Seccomp Syscall Filter Sandbox (2D)

Default Policy Preset

Syscall Categories (click to cycle)

Syscall rate 6/s

● PROCESS RUNNING

Live Telemetry

Syscalls / sec

0.0

Threat score

Allowed (total)

Blocked (total)

How it works

Linux seccomp-BPF lets a process install a small filter program the kernel runs on every syscall before it executes. Given the syscall number, architecture and arguments, the filter returns one action:

action = F(syscall_nr, args)  ∈  { ALLOW, LOG/TRAP, KILL_PROCESS }
F is evaluated in the kernel, once per syscall, before it runs

This sandbox groups real syscalls into six categories and lets you set the policy per category:

Allow (green node) — the syscall executes normally, the dot flies on through to the kernel.
Trap-log (amber) — mirrors SECCOMP_RET_TRAP/RET_LOG: the call is denied and audited, but the process keeps running.
Kill (red) — mirrors SECCOMP_RET_KILL_PROCESS: the very first matching syscall terminates the whole process immediately (a container runtime would then restart it).

The Attack Burst button fires a rapid run of genuinely dangerous syscalls (ptrace, mount, setuid, execve) — exactly the calls a container-escape or privilege-escalation exploit needs. If those categories are set to Allow, each one that reaches the kernel raises the threat score; set them to Kill or Trap-log and the burst is contained (or the process is sacrificed) before it can do damage. This is the same trade-off real sandbox profiles (Docker's default seccomp profile, gVisor, Firecracker jailers) make: a tighter allow-list is safer but breaks more legitimate software. Drag the view to pan and scroll/pinch to zoom.

Allow — reaches kernel Trap-log — denied & audited Kill — process terminated

drag to pan · scroll to zoom