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 ovs_16aligned_ip6_hdr *nh;
171 nh = ofpbuf_try_pull(packet, sizeof *nh);
176 nexthdr = nh->ip6_nxt;
178 memcpy(&flow->ipv6_src, &nh->ip6_src, sizeof flow->ipv6_src);
179 memcpy(&flow->ipv6_dst, &nh->ip6_dst, sizeof flow->ipv6_dst);
181 tc_flow = get_16aligned_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 ovs_16aligned_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_16aligned_be32(&nh->ip_src);
433 flow->nw_dst = get_16aligned_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 = get_16aligned_be32(&arp->ar_spa);
492 flow->nw_dst = get_16aligned_be32(&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 /* Clear the metadata and register wildcard masks. They are not packet
618 flow_wildcards_clear_non_packet_fields(struct flow_wildcards *wc)
620 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
621 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
624 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
627 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
629 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
632 for (i = 0; i < FLOW_U32S; i++) {
640 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
641 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
642 * in 'src1' or 'src2' or both. */
644 flow_wildcards_and(struct flow_wildcards *dst,
645 const struct flow_wildcards *src1,
646 const struct flow_wildcards *src2)
648 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
649 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
650 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
653 for (i = 0; i < FLOW_U32S; i++) {
654 dst_u32[i] = src1_u32[i] & src2_u32[i];
658 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
659 * is, a bit or a field is wildcarded in 'dst' if it is neither
660 * wildcarded in 'src1' nor 'src2'. */
662 flow_wildcards_or(struct flow_wildcards *dst,
663 const struct flow_wildcards *src1,
664 const struct flow_wildcards *src2)
666 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
667 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
668 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
671 for (i = 0; i < FLOW_U32S; i++) {
672 dst_u32[i] = src1_u32[i] | src2_u32[i];
676 /* Perform a bitwise OR of miniflow 'src' flow data with the equivalent
677 * fields in 'dst', storing the result in 'dst'. */
679 flow_union_with_miniflow(struct flow *dst, const struct miniflow *src)
681 uint32_t *dst_u32 = (uint32_t *) dst;
686 for (i = 0; i < MINI_N_MAPS; i++) {
689 for (map = src->map[i]; map; map = zero_rightmost_1bit(map)) {
690 dst_u32[raw_ctz(map) + i * 32] |= src->values[ofs++];
695 /* Fold minimask 'mask''s wildcard mask into 'wc's wildcard mask. */
697 flow_wildcards_fold_minimask(struct flow_wildcards *wc,
698 const struct minimask *mask)
700 flow_union_with_miniflow(&wc->masks, &mask->masks);
703 /* Returns a hash of the wildcards in 'wc'. */
705 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
707 return flow_hash(&wc->masks, basis);
710 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
713 flow_wildcards_equal(const struct flow_wildcards *a,
714 const struct flow_wildcards *b)
716 return flow_equal(&a->masks, &b->masks);
719 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
720 * 'b', false otherwise. */
722 flow_wildcards_has_extra(const struct flow_wildcards *a,
723 const struct flow_wildcards *b)
725 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
726 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
729 for (i = 0; i < FLOW_U32S; i++) {
730 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
737 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
738 * in 'wc' do not need to be equal in 'a' and 'b'. */
740 flow_equal_except(const struct flow *a, const struct flow *b,
741 const struct flow_wildcards *wc)
743 const uint32_t *a_u32 = (const uint32_t *) a;
744 const uint32_t *b_u32 = (const uint32_t *) b;
745 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
748 for (i = 0; i < FLOW_U32S; i++) {
749 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
756 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
757 * (A 0-bit indicates a wildcard bit.) */
759 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
761 wc->masks.regs[idx] = mask;
764 /* Hashes 'flow' based on its L2 through L4 protocol information. */
766 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
771 struct in6_addr ipv6_addr;
776 uint8_t eth_addr[ETH_ADDR_LEN];
782 memset(&fields, 0, sizeof fields);
783 for (i = 0; i < ETH_ADDR_LEN; i++) {
784 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
786 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
787 fields.eth_type = flow->dl_type;
789 /* UDP source and destination port are not taken into account because they
790 * will not necessarily be symmetric in a bidirectional flow. */
