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/types.h>
10#include <linux/refcount.h>
11#include <linux/list.h>
12#include <linux/rbtree.h>
13#include <linux/mutex.h>
14#include <linux/spinlock.h>
15#include <linux/slab.h>
16#include <uapi/linux/btrfs_tree.h>
17#include "fs.h"
18#include "messages.h"
19
20struct btrfs_trans_handle;
21struct btrfs_fs_info;
22
23/* these are the possible values of struct btrfs_delayed_ref_node->action */
24enum btrfs_delayed_ref_action {
25 /* Add one backref to the tree */
26 BTRFS_ADD_DELAYED_REF = 1,
27 /* Delete one backref from the tree */
28 BTRFS_DROP_DELAYED_REF,
29 /* Record a full extent allocation */
30 BTRFS_ADD_DELAYED_EXTENT,
31 /* Not changing ref count on head ref */
32 BTRFS_UPDATE_DELAYED_HEAD,
33} __packed;
34
35struct btrfs_data_ref {
36 /* For EXTENT_DATA_REF */
37
38 /* Inode which refers to this data extent */
39 u64 objectid;
40
41 /*
42 * file_offset - extent_offset
43 *
44 * file_offset is the key.offset of the EXTENT_DATA key.
45 * extent_offset is btrfs_file_extent_offset() of the EXTENT_DATA data.
46 */
47 u64 offset;
48};
49
50struct btrfs_tree_ref {
51 /*
52 * Level of this tree block.
53 *
54 * Shared for skinny (TREE_BLOCK_REF) and normal tree ref.
55 */
56 int level;
57
58 /* For non-skinny metadata, no special member needed */
59};
60
61struct btrfs_delayed_ref_node {
62 struct rb_node ref_node;
63 /*
64 * If action is BTRFS_ADD_DELAYED_REF, also link this node to
65 * ref_head->ref_add_list, then we do not need to iterate the
66 * refs rbtree in the corresponding delayed ref head
67 * (struct btrfs_delayed_ref_head::ref_tree).
68 */
69 struct list_head add_list;
70
71 /* the starting bytenr of the extent */
72 u64 bytenr;
73
74 /* the size of the extent */
75 u64 num_bytes;
76
77 /* seq number to keep track of insertion order */
78 u64 seq;
79
80 /* The ref_root for this ref */
81 u64 ref_root;
82
83 /*
84 * The parent for this ref, if this isn't set the ref_root is the
85 * reference owner.
86 */
87 u64 parent;
88
89 /* ref count on this data structure */
90 refcount_t refs;
91
92 /*
93 * how many refs is this entry adding or deleting. For
94 * head refs, this may be a negative number because it is keeping
95 * track of the total mods done to the reference count.
96 * For individual refs, this will always be a positive number
97 *
98 * It may be more than one, since it is possible for a single
99 * parent to have more than one ref on an extent
100 */
101 int ref_mod;
102
103 unsigned int action:8;
104 unsigned int type:8;
105
106 union {
107 struct btrfs_tree_ref tree_ref;
108 struct btrfs_data_ref data_ref;
109 };
110};
111
112struct btrfs_delayed_extent_op {
113 struct btrfs_disk_key key;
114 bool update_key;
115 bool update_flags;
116 u64 flags_to_set;
117};
118
119/*
120 * the head refs are used to hold a lock on a given extent, which allows us
121 * to make sure that only one process is running the delayed refs
122 * at a time for a single extent. They also store the sum of all the
123 * reference count modifications we've queued up.
124 */
125struct btrfs_delayed_ref_head {
126 u64 bytenr;
127 u64 num_bytes;
128 /*
129 * the mutex is held while running the refs, and it is also
130 * held when checking the sum of reference modifications.
131 */
132 struct mutex mutex;
133
134 refcount_t refs;
135
136 /* Protects 'ref_tree' and 'ref_add_list'. */
137 spinlock_t lock;
138 struct rb_root_cached ref_tree;
139 /* accumulate add BTRFS_ADD_DELAYED_REF nodes to this ref_add_list. */
140 struct list_head ref_add_list;
141
142 struct btrfs_delayed_extent_op *extent_op;
143
144 /*
145 * This is used to track the final ref_mod from all the refs associated
146 * with this head ref, this is not adjusted as delayed refs are run,
147 * this is meant to track if we need to do the csum accounting or not.
148 */
149 int total_ref_mod;
150
151 /*
152 * This is the current outstanding mod references for this bytenr. This
153 * is used with lookup_extent_info to get an accurate reference count
154 * for a bytenr, so it is adjusted as delayed refs are run so that any
155 * on disk reference count + ref_mod is accurate.
156 */
157 int ref_mod;
158
159 /*
160 * The root that triggered the allocation when must_insert_reserved is
161 * set to true.
162 */
163 u64 owning_root;
164
165 /*
166 * Track reserved bytes when setting must_insert_reserved. On success
167 * or cleanup, we will need to free the reservation.
