xref: /linux/net/openvswitch/flow_netlink.c (revision 32786fdc9506aeba98278c1844d4bfb766863832)
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 = 0;
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_64bit(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(4)   /* 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_LABELS */
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(u32) },
355 	[OVS_KEY_ATTR_CT_ZONE]	 = { .len = sizeof(u16) },
356 	[OVS_KEY_ATTR_CT_MARK]	 = { .len = sizeof(u32) },
357 	[OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
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 	bool ttl = false, ipv4 = false, ipv6 = false;
552 	__be16 tun_flags = 0;
553 	int opts_type = 0;
554 	struct nlattr *a;
555 	int rem;
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 			 OVS_TUNNEL_KEY_ATTR_PAD))
725 		return -EMSGSIZE;
726 	switch (tun_proto) {
727 	case AF_INET:
728 		if (output->u.ipv4.src &&
729 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
730 				    output->u.ipv4.src))
731 			return -EMSGSIZE;
732 		if (output->u.ipv4.dst &&
733 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
734 				    output->u.ipv4.dst))
735 			return -EMSGSIZE;
736 		break;
737 	case AF_INET6:
738 		if (!ipv6_addr_any(&output->u.ipv6.src) &&
739 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
740 				     &output->u.ipv6.src))
741 			return -EMSGSIZE;
742 		if (!ipv6_addr_any(&output->u.ipv6.dst) &&
743 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
744 				     &output->u.ipv6.dst))
745 			return -EMSGSIZE;
746 		break;
747 	}
748 	if (output->tos &&
749 	    nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
750 		return -EMSGSIZE;
751 	if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
752 		return -EMSGSIZE;
753 	if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
754 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
755 		return -EMSGSIZE;
756 	if ((output->tun_flags & TUNNEL_CSUM) &&
757 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
758 		return -EMSGSIZE;
759 	if (output->tp_src &&
760 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
761 		return -EMSGSIZE;
762 	if (output->tp_dst &&
763 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
764 		return -EMSGSIZE;
765 	if ((output->tun_flags & TUNNEL_OAM) &&
766 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
767 		return -EMSGSIZE;
768 	if (swkey_tun_opts_len) {
769 		if (output->tun_flags & TUNNEL_GENEVE_OPT &&
770 		    nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
771 			    swkey_tun_opts_len, tun_opts))
772 			return -EMSGSIZE;
773 		else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
774 			 vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
775 			return -EMSGSIZE;
776 	}
777 
778 	return 0;
779 }
780 
781 static int ip_tun_to_nlattr(struct sk_buff *skb,
782 			    const struct ip_tunnel_key *output,
783 			    const void *tun_opts, int swkey_tun_opts_len,
784 			    unsigned short tun_proto)
785 {
786 	struct nlattr *nla;
787 	int err;
788 
789 	nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
790 	if (!nla)
791 		return -EMSGSIZE;
792 
793 	err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
794 				 tun_proto);
795 	if (err)
796 		return err;
797 
798 	nla_nest_end(skb, nla);
799 	return 0;
800 }
801 
802 int ovs_nla_put_tunnel_info(struct sk_buff *skb,
803 			    struct ip_tunnel_info *tun_info)
804 {
805 	return __ip_tun_to_nlattr(skb, &tun_info->key,
806 				  ip_tunnel_info_opts(tun_info),
807 				  tun_info->options_len,
808 				  ip_tunnel_info_af(tun_info));
809 }
810 
811 static int encode_vlan_from_nlattrs(struct sw_flow_match *match,
812 				    const struct nlattr *a[],
813 				    bool is_mask, bool inner)
814 {
815 	__be16 tci = 0;
816 	__be16 tpid = 0;
817 
818 	if (a[OVS_KEY_ATTR_VLAN])
819 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
820 
821 	if (a[OVS_KEY_ATTR_ETHERTYPE])
822 		tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
823 
824 	if (likely(!inner)) {
825 		SW_FLOW_KEY_PUT(match, eth.vlan.tpid, tpid, is_mask);
826 		SW_FLOW_KEY_PUT(match, eth.vlan.tci, tci, is_mask);
827 	} else {
828 		SW_FLOW_KEY_PUT(match, eth.cvlan.tpid, tpid, is_mask);
829 		SW_FLOW_KEY_PUT(match, eth.cvlan.tci, tci, is_mask);
830 	}
831 	return 0;
832 }
833 
834 static int validate_vlan_from_nlattrs(const struct sw_flow_match *match,
835 				      u64 key_attrs, bool inner,
836 				      const struct nlattr **a, bool log)
837 {
838 	__be16 tci = 0;
839 
840 	if (!((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
841 	      (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
842 	       eth_type_vlan(nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE])))) {
843 		/* Not a VLAN. */
844 		return 0;
845 	}
846 
847 	if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
848 	      (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
849 		OVS_NLERR(log, "Invalid %s frame", (inner) ? "C-VLAN" : "VLAN");
850 		return -EINVAL;
851 	}
852 
853 	if (a[OVS_KEY_ATTR_VLAN])
854 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
855 
856 	if (!(tci & htons(VLAN_TAG_PRESENT))) {
857 		if (tci) {
858 			OVS_NLERR(log, "%s TCI does not have VLAN_TAG_PRESENT bit set.",
859 				  (inner) ? "C-VLAN" : "VLAN");
860 			return -EINVAL;
861 		} else if (nla_len(a[OVS_KEY_ATTR_ENCAP])) {
862 			/* Corner case for truncated VLAN header. */
863 			OVS_NLERR(log, "Truncated %s header has non-zero encap attribute.",
864 				  (inner) ? "C-VLAN" : "VLAN");
865 			return -EINVAL;
866 		}
867 	}
868 
869 	return 1;
870 }
871 
872 static int validate_vlan_mask_from_nlattrs(const struct sw_flow_match *match,
873 					   u64 key_attrs, bool inner,
874 					   const struct nlattr **a, bool log)
875 {
876 	__be16 tci = 0;
877 	__be16 tpid = 0;
878 	bool encap_valid = !!(match->key->eth.vlan.tci &
879 			      htons(VLAN_TAG_PRESENT));
880 	bool i_encap_valid = !!(match->key->eth.cvlan.tci &
881 				htons(VLAN_TAG_PRESENT));
882 
883 	if (!(key_attrs & (1 << OVS_KEY_ATTR_ENCAP))) {
884 		/* Not a VLAN. */
885 		return 0;
886 	}
887 
888 	if ((!inner && !encap_valid) || (inner && !i_encap_valid)) {
889 		OVS_NLERR(log, "Encap mask attribute is set for non-%s frame.",
890 			  (inner) ? "C-VLAN" : "VLAN");
891 		return -EINVAL;
892 	}
893 
894 	if (a[OVS_KEY_ATTR_VLAN])
895 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
896 
897 	if (a[OVS_KEY_ATTR_ETHERTYPE])
898 		tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
899 
900 	if (tpid != htons(0xffff)) {
901 		OVS_NLERR(log, "Must have an exact match on %s TPID (mask=%x).",
902 			  (inner) ? "C-VLAN" : "VLAN", ntohs(tpid));
903 		return -EINVAL;
904 	}
905 	if (!(tci & htons(VLAN_TAG_PRESENT))) {
906 		OVS_NLERR(log, "%s TCI mask does not have exact match for VLAN_TAG_PRESENT bit.",
907 			  (inner) ? "C-VLAN" : "VLAN");
908 		return -EINVAL;
909 	}
910 
911 	return 1;
912 }
913 
914 static int __parse_vlan_from_nlattrs(struct sw_flow_match *match,
915 				     u64 *key_attrs, bool inner,
916 				     const struct nlattr **a, bool is_mask,
917 				     bool log)
918 {
919 	int err;
920 	const struct nlattr *encap;
921 
922 	if (!is_mask)
923 		err = validate_vlan_from_nlattrs(match, *key_attrs, inner,
924 						 a, log);
925 	else
926 		err = validate_vlan_mask_from_nlattrs(match, *key_attrs, inner,
927 						      a, log);
928 	if (err <= 0)
929 		return err;
930 
931 	err = encode_vlan_from_nlattrs(match, a, is_mask, inner);
932 	if (err)
933 		return err;
934 
935 	*key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
936 	*key_attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
937 	*key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
938 
939 	encap = a[OVS_KEY_ATTR_ENCAP];
940 
941 	if (!is_mask)
942 		err = parse_flow_nlattrs(encap, a, key_attrs, log);
943 	else
944 		err = parse_flow_mask_nlattrs(encap, a, key_attrs, log);
945 
946 	return err;
947 }
948 
949 static int parse_vlan_from_nlattrs(struct sw_flow_match *match,
950 				   u64 *key_attrs, const struct nlattr **a,
951 				   bool is_mask, bool log)
952 {
953 	int err;
954 	bool encap_valid = false;
955 
956 	err = __parse_vlan_from_nlattrs(match, key_attrs, false, a,
957 					is_mask, log);
958 	if (err)
959 		return err;
960 
961 	encap_valid = !!(match->key->eth.vlan.tci & htons(VLAN_TAG_PRESENT));
962 	if (encap_valid) {
963 		err = __parse_vlan_from_nlattrs(match, key_attrs, true, a,
964 						is_mask, log);
965 		if (err)
966 			return err;
967 	}
968 
969 	return 0;
970 }
971 
972 static int parse_eth_type_from_nlattrs(struct sw_flow_match *match,
973 				       u64 *attrs, const struct nlattr **a,
