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>
45 #include <net/ip_tunnels.h>
47 #include <net/ndisc.h>
49 u64 ovs_flow_used_time(unsigned long flow_jiffies)
51 struct timespec cur_ts;
54 ktime_get_ts(&cur_ts);
55 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
56 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
57 cur_ts.tv_nsec / NSEC_PER_MSEC;
59 return cur_ms - idle_ms;
62 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
64 void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb)
66 struct flow_stats *stats;
67 __be16 tcp_flags = flow->key.tp.flags;
68 int node = numa_node_id();
70 stats = rcu_dereference(flow->stats[node]);
72 /* Check if already have node-specific stats. */
74 spin_lock(&stats->lock);
75 /* Mark if we write on the pre-allocated stats. */
76 if (node == 0 && unlikely(flow->stats_last_writer != node))
77 flow->stats_last_writer = node;
79 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
80 spin_lock(&stats->lock);
82 /* If the current NUMA-node is the only writer on the
83 * pre-allocated stats keep using them.
85 if (unlikely(flow->stats_last_writer != node)) {
86 /* A previous locker may have already allocated the
87 * stats, so we need to check again. If node-specific
88 * stats were already allocated, we update the pre-
89 * allocated stats as we have already locked them.
91 if (likely(flow->stats_last_writer != NUMA_NO_NODE)
92 && likely(!rcu_dereference(flow->stats[node]))) {
93 /* Try to allocate node-specific stats. */
94 struct flow_stats *new_stats;
97 kmem_cache_alloc_node(flow_stats_cache,
101 if (likely(new_stats)) {
102 new_stats->used = jiffies;
103 new_stats->packet_count = 1;
104 new_stats->byte_count = skb->len;
105 new_stats->tcp_flags = tcp_flags;
106 spin_lock_init(&new_stats->lock);
108 rcu_assign_pointer(flow->stats[node],
113 flow->stats_last_writer = node;
117 stats->used = jiffies;
118 stats->packet_count++;
119 stats->byte_count += skb->len;
120 stats->tcp_flags |= tcp_flags;
122 spin_unlock(&stats->lock);
125 /* Called with ovs_mutex. */
126 void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
127 unsigned long *used, __be16 *tcp_flags)
133 memset(ovs_stats, 0, sizeof(*ovs_stats));
135 for_each_node(node) {
136 struct flow_stats *stats = ovsl_dereference(flow->stats[node]);
139 /* Local CPU may write on non-local stats, so we must
140 * block bottom-halves here.
142 spin_lock_bh(&stats->lock);
143 if (!*used || time_after(stats->used, *used))
145 *tcp_flags |= stats->tcp_flags;
146 ovs_stats->n_packets += stats->packet_count;
147 ovs_stats->n_bytes += stats->byte_count;
148 spin_unlock_bh(&stats->lock);
153 void ovs_flow_stats_clear(struct sw_flow *flow)
157 for_each_node(node) {
158 struct flow_stats *stats = rcu_dereference(flow->stats[node]);
161 spin_lock_bh(&stats->lock);
163 stats->packet_count = 0;
164 stats->byte_count = 0;
165 stats->tcp_flags = 0;
166 spin_unlock_bh(&stats->lock);
171 static int check_header(struct sk_buff *skb, int len)
173 if (unlikely(skb->len < len))
175 if (unlikely(!pskb_may_pull(skb, len)))
180 static bool arphdr_ok(struct sk_buff *skb)
182 return pskb_may_pull(skb, skb_network_offset(skb) +
183 sizeof(struct arp_eth_header));
186 static int check_iphdr(struct sk_buff *skb)
188 unsigned int nh_ofs = skb_network_offset(skb);
192 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
196 ip_len = ip_hdrlen(skb);
197 if (unlikely(ip_len < sizeof(struct iphdr) ||
198 skb->len < nh_ofs + ip_len))
201 skb_set_transport_header(skb, nh_ofs + ip_len);
205 static bool tcphdr_ok(struct sk_buff *skb)
207 int th_ofs = skb_transport_offset(skb);
210 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
213 tcp_len = tcp_hdrlen(skb);
214 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
215 skb->len < th_ofs + tcp_len))
221 static bool udphdr_ok(struct sk_buff *skb)
223 return pskb_may_pull(skb, skb_transport_offset(skb) +
224 sizeof(struct udphdr));
227 static bool sctphdr_ok(struct sk_buff *skb)
229 return pskb_may_pull(skb, skb_transport_offset(skb) +
230 sizeof(struct sctphdr));
233 static bool icmphdr_ok(struct sk_buff *skb)
235 return pskb_may_pull(skb, skb_transport_offset(skb) +
236 sizeof(struct icmphdr));
239 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
241 unsigned int nh_ofs = skb_network_offset(skb);
249 err = check_header(skb, nh_ofs + sizeof(*nh));
254 nexthdr = nh->nexthdr;
255 payload_ofs = (u8 *)(nh + 1) - skb->data;
257 key->ip.proto = NEXTHDR_NONE;
258 key->ip.tos = ipv6_get_dsfield(nh);
259 key->ip.ttl = nh->hop_limit;
260 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
261 key->ipv6.addr.src = nh->saddr;
262 key->ipv6.addr.dst = nh->daddr;
