2 * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 #include <sys/types.h>
22 #include <netinet/in.h>
23 #include <netinet/icmp6.h>
24 #include <netinet/ip6.h>
28 #include "byte-order.h"
31 #include "dynamic-string.h"
36 #include "openflow/openflow.h"
38 #include "unaligned.h"
41 VLOG_DEFINE_THIS_MODULE(flow);
43 COVERAGE_DEFINE(flow_extract);
44 COVERAGE_DEFINE(miniflow_malloc);
46 static struct arp_eth_header *
47 pull_arp(struct ofpbuf *packet)
49 return ofpbuf_try_pull(packet, ARP_ETH_HEADER_LEN);
52 static struct ip_header *
53 pull_ip(struct ofpbuf *packet)
55 if (packet->size >= IP_HEADER_LEN) {
56 struct ip_header *ip = packet->data;
57 int ip_len = IP_IHL(ip->ip_ihl_ver) * 4;
58 if (ip_len >= IP_HEADER_LEN && packet->size >= ip_len) {
59 return ofpbuf_pull(packet, ip_len);
65 static struct tcp_header *
66 pull_tcp(struct ofpbuf *packet)
68 if (packet->size >= TCP_HEADER_LEN) {
69 struct tcp_header *tcp = packet->data;
70 int tcp_len = TCP_OFFSET(tcp->tcp_ctl) * 4;
71 if (tcp_len >= TCP_HEADER_LEN && packet->size >= tcp_len) {
72 return ofpbuf_pull(packet, tcp_len);
78 static struct udp_header *
79 pull_udp(struct ofpbuf *packet)
81 return ofpbuf_try_pull(packet, UDP_HEADER_LEN);
84 static struct icmp_header *
85 pull_icmp(struct ofpbuf *packet)
87 return ofpbuf_try_pull(packet, ICMP_HEADER_LEN);
90 static struct icmp6_hdr *
91 pull_icmpv6(struct ofpbuf *packet)
93 return ofpbuf_try_pull(packet, sizeof(struct icmp6_hdr));
97 parse_mpls(struct ofpbuf *b, struct flow *flow)
101 while ((mh = ofpbuf_try_pull(b, sizeof *mh))) {
102 if (flow->mpls_depth++ == 0) {
103 flow->mpls_lse = mh->mpls_lse;
105 if (mh->mpls_lse & htonl(MPLS_BOS_MASK)) {
112 parse_vlan(struct ofpbuf *b, struct flow *flow)
115 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
119 if (b->size >= sizeof(struct qtag_prefix) + sizeof(ovs_be16)) {
120 struct qtag_prefix *qp = ofpbuf_pull(b, sizeof *qp);
121 flow->vlan_tci = qp->tci | htons(VLAN_CFI);
126 parse_ethertype(struct ofpbuf *b)
128 struct llc_snap_header *llc;
131 proto = *(ovs_be16 *) ofpbuf_pull(b, sizeof proto);
132 if (ntohs(proto) >= ETH_TYPE_MIN) {
136 if (b->size < sizeof *llc) {
137 return htons(FLOW_DL_TYPE_NONE);
141 if (llc->llc.llc_dsap != LLC_DSAP_SNAP
142 || llc->llc.llc_ssap != LLC_SSAP_SNAP
143 || llc->llc.llc_cntl != LLC_CNTL_SNAP
144 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
145 sizeof llc->snap.snap_org)) {
146 return htons(FLOW_DL_TYPE_NONE);
149 ofpbuf_pull(b, sizeof *llc);
151 if (ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN) {
152 return llc->snap.snap_type;
155 return htons(FLOW_DL_TYPE_NONE);
159 parse_ipv6(struct ofpbuf *packet, struct flow *flow)
161 const struct ip6_hdr *nh;
165 nh = ofpbuf_try_pull(packet, sizeof *nh);
170 nexthdr = nh->ip6_nxt;
172 flow->ipv6_src = nh->ip6_src;
173 flow->ipv6_dst = nh->ip6_dst;
175 tc_flow = get_unaligned_be32(&nh->ip6_flow);
176 flow->nw_tos = ntohl(tc_flow) >> 20;
177 flow->ipv6_label = tc_flow & htonl(IPV6_LABEL_MASK);
178 flow->nw_ttl = nh->ip6_hlim;
179 flow->nw_proto = IPPROTO_NONE;
182 if ((nexthdr != IPPROTO_HOPOPTS)
183 && (nexthdr != IPPROTO_ROUTING)
184 && (nexthdr != IPPROTO_DSTOPTS)
185 && (nexthdr != IPPROTO_AH)
186 && (nexthdr != IPPROTO_FRAGMENT)) {
187 /* It's either a terminal header (e.g., TCP, UDP) or one we
188 * don't understand. In either case, we're done with the
189 * packet, so use it to fill in 'nw_proto'. */
193 /* We only verify that at least 8 bytes of the next header are
194 * available, but many of these headers are longer. Ensure that