974 				       bool is_mask, bool log)
975 {
976 	__be16 eth_type;
977 
978 	eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
979 	if (is_mask) {
980 		/* Always exact match EtherType. */
981 		eth_type = htons(0xffff);
982 	} else if (!eth_proto_is_802_3(eth_type)) {
983 		OVS_NLERR(log, "EtherType %x is less than min %x",
984 				ntohs(eth_type), ETH_P_802_3_MIN);
985 		return -EINVAL;
986 	}
987 
988 	SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
989 	*attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
990 	return 0;
991 }
992 
993 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
994 				 u64 *attrs, const struct nlattr **a,
995 				 bool is_mask, bool log)
996 {
997 	u8 mac_proto = MAC_PROTO_ETHERNET;
998 
999 	if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
1000 		u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
1001 
1002 		SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
1003 		*attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
1004 	}
1005 
1006 	if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
1007 		u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
1008 
1009 		SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
1010 		*attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
1011 	}
1012 
1013 	if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1014 		SW_FLOW_KEY_PUT(match, phy.priority,
1015 			  nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1016 		*attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1017 	}
1018 
1019 	if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1020 		u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1021 
1022 		if (is_mask) {
1023 			in_port = 0xffffffff; /* Always exact match in_port. */
1024 		} else if (in_port >= DP_MAX_PORTS) {
1025 			OVS_NLERR(log, "Port %d exceeds max allowable %d",
1026 				  in_port, DP_MAX_PORTS);
1027 			return -EINVAL;
1028 		}
1029 
1030 		SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1031 		*attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1032 	} else if (!is_mask) {
1033 		SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1034 	}
1035 
1036 	if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1037 		uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1038 
1039 		SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1040 		*attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1041 	}
1042 	if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1043 		if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1044 				       is_mask, log) < 0)
1045 			return -EINVAL;
1046 		*attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1047 	}
1048 
1049 	if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
1050 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
1051 		u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
1052 
1053 		if (ct_state & ~CT_SUPPORTED_MASK) {
1054 			OVS_NLERR(log, "ct_state flags %08x unsupported",
1055 				  ct_state);
1056 			return -EINVAL;
1057 		}
1058 
1059 		SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
1060 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
1061 	}
1062 	if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
1063 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
1064 		u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
1065 
1066 		SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
1067 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
1068 	}
1069 	if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
1070 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
1071 		u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
1072 
1073 		SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
1074 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
1075 	}
1076 	if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
1077 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
1078 		const struct ovs_key_ct_labels *cl;
1079 
1080 		cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
1081 		SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
1082 				   sizeof(*cl), is_mask);
1083 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
1084 	}
1085 
1086 	/* For layer 3 packets the Ethernet type is provided
1087 	 * and treated as metadata but no MAC addresses are provided.
1088 	 */
1089 	if (!(*attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1090 	    (*attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)))
1091 		mac_proto = MAC_PROTO_NONE;
1092 
1093 	/* Always exact match mac_proto */
1094 	SW_FLOW_KEY_PUT(match, mac_proto, is_mask ? 0xff : mac_proto, is_mask);
1095 
1096 	if (mac_proto == MAC_PROTO_NONE)
1097 		return parse_eth_type_from_nlattrs(match, attrs, a, is_mask,
1098 						   log);
1099 
1100 	return 0;
1101 }
1102 
1103 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
1104 				u64 attrs, const struct nlattr **a,
1105 				bool is_mask, bool log)
1106 {
1107 	int err;
1108 
1109 	err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
1110 	if (err)
1111 		return err;
1112 
1113 	if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
1114 		const struct ovs_key_ethernet *eth_key;
1115 
1116 		eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1117 		SW_FLOW_KEY_MEMCPY(match, eth.src,
1118 				eth_key->eth_src, ETH_ALEN, is_mask);
1119 		SW_FLOW_KEY_MEMCPY(match, eth.dst,
1120 				eth_key->eth_dst, ETH_ALEN, is_mask);
1121 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1122 
1123 		if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
1124 			/* VLAN attribute is always parsed before getting here since it
1125 			 * may occur multiple times.
1126 			 */
1127 			OVS_NLERR(log, "VLAN attribute unexpected.");
1128 			return -EINVAL;
1129 		}
1130 
1131 		if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1132 			err = parse_eth_type_from_nlattrs(match, &attrs, a, is_mask,
1133 							  log);
1134 			if (err)
1135 				return err;
1136 		} else if (!is_mask) {
1137 			SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1138 		}
1139 	} else if (!match->key->eth.type) {
1140 		OVS_NLERR(log, "Either Ethernet header or EtherType is required.");
1141 		return -EINVAL;
1142 	}
1143 
1144 	if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1145 		const struct ovs_key_ipv4 *ipv4_key;
1146 
1147 		ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1148 		if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1149 			OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
1150 				  ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1151 			return -EINVAL;
1152 		}
1153 		SW_FLOW_KEY_PUT(match, ip.proto,
1154 				ipv4_key->ipv4_proto, is_mask);
1155 		SW_FLOW_KEY_PUT(match, ip.tos,
1156 				ipv4_key->ipv4_tos, is_mask);
1157 		SW_FLOW_KEY_PUT(match, ip.ttl,
1158 				ipv4_key->ipv4_ttl, is_mask);
1159 		SW_FLOW_KEY_PUT(match, ip.frag,
1160 				ipv4_key->ipv4_frag, is_mask);
1161 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1162 				ipv4_key->ipv4_src, is_mask);
1163 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1164 				ipv4_key->ipv4_dst, is_mask);
1165 		attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1166 	}
1167 
1168 	if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
1169 		const struct ovs_key_ipv6 *ipv6_key;
1170 
1171 		ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1172 		if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1173 			OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
1174 				  ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1175 			return -EINVAL;
1176 		}
1177 
1178 		if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
1179 			OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
1180 				  ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
1181 			return -EINVAL;
1182 		}
1183 
1184 		SW_FLOW_KEY_PUT(match, ipv6.label,
1185 				ipv6_key->ipv6_label, is_mask);
1186 		SW_FLOW_KEY_PUT(match, ip.proto,
1187 				ipv6_key->ipv6_proto, is_mask);
1188 		SW_FLOW_KEY_PUT(match, ip.tos,
1189 				ipv6_key->ipv6_tclass, is_mask);
1190 		SW_FLOW_KEY_PUT(match, ip.ttl,
1191 				ipv6_key->ipv6_hlimit, is_mask);
1192 		SW_FLOW_KEY_PUT(match, ip.frag,
1193 				ipv6_key->ipv6_frag, is_mask);
1194 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1195 				ipv6_key->ipv6_src,
1196 				sizeof(match->key->ipv6.addr.src),
1197 				is_mask);
1198 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1199 				ipv6_key->ipv6_dst,
1200 				sizeof(match->key->ipv6.addr.dst),
1201 				is_mask);
1202 
1203 		attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1204 	}
1205 
1206 	if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1207 		const struct ovs_key_arp *arp_key;
