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/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
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 "ulist.h"
10#include "delayed-ref.h"
11
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
40/*
41 * Record a dirty extent, and info qgroup to update quota on it
42 * TODO: Use kmem cache to alloc it.
43 */
44struct btrfs_qgroup_extent_record {
45 struct rb_node node;
46 u64 bytenr;
47 u64 num_bytes;
48 struct ulist *old_roots;
49};
50
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 */
72enum btrfs_qgroup_rsv_type {
73 BTRFS_QGROUP_RSV_DATA = 0,
74 BTRFS_QGROUP_RSV_META_PERTRANS,
75 BTRFS_QGROUP_RSV_META_PREALLOC,
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 */
89struct btrfs_qgroup_rsv {
90 u64 values[BTRFS_QGROUP_RSV_LAST];
91};
92
93/*
94 * one struct for each qgroup, organized in fs_info->qgroup_tree.
95 */
96struct 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 */
119 struct btrfs_qgroup_rsv rsv;
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
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
144int btrfs_quota_enable(struct btrfs_trans_handle *trans,
145 struct btrfs_fs_info *fs_info);
146int btrfs_quota_disable(struct btrfs_trans_handle *trans,
147 struct btrfs_fs_info *fs_info);
148int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
149void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
150int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
151 bool interruptible);
152int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
153 struct btrfs_fs_info *fs_info, u64 src, u64 dst);
154int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
155 struct btrfs_fs_info *fs_info, u64 src, u64 dst);
156int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
157 struct btrfs_fs_info *fs_info, u64 qgroupid);
158int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
159 struct btrfs_fs_info *fs_info, u64 qgroupid);
160int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
161 struct btrfs_fs_info *fs_info, u64 qgroupid,
162 struct btrfs_qgroup_limit *limit);
163int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info);
164void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info);
165struct btrfs_delayed_extent_op;
166
167/*
168 * Inform qgroup to trace one dirty extent, its info is recorded in @record.
169 * So qgroup can account it at transaction committing time.
170 *
171 * No lock version, caller must acquire delayed ref lock and allocated memory,
172 * then call btrfs_qgroup_trace_extent_post() after exiting lock context.
173 *
174 * Return 0 for success insert
175 * Return >0 for existing record, caller can free @record safely.
176 * Error is not possible
177 */
178int btrfs_qgroup_trace_extent_nolock(
179 struct btrfs_fs_info *fs_info,
180 struct btrfs_delayed_ref_root *delayed_refs,
181 struct btrfs_qgroup_extent_record *record);
182
183/*
184 * Post handler after qgroup_trace_extent_nolock().
185 *
186 * NOTE: Current qgroup does the expensive backref walk at transaction
187 * committing time with TRANS_STATE_COMMIT_DOING, this blocks incoming
188 * new transaction.
189 * This is designed to allow btrfs_find_all_roots() to get correct new_roots
190 * result.
191 *
192 * However for old_roots there is no need to do backref walk at that time,
193 * since we search commit roots to walk backref and result will always be
194 * correct.
195 *
196 * Due to the nature of no lock version, we can't do backref there.
197 * So we must call btrfs_qgroup_trace_extent_post() after exiting
198 * spinlock context.
199 *
200 * TODO: If we can fix and prove btrfs_find_all_roots() can get correct result
201 * using current root, then we can move all expensive backref walk out of
202 * transaction committing, but not now as qgroup accounting will be wrong again.
203 */
204int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
205 struct btrfs_qgroup_extent_record *qrecord);
206
207/*
208 * Inform qgroup to trace one dirty extent, specified by @bytenr and
209 * @num_bytes.
210 * So qgroup can account it at commit trans time.
211 *
212 * Better encapsulated version, with memory allocation and backref walk for
213 * commit roots.
214 * So this can sleep.
215 *
216 * Return 0 if the operation is done.
