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. */
115 struct miniflow maps;
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 != 33)
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_set_map(MF, OFS) \
136 if ((OFS) < FLOW_TNL_U64S) { \
137 MINIFLOW_ASSERT(!(MF.maps.tnl_map & (UINT64_MAX << (OFS))) \
138 && !MF.maps.pkt_map); \
139 MF.maps.tnl_map |= UINT64_C(1) << (OFS); \
141 MINIFLOW_ASSERT(!(MF.maps.pkt_map \
142 & UINT64_MAX << ((OFS) - FLOW_TNL_U64S))); \
143 MF.maps.pkt_map |= UINT64_C(1) << ((OFS) - FLOW_TNL_U64S); \
146 #define miniflow_assert_in_map(MF, OFS) \
147 if ((OFS) < FLOW_TNL_U64S) { \
148 MINIFLOW_ASSERT(MF.maps.tnl_map & UINT64_C(1) << (OFS) \
149 && !(MF.maps.tnl_map & UINT64_MAX << ((OFS) + 1)) \
150 && !MF.maps.pkt_map); \
152 MINIFLOW_ASSERT(MF.maps.pkt_map & UINT64_C(1) << ((OFS) - FLOW_TNL_U64S) \
153 && !(MF.maps.pkt_map & UINT64_MAX << ((OFS) - FLOW_TNL_U64S + 1))); \
156 #define miniflow_push_uint64_(MF, OFS, VALUE) \
158 MINIFLOW_ASSERT(MF.data < MF.end && (OFS) % 8 == 0); \
159 *MF.data++ = VALUE; \
160 miniflow_set_map(MF, OFS / 8); \
163 #define miniflow_push_be64_(MF, OFS, VALUE) \
164 miniflow_push_uint64_(MF, OFS, (OVS_FORCE uint64_t)(VALUE))
166 #define miniflow_push_uint32_(MF, OFS, VALUE) \
168 MINIFLOW_ASSERT(MF.data < MF.end); \
170 if ((OFS) % 8 == 0) { \
171 miniflow_set_map(MF, OFS / 8); \
172 *(uint32_t *)MF.data = VALUE; \
173 } else if ((OFS) % 8 == 4) { \
174 miniflow_assert_in_map(MF, OFS / 8); \
175 *((uint32_t *)MF.data + 1) = VALUE; \
180 #define miniflow_push_be32_(MF, OFS, VALUE) \
181 miniflow_push_uint32_(MF, OFS, (OVS_FORCE uint32_t)(VALUE))
183 #define miniflow_push_uint16_(MF, OFS, VALUE) \
185 MINIFLOW_ASSERT(MF.data < MF.end); \
187 if ((OFS) % 8 == 0) { \
188 miniflow_set_map(MF, OFS / 8); \
189 *(uint16_t *)MF.data = VALUE; \
190 } else if ((OFS) % 8 == 2) { \
191 miniflow_assert_in_map(MF, OFS / 8); \
192 *((uint16_t *)MF.data + 1) = VALUE; \
193 } else if ((OFS) % 8 == 4) { \
194 miniflow_assert_in_map(MF, OFS / 8); \
195 *((uint16_t *)MF.data + 2) = VALUE; \
196 } else if ((OFS) % 8 == 6) { \
197 miniflow_assert_in_map(MF, OFS / 8); \
198 *((uint16_t *)MF.data + 3) = VALUE; \
203 #define miniflow_pad_to_64_(MF, OFS) \
205 MINIFLOW_ASSERT((OFS) % 8 != 0); \
206 miniflow_assert_in_map(MF, OFS / 8); \
208 memset((uint8_t *)MF.data + (OFS) % 8, 0, 8 - (OFS) % 8); \
212 #define miniflow_push_be16_(MF, OFS, VALUE) \
213 miniflow_push_uint16_(MF, OFS, (OVS_FORCE uint16_t)VALUE);
215 #define miniflow_set_maps(MF, OFS, N_WORDS) \
217 size_t ofs = (OFS); \
218 size_t n_words = (N_WORDS); \
219 uint64_t n_words_mask = UINT64_MAX >> (64 - n_words); \
221 MINIFLOW_ASSERT(n_words && MF.data + n_words <= MF.end); \
222 if (ofs < FLOW_TNL_U64S) { \
223 MINIFLOW_ASSERT(!(MF.maps.tnl_map & UINT64_MAX << ofs) \
224 && !MF.maps.pkt_map); \
225 MF.maps.tnl_map |= n_words_mask << ofs; \
226 if (n_words > FLOW_TNL_U64S - ofs) { \
227 MF.maps.pkt_map |= n_words_mask >> (FLOW_TNL_U64S - ofs); \
230 ofs -= FLOW_TNL_U64S; \
231 MINIFLOW_ASSERT(!(MF.maps.pkt_map & (UINT64_MAX << ofs))); \
232 MF.maps.pkt_map |= n_words_mask << ofs; \
236 /* Data at 'valuep' may be unaligned. */
237 #define miniflow_push_words_(MF, OFS, VALUEP, N_WORDS) \
239 MINIFLOW_ASSERT((OFS) % 8 == 0); \
240 miniflow_set_maps(MF, (OFS) / 8, (N_WORDS)); \
241 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof *MF.data); \
242 MF.data += (N_WORDS); \
245 /* Push 32-bit words padded to 64-bits. */
246 #define miniflow_push_words_32_(MF, OFS, VALUEP, N_WORDS) \
248 miniflow_set_maps(MF, (OFS) / 8, DIV_ROUND_UP(N_WORDS, 2)); \
249 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof(uint32_t)); \
250 MF.data += DIV_ROUND_UP(N_WORDS, 2); \
251 if ((N_WORDS) & 1) { \
252 *((uint32_t *)MF.data - 1) = 0; \
256 /* Data at 'valuep' may be unaligned. */
257 /* MACs start 64-aligned, and must be followed by other data or padding. */
258 #define miniflow_push_macs_(MF, OFS, VALUEP) \
260 miniflow_set_maps(MF, (OFS) / 8, 2); \
261 memcpy(MF.data, (VALUEP), 2 * ETH_ADDR_LEN); \
262 MF.data += 1; /* First word only. */ \
265 #define miniflow_push_uint32(MF, FIELD, VALUE) \
266 miniflow_push_uint32_(MF, offsetof(struct flow, FIELD), VALUE)
268 #define miniflow_push_be32(MF, FIELD, VALUE) \
269 miniflow_push_be32_(MF, offsetof(struct flow, FIELD), VALUE)
271 #define miniflow_push_uint16(MF, FIELD, VALUE) \
272 miniflow_push_uint16_(MF, offsetof(struct flow, FIELD), VALUE)
274 #define miniflow_push_be16(MF, FIELD, VALUE) \
275 miniflow_push_be16_(MF, offsetof(struct flow, FIELD), VALUE)
277 #define miniflow_pad_to_64(MF, FIELD) \
278 miniflow_pad_to_64_(MF, offsetof(struct flow, FIELD))
280 #define miniflow_push_words(MF, FIELD, VALUEP, N_WORDS) \
281 miniflow_push_words_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
283 #define miniflow_push_words_32(MF, FIELD, VALUEP, N_WORDS) \
284 miniflow_push_words_32_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
286 #define miniflow_push_macs(MF, FIELD, VALUEP) \
287 miniflow_push_macs_(MF, offsetof(struct flow, FIELD), VALUEP)
289 /* Pulls the MPLS headers at '*datap' and returns the count of them. */
291 parse_mpls(const void **datap, size_t *sizep)
293 const struct mpls_hdr *mh;
296 while ((mh = data_try_pull(datap, sizep, sizeof *mh))) {
298 if (mh->mpls_lse.lo & htons(1 << MPLS_BOS_SHIFT)) {
302 return MIN(count, FLOW_MAX_MPLS_LABELS);
305 static inline ovs_be16
306 parse_vlan(const void **datap, size_t *sizep)
308 const struct eth_header *eth = *datap;
311 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
315 data_pull(datap, sizep, ETH_ADDR_LEN * 2);
317 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
318 if (OVS_LIKELY(*sizep
319 >= sizeof(struct qtag_prefix) + sizeof(ovs_be16))) {
320 const struct qtag_prefix *qp = data_pull(datap, sizep, sizeof *qp);
321 return qp->tci | htons(VLAN_CFI);
327 static inline ovs_be16
328 parse_ethertype(const void **datap, size_t *sizep)
330 const struct llc_snap_header *llc;
333 proto = *(ovs_be16 *) data_pull(datap, sizep, sizeof proto);
334 if (OVS_LIKELY(ntohs(proto) >= ETH_TYPE_MIN)) {
338 if (OVS_UNLIKELY(*sizep < sizeof *llc)) {
339 return htons(FLOW_DL_TYPE_NONE);
343 if (OVS_UNLIKELY(llc->llc.llc_dsap != LLC_DSAP_SNAP
344 || llc->llc.llc_ssap != LLC_SSAP_SNAP
345 || llc->llc.llc_cntl != LLC_CNTL_SNAP
346 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
347 sizeof llc->snap.snap_org))) {
348 return htons(FLOW_DL_TYPE_NONE);
351 data_pull(datap, sizep, sizeof *llc);
353 if (OVS_LIKELY(ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN)) {
354 return llc->snap.snap_type;
357 return htons(FLOW_DL_TYPE_NONE);
361 parse_icmpv6(const void **datap, size_t *sizep, const struct icmp6_hdr *icmp,
362 const struct in6_addr **nd_target,
363 uint8_t arp_buf[2][ETH_ADDR_LEN])
365 if (icmp->icmp6_code == 0 &&
366 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
367 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
369 *nd_target = data_try_pull(datap, sizep, sizeof **nd_target);
370 if (OVS_UNLIKELY(!*nd_target)) {
374 while (*sizep >= 8) {
375 /* The minimum size of an option is 8 bytes, which also is
376 * the size of Ethernet link-layer options. */
377 const struct nd_opt_hdr *nd_opt = *datap;
378 int opt_len = nd_opt->nd_opt_len * 8;
380 if (!opt_len || opt_len > *sizep) {
384 /* Store the link layer address if the appropriate option is
385 * provided. It is considered an error if the same link
386 * layer option is specified twice. */
387 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
389 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[0]))) {
390 memcpy(arp_buf[0], nd_opt + 1, ETH_ADDR_LEN);
394 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
396 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[1]))) {
397 memcpy(arp_buf[1], nd_opt + 1, ETH_ADDR_LEN);
403 if (OVS_UNLIKELY(!data_try_pull(datap, sizep, opt_len))) {
415 /* Initializes 'flow' members from 'packet' and 'md'
417 * Initializes 'packet' header l2 pointer to the start of the Ethernet
418 * header, and the layer offsets as follows:
420 * - packet->l2_5_ofs to the start of the MPLS shim header, or UINT16_MAX
421 * when there is no MPLS shim header.
