Box fails to start with exit code 159 (SIGSYS/seccomp) on Linux kernel 6.17 — nanosleep syscall blocked

[winni2k]

Running examples/python/01_getting_started/run_codebox.py crashes on my system. I had claude code help me draft this bug report.

Summary

All boxes fail to start with exit code 159 (SIGSYS) on Linux kernel 6.17.9. The boxlite-shim process is killed by the seccomp filter because the nanosleep syscall
is not in the allowlist.

Environment

• OS: Pop!_OS (Ubuntu-based), Linux 6.17.9-76061709-generic x86_64
• boxlite SDK: 0.5.13 (also reproduced on 0.5.12)
• Python: 3.14.2
• KVM: /dev/kvm is accessible (read+write)

Steps to Reproduce

uv run python examples/python/01_getting_started/run_codebox.py

Error Output

RuntimeError: engine reported an error: Box 01KHESHRNKZEKPFJZTR7Q1NQBP failed to start

Exit code: 159 (unknown signal)

Boxlite Logs

From ~/.boxlite/logs/boxlite.log.2026-02-14:

2026-02-14T19:20:26.613942Z INFO boxlite::jailer::command: Building bwrap-isolated command
2026-02-14T19:20:26.649097Z INFO boxlite::jailer::command: Using copied shim binary (Firecracker pattern)
original=/home/winni/proj/sandi/.venv/lib/python3.14/site-packages/boxlite/runtime/boxlite-shim
copied=/home/winni/.boxlite/boxes/01KHESHRNKZEKPFJZTR7Q1NQBP/bin/boxlite-shim
2026-02-14T19:20:26.650443Z INFO boxlite::vmm::controller::shim: boxlite-shim subprocess spawned box_id=01KHESHRNKZEKPFJZTR7Q1NQBP pid=10509
shim_spawn_duration_ms=40
2026-02-14T19:20:26.650568Z INFO boxlite::vmm::controller::shim: VM subprocess started successfully box_id=01KHESHRNKZEKPFJZTR7Q1NQBP
2026-02-14T19:20:27.151516Z WARN boxlite::litebox::init::tasks::guest_connect: VM subprocess exited unexpectedly during startup pid=10509 exit_code=159
2026-02-14T19:20:27.151577Z ERROR boxlite::litebox::init::tasks: Task failed: engine reported an error: Box 01KHESHRNKZEKPFJZTR7Q1NQBP failed to start

Root Cause Analysis

I traced the shim process with strace -f and found the cause. The shim's threads are killed by a seccomp filter that blocks the nanosleep syscall:

--- SIGSYS {si_signo=SIGSYS, si_code=SYS_SECCOMP, si_call_addr=0x74a74d818497, si_syscall=__NR_nanosleep, si_arch=AUDIT_ARCH_X86_64} ---
+++ killed by SIGSYS (core dumped) +++

Exit code 159 = 128 + 31 (SIGSYS). The seccomp filter applied by the bwrap jailer does not include nanosleep in its allowlist. On kernel 6.17, something in
libkrun's threads calls nanosleep (rather than clock_nanosleep), triggering the kill.

Full strace excerpt showing the cascade — one thread hits the seccomp violation, then all threads are killed:

10047 rt_sigaction(SIGSYS, NULL, {sa_handler=SIG_DFL, sa_mask=[], sa_flags=0}, 8) = 0
10049 --- SIGSYS {si_signo=SIGSYS, si_code=SYS_SECCOMP, si_call_addr=0x74a74d818497, si_syscall=__NR_nanosleep, si_arch=AUDIT_ARCH_X86_64} ---
10054 +++ killed by SIGSYS (core dumped) +++
10053 +++ killed by SIGSYS (core dumped) +++
10052 +++ killed by SIGSYS (core dumped) +++
10051 +++ killed by SIGSYS (core dumped) +++
10050 +++ killed by SIGSYS (core dumped) +++
10049 +++ killed by SIGSYS (core dumped) +++
10048 +++ killed by SIGSYS (core dumped) +++
10047 +++ killed by SIGSYS (core dumped) +++
10046 exit_group(159)
10045 exit_group(159)

Suggested Fix

Add nanosleep to the seccomp allowlist in the jailer/bwrap configuration (alongside the existing clock_nanosleep entry).

[DorianZheng]

Hi @winni2k. I have disabled seccomp by default in v0.5.13 and am revamping the seccomp design. You can disable seccomp for testing by now through SecurityOptions. Will fix it ASAP and enable seccomp by default after that.