2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 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"
40 #include "unaligned.h"
42 COVERAGE_DEFINE(flow_extract);
43 COVERAGE_DEFINE(miniflow_malloc);
45 /* U32 indices for segmented flow classification. */
46 const uint8_t flow_segment_u32s[4] = {
47 FLOW_SEGMENT_1_ENDS_AT / 4,
48 FLOW_SEGMENT_2_ENDS_AT / 4,
49 FLOW_SEGMENT_3_ENDS_AT / 4,
53 static struct arp_eth_header *
54 pull_arp(struct ofpbuf *packet)
56 return ofpbuf_try_pull(packet, ARP_ETH_HEADER_LEN);
59 static struct ip_header *
60 pull_ip(struct ofpbuf *packet)
62 if (packet->size >= IP_HEADER_LEN) {
63 struct ip_header *ip = packet->data;
64 int ip_len = IP_IHL(ip->ip_ihl_ver) * 4;
65 if (ip_len >= IP_HEADER_LEN && packet->size >= ip_len) {
66 return ofpbuf_pull(packet, ip_len);
72 static struct tcp_header *
73 pull_tcp(struct ofpbuf *packet)
75 if (packet->size >= TCP_HEADER_LEN) {
76 struct tcp_header *tcp = packet->data;
77 int tcp_len = TCP_OFFSET(tcp->tcp_ctl) * 4;
78 if (tcp_len >= TCP_HEADER_LEN && packet->size >= tcp_len) {
79 return ofpbuf_pull(packet, tcp_len);
85 static struct udp_header *
86 pull_udp(struct ofpbuf *packet)
88 return ofpbuf_try_pull(packet, UDP_HEADER_LEN);
91 static struct sctp_header *
92 pull_sctp(struct ofpbuf *packet)
94 return ofpbuf_try_pull(packet, SCTP_HEADER_LEN);
97 static struct icmp_header *
98 pull_icmp(struct ofpbuf *packet)
100 return ofpbuf_try_pull(packet, ICMP_HEADER_LEN);
103 static struct icmp6_hdr *
104 pull_icmpv6(struct ofpbuf *packet)
106 return ofpbuf_try_pull(packet, sizeof(struct icmp6_hdr));
110 parse_mpls(struct ofpbuf *b, struct flow *flow)
115 while ((mh = ofpbuf_try_pull(b, sizeof *mh))) {
116 if (idx < FLOW_MAX_MPLS_LABELS) {
117 flow->mpls_lse[idx++] = mh->mpls_lse;
119 if (mh->mpls_lse & htonl(MPLS_BOS_MASK)) {
126 parse_vlan(struct ofpbuf *b, struct flow *flow)
129 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
133 if (b->size >= sizeof(struct qtag_prefix) + sizeof(ovs_be16)) {
134 struct qtag_prefix *qp = ofpbuf_pull(b, sizeof *qp);
135 flow->vlan_tci = qp->tci | htons(VLAN_CFI);
140 parse_ethertype(struct ofpbuf *b)
142 struct llc_snap_header *llc;
145 proto = *(ovs_be16 *) ofpbuf_pull(b, sizeof proto);
146 if (ntohs(proto) >= ETH_TYPE_MIN) {
150 if (b->size < sizeof *llc) {
151 return htons(FLOW_DL_TYPE_NONE);
155 if (llc->llc.llc_dsap != LLC_DSAP_SNAP
156 || llc->llc.llc_ssap != LLC_SSAP_SNAP
157 || llc->llc.llc_cntl != LLC_CNTL_SNAP
158 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
159 sizeof llc->snap.snap_org)) {
160 return htons(FLOW_DL_TYPE_NONE);
163 ofpbuf_pull(b, sizeof *llc);
165 if (ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN) {
166 return llc->snap.snap_type;
169 return htons(FLOW_DL_TYPE_NONE);
173 parse_ipv6(struct ofpbuf *packet, struct flow *flow)
175 const struct ovs_16aligned_ip6_hdr *nh;
179 nh = ofpbuf_try_pull(packet, sizeof *nh);
184 nexthdr = nh->ip6_nxt;
186 memcpy(&flow->ipv6_src, &nh->ip6_src, sizeof flow->ipv6_src);
187 memcpy(&flow->ipv6_dst, &nh->ip6_dst, sizeof flow->ipv6_dst);
189 tc_flow = get_16aligned_be32(&nh->ip6_flow);
190 flow->nw_tos = ntohl(tc_flow) >> 20;
191 flow->ipv6_label = tc_flow & htonl(IPV6_LABEL_MASK);
192 flow->nw_ttl = nh->ip6_hlim;
193 flow->nw_proto = IPPROTO_NONE;
196 if ((nexthdr != IPPROTO_HOPOPTS)
197 && (nexthdr != IPPROTO_ROUTING)
198 && (nexthdr != IPPROTO_DSTOPTS)
199 && (nexthdr != IPPROTO_AH)
200 && (nexthdr != IPPROTO_FRAGMENT)) {
201 /* It's either a terminal header (e.g., TCP, UDP) or one we
202 * don't understand. In either case, we're done with the
203 * packet, so use it to fill in 'nw_proto'. */
207 /* We only verify that at least 8 bytes of the next header are
208 * available, but many of these headers are longer. Ensure that
209 * accesses within the extension header are within those first 8
210 * bytes. All extension headers are required to be at least 8
212 if (packet->size < 8) {
216 if ((nexthdr == IPPROTO_HOPOPTS)
217 || (nexthdr == IPPROTO_ROUTING)
218 || (nexthdr == IPPROTO_DSTOPTS)) {
219 /* These headers, while different, have the fields we care about
220 * in the same location and with the same interpretation. */
221 const struct ip6_ext *ext_hdr = packet->data;
222 nexthdr = ext_hdr->ip6e_nxt;
223 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 1) * 8)) {
226 } else if (nexthdr == IPPROTO_AH) {
227 /* A standard AH definition isn't available, but the fields
228 * we care about are in the same location as the generic
229 * option header--only the header length is calculated
231 const struct ip6_ext *ext_hdr = packet->data;
232 nexthdr = ext_hdr->ip6e_nxt;
233 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 2) * 4)) {
236 } else if (nexthdr == IPPROTO_FRAGMENT) {
237 const struct ovs_16aligned_ip6_frag *frag_hdr = packet->data;
239 nexthdr = frag_hdr->ip6f_nxt;
240 if (!ofpbuf_try_pull(packet, sizeof *frag_hdr)) {
244 /* We only process the first fragment. */
245 if (frag_hdr->ip6f_offlg != htons(0)) {
246 flow->nw_frag = FLOW_NW_FRAG_ANY;
247 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
248 flow->nw_frag |= FLOW_NW_FRAG_LATER;
249 nexthdr = IPPROTO_FRAGMENT;
256 flow->nw_proto = nexthdr;
261 parse_tcp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
263 const struct tcp_header *tcp = pull_tcp(b);
265 flow->tp_src = tcp->tcp_src;
266 flow->tp_dst = tcp->tcp_dst;
267 flow->tcp_flags = tcp->tcp_ctl & htons(0x0fff);
268 packet->l7 = b->data;
273 parse_udp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
275 const struct udp_header *udp = pull_udp(b);
277 flow->tp_src = udp->udp_src;
278 flow->tp_dst = udp->udp_dst;
279 packet->l7 = b->data;
284 parse_sctp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
