This section is inspired by the Kernel Self Protection Project (KSPP). It attempts to implement all recommended Linux kernel settings by the KSPP and many more sources.
sysctl settings are configured via the /usr/lib/sysctl.d/990-security-misc.conf
configuration file and significant hardening is applied to a myriad of components.
Kernel space:
-
Restrict access to kernel addresses through the use of kernel pointers regardless of user privileges.
-
Restrict access to the kernel logs to
CAP_SYSLOG
as they often contain sensitive information. -
Prevent kernel information leaks in the console during boot.
-
Restrict usage of
bpf()
toCAP_BPF
to prevent the loading of BPF programs by unprivileged users. -
Restrict loading TTY line disciplines to
CAP_SYS_MODULE
. -
Restrict the
userfaultfd()
syscall toCAP_SYS_PTRACE
, which reduces the likelihood of use-after-free exploits. -
Disable
kexec
as it can be used to replace the running kernel. -
Entirely disable the SysRq key so that the Secure Attention Key (SAK) can no longer be utilized. See documentation.
-
Optional - Restrict user namespaces to
CAP_SYS_ADMIN
as they can lead to substantial privilege escalation. -
Optional - Disable all use of user namespaces.
-
Restrict kernel profiling and the performance events system to
CAP_PERFMON
. -
Force the kernel to panic on both "oopses", which can potentially indicate and thwart certain kernel exploitation attempts, and also kernel warnings in the
WARN()
path. -
Optional - Force immediate reboot on the occurrence of a single kernel panic and also (when using Linux kernel >= 6.2) limit the number of allowed panics to one.
-
Disable the use of legacy TIOCSTI operations which can be used to inject keypresses.
-
Disable asynchronous I/O (when using Linux kernel >= 6.6) as
io_uring
has been the source of numerous kernel exploits.
User space:
-
Restrict usage of
ptrace()
to only processes withCAP_SYS_PTRACE
as it enables programs to inspect and modify other active processes. Optional - Disable usage ofptrace()
by all processes. -
Maximize the bits of entropy used for mmap ASLR across all CPU architectures.
-
Prevent hardlink and symlink TOCTOU races in world-writable directories.
-
Disallow unintentional writes to files in world-writable directories unless they are owned by the directory owner to mitigate some data spoofing attacks.
-
Randomize the addresses (ASLR) for mmap base, stack, VDSO pages, and heap.
-
Raise the minimum address a process can request for memory mapping to 64KB to protect against kernel null pointer dereference vulnerabilities.
-
Increase the maximum number of memory map areas a process is able to utilize to 1,048,576.
-
Optional - Disallow registering interpreters for various (miscellaneous) binary formats based on a magic number or their file extension to prevent unintended code execution. See issue: #267
Core dumps:
- Disable core dump files and prevent their creation. If core dump files are
enabled, they will be named based on
core.PID
instead of the defaultcore
.
Swap space:
- Limit the copying of potentially sensitive content in memory to the swap device.
Networking:
-
Enable hardening of the BPF JIT compiler protect against JIT spraying.
-
Enable TCP SYN cookie protection to assist against SYN flood attacks.
-
Protect against TCP time-wait assassination hazards.
-
Enable reverse path filtering (source validation) of packets received from all interfaces to prevent IP spoofing.
-
Disable ICMP redirect acceptance and redirect sending messages to prevent man-in-the-middle attacks and minimize information disclosure.
-
Ignore ICMP echo requests to prevent clock fingerprinting and Smurf attacks.
-
Ignore bogus ICMP error responses.
-
Disable source routing which allows users to redirect network traffic that can result in man-in-the-middle attacks.
-
Do not accept IPv6 router advertisements and solicitations.
-
Optional - Disable SACK and DSACK as they have historically been a known vector for exploitation.
-
Disable TCP timestamps as they can allow detecting the system time.
-
Optional - Log packets with impossible source or destination addresses to enable further inspection and analysis.
-
Optional - Enable IPv6 Privacy Extensions.