195 * accesses within the extension header are within those first 8
196 * bytes. All extension headers are required to be at least 8
198 if (packet->size < 8) {
202 if ((nexthdr == IPPROTO_HOPOPTS)
203 || (nexthdr == IPPROTO_ROUTING)
204 || (nexthdr == IPPROTO_DSTOPTS)) {
205 /* These headers, while different, have the fields we care about
206 * in the same location and with the same interpretation. */
207 const struct ip6_ext *ext_hdr = packet->data;
208 nexthdr = ext_hdr->ip6e_nxt;
209 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 1) * 8)) {
212 } else if (nexthdr == IPPROTO_AH) {
213 /* A standard AH definition isn't available, but the fields
214 * we care about are in the same location as the generic
215 * option header--only the header length is calculated
217 const struct ip6_ext *ext_hdr = packet->data;
218 nexthdr = ext_hdr->ip6e_nxt;
219 if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 2) * 4)) {
222 } else if (nexthdr == IPPROTO_FRAGMENT) {
223 const struct ip6_frag *frag_hdr = packet->data;
225 nexthdr = frag_hdr->ip6f_nxt;
226 if (!ofpbuf_try_pull(packet, sizeof *frag_hdr)) {
230 /* We only process the first fragment. */
231 if (frag_hdr->ip6f_offlg != htons(0)) {
232 flow->nw_frag = FLOW_NW_FRAG_ANY;
233 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
234 flow->nw_frag |= FLOW_NW_FRAG_LATER;
235 nexthdr = IPPROTO_FRAGMENT;
242 flow->nw_proto = nexthdr;
247 parse_tcp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
249 const struct tcp_header *tcp = pull_tcp(b);
251 flow->tp_src = tcp->tcp_src;
252 flow->tp_dst = tcp->tcp_dst;
253 packet->l7 = b->data;
258 parse_udp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow)
260 const struct udp_header *udp = pull_udp(b);
262 flow->tp_src = udp->udp_src;
263 flow->tp_dst = udp->udp_dst;
264 packet->l7 = b->data;
269 parse_icmpv6(struct ofpbuf *b, struct flow *flow)
271 const struct icmp6_hdr *icmp = pull_icmpv6(b);
277 /* The ICMPv6 type and code fields use the 16-bit transport port
278 * fields, so we need to store them in 16-bit network byte order. */
279 flow->tp_src = htons(icmp->icmp6_type);
280 flow->tp_dst = htons(icmp->icmp6_code);
282 if (icmp->icmp6_code == 0 &&
283 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
284 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
285 const struct in6_addr *nd_target;
287 nd_target = ofpbuf_try_pull(b, sizeof *nd_target);
291 flow->nd_target = *nd_target;
293 while (b->size >= 8) {
294 /* The minimum size of an option is 8 bytes, which also is
295 * the size of Ethernet link-layer options. */
296 const struct nd_opt_hdr *nd_opt = b->data;
297 int opt_len = nd_opt->nd_opt_len * 8;
299 if (!opt_len || opt_len > b->size) {
303 /* Store the link layer address if the appropriate option is
304 * provided. It is considered an error if the same link
305 * layer option is specified twice. */
306 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
308 if (eth_addr_is_zero(flow->arp_sha)) {
309 memcpy(flow->arp_sha, nd_opt + 1, ETH_ADDR_LEN);
313 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
315 if (eth_addr_is_zero(flow->arp_tha)) {
316 memcpy(flow->arp_tha, nd_opt + 1, ETH_ADDR_LEN);
322 if (!ofpbuf_try_pull(b, opt_len)) {
331 memset(&flow->nd_target, 0, sizeof(flow->nd_target));
332 memset(flow->arp_sha, 0, sizeof(flow->arp_sha));
333 memset(flow->arp_tha, 0, sizeof(flow->arp_tha));
339 /* Initializes 'flow' members from 'packet', 'skb_priority', 'tnl', and
342 * Initializes 'packet' header pointers as follows:
344 * - packet->l2 to the start of the Ethernet header.
