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);
351 flow->in_port = ofp_in_port;
352 flow->skb_priority = skb_priority;
359 if (b.size < sizeof *eth) {
365 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
366 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
368 /* dl_type, vlan_tci. */
369 ofpbuf_pull(&b, ETH_ADDR_LEN * 2);
370 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
371 parse_vlan(&b, flow);
373 flow->dl_type = parse_ethertype(&b);
377 if (flow->dl_type == htons(ETH_TYPE_IP)) {
378 const struct ip_header *nh = pull_ip(&b);
382 flow->nw_src = get_unaligned_be32(&nh->ip_src);
383 flow->nw_dst = get_unaligned_be32(&nh->ip_dst);
384 flow->nw_proto = nh->ip_proto;
386 flow->nw_tos = nh->ip_tos;
387 if (IP_IS_FRAGMENT(nh->ip_frag_off)) {
388 flow->nw_frag = FLOW_NW_FRAG_ANY;
389 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
390 flow->nw_frag |= FLOW_NW_FRAG_LATER;
393 flow->nw_ttl = nh->ip_ttl;
395 if (!(nh->ip_frag_off & htons(IP_FRAG_OFF_MASK))) {
396 if (flow->nw_proto == IPPROTO_TCP) {
397 parse_tcp(packet, &b, flow);
398 } else if (flow->nw_proto == IPPROTO_UDP) {
399 parse_udp(packet, &b, flow);
400 } else if (flow->nw_proto == IPPROTO_ICMP) {
401 const struct icmp_header *icmp = pull_icmp(&b);
403 flow->tp_src = htons(icmp->icmp_type);
404 flow->tp_dst = htons(icmp->icmp_code);
410 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
411 if (parse_ipv6(&b, flow)) {
416 if (flow->nw_proto == IPPROTO_TCP) {
417 parse_tcp(packet, &b, flow);
418 } else if (flow->nw_proto == IPPROTO_UDP) {
419 parse_udp(packet, &b, flow);
420 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
421 if (parse_icmpv6(&b, flow)) {
425 } else if (flow->dl_type == htons(ETH_TYPE_ARP)) {
426 const struct arp_eth_header *arp = pull_arp(&b);
427 if (arp && arp->ar_hrd == htons(1)
428 && arp->ar_pro == htons(ETH_TYPE_IP)
429 && arp->ar_hln == ETH_ADDR_LEN
430 && arp->ar_pln == 4) {
431 /* We only match on the lower 8 bits of the opcode. */
432 if (ntohs(arp->ar_op) <= 0xff) {
433 flow->nw_proto = ntohs(arp->ar_op);
436 if ((flow->nw_proto == ARP_OP_REQUEST)
437 || (flow->nw_proto == ARP_OP_REPLY)) {
438 flow->nw_src = arp->ar_spa;
439 flow->nw_dst = arp->ar_tpa;
440 memcpy(flow->arp_sha, arp->ar_sha, ETH_ADDR_LEN);
441 memcpy(flow->arp_tha, arp->ar_tha, ETH_ADDR_LEN);
447 /* For every bit of a field that is wildcarded in 'wildcards', sets the
448 * corresponding bit in 'flow' to zero. */
450 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
452 uint32_t *flow_u32 = (uint32_t *) flow;
453 const uint32_t *wc_u32 = (const uint32_t *) &wildcards->masks;
456 for (i = 0; i < FLOW_U32S; i++) {
457 flow_u32[i] &= wc_u32[i];
461 /* Initializes 'fmd' with the metadata found in 'flow'. */
463 flow_get_metadata(const struct flow *flow, struct flow_metadata *fmd)
465 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 17);
467 fmd->tun_id = flow->tunnel.tun_id;
468 fmd->metadata = flow->metadata;
469 memcpy(fmd->regs, flow->regs, sizeof fmd->regs);
470 fmd->in_port = flow->in_port;
474 flow_to_string(const struct flow *flow)
476 struct ds ds = DS_EMPTY_INITIALIZER;
477 flow_format(&ds, flow);
481 static void format_tunnel_flags(uint16_t flags, struct ds *ds)
483 flags &= ~FLOW_TNL_F_KEY;
485 if (flags & FLOW_TNL_F_DONT_FRAGMENT) {
486 ds_put_cstr(ds, ",df");
487 flags &= ~FLOW_TNL_F_DONT_FRAGMENT;
490 if (flags & FLOW_TNL_F_CSUM) {
491 ds_put_cstr(ds, ",csum");
492 flags &= ~FLOW_TNL_F_CSUM;
496 ds_put_format(ds, ",flags:%#"PRIx16, flags);
501 flow_format(struct ds *ds, const struct flow *flow)
503 ds_put_format(ds, "priority:%"PRIu32, flow->skb_priority);
505 if (flow->tunnel.ip_dst || flow->tunnel.tun_id) {
506 ds_put_cstr(ds, ",tunnel(");
507 ds_put_format(ds, IP_FMT"->"IP_FMT, IP_ARGS(&flow->tunnel.ip_src),
508 IP_ARGS(&flow->tunnel.ip_dst));
510 if (flow->tunnel.flags & FLOW_TNL_F_KEY) {
511 ds_put_format(ds, ",key:%#"PRIx64, ntohll(flow->tunnel.tun_id));
513 ds_put_format(ds, ",tos:%#"PRIx8",ttl:%"PRIu8, flow->tunnel.ip_tos,
514 flow->tunnel.ip_ttl);
515 format_tunnel_flags(flow->tunnel.flags, ds);
516 ds_put_char(ds, ')');
519 ds_put_format(ds, ",metadata:%#"PRIx64
520 ",in_port:%04"PRIx16,
521 ntohll(flow->metadata),
524 ds_put_format(ds, ",tci(");
525 if (flow->vlan_tci) {
526 ds_put_format(ds, "vlan:%"PRIu16",pcp:%d",
527 vlan_tci_to_vid(flow->vlan_tci),
528 vlan_tci_to_pcp(flow->vlan_tci));
530 ds_put_char(ds, '0');
532 ds_put_format(ds, ") mac("ETH_ADDR_FMT"->"ETH_ADDR_FMT
534 ETH_ADDR_ARGS(flow->dl_src),
535 ETH_ADDR_ARGS(flow->dl_dst),
536 ntohs(flow->dl_type));
538 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
539 ds_put_format(ds, " label:%#"PRIx32" proto:%"PRIu8" tos:%#"PRIx8
540 " ttl:%"PRIu8" ipv6(",
541 ntohl(flow->ipv6_label), flow->nw_proto,
542 flow->nw_tos, flow->nw_ttl);
543 print_ipv6_addr(ds, &flow->ipv6_src);
544 ds_put_cstr(ds, "->");
545 print_ipv6_addr(ds, &flow->ipv6_dst);
546 ds_put_char(ds, ')');
547 } else if (flow->dl_type == htons(ETH_TYPE_IP) ||
548 flow->dl_type == htons(ETH_TYPE_ARP)) {
549 ds_put_format(ds, " proto:%"PRIu8" tos:%#"PRIx8" ttl:%"PRIu8
550 " ip("IP_FMT"->"IP_FMT")",
551 flow->nw_proto, flow->nw_tos, flow->nw_ttl,
552 IP_ARGS(&flow->nw_src), IP_ARGS(&flow->nw_dst));
555 ds_put_format(ds, " frag(%s)",
556 flow->nw_frag == FLOW_NW_FRAG_ANY ? "first"
557 : flow->nw_frag == (FLOW_NW_FRAG_ANY | FLOW_NW_FRAG_LATER)
558 ? "later" : "<error>");
560 if (flow->tp_src || flow->tp_dst) {
561 ds_put_format(ds, " port(%"PRIu16"->%"PRIu16")",
562 ntohs(flow->tp_src), ntohs(flow->tp_dst));
564 if (!eth_addr_is_zero(flow->arp_sha) || !eth_addr_is_zero(flow->arp_tha)) {
565 ds_put_format(ds, " arp_ha("ETH_ADDR_FMT"->"ETH_ADDR_FMT")",
566 ETH_ADDR_ARGS(flow->arp_sha),
567 ETH_ADDR_ARGS(flow->arp_tha));
572 flow_print(FILE *stream, const struct flow *flow)
574 char *s = flow_to_string(flow);
579 /* flow_wildcards functions. */
581 /* Initializes 'wc' as a set of wildcards that matches every packet. */
583 flow_wildcards_init_catchall(struct flow_wildcards *wc)
585 memset(&wc->masks, 0, sizeof wc->masks);
588 /* Initializes 'wc' as an exact-match set of wildcards; that is, 'wc' does not
589 * wildcard any bits or fields. */
591 flow_wildcards_init_exact(struct flow_wildcards *wc)
593 memset(&wc->masks, 0xff, sizeof wc->masks);
594 memset(wc->masks.zeros, 0, sizeof wc->masks.zeros);
597 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
600 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
602 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
605 for (i = 0; i < FLOW_U32S; i++) {
613 /* Initializes 'dst' as the combination of wildcards in 'src1' and 'src2'.
