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 "drop.h"
42 #include "flow_netlink.h"
43 
44 struct ovs_len_tbl {
45 	int len;
46 	const struct ovs_len_tbl *next;
47 };
48 
49 #define OVS_ATTR_NESTED -1
50 #define OVS_ATTR_VARIABLE -2
51 #define OVS_COPY_ACTIONS_MAX_DEPTH 16
52 
53 static bool actions_may_change_flow(const struct nlattr *actions)
54 {
55 	struct nlattr *nla;
56 	int rem;
57 
58 	nla_for_each_nested(nla, actions, rem) {
59 		u16 action = nla_type(nla);
60 
61 		switch (action) {
62 		case OVS_ACTION_ATTR_OUTPUT:
63 		case OVS_ACTION_ATTR_RECIRC:
64 		case OVS_ACTION_ATTR_TRUNC:
65 		case OVS_ACTION_ATTR_USERSPACE:
66 		case OVS_ACTION_ATTR_DROP:
67 			break;
68 
69 		case OVS_ACTION_ATTR_CT:
70 		case OVS_ACTION_ATTR_CT_CLEAR:
71 		case OVS_ACTION_ATTR_HASH:
72 		case OVS_ACTION_ATTR_POP_ETH:
73 		case OVS_ACTION_ATTR_POP_MPLS:
74 		case OVS_ACTION_ATTR_POP_NSH:
75 		case OVS_ACTION_ATTR_POP_VLAN:
76 		case OVS_ACTION_ATTR_PUSH_ETH:
77 		case OVS_ACTION_ATTR_PUSH_MPLS:
78 		case OVS_ACTION_ATTR_PUSH_NSH:
79 		case OVS_ACTION_ATTR_PUSH_VLAN:
80 		case OVS_ACTION_ATTR_SAMPLE:
81 		case OVS_ACTION_ATTR_SET:
82 		case OVS_ACTION_ATTR_SET_MASKED:
83 		case OVS_ACTION_ATTR_METER:
84 		case OVS_ACTION_ATTR_CHECK_PKT_LEN:
85 		case OVS_ACTION_ATTR_ADD_MPLS:
86 		case OVS_ACTION_ATTR_DEC_TTL:
87 		default:
88 			return true;
89 		}
90 	}
91 	return false;
92 }
93 
94 static void update_range(struct sw_flow_match *match,
95 			 size_t offset, size_t size, bool is_mask)
96 {
97 	struct sw_flow_key_range *range;
98 	size_t start = rounddown(offset, sizeof(long));
99 	size_t end = roundup(offset + size, sizeof(long));
100 
101 	if (!is_mask)
102 		range = &match->range;
103 	else
104 		range = &match->mask->range;
105 
106 	if (range->start == range->end) {
107 		range->start = start;
108 		range->end = end;
109 		return;
110 	}
111 
112 	if (range->start > start)
113 		range->start = start;
114 
115 	if (range->end < end)
116 		range->end = end;
117 }
118 
119 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
120 	do { \
121 		update_range(match, offsetof(struct sw_flow_key, field),    \
122 			     sizeof((match)->key->field), is_mask);	    \
123 		if (is_mask)						    \
124 			(match)->mask->key.field = value;		    \
125 		else							    \
126 			(match)->key->field = value;		            \
127 	} while (0)
128 
129 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)	    \
130 	do {								    \
131 		update_range(match, offset, len, is_mask);		    \
132 		if (is_mask)						    \
133 			memcpy((u8 *)&(match)->mask->key + offset, value_p, \
134 			       len);					   \
135 		else							    \
136 			memcpy((u8 *)(match)->key + offset, value_p, len);  \
137 	} while (0)
138 
139 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)		      \
140 	SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
141 				  value_p, len, is_mask)
142 
143 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)		    \
144 	do {								    \
145 		update_range(match, offsetof(struct sw_flow_key, field),    \
146 			     sizeof((match)->key->field), is_mask);	    \
147 		if (is_mask)						    \
148 			memset((u8 *)&(match)->mask->key.field, value,      \
149 			       sizeof((match)->mask->key.field));	    \
150 		else							    \
151 			memset((u8 *)&(match)->key->field, value,           \
152 			       sizeof((match)->key->field));                \
153 	} while (0)
154 
155 static bool match_validate(const struct sw_flow_match *match,
156 			   u64 key_attrs, u64 mask_attrs, bool log)
157 {
158 	u64 key_expected = 0;
159 	u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
160 
161 	/* The following mask attributes allowed only if they
162 	 * pass the validation tests. */
163 	mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
164 			| (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)
165 			| (1 << OVS_KEY_ATTR_IPV6)
166 			| (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)
167 			| (1 << OVS_KEY_ATTR_TCP)
168 			| (1 << OVS_KEY_ATTR_TCP_FLAGS)
169 			| (1 << OVS_KEY_ATTR_UDP)
170 			| (1 << OVS_KEY_ATTR_SCTP)
171 			| (1 << OVS_KEY_ATTR_ICMP)
172 			| (1 << OVS_KEY_ATTR_ICMPV6)
173 			| (1 << OVS_KEY_ATTR_ARP)
174 			| (1 << OVS_KEY_ATTR_ND)
175 			| (1 << OVS_KEY_ATTR_MPLS)
176 			| (1 << OVS_KEY_ATTR_NSH));
177 
178 	/* Always allowed mask fields. */
179 	mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
180 		       | (1 << OVS_KEY_ATTR_IN_PORT)
181 		       | (1 << OVS_KEY_ATTR_ETHERTYPE));
182 
183 	/* Check key attributes. */
184 	if (match->key->eth.type == htons(ETH_P_ARP)
185 			|| match->key->eth.type == htons(ETH_P_RARP)) {
186 		key_expected |= 1 << OVS_KEY_ATTR_ARP;
187 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
188 			mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
189 	}
190 
191 	if (eth_p_mpls(match->key->eth.type)) {
192 		key_expected |= 1 << OVS_KEY_ATTR_MPLS;
193 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
194 			mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
195 	}
196 
197 	if (match->key->eth.type == htons(ETH_P_IP)) {
198 		key_expected |= 1 << OVS_KEY_ATTR_IPV4;
199 		if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
200 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
201 			mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4;
202 		}
203 
204 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
205 			if (match->key->ip.proto == IPPROTO_UDP) {
206 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
207 				if (match->mask && (match->mask->key.ip.proto == 0xff))
208 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
209 			}
210 
211 			if (match->key->ip.proto == IPPROTO_SCTP) {
212 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
213 				if (match->mask && (match->mask->key.ip.proto == 0xff))
214 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
215 			}
216 
217 			if (match->key->ip.proto == IPPROTO_TCP) {
218 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
219 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
220 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
221 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
222 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
223 				}
224 			}
225 
226 			if (match->key->ip.proto == IPPROTO_ICMP) {
227 				key_expected |= 1 << OVS_KEY_ATTR_ICMP;
228 				if (match->mask && (match->mask->key.ip.proto == 0xff))
229 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
230 			}
231 		}
232 	}
233 
234 	if (match->key->eth.type == htons(ETH_P_IPV6)) {
235 		key_expected |= 1 << OVS_KEY_ATTR_IPV6;
236 		if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
237 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
238 			mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6;
239 		}
240 
241 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
242 			if (match->key->ip.proto == IPPROTO_UDP) {
243 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
244 				if (match->mask && (match->mask->key.ip.proto == 0xff))
245 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
246 			}
247 
248 			if (match->key->ip.proto == IPPROTO_SCTP) {
249 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
250 				if (match->mask && (match->mask->key.ip.proto == 0xff))
251 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
252 			}
253 
254 			if (match->key->ip.proto == IPPROTO_TCP) {
255 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
256 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
257 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
258 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
259 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
260 				}
261 			}
262 
263 			if (match->key->ip.proto == IPPROTO_ICMPV6) {
264 				key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
265 				if (match->mask && (match->mask->key.ip.proto == 0xff))
266 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
267 
268 				if (match->key->tp.src ==
269 						htons(NDISC_NEIGHBOUR_SOLICITATION) ||
270 				    match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
271 					key_expected |= 1 << OVS_KEY_ATTR_ND;
272 					/* Original direction conntrack tuple
273 					 * uses the same space as the ND fields
274 					 * in the key, so both are not allowed
275 					 * at the same time.
276 					 */
277 					mask_allowed &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
278 					if (match->mask && (match->mask->key.tp.src == htons(0xff)))
279 						mask_allowed |= 1 << OVS_KEY_ATTR_ND;
280 				}
281 			}
282 		}
283 	}
284 
285 	if (match->key->eth.type == htons(ETH_P_NSH)) {
286 		key_expected |= 1 << OVS_KEY_ATTR_NSH;
287 		if (match->mask &&
288 		    match->mask->key.eth.type == htons(0xffff)) {
289 			mask_allowed |= 1 << OVS_KEY_ATTR_NSH;
290 		}
291 	}
292 
293 	if ((key_attrs & key_expected) != key_expected) {
294 		/* Key attributes check failed. */
295 		OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
296 			  (unsigned long long)key_attrs,
297 			  (unsigned long long)key_expected);
298 		return false;
299 	}
300 
301 	if ((mask_attrs & mask_allowed) != mask_attrs) {
302 		/* Mask attributes check failed. */
303 		OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
304 			  (unsigned long long)mask_attrs,
305 			  (unsigned long long)mask_allowed);
306 		return false;
307 	}
308 
309 	return true;
310 }
311 
312 size_t ovs_tun_key_attr_size(void)
313 {
314 	/* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
315 	 * updating this function.
316 	 */
317 	return    nla_total_size_64bit(8) /* OVS_TUNNEL_KEY_ATTR_ID */
318 		+ nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
319 		+ nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_DST */
320 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
321 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
322 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
323 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
324 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
325 		+ nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
326 		/* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS and
327 		 * OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS is mutually exclusive with
328 		 * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
329 		 */
330 		+ nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
331 		+ nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
332 }
333 
334 static size_t ovs_nsh_key_attr_size(void)
335 {
336 	/* Whenever adding new OVS_NSH_KEY_ FIELDS, we should consider
337 	 * updating this function.
338 	 */
339 	return  nla_total_size(NSH_BASE_HDR_LEN) /* OVS_NSH_KEY_ATTR_BASE */
340 		/* OVS_NSH_KEY_ATTR_MD1 and OVS_NSH_KEY_ATTR_MD2 are
341 		 * mutually exclusive, so the bigger one can cover
342 		 * the small one.
343 		 */
344 		+ nla_total_size(NSH_CTX_HDRS_MAX_LEN);
345 }
346 
347 size_t ovs_key_attr_size(void)
348 {
349 	/* Whenever adding new OVS_KEY_ FIELDS, we should consider
350 	 * updating this function.
351 	 */
352 	BUILD_BUG_ON(OVS_KEY_ATTR_MAX != 32);
353 
354 	return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
355 		+ nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
356 		  + ovs_tun_key_attr_size()
357 		+ nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
358 		+ nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
359 		+ nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
360 		+ nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
361 		+ nla_total_size(4)   /* OVS_KEY_ATTR_CT_STATE */
362 		+ nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
363 		+ nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
364 		+ nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABELS */
365 		+ nla_total_size(40)  /* OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6 */
366 		+ nla_total_size(0)   /* OVS_KEY_ATTR_NSH */
367 		  + ovs_nsh_key_attr_size()
368 		+ nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
369 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
370 		+ nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
371 		+ nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
372 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
373 		+ nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
374 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
375 		+ nla_total_size(28)  /* OVS_KEY_ATTR_ND */
376 		+ nla_total_size(2);  /* OVS_KEY_ATTR_IPV6_EXTHDRS */
377 }
378 
379 static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
380 	[OVS_VXLAN_EXT_GBP]	    = { .len = sizeof(u32) },
381 };
382 
383 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
384 	[OVS_TUNNEL_KEY_ATTR_ID]	    = { .len = sizeof(u64) },
385 	[OVS_TUNNEL_KEY_ATTR_IPV4_SRC]	    = { .len = sizeof(u32) },
386 	[OVS_TUNNEL_KEY_ATTR_IPV4_DST]	    = { .len = sizeof(u32) },
387 	[OVS_TUNNEL_KEY_ATTR_TOS]	    = { .len = 1 },
388 	[OVS_TUNNEL_KEY_ATTR_TTL]	    = { .len = 1 },
389 	[OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
390 	[OVS_TUNNEL_KEY_ATTR_CSUM]	    = { .len = 0 },
391 	[OVS_TUNNEL_KEY_ATTR_TP_SRC]	    = { .len = sizeof(u16) },
392 	[OVS_TUNNEL_KEY_ATTR_TP_DST]	    = { .len = sizeof(u16) },
393 	[OVS_TUNNEL_KEY_ATTR_OAM]	    = { .len = 0 },
394 	[OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_VARIABLE },
395 	[OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED,
396 						.next = ovs_vxlan_ext_key_lens },
397 	[OVS_TUNNEL_KEY_ATTR_IPV6_SRC]      = { .len = sizeof(struct in6_addr) },
398 	[OVS_TUNNEL_KEY_ATTR_IPV6_DST]      = { .len = sizeof(struct in6_addr) },
399 	[OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS]   = { .len = OVS_ATTR_VARIABLE },
400 	[OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE]   = { .len = 0 },
401 };
402 
403 static const struct ovs_len_tbl
404 ovs_nsh_key_attr_lens[OVS_NSH_KEY_ATTR_MAX + 1] = {
405 	[OVS_NSH_KEY_ATTR_BASE] = { .len = sizeof(struct ovs_nsh_key_base) },
406 	[OVS_NSH_KEY_ATTR_MD1]  = { .len = sizeof(struct ovs_nsh_key_md1) },
407 	[OVS_NSH_KEY_ATTR_MD2]  = { .len = OVS_ATTR_VARIABLE },
408 };
409 
410 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
411 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
412 	[OVS_KEY_ATTR_ENCAP]	 = { .len = OVS_ATTR_NESTED },
413 	[OVS_KEY_ATTR_PRIORITY]	 = { .len = sizeof(u32) },
414 	[OVS_KEY_ATTR_IN_PORT]	 = { .len = sizeof(u32) },
415 	[OVS_KEY_ATTR_SKB_MARK]	 = { .len = sizeof(u32) },
416 	[OVS_KEY_ATTR_ETHERNET]	 = { .len = sizeof(struct ovs_key_ethernet) },
417 	[OVS_KEY_ATTR_VLAN]	 = { .len = sizeof(__be16) },
418 	[OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
419 	[OVS_KEY_ATTR_IPV4]	 = { .len = sizeof(struct ovs_key_ipv4) },
420 	[OVS_KEY_ATTR_IPV6]	 = { .len = sizeof(struct ovs_key_ipv6) },
421 	[OVS_KEY_ATTR_TCP]	 = { .len = sizeof(struct ovs_key_tcp) },
422 	[OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
423 	[OVS_KEY_ATTR_UDP]	 = { .len = sizeof(struct ovs_key_udp) },
424 	[OVS_KEY_ATTR_SCTP]	 = { .len = sizeof(struct ovs_key_sctp) },
425 	[OVS_KEY_ATTR_ICMP]	 = { .len = sizeof(struct ovs_key_icmp) },
426 	[OVS_KEY_ATTR_ICMPV6]	 = { .len = sizeof(struct ovs_key_icmpv6) },
427 	[OVS_KEY_ATTR_ARP]	 = { .len = sizeof(struct ovs_key_arp) },
428 	[OVS_KEY_ATTR_ND]	 = { .len = sizeof(struct ovs_key_nd) },
429 	[OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
430 	[OVS_KEY_ATTR_DP_HASH]	 = { .len = sizeof(u32) },
431 	[OVS_KEY_ATTR_TUNNEL]	 = { .len = OVS_ATTR_NESTED,
432 				     .next = ovs_tunnel_key_lens, },
433 	[OVS_KEY_ATTR_MPLS]	 = { .len = OVS_ATTR_VARIABLE },
434 	[OVS_KEY_ATTR_CT_STATE]	 = { .len = sizeof(u32) },
435 	[OVS_KEY_ATTR_CT_ZONE]	 = { .len = sizeof(u16) },
436 	[OVS_KEY_ATTR_CT_MARK]	 = { .len = sizeof(u32) },
437 	[OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
438 	[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4] = {
439 		.len = sizeof(struct ovs_key_ct_tuple_ipv4) },
440 	[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6] = {
441 		.len = sizeof(struct ovs_key_ct_tuple_ipv6) },
442 	[OVS_KEY_ATTR_NSH]       = { .len = OVS_ATTR_NESTED,
443 				     .next = ovs_nsh_key_attr_lens, },
444 	[OVS_KEY_ATTR_IPV6_EXTHDRS] = {
445 		.len = sizeof(struct ovs_key_ipv6_exthdrs) },
446 };
447 
448 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
449 {
450 	return expected_len == attr_len ||
451 	       expected_len == OVS_ATTR_NESTED ||
452 	       expected_len == OVS_ATTR_VARIABLE;
453 }
454 
455 static bool is_all_zero(const u8 *fp, size_t size)
456 {
457 	int i;
458 
459 	if (!fp)
460 		return false;
461 
462 	for (i = 0; i < size; i++)
463 		if (fp[i])
464 			return false;
465 
466 	return true;
467 }
468 
469 static int __parse_flow_nlattrs(const struct nlattr *attr,
470 				const struct nlattr *a[],
471 				u64 *attrsp, bool log, bool nz)
472 {
473 	const struct nlattr *nla;
474 	u64 attrs;
475 	int rem;
476 
477 	attrs = *attrsp;
478 	nla_for_each_nested(nla, attr, rem) {
479 		u16 type = nla_type(nla);
480 		int expected_len;
481 
482 		if (type > OVS_KEY_ATTR_MAX) {
483 			OVS_NLERR(log, "Key type %d is out of range max %d",
484 				  type, OVS_KEY_ATTR_MAX);
485 			return -EINVAL;
486 		}
487 
488 		if (type == OVS_KEY_ATTR_PACKET_TYPE ||
489 		    type == OVS_KEY_ATTR_ND_EXTENSIONS ||
490 		    type == OVS_KEY_ATTR_TUNNEL_INFO) {
491 			OVS_NLERR(log, "Key type %d is not supported", type);
492 			return -EINVAL;
493 		}
494 
495 		if (attrs & (1ULL << type)) {
496 			OVS_NLERR(log, "Duplicate key (type %d).", type);
497 			return -EINVAL;
498 		}
499 
500 		expected_len = ovs_key_lens[type].len;
501 		if (!check_attr_len(nla_len(nla), expected_len)) {
502 			OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
503 				  type, nla_len(nla), expected_len);
504 			return -EINVAL;
505 		}
506 
507 		if (!nz || !is_all_zero(nla_data(nla), nla_len(nla))) {
508 			attrs |= 1ULL << type;
509 			a[type] = nla;
510 		}
511 	}
512 	if (rem) {
513 		OVS_NLERR(log, "Message has %d unknown bytes.", rem);
514 		return -EINVAL;
515 	}
516 
517 	*attrsp = attrs;
518 	return 0;
519 }
520 
521 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
522 				   const struct nlattr *a[], u64 *attrsp,
523 				   bool log)
524 {
525 	return __parse_flow_nlattrs(attr, a, attrsp, log, true);
526 }
527 
528 int parse_flow_nlattrs(const struct nlattr *attr, const struct nlattr *a[],
529 		       u64 *attrsp, bool log)
530 {
531 	return __parse_flow_nlattrs(attr, a, attrsp, log, false);
532 }
533 
534 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
535 				     struct sw_flow_match *match, bool is_mask,
536 				     bool log)
537 {
538 	unsigned long opt_key_offset;
539 
540 	if (nla_len(a) > sizeof(match->key->tun_opts)) {
541 		OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
542 			  nla_len(a), sizeof(match->key->tun_opts));
543 		return -EINVAL;
544 	}
545 
546 	if (nla_len(a) % 4 != 0) {
547 		OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
548 			  nla_len(a));
549 		return -EINVAL;
550 	}
551 
552 	/* We need to record the length of the options passed
553 	 * down, otherwise packets with the same format but
554 	 * additional options will be silently matched.
