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/sctp.h>
36 #include <linux/smp.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/icmp.h>
40 #include <linux/icmpv6.h>
41 #include <linux/rculist.h>
44 #include <net/ndisc.h>
48 #include "flow_netlink.h"
52 u64 ovs_flow_used_time(unsigned long flow_jiffies)
54 struct timespec cur_ts;
57 ktime_get_ts(&cur_ts);
58 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
59 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
60 cur_ts.tv_nsec / NSEC_PER_MSEC;
62 return cur_ms - idle_ms;
65 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
67 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
70 struct flow_stats *stats;
71 int node = numa_node_id();
73 stats = rcu_dereference(flow->stats[node]);
75 /* Check if already have node-specific stats. */
77 spin_lock(&stats->lock);
78 /* Mark if we write on the pre-allocated stats. */
79 if (node == 0 && unlikely(flow->stats_last_writer != node))
80 flow->stats_last_writer = node;
82 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
83 spin_lock(&stats->lock);
85 /* If the current NUMA-node is the only writer on the
86 * pre-allocated stats keep using them.
88 if (unlikely(flow->stats_last_writer != node)) {
89 /* A previous locker may have already allocated the
90 * stats, so we need to check again. If node-specific
91 * stats were already allocated, we update the pre-
92 * allocated stats as we have already locked them.
94 if (likely(flow->stats_last_writer != NUMA_NO_NODE)
95 && likely(!rcu_dereference(flow->stats[node]))) {
96 /* Try to allocate node-specific stats. */
97 struct flow_stats *new_stats;
100 kmem_cache_alloc_node(flow_stats_cache,
104 if (likely(new_stats)) {
105 new_stats->used = jiffies;
106 new_stats->packet_count = 1;
107 new_stats->byte_count = skb->len;
108 new_stats->tcp_flags = tcp_flags;
109 spin_lock_init(&new_stats->lock);
111 rcu_assign_pointer(flow->stats[node],
116 flow->stats_last_writer = node;
120 stats->used = jiffies;
121 stats->packet_count++;
122 stats->byte_count += skb->len;
123 stats->tcp_flags |= tcp_flags;
125 spin_unlock(&stats->lock);
128 /* Must be called with rcu_read_lock or ovs_mutex. */
129 void ovs_flow_stats_get(const struct sw_flow *flow,
130 struct ovs_flow_stats *ovs_stats,
131 unsigned long *used, __be16 *tcp_flags)
137 memset(ovs_stats, 0, sizeof(*ovs_stats));
139 for_each_node(node) {
140 struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[node]);
143 /* Local CPU may write on non-local stats, so we must
144 * block bottom-halves here.
146 spin_lock_bh(&stats->lock);
147 if (!*used || time_after(stats->used, *used))
149 *tcp_flags |= stats->tcp_flags;
150 ovs_stats->n_packets += stats->packet_count;
151 ovs_stats->n_bytes += stats->byte_count;
152 spin_unlock_bh(&stats->lock);
157 /* Called with ovs_mutex. */
158 void ovs_flow_stats_clear(struct sw_flow *flow)
162 for_each_node(node) {
163 struct flow_stats *stats = ovsl_dereference(flow->stats[node]);
166 spin_lock_bh(&stats->lock);
168 stats->packet_count = 0;
169 stats->byte_count = 0;
170 stats->tcp_flags = 0;
171 spin_unlock_bh(&stats->lock);
176 static int check_header(struct sk_buff *skb, int len)
178 if (unlikely(skb->len < len))
180 if (unlikely(!pskb_may_pull(skb, len)))
185 static bool arphdr_ok(struct sk_buff *skb)
187 return pskb_may_pull(skb, skb_network_offset(skb) +
188 sizeof(struct arp_eth_header));
191 static int check_iphdr(struct sk_buff *skb)
193 unsigned int nh_ofs = skb_network_offset(skb);
197 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
201 ip_len = ip_hdrlen(skb);
202 if (unlikely(ip_len < sizeof(struct iphdr) ||
203 skb->len < nh_ofs + ip_len))
206 skb_set_transport_header(skb, nh_ofs + ip_len);
210 static bool tcphdr_ok(struct sk_buff *skb)
212 int th_ofs = skb_transport_offset(skb);
215 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
218 tcp_len = tcp_hdrlen(skb);
219 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
220 skb->len < th_ofs + tcp_len))
226 static bool udphdr_ok(struct sk_buff *skb)
228 return pskb_may_pull(skb, skb_transport_offset(skb) +
229 sizeof(struct udphdr));
232 static bool sctphdr_ok(struct sk_buff *skb)
234 return pskb_may_pull(skb, skb_transport_offset(skb) +
235 sizeof(struct sctphdr));
238 static bool icmphdr_ok(struct sk_buff *skb)
