1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Copyright (C) 2014 Facebook. All rights reserved.
9 #include <linux/spinlock.h>
10 #include <linux/rbtree.h>
12 #include "delayed-ref.h"
15 * Btrfs qgroup overview
17 * Btrfs qgroup splits into 3 main part:
19 * Reserve metadata/data space for incoming operations
20 * Affect how qgroup limit works
23 * Tell btrfs qgroup to trace dirty extents.
25 * Dirty extents including:
26 * - Newly allocated extents
27 * - Extents going to be deleted (in this trans)
28 * - Extents whose owner is going to be modified
30 * This is the main part affects whether qgroup numbers will stay
32 * Btrfs qgroup can trace clean extents and won't cause any problem,
33 * but it will consume extra CPU time, it should be avoided if possible.
36 * Btrfs qgroup will updates its numbers, based on dirty extents traced
39 * Normally at qgroup rescan and transaction commit time.
43 * Special performance optimization for balance.
45 * For balance, we need to swap subtree of subvolume and reloc trees.
46 * In theory, we need to trace all subtree blocks of both subvolume and reloc
47 * trees, since their owner has changed during such swap.
49 * However since balance has ensured that both subtrees are containing the
50 * same contents and have the same tree structures, such swap won't cause
51 * qgroup number change.
53 * But there is a race window between subtree swap and transaction commit,
54 * during that window, if we increase/decrease tree level or merge/split tree
55 * blocks, we still need to trace the original subtrees.
57 * So for balance, we use a delayed subtree tracing, whose workflow is:
59 * 1) Record the subtree root block get swapped.
61 * During subtree swap:
64 * reloc tree subvolume tree X
69 * NC ND OE OF OC OD OE OF
71 * In this case, NA and OA are going to be swapped, record (NA, OA) into
74 * 2) After subtree swap.
75 * reloc tree subvolume tree X
80 * OC OD OE OF NC ND OE OF
82 * 3a) COW happens for OB
83 * If we are going to COW tree block OB, we check OB's bytenr against
84 * tree X's swapped_blocks structure.
85 * If it doesn't fit any, nothing will happen.
87 * 3b) COW happens for NA
88 * Check NA's bytenr against tree X's swapped_blocks, and get a hit.
89 * Then we do subtree scan on both subtrees OA and NA.
90 * Resulting 6 tree blocks to be scanned (OA, OC, OD, NA, NC, ND).
92 * Then no matter what we do to subvolume tree X, qgroup numbers will
94 * Then NA's record gets removed from X's swapped_blocks.
96 * 4) Transaction commit
97 * Any record in X's swapped_blocks gets removed, since there is no
98 * modification to the swapped subtrees, no need to trigger heavy qgroup
99 * subtree rescan for them.
103 * Record a dirty extent, and info qgroup to update quota on it
104 * TODO: Use kmem cache to alloc it.
106 struct btrfs_qgroup_extent_record {
110 struct ulist *old_roots;
113 struct btrfs_qgroup_swapped_block {
119 /* bytenr/generation of the tree block in subvolume tree after swap */
121 u64 subvol_generation;
123 /* bytenr/generation of the tree block in reloc tree after swap */
125 u64 reloc_generation;
128 struct btrfs_key first_key;
132 * Qgroup reservation types:
135 * space reserved for data
138 * Space reserved for metadata (per-transaction)
139 * Due to the fact that qgroup data is only updated at transaction commit
140 * time, reserved space for metadata must be kept until transaction
142 * Any metadata reserved that are used in btrfs_start_transaction() should
146 * There are cases where metadata space is reserved before starting
147 * transaction, and then btrfs_join_transaction() to get a trans handle.
148 * Any metadata reserved for such usage should be of this type.
149 * And after join_transaction() part (or all) of such reservation should
150 * be converted into META_PERTRANS.
