]> git.proxmox.com Git - mirror_ovs.git/blob - lib/odp-util.c
projects / mirror_ovs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
tunneling: Add userspace tunnel support for Geneve.
[mirror_ovs.git] / lib / odp-util.c
1 /*
2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 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
28 #include "byte-order.h"
29 #include "coverage.h"
30 #include "dpif.h"
31 #include "dynamic-string.h"
32 #include "flow.h"
33 #include "netlink.h"
34 #include "ofpbuf.h"
35 #include "packets.h"
36 #include "simap.h"
37 #include "timeval.h"
38 #include "unaligned.h"
39 #include "util.h"
40 #include "openvswitch/vlog.h"
41
42 VLOG_DEFINE_THIS_MODULE(odp_util);
43
44 /* The interface between userspace and kernel uses an "OVS_*" prefix.
45 * Since this is fairly non-specific for the OVS userspace components,
46 * "ODP_*" (Open vSwitch Datapath) is used as the prefix for
47 * interactions with the datapath.
48 */
49
50 /* The set of characters that may separate one action or one key attribute
51 * from another. */
52 static const char *delimiters = ", \t\r\n";
53
54 static int parse_odp_key_mask_attr(const char *, const struct simap *port_names,
55 struct ofpbuf *, struct ofpbuf *);
56 static void format_odp_key_attr(const struct nlattr *a,
57 const struct nlattr *ma,
58 const struct hmap *portno_names, struct ds *ds,
59 bool verbose);
60
61 /* Returns one the following for the action with the given OVS_ACTION_ATTR_*
62 * 'type':
63 *
64 * - For an action whose argument has a fixed length, returned that
65 * nonnegative length in bytes.
66 *
67 * - For an action with a variable-length argument, returns -2.
68 *
69 * - For an invalid 'type', returns -1. */
70 static int
71 odp_action_len(uint16_t type)
72 {
73 if (type > OVS_ACTION_ATTR_MAX) {
74 return -1;
75 }
76
77 switch ((enum ovs_action_attr) type) {
78 case OVS_ACTION_ATTR_OUTPUT: return sizeof(uint32_t);
79 case OVS_ACTION_ATTR_TUNNEL_PUSH: return -2;
80 case OVS_ACTION_ATTR_TUNNEL_POP: return sizeof(uint32_t);
81 case OVS_ACTION_ATTR_USERSPACE: return -2;
82 case OVS_ACTION_ATTR_PUSH_VLAN: return sizeof(struct ovs_action_push_vlan);
83 case OVS_ACTION_ATTR_POP_VLAN: return 0;
84 case OVS_ACTION_ATTR_PUSH_MPLS: return sizeof(struct ovs_action_push_mpls);
85 case OVS_ACTION_ATTR_POP_MPLS: return sizeof(ovs_be16);
86 case OVS_ACTION_ATTR_RECIRC: return sizeof(uint32_t);
87 case OVS_ACTION_ATTR_HASH: return sizeof(struct ovs_action_hash);
88 case OVS_ACTION_ATTR_SET: return -2;
89 case OVS_ACTION_ATTR_SET_MASKED: return -2;
90 case OVS_ACTION_ATTR_SAMPLE: return -2;
91
92 case OVS_ACTION_ATTR_UNSPEC:
93 case __OVS_ACTION_ATTR_MAX:
94 return -1;
95 }
96
97 return -1;
98 }
99
100 /* Returns a string form of 'attr'. The return value is either a statically
101 * allocated constant string or the 'bufsize'-byte buffer 'namebuf'. 'bufsize'
102 * should be at least OVS_KEY_ATTR_BUFSIZE. */
103 enum { OVS_KEY_ATTR_BUFSIZE = 3 + INT_STRLEN(unsigned int) + 1 };
104 static const char *
105 ovs_key_attr_to_string(enum ovs_key_attr attr, char *namebuf, size_t bufsize)
106 {
107 switch (attr) {
108 case OVS_KEY_ATTR_UNSPEC: return "unspec";
109 case OVS_KEY_ATTR_ENCAP: return "encap";
110 case OVS_KEY_ATTR_PRIORITY: return "skb_priority";
111 case OVS_KEY_ATTR_SKB_MARK: return "skb_mark";
112 case OVS_KEY_ATTR_TUNNEL: return "tunnel";
113 case OVS_KEY_ATTR_IN_PORT: return "in_port";
114 case OVS_KEY_ATTR_ETHERNET: return "eth";
115 case OVS_KEY_ATTR_VLAN: return "vlan";
116 case OVS_KEY_ATTR_ETHERTYPE: return "eth_type";
117 case OVS_KEY_ATTR_IPV4: return "ipv4";
118 case OVS_KEY_ATTR_IPV6: return "ipv6";
119 case OVS_KEY_ATTR_TCP: return "tcp";
120 case OVS_KEY_ATTR_TCP_FLAGS: return "tcp_flags";
121 case OVS_KEY_ATTR_UDP: return "udp";
122 case OVS_KEY_ATTR_SCTP: return "sctp";
123 case OVS_KEY_ATTR_ICMP: return "icmp";
124 case OVS_KEY_ATTR_ICMPV6: return "icmpv6";
125 case OVS_KEY_ATTR_ARP: return "arp";
126 case OVS_KEY_ATTR_ND: return "nd";
127 case OVS_KEY_ATTR_MPLS: return "mpls";
128 case OVS_KEY_ATTR_DP_HASH: return "dp_hash";
129 case OVS_KEY_ATTR_RECIRC_ID: return "recirc_id";
130
131 case __OVS_KEY_ATTR_MAX:
132 default:
133 snprintf(namebuf, bufsize, "key%u", (unsigned int) attr);
134 return namebuf;
135 }
136 }
137
138 static void
139 format_generic_odp_action(struct ds *ds, const struct nlattr *a)
140 {
141 size_t len = nl_attr_get_size(a);
142
143 ds_put_format(ds, "action%"PRId16, nl_attr_type(a));
144 if (len) {
145 const uint8_t *unspec;
146 unsigned int i;
147
148 unspec = nl_attr_get(a);
149 for (i = 0; i < len; i++) {
150 ds_put_char(ds, i ? ' ': '(');
151 ds_put_format(ds, "%02x", unspec[i]);
152 }
153 ds_put_char(ds, ')');
154 }
155 }
156
157 static void
158 format_odp_sample_action(struct ds *ds, const struct nlattr *attr)
159 {
160 static const struct nl_policy ovs_sample_policy[] = {
161 [OVS_SAMPLE_ATTR_PROBABILITY] = { .type = NL_A_U32 },
162 [OVS_SAMPLE_ATTR_ACTIONS] = { .type = NL_A_NESTED }
163 };
164 struct nlattr *a[ARRAY_SIZE(ovs_sample_policy)];
165 double percentage;
166 const struct nlattr *nla_acts;
167 int len;
168
169 ds_put_cstr(ds, "sample");
170
171 if (!nl_parse_nested(attr, ovs_sample_policy, a, ARRAY_SIZE(a))) {
172 ds_put_cstr(ds, "(error)");
173 return;
174 }
175
176 percentage = (100.0 * nl_attr_get_u32(a[OVS_SAMPLE_ATTR_PROBABILITY])) /
177 UINT32_MAX;
178
179 ds_put_format(ds, "(sample=%.1f%%,", percentage);
180
181 ds_put_cstr(ds, "actions(");
182 nla_acts = nl_attr_get(a[OVS_SAMPLE_ATTR_ACTIONS]);
183 len = nl_attr_get_size(a[OVS_SAMPLE_ATTR_ACTIONS]);
184 format_odp_actions(ds, nla_acts, len);
185 ds_put_format(ds, "))");
186 }
187
188 static const char *
189 slow_path_reason_to_string(uint32_t reason)
190 {
191 switch ((enum slow_path_reason) reason) {
192 #define SPR(ENUM, STRING, EXPLANATION) case ENUM: return STRING;
193 SLOW_PATH_REASONS
194 #undef SPR
195 }
196
197 return NULL;
198 }
199
200 const char *
201 slow_path_reason_to_explanation(enum slow_path_reason reason)
202 {
203 switch (reason) {
204 #define SPR(ENUM, STRING, EXPLANATION) case ENUM: return EXPLANATION;
205 SLOW_PATH_REASONS
206 #undef SPR
207 }
208
209 return "<unknown>";
210 }
211
212 static int
213 parse_flags(const char *s, const char *(*bit_to_string)(uint32_t),
214 uint32_t *res_flags, uint32_t allowed, uint32_t *res_mask)
215 {
216 uint32_t result = 0;
217 int n;
218
219 /* Parse masked flags in numeric format? */
220 if (res_mask && ovs_scan(s, "%"SCNi32"/%"SCNi32"%n",
221 res_flags, res_mask, &n) && n > 0) {
222 if (*res_flags & ~allowed || *res_mask & ~allowed) {
223 return -EINVAL;
224 }
225 return n;
226 }
227
228 n = 0;
229
230 if (res_mask && (*s == '+' || *s == '-')) {
231 uint32_t flags = 0, mask = 0;
232
233 /* Parse masked flags. */
234 while (s[n] != ')') {
235 bool set;
236 uint32_t bit;
237 int name_len;
238
239 if (s[n] == '+') {
240 set = true;
241 } else if (s[n] == '-') {
242 set = false;
243 } else {
244 return -EINVAL;
245 }
246 n++;
247
248 name_len = strcspn(s + n, "+-)");
249
250 for (bit = 1; bit; bit <<= 1) {
251 const char *fname = bit_to_string(bit);
252 size_t len;
253
254 if (!fname) {
255 continue;
256 }
257
258 len = strlen(fname);
259 if (len != name_len) {
260 continue;
261 }
262 if (!strncmp(s + n, fname, len)) {
263 if (mask & bit) {
264 /* bit already set. */
265 return -EINVAL;
266 }
267 if (!(bit & allowed)) {
268 return -EINVAL;
269 }
270 if (set) {
271 flags |= bit;
272 }
273 mask |= bit;
274 break;
275 }
276 }
277
278 if (!bit) {
279 return -EINVAL; /* Unknown flag name */
280 }
281 s += name_len;
282 }
283
284 *res_flags = flags;
285 *res_mask = mask;
286 return n;
287 }
288
289 /* Parse unmasked flags. If a flag is present, it is set, otherwise
290 * it is not set. */
291 while (s[n] != ')') {
292 unsigned long long int flags;
293 uint32_t bit;
294 int n0;
295
296 if (ovs_scan(&s[n], "%lli%n", &flags, &n0)) {
297 if (flags & ~allowed) {
298 return -EINVAL;
299 }
300 n += n0 + (s[n + n0] == ',');
301 result |= flags;
302 continue;
303 }
304
305 for (bit = 1; bit; bit <<= 1) {
306 const char *name = bit_to_string(bit);
307 size_t len;
308
309 if (!name) {
310 continue;
311 }
312
313 len = strlen(name);
314 if (!strncmp(s + n, name, len) &&
315 (s[n + len] == ',' || s[n + len] == ')')) {
316 if (!(bit & allowed)) {
317 return -EINVAL;
318 }
319 result |= bit;
320 n += len + (s[n + len] == ',');
321 break;
322 }
323 }
324
325 if (!bit) {
326 return -EINVAL;
327 }
328 }
329
330 *res_flags = result;
331 if (res_mask) {
332 *res_mask = UINT32_MAX;
333 }
334 return n;
335 }
336
337 static void
338 format_odp_userspace_action(struct ds *ds, const struct nlattr *attr)
339 {
340 static const struct nl_policy ovs_userspace_policy[] = {
341 [OVS_USERSPACE_ATTR_PID] = { .type = NL_A_U32 },
342 [OVS_USERSPACE_ATTR_USERDATA] = { .type = NL_A_UNSPEC,
343 .optional = true },
344 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = { .type = NL_A_U32,
345 .optional = true },
346 };
347 struct nlattr *a[ARRAY_SIZE(ovs_userspace_policy)];
348 const struct nlattr *userdata_attr;
349 const struct nlattr *tunnel_out_port_attr;
350
351 if (!nl_parse_nested(attr, ovs_userspace_policy, a, ARRAY_SIZE(a))) {
352 ds_put_cstr(ds, "userspace(error)");
353 return;
354 }
355
356 ds_put_format(ds, "userspace(pid=%"PRIu32,
357 nl_attr_get_u32(a[OVS_USERSPACE_ATTR_PID]));
358
359 userdata_attr = a[OVS_USERSPACE_ATTR_USERDATA];
360
361 if (userdata_attr) {
362 const uint8_t *userdata = nl_attr_get(userdata_attr);
363 size_t userdata_len = nl_attr_get_size(userdata_attr);
364 bool userdata_unspec = true;
365 union user_action_cookie cookie;
366
367 if (userdata_len >= sizeof cookie.type
368 && userdata_len <= sizeof cookie) {
369
370 memset(&cookie, 0, sizeof cookie);
371 memcpy(&cookie, userdata, userdata_len);
372
373 userdata_unspec = false;
374
375 if (userdata_len == sizeof cookie.sflow
376 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
377 ds_put_format(ds, ",sFlow("
378 "vid=%"PRIu16",pcp=%"PRIu8",output=%"PRIu32")",
379 vlan_tci_to_vid(cookie.sflow.vlan_tci),
380 vlan_tci_to_pcp(cookie.sflow.vlan_tci),
381 cookie.sflow.output);
382 } else if (userdata_len == sizeof cookie.slow_path
383 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
384 ds_put_cstr(ds, ",slow_path(");
385 format_flags(ds, slow_path_reason_to_string,
386 cookie.slow_path.reason, ',');
387 ds_put_format(ds, ")");
388 } else if (userdata_len == sizeof cookie.flow_sample
389 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
390 ds_put_format(ds, ",flow_sample(probability=%"PRIu16
391 ",collector_set_id=%"PRIu32
392 ",obs_domain_id=%"PRIu32
393 ",obs_point_id=%"PRIu32")",
394 cookie.flow_sample.probability,
395 cookie.flow_sample.collector_set_id,
396 cookie.flow_sample.obs_domain_id,
397 cookie.flow_sample.obs_point_id);
398 } else if (userdata_len >= sizeof cookie.ipfix
399 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
400 ds_put_format(ds, ",ipfix(output_port=%"PRIu32")",
401 cookie.ipfix.output_odp_port);
402 } else {
403 userdata_unspec = true;
404 }
405 }
406
407 if (userdata_unspec) {
408 size_t i;
409 ds_put_format(ds, ",userdata(");
410 for (i = 0; i < userdata_len; i++) {
411 ds_put_format(ds, "%02x", userdata[i]);
412 }
413 ds_put_char(ds, ')');
414 }
415 }
416
417 tunnel_out_port_attr = a[OVS_USERSPACE_ATTR_EGRESS_TUN_PORT];
418 if (tunnel_out_port_attr) {
419 ds_put_format(ds, ",tunnel_out_port=%"PRIu32,
420 nl_attr_get_u32(tunnel_out_port_attr));
421 }
422
423 ds_put_char(ds, ')');
424 }
425
426 static void
427 format_vlan_tci(struct ds *ds, ovs_be16 tci, ovs_be16 mask, bool verbose)
428 {
429 if (verbose || vlan_tci_to_vid(tci) || vlan_tci_to_vid(mask)) {
430 ds_put_format(ds, "vid=%"PRIu16, vlan_tci_to_vid(tci));
431 if (vlan_tci_to_vid(mask) != VLAN_VID_MASK) { /* Partially masked. */
432 ds_put_format(ds, "/0x%"PRIx16, vlan_tci_to_vid(mask));
433 };
434 ds_put_char(ds, ',');
435 }
436 if (verbose || vlan_tci_to_pcp(tci) || vlan_tci_to_pcp(mask)) {
437 ds_put_format(ds, "pcp=%d", vlan_tci_to_pcp(tci));
438 if (vlan_tci_to_pcp(mask) != (VLAN_PCP_MASK >> VLAN_PCP_SHIFT)) {
439 ds_put_format(ds, "/0x%x", vlan_tci_to_pcp(mask));
440 }
441 ds_put_char(ds, ',');
442 }
443 if (!(tci & htons(VLAN_CFI))) {
444 ds_put_cstr(ds, "cfi=0");
445 ds_put_char(ds, ',');
446 }
447 ds_chomp(ds, ',');
448 }
449
450 static void
451 format_mpls_lse(struct ds *ds, ovs_be32 mpls_lse)
452 {
453 ds_put_format(ds, "label=%"PRIu32",tc=%d,ttl=%d,bos=%d",
454 mpls_lse_to_label(mpls_lse),
455 mpls_lse_to_tc(mpls_lse),
456 mpls_lse_to_ttl(mpls_lse),
457 mpls_lse_to_bos(mpls_lse));
458 }
459
460 static void
461 format_mpls(struct ds *ds, const struct ovs_key_mpls *mpls_key,
462 const struct ovs_key_mpls *mpls_mask, int n)
463 {
464 if (n == 1) {
465 ovs_be32 key = mpls_key->mpls_lse;
466
467 if (mpls_mask == NULL) {
468 format_mpls_lse(ds, key);
469 } else {
470 ovs_be32 mask = mpls_mask->mpls_lse;
471
472 ds_put_format(ds, "label=%"PRIu32"/0x%x,tc=%d/%x,ttl=%d/0x%x,bos=%d/%x",
473 mpls_lse_to_label(key), mpls_lse_to_label(mask),
474 mpls_lse_to_tc(key), mpls_lse_to_tc(mask),
475 mpls_lse_to_ttl(key), mpls_lse_to_ttl(mask),
476 mpls_lse_to_bos(key), mpls_lse_to_bos(mask));
477 }
478 } else {
479 int i;
480
481 for (i = 0; i < n; i++) {
482 ds_put_format(ds, "lse%d=%#"PRIx32,
483 i, ntohl(mpls_key[i].mpls_lse));
484 if (mpls_mask) {
485 ds_put_format(ds, "/%#"PRIx32, ntohl(mpls_mask[i].mpls_lse));
486 }
487 ds_put_char(ds, ',');
488 }
489 ds_chomp(ds, ',');
490 }
491 }
492
493 static void
494 format_odp_recirc_action(struct ds *ds, uint32_t recirc_id)
495 {
496 ds_put_format(ds, "recirc(%"PRIu32")", recirc_id);
497 }
498
499 static void
500 format_odp_hash_action(struct ds *ds, const struct ovs_action_hash *hash_act)
501 {
502 ds_put_format(ds, "hash(");
503
504 if (hash_act->hash_alg == OVS_HASH_ALG_L4) {
505 ds_put_format(ds, "hash_l4(%"PRIu32")", hash_act->hash_basis);
506 } else {
507 ds_put_format(ds, "Unknown hash algorithm(%"PRIu32")",
508 hash_act->hash_alg);
509 }
510 ds_put_format(ds, ")");
511 }
512
513 static const void *
514 format_udp_tnl_push_header(struct ds *ds, const struct ip_header *ip)
515 {
516 const struct udp_header *udp;
517
518 udp = (const struct udp_header *) (ip + 1);
519 ds_put_format(ds, "udp(src=%"PRIu16",dst=%"PRIu16"),",
520 ntohs(udp->udp_src), ntohs(udp->udp_dst));
521
522 return udp + 1;
523 }
524
525 static void
526 format_odp_tnl_push_header(struct ds *ds, struct ovs_action_push_tnl *data)
527 {
528 const struct eth_header *eth;
529 const struct ip_header *ip;
530 const void *l3;
531
532 eth = (const struct eth_header *)data->header;
533
534 l3 = eth + 1;
535 ip = (const struct ip_header *)l3;
536
537 /* Ethernet */
538 ds_put_format(ds, "header(size=%"PRIu8",type=%"PRIu8",eth(dst=",
539 data->header_len, data->tnl_type);
540 ds_put_format(ds, ETH_ADDR_FMT, ETH_ADDR_ARGS(eth->eth_dst));
541 ds_put_format(ds, ",src=");
542 ds_put_format(ds, ETH_ADDR_FMT, ETH_ADDR_ARGS(eth->eth_src));
543 ds_put_format(ds, ",dl_type=0x%04"PRIx16"),", ntohs(eth->eth_type));
544
545 /* IPv4 */
546 ds_put_format(ds, "ipv4(src="IP_FMT",dst="IP_FMT",proto=%"PRIu8
547 ",tos=%#"PRIx8",ttl=%"PRIu8",frag=0x%"PRIx16"),",
548 IP_ARGS(get_16aligned_be32(&ip->ip_src)),
549 IP_ARGS(get_16aligned_be32(&ip->ip_dst)),
550 ip->ip_proto, ip->ip_tos,
551 ip->ip_ttl,
552 ip->ip_frag_off);
553
554 if (data->tnl_type == OVS_VPORT_TYPE_VXLAN) {
555 const struct vxlanhdr *vxh;
556
557 vxh = format_udp_tnl_push_header(ds, ip);
558
559 ds_put_format(ds, "vxlan(flags=0x%"PRIx32",vni=0x%"PRIx32")",
560 ntohl(get_16aligned_be32(&vxh->vx_flags)),
561 ntohl(get_16aligned_be32(&vxh->vx_vni)) >> 8);
562 } else if (data->tnl_type == OVS_VPORT_TYPE_GENEVE) {
563 const struct genevehdr *gnh;
564
565 gnh = format_udp_tnl_push_header(ds, ip);
566
567 ds_put_format(ds, "geneve(%svni=0x%"PRIx32")",
568 gnh->oam ? "oam," : "",
569 ntohl(get_16aligned_be32(&gnh->vni)) >> 8);
570 } else if (data->tnl_type == OVS_VPORT_TYPE_GRE) {
571 const struct gre_base_hdr *greh;
572 ovs_16aligned_be32 *options;
573 void *l4;
574
575 l4 = ((uint8_t *)l3 + sizeof(struct ip_header));
576 greh = (const struct gre_base_hdr *) l4;
577
578 ds_put_format(ds, "gre((flags=0x%"PRIx16",proto=0x%"PRIx16")",
579 greh->flags, ntohs(greh->protocol));
580 options = (ovs_16aligned_be32 *)(greh + 1);
581 if (greh->flags & htons(GRE_CSUM)) {
582 ds_put_format(ds, ",csum=0x%"PRIx16, ntohs(*((ovs_be16 *)options)));
583 options++;
584 }
585 if (greh->flags & htons(GRE_KEY)) {
586 ds_put_format(ds, ",key=0x%"PRIx32, ntohl(get_16aligned_be32(options)));
587 options++;
588 }
589 if (greh->flags & htons(GRE_SEQ)) {
590 ds_put_format(ds, ",seq=0x%"PRIx32, ntohl(get_16aligned_be32(options)));
591 options++;
592 }
593 ds_put_format(ds, ")");
594 }
595 ds_put_format(ds, ")");
596 }
597
598 static void
599 format_odp_tnl_push_action(struct ds *ds, const struct nlattr *attr)
600 {
601 struct ovs_action_push_tnl *data;
602
603 data = (struct ovs_action_push_tnl *) nl_attr_get(attr);
604
605 ds_put_format(ds, "tnl_push(tnl_port(%"PRIu32"),", data->tnl_port);
606 format_odp_tnl_push_header(ds, data);
607 ds_put_format(ds, ",out_port(%"PRIu32"))", data->out_port);
608 }
609
610 static void
611 format_odp_action(struct ds *ds, const struct nlattr *a)
612 {
613 int expected_len;
614 enum ovs_action_attr type = nl_attr_type(a);
615 const struct ovs_action_push_vlan *vlan;
616 size_t size;
617
618 expected_len = odp_action_len(nl_attr_type(a));
619 if (expected_len != -2 && nl_attr_get_size(a) != expected_len) {
620 ds_put_format(ds, "bad length %"PRIuSIZE", expected %d for: ",
621 nl_attr_get_size(a), expected_len);
622 format_generic_odp_action(ds, a);
623 return;
624 }
625
626 switch (type) {
627 case OVS_ACTION_ATTR_OUTPUT:
628 ds_put_format(ds, "%"PRIu32, nl_attr_get_u32(a));
629 break;
630 case OVS_ACTION_ATTR_TUNNEL_POP:
631 ds_put_format(ds, "tnl_pop(%"PRIu32")", nl_attr_get_u32(a));
632 break;
633 case OVS_ACTION_ATTR_TUNNEL_PUSH:
634 format_odp_tnl_push_action(ds, a);
635 break;
636 case OVS_ACTION_ATTR_USERSPACE:
637 format_odp_userspace_action(ds, a);
638 break;
639 case OVS_ACTION_ATTR_RECIRC:
640 format_odp_recirc_action(ds, nl_attr_get_u32(a));
641 break;
642 case OVS_ACTION_ATTR_HASH:
643 format_odp_hash_action(ds, nl_attr_get(a));
644 break;
645 case OVS_ACTION_ATTR_SET_MASKED:
646 a = nl_attr_get(a);
647 size = nl_attr_get_size(a) / 2;
648 ds_put_cstr(ds, "set(");
649
650 /* Masked set action not supported for tunnel key, which is bigger. */
651 if (size <= sizeof(struct ovs_key_ipv6)) {
652 struct nlattr attr[1 + DIV_ROUND_UP(sizeof(struct ovs_key_ipv6),
653 sizeof(struct nlattr))];
654 struct nlattr mask[1 + DIV_ROUND_UP(sizeof(struct ovs_key_ipv6),
655 sizeof(struct nlattr))];
656
657 mask->nla_type = attr->nla_type = nl_attr_type(a);
658 mask->nla_len = attr->nla_len = NLA_HDRLEN + size;
659 memcpy(attr + 1, (char *)(a + 1), size);
660 memcpy(mask + 1, (char *)(a + 1) + size, size);
661 format_odp_key_attr(attr, mask, NULL, ds, false);
662 } else {
663 format_odp_key_attr(a, NULL, NULL, ds, false);
664 }
665 ds_put_cstr(ds, ")");
666 break;
667 case OVS_ACTION_ATTR_SET:
668 ds_put_cstr(ds, "set(");
669 format_odp_key_attr(nl_attr_get(a), NULL, NULL, ds, true);
670 ds_put_cstr(ds, ")");
671 break;
672 case OVS_ACTION_ATTR_PUSH_VLAN:
673 vlan = nl_attr_get(a);
674 ds_put_cstr(ds, "push_vlan(");
675 if (vlan->vlan_tpid != htons(ETH_TYPE_VLAN)) {
676 ds_put_format(ds, "tpid=0x%04"PRIx16",", ntohs(vlan->vlan_tpid));
677 }
678 format_vlan_tci(ds, vlan->vlan_tci, OVS_BE16_MAX, false);
679 ds_put_char(ds, ')');
680 break;
681 case OVS_ACTION_ATTR_POP_VLAN:
682 ds_put_cstr(ds, "pop_vlan");
683 break;
684 case OVS_ACTION_ATTR_PUSH_MPLS: {
685 const struct ovs_action_push_mpls *mpls = nl_attr_get(a);
686 ds_put_cstr(ds, "push_mpls(");
687 format_mpls_lse(ds, mpls->mpls_lse);
688 ds_put_format(ds, ",eth_type=0x%"PRIx16")", ntohs(mpls->mpls_ethertype));
689 break;
690 }
691 case OVS_ACTION_ATTR_POP_MPLS: {
692 ovs_be16 ethertype = nl_attr_get_be16(a);
693 ds_put_format(ds, "pop_mpls(eth_type=0x%"PRIx16")", ntohs(ethertype));
694 break;
695 }
696 case OVS_ACTION_ATTR_SAMPLE:
697 format_odp_sample_action(ds, a);
698 break;
699 case OVS_ACTION_ATTR_UNSPEC:
700 case __OVS_ACTION_ATTR_MAX:
701 default:
702 format_generic_odp_action(ds, a);
703 break;
704 }
705 }
706
707 void
708 format_odp_actions(struct ds *ds, const struct nlattr *actions,
709 size_t actions_len)
710 {
711 if (actions_len) {
712 const struct nlattr *a;
713 unsigned int left;
714
715 NL_ATTR_FOR_EACH (a, left, actions, actions_len) {
716 if (a != actions) {
717 ds_put_char(ds, ',');
718 }
719 format_odp_action(ds, a);
720 }
721 if (left) {
722 int i;
723
724 if (left == actions_len) {
725 ds_put_cstr(ds, "<empty>");
726 }
727 ds_put_format(ds, ",***%u leftover bytes*** (", left);
728 for (i = 0; i < left; i++) {
729 ds_put_format(ds, "%02x", ((const uint8_t *) a)[i]);
730 }
731 ds_put_char(ds, ')');
732 }
733 } else {
734 ds_put_cstr(ds, "drop");
735 }
736 }
737
738 /* Separate out parse_odp_userspace_action() function. */
739 static int
740 parse_odp_userspace_action(const char *s, struct ofpbuf *actions)
741 {
742 uint32_t pid;
743 union user_action_cookie cookie;
744 struct ofpbuf buf;
745 odp_port_t tunnel_out_port;
746 int n = -1;
747 void *user_data = NULL;
748 size_t user_data_size = 0;
749
750 if (!ovs_scan(s, "userspace(pid=%"SCNi32"%n", &pid, &n)) {
751 return -EINVAL;
752 }
753
754 {
755 uint32_t output;
756 uint32_t probability;
757 uint32_t collector_set_id;
758 uint32_t obs_domain_id;
759 uint32_t obs_point_id;
760 int vid, pcp;
761 int n1 = -1;
762 if (ovs_scan(&s[n], ",sFlow(vid=%i,"
763 "pcp=%i,output=%"SCNi32")%n",
764 &vid, &pcp, &output, &n1)) {
765 uint16_t tci;
766
767 n += n1;
768 tci = vid | (pcp << VLAN_PCP_SHIFT);
769 if (tci) {
770 tci |= VLAN_CFI;
771 }
772
773 cookie.type = USER_ACTION_COOKIE_SFLOW;
774 cookie.sflow.vlan_tci = htons(tci);
775 cookie.sflow.output = output;
776 user_data = &cookie;
777 user_data_size = sizeof cookie.sflow;
778 } else if (ovs_scan(&s[n], ",slow_path(%n",
779 &n1)) {
780 int res;
781
782 n += n1;
783 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
784 cookie.slow_path.unused = 0;
785 cookie.slow_path.reason = 0;
786
787 res = parse_flags(&s[n], slow_path_reason_to_string,
788 &cookie.slow_path.reason,
789 SLOW_PATH_REASON_MASK, NULL);
790 if (res < 0 || s[n + res] != ')') {
791 return res;
792 }
793 n += res + 1;
794
795 user_data = &cookie;
796 user_data_size = sizeof cookie.slow_path;
797 } else if (ovs_scan(&s[n], ",flow_sample(probability=%"SCNi32","
798 "collector_set_id=%"SCNi32","
799 "obs_domain_id=%"SCNi32","
800 "obs_point_id=%"SCNi32")%n",
801 &probability, &collector_set_id,
802 &obs_domain_id, &obs_point_id, &n1)) {
803 n += n1;
804
805 cookie.type = USER_ACTION_COOKIE_FLOW_SAMPLE;
806 cookie.flow_sample.probability = probability;
807 cookie.flow_sample.collector_set_id = collector_set_id;
808 cookie.flow_sample.obs_domain_id = obs_domain_id;
809 cookie.flow_sample.obs_point_id = obs_point_id;
810 user_data = &cookie;
811 user_data_size = sizeof cookie.flow_sample;
812 } else if (ovs_scan(&s[n], ",ipfix(output_port=%"SCNi32")%n",
813 &output, &n1) ) {
814 n += n1;
815 cookie.type = USER_ACTION_COOKIE_IPFIX;
816 cookie.ipfix.output_odp_port = u32_to_odp(output);
817 user_data = &cookie;
818 user_data_size = sizeof cookie.ipfix;
819 } else if (ovs_scan(&s[n], ",userdata(%n",
820 &n1)) {
821 char *end;
822
823 n += n1;
824 ofpbuf_init(&buf, 16);
825 end = ofpbuf_put_hex(&buf, &s[n], NULL);
826 if (end[0] != ')') {
827 return -EINVAL;
828 }
829 user_data = buf.data;
830 user_data_size = buf.size;
831 n = (end + 1) - s;
832 }
833 }
834
835 {
836 int n1 = -1;
837 if (ovs_scan(&s[n], ",tunnel_out_port=%"SCNi32")%n",
838 &tunnel_out_port, &n1)) {
839 odp_put_userspace_action(pid, user_data, user_data_size, tunnel_out_port, actions);
840 return n + n1;
841 } else if (s[n] == ')') {
842 odp_put_userspace_action(pid, user_data, user_data_size, ODPP_NONE, actions);
843 return n + 1;
844 }
845 }
846
847 return -EINVAL;
848 }
849
850 static int
851 ovs_parse_tnl_push(const char *s, struct ovs_action_push_tnl *data)
852 {
853 struct eth_header *eth;
854 struct ip_header *ip;
855 struct udp_header *udp;
856 struct gre_base_hdr *greh;
857 uint16_t gre_proto, dl_type, udp_src, udp_dst;
858 ovs_be32 sip, dip;
859 uint32_t tnl_type = 0, header_len = 0;
860 void *l3, *l4;
861 int n = 0;
862
863 if (!ovs_scan_len(s, &n, "tnl_push(tnl_port(%"SCNi32"),", &data->tnl_port)) {
864 return -EINVAL;
865 }
866 eth = (struct eth_header *) data->header;
867 l3 = (data->header + sizeof *eth);
868 l4 = ((uint8_t *) l3 + sizeof (struct ip_header));
869 ip = (struct ip_header *) l3;
870 if (!ovs_scan_len(s, &n, "header(size=%"SCNi32",type=%"SCNi32","
871 "eth(dst="ETH_ADDR_SCAN_FMT",",
872 &data->header_len,
873 &data->tnl_type,
874 ETH_ADDR_SCAN_ARGS(eth->eth_dst))) {
875 return -EINVAL;
876 }
877
878 if (!ovs_scan_len(s, &n, "src="ETH_ADDR_SCAN_FMT",",
879 ETH_ADDR_SCAN_ARGS(eth->eth_src))) {
880 return -EINVAL;
881 }
882 if (!ovs_scan_len(s, &n, "dl_type=0x%"SCNx16"),", &dl_type)) {
883 return -EINVAL;
884 }
885 eth->eth_type = htons(dl_type);
886
887 /* IPv4 */
888 if (!ovs_scan_len(s, &n, "ipv4(src="IP_SCAN_FMT",dst="IP_SCAN_FMT",proto=%"SCNi8
889 ",tos=%"SCNi8",ttl=%"SCNi8",frag=0x%"SCNx16"),",
890 IP_SCAN_ARGS(&sip),
891 IP_SCAN_ARGS(&dip),
892 &ip->ip_proto, &ip->ip_tos,
893 &ip->ip_ttl, &ip->ip_frag_off)) {
894 return -EINVAL;
895 }
896 put_16aligned_be32(&ip->ip_src, sip);
897 put_16aligned_be32(&ip->ip_dst, dip);
898
899 /* Tunnel header */
900 udp = (struct udp_header *) l4;
901 greh = (struct gre_base_hdr *) l4;
902 if (ovs_scan_len(s, &n, "udp(src=%"SCNi16",dst=%"SCNi16"),",
903 &udp_src, &udp_dst)) {
904 uint32_t vx_flags, vni;
905
906 udp->udp_src = htons(udp_src);
907 udp->udp_dst = htons(udp_dst);
908 udp->udp_len = 0;
909 udp->udp_csum = 0;
910
911 if (ovs_scan_len(s, &n, "vxlan(flags=0x%"SCNx32",vni=0x%"SCNx32"))",
912 &vx_flags, &vni)) {
913 struct vxlanhdr *vxh = (struct vxlanhdr *) (udp + 1);
914
915 put_16aligned_be32(&vxh->vx_flags, htonl(vx_flags));
916 put_16aligned_be32(&vxh->vx_vni, htonl(vni << 8));
917 tnl_type = OVS_VPORT_TYPE_VXLAN;
918 header_len = sizeof *eth + sizeof *ip +
919 sizeof *udp + sizeof *vxh;
920 } else if (ovs_scan_len(s, &n, "geneve(")) {
921 struct genevehdr *gnh = (struct genevehdr *) (udp + 1);
922
923 if (ovs_scan_len(s, &n, "oam,")) {
924 gnh->oam = 1;
925 }
926 if (!ovs_scan_len(s, &n, "vni=0x%"SCNx32"))", &vni)) {
927 return -EINVAL;
928 }
929 gnh->proto_type = htons(ETH_TYPE_TEB);
930 put_16aligned_be32(&gnh->vni, htonl(vni << 8));
931 tnl_type = OVS_VPORT_TYPE_GENEVE;
932 header_len = sizeof *eth + sizeof *ip +
933 sizeof *udp + sizeof *gnh;
934 } else {
935 return -EINVAL;
936 }
937 } else if (ovs_scan_len(s, &n, "gre((flags=0x%"SCNx16",proto=0x%"SCNx16")",
938 &greh->flags, &gre_proto)){
939
940 tnl_type = OVS_VPORT_TYPE_GRE;
941 greh->protocol = htons(gre_proto);
942 ovs_16aligned_be32 *options = (ovs_16aligned_be32 *) (greh + 1);
943
944 if (greh->flags & htons(GRE_CSUM)) {
945 uint16_t csum;
946
947 if (!ovs_scan_len(s, &n, ",csum=0x%"SCNx16, &csum)) {
948 return -EINVAL;
949 }
950
951 memset(options, 0, sizeof *options);
952 *((ovs_be16 *)options) = htons(csum);
953 options++;
954 }
955 if (greh->flags & htons(GRE_KEY)) {
956 uint32_t key;
957
958 if (!ovs_scan_len(s, &n, ",key=0x%"SCNx32, &key)) {
959 return -EINVAL;
960 }
961
962 put_16aligned_be32(options, htonl(key));
963 options++;
964 }
965 if (greh->flags & htons(GRE_SEQ)) {
966 uint32_t seq;
967
968 if (!ovs_scan_len(s, &n, ",seq=0x%"SCNx32, &seq)) {
969 return -EINVAL;
970 }
971 put_16aligned_be32(options, htonl(seq));
972 options++;
973 }
974
975 if (!ovs_scan_len(s, &n, "))")) {
976 return -EINVAL;
977 }
978
979 header_len = sizeof *eth + sizeof *ip +
980 ((uint8_t *) options - (uint8_t *) greh);
981 } else {
982 return -EINVAL;
983 }
984
985 /* check tunnel meta data. */
986 if (data->tnl_type != tnl_type) {
987 return -EINVAL;
988 }
989 if (data->header_len != header_len) {
990 return -EINVAL;
991 }
992
993 /* Out port */
994 if (!ovs_scan_len(s, &n, ",out_port(%"SCNi32"))", &data->out_port)) {
995 return -EINVAL;
996 }
997
998 return n;
999 }
1000
1001 static int
1002 parse_odp_action(const char *s, const struct simap *port_names,
1003 struct ofpbuf *actions)
1004 {
1005 {
1006 uint32_t port;
1007 int n;
1008
1009 if (ovs_scan(s, "%"SCNi32"%n", &port, &n)) {
1010 nl_msg_put_u32(actions, OVS_ACTION_ATTR_OUTPUT, port);
1011 return n;
1012 }
1013 }
1014
1015 if (port_names) {
1016 int len = strcspn(s, delimiters);
1017 struct simap_node *node;
1018
1019 node = simap_find_len(port_names, s, len);
1020 if (node) {
1021 nl_msg_put_u32(actions, OVS_ACTION_ATTR_OUTPUT, node->data);
1022 return len;
1023 }
1024 }
1025
1026 {
1027 uint32_t recirc_id;
1028 int n = -1;
1029
1030 if (ovs_scan(s, "recirc(%"PRIu32")%n", &recirc_id, &n)) {
1031 nl_msg_put_u32(actions, OVS_ACTION_ATTR_RECIRC, recirc_id);
1032 return n;
1033 }
1034 }
1035
1036 if (!strncmp(s, "userspace(", 10)) {
1037 return parse_odp_userspace_action(s, actions);
1038 }
1039
1040 if (!strncmp(s, "set(", 4)) {
1041 size_t start_ofs;
1042 int retval;
1043 struct nlattr mask[128 / sizeof(struct nlattr)];
1044 struct ofpbuf maskbuf;
1045 struct nlattr *nested, *key;
1046 size_t size;
1047
1048 /* 'mask' is big enough to hold any key. */
1049 ofpbuf_use_stack(&maskbuf, mask, sizeof mask);
1050
1051 start_ofs = nl_msg_start_nested(actions, OVS_ACTION_ATTR_SET);
1052 retval = parse_odp_key_mask_attr(s + 4, port_names, actions, &maskbuf);
1053 if (retval < 0) {
1054 return retval;
1055 }
1056 if (s[retval + 4] != ')') {
1057 return -EINVAL;
1058 }
1059
1060 nested = ofpbuf_at_assert(actions, start_ofs, sizeof *nested);
1061 key = nested + 1;
1062
1063 size = nl_attr_get_size(mask);
1064 if (size == nl_attr_get_size(key)) {
1065 /* Change to masked set action if not fully masked. */
1066 if (!is_all_ones(mask + 1, size)) {
1067 key->nla_len += size;
1068 ofpbuf_put(actions, mask + 1, size);
1069 /* 'actions' may have been reallocated by ofpbuf_put(). */
1070 nested = ofpbuf_at_assert(actions, start_ofs, sizeof *nested);
1071 nested->nla_type = OVS_ACTION_ATTR_SET_MASKED;
1072 }
1073 }
1074
1075 nl_msg_end_nested(actions, start_ofs);
1076 return retval + 5;
1077 }
1078
1079 {
1080 struct ovs_action_push_vlan push;
1081 int tpid = ETH_TYPE_VLAN;
1082 int vid, pcp;
1083 int cfi = 1;
1084 int n = -1;
1085
1086 if (ovs_scan(s, "push_vlan(vid=%i,pcp=%i)%n", &vid, &pcp, &n)
1087 || ovs_scan(s, "push_vlan(vid=%i,pcp=%i,cfi=%i)%n",
1088 &vid, &pcp, &cfi, &n)
1089 || ovs_scan(s, "push_vlan(tpid=%i,vid=%i,pcp=%i)%n",
1090 &tpid, &vid, &pcp, &n)
1091 || ovs_scan(s, "push_vlan(tpid=%i,vid=%i,pcp=%i,cfi=%i)%n",
1092 &tpid, &vid, &pcp, &cfi, &n)) {
1093 push.vlan_tpid = htons(tpid);
1094 push.vlan_tci = htons((vid << VLAN_VID_SHIFT)
1095 | (pcp << VLAN_PCP_SHIFT)
1096 | (cfi ? VLAN_CFI : 0));
1097 nl_msg_put_unspec(actions, OVS_ACTION_ATTR_PUSH_VLAN,
1098 &push, sizeof push);
1099
1100 return n;
1101 }
1102 }
1103
1104 if (!strncmp(s, "pop_vlan", 8)) {
1105 nl_msg_put_flag(actions, OVS_ACTION_ATTR_POP_VLAN);
1106 return 8;
1107 }
1108
1109 {
1110 double percentage;
1111 int n = -1;
1112
1113 if (ovs_scan(s, "sample(sample=%lf%%,actions(%n", &percentage, &n)
1114 && percentage >= 0. && percentage <= 100.0) {
1115 size_t sample_ofs, actions_ofs;
1116 double probability;
1117
1118 probability = floor(UINT32_MAX * (percentage / 100.0) + .5);
1119 sample_ofs = nl_msg_start_nested(actions, OVS_ACTION_ATTR_SAMPLE);
1120 nl_msg_put_u32(actions, OVS_SAMPLE_ATTR_PROBABILITY,
1121 (probability <= 0 ? 0
1122 : probability >= UINT32_MAX ? UINT32_MAX
1123 : probability));
1124
1125 actions_ofs = nl_msg_start_nested(actions,
1126 OVS_SAMPLE_ATTR_ACTIONS);
1127 for (;;) {
1128 int retval;
1129
1130 n += strspn(s + n, delimiters);
1131 if (s[n] == ')') {
1132 break;
1133 }
1134
1135 retval = parse_odp_action(s + n, port_names, actions);
1136 if (retval < 0) {
1137 return retval;
1138 }
1139 n += retval;
1140 }
1141 nl_msg_end_nested(actions, actions_ofs);
1142 nl_msg_end_nested(actions, sample_ofs);
1143
1144 return s[n + 1] == ')' ? n + 2 : -EINVAL;
1145 }
1146 }
1147
1148 {
1149 uint32_t port;
1150 int n;
1151
1152 if (ovs_scan(s, "tnl_pop(%"SCNi32")%n", &port, &n)) {
1153 nl_msg_put_u32(actions, OVS_ACTION_ATTR_TUNNEL_POP, port);
1154 return n;
1155 }
1156 }
1157
1158 {
1159 struct ovs_action_push_tnl data;
1160 int n;
1161
1162 n = ovs_parse_tnl_push(s, &data);
1163 if (n > 0) {
1164 odp_put_tnl_push_action(actions, &data);
1165 return n;
1166 } else if (n < 0) {
1167 return n;
1168 }
1169 }
1170 return -EINVAL;
1171 }
1172
1173 /* Parses the string representation of datapath actions, in the format output
1174 * by format_odp_action(). Returns 0 if successful, otherwise a positive errno
1175 * value. On success, the ODP actions are appended to 'actions' as a series of
1176 * Netlink attributes. On failure, no data is appended to 'actions'. Either
1177 * way, 'actions''s data might be reallocated. */
1178 int
1179 odp_actions_from_string(const char *s, const struct simap *port_names,
1180 struct ofpbuf *actions)
1181 {
1182 size_t old_size;
1183
1184 if (!strcasecmp(s, "drop")) {
1185 return 0;
1186 }
1187
1188 old_size = actions->size;
1189 for (;;) {
1190 int retval;
1191
1192 s += strspn(s, delimiters);
1193 if (!*s) {
1194 return 0;
1195 }
1196
1197 retval = parse_odp_action(s, port_names, actions);
1198 if (retval < 0 || !strchr(delimiters, s[retval])) {
1199 actions->size = old_size;
1200 return -retval;
1201 }
1202 s += retval;
1203 }
1204
1205 return 0;
1206 }
1207 \f
1208 /* Returns the correct length of the payload for a flow key attribute of the
1209 * specified 'type', -1 if 'type' is unknown, or -2 if the attribute's payload
1210 * is variable length. */
1211 static int
1212 odp_flow_key_attr_len(uint16_t type)
1213 {
1214 if (type > OVS_KEY_ATTR_MAX) {
1215 return -1;
1216 }
1217
1218 switch ((enum ovs_key_attr) type) {
1219 case OVS_KEY_ATTR_ENCAP: return -2;
1220 case OVS_KEY_ATTR_PRIORITY: return 4;
1221 case OVS_KEY_ATTR_SKB_MARK: return 4;
1222 case OVS_KEY_ATTR_DP_HASH: return 4;
1223 case OVS_KEY_ATTR_RECIRC_ID: return 4;
1224 case OVS_KEY_ATTR_TUNNEL: return -2;
1225 case OVS_KEY_ATTR_IN_PORT: return 4;
1226 case OVS_KEY_ATTR_ETHERNET: return sizeof(struct ovs_key_ethernet);
1227 case OVS_KEY_ATTR_VLAN: return sizeof(ovs_be16);
1228 case OVS_KEY_ATTR_ETHERTYPE: return 2;
1229 case OVS_KEY_ATTR_MPLS: return -2;
1230 case OVS_KEY_ATTR_IPV4: return sizeof(struct ovs_key_ipv4);
1231 case OVS_KEY_ATTR_IPV6: return sizeof(struct ovs_key_ipv6);
1232 case OVS_KEY_ATTR_TCP: return sizeof(struct ovs_key_tcp);
1233 case OVS_KEY_ATTR_TCP_FLAGS: return 2;
1234 case OVS_KEY_ATTR_UDP: return sizeof(struct ovs_key_udp);
1235 case OVS_KEY_ATTR_SCTP: return sizeof(struct ovs_key_sctp);
1236 case OVS_KEY_ATTR_ICMP: return sizeof(struct ovs_key_icmp);
1237 case OVS_KEY_ATTR_ICMPV6: return sizeof(struct ovs_key_icmpv6);
1238 case OVS_KEY_ATTR_ARP: return sizeof(struct ovs_key_arp);
1239 case OVS_KEY_ATTR_ND: return sizeof(struct ovs_key_nd);
1240
1241 case OVS_KEY_ATTR_UNSPEC:
1242 case __OVS_KEY_ATTR_MAX:
1243 return -1;
1244 }
1245
1246 return -1;
1247 }
1248
1249 static void
1250 format_generic_odp_key(const struct nlattr *a, struct ds *ds)
1251 {
1252 size_t len = nl_attr_get_size(a);
1253 if (len) {
1254 const uint8_t *unspec;
1255 unsigned int i;
1256
1257 unspec = nl_attr_get(a);
1258 for (i = 0; i < len; i++) {
1259 if (i) {
1260 ds_put_char(ds, ' ');
1261 }
1262 ds_put_format(ds, "%02x", unspec[i]);
1263 }
1264 }
1265 }
1266
1267 static const char *
1268 ovs_frag_type_to_string(enum ovs_frag_type type)
1269 {
1270 switch (type) {
1271 case OVS_FRAG_TYPE_NONE:
1272 return "no";
1273 case OVS_FRAG_TYPE_FIRST:
1274 return "first";
1275 case OVS_FRAG_TYPE_LATER:
1276 return "later";
1277 case __OVS_FRAG_TYPE_MAX:
1278 default:
1279 return "<error>";
1280 }
1281 }
1282
1283 static int
1284 tunnel_key_attr_len(int type)
1285 {
1286 switch (type) {
1287 case OVS_TUNNEL_KEY_ATTR_ID: return 8;
1288 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC: return 4;
1289 case OVS_TUNNEL_KEY_ATTR_IPV4_DST: return 4;
1290 case OVS_TUNNEL_KEY_ATTR_TOS: return 1;
1291 case OVS_TUNNEL_KEY_ATTR_TTL: return 1;
1292 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT: return 0;
1293 case OVS_TUNNEL_KEY_ATTR_CSUM: return 0;
1294 case OVS_TUNNEL_KEY_ATTR_TP_SRC: return 2;
1295 case OVS_TUNNEL_KEY_ATTR_TP_DST: return 2;
1296 case OVS_TUNNEL_KEY_ATTR_OAM: return 0;
1297 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS: return -2;
1298 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS: return -2;
1299 case __OVS_TUNNEL_KEY_ATTR_MAX:
1300 return -1;
1301 }
1302 return -1;
1303 }
1304
1305 #define GENEVE_OPT(class, type) ((OVS_FORCE uint32_t)(class) << 8 | (type))
1306 static int
1307 parse_geneve_opts(const struct nlattr *attr)
1308 {
1309 int opts_len = nl_attr_get_size(attr);
1310 const struct geneve_opt *opt = nl_attr_get(attr);
1311
1312 while (opts_len > 0) {
1313 int len;
1314
1315 if (opts_len < sizeof(*opt)) {
1316 return -EINVAL;
1317 }
1318
1319 len = sizeof(*opt) + opt->length * 4;
1320 if (len > opts_len) {
1321 return -EINVAL;
1322 }
1323
1324 switch (GENEVE_OPT(opt->opt_class, opt->type)) {
1325 default:
1326 if (opt->type & GENEVE_CRIT_OPT_TYPE) {
1327 return -EINVAL;
1328 }
1329 };
1330
1331 opt = opt + len / sizeof(*opt);
1332 opts_len -= len;
1333 };
1334
1335 return 0;
1336 }
1337
1338 enum odp_key_fitness
1339 odp_tun_key_from_attr(const struct nlattr *attr, struct flow_tnl *tun)
1340 {
1341 unsigned int left;
1342 const struct nlattr *a;
1343 bool ttl = false;
1344 bool unknown = false;
1345
1346 NL_NESTED_FOR_EACH(a, left, attr) {
1347 uint16_t type = nl_attr_type(a);
1348 size_t len = nl_attr_get_size(a);
1349 int expected_len = tunnel_key_attr_len(type);
1350
1351 if (len != expected_len && expected_len >= 0) {
1352 return ODP_FIT_ERROR;
1353 }
1354
1355 switch (type) {
1356 case OVS_TUNNEL_KEY_ATTR_ID:
1357 tun->tun_id = nl_attr_get_be64(a);
1358 tun->flags |= FLOW_TNL_F_KEY;
1359 break;
1360 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1361 tun->ip_src = nl_attr_get_be32(a);
1362 break;
1363 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1364 tun->ip_dst = nl_attr_get_be32(a);
1365 break;
1366 case OVS_TUNNEL_KEY_ATTR_TOS:
1367 tun->ip_tos = nl_attr_get_u8(a);
1368 break;
1369 case OVS_TUNNEL_KEY_ATTR_TTL:
1370 tun->ip_ttl = nl_attr_get_u8(a);
1371 ttl = true;
1372 break;
1373 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1374 tun->flags |= FLOW_TNL_F_DONT_FRAGMENT;
1375 break;
1376 case OVS_TUNNEL_KEY_ATTR_CSUM:
1377 tun->flags |= FLOW_TNL_F_CSUM;
1378 break;
1379 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
1380 tun->tp_src = nl_attr_get_be16(a);
1381 break;
1382 case OVS_TUNNEL_KEY_ATTR_TP_DST:
1383 tun->tp_dst = nl_attr_get_be16(a);
1384 break;
1385 case OVS_TUNNEL_KEY_ATTR_OAM:
1386 tun->flags |= FLOW_TNL_F_OAM;
1387 break;
1388 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS: {
1389 static const struct nl_policy vxlan_opts_policy[] = {
1390 [OVS_VXLAN_EXT_GBP] = { .type = NL_A_U32 },
1391 };
1392 struct nlattr *ext[ARRAY_SIZE(vxlan_opts_policy)];
1393
1394 if (!nl_parse_nested(a, vxlan_opts_policy, ext, ARRAY_SIZE(ext))) {
1395 return ODP_FIT_ERROR;
1396 }
1397
1398 if (ext[OVS_VXLAN_EXT_GBP]) {
1399 uint32_t gbp = nl_attr_get_u32(ext[OVS_VXLAN_EXT_GBP]);
1400
1401 tun->gbp_id = htons(gbp & 0xFFFF);
1402 tun->gbp_flags = (gbp >> 16) & 0xFF;
1403 }
1404
1405 break;
1406 }
1407 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS: {
1408 if (parse_geneve_opts(a)) {
1409 return ODP_FIT_ERROR;
1410 }
1411 /* It is necessary to reproduce options exactly (including order)
1412 * so it's easiest to just echo them back. */
1413 unknown = true;
1414 break;
1415 }
1416 default:
1417 /* Allow this to show up as unexpected, if there are unknown
1418 * tunnel attribute, eventually resulting in ODP_FIT_TOO_MUCH. */
1419 unknown = true;
1420 break;
1421 }
1422 }
1423
1424 if (!ttl) {
1425 return ODP_FIT_ERROR;
1426 }
1427 if (unknown) {
1428 return ODP_FIT_TOO_MUCH;
1429 }
1430 return ODP_FIT_PERFECT;
1431 }
1432
1433 static void
1434 tun_key_to_attr(struct ofpbuf *a, const struct flow_tnl *tun_key)
1435 {
1436 size_t tun_key_ofs;
1437
1438 tun_key_ofs = nl_msg_start_nested(a, OVS_KEY_ATTR_TUNNEL);
1439
1440 /* tun_id != 0 without FLOW_TNL_F_KEY is valid if tun_key is a mask. */
1441 if (tun_key->tun_id || tun_key->flags & FLOW_TNL_F_KEY) {
1442 nl_msg_put_be64(a, OVS_TUNNEL_KEY_ATTR_ID, tun_key->tun_id);
1443 }
1444 if (tun_key->ip_src) {
1445 nl_msg_put_be32(a, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, tun_key->ip_src);
1446 }
1447 if (tun_key->ip_dst) {
1448 nl_msg_put_be32(a, OVS_TUNNEL_KEY_ATTR_IPV4_DST, tun_key->ip_dst);
1449 }
1450 if (tun_key->ip_tos) {
1451 nl_msg_put_u8(a, OVS_TUNNEL_KEY_ATTR_TOS, tun_key->ip_tos);
1452 }
1453 nl_msg_put_u8(a, OVS_TUNNEL_KEY_ATTR_TTL, tun_key->ip_ttl);
1454 if (tun_key->flags & FLOW_TNL_F_DONT_FRAGMENT) {
1455 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT);
1456 }
1457 if (tun_key->flags & FLOW_TNL_F_CSUM) {
1458 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_CSUM);
1459 }
1460 if (tun_key->tp_src) {
1461 nl_msg_put_be16(a, OVS_TUNNEL_KEY_ATTR_TP_SRC, tun_key->tp_src);
1462 }
1463 if (tun_key->tp_dst) {
1464 nl_msg_put_be16(a, OVS_TUNNEL_KEY_ATTR_TP_DST, tun_key->tp_dst);
1465 }
1466 if (tun_key->flags & FLOW_TNL_F_OAM) {
1467 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_OAM);
1468 }
1469 if (tun_key->gbp_flags || tun_key->gbp_id) {
1470 size_t vxlan_opts_ofs;
1471
1472 vxlan_opts_ofs = nl_msg_start_nested(a, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
1473 nl_msg_put_u32(a, OVS_VXLAN_EXT_GBP,
1474 (tun_key->gbp_flags << 16) | ntohs(tun_key->gbp_id));
1475 nl_msg_end_nested(a, vxlan_opts_ofs);
1476 }
1477
1478 nl_msg_end_nested(a, tun_key_ofs);
1479 }
1480
1481 static bool
1482 odp_mask_attr_is_wildcard(const struct nlattr *ma)
1483 {
1484 return is_all_zeros(nl_attr_get(ma), nl_attr_get_size(ma));
1485 }
1486
1487 static bool
1488 odp_mask_is_exact(enum ovs_key_attr attr, const void *mask, size_t size)
1489 {
1490 if (attr == OVS_KEY_ATTR_TCP_FLAGS) {
1491 return TCP_FLAGS(*(ovs_be16 *)mask) == TCP_FLAGS(OVS_BE16_MAX);
1492 }
1493 if (attr == OVS_KEY_ATTR_IPV6) {
1494 const struct ovs_key_ipv6 *ipv6_mask = mask;
1495
1496 return
1497 ((ipv6_mask->ipv6_label & htonl(IPV6_LABEL_MASK))
1498 == htonl(IPV6_LABEL_MASK))
1499 && ipv6_mask->ipv6_proto == UINT8_MAX
1500 && ipv6_mask->ipv6_tclass == UINT8_MAX
1501 && ipv6_mask->ipv6_hlimit == UINT8_MAX
1502 && ipv6_mask->ipv6_frag == UINT8_MAX
1503 && ipv6_mask_is_exact((const struct in6_addr *)ipv6_mask->ipv6_src)
1504 && ipv6_mask_is_exact((const struct in6_addr *)ipv6_mask->ipv6_dst);
1505 }
1506 if (attr == OVS_KEY_ATTR_TUNNEL) {
1507 const struct flow_tnl *tun_mask = mask;
1508
1509 return tun_mask->flags == FLOW_TNL_F_MASK
1510 && tun_mask->tun_id == OVS_BE64_MAX
1511 && tun_mask->ip_src == OVS_BE32_MAX
1512 && tun_mask->ip_dst == OVS_BE32_MAX
1513 && tun_mask->ip_tos == UINT8_MAX
1514 && tun_mask->ip_ttl == UINT8_MAX
1515 && tun_mask->tp_src == OVS_BE16_MAX
1516 && tun_mask->tp_dst == OVS_BE16_MAX
1517 && tun_mask->gbp_id == OVS_BE16_MAX
1518 && tun_mask->gbp_flags == UINT8_MAX;
1519 }
1520
1521 if (attr == OVS_KEY_ATTR_ARP) {
1522 /* ARP key has padding, ignore it. */
1523 BUILD_ASSERT_DECL(sizeof(struct ovs_key_arp) == 24);
1524 BUILD_ASSERT_DECL(offsetof(struct ovs_key_arp, arp_tha) == 10 + 6);
1525 size = offsetof(struct ovs_key_arp, arp_tha) + ETH_ADDR_LEN;
1526 ovs_assert(((uint16_t *)mask)[size/2] == 0);
1527 }
1528
1529 return is_all_ones(mask, size);
1530 }
1531
1532 static bool
1533 odp_mask_attr_is_exact(const struct nlattr *ma)
1534 {
1535 struct flow_tnl tun_mask;
1536 enum ovs_key_attr attr = nl_attr_type(ma);
1537 const void *mask;
1538 size_t size;
1539
1540 if (attr == OVS_KEY_ATTR_TUNNEL) {
1541 memset(&tun_mask, 0, sizeof tun_mask);
1542 odp_tun_key_from_attr(ma, &tun_mask);
1543 mask = &tun_mask;
1544 size = sizeof tun_mask;
1545 } else {
1546 mask = nl_attr_get(ma);
1547 size = nl_attr_get_size(ma);
1548 }
1549
1550 return odp_mask_is_exact(attr, mask, size);
1551 }
1552
1553 void
1554 odp_portno_names_set(struct hmap *portno_names, odp_port_t port_no,
1555 char *port_name)
1556 {
1557 struct odp_portno_names *odp_portno_names;
1558
1559 odp_portno_names = xmalloc(sizeof *odp_portno_names);
1560 odp_portno_names->port_no = port_no;
1561 odp_portno_names->name = xstrdup(port_name);
1562 hmap_insert(portno_names, &odp_portno_names->hmap_node,
1563 hash_odp_port(port_no));
1564 }
1565
1566 static char *
1567 odp_portno_names_get(const struct hmap *portno_names, odp_port_t port_no)
1568 {
1569 struct odp_portno_names *odp_portno_names;
1570
1571 HMAP_FOR_EACH_IN_BUCKET (odp_portno_names, hmap_node,
1572 hash_odp_port(port_no), portno_names) {
1573 if (odp_portno_names->port_no == port_no) {
1574 return odp_portno_names->name;
1575 }
1576 }
1577 return NULL;
1578 }
1579
1580 void
1581 odp_portno_names_destroy(struct hmap *portno_names)
1582 {
1583 struct odp_portno_names *odp_portno_names, *odp_portno_names_next;
1584 HMAP_FOR_EACH_SAFE (odp_portno_names, odp_portno_names_next,
1585 hmap_node, portno_names) {
1586 hmap_remove(portno_names, &odp_portno_names->hmap_node);
1587 free(odp_portno_names->name);
1588 free(odp_portno_names);
1589 }
1590 }
1591
1592 /* Format helpers. */
1593
1594 static void
1595 format_eth(struct ds *ds, const char *name, const uint8_t key[ETH_ADDR_LEN],
1596 const uint8_t (*mask)[ETH_ADDR_LEN], bool verbose)
1597 {
1598 bool mask_empty = mask && eth_addr_is_zero(*mask);
1599
1600 if (verbose || !mask_empty) {
1601 bool mask_full = !mask || eth_mask_is_exact(*mask);
1602
1603 if (mask_full) {
1604 ds_put_format(ds, "%s="ETH_ADDR_FMT",", name, ETH_ADDR_ARGS(key));
1605 } else {
1606 ds_put_format(ds, "%s=", name);
1607 eth_format_masked(key, *mask, ds);
1608 ds_put_char(ds, ',');
1609 }
1610 }
1611 }
1612
1613 static void
1614 format_be64(struct ds *ds, const char *name, ovs_be64 key,
1615 const ovs_be64 *mask, bool verbose)
1616 {
1617 bool mask_empty = mask && !*mask;
1618
1619 if (verbose || !mask_empty) {
1620 bool mask_full = !mask || *mask == OVS_BE64_MAX;
1621
1622 ds_put_format(ds, "%s=0x%"PRIx64, name, ntohll(key));
1623 if (!mask_full) { /* Partially masked. */
1624 ds_put_format(ds, "/%#"PRIx64, ntohll(*mask));
1625 }
1626 ds_put_char(ds, ',');
1627 }
1628 }
1629
1630 static void
1631 format_ipv4(struct ds *ds, const char *name, ovs_be32 key,
1632 const ovs_be32 *mask, bool verbose)
1633 {
1634 bool mask_empty = mask && !*mask;
1635
1636 if (verbose || !mask_empty) {
1637 bool mask_full = !mask || *mask == OVS_BE32_MAX;
1638
1639 ds_put_format(ds, "%s="IP_FMT, name, IP_ARGS(key));
1640 if (!mask_full) { /* Partially masked. */
1641 ds_put_format(ds, "/"IP_FMT, IP_ARGS(*mask));
1642 }
1643 ds_put_char(ds, ',');
1644 }
1645 }
1646
1647 static void
1648 format_ipv6(struct ds *ds, const char *name, const ovs_be32 key_[4],
1649 const ovs_be32 (*mask_)[4], bool verbose)
1650 {
1651 char buf[INET6_ADDRSTRLEN];
1652 const struct in6_addr *key = (const struct in6_addr *)key_;
1653 const struct in6_addr *mask = mask_ ? (const struct in6_addr *)*mask_
1654 : NULL;
1655 bool mask_empty = mask && ipv6_mask_is_any(mask);
1656
1657 if (verbose || !mask_empty) {
1658 bool mask_full = !mask || ipv6_mask_is_exact(mask);
1659
1660 inet_ntop(AF_INET6, key, buf, sizeof buf);
1661 ds_put_format(ds, "%s=%s", name, buf);
1662 if (!mask_full) { /* Partially masked. */
1663 inet_ntop(AF_INET6, mask, buf, sizeof buf);
1664 ds_put_format(ds, "/%s", buf);
1665 }
1666 ds_put_char(ds, ',');
1667 }
1668 }
1669
1670 static void
1671 format_ipv6_label(struct ds *ds, const char *name, ovs_be32 key,
1672 const ovs_be32 *mask, bool verbose)
1673 {
1674 bool mask_empty = mask && !*mask;
1675
1676 if (verbose || !mask_empty) {
1677 bool mask_full = !mask
1678 || (*mask & htonl(IPV6_LABEL_MASK)) == htonl(IPV6_LABEL_MASK);
1679
1680 ds_put_format(ds, "%s=%#"PRIx32, name, ntohl(key));
1681 if (!mask_full) { /* Partially masked. */
1682 ds_put_format(ds, "/%#"PRIx32, ntohl(*mask));
1683 }
1684 ds_put_char(ds, ',');
1685 }
1686 }
1687
1688 static void
1689 format_u8x(struct ds *ds, const char *name, uint8_t key,
1690 const uint8_t *mask, bool verbose)
1691 {
1692 bool mask_empty = mask && !*mask;
1693
1694 if (verbose || !mask_empty) {
