2 * Copyright (c) 2008, 2009, 2010, 2011, 2012 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 icmp_header *
85 pull_icmp(struct ofpbuf *packet)
87 return ofpbuf_try_pull(packet, ICMP_HEADER_LEN);
90 static struct icmp6_hdr *
91 pull_icmpv6(struct ofpbuf *packet)
93 return ofpbuf_try_pull(packet, sizeof(struct icmp6_hdr));
97 parse_vlan(struct ofpbuf *b, struct flow *flow)
100 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
104 if (b->size >= sizeof(struct qtag_prefix) + sizeof(ovs_be16)) {
105 struct qtag_prefix *qp = ofpbuf_pull(b, sizeof *qp);
106 flow->vlan_tci = qp->tci | htons(VLAN_CFI);
111 parse_ethertype(struct ofpbuf *b)
113 struct llc_snap_header *llc;
116 proto = *(ovs_be16 *) ofpbuf_pull(b, sizeof proto);
117 if (ntohs(proto) >= ETH_TYPE_MIN) {
121 if (b->size < sizeof *llc) {
122 return htons(FLOW_DL_TYPE_NONE);
126 if (llc->llc.llc_dsap != LLC_DSAP_SNAP
127 || llc->llc.llc_ssap != LLC_SSAP_SNAP
128 || llc->llc.llc_cntl != LLC_CNTL_SNAP
129 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
130 sizeof llc->snap.snap_org)) {
131 return htons(FLOW_DL_TYPE_NONE);
134 ofpbuf_pull(b, sizeof *llc);
135 return llc->snap.snap_type;
139 parse_ipv6(struct ofpbuf *packet, struct flow *flow)
141 const struct ip6_hdr *nh;
145 nh = ofpbuf_try_pull(packet, sizeof *nh);
150 nexthdr = nh->ip6_nxt;
152 flow->ipv6_src = nh->ip6_src;
153 flow->ipv6_dst = nh->ip6_dst;
155 tc_flow = get_unaligned_be32(&nh->ip6_flow);
156 flow->nw_tos = ntohl(tc_flow) >> 20;
157 flow->ipv6_label = tc_flow & htonl(IPV6_LABEL_MASK);
158 flow->nw_ttl = nh->ip6_hlim;
159 flow->nw_proto = IPPROTO_NONE;
162 if ((nexthdr != IPPROTO_HOPOPTS)
163 && (nexthdr != IPPROTO_ROUTING)
164 && (nexthdr != IPPROTO_DSTOPTS)
165 && (nexthdr != IPPROTO_AH)
166 && (nexthdr != IPPROTO_FRAGMENT)) {
167 /* It's either a terminal header (e.g., TCP, UDP) or one we
168 * don't understand. In either case, we're done with the
169 * packet, so use it to fill in 'nw_proto'. */
173 /* We only verify that at least 8 bytes of the next header are
174 * available, but many of these headers are longer. Ensure that
175 * accesses within the extension header are within those first 8
176 * bytes. All extension headers are required to be at least 8
178 if (packet->size < 8) {
182 if ((nexthdr == IPPROTO_HOPOPTS)
183 || (nexthdr == IPPROTO_ROUTING)
184 || (nexthdr == IPPROTO_DSTOPTS)) {
185 /* These headers, while different, have the fields we care about
186 * in the same location and with the same interpretation. */
187 const struct ip6_ext *ext_hdr = packet->data;
188 nexthdr = ext_hdr->ip6e_nxt;
189 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 1) * 8)) {
192 } else if (nexthdr == IPPROTO_AH) {
193 /* A standard AH definition isn't available, but the fields
194 * we care about are in the same location as the generic
195 * option header--only the header length is calculated
197 const struct ip6_ext *ext_hdr = packet->data;
198 nexthdr = ext_hdr->ip6e_nxt;
199 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 2) * 4)) {
202 } else if (nexthdr == IPPROTO_FRAGMENT) {
203 const struct ip6_frag *frag_hdr = packet->data;
205 nexthdr = frag_hdr->ip6f_nxt;
206 if (!ofpbuf_try_pull(packet, sizeof *frag_hdr)) {
210 /* We only process the first fragment. */
211 if (frag_hdr->ip6f_offlg != htons(0)) {
212 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) == htons(0)) {
213 flow->nw_frag = FLOW_NW_FRAG_ANY;
215 flow->nw_frag |= FLOW_NW_FRAG_LATER;
216 nexthdr = IPPROTO_FRAGMENT;
223 flow->nw_proto = nexthdr;
228 parse_tcp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
230 const struct tcp_header *tcp = pull_tcp(b);
