Hi,
I found suspicious behavior with CGI using PATH_INFO when
it is on the client authenticated SSL.
My situation is as follows.
/foo
SSLVerifyClient none
/bar
SSLVerifyClient require
Assume here're some files under /foo.
a.html - plain html file
b.cgi - regular cgi (not nph)
Then, I get somes file from browser.
What I could get without certificates are...
/foo/a.html
/foo/b.cgi
/foo/b.cgi/foo <--- (1)
What SSL session renegotiation runs, so I needed certificates
to access are...
/bar
/foo/b.cgi/bar <--- (2)
Because /foo is configured as "SSLVerifyClient none", so I expect
it does not need any certificates.
What is different between (1) and (2)?
I could see the request is processed twice because the second
request (protocol="INCLUDED") is generated in ap_add_cgi_vars().
The second request issued with uri=path_info, so that it makes
SSL renegotiation only when path_info is under "/bar".
util_script.c
370 if (r->path_info && r->path_info[0]) {
371 /*
372 * To get PATH_TRANSLATED, treat PATH_INFO as a URI path.
373 * Need to re-escape it for this, since the entire URI was
374 * un-escaped before we determined where the PATH_INFO began.
375 */
376 request_rec *pa_req;
377
378 pa_req = ap_sub_req_lookup_uri(ap_escape_uri(r->pool, r->path_info),
r,
379 NULL);
Does someone have any idea for this?
Masaoki Kobayashi
