xref: /openbmc/linux/fs/btrfs/delayed-ref.h (revision 4c004ed9)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Copyright (C) 2008 Oracle.  All rights reserved.
4  */
5 
6 #ifndef BTRFS_DELAYED_REF_H
7 #define BTRFS_DELAYED_REF_H
8 
9 #include <linux/refcount.h>
10 
11 /* these are the possible values of struct btrfs_delayed_ref_node->action */
12 #define BTRFS_ADD_DELAYED_REF    1 /* add one backref to the tree */
13 #define BTRFS_DROP_DELAYED_REF   2 /* delete one backref from the tree */
14 #define BTRFS_ADD_DELAYED_EXTENT 3 /* record a full extent allocation */
15 #define BTRFS_UPDATE_DELAYED_HEAD 4 /* not changing ref count on head ref */
16 
17 struct btrfs_delayed_ref_node {
18 	struct rb_node ref_node;
19 	/*
20 	 * If action is BTRFS_ADD_DELAYED_REF, also link this node to
21 	 * ref_head->ref_add_list, then we do not need to iterate the
22 	 * whole ref_head->ref_list to find BTRFS_ADD_DELAYED_REF nodes.
23 	 */
24 	struct list_head add_list;
25 
26 	/* the starting bytenr of the extent */
27 	u64 bytenr;
28 
29 	/* the size of the extent */
30 	u64 num_bytes;
31 
32 	/* seq number to keep track of insertion order */
33 	u64 seq;
34 
35 	/* ref count on this data structure */
36 	refcount_t refs;
37 
38 	/*
39 	 * how many refs is this entry adding or deleting.  For
40 	 * head refs, this may be a negative number because it is keeping
41 	 * track of the total mods done to the reference count.
42 	 * For individual refs, this will always be a positive number
43 	 *
44 	 * It may be more than one, since it is possible for a single
45 	 * parent to have more than one ref on an extent
46 	 */
47 	int ref_mod;
48 
49 	unsigned int action:8;
50 	unsigned int type:8;
51 };
52 
53 struct btrfs_delayed_extent_op {
54 	struct btrfs_disk_key key;
55 	u8 level;
56 	bool update_key;
57 	bool update_flags;
58 	u64 flags_to_set;
59 };
60 
61 /*
62  * the head refs are used to hold a lock on a given extent, which allows us
63  * to make sure that only one process is running the delayed refs
64  * at a time for a single extent.  They also store the sum of all the
65  * reference count modifications we've queued up.
66  */
67 struct btrfs_delayed_ref_head {
68 	u64 bytenr;
69 	u64 num_bytes;
70 	/*
71 	 * For insertion into struct btrfs_delayed_ref_root::href_root.
72 	 * Keep it in the same cache line as 'bytenr' for more efficient
73 	 * searches in the rbtree.
74 	 */
75 	struct rb_node href_node;
76 	/*
77 	 * the mutex is held while running the refs, and it is also
78 	 * held when checking the sum of reference modifications.
79 	 */
80 	struct mutex mutex;
81 
82 	refcount_t refs;
83 
84 	/* Protects 'ref_tree' and 'ref_add_list'. */
85 	spinlock_t lock;
86 	struct rb_root_cached ref_tree;
87 	/* accumulate add BTRFS_ADD_DELAYED_REF nodes to this ref_add_list. */
88 	struct list_head ref_add_list;
89 
90 	struct btrfs_delayed_extent_op *extent_op;
91 
92 	/*
93 	 * This is used to track the final ref_mod from all the refs associated
94 	 * with this head ref, this is not adjusted as delayed refs are run,
95 	 * this is meant to track if we need to do the csum accounting or not.
96 	 */
97 	int total_ref_mod;
98 
99 	/*
100 	 * This is the current outstanding mod references for this bytenr.  This
101 	 * is used with lookup_extent_info to get an accurate reference count
102 	 * for a bytenr, so it is adjusted as delayed refs are run so that any
103 	 * on disk reference count + ref_mod is accurate.
104 	 */
105 	int ref_mod;
106 
107 	/*
108 	 * when a new extent is allocated, it is just reserved in memory
109 	 * The actual extent isn't inserted into the extent allocation tree
110 	 * until the delayed ref is processed.  must_insert_reserved is
111 	 * used to flag a delayed ref so the accounting can be updated
112 	 * when a full insert is done.
113 	 *
114 	 * It is possible the extent will be freed before it is ever
115 	 * inserted into the extent allocation tree.  In this case
116 	 * we need to update the in ram accounting to properly reflect
117 	 * the free has happened.