232 flow->tp_src = tcp->tcp_src;
233 flow->tp_dst = tcp->tcp_dst;
234 packet->l7 = b->data;
239 parse_udp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
241 const struct udp_header *udp = pull_udp(b);
243 flow->tp_src = udp->udp_src;
244 flow->tp_dst = udp->udp_dst;
245 packet->l7 = b->data;
250 parse_icmpv6(struct ofpbuf *b, struct flow *flow)
252 const struct icmp6_hdr *icmp = pull_icmpv6(b);
258 /* The ICMPv6 type and code fields use the 16-bit transport port
259 * fields, so we need to store them in 16-bit network byte order. */
260 flow->tp_src = htons(icmp->icmp6_type);
261 flow->tp_dst = htons(icmp->icmp6_code);
263 if (icmp->icmp6_code == 0 &&
264 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
265 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
266 const struct in6_addr *nd_target;
268 nd_target = ofpbuf_try_pull(b, sizeof *nd_target);
272 flow->nd_target = *nd_target;
274 while (b->size >= 8) {
275 /* The minimum size of an option is 8 bytes, which also is
276 * the size of Ethernet link-layer options. */
277 const struct nd_opt_hdr *nd_opt = b->data;
278 int opt_len = nd_opt->nd_opt_len * 8;
280 if (!opt_len || opt_len > b->size) {
284 /* Store the link layer address if the appropriate option is
285 * provided. It is considered an error if the same link
286 * layer option is specified twice. */
287 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
289 if (eth_addr_is_zero(flow->arp_sha)) {
290 memcpy(flow->arp_sha, nd_opt + 1, ETH_ADDR_LEN);
294 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
296 if (eth_addr_is_zero(flow->arp_tha)) {
297 memcpy(flow->arp_tha, nd_opt + 1, ETH_ADDR_LEN);
303 if (!ofpbuf_try_pull(b, opt_len)) {
312 memset(&flow->nd_target, 0, sizeof(flow->nd_target));
313 memset(flow->arp_sha, 0, sizeof(flow->arp_sha));
314 memset(flow->arp_tha, 0, sizeof(flow->arp_tha));
320 /* Initializes 'flow' members from 'packet', 'skb_priority', 'tnl', and
323 * Initializes 'packet' header pointers as follows:
325 * - packet->l2 to the start of the Ethernet header.
327 * - packet->l3 to just past the Ethernet header, or just past the
328 * vlan_header if one is present, to the first byte of the payload of the
331 * - packet->l4 to just past the IPv4 header, if one is present and has a
332 * correct length, and otherwise NULL.
334 * - packet->l7 to just past the TCP or UDP or ICMP header, if one is
335 * present and has a correct length, and otherwise NULL.
338 flow_extract(struct ofpbuf *packet, uint32_t skb_priority, uint32_t skb_mark,
339 const struct flow_tnl *tnl, uint16_t ofp_in_port,
342 struct ofpbuf b = *packet;
343 struct eth_header *eth;
345 COVERAGE_INC(flow_extract);
347 memset(flow, 0, sizeof *flow);
350 ovs_assert(tnl != &flow->tunnel);
353 flow->in_port = ofp_in_port;
354 flow->skb_priority = skb_priority;
355 flow->skb_mark = skb_mark;
362 if (b.size < sizeof *eth) {
368 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
369 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
371 /* dl_type, vlan_tci. */
372 ofpbuf_pull(&b, ETH_ADDR_LEN * 2);
373 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
374 parse_vlan(&b, flow);
376 flow->dl_type = parse_ethertype(&b);
379 flow_extract_l3_onwards(packet, flow, flow->dl_type);
382 /* Initializes l3 and higher 'flow' members from 'packet'
384 * This should be called by or after flow_extract()
386 * Initializes 'packet' header pointers as follows:
388 * - packet->l4 to just past the IPv4 header, if one is present and has a
389 * correct length, and otherwise NULL.
391 * - packet->l7 to just past the TCP or UDP or ICMP header, if one is
392 * present and has a correct length, and otherwise NULL.
