1 /*
2  * Copyright (c) 2007-2017 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 #include <net/tun_proto.h>
52 #include <net/erspan.h>
53 
54 #include "flow_netlink.h"
55 
56 struct ovs_len_tbl {
57 	int len;
58 	const struct ovs_len_tbl *next;
59 };
60 
61 #define OVS_ATTR_NESTED -1
62 #define OVS_ATTR_VARIABLE -2
63 
64 static bool actions_may_change_flow(const struct nlattr *actions)
65 {
66 	struct nlattr *nla;
67 	int rem;
68 
69 	nla_for_each_nested(nla, actions, rem) {
70 		u16 action = nla_type(nla);
71 
72 		switch (action) {
73 		case OVS_ACTION_ATTR_OUTPUT:
74 		case OVS_ACTION_ATTR_RECIRC:
75 		case OVS_ACTION_ATTR_TRUNC:
76 		case OVS_ACTION_ATTR_USERSPACE:
77 			break;
78 
79 		case OVS_ACTION_ATTR_CT:
80 		case OVS_ACTION_ATTR_CT_CLEAR:
81 		case OVS_ACTION_ATTR_HASH:
82 		case OVS_ACTION_ATTR_POP_ETH:
83 		case OVS_ACTION_ATTR_POP_MPLS:
84 		case OVS_ACTION_ATTR_POP_NSH:
85 		case OVS_ACTION_ATTR_POP_VLAN:
86 		case OVS_ACTION_ATTR_PUSH_ETH:
87 		case OVS_ACTION_ATTR_PUSH_MPLS:
88 		case OVS_ACTION_ATTR_PUSH_NSH:
89 		case OVS_ACTION_ATTR_PUSH_VLAN:
90 		case OVS_ACTION_ATTR_SAMPLE:
91 		case OVS_ACTION_ATTR_SET:
92 		case OVS_ACTION_ATTR_SET_MASKED:
93 		case OVS_ACTION_ATTR_METER:
94 		default:
95 			return true;
96 		}
97 	}
98 	return false;
99 }
100 
101 static void update_range(struct sw_flow_match *match,
102 			 size_t offset, size_t size, bool is_mask)
103 {
104 	struct sw_flow_key_range *range;
105 	size_t start = rounddown(offset, sizeof(long));
106 	size_t end = roundup(offset + size, sizeof(long));
107 
108 	if (!is_mask)
109 		range = &match->range;
110 	else
111 		range = &match->mask->range;
112 
113 	if (range->start == range->end) {
114 		range->start = start;
115 		range->end = end;
116 		return;
117 	}
118 
119 	if (range->start > start)
120 		range->start = start;
121 
122 	if (range->end < end)
123 		range->end = end;
124 }
125 
126 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
127 	do { \
128 		update_range(match, offsetof(struct sw_flow_key, field),    \
129 			     sizeof((match)->key->field), is_mask);	    \
130 		if (is_mask)						    \
131 			(match)->mask->key.field = value;		    \
132 		else							    \
133 			(match)->key->field = value;		            \
134 	} while (0)
135 
136 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)	    \
137 	do {								    \
138 		update_range(match, offset, len, is_mask);		    \
139 		if (is_mask)						    \
140 			memcpy((u8 *)&(match)->mask->key + offset, value_p, \
141 			       len);					   \
142 		else							    \
143 			memcpy((u8 *)(match)->key + offset, value_p, len);  \
144 	} while (0)
145 
146 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)		      \
147 	SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
148 				  value_p, len, is_mask)
149 
150 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)		    \
151 	do {								    \
152 		update_range(match, offsetof(struct sw_flow_key, field),    \
153 			     sizeof((match)->key->field), is_mask);	    \
154 		if (is_mask)						    \
155 			memset((u8 *)&(match)->mask->key.field, value,      \
156 			       sizeof((match)->mask->key.field));	    \
157 		else							    \
158 			memset((u8 *)&(match)->key->field, value,           \
159 			       sizeof((match)->key->field));                \
160 	} while (0)
161 
162 static bool match_validate(const struct sw_flow_match *match,
163 			   u64 key_attrs, u64 mask_attrs, bool log)
164 {
165 	u64 key_expected = 0;
166 	u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
167 
168 	/* The following mask attributes allowed only if they
169 	 * pass the validation tests. */
170 	mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
171 			| (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)
172 			| (1 << OVS_KEY_ATTR_IPV6)
173 			| (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)
174 			| (1 << OVS_KEY_ATTR_TCP)
175 			| (1 << OVS_KEY_ATTR_TCP_FLAGS)
176 			| (1 << OVS_KEY_ATTR_UDP)
177 			| (1 << OVS_KEY_ATTR_SCTP)
178 			| (1 << OVS_KEY_ATTR_ICMP)
179 			| (1 << OVS_KEY_ATTR_ICMPV6)
180 			| (1 << OVS_KEY_ATTR_ARP)
181 			| (1 << OVS_KEY_ATTR_ND)
182 			| (1 << OVS_KEY_ATTR_MPLS)
183 			| (1 << OVS_KEY_ATTR_NSH));
184 
185 	/* Always allowed mask fields. */
186 	mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
187 		       | (1 << OVS_KEY_ATTR_IN_PORT)
188 		       | (1 << OVS_KEY_ATTR_ETHERTYPE));
189 
190 	/* Check key attributes. */
191 	if (match->key->eth.type == htons(ETH_P_ARP)
192 			|| match->key->eth.type == htons(ETH_P_RARP)) {
193 		key_expected |= 1 << OVS_KEY_ATTR_ARP;
194 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
195 			mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
196 	}
197 
198 	if (eth_p_mpls(match->key->eth.type)) {
199 		key_expected |= 1 << OVS_KEY_ATTR_MPLS;
200 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
201 			mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
202 	}
203 
204 	if (match->key->eth.type == htons(ETH_P_IP)) {
205 		key_expected |= 1 << OVS_KEY_ATTR_IPV4;
206 		if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
207 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
208 			mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4;
209 		}
210 
211 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
212 			if (match->key->ip.proto == IPPROTO_UDP) {
213 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
214 				if (match->mask && (match->mask->key.ip.proto == 0xff))
215 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
216 			}
217 
218 			if (match->key->ip.proto == IPPROTO_SCTP) {
219 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
220 				if (match->mask && (match->mask->key.ip.proto == 0xff))
221 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
222 			}
223 
224 			if (match->key->ip.proto == IPPROTO_TCP) {
225 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
226 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
227 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
228 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
229 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
230 				}
231 			}
232 
233 			if (match->key->ip.proto == IPPROTO_ICMP) {
234 				key_expected |= 1 << OVS_KEY_ATTR_ICMP;
235 				if (match->mask && (match->mask->key.ip.proto == 0xff))
236 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
237 			}
238 		}
239 	}
240 
241 	if (match->key->eth.type == htons(ETH_P_IPV6)) {
242 		key_expected |= 1 << OVS_KEY_ATTR_IPV6;
243 		if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
244 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
245 			mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6;
246 		}
247 
248 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
249 			if (match->key->ip.proto == IPPROTO_UDP) {
250 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
251 				if (match->mask && (match->mask->key.ip.proto == 0xff))
252 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
253 			}
254 
255 			if (match->key->ip.proto == IPPROTO_SCTP) {
256 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
257 				if (match->mask && (match->mask->key.ip.proto == 0xff))
258 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
259 			}
260 
261 			if (match->key->ip.proto == IPPROTO_TCP) {
262 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
263 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
264 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
265 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
266 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
267 				}
268 			}
269 
270 			if (match->key->ip.proto == IPPROTO_ICMPV6) {
271 				key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
272 				if (match->mask && (match->mask->key.ip.proto == 0xff))
273 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
274 
275 				if (match->key->tp.src ==
276 						htons(NDISC_NEIGHBOUR_SOLICITATION) ||
277 				    match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
278 					key_expected |= 1 << OVS_KEY_ATTR_ND;
279 					/* Original direction conntrack tuple
280 					 * uses the same space as the ND fields
281 					 * in the key, so both are not allowed
282 					 * at the same time.
283 					 */
284 					mask_allowed &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
285 					if (match->mask && (match->mask->key.tp.src == htons(0xff)))
286 						mask_allowed |= 1 << OVS_KEY_ATTR_ND;
287 				}
288 			}
289 		}
290 	}
291 
292 	if (match->key->eth.type == htons(ETH_P_NSH)) {
293 		key_expected |= 1 << OVS_KEY_ATTR_NSH;
294 		if (match->mask &&
295 		    match->mask->key.eth.type == htons(0xffff)) {
296 			mask_allowed |= 1 << OVS_KEY_ATTR_NSH;
297 		}
298 	}
299 
300 	if ((key_attrs & key_expected) != key_expected) {
301 		/* Key attributes check failed. */
302 		OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
303 			  (unsigned long long)key_attrs,
304 			  (unsigned long long)key_expected);
305 		return false;
306 	}
307 
308 	if ((mask_attrs & mask_allowed) != mask_attrs) {
309 		/* Mask attributes check failed. */
310 		OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
311 			  (unsigned long long)mask_attrs,
312 			  (unsigned long long)mask_allowed);
313 		return false;
314 	}
315 
316 	return true;
317 }
318 
319 size_t ovs_tun_key_attr_size(void)
320 {
321 	/* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
322 	 * updating this function.
323 	 */
324 	return    nla_total_size_64bit(8) /* OVS_TUNNEL_KEY_ATTR_ID */
325 		+ nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
326 		+ nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_DST */
327 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
328 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
329 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
330 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
331 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
332 		+ nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
333 		/* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS is mutually exclusive with
334 		 * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
335 		 */
336 		+ nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
337 		+ nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_DST */
338 		+ nla_total_size(4);   /* OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS */
339 }
340 
341 static size_t ovs_nsh_key_attr_size(void)
342 {
343 	/* Whenever adding new OVS_NSH_KEY_ FIELDS, we should consider
344 	 * updating this function.
345 	 */
346 	return  nla_total_size(NSH_BASE_HDR_LEN) /* OVS_NSH_KEY_ATTR_BASE */
347 		/* OVS_NSH_KEY_ATTR_MD1 and OVS_NSH_KEY_ATTR_MD2 are
348 		 * mutually exclusive, so the bigger one can cover
349 		 * the small one.
350 		 */
351 		+ nla_total_size(NSH_CTX_HDRS_MAX_LEN);
352 }
353 
354 size_t ovs_key_attr_size(void)
355 {
356 	/* Whenever adding new OVS_KEY_ FIELDS, we should consider
357 	 * updating this function.
