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"
40 COVERAGE_DEFINE(flow_extract);
41 COVERAGE_DEFINE(miniflow_malloc);
43 static struct arp_eth_header *
44 pull_arp(struct ofpbuf *packet)
46 return ofpbuf_try_pull(packet, ARP_ETH_HEADER_LEN);
49 static struct ip_header *
50 pull_ip(struct ofpbuf *packet)
52 if (packet->size >= IP_HEADER_LEN) {
53 struct ip_header *ip = packet->data;
54 int ip_len = IP_IHL(ip->ip_ihl_ver) * 4;
55 if (ip_len >= IP_HEADER_LEN && packet->size >= ip_len) {
56 return ofpbuf_pull(packet, ip_len);
62 static struct tcp_header *
63 pull_tcp(struct ofpbuf *packet)
65 if (packet->size >= TCP_HEADER_LEN) {
66 struct tcp_header *tcp = packet->data;
67 int tcp_len = TCP_OFFSET(tcp->tcp_ctl) * 4;
68 if (tcp_len >= TCP_HEADER_LEN && packet->size >= tcp_len) {
69 return ofpbuf_pull(packet, tcp_len);
75 static struct udp_header *
76 pull_udp(struct ofpbuf *packet)
78 return ofpbuf_try_pull(packet, UDP_HEADER_LEN);
81 static struct sctp_header *
82 pull_sctp(struct ofpbuf *packet)
84 return ofpbuf_try_pull(packet, SCTP_HEADER_LEN);
87 static struct icmp_header *
88 pull_icmp(struct ofpbuf *packet)
90 return ofpbuf_try_pull(packet, ICMP_HEADER_LEN);
93 static struct icmp6_hdr *
94 pull_icmpv6(struct ofpbuf *packet)
96 return ofpbuf_try_pull(packet, sizeof(struct icmp6_hdr));
100 parse_mpls(struct ofpbuf *b, struct flow *flow)
105 while ((mh = ofpbuf_try_pull(b, sizeof *mh))) {
108 flow->mpls_lse = mh->mpls_lse;
110 if (mh->mpls_lse & htonl(MPLS_BOS_MASK)) {
117 parse_vlan(struct ofpbuf *b, struct flow *flow)
120 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
124 if (b->size >= sizeof(struct qtag_prefix) + sizeof(ovs_be16)) {
125 struct qtag_prefix *qp = ofpbuf_pull(b, sizeof *qp);
126 flow->vlan_tci = qp->tci | htons(VLAN_CFI);
131 parse_ethertype(struct ofpbuf *b)
133 struct llc_snap_header *llc;
136 proto = *(ovs_be16 *) ofpbuf_pull(b, sizeof proto);
137 if (ntohs(proto) >= ETH_TYPE_MIN) {
141 if (b->size < sizeof *llc) {
142 return htons(FLOW_DL_TYPE_NONE);
146 if (llc->llc.llc_dsap != LLC_DSAP_SNAP
147 || llc->llc.llc_ssap != LLC_SSAP_SNAP
148 || llc->llc.llc_cntl != LLC_CNTL_SNAP
149 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
150 sizeof llc->snap.snap_org)) {
151 return htons(FLOW_DL_TYPE_NONE);
154 ofpbuf_pull(b, sizeof *llc);
156 if (ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN) {
157 return llc->snap.snap_type;
160 return htons(FLOW_DL_TYPE_NONE);
164 parse_ipv6(struct ofpbuf *packet, struct flow *flow)
166 const struct ovs_16aligned_ip6_hdr *nh;
170 nh = ofpbuf_try_pull(packet, sizeof *nh);
175 nexthdr = nh->ip6_nxt;
177 memcpy(&flow->ipv6_src, &nh->ip6_src, sizeof flow->ipv6_src);
178 memcpy(&flow->ipv6_dst, &nh->ip6_dst, sizeof flow->ipv6_dst);
180 tc_flow = get_16aligned_be32(&nh->ip6_flow);
181 flow->nw_tos = ntohl(tc_flow) >> 20;
182 flow->ipv6_label = tc_flow & htonl(IPV6_LABEL_MASK);
183 flow->nw_ttl = nh->ip6_hlim;
184 flow->nw_proto = IPPROTO_NONE;
187 if ((nexthdr != IPPROTO_HOPOPTS)
188 && (nexthdr != IPPROTO_ROUTING)
189 && (nexthdr != IPPROTO_DSTOPTS)
190 && (nexthdr != IPPROTO_AH)
191 && (nexthdr != IPPROTO_FRAGMENT)) {
192 /* It's either a terminal header (e.g., TCP, UDP) or one we
193 * don't understand. In either case, we're done with the
194 * packet, so use it to fill in 'nw_proto'. */
198 /* We only verify that at least 8 bytes of the next header are
199 * available, but many of these headers are longer. Ensure that
200 * accesses within the extension header are within those first 8
201 * bytes. All extension headers are required to be at least 8
203 if (packet->size < 8) {
207 if ((nexthdr == IPPROTO_HOPOPTS)
208 || (nexthdr == IPPROTO_ROUTING)
209 || (nexthdr == IPPROTO_DSTOPTS)) {
210 /* These headers, while different, have the fields we care about
211 * in the same location and with the same interpretation. */
