xref: /linux/fs/btrfs/qgroup.h (revision 2fe3c78a2c26dd5ee811024a1b7d6cfb4d654319)
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 	u64 bytenr;
129 	u64 num_bytes;
130 
131 	/*
132 	 * For qgroup reserved data space freeing.
133 	 *
134 	 * @data_rsv_refroot and @data_rsv will be recorded after
135 	 * BTRFS_ADD_DELAYED_EXTENT is called.
136 	 * And will be used to free reserved qgroup space at
137 	 * transaction commit time.
138 	 */
139 	u32 data_rsv;		/* reserved data space needs to be freed */
140 	u64 data_rsv_refroot;	/* which root the reserved data belongs to */
141 	struct ulist *old_roots;
142 };
143 
144 struct btrfs_qgroup_swapped_block {
145 	struct rb_node node;
146 
147 	int level;
148 	bool trace_leaf;
149 
150 	/* bytenr/generation of the tree block in subvolume tree after swap */
151 	u64 subvol_bytenr;
152 	u64 subvol_generation;
153 
154 	/* bytenr/generation of the tree block in reloc tree after swap */
155 	u64 reloc_bytenr;
156 	u64 reloc_generation;
157 
158 	u64 last_snapshot;
159 	struct btrfs_key first_key;
160 };
161 
162 /*
163  * Qgroup reservation types:
164  *
165  * DATA:
166  *	space reserved for data
167  *
168  * META_PERTRANS:
169  * 	Space reserved for metadata (per-transaction)
170  * 	Due to the fact that qgroup data is only updated at transaction commit
171  * 	time, reserved space for metadata must be kept until transaction
172  * 	commits.
173  * 	Any metadata reserved that are used in btrfs_start_transaction() should
174  * 	be of this type.
175  *
176  * META_PREALLOC:
177  *	There are cases where metadata space is reserved before starting
178  *	transaction, and then btrfs_join_transaction() to get a trans handle.
179  *	Any metadata reserved for such usage should be of this type.
180  *	And after join_transaction() part (or all) of such reservation should
181  *	be converted into META_PERTRANS.
182  */
183 enum btrfs_qgroup_rsv_type {
184 	BTRFS_QGROUP_RSV_DATA,
185 	BTRFS_QGROUP_RSV_META_PERTRANS,
186 	BTRFS_QGROUP_RSV_META_PREALLOC,
187 	BTRFS_QGROUP_RSV_LAST,
188 };
189 
190 /*
191  * Represents how many bytes we have reserved for this qgroup.
192  *
193  * Each type should have different reservation behavior.
194  * E.g, data follows its io_tree flag modification, while
195  * *currently* meta is just reserve-and-clear during transaction.
196  *
197  * TODO: Add new type for reservation which can survive transaction commit.
198  * Current metadata reservation behavior is not suitable for such case.
199  */
200 struct btrfs_qgroup_rsv {
201 	u64 values[BTRFS_QGROUP_RSV_LAST];
202 };
203 
204 /*
205  * one struct for each qgroup, organized in fs_info->qgroup_tree.
206  */
207 struct btrfs_qgroup {
208 	u64 qgroupid;
209 
210 	/*
211 	 * state
212 	 */
213 	u64 rfer;	/* referenced */
214 	u64 rfer_cmpr;	/* referenced compressed */
215 	u64 excl;	/* exclusive */
216 	u64 excl_cmpr;	/* exclusive compressed */
217 
218 	/*
219 	 * limits
220 	 */
221 	u64 lim_flags;	/* which limits are set */
222 	u64 max_rfer;
223 	u64 max_excl;
224 	u64 rsv_rfer;
225 	u64 rsv_excl;
226 
227 	/*
228 	 * reservation tracking
229 	 */
230 	struct btrfs_qgroup_rsv rsv;
231 
232 	/*
233 	 * lists
234 	 */
235 	struct list_head groups;  /* groups this group is member of */
236 	struct list_head members; /* groups that are members of this group */
237 	struct list_head dirty;   /* dirty groups */
238 
239 	/*
240 	 * For qgroup iteration usage.
