]> git.proxmox.com Git - ovs.git/blame - datapath/flow_netlink.c
projects / ovs.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
datapath: do not use vport type to determine presence of Geneve attributes
[ovs.git] / datapath / flow_netlink.c
CommitLineData
a097c0b2 1/*
05499369 2 * Copyright (c) 2007-2014 Nicira, Inc.
a097c0b2
PS		 3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
0a0857df
JP		 19#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
a097c0b2
PS		 21#include "flow.h"
22#include "datapath.h"
ccf43786 23#include "mpls.h"
a097c0b2
PS		 24#include <linux/uaccess.h>
25#include <linux/netdevice.h>
26#include <linux/etherdevice.h>
27#include <linux/if_ether.h>
28#include <linux/if_vlan.h>
29#include <net/llc_pdu.h>
30#include <linux/kernel.h>
31#include <linux/jhash.h>
32#include <linux/jiffies.h>
33#include <linux/llc.h>
34#include <linux/module.h>
35#include <linux/in.h>
36#include <linux/rcupdate.h>
37#include <linux/if_arp.h>
38#include <linux/ip.h>
39#include <linux/ipv6.h>
40#include <linux/sctp.h>
41#include <linux/tcp.h>
42#include <linux/udp.h>
43#include <linux/icmp.h>
44#include <linux/icmpv6.h>
45#include <linux/rculist.h>
c1fc1411 46#include <net/geneve.h>
a097c0b2 47#include <net/ip.h>
29c71cfa 48#include <net/ip_tunnels.h>
a097c0b2
PS		 49#include <net/ipv6.h>
50#include <net/ndisc.h>
51
52#include "flow_netlink.h"
53
54static void update_range__(struct sw_flow_match *match,
55 size_t offset, size_t size, bool is_mask)
56{
57 struct sw_flow_key_range *range = NULL;
58 size_t start = rounddown(offset, sizeof(long));
59 size_t end = roundup(offset + size, sizeof(long));
60
61 if (!is_mask)
62 range = &match->range;
63 else if (match->mask)
64 range = &match->mask->range;
65
66 if (!range)
67 return;
68
69 if (range->start == range->end) {
70 range->start = start;
71 range->end = end;
72 return;
73 }
74
75 if (range->start > start)
76 range->start = start;
77
78 if (range->end < end)
79 range->end = end;
80}
81
82#define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
83 do { \
84 update_range__(match, offsetof(struct sw_flow_key, field), \
85 sizeof((match)->key->field), is_mask); \
86 if (is_mask) { \
87 if ((match)->mask) \
88 (match)->mask->key.field = value; \
89 } else { \
90 (match)->key->field = value; \
91 } \
92 } while (0)
93
c1fc1411 94#define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask) \
a097c0b2 95 do { \
c1fc1411 96 update_range__(match, offset, len, is_mask); \
a097c0b2
PS		 97 if (is_mask) { \
98 if ((match)->mask) \
c1fc1411 99 memcpy((u8 *)&(match)->mask->key + offset, value_p, len);\
a097c0b2 100 } else { \
c1fc1411 101 memcpy((u8 *)(match)->key + offset, value_p, len); \
a097c0b2
PS		 102 } \
103 } while (0)
104
c1fc1411
JG		 105#define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
106 SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
107 value_p, len, is_mask)
108
0b496cda
DDP		 109#define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask) \
110 do { \
111 update_range__(match, offsetof(struct sw_flow_key, field), \
112 sizeof((match)->key->field), is_mask); \
113 if (is_mask) { \
114 if ((match)->mask) \
115 memset((u8 *)&(match)->mask->key.field, value,\
116 sizeof((match)->mask->key.field)); \
117 } else { \
118 memset((u8 *)&(match)->key->field, value, \
119 sizeof((match)->key->field)); \
120 } \
121 } while (0)
122
a097c0b2
PS		 123static bool match_validate(const struct sw_flow_match *match,
124 u64 key_attrs, u64 mask_attrs)
125{
126 u64 key_expected = 1ULL << OVS_KEY_ATTR_ETHERNET;
127 u64 mask_allowed = key_attrs; /* At most allow all key attributes */
128
129 /* The following mask attributes allowed only if they
130 * pass the validation tests. */
131 mask_allowed &= ~((1ULL << OVS_KEY_ATTR_IPV4)
132 | (1ULL << OVS_KEY_ATTR_IPV6)
133 | (1ULL << OVS_KEY_ATTR_TCP)
dc235f7f 134 | (1ULL << OVS_KEY_ATTR_TCP_FLAGS)
a097c0b2
PS		 135 | (1ULL << OVS_KEY_ATTR_UDP)
136 | (1ULL << OVS_KEY_ATTR_SCTP)
137 | (1ULL << OVS_KEY_ATTR_ICMP)
138 | (1ULL << OVS_KEY_ATTR_ICMPV6)
139 | (1ULL << OVS_KEY_ATTR_ARP)
ccf43786
SH		 140 | (1ULL << OVS_KEY_ATTR_ND)
141 | (1ULL << OVS_KEY_ATTR_MPLS));
a097c0b2
PS		 142
143 /* Always allowed mask fields. */
144 mask_allowed |= ((1ULL << OVS_KEY_ATTR_TUNNEL)
145 | (1ULL << OVS_KEY_ATTR_IN_PORT)
146 | (1ULL << OVS_KEY_ATTR_ETHERTYPE));
147
148 /* Check key attributes. */
149 if (match->key->eth.type == htons(ETH_P_ARP)
150 || match->key->eth.type == htons(ETH_P_RARP)) {
151 key_expected |= 1ULL << OVS_KEY_ATTR_ARP;
152 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
153 mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
154 }
155
ccf43786
SH		 156
157 if (eth_p_mpls(match->key->eth.type)) {
158 key_expected |= 1ULL << OVS_KEY_ATTR_MPLS;
159 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
160 mask_allowed |= 1ULL << OVS_KEY_ATTR_MPLS;
161 }
162
a097c0b2
PS		 163 if (match->key->eth.type == htons(ETH_P_IP)) {
164 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
165 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
166 mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV4;
167
168 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
169 if (match->key->ip.proto == IPPROTO_UDP) {
170 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
171 if (match->mask && (match->mask->key.ip.proto == 0xff))
172 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
173 }
174
175 if (match->key->ip.proto == IPPROTO_SCTP) {
176 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
177 if (match->mask && (match->mask->key.ip.proto == 0xff))
178 mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
179 }
180
181 if (match->key->ip.proto == IPPROTO_TCP) {
182 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
dc235f7f
JR		 183 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
184 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
a097c0b2 185 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
dc235f7f
JR		 186 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
187 }
a097c0b2
PS		 188 }
189
190 if (match->key->ip.proto == IPPROTO_ICMP) {
191 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
192 if (match->mask && (match->mask->key.ip.proto == 0xff))
193 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
194 }
195 }
196 }
197
198 if (match->key->eth.type == htons(ETH_P_IPV6)) {
199 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
200 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
201 mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
202
203 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
204 if (match->key->ip.proto == IPPROTO_UDP) {
205 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
206 if (match->mask && (match->mask->key.ip.proto == 0xff))
207 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
208 }
209
210 if (match->key->ip.proto == IPPROTO_SCTP) {
211 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
212 if (match->mask && (match->mask->key.ip.proto == 0xff))
213 mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
214 }
215
216 if (match->key->ip.proto == IPPROTO_TCP) {
217 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
dc235f7f
JR		 218 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
219 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
a097c0b2 220 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
dc235f7f
JR		 221 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
222 }
a097c0b2
PS		 223 }
224
225 if (match->key->ip.proto == IPPROTO_ICMPV6) {
226 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
227 if (match->mask && (match->mask->key.ip.proto == 0xff))
228 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
