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_mpls(struct ofpbuf *b, struct flow *flow)
101 while ((mh = ofpbuf_try_pull(b, sizeof *mh))) {
102 if (flow->mpls_depth++ == 0) {
103 flow->mpls_lse = mh->mpls_lse;
105 if (mh->mpls_lse & htonl(MPLS_BOS_MASK)) {
112 parse_vlan(struct ofpbuf *b, struct flow *flow)
115 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
119 if (b->size >= sizeof(struct qtag_prefix) + sizeof(ovs_be16)) {
120 struct qtag_prefix *qp = ofpbuf_pull(b, sizeof *qp);
121 flow->vlan_tci = qp->tci | htons(VLAN_CFI);
126 parse_ethertype(struct ofpbuf *b)
128 struct llc_snap_header *llc;
131 proto = *(ovs_be16 *) ofpbuf_pull(b, sizeof proto);
132 if (ntohs(proto) >= ETH_TYPE_MIN) {
136 if (b->size < sizeof *llc) {
137 return htons(FLOW_DL_TYPE_NONE);
141 if (llc->llc.llc_dsap != LLC_DSAP_SNAP
142 || llc->llc.llc_ssap != LLC_SSAP_SNAP
143 || llc->llc.llc_cntl != LLC_CNTL_SNAP
144 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
145 sizeof llc->snap.snap_org)) {
146 return htons(FLOW_DL_TYPE_NONE);
149 ofpbuf_pull(b, sizeof *llc);
151 if (ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN) {
152 return llc->snap.snap_type;
155 return htons(FLOW_DL_TYPE_NONE);
159 parse_ipv6(struct ofpbuf *packet, struct flow *flow)
161 const struct ip6_hdr *nh;
165 nh = ofpbuf_try_pull(packet, sizeof *nh);
170 nexthdr = nh->ip6_nxt;
172 flow->ipv6_src = nh->ip6_src;
173 flow->ipv6_dst = nh->ip6_dst;
175 tc_flow = get_unaligned_be32(&nh->ip6_flow);
176 flow->nw_tos = ntohl(tc_flow) >> 20;
177 flow->ipv6_label = tc_flow & htonl(IPV6_LABEL_MASK);
178 flow->nw_ttl = nh->ip6_hlim;
179 flow->nw_proto = IPPROTO_NONE;
182 if ((nexthdr != IPPROTO_HOPOPTS)
183 && (nexthdr != IPPROTO_ROUTING)
184 && (nexthdr != IPPROTO_DSTOPTS)
185 && (nexthdr != IPPROTO_AH)
186 && (nexthdr != IPPROTO_FRAGMENT)) {
187 /* It's either a terminal header (e.g., TCP, UDP) or one we
188 * don't understand. In either case, we're done with the
189 * packet, so use it to fill in 'nw_proto'. */
193 /* We only verify that at least 8 bytes of the next header are
194 * available, but many of these headers are longer. Ensure that
195 * accesses within the extension header are within those first 8
196 * bytes. All extension headers are required to be at least 8
198 if (packet->size < 8) {
202 if ((nexthdr == IPPROTO_HOPOPTS)
203 || (nexthdr == IPPROTO_ROUTING)
204 || (nexthdr == IPPROTO_DSTOPTS)) {
205 /* These headers, while different, have the fields we care about
206 * in the same location and with the same interpretation. */
207 const struct ip6_ext *ext_hdr = packet->data;
208 nexthdr = ext_hdr->ip6e_nxt;
209 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 1) * 8)) {
212 } else if (nexthdr == IPPROTO_AH) {
213 /* A standard AH definition isn't available, but the fields
214 * we care about are in the same location as the generic
215 * option header--only the header length is calculated
217 const struct ip6_ext *ext_hdr = packet->data;
218 nexthdr = ext_hdr->ip6e_nxt;
219 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 2) * 4)) {
222 } else if (nexthdr == IPPROTO_FRAGMENT) {
223 const struct ip6_frag *frag_hdr = packet->data;
225 nexthdr = frag_hdr->ip6f_nxt;
226 if (!ofpbuf_try_pull(packet, sizeof *frag_hdr)) {
230 /* We only process the first fragment. */
231 if (frag_hdr->ip6f_offlg != htons(0)) {
232 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) == htons(0)) {
233 flow->nw_frag = FLOW_NW_FRAG_ANY;
235 flow->nw_frag |= FLOW_NW_FRAG_LATER;
236 nexthdr = IPPROTO_FRAGMENT;
243 flow->nw_proto = nexthdr;
248 parse_tcp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
250 const struct tcp_header *tcp = pull_tcp(b);
252 flow->tp_src = tcp->tcp_src;
