Bug: 294236838

Clone this repo:
  1. ac14a1c Update Android.bp by running cargo_embargo by James Farrell · 6 days ago main master
  2. 10ede79 Migrate to cargo_embargo. am: 8f17a51220 am: 0f4e628dca am: 59d0bd7b6b by Andrew Walbran · 7 months ago
  3. 4feb45f Migrate to cargo_embargo. am: 8f17a51220 am: 9a226338a6 am: f1ff62b24c by Andrew Walbran · 7 months ago
  4. 59d0bd7 Migrate to cargo_embargo. am: 8f17a51220 am: 0f4e628dca by Andrew Walbran · 7 months ago
  5. f1ff62b Migrate to cargo_embargo. am: 8f17a51220 am: 9a226338a6 by Andrew Walbran · 7 months ago

vhost-device-vsock

Design

The crate introduces a vhost-device-vsock device that enables communication between an application running in the guest i.e inside a VM and an application running on the host i.e outside the VM. The application running in the guest communicates over VM sockets i.e over AF_VSOCK sockets. The application running on the host connects to a unix socket on the host i.e communicates over AF_UNIX sockets. The main components of the crate are split into various files as described below:

  • packet.rs
    • Introduces the VsockPacket structure that represents a single vsock packet processing methods.
  • rxops.rs
    • Introduces various vsock operations that are enqueued into the rxqueue to be sent to the guest. Exposes a RxOps structure.
  • rxqueue.rs
    • rxqueue contains the pending rx operations corresponding to that connection. The queue is represented as a bitmap as we handle connection-oriented connections. The module contains various queue manipulation methods. Exposes a RxQueue structure.
  • thread_backend.rs
    • Multiplexes connections between host and guest and calls into per connection methods that are responsible for processing data and packets corresponding to the connection. Exposes a VsockThreadBackend structure.
  • txbuf.rs
    • Module to buffer data that is sent from the guest to the host. The module exposes a LocalTxBuf structure.
  • vhost_user_vsock_thread.rs
    • Module exposes a VhostUserVsockThread structure. It also handles new host initiated connections and provides interfaces for registering host connections with the epoll fd. Also provides interfaces for iterating through the rx and tx queues.
  • vsock_conn.rs
    • Module introduces a VsockConnection structure that represents a single vsock connection between the guest and the host. It also processes packets according to their type.
  • vhu_vsock.rs
    • exposes the main vhost user vsock backend interface.

Usage

Run the vhost-device-vsock device:

vhost-device-vsock --guest-cid=<CID assigned to the guest> \
  --socket=<path to the Unix socket to be created to communicate with the VMM via the vhost-user protocol> \
  --uds-path=<path to the Unix socket to communicate with the guest via the virtio-vsock device> \
  [--tx-buffer-size=<size of the buffer used for the TX virtqueue (guest->host packets)>] \
  [--groups=<list of group names to which the device belongs concatenated with '+' delimiter>]

or

vhost-device-vsock --vm guest_cid=<CID assigned to the guest>,socket=<path to the Unix socket to be created to communicate with the VMM via the vhost-user protocol>,uds-path=<path to the Unix socket to communicate with the guest via the virtio-vsock device>[,tx-buffer-size=<size of the buffer used for the TX virtqueue (guest->host packets)>][,groups=<list of group names to which the device belongs concatenated with '+' delimiter>]

Specify the --vm argument multiple times to specify multiple devices like this:

vhost-device-vsock \
--vm guest-cid=3,socket=/tmp/vhost3.socket,uds-path=/tmp/vm3.vsock,groups=group1+groupA \
--vm guest-cid=4,socket=/tmp/vhost4.socket,uds-path=/tmp/vm4.vsock,tx-buffer-size=32768

Or use a configuration file:

vhost-device-vsock --config=<path to the local yaml configuration file>

Configuration file example:

vms:
    - guest_cid: 3
      socket: /tmp/vhost3.socket
      uds_path: /tmp/vm3.sock
      tx_buffer_size: 65536
      groups: group1+groupA
    - guest_cid: 4
      socket: /tmp/vhost4.socket
      uds_path: /tmp/vm4.sock
      tx_buffer_size: 32768
      groups: group2+groupB

