1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2007-2014 Nicira, Inc.
6 #include <linux/uaccess.h>
7 #include <linux/netdevice.h>
8 #include <linux/etherdevice.h>
9 #include <linux/if_ether.h>
10 #include <linux/if_vlan.h>
11 #include <net/llc_pdu.h>
12 #include <linux/kernel.h>
13 #include <linux/jhash.h>
14 #include <linux/jiffies.h>
15 #include <linux/llc.h>
16 #include <linux/module.h>
18 #include <linux/rcupdate.h>
19 #include <linux/cpumask.h>
20 #include <linux/if_arp.h>
22 #include <linux/ipv6.h>
23 #include <linux/mpls.h>
24 #include <linux/sctp.h>
25 #include <linux/smp.h>
26 #include <linux/tcp.h>
27 #include <linux/udp.h>
28 #include <linux/icmp.h>
29 #include <linux/icmpv6.h>
30 #include <linux/rculist.h>
32 #include <net/ip_tunnels.h>
35 #include <net/ndisc.h>
37 #include <net/pkt_cls.h>
38 #include <net/netfilter/nf_conntrack_zones.h>
40 #include "conntrack.h"
43 #include "flow_netlink.h"
46 u64 ovs_flow_used_time(unsigned long flow_jiffies)
48 struct timespec64 cur_ts;
51 ktime_get_ts64(&cur_ts);
52 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
53 cur_ms = (u64)(u32)cur_ts.tv_sec * MSEC_PER_SEC +
54 cur_ts.tv_nsec / NSEC_PER_MSEC;
56 return cur_ms - idle_ms;
59 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
61 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
62 const struct sk_buff *skb)
64 struct sw_flow_stats *stats;
65 unsigned int cpu = smp_processor_id();
66 int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
68 stats = rcu_dereference(flow->stats[cpu]);
70 /* Check if already have CPU-specific stats. */
72 spin_lock(&stats->lock);
73 /* Mark if we write on the pre-allocated stats. */
74 if (cpu == 0 && unlikely(flow->stats_last_writer != cpu))
75 flow->stats_last_writer = cpu;
77 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
78 spin_lock(&stats->lock);
80 /* If the current CPU is the only writer on the
81 * pre-allocated stats keep using them.
83 if (unlikely(flow->stats_last_writer != cpu)) {
84 /* A previous locker may have already allocated the
85 * stats, so we need to check again. If CPU-specific
86 * stats were already allocated, we update the pre-
87 * allocated stats as we have already locked them.
89 if (likely(flow->stats_last_writer != -1) &&
90 likely(!rcu_access_pointer(flow->stats[cpu]))) {
91 /* Try to allocate CPU-specific stats. */
92 struct sw_flow_stats *new_stats;
95 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 = len;
105 new_stats->tcp_flags = tcp_flags;
106 spin_lock_init(&new_stats->lock);
108 rcu_assign_pointer(flow->stats[cpu],
111 flow->cpu_used_mask);
115 flow->stats_last_writer = cpu;
119 stats->used = jiffies;
120 stats->packet_count++;
121 stats->byte_count += 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 /* We open code this to make sure cpu 0 is always considered */
139 for (cpu = 0; cpu < nr_cpu_ids;
140 cpu = cpumask_next(cpu, flow->cpu_used_mask)) {
141 struct sw_flow_stats *stats = rcu_dereference_ovsl(flow->stats[cpu]);
144 /* Local CPU may write on non-local stats, so we must
145 * block bottom-halves here.
