2 * Copyright (c) 2007-2014 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
19 #include <linux/uaccess.h>
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/if_ether.h>
23 #include <linux/if_vlan.h>
24 #include <net/llc_pdu.h>
25 #include <linux/kernel.h>
26 #include <linux/jhash.h>
27 #include <linux/jiffies.h>
28 #include <linux/llc.h>
29 #include <linux/module.h>
31 #include <linux/rcupdate.h>
32 #include <linux/if_arp.h>
34 #include <linux/ipv6.h>
35 #include <linux/mpls.h>
36 #include <linux/sctp.h>
37 #include <linux/smp.h>
38 #include <linux/tcp.h>
39 #include <linux/udp.h>
40 #include <linux/icmp.h>
41 #include <linux/icmpv6.h>
42 #include <linux/rculist.h>
44 #include <net/ip_tunnels.h>
47 #include <net/ndisc.h>
51 #include "flow_netlink.h"
52 #include "conntrack.h"
54 u64 ovs_flow_used_time(unsigned long flow_jiffies)
56 struct timespec cur_ts;
59 ktime_get_ts(&cur_ts);
60 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
61 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
62 cur_ts.tv_nsec / NSEC_PER_MSEC;
64 return cur_ms - idle_ms;
67 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
69 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
70 const struct sk_buff *skb)
72 struct flow_stats *stats;
73 int node = numa_node_id();
74 int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
76 stats = rcu_dereference(flow->stats[node]);
78 /* Check if already have node-specific stats. */
80 spin_lock(&stats->lock);
81 /* Mark if we write on the pre-allocated stats. */
82 if (node == 0 && unlikely(flow->stats_last_writer != node))
83 flow->stats_last_writer = node;
85 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
86 spin_lock(&stats->lock);
88 /* If the current NUMA-node is the only writer on the
89 * pre-allocated stats keep using them.
91 if (unlikely(flow->stats_last_writer != node)) {
92 /* A previous locker may have already allocated the
93 * stats, so we need to check again. If node-specific
94 * stats were already allocated, we update the pre-
95 * allocated stats as we have already locked them.
97 if (likely(flow->stats_last_writer != NUMA_NO_NODE)
98 && likely(!rcu_access_pointer(flow->stats[node]))) {
99 /* Try to allocate node-specific stats. */
100 struct flow_stats *new_stats;
103 kmem_cache_alloc_node(flow_stats_cache,
109 if (likely(new_stats)) {
110 new_stats->used = jiffies;
111 new_stats->packet_count = 1;
112 new_stats->byte_count = len;
113 new_stats->tcp_flags = tcp_flags;
114 spin_lock_init(&new_stats->lock);
116 rcu_assign_pointer(flow->stats[node],
121 flow->stats_last_writer = node;
125 stats->used = jiffies;
126 stats->packet_count++;
127 stats->byte_count += len;
128 stats->tcp_flags |= tcp_flags;
130 spin_unlock(&stats->lock);
133 /* Must be called with rcu_read_lock or ovs_mutex. */
134 void ovs_flow_stats_get(const struct sw_flow *flow,
135 struct ovs_flow_stats *ovs_stats,
136 unsigned long *used, __be16 *tcp_flags)
142 memset(ovs_stats, 0, sizeof(*ovs_stats));
144 for_each_node(node) {
145 struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[node]);
148 /* Local CPU may write on non-local stats, so we must
149 * block bottom-halves here.
