System Hardening: Best practices associated with the configuration of an embedded Linux based operating system. This section includes both hardening of the kernel itself, as well as specific configurations and patches used to protect against known vulnerabilities within the build and configuration of the root filesystem.
At the Kernel level, we must ensure that no console can be launched. It could be used to change the behavior of the system or to have more information about it. Another aspect is the protection of the memory used by the Kernel.
The next sub-sections contain information on various kernel configuration options to enhance the security in the kernel (3.10.17) and also for applications compiled to take advantage of these security features. Additionally, there are also configuration options that protect from known vulnerable configuration options. Here's a high level summary of various kernel configurations that shall be required for deployment.
Kernel Version
The choice of kernel version for the AGL system is essential to its security. Depending on the type of board and eventual production system, different kernel versions are used. For example, one of the systems under study uses the Linux kernel version 3.10, while another uses the Linux kernel version 4.4. For the Linux kernel version 3.10.31, there are 25 known vulnerabilities. These vulnerabilities would allow an attacker to gain privileges, bypass access restrictions, allow memory to be corrupted, or cause denial of service. In contrast, the Linux kernel version of 4.4 has many fewer known vulnerabilities. For this reason, we would in general recommend the later kernel version as a basis for the platform.
Note that, although there are fewer known vulnerabilities in the most recent kernel versions there may be many unknown vulnerabilities underlying. A rule of thumb is to update the kernel as much as possible to avoid the problems you do know, but you should not be complacent in the trust that you place in it. A defense-in-depth approach would then apply.
If there are constraints and dependencies in upgrading to a newer kernel version (e.g. device drivers, board support providers) and you are forced to an older Linux kernel version, there need to be additional provisions made to reduce the risk of kernel exploits, which can include memory monitoring, watch-dog services, and system call hooking. In this case, further defense-in-depth techniques may be required to mitigate the risk of attacks to known vulnerabilities, which can also include runtime integrity verification of components that are vulnerable to tampering.
Kernel Build Configuration
The kernel build configuration is extremely important for determining the level
of access to services and to reduce the breadth of the attack surface. Linux
contains a great and flexible number of capabilities and this is only controlled
through the build configuration. For example, the CONFIG_MODULES parameter
allows kernel modules to be loaded at runtime extending the capabilities of the
kernel. This capability needs to be either inhibited or controlled at runtime
through other configuration parameters. For example, CONFIG_MODULE_SIG_FORCE=y
ensures that only signed modules are loaded. There is a very large number of
kernel configuration parameters, and these are discussed in detail in this
section.
General configuration
Mandatory Access Control
Kernel should controls access with labels and policy.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-General-MAC-1 | CONFIG_IP_NF_SECURITY | m |
| Kernel-General-MAC-2 | CONFIG_IP6_NF_SECURITY | m |
| Kernel-General-MAC-3 | CONFIG_EXT2_FS_SECURITY | y |
| Kernel-General-MAC-4 | CONFIG_EXT3_FS_SECURITY | y |
| Kernel-General-MAC-5 | CONFIG_EXT4_FS_SECURITY | y |
| Kernel-General-MAC-6 | CONFIG_SECURITY | y |
| Kernel-General-MAC-7 | CONFIG_SECURITY_SMACK | y |
| Kernel-General-MAC-8 | CONFIG_TMPFS_XATTR | y |
Please also refer to the Mandatory Access Control documentation in Platform. You can also find useful documentation and links on wikipedia about MAC and about SMACK.
Disable kexec
Kexec is a system call that enables you to load and boot into another kernel from the currently running kernel. This feature is not required in a production environment.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-General-kexec-1 | CONFIG_KEXEC |
n |
kexec can load arbitrary kernels but signing of new kernel can be enforced like it is can be enforced for new modules.
Disable kernel IP auto-configuration
It is preferable to have an IP configuration performed using a user-space tool as these tend to have more validation. We do not want the network interface coming up until the system has come up properly.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-General-IPAutoConf-1 | CONFIG_IP_PNP |
n |
Disable Sysctl syscall support
Enabling this will result in code being included that is hard to maintain and not well tested.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-General-SysCtl_SysCall-1 | CONFIG_SYSCTL_SYSCALL |
n |
Disable Legacy Linux Support
There are some Kernel Configs which are present only to support legacy binaries.
