1 /*
2  * Copyright (c) 2007-2013 Nicira, Inc.
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 
19 #include "flow.h"
20 #include "datapath.h"
21 #include <linux/uaccess.h>
22 #include <linux/netdevice.h>
23 #include <linux/etherdevice.h>
24 #include <linux/if_ether.h>
25 #include <linux/if_vlan.h>
26 #include <net/llc_pdu.h>
27 #include <linux/kernel.h>
28 #include <linux/jhash.h>
29 #include <linux/jiffies.h>
30 #include <linux/llc.h>
31 #include <linux/module.h>
32 #include <linux/in.h>
33 #include <linux/rcupdate.h>
34 #include <linux/if_arp.h>
35 #include <linux/ip.h>
36 #include <linux/ipv6.h>
37 #include <linux/sctp.h>
38 #include <linux/tcp.h>
39 #include <linux/udp.h>
40 #include <linux/icmp.h>
41 #include <linux/icmpv6.h>
42 #include <linux/rculist.h>
43 #include <net/ip.h>
44 #include <net/ipv6.h>
45 #include <net/ndisc.h>
46 
47 #include "flow_netlink.h"
48 
49 static void update_range__(struct sw_flow_match *match,
50 			   size_t offset, size_t size, bool is_mask)
51 {
52 	struct sw_flow_key_range *range = NULL;
53 	size_t start = rounddown(offset, sizeof(long));
54 	size_t end = roundup(offset + size, sizeof(long));
55 
56 	if (!is_mask)
57 		range = &match->range;
58 	else if (match->mask)
59 		range = &match->mask->range;
60 
61 	if (!range)
62 		return;
63 
64 	if (range->start == range->end) {
65 		range->start = start;
66 		range->end = end;
67 		return;
68 	}
69 
70 	if (range->start > start)
71 		range->start = start;
72 
73 	if (range->end < end)
74 		range->end = end;
75 }
76 
77 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
78 	do { \
79 		update_range__(match, offsetof(struct sw_flow_key, field),  \
80 				     sizeof((match)->key->field), is_mask); \
81 		if (is_mask) {						    \
82 			if ((match)->mask)				    \
83 				(match)->mask->key.field = value;	    \
84 		} else {                                                    \
85 			(match)->key->field = value;		            \
86 		}                                                           \
87 	} while (0)
88 
89 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
90 	do { \
91 		update_range__(match, offsetof(struct sw_flow_key, field),  \
92 				len, is_mask);                              \
93 		if (is_mask) {						    \
94 			if ((match)->mask)				    \
95 				memcpy(&(match)->mask->key.field, value_p, len);\
96 		} else {                                                    \
97 			memcpy(&(match)->key->field, value_p, len);         \
98 		}                                                           \
99 	} while (0)
100 
101 static u16 range_n_bytes(const struct sw_flow_key_range *range)
102 {
103 	return range->end - range->start;
104 }
105 
106 static bool match_validate(const struct sw_flow_match *match,
107 			   u64 key_attrs, u64 mask_attrs)
108 {
109 	u64 key_expected = 1 << OVS_KEY_ATTR_ETHERNET;
110 	u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
111 
112 	/* The following mask attributes allowed only if they
113 	 * pass the validation tests. */
114 	mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
115 			| (1 << OVS_KEY_ATTR_IPV6)
116 			| (1 << OVS_KEY_ATTR_TCP)
117 			| (1 << OVS_KEY_ATTR_TCP_FLAGS)
118 			| (1 << OVS_KEY_ATTR_UDP)
119 			| (1 << OVS_KEY_ATTR_SCTP)
120 			| (1 << OVS_KEY_ATTR_ICMP)
121 			| (1 << OVS_KEY_ATTR_ICMPV6)
122 			| (1 << OVS_KEY_ATTR_ARP)
123 			| (1 << OVS_KEY_ATTR_ND));
124 
125 	/* Always allowed mask fields. */
126 	mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
127 		       | (1 << OVS_KEY_ATTR_IN_PORT)
128 		       | (1 << OVS_KEY_ATTR_ETHERTYPE));
129 
130 	/* Check key attributes. */
131 	if (match->key->eth.type == htons(ETH_P_ARP)
132 			|| match->key->eth.type == htons(ETH_P_RARP)) {
133 		key_expected |= 1 << OVS_KEY_ATTR_ARP;
134 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
135 			mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
136 	}
137 
138 	if (match->key->eth.type == htons(ETH_P_IP)) {
139 		key_expected |= 1 << OVS_KEY_ATTR_IPV4;
140 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
141 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
142 
143 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
144 			if (match->key->ip.proto == IPPROTO_UDP) {
145 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
146 				if (match->mask && (match->mask->key.ip.proto == 0xff))
147 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
148 			}
149 
150 			if (match->key->ip.proto == IPPROTO_SCTP) {
151 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
152 				if (match->mask && (match->mask->key.ip.proto == 0xff))
153 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
154 			}
155 
156 			if (match->key->ip.proto == IPPROTO_TCP) {
157 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
158 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
159 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
160 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
161 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
162 				}
163 			}
164 
165 			if (match->key->ip.proto == IPPROTO_ICMP) {
166 				key_expected |= 1 << OVS_KEY_ATTR_ICMP;
167 				if (match->mask && (match->mask->key.ip.proto == 0xff))
168 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
169 			}
170 		}
171 	}
172 
173 	if (match->key->eth.type == htons(ETH_P_IPV6)) {
174 		key_expected |= 1 << OVS_KEY_ATTR_IPV6;
175 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
176 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
177 
178 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
179 			if (match->key->ip.proto == IPPROTO_UDP) {
180 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
181 				if (match->mask && (match->mask->key.ip.proto == 0xff))
182 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
183 			}
184 
185 			if (match->key->ip.proto == IPPROTO_SCTP) {
186 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
187 				if (match->mask && (match->mask->key.ip.proto == 0xff))
188 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
189 			}
190 
191 			if (match->key->ip.proto == IPPROTO_TCP) {
192 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
193 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
194 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
195 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
196 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
197 				}
198 			}
199 
200 			if (match->key->ip.proto == IPPROTO_ICMPV6) {
201 				key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
202 				if (match->mask && (match->mask->key.ip.proto == 0xff))
203 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
204 
205 				if (match->key->ipv6.tp.src ==
206 						htons(NDISC_NEIGHBOUR_SOLICITATION) ||
207 				    match->key->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
208 					key_expected |= 1 << OVS_KEY_ATTR_ND;
209 					if (match->mask && (match->mask->key.ipv6.tp.src == htons(0xffff)))
210 						mask_allowed |= 1 << OVS_KEY_ATTR_ND;
211 				}
212 			}
213 		}
214 	}
215 
216 	if ((key_attrs & key_expected) != key_expected) {
217 		/* Key attributes check failed. */
218 		OVS_NLERR("Missing expected key attributes (key_attrs=%llx, expected=%llx).\n",
219 				key_attrs, key_expected);
220 		return false;
221 	}
222 
