Re: [RFC PATCH net-next 12/15] ipv4: Add "in hardware" indication to routes

From: Ido Schimmel <idosch@idosch.org>
To: David Ahern <dsahern@gmail.com>
Cc: Jiri Pirko <jiri@resnulli.us>,
	Roopa Prabhu <roopa@cumulusnetworks.com>,
	netdev <netdev@vger.kernel.org>,
	David Miller <davem@davemloft.net>,
	Jiri Pirko <jiri@mellanox.com>,
	Jakub Kicinski <jakub.kicinski@netronome.com>,
	Saeed Mahameed <saeedm@mellanox.com>, mlxsw <mlxsw@mellanox.com>,
	Ido Schimmel <idosch@mellanox.com>
Subject: Re: [RFC PATCH net-next 12/15] ipv4: Add "in hardware" indication to routes
Date: Thu, 3 Oct 2019 08:37:50 +0300	[thread overview]
Message-ID: <20191003053750.GC4325@splinter> (raw)
In-Reply-To: <1eea9e93-dbd9-8b50-9bf1-f8f6c6842dcc@gmail.com>

On Wed, Oct 02, 2019 at 08:34:22PM -0600, David Ahern wrote:
> On 10/2/19 12:21 PM, Jiri Pirko wrote:
> >>> This patch adds an "in hardware" indication to IPv4 routes, so that
> >>> users will have better visibility into the offload process. In the
> >>> future IPv6 will be extended with this indication as well.
> >>>
> >>> 'struct fib_alias' is extended with a new field that indicates if
> >>> the route resides in hardware or not. Note that the new field is added
> >>> in the 6 bytes hole and therefore the struct still fits in a single
> >>> cache line [1].
> >>>
> >>> Capable drivers are expected to invoke fib_alias_in_hw_{set,clear}()
> >>> with the route's key in order to set / clear the "in hardware
> >>> indication".
> >>>
> >>> The new indication is dumped to user space via a new flag (i.e.,
> >>> 'RTM_F_IN_HW') in the 'rtm_flags' field in the ancillary header.
> >>>
> >>
> >> nice series Ido. why not call this RTM_F_OFFLOAD to keep it consistent
> >> with the nexthop offload indication ?.
> > 
> > See the second paragraph of this description.
> 
> I read it multiple times. It does not explain why RTM_F_OFFLOAD is not
> used. Unless there is good reason RTM_F_OFFLOAD should be the name for
> consistency with all of the other OFFLOAD flags.

David, I'm not sure I understand the issue. You want the flag to be
called "RTM_F_OFFLOAD" to be consistent with "RTNH_F_OFFLOAD"? Are you
OK with iproute2 displaying it as "in_hw"? Displaying it as "offload" is
really wrong for the reasons I mentioned above. Host routes (for
example) do not offload anything from the kernel, they just reside in
hardware and trap packets...

The above is at least consistent with tc where we already have
"TCA_CLS_FLAGS_IN_HW".

> I realize rtm_flags is overloaded and the lower 8 bits contains RTNH_F
> flags, but that can be managed with good documentation - that RTNH_F
> is for the nexthop and RTM_F is for the prefix.

Are you talking about documenting the display strings in "ip-route" man
page or something else? If we stick with "offload" and "in_hw" then they
should probably be documented there to avoid confusion.