{ config, lib, pkgs, ... }: let cfg = config.security.audit; failureModes = { silent = 0; printk = 1; panic = 2; }; # The order of the fixed rules is determined by augenrules(8) rules = pkgs.writeTextDir "audit.rules" '' -D -b ${toString cfg.backlogLimit} -f ${toString failureModes.${cfg.failureMode}} -r ${toString cfg.rateLimit} ${lib.concatLines cfg.rules} -e ${if cfg.enable == "lock" then "2" else "1"} ''; in { options = { security.audit = { enable = lib.mkOption { type = lib.types.enum [ false true "lock" ]; default = false; description = '' Whether to enable the Linux audit system. The special `lock` value can be used to enable auditing and prevent disabling it until a restart. Be careful about locking this, as it will prevent you from changing your audit configuration until you restart. If possible, test your configuration using build-vm beforehand. ''; }; failureMode = lib.mkOption { type = lib.types.enum [ "silent" "printk" "panic" ]; default = "printk"; description = "How to handle critical errors in the auditing system"; }; backlogLimit = lib.mkOption { type = lib.types.int; # Significantly increase from the kernel default of 64 because a # normal systems generates way more logs. default = 1024; description = '' The maximum number of outstanding audit buffers allowed; exceeding this is considered a failure and handled in a manner specified by failureMode. ''; }; rateLimit = lib.mkOption { type = lib.types.int; default = 0; description = '' The maximum messages per second permitted before triggering a failure as specified by failureMode. Setting it to zero disables the limit. ''; }; rules = lib.mkOption { type = lib.types.listOf lib.types.str; # (types.either types.str (types.submodule rule)); default = [ ]; example = [ "-a exit,always -F arch=b64 -S execve" ]; description = '' The ordered audit rules, with each string appearing as one line of the audit.rules file. ''; }; }; }; config = lib.mkIf (cfg.enable == "lock" || cfg.enable) { boot.kernelParams = [ # A lot of audit events happen before the systemd service starts. Thus # enable it via the kernel commandline to have the audit subsystem ready # as soon as the kernel starts. "audit=1" # Also set the backlog limit because the kernel default is too small to # capture all of them before the service starts. "audit_backlog_limit=${toString cfg.backlogLimit}" ]; environment.systemPackages = [ pkgs.audit ]; # upstream contains a audit-rules.service, which uses augenrules. # That script does not handle cleanup correctly and insists on loading from /etc/audit. # So, instead we have our own service for loading rules. systemd.services.audit-rules-nixos = { description = "Load Audit Rules"; wantedBy = [ "sysinit.target" ]; before = [ "sysinit.target" "shutdown.target" ]; conflicts = [ "shutdown.target" ]; unitConfig = { DefaultDependencies = false; ConditionVirtualization = "!container"; ConditionKernelCommandLine = [ "!audit=0" "!audit=off" ]; }; serviceConfig = { Type = "oneshot"; RemainAfterExit = true; ExecStart = "${lib.getExe' pkgs.audit "auditctl"} -R ${rules}/audit.rules"; ExecStopPost = [ # Disable auditing "${lib.getExe' pkgs.audit "auditctl"} -e 0" # Delete all rules "${lib.getExe' pkgs.audit "auditctl"} -D" ]; }; }; }; }