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_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, OFFSETOFEND(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 (flow_tnl_dst_is_set(&md
->tunnel
)) {
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
));
480 if (md
->recirc_id
|| md
->ct_state
) {
481 miniflow_push_uint32(mf
, recirc_id
, md
->recirc_id
);
482 miniflow_push_uint16(mf
, ct_state
, md
->ct_state
);
483 miniflow_push_uint16(mf
, ct_zone
, md
->ct_zone
);
487 miniflow_push_uint32(mf
, ct_mark
, md
->ct_mark
);
488 miniflow_pad_to_64(mf
, ct_mark
);
490 if (!ovs_u128_is_zero(&md
->ct_label
)) {
491 miniflow_push_words(mf
, ct_label
, &md
->ct_label
,
492 sizeof md
->ct_label
/ sizeof(uint64_t));
496 /* Initialize packet's layer pointer and offsets. */
498 dp_packet_reset_offsets(packet
);
500 /* Must have full Ethernet header to proceed. */
501 if (OVS_UNLIKELY(size
< sizeof(struct eth_header
))) {
507 ASSERT_SEQUENTIAL(dl_dst
, dl_src
);
508 miniflow_push_macs(mf
, dl_dst
, data
);
509 /* dl_type, vlan_tci. */
510 vlan_tci
= parse_vlan(&data
, &size
);
511 dl_type
= parse_ethertype(&data
, &size
);
512 miniflow_push_be16(mf
, dl_type
, dl_type
);
513 miniflow_push_be16(mf
, vlan_tci
, vlan_tci
);
517 if (OVS_UNLIKELY(eth_type_mpls(dl_type
))) {
519 const void *mpls
= data
;
521 packet
->l2_5_ofs
= (char *)data
- l2
;
522 count
= parse_mpls(&data
, &size
);
523 miniflow_push_words_32(mf
, mpls_lse
, mpls
, count
);
527 packet
->l3_ofs
= (char *)data
- l2
;
530 if (OVS_LIKELY(dl_type
== htons(ETH_TYPE_IP
))) {
531 const struct ip_header
*nh
= data
;
535 if (OVS_UNLIKELY(size
< IP_HEADER_LEN
)) {
538 ip_len
= IP_IHL(nh
->ip_ihl_ver
) * 4;
540 if (OVS_UNLIKELY(ip_len
< IP_HEADER_LEN
)) {
543 if (OVS_UNLIKELY(size
< ip_len
)) {
546 tot_len
= ntohs(nh
->ip_tot_len
);
547 if (OVS_UNLIKELY(tot_len
> size
)) {
550 if (OVS_UNLIKELY(size
- tot_len
> UINT8_MAX
)) {
553 dp_packet_set_l2_pad_size(packet
, size
- tot_len
);
554 size
= tot_len
; /* Never pull padding. */
556 /* Push both source and destination address at once. */
557 miniflow_push_words(mf
, nw_src
, &nh
->ip_src
, 1);
559 miniflow_push_be32(mf
, ipv6_label
, 0); /* Padding for IPv4. */
563 nw_proto
= nh
->ip_proto
;
564 if (OVS_UNLIKELY(IP_IS_FRAGMENT(nh
->ip_frag_off
))) {
565 nw_frag
= FLOW_NW_FRAG_ANY
;
566 if (nh
->ip_frag_off
& htons(IP_FRAG_OFF_MASK
)) {
567 nw_frag
|= FLOW_NW_FRAG_LATER
;
570 data_pull(&data
, &size
, ip_len
);
571 } else if (dl_type
== htons(ETH_TYPE_IPV6
)) {
572 const struct ovs_16aligned_ip6_hdr
*nh
;
576 if (OVS_UNLIKELY(size
< sizeof *nh
)) {
579 nh
= data_pull(&data
, &size
, sizeof *nh
);
581 plen
= ntohs(nh
->ip6_plen
);
582 if (OVS_UNLIKELY(plen
> size
)) {
585 /* Jumbo Payload option not supported yet. */
586 if (OVS_UNLIKELY(size
- plen
> UINT8_MAX
)) {
589 dp_packet_set_l2_pad_size(packet
, size
- plen
);
590 size
= plen
; /* Never pull padding. */
592 miniflow_push_words(mf
, ipv6_src
, &nh
->ip6_src
,
593 sizeof nh
->ip6_src
/ 8);
594 miniflow_push_words(mf
, ipv6_dst
, &nh
->ip6_dst
,
595 sizeof nh
->ip6_dst
/ 8);
597 tc_flow
= get_16aligned_be32(&nh
->ip6_flow
);
599 ovs_be32 label
= tc_flow
& htonl(IPV6_LABEL_MASK
);
600 miniflow_push_be32(mf
, ipv6_label
, label
);
603 nw_tos
= ntohl(tc_flow
) >> 20;
604 nw_ttl
= nh
->ip6_hlim
;
605 nw_proto
= nh
->ip6_nxt
;
608 if (OVS_LIKELY((nw_proto
!= IPPROTO_HOPOPTS
)
609 && (nw_proto
!= IPPROTO_ROUTING
)
610 && (nw_proto
!= IPPROTO_DSTOPTS
)
611 && (nw_proto
!= IPPROTO_AH
)
612 && (nw_proto
!= IPPROTO_FRAGMENT
))) {
613 /* It's either a terminal header (e.g., TCP, UDP) or one we
614 * don't understand. In either case, we're done with the
615 * packet, so use it to fill in 'nw_proto'. */
619 /* We only verify that at least 8 bytes of the next header are
620 * available, but many of these headers are longer. Ensure that
621 * accesses within the extension header are within those first 8
622 * bytes. All extension headers are required to be at least 8
624 if (OVS_UNLIKELY(size
< 8)) {
628 if ((nw_proto
== IPPROTO_HOPOPTS
)
629 || (nw_proto
== IPPROTO_ROUTING
)
630 || (nw_proto
== IPPROTO_DSTOPTS
)) {
631 /* These headers, while different, have the fields we care
632 * about in the same location and with the same
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
+ 1) * 8))) {
640 } else if (nw_proto
== IPPROTO_AH
) {
641 /* A standard AH definition isn't available, but the fields
642 * we care about are in the same location as the generic
643 * option header--only the header length is calculated
645 const struct ip6_ext
*ext_hdr
= data
;
646 nw_proto
= ext_hdr
->ip6e_nxt
;
647 if (OVS_UNLIKELY(!data_try_pull(&data
, &size
,
648 (ext_hdr
->ip6e_len
+ 2) * 4))) {
651 } else if (nw_proto
== IPPROTO_FRAGMENT
) {
652 const struct ovs_16aligned_ip6_frag
*frag_hdr
= data
;
654 nw_proto
= frag_hdr
->ip6f_nxt
;
655 if (!data_try_pull(&data
, &size
, sizeof *frag_hdr
)) {
659 /* We only process the first fragment. */
660 if (frag_hdr
->ip6f_offlg
!= htons(0)) {
661 nw_frag
= FLOW_NW_FRAG_ANY
;
662 if ((frag_hdr
->ip6f_offlg
& IP6F_OFF_MASK
) != htons(0)) {
663 nw_frag
|= FLOW_NW_FRAG_LATER
;
664 nw_proto
= IPPROTO_FRAGMENT
;
671 if (dl_type
== htons(ETH_TYPE_ARP
) ||
672 dl_type
== htons(ETH_TYPE_RARP
)) {
673 struct eth_addr arp_buf
[2];
674 const struct arp_eth_header
*arp
= (const struct arp_eth_header
*)
675 data_try_pull(&data
, &size
, ARP_ETH_HEADER_LEN
);
677 if (OVS_LIKELY(arp
) && OVS_LIKELY(arp
->ar_hrd
== htons(1))
678 && OVS_LIKELY(arp
->ar_pro
== htons(ETH_TYPE_IP
))
679 && OVS_LIKELY(arp
->ar_hln
== ETH_ADDR_LEN
)
680 && OVS_LIKELY(arp
->ar_pln
== 4)) {
681 miniflow_push_be32(mf
, nw_src
,
682 get_16aligned_be32(&arp
->ar_spa
));
683 miniflow_push_be32(mf
, nw_dst
,
684 get_16aligned_be32(&arp
->ar_tpa
));
686 /* We only match on the lower 8 bits of the opcode. */
687 if (OVS_LIKELY(ntohs(arp
->ar_op
) <= 0xff)) {
688 miniflow_push_be32(mf
, ipv6_label
, 0); /* Pad with ARP. */
689 miniflow_push_be32(mf
, nw_frag
, htonl(ntohs(arp
->ar_op
)));
692 /* Must be adjacent. */
693 ASSERT_SEQUENTIAL(arp_sha
, arp_tha
);
695 arp_buf
[0] = arp
->ar_sha
;
696 arp_buf
[1] = arp
->ar_tha
;
697 miniflow_push_macs(mf
, arp_sha
, arp_buf
);
698 miniflow_pad_to_64(mf
, arp_tha
);
704 packet
->l4_ofs
= (char *)data
- l2
;
705 miniflow_push_be32(mf
, nw_frag
,
706 BYTES_TO_BE32(nw_frag
, nw_tos
, nw_ttl
, nw_proto
));
708 if (OVS_LIKELY(!(nw_frag
& FLOW_NW_FRAG_LATER
))) {
709 if (OVS_LIKELY(nw_proto
== IPPROTO_TCP
)) {
710 if (OVS_LIKELY(size
>= TCP_HEADER_LEN
)) {
711 const struct tcp_header
*tcp
= data
;
713 miniflow_push_be32(mf
, arp_tha
.ea
[2], 0);
714 miniflow_push_be32(mf
, tcp_flags
,
715 TCP_FLAGS_BE32(tcp
->tcp_ctl
));
716 miniflow_push_be16(mf
, tp_src
, tcp
->tcp_src
);
717 miniflow_push_be16(mf
, tp_dst
, tcp
->tcp_dst
);
718 miniflow_pad_to_64(mf
, tp_dst
);
720 } else if (OVS_LIKELY(nw_proto
== IPPROTO_UDP
)) {
721 if (OVS_LIKELY(size
>= UDP_HEADER_LEN
)) {
722 const struct udp_header
*udp
= data
;
724 miniflow_push_be16(mf
, tp_src
, udp
->udp_src
);
725 miniflow_push_be16(mf
, tp_dst
, udp
->udp_dst
);
726 miniflow_pad_to_64(mf
, tp_dst
);
728 } else if (OVS_LIKELY(nw_proto
== IPPROTO_SCTP
)) {
729 if (OVS_LIKELY(size
>= SCTP_HEADER_LEN
)) {
730 const struct sctp_header
*sctp
= data
;
732 miniflow_push_be16(mf
, tp_src
, sctp
->sctp_src
);
733 miniflow_push_be16(mf
, tp_dst
, sctp
->sctp_dst
);
734 miniflow_pad_to_64(mf
, tp_dst
);
736 } else if (OVS_LIKELY(nw_proto
== IPPROTO_ICMP
)) {
737 if (OVS_LIKELY(size
>= ICMP_HEADER_LEN
)) {
738 const struct icmp_header
*icmp
= data
;
740 miniflow_push_be16(mf
, tp_src
, htons(icmp
->icmp_type
));
741 miniflow_push_be16(mf
, tp_dst
, htons(icmp
->icmp_code
));
742 miniflow_pad_to_64(mf
, tp_dst
);
744 } else if (OVS_LIKELY(nw_proto
== IPPROTO_IGMP
)) {
745 if (OVS_LIKELY(size
>= IGMP_HEADER_LEN
)) {
746 const struct igmp_header
*igmp
= data
;
748 miniflow_push_be16(mf
, tp_src
, htons(igmp
->igmp_type
));
749 miniflow_push_be16(mf
, tp_dst
, htons(igmp
->igmp_code
));
750 miniflow_push_be32(mf
, igmp_group_ip4
,
751 get_16aligned_be32(&igmp
->group
));
753 } else if (OVS_LIKELY(nw_proto
== IPPROTO_ICMPV6
)) {
754 if (OVS_LIKELY(size
>= sizeof(struct icmp6_hdr
))) {
755 const struct in6_addr
*nd_target
= NULL
;
756 struct eth_addr arp_buf
[2] = { { { { 0 } } } };
757 const struct icmp6_hdr
*icmp
= data_pull(&data
, &size
,
759 parse_icmpv6(&data
, &size
, icmp
, &nd_target
, arp_buf
);
761 miniflow_push_words(mf
, nd_target
, nd_target
,
762 sizeof *nd_target
/ sizeof(uint64_t));
764 miniflow_push_macs(mf
, arp_sha
, arp_buf
);
765 miniflow_pad_to_64(mf
, arp_tha
);
766 miniflow_push_be16(mf
, tp_src
, htons(icmp
->icmp6_type
));
767 miniflow_push_be16(mf
, tp_dst
, htons(icmp
->icmp6_code
));
768 miniflow_pad_to_64(mf
, tp_dst
);
776 /* For every bit of a field that is wildcarded in 'wildcards', sets the
777 * corresponding bit in 'flow' to zero. */
779 flow_zero_wildcards(struct flow
*flow
, const struct flow_wildcards
*wildcards
)
781 uint64_t *flow_u64
= (uint64_t *) flow
;
782 const uint64_t *wc_u64
= (const uint64_t *) &wildcards
->masks
;
785 for (i
= 0; i
< FLOW_U64S
; i
++) {
786 flow_u64
[i
] &= wc_u64
[i
];
791 flow_unwildcard_tp_ports(const struct flow
*flow
, struct flow_wildcards
*wc
)
793 if (flow
->nw_proto
!= IPPROTO_ICMP
) {
794 memset(&wc
->masks
.tp_src
, 0xff, sizeof wc
->masks
.tp_src
);
795 memset(&wc
->masks
.tp_dst
, 0xff, sizeof wc
->masks
.tp_dst
);
797 wc
->masks
.tp_src
= htons(0xff);
798 wc
->masks
.tp_dst
= htons(0xff);
802 /* Initializes 'flow_metadata' with the metadata found in 'flow'. */
804 flow_get_metadata(const struct flow
*flow
, struct match
*flow_metadata
)
808 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 35);
810 match_init_catchall(flow_metadata
);
811 if (flow
->tunnel
.tun_id
!= htonll(0)) {
812 match_set_tun_id(flow_metadata
, flow
->tunnel
.tun_id
);
814 if (flow
->tunnel
.flags
& FLOW_TNL_PUB_F_MASK
) {
815 match_set_tun_flags(flow_metadata
,
816 flow
->tunnel
.flags
& FLOW_TNL_PUB_F_MASK
);
818 if (flow
->tunnel
.ip_src
) {
819 match_set_tun_src(flow_metadata
, flow
->tunnel
.ip_src
);
821 if (flow
->tunnel
.ip_dst
) {
822 match_set_tun_dst(flow_metadata
, flow
->tunnel
.ip_dst
);
824 if (ipv6_addr_is_set(&flow
->tunnel
.ipv6_src
)) {
825 match_set_tun_ipv6_src(flow_metadata
, &flow
->tunnel
.ipv6_src
);
827 if (ipv6_addr_is_set(&flow
->tunnel
.ipv6_dst
)) {
828 match_set_tun_ipv6_dst(flow_metadata
, &flow
->tunnel
.ipv6_dst
);
830 if (flow
->tunnel
.gbp_id
!= htons(0)) {
831 match_set_tun_gbp_id(flow_metadata
, flow
->tunnel
.gbp_id
);
833 if (flow
->tunnel
.gbp_flags
) {
834 match_set_tun_gbp_flags(flow_metadata
, flow
->tunnel
.gbp_flags
);
836 tun_metadata_get_fmd(&flow
->tunnel
, flow_metadata
);
837 if (flow
->metadata
!= htonll(0)) {
838 match_set_metadata(flow_metadata
, flow
->metadata
);
841 for (i
= 0; i
< FLOW_N_REGS
; i
++) {
843 match_set_reg(flow_metadata
, i
, flow
->regs
[i
]);
847 if (flow
->pkt_mark
!= 0) {
848 match_set_pkt_mark(flow_metadata
, flow
->pkt_mark
);
851 match_set_in_port(flow_metadata
, flow
->in_port
.ofp_port
);
852 if (flow
->ct_state
!= 0) {
853 match_set_ct_state(flow_metadata
, flow
->ct_state
);
855 if (flow
->ct_zone
!= 0) {
856 match_set_ct_zone(flow_metadata
, flow
->ct_zone
);
858 if (flow
->ct_mark
!= 0) {
859 match_set_ct_mark(flow_metadata
, flow
->ct_mark
);
861 if (!ovs_u128_is_zero(&flow
->ct_label
)) {
862 match_set_ct_label(flow_metadata
, flow
->ct_label
);
866 const char *ct_state_to_string(uint32_t state
)
891 flow_to_string(const struct flow
*flow
)
893 struct ds ds
= DS_EMPTY_INITIALIZER
;
894 flow_format(&ds
, flow
);
899 flow_tun_flag_to_string(uint32_t flags
)
902 case FLOW_TNL_F_DONT_FRAGMENT
:
904 case FLOW_TNL_F_CSUM
:
916 format_flags(struct ds
*ds
, const char *(*bit_to_string
)(uint32_t),
917 uint32_t flags
, char del
)
922 ds_put_char(ds
, '0');
926 uint32_t bit
= rightmost_1bit(flags
);
929 s
= bit_to_string(bit
);
931 ds_put_format(ds
, "%s%c", s
, del
);
940 ds_put_format(ds
, "0x%"PRIx32
"%c", bad
, del
);
946 format_flags_masked(struct ds
*ds
, const char *name
,
947 const char *(*bit_to_string
)(uint32_t), uint32_t flags
,
948 uint32_t mask
, uint32_t max_mask
)
951 ds_put_format(ds
, "%s=", name
);
954 if (mask
== max_mask
) {
955 format_flags(ds
, bit_to_string
, flags
, '|');
960 ds_put_cstr(ds
, "0/0");
965 uint32_t bit
= rightmost_1bit(mask
);
966 const char *s
= bit_to_string(bit
);
968 ds_put_format(ds
, "%s%s", (flags
& bit
) ? "+" : "-",
969 s
? s
: "[Unknown]");
974 /* Scans a string 's' of flags to determine their numerical value and
975 * returns the number of characters parsed using 'bit_to_string' to
976 * lookup flag names. Scanning continues until the character 'end' is
979 * In the event of a failure, a negative error code will be returned. In
980 * addition, if 'res_string' is non-NULL then a descriptive string will
981 * be returned incorporating the identifying string 'field_name'. This
982 * error string must be freed by the caller.