240 return pskb_may_pull(skb, skb_transport_offset(skb) +
241 sizeof(struct icmphdr));
244 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
246 unsigned int nh_ofs = skb_network_offset(skb);
254 err = check_header(skb, nh_ofs + sizeof(*nh));
259 nexthdr = nh->nexthdr;
260 payload_ofs = (u8 *)(nh + 1) - skb->data;
262 key->ip.proto = NEXTHDR_NONE;
263 key->ip.tos = ipv6_get_dsfield(nh);
264 key->ip.ttl = nh->hop_limit;
265 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
266 key->ipv6.addr.src = nh->saddr;
267 key->ipv6.addr.dst = nh->daddr;
269 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
270 if (unlikely(payload_ofs < 0))
274 if (frag_off & htons(~0x7))
275 key->ip.frag = OVS_FRAG_TYPE_LATER;
277 key->ip.frag = OVS_FRAG_TYPE_FIRST;
280 nh_len = payload_ofs - nh_ofs;
281 skb_set_transport_header(skb, nh_ofs + nh_len);
282 key->ip.proto = nexthdr;
286 static bool icmp6hdr_ok(struct sk_buff *skb)
288 return pskb_may_pull(skb, skb_transport_offset(skb) +
289 sizeof(struct icmp6hdr));
292 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
295 __be16 eth_type; /* ETH_P_8021Q */
298 struct qtag_prefix *qp;
300 if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
303 if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
307 qp = (struct qtag_prefix *) skb->data;
308 key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
309 __skb_pull(skb, sizeof(struct qtag_prefix));
314 static __be16 parse_ethertype(struct sk_buff *skb)
316 struct llc_snap_hdr {
317 u8 dsap; /* Always 0xAA */
318 u8 ssap; /* Always 0xAA */
323 struct llc_snap_hdr *llc;
326 proto = *(__be16 *) skb->data;
327 __skb_pull(skb, sizeof(__be16));
329 if (ntohs(proto) >= ETH_P_802_3_MIN)
332 if (skb->len < sizeof(struct llc_snap_hdr))
333 return htons(ETH_P_802_2);
335 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
338 llc = (struct llc_snap_hdr *) skb->data;
339 if (llc->dsap != LLC_SAP_SNAP ||
340 llc->ssap != LLC_SAP_SNAP ||
341 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
342 return htons(ETH_P_802_2);
344 __skb_pull(skb, sizeof(struct llc_snap_hdr));
346 if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
347 return llc->ethertype;
349 return htons(ETH_P_802_2);
352 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
355 struct icmp6hdr *icmp = icmp6_hdr(skb);
357 /* The ICMPv6 type and code fields use the 16-bit transport port
358 * fields, so we need to store them in 16-bit network byte order.
360 key->tp.src = htons(icmp->icmp6_type);
361 key->tp.dst = htons(icmp->icmp6_code);
363 if (icmp->icmp6_code == 0 &&
364 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
365 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
366 int icmp_len = skb->len - skb_transport_offset(skb);
370 /* In order to process neighbor discovery options, we need the
373 if (unlikely(icmp_len < sizeof(*nd)))
376 if (unlikely(skb_linearize(skb)))
379 nd = (struct nd_msg *)skb_transport_header(skb);
380 key->ipv6.nd.target = nd->target;
382 icmp_len -= sizeof(*nd);
384 while (icmp_len >= 8) {
385 struct nd_opt_hdr *nd_opt =
386 (struct nd_opt_hdr *)(nd->opt + offset);
387 int opt_len = nd_opt->nd_opt_len * 8;
389 if (unlikely(!opt_len || opt_len > icmp_len))
392 /* Store the link layer address if the appropriate
393 * option is provided. It is considered an error if
394 * the same link layer option is specified twice.
396 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
398 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
400 ether_addr_copy(key->ipv6.nd.sll,
401 &nd->opt[offset+sizeof(*nd_opt)]);
402 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
404 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
406 ether_addr_copy(key->ipv6.nd.tll,
407 &nd->opt[offset+sizeof(*nd_opt)]);
418 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
419 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
420 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
426 * key_extract - extracts a flow key from an Ethernet frame.
427 * @skb: sk_buff that contains the frame, with skb->data pointing to the
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 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
452 skb_reset_mac_header(skb);
454 /* Link layer. We are guaranteed to have at least the 14 byte Ethernet
455 * header in the linear data area.