253 flow->tp_dst = tcp->tcp_dst;
254 packet->l7 = b->data;
259 parse_udp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
261 const struct udp_header *udp = pull_udp(b);
263 flow->tp_src = udp->udp_src;
264 flow->tp_dst = udp->udp_dst;
265 packet->l7 = b->data;
270 parse_icmpv6(struct ofpbuf *b, struct flow *flow)
272 const struct icmp6_hdr *icmp = pull_icmpv6(b);
278 /* The ICMPv6 type and code fields use the 16-bit transport port
279 * fields, so we need to store them in 16-bit network byte order. */
280 flow->tp_src = htons(icmp->icmp6_type);
281 flow->tp_dst = htons(icmp->icmp6_code);
283 if (icmp->icmp6_code == 0 &&
284 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
285 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
286 const struct in6_addr *nd_target;
288 nd_target = ofpbuf_try_pull(b, sizeof *nd_target);
292 flow->nd_target = *nd_target;
294 while (b->size >= 8) {
295 /* The minimum size of an option is 8 bytes, which also is
296 * the size of Ethernet link-layer options. */
297 const struct nd_opt_hdr *nd_opt = b->data;
298 int opt_len = nd_opt->nd_opt_len * 8;
300 if (!opt_len || opt_len > b->size) {
304 /* Store the link layer address if the appropriate option is
305 * provided. It is considered an error if the same link
306 * layer option is specified twice. */
307 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
309 if (eth_addr_is_zero(flow->arp_sha)) {
310 memcpy(flow->arp_sha, nd_opt + 1, ETH_ADDR_LEN);
314 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
316 if (eth_addr_is_zero(flow->arp_tha)) {
317 memcpy(flow->arp_tha, nd_opt + 1, ETH_ADDR_LEN);
323 if (!ofpbuf_try_pull(b, opt_len)) {
332 memset(&flow->nd_target, 0, sizeof(flow->nd_target));
333 memset(flow->arp_sha, 0, sizeof(flow->arp_sha));
334 memset(flow->arp_tha, 0, sizeof(flow->arp_tha));
340 /* Initializes 'flow' members from 'packet', 'skb_priority', 'tnl', and
343 * Initializes 'packet' header pointers as follows:
345 * - packet->l2 to the start of the Ethernet header.
347 * - packet->l2_5 to the start of the MPLS shim header.
349 * - packet->l3 to just past the Ethernet header, or just past the
350 * vlan_header if one is present, to the first byte of the payload of the
353 * - packet->l4 to just past the IPv4 header, if one is present and has a
354 * correct length, and otherwise NULL.
356 * - packet->l7 to just past the TCP or UDP or ICMP header, if one is
357 * present and has a correct length, and otherwise NULL.
360 flow_extract(struct ofpbuf *packet, uint32_t skb_priority, uint32_t skb_mark,
361 const struct flow_tnl *tnl, uint16_t ofp_in_port,
364 struct ofpbuf b = *packet;
365 struct eth_header *eth;
367 COVERAGE_INC(flow_extract);
369 memset(flow, 0, sizeof *flow);
372 ovs_assert(tnl != &flow->tunnel);
375 flow->in_port = ofp_in_port;
376 flow->skb_priority = skb_priority;
377 flow->skb_mark = skb_mark;
385 if (b.size < sizeof *eth) {
391 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
392 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
394 /* dl_type, vlan_tci. */
395 ofpbuf_pull(&b, ETH_ADDR_LEN * 2);
396 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
397 parse_vlan(&b, flow);
399 flow->dl_type = parse_ethertype(&b);
401 /* Parse mpls, copy l3 ttl. */
402 if (eth_type_mpls(flow->dl_type)) {
403 packet->l2_5 = b.data;
404 parse_mpls(&b, flow);
408 flow_extract_l3_onwards(packet, flow, flow->dl_type);
411 /* Initializes l3 and higher 'flow' members from 'packet'
413 * This should be called by or after flow_extract()
415 * Initializes 'packet' header pointers as follows:
417 * - packet->l4 to just past the IPv4 header, if one is present and has a
418 * correct length, and otherwise NULL.