151 spin_lock_bh(&stats->lock);
152 if (!*used || time_after(stats->used, *used))
154 *tcp_flags |= stats->tcp_flags;
155 ovs_stats->n_packets += stats->packet_count;
156 ovs_stats->n_bytes += stats->byte_count;
157 spin_unlock_bh(&stats->lock);
162 /* Called with ovs_mutex. */
163 void ovs_flow_stats_clear(struct sw_flow *flow)
167 for_each_node(node) {
168 struct flow_stats *stats = ovsl_dereference(flow->stats[node]);
171 spin_lock_bh(&stats->lock);
173 stats->packet_count = 0;
174 stats->byte_count = 0;
175 stats->tcp_flags = 0;
176 spin_unlock_bh(&stats->lock);
181 static int check_header(struct sk_buff *skb, int len)
183 if (unlikely(skb->len < len))
185 if (unlikely(!pskb_may_pull(skb, len)))
190 static bool arphdr_ok(struct sk_buff *skb)
192 return pskb_may_pull(skb, skb_network_offset(skb) +
193 sizeof(struct arp_eth_header));
196 static int check_iphdr(struct sk_buff *skb)
198 unsigned int nh_ofs = skb_network_offset(skb);
202 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
206 ip_len = ip_hdrlen(skb);
207 if (unlikely(ip_len < sizeof(struct iphdr) ||
208 skb->len < nh_ofs + ip_len))
211 skb_set_transport_header(skb, nh_ofs + ip_len);
215 static bool tcphdr_ok(struct sk_buff *skb)
217 int th_ofs = skb_transport_offset(skb);
220 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
223 tcp_len = tcp_hdrlen(skb);
224 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
225 skb->len < th_ofs + tcp_len))
231 static bool udphdr_ok(struct sk_buff *skb)
233 return pskb_may_pull(skb, skb_transport_offset(skb) +
234 sizeof(struct udphdr));
237 static bool sctphdr_ok(struct sk_buff *skb)
239 return pskb_may_pull(skb, skb_transport_offset(skb) +
240 sizeof(struct sctphdr));
243 static bool icmphdr_ok(struct sk_buff *skb)
245 return pskb_may_pull(skb, skb_transport_offset(skb) +
246 sizeof(struct icmphdr));
249 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
251 unsigned int nh_ofs = skb_network_offset(skb);
259 err = check_header(skb, nh_ofs + sizeof(*nh));
264 nexthdr = nh->nexthdr;
265 payload_ofs = (u8 *)(nh + 1) - skb->data;
267 key->ip.proto = NEXTHDR_NONE;
268 key->ip.tos = ipv6_get_dsfield(nh);
269 key->ip.ttl = nh->hop_limit;
270 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
271 key->ipv6.addr.src = nh->saddr;
272 key->ipv6.addr.dst = nh->daddr;
274 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
275 if (unlikely(payload_ofs < 0))
279 if (frag_off & htons(~0x7))
280 key->ip.frag = OVS_FRAG_TYPE_LATER;
282 key->ip.frag = OVS_FRAG_TYPE_FIRST;
284 key->ip.frag = OVS_FRAG_TYPE_NONE;
287 nh_len = payload_ofs - nh_ofs;
288 skb_set_transport_header(skb, nh_ofs + nh_len);
289 key->ip.proto = nexthdr;
293 static bool icmp6hdr_ok(struct sk_buff *skb)
295 return pskb_may_pull(skb, skb_transport_offset(skb) +
296 sizeof(struct icmp6hdr));
299 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
302 __be16 eth_type; /* ETH_P_8021Q */
305 struct qtag_prefix *qp;
307 if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
310 if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
314 qp = (struct qtag_prefix *) skb->data;
315 key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
316 __skb_pull(skb, sizeof(struct qtag_prefix));
321 static __be16 parse_ethertype(struct sk_buff *skb)
323 struct llc_snap_hdr {
324 u8 dsap; /* Always 0xAA */
325 u8 ssap; /* Always 0xAA */
330 struct llc_snap_hdr *llc;
333 proto = *(__be16 *) skb->data;
334 __skb_pull(skb, sizeof(__be16));
336 if (eth_proto_is_802_3(proto))
339 if (skb->len < sizeof(struct llc_snap_hdr))
340 return htons(ETH_P_802_2);
342 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
345 llc = (struct llc_snap_hdr *) skb->data;
346 if (llc->dsap != LLC_SAP_SNAP ||
347 llc->ssap != LLC_SAP_SNAP ||
348 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
349 return htons(ETH_P_802_2);
351 __skb_pull(skb, sizeof(struct llc_snap_hdr));
353 if (eth_proto_is_802_3(llc->ethertype))
354 return llc->ethertype;
356 return htons(ETH_P_802_2);
359 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
362 struct icmp6hdr *icmp = icmp6_hdr(skb);
364 /* The ICMPv6 type and code fields use the 16-bit transport port
365 * fields, so we need to store them in 16-bit network byte order.
367 key->tp.src = htons(icmp->icmp6_type);
368 key->tp.dst = htons(icmp->icmp6_code);
369 memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
371 if (icmp->icmp6_code == 0 &&
372 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
373 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
374 int icmp_len = skb->len - skb_transport_offset(skb);
378 /* In order to process neighbor discovery options, we need the
381 if (unlikely(icmp_len < sizeof(*nd)))
384 if (unlikely(skb_linearize(skb)))
387 nd = (struct nd_msg *)skb_transport_header(skb);
388 key->ipv6.nd.target = nd->target;
390 icmp_len -= sizeof(*nd);
392 while (icmp_len >= 8) {
393 struct nd_opt_hdr *nd_opt =
394 (struct nd_opt_hdr *)(nd->opt + offset);
395 int opt_len = nd_opt->nd_opt_len * 8;
397 if (unlikely(!opt_len || opt_len > icmp_len))
400 /* Store the link layer address if the appropriate
401 * option is provided. It is considered an error if
402 * the same link layer option is specified twice.
