64 dst,
334 struct btrfs_qgroup_list *prealloc);
335 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
336 u64 dst);
337 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
338 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
339 int btrfs_qgroup_cleanup_dropped_subvolume(struct btrfs_fs_info *fs_info, u64 subvolid);
340 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
341 struct btrfs_qgroup_limit *limit);
342 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info);
343 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info);
344
345 int btrfs_qgroup_trace_extent_nolock(
346 struct btrfs_fs_info *fs_info,
347 struct btrfs_delayed_ref_root *delayed_refs,
348 struct btrfs_qgroup_extent_record *record);
349 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
350 struct btrfs_qgroup_extent_record *qrecord);
351 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
352 u64 num_bytes);
353 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
354 struct extent_buffer *eb);
355 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
356 struct extent_buffer *root_eb,
357 u64 root_gen, int root_level);
358 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
359 u64 num_bytes, struct ulist *old_roots,
360 struct ulist *new_roots);
361 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans);
362 int btrfs_run_qgroups(struct btrfs_trans_handle *trans);
363 int btrfs_qgroup_check_inherit(struct btrfs_fs_info *fs_info,
364 struct btrfs_qgroup_inherit *inherit,
365 size_t size);
366 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
367 u64 objectid, u64 inode_rootid,
368 struct btrfs_qgroup_inherit *inherit);
369 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
370 u64 ref_root, u64 num_bytes,
371 enum btrfs_qgroup_rsv_type type);
372
373 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
374 int btrfs_verify_qgroup_counts(const struct btrfs_fs_info *fs_info, u64 qgroupid,
375 u64 rfer, u64 excl);
376 #endif
377
378 /* New io_tree based accurate qgroup reserve API */
379 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
380 struct extent_changeset **reserved, u64 start, u64 len);
381 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len, u64 *released);
382 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
383 struct extent_changeset *reserved, u64 start,
384 u64 len, u64 *freed);
385 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
386 enum btrfs_qgroup_rsv_type type, bool enforce);
387 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
388 enum btrfs_qgroup_rsv_type type, bool enforce,
389 bool noflush);
390 /* Reserve metadata space for pertrans and prealloc type */
btrfs_qgroup_reserve_meta_pertrans(struct btrfs_root * root,int num_bytes,bool enforce)391 static inline int btrfs_qgroup_reserve_meta_pertrans(struct btrfs_root *root,
392 int num_bytes, bool enforce)
393 {
394 return __btrfs_qgroup_reserve_meta(root, num_bytes,
395 BTRFS_QGROUP_RSV_META_PERTRANS,
396 enforce, false);
397 }
btrfs_qgroup_reserve_meta_prealloc(struct btrfs_root * root,int num_bytes,bool enforce,bool noflush)398 static inline int btrfs_qgroup_reserve_meta_prealloc(struct btrfs_root *root,
399 int num_bytes, bool enforce,
400 bool noflush)
401 {
402 return __btrfs_qgroup_reserve_meta(root, num_bytes,
403 BTRFS_QGROUP_RSV_META_PREALLOC,
404 enforce, noflush);
405 }
406
407 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
408 enum btrfs_qgroup_rsv_type type);
409
410 /* Free per-transaction meta reservation for error handling */
btrfs_qgroup_free_meta_pertrans(struct btrfs_root * root,int num_bytes)411 static inline void btrfs_qgroup_free_meta_pertrans(struct btrfs_root *root,
412 int num_bytes)
413 {
414 __btrfs_qgroup_free_meta(root, num_bytes,
415 BTRFS_QGROUP_RSV_META_PERTRANS);
416 }
417
418 /* Pre-allocated meta reservation can be freed at need */
btrfs_qgroup_free_meta_prealloc(struct btrfs_root * root,int num_bytes)419 static inline void btrfs_qgroup_free_meta_prealloc(struct btrfs_root *root,
420 int num_bytes)
421 {
422 __btrfs_qgroup_free_meta(root, num_bytes,
423 BTRFS_QGROUP_RSV_META_PREALLOC);
424 }
425
426 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root);
427 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes);
428 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode);
429
430 /* btrfs_qgroup_swapped_blocks related functions */
431 void btrfs_qgroup_init_swapped_blocks(
432 struct btrfs_qgroup_swapped_blocks *swapped_blocks);
433
434 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root);
435 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
436 struct btrfs_root *subvol_root,
437 struct btrfs_block_group *bg,
438 struct extent_buffer *subvol_parent, int subvol_slot,
439 struct extent_buffer *reloc_parent, int reloc_slot,
440 u64 last_snapshot);
441 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
442 struct btrfs_root *root, struct extent_buffer *eb);
443 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans);
444 bool btrfs_check_quota_leak(const struct btrfs_fs_info *fs_info);
445 void btrfs_free_squota_rsv(struct btrfs_fs_info *fs_info, u64 root, u64 rsv_bytes);
446 int btrfs_record_squota_delta(struct btrfs_fs_info *fs_info,
447 const struct btrfs_squota_delta *delta);
448
449 #endif
450