Loading...
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/spinlock.h>
10#include <linux/rbtree.h>
11#include <linux/kobject.h>
12#include "ulist.h"
13#include "delayed-ref.h"
14#include "misc.h"
15
16/*
17 * Btrfs qgroup overview
18 *
19 * Btrfs qgroup splits into 3 main part:
20 * 1) Reserve
21 * Reserve metadata/data space for incoming operations
22 * Affect how qgroup limit works
23 *
24 * 2) Trace
25 * Tell btrfs qgroup to trace dirty extents.
26 *
27 * Dirty extents including:
28 * - Newly allocated extents
29 * - Extents going to be deleted (in this trans)
30 * - Extents whose owner is going to be modified
31 *
32 * This is the main part affects whether qgroup numbers will stay
33 * consistent.
34 * Btrfs qgroup can trace clean extents and won't cause any problem,
35 * but it will consume extra CPU time, it should be avoided if possible.
36 *
37 * 3) Account
38 * Btrfs qgroup will updates its numbers, based on dirty extents traced
39 * in previous step.
40 *
41 * Normally at qgroup rescan and transaction commit time.
42 */
43
44/*
45 * Special performance optimization for balance.
46 *
47 * For balance, we need to swap subtree of subvolume and reloc trees.
48 * In theory, we need to trace all subtree blocks of both subvolume and reloc
49 * trees, since their owner has changed during such swap.
50 *
51 * However since balance has ensured that both subtrees are containing the
52 * same contents and have the same tree structures, such swap won't cause
53 * qgroup number change.
54 *
55 * But there is a race window between subtree swap and transaction commit,
56 * during that window, if we increase/decrease tree level or merge/split tree
57 * blocks, we still need to trace the original subtrees.
58 *
59 * So for balance, we use a delayed subtree tracing, whose workflow is:
60 *
61 * 1) Record the subtree root block get swapped.
62 *
63 * During subtree swap:
64 * O = Old tree blocks
65 * N = New tree blocks
66 * reloc tree subvolume tree X
67 * Root Root
68 * / \ / \
69 * NA OB OA OB
70 * / | | \ / | | \
71 * NC ND OE OF OC OD OE OF
72 *
73 * In this case, NA and OA are going to be swapped, record (NA, OA) into
74 * subvolume tree X.
75 *
76 * 2) After subtree swap.
77 * reloc tree subvolume tree X
78 * Root Root
79 * / \ / \
80 * OA OB NA OB
81 * / | | \ / | | \
82 * OC OD OE OF NC ND OE OF
83 *
84 * 3a) COW happens for OB
85 * If we are going to COW tree block OB, we check OB's bytenr against
86 * tree X's swapped_blocks structure.
87 * If it doesn't fit any, nothing will happen.
88 *
89 * 3b) COW happens for NA
90 * Check NA's bytenr against tree X's swapped_blocks, and get a hit.
91 * Then we do subtree scan on both subtrees OA and NA.
92 * Resulting 6 tree blocks to be scanned (OA, OC, OD, NA, NC, ND).
93 *
94 * Then no matter what we do to subvolume tree X, qgroup numbers will
95 * still be correct.
96 * Then NA's record gets removed from X's swapped_blocks.
97 *
98 * 4) Transaction commit
99 * Any record in X's swapped_blocks gets removed, since there is no
100 * modification to the swapped subtrees, no need to trigger heavy qgroup
101 * subtree rescan for them.
102 */
103
104#define BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN (1UL << 3)
105#define BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING (1UL << 4)
106
107/*
108 * Record a dirty extent, and info qgroup to update quota on it
109 * TODO: Use kmem cache to alloc it.
110 */
111struct btrfs_qgroup_extent_record {
112 struct rb_node node;
113 u64 bytenr;
114 u64 num_bytes;
115
116 /*
117 * For qgroup reserved data space freeing.
118 *
119 * @data_rsv_refroot and @data_rsv will be recorded after
120 * BTRFS_ADD_DELAYED_EXTENT is called.
121 * And will be used to free reserved qgroup space at
122 * transaction commit time.
