Firejail

Firejail is an easy to use Setuid sandbox program that reduces the risk of security breaches by restricting the running environment of untrusted applications using Linux namespaces, seccomp-bpf and Linux capabilities.

Warning: Running untrusted code is never safe, sandboxing cannot change this.

Installation

Install either firejail, or the firejail-git^AUR package. A GUI application for use with Firejail is also available, firetools.

Note: For information about user_namespaces(7) support in Arch Linux kernels, see Security#Sandboxing applications. Firejail can use it even if it is disabled.

Warning: While upstream is gradually adopting whitelists (cf /etc/firejail/firefox.profile), most of the supplied profiles still rely heavily on blacklists. This means that anything not explicitly forbidden by the profile will be accessible to the application. For example, if you have btrfs snapshots available in /mnt/btrfs, a jailed program may be forbidden from accessing $HOME/.ssh, but would still be able to access /mnt/btrfs/@some-snapshot/$HOME/.ssh. Make sure to audit your profiles; see #Testing profiles.

Configuration

Most users will not require any custom configuration and can proceed to #Usage.

Firejail uses profiles to set the security protections for each of the applications executed inside of it - you can find the default profiles in /etc/firejail/application.profile. Should you require custom profiles for applications not included, or wish to modify the defaults, you may place new rules or copies of the defaults in the ~/.config/firejail directory. You may have multiple custom profile files for a single application, and you may share the same profile file among several applications.

If firejail does not have a profile for a particular application, it uses its restrictive system-wide default profile. This can result in the application not functioning as desired, without first creating a custom and less restrictive profile.

Refer to firejail-profile(5).

Usage

To execute an application using firejail's default protections for that application (the default profile), execute the following:

$ firejail program_name

One-time additions to the default profile can be added as command line options (see firejail(1)). For example, to execute okular with seccomp protection, execute the following:

$ firejail --seccomp okular

You may define multiple non-default profiles for a single program. Once you create your profile file, you can use it by executing:

$ firejail --profile=/absolute/path/to/profile program_name

Using Firejail by default

To use Firejail by default for all applications for which it has profiles, run the firecfg tool with sudo:

$ sudo firecfg

This creates symbolic links in /usr/local/bin/ pointing to /usr/bin/firejail for programs for which Firejail has default or self-created profiles. Note that firecfg(1) only symlinks the programs listed in /etc/firejail/firecfg.config. Certain CLI programs are absent, such as: tar, curl, and git. These need to be symlinked manually. See Profiles not in firecfg #2507 for why they are not included. firecfg additionally adds the current user to Firejail user access database and checks the /usr/share/applications/*.desktop files if they contain the full path to the respective executable, removes the full path and copies them to ~/.local/share/applications/. This ensures that the symlinks in /usr/local/bin/ will be used, which prevents Firejail getting bypassed. If sudo is not installed on your system, you should execute:

# firecfg

as root and

$ firecfg --fix

as user in order to fix the .desktop files.

There may be cases for which you need to manually modify the Exec= line of the .desktop file in ~/.local/share/applications/ to explicitly call Firejail.

Tip: A pacman hook can be used to automatically run firecfg on pacman operations:

/etc/pacman.d/hooks/firejail.hook

[Trigger]
Type = Path
Operation = Install
Operation = Upgrade
Operation = Remove
Target = usr/bin/*
Target = usr/share/applications/*.desktop

[Action]
Description = Configure symlinks in /usr/local/bin based on firecfg.config...
When = PostTransaction
Depends = firejail
Exec = /bin/sh -c 'firecfg >/dev/null 2>&1'

To manually map individual applications, execute:

# ln -s /usr/bin/firejail /usr/local/bin/application

Note:

/usr/local/bin must be set before /usr/bin in the PATH environment variable. The default Arch Linux /etc/profile script satisfies this precondition [1].
To run a symbolic program with custom Firejail setting, simple prefix firejail as seen in #Usage.
For a daemon, you will need to overwrite the systemd unit file for that daemon to call firejail, see systemd#Editing provided units.
Symbolic links to gzip and xz interfere with makepkg's ability to preload libfakeroot.so. See BBS#230913.

Use with hardened_malloc

Note: Some programs will not work using libhardened_malloc.so, such as PyCharm, Firefox.

hardened_malloc^AUR is a hardened implementation of glibc's malloc() allocator, originally written for Android but extended for use on the desktop. While not integrated into glibc yet, it can be used selectively with LD_PRELOAD. The proper way to launch an application within firejail using hardened_malloc is demonstrated below. To make it permanent, you would need to create your own entry in /usr/local/bin/ for the desired application.

