Security

From ArchWiki
(Redirected from Linux-grsec)
Jump to navigation Jump to search

This article contains recommendations and best practices for hardening an Arch Linux system.

Contents

Concepts

  • It is possible to tighten the security so much as to make your system unusable. The trick is to secure it without overdoing it.
  • There are many other things that can be done to heighten the security, but the biggest threat is, and will always be, the user. When you think security, you have to think layers. When one layer is breached, another should stop the attack. But you can never make the system 100% secure unless you unplug the machine from all networks, lock it in a safe and never use it.
  • Be a little paranoid. It helps. And be suspicious. If anything sounds too good to be true, it probably is!
  • The principle of least privilege: each part of a system should only be able to access what is required to use it, and nothing more.

Passwords

Passwords are key to a secure Linux system. They secure your user accounts, encrypted filesystems, and SSH/GPG keys. They are the main way a computer chooses to trust the person using it, so a big part of security is just about picking secure passwords and protecting them.

Choosing secure passwords

Passwords must be complex enough to not be easily guessed from e.g. personal information, or cracked using methods like social engineering or brute-force attacks. The tenets of strong passwords are based on length and randomness. In cryptography the quality of a password is referred to as its entropic security.

Insecure passwords include those containing:

  • Personally identifiable information (e.g., your dog's name, date of birth, area code, favorite video game)
  • Simple character substitutions on words (e.g., k1araj0hns0n), as modern dictionary attacks can easily work with these
  • Root "words" or common strings followed or preceded by added numbers, symbols, or characters (e.g., DG091101%)
  • Common phrases or strings of dictionary words (e.g. photocopyhauntbranchexpose) including with character substitution (e.g. Ph0toc0pyh4uN7br@nch3xp*se)
  • Any of the most common passwords

The best choice for a password is something long (the longer, the better) and generated from a random source. It is important to use a long password. Weak hash algorithms allow an 8-character password hash to be compromised in just a few hours.

Tools like pwgen or apgAUR can generate random passwords. However, these passwords can be difficult to memorize. One memorization technique (for ones typed often) is to generate a long password and memorize a minimally secure number of characters, temporarily writing down the full generated string. Over time, increase the number of characters typed - until the password is ingrained in muscle memory and need not be remembered. This technique is more difficult, but can provide confidence that a password will not turn up in wordlists or "intelligent" brute force attacks that combine words and substitute characters.

One technique for memorizing a password is to use a mnemonic phase, where each word in the phase reminds you of the next character in the password. Take for instance “the girl is walking down the rainy street” could be translated to t6!WdtR5 or, less simply, t&6!RrlW@dtR,57. This approach could make it easier to remember a password, but note that the various letters have very different probabilities of being found at the start of words (Wikipedia:Letter frequency).

Another effective technique can be to write randomly generated passwords down and store them in a safe place, such as in a wallet, purse or document safe. Most people do a generally good job of protecting their physical valuables from attack, and it is easier for most people to understand physical security best practices compared to digital security practices. Bruce Schneier has endorsed this technique.

It is also very effective to combine the mnemonic and random technique by saving long randomly generated passwords with a password manager, which will be in turn accessed with a memorable "master password" that must be used only for that purpose. The master password must be memorized and never saved. This requires the password manager to be installed on a system to easily access the password (which could be seen as an inconvenience or a security feature, depending on the situation). Some password managers also have smartphone apps which can be used to display passwords for manual entry on systems without that password manager installed. Note that a password manager introduces a single point of failure if you ever forget the master password.

Formerly, it was effective to use a memorable long series of unrelated words as a password. The theory is that if a sufficiently long phrase is used, the gained entropy from the password's length can counter the lost entropy from the use of dictionary words. This XKCD comic demonstrates the entropy tradeoff of this method. However, password crackers have caught on to this trick and will generate wordlists containing billions of permutations and variants of dictionary words, reducing the effective entropy of the password.

See Bruce Schneier's article Choosing Secure Passwords, The passphrase FAQ or Wikipedia:Password strength for some additional background.

Maintaining passwords

Once you pick a strong password, be sure to keep it safe. Watch out for keyloggers (software and hardware), screen loggers, social engineering, shoulder surfing, and avoid reusing passwords so insecure servers cannot leak more information than necessary. Password managers can help manage large numbers of complex passwords: if you are copy-pasting the stored passwords from the manager to the applications that need them, make sure to clear the copy buffer every time, and ensure they are not saved in any kind of log (e.g. do not paste them in plain terminal commands, which would store them in files like .bash_history).

As a rule, do not pick insecure passwords just because secure ones are harder to remember. Passwords are a balancing act. It is better to have an encrypted database of secure passwords, guarded behind a key and one strong master password, than it is to have many similar weak passwords. Writing passwords down is perhaps equally effective [1], avoiding potential vulnerabilities in software solutions while requiring physical security.

Another aspect of the strength of the passphrase is that it must not be easily recoverable from other places.

If you use the same passphrase for disk encryption as you use for your login password (useful e.g. to auto-mount the encrypted partition or folder on login), make sure that /etc/shadow either also ends up on an encrypted partition, or uses a strong hash algorithm (i.e. sha512/bcrypt, not md5) for the stored password hash (see SHA password hashes for more information).

