2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 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>
27 #include "byte-order.h"
29 #include "dynamic-string.h"
39 VLOG_DEFINE_THIS_MODULE(odp_util);
41 /* The interface between userspace and kernel uses an "OVS_*" prefix.
42 * Since this is fairly non-specific for the OVS userspace components,
43 * "ODP_*" (Open vSwitch Datapath) is used as the prefix for
44 * interactions with the datapath.
47 /* The set of characters that may separate one action or one key attribute
49 static const char *delimiters = ", \t\r\n";
51 static int parse_odp_key_attr(const char *, const struct simap *port_names,
53 static void format_odp_key_attr(const struct nlattr *a, struct ds *ds);
55 /* Returns one the following for the action with the given OVS_ACTION_ATTR_*
58 * - For an action whose argument has a fixed length, returned that
59 * nonnegative length in bytes.
61 * - For an action with a variable-length argument, returns -2.
63 * - For an invalid 'type', returns -1. */
65 odp_action_len(uint16_t type)
67 if (type > OVS_ACTION_ATTR_MAX) {
71 switch ((enum ovs_action_attr) type) {
72 case OVS_ACTION_ATTR_OUTPUT: return sizeof(uint32_t);
73 case OVS_ACTION_ATTR_USERSPACE: return -2;
74 case OVS_ACTION_ATTR_PUSH_VLAN: return sizeof(struct ovs_action_push_vlan);
75 case OVS_ACTION_ATTR_POP_VLAN: return 0;
76 case OVS_ACTION_ATTR_PUSH_MPLS: return sizeof(struct ovs_action_push_mpls);
77 case OVS_ACTION_ATTR_POP_MPLS: return sizeof(ovs_be16);
78 case OVS_ACTION_ATTR_SET: return -2;
79 case OVS_ACTION_ATTR_SAMPLE: return -2;
81 case OVS_ACTION_ATTR_UNSPEC:
82 case __OVS_ACTION_ATTR_MAX:
90 ovs_key_attr_to_string(enum ovs_key_attr attr)
92 static char unknown_attr[3 + INT_STRLEN(unsigned int) + 1];
95 case OVS_KEY_ATTR_UNSPEC: return "unspec";
96 case OVS_KEY_ATTR_ENCAP: return "encap";
97 case OVS_KEY_ATTR_PRIORITY: return "skb_priority";
98 case OVS_KEY_ATTR_SKB_MARK: return "skb_mark";
99 case OVS_KEY_ATTR_TUNNEL: return "tunnel";
100 case OVS_KEY_ATTR_IN_PORT: return "in_port";
101 case OVS_KEY_ATTR_ETHERNET: return "eth";
102 case OVS_KEY_ATTR_VLAN: return "vlan";
103 case OVS_KEY_ATTR_ETHERTYPE: return "eth_type";
104 case OVS_KEY_ATTR_IPV4: return "ipv4";
105 case OVS_KEY_ATTR_IPV6: return "ipv6";
106 case OVS_KEY_ATTR_TCP: return "tcp";
107 case OVS_KEY_ATTR_UDP: return "udp";
108 case OVS_KEY_ATTR_ICMP: return "icmp";
109 case OVS_KEY_ATTR_ICMPV6: return "icmpv6";
110 case OVS_KEY_ATTR_ARP: return "arp";
111 case OVS_KEY_ATTR_ND: return "nd";
112 case OVS_KEY_ATTR_MPLS: return "mpls";
114 case __OVS_KEY_ATTR_MAX:
116 snprintf(unknown_attr, sizeof unknown_attr, "key%u",
117 (unsigned int) attr);
123 format_generic_odp_action(struct ds *ds, const struct nlattr *a)
125 size_t len = nl_attr_get_size(a);
127 ds_put_format(ds, "action%"PRId16, nl_attr_type(a));
129 const uint8_t *unspec;
132 unspec = nl_attr_get(a);
133 for (i = 0; i < len; i++) {
134 ds_put_char(ds, i ? ' ': '(');
135 ds_put_format(ds, "%02x", unspec[i]);
137 ds_put_char(ds, ')');
142 format_odp_sample_action(struct ds *ds, const struct nlattr *attr)
144 static const struct nl_policy ovs_sample_policy[] = {
145 [OVS_SAMPLE_ATTR_PROBABILITY] = { .type = NL_A_U32 },
146 [OVS_SAMPLE_ATTR_ACTIONS] = { .type = NL_A_NESTED }
148 struct nlattr *a[ARRAY_SIZE(ovs_sample_policy)];
150 const struct nlattr *nla_acts;
153 ds_put_cstr(ds, "sample");
155 if (!nl_parse_nested(attr, ovs_sample_policy, a, ARRAY_SIZE(a))) {
156 ds_put_cstr(ds, "(error)");
160 percentage = (100.0 * nl_attr_get_u32(a[OVS_SAMPLE_ATTR_PROBABILITY])) /
163 ds_put_format(ds, "(sample=%.1f%%,", percentage);
165 ds_put_cstr(ds, "actions(");
166 nla_acts = nl_attr_get(a[OVS_SAMPLE_ATTR_ACTIONS]);
167 len = nl_attr_get_size(a[OVS_SAMPLE_ATTR_ACTIONS]);
168 format_odp_actions(ds, nla_acts, len);
169 ds_put_format(ds, "))");
173 slow_path_reason_to_string(enum slow_path_reason reason)
182 case SLOW_CONTROLLER:
190 static enum slow_path_reason
191 string_to_slow_path_reason(const char *string)
193 enum slow_path_reason i;
195 for (i = 1; i < __SLOW_MAX; i++) {
196 if (!strcmp(string, slow_path_reason_to_string(i))) {
205 parse_flags(const char *s, const char *(*bit_to_string)(uint32_t),
216 while (s[n] != ')') {
217 unsigned long long int flags;
221 if (sscanf(&s[n], "%lli%n", &flags, &n0) > 0 && n0 > 0) {
222 n += n0 + (s[n + n0] == ',');
227 for (bit = 1; bit; bit <<= 1) {
228 const char *name = bit_to_string(bit);
236 if (!strncmp(s + n, name, len) &&
237 (s[n + len] == ',' || s[n + len] == ')')) {
239 n += len + (s[n + len] == ',');
255 format_odp_userspace_action(struct ds *ds, const struct nlattr *attr)
257 static const struct nl_policy ovs_userspace_policy[] = {
258 [OVS_USERSPACE_ATTR_PID] = { .type = NL_A_U32 },
259 [OVS_USERSPACE_ATTR_USERDATA] = { .type = NL_A_UNSPEC,
262 struct nlattr *a[ARRAY_SIZE(ovs_userspace_policy)];
263 const struct nlattr *userdata_attr;
265 if (!nl_parse_nested(attr, ovs_userspace_policy, a, ARRAY_SIZE(a))) {
266 ds_put_cstr(ds, "userspace(error)");
270 ds_put_format(ds, "userspace(pid=%"PRIu32,
271 nl_attr_get_u32(a[OVS_USERSPACE_ATTR_PID]));
273 userdata_attr = a[OVS_USERSPACE_ATTR_USERDATA];
276 const uint8_t *userdata = nl_attr_get(userdata_attr);
277 size_t userdata_len = nl_attr_get_size(userdata_attr);
278 bool userdata_unspec = true;
279 union user_action_cookie cookie;
281 if (userdata_len >= sizeof cookie.type
282 && userdata_len <= sizeof cookie) {
284 memset(&cookie, 0, sizeof cookie);
285 memcpy(&cookie, userdata, userdata_len);
287 userdata_unspec = false;
289 if (userdata_len == sizeof cookie.sflow
290 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
291 ds_put_format(ds, ",sFlow("
292 "vid=%"PRIu16",pcp=%"PRIu8",output=%"PRIu32")",
293 vlan_tci_to_vid(cookie.sflow.vlan_tci),
294 vlan_tci_to_pcp(cookie.sflow.vlan_tci),
295 cookie.sflow.output);
296 } else if (userdata_len == sizeof cookie.slow_path
297 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
299 reason = slow_path_reason_to_string(cookie.slow_path.reason);
300 reason = reason ? reason : "";
301 ds_put_format(ds, ",slow_path(%s)", reason);
302 } else if (userdata_len == sizeof cookie.flow_sample
303 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
304 ds_put_format(ds, ",flow_sample(probability=%"PRIu16
305 ",collector_set_id=%"PRIu32
306 ",obs_domain_id=%"PRIu32
307 ",obs_point_id=%"PRIu32")",
308 cookie.flow_sample.probability,
309 cookie.flow_sample.collector_set_id,
310 cookie.flow_sample.obs_domain_id,
311 cookie.flow_sample.obs_point_id);
312 } else if (userdata_len == sizeof cookie.ipfix
313 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
314 ds_put_format(ds, ",ipfix");
316 userdata_unspec = true;
320 if (userdata_unspec) {
322 ds_put_format(ds, ",userdata(");
323 for (i = 0; i < userdata_len; i++) {
324 ds_put_format(ds, "%02x", userdata[i]);
326 ds_put_char(ds, ')');
330 ds_put_char(ds, ')');
334 format_vlan_tci(struct ds *ds, ovs_be16 vlan_tci)
336 ds_put_format(ds, "vid=%"PRIu16",pcp=%d",
337 vlan_tci_to_vid(vlan_tci),
338 vlan_tci_to_pcp(vlan_tci));
339 if (!(vlan_tci & htons(VLAN_CFI))) {
340 ds_put_cstr(ds, ",cfi=0");
345 format_mpls_lse(struct ds *ds, ovs_be32 mpls_lse)
347 ds_put_format(ds, "label=%"PRIu32",tc=%d,ttl=%d,bos=%d",
348 mpls_lse_to_label(mpls_lse),
349 mpls_lse_to_tc(mpls_lse),
350 mpls_lse_to_ttl(mpls_lse),
351 mpls_lse_to_bos(mpls_lse));
355 format_odp_action(struct ds *ds, const struct nlattr *a)
358 enum ovs_action_attr type = nl_attr_type(a);
359 const struct ovs_action_push_vlan *vlan;
361 expected_len = odp_action_len(nl_attr_type(a));
362 if (expected_len != -2 && nl_attr_get_size(a) != expected_len) {
363 ds_put_format(ds, "bad length %zu, expected %d for: ",
364 nl_attr_get_size(a), expected_len);
365 format_generic_odp_action(ds, a);
370 case OVS_ACTION_ATTR_OUTPUT:
371 ds_put_format(ds, "%"PRIu32, nl_attr_get_u32(a));