212 const struct ip6_ext *ext_hdr = packet->data;
213 nexthdr = ext_hdr->ip6e_nxt;
214 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 1) * 8)) {
217 } else if (nexthdr == IPPROTO_AH) {
218 /* A standard AH definition isn't available, but the fields
219 * we care about are in the same location as the generic
220 * option header--only the header length is calculated
222 const struct ip6_ext *ext_hdr = packet->data;
223 nexthdr = ext_hdr->ip6e_nxt;
224 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 2) * 4)) {
227 } else if (nexthdr == IPPROTO_FRAGMENT) {
228 const struct ovs_16aligned_ip6_frag *frag_hdr = packet->data;
230 nexthdr = frag_hdr->ip6f_nxt;
231 if (!ofpbuf_try_pull(packet, sizeof *frag_hdr)) {
235 /* We only process the first fragment. */
236 if (frag_hdr->ip6f_offlg != htons(0)) {
237 flow->nw_frag = FLOW_NW_FRAG_ANY;
238 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
239 flow->nw_frag |= FLOW_NW_FRAG_LATER;
240 nexthdr = IPPROTO_FRAGMENT;
247 flow->nw_proto = nexthdr;
252 parse_tcp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
254 const struct tcp_header *tcp = pull_tcp(b);
256 flow->tp_src = tcp->tcp_src;
257 flow->tp_dst = tcp->tcp_dst;
258 packet->l7 = b->data;
263 parse_udp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
265 const struct udp_header *udp = pull_udp(b);
267 flow->tp_src = udp->udp_src;
268 flow->tp_dst = udp->udp_dst;
269 packet->l7 = b->data;
274 parse_sctp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
276 const struct sctp_header *sctp = pull_sctp(b);
278 flow->tp_src = sctp->sctp_src;
279 flow->tp_dst = sctp->sctp_dst;
280 packet->l7 = b->data;
285 parse_icmpv6(struct ofpbuf *b, struct flow *flow)
287 const struct icmp6_hdr *icmp = pull_icmpv6(b);
293 /* The ICMPv6 type and code fields use the 16-bit transport port
294 * fields, so we need to store them in 16-bit network byte order. */
295 flow->tp_src = htons(icmp->icmp6_type);
296 flow->tp_dst = htons(icmp->icmp6_code);
298 if (icmp->icmp6_code == 0 &&
299 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
300 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
301 const struct in6_addr *nd_target;
303 nd_target = ofpbuf_try_pull(b, sizeof *nd_target);
307 flow->nd_target = *nd_target;
309 while (b->size >= 8) {
310 /* The minimum size of an option is 8 bytes, which also is
311 * the size of Ethernet link-layer options. */
312 const struct nd_opt_hdr *nd_opt = b->data;
313 int opt_len = nd_opt->nd_opt_len * 8;
315 if (!opt_len || opt_len > b->size) {
319 /* Store the link layer address if the appropriate option is
320 * provided. It is considered an error if the same link
321 * layer option is specified twice. */
322 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
324 if (eth_addr_is_zero(flow->arp_sha)) {
325 memcpy(flow->arp_sha, nd_opt + 1, ETH_ADDR_LEN);
329 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
331 if (eth_addr_is_zero(flow->arp_tha)) {
332 memcpy(flow->arp_tha, nd_opt + 1, ETH_ADDR_LEN);
338 if (!ofpbuf_try_pull(b, opt_len)) {
347 memset(&flow->nd_target, 0, sizeof(flow->nd_target));
348 memset(flow->arp_sha, 0, sizeof(flow->arp_sha));
349 memset(flow->arp_tha, 0, sizeof(flow->arp_tha));
355 /* Initializes 'flow' members from 'packet', 'skb_priority', 'tnl', and
358 * Initializes 'packet' header pointers as follows:
360 * - packet->l2 to the start of the Ethernet header.
362 * - packet->l2_5 to the start of the MPLS shim header.
364 * - packet->l3 to just past the Ethernet header, or just past the
365 * vlan_header if one is present, to the first byte of the payload of the
368 * - packet->l4 to just past the IPv4 header, if one is present and has a
369 * correct length, and otherwise NULL.
371 * - packet->l7 to just past the TCP/UDP/SCTP/ICMP header, if one is
372 * present and has a correct length, and otherwise NULL.
