2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 #include <sys/types.h>
22 #include <netinet/in.h>
23 #include <netinet/icmp6.h>
24 #include <netinet/ip6.h>
28 #include "byte-order.h"
31 #include "dynamic-string.h"
35 #include "dp-packet.h"
36 #include "openflow/openflow.h"
40 #include "unaligned.h"
42 COVERAGE_DEFINE(flow_extract);
43 COVERAGE_DEFINE(miniflow_malloc);
45 /* U64 indices for segmented flow classification. */
46 const uint8_t flow_segment_u64s[4] = {
47 FLOW_SEGMENT_1_ENDS_AT / sizeof(uint64_t),
48 FLOW_SEGMENT_2_ENDS_AT / sizeof(uint64_t),
49 FLOW_SEGMENT_3_ENDS_AT / sizeof(uint64_t),
53 /* Asserts that field 'f1' follows immediately after 'f0' in struct flow,
54 * without any intervening padding. */
55 #define ASSERT_SEQUENTIAL(f0, f1) \
56 BUILD_ASSERT_DECL(offsetof(struct flow, f0) \
57 + MEMBER_SIZEOF(struct flow, f0) \
58 == offsetof(struct flow, f1))
60 /* Asserts that fields 'f0' and 'f1' are in the same 32-bit aligned word within
62 #define ASSERT_SAME_WORD(f0, f1) \
63 BUILD_ASSERT_DECL(offsetof(struct flow, f0) / 4 \
64 == offsetof(struct flow, f1) / 4)
66 /* Asserts that 'f0' and 'f1' are both sequential and within the same 32-bit
67 * aligned word in struct flow. */
68 #define ASSERT_SEQUENTIAL_SAME_WORD(f0, f1) \
69 ASSERT_SEQUENTIAL(f0, f1); \
70 ASSERT_SAME_WORD(f0, f1)
72 /* miniflow_extract() assumes the following to be true to optimize the
73 * extraction process. */
74 ASSERT_SEQUENTIAL_SAME_WORD(dl_type, vlan_tci);
76 ASSERT_SEQUENTIAL_SAME_WORD(nw_frag, nw_tos);
77 ASSERT_SEQUENTIAL_SAME_WORD(nw_tos, nw_ttl);
78 ASSERT_SEQUENTIAL_SAME_WORD(nw_ttl, nw_proto);
80 /* TCP flags in the middle of a BE64, zeroes in the other half. */
81 BUILD_ASSERT_DECL(offsetof(struct flow, tcp_flags) % 8 == 4);
84 #define TCP_FLAGS_BE32(tcp_ctl) ((OVS_FORCE ovs_be32)TCP_FLAGS_BE16(tcp_ctl) \
87 #define TCP_FLAGS_BE32(tcp_ctl) ((OVS_FORCE ovs_be32)TCP_FLAGS_BE16(tcp_ctl))
90 ASSERT_SEQUENTIAL_SAME_WORD(tp_src, tp_dst);
92 /* Removes 'size' bytes from the head end of '*datap', of size '*sizep', which
93 * must contain at least 'size' bytes of data. Returns the first byte of data
95 static inline const void *
96 data_pull(const void **datap, size_t *sizep, size_t size)
98 const char *data = *datap;
104 /* If '*datap' has at least 'size' bytes of data, removes that many bytes from
105 * the head end of '*datap' and returns the first byte removed. Otherwise,
106 * returns a null pointer without modifying '*datap'. */
107 static inline const void *
108 data_try_pull(const void **datap, size_t *sizep, size_t size)
110 return OVS_LIKELY(*sizep >= size) ? data_pull(datap, sizep, size) : NULL;
113 /* Context for pushing data to a miniflow. */
117 uint64_t * const end;
120 /* miniflow_push_* macros allow filling in a miniflow data values in order.
121 * Assertions are needed only when the layout of the struct flow is modified.
122 * 'ofs' is a compile-time constant, which allows most of the code be optimized
123 * away. Some GCC versions gave warnings on ALWAYS_INLINE, so these are
124 * defined as macros. */
126 #if (FLOW_WC_SEQ != 35)
127 #define MINIFLOW_ASSERT(X) ovs_assert(X)
128 BUILD_MESSAGE("FLOW_WC_SEQ changed: miniflow_extract() will have runtime "
129 "assertions enabled. Consider updating FLOW_WC_SEQ after "
132 #define MINIFLOW_ASSERT(X)
135 /* True if 'IDX' and higher bits are not set. */
136 #define ASSERT_FLOWMAP_NOT_SET(FM, IDX) \
138 MINIFLOW_ASSERT(!((FM)->bits[(IDX) / MAP_T_BITS] & \
139 (MAP_MAX << ((IDX) % MAP_T_BITS)))); \
140 for (size_t i = (IDX) / MAP_T_BITS + 1; i < FLOWMAP_UNITS; i++) { \
141 MINIFLOW_ASSERT(!(FM)->bits[i]); \
145 #define miniflow_set_map(MF, OFS) \
147 ASSERT_FLOWMAP_NOT_SET(&MF.map, (OFS)); \
148 flowmap_set(&MF.map, (OFS), 1); \
151 #define miniflow_assert_in_map(MF, OFS) \
152 MINIFLOW_ASSERT(flowmap_is_set(&MF.map, (OFS))); \
153 ASSERT_FLOWMAP_NOT_SET(&MF.map, (OFS) + 1)
155 #define miniflow_push_uint64_(MF, OFS, VALUE) \
157 MINIFLOW_ASSERT(MF.data < MF.end && (OFS) % 8 == 0); \
158 *MF.data++ = VALUE; \
159 miniflow_set_map(MF, OFS / 8); \
162 #define miniflow_push_be64_(MF, OFS, VALUE) \
163 miniflow_push_uint64_(MF, OFS, (OVS_FORCE uint64_t)(VALUE))
165 #define miniflow_push_uint32_(MF, OFS, VALUE) \
167 MINIFLOW_ASSERT(MF.data < MF.end); \
169 if ((OFS) % 8 == 0) { \
170 miniflow_set_map(MF, OFS / 8); \
171 *(uint32_t *)MF.data = VALUE; \
172 } else if ((OFS) % 8 == 4) { \
173 miniflow_assert_in_map(MF, OFS / 8); \
174 *((uint32_t *)MF.data + 1) = VALUE; \
179 #define miniflow_push_be32_(MF, OFS, VALUE) \
180 miniflow_push_uint32_(MF, OFS, (OVS_FORCE uint32_t)(VALUE))
182 #define miniflow_push_uint16_(MF, OFS, VALUE) \
184 MINIFLOW_ASSERT(MF.data < MF.end); \
186 if ((OFS) % 8 == 0) { \
187 miniflow_set_map(MF, OFS / 8); \
188 *(uint16_t *)MF.data = VALUE; \
189 } else if ((OFS) % 8 == 2) { \
190 miniflow_assert_in_map(MF, OFS / 8); \
191 *((uint16_t *)MF.data + 1) = VALUE; \
192 } else if ((OFS) % 8 == 4) { \
193 miniflow_assert_in_map(MF, OFS / 8); \
194 *((uint16_t *)MF.data + 2) = VALUE; \
195 } else if ((OFS) % 8 == 6) { \
196 miniflow_assert_in_map(MF, OFS / 8); \
197 *((uint16_t *)MF.data + 3) = VALUE; \
202 #define miniflow_push_uint8_(MF, OFS, VALUE) \
204 MINIFLOW_ASSERT(MF.data < MF.end); \
206 if ((OFS) % 8 == 0) { \
207 miniflow_set_map(MF, OFS / 8); \
208 *(uint8_t *)MF.data = VALUE; \
209 } else if ((OFS) % 8 == 7) { \
210 miniflow_assert_in_map(MF, OFS / 8); \
211 *((uint8_t *)MF.data + 7) = VALUE; \
214 miniflow_assert_in_map(MF, OFS / 8); \
215 *((uint8_t *)MF.data + ((OFS) % 8)) = VALUE; \
219 #define miniflow_pad_to_64_(MF, OFS) \
221 MINIFLOW_ASSERT((OFS) % 8 != 0); \
222 miniflow_assert_in_map(MF, OFS / 8); \
224 memset((uint8_t *)MF.data + (OFS) % 8, 0, 8 - (OFS) % 8); \
228 #define miniflow_push_be16_(MF, OFS, VALUE) \
229 miniflow_push_uint16_(MF, OFS, (OVS_FORCE uint16_t)VALUE);
231 #define miniflow_push_be8_(MF, OFS, VALUE) \
232 miniflow_push_uint8_(MF, OFS, (OVS_FORCE uint8_t)VALUE);
234 #define miniflow_set_maps(MF, OFS, N_WORDS) \
236 size_t ofs = (OFS); \
237 size_t n_words = (N_WORDS); \
239 MINIFLOW_ASSERT(n_words && MF.data + n_words <= MF.end); \
240 ASSERT_FLOWMAP_NOT_SET(&MF.map, ofs); \
241 flowmap_set(&MF.map, ofs, n_words); \
244 /* Data at 'valuep' may be unaligned. */
245 #define miniflow_push_words_(MF, OFS, VALUEP, N_WORDS) \
247 MINIFLOW_ASSERT((OFS) % 8 == 0); \
248 miniflow_set_maps(MF, (OFS) / 8, (N_WORDS)); \
249 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof *MF.data); \
250 MF.data += (N_WORDS); \
253 /* Push 32-bit words padded to 64-bits. */
254 #define miniflow_push_words_32_(MF, OFS, VALUEP, N_WORDS) \
256 miniflow_set_maps(MF, (OFS) / 8, DIV_ROUND_UP(N_WORDS, 2)); \
257 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof(uint32_t)); \
258 MF.data += DIV_ROUND_UP(N_WORDS, 2); \
259 if ((N_WORDS) & 1) { \
260 *((uint32_t *)MF.data - 1) = 0; \
264 /* Data at 'valuep' may be unaligned. */
265 /* MACs start 64-aligned, and must be followed by other data or padding. */
266 #define miniflow_push_macs_(MF, OFS, VALUEP) \
268 miniflow_set_maps(MF, (OFS) / 8, 2); \
269 memcpy(MF.data, (VALUEP), 2 * ETH_ADDR_LEN); \
270 MF.data += 1; /* First word only. */ \
273 #define miniflow_push_uint32(MF, FIELD, VALUE) \
274 miniflow_push_uint32_(MF, offsetof(struct flow, FIELD), VALUE)
276 #define miniflow_push_be32(MF, FIELD, VALUE) \
277 miniflow_push_be32_(MF, offsetof(struct flow, FIELD), VALUE)
279 #define miniflow_push_uint16(MF, FIELD, VALUE) \
280 miniflow_push_uint16_(MF, offsetof(struct flow, FIELD), VALUE)
282 #define miniflow_push_be16(MF, FIELD, VALUE) \
283 miniflow_push_be16_(MF, offsetof(struct flow, FIELD), VALUE)
285 #define miniflow_push_uint8(MF, FIELD, VALUE) \
286 miniflow_push_uint8_(MF, offsetof(struct flow, FIELD), VALUE)
288 #define miniflow_pad_to_64(MF, FIELD) \
289 miniflow_pad_to_64_(MF, OFFSETOFEND(struct flow, FIELD))
