2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015 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"
35 #include "dp-packet.h"
36 #include "openflow/openflow.h"
40 #include "unaligned.h"
42 COVERAGE_DEFINE(flow_extract);
43 COVERAGE_DEFINE(miniflow_malloc);
45 /* U64 indices for segmented flow classification. */
46 const uint8_t flow_segment_u64s[4] = {
47 FLOW_SEGMENT_1_ENDS_AT / sizeof(uint64_t),
48 FLOW_SEGMENT_2_ENDS_AT / sizeof(uint64_t),
49 FLOW_SEGMENT_3_ENDS_AT / sizeof(uint64_t),
53 /* Asserts that field 'f1' follows immediately after 'f0' in struct flow,
54 * without any intervening padding. */
55 #define ASSERT_SEQUENTIAL(f0, f1) \
56 BUILD_ASSERT_DECL(offsetof(struct flow, f0) \
57 + MEMBER_SIZEOF(struct flow, f0) \
58 == offsetof(struct flow, f1))
60 /* Asserts that fields 'f0' and 'f1' are in the same 32-bit aligned word within
62 #define ASSERT_SAME_WORD(f0, f1) \
63 BUILD_ASSERT_DECL(offsetof(struct flow, f0) / 4 \
64 == offsetof(struct flow, f1) / 4)
66 /* Asserts that 'f0' and 'f1' are both sequential and within the same 32-bit
67 * aligned word in struct flow. */
68 #define ASSERT_SEQUENTIAL_SAME_WORD(f0, f1) \
69 ASSERT_SEQUENTIAL(f0, f1); \
70 ASSERT_SAME_WORD(f0, f1)
72 /* miniflow_extract() assumes the following to be true to optimize the
73 * extraction process. */
74 ASSERT_SEQUENTIAL_SAME_WORD(dl_type, vlan_tci);
76 ASSERT_SEQUENTIAL_SAME_WORD(nw_frag, nw_tos);
77 ASSERT_SEQUENTIAL_SAME_WORD(nw_tos, nw_ttl);
78 ASSERT_SEQUENTIAL_SAME_WORD(nw_ttl, nw_proto);
80 /* TCP flags in the middle of a BE64, zeroes in the other half. */
81 BUILD_ASSERT_DECL(offsetof(struct flow, tcp_flags) % 8 == 4);
84 #define TCP_FLAGS_BE32(tcp_ctl) ((OVS_FORCE ovs_be32)TCP_FLAGS_BE16(tcp_ctl) \
87 #define TCP_FLAGS_BE32(tcp_ctl) ((OVS_FORCE ovs_be32)TCP_FLAGS_BE16(tcp_ctl))
90 ASSERT_SEQUENTIAL_SAME_WORD(tp_src, tp_dst);
92 /* Removes 'size' bytes from the head end of '*datap', of size '*sizep', which
93 * must contain at least 'size' bytes of data. Returns the first byte of data
95 static inline const void *
96 data_pull(const void **datap, size_t *sizep, size_t size)
98 const char *data = *datap;
104 /* If '*datap' has at least 'size' bytes of data, removes that many bytes from
105 * the head end of '*datap' and returns the first byte removed. Otherwise,
106 * returns a null pointer without modifying '*datap'. */
107 static inline const void *
108 data_try_pull(const void **datap, size_t *sizep, size_t size)
110 return OVS_LIKELY(*sizep >= size) ? data_pull(datap, sizep, size) : NULL;
113 /* Context for pushing data to a miniflow. */
117 uint64_t * const end;
120 /* miniflow_push_* macros allow filling in a miniflow data values in order.
121 * Assertions are needed only when the layout of the struct flow is modified.
122 * 'ofs' is a compile-time constant, which allows most of the code be optimized
123 * away. Some GCC versions gave warnings on ALWAYS_INLINE, so these are
124 * defined as macros. */
126 #if (FLOW_WC_SEQ != 32)
127 #define MINIFLOW_ASSERT(X) ovs_assert(X)
128 BUILD_MESSAGE("FLOW_WC_SEQ changed: miniflow_extract() will have runtime "
129 "assertions enabled. Consider updating FLOW_WC_SEQ after "
132 #define MINIFLOW_ASSERT(X)
135 #define miniflow_push_uint64_(MF, OFS, VALUE) \
137 MINIFLOW_ASSERT(MF.data < MF.end && (OFS) % 8 == 0 \
138 && !(MF.map & (UINT64_MAX << (OFS) / 8))); \
139 *MF.data++ = VALUE; \
140 MF.map |= UINT64_C(1) << (OFS) / 8; \
143 #define miniflow_push_be64_(MF, OFS, VALUE) \
144 miniflow_push_uint64_(MF, OFS, (OVS_FORCE uint64_t)(VALUE))
146 #define miniflow_push_uint32_(MF, OFS, VALUE) \
148 MINIFLOW_ASSERT(MF.data < MF.end && \
149 (((OFS) % 8 == 0 && !(MF.map & (UINT64_MAX << (OFS) / 8))) \
150 || ((OFS) % 8 == 4 && MF.map & (UINT64_C(1) << (OFS) / 8) \
151 && !(MF.map & (UINT64_MAX << ((OFS) / 8 + 1)))))); \
153 if ((OFS) % 8 == 0) { \
154 *(uint32_t *)MF.data = VALUE; \
155 MF.map |= UINT64_C(1) << (OFS) / 8; \
156 } else if ((OFS) % 8 == 4) { \
157 *((uint32_t *)MF.data + 1) = VALUE; \
162 #define miniflow_push_be32_(MF, OFS, VALUE) \
163 miniflow_push_uint32_(MF, OFS, (OVS_FORCE uint32_t)(VALUE))
165 #define miniflow_push_uint16_(MF, OFS, VALUE) \
167 MINIFLOW_ASSERT(MF.data < MF.end && \
168 (((OFS) % 8 == 0 && !(MF.map & (UINT64_MAX << (OFS) / 8))) \
169 || ((OFS) % 2 == 0 && MF.map & (UINT64_C(1) << (OFS) / 8) \
170 && !(MF.map & (UINT64_MAX << ((OFS) / 8 + 1)))))); \
172 if ((OFS) % 8 == 0) { \
173 *(uint16_t *)MF.data = VALUE; \
174 MF.map |= UINT64_C(1) << (OFS) / 8; \
175 } else if ((OFS) % 8 == 2) { \
176 *((uint16_t *)MF.data + 1) = VALUE; \
177 } else if ((OFS) % 8 == 4) { \
178 *((uint16_t *)MF.data + 2) = VALUE; \
179 } else if ((OFS) % 8 == 6) { \
180 *((uint16_t *)MF.data + 3) = VALUE; \
185 #define miniflow_pad_to_64_(MF, OFS) \
187 MINIFLOW_ASSERT((OFS) % 8 != 0); \
188 MINIFLOW_ASSERT(MF.map & (UINT64_C(1) << (OFS) / 8)); \
189 MINIFLOW_ASSERT(!(MF.map & (UINT64_MAX << ((OFS) / 8 + 1)))); \
191 memset((uint8_t *)MF.data + (OFS) % 8, 0, 8 - (OFS) % 8); \
195 #define miniflow_push_be16_(MF, OFS, VALUE) \
196 miniflow_push_uint16_(MF, OFS, (OVS_FORCE uint16_t)VALUE);
198 /* Data at 'valuep' may be unaligned. */
199 #define miniflow_push_words_(MF, OFS, VALUEP, N_WORDS) \
201 int ofs64 = (OFS) / 8; \
203 MINIFLOW_ASSERT(MF.data + (N_WORDS) <= MF.end && (OFS) % 8 == 0 \
204 && !(MF.map & (UINT64_MAX << ofs64))); \
206 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof *MF.data); \
207 MF.data += (N_WORDS); \
208 MF.map |= ((UINT64_MAX >> (64 - (N_WORDS))) << ofs64); \
211 /* Push 32-bit words padded to 64-bits. */
212 #define miniflow_push_words_32_(MF, OFS, VALUEP, N_WORDS) \
214 int ofs64 = (OFS) / 8; \
216 MINIFLOW_ASSERT(MF.data + DIV_ROUND_UP(N_WORDS, 2) <= MF.end \
218 && !(MF.map & (UINT64_MAX << ofs64))); \
220 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof(uint32_t)); \
221 MF.data += DIV_ROUND_UP(N_WORDS, 2); \
222 MF.map |= ((UINT64_MAX >> (64 - DIV_ROUND_UP(N_WORDS, 2))) << ofs64); \
223 if ((N_WORDS) & 1) { \
224 *((uint32_t *)MF.data - 1) = 0; \
228 /* Data at 'valuep' may be unaligned. */
229 /* MACs start 64-aligned, and must be followed by other data or padding. */
230 #define miniflow_push_macs_(MF, OFS, VALUEP) \
232 int ofs64 = (OFS) / 8; \
234 MINIFLOW_ASSERT(MF.data + 2 <= MF.end && (OFS) % 8 == 0 \
235 && !(MF.map & (UINT64_MAX << ofs64))); \
237 memcpy(MF.data, (VALUEP), 2 * ETH_ADDR_LEN); \
238 MF.data += 1; /* First word only. */ \
239 MF.map |= UINT64_C(3) << ofs64; /* Both words. */ \
242 #define miniflow_push_uint32(MF, FIELD, VALUE) \
243 miniflow_push_uint32_(MF, offsetof(struct flow, FIELD), VALUE)
245 #define miniflow_push_be32(MF, FIELD, VALUE) \
246 miniflow_push_be32_(MF, offsetof(struct flow, FIELD), VALUE)
248 #define miniflow_push_uint16(MF, FIELD, VALUE) \
249 miniflow_push_uint16_(MF, offsetof(struct flow, FIELD), VALUE)
251 #define miniflow_push_be16(MF, FIELD, VALUE) \
252 miniflow_push_be16_(MF, offsetof(struct flow, FIELD), VALUE)
