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"
32 #include "openvswitch/dynamic-string.h"
35 #include "openvswitch/match.h"
36 #include "dp-packet.h"
37 #include "openflow/openflow.h"
41 #include "unaligned.h"
44 COVERAGE_DEFINE(flow_extract);
45 COVERAGE_DEFINE(miniflow_malloc);
47 /* U64 indices for segmented flow classification. */
48 const uint8_t flow_segment_u64s[4] = {
49 FLOW_SEGMENT_1_ENDS_AT / sizeof(uint64_t),
50 FLOW_SEGMENT_2_ENDS_AT / sizeof(uint64_t),
51 FLOW_SEGMENT_3_ENDS_AT / sizeof(uint64_t),
55 /* Asserts that field 'f1' follows immediately after 'f0' in struct flow,
56 * without any intervening padding. */
57 #define ASSERT_SEQUENTIAL(f0, f1) \
58 BUILD_ASSERT_DECL(offsetof(struct flow, f0) \
59 + MEMBER_SIZEOF(struct flow, f0) \
60 == offsetof(struct flow, f1))
62 /* Asserts that fields 'f0' and 'f1' are in the same 32-bit aligned word within
64 #define ASSERT_SAME_WORD(f0, f1) \
65 BUILD_ASSERT_DECL(offsetof(struct flow, f0) / 4 \
66 == offsetof(struct flow, f1) / 4)
68 /* Asserts that 'f0' and 'f1' are both sequential and within the same 32-bit
69 * aligned word in struct flow. */
70 #define ASSERT_SEQUENTIAL_SAME_WORD(f0, f1) \
71 ASSERT_SEQUENTIAL(f0, f1); \
72 ASSERT_SAME_WORD(f0, f1)
74 /* miniflow_extract() assumes the following to be true to optimize the
75 * extraction process. */
76 ASSERT_SEQUENTIAL_SAME_WORD(dl_type, vlan_tci);
78 ASSERT_SEQUENTIAL_SAME_WORD(nw_frag, nw_tos);
79 ASSERT_SEQUENTIAL_SAME_WORD(nw_tos, nw_ttl);
80 ASSERT_SEQUENTIAL_SAME_WORD(nw_ttl, nw_proto);
82 /* TCP flags in the middle of a BE64, zeroes in the other half. */
83 BUILD_ASSERT_DECL(offsetof(struct flow, tcp_flags) % 8 == 4);
86 #define TCP_FLAGS_BE32(tcp_ctl) ((OVS_FORCE ovs_be32)TCP_FLAGS_BE16(tcp_ctl) \
89 #define TCP_FLAGS_BE32(tcp_ctl) ((OVS_FORCE ovs_be32)TCP_FLAGS_BE16(tcp_ctl))
92 ASSERT_SEQUENTIAL_SAME_WORD(tp_src, tp_dst);
94 /* Removes 'size' bytes from the head end of '*datap', of size '*sizep', which
95 * must contain at least 'size' bytes of data. Returns the first byte of data
97 static inline const void *
98 data_pull(const void **datap, size_t *sizep, size_t size)
100 const char *data = *datap;
101 *datap = data + size;
106 /* If '*datap' has at least 'size' bytes of data, removes that many bytes from
107 * the head end of '*datap' and returns the first byte removed. Otherwise,
108 * returns a null pointer without modifying '*datap'. */
109 static inline const void *
110 data_try_pull(const void **datap, size_t *sizep, size_t size)
112 return OVS_LIKELY(*sizep >= size) ? data_pull(datap, sizep, size) : NULL;
115 /* Context for pushing data to a miniflow. */
119 uint64_t * const end;
122 /* miniflow_push_* macros allow filling in a miniflow data values in order.
123 * Assertions are needed only when the layout of the struct flow is modified.
124 * 'ofs' is a compile-time constant, which allows most of the code be optimized
125 * away. Some GCC versions gave warnings on ALWAYS_INLINE, so these are
126 * defined as macros. */
128 #if (FLOW_WC_SEQ != 36)
129 #define MINIFLOW_ASSERT(X) ovs_assert(X)
130 BUILD_MESSAGE("FLOW_WC_SEQ changed: miniflow_extract() will have runtime "
131 "assertions enabled. Consider updating FLOW_WC_SEQ after "
134 #define MINIFLOW_ASSERT(X)
137 /* True if 'IDX' and higher bits are not set. */
138 #define ASSERT_FLOWMAP_NOT_SET(FM, IDX) \
140 MINIFLOW_ASSERT(!((FM)->bits[(IDX) / MAP_T_BITS] & \
141 (MAP_MAX << ((IDX) % MAP_T_BITS)))); \
142 for (size_t i = (IDX) / MAP_T_BITS + 1; i < FLOWMAP_UNITS; i++) { \
143 MINIFLOW_ASSERT(!(FM)->bits[i]); \
147 #define miniflow_set_map(MF, OFS) \
149 ASSERT_FLOWMAP_NOT_SET(&MF.map, (OFS)); \
150 flowmap_set(&MF.map, (OFS), 1); \
153 #define miniflow_assert_in_map(MF, OFS) \
154 MINIFLOW_ASSERT(flowmap_is_set(&MF.map, (OFS))); \
155 ASSERT_FLOWMAP_NOT_SET(&MF.map, (OFS) + 1)
157 #define miniflow_push_uint64_(MF, OFS, VALUE) \
159 MINIFLOW_ASSERT(MF.data < MF.end && (OFS) % 8 == 0); \
160 *MF.data++ = VALUE; \
161 miniflow_set_map(MF, OFS / 8); \
164 #define miniflow_push_be64_(MF, OFS, VALUE) \
165 miniflow_push_uint64_(MF, OFS, (OVS_FORCE uint64_t)(VALUE))
167 #define miniflow_push_uint32_(MF, OFS, VALUE) \
169 MINIFLOW_ASSERT(MF.data < MF.end); \
171 if ((OFS) % 8 == 0) { \
172 miniflow_set_map(MF, OFS / 8); \
173 *(uint32_t *)MF.data = VALUE; \
174 } else if ((OFS) % 8 == 4) { \
175 miniflow_assert_in_map(MF, OFS / 8); \
176 *((uint32_t *)MF.data + 1) = VALUE; \
181 #define miniflow_push_be32_(MF, OFS, VALUE) \
182 miniflow_push_uint32_(MF, OFS, (OVS_FORCE uint32_t)(VALUE))
184 #define miniflow_push_uint16_(MF, OFS, VALUE) \
186 MINIFLOW_ASSERT(MF.data < MF.end); \
188 if ((OFS) % 8 == 0) { \
189 miniflow_set_map(MF, OFS / 8); \
190 *(uint16_t *)MF.data = VALUE; \
191 } else if ((OFS) % 8 == 2) { \
192 miniflow_assert_in_map(MF, OFS / 8); \
193 *((uint16_t *)MF.data + 1) = VALUE; \
194 } else if ((OFS) % 8 == 4) { \
195 miniflow_assert_in_map(MF, OFS / 8); \
196 *((uint16_t *)MF.data + 2) = VALUE; \
197 } else if ((OFS) % 8 == 6) { \
198 miniflow_assert_in_map(MF, OFS / 8); \
199 *((uint16_t *)MF.data + 3) = VALUE; \
204 #define miniflow_push_uint8_(MF, OFS, VALUE) \
206 MINIFLOW_ASSERT(MF.data < MF.end); \
208 if ((OFS) % 8 == 0) { \
209 miniflow_set_map(MF, OFS / 8); \
210 *(uint8_t *)MF.data = VALUE; \
211 } else if ((OFS) % 8 == 7) { \
212 miniflow_assert_in_map(MF, OFS / 8); \
213 *((uint8_t *)MF.data + 7) = VALUE; \
216 miniflow_assert_in_map(MF, OFS / 8); \
217 *((uint8_t *)MF.data + ((OFS) % 8)) = VALUE; \
221 #define miniflow_pad_to_64_(MF, OFS) \
223 MINIFLOW_ASSERT((OFS) % 8 != 0); \
224 miniflow_assert_in_map(MF, OFS / 8); \
226 memset((uint8_t *)MF.data + (OFS) % 8, 0, 8 - (OFS) % 8); \
230 #define miniflow_pad_from_64_(MF, OFS) \
232 MINIFLOW_ASSERT(MF.data < MF.end); \
234 MINIFLOW_ASSERT((OFS) % 8 != 0); \
235 miniflow_set_map(MF, OFS / 8); \
237 memset((uint8_t *)MF.data, 0, (OFS) % 8); \
240 #define miniflow_push_be16_(MF, OFS, VALUE) \
241 miniflow_push_uint16_(MF, OFS, (OVS_FORCE uint16_t)VALUE);
243 #define miniflow_push_be8_(MF, OFS, VALUE) \
244 miniflow_push_uint8_(MF, OFS, (OVS_FORCE uint8_t)VALUE);
246 #define miniflow_set_maps(MF, OFS, N_WORDS) \
248 size_t ofs = (OFS); \
249 size_t n_words = (N_WORDS); \
251 MINIFLOW_ASSERT(n_words && MF.data + n_words <= MF.end); \
252 ASSERT_FLOWMAP_NOT_SET(&MF.map, ofs); \
253 flowmap_set(&MF.map, ofs, n_words); \
256 /* Data at 'valuep' may be unaligned. */
257 #define miniflow_push_words_(MF, OFS, VALUEP, N_WORDS) \
259 MINIFLOW_ASSERT((OFS) % 8 == 0); \
260 miniflow_set_maps(MF, (OFS) / 8, (N_WORDS)); \
261 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof *MF.data); \
262 MF.data += (N_WORDS); \
265 /* Push 32-bit words padded to 64-bits. */
266 #define miniflow_push_words_32_(MF, OFS, VALUEP, N_WORDS) \
268 miniflow_set_maps(MF, (OFS) / 8, DIV_ROUND_UP(N_WORDS, 2)); \
269 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof(uint32_t)); \
270 MF.data += DIV_ROUND_UP(N_WORDS, 2); \
271 if ((N_WORDS) & 1) { \
272 *((uint32_t *)MF.data - 1) = 0; \
276 /* Data at 'valuep' may be unaligned. */
277 /* MACs start 64-aligned, and must be followed by other data or padding. */
278 #define miniflow_push_macs_(MF, OFS, VALUEP) \
280 miniflow_set_maps(MF, (OFS) / 8, 2); \
281 memcpy(MF.data, (VALUEP), 2 * ETH_ADDR_LEN); \
282 MF.data += 1; /* First word only. */ \
285 #define miniflow_push_uint32(MF, FIELD, VALUE) \
286 miniflow_push_uint32_(MF, offsetof(struct flow, FIELD), VALUE)
288 #define miniflow_push_be32(MF, FIELD, VALUE) \
289 miniflow_push_be32_(MF, offsetof(struct flow, FIELD), VALUE)
291 #define miniflow_push_uint16(MF, FIELD, VALUE) \
292 miniflow_push_uint16_(MF, offsetof(struct flow, FIELD), VALUE)
294 #define miniflow_push_be16(MF, FIELD, VALUE) \
295 miniflow_push_be16_(MF, offsetof(struct flow, FIELD), VALUE)
297 #define miniflow_push_uint8(MF, FIELD, VALUE) \
298 miniflow_push_uint8_(MF, offsetof(struct flow, FIELD), VALUE)
300 #define miniflow_pad_to_64(MF, FIELD) \
