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 /* miniflow_extract() assumes the following to be true to optimize the
54 * extraction process. */
55 BUILD_ASSERT_DECL(offsetof(struct flow
, dl_type
) + 2
56 == offsetof(struct flow
, vlan_tci
) &&
57 offsetof(struct flow
, dl_type
) / 4
58 == offsetof(struct flow
, vlan_tci
) / 4 );
60 BUILD_ASSERT_DECL(offsetof(struct flow
, nw_frag
) + 3
61 == offsetof(struct flow
, nw_proto
) &&
62 offsetof(struct flow
, nw_tos
) + 2
63 == offsetof(struct flow
, nw_proto
) &&
64 offsetof(struct flow
, nw_ttl
) + 1
65 == offsetof(struct flow
, nw_proto
) &&
66 offsetof(struct flow
, nw_frag
) / 4
67 == offsetof(struct flow
, nw_tos
) / 4 &&
68 offsetof(struct flow
, nw_ttl
) / 4
69 == offsetof(struct flow
, nw_tos
) / 4 &&
70 offsetof(struct flow
, nw_proto
) / 4
71 == offsetof(struct flow
, nw_tos
) / 4);
73 /* TCP flags in the middle of a BE64, zeroes in the other half. */
74 BUILD_ASSERT_DECL(offsetof(struct flow
, tcp_flags
) % 8 == 4);
77 #define TCP_FLAGS_BE32(tcp_ctl) ((OVS_FORCE ovs_be32)TCP_FLAGS_BE16(tcp_ctl) \
80 #define TCP_FLAGS_BE32(tcp_ctl) ((OVS_FORCE ovs_be32)TCP_FLAGS_BE16(tcp_ctl))
83 BUILD_ASSERT_DECL(offsetof(struct flow
, tp_src
) + 2
84 == offsetof(struct flow
, tp_dst
) &&
85 offsetof(struct flow
, tp_src
) / 4
86 == offsetof(struct flow
, tp_dst
) / 4);
88 /* Removes 'size' bytes from the head end of '*datap', of size '*sizep', which
89 * must contain at least 'size' bytes of data. Returns the first byte of data
91 static inline const void *
92 data_pull(void **datap
, size_t *sizep
, size_t size
)
94 char *data
= (char *)*datap
;
100 /* If '*datap' has at least 'size' bytes of data, removes that many bytes from
101 * the head end of '*datap' and returns the first byte removed. Otherwise,
102 * returns a null pointer without modifying '*datap'. */
103 static inline const void *
104 data_try_pull(void **datap
, size_t *sizep
, size_t size
)
106 return OVS_LIKELY(*sizep
>= size
) ? data_pull(datap
, sizep
, size
) : NULL
;
109 /* Context for pushing data to a miniflow. */
113 uint64_t * const end
;
116 /* miniflow_push_* macros allow filling in a miniflow data values in order.
117 * Assertions are needed only when the layout of the struct flow is modified.
118 * 'ofs' is a compile-time constant, which allows most of the code be optimized
119 * away. Some GCC versions gave warnings on ALWAYS_INLINE, so these are
120 * defined as macros. */
122 #if (FLOW_WC_SEQ != 31)
123 #define MINIFLOW_ASSERT(X) ovs_assert(X)
124 BUILD_MESSAGE("FLOW_WC_SEQ changed: miniflow_extract() will have runtime "
125 "assertions enabled. Consider updating FLOW_WC_SEQ after "
128 #define MINIFLOW_ASSERT(X)
131 #define miniflow_push_uint64_(MF, OFS, VALUE) \
133 MINIFLOW_ASSERT(MF.data < MF.end && (OFS) % 8 == 0 \
134 && !(MF.map & (UINT64_MAX << (OFS) / 8))); \
135 *MF.data++ = VALUE; \
136 MF.map |= UINT64_C(1) << (OFS) / 8; \
139 #define miniflow_push_be64_(MF, OFS, VALUE) \
140 miniflow_push_uint64_(MF, OFS, (OVS_FORCE uint64_t)(VALUE))
142 #define miniflow_push_uint32_(MF, OFS, VALUE) \
144 MINIFLOW_ASSERT(MF.data < MF.end && \
145 (((OFS) % 8 == 0 && !(MF.map & (UINT64_MAX << (OFS) / 8))) \
146 || ((OFS) % 8 == 4 && MF.map & (UINT64_C(1) << (OFS) / 8) \
147 && !(MF.map & (UINT64_MAX << ((OFS) / 8 + 1)))))); \
149 if ((OFS) % 8 == 0) { \
150 *(uint32_t *)MF.data = VALUE; \
151 MF.map |= UINT64_C(1) << (OFS) / 8; \
152 } else if ((OFS) % 8 == 4) { \
153 *((uint32_t *)MF.data + 1) = VALUE; \
158 #define miniflow_push_be32_(MF, OFS, VALUE) \
159 miniflow_push_uint32_(MF, OFS, (OVS_FORCE uint32_t)(VALUE))
161 #define miniflow_push_uint16_(MF, OFS, VALUE) \
163 MINIFLOW_ASSERT(MF.data < MF.end && \
164 (((OFS) % 8 == 0 && !(MF.map & (UINT64_MAX << (OFS) / 8))) \
165 || ((OFS) % 2 == 0 && MF.map & (UINT64_C(1) << (OFS) / 8) \
166 && !(MF.map & (UINT64_MAX << ((OFS) / 8 + 1)))))); \
168 if ((OFS) % 8 == 0) { \
169 *(uint16_t *)MF.data = VALUE; \
170 MF.map |= UINT64_C(1) << (OFS) / 8; \
171 } else if ((OFS) % 8 == 2) { \
172 *((uint16_t *)MF.data + 1) = VALUE; \
173 } else if ((OFS) % 8 == 4) { \
174 *((uint16_t *)MF.data + 2) = VALUE; \
175 } else if ((OFS) % 8 == 6) { \
176 *((uint16_t *)MF.data + 3) = VALUE; \
181 #define miniflow_pad_to_64_(MF, OFS) \
183 MINIFLOW_ASSERT((OFS) % 8 != 0); \
184 MINIFLOW_ASSERT(MF.map & (UINT64_C(1) << (OFS) / 8)); \
185 MINIFLOW_ASSERT(!(MF.map & (UINT64_MAX << ((OFS) / 8 + 1)))); \
187 memset((uint8_t *)MF.data + (OFS) % 8, 0, 8 - (OFS) % 8); \
191 #define miniflow_push_be16_(MF, OFS, VALUE) \
192 miniflow_push_uint16_(MF, OFS, (OVS_FORCE uint16_t)VALUE);
194 /* Data at 'valuep' may be unaligned. */
195 #define miniflow_push_words_(MF, OFS, VALUEP, N_WORDS) \
197 int ofs64 = (OFS) / 8; \
199 MINIFLOW_ASSERT(MF.data + (N_WORDS) <= MF.end && (OFS) % 8 == 0 \
200 && !(MF.map & (UINT64_MAX << ofs64))); \
202 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof *MF.data); \
203 MF.data += (N_WORDS); \
204 MF.map |= ((UINT64_MAX >> (64 - (N_WORDS))) << ofs64); \
207 /* Push 32-bit words padded to 64-bits. */
208 #define miniflow_push_words_32_(MF, OFS, VALUEP, N_WORDS) \
210 int ofs64 = (OFS) / 8; \
212 MINIFLOW_ASSERT(MF.data + DIV_ROUND_UP(N_WORDS, 2) <= MF.end \
214 && !(MF.map & (UINT64_MAX << ofs64))); \
216 memcpy(MF.data, (VALUEP), (N_WORDS) * sizeof(uint32_t)); \
217 MF.data += DIV_ROUND_UP(N_WORDS, 2); \
218 MF.map |= ((UINT64_MAX >> (64 - DIV_ROUND_UP(N_WORDS, 2))) << ofs64); \
219 if ((N_WORDS) & 1) { \
220 *((uint32_t *)MF.data - 1) = 0; \
224 /* Data at 'valuep' may be unaligned. */
225 /* MACs start 64-aligned, and must be followed by other data or padding. */
226 #define miniflow_push_macs_(MF, OFS, VALUEP) \
228 int ofs64 = (OFS) / 8; \
230 MINIFLOW_ASSERT(MF.data + 2 <= MF.end && (OFS) % 8 == 0 \
231 && !(MF.map & (UINT64_MAX << ofs64))); \
233 memcpy(MF.data, (VALUEP), 2 * ETH_ADDR_LEN); \
234 MF.data += 1; /* First word only. */ \
235 MF.map |= UINT64_C(3) << ofs64; /* Both words. */ \
238 #define miniflow_push_uint32(MF, FIELD, VALUE) \
239 miniflow_push_uint32_(MF, offsetof(struct flow, FIELD), VALUE)
241 #define miniflow_push_be32(MF, FIELD, VALUE) \
242 miniflow_push_be32_(MF, offsetof(struct flow, FIELD), VALUE)
244 #define miniflow_push_uint16(MF, FIELD, VALUE) \
245 miniflow_push_uint16_(MF, offsetof(struct flow, FIELD), VALUE)
247 #define miniflow_push_be16(MF, FIELD, VALUE) \
248 miniflow_push_be16_(MF, offsetof(struct flow, FIELD), VALUE)
250 #define miniflow_pad_to_64(MF, FIELD) \
251 miniflow_pad_to_64_(MF, offsetof(struct flow, FIELD))
253 #define miniflow_push_words(MF, FIELD, VALUEP, N_WORDS) \
254 miniflow_push_words_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
256 #define miniflow_push_words_32(MF, FIELD, VALUEP, N_WORDS) \
257 miniflow_push_words_32_(MF, offsetof(struct flow, FIELD), VALUEP, N_WORDS)
259 #define miniflow_push_macs(MF, FIELD, VALUEP) \
260 miniflow_push_macs_(MF, offsetof(struct flow, FIELD), VALUEP)
262 /* Pulls the MPLS headers at '*datap' and returns the count of them. */
264 parse_mpls(void **datap
, size_t *sizep
)
266 const struct mpls_hdr
*mh
;
269 while ((mh
= data_try_pull(datap
, sizep
, sizeof *mh
))) {
271 if (mh
->mpls_lse
.lo
& htons(1 << MPLS_BOS_SHIFT
)) {
275 return MIN(count
, FLOW_MAX_MPLS_LABELS
);
278 static inline ovs_be16
279 parse_vlan(void **datap
, size_t *sizep
)
281 const struct eth_header
*eth
= *datap
;
284 ovs_be16 eth_type
; /* ETH_TYPE_VLAN */
288 data_pull(datap
, sizep
, ETH_ADDR_LEN
* 2);
290 if (eth