Mitigations for known CPU vulnerabilities are enabled in their strictest form
and simultaneous multithreading (SMT) is disabled. See the
/etc/default/grub.d/40_cpu_mitigations.cfg
configuration file.
Boot parameters relating to kernel hardening, DMA mitigations, and entropy
generation are outlined in the /etc/default/grub.d/40_kernel_hardening.cfg
configuration file.
Kernel space:
-
Disable merging of slabs with similar size, which reduces the risk of triggering heap overflows and limits influencing slab cache layout.
-
Enable sanity checks and red zoning via slab debugging. This will implicitly disable kernel pointer hashing, leaking very sensitive information to root.
-
Enable memory zeroing at both allocation and free time, which mitigates some use-after-free vulnerabilities by erasing sensitive information in memory.
-
Enable the kernel page allocator to randomize free lists to limit some data exfiltration and ROP attacks, especially during the early boot process.
-
Enable kernel page table isolation to increase KASLR effectiveness and also mitigate the Meltdown CPU vulnerability.
-
Enable randomization of the kernel stack offset on syscall entries to harden against memory corruption attacks.
-
Disable vsyscalls as they are vulnerable to ROP attacks and have now been replaced by vDSO.
-
Restrict access to debugfs by not registering the file system since it can contain sensitive information.
-
Force kernel panics on "oopses" to potentially indicate and thwart certain kernel exploitation attempts.
-
Optional - Modify the machine check exception handler.
-
Prevent sensitive kernel information leaks in the console during boot.
-
Enable the kernel Electric-Fence sampling-based memory safety error detector which can identify heap out-of-bounds access, use-after-free, and invalid-free errors.
-
Disable 32-bit vDSO mappings as they are a legacy compatibility feature.
-
Optional - Use kCFI as the default CFI implementation (when using Linux kernel >= 6.2) since it may be slightly more resilient to attacks that are able to write arbitrary executables in memory.
-
Optional - Disable support for all x86 processes and syscalls (when using Linux kernel >= 6.7) to reduce attack surface.
Direct memory access:
-
Enable strict IOMMU translation to protect against some DMA attacks via the use of both CPU manufacturer-specific drivers and kernel settings.
-
Clear the busmaster bit on all PCI bridges during the EFI hand-off, which disables DMA before the IOMMU is configured. May cause boot failure on certain hardware.
Entropy:
-
Do not credit the CPU or bootloader as entropy sources at boot in order to maximize the absolute quantity of entropy in the combined pool.
-
Obtain more entropy at boot from RAM as the runtime memory allocator is being initialized.
Networking:
- Optional - Disable the entire IPv6 stack to reduce attack surface.
- The bits of entropy used for mmap ASLR for all CPU architectures are maxed
out via
/usr/libexec/security-misc/mmap-rnd-bits
(set to the values ofCONFIG_ARCH_MMAP_RND_BITS_MAX
andCONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX
that the kernel was built with), therefore improving its effectiveness.
Summary:
security-misc
is in full compliance with KSPP recommendations wherever feasible. However,
there are a few cases of partial or non-compliance due to technical limitations.
Full compliance:
More than 30 kernel boot parameters and over 30 sysctl settings are fully aligned with the KSPP's recommendations.
Partial compliance:
sysctl kernel.yama.ptrace_scope=3
Completely disables ptrace()
. Can be enabled easily if needed.
sysctl kernel.panic=-1
Forces an immediate reboot on kernel panic. This can be enabled, but it may lead to unexpected system crashes.
Non-compliance:
sysctl user.max_user_namespaces=0
Disables user namespaces entirely. Not recommended due to the potential for widespread breakages.
sysctl fs.binfmt_misc.status=0
Disables the registration of interpreters for miscellaneous binary formats. Currently not feasible due to compatibility issues with Firefox.
Not yet implemented due to issues:
- https://forums.whonix.org/t/enforce-kernel-module-software-signature-verification-module-signing-disallow-kernel-module-loading-by-default/7880/64
- dell/dkms#359
See:
/etc/default/grub.d/40_signed_modules.cfg
Not yet implemented due to issues:
A systemd service dynamically sets the kernel parameter modules_disabled
to 1,
preventing new modules from being loaded. Since this isn't configured directly
within systemctl, it does not break the loading of legitimate and necessary
modules for the user, like drivers etc., given they are plugged in on startup.