614 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded in
615 * 'src1' or 'src2' or both. */
617 flow_wildcards_combine(struct flow_wildcards *dst,
618 const struct flow_wildcards *src1,
619 const struct flow_wildcards *src2)
621 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
622 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
623 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
626 for (i = 0; i < FLOW_U32S; i++) {
627 dst_u32[i] = src1_u32[i] & src2_u32[i];
631 /* Returns a hash of the wildcards in 'wc'. */
633 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
635 return flow_hash(&wc->masks, basis);;
638 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
641 flow_wildcards_equal(const struct flow_wildcards *a,
642 const struct flow_wildcards *b)
644 return flow_equal(&a->masks, &b->masks);
647 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
648 * 'b', false otherwise. */
650 flow_wildcards_has_extra(const struct flow_wildcards *a,
651 const struct flow_wildcards *b)
653 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
654 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
657 for (i = 0; i < FLOW_U32S; i++) {
658 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
665 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
666 * in 'wc' do not need to be equal in 'a' and 'b'. */
668 flow_equal_except(const struct flow *a, const struct flow *b,
669 const struct flow_wildcards *wc)
671 const uint32_t *a_u32 = (const uint32_t *) a;
672 const uint32_t *b_u32 = (const uint32_t *) b;
673 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
676 for (i = 0; i < FLOW_U32S; i++) {
677 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
684 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
685 * (A 0-bit indicates a wildcard bit.) */
687 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
689 wc->masks.regs[idx] = mask;
692 /* Hashes 'flow' based on its L2 through L4 protocol information. */
694 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
699 struct in6_addr ipv6_addr;
704 uint8_t eth_addr[ETH_ADDR_LEN];
710 memset(&fields, 0, sizeof fields);
711 for (i = 0; i < ETH_ADDR_LEN; i++) {
712 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
714 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
715 fields.eth_type = flow->dl_type;
717 /* UDP source and destination port are not taken into account because they
718 * will not necessarily be symmetric in a bidirectional flow. */
719 if (fields.eth_type == htons(ETH_TYPE_IP)) {
720 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
721 fields.ip_proto = flow->nw_proto;
722 if (fields.ip_proto == IPPROTO_TCP) {
723 fields.tp_port = flow->tp_src ^ flow->tp_dst;
725 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
726 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
727 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
728 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
730 for (i=0; i<16; i++) {
731 ipv6_addr[i] = a[i] ^ b[i];
733 fields.ip_proto = flow->nw_proto;
734 if (fields.ip_proto == IPPROTO_TCP) {
735 fields.tp_port = flow->tp_src ^ flow->tp_dst;
738 return hash_bytes(&fields, sizeof fields, basis);
741 /* Hashes the portions of 'flow' designated by 'fields'. */
743 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
748 case NX_HASH_FIELDS_ETH_SRC:
749 return hash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
751 case NX_HASH_FIELDS_SYMMETRIC_L4:
752 return flow_hash_symmetric_l4(flow, basis);
758 /* Returns a string representation of 'fields'. */
760 flow_hash_fields_to_str(enum nx_hash_fields fields)
763 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
764 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
765 default: return "<unknown>";
769 /* Returns true if the value of 'fields' is supported. Otherwise false. */
771 flow_hash_fields_valid(enum nx_hash_fields fields)
773 return fields == NX_HASH_FIELDS_ETH_SRC
774 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
777 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
778 * OpenFlow 1.0 "dl_vlan" value:
780 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
781 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
782 * 'flow' previously matched packets without a VLAN header).
784 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
785 * without a VLAN tag.
