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"
43 COVERAGE_DEFINE(flow_extract);
44 COVERAGE_DEFINE(miniflow_malloc);
46 /* U64 indices for segmented flow classification. */
47 const uint8_t flow_segment_u64s[4] = {
48 FLOW_SEGMENT_1_ENDS_AT / sizeof(uint64_t),
49 FLOW_SEGMENT_2_ENDS_AT / sizeof(uint64_t),
50 FLOW_SEGMENT_3_ENDS_AT / sizeof(uint64_t),
54 /* Asserts that field 'f1' follows immediately after 'f0' in struct flow,
55 * without any intervening padding. */
56 #define ASSERT_SEQUENTIAL(f0, f1) \
57 BUILD_ASSERT_DECL(offsetof(struct flow, f0) \
58 + MEMBER_SIZEOF(struct flow, f0) \
59 == offsetof(struct flow, f1))
61 /* Asserts that fields 'f0' and 'f1' are in the same 32-bit aligned word within
63 #define ASSERT_SAME_WORD(f0, f1) \
64 BUILD_ASSERT_DECL(offsetof(struct flow, f0) / 4 \
65 == offsetof(struct flow, f1) / 4)
67 /* Asserts that 'f0' and 'f1' are both sequential and within the same 32-bit
68 * aligned word in struct flow. */
69 #define ASSERT_SEQUENTIAL_SAME_WORD(f0, f1) \
70 ASSERT_SEQUENTIAL(f0, f1); \
71 ASSERT_SAME_WORD(f0, f1)
73 /* miniflow_extract() assumes the following to be true to optimize the
74 * extraction process. */
75 ASSERT_SEQUENTIAL_SAME_WORD(dl_type, vlan_tci);
77 ASSERT_SEQUENTIAL_SAME_WORD(nw_frag, nw_tos);
78 ASSERT_SEQUENTIAL_SAME_WORD(nw_tos, nw_ttl);
79 ASSERT_SEQUENTIAL_SAME_WORD(nw_ttl, nw_proto);
81 /* TCP flags in the middle of a BE64, zeroes in the other half. */
82 BUILD_ASSERT_DECL(offsetof(struct flow, tcp_flags) % 8 == 4);
85 #define TCP_FLAGS_BE32(tcp_ctl) ((OVS_FORCE ovs_be32)TCP_FLAGS_BE16(tcp_ctl) \
88 #define TCP_FLAGS_BE32(tcp_ctl) ((OVS_FORCE ovs_be32)TCP_FLAGS_BE16(tcp_ctl))
91 ASSERT_SEQUENTIAL_SAME_WORD(tp_src, tp_dst);
93 /* Removes 'size' bytes from the head end of '*datap', of size '*sizep', which
94 * must contain at least 'size' bytes of data. Returns the first byte of data
96 static inline const void *
97 data_pull(const void **datap, size_t *sizep, size_t size)
99 const char *data = *datap;
100 *datap = data + size;
105 /* If '*datap' has at least 'size' bytes of data, removes that many bytes from
106 * the head end of '*datap' and returns the first byte removed. Otherwise,
107 * returns a null pointer without modifying '*datap'. */
108 static inline const void *
109 data_try_pull(const void **datap, size_t *sizep, size_t size)
111 return OVS_LIKELY(*sizep >= size) ? data_pull(datap, sizep, size) : NULL;
114 /* Context for pushing data to a miniflow. */
118 uint64_t * const end;
121 /* miniflow_push_* macros allow filling in a miniflow data values in order.
122 * Assertions are needed only when the layout of the struct flow is modified.
123 * 'ofs' is a compile-time constant, which allows most of the code be optimized
124 * away. Some GCC versions gave warnings on ALWAYS_INLINE, so these are
125 * defined as macros. */
127 #if (FLOW_WC_SEQ != 36)
128 #define MINIFLOW_ASSERT(X) ovs_assert(X)
129 BUILD_MESSAGE("FLOW_WC_SEQ changed: miniflow_extract() will have runtime "
130 "assertions enabled. Consider updating FLOW_WC_SEQ after "
133 #define MINIFLOW_ASSERT(X)
136 /* True if 'IDX' and higher bits are not set. */
137 #define ASSERT_FLOWMAP_NOT_SET(FM, IDX) \
139 MINIFLOW_ASSERT(!((FM)->bits[(IDX) / MAP_T_BITS] & \
140 (MAP_MAX << ((IDX) % MAP_T_BITS)))); \
141 for (size_t i = (IDX) / MAP_T_BITS + 1; i < FLOWMAP_UNITS; i++) { \
142 MINIFLOW_ASSERT(!(FM)->bits[i]); \
146 #define miniflow_set_map(MF, OFS) \
148 ASSERT_FLOWMAP_NOT_SET(&MF.map, (OFS)); \
149 flowmap_set(&MF.map, (OFS), 1); \
152 #define miniflow_assert_in_map(MF, OFS) \
153 MINIFLOW_ASSERT(flowmap_is_set(&MF.map, (OFS))); \
154 ASSERT_FLOWMAP_NOT_SET(&MF.map, (OFS) + 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_push_uint8_(MF, OFS, VALUE) \
205 MINIFLOW_ASSERT(MF.data < MF.end); \
207 if ((OFS) % 8 == 0) { \
208 miniflow_set_map(MF, OFS / 8); \
209 *(uint8_t *)MF.data = VALUE; \
210 } else if ((OFS) % 8 == 7) { \
211 miniflow_assert_in_map(MF, OFS / 8); \
212 *((uint8_t *)MF.data + 7) = VALUE; \
215 miniflow_assert_in_map(MF, OFS / 8); \
216 *((uint8_t *)MF.data + ((OFS) % 8)) = VALUE; \
220 #define miniflow_pad_to_64_(MF, OFS) \
222 MINIFLOW_ASSERT((OFS) % 8 != 0); \
223 miniflow_assert_in_map(MF, OFS / 8); \
225 memset((uint8_t *)MF.data + (OFS) % 8, 0, 8 - (OFS) % 8); \
229 #define miniflow_pad_from_64_(MF, OFS) \
231 MINIFLOW_ASSERT(MF.data < MF.end); \
233 MINIFLOW_ASSERT((OFS) % 8 != 0); \
234 miniflow_set_map(MF, OFS / 8); \
236 memset((uint8_t *)MF.data, 0, (OFS) % 8); \
239 #define miniflow_push_be16_(MF, OFS, VALUE) \
240 miniflow_push_uint16_(MF, OFS, (OVS_FORCE uint16_t)VALUE);
242 #define miniflow_push_be8_(MF, OFS, VALUE) \
243 miniflow_push_uint8_(MF, OFS, (OVS_FORCE uint8_t)VALUE);
245 #define miniflow_set_maps(MF, OFS, N_WORDS) \
247 size_t ofs = (OFS); \
248 size_t n_words = (N_WORDS); \
250 MINIFLOW_ASSERT(n_words && MF.data + n_words <= MF.end); \
251 ASSERT_FLOWMAP_NOT_SET(&MF.map, ofs); \
252 flowmap_set(&MF.map, ofs, n_words); \
255 /* Data at 'valuep' may be unaligned. */
256 #define miniflow_push_words_(MF, OFS, VALUEP, N_WORDS) \
258 MINIFLOW_ASSERT((OFS) % 8 == 0); \
259 miniflow_set_maps(MF, (OFS) / 8, (N_WORDS)); \
260 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof *MF.data); \
261 MF.data += (N_WORDS); \
264 /* Push 32-bit words padded to 64-bits. */
265 #define miniflow_push_words_32_(MF, OFS, VALUEP, N_WORDS) \
267 miniflow_set_maps(MF, (OFS) / 8, DIV_ROUND_UP(N_WORDS, 2)); \
268 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof(uint32_t)); \
269 MF.data += DIV_ROUND_UP(N_WORDS, 2); \
270 if ((N_WORDS) & 1) { \
271 *((uint32_t *)MF.data - 1) = 0; \
275 /* Data at 'valuep' may be unaligned. */
276 /* MACs start 64-aligned, and must be followed by other data or padding. */
277 #define miniflow_push_macs_(MF, OFS, VALUEP) \
279 miniflow_set_maps(MF, (OFS) / 8, 2); \
280 memcpy(MF.data, (VALUEP), 2 * ETH_ADDR_LEN); \
281 MF.data += 1; /* First word only. */ \
284 #define miniflow_push_uint32(MF, FIELD, VALUE) \
285 miniflow_push_uint32_(MF, offsetof(struct flow, FIELD), VALUE)
287 #define miniflow_push_be32(MF, FIELD, VALUE) \
288 miniflow_push_be32_(MF, offsetof(struct flow, FIELD), VALUE)
290 #define miniflow_push_uint16(MF, FIELD, VALUE) \
291 miniflow_push_uint16_(MF, offsetof(struct flow, FIELD), VALUE)
293 #define miniflow_push_be16(MF, FIELD, VALUE) \
294 miniflow_push_be16_(MF, offsetof(struct flow, FIELD), VALUE)
296 #define miniflow_push_uint8(MF, FIELD, VALUE) \
297 miniflow_push_uint8_(MF, offsetof(struct flow, FIELD), VALUE)
299 #define miniflow_pad_to_64(MF, FIELD) \
300 miniflow_pad_to_64_(MF, OFFSETOFEND(struct flow, FIELD))
302 #define miniflow_pad_from_64(MF, FIELD) \
303 miniflow_pad_from_64_(MF, offsetof(struct flow, FIELD))
305 #define miniflow_push_words(MF, FIELD, VALUEP, N_WORDS) \
306 miniflow_push_words_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
308 #define miniflow_push_words_32(MF, FIELD, VALUEP, N_WORDS) \
309 miniflow_push_words_32_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
311 #define miniflow_push_macs(MF, FIELD, VALUEP) \
312 miniflow_push_macs_(MF, offsetof(struct flow, FIELD), VALUEP)
314 /* Pulls the MPLS headers at '*datap' and returns the count of them. */
316 parse_mpls(const void **datap, size_t *sizep)
318 const struct mpls_hdr *mh;
321 while ((mh = data_try_pull(datap, sizep, sizeof *mh))) {
323 if (mh->mpls_lse.lo & htons(1 << MPLS_BOS_SHIFT)) {
327 return MIN(count, FLOW_MAX_MPLS_LABELS);
330 static inline ovs_be16
331 parse_vlan(const void **datap, size_t *sizep)
333 const struct eth_header *eth = *datap;
336 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
340 data_pull(datap, sizep, ETH_ADDR_LEN * 2);
342 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
343 if (OVS_LIKELY(*sizep
344 >= sizeof(struct qtag_prefix) + sizeof(ovs_be16))) {
345 const struct qtag_prefix *qp = data_pull(datap, sizep, sizeof *qp);
346 return qp->tci | htons(VLAN_CFI);
352 static inline ovs_be16
353 parse_ethertype(const void **datap, size_t *sizep)
355 const struct llc_snap_header *llc;
358 proto = *(ovs_be16 *) data_pull(datap, sizep, sizeof proto);
359 if (OVS_LIKELY(ntohs(proto) >= ETH_TYPE_MIN)) {
363 if (OVS_UNLIKELY(*sizep < sizeof *llc)) {
