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