->eth_type
== htons(ETH_TYPE_VLAN
)) {
291 if (OVS_LIKELY(*sizep
292 >= sizeof(struct qtag_prefix
) + sizeof(ovs_be16
))) {
293 const struct qtag_prefix
*qp
= data_pull(datap
, sizep
, sizeof *qp
);
294 return qp
->tci
| htons(VLAN_CFI
);
300 static inline ovs_be16
301 parse_ethertype(void **datap
, size_t *sizep
)
303 const struct llc_snap_header
*llc
;
306 proto
= *(ovs_be16
*) data_pull(datap
, sizep
, sizeof proto
);
307 if (OVS_LIKELY(ntohs(proto
) >= ETH_TYPE_MIN
)) {
311 if (OVS_UNLIKELY(*sizep
< sizeof *llc
)) {
312 return htons(FLOW_DL_TYPE_NONE
);
316 if (OVS_UNLIKELY(llc
->llc
.llc_dsap
!= LLC_DSAP_SNAP
317 || llc
->llc
.llc_ssap
!= LLC_SSAP_SNAP
318 || llc
->llc
.llc_cntl
!= LLC_CNTL_SNAP
319 || memcmp(llc
->snap
.snap_org
, SNAP_ORG_ETHERNET
,
320 sizeof llc
->snap
.snap_org
))) {
321 return htons(FLOW_DL_TYPE_NONE
);
324 data_pull(datap
, sizep
, sizeof *llc
);
326 if (OVS_LIKELY(ntohs(llc
->snap
.snap_type
) >= ETH_TYPE_MIN
)) {
327 return llc
->snap
.snap_type
;
330 return htons(FLOW_DL_TYPE_NONE
);
334 parse_icmpv6(void **datap
, size_t *sizep
, const struct icmp6_hdr
*icmp
,
335 const struct in6_addr
**nd_target
,
336 uint8_t arp_buf
[2][ETH_ADDR_LEN
])
338 if (icmp
->icmp6_code
== 0 &&
339 (icmp
->icmp6_type
== ND_NEIGHBOR_SOLICIT
||
340 icmp
->icmp6_type
== ND_NEIGHBOR_ADVERT
)) {
342 *nd_target
= data_try_pull(datap
, sizep
, sizeof **nd_target
);
343 if (OVS_UNLIKELY(!*nd_target
)) {
347 while (*sizep
>= 8) {
348 /* The minimum size of an option is 8 bytes, which also is
349 * the size of Ethernet link-layer options. */
350 const struct nd_opt_hdr
*nd_opt
= *datap
;
351 int opt_len
= nd_opt
->nd_opt_len
* 8;
353 if (!opt_len
|| opt_len
> *sizep
) {
357 /* Store the link layer address if the appropriate option is
358 * provided. It is considered an error if the same link
359 * layer option is specified twice. */
360 if (nd_opt
->nd_opt_type
== ND_OPT_SOURCE_LINKADDR
362 if (OVS_LIKELY(eth_addr_is_zero(arp_buf
[0]))) {
363 memcpy(arp_buf
[0], nd_opt
+ 1, ETH_ADDR_LEN
);
367 } else if (nd_opt
->nd_opt_type
== ND_OPT_TARGET_LINKADDR
369 if (OVS_LIKELY(eth_addr_is_zero(arp_buf
[1]))) {
370 memcpy(arp_buf
[1], nd_opt
+ 1, ETH_ADDR_LEN
);
376 if (OVS_UNLIKELY(!data_try_pull(datap
, sizep
, opt_len
))) {
388 /* Initializes 'flow' members from 'packet' and 'md'
390 * Initializes 'packet' header l2 pointer to the start of the Ethernet
391 * header, and the layer offsets as follows:
393 * - packet->l2_5_ofs to the start of the MPLS shim header, or UINT16_MAX
394 * when there is no MPLS shim header.
396 * - packet->l3_ofs to just past the Ethernet header, or just past the
397 * vlan_header if one is present, to the first byte of the payload of the
398 * Ethernet frame. UINT16_MAX if the frame is too short to contain an
401 * - packet->l4_ofs to just past the IPv4 header, if one is present and
402 * has at least the content used for the fields of interest for the flow,
403 * otherwise UINT16_MAX.
406 flow_extract(struct dp_packet
*packet
, struct flow
*flow
)
410 uint64_t buf
[FLOW_U64S
];
413 COVERAGE_INC(flow_extract
);
415 miniflow_initialize(&m
.mf
, m
.buf
);
416 miniflow_extract(packet
, &m
.mf
);
417 miniflow_expand(&m
.mf
, flow
);
420 /* Caller is responsible for initializing 'dst' with enough storage for
421 * FLOW_U64S * 8 bytes. */
423 miniflow_extract(struct dp_packet
*packet
, struct miniflow
*dst
)
425 const struct pkt_metadata
*md
= &packet
->md
;
426 void *data
= dp_packet_data(packet
);
427 size_t size
= dp_packet_size(packet
);
428 uint64_t *values
= miniflow_values(dst
);
429 struct mf_ctx mf
= { 0, values
, values
+ FLOW_U64S
};
432 uint8_t nw_frag
, nw_tos
, nw_ttl
, nw_proto
;
435 if (md
->tunnel
.ip_dst
) {
436 miniflow_push_words(mf
, tunnel
, &md
->tunnel
,
437 sizeof md
->tunnel
/ sizeof(uint64_t));
439 if (md
->skb_priority
|| md
->pkt_mark
) {
440 miniflow_push_uint32(mf
, skb_priority
, md
->skb_priority
);
441 miniflow_push_uint32(mf
, pkt_mark
, md
->pkt_mark
);
443 miniflow_push_uint32(mf
, dp_hash
, md
->dp_hash
);
444 miniflow_push_uint32(mf
, in_port
, odp_to_u32(md
->in_port
.odp_port
));
446 miniflow_push_uint32(mf
, recirc_id
, md
->recirc_id
);
447 miniflow_pad_to_64(mf
, conj_id
);
450 /* Initialize packet's layer pointer and offsets. */
452 dp_packet_reset_offsets(packet
);
454 /* Must have full Ethernet header to proceed. */
455 if (OVS_UNLIKELY(size
< sizeof(struct eth_header
))) {
461 BUILD_ASSERT(offsetof(struct flow
, dl_dst
) + 6
462 == offsetof(struct flow
, dl_src
));
463 miniflow_push_macs(mf
, dl_dst
, data
);
464 /* dl_type, vlan_tci. */
465 vlan_tci
= parse_vlan(&data
, &size
);
466 dl_type
= parse_ethertype(&data
, &size
);
467 miniflow_push_be16(mf
, dl_type
, dl_type
);
468 miniflow_push_be16(mf
, vlan_tci
, vlan_tci
);
472 if (OVS_UNLIKELY(eth_type_mpls(dl_type
))) {
474 const void *mpls
= data
;
476 packet
->l2_5_ofs
= (char *)data
- l2
;
477 count
= parse_mpls(&data
, &size
);
478 miniflow_push_words_32(mf
, mpls_lse
, mpls
, count
);
482 packet
->l3_ofs
= (char *)data
- l2
;
485 if (OVS_LIKELY(dl_type
== htons(ETH_TYPE_IP
))) {
486 const struct ip_header
*nh
= data
;
490 if (OVS_UNLIKELY(size
< IP_HEADER_LEN
)) {
493 ip_len
= IP_IHL(nh
->ip_ihl_ver
) * 4;
495 if (OVS_UNLIKELY(ip_len
< IP_HEADER_LEN
)) {
498 if (OVS_UNLIKELY(size
< ip_len
)) {
501 tot_len
= ntohs(nh
->ip_tot_len
);
502 if (OVS_UNLIKELY(tot_len
> size
)) {
505 if (OVS_UNLIKELY(size
- tot_len
> UINT8_MAX
)) {
508 dp_packet_set_l2_pad_size(packet
, size
- tot_len
);
509 size
= tot_len
; /* Never pull padding. */
511 /* Push both source and destination address at once. */
512 miniflow_push_words(mf
, nw_src
, &nh
->ip_src
, 1);
514 miniflow_push_be32(mf
, ipv6_label
, 0); /* Padding for IPv4. */
518 nw_proto
= nh
->ip_proto
;
519 if (OVS_UNLIKELY(IP_IS_FRAGMENT(nh
->ip_frag_off
))) {
520 nw_frag
= FLOW_NW_FRAG_ANY
;
521 if (nh
->ip_frag_off
& htons(IP_FRAG_OFF_MASK
)) {
522 nw_frag
|= FLOW_NW_FRAG_LATER
;
525 data_pull(&data
, &size
, ip_len
);
526 } else if (dl_type
== htons(ETH_TYPE_IPV6
)) {
527 const struct ovs_16aligned_ip6_hdr
*nh
;
531 if (OVS_UNLIKELY(size
< sizeof *nh
)) {
534 nh
= data_pull(&data
, &size
, sizeof *nh
);
536 plen
= ntohs(nh
->ip6_plen
);
537 if (OVS_UNLIKELY(plen
> size
)) {
540 /* Jumbo Payload option not supported yet. */
541 if (OVS_UNLIKELY(size
- plen
> UINT8_MAX
)) {
544 dp_packet_set_l2_pad_size(packet
, size
- plen
);
545 size
= plen
; /* Never pull padding. */
547 miniflow_push_words(mf
, ipv6_src
, &nh
->ip6_src
,
548 sizeof nh
->ip6_src
/ 8);
549 miniflow_push_words(mf
, ipv6_dst
, &nh
->ip6_dst
,
550 sizeof nh
->ip6_dst
/ 8);
552 tc_flow
= get_16aligned_be32(&nh
->ip6_flow
);
554 ovs_be32 label
= tc_flow
& htonl(IPV6_LABEL_MASK
);
555 miniflow_push_be32(mf
, ipv6_label
, label
);
558 nw_tos
= ntohl(tc_flow
) >> 20;
559 nw_ttl
= nh
->ip6_hlim
;
560 nw_proto
= nh
->ip6_nxt
;
563 if (OVS_LIKELY((nw_proto
!= IPPROTO_HOPOPTS
)
564 && (nw_proto
!= IPPROTO_ROUTING
)
565 && (nw_proto
!= IPPROTO_DSTOPTS
)
566 && (nw_proto
!= IPPROTO_AH
)
567 && (nw_proto
!= IPPROTO_FRAGMENT
))) {
568 /* It's either a terminal header (e.g., TCP, UDP) or one we
569 * don't understand. In either case, we're done with the
570 * packet, so use it to fill in 'nw_proto'. */
574 /* We only verify that at least 8 bytes of the next header are
575 * available, but many of these headers are longer. Ensure that
576 * accesses within the extension header are within those first 8
577 * bytes. All extension headers are required to be at least 8
