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)
108 while ((mh = ofpbuf_try_pull(b, sizeof *mh))) {
111 flow->mpls_lse = mh->mpls_lse;
113 if (mh->mpls_lse & htonl(MPLS_BOS_MASK)) {
120 parse_vlan(struct ofpbuf *b, struct flow *flow)
123 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
127 if (b->size >= sizeof(struct qtag_prefix) + sizeof(ovs_be16)) {
128 struct qtag_prefix *qp = ofpbuf_pull(b, sizeof *qp);
129 flow->vlan_tci = qp->tci | htons(VLAN_CFI);
134 parse_ethertype(struct ofpbuf *b)
136 struct llc_snap_header *llc;
139 proto = *(ovs_be16 *) ofpbuf_pull(b, sizeof proto);
140 if (ntohs(proto) >= ETH_TYPE_MIN) {
144 if (b->size < sizeof *llc) {
145 return htons(FLOW_DL_TYPE_NONE);
149 if (llc->llc.llc_dsap != LLC_DSAP_SNAP
150 || llc->llc.llc_ssap != LLC_SSAP_SNAP
151 || llc->llc.llc_cntl != LLC_CNTL_SNAP
152 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
153 sizeof llc->snap.snap_org)) {
154 return htons(FLOW_DL_TYPE_NONE);
157 ofpbuf_pull(b, sizeof *llc);
159 if (ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN) {
160 return llc->snap.snap_type;
163 return htons(FLOW_DL_TYPE_NONE);
167 parse_ipv6(struct ofpbuf *packet, struct flow *flow)
169 const struct ovs_16aligned_ip6_hdr *nh;
173 nh = ofpbuf_try_pull(packet, sizeof *nh);
178 nexthdr = nh->ip6_nxt;
180 memcpy(&flow->ipv6_src, &nh->ip6_src, sizeof flow->ipv6_src);
181 memcpy(&flow->ipv6_dst, &nh->ip6_dst, sizeof flow->ipv6_dst);
183 tc_flow = get_16aligned_be32(&nh->ip6_flow);
184 flow->nw_tos = ntohl(tc_flow) >> 20;
185 flow->ipv6_label = tc_flow & htonl(IPV6_LABEL_MASK);
186 flow->nw_ttl = nh->ip6_hlim;
187 flow->nw_proto = IPPROTO_NONE;
190 if ((nexthdr != IPPROTO_HOPOPTS)
191 && (nexthdr != IPPROTO_ROUTING)
192 && (nexthdr != IPPROTO_DSTOPTS)
193 && (nexthdr != IPPROTO_AH)
194 && (nexthdr != IPPROTO_FRAGMENT)) {
195 /* It's either a terminal header (e.g., TCP, UDP) or one we
196 * don't understand. In either case, we're done with the
197 * packet, so use it to fill in 'nw_proto'. */
201 /* We only verify that at least 8 bytes of the next header are
202 * available, but many of these headers are longer. Ensure that
203 * accesses within the extension header are within those first 8
204 * bytes. All extension headers are required to be at least 8
206 if (packet->size < 8) {
210 if ((nexthdr == IPPROTO_HOPOPTS)
211 || (nexthdr == IPPROTO_ROUTING)
212 || (nexthdr == IPPROTO_DSTOPTS)) {
213 /* These headers, while different, have the fields we care about
214 * in the same location and with the same interpretation. */
215 const struct ip6_ext *ext_hdr = packet->data;
216 nexthdr = ext_hdr->ip6e_nxt;
217 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 1) * 8)) {
220 } else if (nexthdr == IPPROTO_AH) {
221 /* A standard AH definition isn't available, but the fields
222 * we care about are in the same location as the generic
223 * option header--only the header length is calculated
225 const struct ip6_ext *ext_hdr = packet->data;
226 nexthdr = ext_hdr->ip6e_nxt;
227 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 2) * 4)) {
230 } else if (nexthdr == IPPROTO_FRAGMENT) {
231 const struct ovs_16aligned_ip6_frag *frag_hdr = packet->data;
233 nexthdr = frag_hdr->ip6f_nxt;
234 if (!ofpbuf_try_pull(packet, sizeof *frag_hdr)) {
238 /* We only process the first fragment. */
239 if (frag_hdr->ip6f_offlg != htons(0)) {
240 flow->nw_frag = FLOW_NW_FRAG_ANY;
241 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
242 flow->nw_frag |= FLOW_NW_FRAG_LATER;
243 nexthdr = IPPROTO_FRAGMENT;
250 flow->nw_proto = nexthdr;
255 parse_tcp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
257 const struct tcp_header *tcp = pull_tcp(b);
259 flow->tp_src = tcp->tcp_src;
260 flow->tp_dst = tcp->tcp_dst;
261 packet->l7 = b->data;
266 parse_udp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
268 const struct udp_header *udp = pull_udp(b);
270 flow->tp_src = udp->udp_src;
271 flow->tp_dst = udp->udp_dst;
272 packet->l7 = b->data;
277 parse_sctp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
279 const struct sctp_header *sctp = pull_sctp(b);
281 flow->tp_src = sctp->sctp_src;
282 flow->tp_dst = sctp->sctp_dst;
283 packet->l7 = b->data;
288 parse_icmpv6(struct ofpbuf *b, struct flow *flow)
290 const struct icmp6_hdr *icmp = pull_icmpv6(b);