254 #define miniflow_pad_to_64(MF, FIELD) \
255 miniflow_pad_to_64_(MF, offsetof(struct flow, FIELD))
257 #define miniflow_push_words(MF, FIELD, VALUEP, N_WORDS) \
258 miniflow_push_words_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
260 #define miniflow_push_words_32(MF, FIELD, VALUEP, N_WORDS) \
261 miniflow_push_words_32_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
263 #define miniflow_push_macs(MF, FIELD, VALUEP) \
264 miniflow_push_macs_(MF, offsetof(struct flow, FIELD), VALUEP)
266 /* Pulls the MPLS headers at '*datap' and returns the count of them. */
268 parse_mpls(const void **datap, size_t *sizep)
270 const struct mpls_hdr *mh;
273 while ((mh = data_try_pull(datap, sizep, sizeof *mh))) {
275 if (mh->mpls_lse.lo & htons(1 << MPLS_BOS_SHIFT)) {
279 return MIN(count, FLOW_MAX_MPLS_LABELS);
282 static inline ovs_be16
283 parse_vlan(const void **datap, size_t *sizep)
285 const struct eth_header *eth = *datap;
288 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
292 data_pull(datap, sizep, ETH_ADDR_LEN * 2);
294 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
295 if (OVS_LIKELY(*sizep
296 >= sizeof(struct qtag_prefix) + sizeof(ovs_be16))) {
297 const struct qtag_prefix *qp = data_pull(datap, sizep, sizeof *qp);
298 return qp->tci | htons(VLAN_CFI);
304 static inline ovs_be16
305 parse_ethertype(const void **datap, size_t *sizep)
307 const struct llc_snap_header *llc;
310 proto = *(ovs_be16 *) data_pull(datap, sizep, sizeof proto);
311 if (OVS_LIKELY(ntohs(proto) >= ETH_TYPE_MIN)) {
315 if (OVS_UNLIKELY(*sizep < sizeof *llc)) {
316 return htons(FLOW_DL_TYPE_NONE);
320 if (OVS_UNLIKELY(llc->llc.llc_dsap != LLC_DSAP_SNAP
321 || llc->llc.llc_ssap != LLC_SSAP_SNAP
322 || llc->llc.llc_cntl != LLC_CNTL_SNAP
323 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
324 sizeof llc->snap.snap_org))) {
325 return htons(FLOW_DL_TYPE_NONE);
328 data_pull(datap, sizep, sizeof *llc);
330 if (OVS_LIKELY(ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN)) {
331 return llc->snap.snap_type;
334 return htons(FLOW_DL_TYPE_NONE);
338 parse_icmpv6(const void **datap, size_t *sizep, const struct icmp6_hdr *icmp,
339 const struct in6_addr **nd_target,
340 uint8_t arp_buf[2][ETH_ADDR_LEN])
342 if (icmp->icmp6_code == 0 &&
343 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
344 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
346 *nd_target = data_try_pull(datap, sizep, sizeof **nd_target);
347 if (OVS_UNLIKELY(!*nd_target)) {
351 while (*sizep >= 8) {
352 /* The minimum size of an option is 8 bytes, which also is
353 * the size of Ethernet link-layer options. */
354 const struct nd_opt_hdr *nd_opt = *datap;
355 int opt_len = nd_opt->nd_opt_len * 8;
357 if (!opt_len || opt_len > *sizep) {
361 /* Store the link layer address if the appropriate option is
362 * provided. It is considered an error if the same link
363 * layer option is specified twice. */
364 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
366 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[0]))) {
367 memcpy(arp_buf[0], nd_opt + 1, ETH_ADDR_LEN);
371 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
373 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[1]))) {
374 memcpy(arp_buf[1], nd_opt + 1, ETH_ADDR_LEN);
380 if (OVS_UNLIKELY(!data_try_pull(datap, sizep, opt_len))) {
392 /* Initializes 'flow' members from 'packet' and 'md'
394 * Initializes 'packet' header l2 pointer to the start of the Ethernet
395 * header, and the layer offsets as follows:
397 * - packet->l2_5_ofs to the start of the MPLS shim header, or UINT16_MAX
398 * when there is no MPLS shim header.
400 * - packet->l3_ofs to just past the Ethernet header, or just past the
401 * vlan_header if one is present, to the first byte of the payload of the
402 * Ethernet frame. UINT16_MAX if the frame is too short to contain an
405 * - packet->l4_ofs to just past the IPv4 header, if one is present and
406 * has at least the content used for the fields of interest for the flow,
407 * otherwise UINT16_MAX.
410 flow_extract(struct dp_packet *packet, struct flow *flow)
414 uint64_t buf[FLOW_U64S];
417 COVERAGE_INC(flow_extract);
419 miniflow_extract(packet, &m.mf);
420 miniflow_expand(&m.mf, flow);
423 /* Caller is responsible for initializing 'dst' with enough storage for
424 * FLOW_U64S * 8 bytes. */
426 miniflow_extract(struct dp_packet *packet, struct miniflow *dst)
428 const struct pkt_metadata *md = &packet->md;
429 const void *data = dp_packet_data(packet);
430 size_t size = dp_packet_size(packet);
431 uint64_t *values = dst->values;
432 struct mf_ctx mf = { 0, values, values + FLOW_U64S };
435 uint8_t nw_frag, nw_tos, nw_ttl, nw_proto;
438 if (md->tunnel.ip_dst) {
439 miniflow_push_words(mf, tunnel, &md->tunnel,
440 offsetof(struct flow_tnl, metadata) /
442 if (md->tunnel.metadata.opt_map) {
443 miniflow_push_words(mf, tunnel.metadata, &md->tunnel.metadata,
444 sizeof md->tunnel.metadata / sizeof(uint64_t));
447 if (md->skb_priority || md->pkt_mark) {
448 miniflow_push_uint32(mf, skb_priority, md->skb_priority);
449 miniflow_push_uint32(mf, pkt_mark, md->pkt_mark);
451 miniflow_push_uint32(mf, dp_hash, md->dp_hash);
452 miniflow_push_uint32(mf, in_port, odp_to_u32(md->in_port.odp_port));
454 miniflow_push_uint32(mf, recirc_id, md->recirc_id);
455 miniflow_pad_to_64(mf, conj_id);
458 /* Initialize packet's layer pointer and offsets. */
460 dp_packet_reset_offsets(packet);
462 /* Must have full Ethernet header to proceed. */
463 if (OVS_UNLIKELY(size < sizeof(struct eth_header))) {
469 ASSERT_SEQUENTIAL(dl_dst, dl_src);
470 miniflow_push_macs(mf, dl_dst, data);
471 /* dl_type, vlan_tci. */
472 vlan_tci = parse_vlan(&data, &size);
473 dl_type = parse_ethertype(&data, &size);
474 miniflow_push_be16(mf, dl_type, dl_type);
475 miniflow_push_be16(mf, vlan_tci, vlan_tci);
479 if (OVS_UNLIKELY(eth_type_mpls(dl_type))) {
481 const void *mpls = data;
483 packet->l2_5_ofs = (char *)data - l2;
484 count = parse_mpls(&data, &size);
485 miniflow_push_words_32(mf, mpls_lse, mpls, count);
489 packet->l3_ofs = (char *)data - l2;
492 if (OVS_LIKELY(dl_type == htons(ETH_TYPE_IP))) {
493 const struct ip_header *nh = data;
497 if (OVS_UNLIKELY(size < IP_HEADER_LEN)) {
500 ip_len = IP_IHL(nh->ip_ihl_ver) * 4;
502 if (OVS_UNLIKELY(ip_len < IP_HEADER_LEN)) {
505 if (OVS_UNLIKELY(size < ip_len)) {
508 tot_len = ntohs(nh->ip_tot_len);
509 if (OVS_UNLIKELY(tot_len > size)) {
512 if (OVS_UNLIKELY(size - tot_len > UINT8_MAX)) {
515 dp_packet_set_l2_pad_size(packet, size - tot_len);
516 size = tot_len; /* Never pull padding. */
518 /* Push both source and destination address at once. */
519 miniflow_push_words(mf, nw_src, &nh->ip_src, 1);
521 miniflow_push_be32(mf, ipv6_label, 0); /* Padding for IPv4. */
525 nw_proto = nh->ip_proto;
526 if (OVS_UNLIKELY(IP_IS_FRAGMENT(nh->ip_frag_off))) {
527 nw_frag = FLOW_NW_FRAG_ANY;
528 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