364 return htons(FLOW_DL_TYPE_NONE);
368 if (OVS_UNLIKELY(llc->llc.llc_dsap != LLC_DSAP_SNAP
369 || llc->llc.llc_ssap != LLC_SSAP_SNAP
370 || llc->llc.llc_cntl != LLC_CNTL_SNAP
371 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
372 sizeof llc->snap.snap_org))) {
373 return htons(FLOW_DL_TYPE_NONE);
376 data_pull(datap, sizep, sizeof *llc);
378 if (OVS_LIKELY(ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN)) {
379 return llc->snap.snap_type;
382 return htons(FLOW_DL_TYPE_NONE);
386 parse_icmpv6(const void **datap, size_t *sizep, const struct icmp6_hdr *icmp,
387 const struct in6_addr **nd_target,
388 struct eth_addr arp_buf[2])
390 if (icmp->icmp6_code == 0 &&
391 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
392 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
394 *nd_target = data_try_pull(datap, sizep, sizeof **nd_target);
395 if (OVS_UNLIKELY(!*nd_target)) {
399 while (*sizep >= 8) {
400 /* The minimum size of an option is 8 bytes, which also is
401 * the size of Ethernet link-layer options. */
402 const struct ovs_nd_opt *nd_opt = *datap;
403 int opt_len = nd_opt->nd_opt_len * ND_OPT_LEN;
405 if (!opt_len || opt_len > *sizep) {
409 /* Store the link layer address if the appropriate option is
410 * provided. It is considered an error if the same link
411 * layer option is specified twice. */
412 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
414 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[0]))) {
415 arp_buf[0] = nd_opt->nd_opt_mac;
419 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
421 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[1]))) {
422 arp_buf[1] = nd_opt->nd_opt_mac;
428 if (OVS_UNLIKELY(!data_try_pull(datap, sizep, opt_len))) {
438 arp_buf[0] = eth_addr_zero;
439 arp_buf[1] = eth_addr_zero;
442 /* Initializes 'flow' members from 'packet' and 'md'
444 * Initializes 'packet' header l2 pointer to the start of the Ethernet
445 * header, and the layer offsets as follows:
447 * - packet->l2_5_ofs to the start of the MPLS shim header, or UINT16_MAX
448 * when there is no MPLS shim header.
450 * - packet->l3_ofs to just past the Ethernet header, or just past the
451 * vlan_header if one is present, to the first byte of the payload of the
452 * Ethernet frame. UINT16_MAX if the frame is too short to contain an
455 * - packet->l4_ofs to just past the IPv4 header, if one is present and
456 * has at least the content used for the fields of interest for the flow,
457 * otherwise UINT16_MAX.
460 flow_extract(struct dp_packet *packet, struct flow *flow)
464 uint64_t buf[FLOW_U64S];
467 COVERAGE_INC(flow_extract);
469 miniflow_extract(packet, &m.mf);
470 miniflow_expand(&m.mf, flow);
473 /* Caller is responsible for initializing 'dst' with enough storage for
474 * FLOW_U64S * 8 bytes. */
476 miniflow_extract(struct dp_packet *packet, struct miniflow *dst)
478 const struct pkt_metadata *md = &packet->md;
479 const void *data = dp_packet_data(packet);
480 size_t size = dp_packet_size(packet);
481 uint64_t *values = miniflow_values(dst);
482 struct mf_ctx mf = { FLOWMAP_EMPTY_INITIALIZER, values,
483 values + FLOW_U64S };
486 uint8_t nw_frag, nw_tos, nw_ttl, nw_proto;
489 if (flow_tnl_dst_is_set(&md->tunnel)) {
490 miniflow_push_words(mf, tunnel, &md->tunnel,
491 offsetof(struct flow_tnl, metadata) /
494 if (!(md->tunnel.flags & FLOW_TNL_F_UDPIF)) {
495 if (md->tunnel.metadata.present.map) {
496 miniflow_push_words(mf, tunnel.metadata, &md->tunnel.metadata,
497 sizeof md->tunnel.metadata /
501 if (md->tunnel.metadata.present.len) {
502 miniflow_push_words(mf, tunnel.metadata.present,
503 &md->tunnel.metadata.present, 1);
504 miniflow_push_words(mf, tunnel.metadata.opts.gnv,
505 md->tunnel.metadata.opts.gnv,
506 DIV_ROUND_UP(md->tunnel.metadata.present.len,
511 if (md->skb_priority || md->pkt_mark) {
512 miniflow_push_uint32(mf, skb_priority, md->skb_priority);
513 miniflow_push_uint32(mf, pkt_mark, md->pkt_mark);
515 miniflow_push_uint32(mf, dp_hash, md->dp_hash);
516 miniflow_push_uint32(mf, in_port, odp_to_u32(md->in_port.odp_port));
517 if (md->recirc_id || md->ct_state) {
518 miniflow_push_uint32(mf, recirc_id, md->recirc_id);
519 miniflow_push_uint16(mf, ct_state, md->ct_state);
520 miniflow_push_uint16(mf, ct_zone, md->ct_zone);
524 miniflow_push_uint32(mf, ct_mark, md->ct_mark);
525 miniflow_pad_to_64(mf, ct_mark);
527 if (!ovs_u128_is_zero(md->ct_label)) {
528 miniflow_push_words(mf, ct_label, &md->ct_label,
529 sizeof md->ct_label / sizeof(uint64_t));
533 /* Initialize packet's layer pointer and offsets. */
535 dp_packet_reset_offsets(packet);
537 /* Must have full Ethernet header to proceed. */
538 if (OVS_UNLIKELY(size < sizeof(struct eth_header))) {
544 ASSERT_SEQUENTIAL(dl_dst, dl_src);
545 miniflow_push_macs(mf, dl_dst, data);
546 /* dl_type, vlan_tci. */
547 vlan_tci = parse_vlan(&data, &size);
548 dl_type = parse_ethertype(&data, &size);
549 miniflow_push_be16(mf, dl_type, dl_type);
550 miniflow_push_be16(mf, vlan_tci, vlan_tci);
554 if (OVS_UNLIKELY(eth_type_mpls(dl_type))) {
556 const void *mpls = data;
558 packet->l2_5_ofs = (char *)data - l2;
559 count = parse_mpls(&data, &size);
560 miniflow_push_words_32(mf, mpls_lse, mpls, count);
564 packet->l3_ofs = (char *)data - l2;
567 if (OVS_LIKELY(dl_type == htons(ETH_TYPE_IP))) {
568 const struct ip_header *nh = data;
572 if (OVS_UNLIKELY(size < IP_HEADER_LEN)) {
575 ip_len = IP_IHL(nh->ip_ihl_ver) * 4;
577 if (OVS_UNLIKELY(ip_len < IP_HEADER_LEN)) {
580 if (OVS_UNLIKELY(size < ip_len)) {
583 tot_len = ntohs(nh->ip_tot_len);
584 if (OVS_UNLIKELY(tot_len > size)) {
587 if (OVS_UNLIKELY(size - tot_len > UINT8_MAX)) {
590 dp_packet_set_l2_pad_size(packet, size - tot_len);
591 size = tot_len; /* Never pull padding. */
593 /* Push both source and destination address at once. */
594 miniflow_push_words(mf, nw_src, &nh->ip_src, 1);
596 miniflow_push_be32(mf, ipv6_label, 0); /* Padding for IPv4. */
600 nw_proto = nh->ip_proto;
601 if (OVS_UNLIKELY(IP_IS_FRAGMENT(nh->ip_frag_off))) {
602 nw_frag = FLOW_NW_FRAG_ANY;
603 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
604 nw_frag |= FLOW_NW_FRAG_LATER;
607 data_pull(&data, &size, ip_len);
608 } else if (dl_type == htons(ETH_TYPE_IPV6)) {
609 const struct ovs_16aligned_ip6_hdr *nh;
613 if (OVS_UNLIKELY(size < sizeof *nh)) {
616 nh = data_pull(&data, &size, sizeof *nh);
618 plen = ntohs(nh->ip6_plen);
619 if (OVS_UNLIKELY(plen > size)) {
622 /* Jumbo Payload option not supported yet. */
623 if (OVS_UNLIKELY(size - plen > UINT8_MAX)) {
626 dp_packet_set_l2_pad_size(packet, size - plen);
627 size = plen; /* Never pull padding. */
629 miniflow_push_words(mf, ipv6_src, &nh->ip6_src,
630 sizeof nh->ip6_src / 8);
631 miniflow_push_words(mf, ipv6_dst, &nh->ip6_dst,
632 sizeof nh->ip6_dst / 8);
634 tc_flow = get_16aligned_be32(&nh->ip6_flow);
636 ovs_be32 label = tc_flow & htonl(IPV6_LABEL_MASK);
637 miniflow_push_be32(mf, ipv6_label, label);
640 nw_tos = ntohl(tc_flow) >> 20;
641 nw_ttl = nh->ip6_hlim;
642 nw_proto = nh->ip6_nxt;
645 if (OVS_LIKELY((nw_proto != IPPROTO_HOPOPTS)
646 && (nw_proto != IPPROTO_ROUTING)
647 && (nw_proto != IPPROTO_DSTOPTS)
648 && (nw_proto != IPPROTO_AH)
649 && (nw_proto != IPPROTO_FRAGMENT))) {
650 /* It's either a terminal header (e.g., TCP, UDP) or one we
651 * don't understand. In either case, we're done with the
652 * packet, so use it to fill in 'nw_proto'. */
656 /* We only verify that at least 8 bytes of the next header are
657 * available, but many of these headers are longer. Ensure that
658 * accesses within the extension header are within those first 8
659 * bytes. All extension headers are required to be at least 8
661 if (OVS_UNLIKELY(size < 8)) {
665 if ((nw_proto == IPPROTO_HOPOPTS)
666 || (nw_proto == IPPROTO_ROUTING)
667 || (nw_proto == IPPROTO_DSTOPTS)) {
668 /* These headers, while different, have the fields we care
669 * about in the same location and with the same
671 const struct ip6_ext *ext_hdr = data;
672 nw_proto = ext_hdr->ip6e_nxt;
673 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
674 (ext_hdr->ip6e_len + 1) * 8))) {
677 } else if (nw_proto == IPPROTO_AH) {
678 /* A standard AH definition isn't available, but the fields
679 * we care about are in the same location as the generic
