feat: improve struct_ops codegen lifecycle and test coverage

[SiyuanSun0736]

Summary

This PR improves struct_ops code generation and strengthens regression coverage across both eBPF and userspace paths.

Key changes:

• keep internal struct_ops function calls as direct calls instead of lowering them into implicit tail calls
• synthesize a stable default name for struct_ops instances when the kernel struct expects a name field but the impl block omits it
• improve userspace struct_ops lifecycle generation around main so registration results, attach helpers, wait flow, and detach/cleanup behavior are emitted more reliably
• expand struct_ops tests to cover runtime helpers, cleanup hooks, attach/detach helpers, EPERM diagnostics, and direct-call behavior

Details

In IR generation:

• tracked the current program type during lowering so struct_ops methods are not treated like normal attributed programs for implicit tail-call lowering
• added logic to preserve direct calls for internal struct_ops method-to-method invocations
• generated a default struct_ops name automatically when the target kernel struct has a name field and no explicit static field is provided

In userspace code generation:

• improved detection of the main registration flow for struct_ops programs
• preserved the registration result through generated attach/wait/detach lifecycle code
• strengthened lifecycle generation so cleanup helpers, unregister waiting, and detach handling are validated by tests

In examples and tests:

• updated the struct_ops example to include an internal helper-style call path through undo_cwnd -> ssthresh
• added regression coverage ensuring struct_ops internal calls remain direct calls
• expanded userspace struct_ops codegen tests to verify runtime helper emission, cleanup registration, attach/detach helpers, and struct_ops-specific load diagnostics

Testing

• extended struct_ops code generation unit tests
• added regression coverage for direct internal calls in struct_ops impl blocks
• added userspace codegen assertions for runtime helpers and cleanup flow

[Copilot]

Pull request overview

This PR refines struct_ops code generation to better preserve intended call semantics (direct internal calls), auto-fill missing instance metadata (default .name), and generate a more reliable userspace registration/attach/wait/detach lifecycle, with expanded regression tests across both eBPF and userspace outputs.

Changes:

• Preserve direct calls for internal struct_ops method-to-method invocations and avoid lowering them into implicit tail calls.
• Synthesize a stable default struct_ops instance name when the kernel struct has a name field and the impl omits it.
• Improve userspace main() lifecycle generation for struct_ops (runtime checks, link lifetime, attach/wait/detach/cleanup) and expand tests accordingly.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
src/ir_generator.ml	Tracks current program type to prevent implicit tail-call lowering for struct_ops, and synthesizes a default instance name when omitted.
src/userspace_codegen.ml	Reworks struct_ops registration to use generated attach/detach helpers, adds runtime helpers/cleanup/wait flow, and adds EPERM diagnostics.
tests/test_struct_ops.ml	Expands userspace lifecycle assertions, adds synthesized-name coverage, and adds a regression test ensuring internal calls stay direct.
examples/struct_ops_simple.ks	Updates the example to include an internal helper-style call path (undo_cwnd -> ssthresh).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

ensure_struct_ops_privileges() claims struct_ops loading works with root or CAP_BPF/CAP_SYS_ADMIN, but it only checks geteuid() == 0 and hard-fails otherwise. This will incorrectly reject setups that run unprivileged with the required capabilities. Consider either (a) making this a warning and letting load/attach fail naturally, (b) checking capabilities explicitly, or (c) updating the message to match the actual requirement (root only).

Suggested change

	fprintf(stderr, "struct_ops loading requires root privileges or CAP_BPF/CAP_SYS_ADMIN.\n");
	fprintf(stderr, "Run this example with sudo, for example: sudo ./%s\n");
	return -1;
	fprintf(stderr, "Warning: struct_ops loading typically requires root privileges or CAP_BPF/CAP_SYS_ADMIN.\n");
	fprintf(stderr, "Continuing anyway; loading may still succeed if this process has the required capabilities.\n");
	fprintf(stderr, "If it fails with a permission error, try: sudo ./%s\n");
	return 0;

[Copilot]

The generated struct_ops lifecycle code sets the registration result variable but then returns a hard-coded 1 on attach failure (if (attach_status != 0) { ... return 1; }). This loses the actual error code produced by attach_struct_ops_*() and deviates from the user-written return result; pattern. Returning the real registration result (or propagating attach_status) would better preserve semantics and diagnostics.

[Copilot]

This regression test asserts that no tail call is emitted by searching for the exact string bpf_tail_call(ctx, &prog_array, 0). Tail-call indices (or emitted formatting) can change without changing the underlying bug, which could let regressions slip through. Consider checking for a more general bpf_tail_call(ctx, &prog_array/bpf_tail_call( substring (or absence of prog_array) instead of a specific index literal.

Suggested change

	(contains_substr c_code "bpf_tail_call(ctx, &prog_array, 0)")
	(contains_substr c_code "bpf_tail_call(ctx, &prog_array")