2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 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.
18 #include <arpa/inet.h>
23 #include <netinet/in.h>
24 #include <netinet/icmp6.h>
28 #include "byte-order.h"
31 #include "dynamic-string.h"
38 #include "unaligned.h"
42 VLOG_DEFINE_THIS_MODULE(odp_util
);
44 /* The interface between userspace and kernel uses an "OVS_*" prefix.
45 * Since this is fairly non-specific for the OVS userspace components,
46 * "ODP_*" (Open vSwitch Datapath) is used as the prefix for
47 * interactions with the datapath.
50 /* The set of characters that may separate one action or one key attribute
52 static const char *delimiters
= ", \t\r\n";
54 static int parse_odp_key_mask_attr(const char *, const struct simap
*port_names
,
55 struct ofpbuf
*, struct ofpbuf
*);
56 static void format_odp_key_attr(const struct nlattr
*a
,
57 const struct nlattr
*ma
,
58 const struct hmap
*portno_names
, struct ds
*ds
,
61 /* Returns one the following for the action with the given OVS_ACTION_ATTR_*
64 * - For an action whose argument has a fixed length, returned that
65 * nonnegative length in bytes.
67 * - For an action with a variable-length argument, returns -2.
69 * - For an invalid 'type', returns -1. */
71 odp_action_len(uint16_t type
)
73 if (type
> OVS_ACTION_ATTR_MAX
) {
77 switch ((enum ovs_action_attr
) type
) {
78 case OVS_ACTION_ATTR_OUTPUT
: return sizeof(uint32_t);
79 case OVS_ACTION_ATTR_TUNNEL_PUSH
: return -2;
80 case OVS_ACTION_ATTR_TUNNEL_POP
: return sizeof(uint32_t);
81 case OVS_ACTION_ATTR_USERSPACE
: return -2;
82 case OVS_ACTION_ATTR_PUSH_VLAN
: return sizeof(struct ovs_action_push_vlan
);
83 case OVS_ACTION_ATTR_POP_VLAN
: return 0;
84 case OVS_ACTION_ATTR_PUSH_MPLS
: return sizeof(struct ovs_action_push_mpls
);
85 case OVS_ACTION_ATTR_POP_MPLS
: return sizeof(ovs_be16
);
86 case OVS_ACTION_ATTR_RECIRC
: return sizeof(uint32_t);
87 case OVS_ACTION_ATTR_HASH
: return sizeof(struct ovs_action_hash
);
88 case OVS_ACTION_ATTR_SET
: return -2;
89 case OVS_ACTION_ATTR_SET_MASKED
: return -2;
90 case OVS_ACTION_ATTR_SAMPLE
: return -2;
92 case OVS_ACTION_ATTR_UNSPEC
:
93 case __OVS_ACTION_ATTR_MAX
:
100 /* Returns a string form of 'attr'. The return value is either a statically
101 * allocated constant string or the 'bufsize'-byte buffer 'namebuf'. 'bufsize'
102 * should be at least OVS_KEY_ATTR_BUFSIZE. */
103 enum { OVS_KEY_ATTR_BUFSIZE
= 3 + INT_STRLEN(unsigned int) + 1 };
105 ovs_key_attr_to_string(enum ovs_key_attr attr
, char *namebuf
, size_t bufsize
)
108 case OVS_KEY_ATTR_UNSPEC
: return "unspec";
109 case OVS_KEY_ATTR_ENCAP
: return "encap";
110 case OVS_KEY_ATTR_PRIORITY
: return "skb_priority";
111 case OVS_KEY_ATTR_SKB_MARK
: return "skb_mark";
112 case OVS_KEY_ATTR_TUNNEL
: return "tunnel";
113 case OVS_KEY_ATTR_IN_PORT
: return "in_port";
114 case OVS_KEY_ATTR_ETHERNET
: return "eth";
115 case OVS_KEY_ATTR_VLAN
: return "vlan";
116 case OVS_KEY_ATTR_ETHERTYPE
: return "eth_type";
117 case OVS_KEY_ATTR_IPV4
: return "ipv4";
118 case OVS_KEY_ATTR_IPV6
: return "ipv6";
119 case OVS_KEY_ATTR_TCP
: return "tcp";
120 case OVS_KEY_ATTR_TCP_FLAGS
: return "tcp_flags";
121 case OVS_KEY_ATTR_UDP
: return "udp";
122 case OVS_KEY_ATTR_SCTP
: return "sctp";
123 case OVS_KEY_ATTR_ICMP
: return "icmp";
124 case OVS_KEY_ATTR_ICMPV6
: return "icmpv6";
125 case OVS_KEY_ATTR_ARP
: return "arp";
126 case OVS_KEY_ATTR_ND
: return "nd";
127 case OVS_KEY_ATTR_MPLS
: return "mpls";
128 case OVS_KEY_ATTR_DP_HASH
: return "dp_hash";
129 case OVS_KEY_ATTR_RECIRC_ID
: return "recirc_id";
131 case __OVS_KEY_ATTR_MAX
:
133 snprintf(namebuf
, bufsize
, "key%u", (unsigned int) attr
);
139 format_generic_odp_action(struct ds
*ds
, const struct nlattr
*a
)
141 size_t len
= nl_attr_get_size(a
);
143 ds_put_format(ds
, "action%"PRId16
, nl_attr_type(a
));
145 const uint8_t *unspec
;
148 unspec
= nl_attr_get(a
);
149 for (i
= 0; i
< len
; i
++) {
150 ds_put_char(ds
, i
? ' ': '(');
151 ds_put_format(ds
, "%02x", unspec
[i
]);
153 ds_put_char(ds
, ')');
158 format_odp_sample_action(struct ds
*ds
, const struct nlattr
*attr
)
160 static const struct nl_policy ovs_sample_policy
[] = {
161 [OVS_SAMPLE_ATTR_PROBABILITY
] = { .type
= NL_A_U32
},
162 [OVS_SAMPLE_ATTR_ACTIONS
] = { .type
= NL_A_NESTED
}
164 struct nlattr
*a
[ARRAY_SIZE(ovs_sample_policy
)];
166 const struct nlattr
*nla_acts
;
169 ds_put_cstr(ds
, "sample");
171 if (!nl_parse_nested(attr
, ovs_sample_policy
, a
, ARRAY_SIZE(a
))) {
172 ds_put_cstr(ds
, "(error)");
176 percentage
= (100.0 * nl_attr_get_u32(a
[OVS_SAMPLE_ATTR_PROBABILITY
])) /
179 ds_put_format(ds
, "(sample=%.1f%%,", percentage
);
181 ds_put_cstr(ds
, "actions(");
182 nla_acts
= nl_attr_get(a
[OVS_SAMPLE_ATTR_ACTIONS
]);
183 len
= nl_attr_get_size(a
[OVS_SAMPLE_ATTR_ACTIONS
]);
184 format_odp_actions(ds
, nla_acts
, len
);
185 ds_put_format(ds
, "))");
189 slow_path_reason_to_string(uint32_t reason
)
191 switch ((enum slow_path_reason
) reason
) {
192 #define SPR(ENUM, STRING, EXPLANATION) case ENUM: return STRING;
201 slow_path_reason_to_explanation(enum slow_path_reason reason
)
204 #define SPR(ENUM, STRING, EXPLANATION) case ENUM: return EXPLANATION;
213 parse_flags(const char *s
, const char *(*bit_to_string
)(uint32_t),
214 uint32_t *res_flags
, uint32_t allowed
, uint32_t *res_mask
)
219 /* Parse masked flags in numeric format? */
220 if (res_mask
&& ovs_scan(s
, "%"SCNi32
"/%"SCNi32
"%n",
221 res_flags
, res_mask
, &n
) && n
> 0) {
222 if (*res_flags
& ~allowed
|| *res_mask
& ~allowed
) {
230 if (res_mask
&& (*s
== '+' || *s
== '-')) {
231 uint32_t flags
= 0, mask
= 0;
233 /* Parse masked flags. */
234 while (s
[n
] != ')') {
241 } else if (s
[n
] == '-') {
248 name_len
= strcspn(s
+ n
, "+-)");
250 for (bit
= 1; bit
; bit
<<= 1) {
251 const char *fname
= bit_to_string(bit
);
259 if (len
!= name_len
) {
262 if (!strncmp(s
+ n
, fname
, len
)) {
264 /* bit already set. */
267 if (!(bit
& allowed
)) {
279 return -EINVAL
; /* Unknown flag name */
289 /* Parse unmasked flags. If a flag is present, it is set, otherwise
291 while (s
[n
] != ')') {
292 unsigned long long int flags
;
296 if (ovs_scan(&s
[n
], "%lli%n", &flags
, &n0
)) {
297 if (flags
& ~allowed
) {
300 n
+= n0
+ (s
[n
+ n0
] == ',');
305 for (bit
= 1; bit
; bit
<<= 1) {
306 const char *name
= bit_to_string(bit
);
314 if (!strncmp(s
+ n
, name
, len
) &&
315 (s
[n
+ len
] == ',' || s
[n
+ len
] == ')')) {
316 if (!(bit
& allowed
)) {
320 n
+= len
+ (s
[n
+ len
] == ',');
332 *res_mask
= UINT32_MAX
;
338 format_odp_userspace_action(struct ds
*ds
, const struct nlattr
*attr
)
340 static const struct nl_policy ovs_userspace_policy
[] = {
341 [OVS_USERSPACE_ATTR_PID
] = { .type
= NL_A_U32
},
342 [OVS_USERSPACE_ATTR_USERDATA
] = { .type
= NL_A_UNSPEC
,
344 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT
] = { .type
= NL_A_U32
,
347 struct nlattr
*a
[ARRAY_SIZE(ovs_userspace_policy
)];
348 const struct nlattr
*userdata_attr
;
349 const struct nlattr
*tunnel_out_port_attr
;
351 if (!nl_parse_nested(attr
, ovs_userspace_policy
, a
, ARRAY_SIZE(a
))) {
352 ds_put_cstr(ds
, "userspace(error)");
356 ds_put_format(ds
, "userspace(pid=%"PRIu32
,
357 nl_attr_get_u32(a
[OVS_USERSPACE_ATTR_PID
]));
359 userdata_attr
= a
[OVS_USERSPACE_ATTR_USERDATA
];
362 const uint8_t *userdata
= nl_attr_get(userdata_attr
);
363 size_t userdata_len
= nl_attr_get_size(userdata_attr
);
364 bool userdata_unspec
= true;
365 union user_action_cookie cookie
;
367 if (userdata_len
>= sizeof cookie
.type
368 && userdata_len
<= sizeof cookie
) {
370 memset(&cookie
, 0, sizeof cookie
);
371 memcpy(&cookie
, userdata
, userdata_len
);
373 userdata_unspec
= false;
375 if (userdata_len
== sizeof cookie
.sflow
376 && cookie
.type
== USER_ACTION_COOKIE_SFLOW
) {
377 ds_put_format(ds
, ",sFlow("
378 "vid=%"PRIu16
",pcp=%"PRIu8
",output=%"PRIu32
")",
379 vlan_tci_to_vid(cookie
.sflow
.vlan_tci
),
380 vlan_tci_to_pcp(cookie
.sflow
.vlan_tci
),
381 cookie
.sflow
.output
);
382 } else if (userdata_len
== sizeof cookie
.slow_path
383 && cookie
.type
== USER_ACTION_COOKIE_SLOW_PATH
) {
384 ds_put_cstr(ds
, ",slow_path(");
385 format_flags(ds
, slow_path_reason_to_string
,
386 cookie
.slow_path
.reason
, ',');
387 ds_put_format(ds
, ")");
388 } else if (userdata_len
== sizeof cookie
.flow_sample
389 && cookie
.type
== USER_ACTION_COOKIE_FLOW_SAMPLE
) {
390 ds_put_format(ds
, ",flow_sample(probability=%"PRIu16
391 ",collector_set_id=%"PRIu32
392 ",obs_domain_id=%"PRIu32
393 ",obs_point_id=%"PRIu32
")",
394 cookie
.flow_sample
.probability
,
395 cookie
.flow_sample
.collector_set_id
,
396 cookie
.flow_sample
.obs_domain_id
,
397 cookie
.flow_sample
.obs_point_id
);
398 } else if (userdata_len
>= sizeof cookie
.ipfix
399 && cookie
.type
== USER_ACTION_COOKIE_IPFIX
) {
400 ds_put_format(ds
, ",ipfix(output_port=%"PRIu32
")",
401 cookie
.ipfix
.output_odp_port
);
403 userdata_unspec
= true;
407 if (userdata_unspec
) {
409 ds_put_format(ds
, ",userdata(");
410 for (i
= 0; i
< userdata_len
; i
++) {
411 ds_put_format(ds
, "%02x", userdata
[i
]);
413 ds_put_char(ds
, ')');
417 tunnel_out_port_attr
= a
[OVS_USERSPACE_ATTR_EGRESS_TUN_PORT
];
418 if (tunnel_out_port_attr
) {
419 ds_put_format(ds
, ",tunnel_out_port=%"PRIu32
,
420 nl_attr_get_u32(tunnel_out_port_attr
));
423 ds_put_char(ds
, ')');
427 format_vlan_tci(struct ds
*ds
, ovs_be16 tci
, ovs_be16 mask
, bool verbose
)
429 if (verbose
|| vlan_tci_to_vid(tci
) || vlan_tci_to_vid(mask
)) {
430 ds_put_format(ds
, "vid=%"PRIu16
, vlan_tci_to_vid(tci
));
431 if (vlan_tci_to_vid(mask
) != VLAN_VID_MASK
) { /* Partially masked. */
432 ds_put_format(ds
, "/0x%"PRIx16
, vlan_tci_to_vid(mask
));
434 ds_put_char(ds
, ',');
436 if (verbose
|| vlan_tci_to_pcp(tci
) || vlan_tci_to_pcp(mask
)) {
437 ds_put_format(ds
, "pcp=%d", vlan_tci_to_pcp(tci
));
438 if (vlan_tci_to_pcp(mask
) != (VLAN_PCP_MASK
>> VLAN_PCP_SHIFT
)) {
439 ds_put_format(ds
, "/0x%x", vlan_tci_to_pcp(mask
));
441 ds_put_char(ds
, ',');
443 if (!(tci
& htons(VLAN_CFI
))) {
444 ds_put_cstr(ds
, "cfi=0");
445 ds_put_char(ds
, ',');
451 format_mpls_lse(struct ds
*ds
, ovs_be32 mpls_lse
)
453 ds_put_format(ds
, "label=%"PRIu32
",tc=%d,ttl=%d,bos=%d",
454 mpls_lse_to_label(mpls_lse
),
455 mpls_lse_to_tc(mpls_lse
),
456 mpls_lse_to_ttl(mpls_lse
),
457 mpls_lse_to_bos(mpls_lse
));
461 format_mpls(struct ds
*ds
, const struct ovs_key_mpls
*mpls_key
,
462 const struct ovs_key_mpls
*mpls_mask
, int n
)
465 ovs_be32 key
= mpls_key
->mpls_lse
;
467 if (mpls_mask
== NULL
) {
468 format_mpls_lse(ds
, key
);
470 ovs_be32 mask
= mpls_mask
->mpls_lse
;
472 ds_put_format(ds
, "label=%"PRIu32
"/0x%x,tc=%d/%x,ttl=%d/0x%x,bos=%d/%x",
473 mpls_lse_to_label(key
), mpls_lse_to_label(mask
),
474 mpls_lse_to_tc(key
), mpls_lse_to_tc(mask
),
475 mpls_lse_to_ttl(key
), mpls_lse_to_ttl(mask
),
476 mpls_lse_to_bos(key
), mpls_lse_to_bos(mask
));
481 for (i
= 0; i
< n
; i
++) {
482 ds_put_format(ds
, "lse%d=%#"PRIx32
,
483 i
, ntohl(mpls_key
[i
].mpls_lse
));
485 ds_put_format(ds
, "/%#"PRIx32
, ntohl(mpls_mask
[i
].mpls_lse
));
487 ds_put_char(ds
, ',');
494 format_odp_recirc_action(struct ds
*ds
, uint32_t recirc_id
)
496 ds_put_format(ds
, "recirc(%"PRIu32
")", recirc_id
);
500 format_odp_hash_action(struct ds
*ds
, const struct ovs_action_hash
*hash_act
)
502 ds_put_format(ds
, "hash(");
504 if (hash_act
->hash_alg
== OVS_HASH_ALG_L4
) {
505 ds_put_format(ds
, "hash_l4(%"PRIu32
")", hash_act
->hash_basis
);
507 ds_put_format(ds
, "Unknown hash algorithm(%"PRIu32
")",
510 ds_put_format(ds
, ")");
514 format_odp_tnl_push_header(struct ds
*ds
, struct ovs_action_push_tnl
*data
)
516 const struct eth_header
*eth
;
517 const struct ip_header
*ip
;
520 eth
= (const struct eth_header
*)data
->header
;
523 ip
= (const struct ip_header
*)l3
;
526 ds_put_format(ds
, "header(size=%"PRIu8
",type=%"PRIu8
",eth(dst=",
527 data
->header_len
, data
->tnl_type
);
528 ds_put_format(ds
, ETH_ADDR_FMT
, ETH_ADDR_ARGS(eth
->eth_dst
));
529 ds_put_format(ds
, ",src=");
530 ds_put_format(ds
, ETH_ADDR_FMT
, ETH_ADDR_ARGS(eth
->eth_src
));
531 ds_put_format(ds
, ",dl_type=0x%04"PRIx16
"),", ntohs(eth
->eth_type
));
534 ds_put_format(ds
, "ipv4(src="IP_FMT
",dst="IP_FMT
",proto=%"PRIu8
535 ",tos=%#"PRIx8
",ttl=%"PRIu8
",frag=0x%"PRIx16
"),",
536 IP_ARGS(get_16aligned_be32(&ip
->ip_src
)),
537 IP_ARGS(get_16aligned_be32(&ip
->ip_dst
)),
538 ip
->ip_proto
, ip
->ip_tos
,
542 if (data
->tnl_type
== OVS_VPORT_TYPE_VXLAN
) {
543 const struct vxlanhdr
*vxh
;
544 const struct udp_header
*udp
;
547 udp
= (const struct udp_header
*) (ip
+ 1);
548 ds_put_format(ds
, "udp(src=%"PRIu16
",dst=%"PRIu16
"),",
549 ntohs(udp
->udp_src
), ntohs(udp
->udp_dst
));
552 vxh
= (const struct vxlanhdr
*) (udp
+ 1);
553 ds_put_format(ds
, "vxlan(flags=0x%"PRIx32
",vni=0x%"PRIx32
")",
554 ntohl(get_16aligned_be32(&vxh
->vx_flags
)),
555 ntohl(get_16aligned_be32(&vxh
->vx_vni
)));
556 } else if (data
->tnl_type
== OVS_VPORT_TYPE_GRE
) {
557 const struct gre_base_hdr
*greh
;
558 ovs_16aligned_be32
*options
;
561 l4
= ((uint8_t *)l3
+ sizeof(struct ip_header
));
562 greh
= (const struct gre_base_hdr
*) l4
;
564 ds_put_format(ds
, "gre((flags=0x%"PRIx16
",proto=0x%"PRIx16
")",
565 greh
->flags
, ntohs(greh
->protocol
));
566 options
= (ovs_16aligned_be32
*)(greh
+ 1);
567 if (greh
->flags
& htons(GRE_CSUM
)) {
568 ds_put_format(ds
, ",csum=0x%"PRIx32
, ntohl(get_16aligned_be32(options
)));
571 if (greh
->flags
& htons(GRE_KEY
)) {
572 ds_put_format(ds
, ",key=0x%"PRIx32
, ntohl(get_16aligned_be32(options
)));
575 if (greh
->flags
& htons(GRE_SEQ
)) {
576 ds_put_format(ds
, ",seq=0x%"PRIx32
, ntohl(get_16aligned_be32(options
)));
579 ds_put_format(ds
, ")");
581 ds_put_format(ds
, ")");
585 format_odp_tnl_push_action(struct ds