291 #define miniflow_push_words(MF, FIELD, VALUEP, N_WORDS) \
292 miniflow_push_words_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
294 #define miniflow_push_words_32(MF, FIELD, VALUEP, N_WORDS) \
295 miniflow_push_words_32_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
297 #define miniflow_push_macs(MF, FIELD, VALUEP) \
298 miniflow_push_macs_(MF, offsetof(struct flow, FIELD), VALUEP)
300 /* Pulls the MPLS headers at '*datap' and returns the count of them. */
302 parse_mpls(const void **datap, size_t *sizep)
304 const struct mpls_hdr *mh;
307 while ((mh = data_try_pull(datap, sizep, sizeof *mh))) {
309 if (mh->mpls_lse.lo & htons(1 << MPLS_BOS_SHIFT)) {
313 return MIN(count, FLOW_MAX_MPLS_LABELS);
316 static inline ovs_be16
317 parse_vlan(const void **datap, size_t *sizep)
319 const struct eth_header *eth = *datap;
322 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
326 data_pull(datap, sizep, ETH_ADDR_LEN * 2);
328 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
329 if (OVS_LIKELY(*sizep
330 >= sizeof(struct qtag_prefix) + sizeof(ovs_be16))) {
331 const struct qtag_prefix *qp = data_pull(datap, sizep, sizeof *qp);
332 return qp->tci | htons(VLAN_CFI);
338 static inline ovs_be16
339 parse_ethertype(const void **datap, size_t *sizep)
341 const struct llc_snap_header *llc;
344 proto = *(ovs_be16 *) data_pull(datap, sizep, sizeof proto);
345 if (OVS_LIKELY(ntohs(proto) >= ETH_TYPE_MIN)) {
349 if (OVS_UNLIKELY(*sizep < sizeof *llc)) {
350 return htons(FLOW_DL_TYPE_NONE);
354 if (OVS_UNLIKELY(llc->llc.llc_dsap != LLC_DSAP_SNAP
355 || llc->llc.llc_ssap != LLC_SSAP_SNAP
356 || llc->llc.llc_cntl != LLC_CNTL_SNAP
357 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
358 sizeof llc->snap.snap_org))) {
359 return htons(FLOW_DL_TYPE_NONE);
362 data_pull(datap, sizep, sizeof *llc);
364 if (OVS_LIKELY(ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN)) {
365 return llc->snap.snap_type;
368 return htons(FLOW_DL_TYPE_NONE);
372 parse_icmpv6(const void **datap, size_t *sizep, const struct icmp6_hdr *icmp,
373 const struct in6_addr **nd_target,
374 struct eth_addr arp_buf[2])
376 if (icmp->icmp6_code == 0 &&
377 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
378 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
380 *nd_target = data_try_pull(datap, sizep, sizeof **nd_target);
381 if (OVS_UNLIKELY(!*nd_target)) {
385 while (*sizep >= 8) {
386 /* The minimum size of an option is 8 bytes, which also is
387 * the size of Ethernet link-layer options. */
388 const struct ovs_nd_opt *nd_opt = *datap;
389 int opt_len = nd_opt->nd_opt_len * ND_OPT_LEN;
391 if (!opt_len || opt_len > *sizep) {
395 /* Store the link layer address if the appropriate option is
396 * provided. It is considered an error if the same link
397 * layer option is specified twice. */
398 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
400 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[0]))) {
401 arp_buf[0] = nd_opt->nd_opt_mac;
405 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
407 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[1]))) {
408 arp_buf[1] = nd_opt->nd_opt_mac;
414 if (OVS_UNLIKELY(!data_try_pull(datap, sizep, opt_len))) {
424 arp_buf[0] = eth_addr_zero;
425 arp_buf[1] = eth_addr_zero;
428 /* Initializes 'flow' members from 'packet' and 'md'
430 * Initializes 'packet' header l2 pointer to the start of the Ethernet
431 * header, and the layer offsets as follows:
433 * - packet->l2_5_ofs to the start of the MPLS shim header, or UINT16_MAX
434 * when there is no MPLS shim header.
436 * - packet->l3_ofs to just past the Ethernet header, or just past the
437 * vlan_header if one is present, to the first byte of the payload of the
438 * Ethernet frame. UINT16_MAX if the frame is too short to contain an
441 * - packet->l4_ofs to just past the IPv4 header, if one is present and
442 * has at least the content used for the fields of interest for the flow,
443 * otherwise UINT16_MAX.
446 flow_extract(struct dp_packet *packet, struct flow *flow)
450 uint64_t buf[FLOW_U64S];
453 COVERAGE_INC(flow_extract);
455 miniflow_extract(packet, &m.mf);
456 miniflow_expand(&m.mf, flow);
459 /* Caller is responsible for initializing 'dst' with enough storage for
460 * FLOW_U64S * 8 bytes. */
462 miniflow_extract(struct dp_packet *packet, struct miniflow *dst)
464 const struct pkt_metadata *md = &packet->md;
465 const void *data = dp_packet_data(packet);
466 size_t size = dp_packet_size(packet);
467 uint64_t *values = miniflow_values(dst);
468 struct mf_ctx mf = { FLOWMAP_EMPTY_INITIALIZER, values,
469 values + FLOW_U64S };
472 uint8_t nw_frag, nw_tos, nw_ttl, nw_proto;
475 if (flow_tnl_dst_is_set(&md->tunnel)) {
476 miniflow_push_words(mf, tunnel, &md->tunnel,
477 offsetof(struct flow_tnl, metadata) /
480 if (!(md->tunnel.flags & FLOW_TNL_F_UDPIF)) {
481 if (md->tunnel.metadata.present.map) {
482 miniflow_push_words(mf, tunnel.metadata, &md->tunnel.metadata,
483 sizeof md->tunnel.metadata /
487 if (md->tunnel.metadata.present.len) {
488 miniflow_push_words(mf, tunnel.metadata.present,
489 &md->tunnel.metadata.present, 1);
490 miniflow_push_words(mf, tunnel.metadata.opts.gnv,
491 md->tunnel.metadata.opts.gnv,
492 DIV_ROUND_UP(md->tunnel.metadata.present.len,
497 if (md->skb_priority || md->pkt_mark) {
498 miniflow_push_uint32(mf, skb_priority, md->skb_priority);
499 miniflow_push_uint32(mf, pkt_mark, md->pkt_mark);
501 miniflow_push_uint32(mf, dp_hash, md->dp_hash);
502 miniflow_push_uint32(mf, in_port, odp_to_u32(md->in_port.odp_port));
503 if (md->recirc_id || md->ct_state) {
504 miniflow_push_uint32(mf, recirc_id, md->recirc_id);
505 miniflow_push_uint16(mf, ct_state, md->ct_state);
506 miniflow_push_uint16(mf, ct_zone, md->ct_zone);
510 miniflow_push_uint32(mf, ct_mark, md->ct_mark);
511 miniflow_pad_to_64(mf, ct_mark);
513 if (!ovs_u128_is_zero(&md->ct_label)) {
514 miniflow_push_words(mf, ct_label, &md->ct_label,
515 sizeof md->ct_label / sizeof(uint64_t));
519 /* Initialize packet's layer pointer and offsets. */
521 dp_packet_reset_offsets(packet);
523 /* Must have full Ethernet header to proceed. */
524 if (OVS_UNLIKELY(size < sizeof(struct eth_header))) {
530 ASSERT_SEQUENTIAL(dl_dst, dl_src);
531 miniflow_push_macs(mf, dl_dst, data);
532 /* dl_type, vlan_tci. */
533 vlan_tci = parse_vlan(&data, &size);
534 dl_type = parse_ethertype(&data, &size);
535 miniflow_push_be16(mf, dl_type, dl_type);
536 miniflow_push_be16(mf, vlan_tci, vlan_tci);
540 if (OVS_UNLIKELY(eth_type_mpls(dl_type))) {
542 const void *mpls = data;
544 packet->l2_5_ofs = (char *)data - l2;
545 count = parse_mpls(&data, &size);
546 miniflow_push_words_32(mf, mpls_lse, mpls, count);
550 packet->l3_ofs = (char *)data - l2;
553 if (OVS_LIKELY(dl_type == htons(ETH_TYPE_IP))) {
554 const struct ip_header *nh = data;
558 if (OVS_UNLIKELY(size < IP_HEADER_LEN)) {
561 ip_len = IP_IHL(nh->ip_ihl_ver) * 4;
563 if (OVS_UNLIKELY(ip_len < IP_HEADER_LEN)) {
566 if (OVS_UNLIKELY(size < ip_len)) {
569 tot_len = ntohs(nh->ip_tot_len);
570 if (OVS_UNLIKELY(tot_len > size)) {
573 if (OVS_UNLIKELY(size - tot_len > UINT8_MAX)) {
576 dp_packet_set_l2_pad_size(packet, size - tot_len);
577 size = tot_len; /* Never pull padding. */
579 /* Push both source and destination address at once. */
580 miniflow_push_words(mf, nw_src, &nh->ip_src, 1);
582 miniflow_push_be32(mf, ipv6_label, 0); /* Padding for IPv4. */
586 nw_proto = nh->ip_proto;
587 if (OVS_UNLIKELY(IP_IS_FRAGMENT(nh->ip_frag_off))) {
588 nw_frag = FLOW_NW_FRAG_ANY;
589 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
590 nw_frag |= FLOW_NW_FRAG_LATER;
593 data_pull(&data, &size, ip_len);
594 } else if (dl_type == htons(ETH_TYPE_IPV6)) {
595 const struct ovs_16aligned_ip6_hdr *nh;
599 if (OVS_UNLIKELY(size < sizeof *nh)) {
602 nh = data_pull(&data, &size, sizeof *nh);
604 plen = ntohs(nh->ip6_plen);
605 if (OVS_UNLIKELY(plen > size)) {