286 const struct sctp_header *sctp = pull_sctp(b);
288 flow->tp_src = sctp->sctp_src;
289 flow->tp_dst = sctp->sctp_dst;
290 packet->l7 = b->data;
295 parse_icmpv6(struct ofpbuf *b, struct flow *flow)
297 const struct icmp6_hdr *icmp = pull_icmpv6(b);
303 /* The ICMPv6 type and code fields use the 16-bit transport port
304 * fields, so we need to store them in 16-bit network byte order. */
305 flow->tp_src = htons(icmp->icmp6_type);
306 flow->tp_dst = htons(icmp->icmp6_code);
308 if (icmp->icmp6_code == 0 &&
309 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
310 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
311 const struct in6_addr *nd_target;
313 nd_target = ofpbuf_try_pull(b, sizeof *nd_target);
317 flow->nd_target = *nd_target;
319 while (b->size >= 8) {
320 /* The minimum size of an option is 8 bytes, which also is
321 * the size of Ethernet link-layer options. */
322 const struct nd_opt_hdr *nd_opt = b->data;
323 int opt_len = nd_opt->nd_opt_len * 8;
325 if (!opt_len || opt_len > b->size) {
329 /* Store the link layer address if the appropriate option is
330 * provided. It is considered an error if the same link
331 * layer option is specified twice. */
332 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
334 if (eth_addr_is_zero(flow->arp_sha)) {
335 memcpy(flow->arp_sha, nd_opt + 1, ETH_ADDR_LEN);
339 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
341 if (eth_addr_is_zero(flow->arp_tha)) {
342 memcpy(flow->arp_tha, nd_opt + 1, ETH_ADDR_LEN);
348 if (!ofpbuf_try_pull(b, opt_len)) {
357 memset(&flow->nd_target, 0, sizeof(flow->nd_target));
358 memset(flow->arp_sha, 0, sizeof(flow->arp_sha));
359 memset(flow->arp_tha, 0, sizeof(flow->arp_tha));
365 /* Initializes 'flow' members from 'packet' and 'md'
367 * Initializes 'packet' header pointers as follows:
369 * - packet->l2 to the start of the Ethernet header.
371 * - packet->l2_5 to the start of the MPLS shim header.
373 * - packet->l3 to just past the Ethernet header, or just past the
374 * vlan_header if one is present, to the first byte of the payload of the
377 * - packet->l4 to just past the IPv4 header, if one is present and has a
378 * correct length, and otherwise NULL.
380 * - packet->l7 to just past the TCP/UDP/SCTP/ICMP header, if one is
381 * present and has a correct length, and otherwise NULL.
384 flow_extract(struct ofpbuf *packet, const struct pkt_metadata *md,
387 struct ofpbuf b = *packet;
388 struct eth_header *eth;
390 COVERAGE_INC(flow_extract);
392 memset(flow, 0, sizeof *flow);
395 flow->tunnel = md->tunnel;
396 flow->in_port = md->in_port;
397 flow->skb_priority = md->skb_priority;
398 flow->pkt_mark = md->pkt_mark;
407 if (b.size < sizeof *eth) {
413 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
414 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
416 /* dl_type, vlan_tci. */
417 ofpbuf_pull(&b, ETH_ADDR_LEN * 2);
418 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
419 parse_vlan(&b, flow);
421 flow->dl_type = parse_ethertype(&b);
423 /* Parse mpls, copy l3 ttl. */
424 if (eth_type_mpls(flow->dl_type)) {
425 packet->l2_5 = b.data;
426 parse_mpls(&b, flow);
431 if (flow->dl_type == htons(ETH_TYPE_IP)) {
432 const struct ip_header *nh = pull_ip(&b);
436 flow->nw_src = get_16aligned_be32(&nh->ip_src);
437 flow->nw_dst = get_16aligned_be32(&nh->ip_dst);
438 flow->nw_proto = nh->ip_proto;
440 flow->nw_tos = nh->ip_tos;
441 if (IP_IS_FRAGMENT(nh->ip_frag_off)) {
442 flow->nw_frag = FLOW_NW_FRAG_ANY;
443 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
444 flow->nw_frag |= FLOW_NW_FRAG_LATER;
447 flow->nw_ttl = nh->ip_ttl;
449 if (!(nh->ip_frag_off & htons(IP_FRAG_OFF_MASK))) {
450 if (flow->nw_proto == IPPROTO_TCP) {
451 parse_tcp(packet, &b, flow);
452 } else if (flow->nw_proto == IPPROTO_UDP) {
453 parse_udp(packet, &b, flow);
454 } else if (flow->nw_proto == IPPROTO_SCTP) {
455 parse_sctp(packet, &b, flow);
456 } else if (flow->nw_proto == IPPROTO_ICMP) {
457 const struct icmp_header *icmp = pull_icmp(&b);
459 flow->tp_src = htons(icmp->icmp_type);
460 flow->tp_dst = htons(icmp->icmp_code);
466 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
467 if (parse_ipv6(&b, flow)) {
472 if (flow->nw_proto == IPPROTO_TCP) {
473 parse_tcp(packet, &b, flow);
474 } else if (flow->nw_proto == IPPROTO_UDP) {
475 parse_udp(packet, &b, flow);
476 } else if (flow->nw_proto == IPPROTO_SCTP) {
477 parse_sctp(packet, &b, flow);
478 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
479 if (parse_icmpv6(&b, flow)) {
483 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
484 flow->dl_type == htons(ETH_TYPE_RARP)) {
485 const struct arp_eth_header *arp = pull_arp(&b);
486 if (arp && arp->ar_hrd == htons(1)
487 && arp->ar_pro == htons(ETH_TYPE_IP)
488 && arp->ar_hln == ETH_ADDR_LEN
489 && arp->ar_pln == 4) {
490 /* We only match on the lower 8 bits of the opcode. */
491 if (ntohs(arp->ar_op) <= 0xff) {
492 flow->nw_proto = ntohs(arp->ar_op);
495 flow->nw_src = get_16aligned_be32(&arp->ar_spa);
496 flow->nw_dst = get_16aligned_be32(&arp->ar_tpa);
497 memcpy(flow->arp_sha, arp->ar_sha, ETH_ADDR_LEN);
498 memcpy(flow->arp_tha, arp->ar_tha, ETH_ADDR_LEN);
503 /* For every bit of a field that is wildcarded in 'wildcards', sets the
504 * corresponding bit in 'flow' to zero. */
506 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
508 uint32_t *flow_u32 = (uint32_t *) flow;
509 const uint32_t *wc_u32 = (const uint32_t *) &wildcards->masks;
512 for (i = 0; i < FLOW_U32S; i++) {
513 flow_u32[i] &= wc_u32[i];
518 flow_unwildcard_tp_ports(const struct flow *flow, struct flow_wildcards *wc)
520 if (flow->nw_proto != IPPROTO_ICMP) {
521 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