Conntrack: Deactivates Netfilter's connection tracking helper module which
increases kernel attack surface by enabling superfluous functionality such
as IRC parsing in the kernel. See /etc/modprobe.d/30_security-misc_conntrack.conf
.
Certain kernel modules are blacklisted by default to reduce attack surface via
/etc/modprobe.d/30_security-misc_blacklist.conf
. Blacklisting prevents kernel
modules from automatically starting.
-
CD-ROM/DVD: Blacklist modules required for CD-ROM/DVD devices.
-
Miscellaneous: Blacklist an assortment of other modules to prevent them from automatically loading.
Specific kernel modules are entirely disabled to reduce attack surface via
/etc/modprobe.d/30_security-misc_disable.conf
. Disabling prohibits kernel
modules from starting. This approach should not be considered comprehensive;
rather, it is a form of badness enumeration. Any potential candidates for future
disabling should first be blacklisted for a suitable amount of time.
Hardware modules:
-
Optional - Bluetooth: Disabled to reduce attack surface.
-
FireWire (IEEE 1394): Disabled as they are often vulnerable to DMA attacks.
-
GPS: Disable GPS-related modules such as those required for Global Navigation Satellite Systems (GNSS).
-
Optional - Intel Management Engine (ME): Provides some disabling of the interface between the Intel ME and the OS. May lead to breakages in places such as firmware updates, security, power management, display, and DRM. See discussion: #239
-
Intel Platform Monitoring Technology (PMT) Telemetry: Disable some functionality of the Intel PMT components.
-
Thunderbolt: Disabled as they are often vulnerable to DMA attacks.
File system modules:
-
File Systems: Disable uncommon and legacy file systems.
-
Network File Systems: Disable uncommon and legacy network file systems.
Networking modules:
- Network Protocols: A wide array of uncommon and legacy network protocols and drivers are disabled.
Miscellaneous modules:
-
Amateur Radios: Disabled to reduce attack surface.
-
Optional - CPU MSRs: Disabled as can be abused to write to arbitrary memory.
-
Floppy Disks: Disabled to reduce attack surface.
-
Framebuffer (fbdev): Disabled as these drivers are well-known to be buggy, cause kernel panics, and are generally only used by legacy devices.
-
Replaced Modules: Disabled legacy drivers that have been entirely replaced and superseded by newer drivers.
-
Optional - USB Video Device Class: Disables the USB-based video streaming driver for devices like some webcams and digital camcorders.
-
Vivid: Disabled to reduce attack surface given previous vulnerabilities.
- A systemd service clears the System.map file on boot as these contain kernel pointers. The file is completely overwritten with zeroes to ensure it cannot be recovered. See:
/etc/kernel/postinst.d/30_remove-system-map
/lib/systemd/system/remove-system-map.service
/usr/libexec/security-misc/remove-system.map
- Coredumps are disabled as they may contain important information such as encryption keys or passwords. See:
/etc/security/limits.d/30_security-misc.conf
/etc/sysctl.d/30_security-misc.conf
/lib/systemd/coredump.conf.d/30_security-misc.conf
- An initramfs hook sets the sysctl values in
/etc/sysctl.conf
and/etc/sysctl.d
before init is executed so sysctl hardening is enabled as early as possible. This is implemented forinitramfs-tools
only because this is not needed fordracut
asdracut
does that by default, at least onsystemd
enabled systems. Not researched for non-systemd
systems by the author of this part of the readme.
Not yet implemented due to issues:
-
Unlike version 4, IPv6 addresses can provide information not only about the originating network but also the originating device. We prevent this from happening by enabling the respective privacy extensions for IPv6.
-
In addition, we deny the capability to track the originating device in the network at all, by using randomized MAC addresses per connection by default.
