X-Git-Url: http://git.vanrenterghem.biz/git.ikiwiki.info.git/blobdiff_plain/7cd083dc66f80a3c27e8717aa0da1465b4e9b9ef..8fa04f56fccf6a331a66eb896de1366522cd8562:/doc/bugs/Running_on_an_alternative_port_fails.mdwn

diff --git a/doc/bugs/Running_on_an_alternative_port_fails.mdwn b/doc/bugs/Running_on_an_alternative_port_fails.mdwn
index c6e09171d..942700ba3 100644
--- a/doc/bugs/Running_on_an_alternative_port_fails.mdwn
+++ b/doc/bugs/Running_on_an_alternative_port_fails.mdwn
@@ -5,3 +5,89 @@ For example if you run on 'http://my.gear.xxx:8080/' then after clicking login (
  -- fergus
 
 NB: both the 'url' and the 'cgiurl' include the port and removing the port element provides the expected functionality.
+
+---
+
+> I tried to reproduce this by making my laptop's web server use port
+> 8080. Set up ikiwiki to use that in cgiurl and url, and had
+> no problem with either openid or password auth login.
+> 
+> Ikiwiki has had some changes in this area in the past year; you don't say
+> what version you were using. It could also be a problem with your web
+> server, conceviably, if didn't correctly communicate the port to the cgi
+> program. --[[Joey]]
+
+---
+
+>> I did think of that so threw a 'printenv' script to check the port was arriving 
+right.
+
+>>>     SERVER_PORT=8181  
+>>>     HTTP_HOST=zippy0.ie0.cobbled.net  
+
+[ ... ]
+
+>>>> In apache, `HTTP_HOST` includes the port. This is not part of the CGI
+>>>> spec it seems, but perl's `CGI` module seems to rely on it,
+>>>> in `virtual_port`:
+
+>>>>>     my $vh = $self->http('x_forwarded_host') || $self->http('host');
+>>>>>     my $protocol = $self->protocol;
+>>>>>     if ($vh) {
+>>>>>        return ($vh =~ /:(\d+)$/)[0] || ($protocol eq 'https' ? 443 : 80);
+
+>>>> The `CGI` module only looks at `SERVER_PORT` when there's no
+>>>> `HTTP_HOST`. So this is either a bug in perl's CGI or thttpd.
+>>>> --[[Joey]]
+
+[ ... ]
+
+---
+
+>>>>> This is interesting.  If HTTP_HOST is wrong then
+
+>>>>> 0. the client header must be wrong (i.e. not including the PORT)
+>>>>> 0. `perl`'s doing something bad[tm] (or at least lazy)
+>>>>> 0. `apache` is adding it
+>>>>> 0. `thttpd` is stripping it
+
+>>>>> Quick hack shows that `thttpd` must be stripping the port
+number from the `Host:` header.  That can be fixed.
+
+>>>>> Thanks for the assist. -- fergus
+
+---
+
+Patch for `thttpd-2.25b` for posterity and completeness
+
+[[!format patch """
+
+diff --git a/libhttpd.c b/libhttpd.c
+index 73689be..039b7e3 100644
+--- a/libhttpd.c
++++ b/libhttpd.c
+@@ -2074,9 +2074,6 @@ httpd_parse_request( httpd_conn* hc )
+ 		cp = &buf[5];
+ 		cp += strspn( cp, " \t" );
+ 		hc->hdrhost = cp;
+-		cp = strchr( hc->hdrhost, ':' );
+-		if ( cp != (char*) 0 )
+-		    *cp = '\0';
+ 		if ( strchr( hc->hdrhost, '/' ) != (char*) 0 || hc->hdrhost[0] == '.' )
+ 		    {
+ 		    httpd_send_err( hc, 400, httpd_err400title, "", httpd_err400form, "" );
+
+"""]]
+
+-- fergus
+
+---
+
+I've gone ahead and filed a bug on CGI.pm too:
+<https://rt.cpan.org/Ticket/Display.html?id=72678> --[[Joey]] 
+
+---
+
+That'll be an interesting discussion as I'd suggest that HTTP_ headers are defined in the CGI specification as client headers and thus what `thttpd` is doing is wrong (i.e. mangling the client's own representation).  Whether a CGI client should trust HTTP_ header over the server is probably already settled by convention.
+
+-- fergus