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 != 33)
127 #define MINIFLOW_ASSERT(X) ovs_assert(X)
128 BUILD_MESSAGE("FLOW_WC_SEQ changed: miniflow_extract() will have runtime "
129 "assertions enabled. Consider updating FLOW_WC_SEQ after "
132 #define MINIFLOW_ASSERT(X)
135 /* 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 (FLOWMAP_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_pad_to_64_(MF, OFS) \
204 MINIFLOW_ASSERT((OFS) % 8 != 0); \
205 miniflow_assert_in_map(MF, OFS / 8); \
207 memset((uint8_t *)MF.data + (OFS) % 8, 0, 8 - (OFS) % 8); \
211 #define miniflow_push_be16_(MF, OFS, VALUE) \
212 miniflow_push_uint16_(MF, OFS, (OVS_FORCE uint16_t)VALUE);
214 #define miniflow_set_maps(MF, OFS, N_WORDS) \
216 size_t ofs = (OFS); \
217 size_t n_words = (N_WORDS); \
219 MINIFLOW_ASSERT(n_words && MF.data + n_words <= MF.end); \
220 ASSERT_FLOWMAP_NOT_SET(&MF.map, ofs); \
221 flowmap_set(&MF.map, ofs, n_words); \
224 /* Data at 'valuep' may be unaligned. */
225 #define miniflow_push_words_(MF, OFS, VALUEP, N_WORDS) \
227 MINIFLOW_ASSERT((OFS) % 8 == 0); \
228 miniflow_set_maps(MF, (OFS) / 8, (N_WORDS)); \
229 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof *MF.data); \
230 MF.data += (N_WORDS); \
233 /* Push 32-bit words padded to 64-bits. */
234 #define miniflow_push_words_32_(MF, OFS, VALUEP, N_WORDS) \
236 miniflow_set_maps(MF, (OFS) / 8, DIV_ROUND_UP(N_WORDS, 2)); \
237 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof(uint32_t)); \
238 MF.data += DIV_ROUND_UP(N_WORDS, 2); \
239 if ((N_WORDS) & 1) { \
240 *((uint32_t *)MF.data - 1) = 0; \
244 /* Data at 'valuep' may be unaligned. */
245 /* MACs start 64-aligned, and must be followed by other data or padding. */
246 #define miniflow_push_macs_(MF, OFS, VALUEP) \
248 miniflow_set_maps(MF, (OFS) / 8, 2); \
249 memcpy(MF.data, (VALUEP), 2 * ETH_ADDR_LEN); \
250 MF.data += 1; /* First word only. */ \
253 #define miniflow_push_uint32(MF, FIELD, VALUE) \
254 miniflow_push_uint32_(MF, offsetof(struct flow, FIELD), VALUE)
256 #define miniflow_push_be32(MF, FIELD, VALUE) \
257 miniflow_push_be32_(MF, offsetof(struct flow, FIELD), VALUE)
259 #define miniflow_push_uint16(MF, FIELD, VALUE) \
260 miniflow_push_uint16_(MF, offsetof(struct flow, FIELD), VALUE)
262 #define miniflow_push_be16(MF, FIELD, VALUE) \
263 miniflow_push_be16_(MF, offsetof(struct flow, FIELD), VALUE)
265 #define miniflow_pad_to_64(MF, FIELD) \
266 miniflow_pad_to_64_(MF, offsetof(struct flow, FIELD))
268 #define miniflow_push_words(MF, FIELD, VALUEP, N_WORDS) \
269 miniflow_push_words_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
271 #define miniflow_push_words_32(MF, FIELD, VALUEP, N_WORDS) \
272 miniflow_push_words_32_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
274 #define miniflow_push_macs(MF, FIELD, VALUEP) \
275 miniflow_push_macs_(MF, offsetof(struct flow, FIELD), VALUEP)
277 /* Pulls the MPLS headers at '*datap' and returns the count of them. */
279 parse_mpls(const void **datap, size_t *sizep)
281 const struct mpls_hdr *mh;
284 while ((mh = data_try_pull(datap, sizep, sizeof *mh))) {
286 if (mh->mpls_lse.lo & htons(1 << MPLS_BOS_SHIFT)) {
290 return MIN(count, FLOW_MAX_MPLS_LABELS);
293 static inline ovs_be16
294 parse_vlan(const void **datap, size_t *sizep)
296 const struct eth_header *eth = *datap;
299 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
303 data_pull(datap, sizep, ETH_ADDR_LEN * 2);
305 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
306 if (OVS_LIKELY(*sizep
307 >= sizeof(struct qtag_prefix) + sizeof(ovs_be16))) {
308 const struct qtag_prefix *qp = data_pull(datap, sizep, sizeof *qp);
309 return qp->tci | htons(VLAN_CFI);
315 static inline ovs_be16
316 parse_ethertype(const void **datap, size_t *sizep)
318 const struct llc_snap_header *llc;
321 proto = *(ovs_be16 *) data_pull(datap, sizep, sizeof proto);
322 if (OVS_LIKELY(ntohs(proto) >= ETH_TYPE_MIN)) {
326 if (OVS_UNLIKELY(*sizep < sizeof *llc)) {
327 return htons(FLOW_DL_TYPE_NONE);
331 if (OVS_UNLIKELY(llc->llc.llc_dsap != LLC_DSAP_SNAP
332 || llc->llc.llc_ssap != LLC_SSAP_SNAP
333 || llc->llc.llc_cntl != LLC_CNTL_SNAP
334 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
335 sizeof llc->snap.snap_org))) {
336 return htons(FLOW_DL_TYPE_NONE);
339 data_pull(datap, sizep, sizeof *llc);
341 if (OVS_LIKELY(ntohs(llc->snap.snap_type) >= ETH_TYPE_MIN)) {
342 return llc->snap.snap_type;
345 return htons(FLOW_DL_TYPE_NONE);
349 parse_icmpv6(const void **datap, size_t *sizep, const struct icmp6_hdr *icmp,
350 const struct in6_addr **nd_target,
351 struct eth_addr arp_buf[2])
353 if (icmp->icmp6_code == 0 &&
354 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
355 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
357 *nd_target = data_try_pull(datap, sizep, sizeof **nd_target);
358 if (OVS_UNLIKELY(!*nd_target)) {
362 while (*sizep >= 8) {
363 /* The minimum size of an option is 8 bytes, which also is
364 * the size of Ethernet link-layer options. */
365 const struct ovs_nd_opt *nd_opt = *datap;
366 int opt_len = nd_opt->nd_opt_len * ND_OPT_LEN;
368 if (!opt_len || opt_len > *sizep) {
372 /* Store the link layer address if the appropriate option is
373 * provided. It is considered an error if the same link
374 * layer option is specified twice. */
375 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
377 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[0]))) {
378 arp_buf[0] = nd_opt->nd_opt_mac;
382 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
384 if (OVS_LIKELY(eth_addr_is_zero(arp_buf[1]))) {
385 arp_buf[1] = nd_opt->nd_opt_mac;
391 if (OVS_UNLIKELY(!data_try_pull(datap, sizep, opt_len))) {
401 arp_buf[0] = eth_addr_zero;
402 arp_buf[1] = eth_addr_zero;
405 /* Initializes 'flow' members from 'packet' and 'md'
407 * Initializes 'packet' header l2 pointer to the start of the Ethernet
408 * header, and the layer offsets as follows:
410 * - packet->l2_5_ofs to the start of the MPLS shim header, or UINT16_MAX
411 * when there is no MPLS shim header.
413 * - packet->l3_ofs to just past the Ethernet header, or just past the
414 * vlan_header if one is present, to the first byte of the payload of the
415 * Ethernet frame. UINT16_MAX if the frame is too short to contain an
418 * - packet->l4_ofs to just past the IPv4 header, if one is present and
419 * has at least the content used for the fields of interest for the flow,
420 * otherwise UINT16_MAX.
