]> git.proxmox.com Git - ovs.git/blob - lib/odp-util.c
ofproto: Add NXM_NX_TUN_GBP_ID and NXM_NX_TUN_GBP_FLAGS
[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 void
514 format_odp_tnl_push_header(struct ds *ds, struct ovs_action_push_tnl *data)
515 {
516 const struct eth_header *eth;
517 const struct ip_header *ip;
518 const void *l3;
519
520 eth = (const struct eth_header *)data->header;
521
522 l3 = eth + 1;
523 ip = (const struct ip_header *)l3;
524
525 /* Ethernet */
526 ds_put_format(ds, "header(size=%"PRIu8",type=%"PRIu8",eth(dst=",
527 data->header_len, data->tnl_type);
528 ds_put_format(ds, ETH_ADDR_FMT, ETH_ADDR_ARGS(eth->eth_dst));
529 ds_put_format(ds, ",src=");
530 ds_put_format(ds, ETH_ADDR_FMT, ETH_ADDR_ARGS(eth->eth_src));
531 ds_put_format(ds, ",dl_type=0x%04"PRIx16"),", ntohs(eth->eth_type));
532
533 /* IPv4 */
534 ds_put_format(ds, "ipv4(src="IP_FMT",dst="IP_FMT",proto=%"PRIu8
535 ",tos=%#"PRIx8",ttl=%"PRIu8",frag=0x%"PRIx16"),",
536 IP_ARGS(get_16aligned_be32(&ip->ip_src)),
537 IP_ARGS(get_16aligned_be32(&ip->ip_dst)),
538 ip->ip_proto, ip->ip_tos,
539 ip->ip_ttl,
540 ip->ip_frag_off);
541
542 if (data->tnl_type == OVS_VPORT_TYPE_VXLAN) {
543 const struct vxlanhdr *vxh;
544 const struct udp_header *udp;
545
546 /* UDP */
547 udp = (const struct udp_header *) (ip + 1);
548 ds_put_format(ds, "udp(src=%"PRIu16",dst=%"PRIu16"),",
549 ntohs(udp->udp_src), ntohs(udp->udp_dst));
550
551 /* VxLan */
552 vxh = (const struct vxlanhdr *) (udp + 1);
553 ds_put_format(ds, "vxlan(flags=0x%"PRIx32",vni=0x%"PRIx32")",
554 ntohl(get_16aligned_be32(&vxh->vx_flags)),
555 ntohl(get_16aligned_be32(&vxh->vx_vni)));
556 } else if (data->tnl_type == OVS_VPORT_TYPE_GRE) {
557 const struct gre_base_hdr *greh;
558 ovs_16aligned_be32 *options;
559 void *l4;
560
561 l4 = ((uint8_t *)l3 + sizeof(struct ip_header));
562 greh = (const struct gre_base_hdr *) l4;
563
564 ds_put_format(ds, "gre((flags=0x%"PRIx16",proto=0x%"PRIx16")",
565 greh->flags, ntohs(greh->protocol));
566 options = (ovs_16aligned_be32 *)(greh + 1);
567 if (greh->flags & htons(GRE_CSUM)) {
568 ds_put_format(ds, ",csum=0x%"PRIx32, ntohl(get_16aligned_be32(options)));
569 options++;
570 }
571 if (greh->flags & htons(GRE_KEY)) {
572 ds_put_format(ds, ",key=0x%"PRIx32, ntohl(get_16aligned_be32(options)));
573 options++;
574 }
575 if (greh->flags & htons(GRE_SEQ)) {
576 ds_put_format(ds, ",seq=0x%"PRIx32, ntohl(get_16aligned_be32(options)));
577 options++;
578 }
579 ds_put_format(ds, ")");
580 }
581 ds_put_format(ds, ")");
582 }
583
584 static void
585 format_odp_tnl_push_action(struct ds *ds, const struct nlattr *attr)
586 {
587 struct ovs_action_push_tnl *data;
588
589 data = (struct ovs_action_push_tnl *) nl_attr_get(attr);
590
591 ds_put_format(ds, "tnl_push(tnl_port(%"PRIu32"),", data->tnl_port);
592 format_odp_tnl_push_header(ds, data);
593 ds_put_format(ds, ",out_port(%"PRIu32"))", data->out_port);
594 }
595
596 static void
597 format_odp_action(struct ds *ds, const struct nlattr *a)
598 {
599 int expected_len;
600 enum ovs_action_attr type = nl_attr_type(a);
601 const struct ovs_action_push_vlan *vlan;
602 size_t size;
603
604 expected_len = odp_action_len(nl_attr_type(a));
605 if (expected_len != -2 && nl_attr_get_size(a) != expected_len) {
606 ds_put_format(ds, "bad length %"PRIuSIZE", expected %d for: ",
607 nl_attr_get_size(a), expected_len);
608 format_generic_odp_action(ds, a);
609 return;
610 }
611
612 switch (type) {
613 case OVS_ACTION_ATTR_OUTPUT:
614 ds_put_format(ds, "%"PRIu32, nl_attr_get_u32(a));
615 break;
616 case OVS_ACTION_ATTR_TUNNEL_POP:
617 ds_put_format(ds, "tnl_pop(%"PRIu32")", nl_attr_get_u32(a));
618 break;
619 case OVS_ACTION_ATTR_TUNNEL_PUSH:
620 format_odp_tnl_push_action(ds, a);
621 break;
622 case OVS_ACTION_ATTR_USERSPACE:
623 format_odp_userspace_action(ds, a);
624 break;
625 case OVS_ACTION_ATTR_RECIRC:
626 format_odp_recirc_action(ds, nl_attr_get_u32(a));
627 break;
628 case OVS_ACTION_ATTR_HASH:
629 format_odp_hash_action(ds, nl_attr_get(a));
630 break;
631 case OVS_ACTION_ATTR_SET_MASKED:
632 a = nl_attr_get(a);
633 size = nl_attr_get_size(a) / 2;
634 ds_put_cstr(ds, "set(");
635
636 /* Masked set action not supported for tunnel key, which is bigger. */
637 if (size <= sizeof(struct ovs_key_ipv6)) {
638 struct nlattr attr[1 + DIV_ROUND_UP(sizeof(struct ovs_key_ipv6),
639 sizeof(struct nlattr))];
640 struct nlattr mask[1 + DIV_ROUND_UP(sizeof(struct ovs_key_ipv6),
641 sizeof(struct nlattr))];
642
643 mask->nla_type = attr->nla_type = nl_attr_type(a);
644 mask->nla_len = attr->nla_len = NLA_HDRLEN + size;
645 memcpy(attr + 1, (char *)(a + 1), size);
646 memcpy(mask + 1, (char *)(a + 1) + size, size);
647 format_odp_key_attr(attr, mask, NULL, ds, false);
648 } else {
649 format_odp_key_attr(a, NULL, NULL, ds, false);
650 }
651 ds_put_cstr(ds, ")");
652 break;
653 case OVS_ACTION_ATTR_SET:
654 ds_put_cstr(ds, "set(");
655 format_odp_key_attr(nl_attr_get(a), NULL, NULL, ds, true);
656 ds_put_cstr(ds, ")");
657 break;
658 case OVS_ACTION_ATTR_PUSH_VLAN:
659 vlan = nl_attr_get(a);
660 ds_put_cstr(ds, "push_vlan(");
661 if (vlan->vlan_tpid != htons(ETH_TYPE_VLAN)) {
662 ds_put_format(ds, "tpid=0x%04"PRIx16",", ntohs(vlan->vlan_tpid));
663 }
664 format_vlan_tci(ds, vlan->vlan_tci, OVS_BE16_MAX, false);
665 ds_put_char(ds, ')');
666 break;
667 case OVS_ACTION_ATTR_POP_VLAN:
668 ds_put_cstr(ds, "pop_vlan");
669 break;
670 case OVS_ACTION_ATTR_PUSH_MPLS: {
671 const struct ovs_action_push_mpls *mpls = nl_attr_get(a);
672 ds_put_cstr(ds, "push_mpls(");
673 format_mpls_lse(ds, mpls->mpls_lse);
674 ds_put_format(ds, ",eth_type=0x%"PRIx16")", ntohs(mpls->mpls_ethertype));
675 break;
676 }
677 case OVS_ACTION_ATTR_POP_MPLS: {
678 ovs_be16 ethertype = nl_attr_get_be16(a);
679 ds_put_format(ds, "pop_mpls(eth_type=0x%"PRIx16")", ntohs(ethertype));
680 break;
681 }
682 case OVS_ACTION_ATTR_SAMPLE:
683 format_odp_sample_action(ds, a);
684 break;
685 case OVS_ACTION_ATTR_UNSPEC:
686 case __OVS_ACTION_ATTR_MAX:
687 default:
688 format_generic_odp_action(ds, a);
689 break;
690 }
691 }
692
693 void
694 format_odp_actions(struct ds *ds, const struct nlattr *actions,
695 size_t actions_len)
696 {
697 if (actions_len) {
698 const struct nlattr *a;
699 unsigned int left;
700
701 NL_ATTR_FOR_EACH (a, left, actions, actions_len) {
702 if (a != actions) {
703 ds_put_char(ds, ',');
704 }
705 format_odp_action(ds, a);
706 }
707 if (left) {
708 int i;
709
710 if (left == actions_len) {
711 ds_put_cstr(ds, "<empty>");
712 }
713 ds_put_format(ds, ",***%u leftover bytes*** (", left);
714 for (i = 0; i < left; i++) {
715 ds_put_format(ds, "%02x", ((const uint8_t *) a)[i]);
716 }
717 ds_put_char(ds, ')');
718 }
719 } else {
720 ds_put_cstr(ds, "drop");
721 }
722 }
723
724 /* Separate out parse_odp_userspace_action() function. */
725 static int
726 parse_odp_userspace_action(const char *s, struct ofpbuf *actions)
727 {
728 uint32_t pid;
729 union user_action_cookie cookie;
730 struct ofpbuf buf;
731 odp_port_t tunnel_out_port;
732 int n = -1;
733 void *user_data = NULL;
734 size_t user_data_size = 0;
735
736 if (!ovs_scan(s, "userspace(pid=%"SCNi32"%n", &pid, &n)) {
737 return -EINVAL;
738 }
739
740 {
741 uint32_t output;
742 uint32_t probability;
743 uint32_t collector_set_id;
744 uint32_t obs_domain_id;
745 uint32_t obs_point_id;
746 int vid, pcp;
747 int n1 = -1;
748 if (ovs_scan(&s[n], ",sFlow(vid=%i,"
749 "pcp=%i,output=%"SCNi32")%n",
750 &vid, &pcp, &output, &n1)) {
751 uint16_t tci;
752
753 n += n1;
754 tci = vid | (pcp << VLAN_PCP_SHIFT);
755 if (tci) {
756 tci |= VLAN_CFI;
757 }
758
759 cookie.type = USER_ACTION_COOKIE_SFLOW;
760 cookie.sflow.vlan_tci = htons(tci);
761 cookie.sflow.output = output;
762 user_data = &cookie;
763 user_data_size = sizeof cookie.sflow;
764 } else if (ovs_scan(&s[n], ",slow_path(%n",
765 &n1)) {
766 int res;
767
768 n += n1;
769 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
770 cookie.slow_path.unused = 0;
771 cookie.slow_path.reason = 0;
772
773 res = parse_flags(&s[n], slow_path_reason_to_string,
774 &cookie.slow_path.reason,
775 SLOW_PATH_REASON_MASK, NULL);
776 if (res < 0 || s[n + res] != ')') {
777 return res;
778 }
779 n += res + 1;
780
781 user_data = &cookie;
782 user_data_size = sizeof cookie.slow_path;
783 } else if (ovs_scan(&s[n], ",flow_sample(probability=%"SCNi32","
784 "collector_set_id=%"SCNi32","
785 "obs_domain_id=%"SCNi32","
786 "obs_point_id=%"SCNi32")%n",
787 &probability, &collector_set_id,
788 &obs_domain_id, &obs_point_id, &n1)) {
789 n += n1;
790
791 cookie.type = USER_ACTION_COOKIE_FLOW_SAMPLE;
792 cookie.flow_sample.probability = probability;
793 cookie.flow_sample.collector_set_id = collector_set_id;
794 cookie.flow_sample.obs_domain_id = obs_domain_id;
795 cookie.flow_sample.obs_point_id = obs_point_id;
796 user_data = &cookie;
797 user_data_size = sizeof cookie.flow_sample;
798 } else if (ovs_scan(&s[n], ",ipfix(output_port=%"SCNi32")%n",
799 &output, &n1) ) {
800 n += n1;
801 cookie.type = USER_ACTION_COOKIE_IPFIX;
802 cookie.ipfix.output_odp_port = u32_to_odp(output);
803 user_data = &cookie;
804 user_data_size = sizeof cookie.ipfix;
805 } else if (ovs_scan(&s[n], ",userdata(%n",
806 &n1)) {
807 char *end;
808
809 n += n1;
810 ofpbuf_init(&buf, 16);
811 end = ofpbuf_put_hex(&buf, &s[n], NULL);
812 if (end[0] != ')') {
813 return -EINVAL;
814 }
815 user_data = ofpbuf_data(&buf);
816 user_data_size = ofpbuf_size(&buf);
817 n = (end + 1) - s;
818 }
819 }
820
821 {
822 int n1 = -1;
823 if (ovs_scan(&s[n], ",tunnel_out_port=%"SCNi32")%n",
824 &tunnel_out_port, &n1)) {
825 odp_put_userspace_action(pid, user_data, user_data_size, tunnel_out_port, actions);
826 return n + n1;
827 } else if (s[n] == ')') {
828 odp_put_userspace_action(pid, user_data, user_data_size, ODPP_NONE, actions);
829 return n + 1;
830 }
831 }
832
833 return -EINVAL;
834 }
835
836 static int
837 ovs_parse_tnl_push(const char *s, struct ovs_action_push_tnl *data)
838 {
839 struct eth_header *eth;
840 struct ip_header *ip;
841 struct udp_header *udp;
842 struct gre_base_hdr *greh;
843 uint16_t gre_proto, dl_type, udp_src, udp_dst;
844 ovs_be32 sip, dip;
845 uint32_t tnl_type = 0, header_len = 0;
846 void *l3, *l4;
847 int n = 0;
848
849 if (!ovs_scan_len(s, &n, "tnl_push(tnl_port(%"SCNi32"),", &data->tnl_port)) {
850 return -EINVAL;
851 }
852 eth = (struct eth_header *) data->header;
853 l3 = (data->header + sizeof *eth);
854 l4 = ((uint8_t *) l3 + sizeof (struct ip_header));
855 ip = (struct ip_header *) l3;
856 if (!ovs_scan_len(s, &n, "header(size=%"SCNi32",type=%"SCNi32","
857 "eth(dst="ETH_ADDR_SCAN_FMT",",
858 &data->header_len,
859 &data->tnl_type,
860 ETH_ADDR_SCAN_ARGS(eth->eth_dst))) {
861 return -EINVAL;
862 }
863
864 if (!ovs_scan_len(s, &n, "src="ETH_ADDR_SCAN_FMT",",
865 ETH_ADDR_SCAN_ARGS(eth->eth_src))) {
866 return -EINVAL;
867 }
868 if (!ovs_scan_len(s, &n, "dl_type=0x%"SCNx16"),", &dl_type)) {
869 return -EINVAL;
870 }
871 eth->eth_type = htons(dl_type);
872
873 /* IPv4 */
874 if (!ovs_scan_len(s, &n, "ipv4(src="IP_SCAN_FMT",dst="IP_SCAN_FMT",proto=%"SCNi8
875 ",tos=%"SCNi8",ttl=%"SCNi8",frag=0x%"SCNx16"),",
876 IP_SCAN_ARGS(&sip),
877 IP_SCAN_ARGS(&dip),
878 &ip->ip_proto, &ip->ip_tos,
879 &ip->ip_ttl, &ip->ip_frag_off)) {
880 return -EINVAL;
881 }
882 put_16aligned_be32(&ip->ip_src, sip);
883 put_16aligned_be32(&ip->ip_dst, dip);
884
885 /* Tunnel header */
886 udp = (struct udp_header *) l4;
887 greh = (struct gre_base_hdr *) l4;
888 if (ovs_scan_len(s, &n, "udp(src=%"SCNi16",dst=%"SCNi16"),",
889 &udp_src, &udp_dst)) {
890 struct vxlanhdr *vxh;
891 uint32_t vx_flags, vx_vni;
892
893 udp->udp_src = htons(udp_src);
894 udp->udp_dst = htons(udp_dst);
895 udp->udp_len = 0;
896 udp->udp_csum = 0;
897
898 vxh = (struct vxlanhdr *) (udp + 1);
899 if (!ovs_scan_len(s, &n, "vxlan(flags=0x%"SCNx32",vni=0x%"SCNx32"))",
900 &vx_flags, &vx_vni)) {
901 return -EINVAL;
902 }
903 put_16aligned_be32(&vxh->vx_flags, htonl(vx_flags));
904 put_16aligned_be32(&vxh->vx_vni, htonl(vx_vni));
905 tnl_type = OVS_VPORT_TYPE_VXLAN;
906 header_len = sizeof *eth + sizeof *ip +
907 sizeof *udp + sizeof *vxh;
908 } else if (ovs_scan_len(s, &n, "gre((flags=0x%"SCNx16",proto=0x%"SCNx16")",
909 &greh->flags, &gre_proto)){
910
911 tnl_type = OVS_VPORT_TYPE_GRE;
912 greh->protocol = htons(gre_proto);
913 ovs_16aligned_be32 *options = (ovs_16aligned_be32 *) (greh + 1);
914
915 if (greh->flags & htons(GRE_CSUM)) {
916 uint32_t csum;
917
918 if (!ovs_scan_len(s, &n, ",csum=0x%"SCNx32, &csum)) {
919 return -EINVAL;
920 }
921 put_16aligned_be32(options, htonl(csum));
922 options++;
923 }
924 if (greh->flags & htons(GRE_KEY)) {
925 uint32_t key;
926
927 if (!ovs_scan_len(s, &n, ",key=0x%"SCNx32, &key)) {
928 return -EINVAL;
929 }
930
931 put_16aligned_be32(options, htonl(key));
932 options++;
933 }
934 if (greh->flags & htons(GRE_SEQ)) {
935 uint32_t seq;
936
937 if (!ovs_scan_len(s, &n, ",seq=0x%"SCNx32, &seq)) {
938 return -EINVAL;
939 }
940 put_16aligned_be32(options, htonl(seq));
941 options++;
942 }
943
944 if (!ovs_scan_len(s, &n, "))")) {
945 return -EINVAL;
946 }
947
948 header_len = sizeof *eth + sizeof *ip +
949 ((uint8_t *) options - (uint8_t *) greh);
950 } else {
951 return -EINVAL;
952 }
953
954 /* check tunnel meta data. */
955 if (data->tnl_type != tnl_type) {
956 return -EINVAL;
957 }
958 if (data->header_len != header_len) {
959 return -EINVAL;
960 }
961
962 /* Out port */
963 if (!ovs_scan_len(s, &n, ",out_port(%"SCNi32"))", &data->out_port)) {
964 return -EINVAL;
965 }
966
967 return n;
968 }
969
970 static int
971 parse_odp_action(const char *s, const struct simap *port_names,
972 struct ofpbuf *actions)
973 {
974 {
975 uint32_t port;
976 int n;
977
978 if (ovs_scan(s, "%"SCNi32"%n", &port, &n)) {
979 nl_msg_put_u32(actions, OVS_ACTION_ATTR_OUTPUT, port);
980 return n;
981 }
982 }
983
984 if (port_names) {
985 int len = strcspn(s, delimiters);
986 struct simap_node *node;
987
988 node = simap_find_len(port_names, s, len);
989 if (node) {
990 nl_msg_put_u32(actions, OVS_ACTION_ATTR_OUTPUT, node->data);
991 return len;
992 }
993 }
994
995 {
996 uint32_t recirc_id;
997 int n = -1;
998
999 if (ovs_scan(s, "recirc(%"PRIu32")%n", &recirc_id, &n)) {
