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);
150 return llc->snap.snap_type;
154 parse_ipv6(struct ofpbuf *packet, struct flow *flow)
156 const struct ip6_hdr *nh;
160 nh = ofpbuf_try_pull(packet, sizeof *nh);
165 nexthdr = nh->ip6_nxt;
167 flow->ipv6_src = nh->ip6_src;
168 flow->ipv6_dst = nh->ip6_dst;
170 tc_flow = get_unaligned_be32(&nh->ip6_flow);
171 flow->nw_tos = ntohl(tc_flow) >> 20;
172 flow->ipv6_label = tc_flow & htonl(IPV6_LABEL_MASK);
173 flow->nw_ttl = nh->ip6_hlim;
174 flow->nw_proto = IPPROTO_NONE;
177 if ((nexthdr != IPPROTO_HOPOPTS)
178 && (nexthdr != IPPROTO_ROUTING)
179 && (nexthdr != IPPROTO_DSTOPTS)
180 && (nexthdr != IPPROTO_AH)
181 && (nexthdr != IPPROTO_FRAGMENT)) {
182 /* It's either a terminal header (e.g., TCP, UDP) or one we
183 * don't understand. In either case, we're done with the
184 * packet, so use it to fill in 'nw_proto'. */
188 /* We only verify that at least 8 bytes of the next header are
189 * available, but many of these headers are longer. Ensure that
190 * accesses within the extension header are within those first 8
191 * bytes. All extension headers are required to be at least 8
193 if (packet->size < 8) {
197 if ((nexthdr == IPPROTO_HOPOPTS)
198 || (nexthdr == IPPROTO_ROUTING)
199 || (nexthdr == IPPROTO_DSTOPTS)) {
200 /* These headers, while different, have the fields we care about
201 * in the same location and with the same interpretation. */
202 const struct ip6_ext *ext_hdr = packet->data;
203 nexthdr = ext_hdr->ip6e_nxt;
204 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 1) * 8)) {
207 } else if (nexthdr == IPPROTO_AH) {
208 /* A standard AH definition isn't available, but the fields
209 * we care about are in the same location as the generic
210 * option header--only the header length is calculated
212 const struct ip6_ext *ext_hdr = packet->data;
213 nexthdr = ext_hdr->ip6e_nxt;
214 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 2) * 4)) {
217 } else if (nexthdr == IPPROTO_FRAGMENT) {
218 const struct ip6_frag *frag_hdr = packet->data;
220 nexthdr = frag_hdr->ip6f_nxt;
221 if (!ofpbuf_try_pull(packet, sizeof *frag_hdr)) {
225 /* We only process the first fragment. */
226 if (frag_hdr->ip6f_offlg != htons(0)) {
227 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) == htons(0)) {
228 flow->nw_frag = FLOW_NW_FRAG_ANY;
230 flow->nw_frag |= FLOW_NW_FRAG_LATER;
231 nexthdr = IPPROTO_FRAGMENT;
238 flow->nw_proto = nexthdr;
243 parse_tcp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
245 const struct tcp_header *tcp = pull_tcp(b);
247 flow->tp_src = tcp->tcp_src;
248 flow->tp_dst = tcp->tcp_dst;
249 packet->l7 = b->data;
254 parse_udp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
256 const struct udp_header *udp = pull_udp(b);
258 flow->tp_src = udp->udp_src;
259 flow->tp_dst = udp->udp_dst;
260 packet->l7 = b->data;
265 parse_icmpv6(struct ofpbuf *b, struct flow *flow)
267 const struct icmp6_hdr *icmp = pull_icmpv6(b);
273 /* The ICMPv6 type and code fields use the 16-bit transport port
274 * fields, so we need to store them in 16-bit network byte order. */
275 flow->tp_src = htons(icmp->icmp6_type);
276 flow->tp_dst = htons(icmp->icmp6_code);
278 if (icmp->icmp6_code == 0 &&
279 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
280 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
281 const struct in6_addr *nd_target;
283 nd_target = ofpbuf_try_pull(b, sizeof *nd_target);
287 flow->nd_target = *nd_target;
289 while (b->size >= 8) {
290 /* The minimum size of an option is 8 bytes, which also is
291 * the size of Ethernet link-layer options. */
292 const struct nd_opt_hdr *nd_opt = b->data;
293 int opt_len = nd_opt->nd_opt_len * 8;
295 if (!opt_len || opt_len > b->size) {
299 /* Store the link layer address if the appropriate option is
300 * provided. It is considered an error if the same link
301 * layer option is specified twice. */
302 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
304 if (eth_addr_is_zero(flow->arp_sha)) {
305 memcpy(flow->arp_sha, nd_opt + 1, ETH_ADDR_LEN);
309 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
311 if (eth_addr_is_zero(flow->arp_tha)) {
312 memcpy(flow->arp_tha, nd_opt + 1, ETH_ADDR_LEN);
318 if (!ofpbuf_try_pull(b, opt_len)) {
327 memset(&flow->nd_target, 0, sizeof(flow->nd_target));
328 memset(flow->arp_sha, 0, sizeof(flow->arp_sha));
329 memset(flow->arp_tha, 0, sizeof(flow->arp_tha));
335 /* Initializes 'flow' members from 'packet', 'skb_priority', 'tnl', and
338 * Initializes 'packet' header pointers as follows:
340 * - packet->l2 to the start of the Ethernet header.