373 case OVS_ACTION_ATTR_USERSPACE:
374 format_odp_userspace_action(ds, a);
376 case OVS_ACTION_ATTR_SET:
377 ds_put_cstr(ds, "set(");
378 format_odp_key_attr(nl_attr_get(a), ds);
379 ds_put_cstr(ds, ")");
381 case OVS_ACTION_ATTR_PUSH_VLAN:
382 vlan = nl_attr_get(a);
383 ds_put_cstr(ds, "push_vlan(");
384 if (vlan->vlan_tpid != htons(ETH_TYPE_VLAN)) {
385 ds_put_format(ds, "tpid=0x%04"PRIx16",", ntohs(vlan->vlan_tpid));
387 format_vlan_tci(ds, vlan->vlan_tci);
388 ds_put_char(ds, ')');
390 case OVS_ACTION_ATTR_POP_VLAN:
391 ds_put_cstr(ds, "pop_vlan");
393 case OVS_ACTION_ATTR_PUSH_MPLS: {
394 const struct ovs_action_push_mpls *mpls = nl_attr_get(a);
395 ds_put_cstr(ds, "push_mpls(");
396 format_mpls_lse(ds, mpls->mpls_lse);
397 ds_put_format(ds, ",eth_type=0x%"PRIx16")", ntohs(mpls->mpls_ethertype));
400 case OVS_ACTION_ATTR_POP_MPLS: {
401 ovs_be16 ethertype = nl_attr_get_be16(a);
402 ds_put_format(ds, "pop_mpls(eth_type=0x%"PRIx16")", ntohs(ethertype));
405 case OVS_ACTION_ATTR_SAMPLE:
406 format_odp_sample_action(ds, a);
408 case OVS_ACTION_ATTR_UNSPEC:
409 case __OVS_ACTION_ATTR_MAX:
411 format_generic_odp_action(ds, a);
417 format_odp_actions(struct ds *ds, const struct nlattr *actions,
421 const struct nlattr *a;
424 NL_ATTR_FOR_EACH (a, left, actions, actions_len) {
426 ds_put_char(ds, ',');
428 format_odp_action(ds, a);
433 if (left == actions_len) {
434 ds_put_cstr(ds, "<empty>");
436 ds_put_format(ds, ",***%u leftover bytes*** (", left);
437 for (i = 0; i < left; i++) {
438 ds_put_format(ds, "%02x", ((const uint8_t *) a)[i]);
440 ds_put_char(ds, ')');
443 ds_put_cstr(ds, "drop");
448 parse_odp_action(const char *s, const struct simap *port_names,
449 struct ofpbuf *actions)
451 /* Many of the sscanf calls in this function use oversized destination
452 * fields because some sscanf() implementations truncate the range of %i
453 * directives, so that e.g. "%"SCNi16 interprets input of "0xfedc" as a
454 * value of 0x7fff. The other alternatives are to allow only a single
455 * radix (e.g. decimal or hexadecimal) or to write more sophisticated
458 * The tun_id parser has to use an alternative approach because there is no
459 * type larger than 64 bits. */
462 unsigned long long int port;
465 if (sscanf(s, "%lli%n", &port, &n) > 0 && n > 0) {
466 nl_msg_put_u32(actions, OVS_ACTION_ATTR_OUTPUT, port);
472 int len = strcspn(s, delimiters);
473 struct simap_node *node;
475 node = simap_find_len(port_names, s, len);
477 nl_msg_put_u32(actions, OVS_ACTION_ATTR_OUTPUT, node->data);
483 unsigned long long int pid;
484 unsigned long long int output;
485 unsigned long long int probability;
486 unsigned long long int collector_set_id;
487 unsigned long long int obs_domain_id;
488 unsigned long long int obs_point_id;
492 if (sscanf(s, "userspace(pid=%lli)%n", &pid, &n) > 0 && n > 0) {
493 odp_put_userspace_action(pid, NULL, 0, actions);
495 } else if (sscanf(s, "userspace(pid=%lli,sFlow(vid=%i,"
496 "pcp=%i,output=%lli))%n",
497 &pid, &vid, &pcp, &output, &n) > 0 && n > 0) {
498 union user_action_cookie cookie;
501 tci = vid | (pcp << VLAN_PCP_SHIFT);
506 cookie.type = USER_ACTION_COOKIE_SFLOW;
507 cookie.sflow.vlan_tci = htons(tci);
508 cookie.sflow.output = output;
509 odp_put_userspace_action(pid, &cookie, sizeof cookie.sflow,
512 } else if (sscanf(s, "userspace(pid=%lli,slow_path(%n", &pid, &n) > 0
514 union user_action_cookie cookie;
517 if (s[n] == ')' && s[n + 1] == ')') {
520 } else if (sscanf(s + n, "%31[^)]))", reason) > 0) {
521 n += strlen(reason) + 2;
526 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
527 cookie.slow_path.unused = 0;
528 cookie.slow_path.reason = string_to_slow_path_reason(reason);
530 if (reason[0] && !cookie.slow_path.reason) {
534 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path,
537 } else if (sscanf(s, "userspace(pid=%lli,flow_sample(probability=%lli,"
538 "collector_set_id=%lli,obs_domain_id=%lli,"
539 "obs_point_id=%lli))%n",
540 &pid, &probability, &collector_set_id,
541 &obs_domain_id, &obs_point_id, &n) > 0 && n > 0) {
542 union user_action_cookie cookie;
544 cookie.type = USER_ACTION_COOKIE_FLOW_SAMPLE;
545 cookie.flow_sample.probability = probability;
546 cookie.flow_sample.collector_set_id = collector_set_id;
547 cookie.flow_sample.obs_domain_id = obs_domain_id;
548 cookie.flow_sample.obs_point_id = obs_point_id;
549 odp_put_userspace_action(pid, &cookie, sizeof cookie.flow_sample,
552 } else if (sscanf(s, "userspace(pid=%lli,ipfix)%n", &pid, &n) > 0
554 union user_action_cookie cookie;
556 cookie.type = USER_ACTION_COOKIE_IPFIX;
557 odp_put_userspace_action(pid, &cookie, sizeof cookie.ipfix,
560 } else if (sscanf(s, "userspace(pid=%lli,userdata(%n", &pid, &n) > 0
565 ofpbuf_init(&buf, 16);
566 end = ofpbuf_put_hex(&buf, &s[n], NULL);
567 if (end[0] == ')' && end[1] == ')') {
568 odp_put_userspace_action(pid, buf.data, buf.size, actions);
570 return (end + 2) - s;
575 if (!strncmp(s, "set(", 4)) {
579 start_ofs = nl_msg_start_nested(actions, OVS_ACTION_ATTR_SET);
580 retval = parse_odp_key_attr(s + 4, port_names, actions);
584 if (s[retval + 4] != ')') {
587 nl_msg_end_nested(actions, start_ofs);
592 struct ovs_action_push_vlan push;
593 int tpid = ETH_TYPE_VLAN;
598 if ((sscanf(s, "push_vlan(vid=%i,pcp=%i)%n", &vid, &pcp, &n) > 0
600 || (sscanf(s, "push_vlan(vid=%i,pcp=%i,cfi=%i)%n",
601 &vid, &pcp, &cfi, &n) > 0 && n > 0)
602 || (sscanf(s, "push_vlan(tpid=%i,vid=%i,pcp=%i)%n",
603 &tpid, &vid, &pcp, &n) > 0 && n > 0)
604 || (sscanf(s, "push_vlan(tpid=%i,vid=%i,pcp=%i,cfi=%i)%n",
605 &tpid, &vid, &pcp, &cfi, &n) > 0 && n > 0)) {
606 push.vlan_tpid = htons(tpid);
607 push.vlan_tci = htons((vid << VLAN_VID_SHIFT)
608 | (pcp << VLAN_PCP_SHIFT)
609 | (cfi ? VLAN_CFI : 0));
610 nl_msg_put_unspec(actions, OVS_ACTION_ATTR_PUSH_VLAN,
617 if (!strncmp(s, "pop_vlan", 8)) {
618 nl_msg_put_flag(actions, OVS_ACTION_ATTR_POP_VLAN);
626 if (sscanf(s, "sample(sample=%lf%%,actions(%n", &percentage, &n) > 0
627 && percentage >= 0. && percentage <= 100.0
629 size_t sample_ofs, actions_ofs;
632 probability = floor(UINT32_MAX * (percentage / 100.0) + .5);
633 sample_ofs = nl_msg_start_nested(actions, OVS_ACTION_ATTR_SAMPLE);
634 nl_msg_put_u32(actions, OVS_SAMPLE_ATTR_PROBABILITY,
635 (probability <= 0 ? 0
636 : probability >= UINT32_MAX ? UINT32_MAX
639 actions_ofs = nl_msg_start_nested(actions,
640 OVS_SAMPLE_ATTR_ACTIONS);
644 n += strspn(s + n, delimiters);
649 retval = parse_odp_action(s + n, port_names, actions);
655 nl_msg_end_nested(actions, actions_ofs);
656 nl_msg_end_nested(actions, sample_ofs);
658 return s[n + 1] == ')' ? n + 2 : -EINVAL;
665 /* Parses the string representation of datapath actions, in the format output
666 * by format_odp_action(). Returns 0 if successful, otherwise a positive errno
667 * value. On success, the ODP actions are appended to 'actions' as a series of
668 * Netlink attributes. On failure, no data is appended to 'actions'. Either
669 * way, 'actions''s data might be reallocated. */
671 odp_actions_from_string(const char *s, const struct simap *port_names,
672 struct ofpbuf *actions)
676 if (!strcasecmp(s, "drop")) {
680 old_size = actions->size;
684 s += strspn(s, delimiters);
689 retval = parse_odp_action(s, port_names, actions);
690 if (retval < 0 || !strchr(delimiters, s[retval])) {
691 actions->size = old_size;
700 /* Returns the correct length of the payload for a flow key attribute of the
701 * specified 'type', -1 if 'type' is unknown, or -2 if the attribute's payload
702 * is variable length. */
704 odp_flow_key_attr_len(uint16_t type)
706 if (type > OVS_KEY_ATTR_MAX) {
710 switch ((enum ovs_key_attr) type) {
711 case OVS_KEY_ATTR_ENCAP: return -2;
712 case OVS_KEY_ATTR_PRIORITY: return 4;
713 case OVS_KEY_ATTR_SKB_MARK: return 4;
714 case OVS_KEY_ATTR_TUNNEL: return -2;
715 case OVS_KEY_ATTR_IN_PORT: return 4;
716 case OVS_KEY_ATTR_ETHERNET: return sizeof(struct ovs_key_ethernet);
