No Buffer Space Available -> Loose wan connection?
-
What is all this:
miniupnpd[1017]: sendto(udp_notify): No buffer space available
When I get this, my connections on pfsense dies but I can still ping my pfsense box from an external ip. What is this buffer space? Is it because I do not have enough memory? I'm on a Soekris 4801 /w 128mb ram.
-
Look at your states under status > system and let us know how many states you have when this happens if you can replicate the error again.
-
Hi,
When I get this problem, I have around 300 states.
290/10000
Aug 27 21:11:51 last message repeated 32 times
Aug 27 21:10:56 hostapd: ath0: STA 00:13:02:xx:xx:xx IEEE 802.11: deassociated
Aug 27 21:09:51 last message repeated 15 times
Aug 27 21:09:21 miniupnpd[1017]: sendto(udp_notify): No buffer space available -
What version are you running? Make sure that if it has built in uPNP you aren't using the package. If you have a the traffic shaper turned on, try turning it off and see if you get this message with the shaper off. You can get this message because you have an error in your shaper rules. Also execute this command and see how your total compares to the max netstat -m | grep "mbuf clusters".
If you think it is a problem with the memory buffers, you can tweak them like this (untested by myself):
Add to /boot/loader.conf
kern.ipc.nmbclusters="yournumhere" # Set the number of mbuf clusters
kern.ipc.nmbufs="numberabove*4" # Set the number of mbufs = 4 * nmbclustersI don't have many ideas as to what might cause that. You can check if you are running out of memory by using the "top" command in the shell. Just ssh in and, press 8, and type top and press enter. Control c to quit. Are you running torrents when this happens?
-
I have no traffic shaper running but I do have torrents running when this happens. I will look at the memory and maybe try to tweak the parameters you provided.
It's not really the wan connection dying actually. I can't contact the pfsense box from internal network. Ping 192.168.1.1 results in timeouts. Btw, I'm connected wireless though an atheros A card.
-
Also remember to look at UPNP, if it is on, try testing some torrents with it turned off and see if your results differ at all.
-
It looks this this happens with HEAVY traffic (100% cpu) and with MINIUPNP running. I tried the fix in the SVN today (http://cvstrac.pfsense.com/chngview?cn=19407) and it still does not help. Something to do with buffers, memory…
HRM!
-
What are your machine stats? CPU, RAM, etc…
-
It is just a WRAP 1E with 128mb of ram and Soekris net 4801 with 12mb of ram. Same config, two different machines and same problem.
-
Edit /boot/loader.conf
Add to the file:
kern.ipc.nmbclusters="0"
Then save the changes and reboot the firewall.
-
I did put this into the /boot/loader.conf already and it still doesn't work. Once mini-upnp complain about no buffer space, the firewall is dead. I have miniupnp listening on WLAN and not LAN. Any clues?
-
ARGG..this is very annoying. I already stopped mini-upnpd but under heavy traffic, connection cuts out but i'm still connected via wireless. It seems like a memory allocation problem and nmbufs in the loader.conf is not helping! I don't care if this thing net4801 is slower than my linksys wrt54g router but the linksys is VERY stable and pfsense + net4801 is one unstable setup. Any clues on how I can debug this so we can trace down the problem?? Thanks!
-
Your bios is the latest and greatest on the WRAP?
-
On two different boards, althought similar hardware, it acts up on the NET4801 and WRAP. Both are being used as firewall/routers/ap so wan from ethernet and Atheros abg card in the mini-pci slot. Hope someone can come up with more memory tuning parameters.
-
Arg, i'm getting this error again on my via 1ghz eden box still. It's got 512mb of ram and cpu utilization when using dual wan and downloading at 1MB/sec is only 25-40% usage.
Dec 2 21:51:46 miniupnpd[4936]: sendto(udp_notify): No buffer space available
I swear there is a bug but no one seems to be seeing it. I checked and the kernel no mnbclusters is in the loader.conf so that isn't the problem.
PLEASE help!
Thanks!
-
Check /boot/loader.conf
It should contain:
kern.ipc.nmbclusters="0"
-
I do have that and miniupnpd is still crapping out. Makes me think if it is the daemon itself.
Here is my loader.conf:
via:/boot# more loader.conf
hw.ata.atapi_dma="0"
hw.ata.ata_dma="0"
loader_color="NO"
autoboot_delay="5"
hw.ata.wc="0"
kern.ipc.nmbclusters="0"
beastie_disable="YES" -
I do have that and miniupnpd is still crapping out. Makes me think if it is the daemon itself.
Sounds like it. I would bet an application on the network is interacting with it and somehow is triggering a bug that happens to be making FreeBSD run out of resources.
-
It would be helpful if you could track down the application that causes that error. One way might be a tcpdump leading up to and at the time the error. Although I looked at the miniupnpd code and its just sending out a broadcast of miniupnpd's known services. This could be an issue relating to network issues and not the box or miniupnpd. I came across a thread while searching google
http://unix.derkeiler.com/Mailing-Lists/FreeBSD/isp/2003-07/0172.html
-
I got an email back from Thomas the miniupnpd author stating
The error indicate that the send buffer for the network interface is
full. It can only happen when the network interface itself is down or
having problems sending packets.