1208 
1209 		arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1210 		if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1211 			OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1212 				  arp_key->arp_op);
1213 			return -EINVAL;
1214 		}
1215 
1216 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1217 				arp_key->arp_sip, is_mask);
1218 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1219 			arp_key->arp_tip, is_mask);
1220 		SW_FLOW_KEY_PUT(match, ip.proto,
1221 				ntohs(arp_key->arp_op), is_mask);
1222 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1223 				arp_key->arp_sha, ETH_ALEN, is_mask);
1224 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1225 				arp_key->arp_tha, ETH_ALEN, is_mask);
1226 
1227 		attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1228 	}
1229 
1230 	if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1231 		const struct ovs_key_mpls *mpls_key;
1232 
1233 		mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1234 		SW_FLOW_KEY_PUT(match, mpls.top_lse,
1235 				mpls_key->mpls_lse, is_mask);
1236 
1237 		attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1238 	 }
1239 
1240 	if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1241 		const struct ovs_key_tcp *tcp_key;
1242 
1243 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1244 		SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1245 		SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1246 		attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1247 	}
1248 
1249 	if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1250 		SW_FLOW_KEY_PUT(match, tp.flags,
1251 				nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1252 				is_mask);
1253 		attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1254 	}
1255 
1256 	if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1257 		const struct ovs_key_udp *udp_key;
1258 
1259 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1260 		SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1261 		SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1262 		attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1263 	}
1264 
1265 	if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1266 		const struct ovs_key_sctp *sctp_key;
1267 
1268 		sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1269 		SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1270 		SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1271 		attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1272 	}
1273 
1274 	if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1275 		const struct ovs_key_icmp *icmp_key;
1276 
1277 		icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1278 		SW_FLOW_KEY_PUT(match, tp.src,
1279 				htons(icmp_key->icmp_type), is_mask);
1280 		SW_FLOW_KEY_PUT(match, tp.dst,
1281 				htons(icmp_key->icmp_code), is_mask);
1282 		attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1283 	}
1284 
1285 	if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1286 		const struct ovs_key_icmpv6 *icmpv6_key;
1287 
1288 		icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1289 		SW_FLOW_KEY_PUT(match, tp.src,
1290 				htons(icmpv6_key->icmpv6_type), is_mask);
1291 		SW_FLOW_KEY_PUT(match, tp.dst,
1292 				htons(icmpv6_key->icmpv6_code), is_mask);
1293 		attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1294 	}
1295 
1296 	if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1297 		const struct ovs_key_nd *nd_key;
1298 
1299 		nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1300 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1301 			nd_key->nd_target,
1302 			sizeof(match->key->ipv6.nd.target),
1303 			is_mask);
1304 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1305 			nd_key->nd_sll, ETH_ALEN, is_mask);
1306 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1307 				nd_key->nd_tll, ETH_ALEN, is_mask);
1308 		attrs &= ~(1 << OVS_KEY_ATTR_ND);
1309 	}
1310 
1311 	if (attrs != 0) {
1312 		OVS_NLERR(log, "Unknown key attributes %llx",
1313 			  (unsigned long long)attrs);
1314 		return -EINVAL;
1315 	}
1316 
1317 	return 0;
1318 }
1319 
1320 static void nlattr_set(struct nlattr *attr, u8 val,
1321 		       const struct ovs_len_tbl *tbl)
1322 {
1323 	struct nlattr *nla;
1324 	int rem;
1325 
1326 	/* The nlattr stream should already have been validated */
1327 	nla_for_each_nested(nla, attr, rem) {
1328 		if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
1329 			if (tbl[nla_type(nla)].next)
1330 				tbl = tbl[nla_type(nla)].next;
1331 			nlattr_set(nla, val, tbl);
1332 		} else {
1333 			memset(nla_data(nla), val, nla_len(nla));
1334 		}
1335 
1336 		if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1337 			*(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1338 	}
1339 }
1340 
1341 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1342 {
1343 	nlattr_set(attr, val, ovs_key_lens);
1344 }
1345 
1346 /**
1347  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1348  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1349  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1350  * does not include any don't care bit.
1351  * @net: Used to determine per-namespace field support.
1352  * @match: receives the extracted flow match information.
1353  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1354  * sequence. The fields should of the packet that triggered the creation
1355  * of this flow.
1356  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1357  * attribute specifies the mask field of the wildcarded flow.
1358  * @log: Boolean to allow kernel error logging.  Normally true, but when
1359  * probing for feature compatibility this should be passed in as false to
1360  * suppress unnecessary error logging.
1361  */
1362 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1363 		      const struct nlattr *nla_key,
1364 		      const struct nlattr *nla_mask,
1365 		      bool log)
1366 {
1367 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1368 	struct nlattr *newmask = NULL;
1369 	u64 key_attrs = 0;
1370 	u64 mask_attrs = 0;
1371 	int err;
1372 
1373 	err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1374 	if (err)
1375 		return err;
1376 
1377 	err = parse_vlan_from_nlattrs(match, &key_attrs, a, false, log);
1378 	if (err)
1379 		return err;
1380 
1381 	err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1382 	if (err)
1383 		return err;
1384 
1385 	if (match->mask) {
1386 		if (!nla_mask) {
1387 			/* Create an exact match mask. We need to set to 0xff
1388 			 * all the 'match->mask' fields that have been touched
1389 			 * in 'match->key'. We cannot simply memset
1390 			 * 'match->mask', because padding bytes and fields not
1391 			 * specified in 'match->key' should be left to 0.
1392 			 * Instead, we use a stream of netlink attributes,
1393 			 * copied from 'key' and set to 0xff.
1394 			 * ovs_key_from_nlattrs() will take care of filling
1395 			 * 'match->mask' appropriately.
1396 			 */
1397 			newmask = kmemdup(nla_key,
1398 					  nla_total_size(nla_len(nla_key)),
1399 					  GFP_KERNEL);
1400 			if (!newmask)
1401 				return -ENOMEM;
1402 
1403 			mask_set_nlattr(newmask, 0xff);
1404 
1405 			/* The userspace does not send tunnel attributes that
1406 			 * are 0, but we should not wildcard them nonetheless.
1407 			 */
1408 			if (match->key->tun_proto)
1409 				SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1410 							 0xff, true);
1411 
1412 			nla_mask = newmask;
1413 		}
1414 
1415 		err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1416 		if (err)
1417 			goto free_newmask;
1418 
1419 		/* Always match on tci. */
1420 		SW_FLOW_KEY_PUT(match, eth.vlan.tci, htons(0xffff), true);
1421 		SW_FLOW_KEY_PUT(match, eth.cvlan.tci, htons(0xffff), true);
1422 
1423 		err = parse_vlan_from_nlattrs(match, &mask_attrs, a, true, log);
1424 		if (err)
1425 			goto free_newmask;
1426 
1427 		err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1428 					   log);
1429 		if (err)
1430 			goto free_newmask;
1431 	}
1432 
1433 	if (!match_validate(match, key_attrs, mask_attrs, log))
1434 		err = -EINVAL;
1435 
1436 free_newmask:
1437 	kfree(newmask);
1438 	return err;
1439 }
1440 
1441 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1442 {
1443 	size_t len;
1444 
1445 	if (!attr)
1446 		return 0;
1447 
1448 	len = nla_len(attr);
1449 	if (len < 1 || len > MAX_UFID_LENGTH) {
1450 		OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1451 			  nla_len(attr), MAX_UFID_LENGTH);
1452 		return 0;
1453 	}
1454 
1455 	return len;
1456 }
1457 
1458 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1459  * or false otherwise.