342 * - packet->l2_5 to the start of the MPLS shim header.
344 * - packet->l3 to just past the Ethernet header, or just past the
345 * vlan_header if one is present, to the first byte of the payload of the
348 * - packet->l4 to just past the IPv4 header, if one is present and has a
349 * correct length, and otherwise NULL.
351 * - packet->l7 to just past the TCP or UDP or ICMP header, if one is
352 * present and has a correct length, and otherwise NULL.
355 flow_extract(struct ofpbuf *packet, uint32_t skb_priority, uint32_t skb_mark,
356 const struct flow_tnl *tnl, uint16_t ofp_in_port,
359 struct ofpbuf b = *packet;
360 struct eth_header *eth;
362 COVERAGE_INC(flow_extract);
364 memset(flow, 0, sizeof *flow);
367 ovs_assert(tnl != &flow->tunnel);
370 flow->in_port = ofp_in_port;
371 flow->skb_priority = skb_priority;
372 flow->skb_mark = skb_mark;
380 if (b.size < sizeof *eth) {
386 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
387 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
389 /* dl_type, vlan_tci. */
390 ofpbuf_pull(&b, ETH_ADDR_LEN * 2);
391 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
392 parse_vlan(&b, flow);
394 flow->dl_type = parse_ethertype(&b);
396 /* Parse mpls, copy l3 ttl. */
397 if (eth_type_mpls(flow->dl_type)) {
398 packet->l2_5 = b.data;
399 parse_mpls(&b, flow);
403 flow_extract_l3_onwards(packet, flow, flow->dl_type);
406 /* Initializes l3 and higher 'flow' members from 'packet'
408 * This should be called by or after flow_extract()
410 * Initializes 'packet' header pointers as follows:
412 * - packet->l4 to just past the IPv4 header, if one is present and has a
413 * correct length, and otherwise NULL.
415 * - packet->l7 to just past the TCP or UDP or ICMP header, if one is
416 * present and has a correct length, and otherwise NULL.
