1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 *  fs/ext4/extents_status.h
  4 *
  5 * Written by Yongqiang Yang <xiaoqiangnk@gmail.com>
  6 * Modified by
  7 *	Allison Henderson <achender@linux.vnet.ibm.com>
  8 *	Zheng Liu <wenqing.lz@taobao.com>
  9 *
 10 */
 11
 12#ifndef _EXT4_EXTENTS_STATUS_H
 13#define _EXT4_EXTENTS_STATUS_H
 14
 15/*
 16 * Turn on ES_DEBUG__ to get lots of info about extent status operations.
 17 */
 18#ifdef ES_DEBUG__
 19#define es_debug(fmt, ...)	printk(fmt, ##__VA_ARGS__)
 20#else
 21#define es_debug(fmt, ...)	no_printk(fmt, ##__VA_ARGS__)
 22#endif
 23
 24/*
 25 * With ES_AGGRESSIVE_TEST defined, the result of es caching will be
 26 * checked with old map_block's result.
 27 */
 28#define ES_AGGRESSIVE_TEST__
 29
 30/*
 31 * These flags live in the high bits of extent_status.es_pblk
 32 */
 33enum {
 34	ES_WRITTEN_B,
 35	ES_UNWRITTEN_B,
 36	ES_DELAYED_B,
 37	ES_HOLE_B,
 38	ES_REFERENCED_B,
 39	ES_FLAGS
 40};
 41
 42#define ES_SHIFT (sizeof(ext4_fsblk_t)*8 - ES_FLAGS)
 43#define ES_MASK (~((ext4_fsblk_t)0) << ES_SHIFT)
 44
 45#define EXTENT_STATUS_WRITTEN	(1 << ES_WRITTEN_B)
 46#define EXTENT_STATUS_UNWRITTEN (1 << ES_UNWRITTEN_B)
 47#define EXTENT_STATUS_DELAYED	(1 << ES_DELAYED_B)
 48#define EXTENT_STATUS_HOLE	(1 << ES_HOLE_B)
 49#define EXTENT_STATUS_REFERENCED	(1 << ES_REFERENCED_B)
 50
 51#define ES_TYPE_MASK	((ext4_fsblk_t)(EXTENT_STATUS_WRITTEN | \
 52			  EXTENT_STATUS_UNWRITTEN | \
 53			  EXTENT_STATUS_DELAYED | \
 54			  EXTENT_STATUS_HOLE) << ES_SHIFT)
 55
 56struct ext4_sb_info;
 57struct ext4_extent;
 58
 59struct extent_status {
 60	struct rb_node rb_node;
 61	ext4_lblk_t es_lblk;	/* first logical block extent covers */
 62	ext4_lblk_t es_len;	/* length of extent in block */
 63	ext4_fsblk_t es_pblk;	/* first physical block */
 64};
 65
 66struct ext4_es_tree {
 67	struct rb_root root;
 68	struct extent_status *cache_es;	/* recently accessed extent */
 69};
 70
 71struct ext4_es_stats {
 72	unsigned long es_stats_shrunk;
 73	struct percpu_counter es_stats_cache_hits;
 74	struct percpu_counter es_stats_cache_misses;
 75	u64 es_stats_scan_time;
 76	u64 es_stats_max_scan_time;
 77	struct percpu_counter es_stats_all_cnt;
 78	struct percpu_counter es_stats_shk_cnt;
 79};
 80
 81/*
 82 * Pending cluster reservations for bigalloc file systems
 83 *
 84 * A cluster with a pending reservation is a logical cluster shared by at
 85 * least one extent in the extents status tree with delayed and unwritten
 86 * status and at least one other written or unwritten extent.  The
 87 * reservation is said to be pending because a cluster reservation would
 88 * have to be taken in the event all blocks in the cluster shared with
 89 * written or unwritten extents were deleted while the delayed and
 90 * unwritten blocks remained.
 91 *
 92 * The set of pending cluster reservations is an auxiliary data structure
 93 * used with the extents status tree to implement reserved cluster/block
 94 * accounting for bigalloc file systems.  The set is kept in memory and
 95 * records all pending cluster reservations.
 96 *
 97 * Its primary function is to avoid the need to read extents from the
 98 * disk when invalidating pages as a result of a truncate, punch hole, or
 99 * collapse range operation.  Page invalidation requires a decrease in the
100 * reserved cluster count if it results in the removal of all delayed
101 * and unwritten extents (blocks) from a cluster that is not shared with a
102 * written or unwritten extent, and no decrease otherwise.  Determining
103 * whether the cluster is shared can be done by searching for a pending
104 * reservation on it.
105 *
106 * Secondarily, it provides a potentially faster method for determining
107 * whether the reserved cluster count should be increased when a physical
108 * cluster is deallocated as a result of a truncate, punch hole, or
109 * collapse range operation.  The necessary information is also present
110 * in the extents status tree, but might be more rapidly accessed in
111 * the pending reservation set in many cases due to smaller size.
112 *
113 * The pending cluster reservation set is implemented as a red-black tree
114 * with the goal of minimizing per page search time overhead.
115 */
116
117struct pending_reservation {
118	struct rb_node rb_node;
119	ext4_lblk_t lclu;
120};
121
122struct ext4_pending_tree {
123	struct rb_root root;
124};
125
126extern int __init ext4_init_es(void);
127extern void ext4_exit_es(void);
128extern void ext4_es_init_tree(struct ext4_es_tree *tree);
129
130extern int ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk,
131				 ext4_lblk_t len, ext4_fsblk_t pblk,
132				 unsigned int status);
133extern void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk,
134				 ext4_lblk_t len, ext4_fsblk_t pblk,
135				 unsigned int status);
136extern int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
137				 ext4_lblk_t len);
138extern void ext4_es_find_extent_range(struct inode *inode,
139				      int (*match_fn)(struct extent_status *es),
140				      ext4_lblk_t lblk, ext4_lblk_t end,
141				      struct extent_status *es);
142extern int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk,
143				 ext4_lblk_t *next_lblk,
144				 struct extent_status *es);
145extern bool ext4_es_scan_range(struct inode *inode,
146			       int (*matching_fn)(struct extent_status *es),
147			       ext4_lblk_t lblk, ext4_lblk_t end);
148extern bool ext4_es_scan_clu(struct inode *inode,
149			     int (*matching_fn)(struct extent_status *es),
150			     ext4_lblk_t lblk);
151
152static inline unsigned int ext4_es_status(struct extent_status *es)
153{
154	return es->es_pblk >> ES_SHIFT;
155}
156
157static inline unsigned int ext4_es_type(struct extent_status *es)
158{
159	return (es->es_pblk & ES_TYPE_MASK) >> ES_SHIFT;
160}
161
162static inline int ext4_es_is_written(struct extent_status *es)
163{
164	return (ext4_es_type(es) & EXTENT_STATUS_WRITTEN) != 0;
165}
166
167static inline int ext4_es_is_unwritten(struct extent_status *es)
168{
169	return (ext4_es_type(es) & EXTENT_STATUS_UNWRITTEN) != 0;
170}
171
172static inline int ext4_es_is_delayed(struct extent_status *es)
173{
174	return (ext4_es_type(es) & EXTENT_STATUS_DELAYED) != 0;
175}
176
177static inline int ext4_es_is_hole(struct extent_status *es)
178{
179	return (ext4_es_type(es) & EXTENT_STATUS_HOLE) != 0;
180}
181
182static inline int ext4_es_is_mapped(struct extent_status *es)
183{
184	return (ext4_es_is_written(es) || ext4_es_is_unwritten(es));
185}
186
187static inline int ext4_es_is_delonly(struct extent_status *es)
188{
189	return (ext4_es_is_delayed(es) && !ext4_es_is_unwritten(es));
190}
191
192static inline void ext4_es_set_referenced(struct extent_status *es)
193{
194	es->es_pblk |= ((ext4_fsblk_t)EXTENT_STATUS_REFERENCED) << ES_SHIFT;
195}
196
197static inline void ext4_es_clear_referenced(struct extent_status *es)
198{
199	es->es_pblk &= ~(((ext4_fsblk_t)EXTENT_STATUS_REFERENCED) << ES_SHIFT);
200}
201
202static inline int ext4_es_is_referenced(struct extent_status *es)
203{
204	return (ext4_es_status(es) & EXTENT_STATUS_REFERENCED) != 0;
205}
206
207static inline ext4_fsblk_t ext4_es_pblock(struct extent_status *es)
208{
209	return es->es_pblk & ~ES_MASK;
210}
211
212static inline ext4_fsblk_t ext4_es_show_pblock(struct extent_status *es)
213{
214	ext4_fsblk_t pblock = ext4_es_pblock(es);
215	return pblock == ~ES_MASK ? 0 : pblock;
216}
217
218static inline void ext4_es_store_pblock(struct extent_status *es,
219					ext4_fsblk_t pb)
220{
221	ext4_fsblk_t block;
222
223	block = (pb & ~ES_MASK) | (es->es_pblk & ES_MASK);
224	es->es_pblk = block;
225}
226
227static inline void ext4_es_store_status(struct extent_status *es,
228					unsigned int status)
229{
230	es->es_pblk = (((ext4_fsblk_t)status << ES_SHIFT) & ES_MASK) |
231		      (es->es_pblk & ~ES_MASK);
232}
233
234static inline void ext4_es_store_pblock_status(struct extent_status *es,
235					       ext4_fsblk_t pb,
236					       unsigned int status)
237{
238	es->es_pblk = (((ext4_fsblk_t)status << ES_SHIFT) & ES_MASK) |
239		      (pb & ~ES_MASK);
240}
241
242extern int ext4_es_register_shrinker(struct ext4_sb_info *sbi);
243extern void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi);
244
245extern int ext4_seq_es_shrinker_info_show(struct seq_file *seq, void *v);
246
247extern int __init ext4_init_pending(void);
248extern void ext4_exit_pending(void);
249extern void ext4_init_pending_tree(struct ext4_pending_tree *tree);
250extern void ext4_remove_pending(struct inode *inode, ext4_lblk_t lblk);
251extern bool ext4_is_pending(struct inode *inode, ext4_lblk_t lblk);
252extern int ext4_es_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk,
253					bool allocated);
254extern unsigned int ext4_es_delayed_clu(struct inode *inode, ext4_lblk_t lblk,
255					ext4_lblk_t len);
256extern void ext4_clear_inode_es(struct inode *inode);
257
258#endif /* _EXT4_EXTENTS_STATUS_H */