717 case OVS_KEY_ATTR_VLAN: return sizeof(ovs_be16);
718 case OVS_KEY_ATTR_ETHERTYPE: return 2;
719 case OVS_KEY_ATTR_MPLS: return sizeof(struct ovs_key_mpls);
720 case OVS_KEY_ATTR_IPV4: return sizeof(struct ovs_key_ipv4);
721 case OVS_KEY_ATTR_IPV6: return sizeof(struct ovs_key_ipv6);
722 case OVS_KEY_ATTR_TCP: return sizeof(struct ovs_key_tcp);
723 case OVS_KEY_ATTR_UDP: return sizeof(struct ovs_key_udp);
724 case OVS_KEY_ATTR_ICMP: return sizeof(struct ovs_key_icmp);
725 case OVS_KEY_ATTR_ICMPV6: return sizeof(struct ovs_key_icmpv6);
726 case OVS_KEY_ATTR_ARP: return sizeof(struct ovs_key_arp);
727 case OVS_KEY_ATTR_ND: return sizeof(struct ovs_key_nd);
729 case OVS_KEY_ATTR_UNSPEC:
730 case __OVS_KEY_ATTR_MAX:
738 format_generic_odp_key(const struct nlattr *a, struct ds *ds)
740 size_t len = nl_attr_get_size(a);
742 const uint8_t *unspec;
745 unspec = nl_attr_get(a);
746 for (i = 0; i < len; i++) {
747 ds_put_char(ds, i ? ' ': '(');
748 ds_put_format(ds, "%02x", unspec[i]);
750 ds_put_char(ds, ')');
755 ovs_frag_type_to_string(enum ovs_frag_type type)
758 case OVS_FRAG_TYPE_NONE:
760 case OVS_FRAG_TYPE_FIRST:
762 case OVS_FRAG_TYPE_LATER:
764 case __OVS_FRAG_TYPE_MAX:
771 tunnel_key_attr_len(int type)
774 case OVS_TUNNEL_KEY_ATTR_ID: return 8;
775 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC: return 4;
776 case OVS_TUNNEL_KEY_ATTR_IPV4_DST: return 4;
777 case OVS_TUNNEL_KEY_ATTR_TOS: return 1;
778 case OVS_TUNNEL_KEY_ATTR_TTL: return 1;
779 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT: return 0;
780 case OVS_TUNNEL_KEY_ATTR_CSUM: return 0;
781 case __OVS_TUNNEL_KEY_ATTR_MAX:
787 static enum odp_key_fitness
788 tun_key_from_attr(const struct nlattr *attr, struct flow_tnl *tun)
791 const struct nlattr *a;
793 bool unknown = false;
795 NL_NESTED_FOR_EACH(a, left, attr) {
796 uint16_t type = nl_attr_type(a);
797 size_t len = nl_attr_get_size(a);
798 int expected_len = tunnel_key_attr_len(type);
800 if (len != expected_len && expected_len >= 0) {
801 return ODP_FIT_ERROR;
805 case OVS_TUNNEL_KEY_ATTR_ID:
806 tun->tun_id = nl_attr_get_be64(a);
807 tun->flags |= FLOW_TNL_F_KEY;
809 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
810 tun->ip_src = nl_attr_get_be32(a);
812 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
813 tun->ip_dst = nl_attr_get_be32(a);
815 case OVS_TUNNEL_KEY_ATTR_TOS:
816 tun->ip_tos = nl_attr_get_u8(a);
818 case OVS_TUNNEL_KEY_ATTR_TTL:
819 tun->ip_ttl = nl_attr_get_u8(a);
822 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
823 tun->flags |= FLOW_TNL_F_DONT_FRAGMENT;
825 case OVS_TUNNEL_KEY_ATTR_CSUM:
826 tun->flags |= FLOW_TNL_F_CSUM;
829 /* Allow this to show up as unexpected, if there are unknown
830 * tunnel attribute, eventually resulting in ODP_FIT_TOO_MUCH. */
837 return ODP_FIT_ERROR;
840 return ODP_FIT_TOO_MUCH;
842 return ODP_FIT_PERFECT;
846 tun_key_to_attr(struct ofpbuf *a, const struct flow_tnl *tun_key)
850 tun_key_ofs = nl_msg_start_nested(a, OVS_KEY_ATTR_TUNNEL);
852 if (tun_key->flags & FLOW_TNL_F_KEY) {
853 nl_msg_put_be64(a, OVS_TUNNEL_KEY_ATTR_ID, tun_key->tun_id);
855 if (tun_key->ip_src) {
856 nl_msg_put_be32(a, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, tun_key->ip_src);
858 if (tun_key->ip_dst) {
859 nl_msg_put_be32(a, OVS_TUNNEL_KEY_ATTR_IPV4_DST, tun_key->ip_dst);
861 if (tun_key->ip_tos) {
862 nl_msg_put_u8(a, OVS_TUNNEL_KEY_ATTR_TOS, tun_key->ip_tos);
864 nl_msg_put_u8(a, OVS_TUNNEL_KEY_ATTR_TTL, tun_key->ip_ttl);
865 if (tun_key->flags & FLOW_TNL_F_DONT_FRAGMENT) {
866 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT);
868 if (tun_key->flags & FLOW_TNL_F_CSUM) {
869 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_CSUM);
872 nl_msg_end_nested(a, tun_key_ofs);
876 format_odp_key_attr(const struct nlattr *a, struct ds *ds)
878 const struct ovs_key_ethernet *eth_key;
879 const struct ovs_key_ipv4 *ipv4_key;
880 const struct ovs_key_ipv6 *ipv6_key;
881 const struct ovs_key_tcp *tcp_key;
882 const struct ovs_key_udp *udp_key;
883 const struct ovs_key_icmp *icmp_key;
884 const struct ovs_key_icmpv6 *icmpv6_key;
885 const struct ovs_key_arp *arp_key;
886 const struct ovs_key_nd *nd_key;
887 struct flow_tnl tun_key;
888 enum ovs_key_attr attr = nl_attr_type(a);
891 ds_put_cstr(ds, ovs_key_attr_to_string(attr));
892 expected_len = odp_flow_key_attr_len(nl_attr_type(a));
893 if (expected_len != -2 && nl_attr_get_size(a) != expected_len) {
894 ds_put_format(ds, "(bad length %zu, expected %d)",
896 odp_flow_key_attr_len(nl_attr_type(a)));
897 format_generic_odp_key(a, ds);
902 case OVS_KEY_ATTR_ENCAP:
903 ds_put_cstr(ds, "(");
904 if (nl_attr_get_size(a)) {
905 odp_flow_key_format(nl_attr_get(a), nl_attr_get_size(a), ds);
907 ds_put_char(ds, ')');
910 case OVS_KEY_ATTR_PRIORITY:
911 ds_put_format(ds, "(%#"PRIx32")", nl_attr_get_u32(a));
914 case OVS_KEY_ATTR_SKB_MARK:
915 ds_put_format(ds, "(%#"PRIx32")", nl_attr_get_u32(a));
918 case OVS_KEY_ATTR_TUNNEL:
919 memset(&tun_key, 0, sizeof tun_key);
920 if (tun_key_from_attr(a, &tun_key) == ODP_FIT_ERROR) {
921 ds_put_format(ds, "(error)");
923 ds_put_format(ds, "(tun_id=0x%"PRIx64",src="IP_FMT",dst="IP_FMT","
924 "tos=0x%"PRIx8",ttl=%"PRIu8",flags(",
925 ntohll(tun_key.tun_id),
926 IP_ARGS(tun_key.ip_src),
927 IP_ARGS(tun_key.ip_dst),
928 tun_key.ip_tos, tun_key.ip_ttl);
930 format_flags(ds, flow_tun_flag_to_string,
931 (uint32_t) tun_key.flags, ',');
932 ds_put_format(ds, "))");
936 case OVS_KEY_ATTR_IN_PORT:
937 ds_put_format(ds, "(%"PRIu32")", nl_attr_get_u32(a));
940 case OVS_KEY_ATTR_ETHERNET:
941 eth_key = nl_attr_get(a);
942 ds_put_format(ds, "(src="ETH_ADDR_FMT",dst="ETH_ADDR_FMT")",
943 ETH_ADDR_ARGS(eth_key->eth_src),
944 ETH_ADDR_ARGS(eth_key->eth_dst));
947 case OVS_KEY_ATTR_VLAN:
948 ds_put_char(ds, '(');
949 format_vlan_tci(ds, nl_attr_get_be16(a));
950 ds_put_char(ds, ')');
953 case OVS_KEY_ATTR_MPLS: {
954 const struct ovs_key_mpls *mpls_key = nl_attr_get(a);
955 ds_put_char(ds, '(');
956 format_mpls_lse(ds, mpls_key->mpls_top_lse);
957 ds_put_char(ds, ')');
961 case OVS_KEY_ATTR_ETHERTYPE:
962 ds_put_format(ds, "(0x%04"PRIx16")",
963 ntohs(nl_attr_get_be16(a)));
966 case OVS_KEY_ATTR_IPV4:
967 ipv4_key = nl_attr_get(a);
968 ds_put_format(ds, "(src="IP_FMT",dst="IP_FMT",proto=%"PRIu8
969 ",tos=%#"PRIx8",ttl=%"PRIu8",frag=%s)",
970 IP_ARGS(ipv4_key->ipv4_src),
971 IP_ARGS(ipv4_key->ipv4_dst),
972 ipv4_key->ipv4_proto, ipv4_key->ipv4_tos,
974 ovs_frag_type_to_string(ipv4_key->ipv4_frag));
977 case OVS_KEY_ATTR_IPV6: {
978 char src_str[INET6_ADDRSTRLEN];
979 char dst_str[INET6_ADDRSTRLEN];
981 ipv6_key = nl_attr_get(a);
982 inet_ntop(AF_INET6, ipv6_key->ipv6_src, src_str, sizeof src_str);
983 inet_ntop(AF_INET6, ipv6_key->ipv6_dst, dst_str, sizeof dst_str);
985 ds_put_format(ds, "(src=%s,dst=%s,label=%#"PRIx32",proto=%"PRIu8
986 ",tclass=%#"PRIx8",hlimit=%"PRIu8",frag=%s)",
987 src_str, dst_str, ntohl(ipv6_key->ipv6_label),
988 ipv6_key->ipv6_proto, ipv6_key->ipv6_tclass,
989 ipv6_key->ipv6_hlimit,
990 ovs_frag_type_to_string(ipv6_key->ipv6_frag));
994 case OVS_KEY_ATTR_TCP:
995 tcp_key = nl_attr_get(a);
996 ds_put_format(ds, "(src=%"PRIu16",dst=%"PRIu16")",
997 ntohs(tcp_key->tcp_src), ntohs(tcp_key->tcp_dst));
1000 case OVS_KEY_ATTR_UDP:
1001 udp_key = nl_attr_get(a);
1002 ds_put_format(ds, "(src=%"PRIu16",dst=%"PRIu16")",
1003 ntohs(udp_key->udp_src), ntohs(udp_key->udp_dst));
1006 case OVS_KEY_ATTR_ICMP:
1007 icmp_key = nl_attr_get(a);
1008 ds_put_format(ds, "(type=%"PRIu8",code=%"PRIu8")",
1009 icmp_key->icmp_type, icmp_key->icmp_code);
1012 case OVS_KEY_ATTR_ICMPV6:
1013 icmpv6_key = nl_attr_get(a);
1014 ds_put_format(ds, "(type=%"PRIu8",code=%"PRIu8")",
1015 icmpv6_key->icmpv6_type, icmpv6_key->icmpv6_code);
1018 case OVS_KEY_ATTR_ARP:
1019 arp_key = nl_attr_get(a);
1020 ds_put_format(ds, "(sip="IP_FMT",tip="IP_FMT",op=%"PRIu16","
1021 "sha="ETH_ADDR_FMT",tha="ETH_ADDR_FMT")",