152 enum btrfs_qgroup_rsv_type {
153 BTRFS_QGROUP_RSV_DATA,
154 BTRFS_QGROUP_RSV_META_PERTRANS,
155 BTRFS_QGROUP_RSV_META_PREALLOC,
156 BTRFS_QGROUP_RSV_LAST,
160 * Represents how many bytes we have reserved for this qgroup.
162 * Each type should have different reservation behavior.
163 * E.g, data follows its io_tree flag modification, while
164 * *currently* meta is just reserve-and-clear during transaction.
166 * TODO: Add new type for reservation which can survive transaction commit.
167 * Current metadata reservation behavior is not suitable for such case.
169 struct btrfs_qgroup_rsv {
170 u64 values[BTRFS_QGROUP_RSV_LAST];
174 * one struct for each qgroup, organized in fs_info->qgroup_tree.
176 struct btrfs_qgroup {
182 u64 rfer; /* referenced */
183 u64 rfer_cmpr; /* referenced compressed */
184 u64 excl; /* exclusive */
185 u64 excl_cmpr; /* exclusive compressed */
190 u64 lim_flags; /* which limits are set */
197 * reservation tracking
199 struct btrfs_qgroup_rsv rsv;
204 struct list_head groups; /* groups this group is member of */
205 struct list_head members; /* groups that are members of this group */
206 struct list_head dirty; /* dirty groups */
207 struct rb_node node; /* tree of qgroups */
210 * temp variables for accounting operations
211 * Refer to qgroup_shared_accounting() for details.
218 * For qgroup event trace points only
220 #define QGROUP_RESERVE (1<<0)
221 #define QGROUP_RELEASE (1<<1)
222 #define QGROUP_FREE (1<<2)
224 int btrfs_quota_enable(struct btrfs_fs_info *fs_info);
225 int btrfs_quota_disable(struct btrfs_fs_info *fs_info);
226 int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
227 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
228 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
230 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
232 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
234 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
235 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
236 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
237 struct btrfs_qgroup_limit *limit);
238 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info);
239 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info);
240 struct btrfs_delayed_extent_op;
243 * Inform qgroup to trace one dirty extent, its info is recorded in @record.
244 * So qgroup can account it at transaction committing time.
246 * No lock version, caller must acquire delayed ref lock and allocated memory,
247 * then call btrfs_qgroup_trace_extent_post() after exiting lock context.
249 * Return 0 for success insert
250 * Return >0 for existing record, caller can free @record safely.
251 * Error is not possible
253 int btrfs_qgroup_trace_extent_nolock(
254 struct btrfs_fs_info *fs_info,
255 struct btrfs_delayed_ref_root *delayed_refs,
256 struct btrfs_qgroup_extent_record *record);
259 * Post handler after qgroup_trace_extent_nolock().
261 * NOTE: Current qgroup does the expensive backref walk at transaction
262 * committing time with TRANS_STATE_COMMIT_DOING, this blocks incoming
264 * This is designed to allow btrfs_find_all_roots() to get correct new_roots
267 * However for old_roots there is no need to do backref walk at that time,
268 * since we search commit roots to walk backref and result will always be
271 * Due to the nature of no lock version, we can't do backref there.
272 * So we must call btrfs_qgroup_trace_extent_post() after exiting
275 * TODO: If we can fix and prove btrfs_find_all_roots() can get correct result
276 * using current root, then we can move all expensive backref walk out of
277 * transaction committing, but not now as qgroup accounting will be wrong again.
279 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
280 struct btrfs_qgroup_extent_record *qrecord);
283 * Inform qgroup to trace one dirty extent, specified by @bytenr and
285 * So qgroup can account it at commit trans time.
287 * Better encapsulated version, with memory allocation and backref walk for
291 * Return 0 if the operation is done.
292 * Return <0 for error, like memory allocation failure or invalid parameter
295 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
296 u64 num_bytes, gfp_t gfp_flag);
299 * Inform qgroup to trace all leaf items of data
301 * Return 0 for success
302 * Return <0 for error(ENOMEM)
304 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
305 struct extent_buffer *eb);
307 * Inform qgroup to trace a whole subtree, including all its child tree
309 * The root tree block is specified by @root_eb.
