2 * Copyright (c) 2007-2013 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
21 #include <linux/uaccess.h>
22 #include <linux/netdevice.h>
23 #include <linux/etherdevice.h>
24 #include <linux/if_ether.h>
25 #include <linux/if_vlan.h>
26 #include <net/llc_pdu.h>
27 #include <linux/kernel.h>
28 #include <linux/jhash.h>
29 #include <linux/jiffies.h>
30 #include <linux/llc.h>
31 #include <linux/module.h>
33 #include <linux/rcupdate.h>
34 #include <linux/if_arp.h>
36 #include <linux/ipv6.h>
37 #include <linux/sctp.h>
38 #include <linux/smp.h>
39 #include <linux/tcp.h>
40 #include <linux/udp.h>
41 #include <linux/icmp.h>
42 #include <linux/icmpv6.h>
43 #include <linux/rculist.h>
46 #include <net/ndisc.h>
50 u64 ovs_flow_used_time(unsigned long flow_jiffies)
52 struct timespec cur_ts;
55 ktime_get_ts(&cur_ts);
56 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
57 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
58 cur_ts.tv_nsec / NSEC_PER_MSEC;
60 return cur_ms - idle_ms;
63 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
65 void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb)
67 struct flow_stats *stats;
70 if (!flow->stats.is_percpu)
71 stats = flow->stats.stat;
73 stats = this_cpu_ptr(flow->stats.cpu_stats);
75 if ((flow->key.eth.type == htons(ETH_P_IP) ||
76 flow->key.eth.type == htons(ETH_P_IPV6)) &&
77 flow->key.ip.frag != OVS_FRAG_TYPE_LATER &&
78 flow->key.ip.proto == IPPROTO_TCP &&
79 likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
80 tcp_flags = TCP_FLAGS_BE16(tcp_hdr(skb));
83 spin_lock(&stats->lock);
84 stats->used = jiffies;
85 stats->packet_count++;
86 stats->byte_count += skb->len;
87 stats->tcp_flags |= tcp_flags;
88 spin_unlock(&stats->lock);
91 static void stats_read(struct flow_stats *stats, bool lock_bh,
92 struct ovs_flow_stats *ovs_stats,
93 unsigned long *used, __be16 *tcp_flags)
96 spin_lock_bh(&stats->lock);
98 spin_lock(&stats->lock);
100 if (time_after(stats->used, *used))
102 *tcp_flags |= stats->tcp_flags;
103 ovs_stats->n_packets += stats->packet_count;
104 ovs_stats->n_bytes += stats->byte_count;
107 spin_unlock_bh(&stats->lock);
109 spin_unlock(&stats->lock);
112 void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
113 unsigned long *used, __be16 *tcp_flags)
119 memset(ovs_stats, 0, sizeof(*ovs_stats));
121 if (!flow->stats.is_percpu) {
122 stats_read(flow->stats.stat, true, ovs_stats, used, tcp_flags);
126 for_each_possible_cpu(cpu) {
127 struct flow_stats *stats;
130 stats = per_cpu_ptr(flow->stats.cpu_stats, cpu);
131 lock_bh = (cpu == cur_cpu);
132 stats_read(stats, lock_bh, ovs_stats, used, tcp_flags);
138 static void stats_reset(struct flow_stats *stats, bool lock_bh)
141 spin_lock_bh(&stats->lock);
143 spin_lock(&stats->lock);
146 stats->packet_count = 0;
147 stats->byte_count = 0;
148 stats->tcp_flags = 0;
151 spin_unlock_bh(&stats->lock);
153 spin_unlock(&stats->lock);
156 void ovs_flow_stats_clear(struct sw_flow *flow)
160 if (!flow->stats.is_percpu) {
161 stats_reset(flow->stats.stat, true);
165 for_each_possible_cpu(cpu) {
168 lock_bh = (cpu == cur_cpu);
169 stats_reset(per_cpu_ptr(flow->stats.cpu_stats, cpu), lock_bh);
175 static int check_header(struct sk_buff *skb, int len)
177 if (unlikely(skb->len < len))
179 if (unlikely(!pskb_may_pull(skb, len)))
184 static bool arphdr_ok(struct sk_buff *skb)
186 return pskb_may_pull(skb, skb_network_offset(skb) +
