xref: /linux/net/openvswitch/flow_netlink.c (revision f94c1a114ac209977bdf5ca841b98424295ab1f0)
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 		case OVS_ACTION_ATTR_PSAMPLE:
68 			break;
69 
70 		case OVS_ACTION_ATTR_CT:
71 		case OVS_ACTION_ATTR_CT_CLEAR:
72 		case OVS_ACTION_ATTR_HASH:
73 		case OVS_ACTION_ATTR_POP_ETH:
74 		case OVS_ACTION_ATTR_POP_MPLS:
75 		case OVS_ACTION_ATTR_POP_NSH:
76 		case OVS_ACTION_ATTR_POP_VLAN:
77 		case OVS_ACTION_ATTR_PUSH_ETH:
78 		case OVS_ACTION_ATTR_PUSH_MPLS:
79 		case OVS_ACTION_ATTR_PUSH_NSH:
80 		case OVS_ACTION_ATTR_PUSH_VLAN:
81 		case OVS_ACTION_ATTR_SAMPLE:
82 		case OVS_ACTION_ATTR_SET:
83 		case OVS_ACTION_ATTR_SET_MASKED:
84 		case OVS_ACTION_ATTR_METER:
85 		case OVS_ACTION_ATTR_CHECK_PKT_LEN:
86 		case OVS_ACTION_ATTR_ADD_MPLS:
87 		case OVS_ACTION_ATTR_DEC_TTL:
88 		default:
89 			return true;
90 		}
91 	}
92 	return false;
93 }
94 
95 static void update_range(struct sw_flow_match *match,
96 			 size_t offset, size_t size, bool is_mask)
97 {
98 	struct sw_flow_key_range *range;
99 	size_t start = rounddown(offset, sizeof(long));
100 	size_t end = roundup(offset + size, sizeof(long));
101 
102 	if (!is_mask)
103 		range = &match->range;
104 	else
105 		range = &match->mask->range;
106 
107 	if (range->start == range->end) {
108 		range->start = start;
109 		range->end = end;
110 		return;
111 	}
112 
113 	if (range->start > start)
114 		range->start = start;
115 
116 	if (range->end < end)
117 		range->end = end;
118 }
119 
120 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
121 	do { \
122 		update_range(match, offsetof(struct sw_flow_key, field),    \
123 			     sizeof((match)->key->field), is_mask);	    \
124 		if (is_mask)						    \
125 			(match)->mask->key.field = value;		    \
126 		else							    \
127 			(match)->key->field = value;		            \
128 	} while (0)
129 
130 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)	    \
131 	do {								    \
132 		update_range(match, offset, len, is_mask);		    \
133 		if (is_mask)						    \
134 			memcpy((u8 *)&(match)->mask->key + offset, value_p, \
135 			       len);					   \
136 		else							    \
137 			memcpy((u8 *)(match)->key + offset, value_p, len);  \
138 	} while (0)
139 
140 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)		      \
141 	SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
142 				  value_p, len, is_mask)
143 
144 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)		    \
145 	do {								    \
146 		update_range(match, offsetof(struct sw_flow_key, field),    \
147 			     sizeof((match)->key->field), is_mask);	    \
148 		if (is_mask)						    \
149 			memset((u8 *)&(match)->mask->key.field, value,      \
150 			       sizeof((match)->mask->key.field));	    \
151 		else							    \
152 			memset((u8 *)&(match)->key->field, value,           \
153 			       sizeof((match)->key->field));                \
154 	} while (0)
155 
156 #define SW_FLOW_KEY_BITMAP_COPY(match, field, value_p, nbits, is_mask) ({     \
157 	update_range(match, offsetof(struct sw_flow_key, field),	      \
158 		     bitmap_size(nbits), is_mask);			      \
159 	bitmap_copy(is_mask ? (match)->mask->key.field : (match)->key->field, \
160 		    value_p, nbits);					      \
161 })
162 
163 static bool match_validate(const struct sw_flow_match *match,
164 			   u64 key_attrs, u64 mask_attrs, bool log)
165 {
166 	u64 key_expected = 0;
167 	u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
168 
169 	/* The following mask attributes allowed only if they
170 	 * pass the validation tests. */
171 	mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
172 			| (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)
173 			| (1 << OVS_KEY_ATTR_IPV6)
174 			| (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)
175 			| (1 << OVS_KEY_ATTR_TCP)
176 			| (1 << OVS_KEY_ATTR_TCP_FLAGS)
177 			| (1 << OVS_KEY_ATTR_UDP)
178 			| (1 << OVS_KEY_ATTR_SCTP)
179 			| (1 << OVS_KEY_ATTR_ICMP)
180 			| (1 << OVS_KEY_ATTR_ICMPV6)
181 			| (1 << OVS_KEY_ATTR_ARP)
182 			| (1 << OVS_KEY_ATTR_ND)
183 			| (1 << OVS_KEY_ATTR_MPLS)
184 			| (1 << OVS_KEY_ATTR_NSH));
185 
186 	/* Always allowed mask fields. */
187 	mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
188 		       | (1 << OVS_KEY_ATTR_IN_PORT)
189 		       | (1 << OVS_KEY_ATTR_ETHERTYPE));
190 
191 	/* Check key attributes. */
192 	if (match->key->eth.type == htons(ETH_P_ARP)
193 			|| match->key->eth.type == htons(ETH_P_RARP)) {
194 		key_expected |= 1 << OVS_KEY_ATTR_ARP;
195 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
196 			mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
197 	}
198 
199 	if (eth_p_mpls(match->key->eth.type)) {
200 		key_expected |= 1 << OVS_KEY_ATTR_MPLS;
201 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
202 			mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
203 	}
204 
205 	if (match->key->eth.type == htons(ETH_P_IP)) {
206 		key_expected |= 1 << OVS_KEY_ATTR_IPV4;
207 		if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
208 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
209 			mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4;
210 		}
211 
212 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
213 			if (match->key->ip.proto == IPPROTO_UDP) {
214 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
215 				if (match->mask && (match->mask->key.ip.proto == 0xff))
216 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
217 			}
218 
219 			if (match->key->ip.proto == IPPROTO_SCTP) {
220 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
221 				if (match->mask && (match->mask->key.ip.proto == 0xff))
222 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
223 			}
224 
225 			if (match->key->ip.proto == IPPROTO_TCP) {
226 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
227 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
228 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
229 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
230 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
231 				}
232 			}
233 
234 			if (match->key->ip.proto == IPPROTO_ICMP) {
235 				key_expected |= 1 << OVS_KEY_ATTR_ICMP;
236 				if (match->mask && (match->mask->key.ip.proto == 0xff))
237 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
238 			}
239 		}
240 	}
241 
242 	if (match->key->eth.type == htons(ETH_P_IPV6)) {
243 		key_expected |= 1 << OVS_KEY_ATTR_IPV6;
244 		if (match->mask && match->mask->key.eth.type == htons(0xffff)) {
245 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
246 			mask_allowed |= 1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6;
247 		}
248 
249 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
250 			if (match->key->ip.proto == IPPROTO_UDP) {
251 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
252 				if (match->mask && (match->mask->key.ip.proto == 0xff))
253 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
254 			}
255 
256 			if (match->key->ip.proto == IPPROTO_SCTP) {
257 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
258 				if (match->mask && (match->mask->key.ip.proto == 0xff))
259 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
260 			}
261 
262 			if (match->key->ip.proto == IPPROTO_TCP) {
263 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
264 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
265 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
266 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
267 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
268 				}
269 			}
270 
271 			if (match->key->ip.proto == IPPROTO_ICMPV6) {
272 				key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
273 				if (match->mask && (match->mask->key.ip.proto == 0xff))
274 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
275 
276 				if (match->key->tp.src ==
277 						htons(NDISC_NEIGHBOUR_SOLICITATION) ||
278 				    match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
279 					key_expected |= 1 << OVS_KEY_ATTR_ND;
280 					/* Original direction conntrack tuple
281 					 * uses the same space as the ND fields
282 					 * in the key, so both are not allowed
283 					 * at the same time.
284 					 */
285 					mask_allowed &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
286 					if (match->mask && (match->mask->key.tp.src == htons(0xff)))
287 						mask_allowed |= 1 << OVS_KEY_ATTR_ND;
288 				}
289 			}
290 		}
291 	}
292 
293 	if (match->key->eth.type == htons(ETH_P_NSH)) {
294 		key_expected |= 1 << OVS_KEY_ATTR_NSH;
295 		if (match->mask &&
296 		    match->mask->key.eth.type == htons(0xffff)) {
297 			mask_allowed |= 1 << OVS_KEY_ATTR_NSH;
298 		}
299 	}
300 
301 	if ((key_attrs & key_expected) != key_expected) {
302 		/* Key attributes check failed. */
303 		OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
304 			  (unsigned long long)key_attrs,
305 			  (unsigned long long)key_expected);
306 		return false;
307 	}
308 
309 	if ((mask_attrs & mask_allowed) != mask_attrs) {
310 		/* Mask attributes check failed. */
311 		OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
312 			  (unsigned long long)mask_attrs,
313 			  (unsigned long long)mask_allowed);
314 		return false;
315 	}
316 
317 	return true;
318 }
319 
320 size_t ovs_tun_key_attr_size(void)
321 {
322 	/* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
323 	 * updating this function.
324 	 */
325 	return    nla_total_size_64bit(8) /* OVS_TUNNEL_KEY_ATTR_ID */
326 		+ nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
327 		+ nla_total_size(16)   /* OVS_TUNNEL_KEY_ATTR_IPV[46]_DST */
328 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
329 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
330 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
331 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
332 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
333 		+ nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
334 		/* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS and
335 		 * OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS is mutually exclusive with
336 		 * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
337 		 */
338 		+ nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
339 		+ nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
340 }
341 
342 static size_t ovs_nsh_key_attr_size(void)
343 {
344 	/* Whenever adding new OVS_NSH_KEY_ FIELDS, we should consider
345 	 * updating this function.
346 	 */
347 	return  nla_total_size(NSH_BASE_HDR_LEN) /* OVS_NSH_KEY_ATTR_BASE */
348 		/* OVS_NSH_KEY_ATTR_MD1 and OVS_NSH_KEY_ATTR_MD2 are
349 		 * mutually exclusive, so the bigger one can cover
350 		 * the small one.
351 		 */
352 		+ nla_total_size(NSH_CTX_HDRS_MAX_LEN);
353 }
354 
355 size_t ovs_key_attr_size(void)
356 {
357 	/* Whenever adding new OVS_KEY_ FIELDS, we should consider
358 	 * updating this function.