375 flow_extract(struct ofpbuf *packet, uint32_t skb_priority, uint32_t pkt_mark,
376 const struct flow_tnl *tnl, const union flow_in_port *in_port,
379 struct ofpbuf b = *packet;
380 struct eth_header *eth;
382 COVERAGE_INC(flow_extract);
384 memset(flow, 0, sizeof *flow);
387 ovs_assert(tnl != &flow->tunnel);
391 flow->in_port = *in_port;
393 flow->skb_priority = skb_priority;
394 flow->pkt_mark = pkt_mark;
402 if (b.size < sizeof *eth) {
408 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
409 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
411 /* dl_type, vlan_tci. */
412 ofpbuf_pull(&b, ETH_ADDR_LEN * 2);
413 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
414 parse_vlan(&b, flow);
416 flow->dl_type = parse_ethertype(&b);
418 /* Parse mpls, copy l3 ttl. */
419 if (eth_type_mpls(flow->dl_type)) {
420 packet->l2_5 = b.data;
421 parse_mpls(&b, flow);
426 if (flow->dl_type == htons(ETH_TYPE_IP)) {
427 const struct ip_header *nh = pull_ip(&b);
431 flow->nw_src = get_16aligned_be32(&nh->ip_src);
432 flow->nw_dst = get_16aligned_be32(&nh->ip_dst);
433 flow->nw_proto = nh->ip_proto;
435 flow->nw_tos = nh->ip_tos;
436 if (IP_IS_FRAGMENT(nh->ip_frag_off)) {
437 flow->nw_frag = FLOW_NW_FRAG_ANY;
438 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
439 flow->nw_frag |= FLOW_NW_FRAG_LATER;
442 flow->nw_ttl = nh->ip_ttl;
444 if (!(nh->ip_frag_off & htons(IP_FRAG_OFF_MASK))) {
445 if (flow->nw_proto == IPPROTO_TCP) {
446 parse_tcp(packet, &b, flow);
447 } else if (flow->nw_proto == IPPROTO_UDP) {
448 parse_udp(packet, &b, flow);
449 } else if (flow->nw_proto == IPPROTO_SCTP) {
450 parse_sctp(packet, &b, flow);
451 } else if (flow->nw_proto == IPPROTO_ICMP) {
452 const struct icmp_header *icmp = pull_icmp(&b);
454 flow->tp_src = htons(icmp->icmp_type);
455 flow->tp_dst = htons(icmp->icmp_code);
461 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
462 if (parse_ipv6(&b, flow)) {
467 if (flow->nw_proto == IPPROTO_TCP) {
468 parse_tcp(packet, &b, flow);
469 } else if (flow->nw_proto == IPPROTO_UDP) {
470 parse_udp(packet, &b, flow);
471 } else if (flow->nw_proto == IPPROTO_SCTP) {
472 parse_sctp(packet, &b, flow);
473 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
474 if (parse_icmpv6(&b, flow)) {
478 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
479 flow->dl_type == htons(ETH_TYPE_RARP)) {
480 const struct arp_eth_header *arp = pull_arp(&b);
481 if (arp && arp->ar_hrd == htons(1)
482 && arp->ar_pro == htons(ETH_TYPE_IP)
483 && arp->ar_hln == ETH_ADDR_LEN
484 && arp->ar_pln == 4) {
485 /* We only match on the lower 8 bits of the opcode. */
486 if (ntohs(arp->ar_op) <= 0xff) {
487 flow->nw_proto = ntohs(arp->ar_op);
490 flow->nw_src = get_16aligned_be32(&arp->ar_spa);
491 flow->nw_dst = get_16aligned_be32(&arp->ar_tpa);
492 memcpy(flow->arp_sha, arp->ar_sha, ETH_ADDR_LEN);
493 memcpy(flow->arp_tha, arp->ar_tha, ETH_ADDR_LEN);
498 /* For every bit of a field that is wildcarded in 'wildcards', sets the
499 * corresponding bit in 'flow' to zero. */
501 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
503 uint32_t *flow_u32 = (uint32_t *) flow;
504 const uint32_t *wc_u32 = (const uint32_t *) &wildcards->masks;
507 for (i = 0; i < FLOW_U32S; i++) {
508 flow_u32[i] &= wc_u32[i];
512 /* Initializes 'fmd' with the metadata found in 'flow'. */
514 flow_get_metadata(const struct flow *flow, struct flow_metadata *fmd)
516 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 21);
518 fmd->tun_id = flow->tunnel.tun_id;
519 fmd->tun_src = flow->tunnel.ip_src;
520 fmd->tun_dst = flow->tunnel.ip_dst;
521 fmd->metadata = flow->metadata;
522 memcpy(fmd->regs, flow->regs, sizeof fmd->regs);
523 fmd->pkt_mark = flow->pkt_mark;
524 fmd->in_port = flow->in_port.ofp_port;
528 flow_to_string(const struct flow *flow)
530 struct ds ds = DS_EMPTY_INITIALIZER;
531 flow_format(&ds, flow);
536 flow_tun_flag_to_string(uint32_t flags)
539 case FLOW_TNL_F_DONT_FRAGMENT:
541 case FLOW_TNL_F_CSUM:
551 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
552 uint32_t flags, char del)
560 uint32_t bit = rightmost_1bit(flags);
563 s = bit_to_string(bit);
565 ds_put_format(ds, "%s%c", s, del);
574 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
580 flow_format(struct ds *ds, const struct flow *flow)
584 match_wc_init(&match, flow);
585 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
589 flow_print(FILE *stream, const struct flow *flow)
591 char *s = flow_to_string(flow);
596 /* flow_wildcards functions. */
598 /* Initializes 'wc' as a set of wildcards that matches every packet. */
600 flow_wildcards_init_catchall(struct flow_wildcards *wc)
602 memset(&wc->masks, 0, sizeof wc->masks);
605 /* Initializes 'wc' as an exact-match set of wildcards; that is, 'wc' does not
606 * wildcard any bits or fields. */
608 flow_wildcards_init_exact(struct flow_wildcards *wc)
