Re: [RFC PATCH 2/8] btrfs: move kfree out of btrfs_create_qgroup's cleanup path

[Qu Wenruo <wqu@suse.com>]

在 2025/11/13 05:19, Gladyshev Ilya 写道:
> Relocate kfree() from the generic cleanup path to the specific error
> exit where the allocation could leak. This prepares for future
> simplification by allowing removal of the 'out' label and use of
> mutex_guard for cleaner resource management.
> 
> Signed-off-by: Gladyshev Ilya <foxido@foxido.dev>
> ---
>   fs/btrfs/qgroup.c | 8 ++++----
>   1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/fs/btrfs/qgroup.c b/fs/btrfs/qgroup.c
> index 9904bcfd3a60..a8474d0a9c58 100644
> --- a/fs/btrfs/qgroup.c
> +++ b/fs/btrfs/qgroup.c
> @@ -1659,7 +1659,7 @@ int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
>   	struct btrfs_fs_info *fs_info = trans->fs_info;
>   	struct btrfs_root *quota_root;
>   	struct btrfs_qgroup *qgroup;
> -	struct btrfs_qgroup *prealloc = NULL;
> +	struct btrfs_qgroup *prealloc;
>   	int ret = 0;
>   
>   	mutex_lock(&fs_info->qgroup_ioctl_lock);
> @@ -1681,18 +1681,18 @@ int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
>   	}
>   
>   	ret = add_qgroup_item(trans, quota_root, qgroupid);
> -	if (ret)
> +	if (ret) {
> +		kfree(prealloc);
>   		goto out;
> +	}
>   
>   	spin_lock(&fs_info->qgroup_lock);
>   	qgroup = add_qgroup_rb(fs_info, prealloc, qgroupid);
>   	spin_unlock(&fs_info->qgroup_lock);
> -	prealloc = NULL;
>   
>   	ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
>   out:
>   	mutex_unlock(&fs_info->qgroup_ioctl_lock);

You're not on the latest for-next branch, which has the following patch 
applied doing the extra sanity checks:

https://lore.kernel.org/linux-btrfs/20251024102143.236665-5-mssola@mssola.com/

With the extra ASSERT()s, the old code makes more sense.

Thanks,
Qu
> -	kfree(prealloc);
>   	return ret;
>   }
>