See:
/usr/lib/NetworkManager/conf.d/80_ipv6-privacy.conf
/usr/lib/NetworkManager/conf.d/80_randomize-mac.conf
/usr/lib/systemd/networkd.conf.d/80_ipv6-privacy-extensions.conf
-
Default Behavior: Although Bluetooth capability is 'enabled' in the kernel, security-misc deviates from the usual behavior by starting with Bluetooth turned off at system start. This setting remains until the user explicitly opts to activate Bluetooth.
-
User Control: Users have the freedom to easily switch Bluetooth on and off in the usual way, exercising their own discretion. This can be done via the Bluetooth toggle through the usual way, that is either through GUI settings application or command line commands.
-
Enhanced Privacy Settings: We enforce more private defaults for Bluetooth connections. This includes the use of private addresses and strict timeout settings for discoverability and visibility.
-
Security Considerations: Despite these measures, it's important to note that Bluetooth technology, by its nature, may still be prone to exploits due to its history of security vulnerabilities. Thus, we recommend users to opt-out of using Bluetooth when possible.
- See configuration:
/etc/bluetooth/30_security-misc.conf
- For more information and discussion: GitHub Pull Request
-
Disabling Bluetooth: This means the absence of the Bluetooth kernel module. When disabled, Bluetooth is non-existent in the system - it cannot be seen, set, configured, or interacted with in any way.
-
Turning Bluetooth On/Off: This refers to a software toggle. Normally, on Debian systems, Bluetooth is 'on' when the system boots up. It actively searches for known devices to auto-connect and may be discoverable or visible under certain conditions. Our default ensures that Bluetooth is off on startup. However, it remains 'enabled' in the kernel, meaning the kernel can use the Bluetooth protocol and has the necessary modules.
-
Turning Bluetooth On: Simply click the Bluetooth button in the settings application or on the tray, and switch the toggle. It's a straightforward action that can be completed in less than a second.
-
Turning Bluetooth Off: Follow the same procedure as turning it on but switch the toggle to the off position.
-
The
jitterentropy_rng
kernel module is loaded as early as possible during boot to gather more entropy via the/usr/lib/modules-load.d/30_security-misc.conf
configuration file. -
Distrusts the CPU for initial entropy at boot as it is not possible to audit, may contain weaknesses or a backdoor. Similarly, do not credit the bootloader seed for initial entropy. For references, see:
/etc/default/grub.d/40_kernel_hardening.cfg
-
Gathers more entropy during boot if using the linux-hardened kernel patch.
A systemd service is triggered on boot to remount all sensitive partitions and directories with significantly more secure hardened mount options. Since this would require manual tuning for a given specific system, we handle it by creating a very solid configuration file for that very system on package installation.
Not enabled by default yet. In development. Help welcome.
- https://www.kicksecure.com/wiki/Dev/remount-secure
- #157
- https://forums.whonix.org/t/re-mount-home-and-other-with-noexec-and-nosuid-among-other-useful-mount-options-for-better-security/
-
su
is restricted to only users within the groupsudo
which prevents users from usingsu
to gain root access or to switch user accounts -/usr/share/pam-configs/wheel-security-misc
(which results in a change in file/etc/pam.d/common-auth
). -
Add user
root
to groupsudo
. This is required due to the above restriction so that logging in from a virtual console is still possible -debian/security-misc.postinst
-
Abort login for users with locked passwords -
/usr/libexec/security-misc/pam-abort-on-locked-password
. -
Logging into the root account from a virtual, serial, or other console is prevented by shipping an existing and empty
/etc/securetty
file (deletion of/etc/securetty
has a different effect).
This package does not yet automatically lock the root account password. It is not clear if this would be sane in such a package, although it is recommended to lock and expire the root account.
In new Kicksecure builds, the root account will be locked by package dist-base-files.
See:
- https://www.kicksecure.com/wiki/Root
- https://www.kicksecure.com/wiki/Dev/Permissions
- https://forums.whonix.org/t/restrict-root-access/7658
However, a locked root password will break rescue and emergency shell. Therefore, this package enables passwordless rescue and emergency shell. This is the same solution that Debian will likely adopt for the Debian installer: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=802211
See:
/etc/systemd/system/emergency.service.d/override.conf
/etc/systemd/system/rescue.service.d/override.conf
Adverse security effects can be prevented by setting up BIOS password protection, GRUB password protection, and/or full disk encryption.
