xref: /openbmc/linux/net/openvswitch/flow_netlink.c (revision 3f2a3050b4a3e7f32fc0ea3c9b0183090ae00522)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2007-2017 Nicira, Inc.
4  */
5 
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 
8 #include "flow.h"
9 #include "datapath.h"
10 #include <linux/uaccess.h>
11 #include <linux/netdevice.h>
12 #include <linux/etherdevice.h>
13 #include <linux/if_ether.h>
14 #include <linux/if_vlan.h>
15 #include <net/llc_pdu.h>
16 #include <linux/kernel.h>
17 #include <linux/jhash.h>
18 #include <linux/jiffies.h>
19 #include <linux/llc.h>
20 #include <linux/module.h>
21 #include <linux/in.h>
22 #include <linux/rcupdate.h>
23 #include <linux/if_arp.h>
24 #include <linux/ip.h>
25 #include <linux/ipv6.h>
26 #include <linux/sctp.h>
27 #include <linux/tcp.h>
28 #include <linux/udp.h>
29 #include <linux/icmp.h>
30 #include <linux/icmpv6.h>
31 #include <linux/rculist.h>
32 #include <net/geneve.h>
33 #include <net/ip.h>
34 #include <net/ipv6.h>
35 #include <net/ndisc.h>
36 #include <net/mpls.h>
37 #include <net/vxlan.h>
38 #include <net/tun_proto.h>
39 #include <net/erspan.h>
40 
41 #include "flow_netlink.h"
42 
43 struct ovs_len_tbl {
44 	int len;
45 	const struct ovs_len_tbl *next;
46 };
47 
48 #define OVS_ATTR_NESTED -1
49 #define OVS_ATTR_VARIABLE -2
50 
51 static bool actions_may_change_flow(const struct nlattr *actions)
52 {
53 	struct nlattr *nla;
54 	int rem;
55 
56 	nla_for_each_nested(nla, actions, rem) {
57 		u16 action = nla_type(nla);
58 
59 		switch (action) {
60 		case OVS_ACTION_ATTR_OUTPUT:
61 		case OVS_ACTION_ATTR_RECIRC:
62 		case OVS_ACTION_ATTR_TRUNC:
63 		case OVS_ACTION_ATTR_USERSPACE:
64 			break;
65 
66 		case OVS_ACTION_ATTR_CT:
67 		case OVS_ACTION_ATTR_CT_CLEAR:
68 		case OVS_ACTION_ATTR_HASH:
69 		case OVS_ACTION_ATTR_POP_ETH:
70 		case OVS_ACTION_ATTR_POP_MPLS:
71 		case OVS_ACTION_ATTR_POP_NSH:
72 		case OVS_ACTION_ATTR_POP_VLAN:
73 		case OVS_ACTION_ATTR_PUSH_ETH:
74 		case OVS_ACTION_ATTR_PUSH_MPLS:
75 		case OVS_ACTION_ATTR_PUSH_NSH:
76 		case OVS_ACTION_ATTR_PUSH_VLAN:
77 		case OVS_ACTION_ATTR_SAMPLE:
78 		case OVS_ACTION_ATTR_SET:
79 		case OVS_ACTION_ATTR_SET_MASKED:
80 		case OVS_ACTION_ATTR_METER:
81 		case OVS_ACTION_ATTR_CHECK_PKT_LEN:
82 		case OVS_ACTION_ATTR_ADD_MPLS:
83 		case OVS_ACTION_ATTR_DEC_TTL:
84 		default:
85 			return true;
86 		}
87 	}
88 	return false;
89 }
90 
91 static void update_range(struct sw_flow_match *match,
92 			 size_t offset, size_t size, bool is_mask)
93 {
94 	struct sw_flow_key_range *range;
95 	size_t start = rounddown(offset, sizeof(long));
96 	size_t end = roundup(offset + size, sizeof(long));
97 
98 	if (!is_mask)
99 		range = &match->range;
100 	else
101 		range = &match->mask->range;
102 
103 	if (range->start == range->end) {
104 		range->start = start;
105 		range->end = end;
106 		return;
107 	}
108 
109 	if (range->start > start)
110 		range->start = start;
111 
112 	if (range->end < end)
113 		range->end = end;
114 }
115 
116 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
117 	do { \
118 		update_range(match, offsetof(struct sw_flow_key, field),    \
119 			     sizeof((match)->key->field), is_mask);	    \
120 		if (is_mask)						    \
121 			(match)->mask->key.field = value;		    \
122 		else							    \
123 			(match)->key->field = value;		            \
124 	} while (0)
125 
126 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)	    \
127 	do {								    \
128 		update_range(match, offset, len, is_mask);		    \
129 		if (is_mask)						    \
130 			memcpy((u8 *)&(match)->mask->key + offset, value_p, \
131 			       len);					   \
132 		else							    \
133 			memcpy((u8 *)(match)->key + offset, value_p, len);  \
134 	} while (0)
135 
136 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)		      \
137 	SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
138 				  value_p, len, is_mask)
139 
140 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)		    \
141 	do {								    \
142 		update_range(match, offsetof(struct sw_flow_key, field),    \
143 			     sizeof((match)->key->field), is_mask);	    \
144 		if (is_mask)						    \
145 			memset((u8 *)&(match)->mask->key.field, value,      \
146 			       sizeof((match)->mask->key.field));	    \
147 		else							    \
148 			memset((u8 *)&(match)->key->field, value,           \
149 			       sizeof((match)->key->field));                \
150 	} while (0)
151 
152 static bool match_validate(const struct sw_flow_match *match,
153 			   u64 key_attrs, u64 mask_attrs, bool log)
154 {
155 	u64 key_expected = 0;
156 	u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
157 
158 	/* The following mask attributes allowed only if they
159 	 * pass the validation tests. */
160 	mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
161 			| (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)
162 			| (1 << OVS_KEY_ATTR_IPV6)
163 			| (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)
164 			| (1 << OVS_KEY_ATTR_TCP)
165 			| (1 << OVS_KEY_ATTR_TCP_FLAGS)
166 			| (1 << OVS_KEY_ATTR_UDP)
167 			| (1 << OVS_KEY_ATTR_SCTP)
168 			| (1 << OVS_KEY_ATTR_ICMP)
169 			| (1 << OVS_KEY_ATTR_ICMPV6)
170 			| (1 << OVS_KEY_ATTR_ARP)
171 			| (1 << OVS_KEY_ATTR_ND)
172 			| (1 << OVS_KEY_ATTR_MPLS)
173 			| (1 << OVS_KEY_ATTR_NSH));
174 
175 	/* Always allowed mask fields. */
176 	mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
177 		       | (1 << OVS_KEY_ATTR_IN_PORT)
178 		       | (1 << OVS_KEY_ATTR_ETHERTYPE));
179 
180 	/* Check key attributes. */
181 	if (match->key->eth.type == htons(ETH_P_ARP)
182 			|| match->key->eth.type == htons(ETH_P_RARP)) {
183 		key_expected |= 1 << OVS_KEY_ATTR_ARP;
184 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
185 			mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
186 	}
187 
188 	if (eth_p_mpls(match->key->eth.type)) {
189 		key_expected |= 1 << OVS_KEY_ATTR_MPLS;
190 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
191 			mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
192 	}
193 
194 	if (match->key->eth.type == htons(ETH_P_IP)) {
195 		key_expected |= 1 << OVS_KEY_ATTR_IPV4;
196 		if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
197 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
198 			mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4;
199 		}
200 
201 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
202 			if (match->key->ip.proto == IPPROTO_UDP) {
203 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
204 				if (match->mask && (match->mask->key.ip.proto == 0xff))
205 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
206 			}
207 
208 			if (match->key->ip.proto == IPPROTO_SCTP) {
209 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
210 				if (match->mask && (match->mask->key.ip.proto == 0xff))
211 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
212 			}
213 
214 			if (match->key->ip.proto == IPPROTO_TCP) {
215 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
216 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
217 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
218 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
219 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
220 				}
221 			}
222 
223 			if (match->key->ip.proto == IPPROTO_ICMP) {
224 				key_expected |= 1 << OVS_KEY_ATTR_ICMP;
225 				if (match->mask && (match->mask->key.ip.proto == 0xff))
226 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
227 			}
228 		}
229 	}
230 
231 	if (match->key->eth.type == htons(ETH_P_IPV6)) {
232 		key_expected |= 1 << OVS_KEY_ATTR_IPV6;
233 		if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
234 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
235 			mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6;
236 		}
237 
238 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
239 			if (match->key->ip.proto == IPPROTO_UDP) {
240 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
241 				if (match->mask && (match->mask->key.ip.proto == 0xff))
242 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
243 			}
244 
245 			if (match->key->ip.proto == IPPROTO_SCTP) {
246 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
247 				if (match->mask && (match->mask->key.ip.proto == 0xff))
248 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
249 			}
250 
251 			if (match->key->ip.proto == IPPROTO_TCP) {
252 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
253 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
254 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
255 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
256 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
257 				}
258 			}
259 
260 			if (match->key->ip.proto == IPPROTO_ICMPV6) {
261 				key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
262 				if (match->mask && (match->mask->key.ip.proto == 0xff))
263 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
264 
265 				if (match->key->tp.src ==
266 						htons(NDISC_NEIGHBOUR_SOLICITATION) ||
267 				    match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
268 					key_expected |= 1 << OVS_KEY_ATTR_ND;
269 					/* Original direction conntrack tuple
270 					 * uses the same space as the ND fields
271 					 * in the key, so both are not allowed
272 					 * at the same time.
273 					 */
274 					mask_allowed &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
275 					if (match->mask && (match->mask->key.tp.src == htons(0xff)))
276 						mask_allowed |= 1 << OVS_KEY_ATTR_ND;
277 				}
278 			}
279 		}
280 	}
281 
282 	if (match->key->eth.type == htons(ETH_P_NSH)) {
283 		key_expected |= 1 << OVS_KEY_ATTR_NSH;
284 		if (match->mask &&
285 		    match->mask->key.eth.type == htons(0xffff)) {
286 			mask_allowed |= 1 << OVS_KEY_ATTR_NSH;
287 		}
288 	}
289 
290 	if ((key_attrs & key_expected) != key_expected) {
291 		/* Key attributes check failed. */
292 		OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
293 			  (unsigned long long)key_attrs,
294 			  (unsigned long long)key_expected);
295 		return false;
296 	}
297 
298 	if ((mask_attrs & mask_allowed) != mask_attrs) {
299 		/* Mask attributes check failed. */
300 		OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
301 			  (unsigned long long)mask_attrs,
302 			  (unsigned long long)mask_allowed);
303 		return false;
304 	}
305 
306 	return true;
307 }
308 
309 size_t ovs_tun_key_attr_size(void)
310 {
311 	/* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
312 	 * updating this function.
313 	 */
314 	return    nla_total_size_64bit(8) /* OVS_TUNNEL_KEY_ATTR_ID */
315 		+ nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
316 		+ nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_DST */
317 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
318 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
319 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
320 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
321 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
322 		+ nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
323 		/* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS and
324 		 * OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS is mutually exclusive with
325 		 * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
326 		 */
327 		+ nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
328 		+ nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
329 }
330 
331 static size_t ovs_nsh_key_attr_size(void)
332 {
333 	/* Whenever adding new OVS_NSH_KEY_ FIELDS, we should consider
334 	 * updating this function.
335 	 */
336 	return  nla_total_size(NSH_BASE_HDR_LEN) /* OVS_NSH_KEY_ATTR_BASE */
337 		/* OVS_NSH_KEY_ATTR_MD1 and OVS_NSH_KEY_ATTR_MD2 are
338 		 * mutually exclusive, so the bigger one can cover
339 		 * the small one.
340 		 */
341 		+ nla_total_size(NSH_CTX_HDRS_MAX_LEN);
342 }
343 
344 size_t ovs_key_attr_size(void)
345 {
346 	/* Whenever adding new OVS_KEY_ FIELDS, we should consider
347 	 * updating this function.