984 * Upon success, the flag values will be stored in 'res_flags' and
985 * optionally 'res_mask', if it is non-NULL (if it is NULL then any masks
986 * present in the original string will be considered an error). The
987 * caller may restrict the acceptable set of values through the mask
990 parse_flags(const char *s
, const char *(*bit_to_string
)(uint32_t),
991 char end
, const char *field_name
, char **res_string
,
992 uint32_t *res_flags
, uint32_t allowed
, uint32_t *res_mask
)
997 /* Parse masked flags in numeric format? */
998 if (res_mask
&& ovs_scan(s
, "%"SCNi32
"/%"SCNi32
"%n",
999 res_flags
, res_mask
, &n
) && n
> 0) {
1000 if (*res_flags
& ~allowed
|| *res_mask
& ~allowed
) {
1008 if (res_mask
&& (*s
== '+' || *s
== '-')) {
1009 uint32_t flags
= 0, mask
= 0;
1011 /* Parse masked flags. */
1012 while (s
[0] != end
) {
1019 } else if (s
[0] == '-') {
1023 *res_string
= xasprintf("%s: %s must be preceded by '+' "
1024 "(for SET) or '-' (NOT SET)", s
,
1032 for (bit
= 1; bit
; bit
<<= 1) {
1033 const char *fname
= bit_to_string(bit
);
1039 len
= strlen(fname
);
1040 if (strncmp(s
, fname
, len
) ||
1041 (s
[len
] != '+' && s
[len
] != '-' && s
[len
] != end
)) {
1046 /* bit already set. */
1048 *res_string
= xasprintf("%s: Each %s flag can be "
1049 "specified only once", s
,
1054 if (!(bit
& allowed
)) {
1076 /* Parse unmasked flags. If a flag is present, it is set, otherwise
1078 while (s
[n
] != end
) {
1079 unsigned long long int flags
;
1083 if (ovs_scan(&s
[n
], "%lli%n", &flags
, &n0
)) {
1084 if (flags
& ~allowed
) {
1087 n
+= n0
+ (s
[n
+ n0
] == '|');
1092 for (bit
= 1; bit
; bit
<<= 1) {
1093 const char *name
= bit_to_string(bit
);
1101 if (!strncmp(s
+ n
, name
, len
) &&
1102 (s
[n
+ len
] == '|' || s
[n
+ len
] == end
)) {
1103 if (!(bit
& allowed
)) {
1107 n
+= len
+ (s
[n
+ len
] == '|');
1117 *res_flags
= result
;
1119 *res_mask
= UINT32_MAX
;
1128 *res_string
= xasprintf("%s: unknown %s flag(s)", s
, field_name
);
1134 flow_format(struct ds
*ds
, const struct flow
*flow
)
1137 struct flow_wildcards
*wc
= &match
.wc
;
1139 match_wc_init(&match
, flow
);
1141 /* As this function is most often used for formatting a packet in a
1142 * packet-in message, skip formatting the packet context fields that are
1143 * all-zeroes to make the print-out easier on the eyes. This means that a
1144 * missing context field implies a zero value for that field. This is
1145 * similar to OpenFlow encoding of these fields, as the specification
1146 * states that all-zeroes context fields should not be encoded in the
1147 * packet-in messages. */
1148 if (!flow
->in_port
.ofp_port
) {
1149 WC_UNMASK_FIELD(wc
, in_port
);
1151 if (!flow
->skb_priority
) {
1152 WC_UNMASK_FIELD(wc
, skb_priority
);
1154 if (!flow
->pkt_mark
) {
1155 WC_UNMASK_FIELD(wc
, pkt_mark
);
1157 if (!flow
->recirc_id
) {
1158 WC_UNMASK_FIELD(wc
, recirc_id
);
1160 if (!flow
->dp_hash
) {
1161 WC_UNMASK_FIELD(wc
, dp_hash
);
1163 if (!flow
->ct_state
) {
1164 WC_UNMASK_FIELD(wc
, ct_state
);
1166 if (!flow
->ct_zone
) {
1167 WC_UNMASK_FIELD(wc
, ct_zone
);
1169 if (!flow
->ct_mark
) {
1170 WC_UNMASK_FIELD(wc
, ct_mark
);
1172 if (ovs_u128_is_zero(&flow
->ct_label
)) {
1173 WC_UNMASK_FIELD(wc
, ct_label
);
1175 for (int i
= 0; i
< FLOW_N_REGS
; i
++) {
1176 if (!flow
->regs
[i
]) {
1177 WC_UNMASK_FIELD(wc
, regs
[i
]);
1180 if (!flow
->metadata
) {
1181 WC_UNMASK_FIELD(wc
, metadata
);
1184 match_format(&match
, ds
, OFP_DEFAULT_PRIORITY
);
1188 flow_print(FILE *stream
, const struct flow
*flow
)
1190 char *s
= flow_to_string(flow
);
1195 /* flow_wildcards functions. */
1197 /* Initializes 'wc' as a set of wildcards that matches every packet. */
1199 flow_wildcards_init_catchall(struct flow_wildcards
*wc
)
1201 memset(&wc
->masks
, 0, sizeof wc
->masks
);
1204 /* Converts a flow into flow wildcards. It sets the wildcard masks based on
1205 * the packet headers extracted to 'flow'. It will not set the mask for fields
1206 * that do not make sense for the packet type. OpenFlow-only metadata is
1207 * wildcarded, but other metadata is unconditionally exact-matched. */
1208 void flow_wildcards_init_for_packet(struct flow_wildcards
*wc
,
1209 const struct flow
*flow
)
1211 memset(&wc
->masks
, 0x0, sizeof wc
->masks
);
1213 /* Update this function whenever struct flow changes. */
1214 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 35);
1216 if (flow_tnl_dst_is_set(&flow
->tunnel
)) {
1217 if (flow
->tunnel
.flags
& FLOW_TNL_F_KEY
) {
1218 WC_MASK_FIELD(wc
, tunnel
.tun_id
);
1220 WC_MASK_FIELD(wc
, tunnel
.ip_src
);
1221 WC_MASK_FIELD(wc
, tunnel
.ip_dst
);
1222 WC_MASK_FIELD(wc
, tunnel
.ipv6_src
);
1223 WC_MASK_FIELD(wc
, tunnel
.ipv6_dst
);
1224 WC_MASK_FIELD(wc
, tunnel
.flags
);
1225 WC_MASK_FIELD(wc
, tunnel
.ip_tos
);
1226 WC_MASK_FIELD(wc
, tunnel
.ip_ttl
);
1227 WC_MASK_FIELD(wc
, tunnel
.tp_src
);
1228 WC_MASK_FIELD(wc
, tunnel
.tp_dst
);
1229 WC_MASK_FIELD(wc
, tunnel
.gbp_id
);
1230 WC_MASK_FIELD(wc
, tunnel
.gbp_flags
);
1232 if (!(flow
->tunnel
.flags
& FLOW_TNL_F_UDPIF
)) {
1233 if (flow
->tunnel
.metadata
.present
.map
) {
1234 wc
->masks
.tunnel
.metadata
.present
.map
=
1235 flow
->tunnel
.metadata
.present
.map
;
1236 WC_MASK_FIELD(wc
, tunnel
.metadata
.opts
.u8
);
1239 WC_MASK_FIELD(wc
, tunnel
.metadata
.present
.len
);
1240 memset(wc
->masks
.tunnel
.metadata
.opts
.gnv
, 0xff,
1241 flow
->tunnel
.metadata
.present
.len
);
1243 } else if (flow
->tunnel
.tun_id
) {
1244 WC_MASK_FIELD(wc
, tunnel
.tun_id
);
1247 /* metadata, regs, and conj_id wildcarded. */
1249 WC_MASK_FIELD(wc
, skb_priority
);
1250 WC_MASK_FIELD(wc
, pkt_mark
);
1251 WC_MASK_FIELD(wc
, ct_state
);
1252 WC_MASK_FIELD(wc
, ct_zone
);
1253 WC_MASK_FIELD(wc
, ct_mark
);
1254 WC_MASK_FIELD(wc
, ct_label
);
1255 WC_MASK_FIELD(wc
, recirc_id
);
1256 WC_MASK_FIELD(wc
, dp_hash
);
1257 WC_MASK_FIELD(wc
, in_port
);
1259 /* actset_output wildcarded. */
1261 WC_MASK_FIELD(wc
, dl_dst
);
1262 WC_MASK_FIELD(wc
, dl_src
);
1263 WC_MASK_FIELD(wc
, dl_type
);
1264 WC_MASK_FIELD(wc
, vlan_tci
);
1266 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
1267 WC_MASK_FIELD(wc
, nw_src
);
1268 WC_MASK_FIELD(wc
, nw_dst
);
1269 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1270 WC_MASK_FIELD(wc
, ipv6_src
);
1271 WC_MASK_FIELD(wc
, ipv6_dst
);
1272 WC_MASK_FIELD(wc
, ipv6_label
);
1273 } else if (flow
->dl_type
== htons(ETH_TYPE_ARP
) ||
1274 flow
->dl_type
== htons(ETH_TYPE_RARP
)) {
1275 WC_MASK_FIELD(wc
, nw_src
);
1276 WC_MASK_FIELD(wc
, nw_dst
);
1277 WC_MASK_FIELD(wc
, nw_proto
);
1278 WC_MASK_FIELD(wc
, arp_sha
);
1279 WC_MASK_FIELD(wc
, arp_tha
);
1281 } else if (eth_type_mpls(flow
->dl_type
)) {
1282 for (int i
= 0; i
< FLOW_MAX_MPLS_LABELS
; i
++) {
1283 WC_MASK_FIELD(wc
, mpls_lse
[i
]);
1284 if (flow
->mpls_lse
[i
] & htonl(MPLS_BOS_MASK
)) {
1290 return; /* Unknown ethertype. */
1294 WC_MASK_FIELD(wc
, nw_frag
);
1295 WC_MASK_FIELD(wc
, nw_tos
);
1296 WC_MASK_FIELD(wc
, nw_ttl
);
1297 WC_MASK_FIELD(wc
, nw_proto
);
1299 /* No transport layer header in later fragments. */
1300 if (!(flow
->nw_frag
& FLOW_NW_FRAG_LATER
) &&
1301 (flow
->nw_proto
== IPPROTO_ICMP
||
1302 flow
->nw_proto
== IPPROTO_ICMPV6
||
1303 flow
->nw_proto
== IPPROTO_TCP
||
1304 flow
->nw_proto
== IPPROTO_UDP
||
1305 flow
->nw_proto
== IPPROTO_SCTP
||
1306 flow
->nw_proto
== IPPROTO_IGMP
)) {
1307 WC_MASK_FIELD(wc
, tp_src
);
1308 WC_MASK_FIELD(wc
, tp_dst
);
1310 if (flow
->nw_proto
== IPPROTO_TCP
) {
1311 WC_MASK_FIELD(wc
, tcp_flags
);
1312 } else if (flow
->nw_proto
== IPPROTO_ICMPV6
) {
1313 WC_MASK_FIELD(wc
, arp_sha
);
1314 WC_MASK_FIELD(wc
, arp_tha
);
1315 WC_MASK_FIELD(wc
, nd_target
);
1316 } else if (flow
->nw_proto
== IPPROTO_IGMP
) {
1317 WC_MASK_FIELD(wc
, igmp_group_ip4
);
1322 /* Return a map of possible fields for a packet of the same type as 'flow'.
