Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <a2821ac8-6903-372e-36a4-9667b9fceac2@gmail.com>
Date:   Tue, 10 Jan 2023 17:32:15 +0000
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.6.1
Subject: Re: [PATCH] xfrm: compat: change expression for switch in
 xfrm_xlate64
Content-Language: en-US
To:     Anastasia Belova <abelova@astralinux.ru>,
        Steffen Klassert <steffen.klassert@secunet.com>
Cc:     Herbert Xu <herbert@gondor.apana.org.au>,
        "David S. Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>,
        Jakub Kicinski <kuba@kernel.org>,
        Paolo Abeni <pabeni@redhat.com>, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org, lvc-project@linuxtesting.org
References: <20230110091450.21696-1-abelova@astralinux.ru>
From:   Dmitry Safonov <0x7f454c46@gmail.com>
In-Reply-To: <20230110091450.21696-1-abelova@astralinux.ru>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 1/10/23 09:14, Anastasia Belova wrote:
> Compare XFRM_MSG_NEWSPDINFO (value from netlink
> configuration messages enum) with nlh_src->nlmsg_type
> instead of nlh_src->nlmsg_type - XFRM_MSG_BASE.
> 
> Found by Linux Verification Center (linuxtesting.org) with SVACE.

Nice find!

Looking at the details:
xfrm_xlate64() is for translating 64-bit kernel-issued message to 32-bit
userspace ABI. The message is XFRM_MSG_NEWSPDINFO and there's a selftest
that checks if the attributes (thresh valued) were correctly translated
in tools/testing/selftests/net/ipsec.c

So, I was interested in how did it go unnoticed?
The switch here is to differ XFRMA_* attributes from XFRMA_SPD_*
attributes, which can be just copied as they are as they occupy the same
space on 64-bit as well as on 32-bit.

enum xfrm_spdattr_type_t {
	XFRMA_SPD_UNSPEC,
	XFRMA_SPD_INFO,
	XFRMA_SPD_HINFO,
	XFRMA_SPD_IPV4_HTHRESH,
	XFRMA_SPD_IPV6_HTHRESH,
	__XFRMA_SPD_MAX
};
attributes went through xfrm_xlate64_attr() instead of just being
copied. That worked in result as
	case XFRMA_UNSPEC:
	case XFRMA_ALG_AUTH:
	case XFRMA_ALG_CRYPT:
	case XFRMA_ALG_COMP:
	case XFRMA_ENCAP:
	case XFRMA_TMPL:
		return xfrm_nla_cpy(dst, src, nla_len(src));
are equal by value (XFRMA_UNSPEC == 0 == XFRMA_SPD_UNSPEC) and so on.
So, in result, even with this typo the code worked.

What about the reverse case, what was being copied by this
XFRM_MSG_NEWSPDINFO case?
XFRM_MSG_NEWSPDINFO == 0x24
So, before this patch (XFRM_MSG_NEWSPDINFO - XFRM_MSG_BASE) == 0x14 ==
XFRM_MSG_DELPOLICY type of messages would fit this switch case.

Luckily enough, kernel doesn't send back XFRM_MSG_DELPOLICY messages to
userspace. That's how it went unnoticed, by unexpectedly still working.


> Fixes: 4e9505064f58 ("net/xfrm/compat: Copy xfrm_spdattr_type_t atributes")
> Signed-off-by: Anastasia Belova <abelova@astralinux.ru>

Thanks for the patch,
Acked-by: Dmitry Safonov <0x7f454c46@gmail.com>
Tested-by: Dmitry Safonov <0x7f454c46@gmail.com>

> ---
>  net/xfrm/xfrm_compat.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/net/xfrm/xfrm_compat.c b/net/xfrm/xfrm_compat.c
> index a0f62fa02e06..12405aa5bce8 100644
> --- a/net/xfrm/xfrm_compat.c
> +++ b/net/xfrm/xfrm_compat.c
> @@ -302,7 +302,7 @@ static int xfrm_xlate64(struct sk_buff *dst, const struct nlmsghdr *nlh_src)
>  	nla_for_each_attr(nla, attrs, len, remaining) {
>  		int err;
>  
> -		switch (type) {
> +		switch (nlh_src->nlmsg_type) {
>  		case XFRM_MSG_NEWSPDINFO:
>  			err = xfrm_nla_cpy(dst, nla, nla_len(nla));
>  			break;