1000 nl_msg_put_u32(actions, OVS_ACTION_ATTR_RECIRC, recirc_id);
1001 return n;
1002 }
1003 }
1004
1005 if (!strncmp(s, "userspace(", 10)) {
1006 return parse_odp_userspace_action(s, actions);
1007 }
1008
1009 if (!strncmp(s, "set(", 4)) {
1010 size_t start_ofs;
1011 int retval;
1012 struct nlattr mask[128 / sizeof(struct nlattr)];
1013 struct ofpbuf maskbuf;
1014 struct nlattr *nested, *key;
1015 size_t size;
1016
1017 /* 'mask' is big enough to hold any key. */
1018 ofpbuf_use_stack(&maskbuf, mask, sizeof mask);
1019
1020 start_ofs = nl_msg_start_nested(actions, OVS_ACTION_ATTR_SET);
1021 retval = parse_odp_key_mask_attr(s + 4, port_names, actions, &maskbuf);
1022 if (retval < 0) {
1023 return retval;
1024 }
1025 if (s[retval + 4] != ')') {
1026 return -EINVAL;
1027 }
1028
1029 nested = ofpbuf_at_assert(actions, start_ofs, sizeof *nested);
1030 key = nested + 1;
1031
1032 size = nl_attr_get_size(mask);
1033 if (size == nl_attr_get_size(key)) {
1034 /* Change to masked set action if not fully masked. */
1035 if (!is_all_ones(mask + 1, size)) {
1036 key->nla_len += size;
1037 ofpbuf_put(actions, mask + 1, size);
1038 /* 'actions' may have been reallocated by ofpbuf_put(). */
1039 nested = ofpbuf_at_assert(actions, start_ofs, sizeof *nested);
1040 nested->nla_type = OVS_ACTION_ATTR_SET_MASKED;
1041 }
1042 }
1043
1044 nl_msg_end_nested(actions, start_ofs);
1045 return retval + 5;
1046 }
1047
1048 {
1049 struct ovs_action_push_vlan push;
1050 int tpid = ETH_TYPE_VLAN;
1051 int vid, pcp;
1052 int cfi = 1;
1053 int n = -1;
1054
1055 if (ovs_scan(s, "push_vlan(vid=%i,pcp=%i)%n", &vid, &pcp, &n)
1056 || ovs_scan(s, "push_vlan(vid=%i,pcp=%i,cfi=%i)%n",
1057 &vid, &pcp, &cfi, &n)
1058 || ovs_scan(s, "push_vlan(tpid=%i,vid=%i,pcp=%i)%n",
1059 &tpid, &vid, &pcp, &n)
1060 || ovs_scan(s, "push_vlan(tpid=%i,vid=%i,pcp=%i,cfi=%i)%n",
1061 &tpid, &vid, &pcp, &cfi, &n)) {
1062 push.vlan_tpid = htons(tpid);
1063 push.vlan_tci = htons((vid << VLAN_VID_SHIFT)
1064 | (pcp << VLAN_PCP_SHIFT)
1065 | (cfi ? VLAN_CFI : 0));
1066 nl_msg_put_unspec(actions, OVS_ACTION_ATTR_PUSH_VLAN,
1067 &push, sizeof push);
1068
1069 return n;
1070 }
1071 }
1072
1073 if (!strncmp(s, "pop_vlan", 8)) {
1074 nl_msg_put_flag(actions, OVS_ACTION_ATTR_POP_VLAN);
1075 return 8;
1076 }
1077
1078 {
1079 double percentage;
1080 int n = -1;
1081
1082 if (ovs_scan(s, "sample(sample=%lf%%,actions(%n", &percentage, &n)
1083 && percentage >= 0. && percentage <= 100.0) {
1084 size_t sample_ofs, actions_ofs;
1085 double probability;
1086
1087 probability = floor(UINT32_MAX * (percentage / 100.0) + .5);
1088 sample_ofs = nl_msg_start_nested(actions, OVS_ACTION_ATTR_SAMPLE);
1089 nl_msg_put_u32(actions, OVS_SAMPLE_ATTR_PROBABILITY,
1090 (probability <= 0 ? 0
1091 : probability >= UINT32_MAX ? UINT32_MAX
1092 : probability));
1093
1094 actions_ofs = nl_msg_start_nested(actions,
1095 OVS_SAMPLE_ATTR_ACTIONS);
1096 for (;;) {
1097 int retval;
1098
1099 n += strspn(s + n, delimiters);
1100 if (s[n] == ')') {
1101 break;
1102 }
1103
1104 retval = parse_odp_action(s + n, port_names, actions);
1105 if (retval < 0) {
1106 return retval;
1107 }
1108 n += retval;
1109 }
1110 nl_msg_end_nested(actions, actions_ofs);
1111 nl_msg_end_nested(actions, sample_ofs);
1112
1113 return s[n + 1] == ')' ? n + 2 : -EINVAL;
1114 }
1115 }
1116
1117 {
1118 uint32_t port;
1119 int n;
1120
1121 if (ovs_scan(s, "tnl_pop(%"SCNi32")%n", &port, &n)) {
1122 nl_msg_put_u32(actions, OVS_ACTION_ATTR_TUNNEL_POP, port);
1123 return n;
1124 }
1125 }
1126
1127 {
1128 struct ovs_action_push_tnl data;
1129 int n;
1130
1131 n = ovs_parse_tnl_push(s, &data);
1132 if (n > 0) {
1133 odp_put_tnl_push_action(actions, &data);
1134 return n;
1135 } else if (n < 0) {
1136 return n;
1137 }
1138 }
1139 return -EINVAL;
1140 }
1141
1142 /* Parses the string representation of datapath actions, in the format output
1143 * by format_odp_action(). Returns 0 if successful, otherwise a positive errno
1144 * value. On success, the ODP actions are appended to 'actions' as a series of
1145 * Netlink attributes. On failure, no data is appended to 'actions'. Either
1146 * way, 'actions''s data might be reallocated. */
1147 int
1148 odp_actions_from_string(const char *s, const struct simap *port_names,
1149 struct ofpbuf *actions)
1150 {
1151 size_t old_size;
1152
1153 if (!strcasecmp(s, "drop")) {
1154 return 0;
1155 }
1156
1157 old_size = ofpbuf_size(actions);
1158 for (;;) {
1159 int retval;
1160
1161 s += strspn(s, delimiters);
1162 if (!*s) {
1163 return 0;
1164 }
1165
1166 retval = parse_odp_action(s, port_names, actions);
1167 if (retval < 0 || !strchr(delimiters, s[retval])) {
1168 ofpbuf_set_size(actions, old_size);
1169 return -retval;
1170 }
1171 s += retval;
1172 }
1173
1174 return 0;
1175 }
1176 \f
1177 /* Returns the correct length of the payload for a flow key attribute of the
1178 * specified 'type', -1 if 'type' is unknown, or -2 if the attribute's payload
1179 * is variable length. */
1180 static int
1181 odp_flow_key_attr_len(uint16_t type)
1182 {
1183 if (type > OVS_KEY_ATTR_MAX) {
1184 return -1;
1185 }
1186
1187 switch ((enum ovs_key_attr) type) {
1188 case OVS_KEY_ATTR_ENCAP: return -2;
1189 case OVS_KEY_ATTR_PRIORITY: return 4;
1190 case OVS_KEY_ATTR_SKB_MARK: return 4;
1191 case OVS_KEY_ATTR_DP_HASH: return 4;
1192 case OVS_KEY_ATTR_RECIRC_ID: return 4;
1193 case OVS_KEY_ATTR_TUNNEL: return -2;
1194 case OVS_KEY_ATTR_IN_PORT: return 4;
1195 case OVS_KEY_ATTR_ETHERNET: return sizeof(struct ovs_key_ethernet);
1196 case OVS_KEY_ATTR_VLAN: return sizeof(ovs_be16);
1197 case OVS_KEY_ATTR_ETHERTYPE: return 2;
1198 case OVS_KEY_ATTR_MPLS: return -2;
1199 case OVS_KEY_ATTR_IPV4: return sizeof(struct ovs_key_ipv4);
1200 case OVS_KEY_ATTR_IPV6: return sizeof(struct ovs_key_ipv6);
1201 case OVS_KEY_ATTR_TCP: return sizeof(struct ovs_key_tcp);
1202 case OVS_KEY_ATTR_TCP_FLAGS: return 2;
1203 case OVS_KEY_ATTR_UDP: return sizeof(struct ovs_key_udp);
1204 case OVS_KEY_ATTR_SCTP: return sizeof(struct ovs_key_sctp);
1205 case OVS_KEY_ATTR_ICMP: return sizeof(struct ovs_key_icmp);
1206 case OVS_KEY_ATTR_ICMPV6: return sizeof(struct ovs_key_icmpv6);
1207 case OVS_KEY_ATTR_ARP: return sizeof(struct ovs_key_arp);
1208 case OVS_KEY_ATTR_ND: return sizeof(struct ovs_key_nd);
1209
1210 case OVS_KEY_ATTR_UNSPEC:
1211 case __OVS_KEY_ATTR_MAX:
1212 return -1;
1213 }
1214
1215 return -1;
1216 }
1217
1218 static void
1219 format_generic_odp_key(const struct nlattr *a, struct ds *ds)
1220 {
1221 size_t len = nl_attr_get_size(a);
1222 if (len) {
1223 const uint8_t *unspec;
1224 unsigned int i;
1225
1226 unspec = nl_attr_get(a);
1227 for (i = 0; i < len; i++) {
1228 if (i) {
1229 ds_put_char(ds, ' ');
1230 }
1231 ds_put_format(ds, "%02x", unspec[i]);
1232 }
1233 }
1234 }
1235
1236 static const char *
1237 ovs_frag_type_to_string(enum ovs_frag_type type)
1238 {
1239 switch (type) {
1240 case OVS_FRAG_TYPE_NONE:
1241 return "no";
1242 case OVS_FRAG_TYPE_FIRST:
1243 return "first";
1244 case OVS_FRAG_TYPE_LATER:
1245 return "later";
1246 case __OVS_FRAG_TYPE_MAX:
1247 default:
1248 return "<error>";
1249 }
1250 }
1251
1252 static int
1253 tunnel_key_attr_len(int type)
1254 {
1255 switch (type) {
1256 case OVS_TUNNEL_KEY_ATTR_ID: return 8;
1257 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC: return 4;
1258 case OVS_TUNNEL_KEY_ATTR_IPV4_DST: return 4;
1259 case OVS_TUNNEL_KEY_ATTR_TOS: return 1;
1260 case OVS_TUNNEL_KEY_ATTR_TTL: return 1;
1261 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT: return 0;
1262 case OVS_TUNNEL_KEY_ATTR_CSUM: return 0;
1263 case OVS_TUNNEL_KEY_ATTR_TP_SRC: return 2;
1264 case OVS_TUNNEL_KEY_ATTR_TP_DST: return 2;
1265 case OVS_TUNNEL_KEY_ATTR_OAM: return 0;
1266 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS: return -2;
1267 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS: return -2;
1268 case __OVS_TUNNEL_KEY_ATTR_MAX:
1269 return -1;
1270 }
1271 return -1;
1272 }
1273
1274 #define GENEVE_OPT(class, type) ((OVS_FORCE uint32_t)(class) << 8 | (type))
1275 static int
1276 parse_geneve_opts(const struct nlattr *attr)
1277 {
1278 int opts_len = nl_attr_get_size(attr);
1279 const struct geneve_opt *opt = nl_attr_get(attr);
1280
1281 while (opts_len > 0) {
1282 int len;
1283
1284 if (opts_len < sizeof(*opt)) {
1285 return -EINVAL;
1286 }
1287
1288 len = sizeof(*opt) + opt->length * 4;
1289 if (len > opts_len) {
1290 return -EINVAL;
1291 }
1292
1293 switch (GENEVE_OPT(opt->opt_class, opt->type)) {
1294 default:
1295 if (opt->type & GENEVE_CRIT_OPT_TYPE) {
1296 return -EINVAL;
1297 }
1298 };
1299
1300 opt = opt + len / sizeof(*opt);
1301 opts_len -= len;
1302 };
1303
1304 return 0;
1305 }
1306
1307 enum odp_key_fitness
1308 odp_tun_key_from_attr(const struct nlattr *attr, struct flow_tnl *tun)
1309 {
1310 unsigned int left;
1311 const struct nlattr *a;
1312 bool ttl = false;
1313 bool unknown = false;
1314
1315 NL_NESTED_FOR_EACH(a, left, attr) {
1316 uint16_t type = nl_attr_type(a);
1317 size_t len = nl_attr_get_size(a);
1318 int expected_len = tunnel_key_attr_len(type);
1319
1320 if (len != expected_len && expected_len >= 0) {
1321 return ODP_FIT_ERROR;
1322 }
1323
1324 switch (type) {
1325 case OVS_TUNNEL_KEY_ATTR_ID:
1326 tun->tun_id = nl_attr_get_be64(a);
1327 tun->flags |= FLOW_TNL_F_KEY;
1328 break;
1329 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1330 tun->ip_src = nl_attr_get_be32(a);
1331 break;
1332 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1333 tun->ip_dst = nl_attr_get_be32(a);
1334 break;
1335 case OVS_TUNNEL_KEY_ATTR_TOS:
1336 tun->ip_tos = nl_attr_get_u8(a);
1337 break;
1338 case OVS_TUNNEL_KEY_ATTR_TTL:
1339 tun->ip_ttl = nl_attr_get_u8(a);
1340 ttl = true;
1341 break;
1342 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1343 tun->flags |= FLOW_TNL_F_DONT_FRAGMENT;
1344 break;
1345 case OVS_TUNNEL_KEY_ATTR_CSUM:
1346 tun->flags |= FLOW_TNL_F_CSUM;
1347 break;
1348 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
1349 tun->tp_src = nl_attr_get_be16(a);
1350 break;
1351 case OVS_TUNNEL_KEY_ATTR_TP_DST:
1352 tun->tp_dst = nl_attr_get_be16(a);
1353 break;
1354 case OVS_TUNNEL_KEY_ATTR_OAM:
1355 tun->flags |= FLOW_TNL_F_OAM;
1356 break;
1357 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS: {
1358 static const struct nl_policy vxlan_opts_policy[] = {
1359 [OVS_VXLAN_EXT_GBP] = { .type = NL_A_U32 },
1360 };
1361 struct nlattr *ext[ARRAY_SIZE(vxlan_opts_policy)];
1362
1363 if (!nl_parse_nested(a, vxlan_opts_policy, ext, ARRAY_SIZE(ext))) {
1364 return ODP_FIT_ERROR;
1365 }
1366
1367 if (ext[OVS_VXLAN_EXT_GBP]) {
1368 uint32_t gbp = nl_attr_get_u32(ext[OVS_VXLAN_EXT_GBP]);
1369
1370 tun->gbp_id = htons(gbp & 0xFFFF);
1371 tun->gbp_flags = (gbp >> 16) & 0xFF;
1372 }
1373
1374 break;
1375 }
1376 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS: {
1377 if (parse_geneve_opts(a)) {
1378 return ODP_FIT_ERROR;
1379 }
1380 /* It is necessary to reproduce options exactly (including order)
1381 * so it's easiest to just echo them back. */
1382 unknown = true;
1383 break;
1384 }
1385 default:
1386 /* Allow this to show up as unexpected, if there are unknown
1387 * tunnel attribute, eventually resulting in ODP_FIT_TOO_MUCH. */
1388 unknown = true;
1389 break;
1390 }
1391 }
1392
1393 if (!ttl) {
1394 return ODP_FIT_ERROR;
1395 }
1396 if (unknown) {
1397 return ODP_FIT_TOO_MUCH;
1398 }
1399 return ODP_FIT_PERFECT;
1400 }
1401
1402 static void
1403 tun_key_to_attr(struct ofpbuf *a, const struct flow_tnl *tun_key)
1404 {
1405 size_t tun_key_ofs;
1406
1407 tun_key_ofs = nl_msg_start_nested(a, OVS_KEY_ATTR_TUNNEL);
1408
1409 /* tun_id != 0 without FLOW_TNL_F_KEY is valid if tun_key is a mask. */
1410 if (tun_key->tun_id || tun_key->flags & FLOW_TNL_F_KEY) {
1411 nl_msg_put_be64(a, OVS_TUNNEL_KEY_ATTR_ID, tun_key->tun_id);
1412 }
1413 if (tun_key->ip_src) {
1414 nl_msg_put_be32(a, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, tun_key->ip_src);
1415 }
1416 if (tun_key->ip_dst) {
1417 nl_msg_put_be32(a, OVS_TUNNEL_KEY_ATTR_IPV4_DST, tun_key->ip_dst);
1418 }
1419 if (tun_key->ip_tos) {
1420 nl_msg_put_u8(a, OVS_TUNNEL_KEY_ATTR_TOS, tun_key->ip_tos);
1421 }
1422 nl_msg_put_u8(a, OVS_TUNNEL_KEY_ATTR_TTL, tun_key->ip_ttl);
1423 if (tun_key->flags & FLOW_TNL_F_DONT_FRAGMENT) {
1424 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT);
1425 }
1426 if (tun_key->flags & FLOW_TNL_F_CSUM) {
1427 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_CSUM);
1428 }
1429 if (tun_key->tp_src) {
1430 nl_msg_put_be16(a, OVS_TUNNEL_KEY_ATTR_TP_SRC, tun_key->tp_src);
1431 }
1432 if (tun_key->tp_dst) {
1433 nl_msg_put_be16(a, OVS_TUNNEL_KEY_ATTR_TP_DST, tun_key->tp_dst);
1434 }
1435 if (tun_key->flags & FLOW_TNL_F_OAM) {
1436 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_OAM);
1437 }
1438 if (tun_key->gbp_flags || tun_key->gbp_id) {
1439 size_t vxlan_opts_ofs;
1440
1441 vxlan_opts_ofs = nl_msg_start_nested(a, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
1442 nl_msg_put_u32(a, OVS_VXLAN_EXT_GBP,
1443 (tun_key->gbp_flags << 16) | ntohs(tun_key->gbp_id));
1444 nl_msg_end_nested(a, vxlan_opts_ofs);
1445 }
1446
1447 nl_msg_end_nested(a, tun_key_ofs);
1448 }
1449
1450 static bool
1451 odp_mask_attr_is_wildcard(const struct nlattr *ma)
1452 {
1453 return is_all_zeros(nl_attr_get(ma), nl_attr_get_size(ma));
1454 }
1455
1456 static bool
1457 odp_mask_is_exact(enum ovs_key_attr attr, const void *mask, size_t size)
1458 {
1459 if (attr == OVS_KEY_ATTR_TCP_FLAGS) {
1460 return TCP_FLAGS(*(ovs_be16 *)mask) == TCP_FLAGS(OVS_BE16_MAX);
1461 }
1462 if (attr == OVS_KEY_ATTR_IPV6) {
1463 const struct ovs_key_ipv6 *ipv6_mask = mask;
1464
1465 return
1466 ((ipv6_mask->ipv6_label & htonl(IPV6_LABEL_MASK))
1467 == htonl(IPV6_LABEL_MASK))
1468 && ipv6_mask->ipv6_proto == UINT8_MAX
1469 && ipv6_mask->ipv6_tclass == UINT8_MAX
1470 && ipv6_mask->ipv6_hlimit == UINT8_MAX
1471 && ipv6_mask->ipv6_frag == UINT8_MAX
1472 && ipv6_mask_is_exact((const struct in6_addr *)ipv6_mask->ipv6_src)
1473 && ipv6_mask_is_exact((const struct in6_addr *)ipv6_mask->ipv6_dst);
1474 }
1475 if (attr == OVS_KEY_ATTR_TUNNEL) {
1476 const struct flow_tnl *tun_mask = mask;
1477
1478 return tun_mask->flags == FLOW_TNL_F_MASK
