AF_VSOCK support

[totaam]

When using the ProxyServer, or just when connecting to/from a virtual machine in general, it would be very nice to support VSOCK.
One could also envisage running virtual machines without any networking whatsoever (or even the host, and let the VM have the NIC instead), only the VSOCK would be used for display forwarding.

Links:

• virtio-vsock @ kvm forum 2015
• Add virtio transport for AF_VSOCK

Usage:

qemu ... -device vhost-vsock-pci,id=vhost-vsock-pci0

.svm_family = AF_VSOCK,
.svm_cid = VMADDR_CID_HOST,
.svm_port = 1234,

Not sure about the kernel + python support.

/usr/include/bits/socket.h:#define PF_VSOCK	40	/* vSockets.  */
/usr/include/bits/socket.h:#define AF_VSOCK	PF_VSOCK

This may require cythonizing:

>>> s=socket.socket(40)
>>> s.bind((1234, ))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib64/python2.7/socket.py", line 228, in meth
    return getattr(self._sock,name)(*args)
socket.error: getsockaddrarg: bad family
>>> 

[totaam]

Links:

• Porting Existing Socket Applications
• is there a working vmci example? (for the old vmci interface the new vsock is based on)
• VMCI Sockets Programming Guide
• QEMU Virtio Vsock

[totaam]

2016-04-02 17:48:08: antoine uploaded file vsock.patch (8.1 KiB)

patch attempting to force AF_VSOCK into the python socket class

[totaam]

The vsock bits have been reverted in December 2015: VSOCK: revert virtio-vsock until device spec is finalized, which makes it very hard to test. And it's not immediately obvious because you get this unhelpful error:

qemu-system-x86_64: -device vhost-vsock-pci,id=vh0,addr=4.0,guest-cid=4: vhost-vsock: failed to open vhost device: Unknown error -2

You then have to figure out with strace that the device does not exist..

To fix this, you either have to download the kernel tree they are using (see [http://qemu-project.org/Features/VirtioVsock]), or a newer kernel source and apply RFC,v5,4/5 VSOCK: Introduce vhost_vsock.ko. (for example the Fedora rawhide kernels - the patch applies cleanly there still, bar the maintainers file which can be skipped anyway)

[totaam]

2016-04-04 06:06:54: antoine uploaded file vsock-v3.patch (17.3 KiB)

updated patch

[totaam]

Once I figured out that you MUST use the kernel and qemu versions from [http://qemu-project.org/Features/VirtioVsock] (other combinations may or may not work - this one definitely does, the patched kernel is needed at both ends too), it wasn't too difficult: done in r12309.
It is very fast. The latency goes from ~12ms to 0. Packets come in 256KB chunks at about 1.1GB/s with fairly low CPU usage.

---

Usage:

• on the host, load the vsock module and make it accessible by non-privileged users (better than running qemu as root... will do for now):

modprobe vhost_vsock
chmod 666 /dev/vhost-vsock

• start a session with a vsock bind (here using a specific port, "auto" will assign one):

xpra start :100 --start-child="xterm" --bind-vsock=auto:2000

• start a guest in qemu, use the vhost-vsock option vhost-vsock-pci,id=vhost-vsock-pci0,addr=M.0,guest-cid=N, ie:

/usr/local/bin/qemu-system-x86_64 --enable-kvm -m 2048 -name Fedora-24 \
    -drive file=./Fedora-24-x86_64.raw,if=virtio,index=0,format=raw \
    -device vhost-vsock-pci,id=vhost-vsock-pci0,addr=4.0,guest-cid=4

• in the guest, connect to the vsock using:

modprobe vsock
xpra attach vsock:host:2000

Note: I thought the connections using vsock were from guest to host only, but that is not the case: Re: AF_VSOCK status (will need re-testing).

---

@smo: let's discuss.

Debugging:

• -d network
• test scripts from r12310
• netcat-vsock: [https://github.com/stefanha/linux/blob/vsock-extras/nc-vsock.c]

[totaam]

See also #1159 comment:1 for the new vsock-auth= syntax.

[totaam]

Conversely, you can run a server in the guest and connect to it using the assigned CID.
Re-using the same qemu command line as comment:4 (guest-cid=4).
The guest can start a server:

xpra start :100 --start-child="xterm" --bind-vsock=auto:10000

And we can connect to it from the host using:

xpra attach vhost:4:10000

[totaam]

r12698 adds it to the man page.

[totaam]

VSOCK is included in kernel 4.8, here was the pull request: vhost: new features for 4.8

[totaam]

2016-11-18 00:41:29: smo commented

---

Just a quick update to this old ticket.

Looks like the kernel modules are there now with Fedora 24 which is what I run on my workstation.

After loading vhost_vsock module and trying to make use of it in qemu results in

'vhost-vsock-pci' is not a valid device model name

Does qemu perhaps have to be configured with an option to enable vsock support?

[totaam]

2016-11-18 06:29:54: antoine uploaded file kvm-AF_VSOCK-status.email (0.8 KiB)

email sent to the AF_VSOCK maintainer and kvm ML

[totaam]

Seems that even qemu 2.7 in Fedora 25 does not have vsock builtin.
I've pinged an email to the maintainer. (attached)

[totaam]

2016-11-19 04:26:23: antoine commented

---

Got the answer from the 2.8 release notes [http://wiki.qemu.org/ChangeLog/2.8#virtio]: New device vhost-vsock.

You should be able to test with the 2.8 RC.

[totaam]

Reply: Re: AF_VSOCK status in qemu upstream: you can use qemu.git/master .. I have also made a copr repository for Fedora available here with qemu-kvm, kernel, and other packages with virtio-vsock support : [https://copr.fedorainfracloud.org/coprs/stefanha/vsock/]

[totaam]

This has been tested, see #1387 for a follow up - closing.