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
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>
51 u64 ovs_flow_used_time(unsigned long flow_jiffies)
53 struct timespec cur_ts;
56 ktime_get_ts(&cur_ts);
57 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
58 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
59 cur_ts.tv_nsec / NSEC_PER_MSEC;
61 return cur_ms - idle_ms;
64 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
66 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
69 struct flow_stats *stats;
70 int node = numa_node_id();
72 stats = rcu_dereference(flow->stats[node]);
74 /* Check if already have node-specific stats. */
76 spin_lock(&stats->lock);
77 /* Mark if we write on the pre-allocated stats. */
78 if (node == 0 && unlikely(flow->stats_last_writer != node))
79 flow->stats_last_writer = node;
81 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
82 spin_lock(&stats->lock);
84 /* If the current NUMA-node is the only writer on the
85 * pre-allocated stats keep using them.
87 if (unlikely(flow->stats_last_writer != node)) {
88 /* A previous locker may have already allocated the
89 * stats, so we need to check again. If node-specific
90 * stats were already allocated, we update the pre-
91 * allocated stats as we have already locked them.
93 if (likely(flow->stats_last_writer != NUMA_NO_NODE)
94 && likely(!rcu_dereference(flow->stats[node]))) {
95 /* Try to allocate node-specific stats. */
96 struct flow_stats *new_stats;
99 kmem_cache_alloc_node(flow_stats_cache,
103 if (likely(new_stats)) {
104 new_stats->used = jiffies;
105 new_stats->packet_count = 1;
106 new_stats->byte_count = skb->len;
107 new_stats->tcp_flags = tcp_flags;
108 spin_lock_init(&new_stats->lock);
110 rcu_assign_pointer(flow->stats[node],
115 flow->stats_last_writer = node;
119 stats->used = jiffies;
120 stats->packet_count++;
121 stats->byte_count += skb->len;
122 stats->tcp_flags |= tcp_flags;
124 spin_unlock(&stats->lock);
127 /* Must be called with rcu_read_lock or ovs_mutex. */
128 void ovs_flow_stats_get(const struct sw_flow *flow,
129 struct ovs_flow_stats *ovs_stats,
130 unsigned long *used, __be16 *tcp_flags)
136 memset(ovs_stats, 0, sizeof(*ovs_stats));
138 for_each_node(node) {
139 struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[node]);
142 /* Local CPU may write on non-local stats, so we must
143 * block bottom-halves here.
145 spin_lock_bh(&stats->lock);
146 if (!*used || time_after(stats->used, *used))
148 *tcp_flags |= stats->tcp_flags;
149 ovs_stats->n_packets += stats->packet_count;
150 ovs_stats->n_bytes += stats->byte_count;
151 spin_unlock_bh(&stats->lock);
156 /* Called with ovs_mutex. */
157 void ovs_flow_stats_clear(struct sw_flow *flow)
161 for_each_node(node) {
162 struct flow_stats *stats = ovsl_dereference(flow->stats[node]);
165 spin_lock_bh(&stats->lock);
167 stats->packet_count = 0;
168 stats->byte_count = 0;
169 stats->tcp_flags = 0;
170 spin_unlock_bh(&stats->lock);
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;
278 key->ip.frag = OVS_FRAG_TYPE_NONE;
281 nh_len = payload_ofs - nh_ofs;
282 skb_set_transport_header(skb, nh_ofs + nh_len);
283 key->ip.proto = nexthdr;
287 static bool icmp6hdr_ok(struct sk_buff *skb)
289 return pskb_may_pull(skb, skb_transport_offset(skb) +
290 sizeof(struct icmp6hdr));
293 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
296 __be16 eth_type; /* ETH_P_8021Q */
299 struct qtag_prefix *qp;
301 if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
304 if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
308 qp = (struct qtag_prefix *) skb->data;
309 key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
310 __skb_pull(skb, sizeof(struct qtag_prefix));
315 static __be16 parse_ethertype(struct sk_buff *skb)
317 struct llc_snap_hdr {
318 u8 dsap; /* Always 0xAA */
319 u8 ssap; /* Always 0xAA */
324 struct llc_snap_hdr *llc;
327 proto = *(__be16 *) skb->data;
328 __skb_pull(skb, sizeof(__be16));
330 if (ntohs(proto) >= ETH_P_802_3_MIN)
333 if (skb->len < sizeof(struct llc_snap_hdr))
334 return htons(ETH_P_802_2);
336 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
339 llc = (struct llc_snap_hdr *) skb->data;
340 if (llc->dsap != LLC_SAP_SNAP ||
341 llc->ssap != LLC_SAP_SNAP ||
342 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
343 return htons(ETH_P_802_2);
345 __skb_pull(skb, sizeof(struct llc_snap_hdr));
347 if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
348 return llc->ethertype;
350 return htons(ETH_P_802_2);
353 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
356 struct icmp6hdr *icmp = icmp6_hdr(skb);
358 /* The ICMPv6 type and code fields use the 16-bit transport port
359 * fields, so we need to store them in 16-bit network byte order.
