fs/ext4/extents_status.h

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  *  fs/ext4/extents_status.h
0004  *
0005  * Written by Yongqiang Yang <xiaoqiangnk@gmail.com>
0006  * Modified by
0007  *  Allison Henderson <achender@linux.vnet.ibm.com>
0008  *  Zheng Liu <wenqing.lz@taobao.com>
0009  *
0010  */
0011
0012 #ifndef _EXT4_EXTENTS_STATUS_H
0013 #define _EXT4_EXTENTS_STATUS_H
0014
0015 /*
0016  * Turn on ES_DEBUG__ to get lots of info about extent status operations.
0017  */
0018 #ifdef ES_DEBUG__
0019 #define es_debug(fmt, ...)  printk(fmt, ##__VA_ARGS__)
0020 #else
0021 #define es_debug(fmt, ...)  no_printk(fmt, ##__VA_ARGS__)
0022 #endif
0023
0024 /*
0025  * With ES_AGGRESSIVE_TEST defined, the result of es caching will be
0026  * checked with old map_block's result.
0027  */
0028 #define ES_AGGRESSIVE_TEST__
0029
0030 /*
0031  * These flags live in the high bits of extent_status.es_pblk
0032  */
0033 enum {
0034     ES_WRITTEN_B,
0035     ES_UNWRITTEN_B,
0036     ES_DELAYED_B,
0037     ES_HOLE_B,
0038     ES_REFERENCED_B,
0039     ES_FLAGS
0040 };
0041
0042 #define ES_SHIFT (sizeof(ext4_fsblk_t)*8 - ES_FLAGS)
0043 #define ES_MASK (~((ext4_fsblk_t)0) << ES_SHIFT)
0044
0045 #define EXTENT_STATUS_WRITTEN   (1 << ES_WRITTEN_B)
0046 #define EXTENT_STATUS_UNWRITTEN (1 << ES_UNWRITTEN_B)
0047 #define EXTENT_STATUS_DELAYED   (1 << ES_DELAYED_B)
0048 #define EXTENT_STATUS_HOLE  (1 << ES_HOLE_B)
0049 #define EXTENT_STATUS_REFERENCED    (1 << ES_REFERENCED_B)
0050
0051 #define ES_TYPE_MASK    ((ext4_fsblk_t)(EXTENT_STATUS_WRITTEN | \
0052               EXTENT_STATUS_UNWRITTEN | \
0053               EXTENT_STATUS_DELAYED | \
0054               EXTENT_STATUS_HOLE) << ES_SHIFT)
0055
0056 struct ext4_sb_info;
0057 struct ext4_extent;
0058
0059 struct extent_status {
0060     struct rb_node rb_node;
0061     ext4_lblk_t es_lblk;    /* first logical block extent covers */
0062     ext4_lblk_t es_len; /* length of extent in block */
0063     ext4_fsblk_t es_pblk;   /* first physical block */
0064 };
0065
0066 struct ext4_es_tree {
0067     struct rb_root root;
0068     struct extent_status *cache_es; /* recently accessed extent */
0069 };
0070
0071 struct ext4_es_stats {
0072     unsigned long es_stats_shrunk;
0073     struct percpu_counter es_stats_cache_hits;
0074     struct percpu_counter es_stats_cache_misses;
0075     u64 es_stats_scan_time;
0076     u64 es_stats_max_scan_time;
0077     struct percpu_counter es_stats_all_cnt;
0078     struct percpu_counter es_stats_shk_cnt;
0079 };
0080
0081 /*
0082  * Pending cluster reservations for bigalloc file systems
0083  *
0084  * A cluster with a pending reservation is a logical cluster shared by at
0085  * least one extent in the extents status tree with delayed and unwritten
0086  * status and at least one other written or unwritten extent.  The
0087  * reservation is said to be pending because a cluster reservation would
0088  * have to be taken in the event all blocks in the cluster shared with
0089  * written or unwritten extents were deleted while the delayed and
0090  * unwritten blocks remained.
