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"
40 #include "openvswitch/vlog.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_udp_tnl_push_header(struct ds
*ds
, const struct ip_header
*ip
)
516 const struct udp_header
*udp
;
518 udp
= (const struct udp_header
*) (ip
+ 1);
519 ds_put_format(ds
, "udp(src=%"PRIu16
",dst=%"PRIu16
"),",
520 ntohs(udp
->udp_src
), ntohs(udp
->udp_dst
));
526 format_odp_tnl_push_header(struct ds
*ds
, struct ovs_action_push_tnl
*data
)
528 const struct eth_header
*eth
;
529 const struct ip_header
*ip
;
532 eth
= (const struct eth_header
*)data
->header
;
535 ip
= (const struct ip_header
*)l3
;
538 ds_put_format(ds
, "header(size=%"PRIu8
",type=%"PRIu8
",eth(dst=",
539 data
->header_len
, data
->tnl_type
);
540 ds_put_format(ds
, ETH_ADDR_FMT
, ETH_ADDR_ARGS(eth
->eth_dst
));
541 ds_put_format(ds
, ",src=");
542 ds_put_format(ds
, ETH_ADDR_FMT
, ETH_ADDR_ARGS(eth
->eth_src
));
543 ds_put_format(ds
, ",dl_type=0x%04"PRIx16
"),", ntohs(eth
->eth_type
));
546 ds_put_format(ds
, "ipv4(src="IP_FMT
",dst="IP_FMT
",proto=%"PRIu8
547 ",tos=%#"PRIx8
",ttl=%"PRIu8
",frag=0x%"PRIx16
"),",
548 IP_ARGS(get_16aligned_be32(&ip
->ip_src
)),
549 IP_ARGS(get_16aligned_be32(&ip
->ip_dst
)),
550 ip
->ip_proto
, ip
->ip_tos
,
554 if (data
->tnl_type
== OVS_VPORT_TYPE_VXLAN
) {
555 const struct vxlanhdr
*vxh
;
557 vxh
= format_udp_tnl_push_header(ds
, ip
);
559 ds_put_format(ds
, "vxlan(flags=0x%"PRIx32
",vni=0x%"PRIx32
")",
560 ntohl(get_16aligned_be32(&vxh
->vx_flags
)),
561 ntohl(get_16aligned_be32(&vxh
->vx_vni
)) >> 8);
562 } else if (data
->tnl_type
== OVS_VPORT_TYPE_GENEVE
) {
563 const struct genevehdr
*gnh
;
565 gnh
= format_udp_tnl_push_header(ds
, ip
);
567 ds_put_format(ds
, "geneve(%svni=0x%"PRIx32
")",
568 gnh
->oam
? "oam," : "",
569 ntohl(get_16aligned_be32(&gnh
->vni
)) >> 8);
570 } else if (data
->tnl_type
== OVS_VPORT_TYPE_GRE
) {
571 const struct gre_base_hdr
*greh
;
572 ovs_16aligned_be32
*options
;
575 l4
= ((uint8_t *)l3
+ sizeof(struct ip_header
));
576 greh
= (const struct gre_base_hdr
*) l4
;
578 ds_put_format(ds
, "gre((flags=0x%"PRIx16
",proto=0x%"PRIx16
")",
579 greh
->flags
, ntohs(greh
->protocol
));
580 options
= (ovs_16aligned_be32
*)(greh
+ 1);
581 if (greh
->flags
& htons(GRE_CSUM
)) {
582 ds_put_format(ds
, ",csum=0x%"PRIx16
, ntohs(*((ovs_be16
*)options
)));
585 if (greh
->flags
& htons(GRE_KEY
)) {
586 ds_put_format(ds
, ",key=0x%"PRIx32
, ntohl(get_16aligned_be32(options
)));
589 if (greh
->flags
& htons(GRE_SEQ
)) {
590 ds_put_format(ds
, ",seq=0x%"PRIx32
, ntohl(get_16aligned_be32(options
)));
593 ds_put_format(ds
, ")");
595 ds_put_format(ds
, ")");
599 format_odp_tnl_push_action(struct ds
*ds
, const struct nlattr
*attr
)
601 struct ovs_action_push_tnl
*data
;
603 data
= (struct ovs_action_push_tnl
*) nl_attr_get(attr
);
605 ds_put_format(ds
, "tnl_push(tnl_port(%"PRIu32
"),", data
->tnl_port
);
606 format_odp_tnl_push_header(ds
, data
);
607 ds_put_format(ds
, ",out_port(%"PRIu32
"))", data
->out_port
);
611 format_odp_action(struct ds
*ds
, const struct nlattr
*a
)
614 enum ovs_action_attr type
= nl_attr_type(a
);
615 const struct ovs_action_push_vlan
*vlan
;
618 expected_len
= odp_action_len(nl_attr_type(a
));
619 if (expected_len
!= -2 && nl_attr_get_size(a
) != expected_len
) {
620 ds_put_format(ds
, "bad length %"PRIuSIZE
", expected %d for: ",
621 nl_attr_get_size(a
), expected_len
);
622 format_generic_odp_action(ds
, a
);
627 case OVS_ACTION_ATTR_OUTPUT
:
628 ds_put_format(ds
, "%"PRIu32
, nl_attr_get_u32(a
));
630 case OVS_ACTION_ATTR_TUNNEL_POP
:
631 ds_put_format(ds
, "tnl_pop(%"PRIu32
")", nl_attr_get_u32(a
));
633 case OVS_ACTION_ATTR_TUNNEL_PUSH
:
634 format_odp_tnl_push_action(ds
, a
);
636 case OVS_ACTION_ATTR_USERSPACE
:
637 format_odp_userspace_action(ds
, a
);
639 case OVS_ACTION_ATTR_RECIRC
:
640 format_odp_recirc_action(ds
, nl_attr_get_u32(a
));
642 case OVS_ACTION_ATTR_HASH
:
643 format_odp_hash_action(ds
, nl_attr_get(a
));
645 case OVS_ACTION_ATTR_SET_MASKED
:
647 size
= nl_attr_get_size(a
) / 2;
648 ds_put_cstr(ds
, "set(");
650 /* Masked set action not supported for tunnel key, which is bigger. */
651 if (size
<= sizeof(struct ovs_key_ipv6
)) {
652 struct nlattr attr
[1 + DIV_ROUND_UP(sizeof(struct ovs_key_ipv6
),
653 sizeof(struct nlattr
))];
654 struct nlattr mask
[1 + DIV_ROUND_UP(sizeof(struct ovs_key_ipv6
),
655 sizeof(struct nlattr
))];
657 mask
->nla_type
= attr
->nla_type
= nl_attr_type(a
);
658 mask
->nla_len
= attr
->nla_len
= NLA_HDRLEN
+ size
;
659 memcpy(attr
+ 1, (char *)(a
+ 1), size
);
660 memcpy(mask
+ 1, (char *)(a
+ 1) + size
, size
);
661 format_odp_key_attr(attr
, mask
, NULL
, ds
, false);
663 format_odp_key_attr(a
, NULL
, NULL
, ds
, false);
665 ds_put_cstr(ds
, ")");
667 case OVS_ACTION_ATTR_SET
:
668 ds_put_cstr(ds
, "set(");
669 format_odp_key_attr(nl_attr_get(a
), NULL
, NULL
, ds
, true);
670 ds_put_cstr(ds
, ")");
672 case OVS_ACTION_ATTR_PUSH_VLAN
:
673 vlan
= nl_attr_get(a
);
674 ds_put_cstr(ds
, "push_vlan(");
675 if (vlan
->vlan_tpid
!= htons(ETH_TYPE_VLAN
)) {
676 ds_put_format(ds
, "tpid=0x%04"PRIx16
",", ntohs(vlan
->vlan_tpid
));
678 format_vlan_tci(ds
, vlan
->vlan_tci
, OVS_BE16_MAX
, false);
679 ds_put_char(ds
, ')');
681 case OVS_ACTION_ATTR_POP_VLAN
:
682 ds_put_cstr(ds
, "pop_vlan");
684 case OVS_ACTION_ATTR_PUSH_MPLS
: {
685 const struct ovs_action_push_mpls
*mpls
= nl_attr_get(a
);
686 ds_put_cstr(ds
, "push_mpls(");
687 format_mpls_lse(ds
, mpls
->mpls_lse
);
688 ds_put_format(ds
, ",eth_type=0x%"PRIx16
")", ntohs(mpls
->mpls_ethertype
));
691 case OVS_ACTION_ATTR_POP_MPLS
: {
692 ovs_be16 ethertype
= nl_attr_get_be16(a
);
693 ds_put_format(ds
, "pop_mpls(eth_type=0x%"PRIx16
")", ntohs(ethertype
));
696 case OVS_ACTION_ATTR_SAMPLE
:
697 format_odp_sample_action(ds
, a
);
699 case OVS_ACTION_ATTR_UNSPEC
:
700 case __OVS_ACTION_ATTR_MAX
:
702 format_generic_odp_action(ds
, a
);
708 format_odp_actions(struct ds
*ds
, const struct nlattr
*actions
,
712 const struct nlattr
*a
;
715 NL_ATTR_FOR_EACH (a
, left
, actions
, actions_len
) {
717 ds_put_char(ds
, ',');
719 format_odp_action(ds
, a
);
724 if (left
== actions_len
) {
725 ds_put_cstr(ds
, "<empty>");
727 ds_put_format(ds
, ",***%u leftover bytes*** (", left
);
728 for (i
= 0; i
< left
; i
++) {
729 ds_put_format(ds
, "%02x", ((const uint8_t *) a
)[i
]);
731 ds_put_char(ds
, ')');
734 ds_put_cstr(ds
, "drop");
738 /* Separate out parse_odp_userspace_action() function. */
740 parse_odp_userspace_action(const char *s
, struct ofpbuf
*actions
)
743 union user_action_cookie cookie
;
745 odp_port_t tunnel_out_port
;
747 void *user_data
= NULL
;
748 size_t user_data_size
= 0;
750 if (!ovs_scan(s
, "userspace(pid=%"SCNi32
"%n", &pid
, &n
)) {
756 uint32_t probability
;
757 uint32_t collector_set_id
;
758 uint32_t obs_domain_id
;
759 uint32_t obs_point_id
;
762 if (ovs_scan(&s
[n
], ",sFlow(vid=%i,"
763 "pcp=%i,output=%"SCNi32
")%n",
764 &vid
, &pcp
, &output
, &n1
)) {
768 tci
= vid
| (pcp
<< VLAN_PCP_SHIFT
);
773 cookie
.type
= USER_ACTION_COOKIE_SFLOW
;
774 cookie
.sflow
.vlan_tci
= htons(tci
);
775 cookie
.sflow
.output
= output
;
777 user_data_size
= sizeof cookie
.sflow
;
778 } else if (ovs_scan(&s
[n
], ",slow_path(%n",
783 cookie
.type
= USER_ACTION_COOKIE_SLOW_PATH
;
784 cookie
.slow_path
.unused
= 0;
785 cookie
.slow_path
.reason
= 0;
787 res
= parse_flags(&s
[n
], slow_path_reason_to_string
,
788 &cookie
.slow_path
.reason
,
789 SLOW_PATH_REASON_MASK
, NULL
);
790 if (res
< 0 || s
[n
+ res
] != ')') {
796 user_data_size
= sizeof cookie
.slow_path
;
797 } else if (ovs_scan(&s
[n
], ",flow_sample(probability=%"SCNi32
","
798 "collector_set_id=%"SCNi32
","
799 "obs_domain_id=%"SCNi32
","
800 "obs_point_id=%"SCNi32
")%n",
801 &probability
, &collector_set_id
,
802 &obs_domain_id
, &obs_point_id
, &n1
)) {
805 cookie
.type
= USER_ACTION_COOKIE_FLOW_SAMPLE
;
806 cookie
.flow_sample
.probability
= probability
;
807 cookie
.flow_sample
.collector_set_id
= collector_set_id
;
808 cookie
.flow_sample
.obs_domain_id
= obs_domain_id
;
809 cookie
.flow_sample
.obs_point_id
= obs_point_id
;
811 user_data_size
= sizeof cookie
.flow_sample
;
812 } else if (ovs_scan(&s
[n
], ",ipfix(output_port=%"SCNi32
")%n",
815 cookie
.type
= USER_ACTION_COOKIE_IPFIX
;
816 cookie
.ipfix
.output_odp_port
= u32_to_odp(output
);
818 user_data_size
= sizeof cookie
.ipfix
;
819 } else if (ovs_scan(&s
[n
], ",userdata(%n",
824 ofpbuf_init(&buf
, 16);
825 end
= ofpbuf_put_hex(&buf
, &s
[n
], NULL
);
829 user_data
= buf
.data
;
830 user_data_size
= buf
.size
;
837 if (ovs_scan(&s
[n
], ",tunnel_out_port=%"SCNi32
")%n",
838 &tunnel_out_port
, &n1
)) {
839 odp_put_userspace_action(pid
, user_data
, user_data_size
, tunnel_out_port
, actions
);
841 } else if (s
[n
] == ')') {
842 odp_put_userspace_action(pid
, user_data
, user_data_size
, ODPP_NONE
, actions
);
851 ovs_parse_tnl_push(const char *s
, struct ovs_action_push_tnl
*data
)
853 struct eth_header
*eth
;
854 struct ip_header
*ip
;
855 struct udp_header
*udp
;
856 struct gre_base_hdr
*greh
;
857 uint16_t gre_proto
, dl_type
, udp_src
, udp_dst
;
859 uint32_t tnl_type
= 0, header_len
= 0;
863 if (!ovs_scan_len(s
, &n
, "tnl_push(tnl_port(%"SCNi32
"),", &data
->tnl_port
)) {
866 eth
= (struct eth_header
*) data
->header
;
867 l3
= (data
->header
+ sizeof *eth
);
868 l4
= ((uint8_t *) l3
+ sizeof (struct ip_header
));
869 ip
= (struct ip_header
*) l3
;
870 if (!ovs_scan_len(s
, &n
, "header(size=%"SCNi32
",type=%"SCNi32
","
871 "eth(dst="ETH_ADDR_SCAN_FMT
",",
874 ETH_ADDR_SCAN_ARGS(eth
->eth_dst
))) {
878 if (!ovs_scan_len(s
, &n
, "src="ETH_ADDR_SCAN_FMT
",",
879 ETH_ADDR_SCAN_ARGS(eth
->eth_src
))) {
882 if (!ovs_scan_len(s
, &n
, "dl_type=0x%"SCNx16
"),", &dl_type
)) {
885 eth
->eth_type
= htons(dl_type
);
888 if (!ovs_scan_len(s
, &n
, "ipv4(src="IP_SCAN_FMT
",dst="IP_SCAN_FMT
",proto=%"SCNi8
889 ",tos=%"SCNi8
",ttl=%"SCNi8
",frag=0x%"SCNx16
"),",
892 &ip
->ip_proto
, &ip
->ip_tos
,
893 &ip
->ip_ttl
, &ip
->ip_frag_off
)) {
896 put_16aligned_be32(&ip
->ip_src
, sip
);
897 put_16aligned_be32(&ip
->ip_dst
, dip
);
900 udp
= (struct udp_header
*) l4
;
901 greh
= (struct gre_base_hdr
*) l4
;
902 if (ovs_scan_len(s
, &n
, "udp(src=%"SCNi16
",dst=%"SCNi16
"),",
903 &udp_src
, &udp_dst
)) {
904 uint32_t vx_flags
, vni
;
906 udp
->udp_src
= htons(udp_src
);
907 udp
->udp_dst
= htons(udp_dst
);
911 if (ovs_scan_len(s
, &n
, "vxlan(flags=0x%"SCNx32
",vni=0x%"SCNx32
"))",
913 struct vxlanhdr
*vxh
= (struct vxlanhdr
*) (udp
+ 1);
915 put_16aligned_be32(&vxh
->vx_flags
, htonl(vx_flags
));
916 put_16aligned_be32(&vxh
->vx_vni
, htonl(vni
<< 8));
917 tnl_type
= OVS_VPORT_TYPE_VXLAN
;
918 header_len
= sizeof *eth
+ sizeof *ip
+
919 sizeof *udp
+ sizeof *vxh
;
920 } else if (ovs_scan_len(s
, &n
, "geneve(")) {
921 struct genevehdr
*gnh
= (struct genevehdr
*) (udp
+ 1);
923 if (ovs_scan_len(s
, &n
, "oam,")) {
926 if (!ovs_scan_len(s
, &n
, "vni=0x%"SCNx32
"))", &vni
)) {
929 gnh
->proto_type
= htons(ETH_TYPE_TEB
);
930 put_16aligned_be32(&gnh
->vni
, htonl(vni
<< 8));
931 tnl_type
= OVS_VPORT_TYPE_GENEVE
;
932 header_len
= sizeof *eth
+ sizeof *ip
+
933 sizeof *udp
+ sizeof *gnh
;
937 } else if (ovs_scan_len(s
, &n
, "gre((flags=0x%"SCNx16
",proto=0x%"SCNx16
")",
938 &greh
->flags
, &gre_proto
)){
940 tnl_type
= OVS_VPORT_TYPE_GRE
;
941 greh
->protocol
= htons(gre_proto
);
942 ovs_16aligned_be32
*options
= (ovs_16aligned_be32
*) (greh
+ 1);
944 if (greh
->flags
& htons(GRE_CSUM
)) {
947 if (!ovs_scan_len(s
, &n
, ",csum=0x%"SCNx16
, &csum
)) {
951 memset(options
, 0, sizeof *options
);
952 *((ovs_be16
*)options
) = htons(csum
);
955 if (greh
->flags
& htons(GRE_KEY
)) {
958 if (!ovs_scan_len(s
, &n
, ",key=0x%"SCNx32
, &key
)) {
962 put_16aligned_be32(options
, htonl(key
));
965 if (greh
->flags
& htons(GRE_SEQ
)) {
968 if (!ovs_scan_len(s
, &n
, ",seq=0x%"SCNx32
, &seq
)) {
971 put_16aligned_be32(options
, htonl(seq
));
975 if (!ovs_scan_len(s
, &n
, "))")) {
979 header_len
= sizeof *eth
+ sizeof *ip
+
980 ((uint8_t *) options
- (uint8_t *) greh
);
985 /* check tunnel meta data. */
986 if (data
->tnl_type
!= tnl_type
) {
989 if (data
->header_len
!= header_len
) {
994 if (!ovs_scan_len(s
, &n
, ",out_port(%"SCNi32
"))", &data
->out_port
)) {
1002 parse_odp_action(const char *s
, const struct simap
*port_names
,
1003 struct ofpbuf
*actions
)
1009 if (ovs_scan(s
, "%"SCNi32
"%n", &port
, &n
)) {
1010 nl_msg_put_u32(actions
, OVS_ACTION_ATTR_OUTPUT
, port
);
1016 int len
= strcspn(s
, delimiters
);
1017 struct simap_node
*node
;
1019 node
= simap_find_len(port_names
, s
, len
);
1021 nl_msg_put_u32(actions
, OVS_ACTION_ATTR_OUTPUT
, node
->data
);
1030 if (ovs_scan(s