241 	 *
242 	 * The iteration list should always be empty until qgroup_iterator_add()
243 	 * is called.  And should be reset to empty after the iteration is
244 	 * finished.
245 	 */
246 	struct list_head iterator;
247 
248 	/*
249 	 * For nested iterator usage.
250 	 *
251 	 * Here we support at most one level of nested iterator calls like:
252 	 *
253 	 *	LIST_HEAD(all_qgroups);
254 	 *	{
255 	 *		LIST_HEAD(local_qgroups);
256 	 *		qgroup_iterator_add(local_qgroups, qg);
257 	 *		qgroup_iterator_nested_add(all_qgroups, qg);
258 	 *		do_some_work(local_qgroups);
259 	 *		qgroup_iterator_clean(local_qgroups);
260 	 *	}
261 	 *	do_some_work(all_qgroups);
262 	 *	qgroup_iterator_nested_clean(all_qgroups);
263 	 */
264 	struct list_head nested_iterator;
265 	struct rb_node node;	  /* tree of qgroups */
266 
267 	/*
268 	 * temp variables for accounting operations
269 	 * Refer to qgroup_shared_accounting() for details.
270 	 */
271 	u64 old_refcnt;
272 	u64 new_refcnt;
273 
274 	/*
275 	 * Sysfs kobjectid
276 	 */
277 	struct kobject kobj;
278 };
279 
280 /* Glue structure to represent the relations between qgroups. */
281 struct btrfs_qgroup_list {
282 	struct list_head next_group;
283 	struct list_head next_member;
284 	struct btrfs_qgroup *group;
285 	struct btrfs_qgroup *member;
286 };
287 
288 struct btrfs_squota_delta {
289 	/* The fstree root this delta counts against. */
290 	u64 root;
291 	/* The number of bytes in the extent being counted. */
292 	u64 num_bytes;
293 	/* The generation the extent was created in. */
294 	u64 generation;
295 	/* Whether we are using or freeing the extent. */
296 	bool is_inc;
297 	/* Whether the extent is data or metadata. */
298 	bool is_data;
299 };
300 
301 static inline u64 btrfs_qgroup_subvolid(u64 qgroupid)
302 {
303 	return (qgroupid & ((1ULL << BTRFS_QGROUP_LEVEL_SHIFT) - 1));
304 }
305 
306 /*
307  * For qgroup event trace points only
308  */
309 enum {
310 	ENUM_BIT(QGROUP_RESERVE),
311 	ENUM_BIT(QGROUP_RELEASE),
312 	ENUM_BIT(QGROUP_FREE),
313 };
314 
315 enum btrfs_qgroup_mode {
316 	BTRFS_QGROUP_MODE_DISABLED,
317 	BTRFS_QGROUP_MODE_FULL,
318 	BTRFS_QGROUP_MODE_SIMPLE
319 };
320 
321 enum btrfs_qgroup_mode btrfs_qgroup_mode(const struct btrfs_fs_info *fs_info);
322 bool btrfs_qgroup_enabled(const struct btrfs_fs_info *fs_info);
323 bool btrfs_qgroup_full_accounting(const struct btrfs_fs_info *fs_info);
324 int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
325 		       struct btrfs_ioctl_quota_ctl_args *quota_ctl_args);
326 int btrfs_quota_disable(struct btrfs_fs_info *fs_info);
327 int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
328 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
329 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
330 				     bool interruptible);
331 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src, u64 dst,
332 			      struct btrfs_qgroup_list *prealloc);
333 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
334 			      u64 dst);
335 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
336 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
337 int btrfs_qgroup_cleanup_dropped_subvolume(struct btrfs_fs_info *fs_info, u64 subvolid);
338 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
339 		       struct btrfs_qgroup_limit *limit);
340 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info);
341 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info);
342 
343 int btrfs_qgroup_trace_extent_nolock(
344 		struct btrfs_fs_info *fs_info,
345 		struct btrfs_delayed_ref_root *delayed_refs,
346 		struct btrfs_qgroup_extent_record *record);
347 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
348 				   struct btrfs_qgroup_extent_record *qrecord);
349 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
350 			      u64 num_bytes);
351 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
352 				  struct extent_buffer *eb);
353 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
354 			       struct extent_buffer *root_eb,