419 flow_extract_l3_onwards(struct ofpbuf *packet, struct flow *flow,
424 ofpbuf_use_const(&b, packet->l3, packet->size -
425 (size_t)((char *)packet->l3 - (char *)packet->l2));
428 if (dl_type == htons(ETH_TYPE_IP)) {
429 const struct ip_header *nh = pull_ip(&b);
433 flow->nw_src = get_unaligned_be32(&nh->ip_src);
434 flow->nw_dst = get_unaligned_be32(&nh->ip_dst);
435 flow->nw_proto = nh->ip_proto;
437 flow->nw_tos = nh->ip_tos;
438 if (IP_IS_FRAGMENT(nh->ip_frag_off)) {
439 flow->nw_frag = FLOW_NW_FRAG_ANY;
440 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
441 flow->nw_frag |= FLOW_NW_FRAG_LATER;
444 flow->nw_ttl = nh->ip_ttl;
446 if (!(nh->ip_frag_off & htons(IP_FRAG_OFF_MASK))) {
447 if (flow->nw_proto == IPPROTO_TCP) {
448 parse_tcp(packet, &b, flow);
449 } else if (flow->nw_proto == IPPROTO_UDP) {
450 parse_udp(packet, &b, flow);
451 } else if (flow->nw_proto == IPPROTO_ICMP) {
452 const struct icmp_header *icmp = pull_icmp(&b);
454 flow->tp_src = htons(icmp->icmp_type);
455 flow->tp_dst = htons(icmp->icmp_code);
461 } else if (dl_type == htons(ETH_TYPE_IPV6)) {
462 if (parse_ipv6(&b, flow)) {
467 if (flow->nw_proto == IPPROTO_TCP) {
468 parse_tcp(packet, &b, flow);
469 } else if (flow->nw_proto == IPPROTO_UDP) {
470 parse_udp(packet, &b, flow);
471 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
472 if (parse_icmpv6(&b, flow)) {
476 } else if (dl_type == htons(ETH_TYPE_ARP) ||
477 dl_type == htons(ETH_TYPE_RARP)) {
478 const struct arp_eth_header *arp = pull_arp(&b);
479 if (arp && arp->ar_hrd == htons(1)
480 && arp->ar_pro == htons(ETH_TYPE_IP)
481 && arp->ar_hln == ETH_ADDR_LEN
482 && arp->ar_pln == 4) {
483 /* We only match on the lower 8 bits of the opcode. */
484 if (ntohs(arp->ar_op) <= 0xff) {
485 flow->nw_proto = ntohs(arp->ar_op);
488 flow->nw_src = arp->ar_spa;
489 flow->nw_dst = arp->ar_tpa;
490 memcpy(flow->arp_sha, arp->ar_sha, ETH_ADDR_LEN);
491 memcpy(flow->arp_tha, arp->ar_tha, ETH_ADDR_LEN);
496 /* For every bit of a field that is wildcarded in 'wildcards', sets the
497 * corresponding bit in 'flow' to zero. */
499 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
501 uint32_t *flow_u32 = (uint32_t *) flow;
502 const uint32_t *wc_u32 = (const uint32_t *) &wildcards->masks;
505 for (i = 0; i < FLOW_U32S; i++) {
506 flow_u32[i] &= wc_u32[i];
510 /* Initializes 'fmd' with the metadata found in 'flow'. */
512 flow_get_metadata(const struct flow *flow, struct flow_metadata *fmd)
514 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 19);
516 fmd->tun_id = flow->tunnel.tun_id;
517 fmd->metadata = flow->metadata;
518 memcpy(fmd->regs, flow->regs, sizeof fmd->regs);
519 fmd->in_port = flow->in_port;
523 flow_to_string(const struct flow *flow)
525 struct ds ds = DS_EMPTY_INITIALIZER;
526 flow_format(&ds, flow);
531 flow_tun_flag_to_string(uint32_t flags)
534 case FLOW_TNL_F_DONT_FRAGMENT:
536 case FLOW_TNL_F_CSUM:
546 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
547 uint32_t flags, char del)
555 uint32_t bit = rightmost_1bit(flags);
558 s = bit_to_string(bit);
560 ds_put_format(ds, "%s%c", s, del);
569 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
575 flow_format(struct ds *ds, const struct flow *flow)
579 match_wc_init(&match, flow);
580 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
584 flow_print(FILE *stream, const struct flow *flow)
586 char *s = flow_to_string(flow);
591 /* flow_wildcards functions. */
593 /* Initializes 'wc' as a set of wildcards that matches every packet. */
595 flow_wildcards_init_catchall(struct flow_wildcards *wc)
597 memset(&wc->masks, 0, sizeof wc->masks);
600 /* Initializes 'wc' as an exact-match set of wildcards; that is, 'wc' does not
601 * wildcard any bits or fields. */
603 flow_wildcards_init_exact(struct flow_wildcards *wc)
605 memset(&wc->masks, 0xff, sizeof wc->masks);
606 memset(wc->masks.zeros, 0, sizeof wc->masks.zeros);
609 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
612 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
614 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
617 for (i = 0; i < FLOW_U32S; i++) {
625 /* Initializes 'dst' as the combination of wildcards in 'src1' and 'src2'.