223 	if ((mask_attrs & mask_allowed) != mask_attrs) {
224 		/* Mask attributes check failed. */
225 		OVS_NLERR("Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).\n",
226 				mask_attrs, mask_allowed);
227 		return false;
228 	}
229 
230 	return true;
231 }
232 
233 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
234 static const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
235 	[OVS_KEY_ATTR_ENCAP] = -1,
236 	[OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
237 	[OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
238 	[OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
239 	[OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
240 	[OVS_KEY_ATTR_VLAN] = sizeof(__be16),
241 	[OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
242 	[OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
243 	[OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
244 	[OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
245 	[OVS_KEY_ATTR_TCP_FLAGS] = sizeof(__be16),
246 	[OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
247 	[OVS_KEY_ATTR_SCTP] = sizeof(struct ovs_key_sctp),
248 	[OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
249 	[OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
250 	[OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
251 	[OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
252 	[OVS_KEY_ATTR_TUNNEL] = -1,
253 };
254 
255 static bool is_all_zero(const u8 *fp, size_t size)
256 {
257 	int i;
258 
259 	if (!fp)
260 		return false;
261 
262 	for (i = 0; i < size; i++)
263 		if (fp[i])
264 			return false;
265 
266 	return true;
267 }
268 
269 static int __parse_flow_nlattrs(const struct nlattr *attr,
270 				const struct nlattr *a[],
271 				u64 *attrsp, bool nz)
272 {
273 	const struct nlattr *nla;
274 	u64 attrs;
275 	int rem;
276 
277 	attrs = *attrsp;
278 	nla_for_each_nested(nla, attr, rem) {
279 		u16 type = nla_type(nla);
280 		int expected_len;
281 
282 		if (type > OVS_KEY_ATTR_MAX) {
283 			OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
284 				  type, OVS_KEY_ATTR_MAX);
285 			return -EINVAL;
286 		}
287 
288 		if (attrs & (1 << type)) {
289 			OVS_NLERR("Duplicate key attribute (type %d).\n", type);
290 			return -EINVAL;
291 		}
292 
293 		expected_len = ovs_key_lens[type];
294 		if (nla_len(nla) != expected_len && expected_len != -1) {
295 			OVS_NLERR("Key attribute has unexpected length (type=%d"
296 				  ", length=%d, expected=%d).\n", type,
297 				  nla_len(nla), expected_len);
298 			return -EINVAL;
299 		}
300 
301 		if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
302 			attrs |= 1 << type;
303 			a[type] = nla;
304 		}
305 	}
306 	if (rem) {
307 		OVS_NLERR("Message has %d unknown bytes.\n", rem);
308 		return -EINVAL;
309 	}
310 
311 	*attrsp = attrs;
312 	return 0;
313 }
314 
315 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
316 				   const struct nlattr *a[], u64 *attrsp)
317 {
318 	return __parse_flow_nlattrs(attr, a, attrsp, true);
319 }
320 
321 static int parse_flow_nlattrs(const struct nlattr *attr,
322 			      const struct nlattr *a[], u64 *attrsp)
323 {
324 	return __parse_flow_nlattrs(attr, a, attrsp, false);
325 }
326 
327 static int ipv4_tun_from_nlattr(const struct nlattr *attr,
328 				struct sw_flow_match *match, bool is_mask)
329 {
330 	struct nlattr *a;
331 	int rem;
332 	bool ttl = false;
333 	__be16 tun_flags = 0;
334 
335 	nla_for_each_nested(a, attr, rem) {
336 		int type = nla_type(a);
337 		static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
338 			[OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
339 			[OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
340 			[OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
341 			[OVS_TUNNEL_KEY_ATTR_TOS] = 1,
342 			[OVS_TUNNEL_KEY_ATTR_TTL] = 1,
343 			[OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
344 			[OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
345 		};
346 
347 		if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
348 			OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
349 			type, OVS_TUNNEL_KEY_ATTR_MAX);
350 			return -EINVAL;
351 		}
352 
353 		if (ovs_tunnel_key_lens[type] != nla_len(a)) {
354 			OVS_NLERR("IPv4 tunnel attribute type has unexpected "
355 				  " length (type=%d, length=%d, expected=%d).\n",
356 				  type, nla_len(a), ovs_tunnel_key_lens[type]);
357 			return -EINVAL;
358 		}
359 
360 		switch (type) {
361 		case OVS_TUNNEL_KEY_ATTR_ID:
362 			SW_FLOW_KEY_PUT(match, tun_key.tun_id,
363 					nla_get_be64(a), is_mask);
364 			tun_flags |= TUNNEL_KEY;
365 			break;
366 		case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
367 			SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
368 					nla_get_be32(a), is_mask);
369 			break;
370 		case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
371 			SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
372 					nla_get_be32(a), is_mask);
373 			break;
374 		case OVS_TUNNEL_KEY_ATTR_TOS:
375 			SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
376 					nla_get_u8(a), is_mask);
377 			break;
378 		case OVS_TUNNEL_KEY_ATTR_TTL:
379 			SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
380 					nla_get_u8(a), is_mask);
381 			ttl = true;
382 			break;
383 		case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
384 			tun_flags |= TUNNEL_DONT_FRAGMENT;
385 			break;
386 		case OVS_TUNNEL_KEY_ATTR_CSUM:
387 			tun_flags |= TUNNEL_CSUM;
388 			break;
389 		default:
390 			return -EINVAL;
391 		}
392 	}
393 
394 	SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
395 
396 	if (rem > 0) {
397 		OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
398 		return -EINVAL;
399 	}
400 
401 	if (!is_mask) {
402 		if (!match->key->tun_key.ipv4_dst) {
403 			OVS_NLERR("IPv4 tunnel destination address is zero.\n");
404 			return -EINVAL;
405 		}
406 
407 		if (!ttl) {
408 			OVS_NLERR("IPv4 tunnel TTL not specified.\n");
409 			return -EINVAL;
410 		}
411 	}
412 
413 	return 0;
414 }
415 
416 static int ipv4_tun_to_nlattr(struct sk_buff *skb,
417 			      const struct ovs_key_ipv4_tunnel *tun_key,
418 			      const struct ovs_key_ipv4_tunnel *output)
419 {
420 	struct nlattr *nla;
421 
422 	nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
423 	if (!nla)
424 		return -EMSGSIZE;
425 
426 	if (output->tun_flags & TUNNEL_KEY &&
427 	    nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
428 		return -EMSGSIZE;
429 	if (output->ipv4_src &&
430 		nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
431 		return -EMSGSIZE;
432 	if (output->ipv4_dst &&
433 		nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
434 		return -EMSGSIZE;
435 	if (output->ipv4_tos &&
436 		nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
437 		return -EMSGSIZE;
438 	if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
439 		return -EMSGSIZE;
440 	if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
441 		nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
442 		return -EMSGSIZE;
443 	if ((output->tun_flags & TUNNEL_CSUM) &&
444 		nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
445 		return -EMSGSIZE;
446 
447 	nla_nest_end(skb, nla);
448 	return 0;
449 }
450 
451 
452 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
453 				 const struct nlattr **a, bool is_mask)