296 /* The ICMPv6 type and code fields use the 16-bit transport port
297 * fields, so we need to store them in 16-bit network byte order. */
298 flow->tp_src = htons(icmp->icmp6_type);
299 flow->tp_dst = htons(icmp->icmp6_code);
301 if (icmp->icmp6_code == 0 &&
302 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
303 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
304 const struct in6_addr *nd_target;
306 nd_target = ofpbuf_try_pull(b, sizeof *nd_target);
310 flow->nd_target = *nd_target;
312 while (b->size >= 8) {
313 /* The minimum size of an option is 8 bytes, which also is
314 * the size of Ethernet link-layer options. */
315 const struct nd_opt_hdr *nd_opt = b->data;
316 int opt_len = nd_opt->nd_opt_len * 8;
318 if (!opt_len || opt_len > b->size) {
322 /* Store the link layer address if the appropriate option is
323 * provided. It is considered an error if the same link
324 * layer option is specified twice. */
325 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
327 if (eth_addr_is_zero(flow->arp_sha)) {
328 memcpy(flow->arp_sha, nd_opt + 1, ETH_ADDR_LEN);
332 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
334 if (eth_addr_is_zero(flow->arp_tha)) {
335 memcpy(flow->arp_tha, nd_opt + 1, ETH_ADDR_LEN);
341 if (!ofpbuf_try_pull(b, opt_len)) {
350 memset(&flow->nd_target, 0, sizeof(flow->nd_target));
351 memset(flow->arp_sha, 0, sizeof(flow->arp_sha));
352 memset(flow->arp_tha, 0, sizeof(flow->arp_tha));
358 /* Initializes 'flow' members from 'packet', 'skb_priority', 'tnl', and
361 * Initializes 'packet' header pointers as follows:
363 * - packet->l2 to the start of the Ethernet header.
365 * - packet->l2_5 to the start of the MPLS shim header.
367 * - packet->l3 to just past the Ethernet header, or just past the
368 * vlan_header if one is present, to the first byte of the payload of the
371 * - packet->l4 to just past the IPv4 header, if one is present and has a
372 * correct length, and otherwise NULL.
374 * - packet->l7 to just past the TCP/UDP/SCTP/ICMP header, if one is
375 * present and has a correct length, and otherwise NULL.
378 flow_extract(struct ofpbuf *packet, uint32_t skb_priority, uint32_t pkt_mark,
379 const struct flow_tnl *tnl, const union flow_in_port *in_port,
382 struct ofpbuf b = *packet;
383 struct eth_header *eth;
385 COVERAGE_INC(flow_extract);
387 memset(flow, 0, sizeof *flow);
390 ovs_assert(tnl != &flow->tunnel);
394 flow->in_port = *in_port;
396 flow->skb_priority = skb_priority;
397 flow->pkt_mark = pkt_mark;
405 if (b.size < sizeof *eth) {
411 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
412 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
414 /* dl_type, vlan_tci. */
415 ofpbuf_pull(&b, ETH_ADDR_LEN * 2);
416 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
417 parse_vlan(&b, flow);
419 flow->dl_type = parse_ethertype(&b);
421 /* Parse mpls, copy l3 ttl. */
422 if (eth_type_mpls(flow->dl_type)) {
423 packet->l2_5 = b.data;
424 parse_mpls(&b, flow);
429 if (flow->dl_type == htons(ETH_TYPE_IP)) {
430 const struct ip_header *nh = pull_ip(&b);
434 flow->nw_src = get_16aligned_be32(&nh->ip_src);
435 flow->nw_dst = get_16aligned_be32(&nh->ip_dst);
436 flow->nw_proto = nh->ip_proto;
438 flow->nw_tos = nh->ip_tos;
439 if (IP_IS_FRAGMENT(nh->ip_frag_off)) {
440 flow->nw_frag = FLOW_NW_FRAG_ANY;
441 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
442 flow->nw_frag |= FLOW_NW_FRAG_LATER;
445 flow->nw_ttl = nh->ip_ttl;
447 if (!(nh->ip_frag_off & htons(IP_FRAG_OFF_MASK))) {
448 if (flow->nw_proto == IPPROTO_TCP) {
449 parse_tcp(packet, &b, flow);
450 } else if (flow->nw_proto == IPPROTO_UDP) {
451 parse_udp(packet, &b, flow);
452 } else if (flow->nw_proto == IPPROTO_SCTP) {
453 parse_sctp(packet, &b, flow);
454 } else if (flow->nw_proto == IPPROTO_ICMP) {
455 const struct icmp_header *icmp = pull_icmp(&b);
457 flow->tp_src = htons(icmp->icmp_type);
458 flow->tp_dst = htons(icmp->icmp_code);
464 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
465 if (parse_ipv6(&b, flow)) {
470 if (flow->nw_proto == IPPROTO_TCP) {
471 parse_tcp(packet, &b, flow);
472 } else if (flow->nw_proto == IPPROTO_UDP) {
473 parse_udp(packet, &b, flow);
474 } else if (flow->nw_proto == IPPROTO_SCTP) {
475 parse_sctp(packet, &b, flow);
476 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