359 	 */
360 	BUILD_BUG_ON(OVS_KEY_ATTR_MAX != 32);
361 
362 	return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
363 		+ nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
364 		  + ovs_tun_key_attr_size()
365 		+ nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
366 		+ nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
367 		+ nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
368 		+ nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
369 		+ nla_total_size(4)   /* OVS_KEY_ATTR_CT_STATE */
370 		+ nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
371 		+ nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
372 		+ nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABELS */
373 		+ nla_total_size(40)  /* OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6 */
374 		+ nla_total_size(0)   /* OVS_KEY_ATTR_NSH */
375 		  + ovs_nsh_key_attr_size()
376 		+ nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
377 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
378 		+ nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
379 		+ nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
380 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
381 		+ nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
382 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
383 		+ nla_total_size(28)  /* OVS_KEY_ATTR_ND */
384 		+ nla_total_size(2);  /* OVS_KEY_ATTR_IPV6_EXTHDRS */
385 }
386 
387 static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
388 	[OVS_VXLAN_EXT_GBP]	    = { .len = sizeof(u32) },
389 };
390 
391 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
392 	[OVS_TUNNEL_KEY_ATTR_ID]	    = { .len = sizeof(u64) },
393 	[OVS_TUNNEL_KEY_ATTR_IPV4_SRC]	    = { .len = sizeof(u32) },
394 	[OVS_TUNNEL_KEY_ATTR_IPV4_DST]	    = { .len = sizeof(u32) },
395 	[OVS_TUNNEL_KEY_ATTR_TOS]	    = { .len = 1 },
396 	[OVS_TUNNEL_KEY_ATTR_TTL]	    = { .len = 1 },
397 	[OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
398 	[OVS_TUNNEL_KEY_ATTR_CSUM]	    = { .len = 0 },
399 	[OVS_TUNNEL_KEY_ATTR_TP_SRC]	    = { .len = sizeof(u16) },
400 	[OVS_TUNNEL_KEY_ATTR_TP_DST]	    = { .len = sizeof(u16) },
401 	[OVS_TUNNEL_KEY_ATTR_OAM]	    = { .len = 0 },
402 	[OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_VARIABLE },
403 	[OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED,
404 						.next = ovs_vxlan_ext_key_lens },
405 	[OVS_TUNNEL_KEY_ATTR_IPV6_SRC]      = { .len = sizeof(struct in6_addr) },
406 	[OVS_TUNNEL_KEY_ATTR_IPV6_DST]      = { .len = sizeof(struct in6_addr) },
407 	[OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS]   = { .len = OVS_ATTR_VARIABLE },
408 	[OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE]   = { .len = 0 },
409 };
410 
411 static const struct ovs_len_tbl
412 ovs_nsh_key_attr_lens[OVS_NSH_KEY_ATTR_MAX + 1] = {
413 	[OVS_NSH_KEY_ATTR_BASE] = { .len = sizeof(struct ovs_nsh_key_base) },
414 	[OVS_NSH_KEY_ATTR_MD1]  = { .len = sizeof(struct ovs_nsh_key_md1) },
415 	[OVS_NSH_KEY_ATTR_MD2]  = { .len = OVS_ATTR_VARIABLE },
416 };
417 
418 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
419 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
420 	[OVS_KEY_ATTR_ENCAP]	 = { .len = OVS_ATTR_NESTED },
421 	[OVS_KEY_ATTR_PRIORITY]	 = { .len = sizeof(u32) },
422 	[OVS_KEY_ATTR_IN_PORT]	 = { .len = sizeof(u32) },
423 	[OVS_KEY_ATTR_SKB_MARK]	 = { .len = sizeof(u32) },
424 	[OVS_KEY_ATTR_ETHERNET]	 = { .len = sizeof(struct ovs_key_ethernet) },
425 	[OVS_KEY_ATTR_VLAN]	 = { .len = sizeof(__be16) },
426 	[OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
427 	[OVS_KEY_ATTR_IPV4]	 = { .len = sizeof(struct ovs_key_ipv4) },
428 	[OVS_KEY_ATTR_IPV6]	 = { .len = sizeof(struct ovs_key_ipv6) },
429 	[OVS_KEY_ATTR_TCP]	 = { .len = sizeof(struct ovs_key_tcp) },
430 	[OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
431 	[OVS_KEY_ATTR_UDP]	 = { .len = sizeof(struct ovs_key_udp) },
432 	[OVS_KEY_ATTR_SCTP]	 = { .len = sizeof(struct ovs_key_sctp) },
433 	[OVS_KEY_ATTR_ICMP]	 = { .len = sizeof(struct ovs_key_icmp) },
434 	[OVS_KEY_ATTR_ICMPV6]	 = { .len = sizeof(struct ovs_key_icmpv6) },
435 	[OVS_KEY_ATTR_ARP]	 = { .len = sizeof(struct ovs_key_arp) },
436 	[OVS_KEY_ATTR_ND]	 = { .len = sizeof(struct ovs_key_nd) },
437 	[OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
438 	[OVS_KEY_ATTR_DP_HASH]	 = { .len = sizeof(u32) },
439 	[OVS_KEY_ATTR_TUNNEL]	 = { .len = OVS_ATTR_NESTED,
440 				     .next = ovs_tunnel_key_lens, },
441 	[OVS_KEY_ATTR_MPLS]	 = { .len = OVS_ATTR_VARIABLE },
442 	[OVS_KEY_ATTR_CT_STATE]	 = { .len = sizeof(u32) },
443 	[OVS_KEY_ATTR_CT_ZONE]	 = { .len = sizeof(u16) },
444 	[OVS_KEY_ATTR_CT_MARK]	 = { .len = sizeof(u32) },
445 	[OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
446 	[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4] = {
447 		.len = sizeof(struct ovs_key_ct_tuple_ipv4) },
448 	[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6] = {
449 		.len = sizeof(struct ovs_key_ct_tuple_ipv6) },
450 	[OVS_KEY_ATTR_NSH]       = { .len = OVS_ATTR_NESTED,
451 				     .next = ovs_nsh_key_attr_lens, },
452 	[OVS_KEY_ATTR_IPV6_EXTHDRS] = {
453 		.len = sizeof(struct ovs_key_ipv6_exthdrs) },
454 };
455 
456 static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
457 {
458 	return expected_len == attr_len ||
459 	       expected_len == OVS_ATTR_NESTED ||
460 	       expected_len == OVS_ATTR_VARIABLE;
461 }
462 
463 static bool is_all_zero(const u8 *fp, size_t size)
464 {
465 	int i;
466 
467 	if (!fp)
468 		return false;
469 
470 	for (i = 0; i < size; i++)
471 		if (fp[i])
472 			return false;
473 
474 	return true;
475 }
476 
477 static int __parse_flow_nlattrs(const struct nlattr *attr,
478 				const struct nlattr *a[],
479 				u64 *attrsp, bool log, bool nz)
480 {
481 	const struct nlattr *nla;
482 	u64 attrs;
483 	int rem;
484 
485 	attrs = *attrsp;
486 	nla_for_each_nested(nla, attr, rem) {
487 		u16 type = nla_type(nla);
488 		int expected_len;
489 
490 		if (type > OVS_KEY_ATTR_MAX) {
491 			OVS_NLERR(log, "Key type %d is out of range max %d",
492 				  type, OVS_KEY_ATTR_MAX);
493 			return -EINVAL;
494 		}
495 
496 		if (type == OVS_KEY_ATTR_PACKET_TYPE ||
497 		    type == OVS_KEY_ATTR_ND_EXTENSIONS ||
498 		    type == OVS_KEY_ATTR_TUNNEL_INFO) {
499 			OVS_NLERR(log, "Key type %d is not supported", type);
500 			return -EINVAL;
501 		}
502 
503 		if (attrs & (1ULL << type)) {
504 			OVS_NLERR(log, "Duplicate key (type %d).", type);
505 			return -EINVAL;
506 		}
507 
508 		expected_len = ovs_key_lens[type].len;
509 		if (!check_attr_len(nla_len(nla), expected_len)) {
510 			OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
511 				  type, nla_len(nla), expected_len);
512 			return -EINVAL;
513 		}
514 
515 		if (!nz || !is_all_zero(nla_data(nla), nla_len(nla))) {
516 			attrs |= 1ULL << type;
517 			a[type] = nla;
518 		}
519 	}
520 	if (rem) {
521 		OVS_NLERR(log, "Message has %d unknown bytes.", rem);
522 		return -EINVAL;
523 	}
524 
525 	*attrsp = attrs;
526 	return 0;
527 }
528 
529 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
530 				   const struct nlattr *a[], u64 *attrsp,
531 				   bool log)
532 {
533 	return __parse_flow_nlattrs(attr, a, attrsp, log, true);
534 }
535 
536 int parse_flow_nlattrs(const struct nlattr *attr, const struct nlattr *a[],
537 		       u64 *attrsp, bool log)
538 {
539 	return __parse_flow_nlattrs(attr, a, attrsp, log, false);
540 }
541 
542 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
543 				     struct sw_flow_match *match, bool is_mask,
544 				     bool log)
545 {
546 	unsigned long opt_key_offset;
547 
548 	if (nla_len(a) > sizeof(match->key->tun_opts)) {
549 		OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
550 			  nla_len(a), sizeof(match->key->tun_opts));
551 		return -EINVAL;
552 	}
553 
554 	if (nla_len(a) % 4 != 0) {
555 		OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
556 			  nla_len(a));
557 		return -EINVAL;
558 	}
559 
560 	/* We need to record the length of the options passed
561 	 * down, otherwise packets with the same format but
562 	 * additional options will be silently matched.
563 	 */
564 	if (!is_mask) {
565 		SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
566 				false);
567 	} else {
568 		/* This is somewhat unusual because it looks at
569 		 * both the key and mask while parsing the
570 		 * attributes (and by extension assumes the key
571 		 * is parsed first). Normally, we would verify
572 		 * that each is the correct length and that the
573 		 * attributes line up in the validate function.
574 		 * However, that is difficult because this is
575 		 * variable length and we won't have the
576 		 * information later.
577 		 */
578 		if (match->key->tun_opts_len != nla_len(a)) {
579 			OVS_NLERR(log, "Geneve option len %d != mask len %d",
580 				  match->key->tun_opts_len, nla_len(a));
581 			return -EINVAL;
582 		}
583 
584 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
585 	}
586 
587 	opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
588 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
589 				  nla_len(a), is_mask);
590 	return 0;
591 }
592 
593 static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
594 				     struct sw_flow_match *match, bool is_mask,
595 				     bool log)
596 {
597 	struct nlattr *a;
598 	int rem;
599 	unsigned long opt_key_offset;
600 	struct vxlan_metadata opts;
601 
602 	BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
603 
604 	memset(&opts, 0, sizeof(opts));
605 	nla_for_each_nested(a, attr, rem) {
606 		int type = nla_type(a);
607 
608 		if (type > OVS_VXLAN_EXT_MAX) {
609 			OVS_NLERR(log, "VXLAN extension %d out of range max %d",
610 				  type, OVS_VXLAN_EXT_MAX);
611 			return -EINVAL;
612 		}
613 
614 		if (!check_attr_len(nla_len(a),
615 				    ovs_vxlan_ext_key_lens[type].len)) {
616 			OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
617 				  type, nla_len(a),
618 				  ovs_vxlan_ext_key_lens[type].len);
619 			return -EINVAL;
620 		}
621 
622 		switch (type) {
623 		case OVS_VXLAN_EXT_GBP:
624 			opts.gbp = nla_get_u32(a);
625 			break;
626 		default:
627 			OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
628 				  type);
629 			return -EINVAL;
630 		}
631 	}
632 	if (rem) {
633 		OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
634 			  rem);
635 		return -EINVAL;
636 	}
637 
638 	if (!is_mask)
639 		SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
640 	else
641 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
642 
643 	opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
644 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
645 				  is_mask);
646 	return 0;
647 }
648 
649 static int erspan_tun_opt_from_nlattr(const struct nlattr *a,
650 				      struct sw_flow_match *match, bool is_mask,
651 				      bool log)
652 {
653 	unsigned long opt_key_offset;
654 
655 	BUILD_BUG_ON(sizeof(struct erspan_metadata) >
656 		     sizeof(match->key->tun_opts));
657 
658 	if (nla_len(a) > sizeof(match->key->tun_opts)) {
659 		OVS_NLERR(log, "ERSPAN option length err (len %d, max %zu).",
660 			  nla_len(a), sizeof(match->key->tun_opts));
661 		return -EINVAL;
662 	}
663 
664 	if (!is_mask)
665 		SW_FLOW_KEY_PUT(match, tun_opts_len,
666 				sizeof(struct erspan_metadata), false);
667 	else
668 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
669 
670 	opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
671 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
672 				  nla_len(a), is_mask);
673 	return 0;
674 }
675 
676 static int ip_tun_from_nlattr(const struct nlattr *attr,
677 			      struct sw_flow_match *match, bool is_mask,
678 			      bool log)
679 {
680 	bool ttl = false, ipv4 = false, ipv6 = false;
681 	IP_TUNNEL_DECLARE_FLAGS(tun_flags) = { };
682 	bool info_bridge_mode = false;
683 	int opts_type = 0;
684 	struct nlattr *a;
685 	int rem;
686 
687 	nla_for_each_nested(a, attr, rem) {
688 		int type = nla_type(a);
689 		int err;
690 
691 		if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
692 			OVS_NLERR(log, "Tunnel attr %d out of range max %d",
693 				  type, OVS_TUNNEL_KEY_ATTR_MAX);
694 			return -EINVAL;
695 		}
696 
697 		if (!check_attr_len(nla_len(a),
698 				    ovs_tunnel_key_lens[type].len)) {
699 			OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
700 				  type, nla_len(a), ovs_tunnel_key_lens[type].len);
