Description of problem:

For the latest kernels (2.6.7-+) with ethernet cards such as
those controlled by the e1000 driver, the card (not the kernel)
computes the IP & UDP checksums, so when tcpdump gets the packet,
they will be incorrect, and tcpdump reports a "bad udp cksum"
for the packet, which can be confusing. Bug 129778 was incorrectly
raised as a kernel bug on this issue.

David Miller, the kernel developer I consulted on this, writes:
"
The value in the checksum field is the pseudo-checksum, which
the card uses to compute the final checksum.  It could be
different on every single packet.

The AF_PACKET socket, which tcpdump uses to capture packets,
does have available a TP_STATUS_CSUMNOTREADY flag it can
check for this purpose.  But it is only available when
mmap()'s packet capturing is in use.
"

So tcpdump should be using this flag to check if the checksum
has been computed before complaining about 'bad udp cksum' .

How reproducible:
100% - on every UDP packet

Steps to Reproduce:
1. tcpdump an interface for a card such as Intel 82540EP (e1000
driver) with a 2.6.7+ kernel.
2.
3.
  
Actual results:

All UDP packets show "[bad udp cksum xxxx!]" .

Expected results:

TCP should not show this error message - perhaps
it should show something like "[cksum not ready]" .

Additional info: