]> git.proxmox.com Git - ovs.git/blob - lib/odp-util.c
projects / ovs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
odp-util: Geneve netlink decoding.
[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 const char *hex_chars = "0123456789abcdefABCDEF";
55
56 struct attr_len_tbl {
57 int len;
58 const struct attr_len_tbl *next;
59 int next_max;
60 };
61 #define ATTR_LEN_INVALID -1
62 #define ATTR_LEN_VARIABLE -2
63 #define ATTR_LEN_NESTED -3
64
65 static int parse_odp_key_mask_attr(const char *, const struct simap *port_names,
66 struct ofpbuf *, struct ofpbuf *);
67 static void format_odp_key_attr(const struct nlattr *a,
68 const struct nlattr *ma,
69 const struct hmap *portno_names, struct ds *ds,
70 bool verbose);
71
72 static struct nlattr *generate_all_wildcard_mask(const struct attr_len_tbl tbl[],
73 int max, struct ofpbuf *,
74 const struct nlattr *key);
75 /* Returns one the following for the action with the given OVS_ACTION_ATTR_*
76 * 'type':
77 *
78 * - For an action whose argument has a fixed length, returned that
79 * nonnegative length in bytes.
80 *
81 * - For an action with a variable-length argument, returns ATTR_LEN_VARIABLE.
82 *
83 * - For an invalid 'type', returns ATTR_LEN_INVALID. */
84 static int
85 odp_action_len(uint16_t type)
86 {
87 if (type > OVS_ACTION_ATTR_MAX) {
88 return -1;
89 }
90
91 switch ((enum ovs_action_attr) type) {
92 case OVS_ACTION_ATTR_OUTPUT: return sizeof(uint32_t);
93 case OVS_ACTION_ATTR_TUNNEL_PUSH: return ATTR_LEN_VARIABLE;
94 case OVS_ACTION_ATTR_TUNNEL_POP: return sizeof(uint32_t);
95 case OVS_ACTION_ATTR_USERSPACE: return ATTR_LEN_VARIABLE;
96 case OVS_ACTION_ATTR_PUSH_VLAN: return sizeof(struct ovs_action_push_vlan);
97 case OVS_ACTION_ATTR_POP_VLAN: return 0;
98 case OVS_ACTION_ATTR_PUSH_MPLS: return sizeof(struct ovs_action_push_mpls);
99 case OVS_ACTION_ATTR_POP_MPLS: return sizeof(ovs_be16);
100 case OVS_ACTION_ATTR_RECIRC: return sizeof(uint32_t);
101 case OVS_ACTION_ATTR_HASH: return sizeof(struct ovs_action_hash);
102 case OVS_ACTION_ATTR_SET: return ATTR_LEN_VARIABLE;
103 case OVS_ACTION_ATTR_SET_MASKED: return ATTR_LEN_VARIABLE;
104 case OVS_ACTION_ATTR_SAMPLE: return ATTR_LEN_VARIABLE;
105
106 case OVS_ACTION_ATTR_UNSPEC:
107 case __OVS_ACTION_ATTR_MAX:
108 return ATTR_LEN_INVALID;
109 }
110
111 return ATTR_LEN_INVALID;
112 }
113
114 /* Returns a string form of 'attr'. The return value is either a statically
115 * allocated constant string or the 'bufsize'-byte buffer 'namebuf'. 'bufsize'
116 * should be at least OVS_KEY_ATTR_BUFSIZE. */
117 enum { OVS_KEY_ATTR_BUFSIZE = 3 + INT_STRLEN(unsigned int) + 1 };
118 static const char *
119 ovs_key_attr_to_string(enum ovs_key_attr attr, char *namebuf, size_t bufsize)
120 {
121 switch (attr) {
122 case OVS_KEY_ATTR_UNSPEC: return "unspec";
123 case OVS_KEY_ATTR_ENCAP: return "encap";
124 case OVS_KEY_ATTR_PRIORITY: return "skb_priority";
125 case OVS_KEY_ATTR_SKB_MARK: return "skb_mark";
126 case OVS_KEY_ATTR_TUNNEL: return "tunnel";
127 case OVS_KEY_ATTR_IN_PORT: return "in_port";
128 case OVS_KEY_ATTR_ETHERNET: return "eth";
129 case OVS_KEY_ATTR_VLAN: return "vlan";
130 case OVS_KEY_ATTR_ETHERTYPE: return "eth_type";
131 case OVS_KEY_ATTR_IPV4: return "ipv4";
132 case OVS_KEY_ATTR_IPV6: return "ipv6";
133 case OVS_KEY_ATTR_TCP: return "tcp";
134 case OVS_KEY_ATTR_TCP_FLAGS: return "tcp_flags";
135 case OVS_KEY_ATTR_UDP: return "udp";
136 case OVS_KEY_ATTR_SCTP: return "sctp";
137 case OVS_KEY_ATTR_ICMP: return "icmp";
138 case OVS_KEY_ATTR_ICMPV6: return "icmpv6";
139 case OVS_KEY_ATTR_ARP: return "arp";
140 case OVS_KEY_ATTR_ND: return "nd";
141 case OVS_KEY_ATTR_MPLS: return "mpls";
142 case OVS_KEY_ATTR_DP_HASH: return "dp_hash";
143 case OVS_KEY_ATTR_RECIRC_ID: return "recirc_id";
144
145 case __OVS_KEY_ATTR_MAX:
146 default:
147 snprintf(namebuf, bufsize, "key%u", (unsigned int) attr);
148 return namebuf;
149 }
150 }
151
152 static void
153 format_generic_odp_action(struct ds *ds, const struct nlattr *a)
154 {
155 size_t len = nl_attr_get_size(a);
156
157 ds_put_format(ds, "action%"PRId16, nl_attr_type(a));
158 if (len) {
159 const uint8_t *unspec;
160 unsigned int i;
161
162 unspec = nl_attr_get(a);
163 for (i = 0; i < len; i++) {
164 ds_put_char(ds, i ? ' ': '(');
165 ds_put_format(ds, "%02x", unspec[i]);
166 }
167 ds_put_char(ds, ')');
168 }
169 }
170
171 static void
172 format_odp_sample_action(struct ds *ds, const struct nlattr *attr)
173 {
174 static const struct nl_policy ovs_sample_policy[] = {
175 [OVS_SAMPLE_ATTR_PROBABILITY] = { .type = NL_A_U32 },
176 [OVS_SAMPLE_ATTR_ACTIONS] = { .type = NL_A_NESTED }
177 };
178 struct nlattr *a[ARRAY_SIZE(ovs_sample_policy)];
179 double percentage;
180 const struct nlattr *nla_acts;
181 int len;
182
183 ds_put_cstr(ds, "sample");
184
185 if (!nl_parse_nested(attr, ovs_sample_policy, a, ARRAY_SIZE(a))) {
186 ds_put_cstr(ds, "(error)");
187 return;
188 }
189
190 percentage = (100.0 * nl_attr_get_u32(a[OVS_SAMPLE_ATTR_PROBABILITY])) /
191 UINT32_MAX;
192
193 ds_put_format(ds, "(sample=%.1f%%,", percentage);
194
195 ds_put_cstr(ds, "actions(");
196 nla_acts = nl_attr_get(a[OVS_SAMPLE_ATTR_ACTIONS]);
197 len = nl_attr_get_size(a[OVS_SAMPLE_ATTR_ACTIONS]);
198 format_odp_actions(ds, nla_acts, len);
199 ds_put_format(ds, "))");
200 }
201
202 static const char *
203 slow_path_reason_to_string(uint32_t reason)
204 {
205 switch ((enum slow_path_reason) reason) {
206 #define SPR(ENUM, STRING, EXPLANATION) case ENUM: return STRING;
207 SLOW_PATH_REASONS
208 #undef SPR
209 }
210
211 return NULL;
212 }
213
214 const char *
215 slow_path_reason_to_explanation(enum slow_path_reason reason)
216 {
217 switch (reason) {
218 #define SPR(ENUM, STRING, EXPLANATION) case ENUM: return EXPLANATION;
219 SLOW_PATH_REASONS
220 #undef SPR
221 }
222
223 return "<unknown>";
224 }
225
226 static int
227 parse_flags(const char *s, const char *(*bit_to_string)(uint32_t),
228 uint32_t *res_flags, uint32_t allowed, uint32_t *res_mask)
229 {
230 uint32_t result = 0;
231 int n;
232
233 /* Parse masked flags in numeric format? */
234 if (res_mask && ovs_scan(s, "%"SCNi32"/%"SCNi32"%n",
235 res_flags, res_mask, &n) && n > 0) {
236 if (*res_flags & ~allowed || *res_mask & ~allowed) {
237 return -EINVAL;
238 }
239 return n;
240 }
241
242 n = 0;
243
244 if (res_mask && (*s == '+' || *s == '-')) {
245 uint32_t flags = 0, mask = 0;
246
247 /* Parse masked flags. */
248 while (s[0] != ')') {
249 bool set;
250 uint32_t bit;
251 int name_len;
252
253 if (s[0] == '+') {
254 set = true;
255 } else if (s[0] == '-') {
256 set = false;
257 } else {
258 return -EINVAL;
259 }
260 s++;
261 n++;
262
263 name_len = strcspn(s, "+-)");
264
265 for (bit = 1; bit; bit <<= 1) {
266 const char *fname = bit_to_string(bit);
267 size_t len;
268
269 if (!fname) {
270 continue;
271 }
272
273 len = strlen(fname);
274 if (len != name_len) {
275 continue;
276 }
277 if (!strncmp(s, fname, len)) {
278 if (mask & bit) {
279 /* bit already set. */
280 return -EINVAL;
281 }
282 if (!(bit & allowed)) {
283 return -EINVAL;
284 }
285 if (set) {
286 flags |= bit;
287 }
288 mask |= bit;
289 break;
290 }
291 }
292
293 if (!bit) {
294 return -EINVAL; /* Unknown flag name */
295 }
296 s += name_len;
297 n += name_len;
298 }
299
300 *res_flags = flags;
301 *res_mask = mask;
302 return n;
303 }
304
305 /* Parse unmasked flags. If a flag is present, it is set, otherwise
306 * it is not set. */
307 while (s[n] != ')') {
308 unsigned long long int flags;
309 uint32_t bit;
310 int n0;
311
312 if (ovs_scan(&s[n], "%lli%n", &flags, &n0)) {
313 if (flags & ~allowed) {
314 return -EINVAL;
315 }
316 n += n0 + (s[n + n0] == ',');
317 result |= flags;
318 continue;
319 }
320
321 for (bit = 1; bit; bit <<= 1) {
322 const char *name = bit_to_string(bit);
323 size_t len;
324
325 if (!name) {
326 continue;
327 }
328
329 len = strlen(name);
330 if (!strncmp(s + n, name, len) &&
331 (s[n + len] == ',' || s[n + len] == ')')) {
332 if (!(bit & allowed)) {
333 return -EINVAL;
334 }
335 result |= bit;
336 n += len + (s[n + len] == ',');
337 break;
338 }
339 }
340
341 if (!bit) {
342 return -EINVAL;
343 }
344 }
345
346 *res_flags = result;
347 if (res_mask) {
348 *res_mask = UINT32_MAX;
349 }
350 return n;
351 }
352
353 static void
354 format_odp_userspace_action(struct ds *ds, const struct nlattr *attr)
355 {
356 static const struct nl_policy ovs_userspace_policy[] = {
357 [OVS_USERSPACE_ATTR_PID] = { .type = NL_A_U32 },
358 [OVS_USERSPACE_ATTR_USERDATA] = { .type = NL_A_UNSPEC,
359 .optional = true },
360 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = { .type = NL_A_U32,
361 .optional = true },
362 };
363 struct nlattr *a[ARRAY_SIZE(ovs_userspace_policy)];
364 const struct nlattr *userdata_attr;
365 const struct nlattr *tunnel_out_port_attr;
366
367 if (!nl_parse_nested(attr, ovs_userspace_policy, a, ARRAY_SIZE(a))) {
368 ds_put_cstr(ds, "userspace(error)");
369 return;
370 }
371
372 ds_put_format(ds, "userspace(pid=%"PRIu32,
373 nl_attr_get_u32(a[OVS_USERSPACE_ATTR_PID]));
374
375 userdata_attr = a[OVS_USERSPACE_ATTR_USERDATA];
376
377 if (userdata_attr) {
378 const uint8_t *userdata = nl_attr_get(userdata_attr);
379 size_t userdata_len = nl_attr_get_size(userdata_attr);
380 bool userdata_unspec = true;
381 union user_action_cookie cookie;
382
383 if (userdata_len >= sizeof cookie.type
384 && userdata_len <= sizeof cookie) {
385
386 memset(&cookie, 0, sizeof cookie);
387 memcpy(&cookie, userdata, userdata_len);
388
389 userdata_unspec = false;
390
391 if (userdata_len == sizeof cookie.sflow
392 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
393 ds_put_format(ds, ",sFlow("
394 "vid=%"PRIu16",pcp=%"PRIu8",output=%"PRIu32")",
395 vlan_tci_to_vid(cookie.sflow.vlan_tci),
396 vlan_tci_to_pcp(cookie.sflow.vlan_tci),
397 cookie.sflow.output);
398 } else if (userdata_len == sizeof cookie.slow_path
399 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
400 ds_put_cstr(ds, ",slow_path(");
401 format_flags(ds, slow_path_reason_to_string,
402 cookie.slow_path.reason, ',');
403 ds_put_format(ds, ")");
404 } else if (userdata_len == sizeof cookie.flow_sample
405 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
406 ds_put_format(ds, ",flow_sample(probability=%"PRIu16
407 ",collector_set_id=%"PRIu32
408 ",obs_domain_id=%"PRIu32
409 ",obs_point_id=%"PRIu32")",
410 cookie.flow_sample.probability,
411 cookie.flow_sample.collector_set_id,
412 cookie.flow_sample.obs_domain_id,
413 cookie.flow_sample.obs_point_id);
414 } else if (userdata_len >= sizeof cookie.ipfix
415 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
416 ds_put_format(ds, ",ipfix(output_port=%"PRIu32")",
417 cookie.ipfix.output_odp_port);
418 } else {
419 userdata_unspec = true;
420 }
421 }
422
423 if (userdata_unspec) {
424 size_t i;
425 ds_put_format(ds, ",userdata(");
426 for (i = 0; i < userdata_len; i++) {
427 ds_put_format(ds, "%02x", userdata[i]);
428 }
429 ds_put_char(ds, ')');
430 }
431 }
432
433 tunnel_out_port_attr = a[OVS_USERSPACE_ATTR_EGRESS_TUN_PORT];
434 if (tunnel_out_port_attr) {
435 ds_put_format(ds, ",tunnel_out_port=%"PRIu32,
436 nl_attr_get_u32(tunnel_out_port_attr));
437 }
438
439 ds_put_char(ds, ')');
440 }
441
442 static void
443 format_vlan_tci(struct ds *ds, ovs_be16 tci, ovs_be16 mask, bool verbose)
444 {
445 if (verbose || vlan_tci_to_vid(tci) || vlan_tci_to_vid(mask)) {
446 ds_put_format(ds, "vid=%"PRIu16, vlan_tci_to_vid(tci));
447 if (vlan_tci_to_vid(mask) != VLAN_VID_MASK) { /* Partially masked. */
448 ds_put_format(ds, "/0x%"PRIx16, vlan_tci_to_vid(mask));
449 };
450 ds_put_char(ds, ',');
451 }
452 if (verbose || vlan_tci_to_pcp(tci) || vlan_tci_to_pcp(mask)) {
453 ds_put_format(ds, "pcp=%d", vlan_tci_to_pcp(tci));
454 if (vlan_tci_to_pcp(mask) != (VLAN_PCP_MASK >> VLAN_PCP_SHIFT)) {
455 ds_put_format(ds, "/0x%x", vlan_tci_to_pcp(mask));
456 }
457 ds_put_char(ds, ',');
458 }
459 if (!(tci & htons(VLAN_CFI))) {
460 ds_put_cstr(ds, "cfi=0");
461 ds_put_char(ds, ',');
462 }
463 ds_chomp(ds, ',');
464 }
465
466 static void
467 format_mpls_lse(struct ds *ds, ovs_be32 mpls_lse)
468 {
469 ds_put_format(ds, "label=%"PRIu32",tc=%d,ttl=%d,bos=%d",
470 mpls_lse_to_label(mpls_lse),
471 mpls_lse_to_tc(mpls_lse),
472 mpls_lse_to_ttl(mpls_lse),
473 mpls_lse_to_bos(mpls_lse));
474 }
475
476 static void
477 format_mpls(struct ds *ds, const struct ovs_key_mpls *mpls_key,
478 const struct ovs_key_mpls *mpls_mask, int n)
479 {
480 if (n == 1) {
481 ovs_be32 key = mpls_key->mpls_lse;
482
483 if (mpls_mask == NULL) {
484 format_mpls_lse(ds, key);
485 } else {
486 ovs_be32 mask = mpls_mask->mpls_lse;
487
488 ds_put_format(ds, "label=%"PRIu32"/0x%x,tc=%d/%x,ttl=%d/0x%x,bos=%d/%x",
489 mpls_lse_to_label(key), mpls_lse_to_label(mask),
490 mpls_lse_to_tc(key), mpls_lse_to_tc(mask),
491 mpls_lse_to_ttl(key), mpls_lse_to_ttl(mask),
492 mpls_lse_to_bos(key), mpls_lse_to_bos(mask));
493 }
494 } else {
495 int i;
496
497 for (i = 0; i < n; i++) {
498 ds_put_format(ds, "lse%d=%#"PRIx32,
499 i, ntohl(mpls_key[i].mpls_lse));
500 if (mpls_mask) {
501 ds_put_format(ds, "/%#"PRIx32, ntohl(mpls_mask[i].mpls_lse));
502 }
503 ds_put_char(ds, ',');
504 }
505 ds_chomp(ds, ',');
506 }
507 }
508
509 static void
510 format_odp_recirc_action(struct ds *ds, uint32_t recirc_id)
511 {
512 ds_put_format(ds, "recirc(%#"PRIx32")", recirc_id);
513 }
514
515 static void
516 format_odp_hash_action(struct ds *ds, const struct ovs_action_hash *hash_act)
517 {
518 ds_put_format(ds, "hash(");
519
520 if (hash_act->hash_alg == OVS_HASH_ALG_L4) {
521 ds_put_format(ds, "hash_l4(%"PRIu32")", hash_act->hash_basis);
522 } else {
523 ds_put_format(ds, "Unknown hash algorithm(%"PRIu32")",
524 hash_act->hash_alg);
525 }
526 ds_put_format(ds, ")");
527 }
528
529 static const void *
530 format_udp_tnl_push_header(struct ds *ds, const struct ip_header *ip)
531 {
532 const struct udp_header *udp;
533
534 udp = (const struct udp_header *) (ip + 1);
535 ds_put_format(ds, "udp(src=%"PRIu16",dst=%"PRIu16",csum=0x%"PRIx16"),",
536 ntohs(udp->udp_src), ntohs(udp->udp_dst),
537 ntohs(udp->udp_csum));
538
539 return udp + 1;
540 }
541
542 static void
543 format_odp_tnl_push_header(struct ds *ds, struct ovs_action_push_tnl *data)
544 {
545 const struct eth_header *eth;
546 const struct ip_header *ip;
547 const void *l3;
548
549 eth = (const struct eth_header *)data->header;
550
551 l3 = eth + 1;
552 ip = (const struct ip_header *)l3;
553
554 /* Ethernet */
555 ds_put_format(ds, "header(size=%"PRIu8",type=%"PRIu8",eth(dst=",
556 data->header_len, data->tnl_type);
557 ds_put_format(ds, ETH_ADDR_FMT, ETH_ADDR_ARGS(eth->eth_dst));
558 ds_put_format(ds, ",src=");
559 ds_put_format(ds, ETH_ADDR_FMT, ETH_ADDR_ARGS(eth->eth_src));
560 ds_put_format(ds, ",dl_type=0x%04"PRIx16"),", ntohs(eth->eth_type));
561
562 /* IPv4 */
563 ds_put_format(ds, "ipv4(src="IP_FMT",dst="IP_FMT",proto=%"PRIu8
564 ",tos=%#"PRIx8",ttl=%"PRIu8",frag=0x%"PRIx16"),",
565 IP_ARGS(get_16aligned_be32(&ip->ip_src)),
566 IP_ARGS(get_16aligned_be32(&ip->ip_dst)),
567 ip->ip_proto, ip->ip_tos,
568 ip->ip_ttl,
569 ip->ip_frag_off);
570
571 if (data->tnl_type == OVS_VPORT_TYPE_VXLAN) {
572 const struct vxlanhdr *vxh;
573
574 vxh = format_udp_tnl_push_header(ds, ip);
575
576 ds_put_format(ds, "vxlan(flags=0x%"PRIx32",vni=0x%"PRIx32")",
577 ntohl(get_16aligned_be32(&vxh->vx_flags)),
578 ntohl(get_16aligned_be32(&vxh->vx_vni)) >> 8);
579 } else if (data->tnl_type == OVS_VPORT_TYPE_GENEVE) {
580 const struct genevehdr *gnh;
581
582 gnh = format_udp_tnl_push_header(ds, ip);
583
584 ds_put_format(ds, "geneve(%svni=0x%"PRIx32")",
585 gnh->oam ? "oam," : "",
586 ntohl(get_16aligned_be32(&gnh->vni)) >> 8);
587 } else if (data->tnl_type == OVS_VPORT_TYPE_GRE) {
588 const struct gre_base_hdr *greh;
589 ovs_16aligned_be32 *options;
590 void *l4;
591
592 l4 = ((uint8_t *)l3 + sizeof(struct ip_header));
593 greh = (const struct gre_base_hdr *) l4;
594
595 ds_put_format(ds, "gre((flags=0x%"PRIx16",proto=0x%"PRIx16")",
596 ntohs(greh->flags), ntohs(greh->protocol));
597 options = (ovs_16aligned_be32 *)(greh + 1);
598 if (greh->flags & htons(GRE_CSUM)) {
599 ds_put_format(ds, ",csum=0x%"PRIx16, ntohs(*((ovs_be16 *)options)));
600 options++;
601 }
602 if (greh->flags & htons(GRE_KEY)) {
603 ds_put_format(ds, ",key=0x%"PRIx32, ntohl(get_16aligned_be32(options)));
604 options++;
605 }
606 if (greh->flags & htons(GRE_SEQ)) {
607 ds_put_format(ds, ",seq=0x%"PRIx32, ntohl(get_16aligned_be32(options)));
608 options++;
609 }
610 ds_put_format(ds, ")");
611 }
612 ds_put_format(ds, ")");
613 }
614
615 static void
616 format_odp_tnl_push_action(struct ds *ds, const struct nlattr *attr)
617 {
618 struct ovs_action_push_tnl *data;
619
620 data = (struct ovs_action_push_tnl *) nl_attr_get(attr);
621
622 ds_put_format(ds, "tnl_push(tnl_port(%"PRIu32"),", data->tnl_port);
623 format_odp_tnl_push_header(ds, data);
624 ds_put_format(ds, ",out_port(%"PRIu32"))", data->out_port);
625 }
626
627 static void
628 format_odp_action(struct ds *ds, const struct nlattr *a)
629 {
630 int expected_len;
631 enum ovs_action_attr type = nl_attr_type(a);
632 const struct ovs_action_push_vlan *vlan;
633 size_t size;
634
635 expected_len = odp_action_len(nl_attr_type(a));
636 if (expected_len != ATTR_LEN_VARIABLE &&
637 nl_attr_get_size(a) != expected_len) {
638 ds_put_format(ds, "bad length %"PRIuSIZE", expected %d for: ",
639 nl_attr_get_size(a), expected_len);
640 format_generic_odp_action(ds, a);
641 return;
642 }
643
644 switch (type) {
645 case OVS_ACTION_ATTR_OUTPUT:
646 ds_put_format(ds, "%"PRIu32, nl_attr_get_u32(a));
647 break;
648 case OVS_ACTION_ATTR_TUNNEL_POP:
649 ds_put_format(ds, "tnl_pop(%"PRIu32")", nl_attr_get_u32(a));
650 break;
651 case OVS_ACTION_ATTR_TUNNEL_PUSH:
652 format_odp_tnl_push_action(ds, a);
653 break;
654 case OVS_ACTION_ATTR_USERSPACE:
655 format_odp_userspace_action(ds, a);
656 break;
657 case OVS_ACTION_ATTR_RECIRC:
658 format_odp_recirc_action(ds, nl_attr_get_u32(a));
659 break;
660 case OVS_ACTION_ATTR_HASH:
661 format_odp_hash_action(ds, nl_attr_get(a));
662 break;
663 case OVS_ACTION_ATTR_SET_MASKED:
664 a = nl_attr_get(a);
665 size = nl_attr_get_size(a) / 2;
666 ds_put_cstr(ds, "set(");
667
668 /* Masked set action not supported for tunnel key, which is bigger. */
669 if (size <= sizeof(struct ovs_key_ipv6)) {
670 struct nlattr attr[1 + DIV_ROUND_UP(sizeof(struct ovs_key_ipv6),
671 sizeof(struct nlattr))];
672 struct nlattr mask[1 + DIV_ROUND_UP(sizeof(struct ovs_key_ipv6),
673 sizeof(struct nlattr))];
674
675 mask->nla_type = attr->nla_type = nl_attr_type(a);
676 mask->nla_len = attr->nla_len = NLA_HDRLEN + size;
677 memcpy(attr + 1, (char *)(a + 1), size);
678 memcpy(mask + 1, (char *)(a + 1) + size, size);
679 format_odp_key_attr(attr, mask, NULL, ds, false);
680 } else {
681 format_odp_key_attr(a, NULL, NULL, ds, false);
682 }
683 ds_put_cstr(ds, ")");
684 break;
685 case OVS_ACTION_ATTR_SET:
686 ds_put_cstr(ds, "set(");
687 format_odp_key_attr(nl_attr_get(a), NULL, NULL, ds, true);
688 ds_put_cstr(ds, ")");
689 break;
690 case OVS_ACTION_ATTR_PUSH_VLAN:
691 vlan = nl_attr_get(a);
692 ds_put_cstr(ds, "push_vlan(");
693 if (vlan->vlan_tpid != htons(ETH_TYPE_VLAN)) {
694 ds_put_format(ds, "tpid=0x%04"PRIx16",", ntohs(vlan->vlan_tpid));
695 }
696 format_vlan_tci(ds, vlan->vlan_tci, OVS_BE16_MAX, false);
697 ds_put_char(ds, ')');
698 break;
699 case OVS_ACTION_ATTR_POP_VLAN:
700 ds_put_cstr(ds, "pop_vlan");
701 break;
702 case OVS_ACTION_ATTR_PUSH_MPLS: {
703 const struct ovs_action_push_mpls *mpls = nl_attr_get(a);
704 ds_put_cstr(ds, "push_mpls(");
705 format_mpls_lse(ds, mpls->mpls_lse);
706 ds_put_format(ds, ",eth_type=0x%"PRIx16")", ntohs(mpls->mpls_ethertype));
707 break;
708 }
709 case OVS_ACTION_ATTR_POP_MPLS: {
710 ovs_be16 ethertype = nl_attr_get_be16(a);
711 ds_put_format(ds, "pop_mpls(eth_type=0x%"PRIx16")", ntohs(ethertype));
712 break;
713 }
714 case OVS_ACTION_ATTR_SAMPLE:
715 format_odp_sample_action(ds, a);
716 break;
717 case OVS_ACTION_ATTR_UNSPEC:
718 case __OVS_ACTION_ATTR_MAX:
719 default:
720 format_generic_odp_action(ds, a);
721 break;
722 }
723 }
724
725 void
726 format_odp_actions(struct ds *ds, const struct nlattr *actions,
727 size_t actions_len)
728 {
729 if (actions_len) {
730 const struct nlattr *a;
731 unsigned int left;
732
733 NL_ATTR_FOR_EACH (a, left, actions, actions_len) {
734 if (a != actions) {
735 ds_put_char(ds, ',');
736 }
737 format_odp_action(ds, a);
738 }
739 if (left) {
740 int i;
741
742 if (left == actions_len) {
743 ds_put_cstr(ds, "<empty>");
744 }
745 ds_put_format(ds, ",***%u leftover bytes*** (", left);
746 for (i = 0; i < left; i++) {
747 ds_put_format(ds, "%02x", ((const uint8_t *) a)[i]);
748 }
749 ds_put_char(ds, ')');
750 }
751 } else {
752 ds_put_cstr(ds, "drop");
753 }
754 }
755
756 /* Separate out parse_odp_userspace_action() function. */
757 static int
758 parse_odp_userspace_action(const char *s, struct ofpbuf *actions)
759 {
760 uint32_t pid;
761 union user_action_cookie cookie;
762 struct ofpbuf buf;
763 odp_port_t tunnel_out_port;
764 int n = -1;
765 void *user_data = NULL;
766 size_t user_data_size = 0;
767
768 if (!ovs_scan(s, "userspace(pid=%"SCNi32"%n", &pid, &n)) {
769 return -EINVAL;
770 }
771
772 {
773 uint32_t output;
774 uint32_t probability;
775 uint32_t collector_set_id;
776 uint32_t obs_domain_id;
777 uint32_t obs_point_id;
778 int vid, pcp;
779 int n1 = -1;
780 if (ovs_scan(&s[n], ",sFlow(vid=%i,"
781 "pcp=%i,output=%"SCNi32")%n",
782 &vid, &pcp, &output, &n1)) {
783 uint16_t tci;
784
785 n += n1;
786 tci = vid | (pcp << VLAN_PCP_SHIFT);
787 if (tci) {
788 tci |= VLAN_CFI;
789 }
790
791 cookie.type = USER_ACTION_COOKIE_SFLOW;
792 cookie.sflow.vlan_tci = htons(tci);
793 cookie.sflow.output = output;
794 user_data = &cookie;
795 user_data_size = sizeof cookie.sflow;
796 } else if (ovs_scan(&s[n], ",slow_path(%n",
797 &n1)) {
798 int res;
799
800 n += n1;
801 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
802 cookie.slow_path.unused = 0;
803 cookie.slow_path.reason = 0;
804
805 res = parse_flags(&s[n], slow_path_reason_to_string,
806 &cookie.slow_path.reason,
807 SLOW_PATH_REASON_MASK, NULL);
808 if (res < 0 || s[n + res] != ')') {
809 return res;
810 }
811 n += res + 1;
812
813 user_data = &cookie;
814 user_data_size = sizeof cookie.slow_path;
815 } else if (ovs_scan(&s[n], ",flow_sample(probability=%"SCNi32","
816 "collector_set_id=%"SCNi32","
817 "obs_domain_id=%"SCNi32","
818 "obs_point_id=%"SCNi32")%n",
819 &probability, &collector_set_id,
820 &obs_domain_id, &obs_point_id, &n1)) {
821 n += n1;
822
823 cookie.type = USER_ACTION_COOKIE_FLOW_SAMPLE;
824 cookie.flow_sample.probability = probability;
825 cookie.flow_sample.collector_set_id = collector_set_id;
826 cookie.flow_sample.obs_domain_id = obs_domain_id;
827 cookie.flow_sample.obs_point_id = obs_point_id;
828 user_data = &cookie;
829 user_data_size = sizeof cookie.flow_sample;
830 } else if (ovs_scan(&s[n], ",ipfix(output_port=%"SCNi32")%n",
831 &output, &n1) ) {
832 n += n1;
833 cookie.type = USER_ACTION_COOKIE_IPFIX;
834 cookie.ipfix.output_odp_port = u32_to_odp(output);
835 user_data = &cookie;
836 user_data_size = sizeof cookie.ipfix;
837 } else if (ovs_scan(&s[n], ",userdata(%n",
838 &n1)) {
839 char *end;
840
841 n += n1;
842 ofpbuf_init(&buf, 16);
843 end = ofpbuf_put_hex(&buf, &s[n], NULL);
844 if (end[0] != ')') {
845 return -EINVAL;
846 }
847 user_data = buf.data;
848 user_data_size = buf.size;
849 n = (end + 1) - s;
850 }
851 }
852
853 {
854 int n1 = -1;
855 if (ovs_scan(&s[n], ",tunnel_out_port=%"SCNi32")%n",
856 &tunnel_out_port, &n1)) {
857 odp_put_userspace_action(pid, user_data, user_data_size, tunnel_out_port, actions);
858 return n + n1;
859 } else if (s[n] == ')') {
860 odp_put_userspace_action(pid, user_data, user_data_size, ODPP_NONE, actions);
861 return n + 1;
862 }
863 }
864
865 return -EINVAL;
866 }
867
868 static int
869 ovs_parse_tnl_push(const char *s, struct ovs_action_push_tnl *data)
870 {
871 struct eth_header *eth;
872 struct ip_header *ip;
873 struct udp_header *udp;
874 struct gre_base_hdr *greh;
875 uint16_t gre_proto, gre_flags, dl_type, udp_src, udp_dst, csum;
876 ovs_be32 sip, dip;
877 uint32_t tnl_type = 0, header_len = 0;
878 void *l3, *l4;
879 int n = 0;
880
881 if (!ovs_scan_len(s, &n, "tnl_push(tnl_port(%"SCNi32"),", &data->tnl_port)) {
882 return -EINVAL;
883 }
884 eth = (struct eth_header *) data->header;
885 l3 = (data->header + sizeof *eth);
886 l4 = ((uint8_t *) l3 + sizeof (struct ip_header));
887 ip = (struct ip_header *) l3;
888 if (!ovs_scan_len(s, &n, "header(size=%"SCNi32",type=%"SCNi32","
889 "eth(dst="ETH_ADDR_SCAN_FMT",",
890 &data->header_len,
891 &data->tnl_type,
892 ETH_ADDR_SCAN_ARGS(eth->eth_dst))) {
893 return -EINVAL;
894 }
895
896 if (!ovs_scan_len(s, &n, "src="ETH_ADDR_SCAN_FMT",",
897 ETH_ADDR_SCAN_ARGS(eth->eth_src))) {
898 return -EINVAL;
899 }
900 if (!ovs_scan_len(s, &n, "dl_type=0x%"SCNx16"),", &dl_type)) {
901 return -EINVAL;
902 }
903 eth->eth_type = htons(dl_type);
904
905 /* IPv4 */
906 if (!ovs_scan_len(s, &n, "ipv4(src="IP_SCAN_FMT",dst="IP_SCAN_FMT",proto=%"SCNi8
907 ",tos=%"SCNi8",ttl=%"SCNi8",frag=0x%"SCNx16"),",
908 IP_SCAN_ARGS(&sip),
909 IP_SCAN_ARGS(&dip),
910 &ip->ip_proto, &ip->ip_tos,
911 &ip->ip_ttl, &ip->ip_frag_off)) {
912 return -EINVAL;
913 }
914 put_16aligned_be32(&ip->ip_src, sip);
915 put_16aligned_be32(&ip->ip_dst, dip);
916
917 /* Tunnel header */
918 udp = (struct udp_header *) l4;
919 greh = (struct gre_base_hdr *) l4;
920 if (ovs_scan_len(s, &n, "udp(src=%"SCNi16",dst=%"SCNi16",csum=0x%"SCNx16"),",
921 &udp_src, &udp_dst, &csum)) {
922 uint32_t vx_flags, vni;
923
924 udp->udp_src = htons(udp_src);
925 udp->udp_dst = htons(udp_dst);
926 udp->udp_len = 0;
927 udp->udp_csum = htons(csum);
928
929 if (ovs_scan_len(s, &n, "vxlan(flags=0x%"SCNx32",vni=0x%"SCNx32"))",
930 &vx_flags, &vni)) {
931 struct vxlanhdr *vxh = (struct vxlanhdr *) (udp + 1);
932
933 put_16aligned_be32(&vxh->vx_flags, htonl(vx_flags));
934 put_16aligned_be32(&vxh->vx_vni, htonl(vni << 8));
935 tnl_type = OVS_VPORT_TYPE_VXLAN;
936 header_len = sizeof *eth + sizeof *ip +
937 sizeof *udp + sizeof *vxh;
938 } else if (ovs_scan_len(s, &n, "geneve(")) {
939 struct genevehdr *gnh = (struct genevehdr *) (udp + 1);
940
941 memset(gnh, 0, sizeof *gnh);
942 if (ovs_scan_len(s, &n, "oam,")) {
943 gnh->oam = 1;
944 }
945 if (!ovs_scan_len(s, &n, "vni=0x%"SCNx32"))", &vni)) {
946 return -EINVAL;
947 }
948 gnh->proto_type = htons(ETH_TYPE_TEB);
949 put_16aligned_be32(&gnh->vni, htonl(vni << 8));
950 tnl_type = OVS_VPORT_TYPE_GENEVE;
951 header_len = sizeof *eth + sizeof *ip +
952 sizeof *udp + sizeof *gnh;
953 } else {
954 return -EINVAL;
955 }
956 } else if (ovs_scan_len(s, &n, "gre((flags=0x%"SCNx16",proto=0x%"SCNx16")",
957 &gre_flags, &gre_proto)){
958
959 tnl_type = OVS_VPORT_TYPE_GRE;
960 greh->flags = htons(gre_flags);
961 greh->protocol = htons(gre_proto);
962 ovs_16aligned_be32 *options = (ovs_16aligned_be32 *) (greh + 1);
963
964 if (greh->flags & htons(GRE_CSUM)) {
965 if (!ovs_scan_len(s, &n, ",csum=0x%"SCNx16, &csum)) {
966 return -EINVAL;
967 }
968
969 memset(options, 0, sizeof *options);
970 *((ovs_be16 *)options) = htons(csum);
971 options++;
972 }
973 if (greh->flags & htons(GRE_KEY)) {
974 uint32_t key;
975
976 if (!ovs_scan_len(s, &n, ",key=0x%"SCNx32, &key)) {
977 return -EINVAL;
978 }
979
980 put_16aligned_be32(options, htonl(key));
981 options++;
982 }
983 if (greh->flags & htons(GRE_SEQ)) {
984 uint32_t seq;
985
986 if (!ovs_scan_len(s, &n, ",seq=0x%"SCNx32, &seq)) {
987 return -EINVAL;
988 }
989 put_16aligned_be32(options, htonl(seq));
990 options++;
991 }
992
993 if (!ovs_scan_len(s, &n, "))")) {
994 return -EINVAL;
995 }
996
997 header_len = sizeof *eth + sizeof *ip +
998 ((uint8_t *) options - (uint8_t *) greh);
999 } else {
1000 return -EINVAL;
1001 }
1002
1003 /* check tunnel meta data. */
1004 if (data->tnl_type != tnl_type) {
1005 return -EINVAL;
1006 }
1007 if (data->header_len != header_len) {
1008 return -EINVAL;
1009 }
1010
1011 /* Out port */
1012 if (!ovs_scan_len(s, &n, ",out_port(%"SCNi32"))", &data->out_port)) {
1013 return -EINVAL;
1014 }
1015
1016 return n;
1017 }
1018
1019 static int
1020 parse_odp_action(const char *s, const struct simap *port_names,
1021 struct ofpbuf *actions)
1022 {
1023 {
1024 uint32_t port;
1025 int n;
1026
1027 if (ovs_scan(s, "%"SCNi32"%n", &port, &n)) {
1028 nl_msg_put_u32(actions, OVS_ACTION_ATTR_OUTPUT, port);
1029 return n;
1030 }
1031 }
1032
1033 if (port_names) {
1034 int len = strcspn(s, delimiters);
1035 struct simap_node *node;
1036
1037 node = simap_find_len(port_names, s, len);
1038 if (node) {
1039 nl_msg_put_u32(actions, OVS_ACTION_ATTR_OUTPUT, node->data);
1040 return len;
1041 }
1042 }
1043
1044 {
1045 uint32_t recirc_id;
1046 int n = -1;
1047
1048 if (ovs_scan(s, "recirc(%"PRIu32")%n", &recirc_id, &n)) {
1049 nl_msg_put_u32(actions, OVS_ACTION_ATTR_RECIRC, recirc_id);
1050 return n;
1051 }
1052 }
1053
1054 if (!strncmp(s, "userspace(", 10)) {
1055 return parse_odp_userspace_action(s, actions);
1056 }
1057
1058 if (!strncmp(s, "set(", 4)) {
1059 size_t start_ofs;
1060 int retval;
1061 struct nlattr mask[128 / sizeof(struct nlattr)];
1062 struct ofpbuf maskbuf;
1063 struct nlattr *nested, *key;
1064 size_t size;
1065
1066 /* 'mask' is big enough to hold any key. */
1067 ofpbuf_use_stack(&maskbuf, mask, sizeof mask);
1068
1069 start_ofs = nl_msg_start_nested(actions, OVS_ACTION_ATTR_SET);
1070 retval = parse_odp_key_mask_attr(s + 4, port_names, actions, &maskbuf);
1071 if (retval < 0) {
1072 return retval;
1073 }
1074 if (s[retval + 4] != ')') {
1075 return -EINVAL;
1076 }
1077
1078 nested = ofpbuf_at_assert(actions, start_ofs, sizeof *nested);
1079 key = nested + 1;
1080
1081 size = nl_attr_get_size(mask);
1082 if (size == nl_attr_get_size(key)) {
1083 /* Change to masked set action if not fully masked. */
1084 if (!is_all_ones(mask + 1, size)) {
1085 key->nla_len += size;
1086 ofpbuf_put(actions, mask + 1, size);
1087 /* 'actions' may have been reallocated by ofpbuf_put(). */
1088 nested = ofpbuf_at_assert(actions, start_ofs, sizeof *nested);
1089 nested->nla_type = OVS_ACTION_ATTR_SET_MASKED;
1090 }
1091 }
1092
1093 nl_msg_end_nested(actions, start_ofs);
1094 return retval + 5;
1095 }
1096
1097 {
1098 struct ovs_action_push_vlan push;
1099 int tpid = ETH_TYPE_VLAN;
1100 int vid, pcp;
1101 int cfi = 1;
1102 int n = -1;
1103
1104 if (ovs_scan(s, "push_vlan(vid=%i,pcp=%i)%n", &vid, &pcp, &n)
1105 || ovs_scan(s, "push_vlan(vid=%i,pcp=%i,cfi=%i)%n",
1106 &vid, &pcp, &cfi, &n)
1107 || ovs_scan(s, "push_vlan(tpid=%i,vid=%i,pcp=%i)%n",
1108 &tpid, &vid, &pcp, &n)
1109 || ovs_scan(s, "push_vlan(tpid=%i,vid=%i,pcp=%i,cfi=%i)%n",
1110 &tpid, &vid, &pcp, &cfi, &n)) {
1111 push.vlan_tpid = htons(tpid);
1112 push.vlan_tci = htons((vid << VLAN_VID_SHIFT)
1113 | (pcp << VLAN_PCP_SHIFT)
1114 | (cfi ? VLAN_CFI : 0));
1115 nl_msg_put_unspec(actions, OVS_ACTION_ATTR_PUSH_VLAN,
1116 &push, sizeof push);
1117
1118 return n;
1119 }
1120 }
1121
1122 if (!strncmp(s, "pop_vlan", 8)) {
1123 nl_msg_put_flag(actions, OVS_ACTION_ATTR_POP_VLAN);
1124 return 8;
1125 }
1126
1127 {
1128 double percentage;
1129 int n = -1;
1130
1131 if (ovs_scan(s, "sample(sample=%lf%%,actions(%n", &percentage, &n)
1132 && percentage >= 0. && percentage <= 100.0) {
1133 size_t sample_ofs, actions_ofs;
1134 double probability;
1135
1136 probability = floor(UINT32_MAX * (percentage / 100.0) + .5);
1137 sample_ofs = nl_msg_start_nested(actions, OVS_ACTION_ATTR_SAMPLE);
1138 nl_msg_put_u32(actions, OVS_SAMPLE_ATTR_PROBABILITY,
1139 (probability <= 0 ? 0
1140 : probability >= UINT32_MAX ? UINT32_MAX
1141 : probability));
1142
1143 actions_ofs = nl_msg_start_nested(actions,
1144 OVS_SAMPLE_ATTR_ACTIONS);
1145 for (;;) {
1146 int retval;
1147
1148 n += strspn(s + n, delimiters);
1149 if (s[n] == ')') {
1150 break;
1151 }
1152
1153 retval = parse_odp_action(s + n, port_names, actions);
1154 if (retval < 0) {
1155 return retval;
1156 }
1157 n += retval;
1158 }
1159 nl_msg_end_nested(actions, actions_ofs);
1160 nl_msg_end_nested(actions, sample_ofs);
1161
1162 return s[n + 1] == ')' ? n + 2 : -EINVAL;
1163 }
1164 }
1165
1166 {
1167 uint32_t port;
1168 int n;
1169
1170 if (ovs_scan(s, "tnl_pop(%"SCNi32")%n", &port, &n)) {
1171 nl_msg_put_u32(actions, OVS_ACTION_ATTR_TUNNEL_POP, port);
1172 return n;
1173 }
1174 }
1175
1176 {
1177 struct ovs_action_push_tnl data;
1178 int n;
1179
1180 n = ovs_parse_tnl_push(s, &data);
1181 if (n > 0) {
1182 odp_put_tnl_push_action(actions, &data);
1183 return n;
1184 } else if (n < 0) {
1185 return n;
1186 }
1187 }
1188 return -EINVAL;
1189 }
1190
1191 /* Parses the string representation of datapath actions, in the format output
1192 * by format_odp_action(). Returns 0 if successful, otherwise a positive errno
1193 * value. On success, the ODP actions are appended to 'actions' as a series of
1194 * Netlink attributes. On failure, no data is appended to 'actions'. Either
1195 * way, 'actions''s data might be reallocated. */
1196 int
1197 odp_actions_from_string(const char *s, const struct simap *port_names,
1198 struct ofpbuf *actions)
1199 {
1200 size_t old_size;
1201
1202 if (!strcasecmp(s, "drop")) {
1203 return 0;
1204 }
1205
1206 old_size = actions->size;
1207 for (;;) {
1208 int retval;
1209
1210 s += strspn(s, delimiters);
1211 if (!*s) {
1212 return 0;
1213 }
1214
1215 retval = parse_odp_action(s, port_names, actions);
1216 if (retval < 0 || !strchr(delimiters, s[retval])) {
1217 actions->size = old_size;
1218 return -retval;
1219 }
1220 s += retval;
1221 }
1222
1223 return 0;
1224 }
1225 \f
1226 static const struct attr_len_tbl ovs_vxlan_ext_attr_lens[OVS_VXLAN_EXT_MAX + 1] = {
1227 [OVS_VXLAN_EXT_GBP] = { .len = 4 },
1228 };
1229
1230 static const struct attr_len_tbl ovs_tun_key_attr_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1231 [OVS_TUNNEL_KEY_ATTR_ID] = { .len = 8 },
1232 [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = { .len = 4 },
1233 [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = { .len = 4 },
1234 [OVS_TUNNEL_KEY_ATTR_TOS] = { .len = 1 },
1235 [OVS_TUNNEL_KEY_ATTR_TTL] = { .len = 1 },
1236 [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
1237 [OVS_TUNNEL_KEY_ATTR_CSUM] = { .len = 0 },
1238 [OVS_TUNNEL_KEY_ATTR_TP_SRC] = { .len = 2 },
1239 [OVS_TUNNEL_KEY_ATTR_TP_DST] = { .len = 2 },
1240 [OVS_TUNNEL_KEY_ATTR_OAM] = { .len = 0 },
1241 [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS] = { .len = ATTR_LEN_VARIABLE },
1242 [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS] = { .len = ATTR_LEN_NESTED,
1243 .next = ovs_vxlan_ext_attr_lens ,
1244 .next_max = OVS_VXLAN_EXT_MAX},
1245 };
1246
1247 static const struct attr_len_tbl ovs_flow_key_attr_lens[OVS_KEY_ATTR_MAX + 1] = {
1248 [OVS_KEY_ATTR_ENCAP] = { .len = ATTR_LEN_NESTED },
1249 [OVS_KEY_ATTR_PRIORITY] = { .len = 4 },
1250 [OVS_KEY_ATTR_SKB_MARK] = { .len = 4 },
1251 [OVS_KEY_ATTR_DP_HASH] = { .len = 4 },
1252 [OVS_KEY_ATTR_RECIRC_ID] = { .len = 4 },
1253 [OVS_KEY_ATTR_TUNNEL] = { .len = ATTR_LEN_NESTED,
1254 .next = ovs_tun_key_attr_lens,
1255 .next_max = OVS_TUNNEL_KEY_ATTR_MAX },
1256 [OVS_KEY_ATTR_IN_PORT] = { .len = 4 },
1257 [OVS_KEY_ATTR_ETHERNET] = { .len = sizeof(struct ovs_key_ethernet) },
1258 [OVS_KEY_ATTR_VLAN] = { .len = 2 },
1259 [OVS_KEY_ATTR_ETHERTYPE] = { .len = 2 },
1260 [OVS_KEY_ATTR_MPLS] = { .len = ATTR_LEN_VARIABLE },
1261 [OVS_KEY_ATTR_IPV4] = { .len = sizeof(struct ovs_key_ipv4) },
1262 [OVS_KEY_ATTR_IPV6] = { .len = sizeof(struct ovs_key_ipv6) },
1263 [OVS_KEY_ATTR_TCP] = { .len = sizeof(struct ovs_key_tcp) },
1264 [OVS_KEY_ATTR_TCP_FLAGS] = { .len = 2 },
1265 [OVS_KEY_ATTR_UDP] = { .len = sizeof(struct ovs_key_udp) },
1266 [OVS_KEY_ATTR_SCTP] = { .len = sizeof(struct ovs_key_sctp) },
1267 [OVS_KEY_ATTR_ICMP] = { .len = sizeof(struct ovs_key_icmp) },
1268 [OVS_KEY_ATTR_ICMPV6] = { .len = sizeof(struct ovs_key_icmpv6) },
1269 [OVS_KEY_ATTR_ARP] = { .len = sizeof(struct ovs_key_arp) },
1270 [OVS_KEY_ATTR_ND] = { .len = sizeof(struct ovs_key_nd) },
1271 };
1272
1273 /* Returns the correct length of the payload for a flow key attribute of the
1274 * specified 'type', ATTR_LEN_INVALID if 'type' is unknown, ATTR_LEN_VARIABLE
1275 * if the attribute's payload is variable length, or ATTR_LEN_NESTED if the
1276 * payload is a nested type. */
1277 static int
1278 odp_key_attr_len(const struct attr_len_tbl tbl[], int max_len, uint16_t type)
1279 {
1280 if (type > max_len) {
1281 return ATTR_LEN_INVALID;
1282 }
1283
1284 return tbl[type].len;
1285 }
1286
1287 static void
1288 format_generic_odp_key(const struct nlattr *a, struct ds *ds)
1289 {
1290 size_t len = nl_attr_get_size(a);
1291 if (len) {
1292 const uint8_t *unspec;
1293 unsigned int i;
1294
1295 unspec = nl_attr_get(a);
1296 for (i = 0; i < len; i++) {
1297 if (i) {
1298 ds_put_char(ds, ' ');
1299 }
1300 ds_put_format(ds, "%02x", unspec[i]);
1301 }
1302 }
1303 }
1304
1305 static const char *
1306 ovs_frag_type_to_string(enum ovs_frag_type type)
1307 {
1308 switch (type) {
1309 case OVS_FRAG_TYPE_NONE:
1310 return "no";
1311 case OVS_FRAG_TYPE_FIRST:
1312 return "first";
1313 case OVS_FRAG_TYPE_LATER:
1314 return "later";
1315 case __OVS_FRAG_TYPE_MAX:
1316 default:
1317 return "<error>";
1318 }
1319 }
1320
1321 #define GENEVE_OPT(class, type) ((OVS_FORCE uint32_t)(class) << 8 | (type))
1322 static int
1323 parse_geneve_opts(const struct nlattr *attr)
1324 {
1325 int opts_len = nl_attr_get_size(attr);
1326 const struct geneve_opt *opt = nl_attr_get(attr);
1327
1328 while (opts_len > 0) {
1329 int len;
1330
1331 if (opts_len < sizeof(*opt)) {
1332 return -EINVAL;
1333 }
1334
1335 len = sizeof(*opt) + opt->length * 4;
1336 if (len > opts_len) {
1337 return -EINVAL;
1338 }
1339
1340 switch (GENEVE_OPT(opt->opt_class, opt->type)) {
1341 default:
1342 if (opt->type & GENEVE_CRIT_OPT_TYPE) {
1343 return -EINVAL;
1344 }
1345 };
1346
1347 opt = opt + len / sizeof(*opt);
1348 opts_len -= len;
1349 };
1350
1351 return 0;
1352 }
1353
1354 enum odp_key_fitness
1355 odp_tun_key_from_attr(const struct nlattr *attr, struct flow_tnl *tun)
1356 {
1357 unsigned int left;
1358 const struct nlattr *a;
1359 bool ttl = false;
1360 bool unknown = false;
1361
1362 NL_NESTED_FOR_EACH(a, left, attr) {
1363 uint16_t type = nl_attr_type(a);
1364 size_t len = nl_attr_get_size(a);
1365 int expected_len = odp_key_attr_len(ovs_tun_key_attr_lens,
1366 OVS_TUNNEL_ATTR_MAX, type);
1367
1368 if (len != expected_len && expected_len >= 0) {
1369 return ODP_FIT_ERROR;
1370 }
1371
1372 switch (type) {
1373 case OVS_TUNNEL_KEY_ATTR_ID:
1374 tun->tun_id = nl_attr_get_be64(a);
1375 tun->flags |= FLOW_TNL_F_KEY;
1376 break;
1377 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1378 tun->ip_src = nl_attr_get_be32(a);
1379 break;
1380 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1381 tun->ip_dst = nl_attr_get_be32(a);
1382 break;
1383 case OVS_TUNNEL_KEY_ATTR_TOS:
1384 tun->ip_tos = nl_attr_get_u8(a);
1385 break;
1386 case OVS_TUNNEL_KEY_ATTR_TTL:
1387 tun->ip_ttl = nl_attr_get_u8(a);
1388 ttl = true;
1389 break;
1390 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1391 tun->flags |= FLOW_TNL_F_DONT_FRAGMENT;
1392 break;
1393 case OVS_TUNNEL_KEY_ATTR_CSUM:
1394 tun->flags |= FLOW_TNL_F_CSUM;
1395 break;
1396 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
1397 tun->tp_src = nl_attr_get_be16(a);
1398 break;
1399 case OVS_TUNNEL_KEY_ATTR_TP_DST:
1400 tun->tp_dst = nl_attr_get_be16(a);
1401 break;
1402 case OVS_TUNNEL_KEY_ATTR_OAM:
1403 tun->flags |= FLOW_TNL_F_OAM;
1404 break;
1405 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS: {
1406 static const struct nl_policy vxlan_opts_policy[] = {
1407 [OVS_VXLAN_EXT_GBP] = { .type = NL_A_U32 },
1408 };
1409 struct nlattr *ext[ARRAY_SIZE(vxlan_opts_policy)];
1410
1411 if (!nl_parse_nested(a, vxlan_opts_policy, ext, ARRAY_SIZE(ext))) {
1412 return ODP_FIT_ERROR;
1413 }
1414
1415 if (ext[OVS_VXLAN_EXT_GBP]) {
1416 uint32_t gbp = nl_attr_get_u32(ext[OVS_VXLAN_EXT_GBP]);
1417
1418 tun->gbp_id = htons(gbp & 0xFFFF);
1419 tun->gbp_flags = (gbp >> 16) & 0xFF;
1420 }
1421
1422 break;
1423 }
1424 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS: {
1425 if (parse_geneve_opts(a)) {
1426 return ODP_FIT_ERROR;
1427 }
1428 /* It is necessary to reproduce options exactly (including order)
1429 * so it's easiest to just echo them back. */
1430 unknown = true;
1431 break;
1432 }
1433 default:
1434 /* Allow this to show up as unexpected, if there are unknown
1435 * tunnel attribute, eventually resulting in ODP_FIT_TOO_MUCH. */
1436 unknown = true;
1437 break;
1438 }
1439 }
1440
1441 if (!ttl) {
1442 return ODP_FIT_ERROR;
1443 }
1444 if (unknown) {
1445 return ODP_FIT_TOO_MUCH;
1446 }
1447 return ODP_FIT_PERFECT;
1448 }
1449
1450 static void
1451 tun_key_to_attr(struct ofpbuf *a, const struct flow_tnl *tun_key)
1452 {
1453 size_t tun_key_ofs;
1454
1455 tun_key_ofs = nl_msg_start_nested(a, OVS_KEY_ATTR_TUNNEL);
1456
1457 /* tun_id != 0 without FLOW_TNL_F_KEY is valid if tun_key is a mask. */
1458 if (tun_key->tun_id || tun_key->flags & FLOW_TNL_F_KEY) {
1459 nl_msg_put_be64(a, OVS_TUNNEL_KEY_ATTR_ID, tun_key->tun_id);
1460 }
1461 if (tun_key->ip_src) {
1462 nl_msg_put_be32(a, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, tun_key->ip_src);
1463 }
1464 if (tun_key->ip_dst) {
1465 nl_msg_put_be32(a, OVS_TUNNEL_KEY_ATTR_IPV4_DST, tun_key->ip_dst);
1466 }
1467 if (tun_key->ip_tos) {
1468 nl_msg_put_u8(a, OVS_TUNNEL_KEY_ATTR_TOS, tun_key->ip_tos);
1469 }
1470 nl_msg_put_u8(a, OVS_TUNNEL_KEY_ATTR_TTL, tun_key->ip_ttl);
1471 if (tun_key->flags & FLOW_TNL_F_DONT_FRAGMENT) {
1472 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT);
1473 }
1474 if (tun_key->flags & FLOW_TNL_F_CSUM) {
1475 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_CSUM);
1476 }
1477 if (tun_key->tp_src) {
1478 nl_msg_put_be16(a, OVS_TUNNEL_KEY_ATTR_TP_SRC, tun_key->tp_src);
1479 }
1480 if (tun_key->tp_dst) {
1481 nl_msg_put_be16(a, OVS_TUNNEL_KEY_ATTR_TP_DST, tun_key->tp_dst);
1482 }
1483 if (tun_key->flags & FLOW_TNL_F_OAM) {
1484 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_OAM);
1485 }
1486 if (tun_key->gbp_flags || tun_key->gbp_id) {
1487 size_t vxlan_opts_ofs;
1488
1489 vxlan_opts_ofs = nl_msg_start_nested(a, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
1490 nl_msg_put_u32(a, OVS_VXLAN_EXT_GBP,
1491 (tun_key->gbp_flags << 16) | ntohs(tun_key->gbp_id));
1492 nl_msg_end_nested(a, vxlan_opts_ofs);
1493 }
1494
1495 nl_msg_end_nested(a, tun_key_ofs);
1496 }
1497
1498 static bool
1499 odp_mask_attr_is_wildcard(const struct nlattr *ma)
1500 {
1501 return is_all_zeros(nl_attr_get(ma), nl_attr_get_size(ma));
1502 }
1503
1504 static bool
1505 odp_mask_is_exact(enum ovs_key_attr attr, const void *mask, size_t size)
1506 {
1507 if (attr == OVS_KEY_ATTR_TCP_FLAGS) {
1508 return TCP_FLAGS(*(ovs_be16 *)mask) == TCP_FLAGS(OVS_BE16_MAX);
1509 }
1510 if (attr == OVS_KEY_ATTR_IPV6) {
1511 const struct ovs_key_ipv6 *ipv6_mask = mask;
1512
1513 return
1514 ((ipv6_mask->ipv6_label & htonl(IPV6_LABEL_MASK))
1515 == htonl(IPV6_LABEL_MASK))
1516 && ipv6_mask->ipv6_proto == UINT8_MAX
1517 && ipv6_mask->ipv6_tclass == UINT8_MAX
1518 && ipv6_mask->ipv6_hlimit == UINT8_MAX
1519 && ipv6_mask->ipv6_frag == UINT8_MAX
1520 && ipv6_mask_is_exact((const struct in6_addr *)ipv6_mask->ipv6_src)
1521 && ipv6_mask_is_exact((const struct in6_addr *)ipv6_mask->ipv6_dst);
1522 }
1523 if (attr == OVS_KEY_ATTR_TUNNEL) {
1524 const struct flow_tnl *tun_mask = mask;
1525
1526 return tun_mask->flags == FLOW_TNL_F_MASK
1527 && tun_mask->tun_id == OVS_BE64_MAX
1528 && tun_mask->ip_src == OVS_BE32_MAX
1529 && tun_mask->ip_dst == OVS_BE32_MAX
1530 && tun_mask->ip_tos == UINT8_MAX
1531 && tun_mask->ip_ttl == UINT8_MAX
1532 && tun_mask->tp_src == OVS_BE16_MAX
1533 && tun_mask->tp_dst == OVS_BE16_MAX
1534 && tun_mask->gbp_id == OVS_BE16_MAX
1535 && tun_mask->gbp_flags == UINT8_MAX;
1536 }
1537
1538 if (attr == OVS_KEY_ATTR_ARP) {
1539 /* ARP key has padding, ignore it. */
1540 BUILD_ASSERT_DECL(sizeof(struct ovs_key_arp) == 24);
1541 BUILD_ASSERT_DECL(offsetof(struct ovs_key_arp, arp_tha) == 10 + 6);
1542 size = offsetof(struct ovs_key_arp, arp_tha) + ETH_ADDR_LEN;
1543 ovs_assert(((uint16_t *)mask)[size/2] == 0);
1544 }
1545
1546 return is_all_ones(mask, size);
1547 }
1548
1549 static bool
1550 odp_mask_attr_is_exact(const struct nlattr *ma)
1551 {
1552 struct flow_tnl tun_mask;
1553 enum ovs_key_attr attr = nl_attr_type(ma);
1554 const void *mask;
1555 size_t size;
1556
1557 if (attr == OVS_KEY_ATTR_TUNNEL) {
1558 memset(&tun_mask, 0, sizeof tun_mask);
1559 odp_tun_key_from_attr(ma, &tun_mask);
1560 mask = &tun_mask;
1561 size = sizeof tun_mask;
1562 } else {
1563 mask = nl_attr_get(ma);
1564 size = nl_attr_get_size(ma);
1565 }
1566
1567 return odp_mask_is_exact(attr, mask, size);
1568 }
1569
1570 void
1571 odp_portno_names_set(struct hmap *portno_names, odp_port_t port_no,
1572 char *port_name)
1573 {
1574 struct odp_portno_names *odp_portno_names;
1575
1576 odp_portno_names = xmalloc(sizeof *odp_portno_names);
1577 odp_portno_names->port_no = port_no;
1578 odp_portno_names->name = xstrdup(port_name);
1579 hmap_insert(portno_names, &odp_portno_names->hmap_node,
1580 hash_odp_port(port_no));
1581 }
1582
1583 static char *
1584 odp_portno_names_get(const struct hmap *portno_names, odp_port_t port_no)
1585 {
1586 struct odp_portno_names *odp_portno_names;
1587
1588 HMAP_FOR_EACH_IN_BUCKET (odp_portno_names, hmap_node,
1589 hash_odp_port(port_no), portno_names) {
1590 if (odp_portno_names->port_no == port_no) {
1591 return odp_portno_names->name;
1592 }
1593 }
1594 return NULL;
1595 }
1596
1597 void
1598 odp_portno_names_destroy(struct hmap *portno_names)
1599 {
1600 struct odp_portno_names *odp_portno_names, *odp_portno_names_next;
1601 HMAP_FOR_EACH_SAFE (odp_portno_names, odp_portno_names_next,
1602 hmap_node, portno_names) {
1603 hmap_remove(portno_names, &odp_portno_names->hmap_node);
1604 free(odp_portno_names->name);
1605 free(odp_portno_names);
1606 }
1607 }
1608
1609 /* Format helpers. */
1610
1611 static void
1612 format_eth(struct ds *ds, const char *name, const uint8_t key[ETH_ADDR_LEN],
1613 const uint8_t (*mask)[ETH_ADDR_LEN], bool verbose)
1614 {
1615 bool mask_empty = mask && eth_addr_is_zero(*mask);
1616
1617 if (verbose || !mask_empty) {
1618 bool mask_full = !mask || eth_mask_is_exact(*mask);
1619
1620 if (mask_full) {
1621 ds_put_format(ds, "%s="ETH_ADDR_FMT",", name, ETH_ADDR_ARGS(key));
1622 } else {
1623 ds_put_format(ds, "%s=", name);
1624 eth_format_masked(key, *mask, ds);
1625 ds_put_char(ds, ',');
1626 }
1627 }
1628 }
1629
1630 static void
1631 format_be64(struct ds *ds, const char *name, ovs_be64 key,
1632 const ovs_be64 *mask, bool verbose)
1633 {
1634 bool mask_empty = mask && !*mask;
1635
1636 if (verbose || !mask_empty) {
1637 bool mask_full = !mask || *mask == OVS_BE64_MAX;
1638
1639 ds_put_format(ds, "%s=0x%"PRIx64, name, ntohll(key));
1640 if (!mask_full) { /* Partially masked. */
1641 ds_put_format(ds, "/%#"PRIx64, ntohll(*mask));
1642 }
1643 ds_put_char(ds, ',');
1644 }
1645 }
1646
1647 static void
1648 format_ipv4(struct ds *ds, const char *name, ovs_be32 key,
1649 const ovs_be32 *mask, bool verbose)
1650 {
1651 bool mask_empty = mask && !*mask;
1652
1653 if (verbose || !mask_empty) {
1654 bool mask_full = !mask || *mask == OVS_BE32_MAX;
1655
1656 ds_put_format(ds, "%s="IP_FMT, name, IP_ARGS(key));
1657 if (!mask_full) { /* Partially masked. */
1658 ds_put_format(ds, "/"IP_FMT, IP_ARGS(*mask));
1659 }
1660 ds_put_char(ds, ',');
1661 }
1662 }
1663
1664 static void
1665 format_ipv6(struct ds *ds, const char *name, const ovs_be32 key_[4],
1666 const ovs_be32 (*mask_)[4], bool verbose)
1667 {
1668 char buf[INET6_ADDRSTRLEN];
1669 const struct in6_addr *key = (const struct in6_addr *)key_;
1670 const struct in6_addr *mask = mask_ ? (const struct in6_addr *)*mask_
1671 : NULL;
1672 bool mask_empty = mask && ipv6_mask_is_any(mask);
1673
1674 if (verbose || !mask_empty) {
1675 bool mask_full = !mask || ipv6_mask_is_exact(mask);
1676
1677 inet_ntop(AF_INET6, key, buf, sizeof buf);
1678 ds_put_format(ds, "%s=%s", name, buf);
1679 if (!mask_full) { /* Partially masked. */
1680 inet_ntop(AF_INET6, mask, buf, sizeof buf);
1681 ds_put_format(ds, "/%s", buf);
1682 }
1683 ds_put_char(ds, ',');
1684 }
1685 }
1686
1687 static void
1688 format_ipv6_label(struct ds *ds, const char *name, ovs_be32 key,
1689 const ovs_be32 *mask, bool verbose)
1690 {
1691 bool mask_empty = mask && !*mask;
1692
1693 if (verbose || !mask_empty) {
1694 bool mask_full = !mask
1695 || (*mask & htonl(IPV6_LABEL_MASK)) == htonl(IPV6_LABEL_MASK);
1696
1697 ds_put_format(ds, "%s=%#"PRIx32, name, ntohl(key));
1698 if (!mask_full) { /* Partially masked. */
1699 ds_put_format(ds, "/%#"PRIx32, ntohl(*mask));
1700 }
1701 ds_put_char(ds, ',');
1702 }
1703 }
1704
1705 static void
1706 format_u8x(struct ds *ds, const char *name, uint8_t key,
1707 const uint8_t *mask, bool verbose)
1708 {
1709 bool mask_empty = mask && !*mask;
1710
1711 if (verbose || !mask_empty) {
1712 bool mask_full = !mask || *mask == UINT8_MAX;
1713
1714 ds_put_format(ds, "%s=%#"PRIx8, name, key);
1715 if (!mask_full) { /* Partially masked. */
1716 ds_put_format(ds, "/%#"PRIx8, *mask);
1717 }
1718 ds_put_char(ds, ',');
1719 }
1720 }
1721
1722 static void
1723 format_u8u(struct ds *ds, const char *name, uint8_t key,
1724 const uint8_t *mask, bool verbose)
1725 {
1726 bool mask_empty = mask && !*mask;
1727
1728 if (verbose || !mask_empty) {
1729 bool mask_full = !mask || *mask == UINT8_MAX;
1730
1731 ds_put_format(ds, "%s=%"PRIu8, name, key);
1732 if (!mask_full) { /* Partially masked. */
1733 ds_put_format(ds, "/%#"PRIx8, *mask);
1734 }
1735 ds_put_char(ds, ',');
1736 }
1737 }
1738
1739 static void
1740 format_be16(struct ds *ds, const char *name, ovs_be16 key,
1741 const ovs_be16 *mask, bool verbose)
1742 {
1743 bool mask_empty = mask && !*mask;
1744
1745 if (verbose || !mask_empty) {
1746 bool mask_full = !mask || *mask == OVS_BE16_MAX;
1747
1748 ds_put_format(ds, "%s=%"PRIu16, name, ntohs(key));
1749 if (!mask_full) { /* Partially masked. */
1750 ds_put_format(ds, "/%#"PRIx16, ntohs(*mask));
1751 }
1752 ds_put_char(ds, ',');
1753 }
1754 }
1755
1756 static void
1757 format_be16x(struct ds *ds, const char *name, ovs_be16 key,
1758 const ovs_be16 *mask, bool verbose)
1759 {
1760 bool mask_empty = mask && !*mask;
1761
1762 if (verbose || !mask_empty) {
1763 bool mask_full = !mask || *mask == OVS_BE16_MAX;
1764
1765 ds_put_format(ds, "%s=%#"PRIx16, name, ntohs(key));
1766 if (!mask_full) { /* Partially masked. */
1767 ds_put_format(ds, "/%#"PRIx16, ntohs(*mask));
1768 }
1769 ds_put_char(ds, ',');
1770 }
1771 }
1772
1773 static void
1774 format_tun_flags(struct ds *ds, const char *name, uint16_t key,
1775 const uint16_t *mask, bool verbose)
1776 {
1777 bool mask_empty = mask && !*mask;
1778
1779 if (verbose || !mask_empty) {
1780 bool mask_full = !mask || (*mask & FLOW_TNL_F_MASK) == FLOW_TNL_F_MASK;
1781
1782 ds_put_cstr(ds, name);
1783 ds_put_char(ds, '(');
1784 if (!mask_full) { /* Partially masked. */
1785 format_flags_masked(ds, NULL, flow_tun_flag_to_string, key, *mask);
1786 } else { /* Fully masked. */
1787 format_flags(ds, flow_tun_flag_to_string, key, ',');
1788 }
1789 ds_put_cstr(ds, "),");
1790 }
1791 }
1792
1793 static bool
1794 check_attr_len(struct ds *ds, const struct nlattr *a, const struct nlattr *ma,
1795 const struct attr_len_tbl tbl[], int max_len, bool need_key)
1796 {
1797 int expected_len;
1798
1799 expected_len = odp_key_attr_len(tbl, max_len, nl_attr_type(a));
1800 if (expected_len != ATTR_LEN_VARIABLE &&
1801 expected_len != ATTR_LEN_NESTED) {
1802
1803 bool bad_key_len = nl_attr_get_size(a) != expected_len;
1804 bool bad_mask_len = ma && nl_attr_get_size(ma) != expected_len;
1805
1806 if (bad_key_len || bad_mask_len) {
1807 if (need_key) {
1808 ds_put_format(ds, "key%u", nl_attr_type(a));
1809 }
1810 if (bad_key_len) {
1811 ds_put_format(ds, "(bad key length %"PRIuSIZE", expected %d)(",
1812 nl_attr_get_size(a), expected_len);
1813 }
1814 format_generic_odp_key(a, ds);
1815 if (ma) {
1816 ds_put_char(ds, '/');
1817 if (bad_mask_len) {
1818 ds_put_format(ds, "(bad mask length %"PRIuSIZE", expected %d)(",
1819 nl_attr_get_size(ma), expected_len);
1820 }
1821 format_generic_odp_key(ma, ds);
1822 }
1823 ds_put_char(ds, ')');
1824 return false;
1825 }
1826 }
1827
1828 return true;
1829 }
1830
1831 static void
1832 format_unknown_key(struct ds *ds, const struct nlattr *a,
1833 const struct nlattr *ma)
1834 {
1835 ds_put_format(ds, "key%u(", nl_attr_type(a));
1836 format_generic_odp_key(a, ds);
1837 if (ma && !odp_mask_attr_is_exact(ma)) {
1838 ds_put_char(ds, '/');
1839 format_generic_odp_key(ma, ds);
1840 }
1841 ds_put_cstr(ds, "),");
1842 }
1843
1844 static void
1845 format_odp_tun_vxlan_opt(const struct nlattr *attr,
1846 const struct nlattr *mask_attr, struct ds *ds,
1847 bool verbose)
1848 {
1849 unsigned int left;
1850 const struct nlattr *a;
1851 struct ofpbuf ofp;
1852
1853 ofpbuf_init(&ofp, 100);
1854 NL_NESTED_FOR_EACH(a, left, attr) {
1855 uint16_t type = nl_attr_type(a);
1856 const struct nlattr *ma = NULL;
1857
1858 if (mask_attr) {
1859 ma = nl_attr_find__(nl_attr_get(mask_attr),
1860 nl_attr_get_size(mask_attr), type);
1861 if (!ma) {
1862 ma = generate_all_wildcard_mask(ovs_vxlan_ext_attr_lens,
1863 OVS_VXLAN_EXT_MAX,
1864 &ofp, a);
1865 }
1866 }
1867
1868 if (!check_attr_len(ds, a, ma, ovs_vxlan_ext_attr_lens,
1869 OVS_VXLAN_EXT_MAX, true)) {
1870 continue;
1871 }
1872
1873 switch (type) {
1874 case OVS_VXLAN_EXT_GBP: {
1875 uint32_t key = nl_attr_get_u32(a);
1876 ovs_be16 id, id_mask;
1877 uint8_t flags, flags_mask;
1878
1879 id = htons(key & 0xFFFF);
1880 flags = (key >> 16) & 0xFF;
1881 if (ma) {
1882 uint32_t mask = nl_attr_get_u32(ma);
1883 id_mask = htons(mask & 0xFFFF);
1884 flags_mask = (mask >> 16) & 0xFF;
1885 }
1886
1887 ds_put_cstr(ds, "gbp(");
1888 format_be16(ds, "id", id, ma ? &id_mask : NULL, verbose);
1889 format_u8x(ds, "flags", flags, ma ? &flags_mask : NULL, verbose);
1890 ds_chomp(ds, ',');
1891 ds_put_cstr(ds, "),");
1892 break;
1893 }
1894
1895 default:
1896 format_unknown_key(ds, a, ma);
1897 }
1898 ofpbuf_clear(&ofp);
1899 }
1900
1901 ds_chomp(ds, ',');
1902 ofpbuf_uninit(&ofp);
1903 }
1904
1905 #define MASK(PTR, FIELD) PTR ? &PTR->FIELD : NULL
1906
1907 static void
1908 format_odp_tun_geneve(const struct nlattr *attr,
1909 const struct nlattr *mask_attr, struct ds *ds,
1910 bool verbose)
1911 {
1912 int opts_len = nl_attr_get_size(attr);
1913 const struct geneve_opt *opt = nl_attr_get(attr);
1914 const struct geneve_opt *mask = mask_attr ?
1915 nl_attr_get(mask_attr) : NULL;
1916
1917 if (mask && nl_attr_get_size(attr) != nl_attr_get_size(mask_attr)) {
1918 ds_put_format(ds, "value len %"PRIuSIZE" different from mask len %"PRIuSIZE,
1919 nl_attr_get_size(attr), nl_attr_get_size(mask_attr));
1920 return;
1921 }
1922
1923 while (opts_len > 0) {
1924 unsigned int len;
1925 uint8_t data_len, data_len_mask;
1926
1927 if (opts_len < sizeof *opt) {
1928 ds_put_format(ds, "opt len %u less than minimum %"PRIuSIZE,
1929 opts_len, sizeof *opt);
1930 return;
1931 }
1932
1933 data_len = opt->length * 4;
1934 if (mask) {
1935 if (mask->length == 0x1f) {
1936 data_len_mask = UINT8_MAX;
1937 } else {
1938 data_len_mask = mask->length;
1939 }
1940 }
1941 len = sizeof *opt + data_len;
1942 if (len > opts_len) {
1943 ds_put_format(ds, "opt len %u greater than remaining %u",
1944 len, opts_len);
1945 return;
1946 }
1947
1948 ds_put_char(ds, '{');
1949 format_be16x(ds, "class", opt->opt_class, MASK(mask, opt_class),
1950 verbose);
1951 format_u8x(ds, "type", opt->type, MASK(mask, type), verbose);
1952 format_u8u(ds, "len", data_len, mask ? &data_len_mask : NULL, verbose);
1953 if (verbose || !mask || !is_all_zeros(mask + 1, data_len)) {
1954 ds_put_hex(ds, opt + 1, data_len);
1955 if (mask && !is_all_ones(mask + 1, data_len)) {
1956 ds_put_char(ds, '/');
1957 ds_put_hex(ds, mask + 1, data_len);
1958 }
1959 } else {
1960 ds_chomp(ds, ',');
1961 }
1962 ds_put_char(ds, '}');
1963
1964 opt += len / sizeof(*opt);
1965 if (mask) {
1966 mask += len / sizeof(*opt);
1967 }
1968 opts_len -= len;
1969 };
1970 }
1971
1972 static void
1973 format_odp_tun_attr(const struct nlattr *attr, const struct nlattr *mask_attr,
1974 struct ds *ds, bool verbose)
1975 {
1976 unsigned int left;
1977 const struct nlattr *a;
1978 uint16_t flags = 0;
1979 uint16_t mask_flags = 0;
1980 struct ofpbuf ofp;
1981
1982 ofpbuf_init(&ofp, 100);
1983 NL_NESTED_FOR_EACH(a, left, attr) {
1984 enum ovs_tunnel_key_attr type = nl_attr_type(a);
1985 const struct nlattr *ma = NULL;
1986
1987 if (mask_attr) {
1988 ma = nl_attr_find__(nl_attr_get(mask_attr),
1989 nl_attr_get_size(mask_attr), type);
1990 if (!ma) {
1991 ma = generate_all_wildcard_mask(ovs_tun_key_attr_lens,
1992 OVS_TUNNEL_KEY_ATTR_MAX,
1993 &ofp, a);
1994 }
1995 }
1996
1997 if (!check_attr_len(ds, a, ma, ovs_tun_key_attr_lens,
1998 OVS_TUNNEL_KEY_ATTR_MAX, true)) {
1999 continue;
2000 }
2001
2002 switch (type) {
2003 case OVS_TUNNEL_KEY_ATTR_ID:
2004 format_be64(ds, "tun_id", nl_attr_get_be64(a),
2005 ma ? nl_attr_get(ma) : NULL, verbose);
2006 flags |= FLOW_TNL_F_KEY;
2007 if (ma) {
2008 mask_flags |= FLOW_TNL_F_KEY;
2009 }
2010 break;
2011 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
2012 format_ipv4(ds, "src", nl_attr_get_be32(a),
2013 ma ? nl_attr_get(ma) : NULL, verbose);
2014 break;
2015 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
2016 format_ipv4(ds, "dst", nl_attr_get_be32(a),
2017 ma ? nl_attr_get(ma) : NULL, verbose);
2018 break;
2019 case OVS_TUNNEL_KEY_ATTR_TOS:
2020 format_u8x(ds, "tos", nl_attr_get_u8(a),
2021 ma ? nl_attr_get(ma) : NULL, verbose);
2022 break;
2023 case OVS_TUNNEL_KEY_ATTR_TTL:
2024 format_u8u(ds, "ttl", nl_attr_get_u8(a),
2025 ma ? nl_attr_get(ma) : NULL, verbose);
2026 break;
2027 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
2028 flags |= FLOW_TNL_F_DONT_FRAGMENT;
2029 break;
2030 case OVS_TUNNEL_KEY_ATTR_CSUM:
2031 flags |= FLOW_TNL_F_CSUM;
2032 break;
2033 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
2034 format_be16(ds, "tp_src", nl_attr_get_be16(a),
2035 ma ? nl_attr_get(ma) : NULL, verbose);
2036 break;
2037 case OVS_TUNNEL_KEY_ATTR_TP_DST:
2038 format_be16(ds, "tp_dst", nl_attr_get_be16(a),
2039 ma ? nl_attr_get(ma) : NULL, verbose);
2040 break;
2041 case OVS_TUNNEL_KEY_ATTR_OAM:
2042 flags |= FLOW_TNL_F_OAM;
2043 break;
2044 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
2045 ds_put_cstr(ds, "vxlan(");
2046 format_odp_tun_vxlan_opt(a, ma, ds, verbose);
2047 ds_put_cstr(ds, "),");
2048 break;
2049 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
2050 ds_put_cstr(ds, "geneve(");
2051 format_odp_tun_geneve(a, ma, ds, verbose);
2052 ds_put_cstr(ds, "),");
2053 break;
2054 case __OVS_TUNNEL_KEY_ATTR_MAX:
2055 default:
2056 format_unknown_key(ds, a, ma);
2057 }
2058 ofpbuf_clear(&ofp);
2059 }
2060
2061 /* Flags can have a valid mask even if the attribute is not set, so
2062 * we need to collect these separately. */
2063 if (mask_attr) {
2064 NL_NESTED_FOR_EACH(a, left, mask_attr) {
2065 switch (nl_attr_type(a)) {
2066 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
2067 mask_flags |= FLOW_TNL_F_DONT_FRAGMENT;
2068 break;
2069 case OVS_TUNNEL_KEY_ATTR_CSUM:
2070 mask_flags |= FLOW_TNL_F_CSUM;
2071 break;
2072 case OVS_TUNNEL_KEY_ATTR_OAM:
2073 mask_flags |= FLOW_TNL_F_OAM;
2074 break;
2075 }
2076 }
2077 }
2078
2079 format_tun_flags(ds, "flags", flags, mask_attr ? &mask_flags : NULL,
2080 verbose);
2081 ds_chomp(ds, ',');
2082 ofpbuf_uninit(&ofp);
2083 }
2084
2085 static void
2086 format_frag(struct ds *ds, const char *name, uint8_t key,
2087 const uint8_t *mask, bool verbose)
2088 {
2089 bool mask_empty = mask && !*mask;
2090
2091 /* ODP frag is an enumeration field; partial masks are not meaningful. */
2092 if (verbose || !mask_empty) {
2093 bool mask_full = !mask || *mask == UINT8_MAX;
2094
2095 if (!mask_full) { /* Partially masked. */
2096 ds_put_format(ds, "error: partial mask not supported for frag (%#"
2097 PRIx8"),", *mask);
2098 } else {
2099 ds_put_format(ds, "%s=%s,", name, ovs_frag_type_to_string(key));
2100 }
2101 }
2102 }
2103
2104 static void
2105 format_odp_key_attr(const struct nlattr *a, const struct nlattr *ma,
2106 const struct hmap *portno_names, struct ds *ds,
2107 bool verbose)
2108 {
2109 enum ovs_key_attr attr = nl_attr_type(a);
2110 char namebuf[OVS_KEY_ATTR_BUFSIZE];
2111 bool is_exact;
2112
2113 is_exact = ma ? odp_mask_attr_is_exact(ma) : true;
2114
2115 ds_put_cstr(ds, ovs_key_attr_to_string(attr, namebuf, sizeof namebuf));
2116
2117 if (!check_attr_len(ds, a, ma, ovs_flow_key_attr_lens,
2118 OVS_KEY_ATTR_MAX, false)) {
2119 return;
2120 }
2121
2122 ds_put_char(ds, '(');
2123 switch (attr) {
2124 case OVS_KEY_ATTR_ENCAP:
2125 if (ma && nl_attr_get_size(ma) && nl_attr_get_size(a)) {
2126 odp_flow_format(nl_attr_get(a), nl_attr_get_size(a),
2127 nl_attr_get(ma), nl_attr_get_size(ma), NULL, ds,
2128 verbose);
2129 } else if (nl_attr_get_size(a)) {
2130 odp_flow_format(nl_attr_get(a), nl_attr_get_size(a), NULL, 0, NULL,
2131 ds, verbose);
2132 }
2133 break;
2134
2135 case OVS_KEY_ATTR_PRIORITY:
2136 case OVS_KEY_ATTR_SKB_MARK:
2137 case OVS_KEY_ATTR_DP_HASH:
2138 case OVS_KEY_ATTR_RECIRC_ID:
2139 ds_put_format(ds, "%#"PRIx32, nl_attr_get_u32(a));
2140 if (!is_exact) {
2141 ds_put_format(ds, "/%#"PRIx32, nl_attr_get_u32(ma));
2142 }
2143 break;
2144
2145 case OVS_KEY_ATTR_TUNNEL:
2146 format_odp_tun_attr(a, ma, ds, verbose);
2147 break;
2148
2149 case OVS_KEY_ATTR_IN_PORT:
2150 if (portno_names && verbose && is_exact) {
2151 char *name = odp_portno_names_get(portno_names,
2152 u32_to_odp(nl_attr_get_u32(a)));
2153 if (name) {
2154 ds_put_format(ds, "%s", name);
2155 } else {
2156 ds_put_format(ds, "%"PRIu32, nl_attr_get_u32(a));
2157 }
2158 } else {
2159 ds_put_format(ds, "%"PRIu32, nl_attr_get_u32(a));
2160 if (!is_exact) {
2161 ds_put_format(ds, "/%#"PRIx32, nl_attr_get_u32(ma));
2162 }
2163 }
2164 break;
2165
2166 case OVS_KEY_ATTR_ETHERNET: {
2167 const struct ovs_key_ethernet *mask = ma ? nl_attr_get(ma) : NULL;
2168 const struct ovs_key_ethernet *key = nl_attr_get(a);
2169
2170 format_eth(ds, "src", key->eth_src, MASK(mask, eth_src), verbose);
2171 format_eth(ds, "dst", key->eth_dst, MASK(mask, eth_dst), verbose);
2172 ds_chomp(ds, ',');
2173 break;
2174 }
2175 case OVS_KEY_ATTR_VLAN:
2176 format_vlan_tci(ds, nl_attr_get_be16(a),
2177 ma ? nl_attr_get_be16(ma) : OVS_BE16_MAX, verbose);
2178 break;
2179
2180 case OVS_KEY_ATTR_MPLS: {
2181 const struct ovs_key_mpls *mpls_key = nl_attr_get(a);
2182 const struct ovs_key_mpls *mpls_mask = NULL;
2183 size_t size = nl_attr_get_size(a);
2184
2185 if (!size || size % sizeof *mpls_key) {
2186 ds_put_format(ds, "(bad key length %"PRIuSIZE")", size);
2187 return;
2188 }
2189 if (!is_exact) {
2190 mpls_mask = nl_attr_get(ma);
2191 if (size != nl_attr_get_size(ma)) {
2192 ds_put_format(ds, "(key length %"PRIuSIZE" != "
2193 "mask length %"PRIuSIZE")",
2194 size, nl_attr_get_size(ma));
2195 return;
2196 }
2197 }
2198 format_mpls(ds, mpls_key, mpls_mask, size / sizeof *mpls_key);
2199 break;
2200 }
2201 case OVS_KEY_ATTR_ETHERTYPE:
2202 ds_put_format(ds, "0x%04"PRIx16, ntohs(nl_attr_get_be16(a)));
2203 if (!is_exact) {
2204 ds_put_format(ds, "/0x%04"PRIx16, ntohs(nl_attr_get_be16(ma)));
2205 }
2206 break;
2207
2208 case OVS_KEY_ATTR_IPV4: {
2209 const struct ovs_key_ipv4 *key = nl_attr_get(a);
2210 const struct ovs_key_ipv4 *mask = ma ? nl_attr_get(ma) : NULL;
2211
2212 format_ipv4(ds, "src", key->ipv4_src, MASK(mask, ipv4_src), verbose);
2213 format_ipv4(ds, "dst", key->ipv4_dst, MASK(mask, ipv4_dst), verbose);
2214 format_u8u(ds, "proto", key->ipv4_proto, MASK(mask, ipv4_proto),
2215 verbose);
2216 format_u8x(ds, "tos", key->ipv4_tos, MASK(mask, ipv4_tos), verbose);
2217 format_u8u(ds, "ttl", key->ipv4_ttl, MASK(mask, ipv4_ttl), verbose);
2218 format_frag(ds, "frag", key->ipv4_frag, MASK(mask, ipv4_frag),
2219 verbose);
2220 ds_chomp(ds, ',');
2221 break;
2222 }
2223 case OVS_KEY_ATTR_IPV6: {
2224 const struct ovs_key_ipv6 *key = nl_attr_get(a);
2225 const struct ovs_key_ipv6 *mask = ma ? nl_attr_get(ma) : NULL;
2226
2227 format_ipv6(ds, "src", key->ipv6_src, MASK(mask, ipv6_src), verbose);
2228 format_ipv6(ds, "dst", key->ipv6_dst, MASK(mask, ipv6_dst), verbose);
2229 format_ipv6_label(ds, "label", key->ipv6_label, MASK(mask, ipv6_label),
2230 verbose);
2231 format_u8u(ds, "proto", key->ipv6_proto, MASK(mask, ipv6_proto),
2232 verbose);
2233 format_u8x(ds, "tclass", key->ipv6_tclass, MASK(mask, ipv6_tclass),
2234 verbose);
2235 format_u8u(ds, "hlimit", key->ipv6_hlimit, MASK(mask, ipv6_hlimit),
2236 verbose);
2237 format_frag(ds, "frag", key->ipv6_frag, MASK(mask, ipv6_frag),
2238 verbose);
2239 ds_chomp(ds, ',');
2240 break;
2241 }
2242 /* These have the same structure and format. */
2243 case OVS_KEY_ATTR_TCP:
2244 case OVS_KEY_ATTR_UDP:
2245 case OVS_KEY_ATTR_SCTP: {
2246 const struct ovs_key_tcp *key = nl_attr_get(a);
2247 const struct ovs_key_tcp *mask = ma ? nl_attr_get(ma) : NULL;
2248
2249 format_be16(ds, "src", key->tcp_src, MASK(mask, tcp_src), verbose);
2250 format_be16(ds, "dst", key->tcp_dst, MASK(mask, tcp_dst), verbose);
2251 ds_chomp(ds, ',');
2252 break;
2253 }
2254 case OVS_KEY_ATTR_TCP_FLAGS:
2255 if (!is_exact) {
2256 format_flags_masked(ds, NULL, packet_tcp_flag_to_string,
2257 ntohs(nl_attr_get_be16(a)),
2258 ntohs(nl_attr_get_be16(ma)));
2259 } else {
2260 format_flags(ds, packet_tcp_flag_to_string,
2261 ntohs(nl_attr_get_be16(a)), ',');
2262 }
2263 break;
2264
2265 case OVS_KEY_ATTR_ICMP: {
2266 const struct ovs_key_icmp *key = nl_attr_get(a);
2267 const struct ovs_key_icmp *mask = ma ? nl_attr_get(ma) : NULL;
2268
2269 format_u8u(ds, "type", key->icmp_type, MASK(mask, icmp_type), verbose);
2270 format_u8u(ds, "code", key->icmp_code, MASK(mask, icmp_code), verbose);
2271 ds_chomp(ds, ',');
2272 break;
2273 }
2274 case OVS_KEY_ATTR_ICMPV6: {
2275 const struct ovs_key_icmpv6 *key = nl_attr_get(a);
2276 const struct ovs_key_icmpv6 *mask = ma ? nl_attr_get(ma) : NULL;
2277
2278 format_u8u(ds, "type", key->icmpv6_type, MASK(mask, icmpv6_type),
2279 verbose);
2280 format_u8u(ds, "code", key->icmpv6_code, MASK(mask, icmpv6_code),
2281 verbose);
2282 ds_chomp(ds, ',');
2283 break;
2284 }
2285 case OVS_KEY_ATTR_ARP: {
2286 const struct ovs_key_arp *mask = ma ? nl_attr_get(ma) : NULL;
2287 const struct ovs_key_arp *key = nl_attr_get(a);
2288
2289 format_ipv4(ds, "sip", key->arp_sip, MASK(mask, arp_sip), verbose);
2290 format_ipv4(ds, "tip", key->arp_tip, MASK(mask, arp_tip), verbose);
2291 format_be16(ds, "op", key->arp_op, MASK(mask, arp_op), verbose);
2292 format_eth(ds, "sha", key->arp_sha, MASK(mask, arp_sha), verbose);
2293 format_eth(ds, "tha", key->arp_tha, MASK(mask, arp_tha), verbose);
2294 ds_chomp(ds, ',');
2295 break;
2296 }
2297 case OVS_KEY_ATTR_ND: {
2298 const struct ovs_key_nd *mask = ma ? nl_attr_get(ma) : NULL;
2299 const struct ovs_key_nd *key = nl_attr_get(a);
2300
2301 format_ipv6(ds, "target", key->nd_target, MASK(mask, nd_target),
2302 verbose);
2303 format_eth(ds, "sll", key->nd_sll, MASK(mask, nd_sll), verbose);
2304 format_eth(ds, "tll", key->nd_tll, MASK(mask, nd_tll), verbose);
2305
2306 ds_chomp(ds, ',');
2307 break;
2308 }
2309 case OVS_KEY_ATTR_UNSPEC:
2310 case __OVS_KEY_ATTR_MAX:
2311 default:
2312 format_generic_odp_key(a, ds);
2313 if (!is_exact) {
2314 ds_put_char(ds, '/');
2315 format_generic_odp_key(ma, ds);
2316 }
2317 break;
2318 }
2319 ds_put_char(ds, ')');
2320 }
2321
2322 static struct nlattr *
2323 generate_all_wildcard_mask(const struct attr_len_tbl tbl[], int max,
2324 struct ofpbuf *ofp, const struct nlattr *key)
2325 {
2326 const struct nlattr *a;
2327 unsigned int left;
2328 int type = nl_attr_type(key);
2329 int size = nl_attr_get_size(key);
2330
2331 if (odp_key_attr_len(tbl, max, type) != ATTR_LEN_NESTED) {
2332 nl_msg_put_unspec_zero(ofp, type, size);
2333 } else {
2334 size_t nested_mask;
2335
2336 if (tbl[type].next) {
2337 tbl = tbl[type].next;
2338 max = tbl[type].next_max;
2339 }
2340
2341 nested_mask = nl_msg_start_nested(ofp, type);
2342 NL_ATTR_FOR_EACH(a, left, key, nl_attr_get_size(key)) {
2343 generate_all_wildcard_mask(tbl, max, ofp, nl_attr_get(a));
2344 }
2345 nl_msg_end_nested(ofp, nested_mask);
2346 }
2347
2348 return ofp->base;
2349 }
2350
2351 int
2352 odp_ufid_from_string(const char *s_, ovs_u128 *ufid)
2353 {
2354 const char *s = s_;
2355
2356 if (ovs_scan(s, "ufid:")) {
2357 size_t n;
2358
2359 s += 5;
2360 if (ovs_scan(s, "0x")) {
2361 s += 2;
2362 }
2363
2364 n = strspn(s, hex_chars);
2365 if (n != 32) {
2366 return -EINVAL;
2367 }
2368
2369 if (!ovs_scan(s, "%16"SCNx64"%16"SCNx64, &ufid->u64.hi,
2370 &ufid->u64.lo)) {
2371 return -EINVAL;
2372 }
2373 s += n;
2374 s += strspn(s, delimiters);
2375
2376 return s - s_;
2377 }
2378
2379 return 0;
2380 }
2381
2382 void
2383 odp_format_ufid(const ovs_u128 *ufid, struct ds *ds)
2384 {
2385 ds_put_format(ds, "ufid:%016"PRIx64"%016"PRIx64, ufid->u64.hi,
2386 ufid->u64.lo);
2387 }
2388
2389 /* Appends to 'ds' a string representation of the 'key_len' bytes of
2390 * OVS_KEY_ATTR_* attributes in 'key'. If non-null, additionally formats the
2391 * 'mask_len' bytes of 'mask' which apply to 'key'. If 'portno_names' is
2392 * non-null and 'verbose' is true, translates odp port number to its name. */
2393 void
2394 odp_flow_format(const struct nlattr *key, size_t key_len,
2395 const struct nlattr *mask, size_t mask_len,
2396 const struct hmap *portno_names, struct ds *ds, bool verbose)
2397 {
2398 if (key_len) {
2399 const struct nlattr *a;
2400 unsigned int left;
2401 bool has_ethtype_key = false;
2402 const struct nlattr *ma = NULL;
2403 struct ofpbuf ofp;
2404 bool first_field = true;
2405
2406 ofpbuf_init(&ofp, 100);
2407 NL_ATTR_FOR_EACH (a, left, key, key_len) {
2408 bool is_nested_attr;
2409 bool is_wildcard = false;
2410 int attr_type = nl_attr_type(a);
2411
2412 if (attr_type == OVS_KEY_ATTR_ETHERTYPE) {
2413 has_ethtype_key = true;
2414 }
2415
2416 is_nested_attr = odp_key_attr_len(ovs_flow_key_attr_lens,
2417 OVS_KEY_ATTR_MAX, attr_type) ==
2418 ATTR_LEN_NESTED;
2419
2420 if (mask && mask_len) {
2421 ma = nl_attr_find__(mask, mask_len, nl_attr_type(a));
2422 is_wildcard = ma ? odp_mask_attr_is_wildcard(ma) : true;
2423 }
2424
2425 if (verbose || !is_wildcard || is_nested_attr) {
2426 if (is_wildcard && !ma) {
2427 ma = generate_all_wildcard_mask(ovs_flow_key_attr_lens,
2428 OVS_KEY_ATTR_MAX,
2429 &ofp, a);
2430 }
2431 if (!first_field) {
2432 ds_put_char(ds, ',');
2433 }
2434 format_odp_key_attr(a, ma, portno_names, ds, verbose);
2435 first_field = false;
2436 }
2437 ofpbuf_clear(&ofp);
2438 }
2439 ofpbuf_uninit(&ofp);
2440
2441 if (left) {
2442 int i;
2443
2444 if (left == key_len) {
2445 ds_put_cstr(ds, "<empty>");
2446 }
2447 ds_put_format(ds, ",***%u leftover bytes*** (", left);
2448 for (i = 0; i < left; i++) {
2449 ds_put_format(ds, "%02x", ((const uint8_t *) a)[i]);
2450 }
2451 ds_put_char(ds, ')');
2452 }
2453 if (!has_ethtype_key) {
2454 ma = nl_attr_find__(mask, mask_len, OVS_KEY_ATTR_ETHERTYPE);
2455 if (ma) {
2456 ds_put_format(ds, ",eth_type(0/0x%04"PRIx16")",
2457 ntohs(nl_attr_get_be16(ma)));
2458 }
2459 }
2460 } else {
2461 ds_put_cstr(ds, "<empty>");
2462 }
2463 }
2464
2465 /* Appends to 'ds' a string representation of the 'key_len' bytes of
2466 * OVS_KEY_ATTR_* attributes in 'key'. */
2467 void
2468 odp_flow_key_format(const struct nlattr *key,
2469 size_t key_len, struct ds *ds)
2470 {
2471 odp_flow_format(key, key_len, NULL, 0, NULL, ds, true);
2472 }
2473
2474 static bool
2475 ovs_frag_type_from_string(const char *s, enum ovs_frag_type *type)
2476 {
2477 if (!strcasecmp(s, "no")) {
2478 *type = OVS_FRAG_TYPE_NONE;
2479 } else if (!strcasecmp(s, "first")) {
2480 *type = OVS_FRAG_TYPE_FIRST;
2481 } else if (!strcasecmp(s, "later")) {
2482 *type = OVS_FRAG_TYPE_LATER;
2483 } else {
2484 return false;
2485 }
2486 return true;
2487 }
2488
2489 /* Parsing. */
2490
2491 static int
2492 scan_eth(const char *s, uint8_t (*key)[ETH_ADDR_LEN],
2493 uint8_t (*mask)[ETH_ADDR_LEN])
2494 {
2495 int n;
2496
2497 if (ovs_scan(s, ETH_ADDR_SCAN_FMT"%n", ETH_ADDR_SCAN_ARGS(*key), &n)) {
2498 int len = n;
2499
2500 if (mask) {
2501 if (ovs_scan(s + len, "/"ETH_ADDR_SCAN_FMT"%n",
2502 ETH_ADDR_SCAN_ARGS(*mask), &n)) {
2503 len += n;
2504 } else {
2505 memset(mask, 0xff, sizeof *mask);
2506 }
2507 }
2508 return len;
2509 }
2510 return 0;
2511 }
2512
2513 static int
2514 scan_ipv4(const char *s, ovs_be32 *key, ovs_be32 *mask)
2515 {
2516 int n;
2517
2518 if (ovs_scan(s, IP_SCAN_FMT"%n", IP_SCAN_ARGS(key), &n)) {
2519 int len = n;
2520
2521 if (mask) {
2522 if (ovs_scan(s + len, "/"IP_SCAN_FMT"%n",
2523 IP_SCAN_ARGS(mask), &n)) {
2524 len += n;
2525 } else {
2526 *mask = OVS_BE32_MAX;
2527 }
2528 }
2529 return len;
2530 }
2531 return 0;
2532 }
2533
2534 static int
2535 scan_ipv6(const char *s, ovs_be32 (*key)[4], ovs_be32 (*mask)[4])
2536 {
2537 int n;
2538 char ipv6_s[IPV6_SCAN_LEN + 1];
2539
2540 if (ovs_scan(s, IPV6_SCAN_FMT"%n", ipv6_s, &n)
2541 && inet_pton(AF_INET6, ipv6_s, key) == 1) {
2542 int len = n;
2543
2544 if (mask) {
2545 if (ovs_scan(s + len, "/"IPV6_SCAN_FMT"%n", ipv6_s, &n)
2546 && inet_pton(AF_INET6, ipv6_s, mask) == 1) {
2547 len += n;
2548 } else {
2549 memset(mask, 0xff, sizeof *mask);
2550 }
2551 }
2552 return len;
2553 }
2554 return 0;
2555 }
2556
2557 static int
2558 scan_ipv6_label(const char *s, ovs_be32 *key, ovs_be32 *mask)
2559 {
2560 int key_, mask_;
2561 int n;
2562
2563 if (ovs_scan(s, "%i%n", &key_, &n)
2564 && (key_ & ~IPV6_LABEL_MASK) == 0) {
2565 int len = n;
2566
2567 *key = htonl(key_);
2568 if (mask) {
2569 if (ovs_scan(s + len, "/%i%n", &mask_, &n)
2570 && (mask_ & ~IPV6_LABEL_MASK) == 0) {
2571 len += n;
2572 *mask = htonl(mask_);
2573 } else {
2574 *mask = htonl(IPV6_LABEL_MASK);
2575 }
2576 }
2577 return len;
2578 }
2579 return 0;
2580 }
2581
2582 static int
2583 scan_u8(const char *s, uint8_t *key, uint8_t *mask)
2584 {
2585 int n;
2586
2587 if (ovs_scan(s, "%"SCNi8"%n", key, &n)) {
2588 int len = n;
2589
2590 if (mask) {
2591 if (ovs_scan(s + len, "/%"SCNi8"%n", mask, &n)) {
2592 len += n;
2593 } else {
2594 *mask = UINT8_MAX;
2595 }
2596 }
2597 return len;
2598 }
2599 return 0;
2600 }
2601
2602 static int
2603 scan_u32(const char *s, uint32_t *key, uint32_t *mask)
2604 {
2605 int n;
2606
2607 if (ovs_scan(s, "%"SCNi32"%n", key, &n)) {
2608 int len = n;
2609
2610 if (mask) {
2611 if (ovs_scan(s + len, "/%"SCNi32"%n", mask, &n)) {
