fix(udp): use IPV6_PMTUDISC_PROBE instead of IP_PMTUDISC_PROBE on v6

[mxinden]

Both resolve to 3, thus not an actual bug.

Caught thanks to @marten-seemann's DF draft.

[Ralith]

Thanks! Never know when we might hit a weird environment where these differ.

[mxinden]

Never know when we might hit a weird environment where these differ.

Agreed. The more complexity we can offload to other crates (here libc) the better.

[mstyura]

Sorry for late comment, maybe it worth adding comment explaining why IP_PMTUDISC_PROBE is chosen over IP_PMTUDISC_DO

Path MTU discovery value   Meaning
              IP_PMTUDISC_WANT           Use per-route settings.
              IP_PMTUDISC_DONT           Never do Path MTU Discovery.
              IP_PMTUDISC_DO             Always do Path MTU Discovery.
              IP_PMTUDISC_PROBE          Set DF but ignore Path MTU.

E.g. Chromium uses later option in context when DF flag is desired, WebRTC also. So there is should be something important which probably deserves explanation why quinn uses IP_PMTUDISC_PROBE rather than IP_PMTUDISC_DO.

[Ralith]

It seems pretty self-explanatory to me? We do our own MTUD, so we don't want the kernel's built-in MTUD confusing things.

[mstyura]

We do our own MTUD, so we don't want the kernel's built-in MTUD confusing things.

Thanks, this is exact explanation I missed! WebRTC uses IP_PMTUDISC_DO option probably due to by default conservative packet size 1200 and lack of it's own MTU discovery. And why Chromium don't use IP_PMTUDISC_PROBE in it's code also unclear even so it obviously has QUIC implementation.

[mxinden]

@mstyura related discussion marten-seemann/draft-seemann-tsvwg-udp-fragmentation#13. Might be worth moving the Chromium discussion there.