423 * - packet->l3_ofs to just past the Ethernet header, or just past the
424 * vlan_header if one is present, to the first byte of the payload of the
425 * Ethernet frame. UINT16_MAX if the frame is too short to contain an
428 * - packet->l4_ofs to just past the IPv4 header, if one is present and
429 * has at least the content used for the fields of interest for the flow,
430 * otherwise UINT16_MAX.
433 flow_extract(struct dp_packet *packet, struct flow *flow)
437 uint64_t buf[FLOW_U64S];
440 COVERAGE_INC(flow_extract);
442 miniflow_extract(packet, &m.mf);
443 miniflow_expand(&m.mf, flow);
446 /* Caller is responsible for initializing 'dst' with enough storage for
447 * FLOW_U64S * 8 bytes. */
449 miniflow_extract(struct dp_packet *packet, struct miniflow *dst)
451 const struct pkt_metadata *md = &packet->md;
452 const void *data = dp_packet_data(packet);
453 size_t size = dp_packet_size(packet);
454 uint64_t *values = miniflow_values(dst);
455 struct mf_ctx mf = { { 0, 0 }, values, values + FLOW_U64S };
458 uint8_t nw_frag, nw_tos, nw_ttl, nw_proto;
461 if (md->tunnel.ip_dst) {
462 miniflow_push_words(mf, tunnel, &md->tunnel,
463 offsetof(struct flow_tnl, metadata) /
465 if (md->tunnel.metadata.opt_map) {
466 miniflow_push_words(mf, tunnel.metadata, &md->tunnel.metadata,
467 sizeof md->tunnel.metadata / sizeof(uint64_t));
470 if (md->skb_priority || md->pkt_mark) {
471 miniflow_push_uint32(mf, skb_priority, md->skb_priority);
472 miniflow_push_uint32(mf, pkt_mark, md->pkt_mark);
474 miniflow_push_uint32(mf, dp_hash, md->dp_hash);
475 miniflow_push_uint32(mf, in_port, odp_to_u32(md->in_port.odp_port));
477 miniflow_push_uint32(mf, recirc_id, md->recirc_id);
478 miniflow_pad_to_64(mf, conj_id);
481 /* Initialize packet's layer pointer and offsets. */
483 dp_packet_reset_offsets(packet);
485 /* Must have full Ethernet header to proceed. */
486 if (OVS_UNLIKELY(size < sizeof(struct eth_header))) {
492 ASSERT_SEQUENTIAL(dl_dst, dl_src);
493 miniflow_push_macs(mf, dl_dst, data);
494 /* dl_type, vlan_tci. */
495 vlan_tci = parse_vlan(&data, &size);
496 dl_type = parse_ethertype(&data, &size);
497 miniflow_push_be16(mf, dl_type, dl_type);
498 miniflow_push_be16(mf, vlan_tci, vlan_tci);
502 if (OVS_UNLIKELY(eth_type_mpls(dl_type))) {
504 const void *mpls = data;
506 packet->l2_5_ofs = (char *)data - l2;
507 count = parse_mpls(&data, &size);
508 miniflow_push_words_32(mf, mpls_lse, mpls, count);
512 packet->l3_ofs = (char *)data - l2;
515 if (OVS_LIKELY(dl_type == htons(ETH_TYPE_IP))) {
516 const struct ip_header *nh = data;
520 if (OVS_UNLIKELY(size < IP_HEADER_LEN)) {
523 ip_len = IP_IHL(nh->ip_ihl_ver) * 4;
525 if (OVS_UNLIKELY(ip_len < IP_HEADER_LEN)) {
528 if (OVS_UNLIKELY(size < ip_len)) {
531 tot_len = ntohs(nh->ip_tot_len);
532 if (OVS_UNLIKELY(tot_len > size)) {
535 if (OVS_UNLIKELY(size - tot_len > UINT8_MAX)) {
538 dp_packet_set_l2_pad_size(packet, size - tot_len);
539 size = tot_len; /* Never pull padding. */
541 /* Push both source and destination address at once. */
542 miniflow_push_words(mf, nw_src, &nh->ip_src, 1);
544 miniflow_push_be32(mf, ipv6_label, 0); /* Padding for IPv4. */
548 nw_proto = nh->ip_proto;
549 if (OVS_UNLIKELY(IP_IS_FRAGMENT(nh->ip_frag_off))) {
550 nw_frag = FLOW_NW_FRAG_ANY;
551 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
552 nw_frag |= FLOW_NW_FRAG_LATER;
555 data_pull(&data, &size, ip_len);
556 } else if (dl_type == htons(ETH_TYPE_IPV6)) {
557 const struct ovs_16aligned_ip6_hdr *nh;
561 if (OVS_UNLIKELY(size < sizeof *nh)) {
564 nh = data_pull(&data, &size, sizeof *nh);
566 plen = ntohs(nh->ip6_plen);
567 if (OVS_UNLIKELY(plen > size)) {
570 /* Jumbo Payload option not supported yet. */
571 if (OVS_UNLIKELY(size - plen > UINT8_MAX)) {
574 dp_packet_set_l2_pad_size(packet, size - plen);
575 size = plen; /* Never pull padding. */
577 miniflow_push_words(mf, ipv6_src, &nh->ip6_src,
578 sizeof nh->ip6_src / 8);
579 miniflow_push_words(mf, ipv6_dst, &nh->ip6_dst,
580 sizeof nh->ip6_dst / 8);
582 tc_flow = get_16aligned_be32(&nh->ip6_flow);
584 ovs_be32 label = tc_flow & htonl(IPV6_LABEL_MASK);
585 miniflow_push_be32(mf, ipv6_label, label);
588 nw_tos = ntohl(tc_flow) >> 20;
589 nw_ttl = nh->ip6_hlim;
590 nw_proto = nh->ip6_nxt;
593 if (OVS_LIKELY((nw_proto != IPPROTO_HOPOPTS)
594 && (nw_proto != IPPROTO_ROUTING)
595 && (nw_proto != IPPROTO_DSTOPTS)
596 && (nw_proto != IPPROTO_AH)
597 && (nw_proto != IPPROTO_FRAGMENT))) {
598 /* It's either a terminal header (e.g., TCP, UDP) or one we
599 * don't understand. In either case, we're done with the
600 * packet, so use it to fill in 'nw_proto'. */
604 /* We only verify that at least 8 bytes of the next header are
605 * available, but many of these headers are longer. Ensure that
606 * accesses within the extension header are within those first 8
607 * bytes. All extension headers are required to be at least 8
609 if (OVS_UNLIKELY(size < 8)) {
613 if ((nw_proto == IPPROTO_HOPOPTS)
614 || (nw_proto == IPPROTO_ROUTING)
615 || (nw_proto == IPPROTO_DSTOPTS)) {
616 /* These headers, while different, have the fields we care
617 * about in the same location and with the same
619 const struct ip6_ext *ext_hdr = data;
620 nw_proto = ext_hdr->ip6e_nxt;
621 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
622 (ext_hdr->ip6e_len + 1) * 8))) {
625 } else if (nw_proto == IPPROTO_AH) {
626 /* A standard AH definition isn't available, but the fields
627 * we care about are in the same location as the generic
628 * option header--only the header length is calculated
630 const struct ip6_ext *ext_hdr = data;
631 nw_proto = ext_hdr->ip6e_nxt;
632 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
633 (ext_hdr->ip6e_len + 2) * 4))) {
636 } else if (nw_proto == IPPROTO_FRAGMENT) {
637 const struct ovs_16aligned_ip6_frag *frag_hdr = data;
639 nw_proto = frag_hdr->ip6f_nxt;
640 if (!data_try_pull(&data, &size, sizeof *frag_hdr)) {
644 /* We only process the first fragment. */
645 if (frag_hdr->ip6f_offlg != htons(0)) {
646 nw_frag = FLOW_NW_FRAG_ANY;
647 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
648 nw_frag |= FLOW_NW_FRAG_LATER;
649 nw_proto = IPPROTO_FRAGMENT;
656 if (dl_type == htons(ETH_TYPE_ARP) ||
657 dl_type == htons(ETH_TYPE_RARP)) {
658 uint8_t arp_buf[2][ETH_ADDR_LEN];
659 const struct arp_eth_header *arp = (const struct arp_eth_header *)
660 data_try_pull(&data, &size, ARP_ETH_HEADER_LEN);
662 if (OVS_LIKELY(arp) && OVS_LIKELY(arp->ar_hrd == htons(1))
663 && OVS_LIKELY(arp->ar_pro == htons(ETH_TYPE_IP))
664 && OVS_LIKELY(arp->ar_hln == ETH_ADDR_LEN)
665 && OVS_LIKELY(arp->ar_pln == 4)) {
