[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Possible kernel bug in tcp.c? - How to catch kernel crash.



How can I catch, or otherwise make the kernel dump a crash log or core dump somewhere?

Do I need to alter these at all?

kernel.core_pattern = core
kernel.core_setuid_ok = 0
kernel.core_uses_pid = 1


-n.

Naoki wrote:
Ohh, should mention this is only happening (so far) to my eepro100 machines, the tg3 boxes are ok.
Will update of that ceases to be the case.
-n.

Naoki wrote:
That's bizarre. 

These machines that are dying are only running HTTP traffic.  Everything else is firewalled off with iptables and no other services are running.
It's occurring daily and really worrying me to the point of reverting to 2.4.22.

Should I attach gdb and try to grab some more info?

-n.

xyzzy hotpop com wrote:
On Monday 31 May 2004 11:15, Naoki wrote:
  
I've got machines crashing under FC2 with kernel 2.6.5-1.358.  Couldn't
catch the exact kernel messages sorry but here are the rough details.

assertion failed kernel messages.

net/ipv4/tcp.c:1568
net/ipv4/tcp.c:1632

The ethernet driver is :

e100: Intel(R) PRO/100 Network Driver, 3.0.17

Crashing ( and hanging with these messages scrolling at a hefty rate)
alsmost daily because of it.
No problem before with FC1 and 2.4 kernel.

Any ideas?
    

Is this failing on a SAMBA connection?

>From what I can see in the code, it is a BUG_TRAP on the MSG_PEEK flag being 
set by the time it reaches the tcp code (line 1632 also traps if MSG_TRUNC is 
set, but I think that the trap at 1568 for MSG_PEEK alone is the relevant 
one).  SMB, CIF, and NCP are the only network file systems that seem to set 
this flag.  Also, if you are running on a Sparc64/Solaris.

Looks like some sort of weird networking problem in the 2.6 kernel.


  



[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]