1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2007-2017 Nicira, Inc.
4  */
5 
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 
8 #include "flow.h"
9 #include "datapath.h"
10 #include <linux/uaccess.h>
11 #include <linux/netdevice.h>
12 #include <linux/etherdevice.h>
13 #include <linux/if_ether.h>
14 #include <linux/if_vlan.h>
15 #include <net/llc_pdu.h>
16 #include <linux/kernel.h>
17 #include <linux/jhash.h>
18 #include <linux/jiffies.h>
19 #include <linux/llc.h>
20 #include <linux/module.h>
21 #include <linux/in.h>
22 #include <linux/rcupdate.h>
23 #include <linux/if_arp.h>
24 #include <linux/ip.h>
25 #include <linux/ipv6.h>
26 #include <linux/sctp.h>
27 #include <linux/tcp.h>
28 #include <linux/udp.h>
29 #include <linux/icmp.h>
30 #include <linux/icmpv6.h>
31 #include <linux/rculist.h>
32 #include <net/geneve.h>
33 #include <net/ip.h>
34 #include <net/ipv6.h>
35 #include <net/ndisc.h>
36 #include <net/mpls.h>
37 #include <net/vxlan.h>
38 #include <net/tun_proto.h>
39 #include <net/erspan.h>
40 
41 #include "flow_netlink.h"
42 
43 struct ovs_len_tbl {
44 	int len;
45 	const struct ovs_len_tbl *next;
46 };
47 
48 #define OVS_ATTR_NESTED -1
49 #define OVS_ATTR_VARIABLE -2
50 
51 static bool actions_may_change_flow(const struct nlattr *actions)
52 {
53 	struct nlattr *nla;
54 	int rem;
55 
56 	nla_for_each_nested(nla, actions, rem) {
57 		u16 action = nla_type(nla);
58 
59 		switch (action) {
60 		case OVS_ACTION_ATTR_OUTPUT:
61 		case OVS_ACTION_ATTR_RECIRC:
62 		case OVS_ACTION_ATTR_TRUNC:
63 		case OVS_ACTION_ATTR_USERSPACE:
64 			break;
65 
66 		case OVS_ACTION_ATTR_CT:
67 		case OVS_ACTION_ATTR_CT_CLEAR:
68 		case OVS_ACTION_ATTR_HASH:
69 		case OVS_ACTION_ATTR_POP_ETH:
70 		case OVS_ACTION_ATTR_POP_MPLS:
71 		case OVS_ACTION_ATTR_POP_NSH:
72 		case OVS_ACTION_ATTR_POP_VLAN:
73 		case OVS_ACTION_ATTR_PUSH_ETH:
74 		case OVS_ACTION_ATTR_PUSH_MPLS:
75 		case OVS_ACTION_ATTR_PUSH_NSH:
76 		case OVS_ACTION_ATTR_PUSH_VLAN:
77 		case OVS_ACTION_ATTR_SAMPLE:
78 		case OVS_ACTION_ATTR_SET:
79 		case OVS_ACTION_ATTR_SET_MASKED:
80 		case OVS_ACTION_ATTR_METER:
81 		case OVS_ACTION_ATTR_CHECK_PKT_LEN:
82 		case OVS_ACTION_ATTR_ADD_MPLS:
83 		case OVS_ACTION_ATTR_DEC_TTL:
84 		default:
85 			return true;
86 		}
87 	}
88 	return false;
89 }
90 
91 static void update_range(struct sw_flow_match *match,
92 			 size_t offset, size_t size, bool is_mask)
93 {
94 	struct sw_flow_key_range *range;
95 	size_t start = rounddown(offset, sizeof(long));
96 	size_t end = roundup(offset + size, sizeof(long));
97 
98 	if (!is_mask)
99 		range = &match->range;
100 	else
101 		range = &match->mask->range;
102 
103 	if (range->start == range->end) {
104 		range->start = start;
105 		range->end = end;
106 		return;
107 	}
108 
109 	if (range->start > start)
110 		range->start = start;
111 
112 	if (range->end < end)
113 		range->end = end;
114 }
115 
116 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
117 	do { \
118 		update_range(match, offsetof(struct sw_flow_key, field),    \
119 			     sizeof((match)->key->field), is_mask);	    \
120 		if (is_mask)						    \
121 			(match)->mask->key.field = value;		    \
122 		else							    \
123 			(match)->key->field = value;		            \
124 	} while (0)
125 
126 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)	    \
127 	do {								    \
128 		update_range(match, offset, len, is_mask);		    \
129 		if (is_mask)						    \
130 			memcpy((u8 *)&(match)->mask->key + offset, value_p, \
131 			       len);					   \
132 		else							    \
133 			memcpy((u8 *)(match)->key + offset, value_p, len);  \
134 	} while (0)
135 
136 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)		      \
137 	SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
138 				  value_p, len, is_mask)
139 
140 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)		    \
141 	do {								    \
142 		update_range(match, offsetof(struct sw_flow_key, field),    \
143 			     sizeof((match)->key->field), is_mask);	    \
144 		if (is_mask)						    \
145 			memset((u8 *)&(match)->mask->key.field, value,      \
146 			       sizeof((match)->mask->key.field));	    \
147 		else							    \
148 			memset((u8 *)&(match)->key->field, value,           \
149 			       sizeof((match)->key->field));                \
150 	} while (0)
151 
152 static bool match_validate(const struct sw_flow_match *match,
153 			   u64 key_attrs, u64 mask_attrs, bool log)
154 {
155 	u64 key_expected = 0;
156 	u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
157 
158 	/* The following mask attributes allowed only if they
159 	 * pass the validation tests. */
160 	mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
161 			| (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)
162 			| (1 << OVS_KEY_ATTR_IPV6)
163 			| (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)
164 			| (1 << OVS_KEY_ATTR_TCP)
165 			| (1 << OVS_KEY_ATTR_TCP_FLAGS)
166 			| (1 << OVS_KEY_ATTR_UDP)
167 			| (1 << OVS_KEY_ATTR_SCTP)
168 			| (1 << OVS_KEY_ATTR_ICMP)
169 			| (1 << OVS_KEY_ATTR_ICMPV6)
170 			| (1 << OVS_KEY_ATTR_ARP)
171 			| (1 << OVS_KEY_ATTR_ND)
172 			| (1 << OVS_KEY_ATTR_MPLS)
173 			| (1 << OVS_KEY_ATTR_NSH));
174 
175 	/* Always allowed mask fields. */
176 	mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
177 		       | (1 << OVS_KEY_ATTR_IN_PORT)
178 		       | (1 << OVS_KEY_ATTR_ETHERTYPE));
179 
180 	/* Check key attributes. */
181 	if (match->key->eth.type == htons(ETH_P_ARP)
182 			|| match->key->eth.type == htons(ETH_P_RARP)) {
183 		key_expected |= 1 << OVS_KEY_ATTR_ARP;
184 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
185 			mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
186 	}
187 
188 	if (eth_p_mpls(match->key->eth.type)) {
189 		key_expected |= 1 << OVS_KEY_ATTR_MPLS;
190 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
191 			mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
192 	}
193 
194 	if (match->key->eth.type == htons(ETH_P_IP)) {
195 		key_expected |= 1 << OVS_KEY_ATTR_IPV4;
196 		if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
197 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
198 			mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4;
199 		}
200 
201 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
202 			if (match->key->ip.proto == IPPROTO_UDP) {
203 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
204 				if (match->mask && (match->mask->key.ip.proto == 0xff))
205 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
206 			}
207 
208 			if (match->key->ip.proto == IPPROTO_SCTP) {
209 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
210 				if (match->mask && (match->mask->key.ip.proto == 0xff))
211 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
212 			}
213 
214 			if (match->key->ip.proto == IPPROTO_TCP) {
215 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
216 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
217 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
218 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
219 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
220 				}
221 			}
222 
223 			if (match->key->ip.proto == IPPROTO_ICMP) {
224 				key_expected |= 1 << OVS_KEY_ATTR_ICMP;
225 				if (match->mask && (match->mask->key.ip.proto == 0xff))
226 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
227 			}
228 		}
229 	}
230 
231 	if (match->key->eth.type == htons(ETH_P_IPV6)) {
232 		key_expected |= 1 << OVS_KEY_ATTR_IPV6;
233 		if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
234 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
235 			mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6;
236 		}
237 
238 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
239 			if (match->key->ip.proto == IPPROTO_UDP) {
240 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
241 				if (match->mask && (match->mask->key.ip.proto == 0xff))
242 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
243 			}
244 
245 			if (match->key->ip.proto == IPPROTO_SCTP) {
246 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
247 				if (match->mask && (match->mask->key.ip.proto == 0xff))
248 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
249 			}
250 
251 			if (match->key->ip.proto == IPPROTO_TCP) {
252 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
253 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
254 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
255 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
256 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
257 				}
258 			}
259 
260 			if (match->key->ip.proto == IPPROTO_ICMPV6) {
261 				key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
262 				if (match->mask && (match->mask->key.ip.proto == 0xff))
263 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
264 
265 				if (match->key->tp.src ==
266 						htons(NDISC_NEIGHBOUR_SOLICITATION) ||
267 				    match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
268 					key_expected |= 1 << OVS_KEY_ATTR_ND;
269 					/* Original direction conntrack tuple
270 					 * uses the same space as the ND fields
271 					 * in the key, so both are not allowed
272 					 * at the same time.
273 					 */
274 					mask_allowed &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
275 					if (match->mask && (match->mask->key.tp.src == htons(0xff)))
276 						mask_allowed |= 1 << OVS_KEY_ATTR_ND;
277 				}
278 			}
279 		}
280 	}
281 
282 	if (match->key->eth.type == htons(ETH_P_NSH)) {
283 		key_expected |= 1 << OVS_KEY_ATTR_NSH;
284 		if (match->mask &&
285 		    match->mask->key.eth.type == htons(0xffff)) {
286 			mask_allowed |= 1 << OVS_KEY_ATTR_NSH;
287 		}
288 	}
289 
290 	if ((key_attrs & key_expected) != key_expected) {
291 		/* Key attributes check failed. */
292 		OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
293 			  (unsigned long long)key_attrs,
294 			  (unsigned long long)key_expected);
295 		return false;
296 	}
297 
298 	if ((mask_attrs & mask_allowed) != mask_attrs) {
299 		/* Mask attributes check failed. */
300 		OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
301 			  (unsigned long long)mask_attrs,
302 			  (unsigned long long)mask_allowed);
303 		return false;
304 	}
305 
306 	return true;
307 }
308 
309 size_t ovs_tun_key_attr_size(void)
310 {
311 	/* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
312 	 * updating this function.
313 	 */
314 	return    nla_total_size_64bit(8) /* OVS_TUNNEL_KEY_ATTR_ID */
315 		+ nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
316 		+ nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_DST */
317 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
318 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
319 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
320 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
321 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
322 		+ nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
323 		/* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS and
324 		 * OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS is mutually exclusive with
325 		 * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
326 		 */
327 		+ nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
328 		+ nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
329 }
330 
331 static size_t ovs_nsh_key_attr_size(void)
332 {
333 	/* Whenever adding new OVS_NSH_KEY_ FIELDS, we should consider
334 	 * updating this function.
335 	 */
336 	return  nla_total_size(NSH_BASE_HDR_LEN) /* OVS_NSH_KEY_ATTR_BASE */
337 		/* OVS_NSH_KEY_ATTR_MD1 and OVS_NSH_KEY_ATTR_MD2 are
338 		 * mutually exclusive, so the bigger one can cover
339 		 * the small one.
340 		 */
341 		+ nla_total_size(NSH_CTX_HDRS_MAX_LEN);
342 }
343 
344 size_t ovs_key_attr_size(void)
345 {
346 	/* Whenever adding new OVS_KEY_ FIELDS, we should consider
347 	 * updating this function.
348 	 */
349 	BUILD_BUG_ON(OVS_KEY_ATTR_MAX != 32);
350 
351 	return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
352 		+ nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
353 		  + ovs_tun_key_attr_size()
354 		+ nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
355 		+ nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
356 		+ nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
357 		+ nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
358 		+ nla_total_size(4)   /* OVS_KEY_ATTR_CT_STATE */
359 		+ nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
360 		+ nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
361 		+ nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABELS */
362 		+ nla_total_size(40)  /* OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6 */
363 		+ nla_total_size(0)   /* OVS_KEY_ATTR_NSH */
364 		  + ovs_nsh_key_attr_size()
365 		+ nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
366 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
367 		+ nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
368 		+ nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
369 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
370 		+ nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
371 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
372 		+ nla_total_size(28)  /* OVS_KEY_ATTR_ND */
373 		+ nla_total_size(2);  /* OVS_KEY_ATTR_IPV6_EXTHDRS */
374 }
375 
376 static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
377 	[OVS_VXLAN_EXT_GBP]	    = { .len = sizeof(u32) },
378 };
379 
380 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
381 	[OVS_TUNNEL_KEY_ATTR_ID]	    = { .len = sizeof(u64) },
382 	[OVS_TUNNEL_KEY_ATTR_IPV4_SRC]	    = { .len = sizeof(u32) },
383 	[OVS_TUNNEL_KEY_ATTR_IPV4_DST]	    = { .len = sizeof(u32) },
384 	[OVS_TUNNEL_KEY_ATTR_TOS]	    = { .len = 1 },
385 	[OVS_TUNNEL_KEY_ATTR_TTL]	    = { .len = 1 },
386 	[OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
387 	[OVS_TUNNEL_KEY_ATTR_CSUM]	    = { .len = 0 },
388 	[OVS_TUNNEL_KEY_ATTR_TP_SRC]	    = { .len = sizeof(u16) },
389 	[OVS_TUNNEL_KEY_ATTR_TP_DST]	    = { .len = sizeof(u16) },
390 	[OVS_TUNNEL_KEY_ATTR_OAM]	    = { .len = 0 },
391 	[OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_VARIABLE },
392 	[OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED,
393 						.next = ovs_vxlan_ext_key_lens },
394 	[OVS_TUNNEL_KEY_ATTR_IPV6_SRC]      = { .len = sizeof(struct in6_addr) },
395 	[OVS_TUNNEL_KEY_ATTR_IPV6_DST]      = { .len = sizeof(struct in6_addr) },
396 	[OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS]   = { .len = OVS_ATTR_VARIABLE },
397 	[OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE]   = { .len = 0 },
398 };
399 
400 static const struct ovs_len_tbl
401 ovs_nsh_key_attr_lens[OVS_NSH_KEY_ATTR_MAX + 1] = {
402 	[OVS_NSH_KEY_ATTR_BASE] = { .len = sizeof(struct ovs_nsh_key_base) },
403 	[OVS_NSH_KEY_ATTR_MD1]  = { .len = sizeof(struct ovs_nsh_key_md1) },
404 	[OVS_NSH_KEY_ATTR_MD2]  = { .len = OVS_ATTR_VARIABLE },
405 };
406 
407 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
408 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
409 	[OVS_KEY_ATTR_ENCAP]	 = { .len = OVS_ATTR_NESTED },
410 	[OVS_KEY_ATTR_PRIORITY]	 = { .len = sizeof(u32) },
411 	[OVS_KEY_ATTR_IN_PORT]	 = { .len = sizeof(u32) },
412 	[OVS_KEY_ATTR_SKB_MARK]	 = { .len = sizeof(u32) },
413 	[OVS_KEY_ATTR_ETHERNET]	 = { .len = sizeof(struct ovs_key_ethernet) },
414 	[OVS_KEY_ATTR_VLAN]	 = { .len = sizeof(__be16) },
415 	[OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
416 	[OVS_KEY_ATTR_IPV4]	 = { .len = sizeof(struct ovs_key_ipv4) },
417 	[OVS_KEY_ATTR_IPV6]	 = { .len = sizeof(struct ovs_key_ipv6) },
418 	[OVS_KEY_ATTR_TCP]	 = { .len = sizeof(struct ovs_key_tcp) },
419 	[OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
420 	[OVS_KEY_ATTR_UDP]	 = { .len = sizeof(struct ovs_key_udp) },
421 	[OVS_KEY_ATTR_SCTP]	 = { .len = sizeof(struct ovs_key_sctp) },
422 	[OVS_KEY_ATTR_ICMP]	 = { .len = sizeof(struct ovs_key_icmp) },
423 	[OVS_KEY_ATTR_ICMPV6]	 = { .len = sizeof(struct ovs_key_icmpv6) },
424 	[OVS_KEY_ATTR_ARP]	 = { .len = sizeof(struct ovs_key_arp) },
425 	[OVS_KEY_ATTR_ND]	 = { .len = sizeof(struct ovs_key_nd) },
426 	[OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
427 	[OVS_KEY_ATTR_DP_HASH]	 = { .len = sizeof(u32) },
428 	[OVS_KEY_ATTR_TUNNEL]	 = { .len = OVS_ATTR_NESTED,
429 				     .next = ovs_tunnel_key_lens, },
430 	[OVS_KEY_ATTR_MPLS]	 = { .len = OVS_ATTR_VARIABLE },
431 	[OVS_KEY_ATTR_CT_STATE]	 = { .len = sizeof(u32) },
432 	[OVS_KEY_ATTR_CT_ZONE]	 = { .len = sizeof(u16) },
433 	[OVS_KEY_ATTR_CT_MARK]	 = { .len = sizeof(u32) },
434 	[OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
435 	[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4] = {
436 		.len = sizeof(struct ovs_key_ct_tuple_ipv4) },
437 	[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6] = {
438 		.len = sizeof(struct ovs_key_ct_tuple_ipv6) },
439 	[OVS_KEY_ATTR_NSH]       = { .len = OVS_ATTR_NESTED,
440 				     .next = ovs_nsh_key_attr_lens, },
441 	[OVS_KEY_ATTR_IPV6_EXTHDRS] = {
442 		.len = sizeof(struct ovs_key_ipv6_exthdrs) },
443 };
444 
445 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
446 {
447 	return expected_len == attr_len ||
448 	       expected_len == OVS_ATTR_NESTED ||
449 	       expected_len == OVS_ATTR_VARIABLE;
450 }
451 
452 static bool is_all_zero(const u8 *fp, size_t size)
453 {
454 	int i;
455 
456 	if (!fp)
457 		return false;
458 
459 	for (i = 0; i < size; i++)
460 		if (fp[i])
461 			return false;
462 
463 	return true;
464 }
465 
466 static int __parse_flow_nlattrs(const struct nlattr *attr,
467 				const struct nlattr *a[],
468 				u64 *attrsp, bool log, bool nz)
469 {
470 	const struct nlattr *nla;
471 	u64 attrs;
472 	int rem;
473 
474 	attrs = *attrsp;
475 	nla_for_each_nested(nla, attr, rem) {
476 		u16 type = nla_type(nla);
477 		int expected_len;
478 
479 		if (type > OVS_KEY_ATTR_MAX) {
480 			OVS_NLERR(log, "Key type %d is out of range max %d",
481 				  type, OVS_KEY_ATTR_MAX);
482 			return -EINVAL;
483 		}
484 
485 		if (type == OVS_KEY_ATTR_PACKET_TYPE ||
486 		    type == OVS_KEY_ATTR_ND_EXTENSIONS ||
487 		    type == OVS_KEY_ATTR_TUNNEL_INFO) {
488 			OVS_NLERR(log, "Key type %d is not supported", type);
489 			return -EINVAL;
490 		}
491 
492 		if (attrs & (1ULL << type)) {
493 			OVS_NLERR(log, "Duplicate key (type %d).", type);
494 			return -EINVAL;
495 		}
496 
497 		expected_len = ovs_key_lens[type].len;
498 		if (!check_attr_len(nla_len(nla), expected_len)) {
499 			OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
500 				  type, nla_len(nla), expected_len);
501 			return -EINVAL;
502 		}
503 
504 		if (!nz || !is_all_zero(nla_data(nla), nla_len(nla))) {
505 			attrs |= 1ULL << type;
506 			a[type] = nla;
507 		}
508 	}
509 	if (rem) {
510 		OVS_NLERR(log, "Message has %d unknown bytes.", rem);
511 		return -EINVAL;
512 	}
513 
514 	*attrsp = attrs;
515 	return 0;
516 }
517 
518 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
519 				   const struct nlattr *a[], u64 *attrsp,
520 				   bool log)
521 {
522 	return __parse_flow_nlattrs(attr, a, attrsp, log, true);
523 }
524 
525 int parse_flow_nlattrs(const struct nlattr *attr, const struct nlattr *a[],
526 		       u64 *attrsp, bool log)
527 {
528 	return __parse_flow_nlattrs(attr, a, attrsp, log, false);
529 }
530 
531 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
532 				     struct sw_flow_match *match, bool is_mask,
533 				     bool log)
534 {
535 	unsigned long opt_key_offset;
536 
537 	if (nla_len(a) > sizeof(match->key->tun_opts)) {
538 		OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
539 			  nla_len(a), sizeof(match->key->tun_opts));
540 		return -EINVAL;
541 	}
542 
543 	if (nla_len(a) % 4 != 0) {
544 		OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
545 			  nla_len(a));
546 		return -EINVAL;
547 	}
548 
549 	/* We need to record the length of the options passed
550 	 * down, otherwise packets with the same format but
551 	 * additional options will be silently matched.
