===============================================================================
Trusted Path Execution (TPE) Linux Kernel Module, Version 2
===============================================================================
About this module
Trusted Path Execution is a security feature that denies users from executing programs that are not owned by root, or are writable. This closes the door on a whole category of exploits where a malicious user tries to execute his or her own code to attack the system.
Since this module doesn't use any kind of ACLs, it works out of the box with no configuration. It isn't complicated to test or deploy to current production systems. Just install it and you're done!
===============================================================================
Supported Kernels
Version 2 has been tested on the following systems (x86_64 only):
Version 1 works on older kernels, as far back as RHEL/CentOS 5 (linux-2.6.18)
This module should work on most linux kernels version 3.10 and above, but has only been verified on the above systems. If you get a compile error or a kernel oops, please contact this module's author.
===============================================================================
Features
Trusted Path Execution; deny execution of non-root owned or writable binaries
$ gcc -o exploit exploit.c $ chmod 755 exploit $ ./exploit -bash: ./exploit: Permission denied
$ dmesg | tail -n1 tpe: Denied untrusted exec of /home/corey/exploit (uid:500) by /bin/bash (uid:500), parents: /usr/sbin/sshd (uid:500), /usr/sbin/sshd (uid:0), /sbin/init (uid:0). Deny reason: directory uid not trusted
See the "Configuration" section below for how to tweak configuration.
Extras features, also ported from grsecurity (but not TPE related):
Restrict non-root users from viewing loaded kernel modules
$ cat /proc/modules cat: /proc/modules: Operation not permitted
Restrict non-root users from viewing the kernel symbol table
$ cat /proc/kallsyms cat: /proc/kallsyms: Operation not permitted
Restrict non-root users from viewing processes they don't own
$ ps auxf USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND corey 22450 0.0 0.7 97780 1804 ? S 14:12 0:00 sshd: corey @pts/1 corey 22451 0.0 0.7 108288 1884 pts/1 Ss 14:12 0:00 _ -bash corey 22474 0.0 0.4 110184 1044 pts/1 R+ 14:12 0:00 _ ps auxf
Restrict users from getting the kernel version
$ uname -a GNU/Linux
Restrict uses from running ptrace operations against their own processes
$ strace -p $$ strace: attach: ptrace(PTRACE_ATTACH, ...): Permission denied
Most of these are just kernel protections, and the user may be able to get the info via other means if you don't take additional precautions.
See the "Configuration" section below for how to enable these extras.
===============================================================================
How it works
This module makes use of the kernel ftrace framework. In a nutshell, ftrace is in charge of the first few bytes of most kernel symbols, and can redirect them to other addresses. TPE instructs ftrace to redirect certain security handlers to TPE code, which then decides whether it's appropriate to continue or not.
===============================================================================
Installation
See the INSTALL file for installation instructions.
===============================================================================
FAQ
See the FAQ file for frequently asked questions.
===============================================================================
Configuration
Although most people will find they don't need to change the default values, you have the option to configure various things in this module at runtime using the sysctl interface to tighten or lax the TPE restrictions. You can see the values in this proc directory:
/proc/sys/tpe/
softmode - log what would be denied but don't actually deny. default off strict - enforce some TPE features even on trusted users. default on check_file - check file owner/mode in addition to directory. default on group_writable - check if the file/directory is group writable. default on kill - kill the offending process and its parent when it gets denied execution from TPE, unless it's root. default off log - whether to log denied execs to the ring buffer. default on log_verbose - log what can be done to allow what was denied. default on log_max - maximum parent processes in a single log entry. default 50 log_floodburst - number of log entries before logging is disabled. default 5 log_floodtime - seconds until re-enabling logging after floodburst. default 5 xattr_soften - check extended attributes for a soften flag. default on paranoid - enforce the trusted path restrictions on root too. default off trusted_apps - a list of files, separated by commas, that would otherwise have had been denied an exec/mmap/mprotect, are instead allowed to proceed. default to empty (off) hardcoded_path - use with caution! a list of directories, separated by colons, that the trusted path will be restricted to; nothing outside this path may be executed/mmaped. default to empty (off) trusted_gid - gid of trusted users whom TPE is not enforced (see strict) default 0 (off) trusted_invert - changes what "trusted_gid" means; if set, users in the gid will be the only ones whom TPE is enforced. default 0 (off) admin_gid - files belonging to this group are treated as if they're owned by root; TPE is not enforced on them. default 0 (off) dmz_gid - users in this gid can't exec anything at all. default 0 (off)
extras/ - directory for additional protections that aren't TPE related.
ignore_softmode - enable extra features even if softmode is on. default off log - whether to log when extra features are denied. default on lsmod - denies non-root users from viewing loaded kernel modules proc_kallsyms - denies non-root users from viewing /proc/kallsyms harden_ptrace - denies non-root users from running ptrace operations hide_uname - denies non-root users from asking the kernel its version ps - denies non-root users from viewing processes they don't own ps_gid - gid of users who aren't restricted by ps. default 0 (off) restrict_setuid - users not in the trusted_gid are denied calls to setuid()
Edit the /etc/sysctl.d/tpe.conf file to change the defaults of these various features.
===============================================================================
Compatibility Issues
Test this module before deploying to a critical system, especially if you're not using a kernel in the above "Supported Kernels" list.
Trusted Path Execution will cause some programs to stop functioning correctly. If it doesn't result in a kernel BUG or other stack-trace to show up in dmesg, it's the program that needs fixing, not this module. There are several ways to work around this problem:
1) Set a file attribute on the offending binary to soften the checks or 2) Tune the module config with the "sysctl" command (see the admin_gid, trusted_gid, and trusted_apps options)
For example:
$ sudo setfattr -n security.tpe -v "soften_mmap" /usr/bin/gnome-session or $ sudo sysctl tpe.trusted_apps="/usr/bin/gnome-session"
The file attributes "soften_
Programs that have been reported to need such measures are:
You can persist these changes by editing the /etc/sysctl.conf file, or adding a conf file of your own in /etc/sysctl.d/
A few notes:
Kernel modules that make use of kernel tracing (ftrace, kprobes, etc) may conflict with this module if they touch the same kernel symbols. The only one I know of is "kpatch".
This module will not work on systems where loadable kernel module support is disabled at compile time (CONFIG_MODULES not being set).
This module requires many LSM hooks in order to work, and will error out when security is disabled at compile time (CONFIG_SECURITY not being set).
If you custom compile your own kernel and want TPE, consider using grsecurity instead of this module.
===============================================================================
BUGS
There are currently no known bugs. If you discover one, please report it to the author.
===============================================================================
Acknowledgements
Looking at the kpatch implementation, I was able to simplify this module.
Trusted Path Execution is a feature of grsecurity, and I originally pulled code from that project to make this module.
===============================================================================
About the Author
website: http://cormander.com/ github: https://github.com/cormander/ email: corman /AT/ cormander /DOT/ com
===============================================================================