2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 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"
40 #include "unaligned.h"
42 COVERAGE_DEFINE(flow_extract);
43 COVERAGE_DEFINE(miniflow_malloc);
45 /* U32 indices for segmented flow classification. */
46 const uint8_t flow_segment_u32s[4] = {
47 FLOW_SEGMENT_1_ENDS_AT / 4,
48 FLOW_SEGMENT_2_ENDS_AT / 4,
49 FLOW_SEGMENT_3_ENDS_AT / 4,
53 /* miniflow_extract() assumes the following to be true to optimize the
54 * extraction process. */
55 BUILD_ASSERT_DECL(offsetof(struct flow, dl_type) + 2
56 == offsetof(struct flow, vlan_tci) &&
57 offsetof(struct flow, dl_type) / 4
58 == offsetof(struct flow, vlan_tci) / 4 );
60 BUILD_ASSERT_DECL(offsetof(struct flow, nw_frag) + 3
61 == offsetof(struct flow, nw_proto) &&
62 offsetof(struct flow, nw_tos) + 2
63 == offsetof(struct flow, nw_proto) &&
64 offsetof(struct flow, nw_ttl) + 1
65 == offsetof(struct flow, nw_proto) &&
66 offsetof(struct flow, nw_frag) / 4
67 == offsetof(struct flow, nw_tos) / 4 &&
68 offsetof(struct flow, nw_ttl) / 4
69 == offsetof(struct flow, nw_tos) / 4 &&
70 offsetof(struct flow, nw_proto) / 4
71 == offsetof(struct flow, nw_tos) / 4);
73 /* TCP flags in the first half of a BE32, zeroes in the other half. */
74 BUILD_ASSERT_DECL(offsetof(struct flow, tcp_flags) + 2
75 == offsetof(struct flow, pad) &&
76 offsetof(struct flow, tcp_flags) / 4
77 == offsetof(struct flow, pad) / 4);
79 #define TCP_FLAGS_BE32(tcp_ctl) ((OVS_FORCE ovs_be32)TCP_FLAGS_BE16(tcp_ctl) \
82 #define TCP_FLAGS_BE32(tcp_ctl) ((OVS_FORCE ovs_be32)TCP_FLAGS_BE16(tcp_ctl))
85 BUILD_ASSERT_DECL(offsetof(struct flow, tp_src) + 2
86 == offsetof(struct flow, tp_dst) &&
87 offsetof(struct flow, tp_src) / 4
88 == offsetof(struct flow, tp_dst) / 4);
90 /* Removes 'size' bytes from the head end of '*datap', of size '*sizep', which
91 * must contain at least 'size' bytes of data. Returns the first byte of data
93 static inline const void *
94 data_pull(void **datap, size_t *sizep, size_t size)
96 char *data = (char *)*datap;
102 /* If '*datap' has at least 'size' bytes of data, removes that many bytes from
103 * the head end of '*datap' and returns the first byte removed. Otherwise,
104 * returns a null pointer without modifying '*datap'. */
105 static inline const void *
106 data_try_pull(void **datap, size_t *sizep, size_t size)
108 return OVS_LIKELY(*sizep >= size) ? data_pull(datap, sizep, size) : NULL;
111 /* Context for pushing data to a miniflow. */
115 uint32_t * const end;
118 /* miniflow_push_* macros allow filling in a miniflow data values in order.
119 * Assertions are needed only when the layout of the struct flow is modified.
120 * 'ofs' is a compile-time constant, which allows most of the code be optimized
121 * away. Some GCC versions gave warnings on ALWAYS_INLINE, so these are
122 * defined as macros. */
124 #if (FLOW_WC_SEQ != 27)
125 #define MINIFLOW_ASSERT(X) ovs_assert(X)
126 BUILD_MESSAGE("FLOW_WC_SEQ changed: miniflow_extract() will have runtime "
127 "assertions enabled. Consider updating FLOW_WC_SEQ after "
130 #define MINIFLOW_ASSERT(X)
133 #define miniflow_push_uint32_(MF, OFS, VALUE) \
135 MINIFLOW_ASSERT(MF.data < MF.end && (OFS) % 4 == 0 \
136 && !(MF.map & (UINT64_MAX << (OFS) / 4))); \
137 *MF.data++ = VALUE; \
138 MF.map |= UINT64_C(1) << (OFS) / 4; \
141 #define miniflow_push_be32_(MF, OFS, VALUE) \
142 miniflow_push_uint32_(MF, OFS, (OVS_FORCE uint32_t)(VALUE))
144 #define miniflow_push_uint16_(MF, OFS, VALUE) \
146 MINIFLOW_ASSERT(MF.data < MF.end && \
147 (((OFS) % 4 == 0 && !(MF.map & (UINT64_MAX << (OFS) / 4))) \
148 || ((OFS) % 4 == 2 && MF.map & (UINT64_C(1) << (OFS) / 4) \
149 && !(MF.map & (UINT64_MAX << ((OFS) / 4 + 1)))))); \
151 if ((OFS) % 4 == 0) { \
152 *(uint16_t *)MF.data = VALUE; \
153 MF.map |= UINT64_C(1) << (OFS) / 4; \
154 } else if ((OFS) % 4 == 2) { \
155 *((uint16_t *)MF.data + 1) = VALUE; \
160 #define miniflow_push_be16_(MF, OFS, VALUE) \
161 miniflow_push_uint16_(MF, OFS, (OVS_FORCE uint16_t)VALUE);
163 /* Data at 'valuep' may be unaligned. */
164 #define miniflow_push_words_(MF, OFS, VALUEP, N_WORDS) \
166 int ofs32 = (OFS) / 4; \
168 MINIFLOW_ASSERT(MF.data + (N_WORDS) <= MF.end && (OFS) % 4 == 0 \
169 && !(MF.map & (UINT64_MAX << ofs32))); \
171 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof *MF.data); \
172 MF.data += (N_WORDS); \
173 MF.map |= ((UINT64_MAX >> (64 - (N_WORDS))) << ofs32); \
176 #define miniflow_push_uint32(MF, FIELD, VALUE) \
177 miniflow_push_uint32_(MF, offsetof(struct flow, FIELD), VALUE)
179 #define miniflow_push_be32(MF, FIELD, VALUE) \
180 miniflow_push_be32_(MF, offsetof(struct flow, FIELD), VALUE)
182 #define miniflow_push_uint32_check(MF, FIELD, VALUE) \
183 { if (OVS_LIKELY(VALUE)) { \
184 miniflow_push_uint32_(MF, offsetof(struct flow, FIELD), VALUE); \
188 #define miniflow_push_be32_check(MF, FIELD, VALUE) \
189 { if (OVS_LIKELY(VALUE)) { \
190 miniflow_push_be32_(MF, offsetof(struct flow, FIELD), VALUE); \
194 #define miniflow_push_uint16(MF, FIELD, VALUE) \
195 miniflow_push_uint16_(MF, offsetof(struct flow, FIELD), VALUE)
197 #define miniflow_push_be16(MF, FIELD, VALUE) \
198 miniflow_push_be16_(MF, offsetof(struct flow, FIELD), VALUE)
200 #define miniflow_push_words(MF, FIELD, VALUEP, N_WORDS) \
201 miniflow_push_words_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
203 /* Pulls the MPLS headers at '*datap' and returns the count of them. */
205 parse_mpls(void **datap, size_t *sizep)
207 const struct mpls_hdr *mh;
210 while ((mh = data_try_pull(datap, sizep, sizeof *mh))) {
212 if (mh->mpls_lse.lo & htons(1 << MPLS_BOS_SHIFT)) {
216 return MIN(count, FLOW_MAX_MPLS_LABELS);
219 static inline ovs_be16
