[LSF/MM/BPF TOPIC] BPF: Various topics

[LSF/MM/BPF TOPIC] BPF: Various topics

[Joe Stringer]

I'd be interested to participate in the BPF track to discuss the
following topics on BPF, motivated from recent experiences in building
Cilium and a Go eBPF library.

* Improving verifier friendliness. Cilium generates complex BPF
programs which can trigger poor verifier behaviour, eg. a program
would be rejected unless we provided a log buffer of 64MB[0]. Would
like to discuss improvements here like allowing the verifier to treat
the user buffer as a ring buffer via opt-in flag, so that users can
bound the verifier log size and capture only the last N relevant bytes
of verification failure.
* Improving the map iteration interactions. During development of
cilium/ebpf go library, we hit issues defining a clean API providing
guarantees around iteration bounding and completeness of dump[1].
These could be improved via kernel extension, such as opt-in ability
to detect when dumping next key leads to iteration reset to beginning
of map. Also interested in potential for kernel assistance on LRU
eviction handling.
* Development of pure-Go eBPF library. Depending on the audience
appetite this could be hallway track but syncing on library status,
next steps, kernel API pain points.
* Any further updates on socket redirection following LPC discussions.
Detail unclear given that this is in flux.
* Future of BPF interaction with other subsystems / unstable BPF
helpers API per recent discussion on the list.[2]

[0] https://github.com/cilium/cilium/issues/9809 ,
https://github.com/cilium/cilium/issues/7770#issuecomment-558343180,
other PRs.
[1] https://github.com/cilium/ebpf/pull/11
[2] https://www.spinics.net/lists/netdev/msg628451.html