History

ipv6: Implement automatic flow label generation on transmit

Automatically generate flow labels for IPv6 packets on transmit. The flow label is computed based on skb_get_hash. The flow label will only automatically be set when it is zero otherwise (i.e. flow label manager hasn't set one). This supports the transmit side functionality of RFC 6438. Added an IPv6 sysctl auto_flowlabels to enable/disable this behavior system wide, and added IPV6_AUTOFLOWLABEL socket option to enable this functionality per socket. By default, auto flowlabels are disabled to avoid possible conflicts with flow label manager, however if this feature proves useful we may want to enable it by default. It should also be noted that FreeBSD has already implemented automatic flow labels (including the sysctl and socket option). In FreeBSD, automatic flow labels default to enabled. Performance impact: Running super_netperf with 200 flows for TCP_RR and UDP_RR for IPv6. Note that in UDP case, __skb_get_hash will be called for every packet with explains slight regression. In the TCP case the hash is saved in the socket so there is no regression. Automatic flow labels disabled: TCP_RR: 86.53% CPU utilization 127/195/322 90/95/99% latencies 1.40498e+06 tps UDP_RR: 90.70% CPU utilization 118/168/243 90/95/99% latencies 1.50309e+06 tps Automatic flow labels enabled: TCP_RR: 85.90% CPU utilization 128/199/337 90/95/99% latencies 1.40051e+06 UDP_RR 92.61% CPU utilization 115/164/236 90/95/99% latencies 1.4687e+06

Tom Herbert committed 12 years ago

Commit cb1ce2e1 parent 19469a811 files · +62 −6Browse files

Unified · Split

11 files
  1. Documentation/networking/ip-sysctl.txt +9 −0
  2. include/linux/ipv6.h +2 −1
  3. include/net/ipv6.h +20 −0
  4. include/net/netns/ipv6.h +1 −0
  5. include/uapi/linux/in6.h +1 −0
  6. net/ipv6/af_inet6.c +1 −0
  7. net/ipv6/ip6_gre.c +5 −2
  8. net/ipv6/ip6_output.c +5 −2
  9. net/ipv6/ip6_tunnel.c +2 −1
  10. net/ipv6/ipv6_sockglue.c +8 −0
  11. net/ipv6/sysctl_net_ipv6.c +8 −0
Documentation/networking/ip-sysctl.txt+9−0
@@ -1132,6 +1132,15 @@ flowlabel_consistency - BOOLEAN · 1131 unmodified lines
1132 FALSE: disabled1132 FALSE: disabled
1133 Default: TRUE1133 Default: TRUE
1134 1134
1135auto_flowlabels - BOOLEAN
1136 Automatically generate flow labels based based on a flow hash
1137 of the packet. This allows intermediate devices, such as routers,
1138 to idenfify packet flows for mechanisms like Equal Cost Multipath
1139 Routing (see RFC 6438).
1140 TRUE: enabled
1141 FALSE: disabled
1142 Default: false
1143
1135anycast_src_echo_reply - BOOLEAN1144anycast_src_echo_reply - BOOLEAN
1136 Controls the use of anycast addresses as source addresses for ICMPv61145 Controls the use of anycast addresses as source addresses for ICMPv6
1137 echo reply1146 echo reply
include/linux/ipv6.h+2−1
@@ -199,7 +199,8 @@ struct ipv6_pinfo { · 198 unmodified lines
199 * 010: prefer public address199 * 010: prefer public address
200 * 100: prefer care-of address200 * 100: prefer care-of address
201 */201 */
202 dontfrag:1;202 dontfrag:1,
203 autoflowlabel:1;
203 __u8 min_hopcount;204 __u8 min_hopcount;
204 __u8 tclass;205 __u8 tclass;
205 __be32 rcv_flowinfo;206 __be32 rcv_flowinfo;
include/net/ipv6.h+20−0
@@ -699,6 +699,26 @@ static inline void ip6_set_txhash(struct sock *sk) · 698 unmodified lines
699 sk->sk_txhash = flow_hash_from_keys(&keys);699 sk->sk_txhash = flow_hash_from_keys(&keys);
700}700}
701 701
702static inline __be32 ip6_make_flowlabel(struct net *net, struct sk_buff *skb,
703 __be32 flowlabel, bool autolabel)
704{
705 if (!flowlabel && (autolabel || net->ipv6.sysctl.auto_flowlabels)) {
706 __be32 hash;
707
708 hash = skb_get_hash(skb);
709
710 /* Since this is being sent on the wire obfuscate hash a bit
711 * to minimize possbility that any useful information to an
712 * attacker is leaked. Only lower 20 bits are relevant.
713 */
714 hash ^= hash >> 12;
715
716 flowlabel = hash & IPV6_FLOWLABEL_MASK;
717 }
718
719 return flowlabel;
720}
721
702/*722/*
703 * Header manipulation723 * Header manipulation
704 */724 */
include/net/netns/ipv6.h+1−0
@@ -28,6 +28,7 @@ struct netns_sysctl_ipv6 { · 27 unmodified lines
28 int ip6_rt_mtu_expires;28 int ip6_rt_mtu_expires;
29 int ip6_rt_min_advmss;29 int ip6_rt_min_advmss;
30 int flowlabel_consistency;30 int flowlabel_consistency;
31 int auto_flowlabels;
31 int icmpv6_time;32 int icmpv6_time;
32 int anycast_src_echo_reply;33 int anycast_src_echo_reply;
33 int fwmark_reflect;34 int fwmark_reflect;
include/uapi/linux/in6.h+1−0
@@ -233,6 +233,7 @@ struct in6_flowlabel_req { · 232 unmodified lines
233#if 0 /* not yet */233#if 0 /* not yet */
234#define IPV6_USE_MIN_MTU 63234#define IPV6_USE_MIN_MTU 63
235#endif235#endif
236#define IPV6_AUTOFLOWLABEL 64
236 237
237/*238/*
238 * Netfilter (1)239 * Netfilter (1)
net/ipv6/af_inet6.c+1−0
@@ -765,6 +765,7 @@ static int __net_init inet6_net_init(struct net *net) · 764 unmodified lines
765 net->ipv6.sysctl.bindv6only = 0;765 net->ipv6.sysctl.bindv6only = 0;
766 net->ipv6.sysctl.icmpv6_time = 1*HZ;766 net->ipv6.sysctl.icmpv6_time = 1*HZ;
767 net->ipv6.sysctl.flowlabel_consistency = 1;767 net->ipv6.sysctl.flowlabel_consistency = 1;
768 net->ipv6.sysctl.auto_flowlabels = 0;
768 atomic_set(&net->ipv6.rt_genid, 0);769 atomic_set(&net->ipv6.rt_genid, 0);
769 770
770 err = ipv6_init_mibs(net);771 err = ipv6_init_mibs(net);
net/ipv6/ip6_gre.c+5−2
@@ -723,7 +723,8 @@ static netdev_tx_t ip6gre_xmit2(struct sk_buff *skb, · 722 unmodified lines
723 * Push down and install the IP header.723 * Push down and install the IP header.
724 */724 */
725 ipv6h = ipv6_hdr(skb);725 ipv6h = ipv6_hdr(skb);
726 ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield), fl6->flowlabel);726 ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield),
727 ip6_make_flowlabel(net, skb, fl6->flowlabel, false));
727 ipv6h->hop_limit = tunnel->parms.hop_limit;728 ipv6h->hop_limit = tunnel->parms.hop_limit;
728 ipv6h->nexthdr = proto;729 ipv6h->nexthdr = proto;
729 ipv6h->saddr = fl6->saddr;730 ipv6h->saddr = fl6->saddr;
