]> git.proxmox.com Git - ovs.git/blob - lib/odp-util.c
datapath: Add support for tun_key to Open vSwitch datapath
[ovs.git] / lib / odp-util.c
1 /*
2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
3 *
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:
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
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.
15 */
16
17 #include <config.h>
18 #include <arpa/inet.h>
19 #include "odp-util.h"
20 #include <errno.h>
21 #include <inttypes.h>
22 #include <math.h>
23 #include <netinet/in.h>
24 #include <netinet/icmp6.h>
25 #include <stdlib.h>
26 #include <string.h>
27 #include "byte-order.h"
28 #include "coverage.h"
29 #include "dynamic-string.h"
30 #include "flow.h"
31 #include "netlink.h"
32 #include "ofpbuf.h"
33 #include "packets.h"
34 #include "simap.h"
35 #include "timeval.h"
36 #include "util.h"
37 #include "vlog.h"
38
39 VLOG_DEFINE_THIS_MODULE(odp_util);
40
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.
45 */
46
47 /* The set of characters that may separate one action or one key attribute
48 * from another. */
49 static const char *delimiters = ", \t\r\n";
50
51 static int parse_odp_key_attr(const char *, const struct simap *port_names,
52 struct ofpbuf *);
53 static void format_odp_key_attr(const struct nlattr *a, struct ds *ds);
54
55 /* Returns one the following for the action with the given OVS_ACTION_ATTR_*
56 * 'type':
57 *
58 * - For an action whose argument has a fixed length, returned that
59 * nonnegative length in bytes.
60 *
61 * - For an action with a variable-length argument, returns -2.
62 *
63 * - For an invalid 'type', returns -1. */
64 static int
65 odp_action_len(uint16_t type)
66 {
67 if (type > OVS_ACTION_ATTR_MAX) {
68 return -1;
69 }
70
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_SET: return -2;
77 case OVS_ACTION_ATTR_SAMPLE: return -2;
78
79 case OVS_ACTION_ATTR_UNSPEC:
80 case __OVS_ACTION_ATTR_MAX:
81 return -1;
82 }
83
84 return -1;
85 }
86
87 static const char *
88 ovs_key_attr_to_string(enum ovs_key_attr attr)
89 {
90 static char unknown_attr[3 + INT_STRLEN(unsigned int) + 1];
91
92 switch (attr) {
93 case OVS_KEY_ATTR_UNSPEC: return "unspec";
94 case OVS_KEY_ATTR_ENCAP: return "encap";
95 case OVS_KEY_ATTR_PRIORITY: return "priority";
96 case OVS_KEY_ATTR_TUN_ID: return "tun_id";
97 case OVS_KEY_ATTR_IPV4_TUNNEL: return "ipv4_tunnel";
98 case OVS_KEY_ATTR_IN_PORT: return "in_port";
99 case OVS_KEY_ATTR_ETHERNET: return "eth";
100 case OVS_KEY_ATTR_VLAN: return "vlan";
101 case OVS_KEY_ATTR_ETHERTYPE: return "eth_type";
102 case OVS_KEY_ATTR_IPV4: return "ipv4";
103 case OVS_KEY_ATTR_IPV6: return "ipv6";
104 case OVS_KEY_ATTR_TCP: return "tcp";
105 case OVS_KEY_ATTR_UDP: return "udp";
106 case OVS_KEY_ATTR_ICMP: return "icmp";
107 case OVS_KEY_ATTR_ICMPV6: return "icmpv6";
108 case OVS_KEY_ATTR_ARP: return "arp";
109 case OVS_KEY_ATTR_ND: return "nd";
110
111 case __OVS_KEY_ATTR_MAX:
112 default:
113 snprintf(unknown_attr, sizeof unknown_attr, "key%u",
114 (unsigned int) attr);
115 return unknown_attr;
116 }
117 }
118
119 static void
120 format_generic_odp_action(struct ds *ds, const struct nlattr *a)
121 {
122 size_t len = nl_attr_get_size(a);
123
124 ds_put_format(ds, "action%"PRId16, nl_attr_type(a));
125 if (len) {
126 const uint8_t *unspec;
127 unsigned int i;
128
129 unspec = nl_attr_get(a);
130 for (i = 0; i < len; i++) {
131 ds_put_char(ds, i ? ' ': '(');
132 ds_put_format(ds, "%02x", unspec[i]);
133 }
134 ds_put_char(ds, ')');
135 }
136 }
137
138 static void
139 format_odp_sample_action(struct ds *ds, const struct nlattr *attr)
140 {
141 static const struct nl_policy ovs_sample_policy[] = {
142 [OVS_SAMPLE_ATTR_PROBABILITY] = { .type = NL_A_U32 },
143 [OVS_SAMPLE_ATTR_ACTIONS] = { .type = NL_A_NESTED }
144 };
145 struct nlattr *a[ARRAY_SIZE(ovs_sample_policy)];
146 double percentage;
147 const struct nlattr *nla_acts;
148 int len;
149
150 ds_put_cstr(ds, "sample");
151
152 if (!nl_parse_nested(attr, ovs_sample_policy, a, ARRAY_SIZE(a))) {
153 ds_put_cstr(ds, "(error)");
154 return;
155 }
156
157 percentage = (100.0 * nl_attr_get_u32(a[OVS_SAMPLE_ATTR_PROBABILITY])) /
158 UINT32_MAX;
159
160 ds_put_format(ds, "(sample=%.1f%%,", percentage);
161
162 ds_put_cstr(ds, "actions(");
163 nla_acts = nl_attr_get(a[OVS_SAMPLE_ATTR_ACTIONS]);
164 len = nl_attr_get_size(a[OVS_SAMPLE_ATTR_ACTIONS]);
165 format_odp_actions(ds, nla_acts, len);
166 ds_put_format(ds, "))");
167 }
168
169 static const char *
170 slow_path_reason_to_string(enum slow_path_reason bit)
171 {
172 switch (bit) {
173 case SLOW_CFM:
174 return "cfm";
175 case SLOW_LACP:
176 return "lacp";
177 case SLOW_STP:
178 return "stp";
179 case SLOW_IN_BAND:
180 return "in_band";
181 case SLOW_CONTROLLER:
182 return "controller";
183 case SLOW_MATCH:
184 return "match";
185 default:
186 return NULL;
187 }
188 }
189
190 static void
191 format_slow_path_reason(struct ds *ds, uint32_t slow)
192 {
193 uint32_t bad = 0;
194
195 while (slow) {
196 uint32_t bit = rightmost_1bit(slow);
197 const char *s;
198
199 s = slow_path_reason_to_string(bit);
200 if (s) {
201 ds_put_format(ds, "%s,", s);
202 } else {
203 bad |= bit;
204 }
205
206 slow &= ~bit;
207 }
208
209 if (bad) {
210 ds_put_format(ds, "0x%"PRIx32",", bad);
211 }
212 ds_chomp(ds, ',');
213 }
214
215 static void
216 format_odp_userspace_action(struct ds *ds, const struct nlattr *attr)
217 {
218 static const struct nl_policy ovs_userspace_policy[] = {
219 [OVS_USERSPACE_ATTR_PID] = { .type = NL_A_U32 },
220 [OVS_USERSPACE_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
221 };
222 struct nlattr *a[ARRAY_SIZE(ovs_userspace_policy)];
223
224 if (!nl_parse_nested(attr, ovs_userspace_policy, a, ARRAY_SIZE(a))) {
225 ds_put_cstr(ds, "userspace(error)");
226 return;
227 }
228
229 ds_put_format(ds, "userspace(pid=%"PRIu32,
230 nl_attr_get_u32(a[OVS_USERSPACE_ATTR_PID]));
231
232 if (a[OVS_USERSPACE_ATTR_USERDATA]) {
233 uint64_t userdata = nl_attr_get_u64(a[OVS_USERSPACE_ATTR_USERDATA]);
234 union user_action_cookie cookie;
235
236 memcpy(&cookie, &userdata, sizeof cookie);
237
238 switch (cookie.type) {
239 case USER_ACTION_COOKIE_SFLOW:
240 ds_put_format(ds, ",sFlow("
241 "vid=%"PRIu16",pcp=%"PRIu8",output=%"PRIu32")",
242 vlan_tci_to_vid(cookie.sflow.vlan_tci),
243 vlan_tci_to_pcp(cookie.sflow.vlan_tci),
244 cookie.sflow.output);
245 break;
246
247 case USER_ACTION_COOKIE_SLOW_PATH:
248 ds_put_cstr(ds, ",slow_path(");
249 if (cookie.slow_path.reason) {
250 format_slow_path_reason(ds, cookie.slow_path.reason);
251 }
252 ds_put_char(ds, ')');
253 break;
254
255 case USER_ACTION_COOKIE_UNSPEC:
256 default:
257 ds_put_format(ds, ",userdata=0x%"PRIx64, userdata);
258 break;
259 }
260 }
261
262 ds_put_char(ds, ')');
263 }
264
265 static void
266 format_vlan_tci(struct ds *ds, ovs_be16 vlan_tci)
267 {
268 ds_put_format(ds, "vid=%"PRIu16",pcp=%d",
269 vlan_tci_to_vid(vlan_tci),
270 vlan_tci_to_pcp(vlan_tci));
271 if (!(vlan_tci & htons(VLAN_CFI))) {
272 ds_put_cstr(ds, ",cfi=0");
273 }
274 }
275
276 static void
277 format_odp_action(struct ds *ds, const struct nlattr *a)
278 {
279 int expected_len;
280 enum ovs_action_attr type = nl_attr_type(a);
281 const struct ovs_action_push_vlan *vlan;
282
283 expected_len = odp_action_len(nl_attr_type(a));
284 if (expected_len != -2 && nl_attr_get_size(a) != expected_len) {
285 ds_put_format(ds, "bad length %zu, expected %d for: ",
286 nl_attr_get_size(a), expected_len);
287 format_generic_odp_action(ds, a);
288 return;
289 }
290
291 switch (type) {
292 case OVS_ACTION_ATTR_OUTPUT:
293 ds_put_format(ds, "%"PRIu16, nl_attr_get_u32(a));
294 break;
295 case OVS_ACTION_ATTR_USERSPACE:
296 format_odp_userspace_action(ds, a);
297 break;
298 case OVS_ACTION_ATTR_SET:
299 ds_put_cstr(ds, "set(");
300 format_odp_key_attr(nl_attr_get(a), ds);
301 ds_put_cstr(ds, ")");
302 break;
303 case OVS_ACTION_ATTR_PUSH_VLAN:
304 vlan = nl_attr_get(a);
305 ds_put_cstr(ds, "push_vlan(");
306 if (vlan->vlan_tpid != htons(ETH_TYPE_VLAN)) {
307 ds_put_format(ds, "tpid=0x%04"PRIx16",", ntohs(vlan->vlan_tpid));
308 }
309 format_vlan_tci(ds, vlan->vlan_tci);
310 ds_put_char(ds, ')');
311 break;
312 case OVS_ACTION_ATTR_POP_VLAN:
313 ds_put_cstr(ds, "pop_vlan");
314 break;
315 case OVS_ACTION_ATTR_SAMPLE:
316 format_odp_sample_action(ds, a);
317 break;
318 case OVS_ACTION_ATTR_UNSPEC:
319 case __OVS_ACTION_ATTR_MAX:
320 default:
321 format_generic_odp_action(ds, a);
322 break;
323 }
324 }
325
326 void
327 format_odp_actions(struct ds *ds, const struct nlattr *actions,
328 size_t actions_len)
329 {
330 if (actions_len) {
331 const struct nlattr *a;
332 unsigned int left;
333
334 NL_ATTR_FOR_EACH (a, left, actions, actions_len) {
335 if (a != actions) {
336 ds_put_char(ds, ',');
337 }
338 format_odp_action(ds, a);
339 }
340 if (left) {
341 int i;
342
343 if (left == actions_len) {
344 ds_put_cstr(ds, "<empty>");
345 }
346 ds_put_format(ds, ",***%u leftover bytes*** (", left);
347 for (i = 0; i < left; i++) {
348 ds_put_format(ds, "%02x", ((const uint8_t *) a)[i]);
349 }
350 ds_put_char(ds, ')');
351 }
352 } else {
353 ds_put_cstr(ds, "drop");
354 }
355 }
356
357 static int
358 parse_odp_action(const char *s, const struct simap *port_names,
359 struct ofpbuf *actions)
360 {
361 /* Many of the sscanf calls in this function use oversized destination
362 * fields because some sscanf() implementations truncate the range of %i
363 * directives, so that e.g. "%"SCNi16 interprets input of "0xfedc" as a
364 * value of 0x7fff. The other alternatives are to allow only a single
365 * radix (e.g. decimal or hexadecimal) or to write more sophisticated
366 * parsers.