552 	 */
553 	if (!is_mask) {
554 		SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
555 				false);
556 	} else {
557 		/* This is somewhat unusual because it looks at
558 		 * both the key and mask while parsing the
559 		 * attributes (and by extension assumes the key
560 		 * is parsed first). Normally, we would verify
561 		 * that each is the correct length and that the
562 		 * attributes line up in the validate function.
563 		 * However, that is difficult because this is
564 		 * variable length and we won't have the
565 		 * information later.
566 		 */
567 		if (match->key->tun_opts_len != nla_len(a)) {
568 			OVS_NLERR(log, "Geneve option len %d != mask len %d",
569 				  match->key->tun_opts_len, nla_len(a));
570 			return -EINVAL;
571 		}
572 
573 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
574 	}
575 
576 	opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
577 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
578 				  nla_len(a), is_mask);
579 	return 0;
580 }
581 
582 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
583 				     struct sw_flow_match *match, bool is_mask,
584 				     bool log)
585 {
586 	struct nlattr *a;
587 	int rem;
588 	unsigned long opt_key_offset;
589 	struct vxlan_metadata opts;
590 
591 	BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
592 
593 	memset(&opts, 0, sizeof(opts));
594 	nla_for_each_nested(a, attr, rem) {
595 		int type = nla_type(a);
596 
597 		if (type > OVS_VXLAN_EXT_MAX) {
598 			OVS_NLERR(log, "VXLAN extension %d out of range max %d",
599 				  type, OVS_VXLAN_EXT_MAX);
600 			return -EINVAL;
601 		}
602 
603 		if (!check_attr_len(nla_len(a),
604 				    ovs_vxlan_ext_key_lens[type].len)) {
605 			OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
606 				  type, nla_len(a),
607 				  ovs_vxlan_ext_key_lens[type].len);
608 			return -EINVAL;
609 		}
610 
611 		switch (type) {
612 		case OVS_VXLAN_EXT_GBP:
613 			opts.gbp = nla_get_u32(a);
614 			break;
615 		default:
616 			OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
617 				  type);
618 			return -EINVAL;
619 		}
620 	}
621 	if (rem) {
622 		OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
623 			  rem);
624 		return -EINVAL;
625 	}
626 
627 	if (!is_mask)
628 		SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
629 	else
630 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
631 
632 	opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
633 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
634 				  is_mask);
635 	return 0;
636 }
637 
638 static int erspan_tun_opt_from_nlattr(const struct nlattr *a,
639 				      struct sw_flow_match *match, bool is_mask,
640 				      bool log)
641 {
642 	unsigned long opt_key_offset;
643 
644 	BUILD_BUG_ON(sizeof(struct erspan_metadata) >
645 		     sizeof(match->key->tun_opts));
646 
647 	if (nla_len(a) > sizeof(match->key->tun_opts)) {
648 		OVS_NLERR(log, "ERSPAN option length err (len %d, max %zu).",
649 			  nla_len(a), sizeof(match->key->tun_opts));
650 		return -EINVAL;
651 	}
652 
653 	if (!is_mask)
654 		SW_FLOW_KEY_PUT(match, tun_opts_len,
655 				sizeof(struct erspan_metadata), false);
656 	else
657 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
658 
659 	opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
660 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
661 				  nla_len(a), is_mask);
662 	return 0;
663 }
664 
665 static int ip_tun_from_nlattr(const struct nlattr *attr,
666 			      struct sw_flow_match *match, bool is_mask,
667 			      bool log)
668 {
669 	bool ttl = false, ipv4 = false, ipv6 = false;
670 	bool info_bridge_mode = false;
671 	__be16 tun_flags = 0;
672 	int opts_type = 0;
673 	struct nlattr *a;
674 	int rem;
675 
676 	nla_for_each_nested(a, attr, rem) {
677 		int type = nla_type(a);
678 		int err;
679 
680 		if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
681 			OVS_NLERR(log, "Tunnel attr %d out of range max %d",
682 				  type, OVS_TUNNEL_KEY_ATTR_MAX);
683 			return -EINVAL;
684 		}
685 
686 		if (!check_attr_len(nla_len(a),
687 				    ovs_tunnel_key_lens[type].len)) {
688 			OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
689 				  type, nla_len(a), ovs_tunnel_key_lens[type].len);
690 			return -EINVAL;
691 		}
692 
693 		switch (type) {
694 		case OVS_TUNNEL_KEY_ATTR_ID:
695 			SW_FLOW_KEY_PUT(match, tun_key.tun_id,
696 					nla_get_be64(a), is_mask);
697 			tun_flags |= TUNNEL_KEY;
698 			break;
699 		case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
700 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
701 					nla_get_in_addr(a), is_mask);
702 			ipv4 = true;
703 			break;
704 		case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
705 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
706 					nla_get_in_addr(a), is_mask);
707 			ipv4 = true;
708 			break;
709 		case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
710 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.src,
711 					nla_get_in6_addr(a), is_mask);
712 			ipv6 = true;
713 			break;
714 		case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
715 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
716 					nla_get_in6_addr(a), is_mask);
717 			ipv6 = true;
718 			break;
719 		case OVS_TUNNEL_KEY_ATTR_TOS:
720 			SW_FLOW_KEY_PUT(match, tun_key.tos,
721 					nla_get_u8(a), is_mask);
722 			break;
723 		case OVS_TUNNEL_KEY_ATTR_TTL:
724 			SW_FLOW_KEY_PUT(match, tun_key.ttl,
725 					nla_get_u8(a), is_mask);
726 			ttl = true;
727 			break;
728 		case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
729 			tun_flags |= TUNNEL_DONT_FRAGMENT;
730 			break;
731 		case OVS_TUNNEL_KEY_ATTR_CSUM:
732 			tun_flags |= TUNNEL_CSUM;
733 			break;
734 		case OVS_TUNNEL_KEY_ATTR_TP_SRC:
735 			SW_FLOW_KEY_PUT(match, tun_key.tp_src,
736 					nla_get_be16(a), is_mask);
737 			break;
738 		case OVS_TUNNEL_KEY_ATTR_TP_DST:
739 			SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
740 					nla_get_be16(a), is_mask);
741 			break;
742 		case OVS_TUNNEL_KEY_ATTR_OAM:
743 			tun_flags |= TUNNEL_OAM;
744 			break;
745 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
746 			if (opts_type) {
747 				OVS_NLERR(log, "Multiple metadata blocks provided");
748 				return -EINVAL;
749 			}
750 
751 			err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
752 			if (err)
753 				return err;
754 
755 			tun_flags |= TUNNEL_GENEVE_OPT;
756 			opts_type = type;
757 			break;
758 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
759 			if (opts_type) {
760 				OVS_NLERR(log, "Multiple metadata blocks provided");
761 				return -EINVAL;
762 			}
763 
764 			err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
765 			if (err)
766 				return err;
767 
768 			tun_flags |= TUNNEL_VXLAN_OPT;
769 			opts_type = type;
770 			break;
771 		case OVS_TUNNEL_KEY_ATTR_PAD:
772 			break;
773 		case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
774 			if (opts_type) {
775 				OVS_NLERR(log, "Multiple metadata blocks provided");
776 				return -EINVAL;
777 			}
778 
779 			err = erspan_tun_opt_from_nlattr(a, match, is_mask,
780 							 log);
781 			if (err)
782 				return err;
783 
784 			tun_flags |= TUNNEL_ERSPAN_OPT;
785 			opts_type = type;
786 			break;
787 		case OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE:
788 			info_bridge_mode = true;
789 			ipv4 = true;
790 			break;
791 		default:
792 			OVS_NLERR(log, "Unknown IP tunnel attribute %d",
793 				  type);
794 			return -EINVAL;
795 		}
796 	}
797 
798 	SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
799 	if (is_mask)
800 		SW_FLOW_KEY_MEMSET_FIELD(match, tun_proto, 0xff, true);
801 	else
802 		SW_FLOW_KEY_PUT(match, tun_proto, ipv6 ? AF_INET6 : AF_INET,
803 				false);
804 
805 	if (rem > 0) {
806 		OVS_NLERR(log, "IP tunnel attribute has %d unknown bytes.",
807 			  rem);
808 		return -EINVAL;
809 	}
810 
811 	if (ipv4 && ipv6) {
812 		OVS_NLERR(log, "Mixed IPv4 and IPv6 tunnel attributes");
813 		return -EINVAL;
814 	}
815 
816 	if (!is_mask) {
817 		if (!ipv4 && !ipv6) {
818 			OVS_NLERR(log, "IP tunnel dst address not specified");
819 			return -EINVAL;
820 		}
821 		if (ipv4) {
822 			if (info_bridge_mode) {
823 				if (match->key->tun_key.u.ipv4.src ||
824 				    match->key->tun_key.u.ipv4.dst ||
825 				    match->key->tun_key.tp_src ||
826 				    match->key->tun_key.tp_dst ||
827 				    match->key->tun_key.ttl ||
828 				    match->key->tun_key.tos ||
829 				    tun_flags & ~TUNNEL_KEY) {
830 					OVS_NLERR(log, "IPv4 tun info is not correct");
831 					return -EINVAL;
832 				}
833 			} else if (!match->key->tun_key.u.ipv4.dst) {
834 				OVS_NLERR(log, "IPv4 tunnel dst address is zero");
835 				return -EINVAL;
836 			}
837 		}
838 		if (ipv6 && ipv6_addr_any(&match->key->tun_key.u.ipv6.dst)) {
839 			OVS_NLERR(log, "IPv6 tunnel dst address is zero");
840 			return -EINVAL;
841 		}
842 
843 		if (!ttl && !info_bridge_mode) {
844 			OVS_NLERR(log, "IP tunnel TTL not specified.");
845 			return -EINVAL;
846 		}
847 	}
848 
849 	return opts_type;
850 }
851 
852 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
853 			       const void *tun_opts, int swkey_tun_opts_len)
854 {
855 	const struct vxlan_metadata *opts = tun_opts;
856 	struct nlattr *nla;
857 
858 	nla = nla_nest_start_noflag(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
859 	if (!nla)
860 		return -EMSGSIZE;
861 
862 	if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
863 		return -EMSGSIZE;
864 
865 	nla_nest_end(skb, nla);
866 	return 0;
867 }
868 
869 static int __ip_tun_to_nlattr(struct sk_buff *skb,
870 			      const struct ip_tunnel_key *output,
871 			      const void *tun_opts, int swkey_tun_opts_len,
872 			      unsigned short tun_proto, u8 mode)
873 {
874 	if (output->tun_flags & TUNNEL_KEY &&
875 	    nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id,
876 			 OVS_TUNNEL_KEY_ATTR_PAD))
877 		return -EMSGSIZE;
878 
879 	if (mode & IP_TUNNEL_INFO_BRIDGE)
880 		return nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE)
881 		       ? -EMSGSIZE : 0;
882 
883 	switch (tun_proto) {
884 	case AF_INET:
885 		if (output->u.ipv4.src &&
886 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
887 				    output->u.ipv4.src))
888 			return -EMSGSIZE;
889 		if (output->u.ipv4.dst &&
890 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
891 				    output->u.ipv4.dst))
892 			return -EMSGSIZE;
893 		break;
894 	case AF_INET6:
895 		if (!ipv6_addr_any(&output->u.ipv6.src) &&
896 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
897 				     &output->u.ipv6.src))
898 			return -EMSGSIZE;
899 		if (!ipv6_addr_any(&output->u.ipv6.dst) &&
900 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
901 				     &output->u.ipv6.dst))
902 			return -EMSGSIZE;
903 		break;
904 	}
905 	if (output->tos &&
906 	    nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
907 		return -EMSGSIZE;
908 	if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
909 		return -EMSGSIZE;
910 	if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
911 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
912 		return -EMSGSIZE;
913 	if ((output->tun_flags & TUNNEL_CSUM) &&
914 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
915 		return -EMSGSIZE;
916 	if (output->tp_src &&
917 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
918 		return -EMSGSIZE;
919 	if (output->tp_dst &&
920 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
921 		return -EMSGSIZE;
922 	if ((output->tun_flags & TUNNEL_OAM) &&
923 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
924 		return -EMSGSIZE;
925 	if (swkey_tun_opts_len) {
926 		if (output->tun_flags & TUNNEL_GENEVE_OPT &&
927 		    nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
928 			    swkey_tun_opts_len, tun_opts))
929 			return -EMSGSIZE;
930 		else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
931 			 vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
932 			return -EMSGSIZE;
933 		else if (output->tun_flags & TUNNEL_ERSPAN_OPT &&
934 			 nla_put(skb, OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS,
935 				 swkey_tun_opts_len, tun_opts))
936 			return -EMSGSIZE;
937 	}
938 
939 	return 0;
940 }
941 
942 static int ip_tun_to_nlattr(struct sk_buff *skb,
943 			    const struct ip_tunnel_key *output,
944 			    const void *tun_opts, int swkey_tun_opts_len,
945 			    unsigned short tun_proto, u8 mode)
946 {
947 	struct nlattr *nla;
948 	int err;
949 
950 	nla = nla_nest_start_noflag(skb, OVS_KEY_ATTR_TUNNEL);
951 	if (!nla)
952 		return -EMSGSIZE;
953 
954 	err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
955 				 tun_proto, mode);
956 	if (err)
957 		return err;
958 
959 	nla_nest_end(skb, nla);
960 	return 0;
961 }
962 
963 int ovs_nla_put_tunnel_info(struct sk_buff *skb,
964 			    struct ip_tunnel_info *tun_info)
965 {
966 	return __ip_tun_to_nlattr(skb, &tun_info->key,
967 				  ip_tunnel_info_opts(tun_info),
968 				  tun_info->options_len,
969 				  ip_tunnel_info_af(tun_info), tun_info->mode);
970 }
971 
972 static int encode_vlan_from_nlattrs(struct sw_flow_match *match,
973 				    const struct nlattr *a[],
974 				    bool is_mask, bool inner)
975 {
976 	__be16 tci = 0;
977 	__be16 tpid = 0;
978 
979 	if (a[OVS_KEY_ATTR_VLAN])
980 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
981 
982 	if (a[OVS_KEY_ATTR_ETHERTYPE])
983 		tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
984 
985 	if (likely(!inner)) {
986 		SW_FLOW_KEY_PUT(match, eth.vlan.tpid, tpid, is_mask);
987 		SW_FLOW_KEY_PUT(match, eth.vlan.tci, tci, is_mask);
988 	} else {
989 		SW_FLOW_KEY_PUT(match, eth.cvlan.tpid, tpid, is_mask);
990 		SW_FLOW_KEY_PUT(match, eth.cvlan.tci, tci, is_mask);
991 	}
992 	return 0;
993 }
994 
995 static int validate_vlan_from_nlattrs(const struct sw_flow_match *match,
996 				      u64 key_attrs, bool inner,
997 				      const struct nlattr **a, bool log)
998 {
999 	__be16 tci = 0;
1000 
1001 	if (!((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1002 	      (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1003 	       eth_type_vlan(nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE])))) {
1004 		/* Not a VLAN. */
1005 		return 0;
1006 	}
1007 
1008 	if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1009 	      (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1010 		OVS_NLERR(log, "Invalid %s frame", (inner) ? "C-VLAN" : "VLAN");
1011 		return -EINVAL;
1012 	}
1013 
1014 	if (a[OVS_KEY_ATTR_VLAN])
1015 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1016 
1017 	if (!(tci & htons(VLAN_CFI_MASK))) {
1018 		if (tci) {
1019 			OVS_NLERR(log, "%s TCI does not have VLAN_CFI_MASK bit set.",
1020 				  (inner) ? "C-VLAN" : "VLAN");
1021 			return -EINVAL;
1022 		} else if (nla_len(a[OVS_KEY_ATTR_ENCAP])) {