This uses pam_access to allow members of group console
to use the console but
restrict everyone else (except members of group console-unrestricted
) from
using the console with ancient, unpopular login methods such as /bin/login
over
networks as this might be exploitable. (CVE-2001-0797)
This is not enabled by default in this package since this package does not know which users should be added to group 'console' and thus, would break console access.
See:
/usr/share/pam-configs/console-lockdown-security-misc
/etc/security/access-security-misc.conf
User accounts are locked after 50 failed login attempts using pam_faillock
.
Informational output during Linux PAM:
- Show failed and remaining password attempts.
- Document unlock procedure if Linux user account got locked.
- Point out that there is no password feedback for
su
. - Explain locked root account if locked.
See:
/usr/share/pam-configs/tally2-security-misc
/usr/libexec/security-misc/pam-info
/usr/libexec/security-misc/pam-abort-on-locked-password
Read, write, and execute access for "others" are removed during package
installation, upgrade, or PAM mkhomedir
for all users who have home folders in
/home
by running, for example:
chmod o-rwx /home/user
This will be done only once per folder in /home
so users who wish to relax
file permissions are free to do so. This is to protect files in a home folder
that were previously created with lax file permissions prior to the installation
of this package.
See:
debian/security-misc.postinst
/usr/libexec/security-misc/permission-lockdown
/usr/share/pam-configs/mkhomedir-security-misc
Default umask
is set to 027
for files created by non-root users such as
user user
. Broken. Disabled. See:
This is done using the PAM module pam_mkhomedir.so umask=027
.
This means files created by non-root users cannot be read by other non-root
users by default. While Permission Lockdown already protects the /home
folder,
this protects other folders such as /tmp
.
group
read permissions are not removed. This is unnecessary due to Debian's
use of User Private Groups (UPGs). See also:
https://wiki.debian.org/UserPrivateGroups
Default umask
is unchanged for root because then configuration files created
in /etc
by the system administrator would be unreadable by "others" and break
applications. Examples include /etc/firefox-esr
and /etc/thunderbird
.
See:
/usr/share/pam-configs/umask-security-misc
A systemd service removes SUID / SGID bits from non-essential binaries as these are often used in privilege escalation attacks.
Various file permissions are reset with more secure and hardened defaults. These include but are not limited to:
- Limiting
/home
and/root
to the root only. - Limiting crontab to root as well as all the configuration files for cron.
- Limiting the configuration for cups and ssh.
- Protecting the information of sudoers from others.
- Protecting various system-relevant files and modules.
permission-hardener
removes SUID / SGID bits from non-essential binaries as
these are often used in privilege escalation attacks. It is enabled by default
and applied at security-misc package installation and upgrade time.
There is also an optional systemd unit which does the same at boot time that
can be enabled by running systemctl enable permission-hardener.service
as
root. The hardening at boot time is not the default because this slows down
the boot process too much.
See:
/usr/bin/permission-hardener
debian/security-misc.postinst
/lib/systemd/system/permission-hardener.service
/etc/permission-hardener.d
- https://forums.whonix.org/t/disable-suid-binaries/7706
- https://www.kicksecure.com/wiki/SUID_Disabler_and_Permission_Hardener
This is not enabled yet because hidepid is not enabled by default.
Calls to pkexec
are redirected to lxqt-sudo
because pkexec
is
incompatible with hidepid=2
.
See:
/usr/bin/pkexec.security-misc
- https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=860040
- https://forums.whonix.org/t/cannot-use-pkexec/8129
- Enables "
apt-get --error-on=any
" which makes apt exit non-zero for transient failures. -/etc/apt/apt.conf.d/40error-on-any
. - Enables APT seccomp-BPF sandboxing -
/etc/apt/apt.conf.d/40sandbox
. - Deactivates previews in Dolphin.
- Deactivates previews in Nautilus -
/usr/share/glib-2.0/schemas/30_security-misc.gschema.override
. - Deactivates thumbnails in Thunar.