555 	 */
556 	if (!is_mask) {
557 		SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
558 				false);
559 	} else {
560 		/* This is somewhat unusual because it looks at
561 		 * both the key and mask while parsing the
562 		 * attributes (and by extension assumes the key
563 		 * is parsed first). Normally, we would verify
564 		 * that each is the correct length and that the
565 		 * attributes line up in the validate function.
566 		 * However, that is difficult because this is
567 		 * variable length and we won't have the
568 		 * information later.
569 		 */
570 		if (match->key->tun_opts_len != nla_len(a)) {
571 			OVS_NLERR(log, "Geneve option len %d != mask len %d",
572 				  match->key->tun_opts_len, nla_len(a));
573 			return -EINVAL;
574 		}
575 
576 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
577 	}
578 
579 	opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
580 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
581 				  nla_len(a), is_mask);
582 	return 0;
583 }
584 
585 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
586 				     struct sw_flow_match *match, bool is_mask,
587 				     bool log)
588 {
589 	struct nlattr *a;
590 	int rem;
591 	unsigned long opt_key_offset;
592 	struct vxlan_metadata opts;
593 
594 	BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
595 
596 	memset(&opts, 0, sizeof(opts));
597 	nla_for_each_nested(a, attr, rem) {
598 		int type = nla_type(a);
599 
600 		if (type > OVS_VXLAN_EXT_MAX) {
601 			OVS_NLERR(log, "VXLAN extension %d out of range max %d",
602 				  type, OVS_VXLAN_EXT_MAX);
603 			return -EINVAL;
604 		}
605 
606 		if (!check_attr_len(nla_len(a),
607 				    ovs_vxlan_ext_key_lens[type].len)) {
608 			OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
609 				  type, nla_len(a),
610 				  ovs_vxlan_ext_key_lens[type].len);
611 			return -EINVAL;
612 		}
613 
614 		switch (type) {
615 		case OVS_VXLAN_EXT_GBP:
616 			opts.gbp = nla_get_u32(a);
617 			break;
618 		default:
619 			OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
620 				  type);
621 			return -EINVAL;
622 		}
623 	}
624 	if (rem) {
625 		OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
626 			  rem);
627 		return -EINVAL;
628 	}
629 
630 	if (!is_mask)
631 		SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
632 	else
633 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
634 
635 	opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
636 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
637 				  is_mask);
638 	return 0;
639 }
640 
641 static int erspan_tun_opt_from_nlattr(const struct nlattr *a,
642 				      struct sw_flow_match *match, bool is_mask,
643 				      bool log)
644 {
645 	unsigned long opt_key_offset;
646 
647 	BUILD_BUG_ON(sizeof(struct erspan_metadata) >
648 		     sizeof(match->key->tun_opts));
649 
650 	if (nla_len(a) > sizeof(match->key->tun_opts)) {
651 		OVS_NLERR(log, "ERSPAN option length err (len %d, max %zu).",
652 			  nla_len(a), sizeof(match->key->tun_opts));
653 		return -EINVAL;
654 	}
655 
656 	if (!is_mask)
657 		SW_FLOW_KEY_PUT(match, tun_opts_len,
658 				sizeof(struct erspan_metadata), false);
659 	else
660 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
661 
662 	opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
663 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
664 				  nla_len(a), is_mask);
665 	return 0;
666 }
667 
668 static int ip_tun_from_nlattr(const struct nlattr *attr,
669 			      struct sw_flow_match *match, bool is_mask,
670 			      bool log)
671 {
672 	bool ttl = false, ipv4 = false, ipv6 = false;
673 	bool info_bridge_mode = false;
674 	__be16 tun_flags = 0;
675 	int opts_type = 0;
676 	struct nlattr *a;
677 	int rem;
678 
679 	nla_for_each_nested(a, attr, rem) {
680 		int type = nla_type(a);
681 		int err;
682 
683 		if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
684 			OVS_NLERR(log, "Tunnel attr %d out of range max %d",
685 				  type, OVS_TUNNEL_KEY_ATTR_MAX);
686 			return -EINVAL;
687 		}
688 
689 		if (!check_attr_len(nla_len(a),
690 				    ovs_tunnel_key_lens[type].len)) {
691 			OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
692 				  type, nla_len(a), ovs_tunnel_key_lens[type].len);
693 			return -EINVAL;
694 		}
695 
696 		switch (type) {
697 		case OVS_TUNNEL_KEY_ATTR_ID:
698 			SW_FLOW_KEY_PUT(match, tun_key.tun_id,
699 					nla_get_be64(a), is_mask);
700 			tun_flags |= TUNNEL_KEY;
701 			break;
702 		case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
703 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
704 					nla_get_in_addr(a), is_mask);
705 			ipv4 = true;
706 			break;
707 		case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
708 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
709 					nla_get_in_addr(a), is_mask);
710 			ipv4 = true;
711 			break;
712 		case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
713 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.src,
714 					nla_get_in6_addr(a), is_mask);
715 			ipv6 = true;
716 			break;
717 		case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
718 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
719 					nla_get_in6_addr(a), is_mask);
720 			ipv6 = true;
721 			break;
722 		case OVS_TUNNEL_KEY_ATTR_TOS:
723 			SW_FLOW_KEY_PUT(match, tun_key.tos,
724 					nla_get_u8(a), is_mask);
725 			break;
726 		case OVS_TUNNEL_KEY_ATTR_TTL:
727 			SW_FLOW_KEY_PUT(match, tun_key.ttl,
728 					nla_get_u8(a), is_mask);
729 			ttl = true;
730 			break;
731 		case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
732 			tun_flags |= TUNNEL_DONT_FRAGMENT;
733 			break;
734 		case OVS_TUNNEL_KEY_ATTR_CSUM:
735 			tun_flags |= TUNNEL_CSUM;
736 			break;
737 		case OVS_TUNNEL_KEY_ATTR_TP_SRC:
738 			SW_FLOW_KEY_PUT(match, tun_key.tp_src,
739 					nla_get_be16(a), is_mask);
740 			break;
741 		case OVS_TUNNEL_KEY_ATTR_TP_DST:
742 			SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
743 					nla_get_be16(a), is_mask);
744 			break;
745 		case OVS_TUNNEL_KEY_ATTR_OAM:
746 			tun_flags |= TUNNEL_OAM;
747 			break;
748 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
749 			if (opts_type) {
750 				OVS_NLERR(log, "Multiple metadata blocks provided");
751 				return -EINVAL;
752 			}
753 
754 			err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
755 			if (err)
756 				return err;
757 
758 			tun_flags |= TUNNEL_GENEVE_OPT;
759 			opts_type = type;
760 			break;
761 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
762 			if (opts_type) {
763 				OVS_NLERR(log, "Multiple metadata blocks provided");
764 				return -EINVAL;
765 			}
766 
767 			err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
768 			if (err)
769 				return err;
770 
771 			tun_flags |= TUNNEL_VXLAN_OPT;
772 			opts_type = type;
773 			break;
774 		case OVS_TUNNEL_KEY_ATTR_PAD:
775 			break;
776 		case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
777 			if (opts_type) {
778 				OVS_NLERR(log, "Multiple metadata blocks provided");
779 				return -EINVAL;
780 			}
781 
782 			err = erspan_tun_opt_from_nlattr(a, match, is_mask,
783 							 log);
784 			if (err)
785 				return err;
786 
787 			tun_flags |= TUNNEL_ERSPAN_OPT;
788 			opts_type = type;
789 			break;
790 		case OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE:
791 			info_bridge_mode = true;
792 			ipv4 = true;
793 			break;
794 		default:
795 			OVS_NLERR(log, "Unknown IP tunnel attribute %d",
796 				  type);
797 			return -EINVAL;
798 		}
799 	}
800 
801 	SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
802 	if (is_mask)
803 		SW_FLOW_KEY_MEMSET_FIELD(match, tun_proto, 0xff, true);
804 	else
805 		SW_FLOW_KEY_PUT(match, tun_proto, ipv6 ? AF_INET6 : AF_INET,
806 				false);
807 
808 	if (rem > 0) {
809 		OVS_NLERR(log, "IP tunnel attribute has %d unknown bytes.",
810 			  rem);
811 		return -EINVAL;
812 	}
813 
814 	if (ipv4 && ipv6) {
815 		OVS_NLERR(log, "Mixed IPv4 and IPv6 tunnel attributes");
816 		return -EINVAL;
817 	}
818 
819 	if (!is_mask) {
820 		if (!ipv4 && !ipv6) {
821 			OVS_NLERR(log, "IP tunnel dst address not specified");
822 			return -EINVAL;
823 		}
824 		if (ipv4) {
825 			if (info_bridge_mode) {
826 				if (match->key->tun_key.u.ipv4.src ||
827 				    match->key->tun_key.u.ipv4.dst ||
828 				    match->key->tun_key.tp_src ||
829 				    match->key->tun_key.tp_dst ||
830 				    match->key->tun_key.ttl ||
831 				    match->key->tun_key.tos ||
832 				    tun_flags & ~TUNNEL_KEY) {
833 					OVS_NLERR(log, "IPv4 tun info is not correct");
834 					return -EINVAL;
835 				}
836 			} else if (!match->key->tun_key.u.ipv4.dst) {
837 				OVS_NLERR(log, "IPv4 tunnel dst address is zero");
838 				return -EINVAL;
839 			}
840 		}
841 		if (ipv6 && ipv6_addr_any(&match->key->tun_key.u.ipv6.dst)) {
842 			OVS_NLERR(log, "IPv6 tunnel dst address is zero");
843 			return -EINVAL;
844 		}
845 
846 		if (!ttl && !info_bridge_mode) {
847 			OVS_NLERR(log, "IP tunnel TTL not specified.");
848 			return -EINVAL;
849 		}
850 	}
851 
852 	return opts_type;
853 }
854 
855 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
856 			       const void *tun_opts, int swkey_tun_opts_len)
857 {
858 	const struct vxlan_metadata *opts = tun_opts;
859 	struct nlattr *nla;
860 
861 	nla = nla_nest_start_noflag(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
862 	if (!nla)
863 		return -EMSGSIZE;
864 
865 	if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
866 		return -EMSGSIZE;
867 
868 	nla_nest_end(skb, nla);
869 	return 0;
870 }
871 
872 static int __ip_tun_to_nlattr(struct sk_buff *skb,
873 			      const struct ip_tunnel_key *output,
874 			      const void *tun_opts, int swkey_tun_opts_len,
875 			      unsigned short tun_proto, u8 mode)
876 {
877 	if (output->tun_flags & TUNNEL_KEY &&
878 	    nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id,
879 			 OVS_TUNNEL_KEY_ATTR_PAD))
880 		return -EMSGSIZE;
881 
882 	if (mode & IP_TUNNEL_INFO_BRIDGE)
883 		return nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE)
884 		       ? -EMSGSIZE : 0;
885 
886 	switch (tun_proto) {
887 	case AF_INET:
888 		if (output->u.ipv4.src &&
889 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
890 				    output->u.ipv4.src))
891 			return -EMSGSIZE;
892 		if (output->u.ipv4.dst &&
893 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
894 				    output->u.ipv4.dst))
895 			return -EMSGSIZE;
896 		break;
897 	case AF_INET6:
898 		if (!ipv6_addr_any(&output->u.ipv6.src) &&
899 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
900 				     &output->u.ipv6.src))
901 			return -EMSGSIZE;
902 		if (!ipv6_addr_any(&output->u.ipv6.dst) &&
903 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
904 				     &output->u.ipv6.dst))
905 			return -EMSGSIZE;
906 		break;
907 	}
908 	if (output->tos &&
909 	    nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
910 		return -EMSGSIZE;
911 	if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
912 		return -EMSGSIZE;
913 	if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
914 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
915 		return -EMSGSIZE;
916 	if ((output->tun_flags & TUNNEL_CSUM) &&
917 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
918 		return -EMSGSIZE;
919 	if (output->tp_src &&
920 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
921 		return -EMSGSIZE;
922 	if (output->tp_dst &&
923 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
924 		return -EMSGSIZE;
925 	if ((output->tun_flags & TUNNEL_OAM) &&
926 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
927 		return -EMSGSIZE;
928 	if (swkey_tun_opts_len) {
929 		if (output->tun_flags & TUNNEL_GENEVE_OPT &&
930 		    nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
931 			    swkey_tun_opts_len, tun_opts))
932 			return -EMSGSIZE;
933 		else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
934 			 vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
935 			return -EMSGSIZE;
936 		else if (output->tun_flags & TUNNEL_ERSPAN_OPT &&
937 			 nla_put(skb, OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS,
938 				 swkey_tun_opts_len, tun_opts))
939 			return -EMSGSIZE;
940 	}
941 
942 	return 0;
943 }
944 
945 static int ip_tun_to_nlattr(struct sk_buff *skb,
946 			    const struct ip_tunnel_key *output,
947 			    const void *tun_opts, int swkey_tun_opts_len,
948 			    unsigned short tun_proto, u8 mode)
949 {
950 	struct nlattr *nla;
951 	int err;
952 
953 	nla = nla_nest_start_noflag(skb, OVS_KEY_ATTR_TUNNEL);
954 	if (!nla)
955 		return -EMSGSIZE;
956 
957 	err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
958 				 tun_proto, mode);
959 	if (err)
960 		return err;
961 
962 	nla_nest_end(skb, nla);
963 	return 0;
964 }
965 
966 int ovs_nla_put_tunnel_info(struct sk_buff *skb,
967 			    struct ip_tunnel_info *tun_info)
968 {
969 	return __ip_tun_to_nlattr(skb, &tun_info->key,
970 				  ip_tunnel_info_opts(tun_info),
971 				  tun_info->options_len,
972 				  ip_tunnel_info_af(tun_info), tun_info->mode);
973 }
974 
975 static int encode_vlan_from_nlattrs(struct sw_flow_match *match,
976 				    const struct nlattr *a[],
977 				    bool is_mask, bool inner)
978 {
979 	__be16 tci = 0;
980 	__be16 tpid = 0;
981 
982 	if (a[OVS_KEY_ATTR_VLAN])
983 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
984 
985 	if (a[OVS_KEY_ATTR_ETHERTYPE])
986 		tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
987 
988 	if (likely(!inner)) {
989 		SW_FLOW_KEY_PUT(match, eth.vlan.tpid, tpid, is_mask);
990 		SW_FLOW_KEY_PUT(match, eth.vlan.tci, tci, is_mask);
991 	} else {
992 		SW_FLOW_KEY_PUT(match, eth.cvlan.tpid, tpid, is_mask);
993 		SW_FLOW_KEY_PUT(match, eth.cvlan.tci, tci, is_mask);
994 	}
995 	return 0;
996 }
997 
998 static int validate_vlan_from_nlattrs(const struct sw_flow_match *match,
999 				      u64 key_attrs, bool inner,
1000 				      const struct nlattr **a, bool log)
1001 {
1002 	__be16 tci = 0;
1003 
1004 	if (!((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1005 	      (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1006 	       eth_type_vlan(nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE])))) {
1007 		/* Not a VLAN. */
1008 		return 0;
1009 	}
1010 
1011 	if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1012 	      (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1013 		OVS_NLERR(log, "Invalid %s frame", (inner) ? "C-VLAN" : "VLAN");
1014 		return -EINVAL;
1015 	}
1016 
1017 	if (a[OVS_KEY_ATTR_VLAN])
1018 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1019 
1020 	if (!(tci & htons(VLAN_CFI_MASK))) {
1021 		if (tci) {
1022 			OVS_NLERR(log, "%s TCI does not have VLAN_CFI_MASK bit set.",
1023 				  (inner) ? "C-VLAN" : "VLAN");
1024 			return -EINVAL;
1025 		} else if (nla_len(a[OVS_KEY_ATTR_ENCAP])) {