168 */
169 u64 reserved_bytes;
170
171 /* Tree block level, for metadata only. */
172 u8 level;
173
174 /*
175 * when a new extent is allocated, it is just reserved in memory
176 * The actual extent isn't inserted into the extent allocation tree
177 * until the delayed ref is processed. must_insert_reserved is
178 * used to flag a delayed ref so the accounting can be updated
179 * when a full insert is done.
180 *
181 * It is possible the extent will be freed before it is ever
182 * inserted into the extent allocation tree. In this case
183 * we need to update the in ram accounting to properly reflect
184 * the free has happened.
185 */
186 bool must_insert_reserved;
187
188 bool is_data;
189 bool is_system;
190 bool processing;
191 /*
192 * Indicate if it's currently in the data structure that tracks head
193 * refs (struct btrfs_delayed_ref_root::head_refs).
194 */
195 bool tracked;
196};
197
198enum btrfs_delayed_ref_flags {
199 /* Indicate that we are flushing delayed refs for the commit */
200 BTRFS_DELAYED_REFS_FLUSHING,
201};
202
203struct btrfs_delayed_ref_root {
204 /*
205 * Track head references.
206 * The keys correspond to the logical address of the extent ("bytenr")
207 * right shifted by fs_info->sectorsize_bits. This is both to get a more
208 * dense index space (optimizes xarray structure) and because indexes in
209 * xarrays are of "unsigned long" type, meaning they are 32 bits wide on
210 * 32 bits platforms, limiting the extent range to 4G which is too low
211 * and makes it unusable (truncated index values) on 32 bits platforms.
212 * Protected by the spinlock 'lock' defined below.
213 */
214 struct xarray head_refs;
215
216 /*
217 * Track dirty extent records.
218 * The keys correspond to the logical address of the extent ("bytenr")
219 * right shifted by fs_info->sectorsize_bits, for same reasons as above.
220 */
221 struct xarray dirty_extents;
222
223 /*
224 * Protects the xarray head_refs, its entries and the following fields:
225 * num_heads, num_heads_ready, pending_csums and run_delayed_start.
226 */
227 spinlock_t lock;
228
229 /* Total number of head refs, protected by the spinlock 'lock'. */
230 unsigned long num_heads;
231
232 /*
233 * Total number of head refs ready for processing, protected by the
234 * spinlock 'lock'.
235 */
236 unsigned long num_heads_ready;
237
238 /*
239 * Track space reserved for deleting csums of data extents.
240 * Protected by the spinlock 'lock'.
241 */
242 u64 pending_csums;
243
244 unsigned long flags;
245
246 /*
247 * Track from which bytenr to start searching ref heads.
248 * Protected by the spinlock 'lock'.
249 */
250 u64 run_delayed_start;
251
252 /*
253 * To make qgroup to skip given root.
254 * This is for snapshot, as btrfs_qgroup_inherit() will manually
255 * modify counters for snapshot and its source, so we should skip
256 * the snapshot in new_root/old_roots or it will get calculated twice
257 */
258 u64 qgroup_to_skip;
259};
260
261enum btrfs_ref_type {
262 BTRFS_REF_NOT_SET,
263 BTRFS_REF_DATA,
264 BTRFS_REF_METADATA,
265} __packed;
266
267struct btrfs_ref {
268 enum btrfs_ref_type type;
269 enum btrfs_delayed_ref_action action;
270
271 /*
272 * Whether this extent should go through qgroup record.
273 *
274 * Normally false, but for certain cases like delayed subtree scan,
275 * setting this flag can hugely reduce qgroup overhead.
276 */
277 bool skip_qgroup;
278
279 u64 bytenr;
280 u64 num_bytes;
281 u64 owning_root;
282
283 /*
284 * The root that owns the reference for this reference, this will be set
285 * or ->parent will be set, depending on what type of reference this is.
286 */
287 u64 ref_root;
288
289 /* Bytenr of the parent tree block */
290 u64 parent;
291 union {
292 struct btrfs_data_ref data_ref;
293 struct btrfs_tree_ref tree_ref;
294 };
295
296#ifdef CONFIG_BTRFS_DEBUG
297 /* Through which root is this modification. */
298 u64 real_root;
299#endif
300};
301
302extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
303extern struct kmem_cache *btrfs_delayed_ref_node_cachep;
304extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
305
306int __init btrfs_delayed_ref_init(void);
307void __cold btrfs_delayed_ref_exit(void);
308
309static inline u64 btrfs_calc_delayed_ref_bytes(const struct btrfs_fs_info *fs_info,
310 int num_delayed_refs)
311{
312 u64 num_bytes;
313
314 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items: num_delayed_refs);
315
316 /*
317 * We have to check the mount option here because we could be enabling
318 * the free space tree for the first time and don't have the compat_ro
319 * option set yet.
320 *
321 * We need extra reservations if we have the free space tree because
322 * we'll have to modify that tree as well.