346 * - packet->l2_5 to the start of the MPLS shim header.
348 * - packet->l3 to just past the Ethernet header, or just past the
349 * vlan_header if one is present, to the first byte of the payload of the
352 * - packet->l4 to just past the IPv4 header, if one is present and has a
353 * correct length, and otherwise NULL.
355 * - packet->l7 to just past the TCP or UDP or ICMP header, if one is
356 * present and has a correct length, and otherwise NULL.
359 flow_extract(struct ofpbuf *packet, uint32_t skb_priority, uint32_t skb_mark,
360 const struct flow_tnl *tnl, uint16_t ofp_in_port,
363 struct ofpbuf b = *packet;
364 struct eth_header *eth;
366 COVERAGE_INC(flow_extract);
368 memset(flow, 0, sizeof *flow);
371 ovs_assert(tnl != &flow->tunnel);
374 flow->in_port = ofp_in_port;
375 flow->skb_priority = skb_priority;
376 flow->skb_mark = skb_mark;
384 if (b.size < sizeof *eth) {
390 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
391 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
393 /* dl_type, vlan_tci. */
394 ofpbuf_pull(&b, ETH_ADDR_LEN * 2);
395 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
396 parse_vlan(&b, flow);
398 flow->dl_type = parse_ethertype(&b);
400 /* Parse mpls, copy l3 ttl. */
401 if (eth_type_mpls(flow->dl_type)) {
402 packet->l2_5 = b.data;
403 parse_mpls(&b, flow);
408 if (flow->dl_type == htons(ETH_TYPE_IP)) {
409 const struct ip_header *nh = pull_ip(&b);
413 flow->nw_src = get_unaligned_be32(&nh->ip_src);
414 flow->nw_dst = get_unaligned_be32(&nh->ip_dst);
415 flow->nw_proto = nh->ip_proto;
417 flow->nw_tos = nh->ip_tos;
418 if (IP_IS_FRAGMENT(nh->ip_frag_off)) {
419 flow->nw_frag = FLOW_NW_FRAG_ANY;
420 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
421 flow->nw_frag |= FLOW_NW_FRAG_LATER;
424 flow->nw_ttl = nh->ip_ttl;
426 if (!(nh->ip_frag_off & htons(IP_FRAG_OFF_MASK))) {
427 if (flow->nw_proto == IPPROTO_TCP) {
428 parse_tcp(packet, &b, flow);
429 } else if (flow->nw_proto == IPPROTO_UDP) {
430 parse_udp(packet, &b, flow);
431 } else if (flow->nw_proto == IPPROTO_ICMP) {
432 const struct icmp_header *icmp = pull_icmp(&b);
434 flow->tp_src = htons(icmp->icmp_type);
435 flow->tp_dst = htons(icmp->icmp_code);
441 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
442 if (parse_ipv6(&b, flow)) {
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_ICMPV6) {
452 if (parse_icmpv6(&b, flow)) {
456 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
457 flow->dl_type == htons(ETH_TYPE_RARP)) {
458 const struct arp_eth_header *arp = pull_arp(&b);
459 if (arp && arp->ar_hrd == htons(1)
460 && arp->ar_pro == htons(ETH_TYPE_IP)
461 && arp->ar_hln == ETH_ADDR_LEN
462 && arp->ar_pln == 4) {
463 /* We only match on the lower 8 bits of the opcode. */
464 if (ntohs(arp->ar_op) <= 0xff) {
465 flow->nw_proto = ntohs(arp->ar_op);
468 flow->nw_src = arp->ar_spa;
469 flow->nw_dst = arp->ar_tpa;
470 memcpy(flow->arp_sha, arp->ar_sha, ETH_ADDR_LEN);
471 memcpy(flow->arp_tha, arp->ar_tha, ETH_ADDR_LEN);
476 /* For every bit of a field that is wildcarded in 'wildcards', sets the
477 * corresponding bit in 'flow' to zero. */
479 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
481 uint32_t *flow_u32 = (uint32_t *) flow;
482 const uint32_t *wc_u32 = (const uint32_t *) &wildcards->masks;
485 for (i = 0; i < FLOW_U32S; i++) {
486 flow_u32[i] &= wc_u32[i];
490 /* Initializes 'fmd' with the metadata found in 'flow'. */
492 flow_get_metadata(const struct flow *flow, struct flow_metadata *fmd)
494 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
496 fmd->tun_id = flow->tunnel.tun_id;
497 fmd->metadata = flow->metadata;
498 memcpy(fmd->regs, flow->regs, sizeof fmd->regs);
499 fmd->in_port = flow->in_port;
503 flow_to_string(const struct flow *flow)
505 struct ds ds = DS_EMPTY_INITIALIZER;
506 flow_format(&ds, flow);
511 flow_tun_flag_to_string(uint32_t flags)
514 case FLOW_TNL_F_DONT_FRAGMENT:
516 case FLOW_TNL_F_CSUM:
526 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
527 uint32_t flags, char del)
535 uint32_t bit = rightmost_1bit(flags);
538 s = bit_to_string(bit);
540 ds_put_format(ds, "%s%c", s, del);
549 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
555 flow_format(struct ds *ds, const struct flow *flow)
559 match_wc_init(&match, flow);