666 miniflow_push_be32(mf, nw_src,
667 get_16aligned_be32(&arp->ar_spa));
668 miniflow_push_be32(mf, nw_dst,
669 get_16aligned_be32(&arp->ar_tpa));
671 /* We only match on the lower 8 bits of the opcode. */
672 if (OVS_LIKELY(ntohs(arp->ar_op) <= 0xff)) {
673 miniflow_push_be32(mf, ipv6_label, 0); /* Pad with ARP. */
674 miniflow_push_be32(mf, nw_frag, htonl(ntohs(arp->ar_op)));
677 /* Must be adjacent. */
678 ASSERT_SEQUENTIAL(arp_sha, arp_tha);
680 memcpy(arp_buf[0], arp->ar_sha, ETH_ADDR_LEN);
681 memcpy(arp_buf[1], arp->ar_tha, ETH_ADDR_LEN);
682 miniflow_push_macs(mf, arp_sha, arp_buf);
683 miniflow_pad_to_64(mf, tcp_flags);
689 packet->l4_ofs = (char *)data - l2;
690 miniflow_push_be32(mf, nw_frag,
691 BYTES_TO_BE32(nw_frag, nw_tos, nw_ttl, nw_proto));
693 if (OVS_LIKELY(!(nw_frag & FLOW_NW_FRAG_LATER))) {
694 if (OVS_LIKELY(nw_proto == IPPROTO_TCP)) {
695 if (OVS_LIKELY(size >= TCP_HEADER_LEN)) {
696 const struct tcp_header *tcp = data;
698 miniflow_push_be32(mf, arp_tha[2], 0);
699 miniflow_push_be32(mf, tcp_flags,
700 TCP_FLAGS_BE32(tcp->tcp_ctl));
701 miniflow_push_be16(mf, tp_src, tcp->tcp_src);
702 miniflow_push_be16(mf, tp_dst, tcp->tcp_dst);
703 miniflow_pad_to_64(mf, igmp_group_ip4);
705 } else if (OVS_LIKELY(nw_proto == IPPROTO_UDP)) {
706 if (OVS_LIKELY(size >= UDP_HEADER_LEN)) {
707 const struct udp_header *udp = data;
709 miniflow_push_be16(mf, tp_src, udp->udp_src);
710 miniflow_push_be16(mf, tp_dst, udp->udp_dst);
711 miniflow_pad_to_64(mf, igmp_group_ip4);
713 } else if (OVS_LIKELY(nw_proto == IPPROTO_SCTP)) {
714 if (OVS_LIKELY(size >= SCTP_HEADER_LEN)) {
715 const struct sctp_header *sctp = data;
717 miniflow_push_be16(mf, tp_src, sctp->sctp_src);
718 miniflow_push_be16(mf, tp_dst, sctp->sctp_dst);
719 miniflow_pad_to_64(mf, igmp_group_ip4);
721 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMP)) {
722 if (OVS_LIKELY(size >= ICMP_HEADER_LEN)) {
723 const struct icmp_header *icmp = data;
725 miniflow_push_be16(mf, tp_src, htons(icmp->icmp_type));
726 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp_code));
727 miniflow_pad_to_64(mf, igmp_group_ip4);
729 } else if (OVS_LIKELY(nw_proto == IPPROTO_IGMP)) {
730 if (OVS_LIKELY(size >= IGMP_HEADER_LEN)) {
731 const struct igmp_header *igmp = data;
733 miniflow_push_be16(mf, tp_src, htons(igmp->igmp_type));
734 miniflow_push_be16(mf, tp_dst, htons(igmp->igmp_code));
735 miniflow_push_be32(mf, igmp_group_ip4,
736 get_16aligned_be32(&igmp->group));
738 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMPV6)) {
739 if (OVS_LIKELY(size >= sizeof(struct icmp6_hdr))) {
740 const struct in6_addr *nd_target = NULL;
741 uint8_t arp_buf[2][ETH_ADDR_LEN];
742 const struct icmp6_hdr *icmp = data_pull(&data, &size,
744 memset(arp_buf, 0, sizeof arp_buf);
745 if (OVS_LIKELY(parse_icmpv6(&data, &size, icmp, &nd_target,
748 miniflow_push_words(mf, nd_target, nd_target,
749 sizeof *nd_target / 8);
751 miniflow_push_macs(mf, arp_sha, arp_buf);
752 miniflow_pad_to_64(mf, tcp_flags);
753 miniflow_push_be16(mf, tp_src, htons(icmp->icmp6_type));
754 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp6_code));
755 miniflow_pad_to_64(mf, igmp_group_ip4);
764 /* For every bit of a field that is wildcarded in 'wildcards', sets the
765 * corresponding bit in 'flow' to zero. */
767 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
769 uint64_t *flow_u64 = (uint64_t *) flow;
770 const uint64_t *wc_u64 = (const uint64_t *) &wildcards->masks;
773 for (i = 0; i < FLOW_U64S; i++) {
774 flow_u64[i] &= wc_u64[i];
779 flow_unwildcard_tp_ports(const struct flow *flow, struct flow_wildcards *wc)
781 if (flow->nw_proto != IPPROTO_ICMP) {
782 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
783 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
785 wc->masks.tp_src = htons(0xff);
786 wc->masks.tp_dst = htons(0xff);
790 /* Initializes 'flow_metadata' with the metadata found in 'flow'. */
792 flow_get_metadata(const struct flow *flow, struct match *flow_metadata)
796 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 33);
798 match_init_catchall(flow_metadata);
799 if (flow->tunnel.tun_id != htonll(0)) {
800 match_set_tun_id(flow_metadata, flow->tunnel.tun_id);
802 if (flow->tunnel.flags & FLOW_TNL_PUB_F_MASK) {
803 match_set_tun_flags(flow_metadata,
804 flow->tunnel.flags & FLOW_TNL_PUB_F_MASK);
806 if (flow->tunnel.ip_src != htonl(0)) {
807 match_set_tun_src(flow_metadata, flow->tunnel.ip_src);
809 if (flow->tunnel.ip_dst != htonl(0)) {
810 match_set_tun_dst(flow_metadata, flow->tunnel.ip_dst);
812 if (flow->tunnel.gbp_id != htons(0)) {
813 match_set_tun_gbp_id(flow_metadata, flow->tunnel.gbp_id);
815 if (flow->tunnel.gbp_flags) {
816 match_set_tun_gbp_flags(flow_metadata, flow->tunnel.gbp_flags);
818 tun_metadata_get_fmd(&flow->tunnel.metadata, flow_metadata);
819 if (flow->metadata != htonll(0)) {
820 match_set_metadata(flow_metadata, flow->metadata);
823 for (i = 0; i < FLOW_N_REGS; i++) {
825 match_set_reg(flow_metadata, i, flow->regs[i]);
829 if (flow->pkt_mark != 0) {
830 match_set_pkt_mark(flow_metadata, flow->pkt_mark);
833 match_set_in_port(flow_metadata, flow->in_port.ofp_port);
837 flow_to_string(const struct flow *flow)
839 struct ds ds = DS_EMPTY_INITIALIZER;
840 flow_format(&ds, flow);
845 flow_tun_flag_to_string(uint32_t flags)
848 case FLOW_TNL_F_DONT_FRAGMENT:
850 case FLOW_TNL_F_CSUM:
862 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
863 uint32_t flags, char del)
868 ds_put_char(ds, '0');
872 uint32_t bit = rightmost_1bit(flags);
875 s = bit_to_string(bit);
877 ds_put_format(ds, "%s%c", s, del);
886 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
892 format_flags_masked(struct ds *ds, const char *name,
893 const char *(*bit_to_string)(uint32_t), uint32_t flags,
894 uint32_t mask, uint32_t max_mask)
897 ds_put_format(ds, "%s=", name);
900 if (mask == max_mask) {
901 format_flags(ds, bit_to_string, flags, '|');
906 ds_put_cstr(ds, "0/0");
911 uint32_t bit = rightmost_1bit(mask);
912 const char *s = bit_to_string(bit);
914 ds_put_format(ds, "%s%s", (flags & bit) ? "+" : "-",
915 s ? s : "[Unknown]");
920 /* Scans a string 's' of flags to determine their numerical value and
921 * returns the number of characters parsed using 'bit_to_string' to
922 * lookup flag names. Scanning continues until the character 'end' is
925 * In the event of a failure, a negative error code will be returned. In
926 * addition, if 'res_string' is non-NULL then a descriptive string will
927 * be returned incorporating the identifying string 'field_name'. This
928 * error string must be freed by the caller.
