Re: [PATCH] virtio_net: set default mtu to 1500 when 'Device maximum MTU' bigger than 1500

From: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
To: "Michael S. Tsirkin" <mst@redhat.com>
Cc: huangml@yusur.tech, zy@yusur.tech,
	Jason Wang <jasowang@redhat.com>,
	"David S. Miller" <davem@davemloft.net>,
	Eric Dumazet <edumazet@google.com>,
	Jakub Kicinski <kuba@kernel.org>, Paolo Abeni <pabeni@redhat.com>,
	"open list:VIRTIO CORE AND NET DRIVERS"
	<virtualization@lists.linux-foundation.org>,
	"open list:NETWORKING DRIVERS" <netdev@vger.kernel.org>,
	open list <linux-kernel@vger.kernel.org>,
	Hao Chen <chenh@yusur.tech>,
	hengqi@linux.alibaba.com,
	Stephen Hemminger <stephen@networkplumber.org>
Subject: Re: [PATCH] virtio_net: set default mtu to 1500 when 'Device maximum MTU' bigger than 1500
Date: Tue, 9 May 2023 09:51:39 +0800	[thread overview]
Message-ID: <1683597099.3423615-2-xuanzhuo@linux.alibaba.com> (raw)
In-Reply-To: <20230508140640-mutt-send-email-mst@kernel.org>

On Mon, 8 May 2023 14:10:07 -0400, "Michael S. Tsirkin" <mst@redhat.com> wrote:
> On Mon, May 08, 2023 at 09:25:48AM -0700, Stephen Hemminger wrote:
> > On Mon, 8 May 2023 06:30:07 -0400
> > "Michael S. Tsirkin" <mst@redhat.com> wrote:
> >
> > > > > > I don't know, in any scenario, when the hardware supports a large mtu, but we do
> > > > > > not want the user to use it by default.
> > > > >
> > > > > When other devices on the same LAN have mtu set to 1500 and
> > > > > won't accept bigger packets.
> > > >
> > > > So, that depends on pmtu/tcp-probe-mtu.
> > > >
> > > > If the os without pmtu/tcp-probe-mtu has a bigger mtu, then it's big packet
> > > > will lost.
> > > >
> > > > Thanks.
> > > >
> > >
> > > pmtu is designed for routing. LAN is supposed to be configured with
> > > a consistent MTU.
> >
> > Virtio is often used with bridging or macvlan which can't support PMTU.
> > PMTU only works when forwarding at layer 3 (ie routing) where there is
> > a IP address to send the ICMP response. If doing L2 forwarding, the
> > only thin the bridge can do is drop the packet.
> >
> > TCP cab recover but detecting an MTU blackhole requires retransmissions.
>
> Exactly. That's why we basically use the MTU advice supplied by device
> by default - it's designed for use-cases of software devices where
> the device might have more information about the MTU than the guest.
> If hardware devices want e.g. a way to communicate support for
> jumbo frames without communicating any information about the LAN,
> a new feature will be needed.

Let's think this question carefully. If necessary, we will try to introduce a
new feature for virtio-net spec to support Jumbo Frame.

Thanks.

>
> --
> MST
>