1026 			/* Corner case for truncated VLAN header. */
1027 			OVS_NLERR(log, "Truncated %s header has non-zero encap attribute.",
1028 				  (inner) ? "C-VLAN" : "VLAN");
1029 			return -EINVAL;
1030 		}
1031 	}
1032 
1033 	return 1;
1034 }
1035 
1036 static int validate_vlan_mask_from_nlattrs(const struct sw_flow_match *match,
1037 					   u64 key_attrs, bool inner,
1038 					   const struct nlattr **a, bool log)
1039 {
1040 	__be16 tci = 0;
1041 	__be16 tpid = 0;
1042 	bool encap_valid = !!(match->key->eth.vlan.tci &
1043 			      htons(VLAN_CFI_MASK));
1044 	bool i_encap_valid = !!(match->key->eth.cvlan.tci &
1045 				htons(VLAN_CFI_MASK));
1046 
1047 	if (!(key_attrs & (1 << OVS_KEY_ATTR_ENCAP))) {
1048 		/* Not a VLAN. */
1049 		return 0;
1050 	}
1051 
1052 	if ((!inner && !encap_valid) || (inner && !i_encap_valid)) {
1053 		OVS_NLERR(log, "Encap mask attribute is set for non-%s frame.",
1054 			  (inner) ? "C-VLAN" : "VLAN");
1055 		return -EINVAL;
1056 	}
1057 
1058 	if (a[OVS_KEY_ATTR_VLAN])
1059 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1060 
1061 	if (a[OVS_KEY_ATTR_ETHERTYPE])
1062 		tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1063 
1064 	if (tpid != htons(0xffff)) {
1065 		OVS_NLERR(log, "Must have an exact match on %s TPID (mask=%x).",
1066 			  (inner) ? "C-VLAN" : "VLAN", ntohs(tpid));
1067 		return -EINVAL;
1068 	}
1069 	if (!(tci & htons(VLAN_CFI_MASK))) {
1070 		OVS_NLERR(log, "%s TCI mask does not have exact match for VLAN_CFI_MASK bit.",
1071 			  (inner) ? "C-VLAN" : "VLAN");
1072 		return -EINVAL;
1073 	}
1074 
1075 	return 1;
1076 }
1077 
1078 static int __parse_vlan_from_nlattrs(struct sw_flow_match *match,
1079 				     u64 *key_attrs, bool inner,
1080 				     const struct nlattr **a, bool is_mask,
1081 				     bool log)
1082 {
1083 	int err;
1084 	const struct nlattr *encap;
1085 
1086 	if (!is_mask)
1087 		err = validate_vlan_from_nlattrs(match, *key_attrs, inner,
1088 						 a, log);
1089 	else
1090 		err = validate_vlan_mask_from_nlattrs(match, *key_attrs, inner,
1091 						      a, log);
1092 	if (err <= 0)
1093 		return err;
1094 
1095 	err = encode_vlan_from_nlattrs(match, a, is_mask, inner);
1096 	if (err)
1097 		return err;
1098 
1099 	*key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1100 	*key_attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
1101 	*key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1102 
1103 	encap = a[OVS_KEY_ATTR_ENCAP];
1104 
1105 	if (!is_mask)
1106 		err = parse_flow_nlattrs(encap, a, key_attrs, log);
1107 	else
1108 		err = parse_flow_mask_nlattrs(encap, a, key_attrs, log);
1109 
1110 	return err;
1111 }
1112 
1113 static int parse_vlan_from_nlattrs(struct sw_flow_match *match,
1114 				   u64 *key_attrs, const struct nlattr **a,
1115 				   bool is_mask, bool log)
1116 {
1117 	int err;
1118 	bool encap_valid = false;
1119 
1120 	err = __parse_vlan_from_nlattrs(match, key_attrs, false, a,
1121 					is_mask, log);
1122 	if (err)
1123 		return err;
1124 
1125 	encap_valid = !!(match->key->eth.vlan.tci & htons(VLAN_CFI_MASK));
1126 	if (encap_valid) {
1127 		err = __parse_vlan_from_nlattrs(match, key_attrs, true, a,
1128 						is_mask, log);
1129 		if (err)
1130 			return err;
1131 	}
1132 
1133 	return 0;
1134 }
1135 
1136 static int parse_eth_type_from_nlattrs(struct sw_flow_match *match,
1137 				       u64 *attrs, const struct nlattr **a,
1138 				       bool is_mask, bool log)
1139 {
1140 	__be16 eth_type;
1141 
1142 	eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1143 	if (is_mask) {
1144 		/* Always exact match EtherType. */
1145 		eth_type = htons(0xffff);
1146 	} else if (!eth_proto_is_802_3(eth_type)) {
1147 		OVS_NLERR(log, "EtherType %x is less than min %x",
1148 				ntohs(eth_type), ETH_P_802_3_MIN);
1149 		return -EINVAL;
1150 	}
1151 
1152 	SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1153 	*attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1154 	return 0;
1155 }
1156 
1157 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
1158 				 u64 *attrs, const struct nlattr **a,
1159 				 bool is_mask, bool log)
1160 {
1161 	u8 mac_proto = MAC_PROTO_ETHERNET;
1162 
1163 	if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
1164 		u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
1165 
1166 		SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
1167 		*attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
1168 	}
1169 
1170 	if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
1171 		u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
1172 
1173 		SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
1174 		*attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
1175 	}
1176 
1177 	if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1178 		SW_FLOW_KEY_PUT(match, phy.priority,
1179 			  nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1180 		*attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1181 	}
1182 
1183 	if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1184 		u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1185 
1186 		if (is_mask) {
1187 			in_port = 0xffffffff; /* Always exact match in_port. */
1188 		} else if (in_port >= DP_MAX_PORTS) {
1189 			OVS_NLERR(log, "Port %d exceeds max allowable %d",
1190 				  in_port, DP_MAX_PORTS);
1191 			return -EINVAL;
1192 		}
1193 
1194 		SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1195 		*attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1196 	} else if (!is_mask) {
1197 		SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1198 	}
1199 
1200 	if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1201 		uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1202 
1203 		SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1204 		*attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1205 	}
1206 	if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1207 		if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1208 				       is_mask, log) < 0)
1209 			return -EINVAL;
1210 		*attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1211 	}
1212 
1213 	if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
1214 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
1215 		u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
1216 
1217 		if (ct_state & ~CT_SUPPORTED_MASK) {
1218 			OVS_NLERR(log, "ct_state flags %08x unsupported",
1219 				  ct_state);
1220 			return -EINVAL;
1221 		}
1222 
1223 		SW_FLOW_KEY_PUT(match, ct_state, ct_state, is_mask);
1224 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
1225 	}
1226 	if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
1227 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
1228 		u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
1229 
1230 		SW_FLOW_KEY_PUT(match, ct_zone, ct_zone, is_mask);
1231 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
1232 	}
1233 	if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
1234 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
1235 		u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
1236 
1237 		SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
1238 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
1239 	}
1240 	if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
1241 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
1242 		const struct ovs_key_ct_labels *cl;
1243 
1244 		cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
1245 		SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
1246 				   sizeof(*cl), is_mask);
1247 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
1248 	}
1249 	if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)) {
1250 		const struct ovs_key_ct_tuple_ipv4 *ct;
1251 
1252 		ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4]);
1253 
1254 		SW_FLOW_KEY_PUT(match, ipv4.ct_orig.src, ct->ipv4_src, is_mask);
1255 		SW_FLOW_KEY_PUT(match, ipv4.ct_orig.dst, ct->ipv4_dst, is_mask);
1256 		SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1257 		SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1258 		SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv4_proto, is_mask);
1259 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4);
1260 	}
1261 	if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)) {
1262 		const struct ovs_key_ct_tuple_ipv6 *ct;
1263 
1264 		ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6]);
1265 
1266 		SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.src, &ct->ipv6_src,
1267 				   sizeof(match->key->ipv6.ct_orig.src),
1268 				   is_mask);
1269 		SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.dst, &ct->ipv6_dst,
1270 				   sizeof(match->key->ipv6.ct_orig.dst),
1271 				   is_mask);
1272 		SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1273 		SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1274 		SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv6_proto, is_mask);
1275 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
1276 	}
1277 
1278 	/* For layer 3 packets the Ethernet type is provided
1279 	 * and treated as metadata but no MAC addresses are provided.
1280 	 */
1281 	if (!(*attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1282 	    (*attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)))
1283 		mac_proto = MAC_PROTO_NONE;
1284 
1285 	/* Always exact match mac_proto */
1286 	SW_FLOW_KEY_PUT(match, mac_proto, is_mask ? 0xff : mac_proto, is_mask);
1287 
1288 	if (mac_proto == MAC_PROTO_NONE)
1289 		return parse_eth_type_from_nlattrs(match, attrs, a, is_mask,
1290 						   log);
1291 
1292 	return 0;
1293 }
1294 
1295 int nsh_hdr_from_nlattr(const struct nlattr *attr,
1296 			struct nshhdr *nh, size_t size)
1297 {
1298 	struct nlattr *a;
1299 	int rem;
1300 	u8 flags = 0;
1301 	u8 ttl = 0;
1302 	int mdlen = 0;
1303 
1304 	/* validate_nsh has check this, so we needn't do duplicate check here
1305 	 */
1306 	if (size < NSH_BASE_HDR_LEN)
1307 		return -ENOBUFS;
1308 
1309 	nla_for_each_nested(a, attr, rem) {
1310 		int type = nla_type(a);
1311 
1312 		switch (type) {
1313 		case OVS_NSH_KEY_ATTR_BASE: {
1314 			const struct ovs_nsh_key_base *base = nla_data(a);
1315 
1316 			flags = base->flags;
1317 			ttl = base->ttl;
1318 			nh->np = base->np;
1319 			nh->mdtype = base->mdtype;
1320 			nh->path_hdr = base->path_hdr;
1321 			break;
1322 		}
1323 		case OVS_NSH_KEY_ATTR_MD1:
1324 			mdlen = nla_len(a);
1325 			if (mdlen > size - NSH_BASE_HDR_LEN)
1326 				return -ENOBUFS;
1327 			memcpy(&nh->md1, nla_data(a), mdlen);
1328 			break;
1329 
1330 		case OVS_NSH_KEY_ATTR_MD2:
1331 			mdlen = nla_len(a);
1332 			if (mdlen > size - NSH_BASE_HDR_LEN)
1333 				return -ENOBUFS;
1334 			memcpy(&nh->md2, nla_data(a), mdlen);
1335 			break;
1336 
1337 		default:
1338 			return -EINVAL;
1339 		}
1340 	}
1341 
1342 	/* nsh header length  = NSH_BASE_HDR_LEN + mdlen */
1343 	nh->ver_flags_ttl_len = 0;
1344 	nsh_set_flags_ttl_len(nh, flags, ttl, NSH_BASE_HDR_LEN + mdlen);
1345 
1346 	return 0;
1347 }
1348 
1349 int nsh_key_from_nlattr(const struct nlattr *attr,
1350 			struct ovs_key_nsh *nsh, struct ovs_key_nsh *nsh_mask)
1351 {
1352 	struct nlattr *a;
1353 	int rem;
1354 
1355 	/* validate_nsh has check this, so we needn't do duplicate check here
1356 	 */
1357 	nla_for_each_nested(a, attr, rem) {
1358 		int type = nla_type(a);
1359 
1360 		switch (type) {
1361 		case OVS_NSH_KEY_ATTR_BASE: {
1362 			const struct ovs_nsh_key_base *base = nla_data(a);
1363 			const struct ovs_nsh_key_base *base_mask = base + 1;
1364 
1365 			nsh->base = *base;
1366 			nsh_mask->base = *base_mask;
1367 			break;
1368 		}
1369 		case OVS_NSH_KEY_ATTR_MD1: {
1370 			const struct ovs_nsh_key_md1 *md1 = nla_data(a);
1371 			const struct ovs_nsh_key_md1 *md1_mask = md1 + 1;
1372 
1373 			memcpy(nsh->context, md1->context, sizeof(*md1));
1374 			memcpy(nsh_mask->context, md1_mask->context,
1375 			       sizeof(*md1_mask));
1376 			break;
1377 		}
1378 		case OVS_NSH_KEY_ATTR_MD2:
1379 			/* Not supported yet */
1380 			return -ENOTSUPP;
1381 		default:
1382 			return -EINVAL;
1383 		}
1384 	}
1385 
1386 	return 0;
1387 }
1388 
1389 static int nsh_key_put_from_nlattr(const struct nlattr *attr,
1390 				   struct sw_flow_match *match, bool is_mask,
1391 				   bool is_push_nsh, bool log)
1392 {
1393 	struct nlattr *a;
1394 	int rem;
1395 	bool has_base = false;
1396 	bool has_md1 = false;
1397 	bool has_md2 = false;
1398 	u8 mdtype = 0;
1399 	int mdlen = 0;
1400 
1401 	if (WARN_ON(is_push_nsh && is_mask))
1402 		return -EINVAL;
1403 
1404 	nla_for_each_nested(a, attr, rem) {
1405 		int type = nla_type(a);
1406 		int i;
1407 
1408 		if (type > OVS_NSH_KEY_ATTR_MAX) {
1409 			OVS_NLERR(log, "nsh attr %d is out of range max %d",
1410 				  type, OVS_NSH_KEY_ATTR_MAX);
1411 			return -EINVAL;
1412 		}
1413 
1414 		if (!check_attr_len(nla_len(a),
1415 				    ovs_nsh_key_attr_lens[type].len)) {
1416 			OVS_NLERR(
1417 			    log,
1418 			    "nsh attr %d has unexpected len %d expected %d",
1419 			    type,
1420 			    nla_len(a),
1421 			    ovs_nsh_key_attr_lens[type].len
1422 			);
1423 			return -EINVAL;
1424 		}
1425 
1426 		switch (type) {
1427 		case OVS_NSH_KEY_ATTR_BASE: {
1428 			const struct ovs_nsh_key_base *base = nla_data(a);
1429 
1430 			has_base = true;
1431 			mdtype = base->mdtype;
1432 			SW_FLOW_KEY_PUT(match, nsh.base.flags,
1433 					base->flags, is_mask);
1434 			SW_FLOW_KEY_PUT(match, nsh.base.ttl,
1435 					base->ttl, is_mask);
1436 			SW_FLOW_KEY_PUT(match, nsh.base.mdtype,
1437 					base->mdtype, is_mask);
1438 			SW_FLOW_KEY_PUT(match, nsh.base.np,
1439 					base->np, is_mask);
1440 			SW_FLOW_KEY_PUT(match, nsh.base.path_hdr,
1441 					base->path_hdr, is_mask);
1442 			break;
1443 		}
1444 		case OVS_NSH_KEY_ATTR_MD1: {
1445 			const struct ovs_nsh_key_md1 *md1 = nla_data(a);
1446 
1447 			has_md1 = true;
1448 			for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++)
1449 				SW_FLOW_KEY_PUT(match, nsh.context[i],
1450 						md1->context[i], is_mask);
1451 			break;
1452 		}
1453 		case OVS_NSH_KEY_ATTR_MD2:
1454 			if (!is_push_nsh) /* Not supported MD type 2 yet */
1455 				return -ENOTSUPP;
1456 
1457 			has_md2 = true;
1458 			mdlen = nla_len(a);
1459 			if (mdlen > NSH_CTX_HDRS_MAX_LEN || mdlen <= 0) {
1460 				OVS_NLERR(
1461 				    log,
1462 				    "Invalid MD length %d for MD type %d",
1463 				    mdlen,
1464 				    mdtype
1465 				);
1466 				return -EINVAL;
1467 			}
1468 			break;
1469 		default:
1470 			OVS_NLERR(log, "Unknown nsh attribute %d",
1471 				  type);
1472 			return -EINVAL;
1473 		}
1474 	}
1475 
1476 	if (rem > 0) {
1477 		OVS_NLERR(log, "nsh attribute has %d unknown bytes.", rem);
1478 		return -EINVAL;
1479 	}
1480 
1481 	if (has_md1 && has_md2) {
1482 		OVS_NLERR(
1483 		    1,
1484 		    "invalid nsh attribute: md1 and md2 are exclusive."
