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_pad_from_64_(MF, OFS) \
230 MINIFLOW_ASSERT(MF.data < MF.end); \
232 MINIFLOW_ASSERT((OFS) % 8 != 0); \
233 miniflow_set_map(MF, OFS / 8); \
235 memset((uint8_t *)MF.data, 0, (OFS) % 8); \
238 #define miniflow_push_be16_(MF, OFS, VALUE) \
239 miniflow_push_uint16_(MF, OFS, (OVS_FORCE uint16_t)VALUE);
241 #define miniflow_push_be8_(MF, OFS, VALUE) \
242 miniflow_push_uint8_(MF, OFS, (OVS_FORCE uint8_t)VALUE);
244 #define miniflow_set_maps(MF, OFS, N_WORDS) \
246 size_t ofs = (OFS); \
247 size_t n_words = (N_WORDS); \
249 MINIFLOW_ASSERT(n_words && MF.data + n_words <= MF.end); \
250 ASSERT_FLOWMAP_NOT_SET(&MF.map, ofs); \
251 flowmap_set(&MF.map, ofs, n_words); \
254 /* Data at 'valuep' may be unaligned. */
255 #define miniflow_push_words_(MF, OFS, VALUEP, N_WORDS) \
257 MINIFLOW_ASSERT((OFS) % 8 == 0); \
258 miniflow_set_maps(MF, (OFS) / 8, (N_WORDS)); \
259 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof *MF.data); \
260 MF.data += (N_WORDS); \
263 /* Push 32-bit words padded to 64-bits. */
264 #define miniflow_push_words_32_(MF, OFS, VALUEP, N_WORDS) \
266 miniflow_set_maps(MF, (OFS) / 8, DIV_ROUND_UP(N_WORDS, 2)); \
267 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof(uint32_t)); \
268 MF.data += DIV_ROUND_UP(N_WORDS, 2); \
269 if ((N_WORDS) & 1) { \
270 *((uint32_t *)MF.data - 1) = 0; \
274 /* Data at 'valuep' may be unaligned. */
275 /* MACs start 64-aligned, and must be followed by other data or padding. */
276 #define miniflow_push_macs_(MF, OFS, VALUEP) \
278 miniflow_set_maps(MF, (OFS) / 8, 2); \
279 memcpy(MF.data, (VALUEP), 2 * ETH_ADDR_LEN); \
280 MF.data += 1; /* First word only. */ \
283 #define miniflow_push_uint32(MF, FIELD, VALUE) \
284 miniflow_push_uint32_(MF, offsetof(struct flow, FIELD), VALUE)
286 #define miniflow_push_be32(MF, FIELD, VALUE) \
287 miniflow_push_be32_(MF, offsetof(struct flow, FIELD), VALUE)
289 #define miniflow_push_uint16(MF, FIELD, VALUE) \
290 miniflow_push_uint16_(MF, offsetof(struct flow, FIELD), VALUE)
292 #define miniflow_push_be16(MF, FIELD, VALUE) \
293 miniflow_push_be16_(MF, offsetof(struct flow, FIELD), VALUE)
295 #define miniflow_push_uint8(MF, FIELD, VALUE) \
296 miniflow_push_uint8_(MF, offsetof(struct flow, FIELD), VALUE)
298 #define miniflow_pad_to_64(MF, FIELD) \
299 miniflow_pad_to_64_(MF, OFFSETOFEND(struct flow, FIELD))
301 #define miniflow_pad_from_64(MF, FIELD) \
302 miniflow_pad_from_64_(MF, offsetof(struct flow, FIELD))
304 #define miniflow_push_words(MF, FIELD, VALUEP, N_WORDS) \
305 miniflow_push_words_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
307 #define miniflow_push_words_32(MF, FIELD, VALUEP, N_WORDS) \
308 miniflow_push_words_32_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
310 #define miniflow_push_macs(MF, FIELD, VALUEP) \
311 miniflow_push_macs_(MF, offsetof(struct flow, FIELD), VALUEP)
313 /* Pulls the MPLS headers at '*datap' and returns the count of them. */
315 parse_mpls(const void **datap, size_t *sizep)
317 const struct mpls_hdr *mh;
320 while ((mh = data_try_pull(datap, sizep, sizeof *mh))) {
322 if (mh->mpls_lse.lo & htons(1 << MPLS_BOS_SHIFT)) {
326 return MIN(count, FLOW_MAX_MPLS_LABELS);
329 static inline ovs_be16
330 parse_vlan(const void **datap, size_t *sizep)
332 const struct eth_header *eth = *datap;
335 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
339 data_pull(datap, sizep, ETH_ADDR_LEN * 2);
341 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
342 if (OVS_LIKELY(*sizep
343 >= sizeof(struct qtag_prefix) + sizeof(ovs_be16))) {
344 const struct qtag_prefix *qp = data_pull(datap, sizep, sizeof *qp);
345 return qp->tci | htons(VLAN_CFI);
351 static inline ovs_be16
352 parse_ethertype(const void **datap, size_t *sizep)
354 const struct llc_snap_header *llc;
357 proto = *(ovs_be16 *) data_pull(datap, sizep, sizeof proto);
358 if (OVS_LIKELY(ntohs(proto) >= ETH_TYPE_MIN)) {
362 if (OVS_UNLIKELY(*sizep < sizeof *llc)) {
363 return htons(FLOW_DL_TYPE_NONE);
367 if (OVS_UNLIKELY(llc->llc.llc_dsap != LLC_DSAP_SNAP
368 || llc->llc.llc_ssap != LLC_SSAP_SNAP
369 || llc->llc.llc_cntl != LLC_CNTL_SNAP
370 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
371 sizeof llc->snap.snap_org))) {
372 return htons(FLOW_DL_TYPE_NONE);
375 data_pull(datap, sizep, sizeof *llc);
377 if (OVS_LIKELY(ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN)) {
378 return llc->snap.snap_type;
381 return htons(FLOW_DL_TYPE_NONE);
385 parse_icmpv6(const void **datap, size_t *sizep, const struct icmp6_hdr *icmp,
386 const struct in6_addr **nd_target,
387 struct eth_addr arp_buf[2])
389 if (icmp->icmp6_code == 0 &&
390 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
391 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
393 *nd_target = data_try_pull(datap, sizep, sizeof **nd_target);
394 if (OVS_UNLIKELY(!*nd_target)) {
398 while (*sizep >= 8) {
399 /* The minimum size of an option is 8 bytes, which also is
400 * the size of Ethernet link-layer options. */
401 const struct ovs_nd_opt *nd_opt = *datap;
402 int opt_len = nd_opt->nd_opt_len * ND_OPT_LEN;
404 if (!opt_len || opt_len > *sizep) {
408 /* Store the link layer address if the appropriate option is
409 * provided. It is considered an error if the same link
410 * layer option is specified twice. */
411 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
413 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[0]))) {
414 arp_buf[0] = nd_opt->nd_opt_mac;
418 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
420 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[1]))) {
421 arp_buf[1] = nd_opt->nd_opt_mac;
427 if (OVS_UNLIKELY(!data_try_pull(datap, sizep, opt_len))) {
437 arp_buf[0] = eth_addr_zero;
438 arp_buf[1] = eth_addr_zero;
441 /* Initializes 'flow' members from 'packet' and 'md'
443 * Initializes 'packet' header l2 pointer to the start of the Ethernet
444 * header, and the layer offsets as follows:
446 * - packet->l2_5_ofs to the start of the MPLS shim header, or UINT16_MAX
447 * when there is no MPLS shim header.
449 * - packet->l3_ofs to just past the Ethernet header, or just past the
450 * vlan_header if one is present, to the first byte of the payload of the
451 * Ethernet frame. UINT16_MAX if the frame is too short to contain an
454 * - packet->l4_ofs to just past the IPv4 header, if one is present and
455 * has at least the content used for the fields of interest for the flow,
456 * otherwise UINT16_MAX.