187 sizeof(struct arp_eth_header));
190 static int check_iphdr(struct sk_buff *skb)
192 unsigned int nh_ofs = skb_network_offset(skb);
196 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
200 ip_len = ip_hdrlen(skb);
201 if (unlikely(ip_len < sizeof(struct iphdr) ||
202 skb->len < nh_ofs + ip_len))
205 skb_set_transport_header(skb, nh_ofs + ip_len);
209 static bool tcphdr_ok(struct sk_buff *skb)
211 int th_ofs = skb_transport_offset(skb);
214 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
217 tcp_len = tcp_hdrlen(skb);
218 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
219 skb->len < th_ofs + tcp_len))
225 static bool udphdr_ok(struct sk_buff *skb)
227 return pskb_may_pull(skb, skb_transport_offset(skb) +
228 sizeof(struct udphdr));
231 static bool sctphdr_ok(struct sk_buff *skb)
233 return pskb_may_pull(skb, skb_transport_offset(skb) +
234 sizeof(struct sctphdr));
237 static bool icmphdr_ok(struct sk_buff *skb)
239 return pskb_may_pull(skb, skb_transport_offset(skb) +
240 sizeof(struct icmphdr));
243 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
245 unsigned int nh_ofs = skb_network_offset(skb);
253 err = check_header(skb, nh_ofs + sizeof(*nh));
258 nexthdr = nh->nexthdr;
259 payload_ofs = (u8 *)(nh + 1) - skb->data;
261 key->ip.proto = NEXTHDR_NONE;
262 key->ip.tos = ipv6_get_dsfield(nh);
263 key->ip.ttl = nh->hop_limit;
264 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
265 key->ipv6.addr.src = nh->saddr;
266 key->ipv6.addr.dst = nh->daddr;
268 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
269 if (unlikely(payload_ofs < 0))
273 if (frag_off & htons(~0x7))
274 key->ip.frag = OVS_FRAG_TYPE_LATER;
276 key->ip.frag = OVS_FRAG_TYPE_FIRST;
279 nh_len = payload_ofs - nh_ofs;
280 skb_set_transport_header(skb, nh_ofs + nh_len);
281 key->ip.proto = nexthdr;
285 static bool icmp6hdr_ok(struct sk_buff *skb)
287 return pskb_may_pull(skb, skb_transport_offset(skb) +
288 sizeof(struct icmp6hdr));
291 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
294 __be16 eth_type; /* ETH_P_8021Q */
297 struct qtag_prefix *qp;
299 if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
302 if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
306 qp = (struct qtag_prefix *) skb->data;
307 key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
308 __skb_pull(skb, sizeof(struct qtag_prefix));
313 static __be16 parse_ethertype(struct sk_buff *skb)
315 struct llc_snap_hdr {
316 u8 dsap; /* Always 0xAA */
317 u8 ssap; /* Always 0xAA */
322 struct llc_snap_hdr *llc;
325 proto = *(__be16 *) skb->data;
326 __skb_pull(skb, sizeof(__be16));
328 if (ntohs(proto) >= ETH_P_802_3_MIN)
331 if (skb->len < sizeof(struct llc_snap_hdr))
332 return htons(ETH_P_802_2);
334 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
337 llc = (struct llc_snap_hdr *) skb->data;
338 if (llc->dsap != LLC_SAP_SNAP ||
339 llc->ssap != LLC_SAP_SNAP ||
340 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
341 return htons(ETH_P_802_2);
343 __skb_pull(skb, sizeof(struct llc_snap_hdr));
345 if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
346 return llc->ethertype;
348 return htons(ETH_P_802_2);
351 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
354 struct icmp6hdr *icmp = icmp6_hdr(skb);
356 /* The ICMPv6 type and code fields use the 16-bit transport port
357 * fields, so we need to store them in 16-bit network byte order.
