Re: [net-next PATCH v3 3/8] net: sched: add cls_u32 offload hooks for netdevs

On 16-02-18 10:24 AM, John Fastabend wrote:

On 16-02-18 04:14 AM, Jamal Hadi Salim wrote:

quoted

On 16-02-17 06:07 PM, John Fastabend wrote:

quoted

[...]

quoted

IMO, it would be better at this early stage to enforce the correct
behavior for future generations.
To follow the netlink semantics which a lot of people are already
trained to think in.

Current netlink behavior is supposed to be:

1) NEW ==> "Create".
Ambigous - could mean a)"create if it doesnt exist" or b) "fail if it
exists otherwise create"
Unfortunately different parts of the kernel often assume some
default from either #a or #b.

But this is already handled by the core cls_api.c code. We never
get to u32_change if the flags are not correct.

Look at the block right above the op call into the classifiers
change() code in cls_api.c. Starting at line 287.

Indeed that would cover s/ware filters but not h/ware. That will
depend on what hardware can do. Really,
all you need is to propagate those flags. Your
driver ndo can ignore them. I know i will need them.

Alternatively: If we say all filters are going to be stored in
s/ware as well i.e the tri-state we talked about then the cls_api
checks will work as well. But when you are talking millions
of filters (such as i deal with) - that may become impractical
(and then you are going to have all kind of clever things to find
whether an EXCLUSIVE will work or not etc).

cheers,
jamal