147 spin_lock_bh(&stats->lock);
148 if (!*used || time_after(stats->used, *used))
150 *tcp_flags |= stats->tcp_flags;
151 ovs_stats->n_packets += stats->packet_count;
152 ovs_stats->n_bytes += stats->byte_count;
153 spin_unlock_bh(&stats->lock);
158 /* Called with ovs_mutex. */
159 void ovs_flow_stats_clear(struct sw_flow *flow)
163 /* We open code this to make sure cpu 0 is always considered */
164 for (cpu = 0; cpu < nr_cpu_ids;
165 cpu = cpumask_next(cpu, flow->cpu_used_mask)) {
166 struct sw_flow_stats *stats = ovsl_dereference(flow->stats[cpu]);
169 spin_lock_bh(&stats->lock);
171 stats->packet_count = 0;
172 stats->byte_count = 0;
173 stats->tcp_flags = 0;
174 spin_unlock_bh(&stats->lock);
179 static int check_header(struct sk_buff *skb, int len)
181 if (unlikely(skb->len < len))
183 if (unlikely(!pskb_may_pull(skb, len)))
188 static bool arphdr_ok(struct sk_buff *skb)
190 return pskb_may_pull(skb, skb_network_offset(skb) +
191 sizeof(struct arp_eth_header));
194 static int check_iphdr(struct sk_buff *skb)
196 unsigned int nh_ofs = skb_network_offset(skb);
200 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
204 ip_len = ip_hdrlen(skb);
205 if (unlikely(ip_len < sizeof(struct iphdr) ||
206 skb->len < nh_ofs + ip_len))
209 skb_set_transport_header(skb, nh_ofs + ip_len);
213 static bool tcphdr_ok(struct sk_buff *skb)
215 int th_ofs = skb_transport_offset(skb);
218 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
221 tcp_len = tcp_hdrlen(skb);
222 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
223 skb->len < th_ofs + tcp_len))
229 static bool udphdr_ok(struct sk_buff *skb)
231 return pskb_may_pull(skb, skb_transport_offset(skb) +
232 sizeof(struct udphdr));
235 static bool sctphdr_ok(struct sk_buff *skb)
237 return pskb_may_pull(skb, skb_transport_offset(skb) +
238 sizeof(struct sctphdr));
241 static bool icmphdr_ok(struct sk_buff *skb)
243 return pskb_may_pull(skb, skb_transport_offset(skb) +
244 sizeof(struct icmphdr));
248 * get_ipv6_ext_hdrs() - Parses packet and sets IPv6 extension header flags.
250 * @skb: buffer where extension header data starts in packet
252 * @ext_hdrs: flags are stored here
254 * OFPIEH12_UNREP is set if more than one of a given IPv6 extension header
255 * is unexpectedly encountered. (Two destination options headers may be
256 * expected and would not cause this bit to be set.)
258 * OFPIEH12_UNSEQ is set if IPv6 extension headers were not in the order
259 * preferred (but not required) by RFC 2460:
261 * When more than one extension header is used in the same packet, it is
262 * recommended that those headers appear in the following order:
264 * Hop-by-Hop Options header
265 * Destination Options header
268 * Authentication header
269 * Encapsulating Security Payload header
270 * Destination Options header
273 static void get_ipv6_ext_hdrs(struct sk_buff *skb, struct ipv6hdr *nh,
276 u8 next_type = nh->nexthdr;
277 unsigned int start = skb_network_offset(skb) + sizeof(struct ipv6hdr);
278 int dest_options_header_count = 0;
282 while (ipv6_ext_hdr(next_type)) {
283 struct ipv6_opt_hdr _hdr, *hp;
287 *ext_hdrs |= OFPIEH12_NONEXT;
292 if (*ext_hdrs & OFPIEH12_ESP)
293 *ext_hdrs |= OFPIEH12_UNREP;
294 if ((*ext_hdrs & ~(OFPIEH12_HOP | OFPIEH12_DEST |
295 OFPIEH12_ROUTER | IPPROTO_FRAGMENT |
