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"
53 u64 ovs_flow_used_time(unsigned long flow_jiffies)
55 struct timespec cur_ts;
58 ktime_get_ts(&cur_ts);
59 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
60 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
61 cur_ts.tv_nsec / NSEC_PER_MSEC;
63 return cur_ms - idle_ms;
66 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
68 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
69 const struct sk_buff *skb)
71 struct flow_stats *stats;
72 int node = numa_node_id();
73 int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
75 stats = rcu_dereference(flow->stats[node]);
77 /* Check if already have node-specific stats. */
79 spin_lock(&stats->lock);
80 /* Mark if we write on the pre-allocated stats. */
81 if (node == 0 && unlikely(flow->stats_last_writer != node))
82 flow->stats_last_writer = node;
84 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
85 spin_lock(&stats->lock);
87 /* If the current NUMA-node is the only writer on the
88 * pre-allocated stats keep using them.
90 if (unlikely(flow->stats_last_writer != node)) {
91 /* A previous locker may have already allocated the
92 * stats, so we need to check again. If node-specific
93 * stats were already allocated, we update the pre-
94 * allocated stats as we have already locked them.
96 if (likely(flow->stats_last_writer != NUMA_NO_NODE)
97 && likely(!rcu_access_pointer(flow->stats[node]))) {
98 /* Try to allocate node-specific stats. */
99 struct flow_stats *new_stats;
102 kmem_cache_alloc_node(flow_stats_cache,
108 if (likely(new_stats)) {
109 new_stats->used = jiffies;
110 new_stats->packet_count = 1;
111 new_stats->byte_count = len;
112 new_stats->tcp_flags = tcp_flags;
113 spin_lock_init(&new_stats->lock);
115 rcu_assign_pointer(flow->stats[node],
120 flow->stats_last_writer = node;
124 stats->used = jiffies;
125 stats->packet_count++;
126 stats->byte_count += len;
127 stats->tcp_flags |= tcp_flags;
129 spin_unlock(&stats->lock);
132 /* Must be called with rcu_read_lock or ovs_mutex. */
133 void ovs_flow_stats_get(const struct sw_flow *flow,
134 struct ovs_flow_stats *ovs_stats,
135 unsigned long *used, __be16 *tcp_flags)
141 memset(ovs_stats, 0, sizeof(*ovs_stats));
143 for_each_node(node) {
144 struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[node]);
147 /* Local CPU may write on non-local stats, so we must
148 * block bottom-halves here.
150 spin_lock_bh(&stats->lock);
151 if (!*used || time_after(stats->used, *used))
153 *tcp_flags |= stats->tcp_flags;
154 ovs_stats->n_packets += stats->packet_count;
155 ovs_stats->n_bytes += stats->byte_count;
156 spin_unlock_bh(&stats->lock);
161 /* Called with ovs_mutex. */
162 void ovs_flow_stats_clear(struct sw_flow *flow)
166 for_each_node(node) {
167 struct flow_stats *stats = ovsl_dereference(flow->stats[node]);
170 spin_lock_bh(&stats->lock);
172 stats->packet_count = 0;
173 stats->byte_count = 0;
174 stats->tcp_flags = 0;
175 spin_unlock_bh(&stats->lock);
180 static int check_header(struct sk_buff *skb, int len)
182 if (unlikely(skb->len < len))
184 if (unlikely(!pskb_may_pull(skb, len)))
189 static bool arphdr_ok(struct sk_buff *skb)
191 return pskb_may_pull(skb, skb_network_offset(skb) +
192 sizeof(struct arp_eth_header));
195 static int check_iphdr(struct sk_buff *skb)
197 unsigned int nh_ofs = skb_network_offset(skb);
201 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
205 ip_len = ip_hdrlen(skb);
206 if (unlikely(ip_len < sizeof(struct iphdr) ||
207 skb->len < nh_ofs + ip_len))
210 skb_set_transport_header(skb, nh_ofs + ip_len);
214 static bool tcphdr_ok(struct sk_buff *skb)
216 int th_ofs = skb_transport_offset(skb);
219 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
222 tcp_len = tcp_hdrlen(skb);
223 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
224 skb->len < th_ofs + tcp_len))
230 static bool udphdr_ok(struct sk_buff *skb)
232 return pskb_may_pull(skb, skb_transport_offset(skb) +
233 sizeof(struct udphdr));
236 static bool sctphdr_ok(struct sk_buff *skb)
238 return pskb_may_pull(skb, skb_transport_offset(skb) +
239 sizeof(struct sctphdr));