367 *
368 * The tun_id parser has to use an alternative approach because there is no
369 * type larger than 64 bits. */
370
371 {
372 unsigned long long int port;
373 int n = -1;
374
375 if (sscanf(s, "%lli%n", &port, &n) > 0 && n > 0) {
376 nl_msg_put_u32(actions, OVS_ACTION_ATTR_OUTPUT, port);
377 return n;
378 }
379 }
380
381 if (port_names) {
382 int len = strcspn(s, delimiters);
383 struct simap_node *node;
384
385 node = simap_find_len(port_names, s, len);
386 if (node) {
387 nl_msg_put_u32(actions, OVS_ACTION_ATTR_OUTPUT, node->data);
388 return len;
389 }
390 }
391
392 {
393 unsigned long long int pid;
394 unsigned long long int output;
395 char userdata_s[32];
396 int vid, pcp;
397 int n = -1;
398
399 if (sscanf(s, "userspace(pid=%lli)%n", &pid, &n) > 0 && n > 0) {
400 odp_put_userspace_action(pid, NULL, actions);
401 return n;
402 } else if (sscanf(s, "userspace(pid=%lli,sFlow(vid=%i,"
403 "pcp=%i,output=%lli))%n",
404 &pid, &vid, &pcp, &output, &n) > 0 && n > 0) {
405 union user_action_cookie cookie;
406 uint16_t tci;
407
408 tci = vid | (pcp << VLAN_PCP_SHIFT);
409 if (tci) {
410 tci |= VLAN_CFI;
411 }
412
413 cookie.type = USER_ACTION_COOKIE_SFLOW;
414 cookie.sflow.vlan_tci = htons(tci);
415 cookie.sflow.output = output;
416 odp_put_userspace_action(pid, &cookie, actions);
417 return n;
418 } else if (sscanf(s, "userspace(pid=%lli,slow_path(%n", &pid, &n) > 0
419 && n > 0) {
420 union user_action_cookie cookie;
421
422 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
423 cookie.slow_path.unused = 0;
424 cookie.slow_path.reason = 0;
425
426 while (s[n] != ')') {
427 uint32_t bit;
428
429 for (bit = 1; bit; bit <<= 1) {
430 const char *reason = slow_path_reason_to_string(bit);
431 size_t len = strlen(reason);
432
433 if (reason
434 && !strncmp(s + n, reason, len)
435 && (s[n + len] == ',' || s[n + len] == ')'))
436 {
437 cookie.slow_path.reason |= bit;
438 n += len + (s[n + len] == ',');
439 break;
440 }
441 }
442
443 if (!bit) {
444 return -EINVAL;
445 }
446 }
447 if (s[n + 1] != ')') {
448 return -EINVAL;
449 }
450 n += 2;
451
452 odp_put_userspace_action(pid, &cookie, actions);
453 return n;
454 } else if (sscanf(s, "userspace(pid=%lli,userdata="
455 "%31[x0123456789abcdefABCDEF])%n", &pid, userdata_s,
456 &n) > 0 && n > 0) {
457 union user_action_cookie cookie;
458 uint64_t userdata;
459
460 userdata = strtoull(userdata_s, NULL, 0);
461 memcpy(&cookie, &userdata, sizeof cookie);
462 odp_put_userspace_action(pid, &cookie, actions);
463 return n;
464 }
465 }
466
467 if (!strncmp(s, "set(", 4)) {
468 size_t start_ofs;
469 int retval;
470
471 start_ofs = nl_msg_start_nested(actions, OVS_ACTION_ATTR_SET);
472 retval = parse_odp_key_attr(s + 4, port_names, actions);
473 if (retval < 0) {
474 return retval;
475 }
476 if (s[retval + 4] != ')') {
477 return -EINVAL;
478 }
479 nl_msg_end_nested(actions, start_ofs);
480 return retval + 5;
481 }
482
483 {
484 struct ovs_action_push_vlan push;
485 int tpid = ETH_TYPE_VLAN;
486 int vid, pcp;
487 int cfi = 1;
488 int n = -1;
489
490 if ((sscanf(s, "push_vlan(vid=%i,pcp=%i)%n", &vid, &pcp, &n) > 0
491 && n > 0)
492 || (sscanf(s, "push_vlan(vid=%i,pcp=%i,cfi=%i)%n",
493 &vid, &pcp, &cfi, &n) > 0 && n > 0)
494 || (sscanf(s, "push_vlan(tpid=%i,vid=%i,pcp=%i)%n",
495 &tpid, &vid, &pcp, &n) > 0 && n > 0)
496 || (sscanf(s, "push_vlan(tpid=%i,vid=%i,pcp=%i,cfi=%i)%n",
497 &tpid, &vid, &pcp, &cfi, &n) > 0 && n > 0)) {
498 push.vlan_tpid = htons(tpid);
499 push.vlan_tci = htons((vid << VLAN_VID_SHIFT)
500 | (pcp << VLAN_PCP_SHIFT)
501 | (cfi ? VLAN_CFI : 0));
502 nl_msg_put_unspec(actions, OVS_ACTION_ATTR_PUSH_VLAN,
503 &push, sizeof push);
504
505 return n;
506 }
507 }
508
509 if (!strncmp(s, "pop_vlan", 8)) {
510 nl_msg_put_flag(actions, OVS_ACTION_ATTR_POP_VLAN);
511 return 8;
512 }
513
514 {
515 double percentage;
516 int n = -1;
517
518 if (sscanf(s, "sample(sample=%lf%%,actions(%n", &percentage, &n) > 0
519 && percentage >= 0. && percentage <= 100.0
520 && n > 0) {
521 size_t sample_ofs, actions_ofs;
522 double probability;
523
524 probability = floor(UINT32_MAX * (percentage / 100.0) + .5);
525 sample_ofs = nl_msg_start_nested(actions, OVS_ACTION_ATTR_SAMPLE);
526 nl_msg_put_u32(actions, OVS_SAMPLE_ATTR_PROBABILITY,
527 (probability <= 0 ? 0
528 : probability >= UINT32_MAX ? UINT32_MAX
529 : probability));
530
531 actions_ofs = nl_msg_start_nested(actions,
532 OVS_SAMPLE_ATTR_ACTIONS);
533 for (;;) {
534 int retval;
535
536 n += strspn(s + n, delimiters);
537 if (s[n] == ')') {
538 break;
539 }
540
541 retval = parse_odp_action(s + n, port_names, actions);
542 if (retval < 0) {
543 return retval;
544 }
545 n += retval;
546 }
547 nl_msg_end_nested(actions, actions_ofs);
548 nl_msg_end_nested(actions, sample_ofs);
549
550 return s[n + 1] == ')' ? n + 2 : -EINVAL;
551 }
552 }
553
554 return -EINVAL;
555 }
556
557 /* Parses the string representation of datapath actions, in the format output
558 * by format_odp_action(). Returns 0 if successful, otherwise a positive errno
559 * value. On success, the ODP actions are appended to 'actions' as a series of
560 * Netlink attributes. On failure, no data is appended to 'actions'. Either
561 * way, 'actions''s data might be reallocated. */
562 int
563 odp_actions_from_string(const char *s, const struct simap *port_names,
564 struct ofpbuf *actions)
565 {
566 size_t old_size;
567
568 if (!strcasecmp(s, "drop")) {
569 return 0;
570 }
571
572 old_size = actions->size;
573 for (;;) {
574 int retval;
575
576 s += strspn(s, delimiters);
577 if (!*s) {
578 return 0;
579 }
580
581 retval = parse_odp_action(s, port_names, actions);
582 if (retval < 0 || !strchr(delimiters, s[retval])) {
583 actions->size = old_size;
584 return -retval;
585 }
586 s += retval;
587 }
588
589 return 0;
590 }
591 \f
592 /* Returns the correct length of the payload for a flow key attribute of the
593 * specified 'type', -1 if 'type' is unknown, or -2 if the attribute's payload
594 * is variable length. */
595 static int
596 odp_flow_key_attr_len(uint16_t type)
597 {
598 if (type > OVS_KEY_ATTR_MAX) {
599 return -1;
600 }
601
602 switch ((enum ovs_key_attr) type) {
603 case OVS_KEY_ATTR_ENCAP: return -2;
604 case OVS_KEY_ATTR_PRIORITY: return 4;
605 case OVS_KEY_ATTR_TUN_ID: return 8;
606 case OVS_KEY_ATTR_IPV4_TUNNEL: return sizeof(struct ovs_key_ipv4_tunnel);
607 case OVS_KEY_ATTR_IN_PORT: return 4;
608 case OVS_KEY_ATTR_ETHERNET: return sizeof(struct ovs_key_ethernet);
609 case OVS_KEY_ATTR_VLAN: return sizeof(ovs_be16);
610 case OVS_KEY_ATTR_ETHERTYPE: return 2;
611 case OVS_KEY_ATTR_IPV4: return sizeof(struct ovs_key_ipv4);
612 case OVS_KEY_ATTR_IPV6: return sizeof(struct ovs_key_ipv6);
613 case OVS_KEY_ATTR_TCP: return sizeof(struct ovs_key_tcp);
614 case OVS_KEY_ATTR_UDP: return sizeof(struct ovs_key_udp);
615 case OVS_KEY_ATTR_ICMP: return sizeof(struct ovs_key_icmp);
616 case OVS_KEY_ATTR_ICMPV6: return sizeof(struct ovs_key_icmpv6);
617 case OVS_KEY_ATTR_ARP: return sizeof(struct ovs_key_arp);
618 case OVS_KEY_ATTR_ND: return sizeof(struct ovs_key_nd);
619
620 case OVS_KEY_ATTR_UNSPEC:
621 case __OVS_KEY_ATTR_MAX:
622 return -1;
623 }
624
625 return -1;
626 }
627
628 static void
629 format_generic_odp_key(const struct nlattr *a, struct ds *ds)
630 {
631 size_t len = nl_attr_get_size(a);
632 if (len) {
633 const uint8_t *unspec;
634 unsigned int i;
635
636 unspec = nl_attr_get(a);
637 for (i = 0; i < len; i++) {
638 ds_put_char(ds, i ? ' ': '(');
639 ds_put_format(ds, "%02x", unspec[i]);
640 }
641 ds_put_char(ds, ')');
642 }
643 }
644