296 OFPIEH12_AUTH | OFPIEH12_UNREP)) ||
297 dest_options_header_count >= 2) {
298 *ext_hdrs |= OFPIEH12_UNSEQ;
300 *ext_hdrs |= OFPIEH12_ESP;
304 if (*ext_hdrs & OFPIEH12_AUTH)
305 *ext_hdrs |= OFPIEH12_UNREP;
307 ~(OFPIEH12_HOP | OFPIEH12_DEST | OFPIEH12_ROUTER |
308 IPPROTO_FRAGMENT | OFPIEH12_UNREP)) ||
309 dest_options_header_count >= 2) {
310 *ext_hdrs |= OFPIEH12_UNSEQ;
312 *ext_hdrs |= OFPIEH12_AUTH;
315 case IPPROTO_DSTOPTS:
316 if (dest_options_header_count == 0) {
318 ~(OFPIEH12_HOP | OFPIEH12_UNREP))
319 *ext_hdrs |= OFPIEH12_UNSEQ;
320 *ext_hdrs |= OFPIEH12_DEST;
321 } else if (dest_options_header_count == 1) {
323 ~(OFPIEH12_HOP | OFPIEH12_DEST |
324 OFPIEH12_ROUTER | OFPIEH12_FRAG |
325 OFPIEH12_AUTH | OFPIEH12_ESP |
327 *ext_hdrs |= OFPIEH12_UNSEQ;
330 *ext_hdrs |= OFPIEH12_UNREP;
332 dest_options_header_count++;
335 case IPPROTO_FRAGMENT:
336 if (*ext_hdrs & OFPIEH12_FRAG)
337 *ext_hdrs |= OFPIEH12_UNREP;
338 if ((*ext_hdrs & ~(OFPIEH12_HOP |
342 dest_options_header_count >= 2) {
343 *ext_hdrs |= OFPIEH12_UNSEQ;
345 *ext_hdrs |= OFPIEH12_FRAG;
348 case IPPROTO_ROUTING:
349 if (*ext_hdrs & OFPIEH12_ROUTER)
350 *ext_hdrs |= OFPIEH12_UNREP;
351 if ((*ext_hdrs & ~(OFPIEH12_HOP |
354 dest_options_header_count >= 2) {
355 *ext_hdrs |= OFPIEH12_UNSEQ;
357 *ext_hdrs |= OFPIEH12_ROUTER;
360 case IPPROTO_HOPOPTS:
361 if (*ext_hdrs & OFPIEH12_HOP)
362 *ext_hdrs |= OFPIEH12_UNREP;
363 /* OFPIEH12_HOP is set to 1 if a hop-by-hop IPv6
364 * extension header is present as the first
365 * extension header in the packet.
368 *ext_hdrs |= OFPIEH12_HOP;
370 *ext_hdrs |= OFPIEH12_UNSEQ;
377 hp = skb_header_pointer(skb, start, sizeof(_hdr), &_hdr);
380 next_type = hp->nexthdr;
381 start += ipv6_optlen(hp);
385 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
387 unsigned short frag_off;
388 unsigned int payload_ofs = 0;
389 unsigned int nh_ofs = skb_network_offset(skb);
392 int err, nexthdr, flags = 0;
394 err = check_header(skb, nh_ofs + sizeof(*nh));
400 get_ipv6_ext_hdrs(skb, nh, &key->ipv6.exthdrs);
402 key->ip.proto = NEXTHDR_NONE;
403 key->ip.tos = ipv6_get_dsfield(nh);
404 key->ip.ttl = nh->hop_limit;
405 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
406 key->ipv6.addr.src = nh->saddr;
407 key->ipv6.addr.dst = nh->daddr;
409 nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
410 if (flags & IP6_FH_F_FRAG) {
412 key->ip.frag = OVS_FRAG_TYPE_LATER;
413 key->ip.proto = NEXTHDR_FRAGMENT;
416 key->ip.frag = OVS_FRAG_TYPE_FIRST;
418 key->ip.frag = OVS_FRAG_TYPE_NONE;
421 /* Delayed handling of error in ipv6_find_hdr() as it
422 * always sets flags and frag_off to a valid value which may be
423 * used to set key->ip.frag above.
425 if (unlikely(nexthdr < 0))
428 nh_len = payload_ofs - nh_ofs;
429 skb_set_transport_header(skb, nh_ofs + nh_len);
430 key->ip.proto = nexthdr;
434 static bool icmp6hdr_ok(struct sk_buff *skb)
436 return pskb_may_pull(skb, skb_transport_offset(skb) +
437 sizeof(struct icmp6hdr));
441 * parse_vlan_tag - Parse vlan tag from vlan header.
442 * @skb: skb containing frame to parse
443 * @key_vh: pointer to parsed vlan tag
444 * @untag_vlan: should the vlan header be removed from the frame
446 * Return: ERROR on memory error.
447 * %0 if it encounters a non-vlan or incomplete packet.
448 * %1 after successfully parsing vlan tag.