348 	 */
349 	BUILD_BUG_ON(OVS_KEY_ATTR_MAX != 32);
350 
351 	return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
352 		+ nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
353 		  + ovs_tun_key_attr_size()
354 		+ nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
355 		+ nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
356 		+ nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
357 		+ nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
358 		+ nla_total_size(4)   /* OVS_KEY_ATTR_CT_STATE */
359 		+ nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
360 		+ nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
361 		+ nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABELS */
362 		+ nla_total_size(40)  /* OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6 */
363 		+ nla_total_size(0)   /* OVS_KEY_ATTR_NSH */
364 		  + ovs_nsh_key_attr_size()
365 		+ nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
366 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
367 		+ nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
368 		+ nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
369 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
370 		+ nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
371 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
372 		+ nla_total_size(28)  /* OVS_KEY_ATTR_ND */
373 		+ nla_total_size(2);  /* OVS_KEY_ATTR_IPV6_EXTHDRS */
374 }
375 
376 static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
377 	[OVS_VXLAN_EXT_GBP]	    = { .len = sizeof(u32) },
378 };
379 
380 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
381 	[OVS_TUNNEL_KEY_ATTR_ID]	    = { .len = sizeof(u64) },
382 	[OVS_TUNNEL_KEY_ATTR_IPV4_SRC]	    = { .len = sizeof(u32) },
383 	[OVS_TUNNEL_KEY_ATTR_IPV4_DST]	    = { .len = sizeof(u32) },
384 	[OVS_TUNNEL_KEY_ATTR_TOS]	    = { .len = 1 },
385 	[OVS_TUNNEL_KEY_ATTR_TTL]	    = { .len = 1 },
386 	[OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
387 	[OVS_TUNNEL_KEY_ATTR_CSUM]	    = { .len = 0 },
388 	[OVS_TUNNEL_KEY_ATTR_TP_SRC]	    = { .len = sizeof(u16) },
389 	[OVS_TUNNEL_KEY_ATTR_TP_DST]	    = { .len = sizeof(u16) },
390 	[OVS_TUNNEL_KEY_ATTR_OAM]	    = { .len = 0 },
391 	[OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_VARIABLE },
392 	[OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED,
393 						.next = ovs_vxlan_ext_key_lens },
394 	[OVS_TUNNEL_KEY_ATTR_IPV6_SRC]      = { .len = sizeof(struct in6_addr) },
395 	[OVS_TUNNEL_KEY_ATTR_IPV6_DST]      = { .len = sizeof(struct in6_addr) },
396 	[OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS]   = { .len = OVS_ATTR_VARIABLE },
397 	[OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE]   = { .len = 0 },
398 };
399 
400 static const struct ovs_len_tbl
401 ovs_nsh_key_attr_lens[OVS_NSH_KEY_ATTR_MAX + 1] = {
402 	[OVS_NSH_KEY_ATTR_BASE] = { .len = sizeof(struct ovs_nsh_key_base) },
403 	[OVS_NSH_KEY_ATTR_MD1]  = { .len = sizeof(struct ovs_nsh_key_md1) },
404 	[OVS_NSH_KEY_ATTR_MD2]  = { .len = OVS_ATTR_VARIABLE },
405 };
406 
407 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
408 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
409 	[OVS_KEY_ATTR_ENCAP]	 = { .len = OVS_ATTR_NESTED },
410 	[OVS_KEY_ATTR_PRIORITY]	 = { .len = sizeof(u32) },
411 	[OVS_KEY_ATTR_IN_PORT]	 = { .len = sizeof(u32) },
412 	[OVS_KEY_ATTR_SKB_MARK]	 = { .len = sizeof(u32) },
413 	[OVS_KEY_ATTR_ETHERNET]	 = { .len = sizeof(struct ovs_key_ethernet) },
414 	[OVS_KEY_ATTR_VLAN]	 = { .len = sizeof(__be16) },
415 	[OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
416 	[OVS_KEY_ATTR_IPV4]	 = { .len = sizeof(struct ovs_key_ipv4) },
417 	[OVS_KEY_ATTR_IPV6]	 = { .len = sizeof(struct ovs_key_ipv6) },
418 	[OVS_KEY_ATTR_TCP]	 = { .len = sizeof(struct ovs_key_tcp) },
419 	[OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
420 	[OVS_KEY_ATTR_UDP]	 = { .len = sizeof(struct ovs_key_udp) },
421 	[OVS_KEY_ATTR_SCTP]	 = { .len = sizeof(struct ovs_key_sctp) },
422 	[OVS_KEY_ATTR_ICMP]	 = { .len = sizeof(struct ovs_key_icmp) },
423 	[OVS_KEY_ATTR_ICMPV6]	 = { .len = sizeof(struct ovs_key_icmpv6) },
424 	[OVS_KEY_ATTR_ARP]	 = { .len = sizeof(struct ovs_key_arp) },
425 	[OVS_KEY_ATTR_ND]	 = { .len = sizeof(struct ovs_key_nd) },
426 	[OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
427 	[OVS_KEY_ATTR_DP_HASH]	 = { .len = sizeof(u32) },
428 	[OVS_KEY_ATTR_TUNNEL]	 = { .len = OVS_ATTR_NESTED,
429 				     .next = ovs_tunnel_key_lens, },
430 	[OVS_KEY_ATTR_MPLS]	 = { .len = OVS_ATTR_VARIABLE },
431 	[OVS_KEY_ATTR_CT_STATE]	 = { .len = sizeof(u32) },
432 	[OVS_KEY_ATTR_CT_ZONE]	 = { .len = sizeof(u16) },
433 	[OVS_KEY_ATTR_CT_MARK]	 = { .len = sizeof(u32) },
434 	[OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
435 	[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4] = {
436 		.len = sizeof(struct ovs_key_ct_tuple_ipv4) },
437 	[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6] = {
438 		.len = sizeof(struct ovs_key_ct_tuple_ipv6) },
439 	[OVS_KEY_ATTR_NSH]       = { .len = OVS_ATTR_NESTED,
440 				     .next = ovs_nsh_key_attr_lens, },
441 	[OVS_KEY_ATTR_IPV6_EXTHDRS] = {
442 		.len = sizeof(struct ovs_key_ipv6_exthdrs) },
443 };
444 
445 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
446 {
447 	return expected_len == attr_len ||
448 	       expected_len == OVS_ATTR_NESTED ||
449 	       expected_len == OVS_ATTR_VARIABLE;
450 }
451 
452 static bool is_all_zero(const u8 *fp, size_t size)
453 {
454 	int i;
455 
456 	if (!fp)
457 		return false;
458 
459 	for (i = 0; i < size; i++)
460 		if (fp[i])
461 			return false;
462 
463 	return true;
464 }
465 
466 static int __parse_flow_nlattrs(const struct nlattr *attr,
467 				const struct nlattr *a[],
468 				u64 *attrsp, bool log, bool nz)
469 {
470 	const struct nlattr *nla;
471 	u64 attrs;
472 	int rem;
473 
474 	attrs = *attrsp;
475 	nla_for_each_nested(nla, attr, rem) {
476 		u16 type = nla_type(nla);
477 		int expected_len;
478 
479 		if (type > OVS_KEY_ATTR_MAX) {
480 			OVS_NLERR(log, "Key type %d is out of range max %d",
481 				  type, OVS_KEY_ATTR_MAX);
482 			return -EINVAL;
483 		}
484 
485 		if (type == OVS_KEY_ATTR_PACKET_TYPE ||
486 		    type == OVS_KEY_ATTR_ND_EXTENSIONS ||
487 		    type == OVS_KEY_ATTR_TUNNEL_INFO) {
488 			OVS_NLERR(log, "Key type %d is not supported", type);
489 			return -EINVAL;
490 		}
491 
492 		if (attrs & (1ULL << type)) {
493 			OVS_NLERR(log, "Duplicate key (type %d).", type);
494 			return -EINVAL;
495 		}
496 
497 		expected_len = ovs_key_lens[type].len;
498 		if (!check_attr_len(nla_len(nla), expected_len)) {
499 			OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
500 				  type, nla_len(nla), expected_len);
501 			return -EINVAL;
502 		}
503 
504 		if (!nz || !is_all_zero(nla_data(nla), nla_len(nla))) {
505 			attrs |= 1ULL << type;
506 			a[type] = nla;
507 		}
508 	}
509 	if (rem) {
510 		OVS_NLERR(log, "Message has %d unknown bytes.", rem);
511 		return -EINVAL;
512 	}
513 
514 	*attrsp = attrs;
515 	return 0;
516 }
517 
518 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
519 				   const struct nlattr *a[], u64 *attrsp,
520 				   bool log)
521 {
522 	return __parse_flow_nlattrs(attr, a, attrsp, log, true);
523 }
524 
525 int parse_flow_nlattrs(const struct nlattr *attr, const struct nlattr *a[],
526 		       u64 *attrsp, bool log)
527 {
528 	return __parse_flow_nlattrs(attr, a, attrsp, log, false);
529 }
530 
531 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
532 				     struct sw_flow_match *match, bool is_mask,
533 				     bool log)
534 {
535 	unsigned long opt_key_offset;
536 
537 	if (nla_len(a) > sizeof(match->key->tun_opts)) {
538 		OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
539 			  nla_len(a), sizeof(match->key->tun_opts));
540 		return -EINVAL;
541 	}
542 
543 	if (nla_len(a) % 4 != 0) {
544 		OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
545 			  nla_len(a));
546 		return -EINVAL;
547 	}
548 
549 	/* We need to record the length of the options passed
550 	 * down, otherwise packets with the same format but
551 	 * additional options will be silently matched.
552 	 */
553 	if (!is_mask) {
554 		SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
555 				false);
556 	} else {
557 		/* This is somewhat unusual because it looks at
558 		 * both the key and mask while parsing the
559 		 * attributes (and by extension assumes the key
560 		 * is parsed first). Normally, we would verify
561 		 * that each is the correct length and that the
562 		 * attributes line up in the validate function.
563 		 * However, that is difficult because this is
564 		 * variable length and we won't have the
565 		 * information later.
566 		 */
567 		if (match->key->tun_opts_len != nla_len(a)) {
568 			OVS_NLERR(log, "Geneve option len %d != mask len %d",
569 				  match->key->tun_opts_len, nla_len(a));
570 			return -EINVAL;
571 		}
572 
573 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
574 	}
575 
576 	opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
577 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
578 				  nla_len(a), is_mask);
579 	return 0;
580 }
581 
582 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
583 				     struct sw_flow_match *match, bool is_mask,
584 				     bool log)
585 {
586 	struct nlattr *a;
587 	int rem;
588 	unsigned long opt_key_offset;
589 	struct vxlan_metadata opts;
590 
591 	BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
592 
593 	memset(&opts, 0, sizeof(opts));
594 	nla_for_each_nested(a, attr, rem) {
595 		int type = nla_type(a);
596 
597 		if (type > OVS_VXLAN_EXT_MAX) {
598 			OVS_NLERR(log, "VXLAN extension %d out of range max %d",
599 				  type, OVS_VXLAN_EXT_MAX);
600 			return -EINVAL;
601 		}
602 
603 		if (!check_attr_len(nla_len(a),
604 				    ovs_vxlan_ext_key_lens[type].len)) {
605 			OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
606 				  type, nla_len(a),
607 				  ovs_vxlan_ext_key_lens[type].len);
608 			return -EINVAL;
609 		}
610 
611 		switch (type) {
612 		case OVS_VXLAN_EXT_GBP:
613 			opts.gbp = nla_get_u32(a);
614 			break;
615 		default:
616 			OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
617 				  type);
618 			return -EINVAL;
619 		}
620 	}
621 	if (rem) {
622 		OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
623 			  rem);
624 		return -EINVAL;
625 	}
626 
627 	if (!is_mask)
628 		SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
629 	else
630 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
631 
632 	opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
633 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
634 				  is_mask);
635 	return 0;
636 }
637 
638 static int erspan_tun_opt_from_nlattr(const struct nlattr *a,
639 				      struct sw_flow_match *match, bool is_mask,
640 				      bool log)
641 {
642 	unsigned long opt_key_offset;
643 
644 	BUILD_BUG_ON(sizeof(struct erspan_metadata) >
645 		     sizeof(match->key->tun_opts));
646 
647 	if (nla_len(a) > sizeof(match->key->tun_opts)) {
648 		OVS_NLERR(log, "ERSPAN option length err (len %d, max %zu).",
649 			  nla_len(a), sizeof(match->key->tun_opts));
650 		return -EINVAL;
651 	}
652 
653 	if (!is_mask)
654 		SW_FLOW_KEY_PUT(match, tun_opts_len,
655 				sizeof(struct erspan_metadata), false);
656 	else
657 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
658 
659 	opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
660 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
661 				  nla_len(a), is_mask);
662 	return 0;
663 }
664 
665 static int ip_tun_from_nlattr(const struct nlattr *attr,
666 			      struct sw_flow_match *match, bool is_mask,
667 			      bool log)
668 {
669 	bool ttl = false, ipv4 = false, ipv6 = false;
670 	bool info_bridge_mode = false;
671 	__be16 tun_flags = 0;
672 	int opts_type = 0;
673 	struct nlattr *a;
674 	int rem;
675 
676 	nla_for_each_nested(a, attr, rem) {
677 		int type = nla_type(a);
678 		int err;
679 
680 		if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
681 			OVS_NLERR(log, "Tunnel attr %d out of range max %d",
682 				  type, OVS_TUNNEL_KEY_ATTR_MAX);
683 			return -EINVAL;
684 		}
685 
686 		if (!check_attr_len(nla_len(a),
687 				    ovs_tunnel_key_lens[type].len)) {
688 			OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
689 				  type, nla_len(a), ovs_tunnel_key_lens[type].len);
690 			return -EINVAL;
691 		}
692 
693 		switch (type) {
694 		case OVS_TUNNEL_KEY_ATTR_ID:
695 			SW_FLOW_KEY_PUT(match, tun_key.tun_id,
696 					nla_get_be64(a), is_mask);
697 			tun_flags |= TUNNEL_KEY;
698 			break;
699 		case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
700 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
701 					nla_get_in_addr(a), is_mask);
702 			ipv4 = true;
703 			break;
704 		case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
705 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
706 					nla_get_in_addr(a), is_mask);
707 			ipv4 = true;
708 			break;
709 		case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
710 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.src,
711 					nla_get_in6_addr(a), is_mask);
712 			ipv6 = true;
713 			break;
714 		case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
715 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
716 					nla_get_in6_addr(a), is_mask);
717 			ipv6 = true;
718 			break;
719 		case OVS_TUNNEL_KEY_ATTR_TOS:
720 			SW_FLOW_KEY_PUT(match, tun_key.tos,
721 					nla_get_u8(a), is_mask);
722 			break;
723 		case OVS_TUNNEL_KEY_ATTR_TTL:
724 			SW_FLOW_KEY_PUT(match, tun_key.ttl,
725 					nla_get_u8(a), is_mask);
726 			ttl = true;
727 			break;
728 		case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
729 			tun_flags |= TUNNEL_DONT_FRAGMENT;
730 			break;
731 		case OVS_TUNNEL_KEY_ATTR_CSUM:
732 			tun_flags |= TUNNEL_CSUM;
733 			break;
734 		case OVS_TUNNEL_KEY_ATTR_TP_SRC:
735 			SW_FLOW_KEY_PUT(match, tun_key.tp_src,
736 					nla_get_be16(a), is_mask);
737 			break;
738 		case OVS_TUNNEL_KEY_ATTR_TP_DST:
739 			SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
740 					nla_get_be16(a), is_mask);
741 			break;
742 		case OVS_TUNNEL_KEY_ATTR_OAM:
743 			tun_flags |= TUNNEL_OAM;
744 			break;
745 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
746 			if (opts_type) {
747 				OVS_NLERR(log, "Multiple metadata blocks provided");
748 				return -EINVAL;
749 			}
750 
751 			err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
752 			if (err)
753 				return err;
754 
755 			tun_flags |= TUNNEL_GENEVE_OPT;
756 			opts_type = type;
757 			break;
758 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
759 			if (opts_type) {
760 				OVS_NLERR(log, "Multiple metadata blocks provided");
761 				return -EINVAL;
762 			}
763 
764 			err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
765 			if (err)
766 				return err;
767 
768 			tun_flags |= TUNNEL_VXLAN_OPT;
769 			opts_type = type;
770 			break;
771 		case OVS_TUNNEL_KEY_ATTR_PAD:
772 			break;
773 		case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
774 			if (opts_type) {
775 				OVS_NLERR(log, "Multiple metadata blocks provided");
776 				return -EINVAL;
777 			}
778 
779 			err = erspan_tun_opt_from_nlattr(a, match, is_mask,
780 							 log);
781 			if (err)
782 				return err;
783 
784 			tun_flags |= TUNNEL_ERSPAN_OPT;
785 			opts_type = type;
786 			break;
787 		case OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE:
788 			info_bridge_mode = true;
789 			ipv4 = true;
790 			break;
791 		default:
792 			OVS_NLERR(log, "Unknown IP tunnel attribute %d",
793 				  type);
794 			return -EINVAL;
795 		}
796 	}
797 
798 	SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
799 	if (is_mask)
800 		SW_FLOW_KEY_MEMSET_FIELD(match, tun_proto, 0xff, true);
801 	else
802 		SW_FLOW_KEY_PUT(match, tun_proto, ipv6 ? AF_INET6 : AF_INET,
803 				false);
804 
805 	if (rem > 0) {
806 		OVS_NLERR(log, "IP tunnel attribute has %d unknown bytes.",
807 			  rem);
808 		return -EINVAL;
809 	}
810 