791 if (fields.eth_type == htons(ETH_TYPE_IP)) {
792 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
793 fields.ip_proto = flow->nw_proto;
794 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
795 fields.tp_port = flow->tp_src ^ flow->tp_dst;
797 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
798 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
799 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
800 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
802 for (i=0; i<16; i++) {
803 ipv6_addr[i] = a[i] ^ b[i];
805 fields.ip_proto = flow->nw_proto;
806 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
807 fields.tp_port = flow->tp_src ^ flow->tp_dst;
810 return jhash_bytes(&fields, sizeof fields, basis);
813 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
815 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
816 enum nx_hash_fields fields)
819 case NX_HASH_FIELDS_ETH_SRC:
820 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
823 case NX_HASH_FIELDS_SYMMETRIC_L4:
824 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
825 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
826 if (flow->dl_type == htons(ETH_TYPE_IP)) {
827 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
828 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
829 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
830 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
831 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
833 if (is_ip_any(flow)) {
834 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
835 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
836 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
838 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
846 /* Hashes the portions of 'flow' designated by 'fields'. */
848 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
853 case NX_HASH_FIELDS_ETH_SRC:
854 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
856 case NX_HASH_FIELDS_SYMMETRIC_L4:
857 return flow_hash_symmetric_l4(flow, basis);
863 /* Returns a string representation of 'fields'. */
865 flow_hash_fields_to_str(enum nx_hash_fields fields)
868 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
869 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
870 default: return "<unknown>";
874 /* Returns true if the value of 'fields' is supported. Otherwise false. */
876 flow_hash_fields_valid(enum nx_hash_fields fields)
878 return fields == NX_HASH_FIELDS_ETH_SRC
879 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
882 /* Returns a hash value for the bits of 'flow' that are active based on
883 * 'wc', given 'basis'. */
885 flow_hash_in_wildcards(const struct flow *flow,
886 const struct flow_wildcards *wc, uint32_t basis)
888 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
889 const uint32_t *flow_u32 = (const uint32_t *) flow;
894 for (i = 0; i < FLOW_U32S; i++) {
895 hash = mhash_add(hash, flow_u32[i] & wc_u32[i]);
897 return mhash_finish(hash, 4 * FLOW_U32S);
900 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
901 * OpenFlow 1.0 "dl_vlan" value:
903 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
904 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
905 * 'flow' previously matched packets without a VLAN header).
907 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
908 * without a VLAN tag.
910 * - Other values of 'vid' should not be used. */
912 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
914 if (vid == htons(OFP10_VLAN_NONE)) {
915 flow->vlan_tci = htons(0);
917 vid &= htons(VLAN_VID_MASK);
918 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
919 flow->vlan_tci |= htons(VLAN_CFI) | vid;
923 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
924 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
927 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
929 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
930 flow->vlan_tci &= ~mask;
931 flow->vlan_tci |= vid & mask;
934 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
937 * This function has no effect on the VLAN ID that 'flow' matches.
939 * After calling this function, 'flow' will not match packets without a VLAN
942 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
945 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
946 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
949 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
950 * as an OpenFlow 1.1 "mpls_label" value. */
952 flow_set_mpls_label(struct flow *flow, ovs_be32 label)
954 set_mpls_lse_label(&flow->mpls_lse, label);
957 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
960 flow_set_mpls_ttl(struct flow *flow, uint8_t ttl)
962 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
965 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