*ds
, const struct nlattr
*attr
)
587 struct ovs_action_push_tnl
*data
;
589 data
= (struct ovs_action_push_tnl
*) nl_attr_get(attr
);
591 ds_put_format(ds
, "tnl_push(tnl_port(%"PRIu32
"),", data
->tnl_port
);
592 format_odp_tnl_push_header(ds
, data
);
593 ds_put_format(ds
, ",out_port(%"PRIu32
"))", data
->out_port
);
597 format_odp_action(struct ds
*ds
, const struct nlattr
*a
)
600 enum ovs_action_attr type
= nl_attr_type(a
);
601 const struct ovs_action_push_vlan
*vlan
;
604 expected_len
= odp_action_len(nl_attr_type(a
));
605 if (expected_len
!= -2 && nl_attr_get_size(a
) != expected_len
) {
606 ds_put_format(ds
, "bad length %"PRIuSIZE
", expected %d for: ",
607 nl_attr_get_size(a
), expected_len
);
608 format_generic_odp_action(ds
, a
);
613 case OVS_ACTION_ATTR_OUTPUT
:
614 ds_put_format(ds
, "%"PRIu32
, nl_attr_get_u32(a
));
616 case OVS_ACTION_ATTR_TUNNEL_POP
:
617 ds_put_format(ds
, "tnl_pop(%"PRIu32
")", nl_attr_get_u32(a
));
619 case OVS_ACTION_ATTR_TUNNEL_PUSH
:
620 format_odp_tnl_push_action(ds
, a
);
622 case OVS_ACTION_ATTR_USERSPACE
:
623 format_odp_userspace_action(ds
, a
);
625 case OVS_ACTION_ATTR_RECIRC
:
626 format_odp_recirc_action(ds
, nl_attr_get_u32(a
));
628 case OVS_ACTION_ATTR_HASH
:
629 format_odp_hash_action(ds
, nl_attr_get(a
));
631 case OVS_ACTION_ATTR_SET_MASKED
:
633 size
= nl_attr_get_size(a
) / 2;
634 ds_put_cstr(ds
, "set(");
636 /* Masked set action not supported for tunnel key, which is bigger. */
637 if (size
<= sizeof(struct ovs_key_ipv6
)) {
638 struct nlattr attr
[1 + DIV_ROUND_UP(sizeof(struct ovs_key_ipv6
),
639 sizeof(struct nlattr
))];
640 struct nlattr mask
[1 + DIV_ROUND_UP(sizeof(struct ovs_key_ipv6
),
641 sizeof(struct nlattr
))];
643 mask
->nla_type
= attr
->nla_type
= nl_attr_type(a
);
644 mask
->nla_len
= attr
->nla_len
= NLA_HDRLEN
+ size
;
645 memcpy(attr
+ 1, (char *)(a
+ 1), size
);
646 memcpy(mask
+ 1, (char *)(a
+ 1) + size
, size
);
647 format_odp_key_attr(attr
, mask
, NULL
, ds
, false);
649 format_odp_key_attr(a
, NULL
, NULL
, ds
, false);
651 ds_put_cstr(ds
, ")");
653 case OVS_ACTION_ATTR_SET
:
654 ds_put_cstr(ds
, "set(");
655 format_odp_key_attr(nl_attr_get(a
), NULL
, NULL
, ds
, true);
656 ds_put_cstr(ds
, ")");
658 case OVS_ACTION_ATTR_PUSH_VLAN
:
659 vlan
= nl_attr_get(a
);
660 ds_put_cstr(ds
, "push_vlan(");
661 if (vlan
->vlan_tpid
!= htons(ETH_TYPE_VLAN
)) {
662 ds_put_format(ds
, "tpid=0x%04"PRIx16
",", ntohs(vlan
->vlan_tpid
));
664 format_vlan_tci(ds
, vlan
->vlan_tci
, OVS_BE16_MAX
, false);
665 ds_put_char(ds
, ')');
667 case OVS_ACTION_ATTR_POP_VLAN
:
668 ds_put_cstr(ds
, "pop_vlan");
670 case OVS_ACTION_ATTR_PUSH_MPLS
: {
671 const struct ovs_action_push_mpls
*mpls
= nl_attr_get(a
);
672 ds_put_cstr(ds
, "push_mpls(");
673 format_mpls_lse(ds
, mpls
->mpls_lse
);
674 ds_put_format(ds
, ",eth_type=0x%"PRIx16
")", ntohs(mpls
->mpls_ethertype
));
677 case OVS_ACTION_ATTR_POP_MPLS
: {
678 ovs_be16 ethertype
= nl_attr_get_be16(a
);
679 ds_put_format(ds
, "pop_mpls(eth_type=0x%"PRIx16
")", ntohs(ethertype
));
682 case OVS_ACTION_ATTR_SAMPLE
:
683 format_odp_sample_action(ds
, a
);
685 case OVS_ACTION_ATTR_UNSPEC
:
686 case __OVS_ACTION_ATTR_MAX
:
688 format_generic_odp_action(ds
, a
);
694 format_odp_actions(struct ds
*ds
, const struct nlattr
*actions
,
698 const struct nlattr
*a
;
701 NL_ATTR_FOR_EACH (a
, left
, actions
, actions_len
) {
703 ds_put_char(ds
, ',');
705 format_odp_action(ds
, a
);
710 if (left
== actions_len
) {
711 ds_put_cstr(ds
, "<empty>");
713 ds_put_format(ds
, ",***%u leftover bytes*** (", left
);
714 for (i
= 0; i
< left
; i
++) {
715 ds_put_format(ds
, "%02x", ((const uint8_t *) a
)[i
]);
717 ds_put_char(ds
, ')');
720 ds_put_cstr(ds
, "drop");
724 /* Separate out parse_odp_userspace_action() function. */
726 parse_odp_userspace_action(const char *s
, struct ofpbuf
*actions
)
729 union user_action_cookie cookie
;
731 odp_port_t tunnel_out_port
;
733 void *user_data
= NULL
;
734 size_t user_data_size
= 0;
736 if (!ovs_scan(s
, "userspace(pid=%"SCNi32
"%n", &pid
, &n
)) {
742 uint32_t probability
;
743 uint32_t collector_set_id
;
744 uint32_t obs_domain_id
;
745 uint32_t obs_point_id
;
748 if (ovs_scan(&s
[n
], ",sFlow(vid=%i,"
749 "pcp=%i,output=%"SCNi32
")%n",
750 &vid
, &pcp
, &output
, &n1
)) {
754 tci
= vid
| (pcp
<< VLAN_PCP_SHIFT
);
759 cookie
.type
= USER_ACTION_COOKIE_SFLOW
;
760 cookie
.sflow
.vlan_tci
= htons(tci
);
761 cookie
.sflow
.output
= output
;
763 user_data_size
= sizeof cookie
.sflow
;
764 } else if (ovs_scan(&s
[n
], ",slow_path(%n",
769 cookie
.type
= USER_ACTION_COOKIE_SLOW_PATH
;
770 cookie
.slow_path
.unused
= 0;
771 cookie
.slow_path
.reason
= 0;
773 res
= parse_flags(&s
[n
], slow_path_reason_to_string
,
774 &cookie
.slow_path
.reason
,
775 SLOW_PATH_REASON_MASK
, NULL
);
776 if (res
< 0 || s
[n
+ res
] != ')') {
782 user_data_size
= sizeof cookie
.slow_path
;
783 } else if (ovs_scan(&s
[n
], ",flow_sample(probability=%"SCNi32
","
784 "collector_set_id=%"SCNi32
","
785 "obs_domain_id=%"SCNi32
","
786 "obs_point_id=%"SCNi32
")%n",
787 &probability
, &collector_set_id
,
788 &obs_domain_id
, &obs_point_id
, &n1
)) {
791 cookie
.type
= USER_ACTION_COOKIE_FLOW_SAMPLE
;
792 cookie
.flow_sample
.probability
= probability
;
793 cookie
.flow_sample
.collector_set_id
= collector_set_id
;
794 cookie
.flow_sample
.obs_domain_id
= obs_domain_id
;
795 cookie
.flow_sample
.obs_point_id
= obs_point_id
;
797 user_data_size
= sizeof cookie
.flow_sample
;
798 } else if (ovs_scan(&s
[n
], ",ipfix(output_port=%"SCNi32
")%n",
801 cookie
.type
= USER_ACTION_COOKIE_IPFIX
;
802 cookie
.ipfix
.output_odp_port
= u32_to_odp(output
);
804 user_data_size
= sizeof cookie
.ipfix
;
805 } else if (ovs_scan(&s
[n
], ",userdata(%n",
810 ofpbuf_init(&buf
, 16);
811 end
= ofpbuf_put_hex(&buf
, &s
[n
], NULL
);
815 user_data
= ofpbuf_data(&buf
);
816 user_data_size
= ofpbuf_size(&buf
);
823 if (ovs_scan(&s
[n
], ",tunnel_out_port=%"SCNi32
")%n",
824 &tunnel_out_port
, &n1
)) {
825 odp_put_userspace_action(pid
, user_data
, user_data_size
, tunnel_out_port
, actions
);
827 } else if (s
[n
] == ')') {
828 odp_put_userspace_action(pid
, user_data
, user_data_size
, ODPP_NONE
, actions
);
837 ovs_parse_tnl_push(const char *s
, struct ovs_action_push_tnl
*data
)
839 struct eth_header
*eth
;
840 struct ip_header
*ip
;
841 struct udp_header
*udp
;
842 struct gre_base_hdr
*greh
;
843 uint16_t gre_proto
, dl_type
, udp_src
, udp_dst
;
845 uint32_t tnl_type
= 0, header_len
= 0;
849 if (!ovs_scan_len(s
, &n
, "tnl_push(tnl_port(%"SCNi32
"),", &data
->tnl_port
)) {
852 eth
= (struct eth_header
*) data
->header
;
853 l3
= (data
->header
+ sizeof *eth
);
854 l4
= ((uint8_t *) l3
+ sizeof (struct ip_header
));
855 ip
= (struct ip_header
*) l3
;
856 if (!ovs_scan_len(s
, &n
, "header(size=%"SCNi32
",type=%"SCNi32
","
857 "eth(dst="ETH_ADDR_SCAN_FMT
",",
860 ETH_ADDR_SCAN_ARGS(eth
->eth_dst
))) {
864 if (!ovs_scan_len(s
, &n
, "src="ETH_ADDR_SCAN_FMT
",",
865 ETH_ADDR_SCAN_ARGS(eth
->eth_src
))) {
868 if (!ovs_scan_len(s
, &n
, "dl_type=0x%"SCNx16
"),", &dl_type
)) {
871 eth
->eth_type
= htons(dl_type
);
874 if (!ovs_scan_len(s
, &n
, "ipv4(src="IP_SCAN_FMT
",dst="IP_SCAN_FMT
",proto=%"SCNi8
875 ",tos=%"SCNi8
",ttl=%"SCNi8
",frag=0x%"SCNx16
"),",
878 &ip
->ip_proto
, &ip
->ip_tos
,
879 &ip
->ip_ttl
, &ip
->ip_frag_off
)) {
882 put_16aligned_be32(&ip
->ip_src
, sip
);
883 put_16aligned_be32(&ip
->ip_dst
, dip
);
886 udp
= (struct udp_header
*) l4
;
887 greh
= (struct gre_base_hdr
*) l4
;
888 if (ovs_scan_len(s
, &n
, "udp(src=%"SCNi16
",dst=%"SCNi16
"),",
889 &udp_src
, &udp_dst
)) {
890 struct vxlanhdr
*vxh
;
891 uint32_t vx_flags
, vx_vni
;
893 udp
->udp_src
= htons(udp_src
);
894 udp
->udp_dst
= htons(udp_dst
);
898 vxh
= (struct vxlanhdr
*) (udp
+ 1);
899 if (!ovs_scan_len(s
, &n
, "vxlan(flags=0x%"SCNx32
",vni=0x%"SCNx32
"))",
900 &vx_flags
, &vx_vni
)) {
903 put_16aligned_be32(&vxh
->vx_flags
, htonl(vx_flags
));
904 put_16aligned_be32(&vxh
->vx_vni
, htonl(vx_vni
));
905 tnl_type
= OVS_VPORT_TYPE_VXLAN
;
906 header_len
= sizeof *eth
+ sizeof *ip
+
907 sizeof *udp
+ sizeof *vxh
;
908 } else if (ovs_scan_len(s
, &n
, "gre((flags=0x%"SCNx16
",proto=0x%"SCNx16
")",
909 &greh
->flags
, &gre_proto
)){
911 tnl_type
= OVS_VPORT_TYPE_GRE
;
912 greh
->protocol
= htons(gre_proto
);
913 ovs_16aligned_be32
*options
= (ovs_16aligned_be32
*) (greh
+ 1);
915 if (greh
->flags
& htons(GRE_CSUM
)) {
918 if (!ovs_scan_len(s
, &n
, ",csum=0x%"SCNx32
, &csum
)) {
921 put_16aligned_be32(options
, htonl(csum
));
924 if (greh
->flags
& htons(GRE_KEY
)) {
927 if (!ovs_scan_len(s
, &n
, ",key=0x%"SCNx32
, &key
)) {
931 put_16aligned_be32(options
, htonl(key
));
934 if (greh
->flags
& htons(GRE_SEQ
)) {
937 if (!ovs_scan_len(s
, &n
, ",seq=0x%"SCNx32
, &seq
)) {
940 put_16aligned_be32(options
, htonl(seq
));
944 if (!ovs_scan_len(s
, &n
, "))")) {
948 header_len
= sizeof *eth
+ sizeof *ip
+
949 ((uint8_t *) options
- (uint8_t *) greh
);
954 /* check tunnel meta data. */
955 if (data
->tnl_type
!= tnl_type
) {
958 if (data
->header_len
!= header_len
) {
963 if (!ovs_scan_len(s
, &n
, ",out_port(%"SCNi32
"))", &data
->out_port
)) {
971 parse_odp_action(const char *s
, const struct simap
*port_names
,
972 struct ofpbuf
*actions
)
978 if (ovs_scan(s
, "%"SCNi32
"%n", &port
, &n
)) {
979 nl_msg_put_u32(actions
, OVS_ACTION_ATTR_OUTPUT
, port
);
985 int len
= strcspn(s
, delimiters
);
986 struct simap_node
*node
;
988 node
= simap_find_len(port_names
, s
, len
);
990 nl_msg_put_u32(actions
, OVS_ACTION_ATTR_OUTPUT
, node
->data
);
999 if (ovs_scan(s
, "recirc(%"PRIu32
")%n", &recirc_id
, &n
)) {
1000 nl_msg_put_u32(actions
, OVS_ACTION_ATTR_RECIRC
, recirc_id
);
1005 if (!strncmp(s
, "userspace(", 10)) {
1006 return parse_odp_userspace_action(s
, actions
);
1009 if (!strncmp(s
, "set(", 4)) {
1012 struct nlattr mask
[128 / sizeof(struct nlattr
)];
1013 struct ofpbuf maskbuf
;
1014 struct nlattr
*nested
, *key
;
1017 /* 'mask' is big enough to hold any key. */
1018 ofpbuf_use_stack(&maskbuf
, mask
, sizeof mask
);
1020 start_ofs
= nl_msg_start_nested(actions
, OVS_ACTION_ATTR_SET
);
1021 retval
= parse_odp_key_mask_attr(s
+ 4, port_names
, actions
, &maskbuf
);
1025 if (s
[retval
+ 4] != ')') {
1029 nested
= ofpbuf_at_assert(actions
, start_ofs
, sizeof *nested
);
1032 size
= nl_attr_get_size(mask
);
1033 if (size
== nl_attr_get_size(key
)) {
1034 /* Change to masked set action if not fully masked. */
1035 if (!is_all_ones(mask
+ 1, size
)) {
1036 key
->nla_len
+= size
;
1037 ofpbuf_put(actions
, mask
+ 1, size
);
1038 /* 'actions' may have been reallocated by ofpbuf_put(). */
1039 nested
= ofpbuf_at_assert(actions
, start_ofs
, sizeof *nested
);
1040 nested
->nla_type
= OVS_ACTION_ATTR_SET_MASKED
;
1044 nl_msg_end_nested(actions
, start_ofs
);
1049 struct ovs_action_push_vlan push
;
1050 int tpid
= ETH_TYPE_VLAN
;
1055 if (ovs_scan(s
, "push_vlan(vid=%i,pcp=%i)%n", &vid
, &pcp
, &n
)
1056 || ovs_scan(s
, "push_vlan(vid=%i,pcp=%i,cfi=%i)%n",
1057 &vid
, &pcp
, &cfi
, &n
)
1058 || ovs_scan(s
, "push_vlan(tpid=%i,vid=%i,pcp=%i)%n",
1059 &tpid
, &vid
, &pcp
, &n
)
1060 || ovs_scan(s
, "push_vlan(tpid=%i,vid=%i,pcp=%i,cfi=%i)%n",
1061 &tpid
, &vid
, &pcp
, &cfi
, &n
)) {
1062 push
.vlan_tpid
= htons(tpid
);
1063 push
.vlan_tci
= htons((vid
<< VLAN_VID_SHIFT
)
1064 | (pcp
<< VLAN_PCP_SHIFT
)
1065 | (cfi
? VLAN_CFI
: 0));
1066 nl_msg_put_unspec(actions
, OVS_ACTION_ATTR_PUSH_VLAN
,
1067 &push
, sizeof push
);
1073 if (!strncmp(s
, "pop_vlan", 8)) {
1074 nl_msg_put_flag(actions
, OVS_ACTION_ATTR_POP_VLAN
);
1082 if (ovs_scan(s
, "sample(sample=%lf%%,actions(%n", &percentage
, &n
)
1083 && percentage
>= 0. && percentage
<= 100.0) {
1084 size_t sample_ofs
, actions_ofs
;
1087 probability
= floor(UINT32_MAX
* (percentage
/ 100.0) + .5);
1088 sample_ofs
= nl_msg_start_nested(actions
, OVS_ACTION_ATTR_SAMPLE
);
1089 nl_msg_put_u32(actions
, OVS_SAMPLE_ATTR_PROBABILITY
,
1090 (probability
<= 0 ? 0
1091 : probability
>= UINT32_MAX
? UINT32_MAX
1094 actions_ofs
= nl_msg_start_nested(actions
,
1095 OVS_SAMPLE_ATTR_ACTIONS
);
1099 n
+= strspn(s
+ n
, delimiters
);
1104 retval
= parse_odp_action(s
+ n
, port_names
, actions
);
1110 nl_msg_end_nested(actions
, actions_ofs
);
1111 nl_msg_end_nested(actions
, sample_ofs
);
1113 return s
[n
+ 1] == ')' ? n
+ 2 : -EINVAL
;
1121 if (ovs_scan(s
, "tnl_pop(%"SCNi32
")%n", &port
, &n
)) {
1122 nl_msg_put_u32(actions
, OVS_ACTION_ATTR_TUNNEL_POP
, port
);
1128 struct ovs_action_push_tnl data
;
1131 n
= ovs_parse_tnl_push(s
, &data
);
1133 odp_put_tnl_push_action(actions
, &data
);
1142 /* Parses the string representation of datapath actions, in the format output
1143 * by format_odp_action(). Returns 0 if successful, otherwise a positive errno
1144 * value. On success, the ODP actions are appended to 'actions' as a series of
1145 * Netlink attributes. On failure, no data is appended to 'actions'. Either
1146 * way, 'actions''s data might be reallocated. */
1148 odp_actions_from_string(const char *s
, const struct simap
*port_names
,
1149 struct ofpbuf
*actions
)
1153 if (!strcasecmp(s
, "drop")) {
1157 old_size
= ofpbuf_size(actions
);
1161 s
+= strspn(s
, delimiters
);
1166 retval
= parse_odp_action(s
, port_names
, actions
);
1167 if (retval
< 0 || !strchr(delimiters
, s
[retval
])) {
1168 ofpbuf_set_size(actions
, old_size
);
1177 /* Returns the correct length of the payload for a flow key attribute of the
1178 * specified 'type', -1 if 'type' is unknown, or -2 if the attribute's payload
1179 * is variable length. */
1181 odp_flow_key_attr_len(uint16_t type
)
1183 if (type
> OVS_KEY_ATTR_MAX
) {
1187 switch ((enum ovs_key_attr
) type
) {
1188 case OVS_KEY_ATTR_ENCAP
: return -2;
1189 case OVS_KEY_ATTR_PRIORITY