608 /* Jumbo Payload option not supported yet. */
609 if (OVS_UNLIKELY(size - plen > UINT8_MAX)) {
612 dp_packet_set_l2_pad_size(packet, size - plen);
613 size = plen; /* Never pull padding. */
615 miniflow_push_words(mf, ipv6_src, &nh->ip6_src,
616 sizeof nh->ip6_src / 8);
617 miniflow_push_words(mf, ipv6_dst, &nh->ip6_dst,
618 sizeof nh->ip6_dst / 8);
620 tc_flow = get_16aligned_be32(&nh->ip6_flow);
622 ovs_be32 label = tc_flow & htonl(IPV6_LABEL_MASK);
623 miniflow_push_be32(mf, ipv6_label, label);
626 nw_tos = ntohl(tc_flow) >> 20;
627 nw_ttl = nh->ip6_hlim;
628 nw_proto = nh->ip6_nxt;
631 if (OVS_LIKELY((nw_proto != IPPROTO_HOPOPTS)
632 && (nw_proto != IPPROTO_ROUTING)
633 && (nw_proto != IPPROTO_DSTOPTS)
634 && (nw_proto != IPPROTO_AH)
635 && (nw_proto != IPPROTO_FRAGMENT))) {
636 /* It's either a terminal header (e.g., TCP, UDP) or one we
637 * don't understand. In either case, we're done with the
638 * packet, so use it to fill in 'nw_proto'. */
642 /* We only verify that at least 8 bytes of the next header are
643 * available, but many of these headers are longer. Ensure that
644 * accesses within the extension header are within those first 8
645 * bytes. All extension headers are required to be at least 8
647 if (OVS_UNLIKELY(size < 8)) {
651 if ((nw_proto == IPPROTO_HOPOPTS)
652 || (nw_proto == IPPROTO_ROUTING)
653 || (nw_proto == IPPROTO_DSTOPTS)) {
654 /* These headers, while different, have the fields we care
655 * about in the same location and with the same
657 const struct ip6_ext *ext_hdr = data;
658 nw_proto = ext_hdr->ip6e_nxt;
659 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
660 (ext_hdr->ip6e_len + 1) * 8))) {
663 } else if (nw_proto == IPPROTO_AH) {
664 /* A standard AH definition isn't available, but the fields
665 * we care about are in the same location as the generic
666 * option header--only the header length is calculated
668 const struct ip6_ext *ext_hdr = data;
669 nw_proto = ext_hdr->ip6e_nxt;
670 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
671 (ext_hdr->ip6e_len + 2) * 4))) {
674 } else if (nw_proto == IPPROTO_FRAGMENT) {
675 const struct ovs_16aligned_ip6_frag *frag_hdr = data;
677 nw_proto = frag_hdr->ip6f_nxt;
678 if (!data_try_pull(&data, &size, sizeof *frag_hdr)) {
682 /* We only process the first fragment. */
683 if (frag_hdr->ip6f_offlg != htons(0)) {
684 nw_frag = FLOW_NW_FRAG_ANY;
685 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
686 nw_frag |= FLOW_NW_FRAG_LATER;
687 nw_proto = IPPROTO_FRAGMENT;
694 if (dl_type == htons(ETH_TYPE_ARP) ||
695 dl_type == htons(ETH_TYPE_RARP)) {
696 struct eth_addr arp_buf[2];
697 const struct arp_eth_header *arp = (const struct arp_eth_header *)
698 data_try_pull(&data, &size, ARP_ETH_HEADER_LEN);
700 if (OVS_LIKELY(arp) && OVS_LIKELY(arp->ar_hrd == htons(1))
701 && OVS_LIKELY(arp->ar_pro == htons(ETH_TYPE_IP))
702 && OVS_LIKELY(arp->ar_hln == ETH_ADDR_LEN)
703 && OVS_LIKELY(arp->ar_pln == 4)) {
704 miniflow_push_be32(mf, nw_src,
705 get_16aligned_be32(&arp->ar_spa));
706 miniflow_push_be32(mf, nw_dst,
707 get_16aligned_be32(&arp->ar_tpa));
709 /* We only match on the lower 8 bits of the opcode. */
710 if (OVS_LIKELY(ntohs(arp->ar_op) <= 0xff)) {
711 miniflow_push_be32(mf, ipv6_label, 0); /* Pad with ARP. */
712 miniflow_push_be32(mf, nw_frag, htonl(ntohs(arp->ar_op)));
715 /* Must be adjacent. */
716 ASSERT_SEQUENTIAL(arp_sha, arp_tha);
718 arp_buf[0] = arp->ar_sha;
719 arp_buf[1] = arp->ar_tha;
720 miniflow_push_macs(mf, arp_sha, arp_buf);
721 miniflow_pad_to_64(mf, arp_tha);
727 packet->l4_ofs = (char *)data - l2;
728 miniflow_push_be32(mf, nw_frag,
729 BYTES_TO_BE32(nw_frag, nw_tos, nw_ttl, nw_proto));
731 if (OVS_LIKELY(!(nw_frag & FLOW_NW_FRAG_LATER))) {
732 if (OVS_LIKELY(nw_proto == IPPROTO_TCP)) {
733 if (OVS_LIKELY(size >= TCP_HEADER_LEN)) {
734 const struct tcp_header *tcp = data;
736 miniflow_push_be32(mf, arp_tha.ea[2], 0);
737 miniflow_push_be32(mf, tcp_flags,
738 TCP_FLAGS_BE32(tcp->tcp_ctl));
739 miniflow_push_be16(mf, tp_src, tcp->tcp_src);
740 miniflow_push_be16(mf, tp_dst, tcp->tcp_dst);
741 miniflow_pad_to_64(mf, tp_dst);
743 } else if (OVS_LIKELY(nw_proto == IPPROTO_UDP)) {
744 if (OVS_LIKELY(size >= UDP_HEADER_LEN)) {
745 const struct udp_header *udp = data;
747 miniflow_push_be16(mf, tp_src, udp->udp_src);
748 miniflow_push_be16(mf, tp_dst, udp->udp_dst);
749 miniflow_pad_to_64(mf, tp_dst);
751 } else if (OVS_LIKELY(nw_proto == IPPROTO_SCTP)) {
752 if (OVS_LIKELY(size >= SCTP_HEADER_LEN)) {
753 const struct sctp_header *sctp = data;
755 miniflow_push_be16(mf, tp_src, sctp->sctp_src);
756 miniflow_push_be16(mf, tp_dst, sctp->sctp_dst);
757 miniflow_pad_to_64(mf, tp_dst);
759 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMP)) {
760 if (OVS_LIKELY(size >= ICMP_HEADER_LEN)) {
761 const struct icmp_header *icmp = data;
763 miniflow_push_be16(mf, tp_src, htons(icmp->icmp_type));
764 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp_code));
765 miniflow_pad_to_64(mf, tp_dst);
767 } else if (OVS_LIKELY(nw_proto == IPPROTO_IGMP)) {
768 if (OVS_LIKELY(size >= IGMP_HEADER_LEN)) {
769 const struct igmp_header *igmp = data;
771 miniflow_push_be16(mf, tp_src, htons(igmp->igmp_type));
772 miniflow_push_be16(mf, tp_dst, htons(igmp->igmp_code));
773 miniflow_push_be32(mf, igmp_group_ip4,
774 get_16aligned_be32(&igmp->group));
776 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMPV6)) {
777 if (OVS_LIKELY(size >= sizeof(struct icmp6_hdr))) {
778 const struct in6_addr *nd_target = NULL;
779 struct eth_addr arp_buf[2] = { { { { 0 } } } };
780 const struct icmp6_hdr *icmp = data_pull(&data, &size,
782 parse_icmpv6(&data, &size, icmp, &nd_target, arp_buf);
784 miniflow_push_words(mf, nd_target, nd_target,
785 sizeof *nd_target / sizeof(uint64_t));
787 miniflow_push_macs(mf, arp_sha, arp_buf);
788 miniflow_pad_to_64(mf, arp_tha);
789 miniflow_push_be16(mf, tp_src, htons(icmp->icmp6_type));
790 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp6_code));
791 miniflow_pad_to_64(mf, tp_dst);
799 /* For every bit of a field that is wildcarded in 'wildcards', sets the
800 * corresponding bit in 'flow' to zero. */
802 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
804 uint64_t *flow_u64 = (uint64_t *) flow;
805 const uint64_t *wc_u64 = (const uint64_t *) &wildcards->masks;
808 for (i = 0; i < FLOW_U64S; i++) {
809 flow_u64[i] &= wc_u64[i];
814 flow_unwildcard_tp_ports(const struct flow *flow, struct flow_wildcards *wc)
816 if (flow->nw_proto != IPPROTO_ICMP) {
817 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
818 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
820 wc->masks.tp_src = htons(0xff);
821 wc->masks.tp_dst = htons(0xff);
825 /* Initializes 'flow_metadata' with the metadata found in 'flow'. */
827 flow_get_metadata(const struct flow *flow, struct match *flow_metadata)
831 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 35);
833 match_init_catchall(flow_metadata);
834 if (flow->tunnel.tun_id != htonll(0)) {
835 match_set_tun_id(flow_metadata, flow->tunnel.tun_id);
837 if (flow->tunnel.flags & FLOW_TNL_PUB_F_MASK) {
838 match_set_tun_flags(flow_metadata,
839 flow->tunnel.flags & FLOW_TNL_PUB_F_MASK);
841 if (flow->tunnel.ip_src) {
842 match_set_tun_src(flow_metadata, flow->tunnel.ip_src);
844 if (flow->tunnel.ip_dst) {
845 match_set_tun_dst(flow_metadata, flow->tunnel.ip_dst);
847 if (ipv6_addr_is_set(&flow->tunnel.ipv6_src)) {
848 match_set_tun_ipv6_src(flow_metadata, &flow->tunnel.ipv6_src);
850 if (ipv6_addr_is_set(&flow->tunnel.ipv6_dst)) {
851 match_set_tun_ipv6_dst(flow_metadata, &flow->tunnel.ipv6_dst);
853 if (flow->tunnel.gbp_id != htons(0)) {
854 match_set_tun_gbp_id(flow_metadata, flow->tunnel.gbp_id);
856 if (flow->tunnel.gbp_flags) {
857 match_set_tun_gbp_flags(flow_metadata, flow->tunnel.gbp_flags);
859 tun_metadata_get_fmd(&flow->tunnel, flow_metadata);
860 if (flow->metadata != htonll(0)) {