242 static bool icmphdr_ok(struct sk_buff *skb)
244 return pskb_may_pull(skb, skb_transport_offset(skb) +
245 sizeof(struct icmphdr));
248 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
250 unsigned int nh_ofs = skb_network_offset(skb);
258 err = check_header(skb, nh_ofs + sizeof(*nh));
263 nexthdr = nh->nexthdr;
264 payload_ofs = (u8 *)(nh + 1) - skb->data;
266 key->ip.proto = NEXTHDR_NONE;
267 key->ip.tos = ipv6_get_dsfield(nh);
268 key->ip.ttl = nh->hop_limit;
269 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
270 key->ipv6.addr.src = nh->saddr;
271 key->ipv6.addr.dst = nh->daddr;
273 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
274 if (unlikely(payload_ofs < 0))
278 if (frag_off & htons(~0x7))
279 key->ip.frag = OVS_FRAG_TYPE_LATER;
281 key->ip.frag = OVS_FRAG_TYPE_FIRST;
283 key->ip.frag = OVS_FRAG_TYPE_NONE;
286 nh_len = payload_ofs - nh_ofs;
287 skb_set_transport_header(skb, nh_ofs + nh_len);
288 key->ip.proto = nexthdr;
292 static bool icmp6hdr_ok(struct sk_buff *skb)
294 return pskb_may_pull(skb, skb_transport_offset(skb) +
295 sizeof(struct icmp6hdr));
298 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
301 __be16 eth_type; /* ETH_P_8021Q */
304 struct qtag_prefix *qp;
306 if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
309 if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
313 qp = (struct qtag_prefix *) skb->data;
314 key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
315 __skb_pull(skb, sizeof(struct qtag_prefix));
320 static __be16 parse_ethertype(struct sk_buff *skb)
322 struct llc_snap_hdr {
323 u8 dsap; /* Always 0xAA */
324 u8 ssap; /* Always 0xAA */
329 struct llc_snap_hdr *llc;
332 proto = *(__be16 *) skb->data;
333 __skb_pull(skb, sizeof(__be16));
335 if (ntohs(proto) >= ETH_P_802_3_MIN)
338 if (skb->len < sizeof(struct llc_snap_hdr))
339 return htons(ETH_P_802_2);
341 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
344 llc = (struct llc_snap_hdr *) skb->data;
345 if (llc->dsap != LLC_SAP_SNAP ||
346 llc->ssap != LLC_SAP_SNAP ||
347 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
348 return htons(ETH_P_802_2);
350 __skb_pull(skb, sizeof(struct llc_snap_hdr));
352 if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
353 return llc->ethertype;
355 return htons(ETH_P_802_2);
358 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
361 struct icmp6hdr *icmp = icmp6_hdr(skb);
363 /* The ICMPv6 type and code fields use the 16-bit transport port
364 * fields, so we need to store them in 16-bit network byte order.
366 key->tp.src = htons(icmp->icmp6_type);
367 key->tp.dst = htons(icmp->icmp6_code);
368 memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
370 if (icmp->icmp6_code == 0 &&
371 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
372 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
373 int icmp_len = skb->len - skb_transport_offset(skb);
377 /* In order to process neighbor discovery options, we need the
380 if (unlikely(icmp_len < sizeof(*nd)))
383 if (unlikely(skb_linearize(skb)))
386 nd = (struct nd_msg *)skb_transport_header(skb);
387 key->ipv6.nd.target = nd->target;
389 icmp_len -= sizeof(*nd);
391 while (icmp_len >= 8) {
392 struct nd_opt_hdr *nd_opt =
393 (struct nd_opt_hdr *)(nd->opt + offset);
394 int opt_len = nd_opt->nd_opt_len * 8;
396 if (unlikely(!opt_len || opt_len > icmp_len))
399 /* Store the link layer address if the appropriate
400 * option is provided. It is considered an error if
401 * the same link layer option is specified twice.
403 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
405 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
407 ether_addr_copy(key->ipv6.nd.sll,
408 &nd->opt[offset+sizeof(*nd_opt)]);
409 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
411 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
413 ether_addr_copy(key->ipv6.nd.tll,
414 &nd->opt[offset+sizeof(*nd_opt)]);
425 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
426 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
427 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
433 * key_extract - extracts a flow key from an Ethernet frame.
