Description of problem: When Kerberos 5 authentication is configured (i.e. via authconfig), sshd's connection-specific child process segfaults (SIGSEGV) when a non-root user attempts to log onto the system. It doesn't matter of the non-root user authenticates locally or via Kerberos 5. The parent sshd instance is unaffected and continues to field connection attempts on port 22. I observed similar behavior on Red Hat Linux 9 with openssh-server version 3.5p1-6.9. From the perspective of the client, the connection is dropped before the user is prompted for a password. The problem can be duplicated by running sshd in debug mode (the "-d" option to sshd) under GDB. A GDB transcript that includes a backtrace is attached to this bug report. Version-Release number of selected component (if applicable): openssh-server-3.4p1-4 How reproducible: Every time Steps to Reproduce: 1. Run authconfig and enable Kerberos 5 authentication. 2. A reboot is NOT required. 3. Restarting sshd is NOT required. 4. Use the ssh client to attempt to log into a non-root account, e.g. ssh user@localhost. 5. It does not matter if the account authenticates via Kerberos 5 or locally via /etc/shadow. 6. To return to normal operations, re-run authconfig and disable Kerberos 5 authentication. Actual results: After the connection encryption is negotiated, the ssh client will lose contact with the sshd server child instance and report "Connection closed by <NAME OR ADDRESS>", e.g. "Connection closed by 127.0.0.1". When running sshd in debug mode under GDB, one will observe a segfault (signal SIGSEV) in malloc_consolidate() from /lib/libc.so.6. Expected results: The user should have been authenticated as customary for ssh, via the pam_krb5 or pam_unix modules. Obviously, the segfault in malloc_consolidate() should not occur. Additional info: The relevant error as reported when running sshd under GDB in debug mode (sshd's debug messages are prefixed with "debug3"): debug3: mm_auth_password: waiting for MONITOR_ANS_AUTHPASSWORD debug3: mm_request_receive_expect entering: type 11 debug3: mm_request_receive entering debug3: monitor_read: checking request 10 Program received signal SIGSEGV, Segmentation fault. 0x40210618 in malloc_consolidate () from /lib/libc.so.6 A backtrace at this point shows the following: #0 0x40210618 in malloc_consolidate () from /lib/libc.so.6 #1 0x4020fd85 in _int_malloc () from /lib/libc.so.6 #2 0x4020f484 in malloc () from /lib/libc.so.6 #3 0x40183442 in krb5_sendto_kdc () from /usr/kerberos/lib/libkrb5.so.3 #4 0x4016c043 in send_as_request () from /usr/kerberos/lib/libkrb5.so.3 #5 0x4016d260 in krb5_get_init_creds () from /usr/kerberos/lib/libkrb5.so.3 #6 0x4016e6a4 in krb5_get_init_creds_password () from /usr/kerberos/lib/libkrb5.so.3 #7 0x404ba36b in pam_sm_authenticate () from /lib/security/pam_krb5.so #8 0x4001bba9 in _pam_dispatch_aux () from /lib/libpam.so.0 #9 0x4001bd33 in _pam_dispatch () from /lib/libpam.so.0 #10 0x400158df in pam_sm_chauthtok () from /lib/security/pam_stack.so #11 0x40014c4e in pam_sm_authenticate () from /lib/security/pam_stack.so #12 0x4001bba9 in _pam_dispatch_aux () from /lib/libpam.so.0 #13 0x4001bd33 in _pam_dispatch () from /lib/libpam.so.0 #14 0x4001d828 in pam_authenticate () from /lib/libpam.so.0 #15 0x0805487a in strcpy () #16 0x08054bbc in strcpy () #17 0x08055619 in strcpy () #18 0x0805f8b1 in strcpy () #19 0x0805f090 in strcpy () #20 0x0805ed55 in strcpy () #21 0x0804d651 in strcpy () #22 0x0804e750 in strcpy () #23 0x401b74ed in __libc_start_main () from /lib/libc.so.6
Created attachment 93465 [details] GDB output for sshd (Red Hat Linux 8, openssh-server-3.4p1-4)
Does it still happen for you with the current Fedora Core?
I do not know. I am sorry if this sounds snotty, but I stopped using Red Hat Linux and its variants after my bug reports went unanswered. If I recall correctly, this was a bug in the portable version of OpenSSH (not in LinuxPAM or glibc) that was fixed around version 3.6.
Thanks anyway.