1485 		);
1486 		return -EINVAL;
1487 	}
1488 
1489 	if (!is_mask) {
1490 		if ((has_md1 && mdtype != NSH_M_TYPE1) ||
1491 		    (has_md2 && mdtype != NSH_M_TYPE2)) {
1492 			OVS_NLERR(1, "nsh attribute has unmatched MD type %d.",
1493 				  mdtype);
1494 			return -EINVAL;
1495 		}
1496 
1497 		if (is_push_nsh &&
1498 		    (!has_base || (!has_md1 && !has_md2))) {
1499 			OVS_NLERR(
1500 			    1,
1501 			    "push_nsh: missing base or metadata attributes"
1502 			);
1503 			return -EINVAL;
1504 		}
1505 	}
1506 
1507 	return 0;
1508 }
1509 
1510 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
1511 				u64 attrs, const struct nlattr **a,
1512 				bool is_mask, bool log)
1513 {
1514 	int err;
1515 
1516 	err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
1517 	if (err)
1518 		return err;
1519 
1520 	if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
1521 		const struct ovs_key_ethernet *eth_key;
1522 
1523 		eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1524 		SW_FLOW_KEY_MEMCPY(match, eth.src,
1525 				eth_key->eth_src, ETH_ALEN, is_mask);
1526 		SW_FLOW_KEY_MEMCPY(match, eth.dst,
1527 				eth_key->eth_dst, ETH_ALEN, is_mask);
1528 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1529 
1530 		if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
1531 			/* VLAN attribute is always parsed before getting here since it
1532 			 * may occur multiple times.
1533 			 */
1534 			OVS_NLERR(log, "VLAN attribute unexpected.");
1535 			return -EINVAL;
1536 		}
1537 
1538 		if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1539 			err = parse_eth_type_from_nlattrs(match, &attrs, a, is_mask,
1540 							  log);
1541 			if (err)
1542 				return err;
1543 		} else if (!is_mask) {
1544 			SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1545 		}
1546 	} else if (!match->key->eth.type) {
1547 		OVS_NLERR(log, "Either Ethernet header or EtherType is required.");
1548 		return -EINVAL;
1549 	}
1550 
1551 	if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1552 		const struct ovs_key_ipv4 *ipv4_key;
1553 
1554 		ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1555 		if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1556 			OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
1557 				  ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1558 			return -EINVAL;
1559 		}
1560 		SW_FLOW_KEY_PUT(match, ip.proto,
1561 				ipv4_key->ipv4_proto, is_mask);
1562 		SW_FLOW_KEY_PUT(match, ip.tos,
1563 				ipv4_key->ipv4_tos, is_mask);
1564 		SW_FLOW_KEY_PUT(match, ip.ttl,
1565 				ipv4_key->ipv4_ttl, is_mask);
1566 		SW_FLOW_KEY_PUT(match, ip.frag,
1567 				ipv4_key->ipv4_frag, is_mask);
1568 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1569 				ipv4_key->ipv4_src, is_mask);
1570 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1571 				ipv4_key->ipv4_dst, is_mask);
1572 		attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1573 	}
1574 
1575 	if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
1576 		const struct ovs_key_ipv6 *ipv6_key;
1577 
1578 		ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1579 		if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1580 			OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
1581 				  ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1582 			return -EINVAL;
1583 		}
1584 
1585 		if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
1586 			OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x)",
1587 				  ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
1588 			return -EINVAL;
1589 		}
1590 
1591 		SW_FLOW_KEY_PUT(match, ipv6.label,
1592 				ipv6_key->ipv6_label, is_mask);
1593 		SW_FLOW_KEY_PUT(match, ip.proto,
1594 				ipv6_key->ipv6_proto, is_mask);
1595 		SW_FLOW_KEY_PUT(match, ip.tos,
1596 				ipv6_key->ipv6_tclass, is_mask);
1597 		SW_FLOW_KEY_PUT(match, ip.ttl,
1598 				ipv6_key->ipv6_hlimit, is_mask);
1599 		SW_FLOW_KEY_PUT(match, ip.frag,
1600 				ipv6_key->ipv6_frag, is_mask);
1601 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1602 				ipv6_key->ipv6_src,
1603 				sizeof(match->key->ipv6.addr.src),
1604 				is_mask);
1605 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1606 				ipv6_key->ipv6_dst,
1607 				sizeof(match->key->ipv6.addr.dst),
1608 				is_mask);
1609 
1610 		attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1611 	}
1612 
1613 	if (attrs & (1ULL << OVS_KEY_ATTR_IPV6_EXTHDRS)) {
1614 		const struct ovs_key_ipv6_exthdrs *ipv6_exthdrs_key;
1615 
1616 		ipv6_exthdrs_key = nla_data(a[OVS_KEY_ATTR_IPV6_EXTHDRS]);
1617 
1618 		SW_FLOW_KEY_PUT(match, ipv6.exthdrs,
1619 				ipv6_exthdrs_key->hdrs, is_mask);
1620 
1621 		attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6_EXTHDRS);
1622 	}
1623 
1624 	if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1625 		const struct ovs_key_arp *arp_key;
1626 
1627 		arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1628 		if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1629 			OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1630 				  arp_key->arp_op);
1631 			return -EINVAL;
1632 		}
1633 
1634 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1635 				arp_key->arp_sip, is_mask);
1636 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1637 			arp_key->arp_tip, is_mask);
1638 		SW_FLOW_KEY_PUT(match, ip.proto,
1639 				ntohs(arp_key->arp_op), is_mask);
1640 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1641 				arp_key->arp_sha, ETH_ALEN, is_mask);
1642 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1643 				arp_key->arp_tha, ETH_ALEN, is_mask);
1644 
1645 		attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1646 	}
1647 
1648 	if (attrs & (1 << OVS_KEY_ATTR_NSH)) {
1649 		if (nsh_key_put_from_nlattr(a[OVS_KEY_ATTR_NSH], match,
1650 					    is_mask, false, log) < 0)
1651 			return -EINVAL;
1652 		attrs &= ~(1 << OVS_KEY_ATTR_NSH);
1653 	}
1654 
1655 	if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1656 		const struct ovs_key_mpls *mpls_key;
1657 		u32 hdr_len;
1658 		u32 label_count, label_count_mask, i;
1659 
1660 		mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1661 		hdr_len = nla_len(a[OVS_KEY_ATTR_MPLS]);
1662 		label_count = hdr_len / sizeof(struct ovs_key_mpls);
1663 
1664 		if (label_count == 0 || label_count > MPLS_LABEL_DEPTH ||
1665 		    hdr_len % sizeof(struct ovs_key_mpls))
1666 			return -EINVAL;
1667 
1668 		label_count_mask =  GENMASK(label_count - 1, 0);
1669 
1670 		for (i = 0 ; i < label_count; i++)
1671 			SW_FLOW_KEY_PUT(match, mpls.lse[i],
1672 					mpls_key[i].mpls_lse, is_mask);
1673 
1674 		SW_FLOW_KEY_PUT(match, mpls.num_labels_mask,
1675 				label_count_mask, is_mask);
1676 
1677 		attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1678 	 }
1679 
1680 	if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1681 		const struct ovs_key_tcp *tcp_key;
1682 
1683 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1684 		SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1685 		SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1686 		attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1687 	}
1688 
1689 	if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1690 		SW_FLOW_KEY_PUT(match, tp.flags,
1691 				nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1692 				is_mask);
1693 		attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1694 	}
1695 
1696 	if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1697 		const struct ovs_key_udp *udp_key;
1698 
1699 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1700 		SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1701 		SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1702 		attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1703 	}
1704 
1705 	if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1706 		const struct ovs_key_sctp *sctp_key;
1707 
1708 		sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1709 		SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1710 		SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1711 		attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1712 	}
1713 
1714 	if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1715 		const struct ovs_key_icmp *icmp_key;
1716 
1717 		icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1718 		SW_FLOW_KEY_PUT(match, tp.src,
1719 				htons(icmp_key->icmp_type), is_mask);
1720 		SW_FLOW_KEY_PUT(match, tp.dst,
1721 				htons(icmp_key->icmp_code), is_mask);
1722 		attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1723 	}
1724 
1725 	if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1726 		const struct ovs_key_icmpv6 *icmpv6_key;
1727 
1728 		icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1729 		SW_FLOW_KEY_PUT(match, tp.src,
1730 				htons(icmpv6_key->icmpv6_type), is_mask);
1731 		SW_FLOW_KEY_PUT(match, tp.dst,
1732 				htons(icmpv6_key->icmpv6_code), is_mask);
1733 		attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1734 	}
1735 
1736 	if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1737 		const struct ovs_key_nd *nd_key;
1738 
1739 		nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1740 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1741 			nd_key->nd_target,
1742 			sizeof(match->key->ipv6.nd.target),
1743 			is_mask);
1744 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1745 			nd_key->nd_sll, ETH_ALEN, is_mask);
1746 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1747 				nd_key->nd_tll, ETH_ALEN, is_mask);
1748 		attrs &= ~(1 << OVS_KEY_ATTR_ND);
1749 	}
1750 
1751 	if (attrs != 0) {
1752 		OVS_NLERR(log, "Unknown key attributes %llx",
1753 			  (unsigned long long)attrs);
1754 		return -EINVAL;
1755 	}
1756 
1757 	return 0;
1758 }
1759 
1760 static void nlattr_set(struct nlattr *attr, u8 val,
1761 		       const struct ovs_len_tbl *tbl)
1762 {
1763 	struct nlattr *nla;
1764 	int rem;
1765 
1766 	/* The nlattr stream should already have been validated */
1767 	nla_for_each_nested(nla, attr, rem) {
1768 		if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
1769 			nlattr_set(nla, val, tbl[nla_type(nla)].next ? : tbl);
1770 		else
1771 			memset(nla_data(nla), val, nla_len(nla));
1772 
1773 		if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1774 			*(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1775 	}
1776 }
1777 
1778 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1779 {
1780 	nlattr_set(attr, val, ovs_key_lens);
1781 }
1782 
1783 /**
1784  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1785  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1786  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1787  * does not include any don't care bit.
1788  * @net: Used to determine per-namespace field support.
1789  * @match: receives the extracted flow match information.
1790  * @nla_key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1791  * sequence. The fields should of the packet that triggered the creation
1792  * of this flow.
1793  * @nla_mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_*
1794  * Netlink attribute specifies the mask field of the wildcarded flow.
1795  * @log: Boolean to allow kernel error logging.  Normally true, but when
1796  * probing for feature compatibility this should be passed in as false to
1797  * suppress unnecessary error logging.
1798  */
1799 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1800 		      const struct nlattr *nla_key,
1801 		      const struct nlattr *nla_mask,
1802 		      bool log)
1803 {
1804 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1805 	struct nlattr *newmask = NULL;
1806 	u64 key_attrs = 0;
1807 	u64 mask_attrs = 0;
1808 	int err;
1809 
1810 	err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1811 	if (err)
1812 		return err;
1813 
1814 	err = parse_vlan_from_nlattrs(match, &key_attrs, a, false, log);
1815 	if (err)
1816 		return err;
1817 
1818 	err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1819 	if (err)
1820 		return err;
1821 
1822 	if (match->mask) {
1823 		if (!nla_mask) {
1824 			/* Create an exact match mask. We need to set to 0xff
1825 			 * all the 'match->mask' fields that have been touched
1826 			 * in 'match->key'. We cannot simply memset
1827 			 * 'match->mask', because padding bytes and fields not
1828 			 * specified in 'match->key' should be left to 0.
1829 			 * Instead, we use a stream of netlink attributes,
1830 			 * copied from 'key' and set to 0xff.
1831 			 * ovs_key_from_nlattrs() will take care of filling
1832 			 * 'match->mask' appropriately.
1833 			 */
1834 			newmask = kmemdup(nla_key,
1835 					  nla_total_size(nla_len(nla_key)),
1836 					  GFP_KERNEL);
1837 			if (!newmask)
1838 				return -ENOMEM;
1839 
1840 			mask_set_nlattr(newmask, 0xff);
1841 
1842 			/* The userspace does not send tunnel attributes that
1843 			 * are 0, but we should not wildcard them nonetheless.
1844 			 */
1845 			if (match->key->tun_proto)
1846 				SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1847 							 0xff, true);
1848 
1849 			nla_mask = newmask;
1850 		}
1851 
1852 		err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1853 		if (err)
1854 			goto free_newmask;
1855 
1856 		/* Always match on tci. */
1857 		SW_FLOW_KEY_PUT(match, eth.vlan.tci, htons(0xffff), true);
1858 		SW_FLOW_KEY_PUT(match, eth.cvlan.tci, htons(0xffff), true);
1859 
1860 		err = parse_vlan_from_nlattrs(match, &mask_attrs, a, true, log);
1861 		if (err)
1862 			goto free_newmask;
1863 
1864 		err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1865 					   log);
1866 		if (err)
1867 			goto free_newmask;
1868 	}
1869 
1870 	if (!match_validate(match, key_attrs, mask_attrs, log))
1871 		err = -EINVAL;
1872 
1873 free_newmask:
1874 	kfree(newmask);
1875 	return err;
1876 }
1877 
1878 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1879 {
1880 	size_t len;
1881 
1882 	if (!attr)
1883 		return 0;
1884 
1885 	len = nla_len(attr);
1886 	if (len < 1 || len > MAX_UFID_LENGTH) {
1887 		OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1888 			  nla_len(attr), MAX_UFID_LENGTH);
1889 		return 0;
1890 	}
1891 
1892 	return len;
1893 }
1894 
1895 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1896  * or false otherwise.
1897  */
1898 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1899 		      bool log)
1900 {
1901 	sfid->ufid_len = get_ufid_len(attr, log);
1902 	if (sfid->ufid_len)
1903 		memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1904 
1905 	return sfid->ufid_len;
1906 }
1907 
1908 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1909 			   const struct sw_flow_key *key, bool log)
1910 {
1911 	struct sw_flow_key *new_key;
1912 
1913 	if (ovs_nla_get_ufid(sfid, ufid, log))
1914 		return 0;
1915 
1916 	/* If UFID was not provided, use unmasked key. */
1917 	new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1918 	if (!new_key)
1919 		return -ENOMEM;
1920 	memcpy(new_key, key, sizeof(*key));
1921 	sfid->unmasked_key = new_key;
1922 
1923 	return 0;
1924 }
1925 
1926 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1927 {
1928 	return attr ? nla_get_u32(attr) : 0;
1929 }
1930 
1931 /**
1932  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1933  * @net: Network namespace.
1934  * @key: Receives extracted in_port, priority, tun_key, skb_mark and conntrack
1935  * metadata.
