[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <1646191a31be339c81f8b335ff40876145a05908.camel@redhat.com>
Date: Mon, 19 Apr 2021 12:00:41 +0200
From: Davide Caratti <dcaratti@...hat.com>
To: "David S. Miller" <davem@...emloft.net>,
Jamal Hadi Salim <jhs@...atatu.com>,
Cong Wang <xiyou.wangcong@...il.com>,
Jiri Pirko <jiri@...nulli.us>,
Jakub Kicinski <kuba@...nel.org>, wenxu <wenxu@...oud.cn>,
Shuang Li <shuali@...hat.com>,
Marcelo Ricardo Leitner <marcelo.leitner@...il.com>,
netdev@...r.kernel.org
Cc: "echaudro@...hat.com" <echaudro@...hat.com>
Subject: Re: [PATCH net] net/sched: sch_frag: fix OOB read while processing
IPv4 fragments
On Fri, 2021-04-16 at 22:29 +0200, Davide Caratti wrote:
[...]
>
> for IPv4 packets, sch_fragment() uses a temporary struct dst_entry. Then,
> in the following call graph:
>
> ip_fragment()
^^ the above line is a typo,
> ip_do_fragment()
> ip_skb_dst_mtu()
> ip_dst_mtu_maybe_forward()
> ip_mtu_locked()
>
> a pointer to that struct is casted as pointer to struct rtable, hence the
> OOB stack access. Fix this, changing the temporary variable used for IPv4
> packets in sch_fragment(), similarly to what is done for IPv6 in the same
> function.
and thanks to Eelco's help I just reproduced a similar splat with
openvswitch. Indeed, ovs_fragment() seems to have the same problem [1];
I will follow-up with a series that fixes both data-paths.
--
davide
[1] https://elixir.bootlin.com/linux/v5.12-rc8/source/net/openvswitch/actions.c#L813
Powered by blists - more mailing lists