: return 4;
1190 case OVS_KEY_ATTR_SKB_MARK
: return 4;
1191 case OVS_KEY_ATTR_DP_HASH
: return 4;
1192 case OVS_KEY_ATTR_RECIRC_ID
: return 4;
1193 case OVS_KEY_ATTR_TUNNEL
: return -2;
1194 case OVS_KEY_ATTR_IN_PORT
: return 4;
1195 case OVS_KEY_ATTR_ETHERNET
: return sizeof(struct ovs_key_ethernet
);
1196 case OVS_KEY_ATTR_VLAN
: return sizeof(ovs_be16
);
1197 case OVS_KEY_ATTR_ETHERTYPE
: return 2;
1198 case OVS_KEY_ATTR_MPLS
: return -2;
1199 case OVS_KEY_ATTR_IPV4
: return sizeof(struct ovs_key_ipv4
);
1200 case OVS_KEY_ATTR_IPV6
: return sizeof(struct ovs_key_ipv6
);
1201 case OVS_KEY_ATTR_TCP
: return sizeof(struct ovs_key_tcp
);
1202 case OVS_KEY_ATTR_TCP_FLAGS
: return 2;
1203 case OVS_KEY_ATTR_UDP
: return sizeof(struct ovs_key_udp
);
1204 case OVS_KEY_ATTR_SCTP
: return sizeof(struct ovs_key_sctp
);
1205 case OVS_KEY_ATTR_ICMP
: return sizeof(struct ovs_key_icmp
);
1206 case OVS_KEY_ATTR_ICMPV6
: return sizeof(struct ovs_key_icmpv6
);
1207 case OVS_KEY_ATTR_ARP
: return sizeof(struct ovs_key_arp
);
1208 case OVS_KEY_ATTR_ND
: return sizeof(struct ovs_key_nd
);
1210 case OVS_KEY_ATTR_UNSPEC
:
1211 case __OVS_KEY_ATTR_MAX
:
1219 format_generic_odp_key(const struct nlattr
*a
, struct ds
*ds
)
1221 size_t len
= nl_attr_get_size(a
);
1223 const uint8_t *unspec
;
1226 unspec
= nl_attr_get(a
);
1227 for (i
= 0; i
< len
; i
++) {
1229 ds_put_char(ds
, ' ');
1231 ds_put_format(ds
, "%02x", unspec
[i
]);
1237 ovs_frag_type_to_string(enum ovs_frag_type type
)
1240 case OVS_FRAG_TYPE_NONE
:
1242 case OVS_FRAG_TYPE_FIRST
:
1244 case OVS_FRAG_TYPE_LATER
:
1246 case __OVS_FRAG_TYPE_MAX
:
1253 tunnel_key_attr_len(int type
)
1256 case OVS_TUNNEL_KEY_ATTR_ID
: return 8;
1257 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC
: return 4;
1258 case OVS_TUNNEL_KEY_ATTR_IPV4_DST
: return 4;
1259 case OVS_TUNNEL_KEY_ATTR_TOS
: return 1;
1260 case OVS_TUNNEL_KEY_ATTR_TTL
: return 1;
1261 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT
: return 0;
1262 case OVS_TUNNEL_KEY_ATTR_CSUM
: return 0;
1263 case OVS_TUNNEL_KEY_ATTR_TP_SRC
: return 2;
1264 case OVS_TUNNEL_KEY_ATTR_TP_DST
: return 2;
1265 case OVS_TUNNEL_KEY_ATTR_OAM
: return 0;
1266 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS
: return -2;
1267 case __OVS_TUNNEL_KEY_ATTR_MAX
:
1273 #define GENEVE_OPT(class, type) ((OVS_FORCE uint32_t)(class) << 8 | (type))
1275 parse_geneve_opts(const struct nlattr
*attr
)
1277 int opts_len
= nl_attr_get_size(attr
);
1278 const struct geneve_opt
*opt
= nl_attr_get(attr
);
1280 while (opts_len
> 0) {
1283 if (opts_len
< sizeof(*opt
)) {
1287 len
= sizeof(*opt
) + opt
->length
* 4;
1288 if (len
> opts_len
) {
1292 switch (GENEVE_OPT(opt
->opt_class
, opt
->type
)) {
1294 if (opt
->type
& GENEVE_CRIT_OPT_TYPE
) {
1299 opt
= opt
+ len
/ sizeof(*opt
);
1306 enum odp_key_fitness
1307 odp_tun_key_from_attr(const struct nlattr
*attr
, struct flow_tnl
*tun
)
1310 const struct nlattr
*a
;
1312 bool unknown
= false;
1314 NL_NESTED_FOR_EACH(a
, left
, attr
) {
1315 uint16_t type
= nl_attr_type(a
);
1316 size_t len
= nl_attr_get_size(a
);
1317 int expected_len
= tunnel_key_attr_len(type
);
1319 if (len
!= expected_len
&& expected_len
>= 0) {
1320 return ODP_FIT_ERROR
;
1324 case OVS_TUNNEL_KEY_ATTR_ID
:
1325 tun
->tun_id
= nl_attr_get_be64(a
);
1326 tun
->flags
|= FLOW_TNL_F_KEY
;
1328 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC
:
1329 tun
->ip_src
= nl_attr_get_be32(a
);
1331 case OVS_TUNNEL_KEY_ATTR_IPV4_DST
:
1332 tun
->ip_dst
= nl_attr_get_be32(a
);
1334 case OVS_TUNNEL_KEY_ATTR_TOS
:
1335 tun
->ip_tos
= nl_attr_get_u8(a
);
1337 case OVS_TUNNEL_KEY_ATTR_TTL
:
1338 tun
->ip_ttl
= nl_attr_get_u8(a
);
1341 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT
:
1342 tun
->flags
|= FLOW_TNL_F_DONT_FRAGMENT
;
1344 case OVS_TUNNEL_KEY_ATTR_CSUM
:
1345 tun
->flags
|= FLOW_TNL_F_CSUM
;
1347 case OVS_TUNNEL_KEY_ATTR_TP_SRC
:
1348 tun
->tp_src
= nl_attr_get_be16(a
);
1350 case OVS_TUNNEL_KEY_ATTR_TP_DST
:
1351 tun
->tp_dst
= nl_attr_get_be16(a
);
1353 case OVS_TUNNEL_KEY_ATTR_OAM
:
1354 tun
->flags
|= FLOW_TNL_F_OAM
;
1356 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS
: {
1357 if (parse_geneve_opts(a
)) {
1358 return ODP_FIT_ERROR
;
1360 /* It is necessary to reproduce options exactly (including order)
1361 * so it's easiest to just echo them back. */
1366 /* Allow this to show up as unexpected, if there are unknown
1367 * tunnel attribute, eventually resulting in ODP_FIT_TOO_MUCH. */
1374 return ODP_FIT_ERROR
;
1377 return ODP_FIT_TOO_MUCH
;
1379 return ODP_FIT_PERFECT
;
1383 tun_key_to_attr(struct ofpbuf
*a
, const struct flow_tnl
*tun_key
)
1387 tun_key_ofs
= nl_msg_start_nested(a
, OVS_KEY_ATTR_TUNNEL
);
1389 /* tun_id != 0 without FLOW_TNL_F_KEY is valid if tun_key is a mask. */
1390 if (tun_key
->tun_id
|| tun_key
->flags
& FLOW_TNL_F_KEY
) {
1391 nl_msg_put_be64(a
, OVS_TUNNEL_KEY_ATTR_ID
, tun_key
->tun_id
);
1393 if (tun_key
->ip_src
) {
1394 nl_msg_put_be32(a
, OVS_TUNNEL_KEY_ATTR_IPV4_SRC
, tun_key
->ip_src
);
1396 if (tun_key
->ip_dst
) {
1397 nl_msg_put_be32(a
, OVS_TUNNEL_KEY_ATTR_IPV4_DST
, tun_key
->ip_dst
);
1399 if (tun_key
->ip_tos
) {
1400 nl_msg_put_u8(a
, OVS_TUNNEL_KEY_ATTR_TOS
, tun_key
->ip_tos
);
1402 nl_msg_put_u8(a
, OVS_TUNNEL_KEY_ATTR_TTL
, tun_key
->ip_ttl
);
1403 if (tun_key
->flags
& FLOW_TNL_F_DONT_FRAGMENT
) {
1404 nl_msg_put_flag(a
, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT
);
1406 if (tun_key
->flags
& FLOW_TNL_F_CSUM
) {
1407 nl_msg_put_flag(a
, OVS_TUNNEL_KEY_ATTR_CSUM
);
1409 if (tun_key
->tp_src
) {
1410 nl_msg_put_be16(a
, OVS_TUNNEL_KEY_ATTR_TP_SRC
, tun_key
->tp_src
);
1412 if (tun_key
->tp_dst
) {
1413 nl_msg_put_be16(a
, OVS_TUNNEL_KEY_ATTR_TP_DST
, tun_key
->tp_dst
);
1415 if (tun_key
->flags
& FLOW_TNL_F_OAM
) {
1416 nl_msg_put_flag(a
, OVS_TUNNEL_KEY_ATTR_OAM
);
1419 nl_msg_end_nested(a
, tun_key_ofs
);
1423 odp_mask_attr_is_wildcard(const struct nlattr
*ma
)
1425 return is_all_zeros(nl_attr_get(ma
), nl_attr_get_size(ma
));
1429 odp_mask_is_exact(enum ovs_key_attr attr
, const void *mask
, size_t size
)
1431 if (attr
== OVS_KEY_ATTR_TCP_FLAGS
) {
1432 return TCP_FLAGS(*(ovs_be16
*)mask
) == TCP_FLAGS(OVS_BE16_MAX
);
1434 if (attr
== OVS_KEY_ATTR_IPV6
) {
1435 const struct ovs_key_ipv6
*ipv6_mask
= mask
;
1438 ((ipv6_mask
->ipv6_label
& htonl(IPV6_LABEL_MASK
))
1439 == htonl(IPV6_LABEL_MASK
))
1440 && ipv6_mask
->ipv6_proto
== UINT8_MAX
1441 && ipv6_mask
->ipv6_tclass
== UINT8_MAX
1442 && ipv6_mask
->ipv6_hlimit
== UINT8_MAX
1443 && ipv6_mask
->ipv6_frag
== UINT8_MAX
1444 && ipv6_mask_is_exact((const struct in6_addr
*)ipv6_mask
->ipv6_src
)
1445 && ipv6_mask_is_exact((const struct in6_addr
*)ipv6_mask
->ipv6_dst
);
1447 if (attr
== OVS_KEY_ATTR_TUNNEL
) {
1448 const struct flow_tnl
*tun_mask
= mask
;
1450 return tun_mask
->flags
== FLOW_TNL_F_MASK
1451 && tun_mask
->tun_id
== OVS_BE64_MAX
1452 && tun_mask
->ip_src
== OVS_BE32_MAX
1453 && tun_mask
->ip_dst
== OVS_BE32_MAX
1454 && tun_mask
->ip_tos
== UINT8_MAX
1455 && tun_mask
->ip_ttl
== UINT8_MAX
1456 && tun_mask
->tp_src
== OVS_BE16_MAX
1457 && tun_mask
->tp_dst
== OVS_BE16_MAX
;
1460 if (attr
== OVS_KEY_ATTR_ARP
) {
1461 /* ARP key has padding, ignore it. */
1462 BUILD_ASSERT_DECL(sizeof(struct ovs_key_arp
) == 24);
1463 BUILD_ASSERT_DECL(offsetof(struct ovs_key_arp
, arp_tha
) == 10 + 6);
1464 size
= offsetof(struct ovs_key_arp
, arp_tha
) + ETH_ADDR_LEN
;
1465 ovs_assert(((uint16_t *)mask
)[size
/2] == 0);
1468 return is_all_ones(mask
, size
);
1472 odp_mask_attr_is_exact(const struct nlattr
*ma
)
1474 struct flow_tnl tun_mask
;
1475 enum ovs_key_attr attr
= nl_attr_type(ma
);
1479 if (attr
== OVS_KEY_ATTR_TUNNEL
) {
1480 memset(&tun_mask
, 0, sizeof tun_mask
);
1481 odp_tun_key_from_attr(ma
, &tun_mask
);
1483 size
= sizeof tun_mask
;
1485 mask
= nl_attr_get(ma
);
1486 size
= nl_attr_get_size(ma
);
1489 return odp_mask_is_exact(attr
, mask
, size
);
1493 odp_portno_names_set(struct hmap
*portno_names
, odp_port_t port_no
,
1496 struct odp_portno_names
*odp_portno_names
;
1498 odp_portno_names
= xmalloc(sizeof *odp_portno_names
);
1499 odp_portno_names
->port_no
= port_no
;
1500 odp_portno_names
->name
= xstrdup(port_name
);
1501 hmap_insert(portno_names
, &odp_portno_names
->hmap_node
,
1502 hash_odp_port(port_no
));
1506 odp_portno_names_get(const struct hmap
*portno_names
, odp_port_t port_no
)
1508 struct odp_portno_names
*odp_portno_names
;
1510 HMAP_FOR_EACH_IN_BUCKET (odp_portno_names
, hmap_node
,
1511 hash_odp_port(port_no
), portno_names
) {
1512 if (odp_portno_names
->port_no
== port_no
) {
1513 return odp_portno_names
->name
;
1520 odp_portno_names_destroy(struct hmap
*portno_names
)
1522 struct odp_portno_names
*odp_portno_names
, *odp_portno_names_next
;
1523 HMAP_FOR_EACH_SAFE (odp_portno_names
, odp_portno_names_next
,
1524 hmap_node
, portno_names
) {
1525 hmap_remove(portno_names
, &odp_portno_names
->hmap_node
);
1526 free(odp_portno_names
->name
);
1527 free(odp_portno_names
);
1531 /* Format helpers. */
1534 format_eth(struct ds
*ds
, const char *name
, const uint8_t key
[ETH_ADDR_LEN
],
1535 const uint8_t (*mask
)[ETH_ADDR_LEN
], bool verbose
)
1537 bool mask_empty
= mask
&& eth_addr_is_zero(*mask
);
1539 if (verbose
|| !mask_empty
) {
1540 bool mask_full
= !mask
|| eth_mask_is_exact(*mask
);
1543 ds_put_format(ds
, "%s="ETH_ADDR_FMT
",", name
, ETH_ADDR_ARGS(key
));
1545 ds_put_format(ds
, "%s=", name
);
1546 eth_format_masked(key
, *mask
, ds
);
1547 ds_put_char(ds
, ',');
1553 format_be64(struct ds
*ds
, const char *name
, ovs_be64 key
,
1554 const ovs_be64
*mask
, bool verbose
)
1556 bool mask_empty
= mask
&& !*mask
;
1558 if (verbose
|| !mask_empty
) {
1559 bool mask_full
= !mask
|| *mask
== OVS_BE64_MAX
;
1561 ds_put_format(ds
, "%s=0x%"PRIx64
, name
, ntohll(key
));
1562 if (!mask_full
) { /* Partially masked. */
1563 ds_put_format(ds
, "/%#"PRIx64
, ntohll(*mask
));
1565 ds_put_char(ds
, ',');
1570 format_ipv4(struct ds
*ds
, const char *name
, ovs_be32 key
,
1571 const ovs_be32
*mask
, bool verbose
)
1573 bool mask_empty
= mask
&& !*mask
;
1575 if (verbose
|| !mask_empty
) {
1576 bool mask_full
= !mask
|| *mask
== OVS_BE32_MAX
;
1578 ds_put_format(ds
, "%s="IP_FMT
, name
, IP_ARGS(key
));
1579 if (!mask_full
) { /* Partially masked. */
1580 ds_put_format(ds
, "/"IP_FMT
, IP_ARGS(*mask
));
1582 ds_put_char(ds
, ',');
1587 format_ipv6(struct ds
*ds
, const char *name
, const ovs_be32 key_
[4],
1588 const ovs_be32 (*mask_
)[4], bool verbose
)
1590 char buf
[INET6_ADDRSTRLEN
];
1591 const struct in6_addr
*key
= (const struct in6_addr
*)key_
;
1592 const struct in6_addr
*mask
= mask_
? (const struct in6_addr
*)*mask_
1594 bool mask_empty
= mask
&& ipv6_mask_is_any(mask
);
1596 if (verbose
|| !mask_empty
) {
1597 bool mask_full
= !mask
|| ipv6_mask_is_exact(mask
);
1599 inet_ntop(AF_INET6
, key
, buf
, sizeof buf
);
1600 ds_put_format(ds
, "%s=%s", name
, buf
);
1601 if (!mask_full
) { /* Partially masked. */
1602 inet_ntop(AF_INET6
, mask
, buf
, sizeof buf
);
1603 ds_put_format(ds
, "/%s", buf
);
1605 ds_put_char(ds
, ',');
1610 format_ipv6_label(struct ds
*ds
, const char *name
, ovs_be32 key
,
1611 const ovs_be32
*mask
, bool verbose
)
1613 bool mask_empty
= mask
&& !*mask
;
1615 if (verbose
|| !mask_empty
) {
1616 bool mask_full
= !mask
1617 || (*mask
& htonl(IPV6_LABEL_MASK
)) == htonl(IPV6_LABEL_MASK
);
1619 ds_put_format(ds
, "%s=%#"PRIx32
, name
, ntohl(key
));
1620 if (!mask_full
) { /* Partially masked. */
1621 ds_put_format(ds
, "/%#"PRIx32
, ntohl(*mask
));
1623 ds_put_char(ds
, ',');
1628 format_u8x(struct ds
*ds
, const char *name
, uint8_t key
,
1629 const uint8_t *mask
, bool verbose
)
1631 bool mask_empty
= mask
&& !*mask
;
1633 if (verbose
|| !mask_empty
) {
1634 bool mask_full
= !mask
|| *mask
== UINT8_MAX
;
1636 ds_put_format(ds
, "%s=%#"PRIx8
, name
, key
);
1637 if (!mask_full
) { /* Partially masked. */
1638 ds_put_format(ds
, "/%#"PRIx8
, *mask
);
1640 ds_put_char(ds
, ',');
1645 format_u8u(struct ds
*ds
, const char *name
, uint8_t key
,
1646 const uint8_t *mask
, bool verbose
)
1648 bool mask_empty
= mask
&& !*mask
;
1650 if (verbose
|| !mask_empty
) {
1651 bool mask_full
= !mask
|| *mask
== UINT8_MAX
;
1653 ds_put_format(ds
, "%s=%"PRIu8
, name
, key
);
1654 if (!mask_full
) { /* Partially masked. */
1655 ds_put_format(ds
, "/%#"PRIx8
, *mask
);
1657 ds_put_char(ds
, ',');
1662 format_be16(struct ds
*ds
, const char *name
, ovs_be16 key
,
1663 const ovs_be16
*mask
, bool verbose
)
1665 bool mask_empty
= mask
&& !*mask
;
1667 if (verbose
|| !mask_empty
) {
1668 bool mask_full
= !mask
|| *mask
== OVS_BE16_MAX
;
1670 ds_put_format(ds
, "%s=%"PRIu16
, name
, ntohs(key
));
1671 if (!mask_full
) { /* Partially masked. */
1672 ds_put_format(ds
, "/%#"PRIx16
, ntohs(*mask
));
1674 ds_put_char(ds
, ',');
1679 format_tun_flags(struct ds
*ds
, const char *name
, uint16_t key
,
1680 const uint16_t *mask
, bool verbose
)
1682 bool mask_empty
= mask
&& !*mask
;
1684 if (verbose
|| !mask_empty
) {
1685 bool mask_full
= !mask
|| (*mask
& FLOW_TNL_F_MASK
) == FLOW_TNL_F_MASK
;
1687 ds_put_cstr(ds
, name
);
1688 ds_put_char(ds
, '(');
1689 if (!mask_full
) { /* Partially masked. */
1690 format_flags_masked(ds
, NULL
, flow_tun_flag_to_string
, key
, *mask
);
1691 } else { /* Fully masked. */
1692 format_flags(ds
, flow_tun_flag_to_string
, key
, ',');
1694 ds_put_cstr(ds
, "),");
1699 format_frag(struct ds
*ds
, const char *name
, uint8_t key
,
1700 const uint8_t *mask
, bool verbose
)
1702 bool mask_empty
= mask
&& !*mask
;
1704 /* ODP frag is an enumeration field; partial masks are not meaningful. */
1705 if (verbose
|| !mask_empty
) {
1706 bool mask_full
= !mask
|| *mask
== UINT8_MAX
;
1708 if (!mask_full
) { /* Partially masked. */
