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>
23 #include <netinet/in.h>
24 #include <netinet/icmp6.h>
25 #include <netinet/ip6.h>
29 #include "byte-order.h"
32 #include "dynamic-string.h"
35 #include "openflow/openflow.h"
37 #include "unaligned.h"
40 VLOG_DEFINE_THIS_MODULE(flow);
42 COVERAGE_DEFINE(flow_extract);
43 COVERAGE_DEFINE(miniflow_malloc);
45 static struct arp_eth_header *
46 pull_arp(struct ofpbuf *packet)
48 return ofpbuf_try_pull(packet, ARP_ETH_HEADER_LEN);
51 static struct ip_header *
52 pull_ip(struct ofpbuf *packet)
54 if (packet->size >= IP_HEADER_LEN) {
55 struct ip_header *ip = packet->data;
56 int ip_len = IP_IHL(ip->ip_ihl_ver) * 4;
57 if (ip_len >= IP_HEADER_LEN && packet->size >= ip_len) {
58 return ofpbuf_pull(packet, ip_len);
64 static struct tcp_header *
65 pull_tcp(struct ofpbuf *packet)
67 if (packet->size >= TCP_HEADER_LEN) {
68 struct tcp_header *tcp = packet->data;
69 int tcp_len = TCP_OFFSET(tcp->tcp_ctl) * 4;
70 if (tcp_len >= TCP_HEADER_LEN && packet->size >= tcp_len) {
71 return ofpbuf_pull(packet, tcp_len);
77 static struct udp_header *
78 pull_udp(struct ofpbuf *packet)
80 return ofpbuf_try_pull(packet, UDP_HEADER_LEN);
83 static struct icmp_header *
84 pull_icmp(struct ofpbuf *packet)
86 return ofpbuf_try_pull(packet, ICMP_HEADER_LEN);
89 static struct icmp6_hdr *
90 pull_icmpv6(struct ofpbuf *packet)
92 return ofpbuf_try_pull(packet, sizeof(struct icmp6_hdr));
96 parse_vlan(struct ofpbuf *b, struct flow *flow)
99 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
103 if (b->size >= sizeof(struct qtag_prefix) + sizeof(ovs_be16)) {
104 struct qtag_prefix *qp = ofpbuf_pull(b, sizeof *qp);
105 flow->vlan_tci = qp->tci | htons(VLAN_CFI);
110 parse_ethertype(struct ofpbuf *b)
112 struct llc_snap_header *llc;
115 proto = *(ovs_be16 *) ofpbuf_pull(b, sizeof proto);
116 if (ntohs(proto) >= ETH_TYPE_MIN) {
120 if (b->size < sizeof *llc) {
121 return htons(FLOW_DL_TYPE_NONE);
125 if (llc->llc.llc_dsap != LLC_DSAP_SNAP
126 || llc->llc.llc_ssap != LLC_SSAP_SNAP
127 || llc->llc.llc_cntl != LLC_CNTL_SNAP
128 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
129 sizeof llc->snap.snap_org)) {
130 return htons(FLOW_DL_TYPE_NONE);
133 ofpbuf_pull(b, sizeof *llc);
134 return llc->snap.snap_type;
138 parse_ipv6(struct ofpbuf *packet, struct flow *flow)
140 const struct ip6_hdr *nh;
144 nh = ofpbuf_try_pull(packet, sizeof *nh);
149 nexthdr = nh->ip6_nxt;
151 flow->ipv6_src = nh->ip6_src;
152 flow->ipv6_dst = nh->ip6_dst;
154 tc_flow = get_unaligned_be32(&nh->ip6_flow);
155 flow->nw_tos = ntohl(tc_flow) >> 20;
156 flow->ipv6_label = tc_flow & htonl(IPV6_LABEL_MASK);
157 flow->nw_ttl = nh->ip6_hlim;
158 flow->nw_proto = IPPROTO_NONE;
161 if ((nexthdr != IPPROTO_HOPOPTS)
162 && (nexthdr != IPPROTO_ROUTING)
163 && (nexthdr != IPPROTO_DSTOPTS)
164 && (nexthdr != IPPROTO_AH)
165 && (nexthdr != IPPROTO_FRAGMENT)) {
166 /* It's either a terminal header (e.g., TCP, UDP) or one we
167 * don't understand. In either case, we're done with the
168 * packet, so use it to fill in 'nw_proto'. */
172 /* We only verify that at least 8 bytes of the next header are
173 * available, but many of these headers are longer. Ensure that
174 * accesses within the extension header are within those first 8
175 * bytes. All extension headers are required to be at least 8
177 if (packet->size < 8) {
181 if ((nexthdr == IPPROTO_HOPOPTS)
182 || (nexthdr == IPPROTO_ROUTING)
183 || (nexthdr == IPPROTO_DSTOPTS)) {
184 /* These headers, while different, have the fields we care about