477 if (parse_icmpv6(&b, flow)) {
481 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
482 flow->dl_type == htons(ETH_TYPE_RARP)) {
483 const struct arp_eth_header *arp = pull_arp(&b);
484 if (arp && arp->ar_hrd == htons(1)
485 && arp->ar_pro == htons(ETH_TYPE_IP)
486 && arp->ar_hln == ETH_ADDR_LEN
487 && arp->ar_pln == 4) {
488 /* We only match on the lower 8 bits of the opcode. */
489 if (ntohs(arp->ar_op) <= 0xff) {
490 flow->nw_proto = ntohs(arp->ar_op);
493 flow->nw_src = get_16aligned_be32(&arp->ar_spa);
494 flow->nw_dst = get_16aligned_be32(&arp->ar_tpa);
495 memcpy(flow->arp_sha, arp->ar_sha, ETH_ADDR_LEN);
496 memcpy(flow->arp_tha, arp->ar_tha, ETH_ADDR_LEN);
501 /* For every bit of a field that is wildcarded in 'wildcards', sets the
502 * corresponding bit in 'flow' to zero. */
504 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
506 uint32_t *flow_u32 = (uint32_t *) flow;
507 const uint32_t *wc_u32 = (const uint32_t *) &wildcards->masks;
510 for (i = 0; i < FLOW_U32S; i++) {
511 flow_u32[i] &= wc_u32[i];
515 /* Initializes 'fmd' with the metadata found in 'flow'. */
517 flow_get_metadata(const struct flow *flow, struct flow_metadata *fmd)
519 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 21);
521 fmd->tun_id = flow->tunnel.tun_id;
522 fmd->tun_src = flow->tunnel.ip_src;
523 fmd->tun_dst = flow->tunnel.ip_dst;
524 fmd->metadata = flow->metadata;
525 memcpy(fmd->regs, flow->regs, sizeof fmd->regs);
526 fmd->pkt_mark = flow->pkt_mark;
527 fmd->in_port = flow->in_port.ofp_port;
531 flow_to_string(const struct flow *flow)
533 struct ds ds = DS_EMPTY_INITIALIZER;
534 flow_format(&ds, flow);
539 flow_tun_flag_to_string(uint32_t flags)
542 case FLOW_TNL_F_DONT_FRAGMENT:
544 case FLOW_TNL_F_CSUM:
554 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
555 uint32_t flags, char del)
563 uint32_t bit = rightmost_1bit(flags);
566 s = bit_to_string(bit);
568 ds_put_format(ds, "%s%c", s, del);
577 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
583 flow_format(struct ds *ds, const struct flow *flow)
587 match_wc_init(&match, flow);
588 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
592 flow_print(FILE *stream, const struct flow *flow)
594 char *s = flow_to_string(flow);
599 /* flow_wildcards functions. */
601 /* Initializes 'wc' as a set of wildcards that matches every packet. */
603 flow_wildcards_init_catchall(struct flow_wildcards *wc)
605 memset(&wc->masks, 0, sizeof wc->masks);
608 /* Initializes 'wc' as an exact-match set of wildcards; that is, 'wc' does not
609 * wildcard any bits or fields. */
611 flow_wildcards_init_exact(struct flow_wildcards *wc)
613 memset(&wc->masks, 0xff, sizeof wc->masks);
616 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
619 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
621 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
624 for (i = 0; i < FLOW_U32S; i++) {
632 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
633 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
634 * in 'src1' or 'src2' or both. */
636 flow_wildcards_and(struct flow_wildcards *dst,
637 const struct flow_wildcards *src1,
638 const struct flow_wildcards *src2)
640 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
641 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
642 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
645 for (i = 0; i < FLOW_U32S; i++) {
646 dst_u32[i] = src1_u32[i] & src2_u32[i];
650 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
651 * is, a bit or a field is wildcarded in 'dst' if it is neither
652 * wildcarded in 'src1' nor 'src2'. */
654 flow_wildcards_or(struct flow_wildcards *dst,
655 const struct flow_wildcards *src1,
656 const struct flow_wildcards *src2)
658 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
659 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
660 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
663 for (i = 0; i < FLOW_U32S; i++) {
664 dst_u32[i] = src1_u32[i] | src2_u32[i];
668 /* Perform a bitwise OR of miniflow 'src' flow data with the equivalent
669 * fields in 'dst', storing the result in 'dst'. */
671 flow_union_with_miniflow(struct flow *dst, const struct miniflow *src)
673 uint32_t *dst_u32 = (uint32_t *) dst;
678 for (i = 0; i < MINI_N_MAPS; i++) {