579 if (OVS_UNLIKELY(size
< 8)) {
583 if ((nw_proto
== IPPROTO_HOPOPTS
)
584 || (nw_proto
== IPPROTO_ROUTING
)
585 || (nw_proto
== IPPROTO_DSTOPTS
)) {
586 /* These headers, while different, have the fields we care
587 * about in the same location and with the same
589 const struct ip6_ext
*ext_hdr
= data
;
590 nw_proto
= ext_hdr
->ip6e_nxt
;
591 if (OVS_UNLIKELY(!data_try_pull(&data
, &size
,
592 (ext_hdr
->ip6e_len
+ 1) * 8))) {
595 } else if (nw_proto
== IPPROTO_AH
) {
596 /* A standard AH definition isn't available, but the fields
597 * we care about are in the same location as the generic
598 * option header--only the header length is calculated
600 const struct ip6_ext
*ext_hdr
= data
;
601 nw_proto
= ext_hdr
->ip6e_nxt
;
602 if (OVS_UNLIKELY(!data_try_pull(&data
, &size
,
603 (ext_hdr
->ip6e_len
+ 2) * 4))) {
606 } else if (nw_proto
== IPPROTO_FRAGMENT
) {
607 const struct ovs_16aligned_ip6_frag
*frag_hdr
= data
;
609 nw_proto
= frag_hdr
->ip6f_nxt
;
610 if (!data_try_pull(&data
, &size
, sizeof *frag_hdr
)) {
614 /* We only process the first fragment. */
615 if (frag_hdr
->ip6f_offlg
!= htons(0)) {
616 nw_frag
= FLOW_NW_FRAG_ANY
;
617 if ((frag_hdr
->ip6f_offlg
& IP6F_OFF_MASK
) != htons(0)) {
618 nw_frag
|= FLOW_NW_FRAG_LATER
;
619 nw_proto
= IPPROTO_FRAGMENT
;
626 if (dl_type
== htons(ETH_TYPE_ARP
) ||
627 dl_type
== htons(ETH_TYPE_RARP
)) {
628 uint8_t arp_buf
[2][ETH_ADDR_LEN
];
629 const struct arp_eth_header
*arp
= (const struct arp_eth_header
*)
630 data_try_pull(&data
, &size
, ARP_ETH_HEADER_LEN
);
632 if (OVS_LIKELY(arp
) && OVS_LIKELY(arp
->ar_hrd
== htons(1))
633 && OVS_LIKELY(arp
->ar_pro
== htons(ETH_TYPE_IP
))
634 && OVS_LIKELY(arp
->ar_hln
== ETH_ADDR_LEN
)
635 && OVS_LIKELY(arp
->ar_pln
== 4)) {
636 miniflow_push_be32(mf
, nw_src
,
637 get_16aligned_be32(&arp
->ar_spa
));
638 miniflow_push_be32(mf
, nw_dst
,
639 get_16aligned_be32(&arp
->ar_tpa
));
641 /* We only match on the lower 8 bits of the opcode. */
642 if (OVS_LIKELY(ntohs(arp
->ar_op
) <= 0xff)) {
643 miniflow_push_be32(mf
, ipv6_label
, 0); /* Pad with ARP. */
644 miniflow_push_be32(mf
, nw_frag
, htonl(ntohs(arp
->ar_op
)));
647 /* Must be adjacent. */
648 BUILD_ASSERT(offsetof(struct flow
, arp_sha
) + 6
649 == offsetof(struct flow
, arp_tha
));
651 memcpy(arp_buf
[0], arp
->ar_sha
, ETH_ADDR_LEN
);
652 memcpy(arp_buf
[1], arp
->ar_tha
, ETH_ADDR_LEN
);
653 miniflow_push_macs(mf
, arp_sha
, arp_buf
);
654 miniflow_pad_to_64(mf
, tcp_flags
);
660 packet
->l4_ofs
= (char *)data
- l2
;
661 miniflow_push_be32(mf
, nw_frag
,
662 BYTES_TO_BE32(nw_frag
, nw_tos
, nw_ttl
, nw_proto
));
664 if (OVS_LIKELY(!(nw_frag
& FLOW_NW_FRAG_LATER
))) {
665 if (OVS_LIKELY(nw_proto
== IPPROTO_TCP
)) {
666 if (OVS_LIKELY(size
>= TCP_HEADER_LEN
)) {
667 const struct tcp_header
*tcp
= data
;
669 miniflow_push_be32(mf
, arp_tha
[2], 0);
670 miniflow_push_be32(mf
, tcp_flags
,
671 TCP_FLAGS_BE32(tcp
->tcp_ctl
));
672 miniflow_push_be16(mf
, tp_src
, tcp
->tcp_src
);
673 miniflow_push_be16(mf
, tp_dst
, tcp
->tcp_dst
);
674 miniflow_pad_to_64(mf
, igmp_group_ip4
);
676 } else if (OVS_LIKELY(nw_proto
== IPPROTO_UDP
)) {
677 if (OVS_LIKELY(size
>= UDP_HEADER_LEN
)) {
678 const struct udp_header
*udp
= data
;
680 miniflow_push_be16(mf
, tp_src
, udp
->udp_src
);
681 miniflow_push_be16(mf
, tp_dst
, udp
->udp_dst
);
682 miniflow_pad_to_64(mf
, igmp_group_ip4
);
684 } else if (OVS_LIKELY(nw_proto
== IPPROTO_SCTP
)) {
685 if (OVS_LIKELY(size
>= SCTP_HEADER_LEN
)) {
686 const struct sctp_header
*sctp
= data
;
688 miniflow_push_be16(mf
, tp_src
, sctp
->sctp_src
);
689 miniflow_push_be16(mf
, tp_dst
, sctp
->sctp_dst
);
690 miniflow_pad_to_64(mf
, igmp_group_ip4
);
692 } else if (OVS_LIKELY(nw_proto
== IPPROTO_ICMP
)) {
693 if (OVS_LIKELY(size
>= ICMP_HEADER_LEN
)) {
694 const struct icmp_header
*icmp
= data
;
696 miniflow_push_be16(mf
, tp_src
, htons(icmp
->icmp_type
));
697 miniflow_push_be16(mf
, tp_dst
, htons(icmp
->icmp_code
));
698 miniflow_pad_to_64(mf
, igmp_group_ip4
);
700 } else if (OVS_LIKELY(nw_proto
== IPPROTO_IGMP
)) {
701 if (OVS_LIKELY(size
>= IGMP_HEADER_LEN
)) {
702 const struct igmp_header
*igmp
= data
;
704 miniflow_push_be16(mf
, tp_src
, htons(igmp
->igmp_type
));
705 miniflow_push_be16(mf
, tp_dst
, htons(igmp
->igmp_code
));
706 miniflow_push_be32(mf
, igmp_group_ip4
,
707 get_16aligned_be32(&igmp
->group
));
709 } else if (OVS_LIKELY(nw_proto
== IPPROTO_ICMPV6
)) {
710 if (OVS_LIKELY(size
>= sizeof(struct icmp6_hdr
))) {
711 const struct in6_addr
*nd_target
= NULL
;
712 uint8_t arp_buf
[2][ETH_ADDR_LEN
];
713 const struct icmp6_hdr
*icmp
= data_pull(&data
, &size
,
715 memset(arp_buf
, 0, sizeof arp_buf
);
716 if (OVS_LIKELY(parse_icmpv6(&data
, &size
, icmp
, &nd_target
,
719 miniflow_push_words(mf
, nd_target
, nd_target
,
720 sizeof *nd_target
/ 8);
722 miniflow_push_macs(mf
, arp_sha
, arp_buf
);
723 miniflow_pad_to_64(mf
, tcp_flags
);
724 miniflow_push_be16(mf
, tp_src
, htons(icmp
->icmp6_type
));
725 miniflow_push_be16(mf
, tp_dst
, htons(icmp
->icmp6_code
));
726 miniflow_pad_to_64(mf
, igmp_group_ip4
);
735 /* For every bit of a field that is wildcarded in 'wildcards', sets the
736 * corresponding bit in 'flow' to zero. */
738 flow_zero_wildcards(struct flow
*flow
, const struct flow_wildcards
*wildcards
)
740 uint64_t *flow_u64
= (uint64_t *) flow
;
741 const uint64_t *wc_u64
= (const uint64_t *) &wildcards
->masks
;
744 for (i
= 0; i
< FLOW_U64S
; i
++) {
745 flow_u64
[i
] &= wc_u64
[i
];
750 flow_unwildcard_tp_ports(const struct flow
*flow
, struct flow_wildcards
*wc
)
752 if (flow
->nw_proto
!= IPPROTO_ICMP
) {
753 memset(&wc
->masks
.tp_src
, 0xff, sizeof wc
->masks
.tp_src
);
754 memset(&wc
->masks
.tp_dst
, 0xff, sizeof wc
->masks
.tp_dst
);
756 wc
->masks
.tp_src
= htons(0xff);
757 wc
->masks
.tp_dst
= htons(0xff);
761 /* Initializes 'fmd' with the metadata found in 'flow'. */
763 flow_get_metadata(const struct flow
*flow
, struct flow_metadata
*fmd
)
765 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 31);
767 fmd
->dp_hash
= flow
->dp_hash
;
768 fmd
->recirc_id
= flow
->recirc_id
;
769 fmd
->tun_id
= flow
->tunnel
.tun_id
;
770 fmd
->tun_src
= flow
->tunnel
.ip_src
;
771 fmd
->tun_dst
= flow
->tunnel
.ip_dst
;
772 fmd
->gbp_id
= flow
->tunnel
.gbp_id
;
773 fmd
->gbp_flags
= flow
->tunnel
.gbp_flags
;
774 fmd
->metadata
= flow
->metadata
;
775 memcpy(fmd
->regs
, flow
->regs
, sizeof fmd
->regs
);
776 fmd
->pkt_mark
= flow
->pkt_mark
;
777 fmd
->in_port
= flow
->in_port
.ofp_port
;
781 flow_to_string(const struct flow
*flow
)
783 struct ds ds
= DS_EMPTY_INITIALIZER
;
784 flow_format(&ds
, flow
);
789 flow_tun_flag_to_string(uint32_t flags
)
792 case FLOW_TNL_F_DONT_FRAGMENT
:
794 case FLOW_TNL_F_CSUM
:
806 format_flags(struct ds
*ds
, const char *(*bit_to_string
)(uint32_t),
807 uint32_t flags
, char del
)
815 uint32_t bit
= rightmost_1bit(flags
);
818 s
= bit_to_string(bit
);
820 ds_put_format(ds
, "%s%c", s
, del
);
829 ds_put_format(ds
, "0x%"PRIx32
"%c", bad
, del
);
835 format_flags_masked(struct ds
*ds
, const char *name
,
836 const char *(*bit_to_string
)(uint32_t), uint32_t flags
,
840 ds_put_format(ds
, "%s=", name
);
843 uint32_t bit
= rightmost_1bit(mask
);
844 const char *s
= bit_to_string(bit
);
846 ds_put_format(ds
, "%s%s", (flags
& bit
) ? "+" : "-",
847 s
? s
: "[Unknown]");
853 flow_format(struct ds
*ds
, const struct flow
*flow
)
856 struct flow_wildcards
*wc
= &match
.wc
;
858 match_wc_init(&match
, flow
);
860 /* As this function is most often used for formatting a packet in a
861 * packet-in message, skip formatting the packet context fields that are
862 * all-zeroes to make the print-out easier on the eyes. This means that a
863 * missing context field implies a zero value for that field. This is
864 * similar to OpenFlow encoding of these fields, as the specification
865 * states that all-zeroes context fields should not be encoded in the
866 * packet-in messages. */
867 if (!flow
->in_port
.ofp_port
) {
868 WC_UNMASK_FIELD(wc
, in_port
);
870 if (!flow
->skb_priority
) {
871 WC_UNMASK_FIELD(wc
, skb_priority
);
873 if (!flow
->pkt_mark
) {
874 WC_UNMASK_FIELD(wc
, pkt_mark
);
876 if (!flow
->recirc_id
) {
877 WC_UNMASK_FIELD(wc
, recirc_id
);
879 if (!flow
->dp_hash
) {
880 WC_UNMASK_FIELD(wc
, dp_hash
);
882 for (int i
= 0; i
< FLOW_N_REGS
; i
++) {
883 if (!flow
->regs
[i
]) {
884 WC_UNMASK_FIELD(wc
, regs
[i
]);
887 if (!flow
->metadata
) {
888 WC_UNMASK_FIELD(wc
, metadata
);
891 match_format(&match
, ds
, OFP_DEFAULT_PRIORITY
);
895 flow_print(FILE *stream
, const struct flow
*flow
)
897 char *s
= flow_to_string(flow
);
902 /* flow_wildcards functions. */
904 /* Initializes 'wc' as a set of wildcards that matches every packet. */
906 flow_wildcards_init_catchall(struct flow_wildcards
*wc
)
908 memset(&wc
->masks
, 0, sizeof wc
->masks
);
911 /* Converts a flow into flow wildcards. It sets the wildcard masks based on
912 * the packet headers extracted to 'flow'. It will not set the mask for fields
913 * that do not make sense for the packet type. OpenFlow-only metadata is
914 * wildcarded, but other metadata is unconditionally exact-matched. */
915 void flow_wildcards_init_for_packet(struct flow_wildcards
*wc
,
916 const struct flow
*flow
)
918 memset(&wc
->masks
, 0x0, sizeof wc
->masks
);
920 /* Update this function whenever struct flow changes. */
921 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 31);
923 if (flow
->tunnel
.ip_dst
) {
924 if (flow
->tunnel
.flags
& FLOW_TNL_F_KEY
) {
925 WC_MASK_FIELD(wc
, tunnel
.tun_id
);
927 WC_MASK_FIELD(wc
, tunnel
.ip_src
);
928 WC_MASK_FIELD(wc
, tunnel
.ip_dst
);
929 WC_MASK_FIELD(wc
, tunnel
.flags
);
930 WC_MASK_FIELD(wc
, tunnel
.ip_tos
);
931 WC_MASK_FIELD(wc
, tunnel
.ip_ttl
);
932 WC_MASK_FIELD(wc
, tunnel
.tp_src
);
933 WC_MASK_FIELD(wc
, tunnel
.tp_dst
);
934 WC_MASK_FIELD(wc
, tunnel
.gbp_id
);
935 WC_MASK_FIELD(wc
, tunnel
.gbp_flags
);
936 } else if (flow
->tunnel
.tun_id
) {
937 WC_MASK_FIELD(wc
, tunnel
.tun_id
);
940 /* metadata, regs, and conj_id wildcarded. */
942 WC_MASK_FIELD(wc
, skb_priority
);
943 WC_MASK_FIELD(wc
, pkt_mark
);
944 WC_MASK_FIELD(wc
, recirc_id
);
945 WC_MASK_FIELD(wc
, dp_hash
);
946 WC_MASK_FIELD(wc
, in_port
);
948 /* actset_output wildcarded. */
950 WC_MASK_FIELD(wc
, dl_dst
);
951 WC_MASK_FIELD(wc
, dl_src
);
952 WC_MASK_FIELD(wc
, dl_type
);
953 WC_MASK_FIELD(wc
, vlan_tci
);
955 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
956 WC_MASK_FIELD(wc
, nw_src
);
957 WC_MASK_FIELD(wc
, nw_dst
);
958 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
959 WC_MASK_FIELD(wc
, ipv6_src
);
960 WC_MASK_FIELD(wc
, ipv6_dst
);
961 WC_MASK_FIELD(wc
, ipv6_label
);
962 } else if (flow
->dl_type
== htons(ETH_TYPE_ARP
) ||
963 flow
->dl_type
== htons(ETH_TYPE_RARP
)) {
964 WC_MASK_FIELD(wc
, nw_src
);
965 WC_MASK_FIELD(wc
, nw_dst
);
966 WC_MASK_FIELD(wc
, nw_proto
);
967 WC_MASK_FIELD(wc
, arp_sha
);
968 WC_MASK_FIELD(wc
, arp_tha
);
970 } else if (eth_type_mpls(flow
->dl_type
)) {
971 for (int i
= 0; i
< FLOW_MAX_MPLS_LABELS
; i
++) {
972 WC_MASK_FIELD(wc
, mpls_lse
[i
]);
973 if (flow
->mpls_lse
[i
] & htonl(MPLS_BOS_MASK
)) {
979 return; /* Unknown ethertype. */
983 WC_MASK_FIELD(wc
, nw_frag
);
984 WC_MASK_FIELD(wc
, nw_tos
);
985 WC_MASK_FIELD(wc
, nw_ttl
);
986 WC_MASK_FIELD(wc
, nw_proto
);
988 /* No transport layer header in later fragments. */
989 if (!(flow
->nw_frag
& FLOW_NW_FRAG_LATER
) &&
990 (flow
->nw_proto
== IPPROTO_ICMP
||
991 flow
->nw_proto
== IPPROTO_ICMPV6
||
992 flow
->nw_proto
== IPPROTO_TCP
||
993 flow
->nw_proto
== IPPROTO_UDP
||
994 flow
->nw_proto
== IPPROTO_SCTP
||
995 flow
->nw_proto
== IPPROTO_IGMP
)) {
996 WC_MASK_FIELD(wc
, tp_src
);
997 WC_MASK_FIELD(wc
, tp_dst
);
999 if (flow
->nw_proto
== IPPROTO_TCP
) {
1000 WC_MASK_FIELD(wc
, tcp_flags
);
1001 } else if (flow
->nw_proto
== IPPROTO_ICMPV6
) {
1002 WC_MASK_FIELD(wc
, arp_sha
);
1003 WC_MASK_FIELD(wc
, arp_tha
);
1004 WC_MASK_FIELD(wc
, nd_target
);
1005 } else if (flow
->nw_proto
== IPPROTO_IGMP
) {
1006 WC_MASK_FIELD(wc
, igmp_group_ip4
);
1011 /* Return a map of possible fields for a packet of the same type as 'flow'.
1012 * Including extra bits in the returned mask is not wrong, it is just less
1015 * This is a less precise version of flow_wildcards_init_for_packet() above. */
1017 flow_wc_map(const struct flow
*flow
)
1019 /* Update this function whenever struct flow changes. */
1020 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 31);
1022 uint64_t map
= (flow
->tunnel
.ip_dst
) ? MINIFLOW_MAP(tunnel
) : 0;
1024 /* Metadata fields that can appear on packet input. */
1025 map
|= MINIFLOW_MAP(skb_priority
) | MINIFLOW_MAP(pkt_mark
)
1026 | MINIFLOW_MAP(recirc_id
) | MINIFLOW_MAP(dp_hash
)
1027 | MINIFLOW_MAP(in_port
)
1028 | MINIFLOW_MAP(dl_dst
) | MINIFLOW_MAP(dl_src
)
1029 | MINIFLOW_MAP(dl_type
) | MINIFLOW_MAP(vlan_tci
);
1031 /* Ethertype-dependent fields. */
1032 if (OVS_LIKELY(flow
->dl_type
== htons(ETH_TYPE_IP
))) {
1033 map
|= MINIFLOW_MAP(nw_src
) | MINIFLOW_MAP(nw_dst
)
1034 | MINIFLOW_MAP(nw_proto
) | MINIFLOW_MAP(nw_frag
)
1035 | MINIFLOW_MAP(nw_tos
) | MINIFLOW_MAP(nw_ttl
);
1036 if (OVS_UNLIKELY(flow
->nw_proto
== IPPROTO_IGMP
)) {
1037 map
|= MINIFLOW_MAP(igmp_group_ip4
);
1039 map
|= MINIFLOW_MAP(tcp_flags
)
1040 | MINIFLOW_MAP(tp_src
) | MINIFLOW_MAP(tp_dst
);
1042 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1043 map
|= MINIFLOW_MAP(ipv6_src
) | MINIFLOW_MAP(ipv6_dst
)
1044 | MINIFLOW_MAP(ipv6_label
)
1045 | MINIFLOW_MAP(nw_proto
) | MINIFLOW_MAP(nw_frag
)
1046 | MINIFLOW_MAP(nw_tos
) | MINIFLOW_MAP(nw_ttl
);
1047 if (OVS_UNLIKELY(flow
->nw_proto
== IPPROTO_ICMPV6
)) {
1048 map
|= MINIFLOW_MAP(nd_target
)
1049 | MINIFLOW_MAP(arp_sha
) | MINIFLOW_MAP(arp_tha
);
1051 map
|= MINIFLOW_MAP(tcp_flags
)
1052 | MINIFLOW_MAP(tp_src
) | MINIFLOW_MAP(tp_dst
);
1054 } else if (eth_type_mpls(flow
->dl_type
)) {
1055 map
|= MINIFLOW_MAP(mpls_lse
);
1056 } else if (flow
->dl_type
== htons(ETH_TYPE_ARP
) ||
1057 flow
->dl_type
== htons(ETH_TYPE_RARP
)) {
1058 map
|= MINIFLOW_MAP(nw_src
) | MINIFLOW_MAP(nw_dst
)
1059 | MINIFLOW_MAP(nw_proto
)
1060 | MINIFLOW_MAP(arp_sha
) | MINIFLOW_MAP(arp_tha
);
1066 /* Clear the metadata and register wildcard masks. They are not packet
1069 flow_wildcards_clear_non_packet_fields(struct flow_wildcards
*wc
)
1071 /* Update this function whenever struct flow changes. */
1072 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 31);
1074 memset(&wc
->masks
.metadata
, 0, sizeof wc
->masks
.metadata
);
1075 memset(&wc
->masks
.regs
, 0, sizeof wc
->masks
.regs
);
1076 wc
->masks
.actset_output
= 0;
1077 wc
->masks
.conj_id
= 0;
1080 /* Returns true if 'wc' matches every packet, false if 'wc' fixes any bits or
1083 flow_wildcards_is_catchall(const struct flow_wildcards
*wc
)
1085 const uint64_t *wc_u64
= (const uint64_t *) &wc
->masks
;
1088 for (i
= 0; i
< FLOW_U64S
; i
++) {
1096 /* Sets 'dst' as the bitwise AND of wildcards in 'src1' and 'src2'.