1695 bool mask_full = !mask || *mask == UINT8_MAX;
1696
1697 ds_put_format(ds, "%s=%#"PRIx8, name, key);
1698 if (!mask_full) { /* Partially masked. */
1699 ds_put_format(ds, "/%#"PRIx8, *mask);
1700 }
1701 ds_put_char(ds, ',');
1702 }
1703 }
1704
1705 static void
1706 format_u8u(struct ds *ds, const char *name, uint8_t key,
1707 const uint8_t *mask, bool verbose)
1708 {
1709 bool mask_empty = mask && !*mask;
1710
1711 if (verbose || !mask_empty) {
1712 bool mask_full = !mask || *mask == UINT8_MAX;
1713
1714 ds_put_format(ds, "%s=%"PRIu8, name, key);
1715 if (!mask_full) { /* Partially masked. */
1716 ds_put_format(ds, "/%#"PRIx8, *mask);
1717 }
1718 ds_put_char(ds, ',');
1719 }
1720 }
1721
1722 static void
1723 format_be16(struct ds *ds, const char *name, ovs_be16 key,
1724 const ovs_be16 *mask, bool verbose)
1725 {
1726 bool mask_empty = mask && !*mask;
1727
1728 if (verbose || !mask_empty) {
1729 bool mask_full = !mask || *mask == OVS_BE16_MAX;
1730
1731 ds_put_format(ds, "%s=%"PRIu16, name, ntohs(key));
1732 if (!mask_full) { /* Partially masked. */
1733 ds_put_format(ds, "/%#"PRIx16, ntohs(*mask));
1734 }
1735 ds_put_char(ds, ',');
1736 }
1737 }
1738
1739 static void
1740 format_tun_flags(struct ds *ds, const char *name, uint16_t key,
1741 const uint16_t *mask, bool verbose)
1742 {
1743 bool mask_empty = mask && !*mask;
1744
1745 if (verbose || !mask_empty) {
1746 bool mask_full = !mask || (*mask & FLOW_TNL_F_MASK) == FLOW_TNL_F_MASK;
1747
1748 ds_put_cstr(ds, name);
1749 ds_put_char(ds, '(');
1750 if (!mask_full) { /* Partially masked. */
1751 format_flags_masked(ds, NULL, flow_tun_flag_to_string, key, *mask);
1752 } else { /* Fully masked. */
1753 format_flags(ds, flow_tun_flag_to_string, key, ',');
1754 }
1755 ds_put_cstr(ds, "),");
1756 }
1757 }
1758
1759 static void
1760 format_frag(struct ds *ds, const char *name, uint8_t key,
1761 const uint8_t *mask, bool verbose)
1762 {
1763 bool mask_empty = mask && !*mask;
1764
1765 /* ODP frag is an enumeration field; partial masks are not meaningful. */
1766 if (verbose || !mask_empty) {
1767 bool mask_full = !mask || *mask == UINT8_MAX;
1768
1769 if (!mask_full) { /* Partially masked. */
1770 ds_put_format(ds, "error: partial mask not supported for frag (%#"
1771 PRIx8"),", *mask);
1772 } else {
1773 ds_put_format(ds, "%s=%s,", name, ovs_frag_type_to_string(key));
1774 }
1775 }
1776 }
1777
1778 #define MASK(PTR, FIELD) PTR ? &PTR->FIELD : NULL
1779
1780 static void
1781 format_odp_key_attr(const struct nlattr *a, const struct nlattr *ma,
1782 const struct hmap *portno_names, struct ds *ds,
1783 bool verbose)
1784 {
1785 enum ovs_key_attr attr = nl_attr_type(a);
1786 char namebuf[OVS_KEY_ATTR_BUFSIZE];
1787 int expected_len;
1788 bool is_exact;
1789
1790 is_exact = ma ? odp_mask_attr_is_exact(ma) : true;
1791
1792 ds_put_cstr(ds, ovs_key_attr_to_string(attr, namebuf, sizeof namebuf));
1793
1794 {
1795 expected_len = odp_flow_key_attr_len(nl_attr_type(a));
1796 if (expected_len != -2) {
1797 bool bad_key_len = nl_attr_get_size(a) != expected_len;
1798 bool bad_mask_len = ma && nl_attr_get_size(ma) != expected_len;
1799
1800 if (bad_key_len || bad_mask_len) {
1801 if (bad_key_len) {
1802 ds_put_format(ds, "(bad key length %"PRIuSIZE", expected %d)(",
1803 nl_attr_get_size(a), expected_len);
1804 }
1805 format_generic_odp_key(a, ds);
1806 if (ma) {
1807 ds_put_char(ds, '/');
1808 if (bad_mask_len) {
1809 ds_put_format(ds, "(bad mask length %"PRIuSIZE", expected %d)(",
1810 nl_attr_get_size(ma), expected_len);
1811 }
1812 format_generic_odp_key(ma, ds);
1813 }
1814 ds_put_char(ds, ')');
1815 return;
1816 }
1817 }
1818 }
1819
1820 ds_put_char(ds, '(');
1821 switch (attr) {
1822 case OVS_KEY_ATTR_ENCAP:
1823 if (ma && nl_attr_get_size(ma) && nl_attr_get_size(a)) {
1824 odp_flow_format(nl_attr_get(a), nl_attr_get_size(a),
1825 nl_attr_get(ma), nl_attr_get_size(ma), NULL, ds,
1826 verbose);
1827 } else if (nl_attr_get_size(a)) {
1828 odp_flow_format(nl_attr_get(a), nl_attr_get_size(a), NULL, 0, NULL,
1829 ds, verbose);
1830 }
1831 break;
1832
1833 case OVS_KEY_ATTR_PRIORITY:
1834 case OVS_KEY_ATTR_SKB_MARK:
1835 case OVS_KEY_ATTR_DP_HASH:
1836 case OVS_KEY_ATTR_RECIRC_ID:
1837 ds_put_format(ds, "%#"PRIx32, nl_attr_get_u32(a));
1838 if (!is_exact) {
1839 ds_put_format(ds, "/%#"PRIx32, nl_attr_get_u32(ma));
1840 }
1841 break;
1842
1843 case OVS_KEY_ATTR_TUNNEL: {
1844 struct flow_tnl key, mask_;
1845 struct flow_tnl *mask = ma ? &mask_ : NULL;
1846
1847 if (mask) {
1848 memset(mask, 0, sizeof *mask);
1849 odp_tun_key_from_attr(ma, mask);
1850 }
1851 memset(&key, 0, sizeof key);
1852 if (odp_tun_key_from_attr(a, &key) == ODP_FIT_ERROR) {
1853 ds_put_format(ds, "error");
1854 return;
1855 }
1856 format_be64(ds, "tun_id", key.tun_id, MASK(mask, tun_id), verbose);
1857 format_ipv4(ds, "src", key.ip_src, MASK(mask, ip_src), verbose);
1858 format_ipv4(ds, "dst", key.ip_dst, MASK(mask, ip_dst), verbose);
1859 format_u8x(ds, "tos", key.ip_tos, MASK(mask, ip_tos), verbose);
1860 format_u8u(ds, "ttl", key.ip_ttl, MASK(mask, ip_ttl), verbose);
1861 format_be16(ds, "tp_src", key.tp_src, MASK(mask, tp_src), verbose);
1862 format_be16(ds, "tp_dst", key.tp_dst, MASK(mask, tp_dst), verbose);
1863 format_be16(ds, "gbp_id", key.gbp_id, MASK(mask, gbp_id), verbose);
1864 format_u8x(ds, "gbp_flags", key.gbp_flags, MASK(mask, gbp_flags), verbose);
1865 format_tun_flags(ds, "flags", key.flags, MASK(mask, flags), verbose);
1866 ds_chomp(ds, ',');
1867 break;
1868 }
1869 case OVS_KEY_ATTR_IN_PORT:
1870 if (portno_names && verbose && is_exact) {
1871 char *name = odp_portno_names_get(portno_names,
1872 u32_to_odp(nl_attr_get_u32(a)));
1873 if (name) {
1874 ds_put_format(ds, "%s", name);
1875 } else {
1876 ds_put_format(ds, "%"PRIu32, nl_attr_get_u32(a));
1877 }
1878 } else {
1879 ds_put_format(ds, "%"PRIu32, nl_attr_get_u32(a));
1880 if (!is_exact) {
1881 ds_put_format(ds, "/%#"PRIx32, nl_attr_get_u32(ma));
1882 }
1883 }
1884 break;
1885
1886 case OVS_KEY_ATTR_ETHERNET: {
1887 const struct ovs_key_ethernet *mask = ma ? nl_attr_get(ma) : NULL;
1888 const struct ovs_key_ethernet *key = nl_attr_get(a);
1889
1890 format_eth(ds, "src", key->eth_src, MASK(mask, eth_src), verbose);
1891 format_eth(ds, "dst", key->eth_dst, MASK(mask, eth_dst), verbose);
1892 ds_chomp(ds, ',');
1893 break;
1894 }
1895 case OVS_KEY_ATTR_VLAN:
1896 format_vlan_tci(ds, nl_attr_get_be16(a),
1897 ma ? nl_attr_get_be16(ma) : OVS_BE16_MAX, verbose);
1898 break;
1899
1900 case OVS_KEY_ATTR_MPLS: {
1901 const struct ovs_key_mpls *mpls_key = nl_attr_get(a);
1902 const struct ovs_key_mpls *mpls_mask = NULL;
1903 size_t size = nl_attr_get_size(a);
1904
1905 if (!size || size % sizeof *mpls_key) {
1906 ds_put_format(ds, "(bad key length %"PRIuSIZE")", size);
1907 return;
1908 }
1909 if (!is_exact) {
1910 mpls_mask = nl_attr_get(ma);
1911 if (size != nl_attr_get_size(ma)) {
1912 ds_put_format(ds, "(key length %"PRIuSIZE" != "
1913 "mask length %"PRIuSIZE")",
1914 size, nl_attr_get_size(ma));
1915 return;
1916 }
1917 }
1918 format_mpls(ds, mpls_key, mpls_mask, size / sizeof *mpls_key);
1919 break;
1920 }
1921 case OVS_KEY_ATTR_ETHERTYPE:
1922 ds_put_format(ds, "0x%04"PRIx16, ntohs(nl_attr_get_be16(a)));
1923 if (!is_exact) {
1924 ds_put_format(ds, "/0x%04"PRIx16, ntohs(nl_attr_get_be16(ma)));
1925 }
1926 break;
1927
1928 case OVS_KEY_ATTR_IPV4: {
1929 const struct ovs_key_ipv4 *key = nl_attr_get(a);
1930 const struct ovs_key_ipv4 *mask = ma ? nl_attr_get(ma) : NULL;
1931
1932 format_ipv4(ds, "src", key->ipv4_src, MASK(mask, ipv4_src), verbose);
1933 format_ipv4(ds, "dst", key->ipv4_dst, MASK(mask, ipv4_dst), verbose);
1934 format_u8u(ds, "proto", key->ipv4_proto, MASK(mask, ipv4_proto),
1935 verbose);
1936 format_u8x(ds, "tos", key->ipv4_tos, MASK(mask, ipv4_tos), verbose);
1937 format_u8u(ds, "ttl", key->ipv4_ttl, MASK(mask, ipv4_ttl), verbose);
1938 format_frag(ds, "frag", key->ipv4_frag, MASK(mask, ipv4_frag),
1939 verbose);
1940 ds_chomp(ds, ',');
1941 break;
1942 }
1943 case OVS_KEY_ATTR_IPV6: {
1944 const struct ovs_key_ipv6 *key = nl_attr_get(a);
1945 const struct ovs_key_ipv6 *mask = ma ? nl_attr_get(ma) : NULL;
1946
1947 format_ipv6(ds, "src", key->ipv6_src, MASK(mask, ipv6_src), verbose);
1948 format_ipv6(ds, "dst", key->ipv6_dst, MASK(mask, ipv6_dst), verbose);
1949 format_ipv6_label(ds, "label", key->ipv6_label, MASK(mask, ipv6_label),
1950 verbose);
1951 format_u8u(ds, "proto", key->ipv6_proto, MASK(mask, ipv6_proto),
1952 verbose);
1953 format_u8x(ds, "tclass", key->ipv6_tclass, MASK(mask, ipv6_tclass),
1954 verbose);
1955 format_u8u(ds, "hlimit", key->ipv6_hlimit, MASK(mask, ipv6_hlimit),
1956 verbose);
1957 format_frag(ds, "frag", key->ipv6_frag, MASK(mask, ipv6_frag),
1958 verbose);
1959 ds_chomp(ds, ',');
1960 break;
1961 }
1962 /* These have the same structure and format. */
1963 case OVS_KEY_ATTR_TCP:
1964 case OVS_KEY_ATTR_UDP:
1965 case OVS_KEY_ATTR_SCTP: {
1966 const struct ovs_key_tcp *key = nl_attr_get(a);
1967 const struct ovs_key_tcp *mask = ma ? nl_attr_get(ma) : NULL;
1968
1969 format_be16(ds, "src", key->tcp_src, MASK(mask, tcp_src), verbose);
1970 format_be16(ds, "dst", key->tcp_dst, MASK(mask, tcp_dst), verbose);
1971 ds_chomp(ds, ',');
1972 break;
1973 }
1974 case OVS_KEY_ATTR_TCP_FLAGS:
1975 if (!is_exact) {
1976 format_flags_masked(ds, NULL, packet_tcp_flag_to_string,
1977 ntohs(nl_attr_get_be16(a)),
1978 ntohs(nl_attr_get_be16(ma)));
1979 } else {
1980 format_flags(ds, packet_tcp_flag_to_string,
1981 ntohs(nl_attr_get_be16(a)), ',');
1982 }
1983 break;
1984
1985 case OVS_KEY_ATTR_ICMP: {
1986 const struct ovs_key_icmp *key = nl_attr_get(a);
1987 const struct ovs_key_icmp *mask = ma ? nl_attr_get(ma) : NULL;
1988
1989 format_u8u(ds, "type", key->icmp_type, MASK(mask, icmp_type), verbose);
1990 format_u8u(ds, "code", key->icmp_code, MASK(mask, icmp_code), verbose);
1991 ds_chomp(ds, ',');
1992 break;
1993 }
1994 case OVS_KEY_ATTR_ICMPV6: {
1995 const struct ovs_key_icmpv6 *key = nl_attr_get(a);
1996 const struct ovs_key_icmpv6 *mask = ma ? nl_attr_get(ma) : NULL;
1997
1998 format_u8u(ds, "type", key->icmpv6_type, MASK(mask, icmpv6_type),
1999 verbose);
2000 format_u8u(ds, "code", key->icmpv6_code, MASK(mask, icmpv6_code),
2001 verbose);
2002 ds_chomp(ds, ',');
2003 break;
2004 }
2005 case OVS_KEY_ATTR_ARP: {
2006 const struct ovs_key_arp *mask = ma ? nl_attr_get(ma) : NULL;
2007 const struct ovs_key_arp *key = nl_attr_get(a);
2008
2009 format_ipv4(ds, "sip", key->arp_sip, MASK(mask, arp_sip), verbose);
2010 format_ipv4(ds, "tip", key->arp_tip, MASK(mask, arp_tip), verbose);
2011 format_be16(ds, "op", key->arp_op, MASK(mask, arp_op), verbose);
2012 format_eth(ds, "sha", key->arp_sha, MASK(mask, arp_sha), verbose);
2013 format_eth(ds, "tha", key->arp_tha, MASK(mask, arp_tha), verbose);
2014 ds_chomp(ds, ',');
2015 break;
2016 }
2017 case OVS_KEY_ATTR_ND: {
2018 const struct ovs_key_nd *mask = ma ? nl_attr_get(ma) : NULL;
2019 const struct ovs_key_nd *key = nl_attr_get(a);
2020
2021 format_ipv6(ds, "target", key->nd_target, MASK(mask, nd_target),
2022 verbose);
2023 format_eth(ds, "sll", key->nd_sll, MASK(mask, nd_sll), verbose);
2024 format_eth(ds, "tll", key->nd_tll, MASK(mask, nd_tll), verbose);
2025
2026 ds_chomp(ds, ',');
2027 break;
2028 }
2029 case OVS_KEY_ATTR_UNSPEC:
2030 case __OVS_KEY_ATTR_MAX:
2031 default:
2032 format_generic_odp_key(a, ds);
2033 if (!is_exact) {
2034 ds_put_char(ds, '/');
2035 format_generic_odp_key(ma, ds);
2036 }
2037 break;
2038 }
2039 ds_put_char(ds, ')');
2040 }
2041
2042 static struct nlattr *
2043 generate_all_wildcard_mask(struct ofpbuf *ofp, const struct nlattr *key)
2044 {
2045 const struct nlattr *a;
2046 unsigned int left;
2047 int type = nl_attr_type(key);
2048 int size = nl_attr_get_size(key);
2049
2050 if (odp_flow_key_attr_len(type) >=0) {
2051 nl_msg_put_unspec_zero(ofp, type, size);
2052 } else {
2053 size_t nested_mask;
2054
2055 nested_mask = nl_msg_start_nested(ofp, type);
2056 NL_ATTR_FOR_EACH(a, left, key, nl_attr_get_size(key)) {
2057 generate_all_wildcard_mask(ofp, nl_attr_get(a));
2058 }
2059 nl_msg_end_nested(ofp, nested_mask);
2060 }
2061
2062 return ofp->base;
2063 }
2064
2065 int
2066 odp_ufid_from_string(const char *s_, ovs_u128 *ufid)
2067 {
2068 const char *s = s_;
2069
2070 if (ovs_scan(s, "ufid:")) {
2071 size_t n;
2072
2073 s += 5;
2074 if (ovs_scan(s, "0x")) {
2075 s += 2;
2076 }
2077
2078 n = strspn(s, "0123456789abcdefABCDEF");
2079 if (n != 32) {
2080 return -EINVAL;
2081 }
2082
2083 if (!ovs_scan(s, "%16"SCNx64"%16"SCNx64, &ufid->u64.hi,
2084 &ufid->u64.lo)) {
2085 return -EINVAL;
2086 }
2087 s += n;
2088 s += strspn(s, delimiters);
2089
2090 return s - s_;
2091 }
2092
2093 return 0;
2094 }
2095
2096 void
2097 odp_format_ufid(const ovs_u128 *ufid, struct ds *ds)
2098 {
2099 ds_put_format(ds, "ufid:%016"PRIx64"%016"PRIx64, ufid->u64.hi,
2100 ufid->u64.lo);
2101 }
2102
2103 /* Appends to 'ds' a string representation of the 'key_len' bytes of
2104 * OVS_KEY_ATTR_* attributes in 'key'. If non-null, additionally formats the
2105 * 'mask_len' bytes of 'mask' which apply to 'key'. If 'portno_names' is
2106 * non-null and 'verbose' is true, translates odp port number to its name. */
2107 void
2108 odp_flow_format(const struct nlattr *key, size_t key_len,
2109 const struct nlattr *mask, size_t mask_len,
2110 const struct hmap *portno_names, struct ds *ds, bool verbose)
2111 {
2112 if (key_len) {
2113 const struct nlattr *a;
2114 unsigned int left;
2115 bool has_ethtype_key = false;
2116 const struct nlattr *ma = NULL;
2117 struct ofpbuf ofp;
2118 bool first_field = true;
2119
2120 ofpbuf_init(&ofp, 100);
2121 NL_ATTR_FOR_EACH (a, left, key, key_len) {
2122 bool is_nested_attr;
2123 bool is_wildcard = false;
2124 int attr_type = nl_attr_type(a);
2125
2126 if (attr_type == OVS_KEY_ATTR_ETHERTYPE) {
2127 has_ethtype_key = true;
2128 }
2129
2130 is_nested_attr = (odp_flow_key_attr_len(attr_type) == -2);
2131
2132 if (mask && mask_len) {
2133 ma = nl_attr_find__(mask, mask_len, nl_attr_type(a));
2134 is_wildcard = ma ? odp_mask_attr_is_wildcard(ma) : true;
2135 }
2136
2137 if (verbose || !is_wildcard || is_nested_attr) {
2138 if (is_wildcard && !ma) {
2139 ma = generate_all_wildcard_mask(&ofp, a);
2140 }
2141 if (!first_field) {
2142 ds_put_char(ds, ',');
2143 }
2144 format_odp_key_attr(a, ma, portno_names, ds, verbose);
2145 first_field = false;
2146 }
2147 ofpbuf_clear(&ofp);
2148 }
2149 ofpbuf_uninit(&ofp);
2150
2151 if (left) {
2152 int i;
2153
2154 if (left == key_len) {
2155 ds_put_cstr(ds, "<empty>");
2156 }
2157 ds_put_format(ds, ",***%u leftover bytes*** (", left);
2158 for (i = 0; i < left; i++) {
2159 ds_put_format(ds, "%02x", ((const uint8_t *) a)[i]);
2160 }
2161 ds_put_char(ds, ')');
2162 }
2163 if (!has_ethtype_key) {
2164 ma = nl_attr_find__(mask, mask_len, OVS_KEY_ATTR_ETHERTYPE);
2165 if (ma) {
2166 ds_put_format(ds, ",eth_type(0/0x%04"PRIx16")",
2167 ntohs(nl_attr_get_be16(ma)));
2168 }
2169 }
2170 } else {
2171 ds_put_cstr(ds, "<empty>");
2172 }
2173 }
2174
2175 /* Appends to 'ds' a string representation of the 'key_len' bytes of
2176 * OVS_KEY_ATTR_* attributes in 'key'. */
2177 void
2178 odp_flow_key_format(const struct nlattr *key,
2179 size_t key_len, struct ds *ds)
2180 {
2181 odp_flow_format(key, key_len, NULL, 0, NULL, ds, true);
2182 }
2183
2184 static bool
2185 ovs_frag_type_from_string(const char *s, enum ovs_frag_type *type)
2186 {
2187 if (!strcasecmp(s, "no")) {
2188 *type = OVS_FRAG_TYPE_NONE;
2189 } else if (!strcasecmp(s, "first")) {
2190 *type = OVS_FRAG_TYPE_FIRST;
2191 } else if (!strcasecmp(s, "later")) {
2192 *type = OVS_FRAG_TYPE_LATER;
2193 } else {
2194 return false;
2195 }
2196 return true;
2197 }
2198
2199 /* Parsing. */
2200
2201 static int
2202 scan_eth(const char *s, uint8_t (*key)[ETH_ADDR_LEN],
2203 uint8_t (*mask)[ETH_ADDR_LEN])
2204 {
2205 int n;
2206
2207 if (ovs_scan(s, ETH_ADDR_SCAN_FMT"%n", ETH_ADDR_SCAN_ARGS(*key), &n)) {
2208 int len = n;
2209
2210 if (mask) {
2211 if (ovs_scan(s + len, "/"ETH_ADDR_SCAN_FMT"%n",
2212 ETH_ADDR_SCAN_ARGS(*mask), &n)) {
2213 len += n;
2214 } else {
2215 memset(mask, 0xff, sizeof *mask);
2216 }
2217 }
2218 return len;
2219 }
2220 return 0;
2221 }
2222
2223 static int