1460  */
1461 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1462 		      bool log)
1463 {
1464 	sfid->ufid_len = get_ufid_len(attr, log);
1465 	if (sfid->ufid_len)
1466 		memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1467 
1468 	return sfid->ufid_len;
1469 }
1470 
1471 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1472 			   const struct sw_flow_key *key, bool log)
1473 {
1474 	struct sw_flow_key *new_key;
1475 
1476 	if (ovs_nla_get_ufid(sfid, ufid, log))
1477 		return 0;
1478 
1479 	/* If UFID was not provided, use unmasked key. */
1480 	new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1481 	if (!new_key)
1482 		return -ENOMEM;
1483 	memcpy(new_key, key, sizeof(*key));
1484 	sfid->unmasked_key = new_key;
1485 
1486 	return 0;
1487 }
1488 
1489 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1490 {
1491 	return attr ? nla_get_u32(attr) : 0;
1492 }
1493 
1494 /**
1495  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1496  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1497  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1498  * sequence.
1499  * @log: Boolean to allow kernel error logging.  Normally true, but when
1500  * probing for feature compatibility this should be passed in as false to
1501  * suppress unnecessary error logging.
1502  *
1503  * This parses a series of Netlink attributes that form a flow key, which must
1504  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1505  * get the metadata, that is, the parts of the flow key that cannot be
1506  * extracted from the packet itself.
1507  */
1508 
1509 int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
1510 			      struct sw_flow_key *key,
1511 			      bool log)
1512 {
1513 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1514 	struct sw_flow_match match;
1515 	u64 attrs = 0;
1516 	int err;
1517 
1518 	err = parse_flow_nlattrs(attr, a, &attrs, log);
1519 	if (err)
1520 		return -EINVAL;
1521 
1522 	memset(&match, 0, sizeof(match));
1523 	match.key = key;
1524 
1525 	memset(&key->ct, 0, sizeof(key->ct));
1526 	key->phy.in_port = DP_MAX_PORTS;
1527 
1528 	return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1529 }
1530 
1531 static int ovs_nla_put_vlan(struct sk_buff *skb, const struct vlan_head *vh,
1532 			    bool is_mask)
1533 {
1534 	__be16 eth_type = !is_mask ? vh->tpid : htons(0xffff);
1535 
1536 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1537 	    nla_put_be16(skb, OVS_KEY_ATTR_VLAN, vh->tci))
1538 		return -EMSGSIZE;
1539 	return 0;
1540 }
1541 
1542 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1543 			     const struct sw_flow_key *output, bool is_mask,
1544 			     struct sk_buff *skb)
1545 {
1546 	struct ovs_key_ethernet *eth_key;
1547 	struct nlattr *nla;
1548 	struct nlattr *encap = NULL;
1549 	struct nlattr *in_encap = NULL;
1550 
1551 	if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1552 		goto nla_put_failure;
1553 
1554 	if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1555 		goto nla_put_failure;
1556 
1557 	if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1558 		goto nla_put_failure;
1559 
1560 	if ((swkey->tun_proto || is_mask)) {
1561 		const void *opts = NULL;
1562 
1563 		if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1564 			opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1565 
1566 		if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
1567 				     swkey->tun_opts_len, swkey->tun_proto))
1568 			goto nla_put_failure;
1569 	}
1570 
1571 	if (swkey->phy.in_port == DP_MAX_PORTS) {
1572 		if (is_mask && (output->phy.in_port == 0xffff))
1573 			if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1574 				goto nla_put_failure;
1575 	} else {
1576 		u16 upper_u16;
1577 		upper_u16 = !is_mask ? 0 : 0xffff;
1578 
1579 		if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1580 				(upper_u16 << 16) | output->phy.in_port))
1581 			goto nla_put_failure;
1582 	}
1583 
1584 	if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1585 		goto nla_put_failure;
1586 
1587 	if (ovs_ct_put_key(output, skb))
1588 		goto nla_put_failure;
1589 
1590 	if (ovs_key_mac_proto(swkey) == MAC_PROTO_ETHERNET) {
1591 		nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1592 		if (!nla)
1593 			goto nla_put_failure;
1594 
1595 		eth_key = nla_data(nla);
1596 		ether_addr_copy(eth_key->eth_src, output->eth.src);
1597 		ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1598 
1599 		if (swkey->eth.vlan.tci || eth_type_vlan(swkey->eth.type)) {
1600 			if (ovs_nla_put_vlan(skb, &output->eth.vlan, is_mask))
1601 				goto nla_put_failure;
1602 			encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1603 			if (!swkey->eth.vlan.tci)
1604 				goto unencap;
1605 
1606 			if (swkey->eth.cvlan.tci || eth_type_vlan(swkey->eth.type)) {
1607 				if (ovs_nla_put_vlan(skb, &output->eth.cvlan, is_mask))
1608 					goto nla_put_failure;
1609 				in_encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1610 				if (!swkey->eth.cvlan.tci)
1611 					goto unencap;
1612 			}
1613 		}
1614 
1615 		if (swkey->eth.type == htons(ETH_P_802_2)) {
1616 			/*
1617 			* Ethertype 802.2 is represented in the netlink with omitted
1618 			* OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1619 			* 0xffff in the mask attribute.  Ethertype can also
1620 			* be wildcarded.
1621 			*/
1622 			if (is_mask && output->eth.type)
1623 				if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1624 							output->eth.type))
1625 					goto nla_put_failure;
1626 			goto unencap;
1627 		}
1628 	}
1629 
1630 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1631 		goto nla_put_failure;
1632 
1633 	if (eth_type_vlan(swkey->eth.type)) {
1634 		/* There are 3 VLAN tags, we don't know anything about the rest
1635 		 * of the packet, so truncate here.
1636 		 */
1637 		WARN_ON_ONCE(!(encap && in_encap));
1638 		goto unencap;
1639 	}
1640 
1641 	if (swkey->eth.type == htons(ETH_P_IP)) {
1642 		struct ovs_key_ipv4 *ipv4_key;
1643 
1644 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1645 		if (!nla)
1646 			goto nla_put_failure;
1647 		ipv4_key = nla_data(nla);
1648 		ipv4_key->ipv4_src = output->ipv4.addr.src;
1649 		ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1650 		ipv4_key->ipv4_proto = output->ip.proto;
1651 		ipv4_key->ipv4_tos = output->ip.tos;
1652 		ipv4_key->ipv4_ttl = output->ip.ttl;
1653 		ipv4_key->ipv4_frag = output->ip.frag;
1654 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1655 		struct ovs_key_ipv6 *ipv6_key;
1656 
1657 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1658 		if (!nla)
1659 			goto nla_put_failure;
1660 		ipv6_key = nla_data(nla);
1661 		memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1662 				sizeof(ipv6_key->ipv6_src));
1663 		memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1664 				sizeof(ipv6_key->ipv6_dst));
1665 		ipv6_key->ipv6_label = output->ipv6.label;
1666 		ipv6_key->ipv6_proto = output->ip.proto;
1667 		ipv6_key->ipv6_tclass = output->ip.tos;
1668 		ipv6_key->ipv6_hlimit = output->ip.ttl;
1669 		ipv6_key->ipv6_frag = output->ip.frag;
1670 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
1671 		   swkey->eth.type == htons(ETH_P_RARP)) {
1672 		struct ovs_key_arp *arp_key;
1673 
1674 		nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1675 		if (!nla)
1676 			goto nla_put_failure;
1677 		arp_key = nla_data(nla);
1678 		memset(arp_key, 0, sizeof(struct ovs_key_arp));
1679 		arp_key->arp_sip = output->ipv4.addr.src;
1680 		arp_key->arp_tip = output->ipv4.addr.dst;
1681 		arp_key->arp_op = htons(output->ip.proto);
1682 		ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1683 		ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1684 	} else if (eth_p_mpls(swkey->eth.type)) {