1479 && tun_mask->tun_id == OVS_BE64_MAX
1480 && tun_mask->ip_src == OVS_BE32_MAX
1481 && tun_mask->ip_dst == OVS_BE32_MAX
1482 && tun_mask->ip_tos == UINT8_MAX
1483 && tun_mask->ip_ttl == UINT8_MAX
1484 && tun_mask->tp_src == OVS_BE16_MAX
1485 && tun_mask->tp_dst == OVS_BE16_MAX
1486 && tun_mask->gbp_id == OVS_BE16_MAX
1487 && tun_mask->gbp_flags == UINT8_MAX;
1488 }
1489
1490 if (attr == OVS_KEY_ATTR_ARP) {
1491 /* ARP key has padding, ignore it. */
1492 BUILD_ASSERT_DECL(sizeof(struct ovs_key_arp) == 24);
1493 BUILD_ASSERT_DECL(offsetof(struct ovs_key_arp, arp_tha) == 10 + 6);
1494 size = offsetof(struct ovs_key_arp, arp_tha) + ETH_ADDR_LEN;
1495 ovs_assert(((uint16_t *)mask)[size/2] == 0);
1496 }
1497
1498 return is_all_ones(mask, size);
1499 }
1500
1501 static bool
1502 odp_mask_attr_is_exact(const struct nlattr *ma)
1503 {
1504 struct flow_tnl tun_mask;
1505 enum ovs_key_attr attr = nl_attr_type(ma);
1506 const void *mask;
1507 size_t size;
1508
1509 if (attr == OVS_KEY_ATTR_TUNNEL) {
1510 memset(&tun_mask, 0, sizeof tun_mask);
1511 odp_tun_key_from_attr(ma, &tun_mask);
1512 mask = &tun_mask;
1513 size = sizeof tun_mask;
1514 } else {
1515 mask = nl_attr_get(ma);
1516 size = nl_attr_get_size(ma);
1517 }
1518
1519 return odp_mask_is_exact(attr, mask, size);
1520 }
1521
1522 void
1523 odp_portno_names_set(struct hmap *portno_names, odp_port_t port_no,
1524 char *port_name)
1525 {
1526 struct odp_portno_names *odp_portno_names;
1527
1528 odp_portno_names = xmalloc(sizeof *odp_portno_names);
1529 odp_portno_names->port_no = port_no;
1530 odp_portno_names->name = xstrdup(port_name);
1531 hmap_insert(portno_names, &odp_portno_names->hmap_node,
1532 hash_odp_port(port_no));
1533 }
1534
1535 static char *
1536 odp_portno_names_get(const struct hmap *portno_names, odp_port_t port_no)
1537 {
1538 struct odp_portno_names *odp_portno_names;
1539
1540 HMAP_FOR_EACH_IN_BUCKET (odp_portno_names, hmap_node,
1541 hash_odp_port(port_no), portno_names) {
1542 if (odp_portno_names->port_no == port_no) {
1543 return odp_portno_names->name;
1544 }
1545 }
1546 return NULL;
1547 }
1548
1549 void
1550 odp_portno_names_destroy(struct hmap *portno_names)
1551 {
1552 struct odp_portno_names *odp_portno_names, *odp_portno_names_next;
1553 HMAP_FOR_EACH_SAFE (odp_portno_names, odp_portno_names_next,
1554 hmap_node, portno_names) {
1555 hmap_remove(portno_names, &odp_portno_names->hmap_node);
1556 free(odp_portno_names->name);
1557 free(odp_portno_names);
1558 }
1559 }
1560
1561 /* Format helpers. */
1562
1563 static void
1564 format_eth(struct ds *ds, const char *name, const uint8_t key[ETH_ADDR_LEN],
1565 const uint8_t (*mask)[ETH_ADDR_LEN], bool verbose)
1566 {
1567 bool mask_empty = mask && eth_addr_is_zero(*mask);
1568
1569 if (verbose || !mask_empty) {
1570 bool mask_full = !mask || eth_mask_is_exact(*mask);
1571
1572 if (mask_full) {
1573 ds_put_format(ds, "%s="ETH_ADDR_FMT",", name, ETH_ADDR_ARGS(key));
1574 } else {
1575 ds_put_format(ds, "%s=", name);
1576 eth_format_masked(key, *mask, ds);
1577 ds_put_char(ds, ',');
1578 }
1579 }
1580 }
1581
1582 static void
1583 format_be64(struct ds *ds, const char *name, ovs_be64 key,
1584 const ovs_be64 *mask, bool verbose)
1585 {
1586 bool mask_empty = mask && !*mask;
1587
1588 if (verbose || !mask_empty) {
1589 bool mask_full = !mask || *mask == OVS_BE64_MAX;
1590
1591 ds_put_format(ds, "%s=0x%"PRIx64, name, ntohll(key));
1592 if (!mask_full) { /* Partially masked. */
1593 ds_put_format(ds, "/%#"PRIx64, ntohll(*mask));
1594 }
1595 ds_put_char(ds, ',');
1596 }
1597 }
1598
1599 static void
1600 format_ipv4(struct ds *ds, const char *name, ovs_be32 key,
1601 const ovs_be32 *mask, bool verbose)
1602 {
1603 bool mask_empty = mask && !*mask;
1604
1605 if (verbose || !mask_empty) {
1606 bool mask_full = !mask || *mask == OVS_BE32_MAX;
1607
1608 ds_put_format(ds, "%s="IP_FMT, name, IP_ARGS(key));
1609 if (!mask_full) { /* Partially masked. */
1610 ds_put_format(ds, "/"IP_FMT, IP_ARGS(*mask));
1611 }
1612 ds_put_char(ds, ',');
1613 }
1614 }
1615
1616 static void
1617 format_ipv6(struct ds *ds, const char *name, const ovs_be32 key_[4],
1618 const ovs_be32 (*mask_)[4], bool verbose)
1619 {
1620 char buf[INET6_ADDRSTRLEN];
1621 const struct in6_addr *key = (const struct in6_addr *)key_;
1622 const struct in6_addr *mask = mask_ ? (const struct in6_addr *)*mask_
1623 : NULL;
1624 bool mask_empty = mask && ipv6_mask_is_any(mask);
1625
1626 if (verbose || !mask_empty) {
1627 bool mask_full = !mask || ipv6_mask_is_exact(mask);
1628
1629 inet_ntop(AF_INET6, key, buf, sizeof buf);
1630 ds_put_format(ds, "%s=%s", name, buf);
1631 if (!mask_full) { /* Partially masked. */
1632 inet_ntop(AF_INET6, mask, buf, sizeof buf);
1633 ds_put_format(ds, "/%s", buf);
1634 }
1635 ds_put_char(ds, ',');
1636 }
1637 }
1638
1639 static void
1640 format_ipv6_label(struct ds *ds, const char *name, ovs_be32 key,
1641 const ovs_be32 *mask, bool verbose)
1642 {
1643 bool mask_empty = mask && !*mask;
1644
1645 if (verbose || !mask_empty) {
1646 bool mask_full = !mask
1647 || (*mask & htonl(IPV6_LABEL_MASK)) == htonl(IPV6_LABEL_MASK);
1648
1649 ds_put_format(ds, "%s=%#"PRIx32, name, ntohl(key));
1650 if (!mask_full) { /* Partially masked. */
1651 ds_put_format(ds, "/%#"PRIx32, ntohl(*mask));
1652 }
1653 ds_put_char(ds, ',');
1654 }
1655 }
1656
1657 static void
1658 format_u8x(struct ds *ds, const char *name, uint8_t key,
1659 const uint8_t *mask, bool verbose)
1660 {
1661 bool mask_empty = mask && !*mask;
1662
1663 if (verbose || !mask_empty) {
1664 bool mask_full = !mask || *mask == UINT8_MAX;
1665
1666 ds_put_format(ds, "%s=%#"PRIx8, name, key);
1667 if (!mask_full) { /* Partially masked. */
1668 ds_put_format(ds, "/%#"PRIx8, *mask);
1669 }
1670 ds_put_char(ds, ',');
1671 }
1672 }
1673
1674 static void
1675 format_u8u(struct ds *ds, const char *name, uint8_t key,
1676 const uint8_t *mask, bool verbose)
1677 {
1678 bool mask_empty = mask && !*mask;
1679
1680 if (verbose || !mask_empty) {
1681 bool mask_full = !mask || *mask == UINT8_MAX;
1682
1683 ds_put_format(ds, "%s=%"PRIu8, name, key);
1684 if (!mask_full) { /* Partially masked. */
1685 ds_put_format(ds, "/%#"PRIx8, *mask);
1686 }
1687 ds_put_char(ds, ',');
1688 }
1689 }
1690
1691 static void
1692 format_be16(struct ds *ds, const char *name, ovs_be16 key,
1693 const ovs_be16 *mask, bool verbose)
1694 {
1695 bool mask_empty = mask && !*mask;
1696
1697 if (verbose || !mask_empty) {
1698 bool mask_full = !mask || *mask == OVS_BE16_MAX;
1699
1700 ds_put_format(ds, "%s=%"PRIu16, name, ntohs(key));
1701 if (!mask_full) { /* Partially masked. */
1702 ds_put_format(ds, "/%#"PRIx16, ntohs(*mask));
1703 }
1704 ds_put_char(ds, ',');
1705 }
1706 }
1707
1708 static void
1709 format_tun_flags(struct ds *ds, const char *name, uint16_t key,
1710 const uint16_t *mask, bool verbose)
1711 {
1712 bool mask_empty = mask && !*mask;
1713
1714 if (verbose || !mask_empty) {
1715 bool mask_full = !mask || (*mask & FLOW_TNL_F_MASK) == FLOW_TNL_F_MASK;
1716
1717 ds_put_cstr(ds, name);
1718 ds_put_char(ds, '(');
1719 if (!mask_full) { /* Partially masked. */
1720 format_flags_masked(ds, NULL, flow_tun_flag_to_string, key, *mask);
1721 } else { /* Fully masked. */
1722 format_flags(ds, flow_tun_flag_to_string, key, ',');
1723 }
1724 ds_put_cstr(ds, "),");
1725 }
1726 }
1727
1728 static void
1729 format_frag(struct ds *ds, const char *name, uint8_t key,
1730 const uint8_t *mask, bool verbose)
1731 {
1732 bool mask_empty = mask && !*mask;
1733
1734 /* ODP frag is an enumeration field; partial masks are not meaningful. */
1735 if (verbose || !mask_empty) {
1736 bool mask_full = !mask || *mask == UINT8_MAX;
1737
1738 if (!mask_full) { /* Partially masked. */
1739 ds_put_format(ds, "error: partial mask not supported for frag (%#"
1740 PRIx8"),", *mask);
1741 } else {
1742 ds_put_format(ds, "%s=%s,", name, ovs_frag_type_to_string(key));
1743 }
1744 }
1745 }
1746
1747 #define MASK(PTR, FIELD) PTR ? &PTR->FIELD : NULL
1748
1749 static void
1750 format_odp_key_attr(const struct nlattr *a, const struct nlattr *ma,
1751 const struct hmap *portno_names, struct ds *ds,
1752 bool verbose)
1753 {
1754 enum ovs_key_attr attr = nl_attr_type(a);
1755 char namebuf[OVS_KEY_ATTR_BUFSIZE];
1756 int expected_len;
1757 bool is_exact;
1758
1759 is_exact = ma ? odp_mask_attr_is_exact(ma) : true;
1760
1761 ds_put_cstr(ds, ovs_key_attr_to_string(attr, namebuf, sizeof namebuf));
1762
1763 {
1764 expected_len = odp_flow_key_attr_len(nl_attr_type(a));
1765 if (expected_len != -2) {
1766 bool bad_key_len = nl_attr_get_size(a) != expected_len;
1767 bool bad_mask_len = ma && nl_attr_get_size(ma) != expected_len;
1768
1769 if (bad_key_len || bad_mask_len) {
1770 if (bad_key_len) {
1771 ds_put_format(ds, "(bad key length %"PRIuSIZE", expected %d)(",
1772 nl_attr_get_size(a), expected_len);
1773 }
1774 format_generic_odp_key(a, ds);
1775 if (ma) {
1776 ds_put_char(ds, '/');
1777 if (bad_mask_len) {
1778 ds_put_format(ds, "(bad mask length %"PRIuSIZE", expected %d)(",
1779 nl_attr_get_size(ma), expected_len);
1780 }
1781 format_generic_odp_key(ma, ds);
1782 }
1783 ds_put_char(ds, ')');
1784 return;
1785 }
1786 }
1787 }
1788
1789 ds_put_char(ds, '(');
1790 switch (attr) {
1791 case OVS_KEY_ATTR_ENCAP:
1792 if (ma && nl_attr_get_size(ma) && nl_attr_get_size(a)) {
1793 odp_flow_format(nl_attr_get(a), nl_attr_get_size(a),
1794 nl_attr_get(ma), nl_attr_get_size(ma), NULL, ds,
1795 verbose);
1796 } else if (nl_attr_get_size(a)) {
1797 odp_flow_format(nl_attr_get(a), nl_attr_get_size(a), NULL, 0, NULL,
1798 ds, verbose);
1799 }
1800 break;
1801
1802 case OVS_KEY_ATTR_PRIORITY:
1803 case OVS_KEY_ATTR_SKB_MARK:
1804 case OVS_KEY_ATTR_DP_HASH:
1805 case OVS_KEY_ATTR_RECIRC_ID:
1806 ds_put_format(ds, "%#"PRIx32, nl_attr_get_u32(a));
1807 if (!is_exact) {
1808 ds_put_format(ds, "/%#"PRIx32, nl_attr_get_u32(ma));
1809 }
1810 break;
1811
1812 case OVS_KEY_ATTR_TUNNEL: {
1813 struct flow_tnl key, mask_;
1814 struct flow_tnl *mask = ma ? &mask_ : NULL;
1815
1816 if (mask) {
1817 memset(mask, 0, sizeof *mask);
1818 odp_tun_key_from_attr(ma, mask);
1819 }
1820 memset(&key, 0, sizeof key);
1821 if (odp_tun_key_from_attr(a, &key) == ODP_FIT_ERROR) {
1822 ds_put_format(ds, "error");
1823 return;
1824 }
1825 format_be64(ds, "tun_id", key.tun_id, MASK(mask, tun_id), verbose);
1826 format_ipv4(ds, "src", key.ip_src, MASK(mask, ip_src), verbose);
1827 format_ipv4(ds, "dst", key.ip_dst, MASK(mask, ip_dst), verbose);
1828 format_u8x(ds, "tos", key.ip_tos, MASK(mask, ip_tos), verbose);
1829 format_u8u(ds, "ttl", key.ip_ttl, MASK(mask, ip_ttl), verbose);
1830 format_be16(ds, "tp_src", key.tp_src, MASK(mask, tp_src), verbose);
1831 format_be16(ds, "tp_dst", key.tp_dst, MASK(mask, tp_dst), verbose);
1832 format_be16(ds, "gbp_id", key.gbp_id, MASK(mask, gbp_id), verbose);
1833 format_u8x(ds, "gbp_flags", key.gbp_flags, MASK(mask, gbp_flags), verbose);
1834 format_tun_flags(ds, "flags", key.flags, MASK(mask, flags), verbose);
1835 ds_chomp(ds, ',');
1836 break;
1837 }
1838 case OVS_KEY_ATTR_IN_PORT:
1839 if (portno_names && verbose && is_exact) {
1840 char *name = odp_portno_names_get(portno_names,
1841 u32_to_odp(nl_attr_get_u32(a)));
1842 if (name) {
1843 ds_put_format(ds, "%s", name);
1844 } else {
1845 ds_put_format(ds, "%"PRIu32, nl_attr_get_u32(a));
1846 }
1847 } else {
1848 ds_put_format(ds, "%"PRIu32, nl_attr_get_u32(a));
1849 if (!is_exact) {
1850 ds_put_format(ds, "/%#"PRIx32, nl_attr_get_u32(ma));
1851 }
1852 }
1853 break;
1854
1855 case OVS_KEY_ATTR_ETHERNET: {
1856 const struct ovs_key_ethernet *mask = ma ? nl_attr_get(ma) : NULL;
1857 const struct ovs_key_ethernet *key = nl_attr_get(a);
1858
1859 format_eth(ds, "src", key->eth_src, MASK(mask, eth_src), verbose);
1860 format_eth(ds, "dst", key->eth_dst, MASK(mask, eth_dst), verbose);
1861 ds_chomp(ds, ',');
1862 break;
1863 }
1864 case OVS_KEY_ATTR_VLAN:
1865 format_vlan_tci(ds, nl_attr_get_be16(a),
1866 ma ? nl_attr_get_be16(ma) : OVS_BE16_MAX, verbose);
1867 break;
1868
1869 case OVS_KEY_ATTR_MPLS: {
1870 const struct ovs_key_mpls *mpls_key = nl_attr_get(a);
1871 const struct ovs_key_mpls *mpls_mask = NULL;
1872 size_t size = nl_attr_get_size(a);
1873
1874 if (!size || size % sizeof *mpls_key) {
1875 ds_put_format(ds, "(bad key length %"PRIuSIZE")", size);
1876 return;
1877 }
1878 if (!is_exact) {
1879 mpls_mask = nl_attr_get(ma);
1880 if (size != nl_attr_get_size(ma)) {
1881 ds_put_format(ds, "(key length %"PRIuSIZE" != "
1882 "mask length %"PRIuSIZE")",
1883 size, nl_attr_get_size(ma));
1884 return;
1885 }
1886 }
1887 format_mpls(ds, mpls_key, mpls_mask, size / sizeof *mpls_key);
1888 break;
1889 }
1890 case OVS_KEY_ATTR_ETHERTYPE:
1891 ds_put_format(ds, "0x%04"PRIx16, ntohs(nl_attr_get_be16(a)));
1892 if (!is_exact) {
1893 ds_put_format(ds, "/0x%04"PRIx16, ntohs(nl_attr_get_be16(ma)));
1894 }
1895 break;
1896
1897 case OVS_KEY_ATTR_IPV4: {
1898 const struct ovs_key_ipv4 *key = nl_attr_get(a);
1899 const struct ovs_key_ipv4 *mask = ma ? nl_attr_get(ma) : NULL;
1900
1901 format_ipv4(ds, "src", key->ipv4_src, MASK(mask, ipv4_src), verbose);
1902 format_ipv4(ds, "dst", key->ipv4_dst, MASK(mask, ipv4_dst), verbose);
1903 format_u8u(ds, "proto", key->ipv4_proto, MASK(mask, ipv4_proto),
1904 verbose);
1905 format_u8x(ds, "tos", key->ipv4_tos, MASK(mask, ipv4_tos), verbose);
1906 format_u8u(ds, "ttl", key->ipv4_ttl, MASK(mask, ipv4_ttl), verbose);
1907 format_frag(ds, "frag", key->ipv4_frag, MASK(mask, ipv4_frag),
1908 verbose);
1909 ds_chomp(ds, ',');
1910 break;
1911 }
1912 case OVS_KEY_ATTR_IPV6: {
1913 const struct ovs_key_ipv6 *key = nl_attr_get(a);
1914 const struct ovs_key_ipv6 *mask = ma ? nl_attr_get(ma) : NULL;
1915
1916 format_ipv6(ds, "src", key->ipv6_src, MASK(mask, ipv6_src), verbose);
1917 format_ipv6(ds, "dst", key->ipv6_dst, MASK(mask, ipv6_dst), verbose);
1918 format_ipv6_label(ds, "label", key->ipv6_label, MASK(mask, ipv6_label),
1919 verbose);
1920 format_u8u(ds, "proto", key->ipv6_proto, MASK(mask, ipv6_proto),
1921 verbose);
1922 format_u8x(ds, "tclass", key->ipv6_tclass, MASK(mask, ipv6_tclass),
1923 verbose);
1924 format_u8u(ds, "hlimit", key->ipv6_hlimit, MASK(mask, ipv6_hlimit),
1925 verbose);
