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"
39 #include "unaligned.h"
41 COVERAGE_DEFINE(flow_extract);
42 COVERAGE_DEFINE(miniflow_malloc);
44 static struct arp_eth_header *
45 pull_arp(struct ofpbuf *packet)
47 return ofpbuf_try_pull(packet, ARP_ETH_HEADER_LEN);
50 static struct ip_header *
51 pull_ip(struct ofpbuf *packet)
53 if (packet->size >= IP_HEADER_LEN) {
54 struct ip_header *ip = packet->data;
55 int ip_len = IP_IHL(ip->ip_ihl_ver) * 4;
56 if (ip_len >= IP_HEADER_LEN && packet->size >= ip_len) {
57 return ofpbuf_pull(packet, ip_len);
63 static struct tcp_header *
64 pull_tcp(struct ofpbuf *packet)
66 if (packet->size >= TCP_HEADER_LEN) {
67 struct tcp_header *tcp = packet->data;
68 int tcp_len = TCP_OFFSET(tcp->tcp_ctl) * 4;
69 if (tcp_len >= TCP_HEADER_LEN && packet->size >= tcp_len) {
70 return ofpbuf_pull(packet, tcp_len);
76 static struct udp_header *
77 pull_udp(struct ofpbuf *packet)
79 return ofpbuf_try_pull(packet, UDP_HEADER_LEN);
82 static struct sctp_header *
83 pull_sctp(struct ofpbuf *packet)
85 return ofpbuf_try_pull(packet, SCTP_HEADER_LEN);
88 static struct icmp_header *
89 pull_icmp(struct ofpbuf *packet)
91 return ofpbuf_try_pull(packet, ICMP_HEADER_LEN);
94 static struct icmp6_hdr *
95 pull_icmpv6(struct ofpbuf *packet)
97 return ofpbuf_try_pull(packet, sizeof(struct icmp6_hdr));
101 parse_mpls(struct ofpbuf *b, struct flow *flow)
106 while ((mh = ofpbuf_try_pull(b, sizeof *mh))) {
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 flow->tcp_flags = tcp->tcp_ctl & htons(0x0fff);
260 packet->l7 = b->data;
265 parse_udp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
267 const struct udp_header *udp = pull_udp(b);
269 flow->tp_src = udp->udp_src;
270 flow->tp_dst = udp->udp_dst;
271 packet->l7 = b->data;
276 parse_sctp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
278 const struct sctp_header *sctp = pull_sctp(b);
280 flow->tp_src = sctp->sctp_src;
281 flow->tp_dst = sctp->sctp_dst;
282 packet->l7 = b->data;
287 parse_icmpv6(struct ofpbuf *b, struct flow *flow)
289 const struct icmp6_hdr *icmp = pull_icmpv6(b);
295 /* The ICMPv6 type and code fields use the 16-bit transport port
296 * fields, so we need to store them in 16-bit network byte order. */
297 flow->tp_src = htons(icmp->icmp6_type);
298 flow->tp_dst = htons(icmp->icmp6_code);
300 if (icmp->icmp6_code == 0 &&
301 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
302 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
303 const struct in6_addr *nd_target;
305 nd_target = ofpbuf_try_pull(b, sizeof *nd_target);
309 flow->nd_target = *nd_target;
311 while (b->size >= 8) {
312 /* The minimum size of an option is 8 bytes, which also is
313 * the size of Ethernet link-layer options. */
314 const struct nd_opt_hdr *nd_opt = b->data;
315 int opt_len = nd_opt->nd_opt_len * 8;
317 if (!opt_len || opt_len > b->size) {
321 /* Store the link layer address if the appropriate option is
322 * provided. It is considered an error if the same link
323 * layer option is specified twice. */
324 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
326 if (eth_addr_is_zero(flow->arp_sha)) {
327 memcpy(flow->arp_sha, nd_opt + 1, ETH_ADDR_LEN);
331 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
333 if (eth_addr_is_zero(flow->arp_tha)) {
334 memcpy(flow->arp_tha, nd_opt + 1, ETH_ADDR_LEN);
340 if (!ofpbuf_try_pull(b, opt_len)) {
349 memset(&flow->nd_target, 0, sizeof(flow->nd_target));
350 memset(flow->arp_sha, 0, sizeof(flow->arp_sha));
351 memset(flow->arp_tha, 0, sizeof(flow->arp_tha));
357 /* Initializes 'flow' members from 'packet', 'skb_priority', 'tnl', and
360 * Initializes 'packet' header pointers as follows:
362 * - packet->l2 to the start of the Ethernet header.
364 * - packet->l2_5 to the start of the MPLS shim header.
366 * - packet->l3 to just past the Ethernet header, or just past the
367 * vlan_header if one is present, to the first byte of the payload of the
370 * - packet->l4 to just past the IPv4 header, if one is present and has a
371 * correct length, and otherwise NULL.
373 * - packet->l7 to just past the TCP/UDP/SCTP/ICMP header, if one is
374 * present and has a correct length, and otherwise NULL.
377 flow_extract(struct ofpbuf *packet, uint32_t skb_priority, uint32_t pkt_mark,
378 const struct flow_tnl *tnl, const union flow_in_port *in_port,
381 struct ofpbuf b = *packet;
382 struct eth_header *eth;
384 COVERAGE_INC(flow_extract);
386 memset(flow, 0, sizeof *flow);
389 ovs_assert(tnl != &flow->tunnel);
393 flow->in_port = *in_port;
395 flow->skb_priority = skb_priority;
396 flow->pkt_mark = pkt_mark;
404 if (b.size < sizeof *eth) {
410 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
411 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
413 /* dl_type, vlan_tci. */
414 ofpbuf_pull(&b, ETH_ADDR_LEN * 2);
415 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
416 parse_vlan(&b, flow);
418 flow->dl_type = parse_ethertype(&b);
420 /* Parse mpls, copy l3 ttl. */
421 if (eth_type_mpls(flow->dl_type)) {
422 packet->l2_5 = b.data;
423 parse_mpls(&b, flow);
428 if (flow->dl_type == htons(ETH_TYPE_IP)) {
429 const struct ip_header *nh = pull_ip(&b);
433 flow->nw_src = get_16aligned_be32(&nh->ip_src);
434 flow->nw_dst = get_16aligned_be32(&nh->ip_dst);
435 flow->nw_proto = nh->ip_proto;
437 flow->nw_tos = nh->ip_tos;
438 if (IP_IS_FRAGMENT(nh->ip_frag_off)) {
439 flow->nw_frag = FLOW_NW_FRAG_ANY;
440 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
441 flow->nw_frag |= FLOW_NW_FRAG_LATER;
444 flow->nw_ttl = nh->ip_ttl;
446 if (!(nh->ip_frag_off & htons(IP_FRAG_OFF_MASK))) {
447 if (flow->nw_proto == IPPROTO_TCP) {
448 parse_tcp(packet, &b, flow);
449 } else if (flow->nw_proto == IPPROTO_UDP) {
450 parse_udp(packet, &b, flow);
451 } else if (flow->nw_proto == IPPROTO_SCTP) {
452 parse_sctp(packet, &b, flow);
453 } else if (flow->nw_proto == IPPROTO_ICMP) {
454 const struct icmp_header *icmp = pull_icmp(&b);
456 flow->tp_src = htons(icmp->icmp_type);
457 flow->tp_dst = htons(icmp->icmp_code);
463 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
464 if (parse_ipv6(&b, flow)) {
469 if (flow->nw_proto == IPPROTO_TCP) {
470 parse_tcp(packet, &b, flow);
471 } else if (flow->nw_proto == IPPROTO_UDP) {
472 parse_udp(packet, &b, flow);
473 } else if (flow->nw_proto == IPPROTO_SCTP) {
474 parse_sctp(packet, &b, flow);
475 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
476 if (parse_icmpv6(&b, flow)) {
480 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
481 flow->dl_type == htons(ETH_TYPE_RARP)) {
482 const struct arp_eth_header *arp = pull_arp(&b);
483 if (arp && arp->ar_hrd == htons(1)
484 && arp->ar_pro == htons(ETH_TYPE_IP)
485 && arp->ar_hln == ETH_ADDR_LEN
486 && arp->ar_pln == 4) {
487 /* We only match on the lower 8 bits of the opcode. */
488 if (ntohs(arp->ar_op) <= 0xff) {
489 flow->nw_proto = ntohs(arp->ar_op);
492 flow->nw_src = get_16aligned_be32(&arp->ar_spa);
493 flow->nw_dst = get_16aligned_be32(&arp->ar_tpa);
494 memcpy(flow->arp_sha, arp->ar_sha, ETH_ADDR_LEN);
495 memcpy(flow->arp_tha, arp->ar_tha, ETH_ADDR_LEN);
500 /* For every bit of a field that is wildcarded in 'wildcards', sets the
501 * corresponding bit in 'flow' to zero. */
503 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
505 uint32_t *flow_u32 = (uint32_t *) flow;
506 const uint32_t *wc_u32 = (const uint32_t *) &wildcards->masks;
509 for (i = 0; i < FLOW_U32S; i++) {
510 flow_u32[i] &= wc_u32[i];
514 /* Initializes 'fmd' with the metadata found in 'flow'. */
516 flow_get_metadata(const struct flow *flow, struct flow_metadata *fmd)
518 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 22);
520 fmd->tun_id = flow->tunnel.tun_id;
521 fmd->tun_src = flow->tunnel.ip_src;
522 fmd->tun_dst = flow->tunnel.ip_dst;
523 fmd->metadata = flow->metadata;
524 memcpy(fmd->regs, flow->regs, sizeof fmd->regs);
525 fmd->pkt_mark = flow->pkt_mark;