1685 		struct ovs_key_mpls *mpls_key;
1686 
1687 		nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1688 		if (!nla)
1689 			goto nla_put_failure;
1690 		mpls_key = nla_data(nla);
1691 		mpls_key->mpls_lse = output->mpls.top_lse;
1692 	}
1693 
1694 	if ((swkey->eth.type == htons(ETH_P_IP) ||
1695 	     swkey->eth.type == htons(ETH_P_IPV6)) &&
1696 	     swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1697 
1698 		if (swkey->ip.proto == IPPROTO_TCP) {
1699 			struct ovs_key_tcp *tcp_key;
1700 
1701 			nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1702 			if (!nla)
1703 				goto nla_put_failure;
1704 			tcp_key = nla_data(nla);
1705 			tcp_key->tcp_src = output->tp.src;
1706 			tcp_key->tcp_dst = output->tp.dst;
1707 			if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1708 					 output->tp.flags))
1709 				goto nla_put_failure;
1710 		} else if (swkey->ip.proto == IPPROTO_UDP) {
1711 			struct ovs_key_udp *udp_key;
1712 
1713 			nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1714 			if (!nla)
1715 				goto nla_put_failure;
1716 			udp_key = nla_data(nla);
1717 			udp_key->udp_src = output->tp.src;
1718 			udp_key->udp_dst = output->tp.dst;
1719 		} else if (swkey->ip.proto == IPPROTO_SCTP) {
1720 			struct ovs_key_sctp *sctp_key;
1721 
1722 			nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1723 			if (!nla)
1724 				goto nla_put_failure;
1725 			sctp_key = nla_data(nla);
1726 			sctp_key->sctp_src = output->tp.src;
1727 			sctp_key->sctp_dst = output->tp.dst;
1728 		} else if (swkey->eth.type == htons(ETH_P_IP) &&
1729 			   swkey->ip.proto == IPPROTO_ICMP) {
1730 			struct ovs_key_icmp *icmp_key;
1731 
1732 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1733 			if (!nla)
1734 				goto nla_put_failure;
1735 			icmp_key = nla_data(nla);
1736 			icmp_key->icmp_type = ntohs(output->tp.src);
1737 			icmp_key->icmp_code = ntohs(output->tp.dst);
1738 		} else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1739 			   swkey->ip.proto == IPPROTO_ICMPV6) {
1740 			struct ovs_key_icmpv6 *icmpv6_key;
1741 
1742 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1743 						sizeof(*icmpv6_key));
1744 			if (!nla)
1745 				goto nla_put_failure;
1746 			icmpv6_key = nla_data(nla);
1747 			icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1748 			icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1749 
1750 			if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1751 			    icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1752 				struct ovs_key_nd *nd_key;
1753 
1754 				nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1755 				if (!nla)
1756 					goto nla_put_failure;
1757 				nd_key = nla_data(nla);
1758 				memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1759 							sizeof(nd_key->nd_target));
1760 				ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1761 				ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1762 			}
1763 		}
1764 	}
1765 
1766 unencap:
1767 	if (in_encap)
1768 		nla_nest_end(skb, in_encap);
1769 	if (encap)
1770 		nla_nest_end(skb, encap);
1771 
1772 	return 0;
1773 
1774 nla_put_failure:
1775 	return -EMSGSIZE;
1776 }
1777 
1778 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1779 		    const struct sw_flow_key *output, int attr, bool is_mask,
1780 		    struct sk_buff *skb)
1781 {
1782 	int err;
1783 	struct nlattr *nla;
1784 
1785 	nla = nla_nest_start(skb, attr);
1786 	if (!nla)
1787 		return -EMSGSIZE;
1788 	err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1789 	if (err)
1790 		return err;
1791 	nla_nest_end(skb, nla);
1792 
1793 	return 0;
1794 }
1795 
1796 /* Called with ovs_mutex or RCU read lock. */
1797 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1798 {
1799 	if (ovs_identifier_is_ufid(&flow->id))
1800 		return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1801 			       flow->id.ufid);
1802 
1803 	return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1804 			       OVS_FLOW_ATTR_KEY, false, skb);
1805 }
1806 
1807 /* Called with ovs_mutex or RCU read lock. */
1808 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1809 {
1810 	return ovs_nla_put_key(&flow->key, &flow->key,
1811 				OVS_FLOW_ATTR_KEY, false, skb);
1812 }
1813 
1814 /* Called with ovs_mutex or RCU read lock. */
1815 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1816 {
1817 	return ovs_nla_put_key(&flow->key, &flow->mask->key,
1818 				OVS_FLOW_ATTR_MASK, true, skb);
1819 }
1820 
1821 #define MAX_ACTIONS_BUFSIZE	(32 * 1024)
1822 
1823 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1824 {
1825 	struct sw_flow_actions *sfa;
1826 
1827 	if (size > MAX_ACTIONS_BUFSIZE) {
1828 		OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1829 		return ERR_PTR(-EINVAL);
1830 	}
1831 
1832 	sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1833 	if (!sfa)
1834 		return ERR_PTR(-ENOMEM);
1835 
1836 	sfa->actions_len = 0;
1837 	return sfa;
1838 }
1839 
1840 static void ovs_nla_free_set_action(const struct nlattr *a)
1841 {
1842 	const struct nlattr *ovs_key = nla_data(a);
1843 	struct ovs_tunnel_info *ovs_tun;
1844 
1845 	switch (nla_type(ovs_key)) {
1846 	case OVS_KEY_ATTR_TUNNEL_INFO:
1847 		ovs_tun = nla_data(ovs_key);
1848 		dst_release((struct dst_entry *)ovs_tun->tun_dst);
1849 		break;
1850 	}
1851 }
1852 
1853 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1854 {
1855 	const struct nlattr *a;
1856 	int rem;
1857 
1858 	if (!sf_acts)
1859 		return;
1860 
1861 	nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1862 		switch (nla_type(a)) {
1863 		case OVS_ACTION_ATTR_SET:
1864 			ovs_nla_free_set_action(a);
1865 			break;
1866 		case OVS_ACTION_ATTR_CT:
1867 			ovs_ct_free_action(a);
1868 			break;
1869 		}
1870 	}
1871 
1872 	kfree(sf_acts);
1873 }
1874 
1875 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1876 {
1877 	ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1878 }
1879 
1880 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1881  * The caller must hold rcu_read_lock for this to be sensible. */
1882 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
1883 {
1884 	call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
1885 }
1886 
1887 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1888 				       int attr_len, bool log)
1889 {
1890 
1891 	struct sw_flow_actions *acts;
1892 	int new_acts_size;
1893 	int req_size = NLA_ALIGN(attr_len);
1894 	int next_offset = offsetof(struct sw_flow_actions, actions) +
1895 					(*sfa)->actions_len;
1896 
1897 	if (req_size <= (ksize(*sfa) - next_offset))
1898 		goto out;
1899 
1900 	new_acts_size = ksize(*sfa) * 2;
1901 
1902 	if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1903 		if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1904 			return ERR_PTR(-EMSGSIZE);
1905 		new_acts_size = MAX_ACTIONS_BUFSIZE;
1906 	}
1907 
1908 	acts = nla_alloc_flow_actions(new_acts_size, log);
1909 	if (IS_ERR(acts))
1910 		return (void *)acts;
1911 
1912 	memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1913 	acts->actions_len = (*sfa)->actions_len;
1914 	acts->orig_len = (*sfa)->orig_len;
1915 	kfree(*sfa);
1916 	*sfa = acts;
1917 
1918 out:
1919 	(*sfa)->actions_len += req_size;
1920 	return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1921 }
1922 
1923 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1924 				   int attrtype, void *data, int len, bool log)
1925 {
1926 	struct nlattr *a;
1927 
1928 	a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1929 	if (IS_ERR(a))
1930 		return a;
1931 
1932 	a->nla_type = attrtype;
1933 	a->nla_len = nla_attr_size(len);
1934 
1935 	if (data)
1936 		memcpy(nla_data(a), data, len);
