Description of problem:
We originally detected this when a customer of ours had a broken firewall between two servers 
trafficing with eachother using SNMP. If the SNMP datagrams were large enough they got fragmented 
and the firewall code corrupted the the udp checksum. The asynchronous code receiving the snmp 
datagrams then blocked forever which wasn't really supposed to happen at all. Looking at the original 
problem with tcpdump we saw that we get both of the datagrams but in the application code never get 
back from a recvmsg() or similar system call. Some googling around and up turns a bug that got fixed 
in 2.6.10, after RHEL4 branched.

I've attached the patches here for completeness and they are as close to the upstream patches as 
possible.

Patch #1 fixes the problem with handling of udp checksum validation
Patch #2 fixes a problem with SOCK_RAW using the code in Patch #1 which breaks someone's pptp 
server among other things according to the urls. We don't use this code but it is included here for 
completeness.

For more information on Patch #1 see, http://permalink.gmane.org/gmane.linux.kernel.commits.head/
44781
For more information on Patch #2, see http://www.gatago.com/linux/kernel/15477600.html


Version-Release number of selected component (if applicable):
2.6.9-42.0.3
  


Additional info:
It would be nice if you guys could integrate these into the next RHEL4 kernel release or something.

Thanks.