Reuven Plevinsky via dev <[email protected]> writes:
> In case the OVS kernel module is not installed, OVS userspace can
> still be used. Currently its absence blocks OVS startup.
>
> Allow it with a warning. Attempting to create a "system" type
> bridge will fail with "No such file or directory" errors in the
> ovs-vswitchd log.
>
> Signed-off-by: Reuven Plevinsky <[email protected]>
> ---
> utilities/ovs-kmod-ctl.8 | 16 +++++++++++-----
> utilities/ovs-kmod-ctl.in | 7 ++++++-
> 2 files changed, 17 insertions(+), 6 deletions(-)
>
> diff --git a/utilities/ovs-kmod-ctl.8 b/utilities/ovs-kmod-ctl.8
> index c36638e79f52..5acffa931e71 100644
> --- a/utilities/ovs-kmod-ctl.8
> +++ b/utilities/ovs-kmod-ctl.8
> @@ -43,9 +43,14 @@ Each of \fBovs\-kmod\-ctl\fR's commands is described
> separately below.
> .
> .PP
> The \fBinsert\fR command loads the Open vSwitch kernel modules, if
> -needed. If this fails, and the Linux bridge module is loaded but no
> -bridges exist, it tries to unload the bridge module and tries loading
> -the Open vSwitch kernel module again.
> +needed. If the \fBopenvswitch\fR kernel module is not installed on
> +the system, the command continues without it, allowing Open vSwitch
> +to operate in userspace\-only mode.
> +.PP
> +If \fBmodprobe\fR\|(8) fails for an installed module, and the Linux
> +bridge module is loaded but no bridges exist, the command tries to
> +unload the bridge module and retries loading the Open vSwitch kernel
> +module.
> .
> .SH "The ``remove'' command"
> .
> @@ -58,8 +63,9 @@ modules.
> .
> \fBovs\-kmod\-ctl\fR exits with status 0 on success and nonzero on
> failure. The \fBinsert\fR command is considered to succeed if kernel
> -modules are already loaded; the \fBremove\fR command is considered to
> -succeed if none of the kernel modules are loaded.
> +modules are already loaded or if the \fBopenvswitch\fR module is not
> +installed (userspace\-only mode); the \fBremove\fR command is considered
> +to succeed if none of the kernel modules are loaded.
> .
> .SH "ENVIRONMENT"
> .
> diff --git a/utilities/ovs-kmod-ctl.in b/utilities/ovs-kmod-ctl.in
> index 19f1009647e2..daa8c42d4e3d 100644
> --- a/utilities/ovs-kmod-ctl.in
> +++ b/utilities/ovs-kmod-ctl.in
> @@ -29,7 +29,12 @@ done
>
> insert_mods () {
> # Try loading openvswitch kernel module.
> - action "Inserting openvswitch module" modprobe openvswitch
> + modinfo openvswitch > /dev/null 2>&1
> + if test $? = 0; then
> + action "Inserting openvswitch module" modprobe openvswitch
> + else
> + log_warning_msg "No openvswitch module, running in userspace mode
> only"
> + fi
This seems to be a "return 0" effect here to avoid doing the check in
insert_mods_if_required - but that is probably where we should do the
detection that the module is available. So I guess it probably is
better to do it in the `insert_kmods_if_required` function::
if ! modinfo openvswitch > /dev/null 2>&1 ; then
log_warning_msg ....
fi
Also, "running in userspace mode only" isn't exactly correct. Rather,
we know that the vswitchd will not be able to use the netlink datapath,
but I think the message here is misleading. For example, AF_XDP is
kindof a hybrid mode that will still work. I suggest something more
neutral, like "openvswitch module isn't available".
> }
>
> insert_kmods_if_required() {
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