811 	if (ipv4 && ipv6) {
812 		OVS_NLERR(log, "Mixed IPv4 and IPv6 tunnel attributes");
813 		return -EINVAL;
814 	}
815 
816 	if (!is_mask) {
817 		if (!ipv4 && !ipv6) {
818 			OVS_NLERR(log, "IP tunnel dst address not specified");
819 			return -EINVAL;
820 		}
821 		if (ipv4) {
822 			if (info_bridge_mode) {
823 				if (match->key->tun_key.u.ipv4.src ||
824 				    match->key->tun_key.u.ipv4.dst ||
825 				    match->key->tun_key.tp_src ||
826 				    match->key->tun_key.tp_dst ||
827 				    match->key->tun_key.ttl ||
828 				    match->key->tun_key.tos ||
829 				    tun_flags & ~TUNNEL_KEY) {
830 					OVS_NLERR(log, "IPv4 tun info is not correct");
831 					return -EINVAL;
832 				}
833 			} else if (!match->key->tun_key.u.ipv4.dst) {
834 				OVS_NLERR(log, "IPv4 tunnel dst address is zero");
835 				return -EINVAL;
836 			}
837 		}
838 		if (ipv6 && ipv6_addr_any(&match->key->tun_key.u.ipv6.dst)) {
839 			OVS_NLERR(log, "IPv6 tunnel dst address is zero");
840 			return -EINVAL;
841 		}
842 
843 		if (!ttl && !info_bridge_mode) {
844 			OVS_NLERR(log, "IP tunnel TTL not specified.");
845 			return -EINVAL;
846 		}
847 	}
848 
849 	return opts_type;
850 }
851 
852 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
853 			       const void *tun_opts, int swkey_tun_opts_len)
854 {
855 	const struct vxlan_metadata *opts = tun_opts;
856 	struct nlattr *nla;
857 
858 	nla = nla_nest_start_noflag(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
859 	if (!nla)
860 		return -EMSGSIZE;
861 
862 	if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
863 		return -EMSGSIZE;
864 
865 	nla_nest_end(skb, nla);
866 	return 0;
867 }
868 
869 static int __ip_tun_to_nlattr(struct sk_buff *skb,
870 			      const struct ip_tunnel_key *output,
871 			      const void *tun_opts, int swkey_tun_opts_len,
872 			      unsigned short tun_proto, u8 mode)
873 {
874 	if (output->tun_flags & TUNNEL_KEY &&
875 	    nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id,
876 			 OVS_TUNNEL_KEY_ATTR_PAD))
877 		return -EMSGSIZE;
878 
879 	if (mode & IP_TUNNEL_INFO_BRIDGE)
880 		return nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE)
881 		       ? -EMSGSIZE : 0;
882 
883 	switch (tun_proto) {
884 	case AF_INET:
885 		if (output->u.ipv4.src &&
886 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
887 				    output->u.ipv4.src))
888 			return -EMSGSIZE;
889 		if (output->u.ipv4.dst &&
890 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
891 				    output->u.ipv4.dst))
892 			return -EMSGSIZE;
893 		break;
894 	case AF_INET6:
895 		if (!ipv6_addr_any(&output->u.ipv6.src) &&
896 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
897 				     &output->u.ipv6.src))
898 			return -EMSGSIZE;
899 		if (!ipv6_addr_any(&output->u.ipv6.dst) &&
900 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
901 				     &output->u.ipv6.dst))
902 			return -EMSGSIZE;
903 		break;
904 	}
905 	if (output->tos &&
906 	    nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
907 		return -EMSGSIZE;
908 	if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
909 		return -EMSGSIZE;
910 	if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
911 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
912 		return -EMSGSIZE;
913 	if ((output->tun_flags & TUNNEL_CSUM) &&
914 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
915 		return -EMSGSIZE;
916 	if (output->tp_src &&
917 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
918 		return -EMSGSIZE;
919 	if (output->tp_dst &&
920 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
921 		return -EMSGSIZE;
922 	if ((output->tun_flags & TUNNEL_OAM) &&
923 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
924 		return -EMSGSIZE;
925 	if (swkey_tun_opts_len) {
926 		if (output->tun_flags & TUNNEL_GENEVE_OPT &&
927 		    nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
928 			    swkey_tun_opts_len, tun_opts))
929 			return -EMSGSIZE;
930 		else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
931 			 vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
932 			return -EMSGSIZE;
933 		else if (output->tun_flags & TUNNEL_ERSPAN_OPT &&
934 			 nla_put(skb, OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS,
935 				 swkey_tun_opts_len, tun_opts))
936 			return -EMSGSIZE;
937 	}
938 
939 	return 0;
940 }
941 
942 static int ip_tun_to_nlattr(struct sk_buff *skb,
943 			    const struct ip_tunnel_key *output,
944 			    const void *tun_opts, int swkey_tun_opts_len,
945 			    unsigned short tun_proto, u8 mode)
946 {
947 	struct nlattr *nla;
948 	int err;
949 
950 	nla = nla_nest_start_noflag(skb, OVS_KEY_ATTR_TUNNEL);
951 	if (!nla)
952 		return -EMSGSIZE;
953 
954 	err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
955 				 tun_proto, mode);
956 	if (err)
957 		return err;
958 
959 	nla_nest_end(skb, nla);
960 	return 0;
961 }
962 
963 int ovs_nla_put_tunnel_info(struct sk_buff *skb,
964 			    struct ip_tunnel_info *tun_info)
965 {
966 	return __ip_tun_to_nlattr(skb, &tun_info->key,
967 				  ip_tunnel_info_opts(tun_info),
968 				  tun_info->options_len,
969 				  ip_tunnel_info_af(tun_info), tun_info->mode);
970 }
971 
972 static int encode_vlan_from_nlattrs(struct sw_flow_match *match,
973 				    const struct nlattr *a[],
974 				    bool is_mask, bool inner)
975 {
976 	__be16 tci = 0;
977 	__be16 tpid = 0;
978 
979 	if (a[OVS_KEY_ATTR_VLAN])
980 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
981 
982 	if (a[OVS_KEY_ATTR_ETHERTYPE])
983 		tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
984 
985 	if (likely(!inner)) {
986 		SW_FLOW_KEY_PUT(match, eth.vlan.tpid, tpid, is_mask);
987 		SW_FLOW_KEY_PUT(match, eth.vlan.tci, tci, is_mask);
988 	} else {
989 		SW_FLOW_KEY_PUT(match, eth.cvlan.tpid, tpid, is_mask);
990 		SW_FLOW_KEY_PUT(match, eth.cvlan.tci, tci, is_mask);
991 	}
992 	return 0;
993 }
994 
995 static int validate_vlan_from_nlattrs(const struct sw_flow_match *match,
996 				      u64 key_attrs, bool inner,
997 				      const struct nlattr **a, bool log)
998 {
999 	__be16 tci = 0;
1000 
1001 	if (!((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1002 	      (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1003 	       eth_type_vlan(nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE])))) {
1004 		/* Not a VLAN. */
1005 		return 0;
1006 	}
1007 
1008 	if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1009 	      (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1010 		OVS_NLERR(log, "Invalid %s frame", (inner) ? "C-VLAN" : "VLAN");
1011 		return -EINVAL;
1012 	}
1013 
1014 	if (a[OVS_KEY_ATTR_VLAN])
1015 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1016 
1017 	if (!(tci & htons(VLAN_CFI_MASK))) {
1018 		if (tci) {
1019 			OVS_NLERR(log, "%s TCI does not have VLAN_CFI_MASK bit set.",
1020 				  (inner) ? "C-VLAN" : "VLAN");
1021 			return -EINVAL;
1022 		} else if (nla_len(a[OVS_KEY_ATTR_ENCAP])) {
1023 			/* Corner case for truncated VLAN header. */
1024 			OVS_NLERR(log, "Truncated %s header has non-zero encap attribute.",
1025 				  (inner) ? "C-VLAN" : "VLAN");
1026 			return -EINVAL;
1027 		}
1028 	}
1029 
1030 	return 1;
1031 }
1032 
1033 static int validate_vlan_mask_from_nlattrs(const struct sw_flow_match *match,
1034 					   u64 key_attrs, bool inner,
1035 					   const struct nlattr **a, bool log)
1036 {
1037 	__be16 tci = 0;
1038 	__be16 tpid = 0;
1039 	bool encap_valid = !!(match->key->eth.vlan.tci &
1040 			      htons(VLAN_CFI_MASK));
1041 	bool i_encap_valid = !!(match->key->eth.cvlan.tci &
1042 				htons(VLAN_CFI_MASK));
1043 
1044 	if (!(key_attrs & (1 << OVS_KEY_ATTR_ENCAP))) {
1045 		/* Not a VLAN. */
1046 		return 0;
1047 	}
1048 
1049 	if ((!inner && !encap_valid) || (inner && !i_encap_valid)) {
1050 		OVS_NLERR(log, "Encap mask attribute is set for non-%s frame.",
1051 			  (inner) ? "C-VLAN" : "VLAN");
1052 		return -EINVAL;
1053 	}
1054 
1055 	if (a[OVS_KEY_ATTR_VLAN])
1056 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1057 
1058 	if (a[OVS_KEY_ATTR_ETHERTYPE])
1059 		tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1060 
1061 	if (tpid != htons(0xffff)) {
1062 		OVS_NLERR(log, "Must have an exact match on %s TPID (mask=%x).",
1063 			  (inner) ? "C-VLAN" : "VLAN", ntohs(tpid));
1064 		return -EINVAL;
1065 	}
1066 	if (!(tci & htons(VLAN_CFI_MASK))) {
1067 		OVS_NLERR(log, "%s TCI mask does not have exact match for VLAN_CFI_MASK bit.",
1068 			  (inner) ? "C-VLAN" : "VLAN");
1069 		return -EINVAL;
1070 	}
1071 
1072 	return 1;
1073 }
1074 
1075 static int __parse_vlan_from_nlattrs(struct sw_flow_match *match,
1076 				     u64 *key_attrs, bool inner,
1077 				     const struct nlattr **a, bool is_mask,
1078 				     bool log)
1079 {
1080 	int err;
1081 	const struct nlattr *encap;
1082 
1083 	if (!is_mask)
1084 		err = validate_vlan_from_nlattrs(match, *key_attrs, inner,
1085 						 a, log);
1086 	else
1087 		err = validate_vlan_mask_from_nlattrs(match, *key_attrs, inner,
1088 						      a, log);
1089 	if (err <= 0)
1090 		return err;
1091 
1092 	err = encode_vlan_from_nlattrs(match, a, is_mask, inner);
1093 	if (err)
1094 		return err;
1095 
1096 	*key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1097 	*key_attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
1098 	*key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1099 
1100 	encap = a[OVS_KEY_ATTR_ENCAP];
1101 
1102 	if (!is_mask)
1103 		err = parse_flow_nlattrs(encap, a, key_attrs, log);
1104 	else
1105 		err = parse_flow_mask_nlattrs(encap, a, key_attrs, log);
1106 
1107 	return err;
1108 }
1109 
1110 static int parse_vlan_from_nlattrs(struct sw_flow_match *match,
1111 				   u64 *key_attrs, const struct nlattr **a,
1112 				   bool is_mask, bool log)
1113 {
1114 	int err;
1115 	bool encap_valid = false;
1116 
1117 	err = __parse_vlan_from_nlattrs(match, key_attrs, false, a,
1118 					is_mask, log);
1119 	if (err)
1120 		return err;
1121 
1122 	encap_valid = !!(match->key->eth.vlan.tci & htons(VLAN_CFI_MASK));
1123 	if (encap_valid) {
1124 		err = __parse_vlan_from_nlattrs(match, key_attrs, true, a,
1125 						is_mask, log);
1126 		if (err)
1127 			return err;
1128 	}
1129 
1130 	return 0;
1131 }
1132 
1133 static int parse_eth_type_from_nlattrs(struct sw_flow_match *match,
1134 				       u64 *attrs, const struct nlattr **a,
1135 				       bool is_mask, bool log)
1136 {
1137 	__be16 eth_type;
1138 
1139 	eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1140 	if (is_mask) {
1141 		/* Always exact match EtherType. */
1142 		eth_type = htons(0xffff);
1143 	} else if (!eth_proto_is_802_3(eth_type)) {
1144 		OVS_NLERR(log, "EtherType %x is less than min %x",
1145 				ntohs(eth_type), ETH_P_802_3_MIN);
1146 		return -EINVAL;
1147 	}
1148 
1149 	SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1150 	*attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1151 	return 0;
1152 }
1153 
1154 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
1155 				 u64 *attrs, const struct nlattr **a,
1156 				 bool is_mask, bool log)
1157 {
1158 	u8 mac_proto = MAC_PROTO_ETHERNET;
1159 
1160 	if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
1161 		u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
1162 
1163 		SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
1164 		*attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
1165 	}
1166 
1167 	if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
1168 		u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
1169 
1170 		SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
1171 		*attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
1172 	}
1173 
1174 	if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1175 		SW_FLOW_KEY_PUT(match, phy.priority,
1176 			  nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1177 		*attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1178 	}
1179 
1180 	if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1181 		u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1182 
1183 		if (is_mask) {
1184 			in_port = 0xffffffff; /* Always exact match in_port. */
1185 		} else if (in_port >= DP_MAX_PORTS) {
1186 			OVS_NLERR(log, "Port %d exceeds max allowable %d",
1187 				  in_port, DP_MAX_PORTS);
1188 			return -EINVAL;
1189 		}
1190 
1191 		SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1192 		*attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1193 	} else if (!is_mask) {
1194 		SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1195 	}
1196 
1197 	if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1198 		uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1199 
1200 		SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1201 		*attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1202 	}
1203 	if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1204 		if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1205 				       is_mask, log) < 0)
1206 			return -EINVAL;
1207 		*attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1208 	}
1209 
1210 	if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
1211 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
1212 		u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
1213 
1214 		if (ct_state & ~CT_SUPPORTED_MASK) {
1215 			OVS_NLERR(log, "ct_state flags %08x unsupported",
1216 				  ct_state);
1217 			return -EINVAL;
1218 		}
1219 
1220 		SW_FLOW_KEY_PUT(match, ct_state, ct_state, is_mask);
1221 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
1222 	}
1223 	if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
1224 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
1225 		u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
1226 
1227 		SW_FLOW_KEY_PUT(match, ct_zone, ct_zone, is_mask);
1228 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
1229 	}
1230 	if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
1231 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
1232 		u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
1233 
1234 		SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
1235 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
1236 	}
1237 	if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
1238 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
1239 		const struct ovs_key_ct_labels *cl;
1240 
1241 		cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
1242 		SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
1243 				   sizeof(*cl), is_mask);
1244 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
1245 	}
1246 	if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)) {
1247 		const struct ovs_key_ct_tuple_ipv4 *ct;
1248 
1249 		ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4]);
1250 
1251 		SW_FLOW_KEY_PUT(match, ipv4.ct_orig.src, ct->ipv4_src, is_mask);
1252 		SW_FLOW_KEY_PUT(match, ipv4.ct_orig.dst, ct->ipv4_dst, is_mask);
1253 		SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1254 		SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1255 		SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv4_proto, is_mask);
1256 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4);
1257 	}
1258 	if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)) {
1259 		const struct ovs_key_ct_tuple_ipv6 *ct;
1260 
1261 		ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6]);
1262 
1263 		SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.src, &ct->ipv6_src,
1264 				   sizeof(match->key->ipv6.ct_orig.src),
1265 				   is_mask);
1266 		SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.dst, &ct->ipv6_dst,
1267 				   sizeof(match->key->ipv6.ct_orig.dst),
1268 				   is_mask);
1269 		SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1270 		SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1271 		SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv6_proto, is_mask);
1272 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
1273 	}
1274 
1275 	/* For layer 3 packets the Ethernet type is provided
1276 	 * and treated as metadata but no MAC addresses are provided.