423 flow_extract(struct dp_packet *packet, struct flow *flow)
427 uint64_t buf[FLOW_U64S];
430 COVERAGE_INC(flow_extract);
432 miniflow_extract(packet, &m.mf);
433 miniflow_expand(&m.mf, flow);
436 /* Caller is responsible for initializing 'dst' with enough storage for
437 * FLOW_U64S * 8 bytes. */
439 miniflow_extract(struct dp_packet *packet, struct miniflow *dst)
441 const struct pkt_metadata *md = &packet->md;
442 const void *data = dp_packet_data(packet);
443 size_t size = dp_packet_size(packet);
444 uint64_t *values = miniflow_values(dst);
445 struct mf_ctx mf = { FLOWMAP_EMPTY_INITIALIZER, values,
446 values + FLOW_U64S };
449 uint8_t nw_frag, nw_tos, nw_ttl, nw_proto;
452 if (md->tunnel.ip_dst) {
453 miniflow_push_words(mf, tunnel, &md->tunnel,
454 offsetof(struct flow_tnl, metadata) /
457 if (!(md->tunnel.flags & FLOW_TNL_F_UDPIF)) {
458 if (md->tunnel.metadata.present.map) {
459 miniflow_push_words(mf, tunnel.metadata, &md->tunnel.metadata,
460 sizeof md->tunnel.metadata /
464 if (md->tunnel.metadata.present.len) {
465 miniflow_push_words(mf, tunnel.metadata.present,
466 &md->tunnel.metadata.present, 1);
467 miniflow_push_words(mf, tunnel.metadata.opts.gnv,
468 md->tunnel.metadata.opts.gnv,
469 DIV_ROUND_UP(md->tunnel.metadata.present.len,
474 if (md->skb_priority || md->pkt_mark) {
475 miniflow_push_uint32(mf, skb_priority, md->skb_priority);
476 miniflow_push_uint32(mf, pkt_mark, md->pkt_mark);
478 miniflow_push_uint32(mf, dp_hash, md->dp_hash);
479 miniflow_push_uint32(mf, in_port, odp_to_u32(md->in_port.odp_port));
481 miniflow_push_uint32(mf, recirc_id, md->recirc_id);
482 miniflow_pad_to_64(mf, conj_id);
485 /* Initialize packet's layer pointer and offsets. */
487 dp_packet_reset_offsets(packet);
489 /* Must have full Ethernet header to proceed. */
490 if (OVS_UNLIKELY(size < sizeof(struct eth_header))) {
496 ASSERT_SEQUENTIAL(dl_dst, dl_src);
497 miniflow_push_macs(mf, dl_dst, data);
498 /* dl_type, vlan_tci. */
499 vlan_tci = parse_vlan(&data, &size);
500 dl_type = parse_ethertype(&data, &size);
501 miniflow_push_be16(mf, dl_type, dl_type);
502 miniflow_push_be16(mf, vlan_tci, vlan_tci);
506 if (OVS_UNLIKELY(eth_type_mpls(dl_type))) {
508 const void *mpls = data;
510 packet->l2_5_ofs = (char *)data - l2;
511 count = parse_mpls(&data, &size);
512 miniflow_push_words_32(mf, mpls_lse, mpls, count);
516 packet->l3_ofs = (char *)data - l2;
519 if (OVS_LIKELY(dl_type == htons(ETH_TYPE_IP))) {
520 const struct ip_header *nh = data;
524 if (OVS_UNLIKELY(size < IP_HEADER_LEN)) {
527 ip_len = IP_IHL(nh->ip_ihl_ver) * 4;
529 if (OVS_UNLIKELY(ip_len < IP_HEADER_LEN)) {
532 if (OVS_UNLIKELY(size < ip_len)) {
535 tot_len = ntohs(nh->ip_tot_len);
536 if (OVS_UNLIKELY(tot_len > size)) {
539 if (OVS_UNLIKELY(size - tot_len > UINT8_MAX)) {
542 dp_packet_set_l2_pad_size(packet, size - tot_len);
543 size = tot_len; /* Never pull padding. */
545 /* Push both source and destination address at once. */
546 miniflow_push_words(mf, nw_src, &nh->ip_src, 1);
548 miniflow_push_be32(mf, ipv6_label, 0); /* Padding for IPv4. */
552 nw_proto = nh->ip_proto;
553 if (OVS_UNLIKELY(IP_IS_FRAGMENT(nh->ip_frag_off))) {
554 nw_frag = FLOW_NW_FRAG_ANY;
555 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
556 nw_frag |= FLOW_NW_FRAG_LATER;
559 data_pull(&data, &size, ip_len);
560 } else if (dl_type == htons(ETH_TYPE_IPV6)) {
561 const struct ovs_16aligned_ip6_hdr *nh;
565 if (OVS_UNLIKELY(size < sizeof *nh)) {
568 nh = data_pull(&data, &size, sizeof *nh);
570 plen = ntohs(nh->ip6_plen);
571 if (OVS_UNLIKELY(plen > size)) {
574 /* Jumbo Payload option not supported yet. */
575 if (OVS_UNLIKELY(size - plen > UINT8_MAX)) {
578 dp_packet_set_l2_pad_size(packet, size - plen);
579 size = plen; /* Never pull padding. */
581 miniflow_push_words(mf, ipv6_src, &nh->ip6_src,
582 sizeof nh->ip6_src / 8);
583 miniflow_push_words(mf, ipv6_dst, &nh->ip6_dst,
584 sizeof nh->ip6_dst / 8);
586 tc_flow = get_16aligned_be32(&nh->ip6_flow);
588 ovs_be32 label = tc_flow & htonl(IPV6_LABEL_MASK);
589 miniflow_push_be32(mf, ipv6_label, label);
592 nw_tos = ntohl(tc_flow) >> 20;
593 nw_ttl = nh->ip6_hlim;
594 nw_proto = nh->ip6_nxt;
597 if (OVS_LIKELY((nw_proto != IPPROTO_HOPOPTS)
598 && (nw_proto != IPPROTO_ROUTING)
599 && (nw_proto != IPPROTO_DSTOPTS)
600 && (nw_proto != IPPROTO_AH)
601 && (nw_proto != IPPROTO_FRAGMENT))) {
602 /* It's either a terminal header (e.g., TCP, UDP) or one we
603 * don't understand. In either case, we're done with the
604 * packet, so use it to fill in 'nw_proto'. */
608 /* We only verify that at least 8 bytes of the next header are
609 * available, but many of these headers are longer. Ensure that
610 * accesses within the extension header are within those first 8
611 * bytes. All extension headers are required to be at least 8
613 if (OVS_UNLIKELY(size < 8)) {
617 if ((nw_proto == IPPROTO_HOPOPTS)
618 || (nw_proto == IPPROTO_ROUTING)
619 || (nw_proto == IPPROTO_DSTOPTS)) {
620 /* These headers, while different, have the fields we care
621 * about in the same location and with the same
623 const struct ip6_ext *ext_hdr = data;
624 nw_proto = ext_hdr->ip6e_nxt;
625 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
626 (ext_hdr->ip6e_len + 1) * 8))) {
629 } else if (nw_proto == IPPROTO_AH) {
630 /* A standard AH definition isn't available, but the fields
631 * we care about are in the same location as the generic
632 * option header--only the header length is calculated
634 const struct ip6_ext *ext_hdr = data;
635 nw_proto = ext_hdr->ip6e_nxt;
636 if (OVS_UNLIKELY(!data_try_pull(&data, &size,
637 (ext_hdr->ip6e_len + 2) * 4))) {
640 } else if (nw_proto == IPPROTO_FRAGMENT) {
641 const struct ovs_16aligned_ip6_frag *frag_hdr = data;
643 nw_proto = frag_hdr->ip6f_nxt;
644 if (!data_try_pull(&data, &size, sizeof *frag_hdr)) {
648 /* We only process the first fragment. */
649 if (frag_hdr->ip6f_offlg != htons(0)) {
650 nw_frag = FLOW_NW_FRAG_ANY;
651 if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) {
652 nw_frag |= FLOW_NW_FRAG_LATER;
653 nw_proto = IPPROTO_FRAGMENT;
660 if (dl_type == htons(ETH_TYPE_ARP) ||
661 dl_type == htons(ETH_TYPE_RARP)) {
662 struct eth_addr arp_buf[2];
663 const struct arp_eth_header *arp = (const struct arp_eth_header *)
664 data_try_pull(&data, &size, ARP_ETH_HEADER_LEN);
666 if (OVS_LIKELY(arp) && OVS_LIKELY(arp->ar_hrd == htons(1))
667 && OVS_LIKELY(arp->ar_pro == htons(ETH_TYPE_IP))
668 && OVS_LIKELY(arp->ar_hln == ETH_ADDR_LEN)
669 && OVS_LIKELY(arp->ar_pln == 4)) {