2224 scan_ipv4(const char *s, ovs_be32 *key, ovs_be32 *mask)
2225 {
2226 int n;
2227
2228 if (ovs_scan(s, IP_SCAN_FMT"%n", IP_SCAN_ARGS(key), &n)) {
2229 int len = n;
2230
2231 if (mask) {
2232 if (ovs_scan(s + len, "/"IP_SCAN_FMT"%n",
2233 IP_SCAN_ARGS(mask), &n)) {
2234 len += n;
2235 } else {
2236 *mask = OVS_BE32_MAX;
2237 }
2238 }
2239 return len;
2240 }
2241 return 0;
2242 }
2243
2244 static int
2245 scan_ipv6(const char *s, ovs_be32 (*key)[4], ovs_be32 (*mask)[4])
2246 {
2247 int n;
2248 char ipv6_s[IPV6_SCAN_LEN + 1];
2249
2250 if (ovs_scan(s, IPV6_SCAN_FMT"%n", ipv6_s, &n)
2251 && inet_pton(AF_INET6, ipv6_s, key) == 1) {
2252 int len = n;
2253
2254 if (mask) {
2255 if (ovs_scan(s + len, "/"IPV6_SCAN_FMT"%n", ipv6_s, &n)
2256 && inet_pton(AF_INET6, ipv6_s, mask) == 1) {
2257 len += n;
2258 } else {
2259 memset(mask, 0xff, sizeof *mask);
2260 }
2261 }
2262 return len;
2263 }
2264 return 0;
2265 }
2266
2267 static int
2268 scan_ipv6_label(const char *s, ovs_be32 *key, ovs_be32 *mask)
2269 {
2270 int key_, mask_;
2271 int n;
2272
2273 if (ovs_scan(s, "%i%n", &key_, &n)
2274 && (key_ & ~IPV6_LABEL_MASK) == 0) {
2275 int len = n;
2276
2277 *key = htonl(key_);
2278 if (mask) {
2279 if (ovs_scan(s + len, "/%i%n", &mask_, &n)
2280 && (mask_ & ~IPV6_LABEL_MASK) == 0) {
2281 len += n;
2282 *mask = htonl(mask_);
2283 } else {
2284 *mask = htonl(IPV6_LABEL_MASK);
2285 }
2286 }
2287 return len;
2288 }
2289 return 0;
2290 }
2291
2292 static int
2293 scan_u8(const char *s, uint8_t *key, uint8_t *mask)
2294 {
2295 int n;
2296
2297 if (ovs_scan(s, "%"SCNi8"%n", key, &n)) {
2298 int len = n;
2299
2300 if (mask) {
2301 if (ovs_scan(s + len, "/%"SCNi8"%n", mask, &n)) {
2302 len += n;
2303 } else {
2304 *mask = UINT8_MAX;
2305 }
2306 }
2307 return len;
2308 }
2309 return 0;
2310 }
2311
2312 static int
2313 scan_u32(const char *s, uint32_t *key, uint32_t *mask)
2314 {
2315 int n;
2316
2317 if (ovs_scan(s, "%"SCNi32"%n", key, &n)) {
2318 int len = n;
2319
2320 if (mask) {
2321 if (ovs_scan(s + len, "/%"SCNi32"%n", mask, &n)) {
2322 len += n;
2323 } else {
2324 *mask = UINT32_MAX;
2325 }
2326 }
2327 return len;
2328 }
2329 return 0;
2330 }
2331
2332 static int
2333 scan_be16(const char *s, ovs_be16 *key, ovs_be16 *mask)
2334 {
2335 uint16_t key_, mask_;
2336 int n;
2337
2338 if (ovs_scan(s, "%"SCNi16"%n", &key_, &n)) {
2339 int len = n;
2340
2341 *key = htons(key_);
2342 if (mask) {
2343 if (ovs_scan(s + len, "/%"SCNi16"%n", &mask_, &n)) {
2344 len += n;
2345 *mask = htons(mask_);
2346 } else {
2347 *mask = OVS_BE16_MAX;
2348 }
2349 }
2350 return len;
2351 }
2352 return 0;
2353 }
2354
2355 static int
2356 scan_be64(const char *s, ovs_be64 *key, ovs_be64 *mask)
2357 {
2358 uint64_t key_, mask_;
2359 int n;
2360
2361 if (ovs_scan(s, "%"SCNi64"%n", &key_, &n)) {
2362 int len = n;
2363
2364 *key = htonll(key_);
2365 if (mask) {
2366 if (ovs_scan(s + len, "/%"SCNi64"%n", &mask_, &n)) {
2367 len += n;
2368 *mask = htonll(mask_);
2369 } else {
2370 *mask = OVS_BE64_MAX;
2371 }
2372 }
2373 return len;
2374 }
2375 return 0;
2376 }
2377
2378 static int
2379 scan_tun_flags(const char *s, uint16_t *key, uint16_t *mask)
2380 {
2381 uint32_t flags, fmask;
2382 int n;
2383
2384 n = parse_flags(s, flow_tun_flag_to_string, &flags,
2385 FLOW_TNL_F_MASK, mask ? &fmask : NULL);
2386 if (n >= 0 && s[n] == ')') {
2387 *key = flags;
2388 if (mask) {
2389 *mask = fmask;
2390 }
2391 return n + 1;
2392 }
2393 return 0;
2394 }
2395
2396 static int
2397 scan_tcp_flags(const char *s, ovs_be16 *key, ovs_be16 *mask)
2398 {
2399 uint32_t flags, fmask;
2400 int n;
2401
2402 n = parse_flags(s, packet_tcp_flag_to_string, &flags,
2403 TCP_FLAGS(OVS_BE16_MAX), mask ? &fmask : NULL);
2404 if (n >= 0) {
2405 *key = htons(flags);
2406 if (mask) {
2407 *mask = htons(fmask);
2408 }
2409 return n;
2410 }
2411 return 0;
2412 }
2413
2414 static int
2415 scan_frag(const char *s, uint8_t *key, uint8_t *mask)
2416 {
2417 int n;
2418 char frag[8];
2419 enum ovs_frag_type frag_type;
2420
2421 if (ovs_scan(s, "%7[a-z]%n", frag, &n)
2422 && ovs_frag_type_from_string(frag, &frag_type)) {
2423 int len = n;
2424
2425 *key = frag_type;
2426 if (mask) {
2427 *mask = UINT8_MAX;
2428 }
2429 return len;
2430 }
2431 return 0;
2432 }
2433
2434 static int
2435 scan_port(const char *s, uint32_t *key, uint32_t *mask,
2436 const struct simap *port_names)
2437 {
2438 int n;
2439
2440 if (ovs_scan(s, "%"SCNi32"%n", key, &n)) {
2441 int len = n;
2442
2443 if (mask) {
2444 if (ovs_scan(s + len, "/%"SCNi32"%n", mask, &n)) {
2445 len += n;
2446 } else {
2447 *mask = UINT32_MAX;
2448 }
2449 }
2450 return len;
2451 } else if (port_names) {
2452 const struct simap_node *node;
2453 int len;
2454
2455 len = strcspn(s, ")");
2456 node = simap_find_len(port_names, s, len);
2457 if (node) {
2458 *key = node->data;
2459
2460 if (mask) {
2461 *mask = UINT32_MAX;
2462 }
2463 return len;
2464 }
2465 }
2466 return 0;
2467 }
2468
2469 /* Helper for vlan parsing. */
2470 struct ovs_key_vlan__ {
2471 ovs_be16 tci;
2472 };
2473
2474 static bool
2475 set_be16_bf(ovs_be16 *bf, uint8_t bits, uint8_t offset, uint16_t value)
2476 {
2477 const uint16_t mask = ((1U << bits) - 1) << offset;
2478
2479 if (value >> bits) {
2480 return false;
2481 }
2482
2483 *bf = htons((ntohs(*bf) & ~mask) | (value << offset));
2484 return true;
2485 }
2486
2487 static int
2488 scan_be16_bf(const char *s, ovs_be16 *key, ovs_be16 *mask, uint8_t bits,
2489 uint8_t offset)
2490 {
2491 uint16_t key_, mask_;
2492 int n;
2493
2494 if (ovs_scan(s, "%"SCNi16"%n", &key_, &n)) {
2495 int len = n;
2496
2497 if (set_be16_bf(key, bits, offset, key_)) {
2498 if (mask) {
2499 if (ovs_scan(s + len, "/%"SCNi16"%n", &mask_, &n)) {
2500 len += n;
2501
2502 if (!set_be16_bf(mask, bits, offset, mask_)) {
2503 return 0;
2504 }
2505 } else {
2506 *mask |= htons(((1U << bits) - 1) << offset);
2507 }
2508 }
2509 return len;
2510 }
2511 }
2512 return 0;
2513 }
2514
2515 static int
2516 scan_vid(const char *s, ovs_be16 *key, ovs_be16 *mask)
2517 {
2518 return scan_be16_bf(s, key, mask, 12, VLAN_VID_SHIFT);
2519 }
2520
2521 static int
2522 scan_pcp(const char *s, ovs_be16 *key, ovs_be16 *mask)
2523 {
2524 return scan_be16_bf(s, key, mask, 3, VLAN_PCP_SHIFT);
2525 }
2526
2527 static int
2528 scan_cfi(const char *s, ovs_be16 *key, ovs_be16 *mask)
2529 {
2530 return scan_be16_bf(s, key, mask, 1, VLAN_CFI_SHIFT);
2531 }
2532
2533 /* For MPLS. */
2534 static bool
2535 set_be32_bf(ovs_be32 *bf, uint8_t bits, uint8_t offset, uint32_t value)
2536 {
2537 const uint32_t mask = ((1U << bits) - 1) << offset;
2538
2539 if (value >> bits) {
2540 return false;
2541 }
2542
2543 *bf = htonl((ntohl(*bf) & ~mask) | (value << offset));
2544 return true;
2545 }
2546
2547 static int
2548 scan_be32_bf(const char *s, ovs_be32 *key, ovs_be32 *mask, uint8_t bits,
2549 uint8_t offset)
2550 {
2551 uint32_t key_, mask_;
2552 int n;
2553
2554 if (ovs_scan(s, "%"SCNi32"%n", &key_, &n)) {
2555 int len = n;
2556
2557 if (set_be32_bf(key, bits, offset, key_)) {
2558 if (mask) {
2559 if (ovs_scan(s + len, "/%"SCNi32"%n", &mask_, &n)) {
2560 len += n;
2561
2562 if (!set_be32_bf(mask, bits, offset, mask_)) {
2563 return 0;
2564 }
2565 } else {
2566 *mask |= htonl(((1U << bits) - 1) << offset);
2567 }
2568 }
2569 return len;
2570 }
2571 }
2572 return 0;
2573 }
2574
2575 static int
2576 scan_mpls_label(const char *s, ovs_be32 *key, ovs_be32 *mask)
2577 {
2578 return scan_be32_bf(s, key, mask, 20, MPLS_LABEL_SHIFT);
2579 }
2580
2581 static int
2582 scan_mpls_tc(const char *s, ovs_be32 *key, ovs_be32 *mask)
2583 {
2584 return scan_be32_bf(s, key, mask, 3, MPLS_TC_SHIFT);
2585 }
2586
2587 static int
2588 scan_mpls_ttl(const char *s, ovs_be32 *key, ovs_be32 *mask)
2589 {
2590 return scan_be32_bf(s, key, mask, 8, MPLS_TTL_SHIFT);
2591 }
2592
2593 static int
2594 scan_mpls_bos(const char *s, ovs_be32 *key, ovs_be32 *mask)
2595 {
2596 return scan_be32_bf(s, key, mask, 1, MPLS_BOS_SHIFT);
2597 }
2598
2599 /* ATTR is compile-time constant, so only the case with correct data type
2600 * will be used. However, the compiler complains about the data type for
2601 * the other cases, so we must cast to make the compiler silent. */
2602 #define SCAN_PUT_ATTR(BUF, ATTR, DATA) \
2603 if ((ATTR) == OVS_KEY_ATTR_TUNNEL) { \
2604 tun_key_to_attr(BUF, (const struct flow_tnl *)(void *)&(DATA)); \
2605 } else { \
2606 nl_msg_put_unspec(BUF, ATTR, &(DATA), sizeof (DATA)); \
2607 }
2608
2609 #define SCAN_IF(NAME) \
2610 if (strncmp(s, NAME, strlen(NAME)) == 0) { \
2611 const char *start = s; \
2612 int len; \
2613 \
2614 s += strlen(NAME)
2615
2616 /* Usually no special initialization is needed. */
2617 #define SCAN_BEGIN(NAME, TYPE) \
2618 SCAN_IF(NAME); \
2619 TYPE skey, smask; \
2620 memset(&skey, 0, sizeof skey); \
2621 memset(&smask, 0, sizeof smask); \
2622 do { \
2623 len = 0;
2624
2625 /* Init as fully-masked as mask will not be scanned. */
2626 #define SCAN_BEGIN_FULLY_MASKED(NAME, TYPE) \
2627 SCAN_IF(NAME); \
2628 TYPE skey, smask; \
2629 memset(&skey, 0, sizeof skey); \
2630 memset(&smask, 0xff, sizeof smask); \
2631 do { \
2632 len = 0;
2633
2634 /* VLAN needs special initialization. */
2635 #define SCAN_BEGIN_INIT(NAME, TYPE, KEY_INIT, MASK_INIT) \
2636 SCAN_IF(NAME); \
2637 TYPE skey = KEY_INIT; \
2638 TYPE smask = MASK_INIT; \
2639 do { \
2640 len = 0;
2641
2642 /* Scan unnamed entry as 'TYPE' */
2643 #define SCAN_TYPE(TYPE, KEY, MASK) \
2644 len = scan_##TYPE(s, KEY, MASK); \
2645 if (len == 0) { \
2646 return -EINVAL; \
2647 } \
2648 s += len
2649
2650 /* Scan named ('NAME') entry 'FIELD' as 'TYPE'. */
2651 #define SCAN_FIELD(NAME, TYPE, FIELD) \
2652 if (strncmp(s, NAME, strlen(NAME)) == 0) { \
2653 s += strlen(NAME); \
2654 SCAN_TYPE(TYPE, &skey.FIELD, mask ? &smask.FIELD : NULL); \
2655 continue; \
2656 }
2657
2658 #define SCAN_FINISH() \
2659 } while (*s++ == ',' && len != 0); \
2660 if (s[-1] != ')') { \
2661 return -EINVAL; \
2662 }
2663
2664 #define SCAN_FINISH_SINGLE() \
2665 } while (false); \
2666 if (*s++ != ')') { \
2667 return -EINVAL; \
2668 }
2669
2670 #define SCAN_PUT(ATTR) \
2671 if (!mask || !is_all_zeros(&smask, sizeof smask)) { \
2672 SCAN_PUT_ATTR(key, ATTR, skey); \
2673 if (mask) { \
2674 SCAN_PUT_ATTR(mask, ATTR, smask); \
2675 } \
2676 }
2677
2678 #define SCAN_END(ATTR) \
2679 SCAN_FINISH(); \
2680 SCAN_PUT(ATTR); \
2681 return s - start; \
2682 }
2683
2684 #define SCAN_END_SINGLE(ATTR) \
2685 SCAN_FINISH_SINGLE(); \
2686 SCAN_PUT(ATTR); \
2687 return s - start; \
2688 }
2689
2690 #define SCAN_SINGLE(NAME, TYPE, SCAN_AS, ATTR) \
2691 SCAN_BEGIN(NAME, TYPE) { \
2692 SCAN_TYPE(SCAN_AS, &skey, &smask); \
2693 } SCAN_END_SINGLE(ATTR)
2694
2695 #define SCAN_SINGLE_FULLY_MASKED(NAME, TYPE, SCAN_AS, ATTR) \
2696 SCAN_BEGIN_FULLY_MASKED(NAME, TYPE) { \
2697 SCAN_TYPE(SCAN_AS, &skey, NULL); \
2698 } SCAN_END_SINGLE(ATTR)
2699
2700 /* scan_port needs one extra argument. */
2701 #define SCAN_SINGLE_PORT(NAME, TYPE, ATTR) \
2702 SCAN_BEGIN(NAME, TYPE) { \
2703 len = scan_port(s, &skey, &smask, port_names); \
2704 if (len == 0) { \
2705 return -EINVAL; \
2706 } \
2707 s += len; \
2708 } SCAN_END_SINGLE(ATTR)
2709
2710 static int
2711 parse_odp_key_mask_attr(const char *s, const struct simap *port_names,
2712 struct ofpbuf *key, struct ofpbuf *mask)
2713 {
2714 SCAN_SINGLE("skb_priority(", uint32_t, u32, OVS_KEY_ATTR_PRIORITY);
2715 SCAN_SINGLE("skb_mark(", uint32_t, u32, OVS_KEY_ATTR_SKB_MARK);
2716 SCAN_SINGLE_FULLY_MASKED("recirc_id(", uint32_t, u32,
2717 OVS_KEY_ATTR_RECIRC_ID);
2718 SCAN_SINGLE("dp_hash(", uint32_t, u32, OVS_KEY_ATTR_DP_HASH);
2719
2720 SCAN_BEGIN("tunnel(", struct flow_tnl) {
2721 SCAN_FIELD("tun_id=", be64, tun_id);
2722 SCAN_FIELD("src=", ipv4, ip_src);
2723 SCAN_FIELD("dst=", ipv4, ip_dst);
2724 SCAN_FIELD("tos=", u8, ip_tos);
2725 SCAN_FIELD("ttl=", u8, ip_ttl);
2726 SCAN_FIELD("tp_src=", be16, tp_src);
2727 SCAN_FIELD("tp_dst=", be16, tp_dst);
2728 SCAN_FIELD("gbp_id=", be16, gbp_id);
2729 SCAN_FIELD("gbp_flags=", u8, gbp_flags);
2730 SCAN_FIELD("flags(", tun_flags, flags);
2731 } SCAN_END(OVS_KEY_ATTR_TUNNEL);
2732
2733 SCAN_SINGLE_PORT("in_port(", uint32_t, OVS_KEY_ATTR_IN_PORT);
2734
2735 SCAN_BEGIN("eth(", struct ovs_key_ethernet) {
2736 SCAN_FIELD("src=", eth, eth_src);
2737 SCAN_FIELD("dst=", eth, eth_dst);
2738 } SCAN_END(OVS_KEY_ATTR_ETHERNET);
2739
2740 SCAN_BEGIN_INIT("vlan(", struct ovs_key_vlan__,
2741 { htons(VLAN_CFI) }, { htons(VLAN_CFI) }) {
2742 SCAN_FIELD("vid=", vid, tci);
2743 SCAN_FIELD("pcp=", pcp, tci);
2744 SCAN_FIELD("cfi=", cfi, tci);
2745 } SCAN_END(OVS_KEY_ATTR_VLAN);
2746
2747 SCAN_SINGLE("eth_type(", ovs_be16, be16, OVS_KEY_ATTR_ETHERTYPE);
2748
2749 SCAN_BEGIN("mpls(", struct ovs_key_mpls) {
2750 SCAN_FIELD("label=", mpls_label, mpls_lse);
2751 SCAN_FIELD("tc=", mpls_tc, mpls_lse);
2752 SCAN_FIELD("ttl=", mpls_ttl, mpls_lse);
2753 SCAN_FIELD("bos=", mpls_bos, mpls_lse);
2754 } SCAN_END(OVS_KEY_ATTR_MPLS);
2755
2756 SCAN_BEGIN("ipv4(", struct ovs_key_ipv4) {
2757 SCAN_FIELD("src=", ipv4, ipv4_src);
2758 SCAN_FIELD("dst=", ipv4, ipv4_dst);
2759 SCAN_FIELD("proto=", u8, ipv4_proto);
2760 SCAN_FIELD("tos=", u8, ipv4_tos);
2761 SCAN_FIELD("ttl=", u8, ipv4_ttl);
2762 SCAN_FIELD("frag=", frag, ipv4_frag);
2763 } SCAN_END(OVS_KEY_ATTR_IPV4);
2764
2765 SCAN_BEGIN("ipv6(", struct ovs_key_ipv6) {
2766 SCAN_FIELD("src=", ipv6, ipv6_src);
2767 SCAN_FIELD("dst=", ipv6, ipv6_dst);
2768 SCAN_FIELD("label=", ipv6_label, ipv6_label);
2769 SCAN_FIELD("proto=", u8, ipv6_proto);
2770 SCAN_FIELD("tclass=", u8, ipv6_tclass);
2771 SCAN_FIELD("hlimit=", u8, ipv6_hlimit);
2772 SCAN_FIELD("frag=", frag, ipv6_frag);
2773 } SCAN_END(OVS_KEY_ATTR_IPV6);
2774
2775 SCAN_BEGIN("tcp(", struct ovs_key_tcp) {
2776 SCAN_FIELD("src=", be16, tcp_src);
2777 SCAN_FIELD("dst=", be16, tcp_dst);
2778 } SCAN_END(OVS_KEY_ATTR_TCP);
2779
2780 SCAN_SINGLE("tcp_flags(", ovs_be16, tcp_flags, OVS_KEY_ATTR_TCP_FLAGS);
2781
2782 SCAN_BEGIN("udp(", struct ovs_key_udp) {
2783 SCAN_FIELD("src=", be16, udp_src);
2784 SCAN_FIELD("dst=", be16, udp_dst);
2785 } SCAN_END(OVS_KEY_ATTR_UDP);
2786
2787 SCAN_BEGIN("sctp(", struct ovs_key_sctp) {
2788 SCAN_FIELD("src=", be16, sctp_src);
2789 SCAN_FIELD("dst=", be16, sctp_dst);
2790 } SCAN_END(OVS_KEY_ATTR_SCTP);
2791
2792 SCAN_BEGIN("icmp(", struct ovs_key_icmp) {
2793 SCAN_FIELD("type=", u8, icmp_type);
2794 SCAN_FIELD("code=", u8, icmp_code);
2795 } SCAN_END(OVS_KEY_ATTR_ICMP);
2796
2797 SCAN_BEGIN("icmpv6(", struct ovs_key_icmpv6) {
2798 SCAN_FIELD("type=", u8, icmpv6_type);
2799 SCAN_FIELD("code=", u8, icmpv6_code);
2800 } SCAN_END(OVS_KEY_ATTR_ICMPV6);
2801
2802 SCAN_BEGIN("arp(", struct ovs_key_arp) {
2803 SCAN_FIELD("sip=", ipv4, arp_sip);
2804 SCAN_FIELD("tip=", ipv4, arp_tip);
2805 SCAN_FIELD("op=", be16, arp_op);
2806 SCAN_FIELD("sha=", eth, arp_sha);
2807 SCAN_FIELD("tha=", eth, arp_tha);
2808 } SCAN_END(OVS_KEY_ATTR_ARP);
2809
2810 SCAN_BEGIN("nd(", struct ovs_key_nd) {
2811 SCAN_FIELD("target=", ipv6, nd_target);
2812 SCAN_FIELD("sll=", eth, nd_sll);
2813 SCAN_FIELD("tll=", eth, nd_tll);
2814 } SCAN_END(OVS_KEY_ATTR_ND);
2815
2816 /* Encap open-coded. */
2817 if (!strncmp(s, "encap(", 6)) {
2818 const char *start = s;
2819 size_t encap, encap_mask = 0;
2820
2821 encap = nl_msg_start_nested(key, OVS_KEY_ATTR_ENCAP);
2822 if (mask) {
2823 encap_mask = nl_msg_start_nested(mask, OVS_KEY_ATTR_ENCAP);
2824 }
2825
2826 s += 6;
2827 for (;;) {
2828 int retval;
2829
2830 s += strspn(s, delimiters);
2831 if (!*s) {
2832 return -EINVAL;
2833 } else if (*s == ')') {
2834 break;
2835 }
2836
2837 retval = parse_odp_key_mask_attr(s, port_names, key, mask);
2838 if (retval < 0) {
2839 return retval;
2840 }
2841 s += retval;
2842 }
2843 s++;
2844
2845 nl_msg_end_nested(key, encap);
2846 if (mask) {
2847 nl_msg_end_nested(mask, encap_mask);
2848 }
2849
2850 return s - start;
2851 }
2852
2853 return -EINVAL;
2854 }
2855
2856 /* Parses the string representation of a datapath flow key, in the
2857 * format output by odp_flow_key_format(). Returns 0 if successful,
2858 * otherwise a positive errno value. On success, the flow key is
2859 * appended to 'key' as a series of Netlink attributes. On failure, no
2860 * data is appended to 'key'. Either way, 'key''s data might be
2861 * reallocated.