229
708fb4c5 230 if (match->key->tp.src ==
a097c0b2 231 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
708fb4c5 232 match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
a097c0b2 233 key_expected |= 1ULL << OVS_KEY_ATTR_ND;
560f3099 234 if (match->mask && (match->mask->key.tp.src == htons(0xff)))
a097c0b2
PS		 235 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
236 }
237 }
238 }
239 }
240
241 if ((key_attrs & key_expected) != key_expected) {
242 /* Key attributes check failed. */
243 OVS_NLERR("Missing expected key attributes (key_attrs=%llx, expected=%llx).\n",
29dd5cb7 244 (unsigned long long)key_attrs, (unsigned long long)key_expected);
a097c0b2
PS		 245 return false;
246 }
247
248 if ((mask_attrs & mask_allowed) != mask_attrs) {
249 /* Mask attributes check failed. */
250 OVS_NLERR("Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).\n",
29dd5cb7 251 (unsigned long long)mask_attrs, (unsigned long long)mask_allowed);
a097c0b2
PS		 252 return false;
253 }
254
255 return true;
256}
257
258/* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute. */
259static const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
260 [OVS_KEY_ATTR_ENCAP] = -1,
261 [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
262 [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
263 [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
264 [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
265 [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
266 [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
267 [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
268 [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
269 [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
dc235f7f 270 [OVS_KEY_ATTR_TCP_FLAGS] = sizeof(__be16),
a097c0b2
PS		 271 [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
272 [OVS_KEY_ATTR_SCTP] = sizeof(struct ovs_key_sctp),
273 [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
274 [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
275 [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
276 [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
7804df20 277 [OVS_KEY_ATTR_DP_HASH] = sizeof(u32),
a6059080 278 [OVS_KEY_ATTR_RECIRC_ID] = sizeof(u32),
a097c0b2 279 [OVS_KEY_ATTR_TUNNEL] = -1,
ccf43786 280 [OVS_KEY_ATTR_MPLS] = sizeof(struct ovs_key_mpls),
a097c0b2
PS		 281};
282
283static bool is_all_zero(const u8 *fp, size_t size)
284{
285 int i;
286
287 if (!fp)
288 return false;
289
290 for (i = 0; i < size; i++)
291 if (fp[i])
292 return false;
293
294 return true;
295}
296
297static int __parse_flow_nlattrs(const struct nlattr *attr,
298 const struct nlattr *a[],
299 u64 *attrsp, bool nz)
300{
301 const struct nlattr *nla;
302 u64 attrs;
303 int rem;
304
305 attrs = *attrsp;
306 nla_for_each_nested(nla, attr, rem) {
307 u16 type = nla_type(nla);
308 int expected_len;
309
310 if (type > OVS_KEY_ATTR_MAX) {
311 OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
312 type, OVS_KEY_ATTR_MAX);
313 return -EINVAL;
314 }
315
316 if (attrs & (1ULL << type)) {
317 OVS_NLERR("Duplicate key attribute (type %d).\n", type);
318 return -EINVAL;
319 }
320
321 expected_len = ovs_key_lens[type];
322 if (nla_len(nla) != expected_len && expected_len != -1) {
323 OVS_NLERR("Key attribute has unexpected length (type=%d"
324 ", length=%d, expected=%d).\n", type,
325 nla_len(nla), expected_len);
326 return -EINVAL;
327 }
328
329 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
330 attrs |= 1ULL << type;
331 a[type] = nla;
332 }
333 }
334 if (rem) {
335 OVS_NLERR("Message has %d unknown bytes.\n", rem);
336 return -EINVAL;
337 }
338
339 *attrsp = attrs;
340 return 0;
341}
342
343static int parse_flow_mask_nlattrs(const struct nlattr *attr,
344 const struct nlattr *a[], u64 *attrsp)
345{
346 return __parse_flow_nlattrs(attr, a, attrsp, true);
347}
348
349static int parse_flow_nlattrs(const struct nlattr *attr,
350 const struct nlattr *a[], u64 *attrsp)
351{
352 return __parse_flow_nlattrs(attr, a, attrsp, false);
353}
354
355static int ipv4_tun_from_nlattr(const struct nlattr *attr,
356 struct sw_flow_match *match, bool is_mask)
357{
358 struct nlattr *a;
359 int rem;
360 bool ttl = false;
361 __be16 tun_flags = 0;
362
363 nla_for_each_nested(a, attr, rem) {
364 int type = nla_type(a);
365 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
366 [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
367 [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
368 [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
369 [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
370 [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
371 [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
372 [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
94872594 373 [OVS_TUNNEL_KEY_ATTR_OAM] = 0,
c1fc1411 374 [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS] = -1,
a097c0b2
PS		 375 };
376
377 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
378 OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
379 type, OVS_TUNNEL_KEY_ATTR_MAX);
380 return -EINVAL;
381 }
382
c1fc1411
JG		 383 if (ovs_tunnel_key_lens[type] != nla_len(a) &&
384 ovs_tunnel_key_lens[type] != -1) {
a097c0b2
PS		 385 OVS_NLERR("IPv4 tunnel attribute type has unexpected "
386 " length (type=%d, length=%d, expected=%d).\n",
387 type, nla_len(a), ovs_tunnel_key_lens[type]);
388 return -EINVAL;
389 }
390
391 switch (type) {
392 case OVS_TUNNEL_KEY_ATTR_ID:
393 SW_FLOW_KEY_PUT(match, tun_key.tun_id,
394 nla_get_be64(a), is_mask);
395 tun_flags |= TUNNEL_KEY;
396 break;
397 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
398 SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
399 nla_get_be32(a), is_mask);
400 break;
401 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
402 SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
403 nla_get_be32(a), is_mask);
404 break;
405 case OVS_TUNNEL_KEY_ATTR_TOS:
406 SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
407 nla_get_u8(a), is_mask);
408 break;
409 case OVS_TUNNEL_KEY_ATTR_TTL:
410 SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
411 nla_get_u8(a), is_mask);
412 ttl = true;
413 break;
414 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
415 tun_flags |= TUNNEL_DONT_FRAGMENT;
416 break;
417 case OVS_TUNNEL_KEY_ATTR_CSUM:
418 tun_flags |= TUNNEL_CSUM;
419 break;
94872594
JG		 420 case OVS_TUNNEL_KEY_ATTR_OAM:
421 tun_flags |= TUNNEL_OAM;
422 break;
c1fc1411 423 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
fa6395df 424 tun_flags |= TUNNEL_OPTIONS_PRESENT;
c1fc1411
JG		 425 if (nla_len(a) > sizeof(match->key->tun_opts)) {
426 OVS_NLERR("Geneve option length exceeds "
427 "maximum size (len %d, max %zu).\n",
428 nla_len(a),
429 sizeof(match->key->tun_opts));
430 return -EINVAL;
431 }
432
433 if (nla_len(a) % 4 != 0) {
434 OVS_NLERR("Geneve option length is not "
435 "a multiple of 4 (len %d).\n",
436 nla_len(a));
437 return -EINVAL;
438 }
439
440 /* We need to record the length of the options passed
441 * down, otherwise packets with the same format but
442 * additional options will be silently matched.
443 */
444 if (!is_mask) {
445 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
446 false);
447 } else {
448 /* This is somewhat unusual because it looks at
449 * both the key and mask while parsing the
450 * attributes (and by extension assumes the key
451 * is parsed first). Normally, we would verify
452 * that each is the correct length and that the
453 * attributes line up in the validate function.
454 * However, that is difficult because this is
455 * variable length and we won't have the
456 * information later.
457 */
458 if (match->key->tun_opts_len != nla_len(a)) {
459 OVS_NLERR("Geneve option key length (%d)"
460 " is different from mask length (%d).",
461 match->key->tun_opts_len, nla_len(a));
462 return -EINVAL;
463 }
464
465 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff,
466 true);
467 }
468
469 SW_FLOW_KEY_MEMCPY_OFFSET(match,
470 (unsigned long)GENEVE_OPTS((struct sw_flow_key *)0,
471 nla_len(a)),
472 nla_data(a), nla_len(a), is_mask);
473 break;
a097c0b2 474 default:
a473df5b 475 OVS_NLERR("Unknown IPv4 tunnel attribute (%d).\n", type);
a097c0b2
PS		 476 return -EINVAL;
477 }
478 }
479
480 SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
481
482 if (rem > 0) {
483 OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
484 return -EINVAL;
485 }
486
487 if (!is_mask) {
488 if (!match->key->tun_key.ipv4_dst) {
489 OVS_NLERR("IPv4 tunnel destination address is zero.\n");
490 return -EINVAL;
491 }
492
493 if (!ttl) {
494 OVS_NLERR("IPv4 tunnel TTL not specified.\n");
495 return -EINVAL;
496 }
497 }
498
499 return 0;
500}
501
502static int ipv4_tun_to_nlattr(struct sk_buff *skb,
c1fc1411
JG		 503 const struct ovs_key_ipv4_tunnel *output,
504 const struct geneve_opt *tun_opts,
505 int swkey_tun_opts_len)
a097c0b2
PS		 506{
507 struct nlattr *nla;
508
509 nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
510 if (!nla)
511 return -EMSGSIZE;
512
513 if (output->tun_flags & TUNNEL_KEY &&
514 nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
515 return -EMSGSIZE;
516 if (output->ipv4_src &&
517 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
518 return -EMSGSIZE;
519 if (output->ipv4_dst &&
520 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
521 return -EMSGSIZE;
522 if (output->ipv4_tos &&
523 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
524 return -EMSGSIZE;
525 if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
526 return -EMSGSIZE;
527 if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
528 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
529 return -EMSGSIZE;
530 if ((output->tun_flags & TUNNEL_CSUM) &&
531 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
532 return -EMSGSIZE;
94872594
JG		 533 if ((output->tun_flags & TUNNEL_OAM) &&
534 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
535 return -EMSGSIZE;
c1fc1411
JG		 536 if (tun_opts &&
537 nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
05499369
BP		 538 swkey_tun_opts_len, tun_opts))
539 return -EMSGSIZE;
a097c0b2
PS		 540
541 nla_nest_end(skb, nla);
542 return 0;
543}
544
545
546static int metadata_from_nlattrs(struct sw_flow_match *match, u64 *attrs,
547 const struct nlattr **a, bool is_mask)
548{
7804df20
AZ		 549 if (*attrs & (1ULL << OVS_KEY_ATTR_DP_HASH)) {
550 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
551
552 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
553 *attrs &= ~(1ULL << OVS_KEY_ATTR_DP_HASH);
554 }
555
a6059080
AZ		 556 if (*attrs & (1ULL << OVS_KEY_ATTR_RECIRC_ID)) {
557 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
558
559 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
560 *attrs &= ~(1ULL << OVS_KEY_ATTR_RECIRC_ID);
561 }
562
a097c0b2
PS		 563 if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
564 SW_FLOW_KEY_PUT(match, phy.priority,
565 nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
566 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
567 }
568
569 if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
570 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
571
a473df5b 572 if (is_mask) {
a097c0b2 573 in_port = 0xffffffff; /* Always exact match in_port. */
a473df5b
JG		 574 } else if (in_port >= DP_MAX_PORTS) {
575 OVS_NLERR("Input port (%d) exceeds maximum allowable (%d).\n",
576 in_port, DP_MAX_PORTS);
a097c0b2 577 return -EINVAL;
a473df5b 578 }
a097c0b2
PS		 579
580 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
581 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
582 } else if (!is_mask) {
583 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
584 }
585
586 if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
587 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
588
589 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
590 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
591 }
592 if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
593 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
594 is_mask))
595 return -EINVAL;
596 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
597 }
598 return 0;
599}
600
df65fec1
JR		 601static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
602 const struct nlattr **a, bool is_mask)
a097c0b2
PS		 603{
604 int err;
a097c0b2
PS		 605
606 err = metadata_from_nlattrs(match, &attrs, a, is_mask);
607 if (err)
608 return err;
609
610 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
611 const struct ovs_key_ethernet *eth_key;
612
613 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
614 SW_FLOW_KEY_MEMCPY(match, eth.src,
615 eth_key->eth_src, ETH_ALEN, is_mask);
616 SW_FLOW_KEY_MEMCPY(match, eth.dst,
617 eth_key->eth_dst, ETH_ALEN, is_mask);
618 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
619 }
620
621 if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
622 __be16 tci;
623
624 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
625 if (!(tci & htons(VLAN_TAG_PRESENT))) {
626 if (is_mask)
627 OVS_NLERR("VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.\n");
628 else
629 OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
630
631 return -EINVAL;
632 }
633
634 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
635 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
636 } else if (!is_mask)
637 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
638
639 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
640 __be16 eth_type;
641
642 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
643 if (is_mask) {
644 /* Always exact match EtherType. */
645 eth_type = htons(0xffff);
646 } else if (ntohs(eth_type) < ETH_P_802_3_MIN) {
647 OVS_NLERR("EtherType is less than minimum (type=%x, min=%x).\n",
648 ntohs(eth_type), ETH_P_802_3_MIN);
649 return -EINVAL;
650 }
651
652 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
653 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
654 } else if (!is_mask) {
655 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
656 }
657
658 if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
659 const struct ovs_key_ipv4 *ipv4_key;
660
661 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
662 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
663 OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
664 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
665 return -EINVAL;
666 }
667 SW_FLOW_KEY_PUT(match, ip.proto,
668 ipv4_key->ipv4_proto, is_mask);
669 SW_FLOW_KEY_PUT(match, ip.tos,
670 ipv4_key->ipv4_tos, is_mask);
671 SW_FLOW_KEY_PUT(match, ip.ttl,
672 ipv4_key->ipv4_ttl, is_mask);
673 SW_FLOW_KEY_PUT(match, ip.frag,
674 ipv4_key->ipv4_frag, is_mask);
675 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
676 ipv4_key->ipv4_src, is_mask);
677 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
678 ipv4_key->ipv4_dst, is_mask);
679 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
680 }
681
682 if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
683 const struct ovs_key_ipv6 *ipv6_key;
684
685 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
686 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
687 OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
688 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
689 return -EINVAL;
690 }
691 SW_FLOW_KEY_PUT(match, ipv6.label,
692 ipv6_key->ipv6_label, is_mask);
693 SW_FLOW_KEY_PUT(match, ip.proto,
694 ipv6_key->ipv6_proto, is_mask);
695 SW_FLOW_KEY_PUT(match, ip.tos,
696 ipv6_key->ipv6_tclass, is_mask);
697 SW_FLOW_KEY_PUT(match, ip.ttl,
698 ipv6_key->ipv6_hlimit, is_mask);
699 SW_FLOW_KEY_PUT(match, ip.frag,
700 ipv6_key->ipv6_frag, is_mask);
701 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
702 ipv6_key->ipv6_src,
703 sizeof(match->key->ipv6.addr.src),
704 is_mask);
705 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
706 ipv6_key->ipv6_dst,
707 sizeof(match->key->ipv6.addr.dst),
708 is_mask);
709
710 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
711 }
712
713 if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
714 const struct ovs_key_arp *arp_key;
715
716 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
717 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
718 OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
719 arp_key->arp_op);
720 return -EINVAL;
721 }
722
723 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
724 arp_key->arp_sip, is_mask);
725 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
726 arp_key->arp_tip, is_mask);
727 SW_FLOW_KEY_PUT(match, ip.proto,
728 ntohs(arp_key->arp_op), is_mask);
729 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
730 arp_key->arp_sha, ETH_ALEN, is_mask);
731 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
732 arp_key->arp_tha, ETH_ALEN, is_mask);
733