118 	 */
119 	bool must_insert_reserved;
120 	bool is_data;
121 	bool is_system;
122 	bool processing;
123 };
124 
125 struct btrfs_delayed_tree_ref {
126 	struct btrfs_delayed_ref_node node;
127 	u64 root;
128 	u64 parent;
129 	int level;
130 };
131 
132 struct btrfs_delayed_data_ref {
133 	struct btrfs_delayed_ref_node node;
134 	u64 root;
135 	u64 parent;
136 	u64 objectid;
137 	u64 offset;
138 };
139 
140 enum btrfs_delayed_ref_flags {
141 	/* Indicate that we are flushing delayed refs for the commit */
142 	BTRFS_DELAYED_REFS_FLUSHING,
143 };
144 
145 struct btrfs_delayed_ref_root {
146 	/* head ref rbtree */
147 	struct rb_root_cached href_root;
148 
149 	/* dirty extent records */
150 	struct rb_root dirty_extent_root;
151 
152 	/* this spin lock protects the rbtree and the entries inside */
153 	spinlock_t lock;
154 
155 	/* how many delayed ref updates we've queued, used by the
156 	 * throttling code
157 	 */
158 	atomic_t num_entries;
159 
160 	/* total number of head nodes in tree */
161 	unsigned long num_heads;
162 
163 	/* total number of head nodes ready for processing */
164 	unsigned long num_heads_ready;
165 
166 	u64 pending_csums;
167 
168 	unsigned long flags;
169 
170 	u64 run_delayed_start;
171 
172 	/*
173 	 * To make qgroup to skip given root.
174 	 * This is for snapshot, as btrfs_qgroup_inherit() will manually
175 	 * modify counters for snapshot and its source, so we should skip
176 	 * the snapshot in new_root/old_roots or it will get calculated twice
177 	 */
178 	u64 qgroup_to_skip;
179 };
180 
181 enum btrfs_ref_type {
182 	BTRFS_REF_NOT_SET,
183 	BTRFS_REF_DATA,
184 	BTRFS_REF_METADATA,
185 	BTRFS_REF_LAST,
186 };
187 
188 struct btrfs_data_ref {
189 	/* For EXTENT_DATA_REF */
190 
191 	/* Original root this data extent belongs to */
192 	u64 owning_root;
193 
194 	/* Inode which refers to this data extent */
195 	u64 ino;
196 
197 	/*
198 	 * file_offset - extent_offset
199 	 *
200 	 * file_offset is the key.offset of the EXTENT_DATA key.
201 	 * extent_offset is btrfs_file_extent_offset() of the EXTENT_DATA data.
202 	 */
203 	u64 offset;
204 };
205 
206 struct btrfs_tree_ref {
207 	/*
208 	 * Level of this tree block
209 	 *
210 	 * Shared for skinny (TREE_BLOCK_REF) and normal tree ref.
211 	 */
212 	int level;
213 
214 	/*
215 	 * Root which owns this tree block.
216 	 *
217 	 * For TREE_BLOCK_REF (skinny metadata, either inline or keyed)
218 	 */
219 	u64 owning_root;
220 
221 	/* For non-skinny metadata, no special member needed */
222 };
223 
224 struct btrfs_ref {
225 	enum btrfs_ref_type type;
226 	int action;
227 
228 	/*
229 	 * Whether this extent should go through qgroup record.
230 	 *
231 	 * Normally false, but for certain cases like delayed subtree scan,
232 	 * setting this flag can hugely reduce qgroup overhead.
233 	 */
234 	bool skip_qgroup;
235 
236 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
237 	/* Through which root is this modification. */
238 	u64 real_root;
239 #endif
240 	u64 bytenr;
241 	u64 len;
242 
243 	/* Bytenr of the parent tree block */
244 	u64 parent;
245 	union {
246 		struct btrfs_data_ref data_ref;
247 		struct btrfs_tree_ref tree_ref;
248 	};
249 };
250 
251 extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
252 extern struct kmem_cache *btrfs_delayed_tree_ref_cachep;
253 extern struct kmem_cache *btrfs_delayed_data_ref_cachep;
254 extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
255 
256 int __init btrfs_delayed_ref_init(void);
257 void __cold btrfs_delayed_ref_exit(void);
258 
259 static inline u64 btrfs_calc_delayed_ref_bytes(const struct btrfs_fs_info *fs_info,
260 					       int num_delayed_refs)
261 {
262 	u64 num_bytes;
263 
264 	num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_delayed_refs);
265 
266 	/*
267 	 * We have to check the mount option here because we could be enabling
268 	 * the free space tree for the first time and don't have the compat_ro
269 	 * option set yet.
270 	 *
271 	 * We need extra reservations if we have the free space tree because
272 	 * we'll have to modify that tree as well.