355 			       u64 root_gen, int root_level);
356 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
357 				u64 num_bytes, struct ulist *old_roots,
358 				struct ulist *new_roots);
359 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans);
360 int btrfs_run_qgroups(struct btrfs_trans_handle *trans);
361 int btrfs_qgroup_check_inherit(struct btrfs_fs_info *fs_info,
362 			       struct btrfs_qgroup_inherit *inherit,
363 			       size_t size);
364 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
365 			 u64 objectid, u64 inode_rootid,
366 			 struct btrfs_qgroup_inherit *inherit);
367 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
368 			       u64 ref_root, u64 num_bytes,
369 			       enum btrfs_qgroup_rsv_type type);
370 
371 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
372 int btrfs_verify_qgroup_counts(const struct btrfs_fs_info *fs_info, u64 qgroupid,
373 			       u64 rfer, u64 excl);
374 #endif
375 
376 /* New io_tree based accurate qgroup reserve API */
377 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
378 			struct extent_changeset **reserved, u64 start, u64 len);
379 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len, u64 *released);
380 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
381 			   struct extent_changeset *reserved, u64 start,
382 			   u64 len, u64 *freed);
383 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
384 			      enum btrfs_qgroup_rsv_type type, bool enforce);
385 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
386 				enum btrfs_qgroup_rsv_type type, bool enforce,
387 				bool noflush);
388 /* Reserve metadata space for pertrans and prealloc type */
389 static inline int btrfs_qgroup_reserve_meta_pertrans(struct btrfs_root *root,
390 				int num_bytes, bool enforce)
391 {
392 	return __btrfs_qgroup_reserve_meta(root, num_bytes,
393 					   BTRFS_QGROUP_RSV_META_PERTRANS,
394 					   enforce, false);
395 }
396 static inline int btrfs_qgroup_reserve_meta_prealloc(struct btrfs_root *root,
397 						     int num_bytes, bool enforce,
398 						     bool noflush)
399 {
400 	return __btrfs_qgroup_reserve_meta(root, num_bytes,
401 					   BTRFS_QGROUP_RSV_META_PREALLOC,
402 					   enforce, noflush);
403 }
404 
405 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
406 			     enum btrfs_qgroup_rsv_type type);
407 
408 /* Free per-transaction meta reservation for error handling */
409 static inline void btrfs_qgroup_free_meta_pertrans(struct btrfs_root *root,
410 						   int num_bytes)
411 {
412 	__btrfs_qgroup_free_meta(root, num_bytes,
413 			BTRFS_QGROUP_RSV_META_PERTRANS);
414 }
415 
416 /* Pre-allocated meta reservation can be freed at need */
417 static inline void btrfs_qgroup_free_meta_prealloc(struct btrfs_root *root,
418 						   int num_bytes)
419 {
420 	__btrfs_qgroup_free_meta(root, num_bytes,
421 			BTRFS_QGROUP_RSV_META_PREALLOC);
422 }
423 
424 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root);
425 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes);
426 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode);
427 
428 /* btrfs_qgroup_swapped_blocks related functions */
429 void btrfs_qgroup_init_swapped_blocks(
430 	struct btrfs_qgroup_swapped_blocks *swapped_blocks);
431 
432 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root);
433 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
434 		struct btrfs_root *subvol_root,
435 		struct btrfs_block_group *bg,
436 		struct extent_buffer *subvol_parent, int subvol_slot,
437 		struct extent_buffer *reloc_parent, int reloc_slot,
438 		u64 last_snapshot);
439 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
440 		struct btrfs_root *root, struct extent_buffer *eb);
441 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans);
442 bool btrfs_check_quota_leak(const struct btrfs_fs_info *fs_info);
443 void btrfs_free_squota_rsv(struct btrfs_fs_info *fs_info, u64 root, u64 rsv_bytes);
444 int btrfs_record_squota_delta(struct btrfs_fs_info *fs_info,
445 			      const struct btrfs_squota_delta *delta);
446 
447 #endif
448