681 for (map = src->map[i]; map; map = zero_rightmost_1bit(map)) {
682 dst_u32[raw_ctz(map) + i * 32] |= src->values[ofs++];
687 /* Fold minimask 'mask''s wildcard mask into 'wc's wildcard mask. */
689 flow_wildcards_fold_minimask(struct flow_wildcards *wc,
690 const struct minimask *mask)
692 flow_union_with_miniflow(&wc->masks, &mask->masks);
695 /* Returns a hash of the wildcards in 'wc'. */
697 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
699 return flow_hash(&wc->masks, basis);
702 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
705 flow_wildcards_equal(const struct flow_wildcards *a,
706 const struct flow_wildcards *b)
708 return flow_equal(&a->masks, &b->masks);
711 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
712 * 'b', false otherwise. */
714 flow_wildcards_has_extra(const struct flow_wildcards *a,
715 const struct flow_wildcards *b)
717 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
718 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
721 for (i = 0; i < FLOW_U32S; i++) {
722 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
729 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
730 * in 'wc' do not need to be equal in 'a' and 'b'. */
732 flow_equal_except(const struct flow *a, const struct flow *b,
733 const struct flow_wildcards *wc)
735 const uint32_t *a_u32 = (const uint32_t *) a;
736 const uint32_t *b_u32 = (const uint32_t *) b;
737 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
740 for (i = 0; i < FLOW_U32S; i++) {
741 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
748 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
749 * (A 0-bit indicates a wildcard bit.) */
751 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
753 wc->masks.regs[idx] = mask;
756 /* Hashes 'flow' based on its L2 through L4 protocol information. */
758 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
763 struct in6_addr ipv6_addr;
768 uint8_t eth_addr[ETH_ADDR_LEN];
774 memset(&fields, 0, sizeof fields);
775 for (i = 0; i < ETH_ADDR_LEN; i++) {
776 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
778 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
779 fields.eth_type = flow->dl_type;
781 /* UDP source and destination port are not taken into account because they
782 * will not necessarily be symmetric in a bidirectional flow. */
783 if (fields.eth_type == htons(ETH_TYPE_IP)) {
784 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
785 fields.ip_proto = flow->nw_proto;
786 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
787 fields.tp_port = flow->tp_src ^ flow->tp_dst;
789 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
790 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
791 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
792 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
794 for (i=0; i<16; i++) {
795 ipv6_addr[i] = a[i] ^ b[i];
797 fields.ip_proto = flow->nw_proto;
798 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
799 fields.tp_port = flow->tp_src ^ flow->tp_dst;
802 return jhash_bytes(&fields, sizeof fields, basis);
805 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
807 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
808 enum nx_hash_fields fields)
811 case NX_HASH_FIELDS_ETH_SRC:
812 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
815 case NX_HASH_FIELDS_SYMMETRIC_L4:
816 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
817 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
818 if (flow->dl_type == htons(ETH_TYPE_IP)) {
819 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
820 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
821 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
822 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
823 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
825 if (is_ip_any(flow)) {
826 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
827 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
828 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
830 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
838 /* Hashes the portions of 'flow' designated by 'fields'. */
840 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
845 case NX_HASH_FIELDS_ETH_SRC:
846 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