301 miniflow_pad_to_64_(MF, OFFSETOFEND(struct flow, FIELD))
303 #define miniflow_pad_from_64(MF, FIELD) \
304 miniflow_pad_from_64_(MF, offsetof(struct flow, FIELD))
306 #define miniflow_push_words(MF, FIELD, VALUEP, N_WORDS) \
307 miniflow_push_words_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
309 #define miniflow_push_words_32(MF, FIELD, VALUEP, N_WORDS) \
310 miniflow_push_words_32_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
312 #define miniflow_push_macs(MF, FIELD, VALUEP) \
313 miniflow_push_macs_(MF, offsetof(struct flow, FIELD), VALUEP)
315 /* Pulls the MPLS headers at '*datap' and returns the count of them. */
317 parse_mpls(const void **datap, size_t *sizep)
319 const struct mpls_hdr *mh;
322 while ((mh = data_try_pull(datap, sizep, sizeof *mh))) {
324 if (mh->mpls_lse.lo & htons(1 << MPLS_BOS_SHIFT)) {
328 return MIN(count, FLOW_MAX_MPLS_LABELS);
331 static inline ovs_be16
332 parse_vlan(const void **datap, size_t *sizep)
334 const struct eth_header *eth = *datap;
337 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
341 data_pull(datap, sizep, ETH_ADDR_LEN * 2);
343 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
344 if (OVS_LIKELY(*sizep
345 >= sizeof(struct qtag_prefix) + sizeof(ovs_be16))) {
346 const struct qtag_prefix *qp = data_pull(datap, sizep, sizeof *qp);
347 return qp->tci | htons(VLAN_CFI);
353 static inline ovs_be16
354 parse_ethertype(const void **datap, size_t *sizep)
356 const struct llc_snap_header *llc;
359 proto = *(ovs_be16 *) data_pull(datap, sizep, sizeof proto);
360 if (OVS_LIKELY(ntohs(proto) >= ETH_TYPE_MIN)) {
364 if (OVS_UNLIKELY(*sizep < sizeof *llc)) {
365 return htons(FLOW_DL_TYPE_NONE);
369 if (OVS_UNLIKELY(llc->llc.llc_dsap != LLC_DSAP_SNAP
370 || llc->llc.llc_ssap != LLC_SSAP_SNAP
371 || llc->llc.llc_cntl != LLC_CNTL_SNAP
372 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
373 sizeof llc->snap.snap_org))) {
374 return htons(FLOW_DL_TYPE_NONE);
377 data_pull(datap, sizep, sizeof *llc);
379 if (OVS_LIKELY(ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN)) {
380 return llc->snap.snap_type;
383 return htons(FLOW_DL_TYPE_NONE);
387 parse_icmpv6(const void **datap, size_t *sizep, const struct icmp6_hdr *icmp,
388 const struct in6_addr **nd_target,
389 struct eth_addr arp_buf[2])
391 if (icmp->icmp6_code == 0 &&
392 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
393 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
395 *nd_target = data_try_pull(datap, sizep, sizeof **nd_target);
396 if (OVS_UNLIKELY(!*nd_target)) {
400 while (*sizep >= 8) {
401 /* The minimum size of an option is 8 bytes, which also is
402 * the size of Ethernet link-layer options. */
403 const struct ovs_nd_opt *nd_opt = *datap;
404 int opt_len = nd_opt->nd_opt_len * ND_OPT_LEN;
406 if (!opt_len || opt_len > *sizep) {
410 /* Store the link layer address if the appropriate option is
411 * provided. It is considered an error if the same link
412 * layer option is specified twice. */
413 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
415 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[0]))) {
416 arp_buf[0] = nd_opt->nd_opt_mac;
420 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
422 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[1]))) {
423 arp_buf[1] = nd_opt->nd_opt_mac;
429 if (OVS_UNLIKELY(!data_try_pull(datap, sizep, opt_len))) {
439 arp_buf[0] = eth_addr_zero;
440 arp_buf[1] = eth_addr_zero;
443 /* Initializes 'flow' members from 'packet' and 'md'
445 * Initializes 'packet' header l2 pointer to the start of the Ethernet
446 * header, and the layer offsets as follows:
448 * - packet->l2_5_ofs to the start of the MPLS shim header, or UINT16_MAX
449 * when there is no MPLS shim header.
451 * - packet->l3_ofs to just past the Ethernet header, or just past the
452 * vlan_header if one is present, to the first byte of the payload of the
453 * Ethernet frame. UINT16_MAX if the frame is too short to contain an
456 * - packet->l4_ofs to just past the IPv4 header, if one is present and
457 * has at least the content used for the fields of interest for the flow,
458 * otherwise UINT16_MAX.
461 flow_extract(struct dp_packet *packet, struct flow *flow)
465 uint64_t buf[FLOW_U64S];
468 COVERAGE_INC(flow_extract);
470 miniflow_extract(packet, &m.mf);
471 miniflow_expand(&m.mf, flow);
474 /* Caller is responsible for initializing 'dst' with enough storage for
475 * FLOW_U64S * 8 bytes. */
477 miniflow_extract(struct dp_packet *packet, struct miniflow *dst)
479 const struct pkt_metadata *md = &packet->md;
480 const void *data = dp_packet_data(packet);
481 size_t size = dp_packet_size(packet);
482 uint64_t *values = miniflow_values(dst);
483 struct mf_ctx mf = { FLOWMAP_EMPTY_INITIALIZER, values,
484 values + FLOW_U64S };
487 uint8_t nw_frag, nw_tos, nw_ttl, nw_proto;
490 if (flow_tnl_dst_is_set(&md->tunnel)) {
491 miniflow_push_words(mf, tunnel, &md->tunnel,
492 offsetof(struct flow_tnl, metadata) /
495 if (!(md->tunnel.flags & FLOW_TNL_F_UDPIF)) {
496 if (md->tunnel.metadata.present.map) {
497 miniflow_push_words(mf, tunnel.metadata, &md->tunnel.metadata,
498 sizeof md->tunnel.metadata /
502 if (md->tunnel.metadata.present.len) {
503 miniflow_push_words(mf, tunnel.metadata.present,
504 &md->tunnel.metadata.present, 1);
505 miniflow_push_words(mf, tunnel.metadata.opts.gnv,
506 md->tunnel.metadata.opts.gnv,
507 DIV_ROUND_UP(md->tunnel.metadata.present.len,
512 if (md->skb_priority || md->pkt_mark) {
513 miniflow_push_uint32(mf, skb_priority, md->skb_priority);
514 miniflow_push_uint32(mf, pkt_mark, md->pkt_mark);
516 miniflow_push_uint32(mf, dp_hash, md->dp_hash);
517 miniflow_push_uint32(mf, in_port, odp_to_u32(md->in_port.odp_port));
518 if (md->recirc_id || md->ct_state) {
519 miniflow_push_uint32(mf, recirc_id, md->recirc_id);
520 miniflow_push_uint16(mf, ct_state, md->ct_state);
521 miniflow_push_uint16(mf, ct_zone, md->ct_zone);
525 miniflow_push_uint32(mf, ct_mark, md->ct_mark);
526 miniflow_pad_to_64(mf, ct_mark);
528 if (!ovs_u128_is_zero(md->ct_label)) {
529 miniflow_push_words(mf, ct_label, &md->ct_label,
530 sizeof md->ct_label / sizeof(uint64_t));
534 /* Initialize packet's layer pointer and offsets. */
536 dp_packet_reset_offsets(packet);
538 /* Must have full Ethernet header to proceed. */
539 if (OVS_UNLIKELY(size < sizeof(struct eth_header))) {
545 ASSERT_SEQUENTIAL(dl_dst, dl_src);
546 miniflow_push_macs(mf, dl_dst, data);
547 /* dl_type, vlan_tci. */
548 vlan_tci = parse_vlan(&data, &size);
549 dl_type = parse_ethertype(&data, &size);
550 miniflow_push_be16(mf, dl_type, dl_type);
551 miniflow_push_be16(mf, vlan_tci, vlan_tci);
555 if (OVS_UNLIKELY(eth_type_mpls(dl_type))) {
557 const void *mpls = data;
559 packet->l2_5_ofs = (char *)data - l2;
560 count = parse_mpls(&data, &size);
561 miniflow_push_words_32(mf, mpls_lse, mpls, count);
565 packet->l3_ofs = (char *)data - l2;
568 if (OVS_LIKELY(dl_type == htons(ETH_TYPE_IP))) {
569 const struct ip_header *nh = data;
573 if (OVS_UNLIKELY(size < IP_HEADER_LEN)) {
576 ip_len = IP_IHL(nh->ip_ihl_ver) * 4;
578 if (OVS_UNLIKELY(ip_len < IP_HEADER_LEN)) {
581 if (OVS_UNLIKELY(size < ip_len)) {
584 tot_len = ntohs(nh->ip_tot_len);
585 if (OVS_UNLIKELY(tot_len > size)) {
588 if (OVS_UNLIKELY(size - tot_len > UINT8_MAX)) {
591 dp_packet_set_l2_pad_size(packet, size - tot_len);
592 size = tot_len; /* Never pull padding. */
594 /* Push both source and destination address at once. */
595 miniflow_push_words(mf, nw_src, &nh->ip_src, 1);
597 miniflow_push_be32(mf, ipv6_label, 0); /* Padding for IPv4. */
601 nw_proto = nh->ip_proto;
602 if (OVS_UNLIKELY(IP_IS_FRAGMENT(nh->ip_frag_off))) {
603 nw_frag = FLOW_NW_FRAG_ANY;
604 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
605 nw_frag |= FLOW_NW_FRAG_LATER;
608 data_pull(&data, &size, ip_len);
609 } else if (dl_type == htons(ETH_TYPE_IPV6)) {
610 const struct ovs_16aligned_ip6_hdr *nh;
614 if (OVS_UNLIKELY(size < sizeof *nh)) {
617 nh = data_pull(&data, &size, sizeof *nh);
619 plen = ntohs(nh->ip6_plen);