610 memset(&wc->masks, 0xff, sizeof wc->masks);
613 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
616 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
618 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
621 for (i = 0; i < FLOW_U32S; i++) {
629 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
630 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
631 * in 'src1' or 'src2' or both. */
633 flow_wildcards_and(struct flow_wildcards *dst,
634 const struct flow_wildcards *src1,
635 const struct flow_wildcards *src2)
637 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
638 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
639 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
642 for (i = 0; i < FLOW_U32S; i++) {
643 dst_u32[i] = src1_u32[i] & src2_u32[i];
647 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
648 * is, a bit or a field is wildcarded in 'dst' if it is neither
649 * wildcarded in 'src1' nor 'src2'. */
651 flow_wildcards_or(struct flow_wildcards *dst,
652 const struct flow_wildcards *src1,
653 const struct flow_wildcards *src2)
655 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
656 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
657 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
660 for (i = 0; i < FLOW_U32S; i++) {
661 dst_u32[i] = src1_u32[i] | src2_u32[i];
665 /* Perform a bitwise OR of miniflow 'src' flow data with the equivalent
666 * fields in 'dst', storing the result in 'dst'. */
668 flow_union_with_miniflow(struct flow *dst, const struct miniflow *src)
670 uint32_t *dst_u32 = (uint32_t *) dst;
675 for (i = 0; i < MINI_N_MAPS; i++) {
678 for (map = src->map[i]; map; map = zero_rightmost_1bit(map)) {
679 dst_u32[raw_ctz(map) + i * 32] |= src->values[ofs++];
684 /* Fold minimask 'mask''s wildcard mask into 'wc's wildcard mask. */
686 flow_wildcards_fold_minimask(struct flow_wildcards *wc,
687 const struct minimask *mask)
689 flow_union_with_miniflow(&wc->masks, &mask->masks);
692 /* Returns a hash of the wildcards in 'wc'. */
694 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
696 return flow_hash(&wc->masks, basis);
699 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
702 flow_wildcards_equal(const struct flow_wildcards *a,
703 const struct flow_wildcards *b)
705 return flow_equal(&a->masks, &b->masks);
708 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
709 * 'b', false otherwise. */
711 flow_wildcards_has_extra(const struct flow_wildcards *a,
712 const struct flow_wildcards *b)
714 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
715 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
718 for (i = 0; i < FLOW_U32S; i++) {
719 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
726 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
727 * in 'wc' do not need to be equal in 'a' and 'b'. */
729 flow_equal_except(const struct flow *a, const struct flow *b,
730 const struct flow_wildcards *wc)
732 const uint32_t *a_u32 = (const uint32_t *) a;
733 const uint32_t *b_u32 = (const uint32_t *) b;
734 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
737 for (i = 0; i < FLOW_U32S; i++) {
738 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
745 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
746 * (A 0-bit indicates a wildcard bit.) */
748 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
750 wc->masks.regs[idx] = mask;
753 /* Hashes 'flow' based on its L2 through L4 protocol information. */
755 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
760 struct in6_addr ipv6_addr;
765 uint8_t eth_addr[ETH_ADDR_LEN];
771 memset(&fields, 0, sizeof fields);
772 for (i = 0; i < ETH_ADDR_LEN; i++) {
773 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
775 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
776 fields.eth_type = flow->dl_type;
778 /* UDP source and destination port are not taken into account because they
779 * will not necessarily be symmetric in a bidirectional flow. */
780 if (fields.eth_type == htons(ETH_TYPE_IP)) {
781 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
782 fields.ip_proto = flow->nw_proto;
783 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
784 fields.tp_port = flow->tp_src ^ flow->tp_dst;
786 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
787 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
788 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
789 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
791 for (i=0; i<16; i++) {
792 ipv6_addr[i] = a[i] ^ b[i];
794 fields.ip_proto = flow->nw_proto;
795 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
796 fields.tp_port = flow->tp_src ^ flow->tp_dst;
799 return jhash_bytes(&fields, sizeof fields, basis);
802 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
804 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