123 */
124 u32 data_rsv; /* reserved data space needs to be freed */
125 u64 data_rsv_refroot; /* which root the reserved data belongs to */
126 struct ulist *old_roots;
127};
128
129struct btrfs_qgroup_swapped_block {
130 struct rb_node node;
131
132 int level;
133 bool trace_leaf;
134
135 /* bytenr/generation of the tree block in subvolume tree after swap */
136 u64 subvol_bytenr;
137 u64 subvol_generation;
138
139 /* bytenr/generation of the tree block in reloc tree after swap */
140 u64 reloc_bytenr;
141 u64 reloc_generation;
142
143 u64 last_snapshot;
144 struct btrfs_key first_key;
145};
146
147/*
148 * Qgroup reservation types:
149 *
150 * DATA:
151 * space reserved for data
152 *
153 * META_PERTRANS:
154 * Space reserved for metadata (per-transaction)
155 * Due to the fact that qgroup data is only updated at transaction commit
156 * time, reserved space for metadata must be kept until transaction
157 * commits.
158 * Any metadata reserved that are used in btrfs_start_transaction() should
159 * be of this type.
160 *
161 * META_PREALLOC:
162 * There are cases where metadata space is reserved before starting
163 * transaction, and then btrfs_join_transaction() to get a trans handle.
164 * Any metadata reserved for such usage should be of this type.
165 * And after join_transaction() part (or all) of such reservation should
166 * be converted into META_PERTRANS.
167 */
168enum btrfs_qgroup_rsv_type {
169 BTRFS_QGROUP_RSV_DATA,
170 BTRFS_QGROUP_RSV_META_PERTRANS,
171 BTRFS_QGROUP_RSV_META_PREALLOC,
172 BTRFS_QGROUP_RSV_LAST,
173};
174
175/*
176 * Represents how many bytes we have reserved for this qgroup.
177 *
178 * Each type should have different reservation behavior.
179 * E.g, data follows its io_tree flag modification, while
180 * *currently* meta is just reserve-and-clear during transaction.
181 *
182 * TODO: Add new type for reservation which can survive transaction commit.
183 * Current metadata reservation behavior is not suitable for such case.
184 */
185struct btrfs_qgroup_rsv {
186 u64 values[BTRFS_QGROUP_RSV_LAST];
187};
188
189/*
190 * one struct for each qgroup, organized in fs_info->qgroup_tree.
191 */
192struct btrfs_qgroup {
193 u64 qgroupid;
194
195 /*
196 * state
197 */
198 u64 rfer; /* referenced */
199 u64 rfer_cmpr; /* referenced compressed */
200 u64 excl; /* exclusive */
201 u64 excl_cmpr; /* exclusive compressed */
202
203 /*
204 * limits
205 */
206 u64 lim_flags; /* which limits are set */
207 u64 max_rfer;
208 u64 max_excl;
209 u64 rsv_rfer;
210 u64 rsv_excl;
211
212 /*
213 * reservation tracking
214 */
215 struct btrfs_qgroup_rsv rsv;
216
217 /*
218 * lists
219 */
220 struct list_head groups; /* groups this group is member of */
221 struct list_head members; /* groups that are members of this group */
222 struct list_head dirty; /* dirty groups */
223 struct rb_node node; /* tree of qgroups */
224
225 /*
226 * temp variables for accounting operations
227 * Refer to qgroup_shared_accounting() for details.
228 */
229 u64 old_refcnt;
230 u64 new_refcnt;
231
232 /*
233 * Sysfs kobjectid
234 */
235 struct kobject kobj;
236};
237
238static inline u64 btrfs_qgroup_subvolid(u64 qgroupid)
239{
240 return (qgroupid & ((1ULL << BTRFS_QGROUP_LEVEL_SHIFT) - 1));
241}
242
243/*
244 * For qgroup event trace points only
245 */
246enum {
247 ENUM_BIT(QGROUP_RESERVE),
248 ENUM_BIT(QGROUP_RELEASE),
249 ENUM_BIT(QGROUP_FREE),
250};
251
252int btrfs_quota_enable(struct btrfs_fs_info *fs_info);
253int btrfs_quota_disable(struct btrfs_fs_info *fs_info);
254int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
255void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
256int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
257 bool interruptible);
258int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
259 u64 dst);
260int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
261 u64 dst);
262int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
263int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
264int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
265 struct btrfs_qgroup_limit *limit);
266int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info);
267void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info);
268struct btrfs_delayed_extent_op;
269
270/*
271 * Inform qgroup to trace one dirty extent, its info is recorded in @record.
272 * So qgroup can account it at transaction committing time.
273 *
274 * No lock version, caller must acquire delayed ref lock and allocated memory,
275 * then call btrfs_qgroup_trace_extent_post() after exiting lock context.