522 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
524 wc->masks.tp_src = htons(0xff);
525 wc->masks.tp_dst = htons(0xff);
529 /* Initializes 'fmd' with the metadata found in 'flow'. */
531 flow_get_metadata(const struct flow *flow, struct flow_metadata *fmd)
533 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 24);
535 fmd->tun_id = flow->tunnel.tun_id;
536 fmd->tun_src = flow->tunnel.ip_src;
537 fmd->tun_dst = flow->tunnel.ip_dst;
538 fmd->metadata = flow->metadata;
539 memcpy(fmd->regs, flow->regs, sizeof fmd->regs);
540 fmd->pkt_mark = flow->pkt_mark;
541 fmd->in_port = flow->in_port.ofp_port;
545 flow_to_string(const struct flow *flow)
547 struct ds ds = DS_EMPTY_INITIALIZER;
548 flow_format(&ds, flow);
553 flow_tun_flag_to_string(uint32_t flags)
556 case FLOW_TNL_F_DONT_FRAGMENT:
558 case FLOW_TNL_F_CSUM:
568 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
569 uint32_t flags, char del)
577 uint32_t bit = rightmost_1bit(flags);
580 s = bit_to_string(bit);
582 ds_put_format(ds, "%s%c", s, del);
591 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
597 format_flags_masked(struct ds *ds, const char *name,
598 const char *(*bit_to_string)(uint32_t), uint32_t flags,
602 ds_put_format(ds, "%s=", name);
605 uint32_t bit = rightmost_1bit(mask);
606 const char *s = bit_to_string(bit);
608 ds_put_format(ds, "%s%s", (flags & bit) ? "+" : "-",
609 s ? s : "[Unknown]");
615 flow_format(struct ds *ds, const struct flow *flow)
619 match_wc_init(&match, flow);
620 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
624 flow_print(FILE *stream, const struct flow *flow)
626 char *s = flow_to_string(flow);
631 /* flow_wildcards functions. */
633 /* Initializes 'wc' as a set of wildcards that matches every packet. */
635 flow_wildcards_init_catchall(struct flow_wildcards *wc)
637 memset(&wc->masks, 0, sizeof wc->masks);
640 /* Clear the metadata and register wildcard masks. They are not packet
643 flow_wildcards_clear_non_packet_fields(struct flow_wildcards *wc)
645 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
646 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
649 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
652 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
654 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
657 for (i = 0; i < FLOW_U32S; i++) {
665 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
666 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
667 * in 'src1' or 'src2' or both. */
669 flow_wildcards_and(struct flow_wildcards *dst,
670 const struct flow_wildcards *src1,
671 const struct flow_wildcards *src2)
673 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
674 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
675 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
678 for (i = 0; i < FLOW_U32S; i++) {
679 dst_u32[i] = src1_u32[i] & src2_u32[i];
683 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
684 * is, a bit or a field is wildcarded in 'dst' if it is neither
685 * wildcarded in 'src1' nor 'src2'. */
687 flow_wildcards_or(struct flow_wildcards *dst,
688 const struct flow_wildcards *src1,
689 const struct flow_wildcards *src2)
691 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
692 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
693 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
696 for (i = 0; i < FLOW_U32S; i++) {
697 dst_u32[i] = src1_u32[i] | src2_u32[i];
701 /* Perform a bitwise OR of miniflow 'src' flow data with the equivalent
702 * fields in 'dst', storing the result in 'dst'. */
704 flow_union_with_miniflow(struct flow *dst, const struct miniflow *src)
706 uint32_t *dst_u32 = (uint32_t *) dst;
707 const uint32_t *p = src->values;
710 for (map = src->map; map; map = zero_rightmost_1bit(map)) {
711 dst_u32[raw_ctz(map)] |= *p++;
715 /* Fold minimask 'mask''s wildcard mask into 'wc's wildcard mask. */
717 flow_wildcards_fold_minimask(struct flow_wildcards *wc,
718 const struct minimask *mask)
720 flow_union_with_miniflow(&wc->masks, &mask->masks);
724 miniflow_get_map_in_range(const struct miniflow *miniflow,
725 uint8_t start, uint8_t end, unsigned int *offset)
727 uint64_t map = miniflow->map;
731 uint64_t msk = (UINT64_C(1) << start) - 1; /* 'start' LSBs set */
732 *offset = count_1bits(map & msk);
735 if (end < FLOW_U32S) {
736 uint64_t msk = (UINT64_C(1) << end) - 1; /* 'end' LSBs set */
742 /* Fold minimask 'mask''s wildcard mask into 'wc's wildcard mask
743 * in range [start, end). */
745 flow_wildcards_fold_minimask_range(struct flow_wildcards *wc,
746 const struct minimask *mask,
747 uint8_t start, uint8_t end)
749 uint32_t *dst_u32 = (uint32_t *)&wc->masks;
751 uint64_t map = miniflow_get_map_in_range(&mask->masks, start, end,
753 const uint32_t *p = mask->masks.values + offset;
755 for (; map; map = zero_rightmost_1bit(map)) {
756 dst_u32[raw_ctz(map)] |= *p++;
760 /* Returns a hash of the wildcards in 'wc'. */
762 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
764 return flow_hash(&wc->masks, basis);
767 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
770 flow_wildcards_equal(const struct flow_wildcards *a,
771 const struct flow_wildcards *b)
773 return flow_equal(&a->masks, &b->masks);
776 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
777 * 'b', false otherwise. */
779 flow_wildcards_has_extra(const struct flow_wildcards *a,
780 const struct flow_wildcards *b)
782 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
783 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
786 for (i = 0; i < FLOW_U32S; i++) {
787 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
794 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