395 flow_extract_l3_onwards(struct ofpbuf *packet, struct flow *flow,
400 ofpbuf_use_const(&b, packet->l3, packet->size -
401 (size_t)((char *)packet->l3 - (char *)packet->l2));
404 if (dl_type == htons(ETH_TYPE_IP)) {
405 const struct ip_header *nh = pull_ip(&b);
409 flow->nw_src = get_unaligned_be32(&nh->ip_src);
410 flow->nw_dst = get_unaligned_be32(&nh->ip_dst);
411 flow->nw_proto = nh->ip_proto;
413 flow->nw_tos = nh->ip_tos;
414 if (IP_IS_FRAGMENT(nh->ip_frag_off)) {
415 flow->nw_frag = FLOW_NW_FRAG_ANY;
416 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
417 flow->nw_frag |= FLOW_NW_FRAG_LATER;
420 flow->nw_ttl = nh->ip_ttl;
422 if (!(nh->ip_frag_off & htons(IP_FRAG_OFF_MASK))) {
423 if (flow->nw_proto == IPPROTO_TCP) {
424 parse_tcp(packet, &b, flow);
425 } else if (flow->nw_proto == IPPROTO_UDP) {
426 parse_udp(packet, &b, flow);
427 } else if (flow->nw_proto == IPPROTO_ICMP) {
428 const struct icmp_header *icmp = pull_icmp(&b);
430 flow->tp_src = htons(icmp->icmp_type);
431 flow->tp_dst = htons(icmp->icmp_code);
437 } else if (dl_type == htons(ETH_TYPE_IPV6)) {
438 if (parse_ipv6(&b, flow)) {
443 if (flow->nw_proto == IPPROTO_TCP) {
444 parse_tcp(packet, &b, flow);
445 } else if (flow->nw_proto == IPPROTO_UDP) {
446 parse_udp(packet, &b, flow);
447 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
448 if (parse_icmpv6(&b, flow)) {
452 } else if (dl_type == htons(ETH_TYPE_ARP) ||
453 dl_type == htons(ETH_TYPE_RARP)) {
454 const struct arp_eth_header *arp = pull_arp(&b);
455 if (arp && arp->ar_hrd == htons(1)
456 && arp->ar_pro == htons(ETH_TYPE_IP)
457 && arp->ar_hln == ETH_ADDR_LEN
458 && arp->ar_pln == 4) {
459 /* We only match on the lower 8 bits of the opcode. */
460 if (ntohs(arp->ar_op) <= 0xff) {
461 flow->nw_proto = ntohs(arp->ar_op);
464 flow->nw_src = arp->ar_spa;
465 flow->nw_dst = arp->ar_tpa;
466 memcpy(flow->arp_sha, arp->ar_sha, ETH_ADDR_LEN);
467 memcpy(flow->arp_tha, arp->ar_tha, ETH_ADDR_LEN);
472 /* For every bit of a field that is wildcarded in 'wildcards', sets the
473 * corresponding bit in 'flow' to zero. */
475 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
477 uint32_t *flow_u32 = (uint32_t *) flow;
478 const uint32_t *wc_u32 = (const uint32_t *) &wildcards->masks;
481 for (i = 0; i < FLOW_U32S; i++) {
482 flow_u32[i] &= wc_u32[i];
486 /* Initializes 'fmd' with the metadata found in 'flow'. */
488 flow_get_metadata(const struct flow *flow, struct flow_metadata *fmd)
490 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 18);
492 fmd->tun_id = flow->tunnel.tun_id;
493 fmd->metadata = flow->metadata;
494 memcpy(fmd->regs, flow->regs, sizeof fmd->regs);
495 fmd->in_port = flow->in_port;
499 flow_to_string(const struct flow *flow)
501 struct ds ds = DS_EMPTY_INITIALIZER;
502 flow_format(&ds, flow);
507 flow_tun_flag_to_string(uint32_t flags)
510 case FLOW_TNL_F_DONT_FRAGMENT:
512 case FLOW_TNL_F_CSUM:
522 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
523 uint32_t flags, char del)
531 uint32_t bit = rightmost_1bit(flags);
534 s = bit_to_string(bit);
536 ds_put_format(ds, "%s%c", s, del);
545 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
551 flow_format(struct ds *ds, const struct flow *flow)
555 match_wc_init(&match, flow);
556 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
560 flow_print(FILE *stream, const struct flow *flow)
562 char *s = flow_to_string(flow);
567 /* flow_wildcards functions. */
569 /* Initializes 'wc' as a set of wildcards that matches every packet. */
571 flow_wildcards_init_catchall(struct flow_wildcards *wc)
573 memset(&wc->masks, 0, sizeof wc->masks);
576 /* Initializes 'wc' as an exact-match set of wildcards; that is, 'wc' does not
577 * wildcard any bits or fields. */
579 flow_wildcards_init_exact(struct flow_wildcards *wc)
581 memset(&wc->masks, 0xff, sizeof wc->masks);
582 memset(wc->masks.zeros, 0, sizeof wc->masks.zeros);
585 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
588 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
590 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
593 for (i = 0; i < FLOW_U32S; i++) {
601 /* Initializes 'dst' as the combination of wildcards in 'src1' and 'src2'.