$ firejail --env=LD_PRELOAD='/usr/lib/libhardened_malloc.so' /usr/bin/firefox

Alternatively, add the following to a custom profile:

env LD_PRELOAD=/usr/lib/libhardened_malloc.so

Profiles that have private-lib will need the following in custom profiles:

private-lib /lib/libhardened_malloc.so

The various environment variables and settings that can be used to tune hardened_malloc can be found on its github page.

Enable AppArmor support

Since 0.9.60-1, Firejail has supported more direct integration with AppArmor through a generic AppArmor profile. During installation, the profile, firejail-default, is placed in /etc/apparmor.d directory, and needs to be loaded into the kernel by running the following command as root:

# apparmor_parser -r /etc/apparmor.d/firejail-default

See firejail(1) § APPARMOR.

Local customizations of the apparmor profile are supported by editing the file /etc/apparmor.d/local/firejail-local

AppArmor is already enabled for a large number of Firejail profiles. There are several ways to enable AppArmor confinement on top of a Firejail security profile:

Pass the --apparmor flag to Firejail in the command line, e.g. $ firejail --apparmor firefox
Use a custom profile and add the apparmor command.
Enable Apparmor globally in /etc/firejail/globals.local and disable as needed through the use of ignore apparmor in /etc/firejail/ProgramName.local.

Note that enabling AppArmor by above methods always means that /etc/apparmor.d/firejail-default is used. If you rather want to use a specific AppArmor profile for an application, you have to use the above mentioned ignore apparmor command. However, that is not recommended, as using both Firejail and AppArmor for the same applications often creates problems.

Verifying Firejail is being used

$ firejail --list

A more comprehensive output is produced by

$ firejail --tree

Creating custom profiles

Whitelists and blacklists

Blacklists are heavily used in various /etc/firejail/*.inc files which are included in most profiles. Blacklists are permissive:

Deny access to a directory or file and permit everything else: blacklist <directory/file>
Disable/undo/ignore blacklisting a directory or file already blacklisted, e.g., in an *.inc file: noblacklist <directory/file>

The order in which they appear in a profile is important: noblacklist directives must be added above blacklist directives.

Whitelists block everything what is not explicitly whitelisted. They should not be used in profiles for applications that need access to random locations (e.g., text editors, image viewers/editors).

Allow access to a directory or file and forbid everything else: whitelist <directory/file>
Disable/undo/ignore whitelisting a directory or file already whitelisted, e.g., in an *.inc file: nowhitelist <directory/file>

The order in which they appear in a profile is important: nowhitelist directives must be added above whitelist directives.

Whitelisting is always done before blacklisting. As mentioned, a whitelist directive blacklists everything else. A blacklist directive is therefore a fallback if there are no whitelist directives or if a whitelist directive is too permissive.

(no)blacklist and (no)whitelist directives are often used in combination. Example: /etc/firejail/disable-programs.inc (which is included in all profiles) contains the directive:

blacklist ${HOME}/.mozilla

in order to block access to that directory for all applications sandboxed by Firejail. /etc/firejail/firefox.profile must disable this directive and must add a whitelist directive to allow access to that directory (as the Firefox profile is a whitelisted profile):

noblacklist ${HOME}/.mozilla
whitelist ${HOME}/.mozilla

Profile writing

The basic process is:

Copy /usr/share/doc/firejail/profile.template to /etc/firejail/ or ~/.config/firejail/ and rename it to ProfileName.profile where ProfileName should match the name of the executable to be sandboxed
Change the line include PROFILE.local to include ProfileName.local
Gradually comment/uncomment the various options while checking at each stage that the application runs inside the new sandbox. Do not change the order of the sections in that template.
Detailed explanations of the possible options for a Firejail profile can be found in the firejail-profile(5) man page
Test the profile for security holes, see #Testing profiles

If you want to create a whitelisted profile (i.e. a profile which contains whitelist directives), you can build a whitelist of permitted locations by executing

$ firejail --build application

Keep in mind that a whitelisted profile is problematic for applications that need to access random locations (like text editors or file managers).

Note:

The idea is to be as restrictive as possible, while still maintaining usability. This may involve sacrificing potentially dangerous functionality and a change in cavalier work habits.
By default, seccomp filters work on a blacklist (which can be found in /usr/share/doc/firejail/syscalls.txt). It is possible to use seccomp.keep to build a custom whitelist of filters for an application. [2]. A convenient way to automate these steps is to execute /usr/lib/firejail/syscalls.sh. If the application is still broken because of missing syscalls, you should follow the instructions at the bottom of /usr/share/doc/firejail/syscalls.txt.