734 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
735 }
736
ccf43786
SH		 737 if (attrs & (1ULL << OVS_KEY_ATTR_MPLS)) {
738 const struct ovs_key_mpls *mpls_key;
739
740 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
741 SW_FLOW_KEY_PUT(match, mpls.top_lse,
742 mpls_key->mpls_lse, is_mask);
743
744 attrs &= ~(1ULL << OVS_KEY_ATTR_MPLS);
62974663 745 }
ccf43786 746
a097c0b2
PS		 747 if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
748 const struct ovs_key_tcp *tcp_key;
749
750 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
708fb4c5
JR		 751 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
752 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
a097c0b2
PS		 753 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
754 }
755
dc235f7f 756 if (attrs & (1ULL << OVS_KEY_ATTR_TCP_FLAGS)) {
cab29271
JS		 757 SW_FLOW_KEY_PUT(match, tp.flags,
758 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
759 is_mask);
dc235f7f
JR		 760 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP_FLAGS);
761 }
762
a097c0b2
PS		 763 if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
764 const struct ovs_key_udp *udp_key;
765
766 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
708fb4c5
JR		 767 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
768 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
a097c0b2
PS		 769 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
770 }
771
772 if (attrs & (1ULL << OVS_KEY_ATTR_SCTP)) {
773 const struct ovs_key_sctp *sctp_key;
774
775 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
708fb4c5
JR		 776 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
777 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
a097c0b2
PS		 778 attrs &= ~(1ULL << OVS_KEY_ATTR_SCTP);
779 }
780
781 if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
782 const struct ovs_key_icmp *icmp_key;
783
784 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
708fb4c5 785 SW_FLOW_KEY_PUT(match, tp.src,
a097c0b2 786 htons(icmp_key->icmp_type), is_mask);
708fb4c5 787 SW_FLOW_KEY_PUT(match, tp.dst,
a097c0b2
PS		 788 htons(icmp_key->icmp_code), is_mask);
789 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
790 }
791
792 if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
793 const struct ovs_key_icmpv6 *icmpv6_key;
794
795 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
708fb4c5 796 SW_FLOW_KEY_PUT(match, tp.src,
a097c0b2 797 htons(icmpv6_key->icmpv6_type), is_mask);
708fb4c5 798 SW_FLOW_KEY_PUT(match, tp.dst,
a097c0b2
PS		 799 htons(icmpv6_key->icmpv6_code), is_mask);
800 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
801 }
802
803 if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
804 const struct ovs_key_nd *nd_key;
805
806 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
807 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
808 nd_key->nd_target,
809 sizeof(match->key->ipv6.nd.target),
810 is_mask);
811 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
812 nd_key->nd_sll, ETH_ALEN, is_mask);
813 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
814 nd_key->nd_tll, ETH_ALEN, is_mask);
815 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
816 }
817
a473df5b
JG		 818 if (attrs != 0) {
819 OVS_NLERR("Unknown key attributes (%llx).\n",
820 (unsigned long long)attrs);
a097c0b2 821 return -EINVAL;
a473df5b 822 }
a097c0b2
PS		 823
824 return 0;
825}
826
62974663 827static void nlattr_set(struct nlattr *attr, u8 val, bool is_attr_mask_key)
a097c0b2 828{
62974663
DDP		 829 struct nlattr *nla;
830 int rem;
a097c0b2 831
62974663
DDP		 832 /* The nlattr stream should already have been validated */
833 nla_for_each_nested(nla, attr, rem) {
834 /* We assume that ovs_key_lens[type] == -1 means that type is a
835 * nested attribute
836 */
837 if (is_attr_mask_key && ovs_key_lens[nla_type(nla)] == -1)
838 nlattr_set(nla, val, false);
839 else
840 memset(nla_data(nla), val, nla_len(nla));
841 }
842}
843
844static void mask_set_nlattr(struct nlattr *attr, u8 val)
845{
846 nlattr_set(attr, val, true);
a097c0b2
PS		 847}
848
849/**
850 * ovs_nla_get_match - parses Netlink attributes into a flow key and
851 * mask. In case the 'mask' is NULL, the flow is treated as exact match
852 * flow. Otherwise, it is treated as a wildcarded flow, except the mask
853 * does not include any don't care bit.
854 * @match: receives the extracted flow match information.
855 * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
856 * sequence. The fields should of the packet that triggered the creation
857 * of this flow.
858 * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
859 * attribute specifies the mask field of the wildcarded flow.
860 */
861int ovs_nla_get_match(struct sw_flow_match *match,
862 const struct nlattr *key,
863 const struct nlattr *mask)
864{
865 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
866 const struct nlattr *encap;
62974663 867 struct nlattr *newmask = NULL;
a097c0b2
PS		 868 u64 key_attrs = 0;
869 u64 mask_attrs = 0;
870 bool encap_valid = false;
871 int err;
872
873 err = parse_flow_nlattrs(key, a, &key_attrs);
874 if (err)
875 return err;
876
877 if ((key_attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
878 (key_attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) &&
879 (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
880 __be16 tci;
881
882 if (!((key_attrs & (1ULL << OVS_KEY_ATTR_VLAN)) &&
883 (key_attrs & (1ULL << OVS_KEY_ATTR_ENCAP)))) {
884 OVS_NLERR("Invalid Vlan frame.\n");
885 return -EINVAL;
886 }
887
888 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
889 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
890 encap = a[OVS_KEY_ATTR_ENCAP];
891 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
892 encap_valid = true;
893
894 if (tci & htons(VLAN_TAG_PRESENT)) {
895 err = parse_flow_nlattrs(encap, a, &key_attrs);
896 if (err)
897 return err;
898 } else if (!tci) {
899 /* Corner case for truncated 802.1Q header. */
900 if (nla_len(encap)) {
901 OVS_NLERR("Truncated 802.1Q header has non-zero encap attribute.\n");
902 return -EINVAL;
903 }
904 } else {
905 OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
906 return -EINVAL;
907 }
908 }
909
df65fec1 910 err = ovs_key_from_nlattrs(match, key_attrs, a, false);
a097c0b2
PS		 911 if (err)
912 return err;
913
62974663
DDP		 914 if (match->mask && !mask) {
915 /* Create an exact match mask. We need to set to 0xff all the
916 * 'match->mask' fields that have been touched in 'match->key'.
917 * We cannot simply memset 'match->mask', because padding bytes
918 * and fields not specified in 'match->key' should be left to 0.
919 * Instead, we use a stream of netlink attributes, copied from
920 * 'key' and set to 0xff: ovs_key_from_nlattrs() will take care
921 * of filling 'match->mask' appropriately.
922 */
923 newmask = kmemdup(key, nla_total_size(nla_len(key)),
924 GFP_KERNEL);
925 if (!newmask)
926 return -ENOMEM;
927
928 mask_set_nlattr(newmask, 0xff);
929
0b496cda
DDP		 930 /* The userspace does not send tunnel attributes that are 0,
931 * but we should not wildcard them nonetheless. */
932 if (match->key->tun_key.ipv4_dst)
933 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key, 0xff, true);
934
62974663
DDP		 935 mask = newmask;
936 }
937
a097c0b2
PS		 938 if (mask) {
939 err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
940 if (err)
62974663 941 goto free_newmask;
a097c0b2 942
62974663 943 if (mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
a097c0b2
PS		 944 __be16 eth_type = 0;
945 __be16 tci = 0;
946
947 if (!encap_valid) {
948 OVS_NLERR("Encap mask attribute is set for non-VLAN frame.\n");
62974663
DDP		 949 err = -EINVAL;
950 goto free_newmask;
a097c0b2
PS		 951 }
952
953 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
954 if (a[OVS_KEY_ATTR_ETHERTYPE])
955 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
956
957 if (eth_type == htons(0xffff)) {
958 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
959 encap = a[OVS_KEY_ATTR_ENCAP];
960 err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
3854ab21
AW		 961 if (err)
962 goto free_newmask;
a097c0b2
PS		 963 } else {
964 OVS_NLERR("VLAN frames must have an exact match on the TPID (mask=%x).\n",
965 ntohs(eth_type));
62974663
DDP		 966 err = -EINVAL;
967 goto free_newmask;
a097c0b2
PS		 968 }
969
970 if (a[OVS_KEY_ATTR_VLAN])
971 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
972
973 if (!(tci & htons(VLAN_TAG_PRESENT))) {
974 OVS_NLERR("VLAN tag present bit must have an exact match (tci_mask=%x).\n", ntohs(tci));
62974663
DDP		 975 err = -EINVAL;
976 goto free_newmask;
a097c0b2
PS		 977 }
978 }
979
df65fec1 980 err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
a097c0b2 981 if (err)
62974663 982 goto free_newmask;
a097c0b2
PS		 983 }
984
985 if (!match_validate(match, key_attrs, mask_attrs))
62974663 986 err = -EINVAL;
a097c0b2 987
62974663
DDP		 988free_newmask:
989 kfree(newmask);
990 return err;
a097c0b2
PS		 991}
992
993/**
994 * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
995 * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
996 * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
997 * sequence.