701 			return -EINVAL;
702 		}
703 
704 		switch (type) {
705 		case OVS_TUNNEL_KEY_ATTR_ID:
706 			SW_FLOW_KEY_PUT(match, tun_key.tun_id,
707 					nla_get_be64(a), is_mask);
708 			__set_bit(IP_TUNNEL_KEY_BIT, tun_flags);
709 			break;
710 		case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
711 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
712 					nla_get_in_addr(a), is_mask);
713 			ipv4 = true;
714 			break;
715 		case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
716 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
717 					nla_get_in_addr(a), is_mask);
718 			ipv4 = true;
719 			break;
720 		case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
721 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.src,
722 					nla_get_in6_addr(a), is_mask);
723 			ipv6 = true;
724 			break;
725 		case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
726 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
727 					nla_get_in6_addr(a), is_mask);
728 			ipv6 = true;
729 			break;
730 		case OVS_TUNNEL_KEY_ATTR_TOS:
731 			SW_FLOW_KEY_PUT(match, tun_key.tos,
732 					nla_get_u8(a), is_mask);
733 			break;
734 		case OVS_TUNNEL_KEY_ATTR_TTL:
735 			SW_FLOW_KEY_PUT(match, tun_key.ttl,
736 					nla_get_u8(a), is_mask);
737 			ttl = true;
738 			break;
739 		case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
740 			__set_bit(IP_TUNNEL_DONT_FRAGMENT_BIT, tun_flags);
741 			break;
742 		case OVS_TUNNEL_KEY_ATTR_CSUM:
743 			__set_bit(IP_TUNNEL_CSUM_BIT, tun_flags);
744 			break;
745 		case OVS_TUNNEL_KEY_ATTR_TP_SRC:
746 			SW_FLOW_KEY_PUT(match, tun_key.tp_src,
747 					nla_get_be16(a), is_mask);
748 			break;
749 		case OVS_TUNNEL_KEY_ATTR_TP_DST:
750 			SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
751 					nla_get_be16(a), is_mask);
752 			break;
753 		case OVS_TUNNEL_KEY_ATTR_OAM:
754 			__set_bit(IP_TUNNEL_OAM_BIT, tun_flags);
755 			break;
756 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
757 			if (opts_type) {
758 				OVS_NLERR(log, "Multiple metadata blocks provided");
759 				return -EINVAL;
760 			}
761 
762 			err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
763 			if (err)
764 				return err;
765 
766 			__set_bit(IP_TUNNEL_GENEVE_OPT_BIT, tun_flags);
767 			opts_type = type;
768 			break;
769 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
770 			if (opts_type) {
771 				OVS_NLERR(log, "Multiple metadata blocks provided");
772 				return -EINVAL;
773 			}
774 
775 			err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
776 			if (err)
777 				return err;
778 
779 			__set_bit(IP_TUNNEL_VXLAN_OPT_BIT, tun_flags);
780 			opts_type = type;
781 			break;
782 		case OVS_TUNNEL_KEY_ATTR_PAD:
783 			break;
784 		case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
785 			if (opts_type) {
786 				OVS_NLERR(log, "Multiple metadata blocks provided");
787 				return -EINVAL;
788 			}
789 
790 			err = erspan_tun_opt_from_nlattr(a, match, is_mask,
791 							 log);
792 			if (err)
793 				return err;
794 
795 			__set_bit(IP_TUNNEL_ERSPAN_OPT_BIT, tun_flags);
796 			opts_type = type;
797 			break;
798 		case OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE:
799 			info_bridge_mode = true;
800 			ipv4 = true;
801 			break;
802 		default:
803 			OVS_NLERR(log, "Unknown IP tunnel attribute %d",
804 				  type);
805 			return -EINVAL;
806 		}
807 	}
808 
809 	SW_FLOW_KEY_BITMAP_COPY(match, tun_key.tun_flags, tun_flags,
810 				__IP_TUNNEL_FLAG_NUM, is_mask);
811 	if (is_mask)
812 		SW_FLOW_KEY_MEMSET_FIELD(match, tun_proto, 0xff, true);
813 	else
814 		SW_FLOW_KEY_PUT(match, tun_proto, ipv6 ? AF_INET6 : AF_INET,
815 				false);
816 
817 	if (rem > 0) {
818 		OVS_NLERR(log, "IP tunnel attribute has %d unknown bytes.",
819 			  rem);
820 		return -EINVAL;
821 	}
822 
823 	if (ipv4 && ipv6) {
824 		OVS_NLERR(log, "Mixed IPv4 and IPv6 tunnel attributes");
825 		return -EINVAL;
826 	}
827 
828 	if (!is_mask) {
829 		if (!ipv4 && !ipv6) {
830 			OVS_NLERR(log, "IP tunnel dst address not specified");
831 			return -EINVAL;
832 		}
833 		if (ipv4) {
834 			if (info_bridge_mode) {
835 				__clear_bit(IP_TUNNEL_KEY_BIT, tun_flags);
836 
837 				if (match->key->tun_key.u.ipv4.src ||
838 				    match->key->tun_key.u.ipv4.dst ||
839 				    match->key->tun_key.tp_src ||
840 				    match->key->tun_key.tp_dst ||
841 				    match->key->tun_key.ttl ||
842 				    match->key->tun_key.tos ||
843 				    !ip_tunnel_flags_empty(tun_flags)) {
844 					OVS_NLERR(log, "IPv4 tun info is not correct");
845 					return -EINVAL;
846 				}
847 			} else if (!match->key->tun_key.u.ipv4.dst) {
848 				OVS_NLERR(log, "IPv4 tunnel dst address is zero");
849 				return -EINVAL;
850 			}
851 		}
852 		if (ipv6 && ipv6_addr_any(&match->key->tun_key.u.ipv6.dst)) {
853 			OVS_NLERR(log, "IPv6 tunnel dst address is zero");
854 			return -EINVAL;
855 		}
856 
857 		if (!ttl && !info_bridge_mode) {
858 			OVS_NLERR(log, "IP tunnel TTL not specified.");
859 			return -EINVAL;
860 		}
861 	}
862 
863 	return opts_type;
864 }
865 
866 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
867 			       const void *tun_opts, int swkey_tun_opts_len)
868 {
869 	const struct vxlan_metadata *opts = tun_opts;
870 	struct nlattr *nla;
871 
872 	nla = nla_nest_start_noflag(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
873 	if (!nla)
874 		return -EMSGSIZE;
875 
876 	if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
877 		return -EMSGSIZE;
878 
879 	nla_nest_end(skb, nla);
880 	return 0;
881 }
882 
883 static int __ip_tun_to_nlattr(struct sk_buff *skb,
884 			      const struct ip_tunnel_key *output,
885 			      const void *tun_opts, int swkey_tun_opts_len,
886 			      unsigned short tun_proto, u8 mode)
887 {
888 	if (test_bit(IP_TUNNEL_KEY_BIT, output->tun_flags) &&
889 	    nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id,
890 			 OVS_TUNNEL_KEY_ATTR_PAD))
891 		return -EMSGSIZE;
892 
893 	if (mode & IP_TUNNEL_INFO_BRIDGE)
894 		return nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE)
895 		       ? -EMSGSIZE : 0;
896 
897 	switch (tun_proto) {
898 	case AF_INET:
899 		if (output->u.ipv4.src &&
900 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
901 				    output->u.ipv4.src))
902 			return -EMSGSIZE;
903 		if (output->u.ipv4.dst &&
904 		    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
905 				    output->u.ipv4.dst))
906 			return -EMSGSIZE;
907 		break;
908 	case AF_INET6:
909 		if (!ipv6_addr_any(&output->u.ipv6.src) &&
910 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
911 				     &output->u.ipv6.src))
912 			return -EMSGSIZE;
913 		if (!ipv6_addr_any(&output->u.ipv6.dst) &&
914 		    nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
915 				     &output->u.ipv6.dst))
916 			return -EMSGSIZE;
917 		break;
918 	}
919 	if (output->tos &&
920 	    nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
921 		return -EMSGSIZE;
922 	if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
923 		return -EMSGSIZE;
924 	if (test_bit(IP_TUNNEL_DONT_FRAGMENT_BIT, output->tun_flags) &&
925 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
926 		return -EMSGSIZE;
927 	if (test_bit(IP_TUNNEL_CSUM_BIT, output->tun_flags) &&
928 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
929 		return -EMSGSIZE;
930 	if (output->tp_src &&
931 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
932 		return -EMSGSIZE;
933 	if (output->tp_dst &&
934 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
935 		return -EMSGSIZE;
936 	if (test_bit(IP_TUNNEL_OAM_BIT, output->tun_flags) &&
937 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
938 		return -EMSGSIZE;
939 	if (swkey_tun_opts_len) {
940 		if (test_bit(IP_TUNNEL_GENEVE_OPT_BIT, output->tun_flags) &&
941 		    nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
942 			    swkey_tun_opts_len, tun_opts))
943 			return -EMSGSIZE;
944 		else if (test_bit(IP_TUNNEL_VXLAN_OPT_BIT,
945 				  output->tun_flags) &&
946 			 vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
947 			return -EMSGSIZE;
948 		else if (test_bit(IP_TUNNEL_ERSPAN_OPT_BIT,
949 				  output->tun_flags) &&
950 			 nla_put(skb, OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS,
951 				 swkey_tun_opts_len, tun_opts))
952 			return -EMSGSIZE;
953 	}
954 
955 	return 0;
956 }
957 
958 static int ip_tun_to_nlattr(struct sk_buff *skb,
959 			    const struct ip_tunnel_key *output,
960 			    const void *tun_opts, int swkey_tun_opts_len,
961 			    unsigned short tun_proto, u8 mode)
962 {
963 	struct nlattr *nla;
964 	int err;
965 
966 	nla = nla_nest_start_noflag(skb, OVS_KEY_ATTR_TUNNEL);
967 	if (!nla)
968 		return -EMSGSIZE;
969 
970 	err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
971 				 tun_proto, mode);
972 	if (err)
973 		return err;
974 
975 	nla_nest_end(skb, nla);
976 	return 0;
977 }
978 
979 int ovs_nla_put_tunnel_info(struct sk_buff *skb,
980 			    struct ip_tunnel_info *tun_info)
981 {
982 	return __ip_tun_to_nlattr(skb, &tun_info->key,
983 				  ip_tunnel_info_opts(tun_info),
984 				  tun_info->options_len,
985 				  ip_tunnel_info_af(tun_info), tun_info->mode);
986 }
987 
988 static int encode_vlan_from_nlattrs(struct sw_flow_match *match,
989 				    const struct nlattr *a[],
990 				    bool is_mask, bool inner)
991 {
992 	__be16 tci = 0;
993 	__be16 tpid = 0;
994 
995 	if (a[OVS_KEY_ATTR_VLAN])
996 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
997 
998 	if (a[OVS_KEY_ATTR_ETHERTYPE])
999 		tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1000 
1001 	if (likely(!inner)) {
1002 		SW_FLOW_KEY_PUT(match, eth.vlan.tpid, tpid, is_mask);
1003 		SW_FLOW_KEY_PUT(match, eth.vlan.tci, tci, is_mask);
1004 	} else {
1005 		SW_FLOW_KEY_PUT(match, eth.cvlan.tpid, tpid, is_mask);
1006 		SW_FLOW_KEY_PUT(match, eth.cvlan.tci, tci, is_mask);
1007 	}
1008 	return 0;
1009 }
1010 
1011 static int validate_vlan_from_nlattrs(const struct sw_flow_match *match,
1012 				      u64 key_attrs, bool inner,
1013 				      const struct nlattr **a, bool log)
1014 {
1015 	__be16 tci = 0;
1016 
1017 	if (!((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1018 	      (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1019 	       eth_type_vlan(nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE])))) {
1020 		/* Not a VLAN. */
1021 		return 0;
1022 	}
1023 
1024 	if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1025 	      (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1026 		OVS_NLERR(log, "Invalid %s frame", (inner) ? "C-VLAN" : "VLAN");
1027 		return -EINVAL;
1028 	}
1029 
1030 	if (a[OVS_KEY_ATTR_VLAN])
1031 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1032 
1033 	if (!(tci & htons(VLAN_CFI_MASK))) {
1034 		if (tci) {
1035 			OVS_NLERR(log, "%s TCI does not have VLAN_CFI_MASK bit set.",
1036 				  (inner) ? "C-VLAN" : "VLAN");
1037 			return -EINVAL;
1038 		} else if (nla_len(a[OVS_KEY_ATTR_ENCAP])) {
1039 			/* Corner case for truncated VLAN header. */
1040 			OVS_NLERR(log, "Truncated %s header has non-zero encap attribute.",
1041 				  (inner) ? "C-VLAN" : "VLAN");
1042 			return -EINVAL;
1043 		}
1044 	}
1045 
1046 	return 1;
1047 }
1048 
1049 static int validate_vlan_mask_from_nlattrs(const struct sw_flow_match *match,
1050 					   u64 key_attrs, bool inner,
1051 					   const struct nlattr **a, bool log)
1052 {
1053 	__be16 tci = 0;
1054 	__be16 tpid = 0;
1055 	bool encap_valid = !!(match->key->eth.vlan.tci &
1056 			      htons(VLAN_CFI_MASK));
1057 	bool i_encap_valid = !!(match->key->eth.cvlan.tci &
1058 				htons(VLAN_CFI_MASK));
1059 
1060 	if (!(key_attrs & (1 << OVS_KEY_ATTR_ENCAP))) {
1061 		/* Not a VLAN. */
1062 		return 0;
1063 	}
1064 
1065 	if ((!inner && !encap_valid) || (inner && !i_encap_valid)) {
1066 		OVS_NLERR(log, "Encap mask attribute is set for non-%s frame.",
1067 			  (inner) ? "C-VLAN" : "VLAN");
1068 		return -EINVAL;
1069 	}
1070 
1071 	if (a[OVS_KEY_ATTR_VLAN])
1072 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1073 
1074 	if (a[OVS_KEY_ATTR_ETHERTYPE])
1075 		tpid = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1076 
1077 	if (tpid != htons(0xffff)) {
1078 		OVS_NLERR(log, "Must have an exact match on %s TPID (mask=%x).",
1079 			  (inner) ? "C-VLAN" : "VLAN", ntohs(tpid));
1080 		return -EINVAL;
1081 	}
1082 	if (!(tci & htons(VLAN_CFI_MASK))) {
1083 		OVS_NLERR(log, "%s TCI mask does not have exact match for VLAN_CFI_MASK bit.",