848 case NX_HASH_FIELDS_SYMMETRIC_L4:
849 return flow_hash_symmetric_l4(flow, basis);
855 /* Returns a string representation of 'fields'. */
857 flow_hash_fields_to_str(enum nx_hash_fields fields)
860 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
861 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
862 default: return "<unknown>";
866 /* Returns true if the value of 'fields' is supported. Otherwise false. */
868 flow_hash_fields_valid(enum nx_hash_fields fields)
870 return fields == NX_HASH_FIELDS_ETH_SRC
871 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
874 /* Returns a hash value for the bits of 'flow' that are active based on
875 * 'wc', given 'basis'. */
877 flow_hash_in_wildcards(const struct flow *flow,
878 const struct flow_wildcards *wc, uint32_t basis)
880 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
881 const uint32_t *flow_u32 = (const uint32_t *) flow;
886 for (i = 0; i < FLOW_U32S; i++) {
887 hash = mhash_add(hash, flow_u32[i] & wc_u32[i]);
889 return mhash_finish(hash, 4 * FLOW_U32S);
892 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
893 * OpenFlow 1.0 "dl_vlan" value:
895 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
896 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
897 * 'flow' previously matched packets without a VLAN header).
899 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
900 * without a VLAN tag.
902 * - Other values of 'vid' should not be used. */
904 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
906 if (vid == htons(OFP10_VLAN_NONE)) {
907 flow->vlan_tci = htons(0);
909 vid &= htons(VLAN_VID_MASK);
910 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
911 flow->vlan_tci |= htons(VLAN_CFI) | vid;
915 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
916 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
919 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
921 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
922 flow->vlan_tci &= ~mask;
923 flow->vlan_tci |= vid & mask;
926 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
929 * This function has no effect on the VLAN ID that 'flow' matches.
931 * After calling this function, 'flow' will not match packets without a VLAN
934 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
937 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
938 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
941 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
942 * as an OpenFlow 1.1 "mpls_label" value. */
944 flow_set_mpls_label(struct flow *flow, ovs_be32 label)
946 set_mpls_lse_label(&flow->mpls_lse, label);
949 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
952 flow_set_mpls_ttl(struct flow *flow, uint8_t ttl)
954 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
957 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
960 flow_set_mpls_tc(struct flow *flow, uint8_t tc)
962 set_mpls_lse_tc(&flow->mpls_lse, tc);
965 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
967 flow_set_mpls_bos(struct flow *flow, uint8_t bos)
969 set_mpls_lse_bos(&flow->mpls_lse, bos);
972 /* Puts into 'b' a packet that flow_extract() would parse as having the given
975 * (This is useful only for testing, obviously, and the packet isn't really
976 * valid. It hasn't got some checksums filled in, for one, and lots of fields
977 * are just zeroed.) */
979 flow_compose(struct ofpbuf *b, const struct flow *flow)
981 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
982 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
983 struct eth_header *eth = b->l2;
984 eth->eth_type = htons(b->size);
988 if (flow->vlan_tci & htons(VLAN_CFI)) {
989 eth_push_vlan(b, flow->vlan_tci);
992 if (flow->dl_type == htons(ETH_TYPE_IP)) {
993 struct ip_header *ip;
995 b->l3 = ip = ofpbuf_put_zeros(b, sizeof *ip);
996 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
997 ip->ip_tos = flow->nw_tos;
998 ip->ip_ttl = flow->nw_ttl;
999 ip->ip_proto = flow->nw_proto;
1000 put_16aligned_be32(&ip->ip_src, flow->nw_src);
1001 put_16aligned_be32(&ip->ip_dst, flow->nw_dst);
1003 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
1004 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
1005 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
1006 ip->ip_frag_off |= htons(100);