273 	 */
274 	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE))
275 		num_bytes *= 2;
276 
277 	return num_bytes;
278 }
279 
280 static inline void btrfs_init_generic_ref(struct btrfs_ref *generic_ref,
281 				int action, u64 bytenr, u64 len, u64 parent)
282 {
283 	generic_ref->action = action;
284 	generic_ref->bytenr = bytenr;
285 	generic_ref->len = len;
286 	generic_ref->parent = parent;
287 }
288 
289 static inline void btrfs_init_tree_ref(struct btrfs_ref *generic_ref,
290 				int level, u64 root, u64 mod_root, bool skip_qgroup)
291 {
292 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
293 	/* If @real_root not set, use @root as fallback */
294 	generic_ref->real_root = mod_root ?: root;
295 #endif
296 	generic_ref->tree_ref.level = level;
297 	generic_ref->tree_ref.owning_root = root;
298 	generic_ref->type = BTRFS_REF_METADATA;
299 	if (skip_qgroup || !(is_fstree(root) &&
300 			     (!mod_root || is_fstree(mod_root))))
301 		generic_ref->skip_qgroup = true;
302 	else
303 		generic_ref->skip_qgroup = false;
304 
305 }
306 
307 static inline void btrfs_init_data_ref(struct btrfs_ref *generic_ref,
308 				u64 ref_root, u64 ino, u64 offset, u64 mod_root,
309 				bool skip_qgroup)
310 {
311 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
312 	/* If @real_root not set, use @root as fallback */
313 	generic_ref->real_root = mod_root ?: ref_root;
314 #endif
315 	generic_ref->data_ref.owning_root = ref_root;
316 	generic_ref->data_ref.ino = ino;
317 	generic_ref->data_ref.offset = offset;
318 	generic_ref->type = BTRFS_REF_DATA;
319 	if (skip_qgroup || !(is_fstree(ref_root) &&
320 			     (!mod_root || is_fstree(mod_root))))
321 		generic_ref->skip_qgroup = true;
322 	else
323 		generic_ref->skip_qgroup = false;
324 }
325 
326 static inline struct btrfs_delayed_extent_op *
327 btrfs_alloc_delayed_extent_op(void)
328 {
329 	return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
330 }
331 
332 static inline void
333 btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
334 {
335 	if (op)
336 		kmem_cache_free(btrfs_delayed_extent_op_cachep, op);
337 }
338 
339 static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
340 {
341 	WARN_ON(refcount_read(&ref->refs) == 0);
342 	if (refcount_dec_and_test(&ref->refs)) {
343 		WARN_ON(!RB_EMPTY_NODE(&ref->ref_node));
344 		switch (ref->type) {
345 		case BTRFS_TREE_BLOCK_REF_KEY:
346 		case BTRFS_SHARED_BLOCK_REF_KEY:
347 			kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
348 			break;
349 		case BTRFS_EXTENT_DATA_REF_KEY:
350 		case BTRFS_SHARED_DATA_REF_KEY:
351 			kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
352 			break;
353 		default:
354 			BUG();
355 		}
356 	}
357 }
358 
359 static inline u64 btrfs_ref_head_to_space_flags(
360 				struct btrfs_delayed_ref_head *head_ref)
361 {
362 	if (head_ref->is_data)
363 		return BTRFS_BLOCK_GROUP_DATA;
364 	else if (head_ref->is_system)
365 		return BTRFS_BLOCK_GROUP_SYSTEM;
366 	return BTRFS_BLOCK_GROUP_METADATA;
367 }
368 
369 static inline void btrfs_put_delayed_ref_head(struct btrfs_delayed_ref_head *head)
370 {
371 	if (refcount_dec_and_test(&head->refs))
372 		kmem_cache_free(btrfs_delayed_ref_head_cachep, head);
373 }
374 
375 int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
376 			       struct btrfs_ref *generic_ref,
377 			       struct btrfs_delayed_extent_op *extent_op);
378 int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
379 			       struct btrfs_ref *generic_ref,
380 			       u64 reserved);
381 int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
382 				u64 bytenr, u64 num_bytes,
383 				struct btrfs_delayed_extent_op *extent_op);
384 void btrfs_merge_delayed_refs(struct btrfs_fs_info *fs_info,
385 			      struct btrfs_delayed_ref_root *delayed_refs,
386 			      struct btrfs_delayed_ref_head *head);
387 
388 struct btrfs_delayed_ref_head *
389 btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
390 			    u64 bytenr);
391 int btrfs_delayed_ref_lock(struct btrfs_delayed_ref_root *delayed_refs,
392 			   struct btrfs_delayed_ref_head *head);
393 static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
394 {
395 	mutex_unlock(&head->mutex);
396 }
397 void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
398 			   struct btrfs_delayed_ref_head *head);
399 
400 struct btrfs_delayed_ref_head *btrfs_select_ref_head(
401 		struct btrfs_delayed_ref_root *delayed_refs);
402 
403 int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq);
404 
405 void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr);
406 void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans);
407 int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
408 				  enum btrfs_reserve_flush_enum flush);
409 void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
410 				       u64 num_bytes);
411 bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info);
412 
413 /*
414  * helper functions to cast a node into its container
415  */
416 static inline struct btrfs_delayed_tree_ref *
417 btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
418 {
419 	return container_of(node, struct btrfs_delayed_tree_ref, node);
420 }
421 
422 static inline struct btrfs_delayed_data_ref *
423 btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
424 {
425 	return container_of(node, struct btrfs_delayed_data_ref, node);
426 }
427 
428 #endif
429