Message ID | 20190402231136.15156-1-mcroce@redhat.com |
---|---|
State | Changes Requested |
Delegated to: | David Miller |
Headers | show |
Series | [net] net: thunderx: don't allow jumbo frames with XDP | expand |
On Wed, 3 Apr 2019 01:11:36 +0200 Matteo Croce <mcroce@redhat.com> wrote: > The thunderx driver forbids to load an eBPF program if the MTU is higher > than 1500 bytes, but this can be circumvented by first loading the eBPF, > and then raising the MTU. > > XDP assumes that SKBs are linear and fit in a single page, this can lead > to undefined behaviours. > Fix this by limiting the MTU to 1500 bytes if an eBPF program is loaded. I find this 1500 bytes limit strange. While XDP does not support frames that is split over multiple pages, it does support larger frames, as long as it can fit within one (e.g 4K) page, minus XDP_PACKET_HEADROOM (256B) and have tail-room for skb_shared_info (320 bytes), which is 4096-256-320 = 3520 bytes. A quick look at this driver it seems you are limited to 2176 bytes (RCV_FRAG_LEN=1536+64+320 + 256) based on how the dma mapping is done. I think the drivers intent is to limit to 1536 bytes, but the DMA mapping area include more. It seems rather suboptimal that the skb_shared_info is included as part of the DMA mapping. > Fixes: 05c773f52b96e ("net: thunderx: Add basic XDP support") > Signed-off-by: Matteo Croce <mcroce@redhat.com> > --- > drivers/net/ethernet/cavium/thunder/nicvf_main.c | 7 +++++++ > 1 file changed, 7 insertions(+) > > diff --git a/drivers/net/ethernet/cavium/thunder/nicvf_main.c b/drivers/net/ethernet/cavium/thunder/nicvf_main.c > index aa2be4807191..fe1d7513f01d 100644 > --- a/drivers/net/ethernet/cavium/thunder/nicvf_main.c > +++ b/drivers/net/ethernet/cavium/thunder/nicvf_main.c > @@ -1578,6 +1578,13 @@ static int nicvf_change_mtu(struct net_device *netdev, int new_mtu) > struct nicvf *nic = netdev_priv(netdev); > int orig_mtu = netdev->mtu; > > + /* For now just support only the usual MTU sized frames */ > + if (nic->xdp_prog && new_mtu > 1500) { > + netdev_warn(netdev, "Jumbo frames not yet supported with XDP, current MTU %d.\n", > + netdev->mtu); > + return -EOPNOTSUPP; > + } > + > netdev->mtu = new_mtu; > > if (!netif_running(netdev))
From: Matteo Croce <mcroce@redhat.com> Date: Wed, 3 Apr 2019 01:11:36 +0200 > The thunderx driver forbids to load an eBPF program if the MTU is higher > than 1500 bytes, but this can be circumvented by first loading the eBPF, > and then raising the MTU. > > XDP assumes that SKBs are linear and fit in a single page, this can lead > to undefined behaviours. > Fix this by limiting the MTU to 1500 bytes if an eBPF program is loaded. > > Fixes: 05c773f52b96e ("net: thunderx: Add basic XDP support") > Signed-off-by: Matteo Croce <mcroce@redhat.com> Please respond to Jesper's feedback about your choice of a limit of 1500. Otherwise I will toss your patch.