968 flow_set_mpls_tc(struct flow *flow, uint8_t tc)
970 set_mpls_lse_tc(&flow->mpls_lse, tc);
973 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
975 flow_set_mpls_bos(struct flow *flow, uint8_t bos)
977 set_mpls_lse_bos(&flow->mpls_lse, bos);
980 /* Puts into 'b' a packet that flow_extract() would parse as having the given
983 * (This is useful only for testing, obviously, and the packet isn't really
984 * valid. It hasn't got some checksums filled in, for one, and lots of fields
985 * are just zeroed.) */
987 flow_compose(struct ofpbuf *b, const struct flow *flow)
989 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
990 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
991 struct eth_header *eth = b->l2;
992 eth->eth_type = htons(b->size);
996 if (flow->vlan_tci & htons(VLAN_CFI)) {
997 eth_push_vlan(b, flow->vlan_tci);
1000 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1001 struct ip_header *ip;
1003 b->l3 = ip = ofpbuf_put_zeros(b, sizeof *ip);
1004 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
1005 ip->ip_tos = flow->nw_tos;
1006 ip->ip_ttl = flow->nw_ttl;
1007 ip->ip_proto = flow->nw_proto;
1008 put_16aligned_be32(&ip->ip_src, flow->nw_src);
1009 put_16aligned_be32(&ip->ip_dst, flow->nw_dst);
1011 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
1012 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
1013 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
1014 ip->ip_frag_off |= htons(100);
1017 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
1018 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1019 if (flow->nw_proto == IPPROTO_TCP) {
1020 struct tcp_header *tcp;
1022 b->l4 = tcp = ofpbuf_put_zeros(b, sizeof *tcp);
1023 tcp->tcp_src = flow->tp_src;
1024 tcp->tcp_dst = flow->tp_dst;
1025 tcp->tcp_ctl = TCP_CTL(0, 5);
1026 } else if (flow->nw_proto == IPPROTO_UDP) {
1027 struct udp_header *udp;
1029 b->l4 = udp = ofpbuf_put_zeros(b, sizeof *udp);
1030 udp->udp_src = flow->tp_src;
1031 udp->udp_dst = flow->tp_dst;
1032 } else if (flow->nw_proto == IPPROTO_SCTP) {
1033 struct sctp_header *sctp;
1035 b->l4 = sctp = ofpbuf_put_zeros(b, sizeof *sctp);
1036 sctp->sctp_src = flow->tp_src;
1037 sctp->sctp_dst = flow->tp_dst;
1038 } else if (flow->nw_proto == IPPROTO_ICMP) {
1039 struct icmp_header *icmp;
1041 b->l4 = icmp = ofpbuf_put_zeros(b, sizeof *icmp);
1042 icmp->icmp_type = ntohs(flow->tp_src);
1043 icmp->icmp_code = ntohs(flow->tp_dst);
1044 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
1049 ip->ip_tot_len = htons((uint8_t *) b->data + b->size
1050 - (uint8_t *) b->l3);
1051 ip->ip_csum = csum(ip, sizeof *ip);
1052 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1054 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1055 flow->dl_type == htons(ETH_TYPE_RARP)) {
1056 struct arp_eth_header *arp;
1058 b->l3 = arp = ofpbuf_put_zeros(b, sizeof *arp);
1059 arp->ar_hrd = htons(1);
1060 arp->ar_pro = htons(ETH_TYPE_IP);
1061 arp->ar_hln = ETH_ADDR_LEN;
1063 arp->ar_op = htons(flow->nw_proto);
1065 if (flow->nw_proto == ARP_OP_REQUEST ||
1066 flow->nw_proto == ARP_OP_REPLY) {
1067 put_16aligned_be32(&arp->ar_spa, flow->nw_src);
1068 put_16aligned_be32(&arp->ar_tpa, flow->nw_dst);
1069 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
1070 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
1074 if (eth_type_mpls(flow->dl_type)) {
1076 push_mpls(b, flow->dl_type, flow->mpls_lse);
1080 /* Compressed flow. */
1083 miniflow_n_values(const struct miniflow *flow)
1088 for (i = 0; i < MINI_N_MAPS; i++) {
1089 n += popcount(flow->map[i]);
1095 miniflow_alloc_values(struct miniflow *flow, int n)
1097 if (n <= MINI_N_INLINE) {
1098 return flow->inline_values;
1100 COVERAGE_INC(miniflow_malloc);
1101 return xmalloc(n * sizeof *flow->values);
1105 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1106 * with miniflow_destroy(). */
1108 miniflow_init(struct miniflow *dst, const struct flow *src)
1110 const uint32_t *src_u32 = (const uint32_t *) src;
1115 /* Initialize dst->map, counting the number of nonzero elements. */
1117 memset(dst->map, 0, sizeof dst->map);
1118 for (i = 0; i < FLOW_U32S; i++) {
1120 dst->map[i / 32] |= 1u << (i % 32);
1125 /* Initialize dst->values. */
1126 dst->values = miniflow_alloc_values(dst, n);
1128 for (i = 0; i < MINI_N_MAPS; i++) {
1131 for (map = dst->map[i]; map; map = zero_rightmost_1bit(map)) {
1132 dst->values[ofs++] = src_u32[raw_ctz(map) + i * 32];
1137 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1138 * with miniflow_destroy(). */
1140 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
1142 int n = miniflow_n_values(src);
1143 memcpy(dst->map, src->map, sizeof dst->map);
1144 dst->values = miniflow_alloc_values(dst, n);
1145 memcpy(dst->values, src->values, n * sizeof *dst->values);
1148 /* Initializes 'dst' with the data in 'src', destroying 'src'.