529 nw_frag |= FLOW_NW_FRAG_LATER;
532 data_pull(&data, &size, ip_len);
533 } else if (dl_type == htons(ETH_TYPE_IPV6)) {
534 const struct ovs_16aligned_ip6_hdr *nh;
538 if (OVS_UNLIKELY(size < sizeof *nh)) {
541 nh = data_pull(&data, &size, sizeof *nh);
543 plen = ntohs(nh->ip6_plen);
544 if (OVS_UNLIKELY(plen > size)) {
547 /* Jumbo Payload option not supported yet. */
548 if (OVS_UNLIKELY(size - plen > UINT8_MAX)) {
551 dp_packet_set_l2_pad_size(packet, size - plen);
552 size = plen; /* Never pull padding. */
554 miniflow_push_words(mf, ipv6_src, &nh->ip6_src,
555 sizeof nh->ip6_src / 8);
556 miniflow_push_words(mf, ipv6_dst, &nh->ip6_dst,
557 sizeof nh->ip6_dst / 8);
559 tc_flow = get_16aligned_be32(&nh->ip6_flow);
561 ovs_be32 label = tc_flow & htonl(IPV6_LABEL_MASK);
562 miniflow_push_be32(mf, ipv6_label, label);
565 nw_tos = ntohl(tc_flow) >> 20;
566 nw_ttl = nh->ip6_hlim;
567 nw_proto = nh->ip6_nxt;
570 if (OVS_LIKELY((nw_proto != IPPROTO_HOPOPTS)
571 && (nw_proto != IPPROTO_ROUTING)
572 && (nw_proto != IPPROTO_DSTOPTS)
573 && (nw_proto != IPPROTO_AH)
574 && (nw_proto != IPPROTO_FRAGMENT))) {
575 /* It's either a terminal header (e.g., TCP, UDP) or one we
576 * don't understand. In either case, we're done with the
577 * packet, so use it to fill in 'nw_proto'. */
581 /* We only verify that at least 8 bytes of the next header are
582 * available, but many of these headers are longer. Ensure that
583 * accesses within the extension header are within those first 8
584 * bytes. All extension headers are required to be at least 8
586 if (OVS_UNLIKELY(size < 8)) {
590 if ((nw_proto == IPPROTO_HOPOPTS)
591 || (nw_proto == IPPROTO_ROUTING)
592 || (nw_proto == IPPROTO_DSTOPTS)) {
593 /* These headers, while different, have the fields we care
594 * about in the same location and with the same
596 const struct ip6_ext *ext_hdr = data;
597 nw_proto = ext_hdr->ip6e_nxt;
598 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
599 (ext_hdr->ip6e_len + 1) * 8))) {
602 } else if (nw_proto == IPPROTO_AH) {
603 /* A standard AH definition isn't available, but the fields
604 * we care about are in the same location as the generic
605 * option header--only the header length is calculated
607 const struct ip6_ext *ext_hdr = data;
608 nw_proto = ext_hdr->ip6e_nxt;
609 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
610 (ext_hdr->ip6e_len + 2) * 4))) {
613 } else if (nw_proto == IPPROTO_FRAGMENT) {
614 const struct ovs_16aligned_ip6_frag *frag_hdr = data;
616 nw_proto = frag_hdr->ip6f_nxt;
617 if (!data_try_pull(&data, &size, sizeof *frag_hdr)) {
621 /* We only process the first fragment. */
622 if (frag_hdr->ip6f_offlg != htons(0)) {
623 nw_frag = FLOW_NW_FRAG_ANY;
624 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
625 nw_frag |= FLOW_NW_FRAG_LATER;
626 nw_proto = IPPROTO_FRAGMENT;
633 if (dl_type == htons(ETH_TYPE_ARP) ||
634 dl_type == htons(ETH_TYPE_RARP)) {
635 uint8_t arp_buf[2][ETH_ADDR_LEN];
636 const struct arp_eth_header *arp = (const struct arp_eth_header *)
637 data_try_pull(&data, &size, ARP_ETH_HEADER_LEN);
639 if (OVS_LIKELY(arp) && OVS_LIKELY(arp->ar_hrd == htons(1))
640 && OVS_LIKELY(arp->ar_pro == htons(ETH_TYPE_IP))
641 && OVS_LIKELY(arp->ar_hln == ETH_ADDR_LEN)
642 && OVS_LIKELY(arp->ar_pln == 4)) {
643 miniflow_push_be32(mf, nw_src,
644 get_16aligned_be32(&arp->ar_spa));
645 miniflow_push_be32(mf, nw_dst,
646 get_16aligned_be32(&arp->ar_tpa));
648 /* We only match on the lower 8 bits of the opcode. */
649 if (OVS_LIKELY(ntohs(arp->ar_op) <= 0xff)) {
650 miniflow_push_be32(mf, ipv6_label, 0); /* Pad with ARP. */
651 miniflow_push_be32(mf, nw_frag, htonl(ntohs(arp->ar_op)));
654 /* Must be adjacent. */
655 ASSERT_SEQUENTIAL(arp_sha, arp_tha);
657 memcpy(arp_buf[0], arp->ar_sha, ETH_ADDR_LEN);
658 memcpy(arp_buf[1], arp->ar_tha, ETH_ADDR_LEN);
659 miniflow_push_macs(mf, arp_sha, arp_buf);
660 miniflow_pad_to_64(mf, tcp_flags);
666 packet->l4_ofs = (char *)data - l2;
667 miniflow_push_be32(mf, nw_frag,
668 BYTES_TO_BE32(nw_frag, nw_tos, nw_ttl, nw_proto));
670 if (OVS_LIKELY(!(nw_frag & FLOW_NW_FRAG_LATER))) {
671 if (OVS_LIKELY(nw_proto == IPPROTO_TCP)) {
672 if (OVS_LIKELY(size >= TCP_HEADER_LEN)) {
673 const struct tcp_header *tcp = data;
675 miniflow_push_be32(mf, arp_tha[2], 0);
676 miniflow_push_be32(mf, tcp_flags,
677 TCP_FLAGS_BE32(tcp->tcp_ctl));
678 miniflow_push_be16(mf, tp_src, tcp->tcp_src);
679 miniflow_push_be16(mf, tp_dst, tcp->tcp_dst);
680 miniflow_pad_to_64(mf, igmp_group_ip4);
682 } else if (OVS_LIKELY(nw_proto == IPPROTO_UDP)) {
683 if (OVS_LIKELY(size >= UDP_HEADER_LEN)) {
684 const struct udp_header *udp = data;
686 miniflow_push_be16(mf, tp_src, udp->udp_src);
687 miniflow_push_be16(mf, tp_dst, udp->udp_dst);
688 miniflow_pad_to_64(mf, igmp_group_ip4);
690 } else if (OVS_LIKELY(nw_proto == IPPROTO_SCTP)) {
691 if (OVS_LIKELY(size >= SCTP_HEADER_LEN)) {
692 const struct sctp_header *sctp = data;
694 miniflow_push_be16(mf, tp_src, sctp->sctp_src);
695 miniflow_push_be16(mf, tp_dst, sctp->sctp_dst);
696 miniflow_pad_to_64(mf, igmp_group_ip4);
698 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMP)) {
699 if (OVS_LIKELY(size >= ICMP_HEADER_LEN)) {
700 const struct icmp_header *icmp = data;
702 miniflow_push_be16(mf, tp_src, htons(icmp->icmp_type));
703 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp_code));
704 miniflow_pad_to_64(mf, igmp_group_ip4);
706 } else if (OVS_LIKELY(nw_proto == IPPROTO_IGMP)) {
707 if (OVS_LIKELY(size >= IGMP_HEADER_LEN)) {
708 const struct igmp_header *igmp = data;
710 miniflow_push_be16(mf, tp_src, htons(igmp->igmp_type));
711 miniflow_push_be16(mf, tp_dst, htons(igmp->igmp_code));
712 miniflow_push_be32(mf, igmp_group_ip4,
713 get_16aligned_be32(&igmp->group));
715 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMPV6)) {
716 if (OVS_LIKELY(size >= sizeof(struct icmp6_hdr))) {
717 const struct in6_addr *nd_target = NULL;
718 uint8_t arp_buf[2][ETH_ADDR_LEN];
719 const struct icmp6_hdr *icmp = data_pull(&data, &size,
721 memset(arp_buf, 0, sizeof arp_buf);
722 if (OVS_LIKELY(parse_icmpv6(&data, &size, icmp, &nd_target,
725 miniflow_push_words(mf, nd_target, nd_target,
726 sizeof *nd_target / 8);
728 miniflow_push_macs(mf, arp_sha, arp_buf);
729 miniflow_pad_to_64(mf, tcp_flags);
730 miniflow_push_be16(mf, tp_src, htons(icmp->icmp6_type));
731 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp6_code));
732 miniflow_pad_to_64(mf, igmp_group_ip4);
741 /* For every bit of a field that is wildcarded in 'wildcards', sets the
742 * corresponding bit in 'flow' to zero. */
744 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
746 uint64_t *flow_u64 = (uint64_t *) flow;
747 const uint64_t *wc_u64 = (const uint64_t *) &wildcards->masks;
750 for (i = 0; i < FLOW_U64S; i++) {
751 flow_u64[i] &= wc_u64[i];
756 flow_unwildcard_tp_ports(const struct flow *flow, struct flow_wildcards *wc)
758 if (flow->nw_proto != IPPROTO_ICMP) {