1084 			  (inner) ? "C-VLAN" : "VLAN");
1085 		return -EINVAL;
1086 	}
1087 
1088 	return 1;
1089 }
1090 
1091 static int __parse_vlan_from_nlattrs(struct sw_flow_match *match,
1092 				     u64 *key_attrs, bool inner,
1093 				     const struct nlattr **a, bool is_mask,
1094 				     bool log)
1095 {
1096 	int err;
1097 	const struct nlattr *encap;
1098 
1099 	if (!is_mask)
1100 		err = validate_vlan_from_nlattrs(match, *key_attrs, inner,
1101 						 a, log);
1102 	else
1103 		err = validate_vlan_mask_from_nlattrs(match, *key_attrs, inner,
1104 						      a, log);
1105 	if (err <= 0)
1106 		return err;
1107 
1108 	err = encode_vlan_from_nlattrs(match, a, is_mask, inner);
1109 	if (err)
1110 		return err;
1111 
1112 	*key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1113 	*key_attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
1114 	*key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1115 
1116 	encap = a[OVS_KEY_ATTR_ENCAP];
1117 
1118 	if (!is_mask)
1119 		err = parse_flow_nlattrs(encap, a, key_attrs, log);
1120 	else
1121 		err = parse_flow_mask_nlattrs(encap, a, key_attrs, log);
1122 
1123 	return err;
1124 }
1125 
1126 static int parse_vlan_from_nlattrs(struct sw_flow_match *match,
1127 				   u64 *key_attrs, const struct nlattr **a,
1128 				   bool is_mask, bool log)
1129 {
1130 	int err;
1131 	bool encap_valid = false;
1132 
1133 	err = __parse_vlan_from_nlattrs(match, key_attrs, false, a,
1134 					is_mask, log);
1135 	if (err)
1136 		return err;
1137 
1138 	encap_valid = !!(match->key->eth.vlan.tci & htons(VLAN_CFI_MASK));
1139 	if (encap_valid) {
1140 		err = __parse_vlan_from_nlattrs(match, key_attrs, true, a,
1141 						is_mask, log);
1142 		if (err)
1143 			return err;
1144 	}
1145 
1146 	return 0;
1147 }
1148 
1149 static int parse_eth_type_from_nlattrs(struct sw_flow_match *match,
1150 				       u64 *attrs, const struct nlattr **a,
1151 				       bool is_mask, bool log)
1152 {
1153 	__be16 eth_type;
1154 
1155 	eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1156 	if (is_mask) {
1157 		/* Always exact match EtherType. */
1158 		eth_type = htons(0xffff);
1159 	} else if (!eth_proto_is_802_3(eth_type)) {
1160 		OVS_NLERR(log, "EtherType %x is less than min %x",
1161 				ntohs(eth_type), ETH_P_802_3_MIN);
1162 		return -EINVAL;
1163 	}
1164 
1165 	SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1166 	*attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1167 	return 0;
1168 }
1169 
1170 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
1171 				 u64 *attrs, const struct nlattr **a,
1172 				 bool is_mask, bool log)
1173 {
1174 	u8 mac_proto = MAC_PROTO_ETHERNET;
1175 
1176 	if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
1177 		u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
1178 
1179 		SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
1180 		*attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
1181 	}
1182 
1183 	if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
1184 		u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
1185 
1186 		SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
1187 		*attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
1188 	}
1189 
1190 	if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1191 		SW_FLOW_KEY_PUT(match, phy.priority,
1192 			  nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1193 		*attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1194 	}
1195 
1196 	if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1197 		u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1198 
1199 		if (is_mask) {
1200 			in_port = 0xffffffff; /* Always exact match in_port. */
1201 		} else if (in_port >= DP_MAX_PORTS) {
1202 			OVS_NLERR(log, "Port %d exceeds max allowable %d",
1203 				  in_port, DP_MAX_PORTS);
1204 			return -EINVAL;
1205 		}
1206 
1207 		SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1208 		*attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1209 	} else if (!is_mask) {
1210 		SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1211 	}
1212 
1213 	if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1214 		uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1215 
1216 		SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1217 		*attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1218 	}
1219 	if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1220 		if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1221 				       is_mask, log) < 0)
1222 			return -EINVAL;
1223 		*attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1224 	}
1225 
1226 	if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
1227 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
1228 		u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
1229 
1230 		if (ct_state & ~CT_SUPPORTED_MASK) {
1231 			OVS_NLERR(log, "ct_state flags %08x unsupported",
1232 				  ct_state);
1233 			return -EINVAL;
1234 		}
1235 
1236 		SW_FLOW_KEY_PUT(match, ct_state, ct_state, is_mask);
1237 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
1238 	}
1239 	if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
1240 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
1241 		u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
1242 
1243 		SW_FLOW_KEY_PUT(match, ct_zone, ct_zone, is_mask);
1244 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
1245 	}
1246 	if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
1247 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
1248 		u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
1249 
1250 		SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
1251 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
1252 	}
1253 	if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
1254 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
1255 		const struct ovs_key_ct_labels *cl;
1256 
1257 		cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
1258 		SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
1259 				   sizeof(*cl), is_mask);
1260 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
1261 	}
1262 	if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4)) {
1263 		const struct ovs_key_ct_tuple_ipv4 *ct;
1264 
1265 		ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4]);
1266 
1267 		SW_FLOW_KEY_PUT(match, ipv4.ct_orig.src, ct->ipv4_src, is_mask);
1268 		SW_FLOW_KEY_PUT(match, ipv4.ct_orig.dst, ct->ipv4_dst, is_mask);
1269 		SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1270 		SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1271 		SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv4_proto, is_mask);
1272 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4);
1273 	}
1274 	if (*attrs & (1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6)) {
1275 		const struct ovs_key_ct_tuple_ipv6 *ct;
1276 
1277 		ct = nla_data(a[OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6]);
1278 
1279 		SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.src, &ct->ipv6_src,
1280 				   sizeof(match->key->ipv6.ct_orig.src),
1281 				   is_mask);
1282 		SW_FLOW_KEY_MEMCPY(match, ipv6.ct_orig.dst, &ct->ipv6_dst,
1283 				   sizeof(match->key->ipv6.ct_orig.dst),
1284 				   is_mask);
1285 		SW_FLOW_KEY_PUT(match, ct.orig_tp.src, ct->src_port, is_mask);
1286 		SW_FLOW_KEY_PUT(match, ct.orig_tp.dst, ct->dst_port, is_mask);
1287 		SW_FLOW_KEY_PUT(match, ct_orig_proto, ct->ipv6_proto, is_mask);
1288 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6);
1289 	}
1290 
1291 	/* For layer 3 packets the Ethernet type is provided
1292 	 * and treated as metadata but no MAC addresses are provided.
1293 	 */
1294 	if (!(*attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1295 	    (*attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)))
1296 		mac_proto = MAC_PROTO_NONE;
1297 
1298 	/* Always exact match mac_proto */
1299 	SW_FLOW_KEY_PUT(match, mac_proto, is_mask ? 0xff : mac_proto, is_mask);
1300 
1301 	if (mac_proto == MAC_PROTO_NONE)
1302 		return parse_eth_type_from_nlattrs(match, attrs, a, is_mask,
1303 						   log);
1304 
1305 	return 0;
1306 }
1307 
1308 /*
1309  * Constructs NSH header 'nh' from attributes of OVS_ACTION_ATTR_PUSH_NSH,
1310  * where 'nh' points to a memory block of 'size' bytes.  It's assumed that
1311  * attributes were previously validated with validate_push_nsh().
1312  */
1313 int nsh_hdr_from_nlattr(const struct nlattr *attr,
1314 			struct nshhdr *nh, size_t size)
1315 {
1316 	struct nlattr *a;
1317 	int rem;
1318 	u8 flags = 0;
1319 	u8 ttl = 0;
1320 	int mdlen = 0;
1321 
1322 	if (size < NSH_BASE_HDR_LEN)
1323 		return -ENOBUFS;
1324 
1325 	nla_for_each_nested(a, attr, rem) {
1326 		int type = nla_type(a);
1327 
1328 		switch (type) {
1329 		case OVS_NSH_KEY_ATTR_BASE: {
1330 			const struct ovs_nsh_key_base *base = nla_data(a);
1331 
1332 			flags = base->flags;
1333 			ttl = base->ttl;
1334 			nh->np = base->np;
1335 			nh->mdtype = base->mdtype;
1336 			nh->path_hdr = base->path_hdr;
1337 			break;
1338 		}
1339 		case OVS_NSH_KEY_ATTR_MD1:
1340 			mdlen = nla_len(a);
1341 			if (mdlen > size - NSH_BASE_HDR_LEN)
1342 				return -ENOBUFS;
1343 			memcpy(&nh->md1, nla_data(a), mdlen);
1344 			break;
1345 
1346 		case OVS_NSH_KEY_ATTR_MD2:
1347 			mdlen = nla_len(a);
1348 			if (mdlen > size - NSH_BASE_HDR_LEN)
1349 				return -ENOBUFS;
1350 			memcpy(&nh->md2, nla_data(a), mdlen);
1351 			break;
1352 
1353 		default:
1354 			return -EINVAL;
1355 		}
1356 	}
1357 
1358 	/* nsh header length  = NSH_BASE_HDR_LEN + mdlen */
1359 	nh->ver_flags_ttl_len = 0;
1360 	nsh_set_flags_ttl_len(nh, flags, ttl, NSH_BASE_HDR_LEN + mdlen);
1361 
1362 	return 0;
1363 }
1364 
1365 static int nsh_key_put_from_nlattr(const struct nlattr *attr,
1366 				   struct sw_flow_match *match, bool is_mask,
1367 				   bool is_push_nsh, bool log)
1368 {
1369 	struct nlattr *a;
1370 	int rem;
1371 	bool has_base = false;
1372 	bool has_md1 = false;
1373 	bool has_md2 = false;
1374 	u8 mdtype = 0;
1375 	int mdlen = 0;
1376 
1377 	if (WARN_ON(is_push_nsh && is_mask))
1378 		return -EINVAL;
1379 
1380 	nla_for_each_nested(a, attr, rem) {
1381 		int type = nla_type(a);
1382 		int i;
1383 
1384 		if (type > OVS_NSH_KEY_ATTR_MAX) {
1385 			OVS_NLERR(log, "nsh attr %d is out of range max %d",
1386 				  type, OVS_NSH_KEY_ATTR_MAX);
1387 			return -EINVAL;
1388 		}
1389 
1390 		if (!check_attr_len(nla_len(a),
1391 				    ovs_nsh_key_attr_lens[type].len)) {
1392 			OVS_NLERR(
1393 			    log,
1394 			    "nsh attr %d has unexpected len %d expected %d",
1395 			    type,
1396 			    nla_len(a),
1397 			    ovs_nsh_key_attr_lens[type].len
1398 			);
1399 			return -EINVAL;
1400 		}
1401 
1402 		switch (type) {
1403 		case OVS_NSH_KEY_ATTR_BASE: {
1404 			const struct ovs_nsh_key_base *base = nla_data(a);
1405 
1406 			has_base = true;
1407 			mdtype = base->mdtype;
1408 			SW_FLOW_KEY_PUT(match, nsh.base.flags,
1409 					base->flags, is_mask);
1410 			SW_FLOW_KEY_PUT(match, nsh.base.ttl,
1411 					base->ttl, is_mask);
1412 			SW_FLOW_KEY_PUT(match, nsh.base.mdtype,
1413 					base->mdtype, is_mask);
1414 			SW_FLOW_KEY_PUT(match, nsh.base.np,
1415 					base->np, is_mask);
1416 			SW_FLOW_KEY_PUT(match, nsh.base.path_hdr,
1417 					base->path_hdr, is_mask);
1418 			break;
1419 		}
1420 		case OVS_NSH_KEY_ATTR_MD1: {
1421 			const struct ovs_nsh_key_md1 *md1 = nla_data(a);
1422 
1423 			has_md1 = true;
1424 			for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++)
1425 				SW_FLOW_KEY_PUT(match, nsh.context[i],
1426 						md1->context[i], is_mask);
1427 			break;
1428 		}
1429 		case OVS_NSH_KEY_ATTR_MD2:
1430 			if (!is_push_nsh) /* Not supported MD type 2 yet */
1431 				return -ENOTSUPP;
1432 
1433 			has_md2 = true;
1434 			mdlen = nla_len(a);
1435 			if (mdlen > NSH_CTX_HDRS_MAX_LEN || mdlen <= 0) {
1436 				OVS_NLERR(
1437 				    log,
1438 				    "Invalid MD length %d for MD type %d",
1439 				    mdlen,
1440 				    mdtype
1441 				);
1442 				return -EINVAL;
1443 			}
1444 			break;
1445 		default:
1446 			OVS_NLERR(log, "Unknown nsh attribute %d",
1447 				  type);
1448 			return -EINVAL;
1449 		}
1450 	}
1451 
1452 	if (rem > 0) {
1453 		OVS_NLERR(log, "nsh attribute has %d unknown bytes.", rem);
1454 		return -EINVAL;
1455 	}
1456 
1457 	if (has_md1 && has_md2) {
1458 		OVS_NLERR(
1459 		    1,
1460 		    "invalid nsh attribute: md1 and md2 are exclusive."