454 {
455 	if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
456 		SW_FLOW_KEY_PUT(match, phy.priority,
457 			  nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
458 		*attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
459 	}
460 
461 	if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
462 		u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
463 
464 		if (is_mask)
465 			in_port = 0xffffffff; /* Always exact match in_port. */
466 		else if (in_port >= DP_MAX_PORTS)
467 			return -EINVAL;
468 
469 		SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
470 		*attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
471 	} else if (!is_mask) {
472 		SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
473 	}
474 
475 	if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
476 		uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
477 
478 		SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
479 		*attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
480 	}
481 	if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
482 		if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
483 					 is_mask))
484 			return -EINVAL;
485 		*attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
486 	}
487 	return 0;
488 }
489 
490 static int ovs_key_from_nlattrs(struct sw_flow_match *match,  u64 attrs,
491 				const struct nlattr **a, bool is_mask)
492 {
493 	int err;
494 	u64 orig_attrs = attrs;
495 
496 	err = metadata_from_nlattrs(match, &attrs, a, is_mask);
497 	if (err)
498 		return err;
499 
500 	if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
501 		const struct ovs_key_ethernet *eth_key;
502 
503 		eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
504 		SW_FLOW_KEY_MEMCPY(match, eth.src,
505 				eth_key->eth_src, ETH_ALEN, is_mask);
506 		SW_FLOW_KEY_MEMCPY(match, eth.dst,
507 				eth_key->eth_dst, ETH_ALEN, is_mask);
508 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
509 	}
510 
511 	if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
512 		__be16 tci;
513 
514 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
515 		if (!(tci & htons(VLAN_TAG_PRESENT))) {
516 			if (is_mask)
517 				OVS_NLERR("VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.\n");
518 			else
519 				OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
520 
521 			return -EINVAL;
522 		}
523 
524 		SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
525 		attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
526 	} else if (!is_mask)
527 		SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
528 
529 	if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
530 		__be16 eth_type;
531 
532 		eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
533 		if (is_mask) {
534 			/* Always exact match EtherType. */
535 			eth_type = htons(0xffff);
536 		} else if (ntohs(eth_type) < ETH_P_802_3_MIN) {
537 			OVS_NLERR("EtherType is less than minimum (type=%x, min=%x).\n",
538 					ntohs(eth_type), ETH_P_802_3_MIN);
539 			return -EINVAL;
540 		}
541 
542 		SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
543 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
544 	} else if (!is_mask) {
545 		SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
546 	}
547 
548 	if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
549 		const struct ovs_key_ipv4 *ipv4_key;
550 
551 		ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
552 		if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
553 			OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
554 				ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
555 			return -EINVAL;
556 		}
557 		SW_FLOW_KEY_PUT(match, ip.proto,
558 				ipv4_key->ipv4_proto, is_mask);
559 		SW_FLOW_KEY_PUT(match, ip.tos,
560 				ipv4_key->ipv4_tos, is_mask);
561 		SW_FLOW_KEY_PUT(match, ip.ttl,
562 				ipv4_key->ipv4_ttl, is_mask);
563 		SW_FLOW_KEY_PUT(match, ip.frag,
564 				ipv4_key->ipv4_frag, is_mask);
565 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
566 				ipv4_key->ipv4_src, is_mask);
567 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
568 				ipv4_key->ipv4_dst, is_mask);
569 		attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
570 	}
571 
572 	if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
573 		const struct ovs_key_ipv6 *ipv6_key;
574 
575 		ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
576 		if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
577 			OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
578 				ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
579 			return -EINVAL;
580 		}
581 		SW_FLOW_KEY_PUT(match, ipv6.label,
582 				ipv6_key->ipv6_label, is_mask);
583 		SW_FLOW_KEY_PUT(match, ip.proto,
584 				ipv6_key->ipv6_proto, is_mask);
585 		SW_FLOW_KEY_PUT(match, ip.tos,
586 				ipv6_key->ipv6_tclass, is_mask);
587 		SW_FLOW_KEY_PUT(match, ip.ttl,
588 				ipv6_key->ipv6_hlimit, is_mask);
589 		SW_FLOW_KEY_PUT(match, ip.frag,
590 				ipv6_key->ipv6_frag, is_mask);
591 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
592 				ipv6_key->ipv6_src,
593 				sizeof(match->key->ipv6.addr.src),
594 				is_mask);
595 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
596 				ipv6_key->ipv6_dst,
597 				sizeof(match->key->ipv6.addr.dst),
598 				is_mask);
599 
600 		attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
601 	}
602 
603 	if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
604 		const struct ovs_key_arp *arp_key;
605 
606 		arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
607 		if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
608 			OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
609 				  arp_key->arp_op);
610 			return -EINVAL;
611 		}
612 
613 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
614 				arp_key->arp_sip, is_mask);
615 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
616 			arp_key->arp_tip, is_mask);
617 		SW_FLOW_KEY_PUT(match, ip.proto,
618 				ntohs(arp_key->arp_op), is_mask);
619 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
620 				arp_key->arp_sha, ETH_ALEN, is_mask);
621 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
622 				arp_key->arp_tha, ETH_ALEN, is_mask);
623 
624 		attrs &= ~(1 << OVS_KEY_ATTR_ARP);
625 	}
626 
627 	if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
628 		const struct ovs_key_tcp *tcp_key;
629 
630 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
631 		if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) {
632 			SW_FLOW_KEY_PUT(match, ipv4.tp.src,
633 					tcp_key->tcp_src, is_mask);
634 			SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
635 					tcp_key->tcp_dst, is_mask);
636 		} else {
637 			SW_FLOW_KEY_PUT(match, ipv6.tp.src,
638 					tcp_key->tcp_src, is_mask);
639 			SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
640 					tcp_key->tcp_dst, is_mask);
641 		}
642 		attrs &= ~(1 << OVS_KEY_ATTR_TCP);
643 	}
644 
645 	if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
646 		if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) {
647 			SW_FLOW_KEY_PUT(match, ipv4.tp.flags,
648 					nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
649 					is_mask);
650 		} else {
651 			SW_FLOW_KEY_PUT(match, ipv6.tp.flags,
652 					nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