1323 * Including extra bits in the returned mask is not wrong, it is just less
1326 * This is a less precise version of flow_wildcards_init_for_packet() above. */
1328 flow_wc_map(const struct flow
*flow
, struct flowmap
*map
)
1330 /* Update this function whenever struct flow changes. */
1331 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 35);
1335 if (flow_tnl_dst_is_set(&flow
->tunnel
)) {
1336 FLOWMAP_SET__(map
, tunnel
, offsetof(struct flow_tnl
, metadata
));
1337 if (!(flow
->tunnel
.flags
& FLOW_TNL_F_UDPIF
)) {
1338 if (flow
->tunnel
.metadata
.present
.map
) {
1339 FLOWMAP_SET(map
, tunnel
.metadata
);
1342 FLOWMAP_SET(map
, tunnel
.metadata
.present
.len
);
1343 FLOWMAP_SET__(map
, tunnel
.metadata
.opts
.gnv
,
1344 flow
->tunnel
.metadata
.present
.len
);
1348 /* Metadata fields that can appear on packet input. */
1349 FLOWMAP_SET(map
, skb_priority
);
1350 FLOWMAP_SET(map
, pkt_mark
);
1351 FLOWMAP_SET(map
, recirc_id
);
1352 FLOWMAP_SET(map
, dp_hash
);
1353 FLOWMAP_SET(map
, in_port
);
1354 FLOWMAP_SET(map
, dl_dst
);
1355 FLOWMAP_SET(map
, dl_src
);
1356 FLOWMAP_SET(map
, dl_type
);
1357 FLOWMAP_SET(map
, vlan_tci
);
1358 FLOWMAP_SET(map
, ct_state
);
1359 FLOWMAP_SET(map
, ct_zone
);
1360 FLOWMAP_SET(map
, ct_mark
);
1361 FLOWMAP_SET(map
, ct_label
);
1363 /* Ethertype-dependent fields. */
1364 if (OVS_LIKELY(flow
->dl_type
== htons(ETH_TYPE_IP
))) {
1365 FLOWMAP_SET(map
, nw_src
);
1366 FLOWMAP_SET(map
, nw_dst
);
1367 FLOWMAP_SET(map
, nw_proto
);
1368 FLOWMAP_SET(map
, nw_frag
);
1369 FLOWMAP_SET(map
, nw_tos
);
1370 FLOWMAP_SET(map
, nw_ttl
);
1372 if (OVS_UNLIKELY(flow
->nw_proto
== IPPROTO_IGMP
)) {
1373 FLOWMAP_SET(map
, igmp_group_ip4
);
1375 FLOWMAP_SET(map
, tcp_flags
);
1376 FLOWMAP_SET(map
, tp_src
);
1377 FLOWMAP_SET(map
, tp_dst
);
1379 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1380 FLOWMAP_SET(map
, ipv6_src
);
1381 FLOWMAP_SET(map
, ipv6_dst
);
1382 FLOWMAP_SET(map
, ipv6_label
);
1383 FLOWMAP_SET(map
, nw_proto
);
1384 FLOWMAP_SET(map
, nw_frag
);
1385 FLOWMAP_SET(map
, nw_tos
);
1386 FLOWMAP_SET(map
, nw_ttl
);
1388 if (OVS_UNLIKELY(flow
->nw_proto
== IPPROTO_ICMPV6
)) {
1389 FLOWMAP_SET(map
, nd_target
);
1390 FLOWMAP_SET(map
, arp_sha
);
1391 FLOWMAP_SET(map
, arp_tha
);
1393 FLOWMAP_SET(map
, tcp_flags
);
1394 FLOWMAP_SET(map
, tp_src
);
1395 FLOWMAP_SET(map
, tp_dst
);
1397 } else if (eth_type_mpls(flow
->dl_type
)) {
1398 FLOWMAP_SET(map
, mpls_lse
);
1399 } else if (flow
->dl_type
== htons(ETH_TYPE_ARP
) ||
1400 flow
->dl_type
== htons(ETH_TYPE_RARP
)) {
1401 FLOWMAP_SET(map
, nw_src
);
1402 FLOWMAP_SET(map
, nw_dst
);
1403 FLOWMAP_SET(map
, nw_proto
);
1404 FLOWMAP_SET(map
, arp_sha
);
1405 FLOWMAP_SET(map
, arp_tha
);
1409 /* Clear the metadata and register wildcard masks. They are not packet
1412 flow_wildcards_clear_non_packet_fields(struct flow_wildcards
*wc
)
1414 /* Update this function whenever struct flow changes. */
1415 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 35);
1417 memset(&wc
->masks
.metadata
, 0, sizeof wc
->masks
.metadata
);
1418 memset(&wc
->masks
.regs
, 0, sizeof wc
->masks
.regs
);
1419 wc
->masks
.actset_output
= 0;
1420 wc
->masks
.conj_id
= 0;
1423 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
1426 flow_wildcards_is_catchall(const struct flow_wildcards
*wc
)
1428 const uint64_t *wc_u64
= (const uint64_t *) &wc
->masks
;
1431 for (i
= 0; i
< FLOW_U64S
; i
++) {
1439 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
1440 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
1441 * in 'src1' or 'src2' or both. */
1443 flow_wildcards_and(struct flow_wildcards
*dst
,
1444 const struct flow_wildcards
*src1
,
1445 const struct flow_wildcards
*src2
)
1447 uint64_t *dst_u64
= (uint64_t *) &dst
->masks
;
1448 const uint64_t *src1_u64
= (const uint64_t *) &src1
->masks
;
1449 const uint64_t *src2_u64
= (const uint64_t *) &src2
->masks
;
1452 for (i
= 0; i
< FLOW_U64S
; i
++) {
1453 dst_u64
[i
] = src1_u64
[i
] & src2_u64
[i
];
1457 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
1458 * is, a bit or a field is wildcarded in 'dst' if it is neither
1459 * wildcarded in 'src1' nor 'src2'. */
1461 flow_wildcards_or(struct flow_wildcards
*dst
,
1462 const struct flow_wildcards
*src1
,
1463 const struct flow_wildcards
*src2
)
1465 uint64_t *dst_u64
= (uint64_t *) &dst
->masks
;
1466 const uint64_t *src1_u64
= (const uint64_t *) &src1
->masks
;
1467 const uint64_t *src2_u64
= (const uint64_t *) &src2
->masks
;
1470 for (i
= 0; i
< FLOW_U64S
; i
++) {
1471 dst_u64
[i
] = src1_u64
[i
] | src2_u64
[i
];
1475 /* Returns a hash of the wildcards in 'wc'. */
1477 flow_wildcards_hash(const struct flow_wildcards
*wc
, uint32_t basis
)
1479 return flow_hash(&wc
->masks
, basis
);
1482 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
1485 flow_wildcards_equal(const struct flow_wildcards
*a
,
1486 const struct flow_wildcards
*b
)
1488 return flow_equal(&a
->masks
, &b
->masks
);
1491 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
1492 * 'b', false otherwise. */
1494 flow_wildcards_has_extra(const struct flow_wildcards
*a
,
1495 const struct flow_wildcards
*b
)
1497 const uint64_t *a_u64
= (const uint64_t *) &a
->masks
;
1498 const uint64_t *b_u64
= (const uint64_t *) &b
->masks
;
1501 for (i
= 0; i
< FLOW_U64S
; i
++) {
1502 if ((a_u64
[i
] & b_u64
[i
]) != b_u64
[i
]) {
1509 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
1510 * in 'wc' do not need to be equal in 'a' and 'b'. */
1512 flow_equal_except(const struct flow
*a
, const struct flow
*b
,
1513 const struct flow_wildcards
*wc
)
1515 const uint64_t *a_u64
= (const uint64_t *) a
;
1516 const uint64_t *b_u64
= (const uint64_t *) b
;
1517 const uint64_t *wc_u64
= (const uint64_t *) &wc
->masks
;
1520 for (i
= 0; i
< FLOW_U64S
; i
++) {
1521 if ((a_u64
[i
] ^ b_u64
[i
]) & wc_u64
[i
]) {
1528 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1529 * (A 0-bit indicates a wildcard bit.) */
1531 flow_wildcards_set_reg_mask(struct flow_wildcards
*wc
, int idx
, uint32_t mask
)
1533 wc
->masks
.regs
[idx
] = mask
;
1536 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1537 * (A 0-bit indicates a wildcard bit.) */
1539 flow_wildcards_set_xreg_mask(struct flow_wildcards
*wc
, int idx
, uint64_t mask
)
1541 flow_set_xreg(&wc
->masks
, idx
, mask
);
1544 /* Calculates the 5-tuple hash from the given miniflow.