759 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
760 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
762 wc->masks.tp_src = htons(0xff);
763 wc->masks.tp_dst = htons(0xff);
767 /* Initializes 'flow_metadata' with the metadata found in 'flow'. */
769 flow_get_metadata(const struct flow *flow, struct match *flow_metadata)
773 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 32);
775 match_init_catchall(flow_metadata);
776 if (flow->tunnel.tun_id != htonll(0)) {
777 match_set_tun_id(flow_metadata, flow->tunnel.tun_id);
779 if (flow->tunnel.ip_src != htonl(0)) {
780 match_set_tun_src(flow_metadata, flow->tunnel.ip_src);
782 if (flow->tunnel.ip_dst != htonl(0)) {
783 match_set_tun_dst(flow_metadata, flow->tunnel.ip_dst);
785 if (flow->tunnel.gbp_id != htons(0)) {
786 match_set_tun_gbp_id(flow_metadata, flow->tunnel.gbp_id);
788 if (flow->tunnel.gbp_flags) {
789 match_set_tun_gbp_flags(flow_metadata, flow->tunnel.gbp_flags);
791 tun_metadata_get_fmd(&flow->tunnel.metadata, flow_metadata);
792 if (flow->metadata != htonll(0)) {
793 match_set_metadata(flow_metadata, flow->metadata);
796 for (i = 0; i < FLOW_N_REGS; i++) {
798 match_set_reg(flow_metadata, i, flow->regs[i]);
802 if (flow->pkt_mark != 0) {
803 match_set_pkt_mark(flow_metadata, flow->pkt_mark);
806 match_set_in_port(flow_metadata, flow->in_port.ofp_port);
810 flow_to_string(const struct flow *flow)
812 struct ds ds = DS_EMPTY_INITIALIZER;
813 flow_format(&ds, flow);
818 flow_tun_flag_to_string(uint32_t flags)
821 case FLOW_TNL_F_DONT_FRAGMENT:
823 case FLOW_TNL_F_CSUM:
835 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
836 uint32_t flags, char del)
844 uint32_t bit = rightmost_1bit(flags);
847 s = bit_to_string(bit);
849 ds_put_format(ds, "%s%c", s, del);
858 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
864 format_flags_masked(struct ds *ds, const char *name,
865 const char *(*bit_to_string)(uint32_t), uint32_t flags,
869 ds_put_format(ds, "%s=", name);
872 uint32_t bit = rightmost_1bit(mask);
873 const char *s = bit_to_string(bit);
875 ds_put_format(ds, "%s%s", (flags & bit) ? "+" : "-",
876 s ? s : "[Unknown]");
882 flow_format(struct ds *ds, const struct flow *flow)
885 struct flow_wildcards *wc = &match.wc;
887 match_wc_init(&match, flow);
889 /* As this function is most often used for formatting a packet in a
890 * packet-in message, skip formatting the packet context fields that are
891 * all-zeroes to make the print-out easier on the eyes. This means that a
892 * missing context field implies a zero value for that field. This is
893 * similar to OpenFlow encoding of these fields, as the specification
894 * states that all-zeroes context fields should not be encoded in the
895 * packet-in messages. */
896 if (!flow->in_port.ofp_port) {
897 WC_UNMASK_FIELD(wc, in_port);
899 if (!flow->skb_priority) {
900 WC_UNMASK_FIELD(wc, skb_priority);
902 if (!flow->pkt_mark) {
903 WC_UNMASK_FIELD(wc, pkt_mark);
905 if (!flow->recirc_id) {
906 WC_UNMASK_FIELD(wc, recirc_id);
908 if (!flow->dp_hash) {
909 WC_UNMASK_FIELD(wc, dp_hash);
911 for (int i = 0; i < FLOW_N_REGS; i++) {
912 if (!flow->regs[i]) {
913 WC_UNMASK_FIELD(wc, regs[i]);
916 if (!flow->metadata) {
917 WC_UNMASK_FIELD(wc, metadata);
920 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
924 flow_print(FILE *stream, const struct flow *flow)
926 char *s = flow_to_string(flow);
931 /* flow_wildcards functions. */
933 /* Initializes 'wc' as a set of wildcards that matches every packet. */
935 flow_wildcards_init_catchall(struct flow_wildcards *wc)
937 memset(&wc->masks, 0, sizeof wc->masks);
940 /* Converts a flow into flow wildcards. It sets the wildcard masks based on
941 * the packet headers extracted to 'flow'. It will not set the mask for fields
942 * that do not make sense for the packet type. OpenFlow-only metadata is
943 * wildcarded, but other metadata is unconditionally exact-matched. */
944 void flow_wildcards_init_for_packet(struct flow_wildcards *wc,
945 const struct flow *flow)
947 memset(&wc->masks, 0x0, sizeof wc->masks);
949 /* Update this function whenever struct flow changes. */
950 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 32);
952 if (flow->tunnel.ip_dst) {
953 if (flow->tunnel.flags & FLOW_TNL_F_KEY) {
954 WC_MASK_FIELD(wc, tunnel.tun_id);
956 WC_MASK_FIELD(wc, tunnel.ip_src);
957 WC_MASK_FIELD(wc, tunnel.ip_dst);
958 WC_MASK_FIELD(wc, tunnel.flags);
959 WC_MASK_FIELD(wc, tunnel.ip_tos);
960 WC_MASK_FIELD(wc, tunnel.ip_ttl);
961 WC_MASK_FIELD(wc, tunnel.tp_src);
962 WC_MASK_FIELD(wc, tunnel.tp_dst);
963 WC_MASK_FIELD(wc, tunnel.gbp_id);
964 WC_MASK_FIELD(wc, tunnel.gbp_flags);
966 if (flow->tunnel.metadata.opt_map) {
967 wc->masks.tunnel.metadata.opt_map = flow->tunnel.metadata.opt_map;
968 WC_MASK_FIELD(wc, tunnel.metadata.opts);
970 } else if (flow->tunnel.tun_id) {
971 WC_MASK_FIELD(wc, tunnel.tun_id);
974 /* metadata, regs, and conj_id wildcarded. */
976 WC_MASK_FIELD(wc, skb_priority);
977 WC_MASK_FIELD(wc, pkt_mark);
978 WC_MASK_FIELD(wc, recirc_id);
979 WC_MASK_FIELD(wc, dp_hash);
980 WC_MASK_FIELD(wc, in_port);
982 /* actset_output wildcarded. */
984 WC_MASK_FIELD(wc, dl_dst);
985 WC_MASK_FIELD(wc, dl_src);
986 WC_MASK_FIELD(wc, dl_type);
987 WC_MASK_FIELD(wc, vlan_tci);
989 if (flow->dl_type == htons(ETH_TYPE_IP)) {
990 WC_MASK_FIELD(wc, nw_src);
991 WC_MASK_FIELD(wc, nw_dst);
992 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
993 WC_MASK_FIELD(wc, ipv6_src);
994 WC_MASK_FIELD(wc, ipv6_dst);
995 WC_MASK_FIELD(wc, ipv6_label);
996 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
997 flow->dl_type == htons(ETH_TYPE_RARP)) {
998 WC_MASK_FIELD(wc, nw_src);
999 WC_MASK_FIELD(wc, nw_dst);
1000 WC_MASK_FIELD(wc, nw_proto);
1001 WC_MASK_FIELD(wc, arp_sha);
1002 WC_MASK_FIELD(wc, arp_tha);
1004 } else if (eth_type_mpls(flow->dl_type)) {
1005 for (int i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1006 WC_MASK_FIELD(wc, mpls_lse[i]);
1007 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1013 return; /* Unknown ethertype. */
1017 WC_MASK_FIELD(wc, nw_frag);
1018 WC_MASK_FIELD(wc, nw_tos);
1019 WC_MASK_FIELD(wc, nw_ttl);
1020 WC_MASK_FIELD(wc, nw_proto);
1022 /* No transport layer header in later fragments. */
1023 if (!(flow->nw_frag & FLOW_NW_FRAG_LATER) &&
1024 (flow->nw_proto == IPPROTO_ICMP ||
1025 flow->nw_proto == IPPROTO_ICMPV6 ||
1026 flow->nw_proto == IPPROTO_TCP ||
1027 flow->nw_proto == IPPROTO_UDP ||
1028 flow->nw_proto == IPPROTO_SCTP ||
1029 flow->nw_proto == IPPROTO_IGMP)) {
1030 WC_MASK_FIELD(wc, tp_src);
1031 WC_MASK_FIELD(wc, tp_dst);
1033 if (flow->nw_proto == IPPROTO_TCP) {
1034 WC_MASK_FIELD(wc, tcp_flags);
1035 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
1036 WC_MASK_FIELD(wc, arp_sha);
1037 WC_MASK_FIELD(wc, arp_tha);
1038 WC_MASK_FIELD(wc, nd_target);
1039 } else if (flow->nw_proto == IPPROTO_IGMP) {
1040 WC_MASK_FIELD(wc, igmp_group_ip4);
1045 /* Return a map of possible fields for a packet of the same type as 'flow'.