2612 len += n;
2613 } else {
2614 *mask = UINT32_MAX;
2615 }
2616 }
2617 return len;
2618 }
2619 return 0;
2620 }
2621
2622 static int
2623 scan_be16(const char *s, ovs_be16 *key, ovs_be16 *mask)
2624 {
2625 uint16_t key_, mask_;
2626 int n;
2627
2628 if (ovs_scan(s, "%"SCNi16"%n", &key_, &n)) {
2629 int len = n;
2630
2631 *key = htons(key_);
2632 if (mask) {
2633 if (ovs_scan(s + len, "/%"SCNi16"%n", &mask_, &n)) {
2634 len += n;
2635 *mask = htons(mask_);
2636 } else {
2637 *mask = OVS_BE16_MAX;
2638 }
2639 }
2640 return len;
2641 }
2642 return 0;
2643 }
2644
2645 static int
2646 scan_be64(const char *s, ovs_be64 *key, ovs_be64 *mask)
2647 {
2648 uint64_t key_, mask_;
2649 int n;
2650
2651 if (ovs_scan(s, "%"SCNi64"%n", &key_, &n)) {
2652 int len = n;
2653
2654 *key = htonll(key_);
2655 if (mask) {
2656 if (ovs_scan(s + len, "/%"SCNi64"%n", &mask_, &n)) {
2657 len += n;
2658 *mask = htonll(mask_);
2659 } else {
2660 *mask = OVS_BE64_MAX;
2661 }
2662 }
2663 return len;
2664 }
2665 return 0;
2666 }
2667
2668 static int
2669 scan_tun_flags(const char *s, uint16_t *key, uint16_t *mask)
2670 {
2671 uint32_t flags, fmask;
2672 int n;
2673
2674 n = parse_flags(s, flow_tun_flag_to_string, &flags,
2675 FLOW_TNL_F_MASK, mask ? &fmask : NULL);
2676 if (n >= 0 && s[n] == ')') {
2677 *key = flags;
2678 if (mask) {
2679 *mask = fmask;
2680 }
2681 return n + 1;
2682 }
2683 return 0;
2684 }
2685
2686 static int
2687 scan_tcp_flags(const char *s, ovs_be16 *key, ovs_be16 *mask)
2688 {
2689 uint32_t flags, fmask;
2690 int n;
2691
2692 n = parse_flags(s, packet_tcp_flag_to_string, &flags,
2693 TCP_FLAGS(OVS_BE16_MAX), mask ? &fmask : NULL);
2694 if (n >= 0) {
2695 *key = htons(flags);
2696 if (mask) {
2697 *mask = htons(fmask);
2698 }
2699 return n;
2700 }
2701 return 0;
2702 }
2703
2704 static int
2705 scan_frag(const char *s, uint8_t *key, uint8_t *mask)
2706 {
2707 int n;
2708 char frag[8];
2709 enum ovs_frag_type frag_type;
2710
2711 if (ovs_scan(s, "%7[a-z]%n", frag, &n)
2712 && ovs_frag_type_from_string(frag, &frag_type)) {
2713 int len = n;
2714
2715 *key = frag_type;
2716 if (mask) {
2717 *mask = UINT8_MAX;
2718 }
2719 return len;
2720 }
2721 return 0;
2722 }
2723
2724 static int
2725 scan_port(const char *s, uint32_t *key, uint32_t *mask,
2726 const struct simap *port_names)
2727 {
2728 int n;
2729
2730 if (ovs_scan(s, "%"SCNi32"%n", key, &n)) {
2731 int len = n;
2732
2733 if (mask) {
2734 if (ovs_scan(s + len, "/%"SCNi32"%n", mask, &n)) {
2735 len += n;
2736 } else {
2737 *mask = UINT32_MAX;
2738 }
2739 }
2740 return len;
2741 } else if (port_names) {
2742 const struct simap_node *node;
2743 int len;
2744
2745 len = strcspn(s, ")");
2746 node = simap_find_len(port_names, s, len);
2747 if (node) {
2748 *key = node->data;
2749
2750 if (mask) {
2751 *mask = UINT32_MAX;
2752 }
2753 return len;
2754 }
2755 }
2756 return 0;
2757 }
2758
2759 /* Helper for vlan parsing. */
2760 struct ovs_key_vlan__ {
2761 ovs_be16 tci;
2762 };
2763
2764 static bool
2765 set_be16_bf(ovs_be16 *bf, uint8_t bits, uint8_t offset, uint16_t value)
2766 {
2767 const uint16_t mask = ((1U << bits) - 1) << offset;
2768
2769 if (value >> bits) {
2770 return false;
2771 }
2772
2773 *bf = htons((ntohs(*bf) & ~mask) | (value << offset));
2774 return true;
2775 }
2776
2777 static int
2778 scan_be16_bf(const char *s, ovs_be16 *key, ovs_be16 *mask, uint8_t bits,
2779 uint8_t offset)
2780 {
2781 uint16_t key_, mask_;
2782 int n;
2783
2784 if (ovs_scan(s, "%"SCNi16"%n", &key_, &n)) {
2785 int len = n;
2786
2787 if (set_be16_bf(key, bits, offset, key_)) {
2788 if (mask) {
2789 if (ovs_scan(s + len, "/%"SCNi16"%n", &mask_, &n)) {
2790 len += n;
2791
2792 if (!set_be16_bf(mask, bits, offset, mask_)) {
2793 return 0;
2794 }
2795 } else {
2796 *mask |= htons(((1U << bits) - 1) << offset);
2797 }
2798 }
2799 return len;
2800 }
2801 }
2802 return 0;
2803 }
2804
2805 static int
2806 scan_vid(const char *s, ovs_be16 *key, ovs_be16 *mask)
2807 {
2808 return scan_be16_bf(s, key, mask, 12, VLAN_VID_SHIFT);
2809 }
2810
2811 static int
2812 scan_pcp(const char *s, ovs_be16 *key, ovs_be16 *mask)
2813 {
2814 return scan_be16_bf(s, key, mask, 3, VLAN_PCP_SHIFT);
2815 }
2816
2817 static int
2818 scan_cfi(const char *s, ovs_be16 *key, ovs_be16 *mask)
2819 {
2820 return scan_be16_bf(s, key, mask, 1, VLAN_CFI_SHIFT);
2821 }
2822
2823 /* For MPLS. */
2824 static bool
2825 set_be32_bf(ovs_be32 *bf, uint8_t bits, uint8_t offset, uint32_t value)
2826 {
2827 const uint32_t mask = ((1U << bits) - 1) << offset;
2828
2829 if (value >> bits) {
2830 return false;
2831 }
2832
2833 *bf = htonl((ntohl(*bf) & ~mask) | (value << offset));
2834 return true;
2835 }
2836
2837 static int
2838 scan_be32_bf(const char *s, ovs_be32 *key, ovs_be32 *mask, uint8_t bits,
2839 uint8_t offset)
2840 {
2841 uint32_t key_, mask_;
2842 int n;
2843
2844 if (ovs_scan(s, "%"SCNi32"%n", &key_, &n)) {
2845 int len = n;
2846
2847 if (set_be32_bf(key, bits, offset, key_)) {
2848 if (mask) {
2849 if (ovs_scan(s + len, "/%"SCNi32"%n", &mask_, &n)) {
2850 len += n;
2851
2852 if (!set_be32_bf(mask, bits, offset, mask_)) {
2853 return 0;
2854 }
2855 } else {
2856 *mask |= htonl(((1U << bits) - 1) << offset);
2857 }
2858 }
2859 return len;
2860 }
2861 }
2862 return 0;
2863 }
2864
2865 static int
2866 scan_mpls_label(const char *s, ovs_be32 *key, ovs_be32 *mask)
2867 {
2868 return scan_be32_bf(s, key, mask, 20, MPLS_LABEL_SHIFT);
2869 }
2870
2871 static int
2872 scan_mpls_tc(const char *s, ovs_be32 *key, ovs_be32 *mask)
2873 {
2874 return scan_be32_bf(s, key, mask, 3, MPLS_TC_SHIFT);
2875 }
2876
2877 static int
2878 scan_mpls_ttl(const char *s, ovs_be32 *key, ovs_be32 *mask)
2879 {
2880 return scan_be32_bf(s, key, mask, 8, MPLS_TTL_SHIFT);
2881 }
2882
2883 static int
2884 scan_mpls_bos(const char *s, ovs_be32 *key, ovs_be32 *mask)
2885 {
2886 return scan_be32_bf(s, key, mask, 1, MPLS_BOS_SHIFT);
2887 }
2888
2889 static int
2890 scan_vxlan_gbp(const char *s, uint32_t *key, uint32_t *mask)
2891 {
2892 const char *s_base = s;
2893 ovs_be16 id, id_mask;
2894 uint8_t flags, flags_mask;
2895
2896 if (!strncmp(s, "id=", 3)) {
2897 s += 3;
2898 s += scan_be16(s, &id, mask ? &id_mask : NULL);
2899 } else if (mask) {
2900 memset(&id_mask, 0, sizeof id_mask);
2901 }
2902
2903 if (s[0] == ',') {
2904 s++;
2905 }
2906 if (!strncmp(s, "flags=", 6)) {
2907 s += 6;
2908 s += scan_u8(s, &flags, mask ? &flags_mask : NULL);
2909 } else if (mask) {
2910 memset(&flags_mask, 0, sizeof flags_mask);
2911 }
2912
2913 if (!strncmp(s, "))", 2)) {
2914 s += 2;
2915
2916 *key = (flags << 16) | ntohs(id);
2917 if (mask) {
2918 *mask = (flags_mask << 16) | ntohs(id_mask);
2919 }
2920
2921 return s - s_base;
2922 }
2923
2924 return 0;
2925 }
2926
2927 struct geneve_scan {
2928 uint8_t d[252];
2929 int len;
2930 };
2931
2932 static int
2933 scan_geneve(const char *s, struct geneve_scan *key, struct geneve_scan *mask)
2934 {
2935 const char *s_base = s;
2936 struct geneve_opt *opt = (struct geneve_opt *)key->d;
2937 struct geneve_opt *opt_mask = (struct geneve_opt *)(mask ? mask->d : NULL);
2938 int len_remain = sizeof key->d;
2939
2940 while (s[0] == '{' && len_remain >= sizeof *opt) {
2941 int data_len = 0;
2942
2943 s++;
2944 len_remain -= sizeof *opt;
2945
2946 if (!strncmp(s, "class=", 6)) {
2947 s += 6;
2948 s += scan_be16(s, &opt->opt_class,
2949 mask ? &opt_mask->opt_class : NULL);
2950 } else if (mask) {
2951 memset(&opt_mask->opt_class, 0, sizeof opt_mask->opt_class);
2952 }
2953
2954 if (s[0] == ',') {
2955 s++;
2956 }
2957 if (!strncmp(s, "type=", 5)) {
2958 s += 5;
2959 s += scan_u8(s, &opt->type, mask ? &opt_mask->type : NULL);
2960 } else if (mask) {
2961 memset(&opt_mask->type, 0, sizeof opt_mask->type);
2962 }
2963
2964 if (s[0] == ',') {
2965 s++;
2966 }
2967 if (!strncmp(s, "len=", 4)) {
2968 uint8_t opt_len, opt_len_mask;
2969 s += 4;
2970 s += scan_u8(s, &opt_len, mask ? &opt_len_mask : NULL);
2971
2972 if (opt_len > 124 || opt_len % 4 || opt_len > len_remain) {
2973 return 0;
2974 }
2975 opt->length = opt_len / 4;
2976 if (mask) {
2977 opt_mask->length = opt_len_mask;
2978 }
2979 data_len = opt_len;
2980 } else if (mask) {
2981 memset(&opt_mask->type, 0, sizeof opt_mask->type);
2982 }
2983
2984 if (s[0] == ',') {
2985 s++;
2986 }
2987 if (parse_int_string(s, (uint8_t *)(opt + 1), data_len, (char **)&s)) {
2988 return 0;
2989 }
2990
2991 if (mask) {
2992 if (s[0] == '/') {
2993 s++;
2994 if (parse_int_string(s, (uint8_t *)(opt_mask + 1),
2995 data_len, (char **)&s)) {
2996 return 0;
2997 }
2998 }
2999 opt_mask->r1 = 0;
3000 opt_mask->r2 = 0;
3001 opt_mask->r3 = 0;
3002 }
3003
3004 if (s[0] == '}') {
3005 s++;
3006 opt += 1 + data_len / 4;
3007 if (mask) {
3008 opt_mask += 1 + data_len / 4;
3009 }
3010 len_remain -= data_len;
3011 }
3012 }
3013
3014 if (s[0] == ')') {
3015 int len = sizeof key->d - len_remain;
3016
3017 s++;
3018 key->len = len;
3019 if (mask) {
3020 mask->len = len;
3021 }
3022 return s - s_base;
3023 }
3024
3025 return 0;
3026 }
3027
3028 static void
3029 tun_flags_to_attr(struct ofpbuf *a, const void *data_)
3030 {
3031 const uint16_t *flags = data_;
3032
3033 if (*flags & FLOW_TNL_F_DONT_FRAGMENT) {
3034 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT);
3035 }
3036 if (*flags & FLOW_TNL_F_CSUM) {
3037 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_CSUM);
3038 }
3039 if (*flags & FLOW_TNL_F_OAM) {
3040 nl_msg_put_flag(a, OVS_TUNNEL_KEY_ATTR_OAM);
3041 }
3042 }
3043
3044 static void
3045 vxlan_gbp_to_attr(struct ofpbuf *a, const void *data_)
3046 {
3047 const uint32_t *gbp = data_;
3048
3049 if (*gbp) {
3050 size_t vxlan_opts_ofs;
3051
3052 vxlan_opts_ofs = nl_msg_start_nested(a, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
3053 nl_msg_put_u32(a, OVS_VXLAN_EXT_GBP, *gbp);
3054 nl_msg_end_nested(a, vxlan_opts_ofs);
3055 }
3056 }
3057
3058 static void
3059 geneve_to_attr(struct ofpbuf *a, const void *data_)
3060 {
3061 const struct geneve_scan *geneve = data_;
3062
3063 nl_msg_put_unspec(a, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS, geneve->d,
3064 geneve->len);
3065 }
3066
3067 #define SCAN_PUT_ATTR(BUF, ATTR, DATA, FUNC) \
3068 { \
3069 unsigned long call_fn = (unsigned long)FUNC; \
3070 if (call_fn) { \
3071 typedef void (*fn)(struct ofpbuf *, const void *); \
3072 fn func = FUNC; \
3073 func(BUF, &(DATA)); \
3074 } else { \
3075 nl_msg_put_unspec(BUF, ATTR, &(DATA), sizeof (DATA)); \
3076 } \
3077 }
3078
3079 #define SCAN_IF(NAME) \
3080 if (strncmp(s, NAME, strlen(NAME)) == 0) { \
3081 const char *start = s; \
3082 int len; \
3083 \
3084 s += strlen(NAME)
3085
3086 /* Usually no special initialization is needed. */
3087 #define SCAN_BEGIN(NAME, TYPE) \
3088 SCAN_IF(NAME); \
3089 TYPE skey, smask; \
3090 memset(&skey, 0, sizeof skey); \
3091 memset(&smask, 0, sizeof smask); \
3092 do { \
3093 len = 0;
3094
3095 /* Init as fully-masked as mask will not be scanned. */
3096 #define SCAN_BEGIN_FULLY_MASKED(NAME, TYPE) \
3097 SCAN_IF(NAME); \
3098 TYPE skey, smask; \
3099 memset(&skey, 0, sizeof skey); \
3100 memset(&smask, 0xff, sizeof smask); \
3101 do { \
3102 len = 0;
3103
3104 /* VLAN needs special initialization. */
3105 #define SCAN_BEGIN_INIT(NAME, TYPE, KEY_INIT, MASK_INIT) \
3106 SCAN_IF(NAME); \
3107 TYPE skey = KEY_INIT; \
3108 TYPE smask = MASK_INIT; \
3109 do { \
3110 len = 0;
3111
3112 /* Scan unnamed entry as 'TYPE' */
3113 #define SCAN_TYPE(TYPE, KEY, MASK) \
3114 len = scan_##TYPE(s, KEY, MASK); \
3115 if (len == 0) { \
3116 return -EINVAL; \
3117 } \
3118 s += len
3119
3120 /* Scan named ('NAME') entry 'FIELD' as 'TYPE'. */
3121 #define SCAN_FIELD(NAME, TYPE, FIELD) \
3122 if (strncmp(s, NAME, strlen(NAME)) == 0) { \
3123 s += strlen(NAME); \
3124 SCAN_TYPE(TYPE, &skey.FIELD, mask ? &smask.FIELD : NULL); \
3125 continue; \
3126 }
3127
3128 #define SCAN_FINISH() \
3129 } while (*s++ == ',' && len != 0); \
3130 if (s[-1] != ')') { \
3131 return -EINVAL; \
3132 }
3133
3134 #define SCAN_FINISH_SINGLE() \
3135 } while (false); \
3136 if (*s++ != ')') { \
3137 return -EINVAL; \
3138 }
3139
3140 /* Beginning of nested attribute. */
3141 #define SCAN_BEGIN_NESTED(NAME, ATTR) \
3142 SCAN_IF(NAME); \
3143 size_t key_offset, mask_offset; \
3144 key_offset = nl_msg_start_nested(key, ATTR); \
3145 if (mask) { \
3146 mask_offset = nl_msg_start_nested(mask, ATTR); \
3147 } \
3148 do { \
3149 len = 0;
3150
3151 #define SCAN_END_NESTED() \
3152 SCAN_FINISH(); \
3153 nl_msg_end_nested(key, key_offset); \
3154 if (mask) { \
3155 nl_msg_end_nested(mask, mask_offset); \
3156 } \
3157 return s - start; \
3158 }
3159
3160 #define SCAN_FIELD_NESTED__(NAME, TYPE, SCAN_AS, ATTR, FUNC) \
3161 if (strncmp(s, NAME, strlen(NAME)) == 0) { \
3162 TYPE skey, smask; \
3163 memset(&skey, 0, sizeof skey); \
3164 memset(&smask, 0xff, sizeof smask); \
3165 s += strlen(NAME); \
3166 SCAN_TYPE(SCAN_AS, &skey, &smask); \
3167 SCAN_PUT(ATTR, FUNC); \
3168 continue; \
3169 }
3170
3171 #define SCAN_FIELD_NESTED(NAME, TYPE, SCAN_AS, ATTR) \
3172 SCAN_FIELD_NESTED__(NAME, TYPE, SCAN_AS, ATTR, NULL)
3173
3174 #define SCAN_FIELD_NESTED_FUNC(NAME, TYPE, SCAN_AS, FUNC) \
3175 SCAN_FIELD_NESTED__(NAME, TYPE, SCAN_AS, 0, FUNC)
3176
3177 #define SCAN_PUT(ATTR, FUNC) \
3178 if (!mask || !is_all_zeros(&smask, sizeof smask)) { \
3179 SCAN_PUT_ATTR(key, ATTR, skey, FUNC); \
3180 if (mask) { \
3181 SCAN_PUT_ATTR(mask, ATTR, smask, FUNC); \
3182 } \
3183 }
3184
3185 #define SCAN_END(ATTR) \
3186 SCAN_FINISH(); \
3187 SCAN_PUT(ATTR, NULL); \
3188 return s - start; \
3189 }
3190
3191 #define SCAN_END_SINGLE(ATTR) \
3192 SCAN_FINISH_SINGLE(); \
3193 SCAN_PUT(ATTR, NULL); \
3194 return s - start; \
3195 }
3196
3197 #define SCAN_SINGLE(NAME, TYPE, SCAN_AS, ATTR) \
3198 SCAN_BEGIN(NAME, TYPE) { \
3199 SCAN_TYPE(SCAN_AS, &skey, &smask); \
3200 } SCAN_END_SINGLE(ATTR)
3201
3202 #define SCAN_SINGLE_FULLY_MASKED(NAME, TYPE, SCAN_AS, ATTR) \
3203 SCAN_BEGIN_FULLY_MASKED(NAME, TYPE) { \
3204 SCAN_TYPE(SCAN_AS, &skey, NULL); \
3205 } SCAN_END_SINGLE(ATTR)
3206
3207 /* scan_port needs one extra argument. */
3208 #define SCAN_SINGLE_PORT(NAME, TYPE, ATTR) \
3209 SCAN_BEGIN(NAME, TYPE) { \
3210 len = scan_port(s, &skey, &smask, port_names); \
3211 if (len == 0) { \
3212 return -EINVAL; \
3213 } \
3214 s += len; \
3215 } SCAN_END_SINGLE(ATTR)
3216
3217 static int
3218 parse_odp_key_mask_attr(const char *s, const struct simap *port_names,
3219 struct ofpbuf *key, struct ofpbuf *mask)
3220 {
3221 if (!strncmp(s, "ufid:", 5)) {
3222 const char *start = s;
3223
3224 /* Skip UFID. */
3225 s += 5;
3226 s += strspn(s, hex_chars);
3227 s += strspn(s, delimiters);
3228
3229 return s - start;
3230 }
3231
3232 SCAN_SINGLE("skb_priority(", uint32_t, u32, OVS_KEY_ATTR_PRIORITY);
3233 SCAN_SINGLE("skb_mark(", uint32_t, u32, OVS_KEY_ATTR_SKB_MARK);
3234 SCAN_SINGLE_FULLY_MASKED("recirc_id(", uint32_t, u32,
3235 OVS_KEY_ATTR_RECIRC_ID);
3236 SCAN_SINGLE("dp_hash(", uint32_t, u32, OVS_KEY_ATTR_DP_HASH);
3237
3238 SCAN_BEGIN_NESTED("tunnel(", OVS_KEY_ATTR_TUNNEL) {
3239 SCAN_FIELD_NESTED("tun_id=", ovs_be64, be64, OVS_TUNNEL_KEY_ATTR_ID);
3240 SCAN_FIELD_NESTED("src=", ovs_be32, ipv4, OVS_TUNNEL_KEY_ATTR_IPV4_SRC);
3241 SCAN_FIELD_NESTED("dst=", ovs_be32, ipv4, OVS_TUNNEL_KEY_ATTR_IPV4_DST);
3242 SCAN_FIELD_NESTED("tos=", uint8_t, u8, OVS_TUNNEL_KEY_ATTR_TOS);
3243 SCAN_FIELD_NESTED("ttl=", uint8_t, u8, OVS_TUNNEL_KEY_ATTR_TTL);
3244 SCAN_FIELD_NESTED("tp_src=", ovs_be16, be16, OVS_TUNNEL_KEY_ATTR_TP_SRC);
3245 SCAN_FIELD_NESTED("tp_dst=", ovs_be16, be16, OVS_TUNNEL_KEY_ATTR_TP_DST);
3246 SCAN_FIELD_NESTED_FUNC("vxlan(gbp(", uint32_t, vxlan_gbp, vxlan_gbp_to_attr);
3247 SCAN_FIELD_NESTED_FUNC("geneve(", struct geneve_scan, geneve,
3248 geneve_to_attr);
3249 SCAN_FIELD_NESTED_FUNC("flags(", uint16_t, tun_flags, tun_flags_to_attr);
3250 } SCAN_END_NESTED();
3251
3252 SCAN_SINGLE_PORT("in_port(", uint32_t, OVS_KEY_ATTR_IN_PORT);
3253
3254 SCAN_BEGIN("eth(", struct ovs_key_ethernet) {
3255 SCAN_FIELD("src=", eth, eth_src);
3256 SCAN_FIELD("dst=", eth, eth_dst);
3257 } SCAN_END(OVS_KEY_ATTR_ETHERNET);
3258
3259 SCAN_BEGIN_INIT("vlan(", struct ovs_key_vlan__,
3260 { htons(VLAN_CFI) }, { htons(VLAN_CFI) }) {
3261 SCAN_FIELD("vid=", vid, tci);
3262 SCAN_FIELD("pcp=", pcp, tci);
3263 SCAN_FIELD("cfi=", cfi, tci);
3264 } SCAN_END(OVS_KEY_ATTR_VLAN);
3265
3266 SCAN_SINGLE("eth_type(", ovs_be16, be16, OVS_KEY_ATTR_ETHERTYPE);
3267
3268 SCAN_BEGIN("mpls(", struct ovs_key_mpls) {
3269 SCAN_FIELD("label=", mpls_label, mpls_lse);
3270 SCAN_FIELD("tc=", mpls_tc, mpls_lse);
3271 SCAN_FIELD("ttl=", mpls_ttl, mpls_lse);
3272 SCAN_FIELD("bos=", mpls_bos, mpls_lse);
3273 } SCAN_END(OVS_KEY_ATTR_MPLS);
3274
3275 SCAN_BEGIN("ipv4(", struct ovs_key_ipv4) {
3276 SCAN_FIELD("src=", ipv4, ipv4_src);
3277 SCAN_FIELD("dst=", ipv4, ipv4_dst);
3278 SCAN_FIELD("proto=", u8, ipv4_proto);
3279 SCAN_FIELD("tos=", u8, ipv4_tos);
3280 SCAN_FIELD("ttl=", u8, ipv4_ttl);
3281 SCAN_FIELD("frag=", frag, ipv4_frag);
3282 } SCAN_END(OVS_KEY_ATTR_IPV4);
3283
3284 SCAN_BEGIN("ipv6(", struct ovs_key_ipv6) {
3285 SCAN_FIELD("src=", ipv6, ipv6_src);
3286 SCAN_FIELD("dst=", ipv6, ipv6_dst);
3287 SCAN_FIELD("label=", ipv6_label, ipv6_label);
3288 SCAN_FIELD("proto=", u8, ipv6_proto);
3289 SCAN_FIELD("tclass=", u8, ipv6_tclass);
3290 SCAN_FIELD("hlimit=", u8, ipv6_hlimit);
3291 SCAN_FIELD("frag=", frag, ipv6_frag);
3292 } SCAN_END(OVS_KEY_ATTR_IPV6);
3293
3294 SCAN_BEGIN("tcp(", struct ovs_key_tcp) {
3295 SCAN_FIELD("src=", be16, tcp_src);
3296 SCAN_FIELD("dst=", be16, tcp_dst);
3297 } SCAN_END(OVS_KEY_ATTR_TCP);
3298
3299 SCAN_SINGLE("tcp_flags(", ovs_be16, tcp_flags, OVS_KEY_ATTR_TCP_FLAGS);
3300
3301 SCAN_BEGIN("udp(", struct ovs_key_udp) {
3302 SCAN_FIELD("src=", be16, udp_src);
3303 SCAN_FIELD("dst=", be16, udp_dst);
3304 } SCAN_END(OVS_KEY_ATTR_UDP);
3305
3306 SCAN_BEGIN("sctp(", struct ovs_key_sctp) {
3307 SCAN_FIELD("src=", be16, sctp_src);
3308 SCAN_FIELD("dst=", be16, sctp_dst);
3309 } SCAN_END(OVS_KEY_ATTR_SCTP);
3310
3311 SCAN_BEGIN("icmp(", struct ovs_key_icmp) {
3312 SCAN_FIELD("type=", u8, icmp_type);
3313 SCAN_FIELD("code=", u8, icmp_code);
3314 } SCAN_END(OVS_KEY_ATTR_ICMP);
3315
3316 SCAN_BEGIN("icmpv6(", struct ovs_key_icmpv6) {
3317 SCAN_FIELD("type=", u8, icmpv6_type);
3318 SCAN_FIELD("code=", u8, icmpv6_code);
3319 } SCAN_END(OVS_KEY_ATTR_ICMPV6);
3320
3321 SCAN_BEGIN("arp(", struct ovs_key_arp) {
3322 SCAN_FIELD("sip=", ipv4, arp_sip);
3323 SCAN_FIELD("tip=", ipv4, arp_tip);
3324 SCAN_FIELD("op=", be16, arp_op);
3325 SCAN_FIELD("sha=", eth, arp_sha);
3326 SCAN_FIELD("tha=", eth, arp_tha);
3327 } SCAN_END(OVS_KEY_ATTR_ARP);
3328
3329 SCAN_BEGIN("nd(", struct ovs_key_nd) {
3330 SCAN_FIELD("target=", ipv6, nd_target);
3331 SCAN_FIELD("sll=", eth, nd_sll);
3332 SCAN_FIELD("tll=", eth, nd_tll);
3333 } SCAN_END(OVS_KEY_ATTR_ND);
3334
3335 /* Encap open-coded. */
3336 if (!strncmp(s, "encap(", 6)) {
3337 const char *start = s;
3338 size_t encap, encap_mask = 0;
3339
3340 encap = nl_msg_start_nested(key, OVS_KEY_ATTR_ENCAP);
3341 if (mask) {
3342 encap_mask = nl_msg_start_nested(mask, OVS_KEY_ATTR_ENCAP);
3343 }
3344
3345 s += 6;
3346 for (;;) {
3347 int retval;
3348
3349 s += strspn(s, delimiters);
3350 if (!*s) {
3351 return -EINVAL;
3352 } else if (*s == ')') {
3353 break;
3354 }
3355
3356 retval = parse_odp_key_mask_attr(s, port_names, key, mask);
3357 if (retval < 0) {
3358 return retval;
3359 }
3360 s += retval;
3361 }
3362 s++;
3363
3364 nl_msg_end_nested(key, encap);
3365 if (mask) {
3366 nl_msg_end_nested(mask, encap_mask);
3367 }
3368
3369 return s - start;
3370 }
3371
3372 return -EINVAL;
3373 }
3374
3375 /* Parses the string representation of a datapath flow key, in the
3376 * format output by odp_flow_key_format(). Returns 0 if successful,
3377 * otherwise a positive errno value. On success, the flow key is
3378 * appended to 'key' as a series of Netlink attributes. On failure, no
3379 * data is appended to 'key'. Either way, 'key''s data might be
3380 * reallocated.