, "recirc(%"PRIu32
")%n", &recirc_id
, &n
)) {
1031 nl_msg_put_u32(actions
, OVS_ACTION_ATTR_RECIRC
, recirc_id
);
1036 if (!strncmp(s
, "userspace(", 10)) {
1037 return parse_odp_userspace_action(s
, actions
);
1040 if (!strncmp(s
, "set(", 4)) {
1043 struct nlattr mask
[128 / sizeof(struct nlattr
)];
1044 struct ofpbuf maskbuf
;
1045 struct nlattr
*nested
, *key
;
1048 /* 'mask' is big enough to hold any key. */
1049 ofpbuf_use_stack(&maskbuf
, mask
, sizeof mask
);
1051 start_ofs
= nl_msg_start_nested(actions
, OVS_ACTION_ATTR_SET
);
1052 retval
= parse_odp_key_mask_attr(s
+ 4, port_names
, actions
, &maskbuf
);
1056 if (s
[retval
+ 4] != ')') {
1060 nested
= ofpbuf_at_assert(actions
, start_ofs
, sizeof *nested
);
1063 size
= nl_attr_get_size(mask
);
1064 if (size
== nl_attr_get_size(key
)) {
1065 /* Change to masked set action if not fully masked. */
1066 if (!is_all_ones(mask
+ 1, size
)) {
1067 key
->nla_len
+= size
;
1068 ofpbuf_put(actions
, mask
+ 1, size
);
1069 /* 'actions' may have been reallocated by ofpbuf_put(). */
1070 nested
= ofpbuf_at_assert(actions
, start_ofs
, sizeof *nested
);
1071 nested
->nla_type
= OVS_ACTION_ATTR_SET_MASKED
;
1075 nl_msg_end_nested(actions
, start_ofs
);
1080 struct ovs_action_push_vlan push
;
1081 int tpid
= ETH_TYPE_VLAN
;
1086 if (ovs_scan(s
, "push_vlan(vid=%i,pcp=%i)%n", &vid
, &pcp
, &n
)
1087 || ovs_scan(s
, "push_vlan(vid=%i,pcp=%i,cfi=%i)%n",
1088 &vid
, &pcp
, &cfi
, &n
)
1089 || ovs_scan(s
, "push_vlan(tpid=%i,vid=%i,pcp=%i)%n",
1090 &tpid
, &vid
, &pcp
, &n
)
1091 || ovs_scan(s
, "push_vlan(tpid=%i,vid=%i,pcp=%i,cfi=%i)%n",
1092 &tpid
, &vid
, &pcp
, &cfi
, &n
)) {
1093 push
.vlan_tpid
= htons(tpid
);
1094 push
.vlan_tci
= htons((vid
<< VLAN_VID_SHIFT
)
1095 | (pcp
<< VLAN_PCP_SHIFT
)
1096 | (cfi
? VLAN_CFI
: 0));
1097 nl_msg_put_unspec(actions
, OVS_ACTION_ATTR_PUSH_VLAN
,
1098 &push
, sizeof push
);
1104 if (!strncmp(s
, "pop_vlan", 8)) {
1105 nl_msg_put_flag(actions
, OVS_ACTION_ATTR_POP_VLAN
);
1113 if (ovs_scan(s
, "sample(sample=%lf%%,actions(%n", &percentage
, &n
)
1114 && percentage
>= 0. && percentage
<= 100.0) {
1115 size_t sample_ofs
, actions_ofs
;
1118 probability
= floor(UINT32_MAX
* (percentage
/ 100.0) + .5);
1119 sample_ofs
= nl_msg_start_nested(actions
, OVS_ACTION_ATTR_SAMPLE
);
1120 nl_msg_put_u32(actions
, OVS_SAMPLE_ATTR_PROBABILITY
,
1121 (probability
<= 0 ? 0
1122 : probability
>= UINT32_MAX
? UINT32_MAX
1125 actions_ofs
= nl_msg_start_nested(actions
,
1126 OVS_SAMPLE_ATTR_ACTIONS
);
1130 n
+= strspn(s
+ n
, delimiters
);
1135 retval
= parse_odp_action(s
+ n
, port_names
, actions
);
1141 nl_msg_end_nested(actions
, actions_ofs
);
1142 nl_msg_end_nested(actions
, sample_ofs
);
1144 return s
[n
+ 1] == ')' ? n
+ 2 : -EINVAL
;
1152 if (ovs_scan(s
, "tnl_pop(%"SCNi32
")%n", &port
, &n
)) {
1153 nl_msg_put_u32(actions
, OVS_ACTION_ATTR_TUNNEL_POP
, port
);
1159 struct ovs_action_push_tnl data
;
1162 n
= ovs_parse_tnl_push(s
, &data
);
1164 odp_put_tnl_push_action(actions
, &data
);
1173 /* Parses the string representation of datapath actions, in the format output
1174 * by format_odp_action(). Returns 0 if successful, otherwise a positive errno
1175 * value. On success, the ODP actions are appended to 'actions' as a series of
1176 * Netlink attributes. On failure, no data is appended to 'actions'. Either
1177 * way, 'actions''s data might be reallocated. */
1179 odp_actions_from_string(const char *s
, const struct simap
*port_names
,
1180 struct ofpbuf
*actions
)
1184 if (!strcasecmp(s
, "drop")) {
1188 old_size
= actions
->size
;
1192 s
+= strspn(s
, delimiters
);
1197 retval
= parse_odp_action(s
, port_names
, actions
);
1198 if (retval
< 0 || !strchr(delimiters
, s
[retval
])) {
1199 actions
->size
= old_size
;
1208 /* Returns the correct length of the payload for a flow key attribute of the
1209 * specified 'type', -1 if 'type' is unknown, or -2 if the attribute's payload
1210 * is variable length. */
1212 odp_flow_key_attr_len(uint16_t type
)
1214 if (type
> OVS_KEY_ATTR_MAX
) {
1218 switch ((enum ovs_key_attr
) type
) {
1219 case OVS_KEY_ATTR_ENCAP
: return -2;
1220 case OVS_KEY_ATTR_PRIORITY
: return 4;
1221 case OVS_KEY_ATTR_SKB_MARK
: return 4;
1222 case OVS_KEY_ATTR_DP_HASH
: return 4;
1223 case OVS_KEY_ATTR_RECIRC_ID
: return 4;
1224 case OVS_KEY_ATTR_TUNNEL
: return -2;
1225 case OVS_KEY_ATTR_IN_PORT
: return 4;
1226 case OVS_KEY_ATTR_ETHERNET
: return sizeof(struct ovs_key_ethernet
);
1227 case OVS_KEY_ATTR_VLAN
: return sizeof(ovs_be16
);
1228 case OVS_KEY_ATTR_ETHERTYPE
: return 2;
1229 case OVS_KEY_ATTR_MPLS
: return -2;
1230 case OVS_KEY_ATTR_IPV4
: return sizeof(struct ovs_key_ipv4
);
1231 case OVS_KEY_ATTR_IPV6
: return sizeof(struct ovs_key_ipv6
);
1232 case OVS_KEY_ATTR_TCP
: return sizeof(struct ovs_key_tcp
);
1233 case OVS_KEY_ATTR_TCP_FLAGS
: return 2;
1234 case OVS_KEY_ATTR_UDP
: return sizeof(struct ovs_key_udp
);
1235 case OVS_KEY_ATTR_SCTP
: return sizeof(struct ovs_key_sctp
);
1236 case OVS_KEY_ATTR_ICMP
: return sizeof(struct ovs_key_icmp
);
1237 case OVS_KEY_ATTR_ICMPV6
: return sizeof(struct ovs_key_icmpv6
);
1238 case OVS_KEY_ATTR_ARP
: return sizeof(struct ovs_key_arp
);
1239 case OVS_KEY_ATTR_ND
: return sizeof(struct ovs_key_nd
);
1241 case OVS_KEY_ATTR_UNSPEC
:
1242 case __OVS_KEY_ATTR_MAX
:
1250 format_generic_odp_key(const struct nlattr
*a
, struct ds
*ds
)
1252 size_t len
= nl_attr_get_size(a
);
1254 const uint8_t *unspec
;
1257 unspec
= nl_attr_get(a
);
1258 for (i
= 0; i
< len
; i
++) {
1260 ds_put_char(ds
, ' ');
1262 ds_put_format(ds
, "%02x", unspec
[i
]);
1268 ovs_frag_type_to_string(enum ovs_frag_type type
)
1271 case OVS_FRAG_TYPE_NONE
:
1273 case OVS_FRAG_TYPE_FIRST
:
1275 case OVS_FRAG_TYPE_LATER
:
1277 case __OVS_FRAG_TYPE_MAX
:
1284 tunnel_key_attr_len(int type
)
1287 case OVS_TUNNEL_KEY_ATTR_ID
: return 8;
1288 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC
: return 4;
1289 case OVS_TUNNEL_KEY_ATTR_IPV4_DST
: return 4;
1290 case OVS_TUNNEL_KEY_ATTR_TOS
: return 1;
1291 case OVS_TUNNEL_KEY_ATTR_TTL
: return 1;
1292 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT
: return 0;
1293 case OVS_TUNNEL_KEY_ATTR_CSUM
: return 0;
1294 case OVS_TUNNEL_KEY_ATTR_TP_SRC
: return 2;
1295 case OVS_TUNNEL_KEY_ATTR_TP_DST
: return 2;
1296 case OVS_TUNNEL_KEY_ATTR_OAM
: return 0;
1297 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS
: return -2;
1298 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS
: return -2;
1299 case __OVS_TUNNEL_KEY_ATTR_MAX
:
1305 #define GENEVE_OPT(class, type) ((OVS_FORCE uint32_t)(class) << 8 | (type))
1307 parse_geneve_opts(const struct nlattr
*attr
)
1309 int opts_len
= nl_attr_get_size(attr
);
1310 const struct geneve_opt
*opt
= nl_attr_get(attr
);
1312 while (opts_len
> 0) {
1315 if (opts_len
< sizeof(*opt
)) {
1319 len
= sizeof(*opt
) + opt
->length
* 4;
1320 if (len
> opts_len
) {
1324 switch (GENEVE_OPT(opt
->opt_class
, opt
->type
)) {
1326 if (opt
->type
& GENEVE_CRIT_OPT_TYPE
) {
1331 opt
= opt
+ len
/ sizeof(*opt
);
1338 enum odp_key_fitness
1339 odp_tun_key_from_attr(const struct nlattr
*attr
, struct flow_tnl
*tun
)
1342 const struct nlattr
*a
;
1344 bool unknown
= false;
1346 NL_NESTED_FOR_EACH(a
, left
, attr
) {
1347 uint16_t type
= nl_attr_type(a
);
1348 size_t len
= nl_attr_get_size(a
);
1349 int expected_len
= tunnel_key_attr_len(type
);
1351 if (len
!= expected_len
&& expected_len
>= 0) {
1352 return ODP_FIT_ERROR
;
1356 case OVS_TUNNEL_KEY_ATTR_ID
:
1357 tun
->tun_id
= nl_attr_get_be64(a
);
1358 tun
->flags
|= FLOW_TNL_F_KEY
;
1360 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC
:
1361 tun
->ip_src
= nl_attr_get_be32(a
);
1363 case OVS_TUNNEL_KEY_ATTR_IPV4_DST
:
1364 tun
->ip_dst
= nl_attr_get_be32(a
);
1366 case OVS_TUNNEL_KEY_ATTR_TOS
:
1367 tun
->ip_tos
= nl_attr_get_u8(a
);
1369 case OVS_TUNNEL_KEY_ATTR_TTL
:
1370 tun
->ip_ttl
= nl_attr_get_u8(a
);
1373 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT
:
1374 tun
->flags
|= FLOW_TNL_F_DONT_FRAGMENT
;
1376 case OVS_TUNNEL_KEY_ATTR_CSUM
:
1377 tun
->flags
|= FLOW_TNL_F_CSUM
;
1379 case OVS_TUNNEL_KEY_ATTR_TP_SRC
:
1380 tun
->tp_src
= nl_attr_get_be16(a
);
1382 case OVS_TUNNEL_KEY_ATTR_TP_DST
:
1383 tun
->tp_dst
= nl_attr_get_be16(a
);
1385 case OVS_TUNNEL_KEY_ATTR_OAM
:
1386 tun
->flags
|= FLOW_TNL_F_OAM
;
1388 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS
: {
1389 static const struct nl_policy vxlan_opts_policy
[] = {
1390 [OVS_VXLAN_EXT_GBP
] = { .type
= NL_A_U32
},
1392 struct nlattr
*ext
[ARRAY_SIZE(vxlan_opts_policy
)];
1394 if (!nl_parse_nested(a
, vxlan_opts_policy
, ext
, ARRAY_SIZE(ext
))) {
1395 return ODP_FIT_ERROR
;
1398 if (ext
[OVS_VXLAN_EXT_GBP
]) {
1399 uint32_t gbp
= nl_attr_get_u32(ext
[OVS_VXLAN_EXT_GBP
]);
1401 tun
->gbp_id
= htons(gbp
& 0xFFFF);
1402 tun
->gbp_flags
= (gbp
>> 16) & 0xFF;
1407 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS
: {
1408 if (parse_geneve_opts(a
)) {
1409 return ODP_FIT_ERROR
;
1411 /* It is necessary to reproduce options exactly (including order)
1412 * so it's easiest to just echo them back. */
1417 /* Allow this to show up as unexpected, if there are unknown
1418 * tunnel attribute, eventually resulting in ODP_FIT_TOO_MUCH. */
1425 return ODP_FIT_ERROR
;
1428 return ODP_FIT_TOO_MUCH
;
1430 return ODP_FIT_PERFECT
;
1434 tun_key_to_attr(struct ofpbuf
*a
, const struct flow_tnl
*tun_key
)
1438 tun_key_ofs
= nl_msg_start_nested(a
, OVS_KEY_ATTR_TUNNEL
);
1440 /* tun_id != 0 without FLOW_TNL_F_KEY is valid if tun_key is a mask. */
1441 if (tun_key
->tun_id
|| tun_key
->flags
& FLOW_TNL_F_KEY
) {
1442 nl_msg_put_be64(a
, OVS_TUNNEL_KEY_ATTR_ID
, tun_key
->tun_id
);
1444 if (tun_key
->ip_src
) {
1445 nl_msg_put_be32(a
, OVS_TUNNEL_KEY_ATTR_IPV4_SRC
, tun_key
->ip_src
);
1447 if (tun_key
->ip_dst
) {
1448 nl_msg_put_be32(a
, OVS_TUNNEL_KEY_ATTR_IPV4_DST
, tun_key
->ip_dst
);
1450 if (tun_key
->ip_tos
) {
1451 nl_msg_put_u8(a
, OVS_TUNNEL_KEY_ATTR_TOS
, tun_key
->ip_tos
);
1453 nl_msg_put_u8(a
, OVS_TUNNEL_KEY_ATTR_TTL
, tun_key
->ip_ttl
);
1454 if (tun_key
->flags
& FLOW_TNL_F_DONT_FRAGMENT
) {
1455 nl_msg_put_flag(a
, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT
);
1457 if (tun_key
->flags
& FLOW_TNL_F_CSUM
) {
1458 nl_msg_put_flag(a
, OVS_TUNNEL_KEY_ATTR_CSUM
);
1460 if (tun_key
->tp_src
) {
1461 nl_msg_put_be16(a
, OVS_TUNNEL_KEY_ATTR_TP_SRC
, tun_key
->tp_src
);
1463 if (tun_key
->tp_dst
) {
1464 nl_msg_put_be16(a
, OVS_TUNNEL_KEY_ATTR_TP_DST
, tun_key
->tp_dst
);
1466 if (tun_key
->flags
& FLOW_TNL_F_OAM
) {
1467 nl_msg_put_flag(a
, OVS_TUNNEL_KEY_ATTR_OAM
);
1469 if (tun_key
->gbp_flags
|| tun_key
->gbp_id
) {
1470 size_t vxlan_opts_ofs
;
1472 vxlan_opts_ofs
= nl_msg_start_nested(a
, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS
);
1473 nl_msg_put_u32(a
, OVS_VXLAN_EXT_GBP
,
1474 (tun_key
->gbp_flags
<< 16) | ntohs(tun_key
->gbp_id
));
1475 nl_msg_end_nested(a
, vxlan_opts_ofs
);
1478 nl_msg_end_nested(a
, tun_key_ofs
);
1482 odp_mask_attr_is_wildcard(const struct nlattr
*ma
)
1484 return is_all_zeros(nl_attr_get(ma
), nl_attr_get_size(ma
));
1488 odp_mask_is_exact(enum ovs_key_attr attr
, const void *mask
, size_t size
)
1490 if (attr
== OVS_KEY_ATTR_TCP_FLAGS
) {
1491 return TCP_FLAGS(*(ovs_be16
*)mask
) == TCP_FLAGS(OVS_BE16_MAX
);
1493 if (attr
== OVS_KEY_ATTR_IPV6
) {
1494 const struct ovs_key_ipv6
*ipv6_mask
= mask
;
1497 ((ipv6_mask
->ipv6_label
& htonl(IPV6_LABEL_MASK
))
1498 == htonl(IPV6_LABEL_MASK
))
1499 && ipv6_mask
->ipv6_proto
== UINT8_MAX
1500 && ipv6_mask
->ipv6_tclass
== UINT8_MAX
1501 && ipv6_mask
->ipv6_hlimit
== UINT8_MAX
1502 && ipv6_mask
->ipv6_frag
== UINT8_MAX
1503 && ipv6_mask_is_exact((const struct in6_addr
*)ipv6_mask
->ipv6_src
)
1504 && ipv6_mask_is_exact((const struct in6_addr
*)ipv6_mask
->ipv6_dst
);
1506 if (attr
== OVS_KEY_ATTR_TUNNEL
) {
1507 const struct flow_tnl
*tun_mask
= mask
;
1509 return tun_mask
->flags
== FLOW_TNL_F_MASK
1510 && tun_mask
->tun_id
== OVS_BE64_MAX
1511 && tun_mask
->ip_src
== OVS_BE32_MAX
1512 && tun_mask
->ip_dst
== OVS_BE32_MAX
1513 && tun_mask
->ip_tos
== UINT8_MAX
1514 && tun_mask
->ip_ttl
== UINT8_MAX
1515 && tun_mask
->tp_src
== OVS_BE16_MAX
1516 && tun_mask
->tp_dst
== OVS_BE16_MAX
1517 && tun_mask
->gbp_id
== OVS_BE16_MAX
1518 && tun_mask
->gbp_flags
== UINT8_MAX
;
1521 if (attr
== OVS_KEY_ATTR_ARP
) {
1522 /* ARP key has padding, ignore it. */
1523 BUILD_ASSERT_DECL(sizeof(struct ovs_key_arp
) == 24);
1524 BUILD_ASSERT_DECL(offsetof(struct ovs_key_arp
, arp_tha
) == 10 + 6);
1525 size
= offsetof(struct ovs_key_arp
, arp_tha
) + ETH_ADDR_LEN
;
1526 ovs_assert(((uint16_t *)mask
)[size
/2] == 0);
1529 return is_all_ones(mask
, size
);
1533 odp_mask_attr_is_exact(const struct nlattr
