Hi, (Sorry for the large CC list, but I am hoping to get a broad approval of the next changes for this in order to avoid previous mistakes. ;) In particular, I'd be very grateful for some input by Stefan considering his knowledge of the Apache codebase and how ... exotic this problems is.) As I mentioned in the #858373 bug report, I started looking at fixing the regression introduced by the 2.2.22-13+deb7u8 upload, part of DLA-841-1. The problem occurs when a CGI(d) ErrorDocument is configured to handle 400 error messages that can be triggered with a simple "GET / HTTP/1.0\n\n". Such a request segfaults Apache in Wheezy right now. I have been able to confirm that there is an unitialized variable that gets carried around. This issue was introduced as part of CVE-2016-8743-aux.patch in the original upload, although I fail to remember now why this hunk is there exactly. It seems to be related to a patch I somewhat blindly and incorrectly merged (see 87r33tqvqs.fsf@curie.anarc.at for details). Unfortunately, re-introducing the protocol initialization code isn't sufficient: it does fix the segfaults, but the ErrorDocument handling is not quite working yet. Instead of seeing the output of the ErrorDocument, after 10 seconds, I get the raw 400 message, doubled with a 500 error document warning: $ echo -ne "GET /foo HTTP/1.0\n\n" | nc localhost 80 HTTP/1.1 400 Bad Request Date: Wed, 19 Jul 2017 19:11:13 GMT Server: Apache/2.2.22 (Debian) Vary: Accept-Encoding Content-Length: 433 Connection: close Content-Type: text/html; charset=iso-8859-1 <!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML 2.0//EN"> <html><head> <title>400 Bad Request</title> </head><body> <h1>Bad Request</h1> <p>Your browser sent a request that this server could not understand.<br /> </p> <p>Additionally, a 500 Internal Server Error error was encountered while trying to use an ErrorDocument to handle the request.</p> <hr> <address>Apache/2.2.22 (Debian) Server at wheezy.raw Port 80</address> </body></html> In the error log, I see: [Wed Jul 19 19:11:23 2017] [error] [client 127.0.0.1] (70007)The timeout specified has expired: Error reading request entity data The first part of the error is mod_reqtimeout kicking in as the request parser stalls on the CGI script. The second part is mod_cgi(d) failing to read the request from the CGI script, obviously. My theory is that there is *still* something wrong with the request parser, even after fixing the r->protocol initialization flaw. I base this theory on the fact that a 404 ErrorDocument works without problem. $ echo -ne "GET /foo HTTP/1.0\r\n\r\n" | nc localhost 80 HTTP/1.1 404 Not Found Date: Wed, 19 Jul 2017 19:13:44 GMT Server: Apache/2.2.22 (Debian) Vary: Accept-Encoding Connection: close Content-Type: text/html Hello, World. Note that I have also tried to see if sending "\r\n" instead of just "\n" in my "hello world" example would work around the issue: it doesn't, unfortunately. I am at a loss as where to go from here, to be honest. The patch (attached) at least fixes the segfault, which resolves the primary issue at hand here (DoS by crashing processes!) but it would be nice to actually fix the ErrorDocument as well.. Any ideas? Thanks in advance, A.
Attachment:
signature.asc
Description: PGP signature