1097 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded
1098 * in 'src1' or 'src2' or both. */
1100 flow_wildcards_and(struct flow_wildcards
*dst
,
1101 const struct flow_wildcards
*src1
,
1102 const struct flow_wildcards
*src2
)
1104 uint64_t *dst_u64
= (uint64_t *) &dst
->masks
;
1105 const uint64_t *src1_u64
= (const uint64_t *) &src1
->masks
;
1106 const uint64_t *src2_u64
= (const uint64_t *) &src2
->masks
;
1109 for (i
= 0; i
< FLOW_U64S
; i
++) {
1110 dst_u64
[i
] = src1_u64
[i
] & src2_u64
[i
];
1114 /* Sets 'dst' as the bitwise OR of wildcards in 'src1' and 'src2'. That
1115 * is, a bit or a field is wildcarded in 'dst' if it is neither
1116 * wildcarded in 'src1' nor 'src2'. */
1118 flow_wildcards_or(struct flow_wildcards
*dst
,
1119 const struct flow_wildcards
*src1
,
1120 const struct flow_wildcards
*src2
)
1122 uint64_t *dst_u64
= (uint64_t *) &dst
->masks
;
1123 const uint64_t *src1_u64
= (const uint64_t *) &src1
->masks
;
1124 const uint64_t *src2_u64
= (const uint64_t *) &src2
->masks
;
1127 for (i
= 0; i
< FLOW_U64S
; i
++) {
1128 dst_u64
[i
] = src1_u64
[i
] | src2_u64
[i
];
1132 /* Returns a hash of the wildcards in 'wc'. */
1134 flow_wildcards_hash(const struct flow_wildcards
*wc
, uint32_t basis
)
1136 return flow_hash(&wc
->masks
, basis
);
1139 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
1142 flow_wildcards_equal(const struct flow_wildcards
*a
,
1143 const struct flow_wildcards
*b
)
1145 return flow_equal(&a
->masks
, &b
->masks
);
1148 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
1149 * 'b', false otherwise. */
1151 flow_wildcards_has_extra(const struct flow_wildcards
*a
,
1152 const struct flow_wildcards
*b
)
1154 const uint64_t *a_u64
= (const uint64_t *) &a
->masks
;
1155 const uint64_t *b_u64
= (const uint64_t *) &b
->masks
;
1158 for (i
= 0; i
< FLOW_U64S
; i
++) {
1159 if ((a_u64
[i
] & b_u64
[i
]) != b_u64
[i
]) {
1166 /* Returns true if 'a' and 'b' are equal, except that 0-bits (wildcarded bits)
1167 * in 'wc' do not need to be equal in 'a' and 'b'. */
1169 flow_equal_except(const struct flow
*a
, const struct flow
*b
,
1170 const struct flow_wildcards
*wc
)
1172 const uint64_t *a_u64
= (const uint64_t *) a
;
1173 const uint64_t *b_u64
= (const uint64_t *) b
;
1174 const uint64_t *wc_u64
= (const uint64_t *) &wc
->masks
;
1177 for (i
= 0; i
< FLOW_U64S
; i
++) {
1178 if ((a_u64
[i
] ^ b_u64
[i
]) & wc_u64
[i
]) {
1185 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1186 * (A 0-bit indicates a wildcard bit.) */
1188 flow_wildcards_set_reg_mask(struct flow_wildcards
*wc
, int idx
, uint32_t mask
)
1190 wc
->masks
.regs
[idx
] = mask
;
1193 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
1194 * (A 0-bit indicates a wildcard bit.) */
1196 flow_wildcards_set_xreg_mask(struct flow_wildcards
*wc
, int idx
, uint64_t mask
)
1198 flow_set_xreg(&wc
->masks
, idx
, mask
);
1201 /* Calculates the 5-tuple hash from the given miniflow.
1202 * This returns the same value as flow_hash_5tuple for the corresponding
1205 miniflow_hash_5tuple(const struct miniflow
*flow
, uint32_t basis
)
1207 uint32_t hash
= basis
;
1210 ovs_be16 dl_type
= MINIFLOW_GET_BE16(flow
, dl_type
);
1212 hash
= hash_add(hash
, MINIFLOW_GET_U8(flow
, nw_proto
));
1214 /* Separate loops for better optimization. */
1215 if (dl_type
== htons(ETH_TYPE_IPV6
)) {
1216 uint64_t map
= MINIFLOW_MAP(ipv6_src
) | MINIFLOW_MAP(ipv6_dst
);
1219 MINIFLOW_FOR_EACH_IN_MAP(value
, flow
, map
) {
1220 hash
= hash_add64(hash
, value
);
1223 hash
= hash_add(hash
, MINIFLOW_GET_U32(flow
, nw_src
));
1224 hash
= hash_add(hash
, MINIFLOW_GET_U32(flow
, nw_dst
));
1226 /* Add both ports at once. */
1227 hash
= hash_add(hash
, MINIFLOW_GET_U32(flow
, tp_src
));
1228 hash
= hash_finish(hash
, 42); /* Arbitrary number. */
1233 BUILD_ASSERT_DECL(offsetof(struct flow
, tp_src
) + 2
1234 == offsetof(struct flow
, tp_dst
) &&
1235 offsetof(struct flow
, tp_src
) / 4
1236 == offsetof(struct flow
, tp_dst
) / 4);
1237 BUILD_ASSERT_DECL(offsetof(struct flow
, ipv6_src
) + 16
1238 == offsetof(struct flow
, ipv6_dst
));
1240 /* Calculates the 5-tuple hash from the given flow. */
1242 flow_hash_5tuple(const struct flow
*flow
, uint32_t basis
)
1244 uint32_t hash
= basis
;
1247 hash
= hash_add(hash
, flow
->nw_proto
);
1249 if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1250 const uint64_t *flow_u64
= (const uint64_t *)flow
;
1251 int ofs
= offsetof(struct flow
, ipv6_src
) / 8;
1252 int end
= ofs
+ 2 * sizeof flow
->ipv6_src
/ 8;
1254 for (;ofs
< end
; ofs
++) {
1255 hash
= hash_add64(hash
, flow_u64
[ofs
]);
1258 hash
= hash_add(hash
, (OVS_FORCE
uint32_t) flow
->nw_src
);
1259 hash
= hash_add(hash
, (OVS_FORCE
uint32_t) flow
->nw_dst
);
1261 /* Add both ports at once. */
1262 hash
= hash_add(hash
,
1263 ((const uint32_t *)flow
)[offsetof(struct flow
, tp_src
)
1264 / sizeof(uint32_t)]);
1265 hash
= hash_finish(hash
, 42); /* Arbitrary number. */
1270 /* Hashes 'flow' based on its L2 through L4 protocol information. */
1272 flow_hash_symmetric_l4(const struct flow
*flow
, uint32_t basis
)
1277 struct in6_addr ipv6_addr
;
1282 uint8_t eth_addr
[ETH_ADDR_LEN
];
1288 memset(&fields
, 0, sizeof fields
);
1289 for (i
= 0; i
< ETH_ADDR_LEN
; i
++) {
1290 fields
.eth_addr
[i
] = flow
->dl_src
[i
] ^ flow
->dl_dst
[i
];
1292 fields
.vlan_tci
= flow
->vlan_tci
& htons(VLAN_VID_MASK
);
1293 fields
.eth_type
= flow
->dl_type
;
1295 /* UDP source and destination port are not taken into account because they
1296 * will not necessarily be symmetric in a bidirectional flow. */
1297 if (fields
.eth_type
== htons(ETH_TYPE_IP
)) {
1298 fields
.ipv4_addr
= flow
->nw_src
^ flow
->nw_dst
;
1299 fields
.ip_proto
= flow
->nw_proto
;
1300 if (fields
.ip_proto
== IPPROTO_TCP
|| fields
.ip_proto
== IPPROTO_SCTP
) {
1301 fields
.tp_port
= flow
->tp_src
^ flow
->tp_dst
;
1303 } else if (fields
.eth_type
== htons(ETH_TYPE_IPV6
)) {
1304 const uint8_t *a
= &flow
->ipv6_src
.s6_addr
[0];
1305 const uint8_t *b
= &flow
->ipv6_dst
.s6_addr
[0];
1306 uint8_t *ipv6_addr
= &fields
.ipv6_addr
.s6_addr
[0];
1308 for (i
=0; i
<16; i
++) {
1309 ipv6_addr
[i
] = a
[i
] ^ b
[i
];
1311 fields
.ip_proto
= flow
->nw_proto
;
1312 if (fields
.ip_proto
== IPPROTO_TCP
|| fields
.ip_proto
== IPPROTO_SCTP
) {
1313 fields
.tp_port
= flow
->tp_src
^ flow
->tp_dst
;
1316 return jhash_bytes(&fields
, sizeof fields
, basis
);
1319 /* Initialize a flow with random fields that matter for nx_hash_fields. */
1321 flow_random_hash_fields(struct flow
*flow
)
1323 uint16_t rnd
= random_uint16();
1325 /* Initialize to all zeros. */
1326 memset(flow
, 0, sizeof *flow
);
1328 eth_addr_random(flow
->dl_src
);
1329 eth_addr_random(flow
->dl_dst
);
1331 flow
->vlan_tci
= (OVS_FORCE ovs_be16
) (random_uint16() & VLAN_VID_MASK
);
1333 /* Make most of the random flows IPv4, some IPv6, and rest random. */
1334 flow
->dl_type
= rnd
< 0x8000 ? htons(ETH_TYPE_IP