805 enum nx_hash_fields fields)
808 case NX_HASH_FIELDS_ETH_SRC:
809 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
812 case NX_HASH_FIELDS_SYMMETRIC_L4:
813 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
814 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
815 if (flow->dl_type == htons(ETH_TYPE_IP)) {
816 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
817 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
818 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
819 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
820 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
822 if (is_ip_any(flow)) {
823 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
824 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
825 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
827 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
835 /* Hashes the portions of 'flow' designated by 'fields'. */
837 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
842 case NX_HASH_FIELDS_ETH_SRC:
843 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
845 case NX_HASH_FIELDS_SYMMETRIC_L4:
846 return flow_hash_symmetric_l4(flow, basis);
852 /* Returns a string representation of 'fields'. */
854 flow_hash_fields_to_str(enum nx_hash_fields fields)
857 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
858 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
859 default: return "<unknown>";
863 /* Returns true if the value of 'fields' is supported. Otherwise false. */
865 flow_hash_fields_valid(enum nx_hash_fields fields)
867 return fields == NX_HASH_FIELDS_ETH_SRC
868 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
871 /* Returns a hash value for the bits of 'flow' that are active based on
872 * 'wc', given 'basis'. */
874 flow_hash_in_wildcards(const struct flow *flow,
875 const struct flow_wildcards *wc, uint32_t basis)
877 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
878 const uint32_t *flow_u32 = (const uint32_t *) flow;
883 for (i = 0; i < FLOW_U32S; i++) {
884 hash = mhash_add(hash, flow_u32[i] & wc_u32[i]);
886 return mhash_finish(hash, 4 * FLOW_U32S);
889 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
890 * OpenFlow 1.0 "dl_vlan" value:
892 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
893 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
894 * 'flow' previously matched packets without a VLAN header).
896 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
897 * without a VLAN tag.
899 * - Other values of 'vid' should not be used. */
901 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
903 if (vid == htons(OFP10_VLAN_NONE)) {
904 flow->vlan_tci = htons(0);
906 vid &= htons(VLAN_VID_MASK);
907 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
908 flow->vlan_tci |= htons(VLAN_CFI) | vid;
912 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
913 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
916 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
918 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
919 flow->vlan_tci &= ~mask;
920 flow->vlan_tci |= vid & mask;
923 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
926 * This function has no effect on the VLAN ID that 'flow' matches.
928 * After calling this function, 'flow' will not match packets without a VLAN
931 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
934 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
935 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
938 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
939 * as an OpenFlow 1.1 "mpls_label" value. */
941 flow_set_mpls_label(struct flow *flow, ovs_be32 label)
943 set_mpls_lse_label(&flow->mpls_lse, label);
946 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
949 flow_set_mpls_ttl(struct flow *flow, uint8_t ttl)
951 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
954 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
957 flow_set_mpls_tc(struct flow *flow, uint8_t tc)
959 set_mpls_lse_tc(&flow->mpls_lse, tc);
962 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
964 flow_set_mpls_bos(struct flow *flow, uint8_t bos)
966 set_mpls_lse_bos(&flow->mpls_lse, bos);
969 /* Puts into 'b' a packet that flow_extract() would parse as having the given
972 * (This is useful only for testing, obviously, and the packet isn't really
973 * valid. It hasn't got some checksums filled in, for one, and lots of fields
974 * are just zeroed.) */
976 flow_compose(struct ofpbuf *b, const struct flow *flow)
978 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
979 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
980 struct eth_header *eth = b->l2;
981 eth->eth_type = htons(b->size);
985 if (flow->vlan_tci & htons(VLAN_CFI)) {
986 eth_push_vlan(b, flow->vlan_tci);
989 if (flow->dl_type == htons(ETH_TYPE_IP)) {
990 struct ip_header *ip;