3381 *
3382 * If 'port_names' is nonnull, it points to an simap that maps from a port name
3383 * to a port number. (Port names may be used instead of port numbers in
3384 * in_port.)
3385 *
3386 * On success, the attributes appended to 'key' are individually syntactically
3387 * valid, but they may not be valid as a sequence. 'key' might, for example,
3388 * have duplicated keys. odp_flow_key_to_flow() will detect those errors. */
3389 int
3390 odp_flow_from_string(const char *s, const struct simap *port_names,
3391 struct ofpbuf *key, struct ofpbuf *mask)
3392 {
3393 const size_t old_size = key->size;
3394 for (;;) {
3395 int retval;
3396
3397 s += strspn(s, delimiters);
3398 if (!*s) {
3399 return 0;
3400 }
3401
3402 retval = parse_odp_key_mask_attr(s, port_names, key, mask);
3403 if (retval < 0) {
3404 key->size = old_size;
3405 return -retval;
3406 }
3407 s += retval;
3408 }
3409
3410 return 0;
3411 }
3412
3413 static uint8_t
3414 ovs_to_odp_frag(uint8_t nw_frag, bool is_mask)
3415 {
3416 if (is_mask) {
3417 /* Netlink interface 'enum ovs_frag_type' is an 8-bit enumeration type,
3418 * not a set of flags or bitfields. Hence, if the struct flow nw_frag
3419 * mask, which is a set of bits, has the FLOW_NW_FRAG_ANY as zero, we
3420 * must use a zero mask for the netlink frag field, and all ones mask
3421 * otherwise. */
3422 return (nw_frag & FLOW_NW_FRAG_ANY) ? UINT8_MAX : 0;
3423 }
3424 return !(nw_frag & FLOW_NW_FRAG_ANY) ? OVS_FRAG_TYPE_NONE
3425 : nw_frag & FLOW_NW_FRAG_LATER ? OVS_FRAG_TYPE_LATER
3426 : OVS_FRAG_TYPE_FIRST;
3427 }
3428
3429 static void get_ethernet_key(const struct flow *, struct ovs_key_ethernet *);
3430 static void put_ethernet_key(const struct ovs_key_ethernet *, struct flow *);
3431 static void get_ipv4_key(const struct flow *, struct ovs_key_ipv4 *,
3432 bool is_mask);
3433 static void put_ipv4_key(const struct ovs_key_ipv4 *, struct flow *,
3434 bool is_mask);
3435 static void get_ipv6_key(const struct flow *, struct ovs_key_ipv6 *,
3436 bool is_mask);
3437 static void put_ipv6_key(const struct ovs_key_ipv6 *, struct flow *,
3438 bool is_mask);
3439 static void get_arp_key(const struct flow *, struct ovs_key_arp *);
3440 static void put_arp_key(const struct ovs_key_arp *, struct flow *);
3441 static void get_nd_key(const struct flow *, struct ovs_key_nd *);
3442 static void put_nd_key(const struct ovs_key_nd *, struct flow *);
3443
3444 /* These share the same layout. */
3445 union ovs_key_tp {
3446 struct ovs_key_tcp tcp;
3447 struct ovs_key_udp udp;
3448 struct ovs_key_sctp sctp;
3449 };
3450
3451 static void get_tp_key(const struct flow *, union ovs_key_tp *);
3452 static void put_tp_key(const union ovs_key_tp *, struct flow *);
3453
3454 static void
3455 odp_flow_key_from_flow__(struct ofpbuf *buf, const struct flow *flow,
3456 const struct flow *mask, odp_port_t odp_in_port,
3457 size_t max_mpls_depth, bool recirc, bool export_mask)
3458 {
3459 struct ovs_key_ethernet *eth_key;
3460 size_t encap;
3461 const struct flow *data = export_mask ? mask : flow;
3462
3463 nl_msg_put_u32(buf, OVS_KEY_ATTR_PRIORITY, data->skb_priority);
3464
3465 if (flow->tunnel.ip_dst || export_mask) {
3466 tun_key_to_attr(buf, &data->tunnel);
3467 }
3468
3469 nl_msg_put_u32(buf, OVS_KEY_ATTR_SKB_MARK, data->pkt_mark);
3470
3471 if (recirc) {
3472 nl_msg_put_u32(buf, OVS_KEY_ATTR_RECIRC_ID, data->recirc_id);
3473 nl_msg_put_u32(buf, OVS_KEY_ATTR_DP_HASH, data->dp_hash);
3474 }
3475
3476 /* Add an ingress port attribute if this is a mask or 'odp_in_port'
3477 * is not the magical value "ODPP_NONE". */
3478 if (export_mask || odp_in_port != ODPP_NONE) {
3479 nl_msg_put_odp_port(buf, OVS_KEY_ATTR_IN_PORT, odp_in_port);
3480 }
3481
3482 eth_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ETHERNET,
3483 sizeof *eth_key);
3484 get_ethernet_key(data, eth_key);
3485
3486 if (flow->vlan_tci != htons(0) || flow->dl_type == htons(ETH_TYPE_VLAN)) {
3487 if (export_mask) {
3488 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, OVS_BE16_MAX);
3489 } else {
3490 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_TYPE_VLAN));
3491 }
3492 nl_msg_put_be16(buf, OVS_KEY_ATTR_VLAN, data->vlan_tci);
3493 encap = nl_msg_start_nested(buf, OVS_KEY_ATTR_ENCAP);
3494 if (flow->vlan_tci == htons(0)) {
3495 goto unencap;
3496 }
3497 } else {
3498 encap = 0;
3499 }
3500
3501 if (ntohs(flow->dl_type) < ETH_TYPE_MIN) {
3502 /* For backwards compatibility with kernels that don't support
3503 * wildcarding, the following convention is used to encode the
3504 * OVS_KEY_ATTR_ETHERTYPE for key and mask:
3505 *
3506 * key mask matches
3507 * -------- -------- -------
3508 * >0x5ff 0xffff Specified Ethernet II Ethertype.
3509 * >0x5ff 0 Any Ethernet II or non-Ethernet II frame.
3510 * <none> 0xffff Any non-Ethernet II frame (except valid
3511 * 802.3 SNAP packet with valid eth_type).
3512 */
3513 if (export_mask) {
3514 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, OVS_BE16_MAX);
3515 }
3516 goto unencap;
3517 }
3518
3519 nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, data->dl_type);
3520
3521 if (flow->dl_type == htons(ETH_TYPE_IP)) {
3522 struct ovs_key_ipv4 *ipv4_key;
3523
3524 ipv4_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_IPV4,
3525 sizeof *ipv4_key);
3526 get_ipv4_key(data, ipv4_key, export_mask);
3527 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
3528 struct ovs_key_ipv6 *ipv6_key;
3529
3530 ipv6_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_IPV6,
3531 sizeof *ipv6_key);
3532 get_ipv6_key(data, ipv6_key, export_mask);
3533 } else if (flow->dl_type == htons(ETH_TYPE_ARP) ||
3534 flow->dl_type == htons(ETH_TYPE_RARP)) {
3535 struct ovs_key_arp *arp_key;
3536
3537 arp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ARP,
3538 sizeof *arp_key);
3539 get_arp_key(data, arp_key);
3540 } else if (eth_type_mpls(flow->dl_type)) {
3541 struct ovs_key_mpls *mpls_key;
3542 int i, n;
3543
3544 n = flow_count_mpls_labels(flow, NULL);
3545 n = MIN(n, max_mpls_depth);
3546 mpls_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_MPLS,
3547 n * sizeof *mpls_key);
3548 for (i = 0; i < n; i++) {
3549 mpls_key[i].mpls_lse = data->mpls_lse[i];
3550 }
3551 }
3552
3553 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3554 if (flow->nw_proto == IPPROTO_TCP) {
3555 union ovs_key_tp *tcp_key;
3556
3557 tcp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_TCP,
3558 sizeof *tcp_key);
3559 get_tp_key(data, tcp_key);
3560 if (data->tcp_flags) {
3561 nl_msg_put_be16(buf, OVS_KEY_ATTR_TCP_FLAGS, data->tcp_flags);
3562 }
3563 } else if (flow->nw_proto == IPPROTO_UDP) {
3564 union ovs_key_tp *udp_key;
3565
3566 udp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_UDP,
3567 sizeof *udp_key);
3568 get_tp_key(data, udp_key);
3569 } else if (flow->nw_proto == IPPROTO_SCTP) {
3570 union ovs_key_tp *sctp_key;
3571
3572 sctp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_SCTP,
3573 sizeof *sctp_key);
3574 get_tp_key(data, sctp_key);
3575 } else if (flow->dl_type == htons(ETH_TYPE_IP)
3576 && flow->nw_proto == IPPROTO_ICMP) {
3577 struct ovs_key_icmp *icmp_key;
3578
3579 icmp_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ICMP,
3580 sizeof *icmp_key);
3581 icmp_key->icmp_type = ntohs(data->tp_src);
3582 icmp_key->icmp_code = ntohs(data->tp_dst);
3583 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)
3584 && flow->nw_proto == IPPROTO_ICMPV6) {
3585 struct ovs_key_icmpv6 *icmpv6_key;
3586
3587 icmpv6_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ICMPV6,
3588 sizeof *icmpv6_key);
3589 icmpv6_key->icmpv6_type = ntohs(data->tp_src);
3590 icmpv6_key->icmpv6_code = ntohs(data->tp_dst);
3591
3592 if (flow->tp_dst == htons(0)
3593 && (flow->tp_src == htons(ND_NEIGHBOR_SOLICIT)
3594 || flow->tp_src == htons(ND_NEIGHBOR_ADVERT))
3595 && (!export_mask || (data->tp_src == htons(0xffff)
3596 && data->tp_dst == htons(0xffff)))) {
3597
3598 struct ovs_key_nd *nd_key;
3599
3600 nd_key = nl_msg_put_unspec_uninit(buf, OVS_KEY_ATTR_ND,
3601 sizeof *nd_key);
3602 memcpy(nd_key->nd_target, &data->nd_target,
3603 sizeof nd_key->nd_target);
3604 memcpy(nd_key->nd_sll, data->arp_sha, ETH_ADDR_LEN);
3605 memcpy(nd_key->nd_tll, data->arp_tha, ETH_ADDR_LEN);
3606 }
3607 }
3608 }
3609
3610 unencap:
3611 if (encap) {
3612 nl_msg_end_nested(buf, encap);
3613 }
3614 }
3615
3616 /* Appends a representation of 'flow' as OVS_KEY_ATTR_* attributes to 'buf'.
3617 * 'flow->in_port' is ignored (since it is likely to be an OpenFlow port
3618 * number rather than a datapath port number). Instead, if 'odp_in_port'
3619 * is anything other than ODPP_NONE, it is included in 'buf' as the input
3620 * port.
3621 *
3622 * 'buf' must have at least ODPUTIL_FLOW_KEY_BYTES bytes of space, or be
3623 * capable of being expanded to allow for that much space.
3624 *
3625 * 'recirc' indicates support for recirculation fields. If this is true, then
3626 * these fields will always be serialised. */
3627 void
3628 odp_flow_key_from_flow(struct ofpbuf *buf, const struct flow *flow,
3629 const struct flow *mask, odp_port_t odp_in_port,
3630 bool recirc)
3631 {
3632 odp_flow_key_from_flow__(buf, flow, mask, odp_in_port, SIZE_MAX, recirc,
3633 false);
3634 }
3635
3636 /* Appends a representation of 'mask' as OVS_KEY_ATTR_* attributes to
3637 * 'buf'. 'flow' is used as a template to determine how to interpret
3638 * 'mask'. For example, the 'dl_type' of 'mask' describes the mask, but
3639 * it doesn't indicate whether the other fields should be interpreted as
3640 * ARP, IPv4, IPv6, etc.
3641 *
3642 * 'buf' must have at least ODPUTIL_FLOW_KEY_BYTES bytes of space, or be
3643 * capable of being expanded to allow for that much space.