787 * - Other values of 'vid' should not be used. */
789 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
791 if (vid == htons(OFP10_VLAN_NONE)) {
792 flow->vlan_tci = htons(0);
794 vid &= htons(VLAN_VID_MASK);
795 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
796 flow->vlan_tci |= htons(VLAN_CFI) | vid;
800 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
801 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
804 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
806 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
807 flow->vlan_tci &= ~mask;
808 flow->vlan_tci |= vid & mask;
811 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
814 * This function has no effect on the VLAN ID that 'flow' matches.
816 * After calling this function, 'flow' will not match packets without a VLAN
819 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
822 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
823 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
826 /* Puts into 'b' a packet that flow_extract() would parse as having the given
829 * (This is useful only for testing, obviously, and the packet isn't really
830 * valid. It hasn't got some checksums filled in, for one, and lots of fields
831 * are just zeroed.) */
833 flow_compose(struct ofpbuf *b, const struct flow *flow)
835 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
836 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
837 struct eth_header *eth = b->l2;
838 eth->eth_type = htons(b->size);
842 if (flow->vlan_tci & htons(VLAN_CFI)) {
843 eth_push_vlan(b, flow->vlan_tci);
846 if (flow->dl_type == htons(ETH_TYPE_IP)) {
847 struct ip_header *ip;
849 b->l3 = ip = ofpbuf_put_zeros(b, sizeof *ip);
850 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
851 ip->ip_tos = flow->nw_tos;
852 ip->ip_proto = flow->nw_proto;
853 ip->ip_src = flow->nw_src;
854 ip->ip_dst = flow->nw_dst;
856 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
857 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
858 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
859 ip->ip_frag_off |= htons(100);
862 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
863 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
864 if (flow->nw_proto == IPPROTO_TCP) {
865 struct tcp_header *tcp;
867 b->l4 = tcp = ofpbuf_put_zeros(b, sizeof *tcp);
868 tcp->tcp_src = flow->tp_src;
869 tcp->tcp_dst = flow->tp_dst;
870 tcp->tcp_ctl = TCP_CTL(0, 5);
871 } else if (flow->nw_proto == IPPROTO_UDP) {
872 struct udp_header *udp;
874 b->l4 = udp = ofpbuf_put_zeros(b, sizeof *udp);
875 udp->udp_src = flow->tp_src;
876 udp->udp_dst = flow->tp_dst;
877 } else if (flow->nw_proto == IPPROTO_ICMP) {
878 struct icmp_header *icmp;
880 b->l4 = icmp = ofpbuf_put_zeros(b, sizeof *icmp);
881 icmp->icmp_type = ntohs(flow->tp_src);
882 icmp->icmp_code = ntohs(flow->tp_dst);
883 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
888 ip->ip_tot_len = htons((uint8_t *) b->data + b->size
889 - (uint8_t *) b->l3);
890 ip->ip_csum = csum(ip, sizeof *ip);
891 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
893 } else if (flow->dl_type == htons(ETH_TYPE_ARP)) {
894 struct arp_eth_header *arp;
896 b->l3 = arp = ofpbuf_put_zeros(b, sizeof *arp);
897 arp->ar_hrd = htons(1);
898 arp->ar_pro = htons(ETH_TYPE_IP);
899 arp->ar_hln = ETH_ADDR_LEN;
901 arp->ar_op = htons(flow->nw_proto);
903 if (flow->nw_proto == ARP_OP_REQUEST ||
904 flow->nw_proto == ARP_OP_REPLY) {
905 arp->ar_spa = flow->nw_src;
906 arp->ar_tpa = flow->nw_dst;
907 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
908 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
913 /* Compressed flow. */
916 miniflow_n_values(const struct miniflow *flow)
921 for (i = 0; i < MINI_N_MAPS; i++) {
922 n += popcount(flow->map[i]);
928 miniflow_alloc_values(struct miniflow *flow, int n)
930 if (n <= MINI_N_INLINE) {
931 return flow->inline_values;
933 COVERAGE_INC(miniflow_malloc);
934 return xmalloc(n * sizeof *flow->values);
938 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
939 * with miniflow_destroy(). */
941 miniflow_init(struct miniflow *dst, const struct flow *src)
