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/spinlock.h>
10 #include <linux/rbtree.h>
11 #include <linux/kobject.h>
12 #include "ulist.h"
13 #include "delayed-ref.h"
14 #include "misc.h"
15 
16 /*
17  * Btrfs qgroup overview
18  *
19  * Btrfs qgroup splits into 3 main part:
20  * 1) Reserve
21  *    Reserve metadata/data space for incoming operations
22  *    Affect how qgroup limit works
23  *
24  * 2) Trace
25  *    Tell btrfs qgroup to trace dirty extents.
26  *
27  *    Dirty extents including:
28  *    - Newly allocated extents
29  *    - Extents going to be deleted (in this trans)
30  *    - Extents whose owner is going to be modified
31  *
32  *    This is the main part affects whether qgroup numbers will stay
33  *    consistent.
34  *    Btrfs qgroup can trace clean extents and won't cause any problem,
35  *    but it will consume extra CPU time, it should be avoided if possible.
36  *
37  * 3) Account
38  *    Btrfs qgroup will updates its numbers, based on dirty extents traced
39  *    in previous step.
40  *
41  *    Normally at qgroup rescan and transaction commit time.
42  */
43 
44 /*
45  * Special performance optimization for balance.
46  *
47  * For balance, we need to swap subtree of subvolume and reloc trees.
48  * In theory, we need to trace all subtree blocks of both subvolume and reloc
49  * trees, since their owner has changed during such swap.
50  *
51  * However since balance has ensured that both subtrees are containing the
52  * same contents and have the same tree structures, such swap won't cause
53  * qgroup number change.
54  *
55  * But there is a race window between subtree swap and transaction commit,
56  * during that window, if we increase/decrease tree level or merge/split tree
57  * blocks, we still need to trace the original subtrees.
58  *
59  * So for balance, we use a delayed subtree tracing, whose workflow is:
60  *
61  * 1) Record the subtree root block get swapped.
62  *
63  *    During subtree swap:
64  *    O = Old tree blocks
65  *    N = New tree blocks
66  *          reloc tree                     subvolume tree X
67  *             Root                               Root
68  *            /    \                             /    \
69  *          NA     OB                          OA      OB
70  *        /  |     |  \                      /  |      |  \
71  *      NC  ND     OE  OF                   OC  OD     OE  OF
72  *
73  *   In this case, NA and OA are going to be swapped, record (NA, OA) into
74  *   subvolume tree X.
75  *
76  * 2) After subtree swap.
77  *          reloc tree                     subvolume tree X
78  *             Root                               Root
79  *            /    \                             /    \
80  *          OA     OB                          NA      OB
81  *        /  |     |  \                      /  |      |  \
82  *      OC  OD     OE  OF                   NC  ND     OE  OF
83  *
84  * 3a) COW happens for OB
85  *     If we are going to COW tree block OB, we check OB's bytenr against
86  *     tree X's swapped_blocks structure.
87  *     If it doesn't fit any, nothing will happen.
88  *
89  * 3b) COW happens for NA
90  *     Check NA's bytenr against tree X's swapped_blocks, and get a hit.
91  *     Then we do subtree scan on both subtrees OA and NA.
92  *     Resulting 6 tree blocks to be scanned (OA, OC, OD, NA, NC, ND).
93  *
94  *     Then no matter what we do to subvolume tree X, qgroup numbers will
95  *     still be correct.
96  *     Then NA's record gets removed from X's swapped_blocks.
97  *
98  * 4)  Transaction commit
99  *     Any record in X's swapped_blocks gets removed, since there is no
100  *     modification to the swapped subtrees, no need to trigger heavy qgroup
101  *     subtree rescan for them.
102  */
103 
104 #define BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN		(1UL << 3)
105 #define BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING		(1UL << 4)
106 
107 /*
108  * Record a dirty extent, and info qgroup to update quota on it
109  * TODO: Use kmem cache to alloc it.
110  */
111 struct btrfs_qgroup_extent_record {
112 	struct rb_node node;
113 	u64 bytenr;
114 	u64 num_bytes;
115 
116 	/*
117 	 * For qgroup reserved data space freeing.
