xref: /linux/net/openvswitch/flow_netlink.c (revision fee6d4c777a125e56de9370db3b2bf359bf958d6)
1 /*
2  * Copyright (c) 2007-2014 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 
21 #include "flow.h"
22 #include "datapath.h"
23 #include <linux/uaccess.h>
24 #include <linux/netdevice.h>
25 #include <linux/etherdevice.h>
26 #include <linux/if_ether.h>
27 #include <linux/if_vlan.h>
28 #include <net/llc_pdu.h>
29 #include <linux/kernel.h>
30 #include <linux/jhash.h>
31 #include <linux/jiffies.h>
32 #include <linux/llc.h>
33 #include <linux/module.h>
34 #include <linux/in.h>
35 #include <linux/rcupdate.h>
36 #include <linux/if_arp.h>
37 #include <linux/ip.h>
38 #include <linux/ipv6.h>
39 #include <linux/sctp.h>
40 #include <linux/tcp.h>
41 #include <linux/udp.h>
42 #include <linux/icmp.h>
43 #include <linux/icmpv6.h>
44 #include <linux/rculist.h>
45 #include <net/geneve.h>
46 #include <net/ip.h>
47 #include <net/ipv6.h>
48 #include <net/ndisc.h>
49 #include <net/mpls.h>
50 #include <net/vxlan.h>
51 
52 #include "flow_netlink.h"
53 
54 struct ovs_len_tbl {
55 	int len;
56 	const struct ovs_len_tbl *next;
57 };
58 
59 #define OVS_ATTR_NESTED -1
60 #define OVS_ATTR_VARIABLE -2
61 
62 static void update_range(struct sw_flow_match *match,
63 			 size_t offset, size_t size, bool is_mask)
64 {
65 	struct sw_flow_key_range *range;
66 	size_t start = rounddown(offset, sizeof(long));
67 	size_t end = roundup(offset + size, sizeof(long));
68 
69 	if (!is_mask)
70 		range = &match->range;
71 	else
72 		range = &match->mask->range;
73 
74 	if (range->start == range->end) {
75 		range->start = start;
76 		range->end = end;
77 		return;
78 	}
79 
80 	if (range->start > start)
81 		range->start = start;
82 
83 	if (range->end < end)
84 		range->end = end;
85 }
86 
87 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
88 	do { \
89 		update_range(match, offsetof(struct sw_flow_key, field),    \
90 			     sizeof((match)->key->field), is_mask);	    \
91 		if (is_mask)						    \
92 			(match)->mask->key.field = value;		    \
93 		else							    \
94 			(match)->key->field = value;		            \
95 	} while (0)
96 
97 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)	    \
98 	do {								    \
99 		update_range(match, offset, len, is_mask);		    \
100 		if (is_mask)						    \
101 			memcpy((u8 *)&(match)->mask->key + offset, value_p, \
102 			       len);					   \
103 		else							    \
104 			memcpy((u8 *)(match)->key + offset, value_p, len);  \
105 	} while (0)
106 
107 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)		      \
108 	SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
109 				  value_p, len, is_mask)
110 
111 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)		    \
112 	do {								    \
113 		update_range(match, offsetof(struct sw_flow_key, field),    \
114 			     sizeof((match)->key->field), is_mask);	    \
115 		if (is_mask)						    \
116 			memset((u8 *)&(match)->mask->key.field, value,      \
117 			       sizeof((match)->mask->key.field));	    \
118 		else							    \
119 			memset((u8 *)&(match)->key->field, value,           \
120 			       sizeof((match)->key->field));                \
121 	} while (0)
122 
123 static bool match_validate(const struct sw_flow_match *match,
124 			   u64 key_attrs, u64 mask_attrs, bool log)
125 {
126 	u64 key_expected = 1 << 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 &= ~((1 << OVS_KEY_ATTR_IPV4)
132 			| (1 << OVS_KEY_ATTR_IPV6)
133 			| (1 << OVS_KEY_ATTR_TCP)
134 			| (1 << OVS_KEY_ATTR_TCP_FLAGS)
135 			| (1 << OVS_KEY_ATTR_UDP)
136 			| (1 << OVS_KEY_ATTR_SCTP)
137 			| (1 << OVS_KEY_ATTR_ICMP)
138 			| (1 << OVS_KEY_ATTR_ICMPV6)
139 			| (1 << OVS_KEY_ATTR_ARP)
140 			| (1 << OVS_KEY_ATTR_ND)
141 			| (1 << OVS_KEY_ATTR_MPLS));
142 
143 	/* Always allowed mask fields. */
144 	mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
145 		       | (1 << OVS_KEY_ATTR_IN_PORT)
146 		       | (1 << 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 |= 1 << OVS_KEY_ATTR_ARP;
152 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
153 			mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
154 	}
155 
156 	if (eth_p_mpls(match->key->eth.type)) {
157 		key_expected |= 1 << OVS_KEY_ATTR_MPLS;
158 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
159 			mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
160 	}
161 
162 	if (match->key->eth.type == htons(ETH_P_IP)) {
163 		key_expected |= 1 << OVS_KEY_ATTR_IPV4;
164 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
165 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
166 
167 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
168 			if (match->key->ip.proto == IPPROTO_UDP) {
169 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
170 				if (match->mask && (match->mask->key.ip.proto == 0xff))
171 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
172 			}
173 
174 			if (match->key->ip.proto == IPPROTO_SCTP) {
175 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
176 				if (match->mask && (match->mask->key.ip.proto == 0xff))
177 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
178 			}
179 
180 			if (match->key->ip.proto == IPPROTO_TCP) {
181 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
182 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
183 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
184 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
185 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
186 				}
187 			}
188 
189 			if (match->key->ip.proto == IPPROTO_ICMP) {
190 				key_expected |= 1 << OVS_KEY_ATTR_ICMP;
191 				if (match->mask && (match->mask->key.ip.proto == 0xff))
192 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
193 			}
194 		}
195 	}
196 
197 	if (match->key->eth.type == htons(ETH_P_IPV6)) {
198 		key_expected |= 1 << OVS_KEY_ATTR_IPV6;
199 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
200 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
201 
202 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
203 			if (match->key->ip.proto == IPPROTO_UDP) {
204 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
205 				if (match->mask && (match->mask->key.ip.proto == 0xff))
206 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
207 			}
208 
209 			if (match->key->ip.proto == IPPROTO_SCTP) {
210 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
211 				if (match->mask && (match->mask->key.ip.proto == 0xff))
212 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
213 			}
214 
215 			if (match->key->ip.proto == IPPROTO_TCP) {
216 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
217 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
218 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
219 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
220 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
221 				}
222 			}
223 
224 			if (match->key->ip.proto == IPPROTO_ICMPV6) {
225 				key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
226 				if (match->mask && (match->mask->key.ip.proto == 0xff))
227 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
228 
229 				if (match->key->tp.src ==
230 						htons(NDISC_NEIGHBOUR_SOLICITATION) ||
231 				    match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
232 					key_expected |= 1 << OVS_KEY_ATTR_ND;
233 					if (match->mask && (match->mask->key.tp.src == htons(0xff)))
234 						mask_allowed |= 1 << OVS_KEY_ATTR_ND;
235 				}
236 			}
237 		}
238 	}
239 
240 	if ((key_attrs & key_expected) != key_expected) {
241 		/* Key attributes check failed. */
242 		OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
243 			  (unsigned long long)key_attrs,
244 			  (unsigned long long)key_expected);
245 		return false;
246 	}
247 
248 	if ((mask_attrs & mask_allowed) != mask_attrs) {
249 		/* Mask attributes check failed. */
250 		OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
251 			  (unsigned long long)mask_attrs,
252 			  (unsigned long long)mask_allowed);
253 		return false;
254 	}
255 
256 	return true;
257 }
258 
259 size_t ovs_tun_key_attr_size(void)
260 {
261 	/* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
262 	 * updating this function.
263 	 */
264 	return    nla_total_size(8)    /* OVS_TUNNEL_KEY_ATTR_ID */
265 		+ nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
266 		+ nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_DST */
267 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
268 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
269 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
270 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
271 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
272 		+ nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
273 		/* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS is mutually exclusive with
274 		 * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
275 		 */
276 		+ nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
277 		+ nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
278 }
279 
280 size_t ovs_key_attr_size(void)
281 {
282 	/* Whenever adding new OVS_KEY_ FIELDS, we should consider
283 	 * updating this function.