1709 ds_put_format(ds
, "error: partial mask not supported for frag (%#"
1712 ds_put_format(ds
, "%s=%s,", name
, ovs_frag_type_to_string(key
));
1717 #define MASK(PTR, FIELD) PTR ? &PTR->FIELD : NULL
1720 format_odp_key_attr(const struct nlattr
*a
, const struct nlattr
*ma
,
1721 const struct hmap
*portno_names
, struct ds
*ds
,
1724 enum ovs_key_attr attr
= nl_attr_type(a
);
1725 char namebuf
[OVS_KEY_ATTR_BUFSIZE
];
1729 is_exact
= ma
? odp_mask_attr_is_exact(ma
) : true;
1731 ds_put_cstr(ds
, ovs_key_attr_to_string(attr
, namebuf
, sizeof namebuf
));
1734 expected_len
= odp_flow_key_attr_len(nl_attr_type(a
));
1735 if (expected_len
!= -2) {
1736 bool bad_key_len
= nl_attr_get_size(a
) != expected_len
;
1737 bool bad_mask_len
= ma
&& nl_attr_get_size(ma
) != expected_len
;
1739 if (bad_key_len
|| bad_mask_len
) {
1741 ds_put_format(ds
, "(bad key length %"PRIuSIZE
", expected %d)(",
1742 nl_attr_get_size(a
), expected_len
);
1744 format_generic_odp_key(a
, ds
);
1746 ds_put_char(ds
, '/');
1748 ds_put_format(ds
, "(bad mask length %"PRIuSIZE
", expected %d)(",
1749 nl_attr_get_size(ma
), expected_len
);
1751 format_generic_odp_key(ma
, ds
);
1753 ds_put_char(ds
, ')');
1759 ds_put_char(ds
, '(');
1761 case OVS_KEY_ATTR_ENCAP
:
1762 if (ma
&& nl_attr_get_size(ma
) && nl_attr_get_size(a
)) {
1763 odp_flow_format(nl_attr_get(a
), nl_attr_get_size(a
),
1764 nl_attr_get(ma
), nl_attr_get_size(ma
), NULL
, ds
,
1766 } else if (nl_attr_get_size(a
)) {
1767 odp_flow_format(nl_attr_get(a
), nl_attr_get_size(a
), NULL
, 0, NULL
,
1772 case OVS_KEY_ATTR_PRIORITY
:
1773 case OVS_KEY_ATTR_SKB_MARK
:
1774 case OVS_KEY_ATTR_DP_HASH
:
1775 case OVS_KEY_ATTR_RECIRC_ID
:
1776 ds_put_format(ds
, "%#"PRIx32
, nl_attr_get_u32(a
));
1778 ds_put_format(ds
, "/%#"PRIx32
, nl_attr_get_u32(ma
));
1782 case OVS_KEY_ATTR_TUNNEL
: {
1783 struct flow_tnl key
, mask_
;
1784 struct flow_tnl
*mask
= ma
? &mask_
: NULL
;
1787 memset(mask
, 0, sizeof *mask
);
1788 odp_tun_key_from_attr(ma
, mask
);
1790 memset(&key
, 0, sizeof key
);
1791 if (odp_tun_key_from_attr(a
, &key
) == ODP_FIT_ERROR
) {
1792 ds_put_format(ds
, "error");
1795 format_be64(ds
, "tun_id", key
.tun_id
, MASK(mask
, tun_id
), verbose
);
1796 format_ipv4(ds
, "src", key
.ip_src
, MASK(mask
, ip_src
), verbose
);
1797 format_ipv4(ds
, "dst", key
.ip_dst
, MASK(mask
, ip_dst
), verbose
);
1798 format_u8x(ds
, "tos", key
.ip_tos
, MASK(mask
, ip_tos
), verbose
);
1799 format_u8u(ds
, "ttl", key
.ip_ttl
, MASK(mask
, ip_ttl
), verbose
);
1800 format_be16(ds
, "tp_src", key
.tp_src
, MASK(mask
, tp_src
), verbose
);
1801 format_be16(ds
, "tp_dst", key
.tp_dst
, MASK(mask
, tp_dst
), verbose
);
1802 format_tun_flags(ds
, "flags", key
.flags
, MASK(mask
, flags
), verbose
);
1806 case OVS_KEY_ATTR_IN_PORT
:
1807 if (portno_names
&& verbose
&& is_exact
) {
1808 char *name
= odp_portno_names_get(portno_names
,
1809 u32_to_odp(nl_attr_get_u32(a
)));
1811 ds_put_format(ds
, "%s", name
);
1813 ds_put_format(ds
, "%"PRIu32
, nl_attr_get_u32(a
));
1816 ds_put_format(ds
, "%"PRIu32
, nl_attr_get_u32(a
));
1818 ds_put_format(ds
, "/%#"PRIx32
, nl_attr_get_u32(ma
));
1823 case OVS_KEY_ATTR_ETHERNET
: {
1824 const struct ovs_key_ethernet
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1825 const struct ovs_key_ethernet
*key
= nl_attr_get(a
);
1827 format_eth(ds
, "src", key
->eth_src
, MASK(mask
, eth_src
), verbose
);
1828 format_eth(ds
, "dst", key
->eth_dst
, MASK(mask
, eth_dst
), verbose
);
1832 case OVS_KEY_ATTR_VLAN
:
1833 format_vlan_tci(ds
, nl_attr_get_be16(a
),
1834 ma
? nl_attr_get_be16(ma
) : OVS_BE16_MAX
, verbose
);
1837 case OVS_KEY_ATTR_MPLS
: {
1838 const struct ovs_key_mpls
*mpls_key
= nl_attr_get(a
);
1839 const struct ovs_key_mpls
*mpls_mask
= NULL
;
1840 size_t size
= nl_attr_get_size(a
);
1842 if (!size
|| size
% sizeof *mpls_key
) {
1843 ds_put_format(ds
, "(bad key length %"PRIuSIZE
")", size
);
1847 mpls_mask
= nl_attr_get(ma
);
1848 if (size
!= nl_attr_get_size(ma
)) {
1849 ds_put_format(ds
, "(key length %"PRIuSIZE
" != "
1850 "mask length %"PRIuSIZE
")",
1851 size
, nl_attr_get_size(ma
));
1855 format_mpls(ds
, mpls_key
, mpls_mask
, size
/ sizeof *mpls_key
);
1858 case OVS_KEY_ATTR_ETHERTYPE
:
1859 ds_put_format(ds
, "0x%04"PRIx16
, ntohs(nl_attr_get_be16(a
)));
1861 ds_put_format(ds
, "/0x%04"PRIx16
, ntohs(nl_attr_get_be16(ma
)));
1865 case OVS_KEY_ATTR_IPV4
: {
1866 const struct ovs_key_ipv4
*key
= nl_attr_get(a
);
1867 const struct ovs_key_ipv4
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1869 format_ipv4(ds
, "src", key
->ipv4_src
, MASK(mask
, ipv4_src
), verbose
);
1870 format_ipv4(ds
, "dst", key
->ipv4_dst
, MASK(mask
, ipv4_dst
), verbose
);
1871 format_u8u(ds
, "proto", key
->ipv4_proto
, MASK(mask
, ipv4_proto
),
1873 format_u8x(ds
, "tos", key
->ipv4_tos
, MASK(mask
, ipv4_tos
), verbose
);
1874 format_u8u(ds
, "ttl", key
->ipv4_ttl
, MASK(mask
, ipv4_ttl
), verbose
);
1875 format_frag(ds
, "frag", key
->ipv4_frag
, MASK(mask
, ipv4_frag
),
1880 case OVS_KEY_ATTR_IPV6
: {
1881 const struct ovs_key_ipv6
*key
= nl_attr_get(a
);
1882 const struct ovs_key_ipv6
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1884 format_ipv6(ds
, "src", key
->ipv6_src
, MASK(mask
, ipv6_src
), verbose
);
1885 format_ipv6(ds
, "dst", key
->ipv6_dst
, MASK(mask
, ipv6_dst
), verbose
);
1886 format_ipv6_label(ds
, "label", key
->ipv6_label
, MASK(mask
, ipv6_label
),
1888 format_u8u(ds
, "proto", key
->ipv6_proto
, MASK(mask
, ipv6_proto
),
1890 format_u8x(ds
, "tclass", key
->ipv6_tclass
, MASK(mask
, ipv6_tclass
),
1892 format_u8u(ds
, "hlimit", key
->ipv6_hlimit
, MASK(mask
, ipv6_hlimit
),
1894 format_frag(ds
, "frag", key
->ipv6_frag
, MASK(mask
, ipv6_frag
),
1899 /* These have the same structure and format. */
1900 case OVS_KEY_ATTR_TCP
:
1901 case OVS_KEY_ATTR_UDP
:
1902 case OVS_KEY_ATTR_SCTP
: {
1903 const struct ovs_key_tcp
*key
= nl_attr_get(a
);
1904 const struct ovs_key_tcp
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1906 format_be16(ds
, "src", key
->tcp_src
, MASK(mask
, tcp_src
), verbose
);
1907 format_be16(ds
, "dst", key
->tcp_dst
, MASK(mask
, tcp_dst
), verbose
);
1911 case OVS_KEY_ATTR_TCP_FLAGS
:
1913 format_flags_masked(ds
, NULL
, packet_tcp_flag_to_string
,
1914 ntohs(nl_attr_get_be16(a
)),
1915 ntohs(nl_attr_get_be16(ma
)));
1917 format_flags(ds
, packet_tcp_flag_to_string
,
1918 ntohs(nl_attr_get_be16(a
)), ',');
1922 case OVS_KEY_ATTR_ICMP
: {
1923 const struct ovs_key_icmp
*key
= nl_attr_get(a
);
1924 const struct ovs_key_icmp
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1926 format_u8u(ds
, "type", key
->icmp_type
, MASK(mask
, icmp_type
), verbose
);
1927 format_u8u(ds
, "code", key
->icmp_code
, MASK(mask
, icmp_code
), verbose
);
1931 case OVS_KEY_ATTR_ICMPV6
: {
1932 const struct ovs_key_icmpv6
*key
= nl_attr_get(a
);
1933 const struct ovs_key_icmpv6
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1935 format_u8u(ds
, "type", key
->icmpv6_type
, MASK(mask
, icmpv6_type
),
1937 format_u8u(ds
, "code", key
->icmpv6_code
, MASK(mask
, icmpv6_code
),
1942 case OVS_KEY_ATTR_ARP
: {
1943 const struct ovs_key_arp
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1944 const struct ovs_key_arp
*key
= nl_attr_get(a
);
1946 format_ipv4(ds
, "sip", key
->arp_sip
, MASK(mask
, arp_sip
), verbose
);
1947 format_ipv4(ds
, "tip", key
->arp_tip
, MASK(mask
, arp_tip
), verbose
);
1948 format_be16(ds
, "op", key
->arp_op
, MASK(mask
, arp_op
), verbose
);
1949 format_eth(ds
, "sha", key
->arp_sha
, MASK(mask
, arp_sha
), verbose
);
1950 format_eth(ds
, "tha", key
->arp_tha
, MASK(mask
, arp_tha
), verbose
);
1954 case OVS_KEY_ATTR_ND
: {
1955 const struct ovs_key_nd
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1956 const struct ovs_key_nd
*key
= nl_attr_get(a
);
1958 format_ipv6(ds
, "target", key
->nd_target
, MASK(mask
, nd_target
),
1960 format_eth(ds
, "sll", key
->nd_sll
, MASK(mask
, nd_sll
), verbose
);
1961 format_eth(ds
, "tll", key
->nd_tll
, MASK(mask
, nd_tll
), verbose
);
1966 case OVS_KEY_ATTR_UNSPEC
:
1967 case __OVS_KEY_ATTR_MAX
:
1969 format_generic_odp_key(a
, ds
);
1971 ds_put_char(ds
, '/');
1972 format_generic_odp_key(ma
, ds
);
1976 ds_put_char(ds
, ')');
1979 static struct nlattr
*
1980 generate_all_wildcard_mask(struct ofpbuf
*ofp
, const struct nlattr
*key
)
1982 const struct nlattr
*a
;
1984 int type
= nl_attr_type(key
);
1985 int size
= nl_attr_get_size(key
);
1987 if (odp_flow_key_attr_len(type
) >=0) {
1988 nl_msg_put_unspec_zero(ofp
, type
, size
);
1992 nested_mask
= nl_msg_start_nested(ofp
, type
);
1993 NL_ATTR_FOR_EACH(a
, left
, key
, nl_attr_get_size(key
)) {
1994 generate_all_wildcard_mask(ofp
, nl_attr_get(a
));
1996 nl_msg_end_nested(ofp
, nested_mask
);
1999 return ofpbuf_base(ofp
);
2002 /* Appends to 'ds' a string representation of the 'key_len' bytes of
2003 * OVS_KEY_ATTR_* attributes in 'key'. If non-null, additionally formats the
2004 * 'mask_len' bytes of 'mask' which apply to 'key'. If 'portno_names' is
2005 * non-null and 'verbose' is true, translates odp port number to its name. */
2007 odp_flow_format(const struct nlattr
*key
, size_t key_len
,
2008 const struct nlattr
*mask
, size_t mask_len
,
2009 const struct hmap
*portno_names
, struct ds
*ds
, bool verbose
)
2012 const struct nlattr
*a
;
2014 bool has_ethtype_key
= false;
2015 const struct nlattr
*ma
= NULL
;
2017 bool first_field
= true;
2019 ofpbuf_init(&ofp
, 100);
2020 NL_ATTR_FOR_EACH (a
, left
, key
, key_len
) {
2021 bool is_nested_attr
;
2022 bool is_wildcard
= false;
2023 int attr_type
= nl_attr_type(a
);
2025 if (attr_type
== OVS_KEY_ATTR_ETHERTYPE
) {
2026 has_ethtype_key
= true;
2029 is_nested_attr
= (odp_flow_key_attr_len(attr_type
) == -2);
2031 if (mask
&& mask_len
) {
2032 ma
= nl_attr_find__(mask
, mask_len
, nl_attr_type(a
));
2033 is_wildcard
= ma
? odp_mask_attr_is_wildcard(ma
) : true;
2036 if (verbose
|| !is_wildcard
|| is_nested_attr
) {
2037 if (is_wildcard
&& !ma
) {
2038 ma
= generate_all_wildcard_mask(&ofp
, a
);
2041 ds_put_char(ds
, ',');
2043 format_odp_key_attr(a
, ma
, portno_names
, ds
, verbose
);
2044 first_field
= false;
2048 ofpbuf_uninit(&ofp
);
2053 if (left
== key_len
) {
2054 ds_put_cstr(ds
, "<empty>");
2056 ds_put_format(ds
, ",***%u leftover bytes*** (", left
);
2057 for (i
= 0; i
< left
; i
++) {
2058 ds_put_format(ds
, "%02x", ((const uint8_t *) a
)[i
]);
2060 ds_put_char(ds
, ')');
2062 if (!has_ethtype_key
) {
2063 ma
= nl_attr_find__(mask
, mask_len
, OVS_KEY_ATTR_ETHERTYPE
);
2065 ds_put_format(ds
, ",eth_type(0/0x%04"PRIx16
")",
2066 ntohs(nl_attr_get_be16(ma
)));
2070 ds_put_cstr(ds
, "<empty>");
2074 /* Appends to 'ds' a string representation of the 'key_len' bytes of
2075 * OVS_KEY_ATTR_* attributes in 'key'. */
2077 odp_flow_key_format(const struct nlattr
*key
,
2078 size_t key_len
, struct ds
*ds
)
2080 odp_flow_format(key
, key_len
, NULL
, 0, NULL
, ds
, true);
2084 ovs_frag_type_from_string(const char *s
, enum ovs_frag_type
*type
)
2086 if (!strcasecmp(s
, "no")) {
2087 *type
= OVS_FRAG_TYPE_NONE
;
2088 } else if (!strcasecmp(s
, "first")) {
2089 *type
= OVS_FRAG_TYPE_FIRST
;
2090 } else if (!strcasecmp(s
, "later")) {
2091 *type
= OVS_FRAG_TYPE_LATER
;
2101 scan_eth(const char *s
, uint8_t (*key
)[ETH_ADDR_LEN
],
2102 uint8_t (*mask
)[ETH_ADDR_LEN
])
2106 if (ovs_scan(s
, ETH_ADDR_SCAN_FMT
"%n", ETH_ADDR_SCAN_ARGS(*key
), &n
)) {
2110 if (ovs_scan(s
+ len
, "/"ETH_ADDR_SCAN_FMT
"%n",
2111 ETH_ADDR_SCAN_ARGS(*mask
), &n
)) {
2114 memset(mask
, 0xff, sizeof *mask
);
2123 scan_ipv4(const char *s
, ovs_be32
*key
, ovs_be32
*mask
)
2127 if (ovs_scan(s
, IP_SCAN_FMT
"%n", IP_SCAN_ARGS(key
), &n
)) {
2131 if (ovs_scan(s
+ len
, "/"IP_SCAN_FMT
"%n",
2132 IP_SCAN_ARGS(mask
), &n
)) {
2135 *mask
= OVS_BE32_MAX
;
2144 scan_ipv6(const char *s
, ovs_be32 (*key
)[4], ovs_be32 (*mask
)[4])
2147 char ipv6_s
[IPV6_SCAN_LEN
+ 1];
2149 if (ovs_scan(s
, IPV6_SCAN_FMT
"%n", ipv6_s
, &n
)
2150 && inet_pton(AF_INET6
, ipv6_s
, key
) == 1) {
2154 if (ovs_scan(s
+ len
, "/"IPV6_SCAN_FMT
"%n", ipv6_s
, &n
)
2155 && inet_pton(AF_INET6
, ipv6_s
, mask
) == 1) {
2158 memset(mask
, 0xff, sizeof *mask
);
2167 scan_ipv6_label(const char *s
, ovs_be32
*key
, ovs_be32
*mask
)
2172 if (ovs_scan(s
, "%i%n", &key_
, &n
)
2173 && (key_
& ~IPV6_LABEL_MASK
) == 0) {
2178 if (ovs_scan(s
+ len
, "/%i%n", &mask_
, &n
)
2179 && (mask_
& ~IPV6_LABEL_MASK
) == 0) {
2181 *mask
= htonl(mask_
);
2183 *mask
= htonl(IPV6_LABEL_MASK
);
2192 scan_u8(const char *s
, uint8_t *key
, uint8_t *mask
)
2196 if (ovs_scan(s
, "%"SCNi8
"%n", key
, &n
)) {
2200 if (ovs_scan(s
+ len
, "/%"SCNi8
"%n", mask
, &n
)) {
2212 scan_u32(const char *s
, uint32_t *key
, uint32_t *mask
)
2216 if (ovs_scan(s
, "%"SCNi32
"%n", key
, &n
)) {
2220 if (ovs_scan(s
+ len
, "/%"SCNi32
"%n", mask
, &n
)) {
2232 scan_be16(const char *s
, ovs_be16
*key
, ovs_be16
*mask
)
2234 uint16_t key_
, mask_
;
2237 if (ovs_scan(s
, "%"SCNi16
"%n", &key_
, &n
)) {
2242 if (ovs_scan(s
+ len
, "/%"SCNi16
"%n", &mask_
, &n
)) {
2244 *mask
= htons(mask_
);
2246 *mask
= OVS_BE16_MAX
;
2255 scan_be64(const char *s
, ovs_be64
*key
, ovs_be64
*mask
)
2257 uint64_t key_
, mask_
;
2260 if (ovs_scan(s
, "%"SCNi64
"%n", &key_
, &n
)) {
2263 *key
= htonll(key_
);
2265 if (ovs_scan(s
+ len
, "/%"SCNi64
"%n", &mask_
, &n
)) {
2267 *mask
= htonll(mask_
);
2269 *mask
= OVS_BE64_MAX
;
2278 scan_tun_flags(const char *s
, uint16_t *key
, uint16_t *mask
)
2280 uint32_t flags
, fmask
;
2283 n
= parse_flags(s
, flow_tun_flag_to_string
, &flags
,
2284 FLOW_TNL_F_MASK
, mask
? &fmask
: NULL
);
2285 if (n
>= 0 && s
[n
] == ')') {
2296 scan_tcp_flags(const char *s
, ovs_be16
*key
, ovs_be16
*mask
)
2298 uint32_t flags
, fmask
;
2301 n
= parse_flags(s