1277 	 */
1278 	if (!(*attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1279 	    (*attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)))
1280 		mac_proto = MAC_PROTO_NONE;
1281 
1282 	/* Always exact match mac_proto */
1283 	SW_FLOW_KEY_PUT(match, mac_proto, is_mask ? 0xff : mac_proto, is_mask);
1284 
1285 	if (mac_proto == MAC_PROTO_NONE)
1286 		return parse_eth_type_from_nlattrs(match, attrs, a, is_mask,
1287 						   log);
1288 
1289 	return 0;
1290 }
1291 
1292 int nsh_hdr_from_nlattr(const struct nlattr *attr,
1293 			struct nshhdr *nh, size_t size)
1294 {
1295 	struct nlattr *a;
1296 	int rem;
1297 	u8 flags = 0;
1298 	u8 ttl = 0;
1299 	int mdlen = 0;
1300 
1301 	/* validate_nsh has check this, so we needn't do duplicate check here
1302 	 */
1303 	if (size < NSH_BASE_HDR_LEN)
1304 		return -ENOBUFS;
1305 
1306 	nla_for_each_nested(a, attr, rem) {
1307 		int type = nla_type(a);
1308 
1309 		switch (type) {
1310 		case OVS_NSH_KEY_ATTR_BASE: {
1311 			const struct ovs_nsh_key_base *base = nla_data(a);
1312 
1313 			flags = base->flags;
1314 			ttl = base->ttl;
1315 			nh->np = base->np;
1316 			nh->mdtype = base->mdtype;
1317 			nh->path_hdr = base->path_hdr;
1318 			break;
1319 		}
1320 		case OVS_NSH_KEY_ATTR_MD1:
1321 			mdlen = nla_len(a);
1322 			if (mdlen > size - NSH_BASE_HDR_LEN)
1323 				return -ENOBUFS;
1324 			memcpy(&nh->md1, nla_data(a), mdlen);
1325 			break;
1326 
1327 		case OVS_NSH_KEY_ATTR_MD2:
1328 			mdlen = nla_len(a);
1329 			if (mdlen > size - NSH_BASE_HDR_LEN)
1330 				return -ENOBUFS;
1331 			memcpy(&nh->md2, nla_data(a), mdlen);
1332 			break;
1333 
1334 		default:
1335 			return -EINVAL;
1336 		}
1337 	}
1338 
1339 	/* nsh header length  = NSH_BASE_HDR_LEN + mdlen */
1340 	nh->ver_flags_ttl_len = 0;
1341 	nsh_set_flags_ttl_len(nh, flags, ttl, NSH_BASE_HDR_LEN + mdlen);
1342 
1343 	return 0;
1344 }
1345 
1346 int nsh_key_from_nlattr(const struct nlattr *attr,
1347 			struct ovs_key_nsh *nsh, struct ovs_key_nsh *nsh_mask)
1348 {
1349 	struct nlattr *a;
1350 	int rem;
1351 
1352 	/* validate_nsh has check this, so we needn't do duplicate check here
1353 	 */
1354 	nla_for_each_nested(a, attr, rem) {
1355 		int type = nla_type(a);
1356 
1357 		switch (type) {
1358 		case OVS_NSH_KEY_ATTR_BASE: {
1359 			const struct ovs_nsh_key_base *base = nla_data(a);
1360 			const struct ovs_nsh_key_base *base_mask = base + 1;
1361 
1362 			nsh->base = *base;
1363 			nsh_mask->base = *base_mask;
1364 			break;
1365 		}
1366 		case OVS_NSH_KEY_ATTR_MD1: {
1367 			const struct ovs_nsh_key_md1 *md1 = nla_data(a);
1368 			const struct ovs_nsh_key_md1 *md1_mask = md1 + 1;
1369 
1370 			memcpy(nsh->context, md1->context, sizeof(*md1));
1371 			memcpy(nsh_mask->context, md1_mask->context,
1372 			       sizeof(*md1_mask));
1373 			break;
1374 		}
1375 		case OVS_NSH_KEY_ATTR_MD2:
1376 			/* Not supported yet */
1377 			return -ENOTSUPP;
1378 		default:
1379 			return -EINVAL;
1380 		}
1381 	}
1382 
1383 	return 0;
1384 }
1385 
1386 static int nsh_key_put_from_nlattr(const struct nlattr *attr,
1387 				   struct sw_flow_match *match, bool is_mask,
1388 				   bool is_push_nsh, bool log)
1389 {
1390 	struct nlattr *a;
1391 	int rem;
1392 	bool has_base = false;
1393 	bool has_md1 = false;
1394 	bool has_md2 = false;
1395 	u8 mdtype = 0;
1396 	int mdlen = 0;
1397 
1398 	if (WARN_ON(is_push_nsh && is_mask))
1399 		return -EINVAL;
1400 
1401 	nla_for_each_nested(a, attr, rem) {
1402 		int type = nla_type(a);
1403 		int i;
1404 
1405 		if (type > OVS_NSH_KEY_ATTR_MAX) {
1406 			OVS_NLERR(log, "nsh attr %d is out of range max %d",
1407 				  type, OVS_NSH_KEY_ATTR_MAX);
1408 			return -EINVAL;
1409 		}
1410 
1411 		if (!check_attr_len(nla_len(a),
1412 				    ovs_nsh_key_attr_lens[type].len)) {
1413 			OVS_NLERR(
1414 			    log,
1415 			    "nsh attr %d has unexpected len %d expected %d",
1416 			    type,
1417 			    nla_len(a),
1418 			    ovs_nsh_key_attr_lens[type].len
1419 			);
1420 			return -EINVAL;
1421 		}
1422 
1423 		switch (type) {
1424 		case OVS_NSH_KEY_ATTR_BASE: {
1425 			const struct ovs_nsh_key_base *base = nla_data(a);
1426 
1427 			has_base = true;
1428 			mdtype = base->mdtype;
1429 			SW_FLOW_KEY_PUT(match, nsh.base.flags,
1430 					base->flags, is_mask);
1431 			SW_FLOW_KEY_PUT(match, nsh.base.ttl,
1432 					base->ttl, is_mask);
1433 			SW_FLOW_KEY_PUT(match, nsh.base.mdtype,
1434 					base->mdtype, is_mask);
1435 			SW_FLOW_KEY_PUT(match, nsh.base.np,
1436 					base->np, is_mask);
1437 			SW_FLOW_KEY_PUT(match, nsh.base.path_hdr,
1438 					base->path_hdr, is_mask);
1439 			break;
1440 		}
1441 		case OVS_NSH_KEY_ATTR_MD1: {
1442 			const struct ovs_nsh_key_md1 *md1 = nla_data(a);
1443 
1444 			has_md1 = true;
1445 			for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++)
1446 				SW_FLOW_KEY_PUT(match, nsh.context[i],
1447 						md1->context[i], is_mask);
1448 			break;
1449 		}
1450 		case OVS_NSH_KEY_ATTR_MD2:
1451 			if (!is_push_nsh) /* Not supported MD type 2 yet */
1452 				return -ENOTSUPP;
1453 
1454 			has_md2 = true;
1455 			mdlen = nla_len(a);
1456 			if (mdlen > NSH_CTX_HDRS_MAX_LEN || mdlen <= 0) {
1457 				OVS_NLERR(
1458 				    log,
1459 				    "Invalid MD length %d for MD type %d",
1460 				    mdlen,
1461 				    mdtype
1462 				);
1463 				return -EINVAL;
1464 			}
1465 			break;
1466 		default:
1467 			OVS_NLERR(log, "Unknown nsh attribute %d",
1468 				  type);
1469 			return -EINVAL;
1470 		}
1471 	}
1472 
1473 	if (rem > 0) {
1474 		OVS_NLERR(log, "nsh attribute has %d unknown bytes.", rem);
1475 		return -EINVAL;
1476 	}
1477 
1478 	if (has_md1 && has_md2) {
1479 		OVS_NLERR(
1480 		    1,
1481 		    "invalid nsh attribute: md1 and md2 are exclusive."
1482 		);
1483 		return -EINVAL;
1484 	}
1485 
1486 	if (!is_mask) {
1487 		if ((has_md1 && mdtype != NSH_M_TYPE1) ||
1488 		    (has_md2 && mdtype != NSH_M_TYPE2)) {
1489 			OVS_NLERR(1, "nsh attribute has unmatched MD type %d.",
1490 				  mdtype);
1491 			return -EINVAL;
1492 		}
1493 
1494 		if (is_push_nsh &&
1495 		    (!has_base || (!has_md1 && !has_md2))) {
1496 			OVS_NLERR(
1497 			    1,
1498 			    "push_nsh: missing base or metadata attributes"
1499 			);
1500 			return -EINVAL;
1501 		}
1502 	}
1503 
1504 	return 0;
1505 }
1506 
1507 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
1508 				u64 attrs, const struct nlattr **a,
1509 				bool is_mask, bool log)
1510 {
1511 	int err;
1512 
1513 	err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
1514 	if (err)
1515 		return err;
1516 
1517 	if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
1518 		const struct ovs_key_ethernet *eth_key;
1519 
1520 		eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1521 		SW_FLOW_KEY_MEMCPY(match, eth.src,
1522 				eth_key->eth_src, ETH_ALEN, is_mask);
1523 		SW_FLOW_KEY_MEMCPY(match, eth.dst,
1524 				eth_key->eth_dst, ETH_ALEN, is_mask);
1525 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1526 
1527 		if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
1528 			/* VLAN attribute is always parsed before getting here since it
1529 			 * may occur multiple times.
1530 			 */
1531 			OVS_NLERR(log, "VLAN attribute unexpected.");
1532 			return -EINVAL;
1533 		}
1534 
1535 		if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1536 			err = parse_eth_type_from_nlattrs(match, &attrs, a, is_mask,
1537 							  log);
1538 			if (err)
1539 				return err;
1540 		} else if (!is_mask) {
1541 			SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1542 		}
1543 	} else if (!match->key->eth.type) {
1544 		OVS_NLERR(log, "Either Ethernet header or EtherType is required.");
1545 		return -EINVAL;
1546 	}
1547 
1548 	if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1549 		const struct ovs_key_ipv4 *ipv4_key;
1550 
1551 		ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1552 		if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1553 			OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
1554 				  ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1555 			return -EINVAL;
1556 		}
1557 		SW_FLOW_KEY_PUT(match, ip.proto,
1558 				ipv4_key->ipv4_proto, is_mask);
1559 		SW_FLOW_KEY_PUT(match, ip.tos,
1560 				ipv4_key->ipv4_tos, is_mask);
1561 		SW_FLOW_KEY_PUT(match, ip.ttl,
1562 				ipv4_key->ipv4_ttl, is_mask);
1563 		SW_FLOW_KEY_PUT(match, ip.frag,
1564 				ipv4_key->ipv4_frag, is_mask);
1565 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1566 				ipv4_key->ipv4_src, is_mask);
1567 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1568 				ipv4_key->ipv4_dst, is_mask);
1569 		attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1570 	}
1571 
1572 	if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
1573 		const struct ovs_key_ipv6 *ipv6_key;
1574 
1575 		ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1576 		if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1577 			OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
1578 				  ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1579 			return -EINVAL;
1580 		}
1581 
1582 		if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
1583 			OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x)",
1584 				  ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
1585 			return -EINVAL;
1586 		}
1587 
1588 		SW_FLOW_KEY_PUT(match, ipv6.label,
1589 				ipv6_key->ipv6_label, is_mask);
1590 		SW_FLOW_KEY_PUT(match, ip.proto,
1591 				ipv6_key->ipv6_proto, is_mask);
1592 		SW_FLOW_KEY_PUT(match, ip.tos,
1593 				ipv6_key->ipv6_tclass, is_mask);
1594 		SW_FLOW_KEY_PUT(match, ip.ttl,
1595 				ipv6_key->ipv6_hlimit, is_mask);
1596 		SW_FLOW_KEY_PUT(match, ip.frag,
1597 				ipv6_key->ipv6_frag, is_mask);
1598 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1599 				ipv6_key->ipv6_src,
1600 				sizeof(match->key->ipv6.addr.src),
1601 				is_mask);
1602 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1603 				ipv6_key->ipv6_dst,
1604 				sizeof(match->key->ipv6.addr.dst),
1605 				is_mask);
1606 
1607 		attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1608 	}
1609 
1610 	if (attrs & (1ULL << OVS_KEY_ATTR_IPV6_EXTHDRS)) {
1611 		const struct ovs_key_ipv6_exthdrs *ipv6_exthdrs_key;
1612 
1613 		ipv6_exthdrs_key = nla_data(a[OVS_KEY_ATTR_IPV6_EXTHDRS]);
1614 
1615 		SW_FLOW_KEY_PUT(match, ipv6.exthdrs,
1616 				ipv6_exthdrs_key->hdrs, is_mask);
1617 
1618 		attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6_EXTHDRS);
1619 	}
1620 
1621 	if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1622 		const struct ovs_key_arp *arp_key;
1623 
1624 		arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1625 		if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1626 			OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1627 				  arp_key->arp_op);
1628 			return -EINVAL;
1629 		}
1630 
1631 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1632 				arp_key->arp_sip, is_mask);
1633 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1634 			arp_key->arp_tip, is_mask);
1635 		SW_FLOW_KEY_PUT(match, ip.proto,
1636 				ntohs(arp_key->arp_op), is_mask);
1637 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1638 				arp_key->arp_sha, ETH_ALEN, is_mask);
1639 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1640 				arp_key->arp_tha, ETH_ALEN, is_mask);
1641 
1642 		attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1643 	}
1644 
1645 	if (attrs & (1 << OVS_KEY_ATTR_NSH)) {
1646 		if (nsh_key_put_from_nlattr(a[OVS_KEY_ATTR_NSH], match,
1647 					    is_mask, false, log) < 0)
1648 			return -EINVAL;
1649 		attrs &= ~(1 << OVS_KEY_ATTR_NSH);
1650 	}
1651 
1652 	if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1653 		const struct ovs_key_mpls *mpls_key;
1654 		u32 hdr_len;
1655 		u32 label_count, label_count_mask, i;
1656 
1657 		mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1658 		hdr_len = nla_len(a[OVS_KEY_ATTR_MPLS]);
1659 		label_count = hdr_len / sizeof(struct ovs_key_mpls);
1660 
1661 		if (label_count == 0 || label_count > MPLS_LABEL_DEPTH ||
1662 		    hdr_len % sizeof(struct ovs_key_mpls))
1663 			return -EINVAL;
1664 
1665 		label_count_mask =  GENMASK(label_count - 1, 0);
1666 
1667 		for (i = 0 ; i < label_count; i++)
1668 			SW_FLOW_KEY_PUT(match, mpls.lse[i],
1669 					mpls_key[i].mpls_lse, is_mask);
1670 
1671 		SW_FLOW_KEY_PUT(match, mpls.num_labels_mask,
1672 				label_count_mask, is_mask);
1673 
1674 		attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1675 	 }
1676 
1677 	if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1678 		const struct ovs_key_tcp *tcp_key;
1679 
1680 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1681 		SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1682 		SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1683 		attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1684 	}
1685 
1686 	if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1687 		SW_FLOW_KEY_PUT(match, tp.flags,
1688 				nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1689 				is_mask);
1690 		attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1691 	}
1692 
1693 	if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1694 		const struct ovs_key_udp *udp_key;
1695 
1696 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1697 		SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1698 		SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1699 		attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1700 	}
1701 
1702 	if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1703 		const struct ovs_key_sctp *sctp_key;
1704 
1705 		sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1706 		SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1707 		SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1708 		attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1709 	}
1710 
1711 	if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1712 		const struct ovs_key_icmp *icmp_key;
1713 
1714 		icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1715 		SW_FLOW_KEY_PUT(match, tp.src,
1716 				htons(icmp_key->icmp_type), is_mask);
1717 		SW_FLOW_KEY_PUT(match, tp.dst,
1718 				htons(icmp_key->icmp_code), is_mask);
1719 		attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1720 	}
1721 
1722 	if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1723 		const struct ovs_key_icmpv6 *icmpv6_key;
1724 
1725 		icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1726 		SW_FLOW_KEY_PUT(match, tp.src,
1727 				htons(icmpv6_key->icmpv6_type), is_mask);
1728 		SW_FLOW_KEY_PUT(match, tp.dst,
1729 				htons(icmpv6_key->icmpv6_code), is_mask);
1730 		attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1731 	}
1732 
1733 	if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1734 		const struct ovs_key_nd *nd_key;
1735 
1736 		nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1737 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1738 			nd_key->nd_target,
1739 			sizeof(match->key->ipv6.nd.target),
1740 			is_mask);
1741 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1742 			nd_key->nd_sll, ETH_ALEN, is_mask);
1743 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1744 				nd_key->nd_tll, ETH_ALEN, is_mask);
1745 		attrs &= ~(1 << OVS_KEY_ATTR_ND);
1746 	}
1747 
1748 	if (attrs != 0) {
1749 		OVS_NLERR(log, "Unknown key attributes %llx",
1750 			  (unsigned long long)attrs);
1751 		return -EINVAL;
1752 	}
1753 
1754 	return 0;
1755 }
1756 
1757 static void nlattr_set(struct nlattr *attr, u8 val,
1758 		       const struct ovs_len_tbl *tbl)
1759 {
1760 	struct nlattr *nla;
1761 	int rem;
1762 
1763 	/* The nlattr stream should already have been validated */
1764 	nla_for_each_nested(nla, attr, rem) {
1765 		if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
1766 			nlattr_set(nla, val, tbl[nla_type(nla)].next ? : tbl);
1767 		else
1768 			memset(nla_data(nla), val, nla_len(nla));
1769 
1770 		if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1771 			*(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1772 	}
1773 }
1774 
1775 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1776 {
1777 	nlattr_set(attr, val, ovs_key_lens);
1778 }
1779 
1780 /**
1781  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1782  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1783  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1784  * does not include any don't care bit.
1785  * @net: Used to determine per-namespace field support.
1786  * @match: receives the extracted flow match information.
1787  * @nla_key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1788  * sequence. The fields should of the packet that triggered the creation
1789  * of this flow.
1790  * @nla_mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_*
1791  * Netlink attribute specifies the mask field of the wildcarded flow.
1792  * @log: Boolean to allow kernel error logging.  Normally true, but when
1793  * probing for feature compatibility this should be passed in as false to
1794  * suppress unnecessary error logging.
1795  */
1796 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1797 		      const struct nlattr *nla_key,
1798 		      const struct nlattr *nla_mask,
1799 		      bool log)
1800 {
1801 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1802 	struct nlattr *newmask = NULL;
1803 	u64 key_attrs = 0;
1804 	u64 mask_attrs = 0;
1805 	int err;
1806 
1807 	err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1808 	if (err)
1809 		return err;
1810 
1811 	err = parse_vlan_from_nlattrs(match, &key_attrs, a, false, log);
1812 	if (err)
1813 		return err;
1814 
1815 	err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1816 	if (err)
1817 		return err;
1818 
1819 	if (match->mask) {
1820 		if (!nla_mask) {
1821 			/* Create an exact match mask. We need to set to 0xff
1822 			 * all the 'match->mask' fields that have been touched
1823 			 * in 'match->key'. We cannot simply memset
1824 			 * 'match->mask', because padding bytes and fields not
1825 			 * specified in 'match->key' should be left to 0.
1826 			 * Instead, we use a stream of netlink attributes,
1827 			 * copied from 'key' and set to 0xff.
1828 			 * ovs_key_from_nlattrs() will take care of filling
1829 			 * 'match->mask' appropriately.
1830 			 */
1831 			newmask = kmemdup(nla_key,
1832 					  nla_total_size(nla_len(nla_key)),
1833 					  GFP_KERNEL);
1834 			if (!newmask)
1835 				return -ENOMEM;
1836 
1837 			mask_set_nlattr(newmask, 0xff);
1838 
1839 			/* The userspace does not send tunnel attributes that
1840 			 * are 0, but we should not wildcard them nonetheless.