1022 IP_ARGS(arp_key->arp_sip), IP_ARGS(arp_key->arp_tip),
1023 ntohs(arp_key->arp_op), ETH_ADDR_ARGS(arp_key->arp_sha),
1024 ETH_ADDR_ARGS(arp_key->arp_tha));
1027 case OVS_KEY_ATTR_ND: {
1028 char target[INET6_ADDRSTRLEN];
1030 nd_key = nl_attr_get(a);
1031 inet_ntop(AF_INET6, nd_key->nd_target, target, sizeof target);
1033 ds_put_format(ds, "(target=%s", target);
1034 if (!eth_addr_is_zero(nd_key->nd_sll)) {
1035 ds_put_format(ds, ",sll="ETH_ADDR_FMT,
1036 ETH_ADDR_ARGS(nd_key->nd_sll));
1038 if (!eth_addr_is_zero(nd_key->nd_tll)) {
1039 ds_put_format(ds, ",tll="ETH_ADDR_FMT,
1040 ETH_ADDR_ARGS(nd_key->nd_tll));
1042 ds_put_char(ds, ')');
1046 case OVS_KEY_ATTR_UNSPEC:
1047 case __OVS_KEY_ATTR_MAX:
1049 format_generic_odp_key(a, ds);
1054 /* Appends to 'ds' a string representation of the 'key_len' bytes of
1055 * OVS_KEY_ATTR_* attributes in 'key'. */
1057 odp_flow_key_format(const struct nlattr *key, size_t key_len, struct ds *ds)
1060 const struct nlattr *a;
1063 NL_ATTR_FOR_EACH (a, left, key, key_len) {
1065 ds_put_char(ds, ',');
1067 format_odp_key_attr(a, ds);
1072 if (left == key_len) {
1073 ds_put_cstr(ds, "<empty>");
1075 ds_put_format(ds, ",***%u leftover bytes*** (", left);
1076 for (i = 0; i < left; i++) {
1077 ds_put_format(ds, "%02x", ((const uint8_t *) a)[i]);
1079 ds_put_char(ds, ')');
1082 ds_put_cstr(ds, "<empty>");
1087 put_nd_key(int n, const char *nd_target_s,
1088 const uint8_t *nd_sll, const uint8_t *nd_tll, struct ofpbuf *key)
1090 struct ovs_key_nd nd_key;
1092 memset(&nd_key, 0, sizeof nd_key);
1093 if (inet_pton(AF_INET6, nd_target_s, nd_key.nd_target) != 1) {
1097 memcpy(nd_key.nd_sll, nd_sll, ETH_ADDR_LEN);
1100 memcpy(nd_key.nd_tll, nd_tll, ETH_ADDR_LEN);
1102 nl_msg_put_unspec(key, OVS_KEY_ATTR_ND, &nd_key, sizeof nd_key);
1107 ovs_frag_type_from_string(const char *s, enum ovs_frag_type *type)
1109 if (!strcasecmp(s, "no")) {
1110 *type = OVS_FRAG_TYPE_NONE;
1111 } else if (!strcasecmp(s, "first")) {
1112 *type = OVS_FRAG_TYPE_FIRST;
1113 } else if (!strcasecmp(s, "later")) {
1114 *type = OVS_FRAG_TYPE_LATER;
1122 mpls_lse_from_components(int mpls_label, int mpls_tc, int mpls_ttl, int mpls_bos)
1124 return (htonl((mpls_label << MPLS_LABEL_SHIFT) |
1125 (mpls_tc << MPLS_TC_SHIFT) |
1126 (mpls_ttl << MPLS_TTL_SHIFT) |
1127 (mpls_bos << MPLS_BOS_SHIFT)));
1131 parse_odp_key_attr(const char *s, const struct simap *port_names,
1134 /* Many of the sscanf calls in this function use oversized destination
1135 * fields because some sscanf() implementations truncate the range of %i
1136 * directives, so that e.g. "%"SCNi16 interprets input of "0xfedc" as a
1137 * value of 0x7fff. The other alternatives are to allow only a single
1138 * radix (e.g. decimal or hexadecimal) or to write more sophisticated
1141 * The tun_id parser has to use an alternative approach because there is no
1142 * type larger than 64 bits. */
1145 unsigned long long int priority;
1148 if (sscanf(s, "skb_priority(%llx)%n", &priority, &n) > 0 && n > 0) {
1149 nl_msg_put_u32(key, OVS_KEY_ATTR_PRIORITY, priority);
1155 unsigned long long int mark;
1158 if (sscanf(s, "skb_mark(%llx)%n", &mark, &n) > 0 && n > 0) {
1159 nl_msg_put_u32(key, OVS_KEY_ATTR_SKB_MARK, mark);
1167 struct flow_tnl tun_key;
1170 if (sscanf(s, "tunnel(tun_id=%31[x0123456789abcdefABCDEF],"
1171 "src="IP_SCAN_FMT",dst="IP_SCAN_FMT
1172 ",tos=%i,ttl=%i,flags%n", tun_id_s,
1173 IP_SCAN_ARGS(&tun_key.ip_src),
1174 IP_SCAN_ARGS(&tun_key.ip_dst), &tos, &ttl,
1179 tun_key.tun_id = htonll(strtoull(tun_id_s, NULL, 0));
1180 tun_key.ip_tos = tos;
1181 tun_key.ip_ttl = ttl;
1182 res = parse_flags(&s[n], flow_tun_flag_to_string, &flags);
1183 tun_key.flags = (uint16_t) flags;
1193 tun_key_to_attr(key, &tun_key);
1199 unsigned long long int in_port;
1202 if (sscanf(s, "in_port(%lli)%n", &in_port, &n) > 0 && n > 0) {
1203 nl_msg_put_u32(key, OVS_KEY_ATTR_IN_PORT, in_port);
1208 if (port_names && !strncmp(s, "in_port(", 8)) {
1210 const struct simap_node *node;
1214 name_len = strcspn(s, ")");
1215 node = simap_find_len(port_names, name, name_len);
1217 nl_msg_put_u32(key, OVS_KEY_ATTR_IN_PORT, node->data);
1218 return 8 + name_len + 1;
1223 struct ovs_key_ethernet eth_key;
1227 "eth(src="ETH_ADDR_SCAN_FMT",dst="ETH_ADDR_SCAN_FMT")%n",
1228 ETH_ADDR_SCAN_ARGS(eth_key.eth_src),
1229 ETH_ADDR_SCAN_ARGS(eth_key.eth_dst), &n) > 0 && n > 0) {
1230 nl_msg_put_unspec(key, OVS_KEY_ATTR_ETHERNET,
1231 ð_key, sizeof eth_key);
1242 if ((sscanf(s, "vlan(vid=%"SCNi16",pcp=%i)%n", &vid, &pcp, &n) > 0
1244 nl_msg_put_be16(key, OVS_KEY_ATTR_VLAN,
1245 htons((vid << VLAN_VID_SHIFT) |
1246 (pcp << VLAN_PCP_SHIFT) |
1249 } else if ((sscanf(s, "vlan(vid=%"SCNi16",pcp=%i,cfi=%i)%n",
1250 &vid, &pcp, &cfi, &n) > 0
1252 nl_msg_put_be16(key, OVS_KEY_ATTR_VLAN,
1253 htons((vid << VLAN_VID_SHIFT) |
1254 (pcp << VLAN_PCP_SHIFT) |
1255 (cfi ? VLAN_CFI : 0)));
1264 if (sscanf(s, "eth_type(%i)%n", ð_type, &n) > 0 && n > 0) {
1265 nl_msg_put_be16(key, OVS_KEY_ATTR_ETHERTYPE, htons(eth_type));
1271 int label, tc, ttl, bos;
1274 if (sscanf(s, "mpls(label=%"SCNi32",tc=%i,ttl=%i,bos=%i)%n",
1275 &label, &tc, &ttl, &bos, &n) > 0 &&
1277 struct ovs_key_mpls *mpls;
1279 mpls = nl_msg_put_unspec_uninit(key, OVS_KEY_ATTR_MPLS,
1281 mpls->mpls_top_lse = mpls_lse_from_components(label, tc, ttl, bos);
1293 enum ovs_frag_type ipv4_frag;
1296 if (sscanf(s, "ipv4(src="IP_SCAN_FMT",dst="IP_SCAN_FMT","
1297 "proto=%i,tos=%i,ttl=%i,frag=%7[a-z])%n",
1298 IP_SCAN_ARGS(&ipv4_src), IP_SCAN_ARGS(&ipv4_dst),
1299 &ipv4_proto, &ipv4_tos, &ipv4_ttl, frag, &n) > 0
1301 && ovs_frag_type_from_string(frag, &ipv4_frag)) {
1302 struct ovs_key_ipv4 ipv4_key;
1304 ipv4_key.ipv4_src = ipv4_src;
1305 ipv4_key.ipv4_dst = ipv4_dst;
1306 ipv4_key.ipv4_proto = ipv4_proto;
1307 ipv4_key.ipv4_tos = ipv4_tos;
1308 ipv4_key.ipv4_ttl = ipv4_ttl;
1309 ipv4_key.ipv4_frag = ipv4_frag;
1310 nl_msg_put_unspec(key, OVS_KEY_ATTR_IPV4,
1311 &ipv4_key, sizeof ipv4_key);
1317 char ipv6_src_s[IPV6_SCAN_LEN + 1];
1318 char ipv6_dst_s[IPV6_SCAN_LEN + 1];
1324 enum ovs_frag_type ipv6_frag;
1327 if (sscanf(s, "ipv6(src="IPV6_SCAN_FMT",dst="IPV6_SCAN_FMT","
1328 "label=%i,proto=%i,tclass=%i,hlimit=%i,frag=%7[a-z])%n",
1329 ipv6_src_s, ipv6_dst_s, &ipv6_label,
1330 &ipv6_proto, &ipv6_tclass, &ipv6_hlimit, frag, &n) > 0
1332 && ovs_frag_type_from_string(frag, &ipv6_frag)) {
1333 struct ovs_key_ipv6 ipv6_key;
1335 if (inet_pton(AF_INET6, ipv6_src_s, &ipv6_key.ipv6_src) != 1 ||
1336 inet_pton(AF_INET6, ipv6_dst_s, &ipv6_key.ipv6_dst) != 1) {
1339 ipv6_key.ipv6_label = htonl(ipv6_label);
1340 ipv6_key.ipv6_proto = ipv6_proto;
1341 ipv6_key.ipv6_tclass = ipv6_tclass;
1342 ipv6_key.ipv6_hlimit = ipv6_hlimit;
1343 ipv6_key.ipv6_frag = ipv6_frag;
1344 nl_msg_put_unspec(key, OVS_KEY_ATTR_IPV6,
1345 &ipv6_key, sizeof ipv6_key);
1355 if (sscanf(s, "tcp(src=%i,dst=%i)%n",&tcp_src, &tcp_dst, &n) > 0
1357 struct ovs_key_tcp tcp_key;
1359 tcp_key.tcp_src = htons(tcp_src);
1360 tcp_key.tcp_dst = htons(tcp_dst);
1361 nl_msg_put_unspec(key, OVS_KEY_ATTR_TCP, &tcp_key, sizeof tcp_key);
1371 if (sscanf(s, "udp(src=%i,dst=%i)%n", &udp_src, &udp_dst, &n) > 0
1373 struct ovs_key_udp udp_key;
1375 udp_key.udp_src = htons(udp_src);
1376 udp_key.udp_dst = htons(udp_dst);
1377 nl_msg_put_unspec(key, OVS_KEY_ATTR_UDP, &udp_key, sizeof udp_key);
1387 if (sscanf(s, "icmp(type=%i,code=%i)%n",
1388 &icmp_type, &icmp_code, &n) > 0
1390 struct ovs_key_icmp icmp_key;
1392 icmp_key.icmp_type = icmp_type;
1393 icmp_key.icmp_code = icmp_code;
1394 nl_msg_put_unspec(key, OVS_KEY_ATTR_ICMP,
1395 &icmp_key, sizeof icmp_key);
1401 struct ovs_key_icmpv6 icmpv6_key;
1404 if (sscanf(s, "icmpv6(type=%"SCNi8",code=%"SCNi8")%n",
1405 &icmpv6_key.icmpv6_type, &icmpv6_key.icmpv6_code,&n) > 0
1407 nl_msg_put_unspec(key, OVS_KEY_ATTR_ICMPV6,
1408 &icmpv6_key, sizeof icmpv6_key);
1417 uint8_t arp_sha[ETH_ADDR_LEN];
1418 uint8_t arp_tha[ETH_ADDR_LEN];