220 parse_vlan(void **datap, size_t *sizep)
222 const struct eth_header *eth = *datap;
225 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
229 data_pull(datap, sizep, ETH_ADDR_LEN * 2);
231 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
232 if (OVS_LIKELY(*sizep
233 >= sizeof(struct qtag_prefix) + sizeof(ovs_be16))) {
234 const struct qtag_prefix *qp = data_pull(datap, sizep, sizeof *qp);
235 return qp->tci | htons(VLAN_CFI);
241 static inline ovs_be16
242 parse_ethertype(void **datap, size_t *sizep)
244 const struct llc_snap_header *llc;
247 proto = *(ovs_be16 *) data_pull(datap, sizep, sizeof proto);
248 if (OVS_LIKELY(ntohs(proto) >= ETH_TYPE_MIN)) {
252 if (OVS_UNLIKELY(*sizep < sizeof *llc)) {
253 return htons(FLOW_DL_TYPE_NONE);
257 if (OVS_UNLIKELY(llc->llc.llc_dsap != LLC_DSAP_SNAP
258 || llc->llc.llc_ssap != LLC_SSAP_SNAP
259 || llc->llc.llc_cntl != LLC_CNTL_SNAP
260 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
261 sizeof llc->snap.snap_org))) {
262 return htons(FLOW_DL_TYPE_NONE);
265 data_pull(datap, sizep, sizeof *llc);
267 if (OVS_LIKELY(ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN)) {
268 return llc->snap.snap_type;
271 return htons(FLOW_DL_TYPE_NONE);
275 parse_icmpv6(void **datap, size_t *sizep, const struct icmp6_hdr *icmp,
276 const struct in6_addr **nd_target,
277 uint8_t arp_buf[2][ETH_ADDR_LEN])
279 if (icmp->icmp6_code == 0 &&
280 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
281 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
283 *nd_target = data_try_pull(datap, sizep, sizeof **nd_target);
284 if (OVS_UNLIKELY(!*nd_target)) {
288 while (*sizep >= 8) {
289 /* The minimum size of an option is 8 bytes, which also is
290 * the size of Ethernet link-layer options. */
291 const struct nd_opt_hdr *nd_opt = *datap;
292 int opt_len = nd_opt->nd_opt_len * 8;
294 if (!opt_len || opt_len > *sizep) {
298 /* Store the link layer address if the appropriate option is
299 * provided. It is considered an error if the same link
300 * layer option is specified twice. */
301 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
303 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[0]))) {
304 memcpy(arp_buf[0], nd_opt + 1, ETH_ADDR_LEN);
308 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
310 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[1]))) {
311 memcpy(arp_buf[1], nd_opt + 1, ETH_ADDR_LEN);
317 if (OVS_UNLIKELY(!data_try_pull(datap, sizep, opt_len))) {
329 /* Initializes 'flow' members from 'packet' and 'md'
331 * Initializes 'packet' header l2 pointer to the start of the Ethernet
332 * header, and the layer offsets as follows:
334 * - packet->l2_5_ofs to the start of the MPLS shim header, or UINT16_MAX
335 * when there is no MPLS shim header.
337 * - packet->l3_ofs to just past the Ethernet header, or just past the
338 * vlan_header if one is present, to the first byte of the payload of the
339 * Ethernet frame. UINT16_MAX if the frame is too short to contain an
342 * - packet->l4_ofs to just past the IPv4 header, if one is present and
343 * has at least the content used for the fields of interest for the flow,
344 * otherwise UINT16_MAX.
347 flow_extract(struct ofpbuf *packet, const struct pkt_metadata *md,
352 uint32_t buf[FLOW_U32S];
355 COVERAGE_INC(flow_extract);
357 miniflow_initialize(&m.mf, m.buf);
358 miniflow_extract(packet, md, &m.mf);
359 miniflow_expand(&m.mf, flow);
362 /* Caller is responsible for initializing 'dst' with enough storage for
363 * FLOW_U32S * 4 bytes. */
365 miniflow_extract(struct ofpbuf *packet, const struct pkt_metadata *md,
366 struct miniflow *dst)
368 void *data = ofpbuf_data(packet);
369 size_t size = ofpbuf_size(packet);
370 uint32_t *values = miniflow_values(dst);
371 struct mf_ctx mf = { 0, values, values + FLOW_U32S };
374 uint8_t nw_frag, nw_tos, nw_ttl, nw_proto;
378 if (md->tunnel.ip_dst) {
379 miniflow_push_words(mf, tunnel, &md->tunnel,
380 sizeof md->tunnel / 4);
382 miniflow_push_uint32_check(mf, skb_priority, md->skb_priority);
383 miniflow_push_uint32_check(mf, pkt_mark, md->pkt_mark);
384 miniflow_push_uint32_check(mf, recirc_id, md->recirc_id);
385 miniflow_push_uint32(mf, in_port, odp_to_u32(md->in_port.odp_port));
388 /* Initialize packet's layer pointer and offsets. */
390 ofpbuf_set_frame(packet, data);
392 /* Must have full Ethernet header to proceed. */
393 if (OVS_UNLIKELY(size < sizeof(struct eth_header))) {
399 BUILD_ASSERT(offsetof(struct flow, dl_dst) + 6
400 == offsetof(struct flow, dl_src));
401 miniflow_push_words(mf, dl_dst, data, ETH_ADDR_LEN * 2 / 4);
402 /* dl_type, vlan_tci. */
403 vlan_tci = parse_vlan(&data, &size);
404 dl_type = parse_ethertype(&data, &size);
405 miniflow_push_be16(mf, dl_type, dl_type);
406 miniflow_push_be16(mf, vlan_tci, vlan_tci);
410 if (OVS_UNLIKELY(eth_type_mpls(dl_type))) {
412 const void *mpls = data;
414 packet->l2_5_ofs = (char *)data - l2;
415 count = parse_mpls(&data, &size);
416 miniflow_push_words(mf, mpls_lse, mpls, count);
420 packet->l3_ofs = (char *)data - l2;
423 if (OVS_LIKELY(dl_type == htons(ETH_TYPE_IP))) {
424 const struct ip_header *nh = data;
427 if (OVS_UNLIKELY(size < IP_HEADER_LEN)) {
430 ip_len = IP_IHL(nh->ip_ihl_ver) * 4;
432 if (OVS_UNLIKELY(ip_len < IP_HEADER_LEN)) {
436 /* Push both source and destination address at once. */
437 miniflow_push_words(mf, nw_src, &nh->ip_src, 2);
441 nw_proto = nh->ip_proto;
442 if (OVS_UNLIKELY(IP_IS_FRAGMENT(nh->ip_frag_off))) {
443 nw_frag = FLOW_NW_FRAG_ANY;
444 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
445 nw_frag |= FLOW_NW_FRAG_LATER;
448 if (OVS_UNLIKELY(size < ip_len)) {
451 data_pull(&data, &size, ip_len);
453 } else if (dl_type == htons(ETH_TYPE_IPV6)) {