1926 format_frag(ds, "frag", key->ipv6_frag, MASK(mask, ipv6_frag),
1927 verbose);
1928 ds_chomp(ds, ',');
1929 break;
1930 }
1931 /* These have the same structure and format. */
1932 case OVS_KEY_ATTR_TCP:
1933 case OVS_KEY_ATTR_UDP:
1934 case OVS_KEY_ATTR_SCTP: {
1935 const struct ovs_key_tcp *key = nl_attr_get(a);
1936 const struct ovs_key_tcp *mask = ma ? nl_attr_get(ma) : NULL;
1937
1938 format_be16(ds, "src", key->tcp_src, MASK(mask, tcp_src), verbose);
1939 format_be16(ds, "dst", key->tcp_dst, MASK(mask, tcp_dst), verbose);
1940 ds_chomp(ds, ',');
1941 break;
1942 }
1943 case OVS_KEY_ATTR_TCP_FLAGS:
1944 if (!is_exact) {
1945 format_flags_masked(ds, NULL, packet_tcp_flag_to_string,
1946 ntohs(nl_attr_get_be16(a)),
1947 ntohs(nl_attr_get_be16(ma)));
1948 } else {
1949 format_flags(ds, packet_tcp_flag_to_string,
1950 ntohs(nl_attr_get_be16(a)), ',');
1951 }
1952 break;
1953
1954 case OVS_KEY_ATTR_ICMP: {
1955 const struct ovs_key_icmp *key = nl_attr_get(a);
1956 const struct ovs_key_icmp *mask = ma ? nl_attr_get(ma) : NULL;
1957
1958 format_u8u(ds, "type", key->icmp_type, MASK(mask, icmp_type), verbose);
1959 format_u8u(ds, "code", key->icmp_code, MASK(mask, icmp_code), verbose);
1960 ds_chomp(ds, ',');
1961 break;
1962 }
1963 case OVS_KEY_ATTR_ICMPV6: {
1964 const struct ovs_key_icmpv6 *key = nl_attr_get(a);
1965 const struct ovs_key_icmpv6 *mask = ma ? nl_attr_get(ma) : NULL;
1966
1967 format_u8u(ds, "type", key->icmpv6_type, MASK(mask, icmpv6_type),
1968 verbose);
1969 format_u8u(ds, "code", key->icmpv6_code, MASK(mask, icmpv6_code),
1970 verbose);
1971 ds_chomp(ds, ',');
1972 break;
1973 }
1974 case OVS_KEY_ATTR_ARP: {
1975 const struct ovs_key_arp *mask = ma ? nl_attr_get(ma) : NULL;
1976 const struct ovs_key_arp *key = nl_attr_get(a);
1977
1978 format_ipv4(ds, "sip", key->arp_sip, MASK(mask, arp_sip), verbose);
1979 format_ipv4(ds, "tip", key->arp_tip, MASK(mask, arp_tip), verbose);
1980 format_be16(ds, "op", key->arp_op, MASK(mask, arp_op), verbose);
1981 format_eth(ds, "sha", key->arp_sha, MASK(mask, arp_sha), verbose);
1982 format_eth(ds, "tha", key->arp_tha, MASK(mask, arp_tha), verbose);
1983 ds_chomp(ds, ',');
1984 break;
1985 }
1986 case OVS_KEY_ATTR_ND: {
1987 const struct ovs_key_nd *mask = ma ? nl_attr_get(ma) : NULL;
1988 const struct ovs_key_nd *key = nl_attr_get(a);
1989
1990 format_ipv6(ds, "target", key->nd_target, MASK(mask, nd_target),
1991 verbose);
1992 format_eth(ds, "sll", key->nd_sll, MASK(mask, nd_sll), verbose);
1993 format_eth(ds, "tll", key->nd_tll, MASK(mask, nd_tll), verbose);
1994
1995 ds_chomp(ds, ',');
1996 break;
1997 }
1998 case OVS_KEY_ATTR_UNSPEC:
1999 case __OVS_KEY_ATTR_MAX:
2000 default:
2001 format_generic_odp_key(a, ds);
2002 if (!is_exact) {
2003 ds_put_char(ds, '/');
2004 format_generic_odp_key(ma, ds);
2005 }
2006 break;
2007 }
2008 ds_put_char(ds, ')');
2009 }
2010
2011 static struct nlattr *
2012 generate_all_wildcard_mask(struct ofpbuf *ofp, const struct nlattr *key)
2013 {
2014 const struct nlattr *a;
2015 unsigned int left;
2016 int type = nl_attr_type(key);
2017 int size = nl_attr_get_size(key);
2018
2019 if (odp_flow_key_attr_len(type) >=0) {
2020 nl_msg_put_unspec_zero(ofp, type, size);
2021 } else {
2022 size_t nested_mask;
2023
2024 nested_mask = nl_msg_start_nested(ofp, type);
2025 NL_ATTR_FOR_EACH(a, left, key, nl_attr_get_size(key)) {
2026 generate_all_wildcard_mask(ofp, nl_attr_get(a));
2027 }
2028 nl_msg_end_nested(ofp, nested_mask);
2029 }
2030
2031 return ofpbuf_base(ofp);
2032 }
2033
2034 int
2035 odp_ufid_from_string(const char *s_, ovs_u128 *ufid)
2036 {
2037 const char *s = s_;
2038
2039 if (ovs_scan(s, "ufid:")) {
2040 size_t n;
2041
2042 s += 5;
2043 if (ovs_scan(s, "0x")) {
2044 s += 2;
2045 }
2046
2047 n = strspn(s, "0123456789abcdefABCDEF");
2048 if (n != 32) {
2049 return -EINVAL;
2050 }
2051
2052 if (!ovs_scan(s, "%16"SCNx64"%16"SCNx64, &ufid->u64.hi,
2053 &ufid->u64.lo)) {
2054 return -EINVAL;
2055 }
2056 s += n;
2057 s += strspn(s, delimiters);
2058
2059 return s - s_;
2060 }
2061
2062 return 0;
2063 }
2064
2065 void
2066 odp_format_ufid(const ovs_u128 *ufid, struct ds *ds)
2067 {
2068 ds_put_format(ds, "ufid:%016"PRIx64"%016"PRIx64, ufid->u64.hi,
2069 ufid->u64.lo);
2070 }
2071
2072 /* Appends to 'ds' a string representation of the 'key_len' bytes of
2073 * OVS_KEY_ATTR_* attributes in 'key'. If non-null, additionally formats the
2074 * 'mask_len' bytes of 'mask' which apply to 'key'. If 'portno_names' is
2075 * non-null and 'verbose' is true, translates odp port number to its name. */
2076 void
2077 odp_flow_format(const struct nlattr *key, size_t key_len,
2078 const struct nlattr *mask, size_t mask_len,
2079 const struct hmap *portno_names, struct ds *ds, bool verbose)
2080 {
2081 if (key_len) {
2082 const struct nlattr *a;
2083 unsigned int left;
2084 bool has_ethtype_key = false;
2085 const struct nlattr *ma = NULL;
2086 struct ofpbuf ofp;
2087 bool first_field = true;
2088
2089 ofpbuf_init(&ofp, 100);
2090 NL_ATTR_FOR_EACH (a, left, key, key_len) {
2091 bool is_nested_attr;
2092 bool is_wildcard = false;
2093 int attr_type = nl_attr_type(a);
2094
2095 if (attr_type == OVS_KEY_ATTR_ETHERTYPE) {
2096 has_ethtype_key = true;
2097 }
2098
2099 is_nested_attr = (odp_flow_key_attr_len(attr_type) == -2);
2100
2101 if (mask && mask_len) {
2102 ma = nl_attr_find__(mask, mask_len, nl_attr_type(a));
2103 is_wildcard = ma ? odp_mask_attr_is_wildcard(ma) : true;
2104 }
2105
2106 if (verbose || !is_wildcard || is_nested_attr) {
2107 if (is_wildcard && !ma) {
2108 ma = generate_all_wildcard_mask(&ofp, a);
2109 }
2110 if (!first_field) {
2111 ds_put_char(ds, ',');
2112 }
2113 format_odp_key_attr(a, ma, portno_names, ds, verbose);
2114 first_field = false;
2115 }
2116 ofpbuf_clear(&ofp);
2117 }
2118 ofpbuf_uninit(&ofp);
2119
2120 if (left) {
2121 int i;
2122
2123 if (left == key_len) {
2124 ds_put_cstr(ds, "<empty>");
2125 }
2126 ds_put_format(ds, ",***%u leftover bytes*** (", left);
2127 for (i = 0; i < left; i++) {
2128 ds_put_format(ds, "%02x", ((const uint8_t *) a)[i]);
2129 }
2130 ds_put_char(ds, ')');
2131 }
2132 if (!has_ethtype_key) {
2133 ma = nl_attr_find__(mask, mask_len, OVS_KEY_ATTR_ETHERTYPE);
2134 if (ma) {
2135 ds_put_format(ds, ",eth_type(0/0x%04"PRIx16")",
2136 ntohs(nl_attr_get_be16(ma)));
2137 }
2138 }
2139 } else {
2140 ds_put_cstr(ds, "<empty>");
2141 }
2142 }
2143
2144 /* Appends to 'ds' a string representation of the 'key_len' bytes of
2145 * OVS_KEY_ATTR_* attributes in 'key'. */
2146 void
2147 odp_flow_key_format(const struct nlattr *key,
2148 size_t key_len, struct ds *ds)
2149 {
2150 odp_flow_format(key, key_len, NULL, 0, NULL, ds, true);
2151 }
2152
2153 static bool
2154 ovs_frag_type_from_string(const char *s, enum ovs_frag_type *type)
2155 {
2156 if (!strcasecmp(s, "no")) {
2157 *type = OVS_FRAG_TYPE_NONE;
2158 } else if (!strcasecmp(s, "first")) {
2159 *type = OVS_FRAG_TYPE_FIRST;
2160 } else if (!strcasecmp(s, "later")) {
2161 *type = OVS_FRAG_TYPE_LATER;
2162 } else {
2163 return false;
2164 }
2165 return true;
2166 }
2167
2168 /* Parsing. */
2169
2170 static int
2171 scan_eth(const char *s, uint8_t (*key)[ETH_ADDR_LEN],
2172 uint8_t (*mask)[ETH_ADDR_LEN])
2173 {
2174 int n;
2175
2176 if (ovs_scan(s, ETH_ADDR_SCAN_FMT"%n", ETH_ADDR_SCAN_ARGS(*key), &n)) {
2177 int len = n;
2178
2179 if (mask) {
2180 if (ovs_scan(s + len, "/"ETH_ADDR_SCAN_FMT"%n",
2181 ETH_ADDR_SCAN_ARGS(*mask), &n)) {
2182 len += n;
2183 } else {
2184 memset(mask, 0xff, sizeof *mask);
2185 }
2186 }
2187 return len;
2188 }
2189 return 0;
2190 }
2191
2192 static int
2193 scan_ipv4(const char *s, ovs_be32 *key, ovs_be32 *mask)
2194 {
2195 int n;
2196
2197 if (ovs_scan(s, IP_SCAN_FMT"%n", IP_SCAN_ARGS(key), &n)) {
2198 int len = n;
2199
2200 if (mask) {
2201 if (ovs_scan(s + len, "/"IP_SCAN_FMT"%n",
2202 IP_SCAN_ARGS(mask), &n)) {
2203 len += n;
2204 } else {
2205 *mask = OVS_BE32_MAX;
2206 }
2207 }
2208 return len;
2209 }
2210 return 0;
2211 }
2212
2213 static int
2214 scan_ipv6(const char *s, ovs_be32 (*key)[4], ovs_be32 (*mask)[4])
2215 {
2216 int n;
2217 char ipv6_s[IPV6_SCAN_LEN + 1];
2218
2219 if (ovs_scan(s, IPV6_SCAN_FMT"%n", ipv6_s, &n)
2220 && inet_pton(AF_INET6, ipv6_s, key) == 1) {
2221 int len = n;
2222
2223 if (mask) {
2224 if (ovs_scan(s + len, "/"IPV6_SCAN_FMT"%n", ipv6_s, &n)
2225 && inet_pton(AF_INET6, ipv6_s, mask) == 1) {
2226 len += n;
2227 } else {
2228 memset(mask, 0xff, sizeof *mask);
2229 }
2230 }
2231 return len;
2232 }
2233 return 0;
2234 }
2235
2236 static int
2237 scan_ipv6_label(const char *s, ovs_be32 *key, ovs_be32 *mask)
2238 {
2239 int key_, mask_;
2240 int n;
2241
2242 if (ovs_scan(s, "%i%n", &key_, &n)
2243 && (key_ & ~IPV6_LABEL_MASK) == 0) {
2244 int len = n;
2245
2246 *key = htonl(key_);
2247 if (mask) {
2248 if (ovs_scan(s + len, "/%i%n", &mask_, &n)
2249 && (mask_ & ~IPV6_LABEL_MASK) == 0) {
2250 len += n;
2251 *mask = htonl(mask_);
2252 } else {
2253 *mask = htonl(IPV6_LABEL_MASK);
2254 }
2255 }
2256 return len;
2257 }
2258 return 0;
2259 }
2260
2261 static int
2262 scan_u8(const char *s, uint8_t *key, uint8_t *mask)
2263 {
2264 int n;
2265
2266 if (ovs_scan(s, "%"SCNi8"%n", key, &n)) {
2267 int len = n;
2268
2269 if (mask) {
2270 if (ovs_scan(s + len, "/%"SCNi8"%n", mask, &n)) {
2271 len += n;
2272 } else {
2273 *mask = UINT8_MAX;
2274 }
2275 }
2276 return len;
2277 }
2278 return 0;
2279 }
2280
2281 static int
2282 scan_u32(const char *s, uint32_t *key, uint32_t *mask)
2283 {
2284 int n;
2285
2286 if (ovs_scan(s, "%"SCNi32"%n", key, &n)) {
2287 int len = n;
2288
2289 if (mask) {
2290 if (ovs_scan(s + len, "/%"SCNi32"%n", mask, &n)) {
2291 len += n;
2292 } else {
2293 *mask = UINT32_MAX;
2294 }
2295 }
2296 return len;
2297 }
2298 return 0;
2299 }
2300
2301 static int
2302 scan_be16(const char *s, ovs_be16 *key, ovs_be16 *mask)
2303 {
2304 uint16_t key_, mask_;
2305 int n;
2306
2307 if (ovs_scan(s, "%"SCNi16"%n", &key_, &n)) {
2308 int len = n;
2309
2310 *key = htons(key_);
2311 if (mask) {
2312 if (ovs_scan(s + len, "/%"SCNi16"%n", &mask_, &n)) {
2313 len += n;
2314 *mask = htons(mask_);
2315 } else {
2316 *mask = OVS_BE16_MAX;
2317 }
2318 }
2319 return len;
2320 }
2321 return 0;
2322 }
2323
2324 static int
2325 scan_be64(const char *s, ovs_be64 *key, ovs_be64 *mask)
2326 {
2327 uint64_t key_, mask_;
2328 int n;
2329
2330 if (ovs_scan(s, "%"SCNi64"%n", &key_, &n)) {
2331 int len = n;
2332
2333 *key = htonll(key_);
2334 if (mask) {
2335 if (ovs_scan(s + len, "/%"SCNi64"%n", &mask_, &n)) {
2336 len += n;
2337 *mask = htonll(mask_);
2338 } else {
2339 *mask = OVS_BE64_MAX;
2340 }
2341 }
2342 return len;
2343 }
2344 return 0;
2345 }
2346
2347 static int
2348 scan_tun_flags(const char *s, uint16_t *key, uint16_t *mask)
2349 {
2350 uint32_t flags, fmask;
2351 int n;
2352
2353 n = parse_flags(s, flow_tun_flag_to_string, &flags,
2354 FLOW_TNL_F_MASK, mask ? &fmask : NULL);
2355 if (n >= 0 && s[n] == ')') {
2356 *key = flags;
2357 if (mask) {
2358 *mask = fmask;
2359 }
2360 return n + 1;
2361 }
2362 return 0;
2363 }
2364
2365 static int
2366 scan_tcp_flags(const char *s, ovs_be16 *key, ovs_be16 *mask)
2367 {
2368 uint32_t flags, fmask;
2369 int n;
2370
2371 n = parse_flags(s, packet_tcp_flag_to_string, &flags,
2372 TCP_FLAGS(OVS_BE16_MAX), mask ? &fmask : NULL);
2373 if (n >= 0) {
2374 *key = htons(flags);
2375 if (mask) {
2376 *mask = htons(fmask);
2377 }
2378 return n;
2379 }
2380 return 0;
2381 }
2382
2383 static int
2384 scan_frag(const char *s, uint8_t *key, uint8_t *mask)
2385 {
2386 int n;
2387 char frag[8];
2388 enum ovs_frag_type frag_type;
2389
2390 if (ovs_scan(s, "%7[a-z]%n", frag, &n)
2391 && ovs_frag_type_from_string(frag, &frag_type)) {
2392 int len = n;
2393
2394 *key = frag_type;
2395 if (mask) {
2396 *mask = UINT8_MAX;
2397 }
2398 return len;
2399 }
2400 return 0;
2401 }
2402
2403 static int
2404 scan_port(const char *s, uint32_t *key, uint32_t *mask,
2405 const struct simap *port_names)
2406 {
2407 int n;
2408
2409 if (ovs_scan(s, "%"SCNi32"%n", key, &n)) {
2410 int len = n;
2411
2412 if (mask) {
2413 if (ovs_scan(s + len, "/%"SCNi32"%n", mask, &n)) {
2414 len += n;
2415 } else {
2416 *mask = UINT32_MAX;
2417 }
2418 }
2419 return len;
2420 } else if (port_names) {
2421 const struct simap_node *node;
2422 int len;
2423
2424 len = strcspn(s, ")");
2425 node = simap_find_len(port_names, s, len);
2426 if (node) {
2427 *key = node->data;
2428
2429 if (mask) {
2430 *mask = UINT32_MAX;
2431 }
2432 return len;
2433 }
2434 }
2435 return 0;
2436 }
2437
2438 /* Helper for vlan parsing. */
2439 struct ovs_key_vlan__ {
2440 ovs_be16 tci;
2441 };
2442
2443 static bool
2444 set_be16_bf(ovs_be16 *bf, uint8_t bits, uint8_t offset, uint16_t value)
2445 {
2446 const uint16_t mask = ((1U << bits) - 1) << offset;
2447
2448 if (value >> bits) {
2449 return false;
2450 }
2451
2452 *bf = htons((ntohs(*bf) & ~mask) | (value << offset));
2453 return true;
2454 }
2455
2456 static int
2457 scan_be16_bf(const char *s, ovs_be16 *key, ovs_be16 *mask, uint8_t bits,
2458 uint8_t offset)
2459 {
2460 uint16_t key_, mask_;
2461 int n;
2462
2463 if (ovs_scan(s, "%"SCNi16"%n", &key_, &n)) {
2464 int len = n;
2465
2466 if (set_be16_bf(key, bits, offset, key_)) {
2467 if (mask) {
2468 if (ovs_scan(s + len, "/%"SCNi16"%n", &mask_, &n)) {
2469 len += n;
2470
2471 if (!set_be16_bf(mask, bits, offset, mask_)) {
2472 return 0;
2473 }
2474 } else {
2475 *mask |= htons(((1U << bits) - 1) << offset);
2476 }
2477 }
2478 return len;
2479 }
2480 }
2481 return 0;
2482 }
2483
2484 static int
2485 scan_vid(const char *s, ovs_be16 *key, ovs_be16 *mask)
2486 {
2487 return scan_be16_bf(s, key, mask, 12, VLAN_VID_SHIFT);
2488 }
2489
2490 static int
2491 scan_pcp(const char *s, ovs_be16 *key, ovs_be16 *mask)
2492 {
2493 return scan_be16_bf(s, key, mask, 3, VLAN_PCP_SHIFT);
2494 }
2495
2496 static int
2497 scan_cfi(const char *s, ovs_be16 *key, ovs_be16 *mask)
2498 {
2499 return scan_be16_bf(s, key, mask, 1, VLAN_CFI_SHIFT);
2500 }
2501
2502 /* For MPLS. */
2503 static bool
2504 set_be32_bf(ovs_be32 *bf, uint8_t bits, uint8_t offset, uint32_t value)
2505 {
2506 const uint32_t mask = ((1U << bits) - 1) << offset;
2507
2508 if (value >> bits) {
2509 return false;
2510 }
2511