1421 if (sscanf(s, "arp(sip="IP_SCAN_FMT",tip="IP_SCAN_FMT","
1422 "op=%i,sha="ETH_ADDR_SCAN_FMT",tha="ETH_ADDR_SCAN_FMT")%n",
1423 IP_SCAN_ARGS(&arp_sip),
1424 IP_SCAN_ARGS(&arp_tip),
1426 ETH_ADDR_SCAN_ARGS(arp_sha),
1427 ETH_ADDR_SCAN_ARGS(arp_tha), &n) > 0 && n > 0) {
1428 struct ovs_key_arp arp_key;
1430 memset(&arp_key, 0, sizeof arp_key);
1431 arp_key.arp_sip = arp_sip;
1432 arp_key.arp_tip = arp_tip;
1433 arp_key.arp_op = htons(arp_op);
1434 memcpy(arp_key.arp_sha, arp_sha, ETH_ADDR_LEN);
1435 memcpy(arp_key.arp_tha, arp_tha, ETH_ADDR_LEN);
1436 nl_msg_put_unspec(key, OVS_KEY_ATTR_ARP, &arp_key, sizeof arp_key);
1442 char nd_target_s[IPV6_SCAN_LEN + 1];
1443 uint8_t nd_sll[ETH_ADDR_LEN];
1444 uint8_t nd_tll[ETH_ADDR_LEN];
1447 if (sscanf(s, "nd(target="IPV6_SCAN_FMT")%n",
1448 nd_target_s, &n) > 0 && n > 0) {
1449 return put_nd_key(n, nd_target_s, NULL, NULL, key);
1451 if (sscanf(s, "nd(target="IPV6_SCAN_FMT",sll="ETH_ADDR_SCAN_FMT")%n",
1452 nd_target_s, ETH_ADDR_SCAN_ARGS(nd_sll), &n) > 0
1454 return put_nd_key(n, nd_target_s, nd_sll, NULL, key);
1456 if (sscanf(s, "nd(target="IPV6_SCAN_FMT",tll="ETH_ADDR_SCAN_FMT")%n",
1457 nd_target_s, ETH_ADDR_SCAN_ARGS(nd_tll), &n) > 0
1459 return put_nd_key(n, nd_target_s, NULL, nd_tll, key);
1461 if (sscanf(s, "nd(target="IPV6_SCAN_FMT",sll="ETH_ADDR_SCAN_FMT","
1462 "tll="ETH_ADDR_SCAN_FMT")%n",
1463 nd_target_s, ETH_ADDR_SCAN_ARGS(nd_sll),
1464 ETH_ADDR_SCAN_ARGS(nd_tll), &n) > 0
1466 return put_nd_key(n, nd_target_s, nd_sll, nd_tll, key);
1470 if (!strncmp(s, "encap(", 6)) {
1471 const char *start = s;
1474 encap = nl_msg_start_nested(key, OVS_KEY_ATTR_ENCAP);
1480 s += strspn(s, ", \t\r\n");
1483 } else if (*s == ')') {
1487 retval = parse_odp_key_attr(s, port_names, key);
1495 nl_msg_end_nested(key, encap);
1503 /* Parses the string representation of a datapath flow key, in the
1504 * format output by odp_flow_key_format(). Returns 0 if successful,
1505 * otherwise a positive errno value. On success, the flow key is
1506 * appended to 'key' as a series of Netlink attributes. On failure, no
1507 * data is appended to 'key'. Either way, 'key''s data might be
1510 * If 'port_names' is nonnull, it points to an simap that maps from a port name
1511 * to a port number. (Port names may be used instead of port numbers in
1514 * On success, the attributes appended to 'key' are individually syntactically
1515 * valid, but they may not be valid as a sequence. 'key' might, for example,
1516 * have duplicated keys. odp_flow_key_to_flow() will detect those errors. */
1518 odp_flow_key_from_string(const char *s, const struct simap *port_names,
1521 const size_t old_size = key->size;
1525 s += strspn(s, delimiters);
1530 retval = parse_odp_key_attr(s, port_names, key);
1532 key->size = old_size;
1542 ovs_to_odp_frag(uint8_t nw_frag)
1544 return (nw_frag == 0 ? OVS_FRAG_TYPE_NONE
1545 : nw_frag == FLOW_NW_FRAG_ANY ? OVS_FRAG_TYPE_FIRST
1546 : OVS_FRAG_TYPE_LATER);
1549 /* Appends a representation of 'flow' as OVS_KEY_ATTR_* attributes to 'buf'.
1550 * 'flow->in_port' is ignored (since it is likely to be an OpenFlow port
1551 * number rather than a datapath port number). Instead, if 'odp_in_port'
1552 * is anything other than OVSP_NONE, it is included in 'buf' as the input
1555 * 'buf' must have at least ODPUTIL_FLOW_KEY_BYTES bytes of space, or be
1556 * capable of being expanded to allow for that much space. */
1558 odp_flow_key_from_flow(struct ofpbuf *buf, const struct flow *flow,
1559 uint32_t odp_in_port)
1561 struct ovs_key_ethernet *eth_key;
1564 if (flow->skb_priority) {
1565 nl_msg_put_u32(buf, OVS_KEY_ATTR_PRIORITY, flow->skb_priority);
1568 if (flow->tunnel.ip_dst) {
1569 tun_key_to_attr(buf, &flow->tunnel);
1572 if (flow->skb_mark) {
1573 nl_msg_put_u32(buf, OVS_KEY_ATTR_SKB_MARK, flow->skb_mark);
1576 if (odp_in_port != OVSP_NONE) {
1577 nl_msg_put_u32(buf, OVS_KEY_ATTR_IN_PORT, odp_in_port);
1580 eth_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ETHERNET,
1582 memcpy(eth_key->eth_src, flow->dl_src, ETH_ADDR_LEN);
1583 memcpy(eth_key->eth_dst, flow->dl_dst, ETH_ADDR_LEN);
1585 if (flow->vlan_tci != htons(0) || flow->dl_type == htons(ETH_TYPE_VLAN)) {
1586 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_TYPE_VLAN));
1587 nl_msg_put_be16(buf, OVS_KEY_ATTR_VLAN, flow->vlan_tci);
1588 encap = nl_msg_start_nested(buf, OVS_KEY_ATTR_ENCAP);
1589 if (flow->vlan_tci == htons(0)) {
1596 if (ntohs(flow->dl_type) < ETH_TYPE_MIN) {
1600 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, flow->dl_type);
1602 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1603 struct ovs_key_ipv4 *ipv4_key;
1605 ipv4_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_IPV4,
1607 ipv4_key->ipv4_src = flow->nw_src;
1608 ipv4_key->ipv4_dst = flow->nw_dst;
1609 ipv4_key->ipv4_proto = flow->nw_proto;
1610 ipv4_key->ipv4_tos = flow->nw_tos;
1611 ipv4_key->ipv4_ttl = flow->nw_ttl;
1612 ipv4_key->ipv4_frag = ovs_to_odp_frag(flow->nw_frag);
1613 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1614 struct ovs_key_ipv6 *ipv6_key;
1616 ipv6_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_IPV6,
1618 memcpy(ipv6_key->ipv6_src, &flow->ipv6_src, sizeof ipv6_key->ipv6_src);
1619 memcpy(ipv6_key->ipv6_dst, &flow->ipv6_dst, sizeof ipv6_key->ipv6_dst);
1620 ipv6_key->ipv6_label = flow->ipv6_label;
1621 ipv6_key->ipv6_proto = flow->nw_proto;
1622 ipv6_key->ipv6_tclass = flow->nw_tos;
1623 ipv6_key->ipv6_hlimit = flow->nw_ttl;
1624 ipv6_key->ipv6_frag = ovs_to_odp_frag(flow->nw_frag);
1625 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1626 flow->dl_type == htons(ETH_TYPE_RARP)) {
1627 struct ovs_key_arp *arp_key;
1629 arp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ARP,
1631 memset(arp_key, 0, sizeof *arp_key);
1632 arp_key->arp_sip = flow->nw_src;
1633 arp_key->arp_tip = flow->nw_dst;
1634 arp_key->arp_op = htons(flow->nw_proto);
1635 memcpy(arp_key->arp_sha, flow->arp_sha, ETH_ADDR_LEN);
1636 memcpy(arp_key->arp_tha, flow->arp_tha, ETH_ADDR_LEN);
1639 if (flow->mpls_depth) {
1640 struct ovs_key_mpls *mpls_key;
1642 mpls_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_MPLS,
1644 mpls_key->mpls_top_lse = flow->mpls_lse;
1647 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1648 if (flow->nw_proto == IPPROTO_TCP) {
1649 struct ovs_key_tcp *tcp_key;
1651 tcp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_TCP,
1653 tcp_key->tcp_src = flow->tp_src;
1654 tcp_key->tcp_dst = flow->tp_dst;
1655 } else if (flow->nw_proto == IPPROTO_UDP) {
1656 struct ovs_key_udp *udp_key;
1658 udp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_UDP,
1660 udp_key->udp_src = flow->tp_src;
1661 udp_key->udp_dst = flow->tp_dst;
1662 } else if (flow->dl_type == htons(ETH_TYPE_IP)
1663 && flow->nw_proto == IPPROTO_ICMP) {
1664 struct ovs_key_icmp *icmp_key;
1666 icmp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ICMP,
1668 icmp_key->icmp_type = ntohs(flow->tp_src);
1669 icmp_key->icmp_code = ntohs(flow->tp_dst);
1670 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)
1671 && flow->nw_proto == IPPROTO_ICMPV6) {
1672 struct ovs_key_icmpv6 *icmpv6_key;
1674 icmpv6_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ICMPV6,
1675 sizeof *icmpv6_key);
1676 icmpv6_key->icmpv6_type = ntohs(flow->tp_src);
1677 icmpv6_key->icmpv6_code = ntohs(flow->tp_dst);
1679 if (icmpv6_key->icmpv6_type == ND_NEIGHBOR_SOLICIT
1680 || icmpv6_key->icmpv6_type == ND_NEIGHBOR_ADVERT) {
1681 struct ovs_key_nd *nd_key;
1683 nd_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ND,
1685 memcpy(nd_key->nd_target, &flow->nd_target,
1686 sizeof nd_key->nd_target);
1687 memcpy(nd_key->nd_sll, flow->arp_sha, ETH_ADDR_LEN);
1688 memcpy(nd_key->nd_tll, flow->arp_tha, ETH_ADDR_LEN);
1695 nl_msg_end_nested(buf, encap);
1700 odp_flow_key_hash(const struct nlattr *key, size_t key_len)
1702 BUILD_ASSERT_DECL(!(NLA_ALIGNTO % sizeof(uint32_t)));