454 const struct ovs_16aligned_ip6_hdr *nh;
457 if (OVS_UNLIKELY(size < sizeof *nh)) {
460 nh = data_pull(&data, &size, sizeof *nh);
462 miniflow_push_words(mf, ipv6_src, &nh->ip6_src,
463 sizeof nh->ip6_src / 4);
464 miniflow_push_words(mf, ipv6_dst, &nh->ip6_dst,
465 sizeof nh->ip6_dst / 4);
467 tc_flow = get_16aligned_be32(&nh->ip6_flow);
469 ovs_be32 label = tc_flow & htonl(IPV6_LABEL_MASK);
470 miniflow_push_be32_check(mf, ipv6_label, label);
473 nw_tos = ntohl(tc_flow) >> 20;
474 nw_ttl = nh->ip6_hlim;
475 nw_proto = nh->ip6_nxt;
478 if (OVS_LIKELY((nw_proto != IPPROTO_HOPOPTS)
479 && (nw_proto != IPPROTO_ROUTING)
480 && (nw_proto != IPPROTO_DSTOPTS)
481 && (nw_proto != IPPROTO_AH)
482 && (nw_proto != IPPROTO_FRAGMENT))) {
483 /* It's either a terminal header (e.g., TCP, UDP) or one we
484 * don't understand. In either case, we're done with the
485 * packet, so use it to fill in 'nw_proto'. */
489 /* We only verify that at least 8 bytes of the next header are
490 * available, but many of these headers are longer. Ensure that
491 * accesses within the extension header are within those first 8
492 * bytes. All extension headers are required to be at least 8
494 if (OVS_UNLIKELY(size < 8)) {
498 if ((nw_proto == IPPROTO_HOPOPTS)
499 || (nw_proto == IPPROTO_ROUTING)
500 || (nw_proto == IPPROTO_DSTOPTS)) {
501 /* These headers, while different, have the fields we care
502 * about in the same location and with the same
504 const struct ip6_ext *ext_hdr = data;
505 nw_proto = ext_hdr->ip6e_nxt;
506 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
507 (ext_hdr->ip6e_len + 1) * 8))) {
510 } else if (nw_proto == IPPROTO_AH) {
511 /* A standard AH definition isn't available, but the fields
512 * we care about are in the same location as the generic
513 * option header--only the header length is calculated
515 const struct ip6_ext *ext_hdr = data;
516 nw_proto = ext_hdr->ip6e_nxt;
517 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
518 (ext_hdr->ip6e_len + 2) * 4))) {
521 } else if (nw_proto == IPPROTO_FRAGMENT) {
522 const struct ovs_16aligned_ip6_frag *frag_hdr = data;
524 nw_proto = frag_hdr->ip6f_nxt;
525 if (!data_try_pull(&data, &size, sizeof *frag_hdr)) {
529 /* We only process the first fragment. */
530 if (frag_hdr->ip6f_offlg != htons(0)) {
531 nw_frag = FLOW_NW_FRAG_ANY;
532 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
533 nw_frag |= FLOW_NW_FRAG_LATER;
534 nw_proto = IPPROTO_FRAGMENT;
541 if (dl_type == htons(ETH_TYPE_ARP) ||
542 dl_type == htons(ETH_TYPE_RARP)) {
543 uint8_t arp_buf[2][ETH_ADDR_LEN];
544 const struct arp_eth_header *arp = (const struct arp_eth_header *)
545 data_try_pull(&data, &size, ARP_ETH_HEADER_LEN);
547 if (OVS_LIKELY(arp) && OVS_LIKELY(arp->ar_hrd == htons(1))
548 && OVS_LIKELY(arp->ar_pro == htons(ETH_TYPE_IP))
549 && OVS_LIKELY(arp->ar_hln == ETH_ADDR_LEN)
550 && OVS_LIKELY(arp->ar_pln == 4)) {
551 miniflow_push_words(mf, nw_src, &arp->ar_spa, 1);
552 miniflow_push_words(mf, nw_dst, &arp->ar_tpa, 1);
554 /* We only match on the lower 8 bits of the opcode. */
555 if (OVS_LIKELY(ntohs(arp->ar_op) <= 0xff)) {
556 miniflow_push_be32(mf, nw_frag, htonl(ntohs(arp->ar_op)));
559 /* Must be adjacent. */
560 BUILD_ASSERT(offsetof(struct flow, arp_sha) + 6
561 == offsetof(struct flow, arp_tha));
563 memcpy(arp_buf[0], arp->ar_sha, ETH_ADDR_LEN);
564 memcpy(arp_buf[1], arp->ar_tha, ETH_ADDR_LEN);
565 miniflow_push_words(mf, arp_sha, arp_buf,
566 ETH_ADDR_LEN * 2 / 4);
572 packet->l4_ofs = (char *)data - l2;
573 miniflow_push_be32(mf, nw_frag,
574 BYTES_TO_BE32(nw_frag, nw_tos, nw_ttl, nw_proto));
576 if (OVS_LIKELY(!(nw_frag & FLOW_NW_FRAG_LATER))) {
577 if (OVS_LIKELY(nw_proto == IPPROTO_TCP)) {
578 if (OVS_LIKELY(size >= TCP_HEADER_LEN)) {
579 const struct tcp_header *tcp = data;
581 miniflow_push_be32(mf, tcp_flags,
582 TCP_FLAGS_BE32(tcp->tcp_ctl));
583 miniflow_push_words(mf, tp_src, &tcp->tcp_src, 1);
585 } else if (OVS_LIKELY(nw_proto == IPPROTO_UDP)) {
586 if (OVS_LIKELY(size >= UDP_HEADER_LEN)) {
587 const struct udp_header *udp = data;
589 miniflow_push_words(mf, tp_src, &udp->udp_src, 1);
591 } else if (OVS_LIKELY(nw_proto == IPPROTO_SCTP)) {
592 if (OVS_LIKELY(size >= SCTP_HEADER_LEN)) {
593 const struct sctp_header *sctp = data;
595 miniflow_push_words(mf, tp_src, &sctp->sctp_src, 1);
597 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMP)) {
598 if (OVS_LIKELY(size >= ICMP_HEADER_LEN)) {
599 const struct icmp_header *icmp = data;
601 miniflow_push_be16(mf, tp_src, htons(icmp->icmp_type));
602 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp_code));
604 } else if (OVS_LIKELY(nw_proto == IPPROTO_IGMP)) {
605 if (OVS_LIKELY(size >= IGMP_HEADER_LEN)) {
606 const struct igmp_header *igmp = data;
608 miniflow_push_be16(mf, tp_src, htons(igmp->igmp_type));
609 miniflow_push_be16(mf, tp_dst, htons(igmp->igmp_code));
610 miniflow_push_be32(mf, igmp_group_ip4,
611 get_16aligned_be32(&igmp->group));
613 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMPV6)) {
614 if (OVS_LIKELY(size >= sizeof(struct icmp6_hdr))) {
615 const struct in6_addr *nd_target = NULL;
616 uint8_t arp_buf[2][ETH_ADDR_LEN];
617 const struct icmp6_hdr *icmp = data_pull(&data, &size,
619 memset(arp_buf, 0, sizeof arp_buf);
620 if (OVS_LIKELY(parse_icmpv6(&data, &size, icmp, &nd_target,
622 miniflow_push_words(mf, arp_sha, arp_buf,
623 ETH_ADDR_LEN * 2 / 4);
625 miniflow_push_words(mf, nd_target, nd_target,
626 sizeof *nd_target / 4);
628 miniflow_push_be16(mf, tp_src, htons(icmp->icmp6_type));
629 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp6_code));
635 miniflow_push_uint32_check(mf, dp_hash, md->dp_hash);