992 b->l3 = ip = ofpbuf_put_zeros(b, sizeof *ip);
993 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
994 ip->ip_tos = flow->nw_tos;
995 ip->ip_ttl = flow->nw_ttl;
996 ip->ip_proto = flow->nw_proto;
997 put_16aligned_be32(&ip->ip_src, flow->nw_src);
998 put_16aligned_be32(&ip->ip_dst, flow->nw_dst);
1000 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
1001 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
1002 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
1003 ip->ip_frag_off |= htons(100);
1006 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
1007 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1008 if (flow->nw_proto == IPPROTO_TCP) {
1009 struct tcp_header *tcp;
1011 b->l4 = tcp = ofpbuf_put_zeros(b, sizeof *tcp);
1012 tcp->tcp_src = flow->tp_src;
1013 tcp->tcp_dst = flow->tp_dst;
1014 tcp->tcp_ctl = TCP_CTL(0, 5);
1015 } else if (flow->nw_proto == IPPROTO_UDP) {
1016 struct udp_header *udp;
1018 b->l4 = udp = ofpbuf_put_zeros(b, sizeof *udp);
1019 udp->udp_src = flow->tp_src;
1020 udp->udp_dst = flow->tp_dst;
1021 } else if (flow->nw_proto == IPPROTO_SCTP) {
1022 struct sctp_header *sctp;
1024 b->l4 = sctp = ofpbuf_put_zeros(b, sizeof *sctp);
1025 sctp->sctp_src = flow->tp_src;
1026 sctp->sctp_dst = flow->tp_dst;
1027 } else if (flow->nw_proto == IPPROTO_ICMP) {
1028 struct icmp_header *icmp;
1030 b->l4 = icmp = ofpbuf_put_zeros(b, sizeof *icmp);
1031 icmp->icmp_type = ntohs(flow->tp_src);
1032 icmp->icmp_code = ntohs(flow->tp_dst);
1033 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
1038 ip->ip_tot_len = htons((uint8_t *) b->data + b->size
1039 - (uint8_t *) b->l3);
1040 ip->ip_csum = csum(ip, sizeof *ip);
1041 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1043 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1044 flow->dl_type == htons(ETH_TYPE_RARP)) {
1045 struct arp_eth_header *arp;
1047 b->l3 = arp = ofpbuf_put_zeros(b, sizeof *arp);
1048 arp->ar_hrd = htons(1);
1049 arp->ar_pro = htons(ETH_TYPE_IP);
1050 arp->ar_hln = ETH_ADDR_LEN;
1052 arp->ar_op = htons(flow->nw_proto);
1054 if (flow->nw_proto == ARP_OP_REQUEST ||
1055 flow->nw_proto == ARP_OP_REPLY) {
1056 put_16aligned_be32(&arp->ar_spa, flow->nw_src);
1057 put_16aligned_be32(&arp->ar_tpa, flow->nw_dst);
1058 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
1059 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
1063 if (eth_type_mpls(flow->dl_type)) {
1065 push_mpls(b, flow->dl_type, flow->mpls_lse);
1069 /* Compressed flow. */
1072 miniflow_n_values(const struct miniflow *flow)
1077 for (i = 0; i < MINI_N_MAPS; i++) {
1078 n += popcount(flow->map[i]);
1084 miniflow_alloc_values(struct miniflow *flow, int n)
1086 if (n <= MINI_N_INLINE) {
1087 return flow->inline_values;
1089 COVERAGE_INC(miniflow_malloc);
1090 return xmalloc(n * sizeof *flow->values);
1094 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
1095 * the caller. The caller must have already initialized 'dst->map' properly
1096 * to indicate the nonzero uint32_t elements of 'src'. 'n' must be the number
1097 * of 1-bits in 'dst->map'.
1099 * This function initializes 'dst->values' (either inline if possible or with
1100 * malloc() otherwise) and copies the nonzero uint32_t elements of 'src' into
1103 miniflow_init__(struct miniflow *dst, const struct flow *src, int n)
1105 const uint32_t *src_u32 = (const uint32_t *) src;
1109 dst->values = miniflow_alloc_values(dst, n);
1111 for (i = 0; i < MINI_N_MAPS; i++) {
1114 for (map = dst->map[i]; map; map = zero_rightmost_1bit(map)) {
1115 dst->values[ofs++] = src_u32[raw_ctz(map) + i * 32];
1120 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1121 * with miniflow_destroy(). */
1123 miniflow_init(struct miniflow *dst, const struct flow *src)
1125 const uint32_t *src_u32 = (const uint32_t *) src;
1129 /* Initialize dst->map, counting the number of nonzero elements. */
1131 memset(dst->map, 0, sizeof dst->map);
1132 for (i = 0; i < FLOW_U32S; i++) {
1134 dst->map[i / 32] |= 1u << (i % 32);
1139 miniflow_init__(dst, src, n);
1142 /* Initializes 'dst' as a copy of 'src', using 'mask->map' as 'dst''s map. The
1143 * caller must eventually free 'dst' with miniflow_destroy(). */
1145 miniflow_init_with_minimask(struct miniflow *dst, const struct flow *src,
1146 const struct minimask *mask)
1148 memcpy(dst->map, mask->masks.map, sizeof dst->map);
1149 miniflow_init__(dst, src, miniflow_n_values(dst));
1152 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1153 * with miniflow_destroy(). */
1155 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
1157 int n = miniflow_n_values(src);
1158 memcpy(dst->map, src->map, sizeof dst->map);
1159 dst->values = miniflow_alloc_values(dst, n);
1160 memcpy(dst->values, src->values, n * sizeof *dst->values);
1163 /* Initializes 'dst' with the data in 'src', destroying 'src'.