620 if (OVS_UNLIKELY(plen > size)) {
623 /* Jumbo Payload option not supported yet. */
624 if (OVS_UNLIKELY(size - plen > UINT8_MAX)) {
627 dp_packet_set_l2_pad_size(packet, size - plen);
628 size = plen; /* Never pull padding. */
630 miniflow_push_words(mf, ipv6_src, &nh->ip6_src,
631 sizeof nh->ip6_src / 8);
632 miniflow_push_words(mf, ipv6_dst, &nh->ip6_dst,
633 sizeof nh->ip6_dst / 8);
635 tc_flow = get_16aligned_be32(&nh->ip6_flow);
637 ovs_be32 label = tc_flow & htonl(IPV6_LABEL_MASK);
638 miniflow_push_be32(mf, ipv6_label, label);
641 nw_tos = ntohl(tc_flow) >> 20;
642 nw_ttl = nh->ip6_hlim;
643 nw_proto = nh->ip6_nxt;
646 if (OVS_LIKELY((nw_proto != IPPROTO_HOPOPTS)
647 && (nw_proto != IPPROTO_ROUTING)
648 && (nw_proto != IPPROTO_DSTOPTS)
649 && (nw_proto != IPPROTO_AH)
650 && (nw_proto != IPPROTO_FRAGMENT))) {
651 /* It's either a terminal header (e.g., TCP, UDP) or one we
652 * don't understand. In either case, we're done with the
653 * packet, so use it to fill in 'nw_proto'. */
657 /* We only verify that at least 8 bytes of the next header are
658 * available, but many of these headers are longer. Ensure that
659 * accesses within the extension header are within those first 8
660 * bytes. All extension headers are required to be at least 8
662 if (OVS_UNLIKELY(size < 8)) {
666 if ((nw_proto == IPPROTO_HOPOPTS)
667 || (nw_proto == IPPROTO_ROUTING)
668 || (nw_proto == IPPROTO_DSTOPTS)) {
669 /* These headers, while different, have the fields we care
670 * about in the same location and with the same
672 const struct ip6_ext *ext_hdr = data;
673 nw_proto = ext_hdr->ip6e_nxt;
674 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
675 (ext_hdr->ip6e_len + 1) * 8))) {
678 } else if (nw_proto == IPPROTO_AH) {
679 /* A standard AH definition isn't available, but the fields
680 * we care about are in the same location as the generic
681 * option header--only the header length is calculated
683 const struct ip6_ext *ext_hdr = data;
684 nw_proto = ext_hdr->ip6e_nxt;
685 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
686 (ext_hdr->ip6e_len + 2) * 4))) {
689 } else if (nw_proto == IPPROTO_FRAGMENT) {
690 const struct ovs_16aligned_ip6_frag *frag_hdr = data;
692 nw_proto = frag_hdr->ip6f_nxt;
693 if (!data_try_pull(&data, &size, sizeof *frag_hdr)) {
697 /* We only process the first fragment. */
698 if (frag_hdr->ip6f_offlg != htons(0)) {
699 nw_frag = FLOW_NW_FRAG_ANY;
700 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
701 nw_frag |= FLOW_NW_FRAG_LATER;
702 nw_proto = IPPROTO_FRAGMENT;
709 if (dl_type == htons(ETH_TYPE_ARP) ||
710 dl_type == htons(ETH_TYPE_RARP)) {
711 struct eth_addr arp_buf[2];
712 const struct arp_eth_header *arp = (const struct arp_eth_header *)
713 data_try_pull(&data, &size, ARP_ETH_HEADER_LEN);
715 if (OVS_LIKELY(arp) && OVS_LIKELY(arp->ar_hrd == htons(1))
716 && OVS_LIKELY(arp->ar_pro == htons(ETH_TYPE_IP))
717 && OVS_LIKELY(arp->ar_hln == ETH_ADDR_LEN)
718 && OVS_LIKELY(arp->ar_pln == 4)) {
719 miniflow_push_be32(mf, nw_src,
720 get_16aligned_be32(&arp->ar_spa));
721 miniflow_push_be32(mf, nw_dst,
722 get_16aligned_be32(&arp->ar_tpa));
724 /* We only match on the lower 8 bits of the opcode. */
725 if (OVS_LIKELY(ntohs(arp->ar_op) <= 0xff)) {
726 miniflow_push_be32(mf, ipv6_label, 0); /* Pad with ARP. */
727 miniflow_push_be32(mf, nw_frag, htonl(ntohs(arp->ar_op)));
730 /* Must be adjacent. */
731 ASSERT_SEQUENTIAL(arp_sha, arp_tha);
733 arp_buf[0] = arp->ar_sha;
734 arp_buf[1] = arp->ar_tha;
735 miniflow_push_macs(mf, arp_sha, arp_buf);
736 miniflow_pad_to_64(mf, arp_tha);
742 packet->l4_ofs = (char *)data - l2;
743 miniflow_push_be32(mf, nw_frag,
744 BYTES_TO_BE32(nw_frag, nw_tos, nw_ttl, nw_proto));
746 if (OVS_LIKELY(!(nw_frag & FLOW_NW_FRAG_LATER))) {
747 if (OVS_LIKELY(nw_proto == IPPROTO_TCP)) {
748 if (OVS_LIKELY(size >= TCP_HEADER_LEN)) {
749 const struct tcp_header *tcp = data;
751 miniflow_push_be32(mf, arp_tha.ea[2], 0);
752 miniflow_push_be32(mf, tcp_flags,
753 TCP_FLAGS_BE32(tcp->tcp_ctl));
754 miniflow_push_be16(mf, tp_src, tcp->tcp_src);
755 miniflow_push_be16(mf, tp_dst, tcp->tcp_dst);
756 miniflow_pad_to_64(mf, tp_dst);
758 } else if (OVS_LIKELY(nw_proto == IPPROTO_UDP)) {
759 if (OVS_LIKELY(size >= UDP_HEADER_LEN)) {
760 const struct udp_header *udp = data;
762 miniflow_push_be16(mf, tp_src, udp->udp_src);
763 miniflow_push_be16(mf, tp_dst, udp->udp_dst);
764 miniflow_pad_to_64(mf, tp_dst);
766 } else if (OVS_LIKELY(nw_proto == IPPROTO_SCTP)) {
767 if (OVS_LIKELY(size >= SCTP_HEADER_LEN)) {
768 const struct sctp_header *sctp = data;
770 miniflow_push_be16(mf, tp_src, sctp->sctp_src);
771 miniflow_push_be16(mf, tp_dst, sctp->sctp_dst);
772 miniflow_pad_to_64(mf, tp_dst);
774 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMP)) {
775 if (OVS_LIKELY(size >= ICMP_HEADER_LEN)) {
776 const struct icmp_header *icmp = data;
778 miniflow_push_be16(mf, tp_src, htons(icmp->icmp_type));
779 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp_code));
780 miniflow_pad_to_64(mf, tp_dst);
782 } else if (OVS_LIKELY(nw_proto == IPPROTO_IGMP)) {
783 if (OVS_LIKELY(size >= IGMP_HEADER_LEN)) {
784 const struct igmp_header *igmp = data;
786 miniflow_push_be16(mf, tp_src, htons(igmp->igmp_type));
787 miniflow_push_be16(mf, tp_dst, htons(igmp->igmp_code));
788 miniflow_push_be32(mf, igmp_group_ip4,
789 get_16aligned_be32(&igmp->group));
791 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMPV6)) {
792 if (OVS_LIKELY(size >= sizeof(struct icmp6_hdr))) {
793 const struct in6_addr *nd_target = NULL;
794 struct eth_addr arp_buf[2] = { { { { 0 } } } };
795 const struct icmp6_hdr *icmp = data_pull(&data, &size,
797 parse_icmpv6(&data, &size, icmp, &nd_target, arp_buf);
799 miniflow_push_words(mf, nd_target, nd_target,
800 sizeof *nd_target / sizeof(uint64_t));
802 miniflow_push_macs(mf, arp_sha, arp_buf);
803 miniflow_pad_to_64(mf, arp_tha);
804 miniflow_push_be16(mf, tp_src, htons(icmp->icmp6_type));
805 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp6_code));
806 miniflow_pad_to_64(mf, tp_dst);
814 /* For every bit of a field that is wildcarded in 'wildcards', sets the
815 * corresponding bit in 'flow' to zero. */
817 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
819 uint64_t *flow_u64 = (uint64_t *) flow;
820 const uint64_t *wc_u64 = (const uint64_t *) &wildcards->masks;
823 for (i = 0; i < FLOW_U64S; i++) {
824 flow_u64[i] &= wc_u64[i];
829 flow_unwildcard_tp_ports(const struct flow *flow, struct flow_wildcards *wc)
831 if (flow->nw_proto != IPPROTO_ICMP) {
832 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
833 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
835 wc->masks.tp_src = htons(0xff);
836 wc->masks.tp_dst = htons(0xff);
840 /* Initializes 'flow_metadata' with the metadata found in 'flow'. */
842 flow_get_metadata(const struct flow *flow, struct match *flow_metadata)
846 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 36);
848 match_init_catchall(flow_metadata);
849 if (flow->tunnel.tun_id != htonll(0)) {
850 match_set_tun_id(flow_metadata, flow->tunnel.tun_id);
852 if (flow->tunnel.flags & FLOW_TNL_PUB_F_MASK) {
853 match_set_tun_flags(flow_metadata,
854 flow->tunnel.flags & FLOW_TNL_PUB_F_MASK);
856 if (flow->tunnel.ip_src) {
857 match_set_tun_src(flow_metadata, flow->tunnel.ip_src);
859 if (flow->tunnel.ip_dst) {
860 match_set_tun_dst(flow_metadata, flow->tunnel.ip_dst);
862 if (ipv6_addr_is_set(&flow->tunnel.ipv6_src)) {
863 match_set_tun_ipv6_src(flow_metadata, &flow->tunnel.ipv6_src);
865 if (ipv6_addr_is_set(&flow->tunnel.ipv6_dst)) {
866 match_set_tun_ipv6_dst(flow_metadata, &flow->tunnel.ipv6_dst);
868 if (flow->tunnel.gbp_id != htons(0)) {
869 match_set_tun_gbp_id(flow_metadata, flow->tunnel.gbp_id);
871 if (flow->tunnel.gbp_flags) {
872 match_set_tun_gbp_flags(flow_metadata, flow->tunnel.gbp_flags);
874 tun_metadata_get_fmd(&flow->tunnel, flow_metadata);
875 if (flow->metadata != htonll(0)) {
876 match_set_metadata(flow_metadata, flow->metadata);
