Netconf 2018 Day 2

Posted Jun 11, 2018 20:28 UTC (Mon) by Cyberax (✭ supporter ✭, #52523)
Parent article: Netconf 2018 Day 2

Can we just skip all this nonsense with eBPF and just go directly to using LLVM bitcode? Heck, just JIT-compile it in the userspace into native code.

/me runs and hides

Netconf 2018 Day 2

Posted Jun 11, 2018 20:29 UTC (Mon) by Cyberax (✭ supporter ✭, #52523) [Link]

On the second thought, maybe we should just embed a V8 JavaScript engine into the kernel? After all, NetBSD did the same with LUA.

Netconf 2018 Day 2

Posted Jun 11, 2018 22:13 UTC (Mon) by nybble41 (subscriber, #55106) [Link]

> Can we ... go directly to using LLVM bitcode? Heck, just JIT-compile it in the userspace into native code.

That doesn't seem so unreasonable to me, provided the JIT-compilation occurs within a restricted user-mode helper and targets a subset of the native machine language amenable to verification (as with NaCl). The verifier, of course, would need to run in a separate user-mode helper context. LLVM bitcode is not fully portable, though—something like WebAssembly might be a better choice.

Netconf 2018 Day 2

Posted Jun 12, 2018 6:50 UTC (Tue) by magfr (subscriber, #16052) [Link]

Isn't there a tongue in cheek law stating that all programs will develop their own internal lisp parser at some point?
Lisp is also nice since it naturally promotes some of the eBPF limitations.