See also "Consoles" part in order to disabling support for legacy binary
formats. The uselib system call, in particular, has no valid use in any
libc6 or uclibc system in recent times. This configuration is supported in
Linux 3.15 and greater and thus should only be disabled for such versions.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-General-LegacyLinux-1 | CONFIG_USELIB |
n |
Disable firmware auto-loading user mode helper
The firmware auto loading helper, which is a utility executed by the kernel on
hotplug events requiring firmware, can to be set setuid. As a result of
this, the helper utility is an attractive target for attackers with control of
physical ports on the device. Disabling this configuration that is supported in
Linux 3.9 and greater.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-General-FirmHelper-1 | CONFIG_FW_LOADER_USER_HELPER |
n |
It doesn't strictly need to be setuid, there is an option of shipping firmware
builtin into kernel without initrd/filesystem.
Enable Kernel Panic on OOPS
When fuzzing the kernel or attempting kernel exploits attackers are likely to trigger kernel OOPSes. Setting the behavior on OOPS to PANIC can impede their progress.
This configuration is supported in Linux 3.5 and greater and thus should only be enabled for such versions.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-General-PanicOnOOPS-1 | CONFIG_PANIC_ON_OOPS |
y |
Disable socket monitoring interface
These monitors can be used to inspect shared file descriptors on Unix Domain sockets or traffic on 'localhost' which is otherwise assumed to be confidential.
The CONFIG_PACKET_DIAG configuration is supported in Linux 3.7 and greater
and thus should only be disabled for such versions.
The CONFIG_UNIX_DIAG configuration is supported in Linux 3.3 and greater
and thus should only be disabled for such versions.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-General-SocketMon-1 | CONFIG_PACKET_DIAG |
n |
| Kernel-General-SocketMon-2 | CONFIG_UNIX_DIAG |
n |
Disable BPF JIT
The BPF JIT can be used to create kernel-payloads from firewall table rules.
This configuration for is supported in Linux 3.16 and greater and thus should only be disabled for such versions.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-General-BPF_JIT-1 | CONFIG_BPF_JIT |
n |
Enable Enforced Module Signing
The kernel should never allow an unprivileged user the ability to load specific kernel modules, since that would provide a facility to unexpectedly extend the available attack surface.
To protect against even privileged users, systems may need to either disable
module loading entirely, or provide signed modules (e.g.
CONFIG_MODULE_SIG_FORCE, or dm-crypt with LoadPin), to keep from having root
load arbitrary kernel code via the module loader interface.
This configuration is supported in Linux 3.7 and greater and thus should only be enabled for such versions.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-General-ModuleSigning-1 | CONFIG_MODULE_SIG_FORCE |
y |
It is also possible to block the loading of modules after startup with "kernel.modules_disabled".
| Domain | Variable name |
Value |
|---|---|---|
| Kernel-General-ModuleSigning-2 | kernel.modules_disabled |
1 |
Disable all USB, PCMCIA (and other hotplug bus) drivers that aren't needed
To reduce the attack surface, the driver enumeration, probe, and operation happen in the kernel. The driver data is parsed by the kernel, so any logic bugs in these drivers can become kernel exploits.
| Domain | Object | State |
|---|---|---|
| Kernel-General-Drivers-1 | USB |
Disabled |
| Kernel-General-Drivers-2 | PCMCIA |
Disabled |
| Kernel-General-Drivers-3 | Other hotplug bus |
Disabled |
Position Independent Executables
| Domain | Improvement |
|---|---|
| Kernel-General-IndependentExec-1 | Kernel or/and platform part ? |
| Domain | compiler and linker options |
State |
|---|---|---|
| Kernel-General-IndependentExec-1 | -pie -fpic |
Enable |
Produce a position independent executable on targets which supports it.
Prevent Overwrite Attacks
-z,relro linking option helps during program load, several ELF memory sections
need to be written by the linker, but can be turned read-only before turning
over control to the program. This prevents some Global Offset Table GOT
overwrite attacks, or in the dtors section of the ELF binary.
| Domain | compiler and linker options |
State |
|---|---|---|
| Kernel-General-OverwriteAttacks-1 | -z,relro |
Enable |
| Kernel-General-OverwriteAttacks-2 | -z,now |
Enable |
During program load, all dynamic symbols are resolved, allowing for the complete
GOT to be marked read-only (due to -z relro above). This prevents GOT
overwrite attacks. For very large application, this can incur some performance
loss during initial load while symbols are resolved, but this shouldn't be an
issue for daemons.
Library linking
| Domain | Improvement |
|---|---|
| Kernel-General-LibraryLinking-1 | Keep this part? |
It is recommended that dynamic linking should generally not be allowed. This will avoid the user from replacing a library with malicious library.
| Domain | Object | Recommendations |
|---|---|---|
| Kernel-General-LibraryLinking-1 | Dynamic linking | Should generally not be allowed. |
Linking everything statically doesn't change anything wrt security as binaries
will live under same user:group as libraries and setuid executables ignore
LD_PRELOAD/LD_LIBRARY_PATH. It also increases RSS footprint and creates
problems with upgrading.