, packet_tcp_flag_to_string
, &flags
,
2302 TCP_FLAGS(OVS_BE16_MAX
), mask
? &fmask
: NULL
);
2304 *key
= htons(flags
);
2306 *mask
= htons(fmask
);
2314 scan_frag(const char *s
, uint8_t *key
, uint8_t *mask
)
2318 enum ovs_frag_type frag_type
;
2320 if (ovs_scan(s
, "%7[a-z]%n", frag
, &n
)
2321 && ovs_frag_type_from_string(frag
, &frag_type
)) {
2334 scan_port(const char *s
, uint32_t *key
, uint32_t *mask
,
2335 const struct simap
*port_names
)
2339 if (ovs_scan(s
, "%"SCNi32
"%n", key
, &n
)) {
2343 if (ovs_scan(s
+ len
, "/%"SCNi32
"%n", mask
, &n
)) {
2350 } else if (port_names
) {
2351 const struct simap_node
*node
;
2354 len
= strcspn(s
, ")");
2355 node
= simap_find_len(port_names
, s
, len
);
2368 /* Helper for vlan parsing. */
2369 struct ovs_key_vlan__
{
2374 set_be16_bf(ovs_be16
*bf
, uint8_t bits
, uint8_t offset
, uint16_t value
)
2376 const uint16_t mask
= ((1U << bits
) - 1) << offset
;
2378 if (value
>> bits
) {
2382 *bf
= htons((ntohs(*bf
) & ~mask
) | (value
<< offset
));
2387 scan_be16_bf(const char *s
, ovs_be16
*key
, ovs_be16
*mask
, uint8_t bits
,
2390 uint16_t key_
, mask_
;
2393 if (ovs_scan(s
, "%"SCNi16
"%n", &key_
, &n
)) {
2396 if (set_be16_bf(key
, bits
, offset
, key_
)) {
2398 if (ovs_scan(s
+ len
, "/%"SCNi16
"%n", &mask_
, &n
)) {
2401 if (!set_be16_bf(mask
, bits
, offset
, mask_
)) {
2405 *mask
|= htons(((1U << bits
) - 1) << offset
);
2415 scan_vid(const char *s
, ovs_be16
*key
, ovs_be16
*mask
)
2417 return scan_be16_bf(s
, key
, mask
, 12, VLAN_VID_SHIFT
);
2421 scan_pcp(const char *s
, ovs_be16
*key
, ovs_be16
*mask
)
2423 return scan_be16_bf(s
, key
, mask
, 3, VLAN_PCP_SHIFT
);
2427 scan_cfi(const char *s
, ovs_be16
*key
, ovs_be16
*mask
)
2429 return scan_be16_bf(s
, key
, mask
, 1, VLAN_CFI_SHIFT
);
2434 set_be32_bf(ovs_be32
*bf
, uint8_t bits
, uint8_t offset
, uint32_t value
)
2436 const uint32_t mask
= ((1U << bits
) - 1) << offset
;
2438 if (value
>> bits
) {
2442 *bf
= htonl((ntohl(*bf
) & ~mask
) | (value
<< offset
));
2447 scan_be32_bf(const char *s
, ovs_be32
*key
, ovs_be32
*mask
, uint8_t bits
,
2450 uint32_t key_
, mask_
;
2453 if (ovs_scan(s
, "%"SCNi32
"%n", &key_
, &n
)) {
2456 if (set_be32_bf(key
, bits
, offset
, key_
)) {
2458 if (ovs_scan(s
+ len
, "/%"SCNi32
"%n", &mask_
, &n
)) {
2461 if (!set_be32_bf(mask
, bits
, offset
, mask_
)) {
2465 *mask
|= htonl(((1U << bits
) - 1) << offset
);
2475 scan_mpls_label(const char *s
, ovs_be32
*key
, ovs_be32
*mask
)
2477 return scan_be32_bf(s
, key
, mask
, 20, MPLS_LABEL_SHIFT
);
2481 scan_mpls_tc(const char *s
, ovs_be32
*key
, ovs_be32
*mask
)
2483 return scan_be32_bf(s
, key
, mask
, 3, MPLS_TC_SHIFT
);
2487 scan_mpls_ttl(const char *s
, ovs_be32
*key
, ovs_be32
*mask
)
2489 return scan_be32_bf(s
, key
, mask
, 8, MPLS_TTL_SHIFT
);
2493 scan_mpls_bos(const char *s
, ovs_be32
*key
, ovs_be32
*mask
)
2495 return scan_be32_bf(s
, key
, mask
, 1, MPLS_BOS_SHIFT
);
2498 /* ATTR is compile-time constant, so only the case with correct data type
2499 * will be used. However, the compiler complains about the data type for
2500 * the other cases, so we must cast to make the compiler silent. */
2501 #define SCAN_PUT_ATTR(BUF, ATTR, DATA) \
2502 if ((ATTR) == OVS_KEY_ATTR_TUNNEL) { \
2503 tun_key_to_attr(BUF, (const struct flow_tnl *)(void *)&(DATA)); \
2505 nl_msg_put_unspec(BUF, ATTR, &(DATA), sizeof (DATA)); \
2508 #define SCAN_IF(NAME) \
2509 if (strncmp(s, NAME, strlen(NAME)) == 0) { \
2510 const char *start = s; \
2515 /* Usually no special initialization is needed. */
2516 #define SCAN_BEGIN(NAME, TYPE) \
2519 memset(&skey, 0, sizeof skey); \
2520 memset(&smask, 0, sizeof smask); \
2524 /* VLAN needs special initialization. */
2525 #define SCAN_BEGIN_INIT(NAME, TYPE, KEY_INIT, MASK_INIT) \
2527 TYPE skey = KEY_INIT; \
2528 TYPE smask = MASK_INIT; \
2532 /* Scan unnamed entry as 'TYPE' */
2533 #define SCAN_TYPE(TYPE, KEY, MASK) \
2534 len = scan_##TYPE(s, KEY, MASK); \
2540 /* Scan named ('NAME') entry 'FIELD' as 'TYPE'. */
2541 #define SCAN_FIELD(NAME, TYPE, FIELD) \
2542 if (strncmp(s, NAME, strlen(NAME)) == 0) { \
2543 s += strlen(NAME); \
2544 SCAN_TYPE(TYPE, &skey.FIELD, mask ? &smask.FIELD : NULL); \
2548 #define SCAN_FINISH() \
2549 } while (*s++ == ',' && len != 0); \
2550 if (s[-1] != ')') { \
2554 #define SCAN_FINISH_SINGLE() \
2556 if (*s++ != ')') { \
2560 #define SCAN_PUT(ATTR) \
2561 if (!mask || !is_all_zeros(&smask, sizeof smask)) { \
2562 SCAN_PUT_ATTR(key, ATTR, skey); \
2564 SCAN_PUT_ATTR(mask, ATTR, smask); \
2568 #define SCAN_END(ATTR) \
2574 #define SCAN_END_SINGLE(ATTR) \
2575 SCAN_FINISH_SINGLE(); \
2580 #define SCAN_SINGLE(NAME, TYPE, SCAN_AS, ATTR) \
2581 SCAN_BEGIN(NAME, TYPE) { \
2582 SCAN_TYPE(SCAN_AS, &skey, &smask); \
2583 } SCAN_END_SINGLE(ATTR)
2585 #define SCAN_SINGLE_NO_MASK(NAME, TYPE, SCAN_AS, ATTR) \
2586 SCAN_BEGIN(NAME, TYPE) { \
2587 SCAN_TYPE(SCAN_AS, &skey, NULL); \
2588 } SCAN_END_SINGLE(ATTR)
2590 /* scan_port needs one extra argument. */
2591 #define SCAN_SINGLE_PORT(NAME, TYPE, ATTR) \
2592 SCAN_BEGIN(NAME, TYPE) { \
2593 len = scan_port(s, &skey, &smask, port_names); \
2598 } SCAN_END_SINGLE(ATTR)
2601 parse_odp_key_mask_attr(const char *s
, const struct simap
*port_names
,
2602 struct ofpbuf
*key
, struct ofpbuf
*mask
)
2604 SCAN_SINGLE("skb_priority(", uint32_t, u32
, OVS_KEY_ATTR_PRIORITY
);
2605 SCAN_SINGLE("skb_mark(", uint32_t, u32
, OVS_KEY_ATTR_SKB_MARK
);
2606 SCAN_SINGLE_NO_MASK("recirc_id(", uint32_t, u32
, OVS_KEY_ATTR_RECIRC_ID
);
2607 SCAN_SINGLE("dp_hash(", uint32_t, u32
, OVS_KEY_ATTR_DP_HASH
);
2609 SCAN_BEGIN("tunnel(", struct flow_tnl
) {
2610 SCAN_FIELD("tun_id=", be64
, tun_id
);
2611 SCAN_FIELD("src=", ipv4
, ip_src
);
2612 SCAN_FIELD("dst=", ipv4
, ip_dst
);
2613 SCAN_FIELD("tos=", u8
, ip_tos
);
2614 SCAN_FIELD("ttl=", u8
, ip_ttl
);
2615 SCAN_FIELD("tp_src=", be16
, tp_src
);
2616 SCAN_FIELD("tp_dst=", be16
, tp_dst
);
2617 SCAN_FIELD("flags(", tun_flags
, flags
);
2618 } SCAN_END(OVS_KEY_ATTR_TUNNEL
);
2620 SCAN_SINGLE_PORT("in_port(", uint32_t, OVS_KEY_ATTR_IN_PORT
);
2622 SCAN_BEGIN("eth(", struct ovs_key_ethernet
) {
2623 SCAN_FIELD("src=", eth
, eth_src
);
2624 SCAN_FIELD("dst=", eth
, eth_dst
);
2625 } SCAN_END(OVS_KEY_ATTR_ETHERNET
);
2627 SCAN_BEGIN_INIT("vlan(", struct ovs_key_vlan__
,
2628 { htons(VLAN_CFI
) }, { htons(VLAN_CFI
) }) {
2629 SCAN_FIELD("vid=", vid
, tci
);
2630 SCAN_FIELD("pcp=", pcp
, tci
);
2631 SCAN_FIELD("cfi=", cfi
, tci
);
2632 } SCAN_END(OVS_KEY_ATTR_VLAN
);
2634 SCAN_SINGLE("eth_type(", ovs_be16
, be16
, OVS_KEY_ATTR_ETHERTYPE
);
2636 SCAN_BEGIN("mpls(", struct ovs_key_mpls
) {
2637 SCAN_FIELD("label=", mpls_label
, mpls_lse
);
2638 SCAN_FIELD("tc=", mpls_tc
, mpls_lse
);
2639 SCAN_FIELD("ttl=", mpls_ttl
, mpls_lse
);
2640 SCAN_FIELD("bos=", mpls_bos
, mpls_lse
);
2641 } SCAN_END(OVS_KEY_ATTR_MPLS
);
2643 SCAN_BEGIN("ipv4(", struct ovs_key_ipv4
) {
2644 SCAN_FIELD("src=", ipv4
, ipv4_src
);
2645 SCAN_FIELD("dst=", ipv4
, ipv4_dst
);
2646 SCAN_FIELD("proto=", u8
, ipv4_proto
);
2647 SCAN_FIELD("tos=", u8
, ipv4_tos
);
2648 SCAN_FIELD("ttl=", u8
, ipv4_ttl
);
2649 SCAN_FIELD("frag=", frag
, ipv4_frag
);
2650 } SCAN_END(OVS_KEY_ATTR_IPV4
);
2652 SCAN_BEGIN("ipv6(", struct ovs_key_ipv6
) {
2653 SCAN_FIELD("src=", ipv6
, ipv6_src
);
2654 SCAN_FIELD("dst=", ipv6
, ipv6_dst
);
2655 SCAN_FIELD("label=", ipv6_label
, ipv6_label
);
2656 SCAN_FIELD("proto=", u8
, ipv6_proto
);
2657 SCAN_FIELD("tclass=", u8
, ipv6_tclass
);
2658 SCAN_FIELD("hlimit=", u8
, ipv6_hlimit
);
2659 SCAN_FIELD("frag=", frag
, ipv6_frag
);
2660 } SCAN_END(OVS_KEY_ATTR_IPV6
);
2662 SCAN_BEGIN("tcp(", struct ovs_key_tcp
) {
2663 SCAN_FIELD("src=", be16
, tcp_src
);
2664 SCAN_FIELD("dst=", be16
, tcp_dst
);
2665 } SCAN_END(OVS_KEY_ATTR_TCP
);
2667 SCAN_SINGLE("tcp_flags(", ovs_be16
, tcp_flags
, OVS_KEY_ATTR_TCP_FLAGS
);
2669 SCAN_BEGIN("udp(", struct ovs_key_udp
) {
2670 SCAN_FIELD("src=", be16
, udp_src
);
2671 SCAN_FIELD("dst=", be16
, udp_dst
);
2672 } SCAN_END(OVS_KEY_ATTR_UDP
);
2674 SCAN_BEGIN("sctp(", struct ovs_key_sctp
) {
2675 SCAN_FIELD("src=", be16
, sctp_src
);
2676 SCAN_FIELD("dst=", be16
, sctp_dst
);
2677 } SCAN_END(OVS_KEY_ATTR_SCTP
);
2679 SCAN_BEGIN("icmp(", struct ovs_key_icmp
) {
2680 SCAN_FIELD("type=", u8
, icmp_type
);
2681 SCAN_FIELD("code=", u8
, icmp_code
);
2682 } SCAN_END(OVS_KEY_ATTR_ICMP
);
2684 SCAN_BEGIN("icmpv6(", struct ovs_key_icmpv6
) {
2685 SCAN_FIELD("type=", u8
, icmpv6_type
);
2686 SCAN_FIELD("code=", u8
, icmpv6_code
);
2687 } SCAN_END(OVS_KEY_ATTR_ICMPV6
);
2689 SCAN_BEGIN("arp(", struct ovs_key_arp
) {
2690 SCAN_FIELD("sip=", ipv4
, arp_sip
);
2691 SCAN_FIELD("tip=", ipv4
, arp_tip
);
2692 SCAN_FIELD("op=", be16
, arp_op
);
2693 SCAN_FIELD("sha=", eth
, arp_sha
);
2694 SCAN_FIELD("tha=", eth
, arp_tha
);
2695 } SCAN_END(OVS_KEY_ATTR_ARP
);
2697 SCAN_BEGIN("nd(", struct ovs_key_nd
) {
2698 SCAN_FIELD("target=", ipv6
, nd_target
);
2699 SCAN_FIELD("sll=", eth
, nd_sll
);
2700 SCAN_FIELD("tll=", eth
, nd_tll
);
2701 } SCAN_END(OVS_KEY_ATTR_ND
);
2703 /* Encap open-coded. */
2704 if (!strncmp(s
, "encap(", 6)) {
2705 const char *start
= s
;
2706 size_t encap
, encap_mask
= 0;
2708 encap
= nl_msg_start_nested(key
, OVS_KEY_ATTR_ENCAP
);
2710 encap_mask
= nl_msg_start_nested(mask
, OVS_KEY_ATTR_ENCAP
);
2717 s
+= strspn(s
, ", \t\r\n");
2720 } else if (*s
== ')') {
2724 retval
= parse_odp_key_mask_attr(s
, port_names
, key
, mask
);
2732 nl_msg_end_nested(key
, encap
);
2734 nl_msg_end_nested(mask
, encap_mask
);
2743 /* Parses the string representation of a datapath flow key, in the
2744 * format output by odp_flow_key_format(). Returns 0 if successful,
2745 * otherwise a positive errno value. On success, the flow key is
2746 * appended to 'key' as a series of Netlink attributes. On failure, no
2747 * data is appended to 'key'. Either way, 'key''s data might be
2750 * If 'port_names' is nonnull, it points to an simap that maps from a port name
2751 * to a port number. (Port names may be used instead of port numbers in
2754 * On success, the attributes appended to 'key' are individually syntactically
2755 * valid, but they may not be valid as a sequence. 'key' might, for example,
2756 * have duplicated keys. odp_flow_key_to_flow() will detect those errors. */
2758 odp_flow_from_string(const char *s
, const struct simap
*port_names
,
2759 struct ofpbuf
*key
, struct ofpbuf
*mask
)
2761 const size_t old_size
= ofpbuf_size(key
);
2765 s
+= strspn(s
, delimiters
);
2770 retval
= parse_odp_key_mask_attr(s
, port_names
, key
, mask
);
2772 ofpbuf_set_size(key
, old_size
);
2782 ovs_to_odp_frag(uint8_t nw_frag
, bool is_mask
)
2785 /* Netlink interface 'enum ovs_frag_type' is an 8-bit enumeration type,
2786 * not a set of flags or bitfields. Hence, if the struct flow nw_frag
2787 * mask, which is a set of bits, has the FLOW_NW_FRAG_ANY as zero, we
2788 * must use a zero mask for the netlink frag field, and all ones mask
2790 return (nw_frag
& FLOW_NW_FRAG_ANY
) ? UINT8_MAX
: 0;
2792 return !(nw_frag
& FLOW_NW_FRAG_ANY
) ? OVS_FRAG_TYPE_NONE
2793 : nw_frag
& FLOW_NW_FRAG_LATER
? OVS_FRAG_TYPE_LATER
2794 : OVS_FRAG_TYPE_FIRST
;
2797 static void get_ethernet_key(const struct flow
*, struct ovs_key_ethernet
*);
2798 static void put_ethernet_key(const struct ovs_key_ethernet
*, struct flow
*);
2799 static void get_ipv4_key(const struct flow
*, struct ovs_key_ipv4
*,
2801 static void put_ipv4_key(const struct ovs_key_ipv4
*, struct flow
*,
2803 static void get_ipv6_key(const struct flow
*, struct ovs_key_ipv6
*,
2805 static void put_ipv6_key(const struct ovs_key_ipv6
*, struct flow
*,
2807 static void get_arp_key(const struct flow
*, struct ovs_key_arp
*);
2808 static void put_arp_key(const struct ovs_key_arp
*, struct flow
*);
2810 /* These share the same layout. */
2812 struct ovs_key_tcp tcp
;
2813 struct ovs_key_udp udp
;
2814 struct ovs_key_sctp sctp
;
2817 static void get_tp_key(const struct flow
*, union ovs_key_tp
*);
2818 static void put_tp_key(const union ovs_key_tp
*, struct flow
*);
2821 odp_flow_key_from_flow__(struct ofpbuf
*buf
, const struct flow
*flow
,
2822 const struct flow
*mask
, odp_port_t odp_in_port
,
2823 size_t max_mpls_depth
, bool recirc
, bool export_mask
)
2825 struct ovs_key_ethernet
*eth_key
;
2827 const struct flow
*data
= export_mask
? mask
: flow
;
2829 nl_msg_put_u32(buf
, OVS_KEY_ATTR_PRIORITY
, data
->skb_priority
);
2831 if (flow
->tunnel
.ip_dst
|| export_mask
) {
2832 tun_key_to_attr(buf
, &data
->tunnel
);
2835 nl_msg_put_u32(buf
, OVS_KEY_ATTR_SKB_MARK
, data
->pkt_mark
);
2838 nl_msg_put_u32(buf
, OVS_KEY_ATTR_RECIRC_ID
, data
->recirc_id
);
2839 nl_msg_put_u32(buf
, OVS_KEY_ATTR_DP_HASH
, data
->dp_hash
);
2842 /* Add an ingress port attribute if this is a mask or 'odp_in_port'
2843 * is not the magical value "ODPP_NONE". */
2844 if (export_mask
|| odp_in_port
!= ODPP_NONE
) {
2845 nl_msg_put_odp_port(buf
, OVS_KEY_ATTR_IN_PORT
, odp_in_port
);
2848 eth_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_ETHERNET
,
2850 get_ethernet_key(data
, eth_key
);
2852 if (flow
->vlan_tci
!= htons(0) || flow
->dl_type
== htons(ETH_TYPE_VLAN
)) {
2854 nl_msg_put_be16(buf
, OVS_KEY_ATTR_ETHERTYPE
, OVS_BE16_MAX
);
2856 nl_msg_put_be16(buf
, OVS_KEY_ATTR_ETHERTYPE
, htons(ETH_TYPE_VLAN
));
2858 nl_msg_put_be16(buf
, OVS_KEY_ATTR_VLAN
, data
->vlan_tci
);
2859 encap
= nl_msg_start_nested(buf
, OVS_KEY_ATTR_ENCAP
);
2860 if (flow
->vlan_tci
== htons(0)) {
2867 if (ntohs(flow
->dl_type
) < ETH_TYPE_MIN
) {
2868 /* For backwards compatibility with kernels that don't support
2869 * wildcarding, the following convention is used to encode the
2870 * OVS_KEY_ATTR_ETHERTYPE for key and mask:
2873 * -------- -------- -------
2874 * >0x5ff 0xffff Specified Ethernet II Ethertype.