284 	 */
285 	BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 26);
286 
287 	return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
288 		+ nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
289 		  + ovs_tun_key_attr_size()
290 		+ nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
291 		+ nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
292 		+ nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
293 		+ nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
294 		+ nla_total_size(1)   /* OVS_KEY_ATTR_CT_STATE */
295 		+ nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
296 		+ nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
297 		+ nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABEL */
298 		+ nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
299 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
300 		+ nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
301 		+ nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
302 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
303 		+ nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
304 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
305 		+ nla_total_size(28); /* OVS_KEY_ATTR_ND */
306 }
307 
308 static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
309 	[OVS_VXLAN_EXT_GBP]	    = { .len = sizeof(u32) },
310 };
311 
312 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
313 	[OVS_TUNNEL_KEY_ATTR_ID]	    = { .len = sizeof(u64) },
314 	[OVS_TUNNEL_KEY_ATTR_IPV4_SRC]	    = { .len = sizeof(u32) },
315 	[OVS_TUNNEL_KEY_ATTR_IPV4_DST]	    = { .len = sizeof(u32) },
316 	[OVS_TUNNEL_KEY_ATTR_TOS]	    = { .len = 1 },
317 	[OVS_TUNNEL_KEY_ATTR_TTL]	    = { .len = 1 },
318 	[OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
319 	[OVS_TUNNEL_KEY_ATTR_CSUM]	    = { .len = 0 },
320 	[OVS_TUNNEL_KEY_ATTR_TP_SRC]	    = { .len = sizeof(u16) },
321 	[OVS_TUNNEL_KEY_ATTR_TP_DST]	    = { .len = sizeof(u16) },
322 	[OVS_TUNNEL_KEY_ATTR_OAM]	    = { .len = 0 },
323 	[OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_VARIABLE },
324 	[OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED,
325 						.next = ovs_vxlan_ext_key_lens },
326 	[OVS_TUNNEL_KEY_ATTR_IPV6_SRC]      = { .len = sizeof(struct in6_addr) },
327 	[OVS_TUNNEL_KEY_ATTR_IPV6_DST]      = { .len = sizeof(struct in6_addr) },
328 };
329 
330 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
331 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
332 	[OVS_KEY_ATTR_ENCAP]	 = { .len = OVS_ATTR_NESTED },
333 	[OVS_KEY_ATTR_PRIORITY]	 = { .len = sizeof(u32) },
334 	[OVS_KEY_ATTR_IN_PORT]	 = { .len = sizeof(u32) },
335 	[OVS_KEY_ATTR_SKB_MARK]	 = { .len = sizeof(u32) },
336 	[OVS_KEY_ATTR_ETHERNET]	 = { .len = sizeof(struct ovs_key_ethernet) },
337 	[OVS_KEY_ATTR_VLAN]	 = { .len = sizeof(__be16) },
338 	[OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
339 	[OVS_KEY_ATTR_IPV4]	 = { .len = sizeof(struct ovs_key_ipv4) },
340 	[OVS_KEY_ATTR_IPV6]	 = { .len = sizeof(struct ovs_key_ipv6) },
341 	[OVS_KEY_ATTR_TCP]	 = { .len = sizeof(struct ovs_key_tcp) },
342 	[OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
343 	[OVS_KEY_ATTR_UDP]	 = { .len = sizeof(struct ovs_key_udp) },
344 	[OVS_KEY_ATTR_SCTP]	 = { .len = sizeof(struct ovs_key_sctp) },
345 	[OVS_KEY_ATTR_ICMP]	 = { .len = sizeof(struct ovs_key_icmp) },
346 	[OVS_KEY_ATTR_ICMPV6]	 = { .len = sizeof(struct ovs_key_icmpv6) },
347 	[OVS_KEY_ATTR_ARP]	 = { .len = sizeof(struct ovs_key_arp) },
348 	[OVS_KEY_ATTR_ND]	 = { .len = sizeof(struct ovs_key_nd) },
349 	[OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
350 	[OVS_KEY_ATTR_DP_HASH]	 = { .len = sizeof(u32) },
351 	[OVS_KEY_ATTR_TUNNEL]	 = { .len = OVS_ATTR_NESTED,
352 				     .next = ovs_tunnel_key_lens, },
353 	[OVS_KEY_ATTR_MPLS]	 = { .len = sizeof(struct ovs_key_mpls) },
354 	[OVS_KEY_ATTR_CT_STATE]	 = { .len = sizeof(u8) },
355 	[OVS_KEY_ATTR_CT_ZONE]	 = { .len = sizeof(u16) },
356 	[OVS_KEY_ATTR_CT_MARK]	 = { .len = sizeof(u32) },
357 	[OVS_KEY_ATTR_CT_LABEL]	 = { .len = sizeof(struct ovs_key_ct_label) },
358 };
359 
360 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
361 {
362 	return expected_len == attr_len ||
363 	       expected_len == OVS_ATTR_NESTED ||
364 	       expected_len == OVS_ATTR_VARIABLE;
365 }
366 
367 static bool is_all_zero(const u8 *fp, size_t size)
368 {
369 	int i;
370 
371 	if (!fp)
372 		return false;
373 
374 	for (i = 0; i < size; i++)
375 		if (fp[i])
376 			return false;
377 
378 	return true;
379 }
380 
381 static int __parse_flow_nlattrs(const struct nlattr *attr,
382 				const struct nlattr *a[],
383 				u64 *attrsp, bool log, bool nz)
384 {
385 	const struct nlattr *nla;
386 	u64 attrs;
387 	int rem;
388 
389 	attrs = *attrsp;
390 	nla_for_each_nested(nla, attr, rem) {
391 		u16 type = nla_type(nla);
392 		int expected_len;
393 
394 		if (type > OVS_KEY_ATTR_MAX) {
395 			OVS_NLERR(log, "Key type %d is out of range max %d",
396 				  type, OVS_KEY_ATTR_MAX);
397 			return -EINVAL;
398 		}
399 
400 		if (attrs & (1 << type)) {
401 			OVS_NLERR(log, "Duplicate key (type %d).", type);
402 			return -EINVAL;
403 		}
404 
405 		expected_len = ovs_key_lens[type].len;
406 		if (!check_attr_len(nla_len(nla), expected_len)) {
407 			OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
408 				  type, nla_len(nla), expected_len);
409 			return -EINVAL;
410 		}
411 
412 		if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
413 			attrs |= 1 << type;
414 			a[type] = nla;
415 		}
416 	}
417 	if (rem) {
418 		OVS_NLERR(log, "Message has %d unknown bytes.", rem);
419 		return -EINVAL;
420 	}
421 
422 	*attrsp = attrs;
423 	return 0;
424 }
425 
426 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
427 				   const struct nlattr *a[], u64 *attrsp,
428 				   bool log)
429 {
430 	return __parse_flow_nlattrs(attr, a, attrsp, log, true);
431 }
432 
433 static int parse_flow_nlattrs(const struct nlattr *attr,
434 			      const struct nlattr *a[], u64 *attrsp,
435 			      bool log)
436 {
437 	return __parse_flow_nlattrs(attr, a, attrsp, log, false);
438 }
439 
440 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
441 				     struct sw_flow_match *match, bool is_mask,
442 				     bool log)
443 {
444 	unsigned long opt_key_offset;
445 
446 	if (nla_len(a) > sizeof(match->key->tun_opts)) {
447 		OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
448 			  nla_len(a), sizeof(match->key->tun_opts));
449 		return -EINVAL;
450 	}
451 
452 	if (nla_len(a) % 4 != 0) {
453 		OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
454 			  nla_len(a));
455 		return -EINVAL;
456 	}
457 
458 	/* We need to record the length of the options passed
459 	 * down, otherwise packets with the same format but
460 	 * additional options will be silently matched.
461 	 */
462 	if (!is_mask) {
463 		SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
464 				false);
465 	} else {
466 		/* This is somewhat unusual because it looks at
467 		 * both the key and mask while parsing the
468 		 * attributes (and by extension assumes the key
469 		 * is parsed first). Normally, we would verify
470 		 * that each is the correct length and that the
471 		 * attributes line up in the validate function.
472 		 * However, that is difficult because this is
473 		 * variable length and we won't have the
474 		 * information later.
475 		 */
476 		if (match->key->tun_opts_len != nla_len(a)) {
477 			OVS_NLERR(log, "Geneve option len %d != mask len %d",
478 				  match->key->tun_opts_len, nla_len(a));
479 			return -EINVAL;
480 		}
481 
482 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
483 	}
484 
485 	opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
486 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
487 				  nla_len(a), is_mask);
488 	return 0;
489 }
490 
491 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
492 				     struct sw_flow_match *match, bool is_mask,
493 				     bool log)
494 {
495 	struct nlattr *a;
496 	int rem;
497 	unsigned long opt_key_offset;
498 	struct vxlan_metadata opts;
499 
500 	BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
501 
502 	memset(&opts, 0, sizeof(opts));
503 	nla_for_each_nested(a, attr, rem) {
504 		int type = nla_type(a);
505 
506 		if (type > OVS_VXLAN_EXT_MAX) {
507 			OVS_NLERR(log, "VXLAN extension %d out of range max %d",
508 				  type, OVS_VXLAN_EXT_MAX);
509 			return -EINVAL;
510 		}
511 
512 		if (!check_attr_len(nla_len(a),
513 				    ovs_vxlan_ext_key_lens[type].len)) {
514 			OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
515 				  type, nla_len(a),
516 				  ovs_vxlan_ext_key_lens[type].len);
517 			return -EINVAL;
518 		}
519 
520 		switch (type) {
521 		case OVS_VXLAN_EXT_GBP:
522 			opts.gbp = nla_get_u32(a);
523 			break;
524 		default:
525 			OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
526 				  type);
527 			return -EINVAL;
528 		}
529 	}
530 	if (rem) {
531 		OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
532 			  rem);
533 		return -EINVAL;
534 	}
535 
536 	if (!is_mask)
537 		SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
538 	else
539 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
540 
541 	opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
542 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
543 				  is_mask);
544 	return 0;
545 }
546 
547 static int ip_tun_from_nlattr(const struct nlattr *attr,
548 			      struct sw_flow_match *match, bool is_mask,
549 			      bool log)
550 {
551 	struct nlattr *a;
552 	int rem;
553 	bool ttl = false;
554 	__be16 tun_flags = 0, ipv4 = false, ipv6 = false;
555 	int opts_type = 0;
556 
557 	nla_for_each_nested(a, attr, rem) {
558 		int type = nla_type(a);
559 		int err;
560 
561 		if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
562 			OVS_NLERR(log, "Tunnel attr %d out of range max %d",
563 				  type, OVS_TUNNEL_KEY_ATTR_MAX);
564 			return -EINVAL;
565 		}
566 
567 		if (!check_attr_len(nla_len(a),
568 				    ovs_tunnel_key_lens[type].len)) {
569 			OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
570 				  type, nla_len(a), ovs_tunnel_key_lens[type].len);
571 			return -EINVAL;
572 		}
573 
574 		switch (type) {
575 		case OVS_TUNNEL_KEY_ATTR_ID:
576 			SW_FLOW_KEY_PUT(match, tun_key.tun_id,
577 					nla_get_be64(a), is_mask);
578 			tun_flags |= TUNNEL_KEY;
579 			break;
580 		case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
581 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
582 					nla_get_in_addr(a), is_mask);
583 			ipv4 = true;
584 			break;
585 		case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
586 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
587 					nla_get_in_addr(a), is_mask);
588 			ipv4 = true;
589 			break;
590 		case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
591 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
592 					nla_get_in6_addr(a), is_mask);
593 			ipv6 = true;
594 			break;
595 		case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
596 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
597 					nla_get_in6_addr(a), is_mask);
598 			ipv6 = true;
599 			break;
600 		case OVS_TUNNEL_KEY_ATTR_TOS:
601 			SW_FLOW_KEY_PUT(match, tun_key.tos,
602 					nla_get_u8(a), is_mask);
603 			break;
604 		case OVS_TUNNEL_KEY_ATTR_TTL:
605 			SW_FLOW_KEY_PUT(match, tun_key.ttl,
606 					nla_get_u8(a), is_mask);
607 			ttl = true;
608 			break;
609 		case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
610 			tun_flags |= TUNNEL_DONT_FRAGMENT;
611 			break;
612 		case OVS_TUNNEL_KEY_ATTR_CSUM:
613 			tun_flags |= TUNNEL_CSUM;
614 			break;
615 		case OVS_TUNNEL_KEY_ATTR_TP_SRC:
616 			SW_FLOW_KEY_PUT(match, tun_key.tp_src,
617 					nla_get_be16(a), is_mask);
618 			break;
619 		case OVS_TUNNEL_KEY_ATTR_TP_DST:
620 			SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
621 					nla_get_be16(a), is_mask);
622 			break;
623 		case OVS_TUNNEL_KEY_ATTR_OAM:
624 			tun_flags |= TUNNEL_OAM;
625 			break;
626 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
627 			if (opts_type) {
628 				OVS_NLERR(log, "Multiple metadata blocks provided");
629 				return -EINVAL;
630 			}
631 
632 			err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
633 			if (err)
634 				return err;
635 
636 			tun_flags |= TUNNEL_GENEVE_OPT;
637 			opts_type = type;
638 			break;
639 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
640 			if (opts_type) {
641 				OVS_NLERR(log, "Multiple metadata blocks provided");
642 				return -EINVAL;
643 			}
644 
645 			err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
646 			if (err)
647 				return err;
648 
649 			tun_flags |= TUNNEL_VXLAN_OPT;
650 			opts_type = type;
651 			break;
652 		default:
653 			OVS_NLERR(log, "Unknown IP tunnel attribute %d",
654 				  type);
655 			return -EINVAL;
656 		}
657 	}
658 
659 	SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
660 	if (is_mask)
661 		SW_FLOW_KEY_MEMSET_FIELD(match, tun_proto, 0xff, true);
662 	else
663 		SW_FLOW_KEY_PUT(match, tun_proto, ipv6 ? AF_INET6 : AF_INET,
664 				false);
665 
666 	if (rem > 0) {
667 		OVS_NLERR(log, "IP tunnel attribute has %d unknown bytes.",
668 			  rem);
669 		return -EINVAL;
670 	}
671 
672 	if (ipv4 && ipv6) {
673 		OVS_NLERR(log, "Mixed IPv4 and IPv6 tunnel attributes");
674 		return -EINVAL;
675 	}
676 
677 	if (!is_mask) {
678 		if (!ipv4 && !ipv6) {
679 			OVS_NLERR(log, "IP tunnel dst address not specified");
680 			return -EINVAL;
681 		}
682 		if (ipv4 && !match->key->tun_key.u.ipv4.dst) {
683 			OVS_NLERR(log, "IPv4 tunnel dst address is zero");
684 			return -EINVAL;
685 		}
686 		if (ipv6 && ipv6_addr_any(&match->key->tun_key.u.ipv6.dst)) {
687 			OVS_NLERR(log, "IPv6 tunnel dst address is zero");
688 			return -EINVAL;
689 		}
690 
691 		if (!ttl) {
692 			OVS_NLERR(log, "IP tunnel TTL not specified.");
693 			return -EINVAL;
694 		}
695 	}
696 
697 	return opts_type;
698 }
699 
700 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
701 			       const void *tun_opts, int swkey_tun_opts_len)
702 {
703 	const struct vxlan_metadata *opts = tun_opts;
704 	struct nlattr *nla;
705 
706 	nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
707 	if (!nla)
708 		return -EMSGSIZE;
709 
710 	if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
711 		return -EMSGSIZE;
712 
713 	nla_nest_end(skb, nla);
714 	return 0;
715 }
716 
717 static int __ip_tun_to_nlattr(struct sk_buff *skb,
718 			      const struct ip_tunnel_key *output,
719 			      const void *tun_opts, int swkey_tun_opts_len,
720 			      unsigned short tun_proto)
721 {
722 	if (output->tun_flags & TUNNEL_KEY &&
723 	    nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
724 		return -EMSGSIZE;
725 	switch (tun_proto) {
726 	case AF_INET:
727 		if (output->u.ipv4.src &&
728 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
729 				    output->u.ipv4.src))
730 			return -EMSGSIZE;
731 		if (output->u.ipv4.dst &&
732 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
733 				    output->u.ipv4.dst))
734 			return -EMSGSIZE;
735 		break;
736 	case AF_INET6:
737 		if (!ipv6_addr_any(&output->u.ipv6.src) &&
738 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
739 				     &output->u.ipv6.src))
740 			return -EMSGSIZE;
741 		if (!ipv6_addr_any(&output->u.ipv6.dst) &&
742 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
743 				     &output->u.ipv6.dst))
744 			return -EMSGSIZE;
745 		break;
746 	}
747 	if (output->tos &&
748 	    nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
749 		return -EMSGSIZE;
750 	if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
751 		return -EMSGSIZE;
752 	if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
753 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
754 		return -EMSGSIZE;
755 	if ((output->tun_flags & TUNNEL_CSUM) &&
756 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
757 		return -EMSGSIZE;
758 	if (output->tp_src &&
759 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
760 		return -EMSGSIZE;
761 	if (output->tp_dst &&
762 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
763 		return -EMSGSIZE;
764 	if ((output->tun_flags & TUNNEL_OAM) &&
765 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
766 		return -EMSGSIZE;
767 	if (tun_opts) {
768 		if (output->tun_flags & TUNNEL_GENEVE_OPT &&
769 		    nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
770 			    swkey_tun_opts_len, tun_opts))
771 			return -EMSGSIZE;
772 		else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
773 			 vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
774 			return -EMSGSIZE;
775 	}
776 
777 	return 0;
778 }
779 
780 static int ip_tun_to_nlattr(struct sk_buff *skb,
781 			    const struct ip_tunnel_key *output,
782 			    const void *tun_opts, int swkey_tun_opts_len,
783 			    unsigned short tun_proto)
784 {
785 	struct nlattr *nla;
786 	int err;
787 
788 	nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
789 	if (!nla)
790 		return -EMSGSIZE;
791 
792 	err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
793 				 tun_proto);
794 	if (err)
795 		return err;
796 
797 	nla_nest_end(skb, nla);
798 	return 0;
799 }
800 
801 int ovs_nla_put_egress_tunnel_key(struct sk_buff *skb,
802 				  const struct ip_tunnel_info *egress_tun_info,
803 				  const void *egress_tun_opts)
804 {
805 	return __ip_tun_to_nlattr(skb, &egress_tun_info->key,
806 				  egress_tun_opts,
807 				  egress_tun_info->options_len,
808 				  ip_tunnel_info_af(egress_tun_info));
809 }
810 
811 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
812 				 u64 *attrs, const struct nlattr **a,
813 				 bool is_mask, bool log)
814 {
815 	if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
816 		u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
817 
818 		SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
819 		*attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
820 	}
821 
822 	if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
823 		u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
824 
825 		SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
826 		*attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
827 	}
828 
829 	if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
830 		SW_FLOW_KEY_PUT(match, phy.priority,
831 			  nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
832 		*attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
833 	}
834 
835 	if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
836 		u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
837 
838 		if (is_mask) {
839 			in_port = 0xffffffff; /* Always exact match in_port. */
840 		} else if (in_port >= DP_MAX_PORTS) {
841 			OVS_NLERR(log, "Port %d exceeds max allowable %d",
842 				  in_port, DP_MAX_PORTS);
843 			return -EINVAL;
844 		}
845 
846 		SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
847 		*attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
848 	} else if (!is_mask) {
849 		SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
850 	}
851 
852 	if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
853 		uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
854 
855 		SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
856 		*attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
857 	}
858 	if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
859 		if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
860 				       is_mask, log) < 0)
861 			return -EINVAL;
862 		*attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
863 	}
864 
865 	if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
866 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
867 		u8 ct_state = nla_get_u8(a[OVS_KEY_ATTR_CT_STATE]);
868 
869 		SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
870 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
871 	}
872 	if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
873 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
874 		u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
875 
876 		SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
877 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
878 	}
879 	if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
880 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
881 		u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
882 
883 		SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
884 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
885 	}
886 	if (*attrs & (1 << OVS_KEY_ATTR_CT_LABEL) &&
887 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABEL)) {
888 		const struct ovs_key_ct_label *cl;
889 
890 		cl = nla_data(a[OVS_KEY_ATTR_CT_LABEL]);
891 		SW_FLOW_KEY_MEMCPY(match, ct.label, cl->ct_label,
892 				   sizeof(*cl), is_mask);
893 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABEL);
894 	}
895 	return 0;
896 }
897 
898 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
899 				u64 attrs, const struct nlattr **a,
900 				bool is_mask, bool log)
901 {
902 	int err;
903 
904 	err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
905 	if (err)
906 		return err;
907 
908 	if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
909 		const struct ovs_key_ethernet *eth_key;
910 
911 		eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
912 		SW_FLOW_KEY_MEMCPY(match, eth.src,
913 				eth_key->eth_src, ETH_ALEN, is_mask);
914 		SW_FLOW_KEY_MEMCPY(match, eth.dst,
915 				eth_key->eth_dst, ETH_ALEN, is_mask);
916 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
917 	}
918 
919 	if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
920 		__be16 tci;
921 
922 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
923 		if (!(tci & htons(VLAN_TAG_PRESENT))) {
924 			if (is_mask)
925 				OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
926 			else
927 				OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
928 
929 			return -EINVAL;
930 		}
931 
932 		SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
933 		attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
934 	}
935 
936 	if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
937 		__be16 eth_type;
938 
939 		eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
940 		if (is_mask) {
941 			/* Always exact match EtherType. */
942 			eth_type = htons(0xffff);