Memory
Restrict access to kernel memory
The /dev/kmem file in Linux systems is directly mapped to kernel virtual memory. This can be disastrous if an attacker gains root access, as the attacker would have direct access to kernel virtual memory.
To disable the /dev/kmem file, which is very infrequently used by applications, the following kernel option should be set in the compile-time kernel configuration:
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Memory-RestrictAccess-1 | CONFIG_DEVKMEM |
n |
In case applications in userspace need /dev/kmem support, it should be available only for authenticated applications.
Disable access to a kernel core dump
This kernel configuration disables access to a kernel core dump from user space. If enabled, it gives attackers a useful view into kernel memory.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Memory-CoreDump-1 | CONFIG_PROC_KCORE |
n |
Disable swap
If not disabled, attackers can enable swap at runtime, add pressure to the memory subsystem and then scour the pages written to swap for useful information.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Memory-Swap-1 | CONFIG_SWAP |
n |
- Enabling swap at runtime require
CAP_SYS_ADMIN. - Swap block device is usually under root:disk.
- Linux never swaps kernel pages.
- If swap disabling is not possible, swap encryption should be enabled.
Disable "Load All Symbols"
There is a /proc/kallsyms file which exposes the kernel memory space address of many kernel symbols (functions, variables, etc...). This information is useful to attackers in identifying kernel versions/configurations and in preparing payloads for the exploits of kernel space.
Both KALLSYMS_ALL and KALLSYMS shall be disabled;
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Memory-LoadAllSymbols-1 | CONFIG_KALLSYMS |
n |
| Kernel-Memory-LoadAllSymbols-2 | CONFIG_KALLSYMS_ALL |
n |
Stack protection
To prevent stack-smashing, similar to the stack protector used for ELF programs in user-space, the kernel can protect its internal stacks as well.
This configuration is supported in Linux 3.11 and greater and thus should only be enabled for such versions.
This configuration also requires building the kernel with the gcc compiler 4.2 or greater.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Memory-Stack-1 | CONFIG_CC_STACKPROTECTOR |
y |
Other defenses include things like shadow stacks.
Disable access to /dev/mem
The /dev/mem file in Linux systems is directly mapped to physical memory. This can be disastrous if an attacker gains root access, as the attacker would have direct access to physical memory through this convenient device file. It may not always be possible to disable such file, as some applications might need such support. In that case, then this device file should be available only for authenticated applications.
This configuration is supported in Linux 4.0 and greater and thus should only be disabled for such versions.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Memory-Access-1 | CONFIG_DEVMEM |
n |
Disable cross-memory attach
Disable the process_vm_*v syscalls which allow one process to peek/poke the virtual memory of another.
This configuration is supported in Linux 3.5 and greater and thus should only be disabled for such versions.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Memory-CrossMemAttach-1 | CROSS_MEMORY_ATTACH |
n |
Stack Smashing Attacks
| Domain | compiler and linker options |
State |
|---|---|---|
| Kernel-Memory-StackSmashing-1 | -fstack-protector-all |
Enable |
Emit extra code to check for buffer overflows, such as stack smashing attacks.
Detect Buffer Overflows
| Domain | compiler options and config name |
Value |
|---|---|---|
| Kernel-Memory-BufferOverflows-1 | -D_FORTIFY_SOURCE |
2 |
| Kernel-Memory-BufferOverflows-2 | CONFIG_FORTIFY_SOURCE |
y |
Helps detect some buffer overflow errors.
Serial
Disable serial console
The serial console should be disabled to prevent an attacker from accessing this powerful interface.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Consoles-Serial-1 | CONFIG_SERIAL_8250 |
n |
| Kernel-Consoles-Serial-2 | CONFIG_SERIAL_8250_CONSOLE |
n |
| Kernel-Consoles-Serial-3 | CONFIG_SERIAL_CORE |
n |
| Kernel-Consoles-Serial-4 | CONFIG_SERIAL_CORE_CONSOLE |
n |
Bake-in the kernel command-line
The kernel command-line is used to control many aspects of the booting kernel, and is prone to tampering as they are passed in RAM with little to no reverse validation on these parameters. To prevent this type of attack, the kernel shall be configured to ignore commands line arguments, and use pre-configured (compile time) options instead.