943 const uint32_t *src_u32 = (const uint32_t *) src;
948 /* Initialize dst->map, counting the number of nonzero elements. */
950 memset(dst->map, 0, sizeof dst->map);
951 for (i = 0; i < FLOW_U32S; i++) {
953 dst->map[i / 32] |= 1u << (i % 32);
958 /* Initialize dst->values. */
959 dst->values = miniflow_alloc_values(dst, n);
961 for (i = 0; i < MINI_N_MAPS; i++) {
964 for (map = dst->map[i]; map; map = zero_rightmost_1bit(map)) {
965 dst->values[ofs++] = src_u32[raw_ctz(map) + i * 32];
970 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
971 * with miniflow_destroy(). */
973 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
975 int n = miniflow_n_values(src);
976 memcpy(dst->map, src->map, sizeof dst->map);
977 dst->values = miniflow_alloc_values(dst, n);
978 memcpy(dst->values, src->values, n * sizeof *dst->values);
981 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
982 * itself resides; the caller is responsible for that. */
984 miniflow_destroy(struct miniflow *flow)
986 if (flow->values != flow->inline_values) {
991 /* Initializes 'dst' as a copy of 'src'. */
993 miniflow_expand(const struct miniflow *src, struct flow *dst)
995 uint32_t *dst_u32 = (uint32_t *) dst;
999 memset(dst_u32, 0, sizeof *dst);
1002 for (i = 0; i < MINI_N_MAPS; i++) {
1005 for (map = src->map[i]; map; map = zero_rightmost_1bit(map)) {
1006 dst_u32[raw_ctz(map) + i * 32] = src->values[ofs++];
1011 static const uint32_t *
1012 miniflow_get__(const struct miniflow *flow, unsigned int u32_ofs)
1014 if (!(flow->map[u32_ofs / 32] & (1u << (u32_ofs % 32)))) {
1015 static const uint32_t zero = 0;
1018 const uint32_t *p = flow->values;
1020 BUILD_ASSERT(MINI_N_MAPS == 2);
1022 p += popcount(flow->map[0] & ((1u << u32_ofs) - 1));
1024 p += popcount(flow->map[0]);
1025 p += popcount(flow->map[1] & ((1u << (u32_ofs - 32)) - 1));
1031 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1032 * were expanded into a "struct flow". */
1034 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1036 return *miniflow_get__(flow, u32_ofs);
1039 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
1040 * expanded into a "struct flow". */
1042 miniflow_get_be16(const struct miniflow *flow, unsigned int u8_ofs)
1044 const uint32_t *u32p = miniflow_get__(flow, u8_ofs / 4);
1045 const ovs_be16 *be16p = (const ovs_be16 *) u32p;
1046 return be16p[u8_ofs % 4 != 0];
1049 /* Returns the VID within the vlan_tci member of the "struct flow" represented
1052 miniflow_get_vid(const struct miniflow *flow)
1054 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
1055 return vlan_tci_to_vid(tci);
1058 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1060 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1064 for (i = 0; i < MINI_N_MAPS; i++) {
1065 if (a->map[i] != b->map[i]) {
1070 return !memcmp(a->values, b->values,
1071 miniflow_n_values(a) * sizeof *a->values);
1074 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1075 * in 'mask', false if they differ. */
1077 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1078 const struct minimask *mask)
1083 p = mask->masks.values;
1084 for (i = 0; i < MINI_N_MAPS; i++) {
1087 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1088 int ofs = raw_ctz(map) + i * 32;
1090 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p) {
1100 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1101 * in 'mask', false if they differ. */
1103 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1104 const struct minimask *mask)
1106 const uint32_t *b_u32 = (const uint32_t *) b;
1110 p = mask->masks.values;
1111 for (i = 0; i < MINI_N_MAPS; i++) {
1114 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1115 int ofs = raw_ctz(map) + i * 32;
1117 if ((miniflow_get(a, ofs) ^ b_u32[ofs]) & *p) {
1127 /* Returns a hash value for 'flow', given 'basis'. */
1129 miniflow_hash(const struct miniflow *flow, uint32_t basis)
1131 BUILD_ASSERT_DECL(MINI_N_MAPS == 2);
1132 return hash_3words(flow->map[0], flow->map[1],
1133 hash_words(flow->values, miniflow_n_values(flow),
1137 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1138 * 'mask', given 'basis'.