930 * Upon success, the flag values will be stored in 'res_flags' and
931 * optionally 'res_mask', if it is non-NULL (if it is NULL then any masks
932 * present in the original string will be considered an error). The
933 * caller may restrict the acceptable set of values through the mask
936 parse_flags(const char *s, const char *(*bit_to_string)(uint32_t),
937 char end, const char *field_name, char **res_string,
938 uint32_t *res_flags, uint32_t allowed, uint32_t *res_mask)
943 /* Parse masked flags in numeric format? */
944 if (res_mask && ovs_scan(s, "%"SCNi32"/%"SCNi32"%n",
945 res_flags, res_mask, &n) && n > 0) {
946 if (*res_flags & ~allowed || *res_mask & ~allowed) {
954 if (res_mask && (*s == '+' || *s == '-')) {
955 uint32_t flags = 0, mask = 0;
957 /* Parse masked flags. */
958 while (s[0] != end) {
965 } else if (s[0] == '-') {
969 *res_string = xasprintf("%s: %s must be preceded by '+' "
970 "(for SET) or '-' (NOT SET)", s,
978 for (bit = 1; bit; bit <<= 1) {
979 const char *fname = bit_to_string(bit);
986 if (strncmp(s, fname, len) ||
987 (s[len] != '+' && s[len] != '-' && s[len] != end)) {
992 /* bit already set. */
994 *res_string = xasprintf("%s: Each %s flag can be "
995 "specified only once", s,
1000 if (!(bit & allowed)) {
1022 /* Parse unmasked flags. If a flag is present, it is set, otherwise
1024 while (s[n] != end) {
1025 unsigned long long int flags;
1029 if (ovs_scan(&s[n], "%lli%n", &flags, &n0)) {
1030 if (flags & ~allowed) {
1033 n += n0 + (s[n + n0] == '|');
1038 for (bit = 1; bit; bit <<= 1) {
1039 const char *name = bit_to_string(bit);
1047 if (!strncmp(s + n, name, len) &&
1048 (s[n + len] == '|' || s[n + len] == end)) {
1049 if (!(bit & allowed)) {
1053 n += len + (s[n + len] == '|');
1063 *res_flags = result;
1065 *res_mask = UINT32_MAX;
1074 *res_string = xasprintf("%s: unknown %s flag(s)", s, field_name);
1080 flow_format(struct ds *ds, const struct flow *flow)
1083 struct flow_wildcards *wc = &match.wc;
1085 match_wc_init(&match, flow);
1087 /* As this function is most often used for formatting a packet in a
1088 * packet-in message, skip formatting the packet context fields that are
1089 * all-zeroes to make the print-out easier on the eyes. This means that a
1090 * missing context field implies a zero value for that field. This is
1091 * similar to OpenFlow encoding of these fields, as the specification
1092 * states that all-zeroes context fields should not be encoded in the
1093 * packet-in messages. */
1094 if (!flow->in_port.ofp_port) {
1095 WC_UNMASK_FIELD(wc, in_port);
1097 if (!flow->skb_priority) {
1098 WC_UNMASK_FIELD(wc, skb_priority);
1100 if (!flow->pkt_mark) {
1101 WC_UNMASK_FIELD(wc, pkt_mark);
1103 if (!flow->recirc_id) {
1104 WC_UNMASK_FIELD(wc, recirc_id);
1106 if (!flow->dp_hash) {
1107 WC_UNMASK_FIELD(wc, dp_hash);
1109 for (int i = 0; i < FLOW_N_REGS; i++) {
1110 if (!flow->regs[i]) {
1111 WC_UNMASK_FIELD(wc, regs[i]);
1114 if (!flow->metadata) {
1115 WC_UNMASK_FIELD(wc, metadata);
1118 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
1122 flow_print(FILE *stream, const struct flow *flow)
1124 char *s = flow_to_string(flow);
1129 /* flow_wildcards functions. */
1131 /* Initializes 'wc' as a set of wildcards that matches every packet. */
1133 flow_wildcards_init_catchall(struct flow_wildcards *wc)
1135 memset(&wc->masks, 0, sizeof wc->masks);
1138 /* Converts a flow into flow wildcards. It sets the wildcard masks based on
1139 * the packet headers extracted to 'flow'. It will not set the mask for fields
1140 * that do not make sense for the packet type. OpenFlow-only metadata is
1141 * wildcarded, but other metadata is unconditionally exact-matched. */
1142 void flow_wildcards_init_for_packet(struct flow_wildcards *wc,
1143 const struct flow *flow)
1145 memset(&wc->masks, 0x0, sizeof wc->masks);
1147 /* Update this function whenever struct flow changes. */
1148 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 33);
1150 if (flow->tunnel.ip_dst) {
1151 if (flow->tunnel.flags & FLOW_TNL_F_KEY) {
1152 WC_MASK_FIELD(wc, tunnel.tun_id);
1154 WC_MASK_FIELD(wc, tunnel.ip_src);
1155 WC_MASK_FIELD(wc, tunnel.ip_dst);
1156 WC_MASK_FIELD(wc, tunnel.flags);
1157 WC_MASK_FIELD(wc, tunnel.ip_tos);
1158 WC_MASK_FIELD(wc, tunnel.ip_ttl);
1159 WC_MASK_FIELD(wc, tunnel.tp_src);
1160 WC_MASK_FIELD(wc, tunnel.tp_dst);
1161 WC_MASK_FIELD(wc, tunnel.gbp_id);
1162 WC_MASK_FIELD(wc, tunnel.gbp_flags);
1164 if (flow->tunnel.metadata.opt_map) {
1165 wc->masks.tunnel.metadata.opt_map = flow->tunnel.metadata.opt_map;
1166 WC_MASK_FIELD(wc, tunnel.metadata.opts);
1168 } else if (flow->tunnel.tun_id) {
1169 WC_MASK_FIELD(wc, tunnel.tun_id);
1172 /* metadata, regs, and conj_id wildcarded. */
1174 WC_MASK_FIELD(wc, skb_priority);
1175 WC_MASK_FIELD(wc, pkt_mark);
1176 WC_MASK_FIELD(wc, recirc_id);
1177 WC_MASK_FIELD(wc, dp_hash);
1178 WC_MASK_FIELD(wc, in_port);
1180 /* actset_output wildcarded. */
1182 WC_MASK_FIELD(wc, dl_dst);
1183 WC_MASK_FIELD(wc, dl_src);
1184 WC_MASK_FIELD(wc, dl_type);
1185 WC_MASK_FIELD(wc, vlan_tci);
1187 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1188 WC_MASK_FIELD(wc, nw_src);
1189 WC_MASK_FIELD(wc, nw_dst);
1190 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1191 WC_MASK_FIELD(wc, ipv6_src);
1192 WC_MASK_FIELD(wc, ipv6_dst);
1193 WC_MASK_FIELD(wc, ipv6_label);
1194 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1195 flow->dl_type == htons(ETH_TYPE_RARP)) {
1196 WC_MASK_FIELD(wc, nw_src);
1197 WC_MASK_FIELD(wc, nw_dst);
1198 WC_MASK_FIELD(wc, nw_proto);
1199 WC_MASK_FIELD(wc, arp_sha);
1200 WC_MASK_FIELD(wc, arp_tha);
1202 } else if (eth_type_mpls(flow->dl_type)) {
1203 for (int i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1204 WC_MASK_FIELD(wc, mpls_lse[i]);
1205 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1211 return; /* Unknown ethertype. */
1215 WC_MASK_FIELD(wc, nw_frag);
1216 WC_MASK_FIELD(wc, nw_tos);
1217 WC_MASK_FIELD(wc, nw_ttl);
1218 WC_MASK_FIELD(wc, nw_proto);
1220 /* No transport layer header in later fragments. */
1221 if (!(flow->nw_frag & FLOW_NW_FRAG_LATER) &&
1222 (flow->nw_proto == IPPROTO_ICMP ||
1223 flow->nw_proto == IPPROTO_ICMPV6 ||
1224 flow->nw_proto == IPPROTO_TCP ||
1225 flow->nw_proto == IPPROTO_UDP ||
1226 flow->nw_proto == IPPROTO_SCTP ||
1227 flow->nw_proto == IPPROTO_IGMP)) {
1228 WC_MASK_FIELD(wc, tp_src);
1229 WC_MASK_FIELD(wc, tp_dst);
1231 if (flow->nw_proto == IPPROTO_TCP) {
1232 WC_MASK_FIELD(wc, tcp_flags);
1233 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
1234 WC_MASK_FIELD(wc, arp_sha);
1235 WC_MASK_FIELD(wc, arp_tha);
1236 WC_MASK_FIELD(wc, nd_target);
1237 } else if (flow->nw_proto == IPPROTO_IGMP) {
1238 WC_MASK_FIELD(wc, igmp_group_ip4);
1243 /* Return a map of possible fields for a packet of the same type as 'flow'.