670 miniflow_push_be32(mf, nw_src,
671 get_16aligned_be32(&arp->ar_spa));
672 miniflow_push_be32(mf, nw_dst,
673 get_16aligned_be32(&arp->ar_tpa));
675 /* We only match on the lower 8 bits of the opcode. */
676 if (OVS_LIKELY(ntohs(arp->ar_op) <= 0xff)) {
677 miniflow_push_be32(mf, ipv6_label, 0); /* Pad with ARP. */
678 miniflow_push_be32(mf, nw_frag, htonl(ntohs(arp->ar_op)));
681 /* Must be adjacent. */
682 ASSERT_SEQUENTIAL(arp_sha, arp_tha);
684 arp_buf[0] = arp->ar_sha;
685 arp_buf[1] = arp->ar_tha;
686 miniflow_push_macs(mf, arp_sha, arp_buf);
687 miniflow_pad_to_64(mf, tcp_flags);
693 packet->l4_ofs = (char *)data - l2;
694 miniflow_push_be32(mf, nw_frag,
695 BYTES_TO_BE32(nw_frag, nw_tos, nw_ttl, nw_proto));
697 if (OVS_LIKELY(!(nw_frag & FLOW_NW_FRAG_LATER))) {
698 if (OVS_LIKELY(nw_proto == IPPROTO_TCP)) {
699 if (OVS_LIKELY(size >= TCP_HEADER_LEN)) {
700 const struct tcp_header *tcp = data;
702 miniflow_push_be32(mf, arp_tha.ea[2], 0);
703 miniflow_push_be32(mf, tcp_flags,
704 TCP_FLAGS_BE32(tcp->tcp_ctl));
705 miniflow_push_be16(mf, tp_src, tcp->tcp_src);
706 miniflow_push_be16(mf, tp_dst, tcp->tcp_dst);
707 miniflow_pad_to_64(mf, igmp_group_ip4);
709 } else if (OVS_LIKELY(nw_proto == IPPROTO_UDP)) {
710 if (OVS_LIKELY(size >= UDP_HEADER_LEN)) {
711 const struct udp_header *udp = data;
713 miniflow_push_be16(mf, tp_src, udp->udp_src);
714 miniflow_push_be16(mf, tp_dst, udp->udp_dst);
715 miniflow_pad_to_64(mf, igmp_group_ip4);
717 } else if (OVS_LIKELY(nw_proto == IPPROTO_SCTP)) {
718 if (OVS_LIKELY(size >= SCTP_HEADER_LEN)) {
719 const struct sctp_header *sctp = data;
721 miniflow_push_be16(mf, tp_src, sctp->sctp_src);
722 miniflow_push_be16(mf, tp_dst, sctp->sctp_dst);
723 miniflow_pad_to_64(mf, igmp_group_ip4);
725 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMP)) {
726 if (OVS_LIKELY(size >= ICMP_HEADER_LEN)) {
727 const struct icmp_header *icmp = data;
729 miniflow_push_be16(mf, tp_src, htons(icmp->icmp_type));
730 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp_code));
731 miniflow_pad_to_64(mf, igmp_group_ip4);
733 } else if (OVS_LIKELY(nw_proto == IPPROTO_IGMP)) {
734 if (OVS_LIKELY(size >= IGMP_HEADER_LEN)) {
735 const struct igmp_header *igmp = data;
737 miniflow_push_be16(mf, tp_src, htons(igmp->igmp_type));
738 miniflow_push_be16(mf, tp_dst, htons(igmp->igmp_code));
739 miniflow_push_be32(mf, igmp_group_ip4,
740 get_16aligned_be32(&igmp->group));
742 } else if (OVS_LIKELY(nw_proto == IPPROTO_ICMPV6)) {
743 if (OVS_LIKELY(size >= sizeof(struct icmp6_hdr))) {
744 const struct in6_addr *nd_target = NULL;
745 struct eth_addr arp_buf[2] = { { { { 0 } } } };
746 const struct icmp6_hdr *icmp = data_pull(&data, &size,
748 parse_icmpv6(&data, &size, icmp, &nd_target, arp_buf);
750 miniflow_push_words(mf, nd_target, nd_target,
751 sizeof *nd_target / sizeof(uint64_t));
753 miniflow_push_macs(mf, arp_sha, arp_buf);
754 miniflow_pad_to_64(mf, tcp_flags);
755 miniflow_push_be16(mf, tp_src, htons(icmp->icmp6_type));
756 miniflow_push_be16(mf, tp_dst, htons(icmp->icmp6_code));
757 miniflow_pad_to_64(mf, igmp_group_ip4);
765 /* For every bit of a field that is wildcarded in 'wildcards', sets the
766 * corresponding bit in 'flow' to zero. */
768 flow_zero_wildcards(struct flow *flow, const struct flow_wildcards *wildcards)
770 uint64_t *flow_u64 = (uint64_t *) flow;
771 const uint64_t *wc_u64 = (const uint64_t *) &wildcards->masks;
774 for (i = 0; i < FLOW_U64S; i++) {
775 flow_u64[i] &= wc_u64[i];
780 flow_unwildcard_tp_ports(const struct flow *flow, struct flow_wildcards *wc)
782 if (flow->nw_proto != IPPROTO_ICMP) {
783 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
784 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
786 wc->masks.tp_src = htons(0xff);
787 wc->masks.tp_dst = htons(0xff);
791 /* Initializes 'flow_metadata' with the metadata found in 'flow'. */
793 flow_get_metadata(const struct flow *flow, struct match *flow_metadata)
797 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 33);
799 match_init_catchall(flow_metadata);
800 if (flow->tunnel.tun_id != htonll(0)) {
801 match_set_tun_id(flow_metadata, flow->tunnel.tun_id);
803 if (flow->tunnel.flags & FLOW_TNL_PUB_F_MASK) {
804 match_set_tun_flags(flow_metadata,
805 flow->tunnel.flags & FLOW_TNL_PUB_F_MASK);
807 if (flow->tunnel.ip_src != htonl(0)) {
808 match_set_tun_src(flow_metadata, flow->tunnel.ip_src);
810 if (flow->tunnel.ip_dst != htonl(0)) {
811 match_set_tun_dst(flow_metadata, flow->tunnel.ip_dst);
813 if (flow->tunnel.gbp_id != htons(0)) {
814 match_set_tun_gbp_id(flow_metadata, flow->tunnel.gbp_id);
816 if (flow->tunnel.gbp_flags) {
817 match_set_tun_gbp_flags(flow_metadata, flow->tunnel.gbp_flags);
819 tun_metadata_get_fmd(&flow->tunnel, flow_metadata);
820 if (flow->metadata != htonll(0)) {
821 match_set_metadata(flow_metadata, flow->metadata);
824 for (i = 0; i < FLOW_N_REGS; i++) {
826 match_set_reg(flow_metadata, i, flow->regs[i]);
830 if (flow->pkt_mark != 0) {
831 match_set_pkt_mark(flow_metadata, flow->pkt_mark);
834 match_set_in_port(flow_metadata, flow->in_port.ofp_port);
838 flow_to_string(const struct flow *flow)
840 struct ds ds = DS_EMPTY_INITIALIZER;
841 flow_format(&ds, flow);
846 flow_tun_flag_to_string(uint32_t flags)
849 case FLOW_TNL_F_DONT_FRAGMENT:
851 case FLOW_TNL_F_CSUM:
863 format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
864 uint32_t flags, char del)
869 ds_put_char(ds, '0');
873 uint32_t bit = rightmost_1bit(flags);
876 s = bit_to_string(bit);
878 ds_put_format(ds, "%s%c", s, del);
887 ds_put_format(ds, "0x%"PRIx32"%c", bad, del);
893 format_flags_masked(struct ds *ds, const char *name,
894 const char *(*bit_to_string)(uint32_t), uint32_t flags,
895 uint32_t mask, uint32_t max_mask)
898 ds_put_format(ds, "%s=", name);
901 if (mask == max_mask) {
902 format_flags(ds, bit_to_string, flags, '|');
907 ds_put_cstr(ds, "0/0");
912 uint32_t bit = rightmost_1bit(mask);
913 const char *s = bit_to_string(bit);
915 ds_put_format(ds, "%s%s", (flags & bit) ? "+" : "-",
916 s ? s : "[Unknown]");
921 /* Scans a string 's' of flags to determine their numerical value and
922 * returns the number of characters parsed using 'bit_to_string' to
923 * lookup flag names. Scanning continues until the character 'end' is
926 * In the event of a failure, a negative error code will be returned. In
927 * addition, if 'res_string' is non-NULL then a descriptive string will
928 * be returned incorporating the identifying string 'field_name'. This
929 * error string must be freed by the caller.
