From: Sakari A. <sak...@ko...> - 2003-06-18 16:50:57
|
Hi, I have a problem where packet delivery just stops while UML (may) still be running otherwise fine. There's an instance of UML running at my home computer (connected with slow ADSL link) and I connect to it using ssh -C from work. Once it only needed pressing enter key for a while to kill networking. What happens is that ssh connection freezes and no further connections are possible (I guess I would get "connection refused" message if it was sending port unreachable ICMP messages). It however still replies to ping and all the processes are there. I've also tried to use ssh to connect to university machines from UML, and reading news with slrn similarly kills packet delivery rather reliably. It's a bit tricky to debug since I haven't been able to reproduce this reliably from home where I would have ability to restart it... :-I Nothing shows up in logs (nor at UML or host side tty log). Has anyone else seen this? UML version is 2.4.20-5um and it's running on Debian Woody installation using kernel 2.4.21 (same problems with 2.4.20 anyway). -- Sakari Ailus sak...@ko... |
From: Jeff D. <jd...@ad...> - 2003-06-18 19:13:35
|
sak...@ko... said: > What happens is that ssh connection freezes and no further > connections are possible (I guess I would get "connection refused" > message if it was sending port unreachable ICMP messages). It however > still replies to ping and all the processes are there. The first thing I would try is tcpdump the host and UML sides of the tap device (or whatever you're using) and see if the same packets are being seen at both ends. Jeff |
From: Sakari A. <sak...@lu...> - 2003-08-24 10:26:37
|
Jeff Dike wrote: >>What happens is that ssh connection freezes and no further >>connections are possible (I guess I would get "connection refused" >>message if it was sending port unreachable ICMP messages). It however >>still replies to ping and all the processes are there. > The first thing I would try is tcpdump the host and UML sides of the tap > device (or whatever you're using) and see if the same packets are being > seen at both ends. I haven't been able to repeat the problem reliably, but there's another which is similar and easier to repeat, which is that name server queries often seem to crash UML. When I run "host retiisi" (retiisi is the name server, I guess it doesn't matter) I often get "sleeping process x got unexpected signal : 11" message to console and UML crashes. It's also necessary to set tun device down and up for it to work again. I haven't been able to set up debugging so I could check where it crashes. I'm now running Debian Woody (3.0) in UML and AFAIR it didn't happen with Potato (2.2). Both host and UML kernel are compiled with GCC version 3.3. -- Sakari Ailus sak...@lu... |
From: Jeff D. <jd...@ad...> - 2003-08-26 21:07:06
|
sak...@lu... said: > When I run "host retiisi" (retiisi is the name server, I guess it > doesn't matter) I often get "sleeping process x got unexpected signal > : 11" message to console and UML crashes. Can I get access to a UML to which this has happened? Failing that, can you make available a filesystem that can reproduce this? Jeff |
From: Sakari A. <sak...@lu...> - 2003-08-27 18:19:32
|
Jeff Dike wrote: > sak...@lu... said: >>When I run "host retiisi" (retiisi is the name server, I guess it >>doesn't matter) I often get "sleeping process x got unexpected signal >>: 11" message to console and UML crashes. > Can I get access to a UML to which this has happened? > Failing that, can you make available a filesystem that can reproduce this? Now this is funny. You told that jail option isn't supported anymore. But I was still using it on 2.4.20... Just before sending this message, I checked if it still crashed without, and it didn't. X-) Oh well... Are there others who are having these problems still using it? I don't know yet if other less easily reproducible problems have gone away with this change. Anyway, I already made a small file system available here (tested crashing, too). Though this might not be useful anymore... http://retiisi.dyndns.org/~sakke/foo/uml/ bar.bz2 - compressed reiserfs root file system config - .config linux - UML 2.4.20-5um, compiled in Debian 3.0 with GCC-3.3 It's a bit broken now but works well enough for testing. Others may also want to try... I hope not very many though, these files are on my home computer which only has 512 kb/s ADSL link... ;) ;) Anyway, to start: ./linux jail init=/bin/bash ubd0=bar eth0=tuntap,tapX To crash, run: ifconfig eth0 IP_ADDRESS_HERE host retiisi.dyndns.org NAMESERVER It might be necessary to run host several times before it happens. Shorter names seem to be better... -- Sakari Ailus sak...@lu... |
From: Sakari A. <sak...@ko...> - 2003-12-13 12:05:20
|
Sakari Ailus wrote: > I don't know yet if other less easily reproducible problems have gone > away with this change. Apparently my UML instance running 2.6.0-test5-um1 lost its connectivity to outside world via network... Host was running kernel 2.4.22. On host side I set the tap device down and up again, after which it started working again. I don't know whether UML receives packets or is it the host which doesn't. Anyway, I don't get any answers to ping nor I can log in via ssh. This happened when the UML had been up for some three months, so apparently it's not very easy to reproduce. Currently I don't have time to investigate much further, I still have my thesis to write... To Alan: The UML had 24 MB of memory reserved for it. It's file systems and swap are directly on block device, COW is not in use. I think that earlier, when I had slower Internet connection (256/128 kb/s -> 512 kb/s), slower host machine (K6-2+, 512 MB RAM; now 2 GHz Athlon XP, 512 MB RAM) and was using jail mode, this happened a lot more often. -- Sakari Ailus sak...@ko... |