> Ah yes, I recall thinking that your patch should go in, but I thereafter
> forgot about it. I've gone ahead and committed your change since it was
> very simple and similar to what we used to do in 2.5.5, only with a
> better heuristic. I'm planning to look into Thorild Selen's patch as--text follows this line--
> well (to see if it is still needed), but I haven't had a chance yet.
It's not enough just checking the last char to see whether something
is an address:
* The letters a-f are valid last characters of hostnames as well as of
IPv6 addresses. Just testing the last char is not enough, at least
not in this case.
* A machine name may possibly end with a digit. Perhaps not likely on
_the_ internet if we assume it to be an FQDN, but maybe on a private
network, for example.
Alternative heuristics:
* If the string contains a colon, then it's an address. (IPv6)
* Otherwise, it's an address only if it consists of nothing but dots
and digits. (IPv4)
* Otherwise, it can't be an address.
This can simply be implemented as:
/* Fail quietly if tok is a hostname (not an address) */
if (! (strchr(tok, ':') /* IPv6 */
|| (strspn(tok, ".0123456789") == strlen(tok)))) /* IPv4 */
return 0;
(What to do when rsync is configured for IPv4 only, but we are given
an IPv6 address to match against? Just considering it a failed match
would perhaps do.)
Also, please consider introducing the other change in my patch;
i.e. replacing
rprintf(FERROR,"malformed address %s\n", tok);
with
rprintf(FERROR,
"error matching address %s: %s\n",
tok,
gai_strerror(gai));
in match_address. This gives a more descriptive error message when
getaddrinfo fails, instead of just telling us that it failed.
Thorild Selén
Datorföreningen Update / Update Computer Club, Uppsala, SE