2862 *
2863 * If 'port_names' is nonnull, it points to an simap that maps from a port name
2864 * to a port number. (Port names may be used instead of port numbers in
2865 * in_port.)
2866 *
2867 * On success, the attributes appended to 'key' are individually syntactically
2868 * valid, but they may not be valid as a sequence. 'key' might, for example,
2869 * have duplicated keys. odp_flow_key_to_flow() will detect those errors. */
2870 int
2871 odp_flow_from_string(const char *s, const struct simap *port_names,
2872 struct ofpbuf *key, struct ofpbuf *mask)
2873 {
2874 const size_t old_size = key->size;
2875 for (;;) {
2876 int retval;
2877
2878 s += strspn(s, delimiters);
2879 if (!*s) {
2880 return 0;
2881 }
2882
2883 retval = parse_odp_key_mask_attr(s, port_names, key, mask);
2884 if (retval < 0) {
2885 key->size = old_size;
2886 return -retval;
2887 }
2888 s += retval;
2889 }
2890
2891 return 0;
2892 }
2893
2894 static uint8_t
2895 ovs_to_odp_frag(uint8_t nw_frag, bool is_mask)
2896 {
2897 if (is_mask) {
2898 /* Netlink interface 'enum ovs_frag_type' is an 8-bit enumeration type,
2899 * not a set of flags or bitfields. Hence, if the struct flow nw_frag
2900 * mask, which is a set of bits, has the FLOW_NW_FRAG_ANY as zero, we
2901 * must use a zero mask for the netlink frag field, and all ones mask
2902 * otherwise. */
2903 return (nw_frag & FLOW_NW_FRAG_ANY) ? UINT8_MAX : 0;
2904 }
2905 return !(nw_frag & FLOW_NW_FRAG_ANY) ? OVS_FRAG_TYPE_NONE
2906 : nw_frag & FLOW_NW_FRAG_LATER ? OVS_FRAG_TYPE_LATER
2907 : OVS_FRAG_TYPE_FIRST;
2908 }
2909
2910 static void get_ethernet_key(const struct flow *, struct ovs_key_ethernet *);
2911 static void put_ethernet_key(const struct ovs_key_ethernet *, struct flow *);
2912 static void get_ipv4_key(const struct flow *, struct ovs_key_ipv4 *,
2913 bool is_mask);
2914 static void put_ipv4_key(const struct ovs_key_ipv4 *, struct flow *,
2915 bool is_mask);
2916 static void get_ipv6_key(const struct flow *, struct ovs_key_ipv6 *,
2917 bool is_mask);
2918 static void put_ipv6_key(const struct ovs_key_ipv6 *, struct flow *,
2919 bool is_mask);
2920 static void get_arp_key(const struct flow *, struct ovs_key_arp *);
2921 static void put_arp_key(const struct ovs_key_arp *, struct flow *);
2922 static void get_nd_key(const struct flow *, struct ovs_key_nd *);
2923 static void put_nd_key(const struct ovs_key_nd *, struct flow *);
2924
2925 /* These share the same layout. */
2926 union ovs_key_tp {
2927 struct ovs_key_tcp tcp;
2928 struct ovs_key_udp udp;
2929 struct ovs_key_sctp sctp;
2930 };
2931
2932 static void get_tp_key(const struct flow *, union ovs_key_tp *);
2933 static void put_tp_key(const union ovs_key_tp *, struct flow *);
2934
2935 static void
2936 odp_flow_key_from_flow__(struct ofpbuf *buf, const struct flow *flow,
2937 const struct flow *mask, odp_port_t odp_in_port,
2938 size_t max_mpls_depth, bool recirc, bool export_mask)
2939 {
2940 struct ovs_key_ethernet *eth_key;
2941 size_t encap;
2942 const struct flow *data = export_mask ? mask : flow;
2943
2944 nl_msg_put_u32(buf, OVS_KEY_ATTR_PRIORITY, data->skb_priority);
2945
2946 if (flow->tunnel.ip_dst || export_mask) {
2947 tun_key_to_attr(buf, &data->tunnel);
2948 }
2949
2950 nl_msg_put_u32(buf, OVS_KEY_ATTR_SKB_MARK, data->pkt_mark);
2951
2952 if (recirc) {
2953 nl_msg_put_u32(buf, OVS_KEY_ATTR_RECIRC_ID, data->recirc_id);
2954 nl_msg_put_u32(buf, OVS_KEY_ATTR_DP_HASH, data->dp_hash);
2955 }
2956
2957 /* Add an ingress port attribute if this is a mask or 'odp_in_port'
2958 * is not the magical value "ODPP_NONE". */
2959 if (export_mask || odp_in_port != ODPP_NONE) {
2960 nl_msg_put_odp_port(buf, OVS_KEY_ATTR_IN_PORT, odp_in_port);
2961 }
2962
2963 eth_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ETHERNET,
2964 sizeof *eth_key);
2965 get_ethernet_key(data, eth_key);
2966
2967 if (flow->vlan_tci != htons(0) || flow->dl_type == htons(ETH_TYPE_VLAN)) {
2968 if (export_mask) {
2969 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, OVS_BE16_MAX);
2970 } else {
2971 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_TYPE_VLAN));
2972 }
2973 nl_msg_put_be16(buf, OVS_KEY_ATTR_VLAN, data->vlan_tci);
2974 encap = nl_msg_start_nested(buf, OVS_KEY_ATTR_ENCAP);
2975 if (flow->vlan_tci == htons(0)) {
2976 goto unencap;
2977 }
2978 } else {
2979 encap = 0;
2980 }
2981
2982 if (ntohs(flow->dl_type) < ETH_TYPE_MIN) {
2983 /* For backwards compatibility with kernels that don't support
2984 * wildcarding, the following convention is used to encode the
2985 * OVS_KEY_ATTR_ETHERTYPE for key and mask:
2986 *
2987 * key mask matches
2988 * -------- -------- -------
2989 * >0x5ff 0xffff Specified Ethernet II Ethertype.
2990 * >0x5ff 0 Any Ethernet II or non-Ethernet II frame.
2991 * <none> 0xffff Any non-Ethernet II frame (except valid
2992 * 802.3 SNAP packet with valid eth_type).
2993 */
2994 if (export_mask) {
2995 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, OVS_BE16_MAX);
2996 }
2997 goto unencap;
2998 }
2999
3000 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, data->dl_type);
3001
3002 if (flow->dl_type == htons(ETH_TYPE_IP)) {
3003 struct ovs_key_ipv4 *ipv4_key;
3004
3005 ipv4_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_IPV4,
3006 sizeof *ipv4_key);
3007 get_ipv4_key(data, ipv4_key, export_mask);
3008 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
3009 struct ovs_key_ipv6 *ipv6_key;
3010
3011 ipv6_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_IPV6,
3012 sizeof *ipv6_key);
3013 get_ipv6_key(data, ipv6_key, export_mask);
3014 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
3015 flow->dl_type == htons(ETH_TYPE_RARP)) {
3016 struct ovs_key_arp *arp_key;
3017
3018 arp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ARP,
3019 sizeof *arp_key);
3020 get_arp_key(data, arp_key);
3021 } else if (eth_type_mpls(flow->dl_type)) {
3022 struct ovs_key_mpls *mpls_key;
3023 int i, n;
3024
3025 n = flow_count_mpls_labels(flow, NULL);
3026 n = MIN(n, max_mpls_depth);
3027 mpls_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_MPLS,
3028 n * sizeof *mpls_key);
3029 for (i = 0; i < n; i++) {
3030 mpls_key[i].mpls_lse = data->mpls_lse[i];
3031 }
3032 }
3033
3034 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3035 if (flow->nw_proto == IPPROTO_TCP) {
3036 union ovs_key_tp *tcp_key;
3037
3038 tcp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_TCP,
3039 sizeof *tcp_key);
3040 get_tp_key(data, tcp_key);
3041 if (data->tcp_flags) {
3042 nl_msg_put_be16(buf, OVS_KEY_ATTR_TCP_FLAGS, data->tcp_flags);
3043 }
3044 } else if (flow->nw_proto == IPPROTO_UDP) {
3045 union ovs_key_tp *udp_key;
3046
3047 udp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_UDP,
3048 sizeof *udp_key);
3049 get_tp_key(data, udp_key);
3050 } else if (flow->nw_proto == IPPROTO_SCTP) {
3051 union ovs_key_tp *sctp_key;
3052
3053 sctp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_SCTP,
3054 sizeof *sctp_key);
3055 get_tp_key(data, sctp_key);
3056 } else if (flow->dl_type == htons(ETH_TYPE_IP)
3057 && flow->nw_proto == IPPROTO_ICMP) {
3058 struct ovs_key_icmp *icmp_key;
3059
3060 icmp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ICMP,
3061 sizeof *icmp_key);
3062 icmp_key->icmp_type = ntohs(data->tp_src);
3063 icmp_key->icmp_code = ntohs(data->tp_dst);
3064 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)
3065 && flow->nw_proto == IPPROTO_ICMPV6) {
3066 struct ovs_key_icmpv6 *icmpv6_key;
3067
3068 icmpv6_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ICMPV6,
3069 sizeof *icmpv6_key);
3070 icmpv6_key->icmpv6_type = ntohs(data->tp_src);
3071 icmpv6_key->icmpv6_code = ntohs(data->tp_dst);
3072
3073 if (flow->tp_dst == htons(0)
3074 && (flow->tp_src == htons(ND_NEIGHBOR_SOLICIT)
3075 || flow->tp_src == htons(ND_NEIGHBOR_ADVERT))
3076 && (!export_mask || (data->tp_src == htons(0xffff)
3077 && data->tp_dst == htons(0xffff)))) {
3078
3079 struct ovs_key_nd *nd_key;
3080
3081 nd_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ND,
3082 sizeof *nd_key);
3083 memcpy(nd_key->nd_target, &data->nd_target,
3084 sizeof nd_key->nd_target);
3085 memcpy(nd_key->nd_sll, data->arp_sha, ETH_ADDR_LEN);
3086 memcpy(nd_key->nd_tll, data->arp_tha, ETH_ADDR_LEN);
3087 }
3088 }
3089 }
3090
3091 unencap:
3092 if (encap) {
3093 nl_msg_end_nested(buf, encap);
3094 }
3095 }
3096
3097 /* Appends a representation of 'flow' as OVS_KEY_ATTR_* attributes to 'buf'.
3098 * 'flow->in_port' is ignored (since it is likely to be an OpenFlow port
3099 * number rather than a datapath port number). Instead, if 'odp_in_port'
3100 * is anything other than ODPP_NONE, it is included in 'buf' as the input
3101 * port.
3102 *
3103 * 'buf' must have at least ODPUTIL_FLOW_KEY_BYTES bytes of space, or be
3104 * capable of being expanded to allow for that much space.
3105 *
3106 * 'recirc' indicates support for recirculation fields. If this is true, then
3107 * these fields will always be serialised. */
3108 void
3109 odp_flow_key_from_flow(struct ofpbuf *buf, const struct flow *flow,
3110 const struct flow *mask, odp_port_t odp_in_port,
3111 bool recirc)
3112 {
3113 odp_flow_key_from_flow__(buf, flow, mask, odp_in_port, SIZE_MAX, recirc,
3114 false);
3115 }
3116
3117 /* Appends a representation of 'mask' as OVS_KEY_ATTR_* attributes to
3118 * 'buf'. 'flow' is used as a template to determine how to interpret
3119 * 'mask'. For example, the 'dl_type' of 'mask' describes the mask, but
3120 * it doesn't indicate whether the other fields should be interpreted as
3121 * ARP, IPv4, IPv6, etc.
3122 *
3123 * 'buf' must have at least ODPUTIL_FLOW_KEY_BYTES bytes of space, or be
3124 * capable of being expanded to allow for that much space.
3125 *
3126 * 'recirc' indicates support for recirculation fields. If this is true, then
3127 * these fields will always be serialised. */
3128 void
3129 odp_flow_key_from_mask(struct ofpbuf *buf, const struct flow *mask,
3130 const struct flow *flow, uint32_t odp_in_port_mask,
3131 size_t max_mpls_depth, bool recirc)
3132 {
3133 odp_flow_key_from_flow__(buf, flow, mask, u32_to_odp(odp_in_port_mask),
3134 max_mpls_depth, recirc, true);
3135 }
3136
3137 /* Generate ODP flow key from the given packet metadata */
3138 void
3139 odp_key_from_pkt_metadata(struct ofpbuf *buf, const struct pkt_metadata *md)
3140 {
3141 nl_msg_put_u32(buf, OVS_KEY_ATTR_PRIORITY, md->skb_priority);
3142
3143 if (md->tunnel.ip_dst) {
3144 tun_key_to_attr(buf, &md->tunnel);
3145 }
3146
3147 nl_msg_put_u32(buf, OVS_KEY_ATTR_SKB_MARK, md->pkt_mark);
3148
3149 /* Add an ingress port attribute if 'odp_in_port' is not the magical
3150 * value "ODPP_NONE". */
3151 if (md->in_port.odp_port != ODPP_NONE) {
3152 nl_msg_put_odp_port(buf, OVS_KEY_ATTR_IN_PORT, md->in_port.odp_port);
3153 }
3154 }
3155
3156 /* Generate packet metadata from the given ODP flow key. */
3157 void
3158 odp_key_to_pkt_metadata(const struct nlattr *key, size_t key_len,
3159 struct pkt_metadata *md)
3160 {
3161 const struct nlattr *nla;
3162 size_t left;
3163 uint32_t wanted_attrs = 1u << OVS_KEY_ATTR_PRIORITY |
3164 1u << OVS_KEY_ATTR_SKB_MARK | 1u << OVS_KEY_ATTR_TUNNEL |
3165 1u << OVS_KEY_ATTR_IN_PORT;
3166
3167 *md = PKT_METADATA_INITIALIZER(ODPP_NONE);
3168
3169 NL_ATTR_FOR_EACH (nla, left, key, key_len) {
3170 uint16_t type = nl_attr_type(nla);
3171 size_t len = nl_attr_get_size(nla);
3172 int expected_len = odp_flow_key_attr_len(type);
3173
3174 if (len != expected_len && expected_len >= 0) {
3175 continue;
3176 }
3177
3178 switch (type) {
3179 case OVS_KEY_ATTR_RECIRC_ID:
3180 md->recirc_id = nl_attr_get_u32(nla);
3181 wanted_attrs &= ~(1u << OVS_KEY_ATTR_RECIRC_ID);
3182 break;
3183 case OVS_KEY_ATTR_DP_HASH:
3184 md->dp_hash = nl_attr_get_u32(nla);
3185 wanted_attrs &= ~(1u << OVS_KEY_ATTR_DP_HASH);
3186 break;
3187 case OVS_KEY_ATTR_PRIORITY:
3188 md->skb_priority = nl_attr_get_u32(nla);
3189 wanted_attrs &= ~(1u << OVS_KEY_ATTR_PRIORITY);
3190 break;
3191 case OVS_KEY_ATTR_SKB_MARK:
3192 md->pkt_mark = nl_attr_get_u32(nla);
3193 wanted_attrs &= ~(1u << OVS_KEY_ATTR_SKB_MARK);
3194 break;
3195 case OVS_KEY_ATTR_TUNNEL: {
3196 enum odp_key_fitness res;
3197
3198 res = odp_tun_key_from_attr(nla, &md->tunnel);
3199 if (res == ODP_FIT_ERROR) {
3200 memset(&md->tunnel, 0, sizeof md->tunnel);
3201 } else if (res == ODP_FIT_PERFECT) {
3202 wanted_attrs &= ~(1u << OVS_KEY_ATTR_TUNNEL);
3203 }
3204 break;
3205 }
3206 case OVS_KEY_ATTR_IN_PORT:
3207 md->in_port.odp_port = nl_attr_get_odp_port(nla);
3208 wanted_attrs &= ~(1u << OVS_KEY_ATTR_IN_PORT);
3209 break;
3210 default:
3211 break;
3212 }
3213
3214 if (!wanted_attrs) {
3215 return; /* Have everything. */
3216 }
3217 }
3218 }
3219
3220 uint32_t
3221 odp_flow_key_hash(const struct nlattr *key, size_t key_len)
3222 {
3223 BUILD_ASSERT_DECL(!(NLA_ALIGNTO % sizeof(uint32_t)));
3224 return hash_words(ALIGNED_CAST(const uint32_t *, key),
3225 key_len / sizeof(uint32_t), 0);
3226 }
3227
3228 static void
3229 log_odp_key_attributes(struct vlog_rate_limit *rl, const char *title,
3230 uint64_t attrs, int out_of_range_attr,
3231 const struct nlattr *key, size_t key_len)
3232 {
3233 struct ds s;
3234 int i;
3235
3236 if (VLOG_DROP_DBG(rl)) {
3237 return;
3238 }
3239
3240 ds_init(&s);
3241 for (i = 0; i < 64; i++) {
3242 if (attrs & (UINT64_C(1) << i)) {
3243 char namebuf[OVS_KEY_ATTR_BUFSIZE];
3244
3245 ds_put_format(&s, " %s",
3246 ovs_key_attr_to_string(i, namebuf, sizeof namebuf));
3247 }
3248 }
3249 if (out_of_range_attr) {
3250 ds_put_format(&s, " %d (and possibly others)", out_of_range_attr);
3251 }
3252
3253 ds_put_cstr(&s, ": ");
3254 odp_flow_key_format(key, key_len, &s);
3255
3256 VLOG_DBG("%s:%s", title, ds_cstr(&s));
3257 ds_destroy(&s);
3258 }
3259
3260 static uint8_t
3261 odp_to_ovs_frag(uint8_t odp_frag, bool is_mask)
3262 {
3263 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3264
3265 if (is_mask) {
3266 return odp_frag ? FLOW_NW_FRAG_MASK : 0;
3267 }
3268
3269 if (odp_frag > OVS_FRAG_TYPE_LATER) {
3270 VLOG_ERR_RL(&rl, "invalid frag %"PRIu8" in flow key", odp_frag);
3271 return 0xff; /* Error. */
3272 }
3273
3274 return (odp_frag == OVS_FRAG_TYPE_NONE) ? 0
3275 : (odp_frag == OVS_FRAG_TYPE_FIRST) ? FLOW_NW_FRAG_ANY
3276 : FLOW_NW_FRAG_ANY | FLOW_NW_FRAG_LATER;
3277 }
3278
3279 static bool
3280 parse_flow_nlattrs(const struct nlattr *key, size_t key_len,
3281 const struct nlattr *attrs[], uint64_t *present_attrsp,
3282 int *out_of_range_attrp)
3283 {
3284 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3285 const struct nlattr *nla;
3286 uint64_t present_attrs;
3287 size_t left;
3288
3289 BUILD_ASSERT(OVS_KEY_ATTR_MAX < CHAR_BIT * sizeof present_attrs);
3290 present_attrs = 0;
3291 *out_of_range_attrp = 0;
3292 NL_ATTR_FOR_EACH (nla, left, key, key_len) {
3293 uint16_t type = nl_attr_type(nla);
3294 size_t len = nl_attr_get_size(nla);
3295 int expected_len = odp_flow_key_attr_len(type);
3296
3297 if (len != expected_len && expected_len >= 0) {
3298 char namebuf[OVS_KEY_ATTR_BUFSIZE];
3299
3300 VLOG_ERR_RL(&rl, "attribute %s has length %"PRIuSIZE" but should have "
3301 "length %d", ovs_key_attr_to_string(type, namebuf,
3302 sizeof namebuf),
3303 len, expected_len);
3304 return false;
3305 }
3306
3307 if (type > OVS_KEY_ATTR_MAX) {
3308 *out_of_range_attrp = type;
3309 } else {
3310 if (present_attrs & (UINT64_C(1) << type)) {
3311 char namebuf[OVS_KEY_ATTR_BUFSIZE];
3312
3313 VLOG_ERR_RL(&rl, "duplicate %s attribute in flow key",
3314 ovs_key_attr_to_string(type,
3315 namebuf, sizeof namebuf));
3316 return false;
3317 }
3318
3319 present_attrs |= UINT64_C(1) << type;
3320 attrs[type] = nla;
3321 }
3322 }
3323 if (left) {
3324 VLOG_ERR_RL(&rl, "trailing garbage in flow key");
3325 return false;
3326 }
3327
3328 *present_attrsp = present_attrs;
3329 return true;
3330 }
3331
3332 static enum odp_key_fitness
3333 check_expectations(uint64_t present_attrs, int out_of_range_attr,
3334 uint64_t expected_attrs,
3335 const struct nlattr *key, size_t key_len)
3336 {
3337 uint64_t missing_attrs;
3338 uint64_t extra_attrs;
3339
3340 missing_attrs = expected_attrs & ~present_attrs;
3341 if (missing_attrs) {
3342 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3343 log_odp_key_attributes(&rl, "expected but not present",
3344 missing_attrs, 0, key, key_len);
3345 return ODP_FIT_TOO_LITTLE;
3346 }
3347
3348 extra_attrs = present_attrs & ~expected_attrs;
3349 if (extra_attrs || out_of_range_attr) {
3350 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3351 log_odp_key_attributes(&rl, "present but not expected",
3352 extra_attrs, out_of_range_attr, key, key_len);
3353 return ODP_FIT_TOO_MUCH;
3354 }
3355
3356 return ODP_FIT_PERFECT;
3357 }
3358
3359 static bool
3360 parse_ethertype(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
3361 uint64_t present_attrs, uint64_t *expected_attrs,
3362 struct flow *flow, const struct flow *src_flow)
3363 {
3364 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3365 bool is_mask = flow != src_flow;
3366
3367 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE)) {
3368 flow->dl_type = nl_attr_get_be16(attrs[OVS_KEY_ATTR_ETHERTYPE]);
3369 if (!is_mask && ntohs(flow->dl_type) < ETH_TYPE_MIN) {
3370 VLOG_ERR_RL(&rl, "invalid Ethertype %"PRIu16" in flow key",
3371 ntohs(flow->dl_type));
3372 return false;
3373 }
3374 if (is_mask && ntohs(src_flow->dl_type) < ETH_TYPE_MIN &&
3375 flow->dl_type != htons(0xffff)) {
3376 return false;
3377 }
3378 *expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE;
3379 } else {
3380 if (!is_mask) {
3381 flow->dl_type = htons(FLOW_DL_TYPE_NONE);
3382 } else if (ntohs(src_flow->dl_type) < ETH_TYPE_MIN) {
3383 /* See comments in odp_flow_key_from_flow__(). */
3384 VLOG_ERR_RL(&rl, "mask expected for non-Ethernet II frame");
3385 return false;
3386 }
3387 }
3388 return true;
3389 }
3390
3391 static enum odp_key_fitness
