[ovs-dev] [PATCH v2 2/2] datapath lisp: Use ovs offload compat functionality.
Pravin B Shelar
pshelar at nicira.com
Fri Sep 6 18:16:23 UTC 2013
OVS already has compat functions to handle GSO packets.
Following patch get rid of GSO packet handling in lisp
and use ovs iptunnel_xmit() function for same.
CC: Lori Jakab <lojakab at cisco.com>
Signed-off-by: Pravin B Shelar <pshelar at nicira.com>
---
datapath/vport-lisp.c | 221 +++++++++++--------------------------------------
1 files changed, 48 insertions(+), 173 deletions(-)
diff --git a/datapath/vport-lisp.c b/datapath/vport-lisp.c
index 2cec115..b1581ef 100644
--- a/datapath/vport-lisp.c
+++ b/datapath/vport-lisp.c
@@ -34,6 +34,7 @@
#include <net/xfrm.h>
#include "datapath.h"
+#include "gso.h"
#include "vport.h"
/*
@@ -162,7 +163,7 @@ static __be64 instance_id_to_tunnel_id(__u8 *iid)
/* Compute source UDP port for outgoing packet.
* Currently we use the flow hash.
*/
-static u16 ovs_tnl_get_src_port(struct sk_buff *skb)
+static u16 get_src_port(struct sk_buff *skb)
{
int low;
int high;
@@ -177,8 +178,7 @@ static u16 ovs_tnl_get_src_port(struct sk_buff *skb)
}
static void lisp_build_header(const struct vport *vport,
- struct sk_buff *skb,
- int tunnel_hlen)
+ struct sk_buff *skb)
{
struct lisp_port *lisp_port = lisp_vport(vport);
struct udphdr *udph = udp_hdr(skb);
@@ -186,7 +186,7 @@ static void lisp_build_header(const struct vport *vport,
const struct ovs_key_ipv4_tunnel *tun_key = OVS_CB(skb)->tun_key;
udph->dest = lisp_port->dst_port;
- udph->source = htons(ovs_tnl_get_src_port(skb));
+ udph->source = htons(get_src_port(skb));
udph->check = 0;
udph->len = htons(skb->len - skb_transport_offset(skb));
@@ -376,90 +376,47 @@ error:
return ERR_PTR(err);
}
-static bool need_linearize(const struct sk_buff *skb)
+static void lisp_fix_segment(struct sk_buff *skb)
{
- int i;
-
- if (unlikely(skb_shinfo(skb)->frag_list))
- return true;
-
- /*
- * Generally speaking we should linearize if there are paged frags.
- * However, if all of the refcounts are 1 we know nobody else can
- * change them from underneath us and we can skip the linearization.
- */
- for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
- if (unlikely(page_count(skb_frag_page(&skb_shinfo(skb)->frags[i])) > 1))
- return true;
+ struct udphdr *udph = udp_hdr(skb);
- return false;
+ udph->len = htons(skb->len - skb_transport_offset(skb));
}
-static struct sk_buff *handle_offloads(struct sk_buff *skb)
+static void handle_offloads(struct sk_buff *skb)
{
- int err;
-
- if (skb_is_gso(skb)) {
- struct sk_buff *nskb;
- char cb[sizeof(skb->cb)];
-
- memcpy(cb, skb->cb, sizeof(cb));
-
- nskb = __skb_gso_segment(skb, 0, false);
- if (IS_ERR(nskb)) {
- err = PTR_ERR(nskb);
- goto error;
- }
-
- consume_skb(skb);
- skb = nskb;
- while (nskb) {
- memcpy(nskb->cb, cb, sizeof(cb));
- nskb = nskb->next;
- }
- } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
- /* Pages aren't locked and could change at any time.
- * If this happens after we compute the checksum, the
- * checksum will be wrong. We linearize now to avoid
- * this problem.
- */
- if (unlikely(need_linearize(skb))) {
- err = __skb_linearize(skb);
- if (unlikely(err))
- goto error;
- }
-
- err = skb_checksum_help(skb);
- if (unlikely(err))
- goto error;
- }
-
- skb->ip_summed = CHECKSUM_NONE;
- return skb;
-
-error:
- return ERR_PTR(err);
+ if (skb_is_gso(skb))
+ OVS_GSO_CB(skb)->fix_segment = lisp_fix_segment;
+ else if (skb->ip_summed != CHECKSUM_PARTIAL)
+ skb->ip_summed = CHECKSUM_NONE;
}
-static int ovs_tnl_send(struct vport *vport, struct sk_buff *skb,
- u8 ipproto, int tunnel_hlen,
- void (*build_header)(const struct vport *,
- struct sk_buff *,
- int tunnel_hlen))
+static int lisp_send(struct vport *vport, struct sk_buff *skb)
{
- int min_headroom;
+ struct net *net = ovs_dp_get_net(vport->dp);
+ int network_offset = skb_network_offset(skb);
struct rtable *rt;
+ int min_headroom;
__be32 saddr;
- int sent_len = 0;
+ __be16 df;
+ int sent_len;
int err;
- struct sk_buff *nskb;
+
+ if (unlikely(!OVS_CB(skb)->tun_key))
+ return -EINVAL;
+
+ if (skb->protocol != htons(ETH_P_IP) &&