276 *
277 * Return 0 for success insert
278 * Return >0 for existing record, caller can free @record safely.
279 * Error is not possible
280 */
281int btrfs_qgroup_trace_extent_nolock(
282 struct btrfs_fs_info *fs_info,
283 struct btrfs_delayed_ref_root *delayed_refs,
284 struct btrfs_qgroup_extent_record *record);
285
286/*
287 * Post handler after qgroup_trace_extent_nolock().
288 *
289 * NOTE: Current qgroup does the expensive backref walk at transaction
290 * committing time with TRANS_STATE_COMMIT_DOING, this blocks incoming
291 * new transaction.
292 * This is designed to allow btrfs_find_all_roots() to get correct new_roots
293 * result.
294 *
295 * However for old_roots there is no need to do backref walk at that time,
296 * since we search commit roots to walk backref and result will always be
297 * correct.
298 *
299 * Due to the nature of no lock version, we can't do backref there.
300 * So we must call btrfs_qgroup_trace_extent_post() after exiting
301 * spinlock context.
302 *
303 * TODO: If we can fix and prove btrfs_find_all_roots() can get correct result
304 * using current root, then we can move all expensive backref walk out of
305 * transaction committing, but not now as qgroup accounting will be wrong again.
306 */
307int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
308 struct btrfs_qgroup_extent_record *qrecord);
309
310/*
311 * Inform qgroup to trace one dirty extent, specified by @bytenr and
312 * @num_bytes.
313 * So qgroup can account it at commit trans time.
314 *
315 * Better encapsulated version, with memory allocation and backref walk for
316 * commit roots.
317 * So this can sleep.
318 *
319 * Return 0 if the operation is done.
320 * Return <0 for error, like memory allocation failure or invalid parameter
321 * (NULL trans)
322 */
323int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
324 u64 num_bytes);
325
326/*
327 * Inform qgroup to trace all leaf items of data
328 *
329 * Return 0 for success
330 * Return <0 for error(ENOMEM)
331 */
332int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
333 struct extent_buffer *eb);
334/*
335 * Inform qgroup to trace a whole subtree, including all its child tree
336 * blocks and data.
337 * The root tree block is specified by @root_eb.
338 *
339 * Normally used by relocation(tree block swap) and subvolume deletion.
340 *
341 * Return 0 for success
342 * Return <0 for error(ENOMEM or tree search error)
343 */
344int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
345 struct extent_buffer *root_eb,
346 u64 root_gen, int root_level);
347int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
348 u64 num_bytes, struct ulist *old_roots,
349 struct ulist *new_roots);
350int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans);
351int btrfs_run_qgroups(struct btrfs_trans_handle *trans);
352int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
353 u64 objectid, struct btrfs_qgroup_inherit *inherit);
354void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
355 u64 ref_root, u64 num_bytes,
356 enum btrfs_qgroup_rsv_type type);
357
358#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
359int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
360 u64 rfer, u64 excl);
361#endif
362
363/* New io_tree based accurate qgroup reserve API */
364int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
365 struct extent_changeset **reserved, u64 start, u64 len);
366int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len);
367int btrfs_qgroup_free_data(struct btrfs_inode *inode,
368 struct extent_changeset *reserved, u64 start,
369 u64 len);
370int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
371 enum btrfs_qgroup_rsv_type type, bool enforce);
372int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
373 enum btrfs_qgroup_rsv_type type, bool enforce,
374 bool noflush);
375/* Reserve metadata space for pertrans and prealloc type */
376static inline int btrfs_qgroup_reserve_meta_pertrans(struct btrfs_root *root,
377 int num_bytes, bool enforce)
378{
379 return __btrfs_qgroup_reserve_meta(root, num_bytes,
380 BTRFS_QGROUP_RSV_META_PERTRANS,
381 enforce, false);
382}
383static inline int btrfs_qgroup_reserve_meta_prealloc(struct btrfs_root *root,
384 int num_bytes, bool enforce,
385 bool noflush)
386{
387 return __btrfs_qgroup_reserve_meta(root, num_bytes,
388 BTRFS_QGROUP_RSV_META_PREALLOC,
389 enforce, noflush);
390}
391
392void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
393 enum btrfs_qgroup_rsv_type type);
394
395/* Free per-transaction meta reservation for error handling */
396static inline void btrfs_qgroup_free_meta_pertrans(struct btrfs_root *root,
397 int num_bytes)
398{
399 __btrfs_qgroup_free_meta(root, num_bytes,
400 BTRFS_QGROUP_RSV_META_PERTRANS);
401}
402
403/* Pre-allocated meta reservation can be freed at need */
404static inline void btrfs_qgroup_free_meta_prealloc(struct btrfs_root *root,
405 int num_bytes)
406{
407 __btrfs_qgroup_free_meta(root, num_bytes,
408 BTRFS_QGROUP_RSV_META_PREALLOC);
409}
410
411/*
412 * Per-transaction meta reservation should be all freed at transaction commit
413 * time
414 */
415void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root);
416
417/*
418 * Convert @num_bytes of META_PREALLOCATED reservation to META_PERTRANS.