420 * - packet->l7 to just past the TCP or UDP or ICMP header, if one is
421 * present and has a correct length, and otherwise NULL.
424 flow_extract_l3_onwards(struct ofpbuf *packet, struct flow *flow,
429 ofpbuf_use_const(&b, packet->l3, packet->size -
430 (size_t)((char *)packet->l3 - (char *)packet->l2));
433 if (dl_type == htons(ETH_TYPE_IP)) {
434 const struct ip_header *nh = pull_ip(&b);
438 flow->nw_src = get_unaligned_be32(&nh->ip_src);
439 flow->nw_dst = get_unaligned_be32(&nh->ip_dst);
440 flow->nw_proto = nh->ip_proto;
442 flow->nw_tos = nh->ip_tos;
443 if (IP_IS_FRAGMENT(nh->ip_frag_off)) {
444 flow->nw_frag = FLOW_NW_FRAG_ANY;
445 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
446 flow->nw_frag |= FLOW_NW_FRAG_LATER;
449 flow->nw_ttl = nh->ip_ttl;
451 if (!(nh->ip_frag_off & htons(IP_FRAG_OFF_MASK))) {
452 if (flow->nw_proto == IPPROTO_TCP) {
453 parse_tcp(packet, &b, flow);
454 } else if (flow->nw_proto == IPPROTO_UDP) {
455 parse_udp(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 (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_ICMPV6) {
477 if (parse_icmpv6(&b, flow)) {
481 } else if (dl_type == htons(ETH_TYPE_ARP) ||
482 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 = arp->ar_spa;
494 flow->nw_dst = 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 == 19);
521 fmd->tun_id = flow->tunnel.tun_id;
522 fmd->metadata = flow->metadata;
523 memcpy(fmd->regs, flow->regs, sizeof fmd->regs);
524 fmd->in_port = flow->in_port;
528 flow_to_string(const struct flow *flow)
530 struct ds ds = DS_EMPTY_INITIALIZER;
531 flow_format(&ds, flow);
536 flow_tun_flag_to_string(uint32_t flags)
539 case FLOW_TNL_F_DONT_FRAGMENT:
541 case FLOW_TNL_F_CSUM:
551 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
552 uint32_t flags, char del)
560 uint32_t bit = rightmost_1bit(flags);
563 s = bit_to_string(bit);
565 ds_put_format(ds, "%s%c", s, del);
574 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
580 flow_format(struct ds *ds, const struct flow *flow)
584 match_wc_init(&match, flow);
585 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
589 flow_print(FILE *stream, const struct flow *flow)
591 char *s = flow_to_string(flow);
596 /* flow_wildcards functions. */
598 /* Initializes 'wc' as a set of wildcards that matches every packet. */
600 flow_wildcards_init_catchall(struct flow_wildcards *wc)
602 memset(&wc->masks, 0, sizeof wc->masks);
605 /* Initializes 'wc' as an exact-match set of wildcards; that is, 'wc' does not
606 * wildcard any bits or fields. */
608 flow_wildcards_init_exact(struct flow_wildcards *wc)
610 memset(&wc->masks, 0xff, sizeof wc->masks);
611 memset(wc->masks.zeros, 0, sizeof wc->masks.zeros);
614 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
617 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
619 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
622 for (i = 0; i < FLOW_U32S; i++) {
630 /* Initializes 'dst' as the combination of wildcards in 'src1' and 'src2'.