) :
1335 rnd
< 0xc000 ? htons(ETH_TYPE_IPV6
) : (OVS_FORCE ovs_be16
)rnd
;
1337 if (dl_type_is_ip_any(flow
->dl_type
)) {
1338 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
1339 flow
->nw_src
= (OVS_FORCE ovs_be32
)random_uint32();
1340 flow
->nw_dst
= (OVS_FORCE ovs_be32
)random_uint32();
1342 random_bytes(&flow
->ipv6_src
, sizeof flow
->ipv6_src
);
1343 random_bytes(&flow
->ipv6_dst
, sizeof flow
->ipv6_dst
);
1345 /* Make most of IP flows TCP, some UDP or SCTP, and rest random. */
1346 rnd
= random_uint16();
1347 flow
->nw_proto
= rnd
< 0x8000 ? IPPROTO_TCP
:
1348 rnd
< 0xc000 ? IPPROTO_UDP
:
1349 rnd
< 0xd000 ? IPPROTO_SCTP
: (uint8_t)rnd
;
1350 if (flow
->nw_proto
== IPPROTO_TCP
||
1351 flow
->nw_proto
== IPPROTO_UDP
||
1352 flow
->nw_proto
== IPPROTO_SCTP
) {
1353 flow
->tp_src
= (OVS_FORCE ovs_be16
)random_uint16();
1354 flow
->tp_dst
= (OVS_FORCE ovs_be16
)random_uint16();
1359 /* Masks the fields in 'wc' that are used by the flow hash 'fields'. */
1361 flow_mask_hash_fields(const struct flow
*flow
, struct flow_wildcards
*wc
,
1362 enum nx_hash_fields fields
)
1365 case NX_HASH_FIELDS_ETH_SRC
:
1366 memset(&wc
->masks
.dl_src
, 0xff, sizeof wc
->masks
.dl_src
);
1369 case NX_HASH_FIELDS_SYMMETRIC_L4
:
1370 memset(&wc
->masks
.dl_src
, 0xff, sizeof wc
->masks
.dl_src
);
1371 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
1372 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
1373 memset(&wc
->masks
.nw_src
, 0xff, sizeof wc
->masks
.nw_src
);
1374 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
1375 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1376 memset(&wc
->masks
.ipv6_src
, 0xff, sizeof wc
->masks
.ipv6_src
);
1377 memset(&wc
->masks
.ipv6_dst
, 0xff, sizeof wc
->masks
.ipv6_dst
);
1379 if (is_ip_any(flow
)) {
1380 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
1381 flow_unwildcard_tp_ports(flow
, wc
);
1383 wc
->masks
.vlan_tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
1391 /* Hashes the portions of 'flow' designated by 'fields'. */
1393 flow_hash_fields(const struct flow
*flow
, enum nx_hash_fields fields
,
1398 case NX_HASH_FIELDS_ETH_SRC
:
1399 return jhash_bytes(flow
->dl_src
, sizeof flow
->dl_src
, basis
);
1401 case NX_HASH_FIELDS_SYMMETRIC_L4
:
1402 return flow_hash_symmetric_l4(flow
, basis
);
1408 /* Returns a string representation of 'fields'. */
1410 flow_hash_fields_to_str(enum nx_hash_fields fields
)
1413 case NX_HASH_FIELDS_ETH_SRC
: return "eth_src";
1414 case NX_HASH_FIELDS_SYMMETRIC_L4
: return "symmetric_l4";
1415 default: return "<unknown>";
1419 /* Returns true if the value of 'fields' is supported. Otherwise false. */
1421 flow_hash_fields_valid(enum nx_hash_fields fields
)
1423 return fields
== NX_HASH_FIELDS_ETH_SRC
1424 || fields
== NX_HASH_FIELDS_SYMMETRIC_L4
;
1427 /* Returns a hash value for the bits of 'flow' that are active based on
1428 * 'wc', given 'basis'. */
1430 flow_hash_in_wildcards(const struct flow
*flow
,
1431 const struct flow_wildcards
*wc
, uint32_t basis
)
1433 const uint64_t *wc_u64
= (const uint64_t *) &wc
->masks
;
1434 const uint64_t *flow_u64
= (const uint64_t *) flow
;
1439 for (i
= 0; i
< FLOW_U64S
; i
++) {
1440 hash
= hash_add64(hash
, flow_u64
[i
] & wc_u64
[i
]);
1442 return hash_finish(hash
, 8 * FLOW_U64S
);
1445 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1446 * OpenFlow 1.0 "dl_vlan" value:
1448 * - If it is in the range 0...4095, 'flow->vlan_tci' is set to match
1449 * that VLAN. Any existing PCP match is unchanged (it becomes 0 if
1450 * 'flow' previously matched packets without a VLAN header).
1452 * - If it is OFP_VLAN_NONE, 'flow->vlan_tci' is set to match a packet
1453 * without a VLAN tag.
1455 * - Other values of 'vid' should not be used. */
1457 flow_set_dl_vlan(struct flow
*flow
, ovs_be16 vid
)
1459 if (vid
== htons(OFP10_VLAN_NONE
)) {
1460 flow
->vlan_tci
= htons(0);
1462 vid
&= htons(VLAN_VID_MASK
);
1463 flow
->vlan_tci
&= ~htons(VLAN_VID_MASK
);
1464 flow
->vlan_tci
|= htons(VLAN_CFI
) | vid
;
1468 /* Sets the VLAN VID that 'flow' matches to 'vid', which is interpreted as an
1469 * OpenFlow 1.2 "vlan_vid" value, that is, the low 13 bits of 'vlan_tci' (VID
1472 flow_set_vlan_vid(struct flow
*flow
, ovs_be16 vid
)
1474 ovs_be16 mask
= htons(VLAN_VID_MASK
| VLAN_CFI
);
1475 flow
->vlan_tci
&= ~mask
;
1476 flow
->vlan_tci
|= vid
& mask
;
1479 /* Sets the VLAN PCP that 'flow' matches to 'pcp', which should be in the
1482 * This function has no effect on the VLAN ID that 'flow' matches.
1484 * After calling this function, 'flow' will not match packets without a VLAN
1487 flow_set_vlan_pcp(struct flow
*flow
, uint8_t pcp
)
1490 flow
->vlan_tci
&= ~htons(VLAN_PCP_MASK
);
1491 flow
->vlan_tci
|= htons((pcp
<< VLAN_PCP_SHIFT
) | VLAN_CFI
);
1494 /* Returns the number of MPLS LSEs present in 'flow'
1496 * Returns 0 if the 'dl_type' of 'flow' is not an MPLS ethernet type.
1497 * Otherwise traverses 'flow''s MPLS label stack stopping at the
1498 * first entry that has the BoS bit set. If no such entry exists then
1499 * the maximum number of LSEs that can be stored in 'flow' is returned.
1502 flow_count_mpls_labels(const struct flow
*flow
, struct flow_wildcards
*wc
)
1504 /* dl_type is always masked. */
1505 if (eth_type_mpls(flow
->dl_type
)) {
1510 for (i
= 0; i
< FLOW_MAX_MPLS_LABELS
; i
++) {
1512 wc
->masks
.mpls_lse
[i
] |= htonl(MPLS_BOS_MASK
);
1514 if (flow
->mpls_lse
[i
] & htonl(MPLS_BOS_MASK
)) {
1517 if (flow
->mpls_lse
[i
]) {
1527 /* Returns the number consecutive of MPLS LSEs, starting at the
1528 * innermost LSE, that are common in 'a' and 'b'.
1530 * 'an' must be flow_count_mpls_labels(a).
1531 * 'bn' must be flow_count_mpls_labels(b).
1534 flow_count_common_mpls_labels(const struct flow
*a
, int an
,
1535 const struct flow
*b
, int bn
,
1536 struct flow_wildcards
*wc
)
1538 int min_n
= MIN(an
, bn
);
1543 int a_last
= an
- 1;
1544 int b_last
= bn
- 1;
1547 for (i
= 0; i
< min_n
; i
++) {
1549 wc
->masks
.mpls_lse
[a_last
- i
] = OVS_BE32_MAX
;
1550 wc
->masks
.mpls_lse
[b_last
- i
] = OVS_BE32_MAX
;
1552 if (a
->mpls_lse
[a_last
- i
] != b
->mpls_lse
[b_last
- i
]) {
1563 /* Adds a new outermost MPLS label to 'flow' and changes 'flow''s Ethernet type
1564 * to 'mpls_eth_type', which must be an MPLS Ethertype.
1566 * If the new label is the first MPLS label in 'flow', it is generated as;
1568 * - label: 2, if 'flow' is IPv6, otherwise 0.
1570 * - TTL: IPv4 or IPv6 TTL, if present and nonzero, otherwise 64.
1572 * - TC: IPv4 or IPv6 TOS, if present, otherwise 0.
1576 * If the new label is the second or later label MPLS label in 'flow', it is
1579 * - label: Copied from outer label.
1581 * - TTL: Copied from outer label.
1583 * - TC: Copied from outer label.
1587 * 'n' must be flow_count_mpls_labels(flow). 'n' must be less than
1588 * FLOW_MAX_MPLS_LABELS (because otherwise flow->mpls_lse[] would overflow).