3644 *
3645 * 'recirc' indicates support for recirculation fields. If this is true, then
3646 * these fields will always be serialised. */
3647 void
3648 odp_flow_key_from_mask(struct ofpbuf *buf, const struct flow *mask,
3649 const struct flow *flow, uint32_t odp_in_port_mask,
3650 size_t max_mpls_depth, bool recirc)
3651 {
3652 odp_flow_key_from_flow__(buf, flow, mask, u32_to_odp(odp_in_port_mask),
3653 max_mpls_depth, recirc, true);
3654 }
3655
3656 /* Generate ODP flow key from the given packet metadata */
3657 void
3658 odp_key_from_pkt_metadata(struct ofpbuf *buf, const struct pkt_metadata *md)
3659 {
3660 nl_msg_put_u32(buf, OVS_KEY_ATTR_PRIORITY, md->skb_priority);
3661
3662 if (md->tunnel.ip_dst) {
3663 tun_key_to_attr(buf, &md->tunnel);
3664 }
3665
3666 nl_msg_put_u32(buf, OVS_KEY_ATTR_SKB_MARK, md->pkt_mark);
3667
3668 /* Add an ingress port attribute if 'odp_in_port' is not the magical
3669 * value "ODPP_NONE". */
3670 if (md->in_port.odp_port != ODPP_NONE) {
3671 nl_msg_put_odp_port(buf, OVS_KEY_ATTR_IN_PORT, md->in_port.odp_port);
3672 }
3673 }
3674
3675 /* Generate packet metadata from the given ODP flow key. */
3676 void
3677 odp_key_to_pkt_metadata(const struct nlattr *key, size_t key_len,
3678 struct pkt_metadata *md)
3679 {
3680 const struct nlattr *nla;
3681 size_t left;
3682 uint32_t wanted_attrs = 1u << OVS_KEY_ATTR_PRIORITY |
3683 1u << OVS_KEY_ATTR_SKB_MARK | 1u << OVS_KEY_ATTR_TUNNEL |
3684 1u << OVS_KEY_ATTR_IN_PORT;
3685
3686 *md = PKT_METADATA_INITIALIZER(ODPP_NONE);
3687
3688 NL_ATTR_FOR_EACH (nla, left, key, key_len) {
3689 uint16_t type = nl_attr_type(nla);
3690 size_t len = nl_attr_get_size(nla);
3691 int expected_len = odp_key_attr_len(ovs_flow_key_attr_lens,
3692 OVS_KEY_ATTR_MAX, type);
3693
3694 if (len != expected_len && expected_len >= 0) {
3695 continue;
3696 }
3697
3698 switch (type) {
3699 case OVS_KEY_ATTR_RECIRC_ID:
3700 md->recirc_id = nl_attr_get_u32(nla);
3701 wanted_attrs &= ~(1u << OVS_KEY_ATTR_RECIRC_ID);
3702 break;
3703 case OVS_KEY_ATTR_DP_HASH:
3704 md->dp_hash = nl_attr_get_u32(nla);
3705 wanted_attrs &= ~(1u << OVS_KEY_ATTR_DP_HASH);
3706 break;
3707 case OVS_KEY_ATTR_PRIORITY:
3708 md->skb_priority = nl_attr_get_u32(nla);
3709 wanted_attrs &= ~(1u << OVS_KEY_ATTR_PRIORITY);
3710 break;
3711 case OVS_KEY_ATTR_SKB_MARK:
3712 md->pkt_mark = nl_attr_get_u32(nla);
3713 wanted_attrs &= ~(1u << OVS_KEY_ATTR_SKB_MARK);
3714 break;
3715 case OVS_KEY_ATTR_TUNNEL: {
3716 enum odp_key_fitness res;
3717
3718 res = odp_tun_key_from_attr(nla, &md->tunnel);
3719 if (res == ODP_FIT_ERROR) {
3720 memset(&md->tunnel, 0, sizeof md->tunnel);
3721 } else if (res == ODP_FIT_PERFECT) {
3722 wanted_attrs &= ~(1u << OVS_KEY_ATTR_TUNNEL);
3723 }
3724 break;
3725 }
3726 case OVS_KEY_ATTR_IN_PORT:
3727 md->in_port.odp_port = nl_attr_get_odp_port(nla);
3728 wanted_attrs &= ~(1u << OVS_KEY_ATTR_IN_PORT);
3729 break;
3730 default:
3731 break;
3732 }
3733
3734 if (!wanted_attrs) {
3735 return; /* Have everything. */
3736 }
3737 }
3738 }
3739
3740 uint32_t
3741 odp_flow_key_hash(const struct nlattr *key, size_t key_len)
3742 {
3743 BUILD_ASSERT_DECL(!(NLA_ALIGNTO % sizeof(uint32_t)));
3744 return hash_words(ALIGNED_CAST(const uint32_t *, key),
3745 key_len / sizeof(uint32_t), 0);
3746 }
3747
3748 static void
3749 log_odp_key_attributes(struct vlog_rate_limit *rl, const char *title,
3750 uint64_t attrs, int out_of_range_attr,
3751 const struct nlattr *key, size_t key_len)
3752 {
3753 struct ds s;
3754 int i;
3755
3756 if (VLOG_DROP_DBG(rl)) {
3757 return;
3758 }
3759
3760 ds_init(&s);
3761 for (i = 0; i < 64; i++) {
3762 if (attrs & (UINT64_C(1) << i)) {
3763 char namebuf[OVS_KEY_ATTR_BUFSIZE];
3764
3765 ds_put_format(&s, " %s",
3766 ovs_key_attr_to_string(i, namebuf, sizeof namebuf));
3767 }
3768 }
3769 if (out_of_range_attr) {
3770 ds_put_format(&s, " %d (and possibly others)", out_of_range_attr);
3771 }
3772
3773 ds_put_cstr(&s, ": ");
3774 odp_flow_key_format(key, key_len, &s);
3775
3776 VLOG_DBG("%s:%s", title, ds_cstr(&s));
3777 ds_destroy(&s);
3778 }
3779
3780 static uint8_t
3781 odp_to_ovs_frag(uint8_t odp_frag, bool is_mask)
3782 {
3783 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3784
3785 if (is_mask) {
3786 return odp_frag ? FLOW_NW_FRAG_MASK : 0;
3787 }
3788
3789 if (odp_frag > OVS_FRAG_TYPE_LATER) {
3790 VLOG_ERR_RL(&rl, "invalid frag %"PRIu8" in flow key", odp_frag);
3791 return 0xff; /* Error. */
3792 }
3793
3794 return (odp_frag == OVS_FRAG_TYPE_NONE) ? 0
3795 : (odp_frag == OVS_FRAG_TYPE_FIRST) ? FLOW_NW_FRAG_ANY
3796 : FLOW_NW_FRAG_ANY | FLOW_NW_FRAG_LATER;
3797 }
3798
3799 static bool
3800 parse_flow_nlattrs(const struct nlattr *key, size_t key_len,
3801 const struct nlattr *attrs[], uint64_t *present_attrsp,
3802 int *out_of_range_attrp)
3803 {
3804 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3805 const struct nlattr *nla;
3806 uint64_t present_attrs;
3807 size_t left;
3808
3809 BUILD_ASSERT(OVS_KEY_ATTR_MAX < CHAR_BIT * sizeof present_attrs);
3810 present_attrs = 0;
3811 *out_of_range_attrp = 0;
3812 NL_ATTR_FOR_EACH (nla, left, key, key_len) {
3813 uint16_t type = nl_attr_type(nla);
3814 size_t len = nl_attr_get_size(nla);
3815 int expected_len = odp_key_attr_len(ovs_flow_key_attr_lens,
3816 OVS_KEY_ATTR_MAX, type);
3817
3818 if (len != expected_len && expected_len >= 0) {
3819 char namebuf[OVS_KEY_ATTR_BUFSIZE];
3820
3821 VLOG_ERR_RL(&rl, "attribute %s has length %"PRIuSIZE" but should have "
3822 "length %d", ovs_key_attr_to_string(type, namebuf,
3823 sizeof namebuf),
3824 len, expected_len);
3825 return false;
3826 }
3827
3828 if (type > OVS_KEY_ATTR_MAX) {
3829 *out_of_range_attrp = type;
3830 } else {
3831 if (present_attrs & (UINT64_C(1) << type)) {
3832 char namebuf[OVS_KEY_ATTR_BUFSIZE];
3833
3834 VLOG_ERR_RL(&rl, "duplicate %s attribute in flow key",
3835 ovs_key_attr_to_string(type,
3836 namebuf, sizeof namebuf));
3837 return false;
3838 }
3839
3840 present_attrs |= UINT64_C(1) << type;
3841 attrs[type] = nla;
3842 }
3843 }
3844 if (left) {
3845 VLOG_ERR_RL(&rl, "trailing garbage in flow key");
3846 return false;
3847 }
3848
3849 *present_attrsp = present_attrs;
3850 return true;
3851 }
3852
3853 static enum odp_key_fitness
3854 check_expectations(uint64_t present_attrs, int out_of_range_attr,
3855 uint64_t expected_attrs,
3856 const struct nlattr *key, size_t key_len)
3857 {
3858 uint64_t missing_attrs;
3859 uint64_t extra_attrs;
3860
3861 missing_attrs = expected_attrs & ~present_attrs;
3862 if (missing_attrs) {
3863 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3864 log_odp_key_attributes(&rl, "expected but not present",
3865 missing_attrs, 0, key, key_len);
3866 return ODP_FIT_TOO_LITTLE;
3867 }
3868
3869 extra_attrs = present_attrs & ~expected_attrs;
3870 if (extra_attrs || out_of_range_attr) {
3871 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3872 log_odp_key_attributes(&rl, "present but not expected",
3873 extra_attrs, out_of_range_attr, key, key_len);
3874 return ODP_FIT_TOO_MUCH;
3875 }
3876
3877 return ODP_FIT_PERFECT;
3878 }
3879
3880 static bool
3881 parse_ethertype(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
3882 uint64_t present_attrs, uint64_t *expected_attrs,
3883 struct flow *flow, const struct flow *src_flow)
3884 {
3885 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3886 bool is_mask = flow != src_flow;
3887
3888 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE)) {
3889 flow->dl_type = nl_attr_get_be16(attrs[OVS_KEY_ATTR_ETHERTYPE]);
3890 if (!is_mask && ntohs(flow->dl_type) < ETH_TYPE_MIN) {
3891 VLOG_ERR_RL(&rl, "invalid Ethertype %"PRIu16" in flow key",
3892 ntohs(flow->dl_type));
3893 return false;
3894 }
3895 if (is_mask && ntohs(src_flow->dl_type) < ETH_TYPE_MIN &&
3896 flow->dl_type != htons(0xffff)) {
3897 return false;
3898 }
3899 *expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ETHERTYPE;
3900 } else {
3901 if (!is_mask) {
3902 flow->dl_type = htons(FLOW_DL_TYPE_NONE);
3903 } else if (ntohs(src_flow->dl_type) < ETH_TYPE_MIN) {
3904 /* See comments in odp_flow_key_from_flow__(). */
3905 VLOG_ERR_RL(&rl, "mask expected for non-Ethernet II frame");
3906 return false;
3907 }
3908 }
3909 return true;
3910 }
3911
3912 static enum odp_key_fitness
3913 parse_l2_5_onward(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
3914 uint64_t present_attrs, int out_of_range_attr,
3915 uint64_t expected_attrs, struct flow *flow,
3916 const struct nlattr *key, size_t key_len,
3917 const struct flow *src_flow)
3918 {
3919 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3920 bool is_mask = src_flow != flow;
3921 const void *check_start = NULL;
3922 size_t check_len = 0;
3923 enum ovs_key_attr expected_bit = 0xff;
3924
3925 if (eth_type_mpls(src_flow->dl_type)) {
3926 if (!is_mask || present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_MPLS)) {
3927 expected_attrs |= (UINT64_C(1) << OVS_KEY_ATTR_MPLS);
3928 }
3929 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_MPLS)) {
3930 size_t size = nl_attr_get_size(attrs[OVS_KEY_ATTR_MPLS]);
3931 const ovs_be32 *mpls_lse = nl_attr_get(attrs[OVS_KEY_ATTR_MPLS]);
3932 int n = size / sizeof(ovs_be32);
3933 int i;
3934
3935 if (!size || size % sizeof(ovs_be32)) {
3936 return ODP_FIT_ERROR;
3937 }
3938 if (flow->mpls_lse[0] && flow->dl_type != htons(0xffff)) {
3939 return ODP_FIT_ERROR;
3940 }
3941
3942 for (i = 0; i < n && i < FLOW_MAX_MPLS_LABELS; i++) {
3943 flow->mpls_lse[i] = mpls_lse[i];
3944 }
3945 if (n > FLOW_MAX_MPLS_LABELS) {
3946 return ODP_FIT_TOO_MUCH;
3947 }
3948
3949 if (!is_mask) {
3950 /* BOS may be set only in the innermost label. */
3951 for (i = 0; i < n - 1; i++) {
3952 if (flow->mpls_lse[i] & htonl(MPLS_BOS_MASK)) {
3953 return ODP_FIT_ERROR;
3954 }
3955 }
3956
3957 /* BOS must be set in the innermost label. */
3958 if (n < FLOW_MAX_MPLS_LABELS
3959 && !(flow->mpls_lse[n - 1] & htonl(MPLS_BOS_MASK))) {
3960 return ODP_FIT_TOO_LITTLE;
3961 }
3962 }
3963 }
3964
3965 goto done;
3966 } else if (src_flow->dl_type == htons(ETH_TYPE_IP)) {
3967 if (!is_mask) {
3968 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IPV4;
3969 }
3970 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IPV4)) {
3971 const struct ovs_key_ipv4 *ipv4_key;
3972
3973 ipv4_key = nl_attr_get(attrs[OVS_KEY_ATTR_IPV4]);
3974 put_ipv4_key(ipv4_key, flow, is_mask);
3975 if (flow->nw_frag > FLOW_NW_FRAG_MASK) {
3976 return ODP_FIT_ERROR;
3977 }
3978 if (is_mask) {
3979 check_start = ipv4_key;
3980 check_len = sizeof *ipv4_key;
3981 expected_bit = OVS_KEY_ATTR_IPV4;
3982 }
3983 }
3984 } else if (src_flow->dl_type == htons(ETH_TYPE_IPV6)) {
3985 if (!is_mask) {
3986 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IPV6;
3987 }
3988 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IPV6)) {
3989 const struct ovs_key_ipv6 *ipv6_key;
3990
3991 ipv6_key = nl_attr_get(attrs[OVS_KEY_ATTR_IPV6]);
3992 put_ipv6_key(ipv6_key, flow, is_mask);
3993 if (flow->nw_frag > FLOW_NW_FRAG_MASK) {
3994 return ODP_FIT_ERROR;
3995 }
3996 if (is_mask) {
3997 check_start = ipv6_key;
3998 check_len = sizeof *ipv6_key;
3999 expected_bit = OVS_KEY_ATTR_IPV6;
4000 }
4001 }
4002 } else if (src_flow->dl_type == htons(ETH_TYPE_ARP) ||
4003 src_flow->dl_type == htons(ETH_TYPE_RARP)) {
4004 if (!is_mask) {
4005 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ARP;
4006 }
4007 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ARP)) {
4008 const struct ovs_key_arp *arp_key;
4009
4010 arp_key = nl_attr_get(attrs[OVS_KEY_ATTR_ARP]);
4011 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
4012 VLOG_ERR_RL(&rl, "unsupported ARP opcode %"PRIu16" in flow "
4013 "key", ntohs(arp_key->arp_op));
4014 return ODP_FIT_ERROR;
4015 }
4016 put_arp_key(arp_key, flow);
4017 if (is_mask) {
4018 check_start = arp_key;
4019 check_len = sizeof *arp_key;
4020 expected_bit = OVS_KEY_ATTR_ARP;
4021 }
4022 }
4023 } else {
4024 goto done;
4025 }
4026 if (check_len > 0) { /* Happens only when 'is_mask'. */
4027 if (!is_all_zeros(check_start, check_len) &&
4028 flow->dl_type != htons(0xffff)) {
4029 return ODP_FIT_ERROR;
4030 } else {
4031 expected_attrs |= UINT64_C(1) << expected_bit;
4032 }
4033 }
4034
4035 expected_bit = OVS_KEY_ATTR_UNSPEC;
4036 if (src_flow->nw_proto == IPPROTO_TCP
4037 && (src_flow->dl_type == htons(ETH_TYPE_IP) ||
4038 src_flow->dl_type == htons(ETH_TYPE_IPV6))
4039 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4040 if (!is_mask) {
4041 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_TCP;
4042 }
4043 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_TCP)) {
4044 const union ovs_key_tp *tcp_key;
4045
4046 tcp_key = nl_attr_get(attrs[OVS_KEY_ATTR_TCP]);
4047 put_tp_key(tcp_key, flow);
4048 expected_bit = OVS_KEY_ATTR_TCP;
4049 }
4050 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_TCP_FLAGS)) {
4051 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_TCP_FLAGS;
4052 flow->tcp_flags = nl_attr_get_be16(attrs[OVS_KEY_ATTR_TCP_FLAGS]);
4053 }
4054 } else if (src_flow->nw_proto == IPPROTO_UDP
4055 && (src_flow->dl_type == htons(ETH_TYPE_IP) ||
4056 src_flow->dl_type == htons(ETH_TYPE_IPV6))
4057 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4058 if (!is_mask) {
4059 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_UDP;
4060 }
4061 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_UDP)) {
4062 const union ovs_key_tp *udp_key;
4063
4064 udp_key = nl_attr_get(attrs[OVS_KEY_ATTR_UDP]);
4065 put_tp_key(udp_key, flow);
4066 expected_bit = OVS_KEY_ATTR_UDP;
4067 }
4068 } else if (src_flow->nw_proto == IPPROTO_SCTP
4069 && (src_flow->dl_type == htons(ETH_TYPE_IP) ||
4070 src_flow->dl_type == htons(ETH_TYPE_IPV6))
4071 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4072 if (!is_mask) {
4073 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_SCTP;
4074 }
4075 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_SCTP)) {
4076 const union ovs_key_tp *sctp_key;
4077
4078 sctp_key = nl_attr_get(attrs[OVS_KEY_ATTR_SCTP]);
4079 put_tp_key(sctp_key, flow);
4080 expected_bit = OVS_KEY_ATTR_SCTP;
4081 }
4082 } else if (src_flow->nw_proto == IPPROTO_ICMP
4083 && src_flow->dl_type == htons(ETH_TYPE_IP)
4084 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4085 if (!is_mask) {
4086 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ICMP;
4087 }
4088 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ICMP)) {
4089 const struct ovs_key_icmp *icmp_key;
4090
4091 icmp_key = nl_attr_get(attrs[OVS_KEY_ATTR_ICMP]);
4092 flow->tp_src = htons(icmp_key->icmp_type);
4093 flow->tp_dst = htons(icmp_key->icmp_code);
4094 expected_bit = OVS_KEY_ATTR_ICMP;
4095 }
4096 } else if (src_flow->nw_proto == IPPROTO_ICMPV6
4097 && src_flow->dl_type == htons(ETH_TYPE_IPV6)
4098 && !(src_flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4099 if (!is_mask) {
4100 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ICMPV6;
4101 }
4102 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ICMPV6)) {
4103 const struct ovs_key_icmpv6 *icmpv6_key;
4104
4105 icmpv6_key = nl_attr_get(attrs[OVS_KEY_ATTR_ICMPV6]);
4106 flow->tp_src = htons(icmpv6_key->icmpv6_type);
4107 flow->tp_dst = htons(icmpv6_key->icmpv6_code);
4108 expected_bit = OVS_KEY_ATTR_ICMPV6;
4109 if (src_flow->tp_dst == htons(0) &&
4110 (src_flow->tp_src == htons(ND_NEIGHBOR_SOLICIT) ||
4111 src_flow->tp_src == htons(ND_NEIGHBOR_ADVERT))) {
4112 if (!is_mask) {
4113 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ND;
4114 }
4115 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ND)) {
4116 const struct ovs_key_nd *nd_key;
4117
4118 nd_key = nl_attr_get(attrs[OVS_KEY_ATTR_ND]);
4119 memcpy(&flow->nd_target, nd_key->nd_target,
4120 sizeof flow->nd_target);
4121 memcpy(flow->arp_sha, nd_key->nd_sll, ETH_ADDR_LEN);
4122 memcpy(flow->arp_tha, nd_key->nd_tll, ETH_ADDR_LEN);
4123 if (is_mask) {
4124 if (!is_all_zeros(nd_key, sizeof *nd_key) &&
4125 (flow->tp_src != htons(0xffff) ||
4126 flow->tp_dst != htons(0xffff))) {
4127 return ODP_FIT_ERROR;
4128 } else {
4129 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ND;
4130 }
4131 }
4132 }
4133 }
4134 }
4135 }
4136 if (is_mask && expected_bit != OVS_KEY_ATTR_UNSPEC) {
4137 if ((flow->tp_src || flow->tp_dst) && flow->nw_proto != 0xff) {
4138 return ODP_FIT_ERROR;
4139 } else {
4140 expected_attrs |= UINT64_C(1) << expected_bit;
4141 }
4142 }
4143
4144 done:
4145 return check_expectations(present_attrs, out_of_range_attr, expected_attrs,
4146 key, key_len);
4147 }
4148
4149 /* Parse 802.1Q header then encapsulated L3 attributes. */
4150 static enum odp_key_fitness
4151 parse_8021q_onward(const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1],
4152 uint64_t present_attrs, int out_of_range_attr,
4153 uint64_t expected_attrs, struct flow *flow,
4154 const struct nlattr *key, size_t key_len,
4155 const struct flow *src_flow)
4156 {
4157 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4158 bool is_mask = src_flow != flow;
4159
4160 const struct nlattr *encap
4161 = (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ENCAP)
4162 ? attrs[OVS_KEY_ATTR_ENCAP] : NULL);
4163 enum odp_key_fitness encap_fitness;
4164 enum odp_key_fitness fitness;
4165
4166 /* Calculate fitness of outer attributes. */
4167 if (!is_mask) {
4168 expected_attrs |= ((UINT64_C(1) << OVS_KEY_ATTR_VLAN) |
4169 (UINT64_C(1) << OVS_KEY_ATTR_ENCAP));
4170 } else {
4171 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN)) {
4172 expected_attrs |= (UINT64_C(1) << OVS_KEY_ATTR_VLAN);
4173 }
4174 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ENCAP)) {
4175 expected_attrs |= (UINT64_C(1) << OVS_KEY_ATTR_ENCAP);
4176 }
4177 }
4178 fitness = check_expectations(present_attrs, out_of_range_attr,
4179 expected_attrs, key, key_len);
4180
4181 /* Set vlan_tci.
4182 * Remove the TPID from dl_type since it's not the real Ethertype. */
4183 flow->dl_type = htons(0);
4184 flow->vlan_tci = (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN)
4185 ? nl_attr_get_be16(attrs[OVS_KEY_ATTR_VLAN])
4186 : htons(0));
4187 if (!is_mask) {
4188 if (!(present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN))) {
4189 return ODP_FIT_TOO_LITTLE;
4190 } else if (flow->vlan_tci == htons(0)) {
4191 /* Corner case for a truncated 802.1Q header. */
4192 if (fitness == ODP_FIT_PERFECT && nl_attr_get_size(encap)) {
4193 return ODP_FIT_TOO_MUCH;
4194 }
4195 return fitness;
4196 } else if (!(flow->vlan_tci & htons(VLAN_CFI))) {
4197 VLOG_ERR_RL(&rl, "OVS_KEY_ATTR_VLAN 0x%04"PRIx16" is nonzero "
4198 "but CFI bit is not set", ntohs(flow->vlan_tci));
4199 return ODP_FIT_ERROR;
4200 }
4201 } else {
4202 if (!(present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ENCAP))) {
4203 return fitness;
4204 }
4205 }
4206
4207 /* Now parse the encapsulated attributes. */
4208 if (!parse_flow_nlattrs(nl_attr_get(encap), nl_attr_get_size(encap),
4209 attrs, &present_attrs, &out_of_range_attr)) {
4210 return ODP_FIT_ERROR;
4211 }
4212 expected_attrs = 0;
4213
4214 if (!parse_ethertype(attrs, present_attrs, &expected_attrs, flow, src_flow)) {
4215 return ODP_FIT_ERROR;
4216 }
4217 encap_fitness = parse_l2_5_onward(attrs, present_attrs, out_of_range_attr,
4218 expected_attrs, flow, key, key_len,
4219 src_flow);
4220
4221 /* The overall fitness is the worse of the outer and inner attributes. */
4222 return MAX(fitness, encap_fitness);
4223 }
4224
4225 static enum odp_key_fitness
4226 odp_flow_key_to_flow__(const struct nlattr *key, size_t key_len,
4227 struct flow *flow, const struct flow *src_flow)
4228 {
4229 const struct nlattr *attrs[OVS_KEY_ATTR_MAX + 1];
4230 uint64_t expected_attrs;
4231 uint64_t present_attrs;
4232 int out_of_range_attr;
4233 bool is_mask = src_flow != flow;
4234
4235 memset(flow, 0, sizeof *flow);
4236
4237 /* Parse attributes. */
4238 if (!parse_flow_nlattrs(key, key_len, attrs, &present_attrs,
4239 &out_of_range_attr)) {
4240 return ODP_FIT_ERROR;
4241 }
4242 expected_attrs = 0;
4243
4244 /* Metadata. */
4245 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_RECIRC_ID)) {
4246 flow->recirc_id = nl_attr_get_u32(attrs[OVS_KEY_ATTR_RECIRC_ID]);
4247 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_RECIRC_ID;
4248 } else if (is_mask) {
4249 /* Always exact match recirc_id if it is not specified. */
4250 flow->recirc_id = UINT32_MAX;
4251 }
4252
4253 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_DP_HASH)) {
4254 flow->dp_hash = nl_attr_get_u32(attrs[OVS_KEY_ATTR_DP_HASH]);
4255 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_DP_HASH;
4256 }
4257 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_PRIORITY)) {
4258 flow->skb_priority = nl_attr_get_u32(attrs[OVS_KEY_ATTR_PRIORITY]);
4259 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_PRIORITY;
4260 }
4261
4262 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_SKB_MARK)) {
4263 flow->pkt_mark = nl_attr_get_u32(attrs[OVS_KEY_ATTR_SKB_MARK]);
4264 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_SKB_MARK;
4265 }
4266
4267 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_TUNNEL)) {
4268 enum odp_key_fitness res;
4269
4270 res = odp_tun_key_from_attr(attrs[OVS_KEY_ATTR_TUNNEL], &flow->tunnel);
4271 if (res == ODP_FIT_ERROR) {
4272 return ODP_FIT_ERROR;
4273 } else if (res == ODP_FIT_PERFECT) {
4274 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_TUNNEL;
4275 }
4276 }
4277
4278 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_IN_PORT)) {
4279 flow->in_port.odp_port
4280 = nl_attr_get_odp_port(attrs[OVS_KEY_ATTR_IN_PORT]);
4281 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_IN_PORT;
4282 } else if (!is_mask) {
4283 flow->in_port.odp_port = ODPP_NONE;
4284 }
4285
4286 /* Ethernet header. */
4287 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_ETHERNET)) {
4288 const struct ovs_key_ethernet *eth_key;
4289
4290 eth_key = nl_attr_get(attrs[OVS_KEY_ATTR_ETHERNET]);
4291 put_ethernet_key(eth_key, flow);
4292 if (is_mask) {
4293 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ETHERNET;
4294 }
4295 }
4296 if (!is_mask) {
4297 expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_ETHERNET;
4298 }
4299
4300 /* Get Ethertype or 802.1Q TPID or FLOW_DL_TYPE_NONE. */
4301 if (!parse_ethertype(attrs, present_attrs, &expected_attrs, flow,
4302 src_flow)) {
4303 return ODP_FIT_ERROR;
4304 }
4305
4306 if (is_mask
4307 ? (src_flow->vlan_tci & htons(VLAN_CFI)) != 0
4308 : src_flow->dl_type == htons(ETH_TYPE_VLAN)) {
4309 return parse_8021q_onward(attrs, present_attrs, out_of_range_attr,
4310 expected_attrs, flow, key, key_len, src_flow);
4311 }
4312 if (is_mask) {
4313 flow->vlan_tci = htons(0xffff);
4314 if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_VLAN)) {
4315 flow->vlan_tci = nl_attr_get_be16(attrs[OVS_KEY_ATTR_VLAN]);
4316 expected_attrs |= (UINT64_C(1) << OVS_KEY_ATTR_VLAN);
4317 }
4318 }
4319 return parse_l2_5_onward(attrs, present_attrs, out_of_range_attr,
4320 expected_attrs, flow, key, key_len, src_flow);
4321 }
4322
4323 /* Converts the 'key_len' bytes of OVS_KEY_ATTR_* attributes in 'key' to a flow
4324 * structure in 'flow'. Returns an ODP_FIT_* value that indicates how well
4325 * 'key' fits our expectations for what a flow key should contain.
4326 *
4327 * The 'in_port' will be the datapath's understanding of the port. The
4328 * caller will need to translate with odp_port_to_ofp_port() if the
4329 * OpenFlow port is needed.
4330 *
4331 * This function doesn't take the packet itself as an argument because none of
4332 * the currently understood OVS_KEY_ATTR_* attributes require it. Currently,
4333 * it is always possible to infer which additional attribute(s) should appear
4334 * by looking at the attributes for lower-level protocols, e.g. if the network
4335 * protocol in OVS_KEY_ATTR_IPV4 or OVS_KEY_ATTR_IPV6 is IPPROTO_TCP then we
4336 * know that a OVS_KEY_ATTR_TCP attribute must appear and that otherwise it
4337 * must be absent. */
4338 enum odp_key_fitness
4339 odp_flow_key_to_flow(const struct nlattr *key, size_t key_len,
4340 struct flow *flow)
4341 {
4342 return odp_flow_key_to_flow__(key, key_len, flow, flow);
4343 }
4344
4345 /* Converts the 'key_len' bytes of OVS_KEY_ATTR_* attributes in 'key' to a mask
4346 * structure in 'mask'. 'flow' must be a previously translated flow
4347 * corresponding to 'mask'. Returns an ODP_FIT_* value that indicates how well
4348 * 'key' fits our expectations for what a flow key should contain. */
4349 enum odp_key_fitness
4350 odp_flow_key_to_mask(const struct nlattr *key, size_t key_len,
4351 struct flow *mask, const struct flow *flow)
4352 {
4353 return odp_flow_key_to_flow__(key, key_len, mask, flow);
4354 }
4355
4356 /* Returns 'fitness' as a string, for use in debug messages. */
4357 const char *
4358 odp_key_fitness_to_string(enum odp_key_fitness fitness)
4359 {
4360 switch (fitness) {
4361 case ODP_FIT_PERFECT:
4362 return "OK";
4363 case ODP_FIT_TOO_MUCH:
4364 return "too_much";
4365 case ODP_FIT_TOO_LITTLE:
4366 return "too_little";
4367 case ODP_FIT_ERROR:
4368 return "error";
4369 default:
4370 return "<unknown>";
4371 }
4372 }
4373
4374 /* Appends an OVS_ACTION_ATTR_USERSPACE action to 'odp_actions' that specifies
4375 * Netlink PID 'pid'. If 'userdata' is nonnull, adds a userdata attribute
4376 * whose contents are the 'userdata_size' bytes at 'userdata' and returns the
4377 * offset within 'odp_actions' of the start of the cookie. (If 'userdata' is
4378 * null, then the return value is not meaningful.) */
4379 size_t
4380 odp_put_userspace_action(uint32_t pid,
4381 const void *userdata, size_t userdata_size,
4382 odp_port_t tunnel_out_port,
4383 struct ofpbuf *odp_actions)
4384 {
4385 size_t userdata_ofs;
4386 size_t offset;
4387
4388 offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_USERSPACE);
4389 nl_msg_put_u32(odp_actions, OVS_USERSPACE_ATTR_PID, pid);
4390 if (userdata) {
4391 userdata_ofs = odp_actions->size + NLA_HDRLEN;
4392
4393 /* The OVS kernel module before OVS 1.11 and the upstream Linux kernel
4394 * module before Linux 3.10 required the userdata to be exactly 8 bytes
4395 * long:
4396 *
4397 * - The kernel rejected shorter userdata with -ERANGE.
