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 /* 125 * Record a dirty extent, and info qgroup to update quota on it 126 */ 127 struct btrfs_qgroup_extent_record { 128 struct rb_node node; 129 u64 bytenr; 130 u64 num_bytes; 131 132 /* 133 * For qgroup reserved data space freeing. 134 * 135 * @data_rsv_refroot and @data_rsv will be recorded after 136 * BTRFS_ADD_DELAYED_EXTENT is called. 137 * And will be used to free reserved qgroup space at 138 * transaction commit time. 139 */ 140 u32 data_rsv; /* reserved data space needs to be freed */ 141 u64 data_rsv_refroot; /* which root the reserved data belongs to */ 142 struct ulist *old_roots; 143 }; 144 145 struct btrfs_qgroup_swapped_block { 146 struct rb_node node; 147 148 int level; 149 bool trace_leaf; 150 151 /* bytenr/generation of the tree block in subvolume tree after swap */ 152 u64 subvol_bytenr; 153 u64 subvol_generation; 154 155 /* bytenr/generation of the tree block in reloc tree after swap */ 156 u64 reloc_bytenr; 157 u64 reloc_generation; 158 159 u64 last_snapshot; 160 struct btrfs_key first_key; 161 }; 162 163 /* 164 * Qgroup reservation types: 165 * 166 * DATA: 167 * space reserved for data 168 * 169 * META_PERTRANS: 170 * Space reserved for metadata (per-transaction) 171 * Due to the fact that qgroup data is only updated at transaction commit 172 * time, reserved space for metadata must be kept until transaction 173 * commits. 174 * Any metadata reserved that are used in btrfs_start_transaction() should 175 * be of this type. 176 * 177 * META_PREALLOC: 178 * There are cases where metadata space is reserved before starting 179 * transaction, and then btrfs_join_transaction() to get a trans handle. 180 * Any metadata reserved for such usage should be of this type. 181 * And after join_transaction() part (or all) of such reservation should 182 * be converted into META_PERTRANS. 183 */ 184 enum btrfs_qgroup_rsv_type { 185 BTRFS_QGROUP_RSV_DATA, 186 BTRFS_QGROUP_RSV_META_PERTRANS, 187 BTRFS_QGROUP_RSV_META_PREALLOC, 188 BTRFS_QGROUP_RSV_LAST, 189 }; 190 191 /* 192 * Represents how many bytes we have reserved for this qgroup. 193 * 194 * Each type should have different reservation behavior. 195 * E.g, data follows its io_tree flag modification, while 196 * *currently* meta is just reserve-and-clear during transaction. 197 * 198 * TODO: Add new type for reservation which can survive transaction commit. 199 * Current metadata reservation behavior is not suitable for such case. 200 */ 201 struct btrfs_qgroup_rsv { 202 u64 values[BTRFS_QGROUP_RSV_LAST]; 203 }; 204 205 /* 206 * one struct for each qgroup, organized in fs_info->qgroup_tree. 207 */ 208 struct btrfs_qgroup { 209 u64 qgroupid; 210 211 /* 212 * state 213 */ 214 u64 rfer; /* referenced */ 215 u64 rfer_cmpr; /* referenced compressed */ 216 u64 excl; /* exclusive */ 217 u64 excl_cmpr; /* exclusive compressed */ 218 219 /* 220 * limits 221 */ 222 u64 lim_flags; /* which limits are set */ 223 u64 max_rfer; 224 u64 max_excl; 225 u64 rsv_rfer; 226 u64 rsv_excl; 227 228 /* 229 * reservation tracking 230 */ 231 struct btrfs_qgroup_rsv rsv; 232 233 /* 234 * lists 235 */ 236 struct list_head groups; /* groups this group is member of */ 237 struct list_head members; /* groups that are members of this group */ 238 struct list_head dirty; /* dirty groups */ 239 240 /* 241 * For qgroup iteration usage. 242 * 243 * The iteration list should always be empty until qgroup_iterator_add() 244 * is called. And should be reset to empty after the iteration is 245 * finished. 246 */ 247 struct list_head iterator; 248 249 /* 250 * For nested iterator usage. 