560 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
564 flow_print(FILE *stream, const struct flow *flow)
566 char *s = flow_to_string(flow);
571 /* flow_wildcards functions. */
573 /* Initializes 'wc' as a set of wildcards that matches every packet. */
575 flow_wildcards_init_catchall(struct flow_wildcards *wc)
577 memset(&wc->masks, 0, sizeof wc->masks);
580 /* Initializes 'wc' as an exact-match set of wildcards; that is, 'wc' does not
581 * wildcard any bits or fields. */
583 flow_wildcards_init_exact(struct flow_wildcards *wc)
585 memset(&wc->masks, 0xff, sizeof wc->masks);
586 memset(wc->masks.zeros, 0, sizeof wc->masks.zeros);
589 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
592 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
594 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
597 for (i = 0; i < FLOW_U32S; i++) {
605 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
606 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
607 * in 'src1' or 'src2' or both. */
609 flow_wildcards_and(struct flow_wildcards *dst,
610 const struct flow_wildcards *src1,
611 const struct flow_wildcards *src2)
613 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
614 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
615 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
618 for (i = 0; i < FLOW_U32S; i++) {
619 dst_u32[i] = src1_u32[i] & src2_u32[i];
623 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
624 * is, a bit or a field is wildcarded in 'dst' if it is neither
625 * wildcarded in 'src1' nor 'src2'. */
627 flow_wildcards_or(struct flow_wildcards *dst,
628 const struct flow_wildcards *src1,
629 const struct flow_wildcards *src2)
631 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
632 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
633 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
636 for (i = 0; i < FLOW_U32S; i++) {
637 dst_u32[i] = src1_u32[i] | src2_u32[i];
641 /* Perform a bitwise OR of miniflow 'src' flow data with the equivalent
642 * fields in 'dst', storing the result in 'dst'. */
644 flow_union_with_miniflow(struct flow *dst, const struct miniflow *src)
646 uint32_t *dst_u32 = (uint32_t *) dst;
651 for (i = 0; i < MINI_N_MAPS; i++) {
654 for (map = src->map[i]; map; map = zero_rightmost_1bit(map)) {
655 dst_u32[raw_ctz(map) + i * 32] |= src->values[ofs++];
660 /* Fold minimask 'mask''s wildcard mask into 'wc's wildcard mask. */
662 flow_wildcards_fold_minimask(struct flow_wildcards *wc,
663 const struct minimask *mask)
665 flow_union_with_miniflow(&wc->masks, &mask->masks);
668 /* Returns a hash of the wildcards in 'wc'. */
670 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
672 return flow_hash(&wc->masks, basis);
675 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
678 flow_wildcards_equal(const struct flow_wildcards *a,
679 const struct flow_wildcards *b)
681 return flow_equal(&a->masks, &b->masks);
684 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
685 * 'b', false otherwise. */
687 flow_wildcards_has_extra(const struct flow_wildcards *a,
688 const struct flow_wildcards *b)
690 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
691 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
694 for (i = 0; i < FLOW_U32S; i++) {
695 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
702 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
703 * in 'wc' do not need to be equal in 'a' and 'b'. */
705 flow_equal_except(const struct flow *a, const struct flow *b,
706 const struct flow_wildcards *wc)
708 const uint32_t *a_u32 = (const uint32_t *) a;
709 const uint32_t *b_u32 = (const uint32_t *) b;
710 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
713 for (i = 0; i < FLOW_U32S; i++) {
714 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
721 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
722 * (A 0-bit indicates a wildcard bit.) */
724 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
726 wc->masks.regs[idx] = mask;
729 /* Hashes 'flow' based on its L2 through L4 protocol information. */
731 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
736 struct in6_addr ipv6_addr;
741 uint8_t eth_addr[ETH_ADDR_LEN];
747 memset(&fields, 0, sizeof fields);
748 for (i = 0; i < ETH_ADDR_LEN; i++) {
749 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
