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/*
 46 * Besides EXTENT_STATUS_REFERENCED, all these extent type masks
 47 * are exclusive, only one type can be set at a time.
 48 */
 49#define EXTENT_STATUS_WRITTEN	(1 << ES_WRITTEN_B)
 50#define EXTENT_STATUS_UNWRITTEN (1 << ES_UNWRITTEN_B)
 51#define EXTENT_STATUS_DELAYED	(1 << ES_DELAYED_B)
 52#define EXTENT_STATUS_HOLE	(1 << ES_HOLE_B)
 53#define EXTENT_STATUS_REFERENCED	(1 << ES_REFERENCED_B)
 54
 55#define ES_TYPE_MASK	((ext4_fsblk_t)(EXTENT_STATUS_WRITTEN | \
 56			  EXTENT_STATUS_UNWRITTEN | \
 57			  EXTENT_STATUS_DELAYED | \
 58			  EXTENT_STATUS_HOLE))
 59
 60#define ES_TYPE_VALID(type)	((type) && !((type) & ((type) - 1)))
 61
 62struct ext4_sb_info;
 63struct ext4_extent;
 64
 65struct extent_status {
 66	struct rb_node rb_node;
 67	ext4_lblk_t es_lblk;	/* first logical block extent covers */
 68	ext4_lblk_t es_len;	/* length of extent in block */
 69	ext4_fsblk_t es_pblk;	/* first physical block */
 70};
 71
 72struct ext4_es_tree {
 73	struct rb_root root;
 74	struct extent_status *cache_es;	/* recently accessed extent */
 75};
 76
 77struct ext4_es_stats {
 78	unsigned long es_stats_shrunk;
 79	struct percpu_counter es_stats_cache_hits;
 80	struct percpu_counter es_stats_cache_misses;
 81	u64 es_stats_scan_time;
 82	u64 es_stats_max_scan_time;
 83	struct percpu_counter es_stats_all_cnt;
 84	struct percpu_counter es_stats_shk_cnt;
 85};
 86
 87/*
 88 * Pending cluster reservations for bigalloc file systems
 89 *
 90 * A cluster with a pending reservation is a logical cluster shared by at
 91 * least one extent in the extents status tree with delayed and unwritten
 92 * status and at least one other written or unwritten extent.  The
 93 * reservation is said to be pending because a cluster reservation would
 94 * have to be taken in the event all blocks in the cluster shared with
 95 * written or unwritten extents were deleted while the delayed and
 96 * unwritten blocks remained.
 97 *
 98 * The set of pending cluster reservations is an auxiliary data structure
 99 * used with the extents status tree to implement reserved cluster/block
100 * accounting for bigalloc file systems.  The set is kept in memory and
101 * records all pending cluster reservations.
102 *
103 * Its primary function is to avoid the need to read extents from the
104 * disk when invalidating pages as a result of a truncate, punch hole, or
105 * collapse range operation.  Page invalidation requires a decrease in the
106 * reserved cluster count if it results in the removal of all delayed
107 * and unwritten extents (blocks) from a cluster that is not shared with a
108 * written or unwritten extent, and no decrease otherwise.  Determining
109 * whether the cluster is shared can be done by searching for a pending
110 * reservation on it.
111 *
112 * Secondarily, it provides a potentially faster method for determining
113 * whether the reserved cluster count should be increased when a physical
114 * cluster is deallocated as a result of a truncate, punch hole, or
115 * collapse range operation.  The necessary information is also present
116 * in the extents status tree, but might be more rapidly accessed in
117 * the pending reservation set in many cases due to smaller size.
118 *
119 * The pending cluster reservation set is implemented as a red-black tree
120 * with the goal of minimizing per page search time overhead.