526 fmd->in_port = flow->in_port.ofp_port;
530 flow_to_string(const struct flow *flow)
532 struct ds ds = DS_EMPTY_INITIALIZER;
533 flow_format(&ds, flow);
538 flow_tun_flag_to_string(uint32_t flags)
541 case FLOW_TNL_F_DONT_FRAGMENT:
543 case FLOW_TNL_F_CSUM:
553 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
554 uint32_t flags, char del)
562 uint32_t bit = rightmost_1bit(flags);
565 s = bit_to_string(bit);
567 ds_put_format(ds, "%s%c", s, del);
576 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
582 flow_format(struct ds *ds, const struct flow *flow)
586 match_wc_init(&match, flow);
587 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
591 flow_print(FILE *stream, const struct flow *flow)
593 char *s = flow_to_string(flow);
598 /* flow_wildcards functions. */
600 /* Initializes 'wc' as a set of wildcards that matches every packet. */
602 flow_wildcards_init_catchall(struct flow_wildcards *wc)
604 memset(&wc->masks, 0, sizeof wc->masks);
607 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
610 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
612 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
615 for (i = 0; i < FLOW_U32S; i++) {
623 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
624 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
625 * in 'src1' or 'src2' or both. */
627 flow_wildcards_and(struct flow_wildcards *dst,
628 const struct flow_wildcards *src1,
629 const struct flow_wildcards *src2)
631 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
632 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
633 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
636 for (i = 0; i < FLOW_U32S; i++) {
637 dst_u32[i] = src1_u32[i] & src2_u32[i];
641 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
642 * is, a bit or a field is wildcarded in 'dst' if it is neither
643 * wildcarded in 'src1' nor 'src2'. */
645 flow_wildcards_or(struct flow_wildcards *dst,
646 const struct flow_wildcards *src1,
647 const struct flow_wildcards *src2)
649 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
650 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
651 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
654 for (i = 0; i < FLOW_U32S; i++) {
655 dst_u32[i] = src1_u32[i] | src2_u32[i];
659 /* Perform a bitwise OR of miniflow 'src' flow data with the equivalent
660 * fields in 'dst', storing the result in 'dst'. */
662 flow_union_with_miniflow(struct flow *dst, const struct miniflow *src)
664 uint32_t *dst_u32 = (uint32_t *) dst;
668 for (map = src->map; map; map = zero_rightmost_1bit(map)) {
669 dst_u32[raw_ctz(map)] |= src->values[ofs++];
673 /* Fold minimask 'mask''s wildcard mask into 'wc's wildcard mask. */
675 flow_wildcards_fold_minimask(struct flow_wildcards *wc,
676 const struct minimask *mask)
678 flow_union_with_miniflow(&wc->masks, &mask->masks);
681 /* Returns a hash of the wildcards in 'wc'. */
683 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
685 return flow_hash(&wc->masks, basis);
688 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
691 flow_wildcards_equal(const struct flow_wildcards *a,
692 const struct flow_wildcards *b)
694 return flow_equal(&a->masks, &b->masks);
697 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
698 * 'b', false otherwise. */
700 flow_wildcards_has_extra(const struct flow_wildcards *a,
701 const struct flow_wildcards *b)
703 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
704 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
707 for (i = 0; i < FLOW_U32S; i++) {
708 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
715 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
716 * in 'wc' do not need to be equal in 'a' and 'b'. */
718 flow_equal_except(const struct flow *a, const struct flow *b,
719 const struct flow_wildcards *wc)
721 const uint32_t *a_u32 = (const uint32_t *) a;
722 const uint32_t *b_u32 = (const uint32_t *) b;
723 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
726 for (i = 0; i < FLOW_U32S; i++) {
727 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
734 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
735 * (A 0-bit indicates a wildcard bit.) */