358 	 */
359 	BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 29);
360 
361 	return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
362 		+ nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
363 		  + ovs_tun_key_attr_size()
364 		+ nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
365 		+ nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
366 		+ nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
367 		+ nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
368 		+ nla_total_size(4)   /* OVS_KEY_ATTR_CT_STATE */
369 		+ nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
370 		+ nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
371 		+ nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABELS */
372 		+ nla_total_size(40)  /* OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6 */
373 		+ nla_total_size(0)   /* OVS_KEY_ATTR_NSH */
374 		  + ovs_nsh_key_attr_size()
375 		+ nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
376 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
377 		+ nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
378 		+ nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
379 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
380 		+ nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
381 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
382 		+ nla_total_size(28); /* OVS_KEY_ATTR_ND */
383 }
384 
385 static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
386 	[OVS_VXLAN_EXT_GBP]	    = { .len = sizeof(u32) },
387 };
388 
389 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
390 	[OVS_TUNNEL_KEY_ATTR_ID]	    = { .len = sizeof(u64) },
391 	[OVS_TUNNEL_KEY_ATTR_IPV4_SRC]	    = { .len = sizeof(u32) },
392 	[OVS_TUNNEL_KEY_ATTR_IPV4_DST]	    = { .len = sizeof(u32) },
393 	[OVS_TUNNEL_KEY_ATTR_TOS]	    = { .len = 1 },
394 	[OVS_TUNNEL_KEY_ATTR_TTL]	    = { .len = 1 },
395 	[OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
396 	[OVS_TUNNEL_KEY_ATTR_CSUM]	    = { .len = 0 },
397 	[OVS_TUNNEL_KEY_ATTR_TP_SRC]	    = { .len = sizeof(u16) },
398 	[OVS_TUNNEL_KEY_ATTR_TP_DST]	    = { .len = sizeof(u16) },
399 	[OVS_TUNNEL_KEY_ATTR_OAM]	    = { .len = 0 },
400 	[OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_VARIABLE },
401 	[OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED,
402 						.next = ovs_vxlan_ext_key_lens },
403 	[OVS_TUNNEL_KEY_ATTR_IPV6_SRC]      = { .len = sizeof(struct in6_addr) },
404 	[OVS_TUNNEL_KEY_ATTR_IPV6_DST]      = { .len = sizeof(struct in6_addr) },
405 	[OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS]   = { .len = sizeof(u32) },
406 };
407 
408 static const struct ovs_len_tbl
409 ovs_nsh_key_attr_lens[OVS_NSH_KEY_ATTR_MAX + 1] = {
410 	[OVS_NSH_KEY_ATTR_BASE] = { .len = sizeof(struct ovs_nsh_key_base) },
411 	[OVS_NSH_KEY_ATTR_MD1]  = { .len = sizeof(struct ovs_nsh_key_md1) },
412 	[OVS_NSH_KEY_ATTR_MD2]  = { .len = OVS_ATTR_VARIABLE },
413 };
414 
415 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
416 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
417 	[OVS_KEY_ATTR_ENCAP]	 = { .len = OVS_ATTR_NESTED },
418 	[OVS_KEY_ATTR_PRIORITY]	 = { .len = sizeof(u32) },
419 	[OVS_KEY_ATTR_IN_PORT]	 = { .len = sizeof(u32) },
420 	[OVS_KEY_ATTR_SKB_MARK]	 = { .len = sizeof(u32) },
421 	[OVS_KEY_ATTR_ETHERNET]	 = { .len = sizeof(struct ovs_key_ethernet) },
422 	[OVS_KEY_ATTR_VLAN]	 = { .len = sizeof(__be16) },
423 	[OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
424 	[OVS_KEY_ATTR_IPV4]	 = { .len = sizeof(struct ovs_key_ipv4) },
425 	[OVS_KEY_ATTR_IPV6]	 = { .len = sizeof(struct ovs_key_ipv6) },
426 	[OVS_KEY_ATTR_TCP]	 = { .len = sizeof(struct ovs_key_tcp) },
427 	[OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
428 	[OVS_KEY_ATTR_UDP]	 = { .len = sizeof(struct ovs_key_udp) },
429 	[OVS_KEY_ATTR_SCTP]	 = { .len = sizeof(struct ovs_key_sctp) },
430 	[OVS_KEY_ATTR_ICMP]	 = { .len = sizeof(struct ovs_key_icmp) },
431 	[OVS_KEY_ATTR_ICMPV6]	 = { .len = sizeof(struct ovs_key_icmpv6) },
432 	[OVS_KEY_ATTR_ARP]	 = { .len = sizeof(struct ovs_key_arp) },
433 	[OVS_KEY_ATTR_ND]	 = { .len = sizeof(struct ovs_key_nd) },
434 	[OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
435 	[OVS_KEY_ATTR_DP_HASH]	 = { .len = sizeof(u32) },
436 	[OVS_KEY_ATTR_TUNNEL]	 = { .len = OVS_ATTR_NESTED,
437 				     .next = ovs_tunnel_key_lens, },
438 	[OVS_KEY_ATTR_MPLS]	 = { .len = sizeof(struct ovs_key_mpls) },
439 	[OVS_KEY_ATTR_CT_STATE]	 = { .len = sizeof(u32) },
440 	[OVS_KEY_ATTR_CT_ZONE]	 = { .len = sizeof(u16) },
441 	[OVS_KEY_ATTR_CT_MARK]	 = { .len = sizeof(u32) },
442 	[OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
443 	[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4] = {
444 		.len = sizeof(struct ovs_key_ct_tuple_ipv4) },
445 	[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6] = {
446 		.len = sizeof(struct ovs_key_ct_tuple_ipv6) },
447 	[OVS_KEY_ATTR_NSH]       = { .len = OVS_ATTR_NESTED,
448 				     .next = ovs_nsh_key_attr_lens, },
449 };
450 
451 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
452 {
453 	return expected_len == attr_len ||
454 	       expected_len == OVS_ATTR_NESTED ||
455 	       expected_len == OVS_ATTR_VARIABLE;
456 }
457 
458 static bool is_all_zero(const u8 *fp, size_t size)
459 {
460 	int i;
461 
462 	if (!fp)
463 		return false;
464 
465 	for (i = 0; i < size; i++)
466 		if (fp[i])
467 			return false;
468 
469 	return true;
470 }
471 
472 static int __parse_flow_nlattrs(const struct nlattr *attr,
473 				const struct nlattr *a[],
474 				u64 *attrsp, bool log, bool nz)
475 {
476 	const struct nlattr *nla;
477 	u64 attrs;
478 	int rem;
479 
480 	attrs = *attrsp;
481 	nla_for_each_nested(nla, attr, rem) {
482 		u16 type = nla_type(nla);
483 		int expected_len;
484 
485 		if (type > OVS_KEY_ATTR_MAX) {
486 			OVS_NLERR(log, "Key type %d is out of range max %d",
487 				  type, OVS_KEY_ATTR_MAX);
488 			return -EINVAL;
489 		}
490 
491 		if (attrs & (1 << type)) {
492 			OVS_NLERR(log, "Duplicate key (type %d).", type);
493 			return -EINVAL;
494 		}
495 
496 		expected_len = ovs_key_lens[type].len;
497 		if (!check_attr_len(nla_len(nla), expected_len)) {
498 			OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
499 				  type, nla_len(nla), expected_len);
500 			return -EINVAL;
501 		}
502 
503 		if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
504 			attrs |= 1 << type;
505 			a[type] = nla;
506 		}
507 	}
508 	if (rem) {
509 		OVS_NLERR(log, "Message has %d unknown bytes.", rem);
510 		return -EINVAL;
511 	}
512 
513 	*attrsp = attrs;
514 	return 0;
515 }
516 
517 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
518 				   const struct nlattr *a[], u64 *attrsp,
519 				   bool log)
520 {
521 	return __parse_flow_nlattrs(attr, a, attrsp, log, true);
522 }
523 
524 int parse_flow_nlattrs(const struct nlattr *attr, const struct nlattr *a[],
525 		       u64 *attrsp, bool log)
526 {
527 	return __parse_flow_nlattrs(attr, a, attrsp, log, false);
528 }
529 
530 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
531 				     struct sw_flow_match *match, bool is_mask,
532 				     bool log)
533 {
534 	unsigned long opt_key_offset;
535 
536 	if (nla_len(a) > sizeof(match->key->tun_opts)) {
537 		OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
538 			  nla_len(a), sizeof(match->key->tun_opts));
539 		return -EINVAL;
540 	}
541 
542 	if (nla_len(a) % 4 != 0) {
543 		OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
544 			  nla_len(a));
545 		return -EINVAL;
546 	}
547 
548 	/* We need to record the length of the options passed
549 	 * down, otherwise packets with the same format but
550 	 * additional options will be silently matched.
551 	 */
552 	if (!is_mask) {
553 		SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
554 				false);
555 	} else {
556 		/* This is somewhat unusual because it looks at
557 		 * both the key and mask while parsing the
558 		 * attributes (and by extension assumes the key
559 		 * is parsed first). Normally, we would verify
560 		 * that each is the correct length and that the
561 		 * attributes line up in the validate function.
562 		 * However, that is difficult because this is
563 		 * variable length and we won't have the
564 		 * information later.
565 		 */
566 		if (match->key->tun_opts_len != nla_len(a)) {
567 			OVS_NLERR(log, "Geneve option len %d != mask len %d",
568 				  match->key->tun_opts_len, nla_len(a));
569 			return -EINVAL;
570 		}
571 
572 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
573 	}
574 
575 	opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
576 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
577 				  nla_len(a), is_mask);
578 	return 0;
579 }
580 
581 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
582 				     struct sw_flow_match *match, bool is_mask,
583 				     bool log)
584 {
585 	struct nlattr *a;
586 	int rem;
587 	unsigned long opt_key_offset;
588 	struct vxlan_metadata opts;
589 
590 	BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
591 
592 	memset(&opts, 0, sizeof(opts));
593 	nla_for_each_nested(a, attr, rem) {
594 		int type = nla_type(a);
595 
596 		if (type > OVS_VXLAN_EXT_MAX) {
597 			OVS_NLERR(log, "VXLAN extension %d out of range max %d",
598 				  type, OVS_VXLAN_EXT_MAX);
599 			return -EINVAL;
600 		}
601 
602 		if (!check_attr_len(nla_len(a),
603 				    ovs_vxlan_ext_key_lens[type].len)) {
604 			OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
605 				  type, nla_len(a),
606 				  ovs_vxlan_ext_key_lens[type].len);
607 			return -EINVAL;
608 		}
609 
610 		switch (type) {
611 		case OVS_VXLAN_EXT_GBP:
612 			opts.gbp = nla_get_u32(a);
613 			break;
614 		default:
615 			OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
616 				  type);
617 			return -EINVAL;
618 		}
619 	}
620 	if (rem) {
621 		OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
622 			  rem);
623 		return -EINVAL;
624 	}
625 
626 	if (!is_mask)
627 		SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
628 	else
629 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
630 
631 	opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
632 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
633 				  is_mask);
634 	return 0;
635 }
636 
637 static int erspan_tun_opt_from_nlattr(const struct nlattr *attr,
638 				      struct sw_flow_match *match, bool is_mask,
639 				      bool log)
640 {
641 	unsigned long opt_key_offset;
642 	struct erspan_metadata opts;
643 
644 	BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
645 
646 	memset(&opts, 0, sizeof(opts));
647 	opts.index = nla_get_be32(attr);
648 
649 	/* Index has only 20-bit */
650 	if (ntohl(opts.index) & ~INDEX_MASK) {
651 		OVS_NLERR(log, "ERSPAN index number %x too large.",
652 			  ntohl(opts.index));
653 		return -EINVAL;
654 	}
655 
656 	SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), is_mask);
657 	opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
658 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
659 				  is_mask);
660 
661 	return 0;
662 }
663 
664 static int ip_tun_from_nlattr(const struct nlattr *attr,
665 			      struct sw_flow_match *match, bool is_mask,
666 			      bool log)
667 {
668 	bool ttl = false, ipv4 = false, ipv6 = false;
669 	__be16 tun_flags = 0;
670 	int opts_type = 0;
671 	struct nlattr *a;
672 	int rem;
673 
674 	nla_for_each_nested(a, attr, rem) {
675 		int type = nla_type(a);
676 		int err;
677 
678 		if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
679 			OVS_NLERR(log, "Tunnel attr %d out of range max %d",
680 				  type, OVS_TUNNEL_KEY_ATTR_MAX);
681 			return -EINVAL;
682 		}
683 
684 		if (!check_attr_len(nla_len(a),
685 				    ovs_tunnel_key_lens[type].len)) {
686 			OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
687 				  type, nla_len(a), ovs_tunnel_key_lens[type].len);
688 			return -EINVAL;
689 		}
690 
691 		switch (type) {
692 		case OVS_TUNNEL_KEY_ATTR_ID:
693 			SW_FLOW_KEY_PUT(match, tun_key.tun_id,
694 					nla_get_be64(a), is_mask);
695 			tun_flags |= TUNNEL_KEY;
696 			break;
697 		case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
698 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
699 					nla_get_in_addr(a), is_mask);
700 			ipv4 = true;
701 			break;
702 		case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
703 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
704 					nla_get_in_addr(a), is_mask);
705 			ipv4 = true;
706 			break;
707 		case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
708 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.src,
709 					nla_get_in6_addr(a), is_mask);
710 			ipv6 = true;
711 			break;
712 		case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
713 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
714 					nla_get_in6_addr(a), is_mask);
715 			ipv6 = true;
716 			break;
717 		case OVS_TUNNEL_KEY_ATTR_TOS:
718 			SW_FLOW_KEY_PUT(match, tun_key.tos,
719 					nla_get_u8(a), is_mask);
720 			break;
721 		case OVS_TUNNEL_KEY_ATTR_TTL:
722 			SW_FLOW_KEY_PUT(match, tun_key.ttl,
723 					nla_get_u8(a), is_mask);
724 			ttl = true;
725 			break;
726 		case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
727 			tun_flags |= TUNNEL_DONT_FRAGMENT;
728 			break;
729 		case OVS_TUNNEL_KEY_ATTR_CSUM:
730 			tun_flags |= TUNNEL_CSUM;
731 			break;
732 		case OVS_TUNNEL_KEY_ATTR_TP_SRC:
733 			SW_FLOW_KEY_PUT(match, tun_key.tp_src,
734 					nla_get_be16(a), is_mask);
735 			break;
736 		case OVS_TUNNEL_KEY_ATTR_TP_DST:
737 			SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
738 					nla_get_be16(a), is_mask);
739 			break;
740 		case OVS_TUNNEL_KEY_ATTR_OAM:
741 			tun_flags |= TUNNEL_OAM;
742 			break;
743 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
744 			if (opts_type) {
745 				OVS_NLERR(log, "Multiple metadata blocks provided");
746 				return -EINVAL;
747 			}
748 
749 			err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
750 			if (err)
751 				return err;
752 
753 			tun_flags |= TUNNEL_GENEVE_OPT;
754 			opts_type = type;
755 			break;
756 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
757 			if (opts_type) {
758 				OVS_NLERR(log, "Multiple metadata blocks provided");
759 				return -EINVAL;
760 			}
761 
762 			err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
763 			if (err)
764 				return err;
765 
766 			tun_flags |= TUNNEL_VXLAN_OPT;
767 			opts_type = type;
768 			break;
769 		case OVS_TUNNEL_KEY_ATTR_PAD:
770 			break;
771 		case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
772 			if (opts_type) {
773 				OVS_NLERR(log, "Multiple metadata blocks provided");
774 				return -EINVAL;
775 			}
776 
777 			err = erspan_tun_opt_from_nlattr(a, match, is_mask, log);
778 			if (err)
779 				return err;
780 
781 			tun_flags |= TUNNEL_ERSPAN_OPT;
782 			opts_type = type;
783 			break;
784 		default:
785 			OVS_NLERR(log, "Unknown IP tunnel attribute %d",
786 				  type);
787 			return -EINVAL;
788 		}
789 	}
790 
791 	SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
792 	if (is_mask)
793 		SW_FLOW_KEY_MEMSET_FIELD(match, tun_proto, 0xff, true);
794 	else
795 		SW_FLOW_KEY_PUT(match, tun_proto, ipv6 ? AF_INET6 : AF_INET,
796 				false);
797 
798 	if (rem > 0) {
799 		OVS_NLERR(log, "IP tunnel attribute has %d unknown bytes.",
800 			  rem);
801 		return -EINVAL;
802 	}
803 