419 *
420 * This is called when preallocated meta reservation needs to be used.
421 * Normally after btrfs_join_transaction() call.
422 */
423void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes);
424
425void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode);
426
427/* btrfs_qgroup_swapped_blocks related functions */
428void btrfs_qgroup_init_swapped_blocks(
429 struct btrfs_qgroup_swapped_blocks *swapped_blocks);
430
431void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root);
432int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
433 struct btrfs_root *subvol_root,
434 struct btrfs_block_group *bg,
435 struct extent_buffer *subvol_parent, int subvol_slot,
436 struct extent_buffer *reloc_parent, int reloc_slot,
437 u64 last_snapshot);
438int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
439 struct btrfs_root *root, struct extent_buffer *eb);
440void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans);
441bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info);
442
443#endif
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
16struct extent_buffer;
17struct extent_changeset;
18struct btrfs_delayed_extent_op;
19struct btrfs_fs_info;
20struct btrfs_root;
21struct btrfs_ioctl_quota_ctl_args;
22struct btrfs_trans_handle;
23struct btrfs_delayed_ref_root;
24struct 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 */
129struct btrfs_qgroup_extent_record {
130 /*
131 * The bytenr of the extent is given by its index in the dirty_extents
132 * xarray of struct btrfs_delayed_ref_root left shifted by
133 * fs_info->sectorsize_bits.
134 */
135
136 u64 num_bytes;
137
138 /*
139 * For qgroup reserved data space freeing.
140 *
141 * @data_rsv_refroot and @data_rsv will be recorded after
142 * BTRFS_ADD_DELAYED_EXTENT is called.
143 * And will be used to free reserved qgroup space at
144 * transaction commit time.
145 */
146 u32 data_rsv; /* reserved data space needs to be freed */
147 u64 data_rsv_refroot; /* which root the reserved data belongs to */
148 struct ulist *old_roots;
149};
150
151struct btrfs_qgroup_swapped_block {
152 struct rb_node node;
153
154 int level;
155 bool trace_leaf;
156
157 /* bytenr/generation of the tree block in subvolume tree after swap */
158 u64 subvol_bytenr;
159 u64 subvol_generation;
160
161 /* bytenr/generation of the tree block in reloc tree after swap */
162 u64 reloc_bytenr;
163 u64 reloc_generation;
164
165 u64 last_snapshot;
166 struct btrfs_key first_key;
167};
168
169/*
170 * Qgroup reservation types:
171 *
172 * DATA:
173 * space reserved for data
174 *
175 * META_PERTRANS:
176 * Space reserved for metadata (per-transaction)
177 * Due to the fact that qgroup data is only updated at transaction commit
178 * time, reserved space for metadata must be kept until transaction
179 * commits.
180 * Any metadata reserved that are used in btrfs_start_transaction() should
181 * be of this type.
182 *
183 * META_PREALLOC:
184 * There are cases where metadata space is reserved before starting
185 * transaction, and then btrfs_join_transaction() to get a trans handle.
186 * Any metadata reserved for such usage should be of this type.
187 * And after join_transaction() part (or all) of such reservation should
188 * be converted into META_PERTRANS.
189 */
190enum btrfs_qgroup_rsv_type {
191 BTRFS_QGROUP_RSV_DATA,
192 BTRFS_QGROUP_RSV_META_PERTRANS,
193 BTRFS_QGROUP_RSV_META_PREALLOC,
194 BTRFS_QGROUP_RSV_LAST,
195};
196
197/*
198 * Represents how many bytes we have reserved for this qgroup.
199 *
200 * Each type should have different reservation behavior.
201 * E.g, data follows its io_tree flag modification, while
202 * *currently* meta is just reserve-and-clear during transaction.
203 *
204 * TODO: Add new type for reservation which can survive transaction commit.