751 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
752 fields.eth_type = flow->dl_type;
754 /* UDP source and destination port are not taken into account because they
755 * will not necessarily be symmetric in a bidirectional flow. */
756 if (fields.eth_type == htons(ETH_TYPE_IP)) {
757 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
758 fields.ip_proto = flow->nw_proto;
759 if (fields.ip_proto == IPPROTO_TCP) {
760 fields.tp_port = flow->tp_src ^ flow->tp_dst;
762 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
763 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
764 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
765 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
767 for (i=0; i<16; i++) {
768 ipv6_addr[i] = a[i] ^ b[i];
770 fields.ip_proto = flow->nw_proto;
771 if (fields.ip_proto == IPPROTO_TCP) {
772 fields.tp_port = flow->tp_src ^ flow->tp_dst;
775 return jhash_bytes(&fields, sizeof fields, basis);
778 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
780 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
781 enum nx_hash_fields fields)
784 case NX_HASH_FIELDS_ETH_SRC:
785 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
788 case NX_HASH_FIELDS_SYMMETRIC_L4:
789 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
790 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
791 if (flow->dl_type == htons(ETH_TYPE_IP)) {
792 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
793 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
794 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
795 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
796 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
798 if (is_ip_any(flow)) {
799 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
800 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
801 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
803 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
811 /* Hashes the portions of 'flow' designated by 'fields'. */
813 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
818 case NX_HASH_FIELDS_ETH_SRC:
819 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
821 case NX_HASH_FIELDS_SYMMETRIC_L4:
822 return flow_hash_symmetric_l4(flow, basis);
828 /* Returns a string representation of 'fields'. */
830 flow_hash_fields_to_str(enum nx_hash_fields fields)
833 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
834 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
835 default: return "<unknown>";
839 /* Returns true if the value of 'fields' is supported. Otherwise false. */
841 flow_hash_fields_valid(enum nx_hash_fields fields)
843 return fields == NX_HASH_FIELDS_ETH_SRC
844 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
847 /* Returns a hash value for the bits of 'flow' that are active based on
848 * 'wc', given 'basis'. */
850 flow_hash_in_wildcards(const struct flow *flow,
851 const struct flow_wildcards *wc, uint32_t basis)
853 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
854 const uint32_t *flow_u32 = (const uint32_t *) flow;
859 for (i = 0; i < FLOW_U32S; i++) {
860 hash = mhash_add(hash, flow_u32[i] & wc_u32[i]);
862 return mhash_finish(hash, 4 * FLOW_U32S);
865 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
866 * OpenFlow 1.0 "dl_vlan" value:
868 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
869 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
870 * 'flow' previously matched packets without a VLAN header).
872 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
873 * without a VLAN tag.
875 * - Other values of 'vid' should not be used. */
877 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
879 if (vid == htons(OFP10_VLAN_NONE)) {
880 flow->vlan_tci = htons(0);
882 vid &= htons(VLAN_VID_MASK);
883 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
884 flow->vlan_tci |= htons(VLAN_CFI) | vid;
888 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
889 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
892 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
894 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
895 flow->vlan_tci &= ~mask;
896 flow->vlan_tci |= vid & mask;
899 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
902 * This function has no effect on the VLAN ID that 'flow' matches.
