Hi On Wed, Apr 1, 2020 at 6:40 PM Andrea Bolognani <abologna@xxxxxxxxxx> wrote: > > On Wed, 2020-04-01 at 18:23 +0200, Michal Privoznik wrote: > > + <change> > > + <summary> > > + qemu: Support migration with SLIRP interface > > + </summary> > > + <description> > > + As <code>libslirp</code> evolves, so does QEMU. The recent change is > > + that instead of peer-to-peer connection between QEMU and > > + <code>slirp-helper</code> process a separate D-BUS bus is created. > > + This enables QEMU to migrate with a SLIRP interface. > > + </description> > > + </change> > > This looks reasonable enough to me, so > > Reviewed-by: Andrea Bolognani <abologna@xxxxxxxxxx> > > but let's maybe wait until tomorrow to push, in order to give > Marc-André a chance to speak his mind. Thanks. That's quite accurate, but it may be misinterpreted: Networking data is still going through a 1-1 dgram unix socket. D-Bus bus is used to communicate with the helper for RPC/management, including migration. After discussions, we decided to switch to a bus, as it permits other tools, such as libvirt or others, to interact with the helpers easily. A bus topology reduces the overall complexity for implementation and debugging. This should slowly become the RPC standard for "multi-process qemu" in general. So libvirt will need a single connection to the bus, and can then talk to the various helper processes. Similarly for qemu or third-party (with security restrictions). Since the p2p dbus support was never actually used, I am not sure we need to explain that there was a transition. Let me suggest: qemu: Support migration with SLIRP helper interface With QEMU 5.0, a new D-Bus backend allows migration of external processes. When needed, libvirt will start a per-vm D-Bus bus, and migrate the slirp-helper along with QEMU. my 2c
