On Thu, 2018-11-08 at 16:16 -0500, David P. Reed wrote:
> Hi Dan -
> I tried the rpm -e command and it said I don't have that module
> installed. I did grep the directories (/var/run/NetworkManager and
> /etc/NetworkManager) recursively for the string "carrier" and found
> nothing there. So that seems not to be the problem.
>
> It's clear from the journalctl log entries that the two non-connected
> interfaces are being "auto-activated" over and over by something.
> They move from prepare->config->ip-config.
> Reason is "none", but I think that's normally what goes on.
>
> One personal theory (that I don't know how to test properly) is that
> the RealTek r8169 driver (which seems to have been demonstrating
> problems on Ubuntu in the past few months, on some hardware) is
> somehow saying the ports are coming online or are online when they
> are not. I'm not sure how NetworkManager gets told about such
> transitions, but it might be a problem. I also wonder about chrony
> somehow trying to probe the interfaces, triggering them by accident.
NetworkManager listens to kernel messages from the driver. Run "ip
link show dev <dev name>" which should look similar to this:
2: enp0s31f6: <NO-CARRIER,BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500
qdisc fq_codel state DOWN mode DEFAULT group default qlen 1000
link/ether aa:bb:cc:dd:ee:ff brd ff:ff:ff:ff:ff:ff
if you see LOWER_UP that means the kernel driver thinks there is a
carrier/link. At that point it's a kernel driver bug if nothing is
plugged into the card's port.
If it doesn't show LOWER_UP, then it's probably some configuration
issue or bug in NM itself.
Dan
> Is there a "no-ignore-carrier" option that would be something to try
> for debugging? or some specific logging option that will tell more
> about why these ports are "auto-activating"?
>
> I have several other Fedora 28 machines running similar
> configurations, but they don't seem to do this. They are different,
> in that their Ethernet hardware is different, but the software setup
> is roughly similar.
>
> Anyway, I'd love to figure this out. I can share the journals if you
> want or get other data, but this level of logging doesn't say very
> much to me other than what I summarize above.
>
> Thanks.
> -----Original Message-----
> From: "Dan Williams" <d...@redhat.com>
> Sent: Wednesday, November 7, 2018 5:55pm
> To: "David P. Reed" <dpr...@deepplum.com>, networkmanager-list@gnome.
> org
> Subject: Re: why does network manager start activating disconnected
> ethernet ports?
>
>
>
> On Wed, 2018-11-07 at 16:23 -0500, David P. Reed wrote:
> > I have a Fedora 28 server in my lab that has multiple network
> > adapters, but only one of them is cabled to a switch. All the
> > software is updated to NetworkManager-1.10.12-1.fc28.
>
> Did you install the Fedora 28 "Server" spin, or something like
> Workstation? Some of the config options are different for different
> variants. That includes installing the "00-server.conf" file into
> /etc/NetworkManager/conf.d for Server installs in some cases.
>
> One thing that does is set "ignore-carrier=*" which, as you may
> guess,
> makes NM ignore the carrier detection of the NIC on all interfaces.
> Mostly for servers with static IPs where connectivity shouldn't
> change
> even if you unplug or trip over the cable.
>
> Try 'rpm -e NetworkManager-config-server' and then restarting NM or
> rebooting and see if that fixes it.
>
> Dan
>
> > However, when I run journalctl -e I typically see pages and pages
> > and
> > pages of DHCPDISCOVER requests on the non-connected ports,
> > constantly
> > retrying. The logs fill very, very fast.
> >
> > I did something yesterday that managed to cause the retrying to
> > stop,
> > leaving the unused ports all in the disconnected state (nmcli
> > device
> > status said disconnected for each one). I thought, hmm... well that
> > problem is now fixed. `journalctl -e` showed everything quiet.
> >
> > Today, I thought I'd check. When I did `journalctl -e` there were
> > no
> > DHCPDISCOVERs issued, and when I did `nmcli device status` the
> > ports
> > were all still "disconnected".
> >
> > But then I did `nmcli con show`. It said the ports were in a
> > disconnected state.
> >
> > However at this point I happened to repeat the `journalctl -e`
> > command, and my goodness - the stream of DHCPDISCOVER requests
> > timing
> > out were a sight to amaze, and the network manager kept
> > transitioning
> > the state of the ports that had no cables correspondingly, trying
> > to
> > get an IP address.
> >
> > This may be my misunderstanding, but the interface knows whether
> > the
> > port is connected to some other port with a cable or if it is not.
> > It
> > seems logical also that merely asking for the connection status
> > shouldn't "turn on" a lot of useless DHCP discovery on disconnected
> > ports.
> >
> > So am I confused or is this a bug?
> >
> > Tell me what you need to find it if so.
> >
> > I'd also like to know how to stop the behavior. I know I could
> > change
> > the ports to not be "auto", but I'd really like to be able to just
> > plug a cable in and start using the port to talk to some other
> > system
> > or switch.
> > _______________________________________________
> > networkmanager-list mailing list
> > networkmanager-list@gnome.org
> > https://mail.gnome.org/mailman/listinfo/networkmanager-list
_______________________________________________
networkmanager-list mailing list
networkmanager-list@gnome.org
https://mail.gnome.org/mailman/listinfo/networkmanager-list