879 for (i = 0; i < FLOW_N_REGS; i++) {
881 match_set_reg(flow_metadata, i, flow->regs[i]);
885 if (flow->pkt_mark != 0) {
886 match_set_pkt_mark(flow_metadata, flow->pkt_mark);
889 match_set_in_port(flow_metadata, flow->in_port.ofp_port);
890 if (flow->ct_state != 0) {
891 match_set_ct_state(flow_metadata, flow->ct_state);
893 if (flow->ct_zone != 0) {
894 match_set_ct_zone(flow_metadata, flow->ct_zone);
896 if (flow->ct_mark != 0) {
897 match_set_ct_mark(flow_metadata, flow->ct_mark);
899 if (!ovs_u128_is_zero(flow->ct_label)) {
900 match_set_ct_label(flow_metadata, flow->ct_label);
904 const char *ct_state_to_string(uint32_t state)
929 flow_to_string(const struct flow *flow)
931 struct ds ds = DS_EMPTY_INITIALIZER;
932 flow_format(&ds, flow);
937 flow_tun_flag_to_string(uint32_t flags)
940 case FLOW_TNL_F_DONT_FRAGMENT:
942 case FLOW_TNL_F_CSUM:
954 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
955 uint32_t flags, char del)
960 ds_put_char(ds, '0');
964 uint32_t bit = rightmost_1bit(flags);
967 s = bit_to_string(bit);
969 ds_put_format(ds, "%s%c", s, del);
978 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
984 format_flags_masked(struct ds *ds, const char *name,
985 const char *(*bit_to_string)(uint32_t), uint32_t flags,
986 uint32_t mask, uint32_t max_mask)
989 ds_put_format(ds, "%s%s=%s", colors.param, name, colors.end);
992 if (mask == max_mask) {
993 format_flags(ds, bit_to_string, flags, '|');
998 ds_put_cstr(ds, "0/0");
1003 uint32_t bit = rightmost_1bit(mask);
1004 const char *s = bit_to_string(bit);
1006 ds_put_format(ds, "%s%s", (flags & bit) ? "+" : "-",
1007 s ? s : "[Unknown]");
1012 /* Scans a string 's' of flags to determine their numerical value and
1013 * returns the number of characters parsed using 'bit_to_string' to
1014 * lookup flag names. Scanning continues until the character 'end' is
1017 * In the event of a failure, a negative error code will be returned. In
1018 * addition, if 'res_string' is non-NULL then a descriptive string will
1019 * be returned incorporating the identifying string 'field_name'. This
1020 * error string must be freed by the caller.
1022 * Upon success, the flag values will be stored in 'res_flags' and
1023 * optionally 'res_mask', if it is non-NULL (if it is NULL then any masks
1024 * present in the original string will be considered an error). The
1025 * caller may restrict the acceptable set of values through the mask
1028 parse_flags(const char *s, const char *(*bit_to_string)(uint32_t),
1029 char end, const char *field_name, char **res_string,
1030 uint32_t *res_flags, uint32_t allowed, uint32_t *res_mask)
1032 uint32_t result = 0;
1035 /* Parse masked flags in numeric format? */
1036 if (res_mask && ovs_scan(s, "%"SCNi32"/%"SCNi32"%n",
1037 res_flags, res_mask, &n) && n > 0) {
1038 if (*res_flags & ~allowed || *res_mask & ~allowed) {
1046 if (res_mask && (*s == '+' || *s == '-')) {
1047 uint32_t flags = 0, mask = 0;
1049 /* Parse masked flags. */
1050 while (s[0] != end) {
1057 } else if (s[0] == '-') {
1061 *res_string = xasprintf("%s: %s must be preceded by '+' "
1062 "(for SET) or '-' (NOT SET)", s,
1070 for (bit = 1; bit; bit <<= 1) {
1071 const char *fname = bit_to_string(bit);
1077 len = strlen(fname);
1078 if (strncmp(s, fname, len) ||
1079 (s[len] != '+' && s[len] != '-' && s[len] != end)) {
1084 /* bit already set. */
1086 *res_string = xasprintf("%s: Each %s flag can be "
1087 "specified only once", s,
1092 if (!(bit & allowed)) {
1114 /* Parse unmasked flags. If a flag is present, it is set, otherwise
1116 while (s[n] != end) {
1117 unsigned long long int flags;
1121 if (ovs_scan(&s[n], "%lli%n", &flags, &n0)) {
1122 if (flags & ~allowed) {
1125 n += n0 + (s[n + n0] == '|');
1130 for (bit = 1; bit; bit <<= 1) {
1131 const char *name = bit_to_string(bit);
1139 if (!strncmp(s + n, name, len) &&
1140 (s[n + len] == '|' || s[n + len] == end)) {
1141 if (!(bit & allowed)) {
1145 n += len + (s[n + len] == '|');
1155 *res_flags = result;
1157 *res_mask = UINT32_MAX;
1166 *res_string = xasprintf("%s: unknown %s flag(s)", s, field_name);
1172 flow_format(struct ds *ds, const struct flow *flow)
1175 struct flow_wildcards *wc = &match.wc;
1177 match_wc_init(&match, flow);
1179 /* As this function is most often used for formatting a packet in a
1180 * packet-in message, skip formatting the packet context fields that are
1181 * all-zeroes to make the print-out easier on the eyes. This means that a
1182 * missing context field implies a zero value for that field. This is
1183 * similar to OpenFlow encoding of these fields, as the specification
1184 * states that all-zeroes context fields should not be encoded in the
1185 * packet-in messages. */
1186 if (!flow->in_port.ofp_port) {
1187 WC_UNMASK_FIELD(wc, in_port);
1189 if (!flow->skb_priority) {
1190 WC_UNMASK_FIELD(wc, skb_priority);
1192 if (!flow->pkt_mark) {
1193 WC_UNMASK_FIELD(wc, pkt_mark);
1195 if (!flow->recirc_id) {
1196 WC_UNMASK_FIELD(wc, recirc_id);
1198 if (!flow->dp_hash) {
1199 WC_UNMASK_FIELD(wc, dp_hash);
1201 if (!flow->ct_state) {
1202 WC_UNMASK_FIELD(wc, ct_state);
1204 if (!flow->ct_zone) {
1205 WC_UNMASK_FIELD(wc, ct_zone);
1207 if (!flow->ct_mark) {
1208 WC_UNMASK_FIELD(wc, ct_mark);
1210 if (ovs_u128_is_zero(flow->ct_label)) {
1211 WC_UNMASK_FIELD(wc, ct_label);
1213 for (int i = 0; i < FLOW_N_REGS; i++) {
1214 if (!flow->regs[i]) {
1215 WC_UNMASK_FIELD(wc, regs[i]);
1218 if (!flow->metadata) {
1219 WC_UNMASK_FIELD(wc, metadata);
1222 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
1226 flow_print(FILE *stream, const struct flow *flow)
1228 char *s = flow_to_string(flow);
1233 /* flow_wildcards functions. */
1235 /* Initializes 'wc' as a set of wildcards that matches every packet. */
1237 flow_wildcards_init_catchall(struct flow_wildcards *wc)
1239 memset(&wc->masks, 0, sizeof wc->masks);
1242 /* Converts a flow into flow wildcards. It sets the wildcard masks based on
1243 * the packet headers extracted to 'flow'. It will not set the mask for fields
1244 * that do not make sense for the packet type. OpenFlow-only metadata is
1245 * wildcarded, but other metadata is unconditionally exact-matched. */
1246 void flow_wildcards_init_for_packet(struct flow_wildcards *wc,
1247 const struct flow *flow)
1249 memset(&wc->masks, 0x0, sizeof wc->masks);
1251 /* Update this function whenever struct flow changes. */
1252 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 36);
1254 if (flow_tnl_dst_is_set(&flow->tunnel)) {
1255 if (flow->tunnel.flags & FLOW_TNL_F_KEY) {
1256 WC_MASK_FIELD(wc, tunnel.tun_id);
1258 WC_MASK_FIELD(wc, tunnel.ip_src);
1259 WC_MASK_FIELD(wc, tunnel.ip_dst);
1260 WC_MASK_FIELD(wc, tunnel.ipv6_src);
1261 WC_MASK_FIELD(wc, tunnel.ipv6_dst);
1262 WC_MASK_FIELD(wc, tunnel.flags);
1263 WC_MASK_FIELD(wc, tunnel.ip_tos);
1264 WC_MASK_FIELD(wc, tunnel.ip_ttl);
1265 WC_MASK_FIELD(wc, tunnel.tp_src);
1266 WC_MASK_FIELD(wc, tunnel.tp_dst);
1267 WC_MASK_FIELD(wc, tunnel.gbp_id);
1268 WC_MASK_FIELD(wc, tunnel.gbp_flags);
1270 if (!(flow->tunnel.flags & FLOW_TNL_F_UDPIF)) {
1271 if (flow->tunnel.metadata.present.map) {
1272 wc->masks.tunnel.metadata.present.map =
1273 flow->tunnel.metadata.present.map;
1274 WC_MASK_FIELD(wc, tunnel.metadata.opts.u8);
1277 WC_MASK_FIELD(wc, tunnel.metadata.present.len);
1278 memset(wc->masks.tunnel.metadata.opts.gnv, 0xff,
1279 flow->tunnel.metadata.present.len);
1281 } else if (flow->tunnel.tun_id) {
1282 WC_MASK_FIELD(wc, tunnel.tun_id);
1285 /* metadata, regs, and conj_id wildcarded. */
1287 WC_MASK_FIELD(wc, skb_priority);
1288 WC_MASK_FIELD(wc, pkt_mark);
1289 WC_MASK_FIELD(wc, ct_state);
1290 WC_MASK_FIELD(wc, ct_zone);
1291 WC_MASK_FIELD(wc, ct_mark);
1292 WC_MASK_FIELD(wc, ct_label);
1293 WC_MASK_FIELD(wc, recirc_id);
1294 WC_MASK_FIELD(wc, dp_hash);
1295 WC_MASK_FIELD(wc, in_port);
1297 /* actset_output wildcarded. */
1299 WC_MASK_FIELD(wc, dl_dst);
1300 WC_MASK_FIELD(wc, dl_src);
1301 WC_MASK_FIELD(wc, dl_type);
1302 WC_MASK_FIELD(wc, vlan_tci);
1304 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1305 WC_MASK_FIELD(wc, nw_src);