Set the kernel command line in the CONFIG_CMDLINE KConfig item and then pass
no arguments from the bootloader.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Consoles-CommandLine-1 | CONFIG_CMDLINE_BOOL |
y |
| Kernel-Consoles-CommandLine-2 | CONFIG_CMDLINE |
"insert kernel command line here" |
| Kernel-Consoles-CommandLine-3 | CONFIG_CMDLINE_OVERRIDE |
y |
It is recommended that any per-device settings (e.g: MAC addresses, serial numbers, etc.) be stored and accessed from read-only memory (or files), and that any such parameters be verified (signature checking) prior to their use.
Disable KGDB
The Linux kernel supports KGDB over USB and console ports. These mechanisms are
controlled by the kgdbdbgp and kgdboc kernel command-line parameters. It is
important to ensure that no shipping product contains a kernel with KGDB
compiled-in.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Consoles-KDBG-1 | CONFIG_KGDB |
n |
Disable magic sysrq support
On a few architectures, you can access a powerful debugger interface from the keyboard. The same powerful interface can be present on the serial console (responding to serial break) of Linux on other architectures. Disable to avoid potentially exposing this powerful backdoor.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Consoles-SysRQ-1 | CONFIG_MAGIC_SYSRQ |
n |
Disable support for binary formats other than ELF
This will make possible to plug wrapper-driven binary formats into the kernel. It enables support for binary formats other than ELF. Providing the ability to use alternate interpreters would assist an attacker in discovering attack vectors.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Consoles-BinaryFormat-1 | CONFIG_BINFMT_MISC |
n |
Debug
No debuggers shall be present on the file system. This includes, but is not
limited to, the GNU Debugger client/server (commonly known in their short form
names such as the gdb and gdbserver executable binaries respectively), the
LLDB next generation debugger or the TCF (Target Communications Framework)
agnostic framework. Including these binaries as part of the file system will
facilitate an attacker's ability to reverse engineer and debug (either locally
or remotely) any process that is currently executing on the device.
Kernel debug symbols
Debug symbols should always be removed from production kernels as they provide a lot of information to attackers.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Debug-Symbols-1 | CONFIG_DEBUG_INFO |
n |
These kernel debug symbols are enabled by other config items in the kernel. Care
should be taken to disable those also. If CONFIG_DEBUG_INFO cannot be
disabled, then enabling CONFIG_DEBUG_INFO_REDUCED is second best.
At least CONFIG_DEBUG_INFO_REDUCED should be always enabled for developers to
convert addresses in oops messages to line numbers.
Disable Kprobes
Kprobes enables you to dynamically break into any kernel routine and collect debugging and performance information non-disruptively. You can trap at almost any kernel code address, specifying a handler routine to be invoked when the breakpoint is hit.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Debug-Kprobes-1 | CONFIG_KPROBES |
n |
Disable Tracing
FTrace enables the kernel to trace every kernel function. Providing kernel trace functionality would assist an attacker in discovering attack vectors.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Debug-Tracing-1 | CONFIG_FTRACE |
n |
Disable Profiling
Profiling and OProfile enables profiling the whole system, include the kernel, kernel modules, libraries, and applications. Providing profiling functionality would assist an attacker in discovering attack vectors.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Debug-Profiling-1 | CONFIG_OPROFILE |
n |
| Kernel-Debug-Profiling-2 | CONFIG_PROFILING |
n |
Disable OOPS print on BUG()
The output from OOPS print can be helpful in Return Oriented Programming (ROP) when trying to determine the effectiveness of an exploit.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Debug-OOPSOnBUG-1 | CONFIG_DEBUG_BUGVERBOSE |
n |
Disable Kernel Debugging
There are development-only branches of code in the kernel enabled by the
DEBUG_KERNEL conf. This should be disabled to compile-out these branches.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Debug-Dev-1 | CONFIG_DEBUG_KERNEL |
n |
| Kernel-Debug-Dev-2 | CONFIG_EMBEDDED |
n |
In some kernel versions, disabling this requires also disabling
CONFIG_EMBEDDED, and CONFIG_EXPERT. Disabling CONFIG_EXPERT makes it
impossible to disable COREDUMP, DEBUG_BUGVERBOSE, NAMESPACES, KALLSYMS
and BUG. In which case it is better to leave this enabled than enable the
others.
Disable the kernel debug filesystem
The kernel debug filesystem presents a lot of useful information and means of manipulation of the kernel to an attacker.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Debug-FileSystem-1 | CONFIG_DEBUG_FS |
n |
Disable BUG() support
The kernel will display backtrace and register information for BUGs and WARNs in kernel space, making it easier for attackers to develop exploits.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Debug-BUG-1 | CONFIG_BUG |
n |
Disable core dumps
Core dumps provide a lot of debug information for hackers. So disabling core dumps are recommended in production builds.
