2.6.9 BUG in tcp_transmit_skb at net/ipv4/tcp_output.c

I've got an error report for a problem with 2.6.9 networking. We seem to be intermittently hitting the above error after at least a couple hours or so of runtime, sometimes it takes up to 14 hrs or so, sometimes it doesn't happen. Network load is minimal.

Anyone have any ideas on what this might be or how I can track it down?
Has it been fixed in later releases?

Thanks for any help you can give,

Chris

Here's the full output of the bug

kernel BUG in tcp_transmit_skb at net/ipv4/tcp_output.c:277!

Oops: Exception in kernel mode, sig: 5 [#1]

SMP NR_CPUS=2
NIP: C00000000033D538 XER: 0000000000000000 LR: C00000000033CAEC
REGS: c00000012d91f410 TRAP: 0700 Not tainted (2.6.9)
MSR: 9000000000029032 EE: 1 PR: 0 FP: 0 ME: 1 IR/DR: 11
TASK: c00000016c8e6ae0[1211] 'peel' THREAD: c00000012d91c000 CPU: 0
GPR00: 0000000000000001 C00000012D91F6E0 C000000000623A60 FFFFFFFFFFFFFF97
GPR04: C000000123582980 0000000000000000 0000000000000000 0000000000000000
GPR08: 0000000000000000 0000000000000000 0000000000000001 C00000000FFC2900
GPR12: C000000123582980 C000000000492400 0000000000000001 000000001C0D9EE8
GPR16: 0000000000000000 0000000000000001 0000000000001015 0000000000009F5C
GPR20: 0000000000000005 0000000000000005 000000000000008F C00000016CA9E3D8
GPR24: 0000000000000020 C0000001235829F0 C00000016CA9E180 C000000123582980
GPR28: C00000016CA9E180 0000000000000218 C000000000530650 C00000016CA9E488 NIP [c00000000033d538] .tcp_transmit_skb+0x7c/0xa44
LR [c00000000033caec] .tcp_rcv_state_process+0xdb4/0x13bc

Call Trace:

[c00000012d91f6e0] [0000000000000028] 0x28 (unreliable) [c00000012d91f800] [c00000000033caec] .tcp_rcv_state_process+0xdb4/0x13bc [c00000012d91f920] [c000000000347190] .tcp_v4_do_rcv+0x164/0x434 [c00000012d91f9f0] [c0000000002fee64] .__release_sock+0x78/0xc4 [c00000012d91fa90] [c0000000002fef38] .release_sock+0x88/0xdc [c00000012d91fb10] [c000000000359cb0] .inet_stream_connect+0x90/0x3ac [c00000012d91fc20] [c0000000002fc9c8] .sys_connect+0x94/0xc8 [c00000012d91fd40] [c000000000319710] .compat_sys_socketcall+0x134/0x2ac [c00000012d91fde0] [c000000000010434] syscall_exit+0x0/0x18 <0>Kernel panic - not syncing: Fatal exception in interrupt <0>Rebooting in 180 seconds..

There's been lots of fixes to the TSO code since 2.6.9. Please test 2.6.11.7 instead.

Cheers,

Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Unfortunately I can't move to the latest. We have a bunch ofkernel modules from various vendors, board support packages that aren't in mainline, etc. We're in the process of moving to 2.6.10 and that requires support from six separate companies. It's a mess.

If it matters, I'm on a G5 with the sungem driver (which I don't think supports TSO). I've already applied the last -ac patch for 2.6.9 which has a number of fixes for 2.6.9 TCP issues, but apparently not all of them were caught.

Any pointers you can give me?

Chris

You could also try going back to 2.6.8.1 since that predates the TSO restructure.

Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt