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 icmp_header *
85 pull_icmp(struct ofpbuf *packet)
87 return ofpbuf_try_pull(packet, ICMP_HEADER_LEN);
90 static struct icmp6_hdr *
91 pull_icmpv6(struct ofpbuf *packet)
93 return ofpbuf_try_pull(packet, sizeof(struct icmp6_hdr));
97 parse_mpls(struct ofpbuf *b, struct flow *flow)
101 while ((mh = ofpbuf_try_pull(b, sizeof *mh))) {
102 if (flow->mpls_depth++ == 0) {
103 flow->mpls_lse = mh->mpls_lse;
105 if (mh->mpls_lse & htonl(MPLS_BOS_MASK)) {
112 parse_vlan(struct ofpbuf *b, struct flow *flow)
115 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
119 if (b->size >= sizeof(struct qtag_prefix) + sizeof(ovs_be16)) {
120 struct qtag_prefix *qp = ofpbuf_pull(b, sizeof *qp);
121 flow->vlan_tci = qp->tci | htons(VLAN_CFI);
126 parse_ethertype(struct ofpbuf *b)
128 struct llc_snap_header *llc;
131 proto = *(ovs_be16 *) ofpbuf_pull(b, sizeof proto);
132 if (ntohs(proto) >= ETH_TYPE_MIN) {
136 if (b->size < sizeof *llc) {
137 return htons(FLOW_DL_TYPE_NONE);
141 if (llc->llc.llc_dsap != LLC_DSAP_SNAP
142 || llc->llc.llc_ssap != LLC_SSAP_SNAP
143 || llc->llc.llc_cntl != LLC_CNTL_SNAP
144 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
145 sizeof llc->snap.snap_org)) {
146 return htons(FLOW_DL_TYPE_NONE);
149 ofpbuf_pull(b, sizeof *llc);
151 if (ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN) {
152 return llc->snap.snap_type;
155 return htons(FLOW_DL_TYPE_NONE);
159 parse_ipv6(struct ofpbuf *packet, struct flow *flow)
161 const struct ip6_hdr *nh;
165 nh = ofpbuf_try_pull(packet, sizeof *nh);
170 nexthdr = nh->ip6_nxt;
172 flow->ipv6_src = nh->ip6_src;
173 flow->ipv6_dst = nh->ip6_dst;
175 tc_flow = get_unaligned_be32(&nh->ip6_flow);
176 flow->nw_tos = ntohl(tc_flow) >> 20;
177 flow->ipv6_label = tc_flow & htonl(IPV6_LABEL_MASK);
178 flow->nw_ttl = nh->ip6_hlim;
179 flow->nw_proto = IPPROTO_NONE;
182 if ((nexthdr != IPPROTO_HOPOPTS)
183 && (nexthdr != IPPROTO_ROUTING)
184 && (nexthdr != IPPROTO_DSTOPTS)
185 && (nexthdr != IPPROTO_AH)
186 && (nexthdr != IPPROTO_FRAGMENT)) {
187 /* It's either a terminal header (e.g., TCP, UDP) or one we
188 * don't understand. In either case, we're done with the
189 * packet, so use it to fill in 'nw_proto'. */
193 /* We only verify that at least 8 bytes of the next header are
194 * available, but many of these headers are longer. Ensure that
195 * accesses within the extension header are within those first 8
196 * bytes. All extension headers are required to be at least 8
198 if (packet->size < 8) {
202 if ((nexthdr == IPPROTO_HOPOPTS)
203 || (nexthdr == IPPROTO_ROUTING)
204 || (nexthdr == IPPROTO_DSTOPTS)) {
205 /* These headers, while different, have the fields we care about
206 * in the same location and with the same interpretation. */
207 const struct ip6_ext *ext_hdr = packet->data;
208 nexthdr = ext_hdr->ip6e_nxt;
209 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 1) * 8)) {
212 } else if (nexthdr == IPPROTO_AH) {
213 /* A standard AH definition isn't available, but the fields
214 * we care about are in the same location as the generic
215 * option header--only the header length is calculated
217 const struct ip6_ext *ext_hdr = packet->data;
218 nexthdr = ext_hdr->ip6e_nxt;
219 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 2) * 4)) {
222 } else if (nexthdr == IPPROTO_FRAGMENT) {
223 const struct ip6_frag *frag_hdr = packet->data;
225 nexthdr = frag_hdr->ip6f_nxt;
226 if (!ofpbuf_try_pull(packet, sizeof *frag_hdr)) {
230 /* We only process the first fragment. */
231 if (frag_hdr->ip6f_offlg != htons(0)) {
232 flow->nw_frag = FLOW_NW_FRAG_ANY;
233 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
234 flow->nw_frag |= FLOW_NW_FRAG_LATER;
235 nexthdr = IPPROTO_FRAGMENT;
242 flow->nw_proto = nexthdr;
247 parse_tcp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
249 const struct tcp_header *tcp = pull_tcp(b);
251 flow->tp_src = tcp->tcp_src;
252 flow->tp_dst = tcp->tcp_dst;
253 packet->l7 = b->data;
258 parse_udp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
260 const struct udp_header *udp = pull_udp(b);
262 flow->tp_src = udp->udp_src;
263 flow->tp_dst = udp->udp_dst;
264 packet->l7 = b->data;
269 parse_icmpv6(struct ofpbuf *b, struct flow *flow)
271 const struct icmp6_hdr *icmp = pull_icmpv6(b);
277 /* The ICMPv6 type and code fields use the 16-bit transport port
278 * fields, so we need to store them in 16-bit network byte order. */
279 flow->tp_src = htons(icmp->icmp6_type);
280 flow->tp_dst = htons(icmp->icmp6_code);
282 if (icmp->icmp6_code == 0 &&
283 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
284 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
285 const struct in6_addr *nd_target;
287 nd_target = ofpbuf_try_pull(b, sizeof *nd_target);
291 flow->nd_target = *nd_target;
293 while (b->size >= 8) {
294 /* The minimum size of an option is 8 bytes, which also is
295 * the size of Ethernet link-layer options. */
296 const struct nd_opt_hdr *nd_opt = b->data;
297 int opt_len = nd_opt->nd_opt_len * 8;
299 if (!opt_len || opt_len > b->size) {
303 /* Store the link layer address if the appropriate option is
304 * provided. It is considered an error if the same link
305 * layer option is specified twice. */
306 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
308 if (eth_addr_is_zero(flow->arp_sha)) {
309 memcpy(flow->arp_sha, nd_opt + 1, ETH_ADDR_LEN);
313 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
315 if (eth_addr_is_zero(flow->arp_tha)) {
316 memcpy(flow->arp_tha, nd_opt + 1, ETH_ADDR_LEN);
322 if (!ofpbuf_try_pull(b, opt_len)) {
331 memset(&flow->nd_target, 0, sizeof(flow->nd_target));
332 memset(flow->arp_sha, 0, sizeof(flow->arp_sha));
333 memset(flow->arp_tha, 0, sizeof(flow->arp_tha));
339 /* Initializes 'flow' members from 'packet', 'skb_priority', 'tnl', and
342 * Initializes 'packet' header pointers as follows:
344 * - packet->l2 to the start of the Ethernet header.
346 * - packet->l2_5 to the start of the MPLS shim header.
348 * - packet->l3 to just past the Ethernet header, or just past the
349 * vlan_header if one is present, to the first byte of the payload of the
352 * - packet->l4 to just past the IPv4 header, if one is present and has a
353 * correct length, and otherwise NULL.
355 * - packet->l7 to just past the TCP or UDP or ICMP header, if one is
356 * present and has a correct length, and otherwise NULL.
359 flow_extract(struct ofpbuf *packet, uint32_t skb_priority, uint32_t pkt_mark,
360 const struct flow_tnl *tnl, const union flow_in_port *in_port,
363 struct ofpbuf b = *packet;
364 struct eth_header *eth;
366 COVERAGE_INC(flow_extract);
368 memset(flow, 0, sizeof *flow);
371 ovs_assert(tnl != &flow->tunnel);
375 flow->in_port = *in_port;
377 flow->skb_priority = skb_priority;
378 flow->pkt_mark = pkt_mark;
386 if (b.size < sizeof *eth) {
392 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
393 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
395 /* dl_type, vlan_tci. */
396 ofpbuf_pull(&b, ETH_ADDR_LEN * 2);
397 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
398 parse_vlan(&b, flow);
400 flow->dl_type = parse_ethertype(&b);
402 /* Parse mpls, copy l3 ttl. */
403 if (eth_type_mpls(flow->dl_type)) {
404 packet->l2_5 = b.data;
405 parse_mpls(&b, flow);
410 if (flow->dl_type == htons(ETH_TYPE_IP)) {
411 const struct ip_header *nh = pull_ip(&b);
415 flow->nw_src = get_unaligned_be32(&nh->ip_src);
416 flow->nw_dst = get_unaligned_be32(&nh->ip_dst);
417 flow->nw_proto = nh->ip_proto;
419 flow->nw_tos = nh->ip_tos;
420 if (IP_IS_FRAGMENT(nh->ip_frag_off)) {
421 flow->nw_frag = FLOW_NW_FRAG_ANY;
422 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
423 flow->nw_frag |= FLOW_NW_FRAG_LATER;
426 flow->nw_ttl = nh->ip_ttl;
428 if (!(nh->ip_frag_off & htons(IP_FRAG_OFF_MASK))) {
429 if (flow->nw_proto == IPPROTO_TCP) {
430 parse_tcp(packet, &b, flow);
431 } else if (flow->nw_proto == IPPROTO_UDP) {
432 parse_udp(packet, &b, flow);
433 } else if (flow->nw_proto == IPPROTO_ICMP) {
434 const struct icmp_header *icmp = pull_icmp(&b);
436 flow->tp_src = htons(icmp->icmp_type);
437 flow->tp_dst = htons(icmp->icmp_code);
443 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
444 if (parse_ipv6(&b, flow)) {
449 if (flow->nw_proto == IPPROTO_TCP) {
450 parse_tcp(packet, &b, flow);
451 } else if (flow->nw_proto == IPPROTO_UDP) {
452 parse_udp(packet, &b, flow);
453 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
454 if (parse_icmpv6(&b, flow)) {
458 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
459 flow->dl_type == htons(ETH_TYPE_RARP)) {
460 const struct arp_eth_header *arp = pull_arp(&b);
461 if (arp && arp->ar_hrd == htons(1)
462 && arp->ar_pro == htons(ETH_TYPE_IP)
463 && arp->ar_hln == ETH_ADDR_LEN
464 && arp->ar_pln == 4) {
465 /* We only match on the lower 8 bits of the opcode. */
466 if (ntohs(arp->ar_op) <= 0xff) {
467 flow->nw_proto = ntohs(arp->ar_op);
470 flow->nw_src = arp->ar_spa;
471 flow->nw_dst = arp->ar_tpa;
472 memcpy(flow->arp_sha, arp->ar_sha, ETH_ADDR_LEN);
473 memcpy(flow->arp_tha, arp->ar_tha, ETH_ADDR_LEN);
478 /* For every bit of a field that is wildcarded in 'wildcards', sets the
479 * corresponding bit in 'flow' to zero. */
481 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
483 uint32_t *flow_u32 = (uint32_t *) flow;
484 const uint32_t *wc_u32 = (const uint32_t *) &wildcards->masks;
487 for (i = 0; i < FLOW_U32S; i++) {
488 flow_u32[i] &= wc_u32[i];
492 /* Initializes 'fmd' with the metadata found in 'flow'. */
494 flow_get_metadata(const struct flow *flow, struct flow_metadata *fmd)
496 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
498 fmd->tun_id = flow->tunnel.tun_id;
499 fmd->tun_src = flow->tunnel.ip_src;
500 fmd->tun_dst = flow->tunnel.ip_dst;
501 fmd->metadata = flow->metadata;
502 memcpy(fmd->regs, flow->regs, sizeof fmd->regs);
503 fmd->pkt_mark = flow->pkt_mark;
504 fmd->in_port = flow->in_port.ofp_port;
508 flow_to_string(const struct flow *flow)
510 struct ds ds = DS_EMPTY_INITIALIZER;
511 flow_format(&ds, flow);
516 flow_tun_flag_to_string(uint32_t flags)
519 case FLOW_TNL_F_DONT_FRAGMENT:
521 case FLOW_TNL_F_CSUM:
531 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
532 uint32_t flags, char del)
540 uint32_t bit = rightmost_1bit(flags);
543 s = bit_to_string(bit);
545 ds_put_format(ds, "%s%c", s, del);
554 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
560 flow_format(struct ds *ds, const struct flow *flow)
564 match_wc_init(&match, flow);
565 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
569 flow_print(FILE *stream, const struct flow *flow)
571 char *s = flow_to_string(flow);
576 /* flow_wildcards functions. */
578 /* Initializes 'wc' as a set of wildcards that matches every packet. */
580 flow_wildcards_init_catchall(struct flow_wildcards *wc)
582 memset(&wc->masks, 0, sizeof wc->masks);
585 /* Initializes 'wc' as an exact-match set of wildcards; that is, 'wc' does not
586 * wildcard any bits or fields. */
588 flow_wildcards_init_exact(struct flow_wildcards *wc)
590 memset(&wc->masks, 0xff, sizeof wc->masks);
591 memset(wc->masks.zeros, 0, sizeof wc->masks.zeros);
594 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
597 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
599 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
602 for (i = 0; i < FLOW_U32S; i++) {
610 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
611 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
612 * in 'src1' or 'src2' or both. */
614 flow_wildcards_and(struct flow_wildcards *dst,
615 const struct flow_wildcards *src1,
616 const struct flow_wildcards *src2)
618 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
619 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
620 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
623 for (i = 0; i < FLOW_U32S; i++) {
624 dst_u32[i] = src1_u32[i] & src2_u32[i];
628 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
629 * is, a bit or a field is wildcarded in 'dst' if it is neither
630 * wildcarded in 'src1' nor 'src2'. */
632 flow_wildcards_or(struct flow_wildcards *dst,
633 const struct flow_wildcards *src1,
634 const struct flow_wildcards *src2)
636 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
637 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
638 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
641 for (i = 0; i < FLOW_U32S; i++) {
642 dst_u32[i] = src1_u32[i] | src2_u32[i];
646 /* Perform a bitwise OR of miniflow 'src' flow data with the equivalent
647 * fields in 'dst', storing the result in 'dst'. */
649 flow_union_with_miniflow(struct flow *dst, const struct miniflow *src)
651 uint32_t *dst_u32 = (uint32_t *) dst;
656 for (i = 0; i < MINI_N_MAPS; i++) {
659 for (map = src->map[i]; map; map = zero_rightmost_1bit(map)) {
660 dst_u32[raw_ctz(map) + i * 32] |= src->values[ofs++];
665 /* Fold minimask 'mask''s wildcard mask into 'wc's wildcard mask. */
667 flow_wildcards_fold_minimask(struct flow_wildcards *wc,
668 const struct minimask *mask)
670 flow_union_with_miniflow(&wc->masks, &mask->masks);
673 /* Returns a hash of the wildcards in 'wc'. */
675 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
677 return flow_hash(&wc->masks, basis);
680 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
683 flow_wildcards_equal(const struct flow_wildcards *a,
684 const struct flow_wildcards *b)
686 return flow_equal(&a->masks, &b->masks);
689 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
690 * 'b', false otherwise. */
692 flow_wildcards_has_extra(const struct flow_wildcards *a,
693 const struct flow_wildcards *b)
695 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
696 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
699 for (i = 0; i < FLOW_U32S; i++) {
700 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
707 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
708 * in 'wc' do not need to be equal in 'a' and 'b'. */
710 flow_equal_except(const struct flow *a, const struct flow *b,
711 const struct flow_wildcards *wc)
713 const uint32_t *a_u32 = (const uint32_t *) a;
714 const uint32_t *b_u32 = (const uint32_t *) b;
715 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
718 for (i = 0; i < FLOW_U32S; i++) {
719 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
726 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
727 * (A 0-bit indicates a wildcard bit.) */
729 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
731 wc->masks.regs[idx] = mask;
734 /* Hashes 'flow' based on its L2 through L4 protocol information. */
736 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
741 struct in6_addr ipv6_addr;
746 uint8_t eth_addr[ETH_ADDR_LEN];
752 memset(&fields, 0, sizeof fields);
753 for (i = 0; i < ETH_ADDR_LEN; i++) {
754 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
756 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
757 fields.eth_type = flow->dl_type;
759 /* UDP source and destination port are not taken into account because they
760 * will not necessarily be symmetric in a bidirectional flow. */
761 if (fields.eth_type == htons(ETH_TYPE_IP)) {
762 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
763 fields.ip_proto = flow->nw_proto;
764 if (fields.ip_proto == IPPROTO_TCP) {
765 fields.tp_port = flow->tp_src ^ flow->tp_dst;
767 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
768 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
769 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
770 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
772 for (i=0; i<16; i++) {
773 ipv6_addr[i] = a[i] ^ b[i];
775 fields.ip_proto = flow->nw_proto;
776 if (fields.ip_proto == IPPROTO_TCP) {
777 fields.tp_port = flow->tp_src ^ flow->tp_dst;
780 return jhash_bytes(&fields, sizeof fields, basis);
783 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
785 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
786 enum nx_hash_fields fields)
789 case NX_HASH_FIELDS_ETH_SRC:
790 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
793 case NX_HASH_FIELDS_SYMMETRIC_L4:
794 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
795 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
796 if (flow->dl_type == htons(ETH_TYPE_IP)) {
797 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
798 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
799 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
800 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
801 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
803 if (is_ip_any(flow)) {
804 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
805 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
806 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
808 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
816 /* Hashes the portions of 'flow' designated by 'fields'. */
818 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
823 case NX_HASH_FIELDS_ETH_SRC:
824 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
826 case NX_HASH_FIELDS_SYMMETRIC_L4:
827 return flow_hash_symmetric_l4(flow, basis);
833 /* Returns a string representation of 'fields'. */
835 flow_hash_fields_to_str(enum nx_hash_fields fields)
838 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
839 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
840 default: return "<unknown>";
844 /* Returns true if the value of 'fields' is supported. Otherwise false. */
846 flow_hash_fields_valid(enum nx_hash_fields fields)
848 return fields == NX_HASH_FIELDS_ETH_SRC
849 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
852 /* Returns a hash value for the bits of 'flow' that are active based on
853 * 'wc', given 'basis'. */
855 flow_hash_in_wildcards(const struct flow *flow,
856 const struct flow_wildcards *wc, uint32_t basis)
858 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
859 const uint32_t *flow_u32 = (const uint32_t *) flow;
864 for (i = 0; i < FLOW_U32S; i++) {
865 hash = mhash_add(hash, flow_u32[i] & wc_u32[i]);
867 return mhash_finish(hash, 4 * FLOW_U32S);
870 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
871 * OpenFlow 1.0 "dl_vlan" value:
873 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
874 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
875 * 'flow' previously matched packets without a VLAN header).
877 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
878 * without a VLAN tag.
880 * - Other values of 'vid' should not be used. */
882 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
884 if (vid == htons(OFP10_VLAN_NONE)) {
885 flow->vlan_tci = htons(0);
887 vid &= htons(VLAN_VID_MASK);
888 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
889 flow->vlan_tci |= htons(VLAN_CFI) | vid;
893 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
894 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
897 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
899 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
900 flow->vlan_tci &= ~mask;
901 flow->vlan_tci |= vid & mask;
904 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
907 * This function has no effect on the VLAN ID that 'flow' matches.
909 * After calling this function, 'flow' will not match packets without a VLAN
912 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
915 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
916 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
919 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
920 * as an OpenFlow 1.1 "mpls_label" value. */
922 flow_set_mpls_label(struct flow *flow, ovs_be32 label)
924 set_mpls_lse_label(&flow->mpls_lse, label);
927 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
930 flow_set_mpls_ttl(struct flow *flow, uint8_t ttl)
932 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
935 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
938 flow_set_mpls_tc(struct flow *flow, uint8_t tc)
940 set_mpls_lse_tc(&flow->mpls_lse, tc);
943 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
945 flow_set_mpls_bos(struct flow *flow, uint8_t bos)
947 set_mpls_lse_bos(&flow->mpls_lse, bos);
950 /* Puts into 'b' a packet that flow_extract() would parse as having the given
953 * (This is useful only for testing, obviously, and the packet isn't really
954 * valid. It hasn't got some checksums filled in, for one, and lots of fields
955 * are just zeroed.) */
957 flow_compose(struct ofpbuf *b, const struct flow *flow)
959 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
960 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
961 struct eth_header *eth = b->l2;
962 eth->eth_type = htons(b->size);
966 if (flow->vlan_tci & htons(VLAN_CFI)) {
967 eth_push_vlan(b, flow->vlan_tci);
970 if (flow->dl_type == htons(ETH_TYPE_IP)) {
971 struct ip_header *ip;
973 b->l3 = ip = ofpbuf_put_zeros(b, sizeof *ip);
974 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
975 ip->ip_tos = flow->nw_tos;
976 ip->ip_ttl = flow->nw_ttl;
977 ip->ip_proto = flow->nw_proto;
978 ip->ip_src = flow->nw_src;
979 ip->ip_dst = flow->nw_dst;
981 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
982 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
983 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
984 ip->ip_frag_off |= htons(100);
987 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
988 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
989 if (flow->nw_proto == IPPROTO_TCP) {
990 struct tcp_header *tcp;
992 b->l4 = tcp = ofpbuf_put_zeros(b, sizeof *tcp);
993 tcp->tcp_src = flow->tp_src;
994 tcp->tcp_dst = flow->tp_dst;
995 tcp->tcp_ctl = TCP_CTL(0, 5);
996 } else if (flow->nw_proto == IPPROTO_UDP) {
997 struct udp_header *udp;
999 b->l4 = udp = ofpbuf_put_zeros(b, sizeof *udp);
1000 udp->udp_src = flow->tp_src;
1001 udp->udp_dst = flow->tp_dst;
1002 } else if (flow->nw_proto == IPPROTO_ICMP) {
1003 struct icmp_header *icmp;
1005 b->l4 = icmp = ofpbuf_put_zeros(b, sizeof *icmp);
1006 icmp->icmp_type = ntohs(flow->tp_src);
1007 icmp->icmp_code = ntohs(flow->tp_dst);
1008 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
1013 ip->ip_tot_len = htons((uint8_t *) b->data + b->size
1014 - (uint8_t *) b->l3);
1015 ip->ip_csum = csum(ip, sizeof *ip);
1016 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1018 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1019 flow->dl_type == htons(ETH_TYPE_RARP)) {
1020 struct arp_eth_header *arp;
1022 b->l3 = arp = ofpbuf_put_zeros(b, sizeof *arp);
1023 arp->ar_hrd = htons(1);
1024 arp->ar_pro = htons(ETH_TYPE_IP);
1025 arp->ar_hln = ETH_ADDR_LEN;
1027 arp->ar_op = htons(flow->nw_proto);
1029 if (flow->nw_proto == ARP_OP_REQUEST ||
1030 flow->nw_proto == ARP_OP_REPLY) {
1031 arp->ar_spa = flow->nw_src;
1032 arp->ar_tpa = flow->nw_dst;
1033 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
1034 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
1038 if (eth_type_mpls(flow->dl_type)) {
1040 push_mpls(b, flow->dl_type, flow->mpls_lse);
1044 /* Compressed flow. */
1047 miniflow_n_values(const struct miniflow *flow)
1052 for (i = 0; i < MINI_N_MAPS; i++) {
1053 n += popcount(flow->map[i]);
1059 miniflow_alloc_values(struct miniflow *flow, int n)
1061 if (n <= MINI_N_INLINE) {
1062 return flow->inline_values;
1064 COVERAGE_INC(miniflow_malloc);
1065 return xmalloc(n * sizeof *flow->values);
1069 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1070 * with miniflow_destroy(). */
1072 miniflow_init(struct miniflow *dst, const struct flow *src)
1074 const uint32_t *src_u32 = (const uint32_t *) src;
1079 /* Initialize dst->map, counting the number of nonzero elements. */
1081 memset(dst->map, 0, sizeof dst->map);
1082 for (i = 0; i < FLOW_U32S; i++) {
1084 dst->map[i / 32] |= 1u << (i % 32);
1089 /* Initialize dst->values. */
1090 dst->values = miniflow_alloc_values(dst, n);
1092 for (i = 0; i < MINI_N_MAPS; i++) {
1095 for (map = dst->map[i]; map; map = zero_rightmost_1bit(map)) {
1096 dst->values[ofs++] = src_u32[raw_ctz(map) + i * 32];
1101 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1102 * with miniflow_destroy(). */
1104 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
1106 int n = miniflow_n_values(src);
1107 memcpy(dst->map, src->map, sizeof dst->map);
1108 dst->values = miniflow_alloc_values(dst, n);
1109 memcpy(dst->values, src->values, n * sizeof *dst->values);
1112 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
1113 * itself resides; the caller is responsible for that. */
1115 miniflow_destroy(struct miniflow *flow)
1117 if (flow->values != flow->inline_values) {
1122 /* Initializes 'dst' as a copy of 'src'. */
1124 miniflow_expand(const struct miniflow *src, struct flow *dst)
1126 memset(dst, 0, sizeof *dst);
1127 flow_union_with_miniflow(dst, src);
1130 static const uint32_t *
1131 miniflow_get__(const struct miniflow *flow, unsigned int u32_ofs)
1133 if (!(flow->map[u32_ofs / 32] & (1u << (u32_ofs % 32)))) {
1134 static const uint32_t zero = 0;
1137 const uint32_t *p = flow->values;
1139 BUILD_ASSERT(MINI_N_MAPS == 2);
1141 p += popcount(flow->map[0] & ((1u << u32_ofs) - 1));
1143 p += popcount(flow->map[0]);
1144 p += popcount(flow->map[1] & ((1u << (u32_ofs - 32)) - 1));
1150 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1151 * were expanded into a "struct flow". */
1153 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1155 return *miniflow_get__(flow, u32_ofs);
1158 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
1159 * expanded into a "struct flow". */
1161 miniflow_get_be16(const struct miniflow *flow, unsigned int u8_ofs)
1163 const uint32_t *u32p = miniflow_get__(flow, u8_ofs / 4);
1164 const ovs_be16 *be16p = (const ovs_be16 *) u32p;
1165 return be16p[u8_ofs % 4 != 0];
1168 /* Returns the VID within the vlan_tci member of the "struct flow" represented
1171 miniflow_get_vid(const struct miniflow *flow)
1173 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
1174 return vlan_tci_to_vid(tci);
1177 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1179 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1183 for (i = 0; i < MINI_N_MAPS; i++) {
1184 if (a->map[i] != b->map[i]) {
1189 return !memcmp(a->values, b->values,
1190 miniflow_n_values(a) * sizeof *a->values);
1193 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1194 * in 'mask', false if they differ. */
1196 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1197 const struct minimask *mask)
1202 p = mask->masks.values;
1203 for (i = 0; i < MINI_N_MAPS; i++) {
1206 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1207 int ofs = raw_ctz(map) + i * 32;
1209 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p) {
1219 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1220 * in 'mask', false if they differ. */
1222 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1223 const struct minimask *mask)
1225 const uint32_t *b_u32 = (const uint32_t *) b;
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) ^ b_u32[ofs]) & *p) {
1246 /* Returns a hash value for 'flow', given 'basis'. */
1248 miniflow_hash(const struct miniflow *flow, uint32_t basis)
1250 BUILD_ASSERT_DECL(MINI_N_MAPS == 2);
1251 return hash_3words(flow->map[0], flow->map[1],
1252 hash_words(flow->values, miniflow_n_values(flow),
1256 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1257 * 'mask', given 'basis'.
1259 * The hash values returned by this function are the same as those returned by
1260 * flow_hash_in_minimask(), only the form of the arguments differ. */
1262 miniflow_hash_in_minimask(const struct miniflow *flow,
1263 const struct minimask *mask, uint32_t basis)
1265 const uint32_t *p = mask->masks.values;
1270 for (i = 0; i < MINI_N_MAPS; i++) {
1273 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1274 int ofs = raw_ctz(map) + i * 32;
1276 hash = mhash_add(hash, miniflow_get(flow, ofs) & *p);
1281 return mhash_finish(hash, (p - mask->masks.values) * 4);
1284 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1285 * 'mask', given 'basis'.
1287 * The hash values returned by this function are the same as those returned by
1288 * miniflow_hash_in_minimask(), only the form of the arguments differ. */
1290 flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
1293 const uint32_t *flow_u32 = (const uint32_t *) flow;
1294 const uint32_t *p = mask->masks.values;
1299 for (i = 0; i < MINI_N_MAPS; i++) {
1302 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1303 int ofs = raw_ctz(map) + i * 32;
1305 hash = mhash_add(hash, flow_u32[ofs] & *p);
1310 return mhash_finish(hash, (p - mask->masks.values) * 4);
1313 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1314 * with minimask_destroy(). */
1316 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1318 miniflow_init(&mask->masks, &wc->masks);
1321 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1322 * with minimask_destroy(). */
1324 minimask_clone(struct minimask *dst, const struct minimask *src)
1326 miniflow_clone(&dst->masks, &src->masks);
1329 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1331 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1332 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1334 minimask_combine(struct minimask *dst_,
1335 const struct minimask *a_, const struct minimask *b_,
1336 uint32_t storage[FLOW_U32S])
1338 struct miniflow *dst = &dst_->masks;
1339 const struct miniflow *a = &a_->masks;
1340 const struct miniflow *b = &b_->masks;
1344 dst->values = storage;
1345 for (i = 0; i < MINI_N_MAPS; i++) {
1349 for (map = a->map[i] & b->map[i]; map;
1350 map = zero_rightmost_1bit(map)) {
1351 int ofs = raw_ctz(map) + i * 32;
1352 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1355 dst->map[i] |= rightmost_1bit(map);
1356 dst->values[n++] = mask;
1362 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1363 * itself resides; the caller is responsible for that. */
1365 minimask_destroy(struct minimask *mask)
1367 miniflow_destroy(&mask->masks);
1370 /* Initializes 'dst' as a copy of 'src'. */
1372 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1374 miniflow_expand(&mask->masks, &wc->masks);
1377 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1378 * were expanded into a "struct flow_wildcards". */
1380 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1382 return miniflow_get(&mask->masks, u32_ofs);
1385 /* Returns the VID mask within the vlan_tci member of the "struct
1386 * flow_wildcards" represented by 'mask'. */
1388 minimask_get_vid_mask(const struct minimask *mask)
1390 return miniflow_get_vid(&mask->masks);
1393 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1395 minimask_equal(const struct minimask *a, const struct minimask *b)
1397 return miniflow_equal(&a->masks, &b->masks);
1400 /* Returns a hash value for 'mask', given 'basis'. */
1402 minimask_hash(const struct minimask *mask, uint32_t basis)
1404 return miniflow_hash(&mask->masks, basis);
1407 /* Returns true if at least one bit is wildcarded in 'a_' but not in 'b_',
1408 * false otherwise. */
1410 minimask_has_extra(const struct minimask *a_, const struct minimask *b_)
1412 const struct miniflow *a = &a_->masks;
1413 const struct miniflow *b = &b_->masks;
1416 for (i = 0; i < MINI_N_MAPS; i++) {
1419 for (map = a->map[i] | b->map[i]; map;
1420 map = zero_rightmost_1bit(map)) {
1421 int ofs = raw_ctz(map) + i * 32;
1422 uint32_t a_u32 = miniflow_get(a, ofs);
1423 uint32_t b_u32 = miniflow_get(b, ofs);
1425 if ((a_u32 & b_u32) != b_u32) {
1434 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
1437 minimask_is_catchall(const struct minimask *mask_)
1439 const struct miniflow *mask = &mask_->masks;
1441 BUILD_ASSERT(MINI_N_MAPS == 2);
1442 return !(mask->map[0] | mask->map[1]);