If you are backing up your password database, make sure that each copy is not stored behind any other passphrase which in turn is stored in it, e.g. an encrypted drive or an authenticated remote storage service, or you will not be able to access it in case of need; a useful trick is to protect the drives or accounts where the database is backed up using a simple cryptographic hash of the master password. Maintain a list of all the backup locations: if one day you fear that the master passphrase has been compromised you will have to change it immediately on all the database backups and the locations protected with keys derived from the master password.

Version-controlling the database in a secure way can be very complicated: if you choose to do it, you must have a way to update the master password of all the database versions. It may not always be immediately clear when the master password is leaked: to reduce the risk of somebody else discovering your password before you realize that it leaked, you may choose to change it on a periodical basis. If you fear that you have lost control over a copy of the database, you will need to change all the passwords contained in it within the time that it may take to brute-force the master password, according to its entropy.

Password hashes

Tango-view-fullscreen.pngThis article or section needs expansion.Tango-view-fullscreen.png

Reason: Mention key derivation functions, in particular PBKDF2, bcrypt and scrypt, how to use them, advantages and disadvantages, especially regarding custom-hardware-based brute-force attacks. (Discuss in Talk:Security#Removal of incorrect warning)

By default, Arch stores the hashed user passwords in the root-only-readable /etc/shadow file, separated from the other user parameters stored in the world-readable /etc/passwd file, see Users and groups#User database. See also #Restricting root.

Passwords are set with the passwd command, which stretches them with the crypt function and then saves them in /etc/shadow. See also SHA password hashes. The passwords are also salted in order to defend them against rainbow table attacks.

See also How are passwords stored in Linux (Understanding hashing with shadow utils).

Enforcing strong passwords with pam_pwquality

pam_pwquality provides protection against Dictionary attacks and helps configure a password policy that can be enforced throughout the system. It is based on pam_cracklib, so it is backwards compatible with its options.

Install the libpwquality package.

Warning: The root account is not affected by this policy by default.
Note: You can use the root account to set a password for a user that bypasses the desired/configured policy. This is useful when setting temporary passwords.

If for example you want to enforce this policy:

  • prompt 2 times for password in case of an error (retry option)
  • 10 characters minimum length (minlen option)
  • at least 6 characters should be different from old password when entering a new one (difok option)
  • at least 1 digit (dcredit option)
  • at least 1 uppercase (ucredit option)
  • at least 1 lowercase (lcredit option)
  • at least 1 other character (ocredit option)
  • cannot contain the words "myservice" and "mydomain"
  • enforce the policy for root

Edit the /etc/pam.d/passwd file to read as:

#%PAM-1.0
password required pam_pwquality.so retry=2 minlen=10 difok=6 dcredit=-1 ucredit=-1 ocredit=-1 lcredit=-1 [badwords=myservice mydomain] enforce_for_root
password required pam_unix.so use_authtok sha512 shadow

The password required pam_unix.so use_authtok instructs the pam_unix module to not prompt for a password but rather to use the one provided by pam_pwquality.

You can refer to the pam_pwquality(8) and pam_unix(8) man pages for more information.

CPU

Microcode

See microcode for information on how to install important security updates for your CPU's microcode.

Hardware vulnerabilities

Some CPUs contain hardware vulnerabilities. See the kernel documentation on hardware vulnerabilities for a list of these vulnerabilities, as well as mitigation selection guides to help customize the kernel to mitigate these vulnerabilities for specific usage scenarios.

To check if you are affected by a known vulnerability, run the following:

$ grep -r . /sys/devices/system/cpu/vulnerabilities/

In most cases, updating the kernel and microcode will mitigate vulnerabilities.

Simultaneous multithreading (hyper-threading)

Simultaneous multithreading (SMT), also called hyper-threading on Intel CPUs, is a hardware feature that may be a source of L1 Terminal Fault and Microarchitectural Data Sampling vulnerabilities. The Linux kernel and microcode updates contain mitigations for known vulnerabilities, but disabling SMT may still be required on certain CPUs if untrusted virtualization guests are present.

SMT can often be disabled in your system's firmware. Consult your motherboard or system documentation for more information. You can also disable SMT in the kernel by adding the following kernel parameters:

l1tf=full,force mds=full,nosmt mitigations=auto,nosmt nosmt=force

Memory

Hardened malloc

hardened_malloc (hardened_mallocAUR, hardened-malloc-gitAUR) is a hardened replacement for glibc's malloc(). The project was originally developed for integration into Android's Bionic and musl by Daniel Micay, of GrapheneOS, but he has also built in support for standard Linux distributions on the x86_64 architecture.

While hardened_malloc is not yet integrated into glibc (assistance and pull requests welcome) it can be used easily with LD_PRELOAD. In testing so far, it only causes issues with a handful of applications if enabled globally in /etc/ld.so.preload. For example, man fails to work properly unless its seccomp environment flag is disabled due to not having getrandom in the standard whitelist, although this can be easily fixed by rebuilding it with the system call added. Since hardened_malloc has a performance cost, you may want to decide which implementation to use on a case-by-case basis based on attack surface and performance needs.

To try it out in a standalone manner, use the hardened-malloc-preload wrapper script, or manually start an application with the proper preload value:

LD_PRELOAD="/usr/lib/libhardened_malloc.so" /usr/bin/firefox

Proper usage with Firejail can be found on its wiki page, and some configurable build options for hardened_malloc can be found on the github repo.

