LAN interface stops working / "sendto: no buffer space available"

swalter · Aug 21, 2007, 1:21 PM

Hello everyone.

I've been using the 'embedded' version of pfSense on a WRAP for my home DSL router since around January. The WAN interface is configured for PPPoE, for the two LAN interfaces bridging is enabled. There are some NAT rules for incoming traffic like SSH and HTTPS, the DNS forwarder, DHCP server and NTP server are running. I have updated the box from 1.0.1 to 1.2-RC1 a couple of weeks ago, but I keep seeing the following problem which has been there practically all the time.

Occasionally, one of the WRAP's LAN interfaces stops working. SSH and web interface are unreachable, pinging the box isn't possible, either. Usually, this happens at a point of time when I don't really have time to investigate, so I often just pull the power cable and reboot the WRAP. But when I grab my nullmodem cable, go down into the cellar where the box is hanging and connect to it, I can see the system is up and alive. Trying to ping a computer that should be reachable via the non-working LAN interface results in "ping: sendto: no buffer space available".

When I last had this problem, the single computer that is attached to the WRAP's second LAN interface was off. I wanted to see if the other interface wasn't working, either, so I turned the computer on. Just doing that seemed to have a positive effect on the first LAN interface (or the bridge as a whole), since I was suddenly able to ping the hosts at the former again. The same happened when, another time, I merely moved the cable from the problematic interface to the other one and back - suddenly, the interface worked again.

Unfortunately, I only see this problem sporadically and cannot actively reproduce it. Maybe someone has an idea what might be the cause of this? If any specific information is needed, let me know.

Stefan Walter

lsf · Aug 29, 2007, 1:55 AM

Try to edit /boot/loader.conf and add the line

kern.ipc.nmbclusters="0"

swalter · Aug 30, 2007, 5:59 PM

Thanks for the hint. It'll take a while to see if it made a change - as mentioned already, the problem occurred only sporadically.

Regards,
Stefan

swalter · Sep 9, 2007, 3:23 PM

Given that the problem hasn't occured again since I changed that sysctl, I suspect that fixed it. If not, I'll post another reply.

Thanks again,
Stefan

bill_mcgonigle · Sep 10, 2007, 4:13 PM

I'm seeing the same here … would this be worth adding to the LiveCD and embedded bootloader options before it goes final?

I think this is it:

http://www.freebsd.org/releases/6.2R/errata.html
"[20070116, update 20070212] Systems with very heavy network activity have been observed to have some problems with the kernel memory allocator. Symptoms are processes that get stuck in zonelimit state, or system livelocks."
"A bugfix has been committed to HEAD and has been merged to RELENG_6."

which sounds like it would affect pfSense boxes at a disproportionately high level (compared with all the BSD boxes out there).

sullrich · Sep 10, 2007, 4:15 PM

Looks like it is time for the installer to include that value.

bill_mcgonigle · Sep 11, 2007, 6:02 PM

@sullrich:

Looks like it is time for the installer to include that value.

Sorry if this is a dumb question, I didn't find the answer in the docs. Is there a facility to execute an arbitrary command on startup such that I could put a sysctl call in the .xml file for a liveCD version?

I think on a non-LiveCD version this could be kludged with a cron entry.

sullrich · Sep 11, 2007, 7:01 PM

Future livecd's will have the value already built in as I commited the changes to CVS.

bill_mcgonigle · Sep 11, 2007, 7:03 PM

@sullrich:

Future livecd's will have the value already built in as I commited the changes to CVS.

Beauty, thanks!

sullrich · Sep 11, 2007, 7:04 PM

Check the snapshot server in a bit and the new ISO should include the changes.