185 * in the same location and with the same interpretation. */
186 const struct ip6_ext *ext_hdr = packet->data;
187 nexthdr = ext_hdr->ip6e_nxt;
188 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 1) * 8)) {
191 } else if (nexthdr == IPPROTO_AH) {
192 /* A standard AH definition isn't available, but the fields
193 * we care about are in the same location as the generic
194 * option header--only the header length is calculated
196 const struct ip6_ext *ext_hdr = packet->data;
197 nexthdr = ext_hdr->ip6e_nxt;
198 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 2) * 4)) {
201 } else if (nexthdr == IPPROTO_FRAGMENT) {
202 const struct ip6_frag *frag_hdr = packet->data;
204 nexthdr = frag_hdr->ip6f_nxt;
205 if (!ofpbuf_try_pull(packet, sizeof *frag_hdr)) {
209 /* We only process the first fragment. */
210 if (frag_hdr->ip6f_offlg != htons(0)) {
211 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) == htons(0)) {
212 flow->nw_frag = FLOW_NW_FRAG_ANY;
214 flow->nw_frag |= FLOW_NW_FRAG_LATER;
215 nexthdr = IPPROTO_FRAGMENT;
222 flow->nw_proto = nexthdr;
227 parse_tcp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
229 const struct tcp_header *tcp = pull_tcp(b);
231 flow->tp_src = tcp->tcp_src;
232 flow->tp_dst = tcp->tcp_dst;
233 packet->l7 = b->data;
238 parse_udp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
240 const struct udp_header *udp = pull_udp(b);
242 flow->tp_src = udp->udp_src;
243 flow->tp_dst = udp->udp_dst;
244 packet->l7 = b->data;
249 parse_icmpv6(struct ofpbuf *b, struct flow *flow)
251 const struct icmp6_hdr *icmp = pull_icmpv6(b);
257 /* The ICMPv6 type and code fields use the 16-bit transport port
258 * fields, so we need to store them in 16-bit network byte order. */
259 flow->tp_src = htons(icmp->icmp6_type);
260 flow->tp_dst = htons(icmp->icmp6_code);
262 if (icmp->icmp6_code == 0 &&
263 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
264 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
265 const struct in6_addr *nd_target;
267 nd_target = ofpbuf_try_pull(b, sizeof *nd_target);
271 flow->nd_target = *nd_target;
273 while (b->size >= 8) {
274 /* The minimum size of an option is 8 bytes, which also is
275 * the size of Ethernet link-layer options. */
276 const struct nd_opt_hdr *nd_opt = b->data;
277 int opt_len = nd_opt->nd_opt_len * 8;
279 if (!opt_len || opt_len > b->size) {
283 /* Store the link layer address if the appropriate option is
284 * provided. It is considered an error if the same link
285 * layer option is specified twice. */
286 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
288 if (eth_addr_is_zero(flow->arp_sha)) {
289 memcpy(flow->arp_sha, nd_opt + 1, ETH_ADDR_LEN);
293 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
295 if (eth_addr_is_zero(flow->arp_tha)) {
296 memcpy(flow->arp_tha, nd_opt + 1, ETH_ADDR_LEN);
302 if (!ofpbuf_try_pull(b, opt_len)) {
311 memset(&flow->nd_target, 0, sizeof(flow->nd_target));
312 memset(flow->arp_sha, 0, sizeof(flow->arp_sha));
313 memset(flow->arp_tha, 0, sizeof(flow->arp_tha));
319 /* Initializes 'flow' members from 'packet', 'skb_priority', 'tnl', and
322 * Initializes 'packet' header pointers as follows:
324 * - packet->l2 to the start of the Ethernet header.
326 * - packet->l3 to just past the Ethernet header, or just past the
327 * vlan_header if one is present, to the first byte of the payload of the
330 * - packet->l4 to just past the IPv4 header, if one is present and has a
331 * correct length, and otherwise NULL.
333 * - packet->l7 to just past the TCP or UDP or ICMP header, if one is
334 * present and has a correct length, and otherwise NULL.
337 flow_extract(struct ofpbuf *packet, uint32_t skb_priority,
338 const struct flow_tnl *tnl, uint16_t ofp_in_port,
341 struct ofpbuf b = *packet;
342 struct eth_header *eth;
344 COVERAGE_INC(flow_extract);
346 memset(flow, 0, sizeof *flow);
349 assert(tnl != &flow->tunnel);
352 flow->in_port = ofp_in_port;
353 flow->skb_priority = skb_priority;
360 if (b.size < sizeof *eth) {
366 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
367 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
369 /* dl_type, vlan_tci. */
370 ofpbuf_pull(&b, ETH_ADDR_LEN * 2);
371 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
372 parse_vlan(&b, flow);
374 flow->dl_type = parse_ethertype(&b);
378 if (flow->dl_type == htons(ETH_TYPE_IP)) {
379 const struct ip_header *nh = pull_ip(&b);
383 flow->nw_src = get_unaligned_be32(&nh->ip_src);
384 flow->nw_dst = get_unaligned_be32(&nh->ip_dst);
385 flow->nw_proto = nh->ip_proto;
387 flow->nw_tos = nh->ip_tos;
388 if (IP_IS_FRAGMENT(nh->ip_frag_off)) {
389 flow->nw_frag = FLOW_NW_FRAG_ANY;
390 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
391 flow->nw_frag |= FLOW_NW_FRAG_LATER;
394 flow->nw_ttl = nh->ip_ttl;
396 if (!(nh->ip_frag_off & htons(IP_FRAG_OFF_MASK))) {
397 if (flow->nw_proto == IPPROTO_TCP) {
398 parse_tcp(packet, &b, flow);
399 } else if (flow->nw_proto == IPPROTO_UDP) {
400 parse_udp(packet, &b, flow);
401 } else if (flow->nw_proto == IPPROTO_ICMP) {
402 const struct icmp_header *icmp = pull_icmp(&b);
404 flow->tp_src = htons(icmp->icmp_type);
405 flow->tp_dst = htons(icmp->icmp_code);
411 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
412 if (parse_ipv6(&b, flow)) {
417 if (flow->nw_proto == IPPROTO_TCP) {
418 parse_tcp(packet, &b, flow);
419 } else if (flow->nw_proto == IPPROTO_UDP) {
420 parse_udp(packet, &b, flow);
421 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
422 if (parse_icmpv6(&b, flow)) {
426 } else if (flow->dl_type == htons(ETH_TYPE_ARP)) {
427 const struct arp_eth_header *arp = pull_arp(&b);
428 if (arp && arp->ar_hrd == htons(1)
429 && arp->ar_pro == htons(ETH_TYPE_IP)
430 && arp->ar_hln == ETH_ADDR_LEN
431 && arp->ar_pln == 4) {
432 /* We only match on the lower 8 bits of the opcode. */
433 if (ntohs(arp->ar_op) <= 0xff) {
434 flow->nw_proto = ntohs(arp->ar_op);
437 if ((flow->nw_proto == ARP_OP_REQUEST)
438 || (flow->nw_proto == ARP_OP_REPLY)) {
439 flow->nw_src = arp->ar_spa;
440 flow->nw_dst = arp->ar_tpa;
441 memcpy(flow->arp_sha, arp->ar_sha, ETH_ADDR_LEN);
442 memcpy(flow->arp_tha, arp->ar_tha, ETH_ADDR_LEN);
448 /* For every bit of a field that is wildcarded in 'wildcards', sets the
449 * corresponding bit in 'flow' to zero. */
451 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
453 uint32_t *flow_u32 = (uint32_t *) flow;
454 const uint32_t *wc_u32 = (const uint32_t *) &wildcards->masks;
457 for (i = 0; i < FLOW_U32S; i++) {
458 flow_u32[i] &= wc_u32[i];
462 /* Initializes 'fmd' with the metadata found in 'flow'. */
464 flow_get_metadata(const struct flow *flow, struct flow_metadata *fmd)
466 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 17);
468 fmd->tun_id = flow->tunnel.tun_id;
469 fmd->metadata = flow->metadata;
470 memcpy(fmd->regs, flow->regs, sizeof fmd->regs);
471 fmd->in_port = flow->in_port;
475 flow_to_string(const struct flow *flow)
477 struct ds ds = DS_EMPTY_INITIALIZER;
478 flow_format(&ds, flow);