121 */
122
123struct pending_reservation {
124	struct rb_node rb_node;
125	ext4_lblk_t lclu;
126};
127
128struct ext4_pending_tree {
129	struct rb_root root;
130};
131
132extern int __init ext4_init_es(void);
133extern void ext4_exit_es(void);
134extern void ext4_es_init_tree(struct ext4_es_tree *tree);
135
136extern void ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk,
137				  ext4_lblk_t len, ext4_fsblk_t pblk,
138				  unsigned int status,
139				  bool delalloc_reserve_used);
140extern void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk,
141				 ext4_lblk_t len, ext4_fsblk_t pblk,
142				 unsigned int status);
143extern void ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
144				  ext4_lblk_t len);
145extern void ext4_es_find_extent_range(struct inode *inode,
146				      int (*match_fn)(struct extent_status *es),
147				      ext4_lblk_t lblk, ext4_lblk_t end,
148				      struct extent_status *es);
149extern int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk,
150				 ext4_lblk_t *next_lblk,
151				 struct extent_status *es);
152extern bool ext4_es_scan_range(struct inode *inode,
153			       int (*matching_fn)(struct extent_status *es),
154			       ext4_lblk_t lblk, ext4_lblk_t end);
155extern bool ext4_es_scan_clu(struct inode *inode,
156			     int (*matching_fn)(struct extent_status *es),
157			     ext4_lblk_t lblk);
158
159static inline unsigned int ext4_es_status(struct extent_status *es)
160{
161	return es->es_pblk >> ES_SHIFT;
162}
163
164static inline unsigned int ext4_es_type(struct extent_status *es)
165{
166	return (es->es_pblk >> ES_SHIFT) & ES_TYPE_MASK;
167}
168
169static inline int ext4_es_is_written(struct extent_status *es)
170{
171	return (ext4_es_type(es) & EXTENT_STATUS_WRITTEN) != 0;
172}
173
174static inline int ext4_es_is_unwritten(struct extent_status *es)
175{
176	return (ext4_es_type(es) & EXTENT_STATUS_UNWRITTEN) != 0;
177}
178
179static inline int ext4_es_is_delayed(struct extent_status *es)
180{
181	return (ext4_es_type(es) & EXTENT_STATUS_DELAYED) != 0;
182}
183
184static inline int ext4_es_is_hole(struct extent_status *es)
185{
186	return (ext4_es_type(es) & EXTENT_STATUS_HOLE) != 0;
187}
188
189static inline int ext4_es_is_mapped(struct extent_status *es)
190{
191	return (ext4_es_is_written(es) || ext4_es_is_unwritten(es));
192}
193
194static inline void ext4_es_set_referenced(struct extent_status *es)
195{
196	es->es_pblk |= ((ext4_fsblk_t)EXTENT_STATUS_REFERENCED) << ES_SHIFT;
197}
198
199static inline void ext4_es_clear_referenced(struct extent_status *es)
200{
201	es->es_pblk &= ~(((ext4_fsblk_t)EXTENT_STATUS_REFERENCED) << ES_SHIFT);
202}
203
204static inline int ext4_es_is_referenced(struct extent_status *es)
205{
206	return (ext4_es_status(es) & EXTENT_STATUS_REFERENCED) != 0;
207}
208
209static inline ext4_fsblk_t ext4_es_pblock(struct extent_status *es)
210{
211	return es->es_pblk & ~ES_MASK;
212}
213
214static inline ext4_fsblk_t ext4_es_show_pblock(struct extent_status *es)
215{
216	ext4_fsblk_t pblock = ext4_es_pblock(es);
217	return pblock == ~ES_MASK ? 0 : pblock;
218}
219
220static inline void ext4_es_store_pblock(struct extent_status *es,
221					ext4_fsblk_t pb)
222{
223	ext4_fsblk_t block;
224
225	block = (pb & ~ES_MASK) | (es->es_pblk & ES_MASK);
226	es->es_pblk = block;
227}
228
229static inline void ext4_es_store_pblock_status(struct extent_status *es,
230					       ext4_fsblk_t pb,
231					       unsigned int status)
232{
233	WARN_ON_ONCE(!ES_TYPE_VALID(status & ES_TYPE_MASK));
234
235	es->es_pblk = (((ext4_fsblk_t)status << ES_SHIFT) & ES_MASK) |
236		      (pb & ~ES_MASK);
237}
238
239extern int ext4_es_register_shrinker(struct ext4_sb_info *sbi);
240extern void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi);
241
242extern int ext4_seq_es_shrinker_info_show(struct seq_file *seq, void *v);
243
244extern int __init ext4_init_pending(void);
245extern void ext4_exit_pending(void);
246extern void ext4_init_pending_tree(struct ext4_pending_tree *tree);
247extern void ext4_remove_pending(struct inode *inode, ext4_lblk_t lblk);
248extern bool ext4_is_pending(struct inode *inode, ext4_lblk_t lblk);
249extern void ext4_es_insert_delayed_extent(struct inode *inode, ext4_lblk_t lblk,
250					  ext4_lblk_t len, bool lclu_allocated,
251					  bool end_allocated);
252extern void ext4_clear_inode_es(struct inode *inode);
253
254#endif /* _EXT4_EXTENTS_STATUS_H */