1244 * Including extra bits in the returned mask is not wrong, it is just less
1247 * This is a less precise version of flow_wildcards_init_for_packet() above. */
1249 flow_wc_map(const struct flow *flow, struct miniflow *map)
1251 /* Update this function whenever struct flow changes. */
1252 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 33);
1255 if (flow->tunnel.ip_dst) {
1256 map->tnl_map = MINIFLOW_TNL_MAP(tunnel);
1257 if (!flow->tunnel.metadata.opt_map) {
1258 map->tnl_map &= ~MINIFLOW_TNL_MAP(tunnel.metadata);
1262 /* Metadata fields that can appear on packet input. */
1263 map->pkt_map = MINIFLOW_PKT_MAP(skb_priority) | MINIFLOW_PKT_MAP(pkt_mark)
1264 | MINIFLOW_PKT_MAP(recirc_id) | MINIFLOW_PKT_MAP(dp_hash)
1265 | MINIFLOW_PKT_MAP(in_port)
1266 | MINIFLOW_PKT_MAP(dl_dst) | MINIFLOW_PKT_MAP(dl_src)
1267 | MINIFLOW_PKT_MAP(dl_type) | MINIFLOW_PKT_MAP(vlan_tci);
1269 /* Ethertype-dependent fields. */
1270 if (OVS_LIKELY(flow->dl_type == htons(ETH_TYPE_IP))) {
1271 map->pkt_map |= MINIFLOW_PKT_MAP(nw_src) | MINIFLOW_PKT_MAP(nw_dst)
1272 | MINIFLOW_PKT_MAP(nw_proto) | MINIFLOW_PKT_MAP(nw_frag)
1273 | MINIFLOW_PKT_MAP(nw_tos) | MINIFLOW_PKT_MAP(nw_ttl);
1274 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_IGMP)) {
1275 map->pkt_map |= MINIFLOW_PKT_MAP(igmp_group_ip4);
1277 map->pkt_map |= MINIFLOW_PKT_MAP(tcp_flags)
1278 | MINIFLOW_PKT_MAP(tp_src) | MINIFLOW_PKT_MAP(tp_dst);
1280 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1281 map->pkt_map |= MINIFLOW_PKT_MAP(ipv6_src) | MINIFLOW_PKT_MAP(ipv6_dst)
1282 | MINIFLOW_PKT_MAP(ipv6_label)
1283 | MINIFLOW_PKT_MAP(nw_proto) | MINIFLOW_PKT_MAP(nw_frag)
1284 | MINIFLOW_PKT_MAP(nw_tos) | MINIFLOW_PKT_MAP(nw_ttl);
1285 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_ICMPV6)) {
1286 map->pkt_map |= MINIFLOW_PKT_MAP(nd_target)
1287 | MINIFLOW_PKT_MAP(arp_sha) | MINIFLOW_PKT_MAP(arp_tha);
1289 map->pkt_map |= MINIFLOW_PKT_MAP(tcp_flags)
1290 | MINIFLOW_PKT_MAP(tp_src) | MINIFLOW_PKT_MAP(tp_dst);
1292 } else if (eth_type_mpls(flow->dl_type)) {
1293 map->pkt_map |= MINIFLOW_PKT_MAP(mpls_lse);
1294 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1295 flow->dl_type == htons(ETH_TYPE_RARP)) {
1296 map->pkt_map |= MINIFLOW_PKT_MAP(nw_src) | MINIFLOW_PKT_MAP(nw_dst)
1297 | MINIFLOW_PKT_MAP(nw_proto)
1298 | MINIFLOW_PKT_MAP(arp_sha) | MINIFLOW_PKT_MAP(arp_tha);
1302 /* Clear the metadata and register wildcard masks. They are not packet
1305 flow_wildcards_clear_non_packet_fields(struct flow_wildcards *wc)
1307 /* Update this function whenever struct flow changes. */
1308 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 33);
1310 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
1311 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
1312 wc->masks.actset_output = 0;
1313 wc->masks.conj_id = 0;
1316 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
1319 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
1321 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1324 for (i = 0; i < FLOW_U64S; i++) {
1332 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
1333 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
1334 * in 'src1' or 'src2' or both. */
1336 flow_wildcards_and(struct flow_wildcards *dst,
1337 const struct flow_wildcards *src1,
1338 const struct flow_wildcards *src2)
1340 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1341 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1342 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1345 for (i = 0; i < FLOW_U64S; i++) {
1346 dst_u64[i] = src1_u64[i] & src2_u64[i];
1350 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
1351 * is, a bit or a field is wildcarded in 'dst' if it is neither
1352 * wildcarded in 'src1' nor 'src2'. */
1354 flow_wildcards_or(struct flow_wildcards *dst,
1355 const struct flow_wildcards *src1,
1356 const struct flow_wildcards *src2)
1358 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1359 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1360 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1363 for (i = 0; i < FLOW_U64S; i++) {
1364 dst_u64[i] = src1_u64[i] | src2_u64[i];
1368 /* Returns a hash of the wildcards in 'wc'. */
1370 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
1372 return flow_hash(&wc->masks, basis);
1375 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
1378 flow_wildcards_equal(const struct flow_wildcards *a,
1379 const struct flow_wildcards *b)
1381 return flow_equal(&a->masks, &b->masks);
1384 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
1385 * 'b', false otherwise. */
1387 flow_wildcards_has_extra(const struct flow_wildcards *a,
1388 const struct flow_wildcards *b)
1390 const uint64_t *a_u64 = (const uint64_t *) &a->masks;
1391 const uint64_t *b_u64 = (const uint64_t *) &b->masks;
1394 for (i = 0; i < FLOW_U64S; i++) {
1395 if ((a_u64[i] & b_u64[i]) != b_u64[i]) {
1402 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
1403 * in 'wc' do not need to be equal in 'a' and 'b'. */
1405 flow_equal_except(const struct flow *a, const struct flow *b,
1406 const struct flow_wildcards *wc)
1408 const uint64_t *a_u64 = (const uint64_t *) a;
1409 const uint64_t *b_u64 = (const uint64_t *) b;
1410 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1413 for (i = 0; i < FLOW_U64S; i++) {
1414 if ((a_u64[i] ^ b_u64[i]) & wc_u64[i]) {
1421 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1422 * (A 0-bit indicates a wildcard bit.) */
1424 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
1426 wc->masks.regs[idx] = mask;
1429 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1430 * (A 0-bit indicates a wildcard bit.) */
1432 flow_wildcards_set_xreg_mask(struct flow_wildcards *wc, int idx, uint64_t mask)
1434 flow_set_xreg(&wc->masks, idx, mask);
1437 /* Calculates the 5-tuple hash from the given miniflow.
1438 * This returns the same value as flow_hash_5tuple for the corresponding
1441 miniflow_hash_5tuple(const struct miniflow *flow, uint32_t basis)
1443 uint32_t hash = basis;
1446 ovs_be16 dl_type = MINIFLOW_GET_BE16(flow, dl_type);
1448 hash = hash_add(hash, MINIFLOW_GET_U8(flow, nw_proto));
1450 /* Separate loops for better optimization. */
1451 if (dl_type == htons(ETH_TYPE_IPV6)) {
1452 struct miniflow maps = { 0, MINIFLOW_PKT_MAP(ipv6_src)
1453 | MINIFLOW_PKT_MAP(ipv6_dst) };
1456 MINIFLOW_FOR_EACH_IN_PKT_MAP(value, flow, maps) {
1457 hash = hash_add64(hash, value);
1460 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_src));
1461 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_dst));
1463 /* Add both ports at once. */
1464 hash = hash_add(hash, MINIFLOW_GET_U32(flow, tp_src));
1465 hash = hash_finish(hash, 42); /* Arbitrary number. */
1470 ASSERT_SEQUENTIAL_SAME_WORD(tp_src, tp_dst);
1471 ASSERT_SEQUENTIAL(ipv6_src, ipv6_dst);
1473 /* Calculates the 5-tuple hash from the given flow. */
1475 flow_hash_5tuple(const struct flow *flow, uint32_t basis)
1477 uint32_t hash = basis;
1480 hash = hash_add(hash, flow->nw_proto);
1482 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1483 const uint64_t *flow_u64 = (const uint64_t *)flow;
1484 int ofs = offsetof(struct flow, ipv6_src) / 8;
1485 int end = ofs + 2 * sizeof flow->ipv6_src / 8;
1487 for (;ofs < end; ofs++) {
1488 hash = hash_add64(hash, flow_u64[ofs]);
1491 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_src);
1492 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_dst);
1494 /* Add both ports at once. */
1495 hash = hash_add(hash,
1496 ((const uint32_t *)flow)[offsetof(struct flow, tp_src)
1497 / sizeof(uint32_t)]);
1498 hash = hash_finish(hash, 42); /* Arbitrary number. */
1503 /* Hashes 'flow' based on its L2 through L4 protocol information. */
1505 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
1510 struct in6_addr ipv6_addr;
1515 uint8_t eth_addr[ETH_ADDR_LEN];
1521 memset(&fields, 0, sizeof fields);
1522 for (i = 0; i < ETH_ADDR_LEN; i++) {
1523 fields.eth_addr[i] = flow->dl_src[i] ^ flow->dl_dst[i];
1525 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
1526 fields.eth_type = flow->dl_type;
1528 /* UDP source and destination port are not taken into account because they
1529 * will not necessarily be symmetric in a bidirectional flow. */
1530 if (fields.eth_type == htons(ETH_TYPE_IP)) {
1531 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
1532 fields.ip_proto = flow->nw_proto;
1533 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1534 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1536 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
1537 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
1538 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
1539 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
1541 for (i=0; i<16; i++) {
1542 ipv6_addr[i] = a[i] ^ b[i];
1544 fields.ip_proto = flow->nw_proto;
1545 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1546 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1549 return jhash_bytes(&fields, sizeof fields, basis);
1552 /* Hashes 'flow' based on its L3 through L4 protocol information */
1554 flow_hash_symmetric_l3l4(const struct flow *flow, uint32_t basis,
1557 uint32_t hash = basis;
1559 /* UDP source and destination port are also taken into account. */