1545 * This returns the same value as flow_hash_5tuple for the corresponding
1548 miniflow_hash_5tuple(const struct miniflow
*flow
, uint32_t basis
)
1550 uint32_t hash
= basis
;
1553 ovs_be16 dl_type
= MINIFLOW_GET_BE16(flow
, dl_type
);
1555 hash
= hash_add(hash
, MINIFLOW_GET_U8(flow
, nw_proto
));
1557 /* Separate loops for better optimization. */
1558 if (dl_type
== htons(ETH_TYPE_IPV6
)) {
1559 struct flowmap map
= FLOWMAP_EMPTY_INITIALIZER
;
1562 FLOWMAP_SET(&map
, ipv6_src
);
1563 FLOWMAP_SET(&map
, ipv6_dst
);
1565 MINIFLOW_FOR_EACH_IN_FLOWMAP(value
, flow
, map
) {
1566 hash
= hash_add64(hash
, value
);
1569 hash
= hash_add(hash
, MINIFLOW_GET_U32(flow
, nw_src
));
1570 hash
= hash_add(hash
, MINIFLOW_GET_U32(flow
, nw_dst
));
1572 /* Add both ports at once. */
1573 hash
= hash_add(hash
, MINIFLOW_GET_U32(flow
, tp_src
));
1574 hash
= hash_finish(hash
, 42); /* Arbitrary number. */
1579 ASSERT_SEQUENTIAL_SAME_WORD(tp_src
, tp_dst
);
1580 ASSERT_SEQUENTIAL(ipv6_src
, ipv6_dst
);
1582 /* Calculates the 5-tuple hash from the given flow. */
1584 flow_hash_5tuple(const struct flow
*flow
, uint32_t basis
)
1586 uint32_t hash
= basis
;
1589 hash
= hash_add(hash
, flow
->nw_proto
);
1591 if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1592 const uint64_t *flow_u64
= (const uint64_t *)flow
;
1593 int ofs
= offsetof(struct flow
, ipv6_src
) / 8;
1594 int end
= ofs
+ 2 * sizeof flow
->ipv6_src
/ 8;
1596 for (;ofs
< end
; ofs
++) {
1597 hash
= hash_add64(hash
, flow_u64
[ofs
]);
1600 hash
= hash_add(hash
, (OVS_FORCE
uint32_t) flow
->nw_src
);
1601 hash
= hash_add(hash
, (OVS_FORCE
uint32_t) flow
->nw_dst
);
1603 /* Add both ports at once. */
1604 hash
= hash_add(hash
,
1605 ((const uint32_t *)flow
)[offsetof(struct flow
, tp_src
)
1606 / sizeof(uint32_t)]);
1607 hash
= hash_finish(hash
, 42); /* Arbitrary number. */
1612 /* Hashes 'flow' based on its L2 through L4 protocol information. */
1614 flow_hash_symmetric_l4(const struct flow
*flow
, uint32_t basis
)
1619 struct in6_addr ipv6_addr
;
1624 struct eth_addr eth_addr
;
1630 memset(&fields
, 0, sizeof fields
);
1631 for (i
= 0; i
< ARRAY_SIZE(fields
.eth_addr
.be16
); i
++) {
1632 fields
.eth_addr
.be16
[i
] = flow
->dl_src
.be16
[i
] ^ flow
->dl_dst
.be16
[i
];
1634 fields
.vlan_tci
= flow
->vlan_tci
& htons(VLAN_VID_MASK
);
1635 fields
.eth_type
= flow
->dl_type
;
1637 /* UDP source and destination port are not taken into account because they
1638 * will not necessarily be symmetric in a bidirectional flow. */
1639 if (fields
.eth_type
== htons(ETH_TYPE_IP
)) {
1640 fields
.ipv4_addr
= flow
->nw_src
^ flow
->nw_dst
;
1641 fields
.ip_proto
= flow
->nw_proto
;
1642 if (fields
.ip_proto
== IPPROTO_TCP
|| fields
.ip_proto
== IPPROTO_SCTP
) {
1643 fields
.tp_port
= flow
->tp_src
^ flow
->tp_dst
;
1645 } else if (fields
.eth_type
== htons(ETH_TYPE_IPV6
)) {
1646 const uint8_t *a
= &flow
->ipv6_src
.s6_addr
[0];
1647 const uint8_t *b
= &flow
->ipv6_dst
.s6_addr
[0];
1648 uint8_t *ipv6_addr
= &fields
.ipv6_addr
.s6_addr
[0];
1650 for (i
=0; i
<16; i
++) {
1651 ipv6_addr
[i
] = a
[i
] ^ b
[i
];
1653 fields
.ip_proto
= flow
->nw_proto
;
1654 if (fields
.ip_proto
== IPPROTO_TCP
|| fields
.ip_proto
== IPPROTO_SCTP
) {
1655 fields
.tp_port
= flow
->tp_src
^ flow
->tp_dst
;
1658 return jhash_bytes(&fields
, sizeof fields
, basis
);
1661 /* Hashes 'flow' based on its L3 through L4 protocol information */
1663 flow_hash_symmetric_l3l4(const struct flow
*flow
, uint32_t basis
,
1666 uint32_t hash
= basis
;
1668 /* UDP source and destination port are also taken into account. */
1669 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
1670 hash
= hash_add(hash
,
1671 (OVS_FORCE
uint32_t) (flow
->nw_src
^ flow
->nw_dst
));
1672 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1673 /* IPv6 addresses are 64-bit aligned inside struct flow. */
1674 const uint64_t *a
= ALIGNED_CAST(uint64_t *, flow
->ipv6_src
.s6_addr
);
1675 const uint64_t *b
= ALIGNED_CAST(uint64_t *, flow
->ipv6_dst
.s6_addr
);
1677 for (int i
= 0; i
< 4; i
++) {
1678 hash
= hash_add64(hash
, a
[i
] ^ b
[i
]);
1681 /* Cannot hash non-IP flows */
1685 hash
= hash_add(hash
, flow
->nw_proto
);
1686 if (flow
->nw_proto
== IPPROTO_TCP
|| flow
->nw_proto
== IPPROTO_SCTP
||
1687 (inc_udp_ports
&& flow
->nw_proto
== IPPROTO_UDP
)) {
1688 hash
= hash_add(hash
,
1689 (OVS_FORCE
uint16_t) (flow
->tp_src
^ flow
->tp_dst
));
1692 return hash_finish(hash
, basis
);
1695 /* Initialize a flow with random fields that matter for nx_hash_fields. */
1697 flow_random_hash_fields(struct flow
*flow
)
1699 uint16_t rnd
= random_uint16();
1701 /* Initialize to all zeros. */
1702 memset(flow
, 0, sizeof *flow
);
1704 eth_addr_random(&flow
->dl_src
);
1705 eth_addr_random(&flow
->dl_dst
);
1707 flow
->vlan_tci
= (OVS_FORCE ovs_be16
) (random_uint16() & VLAN_VID_MASK
);
1709 /* Make most of the random flows IPv4, some IPv6, and rest random. */
1710 flow
->dl_type
= rnd
< 0x8000 ? htons(ETH_TYPE_IP
) :
1711 rnd
< 0xc000 ? htons(ETH_TYPE_IPV6
) : (OVS_FORCE ovs_be16
)rnd
;
1713 if (dl_type_is_ip_any(flow
->dl_type
)) {
1714 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
1715 flow
->nw_src
= (OVS_FORCE ovs_be32
)random_uint32();
1716 flow
->nw_dst
= (OVS_FORCE ovs_be32
)random_uint32();
1718 random_bytes(&flow
->ipv6_src
, sizeof flow
->ipv6_src
);
1719 random_bytes(&flow
->ipv6_dst
, sizeof flow
->ipv6_dst
);
1721 /* Make most of IP flows TCP, some UDP or SCTP, and rest random. */
1722 rnd
= random_uint16();
1723 flow
->nw_proto
= rnd
< 0x8000 ? IPPROTO_TCP
:
1724 rnd
< 0xc000 ? IPPROTO_UDP
:
1725 rnd
< 0xd000 ? IPPROTO_SCTP
: (uint8_t)rnd
;
1726 if (flow
->nw_proto
== IPPROTO_TCP
||
1727 flow
->nw_proto
== IPPROTO_UDP
||
1728 flow
->nw_proto
== IPPROTO_SCTP
) {
1729 flow
->tp_src
= (OVS_FORCE ovs_be16
)random_uint16();
1730 flow
->tp_dst
= (OVS_FORCE ovs_be16
)random_uint16();
1735 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
1737 flow_mask_hash_fields(const struct flow
*flow
, struct flow_wildcards
*wc
,
1738 enum nx_hash_fields fields
)
1741 case NX_HASH_FIELDS_ETH_SRC
:
1742 memset(&wc
->masks
.dl_src
, 0xff, sizeof wc
->masks
.dl_src
);
1745 case NX_HASH_FIELDS_SYMMETRIC_L4
:
1746 memset(&wc
->masks
.dl_src
, 0xff, sizeof wc
->masks
.dl_src
);
1747 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
1748 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
1749 memset(&wc
->masks
.nw_src
, 0xff, sizeof wc
->masks
.nw_src
);
1750 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
1751 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1752 memset(&wc
->masks
.ipv6_src
, 0xff, sizeof wc