*ma
)
1535 struct flow_tnl tun_mask
;
1536 enum ovs_key_attr attr
= nl_attr_type(ma
);
1540 if (attr
== OVS_KEY_ATTR_TUNNEL
) {
1541 memset(&tun_mask
, 0, sizeof tun_mask
);
1542 odp_tun_key_from_attr(ma
, &tun_mask
);
1544 size
= sizeof tun_mask
;
1546 mask
= nl_attr_get(ma
);
1547 size
= nl_attr_get_size(ma
);
1550 return odp_mask_is_exact(attr
, mask
, size
);
1554 odp_portno_names_set(struct hmap
*portno_names
, odp_port_t port_no
,
1557 struct odp_portno_names
*odp_portno_names
;
1559 odp_portno_names
= xmalloc(sizeof *odp_portno_names
);
1560 odp_portno_names
->port_no
= port_no
;
1561 odp_portno_names
->name
= xstrdup(port_name
);
1562 hmap_insert(portno_names
, &odp_portno_names
->hmap_node
,
1563 hash_odp_port(port_no
));
1567 odp_portno_names_get(const struct hmap
*portno_names
, odp_port_t port_no
)
1569 struct odp_portno_names
*odp_portno_names
;
1571 HMAP_FOR_EACH_IN_BUCKET (odp_portno_names
, hmap_node
,
1572 hash_odp_port(port_no
), portno_names
) {
1573 if (odp_portno_names
->port_no
== port_no
) {
1574 return odp_portno_names
->name
;
1581 odp_portno_names_destroy(struct hmap
*portno_names
)
1583 struct odp_portno_names
*odp_portno_names
, *odp_portno_names_next
;
1584 HMAP_FOR_EACH_SAFE (odp_portno_names
, odp_portno_names_next
,
1585 hmap_node
, portno_names
) {
1586 hmap_remove(portno_names
, &odp_portno_names
->hmap_node
);
1587 free(odp_portno_names
->name
);
1588 free(odp_portno_names
);
1592 /* Format helpers. */
1595 format_eth(struct ds
*ds
, const char *name
, const uint8_t key
[ETH_ADDR_LEN
],
1596 const uint8_t (*mask
)[ETH_ADDR_LEN
], bool verbose
)
1598 bool mask_empty
= mask
&& eth_addr_is_zero(*mask
);
1600 if (verbose
|| !mask_empty
) {
1601 bool mask_full
= !mask
|| eth_mask_is_exact(*mask
);
1604 ds_put_format(ds
, "%s="ETH_ADDR_FMT
",", name
, ETH_ADDR_ARGS(key
));
1606 ds_put_format(ds
, "%s=", name
);
1607 eth_format_masked(key
, *mask
, ds
);
1608 ds_put_char(ds
, ',');
1614 format_be64(struct ds
*ds
, const char *name
, ovs_be64 key
,
1615 const ovs_be64
*mask
, bool verbose
)
1617 bool mask_empty
= mask
&& !*mask
;
1619 if (verbose
|| !mask_empty
) {
1620 bool mask_full
= !mask
|| *mask
== OVS_BE64_MAX
;
1622 ds_put_format(ds
, "%s=0x%"PRIx64
, name
, ntohll(key
));
1623 if (!mask_full
) { /* Partially masked. */
1624 ds_put_format(ds
, "/%#"PRIx64
, ntohll(*mask
));
1626 ds_put_char(ds
, ',');
1631 format_ipv4(struct ds
*ds
, const char *name
, ovs_be32 key
,
1632 const ovs_be32
*mask
, bool verbose
)
1634 bool mask_empty
= mask
&& !*mask
;
1636 if (verbose
|| !mask_empty
) {
1637 bool mask_full
= !mask
|| *mask
== OVS_BE32_MAX
;
1639 ds_put_format(ds
, "%s="IP_FMT
, name
, IP_ARGS(key
));
1640 if (!mask_full
) { /* Partially masked. */
1641 ds_put_format(ds
, "/"IP_FMT
, IP_ARGS(*mask
));
1643 ds_put_char(ds
, ',');
1648 format_ipv6(struct ds
*ds
, const char *name
, const ovs_be32 key_
[4],
1649 const ovs_be32 (*mask_
)[4], bool verbose
)
1651 char buf
[INET6_ADDRSTRLEN
];
1652 const struct in6_addr
*key
= (const struct in6_addr
*)key_
;
1653 const struct in6_addr
*mask
= mask_
? (const struct in6_addr
*)*mask_
1655 bool mask_empty
= mask
&& ipv6_mask_is_any(mask
);
1657 if (verbose
|| !mask_empty
) {
1658 bool mask_full
= !mask
|| ipv6_mask_is_exact(mask
);
1660 inet_ntop(AF_INET6
, key
, buf
, sizeof buf
);
1661 ds_put_format(ds
, "%s=%s", name
, buf
);
1662 if (!mask_full
) { /* Partially masked. */
1663 inet_ntop(AF_INET6
, mask
, buf
, sizeof buf
);
1664 ds_put_format(ds
, "/%s", buf
);
1666 ds_put_char(ds
, ',');
1671 format_ipv6_label(struct ds
*ds
, const char *name
, ovs_be32 key
,
1672 const ovs_be32
*mask
, bool verbose
)
1674 bool mask_empty
= mask
&& !*mask
;
1676 if (verbose
|| !mask_empty
) {
1677 bool mask_full
= !mask
1678 || (*mask
& htonl(IPV6_LABEL_MASK
)) == htonl(IPV6_LABEL_MASK
);
1680 ds_put_format(ds
, "%s=%#"PRIx32
, name
, ntohl(key
));
1681 if (!mask_full
) { /* Partially masked. */
1682 ds_put_format(ds
, "/%#"PRIx32
, ntohl(*mask
));
1684 ds_put_char(ds
, ',');
1689 format_u8x(struct ds
*ds
, const char *name
, uint8_t key
,
1690 const uint8_t *mask
, bool verbose
)
1692 bool mask_empty
= mask
&& !*mask
;
1694 if (verbose
|| !mask_empty
) {
1695 bool mask_full
= !mask
|| *mask
== UINT8_MAX
;
1697 ds_put_format(ds
, "%s=%#"PRIx8
, name
, key
);
1698 if (!mask_full
) { /* Partially masked. */
1699 ds_put_format(ds
, "/%#"PRIx8
, *mask
);
1701 ds_put_char(ds
, ',');
1706 format_u8u(struct ds
*ds
, const char *name
, uint8_t key
,
1707 const uint8_t *mask
, bool verbose
)
1709 bool mask_empty
= mask
&& !*mask
;
1711 if (verbose
|| !mask_empty
) {
1712 bool mask_full
= !mask
|| *mask
== UINT8_MAX
;
1714 ds_put_format(ds
, "%s=%"PRIu8
, name
, key
);
1715 if (!mask_full
) { /* Partially masked. */
1716 ds_put_format(ds
, "/%#"PRIx8
, *mask
);
1718 ds_put_char(ds
, ',');
1723 format_be16(struct ds
*ds
, const char *name
, ovs_be16 key
,
1724 const ovs_be16
*mask
, bool verbose
)
1726 bool mask_empty
= mask
&& !*mask
;
1728 if (verbose
|| !mask_empty
) {
1729 bool mask_full
= !mask
|| *mask
== OVS_BE16_MAX
;
1731 ds_put_format(ds
, "%s=%"PRIu16
, name
, ntohs(key
));
1732 if (!mask_full
) { /* Partially masked. */
1733 ds_put_format(ds
, "/%#"PRIx16
, ntohs(*mask
));
1735 ds_put_char(ds
, ',');
1740 format_tun_flags(struct ds
*ds
, const char *name
, uint16_t key
,
1741 const uint16_t *mask
, bool verbose
)
1743 bool mask_empty
= mask
&& !*mask
;
1745 if (verbose
|| !mask_empty
) {
1746 bool mask_full
= !mask
|| (*mask
& FLOW_TNL_F_MASK
) == FLOW_TNL_F_MASK
;
1748 ds_put_cstr(ds
, name
);
1749 ds_put_char(ds
, '(');
1750 if (!mask_full
) { /* Partially masked. */
1751 format_flags_masked(ds
, NULL
, flow_tun_flag_to_string
, key
, *mask
);
1752 } else { /* Fully masked. */
1753 format_flags(ds
, flow_tun_flag_to_string
, key
, ',');
1755 ds_put_cstr(ds
, "),");
1760 format_frag(struct ds
*ds
, const char *name
, uint8_t key
,
1761 const uint8_t *mask
, bool verbose
)
1763 bool mask_empty
= mask
&& !*mask
;
1765 /* ODP frag is an enumeration field; partial masks are not meaningful. */
1766 if (verbose
|| !mask_empty
) {
1767 bool mask_full
= !mask
|| *mask
== UINT8_MAX
;
1769 if (!mask_full
) { /* Partially masked. */
1770 ds_put_format(ds
, "error: partial mask not supported for frag (%#"
1773 ds_put_format(ds
, "%s=%s,", name
, ovs_frag_type_to_string(key
));
1778 #define MASK(PTR, FIELD) PTR ? &PTR->FIELD : NULL
1781 format_odp_key_attr(const struct nlattr
*a
, const struct nlattr
*ma
,
1782 const struct hmap
*portno_names
, struct ds
*ds
,
1785 enum ovs_key_attr attr
= nl_attr_type(a
);
1786 char namebuf
[OVS_KEY_ATTR_BUFSIZE
];
1790 is_exact
= ma
? odp_mask_attr_is_exact(ma
) : true;
1792 ds_put_cstr(ds
, ovs_key_attr_to_string(attr
, namebuf
, sizeof namebuf
));
1795 expected_len
= odp_flow_key_attr_len(nl_attr_type(a
));
1796 if (expected_len
!= -2) {
1797 bool bad_key_len
= nl_attr_get_size(a
) != expected_len
;
1798 bool bad_mask_len
= ma
&& nl_attr_get_size(ma
) != expected_len
;
1800 if (bad_key_len
|| bad_mask_len
) {
1802 ds_put_format(ds
, "(bad key length %"PRIuSIZE
", expected %d)(",
1803 nl_attr_get_size(a
), expected_len
);
1805 format_generic_odp_key(a
, ds
);
1807 ds_put_char(ds
, '/');
1809 ds_put_format(ds
, "(bad mask length %"PRIuSIZE
", expected %d)(",
1810 nl_attr_get_size(ma
), expected_len
);
1812 format_generic_odp_key(ma
, ds
);
1814 ds_put_char(ds
, ')');
1820 ds_put_char(ds
, '(');
1822 case OVS_KEY_ATTR_ENCAP
:
1823 if (ma
&& nl_attr_get_size(ma
) && nl_attr_get_size(a
)) {
1824 odp_flow_format(nl_attr_get(a
), nl_attr_get_size(a
),
1825 nl_attr_get(ma
), nl_attr_get_size(ma
), NULL
, ds
,
1827 } else if (nl_attr_get_size(a
)) {
1828 odp_flow_format(nl_attr_get(a
), nl_attr_get_size(a
), NULL
, 0, NULL
,
1833 case OVS_KEY_ATTR_PRIORITY
:
1834 case OVS_KEY_ATTR_SKB_MARK
:
1835 case OVS_KEY_ATTR_DP_HASH
:
1836 case OVS_KEY_ATTR_RECIRC_ID
:
1837 ds_put_format(ds
, "%#"PRIx32
, nl_attr_get_u32(a
));
1839 ds_put_format(ds
, "/%#"PRIx32
, nl_attr_get_u32(ma
));
1843 case OVS_KEY_ATTR_TUNNEL
: {
1844 struct flow_tnl key
, mask_
;
1845 struct flow_tnl
*mask
= ma
? &mask_
: NULL
;
1848 memset(mask
, 0, sizeof *mask
);
1849 odp_tun_key_from_attr(ma
, mask
);
1851 memset(&key
, 0, sizeof key
);
1852 if (odp_tun_key_from_attr(a
, &key
) == ODP_FIT_ERROR
) {
1853 ds_put_format(ds
, "error");
1856 format_be64(ds
, "tun_id", key
.tun_id
, MASK(mask
, tun_id
), verbose
);
1857 format_ipv4(ds
, "src", key
.ip_src
, MASK(mask
, ip_src
), verbose
);
1858 format_ipv4(ds
, "dst", key
.ip_dst
, MASK(mask
, ip_dst
), verbose
);
1859 format_u8x(ds
, "tos", key
.ip_tos
, MASK(mask
, ip_tos
), verbose
);
1860 format_u8u(ds
, "ttl", key
.ip_ttl
, MASK(mask
, ip_ttl
), verbose
);
1861 format_be16(ds
, "tp_src", key
.tp_src
, MASK(mask
, tp_src
), verbose
);
1862 format_be16(ds
, "tp_dst", key
.tp_dst
, MASK(mask
, tp_dst
), verbose
);
1863 format_be16(ds
, "gbp_id", key
.gbp_id
, MASK(mask
, gbp_id
), verbose
);
1864 format_u8x(ds
, "gbp_flags", key
.gbp_flags
, MASK(mask
, gbp_flags
), verbose
);
1865 format_tun_flags(ds
, "flags", key
.flags
, MASK(mask
, flags
), verbose
);
1869 case OVS_KEY_ATTR_IN_PORT
:
1870 if (portno_names
&& verbose
&& is_exact
) {
1871 char *name
= odp_portno_names_get(portno_names
,
1872 u32_to_odp(nl_attr_get_u32(a
)));
1874 ds_put_format(ds
, "%s", name
);
1876 ds_put_format(ds
, "%"PRIu32
, nl_attr_get_u32(a
));
1879 ds_put_format(ds
, "%"PRIu32
, nl_attr_get_u32(a
));
1881 ds_put_format(ds
, "/%#"PRIx32
, nl_attr_get_u32(ma
));
1886 case OVS_KEY_ATTR_ETHERNET
: {
1887 const struct ovs_key_ethernet
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1888 const struct ovs_key_ethernet
*key
= nl_attr_get(a
);
1890 format_eth(ds
, "src", key
->eth_src
, MASK(mask
, eth_src
), verbose
);
1891 format_eth(ds
, "dst", key
->eth_dst
, MASK(mask
, eth_dst
), verbose
);
1895 case OVS_KEY_ATTR_VLAN
:
1896 format_vlan_tci(ds
, nl_attr_get_be16(a
),
1897 ma
? nl_attr_get_be16(ma
) : OVS_BE16_MAX
, verbose
);
1900 case OVS_KEY_ATTR_MPLS
: {
1901 const struct ovs_key_mpls
*mpls_key
= nl_attr_get(a
);
1902 const struct ovs_key_mpls
*mpls_mask
= NULL
;
1903 size_t size
= nl_attr_get_size(a
);
1905 if (!size
|| size
% sizeof *mpls_key
) {
1906 ds_put_format(ds
, "(bad key length %"PRIuSIZE
")", size
);
1910 mpls_mask
= nl_attr_get(ma
);
1911 if (size
!= nl_attr_get_size(ma
)) {
1912 ds_put_format(ds
, "(key length %"PRIuSIZE
" != "
1913 "mask length %"PRIuSIZE
")",
1914 size
, nl_attr_get_size(ma
));
1918 format_mpls(ds
, mpls_key
, mpls_mask
, size
/ sizeof *mpls_key
);
1921 case OVS_KEY_ATTR_ETHERTYPE
:
1922 ds_put_format(ds
, "0x%04"PRIx16
, ntohs(nl_attr_get_be16(a
)));
1924 ds_put_format(ds
, "/0x%04"PRIx16
, ntohs(nl_attr_get_be16(ma
)));
1928 case OVS_KEY_ATTR_IPV4
: {
1929 const struct ovs_key_ipv4
*key
= nl_attr_get(a
);
1930 const struct ovs_key_ipv4
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1932 format_ipv4(ds
, "src", key
->ipv4_src
, MASK(mask
, ipv4_src
), verbose
);
1933 format_ipv4(ds
, "dst", key
->ipv4_dst
, MASK(mask
, ipv4_dst
), verbose
);
1934 format_u8u(ds
, "proto", key
->ipv4_proto
, MASK(mask
, ipv4_proto
),
1936 format_u8x(ds
, "tos", key
->ipv4_tos
, MASK(mask
, ipv4_tos
), verbose
);
1937 format_u8u(ds
, "ttl", key
->ipv4_ttl
, MASK(mask
, ipv4_ttl
), verbose
);
1938 format_frag(ds
, "frag", key
->ipv4_frag
, MASK(mask
, ipv4_frag
),
1943 case OVS_KEY_ATTR_IPV6
: {
1944 const struct ovs_key_ipv6
*key
= nl_attr_get(a
);
1945 const struct ovs_key_ipv6
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1947 format_ipv6(ds
, "src", key
->ipv6_src
, MASK(mask
, ipv6_src
), verbose
);
1948 format_ipv6(ds
, "dst", key
->ipv6_dst
, MASK(mask
, ipv6_dst
), verbose
);
1949 format_ipv6_label(ds
, "label", key
->ipv6_label
, MASK(mask
, ipv6_label
),
1951 format_u8u(ds
, "proto", key
->ipv6_proto
, MASK(mask
, ipv6_proto
),
1953 format_u8x(ds
, "tclass", key
->ipv6_tclass
, MASK(mask
, ipv6_tclass
),
1955 format_u8u(ds
, "hlimit", key
->ipv6_hlimit
, MASK(mask
, ipv6_hlimit
),
1957 format_frag(ds
, "frag", key
->ipv6_frag
, MASK(mask
, ipv6_frag
),
1962 /* These have the same structure and format. */
1963 case OVS_KEY_ATTR_TCP
:
1964 case OVS_KEY_ATTR_UDP
:
1965 case OVS_KEY_ATTR_SCTP
: {
1966 const struct ovs_key_tcp
*key
= nl_attr_get(a
);
1967 const struct ovs_key_tcp
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1969 format_be16(ds
, "src", key
->tcp_src
, MASK(mask
, tcp_src
), verbose
);
1970 format_be16(ds
, "dst", key
->tcp_dst
, MASK(mask
, tcp_dst
), verbose
);
1974 case OVS_KEY_ATTR_TCP_FLAGS
:
1976 format_flags_masked(ds
, NULL
, packet_tcp_flag_to_string
,
1977 ntohs(nl_attr_get_be16(a
)),
1978 ntohs(nl_attr_get_be16(ma