943 		} else if (!eth_proto_is_802_3(eth_type)) {
944 			OVS_NLERR(log, "EtherType %x is less than min %x",
945 				  ntohs(eth_type), ETH_P_802_3_MIN);
946 			return -EINVAL;
947 		}
948 
949 		SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
950 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
951 	} else if (!is_mask) {
952 		SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
953 	}
954 
955 	if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
956 		const struct ovs_key_ipv4 *ipv4_key;
957 
958 		ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
959 		if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
960 			OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
961 				  ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
962 			return -EINVAL;
963 		}
964 		SW_FLOW_KEY_PUT(match, ip.proto,
965 				ipv4_key->ipv4_proto, is_mask);
966 		SW_FLOW_KEY_PUT(match, ip.tos,
967 				ipv4_key->ipv4_tos, is_mask);
968 		SW_FLOW_KEY_PUT(match, ip.ttl,
969 				ipv4_key->ipv4_ttl, is_mask);
970 		SW_FLOW_KEY_PUT(match, ip.frag,
971 				ipv4_key->ipv4_frag, is_mask);
972 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
973 				ipv4_key->ipv4_src, is_mask);
974 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
975 				ipv4_key->ipv4_dst, is_mask);
976 		attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
977 	}
978 
979 	if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
980 		const struct ovs_key_ipv6 *ipv6_key;
981 
982 		ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
983 		if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
984 			OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
985 				  ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
986 			return -EINVAL;
987 		}
988 
989 		if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
990 			OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
991 				  ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
992 			return -EINVAL;
993 		}
994 
995 		SW_FLOW_KEY_PUT(match, ipv6.label,
996 				ipv6_key->ipv6_label, is_mask);
997 		SW_FLOW_KEY_PUT(match, ip.proto,
998 				ipv6_key->ipv6_proto, is_mask);
999 		SW_FLOW_KEY_PUT(match, ip.tos,
1000 				ipv6_key->ipv6_tclass, is_mask);
1001 		SW_FLOW_KEY_PUT(match, ip.ttl,
1002 				ipv6_key->ipv6_hlimit, is_mask);
1003 		SW_FLOW_KEY_PUT(match, ip.frag,
1004 				ipv6_key->ipv6_frag, is_mask);
1005 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1006 				ipv6_key->ipv6_src,
1007 				sizeof(match->key->ipv6.addr.src),
1008 				is_mask);
1009 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1010 				ipv6_key->ipv6_dst,
1011 				sizeof(match->key->ipv6.addr.dst),
1012 				is_mask);
1013 
1014 		attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1015 	}
1016 
1017 	if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1018 		const struct ovs_key_arp *arp_key;
1019 
1020 		arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1021 		if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1022 			OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1023 				  arp_key->arp_op);
1024 			return -EINVAL;
1025 		}
1026 
1027 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1028 				arp_key->arp_sip, is_mask);
1029 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1030 			arp_key->arp_tip, is_mask);
1031 		SW_FLOW_KEY_PUT(match, ip.proto,
1032 				ntohs(arp_key->arp_op), is_mask);
1033 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1034 				arp_key->arp_sha, ETH_ALEN, is_mask);
1035 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1036 				arp_key->arp_tha, ETH_ALEN, is_mask);
1037 
1038 		attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1039 	}
1040 
1041 	if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1042 		const struct ovs_key_mpls *mpls_key;
1043 
1044 		mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1045 		SW_FLOW_KEY_PUT(match, mpls.top_lse,
1046 				mpls_key->mpls_lse, is_mask);
1047 
1048 		attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1049 	 }
1050 
1051 	if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1052 		const struct ovs_key_tcp *tcp_key;
1053 
1054 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1055 		SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1056 		SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1057 		attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1058 	}
1059 
1060 	if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1061 		SW_FLOW_KEY_PUT(match, tp.flags,
1062 				nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1063 				is_mask);
1064 		attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1065 	}
1066 
1067 	if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1068 		const struct ovs_key_udp *udp_key;
1069 
1070 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1071 		SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1072 		SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1073 		attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1074 	}
1075 
1076 	if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1077 		const struct ovs_key_sctp *sctp_key;
1078 
1079 		sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1080 		SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1081 		SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1082 		attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1083 	}
1084 
1085 	if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1086 		const struct ovs_key_icmp *icmp_key;
1087 
1088 		icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1089 		SW_FLOW_KEY_PUT(match, tp.src,
1090 				htons(icmp_key->icmp_type), is_mask);
1091 		SW_FLOW_KEY_PUT(match, tp.dst,
1092 				htons(icmp_key->icmp_code), is_mask);
1093 		attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1094 	}
1095 
1096 	if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1097 		const struct ovs_key_icmpv6 *icmpv6_key;
1098 
1099 		icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1100 		SW_FLOW_KEY_PUT(match, tp.src,
1101 				htons(icmpv6_key->icmpv6_type), is_mask);
1102 		SW_FLOW_KEY_PUT(match, tp.dst,
1103 				htons(icmpv6_key->icmpv6_code), is_mask);
1104 		attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1105 	}
1106 
1107 	if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1108 		const struct ovs_key_nd *nd_key;
1109 
1110 		nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1111 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1112 			nd_key->nd_target,
1113 			sizeof(match->key->ipv6.nd.target),
1114 			is_mask);
1115 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1116 			nd_key->nd_sll, ETH_ALEN, is_mask);
1117 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1118 				nd_key->nd_tll, ETH_ALEN, is_mask);
1119 		attrs &= ~(1 << OVS_KEY_ATTR_ND);
1120 	}
1121 
1122 	if (attrs != 0) {
1123 		OVS_NLERR(log, "Unknown key attributes %llx",
1124 			  (unsigned long long)attrs);
1125 		return -EINVAL;
1126 	}
1127 
1128 	return 0;
1129 }
1130 
1131 static void nlattr_set(struct nlattr *attr, u8 val,
1132 		       const struct ovs_len_tbl *tbl)
1133 {
1134 	struct nlattr *nla;
1135 	int rem;
1136 
1137 	/* The nlattr stream should already have been validated */
1138 	nla_for_each_nested(nla, attr, rem) {
1139 		if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
1140 			if (tbl[nla_type(nla)].next)
1141 				tbl = tbl[nla_type(nla)].next;
1142 			nlattr_set(nla, val, tbl);
1143 		} else {
1144 			memset(nla_data(nla), val, nla_len(nla));
1145 		}
1146 	}
1147 }
1148 
1149 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1150 {
1151 	nlattr_set(attr, val, ovs_key_lens);
1152 }
1153 
1154 /**
1155  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1156  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1157  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1158  * does not include any don't care bit.
1159  * @net: Used to determine per-namespace field support.
1160  * @match: receives the extracted flow match information.
1161  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1162  * sequence. The fields should of the packet that triggered the creation
1163  * of this flow.
1164  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1165  * attribute specifies the mask field of the wildcarded flow.
1166  * @log: Boolean to allow kernel error logging.  Normally true, but when
1167  * probing for feature compatibility this should be passed in as false to
1168  * suppress unnecessary error logging.
1169  */
1170 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1171 		      const struct nlattr *nla_key,
1172 		      const struct nlattr *nla_mask,
1173 		      bool log)
1174 {
1175 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1176 	const struct nlattr *encap;
1177 	struct nlattr *newmask = NULL;
1178 	u64 key_attrs = 0;
1179 	u64 mask_attrs = 0;
1180 	bool encap_valid = false;
1181 	int err;
1182 
1183 	err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1184 	if (err)
1185 		return err;
1186 
1187 	if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1188 	    (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1189 	    (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1190 		__be16 tci;
1191 
1192 		if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1193 		      (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1194 			OVS_NLERR(log, "Invalid Vlan frame.");
1195 			return -EINVAL;
1196 		}
1197 
1198 		key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1199 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1200 		encap = a[OVS_KEY_ATTR_ENCAP];
1201 		key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1202 		encap_valid = true;
1203 
1204 		if (tci & htons(VLAN_TAG_PRESENT)) {
1205 			err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1206 			if (err)
1207 				return err;
1208 		} else if (!tci) {
1209 			/* Corner case for truncated 802.1Q header. */
1210 			if (nla_len(encap)) {
1211 				OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
1212 				return -EINVAL;
1213 			}
1214 		} else {
1215 			OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
1216 			return  -EINVAL;
1217 		}
1218 	}
1219 
1220 	err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1221 	if (err)
1222 		return err;
1223 
1224 	if (match->mask) {
1225 		if (!nla_mask) {
1226 			/* Create an exact match mask. We need to set to 0xff
1227 			 * all the 'match->mask' fields that have been touched
1228 			 * in 'match->key'. We cannot simply memset
1229 			 * 'match->mask', because padding bytes and fields not
1230 			 * specified in 'match->key' should be left to 0.
1231 			 * Instead, we use a stream of netlink attributes,
1232 			 * copied from 'key' and set to 0xff.
1233 			 * ovs_key_from_nlattrs() will take care of filling
1234 			 * 'match->mask' appropriately.
1235 			 */
1236 			newmask = kmemdup(nla_key,
1237 					  nla_total_size(nla_len(nla_key)),
1238 					  GFP_KERNEL);
1239 			if (!newmask)
1240 				return -ENOMEM;
1241 
1242 			mask_set_nlattr(newmask, 0xff);
1243 
1244 			/* The userspace does not send tunnel attributes that
1245 			 * are 0, but we should not wildcard them nonetheless.