->masks
.ipv6_src
);
1753 memset(&wc
->masks
.ipv6_dst
, 0xff, sizeof wc
->masks
.ipv6_dst
);
1755 if (is_ip_any(flow
)) {
1756 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
1757 flow_unwildcard_tp_ports(flow
, wc
);
1759 wc
->masks
.vlan_tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
1762 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP
:
1763 if (is_ip_any(flow
) && flow
->nw_proto
== IPPROTO_UDP
) {
1764 memset(&wc
->masks
.tp_src
, 0xff, sizeof wc
->masks
.tp_src
);
1765 memset(&wc
->masks
.tp_dst
, 0xff, sizeof wc
->masks
.tp_dst
);
1768 case NX_HASH_FIELDS_SYMMETRIC_L3L4
:
1769 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
1770 memset(&wc
->masks
.nw_src
, 0xff, sizeof wc
->masks
.nw_src
);
1771 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
1772 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1773 memset(&wc
->masks
.ipv6_src
, 0xff, sizeof wc
->masks
.ipv6_src
);
1774 memset(&wc
->masks
.ipv6_dst
, 0xff, sizeof wc
->masks
.ipv6_dst
);
1776 break; /* non-IP flow */
1779 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
1780 if (flow
->nw_proto
== IPPROTO_TCP
|| flow
->nw_proto
== IPPROTO_SCTP
) {
1781 memset(&wc
->masks
.tp_src
, 0xff, sizeof wc
->masks
.tp_src
);
1782 memset(&wc
->masks
.tp_dst
, 0xff, sizeof wc
->masks
.tp_dst
);
1791 /* Hashes the portions of 'flow' designated by 'fields'. */
1793 flow_hash_fields(const struct flow
*flow
, enum nx_hash_fields fields
,
1798 case NX_HASH_FIELDS_ETH_SRC
:
1799 return jhash_bytes(&flow
->dl_src
, sizeof flow
->dl_src
, basis
);
1801 case NX_HASH_FIELDS_SYMMETRIC_L4
:
1802 return flow_hash_symmetric_l4(flow
, basis
);
1804 case NX_HASH_FIELDS_SYMMETRIC_L3L4
:
1805 return flow_hash_symmetric_l3l4(flow
, basis
, false);
1807 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP
:
1808 return flow_hash_symmetric_l3l4(flow
, basis
, true);
1815 /* Returns a string representation of 'fields'. */
1817 flow_hash_fields_to_str(enum nx_hash_fields fields
)
1820 case NX_HASH_FIELDS_ETH_SRC
: return "eth_src";
1821 case NX_HASH_FIELDS_SYMMETRIC_L4
: return "symmetric_l4";
1822 case NX_HASH_FIELDS_SYMMETRIC_L3L4
: return "symmetric_l3l4";
1823 case NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP
: return "symmetric_l3l4+udp";
1824 default: return "<unknown>";
1828 /* Returns true if the value of 'fields' is supported. Otherwise false. */
1830 flow_hash_fields_valid(enum nx_hash_fields fields
)
1832 return fields
== NX_HASH_FIELDS_ETH_SRC
1833 || fields
== NX_HASH_FIELDS_SYMMETRIC_L4
1834 || fields
== NX_HASH_FIELDS_SYMMETRIC_L3L4
1835 || fields
== NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP
;
1838 /* Returns a hash value for the bits of 'flow' that are active based on
1839 * 'wc', given 'basis'. */
1841 flow_hash_in_wildcards(const struct flow
*flow
,
1842 const struct flow_wildcards
*wc
, uint32_t basis
)
1844 const uint64_t *wc_u64
= (const uint64_t *) &wc
->masks
;
1845 const uint64_t *flow_u64
= (const uint64_t *) flow
;
1850 for (i
= 0; i
< FLOW_U64S
; i
++) {
1851 hash
= hash_add64(hash
, flow_u64
[i
] & wc_u64
[i
]);
1853 return hash_finish(hash
, 8 * FLOW_U64S
);
1856 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1857 * OpenFlow 1.0 "dl_vlan" value:
1859 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
1860 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
1861 * 'flow' previously matched packets without a VLAN header).
1863 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
1864 * without a VLAN tag.
1866 * - Other values of 'vid' should not be used. */
1868 flow_set_dl_vlan(struct flow
*flow
, ovs_be16 vid
)
1870 if (vid
== htons(OFP10_VLAN_NONE
)) {
1871 flow
->vlan_tci
= htons(0);
1873 vid
&= htons(VLAN_VID_MASK
);
1874 flow
->vlan_tci
&= ~htons(VLAN_VID_MASK
);
1875 flow
->vlan_tci
|= htons(VLAN_CFI
) | vid
;
1879 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1880 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
1883 flow_set_vlan_vid(struct flow
*flow
, ovs_be16 vid
)
1885 ovs_be16 mask
= htons(VLAN_VID_MASK
| VLAN_CFI
);
1886 flow
->vlan_tci
&= ~mask
;
1887 flow
->vlan_tci
|= vid
& mask
;
1890 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
1893 * This function has no effect on the VLAN ID that 'flow' matches.
1895 * After calling this function, 'flow' will not match packets without a VLAN
1898 flow_set_vlan_pcp(struct flow
*flow
, uint8_t pcp
)
1901 flow
->vlan_tci
&= ~htons(VLAN_PCP_MASK
);
1902 flow
->vlan_tci
|= htons((pcp
<< VLAN_PCP_SHIFT
) | VLAN_CFI
);
1905 /* Returns the number of MPLS LSEs present in 'flow'
1907 * Returns 0 if the 'dl_type' of 'flow' is not an MPLS ethernet type.
1908 * Otherwise traverses 'flow''s MPLS label stack stopping at the
1909 * first entry that has the BoS bit set. If no such entry exists then
1910 * the maximum number of LSEs that can be stored in 'flow' is returned.
1913 flow_count_mpls_labels(const struct flow
*flow
, struct flow_wildcards
*wc
)
1915 /* dl_type is always masked. */
1916 if (eth_type_mpls(flow
->dl_type
)) {
1921 for (i
= 0; i
< FLOW_MAX_MPLS_LABELS
; i
++) {
1923 wc
->masks
.mpls_lse
[i
] |= htonl(MPLS_BOS_MASK
);
1925 if (flow
->mpls_lse
[i
] & htonl(MPLS_BOS_MASK
)) {
1928 if (flow
->mpls_lse
[i
]) {
1938 /* Returns the number consecutive of MPLS LSEs, starting at the
1939 * innermost LSE, that are common in 'a' and 'b'.
1941 * 'an' must be flow_count_mpls_labels(a).
1942 * 'bn' must be flow_count_mpls_labels(b).
1945 flow_count_common_mpls_labels(const struct flow
*a
, int an
,
1946 const struct flow
*b
, int bn
,
1947 struct flow_wildcards
*wc
)
1949 int min_n
= MIN(an
, bn
);
1954 int a_last
= an
- 1;
1955 int b_last
= bn
- 1;
1958 for (i
= 0; i
< min_n
; i
++) {
1960 wc
->masks
.mpls_lse
[a_last
- i
] = OVS_BE32_MAX
;
1961 wc
->masks
.mpls_lse
[b_last
- i
] = OVS_BE32_MAX
;
1963 if (a
->mpls_lse
[a_last
- i
] != b
->mpls_lse
[b_last
- i
]) {
1974 /* Adds a new outermost MPLS label to 'flow' and changes 'flow''s Ethernet type
1975 * to 'mpls_eth_type', which must be an MPLS Ethertype.
1977 * If the new label is the first MPLS label in 'flow', it is generated as;
1979 * - label: 2, if 'flow' is IPv6, otherwise 0.
1981 * - TTL: IPv4 or IPv6 TTL, if present and nonzero, otherwise 64.
1983 * - TC: IPv4 or IPv6 TOS, if present, otherwise 0.
1987 * If the new label is the second or later label MPLS label in 'flow', it is
1990 * - label: Copied from outer label.
1992 * - TTL: Copied from outer label.
1994 * - TC: Copied from outer label.
1998 * 'n' must be flow_count_mpls_labels(flow). 'n' must be less than
1999 * FLOW_MAX_MPLS_LABELS (because otherwise flow->mpls_lse[] would overflow).