323 */
324 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE))
325 num_bytes *= 2;
326
327 return num_bytes;
328}
329
330static inline u64 btrfs_calc_delayed_ref_csum_bytes(const struct btrfs_fs_info *fs_info,
331 int num_csum_items)
332{
333 /*
334 * Deleting csum items does not result in new nodes/leaves and does not
335 * require changing the free space tree, only the csum tree, so this is
336 * all we need.
337 */
338 return btrfs_calc_metadata_size(fs_info, num_items: num_csum_items);
339}
340
341void btrfs_init_tree_ref(struct btrfs_ref *generic_ref, int level, u64 mod_root,
342 bool skip_qgroup);
343void btrfs_init_data_ref(struct btrfs_ref *generic_ref, u64 ino, u64 offset,
344 u64 mod_root, bool skip_qgroup);
345
346static inline struct btrfs_delayed_extent_op *
347btrfs_alloc_delayed_extent_op(void)
348{
349 return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
350}
351
352static inline void
353btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
354{
355 if (op)
356 kmem_cache_free(s: btrfs_delayed_extent_op_cachep, objp: op);
357}
358
359void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref);
360
361static inline u64 btrfs_ref_head_to_space_flags(
362 struct btrfs_delayed_ref_head *head_ref)
363{
364 if (head_ref->is_data)
365 return BTRFS_BLOCK_GROUP_DATA;
366 else if (head_ref->is_system)
367 return BTRFS_BLOCK_GROUP_SYSTEM;
368 return BTRFS_BLOCK_GROUP_METADATA;
369}
370
371static inline void btrfs_put_delayed_ref_head(struct btrfs_delayed_ref_head *head)
372{
373 if (refcount_dec_and_test(r: &head->refs))
374 kmem_cache_free(s: btrfs_delayed_ref_head_cachep, objp: head);
375}
376
377int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
378 struct btrfs_ref *generic_ref,
379 struct btrfs_delayed_extent_op *extent_op);
380int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
381 struct btrfs_ref *generic_ref,
382 u64 reserved);
383int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
384 u64 bytenr, u64 num_bytes, u8 level,
385 struct btrfs_delayed_extent_op *extent_op);
386void btrfs_merge_delayed_refs(struct btrfs_fs_info *fs_info,
387 struct btrfs_delayed_ref_root *delayed_refs,
388 struct btrfs_delayed_ref_head *head);
389
390struct btrfs_delayed_ref_head *
391btrfs_find_delayed_ref_head(const struct btrfs_fs_info *fs_info,
392 struct btrfs_delayed_ref_root *delayed_refs,
393 u64 bytenr);
394static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
395{
396 mutex_unlock(lock: &head->mutex);
397}
398void btrfs_delete_ref_head(const struct btrfs_fs_info *fs_info,
399 struct btrfs_delayed_ref_root *delayed_refs,
400 struct btrfs_delayed_ref_head *head);
401
402struct btrfs_delayed_ref_head *btrfs_select_ref_head(
403 const struct btrfs_fs_info *fs_info,
404 struct btrfs_delayed_ref_root *delayed_refs);
405void btrfs_unselect_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
406 struct btrfs_delayed_ref_head *head);
407struct btrfs_delayed_ref_node *btrfs_select_delayed_ref(struct btrfs_delayed_ref_head *head);
408
409int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq);
410
411void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr_refs, int nr_csums);
412void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans);
413void btrfs_inc_delayed_refs_rsv_bg_inserts(struct btrfs_fs_info *fs_info);
414void btrfs_dec_delayed_refs_rsv_bg_inserts(struct btrfs_fs_info *fs_info);
415void btrfs_inc_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
416void btrfs_dec_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
417int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
418 enum btrfs_reserve_flush_enum flush);
419bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info);
420bool btrfs_find_delayed_tree_ref(struct btrfs_delayed_ref_head *head,
421 u64 root, u64 parent);
422void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans);
423
424static inline u64 btrfs_delayed_ref_owner(const struct btrfs_delayed_ref_node *node)
425{
426 if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
427 node->type == BTRFS_SHARED_DATA_REF_KEY)
428 return node->data_ref.objectid;
429 return node->tree_ref.level;
430}
431
432static inline u64 btrfs_delayed_ref_offset(const struct btrfs_delayed_ref_node *node)
433{
434 if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
435 node->type == BTRFS_SHARED_DATA_REF_KEY)
436 return node->data_ref.offset;
437 return 0;
438}
439
440static inline u8 btrfs_ref_type(const struct btrfs_ref *ref)
441{
442 ASSERT(ref->type == BTRFS_REF_DATA || ref->type == BTRFS_REF_METADATA);
443
444 if (ref->type == BTRFS_REF_DATA) {
445 if (ref->parent)
446 return BTRFS_SHARED_DATA_REF_KEY;
447 else
448 return BTRFS_EXTENT_DATA_REF_KEY;
449 } else {
450 if (ref->parent)
451 return BTRFS_SHARED_BLOCK_REF_KEY;
452 else
453 return BTRFS_TREE_BLOCK_REF_KEY;
454 }
455
456 return 0;
457}
458
459#endif
460

source code of linux/fs/btrfs/delayed-ref.h