931 * Upon success, the flag values will be stored in 'res_flags' and
932 * optionally 'res_mask', if it is non-NULL (if it is NULL then any masks
933 * present in the original string will be considered an error). The
934 * caller may restrict the acceptable set of values through the mask
937 parse_flags(const char *s, const char *(*bit_to_string)(uint32_t),
938 char end, const char *field_name, char **res_string,
939 uint32_t *res_flags, uint32_t allowed, uint32_t *res_mask)
944 /* Parse masked flags in numeric format? */
945 if (res_mask && ovs_scan(s, "%"SCNi32"/%"SCNi32"%n",
946 res_flags, res_mask, &n) && n > 0) {
947 if (*res_flags & ~allowed || *res_mask & ~allowed) {
955 if (res_mask && (*s == '+' || *s == '-')) {
956 uint32_t flags = 0, mask = 0;
958 /* Parse masked flags. */
959 while (s[0] != end) {
966 } else if (s[0] == '-') {
970 *res_string = xasprintf("%s: %s must be preceded by '+' "
971 "(for SET) or '-' (NOT SET)", s,
979 for (bit = 1; bit; bit <<= 1) {
980 const char *fname = bit_to_string(bit);
987 if (strncmp(s, fname, len) ||
988 (s[len] != '+' && s[len] != '-' && s[len] != end)) {
993 /* bit already set. */
995 *res_string = xasprintf("%s: Each %s flag can be "
996 "specified only once", s,
1001 if (!(bit & allowed)) {
1023 /* Parse unmasked flags. If a flag is present, it is set, otherwise
1025 while (s[n] != end) {
1026 unsigned long long int flags;
1030 if (ovs_scan(&s[n], "%lli%n", &flags, &n0)) {
1031 if (flags & ~allowed) {
1034 n += n0 + (s[n + n0] == '|');
1039 for (bit = 1; bit; bit <<= 1) {
1040 const char *name = bit_to_string(bit);
1048 if (!strncmp(s + n, name, len) &&
1049 (s[n + len] == '|' || s[n + len] == end)) {
1050 if (!(bit & allowed)) {
1054 n += len + (s[n + len] == '|');
1064 *res_flags = result;
1066 *res_mask = UINT32_MAX;
1075 *res_string = xasprintf("%s: unknown %s flag(s)", s, field_name);
1081 flow_format(struct ds *ds, const struct flow *flow)
1084 struct flow_wildcards *wc = &match.wc;
1086 match_wc_init(&match, flow);
1088 /* As this function is most often used for formatting a packet in a
1089 * packet-in message, skip formatting the packet context fields that are
1090 * all-zeroes to make the print-out easier on the eyes. This means that a
1091 * missing context field implies a zero value for that field. This is
1092 * similar to OpenFlow encoding of these fields, as the specification
1093 * states that all-zeroes context fields should not be encoded in the
1094 * packet-in messages. */
1095 if (!flow->in_port.ofp_port) {
1096 WC_UNMASK_FIELD(wc, in_port);
1098 if (!flow->skb_priority) {
1099 WC_UNMASK_FIELD(wc, skb_priority);
1101 if (!flow->pkt_mark) {
1102 WC_UNMASK_FIELD(wc, pkt_mark);
1104 if (!flow->recirc_id) {
1105 WC_UNMASK_FIELD(wc, recirc_id);
1107 if (!flow->dp_hash) {
1108 WC_UNMASK_FIELD(wc, dp_hash);
1110 for (int i = 0; i < FLOW_N_REGS; i++) {
1111 if (!flow->regs[i]) {
1112 WC_UNMASK_FIELD(wc, regs[i]);
1115 if (!flow->metadata) {
1116 WC_UNMASK_FIELD(wc, metadata);
1119 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
1123 flow_print(FILE *stream, const struct flow *flow)
1125 char *s = flow_to_string(flow);
1130 /* flow_wildcards functions. */
1132 /* Initializes 'wc' as a set of wildcards that matches every packet. */
1134 flow_wildcards_init_catchall(struct flow_wildcards *wc)
1136 memset(&wc->masks, 0, sizeof wc->masks);
1139 /* Converts a flow into flow wildcards. It sets the wildcard masks based on
1140 * the packet headers extracted to 'flow'. It will not set the mask for fields
1141 * that do not make sense for the packet type. OpenFlow-only metadata is
1142 * wildcarded, but other metadata is unconditionally exact-matched. */
1143 void flow_wildcards_init_for_packet(struct flow_wildcards *wc,
1144 const struct flow *flow)
1146 memset(&wc->masks, 0x0, sizeof wc->masks);
1148 /* Update this function whenever struct flow changes. */
1149 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 33);
1151 if (flow->tunnel.ip_dst) {
1152 if (flow->tunnel.flags & FLOW_TNL_F_KEY) {
1153 WC_MASK_FIELD(wc, tunnel.tun_id);
1155 WC_MASK_FIELD(wc, tunnel.ip_src);
1156 WC_MASK_FIELD(wc, tunnel.ip_dst);
1157 WC_MASK_FIELD(wc, tunnel.flags);
1158 WC_MASK_FIELD(wc, tunnel.ip_tos);
1159 WC_MASK_FIELD(wc, tunnel.ip_ttl);
1160 WC_MASK_FIELD(wc, tunnel.tp_src);
1161 WC_MASK_FIELD(wc, tunnel.tp_dst);
1162 WC_MASK_FIELD(wc, tunnel.gbp_id);
1163 WC_MASK_FIELD(wc, tunnel.gbp_flags);
1165 if (!(flow->tunnel.flags & FLOW_TNL_F_UDPIF)) {
1166 if (flow->tunnel.metadata.present.map) {
1167 wc->masks.tunnel.metadata.present.map =
1168 flow->tunnel.metadata.present.map;
1169 WC_MASK_FIELD(wc, tunnel.metadata.opts.u8);
1172 WC_MASK_FIELD(wc, tunnel.metadata.present.len);
1173 memset(wc->masks.tunnel.metadata.opts.gnv, 0xff,
1174 flow->tunnel.metadata.present.len);
1176 } else if (flow->tunnel.tun_id) {
1177 WC_MASK_FIELD(wc, tunnel.tun_id);
1180 /* metadata, regs, and conj_id wildcarded. */
1182 WC_MASK_FIELD(wc, skb_priority);
1183 WC_MASK_FIELD(wc, pkt_mark);
1184 WC_MASK_FIELD(wc, recirc_id);
1185 WC_MASK_FIELD(wc, dp_hash);
1186 WC_MASK_FIELD(wc, in_port);
1188 /* actset_output wildcarded. */
1190 WC_MASK_FIELD(wc, dl_dst);
1191 WC_MASK_FIELD(wc, dl_src);
1192 WC_MASK_FIELD(wc, dl_type);
1193 WC_MASK_FIELD(wc, vlan_tci);
1195 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1196 WC_MASK_FIELD(wc, nw_src);
1197 WC_MASK_FIELD(wc, nw_dst);
1198 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1199 WC_MASK_FIELD(wc, ipv6_src);
1200 WC_MASK_FIELD(wc, ipv6_dst);
1201 WC_MASK_FIELD(wc, ipv6_label);
1202 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1203 flow->dl_type == htons(ETH_TYPE_RARP)) {
1204 WC_MASK_FIELD(wc, nw_src);
1205 WC_MASK_FIELD(wc, nw_dst);
1206 WC_MASK_FIELD(wc, nw_proto);
1207 WC_MASK_FIELD(wc, arp_sha);
1208 WC_MASK_FIELD(wc, arp_tha);
1210 } else if (eth_type_mpls(flow->dl_type)) {
1211 for (int i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1212 WC_MASK_FIELD(wc, mpls_lse[i]);
1213 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1219 return; /* Unknown ethertype. */
1223 WC_MASK_FIELD(wc, nw_frag);
1224 WC_MASK_FIELD(wc, nw_tos);
1225 WC_MASK_FIELD(wc, nw_ttl);
1226 WC_MASK_FIELD(wc, nw_proto);
1228 /* No transport layer header in later fragments. */
1229 if (!(flow->nw_frag & FLOW_NW_FRAG_LATER) &&
1230 (flow->nw_proto == IPPROTO_ICMP ||
1231 flow->nw_proto == IPPROTO_ICMPV6 ||
1232 flow->nw_proto == IPPROTO_TCP ||
1233 flow->nw_proto == IPPROTO_UDP ||
1234 flow->nw_proto == IPPROTO_SCTP ||
1235 flow->nw_proto == IPPROTO_IGMP)) {
1236 WC_MASK_FIELD(wc, tp_src);
1237 WC_MASK_FIELD(wc, tp_dst);
1239 if (flow->nw_proto == IPPROTO_TCP) {
1240 WC_MASK_FIELD(wc, tcp_flags);
1241 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
1242 WC_MASK_FIELD(wc, arp_sha);
1243 WC_MASK_FIELD(wc, arp_tha);
1244 WC_MASK_FIELD(wc, nd_target);
1245 } else if (flow->nw_proto == IPPROTO_IGMP) {
1246 WC_MASK_FIELD(wc, igmp_group_ip4);
1251 /* Return a map of possible fields for a packet of the same type as 'flow'.