1841 			 */
1842 			if (match->key->tun_proto)
1843 				SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1844 							 0xff, true);
1845 
1846 			nla_mask = newmask;
1847 		}
1848 
1849 		err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1850 		if (err)
1851 			goto free_newmask;
1852 
1853 		/* Always match on tci. */
1854 		SW_FLOW_KEY_PUT(match, eth.vlan.tci, htons(0xffff), true);
1855 		SW_FLOW_KEY_PUT(match, eth.cvlan.tci, htons(0xffff), true);
1856 
1857 		err = parse_vlan_from_nlattrs(match, &mask_attrs, a, true, log);
1858 		if (err)
1859 			goto free_newmask;
1860 
1861 		err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1862 					   log);
1863 		if (err)
1864 			goto free_newmask;
1865 	}
1866 
1867 	if (!match_validate(match, key_attrs, mask_attrs, log))
1868 		err = -EINVAL;
1869 
1870 free_newmask:
1871 	kfree(newmask);
1872 	return err;
1873 }
1874 
1875 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1876 {
1877 	size_t len;
1878 
1879 	if (!attr)
1880 		return 0;
1881 
1882 	len = nla_len(attr);
1883 	if (len < 1 || len > MAX_UFID_LENGTH) {
1884 		OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1885 			  nla_len(attr), MAX_UFID_LENGTH);
1886 		return 0;
1887 	}
1888 
1889 	return len;
1890 }
1891 
1892 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1893  * or false otherwise.
1894  */
1895 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1896 		      bool log)
1897 {
1898 	sfid->ufid_len = get_ufid_len(attr, log);
1899 	if (sfid->ufid_len)
1900 		memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1901 
1902 	return sfid->ufid_len;
1903 }
1904 
1905 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1906 			   const struct sw_flow_key *key, bool log)
1907 {
1908 	struct sw_flow_key *new_key;
1909 
1910 	if (ovs_nla_get_ufid(sfid, ufid, log))
1911 		return 0;
1912 
1913 	/* If UFID was not provided, use unmasked key. */
1914 	new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1915 	if (!new_key)
1916 		return -ENOMEM;
1917 	memcpy(new_key, key, sizeof(*key));
1918 	sfid->unmasked_key = new_key;
1919 
1920 	return 0;
1921 }
1922 
1923 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1924 {
1925 	return attr ? nla_get_u32(attr) : 0;
1926 }
1927 
1928 /**
1929  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1930  * @net: Network namespace.
1931  * @key: Receives extracted in_port, priority, tun_key, skb_mark and conntrack
1932  * metadata.
1933  * @a: Array of netlink attributes holding parsed %OVS_KEY_ATTR_* Netlink
1934  * attributes.
1935  * @attrs: Bit mask for the netlink attributes included in @a.
1936  * @log: Boolean to allow kernel error logging.  Normally true, but when
1937  * probing for feature compatibility this should be passed in as false to
1938  * suppress unnecessary error logging.
1939  *
1940  * This parses a series of Netlink attributes that form a flow key, which must
1941  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1942  * get the metadata, that is, the parts of the flow key that cannot be
1943  * extracted from the packet itself.
1944  *
1945  * This must be called before the packet key fields are filled in 'key'.
1946  */
1947 
1948 int ovs_nla_get_flow_metadata(struct net *net,
1949 			      const struct nlattr *a[OVS_KEY_ATTR_MAX + 1],
1950 			      u64 attrs, struct sw_flow_key *key, bool log)
1951 {
1952 	struct sw_flow_match match;
1953 
1954 	memset(&match, 0, sizeof(match));
1955 	match.key = key;
1956 
1957 	key->ct_state = 0;
1958 	key->ct_zone = 0;
1959 	key->ct_orig_proto = 0;
1960 	memset(&key->ct, 0, sizeof(key->ct));
1961 	memset(&key->ipv4.ct_orig, 0, sizeof(key->ipv4.ct_orig));
1962 	memset(&key->ipv6.ct_orig, 0, sizeof(key->ipv6.ct_orig));
1963 
1964 	key->phy.in_port = DP_MAX_PORTS;
1965 
1966 	return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1967 }
1968 
1969 static int ovs_nla_put_vlan(struct sk_buff *skb, const struct vlan_head *vh,
1970 			    bool is_mask)
1971 {
1972 	__be16 eth_type = !is_mask ? vh->tpid : htons(0xffff);
1973 
1974 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1975 	    nla_put_be16(skb, OVS_KEY_ATTR_VLAN, vh->tci))
1976 		return -EMSGSIZE;
1977 	return 0;
1978 }
1979 
1980 static int nsh_key_to_nlattr(const struct ovs_key_nsh *nsh, bool is_mask,
1981 			     struct sk_buff *skb)
1982 {
1983 	struct nlattr *start;
1984 
1985 	start = nla_nest_start_noflag(skb, OVS_KEY_ATTR_NSH);
1986 	if (!start)
1987 		return -EMSGSIZE;
1988 
1989 	if (nla_put(skb, OVS_NSH_KEY_ATTR_BASE, sizeof(nsh->base), &nsh->base))
1990 		goto nla_put_failure;
1991 
1992 	if (is_mask || nsh->base.mdtype == NSH_M_TYPE1) {
1993 		if (nla_put(skb, OVS_NSH_KEY_ATTR_MD1,
1994 			    sizeof(nsh->context), nsh->context))
1995 			goto nla_put_failure;
1996 	}
1997 
1998 	/* Don't support MD type 2 yet */
1999 
2000 	nla_nest_end(skb, start);
2001 
2002 	return 0;
2003 
2004 nla_put_failure:
2005 	return -EMSGSIZE;
2006 }
2007 
2008 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
2009 			     const struct sw_flow_key *output, bool is_mask,
2010 			     struct sk_buff *skb)
2011 {
2012 	struct ovs_key_ethernet *eth_key;
2013 	struct nlattr *nla;
2014 	struct nlattr *encap = NULL;
2015 	struct nlattr *in_encap = NULL;
2016 
2017 	if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
2018 		goto nla_put_failure;
2019 
2020 	if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
2021 		goto nla_put_failure;
2022 
2023 	if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
2024 		goto nla_put_failure;
2025 
2026 	if ((swkey->tun_proto || is_mask)) {
2027 		const void *opts = NULL;
2028 
2029 		if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
2030 			opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
2031 
2032 		if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
2033 				     swkey->tun_opts_len, swkey->tun_proto, 0))
2034 			goto nla_put_failure;
2035 	}
2036 
2037 	if (swkey->phy.in_port == DP_MAX_PORTS) {
2038 		if (is_mask && (output->phy.in_port == 0xffff))
2039 			if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
2040 				goto nla_put_failure;
2041 	} else {
2042 		u16 upper_u16;
2043 		upper_u16 = !is_mask ? 0 : 0xffff;
2044 
2045 		if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
2046 				(upper_u16 << 16) | output->phy.in_port))
2047 			goto nla_put_failure;
2048 	}
2049 
2050 	if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
2051 		goto nla_put_failure;
2052 
2053 	if (ovs_ct_put_key(swkey, output, skb))
2054 		goto nla_put_failure;
2055 
2056 	if (ovs_key_mac_proto(swkey) == MAC_PROTO_ETHERNET) {
2057 		nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
2058 		if (!nla)
2059 			goto nla_put_failure;
2060 
2061 		eth_key = nla_data(nla);
2062 		ether_addr_copy(eth_key->eth_src, output->eth.src);
2063 		ether_addr_copy(eth_key->eth_dst, output->eth.dst);
2064 
2065 		if (swkey->eth.vlan.tci || eth_type_vlan(swkey->eth.type)) {
2066 			if (ovs_nla_put_vlan(skb, &output->eth.vlan, is_mask))
2067 				goto nla_put_failure;
2068 			encap = nla_nest_start_noflag(skb, OVS_KEY_ATTR_ENCAP);
2069 			if (!swkey->eth.vlan.tci)
2070 				goto unencap;
2071 
2072 			if (swkey->eth.cvlan.tci || eth_type_vlan(swkey->eth.type)) {
2073 				if (ovs_nla_put_vlan(skb, &output->eth.cvlan, is_mask))
2074 					goto nla_put_failure;
2075 				in_encap = nla_nest_start_noflag(skb,
2076 								 OVS_KEY_ATTR_ENCAP);
2077 				if (!swkey->eth.cvlan.tci)
2078 					goto unencap;
2079 			}
2080 		}
2081 
2082 		if (swkey->eth.type == htons(ETH_P_802_2)) {
2083 			/*
2084 			* Ethertype 802.2 is represented in the netlink with omitted
2085 			* OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
2086 			* 0xffff in the mask attribute.  Ethertype can also
2087 			* be wildcarded.
2088 			*/
2089 			if (is_mask && output->eth.type)
2090 				if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
2091 							output->eth.type))
2092 					goto nla_put_failure;
2093 			goto unencap;
2094 		}
2095 	}
2096 
2097 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
2098 		goto nla_put_failure;
2099 
2100 	if (eth_type_vlan(swkey->eth.type)) {
2101 		/* There are 3 VLAN tags, we don't know anything about the rest
2102 		 * of the packet, so truncate here.
2103 		 */
2104 		WARN_ON_ONCE(!(encap && in_encap));
2105 		goto unencap;
2106 	}
2107 
2108 	if (swkey->eth.type == htons(ETH_P_IP)) {
2109 		struct ovs_key_ipv4 *ipv4_key;
2110 
2111 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
2112 		if (!nla)
2113 			goto nla_put_failure;
2114 		ipv4_key = nla_data(nla);
2115 		ipv4_key->ipv4_src = output->ipv4.addr.src;
2116 		ipv4_key->ipv4_dst = output->ipv4.addr.dst;
2117 		ipv4_key->ipv4_proto = output->ip.proto;
2118 		ipv4_key->ipv4_tos = output->ip.tos;
2119 		ipv4_key->ipv4_ttl = output->ip.ttl;
2120 		ipv4_key->ipv4_frag = output->ip.frag;
2121 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
2122 		struct ovs_key_ipv6 *ipv6_key;
2123 		struct ovs_key_ipv6_exthdrs *ipv6_exthdrs_key;
2124 
2125 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
2126 		if (!nla)
2127 			goto nla_put_failure;
2128 		ipv6_key = nla_data(nla);
2129 		memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
2130 				sizeof(ipv6_key->ipv6_src));
2131 		memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
2132 				sizeof(ipv6_key->ipv6_dst));
2133 		ipv6_key->ipv6_label = output->ipv6.label;
2134 		ipv6_key->ipv6_proto = output->ip.proto;
2135 		ipv6_key->ipv6_tclass = output->ip.tos;
2136 		ipv6_key->ipv6_hlimit = output->ip.ttl;
2137 		ipv6_key->ipv6_frag = output->ip.frag;
2138 
2139 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6_EXTHDRS,
2140 				  sizeof(*ipv6_exthdrs_key));
2141 		if (!nla)
2142 			goto nla_put_failure;
2143 		ipv6_exthdrs_key = nla_data(nla);
2144 		ipv6_exthdrs_key->hdrs = output->ipv6.exthdrs;
2145 	} else if (swkey->eth.type == htons(ETH_P_NSH)) {
2146 		if (nsh_key_to_nlattr(&output->nsh, is_mask, skb))
2147 			goto nla_put_failure;
2148 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
2149 		   swkey->eth.type == htons(ETH_P_RARP)) {
2150 		struct ovs_key_arp *arp_key;
2151 
2152 		nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
2153 		if (!nla)
2154 			goto nla_put_failure;
2155 		arp_key = nla_data(nla);
2156 		memset(arp_key, 0, sizeof(struct ovs_key_arp));
2157 		arp_key->arp_sip = output->ipv4.addr.src;
2158 		arp_key->arp_tip = output->ipv4.addr.dst;
2159 		arp_key->arp_op = htons(output->ip.proto);
2160 		ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
2161 		ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
2162 	} else if (eth_p_mpls(swkey->eth.type)) {
2163 		u8 i, num_labels;
2164 		struct ovs_key_mpls *mpls_key;
2165 
2166 		num_labels = hweight_long(output->mpls.num_labels_mask);
2167 		nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS,
2168 				  num_labels * sizeof(*mpls_key));
2169 		if (!nla)
2170 			goto nla_put_failure;
2171 
2172 		mpls_key = nla_data(nla);
2173 		for (i = 0; i < num_labels; i++)
2174 			mpls_key[i].mpls_lse = output->mpls.lse[i];
2175 	}
2176 
2177 	if ((swkey->eth.type == htons(ETH_P_IP) ||
2178 	     swkey->eth.type == htons(ETH_P_IPV6)) &&
2179 	     swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
2180 
2181 		if (swkey->ip.proto == IPPROTO_TCP) {
2182 			struct ovs_key_tcp *tcp_key;
2183 
2184 			nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
2185 			if (!nla)
2186 				goto nla_put_failure;
2187 			tcp_key = nla_data(nla);
2188 			tcp_key->tcp_src = output->tp.src;
2189 			tcp_key->tcp_dst = output->tp.dst;
2190 			if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
2191 					 output->tp.flags))
2192 				goto nla_put_failure;
2193 		} else if (swkey->ip.proto == IPPROTO_UDP) {
2194 			struct ovs_key_udp *udp_key;
2195 
2196 			nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
2197 			if (!nla)
2198 				goto nla_put_failure;
2199 			udp_key = nla_data(nla);
2200 			udp_key->udp_src = output->tp.src;
2201 			udp_key->udp_dst = output->tp.dst;
2202 		} else if (swkey->ip.proto == IPPROTO_SCTP) {
2203 			struct ovs_key_sctp *sctp_key;
2204 
2205 			nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
2206 			if (!nla)
2207 				goto nla_put_failure;
2208 			sctp_key = nla_data(nla);
2209 			sctp_key->sctp_src = output->tp.src;
2210 			sctp_key->sctp_dst = output->tp.dst;
2211 		} else if (swkey->eth.type == htons(ETH_P_IP) &&
2212 			   swkey->ip.proto == IPPROTO_ICMP) {
2213 			struct ovs_key_icmp *icmp_key;
2214 
2215 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
2216 			if (!nla)
2217 				goto nla_put_failure;
2218 			icmp_key = nla_data(nla);
2219 			icmp_key->icmp_type = ntohs(output->tp.src);
2220 			icmp_key->icmp_code = ntohs(output->tp.dst);
2221 		} else if (swkey->eth.type == htons(ETH_P_IPV6) &&
2222 			   swkey->ip.proto == IPPROTO_ICMPV6) {
2223 			struct ovs_key_icmpv6 *icmpv6_key;
2224 
2225 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
2226 						sizeof(*icmpv6_key));
2227 			if (!nla)
2228 				goto nla_put_failure;
2229 			icmpv6_key = nla_data(nla);
2230 			icmpv6_key->icmpv6_type = ntohs(output->tp.src);
2231 			icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
2232 
2233 			if (swkey->tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
2234 			    swkey->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
2235 				struct ovs_key_nd *nd_key;
2236 
2237 				nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
2238 				if (!nla)
2239 					goto nla_put_failure;
2240 				nd_key = nla_data(nla);
2241 				memcpy(nd_key->nd_target, &output->ipv6.nd.target,
2242 							sizeof(nd_key->nd_target));
2243 				ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
2244 				ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
2245 			}
2246 		}
2247 	}
2248 
2249 unencap:
2250 	if (in_encap)
2251 		nla_nest_end(skb, in_encap);
2252 	if (encap)
2253 		nla_nest_end(skb, encap);
2254 
2255 	return 0;
2256 
2257 nla_put_failure:
2258 	return -EMSGSIZE;
2259 }
2260 
2261 int ovs_nla_put_key(const struct sw_flow_key *swkey,
2262 		    const struct sw_flow_key *output, int attr, bool is_mask,
2263 		    struct sk_buff *skb)
2264 {
2265 	int err;
2266 	struct nlattr *nla;
2267 
2268 	nla = nla_nest_start_noflag(skb, attr);
2269 	if (!nla)
2270 		return -EMSGSIZE;
2271 	err = __ovs_nla_put_key(swkey, output, is_mask, skb);
2272 	if (err)