861 match_set_metadata(flow_metadata, flow->metadata);
864 for (i = 0; i < FLOW_N_REGS; i++) {
866 match_set_reg(flow_metadata, i, flow->regs[i]);
870 if (flow->pkt_mark != 0) {
871 match_set_pkt_mark(flow_metadata, flow->pkt_mark);
874 match_set_in_port(flow_metadata, flow->in_port.ofp_port);
875 if (flow->ct_state != 0) {
876 match_set_ct_state(flow_metadata, flow->ct_state);
878 if (flow->ct_zone != 0) {
879 match_set_ct_zone(flow_metadata, flow->ct_zone);
881 if (flow->ct_mark != 0) {
882 match_set_ct_mark(flow_metadata, flow->ct_mark);
884 if (!ovs_u128_is_zero(&flow->ct_label)) {
885 match_set_ct_label(flow_metadata, flow->ct_label);
889 const char *ct_state_to_string(uint32_t state)
914 flow_to_string(const struct flow *flow)
916 struct ds ds = DS_EMPTY_INITIALIZER;
917 flow_format(&ds, flow);
922 flow_tun_flag_to_string(uint32_t flags)
925 case FLOW_TNL_F_DONT_FRAGMENT:
927 case FLOW_TNL_F_CSUM:
939 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
940 uint32_t flags, char del)
945 ds_put_char(ds, '0');
949 uint32_t bit = rightmost_1bit(flags);
952 s = bit_to_string(bit);
954 ds_put_format(ds, "%s%c", s, del);
963 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
969 format_flags_masked(struct ds *ds, const char *name,
970 const char *(*bit_to_string)(uint32_t), uint32_t flags,
971 uint32_t mask, uint32_t max_mask)
974 ds_put_format(ds, "%s=", name);
977 if (mask == max_mask) {
978 format_flags(ds, bit_to_string, flags, '|');
983 ds_put_cstr(ds, "0/0");
988 uint32_t bit = rightmost_1bit(mask);
989 const char *s = bit_to_string(bit);
991 ds_put_format(ds, "%s%s", (flags & bit) ? "+" : "-",
992 s ? s : "[Unknown]");
997 /* Scans a string 's' of flags to determine their numerical value and
998 * returns the number of characters parsed using 'bit_to_string' to
999 * lookup flag names. Scanning continues until the character 'end' is
1002 * In the event of a failure, a negative error code will be returned. In
1003 * addition, if 'res_string' is non-NULL then a descriptive string will
1004 * be returned incorporating the identifying string 'field_name'. This
1005 * error string must be freed by the caller.
1007 * Upon success, the flag values will be stored in 'res_flags' and
1008 * optionally 'res_mask', if it is non-NULL (if it is NULL then any masks
1009 * present in the original string will be considered an error). The
1010 * caller may restrict the acceptable set of values through the mask
1013 parse_flags(const char *s, const char *(*bit_to_string)(uint32_t),
1014 char end, const char *field_name, char **res_string,
1015 uint32_t *res_flags, uint32_t allowed, uint32_t *res_mask)
1017 uint32_t result = 0;
1020 /* Parse masked flags in numeric format? */
1021 if (res_mask && ovs_scan(s, "%"SCNi32"/%"SCNi32"%n",
1022 res_flags, res_mask, &n) && n > 0) {
1023 if (*res_flags & ~allowed || *res_mask & ~allowed) {
1031 if (res_mask && (*s == '+' || *s == '-')) {
1032 uint32_t flags = 0, mask = 0;
1034 /* Parse masked flags. */
1035 while (s[0] != end) {
1042 } else if (s[0] == '-') {
1046 *res_string = xasprintf("%s: %s must be preceded by '+' "
1047 "(for SET) or '-' (NOT SET)", s,
1055 for (bit = 1; bit; bit <<= 1) {
1056 const char *fname = bit_to_string(bit);
1062 len = strlen(fname);
1063 if (strncmp(s, fname, len) ||
1064 (s[len] != '+' && s[len] != '-' && s[len] != end)) {
1069 /* bit already set. */
1071 *res_string = xasprintf("%s: Each %s flag can be "
1072 "specified only once", s,
1077 if (!(bit & allowed)) {
1099 /* Parse unmasked flags. If a flag is present, it is set, otherwise
1101 while (s[n] != end) {
1102 unsigned long long int flags;
1106 if (ovs_scan(&s[n], "%lli%n", &flags, &n0)) {
1107 if (flags & ~allowed) {
1110 n += n0 + (s[n + n0] == '|');
1115 for (bit = 1; bit; bit <<= 1) {
1116 const char *name = bit_to_string(bit);
1124 if (!strncmp(s + n, name, len) &&
1125 (s[n + len] == '|' || s[n + len] == end)) {
1126 if (!(bit & allowed)) {
1130 n += len + (s[n + len] == '|');
1140 *res_flags = result;
1142 *res_mask = UINT32_MAX;
1151 *res_string = xasprintf("%s: unknown %s flag(s)", s, field_name);
1157 flow_format(struct ds *ds, const struct flow *flow)
1160 struct flow_wildcards *wc = &match.wc;
1162 match_wc_init(&match, flow);
1164 /* As this function is most often used for formatting a packet in a
1165 * packet-in message, skip formatting the packet context fields that are
1166 * all-zeroes to make the print-out easier on the eyes. This means that a
1167 * missing context field implies a zero value for that field. This is
1168 * similar to OpenFlow encoding of these fields, as the specification
1169 * states that all-zeroes context fields should not be encoded in the
1170 * packet-in messages. */
1171 if (!flow->in_port.ofp_port) {
1172 WC_UNMASK_FIELD(wc, in_port);
1174 if (!flow->skb_priority) {
1175 WC_UNMASK_FIELD(wc, skb_priority);
1177 if (!flow->pkt_mark) {
1178 WC_UNMASK_FIELD(wc, pkt_mark);
1180 if (!flow->recirc_id) {
1181 WC_UNMASK_FIELD(wc, recirc_id);
1183 if (!flow->dp_hash) {
1184 WC_UNMASK_FIELD(wc, dp_hash);
1186 if (!flow->ct_state) {
1187 WC_UNMASK_FIELD(wc, ct_state);
1189 if (!flow->ct_zone) {
1190 WC_UNMASK_FIELD(wc, ct_zone);
1192 if (!flow->ct_mark) {
1193 WC_UNMASK_FIELD(wc, ct_mark);
1195 if (ovs_u128_is_zero(&flow->ct_label)) {
1196 WC_UNMASK_FIELD(wc, ct_label);
1198 for (int i = 0; i < FLOW_N_REGS; i++) {
1199 if (!flow->regs[i]) {
1200 WC_UNMASK_FIELD(wc, regs[i]);
1203 if (!flow->metadata) {
1204 WC_UNMASK_FIELD(wc, metadata);
1207 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
1211 flow_print(FILE *stream, const struct flow *flow)
1213 char *s = flow_to_string(flow);
1218 /* flow_wildcards functions. */
1220 /* Initializes 'wc' as a set of wildcards that matches every packet. */
1222 flow_wildcards_init_catchall(struct flow_wildcards *wc)
1224 memset(&wc->masks, 0, sizeof wc->masks);
1227 /* Converts a flow into flow wildcards. It sets the wildcard masks based on
1228 * the packet headers extracted to 'flow'. It will not set the mask for fields
1229 * that do not make sense for the packet type. OpenFlow-only metadata is
1230 * wildcarded, but other metadata is unconditionally exact-matched. */
1231 void flow_wildcards_init_for_packet(struct flow_wildcards *wc,
1232 const struct flow *flow)
1234 memset(&wc->masks, 0x0, sizeof wc->masks);
1236 /* Update this function whenever struct flow changes. */
1237 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 35);
1239 if (flow_tnl_dst_is_set(&flow->tunnel)) {
1240 if (flow->tunnel.flags & FLOW_TNL_F_KEY) {
1241 WC_MASK_FIELD(wc, tunnel.tun_id);
1243 WC_MASK_FIELD(wc, tunnel.ip_src);
1244 WC_MASK_FIELD(wc, tunnel.ip_dst);
1245 WC_MASK_FIELD(wc, tunnel.ipv6_src);
1246 WC_MASK_FIELD(wc, tunnel.ipv6_dst);
1247 WC_MASK_FIELD(wc, tunnel.flags);
1248 WC_MASK_FIELD(wc, tunnel.ip_tos);
1249 WC_MASK_FIELD(wc, tunnel.ip_ttl);
1250 WC_MASK_FIELD(wc, tunnel.tp_src);
1251 WC_MASK_FIELD(wc, tunnel.tp_dst);
1252 WC_MASK_FIELD(wc, tunnel.gbp_id);
1253 WC_MASK_FIELD(wc, tunnel.gbp_flags);
1255 if (!(flow->tunnel.flags & FLOW_TNL_F_UDPIF)) {
1256 if (flow->tunnel.metadata.present.map) {
1257 wc->masks.tunnel.metadata.present.map =
1258 flow->tunnel.metadata.present.map;
1259 WC_MASK_FIELD(wc, tunnel.metadata.opts.u8);
1262 WC_MASK_FIELD(wc, tunnel.metadata.present.len);
1263 memset(wc->masks.tunnel.metadata.opts.gnv, 0xff,
1264 flow->tunnel.metadata.present.len);
1266 } else if (flow->tunnel.tun_id) {
1267 WC_MASK_FIELD(wc, tunnel.tun_id);
1270 /* metadata, regs, and conj_id wildcarded. */
1272 WC_MASK_FIELD(wc, skb_priority);
1273 WC_MASK_FIELD(wc, pkt_mark);
1274 WC_MASK_FIELD(wc, ct_state);
1275 WC_MASK_FIELD(wc, ct_zone);
1276 WC_MASK_FIELD(wc, ct_mark);
1277 WC_MASK_FIELD(wc, ct_label);
1278 WC_MASK_FIELD(wc, recirc_id);
1279 WC_MASK_FIELD(wc, dp_hash);
1280 WC_MASK_FIELD(wc, in_port);
1282 /* actset_output wildcarded. */
1284 WC_MASK_FIELD(wc, dl_dst);
1285 WC_MASK_FIELD(wc, dl_src);
