feat(memtrack): support running the memory instrument in the macro-agent sandbox

[GuillaumeLagrange]

Make the memory instrument work when the runner and codspeed-memtrack run inside the macro-agent sandbox — unprivileged, in a user + PID namespace, with a delegated BPF token instead of root. Companion to the platform PR that sets up the token and the exe-gate.

• Attach via bpf()-native links. Switch uprobes to uprobe_multi and the fork tracepoint to tp_btf, so a BPF token can authorize them (the old perf_event_open path needs CAP_PERFMON). eBPF logic unchanged.
• Resolve PIDs in the tracker's PID namespace. eBPF sees global PIDs but memtrack registers namespace-local ones, so nothing matched inside the sandbox. In the init namespace this still resolves to global PIDs.
• Accept a delegated BPF token as a privilege source. Treat a live LIBBPF_BPF_TOKEN_PATH as satisfied privilege instead of requiring root / file-capabilities.

Proven end-to-end on the macro-runner: codspeed run -m memory on an instrumented divan bench runs inside the sandbox and uploads to staging.

Refs COD-3047

[greptile-apps]

Greptile Summary

This PR makes memory tracking work in the macro-agent sandbox. The main changes are:

• Adds token and perf eBPF skeleton variants.
• Uses bpf-native links for token-based attachment.
• Resolves tracked PIDs relative to memtrack's PID namespace.
• Treats delegated BPF tokens as a privilege source.
• Adds flavor-equivalence coverage for the two attach paths.

Confidence Score: 4/5

This is close, but the realloc tracking gap should be fixed before merging.

• Realloc entry probes still compare the host PID against namespace-local tracked PIDs.
• In the sandbox, that can drop realloc events without an error.
• The rest of the token and namespace wiring is narrowly scoped.

crates/memtrack/src/ebpf/c/memtrack.bpf.c

Important Files Changed

Filename	Overview
crates/memtrack/src/ebpf/c/memtrack.bpf.c	Adds attach-mode selection and namespace-aware PID helpers, but the realloc-family entry path still checks the host PID.
crates/memtrack/src/ebpf/memtrack.rs	Adds token/perf skeleton selection, offset-based uprobe attachment, and PID namespace rodata setup.
src/executor/memory/executor.rs	Accepts delegated BPF tokens as an alternative to root or memtrack file capabilities.

Comments Outside Diff (1)

1. crates/memtrack/src/ebpf/c/memtrack.bpf.c, line 319-321 (link)

   Realloc uses host PID

   In the sandbox, userspace registers the namespace-local PID in tracked_pids, but this entry probe still checks tid >> 32, which is the host PID. When those values differ, this branch returns before saving realloc_args, and the return probe has no arguments to emit the realloc event. Realloc allocations are then silently missing from the memory trace.

   Prompt To Fix With AI

   This is a comment left during a code review.
   Path: crates/memtrack/src/ebpf/c/memtrack.bpf.c
   Line: 319-321
   
   Comment:
   **Realloc uses host PID**
   
   In the sandbox, userspace registers the namespace-local PID in `tracked_pids`, but this entry probe still checks `tid >> 32`, which is the host PID. When those values differ, this branch returns before saving `realloc_args`, and the return probe has no arguments to emit the realloc event. Realloc allocations are then silently missing from the memory trace.
   
   How can I resolve this? If you propose a fix, please make it concise.

   

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
crates/memtrack/src/ebpf/c/memtrack.bpf.c:319-321
**Realloc uses host PID**

In the sandbox, userspace registers the namespace-local PID in `tracked_pids`, but this entry probe still checks `tid >> 32`, which is the host PID. When those values differ, this branch returns before saving `realloc_args`, and the return probe has no arguments to emit the realloc event. Realloc allocations are then silently missing from the memory trace.

_{Reviews (4): Last reviewed commit: "feat(memtrack): allow memtrack usage wit..." | Re-trigger Greptile}

[codspeed-hq]

Merging this PR will not alter performance

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

✅ 7 untouched benchmarks

---

_{Comparing cod-3047-support-memory-instrument-on-macro-agent-sandbox (81c4b15) with main (3278ec8)}

[GuillaumeLagrange]

The changes from this PR have been moved to #456, which combines the runner-side work for COD-3012 (hook-based CPU isolation) and COD-3047 (BPF-token memory instrument) into a single PR. Closing in favor of that one.