631 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded in
632 * 'src1' or 'src2' or both. */
634 flow_wildcards_combine(struct flow_wildcards *dst,
635 const struct flow_wildcards *src1,
636 const struct flow_wildcards *src2)
638 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
639 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
640 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
643 for (i = 0; i < FLOW_U32S; i++) {
644 dst_u32[i] = src1_u32[i] & src2_u32[i];
648 /* Returns a hash of the wildcards in 'wc'. */
650 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
652 return flow_hash(&wc->masks, basis);
655 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
658 flow_wildcards_equal(const struct flow_wildcards *a,
659 const struct flow_wildcards *b)
661 return flow_equal(&a->masks, &b->masks);
664 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
665 * 'b', false otherwise. */
667 flow_wildcards_has_extra(const struct flow_wildcards *a,
668 const struct flow_wildcards *b)
670 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
671 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
674 for (i = 0; i < FLOW_U32S; i++) {
675 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
682 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
683 * in 'wc' do not need to be equal in 'a' and 'b'. */
685 flow_equal_except(const struct flow *a, const struct flow *b,
686 const struct flow_wildcards *wc)
688 const uint32_t *a_u32 = (const uint32_t *) a;
689 const uint32_t *b_u32 = (const uint32_t *) b;
690 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
693 for (i = 0; i < FLOW_U32S; i++) {
694 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
701 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
702 * (A 0-bit indicates a wildcard bit.) */
704 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
706 wc->masks.regs[idx] = mask;
709 /* Hashes 'flow' based on its L2 through L4 protocol information. */
711 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
716 struct in6_addr ipv6_addr;
721 uint8_t eth_addr[ETH_ADDR_LEN];
727 memset(&fields, 0, sizeof fields);
728 for (i = 0; i < ETH_ADDR_LEN; i++) {
729 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
731 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
732 fields.eth_type = flow->dl_type;
734 /* UDP source and destination port are not taken into account because they
735 * will not necessarily be symmetric in a bidirectional flow. */
736 if (fields.eth_type == htons(ETH_TYPE_IP)) {
737 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
738 fields.ip_proto = flow->nw_proto;
739 if (fields.ip_proto == IPPROTO_TCP) {
740 fields.tp_port = flow->tp_src ^ flow->tp_dst;
742 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
743 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
744 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
745 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
747 for (i=0; i<16; i++) {
748 ipv6_addr[i] = a[i] ^ b[i];
750 fields.ip_proto = flow->nw_proto;
751 if (fields.ip_proto == IPPROTO_TCP) {
752 fields.tp_port = flow->tp_src ^ flow->tp_dst;
755 return jhash_bytes(&fields, sizeof fields, basis);
758 /* Hashes the portions of 'flow' designated by 'fields'. */
760 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
765 case NX_HASH_FIELDS_ETH_SRC:
766 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
768 case NX_HASH_FIELDS_SYMMETRIC_L4:
769 return flow_hash_symmetric_l4(flow, basis);
775 /* Returns a string representation of 'fields'. */
777 flow_hash_fields_to_str(enum nx_hash_fields fields)
780 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
781 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
782 default: return "<unknown>";
786 /* Returns true if the value of 'fields' is supported. Otherwise false. */
788 flow_hash_fields_valid(enum nx_hash_fields fields)
790 return fields == NX_HASH_FIELDS_ETH_SRC
791 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
794 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
795 * OpenFlow 1.0 "dl_vlan" value:
797 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
798 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
799 * 'flow' previously matched packets without a VLAN header).
801 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
802 * without a VLAN tag.
804 * - Other values of 'vid' should not be used. */
806 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
808 if (vid == htons(OFP10_VLAN_NONE)) {
809 flow->vlan_tci = htons(0);
811 vid &= htons(VLAN_VID_MASK);
812 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
813 flow->vlan_tci |= htons(VLAN_CFI) | vid;
817 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
818 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
821 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
823 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
824 flow->vlan_tci &= ~mask;
825 flow->vlan_tci |= vid & mask;
828 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
831 * This function has no effect on the VLAN ID that 'flow' matches.