653 					is_mask);
654 		}
655 		attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
656 	}
657 
658 	if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
659 		const struct ovs_key_udp *udp_key;
660 
661 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
662 		if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) {
663 			SW_FLOW_KEY_PUT(match, ipv4.tp.src,
664 					udp_key->udp_src, is_mask);
665 			SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
666 					udp_key->udp_dst, is_mask);
667 		} else {
668 			SW_FLOW_KEY_PUT(match, ipv6.tp.src,
669 					udp_key->udp_src, is_mask);
670 			SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
671 					udp_key->udp_dst, is_mask);
672 		}
673 		attrs &= ~(1 << OVS_KEY_ATTR_UDP);
674 	}
675 
676 	if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
677 		const struct ovs_key_sctp *sctp_key;
678 
679 		sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
680 		if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) {
681 			SW_FLOW_KEY_PUT(match, ipv4.tp.src,
682 					sctp_key->sctp_src, is_mask);
683 			SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
684 					sctp_key->sctp_dst, is_mask);
685 		} else {
686 			SW_FLOW_KEY_PUT(match, ipv6.tp.src,
687 					sctp_key->sctp_src, is_mask);
688 			SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
689 					sctp_key->sctp_dst, is_mask);
690 		}
691 		attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
692 	}
693 
694 	if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
695 		const struct ovs_key_icmp *icmp_key;
696 
697 		icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
698 		SW_FLOW_KEY_PUT(match, ipv4.tp.src,
699 				htons(icmp_key->icmp_type), is_mask);
700 		SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
701 				htons(icmp_key->icmp_code), is_mask);
702 		attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
703 	}
704 
705 	if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
706 		const struct ovs_key_icmpv6 *icmpv6_key;
707 
708 		icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
709 		SW_FLOW_KEY_PUT(match, ipv6.tp.src,
710 				htons(icmpv6_key->icmpv6_type), is_mask);
711 		SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
712 				htons(icmpv6_key->icmpv6_code), is_mask);
713 		attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
714 	}
715 
716 	if (attrs & (1 << OVS_KEY_ATTR_ND)) {
717 		const struct ovs_key_nd *nd_key;
718 
719 		nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
720 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
721 			nd_key->nd_target,
722 			sizeof(match->key->ipv6.nd.target),
723 			is_mask);
724 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
725 			nd_key->nd_sll, ETH_ALEN, is_mask);
726 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
727 				nd_key->nd_tll, ETH_ALEN, is_mask);
728 		attrs &= ~(1 << OVS_KEY_ATTR_ND);
729 	}
730 
731 	if (attrs != 0)
732 		return -EINVAL;
733 
734 	return 0;
735 }
736 
737 static void sw_flow_mask_set(struct sw_flow_mask *mask,
738 			     struct sw_flow_key_range *range, u8 val)
739 {
740 	u8 *m = (u8 *)&mask->key + range->start;
741 
742 	mask->range = *range;
743 	memset(m, val, range_n_bytes(range));
744 }
745 
746 /**
747  * ovs_nla_get_match - parses Netlink attributes into a flow key and
748  * mask. In case the 'mask' is NULL, the flow is treated as exact match
749  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
750  * does not include any don't care bit.
751  * @match: receives the extracted flow match information.
752  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
753  * sequence. The fields should of the packet that triggered the creation
754  * of this flow.
755  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
756  * attribute specifies the mask field of the wildcarded flow.
757  */
758 int ovs_nla_get_match(struct sw_flow_match *match,
759 		      const struct nlattr *key,
760 		      const struct nlattr *mask)
761 {
762 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
763 	const struct nlattr *encap;
764 	u64 key_attrs = 0;
765 	u64 mask_attrs = 0;
766 	bool encap_valid = false;
767 	int err;
768 
769 	err = parse_flow_nlattrs(key, a, &key_attrs);
770 	if (err)
771 		return err;
772 
773 	if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
774 	    (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
775 	    (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
776 		__be16 tci;
777 
778 		if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
779 		      (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
780 			OVS_NLERR("Invalid Vlan frame.\n");
781 			return -EINVAL;
782 		}
783 
784 		key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
785 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
786 		encap = a[OVS_KEY_ATTR_ENCAP];
787 		key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
788 		encap_valid = true;
789 
790 		if (tci & htons(VLAN_TAG_PRESENT)) {
791 			err = parse_flow_nlattrs(encap, a, &key_attrs);
792 			if (err)
793 				return err;
794 		} else if (!tci) {
795 			/* Corner case for truncated 802.1Q header. */
796 			if (nla_len(encap)) {
797 				OVS_NLERR("Truncated 802.1Q header has non-zero encap attribute.\n");
798 				return -EINVAL;
799 			}
800 		} else {
801 			OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
802 			return  -EINVAL;
803 		}
804 	}
805 
806 	err = ovs_key_from_nlattrs(match, key_attrs, a, false);
807 	if (err)
808 		return err;
809 
810 	if (mask) {
811 		err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
812 		if (err)
813 			return err;
814 
815 		if (mask_attrs & 1 << OVS_KEY_ATTR_ENCAP)  {
816 			__be16 eth_type = 0;
817 			__be16 tci = 0;
818 
819 			if (!encap_valid) {
820 				OVS_NLERR("Encap mask attribute is set for non-VLAN frame.\n");
821 				return  -EINVAL;
822 			}
823 
824 			mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
825 			if (a[OVS_KEY_ATTR_ETHERTYPE])
826 				eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
827 
828 			if (eth_type == htons(0xffff)) {
829 				mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
830 				encap = a[OVS_KEY_ATTR_ENCAP];
831 				err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
832 			} else {
833 				OVS_NLERR("VLAN frames must have an exact match on the TPID (mask=%x).\n",
834 						ntohs(eth_type));
835 				return -EINVAL;
836 			}
837 
838 			if (a[OVS_KEY_ATTR_VLAN])
839 				tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
840 
841 			if (!(tci & htons(VLAN_TAG_PRESENT))) {
842 				OVS_NLERR("VLAN tag present bit must have an exact match (tci_mask=%x).\n", ntohs(tci));
843 				return -EINVAL;
844 			}
845 		}
846 
847 		err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
848 		if (err)
849 			return err;
850 	} else {
851 		/* Populate exact match flow's key mask. */
852 		if (match->mask)
853 			sw_flow_mask_set(match->mask, &match->range, 0xff);
854 	}
855 
856 	if (!match_validate(match, key_attrs, mask_attrs))
857 		return -EINVAL;
858 
859 	return 0;
860 }
861 
862 /**
863  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
864  * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
865  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
866  * sequence.