459 flow_extract(struct dp_packet *packet, struct flow *flow)
463 uint64_t buf[FLOW_U64S];
466 COVERAGE_INC(flow_extract);
468 miniflow_extract(packet, &m.mf);
469 miniflow_expand(&m.mf, flow);
472 /* Caller is responsible for initializing 'dst' with enough storage for
473 * FLOW_U64S * 8 bytes. */
475 miniflow_extract(struct dp_packet *packet, struct miniflow *dst)
477 const struct pkt_metadata *md = &packet->md;
478 const void *data = dp_packet_data(packet);
479 size_t size = dp_packet_size(packet);
480 uint64_t *values = miniflow_values(dst);
481 struct mf_ctx mf = { FLOWMAP_EMPTY_INITIALIZER, values,
482 values + FLOW_U64S };
485 uint8_t nw_frag, nw_tos, nw_ttl, nw_proto;
488 if (flow_tnl_dst_is_set(&md->tunnel)) {
489 miniflow_push_words(mf, tunnel, &md->tunnel,
490 offsetof(struct flow_tnl, metadata) /
493 if (!(md->tunnel.flags & FLOW_TNL_F_UDPIF)) {
494 if (md->tunnel.metadata.present.map) {
495 miniflow_push_words(mf, tunnel.metadata, &md->tunnel.metadata,
496 sizeof md->tunnel.metadata /
500 if (md->tunnel.metadata.present.len) {
501 miniflow_push_words(mf, tunnel.metadata.present,
502 &md->tunnel.metadata.present, 1);
503 miniflow_push_words(mf, tunnel.metadata.opts.gnv,
504 md->tunnel.metadata.opts.gnv,
505 DIV_ROUND_UP(md->tunnel.metadata.present.len,
510 if (md->skb_priority || md->pkt_mark) {
511 miniflow_push_uint32(mf, skb_priority, md->skb_priority);
512 miniflow_push_uint32(mf, pkt_mark, md->pkt_mark);
514 miniflow_push_uint32(mf, dp_hash, md->dp_hash);
515 miniflow_push_uint32(mf, in_port, odp_to_u32(md->in_port.odp_port));
516 if (md->recirc_id || md->ct_state) {
517 miniflow_push_uint32(mf, recirc_id, md->recirc_id);
518 miniflow_push_uint16(mf, ct_state, md->ct_state);
519 miniflow_push_uint16(mf, ct_zone, md->ct_zone);
523 miniflow_push_uint32(mf, ct_mark, md->ct_mark);
524 miniflow_pad_to_64(mf, ct_mark);
526 if (!ovs_u128_is_zero(&md->ct_label)) {
527 miniflow_push_words(mf, ct_label, &md->ct_label,
528 sizeof md->ct_label / sizeof(uint64_t));
532 /* Initialize packet's layer pointer and offsets. */
534 dp_packet_reset_offsets(packet);
536 /* Must have full Ethernet header to proceed. */
537 if (OVS_UNLIKELY(size < sizeof(struct eth_header))) {
543 ASSERT_SEQUENTIAL(dl_dst, dl_src);
544 miniflow_push_macs(mf, dl_dst, data);
545 /* dl_type, vlan_tci. */
546 vlan_tci = parse_vlan(&data, &size);
547 dl_type = parse_ethertype(&data, &size);
548 miniflow_push_be16(mf, dl_type, dl_type);
549 miniflow_push_be16(mf, vlan_tci, vlan_tci);
553 if (OVS_UNLIKELY(eth_type_mpls(dl_type))) {
555 const void *mpls = data;
557 packet->l2_5_ofs = (char *)data - l2;
558 count = parse_mpls(&data, &size);
559 miniflow_push_words_32(mf, mpls_lse, mpls, count);
563 packet->l3_ofs = (char *)data - l2;
566 if (OVS_LIKELY(dl_type == htons(ETH_TYPE_IP))) {
567 const struct ip_header *nh = data;
571 if (OVS_UNLIKELY(size < IP_HEADER_LEN)) {
574 ip_len = IP_IHL(nh->ip_ihl_ver) * 4;
576 if (OVS_UNLIKELY(ip_len < IP_HEADER_LEN)) {
579 if (OVS_UNLIKELY(size < ip_len)) {
582 tot_len = ntohs(nh->ip_tot_len);
583 if (OVS_UNLIKELY(tot_len > size)) {
586 if (OVS_UNLIKELY(size - tot_len > UINT8_MAX)) {
589 dp_packet_set_l2_pad_size(packet, size - tot_len);
590 size = tot_len; /* Never pull padding. */
592 /* Push both source and destination address at once. */
593 miniflow_push_words(mf, nw_src, &nh->ip_src, 1);
595 miniflow_push_be32(mf, ipv6_label, 0); /* Padding for IPv4. */
599 nw_proto = nh->ip_proto;
600 if (OVS_UNLIKELY(IP_IS_FRAGMENT(nh->ip_frag_off))) {
601 nw_frag = FLOW_NW_FRAG_ANY;
602 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
603 nw_frag |= FLOW_NW_FRAG_LATER;
606 data_pull(&data, &size, ip_len);
607 } else if (dl_type == htons(ETH_TYPE_IPV6)) {
608 const struct ovs_16aligned_ip6_hdr *nh;
612 if (OVS_UNLIKELY(size < sizeof *nh)) {
615 nh = data_pull(&data, &size, sizeof *nh);
617 plen = ntohs(nh->ip6_plen);
618 if (OVS_UNLIKELY(plen > size)) {
621 /* Jumbo Payload option not supported yet. */
622 if (OVS_UNLIKELY(size - plen > UINT8_MAX)) {
625 dp_packet_set_l2_pad_size(packet, size - plen);
626 size = plen; /* Never pull padding. */
628 miniflow_push_words(mf, ipv6_src, &nh->ip6_src,
629 sizeof nh->ip6_src / 8);
630 miniflow_push_words(mf, ipv6_dst, &nh->ip6_dst,
631 sizeof nh->ip6_dst / 8);
633 tc_flow = get_16aligned_be32(&nh->ip6_flow);
635 ovs_be32 label = tc_flow & htonl(IPV6_LABEL_MASK);
636 miniflow_push_be32(mf, ipv6_label, label);
639 nw_tos = ntohl(tc_flow) >> 20;
640 nw_ttl = nh->ip6_hlim;
641 nw_proto = nh->ip6_nxt;
644 if (OVS_LIKELY((nw_proto != IPPROTO_HOPOPTS)
645 && (nw_proto != IPPROTO_ROUTING)
646 && (nw_proto != IPPROTO_DSTOPTS)
647 && (nw_proto != IPPROTO_AH)
648 && (nw_proto != IPPROTO_FRAGMENT))) {
649 /* It's either a terminal header (e.g., TCP, UDP) or one we
650 * don't understand. In either case, we're done with the
651 * packet, so use it to fill in 'nw_proto'. */
655 /* We only verify that at least 8 bytes of the next header are
656 * available, but many of these headers are longer. Ensure that
657 * accesses within the extension header are within those first 8
658 * bytes. All extension headers are required to be at least 8
660 if (OVS_UNLIKELY(size < 8)) {
664 if ((nw_proto == IPPROTO_HOPOPTS)
665 || (nw_proto == IPPROTO_ROUTING)
666 || (nw_proto == IPPROTO_DSTOPTS)) {
667 /* These headers, while different, have the fields we care
668 * about in the same location and with the same
670 const struct ip6_ext *ext_hdr = data;
671 nw_proto = ext_hdr->ip6e_nxt;
672 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
673 (ext_hdr->ip6e_len + 1) * 8))) {
676 } else if (nw_proto == IPPROTO_AH) {
677 /* A standard AH definition isn't available, but the fields
678 * we care about are in the same location as the generic
679 * option header--only the header length is calculated
681 const struct ip6_ext *ext_hdr = data;
682 nw_proto = ext_hdr->ip6e_nxt;
683 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
684 (ext_hdr->ip6e_len + 2) * 4))) {
687 } else if (nw_proto == IPPROTO_FRAGMENT) {
688 const struct ovs_16aligned_ip6_frag *frag_hdr = data;
690 nw_proto = frag_hdr->ip6f_nxt;
691 if (!data_try_pull(&data, &size, sizeof *frag_hdr)) {
695 /* We only process the first fragment. */
696 if (frag_hdr->ip6f_offlg != htons(0)) {
697 nw_frag = FLOW_NW_FRAG_ANY;
698 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
699 nw_frag |= FLOW_NW_FRAG_LATER;
700 nw_proto = IPPROTO_FRAGMENT;
707 if (dl_type == htons(ETH_TYPE_ARP) ||
708 dl_type == htons(ETH_TYPE_RARP)) {
709 struct eth_addr arp_buf[2];
710 const struct arp_eth_header *arp = (const struct arp_eth_header *)
711 data_try_pull(&data, &size, ARP_ETH_HEADER_LEN);
713 if (OVS_LIKELY(arp) && OVS_LIKELY(arp->ar_hrd == htons(1))
714 && OVS_LIKELY(arp->ar_pro == htons(ETH_TYPE_IP))
715 && OVS_LIKELY(arp->ar_hln == ETH_ADDR_LEN)
716 && OVS_LIKELY(arp->ar_pln == 4)) {
717 miniflow_push_be32(mf, nw_src,
718 get_16aligned_be32(&arp->ar_spa));
719 miniflow_push_be32(mf, nw_dst,
