2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 #include <sys/types.h>
22 #include <netinet/in.h>
23 #include <netinet/icmp6.h>
24 #include <netinet/ip6.h>
28 #include "byte-order.h"
31 #include "dynamic-string.h"
36 #include "openflow/openflow.h"
38 #include "unaligned.h"
41 VLOG_DEFINE_THIS_MODULE(flow);
43 COVERAGE_DEFINE(flow_extract);
44 COVERAGE_DEFINE(miniflow_malloc);
46 static struct arp_eth_header *
47 pull_arp(struct ofpbuf *packet)
49 return ofpbuf_try_pull(packet, ARP_ETH_HEADER_LEN);
52 static struct ip_header *
53 pull_ip(struct ofpbuf *packet)
55 if (packet->size >= IP_HEADER_LEN) {
56 struct ip_header *ip = packet->data;
57 int ip_len = IP_IHL(ip->ip_ihl_ver) * 4;
58 if (ip_len >= IP_HEADER_LEN && packet->size >= ip_len) {
59 return ofpbuf_pull(packet, ip_len);
65 static struct tcp_header *
66 pull_tcp(struct ofpbuf *packet)
68 if (packet->size >= TCP_HEADER_LEN) {
69 struct tcp_header *tcp = packet->data;
70 int tcp_len = TCP_OFFSET(tcp->tcp_ctl) * 4;
71 if (tcp_len >= TCP_HEADER_LEN && packet->size >= tcp_len) {
72 return ofpbuf_pull(packet, tcp_len);
78 static struct udp_header *
79 pull_udp(struct ofpbuf *packet)
81 return ofpbuf_try_pull(packet, UDP_HEADER_LEN);
84 static struct sctp_header *
85 pull_sctp(struct ofpbuf *packet)
87 return ofpbuf_try_pull(packet, SCTP_HEADER_LEN);
90 static struct icmp_header *
91 pull_icmp(struct ofpbuf *packet)
93 return ofpbuf_try_pull(packet, ICMP_HEADER_LEN);
96 static struct icmp6_hdr *
97 pull_icmpv6(struct ofpbuf *packet)
99 return ofpbuf_try_pull(packet, sizeof(struct icmp6_hdr));
103 parse_mpls(struct ofpbuf *b, struct flow *flow)
107 while ((mh = ofpbuf_try_pull(b, sizeof *mh))) {
108 if (flow->mpls_depth++ == 0) {
109 flow->mpls_lse = mh->mpls_lse;
111 if (mh->mpls_lse & htonl(MPLS_BOS_MASK)) {
118 parse_vlan(struct ofpbuf *b, struct flow *flow)
121 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
125 if (b->size >= sizeof(struct qtag_prefix) + sizeof(ovs_be16)) {
126 struct qtag_prefix *qp = ofpbuf_pull(b, sizeof *qp);
127 flow->vlan_tci = qp->tci | htons(VLAN_CFI);
132 parse_ethertype(struct ofpbuf *b)
134 struct llc_snap_header *llc;
137 proto = *(ovs_be16 *) ofpbuf_pull(b, sizeof proto);
138 if (ntohs(proto) >= ETH_TYPE_MIN) {
142 if (b->size < sizeof *llc) {
143 return htons(FLOW_DL_TYPE_NONE);
147 if (llc->llc.llc_dsap != LLC_DSAP_SNAP
148 || llc->llc.llc_ssap != LLC_SSAP_SNAP
149 || llc->llc.llc_cntl != LLC_CNTL_SNAP
150 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
151 sizeof llc->snap.snap_org)) {
152 return htons(FLOW_DL_TYPE_NONE);
155 ofpbuf_pull(b, sizeof *llc);
157 if (ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN) {
158 return llc->snap.snap_type;
161 return htons(FLOW_DL_TYPE_NONE);
165 parse_ipv6(struct ofpbuf *packet, struct flow *flow)
167 const struct ip6_hdr *nh;
171 nh = ofpbuf_try_pull(packet, sizeof *nh);
176 nexthdr = nh->ip6_nxt;
178 flow->ipv6_src = nh->ip6_src;
179 flow->ipv6_dst = nh->ip6_dst;
181 tc_flow = get_unaligned_be32(&nh->ip6_flow);
182 flow->nw_tos = ntohl(tc_flow) >> 20;
183 flow->ipv6_label = tc_flow & htonl(IPV6_LABEL_MASK);
184 flow->nw_ttl = nh->ip6_hlim;
185 flow->nw_proto = IPPROTO_NONE;
188 if ((nexthdr != IPPROTO_HOPOPTS)
189 && (nexthdr != IPPROTO_ROUTING)
190 && (nexthdr != IPPROTO_DSTOPTS)
191 && (nexthdr != IPPROTO_AH)
192 && (nexthdr != IPPROTO_FRAGMENT)) {
193 /* It's either a terminal header (e.g., TCP, UDP) or one we
194 * don't understand. In either case, we're done with the
195 * packet, so use it to fill in 'nw_proto'. */
199 /* We only verify that at least 8 bytes of the next header are
200 * available, but many of these headers are longer. Ensure that
201 * accesses within the extension header are within those first 8
202 * bytes. All extension headers are required to be at least 8
204 if (packet->size < 8) {
208 if ((nexthdr == IPPROTO_HOPOPTS)
209 || (nexthdr == IPPROTO_ROUTING)
210 || (nexthdr == IPPROTO_DSTOPTS)) {
211 /* These headers, while different, have the fields we care about
212 * in the same location and with the same interpretation. */
213 const struct ip6_ext *ext_hdr = packet->data;
214 nexthdr = ext_hdr->ip6e_nxt;
215 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 1) * 8)) {
218 } else if (nexthdr == IPPROTO_AH) {
219 /* A standard AH definition isn't available, but the fields
220 * we care about are in the same location as the generic
221 * option header--only the header length is calculated
223 const struct ip6_ext *ext_hdr = packet->data;
224 nexthdr = ext_hdr->ip6e_nxt;
225 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 2) * 4)) {
228 } else if (nexthdr == IPPROTO_FRAGMENT) {
229 const struct ip6_frag *frag_hdr = packet->data;
231 nexthdr = frag_hdr->ip6f_nxt;
232 if (!ofpbuf_try_pull(packet, sizeof *frag_hdr)) {
236 /* We only process the first fragment. */
237 if (frag_hdr->ip6f_offlg != htons(0)) {
238 flow->nw_frag = FLOW_NW_FRAG_ANY;
239 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
240 flow->nw_frag |= FLOW_NW_FRAG_LATER;
241 nexthdr = IPPROTO_FRAGMENT;
248 flow->nw_proto = nexthdr;
253 parse_tcp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
255 const struct tcp_header *tcp = pull_tcp(b);
257 flow->tp_src = tcp->tcp_src;
258 flow->tp_dst = tcp->tcp_dst;
259 packet->l7 = b->data;
264 parse_udp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
266 const struct udp_header *udp = pull_udp(b);
268 flow->tp_src = udp->udp_src;
269 flow->tp_dst = udp->udp_dst;
270 packet->l7 = b->data;
275 parse_sctp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
277 const struct sctp_header *sctp = pull_sctp(b);
279 flow->tp_src = sctp->sctp_src;
280 flow->tp_dst = sctp->sctp_dst;
281 packet->l7 = b->data;
286 parse_icmpv6(struct ofpbuf *b, struct flow *flow)
288 const struct icmp6_hdr *icmp = pull_icmpv6(b);
294 /* The ICMPv6 type and code fields use the 16-bit transport port
295 * fields, so we need to store them in 16-bit network byte order. */
296 flow->tp_src = htons(icmp->icmp6_type);
297 flow->tp_dst = htons(icmp->icmp6_code);
299 if (icmp->icmp6_code == 0 &&
300 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
301 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
302 const struct in6_addr *nd_target;
304 nd_target = ofpbuf_try_pull(b, sizeof *nd_target);
308 flow->nd_target = *nd_target;
310 while (b->size >= 8) {
311 /* The minimum size of an option is 8 bytes, which also is
312 * the size of Ethernet link-layer options. */
313 const struct nd_opt_hdr *nd_opt = b->data;
314 int opt_len = nd_opt->nd_opt_len * 8;
316 if (!opt_len || opt_len > b->size) {
320 /* Store the link layer address if the appropriate option is
321 * provided. It is considered an error if the same link
322 * layer option is specified twice. */
323 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
325 if (eth_addr_is_zero(flow->arp_sha)) {
326 memcpy(flow->arp_sha, nd_opt + 1, ETH_ADDR_LEN);
330 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
332 if (eth_addr_is_zero(flow->arp_tha)) {
333 memcpy(flow->arp_tha, nd_opt + 1, ETH_ADDR_LEN);
339 if (!ofpbuf_try_pull(b, opt_len)) {
348 memset(&flow->nd_target, 0, sizeof(flow->nd_target));
349 memset(flow->arp_sha, 0, sizeof(flow->arp_sha));
350 memset(flow->arp_tha, 0, sizeof(flow->arp_tha));
356 /* Initializes 'flow' members from 'packet', 'skb_priority', 'tnl', and
359 * Initializes 'packet' header pointers as follows:
361 * - packet->l2 to the start of the Ethernet header.
363 * - packet->l2_5 to the start of the MPLS shim header.
365 * - packet->l3 to just past the Ethernet header, or just past the
366 * vlan_header if one is present, to the first byte of the payload of the
369 * - packet->l4 to just past the IPv4 header, if one is present and has a
370 * correct length, and otherwise NULL.
372 * - packet->l7 to just past the TCP/UDP/SCTP/ICMP header, if one is
373 * present and has a correct length, and otherwise NULL.
376 flow_extract(struct ofpbuf *packet, uint32_t skb_priority, uint32_t pkt_mark,
377 const struct flow_tnl *tnl, const union flow_in_port *in_port,
380 struct ofpbuf b = *packet;
381 struct eth_header *eth;
383 COVERAGE_INC(flow_extract);
385 memset(flow, 0, sizeof *flow);
388 ovs_assert(tnl != &flow->tunnel);
392 flow->in_port = *in_port;
394 flow->skb_priority = skb_priority;
395 flow->pkt_mark = pkt_mark;
403 if (b.size < sizeof *eth) {
409 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
410 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
412 /* dl_type, vlan_tci. */
413 ofpbuf_pull(&b, ETH_ADDR_LEN * 2);
414 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
415 parse_vlan(&b, flow);
417 flow->dl_type = parse_ethertype(&b);
419 /* Parse mpls, copy l3 ttl. */
420 if (eth_type_mpls(flow->dl_type)) {
421 packet->l2_5 = b.data;
422 parse_mpls(&b, flow);
427 if (flow->dl_type == htons(ETH_TYPE_IP)) {
428 const struct ip_header *nh = pull_ip(&b);
432 flow->nw_src = get_unaligned_be32(&nh->ip_src);
433 flow->nw_dst = get_unaligned_be32(&nh->ip_dst);
434 flow->nw_proto = nh->ip_proto;
436 flow->nw_tos = nh->ip_tos;
437 if (IP_IS_FRAGMENT(nh->ip_frag_off)) {
438 flow->nw_frag = FLOW_NW_FRAG_ANY;
439 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
440 flow->nw_frag |= FLOW_NW_FRAG_LATER;
443 flow->nw_ttl = nh->ip_ttl;
445 if (!(nh->ip_frag_off & htons(IP_FRAG_OFF_MASK))) {
446 if (flow->nw_proto == IPPROTO_TCP) {
447 parse_tcp(packet, &b, flow);
448 } else if (flow->nw_proto == IPPROTO_UDP) {
449 parse_udp(packet, &b, flow);
450 } else if (flow->nw_proto == IPPROTO_SCTP) {
451 parse_sctp(packet, &b, flow);
452 } else if (flow->nw_proto == IPPROTO_ICMP) {
453 const struct icmp_header *icmp = pull_icmp(&b);
455 flow->tp_src = htons(icmp->icmp_type);
456 flow->tp_dst = htons(icmp->icmp_code);
462 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
463 if (parse_ipv6(&b, flow)) {
468 if (flow->nw_proto == IPPROTO_TCP) {
469 parse_tcp(packet, &b, flow);
470 } else if (flow->nw_proto == IPPROTO_UDP) {
471 parse_udp(packet, &b, flow);
472 } else if (flow->nw_proto == IPPROTO_SCTP) {
473 parse_sctp(packet, &b, flow);
474 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
475 if (parse_icmpv6(&b, flow)) {
479 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
480 flow->dl_type == htons(ETH_TYPE_RARP)) {
481 const struct arp_eth_header *arp = pull_arp(&b);
482 if (arp && arp->ar_hrd == htons(1)
483 && arp->ar_pro == htons(ETH_TYPE_IP)
484 && arp->ar_hln == ETH_ADDR_LEN
485 && arp->ar_pln == 4) {
486 /* We only match on the lower 8 bits of the opcode. */
487 if (ntohs(arp->ar_op) <= 0xff) {
488 flow->nw_proto = ntohs(arp->ar_op);
491 flow->nw_src = arp->ar_spa;
492 flow->nw_dst = arp->ar_tpa;
493 memcpy(flow->arp_sha, arp->ar_sha, ETH_ADDR_LEN);
494 memcpy(flow->arp_tha, arp->ar_tha, ETH_ADDR_LEN);
499 /* For every bit of a field that is wildcarded in 'wildcards', sets the
500 * corresponding bit in 'flow' to zero. */
502 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
504 uint32_t *flow_u32 = (uint32_t *) flow;
505 const uint32_t *wc_u32 = (const uint32_t *) &wildcards->masks;
508 for (i = 0; i < FLOW_U32S; i++) {
509 flow_u32[i] &= wc_u32[i];
513 /* Initializes 'fmd' with the metadata found in 'flow'. */
515 flow_get_metadata(const struct flow *flow, struct flow_metadata *fmd)
517 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
519 fmd->tun_id = flow->tunnel.tun_id;
520 fmd->tun_src = flow->tunnel.ip_src;
521 fmd->tun_dst = flow->tunnel.ip_dst;
522 fmd->metadata = flow->metadata;
523 memcpy(fmd->regs, flow->regs, sizeof fmd->regs);
524 fmd->pkt_mark = flow->pkt_mark;
525 fmd->in_port = flow->in_port.ofp_port;
529 flow_to_string(const struct flow *flow)
531 struct ds ds = DS_EMPTY_INITIALIZER;
532 flow_format(&ds, flow);
537 flow_tun_flag_to_string(uint32_t flags)
540 case FLOW_TNL_F_DONT_FRAGMENT:
542 case FLOW_TNL_F_CSUM:
552 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
553 uint32_t flags, char del)
561 uint32_t bit = rightmost_1bit(flags);
564 s = bit_to_string(bit);
566 ds_put_format(ds, "%s%c", s, del);
575 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
581 flow_format(struct ds *ds, const struct flow *flow)
585 match_wc_init(&match, flow);
586 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
590 flow_print(FILE *stream, const struct flow *flow)
592 char *s = flow_to_string(flow);
597 /* flow_wildcards functions. */
599 /* Initializes 'wc' as a set of wildcards that matches every packet. */
601 flow_wildcards_init_catchall(struct flow_wildcards *wc)
603 memset(&wc->masks, 0, sizeof wc->masks);
606 /* Initializes 'wc' as an exact-match set of wildcards; that is, 'wc' does not
607 * wildcard any bits or fields. */
609 flow_wildcards_init_exact(struct flow_wildcards *wc)
611 memset(&wc->masks, 0xff, sizeof wc->masks);
612 memset(wc->masks.zeros, 0, sizeof wc->masks.zeros);
615 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
618 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
620 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
623 for (i = 0; i < FLOW_U32S; i++) {
631 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
632 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
633 * in 'src1' or 'src2' or both. */
635 flow_wildcards_and(struct flow_wildcards *dst,
636 const struct flow_wildcards *src1,
637 const struct flow_wildcards *src2)
639 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
640 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
641 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
644 for (i = 0; i < FLOW_U32S; i++) {
645 dst_u32[i] = src1_u32[i] & src2_u32[i];
649 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
650 * is, a bit or a field is wildcarded in 'dst' if it is neither
651 * wildcarded in 'src1' nor 'src2'. */
653 flow_wildcards_or(struct flow_wildcards *dst,
654 const struct flow_wildcards *src1,
655 const struct flow_wildcards *src2)
657 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
658 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
659 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
662 for (i = 0; i < FLOW_U32S; i++) {
663 dst_u32[i] = src1_u32[i] | src2_u32[i];
667 /* Perform a bitwise OR of miniflow 'src' flow data with the equivalent
668 * fields in 'dst', storing the result in 'dst'. */
670 flow_union_with_miniflow(struct flow *dst, const struct miniflow *src)
672 uint32_t *dst_u32 = (uint32_t *) dst;
677 for (i = 0; i < MINI_N_MAPS; i++) {
680 for (map = src->map[i]; map; map = zero_rightmost_1bit(map)) {
681 dst_u32[raw_ctz(map) + i * 32] |= src->values[ofs++];
686 /* Fold minimask 'mask''s wildcard mask into 'wc's wildcard mask. */
688 flow_wildcards_fold_minimask(struct flow_wildcards *wc,
689 const struct minimask *mask)
691 flow_union_with_miniflow(&wc->masks, &mask->masks);
694 /* Returns a hash of the wildcards in 'wc'. */
696 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
698 return flow_hash(&wc->masks, basis);
701 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
704 flow_wildcards_equal(const struct flow_wildcards *a,
705 const struct flow_wildcards *b)
707 return flow_equal(&a->masks, &b->masks);
710 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
711 * 'b', false otherwise. */
713 flow_wildcards_has_extra(const struct flow_wildcards *a,
714 const struct flow_wildcards *b)
716 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
717 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
720 for (i = 0; i < FLOW_U32S; i++) {
721 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
728 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
729 * in 'wc' do not need to be equal in 'a' and 'b'. */
731 flow_equal_except(const struct flow *a, const struct flow *b,
732 const struct flow_wildcards *wc)
734 const uint32_t *a_u32 = (const uint32_t *) a;
735 const uint32_t *b_u32 = (const uint32_t *) b;
736 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
739 for (i = 0; i < FLOW_U32S; i++) {
740 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
747 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
748 * (A 0-bit indicates a wildcard bit.) */
750 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
752 wc->masks.regs[idx] = mask;
755 /* Hashes 'flow' based on its L2 through L4 protocol information. */
757 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
762 struct in6_addr ipv6_addr;
767 uint8_t eth_addr[ETH_ADDR_LEN];
773 memset(&fields, 0, sizeof fields);
774 for (i = 0; i < ETH_ADDR_LEN; i++) {
775 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
777 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
778 fields.eth_type = flow->dl_type;
780 /* UDP source and destination port are not taken into account because they
781 * will not necessarily be symmetric in a bidirectional flow. */
782 if (fields.eth_type == htons(ETH_TYPE_IP)) {
783 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
784 fields.ip_proto = flow->nw_proto;
785 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
786 fields.tp_port = flow->tp_src ^ flow->tp_dst;
788 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
789 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
790 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
791 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
793 for (i=0; i<16; i++) {
794 ipv6_addr[i] = a[i] ^ b[i];
796 fields.ip_proto = flow->nw_proto;
797 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
798 fields.tp_port = flow->tp_src ^ flow->tp_dst;
801 return jhash_bytes(&fields, sizeof fields, basis);
804 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
806 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
807 enum nx_hash_fields fields)
810 case NX_HASH_FIELDS_ETH_SRC:
811 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
814 case NX_HASH_FIELDS_SYMMETRIC_L4:
815 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
816 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
817 if (flow->dl_type == htons(ETH_TYPE_IP)) {
818 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
819 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
820 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
821 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
822 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
824 if (is_ip_any(flow)) {
825 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
826 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
827 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
829 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
837 /* Hashes the portions of 'flow' designated by 'fields'. */
839 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
844 case NX_HASH_FIELDS_ETH_SRC:
845 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
847 case NX_HASH_FIELDS_SYMMETRIC_L4:
848 return flow_hash_symmetric_l4(flow, basis);
854 /* Returns a string representation of 'fields'. */
856 flow_hash_fields_to_str(enum nx_hash_fields fields)
859 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
860 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
861 default: return "<unknown>";
865 /* Returns true if the value of 'fields' is supported. Otherwise false. */
867 flow_hash_fields_valid(enum nx_hash_fields fields)
869 return fields == NX_HASH_FIELDS_ETH_SRC
870 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
873 /* Returns a hash value for the bits of 'flow' that are active based on
874 * 'wc', given 'basis'. */
876 flow_hash_in_wildcards(const struct flow *flow,
877 const struct flow_wildcards *wc, uint32_t basis)
879 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
880 const uint32_t *flow_u32 = (const uint32_t *) flow;
885 for (i = 0; i < FLOW_U32S; i++) {
886 hash = mhash_add(hash, flow_u32[i] & wc_u32[i]);
888 return mhash_finish(hash, 4 * FLOW_U32S);
891 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
892 * OpenFlow 1.0 "dl_vlan" value:
894 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
895 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
896 * 'flow' previously matched packets without a VLAN header).
898 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
899 * without a VLAN tag.
901 * - Other values of 'vid' should not be used. */
903 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
905 if (vid == htons(OFP10_VLAN_NONE)) {
906 flow->vlan_tci = htons(0);
908 vid &= htons(VLAN_VID_MASK);
909 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
910 flow->vlan_tci |= htons(VLAN_CFI) | vid;
914 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
915 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
918 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
920 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
921 flow->vlan_tci &= ~mask;
922 flow->vlan_tci |= vid & mask;
925 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
928 * This function has no effect on the VLAN ID that 'flow' matches.
930 * After calling this function, 'flow' will not match packets without a VLAN
933 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
936 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
937 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
940 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
941 * as an OpenFlow 1.1 "mpls_label" value. */
943 flow_set_mpls_label(struct flow *flow, ovs_be32 label)
945 set_mpls_lse_label(&flow->mpls_lse, label);
948 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
951 flow_set_mpls_ttl(struct flow *flow, uint8_t ttl)
953 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
956 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
959 flow_set_mpls_tc(struct flow *flow, uint8_t tc)
961 set_mpls_lse_tc(&flow->mpls_lse, tc);
964 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
966 flow_set_mpls_bos(struct flow *flow, uint8_t bos)
968 set_mpls_lse_bos(&flow->mpls_lse, bos);
971 /* Puts into 'b' a packet that flow_extract() would parse as having the given
974 * (This is useful only for testing, obviously, and the packet isn't really
975 * valid. It hasn't got some checksums filled in, for one, and lots of fields
976 * are just zeroed.) */
978 flow_compose(struct ofpbuf *b, const struct flow *flow)
980 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
981 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
982 struct eth_header *eth = b->l2;
983 eth->eth_type = htons(b->size);
987 if (flow->vlan_tci & htons(VLAN_CFI)) {
988 eth_push_vlan(b, flow->vlan_tci);
991 if (flow->dl_type == htons(ETH_TYPE_IP)) {
992 struct ip_header *ip;
994 b->l3 = ip = ofpbuf_put_zeros(b, sizeof *ip);
995 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
996 ip->ip_tos = flow->nw_tos;
997 ip->ip_ttl = flow->nw_ttl;
998 ip->ip_proto = flow->nw_proto;
999 ip->ip_src = flow->nw_src;
1000 ip->ip_dst = flow->nw_dst;
1002 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
1003 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
1004 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
1005 ip->ip_frag_off |= htons(100);
1008 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
1009 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1010 if (flow->nw_proto == IPPROTO_TCP) {
1011 struct tcp_header *tcp;
1013 b->l4 = tcp = ofpbuf_put_zeros(b, sizeof *tcp);
1014 tcp->tcp_src = flow->tp_src;
1015 tcp->tcp_dst = flow->tp_dst;
1016 tcp->tcp_ctl = TCP_CTL(0, 5);
1017 } else if (flow->nw_proto == IPPROTO_UDP) {
1018 struct udp_header *udp;
1020 b->l4 = udp = ofpbuf_put_zeros(b, sizeof *udp);
1021 udp->udp_src = flow->tp_src;
1022 udp->udp_dst = flow->tp_dst;
1023 } else if (flow->nw_proto == IPPROTO_SCTP) {
1024 struct sctp_header *sctp;
1026 b->l4 = sctp = ofpbuf_put_zeros(b, sizeof *sctp);
1027 sctp->sctp_src = flow->tp_src;
1028 sctp->sctp_dst = flow->tp_dst;
1029 } else if (flow->nw_proto == IPPROTO_ICMP) {
1030 struct icmp_header *icmp;
1032 b->l4 = icmp = ofpbuf_put_zeros(b, sizeof *icmp);
1033 icmp->icmp_type = ntohs(flow->tp_src);
1034 icmp->icmp_code = ntohs(flow->tp_dst);
1035 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
1040 ip->ip_tot_len = htons((uint8_t *) b->data + b->size
1041 - (uint8_t *) b->l3);
1042 ip->ip_csum = csum(ip, sizeof *ip);
1043 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1045 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1046 flow->dl_type == htons(ETH_TYPE_RARP)) {
1047 struct arp_eth_header *arp;
1049 b->l3 = arp = ofpbuf_put_zeros(b, sizeof *arp);
1050 arp->ar_hrd = htons(1);
1051 arp->ar_pro = htons(ETH_TYPE_IP);
1052 arp->ar_hln = ETH_ADDR_LEN;
1054 arp->ar_op = htons(flow->nw_proto);
1056 if (flow->nw_proto == ARP_OP_REQUEST ||
1057 flow->nw_proto == ARP_OP_REPLY) {
1058 arp->ar_spa = flow->nw_src;
1059 arp->ar_tpa = flow->nw_dst;
1060 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
1061 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
1065 if (eth_type_mpls(flow->dl_type)) {
1067 push_mpls(b, flow->dl_type, flow->mpls_lse);
1071 /* Compressed flow. */
1074 miniflow_n_values(const struct miniflow *flow)
1079 for (i = 0; i < MINI_N_MAPS; i++) {
1080 n += popcount(flow->map[i]);
1086 miniflow_alloc_values(struct miniflow *flow, int n)
1088 if (n <= MINI_N_INLINE) {
1089 return flow->inline_values;
1091 COVERAGE_INC(miniflow_malloc);
1092 return xmalloc(n * sizeof *flow->values);
1096 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1097 * with miniflow_destroy(). */
1099 miniflow_init(struct miniflow *dst, const struct flow *src)
1101 const uint32_t *src_u32 = (const uint32_t *) src;
1106 /* Initialize dst->map, counting the number of nonzero elements. */
1108 memset(dst->map, 0, sizeof dst->map);
1109 for (i = 0; i < FLOW_U32S; i++) {
1111 dst->map[i / 32] |= 1u << (i % 32);
1116 /* Initialize dst->values. */
1117 dst->values = miniflow_alloc_values(dst, n);
1119 for (i = 0; i < MINI_N_MAPS; i++) {
1122 for (map = dst->map[i]; map; map = zero_rightmost_1bit(map)) {
1123 dst->values[ofs++] = src_u32[raw_ctz(map) + i * 32];
1128 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1129 * with miniflow_destroy(). */
1131 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
1133 int n = miniflow_n_values(src);
1134 memcpy(dst->map, src->map, sizeof dst->map);
1135 dst->values = miniflow_alloc_values(dst, n);
1136 memcpy(dst->values, src->values, n * sizeof *dst->values);
1139 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
1140 * itself resides; the caller is responsible for that. */
1142 miniflow_destroy(struct miniflow *flow)
1144 if (flow->values != flow->inline_values) {
1149 /* Initializes 'dst' as a copy of 'src'. */
1151 miniflow_expand(const struct miniflow *src, struct flow *dst)
1153 memset(dst, 0, sizeof *dst);
1154 flow_union_with_miniflow(dst, src);
1157 static const uint32_t *
1158 miniflow_get__(const struct miniflow *flow, unsigned int u32_ofs)
1160 if (!(flow->map[u32_ofs / 32] & (1u << (u32_ofs % 32)))) {
1161 static const uint32_t zero = 0;
1164 const uint32_t *p = flow->values;
1166 BUILD_ASSERT(MINI_N_MAPS == 2);
1168 p += popcount(flow->map[0] & ((1u << u32_ofs) - 1));
1170 p += popcount(flow->map[0]);
1171 p += popcount(flow->map[1] & ((1u << (u32_ofs - 32)) - 1));
1177 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1178 * were expanded into a "struct flow". */
1180 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1182 return *miniflow_get__(flow, u32_ofs);
1185 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
1186 * expanded into a "struct flow". */
1188 miniflow_get_be16(const struct miniflow *flow, unsigned int u8_ofs)
1190 const uint32_t *u32p = miniflow_get__(flow, u8_ofs / 4);
1191 const ovs_be16 *be16p = (const ovs_be16 *) u32p;
1192 return be16p[u8_ofs % 4 != 0];
1195 /* Returns the VID within the vlan_tci member of the "struct flow" represented
1198 miniflow_get_vid(const struct miniflow *flow)
1200 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
1201 return vlan_tci_to_vid(tci);
1204 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1206 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1210 for (i = 0; i < MINI_N_MAPS; i++) {
1211 if (a->map[i] != b->map[i]) {
1216 return !memcmp(a->values, b->values,
1217 miniflow_n_values(a) * sizeof *a->values);
1220 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1221 * in 'mask', false if they differ. */
1223 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1224 const struct minimask *mask)
1229 p = mask->masks.values;
1230 for (i = 0; i < MINI_N_MAPS; i++) {
1233 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1234 int ofs = raw_ctz(map) + i * 32;
1236 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p) {
1246 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1247 * in 'mask', false if they differ. */
1249 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1250 const struct minimask *mask)
1252 const uint32_t *b_u32 = (const uint32_t *) b;
1256 p = mask->masks.values;
1257 for (i = 0; i < MINI_N_MAPS; i++) {
1260 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1261 int ofs = raw_ctz(map) + i * 32;
1263 if ((miniflow_get(a, ofs) ^ b_u32[ofs]) & *p) {
1273 /* Returns a hash value for 'flow', given 'basis'. */
1275 miniflow_hash(const struct miniflow *flow, uint32_t basis)
1277 BUILD_ASSERT_DECL(MINI_N_MAPS == 2);
1278 return hash_3words(flow->map[0], flow->map[1],
1279 hash_words(flow->values, miniflow_n_values(flow),
1283 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1284 * 'mask', given 'basis'.
1286 * The hash values returned by this function are the same as those returned by
1287 * flow_hash_in_minimask(), only the form of the arguments differ. */
1289 miniflow_hash_in_minimask(const struct miniflow *flow,
1290 const struct minimask *mask, uint32_t basis)
1292 const uint32_t *p = mask->masks.values;
1297 for (i = 0; i < MINI_N_MAPS; i++) {
1300 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1301 int ofs = raw_ctz(map) + i * 32;
1303 hash = mhash_add(hash, miniflow_get(flow, ofs) & *p);
1308 return mhash_finish(hash, (p - mask->masks.values) * 4);
1311 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1312 * 'mask', given 'basis'.
1314 * The hash values returned by this function are the same as those returned by
1315 * miniflow_hash_in_minimask(), only the form of the arguments differ. */
1317 flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
1320 const uint32_t *flow_u32 = (const uint32_t *) flow;
1321 const uint32_t *p = mask->masks.values;
1326 for (i = 0; i < MINI_N_MAPS; i++) {
1329 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1330 int ofs = raw_ctz(map) + i * 32;
1332 hash = mhash_add(hash, flow_u32[ofs] & *p);
1337 return mhash_finish(hash, (p - mask->masks.values) * 4);
1340 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1341 * with minimask_destroy(). */
1343 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1345 miniflow_init(&mask->masks, &wc->masks);
1348 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1349 * with minimask_destroy(). */
1351 minimask_clone(struct minimask *dst, const struct minimask *src)
1353 miniflow_clone(&dst->masks, &src->masks);
1356 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1358 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1359 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1361 minimask_combine(struct minimask *dst_,
1362 const struct minimask *a_, const struct minimask *b_,
1363 uint32_t storage[FLOW_U32S])
1365 struct miniflow *dst = &dst_->masks;
1366 const struct miniflow *a = &a_->masks;
1367 const struct miniflow *b = &b_->masks;
1371 dst->values = storage;
1372 for (i = 0; i < MINI_N_MAPS; i++) {
1376 for (map = a->map[i] & b->map[i]; map;
1377 map = zero_rightmost_1bit(map)) {
1378 int ofs = raw_ctz(map) + i * 32;
1379 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1382 dst->map[i] |= rightmost_1bit(map);
1383 dst->values[n++] = mask;
1389 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1390 * itself resides; the caller is responsible for that. */
1392 minimask_destroy(struct minimask *mask)
1394 miniflow_destroy(&mask->masks);
1397 /* Initializes 'dst' as a copy of 'src'. */
1399 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1401 miniflow_expand(&mask->masks, &wc->masks);
1404 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1405 * were expanded into a "struct flow_wildcards". */
1407 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1409 return miniflow_get(&mask->masks, u32_ofs);
1412 /* Returns the VID mask within the vlan_tci member of the "struct
1413 * flow_wildcards" represented by 'mask'. */
1415 minimask_get_vid_mask(const struct minimask *mask)
1417 return miniflow_get_vid(&mask->masks);
1420 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1422 minimask_equal(const struct minimask *a, const struct minimask *b)
1424 return miniflow_equal(&a->masks, &b->masks);
1427 /* Returns a hash value for 'mask', given 'basis'. */
1429 minimask_hash(const struct minimask *mask, uint32_t basis)
1431 return miniflow_hash(&mask->masks, basis);
1434 /* Returns true if at least one bit is wildcarded in 'a_' but not in 'b_',
1435 * false otherwise. */
1437 minimask_has_extra(const struct minimask *a_, const struct minimask *b_)
1439 const struct miniflow *a = &a_->masks;
1440 const struct miniflow *b = &b_->masks;
1443 for (i = 0; i < MINI_N_MAPS; i++) {
1446 for (map = a->map[i] | b->map[i]; map;
1447 map = zero_rightmost_1bit(map)) {
1448 int ofs = raw_ctz(map) + i * 32;
1449 uint32_t a_u32 = miniflow_get(a, ofs);
1450 uint32_t b_u32 = miniflow_get(b, ofs);
1452 if ((a_u32 & b_u32) != b_u32) {
1461 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
1464 minimask_is_catchall(const struct minimask *mask_)
1466 const struct miniflow *mask = &mask_->masks;
1468 BUILD_ASSERT(MINI_N_MAPS == 2);
1469 return !(mask->map[0] | mask->map[1]);