434 * @skb: sk_buff that contains the frame, with skb->data pointing to the
436 * @key: output flow key
438 * The caller must ensure that skb->len >= ETH_HLEN.
440 * Returns 0 if successful, otherwise a negative errno value.
442 * Initializes @skb header pointers as follows:
444 * - skb->mac_header: the Ethernet header.
446 * - skb->network_header: just past the Ethernet header, or just past the
447 * VLAN header, to the first byte of the Ethernet payload.
449 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
450 * on output, then just past the IP header, if one is present and
451 * of a correct length, otherwise the same as skb->network_header.
452 * For other key->eth.type values it is left untouched.
454 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
459 /* Flags are always used as part of stats */
462 skb_reset_mac_header(skb);
464 /* Link layer. We are guaranteed to have at least the 14 byte Ethernet
465 * header in the linear data area.
468 ether_addr_copy(key->eth.src, eth->h_source);
469 ether_addr_copy(key->eth.dst, eth->h_dest);
471 __skb_pull(skb, 2 * ETH_ALEN);
472 /* We are going to push all headers that we pull, so no need to
473 * update skb->csum here.
477 if (skb_vlan_tag_present(skb))
478 key->eth.tci = htons(skb->vlan_tci);
479 else if (eth->h_proto == htons(ETH_P_8021Q))
480 if (unlikely(parse_vlan(skb, key)))
483 key->eth.type = parse_ethertype(skb);
484 if (unlikely(key->eth.type == htons(0)))
487 skb_reset_network_header(skb);
488 skb_reset_mac_len(skb);
489 __skb_push(skb, skb->data - skb_mac_header(skb));
492 if (key->eth.type == htons(ETH_P_IP)) {
496 error = check_iphdr(skb);
497 if (unlikely(error)) {
498 memset(&key->ip, 0, sizeof(key->ip));
499 memset(&key->ipv4, 0, sizeof(key->ipv4));
500 if (error == -EINVAL) {
501 skb->transport_header = skb->network_header;
508 key->ipv4.addr.src = nh->saddr;
509 key->ipv4.addr.dst = nh->daddr;
511 key->ip.proto = nh->protocol;
512 key->ip.tos = nh->tos;
513 key->ip.ttl = nh->ttl;
515 offset = nh->frag_off & htons(IP_OFFSET);
517 key->ip.frag = OVS_FRAG_TYPE_LATER;
520 if (nh->frag_off & htons(IP_MF) ||
521 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
522 key->ip.frag = OVS_FRAG_TYPE_FIRST;
524 key->ip.frag = OVS_FRAG_TYPE_NONE;
526 /* Transport layer. */
527 if (key->ip.proto == IPPROTO_TCP) {
528 if (tcphdr_ok(skb)) {
529 struct tcphdr *tcp = tcp_hdr(skb);
530 key->tp.src = tcp->source;
531 key->tp.dst = tcp->dest;
532 key->tp.flags = TCP_FLAGS_BE16(tcp);
534 memset(&key->tp, 0, sizeof(key->tp));
537 } else if (key->ip.proto == IPPROTO_UDP) {
538 if (udphdr_ok(skb)) {
539 struct udphdr *udp = udp_hdr(skb);
540 key->tp.src = udp->source;
541 key->tp.dst = udp->dest;
543 memset(&key->tp, 0, sizeof(key->tp));
545 } else if (key->ip.proto == IPPROTO_SCTP) {
546 if (sctphdr_ok(skb)) {
547 struct sctphdr *sctp = sctp_hdr(skb);
548 key->tp.src = sctp->source;
549 key->tp.dst = sctp->dest;
551 memset(&key->tp, 0, sizeof(key->tp));
553 } else if (key->ip.proto == IPPROTO_ICMP) {
554 if (icmphdr_ok(skb)) {
555 struct icmphdr *icmp = icmp_hdr(skb);
556 /* The ICMP type and code fields use the 16-bit
557 * transport port fields, so we need to store
558 * them in 16-bit network byte order. */
559 key->tp.src = htons(icmp->type);
560 key->tp.dst = htons(icmp->code);
562 memset(&key->tp, 0, sizeof(key->tp));
566 } else if (key->eth.type == htons(ETH_P_ARP) ||
567 key->eth.type == htons(ETH_P_RARP)) {
568 struct arp_eth_header *arp;
569 bool arp_available = arphdr_ok(skb);
571 arp = (struct arp_eth_header *)skb_network_header(skb);
574 arp->ar_hrd == htons(ARPHRD_ETHER) &&
575 arp->ar_pro == htons(ETH_P_IP) &&
576 arp->ar_hln == ETH_ALEN &&
579 /* We only match on the lower 8 bits of the opcode. */
580 if (ntohs(arp->ar_op) <= 0xff)
581 key->ip.proto = ntohs(arp->ar_op);
585 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
586 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
587 ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
588 ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
590 memset(&key->ip, 0, sizeof(key->ip));
591 memset(&key->ipv4, 0, sizeof(key->ipv4));
593 } else if (eth_p_mpls(key->eth.type)) {
594 size_t stack_len = MPLS_HLEN;
596 /* In the presence of an MPLS label stack the end of the L2
597 * header and the beginning of the L3 header differ.
