I recently worked on an issue where a new Microsoft patch for a security bulletin, MS12-006, was correlated to a web-based application occasionally returning errors. Specifically, while using a web-based version of a reporting package off of a version of Sun (Oracle) Java Web Server the browser would occasionally return a Times New Roman-font “Request Timeout” message when users clicked report-generating links. The folks who support the server working on the issue initially had spent time uninstalling recent patches and found that removing KB2585542 (the patch for MS12-006 on Windows 7 clients) from the client machine would stop the issue from occurring. They then thought it might be a Windows client issue, which is where I got involved.
After some investigation it turns out that the problem was actually with the web server not handling split (fragmented) SSL/TLS records properly. The installation of the patch for MS12-006 causes IE to, in some cases, split SSL/TLS records into multiple requests. Since the server couldn’t appropriately handle these requests it thought the client closed the connection before sending a complete request and returned an 408 Request Timeout message.
This was illustrated by the HTTP logs on the server which show incomplete requests being returned 408 messages, similar to what follows. The bold line illustrates a request that correlates with the issue:
IP | Username | Date | Request Header | Status | Content | Length---------+----------+----------------------+-----------------------------+--------+---------+--------
10.0.0.2 | - | 17/Feb/2012:10:03:25 | "P" | 408 | | 14810.0.0.2 | - | 17/Feb/2012:10:01:13 | "POST /TestApplet HTTP/1.1" | 200 | | 13056
After the installation of the patch for MS12-006 there is a setting which allows this new feature to specifically be used only when called for (“optin mode”, the default), all the time (“enabled for all”), or never (“disabled for all”). Setting the option to “disabled for all” kept the issue from being reproduced, and when set to “enabled for all” it always occurred. (See the Registry Information section of 2643584 for more information on these options and how to set them.) By this I showed that it was this SSL/TLS feature causing the issue communicating with the server. Since this appeared to be an issue only with this particular server the team supporting it is now looking into potential solutions on that side.
While it may be tempting to simply disable this feature on clients, this would handicap one of the security enhancements brought along with the MS12-006 patch. Since this issue isn’t likely to be widespread (else lots of HTTPS sessions would be failing) I believe that the servers should be fixed to be compatible with the client’s standard behavior. Specifically, I believe the issue to actually be with the server not having an RFC2246-compliant TLS implementation and thus it should be fixed.
Hopefully this information will be useful to others who run into such an issue. It’s a bit opaque at first, since it looks to the users like the server is having a problem, but to server like the client is occasionally failing to send proper HTTP requests. Trying to troubleshoot this with Fiddler didn’t help either, because the HTTPS proxying would recreate the request, apparently no longer splitting the SSL/TLS record, which kept the error from appearing.
Here’s a few links which are helpful in understanding this issue, in order of usefulness:
1 Comment· MS12-006: Vulnerability in SSL/TLS could allow information disclosure: January 10, 2012
· SSL/TLS Record Fragmentation Support – Microsoft blog article describing the issue in reverse, from a few years ago when SChannel didn’t support SSL/TLS Record Fragmentation.
· Microsoft Security Bulletin MS12-006 – Important, Vulnerability in SSL/TLS Could Allow Information Disclosure (2643584)
· HTTP Status Code 408 Documentation
· RFC2246 – See section 6.2.1.