833 * After calling this function, 'flow' will not match packets without a VLAN
836 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
839 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
840 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
843 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
844 * as an OpenFlow 1.1 "mpls_label" value. */
846 flow_set_mpls_label(struct flow *flow, ovs_be32 label)
848 set_mpls_lse_label(&flow->mpls_lse, label);
851 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
854 flow_set_mpls_ttl(struct flow *flow, uint8_t ttl)
856 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
859 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
862 flow_set_mpls_tc(struct flow *flow, uint8_t tc)
864 set_mpls_lse_tc(&flow->mpls_lse, tc);
867 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
869 flow_set_mpls_bos(struct flow *flow, uint8_t bos)
871 set_mpls_lse_bos(&flow->mpls_lse, bos);
874 /* Puts into 'b' a packet that flow_extract() would parse as having the given
877 * (This is useful only for testing, obviously, and the packet isn't really
878 * valid. It hasn't got some checksums filled in, for one, and lots of fields
879 * are just zeroed.) */
881 flow_compose(struct ofpbuf *b, const struct flow *flow)
883 ovs_be16 inner_dl_type;
885 inner_dl_type = flow_innermost_dl_type(flow);
886 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
887 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
888 struct eth_header *eth = b->l2;
889 eth->eth_type = htons(b->size);
893 if (flow->vlan_tci & htons(VLAN_CFI)) {
894 eth_push_vlan(b, flow->vlan_tci);
897 if (inner_dl_type == htons(ETH_TYPE_IP)) {
898 struct ip_header *ip;
900 b->l3 = ip = ofpbuf_put_zeros(b, sizeof *ip);
901 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
902 ip->ip_tos = flow->nw_tos;
903 ip->ip_ttl = flow->nw_ttl;
904 ip->ip_proto = flow->nw_proto;
905 ip->ip_src = flow->nw_src;
906 ip->ip_dst = flow->nw_dst;
908 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
909 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
910 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
911 ip->ip_frag_off |= htons(100);
914 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
915 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
916 if (flow->nw_proto == IPPROTO_TCP) {
917 struct tcp_header *tcp;
919 b->l4 = tcp = ofpbuf_put_zeros(b, sizeof *tcp);
920 tcp->tcp_src = flow->tp_src;
921 tcp->tcp_dst = flow->tp_dst;
922 tcp->tcp_ctl = TCP_CTL(0, 5);
923 } else if (flow->nw_proto == IPPROTO_UDP) {
924 struct udp_header *udp;
926 b->l4 = udp = ofpbuf_put_zeros(b, sizeof *udp);
927 udp->udp_src = flow->tp_src;
928 udp->udp_dst = flow->tp_dst;
929 } else if (flow->nw_proto == IPPROTO_ICMP) {
930 struct icmp_header *icmp;
932 b->l4 = icmp = ofpbuf_put_zeros(b, sizeof *icmp);
933 icmp->icmp_type = ntohs(flow->tp_src);
934 icmp->icmp_code = ntohs(flow->tp_dst);
935 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
940 ip->ip_tot_len = htons((uint8_t *) b->data + b->size
941 - (uint8_t *) b->l3);
942 ip->ip_csum = csum(ip, sizeof *ip);
943 } else if (inner_dl_type == htons(ETH_TYPE_IPV6)) {
945 } else if (inner_dl_type == htons(ETH_TYPE_ARP) ||
946 inner_dl_type == htons(ETH_TYPE_RARP)) {
947 struct arp_eth_header *arp;
949 b->l3 = arp = ofpbuf_put_zeros(b, sizeof *arp);
950 arp->ar_hrd = htons(1);
951 arp->ar_pro = htons(ETH_TYPE_IP);
952 arp->ar_hln = ETH_ADDR_LEN;
954 arp->ar_op = htons(flow->nw_proto);
956 if (flow->nw_proto == ARP_OP_REQUEST ||
957 flow->nw_proto == ARP_OP_REPLY) {
958 arp->ar_spa = flow->nw_src;
959 arp->ar_tpa = flow->nw_dst;
960 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
961 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
965 if (eth_type_mpls(flow->dl_type)) {
967 push_mpls(b, flow->dl_type, flow->mpls_lse);
971 /* Compressed flow. */
974 miniflow_n_values(const struct miniflow *flow)
979 for (i = 0; i < MINI_N_MAPS; i++) {
980 n += popcount(flow->map[i]);
986 miniflow_alloc_values(struct miniflow *flow, int n)
988 if (n <= MINI_N_INLINE) {
989 return flow->inline_values;
991 COVERAGE_INC(miniflow_malloc);
992 return xmalloc(n * sizeof *flow->values);
996 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