1046 * Including extra bits in the returned mask is not wrong, it is just less
1049 * This is a less precise version of flow_wildcards_init_for_packet() above. */
1051 flow_wc_map(const struct flow *flow)
1053 /* Update this function whenever struct flow changes. */
1054 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 32);
1056 uint64_t map = (flow->tunnel.ip_dst) ? MINIFLOW_MAP(tunnel) : 0;
1058 /* Metadata fields that can appear on packet input. */
1059 map |= MINIFLOW_MAP(skb_priority) | MINIFLOW_MAP(pkt_mark)
1060 | MINIFLOW_MAP(recirc_id) | MINIFLOW_MAP(dp_hash)
1061 | MINIFLOW_MAP(in_port)
1062 | MINIFLOW_MAP(dl_dst) | MINIFLOW_MAP(dl_src)
1063 | MINIFLOW_MAP(dl_type) | MINIFLOW_MAP(vlan_tci);
1065 /* Ethertype-dependent fields. */
1066 if (OVS_LIKELY(flow->dl_type == htons(ETH_TYPE_IP))) {
1067 map |= MINIFLOW_MAP(nw_src) | MINIFLOW_MAP(nw_dst)
1068 | MINIFLOW_MAP(nw_proto) | MINIFLOW_MAP(nw_frag)
1069 | MINIFLOW_MAP(nw_tos) | MINIFLOW_MAP(nw_ttl);
1070 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_IGMP)) {
1071 map |= MINIFLOW_MAP(igmp_group_ip4);
1073 map |= MINIFLOW_MAP(tcp_flags)
1074 | MINIFLOW_MAP(tp_src) | MINIFLOW_MAP(tp_dst);
1076 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1077 map |= MINIFLOW_MAP(ipv6_src) | MINIFLOW_MAP(ipv6_dst)
1078 | MINIFLOW_MAP(ipv6_label)
1079 | MINIFLOW_MAP(nw_proto) | MINIFLOW_MAP(nw_frag)
1080 | MINIFLOW_MAP(nw_tos) | MINIFLOW_MAP(nw_ttl);
1081 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_ICMPV6)) {
1082 map |= MINIFLOW_MAP(nd_target)
1083 | MINIFLOW_MAP(arp_sha) | MINIFLOW_MAP(arp_tha);
1085 map |= MINIFLOW_MAP(tcp_flags)
1086 | MINIFLOW_MAP(tp_src) | MINIFLOW_MAP(tp_dst);
1088 } else if (eth_type_mpls(flow->dl_type)) {
1089 map |= MINIFLOW_MAP(mpls_lse);
1090 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1091 flow->dl_type == htons(ETH_TYPE_RARP)) {
1092 map |= MINIFLOW_MAP(nw_src) | MINIFLOW_MAP(nw_dst)
1093 | MINIFLOW_MAP(nw_proto)
1094 | MINIFLOW_MAP(arp_sha) | MINIFLOW_MAP(arp_tha);
1100 /* Clear the metadata and register wildcard masks. They are not packet
1103 flow_wildcards_clear_non_packet_fields(struct flow_wildcards *wc)
1105 /* Update this function whenever struct flow changes. */
1106 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 32);
1108 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
1109 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
1110 wc->masks.actset_output = 0;
1111 wc->masks.conj_id = 0;
1114 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
1117 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
1119 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1122 for (i = 0; i < FLOW_U64S; i++) {
1130 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
1131 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
1132 * in 'src1' or 'src2' or both. */
1134 flow_wildcards_and(struct flow_wildcards *dst,
1135 const struct flow_wildcards *src1,
1136 const struct flow_wildcards *src2)
1138 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1139 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1140 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1143 for (i = 0; i < FLOW_U64S; i++) {
1144 dst_u64[i] = src1_u64[i] & src2_u64[i];
1148 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
1149 * is, a bit or a field is wildcarded in 'dst' if it is neither
1150 * wildcarded in 'src1' nor 'src2'. */
1152 flow_wildcards_or(struct flow_wildcards *dst,
1153 const struct flow_wildcards *src1,
1154 const struct flow_wildcards *src2)
1156 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1157 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1158 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1161 for (i = 0; i < FLOW_U64S; i++) {
1162 dst_u64[i] = src1_u64[i] | src2_u64[i];
1166 /* Returns a hash of the wildcards in 'wc'. */
1168 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
1170 return flow_hash(&wc->masks, basis);
1173 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
1176 flow_wildcards_equal(const struct flow_wildcards *a,
1177 const struct flow_wildcards *b)
1179 return flow_equal(&a->masks, &b->masks);
1182 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
1183 * 'b', false otherwise. */
1185 flow_wildcards_has_extra(const struct flow_wildcards *a,
1186 const struct flow_wildcards *b)
1188 const uint64_t *a_u64 = (const uint64_t *) &a->masks;
1189 const uint64_t *b_u64 = (const uint64_t *) &b->masks;
1192 for (i = 0; i < FLOW_U64S; i++) {
1193 if ((a_u64[i] & b_u64[i]) != b_u64[i]) {
1200 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
1201 * in 'wc' do not need to be equal in 'a' and 'b'. */
1203 flow_equal_except(const struct flow *a, const struct flow *b,
1204 const struct flow_wildcards *wc)
1206 const uint64_t *a_u64 = (const uint64_t *) a;
1207 const uint64_t *b_u64 = (const uint64_t *) b;
1208 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1211 for (i = 0; i < FLOW_U64S; i++) {
1212 if ((a_u64[i] ^ b_u64[i]) & wc_u64[i]) {
1219 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1220 * (A 0-bit indicates a wildcard bit.) */
1222 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
1224 wc->masks.regs[idx] = mask;
1227 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1228 * (A 0-bit indicates a wildcard bit.) */
1230 flow_wildcards_set_xreg_mask(struct flow_wildcards *wc, int idx, uint64_t mask)
1232 flow_set_xreg(&wc->masks, idx, mask);
1235 /* Calculates the 5-tuple hash from the given miniflow.
