ext4: limit the length of per-inode prealloc list

In the scenario of writing sparse files, the per-inode prealloc list may
be very long, resulting in high overhead for ext4_mb_use_preallocated().
To circumvent this problem, we limit the maximum length of per-inode
prealloc list to 512 and allow users to modify it.

After patching, we observed that the sys ratio of cpu has dropped, and
the system throughput has increased significantly. We created a process
to write the sparse file, and the running time of the process on the
fixed kernel was significantly reduced, as follows:

Running time on unfixed kernel：
[root@TENCENT64 ~]# time taskset 0x01 ./sparse /data1/sparce.dat
real 0m2.051s
user 0m0.008s
sys 0m2.026s

Running time on fixed kernel：
[root@TENCENT64 ~]# time taskset 0x01 ./sparse /data1/sparce.dat
real 0m0.471s
user 0m0.004s
sys 0m0.395s

Signed-off-by: Chunguang Xu <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]
Signed-off-by: Theodore Ts'o <[e-mail]>

Documentation/admin-guide/ext4.rst

···
         multiple of this tuning parameter if the stripe size is not set in the
         ext4 superblock
 
+  mb_max_inode_prealloc
+        The maximum length of per-inode ext4_prealloc_space list.
+
   mb_max_to_scan
         The maximum number of extents the multiblock allocator will search to
         find the best extent.

fs/ext4/ext4.h

···
 	struct timespec64 i_crtime;
 
 	/* mballoc */
+	atomic_t i_prealloc_active;
 	struct list_head i_prealloc_list;
 	spinlock_t i_prealloc_lock;
 
···
 	unsigned int s_mb_stats;
 	unsigned int s_mb_order2_reqs;
 	unsigned int s_mb_group_prealloc;
+	unsigned int s_mb_max_inode_prealloc;
 	unsigned int s_max_dir_size_kb;
 	/* where last allocation was done - for stream allocation */
 	unsigned long s_mb_last_group;
···
 extern ext4_fsblk_t ext4_mb_new_blocks(handle_t *,
 				struct ext4_allocation_request *, int *);
 extern int ext4_mb_reserve_blocks(struct super_block *, int);
-extern void ext4_discard_preallocations(struct inode *);
+extern void ext4_discard_preallocations(struct inode *, unsigned int);
 extern int __init ext4_init_mballoc(void);
 extern void ext4_exit_mballoc(void);
 extern ext4_group_t ext4_mb_prefetch(struct super_block *sb,

fs/ext4/extents.c

···
 	 * i_mutex. So we can safely drop the i_data_sem here.
 	 */
 	BUG_ON(EXT4_JOURNAL(inode) == NULL);
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 	up_write(&EXT4_I(inode)->i_data_sem);
 	*dropped = 1;
 	return 0;
···
 			 * not a good idea to call discard here directly,
 			 * but otherwise we'd need to call it every free().
 			 */
-			ext4_discard_preallocations(inode);
+			ext4_discard_preallocations(inode, 0);
 			if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
 				fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
 			ext4_free_blocks(handle, inode, NULL, newblock,
···
 	}
 
 	down_write(&EXT4_I(inode)->i_data_sem);
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 
 	ret = ext4_es_remove_extent(inode, punch_start,
 				    EXT_MAX_BLOCKS - punch_start);
···
 		up_write(&EXT4_I(inode)->i_data_sem);
 		goto out_stop;
 	}
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 
 	ret = ext4_ext_shift_extents(inode, handle, punch_stop,
 				     punch_stop - punch_start, SHIFT_LEFT);
···
 		goto out_stop;
 
 	down_write(&EXT4_I(inode)->i_data_sem);
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 
 	path = ext4_find_extent(inode, offset_lblk, NULL, 0);
 	if (IS_ERR(path)) {

fs/ext4/file.c

···
 			(atomic_read(&inode->i_writecount) == 1) &&
 			!EXT4_I(inode)->i_reserved_data_blocks) {
 		down_write(&EXT4_I(inode)->i_data_sem);
-		ext4_discard_preallocations(inode);
+		ext4_discard_preallocations(inode, 0);
 		up_write(&EXT4_I(inode)->i_data_sem);
 	}
 	if (is_dx(inode) && filp->private_data)

fs/ext4/indirect.c

···
 	 * i_mutex. So we can safely drop the i_data_sem here.
 	 */
 	BUG_ON(EXT4_JOURNAL(inode) == NULL);
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 	up_write(&EXT4_I(inode)->i_data_sem);
 	*dropped = 1;
 	return 0;

fs/ext4/inode.c

···
 	 */
 	if ((ei->i_reserved_data_blocks == 0) &&
 	    !inode_is_open_for_write(inode))
-		ext4_discard_preallocations(inode);
+		ext4_discard_preallocations(inode, 0);
 }
 
 static int __check_block_validity(struct inode *inode, const char *func,
···
 	if (stop_block > first_block) {
 
 		down_write(&EXT4_I(inode)->i_data_sem);
-		ext4_discard_preallocations(inode);
+		ext4_discard_preallocations(inode, 0);
 
 		ret = ext4_es_remove_extent(inode, first_block,
 					    stop_block - first_block);
···
 
 	down_write(&EXT4_I(inode)->i_data_sem);
 
