bpf: fix skip_tunnel_nodeport_revnat

[lmb]

bpf: fix skip_tunnel_nodeport_revnat

The test contains a bug since it passes min_port and max_port in struct
ip{4,6}_nat_target in network instead of host byte order. It doesn't fail
because of a series of "fortunate" circumstances, which boil down to
invoking the following function

    __u16 __snat_try_keep_port(__u16 start, __u16 end, __u16 val)
   {
       return val >= start && val <= end ? val :
            __snat_clamp_port_range(start, end, (__u16)get_prandom_u32());
   }

with the following arguments:

    __snat_try_keep_port(htons(port), htons(port), port)

since htons(port) <= port <= htons(port) is (almost) never true, we end up
invoking

    __u16 __snat_clamp_port_range(__u16 start, __u16 end,  __u16 val)
   {
       return (val % (__u16)(end - start)) + start;
   }

Since start == end, we end up executing val % 0. This seems to cause UB
which somehow leads to the code just using val. Of course, this all comes
crashing down when trying to modify any of the code in this path. Fix the
test by converting to host byte order as appropriate.

Fixes: c1761f75ec ("bpf: Implement handling of flag_skip_tunnel for revNAT
nodeport traffic") Signed-off-by: Lorenz Bauer <lmb@isovalent.com>

[lmb]

cc @learnitall took me a while to figure this out 😆

[lmb]

/test

[lmb]

/test

[lmb]

/test

[learnitall]

Woah, nice catch. That's quite the bug 😲. My apologies for the extra work here, I owe you one.

[aanm]

Changed release note label to misc since this is not a user-facing change.

[lmb]

/test

[ti-mo]

Removing r2m here, looks like ci-e2e is hitting a failure that doesn't resolve on rerun.

[lmb]

/test

[lmb]

/test

[lmb]

/test

[lmb]

/test