Storage

Data-at-rest encryption

Data-at-rest encryption, preferably full-disk encryption with a strong passphrase, is the only way to guard data against physical recovery. This provides complete security when the computer is turned off or the disks in question are unmounted.

Once the computer is powered on and the drive is mounted, however, its data becomes just as vulnerable as an unencrypted drive. It is therefore best practice to unmount data partitions as soon as they are no longer needed.

Certain programs, like dm-crypt, allow the user to encrypt a loop file as a virtual volume. This is a reasonable alternative to full-disk encryption when only certain parts of the system need be secure.

File systems

The kernel now prevents security issues related to hardlinks and symlinks if the fs.protected_hardlinks and fs.protected_symlinks sysctl switches are enabled, so there is no longer a major security benefit from separating out world-writable directories.

File systems containing world-writable directories can still be kept separate as a coarse way of limiting the damage from disk space exhaustion. However, filling /var or /tmp is enough to take down services. More flexible mechanisms for dealing with this concern exist (like quotas), and some file systems include related features themselves (Btrfs has quotas on subvolumes).

Mount options

Following the principle of least privilege, file systems should be mounted with the most restrictive mount options possible (without losing functionality).

Relevant mount options are:

  • nodev: Do not interpret character or block special devices on the file system.
  • nosuid: Do not allow set-user-identifier or set-group-identifier bits to take effect.
  • noexec: Do not allow direct execution of any binaries on the mounted file system.
    • Setting noexec on /home disallows executable scripts and breaks Wine*, Steam, PyCharm, etc.
    • Some packages (building nvidia-dkms for example) may require exec on /var.
Note: Wine does not need the exec flag for opening Windows executables. It is only needed when Wine itself is installed in /home.

File systems used for data should always be mounted with nodev, nosuid and noexec.

Potential file system mounts to consider:

  • /var
  • /home
  • /dev/shm
  • /tmp
  • /boot

File access permissions

The default file permissions allow read access to almost everything and changing the permissions can hide valuable information from an attacker who gains access to a non-root account such as the http or nobody users.

For example:

# chmod 700 /boot /etc/{iptables,arptables}

The default Umask 0022 can be changed to improve security for newly created files. The NSA RHEL5 Security Guide suggests a umask of 0077 for maximum security, which makes new files not readable by users other than the owner. To change this, see Umask#Set the mask value.

User setup

After installation make a normal user for daily use. Do not use the root user for daily use.

Enforce a delay after a failed login attempt

Add the following line to /etc/pam.d/system-login to add a delay of at least 4 seconds between failed login attempts:

/etc/pam.d/system-login
auth optional pam_faildelay.so delay=4000000

4000000 is the time in microseconds to delay.

Lock out user after three failed login attempts

As of pambase 20200721.1-2, pam_faillock.so is enabled by default to lock out users for 10 minutes after 3 failed login attempts in a 15 minute period (see FS#67644). The lockout only applies to password authentication (e.g. login and sudo), public key authentication over SSH is still accepted. To prevent complete denial-of-service, this lockout is disabled on root.

To unlock a user, do:

$ faillock --reset --user username

By default, the lock mechanism is a file per-user located at /run/faillock/. Deleting or emptying the file unlocks that user - the directory is owned by root, but the file is owned by the user, so the faillock command only empties the file, therefore does not require root.

The module pam_faillock.so can be configured with the file /etc/security/faillock.conf. The lockout parameters:

  • unlock_time — the lockout time (in seconds, default 10 minutes).
  • fail_interval — the time in which failed logins can cause a lockout (in seconds, default 15 minutes).
  • deny — the number of failed logins before lockout (default 3).
Note: deny = 0 will disable the lockout.

No restart is required for changes to take effect. See faillock.conf(5) for further configuration options, such as enabling lockout for the root account, disabling for centralized login (e.g. LDAP), etc.

Limit amount of processes

On systems with many, or untrusted users, it is important to limit the number of processes each can run at once, therefore preventing fork bombs and other denial of service attacks. /etc/security/limits.conf determines how many processes each user, or group can have open, and is empty (except for useful comments) by default. Adding the following lines to this file will limit all users to 100 active processes, unless they use the prlimit command to explicitly raise their maximum to 200 for that session. These values can be changed according to the appropriate number of processes a user should have running, or the hardware of the box you are administrating.

* soft nproc 100
* hard nproc 200

The current number of threads for each user can be found with ps --no-headers -Leo user | sort | uniq --count. This may help with determining appropriate values for the limits.

Run Xorg rootless

Xorg is commonly considered insecure because of its architecture and dated design. Thus it is recommended to avoid running it as root.

See Xorg#Rootless Xorg for more details how to run it without root privileges. Alternatively, use Wayland instead of Xorg.

Restricting root

The root user is, by definition, the most powerful user on a system. Because of this, there are a number of ways to keep the power of the root user while limiting its ability to cause harm, or at least to make root user actions more traceable.

Use sudo instead of su

Merge-arrows-2.pngThis article or section is a candidate for merging with sudo.Merge-arrows-2.png

Notes: There is a dedicated article. (Discuss in Talk:Security#)

Using sudo for privileged access is preferable to su for a number of reasons.