1236 * This returns the same value as flow_hash_5tuple for the corresponding
1239 miniflow_hash_5tuple(const struct miniflow *flow, uint32_t basis)
1241 uint32_t hash = basis;
1244 ovs_be16 dl_type = MINIFLOW_GET_BE16(flow, dl_type);
1246 hash = hash_add(hash, MINIFLOW_GET_U8(flow, nw_proto));
1248 /* Separate loops for better optimization. */
1249 if (dl_type == htons(ETH_TYPE_IPV6)) {
1250 uint64_t map = MINIFLOW_MAP(ipv6_src) | MINIFLOW_MAP(ipv6_dst);
1253 MINIFLOW_FOR_EACH_IN_MAP(value, flow, map) {
1254 hash = hash_add64(hash, value);
1257 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_src));
1258 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_dst));
1260 /* Add both ports at once. */
1261 hash = hash_add(hash, MINIFLOW_GET_U32(flow, tp_src));
1262 hash = hash_finish(hash, 42); /* Arbitrary number. */
1267 ASSERT_SEQUENTIAL_SAME_WORD(tp_src, tp_dst);
1268 ASSERT_SEQUENTIAL(ipv6_src, ipv6_dst);
1270 /* Calculates the 5-tuple hash from the given flow. */
1272 flow_hash_5tuple(const struct flow *flow, uint32_t basis)
1274 uint32_t hash = basis;
1277 hash = hash_add(hash, flow->nw_proto);
1279 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1280 const uint64_t *flow_u64 = (const uint64_t *)flow;
1281 int ofs = offsetof(struct flow, ipv6_src) / 8;
1282 int end = ofs + 2 * sizeof flow->ipv6_src / 8;
1284 for (;ofs < end; ofs++) {
1285 hash = hash_add64(hash, flow_u64[ofs]);
1288 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_src);
1289 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_dst);
1291 /* Add both ports at once. */
1292 hash = hash_add(hash,
1293 ((const uint32_t *)flow)[offsetof(struct flow, tp_src)
1294 / sizeof(uint32_t)]);
1295 hash = hash_finish(hash, 42); /* Arbitrary number. */
1300 /* Hashes 'flow' based on its L2 through L4 protocol information. */
1302 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
1307 struct in6_addr ipv6_addr;
1312 uint8_t eth_addr[ETH_ADDR_LEN];
1318 memset(&fields, 0, sizeof fields);
1319 for (i = 0; i < ETH_ADDR_LEN; i++) {
1320 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
1322 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
1323 fields.eth_type = flow->dl_type;
1325 /* UDP source and destination port are not taken into account because they
1326 * will not necessarily be symmetric in a bidirectional flow. */
1327 if (fields.eth_type == htons(ETH_TYPE_IP)) {
1328 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
1329 fields.ip_proto = flow->nw_proto;
1330 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1331 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1333 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
1334 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
1335 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
1336 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
1338 for (i=0; i<16; i++) {
1339 ipv6_addr[i] = a[i] ^ b[i];
1341 fields.ip_proto = flow->nw_proto;
1342 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1343 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1346 return jhash_bytes(&fields, sizeof fields, basis);
1349 /* Hashes 'flow' based on its L3 through L4 protocol information */
1351 flow_hash_symmetric_l3l4(const struct flow *flow, uint32_t basis,
1354 uint32_t hash = basis;
1356 /* UDP source and destination port are also taken into account. */
1357 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1358 hash = hash_add(hash,
1359 (OVS_FORCE uint32_t) (flow->nw_src ^ flow->nw_dst));
1360 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1361 /* IPv6 addresses are 64-bit aligned inside struct flow. */
1362 const uint64_t *a = ALIGNED_CAST(uint64_t *, flow->ipv6_src.s6_addr);
1363 const uint64_t *b = ALIGNED_CAST(uint64_t *, flow->ipv6_dst.s6_addr);
1365 for (int i = 0; i < 4; i++) {
1366 hash = hash_add64(hash, a[i] ^ b[i]);
1369 /* Cannot hash non-IP flows */
1373 hash = hash_add(hash, flow->nw_proto);
1374 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP ||
1375 (inc_udp_ports && flow->nw_proto == IPPROTO_UDP)) {
1376 hash = hash_add(hash,
1377 (OVS_FORCE uint16_t) (flow->tp_src ^ flow->tp_dst));
1380 return hash_finish(hash, basis);
1383 /* Initialize a flow with random fields that matter for nx_hash_fields. */
1385 flow_random_hash_fields(struct flow *flow)
1387 uint16_t rnd = random_uint16();
1389 /* Initialize to all zeros. */
1390 memset(flow, 0, sizeof *flow);
1392 eth_addr_random(flow->dl_src);
1393 eth_addr_random(flow->dl_dst);
1395 flow->vlan_tci = (OVS_FORCE ovs_be16) (random_uint16() & VLAN_VID_MASK);
1397 /* Make most of the random flows IPv4, some IPv6, and rest random. */
1398 flow->dl_type = rnd < 0x8000 ? htons(ETH_TYPE_IP) :
1399 rnd < 0xc000 ? htons(ETH_TYPE_IPV6) : (OVS_FORCE ovs_be16)rnd;
1401 if (dl_type_is_ip_any(flow->dl_type)) {
1402 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1403 flow->nw_src = (OVS_FORCE ovs_be32)random_uint32();
1404 flow->nw_dst = (OVS_FORCE ovs_be32)random_uint32();
1406 random_bytes(&flow->ipv6_src, sizeof flow->ipv6_src);
1407 random_bytes(&flow->ipv6_dst, sizeof flow->ipv6_dst);
1409 /* Make most of IP flows TCP, some UDP or SCTP, and rest random. */
1410 rnd = random_uint16();
1411 flow->nw_proto = rnd < 0x8000 ? IPPROTO_TCP :
1412 rnd < 0xc000 ? IPPROTO_UDP :
1413 rnd < 0xd000 ? IPPROTO_SCTP : (uint8_t)rnd;
1414 if (flow->nw_proto == IPPROTO_TCP ||
1415 flow->nw_proto == IPPROTO_UDP ||
1416 flow->nw_proto == IPPROTO_SCTP) {
1417 flow->tp_src = (OVS_FORCE ovs_be16)random_uint16();
1418 flow->tp_dst = (OVS_FORCE ovs_be16)random_uint16();
1423 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
1425 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
1426 enum nx_hash_fields fields)
1429 case NX_HASH_FIELDS_ETH_SRC:
1430 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1433 case NX_HASH_FIELDS_SYMMETRIC_L4:
1434 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1435 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1436 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1437 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1438 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1439 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1440 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1441 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1443 if (is_ip_any(flow)) {
1444 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1445 flow_unwildcard_tp_ports(flow, wc);
1447 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1450 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1451 if (is_ip_any(flow) && flow->nw_proto == IPPROTO_UDP) {
1452 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1453 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1456 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1457 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1458 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1459 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1460 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1461 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1462 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1464 break; /* non-IP flow */
1467 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1468 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP) {
1469 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1470 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1479 /* Hashes the portions of 'flow' designated by 'fields'. */
1481 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
1486 case NX_HASH_FIELDS_ETH_SRC:
1487 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
1489 case NX_HASH_FIELDS_SYMMETRIC_L4:
1490 return flow_hash_symmetric_l4(flow, basis);
1492 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1493 return flow_hash_symmetric_l3l4(flow, basis, false);
1495 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1496 return flow_hash_symmetric_l3l4(flow, basis, true);
1503 /* Returns a string representation of 'fields'. */
1505 flow_hash_fields_to_str(enum nx_hash_fields fields)
1508 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
1509 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
1510 case NX_HASH_FIELDS_SYMMETRIC_L3L4: return "symmetric_l3l4";
1511 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP: return "symmetric_l3l4+udp";
1512 default: return "<unknown>";
1516 /* Returns true if the value of 'fields' is supported. Otherwise false. */
1518 flow_hash_fields_valid(enum nx_hash_fields fields)
1520 return fields == NX_HASH_FIELDS_ETH_SRC
1521 || fields == NX_HASH_FIELDS_SYMMETRIC_L4
1522 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4
1523 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP;
1526 /* Returns a hash value for the bits of 'flow' that are active based on
1527 * 'wc', given 'basis'. */
1529 flow_hash_in_wildcards(const struct flow *flow,
1530 const struct flow_wildcards *wc, uint32_t basis)
1532 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1533 const uint64_t *flow_u64 = (const uint64_t *) flow;
1538 for (i = 0; i < FLOW_U64S; i++) {
1539 hash = hash_add64(hash, flow_u64[i] & wc_u64[i]);
1541 return hash_finish(hash, 8 * FLOW_U64S);
1544 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1545 * OpenFlow 1.0 "dl_vlan" value:
1547 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
1548 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
1549 * 'flow' previously matched packets without a VLAN header).
1551 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
1552 * without a VLAN tag.
1554 * - Other values of 'vid' should not be used. */
1556 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
1558 if (vid == htons(OFP10_VLAN_NONE)) {
1559 flow->vlan_tci = htons(0);
1561 vid &= htons(VLAN_VID_MASK);
1562 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1563 flow->vlan_tci |= htons(VLAN_CFI) | vid;
1567 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1568 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
1571 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
1573 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
1574 flow->vlan_tci &= ~mask;
1575 flow->vlan_tci |= vid & mask;
1578 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
1581 * This function has no effect on the VLAN ID that 'flow' matches.
1583 * After calling this function, 'flow' will not match packets without a VLAN
1586 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
1589 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1590 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
1593 /* Returns the number of MPLS LSEs present in 'flow'
1595 * Returns 0 if the 'dl_type' of 'flow' is not an MPLS ethernet type.
1596 * Otherwise traverses 'flow''s MPLS label stack stopping at the
1597 * first entry that has the BoS bit set. If no such entry exists then
1598 * the maximum number of LSEs that can be stored in 'flow' is returned.
1601 flow_count_mpls_labels(const struct flow *flow, struct flow_wildcards *wc)
1603 /* dl_type is always masked. */
1604 if (eth_type_mpls(flow->dl_type)) {
1609 for (i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1611 wc->masks.mpls_lse[i] |= htonl(MPLS_BOS_MASK);
1613 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1616 if (flow->mpls_lse[i]) {
1626 /* Returns the number consecutive of MPLS LSEs, starting at the
1627 * innermost LSE, that are common in 'a' and 'b'.
1629 * 'an' must be flow_count_mpls_labels(a).
1630 * 'bn' must be flow_count_mpls_labels(b).
1633 flow_count_common_mpls_labels(const struct flow *a, int an,
1634 const struct flow *b, int bn,
1635 struct flow_wildcards *wc)
1637 int min_n = MIN(an, bn);
1642 int a_last = an - 1;
1643 int b_last = bn - 1;
1646 for (i = 0; i < min_n; i++) {
1648 wc->masks.mpls_lse[a_last - i] = OVS_BE32_MAX;
1649 wc->masks.mpls_lse[b_last - i] = OVS_BE32_MAX;
1651 if (a->mpls_lse[a_last - i] != b->mpls_lse[b_last - i]) {
1662 /* Adds a new outermost MPLS label to 'flow' and changes 'flow''s Ethernet type
1663 * to 'mpls_eth_type', which must be an MPLS Ethertype.
1665 * If the new label is the first MPLS label in 'flow', it is generated as;
1667 * - label: 2, if 'flow' is IPv6, otherwise 0.
1669 * - TTL: IPv4 or IPv6 TTL, if present and nonzero, otherwise 64.
1671 * - TC: IPv4 or IPv6 TOS, if present, otherwise 0.