599 * Advance network_header to the beginning of the L3
600 * header. mac_len corresponds to the end of the L2 header.
605 error = check_header(skb, skb->mac_len + stack_len);
609 memcpy(&lse, skb_network_header(skb), MPLS_HLEN);
611 if (stack_len == MPLS_HLEN)
612 memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
614 skb_set_network_header(skb, skb->mac_len + stack_len);
615 if (lse & htonl(MPLS_LS_S_MASK))
618 stack_len += MPLS_HLEN;
620 } else if (key->eth.type == htons(ETH_P_IPV6)) {
621 int nh_len; /* IPv6 Header + Extensions */
623 nh_len = parse_ipv6hdr(skb, key);
624 if (unlikely(nh_len < 0)) {
625 memset(&key->ip, 0, sizeof(key->ip));
626 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
627 if (nh_len == -EINVAL) {
628 skb->transport_header = skb->network_header;
636 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
638 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
639 key->ip.frag = OVS_FRAG_TYPE_FIRST;
641 /* Transport layer. */
642 if (key->ip.proto == NEXTHDR_TCP) {
643 if (tcphdr_ok(skb)) {
644 struct tcphdr *tcp = tcp_hdr(skb);
645 key->tp.src = tcp->source;
646 key->tp.dst = tcp->dest;
647 key->tp.flags = TCP_FLAGS_BE16(tcp);
649 memset(&key->tp, 0, sizeof(key->tp));
651 } else if (key->ip.proto == NEXTHDR_UDP) {
652 if (udphdr_ok(skb)) {
653 struct udphdr *udp = udp_hdr(skb);
654 key->tp.src = udp->source;
655 key->tp.dst = udp->dest;
657 memset(&key->tp, 0, sizeof(key->tp));
659 } else if (key->ip.proto == NEXTHDR_SCTP) {
660 if (sctphdr_ok(skb)) {
661 struct sctphdr *sctp = sctp_hdr(skb);
662 key->tp.src = sctp->source;
663 key->tp.dst = sctp->dest;
665 memset(&key->tp, 0, sizeof(key->tp));
667 } else if (key->ip.proto == NEXTHDR_ICMP) {
668 if (icmp6hdr_ok(skb)) {
669 error = parse_icmpv6(skb, key, nh_len);
673 memset(&key->tp, 0, sizeof(key->tp));
680 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
682 return key_extract(skb, key);
685 int ovs_flow_key_extract(const struct ovs_tunnel_info *tun_info,
686 struct sk_buff *skb, struct sw_flow_key *key)
688 /* Extract metadata from packet. */
690 memcpy(&key->tun_key, &tun_info->tunnel, sizeof(key->tun_key));
692 if (tun_info->options) {
693 BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
695 > sizeof(key->tun_opts));
696 memcpy(TUN_METADATA_OPTS(key, tun_info->options_len),
697 tun_info->options, tun_info->options_len);
698 key->tun_opts_len = tun_info->options_len;
700 key->tun_opts_len = 0;
703 key->tun_opts_len = 0;
704 memset(&key->tun_key, 0, sizeof(key->tun_key));
707 key->phy.priority = skb->priority;
708 key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
709 key->phy.skb_mark = skb->mark;
710 key->ovs_flow_hash = 0;
713 return key_extract(skb, key);
716 int ovs_flow_key_extract_userspace(const struct nlattr *attr,
718 struct sw_flow_key *key, bool log)
722 memset(key, 0, OVS_SW_FLOW_KEY_METADATA_SIZE);
724 /* Extract metadata from netlink attributes. */
725 err = ovs_nla_get_flow_metadata(attr, key, log);
729 return key_extract(skb, key);