=====[BEGIN-ACROS-REPORT]===== ========================================================================= ACROS Security Problem Report #2000-04-06-1-PUB ------------------------------------------------------------------------- Bypassing Warnings For Invalid SSL Certificates In Netscape Navigator ========================================================================= FULL REPORT PUBLIC ====== Affected System(s): Netscape Navigator & Communicator Problem: Bypassing Warnings For Invalid SSL Certificates Severity: High Solution: Installing the Personal Security Manager or Installing the newest Netscape Communicator (v4.73) Discovered: April 3, 2000 Vendor notified: April 4, 2000 Last update: May 10, 2000 Published: May 10, 2000 SUMMARY ======= Our team has discovered a flaw in Netscape Navigator that allows bypassing of warning about an invalid SSL certificate. SSL protection is used in most major Internet-based financial services (e-banking, e-commerce). The flaw we have found effectively disables one of the two basic SSL functionalities: to assure users that they are really communicating with the intended web server - and not with a fake one. Using this flaw, the attacker can make users send secret information (like credit card data and passwords) to his web server rather than the real one - EVEN IF THE COMMUNICATION IS PROTECTED BY SSL PROTOCOL. INTRODUCTION (skip this section if you already understand how SSL works) ============ When a web browser tries to connect to a SSL-protected server, a so-called SSL session is established. At the beginning of this session the server presents his SSL certificate containing his public key. At this point, browser checks the certificate for the following conditions (*): 1) Certificate must be issued by a certificate authority trusted by browser (some are default: Verisign, Thawte etc.) 2) Certificate must not be expired (its expiry date:time must be later than the current system date:time on the computer browser is running on) 3) Certificate must be for the server that browser is connecting to (if browser is connecting to www.e-bank.com, the certificate must be for www.e-bank.com) All three conditions must be met for browser to accept the certificate. For every condition not met, browser should display a warning to the user and then user can decide whether connection should be established or not. These three conditions combined provide user with assurance that his browser is really connecting to the correct server and not to some fake server placed on the Internet by malicious individual(s) trying to trick users to give them credit card information, passwords and other secret information. For example, let's take a look at a sample web e-banking system that doesn't use SSL certificates and requires one-time password tokens for user authentication. User connects to http://www.e-bank.com. Browser asks DNS server for IP address of www.e-bank.com and gets 100.100.100.100. Browser then connects to 100.100.100.100 and user is presented with login form asking for his username and one-time password. He enters this data and starts using e-banking services. A simple attack (called web-spoofing) on this system is to attack the DNS server and "poison" its entry for www.e-bank.com with attacker's IP address 99.99.99.99. Attacker sets up a web server at 99.99.99.99 that web-wise looks exactly like the original www.e-bank.com server. User trying to connect to www.e-bank.com will now instead connect to the attacker's server and provide it with his one-time password. Attacker's server will use this password to connect to the real server at 100.100.100.100 and transfer all of the user's money to his secret Swiss bank account ;-). This attack is successfully disabled by using SSL protocol. In that case, when browser falsely connects to www.e-bank.com at 99.99.99.99 rather than to 100.100.100.100, attacker's server must provide a valid certificate for www.e-bank.com, which it can't unless the attacker has stolen the secret key and the certificate from the real server. Let's look at three possibilities: 1) Attacker could issue a certificate for www.e-bank.com himself (on his own CA). That wouldn't work since his CA is not trusted by user's browser. 2) Attacker could use a stolen expired key and certificate (those are often not protected as strongly as valid ones since one could think they can't be used any more). That wouldn't work since browser will notice that certificate is expired. 3) Attacker could use a valid key and certificate for some other site (e.g. www.something.org). That wouldn't work since browser will accept only valid certificates for www.e-bank.com. It would seem that this problem of web-spoofing is successfully solved with SSL certificates. PROBLEM ======= There is a flaw in implementation of SSL certificate checks in Netscape Navigator. The Flaw -------- Netscape Navigator correctly checks the certificate conditions (*) at the beginning of a SSL session it establishes with a certain web server. The flaw is, while this SSL session is still alive, all HTTPS connections to *THAT SERVER'S IP ADDRESS* are assumed to be a part of this session (and therefore certificate conditions are not checked again). Instead of comparing hostnames to those of currently open sessions, Navigator compares IP addresses. Since more than one hostname can have the same IP address, there is a great potential for security breach. This behavior is not in compliance with SSL specification. DEMONSTRATION ============= The following will try to demonstrate the flaw. It is assumed that for redirecting user's web traffic, the attacker will generally use "DNS poisoning" or reconfiguring routers, while in our demonstration we will use the HOSTS file on client computer to get the same effect and make it easier to reproduce the flaw. In this demonstration, we will make Navigator open Thawte's homepage over secure (HTTPS) connection while requesting Verisign's home address at https://www.verisign.com. Thawte's and Verisign's homepages are used as examples - this would work just the same on any other secured web sites. 