1246 			 */
1247 			if (match->key->tun_proto)
1248 				SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1249 							 0xff, true);
1250 
1251 			nla_mask = newmask;
1252 		}
1253 
1254 		err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1255 		if (err)
1256 			goto free_newmask;
1257 
1258 		/* Always match on tci. */
1259 		SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1260 
1261 		if (mask_attrs & 1 << OVS_KEY_ATTR_ENCAP) {
1262 			__be16 eth_type = 0;
1263 			__be16 tci = 0;
1264 
1265 			if (!encap_valid) {
1266 				OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1267 				err = -EINVAL;
1268 				goto free_newmask;
1269 			}
1270 
1271 			mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1272 			if (a[OVS_KEY_ATTR_ETHERTYPE])
1273 				eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1274 
1275 			if (eth_type == htons(0xffff)) {
1276 				mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1277 				encap = a[OVS_KEY_ATTR_ENCAP];
1278 				err = parse_flow_mask_nlattrs(encap, a,
1279 							      &mask_attrs, log);
1280 				if (err)
1281 					goto free_newmask;
1282 			} else {
1283 				OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1284 					  ntohs(eth_type));
1285 				err = -EINVAL;
1286 				goto free_newmask;
1287 			}
1288 
1289 			if (a[OVS_KEY_ATTR_VLAN])
1290 				tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1291 
1292 			if (!(tci & htons(VLAN_TAG_PRESENT))) {
1293 				OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1294 					  ntohs(tci));
1295 				err = -EINVAL;
1296 				goto free_newmask;
1297 			}
1298 		}
1299 
1300 		err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1301 					   log);
1302 		if (err)
1303 			goto free_newmask;
1304 	}
1305 
1306 	if (!match_validate(match, key_attrs, mask_attrs, log))
1307 		err = -EINVAL;
1308 
1309 free_newmask:
1310 	kfree(newmask);
1311 	return err;
1312 }
1313 
1314 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1315 {
1316 	size_t len;
1317 
1318 	if (!attr)
1319 		return 0;
1320 
1321 	len = nla_len(attr);
1322 	if (len < 1 || len > MAX_UFID_LENGTH) {
1323 		OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1324 			  nla_len(attr), MAX_UFID_LENGTH);
1325 		return 0;
1326 	}
1327 
1328 	return len;
1329 }
1330 
1331 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1332  * or false otherwise.
1333  */
1334 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1335 		      bool log)
1336 {
1337 	sfid->ufid_len = get_ufid_len(attr, log);
1338 	if (sfid->ufid_len)
1339 		memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1340 
1341 	return sfid->ufid_len;
1342 }
1343 
1344 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1345 			   const struct sw_flow_key *key, bool log)
1346 {
1347 	struct sw_flow_key *new_key;
1348 
1349 	if (ovs_nla_get_ufid(sfid, ufid, log))
1350 		return 0;
1351 
1352 	/* If UFID was not provided, use unmasked key. */
1353 	new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1354 	if (!new_key)
1355 		return -ENOMEM;
1356 	memcpy(new_key, key, sizeof(*key));
1357 	sfid->unmasked_key = new_key;
1358 
1359 	return 0;
1360 }
1361 
1362 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1363 {
1364 	return attr ? nla_get_u32(attr) : 0;
1365 }
1366 
1367 /**
1368  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1369  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1370  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1371  * sequence.
1372  * @log: Boolean to allow kernel error logging.  Normally true, but when
1373  * probing for feature compatibility this should be passed in as false to
1374  * suppress unnecessary error logging.
1375  *
1376  * This parses a series of Netlink attributes that form a flow key, which must
1377  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1378  * get the metadata, that is, the parts of the flow key that cannot be
1379  * extracted from the packet itself.
1380  */
1381 
1382 int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
1383 			      struct sw_flow_key *key,
1384 			      bool log)
1385 {
1386 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1387 	struct sw_flow_match match;
1388 	u64 attrs = 0;
1389 	int err;
1390 
1391 	err = parse_flow_nlattrs(attr, a, &attrs, log);
1392 	if (err)
1393 		return -EINVAL;
1394 
1395 	memset(&match, 0, sizeof(match));
1396 	match.key = key;
1397 
1398 	memset(&key->ct, 0, sizeof(key->ct));
1399 	key->phy.in_port = DP_MAX_PORTS;
1400 
1401 	return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1402 }
1403 
1404 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1405 			     const struct sw_flow_key *output, bool is_mask,
1406 			     struct sk_buff *skb)
1407 {
1408 	struct ovs_key_ethernet *eth_key;
1409 	struct nlattr *nla, *encap;
1410 
1411 	if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1412 		goto nla_put_failure;
1413 
1414 	if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1415 		goto nla_put_failure;
1416 
1417 	if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1418 		goto nla_put_failure;
1419 
1420 	if ((swkey->tun_proto || is_mask)) {
1421 		const void *opts = NULL;
1422 
1423 		if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1424 			opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1425 
1426 		if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
1427 				     swkey->tun_opts_len, swkey->tun_proto))
1428 			goto nla_put_failure;
1429 	}
1430 
1431 	if (swkey->phy.in_port == DP_MAX_PORTS) {
1432 		if (is_mask && (output->phy.in_port == 0xffff))
1433 			if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1434 				goto nla_put_failure;
1435 	} else {
1436 		u16 upper_u16;
1437 		upper_u16 = !is_mask ? 0 : 0xffff;
1438 
1439 		if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1440 				(upper_u16 << 16) | output->phy.in_port))
1441 			goto nla_put_failure;
1442 	}
1443 
1444 	if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1445 		goto nla_put_failure;
1446 
1447 	if (ovs_ct_put_key(output, skb))
1448 		goto nla_put_failure;
1449 
1450 	nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1451 	if (!nla)
1452 		goto nla_put_failure;
1453 
1454 	eth_key = nla_data(nla);
1455 	ether_addr_copy(eth_key->eth_src, output->eth.src);
1456 	ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1457 
1458 	if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1459 		__be16 eth_type;
1460 		eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1461 		if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1462 		    nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1463 			goto nla_put_failure;
1464 		encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1465 		if (!swkey->eth.tci)
1466 			goto unencap;
1467 	} else
1468 		encap = NULL;
1469 
1470 	if (swkey->eth.type == htons(ETH_P_802_2)) {
1471 		/*
1472 		 * Ethertype 802.2 is represented in the netlink with omitted
1473 		 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1474 		 * 0xffff in the mask attribute.  Ethertype can also
1475 		 * be wildcarded.
1476 		 */
1477 		if (is_mask && output->eth.type)
1478 			if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1479 						output->eth.type))
1480 				goto nla_put_failure;
1481 		goto unencap;
1482 	}
1483 
1484 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1485 		goto nla_put_failure;
1486 
1487 	if (swkey->eth.type == htons(ETH_P_IP)) {
1488 		struct ovs_key_ipv4 *ipv4_key;
1489 
1490 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1491 		if (!nla)
1492 			goto nla_put_failure;
1493 		ipv4_key = nla_data(nla);
1494 		ipv4_key->ipv4_src = output->ipv4.addr.src;
1495 		ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1496 		ipv4_key->ipv4_proto = output->ip.proto;
1497 		ipv4_key->ipv4_tos = output->ip.tos;
1498 		ipv4_key->ipv4_ttl = output->ip.ttl;
1499 		ipv4_key->ipv4_frag = output->ip.frag;
1500 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1501 		struct ovs_key_ipv6 *ipv6_key;
1502 
1503 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1504 		if (!nla)
1505 			goto nla_put_failure;
1506 		ipv6_key = nla_data(nla);
1507 		memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1508 				sizeof(ipv6_key->ipv6_src));
1509 		memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1510 				sizeof(ipv6_key->ipv6_dst));
1511 		ipv6_key->ipv6_label = output->ipv6.label;
1512 		ipv6_key->ipv6_proto = output->ip.proto;
1513 		ipv6_key->ipv6_tclass = output->ip.tos;
1514 		ipv6_key->ipv6_hlimit = output->ip.ttl;
1515 		ipv6_key->ipv6_frag = output->ip.frag;
1516 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
1517 		   swkey->eth.type == htons(ETH_P_RARP)) {
1518 		struct ovs_key_arp *arp_key;
1519 
1520 		nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1521 		if (!nla)
1522 			goto nla_put_failure;
1523 		arp_key = nla_data(nla);
1524 		memset(arp_key, 0, sizeof(struct ovs_key_arp));
1525 		arp_key->arp_sip = output->ipv4.addr.src;
1526 		arp_key->arp_tip = output->ipv4.addr.dst;
1527 		arp_key->arp_op = htons(output->ip.proto);
1528 		ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1529 		ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1530 	} else if (eth_p_mpls(swkey->eth.type)) {
1531 		struct ovs_key_mpls *mpls_key;
1532 
1533 		nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1534 		if (!nla)
1535 			goto nla_put_failure;
1536 		mpls_key = nla_data(nla);
1537 		mpls_key->mpls_lse = output->mpls.top_lse;
1538 	}
1539 
1540 	if ((swkey->eth.type == htons(ETH_P_IP) ||
1541 	     swkey->eth.type == htons(ETH_P_IPV6)) &&
1542 	     swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1543 
1544 		if (swkey->ip.proto == IPPROTO_TCP) {
1545 			struct ovs_key_tcp *tcp_key;
1546 
1547 			nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1548 			if (!nla)
1549 				goto nla_put_failure;
1550 			tcp_key = nla_data(nla);
1551 			tcp_key->tcp_src = output->tp.src;
1552 			tcp_key->tcp_dst = output->tp.dst;
1553 			if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1554 					 output->tp.flags))
1555 				goto nla_put_failure;
1556 		} else if (swkey->ip.proto == IPPROTO_UDP) {
1557 			struct ovs_key_udp *udp_key;
1558 
1559 			nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1560 			if (!nla)
1561 				goto nla_put_failure;
1562 			udp_key = nla_data(nla);
1563 			udp_key->udp_src = output->tp.src;
1564 			udp_key->udp_dst = output->tp.dst;
1565 		} else if (swkey->ip.proto == IPPROTO_SCTP) {
1566 			struct ovs_key_sctp *sctp_key;
1567 
1568 			nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1569 			if (!nla)
1570 				goto nla_put_failure;
1571 			sctp_key = nla_data(nla);
1572 			sctp_key->sctp_src = output->tp.src;
1573 			sctp_key->sctp_dst = output->tp.dst;
1574 		} else if (swkey->eth.type == htons(ETH_P_IP) &&
1575 			   swkey->ip.proto == IPPROTO_ICMP) {
1576 			struct ovs_key_icmp *icmp_key;
1577 
1578 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1579 			if (!nla)
1580 				goto nla_put_failure;
1581 			icmp_key = nla_data(nla);
1582 			icmp_key->icmp_type = ntohs(output->tp.src);
1583 			icmp_key->icmp_code = ntohs(output->tp.dst);
1584 		} else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1585 			   swkey->ip.proto == IPPROTO_ICMPV6) {