1461 		);
1462 		return -EINVAL;
1463 	}
1464 
1465 	if (!is_mask) {
1466 		if ((has_md1 && mdtype != NSH_M_TYPE1) ||
1467 		    (has_md2 && mdtype != NSH_M_TYPE2)) {
1468 			OVS_NLERR(1, "nsh attribute has unmatched MD type %d.",
1469 				  mdtype);
1470 			return -EINVAL;
1471 		}
1472 
1473 		if (is_push_nsh &&
1474 		    (!has_base || (!has_md1 && !has_md2))) {
1475 			OVS_NLERR(
1476 			    1,
1477 			    "push_nsh: missing base or metadata attributes"
1478 			);
1479 			return -EINVAL;
1480 		}
1481 	}
1482 
1483 	return 0;
1484 }
1485 
1486 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
1487 				u64 attrs, const struct nlattr **a,
1488 				bool is_mask, bool log)
1489 {
1490 	int err;
1491 
1492 	err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
1493 	if (err)
1494 		return err;
1495 
1496 	if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
1497 		const struct ovs_key_ethernet *eth_key;
1498 
1499 		eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1500 		SW_FLOW_KEY_MEMCPY(match, eth.src,
1501 				eth_key->eth_src, ETH_ALEN, is_mask);
1502 		SW_FLOW_KEY_MEMCPY(match, eth.dst,
1503 				eth_key->eth_dst, ETH_ALEN, is_mask);
1504 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1505 
1506 		if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
1507 			/* VLAN attribute is always parsed before getting here since it
1508 			 * may occur multiple times.
1509 			 */
1510 			OVS_NLERR(log, "VLAN attribute unexpected.");
1511 			return -EINVAL;
1512 		}
1513 
1514 		if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1515 			err = parse_eth_type_from_nlattrs(match, &attrs, a, is_mask,
1516 							  log);
1517 			if (err)
1518 				return err;
1519 		} else if (!is_mask) {
1520 			SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1521 		}
1522 	} else if (!match->key->eth.type) {
1523 		OVS_NLERR(log, "Either Ethernet header or EtherType is required.");
1524 		return -EINVAL;
1525 	}
1526 
1527 	if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
1528 		const struct ovs_key_ipv4 *ipv4_key;
1529 
1530 		ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1531 		if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1532 			OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
1533 				  ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1534 			return -EINVAL;
1535 		}
1536 		SW_FLOW_KEY_PUT(match, ip.proto,
1537 				ipv4_key->ipv4_proto, is_mask);
1538 		SW_FLOW_KEY_PUT(match, ip.tos,
1539 				ipv4_key->ipv4_tos, is_mask);
1540 		SW_FLOW_KEY_PUT(match, ip.ttl,
1541 				ipv4_key->ipv4_ttl, is_mask);
1542 		SW_FLOW_KEY_PUT(match, ip.frag,
1543 				ipv4_key->ipv4_frag, is_mask);
1544 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1545 				ipv4_key->ipv4_src, is_mask);
1546 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1547 				ipv4_key->ipv4_dst, is_mask);
1548 		attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1549 	}
1550 
1551 	if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
1552 		const struct ovs_key_ipv6 *ipv6_key;
1553 
1554 		ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1555 		if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1556 			OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
1557 				  ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1558 			return -EINVAL;
1559 		}
1560 
1561 		if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
1562 			OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x)",
1563 				  ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
1564 			return -EINVAL;
1565 		}
1566 
1567 		SW_FLOW_KEY_PUT(match, ipv6.label,
1568 				ipv6_key->ipv6_label, is_mask);
1569 		SW_FLOW_KEY_PUT(match, ip.proto,
1570 				ipv6_key->ipv6_proto, is_mask);
1571 		SW_FLOW_KEY_PUT(match, ip.tos,
1572 				ipv6_key->ipv6_tclass, is_mask);
1573 		SW_FLOW_KEY_PUT(match, ip.ttl,
1574 				ipv6_key->ipv6_hlimit, is_mask);
1575 		SW_FLOW_KEY_PUT(match, ip.frag,
1576 				ipv6_key->ipv6_frag, is_mask);
1577 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1578 				ipv6_key->ipv6_src,
1579 				sizeof(match->key->ipv6.addr.src),
1580 				is_mask);
1581 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1582 				ipv6_key->ipv6_dst,
1583 				sizeof(match->key->ipv6.addr.dst),
1584 				is_mask);
1585 
1586 		attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1587 	}
1588 
1589 	if (attrs & (1ULL << OVS_KEY_ATTR_IPV6_EXTHDRS)) {
1590 		const struct ovs_key_ipv6_exthdrs *ipv6_exthdrs_key;
1591 
1592 		ipv6_exthdrs_key = nla_data(a[OVS_KEY_ATTR_IPV6_EXTHDRS]);
1593 
1594 		SW_FLOW_KEY_PUT(match, ipv6.exthdrs,
1595 				ipv6_exthdrs_key->hdrs, is_mask);
1596 
1597 		attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6_EXTHDRS);
1598 	}
1599 
1600 	if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1601 		const struct ovs_key_arp *arp_key;
1602 
1603 		arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1604 		if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1605 			OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
1606 				  arp_key->arp_op);
1607 			return -EINVAL;
1608 		}
1609 
1610 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1611 				arp_key->arp_sip, is_mask);
1612 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1613 			arp_key->arp_tip, is_mask);
1614 		SW_FLOW_KEY_PUT(match, ip.proto,
1615 				ntohs(arp_key->arp_op), is_mask);
1616 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1617 				arp_key->arp_sha, ETH_ALEN, is_mask);
1618 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1619 				arp_key->arp_tha, ETH_ALEN, is_mask);
1620 
1621 		attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1622 	}
1623 
1624 	if (attrs & (1 << OVS_KEY_ATTR_NSH)) {
1625 		if (nsh_key_put_from_nlattr(a[OVS_KEY_ATTR_NSH], match,
1626 					    is_mask, false, log) < 0)
1627 			return -EINVAL;
1628 		attrs &= ~(1 << OVS_KEY_ATTR_NSH);
1629 	}
1630 
1631 	if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1632 		const struct ovs_key_mpls *mpls_key;
1633 		u32 hdr_len;
1634 		u32 label_count, label_count_mask, i;
1635 
1636 		mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1637 		hdr_len = nla_len(a[OVS_KEY_ATTR_MPLS]);
1638 		label_count = hdr_len / sizeof(struct ovs_key_mpls);
1639 
1640 		if (label_count == 0 || label_count > MPLS_LABEL_DEPTH ||
1641 		    hdr_len % sizeof(struct ovs_key_mpls))
1642 			return -EINVAL;
1643 
1644 		label_count_mask =  GENMASK(label_count - 1, 0);
1645 
1646 		for (i = 0 ; i < label_count; i++)
1647 			SW_FLOW_KEY_PUT(match, mpls.lse[i],
1648 					mpls_key[i].mpls_lse, is_mask);
1649 
1650 		SW_FLOW_KEY_PUT(match, mpls.num_labels_mask,
1651 				label_count_mask, is_mask);
1652 
1653 		attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1654 	 }
1655 
1656 	if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1657 		const struct ovs_key_tcp *tcp_key;
1658 
1659 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1660 		SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1661 		SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
1662 		attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1663 	}
1664 
1665 	if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1666 		SW_FLOW_KEY_PUT(match, tp.flags,
1667 				nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1668 				is_mask);
1669 		attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1670 	}
1671 
1672 	if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1673 		const struct ovs_key_udp *udp_key;
1674 
1675 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1676 		SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1677 		SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
1678 		attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1679 	}
1680 
1681 	if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1682 		const struct ovs_key_sctp *sctp_key;
1683 
1684 		sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1685 		SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1686 		SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
1687 		attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1688 	}
1689 
1690 	if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1691 		const struct ovs_key_icmp *icmp_key;
1692 
1693 		icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1694 		SW_FLOW_KEY_PUT(match, tp.src,
1695 				htons(icmp_key->icmp_type), is_mask);
1696 		SW_FLOW_KEY_PUT(match, tp.dst,
1697 				htons(icmp_key->icmp_code), is_mask);
1698 		attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1699 	}
1700 
1701 	if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1702 		const struct ovs_key_icmpv6 *icmpv6_key;
1703 
1704 		icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1705 		SW_FLOW_KEY_PUT(match, tp.src,
1706 				htons(icmpv6_key->icmpv6_type), is_mask);
1707 		SW_FLOW_KEY_PUT(match, tp.dst,
1708 				htons(icmpv6_key->icmpv6_code), is_mask);
1709 		attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1710 	}
1711 
1712 	if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1713 		const struct ovs_key_nd *nd_key;
1714 
1715 		nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1716 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1717 			nd_key->nd_target,
1718 			sizeof(match->key->ipv6.nd.target),
1719 			is_mask);
1720 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1721 			nd_key->nd_sll, ETH_ALEN, is_mask);
1722 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1723 				nd_key->nd_tll, ETH_ALEN, is_mask);
1724 		attrs &= ~(1 << OVS_KEY_ATTR_ND);
1725 	}
1726 
1727 	if (attrs != 0) {
1728 		OVS_NLERR(log, "Unknown key attributes %llx",
1729 			  (unsigned long long)attrs);
1730 		return -EINVAL;
1731 	}
1732 
1733 	return 0;
1734 }
1735 
1736 static void nlattr_set(struct nlattr *attr, u8 val,
1737 		       const struct ovs_len_tbl *tbl)
1738 {
1739 	struct nlattr *nla;
1740 	int rem;
1741 
1742 	/* The nlattr stream should already have been validated */
1743 	nla_for_each_nested(nla, attr, rem) {
1744 		if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
1745 			nlattr_set(nla, val, tbl[nla_type(nla)].next ? : tbl);
1746 		else
1747 			memset(nla_data(nla), val, nla_len(nla));
1748 
1749 		if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1750 			*(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
1751 	}
1752 }
1753 
1754 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1755 {
1756 	nlattr_set(attr, val, ovs_key_lens);
1757 }
1758 
1759 /**
1760  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1761  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1762  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1763  * does not include any don't care bit.
1764  * @net: Used to determine per-namespace field support.
1765  * @match: receives the extracted flow match information.
1766  * @nla_key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1767  * sequence. The fields should of the packet that triggered the creation
1768  * of this flow.
1769  * @nla_mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_*
1770  * Netlink attribute specifies the mask field of the wildcarded flow.
1771  * @log: Boolean to allow kernel error logging.  Normally true, but when
1772  * probing for feature compatibility this should be passed in as false to
1773  * suppress unnecessary error logging.
1774  */
1775 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1776 		      const struct nlattr *nla_key,
1777 		      const struct nlattr *nla_mask,
1778 		      bool log)
1779 {
1780 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1781 	struct nlattr *newmask = NULL;
1782 	u64 key_attrs = 0;
1783 	u64 mask_attrs = 0;
1784 	int err;
1785 
1786 	err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1787 	if (err)
1788 		return err;
1789 
1790 	err = parse_vlan_from_nlattrs(match, &key_attrs, a, false, log);
1791 	if (err)
1792 		return err;
1793 
1794 	err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1795 	if (err)
1796 		return err;
1797 
1798 	if (match->mask) {
1799 		if (!nla_mask) {
1800 			/* Create an exact match mask. We need to set to 0xff
1801 			 * all the 'match->mask' fields that have been touched
1802 			 * in 'match->key'. We cannot simply memset
1803 			 * 'match->mask', because padding bytes and fields not
1804 			 * specified in 'match->key' should be left to 0.
1805 			 * Instead, we use a stream of netlink attributes,
1806 			 * copied from 'key' and set to 0xff.
1807 			 * ovs_key_from_nlattrs() will take care of filling
1808 			 * 'match->mask' appropriately.
1809 			 */
1810 			newmask = kmemdup(nla_key,
1811 					  nla_total_size(nla_len(nla_key)),
1812 					  GFP_KERNEL);
1813 			if (!newmask)
1814 				return -ENOMEM;
1815 
1816 			mask_set_nlattr(newmask, 0xff);
1817 
1818 			/* The userspace does not send tunnel attributes that
1819 			 * are 0, but we should not wildcard them nonetheless.