680 * option header--only the header length is calculated
682 const struct ip6_ext *ext_hdr = data;
683 nw_proto = ext_hdr->ip6e_nxt;
684 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
685 (ext_hdr->ip6e_len + 2) * 4))) {
688 } else if (nw_proto == IPPROTO_FRAGMENT) {
689 const struct ovs_16aligned_ip6_frag *frag_hdr = data;
691 nw_proto = frag_hdr->ip6f_nxt;
692 if (!data_try_pull(&data, &size, sizeof *frag_hdr)) {
696 /* We only process the first fragment. */
697 if (frag_hdr->ip6f_offlg != htons(0)) {
698 nw_frag = FLOW_NW_FRAG_ANY;
699 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
700 nw_frag |= FLOW_NW_FRAG_LATER;
701 nw_proto = IPPROTO_FRAGMENT;
708 if (dl_type == htons(ETH_TYPE_ARP) ||
709 dl_type == htons(ETH_TYPE_RARP)) {
710 struct eth_addr arp_buf[2];
711 const struct arp_eth_header *arp = (const struct arp_eth_header *)
712 data_try_pull(&data, &size, ARP_ETH_HEADER_LEN);
714 if (OVS_LIKELY(arp) && OVS_LIKELY(arp->ar_hrd == htons(1))
715 && OVS_LIKELY(arp->ar_pro == htons(ETH_TYPE_IP))
716 && OVS_LIKELY(arp->ar_hln == ETH_ADDR_LEN)
717 && OVS_LIKELY(arp->ar_pln == 4)) {
718 miniflow_push_be32(mf, nw_src,
719 get_16aligned_be32(&arp->ar_spa));
720 miniflow_push_be32(mf, nw_dst,
721 get_16aligned_be32(&arp->ar_tpa));
723 /* We only match on the lower 8 bits of the opcode. */
724 if (OVS_LIKELY(ntohs(arp->ar_op) <= 0xff)) {
725 miniflow_push_be32(mf, ipv6_label, 0); /* Pad with ARP. */
726 miniflow_push_be32(mf, nw_frag, htonl(ntohs(arp->ar_op)));
729 /* Must be adjacent. */
730 ASSERT_SEQUENTIAL(arp_sha, arp_tha);
732 arp_buf[0] = arp->ar_sha;
733 arp_buf[1] = arp->ar_tha;
734 miniflow_push_macs(mf, arp_sha, arp_buf);
735 miniflow_pad_to_64(mf, arp_tha);
741 packet->l4_ofs = (char *)data - l2;
742 miniflow_push_be32(mf, nw_frag,
743 BYTES_TO_BE32(nw_frag, nw_tos, nw_ttl, nw_proto));
745 if (OVS_LIKELY(!(nw_frag & FLOW_NW_FRAG_LATER))) {
746 if (OVS_LIKELY(nw_proto == IPPROTO_TCP)) {
747 if (OVS_LIKELY(size >= TCP_HEADER_LEN)) {
748 const struct tcp_header *tcp = data;
750 miniflow_push_be32(mf, arp_tha.ea[2], 0);
751 miniflow_push_be32(mf, tcp_flags,
752 TCP_FLAGS_BE32(tcp->tcp_ctl));
753 miniflow_push_be16(mf, tp_src, tcp->tcp_src);
754 miniflow_push_be16(mf, tp_dst, tcp->tcp_dst);
755 miniflow_pad_to_64(mf, tp_dst);
757 } else if (OVS_LIKELY(nw_proto == IPPROTO_UDP)) {
758 if (OVS_LIKELY(size >= UDP_HEADER_LEN)) {
759 const struct udp_header *udp = data;
761 miniflow_push_be16(mf, tp_src, udp->udp_src);
762 miniflow_push_be16(mf, tp_dst, udp->udp_dst);
763 miniflow_pad_to_64(mf, tp_dst);
765 } else if (OVS_LIKELY(nw_proto == IPPROTO_SCTP)) {
766 if (OVS_LIKELY(size >= SCTP_HEADER_LEN)) {
767 const struct sctp_header *sctp = data;
769 miniflow_push_be16(mf, tp_src, sctp->sctp_src);
770 miniflow_push_be16(mf, tp_dst, sctp->sctp_dst);
771 miniflow_pad_to_64(mf, tp_dst);
773 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMP)) {
774 if (OVS_LIKELY(size >= ICMP_HEADER_LEN)) {
775 const struct icmp_header *icmp = data;
777 miniflow_push_be16(mf, tp_src, htons(icmp->icmp_type));
778 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp_code));
779 miniflow_pad_to_64(mf, tp_dst);
781 } else if (OVS_LIKELY(nw_proto == IPPROTO_IGMP)) {
782 if (OVS_LIKELY(size >= IGMP_HEADER_LEN)) {
783 const struct igmp_header *igmp = data;
785 miniflow_push_be16(mf, tp_src, htons(igmp->igmp_type));
786 miniflow_push_be16(mf, tp_dst, htons(igmp->igmp_code));
787 miniflow_push_be32(mf, igmp_group_ip4,
788 get_16aligned_be32(&igmp->group));
790 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMPV6)) {
791 if (OVS_LIKELY(size >= sizeof(struct icmp6_hdr))) {
792 const struct in6_addr *nd_target = NULL;
793 struct eth_addr arp_buf[2] = { { { { 0 } } } };
794 const struct icmp6_hdr *icmp = data_pull(&data, &size,
796 parse_icmpv6(&data, &size, icmp, &nd_target, arp_buf);
798 miniflow_push_words(mf, nd_target, nd_target,
799 sizeof *nd_target / sizeof(uint64_t));
801 miniflow_push_macs(mf, arp_sha, arp_buf);
802 miniflow_pad_to_64(mf, arp_tha);
803 miniflow_push_be16(mf, tp_src, htons(icmp->icmp6_type));
804 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp6_code));
805 miniflow_pad_to_64(mf, tp_dst);
813 /* For every bit of a field that is wildcarded in 'wildcards', sets the
814 * corresponding bit in 'flow' to zero. */
816 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
818 uint64_t *flow_u64 = (uint64_t *) flow;
819 const uint64_t *wc_u64 = (const uint64_t *) &wildcards->masks;
822 for (i = 0; i < FLOW_U64S; i++) {
823 flow_u64[i] &= wc_u64[i];
828 flow_unwildcard_tp_ports(const struct flow *flow, struct flow_wildcards *wc)
830 if (flow->nw_proto != IPPROTO_ICMP) {
831 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
832 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
834 wc->masks.tp_src = htons(0xff);
835 wc->masks.tp_dst = htons(0xff);
839 /* Initializes 'flow_metadata' with the metadata found in 'flow'. */
841 flow_get_metadata(const struct flow *flow, struct match *flow_metadata)
845 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 36);
847 match_init_catchall(flow_metadata);
848 if (flow->tunnel.tun_id != htonll(0)) {
849 match_set_tun_id(flow_metadata, flow->tunnel.tun_id);
851 if (flow->tunnel.flags & FLOW_TNL_PUB_F_MASK) {
852 match_set_tun_flags(flow_metadata,
853 flow->tunnel.flags & FLOW_TNL_PUB_F_MASK);
855 if (flow->tunnel.ip_src) {
856 match_set_tun_src(flow_metadata, flow->tunnel.ip_src);
858 if (flow->tunnel.ip_dst) {
859 match_set_tun_dst(flow_metadata, flow->tunnel.ip_dst);
861 if (ipv6_addr_is_set(&flow->tunnel.ipv6_src)) {
862 match_set_tun_ipv6_src(flow_metadata, &flow->tunnel.ipv6_src);
864 if (ipv6_addr_is_set(&flow->tunnel.ipv6_dst)) {
865 match_set_tun_ipv6_dst(flow_metadata, &flow->tunnel.ipv6_dst);
867 if (flow->tunnel.gbp_id != htons(0)) {
868 match_set_tun_gbp_id(flow_metadata, flow->tunnel.gbp_id);
870 if (flow->tunnel.gbp_flags) {
871 match_set_tun_gbp_flags(flow_metadata, flow->tunnel.gbp_flags);
873 tun_metadata_get_fmd(&flow->tunnel, flow_metadata);
874 if (flow->metadata != htonll(0)) {
875 match_set_metadata(flow_metadata, flow->metadata);
878 for (i = 0; i < FLOW_N_REGS; i++) {
880 match_set_reg(flow_metadata, i, flow->regs[i]);
884 if (flow->pkt_mark != 0) {
885 match_set_pkt_mark(flow_metadata, flow->pkt_mark);
888 match_set_in_port(flow_metadata, flow->in_port.ofp_port);
889 if (flow->ct_state != 0) {
890 match_set_ct_state(flow_metadata, flow->ct_state);
892 if (flow->ct_zone != 0) {
893 match_set_ct_zone(flow_metadata, flow->ct_zone);
895 if (flow->ct_mark != 0) {
896 match_set_ct_mark(flow_metadata, flow->ct_mark);
898 if (!ovs_u128_is_zero(flow->ct_label)) {
899 match_set_ct_label(flow_metadata, flow->ct_label);
903 const char *ct_state_to_string(uint32_t state)
928 flow_to_string(const struct flow *flow)
930 struct ds ds = DS_EMPTY_INITIALIZER;
931 flow_format(&ds, flow);
936 flow_tun_flag_to_string(uint32_t flags)
939 case FLOW_TNL_F_DONT_FRAGMENT:
941 case FLOW_TNL_F_CSUM:
953 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
954 uint32_t flags, char del)
959 ds_put_char(ds, '0');
963 uint32_t bit = rightmost_1bit(flags);
966 s = bit_to_string(bit);
968 ds_put_format(ds, "%s%c", s, del);
977 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
983 format_flags_masked(struct ds *ds, const char *name,
984 const char *(*bit_to_string)(uint32_t), uint32_t flags,
985 uint32_t mask, uint32_t max_mask)
988 ds_put_format(ds, "%s%s=%s", colors.param, name, colors.end);
991 if (mask == max_mask) {
992 format_flags(ds, bit_to_string, flags, '|');
997 ds_put_cstr(ds, "0/0");
1002 uint32_t bit = rightmost_1bit(mask);
1003 const char *s = bit_to_string(bit);
1005 ds_put_format(ds, "%s%s", (flags & bit) ? "+" : "-",
1006 s ? s : "[Unknown]");
1011 /* Scans a string 's' of flags to determine their numerical value and
1012 * returns the number of characters parsed using 'bit_to_string' to
1013 * lookup flag names. Scanning continues until the character 'end' is
1016 * In the event of a failure, a negative error code will be returned. In
1017 * addition, if 'res_string' is non-NULL then a descriptive string will
1018 * be returned incorporating the identifying string 'field_name'. This
1019 * error string must be freed by the caller.