118 	 *
119 	 * @data_rsv_refroot and @data_rsv will be recorded after
120 	 * BTRFS_ADD_DELAYED_EXTENT is called.
121 	 * And will be used to free reserved qgroup space at
122 	 * transaction commit time.
123 	 */
124 	u32 data_rsv;		/* reserved data space needs to be freed */
125 	u64 data_rsv_refroot;	/* which root the reserved data belongs to */
126 	struct ulist *old_roots;
127 };
128 
129 struct btrfs_qgroup_swapped_block {
130 	struct rb_node node;
131 
132 	int level;
133 	bool trace_leaf;
134 
135 	/* bytenr/generation of the tree block in subvolume tree after swap */
136 	u64 subvol_bytenr;
137 	u64 subvol_generation;
138 
139 	/* bytenr/generation of the tree block in reloc tree after swap */
140 	u64 reloc_bytenr;
141 	u64 reloc_generation;
142 
143 	u64 last_snapshot;
144 	struct btrfs_key first_key;
145 };
146 
147 /*
148  * Qgroup reservation types:
149  *
150  * DATA:
151  *	space reserved for data
152  *
153  * META_PERTRANS:
154  * 	Space reserved for metadata (per-transaction)
155  * 	Due to the fact that qgroup data is only updated at transaction commit
156  * 	time, reserved space for metadata must be kept until transaction
157  * 	commits.
158  * 	Any metadata reserved that are used in btrfs_start_transaction() should
159  * 	be of this type.
160  *
161  * META_PREALLOC:
162  *	There are cases where metadata space is reserved before starting
163  *	transaction, and then btrfs_join_transaction() to get a trans handle.
164  *	Any metadata reserved for such usage should be of this type.
165  *	And after join_transaction() part (or all) of such reservation should
166  *	be converted into META_PERTRANS.
167  */
168 enum btrfs_qgroup_rsv_type {
169 	BTRFS_QGROUP_RSV_DATA,
170 	BTRFS_QGROUP_RSV_META_PERTRANS,
171 	BTRFS_QGROUP_RSV_META_PREALLOC,
172 	BTRFS_QGROUP_RSV_LAST,
173 };
174 
175 /*
176  * Represents how many bytes we have reserved for this qgroup.
177  *
178  * Each type should have different reservation behavior.
179  * E.g, data follows its io_tree flag modification, while
180  * *currently* meta is just reserve-and-clear during transaction.
181  *
182  * TODO: Add new type for reservation which can survive transaction commit.
183  * Current metadata reservation behavior is not suitable for such case.
184  */
185 struct btrfs_qgroup_rsv {
186 	u64 values[BTRFS_QGROUP_RSV_LAST];
187 };
188 
189 /*
190  * one struct for each qgroup, organized in fs_info->qgroup_tree.
191  */
192 struct btrfs_qgroup {
193 	u64 qgroupid;
194 
195 	/*
196 	 * state
197 	 */
198 	u64 rfer;	/* referenced */
199 	u64 rfer_cmpr;	/* referenced compressed */
200 	u64 excl;	/* exclusive */
201 	u64 excl_cmpr;	/* exclusive compressed */
202 
203 	/*
204 	 * limits
205 	 */
206 	u64 lim_flags;	/* which limits are set */
207 	u64 max_rfer;
208 	u64 max_excl;
209 	u64 rsv_rfer;
210 	u64 rsv_excl;
211 
212 	/*
213 	 * reservation tracking
214 	 */
215 	struct btrfs_qgroup_rsv rsv;
216 
217 	/*
218 	 * lists
219 	 */
220 	struct list_head groups;  /* groups this group is member of */
221 	struct list_head members; /* groups that are members of this group */
222 	struct list_head dirty;   /* dirty groups */
223 	struct rb_node node;	  /* tree of qgroups */
224 
225 	/*
226 	 * temp variables for accounting operations
227 	 * Refer to qgroup_shared_accounting() for details.