1936  * @a: Array of netlink attributes holding parsed %OVS_KEY_ATTR_* Netlink
1937  * attributes.
1938  * @attrs: Bit mask for the netlink attributes included in @a.
1939  * @log: Boolean to allow kernel error logging.  Normally true, but when
1940  * probing for feature compatibility this should be passed in as false to
1941  * suppress unnecessary error logging.
1942  *
1943  * This parses a series of Netlink attributes that form a flow key, which must
1944  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1945  * get the metadata, that is, the parts of the flow key that cannot be
1946  * extracted from the packet itself.
1947  *
1948  * This must be called before the packet key fields are filled in 'key'.
1949  */
1950 
1951 int ovs_nla_get_flow_metadata(struct net *net,
1952 			      const struct nlattr *a[OVS_KEY_ATTR_MAX + 1],
1953 			      u64 attrs, struct sw_flow_key *key, bool log)
1954 {
1955 	struct sw_flow_match match;
1956 
1957 	memset(&match, 0, sizeof(match));
1958 	match.key = key;
1959 
1960 	key->ct_state = 0;
1961 	key->ct_zone = 0;
1962 	key->ct_orig_proto = 0;
1963 	memset(&key->ct, 0, sizeof(key->ct));
1964 	memset(&key->ipv4.ct_orig, 0, sizeof(key->ipv4.ct_orig));
1965 	memset(&key->ipv6.ct_orig, 0, sizeof(key->ipv6.ct_orig));
1966 
1967 	key->phy.in_port = DP_MAX_PORTS;
1968 
1969 	return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1970 }
1971 
1972 static int ovs_nla_put_vlan(struct sk_buff *skb, const struct vlan_head *vh,
1973 			    bool is_mask)
1974 {
1975 	__be16 eth_type = !is_mask ? vh->tpid : htons(0xffff);
1976 
1977 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1978 	    nla_put_be16(skb, OVS_KEY_ATTR_VLAN, vh->tci))
1979 		return -EMSGSIZE;
1980 	return 0;
1981 }
1982 
1983 static int nsh_key_to_nlattr(const struct ovs_key_nsh *nsh, bool is_mask,
1984 			     struct sk_buff *skb)
1985 {
1986 	struct nlattr *start;
1987 
1988 	start = nla_nest_start_noflag(skb, OVS_KEY_ATTR_NSH);
1989 	if (!start)
1990 		return -EMSGSIZE;
1991 
1992 	if (nla_put(skb, OVS_NSH_KEY_ATTR_BASE, sizeof(nsh->base), &nsh->base))
1993 		goto nla_put_failure;
1994 
1995 	if (is_mask || nsh->base.mdtype == NSH_M_TYPE1) {
1996 		if (nla_put(skb, OVS_NSH_KEY_ATTR_MD1,
1997 			    sizeof(nsh->context), nsh->context))
1998 			goto nla_put_failure;
1999 	}
2000 
2001 	/* Don't support MD type 2 yet */
2002 
2003 	nla_nest_end(skb, start);
2004 
2005 	return 0;
2006 
2007 nla_put_failure:
2008 	return -EMSGSIZE;
2009 }
2010 
2011 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
2012 			     const struct sw_flow_key *output, bool is_mask,
2013 			     struct sk_buff *skb)
2014 {
2015 	struct ovs_key_ethernet *eth_key;
2016 	struct nlattr *nla;
2017 	struct nlattr *encap = NULL;
2018 	struct nlattr *in_encap = NULL;
2019 
2020 	if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
2021 		goto nla_put_failure;
2022 
2023 	if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
2024 		goto nla_put_failure;
2025 
2026 	if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
2027 		goto nla_put_failure;
2028 
2029 	if ((swkey->tun_proto || is_mask)) {
2030 		const void *opts = NULL;
2031 
2032 		if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
2033 			opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
2034 
2035 		if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
2036 				     swkey->tun_opts_len, swkey->tun_proto, 0))
2037 			goto nla_put_failure;
2038 	}
2039 
2040 	if (swkey->phy.in_port == DP_MAX_PORTS) {
2041 		if (is_mask && (output->phy.in_port == 0xffff))
2042 			if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
2043 				goto nla_put_failure;
2044 	} else {
2045 		u16 upper_u16;
2046 		upper_u16 = !is_mask ? 0 : 0xffff;
2047 
2048 		if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
2049 				(upper_u16 << 16) | output->phy.in_port))
2050 			goto nla_put_failure;
2051 	}
2052 
2053 	if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
2054 		goto nla_put_failure;
2055 
2056 	if (ovs_ct_put_key(swkey, output, skb))
2057 		goto nla_put_failure;
2058 
2059 	if (ovs_key_mac_proto(swkey) == MAC_PROTO_ETHERNET) {
2060 		nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
2061 		if (!nla)
2062 			goto nla_put_failure;
2063 
2064 		eth_key = nla_data(nla);
2065 		ether_addr_copy(eth_key->eth_src, output->eth.src);
2066 		ether_addr_copy(eth_key->eth_dst, output->eth.dst);
2067 
2068 		if (swkey->eth.vlan.tci || eth_type_vlan(swkey->eth.type)) {
2069 			if (ovs_nla_put_vlan(skb, &output->eth.vlan, is_mask))
2070 				goto nla_put_failure;
2071 			encap = nla_nest_start_noflag(skb, OVS_KEY_ATTR_ENCAP);
2072 			if (!swkey->eth.vlan.tci)
2073 				goto unencap;
2074 
2075 			if (swkey->eth.cvlan.tci || eth_type_vlan(swkey->eth.type)) {
2076 				if (ovs_nla_put_vlan(skb, &output->eth.cvlan, is_mask))
2077 					goto nla_put_failure;
2078 				in_encap = nla_nest_start_noflag(skb,
2079 								 OVS_KEY_ATTR_ENCAP);
2080 				if (!swkey->eth.cvlan.tci)
2081 					goto unencap;
2082 			}
2083 		}
2084 
2085 		if (swkey->eth.type == htons(ETH_P_802_2)) {
2086 			/*
2087 			* Ethertype 802.2 is represented in the netlink with omitted
2088 			* OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
2089 			* 0xffff in the mask attribute.  Ethertype can also
2090 			* be wildcarded.
2091 			*/
2092 			if (is_mask && output->eth.type)
2093 				if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
2094 							output->eth.type))
2095 					goto nla_put_failure;
2096 			goto unencap;
2097 		}
2098 	}
2099 
2100 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
2101 		goto nla_put_failure;
2102 
2103 	if (eth_type_vlan(swkey->eth.type)) {
2104 		/* There are 3 VLAN tags, we don't know anything about the rest
2105 		 * of the packet, so truncate here.
2106 		 */
2107 		WARN_ON_ONCE(!(encap && in_encap));
2108 		goto unencap;
2109 	}
2110 
2111 	if (swkey->eth.type == htons(ETH_P_IP)) {
2112 		struct ovs_key_ipv4 *ipv4_key;
2113 
2114 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
2115 		if (!nla)
2116 			goto nla_put_failure;
2117 		ipv4_key = nla_data(nla);
2118 		ipv4_key->ipv4_src = output->ipv4.addr.src;
2119 		ipv4_key->ipv4_dst = output->ipv4.addr.dst;
2120 		ipv4_key->ipv4_proto = output->ip.proto;
2121 		ipv4_key->ipv4_tos = output->ip.tos;
2122 		ipv4_key->ipv4_ttl = output->ip.ttl;
2123 		ipv4_key->ipv4_frag = output->ip.frag;
2124 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
2125 		struct ovs_key_ipv6 *ipv6_key;
2126 		struct ovs_key_ipv6_exthdrs *ipv6_exthdrs_key;
2127 
2128 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
2129 		if (!nla)
2130 			goto nla_put_failure;
2131 		ipv6_key = nla_data(nla);
2132 		memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
2133 				sizeof(ipv6_key->ipv6_src));
2134 		memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
2135 				sizeof(ipv6_key->ipv6_dst));
2136 		ipv6_key->ipv6_label = output->ipv6.label;
2137 		ipv6_key->ipv6_proto = output->ip.proto;
2138 		ipv6_key->ipv6_tclass = output->ip.tos;
2139 		ipv6_key->ipv6_hlimit = output->ip.ttl;
2140 		ipv6_key->ipv6_frag = output->ip.frag;
2141 
2142 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6_EXTHDRS,
2143 				  sizeof(*ipv6_exthdrs_key));
2144 		if (!nla)
2145 			goto nla_put_failure;
2146 		ipv6_exthdrs_key = nla_data(nla);
2147 		ipv6_exthdrs_key->hdrs = output->ipv6.exthdrs;
2148 	} else if (swkey->eth.type == htons(ETH_P_NSH)) {
2149 		if (nsh_key_to_nlattr(&output->nsh, is_mask, skb))
2150 			goto nla_put_failure;
2151 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
2152 		   swkey->eth.type == htons(ETH_P_RARP)) {
2153 		struct ovs_key_arp *arp_key;
2154 
2155 		nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
2156 		if (!nla)
2157 			goto nla_put_failure;
2158 		arp_key = nla_data(nla);
2159 		memset(arp_key, 0, sizeof(struct ovs_key_arp));
2160 		arp_key->arp_sip = output->ipv4.addr.src;
2161 		arp_key->arp_tip = output->ipv4.addr.dst;
2162 		arp_key->arp_op = htons(output->ip.proto);
2163 		ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
2164 		ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
2165 	} else if (eth_p_mpls(swkey->eth.type)) {
2166 		u8 i, num_labels;
2167 		struct ovs_key_mpls *mpls_key;
2168 
2169 		num_labels = hweight_long(output->mpls.num_labels_mask);
2170 		nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS,
2171 				  num_labels * sizeof(*mpls_key));
2172 		if (!nla)
2173 			goto nla_put_failure;
2174 
2175 		mpls_key = nla_data(nla);
2176 		for (i = 0; i < num_labels; i++)
2177 			mpls_key[i].mpls_lse = output->mpls.lse[i];
2178 	}
2179 
2180 	if ((swkey->eth.type == htons(ETH_P_IP) ||
2181 	     swkey->eth.type == htons(ETH_P_IPV6)) &&
2182 	     swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
2183 
2184 		if (swkey->ip.proto == IPPROTO_TCP) {
2185 			struct ovs_key_tcp *tcp_key;
2186 
2187 			nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
2188 			if (!nla)
2189 				goto nla_put_failure;
2190 			tcp_key = nla_data(nla);
2191 			tcp_key->tcp_src = output->tp.src;
2192 			tcp_key->tcp_dst = output->tp.dst;
2193 			if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
2194 					 output->tp.flags))
2195 				goto nla_put_failure;
2196 		} else if (swkey->ip.proto == IPPROTO_UDP) {
2197 			struct ovs_key_udp *udp_key;
2198 
2199 			nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
2200 			if (!nla)
2201 				goto nla_put_failure;
2202 			udp_key = nla_data(nla);
2203 			udp_key->udp_src = output->tp.src;
2204 			udp_key->udp_dst = output->tp.dst;
2205 		} else if (swkey->ip.proto == IPPROTO_SCTP) {
2206 			struct ovs_key_sctp *sctp_key;
2207 
2208 			nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
2209 			if (!nla)
2210 				goto nla_put_failure;
2211 			sctp_key = nla_data(nla);
2212 			sctp_key->sctp_src = output->tp.src;
2213 			sctp_key->sctp_dst = output->tp.dst;
2214 		} else if (swkey->eth.type == htons(ETH_P_IP) &&
2215 			   swkey->ip.proto == IPPROTO_ICMP) {
2216 			struct ovs_key_icmp *icmp_key;
2217 
2218 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
2219 			if (!nla)
2220 				goto nla_put_failure;
2221 			icmp_key = nla_data(nla);
2222 			icmp_key->icmp_type = ntohs(output->tp.src);
2223 			icmp_key->icmp_code = ntohs(output->tp.dst);
2224 		} else if (swkey->eth.type == htons(ETH_P_IPV6) &&
2225 			   swkey->ip.proto == IPPROTO_ICMPV6) {
2226 			struct ovs_key_icmpv6 *icmpv6_key;
2227 
2228 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
2229 						sizeof(*icmpv6_key));
2230 			if (!nla)
2231 				goto nla_put_failure;
2232 			icmpv6_key = nla_data(nla);
2233 			icmpv6_key->icmpv6_type = ntohs(output->tp.src);
2234 			icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
2235 
2236 			if (swkey->tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
2237 			    swkey->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
2238 				struct ovs_key_nd *nd_key;
2239 
2240 				nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
2241 				if (!nla)
2242 					goto nla_put_failure;
2243 				nd_key = nla_data(nla);
2244 				memcpy(nd_key->nd_target, &output->ipv6.nd.target,
2245 							sizeof(nd_key->nd_target));
2246 				ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
2247 				ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
2248 			}
2249 		}
2250 	}
2251 
2252 unencap:
2253 	if (in_encap)
2254 		nla_nest_end(skb, in_encap);
2255 	if (encap)
2256 		nla_nest_end(skb, encap);
2257 
2258 	return 0;
2259 
2260 nla_put_failure:
2261 	return -EMSGSIZE;
2262 }
2263 
2264 int ovs_nla_put_key(const struct sw_flow_key *swkey,
2265 		    const struct sw_flow_key *output, int attr, bool is_mask,
2266 		    struct sk_buff *skb)
2267 {
2268 	int err;
2269 	struct nlattr *nla;
2270 
2271 	nla = nla_nest_start_noflag(skb, attr);
2272 	if (!nla)
2273 		return -EMSGSIZE;
2274 	err = __ovs_nla_put_key(swkey, output, is_mask, skb);
2275 	if (err)
2276 		return err;
2277 	nla_nest_end(skb, nla);
2278 
2279 	return 0;
2280 }
2281 
2282 /* Called with ovs_mutex or RCU read lock. */
2283 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
2284 {
2285 	if (ovs_identifier_is_ufid(&flow->id))
2286 		return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
2287 			       flow->id.ufid);
2288 
2289 	return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
2290 			       OVS_FLOW_ATTR_KEY, false, skb);
2291 }
2292 
2293 /* Called with ovs_mutex or RCU read lock. */
2294 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
2295 {
2296 	return ovs_nla_put_key(&flow->key, &flow->key,
2297 				OVS_FLOW_ATTR_KEY, false, skb);
2298 }
2299 
2300 /* Called with ovs_mutex or RCU read lock. */
2301 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
2302 {
2303 	return ovs_nla_put_key(&flow->key, &flow->mask->key,
2304 				OVS_FLOW_ATTR_MASK, true, skb);
2305 }
2306 
2307 #define MAX_ACTIONS_BUFSIZE	(32 * 1024)
2308 
2309 static struct sw_flow_actions *nla_alloc_flow_actions(int size)
2310 {
2311 	struct sw_flow_actions *sfa;
2312 
2313 	WARN_ON_ONCE(size > MAX_ACTIONS_BUFSIZE);
2314 
2315 	sfa = kmalloc(kmalloc_size_roundup(sizeof(*sfa) + size), GFP_KERNEL);
2316 	if (!sfa)
2317 		return ERR_PTR(-ENOMEM);
2318 
2319 	sfa->actions_len = 0;
2320 	return sfa;
2321 }
2322 
2323 static void ovs_nla_free_nested_actions(const struct nlattr *actions, int len);
2324 
2325 static void ovs_nla_free_check_pkt_len_action(const struct nlattr *action)
2326 {
2327 	const struct nlattr *a;
2328 	int rem;
2329 
2330 	nla_for_each_nested(a, action, rem) {
2331 		switch (nla_type(a)) {
2332 		case OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL:
2333 		case OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER:
2334 			ovs_nla_free_nested_actions(nla_data(a), nla_len(a));
2335 			break;
2336 		}
2337 	}
2338 }
2339 
2340 static void ovs_nla_free_clone_action(const struct nlattr *action)
2341 {
2342 	const struct nlattr *a = nla_data(action);
2343 	int rem = nla_len(action);
2344 
2345 	switch (nla_type(a)) {
2346 	case OVS_CLONE_ATTR_EXEC:
2347 		/* The real list of actions follows this attribute. */
2348 		a = nla_next(a, &rem);
2349 		ovs_nla_free_nested_actions(a, rem);
2350 		break;
2351 	}
2352 }
2353 
2354 static void ovs_nla_free_dec_ttl_action(const struct nlattr *action)
2355 {
2356 	const struct nlattr *a = nla_data(action);
2357 
2358 	switch (nla_type(a)) {
2359 	case OVS_DEC_TTL_ATTR_ACTION:
2360 		ovs_nla_free_nested_actions(nla_data(a), nla_len(a));
2361 		break;
2362 	}
2363 }
2364 
2365 static void ovs_nla_free_sample_action(const struct nlattr *action)
2366 {
2367 	const struct nlattr *a = nla_data(action);
2368 	int rem = nla_len(action);
2369 
2370 	switch (nla_type(a)) {
2371 	case OVS_SAMPLE_ATTR_ARG:
2372 		/* The real list of actions follows this attribute. */
2373 		a = nla_next(a, &rem);
2374 		ovs_nla_free_nested_actions(a, rem);
2375 		break;
2376 	}
2377 }
2378 
2379 static void ovs_nla_free_set_action(const struct nlattr *a)
2380 {
2381 	const struct nlattr *ovs_key = nla_data(a);
2382 	struct ovs_tunnel_info *ovs_tun;
2383 
2384 	switch (nla_type(ovs_key)) {
2385 	case OVS_KEY_ATTR_TUNNEL_INFO:
2386 		ovs_tun = nla_data(ovs_key);
2387 		dst_release((struct dst_entry *)ovs_tun->tun_dst);
2388 		break;
2389 	}
2390 }
2391 
2392 static void ovs_nla_free_nested_actions(const struct nlattr *actions, int len)
2393 {
2394 	const struct nlattr *a;
2395 	int rem;
2396 
2397 	/* Whenever new actions are added, the need to update this
2398 	 * function should be considered.