1149 * The caller must eventually free 'dst' with miniflow_destroy(). */
1151 miniflow_move(struct miniflow *dst, struct miniflow *src)
1153 if (src->values == src->inline_values) {
1154 dst->values = dst->inline_values;
1155 memcpy(dst->values, src->values,
1156 miniflow_n_values(src) * sizeof *dst->values);
1158 dst->values = src->values;
1160 memcpy(dst->map, src->map, sizeof dst->map);
1163 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
1164 * itself resides; the caller is responsible for that. */
1166 miniflow_destroy(struct miniflow *flow)
1168 if (flow->values != flow->inline_values) {
1173 /* Initializes 'dst' as a copy of 'src'. */
1175 miniflow_expand(const struct miniflow *src, struct flow *dst)
1177 memset(dst, 0, sizeof *dst);
1178 flow_union_with_miniflow(dst, src);
1181 static const uint32_t *
1182 miniflow_get__(const struct miniflow *flow, unsigned int u32_ofs)
1184 if (!(flow->map[u32_ofs / 32] & (1u << (u32_ofs % 32)))) {
1185 static const uint32_t zero = 0;
1188 const uint32_t *p = flow->values;
1190 BUILD_ASSERT(MINI_N_MAPS == 2);
1192 p += popcount(flow->map[0] & ((1u << u32_ofs) - 1));
1194 p += popcount(flow->map[0]);
1195 p += popcount(flow->map[1] & ((1u << (u32_ofs - 32)) - 1));
1201 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1202 * were expanded into a "struct flow". */
1204 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1206 return *miniflow_get__(flow, u32_ofs);
1209 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
1210 * expanded into a "struct flow". */
1212 miniflow_get_be16(const struct miniflow *flow, unsigned int u8_ofs)
1214 const uint32_t *u32p = miniflow_get__(flow, u8_ofs / 4);
1215 const ovs_be16 *be16p = (const ovs_be16 *) u32p;
1216 return be16p[u8_ofs % 4 != 0];
1219 /* Returns the VID within the vlan_tci member of the "struct flow" represented
1222 miniflow_get_vid(const struct miniflow *flow)
1224 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
1225 return vlan_tci_to_vid(tci);
1228 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1230 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1234 for (i = 0; i < MINI_N_MAPS; i++) {
1235 if (a->map[i] != b->map[i]) {
1240 return !memcmp(a->values, b->values,
1241 miniflow_n_values(a) * sizeof *a->values);
1244 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1245 * in 'mask', false if they differ. */
1247 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1248 const struct minimask *mask)
1253 p = mask->masks.values;
1254 for (i = 0; i < MINI_N_MAPS; i++) {
1257 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1258 int ofs = raw_ctz(map) + i * 32;
1260 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p) {
1270 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1271 * in 'mask', false if they differ. */
1273 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1274 const struct minimask *mask)
1276 const uint32_t *b_u32 = (const uint32_t *) b;
1280 p = mask->masks.values;
1281 for (i = 0; i < MINI_N_MAPS; i++) {
1284 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1285 int ofs = raw_ctz(map) + i * 32;
1287 if ((miniflow_get(a, ofs) ^ b_u32[ofs]) & *p) {
1297 /* Returns a hash value for 'flow', given 'basis'. */
1299 miniflow_hash(const struct miniflow *flow, uint32_t basis)
1301 BUILD_ASSERT_DECL(MINI_N_MAPS == 2);
1302 return hash_3words(flow->map[0], flow->map[1],
1303 hash_words(flow->values, miniflow_n_values(flow),
1307 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1308 * 'mask', given 'basis'.
1310 * The hash values returned by this function are the same as those returned by
1311 * flow_hash_in_minimask(), only the form of the arguments differ. */
1313 miniflow_hash_in_minimask(const struct miniflow *flow,
1314 const struct minimask *mask, uint32_t basis)
1316 const uint32_t *p = mask->masks.values;
1321 for (i = 0; i < MINI_N_MAPS; i++) {
1324 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1325 int ofs = raw_ctz(map) + i * 32;
1327 hash = mhash_add(hash, miniflow_get(flow, ofs) & *p);
1332 return mhash_finish(hash, (p - mask->masks.values) * 4);
1335 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1336 * 'mask', given 'basis'.
