[lustre-devel] [PATCH 16/22] ext4: add warning for directory htree growth

[NeilBrown <neilb@suse.com>]

On Sun, Jul 21 2019, James Simmons wrote:

You have really out-done yourself with the commit message here !!!

I think this allows lustre to generate warnings if any single directory
exceeds some particular size ??
Any be default, the max size is 33% larger than the first directory that
anything is added to??
I guess lustre just uses one big directory??

I appreciate that this might be useful functionality.  I suspect a
better interface is needed.

NeilBrown


> Signed-off-by: James Simmons <jsimmons@infradead.org>
> ---
>  fs/ext4/ext4.h  |  1 +
>  fs/ext4/namei.c | 69 +++++++++++++++++++++++++++++++++++++++++++++++++++++++--
>  fs/ext4/super.c |  2 ++
>  fs/ext4/sysfs.c |  2 ++
>  4 files changed, 72 insertions(+), 2 deletions(-)
>
> diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
> index bf74c7c..5f73e19 100644
> --- a/fs/ext4/ext4.h
> +++ b/fs/ext4/ext4.h
> @@ -1472,6 +1472,7 @@ struct ext4_sb_info {
>  	unsigned int s_mb_group_prealloc;
>  	unsigned long *s_mb_prealloc_table;
>  	unsigned int s_max_dir_size_kb;
> +	unsigned long s_warning_dir_size;
>  	/* where last allocation was done - for stream allocation */
>  	unsigned long s_mb_last_group;
>  	unsigned long s_mb_last_start;
> diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
> index 1b6d22a..9b30cc6 100644
> --- a/fs/ext4/namei.c
> +++ b/fs/ext4/namei.c
> @@ -757,12 +757,20 @@ struct ext4_dir_lock_data {
>  #define ext4_htree_lock_data(l)	((struct ext4_dir_lock_data *)(l)->lk_private)
>  #define ext4_find_entry(dir, name, dirent, inline) \
>  			ext4_find_entry_locked(dir, name, dirent, inline, NULL)
> -#define ext4_add_entry(handle, dentry, inode) \
> -		       ext4_add_entry_locked(handle, dentry, inode, NULL)
>  
>  /* NB: ext4_lblk_t is 32 bits so we use high bits to identify invalid blk */
>  #define EXT4_HTREE_NODE_CHANGED		(0xcafeULL << 32)
>  
> +inline int ext4_add_entry(handle_t *handle, struct dentry *dentry,
> +			  struct inode *inode)
> +{
> +	int ret = ext4_add_entry_locked(handle, dentry, inode, NULL);
> +
> +	if (ret == -ENOBUFS)
> +		ret = 0;
> +	return ret;
> +}
> +
>  static void ext4_htree_event_cb(void *target, void *event)
>  {
>  	u64 *block = (u64 *)target;
> @@ -2612,6 +2620,55 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
>  	return err;
>  }
>  
> +static unsigned long __ext4_max_dir_size(struct dx_frame *frames,
> +					 struct dx_frame *frame,
> +					 struct inode *dir)
> +{
> +	unsigned long max_dir_size;
> +
> +	if (EXT4_SB(dir->i_sb)->s_max_dir_size_kb) {
> +		max_dir_size = EXT4_SB(dir->i_sb)->s_max_dir_size_kb << 10;
> +	} else {
> +		max_dir_size = EXT4_BLOCK_SIZE(dir->i_sb);
> +		while (frame >= frames) {
> +			max_dir_size *= dx_get_limit(frame->entries);
> +			if (frame == frames)
> +				break;
> +			frame--;
> +		}
> +		/* use 75% of max dir size in average */
> +		max_dir_size = max_dir_size / 4 * 3;
> +	}
> +	return max_dir_size;
> +}
> +
> +/*
> + * With hash tree growing, it is easy to hit ENOSPC, but it is hard
> + * to predict when it will happen. let's give administrators warning
> + * when reaching 3/5 and 2/3 of limit
> + */
> +static inline bool dir_size_in_warning_range(struct dx_frame *frames,
> +					     struct dx_frame *frame,
> +					     struct inode *dir)