482 static void format_tunnel_flags(uint16_t flags, struct ds *ds)
484 flags &= ~FLOW_TNL_F_KEY;
486 if (flags & FLOW_TNL_F_DONT_FRAGMENT) {
487 ds_put_cstr(ds, ",df");
488 flags &= ~FLOW_TNL_F_DONT_FRAGMENT;
491 if (flags & FLOW_TNL_F_CSUM) {
492 ds_put_cstr(ds, ",csum");
493 flags &= ~FLOW_TNL_F_CSUM;
497 ds_put_format(ds, ",flags:%#"PRIx16, flags);
502 flow_format(struct ds *ds, const struct flow *flow)
504 ds_put_format(ds, "priority:%"PRIu32, flow->skb_priority);
506 if (flow->tunnel.ip_dst || flow->tunnel.tun_id) {
507 ds_put_cstr(ds, ",tunnel(");
508 ds_put_format(ds, IP_FMT"->"IP_FMT, IP_ARGS(&flow->tunnel.ip_src),
509 IP_ARGS(&flow->tunnel.ip_dst));
511 if (flow->tunnel.flags & FLOW_TNL_F_KEY) {
512 ds_put_format(ds, ",key:%#"PRIx64, ntohll(flow->tunnel.tun_id));
514 ds_put_format(ds, ",tos:%#"PRIx8",ttl:%"PRIu8, flow->tunnel.ip_tos,
515 flow->tunnel.ip_ttl);
516 format_tunnel_flags(flow->tunnel.flags, ds);
517 ds_put_char(ds, ')');
520 ds_put_format(ds, ",metadata:%#"PRIx64
521 ",in_port:%04"PRIx16,
522 ntohll(flow->metadata),
525 ds_put_format(ds, ",tci(");
526 if (flow->vlan_tci) {
527 ds_put_format(ds, "vlan:%"PRIu16",pcp:%d",
528 vlan_tci_to_vid(flow->vlan_tci),
529 vlan_tci_to_pcp(flow->vlan_tci));
531 ds_put_char(ds, '0');
533 ds_put_format(ds, ") mac("ETH_ADDR_FMT"->"ETH_ADDR_FMT
535 ETH_ADDR_ARGS(flow->dl_src),
536 ETH_ADDR_ARGS(flow->dl_dst),
537 ntohs(flow->dl_type));
539 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
540 ds_put_format(ds, " label:%#"PRIx32" proto:%"PRIu8" tos:%#"PRIx8
541 " ttl:%"PRIu8" ipv6(",
542 ntohl(flow->ipv6_label), flow->nw_proto,
543 flow->nw_tos, flow->nw_ttl);
544 print_ipv6_addr(ds, &flow->ipv6_src);
545 ds_put_cstr(ds, "->");
546 print_ipv6_addr(ds, &flow->ipv6_dst);
547 ds_put_char(ds, ')');
548 } else if (flow->dl_type == htons(ETH_TYPE_IP) ||
549 flow->dl_type == htons(ETH_TYPE_ARP)) {
550 ds_put_format(ds, " proto:%"PRIu8" tos:%#"PRIx8" ttl:%"PRIu8
551 " ip("IP_FMT"->"IP_FMT")",
552 flow->nw_proto, flow->nw_tos, flow->nw_ttl,
553 IP_ARGS(&flow->nw_src), IP_ARGS(&flow->nw_dst));
556 ds_put_format(ds, " frag(%s)",
557 flow->nw_frag == FLOW_NW_FRAG_ANY ? "first"
558 : flow->nw_frag == (FLOW_NW_FRAG_ANY | FLOW_NW_FRAG_LATER)
559 ? "later" : "<error>");
561 if (flow->tp_src || flow->tp_dst) {
562 ds_put_format(ds, " port(%"PRIu16"->%"PRIu16")",
563 ntohs(flow->tp_src), ntohs(flow->tp_dst));
565 if (!eth_addr_is_zero(flow->arp_sha) || !eth_addr_is_zero(flow->arp_tha)) {
566 ds_put_format(ds, " arp_ha("ETH_ADDR_FMT"->"ETH_ADDR_FMT")",
567 ETH_ADDR_ARGS(flow->arp_sha),
568 ETH_ADDR_ARGS(flow->arp_tha));
573 flow_print(FILE *stream, const struct flow *flow)
575 char *s = flow_to_string(flow);
580 /* flow_wildcards functions. */
582 /* Initializes 'wc' as a set of wildcards that matches every packet. */
584 flow_wildcards_init_catchall(struct flow_wildcards *wc)
586 memset(&wc->masks, 0, sizeof wc->masks);
589 /* Initializes 'wc' as an exact-match set of wildcards; that is, 'wc' does not
590 * wildcard any bits or fields. */
592 flow_wildcards_init_exact(struct flow_wildcards *wc)
594 memset(&wc->masks, 0xff, sizeof wc->masks);
595 memset(wc->masks.zeros, 0, sizeof wc->masks.zeros);
598 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
601 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
603 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
606 for (i = 0; i < FLOW_U32S; i++) {
614 /* Initializes 'dst' as the combination of wildcards in 'src1' and 'src2'.
