1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Copyright (C) 2014 Facebook. All rights reserved.
10 #include "delayed-ref.h"
13 * Btrfs qgroup overview
15 * Btrfs qgroup splits into 3 main part:
17 * Reserve metadata/data space for incoming operations
18 * Affect how qgroup limit works
21 * Tell btrfs qgroup to trace dirty extents.
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
28 * This is the main part affects whether qgroup numbers will stay
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.
34 * Btrfs qgroup will updates its numbers, based on dirty extents traced
37 * Normally at qgroup rescan and transaction commit time.
41 * Record a dirty extent, and info qgroup to update quota on it
42 * TODO: Use kmem cache to alloc it.
44 struct btrfs_qgroup_extent_record {
48 struct ulist *old_roots;
52 * Qgroup reservation types:
55 * space reserved for data
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
62 * Any metadata reserved that are used in btrfs_start_transaction() should
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.
72 enum 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,
80 * Represents how many bytes we have reserved for this qgroup.
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.
86 * TODO: Add new type for reservation which can survive transaction commit.
87 * Currect metadata reservation behavior is not suitable for such case.
89 struct btrfs_qgroup_rsv {
90 u64 values[BTRFS_QGROUP_RSV_LAST];
94 * one struct for each qgroup, organized in fs_info->qgroup_tree.
102 u64 rfer; /* referenced */
103 u64 rfer_cmpr; /* referenced compressed */
104 u64 excl; /* exclusive */
105 u64 excl_cmpr; /* exclusive compressed */
110 u64 lim_flags; /* which limits are set */
117 * reservation tracking
119 struct btrfs_qgroup_rsv rsv;
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 */
130 * temp variables for accounting operations
131 * Refer to qgroup_shared_accounting() for details.
138 * For qgroup event trace points only
140 #define QGROUP_RESERVE (1<<0)
141 #define QGROUP_RELEASE (1<<1)
142 #define QGROUP_FREE (1<<2)
144 int btrfs_quota_enable(struct btrfs_fs_info *fs_info);
145 int btrfs_quota_disable(struct btrfs_fs_info *fs_info);
146 int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
147 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
148 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
150 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
151 struct btrfs_fs_info *fs_info, u64 src, u64 dst);
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,
155 struct btrfs_fs_info *fs_info, u64 qgroupid);
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;
166 * Inform qgroup to trace one dirty extent, its info is recorded in @record.
167 * So qgroup can account it at transaction committing time.
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.
172 * Return 0 for success insert
173 * Return >0 for existing record, caller can free @record safely.
174 * Error is not possible
176 int btrfs_qgroup_trace_extent_nolock(
177 struct btrfs_fs_info *fs_info,
178 struct btrfs_delayed_ref_root *delayed_refs,
179 struct btrfs_qgroup_extent_record *record);
182 * Post handler after qgroup_trace_extent_nolock().
184 * NOTE: Current qgroup does the expensive backref walk at transaction
185 * committing time with TRANS_STATE_COMMIT_DOING, this blocks incoming
187 * This is designed to allow btrfs_find_all_roots() to get correct new_roots
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
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
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.
202 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
203 struct btrfs_qgroup_extent_record *qrecord);
206 * Inform qgroup to trace one dirty extent, specified by @bytenr and
208 * So qgroup can account it at commit trans time.
210 * Better encapsulated version, with memory allocation and backref walk for
214 * Return 0 if the operation is done.
215 * Return <0 for error, like memory allocation failure or invalid parameter
218 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
219 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
223 * Inform qgroup to trace all leaf items of data
225 * Return 0 for success
226 * Return <0 for error(ENOMEM)
228 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
229 struct btrfs_fs_info *fs_info,
230 struct extent_buffer *eb);
232 * Inform qgroup to trace a whole subtree, including all its child tree
234 * The root tree block is specified by @root_eb.
236 * Normally used by relocation(tree block swap) and subvolume deletion.
238 * Return 0 for success
239 * Return <0 for error(ENOMEM or tree search error)
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);
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);
250 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans);
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);
256 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
257 u64 ref_root, u64 num_bytes,
258 enum btrfs_qgroup_rsv_type type);
259 static inline void btrfs_qgroup_free_delayed_ref(struct btrfs_fs_info *fs_info,
260 u64 ref_root, u64 num_bytes)
262 trace_btrfs_qgroup_free_delayed_ref(fs_info, ref_root, num_bytes);
263 btrfs_qgroup_free_refroot(fs_info, ref_root, num_bytes,
264 BTRFS_QGROUP_RSV_DATA);
267 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
268 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
272 /* New io_tree based accurate qgroup reserve API */
273 int btrfs_qgroup_reserve_data(struct inode *inode,
274 struct extent_changeset **reserved, u64 start, u64 len);
275 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len);
276 int btrfs_qgroup_free_data(struct inode *inode,
277 struct extent_changeset *reserved, u64 start, u64 len);
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)
285 return __btrfs_qgroup_reserve_meta(root, num_bytes,
286 BTRFS_QGROUP_RSV_META_PERTRANS, enforce);
288 static inline int btrfs_qgroup_reserve_meta_prealloc(struct btrfs_root *root,
289 int num_bytes, bool enforce)
291 return __btrfs_qgroup_reserve_meta(root, num_bytes,
292 BTRFS_QGROUP_RSV_META_PREALLOC, enforce);
295 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
296 enum btrfs_qgroup_rsv_type type);
298 /* Free per-transaction meta reservation for error handling */
299 static inline void btrfs_qgroup_free_meta_pertrans(struct btrfs_root *root,
302 __btrfs_qgroup_free_meta(root, num_bytes,
303 BTRFS_QGROUP_RSV_META_PERTRANS);
306 /* Pre-allocated meta reservation can be freed at need */
307 static inline void btrfs_qgroup_free_meta_prealloc(struct btrfs_root *root,
310 __btrfs_qgroup_free_meta(root, num_bytes,
311 BTRFS_QGROUP_RSV_META_PREALLOC);
315 * Per-transaction meta reservation should be all freed at transaction commit
318 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root);
321 * Convert @num_bytes of META_PREALLOCATED reservation to META_PERTRANS.
323 * This is called when preallocated meta reservation needs to be used.
324 * Normally after btrfs_join_transaction() call.
326 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes);
328 void btrfs_qgroup_check_reserved_leak(struct inode *inode);