1023 			/* Corner case for truncated VLAN header. */
1024 			OVS_NLERR(log, "Truncated %s header has non-zero encap attribute.",
1025 				  (inner) ? "C-VLAN" : "VLAN");
1026 			return -EINVAL;
1027 		}
1028 	}
1029 
1030 	return 1;
1031 }
1032 
1033 static int validate_vlan_mask_from_nlattrs(const struct sw_flow_match *match,
1034 					   u64 key_attrs, bool inner,
1035 					   const struct nlattr **a, bool log)
1036 {
1037 	__be16 tci = 0;
1038 	__be16 tpid = 0;
1039 	bool encap_valid = !!(match->key->eth.vlan.tci &
1040 			      htons(VLAN_CFI_MASK));
1041 	bool i_encap_valid = !!(match->key->eth.cvlan.tci &
1042 				htons(VLAN_CFI_MASK));
1043 
1044 	if (!(key_attrs & (1 << OVS_KEY_ATTR_ENCAP))) {
1045 		/* Not a VLAN. */
1046 		return 0;
1047 	}
1048 
1049 	if ((!inner && !encap_valid) || (inner && !i_encap_valid)) {
1050 		OVS_NLERR(log, "Encap mask attribute is set for non-%s frame.",
1051 			  (inner) ? "C-VLAN" : "VLAN");
1052 		return -EINVAL;
1053 	}
1054 
1055 	if (a[OVS_KEY_ATTR_VLAN])
1056 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1057 
1058 	if (a[OVS_KEY_ATTR_ETHERTYPE])
1059 		tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1060 
1061 	if (tpid != htons(0xffff)) {
1062 		OVS_NLERR(log, "Must have an exact match on %s TPID (mask=%x).",
1063 			  (inner) ? "C-VLAN" : "VLAN", ntohs(tpid));
1064 		return -EINVAL;
1065 	}
1066 	if (!(tci & htons(VLAN_CFI_MASK))) {
1067 		OVS_NLERR(log, "%s TCI mask does not have exact match for VLAN_CFI_MASK bit.",
1068 			  (inner) ? "C-VLAN" : "VLAN");
1069 		return -EINVAL;
1070 	}
1071 
1072 	return 1;
1073 }
1074 
1075 static int __parse_vlan_from_nlattrs(struct sw_flow_match *match,
1076 				     u64 *key_attrs, bool inner,
1077 				     const struct nlattr **a, bool is_mask,
1078 				     bool log)
1079 {
1080 	int err;
1081 	const struct nlattr *encap;
1082 
1083 	if (!is_mask)
1084 		err = validate_vlan_from_nlattrs(match, *key_attrs, inner,
1085 						 a, log);
1086 	else
1087 		err = validate_vlan_mask_from_nlattrs(match, *key_attrs, inner,
1088 						      a, log);
1089 	if (err <= 0)
1090 		return err;
1091 
1092 	err = encode_vlan_from_nlattrs(match, a, is_mask, inner);
1093 	if (err)
1094 		return err;
1095 
1096 	*key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1097 	*key_attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
1098 	*key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1099 
1100 	encap = a[OVS_KEY_ATTR_ENCAP];
1101 
1102 	if (!is_mask)
1103 		err = parse_flow_nlattrs(encap, a, key_attrs, log);
1104 	else
1105 		err = parse_flow_mask_nlattrs(encap, a, key_attrs, log);
1106 
1107 	return err;
1108 }
1109 
1110 static int parse_vlan_from_nlattrs(struct sw_flow_match *match,
1111 				   u64 *key_attrs, const struct nlattr **a,
1112 				   bool is_mask, bool log)
1113 {
1114 	int err;
1115 	bool encap_valid = false;
1116 
1117 	err = __parse_vlan_from_nlattrs(match, key_attrs, false, a,
1118 					is_mask, log);
1119 	if (err)
1120 		return err;
1121 
1122 	encap_valid = !!(match->key->eth.vlan.tci & htons(VLAN_CFI_MASK));
1123 	if (encap_valid) {
1124 		err = __parse_vlan_from_nlattrs(match, key_attrs, true, a,
1125 						is_mask, log);
1126 		if (err)
1127 			return err;
1128 	}
1129 
1130 	return 0;
1131 }
1132 
1133 static int parse_eth_type_from_nlattrs(struct sw_flow_match *match,
1134 				       u64 *attrs, const struct nlattr **a,
1135 				       bool is_mask, bool log)
1136 {
1137 	__be16 eth_type;
1138 
1139 	eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1140 	if (is_mask) {
1141 		/* Always exact match EtherType. */
1142 		eth_type = htons(0xffff);
1143 	} else if (!eth_proto_is_802_3(eth_type)) {
1144 		OVS_NLERR(log, "EtherType %x is less than min %x",
1145 				ntohs(eth_type), ETH_P_802_3_MIN);
1146 		return -EINVAL;
1147 	}
1148 
1149 	SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1150 	*attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1151 	return 0;
1152 }
1153 
1154 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
1155 				 u64 *attrs, const struct nlattr **a,
1156 				 bool is_mask, bool log)
1157 {
1158 	u8 mac_proto = MAC_PROTO_ETHERNET;
1159 
1160 	if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
1161 		u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
1162 
1163 		SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
1164 		*attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
1165 	}
1166 
1167 	if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
1168 		u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
1169 
1170 		SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
1171 		*attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
1172 	}
1173 
1174 	if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1175 		SW_FLOW_KEY_PUT(match, phy.priority,
1176 			  nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1177 		*attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1178 	}
1179 
1180 	if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1181 		u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1182 
1183 		if (is_mask) {
1184 			in_port = 0xffffffff; /* Always exact match in_port. */
1185 		} else if (in_port >= DP_MAX_PORTS) {
1186 			OVS_NLERR(log, "Port %d exceeds max allowable %d",
1187 				  in_port, DP_MAX_PORTS);
1188 			return -EINVAL;
1189 		}
1190 
1191 		SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1192 		*attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1193 	} else if (!is_mask) {
1194 		SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1195 	}
1196 
1197 	if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1198 		uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1199 
1200 		SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1201 		*attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1202 	}
1203 	if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1204 		if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1205 				       is_mask, log) < 0)
1206 			return -EINVAL;
1207 		*attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1208 	}
1209 
1210 	if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
1211 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
1212 		u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
1213 
1214 		if (ct_state & ~CT_SUPPORTED_MASK) {
1215 			OVS_NLERR(log, "ct_state flags %08x unsupported",
1216 				  ct_state);
1217 			return -EINVAL;
1218 		}
1219 
1220 		SW_FLOW_KEY_PUT(match, ct_state, ct_state, is_mask);
1221 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
1222 	}
1223 	if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
1224 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
1225 		u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
1226 
1227 		SW_FLOW_KEY_PUT(match, ct_zone, ct_zone, is_mask);
1228 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
1229 	}
1230 	if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
1231 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
1232 		u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
1233 
1234 		SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
1235 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
1236 	}
1237 	if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
1238 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
1239 		const struct ovs_key_ct_labels *cl;
1240 
1241 		cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
1242 		SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
1243 				   sizeof(*cl), is_mask);
1244 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
1245 	}
1246 	if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)) {
1247 		const struct ovs_key_ct_tuple_ipv4 *ct;
1248 
1249 		ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4]);
1250 
1251 		SW_FLOW_KEY_PUT(match, ipv4.ct_orig.src, ct->ipv4_src, is_mask);
1252 		SW_FLOW_KEY_PUT(match, ipv4.ct_orig.dst, ct->ipv4_dst, is_mask);
1253 		SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1254 		SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1255 		SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv4_proto, is_mask);
1256 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4);
1257 	}
1258 	if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)) {
1259 		const struct ovs_key_ct_tuple_ipv6 *ct;
1260 
1261 		ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6]);
1262 
1263 		SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.src, &ct->ipv6_src,
1264 				   sizeof(match->key->ipv6.ct_orig.src),
1265 				   is_mask);
1266 		SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.dst, &ct->ipv6_dst,
1267 				   sizeof(match->key->ipv6.ct_orig.dst),
1268 				   is_mask);
1269 		SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1270 		SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1271 		SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv6_proto, is_mask);
1272 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
1273 	}
1274 
1275 	/* For layer 3 packets the Ethernet type is provided
1276 	 * and treated as metadata but no MAC addresses are provided.
1277 	 */
1278 	if (!(*attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1279 	    (*attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)))
1280 		mac_proto = MAC_PROTO_NONE;
1281 
1282 	/* Always exact match mac_proto */
1283 	SW_FLOW_KEY_PUT(match, mac_proto, is_mask ? 0xff : mac_proto, is_mask);
1284 
1285 	if (mac_proto == MAC_PROTO_NONE)
1286 		return parse_eth_type_from_nlattrs(match, attrs, a, is_mask,
1287 						   log);
1288 
1289 	return 0;
1290 }
1291 
1292 int nsh_hdr_from_nlattr(const struct nlattr *attr,
1293 			struct nshhdr *nh, size_t size)
1294 {
1295 	struct nlattr *a;
1296 	int rem;
1297 	u8 flags = 0;
1298 	u8 ttl = 0;
1299 	int mdlen = 0;
1300 
1301 	/* validate_nsh has check this, so we needn't do duplicate check here
1302 	 */
1303 	if (size < NSH_BASE_HDR_LEN)
1304 		return -ENOBUFS;
1305 
1306 	nla_for_each_nested(a, attr, rem) {
1307 		int type = nla_type(a);
1308 
1309 		switch (type) {
1310 		case OVS_NSH_KEY_ATTR_BASE: {
1311 			const struct ovs_nsh_key_base *base = nla_data(a);
1312 
1313 			flags = base->flags;
1314 			ttl = base->ttl;
1315 			nh->np = base->np;
1316 			nh->mdtype = base->mdtype;
1317 			nh->path_hdr = base->path_hdr;
1318 			break;
1319 		}
1320 		case OVS_NSH_KEY_ATTR_MD1:
1321 			mdlen = nla_len(a);
1322 			if (mdlen > size - NSH_BASE_HDR_LEN)
1323 				return -ENOBUFS;
1324 			memcpy(&nh->md1, nla_data(a), mdlen);
1325 			break;
1326 
1327 		case OVS_NSH_KEY_ATTR_MD2:
1328 			mdlen = nla_len(a);
1329 			if (mdlen > size - NSH_BASE_HDR_LEN)
1330 				return -ENOBUFS;
1331 			memcpy(&nh->md2, nla_data(a), mdlen);
1332 			break;
1333 
1334 		default:
1335 			return -EINVAL;
1336 		}
1337 	}
1338 
1339 	/* nsh header length  = NSH_BASE_HDR_LEN + mdlen */
1340 	nh->ver_flags_ttl_len = 0;
1341 	nsh_set_flags_ttl_len(nh, flags, ttl, NSH_BASE_HDR_LEN + mdlen);
1342 
1343 	return 0;
1344 }
1345 
1346 int nsh_key_from_nlattr(const struct nlattr *attr,
1347 			struct ovs_key_nsh *nsh, struct ovs_key_nsh *nsh_mask)
1348 {
1349 	struct nlattr *a;
1350 	int rem;
1351 
1352 	/* validate_nsh has check this, so we needn't do duplicate check here
1353 	 */
1354 	nla_for_each_nested(a, attr, rem) {
1355 		int type = nla_type(a);
1356 
1357 		switch (type) {
1358 		case OVS_NSH_KEY_ATTR_BASE: {
1359 			const struct ovs_nsh_key_base *base = nla_data(a);
1360 			const struct ovs_nsh_key_base *base_mask = base + 1;
1361 
1362 			nsh->base = *base;
1363 			nsh_mask->base = *base_mask;
1364 			break;
1365 		}
1366 		case OVS_NSH_KEY_ATTR_MD1: {
1367 			const struct ovs_nsh_key_md1 *md1 = nla_data(a);
1368 			const struct ovs_nsh_key_md1 *md1_mask = md1 + 1;
1369 
1370 			memcpy(nsh->context, md1->context, sizeof(*md1));
1371 			memcpy(nsh_mask->context, md1_mask->context,
1372 			       sizeof(*md1_mask));
1373 			break;
1374 		}
1375 		case OVS_NSH_KEY_ATTR_MD2:
1376 			/* Not supported yet */
1377 			return -ENOTSUPP;
1378 		default:
1379 			return -EINVAL;
1380 		}
1381 	}
1382 
1383 	return 0;
1384 }
1385 
1386 static int nsh_key_put_from_nlattr(const struct nlattr *attr,
1387 				   struct sw_flow_match *match, bool is_mask,
1388 				   bool is_push_nsh, bool log)
1389 {
1390 	struct nlattr *a;
1391 	int rem;
1392 	bool has_base = false;
1393 	bool has_md1 = false;
1394 	bool has_md2 = false;
1395 	u8 mdtype = 0;
1396 	int mdlen = 0;
1397 
1398 	if (WARN_ON(is_push_nsh && is_mask))
1399 		return -EINVAL;
1400 
1401 	nla_for_each_nested(a, attr, rem) {
1402 		int type = nla_type(a);
1403 		int i;
1404 
1405 		if (type > OVS_NSH_KEY_ATTR_MAX) {
1406 			OVS_NLERR(log, "nsh attr %d is out of range max %d",
1407 				  type, OVS_NSH_KEY_ATTR_MAX);
1408 			return -EINVAL;
1409 		}
1410 
1411 		if (!check_attr_len(nla_len(a),
1412 				    ovs_nsh_key_attr_lens[type].len)) {
1413 			OVS_NLERR(
1414 			    log,
1415 			    "nsh attr %d has unexpected len %d expected %d",
1416 			    type,
1417 			    nla_len(a),
1418 			    ovs_nsh_key_attr_lens[type].len
1419 			);
1420 			return -EINVAL;
1421 		}
1422 
1423 		switch (type) {
1424 		case OVS_NSH_KEY_ATTR_BASE: {
1425 			const struct ovs_nsh_key_base *base = nla_data(a);
1426 
1427 			has_base = true;
1428 			mdtype = base->mdtype;
1429 			SW_FLOW_KEY_PUT(match, nsh.base.flags,
1430 					base->flags, is_mask);
1431 			SW_FLOW_KEY_PUT(match, nsh.base.ttl,
1432 					base->ttl, is_mask);
1433 			SW_FLOW_KEY_PUT(match, nsh.base.mdtype,
1434 					base->mdtype, is_mask);
1435 			SW_FLOW_KEY_PUT(match, nsh.base.np,
1436 					base->np, is_mask);
1437 			SW_FLOW_KEY_PUT(match, nsh.base.path_hdr,
1438 					base->path_hdr, is_mask);
1439 			break;
1440 		}
1441 		case OVS_NSH_KEY_ATTR_MD1: {
1442 			const struct ovs_nsh_key_md1 *md1 = nla_data(a);
1443 
1444 			has_md1 = true;
1445 			for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++)
1446 				SW_FLOW_KEY_PUT(match, nsh.context[i],
1447 						md1->context[i], is_mask);
1448 			break;
1449 		}
1450 		case OVS_NSH_KEY_ATTR_MD2:
1451 			if (!is_push_nsh) /* Not supported MD type 2 yet */
1452 				return -ENOTSUPP;
1453 
1454 			has_md2 = true;
1455 			mdlen = nla_len(a);
1456 			if (mdlen > NSH_CTX_HDRS_MAX_LEN || mdlen <= 0) {
1457 				OVS_NLERR(
1458 				    log,
1459 				    "Invalid MD length %d for MD type %d",
1460 				    mdlen,
1461 				    mdtype
1462 				);
1463 				return -EINVAL;
1464 			}
1465 			break;
1466 		default:
1467 			OVS_NLERR(log, "Unknown nsh attribute %d",
1468 				  type);
1469 			return -EINVAL;
1470 		}
1471 	}
1472 
1473 	if (rem > 0) {
1474 		OVS_NLERR(log, "nsh attribute has %d unknown bytes.", rem);
1475 		return -EINVAL;
1476 	}
1477 
1478 	if (has_md1 && has_md2) {
1479 		OVS_NLERR(
1480 		    1,
1481 		    "invalid nsh attribute: md1 and md2 are exclusive."