1164 * The caller must eventually free 'dst' with miniflow_destroy(). */
1166 miniflow_move(struct miniflow *dst, struct miniflow *src)
1168 if (src->values == src->inline_values) {
1169 dst->values = dst->inline_values;
1170 memcpy(dst->values, src->values,
1171 miniflow_n_values(src) * sizeof *dst->values);
1173 dst->values = src->values;
1175 memcpy(dst->map, src->map, sizeof dst->map);
1178 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
1179 * itself resides; the caller is responsible for that. */
1181 miniflow_destroy(struct miniflow *flow)
1183 if (flow->values != flow->inline_values) {
1188 /* Initializes 'dst' as a copy of 'src'. */
1190 miniflow_expand(const struct miniflow *src, struct flow *dst)
1192 memset(dst, 0, sizeof *dst);
1193 flow_union_with_miniflow(dst, src);
1196 static const uint32_t *
1197 miniflow_get__(const struct miniflow *flow, unsigned int u32_ofs)
1199 if (!(flow->map[u32_ofs / 32] & (1u << (u32_ofs % 32)))) {
1200 static const uint32_t zero = 0;
1203 const uint32_t *p = flow->values;
1205 BUILD_ASSERT(MINI_N_MAPS == 2);
1207 p += popcount(flow->map[0] & ((1u << u32_ofs) - 1));
1209 p += popcount(flow->map[0]);
1210 p += popcount(flow->map[1] & ((1u << (u32_ofs - 32)) - 1));
1216 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1217 * were expanded into a "struct flow". */
1219 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1221 return *miniflow_get__(flow, u32_ofs);
1224 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
1225 * expanded into a "struct flow". */
1227 miniflow_get_be16(const struct miniflow *flow, unsigned int u8_ofs)
1229 const uint32_t *u32p = miniflow_get__(flow, u8_ofs / 4);
1230 const ovs_be16 *be16p = (const ovs_be16 *) u32p;
1231 return be16p[u8_ofs % 4 != 0];
1234 /* Returns the VID within the vlan_tci member of the "struct flow" represented
1237 miniflow_get_vid(const struct miniflow *flow)
1239 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
1240 return vlan_tci_to_vid(tci);
1243 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1245 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1247 const uint32_t *ap = a->values;
1248 const uint32_t *bp = b->values;
1251 for (i = 0; i < MINI_N_MAPS; i++) {
1252 const uint32_t a_map = a->map[i];
1253 const uint32_t b_map = b->map[i];
1256 if (a_map == b_map) {
1257 for (map = a_map; map; map = zero_rightmost_1bit(map)) {
1258 if (*ap++ != *bp++) {
1263 for (map = a_map | b_map; map; map = zero_rightmost_1bit(map)) {
1264 uint32_t bit = rightmost_1bit(map);
1265 uint32_t a_value = a_map & bit ? *ap++ : 0;
1266 uint32_t b_value = b_map & bit ? *bp++ : 0;
1268 if (a_value != b_value) {
1278 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1279 * in 'mask', false if they differ. */
1281 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1282 const struct minimask *mask)
1287 p = mask->masks.values;
1288 for (i = 0; i < MINI_N_MAPS; i++) {
1291 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1292 int ofs = raw_ctz(map) + i * 32;
1294 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p) {
1304 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1305 * in 'mask', false if they differ. */
1307 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1308 const struct minimask *mask)
1310 const uint32_t *b_u32 = (const uint32_t *) b;
1314 p = mask->masks.values;
1315 for (i = 0; i < MINI_N_MAPS; i++) {
1318 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1319 int ofs = raw_ctz(map) + i * 32;
1321 if ((miniflow_get(a, ofs) ^ b_u32[ofs]) & *p) {
1331 /* Returns a hash value for 'flow', given 'basis'. */
1333 miniflow_hash(const struct miniflow *flow, uint32_t basis)
1335 const uint32_t *p = flow->values;
1336 uint32_t hash = basis;
1339 for (i = 0; i < MINI_N_MAPS; i++) {
1340 uint32_t hash_map = 0;
1343 for (map = flow->map[i]; map; map = zero_rightmost_1bit(map)) {
1345 hash = mhash_add(hash, *p);
1346 hash_map |= rightmost_1bit(map);
1350 hash = mhash_add(hash, hash_map);
1352 return mhash_finish(hash, p - flow->values);
1355 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1356 * 'mask', given 'basis'.