737 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
739 wc->masks.regs[idx] = mask;
742 /* Hashes 'flow' based on its L2 through L4 protocol information. */
744 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
749 struct in6_addr ipv6_addr;
754 uint8_t eth_addr[ETH_ADDR_LEN];
760 memset(&fields, 0, sizeof fields);
761 for (i = 0; i < ETH_ADDR_LEN; i++) {
762 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
764 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
765 fields.eth_type = flow->dl_type;
767 /* UDP source and destination port are not taken into account because they
768 * will not necessarily be symmetric in a bidirectional flow. */
769 if (fields.eth_type == htons(ETH_TYPE_IP)) {
770 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
771 fields.ip_proto = flow->nw_proto;
772 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
773 fields.tp_port = flow->tp_src ^ flow->tp_dst;
775 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
776 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
777 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
778 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
780 for (i=0; i<16; i++) {
781 ipv6_addr[i] = a[i] ^ b[i];
783 fields.ip_proto = flow->nw_proto;
784 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
785 fields.tp_port = flow->tp_src ^ flow->tp_dst;
788 return jhash_bytes(&fields, sizeof fields, basis);
791 /* Initialize a flow with random fields that matter for nx_hash_fields. */
793 flow_random_hash_fields(struct flow *flow)
795 uint16_t rnd = random_uint16();
797 /* Initialize to all zeros. */
798 memset(flow, 0, sizeof *flow);
800 eth_addr_random(flow->dl_src);
801 eth_addr_random(flow->dl_dst);
803 flow->vlan_tci = (OVS_FORCE ovs_be16) (random_uint16() & VLAN_VID_MASK);
805 /* Make most of the random flows IPv4, some IPv6, and rest random. */
806 flow->dl_type = rnd < 0x8000 ? htons(ETH_TYPE_IP) :
807 rnd < 0xc000 ? htons(ETH_TYPE_IPV6) : (OVS_FORCE ovs_be16)rnd;
809 if (dl_type_is_ip_any(flow->dl_type)) {
810 if (flow->dl_type == htons(ETH_TYPE_IP)) {
811 flow->nw_src = (OVS_FORCE ovs_be32)random_uint32();
812 flow->nw_dst = (OVS_FORCE ovs_be32)random_uint32();
814 random_bytes(&flow->ipv6_src, sizeof flow->ipv6_src);
815 random_bytes(&flow->ipv6_dst, sizeof flow->ipv6_dst);
817 /* Make most of IP flows TCP, some UDP or SCTP, and rest random. */
818 rnd = random_uint16();
819 flow->nw_proto = rnd < 0x8000 ? IPPROTO_TCP :
820 rnd < 0xc000 ? IPPROTO_UDP :
821 rnd < 0xd000 ? IPPROTO_SCTP : (uint8_t)rnd;
822 if (flow->nw_proto == IPPROTO_TCP ||
823 flow->nw_proto == IPPROTO_UDP ||
824 flow->nw_proto == IPPROTO_SCTP) {
825 flow->tp_src = (OVS_FORCE ovs_be16)random_uint16();
826 flow->tp_dst = (OVS_FORCE ovs_be16)random_uint16();
831 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
833 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
834 enum nx_hash_fields fields)
837 case NX_HASH_FIELDS_ETH_SRC:
838 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
841 case NX_HASH_FIELDS_SYMMETRIC_L4:
842 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
843 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
844 if (flow->dl_type == htons(ETH_TYPE_IP)) {
845 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
846 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
847 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
848 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
849 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
851 if (is_ip_any(flow)) {
852 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
853 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
854 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
856 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
864 /* Hashes the portions of 'flow' designated by 'fields'. */
866 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
871 case NX_HASH_FIELDS_ETH_SRC:
872 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
874 case NX_HASH_FIELDS_SYMMETRIC_L4:
875 return flow_hash_symmetric_l4(flow, basis);
881 /* Returns a string representation of 'fields'. */
