pf: Further initialize the inner TCP header in pf_test_state_icmp()

[markj]

Summary

Otherwise the pointer &th.th_sum passed to pf_change_ap() points to
uninitialized memory, since the portion of the TCP header that we copy
from the packet doesn't include the initial checksum. This triggers a
KMSAN violation.

I believe the checksum value doesn't actually matter here, so it's safe
to initialize it to zero. That is, I don't think this is a real bug.

[kp]

I see why KMSAN complains, and I agree that'd work to fix it, but I wonder if it wouldn't be better to pass a uint32_t dummy_pc to pf_change_ap() to make it more obvious that we're not manipulating the actual TCP checksum (which may or may not be there in the packet anyway).

[markj]

In D49347#1125439, @kp wrote:

I see why KMSAN complains, and I agree that'd work to fix it, but I wonder if it wouldn't be better to pass a uint32_t dummy_pc to pf_change_ap() to make it more obvious that we're not manipulating the actual TCP checksum (which may or may not be there in the packet anyway).

That's certainly cleaner than what I did, will update.

[markj]

Apply the suggestion of using a dummy checksum value instead.

[kp]

The commit message needs tweaking to reflect this version.
That may already be in your local tree, because I believe git-arc doesn’t update the commit message.

[glebius]

In D49347#1125790, @kp wrote:

The commit message needs tweaking to reflect this version.
That may already be in your local tree, because I believe git-arc doesn’t update the commit message.

Pro-tip: if you change the header of the commit message in git and do exactly same change in the phabricator Web UI to revision title, git-arc would find it.