)));
1980 format_flags(ds
, packet_tcp_flag_to_string
,
1981 ntohs(nl_attr_get_be16(a
)), ',');
1985 case OVS_KEY_ATTR_ICMP
: {
1986 const struct ovs_key_icmp
*key
= nl_attr_get(a
);
1987 const struct ovs_key_icmp
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1989 format_u8u(ds
, "type", key
->icmp_type
, MASK(mask
, icmp_type
), verbose
);
1990 format_u8u(ds
, "code", key
->icmp_code
, MASK(mask
, icmp_code
), verbose
);
1994 case OVS_KEY_ATTR_ICMPV6
: {
1995 const struct ovs_key_icmpv6
*key
= nl_attr_get(a
);
1996 const struct ovs_key_icmpv6
*mask
= ma
? nl_attr_get(ma
) : NULL
;
1998 format_u8u(ds
, "type", key
->icmpv6_type
, MASK(mask
, icmpv6_type
),
2000 format_u8u(ds
, "code", key
->icmpv6_code
, MASK(mask
, icmpv6_code
),
2005 case OVS_KEY_ATTR_ARP
: {
2006 const struct ovs_key_arp
*mask
= ma
? nl_attr_get(ma
) : NULL
;
2007 const struct ovs_key_arp
*key
= nl_attr_get(a
);
2009 format_ipv4(ds
, "sip", key
->arp_sip
, MASK(mask
, arp_sip
), verbose
);
2010 format_ipv4(ds
, "tip", key
->arp_tip
, MASK(mask
, arp_tip
), verbose
);
2011 format_be16(ds
, "op", key
->arp_op
, MASK(mask
, arp_op
), verbose
);
2012 format_eth(ds
, "sha", key
->arp_sha
, MASK(mask
, arp_sha
), verbose
);
2013 format_eth(ds
, "tha", key
->arp_tha
, MASK(mask
, arp_tha
), verbose
);
2017 case OVS_KEY_ATTR_ND
: {
2018 const struct ovs_key_nd
*mask
= ma
? nl_attr_get(ma
) : NULL
;
2019 const struct ovs_key_nd
*key
= nl_attr_get(a
);
2021 format_ipv6(ds
, "target", key
->nd_target
, MASK(mask
, nd_target
),
2023 format_eth(ds
, "sll", key
->nd_sll
, MASK(mask
, nd_sll
), verbose
);
2024 format_eth(ds
, "tll", key
->nd_tll
, MASK(mask
, nd_tll
), verbose
);
2029 case OVS_KEY_ATTR_UNSPEC
:
2030 case __OVS_KEY_ATTR_MAX
:
2032 format_generic_odp_key(a
, ds
);
2034 ds_put_char(ds
, '/');
2035 format_generic_odp_key(ma
, ds
);
2039 ds_put_char(ds
, ')');
2042 static struct nlattr
*
2043 generate_all_wildcard_mask(struct ofpbuf
*ofp
, const struct nlattr
*key
)
2045 const struct nlattr
*a
;
2047 int type
= nl_attr_type(key
);
2048 int size
= nl_attr_get_size(key
);
2050 if (odp_flow_key_attr_len(type
) >=0) {
2051 nl_msg_put_unspec_zero(ofp
, type
, size
);
2055 nested_mask
= nl_msg_start_nested(ofp
, type
);
2056 NL_ATTR_FOR_EACH(a
, left
, key
, nl_attr_get_size(key
)) {
2057 generate_all_wildcard_mask(ofp
, nl_attr_get(a
));
2059 nl_msg_end_nested(ofp
, nested_mask
);
2066 odp_ufid_from_string(const char *s_
, ovs_u128
*ufid
)
2070 if (ovs_scan(s
, "ufid:")) {
2074 if (ovs_scan(s
, "0x")) {
2078 n
= strspn(s
, "0123456789abcdefABCDEF");
2083 if (!ovs_scan(s
, "%16"SCNx64
"%16"SCNx64
, &ufid
->u64
.hi
,
2088 s
+= strspn(s
, delimiters
);
2097 odp_format_ufid(const ovs_u128
*ufid
, struct ds
*ds
)
2099 ds_put_format(ds
, "ufid:%016"PRIx64
"%016"PRIx64
, ufid
->u64
.hi
,
2103 /* Appends to 'ds' a string representation of the 'key_len' bytes of
2104 * OVS_KEY_ATTR_* attributes in 'key'. If non-null, additionally formats the
2105 * 'mask_len' bytes of 'mask' which apply to 'key'. If 'portno_names' is
2106 * non-null and 'verbose' is true, translates odp port number to its name. */
2108 odp_flow_format(const struct nlattr
*key
, size_t key_len
,
2109 const struct nlattr
*mask
, size_t mask_len
,
2110 const struct hmap
*portno_names
, struct ds
*ds
, bool verbose
)
2113 const struct nlattr
*a
;
2115 bool has_ethtype_key
= false;
2116 const struct nlattr
*ma
= NULL
;
2118 bool first_field
= true;
2120 ofpbuf_init(&ofp
, 100);
2121 NL_ATTR_FOR_EACH (a
, left
, key
, key_len
) {
2122 bool is_nested_attr
;
2123 bool is_wildcard
= false;
2124 int attr_type
= nl_attr_type(a
);
2126 if (attr_type
== OVS_KEY_ATTR_ETHERTYPE
) {
2127 has_ethtype_key
= true;
2130 is_nested_attr
= (odp_flow_key_attr_len(attr_type
) == -2);
2132 if (mask
&& mask_len
) {
2133 ma
= nl_attr_find__(mask
, mask_len
, nl_attr_type(a
));
2134 is_wildcard
= ma
? odp_mask_attr_is_wildcard(ma
) : true;
2137 if (verbose
|| !is_wildcard
|| is_nested_attr
) {
2138 if (is_wildcard
&& !ma
) {
2139 ma
= generate_all_wildcard_mask(&ofp
, a
);
2142 ds_put_char(ds
, ',');
2144 format_odp_key_attr(a
, ma
, portno_names
, ds
, verbose
);
2145 first_field
= false;
2149 ofpbuf_uninit(&ofp
);
2154 if (left
== key_len
) {
2155 ds_put_cstr(ds
, "<empty>");
2157 ds_put_format(ds
, ",***%u leftover bytes*** (", left
);
2158 for (i
= 0; i
< left
; i
++) {
2159 ds_put_format(ds
, "%02x", ((const uint8_t *) a
)[i
]);
2161 ds_put_char(ds
, ')');
2163 if (!has_ethtype_key
) {
2164 ma
= nl_attr_find__(mask
, mask_len
, OVS_KEY_ATTR_ETHERTYPE
);
2166 ds_put_format(ds
, ",eth_type(0/0x%04"PRIx16
")",
2167 ntohs(nl_attr_get_be16(ma
)));
2171 ds_put_cstr(ds
, "<empty>");
2175 /* Appends to 'ds' a string representation of the 'key_len' bytes of
2176 * OVS_KEY_ATTR_* attributes in 'key'. */
2178 odp_flow_key_format(const struct nlattr
*key
,
2179 size_t key_len
, struct ds
*ds
)
2181 odp_flow_format(key
, key_len
, NULL
, 0, NULL
, ds
, true);
2185 ovs_frag_type_from_string(const char *s
, enum ovs_frag_type
*type
)
2187 if (!strcasecmp(s
, "no")) {
2188 *type
= OVS_FRAG_TYPE_NONE
;
2189 } else if (!strcasecmp(s
, "first")) {
2190 *type
= OVS_FRAG_TYPE_FIRST
;
2191 } else if (!strcasecmp(s
, "later")) {
2192 *type
= OVS_FRAG_TYPE_LATER
;
2202 scan_eth(const char *s
, uint8_t (*key
)[ETH_ADDR_LEN
],
2203 uint8_t (*mask
)[ETH_ADDR_LEN
])
2207 if (ovs_scan(s
, ETH_ADDR_SCAN_FMT
"%n", ETH_ADDR_SCAN_ARGS(*key
), &n
)) {
2211 if (ovs_scan(s
+ len
, "/"ETH_ADDR_SCAN_FMT
"%n",
2212 ETH_ADDR_SCAN_ARGS(*mask
), &n
)) {
2215 memset(mask
, 0xff, sizeof *mask
);
2224 scan_ipv4(const char *s
, ovs_be32
*key
, ovs_be32
*mask
)
2228 if (ovs_scan(s
, IP_SCAN_FMT
"%n", IP_SCAN_ARGS(key
), &n
)) {
2232 if (ovs_scan(s
+ len
, "/"IP_SCAN_FMT
"%n",
2233 IP_SCAN_ARGS(mask
), &n
)) {
2236 *mask
= OVS_BE32_MAX
;
2245 scan_ipv6(const char *s
, ovs_be32 (*key
)[4], ovs_be32 (*mask
)[4])
2248 char ipv6_s
[IPV6_SCAN_LEN
+ 1];
2250 if (ovs_scan(s
, IPV6_SCAN_FMT
"%n", ipv6_s
, &n
)
2251 && inet_pton(AF_INET6
, ipv6_s
, key
) == 1) {
2255 if (ovs_scan(s
+ len
, "/"IPV6_SCAN_FMT
"%n", ipv6_s
, &n
)
2256 && inet_pton(AF_INET6
, ipv6_s
, mask
) == 1) {
2259 memset(mask
, 0xff, sizeof *mask
);
2268 scan_ipv6_label(const char *s
, ovs_be32
*key
, ovs_be32
*mask
)
2273 if (ovs_scan(s
, "%i%n", &key_
, &n
)
2274 && (key_
& ~IPV6_LABEL_MASK
) == 0) {
2279 if (ovs_scan(s
+ len
, "/%i%n", &mask_
, &n
)
2280 && (mask_
& ~IPV6_LABEL_MASK
) == 0) {
2282 *mask
= htonl(mask_
);
2284 *mask
= htonl(IPV6_LABEL_MASK
);
2293 scan_u8(const char *s
, uint8_t *key
, uint8_t *mask
)
2297 if (ovs_scan(s
, "%"SCNi8
"%n", key
, &n
)) {
2301 if (ovs_scan(s
+ len
, "/%"SCNi8
"%n", mask
, &n
)) {
2313 scan_u32(const char *s
, uint32_t *key
, uint32_t *mask
)
2317 if (ovs_scan(s
, "%"SCNi32
"%n", key
, &n
)) {
2321 if (ovs_scan(s
+ len
, "/%"SCNi32
"%n", mask
, &n
)) {
2333 scan_be16(const char *s
, ovs_be16
*key
, ovs_be16
*mask
)
2335 uint16_t key_
, mask_
;
2338 if (ovs_scan(s
, "%"SCNi16
"%n", &key_
, &n
)) {
2343 if (ovs_scan(s
+ len
, "/%"SCNi16
"%n", &mask_
, &n
)) {
2345 *mask
= htons(mask_
);
2347 *mask
= OVS_BE16_MAX
;
2356 scan_be64(const char *s
, ovs_be64
*key
, ovs_be64
*mask
)
2358 uint64_t key_
, mask_
;
2361 if (ovs_scan(s
, "%"SCNi64
"%n", &key_
, &n
)) {
2364 *key
= htonll(key_
);
2366 if (ovs_scan(s
+ len
, "/%"SCNi64
"%n", &mask_
, &n
)) {
2368 *mask
= htonll(mask_
);
2370 *mask
= OVS_BE64_MAX
;
2379 scan_tun_flags(const char *s
, uint16_t *key
, uint16_t *mask
)
2381 uint32_t flags
, fmask
;
2384 n
= parse_flags(s
, flow_tun_flag_to_string
, &flags
,
2385 FLOW_TNL_F_MASK
, mask
? &fmask
: NULL
);
2386 if (n
>= 0 && s
[n
] == ')') {
2397 scan_tcp_flags(const char *s
, ovs_be16
*key
, ovs_be16
*mask
)
2399 uint32_t flags
, fmask
;
2402 n
= parse_flags(s
, packet_tcp_flag_to_string
, &flags
,
2403 TCP_FLAGS(OVS_BE16_MAX
), mask
? &fmask
: NULL
);
2405 *key
= htons(flags
);
2407 *mask
= htons(fmask
);
2415 scan_frag(const char *s
, uint8_t *key
, uint8_t *mask
)
2419 enum ovs_frag_type frag_type
;
2421 if (ovs_scan(s
, "%7[a-z]%n", frag
, &n
)
2422 && ovs_frag_type_from_string(frag
, &frag_type
)) {
2435 scan_port(const char *s
, uint32_t *key
, uint32_t *mask
,
2436 const struct simap
*port_names
)
2440 if (ovs_scan(s
, "%"SCNi32
"%n", key
, &n
)) {
2444 if (ovs_scan(s
+ len
, "/%"SCNi32
"%n", mask
, &n
)) {
2451 } else if (port_names
) {
2452 const struct simap_node
*node
;
2455 len
= strcspn(s
, ")");
2456 node
= simap_find_len(port_names
, s
, len
);
2469 /* Helper for vlan parsing. */
2470 struct ovs_key_vlan__
{
2475 set_be16_bf(ovs_be16
*bf
, uint8_t bits
, uint8_t offset
, uint16_t value
)
2477 const uint16_t mask
= ((1U << bits
) - 1) << offset
;
2479 if (value
>> bits
) {
2483 *bf
= htons((ntohs(*bf
) & ~mask
) | (value
<< offset
));
2488 scan_be16_bf(const char *s
, ovs_be16
*key
, ovs_be16
*mask
, uint8_t bits
,
2491 uint16_t key_
, mask_
;
2494 if (ovs_scan(s
, "%"SCNi16
"%n", &key_
, &n
)) {
2497 if (set_be16_bf(key
, bits
, offset
, key_
)) {
2499 if (ovs_scan(s
+ len
, "/%"SCNi16
"%n", &mask_
, &n
)) {
2502 if (!set_be16_bf(mask
, bits
, offset
, mask_
)) {
2506 *mask
|= htons(((1U << bits
) - 1) << offset
);
2516 scan_vid(const char *s
, ovs_be16
*key
, ovs_be16
*mask
)
2518 return scan_be16_bf(s
, key
, mask
, 12, VLAN_VID_SHIFT
);
2522 scan_pcp(const char *s
, ovs_be16
*key
, ovs_be16
*mask
)
2524 return scan_be16_bf(s
, key
, mask
, 3, VLAN_PCP_SHIFT
);
2528 scan_cfi(const char *s
, ovs_be16
*key
, ovs_be16
*mask
)
2530 return scan_be16_bf(s
, key
, mask
, 1, VLAN_CFI_SHIFT
);
2535 set_be32_bf(ovs_be32
*bf
, uint8_t bits
, uint8_t offset
, uint32_t value
)
2537 const uint32_t mask
= ((1U << bits
) - 1) << offset
;
2539 if (value
>> bits
) {
2543 *bf
= htonl((ntohl(*bf
) & ~mask
) | (value
<< offset
));
2548 scan_be32_bf(const char *s
, ovs_be32
*key
, ovs_be32
*mask
, uint8_t bits
,
2551 uint32_t key_
, mask_
;
2554 if (ovs_scan(s
, "%"SCNi32
"%n", &key_
, &n
)) {
2557 if (set_be32_bf(key
, bits
, offset
, key_
)) {
2559 if (ovs_scan(s
+ len
, "/%"SCNi32
"%n", &mask_
, &n
)) {
2562 if (!set_be32_bf(mask
, bits
, offset
, mask_
)) {
2566 *mask
|= htonl(((1U << bits
) - 1) << offset
);
2576 scan_mpls_label(const char *s
, ovs_be32
*key
, ovs_be32
*mask
)
2578 return scan_be32_bf(s
, key
, mask
, 20, MPLS_LABEL_SHIFT
);
2582 scan_mpls_tc(const char *s
, ovs_be32
*key
, ovs_be32
*mask
)
2584 return scan_be32_bf(s
, key
, mask
, 3, MPLS_TC_SHIFT
);
2588 scan_mpls_ttl(const char *s
, ovs_be32
*key
, ovs_be32
*mask
)
2590 return scan_be32_bf(s
, key
, mask
, 8, MPLS_TTL_SHIFT
);
2594 scan_mpls_bos(const char *s
, ovs_be32
*key
, ovs_be32
*mask
)
2596 return scan_be32_bf(s
, key
, mask
, 1, MPLS_BOS_SHIFT
);
2599 /* ATTR is compile-time constant, so only the case with correct data type
2600 * will be used. However, the compiler complains about the data type for
2601 * the other cases, so we must cast to make the compiler silent. */
2602 #define SCAN_PUT_ATTR(BUF, ATTR, DATA) \
2603 if ((ATTR) == OVS_KEY_ATTR_TUNNEL) { \
2604 tun_key_to_attr(BUF, (const struct flow_tnl *)(void *)&(DATA)); \
2606 nl_msg_put_unspec(BUF, ATTR, &(DATA), sizeof (DATA)); \
2609 #define SCAN_IF(NAME) \
2610 if (strncmp(s, NAME, strlen(NAME)) == 0) { \
2611 const char *start = s; \
2616 /* Usually no special initialization is needed. */
2617 #define SCAN_BEGIN(NAME, TYPE) \
2620 memset(&skey, 0, sizeof skey); \
2621 memset(&smask, 0, sizeof smask); \
2625 /* Init as fully-masked as mask will not be scanned. */
2626 #define SCAN_BEGIN_FULLY_MASKED(NAME, TYPE) \
2629 memset(&skey, 0, sizeof skey); \
2630 memset(&smask, 0xff, sizeof smask); \
2634 /* VLAN needs special initialization. */
2635 #define SCAN_BEGIN_INIT(NAME, TYPE, KEY_INIT, MASK_INIT) \
2637 TYPE skey = KEY_INIT; \
2638 TYPE smask = MASK_INIT; \
2642 /* Scan unnamed entry as 'TYPE' */
2643 #define SCAN_TYPE(TYPE, KEY, MASK) \
2644 len = scan_##TYPE(s, KEY, MASK); \
2650 /* Scan named ('NAME') entry 'FIELD' as 'TYPE'. */
2651 #define SCAN_FIELD(NAME, TYPE, FIELD) \
2652 if (strncmp(s, NAME, strlen(NAME)) == 0) { \
2653 s += strlen(NAME); \
2654 SCAN_TYPE(TYPE, &skey.FIELD, mask ? &smask.FIELD : NULL); \
2658 #define SCAN_FINISH() \
2659 } while (*s++ == ',' && len != 0); \
2660 if (s[-1] != ')') { \
2664 #define SCAN_FINISH_SINGLE() \
2666 if (*s++ != ')') { \
2670 #define SCAN_PUT(ATTR) \
2671 if (!mask || !is_all_zeros(&smask, sizeof smask)) { \
2672 SCAN_PUT_ATTR(key, ATTR, skey); \
2674 SCAN_PUT_ATTR(mask, ATTR, smask); \
2678 #define SCAN_END(ATTR) \
2684 #define SCAN_END_SINGLE(ATTR) \
2685 SCAN_FINISH_SINGLE(); \
2690 #define SCAN_SINGLE(NAME, TYPE, SCAN_AS, ATTR) \
2691 SCAN_BEGIN(NAME, TYPE) { \
2692 SCAN_TYPE(SCAN_AS, &skey, &smask); \