2512 *bf = htonl((ntohl(*bf) & ~mask) | (value << offset));
2513 return true;
2514 }
2515
2516 static int
2517 scan_be32_bf(const char *s, ovs_be32 *key, ovs_be32 *mask, uint8_t bits,
2518 uint8_t offset)
2519 {
2520 uint32_t key_, mask_;
2521 int n;
2522
2523 if (ovs_scan(s, "%"SCNi32"%n", &key_, &n)) {
2524 int len = n;
2525
2526 if (set_be32_bf(key, bits, offset, key_)) {
2527 if (mask) {
2528 if (ovs_scan(s + len, "/%"SCNi32"%n", &mask_, &n)) {
2529 len += n;
2530
2531 if (!set_be32_bf(mask, bits, offset, mask_)) {
2532 return 0;
2533 }
2534 } else {
2535 *mask |= htonl(((1U << bits) - 1) << offset);
2536 }
2537 }
2538 return len;
2539 }
2540 }
2541 return 0;
2542 }
2543
2544 static int
2545 scan_mpls_label(const char *s, ovs_be32 *key, ovs_be32 *mask)
2546 {
2547 return scan_be32_bf(s, key, mask, 20, MPLS_LABEL_SHIFT);
2548 }
2549
2550 static int
2551 scan_mpls_tc(const char *s, ovs_be32 *key, ovs_be32 *mask)
2552 {
2553 return scan_be32_bf(s, key, mask, 3, MPLS_TC_SHIFT);
2554 }
2555
2556 static int
2557 scan_mpls_ttl(const char *s, ovs_be32 *key, ovs_be32 *mask)
2558 {
2559 return scan_be32_bf(s, key, mask, 8, MPLS_TTL_SHIFT);
2560 }
2561
2562 static int
2563 scan_mpls_bos(const char *s, ovs_be32 *key, ovs_be32 *mask)
2564 {
2565 return scan_be32_bf(s, key, mask, 1, MPLS_BOS_SHIFT);
2566 }
2567
2568 /* ATTR is compile-time constant, so only the case with correct data type
2569 * will be used. However, the compiler complains about the data type for
2570 * the other cases, so we must cast to make the compiler silent. */
2571 #define SCAN_PUT_ATTR(BUF, ATTR, DATA) \
2572 if ((ATTR) == OVS_KEY_ATTR_TUNNEL) { \
2573 tun_key_to_attr(BUF, (const struct flow_tnl *)(void *)&(DATA)); \
2574 } else { \
2575 nl_msg_put_unspec(BUF, ATTR, &(DATA), sizeof (DATA)); \
2576 }
2577
2578 #define SCAN_IF(NAME) \
2579 if (strncmp(s, NAME, strlen(NAME)) == 0) { \
2580 const char *start = s; \
2581 int len; \
2582 \
2583 s += strlen(NAME)
2584
2585 /* Usually no special initialization is needed. */
2586 #define SCAN_BEGIN(NAME, TYPE) \
2587 SCAN_IF(NAME); \
2588 TYPE skey, smask; \
2589 memset(&skey, 0, sizeof skey); \
2590 memset(&smask, 0, sizeof smask); \
2591 do { \
2592 len = 0;
2593
2594 /* VLAN needs special initialization. */
2595 #define SCAN_BEGIN_INIT(NAME, TYPE, KEY_INIT, MASK_INIT) \
2596 SCAN_IF(NAME); \
2597 TYPE skey = KEY_INIT; \
2598 TYPE smask = MASK_INIT; \
2599 do { \
2600 len = 0;
2601
2602 /* Scan unnamed entry as 'TYPE' */
2603 #define SCAN_TYPE(TYPE, KEY, MASK) \
2604 len = scan_##TYPE(s, KEY, MASK); \
2605 if (len == 0) { \
2606 return -EINVAL; \
2607 } \
2608 s += len
2609
2610 /* Scan named ('NAME') entry 'FIELD' as 'TYPE'. */
2611 #define SCAN_FIELD(NAME, TYPE, FIELD) \
2612 if (strncmp(s, NAME, strlen(NAME)) == 0) { \
2613 s += strlen(NAME); \
2614 SCAN_TYPE(TYPE, &skey.FIELD, mask ? &smask.FIELD : NULL); \
2615 continue; \
2616 }
2617
2618 #define SCAN_FINISH() \
2619 } while (*s++ == ',' && len != 0); \
2620 if (s[-1] != ')') { \
2621 return -EINVAL; \
2622 }
2623
2624 #define SCAN_FINISH_SINGLE() \
2625 } while (false); \
2626 if (*s++ != ')') { \
2627 return -EINVAL; \
2628 }
2629
2630 #define SCAN_PUT(ATTR) \
2631 if (!mask || !is_all_zeros(&smask, sizeof smask)) { \
2632 SCAN_PUT_ATTR(key, ATTR, skey); \
2633 if (mask) { \
2634 SCAN_PUT_ATTR(mask, ATTR, smask); \
2635 } \
2636 }
2637
2638 #define SCAN_END(ATTR) \
2639 SCAN_FINISH(); \
2640 SCAN_PUT(ATTR); \
2641 return s - start; \
2642 }
2643
2644 #define SCAN_END_SINGLE(ATTR) \
2645 SCAN_FINISH_SINGLE(); \
2646 SCAN_PUT(ATTR); \
2647 return s - start; \
2648 }
2649
2650 #define SCAN_SINGLE(NAME, TYPE, SCAN_AS, ATTR) \
2651 SCAN_BEGIN(NAME, TYPE) { \
2652 SCAN_TYPE(SCAN_AS, &skey, &smask); \
2653 } SCAN_END_SINGLE(ATTR)
2654
2655 #define SCAN_SINGLE_NO_MASK(NAME, TYPE, SCAN_AS, ATTR) \
2656 SCAN_BEGIN(NAME, TYPE) { \
2657 SCAN_TYPE(SCAN_AS, &skey, NULL); \
2658 } SCAN_END_SINGLE(ATTR)
2659
2660 /* scan_port needs one extra argument. */
2661 #define SCAN_SINGLE_PORT(NAME, TYPE, ATTR) \
2662 SCAN_BEGIN(NAME, TYPE) { \
2663 len = scan_port(s, &skey, &smask, port_names); \
2664 if (len == 0) { \
2665 return -EINVAL; \
2666 } \
2667 s += len; \
2668 } SCAN_END_SINGLE(ATTR)
2669
2670 static int
2671 parse_odp_key_mask_attr(const char *s, const struct simap *port_names,
2672 struct ofpbuf *key, struct ofpbuf *mask)
2673 {
2674 SCAN_SINGLE("skb_priority(", uint32_t, u32, OVS_KEY_ATTR_PRIORITY);
2675 SCAN_SINGLE("skb_mark(", uint32_t, u32, OVS_KEY_ATTR_SKB_MARK);
2676 SCAN_SINGLE_NO_MASK("recirc_id(", uint32_t, u32, OVS_KEY_ATTR_RECIRC_ID);
2677 SCAN_SINGLE("dp_hash(", uint32_t, u32, OVS_KEY_ATTR_DP_HASH);
2678
2679 SCAN_BEGIN("tunnel(", struct flow_tnl) {
2680 SCAN_FIELD("tun_id=", be64, tun_id);
2681 SCAN_FIELD("src=", ipv4, ip_src);
2682 SCAN_FIELD("dst=", ipv4, ip_dst);
2683 SCAN_FIELD("tos=", u8, ip_tos);
2684 SCAN_FIELD("ttl=", u8, ip_ttl);
2685 SCAN_FIELD("tp_src=", be16, tp_src);
2686 SCAN_FIELD("tp_dst=", be16, tp_dst);
2687 SCAN_FIELD("gbp_id=", be16, gbp_id);
2688 SCAN_FIELD("gbp_flags=", u8, gbp_flags);
2689 SCAN_FIELD("flags(", tun_flags, flags);
2690 } SCAN_END(OVS_KEY_ATTR_TUNNEL);
2691
2692 SCAN_SINGLE_PORT("in_port(", uint32_t, OVS_KEY_ATTR_IN_PORT);
2693
2694 SCAN_BEGIN("eth(", struct ovs_key_ethernet) {
2695 SCAN_FIELD("src=", eth, eth_src);
2696 SCAN_FIELD("dst=", eth, eth_dst);
2697 } SCAN_END(OVS_KEY_ATTR_ETHERNET);
2698
2699 SCAN_BEGIN_INIT("vlan(", struct ovs_key_vlan__,
2700 { htons(VLAN_CFI) }, { htons(VLAN_CFI) }) {
2701 SCAN_FIELD("vid=", vid, tci);
2702 SCAN_FIELD("pcp=", pcp, tci);
2703 SCAN_FIELD("cfi=", cfi, tci);
2704 } SCAN_END(OVS_KEY_ATTR_VLAN);
2705
2706 SCAN_SINGLE("eth_type(", ovs_be16, be16, OVS_KEY_ATTR_ETHERTYPE);
2707
2708 SCAN_BEGIN("mpls(", struct ovs_key_mpls) {
2709 SCAN_FIELD("label=", mpls_label, mpls_lse);
2710 SCAN_FIELD("tc=", mpls_tc, mpls_lse);
2711 SCAN_FIELD("ttl=", mpls_ttl, mpls_lse);
2712 SCAN_FIELD("bos=", mpls_bos, mpls_lse);
2713 } SCAN_END(OVS_KEY_ATTR_MPLS);
2714
2715 SCAN_BEGIN("ipv4(", struct ovs_key_ipv4) {
2716 SCAN_FIELD("src=", ipv4, ipv4_src);
2717 SCAN_FIELD("dst=", ipv4, ipv4_dst);
2718 SCAN_FIELD("proto=", u8, ipv4_proto);
2719 SCAN_FIELD("tos=", u8, ipv4_tos);
2720 SCAN_FIELD("ttl=", u8, ipv4_ttl);
2721 SCAN_FIELD("frag=", frag, ipv4_frag);
2722 } SCAN_END(OVS_KEY_ATTR_IPV4);
2723
2724 SCAN_BEGIN("ipv6(", struct ovs_key_ipv6) {
2725 SCAN_FIELD("src=", ipv6, ipv6_src);
2726 SCAN_FIELD("dst=", ipv6, ipv6_dst);
2727 SCAN_FIELD("label=", ipv6_label, ipv6_label);
2728 SCAN_FIELD("proto=", u8, ipv6_proto);
2729 SCAN_FIELD("tclass=", u8, ipv6_tclass);
2730 SCAN_FIELD("hlimit=", u8, ipv6_hlimit);
2731 SCAN_FIELD("frag=", frag, ipv6_frag);
2732 } SCAN_END(OVS_KEY_ATTR_IPV6);
2733
2734 SCAN_BEGIN("tcp(", struct ovs_key_tcp) {
2735 SCAN_FIELD("src=", be16, tcp_src);
2736 SCAN_FIELD("dst=", be16, tcp_dst);
2737 } SCAN_END(OVS_KEY_ATTR_TCP);
2738
2739 SCAN_SINGLE("tcp_flags(", ovs_be16, tcp_flags, OVS_KEY_ATTR_TCP_FLAGS);
2740
2741 SCAN_BEGIN("udp(", struct ovs_key_udp) {
2742 SCAN_FIELD("src=", be16, udp_src);
2743 SCAN_FIELD("dst=", be16, udp_dst);
2744 } SCAN_END(OVS_KEY_ATTR_UDP);
2745
2746 SCAN_BEGIN("sctp(", struct ovs_key_sctp) {
2747 SCAN_FIELD("src=", be16, sctp_src);
2748 SCAN_FIELD("dst=", be16, sctp_dst);
2749 } SCAN_END(OVS_KEY_ATTR_SCTP);
2750
2751 SCAN_BEGIN("icmp(", struct ovs_key_icmp) {
2752 SCAN_FIELD("type=", u8, icmp_type);
2753 SCAN_FIELD("code=", u8, icmp_code);
2754 } SCAN_END(OVS_KEY_ATTR_ICMP);
2755
2756 SCAN_BEGIN("icmpv6(", struct ovs_key_icmpv6) {
2757 SCAN_FIELD("type=", u8, icmpv6_type);
2758 SCAN_FIELD("code=", u8, icmpv6_code);
2759 } SCAN_END(OVS_KEY_ATTR_ICMPV6);
2760
2761 SCAN_BEGIN("arp(", struct ovs_key_arp) {
2762 SCAN_FIELD("sip=", ipv4, arp_sip);
2763 SCAN_FIELD("tip=", ipv4, arp_tip);
2764 SCAN_FIELD("op=", be16, arp_op);
2765 SCAN_FIELD("sha=", eth, arp_sha);
2766 SCAN_FIELD("tha=", eth, arp_tha);
2767 } SCAN_END(OVS_KEY_ATTR_ARP);
2768
2769 SCAN_BEGIN("nd(", struct ovs_key_nd) {
2770 SCAN_FIELD("target=", ipv6, nd_target);
2771 SCAN_FIELD("sll=", eth, nd_sll);
2772 SCAN_FIELD("tll=", eth, nd_tll);
2773 } SCAN_END(OVS_KEY_ATTR_ND);
2774
2775 /* Encap open-coded. */
2776 if (!strncmp(s, "encap(", 6)) {
2777 const char *start = s;
2778 size_t encap, encap_mask = 0;
2779
2780 encap = nl_msg_start_nested(key, OVS_KEY_ATTR_ENCAP);
2781 if (mask) {
2782 encap_mask = nl_msg_start_nested(mask, OVS_KEY_ATTR_ENCAP);
2783 }
2784
2785 s += 6;
2786 for (;;) {
2787 int retval;
2788
2789 s += strspn(s, delimiters);
2790 if (!*s) {
2791 return -EINVAL;
2792 } else if (*s == ')') {
2793 break;
2794 }
2795
2796 retval = parse_odp_key_mask_attr(s, port_names, key, mask);
2797 if (retval < 0) {
2798 return retval;
2799 }
2800 s += retval;
2801 }
2802 s++;
2803
2804 nl_msg_end_nested(key, encap);
2805 if (mask) {
2806 nl_msg_end_nested(mask, encap_mask);
2807 }
2808
2809 return s - start;
2810 }
2811
2812 return -EINVAL;
2813 }
2814
2815 /* Parses the string representation of a datapath flow key, in the
2816 * format output by odp_flow_key_format(). Returns 0 if successful,
2817 * otherwise a positive errno value. On success, the flow key is
2818 * appended to 'key' as a series of Netlink attributes. On failure, no
2819 * data is appended to 'key'. Either way, 'key''s data might be
2820 * reallocated.
2821 *
2822 * If 'port_names' is nonnull, it points to an simap that maps from a port name
2823 * to a port number. (Port names may be used instead of port numbers in
2824 * in_port.)
2825 *
2826 * On success, the attributes appended to 'key' are individually syntactically
2827 * valid, but they may not be valid as a sequence. 'key' might, for example,
2828 * have duplicated keys. odp_flow_key_to_flow() will detect those errors. */
2829 int
2830 odp_flow_from_string(const char *s, const struct simap *port_names,
2831 struct ofpbuf *key, struct ofpbuf *mask)
2832 {
2833 const size_t old_size = ofpbuf_size(key);
2834 for (;;) {
2835 int retval;
2836
2837 s += strspn(s, delimiters);
2838 if (!*s) {
2839 return 0;
2840 }
2841
2842 retval = parse_odp_key_mask_attr(s, port_names, key, mask);
2843 if (retval < 0) {
2844 ofpbuf_set_size(key, old_size);
2845 return -retval;
2846 }
2847 s += retval;
2848 }
2849
2850 return 0;
2851 }
2852
2853 static uint8_t
2854 ovs_to_odp_frag(uint8_t nw_frag, bool is_mask)
2855 {
2856 if (is_mask) {
2857 /* Netlink interface 'enum ovs_frag_type' is an 8-bit enumeration type,
2858 * not a set of flags or bitfields. Hence, if the struct flow nw_frag
2859 * mask, which is a set of bits, has the FLOW_NW_FRAG_ANY as zero, we
2860 * must use a zero mask for the netlink frag field, and all ones mask
2861 * otherwise. */
2862 return (nw_frag & FLOW_NW_FRAG_ANY) ? UINT8_MAX : 0;
2863 }
2864 return !(nw_frag & FLOW_NW_FRAG_ANY) ? OVS_FRAG_TYPE_NONE
2865 : nw_frag & FLOW_NW_FRAG_LATER ? OVS_FRAG_TYPE_LATER
2866 : OVS_FRAG_TYPE_FIRST;
2867 }
2868
2869 static void get_ethernet_key(const struct flow *, struct ovs_key_ethernet *);
2870 static void put_ethernet_key(const struct ovs_key_ethernet *, struct flow *);
2871 static void get_ipv4_key(const struct flow *, struct ovs_key_ipv4 *,
2872 bool is_mask);
2873 static void put_ipv4_key(const struct ovs_key_ipv4 *, struct flow *,
2874 bool is_mask);
2875 static void get_ipv6_key(const struct flow *, struct ovs_key_ipv6 *,
2876 bool is_mask);
2877 static void put_ipv6_key(const struct ovs_key_ipv6 *, struct flow *,
2878 bool is_mask);
2879 static void get_arp_key(const struct flow *, struct ovs_key_arp *);
2880 static void put_arp_key(const struct ovs_key_arp *, struct flow *);
2881 static void get_nd_key(const struct flow *, struct ovs_key_nd *);
2882 static void put_nd_key(const struct ovs_key_nd *, struct flow *);
2883
2884 /* These share the same layout. */
2885 union ovs_key_tp {
2886 struct ovs_key_tcp tcp;
2887 struct ovs_key_udp udp;
2888 struct ovs_key_sctp sctp;
2889 };
2890
2891 static void get_tp_key(const struct flow *, union ovs_key_tp *);
2892 static void put_tp_key(const union ovs_key_tp *, struct flow *);
2893
2894 static void
2895 odp_flow_key_from_flow__(struct ofpbuf *buf, const struct flow *flow,
2896 const struct flow *mask, odp_port_t odp_in_port,
2897 size_t max_mpls_depth, bool recirc, bool export_mask)
2898 {
2899 struct ovs_key_ethernet *eth_key;
2900 size_t encap;
2901 const struct flow *data = export_mask ? mask : flow;
2902
2903 nl_msg_put_u32(buf, OVS_KEY_ATTR_PRIORITY, data->skb_priority);
2904
2905 if (flow->tunnel.ip_dst || export_mask) {
2906 tun_key_to_attr(buf, &data->tunnel);
2907 }
2908
2909 nl_msg_put_u32(buf, OVS_KEY_ATTR_SKB_MARK, data->pkt_mark);
2910
2911 if (recirc) {
2912 nl_msg_put_u32(buf, OVS_KEY_ATTR_RECIRC_ID, data->recirc_id);
2913 nl_msg_put_u32(buf, OVS_KEY_ATTR_DP_HASH, data->dp_hash);
2914 }
2915
2916 /* Add an ingress port attribute if this is a mask or 'odp_in_port'
2917 * is not the magical value "ODPP_NONE". */
2918 if (export_mask || odp_in_port != ODPP_NONE) {
2919 nl_msg_put_odp_port(buf, OVS_KEY_ATTR_IN_PORT, odp_in_port);
2920 }
2921
2922 eth_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ETHERNET,
2923 sizeof *eth_key);
2924 get_ethernet_key(data, eth_key);
2925
2926 if (flow->vlan_tci != htons(0) || flow->dl_type == htons(ETH_TYPE_VLAN)) {
2927 if (export_mask) {
2928 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, OVS_BE16_MAX);
2929 } else {
2930 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_TYPE_VLAN));
2931 }
2932 nl_msg_put_be16(buf, OVS_KEY_ATTR_VLAN, data->vlan_tci);
2933 encap = nl_msg_start_nested(buf, OVS_KEY_ATTR_ENCAP);
2934 if (flow->vlan_tci == htons(0)) {
2935 goto unencap;
2936 }
2937 } else {
2938 encap = 0;
2939 }
2940
2941 if (ntohs(flow->dl_type) < ETH_TYPE_MIN) {
2942 /* For backwards compatibility with kernels that don't support
2943 * wildcarding, the following convention is used to encode the
2944 * OVS_KEY_ATTR_ETHERTYPE for key and mask:
2945 *
2946 * key mask matches
2947 * -------- -------- -------
2948 * >0x5ff 0xffff Specified Ethernet II Ethertype.
2949 * >0x5ff 0 Any Ethernet II or non-Ethernet II frame.
2950 * <none> 0xffff Any non-Ethernet II frame (except valid
2951 * 802.3 SNAP packet with valid eth_type).