1021 * Upon success, the flag values will be stored in 'res_flags' and
1022 * optionally 'res_mask', if it is non-NULL (if it is NULL then any masks
1023 * present in the original string will be considered an error). The
1024 * caller may restrict the acceptable set of values through the mask
1027 parse_flags(const char *s, const char *(*bit_to_string)(uint32_t),
1028 char end, const char *field_name, char **res_string,
1029 uint32_t *res_flags, uint32_t allowed, uint32_t *res_mask)
1031 uint32_t result = 0;
1034 /* Parse masked flags in numeric format? */
1035 if (res_mask && ovs_scan(s, "%"SCNi32"/%"SCNi32"%n",
1036 res_flags, res_mask, &n) && n > 0) {
1037 if (*res_flags & ~allowed || *res_mask & ~allowed) {
1045 if (res_mask && (*s == '+' || *s == '-')) {
1046 uint32_t flags = 0, mask = 0;
1048 /* Parse masked flags. */
1049 while (s[0] != end) {
1056 } else if (s[0] == '-') {
1060 *res_string = xasprintf("%s: %s must be preceded by '+' "
1061 "(for SET) or '-' (NOT SET)", s,
1069 for (bit = 1; bit; bit <<= 1) {
1070 const char *fname = bit_to_string(bit);
1076 len = strlen(fname);
1077 if (strncmp(s, fname, len) ||
1078 (s[len] != '+' && s[len] != '-' && s[len] != end)) {
1083 /* bit already set. */
1085 *res_string = xasprintf("%s: Each %s flag can be "
1086 "specified only once", s,
1091 if (!(bit & allowed)) {
1113 /* Parse unmasked flags. If a flag is present, it is set, otherwise
1115 while (s[n] != end) {
1116 unsigned long long int flags;
1120 if (ovs_scan(&s[n], "%lli%n", &flags, &n0)) {
1121 if (flags & ~allowed) {
1124 n += n0 + (s[n + n0] == '|');
1129 for (bit = 1; bit; bit <<= 1) {
1130 const char *name = bit_to_string(bit);
1138 if (!strncmp(s + n, name, len) &&
1139 (s[n + len] == '|' || s[n + len] == end)) {
1140 if (!(bit & allowed)) {
1144 n += len + (s[n + len] == '|');
1154 *res_flags = result;
1156 *res_mask = UINT32_MAX;
1165 *res_string = xasprintf("%s: unknown %s flag(s)", s, field_name);
1171 flow_format(struct ds *ds, const struct flow *flow)
1174 struct flow_wildcards *wc = &match.wc;
1176 match_wc_init(&match, flow);
1178 /* As this function is most often used for formatting a packet in a
1179 * packet-in message, skip formatting the packet context fields that are
1180 * all-zeroes to make the print-out easier on the eyes. This means that a
1181 * missing context field implies a zero value for that field. This is
1182 * similar to OpenFlow encoding of these fields, as the specification
1183 * states that all-zeroes context fields should not be encoded in the
1184 * packet-in messages. */
1185 if (!flow->in_port.ofp_port) {
1186 WC_UNMASK_FIELD(wc, in_port);
1188 if (!flow->skb_priority) {
1189 WC_UNMASK_FIELD(wc, skb_priority);
1191 if (!flow->pkt_mark) {
1192 WC_UNMASK_FIELD(wc, pkt_mark);
1194 if (!flow->recirc_id) {
1195 WC_UNMASK_FIELD(wc, recirc_id);
1197 if (!flow->dp_hash) {
1198 WC_UNMASK_FIELD(wc, dp_hash);
1200 if (!flow->ct_state) {
1201 WC_UNMASK_FIELD(wc, ct_state);
1203 if (!flow->ct_zone) {
1204 WC_UNMASK_FIELD(wc, ct_zone);
1206 if (!flow->ct_mark) {
1207 WC_UNMASK_FIELD(wc, ct_mark);
1209 if (ovs_u128_is_zero(flow->ct_label)) {
1210 WC_UNMASK_FIELD(wc, ct_label);
1212 for (int i = 0; i < FLOW_N_REGS; i++) {
1213 if (!flow->regs[i]) {
1214 WC_UNMASK_FIELD(wc, regs[i]);
1217 if (!flow->metadata) {
1218 WC_UNMASK_FIELD(wc, metadata);
1221 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
1225 flow_print(FILE *stream, const struct flow *flow)
1227 char *s = flow_to_string(flow);
1232 /* flow_wildcards functions. */
1234 /* Initializes 'wc' as a set of wildcards that matches every packet. */
1236 flow_wildcards_init_catchall(struct flow_wildcards *wc)
1238 memset(&wc->masks, 0, sizeof wc->masks);
1241 /* Converts a flow into flow wildcards. It sets the wildcard masks based on
1242 * the packet headers extracted to 'flow'. It will not set the mask for fields
1243 * that do not make sense for the packet type. OpenFlow-only metadata is
1244 * wildcarded, but other metadata is unconditionally exact-matched. */
1245 void flow_wildcards_init_for_packet(struct flow_wildcards *wc,
1246 const struct flow *flow)
1248 memset(&wc->masks, 0x0, sizeof wc->masks);
1250 /* Update this function whenever struct flow changes. */
1251 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 36);
1253 if (flow_tnl_dst_is_set(&flow->tunnel)) {
1254 if (flow->tunnel.flags & FLOW_TNL_F_KEY) {
1255 WC_MASK_FIELD(wc, tunnel.tun_id);
1257 WC_MASK_FIELD(wc, tunnel.ip_src);
1258 WC_MASK_FIELD(wc, tunnel.ip_dst);
1259 WC_MASK_FIELD(wc, tunnel.ipv6_src);
1260 WC_MASK_FIELD(wc, tunnel.ipv6_dst);
1261 WC_MASK_FIELD(wc, tunnel.flags);
1262 WC_MASK_FIELD(wc, tunnel.ip_tos);
1263 WC_MASK_FIELD(wc, tunnel.ip_ttl);
1264 WC_MASK_FIELD(wc, tunnel.tp_src);
1265 WC_MASK_FIELD(wc, tunnel.tp_dst);
1266 WC_MASK_FIELD(wc, tunnel.gbp_id);
1267 WC_MASK_FIELD(wc, tunnel.gbp_flags);
1269 if (!(flow->tunnel.flags & FLOW_TNL_F_UDPIF)) {
1270 if (flow->tunnel.metadata.present.map) {
1271 wc->masks.tunnel.metadata.present.map =
1272 flow->tunnel.metadata.present.map;
1273 WC_MASK_FIELD(wc, tunnel.metadata.opts.u8);
1276 WC_MASK_FIELD(wc, tunnel.metadata.present.len);
1277 memset(wc->masks.tunnel.metadata.opts.gnv, 0xff,
1278 flow->tunnel.metadata.present.len);
1280 } else if (flow->tunnel.tun_id) {
1281 WC_MASK_FIELD(wc, tunnel.tun_id);
1284 /* metadata, regs, and conj_id wildcarded. */
1286 WC_MASK_FIELD(wc, skb_priority);
1287 WC_MASK_FIELD(wc, pkt_mark);
1288 WC_MASK_FIELD(wc, ct_state);
1289 WC_MASK_FIELD(wc, ct_zone);
1290 WC_MASK_FIELD(wc, ct_mark);
1291 WC_MASK_FIELD(wc, ct_label);
1292 WC_MASK_FIELD(wc, recirc_id);
1293 WC_MASK_FIELD(wc, dp_hash);
1294 WC_MASK_FIELD(wc, in_port);
1296 /* actset_output wildcarded. */
1298 WC_MASK_FIELD(wc, dl_dst);
1299 WC_MASK_FIELD(wc, dl_src);
1300 WC_MASK_FIELD(wc, dl_type);
1301 WC_MASK_FIELD(wc, vlan_tci);
1303 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1304 WC_MASK_FIELD(wc, nw_src);
1305 WC_MASK_FIELD(wc, nw_dst);
1306 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1307 WC_MASK_FIELD(wc, ipv6_src);
1308 WC_MASK_FIELD(wc, ipv6_dst);
1309 WC_MASK_FIELD(wc, ipv6_label);
1310 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1311 flow->dl_type == htons(ETH_TYPE_RARP)) {
1312 WC_MASK_FIELD(wc, nw_src);
1313 WC_MASK_FIELD(wc, nw_dst);
1314 WC_MASK_FIELD(wc, nw_proto);
1315 WC_MASK_FIELD(wc, arp_sha);
1316 WC_MASK_FIELD(wc, arp_tha);
1318 } else if (eth_type_mpls(flow->dl_type)) {
1319 for (int i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1320 WC_MASK_FIELD(wc, mpls_lse[i]);
1321 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1327 return; /* Unknown ethertype. */
1331 WC_MASK_FIELD(wc, nw_frag);
1332 WC_MASK_FIELD(wc, nw_tos);
1333 WC_MASK_FIELD(wc, nw_ttl);
1334 WC_MASK_FIELD(wc, nw_proto);
1336 /* No transport layer header in later fragments. */
1337 if (!(flow->nw_frag & FLOW_NW_FRAG_LATER) &&
1338 (flow->nw_proto == IPPROTO_ICMP ||
1339 flow->nw_proto == IPPROTO_ICMPV6 ||
1340 flow->nw_proto == IPPROTO_TCP ||
1341 flow->nw_proto == IPPROTO_UDP ||
1342 flow->nw_proto == IPPROTO_SCTP ||
1343 flow->nw_proto == IPPROTO_IGMP)) {
1344 WC_MASK_FIELD(wc, tp_src);
1345 WC_MASK_FIELD(wc, tp_dst);
1347 if (flow->nw_proto == IPPROTO_TCP) {
1348 WC_MASK_FIELD(wc, tcp_flags);
1349 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
1350 WC_MASK_FIELD(wc, arp_sha);
1351 WC_MASK_FIELD(wc, arp_tha);
1352 WC_MASK_FIELD(wc, nd_target);
1353 } else if (flow->nw_proto == IPPROTO_IGMP) {
1354 WC_MASK_FIELD(wc, igmp_group_ip4);
1359 /* Return a map of possible fields for a packet of the same type as 'flow'.