1252 * Including extra bits in the returned mask is not wrong, it is just less
1255 * This is a less precise version of flow_wildcards_init_for_packet() above. */
1257 flow_wc_map(const struct flow *flow, struct flowmap *map)
1259 /* Update this function whenever struct flow changes. */
1260 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 33);
1264 if (flow->tunnel.ip_dst) {
1265 FLOWMAP_SET__(map, tunnel, offsetof(struct flow_tnl, metadata));
1266 if (!(flow->tunnel.flags & FLOW_TNL_F_UDPIF)) {
1267 if (flow->tunnel.metadata.present.map) {
1268 FLOWMAP_SET(map, tunnel.metadata);
1271 FLOWMAP_SET(map, tunnel.metadata.present.len);
1272 FLOWMAP_SET__(map, tunnel.metadata.opts.gnv,
1273 flow->tunnel.metadata.present.len);
1277 /* Metadata fields that can appear on packet input. */
1278 FLOWMAP_SET(map, skb_priority);
1279 FLOWMAP_SET(map, pkt_mark);
1280 FLOWMAP_SET(map, recirc_id);
1281 FLOWMAP_SET(map, dp_hash);
1282 FLOWMAP_SET(map, in_port);
1283 FLOWMAP_SET(map, dl_dst);
1284 FLOWMAP_SET(map, dl_src);
1285 FLOWMAP_SET(map, dl_type);
1286 FLOWMAP_SET(map, vlan_tci);
1288 /* Ethertype-dependent fields. */
1289 if (OVS_LIKELY(flow->dl_type == htons(ETH_TYPE_IP))) {
1290 FLOWMAP_SET(map, nw_src);
1291 FLOWMAP_SET(map, nw_dst);
1292 FLOWMAP_SET(map, nw_proto);
1293 FLOWMAP_SET(map, nw_frag);
1294 FLOWMAP_SET(map, nw_tos);
1295 FLOWMAP_SET(map, nw_ttl);
1297 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_IGMP)) {
1298 FLOWMAP_SET(map, igmp_group_ip4);
1300 FLOWMAP_SET(map, tcp_flags);
1301 FLOWMAP_SET(map, tp_src);
1302 FLOWMAP_SET(map, tp_dst);
1304 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1305 FLOWMAP_SET(map, ipv6_src);
1306 FLOWMAP_SET(map, ipv6_dst);
1307 FLOWMAP_SET(map, ipv6_label);
1308 FLOWMAP_SET(map, nw_proto);
1309 FLOWMAP_SET(map, nw_frag);
1310 FLOWMAP_SET(map, nw_tos);
1311 FLOWMAP_SET(map, nw_ttl);
1313 if (OVS_UNLIKELY(flow->nw_proto == IPPROTO_ICMPV6)) {
1314 FLOWMAP_SET(map, nd_target);
1315 FLOWMAP_SET(map, arp_sha);
1316 FLOWMAP_SET(map, arp_tha);
1318 FLOWMAP_SET(map, tcp_flags);
1319 FLOWMAP_SET(map, tp_src);
1320 FLOWMAP_SET(map, tp_dst);
1322 } else if (eth_type_mpls(flow->dl_type)) {
1323 FLOWMAP_SET(map, mpls_lse);
1324 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1325 flow->dl_type == htons(ETH_TYPE_RARP)) {
1326 FLOWMAP_SET(map, nw_src);
1327 FLOWMAP_SET(map, nw_dst);
1328 FLOWMAP_SET(map, nw_proto);
1329 FLOWMAP_SET(map, arp_sha);
1330 FLOWMAP_SET(map, arp_tha);
1334 /* Clear the metadata and register wildcard masks. They are not packet
1337 flow_wildcards_clear_non_packet_fields(struct flow_wildcards *wc)
1339 /* Update this function whenever struct flow changes. */
1340 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 33);
1342 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
1343 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
1344 wc->masks.actset_output = 0;
1345 wc->masks.conj_id = 0;
1348 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
1351 flow_wildcards_is_catchall(const struct flow_wildcards *wc)
1353 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1356 for (i = 0; i < FLOW_U64S; i++) {
1364 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
1365 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
1366 * in 'src1' or 'src2' or both. */
1368 flow_wildcards_and(struct flow_wildcards *dst,
1369 const struct flow_wildcards *src1,
1370 const struct flow_wildcards *src2)
1372 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1373 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1374 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1377 for (i = 0; i < FLOW_U64S; i++) {
1378 dst_u64[i] = src1_u64[i] & src2_u64[i];
1382 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
1383 * is, a bit or a field is wildcarded in 'dst' if it is neither
1384 * wildcarded in 'src1' nor 'src2'. */
1386 flow_wildcards_or(struct flow_wildcards *dst,
1387 const struct flow_wildcards *src1,
1388 const struct flow_wildcards *src2)
1390 uint64_t *dst_u64 = (uint64_t *) &dst->masks;
1391 const uint64_t *src1_u64 = (const uint64_t *) &src1->masks;
1392 const uint64_t *src2_u64 = (const uint64_t *) &src2->masks;
1395 for (i = 0; i < FLOW_U64S; i++) {
1396 dst_u64[i] = src1_u64[i] | src2_u64[i];
1400 /* Returns a hash of the wildcards in 'wc'. */
1402 flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis)
1404 return flow_hash(&wc->masks, basis);
1407 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
1410 flow_wildcards_equal(const struct flow_wildcards *a,
1411 const struct flow_wildcards *b)
1413 return flow_equal(&a->masks, &b->masks);
1416 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
1417 * 'b', false otherwise. */
1419 flow_wildcards_has_extra(const struct flow_wildcards *a,
1420 const struct flow_wildcards *b)
1422 const uint64_t *a_u64 = (const uint64_t *) &a->masks;
1423 const uint64_t *b_u64 = (const uint64_t *) &b->masks;
1426 for (i = 0; i < FLOW_U64S; i++) {
1427 if ((a_u64[i] & b_u64[i]) != b_u64[i]) {
1434 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
1435 * in 'wc' do not need to be equal in 'a' and 'b'. */
1437 flow_equal_except(const struct flow *a, const struct flow *b,
1438 const struct flow_wildcards *wc)
1440 const uint64_t *a_u64 = (const uint64_t *) a;
1441 const uint64_t *b_u64 = (const uint64_t *) b;
1442 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1445 for (i = 0; i < FLOW_U64S; i++) {
1446 if ((a_u64[i] ^ b_u64[i]) & wc_u64[i]) {
1453 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1454 * (A 0-bit indicates a wildcard bit.) */
1456 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
1458 wc->masks.regs[idx] = mask;
1461 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1462 * (A 0-bit indicates a wildcard bit.) */
1464 flow_wildcards_set_xreg_mask(struct flow_wildcards *wc, int idx, uint64_t mask)
1466 flow_set_xreg(&wc->masks, idx, mask);
1469 /* Calculates the 5-tuple hash from the given miniflow.
1470 * This returns the same value as flow_hash_5tuple for the corresponding
1473 miniflow_hash_5tuple(const struct miniflow *flow, uint32_t basis)
1475 uint32_t hash = basis;
1478 ovs_be16 dl_type = MINIFLOW_GET_BE16(flow, dl_type);
1480 hash = hash_add(hash, MINIFLOW_GET_U8(flow, nw_proto));
1482 /* Separate loops for better optimization. */
1483 if (dl_type == htons(ETH_TYPE_IPV6)) {
1484 struct flowmap map = FLOWMAP_EMPTY_INITIALIZER;
1487 FLOWMAP_SET(&map, ipv6_src);
1488 FLOWMAP_SET(&map, ipv6_dst);
1490 MINIFLOW_FOR_EACH_IN_FLOWMAP(value, flow, map) {
1491 hash = hash_add64(hash, value);
1494 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_src));
1495 hash = hash_add(hash, MINIFLOW_GET_U32(flow, nw_dst));
1497 /* Add both ports at once. */
1498 hash = hash_add(hash, MINIFLOW_GET_U32(flow, tp_src));
1499 hash = hash_finish(hash, 42); /* Arbitrary number. */
1504 ASSERT_SEQUENTIAL_SAME_WORD(tp_src, tp_dst);
1505 ASSERT_SEQUENTIAL(ipv6_src, ipv6_dst);
1507 /* Calculates the 5-tuple hash from the given flow. */
1509 flow_hash_5tuple(const struct flow *flow, uint32_t basis)
1511 uint32_t hash = basis;
1514 hash = hash_add(hash, flow->nw_proto);
1516 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1517 const uint64_t *flow_u64 = (const uint64_t *)flow;
1518 int ofs = offsetof(struct flow, ipv6_src) / 8;
1519 int end = ofs + 2 * sizeof flow->ipv6_src / 8;
1521 for (;ofs < end; ofs++) {
1522 hash = hash_add64(hash, flow_u64[ofs]);
1525 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_src);
1526 hash = hash_add(hash, (OVS_FORCE uint32_t) flow->nw_dst);
1528 /* Add both ports at once. */
1529 hash = hash_add(hash,
1530 ((const uint32_t *)flow)[offsetof(struct flow, tp_src)
1531 / sizeof(uint32_t)]);
1532 hash = hash_finish(hash, 42); /* Arbitrary number. */
1537 /* Hashes 'flow' based on its L2 through L4 protocol information. */
1539 flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis)
1544 struct in6_addr ipv6_addr;
1549 struct eth_addr eth_addr;
1555 memset(&fields, 0, sizeof fields);
1556 for (i = 0; i < ARRAY_SIZE(fields.eth_addr.be16); i++) {
1557 fields.eth_addr.be16[i] = flow->dl_src.be16[i] ^ flow->dl_dst.be16[i];
1559 fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK);
1560 fields.eth_type = flow->dl_type;
1562 /* UDP source and destination port are not taken into account because they
1563 * will not necessarily be symmetric in a bidirectional flow. */
1564 if (fields.eth_type == htons(ETH_TYPE_IP)) {
1565 fields.ipv4_addr = flow->nw_src ^ flow->nw_dst;
1566 fields.ip_proto = flow->nw_proto;
1567 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1568 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1570 } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) {
1571 const uint8_t *a = &flow->ipv6_src.s6_addr[0];
1572 const uint8_t *b = &flow->ipv6_dst.s6_addr[0];
1573 uint8_t *ipv6_addr = &fields.ipv6_addr.s6_addr[0];
1575 for (i=0; i<16; i++) {
1576 ipv6_addr[i] = a[i] ^ b[i];
1578 fields.ip_proto = flow->nw_proto;
1579 if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_SCTP) {
1580 fields.tp_port = flow->tp_src ^ flow->tp_dst;
1583 return jhash_bytes(&fields, sizeof fields, basis);
1586 /* Hashes 'flow' based on its L3 through L4 protocol information */
1588 flow_hash_symmetric_l3l4(const struct flow *flow, uint32_t basis,
1591 uint32_t hash = basis;
1593 /* UDP source and destination port are also taken into account. */