1) First, add the following line to the local HOSTS file on the computer running the Navigator and save it: 207.240.177.177 www.verisign.com This will make the computer (and, consequently, the browser) think that IP address of www.verisign.com (which is actually 205.139.94.60) is in fact 207.240.177.177 (which is actually IP address of www.thawte.com). At this point it is important to note that SSL, if correctly implemented, provides protection against such "domain name spoofing", because while the browser will connect to the wrong server, that server will not be able to provide a valid SSL certificate and the SSL session will not be established (not without user being warned about the certificate). 2) Close all instances of Navigator to clean any cached IP addresses. 3) Open Navigator and go to https://www.thawte.com. It works as it should - Thawte's server provides a valid SSL certificate for its hostname (www.thawte.com) and so the SSL session is established. 4) With the same instance of Navigator, go to https://www.verisign.com. Now watch the Thawte's homepage appear again WITHOUT ANY WARNINGS! What happened here? In step 3), Navigator looked up the IP address for www.thawte.com (from the DNS server) and found 207.240.177.177. It tried to establish a SSL session with that IP address and correctly checked all three certificate conditions (*) - indeed, if any of them weren't true, a warning would pop up. In step 4), Navigator looked up the IP address for www.verisign.com (this time from HOSTS file, but it could easily have been from the same DNS server) and found again 207.240.177.177. Now, since there was already one SSL session open with that IP address, Navigator *INCORRECTLY* decided to use that session instead of establishing another one. EXPLOIT ======= This exploit will show how the flaw could be used to gather user's secret information. Assume there is a web bookstore at www.thebookstore.com. Users go to http://www.thebookstore.com (via normal HTTP connection), browse the books and add them to their virtual shopping baskets. At the check-out, they are directed to a secure order form (e.g. https://www.thebookstore.com/order_form.html) where they enter their personal and credit card information which is then submitted (again via secure HTTPS connection) to the server. This is a typical web e-commerce concept. Assume that IP address of www.thebookstore.com is 100.100.100.100. The attacker sets up his own web server with IP address 99.99.99.99 and installs on it a valid SSL certificate for host www.attacker.com (he could have purchased this certificate from e.g. Verisign if he owns the domain attacker.com; he could have stolen the certificate or he could have broken into a web server with a certificate already installed). The attacker makes this web server function as a gateway to www.thebookstore.com - meaning that all requests are forwarded to www.thebookstore.com, so virtually this server "looks and feels" exactly like the real www.thebookstore.com. There is just one difference: the page before the order form (e.g. http://www.thebookstore.com/basket.html) contains a small (1x1) image originating from https://www.attacker.com (secure HTTPS connection). Then, the attacker "poisons" a heavily used DNS server so that it will return 99.99.99.99 for requests about www.thebookstore.com (normally it returns 100.100.100.100). What happens then? All users of that DNS server who will try to visit (via normal HTTP) http://www.thebookstore.com will connect to 99.99.99.99 instead of 100.100.100.100 but will not notice anything because everything will look just the way it should. They will browse the books and add them to their shopping baskets and at check-out, they will be presented with the order form https://www.thebookstore.com/order_form.html. But the previous HTML page containing the hyperlink to the order form will also contain a small (1x1) image with source https://www.attacker.com/a.gif. Navigator will successfully download this image and for that it will establish a SSL session with www.attacker.com. This session then stays open. When the order form is accessed, Navigator tries to establish another SSL session, this time to www.thebookstore.com. Since DNS server claims this server has the same IP address as www.attacker.com (99.99.99.99), Navigator will use the existing SSL session with 99.99.99.99 and will not check the certificate. The result: Navigator is displaying a SECURE ORDER FORM that it believes to be originating from the genuine server www.thebookstore.com while in fact it is originating from the fake one. No warning about an invalid certificate is issued to the user so he also believes to be safe. When user submits his secret information, it goes to (through) the attacker's server where it is collected for massive abuse. For users to notice the foul play they would have to look at the certificate properties while on a "secure" page https://www.thebookstore.com/... The properties would show that the certificate used was issued for host www.attacker.com. Also, monitoring network traffic would show that the server is not at 100.100.100.100 where it should be but rather at 99.99.99.99. It is a very rare practice to check any of these when nothing suspect is happening. Notes ----- It should be noted that in the previous exploit, if the users tried to access https://www.thebookstore.com over secure (HTTPS) connection from the very start, Navigator would issue a warning. It is imperative for the exploit to work that some time *before* the first secure connection to https://www.thebookstore.com a successful secure connection is made to https://www.attacker.com. That's why a valid certificate must be installed on www.attacker.com. Also, it should be noted that Navigator's SSL sessions don't last forever. We haven't been able to predict the duration of these sessions (it seems to be depending on many things like inactivity time, total time etc.) and we also haven't investigated the possible effects of SSL session resuming. SOLUTION ======== Netscape has (even prior to our notification - see the Acknowledgments section) provided a Navigator Add-on called Personal Security Manager (PSM), freely downloadable at: http://www.iplanet.com/downloads/download/detail_128_316.html Installation of PSM, as far as we have tested it, corrects the identified flaw. Netscape Communicator (v4.73) currently includes the fix for this vulnerability. It is available for download at: http://home.netscape.com/download/ WORKAROUND ========== Navigator/Communicator users who can't or don't want to install PSM can use a "manual" method to make sure they are not under attack: When visiting an SSL-protected site, double click on the lock icon (bottom left corner) or the key icon (in older browsers) and see whether the certificate used for the connection is really issued for the correct hostname. E.g. If you visit https://www.verisign.com, make sure the certificate used is issued for www.verisign.com and not for some other hostname. ADVISORY ======== It is important to emphasize that the flaw presented completely compromises SSL's ability to provide strong server authentication and therefore poses a serious threat to Navigator users relying on its SSL protection. Users of web services --------------------- Netscape Navigator/Communicator users who are also users of any critical web services employing Secure Sockets Layer (SSL) protection to provide secrecy and integrity of browser-server communication are strongly advised to install Personal Security Manager or upgrade to Communicator 4.73 and thus disable this vulnerability. Main examples of such critical web services are: - web banking systems (especially the ones using passwords for authentication - even one-time passwords), - web stores (especially the ones accepting credit card data) and - other web-based e-commerce systems. Providers of web services ------------------------- Providers of critical web services employing Secure Sockets Layer (SSL) protection to provide secrecy and integrity of browser-server communication should advise their users to install Personal Security Manager or upgrade to Communicator 4.73 and thus disable this vulnerability. Since this vulnerability allows for the type of attack that can completely bypass the real/original web server, there are no technical countermeasures which providers of web services could deploy at their sites. Web services using client SSL certificates for user authentication ------------------------------------------------------------------ This vulnerability does NOT allow the attacker to steal client's SSL key and thus execute the man-in-the-middle attack on web services using client SSL certificates for user authentication. It still does, however, allow the attacker to place a fake server (an exact copy) and collect other information users provide (including the data in their client SSL certificates). TESTING RESULTS =============== Tests were performed on: Communicator 4.72 - affected Communicator 4.61 - affected Navigator 4.07 - affected ACKNOWLEDGMENTS =============== We would like to acknowledge Netscape (specifically Mr. Bob Lord and Mr. Kevin Murray) for prompt and professional response to our notification of the identified vulnerability and their help in understanding the flaw and "polishing" this report. We would also like to acknowledge Mr. Matthias Suencksen of Germany, who has discovered some aspects of this vulnerability before we did (back in May 1999). REFERENCES ========== Netscape has issued a Security Note about this vulnerability under a title "The Acros-Suencksen SSL Vulnerability" at: http://home.netscape.com/security/notes/index.html SUPPORT ======= For further details about this issue please contact: Mr. Mitja Kolsek ACROS, d.o.o. Stantetova 4 SI - 2000 Maribor, Slovenia phone: +386 41 720 908 e-mail: mitja.kolsek@acros.si PGP Key available at PGP.COM's key server. PGP Fingerprint: A655 F61C 5103 F561 6D30 AAB2 2DD1 562A DISTRIBUTION ============ This report was sent to: - BugTraq mailing list - NTBugTraq mailing list - Win2KSecAdvice mailing list - SI-CERT - ACROS client mailing list DISCLAIMER ========== The information in this report is purely informational and meant only for the purpose of education and protection. ACROS, d.o.o. shall in no event be liable for any damage whatsoever, direct or implied, arising from use or spread of this information. All identifiers (hostnames, IP addresses, company names, individual names etc.) used in examples and exploits are used only for explanatory purposes and have no connection with any real host, company or individual. In no event should it be assumed that use of these names means specific hosts, companies or individuals are vulnerable to any attacks nor does it mean that they consent to being used in any vulnerability tests. The use of information in this report is entirely at user's risk. COPYRIGHT ========= (c) 2000 ACROS, d.o.o., Slovenia. Forwarding and publishing of this document is permitted providing all information between marks "[BEGIN-ACROS-REPORT]" and "[END-ACROS-REPORT]" remains unchanged. =====[END-ACROS-REPORT]=====