228 	 */
229 	u64 old_refcnt;
230 	u64 new_refcnt;
231 
232 	/*
233 	 * Sysfs kobjectid
234 	 */
235 	struct kobject kobj;
236 };
237 
btrfs_qgroup_subvolid(u64 qgroupid)238 static inline u64 btrfs_qgroup_subvolid(u64 qgroupid)
239 {
240 	return (qgroupid & ((1ULL << BTRFS_QGROUP_LEVEL_SHIFT) - 1));
241 }
242 
243 /*
244  * For qgroup event trace points only
245  */
246 enum {
247 	ENUM_BIT(QGROUP_RESERVE),
248 	ENUM_BIT(QGROUP_RELEASE),
249 	ENUM_BIT(QGROUP_FREE),
250 };
251 
252 int btrfs_quota_enable(struct btrfs_fs_info *fs_info);
253 int btrfs_quota_disable(struct btrfs_fs_info *fs_info);
254 int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info);
255 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info);
256 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
257 				     bool interruptible);
258 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
259 			      u64 dst);
260 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
261 			      u64 dst);
262 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
263 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid);
264 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
265 		       struct btrfs_qgroup_limit *limit);
266 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info);
267 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info);
268 struct btrfs_delayed_extent_op;
269 
270 /*
271  * Inform qgroup to trace one dirty extent, its info is recorded in @record.
272  * So qgroup can account it at transaction committing time.
273  *
274  * No lock version, caller must acquire delayed ref lock and allocated memory,
275  * then call btrfs_qgroup_trace_extent_post() after exiting lock context.
276  *
277  * Return 0 for success insert
278  * Return >0 for existing record, caller can free @record safely.
279  * Error is not possible
280  */
281 int btrfs_qgroup_trace_extent_nolock(
282 		struct btrfs_fs_info *fs_info,
283 		struct btrfs_delayed_ref_root *delayed_refs,
284 		struct btrfs_qgroup_extent_record *record);
285 
286 /*
287  * Post handler after qgroup_trace_extent_nolock().
288  *
289  * NOTE: Current qgroup does the expensive backref walk at transaction
290  * committing time with TRANS_STATE_COMMIT_DOING, this blocks incoming
291  * new transaction.
292  * This is designed to allow btrfs_find_all_roots() to get correct new_roots
293  * result.
294  *
295  * However for old_roots there is no need to do backref walk at that time,
296  * since we search commit roots to walk backref and result will always be
297  * correct.
298  *
299  * Due to the nature of no lock version, we can't do backref there.
300  * So we must call btrfs_qgroup_trace_extent_post() after exiting
301  * spinlock context.
302  *
303  * TODO: If we can fix and prove btrfs_find_all_roots() can get correct result
304  * using current root, then we can move all expensive backref walk out of
305  * transaction committing, but not now as qgroup accounting will be wrong again.
306  */
307 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
308 				   struct btrfs_qgroup_extent_record *qrecord);
309 
310 /*
311  * Inform qgroup to trace one dirty extent, specified by @bytenr and
312  * @num_bytes.
313  * So qgroup can account it at commit trans time.
314  *
315  * Better encapsulated version, with memory allocation and backref walk for
316  * commit roots.
317  * So this can sleep.
318  *
319  * Return 0 if the operation is done.
320  * Return <0 for error, like memory allocation failure or invalid parameter
321  * (NULL trans)
322  */
323 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
324 			      u64 num_bytes);
325 
326 /*
327  * Inform qgroup to trace all leaf items of data
328  *
329  * Return 0 for success
330  * Return <0 for error(ENOMEM)
331  */
332 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
333 				  struct extent_buffer *eb);
334 /*
335  * Inform qgroup to trace a whole subtree, including all its child tree
336  * blocks and data.
337  * The root tree block is specified by @root_eb.
338  *
339  * Normally used by relocation(tree block swap) and subvolume deletion.
