The POSIX setuid()
function has complex semantics and platform-specific behavior [[Open Group 04]].
If the process has appropriate privileges,
setuid()
shall set the real user ID, effective user ID, and the saved set-user-ID of the calling process touid
.If the process does not have appropriate privileges, but
uid
is equal to the real user ID or the saved set-user-ID,setuid()
shall set the effective user ID touid
; the real user ID and saved set-user-ID shall remain unchanged.
The meaning of "appropriate privileges" varies from platform to platform. For example, on Solaris appropriate privileges for setuid()
means that the PRIV_PROC_SETID
privilege is in the effective privilege set of the process. On BSD, it means that the effective user ID (EUID) is zero (that is, the process is running as root) or that uid=geteuid()
. On Linux, it means that the process has CAP_SETUID
capability and that setuid(geteuid())
will fail if the effective user ID (EUID) is not equal to 0, the real user ID (RUID), or the saved set-user-ID (SSUID).
Because of this complex behavior, there may be cases where the desired privilege drops are unsuccessful. For example, the range of Linux Kernel versions (2.2.0-2.2.15) is vulnerable to an insufficient privilege attack wherein setuid(getuid())
did not drop privileges as expected when the capability bits were set to zero. As a precautionary measure, subtle behavior and error conditions for the targeted implementation must be carefully noted.
Noncompliant Code Example
This noncompliant code example compiles cleanly on most POSIX systems, but no explicit checks have been made to ensure that privilege relinquishment has succeeded. This may be dangerous depending on the sequence of the preceding privilege changes.
/* Code intended to run with elevated privileges */ /* Temporarily drop privileges */ if (seteuid(getuid()) != 0) { /* Handle error */ } /* Code intended to run with lower privileges */ if (need_more_privileges) { /* Restore privileges */ if (seteuid(0) != 0) { /* Handle error */ } /* Code intended to run with elevated privileges */ } /* ... */ /* Permanently drop privileges */ if (setuid(getuid()) != 0) { /* Handle error */ } /* * Code intended to run with lower privileges, * but if privilege relinquishment failed, * attacker can regain elevated privileges! */
If the program is run as a setuid root program, the state of the UID
s over time might be as follows:
Description |
Code |
EUID |
RUID |
SSUID |
---|---|---|---|---|
program startup |
|
0 |
user |
0 |
temporary drop |
|
user |
user |
0 |
restore |
|
0 |
user |
0 |
permanent drop |
|
user |
user |
user |
restore (attacker) |
|
user |
user |
user |
If the program fails to restore privileges, it will be unable to permanently drop them later:
Description |
Code |
EUID |
RUID |
SSUID |
---|---|---|---|---|
program startup |
|
0 |
user |
0 |
temporary drop |
|
user |
user |
0 |
restore |
|
user |
user |
0 |
permanent drop |
|
user |
user |
0 |
restore (attacker) |
|
0 |
0 |
0 |
Compliant Solution
This compliant solution was implemented in sendmail, a popular mail transfer agent, to determine if superuser privileges were successfully dropped [[Wheeler 03]]. If the setuid()
call succeeds after (supposedly) dropping privileges permanently, privileges were not dropped as intended.
/* Code intended to run with elevated privileges */ /* Temporarily drop privileges */ if (seteuid(getuid()) != 0) { /* Handle error */ } /* Code intended to run with lower privileges */ if (need_more_privileges) { /* Restore Privileges */ if (seteuid(0) != 0) { /* Handle error */ } /* Code intended to run with elevated privileges */ } /* ... */ /* Permanently drop privileges */ if (setuid(getuid()) != 0) { /* Handle error */ } if (setuid(0) != -1) { /* Privileges can be restored, handle error */ } /* * Code intended to run with lower privileges; * attacker cannot regain elevated privileges */
Compliant Solution
A better solution is to ensure that proper privileges exist before attempting to carry out a permanent drop.