1703 return hash_words((const uint32_t *) key, key_len / sizeof(uint32_t), 0);
1707 log_odp_key_attributes(struct vlog_rate_limit *rl, const char *title,
1708 uint64_t attrs, int out_of_range_attr,
1709 const struct nlattr *key, size_t key_len)
1714 if (VLOG_DROP_DBG(rl)) {
1719 for (i = 0; i < 64; i++) {
1720 if (attrs & (UINT64_C(1) << i)) {
1721 ds_put_format(&s, " %s", ovs_key_attr_to_string(i));
1724 if (out_of_range_attr) {
1725 ds_put_format(&s, " %d (and possibly others)", out_of_range_attr);
1728 ds_put_cstr(&s, ": ");
1729 odp_flow_key_format(key, key_len, &s);
1731 VLOG_DBG("%s:%s", title, ds_cstr(&s));
1736 odp_to_ovs_frag(uint8_t odp_frag, struct flow *flow)
1738 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1740 if (odp_frag > OVS_FRAG_TYPE_LATER) {
1741 VLOG_ERR_RL(&rl, "invalid frag %"PRIu8" in flow key", odp_frag);
1745 if (odp_frag != OVS_FRAG_TYPE_NONE) {
1746 flow->nw_frag |= FLOW_NW_FRAG_ANY;
1747 if (odp_frag == OVS_FRAG_TYPE_LATER) {
1748 flow->nw_frag |= FLOW_NW_FRAG_LATER;
1755 parse_flow_nlattrs(const struct nlattr *key, size_t key_len,
1756 const struct nlattr *attrs[], uint64_t *present_attrsp,
1757 int *out_of_range_attrp)
1759 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
1760 const struct nlattr *nla;
1761 uint64_t present_attrs;
1764 BUILD_ASSERT(OVS_KEY_ATTR_MAX < CHAR_BIT * sizeof present_attrs);
1766 *out_of_range_attrp = 0;
1767 NL_ATTR_FOR_EACH (nla, left, key, key_len) {
1768 uint16_t type = nl_attr_type(nla);
1769 size_t len = nl_attr_get_size(nla);
1770 int expected_len = odp_flow_key_attr_len(type);
1772 if (len != expected_len && expected_len >= 0) {
1773 VLOG_ERR_RL(&rl, "attribute %s has length %zu but should have "
1774 "length %d", ovs_key_attr_to_string(type),
1779 if (type > OVS_KEY_ATTR_MAX) {
1780 *out_of_range_attrp = type;
1782 if (present_attrs & (UINT64_C(1) << type)) {
1783 VLOG_ERR_RL(&rl, "duplicate %s attribute in flow key",
1784 ovs_key_attr_to_string(type));
1788 present_attrs |= UINT64_C(1) << type;
1793 VLOG_ERR_RL(&rl, "trailing garbage in flow key");
1797 *present_attrsp = present_attrs;
1801 static enum odp_key_fitness
1802 check_expectations(uint64_t present_attrs, int out_of_range_attr,
1803 uint64_t expected_attrs,
1804 const struct nlattr *key, size_t key_len)
1806 uint64_t missing_attrs;
1807 uint64_t extra_attrs;
1809 missing_attrs = expected_attrs & ~present_attrs;
1810 if (missing_attrs) {
1811 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
1812 log_odp_key_attributes(&rl, "expected but not present",
1813 missing_attrs, 0, key, key_len);
1814 return ODP_FIT_TOO_LITTLE;
1817 extra_attrs = present_attrs & ~expected_attrs;
1818 if (extra_attrs || out_of_range_attr) {
1819 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
1820 log_odp_key_attributes(&rl, "present but not expected",
1821 extra_attrs, out_of_range_attr, key, key_len);
1822 return ODP_FIT_TOO_MUCH;
1825 return ODP_FIT_PERFECT;
1829 parse_ethertype(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
1830 uint64_t present_attrs, uint64_t *expected_attrs,
1833 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1835 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE)) {
1836 flow->dl_type = nl_attr_get_be16(attrs[OVS_KEY_ATTR_ETHERTYPE]);
1837 if (ntohs(flow->dl_type) < 1536) {
1838 VLOG_ERR_RL(&rl, "invalid Ethertype %"PRIu16" in flow key",
1839 ntohs(flow->dl_type));
1842 *expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE;
1844 flow->dl_type = htons(FLOW_DL_TYPE_NONE);
1849 static enum odp_key_fitness
1850 parse_l2_5_onward(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
1851 uint64_t present_attrs, int out_of_range_attr,
1852 uint64_t expected_attrs, struct flow *flow,
1853 const struct nlattr *key, size_t key_len)
1855 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1857 if (eth_type_mpls(flow->dl_type)) {
1858 expected_attrs |= (UINT64_C(1) << OVS_KEY_ATTR_MPLS);
1860 if (!(present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_MPLS))) {
1861 return ODP_FIT_TOO_LITTLE;
1863 flow->mpls_lse = nl_attr_get_be32(attrs[OVS_KEY_ATTR_MPLS]);
1865 } else if (flow->dl_type == htons(ETH_TYPE_IP)) {
1866 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IPV4;
1867 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IPV4)) {
1868 const struct ovs_key_ipv4 *ipv4_key;
1870 ipv4_key = nl_attr_get(attrs[OVS_KEY_ATTR_IPV4]);
1871 flow->nw_src = ipv4_key->ipv4_src;
1872 flow->nw_dst = ipv4_key->ipv4_dst;
1873 flow->nw_proto = ipv4_key->ipv4_proto;
1874 flow->nw_tos = ipv4_key->ipv4_tos;
1875 flow->nw_ttl = ipv4_key->ipv4_ttl;
1876 if (!odp_to_ovs_frag(ipv4_key->ipv4_frag, flow)) {
1877 return ODP_FIT_ERROR;
1880 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1881 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IPV6;
1882 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IPV6)) {
1883 const struct ovs_key_ipv6 *ipv6_key;
1885 ipv6_key = nl_attr_get(attrs[OVS_KEY_ATTR_IPV6]);
1886 memcpy(&flow->ipv6_src, ipv6_key->ipv6_src, sizeof flow->ipv6_src);
1887 memcpy(&flow->ipv6_dst, ipv6_key->ipv6_dst, sizeof flow->ipv6_dst);
1888 flow->ipv6_label = ipv6_key->ipv6_label;
1889 flow->nw_proto = ipv6_key->ipv6_proto;
1890 flow->nw_tos = ipv6_key->ipv6_tclass;
1891 flow->nw_ttl = ipv6_key->ipv6_hlimit;
1892 if (!odp_to_ovs_frag(ipv6_key->ipv6_frag, flow)) {
1893 return ODP_FIT_ERROR;
1896 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
1897 flow->dl_type == htons(ETH_TYPE_RARP)) {
1898 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ARP;
1899 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ARP)) {
1900 const struct ovs_key_arp *arp_key;
1902 arp_key = nl_attr_get(attrs[OVS_KEY_ATTR_ARP]);
1903 flow->nw_src = arp_key->arp_sip;
1904 flow->nw_dst = arp_key->arp_tip;
1905 if (arp_key->arp_op & htons(0xff00)) {
1906 VLOG_ERR_RL(&rl, "unsupported ARP opcode %"PRIu16" in flow "
1907 "key", ntohs(arp_key->arp_op));
1908 return ODP_FIT_ERROR;
1910 flow->nw_proto = ntohs(arp_key->arp_op);
1911 memcpy(flow->arp_sha, arp_key->arp_sha, ETH_ADDR_LEN);
1912 memcpy(flow->arp_tha, arp_key->arp_tha, ETH_ADDR_LEN);
1916 if (flow->nw_proto == IPPROTO_TCP
1917 && (flow->dl_type == htons(ETH_TYPE_IP) ||
1918 flow->dl_type == htons(ETH_TYPE_IPV6))
1919 && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1920 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_TCP;
1921 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_TCP)) {
1922 const struct ovs_key_tcp *tcp_key;
1924 tcp_key = nl_attr_get(attrs[OVS_KEY_ATTR_TCP]);
1925 flow->tp_src = tcp_key->tcp_src;
1926 flow->tp_dst = tcp_key->tcp_dst;
1928 } else if (flow->nw_proto == IPPROTO_UDP
1929 && (flow->dl_type == htons(ETH_TYPE_IP) ||
1930 flow->dl_type == htons(ETH_TYPE_IPV6))
1931 && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1932 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_UDP;
1933 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_UDP)) {
1934 const struct ovs_key_udp *udp_key;
1936 udp_key = nl_attr_get(attrs[OVS_KEY_ATTR_UDP]);
1937 flow->tp_src = udp_key->udp_src;
1938 flow->tp_dst = udp_key->udp_dst;
1940 } else if (flow->nw_proto == IPPROTO_ICMP
1941 && flow->dl_type == htons(ETH_TYPE_IP)
1942 && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1943 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ICMP;
1944 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ICMP)) {
1945 const struct ovs_key_icmp *icmp_key;
1947 icmp_key = nl_attr_get(attrs[OVS_KEY_ATTR_ICMP]);
1948 flow->tp_src = htons(icmp_key->icmp_type);
1949 flow->tp_dst = htons(icmp_key->icmp_code);
1951 } else if (flow->nw_proto == IPPROTO_ICMPV6
1952 && flow->dl_type == htons(ETH_TYPE_IPV6)
1953 && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1954 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ICMPV6;
1955 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ICMPV6)) {
1956 const struct ovs_key_icmpv6 *icmpv6_key;