626 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded in
627 * 'src1' or 'src2' or both. */
629 flow_wildcards_combine(struct flow_wildcards *dst,
630 const struct flow_wildcards *src1,
631 const struct flow_wildcards *src2)
633 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
634 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
635 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
638 for (i = 0; i < FLOW_U32S; i++) {
639 dst_u32[i] = src1_u32[i] & src2_u32[i];
643 /* Returns a hash of the wildcards in 'wc'. */
645 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
647 return flow_hash(&wc->masks, basis);
650 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
653 flow_wildcards_equal(const struct flow_wildcards *a,
654 const struct flow_wildcards *b)
656 return flow_equal(&a->masks, &b->masks);
659 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
660 * 'b', false otherwise. */
662 flow_wildcards_has_extra(const struct flow_wildcards *a,
663 const struct flow_wildcards *b)
665 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
666 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
669 for (i = 0; i < FLOW_U32S; i++) {
670 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
677 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
678 * in 'wc' do not need to be equal in 'a' and 'b'. */
680 flow_equal_except(const struct flow *a, const struct flow *b,
681 const struct flow_wildcards *wc)
683 const uint32_t *a_u32 = (const uint32_t *) a;
684 const uint32_t *b_u32 = (const uint32_t *) b;
685 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
688 for (i = 0; i < FLOW_U32S; i++) {
689 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
696 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
697 * (A 0-bit indicates a wildcard bit.) */
699 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
701 wc->masks.regs[idx] = mask;
704 /* Hashes 'flow' based on its L2 through L4 protocol information. */
706 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
711 struct in6_addr ipv6_addr;
716 uint8_t eth_addr[ETH_ADDR_LEN];
722 memset(&fields, 0, sizeof fields);
723 for (i = 0; i < ETH_ADDR_LEN; i++) {
724 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
726 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
727 fields.eth_type = flow->dl_type;
729 /* UDP source and destination port are not taken into account because they
730 * will not necessarily be symmetric in a bidirectional flow. */
731 if (fields.eth_type == htons(ETH_TYPE_IP)) {
732 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
733 fields.ip_proto = flow->nw_proto;
734 if (fields.ip_proto == IPPROTO_TCP) {
735 fields.tp_port = flow->tp_src ^ flow->tp_dst;
737 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
738 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
739 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
740 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
742 for (i=0; i<16; i++) {
743 ipv6_addr[i] = a[i] ^ b[i];
745 fields.ip_proto = flow->nw_proto;
746 if (fields.ip_proto == IPPROTO_TCP) {
747 fields.tp_port = flow->tp_src ^ flow->tp_dst;
750 return jhash_bytes(&fields, sizeof fields, basis);
753 /* Hashes the portions of 'flow' designated by 'fields'. */
755 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
760 case NX_HASH_FIELDS_ETH_SRC:
761 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
763 case NX_HASH_FIELDS_SYMMETRIC_L4:
764 return flow_hash_symmetric_l4(flow, basis);
770 /* Returns a string representation of 'fields'. */
772 flow_hash_fields_to_str(enum nx_hash_fields fields)
775 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
776 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
777 default: return "<unknown>";
781 /* Returns true if the value of 'fields' is supported. Otherwise false. */
783 flow_hash_fields_valid(enum nx_hash_fields fields)
785 return fields == NX_HASH_FIELDS_ETH_SRC
786 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
789 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
790 * OpenFlow 1.0 "dl_vlan" value:
792 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
793 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
794 * 'flow' previously matched packets without a VLAN header).
796 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
797 * without a VLAN tag.
799 * - Other values of 'vid' should not be used. */
801 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
803 if (vid == htons(OFP10_VLAN_NONE)) {
804 flow->vlan_tci = htons(0);
806 vid &= htons(VLAN_VID_MASK);
807 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
808 flow->vlan_tci |= htons(VLAN_CFI) | vid;
812 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
813 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
816 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
818 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
819 flow->vlan_tci &= ~mask;
820 flow->vlan_tci |= vid & mask;
823 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
826 * This function has no effect on the VLAN ID that 'flow' matches.
