fixed normalization for the single-character trailing component

Thanks Hiroyuki ARAKI for spotting the bug and providing the fix

Signed-off-by: Michael Krelin <hacker@klever.net>

1 files changed, 1 insertions, 1 deletions

diff --git a/lib/util.cc b/lib/util.cc
index 249eeed..4028697 100644
--- a/lib/util.cc
+++ b/lib/util.cc
@@ -279,193 +279,193 @@ namespace opkele {
 	 *   that is - [:alpha:][:digit:]._~-
 	 * - remove dot segments
 	 * - remove empty and default ports
 	 * - if there's no path component, add '/'
 	 */
 	 string rfc_3986_normalize_uri(const string& uri) {
 	     string rv;
 	     string::size_type ns = uri.find_first_not_of(data::_whitespace_chars);
 	     if(ns==string::npos)
 		 throw bad_input(OPKELE_CP_ "Can't normalize empty URI");
 	     string::size_type colon = uri.find(':',ns);
 	     if(colon==string::npos)
 		 throw bad_input(OPKELE_CP_ "No scheme specified in URI");
 	     transform(
 		     uri.begin()+ns, uri.begin()+colon+1,
 		     back_inserter(rv), ::tolower );
 	     bool s;
 	     string::size_type ul = uri.find_last_not_of(data::_whitespace_chars)+1;
 	     if(ul <= (colon+3))
 		 throw bad_input(OPKELE_CP_ "Unexpected end of URI being normalized encountered");
 	     if(uri[colon+1]!='/' || uri[colon+2]!='/')
 		 throw bad_input(OPKELE_CP_ "Unexpected input in URI being normalized after scheme component");
 	     if(rv=="http:")
 		 s = false;
 	     else if(rv=="https:")
 		 s = true;
 	     else{
 		 /* TODO: support more schemes.  e.g. xri. How do we normalize
 		  * xri?
 		  */
 		 rv.append(uri,colon+1,ul-colon-1);
 		 return rv;
 	     }
 	     rv += "//";
 	     string::size_type interesting = uri.find_first_of(":/#?",colon+3);
 	     if(interesting==string::npos) {
 		 transform(
 			 uri.begin()+colon+3,uri.begin()+ul,
 			 back_inserter(rv), ::tolower );
 		 rv += '/'; return rv;
 	     }
 	     transform(
 		     uri.begin()+colon+3,uri.begin()+interesting,
 		     back_inserter(rv), ::tolower );
 	     bool qf = false;
 	     char ic = uri[interesting];
 	     if(ic==':') {
 		 string::size_type ni = uri.find_first_of("/#?%",interesting+1);
 		 const char *nptr = uri.data()+interesting+1;
 		 char *eptr = 0;
 		 long port = strtol(nptr,&eptr,10);
 		 if( (port>0) && (port<65535) && port!=(s?443:80) ) {
 		     char tmp[8];
 		     snprintf(tmp,sizeof(tmp),":%ld",port);
 		     rv += tmp;
 		 }
 		 if(ni==string::npos) {
 		     rv += '/'; return rv;
 		 }
 		 interesting = ni;
 	     }else if(ic!='/') {
 		 rv += '/'; rv += ic;
 		 qf = true;
 		 ++interesting;
 	     }
 	     string::size_type n = interesting;
 	     char tmp[3] = { 0,0,0 };
 	     stack<string::size_type> psegs; psegs.push(rv.length());
 	     string pseg;
 	     for(;n<ul;) {
 		 string::size_type unsafe = uri.find_first_of(qf?"%":"%/?#",n);
 		 if(unsafe==string::npos) {
 		     pseg.append(uri,n,ul-n-1); n = ul-1;
 		 }else{
 		     pseg.append(uri,n,unsafe-n);
 		     n = unsafe;
 		 }
 		 char c = uri[n++];
 		 if(c=='%') {
 		     if((n+1)>=ul)
 			 throw bad_input(OPKELE_CP_ "Unexpected end of URI encountered while parsing percent-encoded character");
 		     tmp[0] = uri[n++];
 		     tmp[1] = uri[n++];
 		     if(!( isxdigit(tmp[0]) && isxdigit(tmp[1]) ))
 			 throw bad_input(OPKELE_CP_ "Invalid percent-encoded character in URI being normalized");
 		     int cc = strtol(tmp,0,16);
 		     if( isalpha(cc) || isdigit(cc) || strchr("._~-",cc) )
 			 pseg += (char)cc;
 		     else{
 			 pseg += '%';
 			 pseg += (char)toupper(tmp[0]); pseg += (char)toupper(tmp[1]);
 		     }
 		 }else if(qf) {
 		     rv += pseg; rv += c;
 		     pseg.clear();
 		 }else if(n>=ul || strchr("?/#",c)) {
-		     if(pseg.empty() || pseg==".") {
+		     if( (unsafe!=string::npos && pseg.empty()) || pseg==".") {
 		     }else if(pseg=="..") {
 			 if(psegs.size()>1) {
 			     rv.resize(psegs.top()); psegs.pop();
 			 }
 		     }else{
 			 psegs.push(rv.length());
 			 if(c!='/') {
 			     pseg += c;
 			     qf = true;
 			 }
 			 rv += '/'; rv += pseg;
 		     }
 		     if(c=='/' && (n>=ul || strchr("?#",uri[n])) ) {
 			 rv += '/';
 			 if(n<ul)
 			     qf = true;
 		     }else if(strchr("?#",c)) {
 			 if(psegs.size()==1 && psegs.top()==rv.length())
 			     rv += '/';
 			 if(pseg.empty())
 			     rv += c;
 			 qf = true;
 		     }
 		     pseg.clear();
 		 }else{
 		     pseg += c;
 		 }
 	     }
 	     if(!pseg.empty()) {
 		 if(!qf) rv += '/';
 		 rv += pseg;
 	     }
 	     return rv;
 	 }
 