  • It keeps a log of which normal privilege user has run each privileged command.
  • The root user password need not be given out to each user who requires root access.
  • sudo prevents users from accidentally running commands as root that do not need root access, because a full root terminal is not created. This aligns with the principle of least privilege.
  • Individual programs may be enabled per user, instead of offering complete root access just to run one command. For example, to give the user alice access to a particular program:
# visudo
/etc/sudoers
alice ALL = NOPASSWD: /path/to/program

Or, individual commands can be allowed for all users. To mount Samba shares from a server as a regular user:

%users ALL=/sbin/mount.cifs,/sbin/umount.cifs

This allows all users who are members of the group users to run the commands /sbin/mount.cifs and /sbin/umount.cifs from any machine (ALL).

Tip:

To use restricted version of nano instead of vi with visudo,

/etc/sudoers
Defaults editor=/usr/bin/rnano

Exporting EDITOR=nano visudo is regarded as a severe security risk since everything can be used as an EDITOR.

Editing files using sudo

See Sudo#Editing files. Alternatively, you can use an editor like rvim or rnano which has restricted capabilities in order to be safe to run as root.

Restricting root login

Once sudo is properly configured, full root access can be heavily restricted or denied without losing much usability. To disable root, but still allowing to use sudo, you can use passwd --lock root.

Allow only certain users

The PAM pam_wheel.so lets you allow only users in the group wheel to login using su. See su#su and wheel.

Denying SSH login

Even if you do not wish to deny root login for local users, it is always good practice to deny root login via SSH. The purpose of this is to add an additional layer of security before a user can completely compromise your system remotely.

Specify acceptable login combinations with access.conf

When someone attempts to log in with PAM, /etc/security/access.conf is checked for the first combination that matches their login properties. Their attempt then fails or succeeds based on the rule for that combination.

+:root:LOCAL
-:root:ALL

Rules can be set for specific groups and users. In this example, the user archie is allowed to login locally, as are all users in the wheel and adm groups. All other logins are rejected:

+:archie:LOCAL
+:(wheel):LOCAL
+:(adm):LOCAL
-:ALL:ALL

Read more at access.conf(5)

Mandatory access control

Mandatory access control (MAC) is a type of security policy that differs significantly from the discretionary access control (DAC) used by default in Arch and most Linux distributions. MAC essentially means that every action a program could perform that affects the system in any way is checked against a security ruleset. This ruleset, in contrast to DAC methods, cannot be modified by users. Using virtually any mandatory access control system will significantly improve the security of your computer, although there are differences in how it can be implemented.

Pathname MAC

Pathname-based access control is a simple form of access control that offers permissions based on the path of a given file. The downside to this style of access control is that permissions are not carried with files if they are moved about the system. On the positive side, pathname-based MAC can be implemented on a much wider range of filesystems, unlike labels-based alternatives.

  • AppArmor is a Canonical-maintained MAC implementation seen as an "easier" alternative to SELinux.
  • Tomoyo is another simple, easy-to-use system offering mandatory access control. It is designed to be both simple in usage and in implementation, requiring very few dependencies.

Labels MAC

Labels-based access control means the extended attributes of a file are used to govern its security permissions. While this system is arguably more flexible in its security offerings than pathname-based MAC, it only works on filesystems that support these extended attributes.

  • SELinux, based on a NSA project to improve Linux security, implements MAC completely separate from system users and roles. It offers an extremely robust multi-level MAC policy implementation that can easily maintain control of a system that grows and changes past its original configuration.

Access Control Lists

Access Control Lists (ACLs) are an alternative to attaching rules directly to the filesystem in some way. ACLs implement access control by checking program actions against a list of permitted behavior.

Kernel hardening

Kernel self-protection / exploit mitigation

The linux-hardened package uses a basic kernel hardening patch set and more security-focused compile-time configuration options than the linux package. A custom build can be made to choose a different compromise between security and performance than the security-leaning defaults.

However, it should be noted that several packages will not work when using this kernel. For example:

If you use an out-of-tree driver such as NVIDIA, you may need to switch to its DKMS package.

Userspace ASLR comparison

The linux-hardened package provides an improved implementation of Address Space Layout Randomization for userspace processes. The paxtest command can be used to obtain an estimate of the provided entropy:

64-bit processes
linux-hardened 5.4.21.a-1-hardened
Anonymous mapping randomization test     : 32 quality bits (guessed)
Heap randomization test (ET_EXEC)        : 40 quality bits (guessed)
Heap randomization test (PIE)            : 40 quality bits (guessed)
Main executable randomization (ET_EXEC)  : 32 quality bits (guessed)
Main executable randomization (PIE)      : 32 quality bits (guessed)
Shared library randomization test        : 32 quality bits (guessed)
VDSO randomization test                  : 32 quality bits (guessed)
Stack randomization test (SEGMEXEC)      : 40 quality bits (guessed)
Stack randomization test (PAGEEXEC)      : 40 quality bits (guessed)
Arg/env randomization test (SEGMEXEC)    : 44 quality bits (guessed)
Arg/env randomization test (PAGEEXEC)    : 44 quality bits (guessed)
Offset to library randomisation (ET_EXEC): 34 quality bits (guessed)
Offset to library randomisation (ET_DYN) : 34 quality bits (guessed)
Randomization under memory exhaustion @~0: 32 bits (guessed)
Randomization under memory exhaustion @0 : 32 bits (guessed)
linux 5.5.5-arch1-1
Anonymous mapping randomization test     : 28 quality bits (guessed)
Heap randomization test (ET_EXEC)        : 28 quality bits (guessed)
Heap randomization test (PIE)            : 28 quality bits (guessed)
Main executable randomization (ET_EXEC)  : 28 quality bits (guessed)
Main executable randomization (PIE)      : 28 quality bits (guessed)
Shared library randomization test        : 28 quality bits (guessed)
VDSO randomization test                  : 20 quality bits (guessed)
Stack randomization test (SEGMEXEC)      : 30 quality bits (guessed)
Stack randomization test (PAGEEXEC)      : 30 quality bits (guessed)
Arg/env randomization test (SEGMEXEC)    : 22 quality bits (guessed)
Arg/env randomization test (PAGEEXEC)    : 22 quality bits (guessed)
Offset to library randomisation (ET_EXEC): 28 quality bits (guessed)
Offset to library randomisation (ET_DYN) : 28 quality bits (guessed)
Randomization under memory exhaustion @~0: 29 bits (guessed)
Randomization under memory exhaustion @0 : 29 bits (guessed)
linux-lts 4.19.101-1-lts
Anonymous mapping randomization test     : 28 quality bits (guessed)
Heap randomization test (ET_EXEC)        : 28 quality bits (guessed)
Heap randomization test (PIE)            : 28 quality bits (guessed)
Main executable randomization (ET_EXEC)  : 28 quality bits (guessed)
Main executable randomization (PIE)      : 28 quality bits (guessed)
Shared library randomization test        : 28 quality bits (guessed)
VDSO randomization test                  : 19 quality bits (guessed)
Stack randomization test (SEGMEXEC)      : 30 quality bits (guessed)
Stack randomization test (PAGEEXEC)      : 30 quality bits (guessed)
Arg/env randomization test (SEGMEXEC)    : 22 quality bits (guessed)
Arg/env randomization test (PAGEEXEC)    : 22 quality bits (guessed)
Offset to library randomisation (ET_EXEC): 28 quality bits (guessed)
Offset to library randomisation (ET_DYN) : 28 quality bits (guessed)
Randomization under memory exhaustion @~0: 28 bits (guessed)
Randomization under memory exhaustion @0 : 28 bits (guessed)
32-bit processes (on an x86_64 kernel)
linux-hardened
Anonymous mapping randomization test     : 16 quality bits (guessed)
Heap randomization test (ET_EXEC)        : 22 quality bits (guessed)
Heap randomization test (PIE)            : 27 quality bits (guessed)
Main executable randomization (ET_EXEC)  : No randomization
Main executable randomization (PIE)      : 18 quality bits (guessed)
Shared library randomization test        : 16 quality bits (guessed)
VDSO randomization test                  : 16 quality bits (guessed)
Stack randomization test (SEGMEXEC)      : 24 quality bits (guessed)
Stack randomization test (PAGEEXEC)      : 24 quality bits (guessed)
Arg/env randomization test (SEGMEXEC)    : 28 quality bits (guessed)
Arg/env randomization test (PAGEEXEC)    : 28 quality bits (guessed)
Offset to library randomisation (ET_EXEC): 18 quality bits (guessed)
Offset to library randomisation (ET_DYN) : 16 quality bits (guessed)
Randomization under memory exhaustion @~0: 18 bits (guessed)
Randomization under memory exhaustion @0 : 18 bits (guessed)
linux
Anonymous mapping randomization test     : 8 quality bits (guessed)
Heap randomization test (ET_EXEC)        : 13 quality bits (guessed)
Heap randomization test (PIE)            : 13 quality bits (guessed)
Main executable randomization (ET_EXEC)  : No randomization
Main executable randomization (PIE)      : 8 quality bits (guessed)
Shared library randomization test        : 8 quality bits (guessed)
VDSO randomization test                  : 8 quality bits (guessed)
Stack randomization test (SEGMEXEC)      : 19 quality bits (guessed)
Stack randomization test (PAGEEXEC)      : 19 quality bits (guessed)
Arg/env randomization test (SEGMEXEC)    : 11 quality bits (guessed)
Arg/env randomization test (PAGEEXEC)    : 11 quality bits (guessed)
Offset to library randomisation (ET_EXEC): 8 quality bits (guessed)
Offset to library randomisation (ET_DYN) : 13 quality bits (guessed)
Randomization under memory exhaustion @~0: No randomization
Randomization under memory exhaustion @0 : No randomization

Restricting access to kernel logs

The kernel logs contain useful information for an attacker trying to exploit kernel vulnerabilities, such as sensitive memory addresses. The kernel.dmesg_restrict flag was to forbid access to the logs without the CAP_SYS_ADMIN capability (which only processes running as root have by default).

/etc/sysctl.d/51-dmesg-restrict.conf
kernel.dmesg_restrict = 1
Tip: This is enabled by default in linux-hardened.