1306 WC_MASK_FIELD(wc, nw_dst);
1307 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1308 WC_MASK_FIELD(wc, ipv6_src);
1309 WC_MASK_FIELD(wc, ipv6_dst);
1310 WC_MASK_FIELD(wc, ipv6_label);
1311 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1312 flow->dl_type == htons(ETH_TYPE_RARP)) {
1313 WC_MASK_FIELD(wc, nw_src);
1314 WC_MASK_FIELD(wc, nw_dst);
1315 WC_MASK_FIELD(wc, nw_proto);
1316 WC_MASK_FIELD(wc, arp_sha);
1317 WC_MASK_FIELD(wc, arp_tha);
1319 } else if (eth_type_mpls(flow->dl_type)) {
1320 for (int i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1321 WC_MASK_FIELD(wc, mpls_lse[i]);
1322 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1328 return; /* Unknown ethertype. */
1332 WC_MASK_FIELD(wc, nw_frag);
1333 WC_MASK_FIELD(wc, nw_tos);
1334 WC_MASK_FIELD(wc, nw_ttl);
1335 WC_MASK_FIELD(wc, nw_proto);
1337 /* No transport layer header in later fragments. */
1338 if (!(flow->nw_frag & FLOW_NW_FRAG_LATER) &&
1339 (flow->nw_proto == IPPROTO_ICMP ||
1340 flow->nw_proto == IPPROTO_ICMPV6 ||
1341 flow->nw_proto == IPPROTO_TCP ||
1342 flow->nw_proto == IPPROTO_UDP ||
1343 flow->nw_proto == IPPROTO_SCTP ||
1344 flow->nw_proto == IPPROTO_IGMP)) {
1345 WC_MASK_FIELD(wc, tp_src);
1346 WC_MASK_FIELD(wc, tp_dst);
1348 if (flow->nw_proto == IPPROTO_TCP) {
1349 WC_MASK_FIELD(wc, tcp_flags);
1350 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
1351 WC_MASK_FIELD(wc, arp_sha);
1352 WC_MASK_FIELD(wc, arp_tha);
1353 WC_MASK_FIELD(wc, nd_target);
1354 } else if (flow->nw_proto == IPPROTO_IGMP) {
1355 WC_MASK_FIELD(wc, igmp_group_ip4);
1360 /* Return a map of possible fields for a packet of the same type as 'flow'.
1361 * Including extra bits in the returned mask is not wrong, it is just less
1364 * This is a less precise version of flow_wildcards_init_for_packet() above. */
1366 flow_wc_map(const struct flow *flow, struct flowmap *map)
1368 /* Update this function whenever struct flow changes. */
1369 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 36);
1373 if (flow_tnl_dst_is_set(&flow->tunnel)) {
1374 FLOWMAP_SET__(map, tunnel, offsetof(struct flow_tnl, metadata));
1375 if (!(flow->tunnel.flags & FLOW_TNL_F_UDPIF)) {
1376 if (flow->tunnel.metadata.present.map) {
1377 FLOWMAP_SET(map, tunnel.metadata);
1380 FLOWMAP_SET(map, tunnel.metadata.present.len);
1381 FLOWMAP_SET__(map, tunnel.metadata.opts.gnv,
1382 flow->tunnel.metadata.present.len);
1386 /* Metadata fields that can appear on packet input. */
1387 FLOWMAP_SET(map, skb_priority);
1388 FLOWMAP_SET(map, pkt_mark);
1389 FLOWMAP_SET(map, recirc_id);
1390 FLOWMAP_SET(map, dp_hash);
1391 FLOWMAP_SET(map, in_port);
1392 FLOWMAP_SET(map, dl_dst);
1393 FLOWMAP_SET(map, dl_src);
1394 FLOWMAP_SET(map, dl_type);
1395 FLOWMAP_SET(map, vlan_tci);
1396 FLOWMAP_SET(map, ct_state);
1397 FLOWMAP_SET(map, ct_zone);
1398 FLOWMAP_SET(map, ct_mark);
1399 FLOWMAP_SET(map, ct_label);
1401 /* Ethertype-dependent fields. */
1402 if (OVS_LIKELY(flow->dl_type == htons(ETH_TYPE_IP))) {
1403 FLOWMAP_SET(map, nw_src);
1404 FLOWMAP_SET(map, nw_dst);
1405 FLOWMAP_SET(map, nw_proto);
1406 FLOWMAP_SET(map, nw_frag);
1407 FLOWMAP_SET(map, nw_tos);
1408 FLOWMAP_SET(map, nw_ttl);
1409 FLOWMAP_SET(map, tp_src);
1410 FLOWMAP_SET(map, tp_dst);
1412 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_IGMP)) {
1413 FLOWMAP_SET(map, igmp_group_ip4);
1415 FLOWMAP_SET(map, tcp_flags);
1417 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1418 FLOWMAP_SET(map, ipv6_src);
1419 FLOWMAP_SET(map, ipv6_dst);
1420 FLOWMAP_SET(map, ipv6_label);
1421 FLOWMAP_SET(map, nw_proto);
1422 FLOWMAP_SET(map, nw_frag);
1423 FLOWMAP_SET(map, nw_tos);
1424 FLOWMAP_SET(map, nw_ttl);
1425 FLOWMAP_SET(map, tp_src);
1426 FLOWMAP_SET(map, tp_dst);
1428 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_ICMPV6)) {
1429 FLOWMAP_SET(map, nd_target);
1430 FLOWMAP_SET(map, arp_sha);
1431 FLOWMAP_SET(map, arp_tha);
1433 FLOWMAP_SET(map, tcp_flags);
1435 } else if (eth_type_mpls(flow->dl_type)) {
1436 FLOWMAP_SET(map, mpls_lse);
1437 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1438 flow->dl_type == htons(ETH_TYPE_RARP)) {
1439 FLOWMAP_SET(map, nw_src);
1440 FLOWMAP_SET(map, nw_dst);
1441 FLOWMAP_SET(map, nw_proto);
1442 FLOWMAP_SET(map, arp_sha);
1443 FLOWMAP_SET(map, arp_tha);
1447 /* Clear the metadata and register wildcard masks. They are not packet
1450 flow_wildcards_clear_non_packet_fields(struct flow_wildcards *wc)
1452 /* Update this function whenever struct flow changes. */
1453 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 36);
1455 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
1456 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
1457 wc->masks.actset_output = 0;
1458 wc->masks.conj_id = 0;
1461 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
1464 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
1466 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1469 for (i = 0; i < FLOW_U64S; i++) {
1477 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
1478 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
1479 * in 'src1' or 'src2' or both. */
1481 flow_wildcards_and(struct flow_wildcards *dst,
1482 const struct flow_wildcards *src1,
1483 const struct flow_wildcards *src2)
1485 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1486 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1487 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1490 for (i = 0; i < FLOW_U64S; i++) {
1491 dst_u64[i] = src1_u64[i] & src2_u64[i];
1495 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
1496 * is, a bit or a field is wildcarded in 'dst' if it is neither
1497 * wildcarded in 'src1' nor 'src2'. */
1499 flow_wildcards_or(struct flow_wildcards *dst,
1500 const struct flow_wildcards *src1,
1501 const struct flow_wildcards *src2)
1503 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1504 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1505 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1508 for (i = 0; i < FLOW_U64S; i++) {
1509 dst_u64[i] = src1_u64[i] | src2_u64[i];
1513 /* Returns a hash of the wildcards in 'wc'. */
1515 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
1517 return flow_hash(&wc->masks, basis);
1520 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
1523 flow_wildcards_equal(const struct flow_wildcards *a,
1524 const struct flow_wildcards *b)
1526 return flow_equal(&a->masks, &b->masks);
1529 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
1530 * 'b', false otherwise. */
1532 flow_wildcards_has_extra(const struct flow_wildcards *a,
1533 const struct flow_wildcards *b)
1535 const uint64_t *a_u64 = (const uint64_t *) &a->masks;
1536 const uint64_t *b_u64 = (const uint64_t *) &b->masks;
1539 for (i = 0; i < FLOW_U64S; i++) {
1540 if ((a_u64[i] & b_u64[i]) != b_u64[i]) {
1547 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
1548 * in 'wc' do not need to be equal in 'a' and 'b'. */
1550 flow_equal_except(const struct flow *a, const struct flow *b,
1551 const struct flow_wildcards *wc)
1553 const uint64_t *a_u64 = (const uint64_t *) a;
1554 const uint64_t *b_u64 = (const uint64_t *) b;
1555 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1558 for (i = 0; i < FLOW_U64S; i++) {
1559 if ((a_u64[i] ^ b_u64[i]) & wc_u64[i]) {
1566 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1567 * (A 0-bit indicates a wildcard bit.) */
1569 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
1571 wc->masks.regs[idx] = mask;
1574 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1575 * (A 0-bit indicates a wildcard bit.) */
1577 flow_wildcards_set_xreg_mask(struct flow_wildcards *wc, int idx, uint64_t mask)
1579 flow_set_xreg(&wc->masks, idx, mask);
1582 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1583 * (A 0-bit indicates a wildcard bit.) */
1585 flow_wildcards_set_xxreg_mask(struct flow_wildcards *wc, int idx,
1588 flow_set_xxreg(&wc->masks, idx, mask);
1591 /* Calculates the 5-tuple hash from the given miniflow.