645 static const char *
646 ovs_frag_type_to_string(enum ovs_frag_type type)
647 {
648 switch (type) {
649 case OVS_FRAG_TYPE_NONE:
650 return "no";
651 case OVS_FRAG_TYPE_FIRST:
652 return "first";
653 case OVS_FRAG_TYPE_LATER:
654 return "later";
655 case __OVS_FRAG_TYPE_MAX:
656 default:
657 return "<error>";
658 }
659 }
660
661 static void
662 format_odp_key_attr(const struct nlattr *a, struct ds *ds)
663 {
664 const struct ovs_key_ethernet *eth_key;
665 const struct ovs_key_ipv4 *ipv4_key;
666 const struct ovs_key_ipv6 *ipv6_key;
667 const struct ovs_key_tcp *tcp_key;
668 const struct ovs_key_udp *udp_key;
669 const struct ovs_key_icmp *icmp_key;
670 const struct ovs_key_icmpv6 *icmpv6_key;
671 const struct ovs_key_arp *arp_key;
672 const struct ovs_key_nd *nd_key;
673 const struct ovs_key_ipv4_tunnel *ipv4_tun_key;
674 enum ovs_key_attr attr = nl_attr_type(a);
675 int expected_len;
676
677 ds_put_cstr(ds, ovs_key_attr_to_string(attr));
678 expected_len = odp_flow_key_attr_len(nl_attr_type(a));
679 if (expected_len != -2 && nl_attr_get_size(a) != expected_len) {
680 ds_put_format(ds, "(bad length %zu, expected %d)",
681 nl_attr_get_size(a),
682 odp_flow_key_attr_len(nl_attr_type(a)));
683 format_generic_odp_key(a, ds);
684 return;
685 }
686
687 switch (attr) {
688 case OVS_KEY_ATTR_ENCAP:
689 ds_put_cstr(ds, "(");
690 if (nl_attr_get_size(a)) {
691 odp_flow_key_format(nl_attr_get(a), nl_attr_get_size(a), ds);
692 }
693 ds_put_char(ds, ')');
694 break;
695
696 case OVS_KEY_ATTR_PRIORITY:
697 ds_put_format(ds, "(%"PRIu32")", nl_attr_get_u32(a));
698 break;
699
700 case OVS_KEY_ATTR_TUN_ID:
701 ds_put_format(ds, "(%#"PRIx64")", ntohll(nl_attr_get_be64(a)));
702 break;
703
704 case OVS_KEY_ATTR_IPV4_TUNNEL:
705 ipv4_tun_key = nl_attr_get(a);
706 ds_put_format(ds, "(tun_id=0x%"PRIx64",flags=0x%"PRIx32
707 ",src="IP_FMT",dst="IP_FMT",tos=0x%"PRIx8",ttl=%"PRIu8")",
708 ntohll(ipv4_tun_key->tun_id), ipv4_tun_key->tun_flags,
709 IP_ARGS(&ipv4_tun_key->ipv4_src),
710 IP_ARGS(&ipv4_tun_key->ipv4_dst),
711 ipv4_tun_key->ipv4_tos, ipv4_tun_key->ipv4_ttl);
712 break;
713
714 case OVS_KEY_ATTR_IN_PORT:
715 ds_put_format(ds, "(%"PRIu32")", nl_attr_get_u32(a));
716 break;
717
718 case OVS_KEY_ATTR_ETHERNET:
719 eth_key = nl_attr_get(a);
720 ds_put_format(ds, "(src="ETH_ADDR_FMT",dst="ETH_ADDR_FMT")",
721 ETH_ADDR_ARGS(eth_key->eth_src),
722 ETH_ADDR_ARGS(eth_key->eth_dst));
723 break;
724
725 case OVS_KEY_ATTR_VLAN:
726 ds_put_char(ds, '(');
727 format_vlan_tci(ds, nl_attr_get_be16(a));
728 ds_put_char(ds, ')');
729 break;
730
731 case OVS_KEY_ATTR_ETHERTYPE:
732 ds_put_format(ds, "(0x%04"PRIx16")",
733 ntohs(nl_attr_get_be16(a)));
734 break;
735
736 case OVS_KEY_ATTR_IPV4:
737 ipv4_key = nl_attr_get(a);
738 ds_put_format(ds, "(src="IP_FMT",dst="IP_FMT",proto=%"PRIu8
739 ",tos=%#"PRIx8",ttl=%"PRIu8",frag=%s)",
740 IP_ARGS(&ipv4_key->ipv4_src),
741 IP_ARGS(&ipv4_key->ipv4_dst),
742 ipv4_key->ipv4_proto, ipv4_key->ipv4_tos,
743 ipv4_key->ipv4_ttl,
744 ovs_frag_type_to_string(ipv4_key->ipv4_frag));
745 break;
746
747 case OVS_KEY_ATTR_IPV6: {
748 char src_str[INET6_ADDRSTRLEN];
749 char dst_str[INET6_ADDRSTRLEN];
750
751 ipv6_key = nl_attr_get(a);
752 inet_ntop(AF_INET6, ipv6_key->ipv6_src, src_str, sizeof src_str);
753 inet_ntop(AF_INET6, ipv6_key->ipv6_dst, dst_str, sizeof dst_str);
754
755 ds_put_format(ds, "(src=%s,dst=%s,label=%#"PRIx32",proto=%"PRIu8
756 ",tclass=%#"PRIx8",hlimit=%"PRIu8",frag=%s)",
757 src_str, dst_str, ntohl(ipv6_key->ipv6_label),
758 ipv6_key->ipv6_proto, ipv6_key->ipv6_tclass,
759 ipv6_key->ipv6_hlimit,
760 ovs_frag_type_to_string(ipv6_key->ipv6_frag));
761 break;
762 }
763
764 case OVS_KEY_ATTR_TCP:
765 tcp_key = nl_attr_get(a);
766 ds_put_format(ds, "(src=%"PRIu16",dst=%"PRIu16")",
767 ntohs(tcp_key->tcp_src), ntohs(tcp_key->tcp_dst));
768 break;
769
770 case OVS_KEY_ATTR_UDP:
771 udp_key = nl_attr_get(a);
772 ds_put_format(ds, "(src=%"PRIu16",dst=%"PRIu16")",
773 ntohs(udp_key->udp_src), ntohs(udp_key->udp_dst));
774 break;
775
776 case OVS_KEY_ATTR_ICMP:
777 icmp_key = nl_attr_get(a);
778 ds_put_format(ds, "(type=%"PRIu8",code=%"PRIu8")",
779 icmp_key->icmp_type, icmp_key->icmp_code);
780 break;
781
782 case OVS_KEY_ATTR_ICMPV6:
783 icmpv6_key = nl_attr_get(a);
784 ds_put_format(ds, "(type=%"PRIu8",code=%"PRIu8")",
785 icmpv6_key->icmpv6_type, icmpv6_key->icmpv6_code);
786 break;
787
788 case OVS_KEY_ATTR_ARP:
789 arp_key = nl_attr_get(a);
790 ds_put_format(ds, "(sip="IP_FMT",tip="IP_FMT",op=%"PRIu16","
791 "sha="ETH_ADDR_FMT",tha="ETH_ADDR_FMT")",
792 IP_ARGS(&arp_key->arp_sip), IP_ARGS(&arp_key->arp_tip),
793 ntohs(arp_key->arp_op), ETH_ADDR_ARGS(arp_key->arp_sha),
794 ETH_ADDR_ARGS(arp_key->arp_tha));
795 break;
796
797 case OVS_KEY_ATTR_ND: {
798 char target[INET6_ADDRSTRLEN];
799
800 nd_key = nl_attr_get(a);
801 inet_ntop(AF_INET6, nd_key->nd_target, target, sizeof target);
802
803 ds_put_format(ds, "(target=%s", target);
804 if (!eth_addr_is_zero(nd_key->nd_sll)) {
805 ds_put_format(ds, ",sll="ETH_ADDR_FMT,
806 ETH_ADDR_ARGS(nd_key->nd_sll));
807 }
808 if (!eth_addr_is_zero(nd_key->nd_tll)) {
809 ds_put_format(ds, ",tll="ETH_ADDR_FMT,
810 ETH_ADDR_ARGS(nd_key->nd_tll));
811 }
812 ds_put_char(ds, ')');
813 break;
814 }
815
816 case OVS_KEY_ATTR_UNSPEC:
817 case __OVS_KEY_ATTR_MAX:
818 default:
819 format_generic_odp_key(a, ds);
820 break;
821 }
822 }
823
824 /* Appends to 'ds' a string representation of the 'key_len' bytes of
825 * OVS_KEY_ATTR_* attributes in 'key'. */
826 void
827 odp_flow_key_format(const struct nlattr *key, size_t key_len, struct ds *ds)
828 {
829 if (key_len) {
830 const struct nlattr *a;
831 unsigned int left;
832
833 NL_ATTR_FOR_EACH (a, left, key, key_len) {
834 if (a != key) {
835 ds_put_char(ds, ',');
836 }
837 format_odp_key_attr(a, ds);
838 }
839 if (left) {
840 int i;
841
842 if (left == key_len) {
843 ds_put_cstr(ds, "<empty>");
844 }
845 ds_put_format(ds, ",***%u leftover bytes*** (", left);
846 for (i = 0; i < left; i++) {
847 ds_put_format(ds, "%02x", ((const uint8_t *) a)[i]);
848 }
849 ds_put_char(ds, ')');
850 }
851 } else {
852 ds_put_cstr(ds, "<empty>");
853 }
854 }
855
856 static int
857 put_nd_key(int n, const char *nd_target_s,
858 const uint8_t *nd_sll, const uint8_t *nd_tll, struct ofpbuf *key)
859 {
860 struct ovs_key_nd nd_key;
861
862 memset(&nd_key, 0, sizeof nd_key);
863 if (inet_pton(AF_INET6, nd_target_s, nd_key.nd_target) != 1) {
864 return -EINVAL;
865 }
866 if (nd_sll) {
867 memcpy(nd_key.nd_sll, nd_sll, ETH_ADDR_LEN);
868 }
869 if (nd_tll) {
870 memcpy(nd_key.nd_tll, nd_tll, ETH_ADDR_LEN);
871 }
872 nl_msg_put_unspec(key, OVS_KEY_ATTR_ND, &nd_key, sizeof nd_key);
873 return n;
874 }
875
876 static bool
877 ovs_frag_type_from_string(const char *s, enum ovs_frag_type *type)
878 {
879 if (!strcasecmp(s, "no")) {
880 *type = OVS_FRAG_TYPE_NONE;
881 } else if (!strcasecmp(s, "first")) {
882 *type = OVS_FRAG_TYPE_FIRST;
883 } else if (!strcasecmp(s, "later")) {
884 *type = OVS_FRAG_TYPE_LATER;
885 } else {
886 return false;
887 }
888 return true;
889 }
890
891 static int
892 parse_odp_key_attr(const char *s, const struct simap *port_names,
893 struct ofpbuf *key)
894 {
895 /* Many of the sscanf calls in this function use oversized destination
896 * fields because some sscanf() implementations truncate the range of %i
897 * directives, so that e.g. "%"SCNi16 interprets input of "0xfedc" as a
898 * value of 0x7fff. The other alternatives are to allow only a single
899 * radix (e.g. decimal or hexadecimal) or to write more sophisticated
900 * parsers.