615 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded in
616 * 'src1' or 'src2' or both. */
618 flow_wildcards_combine(struct flow_wildcards *dst,
619 const struct flow_wildcards *src1,
620 const struct flow_wildcards *src2)
622 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
623 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
624 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
627 for (i = 0; i < FLOW_U32S; i++) {
628 dst_u32[i] = src1_u32[i] & src2_u32[i];
632 /* Returns a hash of the wildcards in 'wc'. */
634 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
636 return flow_hash(&wc->masks, basis);;
639 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
642 flow_wildcards_equal(const struct flow_wildcards *a,
643 const struct flow_wildcards *b)
645 return flow_equal(&a->masks, &b->masks);
648 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
649 * 'b', false otherwise. */
651 flow_wildcards_has_extra(const struct flow_wildcards *a,
652 const struct flow_wildcards *b)
654 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
655 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
658 for (i = 0; i < FLOW_U32S; i++) {
659 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
666 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
667 * in 'wc' do not need to be equal in 'a' and 'b'. */
669 flow_equal_except(const struct flow *a, const struct flow *b,
670 const struct flow_wildcards *wc)
672 const uint32_t *a_u32 = (const uint32_t *) a;
673 const uint32_t *b_u32 = (const uint32_t *) b;
674 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
677 for (i = 0; i < FLOW_U32S; i++) {
678 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
685 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
686 * (A 0-bit indicates a wildcard bit.) */
688 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
690 wc->masks.regs[idx] = mask;
693 /* Hashes 'flow' based on its L2 through L4 protocol information. */
695 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
700 struct in6_addr ipv6_addr;
705 uint8_t eth_addr[ETH_ADDR_LEN];
711 memset(&fields, 0, sizeof fields);
712 for (i = 0; i < ETH_ADDR_LEN; i++) {
713 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
715 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
716 fields.eth_type = flow->dl_type;
718 /* UDP source and destination port are not taken into account because they
719 * will not necessarily be symmetric in a bidirectional flow. */
720 if (fields.eth_type == htons(ETH_TYPE_IP)) {
721 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
722 fields.ip_proto = flow->nw_proto;
723 if (fields.ip_proto == IPPROTO_TCP) {
724 fields.tp_port = flow->tp_src ^ flow->tp_dst;
726 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
727 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
728 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
729 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
731 for (i=0; i<16; i++) {
732 ipv6_addr[i] = a[i] ^ b[i];
734 fields.ip_proto = flow->nw_proto;
735 if (fields.ip_proto == IPPROTO_TCP) {
736 fields.tp_port = flow->tp_src ^ flow->tp_dst;
739 return hash_bytes(&fields, sizeof fields, basis);
742 /* Hashes the portions of 'flow' designated by 'fields'. */
744 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
749 case NX_HASH_FIELDS_ETH_SRC:
750 return hash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
752 case NX_HASH_FIELDS_SYMMETRIC_L4:
753 return flow_hash_symmetric_l4(flow, basis);
759 /* Returns a string representation of 'fields'. */
761 flow_hash_fields_to_str(enum nx_hash_fields fields)
764 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
765 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
766 default: return "<unknown>";
770 /* Returns true if the value of 'fields' is supported. Otherwise false. */
772 flow_hash_fields_valid(enum nx_hash_fields fields)
774 return fields == NX_HASH_FIELDS_ETH_SRC
775 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
778 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
779 * OpenFlow 1.0 "dl_vlan" value:
781 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
782 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
783 * 'flow' previously matched packets without a VLAN header).