0091  *
0092  * The set of pending cluster reservations is an auxiliary data structure
0093  * used with the extents status tree to implement reserved cluster/block
0094  * accounting for bigalloc file systems.  The set is kept in memory and
0095  * records all pending cluster reservations.
0096  *
0097  * Its primary function is to avoid the need to read extents from the
0098  * disk when invalidating pages as a result of a truncate, punch hole, or
0099  * collapse range operation.  Page invalidation requires a decrease in the
0100  * reserved cluster count if it results in the removal of all delayed
0101  * and unwritten extents (blocks) from a cluster that is not shared with a
0102  * written or unwritten extent, and no decrease otherwise.  Determining
0103  * whether the cluster is shared can be done by searching for a pending
0104  * reservation on it.
0105  *
0106  * Secondarily, it provides a potentially faster method for determining
0107  * whether the reserved cluster count should be increased when a physical
0108  * cluster is deallocated as a result of a truncate, punch hole, or
0109  * collapse range operation.  The necessary information is also present
0110  * in the extents status tree, but might be more rapidly accessed in
0111  * the pending reservation set in many cases due to smaller size.
0112  *
0113  * The pending cluster reservation set is implemented as a red-black tree
0114  * with the goal of minimizing per page search time overhead.
0115  */
0116
0117 struct pending_reservation {
0118     struct rb_node rb_node;
0119     ext4_lblk_t lclu;
0120 };
0121
0122 struct ext4_pending_tree {
0123     struct rb_root root;
0124 };
0125
0126 extern int __init ext4_init_es(void);
0127 extern void ext4_exit_es(void);
0128 extern void ext4_es_init_tree(struct ext4_es_tree *tree);
0129
0130 extern int ext4_es_insert_extent(struct inode *inode, ext4_lblk_t lblk,
0131                  ext4_lblk_t len, ext4_fsblk_t pblk,
0132                  unsigned int status);
0133 extern void ext4_es_cache_extent(struct inode *inode, ext4_lblk_t lblk,
0134                  ext4_lblk_t len, ext4_fsblk_t pblk,
0135                  unsigned int status);
0136 extern int ext4_es_remove_extent(struct inode *inode, ext4_lblk_t lblk,
0137                  ext4_lblk_t len);
0138 extern void ext4_es_find_extent_range(struct inode *inode,
0139                       int (*match_fn)(struct extent_status *es),
0140                       ext4_lblk_t lblk, ext4_lblk_t end,
0141                       struct extent_status *es);
0142 extern int ext4_es_lookup_extent(struct inode *inode, ext4_lblk_t lblk,
0143                  ext4_lblk_t *next_lblk,
0144                  struct extent_status *es);
0145 extern bool ext4_es_scan_range(struct inode *inode,
0146                    int (*matching_fn)(struct extent_status *es),
0147                    ext4_lblk_t lblk, ext4_lblk_t end);
0148 extern bool ext4_es_scan_clu(struct inode *inode,
0149                  int (*matching_fn)(struct extent_status *es),
0150                  ext4_lblk_t lblk);
0151
0152 static inline unsigned int ext4_es_status(struct extent_status *es)
0153 {
0154     return es->es_pblk >> ES_SHIFT;
0155 }
0156
0157 static inline unsigned int ext4_es_type(struct extent_status *es)
0158 {
0159     return (es->es_pblk & ES_TYPE_MASK) >> ES_SHIFT;
0160 }
0161
0162 static inline int ext4_es_is_written(struct extent_status *es)
