The MIT Kerberos Development Team is aware of the following vulnerability in the MIT krb5 software. Please do not publicly disseminate this information prior to our public disclosure. We expect to produce patches and a public advisory within the next week -- probably before 21 December 2004. Accelerated release timeline is due to prior public discussion of the bug. Please let me know if you have any comments on this draft advisory. DRAFT -- DO NOT PUBLISH MIT krb5 Security Advisory 2004-004 Original release: 2004-12-xx Topic: heap buffer overflow in libkadm5srv Severity: serious [ NOTE: This vulnerability has been already discussed on a public mailing list, in sufficient detail for someone to infer how to perform an attack. For this reason, patches and advisory will be published as soon as they are ready, with minimal coordination. We anticipate releasing patches and the public advisory within a week -- probably before 21 December 2004. ] SUMMARY ======= The MIT Kerberos 5 administration library (libkadm5srv) contains a heap buffer overflow in password history handling code which could be exploited to execute arbitrary code on a Key Distribution Center (KDC) host. The overflow occurs during a password change of a principal with a certain password history state. An administrator must have performed a certain password policy change in order to create the vulnerable state. (See MITIGATING FACTORS below.) No exploits are known to exist at this time, though a public discussion of the bug took place during the first weeks of December 2004, containing sufficient detail that someone could infer how to perform an attack. Exploitation of this vulnerability is believed to be difficult, due to the limited extent of the overflow. IMPACT ====== An authenticated user, not necessarily one with administrative privileges, could execute arbitrary code on the KDC host, compromising an entire Kerberos realm. [CAN-2004-1189] MITIGATING FACTORS ================== * Typically, only a principal satisfying the following conditions can trigger the buffer overflow upon password change: + have changed its password fewer times than the history count in its password policy + had its password policy's history count subsequently reduced to equal the number of times it has changed its password * There are other means of producing the vulnerable state, though they are significantly more complex and much less likely. All of these other methods involve a reduction of the password history count in a password policy. * A workaround exists (see FIXES). AFFECTED SOFTWARE ================= * KDC software on all releases of MIT krb5, up to and including krb5-1.3.5. The vulnerable library is libkadm5srv. Programs which use the vulnerable functionality of the library include: + kadmind (administration daemon) + kadmin.local (KDC-local administration client) + kadmind4 (krb4 compatibility administration daemon) FIXES ===== * WORKAROUND: Until your KDC programs and libraries have been patched, do not decrease the password history count on any policy in your Kerberos realm. * The upcoming krb5-1.4 release will contain fixes for this problem. * The upcoming krb5-1.3.6 patch release will contain fixes for this problem. * Patches are in development. REFERENCES ========== This announcement and related security advisories may be found on the MIT Kerberos security advisory page at: http://web.mit.edu/kerberos/advisories/index.html The main MIT Kerberos web page is at: http://web.mit.edu/kerberos/index.html CVE CAN-2004-1189: http://cve.mitre.org/cgi-bin/cvename.cgi?name=CAN-2004-1189 Administration library in MIT Kerberos 5 release 1.3.5 and earlier has a heap buffer overflow in code which handles password history, possibly allowing authenticated attackers to execute arbitrary code on a KDC host. ACKNOWLEDGMENTS =============== Thanks to Michael Tautschnig for reporting this problem. Thanks to Chaskiel Grundman and Luke Howard for providing debugging help on the mailing list. DETAILS ======= The vulnerable function is add_to_history() in src/lib/kadm5/srv/svr_principal.c. The password history is stored as a ring buffer, represented as an array of osa_pw_ent_rec, which is adb->old_keys. The "next" pointer is an index into the array, adb->old_key_next, and the length of the array is stored in adb->old_key_len. The array is dynamically resized as needed, and there is no separate head pointer. The policy's history count is stored in pol->pw_hist_num, but the actual maximum number of keys stored in adb->old_keys is pol->pw_hist_num-1, since the "current" key data are also used for history comparisons when a password change occurs. The index value adb->old_key_next is permitted to index to a position one past the end of the array adb->old_keys if adb->old_key_next is less than pol->pw_hist_num-1. This out-of-bounds indexing is usually fixed up when add_to_history() enlarges the array on a subsequent call. If pol->pw_hist_num is reduced to adb->old_key_next after a password change that causes adb->old_key_next to index out of bounds, a subsequent password change will not run the resizing code, and add_to_history() will write a password history entry past the end of the array adb->old_keys. REVISION HISTORY ================ 2004-12-xx original release Copyright (C) 2004 Massachusetts Institute of Technology
This issue is now public.
http://www.securityfocus.com/archive/1/385000
aliz/rphillips, please apply patch.
Followups to public bug 75143 *** This bug has been marked as a duplicate of 75143 ***