DOH: improve encoding/decoding and update related unit test #4598
Conversation
increased strictness of QNAME-encoding, adding error detection for empty labels and names longer than the overall limit; avoided treating DNAME as unexpected;
lib/doh.c
Outdated
| (12 /* header */ | ||
| + (hostlen ? (hostlen + ((host[hostlen-1] == '.') ? 0 : 1)) : 0) | ||
| + 1 /* zero-length root label */ | ||
| + 4); /* QTYPE, QCLASS */ |
There was a problem hiding this comment.
I don't think this needs to deal with a zero hostlen and I think maybe the handling of the trailing dot can be moved outside of the main formula to make this more readable. Maybe like:
size_t expected_len;
DEBUGASSERT(hostlen);
expected_len = 12 + 1 + hostlen + 4;
if(host[hostlen-1]!='.')
hostlen++;There was a problem hiding this comment.
-
If libcurl will never need to query for an RRset at the root, supporting zero hostlen is definitely unnecessary. As I don't know about this, I put it in there, so that this corner case, legitimate from a DNS point of view, would be covered.
For avoidance of doubt, I'ld like to see either "No, it's really overkill" or "Oops, good catch" from you, if you wouldn't mind.
-
Not doing all the estimation during initialization certainly improves readability. I had far too much fun with the brackets.
I think you mean
if(host[hostlen-1]!='.')
expected_len++;
|
Thanks! |
DOH: improve encoding/decoding and update related unit test
A number of bugs in functions doh_encode() and doh_decode() (see below) are addressed.
Unit test 1655 for doh_encode() is updated with more thorough tests
and proofs of detection.
Bugs in function doh_encode():
for encoding in the QNAME, resulting in an invalid DNS query message;
the host name has a trailing dot.
Bug in function doh_decode():
Reported-by: @niallor