1591 flow_push_mpls(struct flow
*flow
, int n
, ovs_be16 mpls_eth_type
,
1592 struct flow_wildcards
*wc
)
1594 ovs_assert(eth_type_mpls(mpls_eth_type
));
1595 ovs_assert(n
< FLOW_MAX_MPLS_LABELS
);
1601 memset(&wc
->masks
.mpls_lse
, 0xff, sizeof *wc
->masks
.mpls_lse
* n
);
1603 for (i
= n
; i
>= 1; i
--) {
1604 flow
->mpls_lse
[i
] = flow
->mpls_lse
[i
- 1];
1606 flow
->mpls_lse
[0] = (flow
->mpls_lse
[1] & htonl(~MPLS_BOS_MASK
));
1608 int label
= 0; /* IPv4 Explicit Null. */
1612 if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1616 if (is_ip_any(flow
)) {
1617 tc
= (flow
->nw_tos
& IP_DSCP_MASK
) >> 2;
1619 wc
->masks
.nw_tos
|= IP_DSCP_MASK
;
1620 wc
->masks
.nw_ttl
= 0xff;
1628 flow
->mpls_lse
[0] = set_mpls_lse_values(ttl
, tc
, 1, htonl(label
));
1630 /* Clear all L3 and L4 fields and dp_hash. */
1631 BUILD_ASSERT(FLOW_WC_SEQ
== 31);
1632 memset((char *) flow
+ FLOW_SEGMENT_2_ENDS_AT
, 0,
1633 sizeof(struct flow
) - FLOW_SEGMENT_2_ENDS_AT
);
1636 flow
->dl_type
= mpls_eth_type
;
1639 /* Tries to remove the outermost MPLS label from 'flow'. Returns true if
1640 * successful, false otherwise. On success, sets 'flow''s Ethernet type to
1643 * 'n' must be flow_count_mpls_labels(flow). */
1645 flow_pop_mpls(struct flow
*flow
, int n
, ovs_be16 eth_type
,
1646 struct flow_wildcards
*wc
)
1651 /* Nothing to pop. */
1653 } else if (n
== FLOW_MAX_MPLS_LABELS
) {
1655 wc
->masks
.mpls_lse
[n
- 1] |= htonl(MPLS_BOS_MASK
);
1657 if (!(flow
->mpls_lse
[n
- 1] & htonl(MPLS_BOS_MASK
))) {
1658 /* Can't pop because don't know what to fill in mpls_lse[n - 1]. */
1664 memset(&wc
->masks
.mpls_lse
[1], 0xff,
1665 sizeof *wc
->masks
.mpls_lse
* (n
- 1));
1667 for (i
= 1; i
< n
; i
++) {
1668 flow
->mpls_lse
[i
- 1] = flow
->mpls_lse
[i
];
1670 flow
->mpls_lse
[n
- 1] = 0;
1671 flow
->dl_type
= eth_type
;
1675 /* Sets the MPLS Label that 'flow' matches to 'label', which is interpreted
1676 * as an OpenFlow 1.1 "mpls_label" value. */
1678 flow_set_mpls_label(struct flow
*flow
, int idx
, ovs_be32 label
)
1680 set_mpls_lse_label(&flow
->mpls_lse
[idx
], label
);
1683 /* Sets the MPLS TTL that 'flow' matches to 'ttl', which should be in the
1686 flow_set_mpls_ttl(struct flow
*flow
, int idx
, uint8_t ttl
)
1688 set_mpls_lse_ttl(&flow
->mpls_lse
[idx
], ttl
);
1691 /* Sets the MPLS TC that 'flow' matches to 'tc', which should be in the
1694 flow_set_mpls_tc(struct flow
*flow
, int idx
, uint8_t tc
)
1696 set_mpls_lse_tc(&flow
->mpls_lse
[idx
], tc
);
1699 /* Sets the MPLS BOS bit that 'flow' matches to which should be 0 or 1. */
1701 flow_set_mpls_bos(struct flow
*flow
, int idx
, uint8_t bos
)
1703 set_mpls_lse_bos(&flow
->mpls_lse
[idx
], bos
);
1706 /* Sets the entire MPLS LSE. */
1708 flow_set_mpls_lse(struct flow
*flow
, int idx
, ovs_be32 lse
)
1710 flow
->mpls_lse
[idx
] = lse
;
1714 flow_compose_l4(struct dp_packet
*p
, const struct flow
*flow
)
1718 if (!(flow
->nw_frag
& FLOW_NW_FRAG_ANY
)
1719 || !(flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
1720 if (flow
->nw_proto
== IPPROTO_TCP
) {
1721 struct tcp_header
*tcp
;
1723 l4_len
= sizeof *tcp
;
1724 tcp
= dp_packet_put_zeros(p
, l4_len
);
1725 tcp
->tcp_src
= flow
->tp_src
;
1726 tcp
->tcp_dst
= flow
->tp_dst
;
1727 tcp
->tcp_ctl
= TCP_CTL(ntohs(flow
->tcp_flags
), 5);
1728 } else if (flow
->nw_proto
== IPPROTO_UDP
) {
1729 struct udp_header
*udp
;
1731 l4_len
= sizeof *udp
;
1732 udp
= dp_packet_put_zeros(p
, l4_len
);
1733 udp
->udp_src
= flow
->tp_src
;
1734 udp
->udp_dst
= flow
->tp_dst
;
1735 } else if (flow
->nw_proto
== IPPROTO_SCTP
) {
1736 struct sctp_header
*sctp
;
1738 l4_len
= sizeof *sctp
;
1739 sctp
= dp_packet_put_zeros(p
, l4_len
);
1740 sctp
->sctp_src
= flow
->tp_src
;
1741 sctp
->sctp_dst
= flow
->tp_dst
;
1742 } else if (flow
->nw_proto
== IPPROTO_ICMP
) {
1743 struct icmp_header
*icmp
;
1745 l4_len
= sizeof *icmp
;
1746 icmp
= dp_packet_put_zeros(p
, l4_len
);
1747 icmp
->icmp_type
= ntohs(flow
->tp_src
);
1748 icmp
->icmp_code
= ntohs(flow
->tp_dst
);
1749 icmp
->icmp_csum
= csum(icmp
, ICMP_HEADER_LEN
);
1750 } else if (flow
->nw_proto
== IPPROTO_IGMP
) {
1751 struct igmp_header
*igmp
;
1753 l4_len
= sizeof *igmp
;
1754 igmp
= dp_packet_put_zeros(p
, l4_len
);
1755 igmp
->igmp_type
= ntohs(flow
->tp_src
);
1756 igmp
->igmp_code
= ntohs(flow
->tp_dst
);
1757 put_16aligned_be32(&igmp
->group
, flow
->igmp_group_ip4
);
1758 igmp
->igmp_csum
= csum(igmp
, IGMP_HEADER_LEN
);
1759 } else if (flow
->nw_proto
== IPPROTO_ICMPV6
) {
1760 struct icmp6_hdr
*icmp
;
1762 l4_len
= sizeof *icmp
;
1763 icmp
= dp_packet_put_zeros(p
, l4_len
);
1764 icmp
->icmp6_type
= ntohs(flow
->tp_src
);
1765 icmp
->icmp6_code
= ntohs(flow
->tp_dst
);
1767 if (icmp
->icmp6_code
== 0 &&
1768 (icmp
->icmp6_type
== ND_NEIGHBOR_SOLICIT
||
1769 icmp
->icmp6_type
== ND_NEIGHBOR_ADVERT
)) {
1770 struct in6_addr
*nd_target
;
1771 struct nd_opt_hdr
*nd_opt
;
1773 l4_len
+= sizeof *nd_target
;
1774 nd_target
= dp_packet_put_zeros(p
, sizeof *nd_target
);
1775 *nd_target
= flow
->nd_target
;
1777 if (!eth_addr_is_zero(flow
->arp_sha
)) {
1779 nd_opt
= dp_packet_put_zeros(p
, 8);
1780 nd_opt
->nd_opt_len
= 1;
1781 nd_opt
->nd_opt_type
= ND_OPT_SOURCE_LINKADDR
;
1782 memcpy(nd_opt
+ 1, flow
->arp_sha
, ETH_ADDR_LEN
);
1784 if (!eth_addr_is_zero(flow
->arp_tha
)) {
1786 nd_opt
= dp_packet_put_zeros(p
, 8);
1787 nd_opt
->nd_opt_len
= 1;
1788 nd_opt
->nd_opt_type
= ND_OPT_TARGET_LINKADDR
;
1789 memcpy(nd_opt
+ 1, flow
->arp_tha
, ETH_ADDR_LEN
);
1792 icmp
->icmp6_cksum
= (OVS_FORCE
uint16_t)
1793 csum(icmp
, (char *)dp_packet_tail(p
) - (char *)icmp
);
1799 /* Puts into 'b' a packet that flow_extract() would parse as having the given
1802 * (This is useful only for testing, obviously, and the packet isn't really
1803 * valid. It hasn't got some checksums filled in, for one, and lots of fields
1804 * are just zeroed.) */
1806 flow_compose(struct dp_packet
*p
, const struct flow
*flow
)
1810 /* eth_compose() sets l3 pointer and makes sure it is 32-bit aligned. */
1811 eth_compose(p
, flow
->dl_dst
, flow
->dl_src
, ntohs(flow
->dl_type
), 0);
1812 if (flow
->dl_type
== htons(FLOW_DL_TYPE_NONE
)) {
1813 struct eth_header
*eth
= dp_packet_l2(p
);
1814 eth
->eth_type
= htons(dp_packet_size(p
));
1818 if (flow
->vlan_tci
& htons(VLAN_CFI
)) {
1819 eth_push_vlan(p
, htons(ETH_TYPE_VLAN
), flow
->vlan_tci
);
1822 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
1823 struct ip_header
*ip
;
1825 ip
= dp_packet_put_zeros(p
, sizeof *ip
);
1826 ip
->ip_ihl_ver
= IP_IHL_VER(5, 4);
1827 ip
->ip_tos
= flow
->nw_tos
;
1828 ip
->ip_ttl
= flow
->nw_ttl
;
1829 ip
->ip_proto
= flow
->nw_proto
;
1830 put_16aligned_be32(&ip
->ip_src
, flow
->nw_src
);
1831 put_16aligned_be32(&ip
->ip_dst
, flow
->nw_dst
);
1833 if (flow
->nw_frag
& FLOW_NW_FRAG_ANY
) {
1834 ip
->ip_frag_off
|= htons(IP_MORE_FRAGMENTS
);
1835 if (flow
->nw_frag
& FLOW_NW_FRAG_LATER
) {
1836 ip
->ip_frag_off
|= htons(100);
1840 dp_packet_set_l4(p
, dp_packet_tail(p
));
1842 l4_len
= flow_compose_l4(p
, flow
);
1844 ip
= dp_packet_l3(p
);
1845 ip
->ip_tot_len
= htons(p
->l4_ofs
- p
->l3_ofs
+ l4_len
);
1846 ip
->ip_csum
= csum(ip
, sizeof *ip
);
1847 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1848 struct ovs_16aligned_ip6_hdr
*nh
;
1850 nh
= dp_packet_put_zeros(p
, sizeof *nh
);
1851 put_16aligned_be32(&nh
->ip6_flow
, htonl(6 << 28) |
1852 htonl(flow
->nw_tos
<< 20) | flow
->ipv6_label
);
1853 nh
->ip6_hlim
= flow
->nw_ttl
;
1854 nh
->ip6_nxt
= flow
->nw_proto
;
1856 memcpy(&nh
->ip6_src
, &flow
->ipv6_src
, sizeof(nh
->ip6_src
));
1857 memcpy(&nh
->ip6_dst
, &flow
->ipv6_dst
, sizeof(nh
->ip6_dst
));
1859 dp_packet_set_l4(p
, dp_packet_tail(p
));
1861 l4_len
= flow_compose_l4(p
, flow
);
1863 nh
= dp_packet_l3(p
);
1864 nh
->ip6_plen
= htons(l4_len
);
1865 } else if (flow
->dl_type
== htons(ETH_TYPE_ARP
) ||
1866 flow
->dl_type
== htons(ETH_TYPE_RARP
)) {
1867 struct arp_eth_header
*arp
;
1869 arp
= dp_packet_put_zeros(p
, sizeof *arp
);
1870 dp_packet_set_l3(p
, arp
);
1871 arp
->ar_hrd
= htons(1);
1872 arp
->ar_pro
= htons(ETH_TYPE_IP
);
1873 arp
->ar_hln
= ETH_ADDR_LEN
;
1875 arp
->ar_op
= htons(flow
->nw_proto
);
1877 if (flow
->nw_proto
== ARP_OP_REQUEST
||
1878 flow
->nw_proto
== ARP_OP_REPLY
) {
1879 put_16aligned_be32(&arp
->ar_spa
, flow
->nw_src
);
1880 put_16aligned_be32(&arp
->ar_tpa
, flow
->nw_dst
);
1881 memcpy(arp
->ar_sha
, flow
->arp_sha
, ETH_ADDR_LEN
);
1882 memcpy(arp
->ar_tha
, flow
->arp_tha
, ETH_ADDR_LEN
);
1886 if (eth_type_mpls(flow
->dl_type
)) {
1889 p
->l2_5_ofs
= p
->l3_ofs
;
1890 for (n
= 1; n
< FLOW_MAX_MPLS_LABELS
; n
++) {
1891 if (flow
->mpls_lse
[n
- 1] & htonl(MPLS_BOS_MASK
)) {
1896 push_mpls(p
, flow
->dl_type
, flow
->mpls_lse
[--n
]);
1901 /* Compressed flow. */
1904 miniflow_n_values(const struct miniflow
*flow
)
1906 return count_1bits(flow
->map
);
1910 miniflow_alloc_values(struct miniflow
*flow
, int n
)
1912 int size
= MINIFLOW_VALUES_SIZE(n
);
1914 if (size
<= sizeof flow
->inline_values
) {
1915 flow
->values_inline
= true;
1916 return flow
->inline_values
;
1918 COVERAGE_INC(miniflow_malloc
);
1919 flow
->values_inline
= false;
1920 flow
->offline_values
= xmalloc(size
);
1921 return flow
->offline_values
;
1925 /* Completes an initialization of 'dst' as a miniflow copy of 'src' begun by
1926 * the caller. The caller must have already initialized 'dst->map' properly
1927 * to indicate the significant uint64_t elements of 'src'. 'n' must be the
1928 * number of 1-bits in 'dst->map'.