904 * After calling this function, 'flow' will not match packets without a VLAN
907 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
910 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
911 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
914 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
915 * as an OpenFlow 1.1 "mpls_label" value. */
917 flow_set_mpls_label(struct flow *flow, ovs_be32 label)
919 set_mpls_lse_label(&flow->mpls_lse, label);
922 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
925 flow_set_mpls_ttl(struct flow *flow, uint8_t ttl)
927 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
930 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
933 flow_set_mpls_tc(struct flow *flow, uint8_t tc)
935 set_mpls_lse_tc(&flow->mpls_lse, tc);
938 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
940 flow_set_mpls_bos(struct flow *flow, uint8_t bos)
942 set_mpls_lse_bos(&flow->mpls_lse, bos);
945 /* Puts into 'b' a packet that flow_extract() would parse as having the given
948 * (This is useful only for testing, obviously, and the packet isn't really
949 * valid. It hasn't got some checksums filled in, for one, and lots of fields
950 * are just zeroed.) */
952 flow_compose(struct ofpbuf *b, const struct flow *flow)
954 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
955 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
956 struct eth_header *eth = b->l2;
957 eth->eth_type = htons(b->size);
961 if (flow->vlan_tci & htons(VLAN_CFI)) {
962 eth_push_vlan(b, flow->vlan_tci);
965 if (flow->dl_type == htons(ETH_TYPE_IP)) {
966 struct ip_header *ip;
968 b->l3 = ip = ofpbuf_put_zeros(b, sizeof *ip);
969 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
970 ip->ip_tos = flow->nw_tos;
971 ip->ip_ttl = flow->nw_ttl;
972 ip->ip_proto = flow->nw_proto;
973 ip->ip_src = flow->nw_src;
974 ip->ip_dst = flow->nw_dst;
976 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
977 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
978 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
979 ip->ip_frag_off |= htons(100);
982 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
983 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
984 if (flow->nw_proto == IPPROTO_TCP) {
985 struct tcp_header *tcp;
987 b->l4 = tcp = ofpbuf_put_zeros(b, sizeof *tcp);
988 tcp->tcp_src = flow->tp_src;
989 tcp->tcp_dst = flow->tp_dst;
990 tcp->tcp_ctl = TCP_CTL(0, 5);
991 } else if (flow->nw_proto == IPPROTO_UDP) {
992 struct udp_header *udp;
994 b->l4 = udp = ofpbuf_put_zeros(b, sizeof *udp);
995 udp->udp_src = flow->tp_src;
996 udp->udp_dst = flow->tp_dst;
997 } else if (flow->nw_proto == IPPROTO_ICMP) {
998 struct icmp_header *icmp;
1000 b->l4 = icmp = ofpbuf_put_zeros(b, sizeof *icmp);
1001 icmp->icmp_type = ntohs(flow->tp_src);
1002 icmp->icmp_code = ntohs(flow->tp_dst);
1003 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
1008 ip->ip_tot_len = htons((uint8_t *) b->data + b->size
1009 - (uint8_t *) b->l3);
1010 ip->ip_csum = csum(ip, sizeof *ip);
1011 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1013 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1014 flow->dl_type == htons(ETH_TYPE_RARP)) {
1015 struct arp_eth_header *arp;
1017 b->l3 = arp = ofpbuf_put_zeros(b, sizeof *arp);
1018 arp->ar_hrd = htons(1);
1019 arp->ar_pro = htons(ETH_TYPE_IP);
1020 arp->ar_hln = ETH_ADDR_LEN;
1022 arp->ar_op = htons(flow->nw_proto);
1024 if (flow->nw_proto == ARP_OP_REQUEST ||
1025 flow->nw_proto == ARP_OP_REPLY) {
1026 arp->ar_spa = flow->nw_src;
1027 arp->ar_tpa = flow->nw_dst;
1028 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
1029 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
1033 if (eth_type_mpls(flow->dl_type)) {
1035 push_mpls(b, flow->dl_type, flow->mpls_lse);
1039 /* Compressed flow. */
1042 miniflow_n_values(const struct miniflow *flow)
1047 for (i = 0; i < MINI_N_MAPS; i++) {
1048 n += popcount(flow->map[i]);
1054 miniflow_alloc_values(struct miniflow *flow, int n)
1056 if (n <= MINI_N_INLINE) {
1057 return flow->inline_values;
1059 COVERAGE_INC(miniflow_malloc);