361 key->tp.src = htons(icmp->icmp6_type);
362 key->tp.dst = htons(icmp->icmp6_code);
363 memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
365 if (icmp->icmp6_code == 0 &&
366 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
367 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
368 int icmp_len = skb->len - skb_transport_offset(skb);
372 /* In order to process neighbor discovery options, we need the
375 if (unlikely(icmp_len < sizeof(*nd)))
378 if (unlikely(skb_linearize(skb)))
381 nd = (struct nd_msg *)skb_transport_header(skb);
382 key->ipv6.nd.target = nd->target;
384 icmp_len -= sizeof(*nd);
386 while (icmp_len >= 8) {
387 struct nd_opt_hdr *nd_opt =
388 (struct nd_opt_hdr *)(nd->opt + offset);
389 int opt_len = nd_opt->nd_opt_len * 8;
391 if (unlikely(!opt_len || opt_len > icmp_len))
394 /* Store the link layer address if the appropriate
395 * option is provided. It is considered an error if
396 * the same link layer option is specified twice.
398 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
400 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
402 ether_addr_copy(key->ipv6.nd.sll,
403 &nd->opt[offset+sizeof(*nd_opt)]);
404 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
406 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
408 ether_addr_copy(key->ipv6.nd.tll,
409 &nd->opt[offset+sizeof(*nd_opt)]);
420 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
421 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
422 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
428 * ovs_flow_extract - extracts a flow key from an Ethernet frame.
429 * @skb: sk_buff that contains the frame, with skb->data pointing to the
431 * @in_port: port number on which @skb was received.
432 * @key: output flow key
434 * The caller must ensure that skb->len >= ETH_HLEN.
436 * Returns 0 if successful, otherwise a negative errno value.
438 * Initializes @skb header pointers as follows:
440 * - skb->mac_header: the Ethernet header.
442 * - skb->network_header: just past the Ethernet header, or just past the
443 * VLAN header, to the first byte of the Ethernet payload.
445 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
446 * on output, then just past the IP header, if one is present and
447 * of a correct length, otherwise the same as skb->network_header.
448 * For other key->eth.type values it is left untouched.
450 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
455 if (OVS_CB(skb)->tun_info) {
456 struct ovs_tunnel_info *tun_info = OVS_CB(skb)->tun_info;
457 memcpy(&key->tun_key, &tun_info->tunnel,
458 sizeof(key->tun_key));
459 if (tun_info->options) {
460 BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) * 8)) - 1
461 > sizeof(key->tun_opts));
462 memcpy(GENEVE_OPTS(key, tun_info->options_len),
463 tun_info->options, tun_info->options_len);
464 key->tun_opts_len = tun_info->options_len;
466 key->tun_opts_len = 0;
469 key->tun_opts_len = 0;
470 memset(&key->tun_key, 0, sizeof(key->tun_key));
473 key->phy.priority = skb->priority;
474 key->phy.in_port = in_port;
475 key->phy.skb_mark = skb->mark;
476 key->ovs_flow_hash = 0;
479 /* Flags are always used as part of stats. */
482 skb_reset_mac_header(skb);
484 /* Link layer. We are guaranteed to have at least the 14 byte Ethernet
485 * header in the linear data area.
488 ether_addr_copy(key->eth.src, eth->h_source);
489 ether_addr_copy(key->eth.dst, eth->h_dest);
491 __skb_pull(skb, 2 * ETH_ALEN);
492 /* We are going to push all headers that we pull, so no need to
493 * update skb->csum here. */
496 if (vlan_tx_tag_present(skb))
497 key->eth.tci = htons(vlan_get_tci(skb));
498 else if (eth->h_proto == htons(ETH_P_8021Q))
499 if (unlikely(parse_vlan(skb, key)))
502 key->eth.type = parse_ethertype(skb);
503 if (unlikely(key->eth.type == htons(0)))
506 skb_reset_network_header(skb);
507 skb_reset_mac_len(skb);
508 __skb_push(skb, skb->data - skb_mac_header(skb));
511 if (key->eth.type == htons(ETH_P_IP)) {
515 error = check_iphdr(skb);
516 if (unlikely(error)) {
517 memset(&key->ip, 0, sizeof(key->ip));
518 memset(&key->ipv4, 0, sizeof(key->ipv4));
519 if (error == -EINVAL) {
520 skb->transport_header = skb->network_header;
527 key->ipv4.addr.src = nh->saddr;
528 key->ipv4.addr.dst = nh->daddr;
