Re: [PATCH 1/3] btrfs: clear BTRFS_ROOT_IN_TRANS_SETUP on early exit from record_root_in_trans()

[Boris Burkov <boris@bur.io>]

On Fri, Sep 18, 2026 at 01:28:09PM +0100, fdmanana@kernel.org wrote:
> From: Filipe Manana <fdmanana@suse.com>
> 
> If we exit early because the transaction that last used the root already
> matches the current transaction, we leave the BTRFS_ROOT_IN_TRANS_SETUP
> bit set in the root (which we just set right before the exit). While this
> does not cause any functional issue, it makes callers of
> btrfs_record_root_in_trans() always lock fs_info->reloc_mutex and call

I found "always" kind of confusing here. It's until one succeeds and
clears the bit right? It kind of makes it sound like it leaks it
forever.

> record_root_in_trans() for nothing, causing unnecessary lock contention.
> One caller of btrfs_record_root_in_trans() is start_transaction(), used to
> start new transaction or joining an existing one, which is a hot path.
> 
> So clear BTRFS_ROOT_IN_TRANS_SETUP on early exit.
> 
> Assisted-by: LLM
> Signed-off-by: Filipe Manana <fdmanana@suse.com>
> ---
>  fs/btrfs/transaction.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
> index ca114235bbe1..c1555621ae4e 100644
> --- a/fs/btrfs/transaction.c
> +++ b/fs/btrfs/transaction.c
> @@ -432,6 +432,7 @@ static int record_root_in_trans(struct btrfs_trans_handle *trans,
>  		spin_lock(&fs_info->fs_roots_radix_lock);
>  		if (btrfs_get_root_last_trans(root) == trans->transid && !force) {
>  			spin_unlock(&fs_info->fs_roots_radix_lock);
> +			clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
>  			return 0;
>  		}
>  		radix_tree_tag_set(&fs_info->fs_roots_radix,
> -- 
> 2.47.2
>