/* Store the privileged ID for later verification */ uid_t privid = geteuid(); /* Code intended to run with elevated privileges */ /* Temporarily drop privileges */ if (seteuid(getuid()) != 0) { /* Handle error */ } /* Code intended to run with lower privileges */ if (need_more_privileges) { /* Restore Privileges */ if (seteuid(privid) != 0) { /* Handle error */ } /* Code intended to run with elevated privileges */ } /* ... */ /* Restore privileges if needed */ if (geteuid() != privid) { if (seteuid(privid) != 0) { /* Handle error */ } } /* Permanently drop privileges */ if (setuid(getuid()) != 0) { /* Handle error */ } if (setuid(0) != -1) { /* Privileges can be restored, handle error */ } /* * Code intended to run with lower privileges; * attacker cannot regain elevated privileges */
Supplemental Group Privileges
Any user, including root, may belong to multiple groups; these are handled by supplemental group privileges. The getgroups()
function returns an array of group IDs, and the setgroups()
function can set the array to an arbitrary array, but usually only if the user has root privileges. While POSIX defines the getgroups()
function, it does not define setgroups()
.
Under normal circumstances setuid() and related calls do not alter the supplemental group privileges. However, a setuid-root program can grant itself supplemental group ids and then relinquish root privileges, in which icase it maintains the supplemental group ids, but lacks the privilege necessary to relinquish them. Consequently, it is recommended that a program relinquish supplemental group privileges immediately before relinquishing root privileges.
POS37-C. Ensure that privilege relinquishment is successful discusses how to drop supplemental group privileges. To ensure that supplemental group privileges are indeed relinquished, you can use the following eql_sups
function:
/* Returns nonzero if the two group lists are equivalent (taking into account that the lists may differ wrt the egid */ int eql_sups(const int cursups_size, const gid_t* const cursups_list, const int targetsups_size, const gid_t* const targetsups_list) { int i; int j; const int n = targetsups_size; const int diff = cursups_size - targetsups_size; const gid_t egid = getegid(); if (diff > 1 || diff < 0 ) { return 0; } for (i=0, j=0; i < n; i++, j++) { if (cursups_list[j] != targetsups_list[i]) { if (cursups_list[j] == egid) { i--; /* skipping j */ } else { return 0; } } } /* If reached here, we're sure i==targetsups_size. Now, either j==cursups_size (skipped the egid or it wasn't there), or we didn't get to the egid yet because it's the last entry in cursups */ return j == cursups_size || (j+1 == cursups_size && cursups_list[j] == egid); }
System-Specific Capabilities
Many systems have nonportable privilege capabilities that, if unchecked, can yield privilege escalation vulnerabitilites.
File System Access Privileges (Linux)
Processes on Linux have two additional values called fsuid
and fsgid
. These indicate the privileges used when accessing files on the file system. These values normally shadow the effective user ID and effective group ID, but the setfsuid()
and setfsgid()
functions allow them to be changed. Since changes to the euid
and egid
normally also apply to fsuid
and fsgid
, a program relinquishing root privileges need not be concerned with setting fsuid
or fsgid
to safe values. However, there has been at least one kernel bug that violated this invariant ([[Chen 02]] and [[Tsafrir 08]]). Consequently, a prudent program will check that fsuid
and fsgid
have harmless values after relinquishing privileges.
Risk Assessment
If privilege relinquishment conditions are left unchecked, any flaw in the program may lead to unintended system compromise corresponding to the more privileged user or group account.
Rule |
Severity |
Likelihood |
Remediation Cost |
Priority |
Level |
---|---|---|---|---|---|
POS37-C |
high |
probable |
low |
P18 |
L1 |
Automated Detection
Klocwork Version 8.0.4.16 can detect violations of this rule with the SV.FIU.PERMISSIONS and SV.USAGERULES.PERMISSIONS checkers.
Related Vulnerabilities
Search for vulnerabilities resulting from the violation of this rule on the CERT website.
References
[[Chen 02]] "Setuid Demystified"
[[Dowd 06]] Chapter 9, "Unix I: Privileges and Files"
[[ISO/IEC PDTR 24772]] "XYO Privilege Sandbox Issues"
[[MITRE 07]] CWE ID 250, "Execution with Unnecessary Privileges," CWE ID 273, "Failure to Check Whether Privileges Were Dropped Successfully"
[[Open Group 04]] setuid()
, getuid()
, seteuid()
[[Tsafrir 08]] "The Murky Issue of Changing Process Identity: Revising 'Setuid Demystified'"
[[Wheeler 03]] Section 7.4, "Minimize Privileges"