- Rationale: lower attack surface when using the file manager
- https://forums.whonix.org/t/disable-preview-in-file-manager-by-default/18904
- Thunderbird is hardened with the following options:
- Displays domain names in punycode to prevent IDN homograph attacks (a form of phishing).
- Strips email client information from sent email headers.
- Strips user time information from sent email headers by replacing the originating time zone with UTC and rounding the timestamp to the nearest minute.
- Disables scripting when viewing PDF files.
- Disables implicit outgoing connections.
- Disables all and any kind of telemetry.
- Security and privacy enhancements for gnupg's config file
/etc/skel/.gnupg/gpg.conf
. See also:
Added in December 2023.
Before sending pull requests to harden arbitrary applications, please note the scope of security-misc is limited to default installed applications in Kicksecure and Whonix. This includes:
- Thunderbird, VLC Media Player, KeePassXC
- Debian Specific System Components (APT, DPKG)
- System Services (NetworkManager IPv6 privacy options, MAC address randomization)
- Actually used development utilities such as
git
.
It will not be possible to review and merge "1500" settings profiles for arbitrary applications outside of this context.
The main objective of security-misc is to harden Kicksecure and its derivatives,
such as Whonix, by implementing robust security settings. It's designed to be
compatible with Debian, reflecting a commitment to clean implementation and
sound design principles. However, it's important to note that security-misc is a
component of Kicksecure, not a substitute for it. The intention isn't to
recreate Kicksecure within security-misc. Instead, specific security
enhancements, like recommending a curated list of security-focused
default packages (e.g., libpam-tmpdir
), should be integrated directly into
those appropriate areas of Kicksecure (e.g. kicksecure-meta-packages
).
Discussion: #154
Added in December 2023.
Maintainability is a key priority [1]. Before modifying settings in the downstream security-misc, it's essential to first engage with upstream developers to propose these changes as defaults. This step should only be bypassed if there's a clear, prior indication from upstream that such changes won't be accepted. Additionally, before implementing any workarounds, consulting with upstream is necessary to avoid future unmaintainable complexity.
If debugging features are disabled, pull requests won't be merged until there is a corresponding pull request for the debug-misc package to re-enable these. This is to avoid configuring the system into a corner where it can no longer be debugged.
[1] https://www.kicksecure.com/wiki/Dev/maintainability
Some hardening is opt-in as it causes too much breakage to be enabled by default.
-
An optional systemd service mounts
/proc
withhidepid=2
at boot to prevent users from seeing another user's processes. This is disabled by default because it is incompatible withpkexec
. It can be enabled by executingsystemctl enable proc-hidepid.service
as root. -
A systemd service restricts
/proc/cpuinfo
,/proc/bus
,/proc/scsi
, and/sys
to the root user. This hides a lot of hardware identifiers from unprivileged users and increases security as/sys
exposes a lot of information that shouldn't be accessible to unprivileged users. As this will break many things, it is disabled by default and can optionally be enabled by executingsystemctl enable hide-hardware-info.service
as root.
-
Hardened malloc compatibility for haveged workaround
/lib/systemd/system/haveged.service.d/30_security-misc.conf
-
Set
dracut
reproducible=yes
setting
/usr/lib/issue.d/20_security-misc.issue
- Linux Kernel Runtime Guard (LKRG)
- tirdad - TCP ISN CPU Information Leak Protection.
- Kicksecure (TM) - a security-hardened Linux Distribution
- And more.
- https://www.kicksecure.com/wiki/Linux_Kernel_Runtime_Guard_LKRG
- https://github.com/Kicksecure/tirdad
- https://www.kicksecure.com
- https://github.com/Kicksecure
Happening primarily in forums.
https://forums.whonix.org/t/kernel-hardening/7296
See https://www.kicksecure.com/wiki/Security-misc#install
Can be build using standard Debian package build tools such as:
dpkg-buildpackage -b
See instructions. (Replace generic-package
with the actual name of this
package security-misc
.)
- A) easy, OR
- B) including verifying software signatures
security-misc
requires donations to
stay alive!