2273 		return err;
2274 	nla_nest_end(skb, nla);
2275 
2276 	return 0;
2277 }
2278 
2279 /* Called with ovs_mutex or RCU read lock. */
2280 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
2281 {
2282 	if (ovs_identifier_is_ufid(&flow->id))
2283 		return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
2284 			       flow->id.ufid);
2285 
2286 	return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
2287 			       OVS_FLOW_ATTR_KEY, false, skb);
2288 }
2289 
2290 /* Called with ovs_mutex or RCU read lock. */
2291 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
2292 {
2293 	return ovs_nla_put_key(&flow->key, &flow->key,
2294 				OVS_FLOW_ATTR_KEY, false, skb);
2295 }
2296 
2297 /* Called with ovs_mutex or RCU read lock. */
2298 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
2299 {
2300 	return ovs_nla_put_key(&flow->key, &flow->mask->key,
2301 				OVS_FLOW_ATTR_MASK, true, skb);
2302 }
2303 
2304 #define MAX_ACTIONS_BUFSIZE	(32 * 1024)
2305 
2306 static struct sw_flow_actions *nla_alloc_flow_actions(int size)
2307 {
2308 	struct sw_flow_actions *sfa;
2309 
2310 	WARN_ON_ONCE(size > MAX_ACTIONS_BUFSIZE);
2311 
2312 	sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
2313 	if (!sfa)
2314 		return ERR_PTR(-ENOMEM);
2315 
2316 	sfa->actions_len = 0;
2317 	return sfa;
2318 }
2319 
2320 static void ovs_nla_free_set_action(const struct nlattr *a)
2321 {
2322 	const struct nlattr *ovs_key = nla_data(a);
2323 	struct ovs_tunnel_info *ovs_tun;
2324 
2325 	switch (nla_type(ovs_key)) {
2326 	case OVS_KEY_ATTR_TUNNEL_INFO:
2327 		ovs_tun = nla_data(ovs_key);
2328 		dst_release((struct dst_entry *)ovs_tun->tun_dst);
2329 		break;
2330 	}
2331 }
2332 
2333 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
2334 {
2335 	const struct nlattr *a;
2336 	int rem;
2337 
2338 	if (!sf_acts)
2339 		return;
2340 
2341 	nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
2342 		switch (nla_type(a)) {
2343 		case OVS_ACTION_ATTR_SET:
2344 			ovs_nla_free_set_action(a);
2345 			break;
2346 		case OVS_ACTION_ATTR_CT:
2347 			ovs_ct_free_action(a);
2348 			break;
2349 		}
2350 	}
2351 
2352 	kfree(sf_acts);
2353 }
2354 
2355 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
2356 {
2357 	ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
2358 }
2359 
2360 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
2361  * The caller must hold rcu_read_lock for this to be sensible. */
2362 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
2363 {
2364 	call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
2365 }
2366 
2367 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
2368 				       int attr_len, bool log)
2369 {
2370 
2371 	struct sw_flow_actions *acts;
2372 	int new_acts_size;
2373 	size_t req_size = NLA_ALIGN(attr_len);
2374 	int next_offset = offsetof(struct sw_flow_actions, actions) +
2375 					(*sfa)->actions_len;
2376 
2377 	if (req_size <= (ksize(*sfa) - next_offset))
2378 		goto out;
2379 
2380 	new_acts_size = max(next_offset + req_size, ksize(*sfa) * 2);
2381 
2382 	if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
2383 		if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size) {
2384 			OVS_NLERR(log, "Flow action size exceeds max %u",
2385 				  MAX_ACTIONS_BUFSIZE);
2386 			return ERR_PTR(-EMSGSIZE);
2387 		}
2388 		new_acts_size = MAX_ACTIONS_BUFSIZE;
2389 	}
2390 
2391 	acts = nla_alloc_flow_actions(new_acts_size);
2392 	if (IS_ERR(acts))
2393 		return (void *)acts;
2394 
2395 	memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
2396 	acts->actions_len = (*sfa)->actions_len;
2397 	acts->orig_len = (*sfa)->orig_len;
2398 	kfree(*sfa);
2399 	*sfa = acts;
2400 
2401 out:
2402 	(*sfa)->actions_len += req_size;
2403 	return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
2404 }
2405 
2406 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
2407 				   int attrtype, void *data, int len, bool log)
2408 {
2409 	struct nlattr *a;
2410 
2411 	a = reserve_sfa_size(sfa, nla_attr_size(len), log);
2412 	if (IS_ERR(a))
2413 		return a;
2414 
2415 	a->nla_type = attrtype;
2416 	a->nla_len = nla_attr_size(len);
2417 
2418 	if (data)
2419 		memcpy(nla_data(a), data, len);
2420 	memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
2421 
2422 	return a;
2423 }
2424 
2425 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
2426 		       int len, bool log)
2427 {
2428 	struct nlattr *a;
2429 
2430 	a = __add_action(sfa, attrtype, data, len, log);
2431 
2432 	return PTR_ERR_OR_ZERO(a);
2433 }
2434 
2435 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
2436 					  int attrtype, bool log)
2437 {
2438 	int used = (*sfa)->actions_len;
2439 	int err;
2440 
2441 	err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
2442 	if (err)
2443 		return err;
2444 
2445 	return used;
2446 }
2447 
2448 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
2449 					 int st_offset)
2450 {
2451 	struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
2452 							       st_offset);
2453 
2454 	a->nla_len = sfa->actions_len - st_offset;
2455 }
2456 
2457 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2458 				  const struct sw_flow_key *key,
2459 				  struct sw_flow_actions **sfa,
2460 				  __be16 eth_type, __be16 vlan_tci,
2461 				  u32 mpls_label_count, bool log);
2462 
2463 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
2464 				    const struct sw_flow_key *key,
2465 				    struct sw_flow_actions **sfa,
2466 				    __be16 eth_type, __be16 vlan_tci,
2467 				    u32 mpls_label_count, bool log, bool last)
2468 {
2469 	const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
2470 	const struct nlattr *probability, *actions;
2471 	const struct nlattr *a;
2472 	int rem, start, err;
2473 	struct sample_arg arg;
2474 
2475 	memset(attrs, 0, sizeof(attrs));
2476 	nla_for_each_nested(a, attr, rem) {
2477 		int type = nla_type(a);
2478 		if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
2479 			return -EINVAL;
2480 		attrs[type] = a;
2481 	}
2482 	if (rem)
2483 		return -EINVAL;
2484 
2485 	probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
2486 	if (!probability || nla_len(probability) != sizeof(u32))
2487 		return -EINVAL;
2488 
2489 	actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
2490 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
2491 		return -EINVAL;
2492 
2493 	/* validation done, copy sample action. */
2494 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
2495 	if (start < 0)
2496 		return start;
2497 
2498 	/* When both skb and flow may be changed, put the sample
2499 	 * into a deferred fifo. On the other hand, if only skb
2500 	 * may be modified, the actions can be executed in place.
2501 	 *
2502 	 * Do this analysis at the flow installation time.
2503 	 * Set 'clone_action->exec' to true if the actions can be
2504 	 * executed without being deferred.
2505 	 *
2506 	 * If the sample is the last action, it can always be excuted
2507 	 * rather than deferred.
2508 	 */
2509 	arg.exec = last || !actions_may_change_flow(actions);
2510 	arg.probability = nla_get_u32(probability);
2511 
2512 	err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_ARG, &arg, sizeof(arg),
2513 				 log);
2514 	if (err)
2515 		return err;
2516 
2517 	err = __ovs_nla_copy_actions(net, actions, key, sfa,
2518 				     eth_type, vlan_tci, mpls_label_count, log);
2519 
2520 	if (err)
2521 		return err;
2522 
2523 	add_nested_action_end(*sfa, start);
2524 
2525 	return 0;
2526 }
2527 
2528 static int validate_and_copy_dec_ttl(struct net *net,
2529 				     const struct nlattr *attr,
2530 				     const struct sw_flow_key *key,
2531 				     struct sw_flow_actions **sfa,
2532 				     __be16 eth_type, __be16 vlan_tci,
2533 				     u32 mpls_label_count, bool log)
2534 {
2535 	const struct nlattr *attrs[OVS_DEC_TTL_ATTR_MAX + 1];
2536 	int start, action_start, err, rem;
2537 	const struct nlattr *a, *actions;
2538 
2539 	memset(attrs, 0, sizeof(attrs));
2540 	nla_for_each_nested(a, attr, rem) {
2541 		int type = nla_type(a);
2542 
2543 		/* Ignore unknown attributes to be future proof. */
2544 		if (type > OVS_DEC_TTL_ATTR_MAX)
2545 			continue;
2546 
2547 		if (!type || attrs[type]) {
2548 			OVS_NLERR(log, "Duplicate or invalid key (type %d).",
2549 				  type);
2550 			return -EINVAL;
2551 		}
2552 
2553 		attrs[type] = a;
2554 	}
2555 
2556 	if (rem) {
2557 		OVS_NLERR(log, "Message has %d unknown bytes.", rem);
2558 		return -EINVAL;
2559 	}
2560 
2561 	actions = attrs[OVS_DEC_TTL_ATTR_ACTION];
2562 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN)) {
2563 		OVS_NLERR(log, "Missing valid actions attribute.");
2564 		return -EINVAL;
2565 	}
2566 
2567 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_DEC_TTL, log);
2568 	if (start < 0)
2569 		return start;
2570 
2571 	action_start = add_nested_action_start(sfa, OVS_DEC_TTL_ATTR_ACTION, log);
2572 	if (action_start < 0)
2573 		return action_start;
2574 
2575 	err = __ovs_nla_copy_actions(net, actions, key, sfa, eth_type,
2576 				     vlan_tci, mpls_label_count, log);
2577 	if (err)
2578 		return err;
2579 
2580 	add_nested_action_end(*sfa, action_start);
2581 	add_nested_action_end(*sfa, start);
2582 	return 0;
2583 }
2584 
2585 static int validate_and_copy_clone(struct net *net,
2586 				   const struct nlattr *attr,
2587 				   const struct sw_flow_key *key,
2588 				   struct sw_flow_actions **sfa,
2589 				   __be16 eth_type, __be16 vlan_tci,
2590 				   u32 mpls_label_count, bool log, bool last)
2591 {
2592 	int start, err;
2593 	u32 exec;
2594 
2595 	if (nla_len(attr) && nla_len(attr) < NLA_HDRLEN)
2596 		return -EINVAL;
2597 
2598 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_CLONE, log);
2599 	if (start < 0)
2600 		return start;
2601 
2602 	exec = last || !actions_may_change_flow(attr);
2603 
2604 	err = ovs_nla_add_action(sfa, OVS_CLONE_ATTR_EXEC, &exec,
2605 				 sizeof(exec), log);
2606 	if (err)
2607 		return err;
2608 
2609 	err = __ovs_nla_copy_actions(net, attr, key, sfa,
2610 				     eth_type, vlan_tci, mpls_label_count, log);
2611 	if (err)
2612 		return err;
2613 
2614 	add_nested_action_end(*sfa, start);
2615 
2616 	return 0;
2617 }
2618 
2619 void ovs_match_init(struct sw_flow_match *match,
2620 		    struct sw_flow_key *key,
2621 		    bool reset_key,
2622 		    struct sw_flow_mask *mask)
2623 {
2624 	memset(match, 0, sizeof(*match));
2625 	match->key = key;
2626 	match->mask = mask;
2627 
2628 	if (reset_key)
2629 		memset(key, 0, sizeof(*key));
2630 
2631 	if (mask) {
2632 		memset(&mask->key, 0, sizeof(mask->key));
2633 		mask->range.start = mask->range.end = 0;
2634 	}
2635 }
2636 
2637 static int validate_geneve_opts(struct sw_flow_key *key)
2638 {
2639 	struct geneve_opt *option;
2640 	int opts_len = key->tun_opts_len;
2641 	bool crit_opt = false;
2642 
2643 	option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
2644 	while (opts_len > 0) {
2645 		int len;
2646 
2647 		if (opts_len < sizeof(*option))
2648 			return -EINVAL;
2649 
2650 		len = sizeof(*option) + option->length * 4;
2651 		if (len > opts_len)
2652 			return -EINVAL;
2653 
2654 		crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
2655 
2656 		option = (struct geneve_opt *)((u8 *)option + len);
2657 		opts_len -= len;
2658 	}
2659 
2660 	key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
2661 
2662 	return 0;
2663 }
2664 
2665 static int validate_and_copy_set_tun(const struct nlattr *attr,
2666 				     struct sw_flow_actions **sfa, bool log)
2667 {
2668 	struct sw_flow_match match;
2669 	struct sw_flow_key key;
2670 	struct metadata_dst *tun_dst;
2671 	struct ip_tunnel_info *tun_info;
2672 	struct ovs_tunnel_info *ovs_tun;
2673 	struct nlattr *a;
2674 	int err = 0, start, opts_type;
2675 	__be16 dst_opt_type;
2676 
2677 	dst_opt_type = 0;
2678 	ovs_match_init(&match, &key, true, NULL);
2679 	opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
2680 	if (opts_type < 0)
2681 		return opts_type;
2682 
2683 	if (key.tun_opts_len) {
2684 		switch (opts_type) {
2685 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
2686 			err = validate_geneve_opts(&key);
2687 			if (err < 0)
2688 				return err;
2689 			dst_opt_type = TUNNEL_GENEVE_OPT;
2690 			break;
2691 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
2692 			dst_opt_type = TUNNEL_VXLAN_OPT;
2693 			break;
2694 		case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
2695 			dst_opt_type = TUNNEL_ERSPAN_OPT;
2696 			break;
2697 		}
2698 	}
2699 
2700 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
2701 	if (start < 0)
2702 		return start;
2703 
2704 	tun_dst = metadata_dst_alloc(key.tun_opts_len, METADATA_IP_TUNNEL,
2705 				     GFP_KERNEL);
2706 
2707 	if (!tun_dst)
2708 		return -ENOMEM;
2709 
2710 	err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
2711 	if (err) {
2712 		dst_release((struct dst_entry *)tun_dst);
2713 		return err;
2714 	}
2715 
2716 	a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
2717 			 sizeof(*ovs_tun), log);
2718 	if (IS_ERR(a)) {
2719 		dst_release((struct dst_entry *)tun_dst);
2720 		return PTR_ERR(a);
2721 	}
2722 
2723 	ovs_tun = nla_data(a);
2724 	ovs_tun->tun_dst = tun_dst;
2725 
2726 	tun_info = &tun_dst->u.tun_info;
2727 	tun_info->mode = IP_TUNNEL_INFO_TX;
2728 	if (key.tun_proto == AF_INET6)
2729 		tun_info->mode |= IP_TUNNEL_INFO_IPV6;
2730 	else if (key.tun_proto == AF_INET && key.tun_key.u.ipv4.dst == 0)
2731 		tun_info->mode |= IP_TUNNEL_INFO_BRIDGE;
2732 	tun_info->key = key.tun_key;
2733 
2734 	/* We need to store the options in the action itself since
2735 	 * everything else will go away after flow setup. We can append
2736 	 * it to tun_info and then point there.
2737 	 */
2738 	ip_tunnel_info_opts_set(tun_info,
2739 				TUN_METADATA_OPTS(&key, key.tun_opts_len),
2740 				key.tun_opts_len, dst_opt_type);
2741 	add_nested_action_end(*sfa, start);
2742 
2743 	return err;
2744 }
2745 
2746 static bool validate_nsh(const struct nlattr *attr, bool is_mask,
2747 			 bool is_push_nsh, bool log)
2748 {
2749 	struct sw_flow_match match;
2750 	struct sw_flow_key key;
2751 	int ret = 0;
2752 
2753 	ovs_match_init(&match, &key, true, NULL);
2754 	ret = nsh_key_put_from_nlattr(attr, &match, is_mask,
2755 				      is_push_nsh, log);
2756 	return !ret;
2757 }
2758 
2759 /* Return false if there are any non-masked bits set.
2760  * Mask follows data immediately, before any netlink padding.