828 * After calling this function, 'flow' will not match packets without a VLAN
831 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
834 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
835 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
838 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
839 * as an OpenFlow 1.1 "mpls_label" value. */
841 flow_set_mpls_label(struct flow *flow, ovs_be32 label)
843 set_mpls_lse_label(&flow->mpls_lse, label);
846 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
849 flow_set_mpls_tc(struct flow *flow, uint8_t tc)
851 set_mpls_lse_tc(&flow->mpls_lse, tc);
854 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
856 flow_set_mpls_bos(struct flow *flow, uint8_t bos)
858 set_mpls_lse_bos(&flow->mpls_lse, bos);
861 /* Puts into 'b' a packet that flow_extract() would parse as having the given
864 * (This is useful only for testing, obviously, and the packet isn't really
865 * valid. It hasn't got some checksums filled in, for one, and lots of fields
866 * are just zeroed.) */
868 flow_compose(struct ofpbuf *b, const struct flow *flow)
870 ovs_be16 inner_dl_type;
872 inner_dl_type = flow_innermost_dl_type(flow);
873 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
874 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
875 struct eth_header *eth = b->l2;
876 eth->eth_type = htons(b->size);
880 if (flow->vlan_tci & htons(VLAN_CFI)) {
881 eth_push_vlan(b, flow->vlan_tci);
884 if (inner_dl_type == htons(ETH_TYPE_IP)) {
885 struct ip_header *ip;
887 b->l3 = ip = ofpbuf_put_zeros(b, sizeof *ip);
888 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
889 ip->ip_tos = flow->nw_tos;
890 ip->ip_ttl = flow->nw_ttl;
891 ip->ip_proto = flow->nw_proto;
892 ip->ip_src = flow->nw_src;
893 ip->ip_dst = flow->nw_dst;
895 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
896 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
897 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
898 ip->ip_frag_off |= htons(100);
901 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
902 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
903 if (flow->nw_proto == IPPROTO_TCP) {
904 struct tcp_header *tcp;
906 b->l4 = tcp = ofpbuf_put_zeros(b, sizeof *tcp);
907 tcp->tcp_src = flow->tp_src;
908 tcp->tcp_dst = flow->tp_dst;
909 tcp->tcp_ctl = TCP_CTL(0, 5);
910 } else if (flow->nw_proto == IPPROTO_UDP) {
911 struct udp_header *udp;
913 b->l4 = udp = ofpbuf_put_zeros(b, sizeof *udp);
914 udp->udp_src = flow->tp_src;
915 udp->udp_dst = flow->tp_dst;
916 } else if (flow->nw_proto == IPPROTO_ICMP) {
917 struct icmp_header *icmp;
919 b->l4 = icmp = ofpbuf_put_zeros(b, sizeof *icmp);
920 icmp->icmp_type = ntohs(flow->tp_src);
921 icmp->icmp_code = ntohs(flow->tp_dst);
922 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
927 ip->ip_tot_len = htons((uint8_t *) b->data + b->size
928 - (uint8_t *) b->l3);
929 ip->ip_csum = csum(ip, sizeof *ip);
930 } else if (inner_dl_type == htons(ETH_TYPE_IPV6)) {
932 } else if (inner_dl_type == htons(ETH_TYPE_ARP) ||
933 inner_dl_type == htons(ETH_TYPE_RARP)) {
934 struct arp_eth_header *arp;
936 b->l3 = arp = ofpbuf_put_zeros(b, sizeof *arp);
937 arp->ar_hrd = htons(1);
938 arp->ar_pro = htons(ETH_TYPE_IP);
939 arp->ar_hln = ETH_ADDR_LEN;
941 arp->ar_op = htons(flow->nw_proto);
943 if (flow->nw_proto == ARP_OP_REQUEST ||
944 flow->nw_proto == ARP_OP_REPLY) {
945 arp->ar_spa = flow->nw_src;
946 arp->ar_tpa = flow->nw_dst;
947 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
948 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
952 if (eth_type_mpls(flow->dl_type)) {
954 push_mpls(b, flow->dl_type, flow->mpls_lse);
958 /* Compressed flow. */
961 miniflow_n_values(const struct miniflow *flow)
966 for (i = 0; i < MINI_N_MAPS; i++) {
967 n += popcount(flow->map[i]);
973 miniflow_alloc_values(struct miniflow *flow, int n)
975 if (n <= MINI_N_INLINE) {
976 return flow->inline_values;
978 COVERAGE_INC(miniflow_malloc);
979 return xmalloc(n * sizeof *flow->values);
983 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
984 * with miniflow_destroy(). */
986 miniflow_init(struct miniflow *dst, const struct flow *src)