2693 } SCAN_END_SINGLE(ATTR)
2695 #define SCAN_SINGLE_FULLY_MASKED(NAME, TYPE, SCAN_AS, ATTR) \
2696 SCAN_BEGIN_FULLY_MASKED(NAME, TYPE) { \
2697 SCAN_TYPE(SCAN_AS, &skey, NULL); \
2698 } SCAN_END_SINGLE(ATTR)
2700 /* scan_port needs one extra argument. */
2701 #define SCAN_SINGLE_PORT(NAME, TYPE, ATTR) \
2702 SCAN_BEGIN(NAME, TYPE) { \
2703 len = scan_port(s, &skey, &smask, port_names); \
2708 } SCAN_END_SINGLE(ATTR)
2711 parse_odp_key_mask_attr(const char *s
, const struct simap
*port_names
,
2712 struct ofpbuf
*key
, struct ofpbuf
*mask
)
2714 SCAN_SINGLE("skb_priority(", uint32_t, u32
, OVS_KEY_ATTR_PRIORITY
);
2715 SCAN_SINGLE("skb_mark(", uint32_t, u32
, OVS_KEY_ATTR_SKB_MARK
);
2716 SCAN_SINGLE_FULLY_MASKED("recirc_id(", uint32_t, u32
,
2717 OVS_KEY_ATTR_RECIRC_ID
);
2718 SCAN_SINGLE("dp_hash(", uint32_t, u32
, OVS_KEY_ATTR_DP_HASH
);
2720 SCAN_BEGIN("tunnel(", struct flow_tnl
) {
2721 SCAN_FIELD("tun_id=", be64
, tun_id
);
2722 SCAN_FIELD("src=", ipv4
, ip_src
);
2723 SCAN_FIELD("dst=", ipv4
, ip_dst
);
2724 SCAN_FIELD("tos=", u8
, ip_tos
);
2725 SCAN_FIELD("ttl=", u8
, ip_ttl
);
2726 SCAN_FIELD("tp_src=", be16
, tp_src
);
2727 SCAN_FIELD("tp_dst=", be16
, tp_dst
);
2728 SCAN_FIELD("gbp_id=", be16
, gbp_id
);
2729 SCAN_FIELD("gbp_flags=", u8
, gbp_flags
);
2730 SCAN_FIELD("flags(", tun_flags
, flags
);
2731 } SCAN_END(OVS_KEY_ATTR_TUNNEL
);
2733 SCAN_SINGLE_PORT("in_port(", uint32_t, OVS_KEY_ATTR_IN_PORT
);
2735 SCAN_BEGIN("eth(", struct ovs_key_ethernet
) {
2736 SCAN_FIELD("src=", eth
, eth_src
);
2737 SCAN_FIELD("dst=", eth
, eth_dst
);
2738 } SCAN_END(OVS_KEY_ATTR_ETHERNET
);
2740 SCAN_BEGIN_INIT("vlan(", struct ovs_key_vlan__
,
2741 { htons(VLAN_CFI
) }, { htons(VLAN_CFI
) }) {
2742 SCAN_FIELD("vid=", vid
, tci
);
2743 SCAN_FIELD("pcp=", pcp
, tci
);
2744 SCAN_FIELD("cfi=", cfi
, tci
);
2745 } SCAN_END(OVS_KEY_ATTR_VLAN
);
2747 SCAN_SINGLE("eth_type(", ovs_be16
, be16
, OVS_KEY_ATTR_ETHERTYPE
);
2749 SCAN_BEGIN("mpls(", struct ovs_key_mpls
) {
2750 SCAN_FIELD("label=", mpls_label
, mpls_lse
);
2751 SCAN_FIELD("tc=", mpls_tc
, mpls_lse
);
2752 SCAN_FIELD("ttl=", mpls_ttl
, mpls_lse
);
2753 SCAN_FIELD("bos=", mpls_bos
, mpls_lse
);
2754 } SCAN_END(OVS_KEY_ATTR_MPLS
);
2756 SCAN_BEGIN("ipv4(", struct ovs_key_ipv4
) {
2757 SCAN_FIELD("src=", ipv4
, ipv4_src
);
2758 SCAN_FIELD("dst=", ipv4
, ipv4_dst
);
2759 SCAN_FIELD("proto=", u8
, ipv4_proto
);
2760 SCAN_FIELD("tos=", u8
, ipv4_tos
);
2761 SCAN_FIELD("ttl=", u8
, ipv4_ttl
);
2762 SCAN_FIELD("frag=", frag
, ipv4_frag
);
2763 } SCAN_END(OVS_KEY_ATTR_IPV4
);
2765 SCAN_BEGIN("ipv6(", struct ovs_key_ipv6
) {
2766 SCAN_FIELD("src=", ipv6
, ipv6_src
);
2767 SCAN_FIELD("dst=", ipv6
, ipv6_dst
);
2768 SCAN_FIELD("label=", ipv6_label
, ipv6_label
);
2769 SCAN_FIELD("proto=", u8
, ipv6_proto
);
2770 SCAN_FIELD("tclass=", u8
, ipv6_tclass
);
2771 SCAN_FIELD("hlimit=", u8
, ipv6_hlimit
);
2772 SCAN_FIELD("frag=", frag
, ipv6_frag
);
2773 } SCAN_END(OVS_KEY_ATTR_IPV6
);
2775 SCAN_BEGIN("tcp(", struct ovs_key_tcp
) {
2776 SCAN_FIELD("src=", be16
, tcp_src
);
2777 SCAN_FIELD("dst=", be16
, tcp_dst
);
2778 } SCAN_END(OVS_KEY_ATTR_TCP
);
2780 SCAN_SINGLE("tcp_flags(", ovs_be16
, tcp_flags
, OVS_KEY_ATTR_TCP_FLAGS
);
2782 SCAN_BEGIN("udp(", struct ovs_key_udp
) {
2783 SCAN_FIELD("src=", be16
, udp_src
);
2784 SCAN_FIELD("dst=", be16
, udp_dst
);
2785 } SCAN_END(OVS_KEY_ATTR_UDP
);
2787 SCAN_BEGIN("sctp(", struct ovs_key_sctp
) {
2788 SCAN_FIELD("src=", be16
, sctp_src
);
2789 SCAN_FIELD("dst=", be16
, sctp_dst
);
2790 } SCAN_END(OVS_KEY_ATTR_SCTP
);
2792 SCAN_BEGIN("icmp(", struct ovs_key_icmp
) {
2793 SCAN_FIELD("type=", u8
, icmp_type
);
2794 SCAN_FIELD("code=", u8
, icmp_code
);
2795 } SCAN_END(OVS_KEY_ATTR_ICMP
);
2797 SCAN_BEGIN("icmpv6(", struct ovs_key_icmpv6
) {
2798 SCAN_FIELD("type=", u8
, icmpv6_type
);
2799 SCAN_FIELD("code=", u8
, icmpv6_code
);
2800 } SCAN_END(OVS_KEY_ATTR_ICMPV6
);
2802 SCAN_BEGIN("arp(", struct ovs_key_arp
) {
2803 SCAN_FIELD("sip=", ipv4
, arp_sip
);
2804 SCAN_FIELD("tip=", ipv4
, arp_tip
);
2805 SCAN_FIELD("op=", be16
, arp_op
);
2806 SCAN_FIELD("sha=", eth
, arp_sha
);
2807 SCAN_FIELD("tha=", eth
, arp_tha
);
2808 } SCAN_END(OVS_KEY_ATTR_ARP
);
2810 SCAN_BEGIN("nd(", struct ovs_key_nd
) {
2811 SCAN_FIELD("target=", ipv6
, nd_target
);
2812 SCAN_FIELD("sll=", eth
, nd_sll
);
2813 SCAN_FIELD("tll=", eth
, nd_tll
);
2814 } SCAN_END(OVS_KEY_ATTR_ND
);
2816 /* Encap open-coded. */
2817 if (!strncmp(s
, "encap(", 6)) {
2818 const char *start
= s
;
2819 size_t encap
, encap_mask
= 0;
2821 encap
= nl_msg_start_nested(key
, OVS_KEY_ATTR_ENCAP
);
2823 encap_mask
= nl_msg_start_nested(mask
, OVS_KEY_ATTR_ENCAP
);
2830 s
+= strspn(s
, delimiters
);
2833 } else if (*s
== ')') {
2837 retval
= parse_odp_key_mask_attr(s
, port_names
, key
, mask
);
2845 nl_msg_end_nested(key
, encap
);
2847 nl_msg_end_nested(mask
, encap_mask
);
2856 /* Parses the string representation of a datapath flow key, in the
2857 * format output by odp_flow_key_format(). Returns 0 if successful,
2858 * otherwise a positive errno value. On success, the flow key is
2859 * appended to 'key' as a series of Netlink attributes. On failure, no
2860 * data is appended to 'key'. Either way, 'key''s data might be
2863 * If 'port_names' is nonnull, it points to an simap that maps from a port name
2864 * to a port number. (Port names may be used instead of port numbers in
2867 * On success, the attributes appended to 'key' are individually syntactically
2868 * valid, but they may not be valid as a sequence. 'key' might, for example,
2869 * have duplicated keys. odp_flow_key_to_flow() will detect those errors. */
2871 odp_flow_from_string(const char *s
, const struct simap
*port_names
,
2872 struct ofpbuf
*key
, struct ofpbuf
*mask
)
2874 const size_t old_size
= key
->size
;
2878 s
+= strspn(s
, delimiters
);
2883 retval
= parse_odp_key_mask_attr(s
, port_names
, key
, mask
);
2885 key
->size
= old_size
;
2895 ovs_to_odp_frag(uint8_t nw_frag
, bool is_mask
)
2898 /* Netlink interface 'enum ovs_frag_type' is an 8-bit enumeration type,
2899 * not a set of flags or bitfields. Hence, if the struct flow nw_frag
2900 * mask, which is a set of bits, has the FLOW_NW_FRAG_ANY as zero, we
2901 * must use a zero mask for the netlink frag field, and all ones mask
2903 return (nw_frag
& FLOW_NW_FRAG_ANY
) ? UINT8_MAX
: 0;
2905 return !(nw_frag
& FLOW_NW_FRAG_ANY
) ? OVS_FRAG_TYPE_NONE
2906 : nw_frag
& FLOW_NW_FRAG_LATER
? OVS_FRAG_TYPE_LATER
2907 : OVS_FRAG_TYPE_FIRST
;
2910 static void get_ethernet_key(const struct flow
*, struct ovs_key_ethernet
*);
2911 static void put_ethernet_key(const struct ovs_key_ethernet
*, struct flow
*);
2912 static void get_ipv4_key(const struct flow
*, struct ovs_key_ipv4
*,
2914 static void put_ipv4_key(const struct ovs_key_ipv4
*, struct flow
*,
2916 static void get_ipv6_key(const struct flow
*, struct ovs_key_ipv6
*,
2918 static void put_ipv6_key(const struct ovs_key_ipv6
*, struct flow
*,
2920 static void get_arp_key(const struct flow
*, struct ovs_key_arp
*);
2921 static void put_arp_key(const struct ovs_key_arp
*, struct flow
*);
2922 static void get_nd_key(const struct flow
*, struct ovs_key_nd
*);
2923 static void put_nd_key(const struct ovs_key_nd
*, struct flow
*);
2925 /* These share the same layout. */
2927 struct ovs_key_tcp tcp
;
2928 struct ovs_key_udp udp
;
2929 struct ovs_key_sctp sctp
;
2932 static void get_tp_key(const struct flow
*, union ovs_key_tp
*);
2933 static void put_tp_key(const union ovs_key_tp
*, struct flow
*);
2936 odp_flow_key_from_flow__(struct ofpbuf
*buf
, const struct flow
*flow
,
2937 const struct flow
*mask
, odp_port_t odp_in_port
,
2938 size_t max_mpls_depth
, bool recirc
, bool export_mask
)
2940 struct ovs_key_ethernet
*eth_key
;
2942 const struct flow
*data
= export_mask
? mask
: flow
;
2944 nl_msg_put_u32(buf
, OVS_KEY_ATTR_PRIORITY
, data
->skb_priority
);
2946 if (flow
->tunnel
.ip_dst
|| export_mask
) {
2947 tun_key_to_attr(buf
, &data
->tunnel
);
2950 nl_msg_put_u32(buf
, OVS_KEY_ATTR_SKB_MARK
, data
->pkt_mark
);
2953 nl_msg_put_u32(buf
, OVS_KEY_ATTR_RECIRC_ID
, data
->recirc_id
);
2954 nl_msg_put_u32(buf
, OVS_KEY_ATTR_DP_HASH
, data
->dp_hash
);
2957 /* Add an ingress port attribute if this is a mask or 'odp_in_port'
2958 * is not the magical value "ODPP_NONE". */
2959 if (export_mask
|| odp_in_port
!= ODPP_NONE
) {
2960 nl_msg_put_odp_port(buf
, OVS_KEY_ATTR_IN_PORT
, odp_in_port
);
2963 eth_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_ETHERNET
,
2965 get_ethernet_key(data
, eth_key
);
2967 if (flow
->vlan_tci
!= htons(0) || flow
->dl_type
== htons(ETH_TYPE_VLAN
)) {
2969 nl_msg_put_be16(buf
, OVS_KEY_ATTR_ETHERTYPE
, OVS_BE16_MAX
);
2971 nl_msg_put_be16(buf
, OVS_KEY_ATTR_ETHERTYPE
, htons(ETH_TYPE_VLAN
));
2973 nl_msg_put_be16(buf
, OVS_KEY_ATTR_VLAN
, data
->vlan_tci
);
2974 encap
= nl_msg_start_nested(buf
, OVS_KEY_ATTR_ENCAP
);
2975 if (flow
->vlan_tci
== htons(0)) {
2982 if (ntohs(flow
->dl_type
) < ETH_TYPE_MIN
) {
2983 /* For backwards compatibility with kernels that don't support
2984 * wildcarding, the following convention is used to encode the
2985 * OVS_KEY_ATTR_ETHERTYPE for key and mask:
2988 * -------- -------- -------
2989 * >0x5ff 0xffff Specified Ethernet II Ethertype.
2990 * >0x5ff 0 Any Ethernet II or non-Ethernet II frame.
2991 * <none> 0xffff Any non-Ethernet II frame (except valid
2992 * 802.3 SNAP packet with valid eth_type).
2995 nl_msg_put_be16(buf
, OVS_KEY_ATTR_ETHERTYPE
, OVS_BE16_MAX
);
3000 nl_msg_put_be16(buf
, OVS_KEY_ATTR_ETHERTYPE
, data
->dl_type
);
3002 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
3003 struct ovs_key_ipv4
*ipv4_key
;
3005 ipv4_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_IPV4
,
3007 get_ipv4_key(data
, ipv4_key
, export_mask
);
3008 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
3009 struct ovs_key_ipv6
*ipv6_key
;
3011 ipv6_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_IPV6
,
3013 get_ipv6_key(data
, ipv6_key
, export_mask
);
3014 } else if (flow
->dl_type
== htons(ETH_TYPE_ARP
) ||
3015 flow
->dl_type
== htons(ETH_TYPE_RARP
)) {
3016 struct ovs_key_arp
*arp_key
;
3018 arp_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_ARP
,
3020 get_arp_key(data
, arp_key
);
3021 } else if (eth_type_mpls(flow
->dl_type
)) {
3022 struct ovs_key_mpls
*mpls_key
;
3025 n
= flow_count_mpls_labels(flow
, NULL
);
3026 n
= MIN(n
, max_mpls_depth
);
3027 mpls_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_MPLS
,
3028 n
* sizeof *mpls_key
);
3029 for (i
= 0; i
< n
; i
++) {
3030 mpls_key
[i
].mpls_lse
= data
->mpls_lse
[i
];
3034 if (is_ip_any(flow
) && !(flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
3035 if (flow
->nw_proto
== IPPROTO_TCP
) {
3036 union ovs_key_tp
*tcp_key
;
3038 tcp_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_TCP
,
3040 get_tp_key(data
, tcp_key
);
3041 if (data
->tcp_flags
) {
3042 nl_msg_put_be16(buf
, OVS_KEY_ATTR_TCP_FLAGS
, data
->tcp_flags
);
3044 } else if (flow
->nw_proto
== IPPROTO_UDP
) {
3045 union ovs_key_tp
*udp_key
;
3047 udp_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_UDP
,
3049 get_tp_key(data
, udp_key
);
3050 } else if (flow
->nw_proto
== IPPROTO_SCTP
) {
3051 union ovs_key_tp
*sctp_key
;
3053 sctp_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_SCTP
,
3055 get_tp_key(data
, sctp_key
);
3056 } else if (flow
->dl_type
== htons(ETH_TYPE_IP
)
3057 && flow
->nw_proto
== IPPROTO_ICMP
) {
3058 struct ovs_key_icmp
*icmp_key
;
3060 icmp_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_ICMP
,
3062 icmp_key
->icmp_type
= ntohs(data
->tp_src
);
3063 icmp_key
->icmp_code
= ntohs(data
->tp_dst
);
3064 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)
3065 && flow
->nw_proto
== IPPROTO_ICMPV6
) {
3066 struct ovs_key_icmpv6
*icmpv6_key
;
3068 icmpv6_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_ICMPV6
,
3069 sizeof *icmpv6_key
);
3070 icmpv6_key
->icmpv6_type
= ntohs(data
->tp_src
);
3071 icmpv6_key
->icmpv6_code
= ntohs(data
->tp_dst
);
3073 if (flow
->tp_dst
== htons(0)
3074 && (flow
->tp_src
== htons(ND_NEIGHBOR_SOLICIT
)
3075 || flow
->tp_src
== htons(ND_NEIGHBOR_ADVERT
))
3076 && (!export_mask
|| (data
->tp_src
== htons(0xffff)
3077 && data
->tp_dst
== htons(0xffff)))) {
3079 struct ovs_key_nd
*nd_key
;
3081 nd_key
= nl_msg_put_unspec_uninit(buf
, OVS_KEY_ATTR_ND
,
3083 memcpy(nd_key
->nd_target
, &data
->nd_target
,
3084 sizeof nd_key
->nd_target
);
3085 memcpy(nd_key
->nd_sll
, data
->arp_sha
, ETH_ADDR_LEN
);
3086 memcpy(nd_key
->nd_tll
, data
->arp_tha
, ETH_ADDR_LEN
);
3093 nl_msg_end_nested(buf
, encap
);
3097 /* Appends a representation of 'flow' as OVS_KEY_ATTR_* attributes to 'buf'.