1586 			struct ovs_key_icmpv6 *icmpv6_key;
1587 
1588 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1589 						sizeof(*icmpv6_key));
1590 			if (!nla)
1591 				goto nla_put_failure;
1592 			icmpv6_key = nla_data(nla);
1593 			icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1594 			icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1595 
1596 			if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1597 			    icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1598 				struct ovs_key_nd *nd_key;
1599 
1600 				nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1601 				if (!nla)
1602 					goto nla_put_failure;
1603 				nd_key = nla_data(nla);
1604 				memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1605 							sizeof(nd_key->nd_target));
1606 				ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1607 				ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1608 			}
1609 		}
1610 	}
1611 
1612 unencap:
1613 	if (encap)
1614 		nla_nest_end(skb, encap);
1615 
1616 	return 0;
1617 
1618 nla_put_failure:
1619 	return -EMSGSIZE;
1620 }
1621 
1622 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1623 		    const struct sw_flow_key *output, int attr, bool is_mask,
1624 		    struct sk_buff *skb)
1625 {
1626 	int err;
1627 	struct nlattr *nla;
1628 
1629 	nla = nla_nest_start(skb, attr);
1630 	if (!nla)
1631 		return -EMSGSIZE;
1632 	err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1633 	if (err)
1634 		return err;
1635 	nla_nest_end(skb, nla);
1636 
1637 	return 0;
1638 }
1639 
1640 /* Called with ovs_mutex or RCU read lock. */
1641 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1642 {
1643 	if (ovs_identifier_is_ufid(&flow->id))
1644 		return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1645 			       flow->id.ufid);
1646 
1647 	return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1648 			       OVS_FLOW_ATTR_KEY, false, skb);
1649 }
1650 
1651 /* Called with ovs_mutex or RCU read lock. */
1652 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1653 {
1654 	return ovs_nla_put_key(&flow->key, &flow->key,
1655 				OVS_FLOW_ATTR_KEY, false, skb);
1656 }
1657 
1658 /* Called with ovs_mutex or RCU read lock. */
1659 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1660 {
1661 	return ovs_nla_put_key(&flow->key, &flow->mask->key,
1662 				OVS_FLOW_ATTR_MASK, true, skb);
1663 }
1664 
1665 #define MAX_ACTIONS_BUFSIZE	(32 * 1024)
1666 
1667 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1668 {
1669 	struct sw_flow_actions *sfa;
1670 
1671 	if (size > MAX_ACTIONS_BUFSIZE) {
1672 		OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1673 		return ERR_PTR(-EINVAL);
1674 	}
1675 
1676 	sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1677 	if (!sfa)
1678 		return ERR_PTR(-ENOMEM);
1679 
1680 	sfa->actions_len = 0;
1681 	return sfa;
1682 }
1683 
1684 static void ovs_nla_free_set_action(const struct nlattr *a)
1685 {
1686 	const struct nlattr *ovs_key = nla_data(a);
1687 	struct ovs_tunnel_info *ovs_tun;
1688 
1689 	switch (nla_type(ovs_key)) {
1690 	case OVS_KEY_ATTR_TUNNEL_INFO:
1691 		ovs_tun = nla_data(ovs_key);
1692 		dst_release((struct dst_entry *)ovs_tun->tun_dst);
1693 		break;
1694 	}
1695 }
1696 
1697 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1698 {
1699 	const struct nlattr *a;
1700 	int rem;
1701 
1702 	if (!sf_acts)
1703 		return;
1704 
1705 	nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1706 		switch (nla_type(a)) {
1707 		case OVS_ACTION_ATTR_SET:
1708 			ovs_nla_free_set_action(a);
1709 			break;
1710 		case OVS_ACTION_ATTR_CT:
1711 			ovs_ct_free_action(a);
1712 			break;
1713 		}
1714 	}
1715 
1716 	kfree(sf_acts);
1717 }
1718 
1719 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1720 {
1721 	ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1722 }
1723 
1724 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1725  * The caller must hold rcu_read_lock for this to be sensible. */
1726 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
1727 {
1728 	call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
1729 }
1730 
1731 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1732 				       int attr_len, bool log)
1733 {
1734 
1735 	struct sw_flow_actions *acts;
1736 	int new_acts_size;
1737 	int req_size = NLA_ALIGN(attr_len);
1738 	int next_offset = offsetof(struct sw_flow_actions, actions) +
1739 					(*sfa)->actions_len;
1740 
1741 	if (req_size <= (ksize(*sfa) - next_offset))
1742 		goto out;
1743 
1744 	new_acts_size = ksize(*sfa) * 2;
1745 
1746 	if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1747 		if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1748 			return ERR_PTR(-EMSGSIZE);
1749 		new_acts_size = MAX_ACTIONS_BUFSIZE;
1750 	}
1751 
1752 	acts = nla_alloc_flow_actions(new_acts_size, log);
1753 	if (IS_ERR(acts))
1754 		return (void *)acts;
1755 
1756 	memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1757 	acts->actions_len = (*sfa)->actions_len;
1758 	acts->orig_len = (*sfa)->orig_len;
1759 	kfree(*sfa);
1760 	*sfa = acts;
1761 
1762 out:
1763 	(*sfa)->actions_len += req_size;
1764 	return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1765 }
1766 
1767 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1768 				   int attrtype, void *data, int len, bool log)
1769 {
1770 	struct nlattr *a;
1771 
1772 	a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1773 	if (IS_ERR(a))
1774 		return a;
1775 
1776 	a->nla_type = attrtype;
1777 	a->nla_len = nla_attr_size(len);
1778 
1779 	if (data)
1780 		memcpy(nla_data(a), data, len);
1781 	memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1782 
1783 	return a;
1784 }
1785 
1786 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1787 		       int len, bool log)
1788 {
1789 	struct nlattr *a;
1790 
1791 	a = __add_action(sfa, attrtype, data, len, log);
1792 
1793 	return PTR_ERR_OR_ZERO(a);
1794 }
1795 
1796 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1797 					  int attrtype, bool log)
1798 {
1799 	int used = (*sfa)->actions_len;
1800 	int err;
1801 
1802 	err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
1803 	if (err)
1804 		return err;
1805 
1806 	return used;
1807 }
1808 
1809 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1810 					 int st_offset)
1811 {
1812 	struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1813 							       st_offset);
1814 
1815 	a->nla_len = sfa->actions_len - st_offset;
1816 }
1817 
1818 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
1819 				  const struct sw_flow_key *key,
1820 				  int depth, struct sw_flow_actions **sfa,
1821 				  __be16 eth_type, __be16 vlan_tci, bool log);
1822 
1823 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
1824 				    const struct sw_flow_key *key, int depth,
1825 				    struct sw_flow_actions **sfa,
1826 				    __be16 eth_type, __be16 vlan_tci, bool log)
1827 {
1828 	const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1829 	const struct nlattr *probability, *actions;
1830 	const struct nlattr *a;
1831 	int rem, start, err, st_acts;
1832 
1833 	memset(attrs, 0, sizeof(attrs));
1834 	nla_for_each_nested(a, attr, rem) {
1835 		int type = nla_type(a);
1836 		if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1837 			return -EINVAL;
1838 		attrs[type] = a;
1839 	}
1840 	if (rem)
1841 		return -EINVAL;
1842 
1843 	probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1844 	if (!probability || nla_len(probability) != sizeof(u32))
1845 		return -EINVAL;
1846 
1847 	actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1848 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1849 		return -EINVAL;
1850 
1851 	/* validation done, copy sample action. */
1852 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1853 	if (start < 0)
1854 		return start;
1855 	err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1856 				 nla_data(probability), sizeof(u32), log);
1857 	if (err)
1858 		return err;
1859 	st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1860 	if (st_acts < 0)
1861 		return st_acts;
1862 
1863 	err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
1864 				     eth_type, vlan_tci, log);
1865 	if (err)
1866 		return err;
1867 
1868 	add_nested_action_end(*sfa, st_acts);
1869 	add_nested_action_end(*sfa, start);
1870 
1871 	return 0;
1872 }
1873 
1874 void ovs_match_init(struct sw_flow_match *match,
1875 		    struct sw_flow_key *key,
1876 		    struct sw_flow_mask *mask)
1877 {
1878 	memset(match, 0, sizeof(*match));
1879 	match->key = key;
1880 	match->mask = mask;
1881 
1882 	memset(key, 0, sizeof(*key));
1883 
1884 	if (mask) {
1885 		memset(&mask->key, 0, sizeof(mask->key));
1886 		mask->range.start = mask->range.end = 0;
1887 	}
1888 }
1889 
1890 static int validate_geneve_opts(struct sw_flow_key *key)
1891 {
1892 	struct geneve_opt *option;
1893 	int opts_len = key->tun_opts_len;
1894 	bool crit_opt = false;
1895 
1896 	option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1897 	while (opts_len > 0) {
1898 		int len;
1899 
1900 		if (opts_len < sizeof(*option))
1901 			return -EINVAL;
1902 
1903 		len = sizeof(*option) + option->length * 4;
1904 		if (len > opts_len)
1905 			return -EINVAL;
1906 
1907 		crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1908 
1909 		option = (struct geneve_opt *)((u8 *)option + len);
1910 		opts_len -= len;
1911 	};
1912 
1913 	key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1914 
1915 	return 0;
1916 }
1917 
1918 static int validate_and_copy_set_tun(const struct nlattr *attr,
1919 				     struct sw_flow_actions **sfa, bool log)
1920 {
1921 	struct sw_flow_match match;
1922 	struct sw_flow_key key;
1923 	struct metadata_dst *tun_dst;
1924 	struct ip_tunnel_info *tun_info;
1925 	struct ovs_tunnel_info *ovs_tun;
1926 	struct nlattr *a;
1927 	int err = 0, start, opts_type;
1928 
1929 	ovs_match_init(&match, &key, NULL);
1930 	opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
1931 	if (opts_type < 0)
1932 		return opts_type;
1933 
1934 	if (key.tun_opts_len) {
1935 		switch (opts_type) {
1936 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1937 			err = validate_geneve_opts(&key);
1938 			if (err < 0)
1939 				return err;
1940 			break;
1941 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1942 			break;
1943 		}
1944 	};
1945 
1946 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1947 	if (start < 0)
1948 		return start;
1949 
1950 	tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1951 	if (!tun_dst)
1952 		return -ENOMEM;
1953 
1954 	a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1955 			 sizeof(*ovs_tun), log);
1956 	if (IS_ERR(a)) {
1957 		dst_release((struct dst_entry *)tun_dst);
1958 		return PTR_ERR(a);
1959 	}
1960 
1961 	ovs_tun = nla_data(a);
1962 	ovs_tun->tun_dst = tun_dst;
1963 
1964 	tun_info = &tun_dst->u.tun_info;
1965 	tun_info->mode = IP_TUNNEL_INFO_TX;
1966 	if (key.tun_proto == AF_INET6)
1967 		tun_info->mode |= IP_TUNNEL_INFO_IPV6;
1968 	tun_info->key = key.tun_key;
1969 
1970 	/* We need to store the options in the action itself since
1971 	 * everything else will go away after flow setup. We can append
1972 	 * it to tun_info and then point there.