359 key->ipv6.tp.src = htons(icmp->icmp6_type);
360 key->ipv6.tp.dst = htons(icmp->icmp6_code);
362 if (icmp->icmp6_code == 0 &&
363 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
364 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
365 int icmp_len = skb->len - skb_transport_offset(skb);
369 /* In order to process neighbor discovery options, we need the
372 if (unlikely(icmp_len < sizeof(*nd)))
375 if (unlikely(skb_linearize(skb)))
378 nd = (struct nd_msg *)skb_transport_header(skb);
379 key->ipv6.nd.target = nd->target;
381 icmp_len -= sizeof(*nd);
383 while (icmp_len >= 8) {
384 struct nd_opt_hdr *nd_opt =
385 (struct nd_opt_hdr *)(nd->opt + offset);
386 int opt_len = nd_opt->nd_opt_len * 8;
388 if (unlikely(!opt_len || opt_len > icmp_len))
391 /* Store the link layer address if the appropriate
392 * option is provided. It is considered an error if
393 * the same link layer option is specified twice.
395 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
397 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
399 memcpy(key->ipv6.nd.sll,
400 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
401 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
403 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
405 memcpy(key->ipv6.nd.tll,
406 &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
417 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
418 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
419 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
425 * ovs_flow_extract - extracts a flow key from an Ethernet frame.
426 * @skb: sk_buff that contains the frame, with skb->data pointing to the
428 * @in_port: port number on which @skb was received.
429 * @key: output flow key
431 * The caller must ensure that skb->len >= ETH_HLEN.
433 * Returns 0 if successful, otherwise a negative errno value.
435 * Initializes @skb header pointers as follows:
437 * - skb->mac_header: the Ethernet header.
439 * - skb->network_header: just past the Ethernet header, or just past the
440 * VLAN header, to the first byte of the Ethernet payload.
442 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
443 * on output, then just past the IP header, if one is present and
444 * of a correct length, otherwise the same as skb->network_header.
445 * For other key->eth.type values it is left untouched.
447 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
452 memset(key, 0, sizeof(*key));
454 key->phy.priority = skb->priority;
455 if (OVS_CB(skb)->tun_key)
456 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
457 key->phy.in_port = in_port;
458 key->phy.skb_mark = skb->mark;
460 skb_reset_mac_header(skb);
462 /* Link layer. We are guaranteed to have at least the 14 byte Ethernet
463 * header in the linear data area.
466 memcpy(key->eth.src, eth->h_source, ETH_ALEN);
467 memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
469 __skb_pull(skb, 2 * ETH_ALEN);
470 /* We are going to push all headers that we pull, so no need to
471 * update skb->csum here. */
473 if (vlan_tx_tag_present(skb))
474 key->eth.tci = htons(vlan_get_tci(skb));
475 else if (eth->h_proto == htons(ETH_P_8021Q))
476 if (unlikely(parse_vlan(skb, key)))
479 key->eth.type = parse_ethertype(skb);
480 if (unlikely(key->eth.type == htons(0)))
483 skb_reset_network_header(skb);
484 __skb_push(skb, skb->data - skb_mac_header(skb));