3098 * 'flow->in_port' is ignored (since it is likely to be an OpenFlow port
3099 * number rather than a datapath port number). Instead, if 'odp_in_port'
3100 * is anything other than ODPP_NONE, it is included in 'buf' as the input
3103 * 'buf' must have at least ODPUTIL_FLOW_KEY_BYTES bytes of space, or be
3104 * capable of being expanded to allow for that much space.
3106 * 'recirc' indicates support for recirculation fields. If this is true, then
3107 * these fields will always be serialised. */
3109 odp_flow_key_from_flow(struct ofpbuf
*buf
, const struct flow
*flow
,
3110 const struct flow
*mask
, odp_port_t odp_in_port
,
3113 odp_flow_key_from_flow__(buf
, flow
, mask
, odp_in_port
, SIZE_MAX
, recirc
,
3117 /* Appends a representation of 'mask' as OVS_KEY_ATTR_* attributes to
3118 * 'buf'. 'flow' is used as a template to determine how to interpret
3119 * 'mask'. For example, the 'dl_type' of 'mask' describes the mask, but
3120 * it doesn't indicate whether the other fields should be interpreted as
3121 * ARP, IPv4, IPv6, etc.
3123 * 'buf' must have at least ODPUTIL_FLOW_KEY_BYTES bytes of space, or be
3124 * capable of being expanded to allow for that much space.
3126 * 'recirc' indicates support for recirculation fields. If this is true, then
3127 * these fields will always be serialised. */
3129 odp_flow_key_from_mask(struct ofpbuf
*buf
, const struct flow
*mask
,
3130 const struct flow
*flow
, uint32_t odp_in_port_mask
,
3131 size_t max_mpls_depth
, bool recirc
)
3133 odp_flow_key_from_flow__(buf
, flow
, mask
, u32_to_odp(odp_in_port_mask
),
3134 max_mpls_depth
, recirc
, true);
3137 /* Generate ODP flow key from the given packet metadata */
3139 odp_key_from_pkt_metadata(struct ofpbuf
*buf
, const struct pkt_metadata
*md
)
3141 nl_msg_put_u32(buf
, OVS_KEY_ATTR_PRIORITY
, md
->skb_priority
);
3143 if (md
->tunnel
.ip_dst
) {
3144 tun_key_to_attr(buf
, &md
->tunnel
);
3147 nl_msg_put_u32(buf
, OVS_KEY_ATTR_SKB_MARK
, md
->pkt_mark
);
3149 /* Add an ingress port attribute if 'odp_in_port' is not the magical
3150 * value "ODPP_NONE". */
3151 if (md
->in_port
.odp_port
!= ODPP_NONE
) {
3152 nl_msg_put_odp_port(buf
, OVS_KEY_ATTR_IN_PORT
, md
->in_port
.odp_port
);
3156 /* Generate packet metadata from the given ODP flow key. */
3158 odp_key_to_pkt_metadata(const struct nlattr
*key
, size_t key_len
,
3159 struct pkt_metadata
*md
)
3161 const struct nlattr
*nla
;
3163 uint32_t wanted_attrs
= 1u << OVS_KEY_ATTR_PRIORITY
|
3164 1u << OVS_KEY_ATTR_SKB_MARK
| 1u << OVS_KEY_ATTR_TUNNEL
|
3165 1u << OVS_KEY_ATTR_IN_PORT
;
3167 *md
= PKT_METADATA_INITIALIZER(ODPP_NONE
);
3169 NL_ATTR_FOR_EACH (nla
, left
, key
, key_len
) {
3170 uint16_t type
= nl_attr_type(nla
);
3171 size_t len
= nl_attr_get_size(nla
);
3172 int expected_len
= odp_flow_key_attr_len(type
);
3174 if (len
!= expected_len
&& expected_len
>= 0) {
3179 case OVS_KEY_ATTR_RECIRC_ID
:
3180 md
->recirc_id
= nl_attr_get_u32(nla
);
3181 wanted_attrs
&= ~(1u << OVS_KEY_ATTR_RECIRC_ID
);
3183 case OVS_KEY_ATTR_DP_HASH
:
3184 md
->dp_hash
= nl_attr_get_u32(nla
);
3185 wanted_attrs
&= ~(1u << OVS_KEY_ATTR_DP_HASH
);
3187 case OVS_KEY_ATTR_PRIORITY
:
3188 md
->skb_priority
= nl_attr_get_u32(nla
);
3189 wanted_attrs
&= ~(1u << OVS_KEY_ATTR_PRIORITY
);
3191 case OVS_KEY_ATTR_SKB_MARK
:
3192 md
->pkt_mark
= nl_attr_get_u32(nla
);
3193 wanted_attrs
&= ~(1u << OVS_KEY_ATTR_SKB_MARK
);
3195 case OVS_KEY_ATTR_TUNNEL
: {
3196 enum odp_key_fitness res
;
3198 res
= odp_tun_key_from_attr(nla
, &md
->tunnel
);
3199 if (res
== ODP_FIT_ERROR
) {
3200 memset(&md
->tunnel
, 0, sizeof md
->tunnel
);
3201 } else if (res
== ODP_FIT_PERFECT
) {
3202 wanted_attrs
&= ~(1u << OVS_KEY_ATTR_TUNNEL
);
3206 case OVS_KEY_ATTR_IN_PORT
:
3207 md
->in_port
.odp_port
= nl_attr_get_odp_port(nla
);
3208 wanted_attrs
&= ~(1u << OVS_KEY_ATTR_IN_PORT
);
3214 if (!wanted_attrs
) {
3215 return; /* Have everything. */
3221 odp_flow_key_hash(const struct nlattr
*key
, size_t key_len
)
3223 BUILD_ASSERT_DECL(!(NLA_ALIGNTO
% sizeof(uint32_t)));
3224 return hash_words(ALIGNED_CAST(const uint32_t *, key
),
3225 key_len
/ sizeof(uint32_t), 0);
3229 log_odp_key_attributes(struct vlog_rate_limit
*rl
, const char *title
,
3230 uint64_t attrs
, int out_of_range_attr
,
3231 const struct nlattr
*key
, size_t key_len
)
3236 if (VLOG_DROP_DBG(rl
)) {
3241 for (i
= 0; i
< 64; i
++) {
3242 if (attrs
& (UINT64_C(1) << i
)) {
3243 char namebuf
[OVS_KEY_ATTR_BUFSIZE
];
3245 ds_put_format(&s
, " %s",
3246 ovs_key_attr_to_string(i
, namebuf
, sizeof namebuf
));
3249 if (out_of_range_attr
) {
3250 ds_put_format(&s
, " %d (and possibly others)", out_of_range_attr
);
3253 ds_put_cstr(&s
, ": ");
3254 odp_flow_key_format(key
, key_len
, &s
);
3256 VLOG_DBG("%s:%s", title
, ds_cstr(&s
));
3261 odp_to_ovs_frag(uint8_t odp_frag
, bool is_mask
)
3263 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3266 return odp_frag
? FLOW_NW_FRAG_MASK
: 0;
3269 if (odp_frag
> OVS_FRAG_TYPE_LATER
) {
3270 VLOG_ERR_RL(&rl
, "invalid frag %"PRIu8
" in flow key", odp_frag
);
3271 return 0xff; /* Error. */
3274 return (odp_frag
== OVS_FRAG_TYPE_NONE
) ? 0
3275 : (odp_frag
== OVS_FRAG_TYPE_FIRST
) ? FLOW_NW_FRAG_ANY
3276 : FLOW_NW_FRAG_ANY
| FLOW_NW_FRAG_LATER
;
3280 parse_flow_nlattrs(const struct nlattr
*key
, size_t key_len
,
3281 const struct nlattr
*attrs
[], uint64_t *present_attrsp
,
3282 int *out_of_range_attrp
)
3284 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(10, 10);
3285 const struct nlattr
*nla
;
3286 uint64_t present_attrs
;
3289 BUILD_ASSERT(OVS_KEY_ATTR_MAX
< CHAR_BIT
* sizeof present_attrs
);
3291 *out_of_range_attrp
= 0;
3292 NL_ATTR_FOR_EACH (nla
, left
, key
, key_len
) {
3293 uint16_t type
= nl_attr_type(nla
);
3294 size_t len
= nl_attr_get_size(nla
);
3295 int expected_len
= odp_flow_key_attr_len(type
);
3297 if (len
!= expected_len
&& expected_len
>= 0) {
3298 char namebuf
[OVS_KEY_ATTR_BUFSIZE
];
3300 VLOG_ERR_RL(&rl
, "attribute %s has length %"PRIuSIZE
" but should have "
3301 "length %d", ovs_key_attr_to_string(type
, namebuf
,
3307 if (type
> OVS_KEY_ATTR_MAX
) {
3308 *out_of_range_attrp
= type
;
3310 if (present_attrs
& (UINT64_C(1) << type
)) {
3311 char namebuf
[OVS_KEY_ATTR_BUFSIZE
];
3313 VLOG_ERR_RL(&rl
, "duplicate %s attribute in flow key",
3314 ovs_key_attr_to_string(type
,
3315 namebuf
, sizeof namebuf
));
3319 present_attrs
|= UINT64_C(1) << type
;
3324 VLOG_ERR_RL(&rl
, "trailing garbage in flow key");
3328 *present_attrsp
= present_attrs
;
3332 static enum odp_key_fitness
3333 check_expectations(uint64_t present_attrs
, int out_of_range_attr
,
3334 uint64_t expected_attrs
,
3335 const struct nlattr
*key
, size_t key_len
)
3337 uint64_t missing_attrs
;
3338 uint64_t extra_attrs
;
3340 missing_attrs
= expected_attrs
& ~present_attrs
;
3341 if (missing_attrs
) {
3342 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(10, 10);
3343 log_odp_key_attributes(&rl
, "expected but not present",
3344 missing_attrs
, 0, key
, key_len
);
3345 return ODP_FIT_TOO_LITTLE
;
3348 extra_attrs
= present_attrs
& ~expected_attrs
;
3349 if (extra_attrs
|| out_of_range_attr
) {
3350 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(10, 10);
3351 log_odp_key_attributes(&rl
, "present but not expected",
3352 extra_attrs
, out_of_range_attr
, key
, key_len
);
3353 return ODP_FIT_TOO_MUCH
;
3356 return ODP_FIT_PERFECT
;
3360 parse_ethertype(const struct nlattr
*attrs
[OVS_KEY_ATTR_MAX
+ 1],
3361 uint64_t present_attrs
, uint64_t *expected_attrs
,
3362 struct flow
*flow
, const struct flow
*src_flow
)
3364 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3365 bool is_mask
= flow
!= src_flow
;
3367 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE
)) {
3368 flow
->dl_type
= nl_attr_get_be16(attrs
[OVS_KEY_ATTR_ETHERTYPE
]);
3369 if (!is_mask
&& ntohs(flow
->dl_type
) < ETH_TYPE_MIN
) {
3370 VLOG_ERR_RL(&rl
, "invalid Ethertype %"PRIu16
" in flow key",
3371 ntohs(flow
->dl_type
));
3374 if (is_mask
&& ntohs(src_flow
->dl_type
) < ETH_TYPE_MIN
&&
3375 flow
->dl_type
!= htons(0xffff)) {
3378 *expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE
;
3381 flow
->dl_type
= htons(FLOW_DL_TYPE_NONE
);
3382 } else if (ntohs(src_flow
->dl_type
) < ETH_TYPE_MIN
) {
3383 /* See comments in odp_flow_key_from_flow__(). */
3384 VLOG_ERR_RL(&rl
, "mask expected for non-Ethernet II frame");
3391 static enum odp_key_fitness
3392 parse_l2_5_onward(const struct nlattr
*attrs
[OVS_KEY_ATTR_MAX
+ 1],
3393 uint64_t present_attrs
, int out_of_range_attr
,
3394 uint64_t expected_attrs
, struct flow
*flow
,
3395 const struct nlattr
*key
, size_t key_len
,
3396 const struct flow
*src_flow
)
3398 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3399 bool is_mask
= src_flow
!= flow
;
3400 const void *check_start
= NULL
;
3401 size_t check_len
= 0;
3402 enum ovs_key_attr expected_bit
= 0xff;
3404 if (eth_type_mpls(src_flow
->dl_type
)) {
3405 if (!is_mask
|| present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_MPLS
)) {
3406 expected_attrs
|= (UINT64_C(1) << OVS_KEY_ATTR_MPLS
);
3408 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_MPLS
)) {
3409 size_t size
= nl_attr_get_size(attrs
[OVS_KEY_ATTR_MPLS
]);
3410 const ovs_be32
*mpls_lse
= nl_attr_get(attrs
[OVS_KEY_ATTR_MPLS
]);
3411 int n
= size
/ sizeof(ovs_be32
);
3414 if (!size
|| size
% sizeof(ovs_be32
)) {
3415 return ODP_FIT_ERROR
;
3417 if (flow
->mpls_lse
[0] && flow
->dl_type
!= htons(0xffff)) {
3418 return ODP_FIT_ERROR
;
3421 for (i
= 0; i
< n
&& i
< FLOW_MAX_MPLS_LABELS
; i
++) {
3422 flow
->mpls_lse
[i
] = mpls_lse
[i
];
3424 if (n
> FLOW_MAX_MPLS_LABELS
) {
3425 return ODP_FIT_TOO_MUCH
;
3429 /* BOS may be set only in the innermost label. */
3430 for (i
= 0; i
< n
- 1; i
++) {
3431 if (flow
->mpls_lse
[i
] & htonl(MPLS_BOS_MASK
)) {
3432 return ODP_FIT_ERROR
;
3436 /* BOS must be set in the innermost label. */
3437 if (n
< FLOW_MAX_MPLS_LABELS
3438 && !(flow
->mpls_lse
[n
- 1] & htonl(MPLS_BOS_MASK
))) {
3439 return ODP_FIT_TOO_LITTLE
;
3445 } else if (src_flow
->dl_type
== htons(ETH_TYPE_IP
)) {
3447 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_IPV4
;
3449 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_IPV4
)) {
3450 const struct ovs_key_ipv4
*ipv4_key
;
3452 ipv4_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_IPV4
]);
3453 put_ipv4_key(ipv4_key
, flow
, is_mask
);
3454 if (flow
->nw_frag
> FLOW_NW_FRAG_MASK
) {
3455 return ODP_FIT_ERROR
;
3458 check_start
= ipv4_key
;
3459 check_len
= sizeof *ipv4_key
;
3460 expected_bit
= OVS_KEY_ATTR_IPV4
;
3463 } else if (src_flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
3465 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_IPV6
;
3467 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_IPV6
)) {
3468 const struct ovs_key_ipv6
*ipv6_key
;
3470 ipv6_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_IPV6
]);
3471 put_ipv6_key(ipv6_key
, flow
, is_mask
);
3472 if (flow
->nw_frag
> FLOW_NW_FRAG_MASK
) {
3473 return ODP_FIT_ERROR
;
3476 check_start
= ipv6_key
;
3477 check_len
= sizeof *ipv6_key
;
3478 expected_bit
= OVS_KEY_ATTR_IPV6
;
3481 } else if (src_flow
->dl_type
== htons(ETH_TYPE_ARP
) ||
3482 src_flow
->dl_type
== htons(ETH_TYPE_RARP
)) {
3484 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ARP
;
3486 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ARP
)) {
3487 const struct ovs_key_arp
*arp_key
;
3489 arp_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_ARP
]);
3490 if (!is_mask
&& (arp_key
->arp_op
& htons(0xff00))) {
3491 VLOG_ERR_RL(&rl
, "unsupported ARP opcode %"PRIu16
" in flow "
3492 "key", ntohs(arp_key
->arp_op
));
3493 return ODP_FIT_ERROR
;
3495 put_arp_key(arp_key
, flow
);
3497 check_start
= arp_key
;
3498 check_len
= sizeof *arp_key
;
3499 expected_bit
= OVS_KEY_ATTR_ARP
;
3505 if (check_len
> 0) { /* Happens only when 'is_mask'. */
3506 if (!is_all_zeros(check_start
, check_len
) &&
3507 flow
->dl_type
!= htons(0xffff)) {
3508 return ODP_FIT_ERROR
;
3510 expected_attrs
|= UINT64_C(1) << expected_bit
;
3514 expected_bit
= OVS_KEY_ATTR_UNSPEC
;
3515 if (src_flow
->nw_proto
== IPPROTO_TCP
3516 && (src_flow
->dl_type
== htons(ETH_TYPE_IP
) ||
3517 src_flow
->dl_type
== htons(ETH_TYPE_IPV6
))
3518 && !(src_flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
3520 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_TCP
;
3522 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_TCP
)) {
3523 const union ovs_key_tp
*tcp_key
;
3525 tcp_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_TCP
]);
3526 put_tp_key(tcp_key
, flow
);
3527 expected_bit
= OVS_KEY_ATTR_TCP
;
3529 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_TCP_FLAGS
)) {
3530 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_TCP_FLAGS
;
3531 flow
->tcp_flags
= nl_attr_get_be16(attrs
[OVS_KEY_ATTR_TCP_FLAGS
]);
3533 } else if (src_flow
->nw_proto
== IPPROTO_UDP
3534 && (src_flow
->dl_type
== htons(ETH_TYPE_IP
) ||
3535 src_flow
->dl_type
== htons(ETH_TYPE_IPV6
))
3536 && !(src_flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
3538 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_UDP
;
3540 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_UDP
)) {
3541 const union ovs_key_tp
*udp_key
;
3543 udp_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_UDP
]);
3544 put_tp_key(udp_key
, flow
);
3545 expected_bit
= OVS_KEY_ATTR_UDP
;
3547 } else if (src_flow
->nw_proto
== IPPROTO_SCTP
3548 && (src_flow
->dl_type
== htons(ETH_TYPE_IP
) ||
3549 src_flow
->dl_type
== htons(ETH_TYPE_IPV6
))
3550 && !(src_flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
3552 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_SCTP
;
3554 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_SCTP
)) {
3555 const union ovs_key_tp
*sctp_key
;
3557 sctp_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_SCTP
]);
3558 put_tp_key(sctp_key
, flow
);
3559 expected_bit
= OVS_KEY_ATTR_SCTP
;
3561 } else if (src_flow
->nw_proto
== IPPROTO_ICMP
3562 && src_flow
->dl_type
== htons(ETH_TYPE_IP
)
3563 && !(src_flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
3565 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ICMP
;
3567 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ICMP
)) {
3568 const struct ovs_key_icmp
*icmp_key
;
3570 icmp_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_ICMP
]);
3571 flow
->tp_src
= htons(icmp_key
->icmp_type
);
3572 flow
->tp_dst
= htons(icmp_key
->icmp_code
);
3573 expected_bit
= OVS_KEY_ATTR_ICMP
;
3575 } else if (src_flow
->nw_proto
== IPPROTO_ICMPV6
3576 && src_flow
->dl_type
== htons(ETH_TYPE_IPV6
)
3577 && !(src_flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
3579 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ICMPV6
;
3581 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ICMPV6
)) {
3582 const struct ovs_key_icmpv6
*icmpv6_key
;
3584 icmpv6_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_ICMPV6
]);
3585 flow
->tp_src
= htons(icmpv6_key
->icmpv6_type
);
3586 flow
->tp_dst
= htons(icmpv6_key
->icmpv6_code
);
3587 expected_bit
= OVS_KEY_ATTR_ICMPV6
;
3588 if (src_flow
->tp_dst
== htons(0) &&
3589 (src_flow
->tp_src
== htons(ND_NEIGHBOR_SOLICIT
) ||
3590 src_flow
->tp_src
== htons(ND_NEIGHBOR_ADVERT
))) {
3592 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ND
;
3594 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ND
)) {
3595 const struct ovs_key_nd
*nd_key
;
3597 nd_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_ND
]);
3598 memcpy(&flow
->nd_target
, nd_key
->nd_target
,
3599 sizeof flow
->nd_target
);
3600 memcpy(flow
->arp_sha
, nd_key
->nd_sll
, ETH_ADDR_LEN
);
3601 memcpy(flow
->arp_tha
, nd_key
->nd_tll
, ETH_ADDR_LEN
);
3603 if (!is_all_zeros(nd_key
, sizeof *nd_key
) &&
3604 (flow
->tp_src
!= htons(0xffff) ||
3605 flow
->tp_dst
!= htons(0xffff))) {
3606 return ODP_FIT_ERROR
;
3608 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ND