720 get_16aligned_be32(&arp->ar_tpa));
722 /* We only match on the lower 8 bits of the opcode. */
723 if (OVS_LIKELY(ntohs(arp->ar_op) <= 0xff)) {
724 miniflow_push_be32(mf, ipv6_label, 0); /* Pad with ARP. */
725 miniflow_push_be32(mf, nw_frag, htonl(ntohs(arp->ar_op)));
728 /* Must be adjacent. */
729 ASSERT_SEQUENTIAL(arp_sha, arp_tha);
731 arp_buf[0] = arp->ar_sha;
732 arp_buf[1] = arp->ar_tha;
733 miniflow_push_macs(mf, arp_sha, arp_buf);
734 miniflow_pad_to_64(mf, arp_tha);
740 packet->l4_ofs = (char *)data - l2;
741 miniflow_push_be32(mf, nw_frag,
742 BYTES_TO_BE32(nw_frag, nw_tos, nw_ttl, nw_proto));
744 if (OVS_LIKELY(!(nw_frag & FLOW_NW_FRAG_LATER))) {
745 if (OVS_LIKELY(nw_proto == IPPROTO_TCP)) {
746 if (OVS_LIKELY(size >= TCP_HEADER_LEN)) {
747 const struct tcp_header *tcp = data;
749 miniflow_push_be32(mf, arp_tha.ea[2], 0);
750 miniflow_push_be32(mf, tcp_flags,
751 TCP_FLAGS_BE32(tcp->tcp_ctl));
752 miniflow_push_be16(mf, tp_src, tcp->tcp_src);
753 miniflow_push_be16(mf, tp_dst, tcp->tcp_dst);
754 miniflow_pad_to_64(mf, tp_dst);
756 } else if (OVS_LIKELY(nw_proto == IPPROTO_UDP)) {
757 if (OVS_LIKELY(size >= UDP_HEADER_LEN)) {
758 const struct udp_header *udp = data;
760 miniflow_push_be16(mf, tp_src, udp->udp_src);
761 miniflow_push_be16(mf, tp_dst, udp->udp_dst);
762 miniflow_pad_to_64(mf, tp_dst);
764 } else if (OVS_LIKELY(nw_proto == IPPROTO_SCTP)) {
765 if (OVS_LIKELY(size >= SCTP_HEADER_LEN)) {
766 const struct sctp_header *sctp = data;
768 miniflow_push_be16(mf, tp_src, sctp->sctp_src);
769 miniflow_push_be16(mf, tp_dst, sctp->sctp_dst);
770 miniflow_pad_to_64(mf, tp_dst);
772 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMP)) {
773 if (OVS_LIKELY(size >= ICMP_HEADER_LEN)) {
774 const struct icmp_header *icmp = data;
776 miniflow_push_be16(mf, tp_src, htons(icmp->icmp_type));
777 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp_code));
778 miniflow_pad_to_64(mf, tp_dst);
780 } else if (OVS_LIKELY(nw_proto == IPPROTO_IGMP)) {
781 if (OVS_LIKELY(size >= IGMP_HEADER_LEN)) {
782 const struct igmp_header *igmp = data;
784 miniflow_push_be16(mf, tp_src, htons(igmp->igmp_type));
785 miniflow_push_be16(mf, tp_dst, htons(igmp->igmp_code));
786 miniflow_push_be32(mf, igmp_group_ip4,
787 get_16aligned_be32(&igmp->group));
789 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMPV6)) {
790 if (OVS_LIKELY(size >= sizeof(struct icmp6_hdr))) {
791 const struct in6_addr *nd_target = NULL;
792 struct eth_addr arp_buf[2] = { { { { 0 } } } };
793 const struct icmp6_hdr *icmp = data_pull(&data, &size,
795 parse_icmpv6(&data, &size, icmp, &nd_target, arp_buf);
797 miniflow_push_words(mf, nd_target, nd_target,
798 sizeof *nd_target / sizeof(uint64_t));
800 miniflow_push_macs(mf, arp_sha, arp_buf);
801 miniflow_pad_to_64(mf, arp_tha);
802 miniflow_push_be16(mf, tp_src, htons(icmp->icmp6_type));
803 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp6_code));
804 miniflow_pad_to_64(mf, tp_dst);
812 /* For every bit of a field that is wildcarded in 'wildcards', sets the
813 * corresponding bit in 'flow' to zero. */
815 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
817 uint64_t *flow_u64 = (uint64_t *) flow;
818 const uint64_t *wc_u64 = (const uint64_t *) &wildcards->masks;
821 for (i = 0; i < FLOW_U64S; i++) {
822 flow_u64[i] &= wc_u64[i];
827 flow_unwildcard_tp_ports(const struct flow *flow, struct flow_wildcards *wc)
829 if (flow->nw_proto != IPPROTO_ICMP) {
830 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
831 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
833 wc->masks.tp_src = htons(0xff);
834 wc->masks.tp_dst = htons(0xff);
838 /* Initializes 'flow_metadata' with the metadata found in 'flow'. */
840 flow_get_metadata(const struct flow *flow, struct match *flow_metadata)
844 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 35);
846 match_init_catchall(flow_metadata);
847 if (flow->tunnel.tun_id != htonll(0)) {
848 match_set_tun_id(flow_metadata, flow->tunnel.tun_id);
850 if (flow->tunnel.flags & FLOW_TNL_PUB_F_MASK) {
851 match_set_tun_flags(flow_metadata,
852 flow->tunnel.flags & FLOW_TNL_PUB_F_MASK);
854 if (flow->tunnel.ip_src) {
855 match_set_tun_src(flow_metadata, flow->tunnel.ip_src);
857 if (flow->tunnel.ip_dst) {
858 match_set_tun_dst(flow_metadata, flow->tunnel.ip_dst);
860 if (ipv6_addr_is_set(&flow->tunnel.ipv6_src)) {
861 match_set_tun_ipv6_src(flow_metadata, &flow->tunnel.ipv6_src);
863 if (ipv6_addr_is_set(&flow->tunnel.ipv6_dst)) {
864 match_set_tun_ipv6_dst(flow_metadata, &flow->tunnel.ipv6_dst);
866 if (flow->tunnel.gbp_id != htons(0)) {
867 match_set_tun_gbp_id(flow_metadata, flow->tunnel.gbp_id);
869 if (flow->tunnel.gbp_flags) {
870 match_set_tun_gbp_flags(flow_metadata, flow->tunnel.gbp_flags);
872 tun_metadata_get_fmd(&flow->tunnel, flow_metadata);
873 if (flow->metadata != htonll(0)) {
874 match_set_metadata(flow_metadata, flow->metadata);
877 for (i = 0; i < FLOW_N_REGS; i++) {
879 match_set_reg(flow_metadata, i, flow->regs[i]);
883 if (flow->pkt_mark != 0) {
884 match_set_pkt_mark(flow_metadata, flow->pkt_mark);
887 match_set_in_port(flow_metadata, flow->in_port.ofp_port);
888 if (flow->ct_state != 0) {
889 match_set_ct_state(flow_metadata, flow->ct_state);
891 if (flow->ct_zone != 0) {
892 match_set_ct_zone(flow_metadata, flow->ct_zone);
894 if (flow->ct_mark != 0) {
895 match_set_ct_mark(flow_metadata, flow->ct_mark);
897 if (!ovs_u128_is_zero(&flow->ct_label)) {
898 match_set_ct_label(flow_metadata, flow->ct_label);
902 const char *ct_state_to_string(uint32_t state)
927 flow_to_string(const struct flow *flow)
929 struct ds ds = DS_EMPTY_INITIALIZER;
930 flow_format(&ds, flow);
935 flow_tun_flag_to_string(uint32_t flags)
938 case FLOW_TNL_F_DONT_FRAGMENT:
940 case FLOW_TNL_F_CSUM:
952 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
953 uint32_t flags, char del)
958 ds_put_char(ds, '0');
962 uint32_t bit = rightmost_1bit(flags);
965 s = bit_to_string(bit);
967 ds_put_format(ds, "%s%c", s, del);
976 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
982 format_flags_masked(struct ds *ds, const char *name,
983 const char *(*bit_to_string)(uint32_t), uint32_t flags,
984 uint32_t mask, uint32_t max_mask)
987 ds_put_format(ds, "%s=", name);
990 if (mask == max_mask) {
991 format_flags(ds, bit_to_string, flags, '|');
996 ds_put_cstr(ds, "0/0");
1001 uint32_t bit = rightmost_1bit(mask);
1002 const char *s = bit_to_string(bit);
1004 ds_put_format(ds, "%s%s", (flags & bit) ? "+" : "-",
1005 s ? s : "[Unknown]");
1010 /* Scans a string 's' of flags to determine their numerical value and
1011 * returns the number of characters parsed using 'bit_to_string' to
1012 * lookup flag names. Scanning continues until the character 'end' is
1015 * In the event of a failure, a negative error code will be returned. In
1016 * addition, if 'res_string' is non-NULL then a descriptive string will
1017 * be returned incorporating the identifying string 'field_name'. This
1018 * error string must be freed by the caller.