458 ether_addr_copy(key->eth.src, eth->h_source);
459 ether_addr_copy(key->eth.dst, eth->h_dest);
461 __skb_pull(skb, 2 * ETH_ALEN);
462 /* We are going to push all headers that we pull, so no need to
463 * update skb->csum here. */
465 if (vlan_tx_tag_present(skb))
466 key->eth.tci = htons(vlan_get_tci(skb));
467 else if (eth->h_proto == htons(ETH_P_8021Q))
468 if (unlikely(parse_vlan(skb, key)))
471 key->eth.type = parse_ethertype(skb);
472 if (unlikely(key->eth.type == htons(0)))
475 skb_reset_network_header(skb);
476 __skb_push(skb, skb->data - skb_mac_header(skb));
479 if (key->eth.type == htons(ETH_P_IP)) {
483 error = check_iphdr(skb);
484 if (unlikely(error)) {
485 if (error == -EINVAL) {
486 skb->transport_header = skb->network_header;
493 key->ipv4.addr.src = nh->saddr;
494 key->ipv4.addr.dst = nh->daddr;
496 key->ip.proto = nh->protocol;
497 key->ip.tos = nh->tos;
498 key->ip.ttl = nh->ttl;
500 offset = nh->frag_off & htons(IP_OFFSET);
502 key->ip.frag = OVS_FRAG_TYPE_LATER;
505 if (nh->frag_off & htons(IP_MF) ||
506 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
507 key->ip.frag = OVS_FRAG_TYPE_FIRST;
509 /* Transport layer. */
510 if (key->ip.proto == IPPROTO_TCP) {
511 if (tcphdr_ok(skb)) {
512 struct tcphdr *tcp = tcp_hdr(skb);
513 key->tp.src = tcp->source;
514 key->tp.dst = tcp->dest;
515 key->tp.flags = TCP_FLAGS_BE16(tcp);
517 } else if (key->ip.proto == IPPROTO_UDP) {
518 if (udphdr_ok(skb)) {
519 struct udphdr *udp = udp_hdr(skb);
520 key->tp.src = udp->source;
521 key->tp.dst = udp->dest;
523 } else if (key->ip.proto == IPPROTO_SCTP) {
524 if (sctphdr_ok(skb)) {
525 struct sctphdr *sctp = sctp_hdr(skb);
526 key->tp.src = sctp->source;
527 key->tp.dst = sctp->dest;
529 } else if (key->ip.proto == IPPROTO_ICMP) {
530 if (icmphdr_ok(skb)) {
531 struct icmphdr *icmp = icmp_hdr(skb);
532 /* The ICMP type and code fields use the 16-bit
533 * transport port fields, so we need to store
534 * them in 16-bit network byte order. */
535 key->tp.src = htons(icmp->type);
536 key->tp.dst = htons(icmp->code);
540 } else if ((key->eth.type == htons(ETH_P_ARP) ||
541 key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
542 bool arp_available = arphdr_ok(skb);
543 struct arp_eth_header *arp;
545 arp = (struct arp_eth_header *)skb_network_header(skb);
548 arp->ar_hrd == htons(ARPHRD_ETHER) &&
549 arp->ar_pro == htons(ETH_P_IP) &&
550 arp->ar_hln == ETH_ALEN &&
552 /* We only match on the lower 8 bits of the opcode. */
553 if (ntohs(arp->ar_op) <= 0xff)
554 key->ip.proto = ntohs(arp->ar_op);
555 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
556 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
557 ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
558 ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
560 } else if (key->eth.type == htons(ETH_P_IPV6)) {
561 int nh_len; /* IPv6 Header + Extensions */
563 nh_len = parse_ipv6hdr(skb, key);
564 if (unlikely(nh_len < 0)) {
565 if (nh_len == -EINVAL) {
566 skb->transport_header = skb->network_header;
574 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
576 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
577 key->ip.frag = OVS_FRAG_TYPE_FIRST;
579 /* Transport layer. */
580 if (key->ip.proto == NEXTHDR_TCP) {
581 if (tcphdr_ok(skb)) {
582 struct tcphdr *tcp = tcp_hdr(skb);
583 key->tp.src = tcp->source;
584 key->tp.dst = tcp->dest;
585 key->tp.flags = TCP_FLAGS_BE16(tcp);
587 } else if (key->ip.proto == NEXTHDR_UDP) {
588 if (udphdr_ok(skb)) {
589 struct udphdr *udp = udp_hdr(skb);
590 key->tp.src = udp->source;
591 key->tp.dst = udp->dest;
593 } else if (key->ip.proto == NEXTHDR_SCTP) {
594 if (sctphdr_ok(skb)) {
595 struct sctphdr *sctp = sctp_hdr(skb);
596 key->tp.src = sctp->source;
597 key->tp.dst = sctp->dest;
599 } else if (key->ip.proto == NEXTHDR_ICMP) {
600 if (icmp6hdr_ok(skb)) {
601 error = parse_icmpv6(skb, key, nh_len);
611 int ovs_flow_key_extract(struct sk_buff *skb, struct sw_flow_key *key)
613 /* Extract metadata from packet. */
615 memset(key, 0, sizeof(*key));
616 if (OVS_CB(skb)->tun_key)
617 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
619 key->phy.priority = skb->priority;
620 key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
621 key->phy.skb_mark = skb->mark;
623 return key_extract(skb, key);
626 int ovs_flow_key_extract_userspace(const struct nlattr *attr,
628 struct sw_flow_key *key)
632 memset(key, 0, sizeof(*key));
633 /* Extract metadata from netlink attributes. */
634 err = ovs_nla_get_flow_metadata(attr, key);
638 return key_extract(skb, key);
641 int ovs_flow_key_extract_recirc(u32 recirc_id,
642 const struct sw_flow_key *key,
644 struct sw_flow_key *new_key)
646 memset(new_key, 0, sizeof(*new_key));
647 memcpy(new_key, key, OVS_SW_FLOW_KEY_METADATA_SIZE);
648 new_key->recirc_id = recirc_id;
649 return key_extract(skb, new_key);