804 	if (ipv4 && ipv6) {
805 		OVS_NLERR(log, "Mixed IPv4 and IPv6 tunnel attributes");
806 		return -EINVAL;
807 	}
808 
809 	if (!is_mask) {
810 		if (!ipv4 && !ipv6) {
811 			OVS_NLERR(log, "IP tunnel dst address not specified");
812 			return -EINVAL;
813 		}
814 		if (ipv4 && !match->key->tun_key.u.ipv4.dst) {
815 			OVS_NLERR(log, "IPv4 tunnel dst address is zero");
816 			return -EINVAL;
817 		}
818 		if (ipv6 && ipv6_addr_any(&match->key->tun_key.u.ipv6.dst)) {
819 			OVS_NLERR(log, "IPv6 tunnel dst address is zero");
820 			return -EINVAL;
821 		}
822 
823 		if (!ttl) {
824 			OVS_NLERR(log, "IP tunnel TTL not specified.");
825 			return -EINVAL;
826 		}
827 	}
828 
829 	return opts_type;
830 }
831 
832 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
833 			       const void *tun_opts, int swkey_tun_opts_len)
834 {
835 	const struct vxlan_metadata *opts = tun_opts;
836 	struct nlattr *nla;
837 
838 	nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
839 	if (!nla)
840 		return -EMSGSIZE;
841 
842 	if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
843 		return -EMSGSIZE;
844 
845 	nla_nest_end(skb, nla);
846 	return 0;
847 }
848 
849 static int __ip_tun_to_nlattr(struct sk_buff *skb,
850 			      const struct ip_tunnel_key *output,
851 			      const void *tun_opts, int swkey_tun_opts_len,
852 			      unsigned short tun_proto)
853 {
854 	if (output->tun_flags & TUNNEL_KEY &&
855 	    nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id,
856 			 OVS_TUNNEL_KEY_ATTR_PAD))
857 		return -EMSGSIZE;
858 	switch (tun_proto) {
859 	case AF_INET:
860 		if (output->u.ipv4.src &&
861 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
862 				    output->u.ipv4.src))
863 			return -EMSGSIZE;
864 		if (output->u.ipv4.dst &&
865 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
866 				    output->u.ipv4.dst))
867 			return -EMSGSIZE;
868 		break;
869 	case AF_INET6:
870 		if (!ipv6_addr_any(&output->u.ipv6.src) &&
871 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
872 				     &output->u.ipv6.src))
873 			return -EMSGSIZE;
874 		if (!ipv6_addr_any(&output->u.ipv6.dst) &&
875 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
876 				     &output->u.ipv6.dst))
877 			return -EMSGSIZE;
878 		break;
879 	}
880 	if (output->tos &&
881 	    nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
882 		return -EMSGSIZE;
883 	if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
884 		return -EMSGSIZE;
885 	if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
886 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
887 		return -EMSGSIZE;
888 	if ((output->tun_flags & TUNNEL_CSUM) &&
889 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
890 		return -EMSGSIZE;
891 	if (output->tp_src &&
892 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
893 		return -EMSGSIZE;
894 	if (output->tp_dst &&
895 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
896 		return -EMSGSIZE;
897 	if ((output->tun_flags & TUNNEL_OAM) &&
898 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
899 		return -EMSGSIZE;
900 	if (swkey_tun_opts_len) {
901 		if (output->tun_flags & TUNNEL_GENEVE_OPT &&
902 		    nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
903 			    swkey_tun_opts_len, tun_opts))
904 			return -EMSGSIZE;
905 		else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
906 			 vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
907 			return -EMSGSIZE;
908 		else if (output->tun_flags & TUNNEL_ERSPAN_OPT &&
909 			 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS,
910 				      ((struct erspan_metadata *)tun_opts)->index))
911 			return -EMSGSIZE;
912 	}
913 
914 	return 0;
915 }
916 
917 static int ip_tun_to_nlattr(struct sk_buff *skb,
918 			    const struct ip_tunnel_key *output,
919 			    const void *tun_opts, int swkey_tun_opts_len,
920 			    unsigned short tun_proto)
921 {
922 	struct nlattr *nla;
923 	int err;
924 
925 	nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
926 	if (!nla)
927 		return -EMSGSIZE;
928 
929 	err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
930 				 tun_proto);
931 	if (err)
932 		return err;
933 
934 	nla_nest_end(skb, nla);
935 	return 0;
936 }
937 
938 int ovs_nla_put_tunnel_info(struct sk_buff *skb,
939 			    struct ip_tunnel_info *tun_info)
940 {
941 	return __ip_tun_to_nlattr(skb, &tun_info->key,
942 				  ip_tunnel_info_opts(tun_info),
943 				  tun_info->options_len,
944 				  ip_tunnel_info_af(tun_info));
945 }
946 
947 static int encode_vlan_from_nlattrs(struct sw_flow_match *match,
948 				    const struct nlattr *a[],
949 				    bool is_mask, bool inner)
950 {
951 	__be16 tci = 0;
952 	__be16 tpid = 0;
953 
954 	if (a[OVS_KEY_ATTR_VLAN])
955 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
956 
957 	if (a[OVS_KEY_ATTR_ETHERTYPE])
958 		tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
959 
960 	if (likely(!inner)) {
961 		SW_FLOW_KEY_PUT(match, eth.vlan.tpid, tpid, is_mask);
962 		SW_FLOW_KEY_PUT(match, eth.vlan.tci, tci, is_mask);
963 	} else {
964 		SW_FLOW_KEY_PUT(match, eth.cvlan.tpid, tpid, is_mask);
965 		SW_FLOW_KEY_PUT(match, eth.cvlan.tci, tci, is_mask);
966 	}
967 	return 0;
968 }
969 
970 static int validate_vlan_from_nlattrs(const struct sw_flow_match *match,
971 				      u64 key_attrs, bool inner,
972 				      const struct nlattr **a, bool log)
973 {
974 	__be16 tci = 0;
975 
976 	if (!((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
977 	      (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
978 	       eth_type_vlan(nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE])))) {
979 		/* Not a VLAN. */
980 		return 0;
981 	}
982 
983 	if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
984 	      (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
985 		OVS_NLERR(log, "Invalid %s frame", (inner) ? "C-VLAN" : "VLAN");
986 		return -EINVAL;
987 	}
988 
989 	if (a[OVS_KEY_ATTR_VLAN])
990 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
991 
992 	if (!(tci & htons(VLAN_TAG_PRESENT))) {
993 		if (tci) {
994 			OVS_NLERR(log, "%s TCI does not have VLAN_TAG_PRESENT bit set.",
995 				  (inner) ? "C-VLAN" : "VLAN");
996 			return -EINVAL;
997 		} else if (nla_len(a[OVS_KEY_ATTR_ENCAP])) {
998 			/* Corner case for truncated VLAN header. */
999 			OVS_NLERR(log, "Truncated %s header has non-zero encap attribute.",
1000 				  (inner) ? "C-VLAN" : "VLAN");
1001 			return -EINVAL;
1002 		}
1003 	}
1004 
1005 	return 1;
1006 }
1007 
1008 static int validate_vlan_mask_from_nlattrs(const struct sw_flow_match *match,
1009 					   u64 key_attrs, bool inner,
1010 					   const struct nlattr **a, bool log)
1011 {
1012 	__be16 tci = 0;
1013 	__be16 tpid = 0;
1014 	bool encap_valid = !!(match->key->eth.vlan.tci &
1015 			      htons(VLAN_TAG_PRESENT));
1016 	bool i_encap_valid = !!(match->key->eth.cvlan.tci &
1017 				htons(VLAN_TAG_PRESENT));
1018 
1019 	if (!(key_attrs & (1 << OVS_KEY_ATTR_ENCAP))) {
1020 		/* Not a VLAN. */
1021 		return 0;
1022 	}
1023 
1024 	if ((!inner && !encap_valid) || (inner && !i_encap_valid)) {
1025 		OVS_NLERR(log, "Encap mask attribute is set for non-%s frame.",
1026 			  (inner) ? "C-VLAN" : "VLAN");
1027 		return -EINVAL;
1028 	}
1029 
1030 	if (a[OVS_KEY_ATTR_VLAN])
1031 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1032 
1033 	if (a[OVS_KEY_ATTR_ETHERTYPE])
1034 		tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1035 
1036 	if (tpid != htons(0xffff)) {
1037 		OVS_NLERR(log, "Must have an exact match on %s TPID (mask=%x).",
1038 			  (inner) ? "C-VLAN" : "VLAN", ntohs(tpid));
1039 		return -EINVAL;
1040 	}
1041 	if (!(tci & htons(VLAN_TAG_PRESENT))) {
1042 		OVS_NLERR(log, "%s TCI mask does not have exact match for VLAN_TAG_PRESENT bit.",
1043 			  (inner) ? "C-VLAN" : "VLAN");
1044 		return -EINVAL;
1045 	}
1046 
1047 	return 1;
1048 }
1049 
1050 static int __parse_vlan_from_nlattrs(struct sw_flow_match *match,
1051 				     u64 *key_attrs, bool inner,
1052 				     const struct nlattr **a, bool is_mask,
1053 				     bool log)
1054 {
1055 	int err;
1056 	const struct nlattr *encap;
1057 
1058 	if (!is_mask)
1059 		err = validate_vlan_from_nlattrs(match, *key_attrs, inner,
1060 						 a, log);
1061 	else
1062 		err = validate_vlan_mask_from_nlattrs(match, *key_attrs, inner,
1063 						      a, log);
1064 	if (err <= 0)
1065 		return err;
1066 
1067 	err = encode_vlan_from_nlattrs(match, a, is_mask, inner);
1068 	if (err)
1069 		return err;
1070 
1071 	*key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1072 	*key_attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
1073 	*key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1074 
1075 	encap = a[OVS_KEY_ATTR_ENCAP];
1076 
1077 	if (!is_mask)
1078 		err = parse_flow_nlattrs(encap, a, key_attrs, log);
1079 	else
1080 		err = parse_flow_mask_nlattrs(encap, a, key_attrs, log);
1081 
1082 	return err;
1083 }
1084 
1085 static int parse_vlan_from_nlattrs(struct sw_flow_match *match,
1086 				   u64 *key_attrs, const struct nlattr **a,
1087 				   bool is_mask, bool log)
1088 {
1089 	int err;
1090 	bool encap_valid = false;
1091 
1092 	err = __parse_vlan_from_nlattrs(match, key_attrs, false, a,
1093 					is_mask, log);
1094 	if (err)
1095 		return err;
1096 
1097 	encap_valid = !!(match->key->eth.vlan.tci & htons(VLAN_TAG_PRESENT));
1098 	if (encap_valid) {
1099 		err = __parse_vlan_from_nlattrs(match, key_attrs, true, a,
1100 						is_mask, log);
1101 		if (err)
1102 			return err;
1103 	}
1104 
1105 	return 0;
1106 }
1107 
1108 static int parse_eth_type_from_nlattrs(struct sw_flow_match *match,
1109 				       u64 *attrs, const struct nlattr **a,
1110 				       bool is_mask, bool log)
1111 {
1112 	__be16 eth_type;
1113 
1114 	eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1115 	if (is_mask) {
1116 		/* Always exact match EtherType. */
1117 		eth_type = htons(0xffff);
1118 	} else if (!eth_proto_is_802_3(eth_type)) {
1119 		OVS_NLERR(log, "EtherType %x is less than min %x",
1120 				ntohs(eth_type), ETH_P_802_3_MIN);
1121 		return -EINVAL;
1122 	}
1123 
1124 	SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1125 	*attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1126 	return 0;
1127 }
1128 
1129 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
1130 				 u64 *attrs, const struct nlattr **a,
1131 				 bool is_mask, bool log)
1132 {
1133 	u8 mac_proto = MAC_PROTO_ETHERNET;
1134 
1135 	if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
1136 		u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
1137 
1138 		SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
1139 		*attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
1140 	}
1141 
1142 	if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
1143 		u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
1144 
1145 		SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
1146 		*attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
1147 	}
1148 
1149 	if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1150 		SW_FLOW_KEY_PUT(match, phy.priority,
1151 			  nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1152 		*attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1153 	}
1154 
1155 	if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1156 		u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1157 
1158 		if (is_mask) {
1159 			in_port = 0xffffffff; /* Always exact match in_port. */
1160 		} else if (in_port >= DP_MAX_PORTS) {
1161 			OVS_NLERR(log, "Port %d exceeds max allowable %d",
1162 				  in_port, DP_MAX_PORTS);
1163 			return -EINVAL;
1164 		}
1165 
1166 		SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1167 		*attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1168 	} else if (!is_mask) {
1169 		SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1170 	}
1171 
1172 	if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1173 		uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1174 
1175 		SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1176 		*attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1177 	}
1178 	if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1179 		if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1180 				       is_mask, log) < 0)
1181 			return -EINVAL;
1182 		*attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1183 	}
1184 
1185 	if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
1186 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
1187 		u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
1188 
1189 		if (ct_state & ~CT_SUPPORTED_MASK) {
1190 			OVS_NLERR(log, "ct_state flags %08x unsupported",
1191 				  ct_state);
1192 			return -EINVAL;
1193 		}
1194 
1195 		SW_FLOW_KEY_PUT(match, ct_state, ct_state, is_mask);
1196 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
1197 	}
1198 	if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
1199 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
1200 		u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
1201 
1202 		SW_FLOW_KEY_PUT(match, ct_zone, ct_zone, is_mask);
1203 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
1204 	}
1205 	if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
1206 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
1207 		u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
1208 
1209 		SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
1210 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
1211 	}
1212 	if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
1213 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
1214 		const struct ovs_key_ct_labels *cl;
1215 
1216 		cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
1217 		SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
1218 				   sizeof(*cl), is_mask);
1219 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
1220 	}
1221 	if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)) {
1222 		const struct ovs_key_ct_tuple_ipv4 *ct;
1223 
1224 		ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4]);
1225 
1226 		SW_FLOW_KEY_PUT(match, ipv4.ct_orig.src, ct->ipv4_src, is_mask);
1227 		SW_FLOW_KEY_PUT(match, ipv4.ct_orig.dst, ct->ipv4_dst, is_mask);
1228 		SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1229 		SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1230 		SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv4_proto, is_mask);
1231 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4);
1232 	}
1233 	if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)) {
1234 		const struct ovs_key_ct_tuple_ipv6 *ct;
1235 
1236 		ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6]);
1237 
1238 		SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.src, &ct->ipv6_src,
1239 				   sizeof(match->key->ipv6.ct_orig.src),
1240 				   is_mask);
1241 		SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.dst, &ct->ipv6_dst,
1242 				   sizeof(match->key->ipv6.ct_orig.dst),
1243 				   is_mask);
1244 		SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1245 		SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1246 		SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv6_proto, is_mask);
1247 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
1248 	}
1249 
1250 	/* For layer 3 packets the Ethernet type is provided
1251 	 * and treated as metadata but no MAC addresses are provided.