998 *
999 * This parses a series of Netlink attributes that form a flow key, which must
1000 * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1001 * get the metadata, that is, the parts of the flow key that cannot be
1002 * extracted from the packet itself.
1003 */
1004
1005int ovs_nla_get_flow_metadata(struct sw_flow *flow,
1006 const struct nlattr *attr)
1007{
1008 struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1009 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1010 u64 attrs = 0;
1011 int err;
1012 struct sw_flow_match match;
1013
1014 flow->key.phy.in_port = DP_MAX_PORTS;
1015 flow->key.phy.priority = 0;
1016 flow->key.phy.skb_mark = 0;
7804df20 1017 flow->key.ovs_flow_hash = 0;
a6059080 1018 flow->key.recirc_id = 0;
a097c0b2
PS		 1019 memset(tun_key, 0, sizeof(flow->key.tun_key));
1020
1021 err = parse_flow_nlattrs(attr, a, &attrs);
1022 if (err)
1023 return -EINVAL;
1024
1025 memset(&match, 0, sizeof(match));
1026 match.key = &flow->key;
1027
1028 err = metadata_from_nlattrs(&match, &attrs, a, false);
1029 if (err)
1030 return err;
1031
1032 return 0;
1033}
1034
c1fc1411 1035int ovs_nla_put_flow(struct datapath *dp, const struct sw_flow_key *swkey,
a097c0b2
PS		 1036 const struct sw_flow_key *output, struct sk_buff *skb)
1037{
1038 struct ovs_key_ethernet *eth_key;
1039 struct nlattr *nla, *encap;
1040 bool is_mask = (swkey != output);
1041
7804df20
AZ		 1042 if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1043 goto nla_put_failure;
1044
a6059080
AZ		 1045 if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1046 goto nla_put_failure;
1047
a097c0b2
PS		 1048 if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1049 goto nla_put_failure;
1050
c1fc1411
JG		 1051 if ((swkey->tun_key.ipv4_dst || is_mask)) {
1052 const struct geneve_opt *opts = NULL;
1053
fa6395df
AA		 1054 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1055 opts = GENEVE_OPTS(output, swkey->tun_opts_len);
c1fc1411
JG		 1056
1057 if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
1058 swkey->tun_opts_len))
1059 goto nla_put_failure;
1060 }
a097c0b2
PS		 1061
1062 if (swkey->phy.in_port == DP_MAX_PORTS) {
1063 if (is_mask && (output->phy.in_port == 0xffff))
1064 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1065 goto nla_put_failure;
1066 } else {
1067 u16 upper_u16;
1068 upper_u16 = !is_mask ? 0 : 0xffff;
1069
1070 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1071 (upper_u16 << 16) | output->phy.in_port))
1072 goto nla_put_failure;
1073 }
1074
1075 if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1076 goto nla_put_failure;
1077
1078 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1079 if (!nla)
1080 goto nla_put_failure;
1081
1082 eth_key = nla_data(nla);
982a47ec
JP		 1083 ether_addr_copy(eth_key->eth_src, output->eth.src);
1084 ether_addr_copy(eth_key->eth_dst, output->eth.dst);
a097c0b2
PS		 1085
1086 if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1087 __be16 eth_type;
1088 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1089 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1090 nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1091 goto nla_put_failure;
1092 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1093 if (!swkey->eth.tci)
1094 goto unencap;
1095 } else
1096 encap = NULL;
1097
1098 if (swkey->eth.type == htons(ETH_P_802_2)) {
1099 /*
1100 * Ethertype 802.2 is represented in the netlink with omitted
1101 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1102 * 0xffff in the mask attribute. Ethertype can also
1103 * be wildcarded.
1104 */
1105 if (is_mask && output->eth.type)
1106 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1107 output->eth.type))
1108 goto nla_put_failure;
1109 goto unencap;
1110 }
1111
1112 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1113 goto nla_put_failure;
1114
1115 if (swkey->eth.type == htons(ETH_P_IP)) {
1116 struct ovs_key_ipv4 *ipv4_key;
1117
1118 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1119 if (!nla)
1120 goto nla_put_failure;
1121 ipv4_key = nla_data(nla);
1122 ipv4_key->ipv4_src = output->ipv4.addr.src;
1123 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1124 ipv4_key->ipv4_proto = output->ip.proto;
1125 ipv4_key->ipv4_tos = output->ip.tos;
1126 ipv4_key->ipv4_ttl = output->ip.ttl;
1127 ipv4_key->ipv4_frag = output->ip.frag;
1128 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1129 struct ovs_key_ipv6 *ipv6_key;
1130
1131 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1132 if (!nla)
1133 goto nla_put_failure;
1134 ipv6_key = nla_data(nla);
1135 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1136 sizeof(ipv6_key->ipv6_src));
1137 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1138 sizeof(ipv6_key->ipv6_dst));
1139 ipv6_key->ipv6_label = output->ipv6.label;
1140 ipv6_key->ipv6_proto = output->ip.proto;
1141 ipv6_key->ipv6_tclass = output->ip.tos;
1142 ipv6_key->ipv6_hlimit = output->ip.ttl;
1143 ipv6_key->ipv6_frag = output->ip.frag;
1144 } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1145 swkey->eth.type == htons(ETH_P_RARP)) {
1146 struct ovs_key_arp *arp_key;
1147
1148 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1149 if (!nla)
1150 goto nla_put_failure;
1151 arp_key = nla_data(nla);
1152 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1153 arp_key->arp_sip = output->ipv4.addr.src;
1154 arp_key->arp_tip = output->ipv4.addr.dst;
1155 arp_key->arp_op = htons(output->ip.proto);
982a47ec
JP		 1156 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1157 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
ccf43786
SH		 1158 } else if (eth_p_mpls(swkey->eth.type)) {
1159 struct ovs_key_mpls *mpls_key;
1160
1161 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1162 if (!nla)
1163 goto nla_put_failure;
1164 mpls_key = nla_data(nla);
1165 mpls_key->mpls_lse = output->mpls.top_lse;
a097c0b2
PS		 1166 }
1167
1168 if ((swkey->eth.type == htons(ETH_P_IP) ||
1169 swkey->eth.type == htons(ETH_P_IPV6)) &&
1170 swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1171
1172 if (swkey->ip.proto == IPPROTO_TCP) {
1173 struct ovs_key_tcp *tcp_key;
1174
1175 nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1176 if (!nla)
1177 goto nla_put_failure;
1178 tcp_key = nla_data(nla);
708fb4c5
JR		 1179 tcp_key->tcp_src = output->tp.src;
1180 tcp_key->tcp_dst = output->tp.dst;
1181 if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1182 output->tp.flags))
1183 goto nla_put_failure;
a097c0b2
PS		 1184 } else if (swkey->ip.proto == IPPROTO_UDP) {
1185 struct ovs_key_udp *udp_key;
1186
1187 nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1188 if (!nla)
1189 goto nla_put_failure;
1190 udp_key = nla_data(nla);
708fb4c5