1286 WC_MASK_FIELD(wc, dl_type);
1287 WC_MASK_FIELD(wc, vlan_tci);
1289 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1290 WC_MASK_FIELD(wc, nw_src);
1291 WC_MASK_FIELD(wc, nw_dst);
1292 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1293 WC_MASK_FIELD(wc, ipv6_src);
1294 WC_MASK_FIELD(wc, ipv6_dst);
1295 WC_MASK_FIELD(wc, ipv6_label);
1296 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1297 flow->dl_type == htons(ETH_TYPE_RARP)) {
1298 WC_MASK_FIELD(wc, nw_src);
1299 WC_MASK_FIELD(wc, nw_dst);
1300 WC_MASK_FIELD(wc, nw_proto);
1301 WC_MASK_FIELD(wc, arp_sha);
1302 WC_MASK_FIELD(wc, arp_tha);
1304 } else if (eth_type_mpls(flow->dl_type)) {
1305 for (int i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1306 WC_MASK_FIELD(wc, mpls_lse[i]);
1307 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1313 return; /* Unknown ethertype. */
1317 WC_MASK_FIELD(wc, nw_frag);
1318 WC_MASK_FIELD(wc, nw_tos);
1319 WC_MASK_FIELD(wc, nw_ttl);
1320 WC_MASK_FIELD(wc, nw_proto);
1322 /* No transport layer header in later fragments. */
1323 if (!(flow->nw_frag & FLOW_NW_FRAG_LATER) &&
1324 (flow->nw_proto == IPPROTO_ICMP ||
1325 flow->nw_proto == IPPROTO_ICMPV6 ||
1326 flow->nw_proto == IPPROTO_TCP ||
1327 flow->nw_proto == IPPROTO_UDP ||
1328 flow->nw_proto == IPPROTO_SCTP ||
1329 flow->nw_proto == IPPROTO_IGMP)) {
1330 WC_MASK_FIELD(wc, tp_src);
1331 WC_MASK_FIELD(wc, tp_dst);
1333 if (flow->nw_proto == IPPROTO_TCP) {
1334 WC_MASK_FIELD(wc, tcp_flags);
1335 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
1336 WC_MASK_FIELD(wc, arp_sha);
1337 WC_MASK_FIELD(wc, arp_tha);
1338 WC_MASK_FIELD(wc, nd_target);
1339 } else if (flow->nw_proto == IPPROTO_IGMP) {
1340 WC_MASK_FIELD(wc, igmp_group_ip4);
1345 /* Return a map of possible fields for a packet of the same type as 'flow'.
1346 * Including extra bits in the returned mask is not wrong, it is just less
1349 * This is a less precise version of flow_wildcards_init_for_packet() above. */
1351 flow_wc_map(const struct flow *flow, struct flowmap *map)
1353 /* Update this function whenever struct flow changes. */
1354 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 35);
1358 if (flow_tnl_dst_is_set(&flow->tunnel)) {
1359 FLOWMAP_SET__(map, tunnel, offsetof(struct flow_tnl, metadata));
1360 if (!(flow->tunnel.flags & FLOW_TNL_F_UDPIF)) {
1361 if (flow->tunnel.metadata.present.map) {
1362 FLOWMAP_SET(map, tunnel.metadata);
1365 FLOWMAP_SET(map, tunnel.metadata.present.len);
1366 FLOWMAP_SET__(map, tunnel.metadata.opts.gnv,
1367 flow->tunnel.metadata.present.len);
1371 /* Metadata fields that can appear on packet input. */
1372 FLOWMAP_SET(map, skb_priority);
1373 FLOWMAP_SET(map, pkt_mark);
1374 FLOWMAP_SET(map, recirc_id);
1375 FLOWMAP_SET(map, dp_hash);
1376 FLOWMAP_SET(map, in_port);
1377 FLOWMAP_SET(map, dl_dst);
1378 FLOWMAP_SET(map, dl_src);
1379 FLOWMAP_SET(map, dl_type);
1380 FLOWMAP_SET(map, vlan_tci);
1381 FLOWMAP_SET(map, ct_state);
1382 FLOWMAP_SET(map, ct_zone);
1383 FLOWMAP_SET(map, ct_mark);
1384 FLOWMAP_SET(map, ct_label);
1386 /* Ethertype-dependent fields. */
1387 if (OVS_LIKELY(flow->dl_type == htons(ETH_TYPE_IP))) {
1388 FLOWMAP_SET(map, nw_src);
1389 FLOWMAP_SET(map, nw_dst);
1390 FLOWMAP_SET(map, nw_proto);
1391 FLOWMAP_SET(map, nw_frag);
1392 FLOWMAP_SET(map, nw_tos);
1393 FLOWMAP_SET(map, nw_ttl);
1395 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_IGMP)) {
1396 FLOWMAP_SET(map, igmp_group_ip4);
1398 FLOWMAP_SET(map, tcp_flags);
1399 FLOWMAP_SET(map, tp_src);
1400 FLOWMAP_SET(map, tp_dst);
1402 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1403 FLOWMAP_SET(map, ipv6_src);
1404 FLOWMAP_SET(map, ipv6_dst);
1405 FLOWMAP_SET(map, ipv6_label);
1406 FLOWMAP_SET(map, nw_proto);
1407 FLOWMAP_SET(map, nw_frag);
1408 FLOWMAP_SET(map, nw_tos);
1409 FLOWMAP_SET(map, nw_ttl);
1411 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_ICMPV6)) {
1412 FLOWMAP_SET(map, nd_target);
1413 FLOWMAP_SET(map, arp_sha);
1414 FLOWMAP_SET(map, arp_tha);
1416 FLOWMAP_SET(map, tcp_flags);
1417 FLOWMAP_SET(map, tp_src);
1418 FLOWMAP_SET(map, tp_dst);
1420 } else if (eth_type_mpls(flow->dl_type)) {
1421 FLOWMAP_SET(map, mpls_lse);
1422 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1423 flow->dl_type == htons(ETH_TYPE_RARP)) {
1424 FLOWMAP_SET(map, nw_src);
1425 FLOWMAP_SET(map, nw_dst);
1426 FLOWMAP_SET(map, nw_proto);
1427 FLOWMAP_SET(map, arp_sha);
1428 FLOWMAP_SET(map, arp_tha);
1432 /* Clear the metadata and register wildcard masks. They are not packet
1435 flow_wildcards_clear_non_packet_fields(struct flow_wildcards *wc)
1437 /* Update this function whenever struct flow changes. */
1438 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 35);
1440 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
1441 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
1442 wc->masks.actset_output = 0;
1443 wc->masks.conj_id = 0;
1446 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
1449 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
1451 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1454 for (i = 0; i < FLOW_U64S; i++) {
1462 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
1463 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
1464 * in 'src1' or 'src2' or both. */
1466 flow_wildcards_and(struct flow_wildcards *dst,
1467 const struct flow_wildcards *src1,
1468 const struct flow_wildcards *src2)
1470 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1471 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1472 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1475 for (i = 0; i < FLOW_U64S; i++) {
1476 dst_u64[i] = src1_u64[i] & src2_u64[i];
1480 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
1481 * is, a bit or a field is wildcarded in 'dst' if it is neither
1482 * wildcarded in 'src1' nor 'src2'. */
1484 flow_wildcards_or(struct flow_wildcards *dst,
1485 const struct flow_wildcards *src1,
1486 const struct flow_wildcards *src2)
1488 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1489 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1490 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1493 for (i = 0; i < FLOW_U64S; i++) {
1494 dst_u64[i] = src1_u64[i] | src2_u64[i];
1498 /* Returns a hash of the wildcards in 'wc'. */
1500 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
1502 return flow_hash(&wc->masks, basis);
1505 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
1508 flow_wildcards_equal(const struct flow_wildcards *a,
1509 const struct flow_wildcards *b)
1511 return flow_equal(&a->masks, &b->masks);
1514 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
1515 * 'b', false otherwise. */
1517 flow_wildcards_has_extra(const struct flow_wildcards *a,
1518 const struct flow_wildcards *b)
1520 const uint64_t *a_u64 = (const uint64_t *) &a->masks;
1521 const uint64_t *b_u64 = (const uint64_t *) &b->masks;
1524 for (i = 0; i < FLOW_U64S; i++) {
1525 if ((a_u64[i] & b_u64[i]) != b_u64[i]) {
1532 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
1533 * in 'wc' do not need to be equal in 'a' and 'b'. */
1535 flow_equal_except(const struct flow *a, const struct flow *b,
1536 const struct flow_wildcards *wc)
1538 const uint64_t *a_u64 = (const uint64_t *) a;
1539 const uint64_t *b_u64 = (const uint64_t *) b;
1540 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1543 for (i = 0; i < FLOW_U64S; i++) {
1544 if ((a_u64[i] ^ b_u64[i]) & wc_u64[i]) {
1551 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1552 * (A 0-bit indicates a wildcard bit.) */
1554 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
1556 wc->masks.regs[idx] = mask;
1559 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1560 * (A 0-bit indicates a wildcard bit.) */
1562 flow_wildcards_set_xreg_mask(struct flow_wildcards *wc, int idx, uint64_t mask)
1564 flow_set_xreg(&wc->masks, idx, mask);
1567 /* Calculates the 5-tuple hash from the given miniflow.