1360 * Including extra bits in the returned mask is not wrong, it is just less
1363 * This is a less precise version of flow_wildcards_init_for_packet() above. */
1365 flow_wc_map(const struct flow *flow, struct flowmap *map)
1367 /* Update this function whenever struct flow changes. */
1368 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 36);
1372 if (flow_tnl_dst_is_set(&flow->tunnel)) {
1373 FLOWMAP_SET__(map, tunnel, offsetof(struct flow_tnl, metadata));
1374 if (!(flow->tunnel.flags & FLOW_TNL_F_UDPIF)) {
1375 if (flow->tunnel.metadata.present.map) {
1376 FLOWMAP_SET(map, tunnel.metadata);
1379 FLOWMAP_SET(map, tunnel.metadata.present.len);
1380 FLOWMAP_SET__(map, tunnel.metadata.opts.gnv,
1381 flow->tunnel.metadata.present.len);
1385 /* Metadata fields that can appear on packet input. */
1386 FLOWMAP_SET(map, skb_priority);
1387 FLOWMAP_SET(map, pkt_mark);
1388 FLOWMAP_SET(map, recirc_id);
1389 FLOWMAP_SET(map, dp_hash);
1390 FLOWMAP_SET(map, in_port);
1391 FLOWMAP_SET(map, dl_dst);
1392 FLOWMAP_SET(map, dl_src);
1393 FLOWMAP_SET(map, dl_type);
1394 FLOWMAP_SET(map, vlan_tci);
1395 FLOWMAP_SET(map, ct_state);
1396 FLOWMAP_SET(map, ct_zone);
1397 FLOWMAP_SET(map, ct_mark);
1398 FLOWMAP_SET(map, ct_label);
1400 /* Ethertype-dependent fields. */
1401 if (OVS_LIKELY(flow->dl_type == htons(ETH_TYPE_IP))) {
1402 FLOWMAP_SET(map, nw_src);
1403 FLOWMAP_SET(map, nw_dst);
1404 FLOWMAP_SET(map, nw_proto);
1405 FLOWMAP_SET(map, nw_frag);
1406 FLOWMAP_SET(map, nw_tos);
1407 FLOWMAP_SET(map, nw_ttl);
1408 FLOWMAP_SET(map, tp_src);
1409 FLOWMAP_SET(map, tp_dst);
1411 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_IGMP)) {
1412 FLOWMAP_SET(map, igmp_group_ip4);
1414 FLOWMAP_SET(map, tcp_flags);
1416 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1417 FLOWMAP_SET(map, ipv6_src);
1418 FLOWMAP_SET(map, ipv6_dst);
1419 FLOWMAP_SET(map, ipv6_label);
1420 FLOWMAP_SET(map, nw_proto);
1421 FLOWMAP_SET(map, nw_frag);
1422 FLOWMAP_SET(map, nw_tos);
1423 FLOWMAP_SET(map, nw_ttl);
1424 FLOWMAP_SET(map, tp_src);
1425 FLOWMAP_SET(map, tp_dst);
1427 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_ICMPV6)) {
1428 FLOWMAP_SET(map, nd_target);
1429 FLOWMAP_SET(map, arp_sha);
1430 FLOWMAP_SET(map, arp_tha);
1432 FLOWMAP_SET(map, tcp_flags);
1434 } else if (eth_type_mpls(flow->dl_type)) {
1435 FLOWMAP_SET(map, mpls_lse);
1436 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1437 flow->dl_type == htons(ETH_TYPE_RARP)) {
1438 FLOWMAP_SET(map, nw_src);
1439 FLOWMAP_SET(map, nw_dst);
1440 FLOWMAP_SET(map, nw_proto);
1441 FLOWMAP_SET(map, arp_sha);
1442 FLOWMAP_SET(map, arp_tha);
1446 /* Clear the metadata and register wildcard masks. They are not packet
1449 flow_wildcards_clear_non_packet_fields(struct flow_wildcards *wc)
1451 /* Update this function whenever struct flow changes. */
1452 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 36);
1454 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
1455 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
1456 wc->masks.actset_output = 0;
1457 wc->masks.conj_id = 0;
1460 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
1463 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
1465 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1468 for (i = 0; i < FLOW_U64S; i++) {
1476 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
1477 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
1478 * in 'src1' or 'src2' or both. */
1480 flow_wildcards_and(struct flow_wildcards *dst,
1481 const struct flow_wildcards *src1,
1482 const struct flow_wildcards *src2)
1484 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1485 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1486 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1489 for (i = 0; i < FLOW_U64S; i++) {
1490 dst_u64[i] = src1_u64[i] & src2_u64[i];
1494 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
1495 * is, a bit or a field is wildcarded in 'dst' if it is neither
1496 * wildcarded in 'src1' nor 'src2'. */
1498 flow_wildcards_or(struct flow_wildcards *dst,
1499 const struct flow_wildcards *src1,
1500 const struct flow_wildcards *src2)
1502 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1503 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1504 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1507 for (i = 0; i < FLOW_U64S; i++) {
1508 dst_u64[i] = src1_u64[i] | src2_u64[i];
1512 /* Returns a hash of the wildcards in 'wc'. */
1514 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
1516 return flow_hash(&wc->masks, basis);
1519 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
1522 flow_wildcards_equal(const struct flow_wildcards *a,
1523 const struct flow_wildcards *b)
1525 return flow_equal(&a->masks, &b->masks);
1528 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
1529 * 'b', false otherwise. */
1531 flow_wildcards_has_extra(const struct flow_wildcards *a,
1532 const struct flow_wildcards *b)
1534 const uint64_t *a_u64 = (const uint64_t *) &a->masks;
1535 const uint64_t *b_u64 = (const uint64_t *) &b->masks;
1538 for (i = 0; i < FLOW_U64S; i++) {
1539 if ((a_u64[i] & b_u64[i]) != b_u64[i]) {
1546 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
1547 * in 'wc' do not need to be equal in 'a' and 'b'. */
1549 flow_equal_except(const struct flow *a, const struct flow *b,
1550 const struct flow_wildcards *wc)
1552 const uint64_t *a_u64 = (const uint64_t *) a;
1553 const uint64_t *b_u64 = (const uint64_t *) b;
1554 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1557 for (i = 0; i < FLOW_U64S; i++) {
1558 if ((a_u64[i] ^ b_u64[i]) & wc_u64[i]) {
1565 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1566 * (A 0-bit indicates a wildcard bit.) */
1568 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
1570 wc->masks.regs[idx] = mask;
1573 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1574 * (A 0-bit indicates a wildcard bit.) */
1576 flow_wildcards_set_xreg_mask(struct flow_wildcards *wc, int idx, uint64_t mask)
1578 flow_set_xreg(&wc->masks, idx, mask);
1581 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1582 * (A 0-bit indicates a wildcard bit.) */
1584 flow_wildcards_set_xxreg_mask(struct flow_wildcards *wc, int idx,
1587 flow_set_xxreg(&wc->masks, idx, mask);
1590 /* Calculates the 5-tuple hash from the given miniflow.