602 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded in
603 * 'src1' or 'src2' or both. */
605 flow_wildcards_combine(struct flow_wildcards *dst,
606 const struct flow_wildcards *src1,
607 const struct flow_wildcards *src2)
609 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
610 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
611 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
614 for (i = 0; i < FLOW_U32S; i++) {
615 dst_u32[i] = src1_u32[i] & src2_u32[i];
619 /* Returns a hash of the wildcards in 'wc'. */
621 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
623 return flow_hash(&wc->masks, basis);
626 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
629 flow_wildcards_equal(const struct flow_wildcards *a,
630 const struct flow_wildcards *b)
632 return flow_equal(&a->masks, &b->masks);
635 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
636 * 'b', false otherwise. */
638 flow_wildcards_has_extra(const struct flow_wildcards *a,
639 const struct flow_wildcards *b)
641 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
642 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
645 for (i = 0; i < FLOW_U32S; i++) {
646 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
653 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
654 * in 'wc' do not need to be equal in 'a' and 'b'. */
656 flow_equal_except(const struct flow *a, const struct flow *b,
657 const struct flow_wildcards *wc)
659 const uint32_t *a_u32 = (const uint32_t *) a;
660 const uint32_t *b_u32 = (const uint32_t *) b;
661 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
664 for (i = 0; i < FLOW_U32S; i++) {
665 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
672 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
673 * (A 0-bit indicates a wildcard bit.) */
675 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
677 wc->masks.regs[idx] = mask;
680 /* Hashes 'flow' based on its L2 through L4 protocol information. */
682 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
687 struct in6_addr ipv6_addr;
692 uint8_t eth_addr[ETH_ADDR_LEN];
698 memset(&fields, 0, sizeof fields);
699 for (i = 0; i < ETH_ADDR_LEN; i++) {
700 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
702 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
703 fields.eth_type = flow->dl_type;
705 /* UDP source and destination port are not taken into account because they
706 * will not necessarily be symmetric in a bidirectional flow. */
707 if (fields.eth_type == htons(ETH_TYPE_IP)) {
708 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
709 fields.ip_proto = flow->nw_proto;
710 if (fields.ip_proto == IPPROTO_TCP) {
711 fields.tp_port = flow->tp_src ^ flow->tp_dst;
713 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
714 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
715 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
716 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
718 for (i=0; i<16; i++) {
719 ipv6_addr[i] = a[i] ^ b[i];
721 fields.ip_proto = flow->nw_proto;
722 if (fields.ip_proto == IPPROTO_TCP) {
723 fields.tp_port = flow->tp_src ^ flow->tp_dst;
726 return jhash_bytes(&fields, sizeof fields, basis);
729 /* Hashes the portions of 'flow' designated by 'fields'. */
731 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
736 case NX_HASH_FIELDS_ETH_SRC:
737 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
739 case NX_HASH_FIELDS_SYMMETRIC_L4:
740 return flow_hash_symmetric_l4(flow, basis);
746 /* Returns a string representation of 'fields'. */
748 flow_hash_fields_to_str(enum nx_hash_fields fields)
751 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
752 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
753 default: return "<unknown>";
757 /* Returns true if the value of 'fields' is supported. Otherwise false. */
759 flow_hash_fields_valid(enum nx_hash_fields fields)
761 return fields == NX_HASH_FIELDS_ETH_SRC
762 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
765 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
766 * OpenFlow 1.0 "dl_vlan" value:
768 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
769 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
770 * 'flow' previously matched packets without a VLAN header).
