Compilation optimizations cause ebpf verify to fail about libbpf-bootstrap HOT 4 CLOSED

It's a generally recognized (in the BPF community) problem. To quote Alexei Starovoitov:

we brought this issue [the brittle nature of optimizations in conjunction with the verifier] to llvm folks many times. Unfortunately to deaf ears. GCC folks are more accommodating. The plan is to have -Overifiable flag that will turn off optimizations that lose information. Once GCC leads the way maybe clang folks will start listening. afaik llvm community lost a bunch of great engineers to AI projects and that shows.

from libbpf-bootstrap.

It's a generally recognized (in the BPF community) problem. To quote Alexei Starovoitov:

we brought this issue [the brittle nature of optimizations in conjunction with the verifier] to llvm folks many times. Unfortunately to deaf ears. GCC folks are more accommodating. The plan is to have -Overifiable flag that will turn off optimizations that lose information. Once GCC leads the way maybe clang folks will start listening. afaik llvm community lost a bunch of great engineers to AI projects and that shows.

Are there any best practices to avoid these problems as much as possible? I tried to look for this information on the Internet, but I couldn't find it.

from libbpf-bootstrap.

Are there any best practices to avoid these problems as much as possible? I tried to look for this information on the Internet, but I couldn't find it.

Not aware of any. Others may know more, though.

from libbpf-bootstrap.

try storing *offset result in local variable and then check that it is within correct range. And then use that local variable instead of dereferencing *offset again in bt->buf[*offset]

from libbpf-bootstrap.