1020 * Upon success, the flag values will be stored in 'res_flags' and
1021 * optionally 'res_mask', if it is non-NULL (if it is NULL then any masks
1022 * present in the original string will be considered an error). The
1023 * caller may restrict the acceptable set of values through the mask
1026 parse_flags(const char *s, const char *(*bit_to_string)(uint32_t),
1027 char end, const char *field_name, char **res_string,
1028 uint32_t *res_flags, uint32_t allowed, uint32_t *res_mask)
1030 uint32_t result = 0;
1033 /* Parse masked flags in numeric format? */
1034 if (res_mask && ovs_scan(s, "%"SCNi32"/%"SCNi32"%n",
1035 res_flags, res_mask, &n) && n > 0) {
1036 if (*res_flags & ~allowed || *res_mask & ~allowed) {
1044 if (res_mask && (*s == '+' || *s == '-')) {
1045 uint32_t flags = 0, mask = 0;
1047 /* Parse masked flags. */
1048 while (s[0] != end) {
1055 } else if (s[0] == '-') {
1059 *res_string = xasprintf("%s: %s must be preceded by '+' "
1060 "(for SET) or '-' (NOT SET)", s,
1068 for (bit = 1; bit; bit <<= 1) {
1069 const char *fname = bit_to_string(bit);
1075 len = strlen(fname);
1076 if (strncmp(s, fname, len) ||
1077 (s[len] != '+' && s[len] != '-' && s[len] != end)) {
1082 /* bit already set. */
1084 *res_string = xasprintf("%s: Each %s flag can be "
1085 "specified only once", s,
1090 if (!(bit & allowed)) {
1112 /* Parse unmasked flags. If a flag is present, it is set, otherwise
1114 while (s[n] != end) {
1115 unsigned long long int flags;
1119 if (ovs_scan(&s[n], "%lli%n", &flags, &n0)) {
1120 if (flags & ~allowed) {
1123 n += n0 + (s[n + n0] == '|');
1128 for (bit = 1; bit; bit <<= 1) {
1129 const char *name = bit_to_string(bit);
1137 if (!strncmp(s + n, name, len) &&
1138 (s[n + len] == '|' || s[n + len] == end)) {
1139 if (!(bit & allowed)) {
1143 n += len + (s[n + len] == '|');
1153 *res_flags = result;
1155 *res_mask = UINT32_MAX;
1164 *res_string = xasprintf("%s: unknown %s flag(s)", s, field_name);
1170 flow_format(struct ds *ds, const struct flow *flow)
1173 struct flow_wildcards *wc = &match.wc;
1175 match_wc_init(&match, flow);
1177 /* As this function is most often used for formatting a packet in a
1178 * packet-in message, skip formatting the packet context fields that are
1179 * all-zeroes to make the print-out easier on the eyes. This means that a
1180 * missing context field implies a zero value for that field. This is
1181 * similar to OpenFlow encoding of these fields, as the specification
1182 * states that all-zeroes context fields should not be encoded in the
1183 * packet-in messages. */
1184 if (!flow->in_port.ofp_port) {
1185 WC_UNMASK_FIELD(wc, in_port);
1187 if (!flow->skb_priority) {
1188 WC_UNMASK_FIELD(wc, skb_priority);
1190 if (!flow->pkt_mark) {
1191 WC_UNMASK_FIELD(wc, pkt_mark);
1193 if (!flow->recirc_id) {
1194 WC_UNMASK_FIELD(wc, recirc_id);
1196 if (!flow->dp_hash) {
1197 WC_UNMASK_FIELD(wc, dp_hash);
1199 if (!flow->ct_state) {
1200 WC_UNMASK_FIELD(wc, ct_state);
1202 if (!flow->ct_zone) {
1203 WC_UNMASK_FIELD(wc, ct_zone);
1205 if (!flow->ct_mark) {
1206 WC_UNMASK_FIELD(wc, ct_mark);
1208 if (ovs_u128_is_zero(&flow->ct_label)) {
1209 WC_UNMASK_FIELD(wc, ct_label);
1211 for (int i = 0; i < FLOW_N_REGS; i++) {
1212 if (!flow->regs[i]) {
1213 WC_UNMASK_FIELD(wc, regs[i]);
1216 if (!flow->metadata) {
1217 WC_UNMASK_FIELD(wc, metadata);
1220 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
1224 flow_print(FILE *stream, const struct flow *flow)
1226 char *s = flow_to_string(flow);
1231 /* flow_wildcards functions. */
1233 /* Initializes 'wc' as a set of wildcards that matches every packet. */
1235 flow_wildcards_init_catchall(struct flow_wildcards *wc)
1237 memset(&wc->masks, 0, sizeof wc->masks);
1240 /* Converts a flow into flow wildcards. It sets the wildcard masks based on
1241 * the packet headers extracted to 'flow'. It will not set the mask for fields
1242 * that do not make sense for the packet type. OpenFlow-only metadata is
1243 * wildcarded, but other metadata is unconditionally exact-matched. */
1244 void flow_wildcards_init_for_packet(struct flow_wildcards *wc,
1245 const struct flow *flow)
1247 memset(&wc->masks, 0x0, sizeof wc->masks);
1249 /* Update this function whenever struct flow changes. */
1250 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 35);
1252 if (flow_tnl_dst_is_set(&flow->tunnel)) {
1253 if (flow->tunnel.flags & FLOW_TNL_F_KEY) {
1254 WC_MASK_FIELD(wc, tunnel.tun_id);
1256 WC_MASK_FIELD(wc, tunnel.ip_src);
1257 WC_MASK_FIELD(wc, tunnel.ip_dst);
1258 WC_MASK_FIELD(wc, tunnel.ipv6_src);
1259 WC_MASK_FIELD(wc, tunnel.ipv6_dst);
1260 WC_MASK_FIELD(wc, tunnel.flags);
1261 WC_MASK_FIELD(wc, tunnel.ip_tos);
1262 WC_MASK_FIELD(wc, tunnel.ip_ttl);
1263 WC_MASK_FIELD(wc, tunnel.tp_src);
1264 WC_MASK_FIELD(wc, tunnel.tp_dst);
1265 WC_MASK_FIELD(wc, tunnel.gbp_id);
1266 WC_MASK_FIELD(wc, tunnel.gbp_flags);
1268 if (!(flow->tunnel.flags & FLOW_TNL_F_UDPIF)) {
1269 if (flow->tunnel.metadata.present.map) {
1270 wc->masks.tunnel.metadata.present.map =
1271 flow->tunnel.metadata.present.map;
1272 WC_MASK_FIELD(wc, tunnel.metadata.opts.u8);
1275 WC_MASK_FIELD(wc, tunnel.metadata.present.len);
1276 memset(wc->masks.tunnel.metadata.opts.gnv, 0xff,
1277 flow->tunnel.metadata.present.len);
1279 } else if (flow->tunnel.tun_id) {
1280 WC_MASK_FIELD(wc, tunnel.tun_id);
1283 /* metadata, regs, and conj_id wildcarded. */
1285 WC_MASK_FIELD(wc, skb_priority);
1286 WC_MASK_FIELD(wc, pkt_mark);
1287 WC_MASK_FIELD(wc, ct_state);
1288 WC_MASK_FIELD(wc, ct_zone);
1289 WC_MASK_FIELD(wc, ct_mark);
1290 WC_MASK_FIELD(wc, ct_label);
1291 WC_MASK_FIELD(wc, recirc_id);
1292 WC_MASK_FIELD(wc, dp_hash);
1293 WC_MASK_FIELD(wc, in_port);
1295 /* actset_output wildcarded. */
1297 WC_MASK_FIELD(wc, dl_dst);
1298 WC_MASK_FIELD(wc, dl_src);
1299 WC_MASK_FIELD(wc, dl_type);
1300 WC_MASK_FIELD(wc, vlan_tci);
1302 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1303 WC_MASK_FIELD(wc, nw_src);
1304 WC_MASK_FIELD(wc, nw_dst);
1305 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1306 WC_MASK_FIELD(wc, ipv6_src);
1307 WC_MASK_FIELD(wc, ipv6_dst);
1308 WC_MASK_FIELD(wc, ipv6_label);
1309 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1310 flow->dl_type == htons(ETH_TYPE_RARP)) {
1311 WC_MASK_FIELD(wc, nw_src);
1312 WC_MASK_FIELD(wc, nw_dst);
1313 WC_MASK_FIELD(wc, nw_proto);
1314 WC_MASK_FIELD(wc, arp_sha);
1315 WC_MASK_FIELD(wc, arp_tha);
1317 } else if (eth_type_mpls(flow->dl_type)) {
1318 for (int i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1319 WC_MASK_FIELD(wc, mpls_lse[i]);
1320 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1326 return; /* Unknown ethertype. */
1330 WC_MASK_FIELD(wc, nw_frag);
1331 WC_MASK_FIELD(wc, nw_tos);
1332 WC_MASK_FIELD(wc, nw_ttl);
1333 WC_MASK_FIELD(wc, nw_proto);
1335 /* No transport layer header in later fragments. */
1336 if (!(flow->nw_frag & FLOW_NW_FRAG_LATER) &&
1337 (flow->nw_proto == IPPROTO_ICMP ||
1338 flow->nw_proto == IPPROTO_ICMPV6 ||
1339 flow->nw_proto == IPPROTO_TCP ||
1340 flow->nw_proto == IPPROTO_UDP ||
1341 flow->nw_proto == IPPROTO_SCTP ||
1342 flow->nw_proto == IPPROTO_IGMP)) {
1343 WC_MASK_FIELD(wc, tp_src);
1344 WC_MASK_FIELD(wc, tp_dst);
1346 if (flow->nw_proto == IPPROTO_TCP) {
1347 WC_MASK_FIELD(wc, tcp_flags);
1348 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
1349 WC_MASK_FIELD(wc, arp_sha);
1350 WC_MASK_FIELD(wc, arp_tha);
1351 WC_MASK_FIELD(wc, nd_target);
1352 } else if (flow->nw_proto == IPPROTO_IGMP) {
1353 WC_MASK_FIELD(wc, igmp_group_ip4);
1358 /* Return a map of possible fields for a packet of the same type as 'flow'.