1594 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1595 hash = hash_add(hash,
1596 (OVS_FORCE uint32_t) (flow->nw_src ^ flow->nw_dst));
1597 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1598 /* IPv6 addresses are 64-bit aligned inside struct flow. */
1599 const uint64_t *a = ALIGNED_CAST(uint64_t *, flow->ipv6_src.s6_addr);
1600 const uint64_t *b = ALIGNED_CAST(uint64_t *, flow->ipv6_dst.s6_addr);
1602 for (int i = 0; i < 4; i++) {
1603 hash = hash_add64(hash, a[i] ^ b[i]);
1606 /* Cannot hash non-IP flows */
1610 hash = hash_add(hash, flow->nw_proto);
1611 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP ||
1612 (inc_udp_ports && flow->nw_proto == IPPROTO_UDP)) {
1613 hash = hash_add(hash,
1614 (OVS_FORCE uint16_t) (flow->tp_src ^ flow->tp_dst));
1617 return hash_finish(hash, basis);
1620 /* Initialize a flow with random fields that matter for nx_hash_fields. */
1622 flow_random_hash_fields(struct flow *flow)
1624 uint16_t rnd = random_uint16();
1626 /* Initialize to all zeros. */
1627 memset(flow, 0, sizeof *flow);
1629 eth_addr_random(&flow->dl_src);
1630 eth_addr_random(&flow->dl_dst);
1632 flow->vlan_tci = (OVS_FORCE ovs_be16) (random_uint16() & VLAN_VID_MASK);
1634 /* Make most of the random flows IPv4, some IPv6, and rest random. */
1635 flow->dl_type = rnd < 0x8000 ? htons(ETH_TYPE_IP) :
1636 rnd < 0xc000 ? htons(ETH_TYPE_IPV6) : (OVS_FORCE ovs_be16)rnd;
1638 if (dl_type_is_ip_any(flow->dl_type)) {
1639 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1640 flow->nw_src = (OVS_FORCE ovs_be32)random_uint32();
1641 flow->nw_dst = (OVS_FORCE ovs_be32)random_uint32();
1643 random_bytes(&flow->ipv6_src, sizeof flow->ipv6_src);
1644 random_bytes(&flow->ipv6_dst, sizeof flow->ipv6_dst);
1646 /* Make most of IP flows TCP, some UDP or SCTP, and rest random. */
1647 rnd = random_uint16();
1648 flow->nw_proto = rnd < 0x8000 ? IPPROTO_TCP :
1649 rnd < 0xc000 ? IPPROTO_UDP :
1650 rnd < 0xd000 ? IPPROTO_SCTP : (uint8_t)rnd;
1651 if (flow->nw_proto == IPPROTO_TCP ||
1652 flow->nw_proto == IPPROTO_UDP ||
1653 flow->nw_proto == IPPROTO_SCTP) {
1654 flow->tp_src = (OVS_FORCE ovs_be16)random_uint16();
1655 flow->tp_dst = (OVS_FORCE ovs_be16)random_uint16();
1660 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
1662 flow_mask_hash_fields(const struct flow *flow, struct flow_wildcards *wc,
1663 enum nx_hash_fields fields)
1666 case NX_HASH_FIELDS_ETH_SRC:
1667 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1670 case NX_HASH_FIELDS_SYMMETRIC_L4:
1671 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1672 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1673 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1674 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1675 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1676 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1677 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1678 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1680 if (is_ip_any(flow)) {
1681 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1682 flow_unwildcard_tp_ports(flow, wc);
1684 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1687 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1688 if (is_ip_any(flow) && flow->nw_proto == IPPROTO_UDP) {
1689 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1690 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1693 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1694 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1695 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1696 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1697 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1698 memset(&wc->masks.ipv6_src, 0xff, sizeof wc->masks.ipv6_src);
1699 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
1701 break; /* non-IP flow */
1704 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1705 if (flow->nw_proto == IPPROTO_TCP || flow->nw_proto == IPPROTO_SCTP) {
1706 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1707 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1716 /* Hashes the portions of 'flow' designated by 'fields'. */
1718 flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields,
1723 case NX_HASH_FIELDS_ETH_SRC:
1724 return jhash_bytes(&flow->dl_src, sizeof flow->dl_src, basis);
1726 case NX_HASH_FIELDS_SYMMETRIC_L4:
1727 return flow_hash_symmetric_l4(flow, basis);
1729 case NX_HASH_FIELDS_SYMMETRIC_L3L4:
1730 return flow_hash_symmetric_l3l4(flow, basis, false);
1732 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP:
1733 return flow_hash_symmetric_l3l4(flow, basis, true);
1740 /* Returns a string representation of 'fields'. */
1742 flow_hash_fields_to_str(enum nx_hash_fields fields)
1745 case NX_HASH_FIELDS_ETH_SRC: return "eth_src";
1746 case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4";
1747 case NX_HASH_FIELDS_SYMMETRIC_L3L4: return "symmetric_l3l4";
1748 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP: return "symmetric_l3l4+udp";
1749 default: return "<unknown>";
1753 /* Returns true if the value of 'fields' is supported. Otherwise false. */
1755 flow_hash_fields_valid(enum nx_hash_fields fields)
1757 return fields == NX_HASH_FIELDS_ETH_SRC
1758 || fields == NX_HASH_FIELDS_SYMMETRIC_L4
1759 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4
1760 || fields == NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP;
1763 /* Returns a hash value for the bits of 'flow' that are active based on
1764 * 'wc', given 'basis'. */
1766 flow_hash_in_wildcards(const struct flow *flow,
1767 const struct flow_wildcards *wc, uint32_t basis)
1769 const uint64_t *wc_u64 = (const uint64_t *) &wc->masks;
1770 const uint64_t *flow_u64 = (const uint64_t *) flow;
1775 for (i = 0; i < FLOW_U64S; i++) {
1776 hash = hash_add64(hash, flow_u64[i] & wc_u64[i]);
1778 return hash_finish(hash, 8 * FLOW_U64S);
1781 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1782 * OpenFlow 1.0 "dl_vlan" value:
1784 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
1785 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
1786 * 'flow' previously matched packets without a VLAN header).
1788 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
1789 * without a VLAN tag.
1791 * - Other values of 'vid' should not be used. */
1793 flow_set_dl_vlan(struct flow *flow, ovs_be16 vid)
1795 if (vid == htons(OFP10_VLAN_NONE)) {
1796 flow->vlan_tci = htons(0);
1798 vid &= htons(VLAN_VID_MASK);
1799 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1800 flow->vlan_tci |= htons(VLAN_CFI) | vid;
1804 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1805 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
1808 flow_set_vlan_vid(struct flow *flow, ovs_be16 vid)
1810 ovs_be16 mask = htons(VLAN_VID_MASK | VLAN_CFI);
1811 flow->vlan_tci &= ~mask;
1812 flow->vlan_tci |= vid & mask;
1815 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
1818 * This function has no effect on the VLAN ID that 'flow' matches.
1820 * After calling this function, 'flow' will not match packets without a VLAN
1823 flow_set_vlan_pcp(struct flow *flow, uint8_t pcp)
1826 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1827 flow->vlan_tci |= htons((pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
1830 /* Returns the number of MPLS LSEs present in 'flow'
1832 * Returns 0 if the 'dl_type' of 'flow' is not an MPLS ethernet type.
1833 * Otherwise traverses 'flow''s MPLS label stack stopping at the
1834 * first entry that has the BoS bit set. If no such entry exists then
1835 * the maximum number of LSEs that can be stored in 'flow' is returned.
1838 flow_count_mpls_labels(const struct flow *flow, struct flow_wildcards *wc)
1840 /* dl_type is always masked. */
1841 if (eth_type_mpls(flow->dl_type)) {
1846 for (i = 0; i < FLOW_MAX_MPLS_LABELS; i++) {
1848 wc->masks.mpls_lse[i] |= htonl(MPLS_BOS_MASK);
1850 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
1853 if (flow->mpls_lse[i]) {
1863 /* Returns the number consecutive of MPLS LSEs, starting at the
1864 * innermost LSE, that are common in 'a' and 'b'.
1866 * 'an' must be flow_count_mpls_labels(a).
1867 * 'bn' must be flow_count_mpls_labels(b).
1870 flow_count_common_mpls_labels(const struct flow *a, int an,
1871 const struct flow *b, int bn,
1872 struct flow_wildcards *wc)
1874 int min_n = MIN(an, bn);
1879 int a_last = an - 1;
1880 int b_last = bn - 1;
1883 for (i = 0; i < min_n; i++) {
1885 wc->masks.mpls_lse[a_last - i] = OVS_BE32_MAX;
1886 wc->masks.mpls_lse[b_last - i] = OVS_BE32_MAX;
1888 if (a->mpls_lse[a_last - i] != b->mpls_lse[b_last - i]) {
1899 /* Adds a new outermost MPLS label to 'flow' and changes 'flow''s Ethernet type
1900 * to 'mpls_eth_type', which must be an MPLS Ethertype.
1902 * If the new label is the first MPLS label in 'flow', it is generated as;
1904 * - label: 2, if 'flow' is IPv6, otherwise 0.
1906 * - TTL: IPv4 or IPv6 TTL, if present and nonzero, otherwise 64.
1908 * - TC: IPv4 or IPv6 TOS, if present, otherwise 0.
1912 * If the new label is the second or later label MPLS label in 'flow', it is
1915 * - label: Copied from outer label.
1917 * - TTL: Copied from outer label.
1919 * - TC: Copied from outer label.
1923 * 'n' must be flow_count_mpls_labels(flow). 'n' must be less than
1924 * FLOW_MAX_MPLS_LABELS (because otherwise flow->mpls_lse[] would overflow).