1937 	memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1938 
1939 	return a;
1940 }
1941 
1942 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1943 		       int len, bool log)
1944 {
1945 	struct nlattr *a;
1946 
1947 	a = __add_action(sfa, attrtype, data, len, log);
1948 
1949 	return PTR_ERR_OR_ZERO(a);
1950 }
1951 
1952 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1953 					  int attrtype, bool log)
1954 {
1955 	int used = (*sfa)->actions_len;
1956 	int err;
1957 
1958 	err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
1959 	if (err)
1960 		return err;
1961 
1962 	return used;
1963 }
1964 
1965 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1966 					 int st_offset)
1967 {
1968 	struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1969 							       st_offset);
1970 
1971 	a->nla_len = sfa->actions_len - st_offset;
1972 }
1973 
1974 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
1975 				  const struct sw_flow_key *key,
1976 				  int depth, struct sw_flow_actions **sfa,
1977 				  __be16 eth_type, __be16 vlan_tci, bool log);
1978 
1979 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
1980 				    const struct sw_flow_key *key, int depth,
1981 				    struct sw_flow_actions **sfa,
1982 				    __be16 eth_type, __be16 vlan_tci, bool log)
1983 {
1984 	const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1985 	const struct nlattr *probability, *actions;
1986 	const struct nlattr *a;
1987 	int rem, start, err, st_acts;
1988 
1989 	memset(attrs, 0, sizeof(attrs));
1990 	nla_for_each_nested(a, attr, rem) {
1991 		int type = nla_type(a);
1992 		if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1993 			return -EINVAL;
1994 		attrs[type] = a;
1995 	}
1996 	if (rem)
1997 		return -EINVAL;
1998 
1999 	probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
2000 	if (!probability || nla_len(probability) != sizeof(u32))
2001 		return -EINVAL;
2002 
2003 	actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
2004 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
2005 		return -EINVAL;
2006 
2007 	/* validation done, copy sample action. */
2008 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
2009 	if (start < 0)
2010 		return start;
2011 	err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
2012 				 nla_data(probability), sizeof(u32), log);
2013 	if (err)
2014 		return err;
2015 	st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
2016 	if (st_acts < 0)
2017 		return st_acts;
2018 
2019 	err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
2020 				     eth_type, vlan_tci, log);
2021 	if (err)
2022 		return err;
2023 
2024 	add_nested_action_end(*sfa, st_acts);
2025 	add_nested_action_end(*sfa, start);
2026 
2027 	return 0;
2028 }
2029 
2030 void ovs_match_init(struct sw_flow_match *match,
2031 		    struct sw_flow_key *key,
2032 		    bool reset_key,
2033 		    struct sw_flow_mask *mask)
2034 {
2035 	memset(match, 0, sizeof(*match));
2036 	match->key = key;
2037 	match->mask = mask;
2038 
2039 	if (reset_key)
2040 		memset(key, 0, sizeof(*key));
2041 
2042 	if (mask) {
2043 		memset(&mask->key, 0, sizeof(mask->key));
2044 		mask->range.start = mask->range.end = 0;
2045 	}
2046 }
2047 
2048 static int validate_geneve_opts(struct sw_flow_key *key)
2049 {
2050 	struct geneve_opt *option;
2051 	int opts_len = key->tun_opts_len;
2052 	bool crit_opt = false;
2053 
2054 	option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
2055 	while (opts_len > 0) {
2056 		int len;
2057 
2058 		if (opts_len < sizeof(*option))
2059 			return -EINVAL;
2060 
2061 		len = sizeof(*option) + option->length * 4;
2062 		if (len > opts_len)
2063 			return -EINVAL;
2064 
2065 		crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
2066 
2067 		option = (struct geneve_opt *)((u8 *)option + len);
2068 		opts_len -= len;
2069 	};
2070 
2071 	key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
2072 
2073 	return 0;
2074 }
2075 
2076 static int validate_and_copy_set_tun(const struct nlattr *attr,
2077 				     struct sw_flow_actions **sfa, bool log)
2078 {
2079 	struct sw_flow_match match;
2080 	struct sw_flow_key key;
2081 	struct metadata_dst *tun_dst;
2082 	struct ip_tunnel_info *tun_info;
2083 	struct ovs_tunnel_info *ovs_tun;
2084 	struct nlattr *a;
2085 	int err = 0, start, opts_type;
2086 
2087 	ovs_match_init(&match, &key, true, NULL);
2088 	opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
2089 	if (opts_type < 0)
2090 		return opts_type;
2091 
2092 	if (key.tun_opts_len) {
2093 		switch (opts_type) {
2094 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
2095 			err = validate_geneve_opts(&key);
2096 			if (err < 0)
2097 				return err;
2098 			break;
2099 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
2100 			break;
2101 		}
2102 	};
2103 
2104 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
2105 	if (start < 0)
2106 		return start;
2107 
2108 	tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
2109 	if (!tun_dst)
2110 		return -ENOMEM;
2111 
2112 	err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
2113 	if (err) {
2114 		dst_release((struct dst_entry *)tun_dst);
2115 		return err;
2116 	}
2117 
2118 	a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
2119 			 sizeof(*ovs_tun), log);
2120 	if (IS_ERR(a)) {
2121 		dst_release((struct dst_entry *)tun_dst);
2122 		return PTR_ERR(a);
2123 	}
2124 
2125 	ovs_tun = nla_data(a);
2126 	ovs_tun->tun_dst = tun_dst;
2127 
2128 	tun_info = &tun_dst->u.tun_info;
2129 	tun_info->mode = IP_TUNNEL_INFO_TX;
2130 	if (key.tun_proto == AF_INET6)
2131 		tun_info->mode |= IP_TUNNEL_INFO_IPV6;
2132 	tun_info->key = key.tun_key;
2133 
2134 	/* We need to store the options in the action itself since
2135 	 * everything else will go away after flow setup. We can append
2136 	 * it to tun_info and then point there.
2137 	 */
2138 	ip_tunnel_info_opts_set(tun_info,
2139 				TUN_METADATA_OPTS(&key, key.tun_opts_len),
2140 				key.tun_opts_len);
2141 	add_nested_action_end(*sfa, start);
2142 
2143 	return err;
2144 }
2145 
2146 /* Return false if there are any non-masked bits set.
2147  * Mask follows data immediately, before any netlink padding.
2148  */
2149 static bool validate_masked(u8 *data, int len)
2150 {
2151 	u8 *mask = data + len;
2152 
2153 	while (len--)
2154 		if (*data++ & ~*mask++)
2155 			return false;
2156 
2157 	return true;
2158 }
2159 
2160 static int validate_set(const struct nlattr *a,
2161 			const struct sw_flow_key *flow_key,
2162 			struct sw_flow_actions **sfa, bool *skip_copy,
2163 			u8 mac_proto, __be16 eth_type, bool masked, bool log)
2164 {
2165 	const struct nlattr *ovs_key = nla_data(a);
2166 	int key_type = nla_type(ovs_key);
2167 	size_t key_len;
2168 
2169 	/* There can be only one key in a action */
2170 	if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2171 		return -EINVAL;
2172 
2173 	key_len = nla_len(ovs_key);
2174 	if (masked)
2175 		key_len /= 2;
2176 
2177 	if (key_type > OVS_KEY_ATTR_MAX ||
2178 	    !check_attr_len(key_len, ovs_key_lens[key_type].len))
2179 		return -EINVAL;
2180 
2181 	if (masked && !validate_masked(nla_data(ovs_key), key_len))
2182 		return -EINVAL;
2183 
2184 	switch (key_type) {
2185 	const struct ovs_key_ipv4 *ipv4_key;
2186 	const struct ovs_key_ipv6 *ipv6_key;
2187 	int err;
2188 
2189 	case OVS_KEY_ATTR_PRIORITY:
2190 	case OVS_KEY_ATTR_SKB_MARK:
2191 	case OVS_KEY_ATTR_CT_MARK:
2192 	case OVS_KEY_ATTR_CT_LABELS:
2193 		break;
2194 
2195 	case OVS_KEY_ATTR_ETHERNET:
2196 		if (mac_proto != MAC_PROTO_ETHERNET)
2197 			return -EINVAL;
2198 
2199 	case OVS_KEY_ATTR_TUNNEL:
2200 		if (masked)
2201 			return -EINVAL; /* Masked tunnel set not supported. */
2202 
2203 		*skip_copy = true;
