On Fri, 6 Oct 2023 13:18:40 +0200 Daniel Borkmann wrote:
> That should be possible with some work this way, agree. I've been toying a bit
> more on this issue, and actually there is an even better way which would cleanly
> solve all use cases and we likely would utilize it for bpf as well in future.
> I wasn't aware of it before, but the drop reason actually has per-subsystem infra
> already which so far only mac80211 and ovs makes use of.
FWIW I'm not sure if leaning into the subsys specific error codes for
something as core as TC is a good direction. I'd think that what
matters to the user is was it an intentional policy drop or some form
of an error / exception. More detailed info can come from stats.
Maybe I'm overly conservative because I don't care about debugging
mac80211 or ovs but do very much care about TC :) And I think Alastair
(bpftrace) is working on auto-prettifying enums when bpftrace outputs
maps. So we can do something like:
$ bpftrace -e 'tracepoint:skb:kfree_skb { @[args->reason] = count(); }'
Attaching 1 probe...
^C
@[SKB_DROP_REASON_TC_INGRESS]: 2
@[SKB_CONSUMED]: 34
^^^^^^^^^^^^ names!!
Auto-magically.
Which will no longer work with the "pack multiple values into
the reason" scheme of subsys-specific values :(
What I'm saying is that there is a trade-off here between providing
as much info as possible vs basic user getting intelligible data..