1560 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1561 hash = hash_add(hash,
1562 (OVS_FORCE uint32_t) (flow->nw_src ^ flow->nw_dst));
1563 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1564 /* IPv6 addresses are 64-bit aligned inside struct flow. */
1565 const uint64_t *a = ALIGNED_CAST(uint64_t *, flow->ipv6_src.s6_addr);
1566 const uint64_t *b = ALIGNED_CAST(uint64_t *, flow->ipv6_dst.s6_addr);
1568 for (int i = 0; i < 4; i++) {
1569 hash = hash_add64(hash, a[i] ^ b[i]);
1572 /* Cannot hash non-IP flows */
1576 hash = hash_add(hash, flow->nw_proto);
1577 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP ||
1578 (inc_udp_ports && flow->nw_proto == IPPROTO_UDP)) {
1579 hash = hash_add(hash,
1580 (OVS_FORCE uint16_t) (flow->tp_src ^ flow->tp_dst));
1583 return hash_finish(hash, basis);
1586 /* Initialize a flow with random fields that matter for nx_hash_fields. */
1588 flow_random_hash_fields(struct flow *flow)
1590 uint16_t rnd = random_uint16();
1592 /* Initialize to all zeros. */
1593 memset(flow, 0, sizeof *flow);
1595 eth_addr_random(flow->dl_src);
1596 eth_addr_random(flow->dl_dst);
1598 flow->vlan_tci = (OVS_FORCE ovs_be16) (random_uint16() & VLAN_VID_MASK);
1600 /* Make most of the random flows IPv4, some IPv6, and rest random. */
1601 flow->dl_type = rnd < 0x8000 ? htons(ETH_TYPE_IP) :
1602 rnd < 0xc000 ? htons(ETH_TYPE_IPV6) : (OVS_FORCE ovs_be16)rnd;
1604 if (dl_type_is_ip_any(flow->dl_type)) {
1605 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1606 flow->nw_src = (OVS_FORCE ovs_be32)random_uint32();
1607 flow->nw_dst = (OVS_FORCE ovs_be32)random_uint32();
1609 random_bytes(&flow->ipv6_src, sizeof flow->ipv6_src);
1610 random_bytes(&flow->ipv6_dst, sizeof flow->ipv6_dst);
1612 /* Make most of IP flows TCP, some UDP or SCTP, and rest random. */
1613 rnd = random_uint16();
1614 flow->nw_proto = rnd < 0x8000 ? IPPROTO_TCP :
1615 rnd < 0xc000 ? IPPROTO_UDP :
1616 rnd < 0xd000 ? IPPROTO_SCTP : (uint8_t)rnd;
1617 if (flow->nw_proto == IPPROTO_TCP ||
1618 flow->nw_proto == IPPROTO_UDP ||
1619 flow->nw_proto == IPPROTO_SCTP) {
1620 flow->tp_src = (OVS_FORCE ovs_be16)random_uint16();
1621 flow->tp_dst = (OVS_FORCE ovs_be16)random_uint16();
1626 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
1628 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
1629 enum nx_hash_fields fields)
1632 case NX_HASH_FIELDS_ETH_SRC:
1633 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1636 case NX_HASH_FIELDS_SYMMETRIC_L4:
1637 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1638 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1639 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1640 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1641 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1642 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1643 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1644 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1646 if (is_ip_any(flow)) {
1647 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1648 flow_unwildcard_tp_ports(flow, wc);
1650 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1653 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1654 if (is_ip_any(flow) && flow->nw_proto == IPPROTO_UDP) {
1655 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1656 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1659 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1660 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1661 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1662 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1663 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1664 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1665 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1667 break; /* non-IP flow */
1670 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1671 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP) {
1672 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1673 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1682 /* Hashes the portions of 'flow' designated by 'fields'. */
1684 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
1689 case NX_HASH_FIELDS_ETH_SRC:
1690 return jhash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
1692 case NX_HASH_FIELDS_SYMMETRIC_L4:
1693 return flow_hash_symmetric_l4(flow, basis);
1695 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1696 return flow_hash_symmetric_l3l4(flow, basis, false);
1698 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1699 return flow_hash_symmetric_l3l4(flow, basis, true);
1706 /* Returns a string representation of 'fields'. */
1708 flow_hash_fields_to_str(enum nx_hash_fields fields)
1711 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
1712 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
1713 case NX_HASH_FIELDS_SYMMETRIC_L3L4: return "symmetric_l3l4";
1714 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP: return "symmetric_l3l4+udp";
1715 default: return "<unknown>";
1719 /* Returns true if the value of 'fields' is supported. Otherwise false. */
1721 flow_hash_fields_valid(enum nx_hash_fields fields)
1723 return fields == NX_HASH_FIELDS_ETH_SRC
1724 || fields == NX_HASH_FIELDS_SYMMETRIC_L4
1725 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4
1726 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP;
1729 /* Returns a hash value for the bits of 'flow' that are active based on
1730 * 'wc', given 'basis'. */
1732 flow_hash_in_wildcards(const struct flow *flow,
1733 const struct flow_wildcards *wc, uint32_t basis)
1735 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1736 const uint64_t *flow_u64 = (const uint64_t *) flow;
1741 for (i = 0; i < FLOW_U64S; i++) {
1742 hash = hash_add64(hash, flow_u64[i] & wc_u64[i]);
1744 return hash_finish(hash, 8 * FLOW_U64S);
1747 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1748 * OpenFlow 1.0 "dl_vlan" value:
1750 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
1751 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
1752 * 'flow' previously matched packets without a VLAN header).
1754 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
1755 * without a VLAN tag.
1757 * - Other values of 'vid' should not be used. */
1759 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
1761 if (vid == htons(OFP10_VLAN_NONE)) {
1762 flow->vlan_tci = htons(0);
1764 vid &= htons(VLAN_VID_MASK);
1765 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1766 flow->vlan_tci |= htons(VLAN_CFI) | vid;
1770 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1771 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
1774 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
1776 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
1777 flow->vlan_tci &= ~mask;
1778 flow->vlan_tci |= vid & mask;
1781 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
1784 * This function has no effect on the VLAN ID that 'flow' matches.
1786 * After calling this function, 'flow' will not match packets without a VLAN
1789 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
1792 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1793 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
1796 /* Returns the number of MPLS LSEs present in 'flow'
1798 * Returns 0 if the 'dl_type' of 'flow' is not an MPLS ethernet type.
1799 * Otherwise traverses 'flow''s MPLS label stack stopping at the
1800 * first entry that has the BoS bit set. If no such entry exists then
1801 * the maximum number of LSEs that can be stored in 'flow' is returned.
1804 flow_count_mpls_labels(const struct flow *flow, struct flow_wildcards *wc)
1806 /* dl_type is always masked. */
1807 if (eth_type_mpls(flow->dl_type)) {
1812 for (i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1814 wc->masks.mpls_lse[i] |= htonl(MPLS_BOS_MASK);
1816 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1819 if (flow->mpls_lse[i]) {
1829 /* Returns the number consecutive of MPLS LSEs, starting at the
1830 * innermost LSE, that are common in 'a' and 'b'.
1832 * 'an' must be flow_count_mpls_labels(a).
1833 * 'bn' must be flow_count_mpls_labels(b).
1836 flow_count_common_mpls_labels(const struct flow *a, int an,
1837 const struct flow *b, int bn,
1838 struct flow_wildcards *wc)
1840 int min_n = MIN(an, bn);
1845 int a_last = an - 1;
1846 int b_last = bn - 1;
1849 for (i = 0; i < min_n; i++) {
1851 wc->masks.mpls_lse[a_last - i] = OVS_BE32_MAX;
1852 wc->masks.mpls_lse[b_last - i] = OVS_BE32_MAX;
1854 if (a->mpls_lse[a_last - i] != b->mpls_lse[b_last - i]) {
1865 /* Adds a new outermost MPLS label to 'flow' and changes 'flow''s Ethernet type
1866 * to 'mpls_eth_type', which must be an MPLS Ethertype.
1868 * If the new label is the first MPLS label in 'flow', it is generated as;
1870 * - label: 2, if 'flow' is IPv6, otherwise 0.
1872 * - TTL: IPv4 or IPv6 TTL, if present and nonzero, otherwise 64.
1874 * - TC: IPv4 or IPv6 TOS, if present, otherwise 0.
1878 * If the new label is the second or later label MPLS label in 'flow', it is
1881 * - label: Copied from outer label.
1883 * - TTL: Copied from outer label.
1885 * - TC: Copied from outer label.
1889 * 'n' must be flow_count_mpls_labels(flow). 'n' must be less than
1890 * FLOW_MAX_MPLS_LABELS (because otherwise flow->mpls_lse[] would overflow).