1568 * This returns the same value as flow_hash_5tuple for the corresponding
1571 miniflow_hash_5tuple(const struct miniflow *flow, uint32_t basis)
1573 uint32_t hash = basis;
1576 ovs_be16 dl_type = MINIFLOW_GET_BE16(flow, dl_type);
1578 hash = hash_add(hash, MINIFLOW_GET_U8(flow, nw_proto));
1580 /* Separate loops for better optimization. */
1581 if (dl_type == htons(ETH_TYPE_IPV6)) {
1582 struct flowmap map = FLOWMAP_EMPTY_INITIALIZER;
1585 FLOWMAP_SET(&map, ipv6_src);
1586 FLOWMAP_SET(&map, ipv6_dst);
1588 MINIFLOW_FOR_EACH_IN_FLOWMAP(value, flow, map) {
1589 hash = hash_add64(hash, value);
1592 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_src));
1593 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_dst));
1595 /* Add both ports at once. */
1596 hash = hash_add(hash, MINIFLOW_GET_U32(flow, tp_src));
1597 hash = hash_finish(hash, 42); /* Arbitrary number. */
1602 ASSERT_SEQUENTIAL_SAME_WORD(tp_src, tp_dst);
1603 ASSERT_SEQUENTIAL(ipv6_src, ipv6_dst);
1605 /* Calculates the 5-tuple hash from the given flow. */
1607 flow_hash_5tuple(const struct flow *flow, uint32_t basis)
1609 uint32_t hash = basis;
1612 hash = hash_add(hash, flow->nw_proto);
1614 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1615 const uint64_t *flow_u64 = (const uint64_t *)flow;
1616 int ofs = offsetof(struct flow, ipv6_src) / 8;
1617 int end = ofs + 2 * sizeof flow->ipv6_src / 8;
1619 for (;ofs < end; ofs++) {
1620 hash = hash_add64(hash, flow_u64[ofs]);
1623 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_src);
1624 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_dst);
1626 /* Add both ports at once. */
1627 hash = hash_add(hash,
1628 ((const uint32_t *)flow)[offsetof(struct flow, tp_src)
1629 / sizeof(uint32_t)]);
1630 hash = hash_finish(hash, 42); /* Arbitrary number. */
1635 /* Hashes 'flow' based on its L2 through L4 protocol information. */
1637 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
1642 struct in6_addr ipv6_addr;
1647 struct eth_addr eth_addr;
1653 memset(&fields, 0, sizeof fields);
1654 for (i = 0; i < ARRAY_SIZE(fields.eth_addr.be16); i++) {
1655 fields.eth_addr.be16[i] = flow->dl_src.be16[i] ^ flow->dl_dst.be16[i];
1657 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
1658 fields.eth_type = flow->dl_type;
1660 /* UDP source and destination port are not taken into account because they
1661 * will not necessarily be symmetric in a bidirectional flow. */
1662 if (fields.eth_type == htons(ETH_TYPE_IP)) {
1663 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
1664 fields.ip_proto = flow->nw_proto;
1665 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1666 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1668 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
1669 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
1670 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
1671 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
1673 for (i=0; i<16; i++) {
1674 ipv6_addr[i] = a[i] ^ b[i];
1676 fields.ip_proto = flow->nw_proto;
1677 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1678 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1681 return jhash_bytes(&fields, sizeof fields, basis);
1684 /* Hashes 'flow' based on its L3 through L4 protocol information */
1686 flow_hash_symmetric_l3l4(const struct flow *flow, uint32_t basis,
1689 uint32_t hash = basis;
1691 /* UDP source and destination port are also taken into account. */
1692 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1693 hash = hash_add(hash,
1694 (OVS_FORCE uint32_t) (flow->nw_src ^ flow->nw_dst));
1695 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1696 /* IPv6 addresses are 64-bit aligned inside struct flow. */
1697 const uint64_t *a = ALIGNED_CAST(uint64_t *, flow->ipv6_src.s6_addr);
1698 const uint64_t *b = ALIGNED_CAST(uint64_t *, flow->ipv6_dst.s6_addr);
1700 for (int i = 0; i < 4; i++) {
1701 hash = hash_add64(hash, a[i] ^ b[i]);
1704 /* Cannot hash non-IP flows */
1708 hash = hash_add(hash, flow->nw_proto);
1709 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP ||
1710 (inc_udp_ports && flow->nw_proto == IPPROTO_UDP)) {
1711 hash = hash_add(hash,
1712 (OVS_FORCE uint16_t) (flow->tp_src ^ flow->tp_dst));
1715 return hash_finish(hash, basis);
1718 /* Initialize a flow with random fields that matter for nx_hash_fields. */
1720 flow_random_hash_fields(struct flow *flow)
1722 uint16_t rnd = random_uint16();
1724 /* Initialize to all zeros. */
1725 memset(flow, 0, sizeof *flow);
1727 eth_addr_random(&flow->dl_src);
1728 eth_addr_random(&flow->dl_dst);
1730 flow->vlan_tci = (OVS_FORCE ovs_be16) (random_uint16() & VLAN_VID_MASK);
1732 /* Make most of the random flows IPv4, some IPv6, and rest random. */
1733 flow->dl_type = rnd < 0x8000 ? htons(ETH_TYPE_IP) :
1734 rnd < 0xc000 ? htons(ETH_TYPE_IPV6) : (OVS_FORCE ovs_be16)rnd;
1736 if (dl_type_is_ip_any(flow->dl_type)) {
1737 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1738 flow->nw_src = (OVS_FORCE ovs_be32)random_uint32();
1739 flow->nw_dst = (OVS_FORCE ovs_be32)random_uint32();
1741 random_bytes(&flow->ipv6_src, sizeof flow->ipv6_src);
1742 random_bytes(&flow->ipv6_dst, sizeof flow->ipv6_dst);
1744 /* Make most of IP flows TCP, some UDP or SCTP, and rest random. */
1745 rnd = random_uint16();
1746 flow->nw_proto = rnd < 0x8000 ? IPPROTO_TCP :
1747 rnd < 0xc000 ? IPPROTO_UDP :
1748 rnd < 0xd000 ? IPPROTO_SCTP : (uint8_t)rnd;
1749 if (flow->nw_proto == IPPROTO_TCP ||
1750 flow->nw_proto == IPPROTO_UDP ||
1751 flow->nw_proto == IPPROTO_SCTP) {
1752 flow->tp_src = (OVS_FORCE ovs_be16)random_uint16();
1753 flow->tp_dst = (OVS_FORCE ovs_be16)random_uint16();
1758 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
1760 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
1761 enum nx_hash_fields fields)
1764 case NX_HASH_FIELDS_ETH_SRC:
1765 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1768 case NX_HASH_FIELDS_SYMMETRIC_L4:
1769 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1770 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1771 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1772 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1773 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1774 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1775 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1776 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1778 if (is_ip_any(flow)) {
1779 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1780 flow_unwildcard_tp_ports(flow, wc);
1782 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1785 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1786 if (is_ip_any(flow) && flow->nw_proto == IPPROTO_UDP) {
1787 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1788 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1791 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1792 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1793 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1794 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1795 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1796 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1797 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1799 break; /* non-IP flow */
1802 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1803 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP) {
1804 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1805 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1814 /* Hashes the portions of 'flow' designated by 'fields'. */
1816 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
1821 case NX_HASH_FIELDS_ETH_SRC:
1822 return jhash_bytes(&flow->dl_src, sizeof flow->dl_src, basis);
1824 case NX_HASH_FIELDS_SYMMETRIC_L4:
1825 return flow_hash_symmetric_l4(flow, basis);
1827 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1828 return flow_hash_symmetric_l3l4(flow, basis, false);
1830 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1831 return flow_hash_symmetric_l3l4(flow, basis, true);
1838 /* Returns a string representation of 'fields'. */
1840 flow_hash_fields_to_str(enum nx_hash_fields fields)
1843 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
1844 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
1845 case NX_HASH_FIELDS_SYMMETRIC_L3L4: return "symmetric_l3l4";
1846 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP: return "symmetric_l3l4+udp";
1847 default: return "<unknown>";
1851 /* Returns true if the value of 'fields' is supported. Otherwise false. */
1853 flow_hash_fields_valid(enum nx_hash_fields fields)
1855 return fields == NX_HASH_FIELDS_ETH_SRC
1856 || fields == NX_HASH_FIELDS_SYMMETRIC_L4
1857 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4
1858 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP;
1861 /* Returns a hash value for the bits of 'flow' that are active based on
1862 * 'wc', given 'basis'. */
1864 flow_hash_in_wildcards(const struct flow *flow,
1865 const struct flow_wildcards *wc, uint32_t basis)
1867 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1868 const uint64_t *flow_u64 = (const uint64_t *) flow;
1873 for (i = 0; i < FLOW_U64S; i++) {
1874 hash = hash_add64(hash, flow_u64[i] & wc_u64[i]);
1876 return hash_finish(hash, 8 * FLOW_U64S);
1879 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1880 * OpenFlow 1.0 "dl_vlan" value:
1882 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
1883 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
1884 * 'flow' previously matched packets without a VLAN header).
1886 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
1887 * without a VLAN tag.
1889 * - Other values of 'vid' should not be used. */
1891 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
1893 if (vid == htons(OFP10_VLAN_NONE)) {
1894 flow->vlan_tci = htons(0);
1896 vid &= htons(VLAN_VID_MASK);
1897 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1898 flow->vlan_tci |= htons(VLAN_CFI) | vid;
1902 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1903 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
1906 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
1908 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
1909 flow->vlan_tci &= ~mask;
1910 flow->vlan_tci |= vid & mask;
1913 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
1916 * This function has no effect on the VLAN ID that 'flow' matches.