1359 * Including extra bits in the returned mask is not wrong, it is just less
1362 * This is a less precise version of flow_wildcards_init_for_packet() above. */
1364 flow_wc_map(const struct flow *flow, struct flowmap *map)
1366 /* Update this function whenever struct flow changes. */
1367 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 35);
1371 if (flow_tnl_dst_is_set(&flow->tunnel)) {
1372 FLOWMAP_SET__(map, tunnel, offsetof(struct flow_tnl, metadata));
1373 if (!(flow->tunnel.flags & FLOW_TNL_F_UDPIF)) {
1374 if (flow->tunnel.metadata.present.map) {
1375 FLOWMAP_SET(map, tunnel.metadata);
1378 FLOWMAP_SET(map, tunnel.metadata.present.len);
1379 FLOWMAP_SET__(map, tunnel.metadata.opts.gnv,
1380 flow->tunnel.metadata.present.len);
1384 /* Metadata fields that can appear on packet input. */
1385 FLOWMAP_SET(map, skb_priority);
1386 FLOWMAP_SET(map, pkt_mark);
1387 FLOWMAP_SET(map, recirc_id);
1388 FLOWMAP_SET(map, dp_hash);
1389 FLOWMAP_SET(map, in_port);
1390 FLOWMAP_SET(map, dl_dst);
1391 FLOWMAP_SET(map, dl_src);
1392 FLOWMAP_SET(map, dl_type);
1393 FLOWMAP_SET(map, vlan_tci);
1394 FLOWMAP_SET(map, ct_state);
1395 FLOWMAP_SET(map, ct_zone);
1396 FLOWMAP_SET(map, ct_mark);
1397 FLOWMAP_SET(map, ct_label);
1399 /* Ethertype-dependent fields. */
1400 if (OVS_LIKELY(flow->dl_type == htons(ETH_TYPE_IP))) {
1401 FLOWMAP_SET(map, nw_src);
1402 FLOWMAP_SET(map, nw_dst);
1403 FLOWMAP_SET(map, nw_proto);
1404 FLOWMAP_SET(map, nw_frag);
1405 FLOWMAP_SET(map, nw_tos);
1406 FLOWMAP_SET(map, nw_ttl);
1408 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_IGMP)) {
1409 FLOWMAP_SET(map, igmp_group_ip4);
1411 FLOWMAP_SET(map, tcp_flags);
1412 FLOWMAP_SET(map, tp_src);
1413 FLOWMAP_SET(map, tp_dst);
1415 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1416 FLOWMAP_SET(map, ipv6_src);
1417 FLOWMAP_SET(map, ipv6_dst);
1418 FLOWMAP_SET(map, ipv6_label);
1419 FLOWMAP_SET(map, nw_proto);
1420 FLOWMAP_SET(map, nw_frag);
1421 FLOWMAP_SET(map, nw_tos);
1422 FLOWMAP_SET(map, nw_ttl);
1424 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_ICMPV6)) {
1425 FLOWMAP_SET(map, nd_target);
1426 FLOWMAP_SET(map, arp_sha);
1427 FLOWMAP_SET(map, arp_tha);
1429 FLOWMAP_SET(map, tcp_flags);
1430 FLOWMAP_SET(map, tp_src);
1431 FLOWMAP_SET(map, tp_dst);
1433 } else if (eth_type_mpls(flow->dl_type)) {
1434 FLOWMAP_SET(map, mpls_lse);
1435 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1436 flow->dl_type == htons(ETH_TYPE_RARP)) {
1437 FLOWMAP_SET(map, nw_src);
1438 FLOWMAP_SET(map, nw_dst);
1439 FLOWMAP_SET(map, nw_proto);
1440 FLOWMAP_SET(map, arp_sha);
1441 FLOWMAP_SET(map, arp_tha);
1445 /* Clear the metadata and register wildcard masks. They are not packet
1448 flow_wildcards_clear_non_packet_fields(struct flow_wildcards *wc)
1450 /* Update this function whenever struct flow changes. */
1451 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 35);
1453 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
1454 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
1455 wc->masks.actset_output = 0;
1456 wc->masks.conj_id = 0;
1459 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
1462 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
1464 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1467 for (i = 0; i < FLOW_U64S; i++) {
1475 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
1476 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
1477 * in 'src1' or 'src2' or both. */
1479 flow_wildcards_and(struct flow_wildcards *dst,
1480 const struct flow_wildcards *src1,
1481 const struct flow_wildcards *src2)
1483 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1484 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1485 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1488 for (i = 0; i < FLOW_U64S; i++) {
1489 dst_u64[i] = src1_u64[i] & src2_u64[i];
1493 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
1494 * is, a bit or a field is wildcarded in 'dst' if it is neither
1495 * wildcarded in 'src1' nor 'src2'. */
1497 flow_wildcards_or(struct flow_wildcards *dst,
1498 const struct flow_wildcards *src1,
1499 const struct flow_wildcards *src2)
1501 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1502 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1503 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1506 for (i = 0; i < FLOW_U64S; i++) {
1507 dst_u64[i] = src1_u64[i] | src2_u64[i];
1511 /* Returns a hash of the wildcards in 'wc'. */
1513 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
1515 return flow_hash(&wc->masks, basis);
1518 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
1521 flow_wildcards_equal(const struct flow_wildcards *a,
1522 const struct flow_wildcards *b)
1524 return flow_equal(&a->masks, &b->masks);
1527 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
1528 * 'b', false otherwise. */
1530 flow_wildcards_has_extra(const struct flow_wildcards *a,
1531 const struct flow_wildcards *b)
1533 const uint64_t *a_u64 = (const uint64_t *) &a->masks;
1534 const uint64_t *b_u64 = (const uint64_t *) &b->masks;
1537 for (i = 0; i < FLOW_U64S; i++) {
1538 if ((a_u64[i] & b_u64[i]) != b_u64[i]) {
1545 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
1546 * in 'wc' do not need to be equal in 'a' and 'b'. */
1548 flow_equal_except(const struct flow *a, const struct flow *b,
1549 const struct flow_wildcards *wc)
1551 const uint64_t *a_u64 = (const uint64_t *) a;
1552 const uint64_t *b_u64 = (const uint64_t *) b;
1553 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1556 for (i = 0; i < FLOW_U64S; i++) {
1557 if ((a_u64[i] ^ b_u64[i]) & wc_u64[i]) {
1564 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1565 * (A 0-bit indicates a wildcard bit.) */
1567 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
1569 wc->masks.regs[idx] = mask;
1572 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1573 * (A 0-bit indicates a wildcard bit.) */
1575 flow_wildcards_set_xreg_mask(struct flow_wildcards *wc, int idx, uint64_t mask)
1577 flow_set_xreg(&wc->masks, idx, mask);
1580 /* Calculates the 5-tuple hash from the given miniflow.