1958 icmpv6_key = nl_attr_get(attrs[OVS_KEY_ATTR_ICMPV6]);
1959 flow->tp_src = htons(icmpv6_key->icmpv6_type);
1960 flow->tp_dst = htons(icmpv6_key->icmpv6_code);
1962 if (flow->tp_src == htons(ND_NEIGHBOR_SOLICIT) ||
1963 flow->tp_src == htons(ND_NEIGHBOR_ADVERT)) {
1964 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ND;
1965 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ND)) {
1966 const struct ovs_key_nd *nd_key;
1968 nd_key = nl_attr_get(attrs[OVS_KEY_ATTR_ND]);
1969 memcpy(&flow->nd_target, nd_key->nd_target,
1970 sizeof flow->nd_target);
1971 memcpy(flow->arp_sha, nd_key->nd_sll, ETH_ADDR_LEN);
1972 memcpy(flow->arp_tha, nd_key->nd_tll, ETH_ADDR_LEN);
1978 return check_expectations(present_attrs, out_of_range_attr, expected_attrs,
1982 /* Parse 802.1Q header then encapsulated L3 attributes. */
1983 static enum odp_key_fitness
1984 parse_8021q_onward(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
1985 uint64_t present_attrs, int out_of_range_attr,
1986 uint64_t expected_attrs, struct flow *flow,
1987 const struct nlattr *key, size_t key_len)
1989 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1991 const struct nlattr *encap
1992 = (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ENCAP)
1993 ? attrs[OVS_KEY_ATTR_ENCAP] : NULL);
1994 enum odp_key_fitness encap_fitness;
1995 enum odp_key_fitness fitness;
1998 /* Calulate fitness of outer attributes. */
1999 expected_attrs |= ((UINT64_C(1) << OVS_KEY_ATTR_VLAN) |
2000 (UINT64_C(1) << OVS_KEY_ATTR_ENCAP));
2001 fitness = check_expectations(present_attrs, out_of_range_attr,
2002 expected_attrs, key, key_len);
2004 /* Get the VLAN TCI value. */
2005 if (!(present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN))) {
2006 return ODP_FIT_TOO_LITTLE;
2008 tci = nl_attr_get_be16(attrs[OVS_KEY_ATTR_VLAN]);
2009 if (tci == htons(0)) {
2010 /* Corner case for a truncated 802.1Q header. */
2011 if (fitness == ODP_FIT_PERFECT && nl_attr_get_size(encap)) {
2012 return ODP_FIT_TOO_MUCH;
2015 } else if (!(tci & htons(VLAN_CFI))) {
2016 VLOG_ERR_RL(&rl, "OVS_KEY_ATTR_VLAN 0x%04"PRIx16" is nonzero "
2017 "but CFI bit is not set", ntohs(tci));
2018 return ODP_FIT_ERROR;
2022 * Remove the TPID from dl_type since it's not the real Ethertype. */
2023 flow->vlan_tci = tci;
2024 flow->dl_type = htons(0);
2026 /* Now parse the encapsulated attributes. */
2027 if (!parse_flow_nlattrs(nl_attr_get(encap), nl_attr_get_size(encap),
2028 attrs, &present_attrs, &out_of_range_attr)) {
2029 return ODP_FIT_ERROR;
2033 if (!parse_ethertype(attrs, present_attrs, &expected_attrs, flow)) {
2034 return ODP_FIT_ERROR;
2036 encap_fitness = parse_l2_5_onward(attrs, present_attrs, out_of_range_attr,
2037 expected_attrs, flow, key, key_len);
2039 /* The overall fitness is the worse of the outer and inner attributes. */
2040 return MAX(fitness, encap_fitness);
2043 /* Converts the 'key_len' bytes of OVS_KEY_ATTR_* attributes in 'key' to a flow
2044 * structure in 'flow'. Returns an ODP_FIT_* value that indicates how well
2045 * 'key' fits our expectations for what a flow key should contain.
2047 * The 'in_port' will be the datapath's understanding of the port. The
2048 * caller will need to translate with odp_port_to_ofp_port() if the
2049 * OpenFlow port is needed.
2051 * This function doesn't take the packet itself as an argument because none of
2052 * the currently understood OVS_KEY_ATTR_* attributes require it. Currently,
2053 * it is always possible to infer which additional attribute(s) should appear
2054 * by looking at the attributes for lower-level protocols, e.g. if the network
2055 * protocol in OVS_KEY_ATTR_IPV4 or OVS_KEY_ATTR_IPV6 is IPPROTO_TCP then we
2056 * know that a OVS_KEY_ATTR_TCP attribute must appear and that otherwise it
2057 * must be absent. */
2058 enum odp_key_fitness
2059 odp_flow_key_to_flow(const struct nlattr *key, size_t key_len,
2062 const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1];
2063 uint64_t expected_attrs;
2064 uint64_t present_attrs;
2065 int out_of_range_attr;
2067 memset(flow, 0, sizeof *flow);
2069 /* Parse attributes. */
2070 if (!parse_flow_nlattrs(key, key_len, attrs, &present_attrs,
2071 &out_of_range_attr)) {
2072 return ODP_FIT_ERROR;
2077 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_PRIORITY)) {
2078 flow->skb_priority = nl_attr_get_u32(attrs[OVS_KEY_ATTR_PRIORITY]);
2079 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_PRIORITY;
2082 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_SKB_MARK)) {
2083 flow->skb_mark = nl_attr_get_u32(attrs[OVS_KEY_ATTR_SKB_MARK]);
2084 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_SKB_MARK;
2087 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_TUNNEL)) {
2088 enum odp_key_fitness res;
2090 res = tun_key_from_attr(attrs[OVS_KEY_ATTR_TUNNEL], &flow->tunnel);
2091 if (res == ODP_FIT_ERROR) {
2092 return ODP_FIT_ERROR;
2093 } else if (res == ODP_FIT_PERFECT) {
2094 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_TUNNEL;
2098 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IN_PORT)) {
2099 flow->in_port = nl_attr_get_u32(attrs[OVS_KEY_ATTR_IN_PORT]);
2100 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IN_PORT;
2102 flow->in_port = OVSP_NONE;
2105 /* Ethernet header. */
2106 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ETHERNET)) {
2107 const struct ovs_key_ethernet *eth_key;
2109 eth_key = nl_attr_get(attrs[OVS_KEY_ATTR_ETHERNET]);
2110 memcpy(flow->dl_src, eth_key->eth_src, ETH_ADDR_LEN);
2111 memcpy(flow->dl_dst, eth_key->eth_dst, ETH_ADDR_LEN);
2113 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ETHERNET;
2115 /* Get Ethertype or 802.1Q TPID or FLOW_DL_TYPE_NONE. */
2116 if (!parse_ethertype(attrs, present_attrs, &expected_attrs, flow)) {
2117 return ODP_FIT_ERROR;
2120 if (flow->dl_type == htons(ETH_TYPE_VLAN)) {
2121 return parse_8021q_onward(attrs, present_attrs, out_of_range_attr,
2122 expected_attrs, flow, key, key_len);
2124 return parse_l2_5_onward(attrs, present_attrs, out_of_range_attr,
2125 expected_attrs, flow, key, key_len);
2128 /* Returns 'fitness' as a string, for use in debug messages. */
2130 odp_key_fitness_to_string(enum odp_key_fitness fitness)
2133 case ODP_FIT_PERFECT:
2135 case ODP_FIT_TOO_MUCH:
2137 case ODP_FIT_TOO_LITTLE:
2138 return "too_little";
2146 /* Appends an OVS_ACTION_ATTR_USERSPACE action to 'odp_actions' that specifies
2147 * Netlink PID 'pid'. If 'userdata' is nonnull, adds a userdata attribute
2148 * whose contents are the 'userdata_size' bytes at 'userdata' and returns the
2149 * offset within 'odp_actions' of the start of the cookie. (If 'userdata' is
2150 * null, then the return value is not meaningful.) */
2152 odp_put_userspace_action(uint32_t pid,
2153 const void *userdata, size_t userdata_size,
2154 struct ofpbuf *odp_actions)
2156 size_t userdata_ofs;
2159 offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_USERSPACE);
2160 nl_msg_put_u32(odp_actions, OVS_USERSPACE_ATTR_PID, pid);
2162 userdata_ofs = odp_actions->size + NLA_HDRLEN;
2163 nl_msg_put_unspec(odp_actions, OVS_USERSPACE_ATTR_USERDATA,
2164 userdata, userdata_size);
2168 nl_msg_end_nested(odp_actions, offset);
2170 return userdata_ofs;
2174 odp_put_tunnel_action(const struct flow_tnl *tunnel,
2175 struct ofpbuf *odp_actions)
2177 size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
2178 tun_key_to_attr(odp_actions, tunnel);
2179 nl_msg_end_nested(odp_actions, offset);
2182 /* The commit_odp_actions() function and its helpers. */
2185 commit_set_action(struct ofpbuf *odp_actions, enum ovs_key_attr key_type,
2186 const void *key, size_t key_size)
2188 size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
2189 nl_msg_put_unspec(odp_actions, key_type, key, key_size);
2190 nl_msg_end_nested(odp_actions, offset);
2194 odp_put_skb_mark_action(const uint32_t skb_mark,
2195 struct ofpbuf *odp_actions)
2197 commit_set_action(odp_actions, OVS_KEY_ATTR_SKB_MARK, &skb_mark,