2761  */
2762 static bool validate_masked(u8 *data, int len)
2763 {
2764 	u8 *mask = data + len;
2765 
2766 	while (len--)
2767 		if (*data++ & ~*mask++)
2768 			return false;
2769 
2770 	return true;
2771 }
2772 
2773 static int validate_set(const struct nlattr *a,
2774 			const struct sw_flow_key *flow_key,
2775 			struct sw_flow_actions **sfa, bool *skip_copy,
2776 			u8 mac_proto, __be16 eth_type, bool masked, bool log)
2777 {
2778 	const struct nlattr *ovs_key = nla_data(a);
2779 	int key_type = nla_type(ovs_key);
2780 	size_t key_len;
2781 
2782 	/* There can be only one key in a action */
2783 	if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2784 		return -EINVAL;
2785 
2786 	key_len = nla_len(ovs_key);
2787 	if (masked)
2788 		key_len /= 2;
2789 
2790 	if (key_type > OVS_KEY_ATTR_MAX ||
2791 	    !check_attr_len(key_len, ovs_key_lens[key_type].len))
2792 		return -EINVAL;
2793 
2794 	if (masked && !validate_masked(nla_data(ovs_key), key_len))
2795 		return -EINVAL;
2796 
2797 	switch (key_type) {
2798 	case OVS_KEY_ATTR_PRIORITY:
2799 	case OVS_KEY_ATTR_SKB_MARK:
2800 	case OVS_KEY_ATTR_CT_MARK:
2801 	case OVS_KEY_ATTR_CT_LABELS:
2802 		break;
2803 
2804 	case OVS_KEY_ATTR_ETHERNET:
2805 		if (mac_proto != MAC_PROTO_ETHERNET)
2806 			return -EINVAL;
2807 		break;
2808 
2809 	case OVS_KEY_ATTR_TUNNEL: {
2810 		int err;
2811 
2812 		if (masked)
2813 			return -EINVAL; /* Masked tunnel set not supported. */
2814 
2815 		*skip_copy = true;
2816 		err = validate_and_copy_set_tun(a, sfa, log);
2817 		if (err)
2818 			return err;
2819 		break;
2820 	}
2821 	case OVS_KEY_ATTR_IPV4: {
2822 		const struct ovs_key_ipv4 *ipv4_key;
2823 
2824 		if (eth_type != htons(ETH_P_IP))
2825 			return -EINVAL;
2826 
2827 		ipv4_key = nla_data(ovs_key);
2828 
2829 		if (masked) {
2830 			const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2831 
2832 			/* Non-writeable fields. */
2833 			if (mask->ipv4_proto || mask->ipv4_frag)
2834 				return -EINVAL;
2835 		} else {
2836 			if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2837 				return -EINVAL;
2838 
2839 			if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2840 				return -EINVAL;
2841 		}
2842 		break;
2843 	}
2844 	case OVS_KEY_ATTR_IPV6: {
2845 		const struct ovs_key_ipv6 *ipv6_key;
2846 
2847 		if (eth_type != htons(ETH_P_IPV6))
2848 			return -EINVAL;
2849 
2850 		ipv6_key = nla_data(ovs_key);
2851 
2852 		if (masked) {
2853 			const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2854 
2855 			/* Non-writeable fields. */
2856 			if (mask->ipv6_proto || mask->ipv6_frag)
2857 				return -EINVAL;
2858 
2859 			/* Invalid bits in the flow label mask? */
2860 			if (ntohl(mask->ipv6_label) & 0xFFF00000)
2861 				return -EINVAL;
2862 		} else {
2863 			if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2864 				return -EINVAL;
2865 
2866 			if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2867 				return -EINVAL;
2868 		}
2869 		if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2870 			return -EINVAL;
2871 
2872 		break;
2873 	}
2874 	case OVS_KEY_ATTR_TCP:
2875 		if ((eth_type != htons(ETH_P_IP) &&
2876 		     eth_type != htons(ETH_P_IPV6)) ||
2877 		    flow_key->ip.proto != IPPROTO_TCP)
2878 			return -EINVAL;
2879 
2880 		break;
2881 
2882 	case OVS_KEY_ATTR_UDP:
2883 		if ((eth_type != htons(ETH_P_IP) &&
2884 		     eth_type != htons(ETH_P_IPV6)) ||
2885 		    flow_key->ip.proto != IPPROTO_UDP)
2886 			return -EINVAL;
2887 
2888 		break;
2889 
2890 	case OVS_KEY_ATTR_MPLS:
2891 		if (!eth_p_mpls(eth_type))
2892 			return -EINVAL;
2893 		break;
2894 
2895 	case OVS_KEY_ATTR_SCTP:
2896 		if ((eth_type != htons(ETH_P_IP) &&
2897 		     eth_type != htons(ETH_P_IPV6)) ||
2898 		    flow_key->ip.proto != IPPROTO_SCTP)
2899 			return -EINVAL;
2900 
2901 		break;
2902 
2903 	case OVS_KEY_ATTR_NSH:
2904 		if (eth_type != htons(ETH_P_NSH))
2905 			return -EINVAL;
2906 		if (!validate_nsh(nla_data(a), masked, false, log))
2907 			return -EINVAL;
2908 		break;
2909 
2910 	default:
2911 		return -EINVAL;
2912 	}
2913 
2914 	/* Convert non-masked non-tunnel set actions to masked set actions. */
2915 	if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2916 		int start, len = key_len * 2;
2917 		struct nlattr *at;
2918 
2919 		*skip_copy = true;
2920 
2921 		start = add_nested_action_start(sfa,
2922 						OVS_ACTION_ATTR_SET_TO_MASKED,
2923 						log);
2924 		if (start < 0)
2925 			return start;
2926 
2927 		at = __add_action(sfa, key_type, NULL, len, log);
2928 		if (IS_ERR(at))
2929 			return PTR_ERR(at);
2930 
2931 		memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2932 		memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2933 		/* Clear non-writeable bits from otherwise writeable fields. */
2934 		if (key_type == OVS_KEY_ATTR_IPV6) {
2935 			struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2936 
2937 			mask->ipv6_label &= htonl(0x000FFFFF);
2938 		}
2939 		add_nested_action_end(*sfa, start);
2940 	}
2941 
2942 	return 0;
2943 }
2944 
2945 static int validate_userspace(const struct nlattr *attr)
2946 {
2947 	static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2948 		[OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2949 		[OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2950 		[OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2951 	};
2952 	struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2953 	int error;
2954 
2955 	error = nla_parse_nested_deprecated(a, OVS_USERSPACE_ATTR_MAX, attr,
2956 					    userspace_policy, NULL);
2957 	if (error)
2958 		return error;
2959 
2960 	if (!a[OVS_USERSPACE_ATTR_PID] ||
2961 	    !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2962 		return -EINVAL;
2963 
2964 	return 0;
2965 }
2966 
2967 static const struct nla_policy cpl_policy[OVS_CHECK_PKT_LEN_ATTR_MAX + 1] = {
2968 	[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN] = {.type = NLA_U16 },
2969 	[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER] = {.type = NLA_NESTED },
2970 	[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL] = {.type = NLA_NESTED },
2971 };
2972 
2973 static int validate_and_copy_check_pkt_len(struct net *net,
2974 					   const struct nlattr *attr,
2975 					   const struct sw_flow_key *key,
2976 					   struct sw_flow_actions **sfa,
2977 					   __be16 eth_type, __be16 vlan_tci,
2978 					   u32 mpls_label_count,
2979 					   bool log, bool last)
2980 {
2981 	const struct nlattr *acts_if_greater, *acts_if_lesser_eq;
2982 	struct nlattr *a[OVS_CHECK_PKT_LEN_ATTR_MAX + 1];
2983 	struct check_pkt_len_arg arg;
2984 	int nested_acts_start;
2985 	int start, err;
2986 
2987 	err = nla_parse_deprecated_strict(a, OVS_CHECK_PKT_LEN_ATTR_MAX,
2988 					  nla_data(attr), nla_len(attr),
2989 					  cpl_policy, NULL);
2990 	if (err)
2991 		return err;
2992 
2993 	if (!a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN] ||
2994 	    !nla_get_u16(a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN]))
2995 		return -EINVAL;
2996 
2997 	acts_if_lesser_eq = a[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL];
2998 	acts_if_greater = a[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER];
2999 
3000 	/* Both the nested action should be present. */
3001 	if (!acts_if_greater || !acts_if_lesser_eq)
3002 		return -EINVAL;
3003 
3004 	/* validation done, copy the nested actions. */
3005 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_CHECK_PKT_LEN,
3006 					log);
3007 	if (start < 0)
3008 		return start;
3009 
3010 	arg.pkt_len = nla_get_u16(a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN]);
3011 	arg.exec_for_lesser_equal =
3012 		last || !actions_may_change_flow(acts_if_lesser_eq);
3013 	arg.exec_for_greater =
3014 		last || !actions_may_change_flow(acts_if_greater);
3015 
3016 	err = ovs_nla_add_action(sfa, OVS_CHECK_PKT_LEN_ATTR_ARG, &arg,
3017 				 sizeof(arg), log);
3018 	if (err)
3019 		return err;
3020 
3021 	nested_acts_start = add_nested_action_start(sfa,
3022 		OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL, log);
3023 	if (nested_acts_start < 0)
3024 		return nested_acts_start;
3025 
3026 	err = __ovs_nla_copy_actions(net, acts_if_lesser_eq, key, sfa,
3027 				     eth_type, vlan_tci, mpls_label_count, log);
3028 
3029 	if (err)
3030 		return err;
3031 
3032 	add_nested_action_end(*sfa, nested_acts_start);
3033 
3034 	nested_acts_start = add_nested_action_start(sfa,
3035 		OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER, log);
3036 	if (nested_acts_start < 0)
3037 		return nested_acts_start;
3038 
3039 	err = __ovs_nla_copy_actions(net, acts_if_greater, key, sfa,
3040 				     eth_type, vlan_tci, mpls_label_count, log);
3041 
3042 	if (err)
3043 		return err;
3044 
3045 	add_nested_action_end(*sfa, nested_acts_start);
3046 	add_nested_action_end(*sfa, start);
3047 	return 0;
3048 }
3049 
3050 static int copy_action(const struct nlattr *from,
3051 		       struct sw_flow_actions **sfa, bool log)
3052 {
3053 	int totlen = NLA_ALIGN(from->nla_len);
3054 	struct nlattr *to;
3055 
3056 	to = reserve_sfa_size(sfa, from->nla_len, log);
3057 	if (IS_ERR(to))
3058 		return PTR_ERR(to);
3059 
3060 	memcpy(to, from, totlen);
3061 	return 0;
3062 }
3063 
3064 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
3065 				  const struct sw_flow_key *key,
3066 				  struct sw_flow_actions **sfa,
3067 				  __be16 eth_type, __be16 vlan_tci,
3068 				  u32 mpls_label_count, bool log)
3069 {
3070 	u8 mac_proto = ovs_key_mac_proto(key);
3071 	const struct nlattr *a;
3072 	int rem, err;
3073 
3074 	nla_for_each_nested(a, attr, rem) {
3075 		/* Expected argument lengths, (u32)-1 for variable length. */
3076 		static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
3077 			[OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
3078 			[OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
3079 			[OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
3080 			[OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
3081 			[OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
3082 			[OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
3083 			[OVS_ACTION_ATTR_POP_VLAN] = 0,
3084 			[OVS_ACTION_ATTR_SET] = (u32)-1,
3085 			[OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
3086 			[OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
3087 			[OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
3088 			[OVS_ACTION_ATTR_CT] = (u32)-1,
3089 			[OVS_ACTION_ATTR_CT_CLEAR] = 0,
3090 			[OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
3091 			[OVS_ACTION_ATTR_PUSH_ETH] = sizeof(struct ovs_action_push_eth),
3092 			[OVS_ACTION_ATTR_POP_ETH] = 0,
3093 			[OVS_ACTION_ATTR_PUSH_NSH] = (u32)-1,
3094 			[OVS_ACTION_ATTR_POP_NSH] = 0,
3095 			[OVS_ACTION_ATTR_METER] = sizeof(u32),
3096 			[OVS_ACTION_ATTR_CLONE] = (u32)-1,
3097 			[OVS_ACTION_ATTR_CHECK_PKT_LEN] = (u32)-1,
3098 			[OVS_ACTION_ATTR_ADD_MPLS] = sizeof(struct ovs_action_add_mpls),
3099 			[OVS_ACTION_ATTR_DEC_TTL] = (u32)-1,
3100 		};
3101 		const struct ovs_action_push_vlan *vlan;
3102 		int type = nla_type(a);
3103 		bool skip_copy;
3104 
3105 		if (type > OVS_ACTION_ATTR_MAX ||
3106 		    (action_lens[type] != nla_len(a) &&
3107 		     action_lens[type] != (u32)-1))
3108 			return -EINVAL;
3109 
3110 		skip_copy = false;
3111 		switch (type) {
3112 		case OVS_ACTION_ATTR_UNSPEC:
3113 			return -EINVAL;
3114 
3115 		case OVS_ACTION_ATTR_USERSPACE:
3116 			err = validate_userspace(a);
3117 			if (err)
3118 				return err;
3119 			break;
3120 
3121 		case OVS_ACTION_ATTR_OUTPUT:
3122 			if (nla_get_u32(a) >= DP_MAX_PORTS)
3123 				return -EINVAL;
3124 			break;
3125 
3126 		case OVS_ACTION_ATTR_TRUNC: {
3127 			const struct ovs_action_trunc *trunc = nla_data(a);
3128 
3129 			if (trunc->max_len < ETH_HLEN)
3130 				return -EINVAL;
3131 			break;
3132 		}
3133 
3134 		case OVS_ACTION_ATTR_HASH: {
3135 			const struct ovs_action_hash *act_hash = nla_data(a);
3136 
3137 			switch (act_hash->hash_alg) {
3138 			case OVS_HASH_ALG_L4:
3139 				break;
3140 			default:
3141 				return  -EINVAL;
3142 			}
3143 
3144 			break;
3145 		}
3146 
3147 		case OVS_ACTION_ATTR_POP_VLAN:
3148 			if (mac_proto != MAC_PROTO_ETHERNET)
3149 				return -EINVAL;
3150 			vlan_tci = htons(0);
3151 			break;
3152 
3153 		case OVS_ACTION_ATTR_PUSH_VLAN:
3154 			if (mac_proto != MAC_PROTO_ETHERNET)
3155 				return -EINVAL;
3156 			vlan = nla_data(a);
3157 			if (!eth_type_vlan(vlan->vlan_tpid))
3158 				return -EINVAL;
3159 			if (!(vlan->vlan_tci & htons(VLAN_CFI_MASK)))
3160 				return -EINVAL;
3161 			vlan_tci = vlan->vlan_tci;
3162 			break;
3163 
3164 		case OVS_ACTION_ATTR_RECIRC:
3165 			break;
3166 
3167 		case OVS_ACTION_ATTR_ADD_MPLS: {
3168 			const struct ovs_action_add_mpls *mpls = nla_data(a);
3169 
3170 			if (!eth_p_mpls(mpls->mpls_ethertype))
3171 				return -EINVAL;
3172 
3173 			if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK) {
3174 				if (vlan_tci & htons(VLAN_CFI_MASK) ||
3175 				    (eth_type != htons(ETH_P_IP) &&
3176 				     eth_type != htons(ETH_P_IPV6) &&
3177 				     eth_type != htons(ETH_P_ARP) &&
3178 				     eth_type != htons(ETH_P_RARP) &&
3179 				     !eth_p_mpls(eth_type)))
3180 					return -EINVAL;
3181 				mpls_label_count++;
3182 			} else {
3183 				if (mac_proto == MAC_PROTO_ETHERNET) {
3184 					mpls_label_count = 1;
3185 					mac_proto = MAC_PROTO_NONE;
3186 				} else {
3187 					mpls_label_count++;
3188 				}
3189 			}
3190 			eth_type = mpls->mpls_ethertype;
3191 			break;
3192 		}
3193 
3194 		case OVS_ACTION_ATTR_PUSH_MPLS: {
3195 			const struct ovs_action_push_mpls *mpls = nla_data(a);
3196 
3197 			if (!eth_p_mpls(mpls->mpls_ethertype))
3198 				return -EINVAL;
3199 			/* Prohibit push MPLS other than to a white list
3200 			 * for packets that have a known tag order.