1060 return xmalloc(n * sizeof *flow->values);
1064 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1065 * with miniflow_destroy(). */
1067 miniflow_init(struct miniflow *dst, const struct flow *src)
1069 const uint32_t *src_u32 = (const uint32_t *) src;
1074 /* Initialize dst->map, counting the number of nonzero elements. */
1076 memset(dst->map, 0, sizeof dst->map);
1077 for (i = 0; i < FLOW_U32S; i++) {
1079 dst->map[i / 32] |= 1u << (i % 32);
1084 /* Initialize dst->values. */
1085 dst->values = miniflow_alloc_values(dst, n);
1087 for (i = 0; i < MINI_N_MAPS; i++) {
1090 for (map = dst->map[i]; map; map = zero_rightmost_1bit(map)) {
1091 dst->values[ofs++] = src_u32[raw_ctz(map) + i * 32];
1096 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1097 * with miniflow_destroy(). */
1099 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
1101 int n = miniflow_n_values(src);
1102 memcpy(dst->map, src->map, sizeof dst->map);
1103 dst->values = miniflow_alloc_values(dst, n);
1104 memcpy(dst->values, src->values, n * sizeof *dst->values);
1107 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
1108 * itself resides; the caller is responsible for that. */
1110 miniflow_destroy(struct miniflow *flow)
1112 if (flow->values != flow->inline_values) {
1117 /* Initializes 'dst' as a copy of 'src'. */
1119 miniflow_expand(const struct miniflow *src, struct flow *dst)
1121 memset(dst, 0, sizeof *dst);
1122 flow_union_with_miniflow(dst, src);
1125 static const uint32_t *
1126 miniflow_get__(const struct miniflow *flow, unsigned int u32_ofs)
1128 if (!(flow->map[u32_ofs / 32] & (1u << (u32_ofs % 32)))) {
1129 static const uint32_t zero = 0;
1132 const uint32_t *p = flow->values;
1134 BUILD_ASSERT(MINI_N_MAPS == 2);
1136 p += popcount(flow->map[0] & ((1u << u32_ofs) - 1));
1138 p += popcount(flow->map[0]);
1139 p += popcount(flow->map[1] & ((1u << (u32_ofs - 32)) - 1));
1145 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1146 * were expanded into a "struct flow". */
1148 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1150 return *miniflow_get__(flow, u32_ofs);
1153 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
1154 * expanded into a "struct flow". */
1156 miniflow_get_be16(const struct miniflow *flow, unsigned int u8_ofs)
1158 const uint32_t *u32p = miniflow_get__(flow, u8_ofs / 4);
1159 const ovs_be16 *be16p = (const ovs_be16 *) u32p;
1160 return be16p[u8_ofs % 4 != 0];
1163 /* Returns the VID within the vlan_tci member of the "struct flow" represented
1166 miniflow_get_vid(const struct miniflow *flow)
1168 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
1169 return vlan_tci_to_vid(tci);
1172 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1174 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1178 for (i = 0; i < MINI_N_MAPS; i++) {
1179 if (a->map[i] != b->map[i]) {
1184 return !memcmp(a->values, b->values,
1185 miniflow_n_values(a) * sizeof *a->values);
1188 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1189 * in 'mask', false if they differ. */
1191 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1192 const struct minimask *mask)
1197 p = mask->masks.values;
1198 for (i = 0; i < MINI_N_MAPS; i++) {
1201 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1202 int ofs = raw_ctz(map) + i * 32;
1204 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p) {
1214 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1215 * in 'mask', false if they differ. */
1217 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1218 const struct minimask *mask)
1220 const uint32_t *b_u32 = (const uint32_t *) b;
1224 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 if ((miniflow_get(a, ofs) ^ b_u32[ofs]) & *p) {
1241 /* Returns a hash value for 'flow', given 'basis'. */
1243 miniflow_hash(const struct miniflow *flow, uint32_t basis)
1245 BUILD_ASSERT_DECL(MINI_N_MAPS == 2);
1246 return hash_3words(flow->map[0], flow->map[1],
1247 hash_words(flow->values, miniflow_n_values(flow),
1251 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1252 * 'mask', given 'basis'.