On Fri, Apr 5, 2019 at 2:20 AM David Miller <davem@davemloft.net> wrote: > > From: Matteo Croce <mcroce@redhat.com> > Date: Wed, 3 Apr 2019 01:11:36 +0200 > > > The thunderx driver forbids to load an eBPF program if the MTU is > > higher than 1500 bytes, but this can be circumvented by first > > loading the eBPF, and then raising the MTU. > > > > XDP assumes that SKBs are linear and fit in a single page, this can > > lead to undefined behaviours. > > Fix this by limiting the MTU to 1500 bytes if an eBPF program is > > loaded. > > > > Fixes: 05c773f52b96e ("net: thunderx: Add basic XDP support") > > Signed-off-by: Matteo Croce <mcroce@redhat.com> > > Please respond to Jesper's feedback about your choice of a limit of > 1500. > > Otherwise I will toss your patch. Hi David ad Jesper, I didn't deliberately choose a limit of 1500, the limit is always set in nicvf_xdp_setup(): /* For now just support only the usual MTU sized frames */ if (prog && (dev->mtu > 1500)) { netdev_warn(dev, "Jumbo frames not yet supported with XDP... I just enforced the same limit in another code path which didn't do the check. If you think that 1500 is a bad value, and I'm sure you're right because there isn't room even for VLAN tagging, I will send a series like: - 1/2 sets the limit to a resonable value - 2/2 enforce the same limit in the two code paths Regards,
On Fri, Apr 5, 2019 at 2:51 AM Matteo Croce <mcroce@redhat.com> wrote: > > On Fri, Apr 5, 2019 at 2:20 AM David Miller <davem@davemloft.net> wrote: > > > > From: Matteo Croce <mcroce@redhat.com> > > Date: Wed, 3 Apr 2019 01:11:36 +0200 > > > > > The thunderx driver forbids to load an eBPF program if the MTU is > > > higher than 1500 bytes, but this can be circumvented by first > > > loading the eBPF, and then raising the MTU. > > > > > > XDP assumes that SKBs are linear and fit in a single page, this can > > > lead to undefined behaviours. > > > Fix this by limiting the MTU to 1500 bytes if an eBPF program is > > > loaded. > > > > > > Fixes: 05c773f52b96e ("net: thunderx: Add basic XDP support") > > > Signed-off-by: Matteo Croce <mcroce@redhat.com> > > > > Please respond to Jesper's feedback about your choice of a limit of > > 1500. > > > > Otherwise I will toss your patch. > > Hi David ad Jesper, > > I didn't deliberately choose a limit of 1500, the limit is always set > in nicvf_xdp_setup(): > > /* For now just support only the usual MTU sized frames */ > if (prog && (dev->mtu > 1500)) { > netdev_warn(dev, "Jumbo frames not yet supported with XDP... > > I just enforced the same limit in another code path which didn't do > the check. > If you think that 1500 is a bad value, and I'm sure you're right because > there isn't room even for VLAN tagging, I will send a series like: > - 1/2 sets the limit to a resonable value > - 2/2 enforce the same limit in the two code paths > > Regards, > -- > Matteo Croce > per aspera ad upstream Hi all, I did some tests and I've found that on this driver, the maximum allowed frame size with XDP is 1530. Frames bigger than 1530 are split around multiple pages, so the driver doesn't even run the bpf on them: /* For XDP, ignore pkts spanning multiple pages */ if (nic->xdp_prog && (cqe_rx->rb_cnt == 1)) { based on this test, I'll send a series with a proper MTU limit which should be, correct me if I'm wrong: 1530 - 14 (eth) - 4 (QinQ) = 1512 bytes. I subtract only the 4 bytes for the QinQ as the NETIF_F_VLAN_CHALLENGED_BIT flag is not set, and the first VLAN tag should not be counted. Regards, -- Matteo Croce per aspera ad upstream