205 * Current metadata reservation behavior is not suitable for such case.
206 */
207struct btrfs_qgroup_rsv {
208 u64 values[BTRFS_QGROUP_RSV_LAST];
209};
210
211/*
212 * one struct for each qgroup, organized in fs_info->qgroup_tree.
213 */
214struct btrfs_qgroup {
215 u64 qgroupid;
216
217 /*
218 * state
219 */
220 u64 rfer; /* referenced */
221 u64 rfer_cmpr; /* referenced compressed */
222 u64 excl; /* exclusive */
223 u64 excl_cmpr; /* exclusive compressed */
224
225 /*
226 * limits
227 */
228 u64 lim_flags; /* which limits are set */
229 u64 max_rfer;
230 u64 max_excl;
231 u64 rsv_rfer;
232 u64 rsv_excl;
233
234 /*
235 * reservation tracking
236 */
237 struct btrfs_qgroup_rsv rsv;
238
239 /*
240 * lists
241 */
242 struct list_head groups; /* groups this group is member of */
243 struct list_head members; /* groups that are members of this group */
244 struct list_head dirty; /* dirty groups */
245
246 /*
247 * For qgroup iteration usage.
248 *
249 * The iteration list should always be empty until qgroup_iterator_add()
250 * is called. And should be reset to empty after the iteration is
251 * finished.
252 */
253 struct list_head iterator;
254
255 /*
256 * For nested iterator usage.
257 *
258 * Here we support at most one level of nested iterator calls like:
259 *
260 * LIST_HEAD(all_qgroups);
261 * {
262 * LIST_HEAD(local_qgroups);
263 * qgroup_iterator_add(local_qgroups, qg);
264 * qgroup_iterator_nested_add(all_qgroups, qg);
265 * do_some_work(local_qgroups);
266 * qgroup_iterator_clean(local_qgroups);
267 * }
268 * do_some_work(all_qgroups);
269 * qgroup_iterator_nested_clean(all_qgroups);
270 */
271 struct list_head nested_iterator;
272 struct rb_node node; /* tree of qgroups */
273
274 /*
275 * temp variables for accounting operations
276 * Refer to qgroup_shared_accounting() for details.
277 */
278 u64 old_refcnt;
279 u64 new_refcnt;
280
281 /*
282 * Sysfs kobjectid
283 */
284 struct kobject kobj;
285};
286
287/* Glue structure to represent the relations between qgroups. */
288struct btrfs_qgroup_list {
289 struct list_head next_group;
290 struct list_head next_member;
291 struct btrfs_qgroup *group;
292 struct btrfs_qgroup *member;
293};
294
295struct btrfs_squota_delta {
296 /* The fstree root this delta counts against. */
297 u64 root;
298 /* The number of bytes in the extent being counted. */
299 u64 num_bytes;
300 /* The generation the extent was created in. */
301 u64 generation;
302 /* Whether we are using or freeing the extent. */
303 bool is_inc;
304 /* Whether the extent is data or metadata. */
305 bool is_data;
306};
307
308static inline u64 btrfs_qgroup_subvolid(u64 qgroupid)
309{
310 return (qgroupid & ((1ULL << BTRFS_QGROUP_LEVEL_SHIFT) - 1));
311}
312
313/*
314 * For qgroup event trace points only
315 */
316enum {
317 ENUM_BIT(QGROUP_RESERVE),
318 ENUM_BIT(QGROUP_RELEASE),
319 ENUM_BIT(QGROUP_FREE),
320};
321
322enum btrfs_qgroup_mode {
323 BTRFS_QGROUP_MODE_DISABLED,
324 BTRFS_QGROUP_MODE_FULL,
325 BTRFS_QGROUP_MODE_SIMPLE
326};
327
328enum btrfs_qgroup_mode btrfs_qgroup_mode(const struct btrfs_fs_info *fs_info);
329bool btrfs_qgroup_enabled(const struct btrfs_fs_info *fs_info);
330bool btrfs_qgroup_full_accounting(const struct btrfs_fs_info *fs_info);
331int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
332 struct btrfs_ioctl_quota_ctl_args *quota_ctl_args);
333int btrfs_quota_disable(struct btrfs_fs_info *fs_info);
334int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
335void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
336int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
337 bool interruptible);
338int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src, u64 dst,
339 struct btrfs_qgroup_list *prealloc);
340int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
341 u64 dst);
342int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
343int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
344int btrfs_qgroup_cleanup_dropped_subvolume(struct btrfs_fs_info *fs_info, u64 subvolid);
345int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
346 struct btrfs_qgroup_limit *limit);
347int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info);
348void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info);
349
350int btrfs_qgroup_trace_extent_nolock(