1930 * Normally the significant elements are the ones that are non-zero. However,
1931 * when a miniflow is initialized from a (mini)mask, the values can be zeroes,
1932 * so that the flow and mask always have the same maps.
1934 * This function initializes values (either inline if possible or with
1935 * malloc() otherwise) and copies the uint64_t elements of 'src' indicated by
1936 * 'dst->map' into it. */
1938 miniflow_init__(struct miniflow
*dst
, const struct flow
*src
, int n
)
1940 const uint64_t *src_u64
= (const uint64_t *) src
;
1941 uint64_t *dst_u64
= miniflow_alloc_values(dst
, n
);
1944 MAP_FOR_EACH_INDEX(idx
, dst
->map
) {
1945 *dst_u64
++ = src_u64
[idx
];
1949 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1950 * with miniflow_destroy().
1951 * Always allocates offline storage. */
1953 miniflow_init(struct miniflow
*dst
, const struct flow
*src
)
1955 const uint64_t *src_u64
= (const uint64_t *) src
;
1959 /* Initialize dst->map, counting the number of nonzero elements. */
1963 for (i
= 0; i
< FLOW_U64S
; i
++) {
1965 dst
->map
|= UINT64_C(1) << i
;
1970 miniflow_init__(dst
, src
, n
);
1973 /* Initializes 'dst' as a copy of 'src', using 'mask->map' as 'dst''s map. The
1974 * caller must eventually free 'dst' with miniflow_destroy(). */
1976 miniflow_init_with_minimask(struct miniflow
*dst
, const struct flow
*src
,
1977 const struct minimask
*mask
)
1979 dst
->map
= mask
->masks
.map
;
1980 miniflow_init__(dst
, src
, miniflow_n_values(dst
));
1983 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
1984 * with miniflow_destroy(). */
1986 miniflow_clone(struct miniflow
*dst
, const struct miniflow
*src
)
1988 int size
= MINIFLOW_VALUES_SIZE(miniflow_n_values(src
));
1991 dst
->map
= src
->map
;
1992 if (size
<= sizeof dst
->inline_values
) {
1993 dst
->values_inline
= true;
1994 values
= dst
->inline_values
;
1996 dst
->values_inline
= false;
1997 COVERAGE_INC(miniflow_malloc
);
1998 dst
->offline_values
= xmalloc(size
);
1999 values
= dst
->offline_values
;
2001 memcpy(values
, miniflow_get_values(src
), size
);
2004 /* Initializes 'dst' as a copy of 'src'. The caller must have allocated
2005 * 'dst' to have inline space all data in 'src'. */
2007 miniflow_clone_inline(struct miniflow
*dst
, const struct miniflow
*src
,
2010 dst
->values_inline
= true;
2011 dst
->map
= src
->map
;
2012 memcpy(dst
->inline_values
, miniflow_get_values(src
),
2013 MINIFLOW_VALUES_SIZE(n_values
));
2016 /* Initializes 'dst' with the data in 'src', destroying 'src'.
2017 * The caller must eventually free 'dst' with miniflow_destroy().
2018 * 'dst' must be regularly sized miniflow, but 'src' can have
2019 * storage for more than the default MINI_N_INLINE inline
2022 miniflow_move(struct miniflow
*dst
, struct miniflow
*src
)
2024 int size
= MINIFLOW_VALUES_SIZE(miniflow_n_values(src
));
2026 dst
->map
= src
->map
;
2027 if (size
<= sizeof dst
->inline_values
) {
2028 dst
->values_inline
= true;
2029 memcpy(dst
->inline_values
, miniflow_get_values(src
), size
);
2030 miniflow_destroy(src
);
2031 } else if (src
->values_inline
) {
2032 dst
->values_inline
= false;
2033 COVERAGE_INC(miniflow_malloc
);
2034 dst
->offline_values
= xmalloc(size
);
2035 memcpy(dst
->offline_values
, src
->inline_values
, size
);
2037 dst
->values_inline
= false;
2038 dst
->offline_values
= src
->offline_values
;
2042 /* Frees any memory owned by 'flow'. Does not free the storage in which 'flow'
2043 * itself resides; the caller is responsible for that. */
2045 miniflow_destroy(struct miniflow
*flow
)
2047 if (!flow
->values_inline
) {
2048 free(flow
->offline_values
);
2052 /* Initializes 'dst' as a copy of 'src'. */
2054 miniflow_expand(const struct miniflow
*src
, struct flow
*dst
)
2056 memset(dst
, 0, sizeof *dst
);
2057 flow_union_with_miniflow(dst
, src
);
2060 /* Returns true if 'a' and 'b' are the equal miniflow, false otherwise. */
2062 miniflow_equal(const struct miniflow
*a
, const struct miniflow
*b
)
2064 const uint64_t *ap
= miniflow_get_values(a
);
2065 const uint64_t *bp
= miniflow_get_values(b
);
2067 if (OVS_LIKELY(a
->map
== b
->map
)) {
2068 int count
= miniflow_n_values(a
);
2070 return !memcmp(ap
, bp
, count
* sizeof *ap
);
2074 for (map
= a
->map
| b
->map
; map
; map
= zero_rightmost_1bit(map
)) {
2075 uint64_t bit
= rightmost_1bit(map
);
2077 if ((a
->map
& bit
? *ap
++ : 0) != (b
->map
& bit
? *bp
++ : 0)) {
2086 /* Returns false if 'a' and 'b' differ at the places where there are 1-bits
2087 * in 'mask', true otherwise. */
2089 miniflow_equal_in_minimask(const struct miniflow
*a
, const struct miniflow
*b
,
2090 const struct minimask
*mask
)
2092 const uint64_t *p
= miniflow_get_values(&mask
->masks
);
2095 MAP_FOR_EACH_INDEX(idx
, mask
->masks
.map
) {
2096 if ((miniflow_get(a
, idx
) ^ miniflow_get(b
, idx
)) & *p
++) {
2104 /* Returns true if 'a' and 'b' are equal at the places where there are 1-bits
2105 * in 'mask', false if they differ. */
2107 miniflow_equal_flow_in_minimask(const struct miniflow
*a
, const struct flow
*b
,
2108 const struct minimask
*mask
)
2110 const uint64_t *b_u64
= (const uint64_t *) b
;
2111 const uint64_t *p
= miniflow_get_values(&mask
->masks
);
2114 MAP_FOR_EACH_INDEX(idx
, mask
->masks
.map
) {
2115 if ((miniflow_get(a
, idx
) ^ b_u64
[idx
]) & *p
++) {
2124 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
2125 * with minimask_destroy(). */
2127 minimask_init(struct minimask
*mask
, const struct flow_wildcards
*wc
)
2129 miniflow_init(&mask
->masks
, &wc
->masks
);
2132 /* Initializes 'dst' as a copy of 'src'. The caller must eventually free 'dst'
2133 * with minimask_destroy(). */
2135 minimask_clone(struct minimask
*dst
, const struct minimask
*src
)
2137 miniflow_clone(&dst
->masks
, &src
->masks
);
2140 /* Initializes 'dst' with the data in 'src', destroying 'src'.
2141 * The caller must eventually free 'dst' with minimask_destroy(). */
2143 minimask_move(struct minimask
*dst
, struct minimask
*src
)
2145 miniflow_move(&dst
->masks
, &src
->masks
);
2148 /* Initializes 'dst_' as the bit-wise "and" of 'a_' and 'b_'.
2150 * The caller must provide room for FLOW_U64S "uint64_t"s in 'storage', for use
2151 * by 'dst_'. The caller must *not* free 'dst_' with minimask_destroy(). */
2153 minimask_combine(struct minimask
*dst_
,
2154 const struct minimask
*a_
, const struct minimask
*b_
,
2155 uint64_t storage
[FLOW_U64S
])
2157 struct miniflow
*dst
= &dst_
->masks
;
2158 uint64_t *dst_values
= storage
;
2159 const struct miniflow
*a
= &a_
->masks
;
2160 const struct miniflow
*b
= &b_
->masks
;
2163 dst
->values_inline
= false;
2164 dst
->offline_values
= storage
;
2167 MAP_FOR_EACH_INDEX(idx
, a
->map
& b
->map
) {
2168 /* Both 'a' and 'b' have non-zero data at 'idx'. */
2169 uint64_t mask
= miniflow_get__(a
, idx
) & miniflow_get__(b
, idx
);
2172 dst
->map
|= UINT64_C(1) << idx
;
2173 *dst_values
++ = mask
;
2178 /* Frees any memory owned by 'mask'. Does not free the storage in which 'mask'
2179 * itself resides; the caller is responsible for that. */
2181 minimask_destroy(struct minimask
*mask
)
2183 miniflow_destroy(&mask
->masks
);
2186 /* Initializes 'dst' as a copy of 'src'. */
2188 minimask_expand(const struct minimask
*mask
, struct flow_wildcards
*wc
)
2190 miniflow_expand(&mask
->masks
, &wc
->masks
);
2193 /* Returns true if 'a' and 'b' are the same flow mask, false otherwise.
2194 * Minimasks may not have zero data values, so for the minimasks to be the
2195 * same, they need to have the same map and the same data values. */
2197 minimask_equal(const struct minimask
*a
, const struct minimask
*b
)
2199 return a
->masks
.map
== b
->masks
.map
&&
2200 !memcmp(miniflow_get_values(&a
->masks
),
2201 miniflow_get_values(&b
->masks
),
2202 count_1bits(a
->masks
.map
) * sizeof *a
->masks
.inline_values
);
2205 /* Returns true if at least one bit matched by 'b' is wildcarded by 'a',
2206 * false otherwise. */
2208 minimask_has_extra(const struct minimask
*a
, const struct minimask
*b
)
2210 const uint64_t *ap
= miniflow_get_values(&a
->masks
);
2211 const uint64_t *bp
= miniflow_get_values(&b
->masks
);
2214 MAP_FOR_EACH_INDEX(idx
, b
->masks
.map
) {
2215 uint64_t b_u64
= *bp
++;
2217 /* 'b_u64' is non-zero, check if the data in 'a' is either zero
2218 * or misses some of the bits in 'b_u64'. */
2219 if (!(a
->masks
.map
& (UINT64_C(1) << idx
))
2220 || ((miniflow_values_get__(ap
, a
->masks
.map
, idx
) & b_u64
)
2222 return true; /* 'a' wildcards some bits 'b' doesn't. */