1482 		);
1483 		return -EINVAL;
1484 	}
1485 
1486 	if (!is_mask) {
1487 		if ((has_md1 && mdtype != NSH_M_TYPE1) ||
1488 		    (has_md2 && mdtype != NSH_M_TYPE2)) {
1489 			OVS_NLERR(1, "nsh attribute has unmatched MD type %d.",
1490 				  mdtype);
1491 			return -EINVAL;
1492 		}
1493 
1494 		if (is_push_nsh &&
1495 		    (!has_base || (!has_md1 && !has_md2))) {
1496 			OVS_NLERR(
1497 			    1,
1498 			    "push_nsh: missing base or metadata attributes"
1499 			);
1500 			return -EINVAL;
1501 		}
1502 	}
1503 
1504 	return 0;
1505 }
1506 
1507 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
1508 				u64 attrs, const struct nlattr **a,
1509 				bool is_mask, bool log)
1510 {
1511 	int err;
1512 
1513 	err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
1514 	if (err)
1515 		return err;
1516 
1517 	if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
1518 		const struct ovs_key_ethernet *eth_key;
1519 
1520 		eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1521 		SW_FLOW_KEY_MEMCPY(match, eth.src,
1522 				eth_key->eth_src, ETH_ALEN, is_mask);
1523 		SW_FLOW_KEY_MEMCPY(match, eth.dst,
1524 				eth_key->eth_dst, ETH_ALEN, is_mask);
1525 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1526 
1527 		if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
1528 			/* VLAN attribute is always parsed before getting here since it
1529 			 * may occur multiple times.
1530 			 */
1531 			OVS_NLERR(log, "VLAN attribute unexpected.");
1532 			return -EINVAL;
1533 		}
1534 
1535 		if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1536 			err = parse_eth_type_from_nlattrs(match, &attrs, a, is_mask,
1537 							  log);
1538 			if (err)
1539 				return err;
1540 		} else if (!is_mask) {
1541 			SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1542 		}
1543 	} else if (!match->key->eth.type) {
1544 		OVS_NLERR(log, "Either Ethernet header or EtherType is required.");
1545 		return -EINVAL;
1546 	}
1547 
1548 	if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1549 		const struct ovs_key_ipv4 *ipv4_key;
1550 
1551 		ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1552 		if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1553 			OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
1554 				  ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1555 			return -EINVAL;
1556 		}
1557 		SW_FLOW_KEY_PUT(match, ip.proto,
1558 				ipv4_key->ipv4_proto, is_mask);
1559 		SW_FLOW_KEY_PUT(match, ip.tos,
1560 				ipv4_key->ipv4_tos, is_mask);
1561 		SW_FLOW_KEY_PUT(match, ip.ttl,
1562 				ipv4_key->ipv4_ttl, is_mask);
1563 		SW_FLOW_KEY_PUT(match, ip.frag,
1564 				ipv4_key->ipv4_frag, is_mask);
1565 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1566 				ipv4_key->ipv4_src, is_mask);
1567 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1568 				ipv4_key->ipv4_dst, is_mask);
1569 		attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1570 	}
1571 
1572 	if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
1573 		const struct ovs_key_ipv6 *ipv6_key;
1574 
1575 		ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1576 		if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1577 			OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
1578 				  ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1579 			return -EINVAL;
1580 		}
1581 
1582 		if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
1583 			OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x)",
1584 				  ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
1585 			return -EINVAL;
1586 		}
1587 
1588 		SW_FLOW_KEY_PUT(match, ipv6.label,
1589 				ipv6_key->ipv6_label, is_mask);
1590 		SW_FLOW_KEY_PUT(match, ip.proto,
1591 				ipv6_key->ipv6_proto, is_mask);
1592 		SW_FLOW_KEY_PUT(match, ip.tos,
1593 				ipv6_key->ipv6_tclass, is_mask);
1594 		SW_FLOW_KEY_PUT(match, ip.ttl,
1595 				ipv6_key->ipv6_hlimit, is_mask);
1596 		SW_FLOW_KEY_PUT(match, ip.frag,
1597 				ipv6_key->ipv6_frag, is_mask);
1598 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1599 				ipv6_key->ipv6_src,
1600 				sizeof(match->key->ipv6.addr.src),
1601 				is_mask);
1602 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1603 				ipv6_key->ipv6_dst,
1604 				sizeof(match->key->ipv6.addr.dst),
1605 				is_mask);
1606 
1607 		attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1608 	}
1609 
1610 	if (attrs & (1ULL << OVS_KEY_ATTR_IPV6_EXTHDRS)) {
1611 		const struct ovs_key_ipv6_exthdrs *ipv6_exthdrs_key;
1612 
1613 		ipv6_exthdrs_key = nla_data(a[OVS_KEY_ATTR_IPV6_EXTHDRS]);
1614 
1615 		SW_FLOW_KEY_PUT(match, ipv6.exthdrs,
1616 				ipv6_exthdrs_key->hdrs, is_mask);
1617 
1618 		attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6_EXTHDRS);
1619 	}
1620 
1621 	if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1622 		const struct ovs_key_arp *arp_key;
1623 
1624 		arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1625 		if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1626 			OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1627 				  arp_key->arp_op);
1628 			return -EINVAL;
1629 		}
1630 
1631 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1632 				arp_key->arp_sip, is_mask);
1633 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1634 			arp_key->arp_tip, is_mask);
1635 		SW_FLOW_KEY_PUT(match, ip.proto,
1636 				ntohs(arp_key->arp_op), is_mask);
1637 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1638 				arp_key->arp_sha, ETH_ALEN, is_mask);
1639 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1640 				arp_key->arp_tha, ETH_ALEN, is_mask);
1641 
1642 		attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1643 	}
1644 
1645 	if (attrs & (1 << OVS_KEY_ATTR_NSH)) {
1646 		if (nsh_key_put_from_nlattr(a[OVS_KEY_ATTR_NSH], match,
1647 					    is_mask, false, log) < 0)
1648 			return -EINVAL;
1649 		attrs &= ~(1 << OVS_KEY_ATTR_NSH);
1650 	}
1651 
1652 	if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1653 		const struct ovs_key_mpls *mpls_key;
1654 		u32 hdr_len;
1655 		u32 label_count, label_count_mask, i;
1656 
1657 		mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1658 		hdr_len = nla_len(a[OVS_KEY_ATTR_MPLS]);
1659 		label_count = hdr_len / sizeof(struct ovs_key_mpls);
1660 
1661 		if (label_count == 0 || label_count > MPLS_LABEL_DEPTH ||
1662 		    hdr_len % sizeof(struct ovs_key_mpls))
1663 			return -EINVAL;
1664 
1665 		label_count_mask =  GENMASK(label_count - 1, 0);
1666 
1667 		for (i = 0 ; i < label_count; i++)
1668 			SW_FLOW_KEY_PUT(match, mpls.lse[i],
1669 					mpls_key[i].mpls_lse, is_mask);
1670 
1671 		SW_FLOW_KEY_PUT(match, mpls.num_labels_mask,
1672 				label_count_mask, is_mask);
1673 
1674 		attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1675 	 }
1676 
1677 	if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1678 		const struct ovs_key_tcp *tcp_key;
1679 
1680 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1681 		SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1682 		SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1683 		attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1684 	}
1685 
1686 	if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1687 		SW_FLOW_KEY_PUT(match, tp.flags,
1688 				nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1689 				is_mask);
1690 		attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1691 	}
1692 
1693 	if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1694 		const struct ovs_key_udp *udp_key;
1695 
1696 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1697 		SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1698 		SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1699 		attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1700 	}
1701 
1702 	if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1703 		const struct ovs_key_sctp *sctp_key;
1704 
1705 		sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1706 		SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1707 		SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1708 		attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1709 	}
1710 
1711 	if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1712 		const struct ovs_key_icmp *icmp_key;
1713 
1714 		icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1715 		SW_FLOW_KEY_PUT(match, tp.src,
1716 				htons(icmp_key->icmp_type), is_mask);
1717 		SW_FLOW_KEY_PUT(match, tp.dst,
1718 				htons(icmp_key->icmp_code), is_mask);
1719 		attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1720 	}
1721 
1722 	if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1723 		const struct ovs_key_icmpv6 *icmpv6_key;
1724 
1725 		icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1726 		SW_FLOW_KEY_PUT(match, tp.src,
1727 				htons(icmpv6_key->icmpv6_type), is_mask);
1728 		SW_FLOW_KEY_PUT(match, tp.dst,
1729 				htons(icmpv6_key->icmpv6_code), is_mask);
1730 		attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1731 	}
1732 
1733 	if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1734 		const struct ovs_key_nd *nd_key;
1735 
1736 		nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1737 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1738 			nd_key->nd_target,
1739 			sizeof(match->key->ipv6.nd.target),
1740 			is_mask);
1741 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1742 			nd_key->nd_sll, ETH_ALEN, is_mask);
1743 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1744 				nd_key->nd_tll, ETH_ALEN, is_mask);
1745 		attrs &= ~(1 << OVS_KEY_ATTR_ND);
1746 	}
1747 
1748 	if (attrs != 0) {
1749 		OVS_NLERR(log, "Unknown key attributes %llx",
1750 			  (unsigned long long)attrs);
1751 		return -EINVAL;
1752 	}
1753 
1754 	return 0;
1755 }
1756 
1757 static void nlattr_set(struct nlattr *attr, u8 val,
1758 		       const struct ovs_len_tbl *tbl)
1759 {
1760 	struct nlattr *nla;
1761 	int rem;
1762 
1763 	/* The nlattr stream should already have been validated */
1764 	nla_for_each_nested(nla, attr, rem) {
1765 		if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
1766 			nlattr_set(nla, val, tbl[nla_type(nla)].next ? : tbl);
1767 		else
1768 			memset(nla_data(nla), val, nla_len(nla));
1769 
1770 		if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1771 			*(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1772 	}
1773 }
1774 
1775 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1776 {
1777 	nlattr_set(attr, val, ovs_key_lens);
1778 }
1779 
1780 /**
1781  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1782  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1783  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1784  * does not include any don't care bit.
1785  * @net: Used to determine per-namespace field support.
1786  * @match: receives the extracted flow match information.
1787  * @nla_key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1788  * sequence. The fields should of the packet that triggered the creation
1789  * of this flow.
1790  * @nla_mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_*
1791  * Netlink attribute specifies the mask field of the wildcarded flow.
1792  * @log: Boolean to allow kernel error logging.  Normally true, but when
1793  * probing for feature compatibility this should be passed in as false to
1794  * suppress unnecessary error logging.
1795  */
1796 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1797 		      const struct nlattr *nla_key,
1798 		      const struct nlattr *nla_mask,
1799 		      bool log)
1800 {
1801 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1802 	struct nlattr *newmask = NULL;
1803 	u64 key_attrs = 0;
1804 	u64 mask_attrs = 0;
1805 	int err;
1806 
1807 	err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1808 	if (err)
1809 		return err;
1810 
1811 	err = parse_vlan_from_nlattrs(match, &key_attrs, a, false, log);
1812 	if (err)
1813 		return err;
1814 
1815 	err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1816 	if (err)
1817 		return err;
1818 
1819 	if (match->mask) {
1820 		if (!nla_mask) {
1821 			/* Create an exact match mask. We need to set to 0xff
1822 			 * all the 'match->mask' fields that have been touched
1823 			 * in 'match->key'. We cannot simply memset
1824 			 * 'match->mask', because padding bytes and fields not
1825 			 * specified in 'match->key' should be left to 0.
1826 			 * Instead, we use a stream of netlink attributes,
1827 			 * copied from 'key' and set to 0xff.
1828 			 * ovs_key_from_nlattrs() will take care of filling
1829 			 * 'match->mask' appropriately.
1830 			 */
1831 			newmask = kmemdup(nla_key,
1832 					  nla_total_size(nla_len(nla_key)),
1833 					  GFP_KERNEL);
1834 			if (!newmask)
1835 				return -ENOMEM;
1836 
1837 			mask_set_nlattr(newmask, 0xff);
1838 
1839 			/* The userspace does not send tunnel attributes that
1840 			 * are 0, but we should not wildcard them nonetheless.
1841 			 */
1842 			if (match->key->tun_proto)
1843 				SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1844 							 0xff, true);
1845 
1846 			nla_mask = newmask;
1847 		}
1848 
1849 		err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1850 		if (err)
1851 			goto free_newmask;
1852 
1853 		/* Always match on tci. */
1854 		SW_FLOW_KEY_PUT(match, eth.vlan.tci, htons(0xffff), true);
1855 		SW_FLOW_KEY_PUT(match, eth.cvlan.tci, htons(0xffff), true);
1856 
1857 		err = parse_vlan_from_nlattrs(match, &mask_attrs, a, true, log);
1858 		if (err)
1859 			goto free_newmask;
1860 
1861 		err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1862 					   log);
1863 		if (err)
1864 			goto free_newmask;
1865 	}
1866 
1867 	if (!match_validate(match, key_attrs, mask_attrs, log))
1868 		err = -EINVAL;
1869 
1870 free_newmask:
1871 	kfree(newmask);
1872 	return err;
1873 }
1874 
1875 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1876 {
1877 	size_t len;
1878 
1879 	if (!attr)
1880 		return 0;
1881 
1882 	len = nla_len(attr);
1883 	if (len < 1 || len > MAX_UFID_LENGTH) {
1884 		OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1885 			  nla_len(attr), MAX_UFID_LENGTH);
1886 		return 0;
1887 	}
1888 
1889 	return len;
1890 }
1891 
1892 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1893  * or false otherwise.