3392 parse_l2_5_onward(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
3393 uint64_t present_attrs, int out_of_range_attr,
3394 uint64_t expected_attrs, struct flow *flow,
3395 const struct nlattr *key, size_t key_len,
3396 const struct flow *src_flow)
3397 {
3398 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3399 bool is_mask = src_flow != flow;
3400 const void *check_start = NULL;
3401 size_t check_len = 0;
3402 enum ovs_key_attr expected_bit = 0xff;
3403
3404 if (eth_type_mpls(src_flow->dl_type)) {
3405 if (!is_mask || present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_MPLS)) {
3406 expected_attrs |= (UINT64_C(1) << OVS_KEY_ATTR_MPLS);
3407 }
3408 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_MPLS)) {
3409 size_t size = nl_attr_get_size(attrs[OVS_KEY_ATTR_MPLS]);
3410 const ovs_be32 *mpls_lse = nl_attr_get(attrs[OVS_KEY_ATTR_MPLS]);
3411 int n = size / sizeof(ovs_be32);
3412 int i;
3413
3414 if (!size || size % sizeof(ovs_be32)) {
3415 return ODP_FIT_ERROR;
3416 }
3417 if (flow->mpls_lse[0] && flow->dl_type != htons(0xffff)) {
3418 return ODP_FIT_ERROR;
3419 }
3420
3421 for (i = 0; i < n && i < FLOW_MAX_MPLS_LABELS; i++) {
3422 flow->mpls_lse[i] = mpls_lse[i];
3423 }
3424 if (n > FLOW_MAX_MPLS_LABELS) {
3425 return ODP_FIT_TOO_MUCH;
3426 }
3427
3428 if (!is_mask) {
3429 /* BOS may be set only in the innermost label. */
3430 for (i = 0; i < n - 1; i++) {
3431 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
3432 return ODP_FIT_ERROR;
3433 }
3434 }
3435
3436 /* BOS must be set in the innermost label. */
3437 if (n < FLOW_MAX_MPLS_LABELS
3438 && !(flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK))) {
3439 return ODP_FIT_TOO_LITTLE;
3440 }
3441 }
3442 }
3443
3444 goto done;
3445 } else if (src_flow->dl_type == htons(ETH_TYPE_IP)) {
3446 if (!is_mask) {
3447 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IPV4;
3448 }
3449 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IPV4)) {
3450 const struct ovs_key_ipv4 *ipv4_key;
3451
3452 ipv4_key = nl_attr_get(attrs[OVS_KEY_ATTR_IPV4]);
3453 put_ipv4_key(ipv4_key, flow, is_mask);
3454 if (flow->nw_frag > FLOW_NW_FRAG_MASK) {
3455 return ODP_FIT_ERROR;
3456 }
3457 if (is_mask) {
3458 check_start = ipv4_key;
3459 check_len = sizeof *ipv4_key;
3460 expected_bit = OVS_KEY_ATTR_IPV4;
3461 }
3462 }
3463 } else if (src_flow->dl_type == htons(ETH_TYPE_IPV6)) {
3464 if (!is_mask) {
3465 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IPV6;
3466 }
3467 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IPV6)) {
3468 const struct ovs_key_ipv6 *ipv6_key;
3469
3470 ipv6_key = nl_attr_get(attrs[OVS_KEY_ATTR_IPV6]);
3471 put_ipv6_key(ipv6_key, flow, is_mask);
3472 if (flow->nw_frag > FLOW_NW_FRAG_MASK) {
3473 return ODP_FIT_ERROR;
3474 }
3475 if (is_mask) {
3476 check_start = ipv6_key;
3477 check_len = sizeof *ipv6_key;
3478 expected_bit = OVS_KEY_ATTR_IPV6;
3479 }
3480 }
3481 } else if (src_flow->dl_type == htons(ETH_TYPE_ARP) ||
3482 src_flow->dl_type == htons(ETH_TYPE_RARP)) {
3483 if (!is_mask) {
3484 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ARP;
3485 }
3486 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ARP)) {
3487 const struct ovs_key_arp *arp_key;
3488
3489 arp_key = nl_attr_get(attrs[OVS_KEY_ATTR_ARP]);
3490 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
3491 VLOG_ERR_RL(&rl, "unsupported ARP opcode %"PRIu16" in flow "
3492 "key", ntohs(arp_key->arp_op));
3493 return ODP_FIT_ERROR;
3494 }
3495 put_arp_key(arp_key, flow);
3496 if (is_mask) {
3497 check_start = arp_key;
3498 check_len = sizeof *arp_key;
3499 expected_bit = OVS_KEY_ATTR_ARP;
3500 }
3501 }
3502 } else {
3503 goto done;
3504 }
3505 if (check_len > 0) { /* Happens only when 'is_mask'. */
3506 if (!is_all_zeros(check_start, check_len) &&
3507 flow->dl_type != htons(0xffff)) {
3508 return ODP_FIT_ERROR;
3509 } else {
3510 expected_attrs |= UINT64_C(1) << expected_bit;
3511 }
3512 }
3513
3514 expected_bit = OVS_KEY_ATTR_UNSPEC;
3515 if (src_flow->nw_proto == IPPROTO_TCP
3516 && (src_flow->dl_type == htons(ETH_TYPE_IP) ||
3517 src_flow->dl_type == htons(ETH_TYPE_IPV6))
3518 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3519 if (!is_mask) {
3520 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_TCP;
3521 }
3522 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_TCP)) {
3523 const union ovs_key_tp *tcp_key;
3524
3525 tcp_key = nl_attr_get(attrs[OVS_KEY_ATTR_TCP]);
3526 put_tp_key(tcp_key, flow);
3527 expected_bit = OVS_KEY_ATTR_TCP;
3528 }
3529 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_TCP_FLAGS)) {
3530 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_TCP_FLAGS;
3531 flow->tcp_flags = nl_attr_get_be16(attrs[OVS_KEY_ATTR_TCP_FLAGS]);
3532 }
3533 } else if (src_flow->nw_proto == IPPROTO_UDP
3534 && (src_flow->dl_type == htons(ETH_TYPE_IP) ||
3535 src_flow->dl_type == htons(ETH_TYPE_IPV6))
3536 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3537 if (!is_mask) {
3538 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_UDP;
3539 }
3540 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_UDP)) {
3541 const union ovs_key_tp *udp_key;
3542
3543 udp_key = nl_attr_get(attrs[OVS_KEY_ATTR_UDP]);
3544 put_tp_key(udp_key, flow);
3545 expected_bit = OVS_KEY_ATTR_UDP;
3546 }
3547 } else if (src_flow->nw_proto == IPPROTO_SCTP
3548 && (src_flow->dl_type == htons(ETH_TYPE_IP) ||
3549 src_flow->dl_type == htons(ETH_TYPE_IPV6))
3550 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3551 if (!is_mask) {
3552 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_SCTP;
3553 }
3554 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_SCTP)) {
3555 const union ovs_key_tp *sctp_key;
3556
3557 sctp_key = nl_attr_get(attrs[OVS_KEY_ATTR_SCTP]);
3558 put_tp_key(sctp_key, flow);
3559 expected_bit = OVS_KEY_ATTR_SCTP;
3560 }
3561 } else if (src_flow->nw_proto == IPPROTO_ICMP
3562 && src_flow->dl_type == htons(ETH_TYPE_IP)
3563 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3564 if (!is_mask) {
3565 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ICMP;
3566 }
3567 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ICMP)) {
3568 const struct ovs_key_icmp *icmp_key;
3569
3570 icmp_key = nl_attr_get(attrs[OVS_KEY_ATTR_ICMP]);
3571 flow->tp_src = htons(icmp_key->icmp_type);
3572 flow->tp_dst = htons(icmp_key->icmp_code);
3573 expected_bit = OVS_KEY_ATTR_ICMP;
3574 }
3575 } else if (src_flow->nw_proto == IPPROTO_ICMPV6
3576 && src_flow->dl_type == htons(ETH_TYPE_IPV6)
3577 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3578 if (!is_mask) {
3579 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ICMPV6;
3580 }
3581 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ICMPV6)) {
3582 const struct ovs_key_icmpv6 *icmpv6_key;
3583
3584 icmpv6_key = nl_attr_get(attrs[OVS_KEY_ATTR_ICMPV6]);
3585 flow->tp_src = htons(icmpv6_key->icmpv6_type);
3586 flow->tp_dst = htons(icmpv6_key->icmpv6_code);
3587 expected_bit = OVS_KEY_ATTR_ICMPV6;
3588 if (src_flow->tp_dst == htons(0) &&
3589 (src_flow->tp_src == htons(ND_NEIGHBOR_SOLICIT) ||
3590 src_flow->tp_src == htons(ND_NEIGHBOR_ADVERT))) {
3591 if (!is_mask) {
3592 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ND;
3593 }
3594 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ND)) {
3595 const struct ovs_key_nd *nd_key;
3596
3597 nd_key = nl_attr_get(attrs[OVS_KEY_ATTR_ND]);
3598 memcpy(&flow->nd_target, nd_key->nd_target,
3599 sizeof flow->nd_target);
3600 memcpy(flow->arp_sha, nd_key->nd_sll, ETH_ADDR_LEN);
3601 memcpy(flow->arp_tha, nd_key->nd_tll, ETH_ADDR_LEN);
3602 if (is_mask) {
3603 if (!is_all_zeros(nd_key, sizeof *nd_key) &&
3604 (flow->tp_src != htons(0xffff) ||
3605 flow->tp_dst != htons(0xffff))) {
3606 return ODP_FIT_ERROR;
3607 } else {
3608 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ND;
3609 }
3610 }
3611 }
3612 }
3613 }
3614 }
3615 if (is_mask && expected_bit != OVS_KEY_ATTR_UNSPEC) {
3616 if ((flow->tp_src || flow->tp_dst) && flow->nw_proto != 0xff) {
3617 return ODP_FIT_ERROR;
3618 } else {
3619 expected_attrs |= UINT64_C(1) << expected_bit;
3620 }
3621 }
3622
3623 done:
3624 return check_expectations(present_attrs, out_of_range_attr, expected_attrs,
3625 key, key_len);
3626 }
3627
3628 /* Parse 802.1Q header then encapsulated L3 attributes. */
3629 static enum odp_key_fitness
3630 parse_8021q_onward(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
3631 uint64_t present_attrs, int out_of_range_attr,
3632 uint64_t expected_attrs, struct flow *flow,
3633 const struct nlattr *key, size_t key_len,
3634 const struct flow *src_flow)
3635 {
3636 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3637 bool is_mask = src_flow != flow;
3638
3639 const struct nlattr *encap
3640 = (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ENCAP)
3641 ? attrs[OVS_KEY_ATTR_ENCAP] : NULL);
3642 enum odp_key_fitness encap_fitness;
3643 enum odp_key_fitness fitness;
3644
3645 /* Calculate fitness of outer attributes. */
3646 if (!is_mask) {
3647 expected_attrs |= ((UINT64_C(1) << OVS_KEY_ATTR_VLAN) |
3648 (UINT64_C(1) << OVS_KEY_ATTR_ENCAP));
3649 } else {
3650 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN)) {
3651 expected_attrs |= (UINT64_C(1) << OVS_KEY_ATTR_VLAN);
3652 }
3653 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ENCAP)) {
3654 expected_attrs |= (UINT64_C(1) << OVS_KEY_ATTR_ENCAP);
3655 }
3656 }
3657 fitness = check_expectations(present_attrs, out_of_range_attr,
3658 expected_attrs, key, key_len);
3659
3660 /* Set vlan_tci.
3661 * Remove the TPID from dl_type since it's not the real Ethertype. */
3662 flow->dl_type = htons(0);
3663 flow->vlan_tci = (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN)
3664 ? nl_attr_get_be16(attrs[OVS_KEY_ATTR_VLAN])
3665 : htons(0));
3666 if (!is_mask) {
3667 if (!(present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN))) {
3668 return ODP_FIT_TOO_LITTLE;
3669 } else if (flow->vlan_tci == htons(0)) {
3670 /* Corner case for a truncated 802.1Q header. */
3671 if (fitness == ODP_FIT_PERFECT && nl_attr_get_size(encap)) {
3672 return ODP_FIT_TOO_MUCH;
3673 }
3674 return fitness;
3675 } else if (!(flow->vlan_tci & htons(VLAN_CFI))) {
3676 VLOG_ERR_RL(&rl, "OVS_KEY_ATTR_VLAN 0x%04"PRIx16" is nonzero "
3677 "but CFI bit is not set", ntohs(flow->vlan_tci));
3678 return ODP_FIT_ERROR;
3679 }
3680 } else {
3681 if (!(present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ENCAP))) {
3682 return fitness;
3683 }
3684 }
3685
3686 /* Now parse the encapsulated attributes. */
3687 if (!parse_flow_nlattrs(nl_attr_get(encap), nl_attr_get_size(encap),
3688 attrs, &present_attrs, &out_of_range_attr)) {
3689 return ODP_FIT_ERROR;
3690 }
3691 expected_attrs = 0;
3692
3693 if (!parse_ethertype(attrs, present_attrs, &expected_attrs, flow, src_flow)) {
3694 return ODP_FIT_ERROR;
3695 }
3696 encap_fitness = parse_l2_5_onward(attrs, present_attrs, out_of_range_attr,
3697 expected_attrs, flow, key, key_len,
3698 src_flow);
3699
3700 /* The overall fitness is the worse of the outer and inner attributes. */
3701 return MAX(fitness, encap_fitness);
3702 }
3703
3704 static enum odp_key_fitness
3705 odp_flow_key_to_flow__(const struct nlattr *key, size_t key_len,
3706 struct flow *flow, const struct flow *src_flow)
3707 {
3708 const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1];
3709 uint64_t expected_attrs;
3710 uint64_t present_attrs;
3711 int out_of_range_attr;
3712 bool is_mask = src_flow != flow;
3713
3714 memset(flow, 0, sizeof *flow);
3715
3716 /* Parse attributes. */
3717 if (!parse_flow_nlattrs(key, key_len, attrs, &present_attrs,
3718 &out_of_range_attr)) {
3719 return ODP_FIT_ERROR;
3720 }
3721 expected_attrs = 0;
3722
3723 /* Metadata. */
3724 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_RECIRC_ID)) {
3725 flow->recirc_id = nl_attr_get_u32(attrs[OVS_KEY_ATTR_RECIRC_ID]);
3726 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_RECIRC_ID;
3727 } else if (is_mask) {
3728 /* Always exact match recirc_id if it is not specified. */
3729 flow->recirc_id = UINT32_MAX;
3730 }
3731
3732 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_DP_HASH)) {
3733 flow->dp_hash = nl_attr_get_u32(attrs[OVS_KEY_ATTR_DP_HASH]);
3734 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_DP_HASH;
3735 }
3736 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_PRIORITY)) {
3737 flow->skb_priority = nl_attr_get_u32(attrs[OVS_KEY_ATTR_PRIORITY]);
3738 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_PRIORITY;
3739 }
3740
3741 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_SKB_MARK)) {
3742 flow->pkt_mark = nl_attr_get_u32(attrs[OVS_KEY_ATTR_SKB_MARK]);
3743 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_SKB_MARK;
3744 }
3745
3746 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_TUNNEL)) {
3747 enum odp_key_fitness res;
3748
3749 res = odp_tun_key_from_attr(attrs[OVS_KEY_ATTR_TUNNEL], &flow->tunnel);
3750 if (res == ODP_FIT_ERROR) {
3751 return ODP_FIT_ERROR;
3752 } else if (res == ODP_FIT_PERFECT) {
3753 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_TUNNEL;
3754 }
3755 }
3756
3757 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IN_PORT)) {
3758 flow->in_port.odp_port
3759 = nl_attr_get_odp_port(attrs[OVS_KEY_ATTR_IN_PORT]);
3760 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IN_PORT;
3761 } else if (!is_mask) {
3762 flow->in_port.odp_port = ODPP_NONE;
3763 }
3764
3765 /* Ethernet header. */
3766 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ETHERNET)) {
3767 const struct ovs_key_ethernet *eth_key;
3768
3769 eth_key = nl_attr_get(attrs[OVS_KEY_ATTR_ETHERNET]);
3770 put_ethernet_key(eth_key, flow);
3771 if (is_mask) {
3772 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ETHERNET;
3773 }
3774 }
3775 if (!is_mask) {
3776 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ETHERNET;
3777 }
3778
3779 /* Get Ethertype or 802.1Q TPID or FLOW_DL_TYPE_NONE. */
3780 if (!parse_ethertype(attrs, present_attrs, &expected_attrs, flow,
3781 src_flow)) {
3782 return ODP_FIT_ERROR;
3783 }
3784
3785 if (is_mask
3786 ? (src_flow->vlan_tci & htons(VLAN_CFI)) != 0
3787 : src_flow->dl_type == htons(ETH_TYPE_VLAN)) {
3788 return parse_8021q_onward(attrs, present_attrs, out_of_range_attr,
3789 expected_attrs, flow, key, key_len, src_flow);
3790 }
3791 if (is_mask) {
3792 flow->vlan_tci = htons(0xffff);
3793 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN)) {
3794 flow->vlan_tci = nl_attr_get_be16(attrs[OVS_KEY_ATTR_VLAN]);
3795 expected_attrs |= (UINT64_C(1) << OVS_KEY_ATTR_VLAN);
3796 }
3797 }
3798 return parse_l2_5_onward(attrs, present_attrs, out_of_range_attr,
3799 expected_attrs, flow, key, key_len, src_flow);
3800 }
3801
3802 /* Converts the 'key_len' bytes of OVS_KEY_ATTR_* attributes in 'key' to a flow
3803 * structure in 'flow'. Returns an ODP_FIT_* value that indicates how well
3804 * 'key' fits our expectations for what a flow key should contain.
3805 *
3806 * The 'in_port' will be the datapath's understanding of the port. The
3807 * caller will need to translate with odp_port_to_ofp_port() if the
3808 * OpenFlow port is needed.
3809 *
3810 * This function doesn't take the packet itself as an argument because none of
3811 * the currently understood OVS_KEY_ATTR_* attributes require it. Currently,
3812 * it is always possible to infer which additional attribute(s) should appear
3813 * by looking at the attributes for lower-level protocols, e.g. if the network
3814 * protocol in OVS_KEY_ATTR_IPV4 or OVS_KEY_ATTR_IPV6 is IPPROTO_TCP then we
3815 * know that a OVS_KEY_ATTR_TCP attribute must appear and that otherwise it
3816 * must be absent. */
3817 enum odp_key_fitness
3818 odp_flow_key_to_flow(const struct nlattr *key, size_t key_len,
3819 struct flow *flow)
3820 {
3821 return odp_flow_key_to_flow__(key, key_len, flow, flow);
3822 }
3823
3824 /* Converts the 'key_len' bytes of OVS_KEY_ATTR_* attributes in 'key' to a mask
3825 * structure in 'mask'. 'flow' must be a previously translated flow
3826 * corresponding to 'mask'. Returns an ODP_FIT_* value that indicates how well
3827 * 'key' fits our expectations for what a flow key should contain. */
3828 enum odp_key_fitness
3829 odp_flow_key_to_mask(const struct nlattr *key, size_t key_len,
3830 struct flow *mask, const struct flow *flow)
3831 {
3832 return odp_flow_key_to_flow__(key, key_len, mask, flow);
3833 }
3834
3835 /* Returns 'fitness' as a string, for use in debug messages. */
3836 const char *
3837 odp_key_fitness_to_string(enum odp_key_fitness fitness)
3838 {
3839 switch (fitness) {
3840 case ODP_FIT_PERFECT:
3841 return "OK";
3842 case ODP_FIT_TOO_MUCH:
3843 return "too_much";
3844 case ODP_FIT_TOO_LITTLE:
3845 return "too_little";
3846 case ODP_FIT_ERROR:
3847 return "error";
3848 default:
3849 return "<unknown>";
3850 }
3851 }
3852
3853 /* Appends an OVS_ACTION_ATTR_USERSPACE action to 'odp_actions' that specifies
3854 * Netlink PID 'pid'. If 'userdata' is nonnull, adds a userdata attribute
3855 * whose contents are the 'userdata_size' bytes at 'userdata' and returns the
3856 * offset within 'odp_actions' of the start of the cookie. (If 'userdata' is
3857 * null, then the return value is not meaningful.) */
3858 size_t
3859 odp_put_userspace_action(uint32_t pid,
3860 const void *userdata, size_t userdata_size,
3861 odp_port_t tunnel_out_port,
3862 struct ofpbuf *odp_actions)
3863 {
3864 size_t userdata_ofs;
3865 size_t offset;
3866
3867 offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_USERSPACE);
3868 nl_msg_put_u32(odp_actions, OVS_USERSPACE_ATTR_PID, pid);
3869 if (userdata) {
3870 userdata_ofs = odp_actions->size + NLA_HDRLEN;
3871
3872 /* The OVS kernel module before OVS 1.11 and the upstream Linux kernel
3873 * module before Linux 3.10 required the userdata to be exactly 8 bytes
3874 * long:
3875 *
3876 * - The kernel rejected shorter userdata with -ERANGE.
3877 *
3878 * - The kernel silently dropped userdata beyond the first 8 bytes.