1592 * This returns the same value as flow_hash_5tuple for the corresponding
1595 miniflow_hash_5tuple(const struct miniflow *flow, uint32_t basis)
1597 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 36);
1598 uint32_t hash = basis;
1601 ovs_be16 dl_type = MINIFLOW_GET_BE16(flow, dl_type);
1604 if (dl_type == htons(ETH_TYPE_IPV6)) {
1605 struct flowmap map = FLOWMAP_EMPTY_INITIALIZER;
1608 FLOWMAP_SET(&map, ipv6_src);
1609 FLOWMAP_SET(&map, ipv6_dst);
1611 MINIFLOW_FOR_EACH_IN_FLOWMAP(value, flow, map) {
1612 hash = hash_add64(hash, value);
1614 } else if (dl_type == htons(ETH_TYPE_IP)
1615 || dl_type == htons(ETH_TYPE_ARP)) {
1616 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_src));
1617 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_dst));
1622 nw_proto = MINIFLOW_GET_U8(flow, nw_proto);
1623 hash = hash_add(hash, nw_proto);
1624 if (nw_proto != IPPROTO_TCP && nw_proto != IPPROTO_UDP
1625 && nw_proto != IPPROTO_SCTP && nw_proto != IPPROTO_ICMP
1626 && nw_proto != IPPROTO_ICMPV6) {
1630 /* Add both ports at once. */
1631 hash = hash_add(hash, MINIFLOW_GET_U32(flow, tp_src));
1634 return hash_finish(hash, 42);
1637 ASSERT_SEQUENTIAL_SAME_WORD(tp_src, tp_dst);
1638 ASSERT_SEQUENTIAL(ipv6_src, ipv6_dst);
1640 /* Calculates the 5-tuple hash from the given flow. */
1642 flow_hash_5tuple(const struct flow *flow, uint32_t basis)
1644 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 36);
1645 uint32_t hash = basis;
1649 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1650 const uint64_t *flow_u64 = (const uint64_t *)flow;
1651 int ofs = offsetof(struct flow, ipv6_src) / 8;
1652 int end = ofs + 2 * sizeof flow->ipv6_src / 8;
1654 for (;ofs < end; ofs++) {
1655 hash = hash_add64(hash, flow_u64[ofs]);
1657 } else if (flow->dl_type == htons(ETH_TYPE_IP)
1658 || flow->dl_type == htons(ETH_TYPE_ARP)) {
1659 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_src);
1660 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_dst);
1665 hash = hash_add(hash, flow->nw_proto);
1666 if (flow->nw_proto != IPPROTO_TCP && flow->nw_proto != IPPROTO_UDP
1667 && flow->nw_proto != IPPROTO_SCTP && flow->nw_proto != IPPROTO_ICMP
1668 && flow->nw_proto != IPPROTO_ICMPV6) {
1672 /* Add both ports at once. */
1673 hash = hash_add(hash,
1674 ((const uint32_t *)flow)[offsetof(struct flow, tp_src)
1675 / sizeof(uint32_t)]);
1678 return hash_finish(hash, 42); /* Arbitrary number. */
1681 /* Hashes 'flow' based on its L2 through L4 protocol information. */
1683 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
1688 struct in6_addr ipv6_addr;
1693 struct eth_addr eth_addr;
1699 memset(&fields, 0, sizeof fields);
1700 for (i = 0; i < ARRAY_SIZE(fields.eth_addr.be16); i++) {
1701 fields.eth_addr.be16[i] = flow->dl_src.be16[i] ^ flow->dl_dst.be16[i];
1703 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
1704 fields.eth_type = flow->dl_type;
1706 /* UDP source and destination port are not taken into account because they
1707 * will not necessarily be symmetric in a bidirectional flow. */
1708 if (fields.eth_type == htons(ETH_TYPE_IP)) {
1709 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
1710 fields.ip_proto = flow->nw_proto;
1711 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1712 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1714 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
1715 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
1716 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
1717 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
1719 for (i=0; i<16; i++) {
1720 ipv6_addr[i] = a[i] ^ b[i];
1722 fields.ip_proto = flow->nw_proto;
1723 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1724 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1727 return jhash_bytes(&fields, sizeof fields, basis);
1730 /* Hashes 'flow' based on its L3 through L4 protocol information */
1732 flow_hash_symmetric_l3l4(const struct flow *flow, uint32_t basis,
1735 uint32_t hash = basis;
1737 /* UDP source and destination port are also taken into account. */
1738 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1739 hash = hash_add(hash,
1740 (OVS_FORCE uint32_t) (flow->nw_src ^ flow->nw_dst));
1741 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1742 /* IPv6 addresses are 64-bit aligned inside struct flow. */
1743 const uint64_t *a = ALIGNED_CAST(uint64_t *, flow->ipv6_src.s6_addr);
1744 const uint64_t *b = ALIGNED_CAST(uint64_t *, flow->ipv6_dst.s6_addr);
1746 for (int i = 0; i < 4; i++) {
1747 hash = hash_add64(hash, a[i] ^ b[i]);
1750 /* Cannot hash non-IP flows */
1754 hash = hash_add(hash, flow->nw_proto);
1755 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP ||
1756 (inc_udp_ports && flow->nw_proto == IPPROTO_UDP)) {
1757 hash = hash_add(hash,
1758 (OVS_FORCE uint16_t) (flow->tp_src ^ flow->tp_dst));
1761 return hash_finish(hash, basis);
1764 /* Initialize a flow with random fields that matter for nx_hash_fields. */
1766 flow_random_hash_fields(struct flow *flow)
1768 uint16_t rnd = random_uint16();
1770 /* Initialize to all zeros. */
1771 memset(flow, 0, sizeof *flow);
1773 eth_addr_random(&flow->dl_src);
1774 eth_addr_random(&flow->dl_dst);
1776 flow->vlan_tci = (OVS_FORCE ovs_be16) (random_uint16() & VLAN_VID_MASK);
1778 /* Make most of the random flows IPv4, some IPv6, and rest random. */
1779 flow->dl_type = rnd < 0x8000 ? htons(ETH_TYPE_IP) :
1780 rnd < 0xc000 ? htons(ETH_TYPE_IPV6) : (OVS_FORCE ovs_be16)rnd;
1782 if (dl_type_is_ip_any(flow->dl_type)) {
1783 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1784 flow->nw_src = (OVS_FORCE ovs_be32)random_uint32();
1785 flow->nw_dst = (OVS_FORCE ovs_be32)random_uint32();
1787 random_bytes(&flow->ipv6_src, sizeof flow->ipv6_src);
1788 random_bytes(&flow->ipv6_dst, sizeof flow->ipv6_dst);
1790 /* Make most of IP flows TCP, some UDP or SCTP, and rest random. */
1791 rnd = random_uint16();
1792 flow->nw_proto = rnd < 0x8000 ? IPPROTO_TCP :
1793 rnd < 0xc000 ? IPPROTO_UDP :
1794 rnd < 0xd000 ? IPPROTO_SCTP : (uint8_t)rnd;
1795 if (flow->nw_proto == IPPROTO_TCP ||
1796 flow->nw_proto == IPPROTO_UDP ||
1797 flow->nw_proto == IPPROTO_SCTP) {
1798 flow->tp_src = (OVS_FORCE ovs_be16)random_uint16();
1799 flow->tp_dst = (OVS_FORCE ovs_be16)random_uint16();
1804 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
1806 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
1807 enum nx_hash_fields fields)
1810 case NX_HASH_FIELDS_ETH_SRC:
1811 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1814 case NX_HASH_FIELDS_SYMMETRIC_L4:
1815 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1816 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1817 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1818 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1819 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1820 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1821 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1822 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1824 if (is_ip_any(flow)) {
1825 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1826 flow_unwildcard_tp_ports(flow, wc);
1828 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1831 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1832 if (is_ip_any(flow) && flow->nw_proto == IPPROTO_UDP) {
1833 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1834 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1837 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1838 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1839 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1840 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1841 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1842 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1843 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1845 break; /* non-IP flow */
1848 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1849 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP) {
1850 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1851 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1860 /* Hashes the portions of 'flow' designated by 'fields'. */
1862 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
1867 case NX_HASH_FIELDS_ETH_SRC:
1868 return jhash_bytes(&flow->dl_src, sizeof flow->dl_src, basis);
1870 case NX_HASH_FIELDS_SYMMETRIC_L4:
1871 return flow_hash_symmetric_l4(flow, basis);
1873 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1874 return flow_hash_symmetric_l3l4(flow, basis, false);
1876 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1877 return flow_hash_symmetric_l3l4(flow, basis, true);
1884 /* Returns a string representation of 'fields'. */
1886 flow_hash_fields_to_str(enum nx_hash_fields fields)
1889 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
1890 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
1891 case NX_HASH_FIELDS_SYMMETRIC_L3L4: return "symmetric_l3l4";
1892 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP: return "symmetric_l3l4+udp";
1893 default: return "<unknown>";
1897 /* Returns true if the value of 'fields' is supported. Otherwise false. */
1899 flow_hash_fields_valid(enum nx_hash_fields fields)
1901 return fields == NX_HASH_FIELDS_ETH_SRC
1902 || fields == NX_HASH_FIELDS_SYMMETRIC_L4
1903 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4
1904 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP;
1907 /* Returns a hash value for the bits of 'flow' that are active based on
1908 * 'wc', given 'basis'. */
1910 flow_hash_in_wildcards(const struct flow *flow,
1911 const struct flow_wildcards *wc, uint32_t basis)
1913 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1914 const uint64_t *flow_u64 = (const uint64_t *) flow;
1919 for (i = 0; i < FLOW_U64S; i++) {
1920 hash = hash_add64(hash, flow_u64[i] & wc_u64[i]);
1922 return hash_finish(hash, 8 * FLOW_U64S);
1925 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1926 * OpenFlow 1.0 "dl_vlan" value:
1928 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
1929 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
1930 * 'flow' previously matched packets without a VLAN header).
1932 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
1933 * without a VLAN tag.
1935 * - Other values of 'vid' should not be used. */
1937 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
1939 if (vid == htons(OFP10_VLAN_NONE)) {
1940 flow->vlan_tci = htons(0);
1942 vid &= htons(VLAN_VID_MASK);
1943 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1944 flow->vlan_tci |= htons(VLAN_CFI) | vid;
1948 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1949 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
1952 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
1954 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
1955 flow->vlan_tci &= ~mask;
1956 flow->vlan_tci |= vid & mask;
1959 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
1962 * This function has no effect on the VLAN ID that 'flow' matches.
