ext4: change LRU to round-robin in extent status tree shrinker

In this commit we discard the lru algorithm for inodes with extent
status tree because it takes significant effort to maintain a lru list
in extent status tree shrinker and the shrinker can take a long time to
scan this lru list in order to reclaim some objects.

We replace the lru ordering with a simple round-robin.  After that we
never need to keep a lru list.  That means that the list needn't be
sorted if the shrinker can not reclaim any objects in the first round.

Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

fs/ext4/ext4.h

@@ -878,10 +878,9 @@ struct ext4_inode_info {
 	/* extents status tree */
 	struct ext4_es_tree i_es_tree;
 	rwlock_t i_es_lock;
-	struct list_head i_es_lru;
+	struct list_head i_es_list;
 	unsigned int i_es_all_nr;	/* protected by i_es_lock */
-	unsigned int i_es_lru_nr;	/* protected by i_es_lock */
-	unsigned long i_touch_when;	/* jiffies of last accessing */
+	unsigned int i_es_shk_nr;	/* protected by i_es_lock */
 
 	/* ialloc */
 	ext4_group_t	i_last_alloc_group;
@@ -1322,10 +1321,11 @@ struct ext4_sb_info {
 
 	/* Reclaim extents from extent status tree */
 	struct shrinker s_es_shrinker;
-	struct list_head s_es_lru;
+	struct list_head s_es_list;
+	long s_es_nr_inode;
 	struct ext4_es_stats s_es_stats;
 	struct mb_cache *s_mb_cache;
-	spinlock_t s_es_lru_lock ____cacheline_aligned_in_smp;
+	spinlock_t s_es_lock ____cacheline_aligned_in_smp;
 
 	/* Ratelimit ext4 messages. */
 	struct ratelimit_state s_err_ratelimit_state;

fs/ext4/extents.c

@@ -4632,7 +4632,7 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
 
 	trace_ext4_ext_map_blocks_exit(inode, flags, map,
 				       err ? err : allocated);
-	ext4_es_lru_add(inode);
+	ext4_es_list_add(inode);
 	return err ? err : allocated;
 }
 
@@ -5191,7 +5191,7 @@ int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 		error = ext4_fill_fiemap_extents(inode, start_blk,
 						 len_blks, fieinfo);
 	}
-	ext4_es_lru_add(inode);
+	ext4_es_list_add(inode);
 	return error;
 }
 

fs/ext4/extents_status.h

@@ -65,14 +65,13 @@ struct ext4_es_tree {
 };
 
 struct ext4_es_stats {
-	unsigned long es_stats_last_sorted;
 	unsigned long es_stats_shrunk;
 	unsigned long es_stats_cache_hits;
 	unsigned long es_stats_cache_misses;
 	u64 es_stats_scan_time;
 	u64 es_stats_max_scan_time;
 	struct percpu_counter es_stats_all_cnt;
-	struct percpu_counter es_stats_lru_cnt;
+	struct percpu_counter es_stats_shk_cnt;
 };
 
 extern int __init ext4_init_es(void);
@@ -151,7 +150,7 @@ static inline void ext4_es_store_pblock_status(struct extent_status *es,
 
 extern int ext4_es_register_shrinker(struct ext4_sb_info *sbi);
 extern void ext4_es_unregister_shrinker(struct ext4_sb_info *sbi);
-extern void ext4_es_lru_add(struct inode *inode);
-extern void ext4_es_lru_del(struct inode *inode);
+extern void ext4_es_list_add(struct inode *inode);
+extern void ext4_es_list_del(struct inode *inode);
 
 #endif /* _EXT4_EXTENTS_STATUS_H */

fs/ext4/inode.c

@@ -486,7 +486,7 @@ int ext4_map_blocks(handle_t *handle, struct inode *inode,
 
 	/* Lookup extent status tree firstly */
 	if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
-		ext4_es_lru_add(inode);
+		ext4_es_list_add(inode);
 		if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
 			map->m_pblk = ext4_es_pblock(&es) +
 					map->m_lblk - es.es_lblk;
@@ -1388,7 +1388,7 @@ static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
 
