Re: [PATCH] btrfs: speedup mount time with force readahead chunk tree

[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]




On 2020/7/1 下午5:24, robbieko wrote:
> From: Robbie Ko <robbieko@xxxxxxxxxxxx>
> 
> When mounting, we always need to read the whole chunk tree,
> when there are too many chunk items, most of the time is
> spent on btrfs_read_chunk_tree, because we only read one
> leaf at a time.

Well, under most case it would be btrfs_read_block_groups(), unless all
data chunks are very compact with just several large data extents.

> 
> We fix this by adding a new readahead mode READA_FORWARD_FORCE,
> which reads all the leaves after the key in the node when
> reading a level 1 node.
> 
> Signed-off-by: Robbie Ko <robbieko@xxxxxxxxxxxx>
> ---
>  fs/btrfs/ctree.c   | 7 +++++--
>  fs/btrfs/ctree.h   | 2 +-
>  fs/btrfs/volumes.c | 1 +
>  3 files changed, 7 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
> index 3a7648bff42c..abb9108e2d7d 100644
> --- a/fs/btrfs/ctree.c
> +++ b/fs/btrfs/ctree.c
> @@ -2194,7 +2194,7 @@ static void reada_for_search(struct btrfs_fs_info *fs_info,
>  			if (nr == 0)
>  				break;
>  			nr--;
> -		} else if (path->reada == READA_FORWARD) {
> +		} else if (path->reada == READA_FORWARD || path->reada == READA_FORWARD_FORCE) {
>  			nr++;
>  			if (nr >= nritems)
>  				break;
> @@ -2205,12 +2205,15 @@ static void reada_for_search(struct btrfs_fs_info *fs_info,
>  				break;
>  		}
>  		search = btrfs_node_blockptr(node, nr);
> -		if ((search <= target && target - search <= 65536) ||
> +		if ((path->reada == READA_FORWARD_FORCE) ||
> +		    (search <= target && target - search <= 65536) ||
>  		    (search > target && search - target <= 65536)) {
>  			readahead_tree_block(fs_info, search);
>  			nread += blocksize;
>  		}
>  		nscan++;
> +		if (path->reada == READA_FORWARD_FORCE)
> +			continue;
>  		if ((nread > 65536 || nscan > 32))
>  			break;
>  	}
> diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
> index d404cce8ae40..808bcbdc9530 100644
> --- a/fs/btrfs/ctree.h
> +++ b/fs/btrfs/ctree.h
> @@ -353,7 +353,7 @@ struct btrfs_node {
>   * The slots array records the index of the item or block pointer
>   * used while walking the tree.
>   */
> -enum { READA_NONE, READA_BACK, READA_FORWARD };
> +enum { READA_NONE, READA_BACK, READA_FORWARD, READA_FORWARD_FORCE };
>  struct btrfs_path {
>  	struct extent_buffer *nodes[BTRFS_MAX_LEVEL];
>  	int slots[BTRFS_MAX_LEVEL];
> diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
> index 0d6e785bcb98..78fd65abff69 100644
> --- a/fs/btrfs/volumes.c
> +++ b/fs/btrfs/volumes.c
> @@ -7043,6 +7043,7 @@ int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
>  	path = btrfs_alloc_path();
>  	if (!path)
>  		return -ENOMEM;
> +	path->reada = READA_FORWARD_FORCE;

Why not just use regular forward readahead?

Mind to share the reason here? Just to force reada for all tree leaves?

Thanks,
Qu

>  
>  	/*
>  	 * uuid_mutex is needed only if we are mounting a sprout FS
> 

Attachment: signature.asc
Description: OpenPGP digital signature


[Index of Archives]     [Linux Filesystems Development]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux