[ovs-dev] [PATCH RFC v2 1/1] netdev-dpdk: add dpdk vhost-user ports

[Flavio Leitner]

On Fri, 24 Apr 2015 09:24:01 -0400
Thomas F Herbert <thomasfherbert at gmail.com> wrote:

> On 4/24/15 9:01 AM, Flavio Leitner wrote:
> > On Fri, 24 Apr 2015 14:17:17 +0300
> > Panu Matilainen <pmatilai at redhat.com> wrote:
> >
> >> Hi,
> >>
> >> A few comments inline...
> >>
> >> On 04/21/2015 01:10 PM, Ciara Loftus wrote:
> >>> This patch adds support for a new port type to the userspace
> >>> datapath called dpdkvhostuser. It adds to the existing
> >>> infrastructure of vhost-cuse, however disables vhost-cuse ports
> >>> as the default port type, in favour of vhost-user ports. Refer
> >>> to the documentation for enabling vhost-cuse ports if desired.
> >>>
> >>> A new dpdkvhostuser port will create a unix domain socket which
> >>> when provided to QEMU is used to facilitate communication between
> >>> the virtio-net device on the VM and the OVS port on the host.
> >>>
> >>> Signed-off-by: Ciara Loftus <ciara.loftus at intel.com>
> > [...]
> >
> >>> diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
> >>> index f69154b..deb8b83 100644
> >>> --- a/lib/netdev-dpdk.c
> >>> +++ b/lib/netdev-dpdk.c
> >>> @@ -101,8 +101,13 @@ BUILD_ASSERT_DECL((MAX_NB_MBUF /
> >>> ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF))
> >>>
> >>>    #define MAX_PKT_BURST 32           /* Max burst size for RX/TX
> >>> */
> >>>
> >>> +#ifdef VHOST_CUSE
> >>>    /* Character device cuse_dev_name. */
> >>>    char *cuse_dev_name = NULL;
> >>> +#else
> >>> +#define VHOST_USER_PORT_SOCK_PATH "/tmp/%s" /* Socket Location
> >>> Template */
> >> Using /tmp for these seems like asking for trouble to me, how about
> >> somewhere /var/run? At least it should be configurable via cli,
> >> similar to the cuse device name.
> > Why using /tmp would be problematic?
> >
> > Anyway, systemd has the RunTimeDirectory= option to make sure the
> > private directory is created (under /run) when the service is
> > started and that's the default RPM %{_rundir} where the
> > sock.db, .pid files and other OVS files are created. It seems you
> > could get the right path using ovs_rundir().
> Flavio, I agree with Panu. 
> The user should be able to specify the
> path name for the unix domain socket on the vswitchd command line
> preferably or by environment variable. This way the symantics would
> be similar to the specification for the DBSOCK for ovsdb-server with
> a default location in the same place as the ovsdb-server. This would
> make it easier to create an RPM with an install procedure via shell.
> It would also make it easier for troubleshooters and folks working
> from a git source install to look in /usr/local/var/run/openvswitch
> for all the socks and pid files etc.

Tom, I wasn't disagreeing.  I was just curious to find out why
using /tmp would be "asking for trouble".  Of course there is
the inconveniences here and there as already pointed, but it is
still a valid temporary directory.

fbl