1894  */
1895 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1896 		      bool log)
1897 {
1898 	sfid->ufid_len = get_ufid_len(attr, log);
1899 	if (sfid->ufid_len)
1900 		memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1901 
1902 	return sfid->ufid_len;
1903 }
1904 
1905 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1906 			   const struct sw_flow_key *key, bool log)
1907 {
1908 	struct sw_flow_key *new_key;
1909 
1910 	if (ovs_nla_get_ufid(sfid, ufid, log))
1911 		return 0;
1912 
1913 	/* If UFID was not provided, use unmasked key. */
1914 	new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1915 	if (!new_key)
1916 		return -ENOMEM;
1917 	memcpy(new_key, key, sizeof(*key));
1918 	sfid->unmasked_key = new_key;
1919 
1920 	return 0;
1921 }
1922 
1923 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1924 {
1925 	return attr ? nla_get_u32(attr) : 0;
1926 }
1927 
1928 /**
1929  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1930  * @net: Network namespace.
1931  * @key: Receives extracted in_port, priority, tun_key, skb_mark and conntrack
1932  * metadata.
1933  * @a: Array of netlink attributes holding parsed %OVS_KEY_ATTR_* Netlink
1934  * attributes.
1935  * @attrs: Bit mask for the netlink attributes included in @a.
1936  * @log: Boolean to allow kernel error logging.  Normally true, but when
1937  * probing for feature compatibility this should be passed in as false to
1938  * suppress unnecessary error logging.
1939  *
1940  * This parses a series of Netlink attributes that form a flow key, which must
1941  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1942  * get the metadata, that is, the parts of the flow key that cannot be
1943  * extracted from the packet itself.
1944  *
1945  * This must be called before the packet key fields are filled in 'key'.
1946  */
1947 
1948 int ovs_nla_get_flow_metadata(struct net *net,
1949 			      const struct nlattr *a[OVS_KEY_ATTR_MAX + 1],
1950 			      u64 attrs, struct sw_flow_key *key, bool log)
1951 {
1952 	struct sw_flow_match match;
1953 
1954 	memset(&match, 0, sizeof(match));
1955 	match.key = key;
1956 
1957 	key->ct_state = 0;
1958 	key->ct_zone = 0;
1959 	key->ct_orig_proto = 0;
1960 	memset(&key->ct, 0, sizeof(key->ct));
1961 	memset(&key->ipv4.ct_orig, 0, sizeof(key->ipv4.ct_orig));
1962 	memset(&key->ipv6.ct_orig, 0, sizeof(key->ipv6.ct_orig));
1963 
1964 	key->phy.in_port = DP_MAX_PORTS;
1965 
1966 	return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1967 }
1968 
1969 static int ovs_nla_put_vlan(struct sk_buff *skb, const struct vlan_head *vh,
1970 			    bool is_mask)
1971 {
1972 	__be16 eth_type = !is_mask ? vh->tpid : htons(0xffff);
1973 
1974 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1975 	    nla_put_be16(skb, OVS_KEY_ATTR_VLAN, vh->tci))
1976 		return -EMSGSIZE;
1977 	return 0;
1978 }
1979 
1980 static int nsh_key_to_nlattr(const struct ovs_key_nsh *nsh, bool is_mask,
1981 			     struct sk_buff *skb)
1982 {
1983 	struct nlattr *start;
1984 
1985 	start = nla_nest_start_noflag(skb, OVS_KEY_ATTR_NSH);
1986 	if (!start)
1987 		return -EMSGSIZE;
1988 
1989 	if (nla_put(skb, OVS_NSH_KEY_ATTR_BASE, sizeof(nsh->base), &nsh->base))
1990 		goto nla_put_failure;
1991 
1992 	if (is_mask || nsh->base.mdtype == NSH_M_TYPE1) {
1993 		if (nla_put(skb, OVS_NSH_KEY_ATTR_MD1,
1994 			    sizeof(nsh->context), nsh->context))
1995 			goto nla_put_failure;
1996 	}
1997 
1998 	/* Don't support MD type 2 yet */
1999 
2000 	nla_nest_end(skb, start);
2001 
2002 	return 0;
2003 
2004 nla_put_failure:
2005 	return -EMSGSIZE;
2006 }
2007 
2008 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
2009 			     const struct sw_flow_key *output, bool is_mask,
2010 			     struct sk_buff *skb)
2011 {
2012 	struct ovs_key_ethernet *eth_key;
2013 	struct nlattr *nla;
2014 	struct nlattr *encap = NULL;
2015 	struct nlattr *in_encap = NULL;
2016 
2017 	if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
2018 		goto nla_put_failure;
2019 
2020 	if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
2021 		goto nla_put_failure;
2022 
2023 	if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
2024 		goto nla_put_failure;
2025 
2026 	if ((swkey->tun_proto || is_mask)) {
2027 		const void *opts = NULL;
2028 
2029 		if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
2030 			opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
2031 
2032 		if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
2033 				     swkey->tun_opts_len, swkey->tun_proto, 0))
2034 			goto nla_put_failure;
2035 	}
2036 
2037 	if (swkey->phy.in_port == DP_MAX_PORTS) {
2038 		if (is_mask && (output->phy.in_port == 0xffff))
2039 			if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
2040 				goto nla_put_failure;
2041 	} else {
2042 		u16 upper_u16;
2043 		upper_u16 = !is_mask ? 0 : 0xffff;
2044 
2045 		if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
2046 				(upper_u16 << 16) | output->phy.in_port))
2047 			goto nla_put_failure;
2048 	}
2049 
2050 	if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
2051 		goto nla_put_failure;
2052 
2053 	if (ovs_ct_put_key(swkey, output, skb))
2054 		goto nla_put_failure;
2055 
2056 	if (ovs_key_mac_proto(swkey) == MAC_PROTO_ETHERNET) {
2057 		nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
2058 		if (!nla)
2059 			goto nla_put_failure;
2060 
2061 		eth_key = nla_data(nla);
2062 		ether_addr_copy(eth_key->eth_src, output->eth.src);
2063 		ether_addr_copy(eth_key->eth_dst, output->eth.dst);
2064 
2065 		if (swkey->eth.vlan.tci || eth_type_vlan(swkey->eth.type)) {
2066 			if (ovs_nla_put_vlan(skb, &output->eth.vlan, is_mask))
2067 				goto nla_put_failure;
2068 			encap = nla_nest_start_noflag(skb, OVS_KEY_ATTR_ENCAP);
2069 			if (!swkey->eth.vlan.tci)
2070 				goto unencap;
2071 
2072 			if (swkey->eth.cvlan.tci || eth_type_vlan(swkey->eth.type)) {
2073 				if (ovs_nla_put_vlan(skb, &output->eth.cvlan, is_mask))
2074 					goto nla_put_failure;
2075 				in_encap = nla_nest_start_noflag(skb,
2076 								 OVS_KEY_ATTR_ENCAP);
2077 				if (!swkey->eth.cvlan.tci)
2078 					goto unencap;
2079 			}
2080 		}
2081 
2082 		if (swkey->eth.type == htons(ETH_P_802_2)) {
2083 			/*
2084 			* Ethertype 802.2 is represented in the netlink with omitted
2085 			* OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
2086 			* 0xffff in the mask attribute.  Ethertype can also
2087 			* be wildcarded.
2088 			*/
2089 			if (is_mask && output->eth.type)
2090 				if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
2091 							output->eth.type))
2092 					goto nla_put_failure;
2093 			goto unencap;
2094 		}
2095 	}
2096 
2097 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
2098 		goto nla_put_failure;
2099 
2100 	if (eth_type_vlan(swkey->eth.type)) {
2101 		/* There are 3 VLAN tags, we don't know anything about the rest
2102 		 * of the packet, so truncate here.
2103 		 */
2104 		WARN_ON_ONCE(!(encap && in_encap));
2105 		goto unencap;
2106 	}
2107 
2108 	if (swkey->eth.type == htons(ETH_P_IP)) {
2109 		struct ovs_key_ipv4 *ipv4_key;
2110 
2111 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
2112 		if (!nla)
2113 			goto nla_put_failure;
2114 		ipv4_key = nla_data(nla);
2115 		ipv4_key->ipv4_src = output->ipv4.addr.src;
2116 		ipv4_key->ipv4_dst = output->ipv4.addr.dst;
2117 		ipv4_key->ipv4_proto = output->ip.proto;
2118 		ipv4_key->ipv4_tos = output->ip.tos;
2119 		ipv4_key->ipv4_ttl = output->ip.ttl;
2120 		ipv4_key->ipv4_frag = output->ip.frag;
2121 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
2122 		struct ovs_key_ipv6 *ipv6_key;
2123 		struct ovs_key_ipv6_exthdrs *ipv6_exthdrs_key;
2124 
2125 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
2126 		if (!nla)
2127 			goto nla_put_failure;
2128 		ipv6_key = nla_data(nla);
2129 		memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
2130 				sizeof(ipv6_key->ipv6_src));
2131 		memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
2132 				sizeof(ipv6_key->ipv6_dst));
2133 		ipv6_key->ipv6_label = output->ipv6.label;
2134 		ipv6_key->ipv6_proto = output->ip.proto;
2135 		ipv6_key->ipv6_tclass = output->ip.tos;
2136 		ipv6_key->ipv6_hlimit = output->ip.ttl;
2137 		ipv6_key->ipv6_frag = output->ip.frag;
2138 
2139 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6_EXTHDRS,
2140 				  sizeof(*ipv6_exthdrs_key));
2141 		if (!nla)
2142 			goto nla_put_failure;
2143 		ipv6_exthdrs_key = nla_data(nla);
2144 		ipv6_exthdrs_key->hdrs = output->ipv6.exthdrs;
2145 	} else if (swkey->eth.type == htons(ETH_P_NSH)) {
2146 		if (nsh_key_to_nlattr(&output->nsh, is_mask, skb))
2147 			goto nla_put_failure;
2148 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
2149 		   swkey->eth.type == htons(ETH_P_RARP)) {
2150 		struct ovs_key_arp *arp_key;
2151 
2152 		nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
2153 		if (!nla)
2154 			goto nla_put_failure;
2155 		arp_key = nla_data(nla);
2156 		memset(arp_key, 0, sizeof(struct ovs_key_arp));
2157 		arp_key->arp_sip = output->ipv4.addr.src;
2158 		arp_key->arp_tip = output->ipv4.addr.dst;
2159 		arp_key->arp_op = htons(output->ip.proto);
2160 		ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
2161 		ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
2162 	} else if (eth_p_mpls(swkey->eth.type)) {
2163 		u8 i, num_labels;
2164 		struct ovs_key_mpls *mpls_key;
2165 
2166 		num_labels = hweight_long(output->mpls.num_labels_mask);
2167 		nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS,
2168 				  num_labels * sizeof(*mpls_key));
2169 		if (!nla)
2170 			goto nla_put_failure;
2171 
2172 		mpls_key = nla_data(nla);
2173 		for (i = 0; i < num_labels; i++)
2174 			mpls_key[i].mpls_lse = output->mpls.lse[i];
2175 	}
2176 
2177 	if ((swkey->eth.type == htons(ETH_P_IP) ||
2178 	     swkey->eth.type == htons(ETH_P_IPV6)) &&
2179 	     swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
2180 
2181 		if (swkey->ip.proto == IPPROTO_TCP) {
2182 			struct ovs_key_tcp *tcp_key;
2183 
2184 			nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
2185 			if (!nla)
2186 				goto nla_put_failure;
2187 			tcp_key = nla_data(nla);
2188 			tcp_key->tcp_src = output->tp.src;
2189 			tcp_key->tcp_dst = output->tp.dst;
2190 			if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
2191 					 output->tp.flags))
2192 				goto nla_put_failure;
2193 		} else if (swkey->ip.proto == IPPROTO_UDP) {
2194 			struct ovs_key_udp *udp_key;
2195 
2196 			nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
2197 			if (!nla)
2198 				goto nla_put_failure;
2199 			udp_key = nla_data(nla);
2200 			udp_key->udp_src = output->tp.src;
2201 			udp_key->udp_dst = output->tp.dst;
2202 		} else if (swkey->ip.proto == IPPROTO_SCTP) {
2203 			struct ovs_key_sctp *sctp_key;
2204 
2205 			nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
2206 			if (!nla)
2207 				goto nla_put_failure;
2208 			sctp_key = nla_data(nla);
2209 			sctp_key->sctp_src = output->tp.src;
2210 			sctp_key->sctp_dst = output->tp.dst;
2211 		} else if (swkey->eth.type == htons(ETH_P_IP) &&
2212 			   swkey->ip.proto == IPPROTO_ICMP) {
2213 			struct ovs_key_icmp *icmp_key;
2214 
2215 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
2216 			if (!nla)
2217 				goto nla_put_failure;
2218 			icmp_key = nla_data(nla);
2219 			icmp_key->icmp_type = ntohs(output->tp.src);
2220 			icmp_key->icmp_code = ntohs(output->tp.dst);
2221 		} else if (swkey->eth.type == htons(ETH_P_IPV6) &&
2222 			   swkey->ip.proto == IPPROTO_ICMPV6) {
2223 			struct ovs_key_icmpv6 *icmpv6_key;
2224 
2225 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
2226 						sizeof(*icmpv6_key));
2227 			if (!nla)
2228 				goto nla_put_failure;
2229 			icmpv6_key = nla_data(nla);
2230 			icmpv6_key->icmpv6_type = ntohs(output->tp.src);
2231 			icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
2232 
2233 			if (swkey->tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
2234 			    swkey->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
2235 				struct ovs_key_nd *nd_key;
2236 
2237 				nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
2238 				if (!nla)
2239 					goto nla_put_failure;
2240 				nd_key = nla_data(nla);
2241 				memcpy(nd_key->nd_target, &output->ipv6.nd.target,
2242 							sizeof(nd_key->nd_target));
2243 				ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
2244 				ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
2245 			}
2246 		}
2247 	}
2248 
2249 unencap:
2250 	if (in_encap)
2251 		nla_nest_end(skb, in_encap);
2252 	if (encap)
2253 		nla_nest_end(skb, encap);
2254 
2255 	return 0;
2256 
2257 nla_put_failure:
2258 	return -EMSGSIZE;
2259 }
2260 
2261 int ovs_nla_put_key(const struct sw_flow_key *swkey,
2262 		    const struct sw_flow_key *output, int attr, bool is_mask,
2263 		    struct sk_buff *skb)
2264 {
2265 	int err;
2266 	struct nlattr *nla;
2267 
2268 	nla = nla_nest_start_noflag(skb, attr);
2269 	if (!nla)
2270 		return -EMSGSIZE;
2271 	err = __ovs_nla_put_key(swkey, output, is_mask, skb);
2272 	if (err)
2273 		return err;
2274 	nla_nest_end(skb, nla);
2275 
2276 	return 0;
2277 }
2278 
2279 /* Called with ovs_mutex or RCU read lock. */
2280 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
2281 {
2282 	if (ovs_identifier_is_ufid(&flow->id))
2283 		return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
2284 			       flow->id.ufid);
2285 
2286 	return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
2287 			       OVS_FLOW_ATTR_KEY, false, skb);
2288 }
2289 
2290 /* Called with ovs_mutex or RCU read lock. */
2291 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
2292 {
2293 	return ovs_nla_put_key(&flow->key, &flow->key,
2294 				OVS_FLOW_ATTR_KEY, false, skb);
2295 }
2296 
2297 /* Called with ovs_mutex or RCU read lock. */
2298 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
2299 {
2300 	return ovs_nla_put_key(&flow->key, &flow->mask->key,
2301 				OVS_FLOW_ATTR_MASK, true, skb);
2302 }
2303 
2304 #define MAX_ACTIONS_BUFSIZE	(32 * 1024)
2305 
2306 static struct sw_flow_actions *nla_alloc_flow_actions(int size)
2307 {
2308 	struct sw_flow_actions *sfa;
2309 
2310 	WARN_ON_ONCE(size > MAX_ACTIONS_BUFSIZE);
2311 
2312 	sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
2313 	if (!sfa)
2314 		return ERR_PTR(-ENOMEM);
2315 
2316 	sfa->actions_len = 0;
2317 	return sfa;
2318 }
2319 
2320 static void ovs_nla_free_nested_actions(const struct nlattr *actions, int len);
2321 
2322 static void ovs_nla_free_check_pkt_len_action(const struct nlattr *action)
2323 {
2324 	const struct nlattr *a;
2325 	int rem;
2326 
2327 	nla_for_each_nested(a, action, rem) {
2328 		switch (nla_type(a)) {
2329 		case OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL:
2330 		case OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER:
2331 			ovs_nla_free_nested_actions(nla_data(a), nla_len(a));
2332 			break;
2333 		}
2334 	}
2335 }
2336 
2337 static void ovs_nla_free_clone_action(const struct nlattr *action)
2338 {
2339 	const struct nlattr *a = nla_data(action);
2340 	int rem = nla_len(action);
2341 
2342 	switch (nla_type(a)) {
2343 	case OVS_CLONE_ATTR_EXEC:
2344 		/* The real list of actions follows this attribute. */
2345 		a = nla_next(a, &rem);
2346 		ovs_nla_free_nested_actions(a, rem);
2347 		break;
2348 	}
2349 }
2350 
2351 static void ovs_nla_free_dec_ttl_action(const struct nlattr *action)
2352 {
2353 	const struct nlattr *a = nla_data(action);
2354 
2355 	switch (nla_type(a)) {
2356 	case OVS_DEC_TTL_ATTR_ACTION:
2357 		ovs_nla_free_nested_actions(nla_data(a), nla_len(a));
2358 		break;
2359 	}
2360 }
2361 
2362 static void ovs_nla_free_sample_action(const struct nlattr *action)
2363 {
2364 	const struct nlattr *a = nla_data(action);
2365 	int rem = nla_len(action);
2366 
2367 	switch (nla_type(a)) {
2368 	case OVS_SAMPLE_ATTR_ARG:
2369 		/* The real list of actions follows this attribute. */
2370 		a = nla_next(a, &rem);
2371 		ovs_nla_free_nested_actions(a, rem);
2372 		break;
2373 	}
2374 }
2375 
2376 static void ovs_nla_free_set_action(const struct nlattr *a)
2377 {
2378 	const struct nlattr *ovs_key = nla_data(a);
2379 	struct ovs_tunnel_info *ovs_tun;
2380 
2381 	switch (nla_type(ovs_key)) {
2382 	case OVS_KEY_ATTR_TUNNEL_INFO:
2383 		ovs_tun = nla_data(ovs_key);
2384 		dst_release((struct dst_entry *)ovs_tun->tun_dst);
2385 		break;
2386 	}
2387 }
2388 
2389 static void ovs_nla_free_nested_actions(const struct nlattr *actions, int len)
2390 {
2391 	const struct nlattr *a;
2392 	int rem;
2393 
2394 	/* Whenever new actions are added, the need to update this
2395 	 * function should be considered.