1358 * The hash values returned by this function are the same as those returned by
1359 * flow_hash_in_minimask(), only the form of the arguments differ. */
1361 miniflow_hash_in_minimask(const struct miniflow *flow,
1362 const struct minimask *mask, uint32_t basis)
1364 const uint32_t *p = mask->masks.values;
1369 for (i = 0; i < MINI_N_MAPS; i++) {
1372 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1374 int ofs = raw_ctz(map) + i * 32;
1375 hash = mhash_add(hash, miniflow_get(flow, ofs) & *p);
1381 return mhash_finish(hash, (p - mask->masks.values) * 4);
1384 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1385 * 'mask', given 'basis'.
1387 * The hash values returned by this function are the same as those returned by
1388 * miniflow_hash_in_minimask(), only the form of the arguments differ. */
1390 flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
1393 const uint32_t *flow_u32;
1394 const uint32_t *p = mask->masks.values;
1399 flow_u32 = (const uint32_t *) flow;
1400 for (i = 0; i < MINI_N_MAPS; i++) {
1403 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1405 hash = mhash_add(hash, flow_u32[raw_ctz(map)] & *p);
1412 return mhash_finish(hash, (p - mask->masks.values) * 4);
1415 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1416 * with minimask_destroy(). */
1418 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1420 miniflow_init(&mask->masks, &wc->masks);
1423 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1424 * with minimask_destroy(). */
1426 minimask_clone(struct minimask *dst, const struct minimask *src)
1428 miniflow_clone(&dst->masks, &src->masks);
1431 /* Initializes 'dst' with the data in 'src', destroying 'src'.
1432 * The caller must eventually free 'dst' with minimask_destroy(). */
1434 minimask_move(struct minimask *dst, struct minimask *src)
1436 miniflow_move(&dst->masks, &src->masks);
1439 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1441 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1442 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1444 minimask_combine(struct minimask *dst_,
1445 const struct minimask *a_, const struct minimask *b_,
1446 uint32_t storage[FLOW_U32S])
1448 struct miniflow *dst = &dst_->masks;
1449 const struct miniflow *a = &a_->masks;
1450 const struct miniflow *b = &b_->masks;
1454 dst->values = storage;
1455 for (i = 0; i < MINI_N_MAPS; i++) {
1459 for (map = a->map[i] & b->map[i]; map;
1460 map = zero_rightmost_1bit(map)) {
1461 int ofs = raw_ctz(map) + i * 32;
1462 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1465 dst->map[i] |= rightmost_1bit(map);
1466 dst->values[n++] = mask;
1472 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1473 * itself resides; the caller is responsible for that. */
1475 minimask_destroy(struct minimask *mask)
1477 miniflow_destroy(&mask->masks);
1480 /* Initializes 'dst' as a copy of 'src'. */
1482 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1484 miniflow_expand(&mask->masks, &wc->masks);
1487 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1488 * were expanded into a "struct flow_wildcards". */
1490 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1492 return miniflow_get(&mask->masks, u32_ofs);
1495 /* Returns the VID mask within the vlan_tci member of the "struct
1496 * flow_wildcards" represented by 'mask'. */
1498 minimask_get_vid_mask(const struct minimask *mask)
1500 return miniflow_get_vid(&mask->masks);
1503 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1505 minimask_equal(const struct minimask *a, const struct minimask *b)
1507 return miniflow_equal(&a->masks, &b->masks);
1510 /* Returns a hash value for 'mask', given 'basis'. */
1512 minimask_hash(const struct minimask *mask, uint32_t basis)
1514 return miniflow_hash(&mask->masks, basis);
1517 /* Returns true if at least one bit is wildcarded in 'a_' but not in 'b_',
1518 * false otherwise. */
1520 minimask_has_extra(const struct minimask *a_, const struct minimask *b_)
1522 const struct miniflow *a = &a_->masks;
1523 const struct miniflow *b = &b_->masks;
1526 for (i = 0; i < MINI_N_MAPS; i++) {
1529 for (map = a->map[i] | b->map[i]; map;
1530 map = zero_rightmost_1bit(map)) {
1531 int ofs = raw_ctz(map) + i * 32;
1532 uint32_t a_u32 = miniflow_get(a, ofs);
1533 uint32_t b_u32 = miniflow_get(b, ofs);
1535 if ((a_u32 & b_u32) != b_u32) {
1544 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
1547 minimask_is_catchall(const struct minimask *mask_)
1549 const struct miniflow *mask = &mask_->masks;
1550 const uint32_t *p = mask->values;
1553 for (i = 0; i < MINI_N_MAPS; i++) {
1556 for (map = mask->map[i]; map; map = zero_rightmost_1bit(map)) {