JR		 1191 udp_key->udp_src = output->tp.src;
1192 udp_key->udp_dst = output->tp.dst;
a097c0b2
PS		 1193 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1194 struct ovs_key_sctp *sctp_key;
1195
1196 nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1197 if (!nla)
1198 goto nla_put_failure;
1199 sctp_key = nla_data(nla);
708fb4c5
JR		 1200 sctp_key->sctp_src = output->tp.src;
1201 sctp_key->sctp_dst = output->tp.dst;
a097c0b2
PS		 1202 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1203 swkey->ip.proto == IPPROTO_ICMP) {
1204 struct ovs_key_icmp *icmp_key;
1205
1206 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1207 if (!nla)
1208 goto nla_put_failure;
1209 icmp_key = nla_data(nla);
708fb4c5
JR		 1210 icmp_key->icmp_type = ntohs(output->tp.src);
1211 icmp_key->icmp_code = ntohs(output->tp.dst);
a097c0b2
PS		 1212 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1213 swkey->ip.proto == IPPROTO_ICMPV6) {
1214 struct ovs_key_icmpv6 *icmpv6_key;
1215
1216 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1217 sizeof(*icmpv6_key));
1218 if (!nla)
1219 goto nla_put_failure;
1220 icmpv6_key = nla_data(nla);
708fb4c5
JR		 1221 icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1222 icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
a097c0b2
PS		 1223
1224 if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1225 icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1226 struct ovs_key_nd *nd_key;
1227
1228 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1229 if (!nla)
1230 goto nla_put_failure;
1231 nd_key = nla_data(nla);
1232 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1233 sizeof(nd_key->nd_target));
982a47ec
JP		 1234 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1235 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
a097c0b2
PS		 1236 }
1237 }
1238 }
1239
1240unencap:
1241 if (encap)
1242 nla_nest_end(skb, encap);
1243
1244 return 0;
1245
1246nla_put_failure:
1247 return -EMSGSIZE;
1248}
1249
1250#define MAX_ACTIONS_BUFSIZE (32 * 1024)
1251
1252struct sw_flow_actions *ovs_nla_alloc_flow_actions(int size)
1253{
1254 struct sw_flow_actions *sfa;
1255
a473df5b
JG		 1256 if (size > MAX_ACTIONS_BUFSIZE) {
1257 OVS_NLERR("Flow action size (%u bytes) exceeds maximum "
1258 "(%u bytes)\n", size, MAX_ACTIONS_BUFSIZE);
a097c0b2 1259 return ERR_PTR(-EINVAL);
a473df5b 1260 }
a097c0b2
PS		 1261
1262 sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1263 if (!sfa)
1264 return ERR_PTR(-ENOMEM);
1265
1266 sfa->actions_len = 0;
1267 return sfa;
1268}
1269
1270/* RCU callback used by ovs_nla_free_flow_actions. */
1271static void rcu_free_acts_callback(struct rcu_head *rcu)
1272{
1273 struct sw_flow_actions *sf_acts = container_of(rcu,
1274 struct sw_flow_actions, rcu);
1275 kfree(sf_acts);
1276}
1277
1278/* Schedules 'sf_acts' to be freed after the next RCU grace period.
1279 * The caller must hold rcu_read_lock for this to be sensible. */
1280void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1281{
1282 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
1283}
1284
1285static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1286 int attr_len)
1287{
1288
1289 struct sw_flow_actions *acts;
1290 int new_acts_size;
1291 int req_size = NLA_ALIGN(attr_len);
1292 int next_offset = offsetof(struct sw_flow_actions, actions) +
1293 (*sfa)->actions_len;
1294
1295 if (req_size <= (ksize(*sfa) - next_offset))
1296 goto out;
1297
1298 new_acts_size = ksize(*sfa) * 2;
1299
1300 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1301 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1302 return ERR_PTR(-EMSGSIZE);
1303 new_acts_size = MAX_ACTIONS_BUFSIZE;
1304 }
1305
1306 acts = ovs_nla_alloc_flow_actions(new_acts_size);
1307 if (IS_ERR(acts))
1308 return (void *)acts;
1309
1310 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1311 acts->actions_len = (*sfa)->actions_len;
1312 kfree(*sfa);
1313 *sfa = acts;
1314
1315out:
1316 (*sfa)->actions_len += req_size;
1317 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1318}
1319
f0cd669f
JG		 1320static struct nlattr *__add_action(struct sw_flow_actions **sfa, int attrtype,
1321 void *data, int len)
a097c0b2
PS		 1322{
1323 struct nlattr *a;
1324
1325 a = reserve_sfa_size(sfa, nla_attr_size(len));
1326 if (IS_ERR(a))
f0cd669f 1327 return a;
a097c0b2
PS		 1328
1329 a->nla_type = attrtype;
1330 a->nla_len = nla_attr_size(len);
1331
1332 if (data)
1333 memcpy(nla_data(a), data, len);
1334 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1335
f0cd669f
JG		 1336 return a;
1337}
1338
1339static int add_action(struct sw_flow_actions **sfa, int attrtype,
1340 void *data, int len)
1341{
1342 struct nlattr *a;
1343
1344 a = __add_action(sfa, attrtype, data, len);
1345 if (IS_ERR(a))
1346 return PTR_ERR(a);
1347
a097c0b2
PS		 1348 return 0;
1349}
1350
1351static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1352 int attrtype)
1353{
1354 int used = (*sfa)->actions_len;
1355 int err;
1356
1357 err = add_action(sfa, attrtype, NULL, 0);
1358 if (err)
1359 return err;
1360
1361 return used;
1362}
1363
1364static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1365 int st_offset)
1366{
1367 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1368 st_offset);
1369
1370 a->nla_len = sfa->actions_len - st_offset;
1371}
1372
ccf43786
SH		 1373static int ovs_nla_copy_actions__(const struct nlattr *attr,
1374 const struct sw_flow_key *key,
1375 int depth, struct sw_flow_actions **sfa,
1376 __be16 eth_type, __be16 vlan_tci);
1377
a097c0b2
PS		 1378static int validate_and_copy_sample(const struct nlattr *attr,
1379 const struct sw_flow_key *key, int depth,
ccf43786
SH		 1380 struct sw_flow_actions **sfa,
1381 __be16 eth_type, __be16 vlan_tci)
a097c0b2
PS		 1382{
1383 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1384 const struct nlattr *probability, *actions;
1385 const struct nlattr *a;
1386 int rem, start, err, st_acts;
1387
1388 memset(attrs, 0, sizeof(attrs));
1389 nla_for_each_nested(a, attr, rem) {
1390 int type = nla_type(a);
1391 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1392 return -EINVAL;
1393 attrs[type] = a;
1394 }
1395 if (rem)
1396 return -EINVAL;
1397
1398 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1399 if (!probability || nla_len(probability) != sizeof(u32))
1400 return -EINVAL;
1401
1402 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1403 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1404 return -EINVAL;
1405
1406 /* validation done, copy sample action. */
1407 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
1408 if (start < 0)
1409 return start;
1410 err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1411 nla_data(probability), sizeof(u32));
1412 if (err)
1413 return err;