2396 	 */
2397 	BUILD_BUG_ON(OVS_ACTION_ATTR_MAX != 23);
2398 
2399 	if (!actions)
2400 		return;
2401 
2402 	nla_for_each_attr(a, actions, len, rem) {
2403 		switch (nla_type(a)) {
2404 		case OVS_ACTION_ATTR_CHECK_PKT_LEN:
2405 			ovs_nla_free_check_pkt_len_action(a);
2406 			break;
2407 
2408 		case OVS_ACTION_ATTR_CLONE:
2409 			ovs_nla_free_clone_action(a);
2410 			break;
2411 
2412 		case OVS_ACTION_ATTR_CT:
2413 			ovs_ct_free_action(a);
2414 			break;
2415 
2416 		case OVS_ACTION_ATTR_DEC_TTL:
2417 			ovs_nla_free_dec_ttl_action(a);
2418 			break;
2419 
2420 		case OVS_ACTION_ATTR_SAMPLE:
2421 			ovs_nla_free_sample_action(a);
2422 			break;
2423 
2424 		case OVS_ACTION_ATTR_SET:
2425 			ovs_nla_free_set_action(a);
2426 			break;
2427 		}
2428 	}
2429 }
2430 
2431 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
2432 {
2433 	if (!sf_acts)
2434 		return;
2435 
2436 	ovs_nla_free_nested_actions(sf_acts->actions, sf_acts->actions_len);
2437 	kfree(sf_acts);
2438 }
2439 
2440 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
2441 {
2442 	ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
2443 }
2444 
2445 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
2446  * The caller must hold rcu_read_lock for this to be sensible. */
2447 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
2448 {
2449 	call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
2450 }
2451 
2452 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
2453 				       int attr_len, bool log)
2454 {
2455 
2456 	struct sw_flow_actions *acts;
2457 	int new_acts_size;
2458 	size_t req_size = NLA_ALIGN(attr_len);
2459 	int next_offset = offsetof(struct sw_flow_actions, actions) +
2460 					(*sfa)->actions_len;
2461 
2462 	if (req_size <= (ksize(*sfa) - next_offset))
2463 		goto out;
2464 
2465 	new_acts_size = max(next_offset + req_size, ksize(*sfa) * 2);
2466 
2467 	if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
2468 		if ((next_offset + req_size) > MAX_ACTIONS_BUFSIZE) {
2469 			OVS_NLERR(log, "Flow action size exceeds max %u",
2470 				  MAX_ACTIONS_BUFSIZE);
2471 			return ERR_PTR(-EMSGSIZE);
2472 		}
2473 		new_acts_size = MAX_ACTIONS_BUFSIZE;
2474 	}
2475 
2476 	acts = nla_alloc_flow_actions(new_acts_size);
2477 	if (IS_ERR(acts))
2478 		return (void *)acts;
2479 
2480 	memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
2481 	acts->actions_len = (*sfa)->actions_len;
2482 	acts->orig_len = (*sfa)->orig_len;
2483 	kfree(*sfa);
2484 	*sfa = acts;
2485 
2486 out:
2487 	(*sfa)->actions_len += req_size;
2488 	return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
2489 }
2490 
2491 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
2492 				   int attrtype, void *data, int len, bool log)
2493 {
2494 	struct nlattr *a;
2495 
2496 	a = reserve_sfa_size(sfa, nla_attr_size(len), log);
2497 	if (IS_ERR(a))
2498 		return a;
2499 
2500 	a->nla_type = attrtype;
2501 	a->nla_len = nla_attr_size(len);
2502 
2503 	if (data)
2504 		memcpy(nla_data(a), data, len);
2505 	memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
2506 
2507 	return a;
2508 }
2509 
2510 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
2511 		       int len, bool log)
2512 {
2513 	struct nlattr *a;
2514 
2515 	a = __add_action(sfa, attrtype, data, len, log);
2516 
2517 	return PTR_ERR_OR_ZERO(a);
2518 }
2519 
2520 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
2521 					  int attrtype, bool log)
2522 {
2523 	int used = (*sfa)->actions_len;
2524 	int err;
2525 
2526 	err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
2527 	if (err)
2528 		return err;
2529 
2530 	return used;
2531 }
2532 
2533 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
2534 					 int st_offset)
2535 {
2536 	struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
2537 							       st_offset);
2538 
2539 	a->nla_len = sfa->actions_len - st_offset;
2540 }
2541 
2542 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2543 				  const struct sw_flow_key *key,
2544 				  struct sw_flow_actions **sfa,
2545 				  __be16 eth_type, __be16 vlan_tci,
2546 				  u32 mpls_label_count, bool log);
2547 
2548 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
2549 				    const struct sw_flow_key *key,
2550 				    struct sw_flow_actions **sfa,
2551 				    __be16 eth_type, __be16 vlan_tci,
2552 				    u32 mpls_label_count, bool log, bool last)
2553 {
2554 	const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
2555 	const struct nlattr *probability, *actions;
2556 	const struct nlattr *a;
2557 	int rem, start, err;
2558 	struct sample_arg arg;
2559 
2560 	memset(attrs, 0, sizeof(attrs));
2561 	nla_for_each_nested(a, attr, rem) {
2562 		int type = nla_type(a);
2563 		if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
2564 			return -EINVAL;
2565 		attrs[type] = a;
2566 	}
2567 	if (rem)
2568 		return -EINVAL;
2569 
2570 	probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
2571 	if (!probability || nla_len(probability) != sizeof(u32))
2572 		return -EINVAL;
2573 
2574 	actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
2575 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
2576 		return -EINVAL;
2577 
2578 	/* validation done, copy sample action. */
2579 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
2580 	if (start < 0)
2581 		return start;
2582 
2583 	/* When both skb and flow may be changed, put the sample
2584 	 * into a deferred fifo. On the other hand, if only skb
2585 	 * may be modified, the actions can be executed in place.
2586 	 *
2587 	 * Do this analysis at the flow installation time.
2588 	 * Set 'clone_action->exec' to true if the actions can be
2589 	 * executed without being deferred.
2590 	 *
2591 	 * If the sample is the last action, it can always be excuted
2592 	 * rather than deferred.
2593 	 */
2594 	arg.exec = last || !actions_may_change_flow(actions);
2595 	arg.probability = nla_get_u32(probability);
2596 
2597 	err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_ARG, &arg, sizeof(arg),
2598 				 log);
2599 	if (err)
2600 		return err;
2601 
2602 	err = __ovs_nla_copy_actions(net, actions, key, sfa,
2603 				     eth_type, vlan_tci, mpls_label_count, log);
2604 
2605 	if (err)
2606 		return err;
2607 
2608 	add_nested_action_end(*sfa, start);
2609 
2610 	return 0;
2611 }
2612 
2613 static int validate_and_copy_dec_ttl(struct net *net,
2614 				     const struct nlattr *attr,
2615 				     const struct sw_flow_key *key,
2616 				     struct sw_flow_actions **sfa,
2617 				     __be16 eth_type, __be16 vlan_tci,
2618 				     u32 mpls_label_count, bool log)
2619 {
2620 	const struct nlattr *attrs[OVS_DEC_TTL_ATTR_MAX + 1];
2621 	int start, action_start, err, rem;
2622 	const struct nlattr *a, *actions;
2623 
2624 	memset(attrs, 0, sizeof(attrs));
2625 	nla_for_each_nested(a, attr, rem) {
2626 		int type = nla_type(a);
2627 
2628 		/* Ignore unknown attributes to be future proof. */
2629 		if (type > OVS_DEC_TTL_ATTR_MAX)
2630 			continue;
2631 
2632 		if (!type || attrs[type]) {
2633 			OVS_NLERR(log, "Duplicate or invalid key (type %d).",
2634 				  type);
2635 			return -EINVAL;
2636 		}
2637 
2638 		attrs[type] = a;
2639 	}
2640 
2641 	if (rem) {
2642 		OVS_NLERR(log, "Message has %d unknown bytes.", rem);
2643 		return -EINVAL;
2644 	}
2645 
2646 	actions = attrs[OVS_DEC_TTL_ATTR_ACTION];
2647 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN)) {
2648 		OVS_NLERR(log, "Missing valid actions attribute.");
2649 		return -EINVAL;
2650 	}
2651 
2652 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_DEC_TTL, log);
2653 	if (start < 0)
2654 		return start;
2655 
2656 	action_start = add_nested_action_start(sfa, OVS_DEC_TTL_ATTR_ACTION, log);
2657 	if (action_start < 0)
2658 		return action_start;
2659 
2660 	err = __ovs_nla_copy_actions(net, actions, key, sfa, eth_type,
2661 				     vlan_tci, mpls_label_count, log);
2662 	if (err)
2663 		return err;
2664 
2665 	add_nested_action_end(*sfa, action_start);
2666 	add_nested_action_end(*sfa, start);
2667 	return 0;
2668 }
2669 
2670 static int validate_and_copy_clone(struct net *net,
2671 				   const struct nlattr *attr,
2672 				   const struct sw_flow_key *key,
2673 				   struct sw_flow_actions **sfa,
2674 				   __be16 eth_type, __be16 vlan_tci,
2675 				   u32 mpls_label_count, bool log, bool last)
2676 {
2677 	int start, err;
2678 	u32 exec;
2679 
2680 	if (nla_len(attr) && nla_len(attr) < NLA_HDRLEN)
2681 		return -EINVAL;
2682 
2683 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_CLONE, log);
2684 	if (start < 0)
2685 		return start;
2686 
2687 	exec = last || !actions_may_change_flow(attr);
2688 
2689 	err = ovs_nla_add_action(sfa, OVS_CLONE_ATTR_EXEC, &exec,
2690 				 sizeof(exec), log);
2691 	if (err)
2692 		return err;
2693 
2694 	err = __ovs_nla_copy_actions(net, attr, key, sfa,
2695 				     eth_type, vlan_tci, mpls_label_count, log);
2696 	if (err)
2697 		return err;
2698 
2699 	add_nested_action_end(*sfa, start);
2700 
2701 	return 0;
2702 }
2703 
2704 void ovs_match_init(struct sw_flow_match *match,
2705 		    struct sw_flow_key *key,
2706 		    bool reset_key,
2707 		    struct sw_flow_mask *mask)
2708 {
2709 	memset(match, 0, sizeof(*match));
2710 	match->key = key;
2711 	match->mask = mask;
2712 
2713 	if (reset_key)
2714 		memset(key, 0, sizeof(*key));
2715 
2716 	if (mask) {
2717 		memset(&mask->key, 0, sizeof(mask->key));
2718 		mask->range.start = mask->range.end = 0;
2719 	}
2720 }
2721 
2722 static int validate_geneve_opts(struct sw_flow_key *key)
2723 {
2724 	struct geneve_opt *option;
2725 	int opts_len = key->tun_opts_len;
2726 	bool crit_opt = false;
2727 
2728 	option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
2729 	while (opts_len > 0) {
2730 		int len;
2731 
2732 		if (opts_len < sizeof(*option))
2733 			return -EINVAL;
2734 
2735 		len = sizeof(*option) + option->length * 4;
2736 		if (len > opts_len)
2737 			return -EINVAL;
2738 
2739 		crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
2740 
2741 		option = (struct geneve_opt *)((u8 *)option + len);
2742 		opts_len -= len;
2743 	}
2744 
2745 	key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
2746 
2747 	return 0;
2748 }
2749 
2750 static int validate_and_copy_set_tun(const struct nlattr *attr,
2751 				     struct sw_flow_actions **sfa, bool log)
2752 {
2753 	struct sw_flow_match match;
2754 	struct sw_flow_key key;
2755 	struct metadata_dst *tun_dst;
2756 	struct ip_tunnel_info *tun_info;
2757 	struct ovs_tunnel_info *ovs_tun;
2758 	struct nlattr *a;
2759 	int err = 0, start, opts_type;
2760 	__be16 dst_opt_type;
2761 
2762 	dst_opt_type = 0;
2763 	ovs_match_init(&match, &key, true, NULL);
2764 	opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
2765 	if (opts_type < 0)
2766 		return opts_type;
2767 
2768 	if (key.tun_opts_len) {
2769 		switch (opts_type) {
2770 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
2771 			err = validate_geneve_opts(&key);
2772 			if (err < 0)
2773 				return err;
2774 			dst_opt_type = TUNNEL_GENEVE_OPT;
2775 			break;
2776 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
2777 			dst_opt_type = TUNNEL_VXLAN_OPT;
2778 			break;
2779 		case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
2780 			dst_opt_type = TUNNEL_ERSPAN_OPT;
2781 			break;
2782 		}
2783 	}
2784 
2785 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
2786 	if (start < 0)
2787 		return start;
2788 
2789 	tun_dst = metadata_dst_alloc(key.tun_opts_len, METADATA_IP_TUNNEL,
2790 				     GFP_KERNEL);
2791 
2792 	if (!tun_dst)
2793 		return -ENOMEM;
2794 
2795 	err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
2796 	if (err) {
2797 		dst_release((struct dst_entry *)tun_dst);
2798 		return err;
2799 	}
2800 
2801 	a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
2802 			 sizeof(*ovs_tun), log);
2803 	if (IS_ERR(a)) {
2804 		dst_release((struct dst_entry *)tun_dst);
2805 		return PTR_ERR(a);
2806 	}
2807 
2808 	ovs_tun = nla_data(a);
2809 	ovs_tun->tun_dst = tun_dst;
2810 
2811 	tun_info = &tun_dst->u.tun_info;
2812 	tun_info->mode = IP_TUNNEL_INFO_TX;
2813 	if (key.tun_proto == AF_INET6)
2814 		tun_info->mode |= IP_TUNNEL_INFO_IPV6;
2815 	else if (key.tun_proto == AF_INET && key.tun_key.u.ipv4.dst == 0)
2816 		tun_info->mode |= IP_TUNNEL_INFO_BRIDGE;
2817 	tun_info->key = key.tun_key;
2818 
2819 	/* We need to store the options in the action itself since
2820 	 * everything else will go away after flow setup. We can append
2821 	 * it to tun_info and then point there.
2822 	 */
2823 	ip_tunnel_info_opts_set(tun_info,
2824 				TUN_METADATA_OPTS(&key, key.tun_opts_len),
2825 				key.tun_opts_len, dst_opt_type);
2826 	add_nested_action_end(*sfa, start);
2827 
2828 	return err;
2829 }
2830 
2831 static bool validate_nsh(const struct nlattr *attr, bool is_mask,
2832 			 bool is_push_nsh, bool log)
2833 {
2834 	struct sw_flow_match match;
2835 	struct sw_flow_key key;
2836 	int ret = 0;
2837 
2838 	ovs_match_init(&match, &key, true, NULL);
2839 	ret = nsh_key_put_from_nlattr(attr, &match, is_mask,
2840 				      is_push_nsh, log);
2841 	return !ret;
2842 }
2843 
2844 /* Return false if there are any non-masked bits set.