1581 * This returns the same value as flow_hash_5tuple for the corresponding
1584 miniflow_hash_5tuple(const struct miniflow *flow, uint32_t basis)
1586 uint32_t hash = basis;
1589 ovs_be16 dl_type = MINIFLOW_GET_BE16(flow, dl_type);
1591 hash = hash_add(hash, MINIFLOW_GET_U8(flow, nw_proto));
1593 /* Separate loops for better optimization. */
1594 if (dl_type == htons(ETH_TYPE_IPV6)) {
1595 struct flowmap map = FLOWMAP_EMPTY_INITIALIZER;
1598 FLOWMAP_SET(&map, ipv6_src);
1599 FLOWMAP_SET(&map, ipv6_dst);
1601 MINIFLOW_FOR_EACH_IN_FLOWMAP(value, flow, map) {
1602 hash = hash_add64(hash, value);
1605 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_src));
1606 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_dst));
1608 /* Add both ports at once. */
1609 hash = hash_add(hash, MINIFLOW_GET_U32(flow, tp_src));
1610 hash = hash_finish(hash, 42); /* Arbitrary number. */
1615 ASSERT_SEQUENTIAL_SAME_WORD(tp_src, tp_dst);
1616 ASSERT_SEQUENTIAL(ipv6_src, ipv6_dst);
1618 /* Calculates the 5-tuple hash from the given flow. */
1620 flow_hash_5tuple(const struct flow *flow, uint32_t basis)
1622 uint32_t hash = basis;
1625 hash = hash_add(hash, flow->nw_proto);
1627 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1628 const uint64_t *flow_u64 = (const uint64_t *)flow;
1629 int ofs = offsetof(struct flow, ipv6_src) / 8;
1630 int end = ofs + 2 * sizeof flow->ipv6_src / 8;
1632 for (;ofs < end; ofs++) {
1633 hash = hash_add64(hash, flow_u64[ofs]);
1636 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_src);
1637 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_dst);
1639 /* Add both ports at once. */
1640 hash = hash_add(hash,
1641 ((const uint32_t *)flow)[offsetof(struct flow, tp_src)
1642 / sizeof(uint32_t)]);
1643 hash = hash_finish(hash, 42); /* Arbitrary number. */
1648 /* Hashes 'flow' based on its L2 through L4 protocol information. */
1650 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
1655 struct in6_addr ipv6_addr;
1660 struct eth_addr eth_addr;
1666 memset(&fields, 0, sizeof fields);
1667 for (i = 0; i < ARRAY_SIZE(fields.eth_addr.be16); i++) {
1668 fields.eth_addr.be16[i] = flow->dl_src.be16[i] ^ flow->dl_dst.be16[i];
1670 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
1671 fields.eth_type = flow->dl_type;
1673 /* UDP source and destination port are not taken into account because they
1674 * will not necessarily be symmetric in a bidirectional flow. */
1675 if (fields.eth_type == htons(ETH_TYPE_IP)) {
1676 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
1677 fields.ip_proto = flow->nw_proto;
1678 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1679 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1681 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
1682 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
1683 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
1684 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
1686 for (i=0; i<16; i++) {
1687 ipv6_addr[i] = a[i] ^ b[i];
1689 fields.ip_proto = flow->nw_proto;
1690 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1691 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1694 return jhash_bytes(&fields, sizeof fields, basis);
1697 /* Hashes 'flow' based on its L3 through L4 protocol information */
1699 flow_hash_symmetric_l3l4(const struct flow *flow, uint32_t basis,
1702 uint32_t hash = basis;
1704 /* UDP source and destination port are also taken into account. */
1705 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1706 hash = hash_add(hash,
1707 (OVS_FORCE uint32_t) (flow->nw_src ^ flow->nw_dst));
1708 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1709 /* IPv6 addresses are 64-bit aligned inside struct flow. */
1710 const uint64_t *a = ALIGNED_CAST(uint64_t *, flow->ipv6_src.s6_addr);
1711 const uint64_t *b = ALIGNED_CAST(uint64_t *, flow->ipv6_dst.s6_addr);
1713 for (int i = 0; i < 4; i++) {
1714 hash = hash_add64(hash, a[i] ^ b[i]);
1717 /* Cannot hash non-IP flows */
1721 hash = hash_add(hash, flow->nw_proto);
1722 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP ||
1723 (inc_udp_ports && flow->nw_proto == IPPROTO_UDP)) {
1724 hash = hash_add(hash,
1725 (OVS_FORCE uint16_t) (flow->tp_src ^ flow->tp_dst));
1728 return hash_finish(hash, basis);
1731 /* Initialize a flow with random fields that matter for nx_hash_fields. */
1733 flow_random_hash_fields(struct flow *flow)
1735 uint16_t rnd = random_uint16();
1737 /* Initialize to all zeros. */
1738 memset(flow, 0, sizeof *flow);
1740 eth_addr_random(&flow->dl_src);
1741 eth_addr_random(&flow->dl_dst);
1743 flow->vlan_tci = (OVS_FORCE ovs_be16) (random_uint16() & VLAN_VID_MASK);
1745 /* Make most of the random flows IPv4, some IPv6, and rest random. */
1746 flow->dl_type = rnd < 0x8000 ? htons(ETH_TYPE_IP) :
1747 rnd < 0xc000 ? htons(ETH_TYPE_IPV6) : (OVS_FORCE ovs_be16)rnd;
1749 if (dl_type_is_ip_any(flow->dl_type)) {
1750 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1751 flow->nw_src = (OVS_FORCE ovs_be32)random_uint32();
1752 flow->nw_dst = (OVS_FORCE ovs_be32)random_uint32();
1754 random_bytes(&flow->ipv6_src, sizeof flow->ipv6_src);
1755 random_bytes(&flow->ipv6_dst, sizeof flow->ipv6_dst);
1757 /* Make most of IP flows TCP, some UDP or SCTP, and rest random. */
1758 rnd = random_uint16();
1759 flow->nw_proto = rnd < 0x8000 ? IPPROTO_TCP :
1760 rnd < 0xc000 ? IPPROTO_UDP :
1761 rnd < 0xd000 ? IPPROTO_SCTP : (uint8_t)rnd;
1762 if (flow->nw_proto == IPPROTO_TCP ||
1763 flow->nw_proto == IPPROTO_UDP ||
1764 flow->nw_proto == IPPROTO_SCTP) {
1765 flow->tp_src = (OVS_FORCE ovs_be16)random_uint16();
1766 flow->tp_dst = (OVS_FORCE ovs_be16)random_uint16();
1771 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
1773 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
1774 enum nx_hash_fields fields)
1777 case NX_HASH_FIELDS_ETH_SRC:
1778 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1781 case NX_HASH_FIELDS_SYMMETRIC_L4:
1782 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1783 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1784 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1785 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1786 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1787 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1788 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1789 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1791 if (is_ip_any(flow)) {
1792 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1793 flow_unwildcard_tp_ports(flow, wc);
1795 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1798 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1799 if (is_ip_any(flow) && flow->nw_proto == IPPROTO_UDP) {
1800 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1801 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1804 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1805 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1806 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1807 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1808 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1809 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1810 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1812 break; /* non-IP flow */
1815 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1816 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP) {
1817 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1818 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1827 /* Hashes the portions of 'flow' designated by 'fields'. */
1829 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
1834 case NX_HASH_FIELDS_ETH_SRC:
1835 return jhash_bytes(&flow->dl_src, sizeof flow->dl_src, basis);
1837 case NX_HASH_FIELDS_SYMMETRIC_L4:
1838 return flow_hash_symmetric_l4(flow, basis);
1840 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1841 return flow_hash_symmetric_l3l4(flow, basis, false);
1843 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1844 return flow_hash_symmetric_l3l4(flow, basis, true);
1851 /* Returns a string representation of 'fields'. */
1853 flow_hash_fields_to_str(enum nx_hash_fields fields)
1856 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
1857 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
1858 case NX_HASH_FIELDS_SYMMETRIC_L3L4: return "symmetric_l3l4";
1859 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP: return "symmetric_l3l4+udp";
1860 default: return "<unknown>";
1864 /* Returns true if the value of 'fields' is supported. Otherwise false. */
1866 flow_hash_fields_valid(enum nx_hash_fields fields)
1868 return fields == NX_HASH_FIELDS_ETH_SRC
1869 || fields == NX_HASH_FIELDS_SYMMETRIC_L4
1870 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4
1871 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP;
1874 /* Returns a hash value for the bits of 'flow' that are active based on
1875 * 'wc', given 'basis'. */
1877 flow_hash_in_wildcards(const struct flow *flow,
1878 const struct flow_wildcards *wc, uint32_t basis)
1880 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1881 const uint64_t *flow_u64 = (const uint64_t *) flow;
1886 for (i = 0; i < FLOW_U64S; i++) {
1887 hash = hash_add64(hash, flow_u64[i] & wc_u64[i]);
1889 return hash_finish(hash, 8 * FLOW_U64S);
1892 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1893 * OpenFlow 1.0 "dl_vlan" value:
1895 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
1896 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
1897 * 'flow' previously matched packets without a VLAN header).
1899 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
1900 * without a VLAN tag.
1902 * - Other values of 'vid' should not be used. */
1904 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
1906 if (vid == htons(OFP10_VLAN_NONE)) {
1907 flow->vlan_tci = htons(0);
1909 vid &= htons(VLAN_VID_MASK);
1910 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1911 flow->vlan_tci |= htons(VLAN_CFI) | vid;
1915 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1916 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
1919 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
1921 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
1922 flow->vlan_tci &= ~mask;
1923 flow->vlan_tci |= vid & mask;
1926 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
1929 * This function has no effect on the VLAN ID that 'flow' matches.
1931 * After calling this function, 'flow' will not match packets without a VLAN
1934 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
1937 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1938 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
1941 /* Returns the number of MPLS LSEs present in 'flow'
1943 * Returns 0 if the 'dl_type' of 'flow' is not an MPLS ethernet type.
1944 * Otherwise traverses 'flow''s MPLS label stack stopping at the
1945 * first entry that has the BoS bit set. If no such entry exists then
1946 * the maximum number of LSEs that can be stored in 'flow' is returned.
1949 flow_count_mpls_labels(const struct flow *flow, struct flow_wildcards *wc)
1951 /* dl_type is always masked. */
1952 if (eth_type_mpls(flow->dl_type)) {
1957 for (i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1959 wc->masks.mpls_lse[i] |= htonl(MPLS_BOS_MASK);
1961 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1964 if (flow->mpls_lse[i]) {
1974 /* Returns the number consecutive of MPLS LSEs, starting at the
1975 * innermost LSE, that are common in 'a' and 'b'.
1977 * 'an' must be flow_count_mpls_labels(a).
1978 * 'bn' must be flow_count_mpls_labels(b).
1981 flow_count_common_mpls_labels(const struct flow *a, int an,
1982 const struct flow *b, int bn,
1983 struct flow_wildcards *wc)
1985 int min_n = MIN(an, bn);
1990 int a_last = an - 1;
1991 int b_last = bn - 1;
1994 for (i = 0; i < min_n; i++) {
1996 wc->masks.mpls_lse[a_last - i] = OVS_BE32_MAX;
1997 wc->masks.mpls_lse[b_last - i] = OVS_BE32_MAX;
1999 if (a->mpls_lse[a_last - i] != b->mpls_lse[b_last - i]) {
2010 /* Adds a new outermost MPLS label to 'flow' and changes 'flow''s Ethernet type
2011 * to 'mpls_eth_type', which must be an MPLS Ethertype.
2013 * If the new label is the first MPLS label in 'flow', it is generated as;
2015 * - label: 2, if 'flow' is IPv6, otherwise 0.