487 if (key->eth.type == htons(ETH_P_IP)) {
491 error = check_iphdr(skb);
492 if (unlikely(error)) {
493 if (error == -EINVAL) {
494 skb->transport_header = skb->network_header;
501 key->ipv4.addr.src = nh->saddr;
502 key->ipv4.addr.dst = nh->daddr;
504 key->ip.proto = nh->protocol;
505 key->ip.tos = nh->tos;
506 key->ip.ttl = nh->ttl;
508 offset = nh->frag_off & htons(IP_OFFSET);
510 key->ip.frag = OVS_FRAG_TYPE_LATER;
513 if (nh->frag_off & htons(IP_MF) ||
514 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
515 key->ip.frag = OVS_FRAG_TYPE_FIRST;
517 /* Transport layer. */
518 if (key->ip.proto == IPPROTO_TCP) {
519 if (tcphdr_ok(skb)) {
520 struct tcphdr *tcp = tcp_hdr(skb);
521 key->ipv4.tp.src = tcp->source;
522 key->ipv4.tp.dst = tcp->dest;
523 key->ipv4.tp.flags = TCP_FLAGS_BE16(tcp);
525 } else if (key->ip.proto == IPPROTO_UDP) {
526 if (udphdr_ok(skb)) {
527 struct udphdr *udp = udp_hdr(skb);
528 key->ipv4.tp.src = udp->source;
529 key->ipv4.tp.dst = udp->dest;
531 } else if (key->ip.proto == IPPROTO_SCTP) {
532 if (sctphdr_ok(skb)) {
533 struct sctphdr *sctp = sctp_hdr(skb);
534 key->ipv4.tp.src = sctp->source;
535 key->ipv4.tp.dst = sctp->dest;
537 } else if (key->ip.proto == IPPROTO_ICMP) {
538 if (icmphdr_ok(skb)) {
539 struct icmphdr *icmp = icmp_hdr(skb);
540 /* The ICMP type and code fields use the 16-bit
541 * transport port fields, so we need to store
542 * them in 16-bit network byte order. */
543 key->ipv4.tp.src = htons(icmp->type);
544 key->ipv4.tp.dst = htons(icmp->code);
548 } else if ((key->eth.type == htons(ETH_P_ARP) ||
549 key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
550 struct arp_eth_header *arp;
552 arp = (struct arp_eth_header *)skb_network_header(skb);
554 if (arp->ar_hrd == htons(ARPHRD_ETHER)
555 && arp->ar_pro == htons(ETH_P_IP)
556 && arp->ar_hln == ETH_ALEN
557 && arp->ar_pln == 4) {
559 /* We only match on the lower 8 bits of the opcode. */
560 if (ntohs(arp->ar_op) <= 0xff)
561 key->ip.proto = ntohs(arp->ar_op);
562 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
563 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
564 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
565 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
567 } else if (key->eth.type == htons(ETH_P_IPV6)) {
568 int nh_len; /* IPv6 Header + Extensions */
570 nh_len = parse_ipv6hdr(skb, key);
571 if (unlikely(nh_len < 0)) {
572 if (nh_len == -EINVAL) {
573 skb->transport_header = skb->network_header;
581 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
583 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
584 key->ip.frag = OVS_FRAG_TYPE_FIRST;
586 /* Transport layer. */
587 if (key->ip.proto == NEXTHDR_TCP) {
588 if (tcphdr_ok(skb)) {
589 struct tcphdr *tcp = tcp_hdr(skb);
590 key->ipv6.tp.src = tcp->source;
591 key->ipv6.tp.dst = tcp->dest;
592 key->ipv6.tp.flags = TCP_FLAGS_BE16(tcp);
594 } else if (key->ip.proto == NEXTHDR_UDP) {
595 if (udphdr_ok(skb)) {
596 struct udphdr *udp = udp_hdr(skb);
597 key->ipv6.tp.src = udp->source;
598 key->ipv6.tp.dst = udp->dest;
600 } else if (key->ip.proto == NEXTHDR_SCTP) {
601 if (sctphdr_ok(skb)) {
602 struct sctphdr *sctp = sctp_hdr(skb);
603 key->ipv6.tp.src = sctp->source;
604 key->ipv6.tp.dst = sctp->dest;
606 } else if (key->ip.proto == NEXTHDR_ICMP) {
607 if (icmp6hdr_ok(skb)) {
608 error = parse_icmpv6(skb, key, nh_len);