1338 * The hash values returned by this function are the same as those returned by
1339 * miniflow_hash_in_minimask(), only the form of the arguments differ. */
1341 flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
1344 const uint32_t *flow_u32 = (const uint32_t *) flow;
1345 const uint32_t *p = mask->masks.values;
1350 for (i = 0; i < MINI_N_MAPS; i++) {
1353 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1354 int ofs = raw_ctz(map) + i * 32;
1356 hash = mhash_add(hash, flow_u32[ofs] & *p);
1361 return mhash_finish(hash, (p - mask->masks.values) * 4);
1364 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1365 * with minimask_destroy(). */
1367 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1369 miniflow_init(&mask->masks, &wc->masks);
1372 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1373 * with minimask_destroy(). */
1375 minimask_clone(struct minimask *dst, const struct minimask *src)
1377 miniflow_clone(&dst->masks, &src->masks);
1380 /* Initializes 'dst' with the data in 'src', destroying 'src'.
1381 * The caller must eventually free 'dst' with minimask_destroy(). */
1383 minimask_move(struct minimask *dst, struct minimask *src)
1385 miniflow_move(&dst->masks, &src->masks);
1388 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1390 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1391 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1393 minimask_combine(struct minimask *dst_,
1394 const struct minimask *a_, const struct minimask *b_,
1395 uint32_t storage[FLOW_U32S])
1397 struct miniflow *dst = &dst_->masks;
1398 const struct miniflow *a = &a_->masks;
1399 const struct miniflow *b = &b_->masks;
1403 dst->values = storage;
1404 for (i = 0; i < MINI_N_MAPS; i++) {
1408 for (map = a->map[i] & b->map[i]; map;
1409 map = zero_rightmost_1bit(map)) {
1410 int ofs = raw_ctz(map) + i * 32;
1411 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1414 dst->map[i] |= rightmost_1bit(map);
1415 dst->values[n++] = mask;
1421 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1422 * itself resides; the caller is responsible for that. */
1424 minimask_destroy(struct minimask *mask)
1426 miniflow_destroy(&mask->masks);
1429 /* Initializes 'dst' as a copy of 'src'. */
1431 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1433 miniflow_expand(&mask->masks, &wc->masks);
1436 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1437 * were expanded into a "struct flow_wildcards". */
1439 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1441 return miniflow_get(&mask->masks, u32_ofs);
1444 /* Returns the VID mask within the vlan_tci member of the "struct
1445 * flow_wildcards" represented by 'mask'. */
1447 minimask_get_vid_mask(const struct minimask *mask)
1449 return miniflow_get_vid(&mask->masks);
1452 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1454 minimask_equal(const struct minimask *a, const struct minimask *b)
1456 return miniflow_equal(&a->masks, &b->masks);
1459 /* Returns a hash value for 'mask', given 'basis'. */
1461 minimask_hash(const struct minimask *mask, uint32_t basis)
1463 return miniflow_hash(&mask->masks, basis);
1466 /* Returns true if at least one bit is wildcarded in 'a_' but not in 'b_',
1467 * false otherwise. */
1469 minimask_has_extra(const struct minimask *a_, const struct minimask *b_)
1471 const struct miniflow *a = &a_->masks;
1472 const struct miniflow *b = &b_->masks;
1475 for (i = 0; i < MINI_N_MAPS; i++) {
1478 for (map = a->map[i] | b->map[i]; map;
1479 map = zero_rightmost_1bit(map)) {
1480 int ofs = raw_ctz(map) + i * 32;
1481 uint32_t a_u32 = miniflow_get(a, ofs);
1482 uint32_t b_u32 = miniflow_get(b, ofs);
1484 if ((a_u32 & b_u32) != b_u32) {
1493 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
1496 minimask_is_catchall(const struct minimask *mask_)
1498 const struct miniflow *mask = &mask_->masks;
1500 BUILD_ASSERT(MINI_N_MAPS == 2);
1501 return !(mask->map[0] | mask->map[1]);