netdev - Re: [PATCH net v2] net: Reset MAC header for direct packet transmission

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 29 Mar 2021 10:51:06 +0200
From:   Eric Dumazet <edumazet@...gle.com>
To:     Kurt Kanzenbach <kurt@...utronix.de>
Cc:     "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Andrii Nakryiko <andriin@...com>,
        Antoine Tenart <atenart@...nel.org>,
        Wei Wang <weiwan@...gle.com>,
        Cong Wang <cong.wang@...edance.com>,
        Taehee Yoo <ap420073@...il.com>,
        Jesper Dangaard Brouer <brouer@...hat.com>,
        Sebastian Andrzej Siewior <bigeasy@...utronix.de>,
        Willem de Bruijn <willemdebruijn.kernel@...il.com>,
        netdev <netdev@...r.kernel.org>
Subject: Re: [PATCH net v2] net: Reset MAC header for direct packet transmission

On Mon, Mar 29, 2021 at 9:17 AM Kurt Kanzenbach <kurt@...utronix.de> wrote:
>
> Reset MAC header in case of using dev_direct_xmit(), e.g. by specifying
> PACKET_QDISC_BYPASS. This is needed, because other code such as the HSR layer
> expects the MAC header to be correctly set.
>
> This has been observed using the following setup:
>
> |$ ip link add name hsr0 type hsr slave1 lan0 slave2 lan1 supervision 45 version 1
> |$ ifconfig hsr0 up
> |$ ./test hsr0
>
> The test binary is using mmap'ed sockets and is specifying the
> PACKET_QDISC_BYPASS socket option.
>
> This patch resolves the following warning on a non-patched kernel:
>
> |[  112.725394] ------------[ cut here ]------------
> |[  112.731418] WARNING: CPU: 1 PID: 257 at net/hsr/hsr_forward.c:560 hsr_forward_skb+0x484/0x568
> |[  112.739962] net/hsr/hsr_forward.c:560: Malformed frame (port_src hsr0)
>
> The MAC header is also reset unconditionally in case of PACKET_QDISC_BYPASS is
> not used at the top of __dev_queue_xmit().
>
> Fixes: d346a3fae3ff ("packet: introduce PACKET_QDISC_BYPASS socket option")
> Signed-off-by: Kurt Kanzenbach <kurt@...utronix.de>
> ---
>
> Changes since v1:
>
>  * Move skb_reset_mac_header() to __dev_direct_xmit()
>  * Add Fixes tag
>  * Target net tree
>
> Previous versions:
>
>  * https://lkml.kernel.org/netdev/20210326154835.21296-1-kurt@linutronix.de/
>
> net/core/dev.c | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/net/core/dev.c b/net/core/dev.c
> index b4c67a5be606..b5088223dc57 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -4297,6 +4297,8 @@ int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
>                      !netif_carrier_ok(dev)))
>                 goto drop;
>
> +       skb_reset_mac_header(skb);
> +
>         skb = validate_xmit_skb_list(skb, dev, &again);
>         if (skb != orig_skb)
>                 goto drop;
> --
> 2.20.1
>

Note that last year, I addressed the issue differently in commit
96cc4b69581db68efc9749ef32e9cf8e0160c509
("macvlan: do not assume mac_header is set in macvlan_broadcast()")
(amended with commit 1712b2fff8c682d145c7889d2290696647d82dab
"macvlan: use skb_reset_mac_header() in macvlan_queue_xmit()")

My reasoning was that in TX path, when ndo_start_xmit() is called, MAC
header is essentially skb->data,
so I was hoping to _remove_ skb_reset_mac_header(skb) eventually from
the fast path (aka __dev_queue_xmit),
because most drivers do not care about MAC header, they just use skb->data.

I understand it is more difficult to review drivers instead of just
adding more code in  __dev_direct_xmit()

In hsr case, I do not really see why the existing check can not be
simply reworked ?

mac_header really makes sense in input path, when some layer wants to
get it after it has been pulled.

diff --git a/net/hsr/hsr_forward.c b/net/hsr/hsr_forward.c
index ed82a470b6e154be28d7e53be57019bccd4a964d..cda495cb1471e23e6666c1f2e9d27a01694f997f
100644
--- a/net/hsr/hsr_forward.c
+++ b/net/hsr/hsr_forward.c
@@ -555,11 +555,7 @@ void hsr_forward_skb(struct sk_buff *skb, struct
hsr_port *port)
 {
        struct hsr_frame_info frame;

-       if (skb_mac_header(skb) != skb->data) {
-               WARN_ONCE(1, "%s:%d: Malformed frame (port_src %s)\n",
-                         __FILE__, __LINE__, port->dev->name);
-               goto out_drop;
-       }
+       skb_reset_mac_header(skb);

        if (fill_frame_info(&frame, skb, port) < 0)
                goto out_drop;