883 flow_hash_fields_to_str(enum nx_hash_fields fields)
886 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
887 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
888 default: return "<unknown>";
892 /* Returns true if the value of 'fields' is supported. Otherwise false. */
894 flow_hash_fields_valid(enum nx_hash_fields fields)
896 return fields == NX_HASH_FIELDS_ETH_SRC
897 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
900 /* Returns a hash value for the bits of 'flow' that are active based on
901 * 'wc', given 'basis'. */
903 flow_hash_in_wildcards(const struct flow *flow,
904 const struct flow_wildcards *wc, uint32_t basis)
906 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
907 const uint32_t *flow_u32 = (const uint32_t *) flow;
912 for (i = 0; i < FLOW_U32S; i++) {
913 hash = mhash_add(hash, flow_u32[i] & wc_u32[i]);
915 return mhash_finish(hash, 4 * FLOW_U32S);
918 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
919 * OpenFlow 1.0 "dl_vlan" value:
921 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
922 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
923 * 'flow' previously matched packets without a VLAN header).
925 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
926 * without a VLAN tag.
928 * - Other values of 'vid' should not be used. */
930 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
932 if (vid == htons(OFP10_VLAN_NONE)) {
933 flow->vlan_tci = htons(0);
935 vid &= htons(VLAN_VID_MASK);
936 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
937 flow->vlan_tci |= htons(VLAN_CFI) | vid;
941 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
942 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
945 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
947 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
948 flow->vlan_tci &= ~mask;
949 flow->vlan_tci |= vid & mask;
952 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
955 * This function has no effect on the VLAN ID that 'flow' matches.
957 * After calling this function, 'flow' will not match packets without a VLAN
960 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
963 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
964 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
967 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
968 * as an OpenFlow 1.1 "mpls_label" value. */
970 flow_set_mpls_label(struct flow *flow, ovs_be32 label)
972 set_mpls_lse_label(&flow->mpls_lse, label);
975 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
978 flow_set_mpls_ttl(struct flow *flow, uint8_t ttl)
980 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
983 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
986 flow_set_mpls_tc(struct flow *flow, uint8_t tc)
988 set_mpls_lse_tc(&flow->mpls_lse, tc);
991 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
993 flow_set_mpls_bos(struct flow *flow, uint8_t bos)
995 set_mpls_lse_bos(&flow->mpls_lse, bos);
998 /* Puts into 'b' a packet that flow_extract() would parse as having the given
1001 * (This is useful only for testing, obviously, and the packet isn't really
1002 * valid. It hasn't got some checksums filled in, for one, and lots of fields
1003 * are just zeroed.) */
1005 flow_compose(struct ofpbuf *b, const struct flow *flow)
1007 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
1008 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
1009 struct eth_header *eth = b->l2;
1010 eth->eth_type = htons(b->size);
1014 if (flow->vlan_tci & htons(VLAN_CFI)) {
1015 eth_push_vlan(b, flow->vlan_tci);
1018 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1019 struct ip_header *ip;
1021 b->l3 = ip = ofpbuf_put_zeros(b, sizeof *ip);
1022 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
1023 ip->ip_tos = flow->nw_tos;
1024 ip->ip_ttl = flow->nw_ttl;
1025 ip->ip_proto = flow->nw_proto;
1026 put_16aligned_be32(&ip->ip_src, flow->nw_src);
1027 put_16aligned_be32(&ip->ip_dst, flow->nw_dst);
1029 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
1030 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
1031 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
1032 ip->ip_frag_off |= htons(100);