867  *
868  * This parses a series of Netlink attributes that form a flow key, which must
869  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
870  * get the metadata, that is, the parts of the flow key that cannot be
871  * extracted from the packet itself.
872  */
873 
874 int ovs_nla_get_flow_metadata(struct sw_flow *flow,
875 			      const struct nlattr *attr)
876 {
877 	struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
878 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
879 	u64 attrs = 0;
880 	int err;
881 	struct sw_flow_match match;
882 
883 	flow->key.phy.in_port = DP_MAX_PORTS;
884 	flow->key.phy.priority = 0;
885 	flow->key.phy.skb_mark = 0;
886 	memset(tun_key, 0, sizeof(flow->key.tun_key));
887 
888 	err = parse_flow_nlattrs(attr, a, &attrs);
889 	if (err)
890 		return -EINVAL;
891 
892 	memset(&match, 0, sizeof(match));
893 	match.key = &flow->key;
894 
895 	err = metadata_from_nlattrs(&match, &attrs, a, false);
896 	if (err)
897 		return err;
898 
899 	return 0;
900 }
901 
902 int ovs_nla_put_flow(const struct sw_flow_key *swkey,
903 		     const struct sw_flow_key *output, struct sk_buff *skb)
904 {
905 	struct ovs_key_ethernet *eth_key;
906 	struct nlattr *nla, *encap;
907 	bool is_mask = (swkey != output);
908 
909 	if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
910 		goto nla_put_failure;
911 
912 	if ((swkey->tun_key.ipv4_dst || is_mask) &&
913 	    ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
914 		goto nla_put_failure;
915 
916 	if (swkey->phy.in_port == DP_MAX_PORTS) {
917 		if (is_mask && (output->phy.in_port == 0xffff))
918 			if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
919 				goto nla_put_failure;
920 	} else {
921 		u16 upper_u16;
922 		upper_u16 = !is_mask ? 0 : 0xffff;
923 
924 		if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
925 				(upper_u16 << 16) | output->phy.in_port))
926 			goto nla_put_failure;
927 	}
928 
929 	if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
930 		goto nla_put_failure;
931 
932 	nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
933 	if (!nla)
934 		goto nla_put_failure;
935 
936 	eth_key = nla_data(nla);
937 	memcpy(eth_key->eth_src, output->eth.src, ETH_ALEN);
938 	memcpy(eth_key->eth_dst, output->eth.dst, ETH_ALEN);
939 
940 	if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
941 		__be16 eth_type;
942 		eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
943 		if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
944 		    nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
945 			goto nla_put_failure;
946 		encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
947 		if (!swkey->eth.tci)
948 			goto unencap;
949 	} else
950 		encap = NULL;
951 
952 	if (swkey->eth.type == htons(ETH_P_802_2)) {
953 		/*
954 		 * Ethertype 802.2 is represented in the netlink with omitted
955 		 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
956 		 * 0xffff in the mask attribute.  Ethertype can also
957 		 * be wildcarded.
958 		 */
959 		if (is_mask && output->eth.type)
960 			if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
961 						output->eth.type))
962 				goto nla_put_failure;
963 		goto unencap;
964 	}
965 
966 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
967 		goto nla_put_failure;
968 
969 	if (swkey->eth.type == htons(ETH_P_IP)) {
970 		struct ovs_key_ipv4 *ipv4_key;
971 
972 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
973 		if (!nla)
974 			goto nla_put_failure;
975 		ipv4_key = nla_data(nla);
976 		ipv4_key->ipv4_src = output->ipv4.addr.src;
977 		ipv4_key->ipv4_dst = output->ipv4.addr.dst;
978 		ipv4_key->ipv4_proto = output->ip.proto;
979 		ipv4_key->ipv4_tos = output->ip.tos;
980 		ipv4_key->ipv4_ttl = output->ip.ttl;
981 		ipv4_key->ipv4_frag = output->ip.frag;
982 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
983 		struct ovs_key_ipv6 *ipv6_key;
984 
985 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
986 		if (!nla)
987 			goto nla_put_failure;
988 		ipv6_key = nla_data(nla);
989 		memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
990 				sizeof(ipv6_key->ipv6_src));
991 		memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
992 				sizeof(ipv6_key->ipv6_dst));
993 		ipv6_key->ipv6_label = output->ipv6.label;
994 		ipv6_key->ipv6_proto = output->ip.proto;
995 		ipv6_key->ipv6_tclass = output->ip.tos;
996 		ipv6_key->ipv6_hlimit = output->ip.ttl;
997 		ipv6_key->ipv6_frag = output->ip.frag;
998 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
999 		   swkey->eth.type == htons(ETH_P_RARP)) {
1000 		struct ovs_key_arp *arp_key;
1001 
1002 		nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1003 		if (!nla)
1004 			goto nla_put_failure;
1005 		arp_key = nla_data(nla);
1006 		memset(arp_key, 0, sizeof(struct ovs_key_arp));
1007 		arp_key->arp_sip = output->ipv4.addr.src;
1008 		arp_key->arp_tip = output->ipv4.addr.dst;
1009 		arp_key->arp_op = htons(output->ip.proto);
1010 		memcpy(arp_key->arp_sha, output->ipv4.arp.sha, ETH_ALEN);
1011 		memcpy(arp_key->arp_tha, output->ipv4.arp.tha, ETH_ALEN);
1012 	}
1013 
1014 	if ((swkey->eth.type == htons(ETH_P_IP) ||
1015 	     swkey->eth.type == htons(ETH_P_IPV6)) &&
1016 	     swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1017 
1018 		if (swkey->ip.proto == IPPROTO_TCP) {
1019 			struct ovs_key_tcp *tcp_key;
1020 
1021 			nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1022 			if (!nla)
1023 				goto nla_put_failure;
1024 			tcp_key = nla_data(nla);
1025 			if (swkey->eth.type == htons(ETH_P_IP)) {
1026 				tcp_key->tcp_src = output->ipv4.tp.src;
1027 				tcp_key->tcp_dst = output->ipv4.tp.dst;
1028 				if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1029 						 output->ipv4.tp.flags))
