btf: don't try to load strings when file is empty

[therealak12]

I opened this PR to address issue#1818.

In my experience, the problem has been occurring since version v0.19.0 (it does not appear in v0.18.0), specifically for the aesni_intel kernel module. The error encountered is:

apply CO-RE relocations: load BTF for kmod aesni_intel: read string section: string table is empty
exit status 1

I've checked stat /sys/kernel/btf/aesni_intel. It shows 24 bytes, which is for the BTF header, but it has no content.

[dylandreimerink]

This seems to be a duplicate / alternative PR for https://github.com/cilium/ebpf/pull/1828/files

[therealak12]

This seems to be a duplicate / alternative PR for https://github.com/cilium/ebpf/pull/1828/files

That one returns a non-empty list while the file is empty. IMHO, this is a better solution.

[lmb]

Thanks for adding the test, that was useful. Rebased this on main and changed it to use the approach where newStringTable deals with emptiness.

[therealak12]

I assumed introducing an outer check on StringLen could help avoid unnecessary function and system calls.

[lmb]

There are no system calls performed, and function calls are not really measurable at this level. Maybe you mean because the original code did some io.Read? That is gone now.