641 /* For every bit of a field that is wildcarded in 'wildcards', sets the
642 * corresponding bit in 'flow' to zero. */
644 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
646 uint32_t *flow_u32 = (uint32_t *) flow;
647 const uint32_t *wc_u32 = (const uint32_t *) &wildcards->masks;
650 for (i = 0; i < FLOW_U32S; i++) {
651 flow_u32[i] &= wc_u32[i];
656 flow_unwildcard_tp_ports(const struct flow *flow, struct flow_wildcards *wc)
658 if (flow->nw_proto != IPPROTO_ICMP) {
659 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
660 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
662 wc->masks.tp_src = htons(0xff);
663 wc->masks.tp_dst = htons(0xff);
667 /* Initializes 'fmd' with the metadata found in 'flow'. */
669 flow_get_metadata(const struct flow *flow, struct flow_metadata *fmd)
671 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 27);
673 fmd->dp_hash = flow->dp_hash;
674 fmd->recirc_id = flow->recirc_id;
675 fmd->tun_id = flow->tunnel.tun_id;
676 fmd->tun_src = flow->tunnel.ip_src;
677 fmd->tun_dst = flow->tunnel.ip_dst;
678 fmd->metadata = flow->metadata;
679 memcpy(fmd->regs, flow->regs, sizeof fmd->regs);
680 fmd->pkt_mark = flow->pkt_mark;
681 fmd->in_port = flow->in_port.ofp_port;
685 flow_to_string(const struct flow *flow)
687 struct ds ds = DS_EMPTY_INITIALIZER;
688 flow_format(&ds, flow);
693 flow_tun_flag_to_string(uint32_t flags)
696 case FLOW_TNL_F_DONT_FRAGMENT:
698 case FLOW_TNL_F_CSUM:
710 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
711 uint32_t flags, char del)
719 uint32_t bit = rightmost_1bit(flags);
722 s = bit_to_string(bit);
724 ds_put_format(ds, "%s%c", s, del);
733 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
739 format_flags_masked(struct ds *ds, const char *name,
740 const char *(*bit_to_string)(uint32_t), uint32_t flags,
744 ds_put_format(ds, "%s=", name);
747 uint32_t bit = rightmost_1bit(mask);
748 const char *s = bit_to_string(bit);
750 ds_put_format(ds, "%s%s", (flags & bit) ? "+" : "-",
751 s ? s : "[Unknown]");
757 flow_format(struct ds *ds, const struct flow *flow)
761 match_wc_init(&match, flow);
762 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
766 flow_print(FILE *stream, const struct flow *flow)
768 char *s = flow_to_string(flow);
773 /* flow_wildcards functions. */
775 /* Initializes 'wc' as a set of wildcards that matches every packet. */
777 flow_wildcards_init_catchall(struct flow_wildcards *wc)
779 memset(&wc->masks, 0, sizeof wc->masks);
782 /* Clear the metadata and register wildcard masks. They are not packet
785 flow_wildcards_clear_non_packet_fields(struct flow_wildcards *wc)
787 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
788 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
791 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
794 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
796 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
799 for (i = 0; i < FLOW_U32S; i++) {
807 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
808 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
809 * in 'src1' or 'src2' or both. */
811 flow_wildcards_and(struct flow_wildcards *dst,
812 const struct flow_wildcards *src1,
813 const struct flow_wildcards *src2)
815 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
816 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
817 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
820 for (i = 0; i < FLOW_U32S; i++) {
821 dst_u32[i] = src1_u32[i] & src2_u32[i];
825 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
826 * is, a bit or a field is wildcarded in 'dst' if it is neither
827 * wildcarded in 'src1' nor 'src2'. */
829 flow_wildcards_or(struct flow_wildcards *dst,
830 const struct flow_wildcards *src1,
831 const struct flow_wildcards *src2)
833 uint32_t *dst_u32 = (uint32_t *) &dst->masks;
834 const uint32_t *src1_u32 = (const uint32_t *) &src1->masks;
835 const uint32_t *src2_u32 = (const uint32_t *) &src2->masks;
838 for (i = 0; i < FLOW_U32S; i++) {
839 dst_u32[i] = src1_u32[i] | src2_u32[i];
843 /* Returns a hash of the wildcards in 'wc'. */
845 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
847 return flow_hash(&wc->masks, basis);
850 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
853 flow_wildcards_equal(const struct flow_wildcards *a,
854 const struct flow_wildcards *b)
856 return flow_equal(&a->masks, &b->masks);
859 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
860 * 'b', false otherwise. */
862 flow_wildcards_has_extra(const struct flow_wildcards *a,
863 const struct flow_wildcards *b)
865 const uint32_t *a_u32 = (const uint32_t *) &a->masks;
866 const uint32_t *b_u32 = (const uint32_t *) &b->masks;
869 for (i = 0; i < FLOW_U32S; i++) {
870 if ((a_u32[i] & b_u32[i]) != b_u32[i]) {
877 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
878 * in 'wc' do not need to be equal in 'a' and 'b'. */
880 flow_equal_except(const struct flow *a, const struct flow *b,
881 const struct flow_wildcards *wc)
883 const uint32_t *a_u32 = (const uint32_t *) a;
884 const uint32_t *b_u32 = (const uint32_t *) b;
885 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
888 for (i = 0; i < FLOW_U32S; i++) {
889 if ((a_u32[i] ^ b_u32[i]) & wc_u32[i]) {
896 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
897 * (A 0-bit indicates a wildcard bit.) */
899 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
901 wc->masks.regs[idx] = mask;
904 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
905 * (A 0-bit indicates a wildcard bit.) */
907 flow_wildcards_set_xreg_mask(struct flow_wildcards *wc, int idx, uint64_t mask)
909 flow_set_xreg(&wc->masks, idx, mask);
912 /* Calculates the 5-tuple hash from the given miniflow.
913 * This returns the same value as flow_hash_5tuple for the corresponding
916 miniflow_hash_5tuple(const struct miniflow *flow, uint32_t basis)
918 uint32_t hash = basis;
921 ovs_be16 dl_type = MINIFLOW_GET_BE16(flow, dl_type);
923 hash = hash_add(hash, MINIFLOW_GET_U8(flow, nw_proto));
925 /* Separate loops for better optimization. */
926 if (dl_type == htons(ETH_TYPE_IPV6)) {
927 uint64_t map = MINIFLOW_MAP(ipv6_src) | MINIFLOW_MAP(ipv6_dst)
928 | MINIFLOW_MAP(tp_src); /* Covers both ports */
931 MINIFLOW_FOR_EACH_IN_MAP(value, flow, map) {
932 hash = hash_add(hash, value);
935 uint64_t map = MINIFLOW_MAP(nw_src) | MINIFLOW_MAP(nw_dst)
936 | MINIFLOW_MAP(tp_src); /* Covers both ports */
939 MINIFLOW_FOR_EACH_IN_MAP(value, flow, map) {
940 hash = hash_add(hash, value);
943 hash = hash_finish(hash, 42); /* Arbitrary number. */
948 BUILD_ASSERT_DECL(offsetof(struct flow, tp_src) + 2
949 == offsetof(struct flow, tp_dst) &&
950 offsetof(struct flow, tp_src) / 4
951 == offsetof(struct flow, tp_dst) / 4);
952 BUILD_ASSERT_DECL(offsetof(struct flow, ipv6_src) + 16
953 == offsetof(struct flow, ipv6_dst));
955 /* Calculates the 5-tuple hash from the given flow. */
957 flow_hash_5tuple(const struct flow *flow, uint32_t basis)
959 uint32_t hash = basis;
962 const uint32_t *flow_u32 = (const uint32_t *)flow;
964 hash = hash_add(hash, flow->nw_proto);
966 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
967 int ofs = offsetof(struct flow, ipv6_src) / 4;
968 int end = ofs + 2 * sizeof flow->ipv6_src / 4;
971 hash = hash_add(hash, flow_u32[ofs++]);
974 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_src);
975 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_dst);
977 hash = hash_add(hash, flow_u32[offsetof(struct flow, tp_src) / 4]);
979 hash = hash_finish(hash, 42); /* Arbitrary number. */
984 /* Hashes 'flow' based on its L2 through L4 protocol information. */
986 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
991 struct in6_addr ipv6_addr;
996 uint8_t eth_addr[ETH_ADDR_LEN];
1002 memset(&fields, 0, sizeof fields);
1003 for (i = 0; i < ETH_ADDR_LEN; i++) {
1004 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
1006 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
1007 fields.eth_type = flow->dl_type;
1009 /* UDP source and destination port are not taken into account because they
1010 * will not necessarily be symmetric in a bidirectional flow. */
1011 if (fields.eth_type == htons(ETH_TYPE_IP)) {
1012 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
1013 fields.ip_proto = flow->nw_proto;
1014 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1015 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1017 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
1018 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
1019 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
1020 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
1022 for (i=0; i<16; i++) {
1023 ipv6_addr[i] = a[i] ^ b[i];
1025 fields.ip_proto = flow->nw_proto;
1026 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1027 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1030 return jhash_bytes(&fields, sizeof fields, basis);
1033 /* Initialize a flow with random fields that matter for nx_hash_fields. */
1035 flow_random_hash_fields(struct flow *flow)
1037 uint16_t rnd = random_uint16();
1039 /* Initialize to all zeros. */
1040 memset(flow, 0, sizeof *flow);
1042 eth_addr_random(flow->dl_src);
1043 eth_addr_random(flow->dl_dst);
1045 flow->vlan_tci = (OVS_FORCE ovs_be16) (random_uint16() & VLAN_VID_MASK);
1047 /* Make most of the random flows IPv4, some IPv6, and rest random. */
1048 flow->dl_type = rnd < 0x8000 ? htons(ETH_TYPE_IP) :
1049 rnd < 0xc000 ? htons(ETH_TYPE_IPV6) : (OVS_FORCE ovs_be16)rnd;
1051 if (dl_type_is_ip_any(flow->dl_type)) {
1052 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1053 flow->nw_src = (OVS_FORCE ovs_be32)random_uint32();
1054 flow->nw_dst = (OVS_FORCE ovs_be32)random_uint32();
1056 random_bytes(&flow->ipv6_src, sizeof flow->ipv6_src);
1057 random_bytes(&flow->ipv6_dst, sizeof flow->ipv6_dst);
1059 /* Make most of IP flows TCP, some UDP or SCTP, and rest random. */
1060 rnd = random_uint16();
1061 flow->nw_proto = rnd < 0x8000 ? IPPROTO_TCP :
1062 rnd < 0xc000 ? IPPROTO_UDP :
1063 rnd < 0xd000 ? IPPROTO_SCTP : (uint8_t)rnd;
1064 if (flow->nw_proto == IPPROTO_TCP ||
1065 flow->nw_proto == IPPROTO_UDP ||
1066 flow->nw_proto == IPPROTO_SCTP) {
1067 flow->tp_src = (OVS_FORCE ovs_be16)random_uint16();
1068 flow->tp_dst = (OVS_FORCE ovs_be16)random_uint16();
1073 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
1075 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
1076 enum nx_hash_fields fields)
1079 case NX_HASH_FIELDS_ETH_SRC:
1080 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1083 case NX_HASH_FIELDS_SYMMETRIC_L4:
1084 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1085 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1086 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1087 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1088 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1089 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1090 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1091 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1093 if (is_ip_any(flow)) {
1094 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1095 flow_unwildcard_tp_ports(flow, wc);
1097 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1105 /* Hashes the portions of 'flow' designated by 'fields'. */
1107 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
1112 case NX_HASH_FIELDS_ETH_SRC:
1113 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
1115 case NX_HASH_FIELDS_SYMMETRIC_L4:
1116 return flow_hash_symmetric_l4(flow, basis);
1122 /* Returns a string representation of 'fields'. */
1124 flow_hash_fields_to_str(enum nx_hash_fields fields)
1127 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
1128 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
1129 default: return "<unknown>";
1133 /* Returns true if the value of 'fields' is supported. Otherwise false. */
1135 flow_hash_fields_valid(enum nx_hash_fields fields)
1137 return fields == NX_HASH_FIELDS_ETH_SRC
1138 || fields == NX_HASH_FIELDS_SYMMETRIC_L4;
1141 /* Returns a hash value for the bits of 'flow' that are active based on
1142 * 'wc', given 'basis'. */
1144 flow_hash_in_wildcards(const struct flow *flow,
1145 const struct flow_wildcards *wc, uint32_t basis)
1147 const uint32_t *wc_u32 = (const uint32_t *) &wc->masks;
1148 const uint32_t *flow_u32 = (const uint32_t *) flow;
1153 for (i = 0; i < FLOW_U32S; i++) {
1154 hash = hash_add(hash, flow_u32[i] & wc_u32[i]);
1156 return hash_finish(hash, 4 * FLOW_U32S);
1159 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1160 * OpenFlow 1.0 "dl_vlan" value:
1162 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
1163 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
1164 * 'flow' previously matched packets without a VLAN header).
1166 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
1167 * without a VLAN tag.
1169 * - Other values of 'vid' should not be used. */
1171 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
1173 if (vid == htons(OFP10_VLAN_NONE)) {
1174 flow->vlan_tci = htons(0);
1176 vid &= htons(VLAN_VID_MASK);
1177 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1178 flow->vlan_tci |= htons(VLAN_CFI) | vid;
1182 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1183 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
1186 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
1188 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
1189 flow->vlan_tci &= ~mask;
1190 flow->vlan_tci |= vid & mask;
1193 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
1196 * This function has no effect on the VLAN ID that 'flow' matches.
1198 * After calling this function, 'flow' will not match packets without a VLAN
1201 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
1204 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1205 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
1208 /* Returns the number of MPLS LSEs present in 'flow'
1210 * Returns 0 if the 'dl_type' of 'flow' is not an MPLS ethernet type.
1211 * Otherwise traverses 'flow''s MPLS label stack stopping at the
1212 * first entry that has the BoS bit set. If no such entry exists then
1213 * the maximum number of LSEs that can be stored in 'flow' is returned.