264 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
265 if (unlikely(payload_ofs < 0))
269 if (frag_off & htons(~0x7))
270 key->ip.frag = OVS_FRAG_TYPE_LATER;
272 key->ip.frag = OVS_FRAG_TYPE_FIRST;
275 nh_len = payload_ofs - nh_ofs;
276 skb_set_transport_header(skb, nh_ofs + nh_len);
277 key->ip.proto = nexthdr;
281 static bool icmp6hdr_ok(struct sk_buff *skb)
283 return pskb_may_pull(skb, skb_transport_offset(skb) +
284 sizeof(struct icmp6hdr));
287 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
290 __be16 eth_type; /* ETH_P_8021Q */
293 struct qtag_prefix *qp;
295 if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
298 if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
302 qp = (struct qtag_prefix *) skb->data;
303 key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
304 __skb_pull(skb, sizeof(struct qtag_prefix));
309 static __be16 parse_ethertype(struct sk_buff *skb)
311 struct llc_snap_hdr {
312 u8 dsap; /* Always 0xAA */
313 u8 ssap; /* Always 0xAA */
318 struct llc_snap_hdr *llc;
321 proto = *(__be16 *) skb->data;
322 __skb_pull(skb, sizeof(__be16));
324 if (ntohs(proto) >= ETH_P_802_3_MIN)
327 if (skb->len < sizeof(struct llc_snap_hdr))
328 return htons(ETH_P_802_2);
330 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
333 llc = (struct llc_snap_hdr *) skb->data;
334 if (llc->dsap != LLC_SAP_SNAP ||
335 llc->ssap != LLC_SAP_SNAP ||
336 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
337 return htons(ETH_P_802_2);
339 __skb_pull(skb, sizeof(struct llc_snap_hdr));
341 if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
342 return llc->ethertype;
344 return htons(ETH_P_802_2);
347 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
350 struct icmp6hdr *icmp = icmp6_hdr(skb);
352 /* The ICMPv6 type and code fields use the 16-bit transport port
353 * fields, so we need to store them in 16-bit network byte order.
355 key->tp.src = htons(icmp->icmp6_type);
356 key->tp.dst = htons(icmp->icmp6_code);
358 if (icmp->icmp6_code == 0 &&
359 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
360 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
361 int icmp_len = skb->len - skb_transport_offset(skb);
365 /* In order to process neighbor discovery options, we need the
368 if (unlikely(icmp_len < sizeof(*nd)))
371 if (unlikely(skb_linearize(skb)))
374 nd = (struct nd_msg *)skb_transport_header(skb);
375 key->ipv6.nd.target = nd->target;
377 icmp_len -= sizeof(*nd);
379 while (icmp_len >= 8) {
380 struct nd_opt_hdr *nd_opt =
381 (struct nd_opt_hdr *)(nd->opt + offset);
382 int opt_len = nd_opt->nd_opt_len * 8;
384 if (unlikely(!opt_len || opt_len > icmp_len))
387 /* Store the link layer address if the appropriate
388 * option is provided. It is considered an error if
389 * the same link layer option is specified twice.
391 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
393 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
395 ether_addr_copy(key->ipv6.nd.sll,
396 &nd->opt[offset+sizeof(*nd_opt)]);
397 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
399 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
401 ether_addr_copy(key->ipv6.nd.tll,
402 &nd->opt[offset+sizeof(*nd_opt)]);
413 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
414 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
415 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
421 * ovs_flow_extract - extracts a flow key from an Ethernet frame.
422 * @skb: sk_buff that contains the frame, with skb->data pointing to the
424 * @in_port: port number on which @skb was received.
425 * @key: output flow key
427 * The caller must ensure that skb->len >= ETH_HLEN.
429 * Returns 0 if successful, otherwise a negative errno value.
431 * Initializes @skb header pointers as follows:
433 * - skb->mac_header: the Ethernet header.
435 * - skb->network_header: just past the Ethernet header, or just past the
436 * VLAN header, to the first byte of the Ethernet payload.
438 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
439 * on output, then just past the IP header, if one is present and
440 * of a correct length, otherwise the same as skb->network_header.
441 * For other key->eth.type values it is left untouched.
443 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
448 memset(key, 0, sizeof(*key));
450 key->phy.priority = skb->priority;
451 if (OVS_CB(skb)->tun_key)
452 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
453 key->phy.in_port = in_port;
454 key->phy.skb_mark = skb->mark;
456 skb_reset_mac_header(skb);
458 /* Link layer. We are guaranteed to have at least the 14 byte Ethernet
459 * header in the linear data area.
