Packing overflow / compression bug

[miekg]

Packing overflow / compression bug

[andrewtj]

Might be barking up the wrong tree here but I think the root cause of this might be the packed length of RRs being miscalculated. In this case the length of the NAPTR record is off by a few bytes as it doesn't include the length bytes for Replacements labels. In types.go if you change:

func (rr *NAPTR) len() int {
        return rr.Hdr.len() + 4 + len(rr.Flags) + len(rr.Service) +
                len(rr.Regexp) + len(rr.Replacement) + 1
}

To:

func (rr *NAPTR) len() int {
        return rr.Hdr.len() + 4 + len(rr.Flags) + len(rr.Service) +
                len(rr.Regexp) + len(rr.Replacement) + CountLabel(rr.Replacement)
}

It should pack okay without compression. I'll take another pass at this tomorrow if @miekg doesn't beat me to it.

[andrewtj]

I was barking up the wrong tree! The length bytes would have gone in place of periods in Replacement.

I think I know why swapping in CountLabel() paved over the problem though. The wire format for Flags, Service and Regexp is character-string so there should be an extra byte for each.

[miekg]

The packLen() function returns the wrong length, which in turns means that some (NAPTR seems an obvious choice) RR returns the wrong length. packLen() is kind of a premature optimization in that tries to get the package length without actually packing it. Doing something ridiculous in msg.go L1347:

msg = make([]byte, dns.packLen()+20) 

Fixes the problem.

Actually tracking down which RR is doing this is quite annoying, but I'll check,

[miekg]

@andrewtj you're comment was almost spot on, the string need a lenght byte and the NAPTR's len function didn't take this into account. Fixed this and added a test.

The test program now works OK even with compress=false