1893 flow_push_mpls(struct flow *flow, int n, ovs_be16 mpls_eth_type,
1894 struct flow_wildcards *wc)
1896 ovs_assert(eth_type_mpls(mpls_eth_type));
1897 ovs_assert(n < FLOW_MAX_MPLS_LABELS);
1903 memset(&wc->masks.mpls_lse, 0xff, sizeof *wc->masks.mpls_lse * n);
1905 for (i = n; i >= 1; i--) {
1906 flow->mpls_lse[i] = flow->mpls_lse[i - 1];
1908 flow->mpls_lse[0] = (flow->mpls_lse[1] & htonl(~MPLS_BOS_MASK));
1910 int label = 0; /* IPv4 Explicit Null. */
1914 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1918 if (is_ip_any(flow)) {
1919 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1921 wc->masks.nw_tos |= IP_DSCP_MASK;
1922 wc->masks.nw_ttl = 0xff;
1930 flow->mpls_lse[0] = set_mpls_lse_values(ttl, tc, 1, htonl(label));
1932 /* Clear all L3 and L4 fields and dp_hash. */
1933 BUILD_ASSERT(FLOW_WC_SEQ == 33);
1934 memset((char *) flow + FLOW_SEGMENT_2_ENDS_AT, 0,
1935 sizeof(struct flow) - FLOW_SEGMENT_2_ENDS_AT);
1938 flow->dl_type = mpls_eth_type;
1941 /* Tries to remove the outermost MPLS label from 'flow'. Returns true if
1942 * successful, false otherwise. On success, sets 'flow''s Ethernet type to
1945 * 'n' must be flow_count_mpls_labels(flow). */
1947 flow_pop_mpls(struct flow *flow, int n, ovs_be16 eth_type,
1948 struct flow_wildcards *wc)
1953 /* Nothing to pop. */
1955 } else if (n == FLOW_MAX_MPLS_LABELS) {
1957 wc->masks.mpls_lse[n - 1] |= htonl(MPLS_BOS_MASK);
1959 if (!(flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK))) {
1960 /* Can't pop because don't know what to fill in mpls_lse[n - 1]. */
1966 memset(&wc->masks.mpls_lse[1], 0xff,
1967 sizeof *wc->masks.mpls_lse * (n - 1));
1969 for (i = 1; i < n; i++) {
1970 flow->mpls_lse[i - 1] = flow->mpls_lse[i];
1972 flow->mpls_lse[n - 1] = 0;
1973 flow->dl_type = eth_type;
1977 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
1978 * as an OpenFlow 1.1 "mpls_label" value. */
1980 flow_set_mpls_label(struct flow *flow, int idx, ovs_be32 label)
1982 set_mpls_lse_label(&flow->mpls_lse[idx], label);
1985 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
1988 flow_set_mpls_ttl(struct flow *flow, int idx, uint8_t ttl)
1990 set_mpls_lse_ttl(&flow->mpls_lse[idx], ttl);
1993 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
1996 flow_set_mpls_tc(struct flow *flow, int idx, uint8_t tc)
1998 set_mpls_lse_tc(&flow->mpls_lse[idx], tc);
2001 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
2003 flow_set_mpls_bos(struct flow *flow, int idx, uint8_t bos)
2005 set_mpls_lse_bos(&flow->mpls_lse[idx], bos);
2008 /* Sets the entire MPLS LSE. */
2010 flow_set_mpls_lse(struct flow *flow, int idx, ovs_be32 lse)
2012 flow->mpls_lse[idx] = lse;
2016 flow_compose_l4(struct dp_packet *p, const struct flow *flow)
2020 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
2021 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
2022 if (flow->nw_proto == IPPROTO_TCP) {
2023 struct tcp_header *tcp;
2025 l4_len = sizeof *tcp;
2026 tcp = dp_packet_put_zeros(p, l4_len);
2027 tcp->tcp_src = flow->tp_src;
2028 tcp->tcp_dst = flow->tp_dst;
2029 tcp->tcp_ctl = TCP_CTL(ntohs(flow->tcp_flags), 5);
2030 } else if (flow->nw_proto == IPPROTO_UDP) {
2031 struct udp_header *udp;
2033 l4_len = sizeof *udp;
2034 udp = dp_packet_put_zeros(p, l4_len);
2035 udp->udp_src = flow->tp_src;
2036 udp->udp_dst = flow->tp_dst;
2037 } else if (flow->nw_proto == IPPROTO_SCTP) {
2038 struct sctp_header *sctp;
2040 l4_len = sizeof *sctp;
2041 sctp = dp_packet_put_zeros(p, l4_len);
2042 sctp->sctp_src = flow->tp_src;
2043 sctp->sctp_dst = flow->tp_dst;
2044 } else if (flow->nw_proto == IPPROTO_ICMP) {
2045 struct icmp_header *icmp;
2047 l4_len = sizeof *icmp;
2048 icmp = dp_packet_put_zeros(p, l4_len);
2049 icmp->icmp_type = ntohs(flow->tp_src);
2050 icmp->icmp_code = ntohs(flow->tp_dst);
2051 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
2052 } else if (flow->nw_proto == IPPROTO_IGMP) {
2053 struct igmp_header *igmp;
2055 l4_len = sizeof *igmp;
2056 igmp = dp_packet_put_zeros(p, l4_len);
2057 igmp->igmp_type = ntohs(flow->tp_src);
2058 igmp->igmp_code = ntohs(flow->tp_dst);
2059 put_16aligned_be32(&igmp->group, flow->igmp_group_ip4);
2060 igmp->igmp_csum = csum(igmp, IGMP_HEADER_LEN);
2061 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
2062 struct icmp6_hdr *icmp;
2064 l4_len = sizeof *icmp;
2065 icmp = dp_packet_put_zeros(p, l4_len);
2066 icmp->icmp6_type = ntohs(flow->tp_src);
2067 icmp->icmp6_code = ntohs(flow->tp_dst);
2069 if (icmp->icmp6_code == 0 &&
2070 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
2071 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
2072 struct in6_addr *nd_target;
2073 struct nd_opt_hdr *nd_opt;
2075 l4_len += sizeof *nd_target;
2076 nd_target = dp_packet_put_zeros(p, sizeof *nd_target);
2077 *nd_target = flow->nd_target;
2079 if (!eth_addr_is_zero(flow->arp_sha)) {
2081 nd_opt = dp_packet_put_zeros(p, 8);
2082 nd_opt->nd_opt_len = 1;
2083 nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
2084 memcpy(nd_opt + 1, flow->arp_sha, ETH_ADDR_LEN);
2086 if (!eth_addr_is_zero(flow->arp_tha)) {
2088 nd_opt = dp_packet_put_zeros(p, 8);
2089 nd_opt->nd_opt_len = 1;
2090 nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR;
2091 memcpy(nd_opt + 1, flow->arp_tha, ETH_ADDR_LEN);
2094 icmp->icmp6_cksum = (OVS_FORCE uint16_t)
2095 csum(icmp, (char *)dp_packet_tail(p) - (char *)icmp);
2101 /* Puts into 'b' a packet that flow_extract() would parse as having the given
2104 * (This is useful only for testing, obviously, and the packet isn't really
2105 * valid. It hasn't got some checksums filled in, for one, and lots of fields
2106 * are just zeroed.) */
2108 flow_compose(struct dp_packet *p, const struct flow *flow)
2112 /* eth_compose() sets l3 pointer and makes sure it is 32-bit aligned. */
2113 eth_compose(p, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
2114 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
2115 struct eth_header *eth = dp_packet_l2(p);
2116 eth->eth_type = htons(dp_packet_size(p));
2120 if (flow->vlan_tci & htons(VLAN_CFI)) {
2121 eth_push_vlan(p, htons(ETH_TYPE_VLAN), flow->vlan_tci);
2124 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2125 struct ip_header *ip;
2127 ip = dp_packet_put_zeros(p, sizeof *ip);
2128 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
2129 ip->ip_tos = flow->nw_tos;
2130 ip->ip_ttl = flow->nw_ttl;
2131 ip->ip_proto = flow->nw_proto;
2132 put_16aligned_be32(&ip->ip_src, flow->nw_src);
2133 put_16aligned_be32(&ip->ip_dst, flow->nw_dst);
2135 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2136 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
2137 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
2138 ip->ip_frag_off |= htons(100);
2142 dp_packet_set_l4(p, dp_packet_tail(p));
2144 l4_len = flow_compose_l4(p, flow);
2146 ip = dp_packet_l3(p);
2147 ip->ip_tot_len = htons(p->l4_ofs - p->l3_ofs + l4_len);
2148 ip->ip_csum = csum(ip, sizeof *ip);
2149 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2150 struct ovs_16aligned_ip6_hdr *nh;
2152 nh = dp_packet_put_zeros(p, sizeof *nh);
2153 put_16aligned_be32(&nh->ip6_flow, htonl(6 << 28) |
2154 htonl(flow->nw_tos << 20) | flow->ipv6_label);
2155 nh->ip6_hlim = flow->nw_ttl;
2156 nh->ip6_nxt = flow->nw_proto;
2158 memcpy(&nh->ip6_src, &flow->ipv6_src, sizeof(nh->ip6_src));
2159 memcpy(&nh->ip6_dst, &flow->ipv6_dst, sizeof(nh->ip6_dst));
2161 dp_packet_set_l4(p, dp_packet_tail(p));
2163 l4_len = flow_compose_l4(p, flow);
2165 nh = dp_packet_l3(p);
2166 nh->ip6_plen = htons(l4_len);
2167 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
2168 flow->dl_type == htons(ETH_TYPE_RARP)) {
2169 struct arp_eth_header *arp;
2171 arp = dp_packet_put_zeros(p, sizeof *arp);
2172 dp_packet_set_l3(p, arp);
2173 arp->ar_hrd = htons(1);
2174 arp->ar_pro = htons(ETH_TYPE_IP);
2175 arp->ar_hln = ETH_ADDR_LEN;
2177 arp->ar_op = htons(flow->nw_proto);
2179 if (flow->nw_proto == ARP_OP_REQUEST ||
2180 flow->nw_proto == ARP_OP_REPLY) {
2181 put_16aligned_be32(&arp->ar_spa, flow->nw_src);
2182 put_16aligned_be32(&arp->ar_tpa, flow->nw_dst);
2183 memcpy(arp->ar_sha, flow->arp_sha, ETH_ADDR_LEN);
2184 memcpy(arp->ar_tha, flow->arp_tha, ETH_ADDR_LEN);
2188 if (eth_type_mpls(flow->dl_type)) {
2191 p->l2_5_ofs = p->l3_ofs;
2192 for (n = 1; n < FLOW_MAX_MPLS_LABELS; n++) {
2193 if (flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK)) {
2198 push_mpls(p, flow->dl_type, flow->mpls_lse[--n]);
2203 /* Compressed flow. */
2205 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
2206 * the caller. The caller must have already computed 'dst->tnl_map' and
2207 * 'dst->pkt_map' properly to indicate the significant uint64_t elements of
2210 * Normally the significant elements are the ones that are non-zero. However,
2211 * when a miniflow is initialized from a (mini)mask, the values can be zeroes,
2212 * so that the flow and mask always have the same maps. */
2214 miniflow_init(struct miniflow *dst, const struct flow *src)
2216 const uint64_t *src_u64 = (const uint64_t *) src;
2217 uint64_t *dst_u64 = miniflow_values(dst);
2220 MAPS_FOR_EACH_INDEX(idx, *dst) {
2221 *dst_u64++ = src_u64[idx];
2225 /* Initialize the maps of 'flow' from 'src'. */
2227 miniflow_map_init(struct miniflow *flow, const struct flow *src)
2229 const uint64_t *src_u64 = (const uint64_t *) src;
2232 /* Initialize map, counting the number of nonzero elements. */
2234 for (i = 0; i < FLOW_TNL_U64S; i++) {
2236 flow->tnl_map |= UINT64_C(1) << i;
2239 src_u64 += FLOW_TNL_U64S;
2241 for (i = 0; i < FLOW_U64S - FLOW_TNL_U64S; i++) {
2243 flow->pkt_map |= UINT64_C(1) << i;
2248 /* Allocates 'n' count of miniflows, consecutive in memory, initializing the
2249 * map of each from 'src'.