2017 * - TTL: IPv4 or IPv6 TTL, if present and nonzero, otherwise 64.
2019 * - TC: IPv4 or IPv6 TOS, if present, otherwise 0.
2023 * If the new label is the second or later label MPLS label in 'flow', it is
2026 * - label: Copied from outer label.
2028 * - TTL: Copied from outer label.
2030 * - TC: Copied from outer label.
2034 * 'n' must be flow_count_mpls_labels(flow). 'n' must be less than
2035 * FLOW_MAX_MPLS_LABELS (because otherwise flow->mpls_lse[] would overflow).
2038 flow_push_mpls(struct flow *flow, int n, ovs_be16 mpls_eth_type,
2039 struct flow_wildcards *wc)
2041 ovs_assert(eth_type_mpls(mpls_eth_type));
2042 ovs_assert(n < FLOW_MAX_MPLS_LABELS);
2048 memset(&wc->masks.mpls_lse, 0xff, sizeof *wc->masks.mpls_lse * n);
2050 for (i = n; i >= 1; i--) {
2051 flow->mpls_lse[i] = flow->mpls_lse[i - 1];
2053 flow->mpls_lse[0] = (flow->mpls_lse[1] & htonl(~MPLS_BOS_MASK));
2055 int label = 0; /* IPv4 Explicit Null. */
2059 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2063 if (is_ip_any(flow)) {
2064 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
2066 wc->masks.nw_tos |= IP_DSCP_MASK;
2067 wc->masks.nw_ttl = 0xff;
2075 flow->mpls_lse[0] = set_mpls_lse_values(ttl, tc, 1, htonl(label));
2077 /* Clear all L3 and L4 fields and dp_hash. */
2078 BUILD_ASSERT(FLOW_WC_SEQ == 35);
2079 memset((char *) flow + FLOW_SEGMENT_2_ENDS_AT, 0,
2080 sizeof(struct flow) - FLOW_SEGMENT_2_ENDS_AT);
2083 flow->dl_type = mpls_eth_type;
2086 /* Tries to remove the outermost MPLS label from 'flow'. Returns true if
2087 * successful, false otherwise. On success, sets 'flow''s Ethernet type to
2090 * 'n' must be flow_count_mpls_labels(flow). */
2092 flow_pop_mpls(struct flow *flow, int n, ovs_be16 eth_type,
2093 struct flow_wildcards *wc)
2098 /* Nothing to pop. */
2100 } else if (n == FLOW_MAX_MPLS_LABELS) {
2102 wc->masks.mpls_lse[n - 1] |= htonl(MPLS_BOS_MASK);
2104 if (!(flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK))) {
2105 /* Can't pop because don't know what to fill in mpls_lse[n - 1]. */
2111 memset(&wc->masks.mpls_lse[1], 0xff,
2112 sizeof *wc->masks.mpls_lse * (n - 1));
2114 for (i = 1; i < n; i++) {
2115 flow->mpls_lse[i - 1] = flow->mpls_lse[i];
2117 flow->mpls_lse[n - 1] = 0;
2118 flow->dl_type = eth_type;
2122 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
2123 * as an OpenFlow 1.1 "mpls_label" value. */
2125 flow_set_mpls_label(struct flow *flow, int idx, ovs_be32 label)
2127 set_mpls_lse_label(&flow->mpls_lse[idx], label);
2130 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
2133 flow_set_mpls_ttl(struct flow *flow, int idx, uint8_t ttl)
2135 set_mpls_lse_ttl(&flow->mpls_lse[idx], ttl);
2138 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
2141 flow_set_mpls_tc(struct flow *flow, int idx, uint8_t tc)
2143 set_mpls_lse_tc(&flow->mpls_lse[idx], tc);
2146 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
2148 flow_set_mpls_bos(struct flow *flow, int idx, uint8_t bos)
2150 set_mpls_lse_bos(&flow->mpls_lse[idx], bos);
2153 /* Sets the entire MPLS LSE. */
2155 flow_set_mpls_lse(struct flow *flow, int idx, ovs_be32 lse)
2157 flow->mpls_lse[idx] = lse;
2161 flow_compose_l4(struct dp_packet *p, const struct flow *flow)
2165 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
2166 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
2167 if (flow->nw_proto == IPPROTO_TCP) {
2168 struct tcp_header *tcp;
2170 l4_len = sizeof *tcp;
2171 tcp = dp_packet_put_zeros(p, l4_len);
2172 tcp->tcp_src = flow->tp_src;
2173 tcp->tcp_dst = flow->tp_dst;
2174 tcp->tcp_ctl = TCP_CTL(ntohs(flow->tcp_flags), 5);
2175 } else if (flow->nw_proto == IPPROTO_UDP) {
2176 struct udp_header *udp;
2178 l4_len = sizeof *udp;
2179 udp = dp_packet_put_zeros(p, l4_len);
2180 udp->udp_src = flow->tp_src;
2181 udp->udp_dst = flow->tp_dst;
2182 } else if (flow->nw_proto == IPPROTO_SCTP) {
2183 struct sctp_header *sctp;
2185 l4_len = sizeof *sctp;
2186 sctp = dp_packet_put_zeros(p, l4_len);
2187 sctp->sctp_src = flow->tp_src;
2188 sctp->sctp_dst = flow->tp_dst;
2189 } else if (flow->nw_proto == IPPROTO_ICMP) {
2190 struct icmp_header *icmp;
2192 l4_len = sizeof *icmp;
2193 icmp = dp_packet_put_zeros(p, l4_len);
2194 icmp->icmp_type = ntohs(flow->tp_src);
2195 icmp->icmp_code = ntohs(flow->tp_dst);
2196 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
2197 } else if (flow->nw_proto == IPPROTO_IGMP) {
2198 struct igmp_header *igmp;
2200 l4_len = sizeof *igmp;
2201 igmp = dp_packet_put_zeros(p, l4_len);
2202 igmp->igmp_type = ntohs(flow->tp_src);
2203 igmp->igmp_code = ntohs(flow->tp_dst);
2204 put_16aligned_be32(&igmp->group, flow->igmp_group_ip4);
2205 igmp->igmp_csum = csum(igmp, IGMP_HEADER_LEN);
2206 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
2207 struct icmp6_hdr *icmp;
2209 l4_len = sizeof *icmp;
2210 icmp = dp_packet_put_zeros(p, l4_len);
2211 icmp->icmp6_type = ntohs(flow->tp_src);
2212 icmp->icmp6_code = ntohs(flow->tp_dst);
2214 if (icmp->icmp6_code == 0 &&
2215 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
2216 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
2217 struct in6_addr *nd_target;
2218 struct ovs_nd_opt *nd_opt;
2220 l4_len += sizeof *nd_target;
2221 nd_target = dp_packet_put_zeros(p, sizeof *nd_target);
2222 *nd_target = flow->nd_target;
2224 if (!eth_addr_is_zero(flow->arp_sha)) {
2226 nd_opt = dp_packet_put_zeros(p, 8);
2227 nd_opt->nd_opt_len = 1;
2228 nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
2229 nd_opt->nd_opt_mac = flow->arp_sha;
2231 if (!eth_addr_is_zero(flow->arp_tha)) {
2233 nd_opt = dp_packet_put_zeros(p, 8);
2234 nd_opt->nd_opt_len = 1;
2235 nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR;
2236 nd_opt->nd_opt_mac = flow->arp_tha;
2239 icmp->icmp6_cksum = (OVS_FORCE uint16_t)
2240 csum(icmp, (char *)dp_packet_tail(p) - (char *)icmp);
2246 /* Puts into 'b' a packet that flow_extract() would parse as having the given
2249 * (This is useful only for testing, obviously, and the packet isn't really
2250 * valid. It hasn't got some checksums filled in, for one, and lots of fields
2251 * are just zeroed.) */
2253 flow_compose(struct dp_packet *p, const struct flow *flow)
2257 /* eth_compose() sets l3 pointer and makes sure it is 32-bit aligned. */
2258 eth_compose(p, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
2259 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
2260 struct eth_header *eth = dp_packet_l2(p);
2261 eth->eth_type = htons(dp_packet_size(p));
2265 if (flow->vlan_tci & htons(VLAN_CFI)) {
2266 eth_push_vlan(p, htons(ETH_TYPE_VLAN), flow->vlan_tci);
2269 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2270 struct ip_header *ip;
2272 ip = dp_packet_put_zeros(p, sizeof *ip);
2273 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
2274 ip->ip_tos = flow->nw_tos;
2275 ip->ip_ttl = flow->nw_ttl;
2276 ip->ip_proto = flow->nw_proto;
2277 put_16aligned_be32(&ip->ip_src, flow->nw_src);
2278 put_16aligned_be32(&ip->ip_dst, flow->nw_dst);
2280 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2281 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
2282 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
2283 ip->ip_frag_off |= htons(100);
2287 dp_packet_set_l4(p, dp_packet_tail(p));
2289 l4_len = flow_compose_l4(p, flow);
2291 ip = dp_packet_l3(p);
2292 ip->ip_tot_len = htons(p->l4_ofs - p->l3_ofs + l4_len);
2293 ip->ip_csum = csum(ip, sizeof *ip);
2294 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2295 struct ovs_16aligned_ip6_hdr *nh;
2297 nh = dp_packet_put_zeros(p, sizeof *nh);
2298 put_16aligned_be32(&nh->ip6_flow, htonl(6 << 28) |
2299 htonl(flow->nw_tos << 20) | flow->ipv6_label);
2300 nh->ip6_hlim = flow->nw_ttl;
2301 nh->ip6_nxt = flow->nw_proto;
2303 memcpy(&nh->ip6_src, &flow->ipv6_src, sizeof(nh->ip6_src));
2304 memcpy(&nh->ip6_dst, &flow->ipv6_dst, sizeof(nh->ip6_dst));
2306 dp_packet_set_l4(p, dp_packet_tail(p));
2308 l4_len = flow_compose_l4(p, flow);
2310 nh = dp_packet_l3(p);
2311 nh->ip6_plen = htons(l4_len);
2312 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
2313 flow->dl_type == htons(ETH_TYPE_RARP)) {
2314 struct arp_eth_header *arp;
2316 arp = dp_packet_put_zeros(p, sizeof *arp);
2317 dp_packet_set_l3(p, arp);
2318 arp->ar_hrd = htons(1);
2319 arp->ar_pro = htons(ETH_TYPE_IP);
2320 arp->ar_hln = ETH_ADDR_LEN;
2322 arp->ar_op = htons(flow->nw_proto);
2324 if (flow->nw_proto == ARP_OP_REQUEST ||
2325 flow->nw_proto == ARP_OP_REPLY) {
2326 put_16aligned_be32(&arp->ar_spa, flow->nw_src);
2327 put_16aligned_be32(&arp->ar_tpa, flow->nw_dst);
2328 arp->ar_sha = flow->arp_sha;
2329 arp->ar_tha = flow->arp_tha;
2333 if (eth_type_mpls(flow->dl_type)) {
2336 p->l2_5_ofs = p->l3_ofs;
2337 for (n = 1; n < FLOW_MAX_MPLS_LABELS; n++) {
2338 if (flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK)) {
2343 push_mpls(p, flow->dl_type, flow->mpls_lse[--n]);
2348 /* Compressed flow. */
2350 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
2351 * the caller. The caller must have already computed 'dst->map' properly to
2352 * indicate the significant uint64_t elements of 'src'.
