1 ext4: introduce aging to extent status tree
3 From: Jan Kara <jack@suse.cz>
5 Introduce a simple aging to extent status tree. Each extent has a
6 REFERENCED bit which gets set when the extent is used. Shrinker then
7 skips entries with referenced bit set and clears the bit. Thus
8 frequently used extents have higher chances of staying in memory.
10 Signed-off-by: Jan Kara <jack@suse.cz>
11 Signed-off-by: Theodore Ts'o <tytso@mit.edu>
13 fs/ext4/extents_status.c | 22 +++++++++++++++++-----
14 fs/ext4/extents_status.h | 35 +++++++++++++++++++++++++++++++----
15 2 files changed, 48 insertions(+), 9 deletions(-)
17 diff --git a/fs/ext4/extents_status.c b/fs/ext4/extents_status.c
18 index 30596498ed0b..e04d45733976 100644
19 --- a/fs/ext4/extents_status.c
20 +++ b/fs/ext4/extents_status.c
21 @@ -382,7 +382,7 @@ static void ext4_es_free_extent(struct inode *inode, struct extent_status *es)
22 static int ext4_es_can_be_merged(struct extent_status *es1,
23 struct extent_status *es2)
25 - if (ext4_es_status(es1) != ext4_es_status(es2))
26 + if (ext4_es_type(es1) != ext4_es_type(es2))
29 if (((__u64) es1->es_len) + es2->es_len > EXT_MAX_BLOCKS) {
30 @@ -425,6 +425,8 @@ ext4_es_try_to_merge_left(struct inode *inode, struct extent_status *es)
31 es1 = rb_entry(node, struct extent_status, rb_node);
32 if (ext4_es_can_be_merged(es1, es)) {
33 es1->es_len += es->es_len;
34 + if (ext4_es_is_referenced(es))
35 + ext4_es_set_referenced(es1);
36 rb_erase(&es->rb_node, &tree->root);
37 ext4_es_free_extent(inode, es);
39 @@ -447,6 +449,8 @@ ext4_es_try_to_merge_right(struct inode *inode, struct extent_status *es)
40 es1 = rb_entry(node, struct extent_status, rb_node);
41 if (ext4_es_can_be_merged(es, es1)) {
42 es->es_len += es1->es_len;
43 + if (ext4_es_is_referenced(es1))
44 + ext4_es_set_referenced(es);
45 rb_erase(node, &tree->root);
46 ext4_es_free_extent(inode, es1);
48 @@ -813,6 +817,8 @@ out:
49 es->es_lblk = es1->es_lblk;
50 es->es_len = es1->es_len;
51 es->es_pblk = es1->es_pblk;
52 + if (!ext4_es_is_referenced(es))
53 + ext4_es_set_referenced(es);
54 stats->es_stats_cache_hits++;
56 stats->es_stats_cache_misses++;
57 @@ -1252,11 +1258,17 @@ static int es_do_reclaim_extents(struct ext4_inode_info *ei, ext4_lblk_t end,
58 * We can't reclaim delayed extent from status tree because
59 * fiemap, bigallic, and seek_data/hole need to use it.
61 - if (!ext4_es_is_delayed(es)) {
62 - rb_erase(&es->rb_node, &tree->root);
63 - ext4_es_free_extent(inode, es);
65 + if (ext4_es_is_delayed(es))
67 + if (ext4_es_is_referenced(es)) {
68 + ext4_es_clear_referenced(es);
72 + rb_erase(&es->rb_node, &tree->root);
73 + ext4_es_free_extent(inode, es);
78 es = rb_entry(node, struct extent_status, rb_node);
79 diff --git a/fs/ext4/extents_status.h b/fs/ext4/extents_status.h
80 index e86b1f34cfec..691b52613ce4 100644
81 --- a/fs/ext4/extents_status.h
82 +++ b/fs/ext4/extents_status.h
83 @@ -34,6 +34,7 @@ enum {
91 @@ -44,6 +45,12 @@ enum {
92 #define EXTENT_STATUS_UNWRITTEN (1 << ES_UNWRITTEN_B)
93 #define EXTENT_STATUS_DELAYED (1 << ES_DELAYED_B)
94 #define EXTENT_STATUS_HOLE (1 << ES_HOLE_B)
95 +#define EXTENT_STATUS_REFERENCED (1 << ES_REFERENCED_B)
97 +#define ES_TYPE_MASK ((ext4_fsblk_t)(EXTENT_STATUS_WRITTEN | \
98 + EXTENT_STATUS_UNWRITTEN | \
99 + EXTENT_STATUS_DELAYED | \
100 + EXTENT_STATUS_HOLE) << ES_SHIFT)
104 @@ -93,24 +100,44 @@ static inline unsigned int ext4_es_status(struct extent_status *es)
105 return es->es_pblk >> ES_SHIFT;
108 +static inline unsigned int ext4_es_type(struct extent_status *es)
110 + return (es->es_pblk & ES_TYPE_MASK) >> ES_SHIFT;
113 static inline int ext4_es_is_written(struct extent_status *es)
115 - return (ext4_es_status(es) & EXTENT_STATUS_WRITTEN) != 0;
116 + return (ext4_es_type(es) & EXTENT_STATUS_WRITTEN) != 0;
119 static inline int ext4_es_is_unwritten(struct extent_status *es)
121 - return (ext4_es_status(es) & EXTENT_STATUS_UNWRITTEN) != 0;
122 + return (ext4_es_type(es) & EXTENT_STATUS_UNWRITTEN) != 0;
125 static inline int ext4_es_is_delayed(struct extent_status *es)
127 - return (ext4_es_status(es) & EXTENT_STATUS_DELAYED) != 0;
128 + return (ext4_es_type(es) & EXTENT_STATUS_DELAYED) != 0;
131 static inline int ext4_es_is_hole(struct extent_status *es)
133 - return (ext4_es_status(es) & EXTENT_STATUS_HOLE) != 0;
134 + return (ext4_es_type(es) & EXTENT_STATUS_HOLE) != 0;
137 +static inline void ext4_es_set_referenced(struct extent_status *es)
139 + es->es_pblk |= ((ext4_fsblk_t)EXTENT_STATUS_REFERENCED) << ES_SHIFT;
142 +static inline void ext4_es_clear_referenced(struct extent_status *es)
144 + es->es_pblk &= ~(((ext4_fsblk_t)EXTENT_STATUS_REFERENCED) << ES_SHIFT);
147 +static inline int ext4_es_is_referenced(struct extent_status *es)
149 + return (ext4_es_status(es) & EXTENT_STATUS_REFERENCED) != 0;
152 static inline ext4_fsblk_t ext4_es_pblock(struct extent_status *es)