2204 		err = validate_and_copy_set_tun(a, sfa, log);
2205 		if (err)
2206 			return err;
2207 		break;
2208 
2209 	case OVS_KEY_ATTR_IPV4:
2210 		if (eth_type != htons(ETH_P_IP))
2211 			return -EINVAL;
2212 
2213 		ipv4_key = nla_data(ovs_key);
2214 
2215 		if (masked) {
2216 			const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2217 
2218 			/* Non-writeable fields. */
2219 			if (mask->ipv4_proto || mask->ipv4_frag)
2220 				return -EINVAL;
2221 		} else {
2222 			if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2223 				return -EINVAL;
2224 
2225 			if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2226 				return -EINVAL;
2227 		}
2228 		break;
2229 
2230 	case OVS_KEY_ATTR_IPV6:
2231 		if (eth_type != htons(ETH_P_IPV6))
2232 			return -EINVAL;
2233 
2234 		ipv6_key = nla_data(ovs_key);
2235 
2236 		if (masked) {
2237 			const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2238 
2239 			/* Non-writeable fields. */
2240 			if (mask->ipv6_proto || mask->ipv6_frag)
2241 				return -EINVAL;
2242 
2243 			/* Invalid bits in the flow label mask? */
2244 			if (ntohl(mask->ipv6_label) & 0xFFF00000)
2245 				return -EINVAL;
2246 		} else {
2247 			if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2248 				return -EINVAL;
2249 
2250 			if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2251 				return -EINVAL;
2252 		}
2253 		if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2254 			return -EINVAL;
2255 
2256 		break;
2257 
2258 	case OVS_KEY_ATTR_TCP:
2259 		if ((eth_type != htons(ETH_P_IP) &&
2260 		     eth_type != htons(ETH_P_IPV6)) ||
2261 		    flow_key->ip.proto != IPPROTO_TCP)
2262 			return -EINVAL;
2263 
2264 		break;
2265 
2266 	case OVS_KEY_ATTR_UDP:
2267 		if ((eth_type != htons(ETH_P_IP) &&
2268 		     eth_type != htons(ETH_P_IPV6)) ||
2269 		    flow_key->ip.proto != IPPROTO_UDP)
2270 			return -EINVAL;
2271 
2272 		break;
2273 
2274 	case OVS_KEY_ATTR_MPLS:
2275 		if (!eth_p_mpls(eth_type))
2276 			return -EINVAL;
2277 		break;
2278 
2279 	case OVS_KEY_ATTR_SCTP:
2280 		if ((eth_type != htons(ETH_P_IP) &&
2281 		     eth_type != htons(ETH_P_IPV6)) ||
2282 		    flow_key->ip.proto != IPPROTO_SCTP)
2283 			return -EINVAL;
2284 
2285 		break;
2286 
2287 	default:
2288 		return -EINVAL;
2289 	}
2290 
2291 	/* Convert non-masked non-tunnel set actions to masked set actions. */
2292 	if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2293 		int start, len = key_len * 2;
2294 		struct nlattr *at;
2295 
2296 		*skip_copy = true;
2297 
2298 		start = add_nested_action_start(sfa,
2299 						OVS_ACTION_ATTR_SET_TO_MASKED,
2300 						log);
2301 		if (start < 0)
2302 			return start;
2303 
2304 		at = __add_action(sfa, key_type, NULL, len, log);
2305 		if (IS_ERR(at))
2306 			return PTR_ERR(at);
2307 
2308 		memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2309 		memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2310 		/* Clear non-writeable bits from otherwise writeable fields. */
2311 		if (key_type == OVS_KEY_ATTR_IPV6) {
2312 			struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2313 
2314 			mask->ipv6_label &= htonl(0x000FFFFF);
2315 		}
2316 		add_nested_action_end(*sfa, start);
2317 	}
2318 
2319 	return 0;
2320 }
2321 
2322 static int validate_userspace(const struct nlattr *attr)
2323 {
2324 	static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2325 		[OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2326 		[OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2327 		[OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2328 	};
2329 	struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2330 	int error;
2331 
2332 	error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2333 				 attr, userspace_policy);
2334 	if (error)
2335 		return error;
2336 
2337 	if (!a[OVS_USERSPACE_ATTR_PID] ||
2338 	    !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2339 		return -EINVAL;
2340 
2341 	return 0;
2342 }
2343 
2344 static int copy_action(const struct nlattr *from,
2345 		       struct sw_flow_actions **sfa, bool log)
2346 {
2347 	int totlen = NLA_ALIGN(from->nla_len);
2348 	struct nlattr *to;
2349 
2350 	to = reserve_sfa_size(sfa, from->nla_len, log);
2351 	if (IS_ERR(to))
2352 		return PTR_ERR(to);
2353 
2354 	memcpy(to, from, totlen);
2355 	return 0;
2356 }
2357 
2358 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2359 				  const struct sw_flow_key *key,
2360 				  int depth, struct sw_flow_actions **sfa,
2361 				  __be16 eth_type, __be16 vlan_tci, bool log)
2362 {
2363 	u8 mac_proto = ovs_key_mac_proto(key);
2364 	const struct nlattr *a;
2365 	int rem, err;
2366 
2367 	if (depth >= SAMPLE_ACTION_DEPTH)
2368 		return -EOVERFLOW;
2369 
2370 	nla_for_each_nested(a, attr, rem) {
2371 		/* Expected argument lengths, (u32)-1 for variable length. */
2372 		static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2373 			[OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2374 			[OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2375 			[OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2376 			[OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2377 			[OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2378 			[OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2379 			[OVS_ACTION_ATTR_POP_VLAN] = 0,
2380 			[OVS_ACTION_ATTR_SET] = (u32)-1,
2381 			[OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2382 			[OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2383 			[OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2384 			[OVS_ACTION_ATTR_CT] = (u32)-1,
2385 			[OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
2386 			[OVS_ACTION_ATTR_PUSH_ETH] = sizeof(struct ovs_action_push_eth),
2387 			[OVS_ACTION_ATTR_POP_ETH] = 0,
2388 		};
2389 		const struct ovs_action_push_vlan *vlan;
2390 		int type = nla_type(a);
2391 		bool skip_copy;
2392 
2393 		if (type > OVS_ACTION_ATTR_MAX ||
2394 		    (action_lens[type] != nla_len(a) &&
2395 		     action_lens[type] != (u32)-1))
2396 			return -EINVAL;
2397 
2398 		skip_copy = false;
2399 		switch (type) {
2400 		case OVS_ACTION_ATTR_UNSPEC:
2401 			return -EINVAL;
2402 
2403 		case OVS_ACTION_ATTR_USERSPACE:
2404 			err = validate_userspace(a);
2405 			if (err)
2406 				return err;
2407 			break;
2408 
2409 		case OVS_ACTION_ATTR_OUTPUT:
2410 			if (nla_get_u32(a) >= DP_MAX_PORTS)
2411 				return -EINVAL;
2412 			break;
2413 
2414 		case OVS_ACTION_ATTR_TRUNC: {
2415 			const struct ovs_action_trunc *trunc = nla_data(a);
2416 
2417 			if (trunc->max_len < ETH_HLEN)
2418 				return -EINVAL;
2419 			break;
2420 		}
2421 
2422 		case OVS_ACTION_ATTR_HASH: {
2423 			const struct ovs_action_hash *act_hash = nla_data(a);
2424 
2425 			switch (act_hash->hash_alg) {
2426 			case OVS_HASH_ALG_L4:
2427 				break;
2428 			default:
2429 				return  -EINVAL;
2430 			}
2431 
2432 			break;
2433 		}
2434 
2435 		case OVS_ACTION_ATTR_POP_VLAN:
2436 			if (mac_proto != MAC_PROTO_ETHERNET)
2437 				return -EINVAL;
2438 			vlan_tci = htons(0);
2439 			break;
2440 
2441 		case OVS_ACTION_ATTR_PUSH_VLAN:
2442 			if (mac_proto != MAC_PROTO_ETHERNET)
2443 				return -EINVAL;
2444 			vlan = nla_data(a);
2445 			if (!eth_type_vlan(vlan->vlan_tpid))
2446 				return -EINVAL;
2447 			if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2448 				return -EINVAL;
2449 			vlan_tci = vlan->vlan_tci;
2450 			break;
2451 
2452 		case OVS_ACTION_ATTR_RECIRC:
2453 			break;
2454 
2455 		case OVS_ACTION_ATTR_PUSH_MPLS: {
2456 			const struct ovs_action_push_mpls *mpls = nla_data(a);
2457 
2458 			if (!eth_p_mpls(mpls->mpls_ethertype))
2459 				return -EINVAL;
2460 			/* Prohibit push MPLS other than to a white list
2461 			 * for packets that have a known tag order.