1964 * After calling this function, 'flow' will not match packets without a VLAN
1967 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
1970 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1971 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
1974 /* Returns the number of MPLS LSEs present in 'flow'
1976 * Returns 0 if the 'dl_type' of 'flow' is not an MPLS ethernet type.
1977 * Otherwise traverses 'flow''s MPLS label stack stopping at the
1978 * first entry that has the BoS bit set. If no such entry exists then
1979 * the maximum number of LSEs that can be stored in 'flow' is returned.
1982 flow_count_mpls_labels(const struct flow *flow, struct flow_wildcards *wc)
1984 /* dl_type is always masked. */
1985 if (eth_type_mpls(flow->dl_type)) {
1990 for (i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1992 wc->masks.mpls_lse[i] |= htonl(MPLS_BOS_MASK);
1994 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1997 if (flow->mpls_lse[i]) {
2007 /* Returns the number consecutive of MPLS LSEs, starting at the
2008 * innermost LSE, that are common in 'a' and 'b'.
2010 * 'an' must be flow_count_mpls_labels(a).
2011 * 'bn' must be flow_count_mpls_labels(b).
2014 flow_count_common_mpls_labels(const struct flow *a, int an,
2015 const struct flow *b, int bn,
2016 struct flow_wildcards *wc)
2018 int min_n = MIN(an, bn);
2023 int a_last = an - 1;
2024 int b_last = bn - 1;
2027 for (i = 0; i < min_n; i++) {
2029 wc->masks.mpls_lse[a_last - i] = OVS_BE32_MAX;
2030 wc->masks.mpls_lse[b_last - i] = OVS_BE32_MAX;
2032 if (a->mpls_lse[a_last - i] != b->mpls_lse[b_last - i]) {
2043 /* Adds a new outermost MPLS label to 'flow' and changes 'flow''s Ethernet type
2044 * to 'mpls_eth_type', which must be an MPLS Ethertype.
2046 * If the new label is the first MPLS label in 'flow', it is generated as;
2048 * - label: 2, if 'flow' is IPv6, otherwise 0.
2050 * - TTL: IPv4 or IPv6 TTL, if present and nonzero, otherwise 64.
2052 * - TC: IPv4 or IPv6 TOS, if present, otherwise 0.
2056 * If the new label is the second or later label MPLS label in 'flow', it is
2059 * - label: Copied from outer label.
2061 * - TTL: Copied from outer label.
2063 * - TC: Copied from outer label.
2067 * 'n' must be flow_count_mpls_labels(flow). 'n' must be less than
2068 * FLOW_MAX_MPLS_LABELS (because otherwise flow->mpls_lse[] would overflow).
2071 flow_push_mpls(struct flow *flow, int n, ovs_be16 mpls_eth_type,
2072 struct flow_wildcards *wc)
2074 ovs_assert(eth_type_mpls(mpls_eth_type));
2075 ovs_assert(n < FLOW_MAX_MPLS_LABELS);
2081 memset(&wc->masks.mpls_lse, 0xff, sizeof *wc->masks.mpls_lse * n);
2083 for (i = n; i >= 1; i--) {
2084 flow->mpls_lse[i] = flow->mpls_lse[i - 1];
2086 flow->mpls_lse[0] = (flow->mpls_lse[1] & htonl(~MPLS_BOS_MASK));
2088 int label = 0; /* IPv4 Explicit Null. */
2092 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2096 if (is_ip_any(flow)) {
2097 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
2099 wc->masks.nw_tos |= IP_DSCP_MASK;
2100 wc->masks.nw_ttl = 0xff;
2108 flow->mpls_lse[0] = set_mpls_lse_values(ttl, tc, 1, htonl(label));
2110 /* Clear all L3 and L4 fields and dp_hash. */
2111 BUILD_ASSERT(FLOW_WC_SEQ == 36);
2112 memset((char *) flow + FLOW_SEGMENT_2_ENDS_AT, 0,
2113 sizeof(struct flow) - FLOW_SEGMENT_2_ENDS_AT);
2116 flow->dl_type = mpls_eth_type;
2119 /* Tries to remove the outermost MPLS label from 'flow'. Returns true if
2120 * successful, false otherwise. On success, sets 'flow''s Ethernet type to
2123 * 'n' must be flow_count_mpls_labels(flow). */
2125 flow_pop_mpls(struct flow *flow, int n, ovs_be16 eth_type,
2126 struct flow_wildcards *wc)
2131 /* Nothing to pop. */
2133 } else if (n == FLOW_MAX_MPLS_LABELS) {
2135 wc->masks.mpls_lse[n - 1] |= htonl(MPLS_BOS_MASK);
2137 if (!(flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK))) {
2138 /* Can't pop because don't know what to fill in mpls_lse[n - 1]. */
2144 memset(&wc->masks.mpls_lse[1], 0xff,
2145 sizeof *wc->masks.mpls_lse * (n - 1));
2147 for (i = 1; i < n; i++) {
2148 flow->mpls_lse[i - 1] = flow->mpls_lse[i];
2150 flow->mpls_lse[n - 1] = 0;
2151 flow->dl_type = eth_type;
2155 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
2156 * as an OpenFlow 1.1 "mpls_label" value. */
2158 flow_set_mpls_label(struct flow *flow, int idx, ovs_be32 label)
2160 set_mpls_lse_label(&flow->mpls_lse[idx], label);
2163 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
2166 flow_set_mpls_ttl(struct flow *flow, int idx, uint8_t ttl)
2168 set_mpls_lse_ttl(&flow->mpls_lse[idx], ttl);
2171 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
2174 flow_set_mpls_tc(struct flow *flow, int idx, uint8_t tc)
2176 set_mpls_lse_tc(&flow->mpls_lse[idx], tc);
2179 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
2181 flow_set_mpls_bos(struct flow *flow, int idx, uint8_t bos)
2183 set_mpls_lse_bos(&flow->mpls_lse[idx], bos);
2186 /* Sets the entire MPLS LSE. */
2188 flow_set_mpls_lse(struct flow *flow, int idx, ovs_be32 lse)
2190 flow->mpls_lse[idx] = lse;
2194 flow_compose_l4(struct dp_packet *p, const struct flow *flow)
2198 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
2199 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
2200 if (flow->nw_proto == IPPROTO_TCP) {
2201 struct tcp_header *tcp;
2203 l4_len = sizeof *tcp;
2204 tcp = dp_packet_put_zeros(p, l4_len);
2205 tcp->tcp_src = flow->tp_src;
2206 tcp->tcp_dst = flow->tp_dst;
2207 tcp->tcp_ctl = TCP_CTL(ntohs(flow->tcp_flags), 5);
2208 } else if (flow->nw_proto == IPPROTO_UDP) {
2209 struct udp_header *udp;
2211 l4_len = sizeof *udp;
2212 udp = dp_packet_put_zeros(p, l4_len);
2213 udp->udp_src = flow->tp_src;
2214 udp->udp_dst = flow->tp_dst;
2215 } else if (flow->nw_proto == IPPROTO_SCTP) {
2216 struct sctp_header *sctp;
2218 l4_len = sizeof *sctp;
2219 sctp = dp_packet_put_zeros(p, l4_len);
2220 sctp->sctp_src = flow->tp_src;
2221 sctp->sctp_dst = flow->tp_dst;
2222 } else if (flow->nw_proto == IPPROTO_ICMP) {
2223 struct icmp_header *icmp;
2225 l4_len = sizeof *icmp;
2226 icmp = dp_packet_put_zeros(p, l4_len);
2227 icmp->icmp_type = ntohs(flow->tp_src);
2228 icmp->icmp_code = ntohs(flow->tp_dst);
2229 /* Checksum has already been zeroed by put_zeros call. */
2230 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
2231 } else if (flow->nw_proto == IPPROTO_IGMP) {
2232 struct igmp_header *igmp;
2234 l4_len = sizeof *igmp;
2235 igmp = dp_packet_put_zeros(p, l4_len);
2236 igmp->igmp_type = ntohs(flow->tp_src);
2237 igmp->igmp_code = ntohs(flow->tp_dst);
2238 put_16aligned_be32(&igmp->group, flow->igmp_group_ip4);
2239 /* Checksum has already been zeroed by put_zeros call. */
2240 igmp->igmp_csum = csum(igmp, IGMP_HEADER_LEN);
2241 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
2242 struct icmp6_hdr *icmp;
2244 l4_len = sizeof *icmp;
2245 icmp = dp_packet_put_zeros(p, l4_len);
2246 icmp->icmp6_type = ntohs(flow->tp_src);
2247 icmp->icmp6_code = ntohs(flow->tp_dst);
2249 if (icmp->icmp6_code == 0 &&
2250 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
2251 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
2252 struct in6_addr *nd_target;
2253 struct ovs_nd_opt *nd_opt;
2255 l4_len += sizeof *nd_target;
2256 nd_target = dp_packet_put_zeros(p, sizeof *nd_target);
2257 *nd_target = flow->nd_target;
2259 if (!eth_addr_is_zero(flow->arp_sha)) {
2261 nd_opt = dp_packet_put_zeros(p, 8);
2262 nd_opt->nd_opt_len = 1;
2263 nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
2264 nd_opt->nd_opt_mac = flow->arp_sha;
2266 if (!eth_addr_is_zero(flow->arp_tha)) {
2268 nd_opt = dp_packet_put_zeros(p, 8);
2269 nd_opt->nd_opt_len = 1;
2270 nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR;
2271 nd_opt->nd_opt_mac = flow->arp_tha;
2274 icmp->icmp6_cksum = (OVS_FORCE uint16_t)
2275 csum(icmp, (char *)dp_packet_tail(p) - (char *)icmp);
2281 /* Puts into 'b' a packet that flow_extract() would parse as having the given
2284 * (This is useful only for testing, obviously, and the packet isn't really
2285 * valid. It hasn't got some checksums filled in, for one, and lots of fields
2286 * are just zeroed.) */
2288 flow_compose(struct dp_packet *p, const struct flow *flow)
2292 /* eth_compose() sets l3 pointer and makes sure it is 32-bit aligned. */
2293 eth_compose(p, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
2294 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
2295 struct eth_header *eth = dp_packet_l2(p);
2296 eth->eth_type = htons(dp_packet_size(p));
2300 if (flow->vlan_tci & htons(VLAN_CFI)) {
2301 eth_push_vlan(p, htons(ETH_TYPE_VLAN), flow->vlan_tci);