2354 * Normally the significant elements are the ones that are non-zero. However,
2355 * when a miniflow is initialized from a (mini)mask, the values can be zeroes,
2356 * so that the flow and mask always have the same maps. */
2358 miniflow_init(struct miniflow *dst, const struct flow *src)
2360 uint64_t *dst_u64 = miniflow_values(dst);
2363 FLOWMAP_FOR_EACH_INDEX(idx, dst->map) {
2364 *dst_u64++ = flow_u64_value(src, idx);
2368 /* Initialize the maps of 'flow' from 'src'. */
2370 miniflow_map_init(struct miniflow *flow, const struct flow *src)
2372 /* Initialize map, counting the number of nonzero elements. */
2373 flowmap_init(&flow->map);
2374 for (size_t i = 0; i < FLOW_U64S; i++) {
2375 if (flow_u64_value(src, i)) {
2376 flowmap_set(&flow->map, i, 1);
2381 /* Allocates 'n' count of miniflows, consecutive in memory, initializing the
2382 * map of each from 'src'.
2383 * Returns the size of the miniflow data. */
2385 miniflow_alloc(struct miniflow *dsts[], size_t n, const struct miniflow *src)
2387 size_t n_values = miniflow_n_values(src);
2388 size_t data_size = MINIFLOW_VALUES_SIZE(n_values);
2389 struct miniflow *dst = xmalloc(n * (sizeof *src + data_size));
2392 COVERAGE_INC(miniflow_malloc);
2394 for (i = 0; i < n; i++) {
2395 *dst = *src; /* Copy maps. */
2397 dst += 1; /* Just past the maps. */
2398 dst = (struct miniflow *)((uint64_t *)dst + n_values); /* Skip data. */
2403 /* Returns a miniflow copy of 'src'. The caller must eventually free() the
2404 * returned miniflow. */
2406 miniflow_create(const struct flow *src)
2408 struct miniflow tmp;
2409 struct miniflow *dst;
2411 miniflow_map_init(&tmp, src);
2413 miniflow_alloc(&dst, 1, &tmp);
2414 miniflow_init(dst, src);
2418 /* Initializes 'dst' as a copy of 'src'. The caller must have allocated
2419 * 'dst' to have inline space for 'n_values' data in 'src'. */
2421 miniflow_clone(struct miniflow *dst, const struct miniflow *src,
2424 *dst = *src; /* Copy maps. */
2425 memcpy(miniflow_values(dst), miniflow_get_values(src),
2426 MINIFLOW_VALUES_SIZE(n_values));
2429 /* Initializes 'dst' as a copy of 'src'. */
2431 miniflow_expand(const struct miniflow *src, struct flow *dst)
2433 memset(dst, 0, sizeof *dst);
2434 flow_union_with_miniflow(dst, src);
2437 /* Returns true if 'a' and 'b' are equal miniflows, false otherwise. */
2439 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
2441 const uint64_t *ap = miniflow_get_values(a);
2442 const uint64_t *bp = miniflow_get_values(b);
2444 /* This is mostly called after a matching hash, so it is highly likely that
2445 * the maps are equal as well. */
2446 if (OVS_LIKELY(flowmap_equal(a->map, b->map))) {
2447 return !memcmp(ap, bp, miniflow_n_values(a) * sizeof *ap);
2451 FLOWMAP_FOR_EACH_INDEX (idx, flowmap_or(a->map, b->map)) {
2452 if ((flowmap_is_set(&a->map, idx) ? *ap++ : 0)
2453 != (flowmap_is_set(&b->map, idx) ? *bp++ : 0)) {
2462 /* Returns false if 'a' and 'b' differ at the places where there are 1-bits
2463 * in 'mask', true otherwise. */
2465 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
2466 const struct minimask *mask)
2468 const uint64_t *p = miniflow_get_values(&mask->masks);
2471 FLOWMAP_FOR_EACH_INDEX(idx, mask->masks.map) {
2472 if ((miniflow_get(a, idx) ^ miniflow_get(b, idx)) & *p++) {
2480 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
2481 * in 'mask', false if they differ. */
2483 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
2484 const struct minimask *mask)
2486 const uint64_t *p = miniflow_get_values(&mask->masks);
2489 FLOWMAP_FOR_EACH_INDEX(idx, mask->masks.map) {
2490 if ((miniflow_get(a, idx) ^ flow_u64_value(b, idx)) & *p++) {
2500 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
2502 miniflow_init(&mask->masks, &wc->masks);
2505 /* Returns a minimask copy of 'wc'. The caller must eventually free the
2506 * returned minimask with free(). */
2508 minimask_create(const struct flow_wildcards *wc)
2510 return (struct minimask *)miniflow_create(&wc->masks);
2513 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
2515 * The caller must provide room for FLOW_U64S "uint64_t"s in 'storage', which
2516 * must follow '*dst_' in memory, for use by 'dst_'. The caller must *not*
2517 * free 'dst_' free(). */
2519 minimask_combine(struct minimask *dst_,
2520 const struct minimask *a_, const struct minimask *b_,
2521 uint64_t storage[FLOW_U64S])
2523 struct miniflow *dst = &dst_->masks;
2524 uint64_t *dst_values = storage;
2525 const struct miniflow *a = &a_->masks;
2526 const struct miniflow *b = &b_->masks;
2529 flowmap_init(&dst->map);
2531 FLOWMAP_FOR_EACH_INDEX(idx, flowmap_and(a->map, b->map)) {
2532 /* Both 'a' and 'b' have non-zero data at 'idx'. */
2533 uint64_t mask = *miniflow_get__(a, idx) & *miniflow_get__(b, idx);
2536 flowmap_set(&dst->map, idx, 1);
2537 *dst_values++ = mask;
2542 /* Initializes 'wc' as a copy of 'mask'. */
2544 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
2546 miniflow_expand(&mask->masks, &wc->masks);
2549 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise.
2550 * Minimasks may not have zero data values, so for the minimasks to be the
2551 * same, they need to have the same map and the same data values. */
2553 minimask_equal(const struct minimask *a, const struct minimask *b)
2555 return !memcmp(a, b, sizeof *a
2556 + MINIFLOW_VALUES_SIZE(miniflow_n_values(&a->masks)));
2559 /* Returns true if at least one bit matched by 'b' is wildcarded by 'a',
2560 * false otherwise. */
2562 minimask_has_extra(const struct minimask *a, const struct minimask *b)
2564 const uint64_t *bp = miniflow_get_values(&b->masks);
2567 FLOWMAP_FOR_EACH_INDEX(idx, b->masks.map) {
2568 uint64_t b_u64 = *bp++;
2570 /* 'b_u64' is non-zero, check if the data in 'a' is either zero
2571 * or misses some of the bits in 'b_u64'. */
2572 if (!MINIFLOW_IN_MAP(&a->masks, idx)
2573 || ((*miniflow_get__(&a->masks, idx) & b_u64) != b_u64)) {
2574 return true; /* 'a' wildcards some bits 'b' doesn't. */