795 * in 'wc' do not need to be equal in 'a' and 'b'. */
797 flow_equal_except(const struct flow *a, const struct flow *b,
798 const struct flow_wildcards *wc)
800 const uint32_t *a_u32 = (const uint32_t *) a;
801 const uint32_t *b_u32 = (const uint32_t *) b;
802 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
805 for (i = 0; i < FLOW_U32S; i++) {
806 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
813 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
814 * (A 0-bit indicates a wildcard bit.) */
816 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
818 wc->masks.regs[idx] = mask;
821 /* Hashes 'flow' based on its L2 through L4 protocol information. */
823 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
828 struct in6_addr ipv6_addr;
833 uint8_t eth_addr[ETH_ADDR_LEN];
839 memset(&fields, 0, sizeof fields);
840 for (i = 0; i < ETH_ADDR_LEN; i++) {
841 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
843 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
844 fields.eth_type = flow->dl_type;
846 /* UDP source and destination port are not taken into account because they
847 * will not necessarily be symmetric in a bidirectional flow. */
848 if (fields.eth_type == htons(ETH_TYPE_IP)) {
849 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
850 fields.ip_proto = flow->nw_proto;
851 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
852 fields.tp_port = flow->tp_src ^ flow->tp_dst;
854 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
855 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
856 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
857 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
859 for (i=0; i<16; i++) {
860 ipv6_addr[i] = a[i] ^ b[i];
862 fields.ip_proto = flow->nw_proto;
863 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
864 fields.tp_port = flow->tp_src ^ flow->tp_dst;
867 return jhash_bytes(&fields, sizeof fields, basis);
870 /* Initialize a flow with random fields that matter for nx_hash_fields. */
872 flow_random_hash_fields(struct flow *flow)
874 uint16_t rnd = random_uint16();
876 /* Initialize to all zeros. */
877 memset(flow, 0, sizeof *flow);
879 eth_addr_random(flow->dl_src);
880 eth_addr_random(flow->dl_dst);
882 flow->vlan_tci = (OVS_FORCE ovs_be16) (random_uint16() & VLAN_VID_MASK);
884 /* Make most of the random flows IPv4, some IPv6, and rest random. */
885 flow->dl_type = rnd < 0x8000 ? htons(ETH_TYPE_IP) :
886 rnd < 0xc000 ? htons(ETH_TYPE_IPV6) : (OVS_FORCE ovs_be16)rnd;
888 if (dl_type_is_ip_any(flow->dl_type)) {
889 if (flow->dl_type == htons(ETH_TYPE_IP)) {
890 flow->nw_src = (OVS_FORCE ovs_be32)random_uint32();
891 flow->nw_dst = (OVS_FORCE ovs_be32)random_uint32();
893 random_bytes(&flow->ipv6_src, sizeof flow->ipv6_src);
894 random_bytes(&flow->ipv6_dst, sizeof flow->ipv6_dst);
896 /* Make most of IP flows TCP, some UDP or SCTP, and rest random. */
897 rnd = random_uint16();
898 flow->nw_proto = rnd < 0x8000 ? IPPROTO_TCP :
899 rnd < 0xc000 ? IPPROTO_UDP :
900 rnd < 0xd000 ? IPPROTO_SCTP : (uint8_t)rnd;
901 if (flow->nw_proto == IPPROTO_TCP ||
902 flow->nw_proto == IPPROTO_UDP ||
903 flow->nw_proto == IPPROTO_SCTP) {
904 flow->tp_src = (OVS_FORCE ovs_be16)random_uint16();
905 flow->tp_dst = (OVS_FORCE ovs_be16)random_uint16();
910 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
912 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
913 enum nx_hash_fields fields)
916 case NX_HASH_FIELDS_ETH_SRC:
917 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
920 case NX_HASH_FIELDS_SYMMETRIC_L4:
921 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
922 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
923 if (flow->dl_type == htons(ETH_TYPE_IP)) {
924 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
925 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
926 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
927 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
928 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
930 if (is_ip_any(flow)) {
931 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
932 flow_unwildcard_tp_ports(flow, wc);
934 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
942 /* Hashes the portions of 'flow' designated by 'fields'. */
944 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
949 case NX_HASH_FIELDS_ETH_SRC:
950 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
952 case NX_HASH_FIELDS_SYMMETRIC_L4:
953 return flow_hash_symmetric_l4(flow, basis);
959 /* Returns a string representation of 'fields'. */
961 flow_hash_fields_to_str(enum nx_hash_fields fields)
964 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
965 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
966 default: return "<unknown>";
970 /* Returns true if the value of 'fields' is supported. Otherwise false. */
972 flow_hash_fields_valid(enum nx_hash_fields fields)
974 return fields == NX_HASH_FIELDS_ETH_SRC
975 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
978 /* Returns a hash value for the bits of 'flow' that are active based on
979 * 'wc', given 'basis'. */
981 flow_hash_in_wildcards(const struct flow *flow,
982 const struct flow_wildcards *wc, uint32_t basis)
984 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
985 const uint32_t *flow_u32 = (const uint32_t *) flow;
990 for (i = 0; i < FLOW_U32S; i++) {
991 hash = mhash_add(hash, flow_u32[i] & wc_u32[i]);
993 return mhash_finish(hash, 4 * FLOW_U32S);
996 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
997 * OpenFlow 1.0 "dl_vlan" value:
999 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
1000 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
1001 * 'flow' previously matched packets without a VLAN header).