1820 			 */
1821 			if (match->key->tun_proto)
1822 				SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1823 							 0xff, true);
1824 
1825 			nla_mask = newmask;
1826 		}
1827 
1828 		err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1829 		if (err)
1830 			goto free_newmask;
1831 
1832 		/* Always match on tci. */
1833 		SW_FLOW_KEY_PUT(match, eth.vlan.tci, htons(0xffff), true);
1834 		SW_FLOW_KEY_PUT(match, eth.cvlan.tci, htons(0xffff), true);
1835 
1836 		err = parse_vlan_from_nlattrs(match, &mask_attrs, a, true, log);
1837 		if (err)
1838 			goto free_newmask;
1839 
1840 		err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1841 					   log);
1842 		if (err)
1843 			goto free_newmask;
1844 	}
1845 
1846 	if (!match_validate(match, key_attrs, mask_attrs, log))
1847 		err = -EINVAL;
1848 
1849 free_newmask:
1850 	kfree(newmask);
1851 	return err;
1852 }
1853 
1854 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1855 {
1856 	size_t len;
1857 
1858 	if (!attr)
1859 		return 0;
1860 
1861 	len = nla_len(attr);
1862 	if (len < 1 || len > MAX_UFID_LENGTH) {
1863 		OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1864 			  nla_len(attr), MAX_UFID_LENGTH);
1865 		return 0;
1866 	}
1867 
1868 	return len;
1869 }
1870 
1871 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1872  * or false otherwise.
1873  */
1874 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1875 		      bool log)
1876 {
1877 	sfid->ufid_len = get_ufid_len(attr, log);
1878 	if (sfid->ufid_len)
1879 		memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1880 
1881 	return sfid->ufid_len;
1882 }
1883 
1884 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1885 			   const struct sw_flow_key *key, bool log)
1886 {
1887 	struct sw_flow_key *new_key;
1888 
1889 	if (ovs_nla_get_ufid(sfid, ufid, log))
1890 		return 0;
1891 
1892 	/* If UFID was not provided, use unmasked key. */
1893 	new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1894 	if (!new_key)
1895 		return -ENOMEM;
1896 	memcpy(new_key, key, sizeof(*key));
1897 	sfid->unmasked_key = new_key;
1898 
1899 	return 0;
1900 }
1901 
1902 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1903 {
1904 	return nla_get_u32_default(attr, 0);
1905 }
1906 
1907 /**
1908  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1909  * @net: Network namespace.
1910  * @key: Receives extracted in_port, priority, tun_key, skb_mark and conntrack
1911  * metadata.
1912  * @a: Array of netlink attributes holding parsed %OVS_KEY_ATTR_* Netlink
1913  * attributes.
1914  * @attrs: Bit mask for the netlink attributes included in @a.
1915  * @log: Boolean to allow kernel error logging.  Normally true, but when
1916  * probing for feature compatibility this should be passed in as false to
1917  * suppress unnecessary error logging.
1918  *
1919  * This parses a series of Netlink attributes that form a flow key, which must
1920  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1921  * get the metadata, that is, the parts of the flow key that cannot be
1922  * extracted from the packet itself.
1923  *
1924  * This must be called before the packet key fields are filled in 'key'.
1925  */
1926 
1927 int ovs_nla_get_flow_metadata(struct net *net,
1928 			      const struct nlattr *a[OVS_KEY_ATTR_MAX + 1],
1929 			      u64 attrs, struct sw_flow_key *key, bool log)
1930 {
1931 	struct sw_flow_match match;
1932 
1933 	memset(&match, 0, sizeof(match));
1934 	match.key = key;
1935 
1936 	key->ct_state = 0;
1937 	key->ct_zone = 0;
1938 	key->ct_orig_proto = 0;
1939 	memset(&key->ct, 0, sizeof(key->ct));
1940 	memset(&key->ipv4.ct_orig, 0, sizeof(key->ipv4.ct_orig));
1941 	memset(&key->ipv6.ct_orig, 0, sizeof(key->ipv6.ct_orig));
1942 
1943 	key->phy.in_port = DP_MAX_PORTS;
1944 
1945 	return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1946 }
1947 
1948 static int ovs_nla_put_vlan(struct sk_buff *skb, const struct vlan_head *vh,
1949 			    bool is_mask)
1950 {
1951 	__be16 eth_type = !is_mask ? vh->tpid : htons(0xffff);
1952 
1953 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1954 	    nla_put_be16(skb, OVS_KEY_ATTR_VLAN, vh->tci))
1955 		return -EMSGSIZE;
1956 	return 0;
1957 }
1958 
1959 static int nsh_key_to_nlattr(const struct ovs_key_nsh *nsh, bool is_mask,
1960 			     struct sk_buff *skb)
1961 {
1962 	struct nlattr *start;
1963 
1964 	start = nla_nest_start_noflag(skb, OVS_KEY_ATTR_NSH);
1965 	if (!start)
1966 		return -EMSGSIZE;
1967 
1968 	if (nla_put(skb, OVS_NSH_KEY_ATTR_BASE, sizeof(nsh->base), &nsh->base))
1969 		goto nla_put_failure;
1970 
1971 	if (is_mask || nsh->base.mdtype == NSH_M_TYPE1) {
1972 		if (nla_put(skb, OVS_NSH_KEY_ATTR_MD1,
1973 			    sizeof(nsh->context), nsh->context))
1974 			goto nla_put_failure;
1975 	}
1976 
1977 	/* Don't support MD type 2 yet */
1978 
1979 	nla_nest_end(skb, start);
1980 
1981 	return 0;
1982 
1983 nla_put_failure:
1984 	return -EMSGSIZE;
1985 }
1986 
1987 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1988 			     const struct sw_flow_key *output, bool is_mask,
1989 			     struct sk_buff *skb)
1990 {
1991 	struct ovs_key_ethernet *eth_key;
1992 	struct nlattr *nla;
1993 	struct nlattr *encap = NULL;
1994 	struct nlattr *in_encap = NULL;
1995 
1996 	if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1997 		goto nla_put_failure;
1998 
1999 	if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
2000 		goto nla_put_failure;
2001 
2002 	if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
2003 		goto nla_put_failure;
2004 
2005 	if ((swkey->tun_proto || is_mask)) {
2006 		const void *opts = NULL;
2007 
2008 		if (ip_tunnel_is_options_present(output->tun_key.tun_flags))
2009 			opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
2010 
2011 		if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
2012 				     swkey->tun_opts_len, swkey->tun_proto, 0))
2013 			goto nla_put_failure;
2014 	}
2015 
2016 	if (swkey->phy.in_port == DP_MAX_PORTS) {
2017 		if (is_mask && (output->phy.in_port == 0xffff))
2018 			if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
2019 				goto nla_put_failure;
2020 	} else {
2021 		u16 upper_u16;
2022 		upper_u16 = !is_mask ? 0 : 0xffff;
2023 
2024 		if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
2025 				(upper_u16 << 16) | output->phy.in_port))
2026 			goto nla_put_failure;
2027 	}
2028 
2029 	if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
2030 		goto nla_put_failure;
2031 
2032 	if (ovs_ct_put_key(swkey, output, skb))
2033 		goto nla_put_failure;
2034 
2035 	if (ovs_key_mac_proto(swkey) == MAC_PROTO_ETHERNET) {
2036 		nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
2037 		if (!nla)
2038 			goto nla_put_failure;
2039 
2040 		eth_key = nla_data(nla);
2041 		ether_addr_copy(eth_key->eth_src, output->eth.src);
2042 		ether_addr_copy(eth_key->eth_dst, output->eth.dst);
2043 
2044 		if (swkey->eth.vlan.tci || eth_type_vlan(swkey->eth.type)) {
2045 			if (ovs_nla_put_vlan(skb, &output->eth.vlan, is_mask))
2046 				goto nla_put_failure;
2047 			encap = nla_nest_start_noflag(skb, OVS_KEY_ATTR_ENCAP);
2048 			if (!swkey->eth.vlan.tci)
2049 				goto unencap;
2050 
2051 			if (swkey->eth.cvlan.tci || eth_type_vlan(swkey->eth.type)) {
2052 				if (ovs_nla_put_vlan(skb, &output->eth.cvlan, is_mask))
2053 					goto nla_put_failure;
2054 				in_encap = nla_nest_start_noflag(skb,
2055 								 OVS_KEY_ATTR_ENCAP);
2056 				if (!swkey->eth.cvlan.tci)
2057 					goto unencap;
2058 			}
2059 		}
2060 
2061 		if (swkey->eth.type == htons(ETH_P_802_2)) {
2062 			/*
2063 			* Ethertype 802.2 is represented in the netlink with omitted
2064 			* OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
2065 			* 0xffff in the mask attribute.  Ethertype can also
2066 			* be wildcarded.
2067 			*/
2068 			if (is_mask && output->eth.type)
2069 				if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
2070 							output->eth.type))
2071 					goto nla_put_failure;
2072 			goto unencap;
2073 		}
2074 	}
2075 
2076 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
2077 		goto nla_put_failure;
2078 
2079 	if (eth_type_vlan(swkey->eth.type)) {
2080 		/* There are 3 VLAN tags, we don't know anything about the rest
2081 		 * of the packet, so truncate here.