;
3615 if (is_mask
&& expected_bit
!= OVS_KEY_ATTR_UNSPEC
) {
3616 if ((flow
->tp_src
|| flow
->tp_dst
) && flow
->nw_proto
!= 0xff) {
3617 return ODP_FIT_ERROR
;
3619 expected_attrs
|= UINT64_C(1) << expected_bit
;
3624 return check_expectations(present_attrs
, out_of_range_attr
, expected_attrs
,
3628 /* Parse 802.1Q header then encapsulated L3 attributes. */
3629 static enum odp_key_fitness
3630 parse_8021q_onward(const struct nlattr
*attrs
[OVS_KEY_ATTR_MAX
+ 1],
3631 uint64_t present_attrs
, int out_of_range_attr
,
3632 uint64_t expected_attrs
, struct flow
*flow
,
3633 const struct nlattr
*key
, size_t key_len
,
3634 const struct flow
*src_flow
)
3636 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3637 bool is_mask
= src_flow
!= flow
;
3639 const struct nlattr
*encap
3640 = (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ENCAP
)
3641 ? attrs
[OVS_KEY_ATTR_ENCAP
] : NULL
);
3642 enum odp_key_fitness encap_fitness
;
3643 enum odp_key_fitness fitness
;
3645 /* Calculate fitness of outer attributes. */
3647 expected_attrs
|= ((UINT64_C(1) << OVS_KEY_ATTR_VLAN
) |
3648 (UINT64_C(1) << OVS_KEY_ATTR_ENCAP
));
3650 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_VLAN
)) {
3651 expected_attrs
|= (UINT64_C(1) << OVS_KEY_ATTR_VLAN
);
3653 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ENCAP
)) {
3654 expected_attrs
|= (UINT64_C(1) << OVS_KEY_ATTR_ENCAP
);
3657 fitness
= check_expectations(present_attrs
, out_of_range_attr
,
3658 expected_attrs
, key
, key_len
);
3661 * Remove the TPID from dl_type since it's not the real Ethertype. */
3662 flow
->dl_type
= htons(0);
3663 flow
->vlan_tci
= (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_VLAN
)
3664 ? nl_attr_get_be16(attrs
[OVS_KEY_ATTR_VLAN
])
3667 if (!(present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_VLAN
))) {
3668 return ODP_FIT_TOO_LITTLE
;
3669 } else if (flow
->vlan_tci
== htons(0)) {
3670 /* Corner case for a truncated 802.1Q header. */
3671 if (fitness
== ODP_FIT_PERFECT
&& nl_attr_get_size(encap
)) {
3672 return ODP_FIT_TOO_MUCH
;
3675 } else if (!(flow
->vlan_tci
& htons(VLAN_CFI
))) {
3676 VLOG_ERR_RL(&rl
, "OVS_KEY_ATTR_VLAN 0x%04"PRIx16
" is nonzero "
3677 "but CFI bit is not set", ntohs(flow
->vlan_tci
));
3678 return ODP_FIT_ERROR
;
3681 if (!(present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ENCAP
))) {
3686 /* Now parse the encapsulated attributes. */
3687 if (!parse_flow_nlattrs(nl_attr_get(encap
), nl_attr_get_size(encap
),
3688 attrs
, &present_attrs
, &out_of_range_attr
)) {
3689 return ODP_FIT_ERROR
;
3693 if (!parse_ethertype(attrs
, present_attrs
, &expected_attrs
, flow
, src_flow
)) {
3694 return ODP_FIT_ERROR
;
3696 encap_fitness
= parse_l2_5_onward(attrs
, present_attrs
, out_of_range_attr
,
3697 expected_attrs
, flow
, key
, key_len
,
3700 /* The overall fitness is the worse of the outer and inner attributes. */
3701 return MAX(fitness
, encap_fitness
);
3704 static enum odp_key_fitness
3705 odp_flow_key_to_flow__(const struct nlattr
*key
, size_t key_len
,
3706 struct flow
*flow
, const struct flow
*src_flow
)
3708 const struct nlattr
*attrs
[OVS_KEY_ATTR_MAX
+ 1];
3709 uint64_t expected_attrs
;
3710 uint64_t present_attrs
;
3711 int out_of_range_attr
;
3712 bool is_mask
= src_flow
!= flow
;
3714 memset(flow
, 0, sizeof *flow
);
3716 /* Parse attributes. */
3717 if (!parse_flow_nlattrs(key
, key_len
, attrs
, &present_attrs
,
3718 &out_of_range_attr
)) {
3719 return ODP_FIT_ERROR
;
3724 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_RECIRC_ID
)) {
3725 flow
->recirc_id
= nl_attr_get_u32(attrs
[OVS_KEY_ATTR_RECIRC_ID
]);
3726 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_RECIRC_ID
;
3727 } else if (is_mask
) {
3728 /* Always exact match recirc_id if it is not specified. */
3729 flow
->recirc_id
= UINT32_MAX
;
3732 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_DP_HASH
)) {
3733 flow
->dp_hash
= nl_attr_get_u32(attrs
[OVS_KEY_ATTR_DP_HASH
]);
3734 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_DP_HASH
;
3736 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_PRIORITY
)) {
3737 flow
->skb_priority
= nl_attr_get_u32(attrs
[OVS_KEY_ATTR_PRIORITY
]);
3738 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_PRIORITY
;
3741 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_SKB_MARK
)) {
3742 flow
->pkt_mark
= nl_attr_get_u32(attrs
[OVS_KEY_ATTR_SKB_MARK
]);
3743 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_SKB_MARK
;
3746 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_TUNNEL
)) {
3747 enum odp_key_fitness res
;
3749 res
= odp_tun_key_from_attr(attrs
[OVS_KEY_ATTR_TUNNEL
], &flow
->tunnel
);
3750 if (res
== ODP_FIT_ERROR
) {
3751 return ODP_FIT_ERROR
;
3752 } else if (res
== ODP_FIT_PERFECT
) {
3753 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_TUNNEL
;
3757 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_IN_PORT
)) {
3758 flow
->in_port
.odp_port
3759 = nl_attr_get_odp_port(attrs
[OVS_KEY_ATTR_IN_PORT
]);
3760 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_IN_PORT
;
3761 } else if (!is_mask
) {
3762 flow
->in_port
.odp_port
= ODPP_NONE
;
3765 /* Ethernet header. */
3766 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_ETHERNET
)) {
3767 const struct ovs_key_ethernet
*eth_key
;
3769 eth_key
= nl_attr_get(attrs
[OVS_KEY_ATTR_ETHERNET
]);
3770 put_ethernet_key(eth_key
, flow
);
3772 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ETHERNET
;
3776 expected_attrs
|= UINT64_C(1) << OVS_KEY_ATTR_ETHERNET
;
3779 /* Get Ethertype or 802.1Q TPID or FLOW_DL_TYPE_NONE. */
3780 if (!parse_ethertype(attrs
, present_attrs
, &expected_attrs
, flow
,
3782 return ODP_FIT_ERROR
;
3786 ? (src_flow
->vlan_tci
& htons(VLAN_CFI
)) != 0
3787 : src_flow
->dl_type
== htons(ETH_TYPE_VLAN
)) {
3788 return parse_8021q_onward(attrs
, present_attrs
, out_of_range_attr
,
3789 expected_attrs
, flow
, key
, key_len
, src_flow
);
3792 flow
->vlan_tci
= htons(0xffff);
3793 if (present_attrs
& (UINT64_C(1) << OVS_KEY_ATTR_VLAN
)) {
3794 flow
->vlan_tci
= nl_attr_get_be16(attrs
[OVS_KEY_ATTR_VLAN
]);
3795 expected_attrs
|= (UINT64_C(1) << OVS_KEY_ATTR_VLAN
);
3798 return parse_l2_5_onward(attrs
, present_attrs
, out_of_range_attr
,
3799 expected_attrs
, flow
, key
, key_len
, src_flow
);
3802 /* Converts the 'key_len' bytes of OVS_KEY_ATTR_* attributes in 'key' to a flow
3803 * structure in 'flow'. Returns an ODP_FIT_* value that indicates how well
3804 * 'key' fits our expectations for what a flow key should contain.
3806 * The 'in_port' will be the datapath's understanding of the port. The
3807 * caller will need to translate with odp_port_to_ofp_port() if the
3808 * OpenFlow port is needed.
3810 * This function doesn't take the packet itself as an argument because none of
3811 * the currently understood OVS_KEY_ATTR_* attributes require it. Currently,
3812 * it is always possible to infer which additional attribute(s) should appear
3813 * by looking at the attributes for lower-level protocols, e.g. if the network
3814 * protocol in OVS_KEY_ATTR_IPV4 or OVS_KEY_ATTR_IPV6 is IPPROTO_TCP then we
3815 * know that a OVS_KEY_ATTR_TCP attribute must appear and that otherwise it
3816 * must be absent. */
3817 enum odp_key_fitness
3818 odp_flow_key_to_flow(const struct nlattr
*key
, size_t key_len
,
3821 return odp_flow_key_to_flow__(key
, key_len
, flow
, flow
);
3824 /* Converts the 'key_len' bytes of OVS_KEY_ATTR_* attributes in 'key' to a mask
3825 * structure in 'mask'. 'flow' must be a previously translated flow
3826 * corresponding to 'mask'. Returns an ODP_FIT_* value that indicates how well
3827 * 'key' fits our expectations for what a flow key should contain. */
3828 enum odp_key_fitness
3829 odp_flow_key_to_mask(const struct nlattr
*key
, size_t key_len
,
3830 struct flow
*mask
, const struct flow
*flow
)
3832 return odp_flow_key_to_flow__(key
, key_len
, mask
, flow
);
3835 /* Returns 'fitness' as a string, for use in debug messages. */
3837 odp_key_fitness_to_string(enum odp_key_fitness fitness
)
3840 case ODP_FIT_PERFECT
:
3842 case ODP_FIT_TOO_MUCH
:
3844 case ODP_FIT_TOO_LITTLE
:
3845 return "too_little";
3853 /* Appends an OVS_ACTION_ATTR_USERSPACE action to 'odp_actions' that specifies
3854 * Netlink PID 'pid'. If 'userdata' is nonnull, adds a userdata attribute
3855 * whose contents are the 'userdata_size' bytes at 'userdata' and returns the
3856 * offset within 'odp_actions' of the start of the cookie. (If 'userdata' is
3857 * null, then the return value is not meaningful.) */
3859 odp_put_userspace_action(uint32_t pid
,
3860 const void *userdata
, size_t userdata_size
,
3861 odp_port_t tunnel_out_port
,
3862 struct ofpbuf
*odp_actions
)
3864 size_t userdata_ofs
;
3867 offset
= nl_msg_start_nested(odp_actions
, OVS_ACTION_ATTR_USERSPACE
);
3868 nl_msg_put_u32(odp_actions
, OVS_USERSPACE_ATTR_PID
, pid
);
3870 userdata_ofs
= odp_actions
->size
+ NLA_HDRLEN
;
3872 /* The OVS kernel module before OVS 1.11 and the upstream Linux kernel
3873 * module before Linux 3.10 required the userdata to be exactly 8 bytes
3876 * - The kernel rejected shorter userdata with -ERANGE.
3878 * - The kernel silently dropped userdata beyond the first 8 bytes.
3880 * Thus, for maximum compatibility, always put at least 8 bytes. (We
3881 * separately disable features that required more than 8 bytes.) */
3882 memcpy(nl_msg_put_unspec_zero(odp_actions
, OVS_USERSPACE_ATTR_USERDATA
,
3883 MAX(8, userdata_size
)),
3884 userdata
, userdata_size
);
3888 if (tunnel_out_port
!= ODPP_NONE
) {
3889 nl_msg_put_odp_port(odp_actions
, OVS_USERSPACE_ATTR_EGRESS_TUN_PORT
,
3892 nl_msg_end_nested(odp_actions
, offset
);
3894 return userdata_ofs
;
3898 odp_put_tunnel_action(const struct flow_tnl
*tunnel
,
3899 struct ofpbuf
*odp_actions
)
3901 size_t offset
= nl_msg_start_nested(odp_actions
, OVS_ACTION_ATTR_SET
);
3902 tun_key_to_attr(odp_actions
, tunnel
);
3903 nl_msg_end_nested(odp_actions
, offset
);
3907 odp_put_tnl_push_action(struct ofpbuf
*odp_actions
,
3908 struct ovs_action_push_tnl
*data
)
3910 int size
= offsetof(struct ovs_action_push_tnl
, header
);
3912 size
+= data
->header_len
;
3913 nl_msg_put_unspec(odp_actions
, OVS_ACTION_ATTR_TUNNEL_PUSH
, data
, size
);
3917 /* The commit_odp_actions() function and its helpers. */
3920 commit_set_action(struct ofpbuf
*odp_actions
, enum ovs_key_attr key_type
,
3921 const void *key
, size_t key_size
)
3923 size_t offset
= nl_msg_start_nested(odp_actions
, OVS_ACTION_ATTR_SET
);
3924 nl_msg_put_unspec(odp_actions
, key_type
, key
, key_size
);
3925 nl_msg_end_nested(odp_actions
, offset
);
3928 /* Masked set actions have a mask following the data within the netlink
3929 * attribute. The unmasked bits in the data will be cleared as the data
3930 * is copied to the action. */
3932 commit_masked_set_action(struct ofpbuf
*odp_actions
,
3933 enum ovs_key_attr key_type
,
3934 const void *key_
, const void *mask_
, size_t key_size
)
3936 size_t offset
= nl_msg_start_nested(odp_actions
,
3937 OVS_ACTION_ATTR_SET_MASKED
);
3938 char *data
= nl_msg_put_unspec_uninit(odp_actions
, key_type
, key_size
* 2);
3939 const char *key
= key_
, *mask
= mask_
;
3941 memcpy(data
+ key_size
, mask
, key_size
);
3942 /* Clear unmasked bits while copying. */
3943 while (key_size
--) {
3944 *data
++ = *key
++ & *mask
++;
3946 nl_msg_end_nested(odp_actions
, offset
);
3949 /* If any of the flow key data that ODP actions can modify are different in
3950 * 'base->tunnel' and 'flow->tunnel', appends a set_tunnel ODP action to
3951 * 'odp_actions' that change the flow tunneling information in key from
3952 * 'base->tunnel' into 'flow->tunnel', and then changes 'base->tunnel' in the
3953 * same way. In other words, operates the same as commit_odp_actions(), but
3954 * only on tunneling information. */
3956 commit_odp_tunnel_action(const struct flow
*flow
, struct flow
*base
,
3957 struct ofpbuf
*odp_actions
)
3959 /* A valid IPV4_TUNNEL must have non-zero ip_dst. */
3960 if (flow
->tunnel
.ip_dst
) {
3961 if (!memcmp(&base
->tunnel
, &flow
->tunnel
, sizeof base
->tunnel
)) {
3964 memcpy(&base
->tunnel
, &flow
->tunnel
, sizeof base
->tunnel
);
3965 odp_put_tunnel_action(&base
->tunnel
, odp_actions
);
3970 commit(enum ovs_key_attr attr
, bool use_masked_set
,
3971 const void *key
, void *base
, void *mask
, size_t size
,
3972 struct ofpbuf
*odp_actions
)
3974 if (memcmp(key
, base
, size
)) {
3975 bool fully_masked
= odp_mask_is_exact(attr
, mask
, size
);
3977 if (use_masked_set
&& !fully_masked
) {
3978 commit_masked_set_action(odp_actions
, attr
, key
, mask
, size
);
3980 if (!fully_masked
) {
3981 memset(mask
, 0xff, size
);
3983 commit_set_action(odp_actions
, attr
, key
, size
);
3985 memcpy(base
, key
, size
);
3988 /* Mask bits are set when we have either read or set the corresponding
3989 * values. Masked bits will be exact-matched, no need to set them
3990 * if the value did not actually change. */
3996 get_ethernet_key(const struct flow
*flow
, struct ovs_key_ethernet
*eth
)
3998 memcpy(eth
->eth_src
, flow
->dl_src
, ETH_ADDR_LEN
);
3999 memcpy(eth
->eth_dst
, flow
->dl_dst
, ETH_ADDR_LEN
);
4003 put_ethernet_key(const struct ovs_key_ethernet
*eth
, struct flow
*flow
)
4005 memcpy(flow
->dl_src
, eth
->eth_src
, ETH_ADDR_LEN
);
4006 memcpy(flow
->dl_dst
, eth
->eth_dst
, ETH_ADDR_LEN
);
4010 commit_set_ether_addr_action(const struct flow
*flow
, struct flow
*base_flow
,
4011 struct ofpbuf
*odp_actions
,
4012 struct flow_wildcards
*wc
,
4015 struct ovs_key_ethernet key
, base
, mask
;
4017 get_ethernet_key(flow
, &key
);
4018 get_ethernet_key(base_flow
, &base
);
4019 get_ethernet_key(&wc
->masks
, &mask
);
4021 if (commit(OVS_KEY_ATTR_ETHERNET
, use_masked
,
4022 &key
, &base
, &mask
, sizeof key
, odp_actions
)) {
4023 put_ethernet_key(&base
, base_flow
);
4024 put_ethernet_key(&mask
, &wc
->masks
);
4029 pop_vlan(struct flow
*base
,
4030 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
)
4032 memset(&wc
->masks
.vlan_tci
, 0xff, sizeof wc
->masks
.vlan_tci
);
4034 if (base
->vlan_tci
& htons(VLAN_CFI
)) {
4035 nl_msg_put_flag(odp_actions
, OVS_ACTION_ATTR_POP_VLAN
);
4041 commit_vlan_action(ovs_be16 vlan_tci
, struct flow
*base
,
4042 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
)
4044 if (base
->vlan_tci
== vlan_tci
) {
4048 pop_vlan(base
, odp_actions
, wc
);
4049 if (vlan_tci
& htons(VLAN_CFI
)) {
4050 struct ovs_action_push_vlan vlan
;
4052 vlan
.vlan_tpid
= htons(ETH_TYPE_VLAN
);
4053 vlan
.vlan_tci
= vlan_tci
;
4054 nl_msg_put_unspec(odp_actions
, OVS_ACTION_ATTR_PUSH_VLAN
,
4055 &vlan
, sizeof vlan
);
4057 base
->vlan_tci
= vlan_tci
;
4060 /* Wildcarding already done at action translation time. */
4062 commit_mpls_action(const struct flow
*flow
, struct flow
*base
,
4063 struct ofpbuf
*odp_actions
)
4065 int base_n
= flow_count_mpls_labels(base
, NULL
);
4066 int flow_n
= flow_count_mpls_labels(flow
, NULL
);
4067 int common_n
= flow_count_common_mpls_labels(flow
, flow_n
, base
, base_n
,
4070 while (base_n
> common_n
) {
4071 if (base_n
- 1 == common_n
&& flow_n
> common_n
) {
4072 /* If there is only one more LSE in base than there are common
4073 * between base and flow; and flow has at least one more LSE than
4074 * is common then the topmost LSE of base may be updated using
4076 struct ovs_key_mpls mpls_key
;
4078 mpls_key
.mpls_lse
= flow
->mpls_lse
[flow_n
- base_n
];
4079 commit_set_action(odp_actions
, OVS_KEY_ATTR_MPLS
,
4080 &mpls_key
, sizeof mpls_key
);
4081 flow_set_mpls_lse(base
, 0, mpls_key
.mpls_lse
);
4084 /* Otherwise, if there more LSEs in base than are common between
4085 * base and flow then pop the topmost one. */
4089 /* If all the LSEs are to be popped and this is not the outermost
4090 * LSE then use ETH_TYPE_MPLS as the ethertype parameter of the
4091 * POP_MPLS action instead of flow->dl_type.