Persistent local customisation

The standard profile layout includes the capability to make persistent local customisations through the inclusion of .local files[3]. Basically, each officially supported profile contains the lines include ProgramName.local and include globals.local. These *.local files might be located in /etc/firejail/ or in ~/.config/firejail/. Since the order of precedence is determined by which is read first, this makes for a very powerful way of making local customisations. For example, with reference this firejail question, to globally enable Apparmor and disable Internet connectivity, one could simply create/edit /etc/firejail/globals.local to include the lines

# enable Apparmor and disable Internet globally
net none
apparmor

Then, to allow, for example, "curl" to connect to the internet, yet still maintain its apparmor confinement, one would create/edit /etc/firejail/curl.local to include the lines.

# enable internet for curl
ignore net

Since curl.local is read before globals.local, ignore net overrides net none, and, as a bonus, the above changes would be persistent across future updates.

Testing profiles

In order to test and audit a Firejail profile, you may find the following to be useful:

firejail --debug $Program > $PathToOutputFile Gives a detailed breakdown of the sandbox
firejail --debug-blacklists $Program and firejail --debug-whitelists $Program show the blacklisted and whitelisted directories and files for the current profile.
firejail --debug-caps gives a list of caps supported by the current Firejail software build. This is useful when building a caps whitelist.
firejail --help for a full list of --debug options
firemon PID monitors the running process. See firemon --help for details
Executing sudo jailcheck tests running sandboxes. See the jailcheck(1) man page for details.
checksec may also be useful in testing which standard security features are being used

Firejail with Xorg

The factual accuracy of this article or section is disputed.

Reason: Why does sandboxing X11 require DNS? (Discuss in Talk:Firejail)

On Xorg any program can listen to all keyboard input and record all screens. The purpose of sandboxing X11 is to restrict this behavior, which is especially problematic for complex programs working with potentially malicious input like browsers.

Xephyr and Xpra allow you to sandbox Xorg. Although Xpra provides full clipboard support, it is recommended to use Xephyr due to the very notable and permanent lag with nested X11 sessions.

For a complete setup with (not ideal) clipboard support (clipboard is still always shared), see Sakaki's Gentoo guide, especially the section about the clipboard and automatic rescaling.

Alternatively, if clipboard support is not needed but windows need to be managed, install a standalone window manager such as Openbox.

xephyr-screen WidthxHeight can be set in /etc/firejail/firejail.config where Width and Height are in pixels and based on your screen resolution.

To open the sandbox:

$ firejail --x11=xephyr --net=device openbox

device is your active network interface, which is needed to ensure that DNS works. Then right click and select your applications to run.

Note: If you use Unbound, dnsmasq, Pdnsd or any other local resolver on 127.0.0.1, you would leave --net=device out of the command as DNS should work automatically.

See the Firejail Wordpress site for a simpler guide.

According to the guide:

The sandbox replaces the regular X11 server with Xpra or Xephyr server. This prevents X11 keyboard loggers and screenshot utilities from accessing the main X11 server.

Note that the statement:

The only way to disable the abstract socket @/tmp/.X11-unix/X0 is by using a network namespace. If for any reasons you cannot use a network namespace, the abstract socket will still be visible inside the sandbox. Hackers can attach keylogger and screenshot programs to this socket.

is incorrect, xserverrc can be edited to -nolisten local, which disables the abstract sockets of X11 and helps isolate it.

Sandboxing a browser

Openbox can be configured to start a certain browser at startup. program.profile is the respective profile contained in /etc/firejail, and --startup "command" is the command line used to start the program. For example, to start Chromium in the sandbox:

$ firejail --x11=xephyr --profile=/etc/firejail/chromium.profile openbox --startup "chromium"

You can control the size of the screen with the parameter:

--xephyr-screen=400x250

Tips and tricks

Hardening Firejail

The security risk of Firejail being a SUID executable can be mitigated by adding the line

force-nonewprivs yes

to /etc/firejail/firejail.config. However, this can break specific applications. On Arch Linux, VirtualBox doesn't start anymore. With the linux-hardened kernel Wireshark and Chromium-based browsers are also affected.

Further hardening measures include creating a special firejail group with adding the user to that group and changing the file mode for the firejail executable. For details see here.

Tip: you can add a pacman hook to automatically change firejail owner and mode:

/etc/pacman.d/hooks/firejail-permissions.hook

[Trigger]
Operation = Install
Operation = Upgrade
Type = Package
Target = firejail
[Action]
Depends = coreutils
Depends = bash
When = PostTransaction
Exec = /usr/bin/sh -c "chown root:firejail /usr/bin/firejail && chmod 4750 /usr/bin/firejail"
Description = Setting /usr/bin/firejail owner to "root:firejail" and mode "4750"

Make sure to create the firejail group and add your user to it.