1414 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
1415 if (st_acts < 0)
1416 return st_acts;
1417
ccf43786
SH		 1418 err = ovs_nla_copy_actions__(actions, key, depth + 1, sfa,
1419 eth_type, vlan_tci);
a097c0b2
PS		 1420 if (err)
1421 return err;
1422
1423 add_nested_action_end(*sfa, st_acts);
1424 add_nested_action_end(*sfa, start);
1425
1426 return 0;
1427}
1428
ccf43786
SH		 1429static int validate_tp_port(const struct sw_flow_key *flow_key,
1430 __be16 eth_type)
a097c0b2 1431{
ccf43786 1432 if ((eth_type == htons(ETH_P_IP) || eth_type == htons(ETH_P_IPV6)) &&
708fb4c5
JR		 1433 (flow_key->tp.src || flow_key->tp.dst))
1434 return 0;
a097c0b2
PS		 1435
1436 return -EINVAL;
1437}
1438
1439void ovs_match_init(struct sw_flow_match *match,
1440 struct sw_flow_key *key,
1441 struct sw_flow_mask *mask)
1442{
1443 memset(match, 0, sizeof(*match));
1444 match->key = key;
1445 match->mask = mask;
1446
1447 memset(key, 0, sizeof(*key));
1448
1449 if (mask) {
1450 memset(&mask->key, 0, sizeof(mask->key));
1451 mask->range.start = mask->range.end = 0;
1452 }
1453}
1454
1455static int validate_and_copy_set_tun(const struct nlattr *attr,
1456 struct sw_flow_actions **sfa)
1457{
1458 struct sw_flow_match match;
1459 struct sw_flow_key key;
f0cd669f
JG		 1460 struct ovs_tunnel_info *tun_info;
1461 struct nlattr *a;
a097c0b2
PS		 1462 int err, start;
1463
1464 ovs_match_init(&match, &key, NULL);
1465 err = ipv4_tun_from_nlattr(nla_data(attr), &match, false);
1466 if (err)
1467 return err;
1468
c1fc1411
JG		 1469 if (key.tun_opts_len) {
1470 struct geneve_opt *option = GENEVE_OPTS(&key,
1471 key.tun_opts_len);
1472 int opts_len = key.tun_opts_len;
1473 bool crit_opt = false;
1474
1475 while (opts_len > 0) {
1476 int len;
1477
1478 if (opts_len < sizeof(*option))
1479 return -EINVAL;
1480
1481 len = sizeof(*option) + option->length * 4;
1482 if (len > opts_len)
1483 return -EINVAL;
1484
1485 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1486
1487 option = (struct geneve_opt *)((u8 *)option + len);
1488 opts_len -= len;
1489 };
1490
1491 key.tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1492 };
1493
a097c0b2
PS		 1494 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
1495 if (start < 0)
1496 return start;
1497
f0cd669f 1498 a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
c1fc1411 1499 sizeof(*tun_info) + key.tun_opts_len);
f0cd669f
JG		 1500 if (IS_ERR(a))
1501 return PTR_ERR(a);
1502
1503 tun_info = nla_data(a);
1504 tun_info->tunnel = key.tun_key;
c1fc1411
JG		 1505 tun_info->options_len = key.tun_opts_len;
1506
1507 if (tun_info->options_len) {
1508 /* We need to store the options in the action itself since
1509 * everything else will go away after flow setup. We can append
1510 * it to tun_info and then point there.
1511 */
1512 tun_info->options = (struct geneve_opt *)(tun_info + 1);
1513 memcpy(tun_info->options, GENEVE_OPTS(&key, key.tun_opts_len),
1514 key.tun_opts_len);
1515 } else {
1516 tun_info->options = NULL;
1517 }
f0cd669f 1518
a097c0b2
PS		 1519 add_nested_action_end(*sfa, start);
1520
1521 return err;
1522}
1523
1524static int validate_set(const struct nlattr *a,
1525 const struct sw_flow_key *flow_key,
1526 struct sw_flow_actions **sfa,
ccf43786 1527 bool *set_tun, __be16 eth_type)
a097c0b2
PS		 1528{
1529 const struct nlattr *ovs_key = nla_data(a);
1530 int key_type = nla_type(ovs_key);
1531
1532 /* There can be only one key in a action */
1533 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1534 return -EINVAL;
1535
1536 if (key_type > OVS_KEY_ATTR_MAX ||
1537 (ovs_key_lens[key_type] != nla_len(ovs_key) &&
1538 ovs_key_lens[key_type] != -1))
1539 return -EINVAL;
1540
1541 switch (key_type) {
1542 const struct ovs_key_ipv4 *ipv4_key;
1543 const struct ovs_key_ipv6 *ipv6_key;
1544 int err;
1545
1546 case OVS_KEY_ATTR_PRIORITY:
1547 case OVS_KEY_ATTR_SKB_MARK:
1548 case OVS_KEY_ATTR_ETHERNET:
1549 break;
1550
1551 case OVS_KEY_ATTR_TUNNEL:
1552 *set_tun = true;
1553 err = validate_and_copy_set_tun(a, sfa);
1554 if (err)
1555 return err;
1556 break;
1557
1558 case OVS_KEY_ATTR_IPV4:
ccf43786 1559 if (eth_type != htons(ETH_P_IP))
a097c0b2
PS		 1560 return -EINVAL;
1561
1562 if (!flow_key->ip.proto)
1563 return -EINVAL;
1564
1565 ipv4_key = nla_data(ovs_key);
1566 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
1567 return -EINVAL;
1568
1569 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
1570 return -EINVAL;
1571
1572 break;
1573
1574 case OVS_KEY_ATTR_IPV6:
ccf43786 1575 if (eth_type != htons(ETH_P_IPV6))
a097c0b2
PS		 1576 return -EINVAL;
1577
1578 if (!flow_key->ip.proto)
1579 return -EINVAL;
1580
1581 ipv6_key = nla_data(ovs_key);
1582 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
1583 return -EINVAL;
1584
1585 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
1586 return -EINVAL;
1587
1588 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
1589 return -EINVAL;
1590
1591 break;
1592
1593 case OVS_KEY_ATTR_TCP:
1594 if (flow_key->ip.proto != IPPROTO_TCP)
1595 return -EINVAL;
1596
ccf43786 1597 return validate_tp_port(flow_key, eth_type);
a097c0b2
PS		 1598
1599 case OVS_KEY_ATTR_UDP:
1600 if (flow_key->ip.proto != IPPROTO_UDP)
1601 return -EINVAL;
1602
ccf43786
SH		 1603 return validate_tp_port(flow_key, eth_type);
1604
1605 case OVS_KEY_ATTR_MPLS:
1606 if (!eth_p_mpls(eth_type))
1607 return -EINVAL;
1608 break;
a097c0b2
PS		 1609
1610 case OVS_KEY_ATTR_SCTP:
1611 if (flow_key->ip.proto != IPPROTO_SCTP)
1612 return -EINVAL;
1613
ccf43786 1614 return validate_tp_port(flow_key, eth_type);
a097c0b2
PS		 1615
1616 default:
1617 return -EINVAL;
1618 }
1619
1620 return 0;
1621}
1622
1623static int validate_userspace(const struct nlattr *attr)
1624{
1625 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
1626 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
1627 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
1628 };
1629 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
1630 int error;
1631
1632 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
1633 attr, userspace_policy);
1634 if (error)
1635 return error;
1636
1637 if (!a[OVS_USERSPACE_ATTR_PID] ||
1638 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
1639 return -EINVAL;
1640
1641 return 0;
1642}
1643
1644static int copy_action(const struct nlattr *from,
1645 struct sw_flow_actions **sfa)
1646{
1647 int totlen = NLA_ALIGN(from->nla_len);
1648 struct nlattr *to;
1649
1650 to = reserve_sfa_size(sfa, from->nla_len);
1651 if (IS_ERR(to))
1652 return PTR_ERR(to);
1653
1654 memcpy(to, from, totlen);
1655 return 0;
1656}
1657
ccf43786
SH		 1658static int ovs_nla_copy_actions__(const struct nlattr *attr,
1659 const struct sw_flow_key *key,
1660 int depth, struct sw_flow_actions **sfa,