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 
 	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 		err = ext4_ext_truncate(handle, inode);

fs/ext4/ioctl.c

···
 	reset_inode_seed(inode);
 	reset_inode_seed(inode_bl);
 
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 
 	err = ext4_mark_inode_dirty(handle, inode);
 	if (err < 0) {

fs/ext4/mballoc.c

···
 	sbi->s_mb_stats = MB_DEFAULT_STATS;
 	sbi->s_mb_stream_request = MB_DEFAULT_STREAM_THRESHOLD;
 	sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
+	sbi->s_mb_max_inode_prealloc = MB_DEFAULT_MAX_INODE_PREALLOC;
 	/*
 	 * The default group preallocation is 512, which for 4k block
 	 * sizes translates to 2 megabytes.  However for bigalloc file
···
 	mb_debug(sb, "preallocated %d for group %u\n", preallocated, group);
 }
 
+static void ext4_mb_mark_pa_deleted(struct super_block *sb,
+				    struct ext4_prealloc_space *pa)
+{
+	struct ext4_inode_info *ei;
+
+	if (pa->pa_deleted) {
+		ext4_warning(sb, "deleted pa, type:%d, pblk:%llu, lblk:%u, len:%d\n",
+			     pa->pa_type, pa->pa_pstart, pa->pa_lstart,
+			     pa->pa_len);
+		return;
+	}
+
+	pa->pa_deleted = 1;
+
+	if (pa->pa_type == MB_INODE_PA) {
+		ei = EXT4_I(pa->pa_inode);
+		atomic_dec(&ei->i_prealloc_active);
+	}
+}
+
 static void ext4_mb_pa_callback(struct rcu_head *head)
 {
 	struct ext4_prealloc_space *pa;
···
 		return;
 	}
 
-	pa->pa_deleted = 1;
+	ext4_mb_mark_pa_deleted(sb, pa);
 	spin_unlock(&pa->pa_lock);
 
 	grp_blk = pa->pa_pstart;
···
 	spin_lock(pa->pa_obj_lock);
 	list_add_rcu(&pa->pa_inode_list, &ei->i_prealloc_list);
 	spin_unlock(pa->pa_obj_lock);
+	atomic_inc(&ei->i_prealloc_active);
 }
 
 /*
···
 		}
 
 		/* seems this one can be freed ... */
-		pa->pa_deleted = 1;
+		ext4_mb_mark_pa_deleted(sb, pa);
 
 		/* we can trust pa_free ... */
 		free += pa->pa_free;
···
  *
  * FIXME!! Make sure it is valid at all the call sites
  */
-void ext4_discard_preallocations(struct inode *inode)
+void ext4_discard_preallocations(struct inode *inode, unsigned int needed)
 {
 	struct ext4_inode_info *ei = EXT4_I(inode);
 	struct super_block *sb = inode->i_sb;
···
 
 	mb_debug(sb, "discard preallocation for inode %lu\n",
 		 inode->i_ino);
-	trace_ext4_discard_preallocations(inode);
+	trace_ext4_discard_preallocations(inode,
+			atomic_read(&ei->i_prealloc_active), needed);
 
 	INIT_LIST_HEAD(&list);
+
+	if (needed == 0)
+		needed = UINT_MAX;
 
 repeat:
 	/* first, collect all pa's in the inode */
 	spin_lock(&ei->i_prealloc_lock);
-	while (!list_empty(&ei->i_prealloc_list)) {
-		pa = list_entry(ei->i_prealloc_list.next,
+	while (!list_empty(&ei->i_prealloc_list) && needed) {
+		pa = list_entry(ei->i_prealloc_list.prev,
 				struct ext4_prealloc_space, pa_inode_list);
 		BUG_ON(pa->pa_obj_lock != &ei->i_prealloc_lock);
 		spin_lock(&pa->pa_lock);
···
 
 		}
 		if (pa->pa_deleted == 0) {
-			pa->pa_deleted = 1;
+			ext4_mb_mark_pa_deleted(sb, pa);
 			spin_unlock(&pa->pa_lock);
 			list_del_rcu(&pa->pa_inode_list);
 			list_add(&pa->u.pa_tmp_list, &list);
+			needed--;
 			continue;
 		}
 
···
 		BUG_ON(pa->pa_type != MB_GROUP_PA);
 