351 struct btrfs_fs_info *fs_info,
352 struct btrfs_delayed_ref_root *delayed_refs,
353 struct btrfs_qgroup_extent_record *record,
354 u64 bytenr);
355int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
356 struct btrfs_qgroup_extent_record *qrecord,
357 u64 bytenr);
358int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
359 u64 num_bytes);
360int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
361 struct extent_buffer *eb);
362int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
363 struct extent_buffer *root_eb,
364 u64 root_gen, int root_level);
365int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
366 u64 num_bytes, struct ulist *old_roots,
367 struct ulist *new_roots);
368int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans);
369int btrfs_run_qgroups(struct btrfs_trans_handle *trans);
370int btrfs_qgroup_check_inherit(struct btrfs_fs_info *fs_info,
371 struct btrfs_qgroup_inherit *inherit,
372 size_t size);
373int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
374 u64 objectid, u64 inode_rootid,
375 struct btrfs_qgroup_inherit *inherit);
376void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
377 u64 ref_root, u64 num_bytes,
378 enum btrfs_qgroup_rsv_type type);
379
380#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
381int btrfs_verify_qgroup_counts(const struct btrfs_fs_info *fs_info, u64 qgroupid,
382 u64 rfer, u64 excl);
383#endif
384
385/* New io_tree based accurate qgroup reserve API */
386int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
387 struct extent_changeset **reserved, u64 start, u64 len);
388int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len, u64 *released);
389int btrfs_qgroup_free_data(struct btrfs_inode *inode,
390 struct extent_changeset *reserved, u64 start,
391 u64 len, u64 *freed);
392int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
393 enum btrfs_qgroup_rsv_type type, bool enforce);
394int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
395 enum btrfs_qgroup_rsv_type type, bool enforce,
396 bool noflush);
397/* Reserve metadata space for pertrans and prealloc type */
398static inline int btrfs_qgroup_reserve_meta_pertrans(struct btrfs_root *root,
399 int num_bytes, bool enforce)
400{
401 return __btrfs_qgroup_reserve_meta(root, num_bytes,
402 BTRFS_QGROUP_RSV_META_PERTRANS,
403 enforce, false);
404}
405static inline int btrfs_qgroup_reserve_meta_prealloc(struct btrfs_root *root,
406 int num_bytes, bool enforce,
407 bool noflush)
408{
409 return __btrfs_qgroup_reserve_meta(root, num_bytes,
410 BTRFS_QGROUP_RSV_META_PREALLOC,
411 enforce, noflush);
412}
413
414void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
415 enum btrfs_qgroup_rsv_type type);
416
417/* Free per-transaction meta reservation for error handling */
418static inline void btrfs_qgroup_free_meta_pertrans(struct btrfs_root *root,
419 int num_bytes)
420{
421 __btrfs_qgroup_free_meta(root, num_bytes,
422 BTRFS_QGROUP_RSV_META_PERTRANS);
423}
424
425/* Pre-allocated meta reservation can be freed at need */
426static inline void btrfs_qgroup_free_meta_prealloc(struct btrfs_root *root,
427 int num_bytes)
428{
429 __btrfs_qgroup_free_meta(root, num_bytes,
430 BTRFS_QGROUP_RSV_META_PREALLOC);
431}
432
433void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root);
434void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes);
435void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode);
436
437/* btrfs_qgroup_swapped_blocks related functions */
438void btrfs_qgroup_init_swapped_blocks(
439 struct btrfs_qgroup_swapped_blocks *swapped_blocks);
440
441void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root);
442int btrfs_qgroup_add_swapped_blocks(struct btrfs_root *subvol_root,
443 struct btrfs_block_group *bg,
444 struct extent_buffer *subvol_parent, int subvol_slot,
445 struct extent_buffer *reloc_parent, int reloc_slot,
446 u64 last_snapshot);
447int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
448 struct btrfs_root *root, struct extent_buffer *eb);
449void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans);
450bool btrfs_check_quota_leak(const struct btrfs_fs_info *fs_info);
451int btrfs_record_squota_delta(struct btrfs_fs_info *fs_info,
452 const struct btrfs_squota_delta *delta);
453
454#endif