772 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
773 * without a VLAN tag.
775 * - Other values of 'vid' should not be used. */
777 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
779 if (vid == htons(OFP10_VLAN_NONE)) {
780 flow->vlan_tci = htons(0);
782 vid &= htons(VLAN_VID_MASK);
783 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
784 flow->vlan_tci |= htons(VLAN_CFI) | vid;
788 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
789 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
792 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
794 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
795 flow->vlan_tci &= ~mask;
796 flow->vlan_tci |= vid & mask;
799 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
802 * This function has no effect on the VLAN ID that 'flow' matches.
804 * After calling this function, 'flow' will not match packets without a VLAN
807 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
810 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
811 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
814 /* Puts into 'b' a packet that flow_extract() would parse as having the given
817 * (This is useful only for testing, obviously, and the packet isn't really
818 * valid. It hasn't got some checksums filled in, for one, and lots of fields
819 * are just zeroed.) */
821 flow_compose(struct ofpbuf *b, const struct flow *flow)
823 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
824 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
825 struct eth_header *eth = b->l2;
826 eth->eth_type = htons(b->size);
830 if (flow->vlan_tci & htons(VLAN_CFI)) {
831 eth_push_vlan(b, flow->vlan_tci);
834 if (flow->dl_type == htons(ETH_TYPE_IP)) {
835 struct ip_header *ip;
837 b->l3 = ip = ofpbuf_put_zeros(b, sizeof *ip);
838 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
839 ip->ip_tos = flow->nw_tos;
840 ip->ip_ttl = flow->nw_ttl;
841 ip->ip_proto = flow->nw_proto;
842 ip->ip_src = flow->nw_src;
843 ip->ip_dst = flow->nw_dst;
845 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
846 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
847 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
848 ip->ip_frag_off |= htons(100);
851 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
852 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
853 if (flow->nw_proto == IPPROTO_TCP) {
854 struct tcp_header *tcp;
856 b->l4 = tcp = ofpbuf_put_zeros(b, sizeof *tcp);
857 tcp->tcp_src = flow->tp_src;
858 tcp->tcp_dst = flow->tp_dst;
859 tcp->tcp_ctl = TCP_CTL(0, 5);
860 } else if (flow->nw_proto == IPPROTO_UDP) {
861 struct udp_header *udp;
863 b->l4 = udp = ofpbuf_put_zeros(b, sizeof *udp);
864 udp->udp_src = flow->tp_src;
865 udp->udp_dst = flow->tp_dst;
866 } else if (flow->nw_proto == IPPROTO_ICMP) {
867 struct icmp_header *icmp;
869 b->l4 = icmp = ofpbuf_put_zeros(b, sizeof *icmp);
870 icmp->icmp_type = ntohs(flow->tp_src);
871 icmp->icmp_code = ntohs(flow->tp_dst);
872 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
877 ip->ip_tot_len = htons((uint8_t *) b->data + b->size
878 - (uint8_t *) b->l3);
879 ip->ip_csum = csum(ip, sizeof *ip);
880 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
882 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
883 flow->dl_type == htons(ETH_TYPE_RARP)) {
884 struct arp_eth_header *arp;
886 b->l3 = arp = ofpbuf_put_zeros(b, sizeof *arp);
887 arp->ar_hrd = htons(1);
888 arp->ar_pro = htons(ETH_TYPE_IP);
889 arp->ar_hln = ETH_ADDR_LEN;
891 arp->ar_op = htons(flow->nw_proto);
893 if (flow->nw_proto == ARP_OP_REQUEST ||
894 flow->nw_proto == ARP_OP_REPLY) {
895 arp->ar_spa = flow->nw_src;
896 arp->ar_tpa = flow->nw_dst;
897 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
898 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
903 /* Compressed flow. */
906 miniflow_n_values(const struct miniflow *flow)
911 for (i = 0; i < MINI_N_MAPS; i++) {
912 n += popcount(flow->map[i]);
918 miniflow_alloc_values(struct miniflow *flow, int n)
920 if (n <= MINI_N_INLINE) {
921 return flow->inline_values;
923 COVERAGE_INC(miniflow_malloc);
924 return xmalloc(n * sizeof *flow->values);
928 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
