ifmon carrier up/down detection insufficient about minibase HOT 6 CLOSED

Could you try running it with this patch?
https://gist.github.com/arsv/ba776748c44e5763af2ff9b93fe70b6f

ifi->ifi_flags & IFF_RUNNING) ? "Up" : "Down"

That's exactly what ifmon monitors:

Also afaik there's no such thing as cable detection, at least at userspace level. It only detects carrier. The cable is unplugged, or cable is plugged but the router is off — it's the same state, "no-carrier".

The problem is not that it cannot detect router-off condition, it can. Rapid changes in carrier state are most likely the problem.

from minibase.

The situation in the first post, with cable plugged in and router turned off. 'ifmon' with the traces patch is running. Now turn on the router:

carrier up on eth0
carrier rising edge
fork 5365 conf-request eth0
+++ conf-request eth0
dhcp: discover: ETIMEDOUT
exit 5365 65280
carrier down on eth0
carrier falling edge
carrier up on eth0
carrier rising edge
fork 5427 conf-request eth0
+++ conf-request eth0
dhcp: discover: ETIMEDOUT
exit 5427 65280

...interesting, is is calling conf-request. Now, unplug the cable:

carrier down on eth0
carrier falling edge

Now plug in the cable:

carrier up on eth0
carrier rising edge
fork 6286 conf-request eth0
+++ conf-request eth0
+++ dhcp-gw 192.168.1.1
+++ dhcp-dns 192.168.43.1 168.126.63.2
exit 6286 0
assuming discontinuous mode dhcp

...and now online!

from minibase.

Ah, I see. The router brings up carrier early but cannot respond to DHCP requests until much later in its boot process. The initial configuration attempt times out before the router boots up, and ifmon never re-tries.

So my current idea on this is to increase re-send timeout and the number of allowed re-sends for dhcp, let it try for several seconds maybe, and also probably add a command to ifctl to force a re-try. Neither is a good solution, but I don't think there's any way for ifmon to know that the router is finally ready.

This may be a valid reason to introduce per-profile dhcp scripts, at least as an option.

from minibase.

Yes, it will need a fix, even if not optimal, as I think this is probably a common situation, the user forgetting to turn on the modem/router, or having to reset it.

from minibase.

I have a workaround, not elegant, but it works. In /etc/net/conf-request, I inserted a call to script /sbin/dhcp-delayed:

#!/bin/sh
IF="$1"
echo "+++ conf-request ${IF}"
dhcp-delayed ${IF} & #fallback
dhcp ${IF} request

Content of /sbin/dhcp-delayed:

#!/bin/sh
IF="$1"
PIDS=$(pidof dhcp-delayed | wc -w) #get 2 when only one running.
[ $PIDS -gt 2 ] && exit
sleep 40
DNS="$(dhcp ${IF} show | grep '^dns ')"
[ "$DNS" ] && exit
echo "Attempting delayed execution: dhcp ${IF} request"
dhcp $IF request

Pretty simple, dhcp-delayed waits long enough so that all the carrier up/downs have finished, and if the "dhcp $IF request" has not succeeded, does it again.

I don't know if anything like this can be put into ifmon, but anyway, the script works for me. 40 seconds is just long enough, might be better to increase it a bit.

from minibase.

Currently dhcp keeps re-trying for 5s. Which is about as high as it should be for this mode of operation. I thought it was lower but turns out I changed the timeout some time ago. Timeouts around 40s call for a different solution. A script like this is very host-specific, I cannot just hard-code some large time constant and expect it to work well for everyone.

I checked a couple of other implementations, dhcpcd retries for 30s by default and udhcpd from Busybox for 20s. That's already high enough to be configurable (prompting per-profile scripts) yet probably too low to handle your case.

Not sure what's the proper solution here. dhcp is designed around the assumption that it can be a short-running tool. And that should be enough for the vast majority of cases, per my understanding of the DHCP protocol. DHCP makes no provisions for servers suddenly appearing long after the clients sent all their requests.

Your script is actually a good idea, but it can be simplified further.
conf-request:

#!/bin/sh
dhcp-delayed $1

dhcp-delayed:

#!/bin/sh
dhcp $1 request && exit 0
sleep 40
dhcp $1 request

Your own custom double-request dhcp client.

from minibase.