3879 *
3880 * Thus, for maximum compatibility, always put at least 8 bytes. (We
3881 * separately disable features that required more than 8 bytes.) */
3882 memcpy(nl_msg_put_unspec_zero(odp_actions, OVS_USERSPACE_ATTR_USERDATA,
3883 MAX(8, userdata_size)),
3884 userdata, userdata_size);
3885 } else {
3886 userdata_ofs = 0;
3887 }
3888 if (tunnel_out_port != ODPP_NONE) {
3889 nl_msg_put_odp_port(odp_actions, OVS_USERSPACE_ATTR_EGRESS_TUN_PORT,
3890 tunnel_out_port);
3891 }
3892 nl_msg_end_nested(odp_actions, offset);
3893
3894 return userdata_ofs;
3895 }
3896
3897 void
3898 odp_put_tunnel_action(const struct flow_tnl *tunnel,
3899 struct ofpbuf *odp_actions)
3900 {
3901 size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
3902 tun_key_to_attr(odp_actions, tunnel);
3903 nl_msg_end_nested(odp_actions, offset);
3904 }
3905
3906 void
3907 odp_put_tnl_push_action(struct ofpbuf *odp_actions,
3908 struct ovs_action_push_tnl *data)
3909 {
3910 int size = offsetof(struct ovs_action_push_tnl, header);
3911
3912 size += data->header_len;
3913 nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_TUNNEL_PUSH, data, size);
3914 }
3915
3916 \f
3917 /* The commit_odp_actions() function and its helpers. */
3918
3919 static void
3920 commit_set_action(struct ofpbuf *odp_actions, enum ovs_key_attr key_type,
3921 const void *key, size_t key_size)
3922 {
3923 size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
3924 nl_msg_put_unspec(odp_actions, key_type, key, key_size);
3925 nl_msg_end_nested(odp_actions, offset);
3926 }
3927
3928 /* Masked set actions have a mask following the data within the netlink
3929 * attribute. The unmasked bits in the data will be cleared as the data
3930 * is copied to the action. */
3931 void
3932 commit_masked_set_action(struct ofpbuf *odp_actions,
3933 enum ovs_key_attr key_type,
3934 const void *key_, const void *mask_, size_t key_size)
3935 {
3936 size_t offset = nl_msg_start_nested(odp_actions,
3937 OVS_ACTION_ATTR_SET_MASKED);
3938 char *data = nl_msg_put_unspec_uninit(odp_actions, key_type, key_size * 2);
3939 const char *key = key_, *mask = mask_;
3940
3941 memcpy(data + key_size, mask, key_size);
3942 /* Clear unmasked bits while copying. */
3943 while (key_size--) {
3944 *data++ = *key++ & *mask++;
3945 }
3946 nl_msg_end_nested(odp_actions, offset);
3947 }
3948
3949 /* If any of the flow key data that ODP actions can modify are different in
3950 * 'base->tunnel' and 'flow->tunnel', appends a set_tunnel ODP action to
3951 * 'odp_actions' that change the flow tunneling information in key from
3952 * 'base->tunnel' into 'flow->tunnel', and then changes 'base->tunnel' in the
3953 * same way. In other words, operates the same as commit_odp_actions(), but
3954 * only on tunneling information. */
3955 void
3956 commit_odp_tunnel_action(const struct flow *flow, struct flow *base,
3957 struct ofpbuf *odp_actions)
3958 {
3959 /* A valid IPV4_TUNNEL must have non-zero ip_dst. */
3960 if (flow->tunnel.ip_dst) {
3961 if (!memcmp(&base->tunnel, &flow->tunnel, sizeof base->tunnel)) {
3962 return;
3963 }
3964 memcpy(&base->tunnel, &flow->tunnel, sizeof base->tunnel);
3965 odp_put_tunnel_action(&base->tunnel, odp_actions);
3966 }
3967 }
3968
3969 static bool
3970 commit(enum ovs_key_attr attr, bool use_masked_set,
3971 const void *key, void *base, void *mask, size_t size,
3972 struct ofpbuf *odp_actions)
3973 {
3974 if (memcmp(key, base, size)) {
3975 bool fully_masked = odp_mask_is_exact(attr, mask, size);
3976
3977 if (use_masked_set && !fully_masked) {
3978 commit_masked_set_action(odp_actions, attr, key, mask, size);
3979 } else {
3980 if (!fully_masked) {
3981 memset(mask, 0xff, size);
3982 }
3983 commit_set_action(odp_actions, attr, key, size);
3984 }
3985 memcpy(base, key, size);
3986 return true;
3987 } else {
3988 /* Mask bits are set when we have either read or set the corresponding
3989 * values. Masked bits will be exact-matched, no need to set them
3990 * if the value did not actually change. */
3991 return false;
3992 }
3993 }
3994
3995 static void
3996 get_ethernet_key(const struct flow *flow, struct ovs_key_ethernet *eth)
3997 {
3998 memcpy(eth->eth_src, flow->dl_src, ETH_ADDR_LEN);
3999 memcpy(eth->eth_dst, flow->dl_dst, ETH_ADDR_LEN);
4000 }
4001
4002 static void
4003 put_ethernet_key(const struct ovs_key_ethernet *eth, struct flow *flow)
4004 {
4005 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
4006 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
4007 }
4008
4009 static void
4010 commit_set_ether_addr_action(const struct flow *flow, struct flow *base_flow,
4011 struct ofpbuf *odp_actions,
4012 struct flow_wildcards *wc,
4013 bool use_masked)
4014 {
4015 struct ovs_key_ethernet key, base, mask;
4016
4017 get_ethernet_key(flow, &key);
4018 get_ethernet_key(base_flow, &base);
4019 get_ethernet_key(&wc->masks, &mask);
4020
4021 if (commit(OVS_KEY_ATTR_ETHERNET, use_masked,
4022 &key, &base, &mask, sizeof key, odp_actions)) {
4023 put_ethernet_key(&base, base_flow);
4024 put_ethernet_key(&mask, &wc->masks);
4025 }
4026 }
4027
4028 static void
4029 pop_vlan(struct flow *base,
4030 struct ofpbuf *odp_actions, struct flow_wildcards *wc)
4031 {
4032 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4033
4034 if (base->vlan_tci & htons(VLAN_CFI)) {
4035 nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_VLAN);
4036 base->vlan_tci = 0;
4037 }
4038 }
4039
4040 static void
4041 commit_vlan_action(ovs_be16 vlan_tci, struct flow *base,
4042 struct ofpbuf *odp_actions, struct flow_wildcards *wc)
4043 {
4044 if (base->vlan_tci == vlan_tci) {
4045 return;
4046 }
4047
4048 pop_vlan(base, odp_actions, wc);
4049 if (vlan_tci & htons(VLAN_CFI)) {
4050 struct ovs_action_push_vlan vlan;
4051
4052 vlan.vlan_tpid = htons(ETH_TYPE_VLAN);
4053 vlan.vlan_tci = vlan_tci;
4054 nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_PUSH_VLAN,
4055 &vlan, sizeof vlan);
4056 }
4057 base->vlan_tci = vlan_tci;
4058 }
4059
4060 /* Wildcarding already done at action translation time. */
4061 static void
4062 commit_mpls_action(const struct flow *flow, struct flow *base,
4063 struct ofpbuf *odp_actions)
4064 {
4065 int base_n = flow_count_mpls_labels(base, NULL);
4066 int flow_n = flow_count_mpls_labels(flow, NULL);
4067 int common_n = flow_count_common_mpls_labels(flow, flow_n, base, base_n,
4068 NULL);
4069
4070 while (base_n > common_n) {
4071 if (base_n - 1 == common_n && flow_n > common_n) {
4072 /* If there is only one more LSE in base than there are common
4073 * between base and flow; and flow has at least one more LSE than
4074 * is common then the topmost LSE of base may be updated using
4075 * set */
4076 struct ovs_key_mpls mpls_key;
4077
4078 mpls_key.mpls_lse = flow->mpls_lse[flow_n - base_n];
4079 commit_set_action(odp_actions, OVS_KEY_ATTR_MPLS,
4080 &mpls_key, sizeof mpls_key);
4081 flow_set_mpls_lse(base, 0, mpls_key.mpls_lse);
4082 common_n++;
4083 } else {
4084 /* Otherwise, if there more LSEs in base than are common between
4085 * base and flow then pop the topmost one. */
4086 ovs_be16 dl_type;
4087 bool popped;
4088
4089 /* If all the LSEs are to be popped and this is not the outermost
4090 * LSE then use ETH_TYPE_MPLS as the ethertype parameter of the
4091 * POP_MPLS action instead of flow->dl_type.
4092 *
4093 * This is because the POP_MPLS action requires its ethertype
4094 * argument to be an MPLS ethernet type but in this case
4095 * flow->dl_type will be a non-MPLS ethernet type.
4096 *
4097 * When the final POP_MPLS action occurs it use flow->dl_type and
4098 * the and the resulting packet will have the desired dl_type. */
4099 if ((!eth_type_mpls(flow->dl_type)) && base_n > 1) {
4100 dl_type = htons(ETH_TYPE_MPLS);
4101 } else {
4102 dl_type = flow->dl_type;
4103 }
4104 nl_msg_put_be16(odp_actions, OVS_ACTION_ATTR_POP_MPLS, dl_type);
4105 popped = flow_pop_mpls(base, base_n, flow->dl_type, NULL);
4106 ovs_assert(popped);
4107 base_n--;
4108 }
4109 }
4110
4111 /* If, after the above popping and setting, there are more LSEs in flow
4112 * than base then some LSEs need to be pushed. */
4113 while (base_n < flow_n) {
4114 struct ovs_action_push_mpls *mpls;
4115
4116 mpls = nl_msg_put_unspec_zero(odp_actions,
4117 OVS_ACTION_ATTR_PUSH_MPLS,
4118 sizeof *mpls);
4119 mpls->mpls_ethertype = flow->dl_type;
4120 mpls->mpls_lse = flow->mpls_lse[flow_n - base_n - 1];
4121 flow_push_mpls(base, base_n, mpls->mpls_ethertype, NULL);
4122 flow_set_mpls_lse(base, 0, mpls->mpls_lse);
4123 base_n++;
4124 }
4125 }
4126
4127 static void
4128 get_ipv4_key(const struct flow *flow, struct ovs_key_ipv4 *ipv4, bool is_mask)
4129 {
4130 ipv4->ipv4_src = flow->nw_src;
4131 ipv4->ipv4_dst = flow->nw_dst;
4132 ipv4->ipv4_proto = flow->nw_proto;
4133 ipv4->ipv4_tos = flow->nw_tos;
4134 ipv4->ipv4_ttl = flow->nw_ttl;
4135 ipv4->ipv4_frag = ovs_to_odp_frag(flow->nw_frag, is_mask);
4136 }
4137
4138 static void
4139 put_ipv4_key(const struct ovs_key_ipv4 *ipv4, struct flow *flow, bool is_mask)
4140 {
4141 flow->nw_src = ipv4->ipv4_src;
4142 flow->nw_dst = ipv4->ipv4_dst;
4143 flow->nw_proto = ipv4->ipv4_proto;
4144 flow->nw_tos = ipv4->ipv4_tos;
4145 flow->nw_ttl = ipv4->ipv4_ttl;
4146 flow->nw_frag = odp_to_ovs_frag(ipv4->ipv4_frag, is_mask);
4147 }
4148
4149 static void
4150 commit_set_ipv4_action(const struct flow *flow, struct flow *base_flow,
4151 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4152 bool use_masked)
4153 {
4154 struct ovs_key_ipv4 key, mask, base;
4155
4156 /* Check that nw_proto and nw_frag remain unchanged. */
4157 ovs_assert(flow->nw_proto == base_flow->nw_proto &&
4158 flow->nw_frag == base_flow->nw_frag);
4159
4160 get_ipv4_key(flow, &key, false);
4161 get_ipv4_key(base_flow, &base, false);
4162 get_ipv4_key(&wc->masks, &mask, true);
4163 mask.ipv4_proto = 0; /* Not writeable. */
4164 mask.ipv4_frag = 0; /* Not writable. */
4165
4166 if (commit(OVS_KEY_ATTR_IPV4, use_masked, &key, &base, &mask, sizeof key,
4167 odp_actions)) {
4168 put_ipv4_key(&base, base_flow, false);
4169 if (mask.ipv4_proto != 0) { /* Mask was changed by commit(). */
4170 put_ipv4_key(&mask, &wc->masks, true);
4171 }
4172 }
4173 }
4174
4175 static void
4176 get_ipv6_key(const struct flow *flow, struct ovs_key_ipv6 *ipv6, bool is_mask)
4177 {
4178 memcpy(ipv6->ipv6_src, &flow->ipv6_src, sizeof ipv6->ipv6_src);
4179 memcpy(ipv6->ipv6_dst, &flow->ipv6_dst, sizeof ipv6->ipv6_dst);
4180 ipv6->ipv6_label = flow->ipv6_label;
4181 ipv6->ipv6_proto = flow->nw_proto;
4182 ipv6->ipv6_tclass = flow->nw_tos;
4183 ipv6->ipv6_hlimit = flow->nw_ttl;
4184 ipv6->ipv6_frag = ovs_to_odp_frag(flow->nw_frag, is_mask);
4185 }
4186
4187 static void
4188 put_ipv6_key(const struct ovs_key_ipv6 *ipv6, struct flow *flow, bool is_mask)
4189 {
4190 memcpy(&flow->ipv6_src, ipv6->ipv6_src, sizeof flow->ipv6_src);
4191 memcpy(&flow->ipv6_dst, ipv6->ipv6_dst, sizeof flow->ipv6_dst);
4192 flow->ipv6_label = ipv6->ipv6_label;
4193 flow->nw_proto = ipv6->ipv6_proto;
4194 flow->nw_tos = ipv6->ipv6_tclass;
4195 flow->nw_ttl = ipv6->ipv6_hlimit;
4196 flow->nw_frag = odp_to_ovs_frag(ipv6->ipv6_frag, is_mask);
4197 }
4198
4199 static void
4200 commit_set_ipv6_action(const struct flow *flow, struct flow *base_flow,
4201 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4202 bool use_masked)
4203 {
4204 struct ovs_key_ipv6 key, mask, base;
4205
4206 /* Check that nw_proto and nw_frag remain unchanged. */
4207 ovs_assert(flow->nw_proto == base_flow->nw_proto &&
4208 flow->nw_frag == base_flow->nw_frag);
4209
4210 get_ipv6_key(flow, &key, false);
4211 get_ipv6_key(base_flow, &base, false);
4212 get_ipv6_key(&wc->masks, &mask, true);
4213 mask.ipv6_proto = 0; /* Not writeable. */
4214 mask.ipv6_frag = 0; /* Not writable. */
4215
4216 if (commit(OVS_KEY_ATTR_IPV6, use_masked, &key, &base, &mask, sizeof key,
4217 odp_actions)) {
4218 put_ipv6_key(&base, base_flow, false);
4219 if (mask.ipv6_proto != 0) { /* Mask was changed by commit(). */
4220 put_ipv6_key(&mask, &wc->masks, true);
4221 }
4222 }
4223 }
4224
4225 static void
4226 get_arp_key(const struct flow *flow, struct ovs_key_arp *arp)
4227 {
4228 /* ARP key has padding, clear it. */
4229 memset(arp, 0, sizeof *arp);
4230
4231 arp->arp_sip = flow->nw_src;
4232 arp->arp_tip = flow->nw_dst;
4233 arp->arp_op = htons(flow->nw_proto);
4234 memcpy(arp->arp_sha, flow->arp_sha, ETH_ADDR_LEN);
4235 memcpy(arp->arp_tha, flow->arp_tha, ETH_ADDR_LEN);
4236 }
4237
4238 static void
4239 put_arp_key(const struct ovs_key_arp *arp, struct flow *flow)
4240 {
4241 flow->nw_src = arp->arp_sip;
4242 flow->nw_dst = arp->arp_tip;
4243 flow->nw_proto = ntohs(arp->arp_op);
4244 memcpy(flow->arp_sha, arp->arp_sha, ETH_ADDR_LEN);
4245 memcpy(flow->arp_tha, arp->arp_tha, ETH_ADDR_LEN);
4246 }
4247
4248 static enum slow_path_reason
4249 commit_set_arp_action(const struct flow *flow, struct flow *base_flow,
4250 struct ofpbuf *odp_actions, struct flow_wildcards *wc)
4251 {
4252 struct ovs_key_arp key, mask, base;
4253
4254 get_arp_key(flow, &key);
4255 get_arp_key(base_flow, &base);
4256 get_arp_key(&wc->masks, &mask);
4257
4258 if (commit(OVS_KEY_ATTR_ARP, true, &key, &base, &mask, sizeof key,
4259 odp_actions)) {
4260 put_arp_key(&base, base_flow);
4261 put_arp_key(&mask, &wc->masks);
4262 return SLOW_ACTION;
4263 }
4264 return 0;
4265 }
4266
4267 static void
4268 get_nd_key(const struct flow *flow, struct ovs_key_nd *nd)
4269 {
4270 memcpy(nd->nd_target, &flow->nd_target, sizeof flow->nd_target);
4271 /* nd_sll and nd_tll are stored in arp_sha and arp_tha, respectively */
4272 memcpy(nd->nd_sll, flow->arp_sha, ETH_ADDR_LEN);
4273 memcpy(nd->nd_tll, flow->arp_tha, ETH_ADDR_LEN);
4274 }
4275
4276 static void
4277 put_nd_key(const struct ovs_key_nd *nd, struct flow *flow)
4278 {
4279 memcpy(&flow->nd_target, &flow->nd_target, sizeof flow->nd_target);
4280 /* nd_sll and nd_tll are stored in arp_sha and arp_tha, respectively */
4281 memcpy(flow->arp_sha, nd->nd_sll, ETH_ADDR_LEN);
4282 memcpy(flow->arp_tha, nd->nd_tll, ETH_ADDR_LEN);
4283 }
4284
4285 static enum slow_path_reason
4286 commit_set_nd_action(const struct flow *flow, struct flow *base_flow,
4287 struct ofpbuf *odp_actions,
4288 struct flow_wildcards *wc, bool use_masked)
4289 {
4290 struct ovs_key_nd key, mask, base;
4291
4292 get_nd_key(flow, &key);
4293 get_nd_key(base_flow, &base);
4294 get_nd_key(&wc->masks, &mask);
4295
4296 if (commit(OVS_KEY_ATTR_ND, use_masked, &key, &base, &mask, sizeof key,
4297 odp_actions)) {
4298 put_nd_key(&base, base_flow);
4299 put_nd_key(&mask, &wc->masks);
4300 return SLOW_ACTION;
4301 }
4302
4303 return 0;
4304 }
4305
4306 static enum slow_path_reason
4307 commit_set_nw_action(const struct flow *flow, struct flow *base,
4308 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4309 bool use_masked)
4310 {
4311 /* Check if 'flow' really has an L3 header. */
4312 if (!flow->nw_proto) {
4313 return 0;
4314 }
4315
4316 switch (ntohs(base->dl_type)) {
4317 case ETH_TYPE_IP:
4318 commit_set_ipv4_action(flow, base, odp_actions, wc, use_masked);
4319 break;
4320
4321 case ETH_TYPE_IPV6:
4322 commit_set_ipv6_action(flow, base, odp_actions, wc, use_masked);
4323 return commit_set_nd_action(flow, base, odp_actions, wc, use_masked);
4324
4325 case ETH_TYPE_ARP:
4326 return commit_set_arp_action(flow, base, odp_actions, wc);
4327 }
4328
4329 return 0;
4330 }
4331
4332 /* TCP, UDP, and SCTP keys have the same layout. */
4333 BUILD_ASSERT_DECL(sizeof(struct ovs_key_tcp) == sizeof(struct ovs_key_udp) &&
4334 sizeof(struct ovs_key_tcp) == sizeof(struct ovs_key_sctp));
4335
4336 static void
4337 get_tp_key(const struct flow *flow, union ovs_key_tp *tp)
4338 {
4339 tp->tcp.tcp_src = flow->tp_src;
4340 tp->tcp.tcp_dst = flow->tp_dst;
4341 }
4342
4343 static void
4344 put_tp_key(const union ovs_key_tp *tp, struct flow *flow)
4345 {
4346 flow->tp_src = tp->tcp.tcp_src;
4347 flow->tp_dst = tp->tcp.tcp_dst;
4348 }
4349
4350 static void
4351 commit_set_port_action(const struct flow *flow, struct flow *base_flow,
4352 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4353 bool use_masked)
4354 {
4355 enum ovs_key_attr key_type;
4356 union ovs_key_tp key, mask, base;
4357
4358 /* Check if 'flow' really has an L3 header. */
4359 if (!flow->nw_proto) {
4360 return;
4361 }
4362
4363 if (!is_ip_any(base_flow)) {
4364 return;
4365 }
4366
4367 if (flow->nw_proto == IPPROTO_TCP) {
4368 key_type = OVS_KEY_ATTR_TCP;
4369 } else if (flow->nw_proto == IPPROTO_UDP) {
4370 key_type = OVS_KEY_ATTR_UDP;
4371 } else if (flow->nw_proto == IPPROTO_SCTP) {
4372 key_type = OVS_KEY_ATTR_SCTP;
4373 } else {
4374 return;
4375 }
4376
4377 get_tp_key(flow, &key);
4378 get_tp_key(base_flow, &base);
4379 get_tp_key(&wc->masks, &mask);
4380
4381 if (commit(key_type, use_masked, &key, &base, &mask, sizeof key,
4382 odp_actions)) {
4383 put_tp_key(&base, base_flow);
4384 put_tp_key(&mask, &wc->masks);
4385 }
4386 }
4387
4388 static void
4389 commit_set_priority_action(const struct flow *flow, struct flow *base_flow,
4390 struct ofpbuf *odp_actions,
4391 struct flow_wildcards *wc,
4392 bool use_masked)
4393 {
4394 uint32_t key, mask, base;
4395
4396 key = flow->skb_priority;
4397 base = base_flow->skb_priority;
4398 mask = wc->masks.skb_priority;
4399
4400 if (commit(OVS_KEY_ATTR_PRIORITY, use_masked, &key, &base, &mask,
4401 sizeof key, odp_actions)) {
4402 base_flow->skb_priority = base;
4403 wc->masks.skb_priority = mask;
4404 }
4405 }
4406
4407 static void
4408 commit_set_pkt_mark_action(const struct flow *flow, struct flow *base_flow,
4409 struct ofpbuf *odp_actions,
4410 struct flow_wildcards *wc,
4411 bool use_masked)
4412 {
4413 uint32_t key, mask, base;
4414
4415 key = flow->pkt_mark;
4416 base = base_flow->pkt_mark;
4417 mask = wc->masks.pkt_mark;
4418
4419 if (commit(OVS_KEY_ATTR_SKB_MARK, use_masked, &key, &base, &mask,
4420 sizeof key, odp_actions)) {
4421 base_flow->pkt_mark = base;
4422 wc->masks.pkt_mark = mask;
4423 }
4424 }
4425
4426 /* If any of the flow key data that ODP actions can modify are different in
4427 * 'base' and 'flow', appends ODP actions to 'odp_actions' that change the flow
4428 * key from 'base' into 'flow', and then changes 'base' the same way. Does not
4429 * commit set_tunnel actions. Users should call commit_odp_tunnel_action()
4430 * in addition to this function if needed. Sets fields in 'wc' that are
4431 * used as part of the action.
4432 *
4433 * Returns a reason to force processing the flow's packets into the userspace
4434 * slow path, if there is one, otherwise 0. */
4435 enum slow_path_reason
4436 commit_odp_actions(const struct flow *flow, struct flow *base,
4437 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4438 bool use_masked)
4439 {
4440 enum slow_path_reason slow;
4441
4442 commit_set_ether_addr_action(flow, base, odp_actions, wc, use_masked);
4443 slow = commit_set_nw_action(flow, base, odp_actions, wc, use_masked);
4444 commit_set_port_action(flow, base, odp_actions, wc, use_masked);
4445 commit_mpls_action(flow, base, odp_actions);
4446 commit_vlan_action(flow->vlan_tci, base, odp_actions, wc);
4447 commit_set_priority_action(flow, base, odp_actions, wc, use_masked);
4448 commit_set_pkt_mark_action(flow, base, odp_actions, wc, use_masked);
4449
4450 return slow;
4451 }
openvswitch mirror