1003 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
1004 * without a VLAN tag.
1006 * - Other values of 'vid' should not be used. */
1008 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
1010 if (vid == htons(OFP10_VLAN_NONE)) {
1011 flow->vlan_tci = htons(0);
1013 vid &= htons(VLAN_VID_MASK);
1014 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1015 flow->vlan_tci |= htons(VLAN_CFI) | vid;
1019 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1020 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
1023 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
1025 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
1026 flow->vlan_tci &= ~mask;
1027 flow->vlan_tci |= vid & mask;
1030 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
1033 * This function has no effect on the VLAN ID that 'flow' matches.
1035 * After calling this function, 'flow' will not match packets without a VLAN
1038 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
1041 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1042 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
1045 /* Returns the number of MPLS LSEs present in 'flow'
1047 * Returns 0 if the 'dl_type' of 'flow' is not an MPLS ethernet type.
1048 * Otherwise traverses 'flow''s MPLS label stack stopping at the
1049 * first entry that has the BoS bit set. If no such entry exists then
1050 * the maximum number of LSEs that can be stored in 'flow' is returned.
1053 flow_count_mpls_labels(const struct flow *flow, struct flow_wildcards *wc)
1056 wc->masks.dl_type = OVS_BE16_MAX;
1058 if (eth_type_mpls(flow->dl_type)) {
1060 int len = FLOW_MAX_MPLS_LABELS;
1062 for (i = 0; i < len; i++) {
1064 wc->masks.mpls_lse[i] |= htonl(MPLS_BOS_MASK);
1066 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1077 /* Returns the number consecutive of MPLS LSEs, starting at the
1078 * innermost LSE, that are common in 'a' and 'b'.
1080 * 'an' must be flow_count_mpls_labels(a).
1081 * 'bn' must be flow_count_mpls_labels(b).
1084 flow_count_common_mpls_labels(const struct flow *a, int an,
1085 const struct flow *b, int bn,
1086 struct flow_wildcards *wc)
1088 int min_n = MIN(an, bn);
1093 int a_last = an - 1;
1094 int b_last = bn - 1;
1097 for (i = 0; i < min_n; i++) {
1099 wc->masks.mpls_lse[a_last - i] = OVS_BE32_MAX;
1100 wc->masks.mpls_lse[b_last - i] = OVS_BE32_MAX;
1102 if (a->mpls_lse[a_last - i] != b->mpls_lse[b_last - i]) {
1113 /* Adds a new outermost MPLS label to 'flow' and changes 'flow''s Ethernet type
1114 * to 'mpls_eth_type', which must be an MPLS Ethertype.
1116 * If the new label is the first MPLS label in 'flow', it is generated as;
1118 * - label: 2, if 'flow' is IPv6, otherwise 0.
1120 * - TTL: IPv4 or IPv6 TTL, if present and nonzero, otherwise 64.
1122 * - TC: IPv4 or IPv6 TOS, if present, otherwise 0.
1126 * If the new label is the second or label MPLS label in 'flow', it is
1129 * - label: Copied from outer label.
1131 * - TTL: Copied from outer label.
1133 * - TC: Copied from outer label.
1137 * 'n' must be flow_count_mpls_labels(flow). 'n' must be less than
1138 * FLOW_MAX_MPLS_LABELS (because otherwise flow->mpls_lse[] would overflow).
1141 flow_push_mpls(struct flow *flow, int n, ovs_be16 mpls_eth_type,
1142 struct flow_wildcards *wc)
1144 ovs_assert(eth_type_mpls(mpls_eth_type));
1145 ovs_assert(n < FLOW_MAX_MPLS_LABELS);
1147 memset(wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1151 for (i = n; i >= 1; i--) {
1152 flow->mpls_lse[i] = flow->mpls_lse[i - 1];
1154 flow->mpls_lse[0] = (flow->mpls_lse[1]
1155 & htonl(~MPLS_BOS_MASK));
1157 int label = 0; /* IPv4 Explicit Null. */
1161 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1165 if (is_ip_any(flow)) {
1166 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1167 wc->masks.nw_tos |= IP_DSCP_MASK;
1172 wc->masks.nw_ttl = 0xff;
1175 flow->mpls_lse[0] = set_mpls_lse_values(ttl, tc, 1, htonl(label));
1177 /* Clear all L3 and L4 fields. */
1178 BUILD_ASSERT(FLOW_WC_SEQ == 24);
1179 memset((char *) flow + FLOW_SEGMENT_2_ENDS_AT, 0,
1180 sizeof(struct flow) - FLOW_SEGMENT_2_ENDS_AT);
1182 flow->dl_type = mpls_eth_type;
1185 /* Tries to remove the outermost MPLS label from 'flow'. Returns true if
1186 * successful, false otherwise. On success, sets 'flow''s Ethernet type to
1189 * 'n' must be flow_count_mpls_labels(flow). */
1191 flow_pop_mpls(struct flow *flow, int n, ovs_be16 eth_type,
1192 struct flow_wildcards *wc)
1197 /* Nothing to pop. */
1199 } else if (n == FLOW_MAX_MPLS_LABELS
1200 && !(flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK))) {
1201 /* Can't pop because we don't know what to fill in mpls_lse[n - 1]. */
1205 memset(wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1206 for (i = 1; i < n; i++) {
1207 flow->mpls_lse[i - 1] = flow->mpls_lse[i];
1209 flow->mpls_lse[n - 1] = 0;
1210 flow->dl_type = eth_type;
1214 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
1215 * as an OpenFlow 1.1 "mpls_label" value. */
1217 flow_set_mpls_label(struct flow *flow, int idx, ovs_be32 label)
1219 set_mpls_lse_label(&flow->mpls_lse[idx], label);
1222 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
1225 flow_set_mpls_ttl(struct flow *flow, int idx, uint8_t ttl)
1227 set_mpls_lse_ttl(&flow->mpls_lse[idx], ttl);
1230 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
1233 flow_set_mpls_tc(struct flow *flow, int idx, uint8_t tc)
1235 set_mpls_lse_tc(&flow->mpls_lse[idx], tc);
1238 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
1240 flow_set_mpls_bos(struct flow *flow, int idx, uint8_t bos)
1242 set_mpls_lse_bos(&flow->mpls_lse[idx], bos);
1245 /* Sets the entire MPLS LSE. */
1247 flow_set_mpls_lse(struct flow *flow, int idx, ovs_be32 lse)
1249 flow->mpls_lse[idx] = lse;
1253 flow_compose_l4(struct ofpbuf *b, const struct flow *flow)