3201 			 */
3202 			if (vlan_tci & htons(VLAN_CFI_MASK) ||
3203 			    (eth_type != htons(ETH_P_IP) &&
3204 			     eth_type != htons(ETH_P_IPV6) &&
3205 			     eth_type != htons(ETH_P_ARP) &&
3206 			     eth_type != htons(ETH_P_RARP) &&
3207 			     !eth_p_mpls(eth_type)))
3208 				return -EINVAL;
3209 			eth_type = mpls->mpls_ethertype;
3210 			mpls_label_count++;
3211 			break;
3212 		}
3213 
3214 		case OVS_ACTION_ATTR_POP_MPLS: {
3215 			__be16  proto;
3216 			if (vlan_tci & htons(VLAN_CFI_MASK) ||
3217 			    !eth_p_mpls(eth_type))
3218 				return -EINVAL;
3219 
3220 			/* Disallow subsequent L2.5+ set actions and mpls_pop
3221 			 * actions once the last MPLS label in the packet is
3222 			 * is popped as there is no check here to ensure that
3223 			 * the new eth type is valid and thus set actions could
3224 			 * write off the end of the packet or otherwise corrupt
3225 			 * it.
3226 			 *
3227 			 * Support for these actions is planned using packet
3228 			 * recirculation.
3229 			 */
3230 			proto = nla_get_be16(a);
3231 
3232 			if (proto == htons(ETH_P_TEB) &&
3233 			    mac_proto != MAC_PROTO_NONE)
3234 				return -EINVAL;
3235 
3236 			mpls_label_count--;
3237 
3238 			if (!eth_p_mpls(proto) || !mpls_label_count)
3239 				eth_type = htons(0);
3240 			else
3241 				eth_type =  proto;
3242 
3243 			break;
3244 		}
3245 
3246 		case OVS_ACTION_ATTR_SET:
3247 			err = validate_set(a, key, sfa,
3248 					   &skip_copy, mac_proto, eth_type,
3249 					   false, log);
3250 			if (err)
3251 				return err;
3252 			break;
3253 
3254 		case OVS_ACTION_ATTR_SET_MASKED:
3255 			err = validate_set(a, key, sfa,
3256 					   &skip_copy, mac_proto, eth_type,
3257 					   true, log);
3258 			if (err)
3259 				return err;
3260 			break;
3261 
3262 		case OVS_ACTION_ATTR_SAMPLE: {
3263 			bool last = nla_is_last(a, rem);
3264 
3265 			err = validate_and_copy_sample(net, a, key, sfa,
3266 						       eth_type, vlan_tci,
3267 						       mpls_label_count,
3268 						       log, last);
3269 			if (err)
3270 				return err;
3271 			skip_copy = true;
3272 			break;
3273 		}
3274 
3275 		case OVS_ACTION_ATTR_CT:
3276 			err = ovs_ct_copy_action(net, a, key, sfa, log);
3277 			if (err)
3278 				return err;
3279 			skip_copy = true;
3280 			break;
3281 
3282 		case OVS_ACTION_ATTR_CT_CLEAR:
3283 			break;
3284 
3285 		case OVS_ACTION_ATTR_PUSH_ETH:
3286 			/* Disallow pushing an Ethernet header if one
3287 			 * is already present */
3288 			if (mac_proto != MAC_PROTO_NONE)
3289 				return -EINVAL;
3290 			mac_proto = MAC_PROTO_ETHERNET;
3291 			break;
3292 
3293 		case OVS_ACTION_ATTR_POP_ETH:
3294 			if (mac_proto != MAC_PROTO_ETHERNET)
3295 				return -EINVAL;
3296 			if (vlan_tci & htons(VLAN_CFI_MASK))
3297 				return -EINVAL;
3298 			mac_proto = MAC_PROTO_NONE;
3299 			break;
3300 
3301 		case OVS_ACTION_ATTR_PUSH_NSH:
3302 			if (mac_proto != MAC_PROTO_ETHERNET) {
3303 				u8 next_proto;
3304 
3305 				next_proto = tun_p_from_eth_p(eth_type);
3306 				if (!next_proto)
3307 					return -EINVAL;
3308 			}
3309 			mac_proto = MAC_PROTO_NONE;
3310 			if (!validate_nsh(nla_data(a), false, true, true))
3311 				return -EINVAL;
3312 			break;
3313 
3314 		case OVS_ACTION_ATTR_POP_NSH: {
3315 			__be16 inner_proto;
3316 
3317 			if (eth_type != htons(ETH_P_NSH))
3318 				return -EINVAL;
3319 			inner_proto = tun_p_to_eth_p(key->nsh.base.np);
3320 			if (!inner_proto)
3321 				return -EINVAL;
3322 			if (key->nsh.base.np == TUN_P_ETHERNET)
3323 				mac_proto = MAC_PROTO_ETHERNET;
3324 			else
3325 				mac_proto = MAC_PROTO_NONE;
3326 			break;
3327 		}
3328 
3329 		case OVS_ACTION_ATTR_METER:
3330 			/* Non-existent meters are simply ignored.  */
3331 			break;
3332 
3333 		case OVS_ACTION_ATTR_CLONE: {
3334 			bool last = nla_is_last(a, rem);
3335 
3336 			err = validate_and_copy_clone(net, a, key, sfa,
3337 						      eth_type, vlan_tci,
3338 						      mpls_label_count,
3339 						      log, last);
3340 			if (err)
3341 				return err;
3342 			skip_copy = true;
3343 			break;
3344 		}
3345 
3346 		case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
3347 			bool last = nla_is_last(a, rem);
3348 
3349 			err = validate_and_copy_check_pkt_len(net, a, key, sfa,
3350 							      eth_type,
3351 							      vlan_tci,
3352 							      mpls_label_count,
3353 							      log, last);
3354 			if (err)
3355 				return err;
3356 			skip_copy = true;
3357 			break;
3358 		}
3359 
3360 		case OVS_ACTION_ATTR_DEC_TTL:
3361 			err = validate_and_copy_dec_ttl(net, a, key, sfa,
3362 							eth_type, vlan_tci,
3363 							mpls_label_count, log);
3364 			if (err)
3365 				return err;
3366 			skip_copy = true;
3367 			break;
3368 
3369 		default:
3370 			OVS_NLERR(log, "Unknown Action type %d", type);
3371 			return -EINVAL;
3372 		}
3373 		if (!skip_copy) {
3374 			err = copy_action(a, sfa, log);
3375 			if (err)
3376 				return err;
3377 		}
3378 	}
3379 
3380 	if (rem > 0)
3381 		return -EINVAL;
3382 
3383 	return 0;
3384 }
3385 
3386 /* 'key' must be the masked key. */
3387 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
3388 			 const struct sw_flow_key *key,
3389 			 struct sw_flow_actions **sfa, bool log)
3390 {
3391 	int err;
3392 	u32 mpls_label_count = 0;
3393 
3394 	*sfa = nla_alloc_flow_actions(min(nla_len(attr), MAX_ACTIONS_BUFSIZE));
3395 	if (IS_ERR(*sfa))
3396 		return PTR_ERR(*sfa);
3397 
3398 	if (eth_p_mpls(key->eth.type))
3399 		mpls_label_count = hweight_long(key->mpls.num_labels_mask);
3400 
3401 	(*sfa)->orig_len = nla_len(attr);
3402 	err = __ovs_nla_copy_actions(net, attr, key, sfa, key->eth.type,
3403 				     key->eth.vlan.tci, mpls_label_count, log);
3404 	if (err)
3405 		ovs_nla_free_flow_actions(*sfa);
3406 
3407 	return err;
3408 }
3409 
3410 static int sample_action_to_attr(const struct nlattr *attr,
3411 				 struct sk_buff *skb)
3412 {
3413 	struct nlattr *start, *ac_start = NULL, *sample_arg;
3414 	int err = 0, rem = nla_len(attr);
3415 	const struct sample_arg *arg;
3416 	struct nlattr *actions;
3417 
3418 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SAMPLE);
3419 	if (!start)
3420 		return -EMSGSIZE;
3421 
3422 	sample_arg = nla_data(attr);
3423 	arg = nla_data(sample_arg);
3424 	actions = nla_next(sample_arg, &rem);
3425 
3426 	if (nla_put_u32(skb, OVS_SAMPLE_ATTR_PROBABILITY, arg->probability)) {
3427 		err = -EMSGSIZE;
3428 		goto out;
3429 	}
3430 
3431 	ac_start = nla_nest_start_noflag(skb, OVS_SAMPLE_ATTR_ACTIONS);
3432 	if (!ac_start) {
3433 		err = -EMSGSIZE;
3434 		goto out;
3435 	}
3436 
3437 	err = ovs_nla_put_actions(actions, rem, skb);
3438 
3439 out:
3440 	if (err) {
3441 		nla_nest_cancel(skb, ac_start);
3442 		nla_nest_cancel(skb, start);
3443 	} else {
3444 		nla_nest_end(skb, ac_start);
3445 		nla_nest_end(skb, start);
3446 	}
3447 
3448 	return err;
3449 }
3450 
3451 static int clone_action_to_attr(const struct nlattr *attr,
3452 				struct sk_buff *skb)
3453 {
3454 	struct nlattr *start;
3455 	int err = 0, rem = nla_len(attr);
3456 
3457 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CLONE);
3458 	if (!start)
3459 		return -EMSGSIZE;
3460 
3461 	/* Skipping the OVS_CLONE_ATTR_EXEC that is always the first attribute. */
3462 	attr = nla_next(nla_data(attr), &rem);
3463 	err = ovs_nla_put_actions(attr, rem, skb);
3464 
3465 	if (err)
3466 		nla_nest_cancel(skb, start);
3467 	else
3468 		nla_nest_end(skb, start);
3469 
3470 	return err;
3471 }
3472 
3473 static int check_pkt_len_action_to_attr(const struct nlattr *attr,
3474 					struct sk_buff *skb)
3475 {
3476 	struct nlattr *start, *ac_start = NULL;
3477 	const struct check_pkt_len_arg *arg;
3478 	const struct nlattr *a, *cpl_arg;
3479 	int err = 0, rem = nla_len(attr);
3480 
3481 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CHECK_PKT_LEN);
3482 	if (!start)
3483 		return -EMSGSIZE;
3484 
3485 	/* The first nested attribute in 'attr' is always
3486 	 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
3487 	 */
3488 	cpl_arg = nla_data(attr);
3489 	arg = nla_data(cpl_arg);
3490 
3491 	if (nla_put_u16(skb, OVS_CHECK_PKT_LEN_ATTR_PKT_LEN, arg->pkt_len)) {
3492 		err = -EMSGSIZE;
3493 		goto out;
3494 	}
3495 
3496 	/* Second nested attribute in 'attr' is always
3497 	 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
3498 	 */
3499 	a = nla_next(cpl_arg, &rem);
3500 	ac_start =  nla_nest_start_noflag(skb,
3501 					  OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL);
3502 	if (!ac_start) {
3503 		err = -EMSGSIZE;
3504 		goto out;
3505 	}
3506 
3507 	err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3508 	if (err) {
3509 		nla_nest_cancel(skb, ac_start);
3510 		goto out;
3511 	} else {
3512 		nla_nest_end(skb, ac_start);
3513 	}
3514 
3515 	/* Third nested attribute in 'attr' is always
3516 	 * OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER.
3517 	 */
3518 	a = nla_next(a, &rem);
3519 	ac_start =  nla_nest_start_noflag(skb,
3520 					  OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER);
3521 	if (!ac_start) {
3522 		err = -EMSGSIZE;
3523 		goto out;
3524 	}
3525 
3526 	err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3527 	if (err) {
3528 		nla_nest_cancel(skb, ac_start);
3529 		goto out;
3530 	} else {
3531 		nla_nest_end(skb, ac_start);
3532 	}
3533 
3534 	nla_nest_end(skb, start);
3535 	return 0;
3536 
3537 out:
3538 	nla_nest_cancel(skb, start);
3539 	return err;
3540 }
3541 
3542 static int dec_ttl_action_to_attr(const struct nlattr *attr,
3543 				  struct sk_buff *skb)
3544 {
3545 	struct nlattr *start, *action_start;
3546 	const struct nlattr *a;
3547 	int err = 0, rem;
3548 
3549 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_DEC_TTL);
3550 	if (!start)
3551 		return -EMSGSIZE;
3552 
3553 	nla_for_each_attr(a, nla_data(attr), nla_len(attr), rem) {
3554 		switch (nla_type(a)) {
3555 		case OVS_DEC_TTL_ATTR_ACTION:
3556 
3557 			action_start = nla_nest_start_noflag(skb, OVS_DEC_TTL_ATTR_ACTION);
3558 			if (!action_start) {
3559 				err = -EMSGSIZE;
3560 				goto out;
3561 			}
3562 
3563 			err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3564 			if (err)
3565 				goto out;
3566 
3567 			nla_nest_end(skb, action_start);
3568 			break;
3569 
3570 		default:
3571 			/* Ignore all other option to be future compatible */
3572 			break;
3573 		}
3574 	}
3575 
3576 	nla_nest_end(skb, start);
3577 	return 0;
3578 
3579 out:
3580 	nla_nest_cancel(skb, start);
3581 	return err;
3582 }
3583 
3584 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
3585 {
3586 	const struct nlattr *ovs_key = nla_data(a);
3587 	int key_type = nla_type(ovs_key);
3588 	struct nlattr *start;
3589 	int err;
3590 
3591 	switch (key_type) {
3592 	case OVS_KEY_ATTR_TUNNEL_INFO: {
3593 		struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
3594 		struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
3595 
3596 		start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SET);
3597 		if (!start)
3598 			return -EMSGSIZE;
3599 
3600 		err =  ip_tun_to_nlattr(skb, &tun_info->key,
3601 					ip_tunnel_info_opts(tun_info),
3602 					tun_info->options_len,
3603 					ip_tunnel_info_af(tun_info), tun_info->mode);
3604 		if (err)
3605 			return err;
3606 		nla_nest_end(skb, start);
3607 		break;
3608 	}
3609 	default:
3610 		if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
3611 			return -EMSGSIZE;
3612 		break;
3613 	}
3614 
3615 	return 0;
3616 }
3617 
3618 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
3619 						struct sk_buff *skb)
3620 {
3621 	const struct nlattr *ovs_key = nla_data(a);
3622 	struct nlattr *nla;
3623 	size_t key_len = nla_len(ovs_key) / 2;
3624 
3625 	/* Revert the conversion we did from a non-masked set action to
3626 	 * masked set action.
3627 	 */
3628 	nla = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SET);
3629 	if (!nla)
3630 		return -EMSGSIZE;
3631 
3632 	if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
3633 		return -EMSGSIZE;
3634 
3635 	nla_nest_end(skb, nla);
3636 	return 0;
3637 }
3638 
3639 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
3640 {
3641 	const struct nlattr *a;
3642 	int rem, err;
3643 
3644 	nla_for_each_attr(a, attr, len, rem) {
3645 		int type = nla_type(a);
3646 
3647 		switch (type) {
3648 		case OVS_ACTION_ATTR_SET:
3649 			err = set_action_to_attr(a, skb);
3650 			if (err)
3651 				return err;
3652 			break;
3653 
3654 		case OVS_ACTION_ATTR_SET_TO_MASKED:
3655 			err = masked_set_action_to_set_action_attr(a, skb);
3656 			if (err)
3657 				return err;
3658 			break;
3659 
3660 		case OVS_ACTION_ATTR_SAMPLE:
3661 			err = sample_action_to_attr(a, skb);
3662 			if (err)
3663 				return err;
3664 			break;
3665 
3666 		case OVS_ACTION_ATTR_CT:
3667 			err = ovs_ct_action_to_attr(nla_data(a), skb);
3668 			if (err)
3669 				return err;
3670 			break;
3671 
3672 		case OVS_ACTION_ATTR_CLONE:
3673 			err = clone_action_to_attr(a, skb);
3674 			if (err)
3675 				return err;
3676 			break;
3677 
3678 		case OVS_ACTION_ATTR_CHECK_PKT_LEN:
3679 			err = check_pkt_len_action_to_attr(a, skb);
3680 			if (err)
3681 				return err;
3682 			break;
3683 
3684 		case OVS_ACTION_ATTR_DEC_TTL:
3685 			err = dec_ttl_action_to_attr(a, skb);
3686 			if (err)
3687 				return err;
3688 			break;
3689 
3690 		default:
3691 			if (nla_put(skb, type, nla_len(a), nla_data(a)))
3692 				return -EMSGSIZE;
3693 			break;
3694 		}
3695 	}
3696 
3697 	return 0;
3698 }
3699