1035 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
1036 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1037 if (flow->nw_proto == IPPROTO_TCP) {
1038 struct tcp_header *tcp;
1040 b->l4 = tcp = ofpbuf_put_zeros(b, sizeof *tcp);
1041 tcp->tcp_src = flow->tp_src;
1042 tcp->tcp_dst = flow->tp_dst;
1043 tcp->tcp_ctl = TCP_CTL(ntohs(flow->tcp_flags), 5);
1044 } else if (flow->nw_proto == IPPROTO_UDP) {
1045 struct udp_header *udp;
1047 b->l4 = udp = ofpbuf_put_zeros(b, sizeof *udp);
1048 udp->udp_src = flow->tp_src;
1049 udp->udp_dst = flow->tp_dst;
1050 } else if (flow->nw_proto == IPPROTO_SCTP) {
1051 struct sctp_header *sctp;
1053 b->l4 = sctp = ofpbuf_put_zeros(b, sizeof *sctp);
1054 sctp->sctp_src = flow->tp_src;
1055 sctp->sctp_dst = flow->tp_dst;
1056 } else if (flow->nw_proto == IPPROTO_ICMP) {
1057 struct icmp_header *icmp;
1059 b->l4 = icmp = ofpbuf_put_zeros(b, sizeof *icmp);
1060 icmp->icmp_type = ntohs(flow->tp_src);
1061 icmp->icmp_code = ntohs(flow->tp_dst);
1062 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
1064 b->l7 = ofpbuf_tail(b);
1068 ip->ip_tot_len = htons((uint8_t *) b->data + b->size
1069 - (uint8_t *) b->l3);
1070 ip->ip_csum = csum(ip, sizeof *ip);
1071 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1073 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1074 flow->dl_type == htons(ETH_TYPE_RARP)) {
1075 struct arp_eth_header *arp;
1077 b->l3 = arp = ofpbuf_put_zeros(b, sizeof *arp);
1078 arp->ar_hrd = htons(1);
1079 arp->ar_pro = htons(ETH_TYPE_IP);
1080 arp->ar_hln = ETH_ADDR_LEN;
1082 arp->ar_op = htons(flow->nw_proto);
1084 if (flow->nw_proto == ARP_OP_REQUEST ||
1085 flow->nw_proto == ARP_OP_REPLY) {
1086 put_16aligned_be32(&arp->ar_spa, flow->nw_src);
1087 put_16aligned_be32(&arp->ar_tpa, flow->nw_dst);
1088 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
1089 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
1093 if (eth_type_mpls(flow->dl_type)) {
1095 push_mpls(b, flow->dl_type, flow->mpls_lse);
1099 /* Compressed flow. */
1102 miniflow_n_values(const struct miniflow *flow)
1104 return count_1bits(flow->map);
1108 miniflow_alloc_values(struct miniflow *flow, int n)
1110 if (n <= MINI_N_INLINE) {
1111 return flow->inline_values;
1113 COVERAGE_INC(miniflow_malloc);
1114 return xmalloc(n * sizeof *flow->values);
1118 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
1119 * the caller. The caller must have already initialized 'dst->map' properly
1120 * to indicate the nonzero uint32_t elements of 'src'. 'n' must be the number
1121 * of 1-bits in 'dst->map'.
1123 * This function initializes 'dst->values' (either inline if possible or with
1124 * malloc() otherwise) and copies the nonzero uint32_t elements of 'src' into
1127 miniflow_init__(struct miniflow *dst, const struct flow *src, int n)
1129 const uint32_t *src_u32 = (const uint32_t *) src;
1133 dst->values = miniflow_alloc_values(dst, n);
1135 for (map = dst->map; map; map = zero_rightmost_1bit(map)) {
1136 dst->values[ofs++] = src_u32[raw_ctz(map)];
1140 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1141 * with miniflow_destroy(). */
1143 miniflow_init(struct miniflow *dst, const struct flow *src)
1145 const uint32_t *src_u32 = (const uint32_t *) src;
1149 /* Initialize dst->map, counting the number of nonzero elements. */
1153 for (i = 0; i < FLOW_U32S; i++) {
1155 dst->map |= UINT64_C(1) << i;
1160 miniflow_init__(dst, src, n);
1163 /* Initializes 'dst' as a copy of 'src', using 'mask->map' as 'dst''s map. The
1164 * caller must eventually free 'dst' with miniflow_destroy(). */
1166 miniflow_init_with_minimask(struct miniflow *dst, const struct flow *src,
1167 const struct minimask *mask)
1169 dst->map = mask->masks.map;
1170 miniflow_init__(dst, src, miniflow_n_values(dst));
1173 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1174 * with miniflow_destroy(). */
1176 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
1178 int n = miniflow_n_values(src);
1179 dst->map = src->map;
1180 dst->values = miniflow_alloc_values(dst, n);
1181 memcpy(dst->values, src->values, n * sizeof *dst->values);
1184 /* Initializes 'dst' with the data in 'src', destroying 'src'.