2304 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2305 struct ip_header *ip;
2307 ip = dp_packet_put_zeros(p, sizeof *ip);
2308 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
2309 ip->ip_tos = flow->nw_tos;
2310 ip->ip_ttl = flow->nw_ttl;
2311 ip->ip_proto = flow->nw_proto;
2312 put_16aligned_be32(&ip->ip_src, flow->nw_src);
2313 put_16aligned_be32(&ip->ip_dst, flow->nw_dst);
2315 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2316 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
2317 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
2318 ip->ip_frag_off |= htons(100);
2322 dp_packet_set_l4(p, dp_packet_tail(p));
2324 l4_len = flow_compose_l4(p, flow);
2326 ip = dp_packet_l3(p);
2327 ip->ip_tot_len = htons(p->l4_ofs - p->l3_ofs + l4_len);
2328 /* Checksum has already been zeroed by put_zeros call. */
2329 ip->ip_csum = csum(ip, sizeof *ip);
2330 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2331 struct ovs_16aligned_ip6_hdr *nh;
2333 nh = dp_packet_put_zeros(p, sizeof *nh);
2334 put_16aligned_be32(&nh->ip6_flow, htonl(6 << 28) |
2335 htonl(flow->nw_tos << 20) | flow->ipv6_label);
2336 nh->ip6_hlim = flow->nw_ttl;
2337 nh->ip6_nxt = flow->nw_proto;
2339 memcpy(&nh->ip6_src, &flow->ipv6_src, sizeof(nh->ip6_src));
2340 memcpy(&nh->ip6_dst, &flow->ipv6_dst, sizeof(nh->ip6_dst));
2342 dp_packet_set_l4(p, dp_packet_tail(p));
2344 l4_len = flow_compose_l4(p, flow);
2346 nh = dp_packet_l3(p);
2347 nh->ip6_plen = htons(l4_len);
2348 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
2349 flow->dl_type == htons(ETH_TYPE_RARP)) {
2350 struct arp_eth_header *arp;
2352 arp = dp_packet_put_zeros(p, sizeof *arp);
2353 dp_packet_set_l3(p, arp);
2354 arp->ar_hrd = htons(1);
2355 arp->ar_pro = htons(ETH_TYPE_IP);
2356 arp->ar_hln = ETH_ADDR_LEN;
2358 arp->ar_op = htons(flow->nw_proto);
2360 if (flow->nw_proto == ARP_OP_REQUEST ||
2361 flow->nw_proto == ARP_OP_REPLY) {
2362 put_16aligned_be32(&arp->ar_spa, flow->nw_src);
2363 put_16aligned_be32(&arp->ar_tpa, flow->nw_dst);
2364 arp->ar_sha = flow->arp_sha;
2365 arp->ar_tha = flow->arp_tha;
2369 if (eth_type_mpls(flow->dl_type)) {
2372 p->l2_5_ofs = p->l3_ofs;
2373 for (n = 1; n < FLOW_MAX_MPLS_LABELS; n++) {
2374 if (flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK)) {
2379 push_mpls(p, flow->dl_type, flow->mpls_lse[--n]);
2384 /* Compressed flow. */
2386 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
2387 * the caller. The caller must have already computed 'dst->map' properly to
2388 * indicate the significant uint64_t elements of 'src'.
2390 * Normally the significant elements are the ones that are non-zero. However,
2391 * when a miniflow is initialized from a (mini)mask, the values can be zeroes,
2392 * so that the flow and mask always have the same maps. */
2394 miniflow_init(struct miniflow *dst, const struct flow *src)
2396 uint64_t *dst_u64 = miniflow_values(dst);
2399 FLOWMAP_FOR_EACH_INDEX(idx, dst->map) {
2400 *dst_u64++ = flow_u64_value(src, idx);
2404 /* Initialize the maps of 'flow' from 'src'. */
2406 miniflow_map_init(struct miniflow *flow, const struct flow *src)
2408 /* Initialize map, counting the number of nonzero elements. */
2409 flowmap_init(&flow->map);
2410 for (size_t i = 0; i < FLOW_U64S; i++) {
2411 if (flow_u64_value(src, i)) {
2412 flowmap_set(&flow->map, i, 1);
2417 /* Allocates 'n' count of miniflows, consecutive in memory, initializing the
2418 * map of each from 'src'.
2419 * Returns the size of the miniflow data. */
2421 miniflow_alloc(struct miniflow *dsts[], size_t n, const struct miniflow *src)
2423 size_t n_values = miniflow_n_values(src);
2424 size_t data_size = MINIFLOW_VALUES_SIZE(n_values);
2425 struct miniflow *dst = xmalloc(n * (sizeof *src + data_size));
2428 COVERAGE_INC(miniflow_malloc);
2430 for (i = 0; i < n; i++) {
2431 *dst = *src; /* Copy maps. */
2433 dst += 1; /* Just past the maps. */
2434 dst = (struct miniflow *)((uint64_t *)dst + n_values); /* Skip data. */
2439 /* Returns a miniflow copy of 'src'. The caller must eventually free() the
2440 * returned miniflow. */
2442 miniflow_create(const struct flow *src)
2444 struct miniflow tmp;
2445 struct miniflow *dst;
2447 miniflow_map_init(&tmp, src);
2449 miniflow_alloc(&dst, 1, &tmp);
2450 miniflow_init(dst, src);
2454 /* Initializes 'dst' as a copy of 'src'. The caller must have allocated
2455 * 'dst' to have inline space for 'n_values' data in 'src'. */
2457 miniflow_clone(struct miniflow *dst, const struct miniflow *src,
2460 *dst = *src; /* Copy maps. */
2461 memcpy(miniflow_values(dst), miniflow_get_values(src),
2462 MINIFLOW_VALUES_SIZE(n_values));
2465 /* Initializes 'dst' as a copy of 'src'. */
2467 miniflow_expand(const struct miniflow *src, struct flow *dst)
2469 memset(dst, 0, sizeof *dst);
2470 flow_union_with_miniflow(dst, src);
2473 /* Returns true if 'a' and 'b' are equal miniflows, false otherwise. */
2475 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
2477 const uint64_t *ap = miniflow_get_values(a);
2478 const uint64_t *bp = miniflow_get_values(b);
2480 /* This is mostly called after a matching hash, so it is highly likely that
2481 * the maps are equal as well. */
2482 if (OVS_LIKELY(flowmap_equal(a->map, b->map))) {
2483 return !memcmp(ap, bp, miniflow_n_values(a) * sizeof *ap);
2487 FLOWMAP_FOR_EACH_INDEX (idx, flowmap_or(a->map, b->map)) {
2488 if ((flowmap_is_set(&a->map, idx) ? *ap++ : 0)
2489 != (flowmap_is_set(&b->map, idx) ? *bp++ : 0)) {
2498 /* Returns false if 'a' and 'b' differ at the places where there are 1-bits
2499 * in 'mask', true otherwise. */
2501 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
2502 const struct minimask *mask)
2504 const uint64_t *p = miniflow_get_values(&mask->masks);
2507 FLOWMAP_FOR_EACH_INDEX(idx, mask->masks.map) {
2508 if ((miniflow_get(a, idx) ^ miniflow_get(b, idx)) & *p++) {
2516 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
2517 * in 'mask', false if they differ. */
2519 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
2520 const struct minimask *mask)
2522 const uint64_t *p = miniflow_get_values(&mask->masks);
2525 FLOWMAP_FOR_EACH_INDEX(idx, mask->masks.map) {
2526 if ((miniflow_get(a, idx) ^ flow_u64_value(b, idx)) & *p++) {
2536 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
2538 miniflow_init(&mask->masks, &wc->masks);
2541 /* Returns a minimask copy of 'wc'. The caller must eventually free the
2542 * returned minimask with free(). */
2544 minimask_create(const struct flow_wildcards *wc)
2546 return (struct minimask *)miniflow_create(&wc->masks);
2549 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
2551 * The caller must provide room for FLOW_U64S "uint64_t"s in 'storage', which
2552 * must follow '*dst_' in memory, for use by 'dst_'. The caller must *not*
2553 * free 'dst_' free(). */
2555 minimask_combine(struct minimask *dst_,
2556 const struct minimask *a_, const struct minimask *b_,
2557 uint64_t storage[FLOW_U64S])
2559 struct miniflow *dst = &dst_->masks;
2560 uint64_t *dst_values = storage;
2561 const struct miniflow *a = &a_->masks;
2562 const struct miniflow *b = &b_->masks;
2565 flowmap_init(&dst->map);
2567 FLOWMAP_FOR_EACH_INDEX(idx, flowmap_and(a->map, b->map)) {
2568 /* Both 'a' and 'b' have non-zero data at 'idx'. */
2569 uint64_t mask = *miniflow_get__(a, idx) & *miniflow_get__(b, idx);
2572 flowmap_set(&dst->map, idx, 1);
2573 *dst_values++ = mask;
2578 /* Initializes 'wc' as a copy of 'mask'. */
2580 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
2582 miniflow_expand(&mask->masks, &wc->masks);
2585 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise.
2586 * Minimasks may not have zero data values, so for the minimasks to be the
2587 * same, they need to have the same map and the same data values. */
2589 minimask_equal(const struct minimask *a, const struct minimask *b)
2591 return !memcmp(a, b, sizeof *a
2592 + MINIFLOW_VALUES_SIZE(miniflow_n_values(&a->masks)));
2595 /* Returns true if at least one bit matched by 'b' is wildcarded by 'a',
2596 * false otherwise. */
2598 minimask_has_extra(const struct minimask *a, const struct minimask *b)
2600 const uint64_t *bp = miniflow_get_values(&b->masks);
2603 FLOWMAP_FOR_EACH_INDEX(idx, b->masks.map) {
2604 uint64_t b_u64 = *bp++;
2606 /* 'b_u64' is non-zero, check if the data in 'a' is either zero
2607 * or misses some of the bits in 'b_u64'. */
2608 if (!MINIFLOW_IN_MAP(&a->masks, idx)
2609 || ((*miniflow_get__(&a->masks, idx) & b_u64) != b_u64)) {
2610 return true; /* 'a' wildcards some bits 'b' doesn't. */