1675 * If the new label is the second or later label MPLS label in 'flow', it is
1678 * - label: Copied from outer label.
1680 * - TTL: Copied from outer label.
1682 * - TC: Copied from outer label.
1686 * 'n' must be flow_count_mpls_labels(flow). 'n' must be less than
1687 * FLOW_MAX_MPLS_LABELS (because otherwise flow->mpls_lse[] would overflow).
1690 flow_push_mpls(struct flow *flow, int n, ovs_be16 mpls_eth_type,
1691 struct flow_wildcards *wc)
1693 ovs_assert(eth_type_mpls(mpls_eth_type));
1694 ovs_assert(n < FLOW_MAX_MPLS_LABELS);
1700 memset(&wc->masks.mpls_lse, 0xff, sizeof *wc->masks.mpls_lse * n);
1702 for (i = n; i >= 1; i--) {
1703 flow->mpls_lse[i] = flow->mpls_lse[i - 1];
1705 flow->mpls_lse[0] = (flow->mpls_lse[1] & htonl(~MPLS_BOS_MASK));
1707 int label = 0; /* IPv4 Explicit Null. */
1711 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1715 if (is_ip_any(flow)) {
1716 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1718 wc->masks.nw_tos |= IP_DSCP_MASK;
1719 wc->masks.nw_ttl = 0xff;
1727 flow->mpls_lse[0] = set_mpls_lse_values(ttl, tc, 1, htonl(label));
1729 /* Clear all L3 and L4 fields and dp_hash. */
1730 BUILD_ASSERT(FLOW_WC_SEQ == 32);
1731 memset((char *) flow + FLOW_SEGMENT_2_ENDS_AT, 0,
1732 sizeof(struct flow) - FLOW_SEGMENT_2_ENDS_AT);
1735 flow->dl_type = mpls_eth_type;
1738 /* Tries to remove the outermost MPLS label from 'flow'. Returns true if
1739 * successful, false otherwise. On success, sets 'flow''s Ethernet type to
1742 * 'n' must be flow_count_mpls_labels(flow). */
1744 flow_pop_mpls(struct flow *flow, int n, ovs_be16 eth_type,
1745 struct flow_wildcards *wc)
1750 /* Nothing to pop. */
1752 } else if (n == FLOW_MAX_MPLS_LABELS) {
1754 wc->masks.mpls_lse[n - 1] |= htonl(MPLS_BOS_MASK);
1756 if (!(flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK))) {
1757 /* Can't pop because don't know what to fill in mpls_lse[n - 1]. */
1763 memset(&wc->masks.mpls_lse[1], 0xff,
1764 sizeof *wc->masks.mpls_lse * (n - 1));
1766 for (i = 1; i < n; i++) {
1767 flow->mpls_lse[i - 1] = flow->mpls_lse[i];
1769 flow->mpls_lse[n - 1] = 0;
1770 flow->dl_type = eth_type;
1774 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
1775 * as an OpenFlow 1.1 "mpls_label" value. */
1777 flow_set_mpls_label(struct flow *flow, int idx, ovs_be32 label)
1779 set_mpls_lse_label(&flow->mpls_lse[idx], label);
1782 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
1785 flow_set_mpls_ttl(struct flow *flow, int idx, uint8_t ttl)
1787 set_mpls_lse_ttl(&flow->mpls_lse[idx], ttl);
1790 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
1793 flow_set_mpls_tc(struct flow *flow, int idx, uint8_t tc)
1795 set_mpls_lse_tc(&flow->mpls_lse[idx], tc);
1798 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
1800 flow_set_mpls_bos(struct flow *flow, int idx, uint8_t bos)
1802 set_mpls_lse_bos(&flow->mpls_lse[idx], bos);
1805 /* Sets the entire MPLS LSE. */
1807 flow_set_mpls_lse(struct flow *flow, int idx, ovs_be32 lse)
1809 flow->mpls_lse[idx] = lse;
1813 flow_compose_l4(struct dp_packet *p, const struct flow *flow)
1817 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
1818 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1819 if (flow->nw_proto == IPPROTO_TCP) {
1820 struct tcp_header *tcp;
1822 l4_len = sizeof *tcp;
1823 tcp = dp_packet_put_zeros(p, l4_len);
1824 tcp->tcp_src = flow->tp_src;
1825 tcp->tcp_dst = flow->tp_dst;
1826 tcp->tcp_ctl = TCP_CTL(ntohs(flow->tcp_flags), 5);
1827 } else if (flow->nw_proto == IPPROTO_UDP) {
1828 struct udp_header *udp;
1830 l4_len = sizeof *udp;
1831 udp = dp_packet_put_zeros(p, l4_len);
1832 udp->udp_src = flow->tp_src;
1833 udp->udp_dst = flow->tp_dst;
1834 } else if (flow->nw_proto == IPPROTO_SCTP) {
1835 struct sctp_header *sctp;
1837 l4_len = sizeof *sctp;
1838 sctp = dp_packet_put_zeros(p, l4_len);
1839 sctp->sctp_src = flow->tp_src;
1840 sctp->sctp_dst = flow->tp_dst;
1841 } else if (flow->nw_proto == IPPROTO_ICMP) {
1842 struct icmp_header *icmp;
1844 l4_len = sizeof *icmp;
1845 icmp = dp_packet_put_zeros(p, l4_len);
1846 icmp->icmp_type = ntohs(flow->tp_src);
1847 icmp->icmp_code = ntohs(flow->tp_dst);
1848 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
1849 } else if (flow->nw_proto == IPPROTO_IGMP) {
1850 struct igmp_header *igmp;
1852 l4_len = sizeof *igmp;
1853 igmp = dp_packet_put_zeros(p, l4_len);
1854 igmp->igmp_type = ntohs(flow->tp_src);
1855 igmp->igmp_code = ntohs(flow->tp_dst);
1856 put_16aligned_be32(&igmp->group, flow->igmp_group_ip4);
1857 igmp->igmp_csum = csum(igmp, IGMP_HEADER_LEN);
1858 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
1859 struct icmp6_hdr *icmp;
1861 l4_len = sizeof *icmp;
1862 icmp = dp_packet_put_zeros(p, l4_len);
1863 icmp->icmp6_type = ntohs(flow->tp_src);
1864 icmp->icmp6_code = ntohs(flow->tp_dst);
1866 if (icmp->icmp6_code == 0 &&
1867 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
1868 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
1869 struct in6_addr *nd_target;
1870 struct nd_opt_hdr *nd_opt;
1872 l4_len += sizeof *nd_target;
1873 nd_target = dp_packet_put_zeros(p, sizeof *nd_target);
1874 *nd_target = flow->nd_target;
1876 if (!eth_addr_is_zero(flow->arp_sha)) {
1878 nd_opt = dp_packet_put_zeros(p, 8);
1879 nd_opt->nd_opt_len = 1;
1880 nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
1881 memcpy(nd_opt + 1, flow->arp_sha, ETH_ADDR_LEN);
1883 if (!eth_addr_is_zero(flow->arp_tha)) {
1885 nd_opt = dp_packet_put_zeros(p, 8);
1886 nd_opt->nd_opt_len = 1;
1887 nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR;
1888 memcpy(nd_opt + 1, flow->arp_tha, ETH_ADDR_LEN);
1891 icmp->icmp6_cksum = (OVS_FORCE uint16_t)
1892 csum(icmp, (char *)dp_packet_tail(p) - (char *)icmp);
1898 /* Puts into 'b' a packet that flow_extract() would parse as having the given
1901 * (This is useful only for testing, obviously, and the packet isn't really
1902 * valid. It hasn't got some checksums filled in, for one, and lots of fields
1903 * are just zeroed.) */
1905 flow_compose(struct dp_packet *p, const struct flow *flow)
1909 /* eth_compose() sets l3 pointer and makes sure it is 32-bit aligned. */
1910 eth_compose(p, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
1911 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
1912 struct eth_header *eth = dp_packet_l2(p);
1913 eth->eth_type = htons(dp_packet_size(p));
1917 if (flow->vlan_tci & htons(VLAN_CFI)) {
1918 eth_push_vlan(p, htons(ETH_TYPE_VLAN), flow->vlan_tci);
1921 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1922 struct ip_header *ip;
1924 ip = dp_packet_put_zeros(p, sizeof *ip);
1925 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
1926 ip->ip_tos = flow->nw_tos;
1927 ip->ip_ttl = flow->nw_ttl;
1928 ip->ip_proto = flow->nw_proto;
1929 put_16aligned_be32(&ip->ip_src, flow->nw_src);
1930 put_16aligned_be32(&ip->ip_dst, flow->nw_dst);
1932 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
1933 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
1934 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
1935 ip->ip_frag_off |= htons(100);
1939 dp_packet_set_l4(p, dp_packet_tail(p));
1941 l4_len = flow_compose_l4(p, flow);
1943 ip = dp_packet_l3(p);
1944 ip->ip_tot_len = htons(p->l4_ofs - p->l3_ofs + l4_len);
1945 ip->ip_csum = csum(ip, sizeof *ip);
1946 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1947 struct ovs_16aligned_ip6_hdr *nh;
1949 nh = dp_packet_put_zeros(p, sizeof *nh);
1950 put_16aligned_be32(&nh->ip6_flow, htonl(6 << 28) |
1951 htonl(flow->nw_tos << 20) | flow->ipv6_label);
1952 nh->ip6_hlim = flow->nw_ttl;
1953 nh->ip6_nxt = flow->nw_proto;
1955 memcpy(&nh->ip6_src, &flow->ipv6_src, sizeof(nh->ip6_src));
1956 memcpy(&nh->ip6_dst, &flow->ipv6_dst, sizeof(nh->ip6_dst));
1958 dp_packet_set_l4(p, dp_packet_tail(p));
1960 l4_len = flow_compose_l4(p, flow);
1962 nh = dp_packet_l3(p);
1963 nh->ip6_plen = htons(l4_len);
1964 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1965 flow->dl_type == htons(ETH_TYPE_RARP)) {
1966 struct arp_eth_header *arp;
1968 arp = dp_packet_put_zeros(p, sizeof *arp);
1969 dp_packet_set_l3(p, arp);
1970 arp->ar_hrd = htons(1);
1971 arp->ar_pro = htons(ETH_TYPE_IP);
1972 arp->ar_hln = ETH_ADDR_LEN;
1974 arp->ar_op = htons(flow->nw_proto);
1976 if (flow->nw_proto == ARP_OP_REQUEST ||
1977 flow->nw_proto == ARP_OP_REPLY) {
1978 put_16aligned_be32(&arp->ar_spa, flow->nw_src);
1979 put_16aligned_be32(&arp->ar_tpa, flow->nw_dst);
1980 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
1981 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
1985 if (eth_type_mpls(flow->dl_type)) {
1988 p->l2_5_ofs = p->l3_ofs;
1989 for (n = 1; n < FLOW_MAX_MPLS_LABELS; n++) {
1990 if (flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK)) {
1995 push_mpls(p, flow->dl_type, flow->mpls_lse[--n]);
2000 /* Compressed flow. */
2003 miniflow_n_values(const struct miniflow *flow)
2005 return count_1bits(flow->map);
2008 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
2009 * the caller. The caller must have already computed 'map' properly
2010 * to indicate the significant uint64_t elements of 'src'.