 	string& strip_uri_fragment_part(string& u) {
 	    string::size_type q = u.find('?'), f = u.find('#');
 	    if(q==string::npos) {
 		if(f!=string::npos)
 		    u.erase(f);
 	    }else{
 		if(f!=string::npos) {
 		    if(f<q)
 			u.erase(f,q-f);
 		    else
 			u.erase(f);
 		}
 	    }
 	    return u;
 	}
 
 	bool uri_matches_realm(const string& uri,const string& realm) {
 	    string nrealm = opkele::util::rfc_3986_normalize_uri(realm);
 	    string nu = opkele::util::rfc_3986_normalize_uri(uri);
 	    string::size_type pr = nrealm.find("://");
 	    string::size_type pu = nu.find("://");
 	    assert(!(pr==string::npos || pu==string::npos));
 	    pr += sizeof("://")-1;
 	    pu += sizeof("://")-1;
 	    if(!strncmp(nrealm.c_str()+pr,"*.",2)) {
 		pr = nrealm.find('.',pr);
 		pu = nu.find('.',pu);
 		assert(pr!=string::npos);
 		if(pu==string::npos)
 		    return false;
 		// TODO: check for overgeneralized realm
 	    }
 	    string::size_type lr = nrealm.length();
 	    string::size_type lu = nu.length();
 	    if( (lu-pu) < (lr-pr) )
 		return false;
 	    pair<const char*,const char*> mp = mismatch(
 		    nrealm.c_str()+pr,nrealm.c_str()+lr,
 		    nu.c_str()+pu);
 	    if( (*(mp.first-1))!='/'
 		    && !strchr("/?#",*mp.second) )
 		return false;
 	    return true;
 	}
 
 	string abi_demangle(const char *mn) {
 #ifndef HAVE_DEMANGLE
 	    return mn;
 #else /* !HAVE_DEMANGLE */
 	    int dstat;
 	    char *demangled = abi::__cxa_demangle(mn,0,0,&dstat);
 	    if(dstat)
 		return mn;
 	    string rv = demangled;
 	    free(demangled);
 	    return rv;
 #endif /* !HAVE_DEMANGLE */
 	}
 
 	string base64_signature(const assoc_t& assoc,const basic_openid_message& om) {
 	    const string& slist = om.get_field("signed");