1591 * This returns the same value as flow_hash_5tuple for the corresponding
1594 miniflow_hash_5tuple(const struct miniflow *flow, uint32_t basis)
1596 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 36);
1597 uint32_t hash = basis;
1600 ovs_be16 dl_type = MINIFLOW_GET_BE16(flow, dl_type);
1603 if (dl_type == htons(ETH_TYPE_IPV6)) {
1604 struct flowmap map = FLOWMAP_EMPTY_INITIALIZER;
1607 FLOWMAP_SET(&map, ipv6_src);
1608 FLOWMAP_SET(&map, ipv6_dst);
1610 MINIFLOW_FOR_EACH_IN_FLOWMAP(value, flow, map) {
1611 hash = hash_add64(hash, value);
1613 } else if (dl_type == htons(ETH_TYPE_IP)
1614 || dl_type == htons(ETH_TYPE_ARP)) {
1615 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_src));
1616 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_dst));
1621 nw_proto = MINIFLOW_GET_U8(flow, nw_proto);
1622 hash = hash_add(hash, nw_proto);
1623 if (nw_proto != IPPROTO_TCP && nw_proto != IPPROTO_UDP
1624 && nw_proto != IPPROTO_SCTP && nw_proto != IPPROTO_ICMP
1625 && nw_proto != IPPROTO_ICMPV6) {
1629 /* Add both ports at once. */
1630 hash = hash_add(hash, MINIFLOW_GET_U32(flow, tp_src));
1633 return hash_finish(hash, 42);
1636 ASSERT_SEQUENTIAL_SAME_WORD(tp_src, tp_dst);
1637 ASSERT_SEQUENTIAL(ipv6_src, ipv6_dst);
1639 /* Calculates the 5-tuple hash from the given flow. */
1641 flow_hash_5tuple(const struct flow *flow, uint32_t basis)
1643 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 36);
1644 uint32_t hash = basis;
1648 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1649 const uint64_t *flow_u64 = (const uint64_t *)flow;
1650 int ofs = offsetof(struct flow, ipv6_src) / 8;
1651 int end = ofs + 2 * sizeof flow->ipv6_src / 8;
1653 for (;ofs < end; ofs++) {
1654 hash = hash_add64(hash, flow_u64[ofs]);
1656 } else if (flow->dl_type == htons(ETH_TYPE_IP)
1657 || flow->dl_type == htons(ETH_TYPE_ARP)) {
1658 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_src);
1659 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_dst);
1664 hash = hash_add(hash, flow->nw_proto);
1665 if (flow->nw_proto != IPPROTO_TCP && flow->nw_proto != IPPROTO_UDP
1666 && flow->nw_proto != IPPROTO_SCTP && flow->nw_proto != IPPROTO_ICMP
1667 && flow->nw_proto != IPPROTO_ICMPV6) {
1671 /* Add both ports at once. */
1672 hash = hash_add(hash,
1673 ((const uint32_t *)flow)[offsetof(struct flow, tp_src)
1674 / sizeof(uint32_t)]);
1677 return hash_finish(hash, 42); /* Arbitrary number. */
1680 /* Hashes 'flow' based on its L2 through L4 protocol information. */
1682 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
1687 struct in6_addr ipv6_addr;
1692 struct eth_addr eth_addr;
1698 memset(&fields, 0, sizeof fields);
1699 for (i = 0; i < ARRAY_SIZE(fields.eth_addr.be16); i++) {
1700 fields.eth_addr.be16[i] = flow->dl_src.be16[i] ^ flow->dl_dst.be16[i];
1702 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
1703 fields.eth_type = flow->dl_type;
1705 /* UDP source and destination port are not taken into account because they
1706 * will not necessarily be symmetric in a bidirectional flow. */
1707 if (fields.eth_type == htons(ETH_TYPE_IP)) {
1708 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
1709 fields.ip_proto = flow->nw_proto;
1710 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1711 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1713 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
1714 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
1715 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
1716 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
1718 for (i=0; i<16; i++) {
1719 ipv6_addr[i] = a[i] ^ b[i];
1721 fields.ip_proto = flow->nw_proto;
1722 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1723 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1726 return jhash_bytes(&fields, sizeof fields, basis);
1729 /* Hashes 'flow' based on its L3 through L4 protocol information */
1731 flow_hash_symmetric_l3l4(const struct flow *flow, uint32_t basis,
1734 uint32_t hash = basis;
1736 /* UDP source and destination port are also taken into account. */
1737 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1738 hash = hash_add(hash,
1739 (OVS_FORCE uint32_t) (flow->nw_src ^ flow->nw_dst));
1740 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1741 /* IPv6 addresses are 64-bit aligned inside struct flow. */
1742 const uint64_t *a = ALIGNED_CAST(uint64_t *, flow->ipv6_src.s6_addr);
1743 const uint64_t *b = ALIGNED_CAST(uint64_t *, flow->ipv6_dst.s6_addr);
1745 for (int i = 0; i < 4; i++) {
1746 hash = hash_add64(hash, a[i] ^ b[i]);
1749 /* Cannot hash non-IP flows */
1753 hash = hash_add(hash, flow->nw_proto);
1754 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP ||
1755 (inc_udp_ports && flow->nw_proto == IPPROTO_UDP)) {
1756 hash = hash_add(hash,
1757 (OVS_FORCE uint16_t) (flow->tp_src ^ flow->tp_dst));
1760 return hash_finish(hash, basis);
1763 /* Initialize a flow with random fields that matter for nx_hash_fields. */
1765 flow_random_hash_fields(struct flow *flow)
1767 uint16_t rnd = random_uint16();
1769 /* Initialize to all zeros. */
1770 memset(flow, 0, sizeof *flow);
1772 eth_addr_random(&flow->dl_src);
1773 eth_addr_random(&flow->dl_dst);
1775 flow->vlan_tci = (OVS_FORCE ovs_be16) (random_uint16() & VLAN_VID_MASK);
1777 /* Make most of the random flows IPv4, some IPv6, and rest random. */
1778 flow->dl_type = rnd < 0x8000 ? htons(ETH_TYPE_IP) :
1779 rnd < 0xc000 ? htons(ETH_TYPE_IPV6) : (OVS_FORCE ovs_be16)rnd;
1781 if (dl_type_is_ip_any(flow->dl_type)) {
1782 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1783 flow->nw_src = (OVS_FORCE ovs_be32)random_uint32();
1784 flow->nw_dst = (OVS_FORCE ovs_be32)random_uint32();
1786 random_bytes(&flow->ipv6_src, sizeof flow->ipv6_src);
1787 random_bytes(&flow->ipv6_dst, sizeof flow->ipv6_dst);
1789 /* Make most of IP flows TCP, some UDP or SCTP, and rest random. */
1790 rnd = random_uint16();
1791 flow->nw_proto = rnd < 0x8000 ? IPPROTO_TCP :
1792 rnd < 0xc000 ? IPPROTO_UDP :
1793 rnd < 0xd000 ? IPPROTO_SCTP : (uint8_t)rnd;
1794 if (flow->nw_proto == IPPROTO_TCP ||
1795 flow->nw_proto == IPPROTO_UDP ||
1796 flow->nw_proto == IPPROTO_SCTP) {
1797 flow->tp_src = (OVS_FORCE ovs_be16)random_uint16();
1798 flow->tp_dst = (OVS_FORCE ovs_be16)random_uint16();
1803 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
1805 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
1806 enum nx_hash_fields fields)
1809 case NX_HASH_FIELDS_ETH_SRC:
1810 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1813 case NX_HASH_FIELDS_SYMMETRIC_L4:
1814 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1815 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1816 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1817 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1818 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1819 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1820 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1821 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1823 if (is_ip_any(flow)) {
1824 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1825 flow_unwildcard_tp_ports(flow, wc);
1827 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1830 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1831 if (is_ip_any(flow) && flow->nw_proto == IPPROTO_UDP) {
1832 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1833 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1836 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1837 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1838 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1839 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1840 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1841 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1842 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1844 break; /* non-IP flow */
1847 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1848 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP) {
1849 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1850 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1859 /* Hashes the portions of 'flow' designated by 'fields'. */
1861 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
1866 case NX_HASH_FIELDS_ETH_SRC:
1867 return jhash_bytes(&flow->dl_src, sizeof flow->dl_src, basis);
1869 case NX_HASH_FIELDS_SYMMETRIC_L4:
1870 return flow_hash_symmetric_l4(flow, basis);
1872 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1873 return flow_hash_symmetric_l3l4(flow, basis, false);
1875 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1876 return flow_hash_symmetric_l3l4(flow, basis, true);
1883 /* Returns a string representation of 'fields'. */
1885 flow_hash_fields_to_str(enum nx_hash_fields fields)
1888 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
1889 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
1890 case NX_HASH_FIELDS_SYMMETRIC_L3L4: return "symmetric_l3l4";
1891 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP: return "symmetric_l3l4+udp";
1892 default: return "<unknown>";
1896 /* Returns true if the value of 'fields' is supported. Otherwise false. */
1898 flow_hash_fields_valid(enum nx_hash_fields fields)
1900 return fields == NX_HASH_FIELDS_ETH_SRC
1901 || fields == NX_HASH_FIELDS_SYMMETRIC_L4
1902 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4
1903 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP;
1906 /* Returns a hash value for the bits of 'flow' that are active based on
1907 * 'wc', given 'basis'. */
1909 flow_hash_in_wildcards(const struct flow *flow,
1910 const struct flow_wildcards *wc, uint32_t basis)
1912 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1913 const uint64_t *flow_u64 = (const uint64_t *) flow;
1918 for (i = 0; i < FLOW_U64S; i++) {
1919 hash = hash_add64(hash, flow_u64[i] & wc_u64[i]);
1921 return hash_finish(hash, 8 * FLOW_U64S);
1924 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1925 * OpenFlow 1.0 "dl_vlan" value:
1927 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
1928 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
1929 * 'flow' previously matched packets without a VLAN header).
1931 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
1932 * without a VLAN tag.
1934 * - Other values of 'vid' should not be used. */
1936 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
1938 if (vid == htons(OFP10_VLAN_NONE)) {
1939 flow->vlan_tci = htons(0);
1941 vid &= htons(VLAN_VID_MASK);
1942 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1943 flow->vlan_tci |= htons(VLAN_CFI) | vid;
1947 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1948 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
1951 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
1953 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
1954 flow->vlan_tci &= ~mask;
1955 flow->vlan_tci |= vid & mask;
1958 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
1961 * This function has no effect on the VLAN ID that 'flow' matches.
1963 * After calling this function, 'flow' will not match packets without a VLAN
1966 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
1969 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1970 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
1973 /* Returns the number of MPLS LSEs present in 'flow'
1975 * Returns 0 if the 'dl_type' of 'flow' is not an MPLS ethernet type.