2399 	 */
2400 	BUILD_BUG_ON(OVS_ACTION_ATTR_MAX != 24);
2401 
2402 	if (!actions)
2403 		return;
2404 
2405 	nla_for_each_attr(a, actions, len, rem) {
2406 		switch (nla_type(a)) {
2407 		case OVS_ACTION_ATTR_CHECK_PKT_LEN:
2408 			ovs_nla_free_check_pkt_len_action(a);
2409 			break;
2410 
2411 		case OVS_ACTION_ATTR_CLONE:
2412 			ovs_nla_free_clone_action(a);
2413 			break;
2414 
2415 		case OVS_ACTION_ATTR_CT:
2416 			ovs_ct_free_action(a);
2417 			break;
2418 
2419 		case OVS_ACTION_ATTR_DEC_TTL:
2420 			ovs_nla_free_dec_ttl_action(a);
2421 			break;
2422 
2423 		case OVS_ACTION_ATTR_SAMPLE:
2424 			ovs_nla_free_sample_action(a);
2425 			break;
2426 
2427 		case OVS_ACTION_ATTR_SET:
2428 			ovs_nla_free_set_action(a);
2429 			break;
2430 		}
2431 	}
2432 }
2433 
2434 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
2435 {
2436 	if (!sf_acts)
2437 		return;
2438 
2439 	ovs_nla_free_nested_actions(sf_acts->actions, sf_acts->actions_len);
2440 	kfree(sf_acts);
2441 }
2442 
2443 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
2444 {
2445 	ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
2446 }
2447 
2448 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
2449  * The caller must hold rcu_read_lock for this to be sensible. */
2450 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
2451 {
2452 	call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
2453 }
2454 
2455 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
2456 				       int attr_len, bool log)
2457 {
2458 
2459 	struct sw_flow_actions *acts;
2460 	int new_acts_size;
2461 	size_t req_size = NLA_ALIGN(attr_len);
2462 	int next_offset = offsetof(struct sw_flow_actions, actions) +
2463 					(*sfa)->actions_len;
2464 
2465 	if (req_size <= (ksize(*sfa) - next_offset))
2466 		goto out;
2467 
2468 	new_acts_size = max(next_offset + req_size, ksize(*sfa) * 2);
2469 
2470 	if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
2471 		if ((next_offset + req_size) > MAX_ACTIONS_BUFSIZE) {
2472 			OVS_NLERR(log, "Flow action size exceeds max %u",
2473 				  MAX_ACTIONS_BUFSIZE);
2474 			return ERR_PTR(-EMSGSIZE);
2475 		}
2476 		new_acts_size = MAX_ACTIONS_BUFSIZE;
2477 	}
2478 
2479 	acts = nla_alloc_flow_actions(new_acts_size);
2480 	if (IS_ERR(acts))
2481 		return (void *)acts;
2482 
2483 	memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
2484 	acts->actions_len = (*sfa)->actions_len;
2485 	acts->orig_len = (*sfa)->orig_len;
2486 	kfree(*sfa);
2487 	*sfa = acts;
2488 
2489 out:
2490 	(*sfa)->actions_len += req_size;
2491 	return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
2492 }
2493 
2494 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
2495 				   int attrtype, void *data, int len, bool log)
2496 {
2497 	struct nlattr *a;
2498 
2499 	a = reserve_sfa_size(sfa, nla_attr_size(len), log);
2500 	if (IS_ERR(a))
2501 		return a;
2502 
2503 	a->nla_type = attrtype;
2504 	a->nla_len = nla_attr_size(len);
2505 
2506 	if (data)
2507 		memcpy(nla_data(a), data, len);
2508 	memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
2509 
2510 	return a;
2511 }
2512 
2513 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
2514 		       int len, bool log)
2515 {
2516 	struct nlattr *a;
2517 
2518 	a = __add_action(sfa, attrtype, data, len, log);
2519 
2520 	return PTR_ERR_OR_ZERO(a);
2521 }
2522 
2523 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
2524 					  int attrtype, bool log)
2525 {
2526 	int used = (*sfa)->actions_len;
2527 	int err;
2528 
2529 	err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
2530 	if (err)
2531 		return err;
2532 
2533 	return used;
2534 }
2535 
2536 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
2537 					 int st_offset)
2538 {
2539 	struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
2540 							       st_offset);
2541 
2542 	a->nla_len = sfa->actions_len - st_offset;
2543 }
2544 
2545 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2546 				  const struct sw_flow_key *key,
2547 				  struct sw_flow_actions **sfa,
2548 				  __be16 eth_type, __be16 vlan_tci,
2549 				  u32 mpls_label_count, bool log,
2550 				  u32 depth);
2551 
2552 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
2553 				    const struct sw_flow_key *key,
2554 				    struct sw_flow_actions **sfa,
2555 				    __be16 eth_type, __be16 vlan_tci,
2556 				    u32 mpls_label_count, bool log, bool last,
2557 				    u32 depth)
2558 {
2559 	const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
2560 	const struct nlattr *probability, *actions;
2561 	const struct nlattr *a;
2562 	int rem, start, err;
2563 	struct sample_arg arg;
2564 
2565 	memset(attrs, 0, sizeof(attrs));
2566 	nla_for_each_nested(a, attr, rem) {
2567 		int type = nla_type(a);
2568 		if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
2569 			return -EINVAL;
2570 		attrs[type] = a;
2571 	}
2572 	if (rem)
2573 		return -EINVAL;
2574 
2575 	probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
2576 	if (!probability || nla_len(probability) != sizeof(u32))
2577 		return -EINVAL;
2578 
2579 	actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
2580 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
2581 		return -EINVAL;
2582 
2583 	/* validation done, copy sample action. */
2584 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
2585 	if (start < 0)
2586 		return start;
2587 
2588 	/* When both skb and flow may be changed, put the sample
2589 	 * into a deferred fifo. On the other hand, if only skb
2590 	 * may be modified, the actions can be executed in place.
2591 	 *
2592 	 * Do this analysis at the flow installation time.
2593 	 * Set 'clone_action->exec' to true if the actions can be
2594 	 * executed without being deferred.
2595 	 *
2596 	 * If the sample is the last action, it can always be excuted
2597 	 * rather than deferred.
2598 	 */
2599 	arg.exec = last || !actions_may_change_flow(actions);
2600 	arg.probability = nla_get_u32(probability);
2601 
2602 	err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_ARG, &arg, sizeof(arg),
2603 				 log);
2604 	if (err)
2605 		return err;
2606 
2607 	err = __ovs_nla_copy_actions(net, actions, key, sfa,
2608 				     eth_type, vlan_tci, mpls_label_count, log,
2609 				     depth + 1);
2610 
2611 	if (err)
2612 		return err;
2613 
2614 	add_nested_action_end(*sfa, start);
2615 
2616 	return 0;
2617 }
2618 
2619 static int validate_and_copy_dec_ttl(struct net *net,
2620 				     const struct nlattr *attr,
2621 				     const struct sw_flow_key *key,
2622 				     struct sw_flow_actions **sfa,
2623 				     __be16 eth_type, __be16 vlan_tci,
2624 				     u32 mpls_label_count, bool log,
2625 				     u32 depth)
2626 {
2627 	const struct nlattr *attrs[OVS_DEC_TTL_ATTR_MAX + 1];
2628 	int start, action_start, err, rem;
2629 	const struct nlattr *a, *actions;
2630 
2631 	memset(attrs, 0, sizeof(attrs));
2632 	nla_for_each_nested(a, attr, rem) {
2633 		int type = nla_type(a);
2634 
2635 		/* Ignore unknown attributes to be future proof. */
2636 		if (type > OVS_DEC_TTL_ATTR_MAX)
2637 			continue;
2638 
2639 		if (!type || attrs[type]) {
2640 			OVS_NLERR(log, "Duplicate or invalid key (type %d).",
2641 				  type);
2642 			return -EINVAL;
2643 		}
2644 
2645 		attrs[type] = a;
2646 	}
2647 
2648 	if (rem) {
2649 		OVS_NLERR(log, "Message has %d unknown bytes.", rem);
2650 		return -EINVAL;
2651 	}
2652 
2653 	actions = attrs[OVS_DEC_TTL_ATTR_ACTION];
2654 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN)) {
2655 		OVS_NLERR(log, "Missing valid actions attribute.");
2656 		return -EINVAL;
2657 	}
2658 
2659 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_DEC_TTL, log);
2660 	if (start < 0)
2661 		return start;
2662 
2663 	action_start = add_nested_action_start(sfa, OVS_DEC_TTL_ATTR_ACTION, log);
2664 	if (action_start < 0)
2665 		return action_start;
2666 
2667 	err = __ovs_nla_copy_actions(net, actions, key, sfa, eth_type,
2668 				     vlan_tci, mpls_label_count, log,
2669 				     depth + 1);
2670 	if (err)
2671 		return err;
2672 
2673 	add_nested_action_end(*sfa, action_start);
2674 	add_nested_action_end(*sfa, start);
2675 	return 0;
2676 }
2677 
2678 static int validate_and_copy_clone(struct net *net,
2679 				   const struct nlattr *attr,
2680 				   const struct sw_flow_key *key,
2681 				   struct sw_flow_actions **sfa,
2682 				   __be16 eth_type, __be16 vlan_tci,
2683 				   u32 mpls_label_count, bool log, bool last,
2684 				   u32 depth)
2685 {
2686 	int start, err;
2687 	u32 exec;
2688 
2689 	if (nla_len(attr) && nla_len(attr) < NLA_HDRLEN)
2690 		return -EINVAL;
2691 
2692 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_CLONE, log);
2693 	if (start < 0)
2694 		return start;
2695 
2696 	exec = last || !actions_may_change_flow(attr);
2697 
2698 	err = ovs_nla_add_action(sfa, OVS_CLONE_ATTR_EXEC, &exec,
2699 				 sizeof(exec), log);
2700 	if (err)
2701 		return err;
2702 
2703 	err = __ovs_nla_copy_actions(net, attr, key, sfa,
2704 				     eth_type, vlan_tci, mpls_label_count, log,
2705 				     depth + 1);
2706 	if (err)
2707 		return err;
2708 
2709 	add_nested_action_end(*sfa, start);
2710 
2711 	return 0;
2712 }
2713 
2714 void ovs_match_init(struct sw_flow_match *match,
2715 		    struct sw_flow_key *key,
2716 		    bool reset_key,
2717 		    struct sw_flow_mask *mask)
2718 {
2719 	memset(match, 0, sizeof(*match));
2720 	match->key = key;
2721 	match->mask = mask;
2722 
2723 	if (reset_key)
2724 		memset(key, 0, sizeof(*key));
2725 
2726 	if (mask) {
2727 		memset(&mask->key, 0, sizeof(mask->key));
2728 		mask->range.start = mask->range.end = 0;
2729 	}
2730 }
2731 
2732 static int validate_geneve_opts(struct sw_flow_key *key)
2733 {
2734 	struct geneve_opt *option;
2735 	int opts_len = key->tun_opts_len;
2736 	bool crit_opt = false;
2737 
2738 	option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
2739 	while (opts_len > 0) {
2740 		int len;
2741 
2742 		if (opts_len < sizeof(*option))
2743 			return -EINVAL;
2744 
2745 		len = sizeof(*option) + option->length * 4;
2746 		if (len > opts_len)
2747 			return -EINVAL;
2748 
2749 		crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
2750 
2751 		option = (struct geneve_opt *)((u8 *)option + len);
2752 		opts_len -= len;
2753 	}
2754 
2755 	key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
2756 
2757 	return 0;
2758 }
2759 
2760 static int validate_and_copy_set_tun(const struct nlattr *attr,
2761 				     struct sw_flow_actions **sfa, bool log)
2762 {
2763 	struct sw_flow_match match;
2764 	struct sw_flow_key key;
2765 	struct metadata_dst *tun_dst;
2766 	struct ip_tunnel_info *tun_info;
2767 	struct ovs_tunnel_info *ovs_tun;
2768 	struct nlattr *a;
2769 	int err = 0, start, opts_type;
2770 	__be16 dst_opt_type;
2771 
2772 	dst_opt_type = 0;
2773 	ovs_match_init(&match, &key, true, NULL);
2774 	opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
2775 	if (opts_type < 0)
2776 		return opts_type;
2777 
2778 	if (key.tun_opts_len) {
2779 		switch (opts_type) {
2780 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
2781 			err = validate_geneve_opts(&key);
2782 			if (err < 0)
2783 				return err;
2784 			dst_opt_type = TUNNEL_GENEVE_OPT;
2785 			break;
2786 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
2787 			dst_opt_type = TUNNEL_VXLAN_OPT;
2788 			break;
2789 		case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
2790 			dst_opt_type = TUNNEL_ERSPAN_OPT;
2791 			break;
2792 		}
2793 	}
2794 
2795 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
2796 	if (start < 0)
2797 		return start;
2798 
2799 	tun_dst = metadata_dst_alloc(key.tun_opts_len, METADATA_IP_TUNNEL,
2800 				     GFP_KERNEL);
2801 
2802 	if (!tun_dst)
2803 		return -ENOMEM;
2804 
2805 	err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
2806 	if (err) {
2807 		dst_release((struct dst_entry *)tun_dst);
2808 		return err;
2809 	}
2810 
2811 	a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
2812 			 sizeof(*ovs_tun), log);
2813 	if (IS_ERR(a)) {
2814 		dst_release((struct dst_entry *)tun_dst);
2815 		return PTR_ERR(a);
2816 	}
2817 
2818 	ovs_tun = nla_data(a);
2819 	ovs_tun->tun_dst = tun_dst;
2820 
2821 	tun_info = &tun_dst->u.tun_info;
2822 	tun_info->mode = IP_TUNNEL_INFO_TX;
2823 	if (key.tun_proto == AF_INET6)
2824 		tun_info->mode |= IP_TUNNEL_INFO_IPV6;
2825 	else if (key.tun_proto == AF_INET && key.tun_key.u.ipv4.dst == 0)
2826 		tun_info->mode |= IP_TUNNEL_INFO_BRIDGE;
2827 	tun_info->key = key.tun_key;
2828 
2829 	/* We need to store the options in the action itself since
2830 	 * everything else will go away after flow setup. We can append
2831 	 * it to tun_info and then point there.
2832 	 */
2833 	ip_tunnel_info_opts_set(tun_info,
2834 				TUN_METADATA_OPTS(&key, key.tun_opts_len),
2835 				key.tun_opts_len, dst_opt_type);
2836 	add_nested_action_end(*sfa, start);
2837 
2838 	return err;
2839 }
2840 
2841 static bool validate_nsh(const struct nlattr *attr, bool is_mask,
2842 			 bool is_push_nsh, bool log)
2843 {
2844 	struct sw_flow_match match;
2845 	struct sw_flow_key key;
2846 	int ret = 0;
2847 
2848 	ovs_match_init(&match, &key, true, NULL);
2849 	ret = nsh_key_put_from_nlattr(attr, &match, is_mask,
2850 				      is_push_nsh, log);
2851 	return !ret;
2852 }
2853 
2854 /* Return false if there are any non-masked bits set.
2855  * Mask follows data immediately, before any netlink padding.