On Fri, 5 Apr 2019 17:45:34 +0200 Matteo Croce <mcroce@redhat.com> wrote: > On Fri, Apr 5, 2019 at 2:51 AM Matteo Croce <mcroce@redhat.com> wrote: > > > > On Fri, Apr 5, 2019 at 2:20 AM David Miller <davem@davemloft.net> wrote: > > > > > > From: Matteo Croce <mcroce@redhat.com> > > > Date: Wed, 3 Apr 2019 01:11:36 +0200 > > > > > > > The thunderx driver forbids to load an eBPF program if the MTU is > > > > higher than 1500 bytes, but this can be circumvented by first > > > > loading the eBPF, and then raising the MTU. > > > > > > > > XDP assumes that SKBs are linear and fit in a single page, this can > > > > lead to undefined behaviours. > > > > Fix this by limiting the MTU to 1500 bytes if an eBPF program is > > > > loaded. > > > > > > > > Fixes: 05c773f52b96e ("net: thunderx: Add basic XDP support") > > > > Signed-off-by: Matteo Croce <mcroce@redhat.com> > > > > > > Please respond to Jesper's feedback about your choice of a limit of > > > 1500. > > > > > > Otherwise I will toss your patch. > > > > Hi David ad Jesper, > > > > I didn't deliberately choose a limit of 1500, the limit is always set > > in nicvf_xdp_setup(): > > > > /* For now just support only the usual MTU sized frames */ > > if (prog && (dev->mtu > 1500)) { > > netdev_warn(dev, "Jumbo frames not yet supported with XDP... > > > > I just enforced the same limit in another code path which didn't do > > the check. > > If you think that 1500 is a bad value, and I'm sure you're right because > > there isn't room even for VLAN tagging, I will send a series like: > > - 1/2 sets the limit to a resonable value > > - 2/2 enforce the same limit in the two code paths > > > > Regards, > > -- > > Matteo Croce > > per aspera ad upstream > > Hi all, > > I did some tests and I've found that on this driver, the maximum > allowed frame size with XDP is 1530. > Frames bigger than 1530 are split around multiple pages, so the driver > doesn't even run the bpf on them: > > /* For XDP, ignore pkts spanning multiple pages */ > if (nic->xdp_prog && (cqe_rx->rb_cnt == 1)) { > > based on this test, I'll send a series with a proper MTU limit which > should be, correct me if I'm wrong: 1530 - 14 (eth) - 4 (QinQ) = 1512 > bytes. > I subtract only the 4 bytes for the QinQ as the > NETIF_F_VLAN_CHALLENGED_BIT flag is not set, and the first VLAN tag > should not be counted. You *do* need to include the first VLAN tag in the calculation. I guess I didn't explain this clear enough on IRC. XDP cannot use VLAN-offloading. As we explain here[1] when running XDP, you need to disable VLAN-offloading (see cmd in [1]), because XDP need the VLAN header to be "inline" in the packet. XDP don't (yet) have access to reading info from the descriptor. [1] https://github.com/xdp-project/xdp-tutorial/tree/master/packet01-parsing#a-note-about-vlan-offloads
diff --git a/drivers/net/ethernet/cavium/thunder/nicvf_main.c b/drivers/net/ethernet/cavium/thunder/nicvf_main.c index aa2be4807191..fe1d7513f01d 100644 --- a/drivers/net/ethernet/cavium/thunder/nicvf_main.c +++ b/drivers/net/ethernet/cavium/thunder/nicvf_main.c @@ -1578,6 +1578,13 @@ static int nicvf_change_mtu(struct net_device *netdev, int new_mtu) struct nicvf *nic = netdev_priv(netdev); int orig_mtu = netdev->mtu; + /* For now just support only the usual MTU sized frames */ + if (nic->xdp_prog && new_mtu > 1500) { + netdev_warn(netdev, "Jumbo frames not yet supported with XDP, current MTU %d.\n", + netdev->mtu); + return -EOPNOTSUPP; + } + netdev->mtu = new_mtu; if (!netif_running(netdev))
The thunderx driver forbids to load an eBPF program if the MTU is higher than 1500 bytes, but this can be circumvented by first loading the eBPF, and then raising the MTU. XDP assumes that SKBs are linear and fit in a single page, this can lead to undefined behaviours. Fix this by limiting the MTU to 1500 bytes if an eBPF program is loaded. Fixes: 05c773f52b96e ("net: thunderx: Add basic XDP support") Signed-off-by: Matteo Croce <mcroce@redhat.com> --- drivers/net/ethernet/cavium/thunder/nicvf_main.c | 7 +++++++ 1 file changed, 7 insertions(+)