785 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
786 * without a VLAN tag.
788 * - Other values of 'vid' should not be used. */
790 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
792 if (vid == htons(OFP10_VLAN_NONE)) {
793 flow->vlan_tci = htons(0);
795 vid &= htons(VLAN_VID_MASK);
796 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
797 flow->vlan_tci |= htons(VLAN_CFI) | vid;
801 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
802 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
805 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
807 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
808 flow->vlan_tci &= ~mask;
809 flow->vlan_tci |= vid & mask;
812 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
815 * This function has no effect on the VLAN ID that 'flow' matches.
817 * After calling this function, 'flow' will not match packets without a VLAN
820 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
823 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
824 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
827 /* Puts into 'b' a packet that flow_extract() would parse as having the given
830 * (This is useful only for testing, obviously, and the packet isn't really
831 * valid. It hasn't got some checksums filled in, for one, and lots of fields
832 * are just zeroed.) */
834 flow_compose(struct ofpbuf *b, const struct flow *flow)
836 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
837 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
838 struct eth_header *eth = b->l2;
839 eth->eth_type = htons(b->size);
843 if (flow->vlan_tci & htons(VLAN_CFI)) {
844 eth_push_vlan(b, flow->vlan_tci);
847 if (flow->dl_type == htons(ETH_TYPE_IP)) {
848 struct ip_header *ip;
850 b->l3 = ip = ofpbuf_put_zeros(b, sizeof *ip);
851 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
852 ip->ip_tos = flow->nw_tos;
853 ip->ip_proto = flow->nw_proto;
854 ip->ip_src = flow->nw_src;
855 ip->ip_dst = flow->nw_dst;
857 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
858 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
859 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
860 ip->ip_frag_off |= htons(100);
863 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
864 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
865 if (flow->nw_proto == IPPROTO_TCP) {
866 struct tcp_header *tcp;
868 b->l4 = tcp = ofpbuf_put_zeros(b, sizeof *tcp);
869 tcp->tcp_src = flow->tp_src;
870 tcp->tcp_dst = flow->tp_dst;
871 tcp->tcp_ctl = TCP_CTL(0, 5);
872 } else if (flow->nw_proto == IPPROTO_UDP) {
873 struct udp_header *udp;
875 b->l4 = udp = ofpbuf_put_zeros(b, sizeof *udp);
876 udp->udp_src = flow->tp_src;
877 udp->udp_dst = flow->tp_dst;
878 } else if (flow->nw_proto == IPPROTO_ICMP) {
879 struct icmp_header *icmp;
881 b->l4 = icmp = ofpbuf_put_zeros(b, sizeof *icmp);
882 icmp->icmp_type = ntohs(flow->tp_src);
883 icmp->icmp_code = ntohs(flow->tp_dst);
884 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
889 ip->ip_tot_len = htons((uint8_t *) b->data + b->size
890 - (uint8_t *) b->l3);
891 ip->ip_csum = csum(ip, sizeof *ip);
892 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
894 } else if (flow->dl_type == htons(ETH_TYPE_ARP)) {
895 struct arp_eth_header *arp;
897 b->l3 = arp = ofpbuf_put_zeros(b, sizeof *arp);
898 arp->ar_hrd = htons(1);
899 arp->ar_pro = htons(ETH_TYPE_IP);
900 arp->ar_hln = ETH_ADDR_LEN;
902 arp->ar_op = htons(flow->nw_proto);
904 if (flow->nw_proto == ARP_OP_REQUEST ||
905 flow->nw_proto == ARP_OP_REPLY) {
906 arp->ar_spa = flow->nw_src;
907 arp->ar_tpa = flow->nw_dst;
908 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
909 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
914 /* Compressed flow. */
917 miniflow_n_values(const struct miniflow *flow)
922 for (i = 0; i < MINI_N_MAPS; i++) {
923 n += popcount(flow->map[i]);
929 miniflow_alloc_values(struct miniflow *flow, int n)
931 if (n <= MINI_N_INLINE) {
932 return flow->inline_values;
934 COVERAGE_INC(miniflow_malloc);
935 return xmalloc(n * sizeof *flow->values);
939 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
940 * with miniflow_destroy(). */