1918 * After calling this function, 'flow' will not match packets without a VLAN
1921 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
1924 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1925 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
1928 /* Returns the number of MPLS LSEs present in 'flow'
1930 * Returns 0 if the 'dl_type' of 'flow' is not an MPLS ethernet type.
1931 * Otherwise traverses 'flow''s MPLS label stack stopping at the
1932 * first entry that has the BoS bit set. If no such entry exists then
1933 * the maximum number of LSEs that can be stored in 'flow' is returned.
1936 flow_count_mpls_labels(const struct flow *flow, struct flow_wildcards *wc)
1938 /* dl_type is always masked. */
1939 if (eth_type_mpls(flow->dl_type)) {
1944 for (i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1946 wc->masks.mpls_lse[i] |= htonl(MPLS_BOS_MASK);
1948 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1951 if (flow->mpls_lse[i]) {
1961 /* Returns the number consecutive of MPLS LSEs, starting at the
1962 * innermost LSE, that are common in 'a' and 'b'.
1964 * 'an' must be flow_count_mpls_labels(a).
1965 * 'bn' must be flow_count_mpls_labels(b).
1968 flow_count_common_mpls_labels(const struct flow *a, int an,
1969 const struct flow *b, int bn,
1970 struct flow_wildcards *wc)
1972 int min_n = MIN(an, bn);
1977 int a_last = an - 1;
1978 int b_last = bn - 1;
1981 for (i = 0; i < min_n; i++) {
1983 wc->masks.mpls_lse[a_last - i] = OVS_BE32_MAX;
1984 wc->masks.mpls_lse[b_last - i] = OVS_BE32_MAX;
1986 if (a->mpls_lse[a_last - i] != b->mpls_lse[b_last - i]) {
1997 /* Adds a new outermost MPLS label to 'flow' and changes 'flow''s Ethernet type
1998 * to 'mpls_eth_type', which must be an MPLS Ethertype.
2000 * If the new label is the first MPLS label in 'flow', it is generated as;
2002 * - label: 2, if 'flow' is IPv6, otherwise 0.
2004 * - TTL: IPv4 or IPv6 TTL, if present and nonzero, otherwise 64.
2006 * - TC: IPv4 or IPv6 TOS, if present, otherwise 0.
2010 * If the new label is the second or later label MPLS label in 'flow', it is
2013 * - label: Copied from outer label.
2015 * - TTL: Copied from outer label.
2017 * - TC: Copied from outer label.
2021 * 'n' must be flow_count_mpls_labels(flow). 'n' must be less than
2022 * FLOW_MAX_MPLS_LABELS (because otherwise flow->mpls_lse[] would overflow).
2025 flow_push_mpls(struct flow *flow, int n, ovs_be16 mpls_eth_type,
2026 struct flow_wildcards *wc)
2028 ovs_assert(eth_type_mpls(mpls_eth_type));
2029 ovs_assert(n < FLOW_MAX_MPLS_LABELS);
2035 memset(&wc->masks.mpls_lse, 0xff, sizeof *wc->masks.mpls_lse * n);
2037 for (i = n; i >= 1; i--) {
2038 flow->mpls_lse[i] = flow->mpls_lse[i - 1];
2040 flow->mpls_lse[0] = (flow->mpls_lse[1] & htonl(~MPLS_BOS_MASK));
2042 int label = 0; /* IPv4 Explicit Null. */
2046 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2050 if (is_ip_any(flow)) {
2051 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
2053 wc->masks.nw_tos |= IP_DSCP_MASK;
2054 wc->masks.nw_ttl = 0xff;
2062 flow->mpls_lse[0] = set_mpls_lse_values(ttl, tc, 1, htonl(label));
2064 /* Clear all L3 and L4 fields and dp_hash. */
2065 BUILD_ASSERT(FLOW_WC_SEQ == 35);
2066 memset((char *) flow + FLOW_SEGMENT_2_ENDS_AT, 0,
2067 sizeof(struct flow) - FLOW_SEGMENT_2_ENDS_AT);
2070 flow->dl_type = mpls_eth_type;
2073 /* Tries to remove the outermost MPLS label from 'flow'. Returns true if
2074 * successful, false otherwise. On success, sets 'flow''s Ethernet type to
2077 * 'n' must be flow_count_mpls_labels(flow). */
2079 flow_pop_mpls(struct flow *flow, int n, ovs_be16 eth_type,
2080 struct flow_wildcards *wc)
2085 /* Nothing to pop. */
2087 } else if (n == FLOW_MAX_MPLS_LABELS) {
2089 wc->masks.mpls_lse[n - 1] |= htonl(MPLS_BOS_MASK);
2091 if (!(flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK))) {
2092 /* Can't pop because don't know what to fill in mpls_lse[n - 1]. */
2098 memset(&wc->masks.mpls_lse[1], 0xff,
2099 sizeof *wc->masks.mpls_lse * (n - 1));
2101 for (i = 1; i < n; i++) {
2102 flow->mpls_lse[i - 1] = flow->mpls_lse[i];
2104 flow->mpls_lse[n - 1] = 0;
2105 flow->dl_type = eth_type;
2109 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
2110 * as an OpenFlow 1.1 "mpls_label" value. */
2112 flow_set_mpls_label(struct flow *flow, int idx, ovs_be32 label)
2114 set_mpls_lse_label(&flow->mpls_lse[idx], label);
2117 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
2120 flow_set_mpls_ttl(struct flow *flow, int idx, uint8_t ttl)
2122 set_mpls_lse_ttl(&flow->mpls_lse[idx], ttl);
2125 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
2128 flow_set_mpls_tc(struct flow *flow, int idx, uint8_t tc)
2130 set_mpls_lse_tc(&flow->mpls_lse[idx], tc);
2133 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
2135 flow_set_mpls_bos(struct flow *flow, int idx, uint8_t bos)
2137 set_mpls_lse_bos(&flow->mpls_lse[idx], bos);
2140 /* Sets the entire MPLS LSE. */
2142 flow_set_mpls_lse(struct flow *flow, int idx, ovs_be32 lse)
2144 flow->mpls_lse[idx] = lse;
2148 flow_compose_l4(struct dp_packet *p, const struct flow *flow)
2152 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
2153 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
2154 if (flow->nw_proto == IPPROTO_TCP) {
2155 struct tcp_header *tcp;
2157 l4_len = sizeof *tcp;
2158 tcp = dp_packet_put_zeros(p, l4_len);
2159 tcp->tcp_src = flow->tp_src;
2160 tcp->tcp_dst = flow->tp_dst;
2161 tcp->tcp_ctl = TCP_CTL(ntohs(flow->tcp_flags), 5);
2162 } else if (flow->nw_proto == IPPROTO_UDP) {
2163 struct udp_header *udp;
2165 l4_len = sizeof *udp;
2166 udp = dp_packet_put_zeros(p, l4_len);
2167 udp->udp_src = flow->tp_src;
2168 udp->udp_dst = flow->tp_dst;
2169 } else if (flow->nw_proto == IPPROTO_SCTP) {
2170 struct sctp_header *sctp;
2172 l4_len = sizeof *sctp;
2173 sctp = dp_packet_put_zeros(p, l4_len);
2174 sctp->sctp_src = flow->tp_src;
2175 sctp->sctp_dst = flow->tp_dst;
2176 } else if (flow->nw_proto == IPPROTO_ICMP) {
2177 struct icmp_header *icmp;
2179 l4_len = sizeof *icmp;
2180 icmp = dp_packet_put_zeros(p, l4_len);
2181 icmp->icmp_type = ntohs(flow->tp_src);
2182 icmp->icmp_code = ntohs(flow->tp_dst);
2183 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
2184 } else if (flow->nw_proto == IPPROTO_IGMP) {
2185 struct igmp_header *igmp;
2187 l4_len = sizeof *igmp;
2188 igmp = dp_packet_put_zeros(p, l4_len);
2189 igmp->igmp_type = ntohs(flow->tp_src);
2190 igmp->igmp_code = ntohs(flow->tp_dst);
2191 put_16aligned_be32(&igmp->group, flow->igmp_group_ip4);
2192 igmp->igmp_csum = csum(igmp, IGMP_HEADER_LEN);
2193 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
2194 struct icmp6_hdr *icmp;
2196 l4_len = sizeof *icmp;
2197 icmp = dp_packet_put_zeros(p, l4_len);
2198 icmp->icmp6_type = ntohs(flow->tp_src);
2199 icmp->icmp6_code = ntohs(flow->tp_dst);
2201 if (icmp->icmp6_code == 0 &&
2202 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
2203 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
2204 struct in6_addr *nd_target;
2205 struct ovs_nd_opt *nd_opt;
2207 l4_len += sizeof *nd_target;
2208 nd_target = dp_packet_put_zeros(p, sizeof *nd_target);
2209 *nd_target = flow->nd_target;
2211 if (!eth_addr_is_zero(flow->arp_sha)) {
2213 nd_opt = dp_packet_put_zeros(p, 8);
2214 nd_opt->nd_opt_len = 1;
2215 nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
2216 nd_opt->nd_opt_mac = flow->arp_sha;
2218 if (!eth_addr_is_zero(flow->arp_tha)) {
2220 nd_opt = dp_packet_put_zeros(p, 8);
2221 nd_opt->nd_opt_len = 1;
2222 nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR;
2223 nd_opt->nd_opt_mac = flow->arp_tha;
2226 icmp->icmp6_cksum = (OVS_FORCE uint16_t)
2227 csum(icmp, (char *)dp_packet_tail(p) - (char *)icmp);
2233 /* Puts into 'b' a packet that flow_extract() would parse as having the given
2236 * (This is useful only for testing, obviously, and the packet isn't really
2237 * valid. It hasn't got some checksums filled in, for one, and lots of fields
2238 * are just zeroed.) */
2240 flow_compose(struct dp_packet *p, const struct flow *flow)
2244 /* eth_compose() sets l3 pointer and makes sure it is 32-bit aligned. */
2245 eth_compose(p, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
2246 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
2247 struct eth_header *eth = dp_packet_l2(p);
2248 eth->eth_type = htons(dp_packet_size(p));
2252 if (flow->vlan_tci & htons(VLAN_CFI)) {
2253 eth_push_vlan(p, htons(ETH_TYPE_VLAN), flow->vlan_tci);
