The eBPF Syscall Scryer: Real-Time Kernel Execve Ward
Instrument the Linux kernel syscall boundary in real-time with bpftrace to intercept and log every binary execution, catching stealthy reverse shells and unlogged processes.
bpftrace -e 'tracepoint:syscalls:sys_enter_execve { printf("[EXECVE] ts=%u pid=%d comm=%s args: ", elapsed / 1000000, pid, comm); join(args->argv); }' The Lineage to Ambient Kernel Observability
In systems architecture, traditional observability requires recompiling binaries, modifying shared libraries (LD_PRELOAD), or injecting intrusive user-space proxies. However, sophisticated attackers, rootkits, and anomalous container processes evade user-space hooks by invoking raw syscalls directly.
The Linux kernel’s Extended Berkeley Packet Filter (eBPF) provides an ambient, non-intrusive probe architecture:
- eBPF programs run sandboxed inside the Linux kernel virtual machine (JIT-compiled to native CPU instructions).
- They attach directly to static kernel tracepoints (
tracepoint:syscalls:sys_enter_execve) or dynamic probes (kprobes). - Execution takes nanoseconds with zero overhead on running production workloads.
The eBPF Syscall Scryer provides true Sight Beyond Sight: it sits inside kernel space, logging every single executable invoked on the machine—including ephemeral commands that exit too quickly for ps aux or top to ever capture.
The Spell Formula
Cast this invocation to attach an eBPF tracepoint probe that logs every command executed on the host or inside any container:
bpftrace -e 'tracepoint:syscalls:sys_enter_execve {
printf("[EXECVE %u ms] PID=%d PPID=%d comm=%s args: ",
elapsed / 1000000, pid, curtask->parent->pid, comm);
join(args->argv);
}'
To trace only suspicious commands executing out of /tmp, /dev/shm, or hidden directories:
bpftrace -e 'tracepoint:syscalls:sys_enter_execve /str(args->filename) == "/tmp" || str(args->filename) == "/dev/shm"/ {
printf("[MALICIOUS PATH DETECTED] PID=%d Binary=%s\n", pid, str(args->filename));
}'
Anatomy of the Spell
1. tracepoint:syscalls:sys_enter_execve
- Hooks directly into the entrypoint of the Linux
sys_execve(2)system call. - Guaranteed to fire regardless of whether the command was run via SSH, Cron, Docker, Kubernetes, or an exploited web service.
2. elapsed / 1000000
- Normalizes high-resolution kernel nanosecond monotonic clocks into human-readable milliseconds.
3. pid, curtask->parent->pid, comm
pid: Process ID executing the binary.curtask->parent->pid: Reaches directly into the kernel’stask_structto identify the Parent PID without user-space race conditions.comm: Name of the parent process.
4. join(args->argv)
- Concatenates the variable-length array of string arguments (
argv) into a single aligned line of telemetry.
Observability Comparison: ps vs. Auditd vs. eBPF
| Dimension | Periodic Polling (ps / top) |
Linux Audit Framework (auditd) |
eBPF Syscall Scryer |
|---|---|---|---|
| Short-Lived Commands (<5ms) | Misses 90% of ephemeral commands | Catches commands but heavy CPU overhead | Zero missed events; nanosecond JIT speed |
| Kernel Overhead | High I/O scanning /proc |
High context-switching penalties | Negligible (<1% CPU consumption) |
| Container Awareness | Blind to container namespace boundaries | Requires complex rule tagging | Hooks kernel boundary across all namespaces |
| Safety Invariants | User-space only | Can overflow buffer queues | Kernel eBPF verifier guarantees zero crash |
Through kernel eBPF tracepoints, the Divination school grants sovereign visibility into every heartbeat of process execution.