| Message ID | 1467274002-61390-3-git-send-email-pshelar@ovn.org |
|---|---|
| State | Superseded |
| Headers | show |
On Thu, Jun 30, 2016 at 1:06 AM, Pravin B Shelar <pshelar@ovn.org> wrote: > Following patch backports updated iptunnel pull function. > Also brings in following upstream fix: > > commit a09a4c8dd1ec7f830e1fb9e59eb72bddc965d168 > Author: Jesse Gross <jesse@kernel.org> > > tunnels: Remove encapsulation offloads on decap. > > If a packet is either locally encapsulated or processed through GRO > it is marked with the offloads that it requires. However, when it is > decapsulated these tunnel offload indications are not removed. This > means that if we receive an encapsulated TCP packet, aggregate it with > GRO, decapsulate, and retransmit the resulting frame on a NIC that does > not support encapsulation, we won't be able to take advantage of hardware > offloads even though it is just a simple TCP packet at this point. > > This fixes the problem by stripping off encapsulation offload indications > when packets are decapsulated. > > The performance impacts of this bug are significant. In a test where a > Geneve encapsulated TCP stream is sent to a hypervisor, GRO'ed, decapsulated, > and bridged to a VM performance is improved by 60% (5Gbps->8Gbps) as a > result of avoiding unnecessary segmentation at the VM tap interface. > > Reported-by: Ramu Ramamurthy <sramamur@linux.vnet.ibm.com> > Fixes: 68c33163 ("v4 GRE: Add TCP segmentation offload for GRE") > Signed-off-by: Jesse Gross <jesse@kernel.org> > Signed-off-by: David S. Miller <davem@davemloft.net> > > Signed-off-by: Pravin B Shelar <pshelar@ovn.org> Acked-by: Jesse Gross <jesse@kernel.org> Since this patch bumps up the version that we'll use our backports for, we'll need to remember that this effectively has a dependency on later patches in the series. Otherwise, we'll have a situation like with 2.4 where using the backport causes a regression in some cases.
diff --git a/acinclude.m4 b/acinclude.m4 index 65d41f0..263c31d 100644 --- a/acinclude.m4 +++ b/acinclude.m4 @@ -432,7 +432,8 @@ AC_DEFUN([OVS_CHECK_LINUX_COMPAT], [ OVS_GREP_IFELSE([$KSRC/include/net/inetpeer.h], [vif], [OVS_DEFINE([HAVE_INETPEER_VIF_SUPPORT])]) - OVS_GREP_IFELSE([$KSRC/include/net/dst_metadata.h], [metadata_dst]) + OVS_GREP_IFELSE([$KSRC/include/net/ip_tunnels.h], [iptunnel_pull_offloads], + [OVS_DEFINE([HAVE_METADATA_DST])]) OVS_GREP_IFELSE([$KSRC/include/linux/net.h], [sock_create_kern.*net], [OVS_DEFINE([HAVE_SOCK_CREATE_KERN_NET])]) diff --git a/datapath/linux/compat/include/net/ip_tunnels.h b/datapath/linux/compat/include/net/ip_tunnels.h index bf67ac4..26c9041 100644 --- a/datapath/linux/compat/include/net/ip_tunnels.h +++ b/datapath/linux/compat/include/net/ip_tunnels.h @@ -339,4 +339,19 @@ struct net *rpl_ip_tunnel_get_link_net(const struct net_device *dev); int rpl___ip_tunnel_change_mtu(struct net_device *dev, int new_mtu, bool strict); #endif +static inline int iptunnel_pull_offloads(struct sk_buff *skb) +{ + if (skb_is_gso(skb)) { + int err; + + err = skb_unclone(skb, GFP_ATOMIC); + if (unlikely(err)) + return err; + skb_shinfo(skb)->gso_type &= ~(NETIF_F_GSO_ENCAP_ALL >> + NETIF_F_GSO_SHIFT); + } + + skb->encapsulation = 0; + return 0; +} #endif /* __NET_IP_TUNNELS_H */ diff --git a/datapath/linux/compat/ip_tunnels_core.c b/datapath/linux/compat/ip_tunnels_core.c index c5fb2b7..ee4fadc 100644 --- a/datapath/linux/compat/ip_tunnels_core.c +++ b/datapath/linux/compat/ip_tunnels_core.c @@ -171,7 +171,7 @@ int rpl___iptunnel_pull_header(struct sk_buff *skb, int hdr_len, skb_set_queue_mapping(skb, 0); skb_scrub_packet(skb, xnet); - return 0; + return iptunnel_pull_offloads(skb); } EXPORT_SYMBOL_GPL(rpl___iptunnel_pull_header);
Following patch backports updated iptunnel pull function. Also brings in following upstream fix: commit a09a4c8dd1ec7f830e1fb9e59eb72bddc965d168 Author: Jesse Gross <jesse@kernel.org> tunnels: Remove encapsulation offloads on decap. If a packet is either locally encapsulated or processed through GRO it is marked with the offloads that it requires. However, when it is decapsulated these tunnel offload indications are not removed. This means that if we receive an encapsulated TCP packet, aggregate it with GRO, decapsulate, and retransmit the resulting frame on a NIC that does not support encapsulation, we won't be able to take advantage of hardware offloads even though it is just a simple TCP packet at this point. This fixes the problem by stripping off encapsulation offload indications when packets are decapsulated. The performance impacts of this bug are significant. In a test where a Geneve encapsulated TCP stream is sent to a hypervisor, GRO'ed, decapsulated, and bridged to a VM performance is improved by 60% (5Gbps->8Gbps) as a result of avoiding unnecessary segmentation at the VM tap interface. Reported-by: Ramu Ramamurthy <sramamur@linux.vnet.ibm.com> Fixes: 68c33163 ("v4 GRE: Add TCP segmentation offload for GRE") Signed-off-by: Jesse Gross <jesse@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Pravin B Shelar <pshelar@ovn.org> --- acinclude.m4 | 3 ++- datapath/linux/compat/include/net/ip_tunnels.h | 15 +++++++++++++++ datapath/linux/compat/ip_tunnels_core.c | 2 +- 3 files changed, 18 insertions(+), 2 deletions(-)