Restricting access to kernel pointers in the proc filesystem

Setting kernel.kptr_restrict to 1 will hide kernel symbol addresses in /proc/kallsyms from regular users without CAP_SYSLOG, making it more difficult for kernel exploits to resolve addresses/symbols dynamically. This will not help that much on a pre-compiled Arch Linux kernel, since a determined attacker could just download the kernel package and get the symbols manually from there, but if you are compiling your own kernel, this can help mitigating local root exploits. This will break some perf commands when used by non-root users (but many perf features require root access anyway). See FS#34323 for more information.

Setting kernel.kptr_restrict to 2 will hide kernel symbol addresses in /proc/kallsyms regardless of privileges.

/etc/sysctl.d/51-kptr-restrict.conf
kernel.kptr_restrict = 1
Note: linux-hardened sets kptr_restrict=2 by default rather than 0.

BPF hardening

BPF is a system used to load and execute bytecode within the kernel dynamically during runtime. It is used in a number of Linux kernel subsystems such as networking (e.g. XDP, tc), tracing (e.g. kprobes, uprobes, tracepoints) and security (e.g. seccomp). It is also useful for advanced network security, performance profiling and dynamic tracing.

BPF was originally an acronym of Berkeley Packet Filter since the original classic BPF was used for packet capture tools for BSD. This eventually evolved into Extended BPF (eBPF), which was shortly afterwards renamed to just BPF (not an acronym). BPF should not be confused with packet filtering tools like iptables or netfilter, although BPF can be used to implement packet filtering tools.

BPF code may be either interpreted or compiled using a Just-In-Time (JIT) compiler. The Arch kernel is built with CONFIG_BPF_JIT_ALWAYS_ON which disables the BPF interpreter and forces all BPF to use JIT compilation. This makes it harder for an attacker to use BPF to escalate attacks that exploit SPECTRE-style vulnerabilities. See the kernel patch which introduced CONFIG_BPF_JIT_ALWAYS_ON for more details.

The kernel includes a hardening feature for JIT-compiled BPF which can mitigate some types of JIT spraying attacks at the cost of performance and the ability to trace and debug many BPF programs. It may be enabled by setting net.core.bpf_jit_harden to 1 (to enable hardening of unprivileged code) or 2 (to enable hardening of all code).

See the net.core.bpf_* settings in the kernel documentation for more details.

Tip: linux-hardened sets net.core.bpf_jit_harden=2 by default rather than 0.

ptrace scope

The ptrace(2) syscall provides a means by which one process (the "tracer") may observe and control the execution of another process (the "tracee"), and examine and change the tracee's memory and registers. ptrace is commonly used by debugging tools including gdb, strace, perf, reptyr and other debuggers. However, it also provides a means by which a malicious process can read data from and take control of other processes.

Arch enables the Yama LSM by default, which provides a kernel.yama.ptrace_scope kernel parameter. This parameter is set to 1 (restricted) by default which prevents tracers from performing a ptrace call on tracees outside of a restricted scope unless the tracer is privileged or has the CAP_SYS_PTRACE capability. This is a significant improvement in security compared to the classic permissions. Without this module, there is no separation between processes running as the same user (in the absence of additional security layers such as pid_namespaces(7)).

Note: By default, you can still use tools which require ptrace by running them as privileged processes, e.g. using sudo.

If you do not need to use debugging tools, consider setting kernel.yama.ptrace_scope to 2 (admin-only) or 3 (no ptrace possible) to harden the system.

hidepid

Tango-view-fullscreen.pngThis article or section needs expansion.Tango-view-fullscreen.png

Reason: Linux 5.8 implemented private instances and new values for hidepid=. (Discuss in Talk:Security#)
Warning:
  • This may cause issues for certain applications like an application running in a sandbox and Xorg (see workaround).
  • This causes issues with D-Bus, PulseAudio and bluetooth when using systemd > 237.64-1.

The kernel has the ability to hide other users' processes, normally accessible via /proc, from unprivileged users by mounting the proc filesystem with the hidepid= and gid= options documented in https://www.kernel.org/doc/html/latest/filesystems/proc.html.

This greatly complicates an intruder's task of gathering information about running processes, whether some daemon runs with elevated privileges, whether other user runs some sensitive program, whether other users run any program at all, makes it impossible to learn whether any user runs a specific program (given the program does not reveal itself by its behaviour), and, as an additional bonus, poorly written programs passing sensitive information via program arguments are now protected against local eavesdroppers.

The proc group, provided by the filesystem package, acts as a whitelist of users authorized to learn other users' process information. If users or services need access to /proc/<pid> directories beyond their own, add them to the group.

For example, to hide process information from other users except those in the proc group:

/etc/fstab
proc	/proc	proc	nosuid,nodev,noexec,hidepid=2,gid=proc	0	0

For user sessions to work correctly, an exception needs to be added for systemd-logind:

/etc/systemd/system/systemd-logind.service.d/hidepid.conf
[Service]
SupplementaryGroups=proc

Restricting module loading

The default Arch kernel has CONFIG_MODULE_SIG_ALL enabled which signs all kernel modules build as part of the linux package. This allows the kernel to restrict modules to be only loaded when they are signed with a valid key, in practical terms this means that all out of tree modules compiled locally or provides by packages such as virtualbox-host-modules-arch cannot be loaded. Kernel module loading can be restricted by setting the kernel parameter module.sig_enforce=1. More information can be found at the kernel documentation.

Disable kexec

Kexec allows replacing the current running kernel.