1140 * The hash values returned by this function are the same as those returned by
1141 * flow_hash_in_minimask(), only the form of the arguments differ. */
1143 miniflow_hash_in_minimask(const struct miniflow *flow,
1144 const struct minimask *mask, uint32_t basis)
1146 const uint32_t *p = mask->masks.values;
1151 for (i = 0; i < MINI_N_MAPS; i++) {
1154 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1155 int ofs = raw_ctz(map) + i * 32;
1157 hash = mhash_add(hash, miniflow_get(flow, ofs) & *p);
1162 return mhash_finish(hash, p - mask->masks.values);
1165 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1166 * 'mask', given 'basis'.
1168 * The hash values returned by this function are the same as those returned by
1169 * miniflow_hash_in_minimask(), only the form of the arguments differ. */
1171 flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
1174 const uint32_t *flow_u32 = (const uint32_t *) flow;
1175 const uint32_t *p = mask->masks.values;
1180 for (i = 0; i < MINI_N_MAPS; i++) {
1183 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1184 int ofs = raw_ctz(map) + i * 32;
1186 hash = mhash_add(hash, flow_u32[ofs] & *p);
1191 return mhash_finish(hash, p - mask->masks.values);
1194 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1195 * with minimask_destroy(). */
1197 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1199 miniflow_init(&mask->masks, &wc->masks);
1202 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1203 * with minimask_destroy(). */
1205 minimask_clone(struct minimask *dst, const struct minimask *src)
1207 miniflow_clone(&dst->masks, &src->masks);
1210 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1212 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1213 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1215 minimask_combine(struct minimask *dst_,
1216 const struct minimask *a_, const struct minimask *b_,
1217 uint32_t storage[FLOW_U32S])
1219 struct miniflow *dst = &dst_->masks;
1220 const struct miniflow *a = &a_->masks;
1221 const struct miniflow *b = &b_->masks;
1225 dst->values = storage;
1226 for (i = 0; i < MINI_N_MAPS; i++) {
1230 for (map = a->map[i] & b->map[i]; map;
1231 map = zero_rightmost_1bit(map)) {
1232 int ofs = raw_ctz(map) + i * 32;
1233 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1236 dst->map[i] |= rightmost_1bit(map);
1237 dst->values[n++] = mask;
1243 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1244 * itself resides; the caller is responsible for that. */
1246 minimask_destroy(struct minimask *mask)
1248 miniflow_destroy(&mask->masks);
1251 /* Initializes 'dst' as a copy of 'src'. */
1253 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1255 miniflow_expand(&mask->masks, &wc->masks);
1258 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1259 * were expanded into a "struct flow_wildcards". */
1261 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1263 return miniflow_get(&mask->masks, u32_ofs);
1266 /* Returns the VID mask within the vlan_tci member of the "struct
1267 * flow_wildcards" represented by 'mask'. */
1269 minimask_get_vid_mask(const struct minimask *mask)
1271 return miniflow_get_vid(&mask->masks);
1274 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1276 minimask_equal(const struct minimask *a, const struct minimask *b)
1278 return miniflow_equal(&a->masks, &b->masks);
1281 /* Returns a hash value for 'mask', given 'basis'. */
1283 minimask_hash(const struct minimask *mask, uint32_t basis)
1285 return miniflow_hash(&mask->masks, basis);
1288 /* Returns true if at least one bit is wildcarded in 'a_' but not in 'b_',
1289 * false otherwise. */
1291 minimask_has_extra(const struct minimask *a_, const struct minimask *b_)
1293 const struct miniflow *a = &a_->masks;
1294 const struct miniflow *b = &b_->masks;
1297 for (i = 0; i < MINI_N_MAPS; i++) {
1300 for (map = a->map[i] | b->map[i]; map;
1301 map = zero_rightmost_1bit(map)) {
1302 int ofs = raw_ctz(map) + i * 32;
1303 uint32_t a_u32 = miniflow_get(a, ofs);
1304 uint32_t b_u32 = miniflow_get(b, ofs);
1306 if ((a_u32 & b_u32) != b_u32) {
1315 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
1318 minimask_is_catchall(const struct minimask *mask_)
1320 const struct miniflow *mask = &mask_->masks;
1322 BUILD_ASSERT(MINI_N_MAPS == 2);
1323 return !(mask->map[0] | mask->map[1]);