2856  */
2857 static bool validate_masked(u8 *data, int len)
2858 {
2859 	u8 *mask = data + len;
2860 
2861 	while (len--)
2862 		if (*data++ & ~*mask++)
2863 			return false;
2864 
2865 	return true;
2866 }
2867 
2868 static int validate_set(const struct nlattr *a,
2869 			const struct sw_flow_key *flow_key,
2870 			struct sw_flow_actions **sfa, bool *skip_copy,
2871 			u8 mac_proto, __be16 eth_type, bool masked, bool log)
2872 {
2873 	const struct nlattr *ovs_key = nla_data(a);
2874 	int key_type = nla_type(ovs_key);
2875 	size_t key_len;
2876 
2877 	/* There can be only one key in a action */
2878 	if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2879 		return -EINVAL;
2880 
2881 	key_len = nla_len(ovs_key);
2882 	if (masked)
2883 		key_len /= 2;
2884 
2885 	if (key_type > OVS_KEY_ATTR_MAX ||
2886 	    !check_attr_len(key_len, ovs_key_lens[key_type].len))
2887 		return -EINVAL;
2888 
2889 	if (masked && !validate_masked(nla_data(ovs_key), key_len))
2890 		return -EINVAL;
2891 
2892 	switch (key_type) {
2893 	case OVS_KEY_ATTR_PRIORITY:
2894 	case OVS_KEY_ATTR_SKB_MARK:
2895 	case OVS_KEY_ATTR_CT_MARK:
2896 	case OVS_KEY_ATTR_CT_LABELS:
2897 		break;
2898 
2899 	case OVS_KEY_ATTR_ETHERNET:
2900 		if (mac_proto != MAC_PROTO_ETHERNET)
2901 			return -EINVAL;
2902 		break;
2903 
2904 	case OVS_KEY_ATTR_TUNNEL: {
2905 		int err;
2906 
2907 		if (masked)
2908 			return -EINVAL; /* Masked tunnel set not supported. */
2909 
2910 		*skip_copy = true;
2911 		err = validate_and_copy_set_tun(a, sfa, log);
2912 		if (err)
2913 			return err;
2914 		break;
2915 	}
2916 	case OVS_KEY_ATTR_IPV4: {
2917 		const struct ovs_key_ipv4 *ipv4_key;
2918 
2919 		if (eth_type != htons(ETH_P_IP))
2920 			return -EINVAL;
2921 
2922 		ipv4_key = nla_data(ovs_key);
2923 
2924 		if (masked) {
2925 			const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2926 
2927 			/* Non-writeable fields. */
2928 			if (mask->ipv4_proto || mask->ipv4_frag)
2929 				return -EINVAL;
2930 		} else {
2931 			if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2932 				return -EINVAL;
2933 
2934 			if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2935 				return -EINVAL;
2936 		}
2937 		break;
2938 	}
2939 	case OVS_KEY_ATTR_IPV6: {
2940 		const struct ovs_key_ipv6 *ipv6_key;
2941 
2942 		if (eth_type != htons(ETH_P_IPV6))
2943 			return -EINVAL;
2944 
2945 		ipv6_key = nla_data(ovs_key);
2946 
2947 		if (masked) {
2948 			const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2949 
2950 			/* Non-writeable fields. */
2951 			if (mask->ipv6_proto || mask->ipv6_frag)
2952 				return -EINVAL;
2953 
2954 			/* Invalid bits in the flow label mask? */
2955 			if (ntohl(mask->ipv6_label) & 0xFFF00000)
2956 				return -EINVAL;
2957 		} else {
2958 			if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2959 				return -EINVAL;
2960 
2961 			if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2962 				return -EINVAL;
2963 		}
2964 		if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2965 			return -EINVAL;
2966 
2967 		break;
2968 	}
2969 	case OVS_KEY_ATTR_TCP:
2970 		if ((eth_type != htons(ETH_P_IP) &&
2971 		     eth_type != htons(ETH_P_IPV6)) ||
2972 		    flow_key->ip.proto != IPPROTO_TCP)
2973 			return -EINVAL;
2974 
2975 		break;
2976 
2977 	case OVS_KEY_ATTR_UDP:
2978 		if ((eth_type != htons(ETH_P_IP) &&
2979 		     eth_type != htons(ETH_P_IPV6)) ||
2980 		    flow_key->ip.proto != IPPROTO_UDP)
2981 			return -EINVAL;
2982 
2983 		break;
2984 
2985 	case OVS_KEY_ATTR_MPLS:
2986 		if (!eth_p_mpls(eth_type))
2987 			return -EINVAL;
2988 		break;
2989 
2990 	case OVS_KEY_ATTR_SCTP:
2991 		if ((eth_type != htons(ETH_P_IP) &&
2992 		     eth_type != htons(ETH_P_IPV6)) ||
2993 		    flow_key->ip.proto != IPPROTO_SCTP)
2994 			return -EINVAL;
2995 
2996 		break;
2997 
2998 	case OVS_KEY_ATTR_NSH:
2999 		if (eth_type != htons(ETH_P_NSH))
3000 			return -EINVAL;
3001 		if (!validate_nsh(nla_data(a), masked, false, log))
3002 			return -EINVAL;
3003 		break;
3004 
3005 	default:
3006 		return -EINVAL;
3007 	}
3008 
3009 	/* Convert non-masked non-tunnel set actions to masked set actions. */
3010 	if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
3011 		int start, len = key_len * 2;
3012 		struct nlattr *at;
3013 
3014 		*skip_copy = true;
3015 
3016 		start = add_nested_action_start(sfa,
3017 						OVS_ACTION_ATTR_SET_TO_MASKED,
3018 						log);
3019 		if (start < 0)
3020 			return start;
3021 
3022 		at = __add_action(sfa, key_type, NULL, len, log);
3023 		if (IS_ERR(at))
3024 			return PTR_ERR(at);
3025 
3026 		memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
3027 		memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
3028 		/* Clear non-writeable bits from otherwise writeable fields. */
3029 		if (key_type == OVS_KEY_ATTR_IPV6) {
3030 			struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
3031 
3032 			mask->ipv6_label &= htonl(0x000FFFFF);
3033 		}
3034 		add_nested_action_end(*sfa, start);
3035 	}
3036 
3037 	return 0;
3038 }
3039 
3040 static int validate_userspace(const struct nlattr *attr)
3041 {
3042 	static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
3043 		[OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
3044 		[OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
3045 		[OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
3046 	};
3047 	struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
3048 	int error;
3049 
3050 	error = nla_parse_nested_deprecated(a, OVS_USERSPACE_ATTR_MAX, attr,
3051 					    userspace_policy, NULL);
3052 	if (error)
3053 		return error;
3054 
3055 	if (!a[OVS_USERSPACE_ATTR_PID] ||
3056 	    !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
3057 		return -EINVAL;
3058 
3059 	return 0;
3060 }
3061 
3062 static const struct nla_policy cpl_policy[OVS_CHECK_PKT_LEN_ATTR_MAX + 1] = {
3063 	[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN] = {.type = NLA_U16 },
3064 	[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER] = {.type = NLA_NESTED },
3065 	[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL] = {.type = NLA_NESTED },
3066 };
3067 
3068 static int validate_and_copy_check_pkt_len(struct net *net,
3069 					   const struct nlattr *attr,
3070 					   const struct sw_flow_key *key,
3071 					   struct sw_flow_actions **sfa,
3072 					   __be16 eth_type, __be16 vlan_tci,
3073 					   u32 mpls_label_count,
3074 					   bool log, bool last, u32 depth)
3075 {
3076 	const struct nlattr *acts_if_greater, *acts_if_lesser_eq;
3077 	struct nlattr *a[OVS_CHECK_PKT_LEN_ATTR_MAX + 1];
3078 	struct check_pkt_len_arg arg;
3079 	int nested_acts_start;
3080 	int start, err;
3081 
3082 	err = nla_parse_deprecated_strict(a, OVS_CHECK_PKT_LEN_ATTR_MAX,
3083 					  nla_data(attr), nla_len(attr),
3084 					  cpl_policy, NULL);
3085 	if (err)
3086 		return err;
3087 
3088 	if (!a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN] ||
3089 	    !nla_get_u16(a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN]))
3090 		return -EINVAL;
3091 
3092 	acts_if_lesser_eq = a[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL];
3093 	acts_if_greater = a[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER];
3094 
3095 	/* Both the nested action should be present. */
3096 	if (!acts_if_greater || !acts_if_lesser_eq)
3097 		return -EINVAL;
3098 
3099 	/* validation done, copy the nested actions. */
3100 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_CHECK_PKT_LEN,
3101 					log);
3102 	if (start < 0)
3103 		return start;
3104 
3105 	arg.pkt_len = nla_get_u16(a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN]);
3106 	arg.exec_for_lesser_equal =
3107 		last || !actions_may_change_flow(acts_if_lesser_eq);
3108 	arg.exec_for_greater =
3109 		last || !actions_may_change_flow(acts_if_greater);
3110 
3111 	err = ovs_nla_add_action(sfa, OVS_CHECK_PKT_LEN_ATTR_ARG, &arg,
3112 				 sizeof(arg), log);
3113 	if (err)
3114 		return err;
3115 
3116 	nested_acts_start = add_nested_action_start(sfa,
3117 		OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL, log);
3118 	if (nested_acts_start < 0)
3119 		return nested_acts_start;
3120 
3121 	err = __ovs_nla_copy_actions(net, acts_if_lesser_eq, key, sfa,
3122 				     eth_type, vlan_tci, mpls_label_count, log,
3123 				     depth + 1);
3124 
3125 	if (err)
3126 		return err;
3127 
3128 	add_nested_action_end(*sfa, nested_acts_start);
3129 
3130 	nested_acts_start = add_nested_action_start(sfa,
3131 		OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER, log);
3132 	if (nested_acts_start < 0)
3133 		return nested_acts_start;
3134 
3135 	err = __ovs_nla_copy_actions(net, acts_if_greater, key, sfa,
3136 				     eth_type, vlan_tci, mpls_label_count, log,
3137 				     depth + 1);
3138 
3139 	if (err)
3140 		return err;
3141 
3142 	add_nested_action_end(*sfa, nested_acts_start);
3143 	add_nested_action_end(*sfa, start);
3144 	return 0;
3145 }
3146 
3147 static int copy_action(const struct nlattr *from,
3148 		       struct sw_flow_actions **sfa, bool log)
3149 {
3150 	int totlen = NLA_ALIGN(from->nla_len);
3151 	struct nlattr *to;
3152 
3153 	to = reserve_sfa_size(sfa, from->nla_len, log);
3154 	if (IS_ERR(to))
3155 		return PTR_ERR(to);
3156 
3157 	memcpy(to, from, totlen);
3158 	return 0;
3159 }
3160 
3161 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
3162 				  const struct sw_flow_key *key,
3163 				  struct sw_flow_actions **sfa,
3164 				  __be16 eth_type, __be16 vlan_tci,
3165 				  u32 mpls_label_count, bool log,
3166 				  u32 depth)
3167 {
3168 	u8 mac_proto = ovs_key_mac_proto(key);
3169 	const struct nlattr *a;
3170 	int rem, err;
3171 
3172 	if (depth > OVS_COPY_ACTIONS_MAX_DEPTH)
3173 		return -EOVERFLOW;
3174 
3175 	nla_for_each_nested(a, attr, rem) {
3176 		/* Expected argument lengths, (u32)-1 for variable length. */
3177 		static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
3178 			[OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
3179 			[OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
3180 			[OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
3181 			[OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
3182 			[OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
3183 			[OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
3184 			[OVS_ACTION_ATTR_POP_VLAN] = 0,
3185 			[OVS_ACTION_ATTR_SET] = (u32)-1,
3186 			[OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
3187 			[OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
3188 			[OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
3189 			[OVS_ACTION_ATTR_CT] = (u32)-1,
3190 			[OVS_ACTION_ATTR_CT_CLEAR] = 0,
3191 			[OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
3192 			[OVS_ACTION_ATTR_PUSH_ETH] = sizeof(struct ovs_action_push_eth),
3193 			[OVS_ACTION_ATTR_POP_ETH] = 0,
3194 			[OVS_ACTION_ATTR_PUSH_NSH] = (u32)-1,
3195 			[OVS_ACTION_ATTR_POP_NSH] = 0,
3196 			[OVS_ACTION_ATTR_METER] = sizeof(u32),
3197 			[OVS_ACTION_ATTR_CLONE] = (u32)-1,
3198 			[OVS_ACTION_ATTR_CHECK_PKT_LEN] = (u32)-1,
3199 			[OVS_ACTION_ATTR_ADD_MPLS] = sizeof(struct ovs_action_add_mpls),
3200 			[OVS_ACTION_ATTR_DEC_TTL] = (u32)-1,
3201 			[OVS_ACTION_ATTR_DROP] = sizeof(u32),
3202 		};
3203 		const struct ovs_action_push_vlan *vlan;
3204 		int type = nla_type(a);
3205 		bool skip_copy;
3206 
3207 		if (type > OVS_ACTION_ATTR_MAX ||
3208 		    (action_lens[type] != nla_len(a) &&
3209 		     action_lens[type] != (u32)-1))
3210 			return -EINVAL;
3211 
3212 		skip_copy = false;
3213 		switch (type) {
3214 		case OVS_ACTION_ATTR_UNSPEC:
3215 			return -EINVAL;
3216 
3217 		case OVS_ACTION_ATTR_USERSPACE:
3218 			err = validate_userspace(a);
3219 			if (err)
3220 				return err;
3221 			break;
3222 
3223 		case OVS_ACTION_ATTR_OUTPUT:
3224 			if (nla_get_u32(a) >= DP_MAX_PORTS)
3225 				return -EINVAL;
3226 			break;
3227 
3228 		case OVS_ACTION_ATTR_TRUNC: {
3229 			const struct ovs_action_trunc *trunc = nla_data(a);
3230 
3231 			if (trunc->max_len < ETH_HLEN)
3232 				return -EINVAL;
3233 			break;
3234 		}
3235 
3236 		case OVS_ACTION_ATTR_HASH: {
3237 			const struct ovs_action_hash *act_hash = nla_data(a);
3238 
3239 			switch (act_hash->hash_alg) {
3240 			case OVS_HASH_ALG_L4:
3241 				fallthrough;
3242 			case OVS_HASH_ALG_SYM_L4:
3243 				break;
3244 			default:
3245 				return  -EINVAL;
3246 			}
3247 
3248 			break;
3249 		}
3250 
3251 		case OVS_ACTION_ATTR_POP_VLAN:
3252 			if (mac_proto != MAC_PROTO_ETHERNET)
3253 				return -EINVAL;
3254 			vlan_tci = htons(0);
3255 			break;
3256 
3257 		case OVS_ACTION_ATTR_PUSH_VLAN:
3258 			if (mac_proto != MAC_PROTO_ETHERNET)
3259 				return -EINVAL;
3260 			vlan = nla_data(a);
3261 			if (!eth_type_vlan(vlan->vlan_tpid))
3262 				return -EINVAL;
3263 			if (!(vlan->vlan_tci & htons(VLAN_CFI_MASK)))
3264 				return -EINVAL;
3265 			vlan_tci = vlan->vlan_tci;
3266 			break;
3267 
3268 		case OVS_ACTION_ATTR_RECIRC:
3269 			break;
3270 
3271 		case OVS_ACTION_ATTR_ADD_MPLS: {
3272 			const struct ovs_action_add_mpls *mpls = nla_data(a);
3273 
3274 			if (!eth_p_mpls(mpls->mpls_ethertype))
3275 				return -EINVAL;
3276 
3277 			if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK) {
3278 				if (vlan_tci & htons(VLAN_CFI_MASK) ||
3279 				    (eth_type != htons(ETH_P_IP) &&
3280 				     eth_type != htons(ETH_P_IPV6) &&
3281 				     eth_type != htons(ETH_P_ARP) &&
3282 				     eth_type != htons(ETH_P_RARP) &&
3283 				     !eth_p_mpls(eth_type)))
3284 					return -EINVAL;
3285 				mpls_label_count++;
3286 			} else {
3287 				if (mac_proto == MAC_PROTO_ETHERNET) {
3288 					mpls_label_count = 1;
3289 					mac_proto = MAC_PROTO_NONE;
3290 				} else {
3291 					mpls_label_count++;
3292 				}
3293 			}
3294 			eth_type = mpls->mpls_ethertype;
3295 			break;
3296 		}
3297 
3298 		case OVS_ACTION_ATTR_PUSH_MPLS: {
3299 			const struct ovs_action_push_mpls *mpls = nla_data(a);
3300 
3301 			if (!eth_p_mpls(mpls->mpls_ethertype))
3302 				return -EINVAL;
3303 			/* Prohibit push MPLS other than to a white list
3304 			 * for packets that have a known tag order.