942 miniflow_init(struct miniflow *dst, const struct flow *src)
944 const uint32_t *src_u32 = (const uint32_t *) src;
949 /* Initialize dst->map, counting the number of nonzero elements. */
951 memset(dst->map, 0, sizeof dst->map);
952 for (i = 0; i < FLOW_U32S; i++) {
954 dst->map[i / 32] |= 1u << (i % 32);
959 /* Initialize dst->values. */
960 dst->values = miniflow_alloc_values(dst, n);
962 for (i = 0; i < MINI_N_MAPS; i++) {
965 for (map = dst->map[i]; map; map = zero_rightmost_1bit(map)) {
966 dst->values[ofs++] = src_u32[raw_ctz(map) + i * 32];
971 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
972 * with miniflow_destroy(). */
974 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
976 int n = miniflow_n_values(src);
977 memcpy(dst->map, src->map, sizeof dst->map);
978 dst->values = miniflow_alloc_values(dst, n);
979 memcpy(dst->values, src->values, n * sizeof *dst->values);
982 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
983 * itself resides; the caller is responsible for that. */
985 miniflow_destroy(struct miniflow *flow)
987 if (flow->values != flow->inline_values) {
992 /* Initializes 'dst' as a copy of 'src'. */
994 miniflow_expand(const struct miniflow *src, struct flow *dst)
996 uint32_t *dst_u32 = (uint32_t *) dst;
1000 memset(dst_u32, 0, sizeof *dst);
1003 for (i = 0; i < MINI_N_MAPS; i++) {
1006 for (map = src->map[i]; map; map = zero_rightmost_1bit(map)) {
1007 dst_u32[raw_ctz(map) + i * 32] = src->values[ofs++];
1012 static const uint32_t *
1013 miniflow_get__(const struct miniflow *flow, unsigned int u32_ofs)
1015 if (!(flow->map[u32_ofs / 32] & (1u << (u32_ofs % 32)))) {
1016 static const uint32_t zero = 0;
1019 const uint32_t *p = flow->values;
1021 BUILD_ASSERT(MINI_N_MAPS == 2);
1023 p += popcount(flow->map[0] & ((1u << u32_ofs) - 1));
1025 p += popcount(flow->map[0]);
1026 p += popcount(flow->map[1] & ((1u << (u32_ofs - 32)) - 1));
1032 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1033 * were expanded into a "struct flow". */
1035 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1037 return *miniflow_get__(flow, u32_ofs);
1040 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
1041 * expanded into a "struct flow". */
1043 miniflow_get_be16(const struct miniflow *flow, unsigned int u8_ofs)
1045 const uint32_t *u32p = miniflow_get__(flow, u8_ofs / 4);
1046 const ovs_be16 *be16p = (const ovs_be16 *) u32p;
1047 return be16p[u8_ofs % 4 != 0];
1050 /* Returns the VID within the vlan_tci member of the "struct flow" represented
1053 miniflow_get_vid(const struct miniflow *flow)
1055 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
1056 return vlan_tci_to_vid(tci);
1059 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1061 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1065 for (i = 0; i < MINI_N_MAPS; i++) {
1066 if (a->map[i] != b->map[i]) {
1071 return !memcmp(a->values, b->values,
1072 miniflow_n_values(a) * sizeof *a->values);
1075 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1076 * in 'mask', false if they differ. */
1078 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1079 const struct minimask *mask)
1084 p = mask->masks.values;
1085 for (i = 0; i < MINI_N_MAPS; i++) {
1088 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1089 int ofs = raw_ctz(map) + i * 32;
1091 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p) {
1101 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1102 * in 'mask', false if they differ. */
1104 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1105 const struct minimask *mask)
1107 const uint32_t *b_u32 = (const uint32_t *) b;
1111 p = mask->masks.values;
1112 for (i = 0; i < MINI_N_MAPS; i++) {
1115 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1116 int ofs = raw_ctz(map) + i * 32;
1118 if ((miniflow_get(a, ofs) ^ b_u32[ofs]) & *p) {
1128 /* Returns a hash value for 'flow', given 'basis'. */
1130 miniflow_hash(const struct miniflow *flow, uint32_t basis)
1132 BUILD_ASSERT_DECL(MINI_N_MAPS == 2);
1133 return hash_3words(flow->map[0], flow->map[1],
1134 hash_words(flow->values, miniflow_n_values(flow),
1138 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1139 * 'mask', given 'basis'.