2002 flow_push_mpls(struct flow
*flow
, int n
, ovs_be16 mpls_eth_type
,
2003 struct flow_wildcards
*wc
)
2005 ovs_assert(eth_type_mpls(mpls_eth_type
));
2006 ovs_assert(n
< FLOW_MAX_MPLS_LABELS
);
2012 memset(&wc
->masks
.mpls_lse
, 0xff, sizeof *wc
->masks
.mpls_lse
* n
);
2014 for (i
= n
; i
>= 1; i
--) {
2015 flow
->mpls_lse
[i
] = flow
->mpls_lse
[i
- 1];
2017 flow
->mpls_lse
[0] = (flow
->mpls_lse
[1] & htonl(~MPLS_BOS_MASK
));
2019 int label
= 0; /* IPv4 Explicit Null. */
2023 if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
2027 if (is_ip_any(flow
)) {
2028 tc
= (flow
->nw_tos
& IP_DSCP_MASK
) >> 2;
2030 wc
->masks
.nw_tos
|= IP_DSCP_MASK
;
2031 wc
->masks
.nw_ttl
= 0xff;
2039 flow
->mpls_lse
[0] = set_mpls_lse_values(ttl
, tc
, 1, htonl(label
));
2041 /* Clear all L3 and L4 fields and dp_hash. */
2042 BUILD_ASSERT(FLOW_WC_SEQ
== 35);
2043 memset((char *) flow
+ FLOW_SEGMENT_2_ENDS_AT
, 0,
2044 sizeof(struct flow
) - FLOW_SEGMENT_2_ENDS_AT
);
2047 flow
->dl_type
= mpls_eth_type
;
2050 /* Tries to remove the outermost MPLS label from 'flow'. Returns true if
2051 * successful, false otherwise. On success, sets 'flow''s Ethernet type to
2054 * 'n' must be flow_count_mpls_labels(flow). */
2056 flow_pop_mpls(struct flow
*flow
, int n
, ovs_be16 eth_type
,
2057 struct flow_wildcards
*wc
)
2062 /* Nothing to pop. */
2064 } else if (n
== FLOW_MAX_MPLS_LABELS
) {
2066 wc
->masks
.mpls_lse
[n
- 1] |= htonl(MPLS_BOS_MASK
);
2068 if (!(flow
->mpls_lse
[n
- 1] & htonl(MPLS_BOS_MASK
))) {
2069 /* Can't pop because don't know what to fill in mpls_lse[n - 1]. */
2075 memset(&wc
->masks
.mpls_lse
[1], 0xff,
2076 sizeof *wc
->masks
.mpls_lse
* (n
- 1));
2078 for (i
= 1; i
< n
; i
++) {
2079 flow
->mpls_lse
[i
- 1] = flow
->mpls_lse
[i
];
2081 flow
->mpls_lse
[n
- 1] = 0;
2082 flow
->dl_type
= eth_type
;
2086 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
2087 * as an OpenFlow 1.1 "mpls_label" value. */
2089 flow_set_mpls_label(struct flow
*flow
, int idx
, ovs_be32 label
)
2091 set_mpls_lse_label(&flow
->mpls_lse
[idx
], label
);
2094 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
2097 flow_set_mpls_ttl(struct flow
*flow
, int idx
, uint8_t ttl
)
2099 set_mpls_lse_ttl(&flow
->mpls_lse
[idx
], ttl
);
2102 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
2105 flow_set_mpls_tc(struct flow
*flow
, int idx
, uint8_t tc
)
2107 set_mpls_lse_tc(&flow
->mpls_lse
[idx
], tc
);
2110 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
2112 flow_set_mpls_bos(struct flow
*flow
, int idx
, uint8_t bos
)
2114 set_mpls_lse_bos(&flow
->mpls_lse
[idx
], bos
);
2117 /* Sets the entire MPLS LSE. */
2119 flow_set_mpls_lse(struct flow
*flow
, int idx
, ovs_be32 lse
)
2121 flow
->mpls_lse
[idx
] = lse
;
2125 flow_compose_l4(struct dp_packet
*p
, const struct flow
*flow
)
2129 if (!(flow
->nw_frag
& FLOW_NW_FRAG_ANY
)
2130 || !(flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
2131 if (flow
->nw_proto
== IPPROTO_TCP
) {
2132 struct tcp_header
*tcp
;
2134 l4_len
= sizeof *tcp
;
2135 tcp
= dp_packet_put_zeros(p
, l4_len
);
2136 tcp
->tcp_src
= flow
->tp_src
;
2137 tcp
->tcp_dst
= flow
->tp_dst
;
2138 tcp
->tcp_ctl
= TCP_CTL(ntohs(flow
->tcp_flags
), 5);
2139 } else if (flow
->nw_proto
== IPPROTO_UDP
) {
2140 struct udp_header
*udp
;
2142 l4_len
= sizeof *udp
;
2143 udp
= dp_packet_put_zeros(p
, l4_len
);
2144 udp
->udp_src
= flow
->tp_src
;
2145 udp
->udp_dst
= flow
->tp_dst
;
2146 } else if (flow
->nw_proto
== IPPROTO_SCTP
) {
2147 struct sctp_header
*sctp
;
2149 l4_len
= sizeof *sctp
;
2150 sctp
= dp_packet_put_zeros(p
, l4_len
);
2151 sctp
->sctp_src
= flow
->tp_src
;
2152 sctp
->sctp_dst
= flow
->tp_dst
;
2153 } else if (flow
->nw_proto
== IPPROTO_ICMP
) {
2154 struct icmp_header
*icmp
;
2156 l4_len
= sizeof *icmp
;
2157 icmp
= dp_packet_put_zeros(p
, l4_len
);
2158 icmp
->icmp_type
= ntohs(flow
->tp_src
);
2159 icmp
->icmp_code
= ntohs(flow
->tp_dst
);
2160 icmp
->icmp_csum
= csum(icmp
, ICMP_HEADER_LEN
);
2161 } else if (flow
->nw_proto
== IPPROTO_IGMP
) {
2162 struct igmp_header
*igmp
;
2164 l4_len
= sizeof *igmp
;
2165 igmp
= dp_packet_put_zeros(p
, l4_len
);
2166 igmp
->igmp_type
= ntohs(flow
->tp_src
);
2167 igmp
->igmp_code
= ntohs(flow
->tp_dst
);
2168 put_16aligned_be32(&igmp
->group
, flow
->igmp_group_ip4
);
2169 igmp
->igmp_csum
= csum(igmp
, IGMP_HEADER_LEN
);
2170 } else if (flow
->nw_proto
== IPPROTO_ICMPV6
) {
2171 struct icmp6_hdr
*icmp
;
2173 l4_len
= sizeof *icmp
;
2174 icmp
= dp_packet_put_zeros(p
, l4_len
);
2175 icmp
->icmp6_type
= ntohs(flow
->tp_src
);
2176 icmp
->icmp6_code
= ntohs(flow
->tp_dst
);
2178 if (icmp
->icmp6_code
== 0 &&
2179 (icmp
->icmp6_type
== ND_NEIGHBOR_SOLICIT
||
2180 icmp
->icmp6_type
== ND_NEIGHBOR_ADVERT
)) {
2181 struct in6_addr
*nd_target
;
2182 struct ovs_nd_opt
*nd_opt
;
2184 l4_len
+= sizeof *nd_target
;
2185 nd_target
= dp_packet_put_zeros(p
, sizeof *nd_target
);
2186 *nd_target
= flow
->nd_target
;
2188 if (!eth_addr_is_zero(flow
->arp_sha
)) {
2190 nd_opt
= dp_packet_put_zeros(p
, 8);
2191 nd_opt
->nd_opt_len
= 1;
2192 nd_opt
->nd_opt_type
= ND_OPT_SOURCE_LINKADDR
;
2193 nd_opt
->nd_opt_mac
= flow
->arp_sha
;
2195 if (!eth_addr_is_zero(flow
->arp_tha
)) {
2197 nd_opt
= dp_packet_put_zeros(p
, 8);
2198 nd_opt
->nd_opt_len
= 1;
2199 nd_opt
->nd_opt_type
= ND_OPT_TARGET_LINKADDR
;
2200 nd_opt
->nd_opt_mac
= flow
->arp_tha
;
2203 icmp
->icmp6_cksum
= (OVS_FORCE
uint16_t)
2204 csum(icmp
, (char *)dp_packet_tail(p
) - (char *)icmp
);
2210 /* Puts into 'b' a packet that flow_extract() would parse as having the given
2213 * (This is useful only for testing, obviously, and the packet isn't really
2214 * valid. It hasn't got some checksums filled in, for one, and lots of fields
2215 * are just zeroed.) */
2217 flow_compose(struct dp_packet
*p
, const struct flow
*flow
)
2221 /* eth_compose() sets l3 pointer and makes sure it is 32-bit aligned. */
2222 eth_compose(p
, flow
->dl_dst
, flow
->dl_src
, ntohs(flow
->dl_type
), 0);
2223 if (flow
->dl_type
== htons(FLOW_DL_TYPE_NONE
)) {
2224 struct eth_header
*eth
= dp_packet_l2(p
);
2225 eth
->eth_type
= htons(dp_packet_size(p
));
2229 if (flow
->vlan_tci
& htons(VLAN_CFI
)) {
2230 eth_push_vlan(p
, htons(ETH_TYPE_VLAN
), flow
->vlan_tci
);
2233 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
2234 struct ip_header
*ip
;
2236 ip
= dp_packet_put_zeros(p
, sizeof *ip
);
2237 ip
->ip_ihl_ver
= IP_IHL_VER(5, 4);
2238 ip
->ip_tos
= flow
->nw_tos
;
2239 ip
->ip_ttl
= flow
->nw_ttl
;
2240 ip
->ip_proto
= flow
->nw_proto
;
2241 put_16aligned_be32(&ip
->ip_src
, flow
->nw_src
);
2242 put_16aligned_be32(&ip
->ip_dst
, flow
->nw_dst
);
2244 if (flow
->nw_frag
& FLOW_NW_FRAG_ANY
) {
2245 ip
->ip_frag_off
|= htons(IP_MORE_FRAGMENTS
);
2246 if (flow
->nw_frag
& FLOW_NW_FRAG_LATER
) {
2247 ip
->ip_frag_off
|= htons(100);
2251 dp_packet_set_l4(p
, dp_packet_tail(p
));
2253 l4_len
= flow_compose_l4(p
, flow
);
2255 ip
= dp_packet_l3(p
);
2256 ip
->ip_tot_len
= htons(p
->l4_ofs
- p
->l3_ofs
+ l4_len
);
2257 ip
->ip_csum
= csum(ip
, sizeof *ip
);
2258 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
2259 struct ovs_16aligned_ip6_hdr
*nh
;
2261 nh
= dp_packet_put_zeros(p
, sizeof *nh
);
2262 put_16aligned_be32(&nh
->ip6_flow
, htonl(6 << 28) |
2263 htonl(flow
->nw_tos
<< 20) | flow
->ipv6_label
);
2264 nh
->ip6_hlim
= flow
->nw_ttl
;
2265 nh
->ip6_nxt
= flow
->nw_proto
;
2267 memcpy(&nh
->ip6_src
, &flow
->ipv6_src
, sizeof(nh
->ip6_src
));
2268 memcpy(&nh
->ip6_dst
, &flow
->ipv6_dst
, sizeof(nh
->ip6_dst
));
2270 dp_packet_set_l4(p
, dp_packet_tail(p
));
2272 l4_len
= flow_compose_l4(p
, flow
);
2274 nh
= dp_packet_l3(p
);
2275 nh
->ip6_plen
= htons(l4_len
);
2276 } else if (flow
->dl_type
== htons(ETH_TYPE_ARP
) ||
2277 flow
->dl_type
== htons(ETH_TYPE_RARP
)) {
2278 struct arp_eth_header
*arp
;
2280 arp
= dp_packet_put_zeros(p
, sizeof *arp
);
2281 dp_packet_set_l3(p
, arp
);
2282 arp
->ar_hrd
= htons(1);
2283 arp
->ar_pro
= htons(ETH_TYPE_IP
);
2284 arp
->ar_hln
= ETH_ADDR_LEN
;
2286 arp
->ar_op
= htons(flow
->nw_proto
);
2288 if (flow
->nw_proto
== ARP_OP_REQUEST
||
2289 flow
->nw_proto
== ARP_OP_REPLY
) {
2290 put_16aligned_be32(&arp
->ar_spa
, flow
->nw_src
);
2291 put_16aligned_be32(&arp
->ar_tpa
, flow
->nw_dst
);
2292 arp
->ar_sha
= flow
->arp_sha
;
2293 arp
->ar_tha
= flow
->arp_tha
;
2297 if (eth_type_mpls(flow
->dl_type
)) {
2300 p
->l2_5_ofs
= p
->l3_ofs
;
2301 for (n
= 1; n
< FLOW_MAX_MPLS_LABELS
; n
++) {
2302 if (flow
->mpls_lse
[n
- 1] & htonl(MPLS_BOS_MASK
)) {
2307 push_mpls(p
, flow
->dl_type
, flow
->mpls_lse
[--n
]);
2312 /* Compressed flow. */
2314 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
2315 * the caller. The caller must have already computed 'dst->map' properly to
2316 * indicate the significant uint64_t elements of 'src'.