1216 flow_count_mpls_labels(const struct flow *flow, struct flow_wildcards *wc)
1219 wc->masks.dl_type = OVS_BE16_MAX;
1221 if (eth_type_mpls(flow->dl_type)) {
1223 int len = FLOW_MAX_MPLS_LABELS;
1225 for (i = 0; i < len; i++) {
1227 wc->masks.mpls_lse[i] |= htonl(MPLS_BOS_MASK);
1229 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1240 /* Returns the number consecutive of MPLS LSEs, starting at the
1241 * innermost LSE, that are common in 'a' and 'b'.
1243 * 'an' must be flow_count_mpls_labels(a).
1244 * 'bn' must be flow_count_mpls_labels(b).
1247 flow_count_common_mpls_labels(const struct flow *a, int an,
1248 const struct flow *b, int bn,
1249 struct flow_wildcards *wc)
1251 int min_n = MIN(an, bn);
1256 int a_last = an - 1;
1257 int b_last = bn - 1;
1260 for (i = 0; i < min_n; i++) {
1262 wc->masks.mpls_lse[a_last - i] = OVS_BE32_MAX;
1263 wc->masks.mpls_lse[b_last - i] = OVS_BE32_MAX;
1265 if (a->mpls_lse[a_last - i] != b->mpls_lse[b_last - i]) {
1276 /* Adds a new outermost MPLS label to 'flow' and changes 'flow''s Ethernet type
1277 * to 'mpls_eth_type', which must be an MPLS Ethertype.
1279 * If the new label is the first MPLS label in 'flow', it is generated as;
1281 * - label: 2, if 'flow' is IPv6, otherwise 0.
1283 * - TTL: IPv4 or IPv6 TTL, if present and nonzero, otherwise 64.
1285 * - TC: IPv4 or IPv6 TOS, if present, otherwise 0.
1289 * If the new label is the second or label MPLS label in 'flow', it is
1292 * - label: Copied from outer label.
1294 * - TTL: Copied from outer label.
1296 * - TC: Copied from outer label.
1300 * 'n' must be flow_count_mpls_labels(flow). 'n' must be less than
1301 * FLOW_MAX_MPLS_LABELS (because otherwise flow->mpls_lse[] would overflow).
1304 flow_push_mpls(struct flow *flow, int n, ovs_be16 mpls_eth_type,
1305 struct flow_wildcards *wc)
1307 ovs_assert(eth_type_mpls(mpls_eth_type));
1308 ovs_assert(n < FLOW_MAX_MPLS_LABELS);
1310 memset(wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1314 for (i = n; i >= 1; i--) {
1315 flow->mpls_lse[i] = flow->mpls_lse[i - 1];
1317 flow->mpls_lse[0] = (flow->mpls_lse[1]
1318 & htonl(~MPLS_BOS_MASK));
1320 int label = 0; /* IPv4 Explicit Null. */
1324 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1328 if (is_ip_any(flow)) {
1329 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1330 wc->masks.nw_tos |= IP_DSCP_MASK;
1335 wc->masks.nw_ttl = 0xff;
1338 flow->mpls_lse[0] = set_mpls_lse_values(ttl, tc, 1, htonl(label));
1340 /* Clear all L3 and L4 fields. */
1341 BUILD_ASSERT(FLOW_WC_SEQ == 27);
1342 memset((char *) flow + FLOW_SEGMENT_2_ENDS_AT, 0,
1343 sizeof(struct flow) - FLOW_SEGMENT_2_ENDS_AT);
1345 flow->dl_type = mpls_eth_type;
1348 /* Tries to remove the outermost MPLS label from 'flow'. Returns true if
1349 * successful, false otherwise. On success, sets 'flow''s Ethernet type to
1352 * 'n' must be flow_count_mpls_labels(flow). */
1354 flow_pop_mpls(struct flow *flow, int n, ovs_be16 eth_type,
1355 struct flow_wildcards *wc)
1360 /* Nothing to pop. */
1362 } else if (n == FLOW_MAX_MPLS_LABELS
1363 && !(flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK))) {
1364 /* Can't pop because we don't know what to fill in mpls_lse[n - 1]. */
1368 memset(wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1369 for (i = 1; i < n; i++) {
1370 flow->mpls_lse[i - 1] = flow->mpls_lse[i];
1372 flow->mpls_lse[n - 1] = 0;
1373 flow->dl_type = eth_type;
1377 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
1378 * as an OpenFlow 1.1 "mpls_label" value. */
1380 flow_set_mpls_label(struct flow *flow, int idx, ovs_be32 label)
1382 set_mpls_lse_label(&flow->mpls_lse[idx], label);
1385 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
1388 flow_set_mpls_ttl(struct flow *flow, int idx, uint8_t ttl)
1390 set_mpls_lse_ttl(&flow->mpls_lse[idx], ttl);
1393 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
1396 flow_set_mpls_tc(struct flow *flow, int idx, uint8_t tc)
1398 set_mpls_lse_tc(&flow->mpls_lse[idx], tc);
1401 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
1403 flow_set_mpls_bos(struct flow *flow, int idx, uint8_t bos)
1405 set_mpls_lse_bos(&flow->mpls_lse[idx], bos);
1408 /* Sets the entire MPLS LSE. */
1410 flow_set_mpls_lse(struct flow *flow, int idx, ovs_be32 lse)
1412 flow->mpls_lse[idx] = lse;
1416 flow_compose_l4(struct ofpbuf *b, const struct flow *flow)
1420 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
1421 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1422 if (flow->nw_proto == IPPROTO_TCP) {
1423 struct tcp_header *tcp;
1425 l4_len = sizeof *tcp;
1426 tcp = ofpbuf_put_zeros(b, l4_len);
1427 tcp->tcp_src = flow->tp_src;
1428 tcp->tcp_dst = flow->tp_dst;
1429 tcp->tcp_ctl = TCP_CTL(ntohs(flow->tcp_flags), 5);
1430 } else if (flow->nw_proto == IPPROTO_UDP) {
1431 struct udp_header *udp;
1433 l4_len = sizeof *udp;
1434 udp = ofpbuf_put_zeros(b, l4_len);
1435 udp->udp_src = flow->tp_src;
1436 udp->udp_dst = flow->tp_dst;
1437 } else if (flow->nw_proto == IPPROTO_SCTP) {
1438 struct sctp_header *sctp;
1440 l4_len = sizeof *sctp;
1441 sctp = ofpbuf_put_zeros(b, l4_len);
1442 sctp->sctp_src = flow->tp_src;
1443 sctp->sctp_dst = flow->tp_dst;
1444 } else if (flow->nw_proto == IPPROTO_ICMP) {
1445 struct icmp_header *icmp;
1447 l4_len = sizeof *icmp;
1448 icmp = ofpbuf_put_zeros(b, l4_len);
1449 icmp->icmp_type = ntohs(flow->tp_src);
1450 icmp->icmp_code = ntohs(flow->tp_dst);
1451 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
1452 } else if (flow->nw_proto == IPPROTO_IGMP) {
1453 struct igmp_header *igmp;
1455 l4_len = sizeof *igmp;
1456 igmp = ofpbuf_put_zeros(b, l4_len);
1457 igmp->igmp_type = ntohs(flow->tp_src);
1458 igmp->igmp_code = ntohs(flow->tp_dst);
1459 put_16aligned_be32(&igmp->group, flow->igmp_group_ip4);