1255 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
1256 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1257 if (flow->nw_proto == IPPROTO_TCP) {
1258 struct tcp_header *tcp;
1260 tcp = ofpbuf_put_zeros(b, sizeof *tcp);
1261 tcp->tcp_src = flow->tp_src;
1262 tcp->tcp_dst = flow->tp_dst;
1263 tcp->tcp_ctl = TCP_CTL(ntohs(flow->tcp_flags), 5);
1264 b->l7 = ofpbuf_tail(b);
1265 } else if (flow->nw_proto == IPPROTO_UDP) {
1266 struct udp_header *udp;
1268 udp = ofpbuf_put_zeros(b, sizeof *udp);
1269 udp->udp_src = flow->tp_src;
1270 udp->udp_dst = flow->tp_dst;
1271 b->l7 = ofpbuf_tail(b);
1272 } else if (flow->nw_proto == IPPROTO_SCTP) {
1273 struct sctp_header *sctp;
1275 sctp = ofpbuf_put_zeros(b, sizeof *sctp);
1276 sctp->sctp_src = flow->tp_src;
1277 sctp->sctp_dst = flow->tp_dst;
1278 b->l7 = ofpbuf_tail(b);
1279 } else if (flow->nw_proto == IPPROTO_ICMP) {
1280 struct icmp_header *icmp;
1282 icmp = ofpbuf_put_zeros(b, sizeof *icmp);
1283 icmp->icmp_type = ntohs(flow->tp_src);
1284 icmp->icmp_code = ntohs(flow->tp_dst);
1285 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
1286 b->l7 = ofpbuf_tail(b);
1287 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
1288 struct icmp6_hdr *icmp;
1290 icmp = ofpbuf_put_zeros(b, sizeof *icmp);
1291 icmp->icmp6_type = ntohs(flow->tp_src);
1292 icmp->icmp6_code = ntohs(flow->tp_dst);
1294 if (icmp->icmp6_code == 0 &&
1295 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
1296 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
1297 struct in6_addr *nd_target;
1298 struct nd_opt_hdr *nd_opt;
1300 nd_target = ofpbuf_put_zeros(b, sizeof *nd_target);
1301 *nd_target = flow->nd_target;
1303 if (!eth_addr_is_zero(flow->arp_sha)) {
1304 nd_opt = ofpbuf_put_zeros(b, 8);
1305 nd_opt->nd_opt_len = 1;
1306 nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
1307 memcpy(nd_opt + 1, flow->arp_sha, ETH_ADDR_LEN);
1309 if (!eth_addr_is_zero(flow->arp_tha)) {
1310 nd_opt = ofpbuf_put_zeros(b, 8);
1311 nd_opt->nd_opt_len = 1;
1312 nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR;
1313 memcpy(nd_opt + 1, flow->arp_tha, ETH_ADDR_LEN);
1316 icmp->icmp6_cksum = (OVS_FORCE uint16_t)
1317 csum(icmp, (char *)ofpbuf_tail(b) - (char *)icmp);
1318 b->l7 = ofpbuf_tail(b);
1323 /* Puts into 'b' a packet that flow_extract() would parse as having the given
1326 * (This is useful only for testing, obviously, and the packet isn't really
1327 * valid. It hasn't got some checksums filled in, for one, and lots of fields
1328 * are just zeroed.) */
1330 flow_compose(struct ofpbuf *b, const struct flow *flow)
1332 /* eth_compose() sets l3 pointer and makes sure it is 32-bit aligned. */
1333 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
1334 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
1335 struct eth_header *eth = b->l2;
1336 eth->eth_type = htons(b->size);
1340 if (flow->vlan_tci & htons(VLAN_CFI)) {
1341 eth_push_vlan(b, htons(ETH_TYPE_VLAN), flow->vlan_tci);
1344 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1345 struct ip_header *ip;
1347 ip = ofpbuf_put_zeros(b, sizeof *ip);
1348 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
1349 ip->ip_tos = flow->nw_tos;
1350 ip->ip_ttl = flow->nw_ttl;
1351 ip->ip_proto = flow->nw_proto;
1352 put_16aligned_be32(&ip->ip_src, flow->nw_src);
1353 put_16aligned_be32(&ip->ip_dst, flow->nw_dst);
1355 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
1356 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
1357 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
1358 ip->ip_frag_off |= htons(100);
1362 b->l4 = ofpbuf_tail(b);
1364 flow_compose_l4(b, flow);
1366 ip->ip_tot_len = htons((uint8_t *) b->data + b->size
1367 - (uint8_t *) b->l3);
1368 ip->ip_csum = csum(ip, sizeof *ip);
1369 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1370 struct ovs_16aligned_ip6_hdr *nh;
1372 nh = ofpbuf_put_zeros(b, sizeof *nh);
1373 put_16aligned_be32(&nh->ip6_flow, htonl(6 << 28) |
1374 htonl(flow->nw_tos << 20) | flow->ipv6_label);
1375 nh->ip6_hlim = flow->nw_ttl;
1376 nh->ip6_nxt = flow->nw_proto;
1378 memcpy(&nh->ip6_src, &flow->ipv6_src, sizeof(nh->ip6_src));
1379 memcpy(&nh->ip6_dst, &flow->ipv6_dst, sizeof(nh->ip6_dst));
1381 b->l4 = ofpbuf_tail(b);
1383 flow_compose_l4(b, flow);
1386 b->l7 ? htons((uint8_t *) b->l7 - (uint8_t *) b->l4) : htons(0);
1387 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1388 flow->dl_type == htons(ETH_TYPE_RARP)) {
1389 struct arp_eth_header *arp;
1391 b->l3 = arp = ofpbuf_put_zeros(b, sizeof *arp);
1392 arp->ar_hrd = htons(1);
1393 arp->ar_pro = htons(ETH_TYPE_IP);
1394 arp->ar_hln = ETH_ADDR_LEN;
1396 arp->ar_op = htons(flow->nw_proto);
1398 if (flow->nw_proto == ARP_OP_REQUEST ||
1399 flow->nw_proto == ARP_OP_REPLY) {
1400 put_16aligned_be32(&arp->ar_spa, flow->nw_src);
1401 put_16aligned_be32(&arp->ar_tpa, flow->nw_dst);
1402 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
1403 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
1407 if (eth_type_mpls(flow->dl_type)) {
1411 for (n = 1; n < FLOW_MAX_MPLS_LABELS; n++) {
1412 if (flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK)) {
1417 push_mpls(b, flow->dl_type, flow->mpls_lse[--n]);
1422 /* Compressed flow. */
1425 miniflow_n_values(const struct miniflow *flow)
1427 return count_1bits(flow->map);
1431 miniflow_alloc_values(struct miniflow *flow, int n)
1433 if (n <= MINI_N_INLINE) {
1434 return flow->inline_values;
1436 COVERAGE_INC(miniflow_malloc);
1437 return xmalloc(n * sizeof *flow->values);
1441 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
1442 * the caller. The caller must have already initialized 'dst->map' properly
1443 * to indicate the significant uint32_t elements of 'src'. 'n' must be the
1444 * number of 1-bits in 'dst->map'.