1185 * The caller must eventually free 'dst' with miniflow_destroy(). */
1187 miniflow_move(struct miniflow *dst, struct miniflow *src)
1189 if (src->values == src->inline_values) {
1190 dst->values = dst->inline_values;
1191 memcpy(dst->values, src->values,
1192 miniflow_n_values(src) * sizeof *dst->values);
1194 dst->values = src->values;
1196 dst->map = src->map;
1199 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
1200 * itself resides; the caller is responsible for that. */
1202 miniflow_destroy(struct miniflow *flow)
1204 if (flow->values != flow->inline_values) {
1209 /* Initializes 'dst' as a copy of 'src'. */
1211 miniflow_expand(const struct miniflow *src, struct flow *dst)
1213 memset(dst, 0, sizeof *dst);
1214 flow_union_with_miniflow(dst, src);
1217 static const uint32_t *
1218 miniflow_get__(const struct miniflow *flow, unsigned int u32_ofs)
1220 if (!(flow->map & (UINT64_C(1) << u32_ofs))) {
1221 static const uint32_t zero = 0;
1224 return flow->values +
1225 count_1bits(flow->map & ((UINT64_C(1) << u32_ofs) - 1));
1228 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1229 * were expanded into a "struct flow". */
1231 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1233 return *miniflow_get__(flow, u32_ofs);
1236 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
1237 * expanded into a "struct flow". */
1239 miniflow_get_be16(const struct miniflow *flow, unsigned int u8_ofs)
1241 const uint32_t *u32p = miniflow_get__(flow, u8_ofs / 4);
1242 const ovs_be16 *be16p = (const ovs_be16 *) u32p;
1243 return be16p[u8_ofs % 4 != 0];
1246 /* Returns the VID within the vlan_tci member of the "struct flow" represented
1249 miniflow_get_vid(const struct miniflow *flow)
1251 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
1252 return vlan_tci_to_vid(tci);
1255 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1257 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1259 const uint32_t *ap = a->values;
1260 const uint32_t *bp = b->values;
1261 const uint64_t a_map = a->map;
1262 const uint64_t b_map = b->map;
1265 if (a_map == b_map) {
1266 for (map = a_map; map; map = zero_rightmost_1bit(map)) {
1267 if (*ap++ != *bp++) {
1272 for (map = a_map | b_map; map; map = zero_rightmost_1bit(map)) {
1273 uint64_t bit = rightmost_1bit(map);
1274 uint64_t a_value = a_map & bit ? *ap++ : 0;
1275 uint64_t b_value = b_map & bit ? *bp++ : 0;
1277 if (a_value != b_value) {
1286 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1287 * in 'mask', false if they differ. */
1289 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1290 const struct minimask *mask)
1295 p = mask->masks.values;
1297 for (map = mask->masks.map; map; map = zero_rightmost_1bit(map)) {
1298 int ofs = raw_ctz(map);
1300 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p) {
1309 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1310 * in 'mask', false if they differ. */
1312 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1313 const struct minimask *mask)
1315 const uint32_t *b_u32 = (const uint32_t *) b;
1319 p = mask->masks.values;
1321 for (map = mask->masks.map; map; map = zero_rightmost_1bit(map)) {
1322 int ofs = raw_ctz(map);
1324 if ((miniflow_get(a, ofs) ^ b_u32[ofs]) & *p) {
1333 /* Returns a hash value for 'flow', given 'basis'. */
1335 miniflow_hash(const struct miniflow *flow, uint32_t basis)
1337 const uint32_t *p = flow->values;
1338 uint32_t hash = basis;
1339 uint64_t hash_map = 0;
1342 for (map = flow->map; map; map = zero_rightmost_1bit(map)) {
1344 hash = mhash_add(hash, *p);
1345 hash_map |= rightmost_1bit(map);
1349 hash = mhash_add(hash, hash_map);
1350 hash = mhash_add(hash, hash_map >> 32);
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;
1370 for (map = mask->masks.map; map; map = zero_rightmost_1bit(map)) {
1372 int ofs = raw_ctz(map);
1373 hash = mhash_add(hash, miniflow_get(flow, ofs) & *p);
1378 return mhash_finish(hash, (p - mask->masks.values) * 4);
1381 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1382 * 'mask', given 'basis'.