4398 *
4399 * - The kernel silently dropped userdata beyond the first 8 bytes.
4400 *
4401 * Thus, for maximum compatibility, always put at least 8 bytes. (We
4402 * separately disable features that required more than 8 bytes.) */
4403 memcpy(nl_msg_put_unspec_zero(odp_actions, OVS_USERSPACE_ATTR_USERDATA,
4404 MAX(8, userdata_size)),
4405 userdata, userdata_size);
4406 } else {
4407 userdata_ofs = 0;
4408 }
4409 if (tunnel_out_port != ODPP_NONE) {
4410 nl_msg_put_odp_port(odp_actions, OVS_USERSPACE_ATTR_EGRESS_TUN_PORT,
4411 tunnel_out_port);
4412 }
4413 nl_msg_end_nested(odp_actions, offset);
4414
4415 return userdata_ofs;
4416 }
4417
4418 void
4419 odp_put_tunnel_action(const struct flow_tnl *tunnel,
4420 struct ofpbuf *odp_actions)
4421 {
4422 size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
4423 tun_key_to_attr(odp_actions, tunnel);
4424 nl_msg_end_nested(odp_actions, offset);
4425 }
4426
4427 void
4428 odp_put_tnl_push_action(struct ofpbuf *odp_actions,
4429 struct ovs_action_push_tnl *data)
4430 {
4431 int size = offsetof(struct ovs_action_push_tnl, header);
4432
4433 size += data->header_len;
4434 nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_TUNNEL_PUSH, data, size);
4435 }
4436
4437 \f
4438 /* The commit_odp_actions() function and its helpers. */
4439
4440 static void
4441 commit_set_action(struct ofpbuf *odp_actions, enum ovs_key_attr key_type,
4442 const void *key, size_t key_size)
4443 {
4444 size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
4445 nl_msg_put_unspec(odp_actions, key_type, key, key_size);
4446 nl_msg_end_nested(odp_actions, offset);
4447 }
4448
4449 /* Masked set actions have a mask following the data within the netlink
4450 * attribute. The unmasked bits in the data will be cleared as the data
4451 * is copied to the action. */
4452 void
4453 commit_masked_set_action(struct ofpbuf *odp_actions,
4454 enum ovs_key_attr key_type,
4455 const void *key_, const void *mask_, size_t key_size)
4456 {
4457 size_t offset = nl_msg_start_nested(odp_actions,
4458 OVS_ACTION_ATTR_SET_MASKED);
4459 char *data = nl_msg_put_unspec_uninit(odp_actions, key_type, key_size * 2);
4460 const char *key = key_, *mask = mask_;
4461
4462 memcpy(data + key_size, mask, key_size);
4463 /* Clear unmasked bits while copying. */
4464 while (key_size--) {
4465 *data++ = *key++ & *mask++;
4466 }
4467 nl_msg_end_nested(odp_actions, offset);
4468 }
4469
4470 /* If any of the flow key data that ODP actions can modify are different in
4471 * 'base->tunnel' and 'flow->tunnel', appends a set_tunnel ODP action to
4472 * 'odp_actions' that change the flow tunneling information in key from
4473 * 'base->tunnel' into 'flow->tunnel', and then changes 'base->tunnel' in the
4474 * same way. In other words, operates the same as commit_odp_actions(), but
4475 * only on tunneling information. */
4476 void
4477 commit_odp_tunnel_action(const struct flow *flow, struct flow *base,
4478 struct ofpbuf *odp_actions)
4479 {
4480 /* A valid IPV4_TUNNEL must have non-zero ip_dst. */
4481 if (flow->tunnel.ip_dst) {
4482 if (!memcmp(&base->tunnel, &flow->tunnel, sizeof base->tunnel)) {
4483 return;
4484 }
4485 memcpy(&base->tunnel, &flow->tunnel, sizeof base->tunnel);
4486 odp_put_tunnel_action(&base->tunnel, odp_actions);
4487 }
4488 }
4489
4490 static bool
4491 commit(enum ovs_key_attr attr, bool use_masked_set,
4492 const void *key, void *base, void *mask, size_t size,
4493 struct ofpbuf *odp_actions)
4494 {
4495 if (memcmp(key, base, size)) {
4496 bool fully_masked = odp_mask_is_exact(attr, mask, size);
4497
4498 if (use_masked_set && !fully_masked) {
4499 commit_masked_set_action(odp_actions, attr, key, mask, size);
4500 } else {
4501 if (!fully_masked) {
4502 memset(mask, 0xff, size);
4503 }
4504 commit_set_action(odp_actions, attr, key, size);
4505 }
4506 memcpy(base, key, size);
4507 return true;
4508 } else {
4509 /* Mask bits are set when we have either read or set the corresponding
4510 * values. Masked bits will be exact-matched, no need to set them
4511 * if the value did not actually change. */
4512 return false;
4513 }
4514 }
4515
4516 static void
4517 get_ethernet_key(const struct flow *flow, struct ovs_key_ethernet *eth)
4518 {
4519 memcpy(eth->eth_src, flow->dl_src, ETH_ADDR_LEN);
4520 memcpy(eth->eth_dst, flow->dl_dst, ETH_ADDR_LEN);
4521 }
4522
4523 static void
4524 put_ethernet_key(const struct ovs_key_ethernet *eth, struct flow *flow)
4525 {
4526 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
4527 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
4528 }
4529
4530 static void
4531 commit_set_ether_addr_action(const struct flow *flow, struct flow *base_flow,
4532 struct ofpbuf *odp_actions,
4533 struct flow_wildcards *wc,
4534 bool use_masked)
4535 {
4536 struct ovs_key_ethernet key, base, mask;
4537
4538 get_ethernet_key(flow, &key);
4539 get_ethernet_key(base_flow, &base);
4540 get_ethernet_key(&wc->masks, &mask);
4541
4542 if (commit(OVS_KEY_ATTR_ETHERNET, use_masked,
4543 &key, &base, &mask, sizeof key, odp_actions)) {
4544 put_ethernet_key(&base, base_flow);
4545 put_ethernet_key(&mask, &wc->masks);
4546 }
4547 }
4548
4549 static void
4550 pop_vlan(struct flow *base,
4551 struct ofpbuf *odp_actions, struct flow_wildcards *wc)
4552 {
4553 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4554
4555 if (base->vlan_tci & htons(VLAN_CFI)) {
4556 nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_VLAN);
4557 base->vlan_tci = 0;
4558 }
4559 }
4560
4561 static void
4562 commit_vlan_action(ovs_be16 vlan_tci, struct flow *base,
4563 struct ofpbuf *odp_actions, struct flow_wildcards *wc)
4564 {
4565 if (base->vlan_tci == vlan_tci) {
4566 return;
4567 }
4568
4569 pop_vlan(base, odp_actions, wc);
4570 if (vlan_tci & htons(VLAN_CFI)) {
4571 struct ovs_action_push_vlan vlan;
4572
4573 vlan.vlan_tpid = htons(ETH_TYPE_VLAN);
4574 vlan.vlan_tci = vlan_tci;
4575 nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_PUSH_VLAN,
4576 &vlan, sizeof vlan);
4577 }
4578 base->vlan_tci = vlan_tci;
4579 }
4580
4581 /* Wildcarding already done at action translation time. */
4582 static void
4583 commit_mpls_action(const struct flow *flow, struct flow *base,
4584 struct ofpbuf *odp_actions)
4585 {
4586 int base_n = flow_count_mpls_labels(base, NULL);
4587 int flow_n = flow_count_mpls_labels(flow, NULL);
4588 int common_n = flow_count_common_mpls_labels(flow, flow_n, base, base_n,
4589 NULL);
4590
4591 while (base_n > common_n) {
4592 if (base_n - 1 == common_n && flow_n > common_n) {
4593 /* If there is only one more LSE in base than there are common
4594 * between base and flow; and flow has at least one more LSE than
4595 * is common then the topmost LSE of base may be updated using
4596 * set */
4597 struct ovs_key_mpls mpls_key;
4598
4599 mpls_key.mpls_lse = flow->mpls_lse[flow_n - base_n];
4600 commit_set_action(odp_actions, OVS_KEY_ATTR_MPLS,
4601 &mpls_key, sizeof mpls_key);
4602 flow_set_mpls_lse(base, 0, mpls_key.mpls_lse);
4603 common_n++;
4604 } else {
4605 /* Otherwise, if there more LSEs in base than are common between
4606 * base and flow then pop the topmost one. */
4607 ovs_be16 dl_type;
4608 bool popped;
4609
4610 /* If all the LSEs are to be popped and this is not the outermost
4611 * LSE then use ETH_TYPE_MPLS as the ethertype parameter of the
4612 * POP_MPLS action instead of flow->dl_type.
4613 *
4614 * This is because the POP_MPLS action requires its ethertype
4615 * argument to be an MPLS ethernet type but in this case
4616 * flow->dl_type will be a non-MPLS ethernet type.
4617 *
4618 * When the final POP_MPLS action occurs it use flow->dl_type and
4619 * the and the resulting packet will have the desired dl_type. */
4620 if ((!eth_type_mpls(flow->dl_type)) && base_n > 1) {
4621 dl_type = htons(ETH_TYPE_MPLS);
4622 } else {
4623 dl_type = flow->dl_type;
4624 }
4625 nl_msg_put_be16(odp_actions, OVS_ACTION_ATTR_POP_MPLS, dl_type);
4626 popped = flow_pop_mpls(base, base_n, flow->dl_type, NULL);
4627 ovs_assert(popped);
4628 base_n--;
4629 }
4630 }
4631
4632 /* If, after the above popping and setting, there are more LSEs in flow
4633 * than base then some LSEs need to be pushed. */
4634 while (base_n < flow_n) {
4635 struct ovs_action_push_mpls *mpls;
4636
4637 mpls = nl_msg_put_unspec_zero(odp_actions,
4638 OVS_ACTION_ATTR_PUSH_MPLS,
4639 sizeof *mpls);
4640 mpls->mpls_ethertype = flow->dl_type;
4641 mpls->mpls_lse = flow->mpls_lse[flow_n - base_n - 1];
4642 flow_push_mpls(base, base_n, mpls->mpls_ethertype, NULL);
4643 flow_set_mpls_lse(base, 0, mpls->mpls_lse);
4644 base_n++;
4645 }
4646 }
4647
4648 static void
4649 get_ipv4_key(const struct flow *flow, struct ovs_key_ipv4 *ipv4, bool is_mask)
4650 {
4651 ipv4->ipv4_src = flow->nw_src;
4652 ipv4->ipv4_dst = flow->nw_dst;
4653 ipv4->ipv4_proto = flow->nw_proto;
4654 ipv4->ipv4_tos = flow->nw_tos;
4655 ipv4->ipv4_ttl = flow->nw_ttl;
4656 ipv4->ipv4_frag = ovs_to_odp_frag(flow->nw_frag, is_mask);
4657 }
4658
4659 static void
4660 put_ipv4_key(const struct ovs_key_ipv4 *ipv4, struct flow *flow, bool is_mask)
4661 {
4662 flow->nw_src = ipv4->ipv4_src;
4663 flow->nw_dst = ipv4->ipv4_dst;
4664 flow->nw_proto = ipv4->ipv4_proto;
4665 flow->nw_tos = ipv4->ipv4_tos;
4666 flow->nw_ttl = ipv4->ipv4_ttl;
4667 flow->nw_frag = odp_to_ovs_frag(ipv4->ipv4_frag, is_mask);
4668 }
4669
4670 static void
4671 commit_set_ipv4_action(const struct flow *flow, struct flow *base_flow,
4672 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4673 bool use_masked)
4674 {
4675 struct ovs_key_ipv4 key, mask, base;
4676
4677 /* Check that nw_proto and nw_frag remain unchanged. */
4678 ovs_assert(flow->nw_proto == base_flow->nw_proto &&
4679 flow->nw_frag == base_flow->nw_frag);
4680
4681 get_ipv4_key(flow, &key, false);
4682 get_ipv4_key(base_flow, &base, false);
4683 get_ipv4_key(&wc->masks, &mask, true);
4684 mask.ipv4_proto = 0; /* Not writeable. */
4685 mask.ipv4_frag = 0; /* Not writable. */
4686
4687 if (commit(OVS_KEY_ATTR_IPV4, use_masked, &key, &base, &mask, sizeof key,
4688 odp_actions)) {
4689 put_ipv4_key(&base, base_flow, false);
4690 if (mask.ipv4_proto != 0) { /* Mask was changed by commit(). */
4691 put_ipv4_key(&mask, &wc->masks, true);
4692 }
4693 }
4694 }
4695
4696 static void
4697 get_ipv6_key(const struct flow *flow, struct ovs_key_ipv6 *ipv6, bool is_mask)
4698 {
4699 memcpy(ipv6->ipv6_src, &flow->ipv6_src, sizeof ipv6->ipv6_src);
4700 memcpy(ipv6->ipv6_dst, &flow->ipv6_dst, sizeof ipv6->ipv6_dst);
4701 ipv6->ipv6_label = flow->ipv6_label;
4702 ipv6->ipv6_proto = flow->nw_proto;
4703 ipv6->ipv6_tclass = flow->nw_tos;
4704 ipv6->ipv6_hlimit = flow->nw_ttl;
4705 ipv6->ipv6_frag = ovs_to_odp_frag(flow->nw_frag, is_mask);
4706 }
4707
4708 static void
4709 put_ipv6_key(const struct ovs_key_ipv6 *ipv6, struct flow *flow, bool is_mask)
4710 {
4711 memcpy(&flow->ipv6_src, ipv6->ipv6_src, sizeof flow->ipv6_src);
4712 memcpy(&flow->ipv6_dst, ipv6->ipv6_dst, sizeof flow->ipv6_dst);
4713 flow->ipv6_label = ipv6->ipv6_label;
4714 flow->nw_proto = ipv6->ipv6_proto;
4715 flow->nw_tos = ipv6->ipv6_tclass;
4716 flow->nw_ttl = ipv6->ipv6_hlimit;
4717 flow->nw_frag = odp_to_ovs_frag(ipv6->ipv6_frag, is_mask);
4718 }
4719
4720 static void
4721 commit_set_ipv6_action(const struct flow *flow, struct flow *base_flow,
4722 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4723 bool use_masked)
4724 {
4725 struct ovs_key_ipv6 key, mask, base;
4726
4727 /* Check that nw_proto and nw_frag remain unchanged. */
4728 ovs_assert(flow->nw_proto == base_flow->nw_proto &&
4729 flow->nw_frag == base_flow->nw_frag);
4730
4731 get_ipv6_key(flow, &key, false);
4732 get_ipv6_key(base_flow, &base, false);
4733 get_ipv6_key(&wc->masks, &mask, true);
4734 mask.ipv6_proto = 0; /* Not writeable. */
4735 mask.ipv6_frag = 0; /* Not writable. */
4736
4737 if (commit(OVS_KEY_ATTR_IPV6, use_masked, &key, &base, &mask, sizeof key,
4738 odp_actions)) {
4739 put_ipv6_key(&base, base_flow, false);
4740 if (mask.ipv6_proto != 0) { /* Mask was changed by commit(). */
4741 put_ipv6_key(&mask, &wc->masks, true);
4742 }
4743 }
4744 }
4745
4746 static void
4747 get_arp_key(const struct flow *flow, struct ovs_key_arp *arp)
4748 {
4749 /* ARP key has padding, clear it. */
4750 memset(arp, 0, sizeof *arp);
4751
4752 arp->arp_sip = flow->nw_src;
4753 arp->arp_tip = flow->nw_dst;
4754 arp->arp_op = htons(flow->nw_proto);
4755 memcpy(arp->arp_sha, flow->arp_sha, ETH_ADDR_LEN);
4756 memcpy(arp->arp_tha, flow->arp_tha, ETH_ADDR_LEN);
4757 }
4758
4759 static void
4760 put_arp_key(const struct ovs_key_arp *arp, struct flow *flow)
4761 {
4762 flow->nw_src = arp->arp_sip;
4763 flow->nw_dst = arp->arp_tip;
4764 flow->nw_proto = ntohs(arp->arp_op);
4765 memcpy(flow->arp_sha, arp->arp_sha, ETH_ADDR_LEN);
4766 memcpy(flow->arp_tha, arp->arp_tha, ETH_ADDR_LEN);
4767 }
4768
4769 static enum slow_path_reason
4770 commit_set_arp_action(const struct flow *flow, struct flow *base_flow,
4771 struct ofpbuf *odp_actions, struct flow_wildcards *wc)
4772 {
4773 struct ovs_key_arp key, mask, base;
4774
4775 get_arp_key(flow, &key);
4776 get_arp_key(base_flow, &base);
4777 get_arp_key(&wc->masks, &mask);
4778
4779 if (commit(OVS_KEY_ATTR_ARP, true, &key, &base, &mask, sizeof key,
4780 odp_actions)) {
4781 put_arp_key(&base, base_flow);
4782 put_arp_key(&mask, &wc->masks);
4783 return SLOW_ACTION;
4784 }
4785 return 0;
4786 }
4787
4788 static void
4789 get_nd_key(const struct flow *flow, struct ovs_key_nd *nd)
4790 {
4791 memcpy(nd->nd_target, &flow->nd_target, sizeof flow->nd_target);
4792 /* nd_sll and nd_tll are stored in arp_sha and arp_tha, respectively */
4793 memcpy(nd->nd_sll, flow->arp_sha, ETH_ADDR_LEN);
4794 memcpy(nd->nd_tll, flow->arp_tha, ETH_ADDR_LEN);
4795 }
4796
4797 static void
4798 put_nd_key(const struct ovs_key_nd *nd, struct flow *flow)
4799 {
4800 memcpy(&flow->nd_target, &flow->nd_target, sizeof flow->nd_target);
4801 /* nd_sll and nd_tll are stored in arp_sha and arp_tha, respectively */
4802 memcpy(flow->arp_sha, nd->nd_sll, ETH_ADDR_LEN);
4803 memcpy(flow->arp_tha, nd->nd_tll, ETH_ADDR_LEN);
4804 }
4805
4806 static enum slow_path_reason
4807 commit_set_nd_action(const struct flow *flow, struct flow *base_flow,
4808 struct ofpbuf *odp_actions,
4809 struct flow_wildcards *wc, bool use_masked)
4810 {
4811 struct ovs_key_nd key, mask, base;
4812
4813 get_nd_key(flow, &key);
4814 get_nd_key(base_flow, &base);
4815 get_nd_key(&wc->masks, &mask);
4816
4817 if (commit(OVS_KEY_ATTR_ND, use_masked, &key, &base, &mask, sizeof key,
4818 odp_actions)) {
4819 put_nd_key(&base, base_flow);
4820 put_nd_key(&mask, &wc->masks);
4821 return SLOW_ACTION;
4822 }
4823
4824 return 0;
4825 }
4826
4827 static enum slow_path_reason
4828 commit_set_nw_action(const struct flow *flow, struct flow *base,
4829 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4830 bool use_masked)
4831 {
4832 /* Check if 'flow' really has an L3 header. */
4833 if (!flow->nw_proto) {
4834 return 0;
4835 }
4836
4837 switch (ntohs(base->dl_type)) {
4838 case ETH_TYPE_IP:
4839 commit_set_ipv4_action(flow, base, odp_actions, wc, use_masked);
4840 break;
4841
4842 case ETH_TYPE_IPV6:
4843 commit_set_ipv6_action(flow, base, odp_actions, wc, use_masked);
4844 return commit_set_nd_action(flow, base, odp_actions, wc, use_masked);
4845
4846 case ETH_TYPE_ARP:
4847 return commit_set_arp_action(flow, base, odp_actions, wc);
4848 }
4849
4850 return 0;
4851 }
4852
4853 /* TCP, UDP, and SCTP keys have the same layout. */
4854 BUILD_ASSERT_DECL(sizeof(struct ovs_key_tcp) == sizeof(struct ovs_key_udp) &&
4855 sizeof(struct ovs_key_tcp) == sizeof(struct ovs_key_sctp));
4856
4857 static void
4858 get_tp_key(const struct flow *flow, union ovs_key_tp *tp)
4859 {
4860 tp->tcp.tcp_src = flow->tp_src;
4861 tp->tcp.tcp_dst = flow->tp_dst;
4862 }
4863
4864 static void
4865 put_tp_key(const union ovs_key_tp *tp, struct flow *flow)
4866 {
4867 flow->tp_src = tp->tcp.tcp_src;
4868 flow->tp_dst = tp->tcp.tcp_dst;
4869 }
4870
4871 static void
4872 commit_set_port_action(const struct flow *flow, struct flow *base_flow,
4873 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4874 bool use_masked)
4875 {
4876 enum ovs_key_attr key_type;
4877 union ovs_key_tp key, mask, base;
4878
4879 /* Check if 'flow' really has an L3 header. */
4880 if (!flow->nw_proto) {
4881 return;
4882 }
4883
4884 if (!is_ip_any(base_flow)) {
4885 return;
4886 }
4887
4888 if (flow->nw_proto == IPPROTO_TCP) {
4889 key_type = OVS_KEY_ATTR_TCP;
4890 } else if (flow->nw_proto == IPPROTO_UDP) {
4891 key_type = OVS_KEY_ATTR_UDP;
4892 } else if (flow->nw_proto == IPPROTO_SCTP) {
4893 key_type = OVS_KEY_ATTR_SCTP;
4894 } else {
4895 return;
4896 }
4897
4898 get_tp_key(flow, &key);
4899 get_tp_key(base_flow, &base);
4900 get_tp_key(&wc->masks, &mask);
4901
4902 if (commit(key_type, use_masked, &key, &base, &mask, sizeof key,
4903 odp_actions)) {
4904 put_tp_key(&base, base_flow);
4905 put_tp_key(&mask, &wc->masks);
4906 }
4907 }
4908
4909 static void
4910 commit_set_priority_action(const struct flow *flow, struct flow *base_flow,
4911 struct ofpbuf *odp_actions,
4912 struct flow_wildcards *wc,
4913 bool use_masked)
4914 {
4915 uint32_t key, mask, base;
4916
4917 key = flow->skb_priority;
4918 base = base_flow->skb_priority;
4919 mask = wc->masks.skb_priority;
4920
4921 if (commit(OVS_KEY_ATTR_PRIORITY, use_masked, &key, &base, &mask,
4922 sizeof key, odp_actions)) {
4923 base_flow->skb_priority = base;
4924 wc->masks.skb_priority = mask;
4925 }
4926 }
4927
4928 static void
4929 commit_set_pkt_mark_action(const struct flow *flow, struct flow *base_flow,
4930 struct ofpbuf *odp_actions,
4931 struct flow_wildcards *wc,
4932 bool use_masked)
4933 {
4934 uint32_t key, mask, base;
4935
4936 key = flow->pkt_mark;
4937 base = base_flow->pkt_mark;
4938 mask = wc->masks.pkt_mark;
4939
4940 if (commit(OVS_KEY_ATTR_SKB_MARK, use_masked, &key, &base, &mask,
4941 sizeof key, odp_actions)) {
4942 base_flow->pkt_mark = base;
4943 wc->masks.pkt_mark = mask;
4944 }
4945 }
4946
4947 /* If any of the flow key data that ODP actions can modify are different in
4948 * 'base' and 'flow', appends ODP actions to 'odp_actions' that change the flow
4949 * key from 'base' into 'flow', and then changes 'base' the same way. Does not
4950 * commit set_tunnel actions. Users should call commit_odp_tunnel_action()
4951 * in addition to this function if needed. Sets fields in 'wc' that are
4952 * used as part of the action.
4953 *
4954 * Returns a reason to force processing the flow's packets into the userspace
4955 * slow path, if there is one, otherwise 0. */
4956 enum slow_path_reason
4957 commit_odp_actions(const struct flow *flow, struct flow *base,
4958 struct ofpbuf *odp_actions, struct flow_wildcards *wc,
4959 bool use_masked)
4960 {
4961 enum slow_path_reason slow;
4962
4963 commit_set_ether_addr_action(flow, base, odp_actions, wc, use_masked);
4964 slow = commit_set_nw_action(flow, base, odp_actions, wc, use_masked);
4965 commit_set_port_action(flow, base, odp_actions, wc, use_masked);
4966 commit_mpls_action(flow, base, odp_actions);
4967 commit_vlan_action(flow->vlan_tci, base, odp_actions, wc);
4968 commit_set_priority_action(flow, base, odp_actions, wc, use_masked);
4969 commit_set_pkt_mark_action(flow, base, odp_actions, wc, use_masked);
4970
4971 return slow;
4972 }
openvswitch packages for PVE