2201 /* If any of the flow key data that ODP actions can modify are different in
2202 * 'base->tunnel' and 'flow->tunnel', appends a set_tunnel ODP action to
2203 * 'odp_actions' that change the flow tunneling information in key from
2204 * 'base->tunnel' into 'flow->tunnel', and then changes 'base->tunnel' in the
2205 * same way. In other words, operates the same as commit_odp_actions(), but
2206 * only on tunneling information. */
2208 commit_odp_tunnel_action(const struct flow *flow, struct flow *base,
2209 struct ofpbuf *odp_actions)
2211 /* A valid IPV4_TUNNEL must have non-zero ip_dst. */
2212 if (flow->tunnel.ip_dst) {
2213 if (!memcmp(&base->tunnel, &flow->tunnel, sizeof base->tunnel)) {
2216 memcpy(&base->tunnel, &flow->tunnel, sizeof base->tunnel);
2217 odp_put_tunnel_action(&base->tunnel, odp_actions);
2222 commit_set_ether_addr_action(const struct flow *flow, struct flow *base,
2223 struct ofpbuf *odp_actions)
2225 struct ovs_key_ethernet eth_key;
2227 if (eth_addr_equals(base->dl_src, flow->dl_src) &&
2228 eth_addr_equals(base->dl_dst, flow->dl_dst)) {
2232 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
2233 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
2235 memcpy(eth_key.eth_src, base->dl_src, ETH_ADDR_LEN);
2236 memcpy(eth_key.eth_dst, base->dl_dst, ETH_ADDR_LEN);
2238 commit_set_action(odp_actions, OVS_KEY_ATTR_ETHERNET,
2239 ð_key, sizeof(eth_key));
2243 commit_vlan_action(const struct flow *flow, struct flow *base,
2244 struct ofpbuf *odp_actions)
2246 if (base->vlan_tci == flow->vlan_tci) {
2250 if (base->vlan_tci & htons(VLAN_CFI)) {
2251 nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_VLAN);
2254 if (flow->vlan_tci & htons(VLAN_CFI)) {
2255 struct ovs_action_push_vlan vlan;
2257 vlan.vlan_tpid = htons(ETH_TYPE_VLAN);
2258 vlan.vlan_tci = flow->vlan_tci;
2259 nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_PUSH_VLAN,
2260 &vlan, sizeof vlan);
2262 base->vlan_tci = flow->vlan_tci;
2266 commit_mpls_action(const struct flow *flow, struct flow *base,
2267 struct ofpbuf *odp_actions)
2269 if (flow->mpls_lse == base->mpls_lse &&
2270 flow->mpls_depth == base->mpls_depth) {
2274 if (flow->mpls_depth < base->mpls_depth) {
2275 if (base->mpls_depth - flow->mpls_depth > 1) {
2276 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
2277 VLOG_WARN_RL(&rl, "Multiple mpls_pop actions reduced to "
2278 " a single mpls_pop action");
2281 nl_msg_put_be16(odp_actions, OVS_ACTION_ATTR_POP_MPLS, flow->dl_type);
2282 } else if (flow->mpls_depth > base->mpls_depth) {
2283 struct ovs_action_push_mpls *mpls;
2285 if (flow->mpls_depth - base->mpls_depth > 1) {
2286 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
2287 VLOG_WARN_RL(&rl, "Multiple mpls_push actions reduced to "
2288 " a single mpls_push action");
2291 mpls = nl_msg_put_unspec_uninit(odp_actions, OVS_ACTION_ATTR_PUSH_MPLS,
2293 memset(mpls, 0, sizeof *mpls);
2294 mpls->mpls_ethertype = flow->dl_type;
2295 mpls->mpls_lse = flow->mpls_lse;
2297 struct ovs_key_mpls mpls_key;
2299 mpls_key.mpls_top_lse = flow->mpls_lse;
2300 commit_set_action(odp_actions, OVS_KEY_ATTR_MPLS,
2301 &mpls_key, sizeof(mpls_key));
2304 base->dl_type = flow->dl_type;
2305 base->mpls_lse = flow->mpls_lse;
2306 base->mpls_depth = flow->mpls_depth;
2310 commit_set_ipv4_action(const struct flow *flow, struct flow *base,
2311 struct ofpbuf *odp_actions)
2313 struct ovs_key_ipv4 ipv4_key;
2315 if (base->nw_src == flow->nw_src &&
2316 base->nw_dst == flow->nw_dst &&
2317 base->nw_tos == flow->nw_tos &&
2318 base->nw_ttl == flow->nw_ttl &&
2319 base->nw_frag == flow->nw_frag) {
2323 ipv4_key.ipv4_src = base->nw_src = flow->nw_src;
2324 ipv4_key.ipv4_dst = base->nw_dst = flow->nw_dst;
2325 ipv4_key.ipv4_tos = base->nw_tos = flow->nw_tos;
2326 ipv4_key.ipv4_ttl = base->nw_ttl = flow->nw_ttl;
2327 ipv4_key.ipv4_proto = base->nw_proto;
2328 ipv4_key.ipv4_frag = ovs_to_odp_frag(base->nw_frag);
2330 commit_set_action(odp_actions, OVS_KEY_ATTR_IPV4,
2331 &ipv4_key, sizeof(ipv4_key));
2335 commit_set_ipv6_action(const struct flow *flow, struct flow *base,
2336 struct ofpbuf *odp_actions)
2338 struct ovs_key_ipv6 ipv6_key;
2340 if (ipv6_addr_equals(&base->ipv6_src, &flow->ipv6_src) &&
2341 ipv6_addr_equals(&base->ipv6_dst, &flow->ipv6_dst) &&
2342 base->ipv6_label == flow->ipv6_label &&
2343 base->nw_tos == flow->nw_tos &&
2344 base->nw_ttl == flow->nw_ttl &&
2345 base->nw_frag == flow->nw_frag) {
2349 base->ipv6_src = flow->ipv6_src;
2350 memcpy(&ipv6_key.ipv6_src, &base->ipv6_src, sizeof(ipv6_key.ipv6_src));
2351 base->ipv6_dst = flow->ipv6_dst;
2352 memcpy(&ipv6_key.ipv6_dst, &base->ipv6_dst, sizeof(ipv6_key.ipv6_dst));
2354 ipv6_key.ipv6_label = base->ipv6_label = flow->ipv6_label;
2355 ipv6_key.ipv6_tclass = base->nw_tos = flow->nw_tos;
2356 ipv6_key.ipv6_hlimit = base->nw_ttl = flow->nw_ttl;
2357 ipv6_key.ipv6_proto = base->nw_proto;
2358 ipv6_key.ipv6_frag = ovs_to_odp_frag(base->nw_frag);
2360 commit_set_action(odp_actions, OVS_KEY_ATTR_IPV6,
2361 &ipv6_key, sizeof(ipv6_key));
2365 commit_set_nw_action(const struct flow *flow, struct flow *base,
2366 struct ofpbuf *odp_actions)
2368 /* Check if flow really have an IP header. */
2369 if (!flow->nw_proto) {
2373 if (base->dl_type == htons(ETH_TYPE_IP)) {
2374 commit_set_ipv4_action(flow, base, odp_actions);
2375 } else if (base->dl_type == htons(ETH_TYPE_IPV6)) {
2376 commit_set_ipv6_action(flow, base, odp_actions);
2381 commit_set_port_action(const struct flow *flow, struct flow *base,
2382 struct ofpbuf *odp_actions)
2384 if (!is_ip_any(base) || (!base->tp_src && !base->tp_dst)) {
2388 if (base->tp_src == flow->tp_src &&
2389 base->tp_dst == flow->tp_dst) {
2393 if (flow->nw_proto == IPPROTO_TCP) {
2394 struct ovs_key_tcp port_key;
2396 port_key.tcp_src = base->tp_src = flow->tp_src;
2397 port_key.tcp_dst = base->tp_dst = flow->tp_dst;
2399 commit_set_action(odp_actions, OVS_KEY_ATTR_TCP,
2400 &port_key, sizeof(port_key));
2402 } else if (flow->nw_proto == IPPROTO_UDP) {
2403 struct ovs_key_udp port_key;
2405 port_key.udp_src = base->tp_src = flow->tp_src;
2406 port_key.udp_dst = base->tp_dst = flow->tp_dst;
2408 commit_set_action(odp_actions, OVS_KEY_ATTR_UDP,
2409 &port_key, sizeof(port_key));
2414 commit_set_priority_action(const struct flow *flow, struct flow *base,
2415 struct ofpbuf *odp_actions)
2417 if (base->skb_priority == flow->skb_priority) {
2420 base->skb_priority = flow->skb_priority;
2422 commit_set_action(odp_actions, OVS_KEY_ATTR_PRIORITY,
2423 &base->skb_priority, sizeof(base->skb_priority));
2427 commit_set_skb_mark_action(const struct flow *flow, struct flow *base,
2428 struct ofpbuf *odp_actions)
2430 if (base->skb_mark == flow->skb_mark) {
2433 base->skb_mark = flow->skb_mark;
2435 odp_put_skb_mark_action(base->skb_mark, odp_actions);
2437 /* If any of the flow key data that ODP actions can modify are different in
2438 * 'base' and 'flow', appends ODP actions to 'odp_actions' that change the flow
2439 * key from 'base' into 'flow', and then changes 'base' the same way. Does not
2440 * commit set_tunnel actions. Users should call commit_odp_tunnel_action()
2441 * in addition to this function if needed. */
2443 commit_odp_actions(const struct flow *flow, struct flow *base,
2444 struct ofpbuf *odp_actions)
2446 commit_set_ether_addr_action(flow, base, odp_actions);
2447 commit_vlan_action(flow, base, odp_actions);
2448 commit_set_nw_action(flow, base, odp_actions);
2449 commit_set_port_action(flow, base, odp_actions);
2450 /* Commiting MPLS actions should occur after committing nw and port
2451 * actions. This is because committing MPLS actions may alter a packet so
2452 * that it is no longer IP and thus nw and port actions are no longer valid.
2454 commit_mpls_action(flow, base, odp_actions);
2455 commit_set_priority_action(flow, base, odp_actions);
2456 commit_set_skb_mark_action(flow, base, odp_actions);