Provides the module Vnetif
which can be used as a replacement for the regular Netif
implementation in Xen and Unix. Stacks built using Vnetif
are connected to a software switch that allows the stacks to communicate as if they were connected to the same LAN.
An example of a unikernel that communicates with itself over Vnetif
can be seen here. An iperf-like performance test is available here. The examples can be compiled for Unix and Xen and do not need access to a real network interface.
opam install mirage-vnetif
First, construct a TCP/IP stack based on vnetif
:
module S = struct
module B = Basic_backend.Make
module V = Vnetif.Make(B)
module E = Ethif.Make(V)
module I = Ipv4.Make(E)(Clock)(OS.Time)
module U = Udp.Make(I)
module T = Tcp.Flow.Make(I)(OS.Time)(Clock)(Random)
module S = Tcpip_stack_direct.Make(C)(OS.Time)(Random)(V)(E)(I)(U)(T)
include S
end
Since we don't have the mirage-tool to help us we have to construct the stack manually. This code would usually be generated in main.ml
by mirage configure --xen/unix
.
let or_error name fn t =
fn t
>>= function
| `Error e -> fail (Failure ("Error starting " ^ name))
| `Ok t -> return t
let create_stack c backend ip netmask gw =
or_error "backend" S.V.connect backend >>= fun netif ->
or_error "ethif" S.E.connect netif >>= fun ethif ->
or_error "ipv4" S.I.connect ethif >>= fun ipv4 ->
or_error "udpv4" S.U.connect ipv4 >>= fun udpv4 ->
or_error "tcpv4" S.T.connect ipv4 >>= fun tcpv4 ->
let config = {
Mirage_types_lwt.name = "stack";
Mirage_types_lwt.console = c;
Mirage_types_lwt.interface = netif;
Mirage_types_lwt.mode = `IPv4 (ip, netmask, gw);
} in
or_error "stack" (S.connect config ethif ipv4 udpv4) tcpv4
We can now create multiple stacks that talk over the same backend. Basic_backend.create
accepts two optional parameters:
use_async_readers
makes thewrite
calls non-blocking. This is necessary to use Vnetif with the Mirage TCP/IP stack.yield
specifies the yield function to use in non-blocking mode. In a unikernel this is typicallyOS.Time.sleep 0.0
, but in a Unix processLwt_main.yield ()
can be used instead.
let () =
(* create async backend with OS.Time.sleep 0.0 as yield *)
let backend = Basic_backend.create ~use_async_readers:true
~yield:(fun() -> OS.Time.sleep 0.0 ) () in
let netmask = Ipaddr.V4.of_string_exn "255.255.255.0" in
let gw = Ipaddr.V4.of_string_exn "10.0.0.1" in
let server_ip = Ipaddr.V4.of_string_exn "10.0.0.100" in
create_stack c backend server_ip netmask [gw] >>= fun server_stack ->
let client_ip = Ipaddr.V4.of_string_exn "10.0.0.101" in
create_stack c backend server_ip netmask [gw] >>= fun client_stack ->
The stacks can now be used as regular Mirage TCP/IP stacks, e.g.:
S.listen_tcpv4 server_stack ~port:80 (fun f -> ...);
S.listen s1
mirage configure --xen/--unix
make