2082 		 */
2083 		WARN_ON_ONCE(!(encap && in_encap));
2084 		goto unencap;
2085 	}
2086 
2087 	if (swkey->eth.type == htons(ETH_P_IP)) {
2088 		struct ovs_key_ipv4 *ipv4_key;
2089 
2090 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
2091 		if (!nla)
2092 			goto nla_put_failure;
2093 		ipv4_key = nla_data(nla);
2094 		ipv4_key->ipv4_src = output->ipv4.addr.src;
2095 		ipv4_key->ipv4_dst = output->ipv4.addr.dst;
2096 		ipv4_key->ipv4_proto = output->ip.proto;
2097 		ipv4_key->ipv4_tos = output->ip.tos;
2098 		ipv4_key->ipv4_ttl = output->ip.ttl;
2099 		ipv4_key->ipv4_frag = output->ip.frag;
2100 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
2101 		struct ovs_key_ipv6 *ipv6_key;
2102 		struct ovs_key_ipv6_exthdrs *ipv6_exthdrs_key;
2103 
2104 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
2105 		if (!nla)
2106 			goto nla_put_failure;
2107 		ipv6_key = nla_data(nla);
2108 		memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
2109 				sizeof(ipv6_key->ipv6_src));
2110 		memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
2111 				sizeof(ipv6_key->ipv6_dst));
2112 		ipv6_key->ipv6_label = output->ipv6.label;
2113 		ipv6_key->ipv6_proto = output->ip.proto;
2114 		ipv6_key->ipv6_tclass = output->ip.tos;
2115 		ipv6_key->ipv6_hlimit = output->ip.ttl;
2116 		ipv6_key->ipv6_frag = output->ip.frag;
2117 
2118 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6_EXTHDRS,
2119 				  sizeof(*ipv6_exthdrs_key));
2120 		if (!nla)
2121 			goto nla_put_failure;
2122 		ipv6_exthdrs_key = nla_data(nla);
2123 		ipv6_exthdrs_key->hdrs = output->ipv6.exthdrs;
2124 	} else if (swkey->eth.type == htons(ETH_P_NSH)) {
2125 		if (nsh_key_to_nlattr(&output->nsh, is_mask, skb))
2126 			goto nla_put_failure;
2127 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
2128 		   swkey->eth.type == htons(ETH_P_RARP)) {
2129 		struct ovs_key_arp *arp_key;
2130 
2131 		nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
2132 		if (!nla)
2133 			goto nla_put_failure;
2134 		arp_key = nla_data(nla);
2135 		memset(arp_key, 0, sizeof(struct ovs_key_arp));
2136 		arp_key->arp_sip = output->ipv4.addr.src;
2137 		arp_key->arp_tip = output->ipv4.addr.dst;
2138 		arp_key->arp_op = htons(output->ip.proto);
2139 		ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
2140 		ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
2141 	} else if (eth_p_mpls(swkey->eth.type)) {
2142 		u8 i, num_labels;
2143 		struct ovs_key_mpls *mpls_key;
2144 
2145 		num_labels = hweight_long(output->mpls.num_labels_mask);
2146 		nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS,
2147 				  num_labels * sizeof(*mpls_key));
2148 		if (!nla)
2149 			goto nla_put_failure;
2150 
2151 		mpls_key = nla_data(nla);
2152 		for (i = 0; i < num_labels; i++)
2153 			mpls_key[i].mpls_lse = output->mpls.lse[i];
2154 	}
2155 
2156 	if ((swkey->eth.type == htons(ETH_P_IP) ||
2157 	     swkey->eth.type == htons(ETH_P_IPV6)) &&
2158 	     swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
2159 
2160 		if (swkey->ip.proto == IPPROTO_TCP) {
2161 			struct ovs_key_tcp *tcp_key;
2162 
2163 			nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
2164 			if (!nla)
2165 				goto nla_put_failure;
2166 			tcp_key = nla_data(nla);
2167 			tcp_key->tcp_src = output->tp.src;
2168 			tcp_key->tcp_dst = output->tp.dst;
2169 			if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
2170 					 output->tp.flags))
2171 				goto nla_put_failure;
2172 		} else if (swkey->ip.proto == IPPROTO_UDP) {
2173 			struct ovs_key_udp *udp_key;
2174 
2175 			nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
2176 			if (!nla)
2177 				goto nla_put_failure;
2178 			udp_key = nla_data(nla);
2179 			udp_key->udp_src = output->tp.src;
2180 			udp_key->udp_dst = output->tp.dst;
2181 		} else if (swkey->ip.proto == IPPROTO_SCTP) {
2182 			struct ovs_key_sctp *sctp_key;
2183 
2184 			nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
2185 			if (!nla)
2186 				goto nla_put_failure;
2187 			sctp_key = nla_data(nla);
2188 			sctp_key->sctp_src = output->tp.src;
2189 			sctp_key->sctp_dst = output->tp.dst;
2190 		} else if (swkey->eth.type == htons(ETH_P_IP) &&
2191 			   swkey->ip.proto == IPPROTO_ICMP) {
2192 			struct ovs_key_icmp *icmp_key;
2193 
2194 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
2195 			if (!nla)
2196 				goto nla_put_failure;
2197 			icmp_key = nla_data(nla);
2198 			icmp_key->icmp_type = ntohs(output->tp.src);
2199 			icmp_key->icmp_code = ntohs(output->tp.dst);
2200 		} else if (swkey->eth.type == htons(ETH_P_IPV6) &&
2201 			   swkey->ip.proto == IPPROTO_ICMPV6) {
2202 			struct ovs_key_icmpv6 *icmpv6_key;
2203 
2204 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
2205 						sizeof(*icmpv6_key));
2206 			if (!nla)
2207 				goto nla_put_failure;
2208 			icmpv6_key = nla_data(nla);
2209 			icmpv6_key->icmpv6_type = ntohs(output->tp.src);
2210 			icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
2211 
2212 			if (swkey->tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
2213 			    swkey->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
2214 				struct ovs_key_nd *nd_key;
2215 
2216 				nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
2217 				if (!nla)
2218 					goto nla_put_failure;
2219 				nd_key = nla_data(nla);
2220 				memcpy(nd_key->nd_target, &output->ipv6.nd.target,
2221 							sizeof(nd_key->nd_target));
2222 				ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
2223 				ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
2224 			}
2225 		}
2226 	}
2227 
2228 unencap:
2229 	if (in_encap)
2230 		nla_nest_end(skb, in_encap);
2231 	if (encap)
2232 		nla_nest_end(skb, encap);
2233 
2234 	return 0;
2235 
2236 nla_put_failure:
2237 	return -EMSGSIZE;
2238 }
2239 
2240 int ovs_nla_put_key(const struct sw_flow_key *swkey,
2241 		    const struct sw_flow_key *output, int attr, bool is_mask,
2242 		    struct sk_buff *skb)
2243 {
2244 	int err;
2245 	struct nlattr *nla;
2246 
2247 	nla = nla_nest_start_noflag(skb, attr);
2248 	if (!nla)
2249 		return -EMSGSIZE;
2250 	err = __ovs_nla_put_key(swkey, output, is_mask, skb);
2251 	if (err)
2252 		return err;
2253 	nla_nest_end(skb, nla);
2254 
2255 	return 0;
2256 }
2257 
2258 /* Called with ovs_mutex or RCU read lock. */
2259 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
2260 {
2261 	if (ovs_identifier_is_ufid(&flow->id))
2262 		return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
2263 			       flow->id.ufid);
2264 
2265 	return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
2266 			       OVS_FLOW_ATTR_KEY, false, skb);
2267 }
2268 
2269 /* Called with ovs_mutex or RCU read lock. */
2270 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
2271 {
2272 	return ovs_nla_put_key(&flow->key, &flow->key,
2273 				OVS_FLOW_ATTR_KEY, false, skb);
2274 }
2275 
2276 /* Called with ovs_mutex or RCU read lock. */
2277 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
2278 {
2279 	return ovs_nla_put_key(&flow->key, &flow->mask->key,
2280 				OVS_FLOW_ATTR_MASK, true, skb);
2281 }
2282 
2283 static struct sw_flow_actions *nla_alloc_flow_actions(int size)
2284 {
2285 	struct sw_flow_actions *sfa;
2286 
2287 	sfa = kmalloc(kmalloc_size_roundup(sizeof(*sfa) + size), GFP_KERNEL);
2288 	if (!sfa)
2289 		return ERR_PTR(-ENOMEM);
2290 
2291 	sfa->actions_len = 0;
2292 	return sfa;
2293 }
2294 
2295 static void ovs_nla_free_nested_actions(const struct nlattr *actions, int len);
2296 
2297 static void ovs_nla_free_check_pkt_len_action(const struct nlattr *action)
2298 {
2299 	const struct nlattr *a;
2300 	int rem;
2301 
2302 	nla_for_each_nested(a, action, rem) {
2303 		switch (nla_type(a)) {
2304 		case OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL:
2305 		case OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER:
2306 			ovs_nla_free_nested_actions(nla_data(a), nla_len(a));
2307 			break;
2308 		}
2309 	}
2310 }
2311 
2312 static void ovs_nla_free_clone_action(const struct nlattr *action)
2313 {
2314 	const struct nlattr *a = nla_data(action);
2315 	int rem = nla_len(action);
2316 
2317 	switch (nla_type(a)) {
2318 	case OVS_CLONE_ATTR_EXEC:
2319 		/* The real list of actions follows this attribute. */
2320 		a = nla_next(a, &rem);
2321 		ovs_nla_free_nested_actions(a, rem);
2322 		break;
2323 	}
2324 }
2325 
2326 static void ovs_nla_free_dec_ttl_action(const struct nlattr *action)
2327 {
2328 	const struct nlattr *a = nla_data(action);
2329 
2330 	switch (nla_type(a)) {
2331 	case OVS_DEC_TTL_ATTR_ACTION:
2332 		ovs_nla_free_nested_actions(nla_data(a), nla_len(a));
2333 		break;
2334 	}
2335 }
2336 
2337 static void ovs_nla_free_sample_action(const struct nlattr *action)
2338 {
2339 	const struct nlattr *a = nla_data(action);
2340 	int rem = nla_len(action);
2341 
2342 	switch (nla_type(a)) {
2343 	case OVS_SAMPLE_ATTR_ARG:
2344 		/* The real list of actions follows this attribute. */
2345 		a = nla_next(a, &rem);
2346 		ovs_nla_free_nested_actions(a, rem);
2347 		break;
2348 	}
2349 }
2350 
2351 static void ovs_nla_free_set_action(const struct nlattr *a)
2352 {
2353 	const struct nlattr *ovs_key = nla_data(a);
2354 	struct ovs_tunnel_info *ovs_tun;
2355 
2356 	switch (nla_type(ovs_key)) {
2357 	case OVS_KEY_ATTR_TUNNEL_INFO:
2358 		ovs_tun = nla_data(ovs_key);
2359 		dst_release((struct dst_entry *)ovs_tun->tun_dst);
2360 		break;
2361 	}
2362 }
2363 
2364 static void ovs_nla_free_nested_actions(const struct nlattr *actions, int len)
2365 {
2366 	const struct nlattr *a;
2367 	int rem;
2368 
2369 	/* Whenever new actions are added, the need to update this
2370 	 * function should be considered.
2371 	 */
2372 	BUILD_BUG_ON(OVS_ACTION_ATTR_MAX != 25);
2373 
2374 	if (!actions)
2375 		return;
2376 
2377 	nla_for_each_attr(a, actions, len, rem) {
2378 		switch (nla_type(a)) {
2379 		case OVS_ACTION_ATTR_CHECK_PKT_LEN:
2380 			ovs_nla_free_check_pkt_len_action(a);
2381 			break;
2382 
2383 		case OVS_ACTION_ATTR_CLONE:
2384 			ovs_nla_free_clone_action(a);
2385 			break;
2386 
2387 		case OVS_ACTION_ATTR_CT:
2388 			ovs_ct_free_action(a);
2389 			break;
2390 
2391 		case OVS_ACTION_ATTR_DEC_TTL:
2392 			ovs_nla_free_dec_ttl_action(a);
2393 			break;
2394 
2395 		case OVS_ACTION_ATTR_SAMPLE:
2396 			ovs_nla_free_sample_action(a);
2397 			break;
2398 
2399 		case OVS_ACTION_ATTR_SET:
2400 			ovs_nla_free_set_action(a);
2401 			break;
2402 		}
2403 	}
2404 }
2405 
2406 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
2407 {
2408 	if (!sf_acts)
2409 		return;
2410 
2411 	ovs_nla_free_nested_actions(sf_acts->actions, sf_acts->actions_len);
2412 	kfree(sf_acts);
2413 }
2414 
2415 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
2416 {
2417 	ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
2418 }
2419 
2420 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
2421  * The caller must hold rcu_read_lock for this to be sensible. */
2422 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
2423 {
2424 	call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
2425 }
2426 
2427 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
2428 				       int attr_len, bool log)
2429 {
2430 
2431 	struct sw_flow_actions *acts;
2432 	int new_acts_size;
2433 	size_t req_size = NLA_ALIGN(attr_len);
2434 	int next_offset = offsetof(struct sw_flow_actions, actions) +
2435 					(*sfa)->actions_len;
2436 
2437 	if (req_size <= (ksize(*sfa) - next_offset))
2438 		goto out;
2439 
2440 	new_acts_size = max(next_offset + req_size, ksize(*sfa) * 2);
2441 
2442 	acts = nla_alloc_flow_actions(new_acts_size);
2443 	if (IS_ERR(acts))
2444 		return ERR_CAST(acts);
2445 
2446 	memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
2447 	acts->actions_len = (*sfa)->actions_len;
2448 	acts->orig_len = (*sfa)->orig_len;
2449 	kfree(*sfa);
2450 	*sfa = acts;
2451 
2452 out:
2453 	(*sfa)->actions_len += req_size;
2454 	return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
2455 }
2456 
2457 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
2458 				   int attrtype, void *data, int len, bool log)
2459 {
2460 	struct nlattr *a;
2461 
2462 	a = reserve_sfa_size(sfa, nla_attr_size(len), log);
2463 	if (IS_ERR(a))
2464 		return a;
2465 
2466 	a->nla_type = attrtype;
2467 	a->nla_len = nla_attr_size(len);
2468 
2469 	if (data)
2470 		memcpy(nla_data(a), data, len);
2471 	memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
2472 
2473 	return a;
2474 }
2475 
2476 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
2477 		       int len, bool log)
2478 {
2479 	struct nlattr *a;
2480 
2481 	a = __add_action(sfa, attrtype, data, len, log);
2482 
2483 	return PTR_ERR_OR_ZERO(a);
2484 }
2485 
2486 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
2487 					  int attrtype, bool log)
2488 {
2489 	int used = (*sfa)->actions_len;
2490 	int err;
2491 
2492 	err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
2493 	if (err)
2494 		return err;
2495 
2496 	return used;
2497 }
2498 
2499 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
2500 					 int st_offset)
2501 {
2502 	struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
2503 							       st_offset);
2504 
2505 	a->nla_len = sfa->actions_len - st_offset;
2506 }
2507 
2508 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2509 				  const struct sw_flow_key *key,
2510 				  struct sw_flow_actions **sfa,
2511 				  __be16 eth_type, __be16 vlan_tci,
2512 				  u32 mpls_label_count, bool log,
2513 				  u32 depth);
2514 
2515 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
2516 				    const struct sw_flow_key *key,
2517 				    struct sw_flow_actions **sfa,
2518 				    __be16 eth_type, __be16 vlan_tci,
2519 				    u32 mpls_label_count, bool log, bool last,
2520 				    u32 depth)
2521 {
2522 	const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
2523 	const struct nlattr *probability, *actions;
2524 	const struct nlattr *a;
2525 	int rem, start, err;
2526 	struct sample_arg arg;
2527 
2528 	memset(attrs, 0, sizeof(attrs));
2529 	nla_for_each_nested(a, attr, rem) {
2530 		int type = nla_type(a);
2531 		if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
2532 			return -EINVAL;
2533 		attrs[type] = a;
2534 	}
2535 	if (rem)
2536 		return -EINVAL;
2537 
2538 	probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
2539 	if (!probability || nla_len(probability) != sizeof(u32))
2540 		return -EINVAL;
2541 
2542 	actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
2543 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
2544 		return -EINVAL;
2545 
2546 	/* validation done, copy sample action. */
2547 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
2548 	if (start < 0)
2549 		return start;
2550 
2551 	/* When both skb and flow may be changed, put the sample
2552 	 * into a deferred fifo. On the other hand, if only skb
2553 	 * may be modified, the actions can be executed in place.
