Transforming functions between compatible hook points (MERGE-RELATED) #88
Description
Part of the optimization done by OPENED is to shift probes around compatible hook points (see compatible_hookpoints on this link) to maximize performance and resource usage.
As far as I understand, we currently assume a probe P is compatible to a hook point H if all helper functions of P are available in H
A few considerations:
- is this really all it takes for the "majority" of cross hook point compilations to succeed?
- Is the transformation always semantically equivalent? Can the context state or even return values of helper functions change depending on the hookpoint attached? e.g., XDP doesn't have a PID vs non-XDP generally has a PID. Uprobe has the original function's address on the instruction pointer, Uretprobe has the BPF trampoline's address.
- Specifically for the current approach: How to deal with helper functions that are named slightly differently but "equivalent" enough amongst themselves?