1252 	 */
1253 	if (!(*attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1254 	    (*attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)))
1255 		mac_proto = MAC_PROTO_NONE;
1256 
1257 	/* Always exact match mac_proto */
1258 	SW_FLOW_KEY_PUT(match, mac_proto, is_mask ? 0xff : mac_proto, is_mask);
1259 
1260 	if (mac_proto == MAC_PROTO_NONE)
1261 		return parse_eth_type_from_nlattrs(match, attrs, a, is_mask,
1262 						   log);
1263 
1264 	return 0;
1265 }
1266 
1267 int nsh_hdr_from_nlattr(const struct nlattr *attr,
1268 			struct nshhdr *nh, size_t size)
1269 {
1270 	struct nlattr *a;
1271 	int rem;
1272 	u8 flags = 0;
1273 	u8 ttl = 0;
1274 	int mdlen = 0;
1275 
1276 	/* validate_nsh has check this, so we needn't do duplicate check here
1277 	 */
1278 	if (size < NSH_BASE_HDR_LEN)
1279 		return -ENOBUFS;
1280 
1281 	nla_for_each_nested(a, attr, rem) {
1282 		int type = nla_type(a);
1283 
1284 		switch (type) {
1285 		case OVS_NSH_KEY_ATTR_BASE: {
1286 			const struct ovs_nsh_key_base *base = nla_data(a);
1287 
1288 			flags = base->flags;
1289 			ttl = base->ttl;
1290 			nh->np = base->np;
1291 			nh->mdtype = base->mdtype;
1292 			nh->path_hdr = base->path_hdr;
1293 			break;
1294 		}
1295 		case OVS_NSH_KEY_ATTR_MD1:
1296 			mdlen = nla_len(a);
1297 			if (mdlen > size - NSH_BASE_HDR_LEN)
1298 				return -ENOBUFS;
1299 			memcpy(&nh->md1, nla_data(a), mdlen);
1300 			break;
1301 
1302 		case OVS_NSH_KEY_ATTR_MD2:
1303 			mdlen = nla_len(a);
1304 			if (mdlen > size - NSH_BASE_HDR_LEN)
1305 				return -ENOBUFS;
1306 			memcpy(&nh->md2, nla_data(a), mdlen);
1307 			break;
1308 
1309 		default:
1310 			return -EINVAL;
1311 		}
1312 	}
1313 
1314 	/* nsh header length  = NSH_BASE_HDR_LEN + mdlen */
1315 	nh->ver_flags_ttl_len = 0;
1316 	nsh_set_flags_ttl_len(nh, flags, ttl, NSH_BASE_HDR_LEN + mdlen);
1317 
1318 	return 0;
1319 }
1320 
1321 int nsh_key_from_nlattr(const struct nlattr *attr,
1322 			struct ovs_key_nsh *nsh, struct ovs_key_nsh *nsh_mask)
1323 {
1324 	struct nlattr *a;
1325 	int rem;
1326 
1327 	/* validate_nsh has check this, so we needn't do duplicate check here
1328 	 */
1329 	nla_for_each_nested(a, attr, rem) {
1330 		int type = nla_type(a);
1331 
1332 		switch (type) {
1333 		case OVS_NSH_KEY_ATTR_BASE: {
1334 			const struct ovs_nsh_key_base *base = nla_data(a);
1335 			const struct ovs_nsh_key_base *base_mask = base + 1;
1336 
1337 			nsh->base = *base;
1338 			nsh_mask->base = *base_mask;
1339 			break;
1340 		}
1341 		case OVS_NSH_KEY_ATTR_MD1: {
1342 			const struct ovs_nsh_key_md1 *md1 = nla_data(a);
1343 			const struct ovs_nsh_key_md1 *md1_mask = md1 + 1;
1344 
1345 			memcpy(nsh->context, md1->context, sizeof(*md1));
1346 			memcpy(nsh_mask->context, md1_mask->context,
1347 			       sizeof(*md1_mask));
1348 			break;
1349 		}
1350 		case OVS_NSH_KEY_ATTR_MD2:
1351 			/* Not supported yet */
1352 			return -ENOTSUPP;
1353 		default:
1354 			return -EINVAL;
1355 		}
1356 	}
1357 
1358 	return 0;
1359 }
1360 
1361 static int nsh_key_put_from_nlattr(const struct nlattr *attr,
1362 				   struct sw_flow_match *match, bool is_mask,
1363 				   bool is_push_nsh, bool log)
1364 {
1365 	struct nlattr *a;
1366 	int rem;
1367 	bool has_base = false;
1368 	bool has_md1 = false;
1369 	bool has_md2 = false;
1370 	u8 mdtype = 0;
1371 	int mdlen = 0;
1372 
1373 	if (WARN_ON(is_push_nsh && is_mask))
1374 		return -EINVAL;
1375 
1376 	nla_for_each_nested(a, attr, rem) {
1377 		int type = nla_type(a);
1378 		int i;
1379 
1380 		if (type > OVS_NSH_KEY_ATTR_MAX) {
1381 			OVS_NLERR(log, "nsh attr %d is out of range max %d",
1382 				  type, OVS_NSH_KEY_ATTR_MAX);
1383 			return -EINVAL;
1384 		}
1385 
1386 		if (!check_attr_len(nla_len(a),
1387 				    ovs_nsh_key_attr_lens[type].len)) {
1388 			OVS_NLERR(
1389 			    log,
1390 			    "nsh attr %d has unexpected len %d expected %d",
1391 			    type,
1392 			    nla_len(a),
1393 			    ovs_nsh_key_attr_lens[type].len
1394 			);
1395 			return -EINVAL;
1396 		}
1397 
1398 		switch (type) {
1399 		case OVS_NSH_KEY_ATTR_BASE: {
1400 			const struct ovs_nsh_key_base *base = nla_data(a);
1401 
1402 			has_base = true;
1403 			mdtype = base->mdtype;
1404 			SW_FLOW_KEY_PUT(match, nsh.base.flags,
1405 					base->flags, is_mask);
1406 			SW_FLOW_KEY_PUT(match, nsh.base.ttl,
1407 					base->ttl, is_mask);
1408 			SW_FLOW_KEY_PUT(match, nsh.base.mdtype,
1409 					base->mdtype, is_mask);
1410 			SW_FLOW_KEY_PUT(match, nsh.base.np,
1411 					base->np, is_mask);
1412 			SW_FLOW_KEY_PUT(match, nsh.base.path_hdr,
1413 					base->path_hdr, is_mask);
1414 			break;
1415 		}
1416 		case OVS_NSH_KEY_ATTR_MD1: {
1417 			const struct ovs_nsh_key_md1 *md1 = nla_data(a);
1418 
1419 			has_md1 = true;
1420 			for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++)
1421 				SW_FLOW_KEY_PUT(match, nsh.context[i],
1422 						md1->context[i], is_mask);
1423 			break;
1424 		}
1425 		case OVS_NSH_KEY_ATTR_MD2:
1426 			if (!is_push_nsh) /* Not supported MD type 2 yet */
1427 				return -ENOTSUPP;
1428 
1429 			has_md2 = true;
1430 			mdlen = nla_len(a);
1431 			if (mdlen > NSH_CTX_HDRS_MAX_LEN || mdlen <= 0) {
1432 				OVS_NLERR(
1433 				    log,
1434 				    "Invalid MD length %d for MD type %d",
1435 				    mdlen,
1436 				    mdtype
1437 				);
1438 				return -EINVAL;
1439 			}
1440 			break;
1441 		default:
1442 			OVS_NLERR(log, "Unknown nsh attribute %d",
1443 				  type);
1444 			return -EINVAL;
1445 		}
1446 	}
1447 
1448 	if (rem > 0) {
1449 		OVS_NLERR(log, "nsh attribute has %d unknown bytes.", rem);
1450 		return -EINVAL;
1451 	}
1452 
1453 	if (has_md1 && has_md2) {
1454 		OVS_NLERR(
1455 		    1,
1456 		    "invalid nsh attribute: md1 and md2 are exclusive."
1457 		);
1458 		return -EINVAL;
1459 	}
1460 
1461 	if (!is_mask) {
1462 		if ((has_md1 && mdtype != NSH_M_TYPE1) ||
1463 		    (has_md2 && mdtype != NSH_M_TYPE2)) {
1464 			OVS_NLERR(1, "nsh attribute has unmatched MD type %d.",
1465 				  mdtype);
1466 			return -EINVAL;
1467 		}
1468 
1469 		if (is_push_nsh &&
1470 		    (!has_base || (!has_md1 && !has_md2))) {
1471 			OVS_NLERR(
1472 			    1,
1473 			    "push_nsh: missing base or metadata attributes"
1474 			);
1475 			return -EINVAL;
1476 		}
1477 	}
1478 
1479 	return 0;
1480 }
1481 
1482 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
1483 				u64 attrs, const struct nlattr **a,
1484 				bool is_mask, bool log)
1485 {
1486 	int err;
1487 
1488 	err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
1489 	if (err)
1490 		return err;
1491 
1492 	if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
1493 		const struct ovs_key_ethernet *eth_key;
1494 
1495 		eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1496 		SW_FLOW_KEY_MEMCPY(match, eth.src,
1497 				eth_key->eth_src, ETH_ALEN, is_mask);
1498 		SW_FLOW_KEY_MEMCPY(match, eth.dst,
1499 				eth_key->eth_dst, ETH_ALEN, is_mask);
1500 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1501 
1502 		if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
1503 			/* VLAN attribute is always parsed before getting here since it
1504 			 * may occur multiple times.