1927 flow_push_mpls(struct flow *flow, int n, ovs_be16 mpls_eth_type,
1928 struct flow_wildcards *wc)
1930 ovs_assert(eth_type_mpls(mpls_eth_type));
1931 ovs_assert(n < FLOW_MAX_MPLS_LABELS);
1937 memset(&wc->masks.mpls_lse, 0xff, sizeof *wc->masks.mpls_lse * n);
1939 for (i = n; i >= 1; i--) {
1940 flow->mpls_lse[i] = flow->mpls_lse[i - 1];
1942 flow->mpls_lse[0] = (flow->mpls_lse[1] & htonl(~MPLS_BOS_MASK));
1944 int label = 0; /* IPv4 Explicit Null. */
1948 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1952 if (is_ip_any(flow)) {
1953 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1955 wc->masks.nw_tos |= IP_DSCP_MASK;
1956 wc->masks.nw_ttl = 0xff;
1964 flow->mpls_lse[0] = set_mpls_lse_values(ttl, tc, 1, htonl(label));
1966 /* Clear all L3 and L4 fields and dp_hash. */
1967 BUILD_ASSERT(FLOW_WC_SEQ == 33);
1968 memset((char *) flow + FLOW_SEGMENT_2_ENDS_AT, 0,
1969 sizeof(struct flow) - FLOW_SEGMENT_2_ENDS_AT);
1972 flow->dl_type = mpls_eth_type;
1975 /* Tries to remove the outermost MPLS label from 'flow'. Returns true if
1976 * successful, false otherwise. On success, sets 'flow''s Ethernet type to
1979 * 'n' must be flow_count_mpls_labels(flow). */
1981 flow_pop_mpls(struct flow *flow, int n, ovs_be16 eth_type,
1982 struct flow_wildcards *wc)
1987 /* Nothing to pop. */
1989 } else if (n == FLOW_MAX_MPLS_LABELS) {
1991 wc->masks.mpls_lse[n - 1] |= htonl(MPLS_BOS_MASK);
1993 if (!(flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK))) {
1994 /* Can't pop because don't know what to fill in mpls_lse[n - 1]. */
2000 memset(&wc->masks.mpls_lse[1], 0xff,
2001 sizeof *wc->masks.mpls_lse * (n - 1));
2003 for (i = 1; i < n; i++) {
2004 flow->mpls_lse[i - 1] = flow->mpls_lse[i];
2006 flow->mpls_lse[n - 1] = 0;
2007 flow->dl_type = eth_type;
2011 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
2012 * as an OpenFlow 1.1 "mpls_label" value. */
2014 flow_set_mpls_label(struct flow *flow, int idx, ovs_be32 label)
2016 set_mpls_lse_label(&flow->mpls_lse[idx], label);
2019 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
2022 flow_set_mpls_ttl(struct flow *flow, int idx, uint8_t ttl)
2024 set_mpls_lse_ttl(&flow->mpls_lse[idx], ttl);
2027 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
2030 flow_set_mpls_tc(struct flow *flow, int idx, uint8_t tc)
2032 set_mpls_lse_tc(&flow->mpls_lse[idx], tc);
2035 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
2037 flow_set_mpls_bos(struct flow *flow, int idx, uint8_t bos)
2039 set_mpls_lse_bos(&flow->mpls_lse[idx], bos);
2042 /* Sets the entire MPLS LSE. */
2044 flow_set_mpls_lse(struct flow *flow, int idx, ovs_be32 lse)
2046 flow->mpls_lse[idx] = lse;
2050 flow_compose_l4(struct dp_packet *p, const struct flow *flow)
2054 if (!(flow->nw_frag & FLOW_NW_FRAG_ANY)
2055 || !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
2056 if (flow->nw_proto == IPPROTO_TCP) {
2057 struct tcp_header *tcp;
2059 l4_len = sizeof *tcp;
2060 tcp = dp_packet_put_zeros(p, l4_len);
2061 tcp->tcp_src = flow->tp_src;
2062 tcp->tcp_dst = flow->tp_dst;
2063 tcp->tcp_ctl = TCP_CTL(ntohs(flow->tcp_flags), 5);
2064 } else if (flow->nw_proto == IPPROTO_UDP) {
2065 struct udp_header *udp;
2067 l4_len = sizeof *udp;
2068 udp = dp_packet_put_zeros(p, l4_len);
2069 udp->udp_src = flow->tp_src;
2070 udp->udp_dst = flow->tp_dst;
2071 } else if (flow->nw_proto == IPPROTO_SCTP) {
2072 struct sctp_header *sctp;
2074 l4_len = sizeof *sctp;
2075 sctp = dp_packet_put_zeros(p, l4_len);
2076 sctp->sctp_src = flow->tp_src;
2077 sctp->sctp_dst = flow->tp_dst;
2078 } else if (flow->nw_proto == IPPROTO_ICMP) {
2079 struct icmp_header *icmp;
2081 l4_len = sizeof *icmp;
2082 icmp = dp_packet_put_zeros(p, l4_len);
2083 icmp->icmp_type = ntohs(flow->tp_src);
2084 icmp->icmp_code = ntohs(flow->tp_dst);
2085 icmp->icmp_csum = csum(icmp, ICMP_HEADER_LEN);
2086 } else if (flow->nw_proto == IPPROTO_IGMP) {
2087 struct igmp_header *igmp;
2089 l4_len = sizeof *igmp;
2090 igmp = dp_packet_put_zeros(p, l4_len);
2091 igmp->igmp_type = ntohs(flow->tp_src);
2092 igmp->igmp_code = ntohs(flow->tp_dst);
2093 put_16aligned_be32(&igmp->group, flow->igmp_group_ip4);
2094 igmp->igmp_csum = csum(igmp, IGMP_HEADER_LEN);
2095 } else if (flow->nw_proto == IPPROTO_ICMPV6) {
2096 struct icmp6_hdr *icmp;
2098 l4_len = sizeof *icmp;
2099 icmp = dp_packet_put_zeros(p, l4_len);
2100 icmp->icmp6_type = ntohs(flow->tp_src);
2101 icmp->icmp6_code = ntohs(flow->tp_dst);
2103 if (icmp->icmp6_code == 0 &&
2104 (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT ||
2105 icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) {
2106 struct in6_addr *nd_target;
2107 struct ovs_nd_opt *nd_opt;
2109 l4_len += sizeof *nd_target;
2110 nd_target = dp_packet_put_zeros(p, sizeof *nd_target);
2111 *nd_target = flow->nd_target;
2113 if (!eth_addr_is_zero(flow->arp_sha)) {
2115 nd_opt = dp_packet_put_zeros(p, 8);
2116 nd_opt->nd_opt_len = 1;
2117 nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
2118 nd_opt->nd_opt_mac = flow->arp_sha;
2120 if (!eth_addr_is_zero(flow->arp_tha)) {
2122 nd_opt = dp_packet_put_zeros(p, 8);
2123 nd_opt->nd_opt_len = 1;
2124 nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR;
2125 nd_opt->nd_opt_mac = flow->arp_tha;
2128 icmp->icmp6_cksum = (OVS_FORCE uint16_t)
2129 csum(icmp, (char *)dp_packet_tail(p) - (char *)icmp);
2135 /* Puts into 'b' a packet that flow_extract() would parse as having the given
2138 * (This is useful only for testing, obviously, and the packet isn't really
2139 * valid. It hasn't got some checksums filled in, for one, and lots of fields
2140 * are just zeroed.) */
2142 flow_compose(struct dp_packet *p, const struct flow *flow)
2146 /* eth_compose() sets l3 pointer and makes sure it is 32-bit aligned. */
2147 eth_compose(p, flow->dl_dst, flow->dl_src, ntohs(flow->dl_type), 0);
2148 if (flow->dl_type == htons(FLOW_DL_TYPE_NONE)) {
2149 struct eth_header *eth = dp_packet_l2(p);
2150 eth->eth_type = htons(dp_packet_size(p));
2154 if (flow->vlan_tci & htons(VLAN_CFI)) {
2155 eth_push_vlan(p, htons(ETH_TYPE_VLAN), flow->vlan_tci);
2158 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2159 struct ip_header *ip;
2161 ip = dp_packet_put_zeros(p, sizeof *ip);
2162 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
2163 ip->ip_tos = flow->nw_tos;
2164 ip->ip_ttl = flow->nw_ttl;
2165 ip->ip_proto = flow->nw_proto;
2166 put_16aligned_be32(&ip->ip_src, flow->nw_src);
2167 put_16aligned_be32(&ip->ip_dst, flow->nw_dst);
2169 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2170 ip->ip_frag_off |= htons(IP_MORE_FRAGMENTS);
2171 if (flow->nw_frag & FLOW_NW_FRAG_LATER) {
2172 ip->ip_frag_off |= htons(100);
2176 dp_packet_set_l4(p, dp_packet_tail(p));
2178 l4_len = flow_compose_l4(p, flow);
2180 ip = dp_packet_l3(p);
2181 ip->ip_tot_len = htons(p->l4_ofs - p->l3_ofs + l4_len);
2182 ip->ip_csum = csum(ip, sizeof *ip);
2183 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2184 struct ovs_16aligned_ip6_hdr *nh;
2186 nh = dp_packet_put_zeros(p, sizeof *nh);
2187 put_16aligned_be32(&nh->ip6_flow, htonl(6 << 28) |
2188 htonl(flow->nw_tos << 20) | flow->ipv6_label);
2189 nh->ip6_hlim = flow->nw_ttl;
2190 nh->ip6_nxt = flow->nw_proto;
2192 memcpy(&nh->ip6_src, &flow->ipv6_src, sizeof(nh->ip6_src));
2193 memcpy(&nh->ip6_dst, &flow->ipv6_dst, sizeof(nh->ip6_dst));
2195 dp_packet_set_l4(p, dp_packet_tail(p));
2197 l4_len = flow_compose_l4(p, flow);
2199 nh = dp_packet_l3(p);
2200 nh->ip6_plen = htons(l4_len);
2201 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
2202 flow->dl_type == htons(ETH_TYPE_RARP)) {
2203 struct arp_eth_header *arp;
2205 arp = dp_packet_put_zeros(p, sizeof *arp);
2206 dp_packet_set_l3(p, arp);
2207 arp->ar_hrd = htons(1);
2208 arp->ar_pro = htons(ETH_TYPE_IP);
2209 arp->ar_hln = ETH_ADDR_LEN;
2211 arp->ar_op = htons(flow->nw_proto);
2213 if (flow->nw_proto == ARP_OP_REQUEST ||
2214 flow->nw_proto == ARP_OP_REPLY) {
2215 put_16aligned_be32(&arp->ar_spa, flow->nw_src);
2216 put_16aligned_be32(&arp->ar_tpa, flow->nw_dst);
2217 arp->ar_sha = flow->arp_sha;
2218 arp->ar_tha = flow->arp_tha;
2222 if (eth_type_mpls(flow->dl_type)) {
2225 p->l2_5_ofs = p->l3_ofs;
2226 for (n = 1; n < FLOW_MAX_MPLS_LABELS; n++) {
2227 if (flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK)) {
2232 push_mpls(p, flow->dl_type, flow->mpls_lse[--n]);
2237 /* Compressed flow. */
2239 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
2240 * the caller. The caller must have already computed 'dst->map' properly to
2241 * indicate the significant uint64_t elements of 'src'.