+ skb->protocol != htons(ETH_P_IPV6)) {
+ kfree_skb(skb);
+ return 0;
+ }
/* Route lookup */
saddr = OVS_CB(skb)->tun_key->ipv4_src;
rt = find_route(ovs_dp_get_net(vport->dp),
&saddr,
OVS_CB(skb)->tun_key->ipv4_dst,
- ipproto,
+ IPPROTO_UDP,
OVS_CB(skb)->tun_key->ipv4_tos,
skb->mark);
if (IS_ERR(rt)) {
@@ -467,11 +424,8 @@ static int ovs_tnl_send(struct vport *vport, struct sk_buff *skb,
goto error;
}
- tunnel_hlen += sizeof(struct iphdr);
-
min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
- + tunnel_hlen
- + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
+ + sizeof(struct iphdr) + LISP_HLEN;
if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
int head_delta = SKB_DATA_ALIGN(min_headroom -
@@ -484,83 +438,27 @@ static int ovs_tnl_send(struct vport *vport, struct sk_buff *skb,
goto err_free_rt;
}
+ skb_pull(skb, network_offset);
+ skb_reset_mac_header(skb);
+ skb_reset_inner_headers(skb);
+
+ __skb_push(skb, LISP_HLEN);
+ skb_reset_transport_header(skb);
+
+ lisp_build_header(vport, skb);
+
/* Offloading */
- nskb = handle_offloads(skb);
- if (IS_ERR(nskb)) {
- err = PTR_ERR(nskb);
- goto err_free_rt;
- }
- skb = nskb;
-
- /* Reset SKB */
- nf_reset(skb);
- secpath_reset(skb);
- skb_dst_drop(skb);
- skb_clear_rxhash(skb);
-
- while (skb) {
- struct sk_buff *next_skb = skb->next;
- struct iphdr *iph;
- int frag_len;
-
- skb->next = NULL;
-
- if (vlan_tx_tag_present(skb)) {
- if (unlikely(!__vlan_put_tag(skb,
- skb->vlan_proto,
- vlan_tx_tag_get(skb))))
- goto next;
-
- vlan_set_tci(skb, 0);
- }
-
- frag_len = skb->len;
- skb_push(skb, tunnel_hlen);
- skb_reset_network_header(skb);
- skb_set_transport_header(skb, sizeof(struct iphdr));
-
- if (next_skb)
- skb_dst_set(skb, dst_clone(&rt_dst(rt)));
- else
- skb_dst_set(skb, &rt_dst(rt));
-
- /* Push Tunnel header. */
- build_header(vport, skb, tunnel_hlen);
-
- /* Push IP header. */
- iph = ip_hdr(skb);
- iph->version = 4;
- iph->ihl = sizeof(struct iphdr) >> 2;
- iph->protocol = ipproto;
- iph->daddr = OVS_CB(skb)->tun_key->ipv4_dst;
- iph->saddr = saddr;
- iph->tos = OVS_CB(skb)->tun_key->ipv4_tos;
- iph->ttl = OVS_CB(skb)->tun_key->ipv4_ttl;
- iph->frag_off = OVS_CB(skb)->tun_key->tun_flags &
+ handle_offloads(skb);
+ skb->local_df = 1;
+
+ df = OVS_CB(skb)->tun_key->tun_flags &
TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
- /*
- * Allow our local IP stack to fragment the outer packet even
- * if the DF bit is set as a last resort. We also need to
- * force selection of an IP ID here with __ip_select_ident(),
- * as ip_select_ident() assumes a proper ID is not needed when
- * when the DF bit is set.
- */
- skb->local_df = 1;
- __ip_select_ident(iph, skb_dst(skb), 0);
-
- memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
-
- err = ip_local_out(skb);
- if (unlikely(net_xmit_eval(err)))
- goto next;
-
- sent_len += frag_len;
-
-next:
- skb = next_skb;
- }
+ sent_len = iptunnel_xmit(net, rt, skb,
+ saddr, OVS_CB(skb)->tun_key->ipv4_dst,
+ IPPROTO_UDP, OVS_CB(skb)->tun_key->ipv4_tos,
+ OVS_CB(skb)->tun_key->ipv4_ttl, df);
- return sent_len;
+ return sent_len > 0 ? sent_len + network_offset : sent_len;
err_free_rt:
ip_rt_put(rt);
@@ -568,29 +466,6 @@ error:
return err;
}
-static int lisp_tnl_send(struct vport *vport, struct sk_buff *skb)
-{
- int tnl_len;
- int network_offset = skb_network_offset(skb);
-
- if (unlikely(!OVS_CB(skb)->tun_key))
- return -EINVAL;
-
- /* We only encapsulate IPv4 and IPv6 packets */
- switch (skb->protocol) {
- case htons(ETH_P_IP):
- case htons(ETH_P_IPV6):
- /* Pop off "inner" Ethernet header */
- skb_pull(skb, network_offset);
- tnl_len = ovs_tnl_send(vport, skb, IPPROTO_UDP,
- LISP_HLEN, lisp_build_header);
- return tnl_len > 0 ? tnl_len + network_offset : tnl_len;
- default:
- kfree_skb(skb);
- return 0;
- }
-}
-
static const char *lisp_get_name(const struct vport *vport)
{
struct lisp_port *lisp_port = lisp_vport(vport);
@@ -603,5 +478,5 @@ const struct vport_ops ovs_lisp_vport_ops = {
.destroy = lisp_tnl_destroy,
.get_name = lisp_get_name,
.get_options = lisp_get_options,
- .send = lisp_tnl_send,
+ .send = lisp_send,
};
--
1.7.1
More information about the dev
mailing list