1505 			 */
1506 			OVS_NLERR(log, "VLAN attribute unexpected.");
1507 			return -EINVAL;
1508 		}
1509 
1510 		if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1511 			err = parse_eth_type_from_nlattrs(match, &attrs, a, is_mask,
1512 							  log);
1513 			if (err)
1514 				return err;
1515 		} else if (!is_mask) {
1516 			SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1517 		}
1518 	} else if (!match->key->eth.type) {
1519 		OVS_NLERR(log, "Either Ethernet header or EtherType is required.");
1520 		return -EINVAL;
1521 	}
1522 
1523 	if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1524 		const struct ovs_key_ipv4 *ipv4_key;
1525 
1526 		ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1527 		if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1528 			OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
1529 				  ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1530 			return -EINVAL;
1531 		}
1532 		SW_FLOW_KEY_PUT(match, ip.proto,
1533 				ipv4_key->ipv4_proto, is_mask);
1534 		SW_FLOW_KEY_PUT(match, ip.tos,
1535 				ipv4_key->ipv4_tos, is_mask);
1536 		SW_FLOW_KEY_PUT(match, ip.ttl,
1537 				ipv4_key->ipv4_ttl, is_mask);
1538 		SW_FLOW_KEY_PUT(match, ip.frag,
1539 				ipv4_key->ipv4_frag, is_mask);
1540 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1541 				ipv4_key->ipv4_src, is_mask);
1542 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1543 				ipv4_key->ipv4_dst, is_mask);
1544 		attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1545 	}
1546 
1547 	if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
1548 		const struct ovs_key_ipv6 *ipv6_key;
1549 
1550 		ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1551 		if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1552 			OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
1553 				  ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1554 			return -EINVAL;
1555 		}
1556 
1557 		if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
1558 			OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x)",
1559 				  ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
1560 			return -EINVAL;
1561 		}
1562 
1563 		SW_FLOW_KEY_PUT(match, ipv6.label,
1564 				ipv6_key->ipv6_label, is_mask);
1565 		SW_FLOW_KEY_PUT(match, ip.proto,
1566 				ipv6_key->ipv6_proto, is_mask);
1567 		SW_FLOW_KEY_PUT(match, ip.tos,
1568 				ipv6_key->ipv6_tclass, is_mask);
1569 		SW_FLOW_KEY_PUT(match, ip.ttl,
1570 				ipv6_key->ipv6_hlimit, is_mask);
1571 		SW_FLOW_KEY_PUT(match, ip.frag,
1572 				ipv6_key->ipv6_frag, is_mask);
1573 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1574 				ipv6_key->ipv6_src,
1575 				sizeof(match->key->ipv6.addr.src),
1576 				is_mask);
1577 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1578 				ipv6_key->ipv6_dst,
1579 				sizeof(match->key->ipv6.addr.dst),
1580 				is_mask);
1581 
1582 		attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1583 	}
1584 
1585 	if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1586 		const struct ovs_key_arp *arp_key;
1587 
1588 		arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1589 		if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1590 			OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1591 				  arp_key->arp_op);
1592 			return -EINVAL;
1593 		}
1594 
1595 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1596 				arp_key->arp_sip, is_mask);
1597 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1598 			arp_key->arp_tip, is_mask);
1599 		SW_FLOW_KEY_PUT(match, ip.proto,
1600 				ntohs(arp_key->arp_op), is_mask);
1601 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1602 				arp_key->arp_sha, ETH_ALEN, is_mask);
1603 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1604 				arp_key->arp_tha, ETH_ALEN, is_mask);
1605 
1606 		attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1607 	}
1608 
1609 	if (attrs & (1 << OVS_KEY_ATTR_NSH)) {
1610 		if (nsh_key_put_from_nlattr(a[OVS_KEY_ATTR_NSH], match,
1611 					    is_mask, false, log) < 0)
1612 			return -EINVAL;
1613 		attrs &= ~(1 << OVS_KEY_ATTR_NSH);
1614 	}
1615 
1616 	if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1617 		const struct ovs_key_mpls *mpls_key;
1618 
1619 		mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1620 		SW_FLOW_KEY_PUT(match, mpls.top_lse,
1621 				mpls_key->mpls_lse, is_mask);
1622 
1623 		attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1624 	 }
1625 
1626 	if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1627 		const struct ovs_key_tcp *tcp_key;
1628 
1629 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1630 		SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1631 		SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1632 		attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1633 	}
1634 
1635 	if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1636 		SW_FLOW_KEY_PUT(match, tp.flags,
1637 				nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1638 				is_mask);
1639 		attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1640 	}
1641 
1642 	if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1643 		const struct ovs_key_udp *udp_key;
1644 
1645 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1646 		SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1647 		SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1648 		attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1649 	}
1650 
1651 	if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1652 		const struct ovs_key_sctp *sctp_key;
1653 
1654 		sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1655 		SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1656 		SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1657 		attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1658 	}
1659 
1660 	if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1661 		const struct ovs_key_icmp *icmp_key;
1662 
1663 		icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1664 		SW_FLOW_KEY_PUT(match, tp.src,
1665 				htons(icmp_key->icmp_type), is_mask);
1666 		SW_FLOW_KEY_PUT(match, tp.dst,
1667 				htons(icmp_key->icmp_code), is_mask);
1668 		attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1669 	}
1670 
1671 	if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1672 		const struct ovs_key_icmpv6 *icmpv6_key;
1673 
1674 		icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1675 		SW_FLOW_KEY_PUT(match, tp.src,
1676 				htons(icmpv6_key->icmpv6_type), is_mask);
1677 		SW_FLOW_KEY_PUT(match, tp.dst,
1678 				htons(icmpv6_key->icmpv6_code), is_mask);
1679 		attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1680 	}
1681 
1682 	if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1683 		const struct ovs_key_nd *nd_key;
1684 
1685 		nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1686 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1687 			nd_key->nd_target,
1688 			sizeof(match->key->ipv6.nd.target),
1689 			is_mask);
1690 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1691 			nd_key->nd_sll, ETH_ALEN, is_mask);
1692 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1693 				nd_key->nd_tll, ETH_ALEN, is_mask);
1694 		attrs &= ~(1 << OVS_KEY_ATTR_ND);
1695 	}
1696 
1697 	if (attrs != 0) {
1698 		OVS_NLERR(log, "Unknown key attributes %llx",
1699 			  (unsigned long long)attrs);
1700 		return -EINVAL;
1701 	}
1702 
1703 	return 0;
1704 }
1705 
1706 static void nlattr_set(struct nlattr *attr, u8 val,
1707 		       const struct ovs_len_tbl *tbl)
1708 {
1709 	struct nlattr *nla;
1710 	int rem;
1711 
1712 	/* The nlattr stream should already have been validated */
1713 	nla_for_each_nested(nla, attr, rem) {
1714 		if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
1715 			if (tbl[nla_type(nla)].next)
1716 				tbl = tbl[nla_type(nla)].next;
1717 			nlattr_set(nla, val, tbl);
1718 		} else {
1719 			memset(nla_data(nla), val, nla_len(nla));
1720 		}
1721 
1722 		if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1723 			*(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1724 	}
1725 }
1726 
1727 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1728 {
1729 	nlattr_set(attr, val, ovs_key_lens);
1730 }
1731 
1732 /**
1733  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1734  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1735  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1736  * does not include any don't care bit.
1737  * @net: Used to determine per-namespace field support.
1738  * @match: receives the extracted flow match information.
1739  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1740  * sequence. The fields should of the packet that triggered the creation
1741  * of this flow.
1742  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1743  * attribute specifies the mask field of the wildcarded flow.
1744  * @log: Boolean to allow kernel error logging.  Normally true, but when
1745  * probing for feature compatibility this should be passed in as false to
1746  * suppress unnecessary error logging.
1747  */
1748 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1749 		      const struct nlattr *nla_key,
1750 		      const struct nlattr *nla_mask,
1751 		      bool log)
1752 {
1753 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1754 	struct nlattr *newmask = NULL;
1755 	u64 key_attrs = 0;
1756 	u64 mask_attrs = 0;
1757 	int err;
1758 
1759 	err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1760 	if (err)
1761 		return err;
1762 
1763 	err = parse_vlan_from_nlattrs(match, &key_attrs, a, false, log);
1764 	if (err)
1765 		return err;
1766 
1767 	err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1768 	if (err)
1769 		return err;
1770 
1771 	if (match->mask) {
1772 		if (!nla_mask) {
1773 			/* Create an exact match mask. We need to set to 0xff
1774 			 * all the 'match->mask' fields that have been touched
1775 			 * in 'match->key'. We cannot simply memset
1776 			 * 'match->mask', because padding bytes and fields not
1777 			 * specified in 'match->key' should be left to 0.
1778 			 * Instead, we use a stream of netlink attributes,
1779 			 * copied from 'key' and set to 0xff.
1780 			 * ovs_key_from_nlattrs() will take care of filling
1781 			 * 'match->mask' appropriately.
1782 			 */
1783 			newmask = kmemdup(nla_key,
1784 					  nla_total_size(nla_len(nla_key)),
1785 					  GFP_KERNEL);
1786 			if (!newmask)
1787 				return -ENOMEM;
1788 
1789 			mask_set_nlattr(newmask, 0xff);
1790 
1791 			/* The userspace does not send tunnel attributes that
1792 			 * are 0, but we should not wildcard them nonetheless.
1793 			 */
1794 			if (match->key->tun_proto)
1795 				SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1796 							 0xff, true);
1797 
1798 			nla_mask = newmask;
1799 		}
1800 
1801 		err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1802 		if (err)
1803 			goto free_newmask;
1804 
1805 		/* Always match on tci. */
1806 		SW_FLOW_KEY_PUT(match, eth.vlan.tci, htons(0xffff), true);
1807 		SW_FLOW_KEY_PUT(match, eth.cvlan.tci, htons(0xffff), true);
1808 
1809 		err = parse_vlan_from_nlattrs(match, &mask_attrs, a, true, log);
1810 		if (err)
1811 			goto free_newmask;
1812 
1813 		err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1814 					   log);
1815 		if (err)
1816 			goto free_newmask;
1817 	}
1818 
1819 	if (!match_validate(match, key_attrs, mask_attrs, log))
1820 		err = -EINVAL;
1821 
1822 free_newmask:
1823 	kfree(newmask);
1824 	return err;
1825 }
1826 
1827 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1828 {
1829 	size_t len;
1830 
1831 	if (!attr)
1832 		return 0;
1833 
1834 	len = nla_len(attr);
1835 	if (len < 1 || len > MAX_UFID_LENGTH) {
1836 		OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1837 			  nla_len(attr), MAX_UFID_LENGTH);
1838 		return 0;
1839 	}
1840 
1841 	return len;
1842 }
1843 
1844 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1845  * or false otherwise.
1846  */
1847 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1848 		      bool log)
1849 {
1850 	sfid->ufid_len = get_ufid_len(attr, log);
1851 	if (sfid->ufid_len)
1852 		memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1853 
1854 	return sfid->ufid_len;
1855 }
1856 
1857 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1858 			   const struct sw_flow_key *key, bool log)
1859 {
1860 	struct sw_flow_key *new_key;
1861 
1862 	if (ovs_nla_get_ufid(sfid, ufid, log))
1863 		return 0;
1864 
1865 	/* If UFID was not provided, use unmasked key. */
1866 	new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1867 	if (!new_key)
1868 		return -ENOMEM;
1869 	memcpy(new_key, key, sizeof(*key));
1870 	sfid->unmasked_key = new_key;
1871 
1872 	return 0;
1873 }
1874 
1875 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1876 {
1877 	return attr ? nla_get_u32(attr) : 0;
1878 }
1879 
1880 /**
1881  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1882  * @net: Network namespace.
1883  * @key: Receives extracted in_port, priority, tun_key, skb_mark and conntrack
1884  * metadata.
1885  * @a: Array of netlink attributes holding parsed %OVS_KEY_ATTR_* Netlink
1886  * attributes.
1887  * @attrs: Bit mask for the netlink attributes included in @a.
1888  * @log: Boolean to allow kernel error logging.  Normally true, but when
1889  * probing for feature compatibility this should be passed in as false to
1890  * suppress unnecessary error logging.
1891  *
1892  * This parses a series of Netlink attributes that form a flow key, which must
1893  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1894  * get the metadata, that is, the parts of the flow key that cannot be
1895  * extracted from the packet itself.
1896  *
1897  * This must be called before the packet key fields are filled in 'key'.
1898  */
1899 
1900 int ovs_nla_get_flow_metadata(struct net *net,
1901 			      const struct nlattr *a[OVS_KEY_ATTR_MAX + 1],
1902 			      u64 attrs, struct sw_flow_key *key, bool log)
1903 {
1904 	struct sw_flow_match match;
1905 
1906 	memset(&match, 0, sizeof(match));
1907 	match.key = key;
1908 
1909 	key->ct_state = 0;
1910 	key->ct_zone = 0;
1911 	key->ct_orig_proto = 0;
1912 	memset(&key->ct, 0, sizeof(key->ct));
1913 	memset(&key->ipv4.ct_orig, 0, sizeof(key->ipv4.ct_orig));
1914 	memset(&key->ipv6.ct_orig, 0, sizeof(key->ipv6.ct_orig));
1915 
1916 	key->phy.in_port = DP_MAX_PORTS;
1917 
1918 	return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1919 }
1920 
1921 static int ovs_nla_put_vlan(struct sk_buff *skb, const struct vlan_head *vh,
1922 			    bool is_mask)
1923 {
1924 	__be16 eth_type = !is_mask ? vh->tpid : htons(0xffff);
1925 
1926 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1927 	    nla_put_be16(skb, OVS_KEY_ATTR_VLAN, vh->tci))
1928 		return -EMSGSIZE;
1929 	return 0;
1930 }
1931 
1932 static int nsh_key_to_nlattr(const struct ovs_key_nsh *nsh, bool is_mask,
1933 			     struct sk_buff *skb)
1934 {
1935 	struct nlattr *start;
1936 
1937 	start = nla_nest_start(skb, OVS_KEY_ATTR_NSH);
1938 	if (!start)
1939 		return -EMSGSIZE;
1940 
1941 	if (nla_put(skb, OVS_NSH_KEY_ATTR_BASE, sizeof(nsh->base), &nsh->base))
1942 		goto nla_put_failure;
1943 
1944 	if (is_mask || nsh->base.mdtype == NSH_M_TYPE1) {
1945 		if (nla_put(skb, OVS_NSH_KEY_ATTR_MD1,
1946 			    sizeof(nsh->context), nsh->context))
1947 			goto nla_put_failure;
1948 	}
1949 
1950 	/* Don't support MD type 2 yet */
1951 
1952 	nla_nest_end(skb, start);
1953 
1954 	return 0;
1955 
1956 nla_put_failure:
1957 	return -EMSGSIZE;
1958 }
1959 
1960 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1961 			     const struct sw_flow_key *output, bool is_mask,
1962 			     struct sk_buff *skb)
1963 {
1964 	struct ovs_key_ethernet *eth_key;
1965 	struct nlattr *nla;
1966 	struct nlattr *encap = NULL;
1967 	struct nlattr *in_encap = NULL;
1968 
1969 	if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1970 		goto nla_put_failure;
1971 
1972 	if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1973 		goto nla_put_failure;
1974 
1975 	if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1976 		goto nla_put_failure;
1977 
1978 	if ((swkey->tun_proto || is_mask)) {
1979 		const void *opts = NULL;
1980 
1981 		if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1982 			opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1983 
1984 		if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
1985 				     swkey->tun_opts_len, swkey->tun_proto))
1986 			goto nla_put_failure;
1987 	}
1988 
1989 	if (swkey->phy.in_port == DP_MAX_PORTS) {
1990 		if (is_mask && (output->phy.in_port == 0xffff))
1991 			if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1992 				goto nla_put_failure;
1993 	} else {
1994 		u16 upper_u16;
1995 		upper_u16 = !is_mask ? 0 : 0xffff;
1996 
1997 		if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1998 				(upper_u16 << 16) | output->phy.in_port))
1999 			goto nla_put_failure;
2000 	}
2001 
2002 	if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
2003 		goto nla_put_failure;
2004 
2005 	if (ovs_ct_put_key(swkey, output, skb))
2006 		goto nla_put_failure;
2007 
2008 	if (ovs_key_mac_proto(swkey) == MAC_PROTO_ETHERNET) {
2009 		nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
2010 		if (!nla)
2011 			goto nla_put_failure;
2012 
2013 		eth_key = nla_data(nla);
2014 		ether_addr_copy(eth_key->eth_src, output->eth.src);
2015 		ether_addr_copy(eth_key->eth_dst, output->eth.dst);
2016 
2017 		if (swkey->eth.vlan.tci || eth_type_vlan(swkey->eth.type)) {
2018 			if (ovs_nla_put_vlan(skb, &output->eth.vlan, is_mask))
2019 				goto nla_put_failure;
2020 			encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
2021 			if (!swkey->eth.vlan.tci)
2022 				goto unencap;
2023 
2024 			if (swkey->eth.cvlan.tci || eth_type_vlan(swkey->eth.type)) {
2025 				if (ovs_nla_put_vlan(skb, &output->eth.cvlan, is_mask))
2026 					goto nla_put_failure;
2027 				in_encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
2028 				if (!swkey->eth.cvlan.tci)
2029 					goto unencap;
2030 			}
2031 		}
2032 
2033 		if (swkey->eth.type == htons(ETH_P_802_2)) {
2034 			/*
2035 			* Ethertype 802.2 is represented in the netlink with omitted
2036 			* OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
2037 			* 0xffff in the mask attribute.  Ethertype can also
2038 			* be wildcarded.