530 key->ip.proto = nh->protocol;
531 key->ip.tos = nh->tos;
532 key->ip.ttl = nh->ttl;
534 offset = nh->frag_off & htons(IP_OFFSET);
536 key->ip.frag = OVS_FRAG_TYPE_LATER;
539 if (nh->frag_off & htons(IP_MF) ||
540 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
541 key->ip.frag = OVS_FRAG_TYPE_FIRST;
543 key->ip.frag = OVS_FRAG_TYPE_NONE;
545 /* Transport layer. */
546 if (key->ip.proto == IPPROTO_TCP) {
547 if (tcphdr_ok(skb)) {
548 struct tcphdr *tcp = tcp_hdr(skb);
549 key->tp.src = tcp->source;
550 key->tp.dst = tcp->dest;
551 key->tp.flags = TCP_FLAGS_BE16(tcp);
553 memset(&key->tp, 0, sizeof(key->tp));
555 } else if (key->ip.proto == IPPROTO_UDP) {
556 if (udphdr_ok(skb)) {
557 struct udphdr *udp = udp_hdr(skb);
558 key->tp.src = udp->source;
559 key->tp.dst = udp->dest;
561 memset(&key->tp, 0, sizeof(key->tp));
563 } else if (key->ip.proto == IPPROTO_SCTP) {
564 if (sctphdr_ok(skb)) {
565 struct sctphdr *sctp = sctp_hdr(skb);
566 key->tp.src = sctp->source;
567 key->tp.dst = sctp->dest;
569 memset(&key->tp, 0, sizeof(key->tp));
571 } else if (key->ip.proto == IPPROTO_ICMP) {
572 if (icmphdr_ok(skb)) {
573 struct icmphdr *icmp = icmp_hdr(skb);
574 /* The ICMP type and code fields use the 16-bit
575 * transport port fields, so we need to store
576 * them in 16-bit network byte order. */
577 key->tp.src = htons(icmp->type);
578 key->tp.dst = htons(icmp->code);
580 memset(&key->tp, 0, sizeof(key->tp));
584 } else if (key->eth.type == htons(ETH_P_ARP) ||
585 key->eth.type == htons(ETH_P_RARP)) {
586 struct arp_eth_header *arp;
588 arp = (struct arp_eth_header *)skb_network_header(skb);
591 && arp->ar_hrd == htons(ARPHRD_ETHER)
592 && arp->ar_pro == htons(ETH_P_IP)
593 && arp->ar_hln == ETH_ALEN
594 && arp->ar_pln == 4) {
596 /* We only match on the lower 8 bits of the opcode. */
597 if (ntohs(arp->ar_op) <= 0xff)
598 key->ip.proto = ntohs(arp->ar_op);
602 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
603 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
604 ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
605 ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
607 memset(&key->ip, 0, sizeof(key->ip));
608 memset(&key->ipv4, 0, sizeof(key->ipv4));
610 } else if (eth_p_mpls(key->eth.type)) {
611 size_t stack_len = MPLS_HLEN;
613 /* In the presence of an MPLS label stack the end of the L2
614 * header and the beginning of the L3 header differ.
616 * Advance network_header to the beginning of the L3
617 * header. mac_len corresponds to the end of the L2 header.
622 error = check_header(skb, skb->mac_len + stack_len);
626 memcpy(&lse, skb_network_header(skb), MPLS_HLEN);
628 if (stack_len == MPLS_HLEN)
629 memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
631 skb_set_network_header(skb, skb->mac_len + stack_len);
632 if (lse & htonl(MPLS_BOS_MASK))
635 stack_len += MPLS_HLEN;
637 } else if (key->eth.type == htons(ETH_P_IPV6)) {
638 int nh_len; /* IPv6 Header + Extensions */
640 nh_len = parse_ipv6hdr(skb, key);
641 if (unlikely(nh_len < 0)) {
642 memset(&key->ip, 0, sizeof(key->ip));
643 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
644 if (nh_len == -EINVAL) {
645 skb->transport_header = skb->network_header;
653 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
655 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
656 key->ip.frag = OVS_FRAG_TYPE_FIRST;
658 /* Transport layer. */
659 if (key->ip.proto == NEXTHDR_TCP) {
660 if (tcphdr_ok(skb)) {
661 struct tcphdr *tcp = tcp_hdr(skb);
662 key->tp.src = tcp->source;
663 key->tp.dst = tcp->dest;
664 key->tp.flags = TCP_FLAGS_BE16(tcp);
666 memset(&key->tp, 0, sizeof(key->tp));
668 } else if (key->ip.proto == NEXTHDR_UDP) {
669 if (udphdr_ok(skb)) {
670 struct udphdr *udp = udp_hdr(skb);
671 key->tp.src = udp->source;
672 key->tp.dst = udp->dest;
674 memset(&key->tp, 0, sizeof(key->tp));
676 } else if (key->ip.proto == NEXTHDR_SCTP) {
677 if (sctphdr_ok(skb)) {
678 struct sctphdr *sctp = sctp_hdr(skb);
679 key->tp.src = sctp->source;
680 key->tp.dst = sctp->dest;
682 memset(&key->tp, 0, sizeof(key->tp));
684 } else if (key->ip.proto == NEXTHDR_ICMP) {
685 if (icmp6hdr_ok(skb)) {
686 error = parse_icmpv6(skb, key, nh_len);
690 memset(&key->tp, 0, sizeof(key->tp));