2952 */
2953 if (export_mask) {
2954 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, OVS_BE16_MAX);
2955 }
2956 goto unencap;
2957 }
2958
2959 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, data->dl_type);
2960
2961 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2962 struct ovs_key_ipv4 *ipv4_key;
2963
2964 ipv4_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_IPV4,
2965 sizeof *ipv4_key);
2966 get_ipv4_key(data, ipv4_key, export_mask);
2967 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2968 struct ovs_key_ipv6 *ipv6_key;
2969
2970 ipv6_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_IPV6,
2971 sizeof *ipv6_key);
2972 get_ipv6_key(data, ipv6_key, export_mask);
2973 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
2974 flow->dl_type == htons(ETH_TYPE_RARP)) {
2975 struct ovs_key_arp *arp_key;
2976
2977 arp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ARP,
2978 sizeof *arp_key);
2979 get_arp_key(data, arp_key);
2980 } else if (eth_type_mpls(flow->dl_type)) {
2981 struct ovs_key_mpls *mpls_key;
2982 int i, n;
2983
2984 n = flow_count_mpls_labels(flow, NULL);
2985 n = MIN(n, max_mpls_depth);
2986 mpls_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_MPLS,
2987 n * sizeof *mpls_key);
2988 for (i = 0; i < n; i++) {
2989 mpls_key[i].mpls_lse = data->mpls_lse[i];
2990 }
2991 }
2992
2993 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
2994 if (flow->nw_proto == IPPROTO_TCP) {
2995 union ovs_key_tp *tcp_key;
2996
2997 tcp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_TCP,
2998 sizeof *tcp_key);
2999 get_tp_key(data, tcp_key);
3000 if (data->tcp_flags) {
3001 nl_msg_put_be16(buf, OVS_KEY_ATTR_TCP_FLAGS, data->tcp_flags);
3002 }
3003 } else if (flow->nw_proto == IPPROTO_UDP) {
3004 union ovs_key_tp *udp_key;
3005
3006 udp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_UDP,
3007 sizeof *udp_key);
3008 get_tp_key(data, udp_key);
3009 } else if (flow->nw_proto == IPPROTO_SCTP) {
3010 union ovs_key_tp *sctp_key;
3011
3012 sctp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_SCTP,
3013 sizeof *sctp_key);
3014 get_tp_key(data, sctp_key);
3015 } else if (flow->dl_type == htons(ETH_TYPE_IP)
3016 && flow->nw_proto == IPPROTO_ICMP) {
3017 struct ovs_key_icmp *icmp_key;
3018
3019 icmp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ICMP,
3020 sizeof *icmp_key);
3021 icmp_key->icmp_type = ntohs(data->tp_src);
3022 icmp_key->icmp_code = ntohs(data->tp_dst);
3023 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)
3024 && flow->nw_proto == IPPROTO_ICMPV6) {
3025 struct ovs_key_icmpv6 *icmpv6_key;
3026
3027 icmpv6_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ICMPV6,
3028 sizeof *icmpv6_key);
3029 icmpv6_key->icmpv6_type = ntohs(data->tp_src);
3030 icmpv6_key->icmpv6_code = ntohs(data->tp_dst);
3031
3032 if (flow->tp_dst == htons(0)
3033 && (flow->tp_src == htons(ND_NEIGHBOR_SOLICIT)
3034 || flow->tp_src == htons(ND_NEIGHBOR_ADVERT))
3035 && (!export_mask || (data->tp_src == htons(0xffff)
3036 && data->tp_dst == htons(0xffff)))) {
3037
3038 struct ovs_key_nd *nd_key;
3039
3040 nd_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ND,
3041 sizeof *nd_key);
3042 memcpy(nd_key->nd_target, &data->nd_target,
3043 sizeof nd_key->nd_target);
3044 memcpy(nd_key->nd_sll, data->arp_sha, ETH_ADDR_LEN);
3045 memcpy(nd_key->nd_tll, data->arp_tha, ETH_ADDR_LEN);
3046 }
3047 }
3048 }
3049
3050 unencap:
3051 if (encap) {
3052 nl_msg_end_nested(buf, encap);
3053 }
3054 }
3055
3056 /* Appends a representation of 'flow' as OVS_KEY_ATTR_* attributes to 'buf'.
3057 * 'flow->in_port' is ignored (since it is likely to be an OpenFlow port
3058 * number rather than a datapath port number). Instead, if 'odp_in_port'
3059 * is anything other than ODPP_NONE, it is included in 'buf' as the input
3060 * port.
3061 *
3062 * 'buf' must have at least ODPUTIL_FLOW_KEY_BYTES bytes of space, or be
3063 * capable of being expanded to allow for that much space.
3064 *
3065 * 'recirc' indicates support for recirculation fields. If this is true, then
3066 * these fields will always be serialised. */
3067 void
3068 odp_flow_key_from_flow(struct ofpbuf *buf, const struct flow *flow,
3069 const struct flow *mask, odp_port_t odp_in_port,
3070 bool recirc)
3071 {
3072 odp_flow_key_from_flow__(buf, flow, mask, odp_in_port, SIZE_MAX, recirc,
3073 false);
3074 }
3075
3076 /* Appends a representation of 'mask' as OVS_KEY_ATTR_* attributes to
3077 * 'buf'. 'flow' is used as a template to determine how to interpret
3078 * 'mask'. For example, the 'dl_type' of 'mask' describes the mask, but
3079 * it doesn't indicate whether the other fields should be interpreted as
3080 * ARP, IPv4, IPv6, etc.
3081 *
3082 * 'buf' must have at least ODPUTIL_FLOW_KEY_BYTES bytes of space, or be
3083 * capable of being expanded to allow for that much space.
3084 *
3085 * 'recirc' indicates support for recirculation fields. If this is true, then
3086 * these fields will always be serialised. */
3087 void
3088 odp_flow_key_from_mask(struct ofpbuf *buf, const struct flow *mask,
3089 const struct flow *flow, uint32_t odp_in_port_mask,
3090 size_t max_mpls_depth, bool recirc)
3091 {
3092 odp_flow_key_from_flow__(buf, flow, mask, u32_to_odp(odp_in_port_mask),
3093 max_mpls_depth, recirc, true);
3094 }
3095
3096 /* Generate ODP flow key from the given packet metadata */
3097 void
3098 odp_key_from_pkt_metadata(struct ofpbuf *buf, const struct pkt_metadata *md)
3099 {
3100 nl_msg_put_u32(buf, OVS_KEY_ATTR_PRIORITY, md->skb_priority);
3101
3102 if (md->tunnel.ip_dst) {
3103 tun_key_to_attr(buf, &md->tunnel);
3104 }
3105
3106 nl_msg_put_u32(buf, OVS_KEY_ATTR_SKB_MARK, md->pkt_mark);
3107
3108 /* Add an ingress port attribute if 'odp_in_port' is not the magical
3109 * value "ODPP_NONE". */
3110 if (md->in_port.odp_port != ODPP_NONE) {
3111 nl_msg_put_odp_port(buf, OVS_KEY_ATTR_IN_PORT, md->in_port.odp_port);
3112 }
3113 }
3114
3115 /* Generate packet metadata from the given ODP flow key. */
3116 void
3117 odp_key_to_pkt_metadata(const struct nlattr *key, size_t key_len,
3118 struct pkt_metadata *md)
3119 {
3120 const struct nlattr *nla;
3121 size_t left;
3122 uint32_t wanted_attrs = 1u << OVS_KEY_ATTR_PRIORITY |
3123 1u << OVS_KEY_ATTR_SKB_MARK | 1u << OVS_KEY_ATTR_TUNNEL |
3124 1u << OVS_KEY_ATTR_IN_PORT;
3125
3126 *md = PKT_METADATA_INITIALIZER(ODPP_NONE);
3127
3128 NL_ATTR_FOR_EACH (nla, left, key, key_len) {
3129 uint16_t type = nl_attr_type(nla);
3130 size_t len = nl_attr_get_size(nla);
3131 int expected_len = odp_flow_key_attr_len(type);
3132
3133 if (len != expected_len && expected_len >= 0) {
3134 continue;
3135 }
3136
3137 switch (type) {
3138 case OVS_KEY_ATTR_RECIRC_ID:
3139 md->recirc_id = nl_attr_get_u32(nla);
3140 wanted_attrs &= ~(1u << OVS_KEY_ATTR_RECIRC_ID);
3141 break;
3142 case OVS_KEY_ATTR_DP_HASH:
3143 md->dp_hash = nl_attr_get_u32(nla);
3144 wanted_attrs &= ~(1u << OVS_KEY_ATTR_DP_HASH);
3145 break;
3146 case OVS_KEY_ATTR_PRIORITY:
3147 md->skb_priority = nl_attr_get_u32(nla);
3148 wanted_attrs &= ~(1u << OVS_KEY_ATTR_PRIORITY);
3149 break;
3150 case OVS_KEY_ATTR_SKB_MARK:
3151 md->pkt_mark = nl_attr_get_u32(nla);
3152 wanted_attrs &= ~(1u << OVS_KEY_ATTR_SKB_MARK);
3153 break;
3154 case OVS_KEY_ATTR_TUNNEL: {
3155 enum odp_key_fitness res;
3156
3157 res = odp_tun_key_from_attr(nla, &md->tunnel);
3158 if (res == ODP_FIT_ERROR) {
3159 memset(&md->tunnel, 0, sizeof md->tunnel);
3160 } else if (res == ODP_FIT_PERFECT) {
3161 wanted_attrs &= ~(1u << OVS_KEY_ATTR_TUNNEL);
3162 }
3163 break;
3164 }
3165 case OVS_KEY_ATTR_IN_PORT:
3166 md->in_port.odp_port = nl_attr_get_odp_port(nla);
3167 wanted_attrs &= ~(1u << OVS_KEY_ATTR_IN_PORT);
3168 break;
3169 default:
3170 break;
3171 }
3172
3173 if (!wanted_attrs) {
3174 return; /* Have everything. */
3175 }
3176 }
3177 }
3178
3179 uint32_t
3180 odp_flow_key_hash(const struct nlattr *key, size_t key_len)
3181 {
3182 BUILD_ASSERT_DECL(!(NLA_ALIGNTO % sizeof(uint32_t)));
3183 return hash_words(ALIGNED_CAST(const uint32_t *, key),
3184 key_len / sizeof(uint32_t), 0);
3185 }
3186
3187 static void
3188 log_odp_key_attributes(struct vlog_rate_limit *rl, const char *title,
3189 uint64_t attrs, int out_of_range_attr,
3190 const struct nlattr *key, size_t key_len)
3191 {
3192 struct ds s;
3193 int i;
3194
3195 if (VLOG_DROP_DBG(rl)) {
3196 return;
3197 }
3198
3199 ds_init(&s);
3200 for (i = 0; i < 64; i++) {
3201 if (attrs & (UINT64_C(1) << i)) {
3202 char namebuf[OVS_KEY_ATTR_BUFSIZE];
3203
3204 ds_put_format(&s, " %s",
3205 ovs_key_attr_to_string(i, namebuf, sizeof namebuf));
3206 }
3207 }
3208 if (out_of_range_attr) {
3209 ds_put_format(&s, " %d (and possibly others)", out_of_range_attr);
3210 }
3211
3212 ds_put_cstr(&s, ": ");
3213 odp_flow_key_format(key, key_len, &s);
3214
3215 VLOG_DBG("%s:%s", title, ds_cstr(&s));
3216 ds_destroy(&s);
3217 }
3218
3219 static uint8_t
3220 odp_to_ovs_frag(uint8_t odp_frag, bool is_mask)
3221 {
3222 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3223
3224 if (is_mask) {
3225 return odp_frag ? FLOW_NW_FRAG_MASK : 0;
3226 }
3227
3228 if (odp_frag > OVS_FRAG_TYPE_LATER) {
3229 VLOG_ERR_RL(&rl, "invalid frag %"PRIu8" in flow key", odp_frag);
3230 return 0xff; /* Error. */
3231 }
3232
3233 return (odp_frag == OVS_FRAG_TYPE_NONE) ? 0
3234 : (odp_frag == OVS_FRAG_TYPE_FIRST) ? FLOW_NW_FRAG_ANY
3235 : FLOW_NW_FRAG_ANY | FLOW_NW_FRAG_LATER;
3236 }
3237
3238 static bool
3239 parse_flow_nlattrs(const struct nlattr *key, size_t key_len,
3240 const struct nlattr *attrs[], uint64_t *present_attrsp,
3241 int *out_of_range_attrp)
3242 {
3243 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3244 const struct nlattr *nla;
3245 uint64_t present_attrs;
3246 size_t left;
3247
3248 BUILD_ASSERT(OVS_KEY_ATTR_MAX < CHAR_BIT * sizeof present_attrs);
3249 present_attrs = 0;
3250 *out_of_range_attrp = 0;
3251 NL_ATTR_FOR_EACH (nla, left, key, key_len) {
3252 uint16_t type = nl_attr_type(nla);
3253 size_t len = nl_attr_get_size(nla);
3254 int expected_len = odp_flow_key_attr_len(type);
3255
3256 if (len != expected_len && expected_len >= 0) {
3257 char namebuf[OVS_KEY_ATTR_BUFSIZE];
3258
3259 VLOG_ERR_RL(&rl, "attribute %s has length %"PRIuSIZE" but should have "
3260 "length %d", ovs_key_attr_to_string(type, namebuf,
3261 sizeof namebuf),
3262 len, expected_len);
3263 return false;
3264 }
3265
3266 if (type > OVS_KEY_ATTR_MAX) {
3267 *out_of_range_attrp = type;
3268 } else {
3269 if (present_attrs & (UINT64_C(1) << type)) {
3270 char namebuf[OVS_KEY_ATTR_BUFSIZE];
3271
3272 VLOG_ERR_RL(&rl, "duplicate %s attribute in flow key",
3273 ovs_key_attr_to_string(type,
3274 namebuf, sizeof namebuf));
3275 return false;
3276 }
3277
3278 present_attrs |= UINT64_C(1) << type;
3279 attrs[type] = nla;
3280 }
3281 }
3282 if (left) {
3283 VLOG_ERR_RL(&rl, "trailing garbage in flow key");
3284 return false;
3285 }
3286
3287 *present_attrsp = present_attrs;
3288 return true;
3289 }
3290
3291 static enum odp_key_fitness
3292 check_expectations(uint64_t present_attrs, int out_of_range_attr,
3293 uint64_t expected_attrs,
3294 const struct nlattr *key, size_t key_len)
3295 {
3296 uint64_t missing_attrs;
3297 uint64_t extra_attrs;
3298
3299 missing_attrs = expected_attrs & ~present_attrs;
3300 if (missing_attrs) {
3301 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3302 log_odp_key_attributes(&rl, "expected but not present",
3303 missing_attrs, 0, key, key_len);
3304 return ODP_FIT_TOO_LITTLE;
3305 }
3306
3307 extra_attrs = present_attrs & ~expected_attrs;
3308 if (extra_attrs || out_of_range_attr) {
3309 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3310 log_odp_key_attributes(&rl, "present but not expected",
3311 extra_attrs, out_of_range_attr, key, key_len);
3312 return ODP_FIT_TOO_MUCH;
3313 }
3314
3315 return ODP_FIT_PERFECT;
3316 }
3317
3318 static bool
3319 parse_ethertype(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
3320 uint64_t present_attrs, uint64_t *expected_attrs,
3321 struct flow *flow, const struct flow *src_flow)
3322 {
3323 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3324 bool is_mask = flow != src_flow;
3325
3326 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE)) {
3327 flow->dl_type = nl_attr_get_be16(attrs[OVS_KEY_ATTR_ETHERTYPE]);
3328 if (!is_mask && ntohs(flow->dl_type) < ETH_TYPE_MIN) {
3329 VLOG_ERR_RL(&rl, "invalid Ethertype %"PRIu16" in flow key",
3330 ntohs(flow->dl_type));
3331 return false;
3332 }
3333 if (is_mask && ntohs(src_flow->dl_type) < ETH_TYPE_MIN &&
3334 flow->dl_type != htons(0xffff)) {
3335 return false;
3336 }
3337 *expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE;
3338 } else {
3339 if (!is_mask) {
3340 flow->dl_type = htons(FLOW_DL_TYPE_NONE);
3341 } else if (ntohs(src_flow->dl_type) < ETH_TYPE_MIN) {
3342 /* See comments in odp_flow_key_from_flow__(). */
3343 VLOG_ERR_RL(&rl, "mask expected for non-Ethernet II frame");
3344 return false;
3345 }
3346 }
3347 return true;
3348 }
3349
3350 static enum odp_key_fitness
3351 parse_l2_5_onward(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
3352 uint64_t present_attrs, int out_of_range_attr,
3353 uint64_t expected_attrs, struct flow *flow,
3354 const struct nlattr *key, size_t key_len,
3355 const struct flow *src_flow)
3356 {
3357 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3358 bool is_mask = src_flow != flow;
3359 const void *check_start = NULL;
3360 size_t check_len = 0;
3361 enum ovs_key_attr expected_bit = 0xff;
3362
3363 if (eth_type_mpls(src_flow->dl_type)) {
3364 if (!is_mask || present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_MPLS)) {
3365 expected_attrs |= (UINT64_C(1) << OVS_KEY_ATTR_MPLS);
3366 }
3367 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_MPLS)) {
3368 size_t size = nl_attr_get_size(attrs[OVS_KEY_ATTR_MPLS]);
3369 const ovs_be32 *mpls_lse = nl_attr_get(attrs[OVS_KEY_ATTR_MPLS]);
3370 int n = size / sizeof(ovs_be32);
3371 int i;
3372
3373 if (!size || size % sizeof(ovs_be32)) {
3374 return ODP_FIT_ERROR;
3375 }
3376 if (flow->mpls_lse[0] && flow->dl_type != htons(0xffff)) {
3377 return ODP_FIT_ERROR;
3378 }
3379
3380 for (i = 0; i < n && i < FLOW_MAX_MPLS_LABELS; i++) {
3381 flow->mpls_lse[i] = mpls_lse[i];
3382 }
3383 if (n > FLOW_MAX_MPLS_LABELS) {
3384 return ODP_FIT_TOO_MUCH;
3385 }
3386
3387 if (!is_mask) {
3388 /* BOS may be set only in the innermost label. */
3389 for (i = 0; i < n - 1; i++) {
3390 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
3391 return ODP_FIT_ERROR;
3392 }
3393 }
3394
3395 /* BOS must be set in the innermost label. */
3396 if (n < FLOW_MAX_MPLS_LABELS
3397 && !(flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK))) {
3398 return ODP_FIT_TOO_LITTLE;
3399 }
3400 }
3401 }
3402
3403 goto done;
3404 } else if (src_flow->dl_type == htons(ETH_TYPE_IP)) {
3405 if (!is_mask) {
3406 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IPV4;
3407 }
3408 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IPV4)) {
3409 const struct ovs_key_ipv4 *ipv4_key;
3410
3411 ipv4_key = nl_attr_get(attrs[OVS_KEY_ATTR_IPV4]);
3412 put_ipv4_key(ipv4_key, flow, is_mask);
3413 if (flow->nw_frag > FLOW_NW_FRAG_MASK) {
3414 return ODP_FIT_ERROR;
3415 }
3416 if (is_mask) {
3417 check_start = ipv4_key;
3418 check_len = sizeof *ipv4_key;
3419 expected_bit = OVS_KEY_ATTR_IPV4;
3420 }
3421 }
3422 } else if (src_flow->dl_type == htons(ETH_TYPE_IPV6)) {