1446 * Normally the significant elements are the ones that are non-zero. However,
1447 * when a miniflow is initialized from a (mini)mask, the values can be zeroes,
1448 * so that the flow and mask always have the same maps.
1450 * This function initializes 'dst->values' (either inline if possible or with
1451 * malloc() otherwise) and copies the uint32_t elements of 'src' indicated by
1452 * 'dst->map' into it. */
1454 miniflow_init__(struct miniflow *dst, const struct flow *src, int n)
1456 const uint32_t *src_u32 = (const uint32_t *) src;
1460 dst->values = miniflow_alloc_values(dst, n);
1462 for (map = dst->map; map; map = zero_rightmost_1bit(map)) {
1463 dst->values[ofs++] = src_u32[raw_ctz(map)];
1467 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1468 * with miniflow_destroy(). */
1470 miniflow_init(struct miniflow *dst, const struct flow *src)
1472 const uint32_t *src_u32 = (const uint32_t *) src;
1476 /* Initialize dst->map, counting the number of nonzero elements. */
1480 for (i = 0; i < FLOW_U32S; i++) {
1482 dst->map |= UINT64_C(1) << i;
1487 miniflow_init__(dst, src, n);
1490 /* Initializes 'dst' as a copy of 'src', using 'mask->map' as 'dst''s map. The
1491 * caller must eventually free 'dst' with miniflow_destroy(). */
1493 miniflow_init_with_minimask(struct miniflow *dst, const struct flow *src,
1494 const struct minimask *mask)
1496 dst->map = mask->masks.map;
1497 miniflow_init__(dst, src, miniflow_n_values(dst));
1500 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1501 * with miniflow_destroy(). */
1503 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
1505 int n = miniflow_n_values(src);
1506 dst->map = src->map;
1507 dst->values = miniflow_alloc_values(dst, n);
1508 memcpy(dst->values, src->values, n * sizeof *dst->values);
1511 /* Initializes 'dst' with the data in 'src', destroying 'src'.
1512 * The caller must eventually free 'dst' with miniflow_destroy(). */
1514 miniflow_move(struct miniflow *dst, struct miniflow *src)
1516 if (src->values == src->inline_values) {
1517 dst->values = dst->inline_values;
1518 memcpy(dst->values, src->values,
1519 miniflow_n_values(src) * sizeof *dst->values);
1521 dst->values = src->values;
1523 dst->map = src->map;
1526 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
1527 * itself resides; the caller is responsible for that. */
1529 miniflow_destroy(struct miniflow *flow)
1531 if (flow->values != flow->inline_values) {
1536 /* Initializes 'dst' as a copy of 'src'. */
1538 miniflow_expand(const struct miniflow *src, struct flow *dst)
1540 memset(dst, 0, sizeof *dst);
1541 flow_union_with_miniflow(dst, src);
1544 static const uint32_t *
1545 miniflow_get__(const struct miniflow *flow, unsigned int u32_ofs)
1547 if (!(flow->map & (UINT64_C(1) << u32_ofs))) {
1548 static const uint32_t zero = 0;
1551 return flow->values +
1552 count_1bits(flow->map & ((UINT64_C(1) << u32_ofs) - 1));
1555 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1556 * were expanded into a "struct flow". */
1558 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1560 return *miniflow_get__(flow, u32_ofs);
1563 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
1564 * expanded into a "struct flow". */
1566 miniflow_get_be16(const struct miniflow *flow, unsigned int u8_ofs)
1568 const uint32_t *u32p = miniflow_get__(flow, u8_ofs / 4);
1569 const ovs_be16 *be16p = (const ovs_be16 *) u32p;
1570 return be16p[u8_ofs % 4 != 0];
1573 /* Returns the VID within the vlan_tci member of the "struct flow" represented
1576 miniflow_get_vid(const struct miniflow *flow)
1578 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
1579 return vlan_tci_to_vid(tci);
1582 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1584 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1586 const uint32_t *ap = a->values;
1587 const uint32_t *bp = b->values;
1588 const uint64_t a_map = a->map;
1589 const uint64_t b_map = b->map;
1592 if (a_map == b_map) {
1593 for (map = a_map; map; map = zero_rightmost_1bit(map)) {
1594 if (*ap++ != *bp++) {
1599 for (map = a_map | b_map; map; map = zero_rightmost_1bit(map)) {
1600 uint64_t bit = rightmost_1bit(map);
1601 uint64_t a_value = a_map & bit ? *ap++ : 0;
1602 uint64_t b_value = b_map & bit ? *bp++ : 0;
1604 if (a_value != b_value) {
1613 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1614 * in 'mask', false if they differ. */
1616 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1617 const struct minimask *mask)
1622 p = mask->masks.values;
1624 for (map = mask->masks.map; map; map = zero_rightmost_1bit(map)) {
1625 int ofs = raw_ctz(map);
1627 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p) {
1636 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1637 * in 'mask', false if they differ. */
1639 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1640 const struct minimask *mask)
1642 const uint32_t *b_u32 = (const uint32_t *) b;
1646 p = mask->masks.values;
1648 for (map = mask->masks.map; map; map = zero_rightmost_1bit(map)) {
1649 int ofs = raw_ctz(map);
1651 if ((miniflow_get(a, ofs) ^ b_u32[ofs]) & *p) {
1660 /* Returns a hash value for 'flow', given 'basis'. */
1662 miniflow_hash(const struct miniflow *flow, uint32_t basis)
1664 const uint32_t *p = flow->values;
1665 uint32_t hash = basis;
1666 uint64_t hash_map = 0;
1669 for (map = flow->map; map; map = zero_rightmost_1bit(map)) {
1671 hash = mhash_add(hash, *p);
1672 hash_map |= rightmost_1bit(map);
1676 hash = mhash_add(hash, hash_map);
1677 hash = mhash_add(hash, hash_map >> 32);
1679 return mhash_finish(hash, p - flow->values);
1682 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1683 * 'mask', given 'basis'.
