[PATCH net] af_packet: fix the tx skb protocol in raw sockets with ETH_P_ALL

[PATCH net] af_packet: fix the tx skb protocol in raw sockets with ETH_P_ALL

[Yoshiki Komachi]

I am using "protocol ip" filters in TC to manipulate TC flower
classifiers, which are only available with "protocol ip". However,
I faced an issue that packets sent via raw sockets with ETH_P_ALL
did not match the ip filters even if they did satisfy the condition
(e.g., DHCP offer from dhcpd).

I have determined that the behavior was caused by an unexpected
value stored in skb->protocol, namely, ETH_P_ALL instead of ETH_P_IP,
when packets were sent via raw sockets with ETH_P_ALL set.

IMHO, storing ETH_P_ALL in skb->protocol is not appropriate for
packets sent via raw sockets because ETH_P_ALL is not a real ether
type used on wire, but a virtual one.

This patch fixes the tx protocol selection in cases of transmission
via raw sockets created with ETH_P_ALL so that it asks the driver to
extract protocol from the Ethernet header.

Signed-off-by: Yoshiki Komachi <komachi.yoshiki@lab.ntt.co.jp>
---
 net/packet/af_packet.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index 8376bc1..9f6170f 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -1852,7 +1852,8 @@ static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
 
 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
 {
-	if (!skb->protocol && sock->type == SOCK_RAW) {
+	if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
+	    sock->type == SOCK_RAW) {
 		skb_reset_mac_header(skb);
 		skb->protocol = dev_parse_header_protocol(skb);
 	}
-- 
1.8.3.1

Re: [PATCH net] af_packet: fix the tx skb protocol in raw sockets with ETH_P_ALL

[Willem de Bruijn]

On Mon, Mar 18, 2019 at 1:41 AM Yoshiki Komachi
<komachi.yoshiki@lab.ntt.co.jp> wrote:
>
> I am using "protocol ip" filters in TC to manipulate TC flower
> classifiers, which are only available with "protocol ip". However,
> I faced an issue that packets sent via raw sockets with ETH_P_ALL
> did not match the ip filters even if they did satisfy the condition
> (e.g., DHCP offer from dhcpd).
>
> I have determined that the behavior was caused by an unexpected
> value stored in skb->protocol, namely, ETH_P_ALL instead of ETH_P_IP,
> when packets were sent via raw sockets with ETH_P_ALL set.
>
> IMHO, storing ETH_P_ALL in skb->protocol is not appropriate for
> packets sent via raw sockets because ETH_P_ALL is not a real ether
> type used on wire, but a virtual one.
>
> This patch fixes the tx protocol selection in cases of transmission
> via raw sockets created with ETH_P_ALL so that it asks the driver to
> extract protocol from the Ethernet header.
>
> Signed-off-by: Yoshiki Komachi <komachi.yoshiki@lab.ntt.co.jp>

Acked-by: Willem de Bruijn <willemb@google.com>

The Fixes tag is missing. Calling sendmsg on a packet socket bound to
ETH_P_ALL goes back a long way. It is a user, not kernel, bug to do
so.

But no more than sending on a socket bound to ETH_P_NONE (0), which
was addressed in commit c72219b75fde ("packet: infer protocol from
ethernet header if unset").

Probing with packet_parse_headers is available only as of commit
75c65772c3 ("net/packet: Ask driver for protocol if not provided by
user") in v5.1-rc1. This fix does not need to go back further than
that, imho.

Re: [PATCH net] af_packet: fix the tx skb protocol in raw sockets with ETH_P_ALL

[David Miller]

From: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
Date: Mon, 18 Mar 2019 10:49:56 -0400

> On Mon, Mar 18, 2019 at 1:41 AM Yoshiki Komachi
> <komachi.yoshiki@lab.ntt.co.jp> wrote:
>>
>> I am using "protocol ip" filters in TC to manipulate TC flower
>> classifiers, which are only available with "protocol ip". However,
>> I faced an issue that packets sent via raw sockets with ETH_P_ALL
>> did not match the ip filters even if they did satisfy the condition
>> (e.g., DHCP offer from dhcpd).
>>
>> I have determined that the behavior was caused by an unexpected
>> value stored in skb->protocol, namely, ETH_P_ALL instead of ETH_P_IP,
>> when packets were sent via raw sockets with ETH_P_ALL set.
>>
>> IMHO, storing ETH_P_ALL in skb->protocol is not appropriate for
>> packets sent via raw sockets because ETH_P_ALL is not a real ether
>> type used on wire, but a virtual one.
>>
>> This patch fixes the tx protocol selection in cases of transmission
>> via raw sockets created with ETH_P_ALL so that it asks the driver to
>> extract protocol from the Ethernet header.
>>
>> Signed-off-by: Yoshiki Komachi <komachi.yoshiki@lab.ntt.co.jp>
> 
> Acked-by: Willem de Bruijn <willemb@google.com>

Applied.

> The Fixes tag is missing. Calling sendmsg on a packet socket bound to
> ETH_P_ALL goes back a long way. It is a user, not kernel, bug to do
> so.
> 
> But no more than sending on a socket bound to ETH_P_NONE (0), which
> was addressed in commit c72219b75fde ("packet: infer protocol from
> ethernet header if unset").
> 
> Probing with packet_parse_headers is available only as of commit
> 75c65772c3 ("net/packet: Ask driver for protocol if not provided by
> user") in v5.1-rc1. This fix does not need to go back further than
> that, imho.

I added 75c65772c3 as the Fixes: tag, and will not queue this up for
-stable.

Thanks for the analysis.

Re: [PATCH net] af_packet: fix the tx skb protocol in raw sockets with ETH_P_ALL

[Maciej Żenczykowski]

> The Fixes tag is missing. Calling sendmsg on a packet socket bound to
> ETH_P_ALL goes back a long way. It is a user, not kernel, bug to do
> so.

I would argue it's not a bug at all - at least not the way you phrased it.

AFAIK it can/could be done correctly via specifying the protocol in
the sll_protocol field of the struct sockaddr_ll passed to sendmsg as
the target address.

Not specifying it would indeed be a bug (and a common one)
Hence I do agree it's better to automate this.

> But no more than sending on a socket bound to ETH_P_NONE (0), which
> was addressed in commit c72219b75fde ("packet: infer protocol from
> ethernet header if unset").

This is also not a bug.  That's simply how you send when you don't
want to receive.

Other approaches add your socket to the receive hooks and slow the system
down (even if you add a bpf filter to unconditionally drop stuff).

ie. this is the *optimal* way to send frames.