2462 			 */
2463 			if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2464 			    (eth_type != htons(ETH_P_IP) &&
2465 			     eth_type != htons(ETH_P_IPV6) &&
2466 			     eth_type != htons(ETH_P_ARP) &&
2467 			     eth_type != htons(ETH_P_RARP) &&
2468 			     !eth_p_mpls(eth_type)))
2469 				return -EINVAL;
2470 			eth_type = mpls->mpls_ethertype;
2471 			break;
2472 		}
2473 
2474 		case OVS_ACTION_ATTR_POP_MPLS:
2475 			if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2476 			    !eth_p_mpls(eth_type))
2477 				return -EINVAL;
2478 
2479 			/* Disallow subsequent L2.5+ set and mpls_pop actions
2480 			 * as there is no check here to ensure that the new
2481 			 * eth_type is valid and thus set actions could
2482 			 * write off the end of the packet or otherwise
2483 			 * corrupt it.
2484 			 *
2485 			 * Support for these actions is planned using packet
2486 			 * recirculation.
2487 			 */
2488 			eth_type = htons(0);
2489 			break;
2490 
2491 		case OVS_ACTION_ATTR_SET:
2492 			err = validate_set(a, key, sfa,
2493 					   &skip_copy, mac_proto, eth_type,
2494 					   false, log);
2495 			if (err)
2496 				return err;
2497 			break;
2498 
2499 		case OVS_ACTION_ATTR_SET_MASKED:
2500 			err = validate_set(a, key, sfa,
2501 					   &skip_copy, mac_proto, eth_type,
2502 					   true, log);
2503 			if (err)
2504 				return err;
2505 			break;
2506 
2507 		case OVS_ACTION_ATTR_SAMPLE:
2508 			err = validate_and_copy_sample(net, a, key, depth, sfa,
2509 						       eth_type, vlan_tci, log);
2510 			if (err)
2511 				return err;
2512 			skip_copy = true;
2513 			break;
2514 
2515 		case OVS_ACTION_ATTR_CT:
2516 			err = ovs_ct_copy_action(net, a, key, sfa, log);
2517 			if (err)
2518 				return err;
2519 			skip_copy = true;
2520 			break;
2521 
2522 		case OVS_ACTION_ATTR_PUSH_ETH:
2523 			/* Disallow pushing an Ethernet header if one
2524 			 * is already present */
2525 			if (mac_proto != MAC_PROTO_NONE)
2526 				return -EINVAL;
2527 			mac_proto = MAC_PROTO_NONE;
2528 			break;
2529 
2530 		case OVS_ACTION_ATTR_POP_ETH:
2531 			if (mac_proto != MAC_PROTO_ETHERNET)
2532 				return -EINVAL;
2533 			if (vlan_tci & htons(VLAN_TAG_PRESENT))
2534 				return -EINVAL;
2535 			mac_proto = MAC_PROTO_ETHERNET;
2536 			break;
2537 
2538 		default:
2539 			OVS_NLERR(log, "Unknown Action type %d", type);
2540 			return -EINVAL;
2541 		}
2542 		if (!skip_copy) {
2543 			err = copy_action(a, sfa, log);
2544 			if (err)
2545 				return err;
2546 		}
2547 	}
2548 
2549 	if (rem > 0)
2550 		return -EINVAL;
2551 
2552 	return 0;
2553 }
2554 
2555 /* 'key' must be the masked key. */
2556 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2557 			 const struct sw_flow_key *key,
2558 			 struct sw_flow_actions **sfa, bool log)
2559 {
2560 	int err;
2561 
2562 	*sfa = nla_alloc_flow_actions(nla_len(attr), log);
2563 	if (IS_ERR(*sfa))
2564 		return PTR_ERR(*sfa);
2565 
2566 	(*sfa)->orig_len = nla_len(attr);
2567 	err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
2568 				     key->eth.vlan.tci, log);
2569 	if (err)
2570 		ovs_nla_free_flow_actions(*sfa);
2571 
2572 	return err;
2573 }
2574 
2575 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2576 {
2577 	const struct nlattr *a;
2578 	struct nlattr *start;
2579 	int err = 0, rem;
2580 
2581 	start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2582 	if (!start)
2583 		return -EMSGSIZE;
2584 
2585 	nla_for_each_nested(a, attr, rem) {
2586 		int type = nla_type(a);
2587 		struct nlattr *st_sample;
2588 
2589 		switch (type) {
2590 		case OVS_SAMPLE_ATTR_PROBABILITY:
2591 			if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2592 				    sizeof(u32), nla_data(a)))
2593 				return -EMSGSIZE;
2594 			break;
2595 		case OVS_SAMPLE_ATTR_ACTIONS:
2596 			st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2597 			if (!st_sample)
2598 				return -EMSGSIZE;
2599 			err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2600 			if (err)
2601 				return err;
2602 			nla_nest_end(skb, st_sample);
2603 			break;
2604 		}
2605 	}
2606 
2607 	nla_nest_end(skb, start);
2608 	return err;
2609 }
2610 
2611 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2612 {
2613 	const struct nlattr *ovs_key = nla_data(a);
2614 	int key_type = nla_type(ovs_key);
2615 	struct nlattr *start;
2616 	int err;
2617 
2618 	switch (key_type) {
2619 	case OVS_KEY_ATTR_TUNNEL_INFO: {
2620 		struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2621 		struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2622 
2623 		start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2624 		if (!start)
2625 			return -EMSGSIZE;
2626 
2627 		err =  ip_tun_to_nlattr(skb, &tun_info->key,
2628 					ip_tunnel_info_opts(tun_info),
2629 					tun_info->options_len,
2630 					ip_tunnel_info_af(tun_info));
2631 		if (err)
2632 			return err;
2633 		nla_nest_end(skb, start);
2634 		break;
2635 	}
2636 	default:
2637 		if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2638 			return -EMSGSIZE;
2639 		break;
2640 	}
2641 
2642 	return 0;
2643 }
2644 
2645 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2646 						struct sk_buff *skb)
2647 {
2648 	const struct nlattr *ovs_key = nla_data(a);
2649 	struct nlattr *nla;
2650 	size_t key_len = nla_len(ovs_key) / 2;
2651 
2652 	/* Revert the conversion we did from a non-masked set action to
2653 	 * masked set action.
2654 	 */
2655 	nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2656 	if (!nla)
2657 		return -EMSGSIZE;
2658 
2659 	if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2660 		return -EMSGSIZE;
2661 
2662 	nla_nest_end(skb, nla);
2663 	return 0;
2664 }
2665 
2666 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2667 {
2668 	const struct nlattr *a;
2669 	int rem, err;
2670 
2671 	nla_for_each_attr(a, attr, len, rem) {
2672 		int type = nla_type(a);
2673 
2674 		switch (type) {
2675 		case OVS_ACTION_ATTR_SET:
2676 			err = set_action_to_attr(a, skb);
2677 			if (err)
2678 				return err;
2679 			break;
2680 
2681 		case OVS_ACTION_ATTR_SET_TO_MASKED:
2682 			err = masked_set_action_to_set_action_attr(a, skb);
2683 			if (err)
2684 				return err;
2685 			break;
2686 
2687 		case OVS_ACTION_ATTR_SAMPLE:
2688 			err = sample_action_to_attr(a, skb);
2689 			if (err)
2690 				return err;
2691 			break;
2692 
2693 		case OVS_ACTION_ATTR_CT:
2694 			err = ovs_ct_action_to_attr(nla_data(a), skb);
2695 			if (err)
2696 				return err;
2697 			break;
2698 
2699 		default:
2700 			if (nla_put(skb, type, nla_len(a), nla_data(a)))
2701 				return -EMSGSIZE;
2702 			break;
2703 		}
2704 	}
2705 
2706 	return 0;
2707 }
2708