Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <3e33c1ba057d39f145bb935b6f92f17dc9cbd207.camel@poochiereds.net>
Subject: Re: [PATCH 1/4] SUNRPC: Don't leak netobj memory when
 gss_read_proxy_verf() fails
From:   Jeff Layton <jlayton@poochiereds.net>
To:     Chuck Lever <chuck.lever@oracle.com>, linux-nfs@vger.kernel.org
Date:   Mon, 28 Nov 2022 08:11:28 -0500
In-Reply-To: <166949611830.106845.15345645610329421030.stgit@klimt.1015granger.net>
References: <166949601705.106845.10614964159272504008.stgit@klimt.1015granger.net>
         <166949611830.106845.15345645610329421030.stgit@klimt.1015granger.net>
Content-Type: text/plain; charset="ISO-8859-15"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.46.1 (3.46.1-1.fc37) 
MIME-Version: 1.0
Precedence: bulk

On Sat, 2022-11-26 at 15:55 -0500, Chuck Lever wrote:
> Fixes: 030d794bf498 ("SUNRPC: Use gssproxy upcall for server RPCGSS authe=
ntication.")
> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
> Cc: <stable@vger.kernel.org>
> ---
>  net/sunrpc/auth_gss/svcauth_gss.c |    9 +++++++--
>  1 file changed, 7 insertions(+), 2 deletions(-)
>=20
> diff --git a/net/sunrpc/auth_gss/svcauth_gss.c b/net/sunrpc/auth_gss/svca=
uth_gss.c
> index bcd74dddbe2d..9a5db285d4ae 100644
> --- a/net/sunrpc/auth_gss/svcauth_gss.c
> +++ b/net/sunrpc/auth_gss/svcauth_gss.c
> @@ -1162,18 +1162,23 @@ static int gss_read_proxy_verf(struct svc_rqst *r=
qstp,
>  		return res;
> =20
>  	inlen =3D svc_getnl(argv);
> -	if (inlen > (argv->iov_len + rqstp->rq_arg.page_len))
> +	if (inlen > (argv->iov_len + rqstp->rq_arg.page_len)) {
> +		kfree(in_handle->data);

I wish this were more obvious in this code. It's not at all evident to
the casual reader that gss_read_common_verf calls dup_netobj here and
that you need to clean up after it. At a bare minimum, we ought to have
a comment to that effect over gss_read_common_verf. While you're in
there, that function is also pretty big to be marked static inline. Can
you change that too? Ditto for gss_read_verf.


>  		return SVC_DENIED;
> +	}
> =20
>  	pages =3D DIV_ROUND_UP(inlen, PAGE_SIZE);
>  	in_token->pages =3D kcalloc(pages, sizeof(struct page *), GFP_KERNEL);
> -	if (!in_token->pages)
> +	if (!in_token->pages) {
> +		kfree(in_handle->data);
>  		return SVC_DENIED;
> +	}
>  	in_token->page_base =3D 0;
>  	in_token->page_len =3D inlen;
>  	for (i =3D 0; i < pages; i++) {
>  		in_token->pages[i] =3D alloc_page(GFP_KERNEL);
>  		if (!in_token->pages[i]) {
> +			kfree(in_handle->data);
>  			gss_free_in_token_pages(in_token);
>  			return SVC_DENIED;
>  		}
>=20
>=20

--=20
Jeff Layton <jlayton@poochiereds.net>