Run VMM (e.g. QEMU):

qemu-system-x86_64 \
  <normal QEMU options> \
  -object memory-backend-file,share=on,id=mem0,size=<Guest RAM size>,mem-path=<Guest RAM file path> \ # size == -m size
  -machine <machine options>,memory-backend=mem0 \
  -chardev socket,id=char0,reconnect=0,path=<vhost-user socket path> \
  -device vhost-user-vsock-pci,chardev=char0

Working example

shell1$ vhost-device-vsock --vm guest-cid=4,uds-path=/tmp/vm4.vsock,socket=/tmp/vhost4.socket

or if you want to configure the TX buffer size

shell1$ vhost-device-vsock --vm guest-cid=4,uds-path=/tmp/vm4.vsock,socket=/tmp/vhost4.socket,tx-buffer-size=65536
shell2$ qemu-system-x86_64 \
          -drive file=vm.qcow2,format=qcow2,if=virtio -smp 2 -m 512M -mem-prealloc \
          -object memory-backend-file,share=on,id=mem0,size=512M,mem-path="/dev/hugepages" \
          -machine q35,accel=kvm,memory-backend=mem0 \
          -chardev socket,id=char0,reconnect=0,path=/tmp/vhost4.socket \
          -device vhost-user-vsock-pci,chardev=char0

Guest listening

iperf

# https://github.com/stefano-garzarella/iperf-vsock
guest$ iperf3 --vsock -s
host$  iperf3 --vsock -c /tmp/vm4.vsock

netcat

guest$ nc --vsock -l 1234

host$  nc -U /tmp/vm4.vsock
CONNECT 1234

Host listening

iperf

# https://github.com/stefano-garzarella/iperf-vsock
host$  iperf3 --vsock -s -B /tmp/vm4.vsock
guest$ iperf3 --vsock -c 2

netcat

host$ nc -l -U /tmp/vm4.vsock_1234

guest$ nc --vsock 2 1234

Sibling VM communication

If you add multiple VMs with their devices configured with at least one common group name, they can communicate with each other. If you don't explicitly specify a group name, a default group will be assigned to the device with name default, and all such devices will be able to communicate with each other. Or you can choose a different list of group names for each device, and only devices with the at least one group in commmon will be able to communicate with each other.

For example, if you have two VMs with CID 3 and 4, you can run the following commands to make them communicate:

shell1$ vhost-device-vsock --vm guest-cid=3,uds-path=/tmp/vm3.vsock,socket=/tmp/vhost3.socket,groups=group1+group2 \
          --vm guest-cid=4,uds-path=/tmp/vm4.vsock,socket=/tmp/vhost4.socket,groups=group1
shell2$ qemu-system-x86_64 \
          -drive file=vm1.qcow2,format=qcow2,if=virtio -smp 2 -m 512M -mem-prealloc \
          -object memory-backend-file,share=on,id=mem0,size=512M,mem-path="/dev/hugepages" \
          -machine q35,accel=kvm,memory-backend=mem0 \
          -chardev socket,id=char0,reconnect=0,path=/tmp/vhost3.socket \
          -device vhost-user-vsock-pci,chardev=char0
shell3$ qemu-system-x86_64 \
          -drive file=vm2.qcow2,format=qcow2,if=virtio -smp 2 -m 512M -mem-prealloc \
          -object memory-backend-file,share=on,id=mem0,size=512M,mem-path="/dev/hugepages2" \
          -machine q35,accel=kvm,memory-backend=mem0 \
          -chardev socket,id=char0,reconnect=0,path=/tmp/vhost4.socket \
          -device vhost-user-vsock-pci,chardev=char0

Please note that here the groups parameter is specified just for clarity, but it is not necessary to specify it if you want to use the default group and make all the devices communicate with one another. It is useful to specify a list of groups when you want fine-grained control over which devices can communicate with each other.

# nc-vsock patched to set `.svm_flags = VMADDR_FLAG_TO_HOST`
guest_cid3$ nc-vsock -l 1234
guest_cid4$ nc-vsock 3 1234

License

This project is licensed under either of