1009 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
1010 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1011 if (flow->nw_proto == IPPROTO_TCP) {
1012 struct tcp_header *tcp;
1014 b->l4 = tcp = ofpbuf_put_zeros(b, sizeof *tcp);
1015 tcp->tcp_src = flow->tp_src;
1016 tcp->tcp_dst = flow->tp_dst;
1017 tcp->tcp_ctl = TCP_CTL(0, 5);
1018 } else if (flow->nw_proto == IPPROTO_UDP) {
1019 struct udp_header *udp;
1021 b->l4 = udp = ofpbuf_put_zeros(b, sizeof *udp);
1022 udp->udp_src = flow->tp_src;
1023 udp->udp_dst = flow->tp_dst;
1024 } else if (flow->nw_proto == IPPROTO_SCTP) {
1025 struct sctp_header *sctp;
1027 b->l4 = sctp = ofpbuf_put_zeros(b, sizeof *sctp);
1028 sctp->sctp_src = flow->tp_src;
1029 sctp->sctp_dst = flow->tp_dst;
1030 } else if (flow->nw_proto == IPPROTO_ICMP) {
1031 struct icmp_header *icmp;
1033 b->l4 = icmp = ofpbuf_put_zeros(b, sizeof *icmp);
1034 icmp->icmp_type = ntohs(flow->tp_src);
1035 icmp->icmp_code = ntohs(flow->tp_dst);
1036 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
1041 ip->ip_tot_len = htons((uint8_t *) b->data + b->size
1042 - (uint8_t *) b->l3);
1043 ip->ip_csum = csum(ip, sizeof *ip);
1044 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1046 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1047 flow->dl_type == htons(ETH_TYPE_RARP)) {
1048 struct arp_eth_header *arp;
1050 b->l3 = arp = ofpbuf_put_zeros(b, sizeof *arp);
1051 arp->ar_hrd = htons(1);
1052 arp->ar_pro = htons(ETH_TYPE_IP);
1053 arp->ar_hln = ETH_ADDR_LEN;
1055 arp->ar_op = htons(flow->nw_proto);
1057 if (flow->nw_proto == ARP_OP_REQUEST ||
1058 flow->nw_proto == ARP_OP_REPLY) {
1059 put_16aligned_be32(&arp->ar_spa, flow->nw_src);
1060 put_16aligned_be32(&arp->ar_tpa, flow->nw_dst);
1061 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
1062 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
1066 if (eth_type_mpls(flow->dl_type)) {
1068 push_mpls(b, flow->dl_type, flow->mpls_lse);
1072 /* Compressed flow. */
1075 miniflow_n_values(const struct miniflow *flow)
1080 for (i = 0; i < MINI_N_MAPS; i++) {
1081 n += popcount(flow->map[i]);
1087 miniflow_alloc_values(struct miniflow *flow, int n)
1089 if (n <= MINI_N_INLINE) {
1090 return flow->inline_values;
1092 COVERAGE_INC(miniflow_malloc);
1093 return xmalloc(n * sizeof *flow->values);
1097 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
1098 * the caller. The caller must have already initialized 'dst->map' properly
1099 * to indicate the nonzero uint32_t elements of 'src'. 'n' must be the number
1100 * of 1-bits in 'dst->map'.
1102 * This function initializes 'dst->values' (either inline if possible or with
1103 * malloc() otherwise) and copies the nonzero uint32_t elements of 'src' into
1106 miniflow_init__(struct miniflow *dst, const struct flow *src, int n)
1108 const uint32_t *src_u32 = (const uint32_t *) src;
1112 dst->values = miniflow_alloc_values(dst, n);
1114 for (i = 0; i < MINI_N_MAPS; i++) {
1117 for (map = dst->map[i]; map; map = zero_rightmost_1bit(map)) {
1118 dst->values[ofs++] = src_u32[raw_ctz(map) + i * 32];
1123 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1124 * with miniflow_destroy(). */
1126 miniflow_init(struct miniflow *dst, const struct flow *src)
1128 const uint32_t *src_u32 = (const uint32_t *) src;
1132 /* Initialize dst->map, counting the number of nonzero elements. */
1134 memset(dst->map, 0, sizeof dst->map);
1135 for (i = 0; i < FLOW_U32S; i++) {
1137 dst->map[i / 32] |= 1u << (i % 32);
1142 miniflow_init__(dst, src, n);
1145 /* Initializes 'dst' as a copy of 'src', using 'mask->map' as 'dst''s map. The
1146 * caller must eventually free 'dst' with miniflow_destroy(). */
1148 miniflow_init_with_minimask(struct miniflow *dst, const struct flow *src,
1149 const struct minimask *mask)
1151 memcpy(dst->map, mask->masks.map, sizeof dst->map);
1152 miniflow_init__(dst, src, miniflow_n_values(dst));
1155 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1156 * with miniflow_destroy(). */
1158 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
1160 int n = miniflow_n_values(src);
1161 memcpy(dst->map, src->map, sizeof dst->map);
1162 dst->values = miniflow_alloc_values(dst, n);
1163 memcpy(dst->values, src->values, n * sizeof *dst->values);
1166 /* Initializes 'dst' with the data in 'src', destroying 'src'.