4093 * This is because the POP_MPLS action requires its ethertype
4094 * argument to be an MPLS ethernet type but in this case
4095 * flow->dl_type will be a non-MPLS ethernet type.
4097 * When the final POP_MPLS action occurs it use flow->dl_type and
4098 * the and the resulting packet will have the desired dl_type. */
4099 if ((!eth_type_mpls(flow
->dl_type
)) && base_n
> 1) {
4100 dl_type
= htons(ETH_TYPE_MPLS
);
4102 dl_type
= flow
->dl_type
;
4104 nl_msg_put_be16(odp_actions
, OVS_ACTION_ATTR_POP_MPLS
, dl_type
);
4105 popped
= flow_pop_mpls(base
, base_n
, flow
->dl_type
, NULL
);
4111 /* If, after the above popping and setting, there are more LSEs in flow
4112 * than base then some LSEs need to be pushed. */
4113 while (base_n
< flow_n
) {
4114 struct ovs_action_push_mpls
*mpls
;
4116 mpls
= nl_msg_put_unspec_zero(odp_actions
,
4117 OVS_ACTION_ATTR_PUSH_MPLS
,
4119 mpls
->mpls_ethertype
= flow
->dl_type
;
4120 mpls
->mpls_lse
= flow
->mpls_lse
[flow_n
- base_n
- 1];
4121 flow_push_mpls(base
, base_n
, mpls
->mpls_ethertype
, NULL
);
4122 flow_set_mpls_lse(base
, 0, mpls
->mpls_lse
);
4128 get_ipv4_key(const struct flow
*flow
, struct ovs_key_ipv4
*ipv4
, bool is_mask
)
4130 ipv4
->ipv4_src
= flow
->nw_src
;
4131 ipv4
->ipv4_dst
= flow
->nw_dst
;
4132 ipv4
->ipv4_proto
= flow
->nw_proto
;
4133 ipv4
->ipv4_tos
= flow
->nw_tos
;
4134 ipv4
->ipv4_ttl
= flow
->nw_ttl
;
4135 ipv4
->ipv4_frag
= ovs_to_odp_frag(flow
->nw_frag
, is_mask
);
4139 put_ipv4_key(const struct ovs_key_ipv4
*ipv4
, struct flow
*flow
, bool is_mask
)
4141 flow
->nw_src
= ipv4
->ipv4_src
;
4142 flow
->nw_dst
= ipv4
->ipv4_dst
;
4143 flow
->nw_proto
= ipv4
->ipv4_proto
;
4144 flow
->nw_tos
= ipv4
->ipv4_tos
;
4145 flow
->nw_ttl
= ipv4
->ipv4_ttl
;
4146 flow
->nw_frag
= odp_to_ovs_frag(ipv4
->ipv4_frag
, is_mask
);
4150 commit_set_ipv4_action(const struct flow
*flow
, struct flow
*base_flow
,
4151 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
,
4154 struct ovs_key_ipv4 key
, mask
, base
;
4156 /* Check that nw_proto and nw_frag remain unchanged. */
4157 ovs_assert(flow
->nw_proto
== base_flow
->nw_proto
&&
4158 flow
->nw_frag
== base_flow
->nw_frag
);
4160 get_ipv4_key(flow
, &key
, false);
4161 get_ipv4_key(base_flow
, &base
, false);
4162 get_ipv4_key(&wc
->masks
, &mask
, true);
4163 mask
.ipv4_proto
= 0; /* Not writeable. */
4164 mask
.ipv4_frag
= 0; /* Not writable. */
4166 if (commit(OVS_KEY_ATTR_IPV4
, use_masked
, &key
, &base
, &mask
, sizeof key
,
4168 put_ipv4_key(&base
, base_flow
, false);
4169 if (mask
.ipv4_proto
!= 0) { /* Mask was changed by commit(). */
4170 put_ipv4_key(&mask
, &wc
->masks
, true);
4176 get_ipv6_key(const struct flow
*flow
, struct ovs_key_ipv6
*ipv6
, bool is_mask
)
4178 memcpy(ipv6
->ipv6_src
, &flow
->ipv6_src
, sizeof ipv6
->ipv6_src
);
4179 memcpy(ipv6
->ipv6_dst
, &flow
->ipv6_dst
, sizeof ipv6
->ipv6_dst
);
4180 ipv6
->ipv6_label
= flow
->ipv6_label
;
4181 ipv6
->ipv6_proto
= flow
->nw_proto
;
4182 ipv6
->ipv6_tclass
= flow
->nw_tos
;
4183 ipv6
->ipv6_hlimit
= flow
->nw_ttl
;
4184 ipv6
->ipv6_frag
= ovs_to_odp_frag(flow
->nw_frag
, is_mask
);
4188 put_ipv6_key(const struct ovs_key_ipv6
*ipv6
, struct flow
*flow
, bool is_mask
)
4190 memcpy(&flow
->ipv6_src
, ipv6
->ipv6_src
, sizeof flow
->ipv6_src
);
4191 memcpy(&flow
->ipv6_dst
, ipv6
->ipv6_dst
, sizeof flow
->ipv6_dst
);
4192 flow
->ipv6_label
= ipv6
->ipv6_label
;
4193 flow
->nw_proto
= ipv6
->ipv6_proto
;
4194 flow
->nw_tos
= ipv6
->ipv6_tclass
;
4195 flow
->nw_ttl
= ipv6
->ipv6_hlimit
;
4196 flow
->nw_frag
= odp_to_ovs_frag(ipv6
->ipv6_frag
, is_mask
);
4200 commit_set_ipv6_action(const struct flow
*flow
, struct flow
*base_flow
,
4201 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
,
4204 struct ovs_key_ipv6 key
, mask
, base
;
4206 /* Check that nw_proto and nw_frag remain unchanged. */
4207 ovs_assert(flow
->nw_proto
== base_flow
->nw_proto
&&
4208 flow
->nw_frag
== base_flow
->nw_frag
);
4210 get_ipv6_key(flow
, &key
, false);
4211 get_ipv6_key(base_flow
, &base
, false);
4212 get_ipv6_key(&wc
->masks
, &mask
, true);
4213 mask
.ipv6_proto
= 0; /* Not writeable. */
4214 mask
.ipv6_frag
= 0; /* Not writable. */
4216 if (commit(OVS_KEY_ATTR_IPV6
, use_masked
, &key
, &base
, &mask
, sizeof key
,
4218 put_ipv6_key(&base
, base_flow
, false);
4219 if (mask
.ipv6_proto
!= 0) { /* Mask was changed by commit(). */
4220 put_ipv6_key(&mask
, &wc
->masks
, true);
4226 get_arp_key(const struct flow
*flow
, struct ovs_key_arp
*arp
)
4228 /* ARP key has padding, clear it. */
4229 memset(arp
, 0, sizeof *arp
);
4231 arp
->arp_sip
= flow
->nw_src
;
4232 arp
->arp_tip
= flow
->nw_dst
;
4233 arp
->arp_op
= htons(flow
->nw_proto
);
4234 memcpy(arp
->arp_sha
, flow
->arp_sha
, ETH_ADDR_LEN
);
4235 memcpy(arp
->arp_tha
, flow
->arp_tha
, ETH_ADDR_LEN
);
4239 put_arp_key(const struct ovs_key_arp
*arp
, struct flow
*flow
)
4241 flow
->nw_src
= arp
->arp_sip
;
4242 flow
->nw_dst
= arp
->arp_tip
;
4243 flow
->nw_proto
= ntohs(arp
->arp_op
);
4244 memcpy(flow
->arp_sha
, arp
->arp_sha
, ETH_ADDR_LEN
);
4245 memcpy(flow
->arp_tha
, arp
->arp_tha
, ETH_ADDR_LEN
);
4248 static enum slow_path_reason
4249 commit_set_arp_action(const struct flow
*flow
, struct flow
*base_flow
,
4250 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
)
4252 struct ovs_key_arp key
, mask
, base
;
4254 get_arp_key(flow
, &key
);
4255 get_arp_key(base_flow
, &base
);
4256 get_arp_key(&wc
->masks
, &mask
);
4258 if (commit(OVS_KEY_ATTR_ARP
, true, &key
, &base
, &mask
, sizeof key
,
4260 put_arp_key(&base
, base_flow
);
4261 put_arp_key(&mask
, &wc
->masks
);
4268 get_nd_key(const struct flow
*flow
, struct ovs_key_nd
*nd
)
4270 memcpy(nd
->nd_target
, &flow
->nd_target
, sizeof flow
->nd_target
);
4271 /* nd_sll and nd_tll are stored in arp_sha and arp_tha, respectively */
4272 memcpy(nd
->nd_sll
, flow
->arp_sha
, ETH_ADDR_LEN
);
4273 memcpy(nd
->nd_tll
, flow
->arp_tha
, ETH_ADDR_LEN
);
4277 put_nd_key(const struct ovs_key_nd
*nd
, struct flow
*flow
)
4279 memcpy(&flow
->nd_target
, &flow
->nd_target
, sizeof flow
->nd_target
);
4280 /* nd_sll and nd_tll are stored in arp_sha and arp_tha, respectively */
4281 memcpy(flow
->arp_sha
, nd
->nd_sll
, ETH_ADDR_LEN
);
4282 memcpy(flow
->arp_tha
, nd
->nd_tll
, ETH_ADDR_LEN
);
4285 static enum slow_path_reason
4286 commit_set_nd_action(const struct flow
*flow
, struct flow
*base_flow
,
4287 struct ofpbuf
*odp_actions
,
4288 struct flow_wildcards
*wc
, bool use_masked
)
4290 struct ovs_key_nd key
, mask
, base
;
4292 get_nd_key(flow
, &key
);
4293 get_nd_key(base_flow
, &base
);
4294 get_nd_key(&wc
->masks
, &mask
);
4296 if (commit(OVS_KEY_ATTR_ND
, use_masked
, &key
, &base
, &mask
, sizeof key
,
4298 put_nd_key(&base
, base_flow
);
4299 put_nd_key(&mask
, &wc
->masks
);
4306 static enum slow_path_reason
4307 commit_set_nw_action(const struct flow
*flow
, struct flow
*base
,
4308 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
,
4311 /* Check if 'flow' really has an L3 header. */
4312 if (!flow
->nw_proto
) {
4316 switch (ntohs(base
->dl_type
)) {
4318 commit_set_ipv4_action(flow
, base
, odp_actions
, wc
, use_masked
);
4322 commit_set_ipv6_action(flow
, base
, odp_actions
, wc
, use_masked
);
4323 return commit_set_nd_action(flow
, base
, odp_actions
, wc
, use_masked
);
4326 return commit_set_arp_action(flow
, base
, odp_actions
, wc
);
4332 /* TCP, UDP, and SCTP keys have the same layout. */
4333 BUILD_ASSERT_DECL(sizeof(struct ovs_key_tcp
) == sizeof(struct ovs_key_udp
) &&
4334 sizeof(struct ovs_key_tcp
) == sizeof(struct ovs_key_sctp
));
4337 get_tp_key(const struct flow
*flow
, union ovs_key_tp
*tp
)
4339 tp
->tcp
.tcp_src
= flow
->tp_src
;
4340 tp
->tcp
.tcp_dst
= flow
->tp_dst
;
4344 put_tp_key(const union ovs_key_tp
*tp
, struct flow
*flow
)
4346 flow
->tp_src
= tp
->tcp
.tcp_src
;
4347 flow
->tp_dst
= tp
->tcp
.tcp_dst
;
4351 commit_set_port_action(const struct flow
*flow
, struct flow
*base_flow
,
4352 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
,
4355 enum ovs_key_attr key_type
;
4356 union ovs_key_tp key
, mask
, base
;
4358 /* Check if 'flow' really has an L3 header. */
4359 if (!flow
->nw_proto
) {
4363 if (!is_ip_any(base_flow
)) {
4367 if (flow
->nw_proto
== IPPROTO_TCP
) {
4368 key_type
= OVS_KEY_ATTR_TCP
;
4369 } else if (flow
->nw_proto
== IPPROTO_UDP
) {
4370 key_type
= OVS_KEY_ATTR_UDP
;
4371 } else if (flow
->nw_proto
== IPPROTO_SCTP
) {
4372 key_type
= OVS_KEY_ATTR_SCTP
;
4377 get_tp_key(flow
, &key
);
4378 get_tp_key(base_flow
, &base
);
4379 get_tp_key(&wc
->masks
, &mask
);
4381 if (commit(key_type
, use_masked
, &key
, &base
, &mask
, sizeof key
,
4383 put_tp_key(&base
, base_flow
);
4384 put_tp_key(&mask
, &wc
->masks
);
4389 commit_set_priority_action(const struct flow
*flow
, struct flow
*base_flow
,
4390 struct ofpbuf
*odp_actions
,
4391 struct flow_wildcards
*wc
,
4394 uint32_t key
, mask
, base
;
4396 key
= flow
->skb_priority
;
4397 base
= base_flow
->skb_priority
;
4398 mask
= wc
->masks
.skb_priority
;
4400 if (commit(OVS_KEY_ATTR_PRIORITY
, use_masked
, &key
, &base
, &mask
,
4401 sizeof key
, odp_actions
)) {
4402 base_flow
->skb_priority
= base
;
4403 wc
->masks
.skb_priority
= mask
;
4408 commit_set_pkt_mark_action(const struct flow
*flow
, struct flow
*base_flow
,
4409 struct ofpbuf
*odp_actions
,
4410 struct flow_wildcards
*wc
,
4413 uint32_t key
, mask
, base
;
4415 key
= flow
->pkt_mark
;
4416 base
= base_flow
->pkt_mark
;
4417 mask
= wc
->masks
.pkt_mark
;
4419 if (commit(OVS_KEY_ATTR_SKB_MARK
, use_masked
, &key
, &base
, &mask
,
4420 sizeof key
, odp_actions
)) {
4421 base_flow
->pkt_mark
= base
;
4422 wc
->masks
.pkt_mark
= mask
;
4426 /* If any of the flow key data that ODP actions can modify are different in
4427 * 'base' and 'flow', appends ODP actions to 'odp_actions' that change the flow
4428 * key from 'base' into 'flow', and then changes 'base' the same way. Does not
4429 * commit set_tunnel actions. Users should call commit_odp_tunnel_action()
4430 * in addition to this function if needed. Sets fields in 'wc' that are
4431 * used as part of the action.
4433 * Returns a reason to force processing the flow's packets into the userspace
4434 * slow path, if there is one, otherwise 0. */
4435 enum slow_path_reason
4436 commit_odp_actions(const struct flow
*flow
, struct flow
*base
,
4437 struct ofpbuf
*odp_actions
, struct flow_wildcards
*wc
,
4440 enum slow_path_reason slow
;
4442 commit_set_ether_addr_action(flow
, base
, odp_actions
, wc
, use_masked
);
4443 slow
= commit_set_nw_action(flow
, base
, odp_actions
, wc
, use_masked
);
4444 commit_set_port_action(flow
, base
, odp_actions
, wc
, use_masked
);
4445 commit_mpls_action(flow
, base
, odp_actions
);
4446 commit_vlan_action(flow
->vlan_tci
, base
, odp_actions
, wc
);
4447 commit_set_priority_action(flow
, base
, odp_actions
, wc
, use_masked
);
4448 commit_set_pkt_mark_action(flow
, base
, odp_actions
, wc
, use_masked
);