929 * with miniflow_destroy(). */
931 miniflow_init(struct miniflow *dst, const struct flow *src)
933 const uint32_t *src_u32 = (const uint32_t *) src;
938 /* Initialize dst->map, counting the number of nonzero elements. */
940 memset(dst->map, 0, sizeof dst->map);
941 for (i = 0; i < FLOW_U32S; i++) {
943 dst->map[i / 32] |= 1u << (i % 32);
948 /* Initialize dst->values. */
949 dst->values = miniflow_alloc_values(dst, n);
951 for (i = 0; i < MINI_N_MAPS; i++) {
954 for (map = dst->map[i]; map; map = zero_rightmost_1bit(map)) {
955 dst->values[ofs++] = src_u32[raw_ctz(map) + i * 32];
960 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
961 * with miniflow_destroy(). */
963 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
965 int n = miniflow_n_values(src);
966 memcpy(dst->map, src->map, sizeof dst->map);
967 dst->values = miniflow_alloc_values(dst, n);
968 memcpy(dst->values, src->values, n * sizeof *dst->values);
971 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
972 * itself resides; the caller is responsible for that. */
974 miniflow_destroy(struct miniflow *flow)
976 if (flow->values != flow->inline_values) {
981 /* Initializes 'dst' as a copy of 'src'. */
983 miniflow_expand(const struct miniflow *src, struct flow *dst)
985 uint32_t *dst_u32 = (uint32_t *) dst;
989 memset(dst_u32, 0, sizeof *dst);
992 for (i = 0; i < MINI_N_MAPS; i++) {
995 for (map = src->map[i]; map; map = zero_rightmost_1bit(map)) {
996 dst_u32[raw_ctz(map) + i * 32] = src->values[ofs++];
1001 static const uint32_t *
1002 miniflow_get__(const struct miniflow *flow, unsigned int u32_ofs)
1004 if (!(flow->map[u32_ofs / 32] & (1u << (u32_ofs % 32)))) {
1005 static const uint32_t zero = 0;
1008 const uint32_t *p = flow->values;
1010 BUILD_ASSERT(MINI_N_MAPS == 2);
1012 p += popcount(flow->map[0] & ((1u << u32_ofs) - 1));
1014 p += popcount(flow->map[0]);
1015 p += popcount(flow->map[1] & ((1u << (u32_ofs - 32)) - 1));
1021 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1022 * were expanded into a "struct flow". */
1024 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1026 return *miniflow_get__(flow, u32_ofs);
1029 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
1030 * expanded into a "struct flow". */
1032 miniflow_get_be16(const struct miniflow *flow, unsigned int u8_ofs)
1034 const uint32_t *u32p = miniflow_get__(flow, u8_ofs / 4);
1035 const ovs_be16 *be16p = (const ovs_be16 *) u32p;
1036 return be16p[u8_ofs % 4 != 0];
1039 /* Returns the VID within the vlan_tci member of the "struct flow" represented
1042 miniflow_get_vid(const struct miniflow *flow)
1044 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
1045 return vlan_tci_to_vid(tci);
1048 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1050 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1054 for (i = 0; i < MINI_N_MAPS; i++) {
1055 if (a->map[i] != b->map[i]) {
1060 return !memcmp(a->values, b->values,
1061 miniflow_n_values(a) * sizeof *a->values);
1064 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1065 * in 'mask', false if they differ. */
1067 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1068 const struct minimask *mask)
1073 p = mask->masks.values;
1074 for (i = 0; i < MINI_N_MAPS; i++) {
1077 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1078 int ofs = raw_ctz(map) + i * 32;
1080 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p) {
1090 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1091 * in 'mask', false if they differ. */
1093 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1094 const struct minimask *mask)
1096 const uint32_t *b_u32 = (const uint32_t *) b;
1100 p = mask->masks.values;
1101 for (i = 0; i < MINI_N_MAPS; i++) {
1104 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1105 int ofs = raw_ctz(map) + i * 32;
1107 if ((miniflow_get(a, ofs) ^ b_u32[ofs]) & *p) {
1117 /* Returns a hash value for 'flow', given 'basis'. */
1119 miniflow_hash(const struct miniflow *flow, uint32_t basis)
1121 BUILD_ASSERT_DECL(MINI_N_MAPS == 2);
1122 return hash_3words(flow->map[0], flow->map[1],
1123 hash_words(flow->values, miniflow_n_values(flow),
1127 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1128 * 'mask', given 'basis'.