997 * with miniflow_destroy(). */
999 miniflow_init(struct miniflow *dst, const struct flow *src)
1001 const uint32_t *src_u32 = (const uint32_t *) src;
1006 /* Initialize dst->map, counting the number of nonzero elements. */
1008 memset(dst->map, 0, sizeof dst->map);
1009 for (i = 0; i < FLOW_U32S; i++) {
1011 dst->map[i / 32] |= 1u << (i % 32);
1016 /* Initialize dst->values. */
1017 dst->values = miniflow_alloc_values(dst, n);
1019 for (i = 0; i < MINI_N_MAPS; i++) {
1022 for (map = dst->map[i]; map; map = zero_rightmost_1bit(map)) {
1023 dst->values[ofs++] = src_u32[raw_ctz(map) + i * 32];
1028 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1029 * with miniflow_destroy(). */
1031 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
1033 int n = miniflow_n_values(src);
1034 memcpy(dst->map, src->map, sizeof dst->map);
1035 dst->values = miniflow_alloc_values(dst, n);
1036 memcpy(dst->values, src->values, n * sizeof *dst->values);
1039 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
1040 * itself resides; the caller is responsible for that. */
1042 miniflow_destroy(struct miniflow *flow)
1044 if (flow->values != flow->inline_values) {
1049 /* Initializes 'dst' as a copy of 'src'. */
1051 miniflow_expand(const struct miniflow *src, struct flow *dst)
1053 uint32_t *dst_u32 = (uint32_t *) dst;
1057 memset(dst_u32, 0, sizeof *dst);
1060 for (i = 0; i < MINI_N_MAPS; i++) {
1063 for (map = src->map[i]; map; map = zero_rightmost_1bit(map)) {
1064 dst_u32[raw_ctz(map) + i * 32] = src->values[ofs++];
1069 static const uint32_t *
1070 miniflow_get__(const struct miniflow *flow, unsigned int u32_ofs)
1072 if (!(flow->map[u32_ofs / 32] & (1u << (u32_ofs % 32)))) {
1073 static const uint32_t zero = 0;
1076 const uint32_t *p = flow->values;
1078 BUILD_ASSERT(MINI_N_MAPS == 2);
1080 p += popcount(flow->map[0] & ((1u << u32_ofs) - 1));
1082 p += popcount(flow->map[0]);
1083 p += popcount(flow->map[1] & ((1u << (u32_ofs - 32)) - 1));
1089 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1090 * were expanded into a "struct flow". */
1092 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1094 return *miniflow_get__(flow, u32_ofs);
1097 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
1098 * expanded into a "struct flow". */
1100 miniflow_get_be16(const struct miniflow *flow, unsigned int u8_ofs)
1102 const uint32_t *u32p = miniflow_get__(flow, u8_ofs / 4);
1103 const ovs_be16 *be16p = (const ovs_be16 *) u32p;
1104 return be16p[u8_ofs % 4 != 0];
1107 /* Returns the VID within the vlan_tci member of the "struct flow" represented
1110 miniflow_get_vid(const struct miniflow *flow)
1112 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
1113 return vlan_tci_to_vid(tci);
1116 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1118 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1122 for (i = 0; i < MINI_N_MAPS; i++) {
1123 if (a->map[i] != b->map[i]) {
1128 return !memcmp(a->values, b->values,
1129 miniflow_n_values(a) * sizeof *a->values);
1132 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1133 * in 'mask', false if they differ. */
1135 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1136 const struct minimask *mask)
1141 p = mask->masks.values;
1142 for (i = 0; i < MINI_N_MAPS; i++) {
1145 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1146 int ofs = raw_ctz(map) + i * 32;
1148 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p) {
1158 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1159 * in 'mask', false if they differ. */
1161 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1162 const struct minimask *mask)
1164 const uint32_t *b_u32 = (const uint32_t *) b;
1168 p = mask->masks.values;
1169 for (i = 0; i < MINI_N_MAPS; i++) {
1172 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1173 int ofs = raw_ctz(map) + i * 32;
1175 if ((miniflow_get(a, ofs) ^ b_u32[ofs]) & *p) {
1185 /* Returns a hash value for 'flow', given 'basis'. */
1187 miniflow_hash(const struct miniflow *flow, uint32_t basis)
1189 BUILD_ASSERT_DECL(MINI_N_MAPS == 2);
1190 return hash_3words(flow->map[0], flow->map[1],
1191 hash_words(flow->values, miniflow_n_values(flow),
1195 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1196 * 'mask', given 'basis'.
