[Dnsmasq-discuss] Bug with EDNS packet size and DNS server files

[Scott Bonar]

All,

I believe I have stumbled on a bug in which the server->edns_pktsz field is not initialized to the daemon->edns_pktsz value
If they are loaded from a server file.  I believe this is because when read_opts() is called the servers_file option is parsed
And loaded into daemon->servers_file, but not read.  After all the options have been parse in read_opts, it then loops
Through all the servers and initializes edns_pktsz to daemon->edns_pktsz, but because the server file has not been
Read yet they are not initialize.  The server file is read later on in the async_event() function.

The one option that I have tested is in the function check_servers(), which called after the read_servers_file(), is as it
loops through the servers, check if edns_pktsz is 0 and if it is initialize it to daemon->edns_pktsz.

The best way I have found to test this is a) use the servers file as defined, b) use the add-subnet option which adds
An EDNS optional record to the DNS request, and c) disable the cache.  Then wireshark the DNS traffic.
You will see the first request has the EDNS packet length set to 0.  Some servers do not like this and return a SERVFAIL,
which triggers a resend, at which point the EDNS packet length is 1024 and the request is successful.

As you can see this is not optimal.

Thoughts? 


Scott Bonar | Cradlepoint 
o: +1.208.489.0029 | sbonar at cradlepoint.com