Re: CVS commit: src/lib/libc/net

[Steffen Nurpmeso <sdaoden%yandex.com@localhost>]

Erik Fair <fair%netbsd.org@localhost> wrote:
 |So, what did we do by default: allow “_” in hostnames when that’s exp\
 |licitly against standard, or not? I read the patch in the PR and coul\
 |dn’t make heads or tails of it.

fwiw i don't think this is generally true, my own one has
a DNS::verifyName() and that does

                // find label border, if any
#if(_ASSUME_LONG)
                wbuf = s(char*,Mem::Utils::find(_dname, _DOT, _dnamelen));
                llen = (wbuf ? s(ui32,wbuf-_dname) : _dnamelen);
#else
                for(wbuf=_dname, llen^=llen;  llen < _dnamelen;  ++llen)
                        if(*(wbuf++) == _DOT)
                                break;
#endif
                if(!llen || llen > 63)
                        goto jesll;

[and]

                // 1033, 1034, 2929 (3.3): rather anything (printable ASCII)
[these numbers are RFCs, vnm_ is enum DNS::VerifyNameMode:
                vnm_none        = 0,    // boundary checks only
                vnm_boundary    = vnm_none,
                vnm_dname       = 1,    // printable ASCII
                vnm_hostname    = 2,    // hostname acc. RFC 1035 + 1123, 2.1
                vnm_srv         = 3,    // vnm_hostname plus _ label start
]
                case vnm_dname:
                        while(llen--) {
                                ui8     c = s(ui8,*(wbuf++));
                                if(!_IS_ASCII_PRINTABLE(c))
                                        goto jeill;
                        }
                        break;

                // hostname acc. 1035:2.3.1, 1123:2.1
[these numbers are RFCs]
                case vnm_hostname:
                case vnm_srv:
                        // may start with letter or digit (latter 1123, 2.1)
                        if(!_IS_ASCII_ALNUM(wbuf[0]))
                                // t_srv may start with underscore (RFC 2782)
                                if(_vnm != vnm_srv || wbuf[0] != _UNDERSCORE)
                                        goto jeill;
                        // may consist of letters, digits and hyphen
                        while(s(si32,--llen) > 0) {
                                char    c = *(++wbuf);
                                if(c != _HYPHEN && !_IS_ASCII_ALNUM(c))
                                        goto jeill;
                        }
                        // but may end with letter or digit only
                        if(wbuf[0] == _HYPHEN)
                                goto jeill;
                        break;
                } // switch(_vnm)

and then

  ?0[sdaoden@wales ]$ git grep IS_ASCII_PRI master
  master:modules/dns/config.h:#define _IS_ASCII_PRINTABLE(CHAR) ((CHAR) > 0x20 && (CHAR) < 0x7F)

So vnm_srv does regulary allow underscore as in

        /*!
        * \var vnm_srv
        * Boundary check (#vnm_none) plus content verification.
        * This is exactly like #vnm_hostname,
        * but additionally allows a label to start with an underscore ("_").
        * RFC 2782 introduced the use of underscores as a leading character for
        * service and protocol names to reduce accidental clashes,
        * and thus we support them as required.
        * Note however that this check is stupid
        * in that \e all labels are allowed to start with underscores \ldots
        */

Oh, it seems not to be completely correct, unfortunately.  I think
i should stop tracking -d again.

--steffen