1030 					goto nla_put_failure;
1031 			} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1032 				tcp_key->tcp_src = output->ipv6.tp.src;
1033 				tcp_key->tcp_dst = output->ipv6.tp.dst;
1034 				if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1035 						 output->ipv6.tp.flags))
1036 					goto nla_put_failure;
1037 			}
1038 		} else if (swkey->ip.proto == IPPROTO_UDP) {
1039 			struct ovs_key_udp *udp_key;
1040 
1041 			nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1042 			if (!nla)
1043 				goto nla_put_failure;
1044 			udp_key = nla_data(nla);
1045 			if (swkey->eth.type == htons(ETH_P_IP)) {
1046 				udp_key->udp_src = output->ipv4.tp.src;
1047 				udp_key->udp_dst = output->ipv4.tp.dst;
1048 			} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1049 				udp_key->udp_src = output->ipv6.tp.src;
1050 				udp_key->udp_dst = output->ipv6.tp.dst;
1051 			}
1052 		} else if (swkey->ip.proto == IPPROTO_SCTP) {
1053 			struct ovs_key_sctp *sctp_key;
1054 
1055 			nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1056 			if (!nla)
1057 				goto nla_put_failure;
1058 			sctp_key = nla_data(nla);
1059 			if (swkey->eth.type == htons(ETH_P_IP)) {
1060 				sctp_key->sctp_src = swkey->ipv4.tp.src;
1061 				sctp_key->sctp_dst = swkey->ipv4.tp.dst;
1062 			} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1063 				sctp_key->sctp_src = swkey->ipv6.tp.src;
1064 				sctp_key->sctp_dst = swkey->ipv6.tp.dst;
1065 			}
1066 		} else if (swkey->eth.type == htons(ETH_P_IP) &&
1067 			   swkey->ip.proto == IPPROTO_ICMP) {
1068 			struct ovs_key_icmp *icmp_key;
1069 
1070 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1071 			if (!nla)
1072 				goto nla_put_failure;
1073 			icmp_key = nla_data(nla);
1074 			icmp_key->icmp_type = ntohs(output->ipv4.tp.src);
1075 			icmp_key->icmp_code = ntohs(output->ipv4.tp.dst);
1076 		} else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1077 			   swkey->ip.proto == IPPROTO_ICMPV6) {
1078 			struct ovs_key_icmpv6 *icmpv6_key;
1079 
1080 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1081 						sizeof(*icmpv6_key));
1082 			if (!nla)
1083 				goto nla_put_failure;
1084 			icmpv6_key = nla_data(nla);
1085 			icmpv6_key->icmpv6_type = ntohs(output->ipv6.tp.src);
1086 			icmpv6_key->icmpv6_code = ntohs(output->ipv6.tp.dst);
1087 
1088 			if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1089 			    icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1090 				struct ovs_key_nd *nd_key;
1091 
1092 				nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1093 				if (!nla)
1094 					goto nla_put_failure;
1095 				nd_key = nla_data(nla);
1096 				memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1097 							sizeof(nd_key->nd_target));
1098 				memcpy(nd_key->nd_sll, output->ipv6.nd.sll, ETH_ALEN);
1099 				memcpy(nd_key->nd_tll, output->ipv6.nd.tll, ETH_ALEN);
1100 			}
1101 		}
1102 	}
1103 
1104 unencap:
1105 	if (encap)
1106 		nla_nest_end(skb, encap);
1107 
1108 	return 0;
1109 
1110 nla_put_failure:
1111 	return -EMSGSIZE;
1112 }
1113 
1114 #define MAX_ACTIONS_BUFSIZE	(32 * 1024)
1115 
1116 struct sw_flow_actions *ovs_nla_alloc_flow_actions(int size)
1117 {
1118 	struct sw_flow_actions *sfa;
1119 
1120 	if (size > MAX_ACTIONS_BUFSIZE)
1121 		return ERR_PTR(-EINVAL);
1122 
1123 	sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1124 	if (!sfa)
1125 		return ERR_PTR(-ENOMEM);
1126 
1127 	sfa->actions_len = 0;
1128 	return sfa;
1129 }
1130 
1131 /* RCU callback used by ovs_nla_free_flow_actions. */
1132 static void rcu_free_acts_callback(struct rcu_head *rcu)
1133 {
1134 	struct sw_flow_actions *sf_acts = container_of(rcu,
1135 			struct sw_flow_actions, rcu);
1136 	kfree(sf_acts);
1137 }
1138 
1139 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1140  * The caller must hold rcu_read_lock for this to be sensible. */
1141 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1142 {
1143 	call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
1144 }
1145 
1146 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1147 				       int attr_len)
1148 {
1149 
1150 	struct sw_flow_actions *acts;
1151 	int new_acts_size;
1152 	int req_size = NLA_ALIGN(attr_len);
1153 	int next_offset = offsetof(struct sw_flow_actions, actions) +
1154 					(*sfa)->actions_len;
1155 
1156 	if (req_size <= (ksize(*sfa) - next_offset))
1157 		goto out;
1158 
1159 	new_acts_size = ksize(*sfa) * 2;
1160 
1161 	if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1162 		if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1163 			return ERR_PTR(-EMSGSIZE);
1164 		new_acts_size = MAX_ACTIONS_BUFSIZE;
1165 	}
1166 
1167 	acts = ovs_nla_alloc_flow_actions(new_acts_size);
1168 	if (IS_ERR(acts))
1169 		return (void *)acts;
1170 
1171 	memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1172 	acts->actions_len = (*sfa)->actions_len;
1173 	kfree(*sfa);
1174 	*sfa = acts;
1175 
1176 out:
1177 	(*sfa)->actions_len += req_size;
1178 	return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1179 }
1180 
1181 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
1182 {
1183 	struct nlattr *a;
1184 
1185 	a = reserve_sfa_size(sfa, nla_attr_size(len));
1186 	if (IS_ERR(a))
1187 		return PTR_ERR(a);
1188 
1189 	a->nla_type = attrtype;
1190 	a->nla_len = nla_attr_size(len);
1191 
1192 	if (data)
1193 		memcpy(nla_data(a), data, len);
1194 	memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1195 
1196 	return 0;
1197 }
1198 
1199 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1200 					  int attrtype)
1201 {
1202 	int used = (*sfa)->actions_len;
1203 	int err;
1204 
1205 	err = add_action(sfa, attrtype, NULL, 0);
1206 	if (err)
1207 		return err;