1167 * The caller must eventually free 'dst' with miniflow_destroy(). */
1169 miniflow_move(struct miniflow *dst, struct miniflow *src)
1171 if (src->values == src->inline_values) {
1172 dst->values = dst->inline_values;
1173 memcpy(dst->values, src->values,
1174 miniflow_n_values(src) * sizeof *dst->values);
1176 dst->values = src->values;
1178 memcpy(dst->map, src->map, sizeof dst->map);
1181 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
1182 * itself resides; the caller is responsible for that. */
1184 miniflow_destroy(struct miniflow *flow)
1186 if (flow->values != flow->inline_values) {
1191 /* Initializes 'dst' as a copy of 'src'. */
1193 miniflow_expand(const struct miniflow *src, struct flow *dst)
1195 memset(dst, 0, sizeof *dst);
1196 flow_union_with_miniflow(dst, src);
1199 static const uint32_t *
1200 miniflow_get__(const struct miniflow *flow, unsigned int u32_ofs)
1202 if (!(flow->map[u32_ofs / 32] & (1u << (u32_ofs % 32)))) {
1203 static const uint32_t zero = 0;
1206 const uint32_t *p = flow->values;
1208 BUILD_ASSERT(MINI_N_MAPS == 2);
1210 p += popcount(flow->map[0] & ((1u << u32_ofs) - 1));
1212 p += popcount(flow->map[0]);
1213 p += popcount(flow->map[1] & ((1u << (u32_ofs - 32)) - 1));
1219 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1220 * were expanded into a "struct flow". */
1222 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1224 return *miniflow_get__(flow, u32_ofs);
1227 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
1228 * expanded into a "struct flow". */
1230 miniflow_get_be16(const struct miniflow *flow, unsigned int u8_ofs)
1232 const uint32_t *u32p = miniflow_get__(flow, u8_ofs / 4);
1233 const ovs_be16 *be16p = (const ovs_be16 *) u32p;
1234 return be16p[u8_ofs % 4 != 0];
1237 /* Returns the VID within the vlan_tci member of the "struct flow" represented
1240 miniflow_get_vid(const struct miniflow *flow)
1242 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
1243 return vlan_tci_to_vid(tci);
1246 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1248 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1250 const uint32_t *ap = a->values;
1251 const uint32_t *bp = b->values;
1254 for (i = 0; i < MINI_N_MAPS; i++) {
1255 const uint32_t a_map = a->map[i];
1256 const uint32_t b_map = b->map[i];
1259 if (a_map == b_map) {
1260 for (map = a_map; map; map = zero_rightmost_1bit(map)) {
1261 if (*ap++ != *bp++) {
1266 for (map = a_map | b_map; map; map = zero_rightmost_1bit(map)) {
1267 uint32_t bit = rightmost_1bit(map);
1268 uint32_t a_value = a_map & bit ? *ap++ : 0;
1269 uint32_t b_value = b_map & bit ? *bp++ : 0;
1271 if (a_value != b_value) {
1281 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1282 * in 'mask', false if they differ. */
1284 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1285 const struct minimask *mask)
1290 p = mask->masks.values;
1291 for (i = 0; i < MINI_N_MAPS; i++) {
1294 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1295 int ofs = raw_ctz(map) + i * 32;
1297 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p) {
1307 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1308 * in 'mask', false if they differ. */
1310 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1311 const struct minimask *mask)
1313 const uint32_t *b_u32 = (const uint32_t *) b;
1317 p = mask->masks.values;
1318 for (i = 0; i < MINI_N_MAPS; i++) {
1321 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1322 int ofs = raw_ctz(map) + i * 32;
1324 if ((miniflow_get(a, ofs) ^ b_u32[ofs]) & *p) {
1334 /* Returns a hash value for 'flow', given 'basis'. */
1336 miniflow_hash(const struct miniflow *flow, uint32_t basis)
1338 const uint32_t *p = flow->values;
1339 uint32_t hash = basis;
1342 for (i = 0; i < MINI_N_MAPS; i++) {
1343 uint32_t hash_map = 0;
1346 for (map = flow->map[i]; map; map = zero_rightmost_1bit(map)) {
1348 hash = mhash_add(hash, *p);
1349 hash_map |= rightmost_1bit(map);
1353 hash = mhash_add(hash, hash_map);
1355 return mhash_finish(hash, p - flow->values);
1358 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1359 * 'mask', given 'basis'.