3423 if (!is_mask) {
3424 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IPV6;
3425 }
3426 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IPV6)) {
3427 const struct ovs_key_ipv6 *ipv6_key;
3428
3429 ipv6_key = nl_attr_get(attrs[OVS_KEY_ATTR_IPV6]);
3430 put_ipv6_key(ipv6_key, flow, is_mask);
3431 if (flow->nw_frag > FLOW_NW_FRAG_MASK) {
3432 return ODP_FIT_ERROR;
3433 }
3434 if (is_mask) {
3435 check_start = ipv6_key;
3436 check_len = sizeof *ipv6_key;
3437 expected_bit = OVS_KEY_ATTR_IPV6;
3438 }
3439 }
3440 } else if (src_flow->dl_type == htons(ETH_TYPE_ARP) ||
3441 src_flow->dl_type == htons(ETH_TYPE_RARP)) {
3442 if (!is_mask) {
3443 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ARP;
3444 }
3445 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ARP)) {
3446 const struct ovs_key_arp *arp_key;
3447
3448 arp_key = nl_attr_get(attrs[OVS_KEY_ATTR_ARP]);
3449 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
3450 VLOG_ERR_RL(&rl, "unsupported ARP opcode %"PRIu16" in flow "
3451 "key", ntohs(arp_key->arp_op));
3452 return ODP_FIT_ERROR;
3453 }
3454 put_arp_key(arp_key, flow);
3455 if (is_mask) {
3456 check_start = arp_key;
3457 check_len = sizeof *arp_key;
3458 expected_bit = OVS_KEY_ATTR_ARP;
3459 }
3460 }
3461 } else {
3462 goto done;
3463 }
3464 if (check_len > 0) { /* Happens only when 'is_mask'. */
3465 if (!is_all_zeros(check_start, check_len) &&
3466 flow->dl_type != htons(0xffff)) {
3467 return ODP_FIT_ERROR;
3468 } else {
3469 expected_attrs |= UINT64_C(1) << expected_bit;
3470 }
3471 }
3472
3473 expected_bit = OVS_KEY_ATTR_UNSPEC;
3474 if (src_flow->nw_proto == IPPROTO_TCP
3475 && (src_flow->dl_type == htons(ETH_TYPE_IP) ||
3476 src_flow->dl_type == htons(ETH_TYPE_IPV6))
3477 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3478 if (!is_mask) {
3479 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_TCP;
3480 }
3481 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_TCP)) {
3482 const union ovs_key_tp *tcp_key;
3483
3484 tcp_key = nl_attr_get(attrs[OVS_KEY_ATTR_TCP]);
3485 put_tp_key(tcp_key, flow);
3486 expected_bit = OVS_KEY_ATTR_TCP;
3487 }
3488 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_TCP_FLAGS)) {
3489 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_TCP_FLAGS;
3490 flow->tcp_flags = nl_attr_get_be16(attrs[OVS_KEY_ATTR_TCP_FLAGS]);
3491 }
3492 } else if (src_flow->nw_proto == IPPROTO_UDP
3493 && (src_flow->dl_type == htons(ETH_TYPE_IP) ||
3494 src_flow->dl_type == htons(ETH_TYPE_IPV6))
3495 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3496 if (!is_mask) {
3497 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_UDP;
3498 }
3499 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_UDP)) {
3500 const union ovs_key_tp *udp_key;
3501
3502 udp_key = nl_attr_get(attrs[OVS_KEY_ATTR_UDP]);
3503 put_tp_key(udp_key, flow);
3504 expected_bit = OVS_KEY_ATTR_UDP;
3505 }
3506 } else if (src_flow->nw_proto == IPPROTO_SCTP
3507 && (src_flow->dl_type == htons(ETH_TYPE_IP) ||
3508 src_flow->dl_type == htons(ETH_TYPE_IPV6))
3509 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3510 if (!is_mask) {
3511 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_SCTP;
3512 }
3513 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_SCTP)) {
3514 const union ovs_key_tp *sctp_key;
3515
3516 sctp_key = nl_attr_get(attrs[OVS_KEY_ATTR_SCTP]);
3517 put_tp_key(sctp_key, flow);
3518 expected_bit = OVS_KEY_ATTR_SCTP;
3519 }
3520 } else if (src_flow->nw_proto == IPPROTO_ICMP
3521 && src_flow->dl_type == htons(ETH_TYPE_IP)
3522 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3523 if (!is_mask) {
3524 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ICMP;
3525 }
3526 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ICMP)) {
3527 const struct ovs_key_icmp *icmp_key;
3528
3529 icmp_key = nl_attr_get(attrs[OVS_KEY_ATTR_ICMP]);
3530 flow->tp_src = htons(icmp_key->icmp_type);
3531 flow->tp_dst = htons(icmp_key->icmp_code);
3532 expected_bit = OVS_KEY_ATTR_ICMP;
3533 }
3534 } else if (src_flow->nw_proto == IPPROTO_ICMPV6
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_ICMPV6;
3539 }
3540 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ICMPV6)) {
3541 const struct ovs_key_icmpv6 *icmpv6_key;
3542
3543 icmpv6_key = nl_attr_get(attrs[OVS_KEY_ATTR_ICMPV6]);
3544 flow->tp_src = htons(icmpv6_key->icmpv6_type);
3545 flow->tp_dst = htons(icmpv6_key->icmpv6_code);
3546 expected_bit = OVS_KEY_ATTR_ICMPV6;
3547 if (src_flow->tp_dst == htons(0) &&
3548 (src_flow->tp_src == htons(ND_NEIGHBOR_SOLICIT) ||
3549 src_flow->tp_src == htons(ND_NEIGHBOR_ADVERT))) {
3550 if (!is_mask) {
3551 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ND;
3552 }
3553 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ND)) {
3554 const struct ovs_key_nd *nd_key;
3555
3556 nd_key = nl_attr_get(attrs[OVS_KEY_ATTR_ND]);
3557 memcpy(&flow->nd_target, nd_key->nd_target,
3558 sizeof flow->nd_target);
3559 memcpy(flow->arp_sha, nd_key->nd_sll, ETH_ADDR_LEN);
3560 memcpy(flow->arp_tha, nd_key->nd_tll, ETH_ADDR_LEN);
3561 if (is_mask) {
3562 if (!is_all_zeros(nd_key, sizeof *nd_key) &&
3563 (flow->tp_src != htons(0xffff) ||
3564 flow->tp_dst != htons(0xffff))) {
3565 return ODP_FIT_ERROR;
3566 } else {
3567 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ND;
3568 }
3569 }
3570 }
3571 }
3572 }
3573 }
3574 if (is_mask && expected_bit != OVS_KEY_ATTR_UNSPEC) {
3575 if ((flow->tp_src || flow->tp_dst) && flow->nw_proto != 0xff) {
3576 return ODP_FIT_ERROR;
3577 } else {
3578 expected_attrs |= UINT64_C(1) << expected_bit;
3579 }
3580 }
3581
3582 done:
3583 return check_expectations(present_attrs, out_of_range_attr, expected_attrs,
3584 key, key_len);
3585 }
3586
3587 /* Parse 802.1Q header then encapsulated L3 attributes. */
3588 static enum odp_key_fitness
3589 parse_8021q_onward(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
3590 uint64_t present_attrs, int out_of_range_attr,
3591 uint64_t expected_attrs, struct flow *flow,
3592 const struct nlattr *key, size_t key_len,
3593 const struct flow *src_flow)
3594 {
3595 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3596 bool is_mask = src_flow != flow;
3597
3598 const struct nlattr *encap
3599 = (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ENCAP)
3600 ? attrs[OVS_KEY_ATTR_ENCAP] : NULL);
3601 enum odp_key_fitness encap_fitness;
3602 enum odp_key_fitness fitness;
3603
3604 /* Calculate fitness of outer attributes. */
3605 if (!is_mask) {
3606 expected_attrs |= ((UINT64_C(1) << OVS_KEY_ATTR_VLAN) |
3607 (UINT64_C(1) << OVS_KEY_ATTR_ENCAP));
3608 } else {
3609 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN)) {
3610 expected_attrs |= (UINT64_C(1) << OVS_KEY_ATTR_VLAN);
3611 }
3612 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ENCAP)) {
3613 expected_attrs |= (UINT64_C(1) << OVS_KEY_ATTR_ENCAP);
3614 }
3615 }
3616 fitness = check_expectations(present_attrs, out_of_range_attr,
3617 expected_attrs, key, key_len);
3618
3619 /* Set vlan_tci.
3620 * Remove the TPID from dl_type since it's not the real Ethertype. */
3621 flow->dl_type = htons(0);
3622 flow->vlan_tci = (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN)
3623 ? nl_attr_get_be16(attrs[OVS_KEY_ATTR_VLAN])
3624 : htons(0));
3625 if (!is_mask) {
3626 if (!(present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN))) {
3627 return ODP_FIT_TOO_LITTLE;
3628 } else if (flow->vlan_tci == htons(0)) {
3629 /* Corner case for a truncated 802.1Q header. */
3630 if (fitness == ODP_FIT_PERFECT && nl_attr_get_size(encap)) {
3631 return ODP_FIT_TOO_MUCH;
3632 }
3633 return fitness;
3634 } else if (!(flow->vlan_tci & htons(VLAN_CFI))) {
3635 VLOG_ERR_RL(&rl, "OVS_KEY_ATTR_VLAN 0x%04"PRIx16" is nonzero "
3636 "but CFI bit is not set", ntohs(flow->vlan_tci));
3637 return ODP_FIT_ERROR;
3638 }
3639 } else {
3640 if (!(present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ENCAP))) {
3641 return fitness;
3642 }
3643 }
3644
3645 /* Now parse the encapsulated attributes. */
3646 if (!parse_flow_nlattrs(nl_attr_get(encap), nl_attr_get_size(encap),
3647 attrs, &present_attrs, &out_of_range_attr)) {
3648 return ODP_FIT_ERROR;
3649 }
3650 expected_attrs = 0;
3651
3652 if (!parse_ethertype(attrs, present_attrs, &expected_attrs, flow, src_flow)) {
3653 return ODP_FIT_ERROR;
3654 }
3655 encap_fitness = parse_l2_5_onward(attrs, present_attrs, out_of_range_attr,
3656 expected_attrs, flow, key, key_len,
3657 src_flow);
3658
3659 /* The overall fitness is the worse of the outer and inner attributes. */
3660 return MAX(fitness, encap_fitness);
3661 }
3662
3663 static enum odp_key_fitness
3664 odp_flow_key_to_flow__(const struct nlattr *key, size_t key_len,
3665 struct flow *flow, const struct flow *src_flow)
3666 {
3667 const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1];
3668 uint64_t expected_attrs;
3669 uint64_t present_attrs;
3670 int out_of_range_attr;
3671 bool is_mask = src_flow != flow;
3672
3673 memset(flow, 0, sizeof *flow);
3674
3675 /* Parse attributes. */
3676 if (!parse_flow_nlattrs(key, key_len, attrs, &present_attrs,
3677 &out_of_range_attr)) {
3678 return ODP_FIT_ERROR;
3679 }
3680 expected_attrs = 0;
3681
3682 /* Metadata. */
3683 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_RECIRC_ID)) {
3684 flow->recirc_id = nl_attr_get_u32(attrs[OVS_KEY_ATTR_RECIRC_ID]);
3685 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_RECIRC_ID;
3686 } else if (is_mask) {
3687 /* Always exact match recirc_id if it is not specified. */
3688 flow->recirc_id = UINT32_MAX;
3689 }
3690
3691 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_DP_HASH)) {
3692 flow->dp_hash = nl_attr_get_u32(attrs[OVS_KEY_ATTR_DP_HASH]);
3693 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_DP_HASH;
3694 }
3695 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_PRIORITY)) {
3696 flow->skb_priority = nl_attr_get_u32(attrs[OVS_KEY_ATTR_PRIORITY]);
3697 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_PRIORITY;
3698 }
3699
3700 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_SKB_MARK)) {
3701 flow->pkt_mark = nl_attr_get_u32(attrs[OVS_KEY_ATTR_SKB_MARK]);
3702 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_SKB_MARK;
3703 }
3704
3705 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_TUNNEL)) {
3706 enum odp_key_fitness res;
3707
3708 res = odp_tun_key_from_attr(attrs[OVS_KEY_ATTR_TUNNEL], &flow->tunnel);
3709 if (res == ODP_FIT_ERROR) {
3710 return ODP_FIT_ERROR;
3711 } else if (res == ODP_FIT_PERFECT) {
3712 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_TUNNEL;
3713 }
3714 }
3715
3716 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IN_PORT)) {
3717 flow->in_port.odp_port
3718 = nl_attr_get_odp_port(attrs[OVS_KEY_ATTR_IN_PORT]);
3719 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IN_PORT;
3720 } else if (!is_mask) {
3721 flow->in_port.odp_port = ODPP_NONE;
3722 }
3723
3724 /* Ethernet header. */
3725 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ETHERNET)) {
3726 const struct ovs_key_ethernet *eth_key;
3727
3728 eth_key = nl_attr_get(attrs[OVS_KEY_ATTR_ETHERNET]);
3729 put_ethernet_key(eth_key, flow);
3730 if (is_mask) {
3731 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ETHERNET;
3732 }
3733 }
3734 if (!is_mask) {
3735 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ETHERNET;
3736 }
3737
3738 /* Get Ethertype or 802.1Q TPID or FLOW_DL_TYPE_NONE. */
3739 if (!parse_ethertype(attrs, present_attrs, &expected_attrs, flow,
3740 src_flow)) {
3741 return ODP_FIT_ERROR;
3742 }
3743
3744 if (is_mask
3745 ? (src_flow->vlan_tci & htons(VLAN_CFI)) != 0
3746 : src_flow->dl_type == htons(ETH_TYPE_VLAN)) {
3747 return parse_8021q_onward(attrs, present_attrs, out_of_range_attr,
3748 expected_attrs, flow, key, key_len, src_flow);
3749 }
3750 if (is_mask) {
3751 flow->vlan_tci = htons(0xffff);
3752 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN)) {
3753 flow->vlan_tci = nl_attr_get_be16(attrs[OVS_KEY_ATTR_VLAN]);
3754 expected_attrs |= (UINT64_C(1) << OVS_KEY_ATTR_VLAN);
3755 }
3756 }
3757 return parse_l2_5_onward(attrs, present_attrs, out_of_range_attr,
3758 expected_attrs, flow, key, key_len, src_flow);
3759 }
3760
3761 /* Converts the 'key_len' bytes of OVS_KEY_ATTR_* attributes in 'key' to a flow
3762 * structure in 'flow'. Returns an ODP_FIT_* value that indicates how well
3763 * 'key' fits our expectations for what a flow key should contain.
3764 *
3765 * The 'in_port' will be the datapath's understanding of the port. The
3766 * caller will need to translate with odp_port_to_ofp_port() if the
3767 * OpenFlow port is needed.
3768 *
3769 * This function doesn't take the packet itself as an argument because none of
3770 * the currently understood OVS_KEY_ATTR_* attributes require it. Currently,
3771 * it is always possible to infer which additional attribute(s) should appear
3772 * by looking at the attributes for lower-level protocols, e.g. if the network
3773 * protocol in OVS_KEY_ATTR_IPV4 or OVS_KEY_ATTR_IPV6 is IPPROTO_TCP then we
3774 * know that a OVS_KEY_ATTR_TCP attribute must appear and that otherwise it
3775 * must be absent. */
3776 enum odp_key_fitness
3777 odp_flow_key_to_flow(const struct nlattr *key, size_t key_len,
3778 struct flow *flow)
3779 {
3780 return odp_flow_key_to_flow__(key, key_len, flow, flow);
3781 }
3782
3783 /* Converts the 'key_len' bytes of OVS_KEY_ATTR_* attributes in 'key' to a mask
3784 * structure in 'mask'. 'flow' must be a previously translated flow
3785 * corresponding to 'mask'. Returns an ODP_FIT_* value that indicates how well
3786 * 'key' fits our expectations for what a flow key should contain. */
3787 enum odp_key_fitness
3788 odp_flow_key_to_mask(const struct nlattr *key, size_t key_len,
3789 struct flow *mask, const struct flow *flow)
3790 {
3791 return odp_flow_key_to_flow__(key, key_len, mask, flow);
3792 }
3793
3794 /* Returns 'fitness' as a string, for use in debug messages. */
3795 const char *
3796 odp_key_fitness_to_string(enum odp_key_fitness fitness)
3797 {
3798 switch (fitness) {
3799 case ODP_FIT_PERFECT:
3800 return "OK";
3801 case ODP_FIT_TOO_MUCH:
3802 return "too_much";
3803 case ODP_FIT_TOO_LITTLE:
3804 return "too_little";
3805 case ODP_FIT_ERROR:
3806 return "error";
3807 default:
3808 return "<unknown>";
3809 }
3810 }
3811
3812 /* Appends an OVS_ACTION_ATTR_USERSPACE action to 'odp_actions' that specifies
3813 * Netlink PID 'pid'. If 'userdata' is nonnull, adds a userdata attribute
3814 * whose contents are the 'userdata_size' bytes at 'userdata' and returns the
3815 * offset within 'odp_actions' of the start of the cookie. (If 'userdata' is
3816 * null, then the return value is not meaningful.) */
3817 size_t
3818 odp_put_userspace_action(uint32_t pid,
3819 const void *userdata, size_t userdata_size,
3820 odp_port_t tunnel_out_port,
3821 struct ofpbuf *odp_actions)
3822 {
3823 size_t userdata_ofs;
3824 size_t offset;
3825
3826 offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_USERSPACE);
3827 nl_msg_put_u32(odp_actions, OVS_USERSPACE_ATTR_PID, pid);
3828 if (userdata) {
3829 userdata_ofs = ofpbuf_size(odp_actions) + NLA_HDRLEN;
3830
3831 /* The OVS kernel module before OVS 1.11 and the upstream Linux kernel
3832 * module before Linux 3.10 required the userdata to be exactly 8 bytes
3833 * long:
3834 *
3835 * - The kernel rejected shorter userdata with -ERANGE.
3836 *
3837 * - The kernel silently dropped userdata beyond the first 8 bytes.