/etc/sysctl.d/51-kexec-restrict.conf
kernel.kexec_load_disabled = 1
Tip: kexec is disabled by default in linux-hardened.

Kernel lockdown mode

Since Linux 5.4 the kernel has gained an optional lockdown feature, intended to strengthen the boundary between UID 0 (root) and the kernel. When enabled some applications may cease to work who rely on low-level access to either hardware or the kernel.

To use lockdown, its LSM must be initialized, check it by running cat /sys/kernel/security/lsm. If the output does not have lockdown in it, set the kernel parameter lsm=lockdown,yama and reboot.

Lockdown has two modes of operation:

  • integrity: kernel features that allow userland to modify the running kernel are disabled (kexec, bpf).
  • confidentiality: kernel features that allow userland to extract confidential information from the kernel are also disabled.

To enable kernel lockdown at runtime, run:

# echo mode > /sys/kernel/security/lockdown

To enable kernel lockdown on boot, use the kernel parameter lockdown=mode.

Note:
  • Kernel lockdown cannot be disabled at runtime.
  • Kernel lockdown disables hibernation.

Sandboxing applications

See also Wikipedia:Sandbox (computer security).

Note: The user namespace configuration item CONFIG_USER_NS is currently enabled in linux (4.14.5 or later), linux-lts (4.14.15 or later) and linux-hardened. Lack of it may prevent certain sandboxing features from being made available to applications.
Warning: Unprivileged user namespace usage (CONFIG_USER_NS_UNPRIVILEGED) is enabled by default in linux (5.1.8 or later), linux-lts (4.19.55-2 or later) and linux-zen (5.1.14.zen1-2 or later) unless the kernel.unprivileged_userns_clone sysctl is set to 0. Since this greatly increases the attack surface for local privilege escalation, it is advised to disable this manually, or use the linux-hardened kernel. For more information see FS#36969.

Firejail

Firejail is an easy to use and simple tool for sandboxing applications and servers alike. Firejail is suggested for browsers and internet facing applications, as well as any servers you may be running.

bubblewrap

bubblewrap is a sandbox application developed from Flatpak with an even smaller resource footprint than Firejail. While it lacks certain features such as file path whitelisting, bubblewrap does offer bind mounts as well as the creation of user/IPC/PID/network/cgroup namespaces and can support both simple and complex sandboxes.

chroots

Manual chroot jails can also be constructed.

Linux containers

Linux Containers are another good option when you need more separation than the other options (short of KVM and VirtualBox) provide. LXC is run on top of the existing kernel in a pseudo-chroot with their own virtual hardware.

Other virtualization options

Using full virtualization options such as VirtualBox, KVM, Xen or Qubes OS (based on Xen) can also improve isolation and security in the event you plan on running risky applications or browsing dangerous websites.

Network and firewalls

Firewalls

While the stock Arch kernel is capable of using Netfilter's iptables and nftables, they are not enabled by default. It is highly recommended to set up some form of firewall to protect the services running on the system. Many resources (including ArchWiki) do not state explicitly which services are worth protecting, so enabling a firewall is a good precaution.

Kernel parameters

Kernel parameters which affect networking can be set using Sysctl. For how to do this, see Sysctl#TCP/IP stack hardening.

SSH

To mitigate brute-force attacks it is recommended to enforce key-based authentication. For OpenSSH, see OpenSSH#Force public key authentication. Alternatively Fail2ban or Sshguard offer lesser forms of protection by monitoring logs and writing firewall rules but open up the potential for a denial of service, since an attacker can spoof packets as if they came from the administrator after identifying their address. Spoofing IP has lines of defense, such as by reverse path filtering and disabling ICMP redirects.

You may want to harden authentication even more by using two-factor authentication. Google Authenticator provides a two-step authentication procedure using one-time passcodes (OTP).

Denying root login is also a good practice, both for tracing intrusions and adding an additional layer of security before root access. For OpenSSH, see OpenSSH#Deny.

DNS

Tango-inaccurate.pngThe factual accuracy of this article or section is disputed.Tango-inaccurate.png

Reason: Your browser might notice DNS spoofing with HSTS. (Discuss in Talk:Security#)

DNS queries are, by default on most systems, sent and received unencrypted and without checking for authentication of receipt from qualified servers. This could then allow man-in-the-middle attacks, whereby an attacker intercepts your DNS queries and modifies the responses to deliver you an IP address leading to a phishing page to collect your valuable information. Neither you nor the browser/other software would be aware since the DNS protocol takes the legitimacy of query results for granted.

DNSSEC is a set of standards in place that requires DNS servers to provide clients with origin authentication of DNS data, authenticated denial of existence, and data integrity. It, however, is not yet widely used. With DNSSEC enabled, an attacker can not make modifications to your DNS queries and the returning results, but would still be able to read them.

DNSCrypt, as well as later alternative protocol developments DNS over TLS and DNS over HTTPS, use cryptography to secure communications with DNS servers. Usually only one protocol is employed on a system level. See Domain name resolution#DNS servers for supporting software.

If you have a domain name, set a Sender Policy Framework policy to combat email spoofing.

Proxies

Proxies are commonly used as an extra layer between applications and the network, sanitizing data from untrusted sources. The attack surface of a small proxy running with lower privileges is significantly smaller than a complex application running with the end user privileges.