2845  * Mask follows data immediately, before any netlink padding.
2846  */
2847 static bool validate_masked(u8 *data, int len)
2848 {
2849 	u8 *mask = data + len;
2850 
2851 	while (len--)
2852 		if (*data++ & ~*mask++)
2853 			return false;
2854 
2855 	return true;
2856 }
2857 
2858 static int validate_set(const struct nlattr *a,
2859 			const struct sw_flow_key *flow_key,
2860 			struct sw_flow_actions **sfa, bool *skip_copy,
2861 			u8 mac_proto, __be16 eth_type, bool masked, bool log)
2862 {
2863 	const struct nlattr *ovs_key = nla_data(a);
2864 	int key_type = nla_type(ovs_key);
2865 	size_t key_len;
2866 
2867 	/* There can be only one key in a action */
2868 	if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2869 		return -EINVAL;
2870 
2871 	key_len = nla_len(ovs_key);
2872 	if (masked)
2873 		key_len /= 2;
2874 
2875 	if (key_type > OVS_KEY_ATTR_MAX ||
2876 	    !check_attr_len(key_len, ovs_key_lens[key_type].len))
2877 		return -EINVAL;
2878 
2879 	if (masked && !validate_masked(nla_data(ovs_key), key_len))
2880 		return -EINVAL;
2881 
2882 	switch (key_type) {
2883 	case OVS_KEY_ATTR_PRIORITY:
2884 	case OVS_KEY_ATTR_SKB_MARK:
2885 	case OVS_KEY_ATTR_CT_MARK:
2886 	case OVS_KEY_ATTR_CT_LABELS:
2887 		break;
2888 
2889 	case OVS_KEY_ATTR_ETHERNET:
2890 		if (mac_proto != MAC_PROTO_ETHERNET)
2891 			return -EINVAL;
2892 		break;
2893 
2894 	case OVS_KEY_ATTR_TUNNEL: {
2895 		int err;
2896 
2897 		if (masked)
2898 			return -EINVAL; /* Masked tunnel set not supported. */
2899 
2900 		*skip_copy = true;
2901 		err = validate_and_copy_set_tun(a, sfa, log);
2902 		if (err)
2903 			return err;
2904 		break;
2905 	}
2906 	case OVS_KEY_ATTR_IPV4: {
2907 		const struct ovs_key_ipv4 *ipv4_key;
2908 
2909 		if (eth_type != htons(ETH_P_IP))
2910 			return -EINVAL;
2911 
2912 		ipv4_key = nla_data(ovs_key);
2913 
2914 		if (masked) {
2915 			const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2916 
2917 			/* Non-writeable fields. */
2918 			if (mask->ipv4_proto || mask->ipv4_frag)
2919 				return -EINVAL;
2920 		} else {
2921 			if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2922 				return -EINVAL;
2923 
2924 			if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2925 				return -EINVAL;
2926 		}
2927 		break;
2928 	}
2929 	case OVS_KEY_ATTR_IPV6: {
2930 		const struct ovs_key_ipv6 *ipv6_key;
2931 
2932 		if (eth_type != htons(ETH_P_IPV6))
2933 			return -EINVAL;
2934 
2935 		ipv6_key = nla_data(ovs_key);
2936 
2937 		if (masked) {
2938 			const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2939 
2940 			/* Non-writeable fields. */
2941 			if (mask->ipv6_proto || mask->ipv6_frag)
2942 				return -EINVAL;
2943 
2944 			/* Invalid bits in the flow label mask? */
2945 			if (ntohl(mask->ipv6_label) & 0xFFF00000)
2946 				return -EINVAL;
2947 		} else {
2948 			if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2949 				return -EINVAL;
2950 
2951 			if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2952 				return -EINVAL;
2953 		}
2954 		if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2955 			return -EINVAL;
2956 
2957 		break;
2958 	}
2959 	case OVS_KEY_ATTR_TCP:
2960 		if ((eth_type != htons(ETH_P_IP) &&
2961 		     eth_type != htons(ETH_P_IPV6)) ||
2962 		    flow_key->ip.proto != IPPROTO_TCP)
2963 			return -EINVAL;
2964 
2965 		break;
2966 
2967 	case OVS_KEY_ATTR_UDP:
2968 		if ((eth_type != htons(ETH_P_IP) &&
2969 		     eth_type != htons(ETH_P_IPV6)) ||
2970 		    flow_key->ip.proto != IPPROTO_UDP)
2971 			return -EINVAL;
2972 
2973 		break;
2974 
2975 	case OVS_KEY_ATTR_MPLS:
2976 		if (!eth_p_mpls(eth_type))
2977 			return -EINVAL;
2978 		break;
2979 
2980 	case OVS_KEY_ATTR_SCTP:
2981 		if ((eth_type != htons(ETH_P_IP) &&
2982 		     eth_type != htons(ETH_P_IPV6)) ||
2983 		    flow_key->ip.proto != IPPROTO_SCTP)
2984 			return -EINVAL;
2985 
2986 		break;
2987 
2988 	case OVS_KEY_ATTR_NSH:
2989 		if (eth_type != htons(ETH_P_NSH))
2990 			return -EINVAL;
2991 		if (!validate_nsh(nla_data(a), masked, false, log))
2992 			return -EINVAL;
2993 		break;
2994 
2995 	default:
2996 		return -EINVAL;
2997 	}
2998 
2999 	/* Convert non-masked non-tunnel set actions to masked set actions. */
3000 	if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
3001 		int start, len = key_len * 2;
3002 		struct nlattr *at;
3003 
3004 		*skip_copy = true;
3005 
3006 		start = add_nested_action_start(sfa,
3007 						OVS_ACTION_ATTR_SET_TO_MASKED,
3008 						log);
3009 		if (start < 0)
3010 			return start;
3011 
3012 		at = __add_action(sfa, key_type, NULL, len, log);
3013 		if (IS_ERR(at))
3014 			return PTR_ERR(at);
3015 
3016 		memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
3017 		memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
3018 		/* Clear non-writeable bits from otherwise writeable fields. */
3019 		if (key_type == OVS_KEY_ATTR_IPV6) {
3020 			struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
3021 
3022 			mask->ipv6_label &= htonl(0x000FFFFF);
3023 		}
3024 		add_nested_action_end(*sfa, start);
3025 	}
3026 
3027 	return 0;
3028 }
3029 
3030 static int validate_userspace(const struct nlattr *attr)
3031 {
3032 	static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
3033 		[OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
3034 		[OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
3035 		[OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
3036 	};
3037 	struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
3038 	int error;
3039 
3040 	error = nla_parse_nested_deprecated(a, OVS_USERSPACE_ATTR_MAX, attr,
3041 					    userspace_policy, NULL);
3042 	if (error)
3043 		return error;
3044 
3045 	if (!a[OVS_USERSPACE_ATTR_PID] ||
3046 	    !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
3047 		return -EINVAL;
3048 
3049 	return 0;
3050 }
3051 
3052 static const struct nla_policy cpl_policy[OVS_CHECK_PKT_LEN_ATTR_MAX + 1] = {
3053 	[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN] = {.type = NLA_U16 },
3054 	[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER] = {.type = NLA_NESTED },
3055 	[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL] = {.type = NLA_NESTED },
3056 };
3057 
3058 static int validate_and_copy_check_pkt_len(struct net *net,
3059 					   const struct nlattr *attr,
3060 					   const struct sw_flow_key *key,
3061 					   struct sw_flow_actions **sfa,
3062 					   __be16 eth_type, __be16 vlan_tci,
3063 					   u32 mpls_label_count,
3064 					   bool log, bool last)
3065 {
3066 	const struct nlattr *acts_if_greater, *acts_if_lesser_eq;
3067 	struct nlattr *a[OVS_CHECK_PKT_LEN_ATTR_MAX + 1];
3068 	struct check_pkt_len_arg arg;
3069 	int nested_acts_start;
3070 	int start, err;
3071 
3072 	err = nla_parse_deprecated_strict(a, OVS_CHECK_PKT_LEN_ATTR_MAX,
3073 					  nla_data(attr), nla_len(attr),
3074 					  cpl_policy, NULL);
3075 	if (err)
3076 		return err;
3077 
3078 	if (!a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN] ||
3079 	    !nla_get_u16(a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN]))
3080 		return -EINVAL;
3081 
3082 	acts_if_lesser_eq = a[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL];
3083 	acts_if_greater = a[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER];
3084 
3085 	/* Both the nested action should be present. */
3086 	if (!acts_if_greater || !acts_if_lesser_eq)
3087 		return -EINVAL;
3088 
3089 	/* validation done, copy the nested actions. */
3090 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_CHECK_PKT_LEN,
3091 					log);
3092 	if (start < 0)
3093 		return start;
3094 
3095 	arg.pkt_len = nla_get_u16(a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN]);
3096 	arg.exec_for_lesser_equal =
3097 		last || !actions_may_change_flow(acts_if_lesser_eq);
3098 	arg.exec_for_greater =
3099 		last || !actions_may_change_flow(acts_if_greater);
3100 
3101 	err = ovs_nla_add_action(sfa, OVS_CHECK_PKT_LEN_ATTR_ARG, &arg,
3102 				 sizeof(arg), log);
3103 	if (err)
3104 		return err;
3105 
3106 	nested_acts_start = add_nested_action_start(sfa,
3107 		OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL, log);
3108 	if (nested_acts_start < 0)
3109 		return nested_acts_start;
3110 
3111 	err = __ovs_nla_copy_actions(net, acts_if_lesser_eq, key, sfa,
3112 				     eth_type, vlan_tci, mpls_label_count, log);
3113 
3114 	if (err)
3115 		return err;
3116 
3117 	add_nested_action_end(*sfa, nested_acts_start);
3118 
3119 	nested_acts_start = add_nested_action_start(sfa,
3120 		OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER, log);
3121 	if (nested_acts_start < 0)
3122 		return nested_acts_start;
3123 
3124 	err = __ovs_nla_copy_actions(net, acts_if_greater, key, sfa,
3125 				     eth_type, vlan_tci, mpls_label_count, log);
3126 
3127 	if (err)
3128 		return err;
3129 
3130 	add_nested_action_end(*sfa, nested_acts_start);
3131 	add_nested_action_end(*sfa, start);
3132 	return 0;
3133 }
3134 
3135 static int copy_action(const struct nlattr *from,
3136 		       struct sw_flow_actions **sfa, bool log)
3137 {
3138 	int totlen = NLA_ALIGN(from->nla_len);
3139 	struct nlattr *to;
3140 
3141 	to = reserve_sfa_size(sfa, from->nla_len, log);
3142 	if (IS_ERR(to))
3143 		return PTR_ERR(to);
3144 
3145 	memcpy(to, from, totlen);
3146 	return 0;
3147 }
3148 
3149 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
3150 				  const struct sw_flow_key *key,
3151 				  struct sw_flow_actions **sfa,
3152 				  __be16 eth_type, __be16 vlan_tci,
3153 				  u32 mpls_label_count, bool log)
3154 {
3155 	u8 mac_proto = ovs_key_mac_proto(key);
3156 	const struct nlattr *a;
3157 	int rem, err;
3158 
3159 	nla_for_each_nested(a, attr, rem) {
3160 		/* Expected argument lengths, (u32)-1 for variable length. */
3161 		static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
3162 			[OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
3163 			[OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
3164 			[OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
3165 			[OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
3166 			[OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
3167 			[OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
3168 			[OVS_ACTION_ATTR_POP_VLAN] = 0,
3169 			[OVS_ACTION_ATTR_SET] = (u32)-1,
3170 			[OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
3171 			[OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
3172 			[OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
3173 			[OVS_ACTION_ATTR_CT] = (u32)-1,
3174 			[OVS_ACTION_ATTR_CT_CLEAR] = 0,
3175 			[OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
3176 			[OVS_ACTION_ATTR_PUSH_ETH] = sizeof(struct ovs_action_push_eth),
3177 			[OVS_ACTION_ATTR_POP_ETH] = 0,
3178 			[OVS_ACTION_ATTR_PUSH_NSH] = (u32)-1,
3179 			[OVS_ACTION_ATTR_POP_NSH] = 0,
3180 			[OVS_ACTION_ATTR_METER] = sizeof(u32),
3181 			[OVS_ACTION_ATTR_CLONE] = (u32)-1,
3182 			[OVS_ACTION_ATTR_CHECK_PKT_LEN] = (u32)-1,
3183 			[OVS_ACTION_ATTR_ADD_MPLS] = sizeof(struct ovs_action_add_mpls),
3184 			[OVS_ACTION_ATTR_DEC_TTL] = (u32)-1,
3185 		};
3186 		const struct ovs_action_push_vlan *vlan;
3187 		int type = nla_type(a);
3188 		bool skip_copy;
3189 
3190 		if (type > OVS_ACTION_ATTR_MAX ||
3191 		    (action_lens[type] != nla_len(a) &&
3192 		     action_lens[type] != (u32)-1))
3193 			return -EINVAL;
3194 
3195 		skip_copy = false;
3196 		switch (type) {
3197 		case OVS_ACTION_ATTR_UNSPEC:
3198 			return -EINVAL;
3199 
3200 		case OVS_ACTION_ATTR_USERSPACE:
3201 			err = validate_userspace(a);
3202 			if (err)
3203 				return err;
3204 			break;
3205 
3206 		case OVS_ACTION_ATTR_OUTPUT:
3207 			if (nla_get_u32(a) >= DP_MAX_PORTS)
3208 				return -EINVAL;
3209 			break;
3210 
3211 		case OVS_ACTION_ATTR_TRUNC: {
3212 			const struct ovs_action_trunc *trunc = nla_data(a);
3213 
3214 			if (trunc->max_len < ETH_HLEN)
3215 				return -EINVAL;
3216 			break;
3217 		}
3218 
3219 		case OVS_ACTION_ATTR_HASH: {
3220 			const struct ovs_action_hash *act_hash = nla_data(a);
3221 
3222 			switch (act_hash->hash_alg) {
3223 			case OVS_HASH_ALG_L4:
3224 				break;
3225 			default:
3226 				return  -EINVAL;
3227 			}
3228 
3229 			break;
3230 		}
3231 
3232 		case OVS_ACTION_ATTR_POP_VLAN:
3233 			if (mac_proto != MAC_PROTO_ETHERNET)
3234 				return -EINVAL;
3235 			vlan_tci = htons(0);
3236 			break;
3237 
3238 		case OVS_ACTION_ATTR_PUSH_VLAN:
3239 			if (mac_proto != MAC_PROTO_ETHERNET)
3240 				return -EINVAL;
3241 			vlan = nla_data(a);
3242 			if (!eth_type_vlan(vlan->vlan_tpid))
3243 				return -EINVAL;
3244 			if (!(vlan->vlan_tci & htons(VLAN_CFI_MASK)))
3245 				return -EINVAL;
3246 			vlan_tci = vlan->vlan_tci;
3247 			break;
3248 
3249 		case OVS_ACTION_ATTR_RECIRC:
3250 			break;
3251 
3252 		case OVS_ACTION_ATTR_ADD_MPLS: {
3253 			const struct ovs_action_add_mpls *mpls = nla_data(a);
3254 
3255 			if (!eth_p_mpls(mpls->mpls_ethertype))
3256 				return -EINVAL;
3257 
3258 			if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK) {
3259 				if (vlan_tci & htons(VLAN_CFI_MASK) ||
3260 				    (eth_type != htons(ETH_P_IP) &&
3261 				     eth_type != htons(ETH_P_IPV6) &&
3262 				     eth_type != htons(ETH_P_ARP) &&
3263 				     eth_type != htons(ETH_P_RARP) &&
3264 				     !eth_p_mpls(eth_type)))
3265 					return -EINVAL;
3266 				mpls_label_count++;
3267 			} else {
3268 				if (mac_proto == MAC_PROTO_ETHERNET) {
3269 					mpls_label_count = 1;
3270 					mac_proto = MAC_PROTO_NONE;
3271 				} else {
3272 					mpls_label_count++;
3273 				}
3274 			}
3275 			eth_type = mpls->mpls_ethertype;
3276 			break;
3277 		}
3278 
3279 		case OVS_ACTION_ATTR_PUSH_MPLS: {
3280 			const struct ovs_action_push_mpls *mpls = nla_data(a);
3281 
3282 			if (!eth_p_mpls(mpls->mpls_ethertype))
3283 				return -EINVAL;
3284 			/* Prohibit push MPLS other than to a white list
3285 			 * for packets that have a known tag order.