450 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
453 struct vlan_head *vh = (struct vlan_head *)skb->data;
455 if (likely(!eth_type_vlan(vh->tpid)))
458 if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
461 if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
465 vh = (struct vlan_head *)skb->data;
466 key_vh->tci = vh->tci | htons(VLAN_CFI_MASK);
467 key_vh->tpid = vh->tpid;
469 if (unlikely(untag_vlan)) {
470 int offset = skb->data - skb_mac_header(skb);
474 __skb_push(skb, offset);
475 err = __skb_vlan_pop(skb, &tci);
476 __skb_pull(skb, offset);
479 __vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
481 __skb_pull(skb, sizeof(struct vlan_head));
486 static void clear_vlan(struct sw_flow_key *key)
488 key->eth.vlan.tci = 0;
489 key->eth.vlan.tpid = 0;
490 key->eth.cvlan.tci = 0;
491 key->eth.cvlan.tpid = 0;
494 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
498 if (skb_vlan_tag_present(skb)) {
499 key->eth.vlan.tci = htons(skb->vlan_tci) | htons(VLAN_CFI_MASK);
500 key->eth.vlan.tpid = skb->vlan_proto;
502 /* Parse outer vlan tag in the non-accelerated case. */
503 res = parse_vlan_tag(skb, &key->eth.vlan, true);
508 /* Parse inner vlan tag. */
509 res = parse_vlan_tag(skb, &key->eth.cvlan, false);
516 static __be16 parse_ethertype(struct sk_buff *skb)
518 struct llc_snap_hdr {
519 u8 dsap; /* Always 0xAA */
520 u8 ssap; /* Always 0xAA */
525 struct llc_snap_hdr *llc;
528 proto = *(__be16 *) skb->data;
529 __skb_pull(skb, sizeof(__be16));
531 if (eth_proto_is_802_3(proto))
534 if (skb->len < sizeof(struct llc_snap_hdr))
535 return htons(ETH_P_802_2);
537 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
540 llc = (struct llc_snap_hdr *) skb->data;
541 if (llc->dsap != LLC_SAP_SNAP ||
542 llc->ssap != LLC_SAP_SNAP ||
543 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
544 return htons(ETH_P_802_2);
546 __skb_pull(skb, sizeof(struct llc_snap_hdr));
548 if (eth_proto_is_802_3(llc->ethertype))
549 return llc->ethertype;
551 return htons(ETH_P_802_2);
554 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
557 struct icmp6hdr *icmp = icmp6_hdr(skb);
559 /* The ICMPv6 type and code fields use the 16-bit transport port
560 * fields, so we need to store them in 16-bit network byte order.
562 key->tp.src = htons(icmp->icmp6_type);
563 key->tp.dst = htons(icmp->icmp6_code);
565 if (icmp->icmp6_code == 0 &&
566 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
567 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
568 int icmp_len = skb->len - skb_transport_offset(skb);
572 memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
574 /* In order to process neighbor discovery options, we need the
577 if (unlikely(icmp_len < sizeof(*nd)))
580 if (unlikely(skb_linearize(skb)))
583 nd = (struct nd_msg *)skb_transport_header(skb);
584 key->ipv6.nd.target = nd->target;
586 icmp_len -= sizeof(*nd);
588 while (icmp_len >= 8) {
589 struct nd_opt_hdr *nd_opt =
590 (struct nd_opt_hdr *)(nd->opt + offset);
591 int opt_len = nd_opt->nd_opt_len * 8;
593 if (unlikely(!opt_len || opt_len > icmp_len))
596 /* Store the link layer address if the appropriate
597 * option is provided. It is considered an error if
598 * the same link layer option is specified twice.
600 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
602 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
604 ether_addr_copy(key->ipv6.nd.sll,
605 &nd->opt[offset+sizeof(*nd_opt)]);
606 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
608 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
610 ether_addr_copy(key->ipv6.nd.tll,
611 &nd->opt[offset+sizeof(*nd_opt)]);
622 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
623 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
624 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
629 static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key)
632 unsigned int nh_ofs = skb_network_offset(skb);
636 err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN);
641 version = nsh_get_ver(nh);
642 length = nsh_hdr_len(nh);
647 err = check_header(skb, nh_ofs + length);
652 key->nsh.base.flags = nsh_get_flags(nh);
653 key->nsh.base.ttl = nsh_get_ttl(nh);
654 key->nsh.base.mdtype = nh->mdtype;
655 key->nsh.base.np = nh->np;
656 key->nsh.base.path_hdr = nh->path_hdr;
657 switch (key->nsh.base.mdtype) {
659 if (length != NSH_M_TYPE1_LEN)
661 memcpy(key->nsh.context, nh->md1.context,
665 memset(key->nsh.context, 0,
676 * key_extract_l3l4 - extracts L3/L4 header information.