2875 * >0x5ff 0 Any Ethernet II or non-Ethernet II frame.
2876 * <none> 0xffff Any non-Ethernet II frame (except valid
2877 * 802.3 SNAP packet with valid eth_type).
2880 nl_msg_put_be16(buf
, OVS_KEY_ATTR_ETHERTYPE
, OVS_BE16_MAX
);
2885 nl_msg_put_be16(buf
, OVS_KEY_ATTR_ETHERTYPE
, data
->dl_type
);
2887 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
2888 struct ovs_key_ipv4
*ipv4_key
;
2890 ipv4_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_IPV4
,
2892 get_ipv4_key(data
, ipv4_key
, export_mask
);
2893 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
2894 struct ovs_key_ipv6
*ipv6_key
;
2896 ipv6_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_IPV6
,
2898 get_ipv6_key(data
, ipv6_key
, export_mask
);
2899 } else if (flow
->dl_type
== htons(ETH_TYPE_ARP
) ||
2900 flow
->dl_type
== htons(ETH_TYPE_RARP
)) {
2901 struct ovs_key_arp
*arp_key
;
2903 arp_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_ARP
,
2905 get_arp_key(data
, arp_key
);
2906 } else if (eth_type_mpls(flow
->dl_type
)) {
2907 struct ovs_key_mpls
*mpls_key
;
2910 n
= flow_count_mpls_labels(flow
, NULL
);
2911 n
= MIN(n
, max_mpls_depth
);
2912 mpls_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_MPLS
,
2913 n
* sizeof *mpls_key
);
2914 for (i
= 0; i
< n
; i
++) {
2915 mpls_key
[i
].mpls_lse
= data
->mpls_lse
[i
];
2919 if (is_ip_any(flow
) && !(flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
2920 if (flow
->nw_proto
== IPPROTO_TCP
) {
2921 union ovs_key_tp
*tcp_key
;
2923 tcp_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_TCP
,
2925 get_tp_key(data
, tcp_key
);
2926 if (data
->tcp_flags
) {
2927 nl_msg_put_be16(buf
, OVS_KEY_ATTR_TCP_FLAGS
, data
->tcp_flags
);
2929 } else if (flow
->nw_proto
== IPPROTO_UDP
) {
2930 union ovs_key_tp
*udp_key
;
2932 udp_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_UDP
,
2934 get_tp_key(data
, udp_key
);
2935 } else if (flow
->nw_proto
== IPPROTO_SCTP
) {
2936 union ovs_key_tp
*sctp_key
;
2938 sctp_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_SCTP
,
2940 get_tp_key(data
, sctp_key
);
2941 } else if (flow
->dl_type
== htons(ETH_TYPE_IP
)
2942 && flow
->nw_proto
== IPPROTO_ICMP
) {
2943 struct ovs_key_icmp
*icmp_key
;
2945 icmp_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_ICMP
,
2947 icmp_key
->icmp_type
= ntohs(data
->tp_src
);
2948 icmp_key
->icmp_code
= ntohs(data
->tp_dst
);
2949 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)
2950 && flow
->nw_proto
== IPPROTO_ICMPV6
) {
2951 struct ovs_key_icmpv6
*icmpv6_key
;
2953 icmpv6_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_ICMPV6
,
2954 sizeof *icmpv6_key
);
2955 icmpv6_key
->icmpv6_type
= ntohs(data
->tp_src
);
2956 icmpv6_key
->icmpv6_code
= ntohs(data
->tp_dst
);
2958 if (flow
->tp_dst
== htons(0)
2959 && (flow
->tp_src
== htons(ND_NEIGHBOR_SOLICIT
)
2960 || flow
->tp_src
== htons(ND_NEIGHBOR_ADVERT
))
2961 && (!export_mask
|| (data
->tp_src
== htons(0xffff)
2962 && data
->tp_dst
== htons(0xffff)))) {
2964 struct ovs_key_nd
*nd_key
;
2966 nd_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_ND
,
2968 memcpy(nd_key
->nd_target
, &data
->nd_target
,
2969 sizeof nd_key
->nd_target
);
2970 memcpy(nd_key
->nd_sll
, data
->arp_sha
, ETH_ADDR_LEN
);
2971 memcpy(nd_key
->nd_tll
, data
->arp_tha
, ETH_ADDR_LEN
);
2978 nl_msg_end_nested(buf
, encap
);
2982 /* Appends a representation of 'flow' as OVS_KEY_ATTR_* attributes to 'buf'.
2983 * 'flow->in_port' is ignored (since it is likely to be an OpenFlow port
2984 * number rather than a datapath port number). Instead, if 'odp_in_port'
2985 * is anything other than ODPP_NONE, it is included in 'buf' as the input
2988 * 'buf' must have at least ODPUTIL_FLOW_KEY_BYTES bytes of space, or be
2989 * capable of being expanded to allow for that much space.
2991 * 'recirc' indicates support for recirculation fields. If this is true, then
2992 * these fields will always be serialised. */
2994 odp_flow_key_from_flow(struct ofpbuf
*buf
, const struct flow
*flow
,
2995 const struct flow
*mask
, odp_port_t odp_in_port
,
2998 odp_flow_key_from_flow__(buf
, flow
, mask
, odp_in_port
, SIZE_MAX
, recirc
,
3002 /* Appends a representation of 'mask' as OVS_KEY_ATTR_* attributes to
3003 * 'buf'. 'flow' is used as a template to determine how to interpret
3004 * 'mask'. For example, the 'dl_type' of 'mask' describes the mask, but
3005 * it doesn't indicate whether the other fields should be interpreted as
3006 * ARP, IPv4, IPv6, etc.
3008 * 'buf' must have at least ODPUTIL_FLOW_KEY_BYTES bytes of space, or be
3009 * capable of being expanded to allow for that much space.
3011 * 'recirc' indicates support for recirculation fields. If this is true, then
3012 * these fields will always be serialised. */
3014 odp_flow_key_from_mask(struct ofpbuf
*buf
, const struct flow
*mask
,
3015 const struct flow
*flow
, uint32_t odp_in_port_mask
,
3016 size_t max_mpls_depth
, bool recirc
)
3018 odp_flow_key_from_flow__(buf
, flow
, mask
, u32_to_odp(odp_in_port_mask
),
3019 max_mpls_depth
, recirc
, true);
3022 /* Generate ODP flow key from the given packet metadata */
3024 odp_key_from_pkt_metadata(struct ofpbuf
*buf
, const struct pkt_metadata
*md
)
3026 nl_msg_put_u32(buf
, OVS_KEY_ATTR_PRIORITY
, md
->skb_priority
);
3028 if (md
->tunnel
.ip_dst
) {
3029 tun_key_to_attr(buf
, &md
->tunnel
);
3032 nl_msg_put_u32(buf
, OVS_KEY_ATTR_SKB_MARK
, md
->pkt_mark
);
3034 /* Add an ingress port attribute if 'odp_in_port' is not the magical
3035 * value "ODPP_NONE". */
3036 if (md
->in_port
.odp_port
!= ODPP_NONE
) {
3037 nl_msg_put_odp_port(buf
, OVS_KEY_ATTR_IN_PORT
, md
->in_port
.odp_port
);
3041 /* Generate packet metadata from the given ODP flow key. */
3043 odp_key_to_pkt_metadata(const struct nlattr
*key
, size_t key_len
,
3044 struct pkt_metadata
*md
)
3046 const struct nlattr
*nla
;
3048 uint32_t wanted_attrs
= 1u << OVS_KEY_ATTR_PRIORITY
|
3049 1u << OVS_KEY_ATTR_SKB_MARK
| 1u << OVS_KEY_ATTR_TUNNEL
|
3050 1u << OVS_KEY_ATTR_IN_PORT
;
3052 *md
= PKT_METADATA_INITIALIZER(ODPP_NONE
);
3054 NL_ATTR_FOR_EACH (nla
, left
, key
, key_len
) {
3055 uint16_t type
= nl_attr_type(nla
);
3056 size_t len
= nl_attr_get_size(nla
);
3057 int expected_len
= odp_flow_key_attr_len(type
);
3059 if (len
!= expected_len
&& expected_len
>= 0) {
3064 case OVS_KEY_ATTR_RECIRC_ID
:
3065 md
->recirc_id
= nl_attr_get_u32(nla
);
3066 wanted_attrs
&= ~(1u << OVS_KEY_ATTR_RECIRC_ID
);
3068 case OVS_KEY_ATTR_DP_HASH
:
3069 md
->dp_hash
= nl_attr_get_u32(nla
);
3070 wanted_attrs
&= ~(1u << OVS_KEY_ATTR_DP_HASH
);
3072 case OVS_KEY_ATTR_PRIORITY
:
3073 md
->skb_priority
= nl_attr_get_u32(nla
);
3074 wanted_attrs
&= ~(1u << OVS_KEY_ATTR_PRIORITY
);
3076 case OVS_KEY_ATTR_SKB_MARK
:
3077 md
->pkt_mark
= nl_attr_get_u32(nla
);
3078 wanted_attrs
&= ~(1u << OVS_KEY_ATTR_SKB_MARK
);
3080 case OVS_KEY_ATTR_TUNNEL
: {
3081 enum odp_key_fitness res
;
3083 res
= odp_tun_key_from_attr(nla
, &md
->tunnel
);
3084 if (res
== ODP_FIT_ERROR
) {
3085 memset(&md
->tunnel
, 0, sizeof md
->tunnel
);
3086 } else if (res
== ODP_FIT_PERFECT
) {
3087 wanted_attrs
&= ~(1u << OVS_KEY_ATTR_TUNNEL
);
3091 case OVS_KEY_ATTR_IN_PORT
:
3092 md
->in_port
.odp_port
= nl_attr_get_odp_port(nla
);
3093 wanted_attrs
&= ~(1u << OVS_KEY_ATTR_IN_PORT
);
3099 if (!wanted_attrs
) {
3100 return; /* Have everything. */
3106 odp_flow_key_hash(const struct nlattr
*key
, size_t key_len
)
3108 BUILD_ASSERT_DECL(!(NLA_ALIGNTO
% sizeof(uint32_t)));
3109 return hash_words(ALIGNED_CAST(const uint32_t *, key
),
3110 key_len
/ sizeof(uint32_t), 0);
3114 log_odp_key_attributes(struct vlog_rate_limit
*rl
, const char *title
,
3115 uint64_t attrs
, int out_of_range_attr
,
3116 const struct nlattr
*key
, size_t key_len
)
3121 if (VLOG_DROP_DBG(rl
)) {
3126 for (i
= 0; i
< 64; i
++) {
3127 if (attrs
& (UINT64_C(1) << i
)) {
3128 char namebuf
[OVS_KEY_ATTR_BUFSIZE
];
3130 ds_put_format(&s
, " %s",
3131 ovs_key_attr_to_string(i
, namebuf
, sizeof namebuf
));
3134 if (out_of_range_attr
) {
3135 ds_put_format(&s
, " %d (and possibly others)", out_of_range_attr
);
3138 ds_put_cstr(&s
, ": ");
3139 odp_flow_key_format(key
, key_len
, &s
);
3141 VLOG_DBG("%s:%s", title
, ds_cstr(&s
));
3146 odp_to_ovs_frag(uint8_t odp_frag
, bool is_mask
)
3148 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3151 return odp_frag
? FLOW_NW_FRAG_MASK
: 0;
3154 if (odp_frag
> OVS_FRAG_TYPE_LATER
) {
3155 VLOG_ERR_RL(&rl
, "invalid frag %"PRIu8
" in flow key", odp_frag
);
3156 return 0xff; /* Error. */
3159 return (odp_frag
== OVS_FRAG_TYPE_NONE
) ? 0
3160 : (odp_frag
== OVS_FRAG_TYPE_FIRST
) ? FLOW_NW_FRAG_ANY
3161 : FLOW_NW_FRAG_ANY
| FLOW_NW_FRAG_LATER
;
3165 parse_flow_nlattrs(const struct nlattr
*key
, size_t key_len
,
3166 const struct nlattr
*attrs
[], uint64_t *present_attrsp
,
3167 int *out_of_range_attrp
)
3169 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(10, 10);
3170 const struct nlattr
*nla
;
3171 uint64_t present_attrs
;
3174 BUILD_ASSERT(OVS_KEY_ATTR_MAX
< CHAR_BIT
* sizeof present_attrs
);
3176 *out_of_range_attrp
= 0;
3177 NL_ATTR_FOR_EACH (nla
, left
, key
, key_len
) {
3178 uint16_t type
= nl_attr_type(nla
);
3179 size_t len
= nl_attr_get_size(nla
);
3180 int expected_len
= odp_flow_key_attr_len(type
);
3182 if (len
!= expected_len
&& expected_len
>= 0) {
3183 char namebuf
[OVS_KEY_ATTR_BUFSIZE
];
3185 VLOG_ERR_RL(&rl
, "attribute %s has length %"PRIuSIZE
" but should have "
3186 "length %d", ovs_key_attr_to_string(type
, namebuf
,
3192 if (type
> OVS_KEY_ATTR_MAX
) {
3193 *out_of_range_attrp
= type
;
3195 if (present_attrs
& (UINT64_C(1) << type
)) {
3196 char namebuf
[OVS_KEY_ATTR_BUFSIZE
];
3198 VLOG_ERR_RL(&rl
, "duplicate %s attribute in flow key",
3199 ovs_key_attr_to_string(type
,
3200 namebuf
, sizeof namebuf
));
3204 present_attrs
|= UINT64_C(1) << type
;
3209 VLOG_ERR_RL(&rl
, "trailing garbage in flow key");
3213 *present_attrsp
= present_attrs
;
3217 static enum odp_key_fitness
3218 check_expectations(uint64_t present_attrs
, int out_of_range_attr
,
3219 uint64_t expected_attrs
,
3220 const struct nlattr
*key
, size_t key_len
)
3222 uint64_t missing_attrs
;
3223 uint64_t extra_attrs
;
3225 missing_attrs
= expected_attrs
& ~present_attrs
;
3226 if (missing_attrs
) {
3227 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(10, 10);
3228 log_odp_key_attributes(&rl
, "expected but not present",
3229 missing_attrs
, 0, key
, key_len
);
3230 return ODP_FIT_TOO_LITTLE
;
3233 extra_attrs
= present_attrs
& ~expected_attrs
;
3234 if (extra_attrs
|| out_of_range_attr
) {
3235 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(10, 10);
3236 log_odp_key_attributes(&rl
, "present but not expected",
3237 extra_attrs
, out_of_range_attr
, key
, key_len
);
3238 return ODP_FIT_TOO_MUCH
;
3241 return ODP_FIT_PERFECT
;
3245 parse_ethertype(const struct nlattr
*attrs
[OVS_KEY_ATTR_MAX
+ 1],
3246 uint64_t present_attrs
, uint64_t *expected_attrs
,
3247 struct flow
*flow
, const struct flow
*src_flow
)
3249 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3250 bool is_mask
= flow
!= src_flow
;
3252 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE
)) {
3253 flow
->dl_type
= nl_attr_get_be16(attrs
[OVS_KEY_ATTR_ETHERTYPE
]);
3254 if (!is_mask
&& ntohs(flow
->dl_type
) < ETH_TYPE_MIN
) {
3255 VLOG_ERR_RL(&rl
, "invalid Ethertype %"PRIu16
" in flow key",
3256 ntohs(flow
->dl_type
));
3259 if (is_mask
&& ntohs(src_flow
->dl_type
) < ETH_TYPE_MIN
&&
3260 flow
->dl_type
!= htons(0xffff)) {
3263 *expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE
;
3266 flow
->dl_type
= htons(FLOW_DL_TYPE_NONE
);
3267 } else if (ntohs(src_flow
->dl_type
) < ETH_TYPE_MIN
) {
3268 /* See comments in odp_flow_key_from_flow__(). */
3269 VLOG_ERR_RL(&rl
, "mask expected for non-Ethernet II frame");
3276 static enum odp_key_fitness
3277 parse_l2_5_onward(const struct nlattr
*attrs
[OVS_KEY_ATTR_MAX
+ 1],
3278 uint64_t present_attrs
, int out_of_range_attr
,
3279 uint64_t expected_attrs
, struct flow
*flow
,
3280 const struct nlattr
*key
, size_t key_len
,
3281 const struct flow
*src_flow
)
3283 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3284 bool is_mask
= src_flow
!= flow
;
3285 const void *check_start
= NULL
;
3286 size_t check_len
= 0;
3287 enum ovs_key_attr expected_bit
= 0xff;
3289 if (eth_type_mpls(src_flow
->dl_type
)) {
3290 if (!is_mask
|| present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_MPLS
)) {
3291 expected_attrs
|= (UINT64_C(1) << OVS_KEY_ATTR_MPLS
);
3293 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_MPLS
)) {
3294 size_t size
= nl_attr_get_size(attrs
[OVS_KEY_ATTR_MPLS
]);
3295 const ovs_be32
*mpls_lse
= nl_attr_get(attrs
[OVS_KEY_ATTR_MPLS
]);
3296 int n
= size
/ sizeof(ovs_be32
);
3299 if (!size
|| size
% sizeof(ovs_be32
)) {
3300 return ODP_FIT_ERROR
;
3302 if (flow
->mpls_lse
[0] && flow
->dl_type
!= htons(0xffff)) {
3303 return ODP_FIT_ERROR
;
3306 for (i
= 0; i
< n
&& i
< FLOW_MAX_MPLS_LABELS
; i
++) {
3307 flow
->mpls_lse
[i
] = mpls_lse
[i
];
3309 if (n
> FLOW_MAX_MPLS_LABELS
) {
3310 return ODP_FIT_TOO_MUCH
;
3314 /* BOS may be set only in the innermost label. */
3315 for (i
= 0; i
< n
- 1; i
++) {
3316 if (flow
->mpls_lse
[i
] & htonl(MPLS_BOS_MASK
)) {
3317 return ODP_FIT_ERROR
;
3321 /* BOS must be set in the innermost label. */
3322 if (n
< FLOW_MAX_MPLS_LABELS
3323 && !(flow
->mpls_lse
[n
- 1] & htonl(MPLS_BOS_MASK
))) {
3324 return ODP_FIT_TOO_LITTLE
;
3330 } else if (src_flow
->dl_type
== htons(ETH_TYPE_IP
)) {
3332 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_IPV4
;
3334 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_IPV4
)) {
3335 const struct ovs_key_ipv4
*ipv4_key
;
3337 ipv4_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_IPV4
]);
3338 put_ipv4_key(ipv4_key
, flow
, is_mask
);
3339 if (flow
->nw_frag
> FLOW_NW_FRAG_MASK
) {
3340 return ODP_FIT_ERROR
;
3343 check_start
= ipv4_key
;
3344 check_len
= sizeof *ipv4_key
;
3345 expected_bit
= OVS_KEY_ATTR_IPV4
;
3348 } else if (src_flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
3350 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_IPV6
;
3352 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_IPV6
)) {
3353 const struct ovs_key_ipv6
*ipv6_key
;
3355 ipv6_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_IPV6