988 const uint32_t *src_u32 = (const uint32_t *) src;
993 /* Initialize dst->map, counting the number of nonzero elements. */
995 memset(dst->map, 0, sizeof dst->map);
996 for (i = 0; i < FLOW_U32S; i++) {
998 dst->map[i / 32] |= 1u << (i % 32);
1003 /* Initialize dst->values. */
1004 dst->values = miniflow_alloc_values(dst, n);
1006 for (i = 0; i < MINI_N_MAPS; i++) {
1009 for (map = dst->map[i]; map; map = zero_rightmost_1bit(map)) {
1010 dst->values[ofs++] = src_u32[raw_ctz(map) + i * 32];
1015 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1016 * with miniflow_destroy(). */
1018 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
1020 int n = miniflow_n_values(src);
1021 memcpy(dst->map, src->map, sizeof dst->map);
1022 dst->values = miniflow_alloc_values(dst, n);
1023 memcpy(dst->values, src->values, n * sizeof *dst->values);
1026 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
1027 * itself resides; the caller is responsible for that. */
1029 miniflow_destroy(struct miniflow *flow)
1031 if (flow->values != flow->inline_values) {
1036 /* Initializes 'dst' as a copy of 'src'. */
1038 miniflow_expand(const struct miniflow *src, struct flow *dst)
1040 uint32_t *dst_u32 = (uint32_t *) dst;
1044 memset(dst_u32, 0, sizeof *dst);
1047 for (i = 0; i < MINI_N_MAPS; i++) {
1050 for (map = src->map[i]; map; map = zero_rightmost_1bit(map)) {
1051 dst_u32[raw_ctz(map) + i * 32] = src->values[ofs++];
1056 static const uint32_t *
1057 miniflow_get__(const struct miniflow *flow, unsigned int u32_ofs)
1059 if (!(flow->map[u32_ofs / 32] & (1u << (u32_ofs % 32)))) {
1060 static const uint32_t zero = 0;
1063 const uint32_t *p = flow->values;
1065 BUILD_ASSERT(MINI_N_MAPS == 2);
1067 p += popcount(flow->map[0] & ((1u << u32_ofs) - 1));
1069 p += popcount(flow->map[0]);
1070 p += popcount(flow->map[1] & ((1u << (u32_ofs - 32)) - 1));
1076 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1077 * were expanded into a "struct flow". */
1079 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1081 return *miniflow_get__(flow, u32_ofs);
1084 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
1085 * expanded into a "struct flow". */
1087 miniflow_get_be16(const struct miniflow *flow, unsigned int u8_ofs)
1089 const uint32_t *u32p = miniflow_get__(flow, u8_ofs / 4);
1090 const ovs_be16 *be16p = (const ovs_be16 *) u32p;
1091 return be16p[u8_ofs % 4 != 0];
1094 /* Returns the VID within the vlan_tci member of the "struct flow" represented
1097 miniflow_get_vid(const struct miniflow *flow)
1099 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
1100 return vlan_tci_to_vid(tci);
1103 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1105 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1109 for (i = 0; i < MINI_N_MAPS; i++) {
1110 if (a->map[i] != b->map[i]) {
1115 return !memcmp(a->values, b->values,
1116 miniflow_n_values(a) * sizeof *a->values);
1119 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1120 * in 'mask', false if they differ. */
1122 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1123 const struct minimask *mask)
1128 p = mask->masks.values;
1129 for (i = 0; i < MINI_N_MAPS; i++) {
1132 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1133 int ofs = raw_ctz(map) + i * 32;
1135 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p) {
1145 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1146 * in 'mask', false if they differ. */
1148 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1149 const struct minimask *mask)
1151 const uint32_t *b_u32 = (const uint32_t *) b;
1155 p = mask->masks.values;
1156 for (i = 0; i < MINI_N_MAPS; i++) {
1159 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1160 int ofs = raw_ctz(map) + i * 32;
1162 if ((miniflow_get(a, ofs) ^ b_u32[ofs]) & *p) {
1172 /* Returns a hash value for 'flow', given 'basis'. */
1174 miniflow_hash(const struct miniflow *flow, uint32_t basis)
1176 BUILD_ASSERT_DECL(MINI_N_MAPS == 2);
1177 return hash_3words(flow->map[0], flow->map[1],
1178 hash_words(flow->values, miniflow_n_values(flow),
1182 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1183 * 'mask', given 'basis'.