0163 {
0164     return (ext4_es_type(es) & EXTENT_STATUS_WRITTEN) != 0;
0165 }
0166
0167 static inline int ext4_es_is_unwritten(struct extent_status *es)
0168 {
0169     return (ext4_es_type(es) & EXTENT_STATUS_UNWRITTEN) != 0;
0170 }
0171
0172 static inline int ext4_es_is_delayed(struct extent_status *es)
0173 {
0174     return (ext4_es_type(es) & EXTENT_STATUS_DELAYED) != 0;
0175 }
0176
0177 static inline int ext4_es_is_hole(struct extent_status *es)
0178 {
0179     return (ext4_es_type(es) & EXTENT_STATUS_HOLE) != 0;
0180 }
0181
0182 static inline int ext4_es_is_mapped(struct extent_status *es)
0183 {
0184     return (ext4_es_is_written(es) || ext4_es_is_unwritten(es));
0185 }
0186
0187 static inline int ext4_es_is_delonly(struct extent_status *es)
0188 {
0189     return (ext4_es_is_delayed(es) && !ext4_es_is_unwritten(es));
0190 }
0191
0192 static inline void ext4_es_set_referenced(struct extent_status *es)
0193 {
0194     es->es_pblk |= ((ext4_fsblk_t)EXTENT_STATUS_REFERENCED) << ES_SHIFT;
0195 }
0196
0197 static inline void ext4_es_clear_referenced(struct extent_status *es)
0198 {
0199     es->es_pblk &= ~(((ext4_fsblk_t)EXTENT_STATUS_REFERENCED) << ES_SHIFT);
0200 }
0201
0202 static inline int ext4_es_is_referenced(struct extent_status *es)
0203 {
0204     return (ext4_es_status(es) & EXTENT_STATUS_REFERENCED) != 0;
0205 }
0206
0207 static inline ext4_fsblk_t ext4_es_pblock(struct extent_status *es)
0208 {
0209     return es->es_pblk & ~ES_MASK;
0210 }
0211
0212 static inline ext4_fsblk_t ext4_es_show_pblock(struct extent_status *es)
0213 {
0214     ext4_fsblk_t pblock = ext4_es_pblock(es);
0215     return pblock == ~ES_MASK ? 0 : pblock;
0216 }
0217
0218 static inline void ext4_es_store_pblock(struct extent_status *es,
0219                     ext4_fsblk_t pb)
0220 {
0221     ext4_fsblk_t block;
0222
0223     block = (pb & ~ES_MASK) | (es->es_pblk & ES_MASK);
0224     es->es_pblk = block;
0225 }
0226
0227 static inline void ext4_es_store_status(struct extent_status *es,
0228                     unsigned int status)
0229 {
0230     es->es_pblk = (((ext4_fsblk_t)status << ES_SHIFT) & ES_MASK) |
0231               (es->es_pblk & ~ES_MASK);
0232 }
0233
0234 static inline void ext4_es_store_pblock_status(struct extent_status *es,
0235                            ext4_fsblk_t pb,
0236                            unsigned int status)
0237 {
0238     es->es_pblk = (((ext4_fsblk_t)status << ES_SHIFT) & ES_MASK) |
0239               (pb & ~ES_MASK);
0240 }
0241
0242 extern int ext4_es_register_shrinker(struct ext4_sb_info *sbi);
0243 extern void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi);
0244
0245 extern int ext4_seq_es_shrinker_info_show(struct seq_file *seq, void *v);
0246
0247 extern int __init ext4_init_pending(void);
0248 extern void ext4_exit_pending(void);
0249 extern void ext4_init_pending_tree(struct ext4_pending_tree *tree);
0250 extern void ext4_remove_pending(struct inode *inode, ext4_lblk_t lblk);
0251 extern bool ext4_is_pending(struct inode *inode, ext4_lblk_t lblk);
0252 extern int ext4_es_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk,
0253                     bool allocated);
0254 extern unsigned int ext4_es_delayed_clu(struct inode *inode, ext4_lblk_t lblk,
0255                     ext4_lblk_t len);
0256 extern void ext4_clear_inode_es(struct inode *inode);
0257
0258 #endif /* _EXT4_EXTENTS_STATUS_H */