1661 __be16 eth_type, __be16 vlan_tci)
a097c0b2
PS		 1662{
1663 const struct nlattr *a;
1664 int rem, err;
1665
1666 if (depth >= SAMPLE_ACTION_DEPTH)
1667 return -EOVERFLOW;
1668
1669 nla_for_each_nested(a, attr, rem) {
1670 /* Expected argument lengths, (u32)-1 for variable length. */
1671 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
1672 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
a6059080 1673 [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
a097c0b2 1674 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
ccf43786
SH		 1675 [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
1676 [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
a097c0b2
PS		 1677 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
1678 [OVS_ACTION_ATTR_POP_VLAN] = 0,
1679 [OVS_ACTION_ATTR_SET] = (u32)-1,
7804df20
AZ		 1680 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
1681 [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash)
a097c0b2
PS		 1682 };
1683 const struct ovs_action_push_vlan *vlan;
1684 int type = nla_type(a);
1685 bool skip_copy;
1686
1687 if (type > OVS_ACTION_ATTR_MAX ||
1688 (action_lens[type] != nla_len(a) &&
1689 action_lens[type] != (u32)-1))
1690 return -EINVAL;
1691
1692 skip_copy = false;
1693 switch (type) {
1694 case OVS_ACTION_ATTR_UNSPEC:
1695 return -EINVAL;
1696
1697 case OVS_ACTION_ATTR_USERSPACE:
1698 err = validate_userspace(a);
1699 if (err)
1700 return err;
1701 break;
1702
1703 case OVS_ACTION_ATTR_OUTPUT:
1704 if (nla_get_u32(a) >= DP_MAX_PORTS)
1705 return -EINVAL;
1706 break;
1707
7804df20
AZ		 1708 case OVS_ACTION_ATTR_HASH: {
1709 const struct ovs_action_hash *act_hash = nla_data(a);
1710
1711 switch (act_hash->hash_alg) {
1712 case OVS_HASH_ALG_L4:
1713 break;
1714 default:
1715 return -EINVAL;
1716 }
1717
1718 break;
1719 }
a097c0b2
PS		 1720
1721 case OVS_ACTION_ATTR_POP_VLAN:
e0b8f73f 1722 vlan_tci = htons(0);
a097c0b2
PS		 1723 break;
1724
1725 case OVS_ACTION_ATTR_PUSH_VLAN:
1726 vlan = nla_data(a);
1727 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
1728 return -EINVAL;
1729 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
1730 return -EINVAL;
ccf43786 1731 vlan_tci = vlan->vlan_tci;
a097c0b2
PS		 1732 break;
1733
a6059080
AZ		 1734 case OVS_ACTION_ATTR_RECIRC:
1735 break;
1736
ccf43786
SH		 1737 case OVS_ACTION_ATTR_PUSH_MPLS: {
1738 const struct ovs_action_push_mpls *mpls = nla_data(a);
1739
1740 if (!eth_p_mpls(mpls->mpls_ethertype))
1741 return -EINVAL;
1742 /* Prohibit push MPLS other than to a white list
1743 * for packets that have a known tag order.
e0b8f73f 1744 */
ccf43786
SH		 1745 if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
1746 (eth_type != htons(ETH_P_IP) &&
1747 eth_type != htons(ETH_P_IPV6) &&
1748 eth_type != htons(ETH_P_ARP) &&
1749 eth_type != htons(ETH_P_RARP) &&
1750 !eth_p_mpls(eth_type)))
1751 return -EINVAL;
1752 eth_type = mpls->mpls_ethertype;
1753 break;
1754 }
1755
1756 case OVS_ACTION_ATTR_POP_MPLS:
1757 if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
1758 !eth_p_mpls(eth_type))
1759 return -EINVAL;
1760
1761 /* Disallow subsequent L2.5+ set and mpls_pop actions
1762 * as there is no check here to ensure that the new
1763 * eth_type is valid and thus set actions could
1764 * write off the end of the packet or otherwise
1765 * corrupt it.
1766 *
1767 * Support for these actions is planned using packet
1768 * recirculation.
1769 */
1770 eth_type = htons(0);
1771 break;
1772
a097c0b2 1773 case OVS_ACTION_ATTR_SET:
ccf43786 1774 err = validate_set(a, key, sfa, &skip_copy, eth_type);
a097c0b2
PS		 1775 if (err)
1776 return err;
1777 break;
1778
1779 case OVS_ACTION_ATTR_SAMPLE:
ccf43786
SH		 1780 err = validate_and_copy_sample(a, key, depth, sfa,
1781 eth_type, vlan_tci);
a097c0b2
PS		 1782 if (err)
1783 return err;
1784 skip_copy = true;
1785 break;
1786
1787 default:
1788 return -EINVAL;
1789 }
1790 if (!skip_copy) {
1791 err = copy_action(a, sfa);
1792 if (err)
1793 return err;
1794 }
1795 }
1796
1797 if (rem > 0)
1798 return -EINVAL;
1799
1800 return 0;
1801}
1802
ccf43786
SH		 1803int ovs_nla_copy_actions(const struct nlattr *attr,
1804 const struct sw_flow_key *key,
1805 struct sw_flow_actions **sfa)
1806{
1807 return ovs_nla_copy_actions__(attr, key, 0, sfa, key->eth.type,
1808 key->eth.tci);
1809}
1810
a097c0b2
PS		 1811static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1812{
1813 const struct nlattr *a;
1814 struct nlattr *start;
1815 int err = 0, rem;
1816
1817 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1818 if (!start)
1819 return -EMSGSIZE;
1820
1821 nla_for_each_nested(a, attr, rem) {
1822 int type = nla_type(a);
1823 struct nlattr *st_sample;
1824
1825 switch (type) {
1826 case OVS_SAMPLE_ATTR_PROBABILITY:
1827 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
1828 sizeof(u32), nla_data(a)))
1829 return -EMSGSIZE;
1830 break;
1831 case OVS_SAMPLE_ATTR_ACTIONS:
1832 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1833 if (!st_sample)
1834 return -EMSGSIZE;
1835 err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
1836 if (err)
1837 return err;
1838 nla_nest_end(skb, st_sample);
1839 break;
1840 }
1841 }
1842
1843 nla_nest_end(skb, start);
1844 return err;
1845}
1846
1847static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1848{
1849 const struct nlattr *ovs_key = nla_data(a);
1850 int key_type = nla_type(ovs_key);
1851 struct nlattr *start;
1852 int err;
1853
1854 switch (key_type) {
f0cd669f
JG		 1855 case OVS_KEY_ATTR_TUNNEL_INFO: {
1856 struct ovs_tunnel_info *tun_info = nla_data(ovs_key);
1857
a097c0b2
PS		 1858 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1859 if (!start)
1860 return -EMSGSIZE;
1861
f0cd669f 1862 err = ipv4_tun_to_nlattr(skb, &tun_info->tunnel,
c1fc1411
JG		 1863 tun_info->options_len ?
1864 tun_info->options : NULL,
1865 tun_info->options_len);
a097c0b2
PS		 1866 if (err)
1867 return err;
1868 nla_nest_end(skb, start);
1869 break;
f0cd669f 1870 }
a097c0b2
PS		 1871 default:
1872 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1873 return -EMSGSIZE;
1874 break;
1875 }
1876
1877 return 0;
1878}
1879
1880int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
1881{
1882 const struct nlattr *a;
1883 int rem, err;
1884
1885 nla_for_each_attr(a, attr, len, rem) {
1886 int type = nla_type(a);
1887
1888 switch (type) {
1889 case OVS_ACTION_ATTR_SET:
1890 err = set_action_to_attr(a, skb);
1891 if (err)
1892 return err;
1893 break;
1894
1895 case OVS_ACTION_ATTR_SAMPLE:
1896 err = sample_action_to_attr(a, skb);
1897 if (err)
1898 return err;
1899 break;
1900 default:
1901 if (nla_put(skb, type, nla_len(a), nla_data(a)))
1902 return -EMSGSIZE;
1903 break;
1904 }
1905 }
1906
1907 return 0;
1908}
openvswitch packages for PVE