340  *
341  * Return 0 for success
342  * Return <0 for error(ENOMEM or tree search error)
343  */
344 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
345 			       struct extent_buffer *root_eb,
346 			       u64 root_gen, int root_level);
347 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
348 				u64 num_bytes, struct ulist *old_roots,
349 				struct ulist *new_roots);
350 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans);
351 int btrfs_run_qgroups(struct btrfs_trans_handle *trans);
352 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
353 			 u64 objectid, struct btrfs_qgroup_inherit *inherit);
354 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
355 			       u64 ref_root, u64 num_bytes,
356 			       enum btrfs_qgroup_rsv_type type);
357 
358 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
359 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
360 			       u64 rfer, u64 excl);
361 #endif
362 
363 /* New io_tree based accurate qgroup reserve API */
364 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
365 			struct extent_changeset **reserved, u64 start, u64 len);
366 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len);
367 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
368 			   struct extent_changeset *reserved, u64 start,
369 			   u64 len);
370 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
371 			      enum btrfs_qgroup_rsv_type type, bool enforce);
372 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
373 				enum btrfs_qgroup_rsv_type type, bool enforce,
374 				bool noflush);
375 /* Reserve metadata space for pertrans and prealloc type */
btrfs_qgroup_reserve_meta_pertrans(struct btrfs_root * root,int num_bytes,bool enforce)376 static inline int btrfs_qgroup_reserve_meta_pertrans(struct btrfs_root *root,
377 				int num_bytes, bool enforce)
378 {
379 	return __btrfs_qgroup_reserve_meta(root, num_bytes,
380 					   BTRFS_QGROUP_RSV_META_PERTRANS,
381 					   enforce, false);
382 }
btrfs_qgroup_reserve_meta_prealloc(struct btrfs_root * root,int num_bytes,bool enforce,bool noflush)383 static inline int btrfs_qgroup_reserve_meta_prealloc(struct btrfs_root *root,
384 						     int num_bytes, bool enforce,
385 						     bool noflush)
386 {
387 	return __btrfs_qgroup_reserve_meta(root, num_bytes,
388 					   BTRFS_QGROUP_RSV_META_PREALLOC,
389 					   enforce, noflush);
390 }
391 
392 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
393 			     enum btrfs_qgroup_rsv_type type);
394 
395 /* Free per-transaction meta reservation for error handling */
btrfs_qgroup_free_meta_pertrans(struct btrfs_root * root,int num_bytes)396 static inline void btrfs_qgroup_free_meta_pertrans(struct btrfs_root *root,
397 						   int num_bytes)
398 {
399 	__btrfs_qgroup_free_meta(root, num_bytes,
400 			BTRFS_QGROUP_RSV_META_PERTRANS);
401 }
402 
403 /* Pre-allocated meta reservation can be freed at need */
btrfs_qgroup_free_meta_prealloc(struct btrfs_root * root,int num_bytes)404 static inline void btrfs_qgroup_free_meta_prealloc(struct btrfs_root *root,
405 						   int num_bytes)
406 {
407 	__btrfs_qgroup_free_meta(root, num_bytes,
408 			BTRFS_QGROUP_RSV_META_PREALLOC);
409 }
410 
411 /*
412  * Per-transaction meta reservation should be all freed at transaction commit
413  * time
414  */
415 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root);
416 
417 /*
418  * Convert @num_bytes of META_PREALLOCATED reservation to META_PERTRANS.
419  *
420  * This is called when preallocated meta reservation needs to be used.
421  * Normally after btrfs_join_transaction() call.
422  */
423 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes);
424 
425 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode);
426 
427 /* btrfs_qgroup_swapped_blocks related functions */
428 void btrfs_qgroup_init_swapped_blocks(
429 	struct btrfs_qgroup_swapped_blocks *swapped_blocks);
430 
431 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root);
432 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
433 		struct btrfs_root *subvol_root,
434 		struct btrfs_block_group *bg,
435 		struct extent_buffer *subvol_parent, int subvol_slot,
436 		struct extent_buffer *reloc_parent, int reloc_slot,
437 		u64 last_snapshot);
438 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
439 		struct btrfs_root *root, struct extent_buffer *eb);
440 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans);
441 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info);
442 
443 #endif
444