1384 * The hash values returned by this function are the same as those returned by
1385 * miniflow_hash_in_minimask(), only the form of the arguments differ. */
1387 flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
1390 const uint32_t *flow_u32 = (const uint32_t *)flow;
1391 const uint32_t *p = mask->masks.values;
1396 for (map = mask->masks.map; map; map = zero_rightmost_1bit(map)) {
1398 hash = mhash_add(hash, flow_u32[raw_ctz(map)] & *p);
1403 return mhash_finish(hash, (p - mask->masks.values) * 4);
1406 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1407 * with minimask_destroy(). */
1409 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1411 miniflow_init(&mask->masks, &wc->masks);
1414 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1415 * with minimask_destroy(). */
1417 minimask_clone(struct minimask *dst, const struct minimask *src)
1419 miniflow_clone(&dst->masks, &src->masks);
1422 /* Initializes 'dst' with the data in 'src', destroying 'src'.
1423 * The caller must eventually free 'dst' with minimask_destroy(). */
1425 minimask_move(struct minimask *dst, struct minimask *src)
1427 miniflow_move(&dst->masks, &src->masks);
1430 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1432 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1433 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1435 minimask_combine(struct minimask *dst_,
1436 const struct minimask *a_, const struct minimask *b_,
1437 uint32_t storage[FLOW_U32S])
1439 struct miniflow *dst = &dst_->masks;
1440 const struct miniflow *a = &a_->masks;
1441 const struct miniflow *b = &b_->masks;
1445 dst->values = storage;
1448 for (map = a->map & b->map; map; map = zero_rightmost_1bit(map)) {
1449 int ofs = raw_ctz(map);
1450 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1453 dst->map |= rightmost_1bit(map);
1454 dst->values[n++] = mask;
1459 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1460 * itself resides; the caller is responsible for that. */
1462 minimask_destroy(struct minimask *mask)
1464 miniflow_destroy(&mask->masks);
1467 /* Initializes 'dst' as a copy of 'src'. */
1469 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1471 miniflow_expand(&mask->masks, &wc->masks);
1474 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1475 * were expanded into a "struct flow_wildcards". */
1477 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1479 return miniflow_get(&mask->masks, u32_ofs);
1482 /* Returns the VID mask within the vlan_tci member of the "struct
1483 * flow_wildcards" represented by 'mask'. */
1485 minimask_get_vid_mask(const struct minimask *mask)
1487 return miniflow_get_vid(&mask->masks);
1490 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1492 minimask_equal(const struct minimask *a, const struct minimask *b)
1494 return miniflow_equal(&a->masks, &b->masks);
1497 /* Returns a hash value for 'mask', given 'basis'. */
1499 minimask_hash(const struct minimask *mask, uint32_t basis)
1501 return miniflow_hash(&mask->masks, basis);
1504 /* Returns true if at least one bit is wildcarded in 'a_' but not in 'b_',
1505 * false otherwise. */
1507 minimask_has_extra(const struct minimask *a_, const struct minimask *b_)
1509 const struct miniflow *a = &a_->masks;
1510 const struct miniflow *b = &b_->masks;
1513 for (map = a->map | b->map; map; map = zero_rightmost_1bit(map)) {
1514 int ofs = raw_ctz(map);
1515 uint32_t a_u32 = miniflow_get(a, ofs);
1516 uint32_t b_u32 = miniflow_get(b, ofs);
1518 if ((a_u32 & b_u32) != b_u32) {
1526 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
1529 minimask_is_catchall(const struct minimask *mask_)
1531 const struct miniflow *mask = &mask_->masks;
1532 const uint32_t *p = mask->values;
1535 for (map = mask->map; map; map = zero_rightmost_1bit(map)) {