2318 * Normally the significant elements are the ones that are non-zero. However,
2319 * when a miniflow is initialized from a (mini)mask, the values can be zeroes,
2320 * so that the flow and mask always have the same maps. */
2322 miniflow_init(struct miniflow
*dst
, const struct flow
*src
)
2324 uint64_t *dst_u64
= miniflow_values(dst
);
2327 FLOWMAP_FOR_EACH_INDEX(idx
, dst
->map
) {
2328 *dst_u64
++ = flow_u64_value(src
, idx
);
2332 /* Initialize the maps of 'flow' from 'src'. */
2334 miniflow_map_init(struct miniflow
*flow
, const struct flow
*src
)
2336 /* Initialize map, counting the number of nonzero elements. */
2337 flowmap_init(&flow
->map
);
2338 for (size_t i
= 0; i
< FLOW_U64S
; i
++) {
2339 if (flow_u64_value(src
, i
)) {
2340 flowmap_set(&flow
->map
, i
, 1);
2345 /* Allocates 'n' count of miniflows, consecutive in memory, initializing the
2346 * map of each from 'src'.
2347 * Returns the size of the miniflow data. */
2349 miniflow_alloc(struct miniflow
*dsts
[], size_t n
, const struct miniflow
*src
)
2351 size_t n_values
= miniflow_n_values(src
);
2352 size_t data_size
= MINIFLOW_VALUES_SIZE(n_values
);
2353 struct miniflow
*dst
= xmalloc(n
* (sizeof *src
+ data_size
));
2356 COVERAGE_INC(miniflow_malloc
);
2358 for (i
= 0; i
< n
; i
++) {
2359 *dst
= *src
; /* Copy maps. */
2361 dst
+= 1; /* Just past the maps. */
2362 dst
= (struct miniflow
*)((uint64_t *)dst
+ n_values
); /* Skip data. */
2367 /* Returns a miniflow copy of 'src'. The caller must eventually free() the
2368 * returned miniflow. */
2370 miniflow_create(const struct flow
*src
)
2372 struct miniflow tmp
;
2373 struct miniflow
*dst
;
2375 miniflow_map_init(&tmp
, src
);
2377 miniflow_alloc(&dst
, 1, &tmp
);
2378 miniflow_init(dst
, src
);
2382 /* Initializes 'dst' as a copy of 'src'. The caller must have allocated
2383 * 'dst' to have inline space for 'n_values' data in 'src'. */
2385 miniflow_clone(struct miniflow
*dst
, const struct miniflow
*src
,
2388 *dst
= *src
; /* Copy maps. */
2389 memcpy(miniflow_values(dst
), miniflow_get_values(src
),
2390 MINIFLOW_VALUES_SIZE(n_values
));
2393 /* Initializes 'dst' as a copy of 'src'. */
2395 miniflow_expand(const struct miniflow
*src
, struct flow
*dst
)
2397 memset(dst
, 0, sizeof *dst
);
2398 flow_union_with_miniflow(dst
, src
);
2401 /* Returns true if 'a' and 'b' are equal miniflows, false otherwise. */
2403 miniflow_equal(const struct miniflow
*a
, const struct miniflow
*b
)
2405 const uint64_t *ap
= miniflow_get_values(a
);
2406 const uint64_t *bp
= miniflow_get_values(b
);
2408 /* This is mostly called after a matching hash, so it is highly likely that
2409 * the maps are equal as well. */
2410 if (OVS_LIKELY(flowmap_equal(a
->map
, b
->map
))) {
2411 return !memcmp(ap
, bp
, miniflow_n_values(a
) * sizeof *ap
);
2415 FLOWMAP_FOR_EACH_INDEX (idx
, flowmap_or(a
->map
, b
->map
)) {
2416 if ((flowmap_is_set(&a
->map
, idx
) ? *ap
++ : 0)
2417 != (flowmap_is_set(&b
->map
, idx
) ? *bp
++ : 0)) {
2426 /* Returns false if 'a' and 'b' differ at the places where there are 1-bits
2427 * in 'mask', true otherwise. */
2429 miniflow_equal_in_minimask(const struct miniflow
*a
, const struct miniflow
*b
,
2430 const struct minimask
*mask
)
2432 const uint64_t *p
= miniflow_get_values(&mask
->masks
);
2435 FLOWMAP_FOR_EACH_INDEX(idx
, mask
->masks
.map
) {
2436 if ((miniflow_get(a
, idx
) ^ miniflow_get(b
, idx
)) & *p
++) {
2444 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
2445 * in 'mask', false if they differ. */
2447 miniflow_equal_flow_in_minimask(const struct miniflow
*a
, const struct flow
*b
,
2448 const struct minimask
*mask
)
2450 const uint64_t *p
= miniflow_get_values(&mask
->masks
);
2453 FLOWMAP_FOR_EACH_INDEX(idx
, mask
->masks
.map
) {
2454 if ((miniflow_get(a
, idx
) ^ flow_u64_value(b
, idx
)) & *p
++) {
2464 minimask_init(struct minimask
*mask
, const struct flow_wildcards
*wc
)
2466 miniflow_init(&mask
->masks
, &wc
->masks
);
2469 /* Returns a minimask copy of 'wc'. The caller must eventually free the
2470 * returned minimask with free(). */
2472 minimask_create(const struct flow_wildcards
*wc
)
2474 return (struct minimask
*)miniflow_create(&wc
->masks
);
2477 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
2479 * The caller must provide room for FLOW_U64S "uint64_t"s in 'storage', which
2480 * must follow '*dst_' in memory, for use by 'dst_'. The caller must *not*
2481 * free 'dst_' free(). */
2483 minimask_combine(struct minimask
*dst_
,
2484 const struct minimask
*a_
, const struct minimask
*b_
,
2485 uint64_t storage
[FLOW_U64S
])
2487 struct miniflow
*dst
= &dst_
->masks
;
2488 uint64_t *dst_values
= storage
;
2489 const struct miniflow
*a
= &a_
->masks
;
2490 const struct miniflow
*b
= &b_
->masks
;
2493 flowmap_init(&dst
->map
);
2495 FLOWMAP_FOR_EACH_INDEX(idx
, flowmap_and(a
->map
, b
->map
)) {
2496 /* Both 'a' and 'b' have non-zero data at 'idx'. */
2497 uint64_t mask
= *miniflow_get__(a
, idx
) & *miniflow_get__(b
, idx
);
2500 flowmap_set(&dst
->map
, idx
, 1);
2501 *dst_values
++ = mask
;
2506 /* Initializes 'wc' as a copy of 'mask'. */
2508 minimask_expand(const struct minimask
*mask
, struct flow_wildcards
*wc
)
2510 miniflow_expand(&mask
->masks
, &wc
->masks
);
2513 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise.
2514 * Minimasks may not have zero data values, so for the minimasks to be the
2515 * same, they need to have the same map and the same data values. */
2517 minimask_equal(const struct minimask
*a
, const struct minimask
*b
)
2519 return !memcmp(a
, b
, sizeof *a
2520 + MINIFLOW_VALUES_SIZE(miniflow_n_values(&a
->masks
)));
2523 /* Returns true if at least one bit matched by 'b' is wildcarded by 'a',
2524 * false otherwise. */
2526 minimask_has_extra(const struct minimask
*a
, const struct minimask
*b
)
2528 const uint64_t *bp
= miniflow_get_values(&b
->masks
);
2531 FLOWMAP_FOR_EACH_INDEX(idx
, b
->masks
.map
) {
2532 uint64_t b_u64
= *bp
++;
2534 /* 'b_u64' is non-zero, check if the data in 'a' is either zero
2535 * or misses some of the bits in 'b_u64'. */
2536 if (!MINIFLOW_IN_MAP(&a
->masks
, idx
)
2537 || ((*miniflow_get__(&a
->masks
, idx
) & b_u64
) != b_u64
)) {
2538 return true; /* 'a' wildcards some bits 'b' doesn't. */