1976 * Otherwise traverses 'flow''s MPLS label stack stopping at the
1977 * first entry that has the BoS bit set. If no such entry exists then
1978 * the maximum number of LSEs that can be stored in 'flow' is returned.
1981 flow_count_mpls_labels(const struct flow *flow, struct flow_wildcards *wc)
1983 /* dl_type is always masked. */
1984 if (eth_type_mpls(flow->dl_type)) {
1989 for (i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1991 wc->masks.mpls_lse[i] |= htonl(MPLS_BOS_MASK);
1993 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1996 if (flow->mpls_lse[i]) {
2006 /* Returns the number consecutive of MPLS LSEs, starting at the
2007 * innermost LSE, that are common in 'a' and 'b'.
2009 * 'an' must be flow_count_mpls_labels(a).
2010 * 'bn' must be flow_count_mpls_labels(b).
2013 flow_count_common_mpls_labels(const struct flow *a, int an,
2014 const struct flow *b, int bn,
2015 struct flow_wildcards *wc)
2017 int min_n = MIN(an, bn);
2022 int a_last = an - 1;
2023 int b_last = bn - 1;
2026 for (i = 0; i < min_n; i++) {
2028 wc->masks.mpls_lse[a_last - i] = OVS_BE32_MAX;
2029 wc->masks.mpls_lse[b_last - i] = OVS_BE32_MAX;
2031 if (a->mpls_lse[a_last - i] != b->mpls_lse[b_last - i]) {
2042 /* Adds a new outermost MPLS label to 'flow' and changes 'flow''s Ethernet type
2043 * to 'mpls_eth_type', which must be an MPLS Ethertype.
2045 * If the new label is the first MPLS label in 'flow', it is generated as;
2047 * - label: 2, if 'flow' is IPv6, otherwise 0.
2049 * - TTL: IPv4 or IPv6 TTL, if present and nonzero, otherwise 64.
2051 * - TC: IPv4 or IPv6 TOS, if present, otherwise 0.
2055 * If the new label is the second or later label MPLS label in 'flow', it is
2058 * - label: Copied from outer label.
2060 * - TTL: Copied from outer label.
2062 * - TC: Copied from outer label.
2066 * 'n' must be flow_count_mpls_labels(flow). 'n' must be less than
2067 * FLOW_MAX_MPLS_LABELS (because otherwise flow->mpls_lse[] would overflow).
2070 flow_push_mpls(struct flow *flow, int n, ovs_be16 mpls_eth_type,
2071 struct flow_wildcards *wc)
2073 ovs_assert(eth_type_mpls(mpls_eth_type));
2074 ovs_assert(n < FLOW_MAX_MPLS_LABELS);
2080 memset(&wc->masks.mpls_lse, 0xff, sizeof *wc->masks.mpls_lse * n);
2082 for (i = n; i >= 1; i--) {
2083 flow->mpls_lse[i] = flow->mpls_lse[i - 1];
2085 flow->mpls_lse[0] = (flow->mpls_lse[1] & htonl(~MPLS_BOS_MASK));
2087 int label = 0; /* IPv4 Explicit Null. */
2091 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2095 if (is_ip_any(flow)) {
2096 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
2098 wc->masks.nw_tos |= IP_DSCP_MASK;
2099 wc->masks.nw_ttl = 0xff;
2107 flow->mpls_lse[0] = set_mpls_lse_values(ttl, tc, 1, htonl(label));
2109 /* Clear all L3 and L4 fields and dp_hash. */
2110 BUILD_ASSERT(FLOW_WC_SEQ == 36);
2111 memset((char *) flow + FLOW_SEGMENT_2_ENDS_AT, 0,
2112 sizeof(struct flow) - FLOW_SEGMENT_2_ENDS_AT);
2115 flow->dl_type = mpls_eth_type;
2118 /* Tries to remove the outermost MPLS label from 'flow'. Returns true if
2119 * successful, false otherwise. On success, sets 'flow''s Ethernet type to
2122 * 'n' must be flow_count_mpls_labels(flow). */
2124 flow_pop_mpls(struct flow *flow, int n, ovs_be16 eth_type,
2125 struct flow_wildcards *wc)
2130 /* Nothing to pop. */
2132 } else if (n == FLOW_MAX_MPLS_LABELS) {
2134 wc->masks.mpls_lse[n - 1] |= htonl(MPLS_BOS_MASK);
2136 if (!(flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK))) {
2137 /* Can't pop because don't know what to fill in mpls_lse[n - 1]. */
2143 memset(&wc->masks.mpls_lse[1], 0xff,
2144 sizeof *wc->masks.mpls_lse * (n - 1));
2146 for (i = 1; i < n; i++) {
2147 flow->mpls_lse[i - 1] = flow->mpls_lse[i];
2149 flow->mpls_lse[n - 1] = 0;
2150 flow->dl_type = eth_type;
2154 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
2155 * as an OpenFlow 1.1 "mpls_label" value. */
2157 flow_set_mpls_label(struct flow *flow, int idx, ovs_be32 label)
2159 set_mpls_lse_label(&flow->mpls_lse[idx], label);
2162 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
2165 flow_set_mpls_ttl(struct flow *flow, int idx, uint8_t ttl)
2167 set_mpls_lse_ttl(&flow->mpls_lse[idx], ttl);
2170 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
2173 flow_set_mpls_tc(struct flow *flow, int idx, uint8_t tc)
2175 set_mpls_lse_tc(&flow->mpls_lse[idx], tc);
2178 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
2180 flow_set_mpls_bos(struct flow *flow, int idx, uint8_t bos)
2182 set_mpls_lse_bos(&flow->mpls_lse[idx], bos);
2185 /* Sets the entire MPLS LSE. */
2187 flow_set_mpls_lse(struct flow *flow, int idx, ovs_be32 lse)
2189 flow->mpls_lse[idx] = lse;
2193 flow_compose_l4(struct dp_packet *p, const struct flow *flow)
2197 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
2198 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
2199 if (flow->nw_proto == IPPROTO_TCP) {
2200 struct tcp_header *tcp;
2202 l4_len = sizeof *tcp;
2203 tcp = dp_packet_put_zeros(p, l4_len);
2204 tcp->tcp_src = flow->tp_src;
2205 tcp->tcp_dst = flow->tp_dst;
2206 tcp->tcp_ctl = TCP_CTL(ntohs(flow->tcp_flags), 5);
2207 } else if (flow->nw_proto == IPPROTO_UDP) {
2208 struct udp_header *udp;
2210 l4_len = sizeof *udp;
2211 udp = dp_packet_put_zeros(p, l4_len);
2212 udp->udp_src = flow->tp_src;
2213 udp->udp_dst = flow->tp_dst;
2214 } else if (flow->nw_proto == IPPROTO_SCTP) {
2215 struct sctp_header *sctp;
2217 l4_len = sizeof *sctp;
2218 sctp = dp_packet_put_zeros(p, l4_len);
2219 sctp->sctp_src = flow->tp_src;
2220 sctp->sctp_dst = flow->tp_dst;
2221 } else if (flow->nw_proto == IPPROTO_ICMP) {
2222 struct icmp_header *icmp;
2224 l4_len = sizeof *icmp;
2225 icmp = dp_packet_put_zeros(p, l4_len);
2226 icmp->icmp_type = ntohs(flow->tp_src);
2227 icmp->icmp_code = ntohs(flow->tp_dst);
2228 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
2229 } else if (flow->nw_proto == IPPROTO_IGMP) {
2230 struct igmp_header *igmp;
2232 l4_len = sizeof *igmp;
2233 igmp = dp_packet_put_zeros(p, l4_len);
2234 igmp->igmp_type = ntohs(flow->tp_src);
2235 igmp->igmp_code = ntohs(flow->tp_dst);
2236 put_16aligned_be32(&igmp->group, flow->igmp_group_ip4);
2237 igmp->igmp_csum = csum(igmp, IGMP_HEADER_LEN);
2238 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
2239 struct icmp6_hdr *icmp;
2241 l4_len = sizeof *icmp;
2242 icmp = dp_packet_put_zeros(p, l4_len);
2243 icmp->icmp6_type = ntohs(flow->tp_src);
2244 icmp->icmp6_code = ntohs(flow->tp_dst);
2246 if (icmp->icmp6_code == 0 &&
2247 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
2248 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
2249 struct in6_addr *nd_target;
2250 struct ovs_nd_opt *nd_opt;
2252 l4_len += sizeof *nd_target;
2253 nd_target = dp_packet_put_zeros(p, sizeof *nd_target);
2254 *nd_target = flow->nd_target;
2256 if (!eth_addr_is_zero(flow->arp_sha)) {
2258 nd_opt = dp_packet_put_zeros(p, 8);
2259 nd_opt->nd_opt_len = 1;
2260 nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
2261 nd_opt->nd_opt_mac = flow->arp_sha;
2263 if (!eth_addr_is_zero(flow->arp_tha)) {
2265 nd_opt = dp_packet_put_zeros(p, 8);
2266 nd_opt->nd_opt_len = 1;
2267 nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR;
2268 nd_opt->nd_opt_mac = flow->arp_tha;
2271 icmp->icmp6_cksum = (OVS_FORCE uint16_t)
2272 csum(icmp, (char *)dp_packet_tail(p) - (char *)icmp);
2278 /* Puts into 'b' a packet that flow_extract() would parse as having the given
2281 * (This is useful only for testing, obviously, and the packet isn't really
2282 * valid. It hasn't got some checksums filled in, for one, and lots of fields
2283 * are just zeroed.) */
2285 flow_compose(struct dp_packet *p, const struct flow *flow)
2289 /* eth_compose() sets l3 pointer and makes sure it is 32-bit aligned. */
2290 eth_compose(p, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
2291 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
2292 struct eth_header *eth = dp_packet_l2(p);
2293 eth->eth_type = htons(dp_packet_size(p));
2297 if (flow->vlan_tci & htons(VLAN_CFI)) {
2298 eth_push_vlan(p, htons(ETH_TYPE_VLAN), flow->vlan_tci);
2301 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2302 struct ip_header *ip;
2304 ip = dp_packet_put_zeros(p, sizeof *ip);
2305 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
2306 ip->ip_tos = flow->nw_tos;