1460 igmp->igmp_csum = csum(igmp, IGMP_HEADER_LEN);
1461 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
1462 struct icmp6_hdr *icmp;
1464 l4_len = sizeof *icmp;
1465 icmp = ofpbuf_put_zeros(b, l4_len);
1466 icmp->icmp6_type = ntohs(flow->tp_src);
1467 icmp->icmp6_code = ntohs(flow->tp_dst);
1469 if (icmp->icmp6_code == 0 &&
1470 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
1471 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
1472 struct in6_addr *nd_target;
1473 struct nd_opt_hdr *nd_opt;
1475 l4_len += sizeof *nd_target;
1476 nd_target = ofpbuf_put_zeros(b, sizeof *nd_target);
1477 *nd_target = flow->nd_target;
1479 if (!eth_addr_is_zero(flow->arp_sha)) {
1481 nd_opt = ofpbuf_put_zeros(b, 8);
1482 nd_opt->nd_opt_len = 1;
1483 nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
1484 memcpy(nd_opt + 1, flow->arp_sha, ETH_ADDR_LEN);
1486 if (!eth_addr_is_zero(flow->arp_tha)) {
1488 nd_opt = ofpbuf_put_zeros(b, 8);
1489 nd_opt->nd_opt_len = 1;
1490 nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR;
1491 memcpy(nd_opt + 1, flow->arp_tha, ETH_ADDR_LEN);
1494 icmp->icmp6_cksum = (OVS_FORCE uint16_t)
1495 csum(icmp, (char *)ofpbuf_tail(b) - (char *)icmp);
1501 /* Puts into 'b' a packet that flow_extract() would parse as having the given
1504 * (This is useful only for testing, obviously, and the packet isn't really
1505 * valid. It hasn't got some checksums filled in, for one, and lots of fields
1506 * are just zeroed.) */
1508 flow_compose(struct ofpbuf *b, const struct flow *flow)
1512 /* eth_compose() sets l3 pointer and makes sure it is 32-bit aligned. */
1513 eth_compose(b, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
1514 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
1515 struct eth_header *eth = ofpbuf_l2(b);
1516 eth->eth_type = htons(ofpbuf_size(b));
1520 if (flow->vlan_tci & htons(VLAN_CFI)) {
1521 eth_push_vlan(b, htons(ETH_TYPE_VLAN), flow->vlan_tci);
1524 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1525 struct ip_header *ip;
1527 ip = ofpbuf_put_zeros(b, sizeof *ip);
1528 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
1529 ip->ip_tos = flow->nw_tos;
1530 ip->ip_ttl = flow->nw_ttl;
1531 ip->ip_proto = flow->nw_proto;
1532 put_16aligned_be32(&ip->ip_src, flow->nw_src);
1533 put_16aligned_be32(&ip->ip_dst, flow->nw_dst);
1535 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
1536 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
1537 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
1538 ip->ip_frag_off |= htons(100);
1542 ofpbuf_set_l4(b, ofpbuf_tail(b));
1544 l4_len = flow_compose_l4(b, flow);
1547 ip->ip_tot_len = htons(b->l4_ofs - b->l3_ofs + l4_len);
1548 ip->ip_csum = csum(ip, sizeof *ip);
1549 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1550 struct ovs_16aligned_ip6_hdr *nh;
1552 nh = ofpbuf_put_zeros(b, sizeof *nh);
1553 put_16aligned_be32(&nh->ip6_flow, htonl(6 << 28) |
1554 htonl(flow->nw_tos << 20) | flow->ipv6_label);
1555 nh->ip6_hlim = flow->nw_ttl;
1556 nh->ip6_nxt = flow->nw_proto;
1558 memcpy(&nh->ip6_src, &flow->ipv6_src, sizeof(nh->ip6_src));
1559 memcpy(&nh->ip6_dst, &flow->ipv6_dst, sizeof(nh->ip6_dst));
1561 ofpbuf_set_l4(b, ofpbuf_tail(b));
1563 l4_len = flow_compose_l4(b, flow);
1566 nh->ip6_plen = htons(l4_len);
1567 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1568 flow->dl_type == htons(ETH_TYPE_RARP)) {
1569 struct arp_eth_header *arp;
1571 arp = ofpbuf_put_zeros(b, sizeof *arp);
1572 ofpbuf_set_l3(b, arp);
1573 arp->ar_hrd = htons(1);
1574 arp->ar_pro = htons(ETH_TYPE_IP);
1575 arp->ar_hln = ETH_ADDR_LEN;
1577 arp->ar_op = htons(flow->nw_proto);
1579 if (flow->nw_proto == ARP_OP_REQUEST ||
1580 flow->nw_proto == ARP_OP_REPLY) {
1581 put_16aligned_be32(&arp->ar_spa, flow->nw_src);
1582 put_16aligned_be32(&arp->ar_tpa, flow->nw_dst);
1583 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
1584 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
1588 if (eth_type_mpls(flow->dl_type)) {
1591 b->l2_5_ofs = b->l3_ofs;
1592 for (n = 1; n < FLOW_MAX_MPLS_LABELS; n++) {
1593 if (flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK)) {
1598 push_mpls(b, flow->dl_type, flow->mpls_lse[--n]);
1603 /* Compressed flow. */
1606 miniflow_n_values(const struct miniflow *flow)
1608 return count_1bits(flow->map);
1612 miniflow_alloc_values(struct miniflow *flow, int n)
1614 int size = MINIFLOW_VALUES_SIZE(n);
1616 if (size <= sizeof flow->inline_values) {
1617 flow->values_inline = true;
1618 return flow->inline_values;
1620 COVERAGE_INC(miniflow_malloc);
1621 flow->values_inline = false;
1622 flow->offline_values = xmalloc(size);
1623 return flow->offline_values;
1627 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
1628 * the caller. The caller must have already initialized 'dst->map' properly
1629 * to indicate the significant uint32_t elements of 'src'. 'n' must be the
1630 * number of 1-bits in 'dst->map'.
1632 * Normally the significant elements are the ones that are non-zero. However,
1633 * when a miniflow is initialized from a (mini)mask, the values can be zeroes,
1634 * so that the flow and mask always have the same maps.
1636 * This function initializes values (either inline if possible or with
1637 * malloc() otherwise) and copies the uint32_t elements of 'src' indicated by
1638 * 'dst->map' into it. */
1640 miniflow_init__(struct miniflow *dst, const struct flow *src, int n)
1642 const uint32_t *src_u32 = (const uint32_t *) src;
1643 uint32_t *dst_u32 = miniflow_alloc_values(dst, n);
1646 for (map = dst->map; map; map = zero_rightmost_1bit(map)) {
1647 *dst_u32++ = src_u32[raw_ctz(map)];
1651 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1652 * with miniflow_destroy().