2256 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2257 struct ip_header *ip;
2259 ip = dp_packet_put_zeros(p, sizeof *ip);
2260 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
2261 ip->ip_tos = flow->nw_tos;
2262 ip->ip_ttl = flow->nw_ttl;
2263 ip->ip_proto = flow->nw_proto;
2264 put_16aligned_be32(&ip->ip_src, flow->nw_src);
2265 put_16aligned_be32(&ip->ip_dst, flow->nw_dst);
2267 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2268 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
2269 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
2270 ip->ip_frag_off |= htons(100);
2274 dp_packet_set_l4(p, dp_packet_tail(p));
2276 l4_len = flow_compose_l4(p, flow);
2278 ip = dp_packet_l3(p);
2279 ip->ip_tot_len = htons(p->l4_ofs - p->l3_ofs + l4_len);
2280 ip->ip_csum = csum(ip, sizeof *ip);
2281 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2282 struct ovs_16aligned_ip6_hdr *nh;
2284 nh = dp_packet_put_zeros(p, sizeof *nh);
2285 put_16aligned_be32(&nh->ip6_flow, htonl(6 << 28) |
2286 htonl(flow->nw_tos << 20) | flow->ipv6_label);
2287 nh->ip6_hlim = flow->nw_ttl;
2288 nh->ip6_nxt = flow->nw_proto;
2290 memcpy(&nh->ip6_src, &flow->ipv6_src, sizeof(nh->ip6_src));
2291 memcpy(&nh->ip6_dst, &flow->ipv6_dst, sizeof(nh->ip6_dst));
2293 dp_packet_set_l4(p, dp_packet_tail(p));
2295 l4_len = flow_compose_l4(p, flow);
2297 nh = dp_packet_l3(p);
2298 nh->ip6_plen = htons(l4_len);
2299 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
2300 flow->dl_type == htons(ETH_TYPE_RARP)) {
2301 struct arp_eth_header *arp;
2303 arp = dp_packet_put_zeros(p, sizeof *arp);
2304 dp_packet_set_l3(p, arp);
2305 arp->ar_hrd = htons(1);
2306 arp->ar_pro = htons(ETH_TYPE_IP);
2307 arp->ar_hln = ETH_ADDR_LEN;
2309 arp->ar_op = htons(flow->nw_proto);
2311 if (flow->nw_proto == ARP_OP_REQUEST ||
2312 flow->nw_proto == ARP_OP_REPLY) {
2313 put_16aligned_be32(&arp->ar_spa, flow->nw_src);
2314 put_16aligned_be32(&arp->ar_tpa, flow->nw_dst);
2315 arp->ar_sha = flow->arp_sha;
2316 arp->ar_tha = flow->arp_tha;
2320 if (eth_type_mpls(flow->dl_type)) {
2323 p->l2_5_ofs = p->l3_ofs;
2324 for (n = 1; n < FLOW_MAX_MPLS_LABELS; n++) {
2325 if (flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK)) {
2330 push_mpls(p, flow->dl_type, flow->mpls_lse[--n]);
2335 /* Compressed flow. */
2337 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
2338 * the caller. The caller must have already computed 'dst->map' properly to
2339 * indicate the significant uint64_t elements of 'src'.
2341 * Normally the significant elements are the ones that are non-zero. However,
2342 * when a miniflow is initialized from a (mini)mask, the values can be zeroes,
2343 * so that the flow and mask always have the same maps. */
2345 miniflow_init(struct miniflow *dst, const struct flow *src)
2347 uint64_t *dst_u64 = miniflow_values(dst);
2350 FLOWMAP_FOR_EACH_INDEX(idx, dst->map) {
2351 *dst_u64++ = flow_u64_value(src, idx);
2355 /* Initialize the maps of 'flow' from 'src'. */
2357 miniflow_map_init(struct miniflow *flow, const struct flow *src)
2359 /* Initialize map, counting the number of nonzero elements. */
2360 flowmap_init(&flow->map);
2361 for (size_t i = 0; i < FLOW_U64S; i++) {
2362 if (flow_u64_value(src, i)) {
2363 flowmap_set(&flow->map, i, 1);
2368 /* Allocates 'n' count of miniflows, consecutive in memory, initializing the
2369 * map of each from 'src'.
2370 * Returns the size of the miniflow data. */
2372 miniflow_alloc(struct miniflow *dsts[], size_t n, const struct miniflow *src)
2374 size_t n_values = miniflow_n_values(src);
2375 size_t data_size = MINIFLOW_VALUES_SIZE(n_values);
2376 struct miniflow *dst = xmalloc(n * (sizeof *src + data_size));
2379 COVERAGE_INC(miniflow_malloc);
2381 for (i = 0; i < n; i++) {
2382 *dst = *src; /* Copy maps. */
2384 dst += 1; /* Just past the maps. */
2385 dst = (struct miniflow *)((uint64_t *)dst + n_values); /* Skip data. */
2390 /* Returns a miniflow copy of 'src'. The caller must eventually free() the
2391 * returned miniflow. */
2393 miniflow_create(const struct flow *src)
2395 struct miniflow tmp;
2396 struct miniflow *dst;
2398 miniflow_map_init(&tmp, src);
2400 miniflow_alloc(&dst, 1, &tmp);
2401 miniflow_init(dst, src);
2405 /* Initializes 'dst' as a copy of 'src'. The caller must have allocated
2406 * 'dst' to have inline space for 'n_values' data in 'src'. */
2408 miniflow_clone(struct miniflow *dst, const struct miniflow *src,
2411 *dst = *src; /* Copy maps. */
2412 memcpy(miniflow_values(dst), miniflow_get_values(src),
2413 MINIFLOW_VALUES_SIZE(n_values));
2416 /* Initializes 'dst' as a copy of 'src'. */
2418 miniflow_expand(const struct miniflow *src, struct flow *dst)
2420 memset(dst, 0, sizeof *dst);
2421 flow_union_with_miniflow(dst, src);
2424 /* Returns true if 'a' and 'b' are equal miniflows, false otherwise. */
2426 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
2428 const uint64_t *ap = miniflow_get_values(a);
2429 const uint64_t *bp = miniflow_get_values(b);
2431 /* This is mostly called after a matching hash, so it is highly likely that
2432 * the maps are equal as well. */
2433 if (OVS_LIKELY(flowmap_equal(a->map, b->map))) {
2434 return !memcmp(ap, bp, miniflow_n_values(a) * sizeof *ap);
2438 FLOWMAP_FOR_EACH_INDEX (idx, flowmap_or(a->map, b->map)) {
2439 if ((flowmap_is_set(&a->map, idx) ? *ap++ : 0)
2440 != (flowmap_is_set(&b->map, idx) ? *bp++ : 0)) {
2449 /* Returns false if 'a' and 'b' differ at the places where there are 1-bits
2450 * in 'mask', true otherwise. */
2452 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
2453 const struct minimask *mask)
2455 const uint64_t *p = miniflow_get_values(&mask->masks);
2458 FLOWMAP_FOR_EACH_INDEX(idx, mask->masks.map) {
2459 if ((miniflow_get(a, idx) ^ miniflow_get(b, idx)) & *p++) {
2467 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
2468 * in 'mask', false if they differ. */
2470 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
2471 const struct minimask *mask)
2473 const uint64_t *p = miniflow_get_values(&mask->masks);
2476 FLOWMAP_FOR_EACH_INDEX(idx, mask->masks.map) {
2477 if ((miniflow_get(a, idx) ^ flow_u64_value(b, idx)) & *p++) {
2487 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
2489 miniflow_init(&mask->masks, &wc->masks);
2492 /* Returns a minimask copy of 'wc'. The caller must eventually free the
2493 * returned minimask with free(). */
2495 minimask_create(const struct flow_wildcards *wc)
2497 return (struct minimask *)miniflow_create(&wc->masks);
2500 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
2502 * The caller must provide room for FLOW_U64S "uint64_t"s in 'storage', which
2503 * must follow '*dst_' in memory, for use by 'dst_'. The caller must *not*
2504 * free 'dst_' free(). */
2506 minimask_combine(struct minimask *dst_,
2507 const struct minimask *a_, const struct minimask *b_,
2508 uint64_t storage[FLOW_U64S])
2510 struct miniflow *dst = &dst_->masks;
2511 uint64_t *dst_values = storage;
2512 const struct miniflow *a = &a_->masks;
2513 const struct miniflow *b = &b_->masks;
2516 flowmap_init(&dst->map);
2518 FLOWMAP_FOR_EACH_INDEX(idx, flowmap_and(a->map, b->map)) {
2519 /* Both 'a' and 'b' have non-zero data at 'idx'. */
2520 uint64_t mask = *miniflow_get__(a, idx) & *miniflow_get__(b, idx);
2523 flowmap_set(&dst->map, idx, 1);
2524 *dst_values++ = mask;
2529 /* Initializes 'wc' as a copy of 'mask'. */
2531 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
2533 miniflow_expand(&mask->masks, &wc->masks);
2536 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise.
2537 * Minimasks may not have zero data values, so for the minimasks to be the
2538 * same, they need to have the same map and the same data values. */
2540 minimask_equal(const struct minimask *a, const struct minimask *b)
2542 return !memcmp(a, b, sizeof *a
2543 + MINIFLOW_VALUES_SIZE(miniflow_n_values(&a->masks)));
2546 /* Returns true if at least one bit matched by 'b' is wildcarded by 'a',
2547 * false otherwise. */
2549 minimask_has_extra(const struct minimask *a, const struct minimask *b)
2551 const uint64_t *bp = miniflow_get_values(&b->masks);
2554 FLOWMAP_FOR_EACH_INDEX(idx, b->masks.map) {
2555 uint64_t b_u64 = *bp++;
2557 /* 'b_u64' is non-zero, check if the data in 'a' is either zero
2558 * or misses some of the bits in 'b_u64'. */
2559 if (!MINIFLOW_IN_MAP(&a->masks, idx)
2560 || ((*miniflow_get__(&a->masks, idx) & b_u64) != b_u64)) {
2561 return true; /* 'a' wildcards some bits 'b' doesn't. */