Re: vsock support for communication between guests

[Robert Hoo <robert.hoo.linux@gmail.com>]

On 11/10/2025 7:10 PM, Stefano Garzarella wrote:
> On Sun, 9 Nov 2025 at 14:13, Robert Hoo <robert.hoo.linux@gmail.com> wrote:
>>
>> On 11/6/2025 10:32 PM, Stefano Garzarella wrote:
>>> On Fri, Oct 10, 2025 at 09:00:21PM +0800, Robert Hoo wrote:
>>>> Hi,
>>>>
>>>> Does vsock support communication between guests?
>>>>  From man page, and my experiment, seems it doesn't.
>>>> But why not?
>>>>
>>>
>>> It depends, vhost-user vsock device, supports it.
>>> See
>>> https://github.com/rust-vmm/vhost-device/tree/main/vhost-device-vsock#sibling-vm-communication
>>>
>>> The vhost-vsock in-kernel device doesn't support it.
>>>
>>> The main problem is that vsock is designed for host<->guest communication, so
>>> implementing a guest<->guest communication is possible, but requires more
>>> configuration (e.g. some kind of firewall, etc.) and also an extension to the
>>> address (see the required
>>> `.svm_flags = VMADDR_FLAG_TO_HOST` in the link).
>>>
>>> The easy way to do that with vhost-vsock, is to use socat in the host to
>>> concatenate 2 VMs (some examples here:
>>> https://stefano-garzarella.github.io/posts/2021-01-22-socat-vsock/)
>>>
>>> Cheers,
>>> Stefano
>>>
>> Nice, thanks Stefano. It sounds ideal for my VM <--> VM communication
>> requirement. I'll read the doc carefully later.
>>
>> BTW, I also found your vsock-bridge
>> (https://github.com/stefano-garzarella/vsock-bridge); but seems its last commit
>> was 5 yrs ago. It's not recommended, is it?
>>
> 
> Oh, that was just a little exercise I did to learn Rust at the time,
> so I'd say no, it's not recommended.
> BTW `socat` supports a similar use case, so related to the example in
> the vsock-bridge's README, you can do the following:
> 
> host$ socat VSOCK-LISTEN:5201 VSOCK-CONNECT:4:5201
> vm_cid3$ iperf --vsock -s
> vm_cid4$ iperf --vsock -c 2
> 
> But yeah, it's not 2 ways like vsock-bridge (i.e. `vm_cid3` can't
> connect to `vm_cid4`).
> 
> Cheers,
> Stefano
> 
Thanks Stefano.