1208 
1209 	return used;
1210 }
1211 
1212 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1213 					 int st_offset)
1214 {
1215 	struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1216 							       st_offset);
1217 
1218 	a->nla_len = sfa->actions_len - st_offset;
1219 }
1220 
1221 static int validate_and_copy_sample(const struct nlattr *attr,
1222 				    const struct sw_flow_key *key, int depth,
1223 				    struct sw_flow_actions **sfa)
1224 {
1225 	const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1226 	const struct nlattr *probability, *actions;
1227 	const struct nlattr *a;
1228 	int rem, start, err, st_acts;
1229 
1230 	memset(attrs, 0, sizeof(attrs));
1231 	nla_for_each_nested(a, attr, rem) {
1232 		int type = nla_type(a);
1233 		if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1234 			return -EINVAL;
1235 		attrs[type] = a;
1236 	}
1237 	if (rem)
1238 		return -EINVAL;
1239 
1240 	probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1241 	if (!probability || nla_len(probability) != sizeof(u32))
1242 		return -EINVAL;
1243 
1244 	actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1245 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1246 		return -EINVAL;
1247 
1248 	/* validation done, copy sample action. */
1249 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
1250 	if (start < 0)
1251 		return start;
1252 	err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1253 			 nla_data(probability), sizeof(u32));
1254 	if (err)
1255 		return err;
1256 	st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
1257 	if (st_acts < 0)
1258 		return st_acts;
1259 
1260 	err = ovs_nla_copy_actions(actions, key, depth + 1, sfa);
1261 	if (err)
1262 		return err;
1263 
1264 	add_nested_action_end(*sfa, st_acts);
1265 	add_nested_action_end(*sfa, start);
1266 
1267 	return 0;
1268 }
1269 
1270 static int validate_tp_port(const struct sw_flow_key *flow_key)
1271 {
1272 	if (flow_key->eth.type == htons(ETH_P_IP)) {
1273 		if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
1274 			return 0;
1275 	} else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
1276 		if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
1277 			return 0;
1278 	}
1279 
1280 	return -EINVAL;
1281 }
1282 
1283 void ovs_match_init(struct sw_flow_match *match,
1284 		    struct sw_flow_key *key,
1285 		    struct sw_flow_mask *mask)
1286 {
1287 	memset(match, 0, sizeof(*match));
1288 	match->key = key;
1289 	match->mask = mask;
1290 
1291 	memset(key, 0, sizeof(*key));
1292 
1293 	if (mask) {
1294 		memset(&mask->key, 0, sizeof(mask->key));
1295 		mask->range.start = mask->range.end = 0;
1296 	}
1297 }
1298 
1299 static int validate_and_copy_set_tun(const struct nlattr *attr,
1300 				     struct sw_flow_actions **sfa)
1301 {
1302 	struct sw_flow_match match;
1303 	struct sw_flow_key key;
1304 	int err, start;
1305 
1306 	ovs_match_init(&match, &key, NULL);
1307 	err = ipv4_tun_from_nlattr(nla_data(attr), &match, false);
1308 	if (err)
1309 		return err;
1310 
1311 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
1312 	if (start < 0)
1313 		return start;
1314 
1315 	err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &match.key->tun_key,
1316 			sizeof(match.key->tun_key));
1317 	add_nested_action_end(*sfa, start);
1318 
1319 	return err;
1320 }
1321 
1322 static int validate_set(const struct nlattr *a,
1323 			const struct sw_flow_key *flow_key,
1324 			struct sw_flow_actions **sfa,
1325 			bool *set_tun)
1326 {
1327 	const struct nlattr *ovs_key = nla_data(a);
1328 	int key_type = nla_type(ovs_key);
1329 
1330 	/* There can be only one key in a action */
1331 	if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1332 		return -EINVAL;
1333 
1334 	if (key_type > OVS_KEY_ATTR_MAX ||
1335 	    (ovs_key_lens[key_type] != nla_len(ovs_key) &&
1336 	     ovs_key_lens[key_type] != -1))
1337 		return -EINVAL;
1338 
1339 	switch (key_type) {
1340 	const struct ovs_key_ipv4 *ipv4_key;
1341 	const struct ovs_key_ipv6 *ipv6_key;
1342 	int err;
1343 
1344 	case OVS_KEY_ATTR_PRIORITY:
1345 	case OVS_KEY_ATTR_SKB_MARK:
1346 	case OVS_KEY_ATTR_ETHERNET:
1347 		break;
1348 
1349 	case OVS_KEY_ATTR_TUNNEL:
1350 		*set_tun = true;
1351 		err = validate_and_copy_set_tun(a, sfa);
1352 		if (err)
1353 			return err;
1354 		break;
1355 
1356 	case OVS_KEY_ATTR_IPV4:
1357 		if (flow_key->eth.type != htons(ETH_P_IP))
1358 			return -EINVAL;
1359 
1360 		if (!flow_key->ip.proto)
1361 			return -EINVAL;
1362 
1363 		ipv4_key = nla_data(ovs_key);
1364 		if (ipv4_key->ipv4_proto != flow_key->ip.proto)
1365 			return -EINVAL;
1366 
1367 		if (ipv4_key->ipv4_frag != flow_key->ip.frag)
1368 			return -EINVAL;
1369 
1370 		break;
1371 
1372 	case OVS_KEY_ATTR_IPV6:
1373 		if (flow_key->eth.type != htons(ETH_P_IPV6))
1374 			return -EINVAL;
1375 
1376 		if (!flow_key->ip.proto)
1377 			return -EINVAL;
1378 
1379 		ipv6_key = nla_data(ovs_key);
1380 		if (ipv6_key->ipv6_proto != flow_key->ip.proto)
1381 			return -EINVAL;
1382 
1383 		if (ipv6_key->ipv6_frag != flow_key->ip.frag)
1384 			return -EINVAL;
1385 
1386 		if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
1387 			return -EINVAL;
1388 
1389 		break;
1390 
1391 	case OVS_KEY_ATTR_TCP:
1392 		if (flow_key->ip.proto != IPPROTO_TCP)
1393 			return -EINVAL;
1394 
1395 		return validate_tp_port(flow_key);
1396 
1397 	case OVS_KEY_ATTR_UDP:
1398 		if (flow_key->ip.proto != IPPROTO_UDP)
1399 			return -EINVAL;
1400 
1401 		return validate_tp_port(flow_key);
1402 
1403 	case OVS_KEY_ATTR_SCTP:
1404 		if (flow_key->ip.proto != IPPROTO_SCTP)
1405 			return -EINVAL;
1406 
1407 		return validate_tp_port(flow_key);
1408 
1409 	default:
1410 		return -EINVAL;
1411 	}
1412 