677 * @skb: sk_buff that contains the frame, with skb->data pointing to the
679 * @key: output flow key
681 * Return: %0 if successful, otherwise a negative errno value.
683 static int key_extract_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
688 if (key->eth.type == htons(ETH_P_IP)) {
692 error = check_iphdr(skb);
693 if (unlikely(error)) {
694 memset(&key->ip, 0, sizeof(key->ip));
695 memset(&key->ipv4, 0, sizeof(key->ipv4));
696 if (error == -EINVAL) {
697 skb->transport_header = skb->network_header;
704 key->ipv4.addr.src = nh->saddr;
705 key->ipv4.addr.dst = nh->daddr;
707 key->ip.proto = nh->protocol;
708 key->ip.tos = nh->tos;
709 key->ip.ttl = nh->ttl;
711 offset = nh->frag_off & htons(IP_OFFSET);
713 key->ip.frag = OVS_FRAG_TYPE_LATER;
714 memset(&key->tp, 0, sizeof(key->tp));
717 if (nh->frag_off & htons(IP_MF) ||
718 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
719 key->ip.frag = OVS_FRAG_TYPE_FIRST;
721 key->ip.frag = OVS_FRAG_TYPE_NONE;
723 /* Transport layer. */
724 if (key->ip.proto == IPPROTO_TCP) {
725 if (tcphdr_ok(skb)) {
726 struct tcphdr *tcp = tcp_hdr(skb);
727 key->tp.src = tcp->source;
728 key->tp.dst = tcp->dest;
729 key->tp.flags = TCP_FLAGS_BE16(tcp);
731 memset(&key->tp, 0, sizeof(key->tp));
734 } else if (key->ip.proto == IPPROTO_UDP) {
735 if (udphdr_ok(skb)) {
736 struct udphdr *udp = udp_hdr(skb);
737 key->tp.src = udp->source;
738 key->tp.dst = udp->dest;
740 memset(&key->tp, 0, sizeof(key->tp));
742 } else if (key->ip.proto == IPPROTO_SCTP) {
743 if (sctphdr_ok(skb)) {
744 struct sctphdr *sctp = sctp_hdr(skb);
745 key->tp.src = sctp->source;
746 key->tp.dst = sctp->dest;
748 memset(&key->tp, 0, sizeof(key->tp));
750 } else if (key->ip.proto == IPPROTO_ICMP) {
751 if (icmphdr_ok(skb)) {
752 struct icmphdr *icmp = icmp_hdr(skb);
753 /* The ICMP type and code fields use the 16-bit
754 * transport port fields, so we need to store
755 * them in 16-bit network byte order. */
756 key->tp.src = htons(icmp->type);
757 key->tp.dst = htons(icmp->code);
759 memset(&key->tp, 0, sizeof(key->tp));
763 } else if (key->eth.type == htons(ETH_P_ARP) ||
764 key->eth.type == htons(ETH_P_RARP)) {
765 struct arp_eth_header *arp;
766 bool arp_available = arphdr_ok(skb);
768 arp = (struct arp_eth_header *)skb_network_header(skb);
771 arp->ar_hrd == htons(ARPHRD_ETHER) &&
772 arp->ar_pro == htons(ETH_P_IP) &&
773 arp->ar_hln == ETH_ALEN &&
776 /* We only match on the lower 8 bits of the opcode. */
777 if (ntohs(arp->ar_op) <= 0xff)
778 key->ip.proto = ntohs(arp->ar_op);
782 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
783 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
784 ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
785 ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
787 memset(&key->ip, 0, sizeof(key->ip));
788 memset(&key->ipv4, 0, sizeof(key->ipv4));
790 } else if (eth_p_mpls(key->eth.type)) {
793 memset(&key->mpls, 0, sizeof(key->mpls));
794 skb_set_inner_network_header(skb, skb->mac_len);
798 error = check_header(skb, skb->mac_len +
799 label_count * MPLS_HLEN);
803 memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
805 if (label_count <= MPLS_LABEL_DEPTH)