Paths containing spaces

If you need to reference, whitelist, or blacklist a directory within a custom profile, such as with palemoon^AUR, you must do so using the absolute path, without encapsulation or escapes:

/home/user/.moonchild productions

Private mode

Firejail also includes a one time private mode, in which no mounts are made in the chroots to your home directory. In doing this, you can execute applications without performing any changes to disk. For example, to execute okular in private mode, do the following:

$ firejail --seccomp --private okular

Experimental improved tools

Some of the Firejail developers recognized issues with the tools it ships with and made their own, improved versions of them.

Warning: As stated in some of the projects descriptions, these tools might be experimental.

firecfg.py, an improved version of firecfg.
fjp, a tool to interact with Firejail profiles.
fireurl, Fixing the firejail URL open issue.
firejail-handler-http, which helps with opening HTTP(S) links properly when sandboxing applications.
firejail-handler-extra, like above but handles other protocols.

Troubleshooting

Firejail can be hard to debug. The symptoms of a misconfigured or otherwise unfitting setup range from random segmentation faults and hangs in the applications to simple error messages.

Some applications are harder to sandbox than others. For example web browsers and Electron applications tend to need more troubleshooting than others since there is much that can go wrong. It is crucial to check the FAQ and open issues first, since debugging can take quite some time.

Tip: Also see the upstream wiki, this blog post and especially its page on debugging Firejail.

Remove Firejail symbolic links

To remove Firejail created symbolic links (e.g. reset to default):

# firecfg --clean

If you do not want to use Firejail for a specific application (e.g., because you prefer to rather confine it with AppArmor), you have to manually remove the related symbolic link:

# rm /usr/local/bin/application

As a subsequent execution of firecfg would re-add the removed symlinks, the respective applications should be commented in /etc/firejail/firecfg.config.

Verify if any leftovers of Desktop entries are still overruled by Firejail.

PulseAudio

Note: Using PulseAudio version 9.0 or later should fix this issue.

If Firejail causes PulseAudio issues with sandboxed applications [4], the following command may be used:

$ firecfg --fix-sound

This commands creates a custom ~/.config/pulse/client.conf file for the current user with enable-shm = no and possible other workarounds.

hidepid

If the system uses the hidepid kernel parameter, Firemon can only be run as root. This, among other things, will cause problems with the Firetools GUI incorrectly reporting "Capabilities", "Protocols" and the status of "Seccomp"[5].

Proprietary Nvidia drivers

Some users report problems when using Firejail and proprietary graphic drivers from NVIDIA (e.g. [6], [7] or [8]). This can often be solved by disabling the noroot Firejail option in the application's profile file.

--net options and Linux kernel >=4.20.0

There is a bug on firejail 0.5.96 with linux >= 4.20.0, see [9] and [10]

Example error message:

$ firejail --noprofile --net=eth0 ls
Parent pid 8521, child pid 8522
Error send: arp.c:182 arp_check: Invalid argument
Error: proc 8521 cannot sync with peer: unexpected EOF
Peer 8522 unexpectedly exited with status 1

Warning: Cannot confine the application using AppArmor

This article or section is a candidate for merging with #Enable AppArmor support.

Notes: The requirement to enable AppArmor should be made explicit in the previous text. (Discuss in Talk:Firejail)

For some applications (e.g. Firefox) starting with Firejail may result in warnings like:

Warning: Cannot confine the application using AppArmor.
Maybe firejail-default AppArmor profile is not loaded into the kernel.
As root, run "aa-enforce firejail-default" to load it.

When running the suggested command you might see:

ERROR: Cache read/write disabled: interface file missing. (Kernel needs AppArmor 2.4 compatibility patch.)

This means that AppArmor is not enabled as a kernel parameter, so you have to set it according to AppArmor#Installation.

/usr/bin/patch: **** Can't open patch file

This means the PKGBUILD uses patch with the -i argument so a whitelist for $SRCDEST in /etc/makepkg.conf is needed.

Create the override patch.local with the value of your $SRCDEST:

whitelist /path/to/makepkg/sources

Changing the PKGBUILD to use stdin also works:

patch -p1 < ../file.patch

Daemonizing/backgrounded processes hang

There is a known issue that prevents processes from daemonizing. There is currently no solution to this except not using Firejail to sandbox the affected application. Because it is a bug within Firejail, no configuration can solve this issue. Fortunately the applications mentioned in the issue usually do not have a large attack surface, so the risks of running them without a sandbox are comparatively low.