2039 			*/
2040 			if (is_mask && output->eth.type)
2041 				if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
2042 							output->eth.type))
2043 					goto nla_put_failure;
2044 			goto unencap;
2045 		}
2046 	}
2047 
2048 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
2049 		goto nla_put_failure;
2050 
2051 	if (eth_type_vlan(swkey->eth.type)) {
2052 		/* There are 3 VLAN tags, we don't know anything about the rest
2053 		 * of the packet, so truncate here.
2054 		 */
2055 		WARN_ON_ONCE(!(encap && in_encap));
2056 		goto unencap;
2057 	}
2058 
2059 	if (swkey->eth.type == htons(ETH_P_IP)) {
2060 		struct ovs_key_ipv4 *ipv4_key;
2061 
2062 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
2063 		if (!nla)
2064 			goto nla_put_failure;
2065 		ipv4_key = nla_data(nla);
2066 		ipv4_key->ipv4_src = output->ipv4.addr.src;
2067 		ipv4_key->ipv4_dst = output->ipv4.addr.dst;
2068 		ipv4_key->ipv4_proto = output->ip.proto;
2069 		ipv4_key->ipv4_tos = output->ip.tos;
2070 		ipv4_key->ipv4_ttl = output->ip.ttl;
2071 		ipv4_key->ipv4_frag = output->ip.frag;
2072 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
2073 		struct ovs_key_ipv6 *ipv6_key;
2074 
2075 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
2076 		if (!nla)
2077 			goto nla_put_failure;
2078 		ipv6_key = nla_data(nla);
2079 		memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
2080 				sizeof(ipv6_key->ipv6_src));
2081 		memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
2082 				sizeof(ipv6_key->ipv6_dst));
2083 		ipv6_key->ipv6_label = output->ipv6.label;
2084 		ipv6_key->ipv6_proto = output->ip.proto;
2085 		ipv6_key->ipv6_tclass = output->ip.tos;
2086 		ipv6_key->ipv6_hlimit = output->ip.ttl;
2087 		ipv6_key->ipv6_frag = output->ip.frag;
2088 	} else if (swkey->eth.type == htons(ETH_P_NSH)) {
2089 		if (nsh_key_to_nlattr(&output->nsh, is_mask, skb))
2090 			goto nla_put_failure;
2091 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
2092 		   swkey->eth.type == htons(ETH_P_RARP)) {
2093 		struct ovs_key_arp *arp_key;
2094 
2095 		nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
2096 		if (!nla)
2097 			goto nla_put_failure;
2098 		arp_key = nla_data(nla);
2099 		memset(arp_key, 0, sizeof(struct ovs_key_arp));
2100 		arp_key->arp_sip = output->ipv4.addr.src;
2101 		arp_key->arp_tip = output->ipv4.addr.dst;
2102 		arp_key->arp_op = htons(output->ip.proto);
2103 		ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
2104 		ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
2105 	} else if (eth_p_mpls(swkey->eth.type)) {
2106 		struct ovs_key_mpls *mpls_key;
2107 
2108 		nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
2109 		if (!nla)
2110 			goto nla_put_failure;
2111 		mpls_key = nla_data(nla);
2112 		mpls_key->mpls_lse = output->mpls.top_lse;
2113 	}
2114 
2115 	if ((swkey->eth.type == htons(ETH_P_IP) ||
2116 	     swkey->eth.type == htons(ETH_P_IPV6)) &&
2117 	     swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
2118 
2119 		if (swkey->ip.proto == IPPROTO_TCP) {
2120 			struct ovs_key_tcp *tcp_key;
2121 
2122 			nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
2123 			if (!nla)
2124 				goto nla_put_failure;
2125 			tcp_key = nla_data(nla);
2126 			tcp_key->tcp_src = output->tp.src;
2127 			tcp_key->tcp_dst = output->tp.dst;
2128 			if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
2129 					 output->tp.flags))
2130 				goto nla_put_failure;
2131 		} else if (swkey->ip.proto == IPPROTO_UDP) {
2132 			struct ovs_key_udp *udp_key;
2133 
2134 			nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
2135 			if (!nla)
2136 				goto nla_put_failure;
2137 			udp_key = nla_data(nla);
2138 			udp_key->udp_src = output->tp.src;
2139 			udp_key->udp_dst = output->tp.dst;
2140 		} else if (swkey->ip.proto == IPPROTO_SCTP) {
2141 			struct ovs_key_sctp *sctp_key;
2142 
2143 			nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
2144 			if (!nla)
2145 				goto nla_put_failure;
2146 			sctp_key = nla_data(nla);
2147 			sctp_key->sctp_src = output->tp.src;
2148 			sctp_key->sctp_dst = output->tp.dst;
2149 		} else if (swkey->eth.type == htons(ETH_P_IP) &&
2150 			   swkey->ip.proto == IPPROTO_ICMP) {
2151 			struct ovs_key_icmp *icmp_key;
2152 
2153 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
2154 			if (!nla)
2155 				goto nla_put_failure;
2156 			icmp_key = nla_data(nla);
2157 			icmp_key->icmp_type = ntohs(output->tp.src);
2158 			icmp_key->icmp_code = ntohs(output->tp.dst);
2159 		} else if (swkey->eth.type == htons(ETH_P_IPV6) &&
2160 			   swkey->ip.proto == IPPROTO_ICMPV6) {
2161 			struct ovs_key_icmpv6 *icmpv6_key;
2162 
2163 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
2164 						sizeof(*icmpv6_key));
2165 			if (!nla)
2166 				goto nla_put_failure;
2167 			icmpv6_key = nla_data(nla);
2168 			icmpv6_key->icmpv6_type = ntohs(output->tp.src);
2169 			icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
2170 
2171 			if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
2172 			    icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
2173 				struct ovs_key_nd *nd_key;
2174 
2175 				nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
2176 				if (!nla)
2177 					goto nla_put_failure;
2178 				nd_key = nla_data(nla);
2179 				memcpy(nd_key->nd_target, &output->ipv6.nd.target,
2180 							sizeof(nd_key->nd_target));
2181 				ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
2182 				ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
2183 			}
2184 		}
2185 	}
2186 
2187 unencap:
2188 	if (in_encap)
2189 		nla_nest_end(skb, in_encap);
2190 	if (encap)
2191 		nla_nest_end(skb, encap);
2192 
2193 	return 0;
2194 
2195 nla_put_failure:
2196 	return -EMSGSIZE;
2197 }
2198 
2199 int ovs_nla_put_key(const struct sw_flow_key *swkey,
2200 		    const struct sw_flow_key *output, int attr, bool is_mask,
2201 		    struct sk_buff *skb)
2202 {
2203 	int err;
2204 	struct nlattr *nla;
2205 
2206 	nla = nla_nest_start(skb, attr);
2207 	if (!nla)
2208 		return -EMSGSIZE;
2209 	err = __ovs_nla_put_key(swkey, output, is_mask, skb);
2210 	if (err)
2211 		return err;
2212 	nla_nest_end(skb, nla);
2213 
2214 	return 0;
2215 }
2216 
2217 /* Called with ovs_mutex or RCU read lock. */
2218 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
2219 {
2220 	if (ovs_identifier_is_ufid(&flow->id))
2221 		return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
2222 			       flow->id.ufid);
2223 
2224 	return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
2225 			       OVS_FLOW_ATTR_KEY, false, skb);
2226 }
2227 
2228 /* Called with ovs_mutex or RCU read lock. */
2229 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
2230 {
2231 	return ovs_nla_put_key(&flow->key, &flow->key,
2232 				OVS_FLOW_ATTR_KEY, false, skb);
2233 }
2234 
2235 /* Called with ovs_mutex or RCU read lock. */
2236 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
2237 {
2238 	return ovs_nla_put_key(&flow->key, &flow->mask->key,
2239 				OVS_FLOW_ATTR_MASK, true, skb);
2240 }
2241 
2242 #define MAX_ACTIONS_BUFSIZE	(32 * 1024)
2243 
2244 static struct sw_flow_actions *nla_alloc_flow_actions(int size)
2245 {
2246 	struct sw_flow_actions *sfa;
2247 
2248 	WARN_ON_ONCE(size > MAX_ACTIONS_BUFSIZE);
2249 
2250 	sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
2251 	if (!sfa)
2252 		return ERR_PTR(-ENOMEM);
2253 
2254 	sfa->actions_len = 0;
2255 	return sfa;
2256 }
2257 
2258 static void ovs_nla_free_set_action(const struct nlattr *a)
2259 {
2260 	const struct nlattr *ovs_key = nla_data(a);
2261 	struct ovs_tunnel_info *ovs_tun;
2262 
2263 	switch (nla_type(ovs_key)) {
2264 	case OVS_KEY_ATTR_TUNNEL_INFO:
2265 		ovs_tun = nla_data(ovs_key);
2266 		dst_release((struct dst_entry *)ovs_tun->tun_dst);
2267 		break;
2268 	}
2269 }
2270 
2271 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
2272 {
2273 	const struct nlattr *a;
2274 	int rem;
2275 
2276 	if (!sf_acts)
2277 		return;
2278 
2279 	nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
2280 		switch (nla_type(a)) {
2281 		case OVS_ACTION_ATTR_SET:
2282 			ovs_nla_free_set_action(a);
2283 			break;
2284 		case OVS_ACTION_ATTR_CT:
2285 			ovs_ct_free_action(a);
2286 			break;
2287 		}
2288 	}
2289 
2290 	kfree(sf_acts);
2291 }
2292 
2293 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
2294 {
2295 	ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
2296 }
2297 
2298 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
2299  * The caller must hold rcu_read_lock for this to be sensible. */
2300 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
2301 {
2302 	call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
2303 }
2304 
2305 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
2306 				       int attr_len, bool log)
2307 {
2308 
2309 	struct sw_flow_actions *acts;
2310 	int new_acts_size;
2311 	int req_size = NLA_ALIGN(attr_len);
2312 	int next_offset = offsetof(struct sw_flow_actions, actions) +
2313 					(*sfa)->actions_len;
2314 
2315 	if (req_size <= (ksize(*sfa) - next_offset))
2316 		goto out;
2317 
2318 	new_acts_size = ksize(*sfa) * 2;
2319 
2320 	if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
2321 		if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size) {
2322 			OVS_NLERR(log, "Flow action size exceeds max %u",
2323 				  MAX_ACTIONS_BUFSIZE);
2324 			return ERR_PTR(-EMSGSIZE);
2325 		}
2326 		new_acts_size = MAX_ACTIONS_BUFSIZE;
2327 	}
2328 
2329 	acts = nla_alloc_flow_actions(new_acts_size);
2330 	if (IS_ERR(acts))
2331 		return (void *)acts;
2332 
2333 	memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
2334 	acts->actions_len = (*sfa)->actions_len;
2335 	acts->orig_len = (*sfa)->orig_len;
2336 	kfree(*sfa);
2337 	*sfa = acts;
2338 
2339 out:
2340 	(*sfa)->actions_len += req_size;
2341 	return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
2342 }
2343 
2344 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
2345 				   int attrtype, void *data, int len, bool log)
2346 {
2347 	struct nlattr *a;
2348 
2349 	a = reserve_sfa_size(sfa, nla_attr_size(len), log);
2350 	if (IS_ERR(a))
2351 		return a;
2352 
2353 	a->nla_type = attrtype;
2354 	a->nla_len = nla_attr_size(len);
2355 
2356 	if (data)
2357 		memcpy(nla_data(a), data, len);
2358 	memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
2359 
2360 	return a;
2361 }
2362 
2363 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
2364 		       int len, bool log)
2365 {
2366 	struct nlattr *a;
2367 
2368 	a = __add_action(sfa, attrtype, data, len, log);
2369 
2370 	return PTR_ERR_OR_ZERO(a);
2371 }
2372 
2373 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
2374 					  int attrtype, bool log)
2375 {
2376 	int used = (*sfa)->actions_len;
2377 	int err;
2378 
2379 	err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
2380 	if (err)
2381 		return err;
2382 
2383 	return used;
2384 }
2385 
2386 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
2387 					 int st_offset)
2388 {
2389 	struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
2390 							       st_offset);
2391 
2392 	a->nla_len = sfa->actions_len - st_offset;
2393 }
2394 
2395 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2396 				  const struct sw_flow_key *key,
2397 				  struct sw_flow_actions **sfa,
2398 				  __be16 eth_type, __be16 vlan_tci, bool log);
2399 
2400 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
2401 				    const struct sw_flow_key *key,
2402 				    struct sw_flow_actions **sfa,
2403 				    __be16 eth_type, __be16 vlan_tci,
2404 				    bool log, bool last)
2405 {
2406 	const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
2407 	const struct nlattr *probability, *actions;
2408 	const struct nlattr *a;
2409 	int rem, start, err;
2410 	struct sample_arg arg;
2411 
2412 	memset(attrs, 0, sizeof(attrs));
2413 	nla_for_each_nested(a, attr, rem) {
2414 		int type = nla_type(a);
2415 		if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
2416 			return -EINVAL;
2417 		attrs[type] = a;
2418 	}
2419 	if (rem)
2420 		return -EINVAL;
2421 
2422 	probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
2423 	if (!probability || nla_len(probability) != sizeof(u32))
2424 		return -EINVAL;
2425 
2426 	actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
2427 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
2428 		return -EINVAL;
2429 
2430 	/* validation done, copy sample action. */
2431 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
2432 	if (start < 0)
2433 		return start;
2434 
2435 	/* When both skb and flow may be changed, put the sample
2436 	 * into a deferred fifo. On the other hand, if only skb
2437 	 * may be modified, the actions can be executed in place.