1973 	 */
1974 	ip_tunnel_info_opts_set(tun_info,
1975 				TUN_METADATA_OPTS(&key, key.tun_opts_len),
1976 				key.tun_opts_len);
1977 	add_nested_action_end(*sfa, start);
1978 
1979 	return err;
1980 }
1981 
1982 /* Return false if there are any non-masked bits set.
1983  * Mask follows data immediately, before any netlink padding.
1984  */
1985 static bool validate_masked(u8 *data, int len)
1986 {
1987 	u8 *mask = data + len;
1988 
1989 	while (len--)
1990 		if (*data++ & ~*mask++)
1991 			return false;
1992 
1993 	return true;
1994 }
1995 
1996 static int validate_set(const struct nlattr *a,
1997 			const struct sw_flow_key *flow_key,
1998 			struct sw_flow_actions **sfa,
1999 			bool *skip_copy, __be16 eth_type, bool masked, bool log)
2000 {
2001 	const struct nlattr *ovs_key = nla_data(a);
2002 	int key_type = nla_type(ovs_key);
2003 	size_t key_len;
2004 
2005 	/* There can be only one key in a action */
2006 	if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2007 		return -EINVAL;
2008 
2009 	key_len = nla_len(ovs_key);
2010 	if (masked)
2011 		key_len /= 2;
2012 
2013 	if (key_type > OVS_KEY_ATTR_MAX ||
2014 	    !check_attr_len(key_len, ovs_key_lens[key_type].len))
2015 		return -EINVAL;
2016 
2017 	if (masked && !validate_masked(nla_data(ovs_key), key_len))
2018 		return -EINVAL;
2019 
2020 	switch (key_type) {
2021 	const struct ovs_key_ipv4 *ipv4_key;
2022 	const struct ovs_key_ipv6 *ipv6_key;
2023 	int err;
2024 
2025 	case OVS_KEY_ATTR_PRIORITY:
2026 	case OVS_KEY_ATTR_SKB_MARK:
2027 	case OVS_KEY_ATTR_CT_MARK:
2028 	case OVS_KEY_ATTR_CT_LABEL:
2029 	case OVS_KEY_ATTR_ETHERNET:
2030 		break;
2031 
2032 	case OVS_KEY_ATTR_TUNNEL:
2033 		if (eth_p_mpls(eth_type))
2034 			return -EINVAL;
2035 
2036 		if (masked)
2037 			return -EINVAL; /* Masked tunnel set not supported. */
2038 
2039 		*skip_copy = true;
2040 		err = validate_and_copy_set_tun(a, sfa, log);
2041 		if (err)
2042 			return err;
2043 		break;
2044 
2045 	case OVS_KEY_ATTR_IPV4:
2046 		if (eth_type != htons(ETH_P_IP))
2047 			return -EINVAL;
2048 
2049 		ipv4_key = nla_data(ovs_key);
2050 
2051 		if (masked) {
2052 			const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2053 
2054 			/* Non-writeable fields. */
2055 			if (mask->ipv4_proto || mask->ipv4_frag)
2056 				return -EINVAL;
2057 		} else {
2058 			if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2059 				return -EINVAL;
2060 
2061 			if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2062 				return -EINVAL;
2063 		}
2064 		break;
2065 
2066 	case OVS_KEY_ATTR_IPV6:
2067 		if (eth_type != htons(ETH_P_IPV6))
2068 			return -EINVAL;
2069 
2070 		ipv6_key = nla_data(ovs_key);
2071 
2072 		if (masked) {
2073 			const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2074 
2075 			/* Non-writeable fields. */
2076 			if (mask->ipv6_proto || mask->ipv6_frag)
2077 				return -EINVAL;
2078 
2079 			/* Invalid bits in the flow label mask? */
2080 			if (ntohl(mask->ipv6_label) & 0xFFF00000)
2081 				return -EINVAL;
2082 		} else {
2083 			if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2084 				return -EINVAL;
2085 
2086 			if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2087 				return -EINVAL;
2088 		}
2089 		if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2090 			return -EINVAL;
2091 
2092 		break;
2093 
2094 	case OVS_KEY_ATTR_TCP:
2095 		if ((eth_type != htons(ETH_P_IP) &&
2096 		     eth_type != htons(ETH_P_IPV6)) ||
2097 		    flow_key->ip.proto != IPPROTO_TCP)
2098 			return -EINVAL;
2099 
2100 		break;
2101 
2102 	case OVS_KEY_ATTR_UDP:
2103 		if ((eth_type != htons(ETH_P_IP) &&
2104 		     eth_type != htons(ETH_P_IPV6)) ||
2105 		    flow_key->ip.proto != IPPROTO_UDP)
2106 			return -EINVAL;
2107 
2108 		break;
2109 
2110 	case OVS_KEY_ATTR_MPLS:
2111 		if (!eth_p_mpls(eth_type))
2112 			return -EINVAL;
2113 		break;
2114 
2115 	case OVS_KEY_ATTR_SCTP:
2116 		if ((eth_type != htons(ETH_P_IP) &&
2117 		     eth_type != htons(ETH_P_IPV6)) ||
2118 		    flow_key->ip.proto != IPPROTO_SCTP)
2119 			return -EINVAL;
2120 
2121 		break;
2122 
2123 	default:
2124 		return -EINVAL;
2125 	}
2126 
2127 	/* Convert non-masked non-tunnel set actions to masked set actions. */
2128 	if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2129 		int start, len = key_len * 2;
2130 		struct nlattr *at;
2131 
2132 		*skip_copy = true;
2133 
2134 		start = add_nested_action_start(sfa,
2135 						OVS_ACTION_ATTR_SET_TO_MASKED,
2136 						log);
2137 		if (start < 0)
2138 			return start;
2139 
2140 		at = __add_action(sfa, key_type, NULL, len, log);
2141 		if (IS_ERR(at))
2142 			return PTR_ERR(at);
2143 
2144 		memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2145 		memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2146 		/* Clear non-writeable bits from otherwise writeable fields. */
2147 		if (key_type == OVS_KEY_ATTR_IPV6) {
2148 			struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2149 
2150 			mask->ipv6_label &= htonl(0x000FFFFF);
2151 		}
2152 		add_nested_action_end(*sfa, start);
2153 	}
2154 
2155 	return 0;
2156 }
2157 
2158 static int validate_userspace(const struct nlattr *attr)
2159 {
2160 	static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2161 		[OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2162 		[OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2163 		[OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2164 	};
2165 	struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2166 	int error;
2167 
2168 	error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2169 				 attr, userspace_policy);
2170 	if (error)
2171 		return error;
2172 
2173 	if (!a[OVS_USERSPACE_ATTR_PID] ||
2174 	    !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2175 		return -EINVAL;
2176 
2177 	return 0;
2178 }
2179 
2180 static int copy_action(const struct nlattr *from,
2181 		       struct sw_flow_actions **sfa, bool log)
2182 {
2183 	int totlen = NLA_ALIGN(from->nla_len);
2184 	struct nlattr *to;
2185 
2186 	to = reserve_sfa_size(sfa, from->nla_len, log);
2187 	if (IS_ERR(to))
2188 		return PTR_ERR(to);
2189 
2190 	memcpy(to, from, totlen);
2191 	return 0;
2192 }
2193 
2194 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2195 				  const struct sw_flow_key *key,
2196 				  int depth, struct sw_flow_actions **sfa,
2197 				  __be16 eth_type, __be16 vlan_tci, bool log)
2198 {
2199 	const struct nlattr *a;
2200 	int rem, err;
2201 
2202 	if (depth >= SAMPLE_ACTION_DEPTH)
2203 		return -EOVERFLOW;
2204 
2205 	nla_for_each_nested(a, attr, rem) {
2206 		/* Expected argument lengths, (u32)-1 for variable length. */
2207 		static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2208 			[OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2209 			[OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2210 			[OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2211 			[OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2212 			[OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2213 			[OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2214 			[OVS_ACTION_ATTR_POP_VLAN] = 0,
2215 			[OVS_ACTION_ATTR_SET] = (u32)-1,
2216 			[OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2217 			[OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2218 			[OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2219 			[OVS_ACTION_ATTR_CT] = (u32)-1,
2220 		};
2221 		const struct ovs_action_push_vlan *vlan;
2222 		int type = nla_type(a);
2223 		bool skip_copy;
2224 
2225 		if (type > OVS_ACTION_ATTR_MAX ||
2226 		    (action_lens[type] != nla_len(a) &&
2227 		     action_lens[type] != (u32)-1))
2228 			return -EINVAL;
2229 
2230 		skip_copy = false;
2231 		switch (type) {
2232 		case OVS_ACTION_ATTR_UNSPEC:
2233 			return -EINVAL;
2234 
2235 		case OVS_ACTION_ATTR_USERSPACE:
2236 			err = validate_userspace(a);
2237 			if (err)
2238 				return err;
2239 			break;
2240 
2241 		case OVS_ACTION_ATTR_OUTPUT:
2242 			if (nla_get_u32(a) >= DP_MAX_PORTS)
2243 				return -EINVAL;
2244 			break;
2245 
2246 		case OVS_ACTION_ATTR_HASH: {
2247 			const struct ovs_action_hash *act_hash = nla_data(a);
2248 
2249 			switch (act_hash->hash_alg) {
2250 			case OVS_HASH_ALG_L4:
2251 				break;
2252 			default:
2253 				return  -EINVAL;
2254 			}
2255 
2256 			break;
2257 		}
2258 
2259 		case OVS_ACTION_ATTR_POP_VLAN:
2260 			vlan_tci = htons(0);
2261 			break;
2262 
2263 		case OVS_ACTION_ATTR_PUSH_VLAN:
2264 			vlan = nla_data(a);
2265 			if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2266 				return -EINVAL;
2267 			if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2268 				return -EINVAL;
2269 			vlan_tci = vlan->vlan_tci;
2270 			break;
2271 
2272 		case OVS_ACTION_ATTR_RECIRC:
2273 			break;
2274 
2275 		case OVS_ACTION_ATTR_PUSH_MPLS: {
2276 			const struct ovs_action_push_mpls *mpls = nla_data(a);
2277 
2278 			if (!eth_p_mpls(mpls->mpls_ethertype))
2279 				return -EINVAL;
2280 			/* Prohibit push MPLS other than to a white list
2281 			 * for packets that have a known tag order.
