Jakub Kicinski <kuba@xxxxxxxxxx> writes: > On Tue, 16 Apr 2024 20:32:14 +0100 Donald Hunter wrote: >> The nfnetlink family uses the directional op model but errors get >> reported using the request value instead of the reply value. > > What's an error in this case ? "Normal" errors come via NLMSG_ERROR Thanks for pointing out what should have been obvious. Looking at it again today, I realise I missed the root cause which was a bug in the extack decoding for directional ops. When I fix that issue, this patch can be dropped. >> diff --git a/tools/net/ynl/lib/nlspec.py b/tools/net/ynl/lib/nlspec.py >> index 6d08ab9e213f..04085bc6365e 100644 >> --- a/tools/net/ynl/lib/nlspec.py >> +++ b/tools/net/ynl/lib/nlspec.py >> @@ -567,6 +567,18 @@ class SpecFamily(SpecElement): >> return op >> return None >> >> + def get_op_by_value(self, value): >> + """ >> + For a given operation value, look up operation spec. Search >> + by response value first then fall back to request value. This >> + is required for handling failure cases. > > Looks like we're only going to need it in NetlinkProtocol, so that's > fine. But let's somehow call out that this is a bit of a hack, so that > people don't feel like this is the more correct way of finding the op > than direct access to rsp_by_value[].