2438 	 *
2439 	 * Do this analysis at the flow installation time.
2440 	 * Set 'clone_action->exec' to true if the actions can be
2441 	 * executed without being deferred.
2442 	 *
2443 	 * If the sample is the last action, it can always be excuted
2444 	 * rather than deferred.
2445 	 */
2446 	arg.exec = last || !actions_may_change_flow(actions);
2447 	arg.probability = nla_get_u32(probability);
2448 
2449 	err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_ARG, &arg, sizeof(arg),
2450 				 log);
2451 	if (err)
2452 		return err;
2453 
2454 	err = __ovs_nla_copy_actions(net, actions, key, sfa,
2455 				     eth_type, vlan_tci, log);
2456 
2457 	if (err)
2458 		return err;
2459 
2460 	add_nested_action_end(*sfa, start);
2461 
2462 	return 0;
2463 }
2464 
2465 void ovs_match_init(struct sw_flow_match *match,
2466 		    struct sw_flow_key *key,
2467 		    bool reset_key,
2468 		    struct sw_flow_mask *mask)
2469 {
2470 	memset(match, 0, sizeof(*match));
2471 	match->key = key;
2472 	match->mask = mask;
2473 
2474 	if (reset_key)
2475 		memset(key, 0, sizeof(*key));
2476 
2477 	if (mask) {
2478 		memset(&mask->key, 0, sizeof(mask->key));
2479 		mask->range.start = mask->range.end = 0;
2480 	}
2481 }
2482 
2483 static int validate_geneve_opts(struct sw_flow_key *key)
2484 {
2485 	struct geneve_opt *option;
2486 	int opts_len = key->tun_opts_len;
2487 	bool crit_opt = false;
2488 
2489 	option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
2490 	while (opts_len > 0) {
2491 		int len;
2492 
2493 		if (opts_len < sizeof(*option))
2494 			return -EINVAL;
2495 
2496 		len = sizeof(*option) + option->length * 4;
2497 		if (len > opts_len)
2498 			return -EINVAL;
2499 
2500 		crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
2501 
2502 		option = (struct geneve_opt *)((u8 *)option + len);
2503 		opts_len -= len;
2504 	};
2505 
2506 	key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
2507 
2508 	return 0;
2509 }
2510 
2511 static int validate_and_copy_set_tun(const struct nlattr *attr,
2512 				     struct sw_flow_actions **sfa, bool log)
2513 {
2514 	struct sw_flow_match match;
2515 	struct sw_flow_key key;
2516 	struct metadata_dst *tun_dst;
2517 	struct ip_tunnel_info *tun_info;
2518 	struct ovs_tunnel_info *ovs_tun;
2519 	struct nlattr *a;
2520 	int err = 0, start, opts_type;
2521 
2522 	ovs_match_init(&match, &key, true, NULL);
2523 	opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
2524 	if (opts_type < 0)
2525 		return opts_type;
2526 
2527 	if (key.tun_opts_len) {
2528 		switch (opts_type) {
2529 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
2530 			err = validate_geneve_opts(&key);
2531 			if (err < 0)
2532 				return err;
2533 			break;
2534 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
2535 			break;
2536 		case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
2537 			break;
2538 		}
2539 	};
2540 
2541 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
2542 	if (start < 0)
2543 		return start;
2544 
2545 	tun_dst = metadata_dst_alloc(key.tun_opts_len, METADATA_IP_TUNNEL,
2546 				     GFP_KERNEL);
2547 
2548 	if (!tun_dst)
2549 		return -ENOMEM;
2550 
2551 	err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
2552 	if (err) {
2553 		dst_release((struct dst_entry *)tun_dst);
2554 		return err;
2555 	}
2556 
2557 	a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
2558 			 sizeof(*ovs_tun), log);
2559 	if (IS_ERR(a)) {
2560 		dst_release((struct dst_entry *)tun_dst);
2561 		return PTR_ERR(a);
2562 	}
2563 
2564 	ovs_tun = nla_data(a);
2565 	ovs_tun->tun_dst = tun_dst;
2566 
2567 	tun_info = &tun_dst->u.tun_info;
2568 	tun_info->mode = IP_TUNNEL_INFO_TX;
2569 	if (key.tun_proto == AF_INET6)
2570 		tun_info->mode |= IP_TUNNEL_INFO_IPV6;
2571 	tun_info->key = key.tun_key;
2572 
2573 	/* We need to store the options in the action itself since
2574 	 * everything else will go away after flow setup. We can append
2575 	 * it to tun_info and then point there.
2576 	 */
2577 	ip_tunnel_info_opts_set(tun_info,
2578 				TUN_METADATA_OPTS(&key, key.tun_opts_len),
2579 				key.tun_opts_len);
2580 	add_nested_action_end(*sfa, start);
2581 
2582 	return err;
2583 }
2584 
2585 static bool validate_nsh(const struct nlattr *attr, bool is_mask,
2586 			 bool is_push_nsh, bool log)
2587 {
2588 	struct sw_flow_match match;
2589 	struct sw_flow_key key;
2590 	int ret = 0;
2591 
2592 	ovs_match_init(&match, &key, true, NULL);
2593 	ret = nsh_key_put_from_nlattr(attr, &match, is_mask,
2594 				      is_push_nsh, log);
2595 	return !ret;
2596 }
2597 
2598 /* Return false if there are any non-masked bits set.
2599  * Mask follows data immediately, before any netlink padding.
2600  */
2601 static bool validate_masked(u8 *data, int len)
2602 {
2603 	u8 *mask = data + len;
2604 
2605 	while (len--)
2606 		if (*data++ & ~*mask++)
2607 			return false;
2608 
2609 	return true;
2610 }
2611 
2612 static int validate_set(const struct nlattr *a,
2613 			const struct sw_flow_key *flow_key,
2614 			struct sw_flow_actions **sfa, bool *skip_copy,
2615 			u8 mac_proto, __be16 eth_type, bool masked, bool log)
2616 {
2617 	const struct nlattr *ovs_key = nla_data(a);
2618 	int key_type = nla_type(ovs_key);
2619 	size_t key_len;
2620 
2621 	/* There can be only one key in a action */
2622 	if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2623 		return -EINVAL;
2624 
2625 	key_len = nla_len(ovs_key);
2626 	if (masked)
2627 		key_len /= 2;
2628 
2629 	if (key_type > OVS_KEY_ATTR_MAX ||
2630 	    !check_attr_len(key_len, ovs_key_lens[key_type].len))
2631 		return -EINVAL;
2632 
2633 	if (masked && !validate_masked(nla_data(ovs_key), key_len))
2634 		return -EINVAL;
2635 
2636 	switch (key_type) {
2637 	const struct ovs_key_ipv4 *ipv4_key;
2638 	const struct ovs_key_ipv6 *ipv6_key;
2639 	int err;
2640 
2641 	case OVS_KEY_ATTR_PRIORITY:
2642 	case OVS_KEY_ATTR_SKB_MARK:
2643 	case OVS_KEY_ATTR_CT_MARK:
2644 	case OVS_KEY_ATTR_CT_LABELS:
2645 		break;
2646 
2647 	case OVS_KEY_ATTR_ETHERNET:
2648 		if (mac_proto != MAC_PROTO_ETHERNET)
2649 			return -EINVAL;
2650 		break;
2651 
2652 	case OVS_KEY_ATTR_TUNNEL:
2653 		if (masked)
2654 			return -EINVAL; /* Masked tunnel set not supported. */
2655 
2656 		*skip_copy = true;
2657 		err = validate_and_copy_set_tun(a, sfa, log);
2658 		if (err)
2659 			return err;
2660 		break;
2661 
2662 	case OVS_KEY_ATTR_IPV4:
2663 		if (eth_type != htons(ETH_P_IP))
2664 			return -EINVAL;
2665 
2666 		ipv4_key = nla_data(ovs_key);
2667 
2668 		if (masked) {
2669 			const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2670 
2671 			/* Non-writeable fields. */
2672 			if (mask->ipv4_proto || mask->ipv4_frag)
2673 				return -EINVAL;
2674 		} else {
2675 			if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2676 				return -EINVAL;
2677 
2678 			if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2679 				return -EINVAL;
2680 		}
2681 		break;
2682 
2683 	case OVS_KEY_ATTR_IPV6:
2684 		if (eth_type != htons(ETH_P_IPV6))
2685 			return -EINVAL;
2686 
2687 		ipv6_key = nla_data(ovs_key);
2688 
2689 		if (masked) {
2690 			const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2691 
2692 			/* Non-writeable fields. */
2693 			if (mask->ipv6_proto || mask->ipv6_frag)
2694 				return -EINVAL;
2695 
2696 			/* Invalid bits in the flow label mask? */
2697 			if (ntohl(mask->ipv6_label) & 0xFFF00000)
2698 				return -EINVAL;
2699 		} else {
2700 			if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2701 				return -EINVAL;
2702 
2703 			if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2704 				return -EINVAL;
2705 		}
2706 		if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2707 			return -EINVAL;
2708 
2709 		break;
2710 
2711 	case OVS_KEY_ATTR_TCP:
2712 		if ((eth_type != htons(ETH_P_IP) &&
2713 		     eth_type != htons(ETH_P_IPV6)) ||
2714 		    flow_key->ip.proto != IPPROTO_TCP)
2715 			return -EINVAL;
2716 
2717 		break;
2718 
2719 	case OVS_KEY_ATTR_UDP:
2720 		if ((eth_type != htons(ETH_P_IP) &&
2721 		     eth_type != htons(ETH_P_IPV6)) ||
2722 		    flow_key->ip.proto != IPPROTO_UDP)
2723 			return -EINVAL;
2724 
2725 		break;
2726 
2727 	case OVS_KEY_ATTR_MPLS:
2728 		if (!eth_p_mpls(eth_type))
2729 			return -EINVAL;
2730 		break;
2731 
2732 	case OVS_KEY_ATTR_SCTP:
2733 		if ((eth_type != htons(ETH_P_IP) &&
2734 		     eth_type != htons(ETH_P_IPV6)) ||
2735 		    flow_key->ip.proto != IPPROTO_SCTP)
2736 			return -EINVAL;
2737 
2738 		break;
2739 
2740 	case OVS_KEY_ATTR_NSH:
2741 		if (eth_type != htons(ETH_P_NSH))
2742 			return -EINVAL;
2743 		if (!validate_nsh(nla_data(a), masked, false, log))
2744 			return -EINVAL;
2745 		break;
2746 
2747 	default:
2748 		return -EINVAL;
2749 	}
2750 
2751 	/* Convert non-masked non-tunnel set actions to masked set actions. */
2752 	if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2753 		int start, len = key_len * 2;
2754 		struct nlattr *at;
2755 
2756 		*skip_copy = true;
2757 
2758 		start = add_nested_action_start(sfa,
2759 						OVS_ACTION_ATTR_SET_TO_MASKED,
2760 						log);
2761 		if (start < 0)
2762 			return start;
2763 
2764 		at = __add_action(sfa, key_type, NULL, len, log);
2765 		if (IS_ERR(at))
2766 			return PTR_ERR(at);
2767 
2768 		memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2769 		memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2770 		/* Clear non-writeable bits from otherwise writeable fields. */
2771 		if (key_type == OVS_KEY_ATTR_IPV6) {
2772 			struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2773 
2774 			mask->ipv6_label &= htonl(0x000FFFFF);
2775 		}
2776 		add_nested_action_end(*sfa, start);
2777 	}
2778 
2779 	return 0;
2780 }
2781 
2782 static int validate_userspace(const struct nlattr *attr)
2783 {
2784 	static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2785 		[OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2786 		[OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2787 		[OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2788 	};
2789 	struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2790 	int error;
2791 
2792 	error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX, attr,
2793 				 userspace_policy, NULL);
2794 	if (error)
2795 		return error;
2796 
2797 	if (!a[OVS_USERSPACE_ATTR_PID] ||
2798 	    !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2799 		return -EINVAL;
2800 
2801 	return 0;
2802 }
2803 
2804 static int copy_action(const struct nlattr *from,
2805 		       struct sw_flow_actions **sfa, bool log)
2806 {
2807 	int totlen = NLA_ALIGN(from->nla_len);
2808 	struct nlattr *to;
2809 
2810 	to = reserve_sfa_size(sfa, from->nla_len, log);
2811 	if (IS_ERR(to))
2812 		return PTR_ERR(to);
2813 
2814 	memcpy(to, from, totlen);
2815 	return 0;
2816 }
2817 
2818 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2819 				  const struct sw_flow_key *key,
2820 				  struct sw_flow_actions **sfa,
2821 				  __be16 eth_type, __be16 vlan_tci, bool log)
2822 {
2823 	u8 mac_proto = ovs_key_mac_proto(key);
2824 	const struct nlattr *a;
2825 	int rem, err;
2826 
2827 	nla_for_each_nested(a, attr, rem) {
2828 		/* Expected argument lengths, (u32)-1 for variable length. */
2829 		static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2830 			[OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2831 			[OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2832 			[OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2833 			[OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2834 			[OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2835 			[OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2836 			[OVS_ACTION_ATTR_POP_VLAN] = 0,
2837 			[OVS_ACTION_ATTR_SET] = (u32)-1,
2838 			[OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2839 			[OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2840 			[OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2841 			[OVS_ACTION_ATTR_CT] = (u32)-1,
2842 			[OVS_ACTION_ATTR_CT_CLEAR] = 0,
2843 			[OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
2844 			[OVS_ACTION_ATTR_PUSH_ETH] = sizeof(struct ovs_action_push_eth),
2845 			[OVS_ACTION_ATTR_POP_ETH] = 0,
2846 			[OVS_ACTION_ATTR_PUSH_NSH] = (u32)-1,
2847 			[OVS_ACTION_ATTR_POP_NSH] = 0,
2848 			[OVS_ACTION_ATTR_METER] = sizeof(u32),
2849 		};
2850 		const struct ovs_action_push_vlan *vlan;
2851 		int type = nla_type(a);
2852 		bool skip_copy;
2853 
2854 		if (type > OVS_ACTION_ATTR_MAX ||
2855 		    (action_lens[type] != nla_len(a) &&
2856 		     action_lens[type] != (u32)-1))
2857 			return -EINVAL;
2858 
2859 		skip_copy = false;
2860 		switch (type) {
2861 		case OVS_ACTION_ATTR_UNSPEC:
2862 			return -EINVAL;
2863 
2864 		case OVS_ACTION_ATTR_USERSPACE:
2865 			err = validate_userspace(a);
2866 			if (err)
2867 				return err;
2868 			break;
2869 
2870 		case OVS_ACTION_ATTR_OUTPUT:
2871 			if (nla_get_u32(a) >= DP_MAX_PORTS)
2872 				return -EINVAL;
2873 			break;
2874 
2875 		case OVS_ACTION_ATTR_TRUNC: {
2876 			const struct ovs_action_trunc *trunc = nla_data(a);
2877 
2878 			if (trunc->max_len < ETH_HLEN)
2879 				return -EINVAL;
2880 			break;
2881 		}
2882 
2883 		case OVS_ACTION_ATTR_HASH: {
2884 			const struct ovs_action_hash *act_hash = nla_data(a);
2885 
2886 			switch (act_hash->hash_alg) {
2887 			case OVS_HASH_ALG_L4:
2888 				break;
2889 			default:
2890 				return  -EINVAL;
2891 			}
2892 
2893 			break;
2894 		}
2895 
2896 		case OVS_ACTION_ATTR_POP_VLAN:
2897 			if (mac_proto != MAC_PROTO_ETHERNET)
2898 				return -EINVAL;
2899 			vlan_tci = htons(0);
2900 			break;
2901 
2902 		case OVS_ACTION_ATTR_PUSH_VLAN:
2903 			if (mac_proto != MAC_PROTO_ETHERNET)
2904 				return -EINVAL;
2905 			vlan = nla_data(a);
2906 			if (!eth_type_vlan(vlan->vlan_tpid))
2907 				return -EINVAL;
2908 			if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2909 				return -EINVAL;
2910 			vlan_tci = vlan->vlan_tci;
2911 			break;
2912 
2913 		case OVS_ACTION_ATTR_RECIRC:
2914 			break;
2915 
2916 		case OVS_ACTION_ATTR_PUSH_MPLS: {
2917 			const struct ovs_action_push_mpls *mpls = nla_data(a);
2918 
2919 			if (!eth_p_mpls(mpls->mpls_ethertype))
2920 				return -EINVAL;
2921 			/* Prohibit push MPLS other than to a white list
2922 			 * for packets that have a known tag order.