806 memcpy(&key->mpls.lse[label_count - 1], &lse,
809 skb_set_inner_network_header(skb, skb->mac_len +
810 label_count * MPLS_HLEN);
811 if (lse & htonl(MPLS_LS_S_MASK))
816 if (label_count > MPLS_LABEL_DEPTH)
817 label_count = MPLS_LABEL_DEPTH;
819 key->mpls.num_labels_mask = GENMASK(label_count - 1, 0);
820 } else if (key->eth.type == htons(ETH_P_IPV6)) {
821 int nh_len; /* IPv6 Header + Extensions */
823 nh_len = parse_ipv6hdr(skb, key);
824 if (unlikely(nh_len < 0)) {
827 memset(&key->ip, 0, sizeof(key->ip));
828 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
831 skb->transport_header = skb->network_header;
840 if (key->ip.frag == OVS_FRAG_TYPE_LATER) {
841 memset(&key->tp, 0, sizeof(key->tp));
844 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
845 key->ip.frag = OVS_FRAG_TYPE_FIRST;
847 /* Transport layer. */
848 if (key->ip.proto == NEXTHDR_TCP) {
849 if (tcphdr_ok(skb)) {
850 struct tcphdr *tcp = tcp_hdr(skb);
851 key->tp.src = tcp->source;
852 key->tp.dst = tcp->dest;
853 key->tp.flags = TCP_FLAGS_BE16(tcp);
855 memset(&key->tp, 0, sizeof(key->tp));
857 } else if (key->ip.proto == NEXTHDR_UDP) {
858 if (udphdr_ok(skb)) {
859 struct udphdr *udp = udp_hdr(skb);
860 key->tp.src = udp->source;
861 key->tp.dst = udp->dest;
863 memset(&key->tp, 0, sizeof(key->tp));
865 } else if (key->ip.proto == NEXTHDR_SCTP) {
866 if (sctphdr_ok(skb)) {
867 struct sctphdr *sctp = sctp_hdr(skb);
868 key->tp.src = sctp->source;
869 key->tp.dst = sctp->dest;
871 memset(&key->tp, 0, sizeof(key->tp));
873 } else if (key->ip.proto == NEXTHDR_ICMP) {
874 if (icmp6hdr_ok(skb)) {
875 error = parse_icmpv6(skb, key, nh_len);
879 memset(&key->tp, 0, sizeof(key->tp));
882 } else if (key->eth.type == htons(ETH_P_NSH)) {
883 error = parse_nsh(skb, key);
891 * key_extract - extracts a flow key from an Ethernet frame.
892 * @skb: sk_buff that contains the frame, with skb->data pointing to the
894 * @key: output flow key
896 * The caller must ensure that skb->len >= ETH_HLEN.
898 * Initializes @skb header fields as follows:
900 * - skb->mac_header: the L2 header.
902 * - skb->network_header: just past the L2 header, or just past the
903 * VLAN header, to the first byte of the L2 payload.
905 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
906 * on output, then just past the IP header, if one is present and
907 * of a correct length, otherwise the same as skb->network_header.
908 * For other key->eth.type values it is left untouched.
910 * - skb->protocol: the type of the data starting at skb->network_header.
911 * Equals to key->eth.type.
913 * Return: %0 if successful, otherwise a negative errno value.
915 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
919 /* Flags are always used as part of stats */
922 skb_reset_mac_header(skb);
926 if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) {
927 if (unlikely(eth_type_vlan(skb->protocol)))
930 skb_reset_network_header(skb);
931 key->eth.type = skb->protocol;
934 ether_addr_copy(key->eth.src, eth->h_source);
935 ether_addr_copy(key->eth.dst, eth->h_dest);
937 __skb_pull(skb, 2 * ETH_ALEN);
938 /* We are going to push all headers that we pull, so no need to
939 * update skb->csum here.