1185 * The hash values returned by this function are the same as those returned by
1186 * flow_hash_in_minimask(), only the form of the arguments differ. */
1188 miniflow_hash_in_minimask(const struct miniflow *flow,
1189 const struct minimask *mask, uint32_t basis)
1191 const uint32_t *p = mask->masks.values;
1196 for (i = 0; i < MINI_N_MAPS; i++) {
1199 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1200 int ofs = raw_ctz(map) + i * 32;
1202 hash = mhash_add(hash, miniflow_get(flow, ofs) & *p);
1207 return mhash_finish(hash, (p - mask->masks.values) * 4);
1210 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1211 * 'mask', given 'basis'.
1213 * The hash values returned by this function are the same as those returned by
1214 * miniflow_hash_in_minimask(), only the form of the arguments differ. */
1216 flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
1219 const uint32_t *flow_u32 = (const uint32_t *) flow;
1220 const uint32_t *p = mask->masks.values;
1225 for (i = 0; i < MINI_N_MAPS; i++) {
1228 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1229 int ofs = raw_ctz(map) + i * 32;
1231 hash = mhash_add(hash, flow_u32[ofs] & *p);
1236 return mhash_finish(hash, (p - mask->masks.values) * 4);
1239 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1240 * with minimask_destroy(). */
1242 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1244 miniflow_init(&mask->masks, &wc->masks);
1247 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1248 * with minimask_destroy(). */
1250 minimask_clone(struct minimask *dst, const struct minimask *src)
1252 miniflow_clone(&dst->masks, &src->masks);
1255 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1257 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1258 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1260 minimask_combine(struct minimask *dst_,
1261 const struct minimask *a_, const struct minimask *b_,
1262 uint32_t storage[FLOW_U32S])
1264 struct miniflow *dst = &dst_->masks;
1265 const struct miniflow *a = &a_->masks;
1266 const struct miniflow *b = &b_->masks;
1270 dst->values = storage;
1271 for (i = 0; i < MINI_N_MAPS; i++) {
1275 for (map = a->map[i] & b->map[i]; map;
1276 map = zero_rightmost_1bit(map)) {
1277 int ofs = raw_ctz(map) + i * 32;
1278 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1281 dst->map[i] |= rightmost_1bit(map);
1282 dst->values[n++] = mask;
1288 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1289 * itself resides; the caller is responsible for that. */
1291 minimask_destroy(struct minimask *mask)
1293 miniflow_destroy(&mask->masks);
1296 /* Initializes 'dst' as a copy of 'src'. */
1298 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1300 miniflow_expand(&mask->masks, &wc->masks);
1303 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1304 * were expanded into a "struct flow_wildcards". */
1306 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1308 return miniflow_get(&mask->masks, u32_ofs);
1311 /* Returns the VID mask within the vlan_tci member of the "struct
1312 * flow_wildcards" represented by 'mask'. */
1314 minimask_get_vid_mask(const struct minimask *mask)
1316 return miniflow_get_vid(&mask->masks);
1319 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1321 minimask_equal(const struct minimask *a, const struct minimask *b)
1323 return miniflow_equal(&a->masks, &b->masks);
1326 /* Returns a hash value for 'mask', given 'basis'. */
1328 minimask_hash(const struct minimask *mask, uint32_t basis)
1330 return miniflow_hash(&mask->masks, basis);
1333 /* Returns true if at least one bit is wildcarded in 'a_' but not in 'b_',
1334 * false otherwise. */
1336 minimask_has_extra(const struct minimask *a_, const struct minimask *b_)
1338 const struct miniflow *a = &a_->masks;
1339 const struct miniflow *b = &b_->masks;
1342 for (i = 0; i < MINI_N_MAPS; i++) {
1345 for (map = a->map[i] | b->map[i]; map;
1346 map = zero_rightmost_1bit(map)) {
1347 int ofs = raw_ctz(map) + i * 32;
1348 uint32_t a_u32 = miniflow_get(a, ofs);
1349 uint32_t b_u32 = miniflow_get(b, ofs);
1351 if ((a_u32 & b_u32) != b_u32) {
1360 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
1363 minimask_is_catchall(const struct minimask *mask_)
1365 const struct miniflow *mask = &mask_->masks;
1367 BUILD_ASSERT(MINI_N_MAPS == 2);
1368 return !(mask->map[0] | mask->map[1]);