bun x / bun run / Bun.spawn crash on Android 10: pidfd_open trapped by seccomp as SIGSYS

[HanSoBored]

Summary

On Android < 12 (kernel < 5.3, e.g. kernel 4.9 / SDK 29), the app-sandbox seccomp policy TRAPs pidfd_open as SIGSYS. Bun calls pidfd_open when spawning child processes, so anything that spawns crashes:

• bun x cowsay ... → exit 134 (SIGABRT)
• bun run <script-with-script> → error: script was terminated by signal SIGSYS
• Bun.spawn() → exit 159

Root cause

Two-part:

1. Bun's spawn path calls pidfd_open, which Android seccomp TRAPs as SIGSYS instead of returning ENOSYS.
2. Bun then installs its own SIGSYS crash handler (SA_RESETHAND) at startup, so the trap lands in Bun's abort path rather than letting the shim's fallback run.

bun -e '...' (no child spawn) works fine; only child-process spawning fails.

Reproduction

bun x cowsay "hi"        # exit 134
bun run hi               # error: script was terminated by signal SIGSYS

Proposed fix

Add a SIGSYS handler to bun-shim.so (same technique as opencode-termux's libseccomp-shim):

• A SIGSYS handler returns -ENOSYS for seccomp-trapped syscalls, so Bun falls back to waitpid.
• Interpose sigaction() so Bun's own SA_RESETHAND crash handler cannot override ours.

Verified on aarch64/Android 10 (kernel 4.9): all child spawn paths work, and tests/run-tests.sh passes 9/9.

Environment

• Termux (aarch64), Android 10, kernel 4.9.186
• Bun 1.3.14

[Happ1ness-dev]

Hello, thank you for reporting this issue and taking your time to write a solution for it.

I'm currently unable to review your PR due to my irl situation, since my whole city is blacked out and I rarely have any electricity/internet.

I'll try to review it once my situation improves.

[HanSoBored]

Hello, thank you for reporting this issue and taking your time to write a solution for it.

I'm currently unable to review your PR due to my irl situation, since my whole city is blacked out and I rarely have any electricity/internet.

I'll try to review it once my situation improves.

Hello, I read the discussion in issue #13, where you decided against switching to the native Android build from upstream Bun and opted to stick with the glibc instead. In my opinion—even though the native Android Bun currently has some minor issues, such as benchmarking slightly slower than the glibc approach—upstream Bun will continue to improve their product; the native Android version is still relatively new and will get better over time. What I’d like to suggest is considering backward compatibility—specifically, allowing bun-termux to run on older kernels like 4.9, rather than sticking to the 5.1 minimum required by upstream Bun. Implementing fallbacks (like statx->fstatat, pidfd_open->waitpid, etc.) seems more rational to me and would help the community, run bun on their older phones.

[Happ1ness-dev]

I might pivot to bionic later with this approach, but currently the focus is on glibc version, specifically for anyone who wants to run glibc bun for any reason.

allowing bun-termux to run on older kernels like 4.9, rather than sticking to the 5.1 minimum required by upstream Bun.

This is already the case in Bun-Termux, in a way. Many people have successfully run it on devices with older kernels.

In my case, I'm running it on Android 12 (MIUI V13.0.7.0.SKRUXM) with kernel version 4.19.157.

[HanSoBored]

In my case, I'm running it on Android 12 (MIUI V13.0.7.0.SKRUXM) with kernel version 4.19.157.

sounds great, because android 12 has whitelisted pidfd_open.

however this patch seems will also solve problem on Issue #10, he got exact same problem, which is bun fails when spawning a new process ( bun --bun x -> abort) on android 9/10.