3838 *
3839 * Thus, for maximum compatibility, always put at least 8 bytes. (We
3840 * separately disable features that required more than 8 bytes.) */
3841 memcpy(nl_msg_put_unspec_zero(odp_actions, OVS_USERSPACE_ATTR_USERDATA,
3842 MAX(8, userdata_size)),
3843 userdata, userdata_size);
3844 } else {
3845 userdata_ofs = 0;
3846 }
3847 if (tunnel_out_port != ODPP_NONE) {
3848 nl_msg_put_odp_port(odp_actions, OVS_USERSPACE_ATTR_EGRESS_TUN_PORT,
3849 tunnel_out_port);
3850 }
3851 nl_msg_end_nested(odp_actions, offset);
3852
3853 return userdata_ofs;
3854 }
3855
3856 void
3857 odp_put_tunnel_action(const struct flow_tnl *tunnel,
3858 struct ofpbuf *odp_actions)
3859 {
3860 size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
3861 tun_key_to_attr(odp_actions, tunnel);
3862 nl_msg_end_nested(odp_actions, offset);
3863 }
3864
3865 void
3866 odp_put_tnl_push_action(struct ofpbuf *odp_actions,
3867 struct ovs_action_push_tnl *data)
3868 {
3869 int size = offsetof(struct ovs_action_push_tnl, header);
3870
3871 size += data->header_len;
3872 nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_TUNNEL_PUSH, data, size);
3873 }
3874
3875 \f
3876 /* The commit_odp_actions() function and its helpers. */
3877
3878 static void
3879 commit_set_action(struct ofpbuf *odp_actions, enum ovs_key_attr key_type,
3880 const void *key, size_t key_size)
3881 {
3882 size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
3883 nl_msg_put_unspec(odp_actions, key_type, key, key_size);
3884 nl_msg_end_nested(odp_actions, offset);
3885 }
3886
3887 /* Masked set actions have a mask following the data within the netlink
3888 * attribute. The unmasked bits in the data will be cleared as the data
3889 * is copied to the action. */
3890 void
3891 commit_masked_set_action(struct ofpbuf *odp_actions,
3892 enum ovs_key_attr key_type,
3893 const void *key_, const void *mask_, size_t key_size)
3894 {
3895 size_t offset = nl_msg_start_nested(odp_actions,
3896 OVS_ACTION_ATTR_SET_MASKED);
3897 char *data = nl_msg_put_unspec_uninit(odp_actions, key_type, key_size * 2);
3898 const char *key = key_, *mask = mask_;
3899
3900 memcpy(data + key_size, mask, key_size);
3901 /* Clear unmasked bits while copying. */
3902 while (key_size--) {
3903 *data++ = *key++ & *mask++;
3904 }
3905 nl_msg_end_nested(odp_actions, offset);
3906 }
3907
3908 /* If any of the flow key data that ODP actions can modify are different in
3909 * 'base->tunnel' and 'flow->tunnel', appends a set_tunnel ODP action to
3910 * 'odp_actions' that change the flow tunneling information in key from
3911 * 'base->tunnel' into 'flow->tunnel', and then changes 'base->tunnel' in the
3912 * same way. In other words, operates the same as commit_odp_actions(), but
3913 * only on tunneling information. */
3914 void
3915 commit_odp_tunnel_action(const struct flow *flow, struct flow *base,
3916 struct ofpbuf *odp_actions)
3917 {
3918 /* A valid IPV4_TUNNEL must have non-zero ip_dst. */
3919 if (flow->tunnel.ip_dst) {
3920 if (!memcmp(&base->tunnel, &flow->tunnel, sizeof base->tunnel)) {
3921 return;
3922 }
3923 memcpy(&base->tunnel, &flow->tunnel, sizeof base->tunnel);
3924 odp_put_tunnel_action(&base->tunnel, odp_actions);
3925 }
3926 }
3927
3928 static bool
3929 commit(enum ovs_key_attr attr, bool use_masked_set,
3930 const void *key, void *base, void *mask, size_t size,
3931 struct ofpbuf *odp_actions)
3932 {
3933 if (memcmp(key, base, size)) {
3934 bool fully_masked = odp_mask_is_exact(attr, mask, size);
3935
3936 if (use_masked_set && !fully_masked) {
3937 commit_masked_set_action(odp_actions, attr, key, mask, size);
3938 } else {
3939 if (!fully_masked) {
3940 memset(mask, 0xff, size);
3941 }
3942 commit_set_action(odp_actions, attr, key, size);
3943 }
3944 memcpy(base, key, size);
3945 return true;
3946 } else {
3947 /* Mask bits are set when we have either read or set the corresponding
3948 * values. Masked bits will be exact-matched, no need to set them
3949 * if the value did not actually change. */
3950 return false;
3951 }
3952 }
3953
3954 static void
3955 get_ethernet_key(const struct flow *flow, struct ovs_key_ethernet *eth)
3956 {
3957 memcpy(eth->eth_src, flow->dl_src, ETH_ADDR_LEN);
3958 memcpy(eth->eth_dst, flow->dl_dst, ETH_ADDR_LEN);
3959 }
3960
3961 static void
3962 put_ethernet_key(const struct ovs_key_ethernet *eth, struct flow *flow)
3963 {
3964 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
3965 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
3966 }
3967
3968 static void
3969 commit_set_ether_addr_action(const struct flow *flow, struct flow *base_flow,
3970 struct ofpbuf *odp_actions,
3971 struct flow_wildcards *wc,
3972 bool use_masked)
3973 {
3974 struct ovs_key_ethernet key, base, mask;
3975
3976 get_ethernet_key(flow, &key);
3977 get_ethernet_key(base_flow, &base);
3978 get_ethernet_key(&wc->masks, &mask);
3979
3980 if (commit(OVS_KEY_ATTR_ETHERNET, use_masked,
3981 &key, &base, &mask, sizeof key, odp_actions)) {
3982 put_ethernet_key(&base, base_flow);
3983 put_ethernet_key(&mask, &wc->masks);
3984 }
3985 }
3986
3987 static void
3988 pop_vlan(struct flow *base,
3989 struct ofpbuf *odp_actions, struct flow_wildcards *wc)
3990 {
3991 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3992
3993 if (base->vlan_tci & htons(VLAN_CFI)) {
3994 nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_VLAN);
3995 base->vlan_tci = 0;
3996 }
3997 }
3998
3999 static void
4000 commit_vlan_action(ovs_be16 vlan_tci, struct flow *base,
4001 struct ofpbuf *odp_actions, struct flow_wildcards *wc)
4002 {
4003 if (base->vlan_tci == vlan_tci) {
4004 return;
4005 }
4006
4007 pop_vlan(base, odp_actions, wc);
4008 if (vlan_tci & htons(VLAN_CFI)) {
4009 struct ovs_action_push_vlan vlan;
4010
4011 vlan.vlan_tpid = htons(ETH_TYPE_VLAN);
4012 vlan.vlan_tci = vlan_tci;
4013 nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_PUSH_VLAN,
4014 &vlan, sizeof vlan);
4015 }
4016 base->vlan_tci = vlan_tci;
4017 }
4018
4019 /* Wildcarding already done at action translation time. */
4020 static void
4021 commit_mpls_action(const struct flow *flow, struct flow *base,
4022 struct ofpbuf *odp_actions)
4023 {
4024 int base_n = flow_count_mpls_labels(base, NULL);
4025 int flow_n = flow_count_mpls_labels(flow, NULL);
4026 int common_n = flow_count_common_mpls_labels(flow, flow_n, base, base_n,
4027 NULL);
4028
4029 while (base_n > common_n) {
4030 if (base_n - 1 == common_n && flow_n > common_n) {
4031 /* If there is only one more LSE in base than there are common
4032 * between base and flow; and flow has at least one more LSE than
4033 * is common then the topmost LSE of base may be updated using
4034 * set */
4035 struct ovs_key_mpls mpls_key;
4036
4037 mpls_key.mpls_lse = flow->mpls_lse[flow_n - base_n];
4038 commit_set_action(odp_actions, OVS_KEY_ATTR_MPLS,
4039 &mpls_key, sizeof mpls_key);
4040 flow_set_mpls_lse(base, 0, mpls_key.mpls_lse);
4041 common_n++;
4042 } else {
4043 /* Otherwise, if there more LSEs in base than are common between
4044 * base and flow then pop the topmost one. */
4045 ovs_be16 dl_type;
4046 bool popped;
4047
4048 /* If all the LSEs are to be popped and this is not the outermost
4049 * LSE then use ETH_TYPE_MPLS as the ethertype parameter of the
4050 * POP_MPLS action instead of flow->dl_type.
4051 *
4052 * This is because the POP_MPLS action requires its ethertype
4053 * argument to be an MPLS ethernet type but in this case
4054 * flow->dl_type will be a non-MPLS ethernet type.
4055 *
4056 * When the final POP_MPLS action occurs it use flow->dl_type and
4057 * the and the resulting packet will have the desired dl_type. */
4058 if ((!eth_type_mpls(flow->dl_type)) && base_n > 1) {
4059 dl_type = htons(ETH_TYPE_MPLS);
4060 } else {
4061 dl_type = flow->dl_type;
4062 }
4063 nl_msg_put_be16(odp_actions, OVS_ACTION_ATTR_POP_MPLS, dl_type);
4064 popped = flow_pop_mpls(base, base_n, flow->dl_type, NULL);
4065 ovs_assert(popped);
4066 base_n--;
4067 }
4068 }
4069
4070 /* If, after the above popping and setting, there are more LSEs in flow
4071 * than base then some LSEs need to be pushed. */
4072 while (base_n < flow_n) {
4073 struct ovs_action_push_mpls *mpls;
4074
4075 mpls = nl_msg_put_unspec_zero(odp_actions,
4076 OVS_ACTION_ATTR_PUSH_MPLS,
4077 sizeof *mpls);
4078 mpls->mpls_ethertype = flow->dl_type;
4079 mpls->mpls_lse = flow->mpls_lse[flow_n - base_n - 1];
4080 flow_push_mpls(base, base_n, mpls->mpls_ethertype, NULL);
4081 flow_set_mpls_lse(base, 0, mpls->mpls_lse);
4082 base_n++;
4083 }
4084 }
4085
4086 static void
4087 get_ipv4_key(const struct flow *flow, struct ovs_key_ipv4 *ipv4, bool is_mask)
4088 {
4089 ipv4->ipv4_src = flow->nw_src;
4090 ipv4->ipv4_dst = flow->nw_dst;
4091 ipv4->ipv4_proto = flow->nw_proto;
4092 ipv4->ipv4_tos = flow->nw_tos;
4093 ipv4->ipv4_ttl = flow->nw_ttl;
4094 ipv4->ipv4_frag = ovs_to_odp_frag(flow->nw_frag, is_mask);
4095 }
4096
4097 static void
4098 put_ipv4_key(const struct ovs_key_ipv4 *ipv4, struct flow *flow, bool is_mask)
4099 {
4100 flow->nw_src = ipv4->ipv4_src;
4101 flow->nw_dst = ipv4->ipv4_dst;
4102 flow->nw_proto = ipv4->ipv4_proto;
4103 flow->nw_tos = ipv4->ipv4_tos;
4104 flow->nw_ttl = ipv4->ipv4_ttl;
4105 flow->nw_frag = odp_to_ovs_frag(ipv4->ipv4_frag, is_mask);
4106 }
4107
4108 static void
4109 commit_set_ipv4_action(const struct flow *flow, struct flow *base_flow,
4110 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4111 bool use_masked)
4112 {
4113 struct ovs_key_ipv4 key, mask, base;
4114
4115 /* Check that nw_proto and nw_frag remain unchanged. */
4116 ovs_assert(flow->nw_proto == base_flow->nw_proto &&
4117 flow->nw_frag == base_flow->nw_frag);
4118
4119 get_ipv4_key(flow, &key, false);
4120 get_ipv4_key(base_flow, &base, false);
4121 get_ipv4_key(&wc->masks, &mask, true);
4122 mask.ipv4_proto = 0; /* Not writeable. */
4123 mask.ipv4_frag = 0; /* Not writable. */
4124
4125 if (commit(OVS_KEY_ATTR_IPV4, use_masked, &key, &base, &mask, sizeof key,
4126 odp_actions)) {
4127 put_ipv4_key(&base, base_flow, false);
4128 if (mask.ipv4_proto != 0) { /* Mask was changed by commit(). */
4129 put_ipv4_key(&mask, &wc->masks, true);
4130 }
4131 }
4132 }
4133
4134 static void
4135 get_ipv6_key(const struct flow *flow, struct ovs_key_ipv6 *ipv6, bool is_mask)
4136 {
4137 memcpy(ipv6->ipv6_src, &flow->ipv6_src, sizeof ipv6->ipv6_src);
4138 memcpy(ipv6->ipv6_dst, &flow->ipv6_dst, sizeof ipv6->ipv6_dst);
4139 ipv6->ipv6_label = flow->ipv6_label;
4140 ipv6->ipv6_proto = flow->nw_proto;
4141 ipv6->ipv6_tclass = flow->nw_tos;
4142 ipv6->ipv6_hlimit = flow->nw_ttl;
4143 ipv6->ipv6_frag = ovs_to_odp_frag(flow->nw_frag, is_mask);
4144 }
4145
4146 static void
4147 put_ipv6_key(const struct ovs_key_ipv6 *ipv6, struct flow *flow, bool is_mask)
4148 {
4149 memcpy(&flow->ipv6_src, ipv6->ipv6_src, sizeof flow->ipv6_src);
4150 memcpy(&flow->ipv6_dst, ipv6->ipv6_dst, sizeof flow->ipv6_dst);
4151 flow->ipv6_label = ipv6->ipv6_label;
4152 flow->nw_proto = ipv6->ipv6_proto;
4153 flow->nw_tos = ipv6->ipv6_tclass;
4154 flow->nw_ttl = ipv6->ipv6_hlimit;
4155 flow->nw_frag = odp_to_ovs_frag(ipv6->ipv6_frag, is_mask);
4156 }
4157
4158 static void
4159 commit_set_ipv6_action(const struct flow *flow, struct flow *base_flow,
4160 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4161 bool use_masked)
4162 {
4163 struct ovs_key_ipv6 key, mask, base;
4164
4165 /* Check that nw_proto and nw_frag remain unchanged. */
4166 ovs_assert(flow->nw_proto == base_flow->nw_proto &&
4167 flow->nw_frag == base_flow->nw_frag);
4168
4169 get_ipv6_key(flow, &key, false);
4170 get_ipv6_key(base_flow, &base, false);
4171 get_ipv6_key(&wc->masks, &mask, true);
4172 mask.ipv6_proto = 0; /* Not writeable. */
4173 mask.ipv6_frag = 0; /* Not writable. */
4174
4175 if (commit(OVS_KEY_ATTR_IPV6, use_masked, &key, &base, &mask, sizeof key,
4176 odp_actions)) {
4177 put_ipv6_key(&base, base_flow, false);
4178 if (mask.ipv6_proto != 0) { /* Mask was changed by commit(). */
4179 put_ipv6_key(&mask, &wc->masks, true);
4180 }
4181 }
4182 }
4183
4184 static void
4185 get_arp_key(const struct flow *flow, struct ovs_key_arp *arp)
4186 {
4187 /* ARP key has padding, clear it. */
4188 memset(arp, 0, sizeof *arp);
4189
4190 arp->arp_sip = flow->nw_src;
4191 arp->arp_tip = flow->nw_dst;
4192 arp->arp_op = htons(flow->nw_proto);
4193 memcpy(arp->arp_sha, flow->arp_sha, ETH_ADDR_LEN);
4194 memcpy(arp->arp_tha, flow->arp_tha, ETH_ADDR_LEN);
4195 }
4196
4197 static void
4198 put_arp_key(const struct ovs_key_arp *arp, struct flow *flow)
4199 {
4200 flow->nw_src = arp->arp_sip;
4201 flow->nw_dst = arp->arp_tip;
4202 flow->nw_proto = ntohs(arp->arp_op);
4203 memcpy(flow->arp_sha, arp->arp_sha, ETH_ADDR_LEN);
4204 memcpy(flow->arp_tha, arp->arp_tha, ETH_ADDR_LEN);
4205 }
4206
4207 static enum slow_path_reason
4208 commit_set_arp_action(const struct flow *flow, struct flow *base_flow,
4209 struct ofpbuf *odp_actions, struct flow_wildcards *wc)
4210 {
4211 struct ovs_key_arp key, mask, base;
4212
4213 get_arp_key(flow, &key);
4214 get_arp_key(base_flow, &base);
4215 get_arp_key(&wc->masks, &mask);
4216
4217 if (commit(OVS_KEY_ATTR_ARP, true, &key, &base, &mask, sizeof key,
4218 odp_actions)) {
4219 put_arp_key(&base, base_flow);
4220 put_arp_key(&mask, &wc->masks);
4221 return SLOW_ACTION;
4222 }
4223 return 0;
4224 }
4225
4226 static void
4227 get_nd_key(const struct flow *flow, struct ovs_key_nd *nd)
4228 {
4229 memcpy(nd->nd_target, &flow->nd_target, sizeof flow->nd_target);
4230 /* nd_sll and nd_tll are stored in arp_sha and arp_tha, respectively */
4231 memcpy(nd->nd_sll, flow->arp_sha, ETH_ADDR_LEN);
4232 memcpy(nd->nd_tll, flow->arp_tha, ETH_ADDR_LEN);
4233 }
4234
4235 static void
4236 put_nd_key(const struct ovs_key_nd *nd, struct flow *flow)
4237 {
4238 memcpy(&flow->nd_target, &flow->nd_target, sizeof flow->nd_target);
4239 /* nd_sll and nd_tll are stored in arp_sha and arp_tha, respectively */
4240 memcpy(flow->arp_sha, nd->nd_sll, ETH_ADDR_LEN);
4241 memcpy(flow->arp_tha, nd->nd_tll, ETH_ADDR_LEN);
4242 }
4243
4244 static enum slow_path_reason
4245 commit_set_nd_action(const struct flow *flow, struct flow *base_flow,
4246 struct ofpbuf *odp_actions,
4247 struct flow_wildcards *wc, bool use_masked)
4248 {
4249 struct ovs_key_nd key, mask, base;
4250
4251 get_nd_key(flow, &key);
4252 get_nd_key(base_flow, &base);
4253 get_nd_key(&wc->masks, &mask);
4254
4255 if (commit(OVS_KEY_ATTR_ND, use_masked, &key, &base, &mask, sizeof key,
4256 odp_actions)) {
4257 put_nd_key(&base, base_flow);
4258 put_nd_key(&mask, &wc->masks);
4259 return SLOW_ACTION;
4260 }
4261
4262 return 0;
4263 }
4264
4265 static enum slow_path_reason
4266 commit_set_nw_action(const struct flow *flow, struct flow *base,
4267 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4268 bool use_masked)
4269 {
4270 /* Check if 'flow' really has an L3 header. */
4271 if (!flow->nw_proto) {
4272 return 0;
4273 }
4274
4275 switch (ntohs(base->dl_type)) {
4276 case ETH_TYPE_IP:
4277 commit_set_ipv4_action(flow, base, odp_actions, wc, use_masked);
4278 break;
4279
4280 case ETH_TYPE_IPV6:
4281 commit_set_ipv6_action(flow, base, odp_actions, wc, use_masked);
4282 return commit_set_nd_action(flow, base, odp_actions, wc, use_masked);
4283
4284 case ETH_TYPE_ARP:
4285 return commit_set_arp_action(flow, base, odp_actions, wc);
4286 }
4287
4288 return 0;
4289 }
4290
4291 /* TCP, UDP, and SCTP keys have the same layout. */
4292 BUILD_ASSERT_DECL(sizeof(struct ovs_key_tcp) == sizeof(struct ovs_key_udp) &&
4293 sizeof(struct ovs_key_tcp) == sizeof(struct ovs_key_sctp));
4294
4295 static void
4296 get_tp_key(const struct flow *flow, union ovs_key_tp *tp)
4297 {
4298 tp->tcp.tcp_src = flow->tp_src;
4299 tp->tcp.tcp_dst = flow->tp_dst;
4300 }
4301
4302 static void
4303 put_tp_key(const union ovs_key_tp *tp, struct flow *flow)
4304 {
4305 flow->tp_src = tp->tcp.tcp_src;
4306 flow->tp_dst = tp->tcp.tcp_dst;
4307 }
4308
4309 static void
4310 commit_set_port_action(const struct flow *flow, struct flow *base_flow,
4311 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4312 bool use_masked)
4313 {
4314 enum ovs_key_attr key_type;
4315 union ovs_key_tp key, mask, base;
4316
4317 /* Check if 'flow' really has an L3 header. */
4318 if (!flow->nw_proto) {
4319 return;
4320 }
4321
4322 if (!is_ip_any(base_flow)) {
4323 return;
4324 }
4325
4326 if (flow->nw_proto == IPPROTO_TCP) {
4327 key_type = OVS_KEY_ATTR_TCP;
4328 } else if (flow->nw_proto == IPPROTO_UDP) {
4329 key_type = OVS_KEY_ATTR_UDP;
4330 } else if (flow->nw_proto == IPPROTO_SCTP) {
4331 key_type = OVS_KEY_ATTR_SCTP;
4332 } else {
4333 return;
4334 }
4335
4336 get_tp_key(flow, &key);
4337 get_tp_key(base_flow, &base);
4338 get_tp_key(&wc->masks, &mask);
4339
4340 if (commit(key_type, use_masked, &key, &base, &mask, sizeof key,
4341 odp_actions)) {
4342 put_tp_key(&base, base_flow);
4343 put_tp_key(&mask, &wc->masks);
4344 }
4345 }
4346
4347 static void
4348 commit_set_priority_action(const struct flow *flow, struct flow *base_flow,
4349 struct ofpbuf *odp_actions,
4350 struct flow_wildcards *wc,
4351 bool use_masked)
4352 {
4353 uint32_t key, mask, base;
4354
4355 key = flow->skb_priority;
4356 base = base_flow->skb_priority;
4357 mask = wc->masks.skb_priority;
4358
4359 if (commit(OVS_KEY_ATTR_PRIORITY, use_masked, &key, &base, &mask,
4360 sizeof key, odp_actions)) {
4361 base_flow->skb_priority = base;
4362 wc->masks.skb_priority = mask;
4363 }
4364 }
4365
4366 static void
4367 commit_set_pkt_mark_action(const struct flow *flow, struct flow *base_flow,
4368 struct ofpbuf *odp_actions,
4369 struct flow_wildcards *wc,
4370 bool use_masked)
4371 {
4372 uint32_t key, mask, base;
4373
4374 key = flow->pkt_mark;
4375 base = base_flow->pkt_mark;
4376 mask = wc->masks.pkt_mark;
4377
4378 if (commit(OVS_KEY_ATTR_SKB_MARK, use_masked, &key, &base, &mask,
4379 sizeof key, odp_actions)) {
4380 base_flow->pkt_mark = base;
4381 wc->masks.pkt_mark = mask;
4382 }
4383 }
4384
4385 /* If any of the flow key data that ODP actions can modify are different in
4386 * 'base' and 'flow', appends ODP actions to 'odp_actions' that change the flow
4387 * key from 'base' into 'flow', and then changes 'base' the same way. Does not
4388 * commit set_tunnel actions. Users should call commit_odp_tunnel_action()
4389 * in addition to this function if needed. Sets fields in 'wc' that are
4390 * used as part of the action.
4391 *
4392 * Returns a reason to force processing the flow's packets into the userspace
4393 * slow path, if there is one, otherwise 0. */
4394 enum slow_path_reason
4395 commit_odp_actions(const struct flow *flow, struct flow *base,
4396 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4397 bool use_masked)
4398 {
4399 enum slow_path_reason slow;
4400
4401 commit_set_ether_addr_action(flow, base, odp_actions, wc, use_masked);
4402 slow = commit_set_nw_action(flow, base, odp_actions, wc, use_masked);
4403 commit_set_port_action(flow, base, odp_actions, wc, use_masked);
4404 commit_mpls_action(flow, base, odp_actions);
4405 commit_vlan_action(flow->vlan_tci, base, odp_actions, wc);
4406 commit_set_priority_action(flow, base, odp_actions, wc, use_masked);
4407 commit_set_pkt_mark_action(flow, base, odp_actions, wc, use_masked);
4408
4409 return slow;
4410 }