1198 * The hash values returned by this function are the same as those returned by
1199 * flow_hash_in_minimask(), only the form of the arguments differ. */
1201 miniflow_hash_in_minimask(const struct miniflow *flow,
1202 const struct minimask *mask, uint32_t basis)
1204 const uint32_t *p = mask->masks.values;
1209 for (i = 0; i < MINI_N_MAPS; i++) {
1212 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1213 int ofs = raw_ctz(map) + i * 32;
1215 hash = mhash_add(hash, miniflow_get(flow, ofs) & *p);
1220 return mhash_finish(hash, (p - mask->masks.values) * 4);
1223 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1224 * 'mask', given 'basis'.
1226 * The hash values returned by this function are the same as those returned by
1227 * miniflow_hash_in_minimask(), only the form of the arguments differ. */
1229 flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
1232 const uint32_t *flow_u32 = (const uint32_t *) flow;
1233 const uint32_t *p = mask->masks.values;
1238 for (i = 0; i < MINI_N_MAPS; i++) {
1241 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1242 int ofs = raw_ctz(map) + i * 32;
1244 hash = mhash_add(hash, flow_u32[ofs] & *p);
1249 return mhash_finish(hash, (p - mask->masks.values) * 4);
1252 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1253 * with minimask_destroy(). */
1255 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1257 miniflow_init(&mask->masks, &wc->masks);
1260 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1261 * with minimask_destroy(). */
1263 minimask_clone(struct minimask *dst, const struct minimask *src)
1265 miniflow_clone(&dst->masks, &src->masks);
1268 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1270 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1271 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1273 minimask_combine(struct minimask *dst_,
1274 const struct minimask *a_, const struct minimask *b_,
1275 uint32_t storage[FLOW_U32S])
1277 struct miniflow *dst = &dst_->masks;
1278 const struct miniflow *a = &a_->masks;
1279 const struct miniflow *b = &b_->masks;
1283 dst->values = storage;
1284 for (i = 0; i < MINI_N_MAPS; i++) {
1288 for (map = a->map[i] & b->map[i]; map;
1289 map = zero_rightmost_1bit(map)) {
1290 int ofs = raw_ctz(map) + i * 32;
1291 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1294 dst->map[i] |= rightmost_1bit(map);
1295 dst->values[n++] = mask;
1301 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1302 * itself resides; the caller is responsible for that. */
1304 minimask_destroy(struct minimask *mask)
1306 miniflow_destroy(&mask->masks);
1309 /* Initializes 'dst' as a copy of 'src'. */
1311 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1313 miniflow_expand(&mask->masks, &wc->masks);
1316 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1317 * were expanded into a "struct flow_wildcards". */
1319 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1321 return miniflow_get(&mask->masks, u32_ofs);
1324 /* Returns the VID mask within the vlan_tci member of the "struct
1325 * flow_wildcards" represented by 'mask'. */
1327 minimask_get_vid_mask(const struct minimask *mask)
1329 return miniflow_get_vid(&mask->masks);
1332 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1334 minimask_equal(const struct minimask *a, const struct minimask *b)
1336 return miniflow_equal(&a->masks, &b->masks);
1339 /* Returns a hash value for 'mask', given 'basis'. */
1341 minimask_hash(const struct minimask *mask, uint32_t basis)
1343 return miniflow_hash(&mask->masks, basis);
1346 /* Returns true if at least one bit is wildcarded in 'a_' but not in 'b_',
1347 * false otherwise. */
1349 minimask_has_extra(const struct minimask *a_, const struct minimask *b_)
1351 const struct miniflow *a = &a_->masks;
1352 const struct miniflow *b = &b_->masks;
1355 for (i = 0; i < MINI_N_MAPS; i++) {
1358 for (map = a->map[i] | b->map[i]; map;
1359 map = zero_rightmost_1bit(map)) {
1360 int ofs = raw_ctz(map) + i * 32;
1361 uint32_t a_u32 = miniflow_get(a, ofs);
1362 uint32_t b_u32 = miniflow_get(b, ofs);
1364 if ((a_u32 & b_u32) != b_u32) {
1373 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
1376 minimask_is_catchall(const struct minimask *mask_)
1378 const struct miniflow *mask = &mask_->masks;
1380 BUILD_ASSERT(MINI_N_MAPS == 2);
1381 return !(mask->map[0] | mask->map[1]);