2307 ip->ip_ttl = flow->nw_ttl;
2308 ip->ip_proto = flow->nw_proto;
2309 put_16aligned_be32(&ip->ip_src, flow->nw_src);
2310 put_16aligned_be32(&ip->ip_dst, flow->nw_dst);
2312 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2313 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
2314 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
2315 ip->ip_frag_off |= htons(100);
2319 dp_packet_set_l4(p, dp_packet_tail(p));
2321 l4_len = flow_compose_l4(p, flow);
2323 ip = dp_packet_l3(p);
2324 ip->ip_tot_len = htons(p->l4_ofs - p->l3_ofs + l4_len);
2325 ip->ip_csum = csum(ip, sizeof *ip);
2326 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2327 struct ovs_16aligned_ip6_hdr *nh;
2329 nh = dp_packet_put_zeros(p, sizeof *nh);
2330 put_16aligned_be32(&nh->ip6_flow, htonl(6 << 28) |
2331 htonl(flow->nw_tos << 20) | flow->ipv6_label);
2332 nh->ip6_hlim = flow->nw_ttl;
2333 nh->ip6_nxt = flow->nw_proto;
2335 memcpy(&nh->ip6_src, &flow->ipv6_src, sizeof(nh->ip6_src));
2336 memcpy(&nh->ip6_dst, &flow->ipv6_dst, sizeof(nh->ip6_dst));
2338 dp_packet_set_l4(p, dp_packet_tail(p));
2340 l4_len = flow_compose_l4(p, flow);
2342 nh = dp_packet_l3(p);
2343 nh->ip6_plen = htons(l4_len);
2344 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
2345 flow->dl_type == htons(ETH_TYPE_RARP)) {
2346 struct arp_eth_header *arp;
2348 arp = dp_packet_put_zeros(p, sizeof *arp);
2349 dp_packet_set_l3(p, arp);
2350 arp->ar_hrd = htons(1);
2351 arp->ar_pro = htons(ETH_TYPE_IP);
2352 arp->ar_hln = ETH_ADDR_LEN;
2354 arp->ar_op = htons(flow->nw_proto);
2356 if (flow->nw_proto == ARP_OP_REQUEST ||
2357 flow->nw_proto == ARP_OP_REPLY) {
2358 put_16aligned_be32(&arp->ar_spa, flow->nw_src);
2359 put_16aligned_be32(&arp->ar_tpa, flow->nw_dst);
2360 arp->ar_sha = flow->arp_sha;
2361 arp->ar_tha = flow->arp_tha;
2365 if (eth_type_mpls(flow->dl_type)) {
2368 p->l2_5_ofs = p->l3_ofs;
2369 for (n = 1; n < FLOW_MAX_MPLS_LABELS; n++) {
2370 if (flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK)) {
2375 push_mpls(p, flow->dl_type, flow->mpls_lse[--n]);
2380 /* Compressed flow. */
2382 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
2383 * the caller. The caller must have already computed 'dst->map' properly to
2384 * indicate the significant uint64_t elements of 'src'.
2386 * Normally the significant elements are the ones that are non-zero. However,
2387 * when a miniflow is initialized from a (mini)mask, the values can be zeroes,
2388 * so that the flow and mask always have the same maps. */
2390 miniflow_init(struct miniflow *dst, const struct flow *src)
2392 uint64_t *dst_u64 = miniflow_values(dst);
2395 FLOWMAP_FOR_EACH_INDEX(idx, dst->map) {
2396 *dst_u64++ = flow_u64_value(src, idx);
2400 /* Initialize the maps of 'flow' from 'src'. */
2402 miniflow_map_init(struct miniflow *flow, const struct flow *src)
2404 /* Initialize map, counting the number of nonzero elements. */
2405 flowmap_init(&flow->map);
2406 for (size_t i = 0; i < FLOW_U64S; i++) {
2407 if (flow_u64_value(src, i)) {
2408 flowmap_set(&flow->map, i, 1);
2413 /* Allocates 'n' count of miniflows, consecutive in memory, initializing the
2414 * map of each from 'src'.
2415 * Returns the size of the miniflow data. */
2417 miniflow_alloc(struct miniflow *dsts[], size_t n, const struct miniflow *src)
2419 size_t n_values = miniflow_n_values(src);
2420 size_t data_size = MINIFLOW_VALUES_SIZE(n_values);
2421 struct miniflow *dst = xmalloc(n * (sizeof *src + data_size));
2424 COVERAGE_INC(miniflow_malloc);
2426 for (i = 0; i < n; i++) {
2427 *dst = *src; /* Copy maps. */
2429 dst += 1; /* Just past the maps. */
2430 dst = (struct miniflow *)((uint64_t *)dst + n_values); /* Skip data. */
2435 /* Returns a miniflow copy of 'src'. The caller must eventually free() the
2436 * returned miniflow. */
2438 miniflow_create(const struct flow *src)
2440 struct miniflow tmp;
2441 struct miniflow *dst;
2443 miniflow_map_init(&tmp, src);
2445 miniflow_alloc(&dst, 1, &tmp);
2446 miniflow_init(dst, src);
2450 /* Initializes 'dst' as a copy of 'src'. The caller must have allocated
2451 * 'dst' to have inline space for 'n_values' data in 'src'. */
2453 miniflow_clone(struct miniflow *dst, const struct miniflow *src,
2456 *dst = *src; /* Copy maps. */
2457 memcpy(miniflow_values(dst), miniflow_get_values(src),
2458 MINIFLOW_VALUES_SIZE(n_values));
2461 /* Initializes 'dst' as a copy of 'src'. */
2463 miniflow_expand(const struct miniflow *src, struct flow *dst)
2465 memset(dst, 0, sizeof *dst);
2466 flow_union_with_miniflow(dst, src);
2469 /* Returns true if 'a' and 'b' are equal miniflows, false otherwise. */
2471 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
2473 const uint64_t *ap = miniflow_get_values(a);
2474 const uint64_t *bp = miniflow_get_values(b);
2476 /* This is mostly called after a matching hash, so it is highly likely that
2477 * the maps are equal as well. */
2478 if (OVS_LIKELY(flowmap_equal(a->map, b->map))) {
2479 return !memcmp(ap, bp, miniflow_n_values(a) * sizeof *ap);
2483 FLOWMAP_FOR_EACH_INDEX (idx, flowmap_or(a->map, b->map)) {
2484 if ((flowmap_is_set(&a->map, idx) ? *ap++ : 0)
2485 != (flowmap_is_set(&b->map, idx) ? *bp++ : 0)) {
2494 /* Returns false if 'a' and 'b' differ at the places where there are 1-bits
2495 * in 'mask', true otherwise. */
2497 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
2498 const struct minimask *mask)
2500 const uint64_t *p = miniflow_get_values(&mask->masks);
2503 FLOWMAP_FOR_EACH_INDEX(idx, mask->masks.map) {
2504 if ((miniflow_get(a, idx) ^ miniflow_get(b, idx)) & *p++) {
2512 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
2513 * in 'mask', false if they differ. */
2515 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
2516 const struct minimask *mask)
2518 const uint64_t *p = miniflow_get_values(&mask->masks);
2521 FLOWMAP_FOR_EACH_INDEX(idx, mask->masks.map) {
2522 if ((miniflow_get(a, idx) ^ flow_u64_value(b, idx)) & *p++) {
2532 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
2534 miniflow_init(&mask->masks, &wc->masks);
2537 /* Returns a minimask copy of 'wc'. The caller must eventually free the
2538 * returned minimask with free(). */
2540 minimask_create(const struct flow_wildcards *wc)
2542 return (struct minimask *)miniflow_create(&wc->masks);
2545 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
2547 * The caller must provide room for FLOW_U64S "uint64_t"s in 'storage', which
2548 * must follow '*dst_' in memory, for use by 'dst_'. The caller must *not*
2549 * free 'dst_' free(). */
2551 minimask_combine(struct minimask *dst_,
2552 const struct minimask *a_, const struct minimask *b_,
2553 uint64_t storage[FLOW_U64S])
2555 struct miniflow *dst = &dst_->masks;
2556 uint64_t *dst_values = storage;
2557 const struct miniflow *a = &a_->masks;
2558 const struct miniflow *b = &b_->masks;
2561 flowmap_init(&dst->map);
2563 FLOWMAP_FOR_EACH_INDEX(idx, flowmap_and(a->map, b->map)) {
2564 /* Both 'a' and 'b' have non-zero data at 'idx'. */
2565 uint64_t mask = *miniflow_get__(a, idx) & *miniflow_get__(b, idx);
2568 flowmap_set(&dst->map, idx, 1);
2569 *dst_values++ = mask;
2574 /* Initializes 'wc' as a copy of 'mask'. */
2576 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
2578 miniflow_expand(&mask->masks, &wc->masks);
2581 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise.
2582 * Minimasks may not have zero data values, so for the minimasks to be the
2583 * same, they need to have the same map and the same data values. */
2585 minimask_equal(const struct minimask *a, const struct minimask *b)
2587 return !memcmp(a, b, sizeof *a
2588 + MINIFLOW_VALUES_SIZE(miniflow_n_values(&a->masks)));
2591 /* Returns true if at least one bit matched by 'b' is wildcarded by 'a',
2592 * false otherwise. */
2594 minimask_has_extra(const struct minimask *a, const struct minimask *b)
2596 const uint64_t *bp = miniflow_get_values(&b->masks);
2599 FLOWMAP_FOR_EACH_INDEX(idx, b->masks.map) {
2600 uint64_t b_u64 = *bp++;
2602 /* 'b_u64' is non-zero, check if the data in 'a' is either zero
2603 * or misses some of the bits in 'b_u64'. */
2604 if (!MINIFLOW_IN_MAP(&a->masks, idx)
2605 || ((*miniflow_get__(&a->masks, idx) & b_u64) != b_u64)) {
2606 return true; /* 'a' wildcards some bits 'b' doesn't. */