901 *
902 * The tun_id parser has to use an alternative approach because there is no
903 * type larger than 64 bits. */
904
905 {
906 unsigned long long int priority;
907 int n = -1;
908
909 if (sscanf(s, "priority(%lli)%n", &priority, &n) > 0 && n > 0) {
910 nl_msg_put_u32(key, OVS_KEY_ATTR_PRIORITY, priority);
911 return n;
912 }
913 }
914
915 {
916 char tun_id_s[32];
917 int n = -1;
918
919 if (sscanf(s, "tun_id(%31[x0123456789abcdefABCDEF])%n",
920 tun_id_s, &n) > 0 && n > 0) {
921 uint64_t tun_id = strtoull(tun_id_s, NULL, 0);
922 nl_msg_put_be64(key, OVS_KEY_ATTR_TUN_ID, htonll(tun_id));
923 return n;
924 }
925 }
926
927 {
928 unsigned long long int in_port;
929 int n = -1;
930
931 if (sscanf(s, "in_port(%lli)%n", &in_port, &n) > 0 && n > 0) {
932 nl_msg_put_u32(key, OVS_KEY_ATTR_IN_PORT, in_port);
933 return n;
934 }
935 }
936
937 if (port_names && !strncmp(s, "in_port(", 8)) {
938 const char *name;
939 const struct simap_node *node;
940 int name_len;
941
942 name = s + 8;
943 name_len = strcspn(s, ")");
944 node = simap_find_len(port_names, name, name_len);
945 if (node) {
946 nl_msg_put_u32(key, OVS_KEY_ATTR_IN_PORT, node->data);
947 return 8 + name_len + 1;
948 }
949 }
950
951 {
952 struct ovs_key_ethernet eth_key;
953 int n = -1;
954
955 if (sscanf(s,
956 "eth(src="ETH_ADDR_SCAN_FMT",dst="ETH_ADDR_SCAN_FMT")%n",
957 ETH_ADDR_SCAN_ARGS(eth_key.eth_src),
958 ETH_ADDR_SCAN_ARGS(eth_key.eth_dst), &n) > 0 && n > 0) {
959 nl_msg_put_unspec(key, OVS_KEY_ATTR_ETHERNET,
960 &eth_key, sizeof eth_key);
961 return n;
962 }
963 }
964
965 {
966 uint16_t vid;
967 int pcp;
968 int cfi;
969 int n = -1;
970
971 if ((sscanf(s, "vlan(vid=%"SCNi16",pcp=%i)%n", &vid, &pcp, &n) > 0
972 && n > 0)) {
973 nl_msg_put_be16(key, OVS_KEY_ATTR_VLAN,
974 htons((vid << VLAN_VID_SHIFT) |
975 (pcp << VLAN_PCP_SHIFT) |
976 VLAN_CFI));
977 return n;
978 } else if ((sscanf(s, "vlan(vid=%"SCNi16",pcp=%i,cfi=%i)%n",
979 &vid, &pcp, &cfi, &n) > 0
980 && n > 0)) {
981 nl_msg_put_be16(key, OVS_KEY_ATTR_VLAN,
982 htons((vid << VLAN_VID_SHIFT) |
983 (pcp << VLAN_PCP_SHIFT) |
984 (cfi ? VLAN_CFI : 0)));
985 return n;
986 }
987 }
988
989 {
990 int eth_type;
991 int n = -1;
992
993 if (sscanf(s, "eth_type(%i)%n", &eth_type, &n) > 0 && n > 0) {
994 nl_msg_put_be16(key, OVS_KEY_ATTR_ETHERTYPE, htons(eth_type));
995 return n;
996 }
997 }
998
999 {
1000 ovs_be32 ipv4_src;
1001 ovs_be32 ipv4_dst;
1002 int ipv4_proto;
1003 int ipv4_tos;
1004 int ipv4_ttl;
1005 char frag[8];
1006 enum ovs_frag_type ipv4_frag;
1007 int n = -1;
1008
1009 if (sscanf(s, "ipv4(src="IP_SCAN_FMT",dst="IP_SCAN_FMT","
1010 "proto=%i,tos=%i,ttl=%i,frag=%7[a-z])%n",
1011 IP_SCAN_ARGS(&ipv4_src), IP_SCAN_ARGS(&ipv4_dst),
1012 &ipv4_proto, &ipv4_tos, &ipv4_ttl, frag, &n) > 0
1013 && n > 0
1014 && ovs_frag_type_from_string(frag, &ipv4_frag)) {
1015 struct ovs_key_ipv4 ipv4_key;
1016
1017 ipv4_key.ipv4_src = ipv4_src;
1018 ipv4_key.ipv4_dst = ipv4_dst;
1019 ipv4_key.ipv4_proto = ipv4_proto;
1020 ipv4_key.ipv4_tos = ipv4_tos;
1021 ipv4_key.ipv4_ttl = ipv4_ttl;
1022 ipv4_key.ipv4_frag = ipv4_frag;
1023 nl_msg_put_unspec(key, OVS_KEY_ATTR_IPV4,
1024 &ipv4_key, sizeof ipv4_key);
1025 return n;
1026 }
1027 }
1028
1029 {
1030 char ipv6_src_s[IPV6_SCAN_LEN + 1];
1031 char ipv6_dst_s[IPV6_SCAN_LEN + 1];
1032 int ipv6_label;
1033 int ipv6_proto;
1034 int ipv6_tclass;
1035 int ipv6_hlimit;
1036 char frag[8];
1037 enum ovs_frag_type ipv6_frag;
1038 int n = -1;
1039
1040 if (sscanf(s, "ipv6(src="IPV6_SCAN_FMT",dst="IPV6_SCAN_FMT","
1041 "label=%i,proto=%i,tclass=%i,hlimit=%i,frag=%7[a-z])%n",
1042 ipv6_src_s, ipv6_dst_s, &ipv6_label,
1043 &ipv6_proto, &ipv6_tclass, &ipv6_hlimit, frag, &n) > 0
1044 && n > 0
1045 && ovs_frag_type_from_string(frag, &ipv6_frag)) {
1046 struct ovs_key_ipv6 ipv6_key;
1047
1048 if (inet_pton(AF_INET6, ipv6_src_s, &ipv6_key.ipv6_src) != 1 ||
1049 inet_pton(AF_INET6, ipv6_dst_s, &ipv6_key.ipv6_dst) != 1) {
1050 return -EINVAL;
1051 }
1052 ipv6_key.ipv6_label = htonl(ipv6_label);
1053 ipv6_key.ipv6_proto = ipv6_proto;
1054 ipv6_key.ipv6_tclass = ipv6_tclass;
1055 ipv6_key.ipv6_hlimit = ipv6_hlimit;
1056 ipv6_key.ipv6_frag = ipv6_frag;
1057 nl_msg_put_unspec(key, OVS_KEY_ATTR_IPV6,
1058 &ipv6_key, sizeof ipv6_key);
1059 return n;
1060 }
1061 }
1062
1063 {
1064 int tcp_src;
1065 int tcp_dst;
1066 int n = -1;
1067
1068 if (sscanf(s, "tcp(src=%i,dst=%i)%n",&tcp_src, &tcp_dst, &n) > 0
1069 && n > 0) {
1070 struct ovs_key_tcp tcp_key;
1071
1072 tcp_key.tcp_src = htons(tcp_src);
1073 tcp_key.tcp_dst = htons(tcp_dst);
1074 nl_msg_put_unspec(key, OVS_KEY_ATTR_TCP, &tcp_key, sizeof tcp_key);
1075 return n;
1076 }
1077 }
1078
1079 {
1080 int udp_src;
1081 int udp_dst;
1082 int n = -1;
1083
1084 if (sscanf(s, "udp(src=%i,dst=%i)%n", &udp_src, &udp_dst, &n) > 0
1085 && n > 0) {
1086 struct ovs_key_udp udp_key;
1087
1088 udp_key.udp_src = htons(udp_src);