1413 	return 0;
1414 }
1415 
1416 static int validate_userspace(const struct nlattr *attr)
1417 {
1418 	static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
1419 		[OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
1420 		[OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
1421 	};
1422 	struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
1423 	int error;
1424 
1425 	error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
1426 				 attr, userspace_policy);
1427 	if (error)
1428 		return error;
1429 
1430 	if (!a[OVS_USERSPACE_ATTR_PID] ||
1431 	    !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
1432 		return -EINVAL;
1433 
1434 	return 0;
1435 }
1436 
1437 static int copy_action(const struct nlattr *from,
1438 		       struct sw_flow_actions **sfa)
1439 {
1440 	int totlen = NLA_ALIGN(from->nla_len);
1441 	struct nlattr *to;
1442 
1443 	to = reserve_sfa_size(sfa, from->nla_len);
1444 	if (IS_ERR(to))
1445 		return PTR_ERR(to);
1446 
1447 	memcpy(to, from, totlen);
1448 	return 0;
1449 }
1450 
1451 int ovs_nla_copy_actions(const struct nlattr *attr,
1452 			 const struct sw_flow_key *key,
1453 			 int depth,
1454 			 struct sw_flow_actions **sfa)
1455 {
1456 	const struct nlattr *a;
1457 	int rem, err;
1458 
1459 	if (depth >= SAMPLE_ACTION_DEPTH)
1460 		return -EOVERFLOW;
1461 
1462 	nla_for_each_nested(a, attr, rem) {
1463 		/* Expected argument lengths, (u32)-1 for variable length. */
1464 		static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
1465 			[OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
1466 			[OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
1467 			[OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
1468 			[OVS_ACTION_ATTR_POP_VLAN] = 0,
1469 			[OVS_ACTION_ATTR_SET] = (u32)-1,
1470 			[OVS_ACTION_ATTR_SAMPLE] = (u32)-1
1471 		};
1472 		const struct ovs_action_push_vlan *vlan;
1473 		int type = nla_type(a);
1474 		bool skip_copy;
1475 
1476 		if (type > OVS_ACTION_ATTR_MAX ||
1477 		    (action_lens[type] != nla_len(a) &&
1478 		     action_lens[type] != (u32)-1))
1479 			return -EINVAL;
1480 
1481 		skip_copy = false;
1482 		switch (type) {
1483 		case OVS_ACTION_ATTR_UNSPEC:
1484 			return -EINVAL;
1485 
1486 		case OVS_ACTION_ATTR_USERSPACE:
1487 			err = validate_userspace(a);
1488 			if (err)
1489 				return err;
1490 			break;
1491 
1492 		case OVS_ACTION_ATTR_OUTPUT:
1493 			if (nla_get_u32(a) >= DP_MAX_PORTS)
1494 				return -EINVAL;
1495 			break;
1496 
1497 
1498 		case OVS_ACTION_ATTR_POP_VLAN:
1499 			break;
1500 
1501 		case OVS_ACTION_ATTR_PUSH_VLAN:
1502 			vlan = nla_data(a);
1503 			if (vlan->vlan_tpid != htons(ETH_P_8021Q))
1504 				return -EINVAL;
1505 			if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
1506 				return -EINVAL;
1507 			break;
1508 
1509 		case OVS_ACTION_ATTR_SET:
1510 			err = validate_set(a, key, sfa, &skip_copy);
1511 			if (err)
1512 				return err;
1513 			break;
1514 
1515 		case OVS_ACTION_ATTR_SAMPLE:
1516 			err = validate_and_copy_sample(a, key, depth, sfa);
1517 			if (err)
1518 				return err;
1519 			skip_copy = true;
1520 			break;
1521 
1522 		default:
1523 			return -EINVAL;
1524 		}
1525 		if (!skip_copy) {
1526 			err = copy_action(a, sfa);
1527 			if (err)
1528 				return err;
1529 		}
1530 	}
1531 
1532 	if (rem > 0)
1533 		return -EINVAL;
1534 
1535 	return 0;
1536 }
1537 
1538 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1539 {
1540 	const struct nlattr *a;
1541 	struct nlattr *start;
1542 	int err = 0, rem;
1543 
1544 	start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1545 	if (!start)
1546 		return -EMSGSIZE;
1547 
1548 	nla_for_each_nested(a, attr, rem) {
1549 		int type = nla_type(a);
1550 		struct nlattr *st_sample;
1551 
1552 		switch (type) {
1553 		case OVS_SAMPLE_ATTR_PROBABILITY:
1554 			if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
1555 				    sizeof(u32), nla_data(a)))
1556 				return -EMSGSIZE;
1557 			break;
1558 		case OVS_SAMPLE_ATTR_ACTIONS:
1559 			st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1560 			if (!st_sample)
1561 				return -EMSGSIZE;
1562 			err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
1563 			if (err)
1564 				return err;
1565 			nla_nest_end(skb, st_sample);
1566 			break;
1567 		}
1568 	}
1569 
1570 	nla_nest_end(skb, start);
1571 	return err;
1572 }
1573 
1574 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1575 {
1576 	const struct nlattr *ovs_key = nla_data(a);
1577 	int key_type = nla_type(ovs_key);
1578 	struct nlattr *start;
1579 	int err;
1580 
1581 	switch (key_type) {
1582 	case OVS_KEY_ATTR_IPV4_TUNNEL:
1583 		start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1584 		if (!start)
1585 			return -EMSGSIZE;
1586 
1587 		err = ipv4_tun_to_nlattr(skb, nla_data(ovs_key),
1588 					     nla_data(ovs_key));
1589 		if (err)
1590 			return err;
1591 		nla_nest_end(skb, start);
1592 		break;
1593 	default:
1594 		if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1595 			return -EMSGSIZE;
1596 		break;
1597 	}
1598 
1599 	return 0;
1600 }
1601 
1602 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
1603 {
1604 	const struct nlattr *a;
1605 	int rem, err;
1606 
1607 	nla_for_each_attr(a, attr, len, rem) {
1608 		int type = nla_type(a);
1609 
1610 		switch (type) {
1611 		case OVS_ACTION_ATTR_SET:
1612 			err = set_action_to_attr(a, skb);
1613 			if (err)
1614 				return err;
1615 			break;
1616 
1617 		case OVS_ACTION_ATTR_SAMPLE:
1618 			err = sample_action_to_attr(a, skb);
1619 			if (err)
1620 				return err;
1621 			break;
1622 		default:
1623 			if (nla_put(skb, type, nla_len(a), nla_data(a)))
1624 				return -EMSGSIZE;
1625 			break;
1626 		}
1627 	}
1628 
1629 	return 0;
1630 }
1631