2923 			 */
2924 			if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2925 			    (eth_type != htons(ETH_P_IP) &&
2926 			     eth_type != htons(ETH_P_IPV6) &&
2927 			     eth_type != htons(ETH_P_ARP) &&
2928 			     eth_type != htons(ETH_P_RARP) &&
2929 			     !eth_p_mpls(eth_type)))
2930 				return -EINVAL;
2931 			eth_type = mpls->mpls_ethertype;
2932 			break;
2933 		}
2934 
2935 		case OVS_ACTION_ATTR_POP_MPLS:
2936 			if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2937 			    !eth_p_mpls(eth_type))
2938 				return -EINVAL;
2939 
2940 			/* Disallow subsequent L2.5+ set and mpls_pop actions
2941 			 * as there is no check here to ensure that the new
2942 			 * eth_type is valid and thus set actions could
2943 			 * write off the end of the packet or otherwise
2944 			 * corrupt it.
2945 			 *
2946 			 * Support for these actions is planned using packet
2947 			 * recirculation.
2948 			 */
2949 			eth_type = htons(0);
2950 			break;
2951 
2952 		case OVS_ACTION_ATTR_SET:
2953 			err = validate_set(a, key, sfa,
2954 					   &skip_copy, mac_proto, eth_type,
2955 					   false, log);
2956 			if (err)
2957 				return err;
2958 			break;
2959 
2960 		case OVS_ACTION_ATTR_SET_MASKED:
2961 			err = validate_set(a, key, sfa,
2962 					   &skip_copy, mac_proto, eth_type,
2963 					   true, log);
2964 			if (err)
2965 				return err;
2966 			break;
2967 
2968 		case OVS_ACTION_ATTR_SAMPLE: {
2969 			bool last = nla_is_last(a, rem);
2970 
2971 			err = validate_and_copy_sample(net, a, key, sfa,
2972 						       eth_type, vlan_tci,
2973 						       log, last);
2974 			if (err)
2975 				return err;
2976 			skip_copy = true;
2977 			break;
2978 		}
2979 
2980 		case OVS_ACTION_ATTR_CT:
2981 			err = ovs_ct_copy_action(net, a, key, sfa, log);
2982 			if (err)
2983 				return err;
2984 			skip_copy = true;
2985 			break;
2986 
2987 		case OVS_ACTION_ATTR_CT_CLEAR:
2988 			break;
2989 
2990 		case OVS_ACTION_ATTR_PUSH_ETH:
2991 			/* Disallow pushing an Ethernet header if one
2992 			 * is already present */
2993 			if (mac_proto != MAC_PROTO_NONE)
2994 				return -EINVAL;
2995 			mac_proto = MAC_PROTO_NONE;
2996 			break;
2997 
2998 		case OVS_ACTION_ATTR_POP_ETH:
2999 			if (mac_proto != MAC_PROTO_ETHERNET)
3000 				return -EINVAL;
3001 			if (vlan_tci & htons(VLAN_TAG_PRESENT))
3002 				return -EINVAL;
3003 			mac_proto = MAC_PROTO_ETHERNET;
3004 			break;
3005 
3006 		case OVS_ACTION_ATTR_PUSH_NSH:
3007 			if (mac_proto != MAC_PROTO_ETHERNET) {
3008 				u8 next_proto;
3009 
3010 				next_proto = tun_p_from_eth_p(eth_type);
3011 				if (!next_proto)
3012 					return -EINVAL;
3013 			}
3014 			mac_proto = MAC_PROTO_NONE;
3015 			if (!validate_nsh(nla_data(a), false, true, true))
3016 				return -EINVAL;
3017 			break;
3018 
3019 		case OVS_ACTION_ATTR_POP_NSH: {
3020 			__be16 inner_proto;
3021 
3022 			if (eth_type != htons(ETH_P_NSH))
3023 				return -EINVAL;
3024 			inner_proto = tun_p_to_eth_p(key->nsh.base.np);
3025 			if (!inner_proto)
3026 				return -EINVAL;
3027 			if (key->nsh.base.np == TUN_P_ETHERNET)
3028 				mac_proto = MAC_PROTO_ETHERNET;
3029 			else
3030 				mac_proto = MAC_PROTO_NONE;
3031 			break;
3032 		}
3033 
3034 		case OVS_ACTION_ATTR_METER:
3035 			/* Non-existent meters are simply ignored.  */
3036 			break;
3037 
3038 		default:
3039 			OVS_NLERR(log, "Unknown Action type %d", type);
3040 			return -EINVAL;
3041 		}
3042 		if (!skip_copy) {
3043 			err = copy_action(a, sfa, log);
3044 			if (err)
3045 				return err;
3046 		}
3047 	}
3048 
3049 	if (rem > 0)
3050 		return -EINVAL;
3051 
3052 	return 0;
3053 }
3054 
3055 /* 'key' must be the masked key. */
3056 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
3057 			 const struct sw_flow_key *key,
3058 			 struct sw_flow_actions **sfa, bool log)
3059 {
3060 	int err;
3061 
3062 	*sfa = nla_alloc_flow_actions(min(nla_len(attr), MAX_ACTIONS_BUFSIZE));
3063 	if (IS_ERR(*sfa))
3064 		return PTR_ERR(*sfa);
3065 
3066 	(*sfa)->orig_len = nla_len(attr);
3067 	err = __ovs_nla_copy_actions(net, attr, key, sfa, key->eth.type,
3068 				     key->eth.vlan.tci, log);
3069 	if (err)
3070 		ovs_nla_free_flow_actions(*sfa);
3071 
3072 	return err;
3073 }
3074 
3075 static int sample_action_to_attr(const struct nlattr *attr,
3076 				 struct sk_buff *skb)
3077 {
3078 	struct nlattr *start, *ac_start = NULL, *sample_arg;
3079 	int err = 0, rem = nla_len(attr);
3080 	const struct sample_arg *arg;
3081 	struct nlattr *actions;
3082 
3083 	start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
3084 	if (!start)
3085 		return -EMSGSIZE;
3086 
3087 	sample_arg = nla_data(attr);
3088 	arg = nla_data(sample_arg);
3089 	actions = nla_next(sample_arg, &rem);
3090 
3091 	if (nla_put_u32(skb, OVS_SAMPLE_ATTR_PROBABILITY, arg->probability)) {
3092 		err = -EMSGSIZE;
3093 		goto out;
3094 	}
3095 
3096 	ac_start = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
3097 	if (!ac_start) {
3098 		err = -EMSGSIZE;
3099 		goto out;
3100 	}
3101 
3102 	err = ovs_nla_put_actions(actions, rem, skb);
3103 
3104 out:
3105 	if (err) {
3106 		nla_nest_cancel(skb, ac_start);
3107 		nla_nest_cancel(skb, start);
3108 	} else {
3109 		nla_nest_end(skb, ac_start);
3110 		nla_nest_end(skb, start);
3111 	}
3112 
3113 	return err;
3114 }
3115 
3116 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
3117 {
3118 	const struct nlattr *ovs_key = nla_data(a);
3119 	int key_type = nla_type(ovs_key);
3120 	struct nlattr *start;
3121 	int err;
3122 
3123 	switch (key_type) {
3124 	case OVS_KEY_ATTR_TUNNEL_INFO: {
3125 		struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
3126 		struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
3127 
3128 		start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
3129 		if (!start)
3130 			return -EMSGSIZE;
3131 
3132 		err =  ip_tun_to_nlattr(skb, &tun_info->key,
3133 					ip_tunnel_info_opts(tun_info),
3134 					tun_info->options_len,
3135 					ip_tunnel_info_af(tun_info));
3136 		if (err)
3137 			return err;
3138 		nla_nest_end(skb, start);
3139 		break;
3140 	}
3141 	default:
3142 		if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
3143 			return -EMSGSIZE;
3144 		break;
3145 	}
3146 
3147 	return 0;
3148 }
3149 
3150 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
3151 						struct sk_buff *skb)
3152 {
3153 	const struct nlattr *ovs_key = nla_data(a);
3154 	struct nlattr *nla;
3155 	size_t key_len = nla_len(ovs_key) / 2;
3156 
3157 	/* Revert the conversion we did from a non-masked set action to
3158 	 * masked set action.
3159 	 */
3160 	nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
3161 	if (!nla)
3162 		return -EMSGSIZE;
3163 
3164 	if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
3165 		return -EMSGSIZE;
3166 
3167 	nla_nest_end(skb, nla);
3168 	return 0;
3169 }
3170 
3171 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
3172 {
3173 	const struct nlattr *a;
3174 	int rem, err;
3175 
3176 	nla_for_each_attr(a, attr, len, rem) {
3177 		int type = nla_type(a);
3178 
3179 		switch (type) {
3180 		case OVS_ACTION_ATTR_SET:
3181 			err = set_action_to_attr(a, skb);
3182 			if (err)
3183 				return err;
3184 			break;
3185 
3186 		case OVS_ACTION_ATTR_SET_TO_MASKED:
3187 			err = masked_set_action_to_set_action_attr(a, skb);
3188 			if (err)
3189 				return err;
3190 			break;
3191 
3192 		case OVS_ACTION_ATTR_SAMPLE:
3193 			err = sample_action_to_attr(a, skb);
3194 			if (err)
3195 				return err;
3196 			break;
3197 
3198 		case OVS_ACTION_ATTR_CT:
3199 			err = ovs_ct_action_to_attr(nla_data(a), skb);
3200 			if (err)
3201 				return err;
3202 			break;
3203 
3204 		default:
3205 			if (nla_put(skb, type, nla_len(a), nla_data(a)))
3206 				return -EMSGSIZE;
3207 			break;
3208 		}
3209 	}
3210 
3211 	return 0;
3212 }
3213