This configuration is supported in Linux 3.7 and greater and thus should only be disabled for such versions.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Debug-CoreDumps-1 | CONFIG_COREDUMP |
n |
Kernel Address Display Restriction
When attackers try to develop "run anywhere" exploits for kernel vulnerabilities, they frequently need to know the location of internal kernel structures. By treating kernel addresses as sensitive information, those locations are not visible to regular local users.
/proc/sys/kernel/kptr_restrict is set to "1" to block the reporting of known kernel address leaks.
| Domain | File name |
Value |
|---|---|---|
| Kernel-Debug-AdressDisplay-1 | /proc/sys/kernel/kptr_restrict |
1 |
Additionally, various files and directories should be readable only by the root
user: /boot/vmlinuz*, /boot/System.map*, /sys/kernel/debug/,
/proc/slabinfo
| Domain | File or Directorie name |
State |
|---|---|---|
| Kernel-Debug-AdressDisplay-1 | /boot/vmlinuz* |
Readable Only for root user |
| Kernel-Debug-AdressDisplay-2 | /boot/System.map* |
Readable Only for root user |
| Kernel-Debug-AdressDisplay-3 | /sys/kernel/debug/ |
Readable Only for root user |
| Kernel-Debug-AdressDisplay-4 | /proc/slabinfo |
Readable Only for root user |
DMESG Restrictions
When attackers try to develop "run anywhere" exploits for vulnerabilities, they
frequently will use dmesg output. By treating dmesg output as sensitive
information, this output is not available to the attacker.
/proc/sys/kernel/dmesg_restrict can be set to "1" to treat dmesg output as sensitive.
| Domain | File name |
Value |
|---|---|---|
| Kernel-Debug-DMESG-1 | /proc/sys/kernel/dmesg_restrict |
1 |
Enable the below compiler and linker options when building user-space applications to avoid stack smashing, buffer overflow attacks.
Disable /proc/config.gz
It is extremely important to not expose the kernel configuration used on a production device to a potential attacker. With access to the kernel config, it could be possible for an attacker to build a custom kernel for the device that may disable critical security features.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-Debug-Config-1 | CONFIG_IKCONFIG |
n |
File System
Disable all file systems not needed
To reduce the attack surface, file system data is parsed by the kernel, so any logic bugs in file system drivers can become kernel exploits.
Disable NFS file system
NFS FileSystems are useful during development phases, but this can be a very helpful way for an attacker to get files when you are in production mode, so we must disable them.
| Domain | Config name |
Value |
|---|---|---|
| Kernel-FileSystems-NFS-1 | CONFIG_NFSD |
n |
| Kernel-FileSystems-NFS-2 | CONFIG_NFS_FS |
n |
Partition Mount Options
There are several security restrictions that can be set on a filesystem when it is mounted. Some common security options include, but are not limited to:
nosuid - Do not allow set-user-identifier or set-group-identifier bits to take
effect.
nodev - Do not interpret character or block special devices on the filesystem.
noexec - Do not allow execution of any binaries on the mounted filesystem.
ro - Mount filesystem as read-only.
The following flags shall be used for mounting common filesystems:
| Domain | Partition |
Value |
|---|---|---|
| Kernel-FileSystems-Mount-1 | /boot |
nosuid, nodev and noexec. |
| Kernel-FileSystems-Mount-2 | /var & /tmp |
In /etc/fstab or vfstab, add nosuid, nodev and noexec. |
| Kernel-FileSystems-Mount-3 | Non-root local | If type is ext2 or ext3 and mount point not '/', add nodev. |
| Kernel-FileSystems-Mount-4 | Removable storage | Add nosuid, nodev and noexec. |
| Kernel-FileSystems-Mount-5 | Temporary storage | Add nosuid, nodev and noexec. |
| Kernel-FileSystems-Mount-6 | /dev/shm |
Add nosuid, nodev and noexec. |
| Kernel-FileSystems-Mount-7 | /dev |
Add nosuid and noexec. |
If CONFIG_DEVTMPFS_MOUNT is set, then the kernel will mount /dev and will not
apply the nosuid, noexec options. Either disable CONFIG_DEVTMPFS_MOUNT or
add a remount with noexec and nosuid options to system startup.
| Domain | Config name |
State or Value |
|---|---|---|
| Kernel-FileSystems-Mount-1 | CONFIG_DEVTMPFS_MOUNT |
Disabled or add remount with noexec and nosuid to system startup. |