2012 * Normally the significant elements are the ones that are non-zero. However,
2013 * when a miniflow is initialized from a (mini)mask, the values can be zeroes,
2014 * so that the flow and mask always have the same maps.
2016 * This function always dynamically allocates a miniflow with the correct
2017 * amount of inline storage and copies the uint64_t elements of 'src' indicated
2018 * by 'map' into it. */
2019 static struct miniflow *
2020 miniflow_init__(const struct flow *src, uint64_t map)
2022 const uint64_t *src_u64 = (const uint64_t *) src;
2023 struct miniflow *dst = xmalloc(sizeof *dst
2024 + MINIFLOW_VALUES_SIZE(count_1bits(map)));
2025 uint64_t *dst_u64 = dst->values;
2028 COVERAGE_INC(miniflow_malloc);
2031 MAP_FOR_EACH_INDEX(idx, map) {
2032 *dst_u64++ = src_u64[idx];
2037 /* Returns a miniflow copy of 'src'. The caller must eventually free the
2038 * returned miniflow with free(). */
2040 miniflow_create(const struct flow *src)
2042 const uint64_t *src_u64 = (const uint64_t *) src;
2046 /* Initialize dst->map, counting the number of nonzero elements. */
2049 for (i = 0; i < FLOW_U64S; i++) {
2051 map |= UINT64_C(1) << i;
2055 return miniflow_init__(src, map);
2058 /* Returns a copy of 'src', using 'mask->map'. The caller must eventually free
2059 * the returned miniflow with free(). */
2061 miniflow_create_with_minimask(const struct flow *src,
2062 const struct minimask *mask)
2064 return miniflow_init__(src, mask->masks.map);
2067 /* Returns a copy of 'src'. The caller must eventually free the returned
2068 * miniflow with free(). */
2070 miniflow_clone(const struct miniflow *src)
2072 struct miniflow *dst;
2073 int n = miniflow_n_values(src);
2075 COVERAGE_INC(miniflow_malloc);
2076 dst = xmalloc(sizeof *dst + MINIFLOW_VALUES_SIZE(n));
2077 miniflow_clone_inline(dst, src, n);
2081 /* Initializes 'dst' as a copy of 'src'. The caller must have allocated
2082 * 'dst' to have inline space for 'n_values' data in 'src'. */
2084 miniflow_clone_inline(struct miniflow *dst, const struct miniflow *src,
2087 dst->map = src->map;
2088 memcpy(dst->values, src->values, MINIFLOW_VALUES_SIZE(n_values));
2091 /* Initializes 'dst' as a copy of 'src'. */
2093 miniflow_expand(const struct miniflow *src, struct flow *dst)
2095 memset(dst, 0, sizeof *dst);
2096 flow_union_with_miniflow(dst, src);
2099 /* Returns true if 'a' and 'b' are equal miniflows, false otherwise. */
2101 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
2103 const uint64_t *ap = a->values;
2104 const uint64_t *bp = b->values;
2106 if (OVS_LIKELY(a->map == b->map)) {
2107 int count = miniflow_n_values(a);
2109 return !memcmp(ap, bp, count * sizeof *ap);
2113 for (map = a->map | b->map; map; map = zero_rightmost_1bit(map)) {
2114 uint64_t bit = rightmost_1bit(map);
2116 if ((a->map & bit ? *ap++ : 0) != (b->map & bit ? *bp++ : 0)) {
2125 /* Returns false if 'a' and 'b' differ at the places where there are 1-bits
2126 * in 'mask', true otherwise. */
2128 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
2129 const struct minimask *mask)
2131 const uint64_t *p = mask->masks.values;
2134 MAP_FOR_EACH_INDEX(idx, mask->masks.map) {
2135 if ((miniflow_get(a, idx) ^ miniflow_get(b, idx)) & *p++) {
2143 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
2144 * in 'mask', false if they differ. */
2146 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
2147 const struct minimask *mask)
2149 const uint64_t *b_u64 = (const uint64_t *) b;
2150 const uint64_t *p = mask->masks.values;
2153 MAP_FOR_EACH_INDEX(idx, mask->masks.map) {
2154 if ((miniflow_get(a, idx) ^ b_u64[idx]) & *p++) {
2163 /* Returns a minimask copy of 'wc'. The caller must eventually free the
2164 * returned minimask with free(). */
2166 minimask_create(const struct flow_wildcards *wc)
2168 return (struct minimask *)miniflow_create(&wc->masks);
2171 /* Returns a copy of 'src'. The caller must eventually free the returned
2172 * minimask with free(). */
2174 minimask_clone(const struct minimask *src)
2176 return (struct minimask *)miniflow_clone(&src->masks);
2179 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
2181 * The caller must provide room for FLOW_U64S "uint64_t"s in 'storage', which
2182 * must follow '*dst_' in memory, for use by 'dst_'. The caller must *not*
2183 * free 'dst_' free(). */
2185 minimask_combine(struct minimask *dst_,
2186 const struct minimask *a_, const struct minimask *b_,
2187 uint64_t storage[FLOW_U64S])
2189 struct miniflow *dst = &dst_->masks;
2190 uint64_t *dst_values = storage;
2191 const struct miniflow *a = &a_->masks;
2192 const struct miniflow *b = &b_->masks;
2196 MAP_FOR_EACH_INDEX(idx, a->map & b->map) {
2197 /* Both 'a' and 'b' have non-zero data at 'idx'. */
2198 uint64_t mask = miniflow_get__(a, idx) & miniflow_get__(b, idx);
2201 dst->map |= UINT64_C(1) << idx;
2202 *dst_values++ = mask;
2207 /* Initializes 'wc' as a copy of 'mask'. */
2209 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
2211 miniflow_expand(&mask->masks, &wc->masks);
2214 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise.
2215 * Minimasks may not have zero data values, so for the minimasks to be the
2216 * same, they need to have the same map and the same data values. */
2218 minimask_equal(const struct minimask *a, const struct minimask *b)
2220 return a->masks.map == b->masks.map &&
2221 !memcmp(a->masks.values, b->masks.values,
2222 count_1bits(a->masks.map) * sizeof *a->masks.values);
2225 /* Returns true if at least one bit matched by 'b' is wildcarded by 'a',
2226 * false otherwise. */
2228 minimask_has_extra(const struct minimask *a, const struct minimask *b)
2230 const uint64_t *ap = a->masks.values;
2231 const uint64_t *bp = b->masks.values;
2234 MAP_FOR_EACH_INDEX(idx, b->masks.map) {
2235 uint64_t b_u64 = *bp++;
2237 /* 'b_u64' is non-zero, check if the data in 'a' is either zero
2238 * or misses some of the bits in 'b_u64'. */
2239 if (!(a->masks.map & (UINT64_C(1) << idx))
2240 || ((miniflow_values_get__(ap, a->masks.map, idx) & b_u64)
2242 return true; /* 'a' wildcards some bits 'b' doesn't. */