404 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
406 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
408 ether_addr_copy(key->ipv6.nd.sll,
409 &nd->opt[offset+sizeof(*nd_opt)]);
410 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
412 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
414 ether_addr_copy(key->ipv6.nd.tll,
415 &nd->opt[offset+sizeof(*nd_opt)]);
426 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
427 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
428 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
434 * key_extract - extracts a flow key from an Ethernet frame.
435 * @skb: sk_buff that contains the frame, with skb->data pointing to the
437 * @key: output flow key
439 * The caller must ensure that skb->len >= ETH_HLEN.
441 * Returns 0 if successful, otherwise a negative errno value.
443 * Initializes @skb header pointers as follows:
445 * - skb->mac_header: the Ethernet header.
447 * - skb->network_header: just past the Ethernet header, or just past the
448 * VLAN header, to the first byte of the Ethernet payload.
450 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
451 * on output, then just past the IP header, if one is present and
452 * of a correct length, otherwise the same as skb->network_header.
453 * For other key->eth.type values it is left untouched.
455 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
460 /* Flags are always used as part of stats */
463 skb_reset_mac_header(skb);
465 /* Link layer. We are guaranteed to have at least the 14 byte Ethernet
466 * header in the linear data area.
469 ether_addr_copy(key->eth.src, eth->h_source);
470 ether_addr_copy(key->eth.dst, eth->h_dest);
472 __skb_pull(skb, 2 * ETH_ALEN);
473 /* We are going to push all headers that we pull, so no need to
474 * update skb->csum here.
478 if (skb_vlan_tag_present(skb))
479 key->eth.tci = htons(skb->vlan_tci);
480 else if (eth->h_proto == htons(ETH_P_8021Q))
481 if (unlikely(parse_vlan(skb, key)))
484 key->eth.type = parse_ethertype(skb);
485 if (unlikely(key->eth.type == htons(0)))
488 skb_reset_network_header(skb);
489 skb_reset_mac_len(skb);
490 __skb_push(skb, skb->data - skb_mac_header(skb));
493 if (key->eth.type == htons(ETH_P_IP)) {
497 error = check_iphdr(skb);
498 if (unlikely(error)) {
499 memset(&key->ip, 0, sizeof(key->ip));
500 memset(&key->ipv4, 0, sizeof(key->ipv4));
501 if (error == -EINVAL) {
502 skb->transport_header = skb->network_header;
509 key->ipv4.addr.src = nh->saddr;
510 key->ipv4.addr.dst = nh->daddr;
512 key->ip.proto = nh->protocol;
513 key->ip.tos = nh->tos;
514 key->ip.ttl = nh->ttl;
516 offset = nh->frag_off & htons(IP_OFFSET);
518 key->ip.frag = OVS_FRAG_TYPE_LATER;
521 if (nh->frag_off & htons(IP_MF) ||
522 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
523 key->ip.frag = OVS_FRAG_TYPE_FIRST;
525 key->ip.frag = OVS_FRAG_TYPE_NONE;
527 /* Transport layer. */
528 if (key->ip.proto == IPPROTO_TCP) {
529 if (tcphdr_ok(skb)) {
530 struct tcphdr *tcp = tcp_hdr(skb);
531 key->tp.src = tcp->source;
532 key->tp.dst = tcp->dest;
533 key->tp.flags = TCP_FLAGS_BE16(tcp);
535 memset(&key->tp, 0, sizeof(key->tp));
538 } else if (key->ip.proto == IPPROTO_UDP) {
539 if (udphdr_ok(skb)) {
540 struct udphdr *udp = udp_hdr(skb);
541 key->tp.src = udp->source;
542 key->tp.dst = udp->dest;
544 memset(&key->tp, 0, sizeof(key->tp));
546 } else if (key->ip.proto == IPPROTO_SCTP) {
547 if (sctphdr_ok(skb)) {
548 struct sctphdr *sctp = sctp_hdr(skb);
549 key->tp.src = sctp->source;
550 key->tp.dst = sctp->dest;
552 memset(&key->tp, 0, sizeof(key->tp));
554 } else if (key->ip.proto == IPPROTO_ICMP) {
555 if (icmphdr_ok(skb)) {
556 struct icmphdr *icmp = icmp_hdr(skb);
557 /* The ICMP type and code fields use the 16-bit
558 * transport port fields, so we need to store
559 * them in 16-bit network byte order. */
560 key->tp.src = htons(icmp->type);
561 key->tp.dst = htons(icmp->code);
563 memset(&key->tp, 0, sizeof(key->tp));
567 } else if (key->eth.type == htons(ETH_P_ARP) ||
568 key->eth.type == htons(ETH_P_RARP)) {
569 struct arp_eth_header *arp;
570 bool arp_available = arphdr_ok(skb);
572 arp = (struct arp_eth_header *)skb_network_header(skb);
575 arp->ar_hrd == htons(ARPHRD_ETHER) &&
576 arp->ar_pro == htons(ETH_P_IP) &&
577 arp->ar_hln == ETH_ALEN &&
580 /* We only match on the lower 8 bits of the opcode. */
581 if (ntohs(arp->ar_op) <= 0xff)
582 key->ip.proto = ntohs(arp->ar_op);
586 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
587 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
588 ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
589 ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
591 memset(&key->ip, 0, sizeof(key->ip));
592 memset(&key->ipv4, 0, sizeof(key->ipv4));
594 } else if (eth_p_mpls(key->eth.type)) {
595 size_t stack_len = MPLS_HLEN;
597 /* In the presence of an MPLS label stack the end of the L2
598 * header and the beginning of the L3 header differ.