1653 * Always allocates offline storage. */
1655 miniflow_init(struct miniflow *dst, const struct flow *src)
1657 const uint32_t *src_u32 = (const uint32_t *) src;
1661 /* Initialize dst->map, counting the number of nonzero elements. */
1665 for (i = 0; i < FLOW_U32S; i++) {
1667 dst->map |= UINT64_C(1) << i;
1672 miniflow_init__(dst, src, n);
1675 /* Initializes 'dst' as a copy of 'src', using 'mask->map' as 'dst''s map. The
1676 * caller must eventually free 'dst' with miniflow_destroy(). */
1678 miniflow_init_with_minimask(struct miniflow *dst, const struct flow *src,
1679 const struct minimask *mask)
1681 dst->map = mask->masks.map;
1682 miniflow_init__(dst, src, miniflow_n_values(dst));
1685 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1686 * with miniflow_destroy(). */
1688 miniflow_clone(struct miniflow *dst, const struct miniflow *src)
1690 int size = MINIFLOW_VALUES_SIZE(miniflow_n_values(src));
1693 dst->map = src->map;
1694 if (size <= sizeof dst->inline_values) {
1695 dst->values_inline = true;
1696 values = dst->inline_values;
1698 dst->values_inline = false;
1699 COVERAGE_INC(miniflow_malloc);
1700 dst->offline_values = xmalloc(size);
1701 values = dst->offline_values;
1703 memcpy(values, miniflow_get_values(src), size);
1706 /* Initializes 'dst' as a copy of 'src'. The caller must have allocated
1707 * 'dst' to have inline space all data in 'src'. */
1709 miniflow_clone_inline(struct miniflow *dst, const struct miniflow *src,
1712 dst->values_inline = true;
1713 dst->map = src->map;
1714 memcpy(dst->inline_values, miniflow_get_values(src),
1715 MINIFLOW_VALUES_SIZE(n_values));
1718 /* Initializes 'dst' with the data in 'src', destroying 'src'.
1719 * The caller must eventually free 'dst' with miniflow_destroy().
1720 * 'dst' must be regularly sized miniflow, but 'src' can have
1721 * storage for more than the default MINI_N_INLINE inline
1724 miniflow_move(struct miniflow *dst, struct miniflow *src)
1726 int size = MINIFLOW_VALUES_SIZE(miniflow_n_values(src));
1728 dst->map = src->map;
1729 if (size <= sizeof dst->inline_values) {
1730 dst->values_inline = true;
1731 memcpy(dst->inline_values, miniflow_get_values(src), size);
1732 miniflow_destroy(src);
1733 } else if (src->values_inline) {
1734 dst->values_inline = false;
1735 COVERAGE_INC(miniflow_malloc);
1736 dst->offline_values = xmalloc(size);
1737 memcpy(dst->offline_values, src->inline_values, size);
1739 dst->values_inline = false;
1740 dst->offline_values = src->offline_values;
1744 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
1745 * itself resides; the caller is responsible for that. */
1747 miniflow_destroy(struct miniflow *flow)
1749 if (!flow->values_inline) {
1750 free(flow->offline_values);
1754 /* Initializes 'dst' as a copy of 'src'. */
1756 miniflow_expand(const struct miniflow *src, struct flow *dst)
1758 memset(dst, 0, sizeof *dst);
1759 flow_union_with_miniflow(dst, src);
1762 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'flow'
1763 * were expanded into a "struct flow". */
1765 miniflow_get(const struct miniflow *flow, unsigned int u32_ofs)
1767 return (flow->map & UINT64_C(1) << u32_ofs)
1768 ? *(miniflow_get_u32_values(flow) +
1769 count_1bits(flow->map & ((UINT64_C(1) << u32_ofs) - 1)))
1773 /* Returns true if 'a' and 'b' are the same flow, false otherwise. */
1775 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
1777 const uint32_t *ap = miniflow_get_u32_values(a);
1778 const uint32_t *bp = miniflow_get_u32_values(b);
1779 const uint64_t a_map = a->map;
1780 const uint64_t b_map = b->map;
1782 if (OVS_LIKELY(a_map == b_map)) {
1783 int count = miniflow_n_values(a);
1785 return !memcmp(ap, bp, count * sizeof *ap);
1789 for (map = a_map | b_map; map; map = zero_rightmost_1bit(map)) {
1790 uint64_t bit = rightmost_1bit(map);
1791 uint64_t a_value = a_map & bit ? *ap++ : 0;
1792 uint64_t b_value = b_map & bit ? *bp++ : 0;
1794 if (a_value != b_value) {
1803 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1804 * in 'mask', false if they differ. */
1806 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
1807 const struct minimask *mask)
1809 const uint32_t *p = miniflow_get_u32_values(&mask->masks);
1812 for (map = mask->masks.map; map; map = zero_rightmost_1bit(map)) {
1813 int ofs = raw_ctz(map);
1815 if ((miniflow_get(a, ofs) ^ miniflow_get(b, ofs)) & *p++) {
1823 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
1824 * in 'mask', false if they differ. */
1826 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
1827 const struct minimask *mask)
1829 const uint32_t *b_u32 = (const uint32_t *) b;
1830 const uint32_t *p = miniflow_get_u32_values(&mask->masks);
1833 for (map = mask->masks.map; map; map = zero_rightmost_1bit(map)) {
1834 int ofs = raw_ctz(map);
1836 if ((miniflow_get(a, ofs) ^ b_u32[ofs]) & *p++) {
1845 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1846 * with minimask_destroy(). */
1848 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
1850 miniflow_init(&mask->masks, &wc->masks);
1853 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1854 * with minimask_destroy(). */
1856 minimask_clone(struct minimask *dst, const struct minimask *src)
1858 miniflow_clone(&dst->masks, &src->masks);
1861 /* Initializes 'dst' with the data in 'src', destroying 'src'.
1862 * The caller must eventually free 'dst' with minimask_destroy(). */
1864 minimask_move(struct minimask *dst, struct minimask *src)
1866 miniflow_move(&dst->masks, &src->masks);
1869 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
1871 * The caller must provide room for FLOW_U32S "uint32_t"s in 'storage', for use
1872 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
1874 minimask_combine(struct minimask *dst_,
1875 const struct minimask *a_, const struct minimask *b_,
1876 uint32_t storage[FLOW_U32S])
1878 struct miniflow *dst = &dst_->masks;
1879 uint32_t *dst_values = storage;
1880 const struct miniflow *a = &a_->masks;
1881 const struct miniflow *b = &b_->masks;
1885 dst->values_inline = false;
1886 dst->offline_values = storage;
1889 for (map = a->map & b->map; map; map = zero_rightmost_1bit(map)) {
1890 int ofs = raw_ctz(map);
1891 uint32_t mask = miniflow_get(a, ofs) & miniflow_get(b, ofs);
1894 dst->map |= rightmost_1bit(map);
1895 dst_values[n++] = mask;
1900 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
1901 * itself resides; the caller is responsible for that. */
1903 minimask_destroy(struct minimask *mask)
1905 miniflow_destroy(&mask->masks);
1908 /* Initializes 'dst' as a copy of 'src'. */
1910 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
1912 miniflow_expand(&mask->masks, &wc->masks);
1915 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
1916 * were expanded into a "struct flow_wildcards". */
1918 minimask_get(const struct minimask *mask, unsigned int u32_ofs)
1920 return miniflow_get(&mask->masks, u32_ofs);
1923 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise. */
1925 minimask_equal(const struct minimask *a, const struct minimask *b)
1927 return miniflow_equal(&a->masks, &b->masks);
1930 /* Returns true if at least one bit matched by 'b' is wildcarded by 'a',
1931 * false otherwise. */
1933 minimask_has_extra(const struct minimask *a, const struct minimask *b)
1935 const uint32_t *p = miniflow_get_u32_values(&b->masks);
1938 for (map = b->masks.map; map; map = zero_rightmost_1bit(map)) {
1939 uint32_t a_u32 = minimask_get(a, raw_ctz(map));
1940 uint32_t b_u32 = *p++;
1942 if ((a_u32 & b_u32) != b_u32) {