Hi
I report here the issue with the previous patch.
The code is now checking against params->tot_len but then it is still
using is_skb_forwardable.
Consider this case where I shrink the packet:
skb->len == 1520
dev->mtu == 1500
params->tot_len == 1480
So the incoming pkt has len 1520, and the out interface has mtu 1500.
In this case fragmentation is not needed because params->tot_len < dev->mtu.
However the code calls is_skb_forwardable and may return false because
skb->len > dev->mtu, resulting in BPF_FIB_LKUP_RET_FRAG_NEEDED.
What I propose is using params->tot_len only if provided, without
falling back to use is_skb_forwardable when provided.
Something like this:
if (params->tot_len > 0) {
if (params->tot_len > mtu)
rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
} else if (!is_skb_forwardable(dev, skb)) {
rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
}
However, doing so we are skipping more relaxed MTU checks inside
is_skb_forwardable, so I'm not sure about this.
Please comment
Il giorno ven 20 nov 2020 alle ore 09:26 Jesper Dangaard Brouer
<brouer@xxxxxxxxxx> ha scritto:
>
> On Wed, 18 Nov 2020 16:29:35 +0100
> Jesper Dangaard Brouer <brouer@xxxxxxxxxx> wrote:
>
> > BPF end-user on Cilium slack-channel (Carlo Carraro) wants to use
> > bpf_fib_lookup for doing MTU-check, but *prior* to extending packet size,
> > by adjusting fib_params 'tot_len' with the packet length plus the
> > expected encap size. (Just like the bpf_check_mtu helper supports). He
> > discovered that for SKB ctx the param->tot_len was not used, instead
> > skb->len was used (via MTU check in is_skb_forwardable()).
> >
> > Fix this by using fib_params 'tot_len' for MTU check. If not provided
> > (e.g. zero) then keep existing behaviour intact.
>
> Carlo pointed out (in slack) that the logic is not correctly
> implemented in this patch.
>
> I will send a V7.
>
>
> > Fixes: 4c79579b44b1 ("bpf: Change bpf_fib_lookup to return lookup status")
> > Reported-by: Carlo Carraro <colrack@xxxxxxxxx>
> > Signed-off-by: Jesper Dangaard Brouer <brouer@xxxxxxxxxx>
> > ---
> > net/core/filter.c | 12 +++++++++++-
> > 1 file changed, 11 insertions(+), 1 deletion(-)
> >
> > diff --git a/net/core/filter.c b/net/core/filter.c
> > index 1ee97fdeea64..ae1fe8e6069a 100644
> > --- a/net/core/filter.c
> > +++ b/net/core/filter.c
> > @@ -5567,10 +5567,20 @@ BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
> >
> > if (!rc) {
> > struct net_device *dev;
> > + u32 mtu;
> >
> > dev = dev_get_by_index_rcu(net, params->ifindex);
> > - if (!is_skb_forwardable(dev, skb))
> > + mtu = dev->mtu;
> > +
> > + /* Using tot_len for L3 MTU check if provided by user. Notice at
> > + * this TC cls_bpf level skb->len contains L2 size, but
> > + * is_skb_forwardable takes that into account.
> > + */
> > + if (params->tot_len > mtu) {
> > rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
> > + } else if (!is_skb_forwardable(dev, skb)) {
> > + rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
> > + }
> > }
> >
> > return rc;
>
> --
> Best regards,
> Jesper Dangaard Brouer
> MSc.CS, Principal Kernel Engineer at Red Hat
> LinkedIn: http://www.linkedin.com/in/brouer
>