2243 * Normally the significant elements are the ones that are non-zero. However,
2244 * when a miniflow is initialized from a (mini)mask, the values can be zeroes,
2245 * so that the flow and mask always have the same maps. */
2247 miniflow_init(struct miniflow *dst, const struct flow *src)
2249 uint64_t *dst_u64 = miniflow_values(dst);
2252 FLOWMAP_FOR_EACH_INDEX(idx, dst->map) {
2253 *dst_u64++ = flow_u64_value(src, idx);
2257 /* Initialize the maps of 'flow' from 'src'. */
2259 miniflow_map_init(struct miniflow *flow, const struct flow *src)
2261 /* Initialize map, counting the number of nonzero elements. */
2262 flowmap_init(&flow->map);
2263 for (size_t i = 0; i < FLOW_U64S; i++) {
2264 if (flow_u64_value(src, i)) {
2265 flowmap_set(&flow->map, i, 1);
2270 /* Allocates 'n' count of miniflows, consecutive in memory, initializing the
2271 * map of each from 'src'.
2272 * Returns the size of the miniflow data. */
2274 miniflow_alloc(struct miniflow *dsts[], size_t n, const struct miniflow *src)
2276 size_t n_values = miniflow_n_values(src);
2277 size_t data_size = MINIFLOW_VALUES_SIZE(n_values);
2278 struct miniflow *dst = xmalloc(n * (sizeof *src + data_size));
2281 COVERAGE_INC(miniflow_malloc);
2283 for (i = 0; i < n; i++) {
2284 *dst = *src; /* Copy maps. */
2286 dst += 1; /* Just past the maps. */
2287 dst = (struct miniflow *)((uint64_t *)dst + n_values); /* Skip data. */
2292 /* Returns a miniflow copy of 'src'. The caller must eventually free() the
2293 * returned miniflow. */
2295 miniflow_create(const struct flow *src)
2297 struct miniflow tmp;
2298 struct miniflow *dst;
2300 miniflow_map_init(&tmp, src);
2302 miniflow_alloc(&dst, 1, &tmp);
2303 miniflow_init(dst, src);
2307 /* Initializes 'dst' as a copy of 'src'. The caller must have allocated
2308 * 'dst' to have inline space for 'n_values' data in 'src'. */
2310 miniflow_clone(struct miniflow *dst, const struct miniflow *src,
2313 *dst = *src; /* Copy maps. */
2314 memcpy(miniflow_values(dst), miniflow_get_values(src),
2315 MINIFLOW_VALUES_SIZE(n_values));
2318 /* Initializes 'dst' as a copy of 'src'. */
2320 miniflow_expand(const struct miniflow *src, struct flow *dst)
2322 memset(dst, 0, sizeof *dst);
2323 flow_union_with_miniflow(dst, src);
2326 /* Returns true if 'a' and 'b' are equal miniflows, false otherwise. */
2328 miniflow_equal(const struct miniflow *a, const struct miniflow *b)
2330 const uint64_t *ap = miniflow_get_values(a);
2331 const uint64_t *bp = miniflow_get_values(b);
2333 /* This is mostly called after a matching hash, so it is highly likely that
2334 * the maps are equal as well. */
2335 if (OVS_LIKELY(flowmap_equal(a->map, b->map))) {
2336 return !memcmp(ap, bp, miniflow_n_values(a) * sizeof *ap);
2340 FLOWMAP_FOR_EACH_INDEX (idx, flowmap_or(a->map, b->map)) {
2341 if ((flowmap_is_set(&a->map, idx) ? *ap++ : 0)
2342 != (flowmap_is_set(&b->map, idx) ? *bp++ : 0)) {
2351 /* Returns false if 'a' and 'b' differ at the places where there are 1-bits
2352 * in 'mask', true otherwise. */
2354 miniflow_equal_in_minimask(const struct miniflow *a, const struct miniflow *b,
2355 const struct minimask *mask)
2357 const uint64_t *p = miniflow_get_values(&mask->masks);
2360 FLOWMAP_FOR_EACH_INDEX(idx, mask->masks.map) {
2361 if ((miniflow_get(a, idx) ^ miniflow_get(b, idx)) & *p++) {
2369 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
2370 * in 'mask', false if they differ. */
2372 miniflow_equal_flow_in_minimask(const struct miniflow *a, const struct flow *b,
2373 const struct minimask *mask)
2375 const uint64_t *p = miniflow_get_values(&mask->masks);
2378 FLOWMAP_FOR_EACH_INDEX(idx, mask->masks.map) {
2379 if ((miniflow_get(a, idx) ^ flow_u64_value(b, idx)) & *p++) {
2389 minimask_init(struct minimask *mask, const struct flow_wildcards *wc)
2391 miniflow_init(&mask->masks, &wc->masks);
2394 /* Returns a minimask copy of 'wc'. The caller must eventually free the
2395 * returned minimask with free(). */
2397 minimask_create(const struct flow_wildcards *wc)
2399 return (struct minimask *)miniflow_create(&wc->masks);
2402 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
2404 * The caller must provide room for FLOW_U64S "uint64_t"s in 'storage', which
2405 * must follow '*dst_' in memory, for use by 'dst_'. The caller must *not*
2406 * free 'dst_' free(). */
2408 minimask_combine(struct minimask *dst_,
2409 const struct minimask *a_, const struct minimask *b_,
2410 uint64_t storage[FLOW_U64S])
2412 struct miniflow *dst = &dst_->masks;
2413 uint64_t *dst_values = storage;
2414 const struct miniflow *a = &a_->masks;
2415 const struct miniflow *b = &b_->masks;
2418 flowmap_init(&dst->map);
2420 FLOWMAP_FOR_EACH_INDEX(idx, flowmap_and(a->map, b->map)) {
2421 /* Both 'a' and 'b' have non-zero data at 'idx'. */
2422 uint64_t mask = *miniflow_get__(a, idx) & *miniflow_get__(b, idx);
2425 flowmap_set(&dst->map, idx, 1);
2426 *dst_values++ = mask;
2431 /* Initializes 'wc' as a copy of 'mask'. */
2433 minimask_expand(const struct minimask *mask, struct flow_wildcards *wc)
2435 miniflow_expand(&mask->masks, &wc->masks);
2438 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise.
2439 * Minimasks may not have zero data values, so for the minimasks to be the
2440 * same, they need to have the same map and the same data values. */
2442 minimask_equal(const struct minimask *a, const struct minimask *b)
2444 return !memcmp(a, b, sizeof *a
2445 + MINIFLOW_VALUES_SIZE(miniflow_n_values(&a->masks)));
2448 /* Returns true if at least one bit matched by 'b' is wildcarded by 'a',
2449 * false otherwise. */
2451 minimask_has_extra(const struct minimask *a, const struct minimask *b)
2453 const uint64_t *bp = miniflow_get_values(&b->masks);
2456 FLOWMAP_FOR_EACH_INDEX(idx, b->masks.map) {
2457 uint64_t b_u64 = *bp++;
2459 /* 'b_u64' is non-zero, check if the data in 'a' is either zero
2460 * or misses some of the bits in 'b_u64'. */
2461 if (!MINIFLOW_IN_MAP(&a->masks, idx)
2462 || ((*miniflow_get__(&a->masks, idx) & b_u64) != b_u64)) {
2463 return true; /* 'a' wildcards some bits 'b' doesn't. */