1089 udp_key.udp_dst = htons(udp_dst);
1090 nl_msg_put_unspec(key, OVS_KEY_ATTR_UDP, &udp_key, sizeof udp_key);
1091 return n;
1092 }
1093 }
1094
1095 {
1096 int icmp_type;
1097 int icmp_code;
1098 int n = -1;
1099
1100 if (sscanf(s, "icmp(type=%i,code=%i)%n",
1101 &icmp_type, &icmp_code, &n) > 0
1102 && n > 0) {
1103 struct ovs_key_icmp icmp_key;
1104
1105 icmp_key.icmp_type = icmp_type;
1106 icmp_key.icmp_code = icmp_code;
1107 nl_msg_put_unspec(key, OVS_KEY_ATTR_ICMP,
1108 &icmp_key, sizeof icmp_key);
1109 return n;
1110 }
1111 }
1112
1113 {
1114 struct ovs_key_icmpv6 icmpv6_key;
1115 int n = -1;
1116
1117 if (sscanf(s, "icmpv6(type=%"SCNi8",code=%"SCNi8")%n",
1118 &icmpv6_key.icmpv6_type, &icmpv6_key.icmpv6_code,&n) > 0
1119 && n > 0) {
1120 nl_msg_put_unspec(key, OVS_KEY_ATTR_ICMPV6,
1121 &icmpv6_key, sizeof icmpv6_key);
1122 return n;
1123 }
1124 }
1125
1126 {
1127 ovs_be32 arp_sip;
1128 ovs_be32 arp_tip;
1129 int arp_op;
1130 uint8_t arp_sha[ETH_ADDR_LEN];
1131 uint8_t arp_tha[ETH_ADDR_LEN];
1132 int n = -1;
1133
1134 if (sscanf(s, "arp(sip="IP_SCAN_FMT",tip="IP_SCAN_FMT","
1135 "op=%i,sha="ETH_ADDR_SCAN_FMT",tha="ETH_ADDR_SCAN_FMT")%n",
1136 IP_SCAN_ARGS(&arp_sip),
1137 IP_SCAN_ARGS(&arp_tip),
1138 &arp_op,
1139 ETH_ADDR_SCAN_ARGS(arp_sha),
1140 ETH_ADDR_SCAN_ARGS(arp_tha), &n) > 0 && n > 0) {
1141 struct ovs_key_arp arp_key;
1142
1143 memset(&arp_key, 0, sizeof arp_key);
1144 arp_key.arp_sip = arp_sip;
1145 arp_key.arp_tip = arp_tip;
1146 arp_key.arp_op = htons(arp_op);
1147 memcpy(arp_key.arp_sha, arp_sha, ETH_ADDR_LEN);
1148 memcpy(arp_key.arp_tha, arp_tha, ETH_ADDR_LEN);
1149 nl_msg_put_unspec(key, OVS_KEY_ATTR_ARP, &arp_key, sizeof arp_key);
1150 return n;
1151 }
1152 }
1153
1154 {
1155 char nd_target_s[IPV6_SCAN_LEN + 1];
1156 uint8_t nd_sll[ETH_ADDR_LEN];
1157 uint8_t nd_tll[ETH_ADDR_LEN];
1158 int n = -1;
1159
1160 if (sscanf(s, "nd(target="IPV6_SCAN_FMT")%n",
1161 nd_target_s, &n) > 0 && n > 0) {
1162 return put_nd_key(n, nd_target_s, NULL, NULL, key);
1163 }
1164 if (sscanf(s, "nd(target="IPV6_SCAN_FMT",sll="ETH_ADDR_SCAN_FMT")%n",
1165 nd_target_s, ETH_ADDR_SCAN_ARGS(nd_sll), &n) > 0
1166 && n > 0) {
1167 return put_nd_key(n, nd_target_s, nd_sll, NULL, key);
1168 }
1169 if (sscanf(s, "nd(target="IPV6_SCAN_FMT",tll="ETH_ADDR_SCAN_FMT")%n",
1170 nd_target_s, ETH_ADDR_SCAN_ARGS(nd_tll), &n) > 0
1171 && n > 0) {
1172 return put_nd_key(n, nd_target_s, NULL, nd_tll, key);
1173 }
1174 if (sscanf(s, "nd(target="IPV6_SCAN_FMT",sll="ETH_ADDR_SCAN_FMT","
1175 "tll="ETH_ADDR_SCAN_FMT")%n",
1176 nd_target_s, ETH_ADDR_SCAN_ARGS(nd_sll),
1177 ETH_ADDR_SCAN_ARGS(nd_tll), &n) > 0
1178 && n > 0) {
1179 return put_nd_key(n, nd_target_s, nd_sll, nd_tll, key);
1180 }
1181 }
1182
1183 if (!strncmp(s, "encap(", 6)) {
1184 const char *start = s;
1185 size_t encap;
1186
1187 encap = nl_msg_start_nested(key, OVS_KEY_ATTR_ENCAP);
1188
1189 s += 6;
1190 for (;;) {
1191 int retval;
1192
1193 s += strspn(s, ", \t\r\n");
1194 if (!*s) {
1195 return -EINVAL;
1196 } else if (*s == ')') {
1197 break;
1198 }
1199
1200 retval = parse_odp_key_attr(s, port_names, key);
1201 if (retval < 0) {
1202 return retval;
1203 }
1204 s += retval;
1205 }
1206 s++;
1207
1208 nl_msg_end_nested(key, encap);
1209
1210 return s - start;
1211 }
1212
1213 return -EINVAL;
1214 }
1215
1216 /* Parses the string representation of a datapath flow key, in the
1217 * format output by odp_flow_key_format(). Returns 0 if successful,
1218 * otherwise a positive errno value. On success, the flow key is
1219 * appended to 'key' as a series of Netlink attributes. On failure, no
1220 * data is appended to 'key'. Either way, 'key''s data might be
1221 * reallocated.
1222 *
1223 * If 'port_names' is nonnull, it points to an simap that maps from a port name
1224 * to a port number. (Port names may be used instead of port numbers in
1225 * in_port.)
1226 *
1227 * On success, the attributes appended to 'key' are individually syntactically
1228 * valid, but they may not be valid as a sequence. 'key' might, for example,
1229 * have duplicated keys. odp_flow_key_to_flow() will detect those errors. */
1230 int
1231 odp_flow_key_from_string(const char *s, const struct simap *port_names,
1232 struct ofpbuf *key)
1233 {
1234 const size_t old_size = key->size;
1235 for (;;) {
1236 int retval;
1237
1238 s += strspn(s, delimiters);
1239 if (!*s) {
1240 return 0;
1241 }
1242
1243 retval = parse_odp_key_attr(s, port_names, key);
1244 if (retval < 0) {
1245 key->size = old_size;
1246 return -retval;
1247 }
1248 s += retval;
1249 }
1250
1251 return 0;
1252 }
1253
1254 static uint8_t
1255 ovs_to_odp_frag(uint8_t nw_frag)
1256 {
1257 return (nw_frag == 0 ? OVS_FRAG_TYPE_NONE
1258 : nw_frag == FLOW_NW_FRAG_ANY ? OVS_FRAG_TYPE_FIRST
1259 : OVS_FRAG_TYPE_LATER);
1260 }
1261
1262 /* Appends a representation of 'flow' as OVS_KEY_ATTR_* attributes to 'buf'.
