cage is a creator of unbreakable padded cells for untrusted applications. Use it when you would like to chroot a program that wants to run as root. (e.g. sendmail, sshd, etc.)
Ordinarily, applications run as root have the capabilities required to break out of their chroot jail; however, before executing your command, cage chroots into the directory you specify, then drops all privileges that would allow your process to bust out.
By default, the removed privileges currently include:
CAP_SETPCAP
Set capabilities on arbitrary processes.
CAP_SYS_MODULE
The ability to insert loadable kernel modules that hijack systems calls.
CAP_SYS_RAWIO
Read/write access to kernel memory, keyboard controller, I/O devices, etc.
CAP_SYS_CHROOT
Allows a process to change its root directory, the obvious way to break chroot.
CAP_SYS_PTRACE
Attach to, and control the execution of, arbitrary processes.
CAP_SYS_ADMIN
Allows mounting and unmounting of filesystems, among other things.
CAP_MKNOD
Allows a process to create device nodes.
Installation is straightforward. Either install the red hat 7.2 RPM, or:
$ tar -zxf cage-0.80.tar.gz
$ cd cage-0.80
$ ./configure
# make install
Then, follow these intructions to set up POSIX 1003.1e capabilities
To run sendmail under /home/sendmail the following command would be used:
cage -- /home/sendmail /usr/sbin/sendmail -bd -q15m
The `--' is only required if the command to be run has dashed arguments.
To run slapd under /home/ldap with only the capabilities required to bind to a privileged port:
cage -c cap_net_bind_service=eip /home/ldap slapd