 		/* seems this one can be freed ... */
-		pa->pa_deleted = 1;
+		ext4_mb_mark_pa_deleted(sb, pa);
 		spin_unlock(&pa->pa_lock);
 
 		list_del_rcu(&pa->pa_inode_list);
···
 }
 
 /*
+ * if per-inode prealloc list is too long, trim some PA
+ */
+static void ext4_mb_trim_inode_pa(struct inode *inode)
+{
+	struct ext4_inode_info *ei = EXT4_I(inode);
+	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
+	int count, delta;
+
+	count = atomic_read(&ei->i_prealloc_active);
+	delta = (sbi->s_mb_max_inode_prealloc >> 2) + 1;
+	if (count > sbi->s_mb_max_inode_prealloc + delta) {
+		count -= sbi->s_mb_max_inode_prealloc;
+		ext4_discard_preallocations(inode, count);
+	}
+}
+
+/*
  * release all resource we used in allocation
  */
 static int ext4_mb_release_context(struct ext4_allocation_context *ac)
 {
+	struct inode *inode = ac->ac_inode;
+	struct ext4_inode_info *ei = EXT4_I(inode);
 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
 	struct ext4_prealloc_space *pa = ac->ac_pa;
 	if (pa) {
···
 				ext4_mb_add_n_trim(ac);
 			}
 		}
+
+		if (pa->pa_type == MB_INODE_PA) {
+			/*
+			 * treat per-inode prealloc list as a lru list, then try
+			 * to trim the least recently used PA.
+			 */
+			spin_lock(pa->pa_obj_lock);
+			list_move(&pa->pa_inode_list, &ei->i_prealloc_list);
+			spin_unlock(pa->pa_obj_lock);
+		}
+
 		ext4_mb_put_pa(ac, ac->ac_sb, pa);
 	}
 	if (ac->ac_bitmap_page)
···
 	if (ac->ac_flags & EXT4_MB_HINT_GROUP_ALLOC)
 		mutex_unlock(&ac->ac_lg->lg_mutex);
 	ext4_mb_collect_stats(ac);
+	ext4_mb_trim_inode_pa(inode);
 	return 0;
 }
 

fs/ext4/mballoc.h

···
  */
 #define MB_DEFAULT_GROUP_PREALLOC	512
 
+/*
+ * maximum length of inode prealloc list
+ */
+#define MB_DEFAULT_MAX_INODE_PREALLOC	512
 
 struct ext4_free_data {
 	/* this links the free block information from sb_info */

fs/ext4/move_extent.c

···
 
 out:
 	if (*moved_len) {
-		ext4_discard_preallocations(orig_inode);
-		ext4_discard_preallocations(donor_inode);
+		ext4_discard_preallocations(orig_inode, 0);
+		ext4_discard_preallocations(donor_inode, 0);
 	}
 
 	ext4_ext_drop_refs(path);

fs/ext4/super.c

···
 	inode_set_iversion(&ei->vfs_inode, 1);
 	spin_lock_init(&ei->i_raw_lock);
 	INIT_LIST_HEAD(&ei->i_prealloc_list);
+	atomic_set(&ei->i_prealloc_active, 0);
 	spin_lock_init(&ei->i_prealloc_lock);
 	ext4_es_init_tree(&ei->i_es_tree);
 	rwlock_init(&ei->i_es_lock);
···
 {
 	invalidate_inode_buffers(inode);
 	clear_inode(inode);
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 	ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 	dquot_drop(inode);
 	if (EXT4_I(inode)->jinode) {

fs/ext4/sysfs.c

···
 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
+EXT4_RW_ATTR_SBI_UI(mb_max_inode_prealloc, s_mb_max_inode_prealloc);
 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
 EXT4_ATTR(trigger_fs_error, 0200, trigger_test_error);
 EXT4_RW_ATTR_SBI_UI(err_ratelimit_interval_ms, s_err_ratelimit_state.interval);
···
 	ATTR_LIST(mb_order2_req),
 	ATTR_LIST(mb_stream_req),
 	ATTR_LIST(mb_group_prealloc),
+	ATTR_LIST(mb_max_inode_prealloc),
 	ATTR_LIST(max_writeback_mb_bump),
 	ATTR_LIST(extent_max_zeroout_kb),
 	ATTR_LIST(trigger_fs_error),

include/trace/events/ext4.h

···
 );
 
 TRACE_EVENT(ext4_discard_preallocations,
-	TP_PROTO(struct inode *inode),
+	TP_PROTO(struct inode *inode, unsigned int len, unsigned int needed),
 
-	TP_ARGS(inode),
+	TP_ARGS(inode, len, needed),
 
 	TP_STRUCT__entry(
-		__field(	dev_t,	dev			)
-		__field(	ino_t,	ino			)
+		__field(	dev_t,		dev		)
+		__field(	ino_t,		ino		)
+		__field(	unsigned int,	len		)
+		__field(	unsigned int,	needed		)
 
 	),
 
 	TP_fast_assign(
 		__entry->dev	= inode->i_sb->s_dev;
 		__entry->ino	= inode->i_ino;
+		__entry->len	= len;
+		__entry->needed	= needed;
 	),
 
-	TP_printk("dev %d,%d ino %lu",
+	TP_printk("dev %d,%d ino %lu len: %u needed %u",
 		  MAJOR(__entry->dev), MINOR(__entry->dev),
-		  (unsigned long) __entry->ino)
+		  (unsigned long) __entry->ino, __entry->len,
+		  __entry->needed)
 );
 
 TRACE_EVENT(ext4_mb_discard_preallocations,