942 if (unlikely(parse_vlan(skb, key)))
945 key->eth.type = parse_ethertype(skb);
946 if (unlikely(key->eth.type == htons(0)))
949 /* Multiple tagged packets need to retain TPID to satisfy
950 * skb_vlan_pop(), which will later shift the ethertype into
953 if (key->eth.cvlan.tci & htons(VLAN_CFI_MASK))
954 skb->protocol = key->eth.cvlan.tpid;
956 skb->protocol = key->eth.type;
958 skb_reset_network_header(skb);
959 __skb_push(skb, skb->data - skb_mac_header(skb));
962 skb_reset_mac_len(skb);
964 /* Fill out L3/L4 key info, if any */
965 return key_extract_l3l4(skb, key);
968 /* In the case of conntrack fragment handling it expects L3 headers,
971 int ovs_flow_key_update_l3l4(struct sk_buff *skb, struct sw_flow_key *key)
973 return key_extract_l3l4(skb, key);
976 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
980 res = key_extract(skb, key);
982 key->mac_proto &= ~SW_FLOW_KEY_INVALID;
987 static int key_extract_mac_proto(struct sk_buff *skb)
989 switch (skb->dev->type) {
991 return MAC_PROTO_ETHERNET;
993 if (skb->protocol == htons(ETH_P_TEB))
994 return MAC_PROTO_ETHERNET;
995 return MAC_PROTO_NONE;
1001 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
1002 struct sk_buff *skb, struct sw_flow_key *key)
1004 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1005 struct tc_skb_ext *tc_ext;
1007 bool post_ct = false, post_ct_snat = false, post_ct_dnat = false;
1011 /* Extract metadata from packet. */
1013 key->tun_proto = ip_tunnel_info_af(tun_info);
1014 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
1016 if (tun_info->options_len) {
1017 BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
1019 > sizeof(key->tun_opts));
1021 ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
1023 key->tun_opts_len = tun_info->options_len;
1025 key->tun_opts_len = 0;
1029 key->tun_opts_len = 0;
1030 memset(&key->tun_key, 0, sizeof(key->tun_key));
1033 key->phy.priority = skb->priority;
1034 key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
1035 key->phy.skb_mark = skb->mark;
1036 key->ovs_flow_hash = 0;
1037 res = key_extract_mac_proto(skb);
1040 key->mac_proto = res;
1042 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1043 if (tc_skb_ext_tc_enabled()) {
1044 tc_ext = skb_ext_find(skb, TC_SKB_EXT);
1045 key->recirc_id = tc_ext && !tc_ext->act_miss ?
1047 OVS_CB(skb)->mru = tc_ext ? tc_ext->mru : 0;
1048 post_ct = tc_ext ? tc_ext->post_ct : false;
1049 post_ct_snat = post_ct ? tc_ext->post_ct_snat : false;
1050 post_ct_dnat = post_ct ? tc_ext->post_ct_dnat : false;
1051 zone = post_ct ? tc_ext->zone : 0;
1059 err = key_extract(skb, key);
1061 ovs_ct_fill_key(skb, key, post_ct); /* Must be after key_extract(). */
1063 if (!skb_get_nfct(skb)) {
1064 key->ct_zone = zone;
1067 key->ct_state &= ~OVS_CS_F_DST_NAT;
1069 key->ct_state &= ~OVS_CS_F_SRC_NAT;
1076 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
1077 struct sk_buff *skb,
1078 struct sw_flow_key *key, bool log)
1080 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1084 err = parse_flow_nlattrs(attr, a, &attrs, log);
1088 /* Extract metadata from netlink attributes. */
1089 err = ovs_nla_get_flow_metadata(net, a, attrs, key, log);
1093 /* key_extract assumes that skb->protocol is set-up for
1094 * layer 3 packets which is the case for other callers,
1095 * in particular packets received from the network stack.
1096 * Here the correct value can be set from the metadata
1098 * For L2 packet key eth type would be zero. skb protocol
1099 * would be set to correct value later during key-extact.
1102 skb->protocol = key->eth.type;
1103 err = key_extract(skb, key);
1107 /* Check that we have conntrack original direction tuple metadata only
1108 * for packets for which it makes sense. Otherwise the key may be
1109 * corrupted due to overlapping key fields.
1111 if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) &&
1112 key->eth.type != htons(ETH_P_IP))
1114 if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) &&
1115 (key->eth.type != htons(ETH_P_IPV6) ||
1116 sw_flow_key_is_nd(key)))