3286 			 */
3287 			if (vlan_tci & htons(VLAN_CFI_MASK) ||
3288 			    (eth_type != htons(ETH_P_IP) &&
3289 			     eth_type != htons(ETH_P_IPV6) &&
3290 			     eth_type != htons(ETH_P_ARP) &&
3291 			     eth_type != htons(ETH_P_RARP) &&
3292 			     !eth_p_mpls(eth_type)))
3293 				return -EINVAL;
3294 			eth_type = mpls->mpls_ethertype;
3295 			mpls_label_count++;
3296 			break;
3297 		}
3298 
3299 		case OVS_ACTION_ATTR_POP_MPLS: {
3300 			__be16  proto;
3301 			if (vlan_tci & htons(VLAN_CFI_MASK) ||
3302 			    !eth_p_mpls(eth_type))
3303 				return -EINVAL;
3304 
3305 			/* Disallow subsequent L2.5+ set actions and mpls_pop
3306 			 * actions once the last MPLS label in the packet is
3307 			 * is popped as there is no check here to ensure that
3308 			 * the new eth type is valid and thus set actions could
3309 			 * write off the end of the packet or otherwise corrupt
3310 			 * it.
3311 			 *
3312 			 * Support for these actions is planned using packet
3313 			 * recirculation.
3314 			 */
3315 			proto = nla_get_be16(a);
3316 
3317 			if (proto == htons(ETH_P_TEB) &&
3318 			    mac_proto != MAC_PROTO_NONE)
3319 				return -EINVAL;
3320 
3321 			mpls_label_count--;
3322 
3323 			if (!eth_p_mpls(proto) || !mpls_label_count)
3324 				eth_type = htons(0);
3325 			else
3326 				eth_type =  proto;
3327 
3328 			break;
3329 		}
3330 
3331 		case OVS_ACTION_ATTR_SET:
3332 			err = validate_set(a, key, sfa,
3333 					   &skip_copy, mac_proto, eth_type,
3334 					   false, log);
3335 			if (err)
3336 				return err;
3337 			break;
3338 
3339 		case OVS_ACTION_ATTR_SET_MASKED:
3340 			err = validate_set(a, key, sfa,
3341 					   &skip_copy, mac_proto, eth_type,
3342 					   true, log);
3343 			if (err)
3344 				return err;
3345 			break;
3346 
3347 		case OVS_ACTION_ATTR_SAMPLE: {
3348 			bool last = nla_is_last(a, rem);
3349 
3350 			err = validate_and_copy_sample(net, a, key, sfa,
3351 						       eth_type, vlan_tci,
3352 						       mpls_label_count,
3353 						       log, last);
3354 			if (err)
3355 				return err;
3356 			skip_copy = true;
3357 			break;
3358 		}
3359 
3360 		case OVS_ACTION_ATTR_CT:
3361 			err = ovs_ct_copy_action(net, a, key, sfa, log);
3362 			if (err)
3363 				return err;
3364 			skip_copy = true;
3365 			break;
3366 
3367 		case OVS_ACTION_ATTR_CT_CLEAR:
3368 			break;
3369 
3370 		case OVS_ACTION_ATTR_PUSH_ETH:
3371 			/* Disallow pushing an Ethernet header if one
3372 			 * is already present */
3373 			if (mac_proto != MAC_PROTO_NONE)
3374 				return -EINVAL;
3375 			mac_proto = MAC_PROTO_ETHERNET;
3376 			break;
3377 
3378 		case OVS_ACTION_ATTR_POP_ETH:
3379 			if (mac_proto != MAC_PROTO_ETHERNET)
3380 				return -EINVAL;
3381 			if (vlan_tci & htons(VLAN_CFI_MASK))
3382 				return -EINVAL;
3383 			mac_proto = MAC_PROTO_NONE;
3384 			break;
3385 
3386 		case OVS_ACTION_ATTR_PUSH_NSH:
3387 			if (mac_proto != MAC_PROTO_ETHERNET) {
3388 				u8 next_proto;
3389 
3390 				next_proto = tun_p_from_eth_p(eth_type);
3391 				if (!next_proto)
3392 					return -EINVAL;
3393 			}
3394 			mac_proto = MAC_PROTO_NONE;
3395 			if (!validate_nsh(nla_data(a), false, true, true))
3396 				return -EINVAL;
3397 			break;
3398 
3399 		case OVS_ACTION_ATTR_POP_NSH: {
3400 			__be16 inner_proto;
3401 
3402 			if (eth_type != htons(ETH_P_NSH))
3403 				return -EINVAL;
3404 			inner_proto = tun_p_to_eth_p(key->nsh.base.np);
3405 			if (!inner_proto)
3406 				return -EINVAL;
3407 			if (key->nsh.base.np == TUN_P_ETHERNET)
3408 				mac_proto = MAC_PROTO_ETHERNET;
3409 			else
3410 				mac_proto = MAC_PROTO_NONE;
3411 			break;
3412 		}
3413 
3414 		case OVS_ACTION_ATTR_METER:
3415 			/* Non-existent meters are simply ignored.  */
3416 			break;
3417 
3418 		case OVS_ACTION_ATTR_CLONE: {
3419 			bool last = nla_is_last(a, rem);
3420 
3421 			err = validate_and_copy_clone(net, a, key, sfa,
3422 						      eth_type, vlan_tci,
3423 						      mpls_label_count,
3424 						      log, last);
3425 			if (err)
3426 				return err;
3427 			skip_copy = true;
3428 			break;
3429 		}
3430 
3431 		case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
3432 			bool last = nla_is_last(a, rem);
3433 
3434 			err = validate_and_copy_check_pkt_len(net, a, key, sfa,
3435 							      eth_type,
3436 							      vlan_tci,
3437 							      mpls_label_count,
3438 							      log, last);
3439 			if (err)
3440 				return err;
3441 			skip_copy = true;
3442 			break;
3443 		}
3444 
3445 		case OVS_ACTION_ATTR_DEC_TTL:
3446 			err = validate_and_copy_dec_ttl(net, a, key, sfa,
3447 							eth_type, vlan_tci,
3448 							mpls_label_count, log);
3449 			if (err)
3450 				return err;
3451 			skip_copy = true;
3452 			break;
3453 
3454 		default:
3455 			OVS_NLERR(log, "Unknown Action type %d", type);
3456 			return -EINVAL;
3457 		}
3458 		if (!skip_copy) {
3459 			err = copy_action(a, sfa, log);
3460 			if (err)
3461 				return err;
3462 		}
3463 	}
3464 
3465 	if (rem > 0)
3466 		return -EINVAL;
3467 
3468 	return 0;
3469 }
3470 
3471 /* 'key' must be the masked key. */
3472 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
3473 			 const struct sw_flow_key *key,
3474 			 struct sw_flow_actions **sfa, bool log)
3475 {
3476 	int err;
3477 	u32 mpls_label_count = 0;
3478 
3479 	*sfa = nla_alloc_flow_actions(min(nla_len(attr), MAX_ACTIONS_BUFSIZE));
3480 	if (IS_ERR(*sfa))
3481 		return PTR_ERR(*sfa);
3482 
3483 	if (eth_p_mpls(key->eth.type))
3484 		mpls_label_count = hweight_long(key->mpls.num_labels_mask);
3485 
3486 	(*sfa)->orig_len = nla_len(attr);
3487 	err = __ovs_nla_copy_actions(net, attr, key, sfa, key->eth.type,
3488 				     key->eth.vlan.tci, mpls_label_count, log);
3489 	if (err)
3490 		ovs_nla_free_flow_actions(*sfa);
3491 
3492 	return err;
3493 }
3494 
3495 static int sample_action_to_attr(const struct nlattr *attr,
3496 				 struct sk_buff *skb)
3497 {
3498 	struct nlattr *start, *ac_start = NULL, *sample_arg;
3499 	int err = 0, rem = nla_len(attr);
3500 	const struct sample_arg *arg;
3501 	struct nlattr *actions;
3502 
3503 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SAMPLE);
3504 	if (!start)
3505 		return -EMSGSIZE;
3506 
3507 	sample_arg = nla_data(attr);
3508 	arg = nla_data(sample_arg);
3509 	actions = nla_next(sample_arg, &rem);
3510 
3511 	if (nla_put_u32(skb, OVS_SAMPLE_ATTR_PROBABILITY, arg->probability)) {
3512 		err = -EMSGSIZE;
3513 		goto out;
3514 	}
3515 
3516 	ac_start = nla_nest_start_noflag(skb, OVS_SAMPLE_ATTR_ACTIONS);
3517 	if (!ac_start) {
3518 		err = -EMSGSIZE;
3519 		goto out;
3520 	}
3521 
3522 	err = ovs_nla_put_actions(actions, rem, skb);
3523 
3524 out:
3525 	if (err) {
3526 		nla_nest_cancel(skb, ac_start);
3527 		nla_nest_cancel(skb, start);
3528 	} else {
3529 		nla_nest_end(skb, ac_start);
3530 		nla_nest_end(skb, start);
3531 	}
3532 
3533 	return err;
3534 }
3535 
3536 static int clone_action_to_attr(const struct nlattr *attr,
3537 				struct sk_buff *skb)
3538 {
3539 	struct nlattr *start;
3540 	int err = 0, rem = nla_len(attr);
3541 
3542 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CLONE);
3543 	if (!start)
3544 		return -EMSGSIZE;
3545 
3546 	/* Skipping the OVS_CLONE_ATTR_EXEC that is always the first attribute. */
3547 	attr = nla_next(nla_data(attr), &rem);
3548 	err = ovs_nla_put_actions(attr, rem, skb);
3549 
3550 	if (err)
3551 		nla_nest_cancel(skb, start);
3552 	else
3553 		nla_nest_end(skb, start);
3554 
3555 	return err;
3556 }
3557 
3558 static int check_pkt_len_action_to_attr(const struct nlattr *attr,
3559 					struct sk_buff *skb)
3560 {
3561 	struct nlattr *start, *ac_start = NULL;
3562 	const struct check_pkt_len_arg *arg;
3563 	const struct nlattr *a, *cpl_arg;
3564 	int err = 0, rem = nla_len(attr);
3565 
3566 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CHECK_PKT_LEN);
3567 	if (!start)
3568 		return -EMSGSIZE;
3569 
3570 	/* The first nested attribute in 'attr' is always
3571 	 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
3572 	 */
3573 	cpl_arg = nla_data(attr);
3574 	arg = nla_data(cpl_arg);
3575 
3576 	if (nla_put_u16(skb, OVS_CHECK_PKT_LEN_ATTR_PKT_LEN, arg->pkt_len)) {
3577 		err = -EMSGSIZE;
3578 		goto out;
3579 	}
3580 
3581 	/* Second nested attribute in 'attr' is always
3582 	 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
3583 	 */
3584 	a = nla_next(cpl_arg, &rem);
3585 	ac_start =  nla_nest_start_noflag(skb,
3586 					  OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL);
3587 	if (!ac_start) {
3588 		err = -EMSGSIZE;
3589 		goto out;
3590 	}
3591 
3592 	err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3593 	if (err) {
3594 		nla_nest_cancel(skb, ac_start);
3595 		goto out;
3596 	} else {
3597 		nla_nest_end(skb, ac_start);
3598 	}
3599 
3600 	/* Third nested attribute in 'attr' is always
3601 	 * OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER.
3602 	 */
3603 	a = nla_next(a, &rem);
3604 	ac_start =  nla_nest_start_noflag(skb,
3605 					  OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER);
3606 	if (!ac_start) {
3607 		err = -EMSGSIZE;
3608 		goto out;
3609 	}
3610 
3611 	err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3612 	if (err) {
3613 		nla_nest_cancel(skb, ac_start);
3614 		goto out;
3615 	} else {
3616 		nla_nest_end(skb, ac_start);
3617 	}
3618 
3619 	nla_nest_end(skb, start);
3620 	return 0;
3621 
3622 out:
3623 	nla_nest_cancel(skb, start);
3624 	return err;
3625 }
3626 
3627 static int dec_ttl_action_to_attr(const struct nlattr *attr,
3628 				  struct sk_buff *skb)
3629 {
3630 	struct nlattr *start, *action_start;
3631 	const struct nlattr *a;
3632 	int err = 0, rem;
3633 
3634 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_DEC_TTL);
3635 	if (!start)
3636 		return -EMSGSIZE;
3637 
3638 	nla_for_each_attr(a, nla_data(attr), nla_len(attr), rem) {
3639 		switch (nla_type(a)) {
3640 		case OVS_DEC_TTL_ATTR_ACTION:
3641 
3642 			action_start = nla_nest_start_noflag(skb, OVS_DEC_TTL_ATTR_ACTION);
3643 			if (!action_start) {
3644 				err = -EMSGSIZE;
3645 				goto out;
3646 			}
3647 
3648 			err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3649 			if (err)
3650 				goto out;
3651 
3652 			nla_nest_end(skb, action_start);
3653 			break;
3654 
3655 		default:
3656 			/* Ignore all other option to be future compatible */
3657 			break;
3658 		}
3659 	}
3660 
3661 	nla_nest_end(skb, start);
3662 	return 0;
3663 
3664 out:
3665 	nla_nest_cancel(skb, start);
3666 	return err;
3667 }
3668 
3669 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
3670 {
3671 	const struct nlattr *ovs_key = nla_data(a);
3672 	int key_type = nla_type(ovs_key);
3673 	struct nlattr *start;
3674 	int err;
3675 
3676 	switch (key_type) {
3677 	case OVS_KEY_ATTR_TUNNEL_INFO: {
3678 		struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
3679 		struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
3680 
3681 		start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SET);
3682 		if (!start)
3683 			return -EMSGSIZE;
3684 
3685 		err =  ip_tun_to_nlattr(skb, &tun_info->key,
3686 					ip_tunnel_info_opts(tun_info),
3687 					tun_info->options_len,
3688 					ip_tunnel_info_af(tun_info), tun_info->mode);
3689 		if (err)
3690 			return err;
3691 		nla_nest_end(skb, start);
3692 		break;
3693 	}
3694 	default:
3695 		if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
3696 			return -EMSGSIZE;
3697 		break;
3698 	}
3699 
3700 	return 0;
3701 }
3702 
3703 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
3704 						struct sk_buff *skb)
3705 {
3706 	const struct nlattr *ovs_key = nla_data(a);
3707 	struct nlattr *nla;
3708 	size_t key_len = nla_len(ovs_key) / 2;
3709 
3710 	/* Revert the conversion we did from a non-masked set action to
3711 	 * masked set action.
3712 	 */
3713 	nla = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SET);
3714 	if (!nla)
3715 		return -EMSGSIZE;
3716 
3717 	if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
3718 		return -EMSGSIZE;
3719 
3720 	nla_nest_end(skb, nla);
3721 	return 0;
3722 }
3723 
3724 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
3725 {
3726 	const struct nlattr *a;
3727 	int rem, err;
3728 
3729 	nla_for_each_attr(a, attr, len, rem) {
3730 		int type = nla_type(a);
3731 
3732 		switch (type) {
3733 		case OVS_ACTION_ATTR_SET:
3734 			err = set_action_to_attr(a, skb);
3735 			if (err)
3736 				return err;
3737 			break;
3738 
3739 		case OVS_ACTION_ATTR_SET_TO_MASKED:
3740 			err = masked_set_action_to_set_action_attr(a, skb);
3741 			if (err)
3742 				return err;
3743 			break;
3744 
3745 		case OVS_ACTION_ATTR_SAMPLE:
3746 			err = sample_action_to_attr(a, skb);
3747 			if (err)
3748 				return err;
3749 			break;
3750 
3751 		case OVS_ACTION_ATTR_CT:
3752 			err = ovs_ct_action_to_attr(nla_data(a), skb);
3753 			if (err)
3754 				return err;
3755 			break;
3756 
3757 		case OVS_ACTION_ATTR_CLONE:
3758 			err = clone_action_to_attr(a, skb);
3759 			if (err)
3760 				return err;
3761 			break;
3762 
3763 		case OVS_ACTION_ATTR_CHECK_PKT_LEN:
3764 			err = check_pkt_len_action_to_attr(a, skb);
3765 			if (err)
3766 				return err;
3767 			break;
3768 
3769 		case OVS_ACTION_ATTR_DEC_TTL:
3770 			err = dec_ttl_action_to_attr(a, skb);
3771 			if (err)
3772 				return err;
3773 			break;
3774 
3775 		default:
3776 			if (nla_put(skb, type, nla_len(a), nla_data(a)))
3777 				return -EMSGSIZE;
3778 			break;
3779 		}
3780 	}
3781 
3782 	return 0;
3783 }
3784