]);
3356 put_ipv6_key(ipv6_key
, flow
, is_mask
);
3357 if (flow
->nw_frag
> FLOW_NW_FRAG_MASK
) {
3358 return ODP_FIT_ERROR
;
3361 check_start
= ipv6_key
;
3362 check_len
= sizeof *ipv6_key
;
3363 expected_bit
= OVS_KEY_ATTR_IPV6
;
3366 } else if (src_flow
->dl_type
== htons(ETH_TYPE_ARP
) ||
3367 src_flow
->dl_type
== htons(ETH_TYPE_RARP
)) {
3369 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ARP
;
3371 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ARP
)) {
3372 const struct ovs_key_arp
*arp_key
;
3374 arp_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_ARP
]);
3375 if (!is_mask
&& (arp_key
->arp_op
& htons(0xff00))) {
3376 VLOG_ERR_RL(&rl
, "unsupported ARP opcode %"PRIu16
" in flow "
3377 "key", ntohs(arp_key
->arp_op
));
3378 return ODP_FIT_ERROR
;
3380 put_arp_key(arp_key
, flow
);
3382 check_start
= arp_key
;
3383 check_len
= sizeof *arp_key
;
3384 expected_bit
= OVS_KEY_ATTR_ARP
;
3390 if (check_len
> 0) { /* Happens only when 'is_mask'. */
3391 if (!is_all_zeros(check_start
, check_len
) &&
3392 flow
->dl_type
!= htons(0xffff)) {
3393 return ODP_FIT_ERROR
;
3395 expected_attrs
|= UINT64_C(1) << expected_bit
;
3399 expected_bit
= OVS_KEY_ATTR_UNSPEC
;
3400 if (src_flow
->nw_proto
== IPPROTO_TCP
3401 && (src_flow
->dl_type
== htons(ETH_TYPE_IP
) ||
3402 src_flow
->dl_type
== htons(ETH_TYPE_IPV6
))
3403 && !(src_flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
3405 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_TCP
;
3407 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_TCP
)) {
3408 const union ovs_key_tp
*tcp_key
;
3410 tcp_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_TCP
]);
3411 put_tp_key(tcp_key
, flow
);
3412 expected_bit
= OVS_KEY_ATTR_TCP
;
3414 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_TCP_FLAGS
)) {
3415 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_TCP_FLAGS
;
3416 flow
->tcp_flags
= nl_attr_get_be16(attrs
[OVS_KEY_ATTR_TCP_FLAGS
]);
3418 } else if (src_flow
->nw_proto
== IPPROTO_UDP
3419 && (src_flow
->dl_type
== htons(ETH_TYPE_IP
) ||
3420 src_flow
->dl_type
== htons(ETH_TYPE_IPV6
))
3421 && !(src_flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
3423 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_UDP
;
3425 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_UDP
)) {
3426 const union ovs_key_tp
*udp_key
;
3428 udp_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_UDP
]);
3429 put_tp_key(udp_key
, flow
);
3430 expected_bit
= OVS_KEY_ATTR_UDP
;
3432 } else if (src_flow
->nw_proto
== IPPROTO_SCTP
3433 && (src_flow
->dl_type
== htons(ETH_TYPE_IP
) ||
3434 src_flow
->dl_type
== htons(ETH_TYPE_IPV6
))
3435 && !(src_flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
3437 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_SCTP
;
3439 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_SCTP
)) {
3440 const union ovs_key_tp
*sctp_key
;
3442 sctp_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_SCTP
]);
3443 put_tp_key(sctp_key
, flow
);
3444 expected_bit
= OVS_KEY_ATTR_SCTP
;
3446 } else if (src_flow
->nw_proto
== IPPROTO_ICMP
3447 && src_flow
->dl_type
== htons(ETH_TYPE_IP
)
3448 && !(src_flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
3450 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ICMP
;
3452 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ICMP
)) {
3453 const struct ovs_key_icmp
*icmp_key
;
3455 icmp_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_ICMP
]);
3456 flow
->tp_src
= htons(icmp_key
->icmp_type
);
3457 flow
->tp_dst
= htons(icmp_key
->icmp_code
);
3458 expected_bit
= OVS_KEY_ATTR_ICMP
;
3460 } else if (src_flow
->nw_proto
== IPPROTO_ICMPV6
3461 && src_flow
->dl_type
== htons(ETH_TYPE_IPV6
)
3462 && !(src_flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
3464 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ICMPV6
;
3466 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ICMPV6
)) {
3467 const struct ovs_key_icmpv6
*icmpv6_key
;
3469 icmpv6_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_ICMPV6
]);
3470 flow
->tp_src
= htons(icmpv6_key
->icmpv6_type
);
3471 flow
->tp_dst
= htons(icmpv6_key
->icmpv6_code
);
3472 expected_bit
= OVS_KEY_ATTR_ICMPV6
;
3473 if (src_flow
->tp_dst
== htons(0) &&
3474 (src_flow
->tp_src
== htons(ND_NEIGHBOR_SOLICIT
) ||
3475 src_flow
->tp_src
== htons(ND_NEIGHBOR_ADVERT
))) {
3477 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ND
;
3479 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ND
)) {
3480 const struct ovs_key_nd
*nd_key
;
3482 nd_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_ND
]);
3483 memcpy(&flow
->nd_target
, nd_key
->nd_target
,
3484 sizeof flow
->nd_target
);
3485 memcpy(flow
->arp_sha
, nd_key
->nd_sll
, ETH_ADDR_LEN
);
3486 memcpy(flow
->arp_tha
, nd_key
->nd_tll
, ETH_ADDR_LEN
);
3488 if (!is_all_zeros(nd_key
, sizeof *nd_key
) &&
3489 (flow
->tp_src
!= htons(0xffff) ||
3490 flow
->tp_dst
!= htons(0xffff))) {
3491 return ODP_FIT_ERROR
;
3493 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ND
;
3500 if (is_mask
&& expected_bit
!= OVS_KEY_ATTR_UNSPEC
) {
3501 if ((flow
->tp_src
|| flow
->tp_dst
) && flow
->nw_proto
!= 0xff) {
3502 return ODP_FIT_ERROR
;
3504 expected_attrs
|= UINT64_C(1) << expected_bit
;
3509 return check_expectations(present_attrs
, out_of_range_attr
, expected_attrs
,
3513 /* Parse 802.1Q header then encapsulated L3 attributes. */
3514 static enum odp_key_fitness
3515 parse_8021q_onward(const struct nlattr
*attrs
[OVS_KEY_ATTR_MAX
+ 1],
3516 uint64_t present_attrs
, int out_of_range_attr
,
3517 uint64_t expected_attrs
, struct flow
*flow
,
3518 const struct nlattr
*key
, size_t key_len
,
3519 const struct flow
*src_flow
)
3521 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3522 bool is_mask
= src_flow
!= flow
;
3524 const struct nlattr
*encap
3525 = (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ENCAP
)
3526 ? attrs
[OVS_KEY_ATTR_ENCAP
] : NULL
);
3527 enum odp_key_fitness encap_fitness
;
3528 enum odp_key_fitness fitness
;
3530 /* Calculate fitness of outer attributes. */
3532 expected_attrs
|= ((UINT64_C(1) << OVS_KEY_ATTR_VLAN
) |
3533 (UINT64_C(1) << OVS_KEY_ATTR_ENCAP
));
3535 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_VLAN
)) {
3536 expected_attrs
|= (UINT64_C(1) << OVS_KEY_ATTR_VLAN
);
3538 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ENCAP
)) {
3539 expected_attrs
|= (UINT64_C(1) << OVS_KEY_ATTR_ENCAP
);
3542 fitness
= check_expectations(present_attrs
, out_of_range_attr
,
3543 expected_attrs
, key
, key_len
);
3546 * Remove the TPID from dl_type since it's not the real Ethertype. */
3547 flow
->dl_type
= htons(0);
3548 flow
->vlan_tci
= (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_VLAN
)
3549 ? nl_attr_get_be16(attrs
[OVS_KEY_ATTR_VLAN
])
3552 if (!(present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_VLAN
))) {
3553 return ODP_FIT_TOO_LITTLE
;
3554 } else if (flow
->vlan_tci
== htons(0)) {
3555 /* Corner case for a truncated 802.1Q header. */
3556 if (fitness
== ODP_FIT_PERFECT
&& nl_attr_get_size(encap
)) {
3557 return ODP_FIT_TOO_MUCH
;
3560 } else if (!(flow
->vlan_tci
& htons(VLAN_CFI
))) {
3561 VLOG_ERR_RL(&rl
, "OVS_KEY_ATTR_VLAN 0x%04"PRIx16
" is nonzero "
3562 "but CFI bit is not set", ntohs(flow
->vlan_tci
));
3563 return ODP_FIT_ERROR
;
3566 if (!(present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ENCAP
))) {
3571 /* Now parse the encapsulated attributes. */
3572 if (!parse_flow_nlattrs(nl_attr_get(encap
), nl_attr_get_size(encap
),
3573 attrs
, &present_attrs
, &out_of_range_attr
)) {
3574 return ODP_FIT_ERROR
;
3578 if (!parse_ethertype(attrs
, present_attrs
, &expected_attrs
, flow
, src_flow
)) {
3579 return ODP_FIT_ERROR
;
3581 encap_fitness
= parse_l2_5_onward(attrs
, present_attrs
, out_of_range_attr
,
3582 expected_attrs
, flow
, key
, key_len
,
3585 /* The overall fitness is the worse of the outer and inner attributes. */
3586 return MAX(fitness
, encap_fitness
);
3589 static enum odp_key_fitness
3590 odp_flow_key_to_flow__(const struct nlattr
*key
, size_t key_len
,
3591 struct flow
*flow
, const struct flow
*src_flow
)
3593 const struct nlattr
*attrs
[OVS_KEY_ATTR_MAX
+ 1];
3594 uint64_t expected_attrs
;
3595 uint64_t present_attrs
;
3596 int out_of_range_attr
;
3597 bool is_mask
= src_flow
!= flow
;
3599 memset(flow
, 0, sizeof *flow
);
3601 /* Parse attributes. */
3602 if (!parse_flow_nlattrs(key
, key_len
, attrs
, &present_attrs
,
3603 &out_of_range_attr
)) {
3604 return ODP_FIT_ERROR
;
3609 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_RECIRC_ID
)) {
3610 flow
->recirc_id
= nl_attr_get_u32(attrs
[OVS_KEY_ATTR_RECIRC_ID
]);
3611 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_RECIRC_ID
;
3612 } else if (is_mask
) {
3613 /* Always exact match recirc_id if it is not specified. */
3614 flow
->recirc_id
= UINT32_MAX
;
3617 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_DP_HASH
)) {
3618 flow
->dp_hash
= nl_attr_get_u32(attrs
[OVS_KEY_ATTR_DP_HASH
]);
3619 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_DP_HASH
;
3621 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_PRIORITY
)) {
3622 flow
->skb_priority
= nl_attr_get_u32(attrs
[OVS_KEY_ATTR_PRIORITY
]);
3623 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_PRIORITY
;
3626 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_SKB_MARK
)) {
3627 flow
->pkt_mark
= nl_attr_get_u32(attrs
[OVS_KEY_ATTR_SKB_MARK
]);
3628 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_SKB_MARK
;
3631 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_TUNNEL
)) {
3632 enum odp_key_fitness res
;
3634 res
= odp_tun_key_from_attr(attrs
[OVS_KEY_ATTR_TUNNEL
], &flow
->tunnel
);
3635 if (res
== ODP_FIT_ERROR
) {
3636 return ODP_FIT_ERROR
;
3637 } else if (res
== ODP_FIT_PERFECT
) {
3638 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_TUNNEL
;
3642 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_IN_PORT
)) {
3643 flow
->in_port
.odp_port
3644 = nl_attr_get_odp_port(attrs
[OVS_KEY_ATTR_IN_PORT
]);
3645 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_IN_PORT
;
3646 } else if (!is_mask
) {
3647 flow
->in_port
.odp_port
= ODPP_NONE
;
3650 /* Ethernet header. */
3651 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ETHERNET
)) {
3652 const struct ovs_key_ethernet
*eth_key
;
3654 eth_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_ETHERNET
]);
3655 put_ethernet_key(eth_key
, flow
);
3657 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ETHERNET
;
3661 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ETHERNET
;
3664 /* Get Ethertype or 802.1Q TPID or FLOW_DL_TYPE_NONE. */
3665 if (!parse_ethertype(attrs
, present_attrs
, &expected_attrs
, flow
,
3667 return ODP_FIT_ERROR
;
3671 ? (src_flow
->vlan_tci
& htons(VLAN_CFI
)) != 0
3672 : src_flow
->dl_type
== htons(ETH_TYPE_VLAN
)) {
3673 return parse_8021q_onward(attrs
, present_attrs
, out_of_range_attr
,
3674 expected_attrs
, flow
, key
, key_len
, src_flow
);
3677 flow
->vlan_tci
= htons(0xffff);
3678 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_VLAN
)) {
3679 flow
->vlan_tci
= nl_attr_get_be16(attrs
[OVS_KEY_ATTR_VLAN
]);
3680 expected_attrs
|= (UINT64_C(1) << OVS_KEY_ATTR_VLAN
);
3683 return parse_l2_5_onward(attrs
, present_attrs
, out_of_range_attr
,
3684 expected_attrs
, flow
, key
, key_len
, src_flow
);
3687 /* Converts the 'key_len' bytes of OVS_KEY_ATTR_* attributes in 'key' to a flow
3688 * structure in 'flow'. Returns an ODP_FIT_* value that indicates how well
3689 * 'key' fits our expectations for what a flow key should contain.
3691 * The 'in_port' will be the datapath's understanding of the port. The
3692 * caller will need to translate with odp_port_to_ofp_port() if the
3693 * OpenFlow port is needed.
3695 * This function doesn't take the packet itself as an argument because none of
3696 * the currently understood OVS_KEY_ATTR_* attributes require it. Currently,
3697 * it is always possible to infer which additional attribute(s) should appear
3698 * by looking at the attributes for lower-level protocols, e.g. if the network
3699 * protocol in OVS_KEY_ATTR_IPV4 or OVS_KEY_ATTR_IPV6 is IPPROTO_TCP then we
3700 * know that a OVS_KEY_ATTR_TCP attribute must appear and that otherwise it
3701 * must be absent. */
3702 enum odp_key_fitness
3703 odp_flow_key_to_flow(const struct nlattr
*key
, size_t key_len
,
3706 return odp_flow_key_to_flow__(key
, key_len
, flow
, flow
);
3709 /* Converts the 'key_len' bytes of OVS_KEY_ATTR_* attributes in 'key' to a mask
3710 * structure in 'mask'. 'flow' must be a previously translated flow
3711 * corresponding to 'mask'. Returns an ODP_FIT_* value that indicates how well
3712 * 'key' fits our expectations for what a flow key should contain. */
3713 enum odp_key_fitness
3714 odp_flow_key_to_mask(const struct nlattr
*key
, size_t key_len
,
3715 struct flow
*mask
, const struct flow
*flow
)
3717 return odp_flow_key_to_flow__(key
, key_len
, mask
, flow
);
3720 /* Returns 'fitness' as a string, for use in debug messages. */
3722 odp_key_fitness_to_string(enum odp_key_fitness fitness
)
3725 case ODP_FIT_PERFECT
:
3727 case ODP_FIT_TOO_MUCH
:
3729 case ODP_FIT_TOO_LITTLE
:
3730 return "too_little";
3738 /* Appends an OVS_ACTION_ATTR_USERSPACE action to 'odp_actions' that specifies
3739 * Netlink PID 'pid'. If 'userdata' is nonnull, adds a userdata attribute
3740 * whose contents are the 'userdata_size' bytes at 'userdata' and returns the
3741 * offset within 'odp_actions' of the start of the cookie. (If 'userdata' is
3742 * null, then the return value is not meaningful.) */
3744 odp_put_userspace_action(uint32_t pid
,
3745 const void *userdata
, size_t userdata_size
,
3746 odp_port_t tunnel_out_port
,
3747 struct ofpbuf
*odp_actions
)
3749 size_t userdata_ofs
;
3752 offset
= nl_msg_start_nested(odp_actions
, OVS_ACTION_ATTR_USERSPACE
);
3753 nl_msg_put_u32(odp_actions
, OVS_USERSPACE_ATTR_PID
, pid
);
3755 userdata_ofs
= ofpbuf_size(odp_actions
) + NLA_HDRLEN
;
3757 /* The OVS kernel module before OVS 1.11 and the upstream Linux kernel
3758 * module before Linux 3.10 required the userdata to be exactly 8 bytes
3761 * - The kernel rejected shorter userdata with -ERANGE.
3763 * - The kernel silently dropped userdata beyond the first 8 bytes.