1361 * The hash values returned by this function are the same as those returned by
1362 * flow_hash_in_minimask(), only the form of the arguments differ. */
1364 miniflow_hash_in_minimask(const struct miniflow *flow,
1365 const struct minimask *mask, uint32_t basis)
1367 const uint32_t *p = mask->masks.values;
1372 for (i = 0; i < MINI_N_MAPS; i++) {
1375 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1377 int ofs = raw_ctz(map) + i * 32;
1378 hash = mhash_add(hash, miniflow_get(flow, ofs) & *p);
1384 return mhash_finish(hash, (p - mask->masks.values) * 4);
1387 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1388 * 'mask', given 'basis'.
1390 * The hash values returned by this function are the same as those returned by
1391 * miniflow_hash_in_minimask(), only the form of the arguments differ. */
1393 flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
1396 const uint32_t *flow_u32;
1397 const uint32_t *p = mask->masks.values;
1402 flow_u32 = (const uint32_t *) flow;
1403 for (i = 0; i < MINI_N_MAPS; i++) {
1406 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1408 hash = mhash_add(hash, flow_u32[raw_ctz(map)] & *p);
1415 return mhash_finish(hash, (p - mask->masks.values) * 4);
1418 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1419 * with minimask_destroy(). */
1421 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1423 miniflow_init(&mask->masks, &wc->masks);
1426 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1427 * with minimask_destroy(). */
1429 minimask_clone(struct minimask *dst, const struct minimask *src)
1431 miniflow_clone(&dst->masks, &src->masks);
1434 /* Initializes 'dst' with the data in 'src', destroying 'src'.
1435 * The caller must eventually free 'dst' with minimask_destroy(). */
1437 minimask_move(struct minimask *dst, struct minimask *src)
1439 miniflow_move(&dst->masks, &src->masks);
1442 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1444 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1445 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1447 minimask_combine(struct minimask *dst_,
1448 const struct minimask *a_, const struct minimask *b_,
1449 uint32_t storage[FLOW_U32S])
1451 struct miniflow *dst = &dst_->masks;
1452 const struct miniflow *a = &a_->masks;
1453 const struct miniflow *b = &b_->masks;
1457 dst->values = storage;
1458 for (i = 0; i < MINI_N_MAPS; i++) {
1462 for (map = a->map[i] & b->map[i]; map;
1463 map = zero_rightmost_1bit(map)) {
1464 int ofs = raw_ctz(map) + i * 32;
1465 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1468 dst->map[i] |= rightmost_1bit(map);
1469 dst->values[n++] = mask;
1475 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1476 * itself resides; the caller is responsible for that. */
1478 minimask_destroy(struct minimask *mask)
1480 miniflow_destroy(&mask->masks);
1483 /* Initializes 'dst' as a copy of 'src'. */
1485 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1487 miniflow_expand(&mask->masks, &wc->masks);
1490 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1491 * were expanded into a "struct flow_wildcards". */
1493 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1495 return miniflow_get(&mask->masks, u32_ofs);
1498 /* Returns the VID mask within the vlan_tci member of the "struct
1499 * flow_wildcards" represented by 'mask'. */
1501 minimask_get_vid_mask(const struct minimask *mask)
1503 return miniflow_get_vid(&mask->masks);
1506 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1508 minimask_equal(const struct minimask *a, const struct minimask *b)
1510 return miniflow_equal(&a->masks, &b->masks);
1513 /* Returns a hash value for 'mask', given 'basis'. */
1515 minimask_hash(const struct minimask *mask, uint32_t basis)
1517 return miniflow_hash(&mask->masks, basis);
1520 /* Returns true if at least one bit is wildcarded in 'a_' but not in 'b_',
1521 * false otherwise. */
1523 minimask_has_extra(const struct minimask *a_, const struct minimask *b_)
1525 const struct miniflow *a = &a_->masks;
1526 const struct miniflow *b = &b_->masks;
1529 for (i = 0; i < MINI_N_MAPS; i++) {
1532 for (map = a->map[i] | b->map[i]; map;
1533 map = zero_rightmost_1bit(map)) {
1534 int ofs = raw_ctz(map) + i * 32;
1535 uint32_t a_u32 = miniflow_get(a, ofs);
1536 uint32_t b_u32 = miniflow_get(b, ofs);
1538 if ((a_u32 & b_u32) != b_u32) {
1547 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
1550 minimask_is_catchall(const struct minimask *mask_)
1552 const struct miniflow *mask = &mask_->masks;
1553 const uint32_t *p = mask->values;
1556 for (i = 0; i < MINI_N_MAPS; i++) {
1559 for (map = mask->map[i]; map; map = zero_rightmost_1bit(map)) {