217 * Return <0 for error, like memory allocation failure or invalid parameter
218 * (NULL trans)
219 */
220int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
221 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
222 gfp_t gfp_flag);
223
224/*
225 * Inform qgroup to trace all leaf items of data
226 *
227 * Return 0 for success
228 * Return <0 for error(ENOMEM)
229 */
230int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
231 struct btrfs_fs_info *fs_info,
232 struct extent_buffer *eb);
233/*
234 * Inform qgroup to trace a whole subtree, including all its child tree
235 * blocks and data.
236 * The root tree block is specified by @root_eb.
237 *
238 * Normally used by relocation(tree block swap) and subvolume deletion.
239 *
240 * Return 0 for success
241 * Return <0 for error(ENOMEM or tree search error)
242 */
243int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
244 struct btrfs_root *root,
245 struct extent_buffer *root_eb,
246 u64 root_gen, int root_level);
247int
248btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
249 struct btrfs_fs_info *fs_info,
250 u64 bytenr, u64 num_bytes,
251 struct ulist *old_roots, struct ulist *new_roots);
252int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans);
253int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
254 struct btrfs_fs_info *fs_info);
255int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
256 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
257 struct btrfs_qgroup_inherit *inherit);
258void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
259 u64 ref_root, u64 num_bytes,
260 enum btrfs_qgroup_rsv_type type);
261static inline void btrfs_qgroup_free_delayed_ref(struct btrfs_fs_info *fs_info,
262 u64 ref_root, u64 num_bytes)
263{
264 trace_btrfs_qgroup_free_delayed_ref(fs_info, ref_root, num_bytes);
265 btrfs_qgroup_free_refroot(fs_info, ref_root, num_bytes,
266 BTRFS_QGROUP_RSV_DATA);
267}
268
269#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
270int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
271 u64 rfer, u64 excl);
272#endif
273
274/* New io_tree based accurate qgroup reserve API */
275int btrfs_qgroup_reserve_data(struct inode *inode,
276 struct extent_changeset **reserved, u64 start, u64 len);
277int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len);
278int btrfs_qgroup_free_data(struct inode *inode,
279 struct extent_changeset *reserved, u64 start, u64 len);
280
281int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
282 enum btrfs_qgroup_rsv_type type, bool enforce);
283/* Reserve metadata space for pertrans and prealloc type */
284static inline int btrfs_qgroup_reserve_meta_pertrans(struct btrfs_root *root,
285 int num_bytes, bool enforce)
286{
287 return __btrfs_qgroup_reserve_meta(root, num_bytes,
288 BTRFS_QGROUP_RSV_META_PERTRANS, enforce);
289}
290static inline int btrfs_qgroup_reserve_meta_prealloc(struct btrfs_root *root,
291 int num_bytes, bool enforce)
292{
293 return __btrfs_qgroup_reserve_meta(root, num_bytes,
294 BTRFS_QGROUP_RSV_META_PREALLOC, enforce);
295}
296
297void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
298 enum btrfs_qgroup_rsv_type type);
299
300/* Free per-transaction meta reservation for error handling */
301static inline void btrfs_qgroup_free_meta_pertrans(struct btrfs_root *root,
302 int num_bytes)
303{
304 __btrfs_qgroup_free_meta(root, num_bytes,
305 BTRFS_QGROUP_RSV_META_PERTRANS);
306}
307
308/* Pre-allocated meta reservation can be freed at need */
309static inline void btrfs_qgroup_free_meta_prealloc(struct btrfs_root *root,
310 int num_bytes)
311{
312 __btrfs_qgroup_free_meta(root, num_bytes,
313 BTRFS_QGROUP_RSV_META_PREALLOC);
314}
315
316/*
317 * Per-transaction meta reservation should be all freed at transaction commit
318 * time
319 */
320void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root);
321
322/*
323 * Convert @num_bytes of META_PREALLOCATED reservation to META_PERTRANS.
324 *
325 * This is called when preallocated meta reservation needs to be used.
326 * Normally after btrfs_join_transaction() call.
327 */
328void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes);
329
330void btrfs_qgroup_check_reserved_leak(struct inode *inode);
331
332#endif