> +{
> +	struct super_block *sb = dir->i_sb;
> +	unsigned long size1, size2;
> +
> +	if (unlikely(!EXT4_SB(sb)->s_warning_dir_size))
> +		EXT4_SB(sb)->s_warning_dir_size =
> +			__ext4_max_dir_size(frames, frame, dir);
> +
> +	size1 = EXT4_SB(sb)->s_warning_dir_size / 16 * 10;
> +	size1 = size1 & ~(EXT4_BLOCK_SIZE(sb) - 1);
> +	size2 = EXT4_SB(sb)->s_warning_dir_size / 16 * 11;
> +	size2 = size2 & ~(EXT4_BLOCK_SIZE(sb) - 1);
> +	if (in_range(dir->i_size, size1, EXT4_BLOCK_SIZE(sb)) ||
> +	    in_range(dir->i_size, size2, EXT4_BLOCK_SIZE(sb)))
> +		return true;
> +
> +	return false;
> +}
> +
>  /*
>   *	ext4_add_entry()
>   *
> @@ -2739,6 +2796,7 @@ static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
>  	struct buffer_head *bh;
>  	struct super_block *sb = dir->i_sb;
>  	struct ext4_dir_entry_2 *de;
> +	bool ret_warn = false;
>  	int restart;
>  	int err;
>  
> @@ -2769,6 +2827,11 @@ static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
>  	/* Block full, should compress but for now just split */
>  	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
>  		       dx_get_count(entries), dx_get_limit(entries)));
> +
> +	if (frame - frames + 1 >= ext4_dir_htree_level(sb) ||
> +	    EXT4_SB(sb)->s_warning_dir_size)
> +		ret_warn = dir_size_in_warning_range(frames, frame, dir);
> +
>  	/* Need to split index? */
>  	if (dx_get_count(entries) == dx_get_limit(entries)) {
>  		ext4_lblk_t newblock;
> @@ -2935,6 +2998,8 @@ static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
>  	 */
>  	if (restart && err == 0)
>  		goto again;
> +	if (err == 0 && ret_warn)
> +		err = -ENOBUFS;
>  	return err;
>  }
>  
> diff --git a/fs/ext4/super.c b/fs/ext4/super.c
> index 07242d7..a3179b2 100644
> --- a/fs/ext4/super.c
> +++ b/fs/ext4/super.c
> @@ -1901,6 +1901,8 @@ static int handle_mount_opt(struct super_block *sb, char *opt, int token,
>  		sbi->s_li_wait_mult = arg;
>  	} else if (token == Opt_max_dir_size_kb) {
>  		sbi->s_max_dir_size_kb = arg;
> +		/* reset s_warning_dir_size and make it re-calculated */
> +		sbi->s_warning_dir_size = 0;
>  	} else if (token == Opt_stripe) {
>  		sbi->s_stripe = arg;
>  	} else if (token == Opt_resuid) {
> diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
> index 3a71a16..575f318 100644
> --- a/fs/ext4/sysfs.c
> +++ b/fs/ext4/sysfs.c
> @@ -182,6 +182,7 @@ static ssize_t journal_task_show(struct ext4_sb_info *sbi, char *buf)
>  EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
>  EXT4_RW_ATTR_SBI_UI(max_dir_size, s_max_dir_size_kb);
>  EXT4_RW_ATTR_SBI_UI(max_dir_size_kb, s_max_dir_size_kb);
> +EXT4_RW_ATTR_SBI_UI(warning_dir_size, s_warning_dir_size);
>  EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
>  EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
>  EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
> @@ -214,6 +215,7 @@ static ssize_t journal_task_show(struct ext4_sb_info *sbi, char *buf)
>  	ATTR_LIST(inode_goal),
>  	ATTR_LIST(max_dir_size),
>  	ATTR_LIST(max_dir_size_kb),
> +	ATTR_LIST(warning_dir_size),
>  	ATTR_LIST(mb_stats),
>  	ATTR_LIST(mb_max_to_scan),
>  	ATTR_LIST(mb_min_to_scan),
> -- 
> 1.8.3.1
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 832 bytes
Desc: not available
URL: <http://lists.lustre.org/pipermail/lustre-devel-lustre.org/attachments/20190722/e1069c6e/attachment-0001.sig>