1685 * The hash values returned by this function are the same as those returned by
1686 * flow_hash_in_minimask(), only the form of the arguments differ. */
1688 miniflow_hash_in_minimask(const struct miniflow *flow,
1689 const struct minimask *mask, uint32_t basis)
1691 const uint32_t *p = mask->masks.values;
1697 for (map = mask->masks.map; map; map = zero_rightmost_1bit(map)) {
1698 hash = mhash_add(hash, miniflow_get(flow, raw_ctz(map)) & *p++);
1701 return mhash_finish(hash, (p - mask->masks.values) * 4);
1704 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1705 * 'mask', given 'basis'.
1707 * The hash values returned by this function are the same as those returned by
1708 * miniflow_hash_in_minimask(), only the form of the arguments differ. */
1710 flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
1713 const uint32_t *flow_u32 = (const uint32_t *)flow;
1714 const uint32_t *p = mask->masks.values;
1719 for (map = mask->masks.map; map; map = zero_rightmost_1bit(map)) {
1720 hash = mhash_add(hash, flow_u32[raw_ctz(map)] & *p++);
1723 return mhash_finish(hash, (p - mask->masks.values) * 4);
1726 /* Returns a hash value for the bits of range [start, end) in 'flow',
1727 * where there are 1-bits in 'mask', given 'hash'.
1729 * The hash values returned by this function are the same as those returned by
1730 * minimatch_hash_range(), only the form of the arguments differ. */
1732 flow_hash_in_minimask_range(const struct flow *flow,
1733 const struct minimask *mask,
1734 uint8_t start, uint8_t end, uint32_t *basis)
1736 const uint32_t *flow_u32 = (const uint32_t *)flow;
1737 unsigned int offset;
1738 uint64_t map = miniflow_get_map_in_range(&mask->masks, start, end,
1740 const uint32_t *p = mask->masks.values + offset;
1741 uint32_t hash = *basis;
1743 for (; map; map = zero_rightmost_1bit(map)) {
1744 hash = mhash_add(hash, flow_u32[raw_ctz(map)] & *p++);
1747 *basis = hash; /* Allow continuation from the unfinished value. */
1748 return mhash_finish(hash, (p - mask->masks.values) * 4);
1752 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1753 * with minimask_destroy(). */
1755 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1757 miniflow_init(&mask->masks, &wc->masks);
1760 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1761 * with minimask_destroy(). */
1763 minimask_clone(struct minimask *dst, const struct minimask *src)
1765 miniflow_clone(&dst->masks, &src->masks);
1768 /* Initializes 'dst' with the data in 'src', destroying 'src'.
1769 * The caller must eventually free 'dst' with minimask_destroy(). */
1771 minimask_move(struct minimask *dst, struct minimask *src)
1773 miniflow_move(&dst->masks, &src->masks);
1776 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1778 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1779 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1781 minimask_combine(struct minimask *dst_,
1782 const struct minimask *a_, const struct minimask *b_,
1783 uint32_t storage[FLOW_U32S])
1785 struct miniflow *dst = &dst_->masks;
1786 const struct miniflow *a = &a_->masks;
1787 const struct miniflow *b = &b_->masks;
1791 dst->values = storage;
1794 for (map = a->map & b->map; map; map = zero_rightmost_1bit(map)) {
1795 int ofs = raw_ctz(map);
1796 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1799 dst->map |= rightmost_1bit(map);
1800 dst->values[n++] = mask;
1805 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1806 * itself resides; the caller is responsible for that. */
1808 minimask_destroy(struct minimask *mask)
1810 miniflow_destroy(&mask->masks);
1813 /* Initializes 'dst' as a copy of 'src'. */
1815 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1817 miniflow_expand(&mask->masks, &wc->masks);
1820 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1821 * were expanded into a "struct flow_wildcards". */
1823 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1825 return miniflow_get(&mask->masks, u32_ofs);
1828 /* Returns the VID mask within the vlan_tci member of the "struct
1829 * flow_wildcards" represented by 'mask'. */
1831 minimask_get_vid_mask(const struct minimask *mask)
1833 return miniflow_get_vid(&mask->masks);
1836 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1838 minimask_equal(const struct minimask *a, const struct minimask *b)
1840 return miniflow_equal(&a->masks, &b->masks);
1843 /* Returns a hash value for 'mask', given 'basis'. */
1845 minimask_hash(const struct minimask *mask, uint32_t basis)
1847 return miniflow_hash(&mask->masks, basis);
1850 /* Returns true if at least one bit is wildcarded in 'a_' but not in 'b_',
1851 * false otherwise. */
1853 minimask_has_extra(const struct minimask *a_, const struct minimask *b_)
1855 const struct miniflow *a = &a_->masks;
1856 const struct miniflow *b = &b_->masks;
1859 for (map = a->map | b->map; map; map = zero_rightmost_1bit(map)) {
1860 int ofs = raw_ctz(map);
1861 uint32_t a_u32 = miniflow_get(a, ofs);
1862 uint32_t b_u32 = miniflow_get(b, ofs);
1864 if ((a_u32 & b_u32) != b_u32) {
1872 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
1875 minimask_is_catchall(const struct minimask *mask_)
1877 const struct miniflow *mask = &mask_->masks;
1878 const uint32_t *p = mask->values;
1881 for (map = mask->map; map; map = zero_rightmost_1bit(map)) {