2282 			 */
2283 			if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2284 			    (eth_type != htons(ETH_P_IP) &&
2285 			     eth_type != htons(ETH_P_IPV6) &&
2286 			     eth_type != htons(ETH_P_ARP) &&
2287 			     eth_type != htons(ETH_P_RARP) &&
2288 			     !eth_p_mpls(eth_type)))
2289 				return -EINVAL;
2290 			eth_type = mpls->mpls_ethertype;
2291 			break;
2292 		}
2293 
2294 		case OVS_ACTION_ATTR_POP_MPLS:
2295 			if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2296 			    !eth_p_mpls(eth_type))
2297 				return -EINVAL;
2298 
2299 			/* Disallow subsequent L2.5+ set and mpls_pop actions
2300 			 * as there is no check here to ensure that the new
2301 			 * eth_type is valid and thus set actions could
2302 			 * write off the end of the packet or otherwise
2303 			 * corrupt it.
2304 			 *
2305 			 * Support for these actions is planned using packet
2306 			 * recirculation.
2307 			 */
2308 			eth_type = htons(0);
2309 			break;
2310 
2311 		case OVS_ACTION_ATTR_SET:
2312 			err = validate_set(a, key, sfa,
2313 					   &skip_copy, eth_type, false, log);
2314 			if (err)
2315 				return err;
2316 			break;
2317 
2318 		case OVS_ACTION_ATTR_SET_MASKED:
2319 			err = validate_set(a, key, sfa,
2320 					   &skip_copy, eth_type, true, log);
2321 			if (err)
2322 				return err;
2323 			break;
2324 
2325 		case OVS_ACTION_ATTR_SAMPLE:
2326 			err = validate_and_copy_sample(net, a, key, depth, sfa,
2327 						       eth_type, vlan_tci, log);
2328 			if (err)
2329 				return err;
2330 			skip_copy = true;
2331 			break;
2332 
2333 		case OVS_ACTION_ATTR_CT:
2334 			err = ovs_ct_copy_action(net, a, key, sfa, log);
2335 			if (err)
2336 				return err;
2337 			skip_copy = true;
2338 			break;
2339 
2340 		default:
2341 			OVS_NLERR(log, "Unknown Action type %d", type);
2342 			return -EINVAL;
2343 		}
2344 		if (!skip_copy) {
2345 			err = copy_action(a, sfa, log);
2346 			if (err)
2347 				return err;
2348 		}
2349 	}
2350 
2351 	if (rem > 0)
2352 		return -EINVAL;
2353 
2354 	return 0;
2355 }
2356 
2357 /* 'key' must be the masked key. */
2358 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2359 			 const struct sw_flow_key *key,
2360 			 struct sw_flow_actions **sfa, bool log)
2361 {
2362 	int err;
2363 
2364 	*sfa = nla_alloc_flow_actions(nla_len(attr), log);
2365 	if (IS_ERR(*sfa))
2366 		return PTR_ERR(*sfa);
2367 
2368 	(*sfa)->orig_len = nla_len(attr);
2369 	err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
2370 				     key->eth.tci, log);
2371 	if (err)
2372 		ovs_nla_free_flow_actions(*sfa);
2373 
2374 	return err;
2375 }
2376 
2377 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2378 {
2379 	const struct nlattr *a;
2380 	struct nlattr *start;
2381 	int err = 0, rem;
2382 
2383 	start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2384 	if (!start)
2385 		return -EMSGSIZE;
2386 
2387 	nla_for_each_nested(a, attr, rem) {
2388 		int type = nla_type(a);
2389 		struct nlattr *st_sample;
2390 
2391 		switch (type) {
2392 		case OVS_SAMPLE_ATTR_PROBABILITY:
2393 			if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2394 				    sizeof(u32), nla_data(a)))
2395 				return -EMSGSIZE;
2396 			break;
2397 		case OVS_SAMPLE_ATTR_ACTIONS:
2398 			st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2399 			if (!st_sample)
2400 				return -EMSGSIZE;
2401 			err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2402 			if (err)
2403 				return err;
2404 			nla_nest_end(skb, st_sample);
2405 			break;
2406 		}
2407 	}
2408 
2409 	nla_nest_end(skb, start);
2410 	return err;
2411 }
2412 
2413 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2414 {
2415 	const struct nlattr *ovs_key = nla_data(a);
2416 	int key_type = nla_type(ovs_key);
2417 	struct nlattr *start;
2418 	int err;
2419 
2420 	switch (key_type) {
2421 	case OVS_KEY_ATTR_TUNNEL_INFO: {
2422 		struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2423 		struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2424 
2425 		start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2426 		if (!start)
2427 			return -EMSGSIZE;
2428 
2429 		err = ip_tun_to_nlattr(skb, &tun_info->key,
2430 				       tun_info->options_len ?
2431 					     ip_tunnel_info_opts(tun_info) : NULL,
2432 				       tun_info->options_len,
2433 				       ip_tunnel_info_af(tun_info));
2434 		if (err)
2435 			return err;
2436 		nla_nest_end(skb, start);
2437 		break;
2438 	}
2439 	default:
2440 		if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2441 			return -EMSGSIZE;
2442 		break;
2443 	}
2444 
2445 	return 0;
2446 }
2447 
2448 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2449 						struct sk_buff *skb)
2450 {
2451 	const struct nlattr *ovs_key = nla_data(a);
2452 	struct nlattr *nla;
2453 	size_t key_len = nla_len(ovs_key) / 2;
2454 
2455 	/* Revert the conversion we did from a non-masked set action to
2456 	 * masked set action.
2457 	 */
2458 	nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2459 	if (!nla)
2460 		return -EMSGSIZE;
2461 
2462 	if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2463 		return -EMSGSIZE;
2464 
2465 	nla_nest_end(skb, nla);
2466 	return 0;
2467 }
2468 
2469 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2470 {
2471 	const struct nlattr *a;
2472 	int rem, err;
2473 
2474 	nla_for_each_attr(a, attr, len, rem) {
2475 		int type = nla_type(a);
2476 
2477 		switch (type) {
2478 		case OVS_ACTION_ATTR_SET:
2479 			err = set_action_to_attr(a, skb);
2480 			if (err)
2481 				return err;
2482 			break;
2483 
2484 		case OVS_ACTION_ATTR_SET_TO_MASKED:
2485 			err = masked_set_action_to_set_action_attr(a, skb);
2486 			if (err)
2487 				return err;
2488 			break;
2489 
2490 		case OVS_ACTION_ATTR_SAMPLE:
2491 			err = sample_action_to_attr(a, skb);
2492 			if (err)
2493 				return err;
2494 			break;
2495 
2496 		case OVS_ACTION_ATTR_CT:
2497 			err = ovs_ct_action_to_attr(nla_data(a), skb);
2498 			if (err)
2499 				return err;
2500 			break;
2501 
2502 		default:
2503 			if (nla_put(skb, type, nla_len(a), nla_data(a)))
2504 				return -EMSGSIZE;
2505 			break;
2506 		}
2507 	}
2508 
2509 	return 0;
2510 }
2511