For example the DNS resolver is implemented in glibc, that is linked with the application (that may be running as root), so a bug in the DNS resolver might lead to a remote code execution. This can be prevented by installing a DNS caching server, such as dnsmasq, which acts as a proxy. [2]

Managing SSL certificates

Merge-arrows-2.pngThis article or section is a candidate for merging with Transport Layer Security.Merge-arrows-2.png

Notes: There is a dedicated article. (Discuss in Talk:Security#)

See OpenSSL and Network Security Services (NSS) for managing custom server-side SSL certificates. Notably, the related Let’s Encrypt project is also supported.

The default internet SSL certificate trustchains are provided by the ca-certificates package and its dependencies. Note that Arch relies on trust-sources (e.g. ca-certificates-mozilla) providing the certificates to be trusted per default by the system.

There may be occasions when you want to deviate from the default. For example, you may read some news and want to distrust a certificate rather than wait until the trust-source providers do. The Arch infrastructure makes such easy:

  1. Obtain the respective certificate in .crt format (Example: view, download; in case of an existing trusted root certificate authority, you may also find it extracted in the system path),
  2. Copy it to /etc/ca-certificates/trust-source/blacklist/ and
  3. Run update-ca-trust as root.

To check the blacklisting works as intended, you may re-open your preferred browser and do so via its GUI, which should show it as untrusted now.

Physical security

Physical access to a computer is root access given enough time and resources. However, a high practical level of security can be obtained by putting up enough barriers.

An attacker can gain full control of your computer on the next boot by simply attaching a malicious IEEE 1394 (FireWire), Thunderbolt or PCI Express device as they are given full memory access.[3][dead link 2020-04-03 ⓘ] There is little you can do from preventing this, or modification of the hardware itself - such as flashing malicious firmware onto a drive. However, the vast majority of attackers will not be this knowledgeable and determined.

#Data-at-rest encryption will prevent access to your data if the computer is stolen, but malicious firmware can be installed to obtain this data upon your next log in by a resourceful attacker.

Locking down BIOS

Adding a password to the BIOS prevents someone from booting into removable media, which is basically the same as having root access to your computer. You should make sure your drive is first in the boot order and disable the other drives from being bootable if you can.

Boot loaders

It is highly important to protect your boot loader. An unprotected boot loader can bypass any login restrictions, e.g. by setting the init=/bin/sh kernel parameter to boot directly to a shell.

Syslinux

Syslinux supports password-protecting your bootloader. It allows you to set either a per-menu-item password or a global bootloader password.

GRUB

GRUB supports bootloader passwords as well. See GRUB/Tips and tricks#Password protection of GRUB menu for details. It also has support for encrypted /boot, which only leaves some parts of the bootloader code unencrypted. GRUB's configuration, kernel and initramfs are encrypted.

Boot partition on removable flash drive

One popular idea is to place the boot partition on a flash drive in order to render the system unbootable without it. Proponents of this idea often use full-disk encryption alongside, and some also use detached encryption headers placed on the boot partition.

This method can also be merged with encrypting /boot.

Automatic logout

If you are using Bash or Zsh, you can set TMOUT for an automatic logout from shells after a timeout.

For example, the following will automatically log out from virtual consoles (but not terminal emulators in X11):

/etc/profile.d/shell-timeout.sh
TMOUT="$(( 60*10 ))";
[ -z "$DISPLAY" ] && export TMOUT;
case $( /usr/bin/tty ) in
	/dev/tty[0-9]*) export TMOUT;;
esac

If you really want EVERY Bash/Zsh prompt (even within X) to timeout, use:

$ export TMOUT="$(( 60*10 ))";

Note that this will not work if there is some command running in the shell (eg.: an SSH session or other shell without TMOUT support). But if you are using VC mostly for restarting frozen GDM/Xorg as root, then this is very useful.

Protect against rogue USB devices

Install USBGuard, which is a software framework that helps to protect your computer against rogue USB devices (a.k.a. BadUSB, PoisonTap or LanTurtle) by implementing basic whitelisting and blacklisting capabilities based on device attributes.

Packages

Authentication

Attacks on package managers are possible without proper use of package signing, and can affect even package managers with proper signature systems. Arch uses package signing by default and relies on a web of trust from 5 trusted master keys. See Pacman-key for details.

Upgrades

It is important to regularly upgrade the system.

Follow vulnerability alerts

Subscribe to the Common Vulnerabilities and Exposure (CVE) Security Alert updates, made available by National Vulnerability Database, and found on the NVD Download webpage. The Arch Linux Security Tracker serves as a particularly useful resource in that it combines Arch Linux Security Advisory (ASA), Arch Linux Vulnerability Group (AVG) and CVE data sets in tabular format. See also Arch Security Team.

You should also consider subscribing to the release notifications for software you use, especially if you install software through means other than the main repositories or AUR. Some software have mailing lists you can subscribe to for security notifications. Source code hosting sites often offer RSS feeds for new releases.

Rebuilding packages

Packages can be rebuilt and stripped of undesired functions and features as a means to reduce attack surface. For example, bzip2 can be rebuilt without bzip2recover in an attempt to circumvent CVE-2016-3189. Custom hardening flags can also be applied either manually or via a wrapper.

See also