251 * 252 * Here we support at most one level of nested iterator calls like: 253 * 254 * LIST_HEAD(all_qgroups); 255 * { 256 * LIST_HEAD(local_qgroups); 257 * qgroup_iterator_add(local_qgroups, qg); 258 * qgroup_iterator_nested_add(all_qgroups, qg); 259 * do_some_work(local_qgroups); 260 * qgroup_iterator_clean(local_qgroups); 261 * } 262 * do_some_work(all_qgroups); 263 * qgroup_iterator_nested_clean(all_qgroups); 264 */ 265 struct list_head nested_iterator; 266 struct rb_node node; /* tree of qgroups */ 267 268 /* 269 * temp variables for accounting operations 270 * Refer to qgroup_shared_accounting() for details. 271 */ 272 u64 old_refcnt; 273 u64 new_refcnt; 274 275 /* 276 * Sysfs kobjectid 277 */ 278 struct kobject kobj; 279 }; 280 281 /* Glue structure to represent the relations between qgroups. */ 282 struct btrfs_qgroup_list { 283 struct list_head next_group; 284 struct list_head next_member; 285 struct btrfs_qgroup *group; 286 struct btrfs_qgroup *member; 287 }; 288 289 struct btrfs_squota_delta { 290 /* The fstree root this delta counts against. */ 291 u64 root; 292 /* The number of bytes in the extent being counted. */ 293 u64 num_bytes; 294 /* The generation the extent was created in. */ 295 u64 generation; 296 /* Whether we are using or freeing the extent. */ 297 bool is_inc; 298 /* Whether the extent is data or metadata. */ 299 bool is_data; 300 }; 301 302 static inline u64 btrfs_qgroup_subvolid(u64 qgroupid) 303 { 304 return (qgroupid & ((1ULL << BTRFS_QGROUP_LEVEL_SHIFT) - 1)); 305 } 306 307 /* 308 * For qgroup event trace points only 309 */ 310 enum { 311 ENUM_BIT(QGROUP_RESERVE), 312 ENUM_BIT(QGROUP_RELEASE), 313 ENUM_BIT(QGROUP_FREE), 314 }; 315 316 enum btrfs_qgroup_mode { 317 BTRFS_QGROUP_MODE_DISABLED, 318 BTRFS_QGROUP_MODE_FULL, 319 BTRFS_QGROUP_MODE_SIMPLE 320 }; 321 322 enum btrfs_qgroup_mode btrfs_qgroup_mode(const struct btrfs_fs_info *fs_info); 323 bool btrfs_qgroup_enabled(const struct btrfs_fs_info *fs_info); 324 bool btrfs_qgroup_full_accounting(const struct btrfs_fs_info *fs_info); 325 int btrfs_quota_enable(struct btrfs_fs_info *fs_info, 326 struct btrfs_ioctl_quota_ctl_args *quota_ctl_args); 327 int btrfs_quota_disable(struct btrfs_fs_info *fs_info); 328 int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info); 329 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info); 330 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info, 331 bool interruptible); 332 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src, u64 dst, 333 struct btrfs_qgroup_list *prealloc); 334 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src, 335 u64 dst); 336 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid); 337 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid); 338 int btrfs_qgroup_cleanup_dropped_subvolume(struct btrfs_fs_info *fs_info, u64 subvolid); 339 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid, 340 struct btrfs_qgroup_limit *limit); 341 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info); 342 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info); 343 344 int btrfs_qgroup_trace_extent_nolock( 345 struct btrfs_fs_info *fs_info, 346 struct btrfs_delayed_ref_root *delayed_refs, 347 struct btrfs_qgroup_extent_record *record); 348 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans, 349 struct btrfs_qgroup_extent_record *qrecord); 350 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr, 351 u64 num_bytes); 352 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans, 353 struct extent_buffer *eb); 354 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans, 355 struct extent_buffer *root_eb, 356 u64 root_gen, int root_level); 357 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr, 358 u64 num_bytes, struct ulist *old_roots, 359 struct ulist *new_roots); 360 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans); 361 int btrfs_run_qgroups(struct btrfs_trans_handle *trans); 362 int btrfs_qgroup_check_inherit(struct btrfs_fs_info *fs_info, 363 struct btrfs_qgroup_inherit *inherit, 364 size_t size); 365 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid, 366 u64 objectid, u64 inode_rootid, 367 struct btrfs_qgroup_inherit *inherit); 368 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info, 369 u64 ref_root, u64 num_bytes, 370 enum btrfs_qgroup_rsv_type type); 371 372 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 373 int btrfs_verify_qgroup_counts(const struct btrfs_fs_info *fs_info, u64 qgroupid, 374 u64 rfer, u64 excl); 375 #endif 376 377 /* New io_tree based accurate qgroup reserve API */ 378 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode, 379 struct extent_changeset **reserved, u64 start, u64 len); 380 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len, u64 *released); 381 int btrfs_qgroup_free_data(struct btrfs_inode *inode, 382 struct extent_changeset *reserved, u64 start, 383 u64 len, u64 *freed); 384 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes, 385 enum btrfs_qgroup_rsv_type type, bool enforce); 386 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes, 387 enum btrfs_qgroup_rsv_type type, bool enforce, 388 bool noflush); 389 /* Reserve metadata space for pertrans and prealloc type */ 390 static inline int btrfs_qgroup_reserve_meta_pertrans(struct btrfs_root *root, 391 int num_bytes, bool enforce) 392 { 393 return __btrfs_qgroup_reserve_meta(root, num_bytes, 394 BTRFS_QGROUP_RSV_META_PERTRANS, 395 enforce, false); 396 } 397 static inline int btrfs_qgroup_reserve_meta_prealloc(struct btrfs_root *root, 398 int num_bytes, bool enforce, 399 bool noflush) 400 { 401 return __btrfs_qgroup_reserve_meta(root, num_bytes, 402 BTRFS_QGROUP_RSV_META_PREALLOC, 403 enforce, noflush); 404 } 405 406 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes, 407 enum btrfs_qgroup_rsv_type type); 408 409 /* Free per-transaction meta reservation for error handling */ 410 static inline void btrfs_qgroup_free_meta_pertrans(struct btrfs_root *root, 411 int num_bytes) 412 { 413 __btrfs_qgroup_free_meta(root, num_bytes, 414 BTRFS_QGROUP_RSV_META_PERTRANS); 415 } 416 417 /* Pre-allocated meta reservation can be freed at need */ 418 static inline void btrfs_qgroup_free_meta_prealloc(struct btrfs_root *root, 419 int num_bytes) 420 { 421 __btrfs_qgroup_free_meta(root, num_bytes, 422 BTRFS_QGROUP_RSV_META_PREALLOC); 423 } 424 425 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root); 426 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes); 427 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode); 428 429 /* btrfs_qgroup_swapped_blocks related functions */ 430 void btrfs_qgroup_init_swapped_blocks( 431 struct btrfs_qgroup_swapped_blocks *swapped_blocks); 432 433 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root); 434 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans, 435 struct btrfs_root *subvol_root, 436 struct btrfs_block_group *bg, 437 struct extent_buffer *subvol_parent, int subvol_slot, 438 struct extent_buffer *reloc_parent, int reloc_slot, 439 u64 last_snapshot); 440 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans, 441 struct btrfs_root *root, struct extent_buffer *eb); 442 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans); 443 bool btrfs_check_quota_leak(const struct btrfs_fs_info *fs_info); 444 void btrfs_free_squota_rsv(struct btrfs_fs_info *fs_info, u64 root, u64 rsv_bytes); 445 int btrfs_record_squota_delta(struct btrfs_fs_info *fs_info, 446 const struct btrfs_squota_delta *delta); 447 448 #endif 449