1130 * The hash values returned by this function are the same as those returned by
1131 * flow_hash_in_minimask(), only the form of the arguments differ. */
1133 miniflow_hash_in_minimask(const struct miniflow *flow,
1134 const struct minimask *mask, uint32_t basis)
1136 const uint32_t *p = mask->masks.values;
1141 for (i = 0; i < MINI_N_MAPS; i++) {
1144 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1145 int ofs = raw_ctz(map) + i * 32;
1147 hash = mhash_add(hash, miniflow_get(flow, ofs) & *p);
1152 return mhash_finish(hash, (p - mask->masks.values) * 4);
1155 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1156 * 'mask', given 'basis'.
1158 * The hash values returned by this function are the same as those returned by
1159 * miniflow_hash_in_minimask(), only the form of the arguments differ. */
1161 flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
1164 const uint32_t *flow_u32 = (const uint32_t *) flow;
1165 const uint32_t *p = mask->masks.values;
1170 for (i = 0; i < MINI_N_MAPS; i++) {
1173 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1174 int ofs = raw_ctz(map) + i * 32;
1176 hash = mhash_add(hash, flow_u32[ofs] & *p);
1181 return mhash_finish(hash, (p - mask->masks.values) * 4);
1184 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1185 * with minimask_destroy(). */
1187 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1189 miniflow_init(&mask->masks, &wc->masks);
1192 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1193 * with minimask_destroy(). */
1195 minimask_clone(struct minimask *dst, const struct minimask *src)
1197 miniflow_clone(&dst->masks, &src->masks);
1200 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1202 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1203 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1205 minimask_combine(struct minimask *dst_,
1206 const struct minimask *a_, const struct minimask *b_,
1207 uint32_t storage[FLOW_U32S])
1209 struct miniflow *dst = &dst_->masks;
1210 const struct miniflow *a = &a_->masks;
1211 const struct miniflow *b = &b_->masks;
1215 dst->values = storage;
1216 for (i = 0; i < MINI_N_MAPS; i++) {
1220 for (map = a->map[i] & b->map[i]; map;
1221 map = zero_rightmost_1bit(map)) {
1222 int ofs = raw_ctz(map) + i * 32;
1223 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1226 dst->map[i] |= rightmost_1bit(map);
1227 dst->values[n++] = mask;
1233 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1234 * itself resides; the caller is responsible for that. */
1236 minimask_destroy(struct minimask *mask)
1238 miniflow_destroy(&mask->masks);
1241 /* Initializes 'dst' as a copy of 'src'. */
1243 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1245 miniflow_expand(&mask->masks, &wc->masks);
1248 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1249 * were expanded into a "struct flow_wildcards". */
1251 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1253 return miniflow_get(&mask->masks, u32_ofs);
1256 /* Returns the VID mask within the vlan_tci member of the "struct
1257 * flow_wildcards" represented by 'mask'. */
1259 minimask_get_vid_mask(const struct minimask *mask)
1261 return miniflow_get_vid(&mask->masks);
1264 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1266 minimask_equal(const struct minimask *a, const struct minimask *b)
1268 return miniflow_equal(&a->masks, &b->masks);
1271 /* Returns a hash value for 'mask', given 'basis'. */
1273 minimask_hash(const struct minimask *mask, uint32_t basis)
1275 return miniflow_hash(&mask->masks, basis);
1278 /* Returns true if at least one bit is wildcarded in 'a_' but not in 'b_',
1279 * false otherwise. */
1281 minimask_has_extra(const struct minimask *a_, const struct minimask *b_)
1283 const struct miniflow *a = &a_->masks;
1284 const struct miniflow *b = &b_->masks;
1287 for (i = 0; i < MINI_N_MAPS; i++) {
1290 for (map = a->map[i] | b->map[i]; map;
1291 map = zero_rightmost_1bit(map)) {
1292 int ofs = raw_ctz(map) + i * 32;
1293 uint32_t a_u32 = miniflow_get(a, ofs);
1294 uint32_t b_u32 = miniflow_get(b, ofs);
1296 if ((a_u32 & b_u32) != b_u32) {
1305 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
1308 minimask_is_catchall(const struct minimask *mask_)
1310 const struct miniflow *mask = &mask_->masks;
1312 BUILD_ASSERT(MINI_N_MAPS == 2);
1313 return !(mask->map[0] | mask->map[1]);