@@ -1174,7 +1175,9 @@ static int ip6gre_header(struct sk_buff *skb, struct net_device *dev, · 444 unmodified lines
1174 struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen);1175 struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen);
1175 __be16 *p = (__be16 *)(ipv6h+1);1176 __be16 *p = (__be16 *)(ipv6h+1);
1176 1177
1177 ip6_flow_hdr(ipv6h, 0, t->fl.u.ip6.flowlabel);1178 ip6_flow_hdr(ipv6h, 0,
1179 ip6_make_flowlabel(dev_net(dev), skb,
1180 t->fl.u.ip6.flowlabel, false));
1178 ipv6h->hop_limit = t->parms.hop_limit;1181 ipv6h->hop_limit = t->parms.hop_limit;
1179 ipv6h->nexthdr = NEXTHDR_GRE;1182 ipv6h->nexthdr = NEXTHDR_GRE;
1180 ipv6h->saddr = t->parms.laddr;1183 ipv6h->saddr = t->parms.laddr;
net/ipv6/ip6_output.c+5−2
@@ -205,7 +205,8 @@ int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6, · 204 unmodified lines
205 if (hlimit < 0)205 if (hlimit < 0)
206 hlimit = ip6_dst_hoplimit(dst);206 hlimit = ip6_dst_hoplimit(dst);
207 207
208 ip6_flow_hdr(hdr, tclass, fl6->flowlabel);208 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
209 np->autoflowlabel));
209 210
210 hdr->payload_len = htons(seg_len);211 hdr->payload_len = htons(seg_len);
211 hdr->nexthdr = proto;212 hdr->nexthdr = proto;
@@ -1569,7 +1570,9 @@ int ip6_push_pending_frames(struct sock *sk) · 1357 unmodified lines
1569 skb_reset_network_header(skb);1570 skb_reset_network_header(skb);
1570 hdr = ipv6_hdr(skb);1571 hdr = ipv6_hdr(skb);
1571 1572
1572 ip6_flow_hdr(hdr, np->cork.tclass, fl6->flowlabel);1573 ip6_flow_hdr(hdr, np->cork.tclass,
1574 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1575 np->autoflowlabel));
1573 hdr->hop_limit = np->cork.hop_limit;1576 hdr->hop_limit = np->cork.hop_limit;
1574 hdr->nexthdr = proto;1577 hdr->nexthdr = proto;
1575 hdr->saddr = fl6->saddr;1578 hdr->saddr = fl6->saddr;
net/ipv6/ip6_tunnel.c+2−1
@@ -1046,7 +1046,8 @@ static int ip6_tnl_xmit2(struct sk_buff *skb, · 1045 unmodified lines
1046 skb_push(skb, sizeof(struct ipv6hdr));1046 skb_push(skb, sizeof(struct ipv6hdr));
1047 skb_reset_network_header(skb);1047 skb_reset_network_header(skb);
1048 ipv6h = ipv6_hdr(skb);1048 ipv6h = ipv6_hdr(skb);
1049 ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield), fl6->flowlabel);1049 ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield),
1050 ip6_make_flowlabel(net, skb, fl6->flowlabel, false));
1050 ipv6h->hop_limit = t->parms.hop_limit;1051 ipv6h->hop_limit = t->parms.hop_limit;
1051 ipv6h->nexthdr = proto;1052 ipv6h->nexthdr = proto;
1052 ipv6h->saddr = fl6->saddr;1053 ipv6h->saddr = fl6->saddr;
net/ipv6/ipv6_sockglue.c+8−0
@@ -834,6 +834,10 @@ pref_skip_coa: · 833 unmodified lines
834 np->dontfrag = valbool;834 np->dontfrag = valbool;
835 retv = 0;835 retv = 0;
836 break;836 break;
837 case IPV6_AUTOFLOWLABEL:
838 np->autoflowlabel = valbool;
839 retv = 0;
840 break;
837 }841 }
838 842
839 release_sock(sk);843 release_sock(sk);
@@ -1273,6 +1277,10 @@ static int do_ipv6_getsockopt(struct sock *sk, int level, int optname, · 433 unmodified lines
1273 val = np->dontfrag;1277 val = np->dontfrag;
1274 break;1278 break;
1275 1279
1280 case IPV6_AUTOFLOWLABEL:
1281 val = np->autoflowlabel;
1282 break;
1283
1276 default:1284 default:
1277 return -ENOPROTOOPT;1285 return -ENOPROTOOPT;
1278 }1286 }
net/ipv6/sysctl_net_ipv6.c+8−0
@@ -38,6 +38,13 @@ static struct ctl_table ipv6_table_template[] = { · 37 unmodified lines
38 .mode = 0644,38 .mode = 0644,
39 .proc_handler = proc_dointvec39 .proc_handler = proc_dointvec
40 },40 },
41 {
42 .procname = "auto_flowlabels",
43 .data = &init_net.ipv6.sysctl.auto_flowlabels,
44 .maxlen = sizeof(int),
45 .mode = 0644,
46 .proc_handler = proc_dointvec
47 },
41 {48 {
42 .procname = "fwmark_reflect",49 .procname = "fwmark_reflect",
43 .data = &init_net.ipv6.sysctl.fwmark_reflect,50 .data = &init_net.ipv6.sysctl.fwmark_reflect,
@@ -74,6 +81,7 @@ static int __net_init ipv6_sysctl_net_init(struct net *net) · 30 unmodified lines
74 ipv6_table[0].data = &net->ipv6.sysctl.bindv6only;81 ipv6_table[0].data = &net->ipv6.sysctl.bindv6only;
75 ipv6_table[1].data = &net->ipv6.sysctl.anycast_src_echo_reply;82 ipv6_table[1].data = &net->ipv6.sysctl.anycast_src_echo_reply;
76 ipv6_table[2].data = &net->ipv6.sysctl.flowlabel_consistency;83 ipv6_table[2].data = &net->ipv6.sysctl.flowlabel_consistency;
84 ipv6_table[3].data = &net->ipv6.sysctl.auto_flowlabels;
77 85
78 ipv6_route_table = ipv6_route_sysctl_init(net);86 ipv6_route_table = ipv6_route_sysctl_init(net);
79 if (!ipv6_route_table)87 if (!ipv6_route_table)