311 * Normally used by relocation(tree block swap) and subvolume deletion.
313 * Return 0 for success
314 * Return <0 for error(ENOMEM or tree search error)
316 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
317 struct extent_buffer *root_eb,
318 u64 root_gen, int root_level);
319 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
320 u64 num_bytes, struct ulist *old_roots,
321 struct ulist *new_roots);
322 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans);
323 int btrfs_run_qgroups(struct btrfs_trans_handle *trans);
324 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
325 u64 objectid, struct btrfs_qgroup_inherit *inherit);
326 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
327 u64 ref_root, u64 num_bytes,
328 enum btrfs_qgroup_rsv_type type);
329 static inline void btrfs_qgroup_free_delayed_ref(struct btrfs_fs_info *fs_info,
330 u64 ref_root, u64 num_bytes)
332 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
334 trace_btrfs_qgroup_free_delayed_ref(fs_info, ref_root, num_bytes);
335 btrfs_qgroup_free_refroot(fs_info, ref_root, num_bytes,
336 BTRFS_QGROUP_RSV_DATA);
339 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
340 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
344 /* New io_tree based accurate qgroup reserve API */
345 int btrfs_qgroup_reserve_data(struct inode *inode,
346 struct extent_changeset **reserved, u64 start, u64 len);
347 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len);
348 int btrfs_qgroup_free_data(struct inode *inode,
349 struct extent_changeset *reserved, u64 start, u64 len);
351 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
352 enum btrfs_qgroup_rsv_type type, bool enforce);
353 /* Reserve metadata space for pertrans and prealloc type */
354 static inline int btrfs_qgroup_reserve_meta_pertrans(struct btrfs_root *root,
355 int num_bytes, bool enforce)
357 return __btrfs_qgroup_reserve_meta(root, num_bytes,
358 BTRFS_QGROUP_RSV_META_PERTRANS, enforce);
360 static inline int btrfs_qgroup_reserve_meta_prealloc(struct btrfs_root *root,
361 int num_bytes, bool enforce)
363 return __btrfs_qgroup_reserve_meta(root, num_bytes,
364 BTRFS_QGROUP_RSV_META_PREALLOC, enforce);
367 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
368 enum btrfs_qgroup_rsv_type type);
370 /* Free per-transaction meta reservation for error handling */
371 static inline void btrfs_qgroup_free_meta_pertrans(struct btrfs_root *root,
374 __btrfs_qgroup_free_meta(root, num_bytes,
375 BTRFS_QGROUP_RSV_META_PERTRANS);
378 /* Pre-allocated meta reservation can be freed at need */
379 static inline void btrfs_qgroup_free_meta_prealloc(struct btrfs_root *root,
382 __btrfs_qgroup_free_meta(root, num_bytes,
383 BTRFS_QGROUP_RSV_META_PREALLOC);
387 * Per-transaction meta reservation should be all freed at transaction commit
390 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root);
393 * Convert @num_bytes of META_PREALLOCATED reservation to META_PERTRANS.
395 * This is called when preallocated meta reservation needs to be used.
396 * Normally after btrfs_join_transaction() call.
398 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes);
400 void btrfs_qgroup_check_reserved_leak(struct inode *inode);
402 /* btrfs_qgroup_swapped_blocks related functions */
403 void btrfs_qgroup_init_swapped_blocks(
404 struct btrfs_qgroup_swapped_blocks *swapped_blocks);
406 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root);
407 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
408 struct btrfs_root *subvol_root,
409 struct btrfs_block_group_cache *bg,
410 struct extent_buffer *subvol_parent, int subvol_slot,
411 struct extent_buffer *reloc_parent, int reloc_slot,
413 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
414 struct btrfs_root *root, struct extent_buffer *eb);