1263 *
1264 * 'buf' must have at least ODPUTIL_FLOW_KEY_BYTES bytes of space, or be
1265 * capable of being expanded to allow for that much space. */
1266 void
1267 odp_flow_key_from_flow(struct ofpbuf *buf, const struct flow *flow)
1268 {
1269 struct ovs_key_ethernet *eth_key;
1270 size_t encap;
1271
1272 if (flow->skb_priority) {
1273 nl_msg_put_u32(buf, OVS_KEY_ATTR_PRIORITY, flow->skb_priority);
1274 }
1275
1276 if (flow->tunnel.tun_id != htonll(0)) {
1277 nl_msg_put_be64(buf, OVS_KEY_ATTR_TUN_ID, flow->tunnel.tun_id);
1278 }
1279
1280 if (flow->in_port != OFPP_NONE && flow->in_port != OFPP_CONTROLLER) {
1281 nl_msg_put_u32(buf, OVS_KEY_ATTR_IN_PORT,
1282 ofp_port_to_odp_port(flow->in_port));
1283 }
1284
1285 eth_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ETHERNET,
1286 sizeof *eth_key);
1287 memcpy(eth_key->eth_src, flow->dl_src, ETH_ADDR_LEN);
1288 memcpy(eth_key->eth_dst, flow->dl_dst, ETH_ADDR_LEN);
1289
1290 if (flow->vlan_tci != htons(0) || flow->dl_type == htons(ETH_TYPE_VLAN)) {
1291 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_TYPE_VLAN));
1292 nl_msg_put_be16(buf, OVS_KEY_ATTR_VLAN, flow->vlan_tci);
1293 encap = nl_msg_start_nested(buf, OVS_KEY_ATTR_ENCAP);
1294 if (flow->vlan_tci == htons(0)) {
1295 goto unencap;
1296 }
1297 } else {
1298 encap = 0;
1299 }
1300
1301 if (ntohs(flow->dl_type) < ETH_TYPE_MIN) {
1302 goto unencap;
1303 }
1304
1305 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, flow->dl_type);
1306
1307 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1308 struct ovs_key_ipv4 *ipv4_key;
1309
1310 ipv4_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_IPV4,
1311 sizeof *ipv4_key);
1312 ipv4_key->ipv4_src = flow->nw_src;
1313 ipv4_key->ipv4_dst = flow->nw_dst;
1314 ipv4_key->ipv4_proto = flow->nw_proto;
1315 ipv4_key->ipv4_tos = flow->nw_tos;
1316 ipv4_key->ipv4_ttl = flow->nw_ttl;
1317 ipv4_key->ipv4_frag = ovs_to_odp_frag(flow->nw_frag);
1318 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1319 struct ovs_key_ipv6 *ipv6_key;
1320
1321 ipv6_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_IPV6,
1322 sizeof *ipv6_key);
1323 memcpy(ipv6_key->ipv6_src, &flow->ipv6_src, sizeof ipv6_key->ipv6_src);
1324 memcpy(ipv6_key->ipv6_dst, &flow->ipv6_dst, sizeof ipv6_key->ipv6_dst);
1325 ipv6_key->ipv6_label = flow->ipv6_label;
1326 ipv6_key->ipv6_proto = flow->nw_proto;
1327 ipv6_key->ipv6_tclass = flow->nw_tos;
1328 ipv6_key->ipv6_hlimit = flow->nw_ttl;
1329 ipv6_key->ipv6_frag = ovs_to_odp_frag(flow->nw_frag);
1330 } else if (flow->dl_type == htons(ETH_TYPE_ARP)) {
1331 struct ovs_key_arp *arp_key;
1332
1333 arp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ARP,
1334 sizeof *arp_key);
1335 memset(arp_key, 0, sizeof *arp_key);
1336 arp_key->arp_sip = flow->nw_src;
1337 arp_key->arp_tip = flow->nw_dst;
1338 arp_key->arp_op = htons(flow->nw_proto);
1339 memcpy(arp_key->arp_sha, flow->arp_sha, ETH_ADDR_LEN);
1340 memcpy(arp_key->arp_tha, flow->arp_tha, ETH_ADDR_LEN);
1341 }
1342
1343 if ((flow->dl_type == htons(ETH_TYPE_IP)
1344 || flow->dl_type == htons(ETH_TYPE_IPV6))
1345 && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1346
1347 if (flow->nw_proto == IPPROTO_TCP) {
1348 struct ovs_key_tcp *tcp_key;
1349
1350 tcp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_TCP,
1351 sizeof *tcp_key);
1352 tcp_key->tcp_src = flow->tp_src;
1353 tcp_key->tcp_dst = flow->tp_dst;
1354 } else if (flow->nw_proto == IPPROTO_UDP) {
1355 struct ovs_key_udp *udp_key;
1356
1357 udp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_UDP,
1358 sizeof *udp_key);
1359 udp_key->udp_src = flow->tp_src;
1360 udp_key->udp_dst = flow->tp_dst;
1361 } else if (flow->dl_type == htons(ETH_TYPE_IP)
1362 && flow->nw_proto == IPPROTO_ICMP) {
1363 struct ovs_key_icmp *icmp_key;
1364
1365 icmp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ICMP,
1366 sizeof *icmp_key);
1367 icmp_key->icmp_type = ntohs(flow->tp_src);
1368 icmp_key->icmp_code = ntohs(flow->tp_dst);
1369 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)
1370 && flow->nw_proto == IPPROTO_ICMPV6) {
1371 struct ovs_key_icmpv6 *icmpv6_key;
1372
1373 icmpv6_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ICMPV6,
1374 sizeof *icmpv6_key);
1375 icmpv6_key->icmpv6_type = ntohs(flow->tp_src);
1376 icmpv6_key->icmpv6_code = ntohs(flow->tp_dst);
1377
1378 if (icmpv6_key->icmpv6_type == ND_NEIGHBOR_SOLICIT
1379 || icmpv6_key->icmpv6_type == ND_NEIGHBOR_ADVERT) {
1380 struct ovs_key_nd *nd_key;
1381
1382 nd_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ND,
1383 sizeof *nd_key);
1384 memcpy(nd_key->nd_target, &flow->nd_target,
1385 sizeof nd_key->nd_target);
1386 memcpy(nd_key->nd_sll, flow->arp_sha, ETH_ADDR_LEN);
1387 memcpy(nd_key->nd_tll, flow->arp_tha, ETH_ADDR_LEN);
1388 }
1389 }
1390 }
1391
1392 unencap:
1393 if (encap) {
1394 nl_msg_end_nested(buf, encap);
1395 }
1396 }
1397
1398 uint32_t
1399 odp_flow_key_hash(const struct nlattr *key, size_t key_len)
1400 {
1401 BUILD_ASSERT_DECL(!(NLA_ALIGNTO % sizeof(uint32_t)));
1402 return hash_words((const uint32_t *) key, key_len / sizeof(uint32_t), 0);
1403 }
1404
1405 static void
1406 log_odp_key_attributes(struct vlog_rate_limit *rl, const char *title,
1407 uint64_t attrs, int out_of_range_attr,
1408 const struct nlattr *key, size_t key_len)
1409 {
1410 struct ds s;
1411 int i;
1412
1413 if (VLOG_DROP_DBG(rl)) {
1414 return;
1415 }
1416
1417 ds_init(&s);
1418 for (i = 0; i < 64; i++) {
1419 if (attrs & (UINT64_C(1) << i)) {
1420 ds_put_format(&s, " %s", ovs_key_attr_to_string(i));
1421 }
1422 }
1423 if (out_of_range_attr) {
1424 ds_put_format(&s, " %d (and possibly others)", out_of_range_attr);
1425 }
1426
1427 ds_put_cstr(&s, ": ");
1428 odp_flow_key_format(key, key_len, &s);
1429
1430 VLOG_DBG("%s:%s", title, ds_cstr(&s));
1431 ds_destroy(&s);
1432 }
1433
1434 static bool
1435 odp_to_ovs_frag(uint8_t odp_frag, struct flow *flow)
1436 {
1437 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1438
1439 if (odp_frag > OVS_FRAG_TYPE_LATER) {
1440 VLOG_ERR_RL(&rl, "invalid frag %"PRIu8" in flow key", odp_frag);
1441 return false;
1442 }
1443
1444 if (odp_frag != OVS_FRAG_TYPE_NONE) {
1445 flow->nw_frag |= FLOW_NW_FRAG_ANY;
1446 if (odp_frag == OVS_FRAG_TYPE_LATER) {
1447 flow->nw_frag |= FLOW_NW_FRAG_LATER;
1448 }
1449 }
1450 return true;
1451 }
1452
1453 static bool
1454 parse_flow_nlattrs(const struct nlattr *key, size_t key_len,
1455 const struct nlattr *attrs[], uint64_t *present_attrsp,
1456 int *out_of_range_attrp)
1457 {
1458 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
1459 const struct nlattr *nla;
1460 uint64_t present_attrs;
1461 size_t left;
1462
1463 present_attrs = 0;
1464 *out_of_range_attrp = 0;
1465 NL_ATTR_FOR_EACH (nla, left, key, key_len) {
1466 uint16_t type = nl_attr_type(nla);
1467 size_t len = nl_attr_get_size(nla);
1468 int expected_len = odp_flow_key_attr_len(type);
1469
1470 if (len != expected_len && expected_len >= 0) {
1471 VLOG_ERR_RL(&rl, "attribute %s has length %zu but should have "
1472 "length %d", ovs_key_attr_to_string(type),
1473 len, expected_len);
1474 return false;
1475 }
1476
1477 if (type >= CHAR_BIT * sizeof present_attrs) {
1478 *out_of_range_attrp = type;
1479 } else {
1480 if (present_attrs & (UINT64_C(1) << type)) {
1481 VLOG_ERR_RL(&rl, "duplicate %s attribute in flow key",
1482 ovs_key_attr_to_string(type));
1483 return false;
1484 }
1485
1486 present_attrs |= UINT64_C(1) << type;
1487 attrs[type] = nla;
1488 }
1489 }
1490 if (left) {
1491 VLOG_ERR_RL(&rl, "trailing garbage in flow key");
1492 return false;
1493 }
1494
1495 *present_attrsp = present_attrs;
1496 return true;
1497 }
1498
1499 static enum odp_key_fitness
1500 check_expectations(uint64_t present_attrs, int out_of_range_attr,
1501 uint64_t expected_attrs,
1502 const struct nlattr *key, size_t key_len)
1503 {
1504 uint64_t missing_attrs;
1505 uint64_t extra_attrs;
1506
1507 missing_attrs = expected_attrs & ~present_attrs;
1508 if (missing_attrs) {
1509 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
1510 log_odp_key_attributes(&rl, "expected but not present",
1511 missing_attrs, 0, key, key_len);
1512 return ODP_FIT_TOO_LITTLE;
1513 }
1514
1515 extra_attrs = present_attrs & ~expected_attrs;
1516 if (extra_attrs || out_of_range_attr) {
1517 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
1518 log_odp_key_attributes(&rl, "present but not expected",