462 ether_addr_copy(key->eth.src, eth->h_source);
463 ether_addr_copy(key->eth.dst, eth->h_dest);
465 __skb_pull(skb, 2 * ETH_ALEN);
466 /* We are going to push all headers that we pull, so no need to
467 * update skb->csum here.
470 if (vlan_tx_tag_present(skb))
471 key->eth.tci = htons(skb->vlan_tci);
472 else if (eth->h_proto == htons(ETH_P_8021Q))
473 if (unlikely(parse_vlan(skb, key)))
476 key->eth.type = parse_ethertype(skb);
477 if (unlikely(key->eth.type == htons(0)))
480 skb_reset_network_header(skb);
481 __skb_push(skb, skb->data - skb_mac_header(skb));
484 if (key->eth.type == htons(ETH_P_IP)) {
488 error = check_iphdr(skb);
489 if (unlikely(error)) {
490 if (error == -EINVAL) {
491 skb->transport_header = skb->network_header;
498 key->ipv4.addr.src = nh->saddr;
499 key->ipv4.addr.dst = nh->daddr;
501 key->ip.proto = nh->protocol;
502 key->ip.tos = nh->tos;
503 key->ip.ttl = nh->ttl;
505 offset = nh->frag_off & htons(IP_OFFSET);
507 key->ip.frag = OVS_FRAG_TYPE_LATER;
510 if (nh->frag_off & htons(IP_MF) ||
511 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
512 key->ip.frag = OVS_FRAG_TYPE_FIRST;
514 /* Transport layer. */
515 if (key->ip.proto == IPPROTO_TCP) {
516 if (tcphdr_ok(skb)) {
517 struct tcphdr *tcp = tcp_hdr(skb);
518 key->tp.src = tcp->source;
519 key->tp.dst = tcp->dest;
520 key->tp.flags = TCP_FLAGS_BE16(tcp);
522 } else if (key->ip.proto == IPPROTO_UDP) {
523 if (udphdr_ok(skb)) {
524 struct udphdr *udp = udp_hdr(skb);
525 key->tp.src = udp->source;
526 key->tp.dst = udp->dest;
528 } else if (key->ip.proto == IPPROTO_SCTP) {
529 if (sctphdr_ok(skb)) {
530 struct sctphdr *sctp = sctp_hdr(skb);
531 key->tp.src = sctp->source;
532 key->tp.dst = sctp->dest;
534 } else if (key->ip.proto == IPPROTO_ICMP) {
535 if (icmphdr_ok(skb)) {
536 struct icmphdr *icmp = icmp_hdr(skb);
537 /* The ICMP type and code fields use the 16-bit
538 * transport port fields, so we need to store
539 * them in 16-bit network byte order. */
540 key->tp.src = htons(icmp->type);
541 key->tp.dst = htons(icmp->code);
545 } else if ((key->eth.type == htons(ETH_P_ARP) ||
546 key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
547 struct arp_eth_header *arp;
549 arp = (struct arp_eth_header *)skb_network_header(skb);
551 if (arp->ar_hrd == htons(ARPHRD_ETHER)
552 && arp->ar_pro == htons(ETH_P_IP)
553 && arp->ar_hln == ETH_ALEN
554 && arp->ar_pln == 4) {
556 /* We only match on the lower 8 bits of the opcode. */
557 if (ntohs(arp->ar_op) <= 0xff)
558 key->ip.proto = ntohs(arp->ar_op);
559 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
560 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
561 ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
562 ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
564 } else if (key->eth.type == htons(ETH_P_IPV6)) {
565 int nh_len; /* IPv6 Header + Extensions */
567 nh_len = parse_ipv6hdr(skb, key);
568 if (unlikely(nh_len < 0)) {
569 if (nh_len == -EINVAL) {
570 skb->transport_header = skb->network_header;
578 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
580 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
581 key->ip.frag = OVS_FRAG_TYPE_FIRST;
583 /* Transport layer. */
584 if (key->ip.proto == NEXTHDR_TCP) {
585 if (tcphdr_ok(skb)) {
586 struct tcphdr *tcp = tcp_hdr(skb);
587 key->tp.src = tcp->source;
588 key->tp.dst = tcp->dest;
589 key->tp.flags = TCP_FLAGS_BE16(tcp);
591 } else if (key->ip.proto == NEXTHDR_UDP) {
592 if (udphdr_ok(skb)) {
593 struct udphdr *udp = udp_hdr(skb);
594 key->tp.src = udp->source;
595 key->tp.dst = udp->dest;
597 } else if (key->ip.proto == NEXTHDR_SCTP) {
598 if (sctphdr_ok(skb)) {
599 struct sctphdr *sctp = sctp_hdr(skb);
600 key->tp.src = sctp->source;
601 key->tp.dst = sctp->dest;
603 } else if (key->ip.proto == NEXTHDR_ICMP) {
604 if (icmp6hdr_ok(skb)) {
605 error = parse_icmpv6(skb, key, nh_len);