2250 * Returns the size of the miniflow data. */
2252 miniflow_alloc(struct miniflow *dsts[], size_t n, const struct miniflow *src)
2254 size_t n_values = miniflow_n_values(src);
2255 size_t data_size = MINIFLOW_VALUES_SIZE(n_values);
2256 struct miniflow *dst = xmalloc(n * (sizeof *src + data_size));
2259 COVERAGE_INC(miniflow_malloc);
2261 for (i = 0; i < n; i++) {
2262 *dst = *src; /* Copy maps. */
2264 dst += 1; /* Just past the maps. */
2265 dst = (struct miniflow *)((uint64_t *)dst + n_values); /* Skip data. */
2270 /* Returns a miniflow copy of 'src'. The caller must eventually free() the
2271 * returned miniflow. */
2273 miniflow_create(const struct flow *src)
2275 struct miniflow tmp;
2276 struct miniflow *dst;
2278 miniflow_map_init(&tmp, src);
2280 miniflow_alloc(&dst, 1, &tmp);
2281 miniflow_init(dst, src);
2285 /* Initializes 'dst' as a copy of 'src'. The caller must have allocated
2286 * 'dst' to have inline space for 'n_values' data in 'src'. */
2288 miniflow_clone(struct miniflow *dst, const struct miniflow *src,
2291 *dst = *src; /* Copy maps. */
2292 memcpy(miniflow_values(dst), miniflow_get_values(src),
2293 MINIFLOW_VALUES_SIZE(n_values));
2296 /* Initializes 'dst' as a copy of 'src'. */
2298 miniflow_expand(const struct miniflow *src, struct flow *dst)
2300 memset(dst, 0, sizeof *dst);
2301 flow_union_with_miniflow(dst, src);
2304 /* Returns true if 'a' and 'b' are equal miniflows, false otherwise. */
2306 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
2308 const uint64_t *ap = miniflow_get_values(a);
2309 const uint64_t *bp = miniflow_get_values(b);
2311 if (OVS_LIKELY(a->tnl_map == b->tnl_map && a->pkt_map == b->pkt_map)) {
2312 return !memcmp(ap, bp, miniflow_n_values(a) * sizeof *ap);
2316 map = a->tnl_map | b->tnl_map;
2317 for (; map; map = zero_rightmost_1bit(map)) {
2318 uint64_t bit = rightmost_1bit(map);
2320 if ((a->tnl_map & bit ? *ap++ : 0)
2321 != (b->tnl_map & bit ? *bp++ : 0)) {
2325 map = a->pkt_map | b->pkt_map;
2326 for (; map; map = zero_rightmost_1bit(map)) {
2327 uint64_t bit = rightmost_1bit(map);
2329 if ((a->pkt_map & bit ? *ap++ : 0)
2330 != (b->pkt_map & bit ? *bp++ : 0)) {
2339 /* Returns false if 'a' and 'b' differ at the places where there are 1-bits
2340 * in 'mask', true otherwise. */
2342 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
2343 const struct minimask *mask)
2345 const uint64_t *p = miniflow_get_values(&mask->masks);
2348 MAPS_FOR_EACH_INDEX(idx, mask->masks) {
2349 if ((miniflow_get(a, idx) ^ miniflow_get(b, idx)) & *p++) {
2357 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
2358 * in 'mask', false if they differ. */
2360 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
2361 const struct minimask *mask)
2363 const uint64_t *b_u64 = (const uint64_t *) b;
2364 const uint64_t *p = miniflow_get_values(&mask->masks);
2367 MAPS_FOR_EACH_INDEX(idx, mask->masks) {
2368 if ((miniflow_get(a, idx) ^ b_u64[idx]) & *p++) {
2378 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
2380 miniflow_init(&mask->masks, &wc->masks);
2383 /* Returns a minimask copy of 'wc'. The caller must eventually free the
2384 * returned minimask with free(). */
2386 minimask_create(const struct flow_wildcards *wc)
2388 return (struct minimask *)miniflow_create(&wc->masks);
2391 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
2393 * The caller must provide room for FLOW_U64S "uint64_t"s in 'storage', which
2394 * must follow '*dst_' in memory, for use by 'dst_'. The caller must *not*
2395 * free 'dst_' free(). */
2397 minimask_combine(struct minimask *dst_,
2398 const struct minimask *a_, const struct minimask *b_,
2399 uint64_t storage[FLOW_U64S])
2401 struct miniflow *dst = &dst_->masks;
2402 uint64_t *dst_values = storage;
2403 const struct miniflow *a = &a_->masks;
2404 const struct miniflow *b = &b_->masks;
2405 const uint64_t *ap = miniflow_get_values(a);
2406 const uint64_t *bp = miniflow_get_values(b);
2410 MAP_FOR_EACH_INDEX(idx, a->tnl_map & b->tnl_map) {
2411 /* Both 'a' and 'b' have non-zero data at 'idx'. */
2412 uint64_t mask = *miniflow_values_get__(ap, a->tnl_map, idx)
2413 & *miniflow_values_get__(bp, b->tnl_map, idx);
2416 dst->tnl_map |= UINT64_C(1) << idx;
2417 *dst_values++ = mask;
2421 ap += count_1bits(a->tnl_map); /* Skip tnl_map values. */
2422 bp += count_1bits(b->tnl_map); /* Skip tnl_map values. */
2423 MAP_FOR_EACH_INDEX(idx, a->pkt_map & b->pkt_map) {
2424 /* Both 'a' and 'b' have non-zero data at 'idx'. */
2425 uint64_t mask = *miniflow_values_get__(ap, a->pkt_map, idx)
2426 & *miniflow_values_get__(bp, b->pkt_map, idx);
2429 dst->pkt_map |= UINT64_C(1) << idx;
2430 *dst_values++ = mask;
2435 /* Initializes 'wc' as a copy of 'mask'. */
2437 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
2439 miniflow_expand(&mask->masks, &wc->masks);
2442 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise.
2443 * Minimasks may not have zero data values, so for the minimasks to be the
2444 * same, they need to have the same map and the same data values. */
2446 minimask_equal(const struct minimask *a, const struct minimask *b)
2448 return a->masks.tnl_map == b->masks.tnl_map
2449 && a->masks.pkt_map == b->masks.pkt_map &&
2450 !memcmp(miniflow_get_values(&a->masks), miniflow_get_values(&b->masks),
2451 MINIFLOW_VALUES_SIZE(miniflow_n_values(&a->masks)));
2454 /* Returns true if at least one bit matched by 'b' is wildcarded by 'a',
2455 * false otherwise. */
2457 minimask_has_extra(const struct minimask *a, const struct minimask *b)
2459 const uint64_t *ap = miniflow_get_values(&a->masks);
2460 const uint64_t *bp = miniflow_get_values(&b->masks);
2463 MAP_FOR_EACH_INDEX(idx, b->masks.tnl_map) {
2464 uint64_t b_u64 = *bp++;
2466 /* 'b_u64' is non-zero, check if the data in 'a' is either zero
2467 * or misses some of the bits in 'b_u64'. */
2468 if (!(a->masks.tnl_map & (UINT64_C(1) << idx))
2469 || ((*miniflow_values_get__(ap, a->masks.tnl_map, idx) & b_u64)
2471 return true; /* 'a' wildcards some bits 'b' doesn't. */
2474 ap += count_1bits(a->masks.tnl_map); /* Skip tnl_map values. */
2475 MAP_FOR_EACH_INDEX(idx, b->masks.pkt_map) {
2476 uint64_t b_u64 = *bp++;
2478 if (!(a->masks.pkt_map & (UINT64_C(1) << idx))
2479 || ((*miniflow_values_get__(ap, a->masks.pkt_map, idx) & b_u64)
2481 return true; /* 'a' wildcards some bits 'b' doesn't. */