On 9/29/22 9:21 AM, Florian Westphal wrote: > Guillaume Nault <gnault@xxxxxxxxxx> wrote: > > [ CC David Ahern ] > >> On Wed, Sep 28, 2022 at 01:39:08PM +0200, Florian Westphal wrote: >>> From: Phil Sutter <phil@xxxxxx> >>> >>> Analogous to commit b575b24b8eee3 ("netfilter: Fix rpfilter >>> dropping vrf packets by mistake") but for nftables fib expression: >>> Add special treatment of VRF devices so that typical reverse path >>> filtering via 'fib saddr . iif oif' expression works as expected. >>> >>> Fixes: f6d0cbcf09c50 ("netfilter: nf_tables: add fib expression") >>> Signed-off-by: Phil Sutter <phil@xxxxxx> >>> Signed-off-by: Florian Westphal <fw@xxxxxxxxx> >>> --- >>> net/ipv4/netfilter/nft_fib_ipv4.c | 3 +++ >>> net/ipv6/netfilter/nft_fib_ipv6.c | 6 +++++- >>> 2 files changed, 8 insertions(+), 1 deletion(-) >>> >>> diff --git a/net/ipv4/netfilter/nft_fib_ipv4.c b/net/ipv4/netfilter/nft_fib_ipv4.c >>> index b75cac69bd7e..7ade04ff972d 100644 >>> --- a/net/ipv4/netfilter/nft_fib_ipv4.c >>> +++ b/net/ipv4/netfilter/nft_fib_ipv4.c >>> @@ -83,6 +83,9 @@ void nft_fib4_eval(const struct nft_expr *expr, struct nft_regs *regs, >>> else >>> oif = NULL; >>> >>> + if (priv->flags & NFTA_FIB_F_IIF) >>> + fl4.flowi4_oif = l3mdev_master_ifindex_rcu(oif); >>> + >> >> Shouldn't we set .flowi4_l3mdev instead of .flowi4_oif? > > No idea. > db53cd3d88dc328dea2e968c9c8d3b4294a8a674 sets both. > rp_filter modules in iptables only set flowi(6)_oif. > > David, can you give advice on what the correct fix is? > > Then we could change all users in netfilter at once rather than the > current collection of random-looking guesses... Old usage is setting flow oif and it gets converted to the L3 device if there is one. The new usage is flow l3mdev, but I only updated places I new I was testing. ie.., use l3mdev in the flow struct if you have a VRF test case for it.