Received-SPF: pass (google.com: domain of linux-kernel+bounces-128685-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Precedence: bulk
MIME-Version: 1.0
References: <20240402180026.2469459-1-snovitoll@gmail.com>
In-Reply-To: <20240402180026.2469459-1-snovitoll@gmail.com>
From: Ryusuke Konishi <konishi.ryusuke@gmail.com>
Date: Wed, 3 Apr 2024 05:55:30 +0900
Message-ID: <CAKFNMonDO_diEo+_JXKgHMTmOPv5T30TALb-ZvsOprcwSJCtFQ@mail.gmail.com>
Subject: Re: [PATCH] fs/nilfs2: prevent int overflow in btree binary search
To: Sabyrzhan Tasbolatov <snovitoll@gmail.com>
Cc: linux-nilfs@vger.kernel.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Wed, Apr 3, 2024 at 3:00=E2=80=AFAM Sabyrzhan Tasbolatov wrote:
>
> Should prevent int overflow if low + high > INT_MAX in big btree with
> nchildren in nilfs_btree_node_lookup() binary search.
>
> Signed-off-by: Sabyrzhan Tasbolatov <snovitoll@gmail.com>
> ---
>  fs/nilfs2/btree.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/fs/nilfs2/btree.c b/fs/nilfs2/btree.c
> index 65659fa03..39ee4fe11 100644
> --- a/fs/nilfs2/btree.c
> +++ b/fs/nilfs2/btree.c
> @@ -300,7 +300,7 @@ static int nilfs_btree_node_lookup(const struct nilfs=
_btree_node *node,
>         index =3D 0;
>         s =3D 0;
>         while (low <=3D high) {
> -               index =3D (low + high) / 2;
> +               index =3D low + (high - low) / 2;
>                 nkey =3D nilfs_btree_node_get_key(node, index);
>                 if (nkey =3D=3D key) {
>                         s =3D 0;
> --
> 2.34.1
>

Hi Sabyrzhan,

Thank you for your interesting patch.

In this function, the value of the variable "high" is initialized with
"nilfs_btree_node_get_nchildren() - 1", and "low" is initialized with
0.

nilfs_btree_node_get_nchildren() returns a value read from a 16-bit
wide field, so it will never exceed U16_MAX.

These index calculations narrow the range between "low" and "high", so
as long as INT_MAX is 32-bit or more, it seems that the calculation of
this intermediate value will not overflow.

So while it's a good overflow avoidance technique, it doesn't seem to
happen in practice.

Am I missing something?

Regards,
Ryusuke Konishi