3305 			 */
3306 			if (vlan_tci & htons(VLAN_CFI_MASK) ||
3307 			    (eth_type != htons(ETH_P_IP) &&
3308 			     eth_type != htons(ETH_P_IPV6) &&
3309 			     eth_type != htons(ETH_P_ARP) &&
3310 			     eth_type != htons(ETH_P_RARP) &&
3311 			     !eth_p_mpls(eth_type)))
3312 				return -EINVAL;
3313 			eth_type = mpls->mpls_ethertype;
3314 			mpls_label_count++;
3315 			break;
3316 		}
3317 
3318 		case OVS_ACTION_ATTR_POP_MPLS: {
3319 			__be16  proto;
3320 			if (vlan_tci & htons(VLAN_CFI_MASK) ||
3321 			    !eth_p_mpls(eth_type))
3322 				return -EINVAL;
3323 
3324 			/* Disallow subsequent L2.5+ set actions and mpls_pop
3325 			 * actions once the last MPLS label in the packet is
3326 			 * popped as there is no check here to ensure that
3327 			 * the new eth type is valid and thus set actions could
3328 			 * write off the end of the packet or otherwise corrupt
3329 			 * it.
3330 			 *
3331 			 * Support for these actions is planned using packet
3332 			 * recirculation.
3333 			 */
3334 			proto = nla_get_be16(a);
3335 
3336 			if (proto == htons(ETH_P_TEB) &&
3337 			    mac_proto != MAC_PROTO_NONE)
3338 				return -EINVAL;
3339 
3340 			mpls_label_count--;
3341 
3342 			if (!eth_p_mpls(proto) || !mpls_label_count)
3343 				eth_type = htons(0);
3344 			else
3345 				eth_type =  proto;
3346 
3347 			break;
3348 		}
3349 
3350 		case OVS_ACTION_ATTR_SET:
3351 			err = validate_set(a, key, sfa,
3352 					   &skip_copy, mac_proto, eth_type,
3353 					   false, log);
3354 			if (err)
3355 				return err;
3356 			break;
3357 
3358 		case OVS_ACTION_ATTR_SET_MASKED:
3359 			err = validate_set(a, key, sfa,
3360 					   &skip_copy, mac_proto, eth_type,
3361 					   true, log);
3362 			if (err)
3363 				return err;
3364 			break;
3365 
3366 		case OVS_ACTION_ATTR_SAMPLE: {
3367 			bool last = nla_is_last(a, rem);
3368 
3369 			err = validate_and_copy_sample(net, a, key, sfa,
3370 						       eth_type, vlan_tci,
3371 						       mpls_label_count,
3372 						       log, last, depth);
3373 			if (err)
3374 				return err;
3375 			skip_copy = true;
3376 			break;
3377 		}
3378 
3379 		case OVS_ACTION_ATTR_CT:
3380 			err = ovs_ct_copy_action(net, a, key, sfa, log);
3381 			if (err)
3382 				return err;
3383 			skip_copy = true;
3384 			break;
3385 
3386 		case OVS_ACTION_ATTR_CT_CLEAR:
3387 			break;
3388 
3389 		case OVS_ACTION_ATTR_PUSH_ETH:
3390 			/* Disallow pushing an Ethernet header if one
3391 			 * is already present */
3392 			if (mac_proto != MAC_PROTO_NONE)
3393 				return -EINVAL;
3394 			mac_proto = MAC_PROTO_ETHERNET;
3395 			break;
3396 
3397 		case OVS_ACTION_ATTR_POP_ETH:
3398 			if (mac_proto != MAC_PROTO_ETHERNET)
3399 				return -EINVAL;
3400 			if (vlan_tci & htons(VLAN_CFI_MASK))
3401 				return -EINVAL;
3402 			mac_proto = MAC_PROTO_NONE;
3403 			break;
3404 
3405 		case OVS_ACTION_ATTR_PUSH_NSH:
3406 			if (mac_proto != MAC_PROTO_ETHERNET) {
3407 				u8 next_proto;
3408 
3409 				next_proto = tun_p_from_eth_p(eth_type);
3410 				if (!next_proto)
3411 					return -EINVAL;
3412 			}
3413 			mac_proto = MAC_PROTO_NONE;
3414 			if (!validate_nsh(nla_data(a), false, true, true))
3415 				return -EINVAL;
3416 			break;
3417 
3418 		case OVS_ACTION_ATTR_POP_NSH: {
3419 			__be16 inner_proto;
3420 
3421 			if (eth_type != htons(ETH_P_NSH))
3422 				return -EINVAL;
3423 			inner_proto = tun_p_to_eth_p(key->nsh.base.np);
3424 			if (!inner_proto)
3425 				return -EINVAL;
3426 			if (key->nsh.base.np == TUN_P_ETHERNET)
3427 				mac_proto = MAC_PROTO_ETHERNET;
3428 			else
3429 				mac_proto = MAC_PROTO_NONE;
3430 			break;
3431 		}
3432 
3433 		case OVS_ACTION_ATTR_METER:
3434 			/* Non-existent meters are simply ignored.  */
3435 			break;
3436 
3437 		case OVS_ACTION_ATTR_CLONE: {
3438 			bool last = nla_is_last(a, rem);
3439 
3440 			err = validate_and_copy_clone(net, a, key, sfa,
3441 						      eth_type, vlan_tci,
3442 						      mpls_label_count,
3443 						      log, last, depth);
3444 			if (err)
3445 				return err;
3446 			skip_copy = true;
3447 			break;
3448 		}
3449 
3450 		case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
3451 			bool last = nla_is_last(a, rem);
3452 
3453 			err = validate_and_copy_check_pkt_len(net, a, key, sfa,
3454 							      eth_type,
3455 							      vlan_tci,
3456 							      mpls_label_count,
3457 							      log, last,
3458 							      depth);
3459 			if (err)
3460 				return err;
3461 			skip_copy = true;
3462 			break;
3463 		}
3464 
3465 		case OVS_ACTION_ATTR_DEC_TTL:
3466 			err = validate_and_copy_dec_ttl(net, a, key, sfa,
3467 							eth_type, vlan_tci,
3468 							mpls_label_count, log,
3469 							depth);
3470 			if (err)
3471 				return err;
3472 			skip_copy = true;
3473 			break;
3474 
3475 		case OVS_ACTION_ATTR_DROP:
3476 			if (!nla_is_last(a, rem))
3477 				return -EINVAL;
3478 			break;
3479 
3480 		default:
3481 			OVS_NLERR(log, "Unknown Action type %d", type);
3482 			return -EINVAL;
3483 		}
3484 		if (!skip_copy) {
3485 			err = copy_action(a, sfa, log);
3486 			if (err)
3487 				return err;
3488 		}
3489 	}
3490 
3491 	if (rem > 0)
3492 		return -EINVAL;
3493 
3494 	return 0;
3495 }
3496 
3497 /* 'key' must be the masked key. */
3498 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
3499 			 const struct sw_flow_key *key,
3500 			 struct sw_flow_actions **sfa, bool log)
3501 {
3502 	int err;
3503 	u32 mpls_label_count = 0;
3504 
3505 	*sfa = nla_alloc_flow_actions(min(nla_len(attr), MAX_ACTIONS_BUFSIZE));
3506 	if (IS_ERR(*sfa))
3507 		return PTR_ERR(*sfa);
3508 
3509 	if (eth_p_mpls(key->eth.type))
3510 		mpls_label_count = hweight_long(key->mpls.num_labels_mask);
3511 
3512 	(*sfa)->orig_len = nla_len(attr);
3513 	err = __ovs_nla_copy_actions(net, attr, key, sfa, key->eth.type,
3514 				     key->eth.vlan.tci, mpls_label_count, log,
3515 				     0);
3516 	if (err)
3517 		ovs_nla_free_flow_actions(*sfa);
3518 
3519 	return err;
3520 }
3521 
3522 static int sample_action_to_attr(const struct nlattr *attr,
3523 				 struct sk_buff *skb)
3524 {
3525 	struct nlattr *start, *ac_start = NULL, *sample_arg;
3526 	int err = 0, rem = nla_len(attr);
3527 	const struct sample_arg *arg;
3528 	struct nlattr *actions;
3529 
3530 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SAMPLE);
3531 	if (!start)
3532 		return -EMSGSIZE;
3533 
3534 	sample_arg = nla_data(attr);
3535 	arg = nla_data(sample_arg);
3536 	actions = nla_next(sample_arg, &rem);
3537 
3538 	if (nla_put_u32(skb, OVS_SAMPLE_ATTR_PROBABILITY, arg->probability)) {
3539 		err = -EMSGSIZE;
3540 		goto out;
3541 	}
3542 
3543 	ac_start = nla_nest_start_noflag(skb, OVS_SAMPLE_ATTR_ACTIONS);
3544 	if (!ac_start) {
3545 		err = -EMSGSIZE;
3546 		goto out;
3547 	}
3548 
3549 	err = ovs_nla_put_actions(actions, rem, skb);
3550 
3551 out:
3552 	if (err) {
3553 		nla_nest_cancel(skb, ac_start);
3554 		nla_nest_cancel(skb, start);
3555 	} else {
3556 		nla_nest_end(skb, ac_start);
3557 		nla_nest_end(skb, start);
3558 	}
3559 
3560 	return err;
3561 }
3562 
3563 static int clone_action_to_attr(const struct nlattr *attr,
3564 				struct sk_buff *skb)
3565 {
3566 	struct nlattr *start;
3567 	int err = 0, rem = nla_len(attr);
3568 
3569 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CLONE);
3570 	if (!start)
3571 		return -EMSGSIZE;
3572 
3573 	/* Skipping the OVS_CLONE_ATTR_EXEC that is always the first attribute. */
3574 	attr = nla_next(nla_data(attr), &rem);
3575 	err = ovs_nla_put_actions(attr, rem, skb);
3576 
3577 	if (err)
3578 		nla_nest_cancel(skb, start);
3579 	else
3580 		nla_nest_end(skb, start);
3581 
3582 	return err;
3583 }
3584 
3585 static int check_pkt_len_action_to_attr(const struct nlattr *attr,
3586 					struct sk_buff *skb)
3587 {
3588 	struct nlattr *start, *ac_start = NULL;
3589 	const struct check_pkt_len_arg *arg;
3590 	const struct nlattr *a, *cpl_arg;
3591 	int err = 0, rem = nla_len(attr);
3592 
3593 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CHECK_PKT_LEN);
3594 	if (!start)
3595 		return -EMSGSIZE;
3596 
3597 	/* The first nested attribute in 'attr' is always
3598 	 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
3599 	 */
3600 	cpl_arg = nla_data(attr);
3601 	arg = nla_data(cpl_arg);
3602 
3603 	if (nla_put_u16(skb, OVS_CHECK_PKT_LEN_ATTR_PKT_LEN, arg->pkt_len)) {
3604 		err = -EMSGSIZE;
3605 		goto out;
3606 	}
3607 
3608 	/* Second nested attribute in 'attr' is always
3609 	 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
3610 	 */
3611 	a = nla_next(cpl_arg, &rem);
3612 	ac_start =  nla_nest_start_noflag(skb,
3613 					  OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL);
3614 	if (!ac_start) {
3615 		err = -EMSGSIZE;
3616 		goto out;
3617 	}
3618 
3619 	err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3620 	if (err) {
3621 		nla_nest_cancel(skb, ac_start);
3622 		goto out;
3623 	} else {
3624 		nla_nest_end(skb, ac_start);
3625 	}
3626 
3627 	/* Third nested attribute in 'attr' is always
3628 	 * OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER.
3629 	 */
3630 	a = nla_next(a, &rem);
3631 	ac_start =  nla_nest_start_noflag(skb,
3632 					  OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER);
3633 	if (!ac_start) {
3634 		err = -EMSGSIZE;
3635 		goto out;
3636 	}
3637 
3638 	err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3639 	if (err) {
3640 		nla_nest_cancel(skb, ac_start);
3641 		goto out;
3642 	} else {
3643 		nla_nest_end(skb, ac_start);
3644 	}
3645 
3646 	nla_nest_end(skb, start);
3647 	return 0;
3648 
3649 out:
3650 	nla_nest_cancel(skb, start);
3651 	return err;
3652 }
3653 
3654 static int dec_ttl_action_to_attr(const struct nlattr *attr,
3655 				  struct sk_buff *skb)
3656 {
3657 	struct nlattr *start, *action_start;
3658 	const struct nlattr *a;
3659 	int err = 0, rem;
3660 
3661 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_DEC_TTL);
3662 	if (!start)
3663 		return -EMSGSIZE;
3664 
3665 	nla_for_each_attr(a, nla_data(attr), nla_len(attr), rem) {
3666 		switch (nla_type(a)) {
3667 		case OVS_DEC_TTL_ATTR_ACTION:
3668 
3669 			action_start = nla_nest_start_noflag(skb, OVS_DEC_TTL_ATTR_ACTION);
3670 			if (!action_start) {
3671 				err = -EMSGSIZE;
3672 				goto out;
3673 			}
3674 
3675 			err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3676 			if (err)
3677 				goto out;
3678 
3679 			nla_nest_end(skb, action_start);
3680 			break;
3681 
3682 		default:
3683 			/* Ignore all other option to be future compatible */
3684 			break;
3685 		}
3686 	}
3687 
3688 	nla_nest_end(skb, start);
3689 	return 0;
3690 
3691 out:
3692 	nla_nest_cancel(skb, start);
3693 	return err;
3694 }
3695 
3696 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
3697 {
3698 	const struct nlattr *ovs_key = nla_data(a);
3699 	int key_type = nla_type(ovs_key);
3700 	struct nlattr *start;
3701 	int err;
3702 
3703 	switch (key_type) {
3704 	case OVS_KEY_ATTR_TUNNEL_INFO: {
3705 		struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
3706 		struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
3707 
3708 		start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SET);
3709 		if (!start)
3710 			return -EMSGSIZE;
3711 
3712 		err =  ip_tun_to_nlattr(skb, &tun_info->key,
3713 					ip_tunnel_info_opts(tun_info),
3714 					tun_info->options_len,
3715 					ip_tunnel_info_af(tun_info), tun_info->mode);
3716 		if (err)
3717 			return err;
3718 		nla_nest_end(skb, start);
3719 		break;
3720 	}
3721 	default:
3722 		if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
3723 			return -EMSGSIZE;
3724 		break;
3725 	}
3726 
3727 	return 0;
3728 }
3729 
3730 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
3731 						struct sk_buff *skb)
3732 {
3733 	const struct nlattr *ovs_key = nla_data(a);
3734 	struct nlattr *nla;
3735 	size_t key_len = nla_len(ovs_key) / 2;
3736 
3737 	/* Revert the conversion we did from a non-masked set action to
3738 	 * masked set action.
3739 	 */
3740 	nla = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SET);
3741 	if (!nla)
3742 		return -EMSGSIZE;
3743 
3744 	if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
3745 		return -EMSGSIZE;
3746 
3747 	nla_nest_end(skb, nla);
3748 	return 0;
3749 }
3750 
3751 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
3752 {
3753 	const struct nlattr *a;
3754 	int rem, err;
3755 
3756 	nla_for_each_attr(a, attr, len, rem) {
3757 		int type = nla_type(a);
3758 
3759 		switch (type) {
3760 		case OVS_ACTION_ATTR_SET:
3761 			err = set_action_to_attr(a, skb);
3762 			if (err)
3763 				return err;
3764 			break;
3765 
3766 		case OVS_ACTION_ATTR_SET_TO_MASKED:
3767 			err = masked_set_action_to_set_action_attr(a, skb);
3768 			if (err)
3769 				return err;
3770 			break;
3771 
3772 		case OVS_ACTION_ATTR_SAMPLE:
3773 			err = sample_action_to_attr(a, skb);
3774 			if (err)
3775 				return err;
3776 			break;
3777 
3778 		case OVS_ACTION_ATTR_CT:
3779 			err = ovs_ct_action_to_attr(nla_data(a), skb);
3780 			if (err)
3781 				return err;
3782 			break;
3783 
3784 		case OVS_ACTION_ATTR_CLONE:
3785 			err = clone_action_to_attr(a, skb);
3786 			if (err)
3787 				return err;
3788 			break;
3789 
3790 		case OVS_ACTION_ATTR_CHECK_PKT_LEN:
3791 			err = check_pkt_len_action_to_attr(a, skb);
3792 			if (err)
3793 				return err;
3794 			break;
3795 
3796 		case OVS_ACTION_ATTR_DEC_TTL:
3797 			err = dec_ttl_action_to_attr(a, skb);
3798 			if (err)
3799 				return err;
3800 			break;
3801 
3802 		default:
3803 			if (nla_put(skb, type, nla_len(a), nla_data(a)))
3804 				return -EMSGSIZE;
3805 			break;
3806 		}
3807 	}
3808 
3809 	return 0;
3810 }
3811