Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Mon, 26 Jun 2023 17:03:15 +0200
From:   Stefano Garzarella <sgarzare@redhat.com>
To:     Bobby Eshleman <bobbyeshleman@gmail.com>
Cc:     Arseniy Krasnov <oxffffaa@gmail.com>,
        Bobby Eshleman <bobby.eshleman@bytedance.com>,
        Stefan Hajnoczi <stefanha@redhat.com>,
        "Michael S. Tsirkin" <mst@redhat.com>,
        Jason Wang <jasowang@redhat.com>,
        Xuan Zhuo <xuanzhuo@linux.alibaba.com>,
        "David S. Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>,
        Jakub Kicinski <kuba@kernel.org>,
        Paolo Abeni <pabeni@redhat.com>,
        "K. Y. Srinivasan" <kys@microsoft.com>,
        Haiyang Zhang <haiyangz@microsoft.com>,
        Wei Liu <wei.liu@kernel.org>, Dexuan Cui <decui@microsoft.com>,
        Bryan Tan <bryantan@vmware.com>,
        Vishnu Dasa <vdasa@vmware.com>,
        VMware PV-Drivers Reviewers <pv-drivers@vmware.com>,
        Dan Carpenter <dan.carpenter@linaro.org>,
        Simon Horman <simon.horman@corigine.com>, kvm@vger.kernel.org,
        virtualization@lists.linux-foundation.org, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-hyperv@vger.kernel.org,
        bpf@vger.kernel.org
Subject: Re: [PATCH RFC net-next v4 6/8] virtio/vsock: support dgrams
Message-ID: <d53tgo4igvz34pycgs36xikjosrncejlzuvh47bszk55milq52@whcyextsxfka>
References: <20230413-b4-vsock-dgram-v4-0-0cebbb2ae899@bytedance.com>
 <20230413-b4-vsock-dgram-v4-6-0cebbb2ae899@bytedance.com>
 <92b3a6df-ded3-6470-39d1-fe0939441abc@gmail.com>
 <ppx75eomyyb354knfkwbwin3il2ot7hf5cefwrt6ztpcbc3pps@q736cq5v4bdh>
 <ZJUho6NbpCgGatap@bullseye>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Disposition: inline
In-Reply-To: <ZJUho6NbpCgGatap@bullseye>
Precedence: bulk

On Fri, Jun 23, 2023 at 04:37:55AM +0000, Bobby Eshleman wrote:
>On Thu, Jun 22, 2023 at 06:09:12PM +0200, Stefano Garzarella wrote:
>> On Sun, Jun 11, 2023 at 11:49:02PM +0300, Arseniy Krasnov wrote:
>> > Hello Bobby!
>> >
>> > On 10.06.2023 03:58, Bobby Eshleman wrote:
>> > > This commit adds support for datagrams over virtio/vsock.
>> > >
>> > > Message boundaries are preserved on a per-skb and per-vq entry basis.
>> >
>> > I'm a little bit confused about the following case: let vhost sends 4097 bytes
>> > datagram to the guest. Guest uses 4096 RX buffers in it's virtio queue, each
>> > buffer has attached empty skb to it. Vhost places first 4096 bytes to the first
>> > buffer of guests RX queue, and 1 last byte to the second buffer. Now IIUC guest
>> > has two skb in it rx queue, and user in guest wants to read data - does it read
>> > 4097 bytes, while guest has two skb - 4096 bytes and 1 bytes? In seqpacket there is
>> > special marker in header which shows where message ends, and how it works here?
>>
>> I think the main difference is that DGRAM is not connection-oriented, so
>> we don't have a stream and we can't split the packet into 2 (maybe we
>> could, but we have no guarantee that the second one for example will be
>> not discarded because there is no space).
>>
>> So I think it is acceptable as a restriction to keep it simple.
>>
>> My only doubt is, should we make the RX buffer size configurable,
>> instead of always using 4k?
>>
>I think that is a really good idea. What mechanism do you imagine?

Some parameter in sysfs?

>
>For sendmsg() with buflen > VQ_BUF_SIZE, I think I'd like -ENOBUFS

For the guest it should be easy since it allocates the buffers, but for
the host?

Maybe we should add a field in the configuration space that reports some
sort of MTU.

Something in addition to what Laura had proposed here:
https://markmail.org/message/ymhz7wllutdxji3e

>returned even though it is uncharacteristic of Linux sockets.
>Alternatively, silently dropping is okay... but seems needlessly
>unhelpful.

UDP takes advantage of IP fragmentation, right?
But what happens if a fragment is lost?

We should try to behave in a similar way.

>
>FYI, this patch is broken for h2g because it requeues partially sent
>skbs, so probably doesn't need much code review until we decided on the
>policy.

Got it.

Thanks,
Stefano