1519 extra_attrs, out_of_range_attr, key, key_len);
1520 return ODP_FIT_TOO_MUCH;
1521 }
1522
1523 return ODP_FIT_PERFECT;
1524 }
1525
1526 static bool
1527 parse_ethertype(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
1528 uint64_t present_attrs, uint64_t *expected_attrs,
1529 struct flow *flow)
1530 {
1531 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1532
1533 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE)) {
1534 flow->dl_type = nl_attr_get_be16(attrs[OVS_KEY_ATTR_ETHERTYPE]);
1535 if (ntohs(flow->dl_type) < 1536) {
1536 VLOG_ERR_RL(&rl, "invalid Ethertype %"PRIu16" in flow key",
1537 ntohs(flow->dl_type));
1538 return false;
1539 }
1540 *expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE;
1541 } else {
1542 flow->dl_type = htons(FLOW_DL_TYPE_NONE);
1543 }
1544 return true;
1545 }
1546
1547 static enum odp_key_fitness
1548 parse_l3_onward(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
1549 uint64_t present_attrs, int out_of_range_attr,
1550 uint64_t expected_attrs, struct flow *flow,
1551 const struct nlattr *key, size_t key_len)
1552 {
1553 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1554
1555 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1556 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IPV4;
1557 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IPV4)) {
1558 const struct ovs_key_ipv4 *ipv4_key;
1559
1560 ipv4_key = nl_attr_get(attrs[OVS_KEY_ATTR_IPV4]);
1561 flow->nw_src = ipv4_key->ipv4_src;
1562 flow->nw_dst = ipv4_key->ipv4_dst;
1563 flow->nw_proto = ipv4_key->ipv4_proto;
1564 flow->nw_tos = ipv4_key->ipv4_tos;
1565 flow->nw_ttl = ipv4_key->ipv4_ttl;
1566 if (!odp_to_ovs_frag(ipv4_key->ipv4_frag, flow)) {
1567 return ODP_FIT_ERROR;
1568 }
1569 }
1570 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1571 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IPV6;
1572 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IPV6)) {
1573 const struct ovs_key_ipv6 *ipv6_key;
1574
1575 ipv6_key = nl_attr_get(attrs[OVS_KEY_ATTR_IPV6]);
1576 memcpy(&flow->ipv6_src, ipv6_key->ipv6_src, sizeof flow->ipv6_src);
1577 memcpy(&flow->ipv6_dst, ipv6_key->ipv6_dst, sizeof flow->ipv6_dst);
1578 flow->ipv6_label = ipv6_key->ipv6_label;
1579 flow->nw_proto = ipv6_key->ipv6_proto;
1580 flow->nw_tos = ipv6_key->ipv6_tclass;
1581 flow->nw_ttl = ipv6_key->ipv6_hlimit;
1582 if (!odp_to_ovs_frag(ipv6_key->ipv6_frag, flow)) {
1583 return ODP_FIT_ERROR;
1584 }
1585 }
1586 } else if (flow->dl_type == htons(ETH_TYPE_ARP)) {
1587 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ARP;
1588 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ARP)) {
1589 const struct ovs_key_arp *arp_key;
1590
1591 arp_key = nl_attr_get(attrs[OVS_KEY_ATTR_ARP]);
1592 flow->nw_src = arp_key->arp_sip;
1593 flow->nw_dst = arp_key->arp_tip;
1594 if (arp_key->arp_op & htons(0xff00)) {
1595 VLOG_ERR_RL(&rl, "unsupported ARP opcode %"PRIu16" in flow "
1596 "key", ntohs(arp_key->arp_op));
1597 return ODP_FIT_ERROR;
1598 }
1599 flow->nw_proto = ntohs(arp_key->arp_op);
1600 memcpy(flow->arp_sha, arp_key->arp_sha, ETH_ADDR_LEN);
1601 memcpy(flow->arp_tha, arp_key->arp_tha, ETH_ADDR_LEN);
1602 }
1603 }
1604
1605 if (flow->nw_proto == IPPROTO_TCP
1606 && (flow->dl_type == htons(ETH_TYPE_IP) ||
1607 flow->dl_type == htons(ETH_TYPE_IPV6))
1608 && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1609 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_TCP;
1610 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_TCP)) {
1611 const struct ovs_key_tcp *tcp_key;
1612
1613 tcp_key = nl_attr_get(attrs[OVS_KEY_ATTR_TCP]);
1614 flow->tp_src = tcp_key->tcp_src;
1615 flow->tp_dst = tcp_key->tcp_dst;
1616 }
1617 } else if (flow->nw_proto == IPPROTO_UDP
1618 && (flow->dl_type == htons(ETH_TYPE_IP) ||
1619 flow->dl_type == htons(ETH_TYPE_IPV6))
1620 && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1621 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_UDP;
1622 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_UDP)) {
1623 const struct ovs_key_udp *udp_key;
1624
1625 udp_key = nl_attr_get(attrs[OVS_KEY_ATTR_UDP]);
1626 flow->tp_src = udp_key->udp_src;
1627 flow->tp_dst = udp_key->udp_dst;
1628 }
1629 } else if (flow->nw_proto == IPPROTO_ICMP
1630 && flow->dl_type == htons(ETH_TYPE_IP)
1631 && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1632 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ICMP;
1633 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ICMP)) {
1634 const struct ovs_key_icmp *icmp_key;
1635
1636 icmp_key = nl_attr_get(attrs[OVS_KEY_ATTR_ICMP]);
1637 flow->tp_src = htons(icmp_key->icmp_type);
1638 flow->tp_dst = htons(icmp_key->icmp_code);
1639 }
1640 } else if (flow->nw_proto == IPPROTO_ICMPV6
1641 && flow->dl_type == htons(ETH_TYPE_IPV6)
1642 && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
1643 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ICMPV6;
1644 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ICMPV6)) {
1645 const struct ovs_key_icmpv6 *icmpv6_key;
1646
1647 icmpv6_key = nl_attr_get(attrs[OVS_KEY_ATTR_ICMPV6]);
1648 flow->tp_src = htons(icmpv6_key->icmpv6_type);
1649 flow->tp_dst = htons(icmpv6_key->icmpv6_code);
1650
1651 if (flow->tp_src == htons(ND_NEIGHBOR_SOLICIT) ||
1652 flow->tp_src == htons(ND_NEIGHBOR_ADVERT)) {
1653 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ND;
1654 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ND)) {
1655 const struct ovs_key_nd *nd_key;
1656
1657 nd_key = nl_attr_get(attrs[OVS_KEY_ATTR_ND]);
1658 memcpy(&flow->nd_target, nd_key->nd_target,
1659 sizeof flow->nd_target);
1660 memcpy(flow->arp_sha, nd_key->nd_sll, ETH_ADDR_LEN);
1661 memcpy(flow->arp_tha, nd_key->nd_tll, ETH_ADDR_LEN);
1662 }
1663 }
1664 }
1665 }
1666
1667 return check_expectations(present_attrs, out_of_range_attr, expected_attrs,
1668 key, key_len);
1669 }
1670
1671 /* Parse 802.1Q header then encapsulated L3 attributes. */
1672 static enum odp_key_fitness
1673 parse_8021q_onward(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
1674 uint64_t present_attrs, int out_of_range_attr,
1675 uint64_t expected_attrs, struct flow *flow,
1676 const struct nlattr *key, size_t key_len)
1677 {
1678 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1679
1680 const struct nlattr *encap
1681 = (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ENCAP)
1682 ? attrs[OVS_KEY_ATTR_ENCAP] : NULL);
1683 enum odp_key_fitness encap_fitness;
1684 enum odp_key_fitness fitness;
1685 ovs_be16 tci;
1686
1687 /* Calulate fitness of outer attributes. */
1688 expected_attrs |= ((UINT64_C(1) << OVS_KEY_ATTR_VLAN) |
1689 (UINT64_C(1) << OVS_KEY_ATTR_ENCAP));
1690 fitness = check_expectations(present_attrs, out_of_range_attr,
1691 expected_attrs, key, key_len);
1692
1693 /* Get the VLAN TCI value. */
1694 if (!(present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN))) {
1695 return ODP_FIT_TOO_LITTLE;
1696 }
1697 tci = nl_attr_get_be16(attrs[OVS_KEY_ATTR_VLAN]);
1698 if (tci == htons(0)) {
1699 /* Corner case for a truncated 802.1Q header. */
1700 if (fitness == ODP_FIT_PERFECT && nl_attr_get_size(encap)) {
1701 return ODP_FIT_TOO_MUCH;
1702 }
1703 return fitness;
1704 } else if (!(tci & htons(VLAN_CFI))) {
1705 VLOG_ERR_RL(&rl, "OVS_KEY_ATTR_VLAN 0x%04"PRIx16" is nonzero "
1706 "but CFI bit is not set", ntohs(tci));
1707 return ODP_FIT_ERROR;
1708 }
1709
1710 /* Set vlan_tci.
1711 * Remove the TPID from dl_type since it's not the real Ethertype. */
1712 flow->vlan_tci = tci;
1713 flow->dl_type = htons(0);
1714
1715 /* Now parse the encapsulated attributes. */
1716 if (!parse_flow_nlattrs(nl_attr_get(encap), nl_attr_get_size(encap),
1717 attrs, &present_attrs, &out_of_range_attr)) {
1718 return ODP_FIT_ERROR;
1719 }
1720 expected_attrs = 0;
1721
1722 if (!parse_ethertype(attrs, present_attrs, &expected_attrs, flow)) {
1723 return ODP_FIT_ERROR;
1724 }
1725 encap_fitness = parse_l3_onward(attrs, present_attrs, out_of_range_attr,
1726 expected_attrs, flow, key, key_len);
1727
1728 /* The overall fitness is the worse of the outer and inner attributes. */
1729 return MAX(fitness, encap_fitness);
1730 }
1731
1732 /* Converts the 'key_len' bytes of OVS_KEY_ATTR_* attributes in 'key' to a flow
1733 * structure in 'flow'. Returns an ODP_FIT_* value that indicates how well
1734 * 'key' fits our expectations for what a flow key should contain.