 	/* Lookup extent status tree firstly */
 	if (ext4_es_lookup_extent(inode, iblock, &es)) {
-		ext4_es_lru_add(inode);
+		ext4_es_list_add(inode);
 		if (ext4_es_is_hole(&es)) {
 			retval = 0;
 			down_read(&EXT4_I(inode)->i_data_sem);

fs/ext4/ioctl.c

@@ -78,8 +78,8 @@ static void swap_inode_data(struct inode *inode1, struct inode *inode2)
 	memswap(&ei1->i_disksize, &ei2->i_disksize, sizeof(ei1->i_disksize));
 	ext4_es_remove_extent(inode1, 0, EXT_MAX_BLOCKS);
 	ext4_es_remove_extent(inode2, 0, EXT_MAX_BLOCKS);
-	ext4_es_lru_del(inode1);
-	ext4_es_lru_del(inode2);
+	ext4_es_list_del(inode1);
+	ext4_es_list_del(inode2);
 
 	isize = i_size_read(inode1);
 	i_size_write(inode1, i_size_read(inode2));

fs/ext4/super.c

@@ -871,10 +871,9 @@ static struct inode *ext4_alloc_inode(struct super_block *sb)
 	spin_lock_init(&ei->i_prealloc_lock);
 	ext4_es_init_tree(&ei->i_es_tree);
 	rwlock_init(&ei->i_es_lock);
-	INIT_LIST_HEAD(&ei->i_es_lru);
+	INIT_LIST_HEAD(&ei->i_es_list);
 	ei->i_es_all_nr = 0;
-	ei->i_es_lru_nr = 0;
-	ei->i_touch_when = 0;
+	ei->i_es_shk_nr = 0;
 	ei->i_reserved_data_blocks = 0;
 	ei->i_reserved_meta_blocks = 0;
 	ei->i_allocated_meta_blocks = 0;
@@ -963,7 +962,7 @@ void ext4_clear_inode(struct inode *inode)
 	dquot_drop(inode);
 	ext4_discard_preallocations(inode);
 	ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
-	ext4_es_lru_del(inode);
+	ext4_es_list_del(inode);
 	if (EXT4_I(inode)->jinode) {
 		jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 					       EXT4_I(inode)->jinode);

include/trace/events/ext4.h

@@ -2450,15 +2450,14 @@ TRACE_EVENT(ext4_collapse_range,
 
 TRACE_EVENT(ext4_es_shrink,
 	TP_PROTO(struct super_block *sb, int nr_shrunk, u64 scan_time,
-		 int skip_precached, int nr_skipped, int retried),
+		 int nr_skipped, int retried),
 
-	TP_ARGS(sb, nr_shrunk, scan_time, skip_precached, nr_skipped, retried),
+	TP_ARGS(sb, nr_shrunk, scan_time, nr_skipped, retried),
 
 	TP_STRUCT__entry(
 		__field(	dev_t,		dev		)
 		__field(	int,		nr_shrunk	)
 		__field(	unsigned long long, scan_time	)
-		__field(	int,		skip_precached	)
 		__field(	int,		nr_skipped	)
 		__field(	int,		retried		)
 	),
@@ -2467,16 +2466,14 @@ TRACE_EVENT(ext4_es_shrink,
 		__entry->dev		= sb->s_dev;
 		__entry->nr_shrunk	= nr_shrunk;
 		__entry->scan_time	= div_u64(scan_time, 1000);
-		__entry->skip_precached = skip_precached;
 		__entry->nr_skipped	= nr_skipped;
 		__entry->retried	= retried;
 	),
 
-	TP_printk("dev %d,%d nr_shrunk %d, scan_time %llu skip_precached %d "
+	TP_printk("dev %d,%d nr_shrunk %d, scan_time %llu "
 		  "nr_skipped %d retried %d",
 		  MAJOR(__entry->dev), MINOR(__entry->dev), __entry->nr_shrunk,
-		  __entry->scan_time, __entry->skip_precached,
-		  __entry->nr_skipped, __entry->retried)
+		  __entry->scan_time, __entry->nr_skipped, __entry->retried)
 );
 
 #endif /* _TRACE_EXT4_H */