Received-SPF: pass (google.com: domain of linux-kernel+bounces-19398-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
Date: Mon, 8 Jan 2024 12:15:04 +0100
From: Florian Westphal <fw@strlen.de>
To: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
Cc: "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,
	kernel@openvz.org
Subject: Re: [PATCH] neighbour: purge nf_bridged skb from foreign device neigh
Message-ID: <20240108111504.GA23297@breakpoint.cc>
References: <20240108085232.95437-1-ptikhomirov@virtuozzo.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20240108085232.95437-1-ptikhomirov@virtuozzo.com>
User-Agent: Mutt/1.10.1 (2018-07-13)

Pavel Tikhomirov <ptikhomirov@virtuozzo.com> wrote:
> An skb can be added to a neigh->arp_queue while waiting for an arp
> reply. Where original skb's skb->dev can be different to neigh's
> neigh->dev. For instance in case of bridging dnated skb from one veth to
> another, the skb would be added to a neigh->arp_queue of the bridge.
> 
> There is no explicit mechanism that prevents the original skb->dev link
> of such skb from being freed under us. For instance neigh_flush_dev does
> not cleanup skbs from different device's neigh queue. But that original
> link can be used and lead to crash on e.g. this stack:
> 
> arp_process
>   neigh_update
>     skb = __skb_dequeue(&neigh->arp_queue)
>       neigh_resolve_output(..., skb)
>         ...
>           br_nf_dev_xmit
>             br_nf_pre_routing_finish_bridge_slow
>               skb->dev = nf_bridge->physindev
>               br_handle_frame_finish
> 
> So let's improve neigh_flush_dev to also purge skbs when device
> equal to their skb->nf_bridge->physindev gets destroyed.

Can we fix this by replacing physindev pointer with plain
ifindex instead?  There are not too many places that need to
peek into the original net_device struct, so I don't think
the additional dev_get_by_index_rcu() would be an issue.