1254 * The hash values returned by this function are the same as those returned by
1255 * flow_hash_in_minimask(), only the form of the arguments differ. */
1257 miniflow_hash_in_minimask(const struct miniflow *flow,
1258 const struct minimask *mask, uint32_t basis)
1260 const uint32_t *p = mask->masks.values;
1265 for (i = 0; i < MINI_N_MAPS; i++) {
1268 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1269 int ofs = raw_ctz(map) + i * 32;
1271 hash = mhash_add(hash, miniflow_get(flow, ofs) & *p);
1276 return mhash_finish(hash, (p - mask->masks.values) * 4);
1279 /* Returns a hash value for the bits of 'flow' where there are 1-bits in
1280 * 'mask', given 'basis'.
1282 * The hash values returned by this function are the same as those returned by
1283 * miniflow_hash_in_minimask(), only the form of the arguments differ. */
1285 flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
1288 const uint32_t *flow_u32 = (const uint32_t *) flow;
1289 const uint32_t *p = mask->masks.values;
1294 for (i = 0; i < MINI_N_MAPS; i++) {
1297 for (map = mask->masks.map[i]; map; map = zero_rightmost_1bit(map)) {
1298 int ofs = raw_ctz(map) + i * 32;
1300 hash = mhash_add(hash, flow_u32[ofs] & *p);
1305 return mhash_finish(hash, (p - mask->masks.values) * 4);
1308 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1309 * with minimask_destroy(). */
1311 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1313 miniflow_init(&mask->masks, &wc->masks);
1316 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1317 * with minimask_destroy(). */
1319 minimask_clone(struct minimask *dst, const struct minimask *src)
1321 miniflow_clone(&dst->masks, &src->masks);
1324 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1326 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1327 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1329 minimask_combine(struct minimask *dst_,
1330 const struct minimask *a_, const struct minimask *b_,
1331 uint32_t storage[FLOW_U32S])
1333 struct miniflow *dst = &dst_->masks;
1334 const struct miniflow *a = &a_->masks;
1335 const struct miniflow *b = &b_->masks;
1339 dst->values = storage;
1340 for (i = 0; i < MINI_N_MAPS; i++) {
1344 for (map = a->map[i] & b->map[i]; map;
1345 map = zero_rightmost_1bit(map)) {
1346 int ofs = raw_ctz(map) + i * 32;
1347 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1350 dst->map[i] |= rightmost_1bit(map);
1351 dst->values[n++] = mask;
1357 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1358 * itself resides; the caller is responsible for that. */
1360 minimask_destroy(struct minimask *mask)
1362 miniflow_destroy(&mask->masks);
1365 /* Initializes 'dst' as a copy of 'src'. */
1367 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1369 miniflow_expand(&mask->masks, &wc->masks);
1372 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1373 * were expanded into a "struct flow_wildcards". */
1375 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1377 return miniflow_get(&mask->masks, u32_ofs);
1380 /* Returns the VID mask within the vlan_tci member of the "struct
1381 * flow_wildcards" represented by 'mask'. */
1383 minimask_get_vid_mask(const struct minimask *mask)
1385 return miniflow_get_vid(&mask->masks);
1388 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1390 minimask_equal(const struct minimask *a, const struct minimask *b)
1392 return miniflow_equal(&a->masks, &b->masks);
1395 /* Returns a hash value for 'mask', given 'basis'. */
1397 minimask_hash(const struct minimask *mask, uint32_t basis)
1399 return miniflow_hash(&mask->masks, basis);
1402 /* Returns true if at least one bit is wildcarded in 'a_' but not in 'b_',
1403 * false otherwise. */
1405 minimask_has_extra(const struct minimask *a_, const struct minimask *b_)
1407 const struct miniflow *a = &a_->masks;
1408 const struct miniflow *b = &b_->masks;
1411 for (i = 0; i < MINI_N_MAPS; i++) {
1414 for (map = a->map[i] | b->map[i]; map;
1415 map = zero_rightmost_1bit(map)) {
1416 int ofs = raw_ctz(map) + i * 32;
1417 uint32_t a_u32 = miniflow_get(a, ofs);
1418 uint32_t b_u32 = miniflow_get(b, ofs);
1420 if ((a_u32 & b_u32) != b_u32) {
1429 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
1432 minimask_is_catchall(const struct minimask *mask_)
1434 const struct miniflow *mask = &mask_->masks;
1436 BUILD_ASSERT(MINI_N_MAPS == 2);
1437 return !(mask->map[0] | mask->map[1]);