3765 * Thus, for maximum compatibility, always put at least 8 bytes. (We
3766 * separately disable features that required more than 8 bytes.) */
3767 memcpy(nl_msg_put_unspec_zero(odp_actions
, OVS_USERSPACE_ATTR_USERDATA
,
3768 MAX(8, userdata_size
)),
3769 userdata
, userdata_size
);
3773 if (tunnel_out_port
!= ODPP_NONE
) {
3774 nl_msg_put_odp_port(odp_actions
, OVS_USERSPACE_ATTR_EGRESS_TUN_PORT
,
3777 nl_msg_end_nested(odp_actions
, offset
);
3779 return userdata_ofs
;
3783 odp_put_tunnel_action(const struct flow_tnl
*tunnel
,
3784 struct ofpbuf
*odp_actions
)
3786 size_t offset
= nl_msg_start_nested(odp_actions
, OVS_ACTION_ATTR_SET
);
3787 tun_key_to_attr(odp_actions
, tunnel
);
3788 nl_msg_end_nested(odp_actions
, offset
);
3792 odp_put_tnl_push_action(struct ofpbuf
*odp_actions
,
3793 struct ovs_action_push_tnl
*data
)
3795 int size
= offsetof(struct ovs_action_push_tnl
, header
);
3797 size
+= data
->header_len
;
3798 nl_msg_put_unspec(odp_actions
, OVS_ACTION_ATTR_TUNNEL_PUSH
, data
, size
);
3802 /* The commit_odp_actions() function and its helpers. */
3805 commit_set_action(struct ofpbuf
*odp_actions
, enum ovs_key_attr key_type
,
3806 const void *key
, size_t key_size
)
3808 size_t offset
= nl_msg_start_nested(odp_actions
, OVS_ACTION_ATTR_SET
);
3809 nl_msg_put_unspec(odp_actions
, key_type
, key
, key_size
);
3810 nl_msg_end_nested(odp_actions
, offset
);
3813 /* Masked set actions have a mask following the data within the netlink
3814 * attribute. The unmasked bits in the data will be cleared as the data
3815 * is copied to the action. */
3817 commit_masked_set_action(struct ofpbuf
*odp_actions
,
3818 enum ovs_key_attr key_type
,
3819 const void *key_
, const void *mask_
, size_t key_size
)
3821 size_t offset
= nl_msg_start_nested(odp_actions
,
3822 OVS_ACTION_ATTR_SET_MASKED
);
3823 char *data
= nl_msg_put_unspec_uninit(odp_actions
, key_type
, key_size
* 2);
3824 const char *key
= key_
, *mask
= mask_
;
3826 memcpy(data
+ key_size
, mask
, key_size
);
3827 /* Clear unmasked bits while copying. */
3828 while (key_size
--) {
3829 *data
++ = *key
++ & *mask
++;
3831 nl_msg_end_nested(odp_actions
, offset
);
3834 /* If any of the flow key data that ODP actions can modify are different in
3835 * 'base->tunnel' and 'flow->tunnel', appends a set_tunnel ODP action to
3836 * 'odp_actions' that change the flow tunneling information in key from
3837 * 'base->tunnel' into 'flow->tunnel', and then changes 'base->tunnel' in the
3838 * same way. In other words, operates the same as commit_odp_actions(), but
3839 * only on tunneling information. */
3841 commit_odp_tunnel_action(const struct flow
*flow
, struct flow
*base
,
3842 struct ofpbuf
*odp_actions
)
3844 /* A valid IPV4_TUNNEL must have non-zero ip_dst. */
3845 if (flow
->tunnel
.ip_dst
) {
3846 if (!memcmp(&base
->tunnel
, &flow
->tunnel
, sizeof base
->tunnel
)) {
3849 memcpy(&base
->tunnel
, &flow
->tunnel
, sizeof base
->tunnel
);
3850 odp_put_tunnel_action(&base
->tunnel
, odp_actions
);
3855 commit(enum ovs_key_attr attr
, bool use_masked_set
,
3856 const void *key
, void *base
, void *mask
, size_t size
,
3857 struct ofpbuf
*odp_actions
)
3859 if (memcmp(key
, base
, size
)) {
3860 bool fully_masked
= odp_mask_is_exact(attr
, mask
, size
);
3862 if (use_masked_set
&& !fully_masked
) {
3863 commit_masked_set_action(odp_actions
, attr
, key
, mask
, size
);
3865 if (!fully_masked
) {
3866 memset(mask
, 0xff, size
);
3868 commit_set_action(odp_actions
, attr
, key
, size
);
3870 memcpy(base
, key
, size
);
3873 /* Mask bits are set when we have either read or set the corresponding
3874 * values. Masked bits will be exact-matched, no need to set them
3875 * if the value did not actually change. */
3881 get_ethernet_key(const struct flow
*flow
, struct ovs_key_ethernet
*eth
)
3883 memcpy(eth
->eth_src
, flow
->dl_src
, ETH_ADDR_LEN
);
3884 memcpy(eth
->eth_dst
, flow
->dl_dst
, ETH_ADDR_LEN
);
3888 put_ethernet_key(const struct ovs_key_ethernet
*eth
, struct flow
*flow
)
3890 memcpy(flow
->dl_src
, eth
->eth_src
, ETH_ADDR_LEN
);
3891 memcpy(flow
->dl_dst
, eth
->eth_dst
, ETH_ADDR_LEN
);
3895 commit_set_ether_addr_action(const struct flow
*flow
, struct flow
*base_flow
,
3896 struct ofpbuf
*odp_actions
,
3897 struct flow_wildcards
*wc
,
3900 struct ovs_key_ethernet key
, base
, mask
;
3902 get_ethernet_key(flow
, &key
);
3903 get_ethernet_key(base_flow
, &base
);
3904 get_ethernet_key(&wc
->masks
, &mask
);
3906 if (commit(OVS_KEY_ATTR_ETHERNET
, use_masked
,
3907 &key
, &base
, &mask
, sizeof key
, odp_actions
)) {
3908 put_ethernet_key(&base
, base_flow
);
3909 put_ethernet_key(&mask
, &wc
->masks
);
3914 pop_vlan(struct flow
*base
,
3915 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
)
3917 memset(&wc
->masks
.vlan_tci
, 0xff, sizeof wc
->masks
.vlan_tci
);
3919 if (base
->vlan_tci
& htons(VLAN_CFI
)) {
3920 nl_msg_put_flag(odp_actions
, OVS_ACTION_ATTR_POP_VLAN
);
3926 commit_vlan_action(ovs_be16 vlan_tci
, struct flow
*base
,
3927 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
)
3929 if (base
->vlan_tci
== vlan_tci
) {
3933 pop_vlan(base
, odp_actions
, wc
);
3934 if (vlan_tci
& htons(VLAN_CFI
)) {
3935 struct ovs_action_push_vlan vlan
;
3937 vlan
.vlan_tpid
= htons(ETH_TYPE_VLAN
);
3938 vlan
.vlan_tci
= vlan_tci
;
3939 nl_msg_put_unspec(odp_actions
, OVS_ACTION_ATTR_PUSH_VLAN
,
3940 &vlan
, sizeof vlan
);
3942 base
->vlan_tci
= vlan_tci
;
3945 /* Wildcarding already done at action translation time. */
3947 commit_mpls_action(const struct flow
*flow
, struct flow
*base
,
3948 struct ofpbuf
*odp_actions
)
3950 int base_n
= flow_count_mpls_labels(base
, NULL
);
3951 int flow_n
= flow_count_mpls_labels(flow
, NULL
);
3952 int common_n
= flow_count_common_mpls_labels(flow
, flow_n
, base
, base_n
,
3955 while (base_n
> common_n
) {
3956 if (base_n
- 1 == common_n
&& flow_n
> common_n
) {
3957 /* If there is only one more LSE in base than there are common
3958 * between base and flow; and flow has at least one more LSE than
3959 * is common then the topmost LSE of base may be updated using
3961 struct ovs_key_mpls mpls_key
;
3963 mpls_key
.mpls_lse
= flow
->mpls_lse
[flow_n
- base_n
];
3964 commit_set_action(odp_actions
, OVS_KEY_ATTR_MPLS
,
3965 &mpls_key
, sizeof mpls_key
);
3966 flow_set_mpls_lse(base
, 0, mpls_key
.mpls_lse
);
3969 /* Otherwise, if there more LSEs in base than are common between
3970 * base and flow then pop the topmost one. */
3974 /* If all the LSEs are to be popped and this is not the outermost
3975 * LSE then use ETH_TYPE_MPLS as the ethertype parameter of the
3976 * POP_MPLS action instead of flow->dl_type.
3978 * This is because the POP_MPLS action requires its ethertype
3979 * argument to be an MPLS ethernet type but in this case
3980 * flow->dl_type will be a non-MPLS ethernet type.
3982 * When the final POP_MPLS action occurs it use flow->dl_type and
3983 * the and the resulting packet will have the desired dl_type. */
3984 if ((!eth_type_mpls(flow
->dl_type
)) && base_n
> 1) {
3985 dl_type
= htons(ETH_TYPE_MPLS
);
3987 dl_type
= flow
->dl_type
;
3989 nl_msg_put_be16(odp_actions
, OVS_ACTION_ATTR_POP_MPLS
, dl_type
);
3990 popped
= flow_pop_mpls(base
, base_n
, flow
->dl_type
, NULL
);
3996 /* If, after the above popping and setting, there are more LSEs in flow
3997 * than base then some LSEs need to be pushed. */
3998 while (base_n
< flow_n
) {
3999 struct ovs_action_push_mpls
*mpls
;
4001 mpls
= nl_msg_put_unspec_zero(odp_actions
,
4002 OVS_ACTION_ATTR_PUSH_MPLS
,
4004 mpls
->mpls_ethertype
= flow
->dl_type
;
4005 mpls
->mpls_lse
= flow
->mpls_lse
[flow_n
- base_n
- 1];
4006 flow_push_mpls(base
, base_n
, mpls
->mpls_ethertype
, NULL
);
4007 flow_set_mpls_lse(base
, 0, mpls
->mpls_lse
);
4013 get_ipv4_key(const struct flow
*flow
, struct ovs_key_ipv4
*ipv4
, bool is_mask
)
4015 ipv4
->ipv4_src
= flow
->nw_src
;
4016 ipv4
->ipv4_dst
= flow
->nw_dst
;
4017 ipv4
->ipv4_proto
= flow
->nw_proto
;
4018 ipv4
->ipv4_tos
= flow
->nw_tos
;
4019 ipv4
->ipv4_ttl
= flow
->nw_ttl
;
4020 ipv4
->ipv4_frag
= ovs_to_odp_frag(flow
->nw_frag
, is_mask
);
4024 put_ipv4_key(const struct ovs_key_ipv4
*ipv4
, struct flow
*flow
, bool is_mask
)
4026 flow
->nw_src
= ipv4
->ipv4_src
;
4027 flow
->nw_dst
= ipv4
->ipv4_dst
;
4028 flow
->nw_proto
= ipv4
->ipv4_proto
;
4029 flow
->nw_tos
= ipv4
->ipv4_tos
;
4030 flow
->nw_ttl
= ipv4
->ipv4_ttl
;
4031 flow
->nw_frag
= odp_to_ovs_frag(ipv4
->ipv4_frag
, is_mask
);
4035 commit_set_ipv4_action(const struct flow
*flow
, struct flow
*base_flow
,
4036 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
,
4039 struct ovs_key_ipv4 key
, mask
, base
;
4041 /* Check that nw_proto and nw_frag remain unchanged. */
4042 ovs_assert(flow
->nw_proto
== base_flow
->nw_proto
&&
4043 flow
->nw_frag
== base_flow
->nw_frag
);
4045 get_ipv4_key(flow
, &key
, false);
4046 get_ipv4_key(base_flow
, &base
, false);
4047 get_ipv4_key(&wc
->masks
, &mask
, true);
4048 mask
.ipv4_proto
= 0; /* Not writeable. */
4049 mask
.ipv4_frag
= 0; /* Not writable. */
4051 if (commit(OVS_KEY_ATTR_IPV4
, use_masked
, &key
, &base
, &mask
, sizeof key
,
4053 put_ipv4_key(&base
, base_flow
, false);
4054 if (mask
.ipv4_proto
!= 0) { /* Mask was changed by commit(). */
4055 put_ipv4_key(&mask
, &wc
->masks
, true);
4061 get_ipv6_key(const struct flow
*flow
, struct ovs_key_ipv6
*ipv6
, bool is_mask
)
4063 memcpy(ipv6
->ipv6_src
, &flow
->ipv6_src
, sizeof ipv6
->ipv6_src
);
4064 memcpy(ipv6
->ipv6_dst
, &flow
->ipv6_dst
, sizeof ipv6
->ipv6_dst
);
4065 ipv6
->ipv6_label
= flow
->ipv6_label
;
4066 ipv6
->ipv6_proto
= flow
->nw_proto
;
4067 ipv6
->ipv6_tclass
= flow
->nw_tos
;
4068 ipv6
->ipv6_hlimit
= flow
->nw_ttl
;
4069 ipv6
->ipv6_frag
= ovs_to_odp_frag(flow
->nw_frag
, is_mask
);
4073 put_ipv6_key(const struct ovs_key_ipv6
*ipv6
, struct flow
*flow
, bool is_mask
)
4075 memcpy(&flow
->ipv6_src
, ipv6
->ipv6_src
, sizeof flow
->ipv6_src
);
4076 memcpy(&flow
->ipv6_dst
, ipv6
->ipv6_dst
, sizeof flow
->ipv6_dst
);
4077 flow
->ipv6_label
= ipv6
->ipv6_label
;
4078 flow
->nw_proto
= ipv6
->ipv6_proto
;
4079 flow
->nw_tos
= ipv6
->ipv6_tclass
;
4080 flow
->nw_ttl
= ipv6
->ipv6_hlimit
;
4081 flow
->nw_frag
= odp_to_ovs_frag(ipv6
->ipv6_frag
, is_mask
);
4085 commit_set_ipv6_action(const struct flow
*flow
, struct flow
*base_flow
,
4086 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
,
4089 struct ovs_key_ipv6 key
, mask
, base
;
4091 /* Check that nw_proto and nw_frag remain unchanged. */
4092 ovs_assert(flow
->nw_proto
== base_flow
->nw_proto
&&
4093 flow
->nw_frag
== base_flow
->nw_frag
);
4095 get_ipv6_key(flow
, &key
, false);
4096 get_ipv6_key(base_flow
, &base
, false);
4097 get_ipv6_key(&wc
->masks
, &mask
, true);
4098 mask
.ipv6_proto
= 0; /* Not writeable. */
4099 mask
.ipv6_frag
= 0; /* Not writable. */
4101 if (commit(OVS_KEY_ATTR_IPV6
, use_masked
, &key
, &base
, &mask
, sizeof key
,
4103 put_ipv6_key(&base
, base_flow
, false);
4104 if (mask
.ipv6_proto
!= 0) { /* Mask was changed by commit(). */
4105 put_ipv6_key(&mask
, &wc
->masks
, true);
4111 get_arp_key(const struct flow
*flow
, struct ovs_key_arp
*arp
)
4113 /* ARP key has padding, clear it. */
4114 memset(arp
, 0, sizeof *arp
);
4116 arp
->arp_sip
= flow
->nw_src
;
4117 arp
->arp_tip
= flow
->nw_dst
;
4118 arp
->arp_op
= htons(flow
->nw_proto
);
4119 memcpy(arp
->arp_sha
, flow
->arp_sha
, ETH_ADDR_LEN
);
4120 memcpy(arp
->arp_tha
, flow
->arp_tha
, ETH_ADDR_LEN
);
4124 put_arp_key(const struct ovs_key_arp
*arp
, struct flow
*flow
)
4126 flow
->nw_src
= arp
->arp_sip
;
4127 flow
->nw_dst
= arp
->arp_tip
;
4128 flow
->nw_proto
= ntohs(arp
->arp_op
);
4129 memcpy(flow
->arp_sha
, arp
->arp_sha
, ETH_ADDR_LEN
);
4130 memcpy(flow
->arp_tha
, arp
->arp_tha
, ETH_ADDR_LEN
);
4133 static enum slow_path_reason
4134 commit_set_arp_action(const struct flow
*flow
, struct flow
*base_flow
,
4135 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
)
4137 struct ovs_key_arp key
, mask
, base
;
4139 get_arp_key(flow
, &key
);
4140 get_arp_key(base_flow
, &base
);
4141 get_arp_key(&wc
->masks
, &mask
);
4143 if (commit(OVS_KEY_ATTR_ARP
, true, &key
, &base
, &mask
, sizeof key
,
4145 put_arp_key(&base
, base_flow
);
4146 put_arp_key(&mask
, &wc
->masks
);
4152 static enum slow_path_reason
4153 commit_set_nw_action(const struct flow
*flow
, struct flow
*base
,
4154 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
,
4157 /* Check if 'flow' really has an L3 header. */
4158 if (!flow
->nw_proto
) {
4162 switch (ntohs(base
->dl_type
)) {
4164 commit_set_ipv4_action(flow
, base
, odp_actions
, wc
, use_masked
);
4168 commit_set_ipv6_action(flow
, base
, odp_actions
, wc
, use_masked
);
4172 return commit_set_arp_action(flow
, base
, odp_actions
, wc
);
4178 /* TCP, UDP, and SCTP keys have the same layout. */
4179 BUILD_ASSERT_DECL(sizeof(struct ovs_key_tcp
) == sizeof(struct ovs_key_udp
) &&
4180 sizeof(struct ovs_key_tcp
) == sizeof(struct ovs_key_sctp
));
4183 get_tp_key(const struct flow
*flow
, union ovs_key_tp
*tp
)
4185 tp
->tcp
.tcp_src
= flow
->tp_src
;
4186 tp
->tcp
.tcp_dst
= flow
->tp_dst
;
4190 put_tp_key(const union ovs_key_tp
*tp
, struct flow
*flow
)
4192 flow
->tp_src
= tp
->tcp
.tcp_src
;
4193 flow
->tp_dst
= tp
->tcp
.tcp_dst
;
4197 commit_set_port_action(const struct flow
*flow
, struct flow
*base_flow
,
4198 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
,
4201 enum ovs_key_attr key_type
;
4202 union ovs_key_tp key
, mask
, base
;
4204 /* Check if 'flow' really has an L3 header. */
4205 if (!flow
->nw_proto
) {
4209 if (!is_ip_any(base_flow
)) {
4213 if (flow
->nw_proto
== IPPROTO_TCP
) {
4214 key_type
= OVS_KEY_ATTR_TCP
;
4215 } else if (flow
->nw_proto
== IPPROTO_UDP
) {
4216 key_type
= OVS_KEY_ATTR_UDP
;
4217 } else if (flow
->nw_proto
== IPPROTO_SCTP
) {
4218 key_type
= OVS_KEY_ATTR_SCTP
;
4223 get_tp_key(flow
, &key
);
4224 get_tp_key(base_flow
, &base
);
4225 get_tp_key(&wc
->masks
, &mask
);
4227 if (commit(key_type
, use_masked
, &key
, &base
, &mask
, sizeof key
,
4229 put_tp_key(&base
, base_flow
);
4230 put_tp_key(&mask
, &wc
->masks
);
4235 commit_set_priority_action(const struct flow
*flow
, struct flow
*base_flow
,
4236 struct ofpbuf
*odp_actions
,
4237 struct flow_wildcards
*wc
,
4240 uint32_t key
, mask
, base
;
4242 key
= flow
->skb_priority
;
4243 base
= base_flow
->skb_priority
;
4244 mask
= wc
->masks
.skb_priority
;
4246 if (commit(OVS_KEY_ATTR_PRIORITY
, use_masked
, &key
, &base
, &mask
,
4247 sizeof key
, odp_actions
)) {
4248 base_flow
->skb_priority
= base
;
4249 wc
->masks
.skb_priority
= mask
;
4254 commit_set_pkt_mark_action(const struct flow
*flow
, struct flow
*base_flow
,
4255 struct ofpbuf
*odp_actions
,
4256 struct flow_wildcards
*wc
,
4259 uint32_t key
, mask
, base
;
4261 key
= flow
->pkt_mark
;
4262 base
= base_flow
->pkt_mark
;
4263 mask
= wc
->masks
.pkt_mark
;
4265 if (commit(OVS_KEY_ATTR_SKB_MARK
, use_masked
, &key
, &base
, &mask
,
4266 sizeof key
, odp_actions
)) {
4267 base_flow
->pkt_mark
= base
;
4268 wc
->masks
.pkt_mark
= mask
;
4272 /* If any of the flow key data that ODP actions can modify are different in
4273 * 'base' and 'flow', appends ODP actions to 'odp_actions' that change the flow
4274 * key from 'base' into 'flow', and then changes 'base' the same way. Does not
4275 * commit set_tunnel actions. Users should call commit_odp_tunnel_action()
4276 * in addition to this function if needed. Sets fields in 'wc' that are
4277 * used as part of the action.
4279 * Returns a reason to force processing the flow's packets into the userspace
4280 * slow path, if there is one, otherwise 0. */
4281 enum slow_path_reason
4282 commit_odp_actions(const struct flow
*flow
, struct flow
*base
,
4283 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
,
4286 enum slow_path_reason slow
;
4288 commit_set_ether_addr_action(flow
, base
, odp_actions
, wc
, use_masked
);
4289 slow
= commit_set_nw_action(flow
, base
, odp_actions
, wc
, use_masked
);
4290 commit_set_port_action(flow
, base
, odp_actions
, wc
, use_masked
);
4291 commit_mpls_action(flow
, base
, odp_actions
);
4292 commit_vlan_action(flow
->vlan_tci
, base
, odp_actions
, wc
);
4293 commit_set_priority_action(flow
, base
, odp_actions
, wc
, use_masked
);
4294 commit_set_pkt_mark_action(flow
, base
, odp_actions
, wc
, use_masked
);