600 * Advance network_header to the beginning of the L3
601 * header. mac_len corresponds to the end of the L2 header.
606 error = check_header(skb, skb->mac_len + stack_len);
610 memcpy(&lse, skb_network_header(skb), MPLS_HLEN);
612 if (stack_len == MPLS_HLEN)
613 memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
615 skb_set_network_header(skb, skb->mac_len + stack_len);
616 if (lse & htonl(MPLS_LS_S_MASK))
619 stack_len += MPLS_HLEN;
621 } else if (key->eth.type == htons(ETH_P_IPV6)) {
622 int nh_len; /* IPv6 Header + Extensions */
624 nh_len = parse_ipv6hdr(skb, key);
625 if (unlikely(nh_len < 0)) {
626 memset(&key->ip, 0, sizeof(key->ip));
627 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
628 if (nh_len == -EINVAL) {
629 skb->transport_header = skb->network_header;
637 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
639 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
640 key->ip.frag = OVS_FRAG_TYPE_FIRST;
642 /* Transport layer. */
643 if (key->ip.proto == NEXTHDR_TCP) {
644 if (tcphdr_ok(skb)) {
645 struct tcphdr *tcp = tcp_hdr(skb);
646 key->tp.src = tcp->source;
647 key->tp.dst = tcp->dest;
648 key->tp.flags = TCP_FLAGS_BE16(tcp);
650 memset(&key->tp, 0, sizeof(key->tp));
652 } else if (key->ip.proto == NEXTHDR_UDP) {
653 if (udphdr_ok(skb)) {
654 struct udphdr *udp = udp_hdr(skb);
655 key->tp.src = udp->source;
656 key->tp.dst = udp->dest;
658 memset(&key->tp, 0, sizeof(key->tp));
660 } else if (key->ip.proto == NEXTHDR_SCTP) {
661 if (sctphdr_ok(skb)) {
662 struct sctphdr *sctp = sctp_hdr(skb);
663 key->tp.src = sctp->source;
664 key->tp.dst = sctp->dest;
666 memset(&key->tp, 0, sizeof(key->tp));
668 } else if (key->ip.proto == NEXTHDR_ICMP) {
669 if (icmp6hdr_ok(skb)) {
670 error = parse_icmpv6(skb, key, nh_len);
674 memset(&key->tp, 0, sizeof(key->tp));
681 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
683 return key_extract(skb, key);
686 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
687 struct sk_buff *skb, struct sw_flow_key *key)
689 /* Extract metadata from packet. */
691 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
693 if (tun_info->options) {
694 BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
696 > sizeof(key->tun_opts));
697 memcpy(TUN_METADATA_OPTS(key, tun_info->options_len),
698 tun_info->options, tun_info->options_len);
699 key->tun_opts_len = tun_info->options_len;
701 key->tun_opts_len = 0;
704 key->tun_opts_len = 0;
705 memset(&key->tun_key, 0, sizeof(key->tun_key));
708 key->phy.priority = skb->priority;
709 key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
710 key->phy.skb_mark = skb->mark;
711 ovs_ct_fill_key(skb, key);
712 key->ovs_flow_hash = 0;
715 return key_extract(skb, key);
718 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
720 struct sw_flow_key *key, bool log)
724 memset(key, 0, OVS_SW_FLOW_KEY_METADATA_SIZE);
726 /* Extract metadata from netlink attributes. */
727 err = ovs_nla_get_flow_metadata(net, attr, key, log);
731 return key_extract(skb, key);