On Wed, Jan 26, 2022 at 4:05 AM Kumar Kartikeya Dwivedi
[off-list ref] wrote:

On Wed, Jan 26, 2022 at 05:12:15PM IST, Lorenzo Bianconi wrote:

quoted

quoted

[ snip to focus on the API ]

quoted

+int br_fdb_find_port_from_ifindex(struct xdp_md *xdp_ctx,
+                           struct bpf_fdb_lookup *opt,
+                           u32 opt__sz)
+{
+ struct xdp_buff *ctx = (struct xdp_buff *)xdp_ctx;
+ struct net_bridge_port *port;
+ struct net_device *dev;
+ int ret = -ENODEV;
+
+ BUILD_BUG_ON(sizeof(struct bpf_fdb_lookup) != NF_BPF_FDB_OPTS_SZ);
+ if (!opt || opt__sz != sizeof(struct bpf_fdb_lookup))
+         return -ENODEV;

Why is the BUILD_BUG_ON needed? Or why is the NF_BPF_FDB_OPTS_SZ
constant even needed?

I added it to be symmetric with respect to ct counterpart

But the constant needs to be an enum, not a define, otherwise it will not be
emitted to BTF, I added it so that one could easily check the struct 'version'
(because sizeof is not relocated in BPF programs).

Without reading the rest of the thread, bpf_core_type_size(struct
bpf_fdb_lookup) would be a CO-RE-relocatable way to get the actual
size of the type in the kernel.

Yes, bpf_core_field_exists and would also work, but the size is fixed anyway and
we need to check it, so it felt better to give it a name and also make it
visible to BPF programs at the same time.

quoted

 [...]

--
Kartikeya