1735 *
1736 * This function doesn't take the packet itself as an argument because none of
1737 * the currently understood OVS_KEY_ATTR_* attributes require it. Currently,
1738 * it is always possible to infer which additional attribute(s) should appear
1739 * by looking at the attributes for lower-level protocols, e.g. if the network
1740 * protocol in OVS_KEY_ATTR_IPV4 or OVS_KEY_ATTR_IPV6 is IPPROTO_TCP then we
1741 * know that a OVS_KEY_ATTR_TCP attribute must appear and that otherwise it
1742 * must be absent. */
1743 enum odp_key_fitness
1744 odp_flow_key_to_flow(const struct nlattr *key, size_t key_len,
1745 struct flow *flow)
1746 {
1747 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1748 const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1];
1749 uint64_t expected_attrs;
1750 uint64_t present_attrs;
1751 int out_of_range_attr;
1752
1753 memset(flow, 0, sizeof *flow);
1754
1755 /* Parse attributes. */
1756 if (!parse_flow_nlattrs(key, key_len, attrs, &present_attrs,
1757 &out_of_range_attr)) {
1758 return ODP_FIT_ERROR;
1759 }
1760 expected_attrs = 0;
1761
1762 /* Metadata. */
1763 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_PRIORITY)) {
1764 flow->skb_priority = nl_attr_get_u32(attrs[OVS_KEY_ATTR_PRIORITY]);
1765 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_PRIORITY;
1766 }
1767
1768 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_TUN_ID)) {
1769 flow->tunnel.tun_id = nl_attr_get_be64(attrs[OVS_KEY_ATTR_TUN_ID]);
1770 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_TUN_ID;
1771 }
1772
1773 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IN_PORT)) {
1774 uint32_t in_port = nl_attr_get_u32(attrs[OVS_KEY_ATTR_IN_PORT]);
1775 if (in_port >= UINT16_MAX || in_port >= OFPP_MAX) {
1776 VLOG_ERR_RL(&rl, "in_port %"PRIu32" out of supported range",
1777 in_port);
1778 return ODP_FIT_ERROR;
1779 }
1780 flow->in_port = odp_port_to_ofp_port(in_port);
1781 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IN_PORT;
1782 } else {
1783 flow->in_port = OFPP_NONE;
1784 }
1785
1786 /* Ethernet header. */
1787 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ETHERNET)) {
1788 const struct ovs_key_ethernet *eth_key;
1789
1790 eth_key = nl_attr_get(attrs[OVS_KEY_ATTR_ETHERNET]);
1791 memcpy(flow->dl_src, eth_key->eth_src, ETH_ADDR_LEN);
1792 memcpy(flow->dl_dst, eth_key->eth_dst, ETH_ADDR_LEN);
1793 }
1794 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ETHERNET;
1795
1796 /* Get Ethertype or 802.1Q TPID or FLOW_DL_TYPE_NONE. */
1797 if (!parse_ethertype(attrs, present_attrs, &expected_attrs, flow)) {
1798 return ODP_FIT_ERROR;
1799 }
1800
1801 if (flow->dl_type == htons(ETH_TYPE_VLAN)) {
1802 return parse_8021q_onward(attrs, present_attrs, out_of_range_attr,
1803 expected_attrs, flow, key, key_len);
1804 }
1805 return parse_l3_onward(attrs, present_attrs, out_of_range_attr,
1806 expected_attrs, flow, key, key_len);
1807 }
1808
1809 /* Returns 'fitness' as a string, for use in debug messages. */
1810 const char *
1811 odp_key_fitness_to_string(enum odp_key_fitness fitness)
1812 {
1813 switch (fitness) {
1814 case ODP_FIT_PERFECT:
1815 return "OK";
1816 case ODP_FIT_TOO_MUCH:
1817 return "too_much";
1818 case ODP_FIT_TOO_LITTLE:
1819 return "too_little";
1820 case ODP_FIT_ERROR:
1821 return "error";
1822 default:
1823 return "<unknown>";
1824 }
1825 }
1826
1827 /* Appends an OVS_ACTION_ATTR_USERSPACE action to 'odp_actions' that specifies
1828 * Netlink PID 'pid'. If 'cookie' is nonnull, adds a userdata attribute whose
1829 * contents contains 'cookie' and returns the offset within 'odp_actions' of
1830 * the start of the cookie. (If 'cookie' is null, then the return value is not
1831 * meaningful.) */
1832 size_t
1833 odp_put_userspace_action(uint32_t pid, const union user_action_cookie *cookie,
1834 struct ofpbuf *odp_actions)
1835 {
1836 size_t offset;
1837
1838 offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_USERSPACE);
1839 nl_msg_put_u32(odp_actions, OVS_USERSPACE_ATTR_PID, pid);
1840 if (cookie) {
1841 nl_msg_put_unspec(odp_actions, OVS_USERSPACE_ATTR_USERDATA,
1842 cookie, sizeof *cookie);
1843 }
1844 nl_msg_end_nested(odp_actions, offset);
1845
1846 return cookie ? odp_actions->size - NLA_ALIGN(sizeof *cookie) : 0;
1847 }
1848 \f
1849 /* The commit_odp_actions() function and its helpers. */
1850
1851 static void
1852 commit_set_action(struct ofpbuf *odp_actions, enum ovs_key_attr key_type,
1853 const void *key, size_t key_size)
1854 {
1855 size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
1856 nl_msg_put_unspec(odp_actions, key_type, key, key_size);
1857 nl_msg_end_nested(odp_actions, offset);
1858 }
1859
1860 static void
1861 commit_set_tun_id_action(const struct flow *flow, struct flow *base,
1862 struct ofpbuf *odp_actions)
1863 {
1864 if (base->tunnel.tun_id == flow->tunnel.tun_id) {
1865 return;
1866 }
1867 base->tunnel.tun_id = flow->tunnel.tun_id;
1868
1869 commit_set_action(odp_actions, OVS_KEY_ATTR_TUN_ID,
1870 &base->tunnel.tun_id, sizeof(base->tunnel.tun_id));
1871 }
1872
1873 static void
1874 commit_set_ether_addr_action(const struct flow *flow, struct flow *base,
1875 struct ofpbuf *odp_actions)
1876 {
1877 struct ovs_key_ethernet eth_key;
1878
1879 if (eth_addr_equals(base->dl_src, flow->dl_src) &&
1880 eth_addr_equals(base->dl_dst, flow->dl_dst)) {
1881 return;
1882 }
1883
1884 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
1885 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
1886
1887 memcpy(eth_key.eth_src, base->dl_src, ETH_ADDR_LEN);
1888 memcpy(eth_key.eth_dst, base->dl_dst, ETH_ADDR_LEN);
1889
1890 commit_set_action(odp_actions, OVS_KEY_ATTR_ETHERNET,
1891 &eth_key, sizeof(eth_key));
1892 }
1893
1894 static void
1895 commit_vlan_action(const struct flow *flow, struct flow *base,
1896 struct ofpbuf *odp_actions)
1897 {
1898 if (base->vlan_tci == flow->vlan_tci) {
1899 return;
1900 }
1901
1902 if (base->vlan_tci & htons(VLAN_CFI)) {
1903 nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_VLAN);
1904 }
1905
1906 if (flow->vlan_tci & htons(VLAN_CFI)) {
1907 struct ovs_action_push_vlan vlan;
1908
1909 vlan.vlan_tpid = htons(ETH_TYPE_VLAN);
1910 vlan.vlan_tci = flow->vlan_tci;
1911 nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_PUSH_VLAN,
1912 &vlan, sizeof vlan);
1913 }
1914 base->vlan_tci = flow->vlan_tci;
1915 }
1916
1917 static void
1918 commit_set_ipv4_action(const struct flow *flow, struct flow *base,
1919 struct ofpbuf *odp_actions)
1920 {
1921 struct ovs_key_ipv4 ipv4_key;
1922
1923 if (base->nw_src == flow->nw_src &&
1924 base->nw_dst == flow->nw_dst &&
1925 base->nw_tos == flow->nw_tos &&
1926 base->nw_ttl == flow->nw_ttl &&
1927 base->nw_frag == flow->nw_frag) {
1928 return;
1929 }
1930
1931 ipv4_key.ipv4_src = base->nw_src = flow->nw_src;
1932 ipv4_key.ipv4_dst = base->nw_dst = flow->nw_dst;
1933 ipv4_key.ipv4_tos = base->nw_tos = flow->nw_tos;
1934 ipv4_key.ipv4_ttl = base->nw_ttl = flow->nw_ttl;
1935 ipv4_key.ipv4_proto = base->nw_proto;
1936 ipv4_key.ipv4_frag = ovs_to_odp_frag(base->nw_frag);
1937
1938 commit_set_action(odp_actions, OVS_KEY_ATTR_IPV4,
1939 &ipv4_key, sizeof(ipv4_key));
1940 }
1941
1942 static void
1943 commit_set_ipv6_action(const struct flow *flow, struct flow *base,
1944 struct ofpbuf *odp_actions)
1945 {
1946 struct ovs_key_ipv6 ipv6_key;
1947
1948 if (ipv6_addr_equals(&base->ipv6_src, &flow->ipv6_src) &&
1949 ipv6_addr_equals(&base->ipv6_dst, &flow->ipv6_dst) &&
1950 base->ipv6_label == flow->ipv6_label &&
1951 base->nw_tos == flow->nw_tos &&
1952 base->nw_ttl == flow->nw_ttl &&
1953 base->nw_frag == flow->nw_frag) {
1954 return;
1955 }
1956
1957 base->ipv6_src = flow->ipv6_src;
1958 memcpy(&ipv6_key.ipv6_src, &base->ipv6_src, sizeof(ipv6_key.ipv6_src));
1959 base->ipv6_dst = flow->ipv6_dst;
1960 memcpy(&ipv6_key.ipv6_dst, &base->ipv6_dst, sizeof(ipv6_key.ipv6_dst));
1961
1962 ipv6_key.ipv6_label = base->ipv6_label = flow->ipv6_label;
1963 ipv6_key.ipv6_tclass = base->nw_tos = flow->nw_tos;
1964 ipv6_key.ipv6_hlimit = base->nw_ttl = flow->nw_ttl;
1965 ipv6_key.ipv6_proto = base->nw_proto;
1966 ipv6_key.ipv6_frag = ovs_to_odp_frag(base->nw_frag);
1967
1968 commit_set_action(odp_actions, OVS_KEY_ATTR_IPV6,
1969 &ipv6_key, sizeof(ipv6_key));
1970 }
1971
1972 static void
1973 commit_set_nw_action(const struct flow *flow, struct flow *base,
1974 struct ofpbuf *odp_actions)
1975 {
1976 /* Check if flow really have an IP header. */
1977 if (!flow->nw_proto) {
1978 return;
1979 }
1980
1981 if (base->dl_type == htons(ETH_TYPE_IP)) {
1982 commit_set_ipv4_action(flow, base, odp_actions);
1983 } else if (base->dl_type == htons(ETH_TYPE_IPV6)) {
1984 commit_set_ipv6_action(flow, base, odp_actions);
1985 }
1986 }
1987
1988 static void
1989 commit_set_port_action(const struct flow *flow, struct flow *base,
1990 struct ofpbuf *odp_actions)
1991 {
1992 if (!base->tp_src && !base->tp_dst) {
1993 return;
1994 }
1995
1996 if (base->tp_src == flow->tp_src &&
1997 base->tp_dst == flow->tp_dst) {
1998 return;
1999 }
2000
2001 if (flow->nw_proto == IPPROTO_TCP) {
2002 struct ovs_key_tcp port_key;
2003
2004 port_key.tcp_src = base->tp_src = flow->tp_src;
2005 port_key.tcp_dst = base->tp_dst = flow->tp_dst;
2006
2007 commit_set_action(odp_actions, OVS_KEY_ATTR_TCP,
2008 &port_key, sizeof(port_key));
2009
2010 } else if (flow->nw_proto == IPPROTO_UDP) {
2011 struct ovs_key_udp port_key;
2012
2013 port_key.udp_src = base->tp_src = flow->tp_src;
2014 port_key.udp_dst = base->tp_dst = flow->tp_dst;
2015
2016 commit_set_action(odp_actions, OVS_KEY_ATTR_UDP,
2017 &port_key, sizeof(port_key));
2018 }
2019 }
2020
2021 static void
2022 commit_set_priority_action(const struct flow *flow, struct flow *base,
2023 struct ofpbuf *odp_actions)
2024 {
2025 if (base->skb_priority == flow->skb_priority) {
2026 return;
2027 }
2028 base->skb_priority = flow->skb_priority;
2029
2030 commit_set_action(odp_actions, OVS_KEY_ATTR_PRIORITY,
2031 &base->skb_priority, sizeof(base->skb_priority));
2032 }
2033
2034 /* If any of the flow key data that ODP actions can modify are different in
2035 * 'base' and 'flow', appends ODP actions to 'odp_actions' that change the flow
2036 * key from 'base' into 'flow', and then changes 'base' the same way. */
2037 void
2038 commit_odp_actions(const struct flow *flow, struct flow *base,
2039 struct ofpbuf *odp_actions)
2040 {
2041 commit_set_tun_id_action(flow, base, odp_actions);
2042 commit_set_ether_addr_action(flow, base, odp_actions);
2043 commit_vlan_action(flow, base, odp_actions);
2044 commit_set_nw_action(flow, base, odp_actions);
2045 commit_set_port_action(flow, base, odp_actions);
2046 commit_set_priority_action(flow, base, odp_actions);
2047 }