1141 * The hash values returned by this function are the same as those returned by
1142 * flow_hash_in_minimask(), only the form of the arguments differ. */
1144 miniflow_hash_in_minimask(const struct miniflow *flow,
1145 const struct minimask *mask, uint32_t basis)
1147 const uint32_t *p = mask->masks.values;
1152 for (i = 0; i < MINI_N_MAPS; i++) {
1155 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1156 int ofs = raw_ctz(map) + i * 32;
1158 hash = mhash_add(hash, miniflow_get(flow, ofs) & *p);
1163 return mhash_finish(hash, p - mask->masks.values);
1166 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1167 * 'mask', given 'basis'.
1169 * The hash values returned by this function are the same as those returned by
1170 * miniflow_hash_in_minimask(), only the form of the arguments differ. */
1172 flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
1175 const uint32_t *flow_u32 = (const uint32_t *) flow;
1176 const uint32_t *p = mask->masks.values;
1181 for (i = 0; i < MINI_N_MAPS; i++) {
1184 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1185 int ofs = raw_ctz(map) + i * 32;
1187 hash = mhash_add(hash, flow_u32[ofs] & *p);
1192 return mhash_finish(hash, p - mask->masks.values);
1195 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1196 * with minimask_destroy(). */
1198 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1200 miniflow_init(&mask->masks, &wc->masks);
1203 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1204 * with minimask_destroy(). */
1206 minimask_clone(struct minimask *dst, const struct minimask *src)
1208 miniflow_clone(&dst->masks, &src->masks);
1211 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1213 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1214 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1216 minimask_combine(struct minimask *dst_,
1217 const struct minimask *a_, const struct minimask *b_,
1218 uint32_t storage[FLOW_U32S])
1220 struct miniflow *dst = &dst_->masks;
1221 const struct miniflow *a = &a_->masks;
1222 const struct miniflow *b = &b_->masks;
1226 dst->values = storage;
1227 for (i = 0; i < MINI_N_MAPS; i++) {
1231 for (map = a->map[i] & b->map[i]; map;
1232 map = zero_rightmost_1bit(map)) {
1233 int ofs = raw_ctz(map) + i * 32;
1234 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1237 dst->map[i] |= rightmost_1bit(map);
1238 dst->values[n++] = mask;
1244 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1245 * itself resides; the caller is responsible for that. */
1247 minimask_destroy(struct minimask *mask)
1249 miniflow_destroy(&mask->masks);
1252 /* Initializes 'dst' as a copy of 'src'. */
1254 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1256 miniflow_expand(&mask->masks, &wc->masks);
1259 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1260 * were expanded into a "struct flow_wildcards". */
1262 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1264 return miniflow_get(&mask->masks, u32_ofs);
1267 /* Returns the VID mask within the vlan_tci member of the "struct
1268 * flow_wildcards" represented by 'mask'. */
1270 minimask_get_vid_mask(const struct minimask *mask)
1272 return miniflow_get_vid(&mask->masks);
1275 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1277 minimask_equal(const struct minimask *a, const struct minimask *b)
1279 return miniflow_equal(&a->masks, &b->masks);
1282 /* Returns a hash value for 'mask', given 'basis'. */
1284 minimask_hash(const struct minimask *mask, uint32_t basis)
1286 return miniflow_hash(&mask->masks, basis);
1289 /* Returns true if at least one bit is wildcarded in 'a_' but not in 'b_',
1290 * false otherwise. */
1292 minimask_has_extra(const struct minimask *a_, const struct minimask *b_)
1294 const struct miniflow *a = &a_->masks;
1295 const struct miniflow *b = &b_->masks;
1298 for (i = 0; i < MINI_N_MAPS; i++) {
1301 for (map = a->map[i] | b->map[i]; map;
1302 map = zero_rightmost_1bit(map)) {
1303 int ofs = raw_ctz(map) + i * 32;
1304 uint32_t a_u32 = miniflow_get(a, ofs);
1305 uint32_t b_u32 = miniflow_get(b, ofs);
1307 if ((a_u32 & b_u32) != b_u32) {
1316 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
1319 minimask_is_catchall(const struct minimask *mask_)
1321 const struct miniflow *mask = &mask_->masks;
1323 BUILD_ASSERT(MINI_N_MAPS == 2);
1324 return !(mask->map[0] | mask->map[1]);