2554 	 *
2555 	 * Do this analysis at the flow installation time.
2556 	 * Set 'clone_action->exec' to true if the actions can be
2557 	 * executed without being deferred.
2558 	 *
2559 	 * If the sample is the last action, it can always be excuted
2560 	 * rather than deferred.
2561 	 */
2562 	arg.exec = last || !actions_may_change_flow(actions);
2563 	arg.probability = nla_get_u32(probability);
2564 
2565 	err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_ARG, &arg, sizeof(arg),
2566 				 log);
2567 	if (err)
2568 		return err;
2569 
2570 	err = __ovs_nla_copy_actions(net, actions, key, sfa,
2571 				     eth_type, vlan_tci, mpls_label_count, log,
2572 				     depth + 1);
2573 
2574 	if (err)
2575 		return err;
2576 
2577 	add_nested_action_end(*sfa, start);
2578 
2579 	return 0;
2580 }
2581 
2582 static int validate_and_copy_dec_ttl(struct net *net,
2583 				     const struct nlattr *attr,
2584 				     const struct sw_flow_key *key,
2585 				     struct sw_flow_actions **sfa,
2586 				     __be16 eth_type, __be16 vlan_tci,
2587 				     u32 mpls_label_count, bool log,
2588 				     u32 depth)
2589 {
2590 	const struct nlattr *attrs[OVS_DEC_TTL_ATTR_MAX + 1];
2591 	int start, action_start, err, rem;
2592 	const struct nlattr *a, *actions;
2593 
2594 	memset(attrs, 0, sizeof(attrs));
2595 	nla_for_each_nested(a, attr, rem) {
2596 		int type = nla_type(a);
2597 
2598 		/* Ignore unknown attributes to be future proof. */
2599 		if (type > OVS_DEC_TTL_ATTR_MAX)
2600 			continue;
2601 
2602 		if (!type || attrs[type]) {
2603 			OVS_NLERR(log, "Duplicate or invalid key (type %d).",
2604 				  type);
2605 			return -EINVAL;
2606 		}
2607 
2608 		attrs[type] = a;
2609 	}
2610 
2611 	if (rem) {
2612 		OVS_NLERR(log, "Message has %d unknown bytes.", rem);
2613 		return -EINVAL;
2614 	}
2615 
2616 	actions = attrs[OVS_DEC_TTL_ATTR_ACTION];
2617 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN)) {
2618 		OVS_NLERR(log, "Missing valid actions attribute.");
2619 		return -EINVAL;
2620 	}
2621 
2622 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_DEC_TTL, log);
2623 	if (start < 0)
2624 		return start;
2625 
2626 	action_start = add_nested_action_start(sfa, OVS_DEC_TTL_ATTR_ACTION, log);
2627 	if (action_start < 0)
2628 		return action_start;
2629 
2630 	err = __ovs_nla_copy_actions(net, actions, key, sfa, eth_type,
2631 				     vlan_tci, mpls_label_count, log,
2632 				     depth + 1);
2633 	if (err)
2634 		return err;
2635 
2636 	add_nested_action_end(*sfa, action_start);
2637 	add_nested_action_end(*sfa, start);
2638 	return 0;
2639 }
2640 
2641 static int validate_and_copy_clone(struct net *net,
2642 				   const struct nlattr *attr,
2643 				   const struct sw_flow_key *key,
2644 				   struct sw_flow_actions **sfa,
2645 				   __be16 eth_type, __be16 vlan_tci,
2646 				   u32 mpls_label_count, bool log, bool last,
2647 				   u32 depth)
2648 {
2649 	int start, err;
2650 	u32 exec;
2651 
2652 	if (nla_len(attr) && nla_len(attr) < NLA_HDRLEN)
2653 		return -EINVAL;
2654 
2655 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_CLONE, log);
2656 	if (start < 0)
2657 		return start;
2658 
2659 	exec = last || !actions_may_change_flow(attr);
2660 
2661 	err = ovs_nla_add_action(sfa, OVS_CLONE_ATTR_EXEC, &exec,
2662 				 sizeof(exec), log);
2663 	if (err)
2664 		return err;
2665 
2666 	err = __ovs_nla_copy_actions(net, attr, key, sfa,
2667 				     eth_type, vlan_tci, mpls_label_count, log,
2668 				     depth + 1);
2669 	if (err)
2670 		return err;
2671 
2672 	add_nested_action_end(*sfa, start);
2673 
2674 	return 0;
2675 }
2676 
2677 void ovs_match_init(struct sw_flow_match *match,
2678 		    struct sw_flow_key *key,
2679 		    bool reset_key,
2680 		    struct sw_flow_mask *mask)
2681 {
2682 	memset(match, 0, sizeof(*match));
2683 	match->key = key;
2684 	match->mask = mask;
2685 
2686 	if (reset_key)
2687 		memset(key, 0, sizeof(*key));
2688 
2689 	if (mask) {
2690 		memset(&mask->key, 0, sizeof(mask->key));
2691 		mask->range.start = mask->range.end = 0;
2692 	}
2693 }
2694 
2695 static int validate_geneve_opts(struct sw_flow_key *key)
2696 {
2697 	struct geneve_opt *option;
2698 	int opts_len = key->tun_opts_len;
2699 	bool crit_opt = false;
2700 
2701 	option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
2702 	while (opts_len > 0) {
2703 		int len;
2704 
2705 		if (opts_len < sizeof(*option))
2706 			return -EINVAL;
2707 
2708 		len = sizeof(*option) + option->length * 4;
2709 		if (len > opts_len)
2710 			return -EINVAL;
2711 
2712 		crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
2713 
2714 		option = (struct geneve_opt *)((u8 *)option + len);
2715 		opts_len -= len;
2716 	}
2717 
2718 	if (crit_opt)
2719 		__set_bit(IP_TUNNEL_CRIT_OPT_BIT, key->tun_key.tun_flags);
2720 
2721 	return 0;
2722 }
2723 
2724 static int validate_and_copy_set_tun(const struct nlattr *attr,
2725 				     struct sw_flow_actions **sfa, bool log)
2726 {
2727 	IP_TUNNEL_DECLARE_FLAGS(dst_opt_type) = { };
2728 	struct sw_flow_match match;
2729 	struct sw_flow_key key;
2730 	struct metadata_dst *tun_dst;
2731 	struct ip_tunnel_info *tun_info;
2732 	struct ovs_tunnel_info *ovs_tun;
2733 	struct nlattr *a;
2734 	int err = 0, start, opts_type;
2735 
2736 	ovs_match_init(&match, &key, true, NULL);
2737 	opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
2738 	if (opts_type < 0)
2739 		return opts_type;
2740 
2741 	if (key.tun_opts_len) {
2742 		switch (opts_type) {
2743 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
2744 			err = validate_geneve_opts(&key);
2745 			if (err < 0)
2746 				return err;
2747 
2748 			__set_bit(IP_TUNNEL_GENEVE_OPT_BIT, dst_opt_type);
2749 			break;
2750 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
2751 			__set_bit(IP_TUNNEL_VXLAN_OPT_BIT, dst_opt_type);
2752 			break;
2753 		case OVS_TUNNEL_KEY_ATTR_ERSPAN_OPTS:
2754 			__set_bit(IP_TUNNEL_ERSPAN_OPT_BIT, dst_opt_type);
2755 			break;
2756 		}
2757 	}
2758 
2759 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
2760 	if (start < 0)
2761 		return start;
2762 
2763 	tun_dst = metadata_dst_alloc(key.tun_opts_len, METADATA_IP_TUNNEL,
2764 				     GFP_KERNEL);
2765 
2766 	if (!tun_dst)
2767 		return -ENOMEM;
2768 
2769 	err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
2770 	if (err) {
2771 		dst_release((struct dst_entry *)tun_dst);
2772 		return err;
2773 	}
2774 
2775 	a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
2776 			 sizeof(*ovs_tun), log);
2777 	if (IS_ERR(a)) {
2778 		dst_release((struct dst_entry *)tun_dst);
2779 		return PTR_ERR(a);
2780 	}
2781 
2782 	ovs_tun = nla_data(a);
2783 	ovs_tun->tun_dst = tun_dst;
2784 
2785 	tun_info = &tun_dst->u.tun_info;
2786 	tun_info->mode = IP_TUNNEL_INFO_TX;
2787 	if (key.tun_proto == AF_INET6)
2788 		tun_info->mode |= IP_TUNNEL_INFO_IPV6;
2789 	else if (key.tun_proto == AF_INET && key.tun_key.u.ipv4.dst == 0)
2790 		tun_info->mode |= IP_TUNNEL_INFO_BRIDGE;
2791 	tun_info->key = key.tun_key;
2792 
2793 	/* We need to store the options in the action itself since
2794 	 * everything else will go away after flow setup. We can append
2795 	 * it to tun_info and then point there.
2796 	 */
2797 	ip_tunnel_info_opts_set(tun_info,
2798 				TUN_METADATA_OPTS(&key, key.tun_opts_len),
2799 				key.tun_opts_len, dst_opt_type);
2800 	add_nested_action_end(*sfa, start);
2801 
2802 	return err;
2803 }
2804 
2805 static bool validate_push_nsh(const struct nlattr *attr, bool log)
2806 {
2807 	struct sw_flow_match match;
2808 	struct sw_flow_key key;
2809 
2810 	ovs_match_init(&match, &key, true, NULL);
2811 	return !nsh_key_put_from_nlattr(attr, &match, false, true, log);
2812 }
2813 
2814 /* Return false if there are any non-masked bits set.
2815  * Mask follows data immediately, before any netlink padding.
2816  */
2817 static bool validate_masked(u8 *data, int len)
2818 {
2819 	u8 *mask = data + len;
2820 
2821 	while (len--)
2822 		if (*data++ & ~*mask++)
2823 			return false;
2824 
2825 	return true;
2826 }
2827 
2828 static int validate_set(const struct nlattr *a,
2829 			const struct sw_flow_key *flow_key,
2830 			struct sw_flow_actions **sfa, bool *skip_copy,
2831 			u8 mac_proto, __be16 eth_type, bool masked, bool log)
2832 {
2833 	const struct nlattr *ovs_key = nla_data(a);
2834 	int key_type = nla_type(ovs_key);
2835 	size_t key_len;
2836 
2837 	/* There can be only one key in a action */
2838 	if (!nla_ok(ovs_key, nla_len(a)) ||
2839 	    nla_total_size(nla_len(ovs_key)) != nla_len(a))
2840 		return -EINVAL;
2841 
2842 	key_len = nla_len(ovs_key);
2843 	if (masked)
2844 		key_len /= 2;
2845 
2846 	if (key_type > OVS_KEY_ATTR_MAX ||
2847 	    !check_attr_len(key_len, ovs_key_lens[key_type].len))
2848 		return -EINVAL;
2849 
2850 	if (masked && !validate_masked(nla_data(ovs_key), key_len))
2851 		return -EINVAL;
2852 
2853 	switch (key_type) {
2854 	case OVS_KEY_ATTR_PRIORITY:
2855 	case OVS_KEY_ATTR_SKB_MARK:
2856 	case OVS_KEY_ATTR_CT_MARK:
2857 	case OVS_KEY_ATTR_CT_LABELS:
2858 		break;
2859 
2860 	case OVS_KEY_ATTR_ETHERNET:
2861 		if (mac_proto != MAC_PROTO_ETHERNET)
2862 			return -EINVAL;
2863 		break;
2864 
2865 	case OVS_KEY_ATTR_TUNNEL: {
2866 		int err;
2867 
2868 		if (masked)
2869 			return -EINVAL; /* Masked tunnel set not supported. */
2870 
2871 		*skip_copy = true;
2872 		err = validate_and_copy_set_tun(a, sfa, log);
2873 		if (err)
2874 			return err;
2875 		break;
2876 	}
2877 	case OVS_KEY_ATTR_IPV4: {
2878 		const struct ovs_key_ipv4 *ipv4_key;
2879 
2880 		if (eth_type != htons(ETH_P_IP))
2881 			return -EINVAL;
2882 
2883 		ipv4_key = nla_data(ovs_key);
2884 
2885 		if (masked) {
2886 			const struct ovs_key_ipv4 *mask = ipv4_key + 1;
2887 
2888 			/* Non-writeable fields. */
2889 			if (mask->ipv4_proto || mask->ipv4_frag)
2890 				return -EINVAL;
2891 		} else {
2892 			if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2893 				return -EINVAL;
2894 
2895 			if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2896 				return -EINVAL;
2897 		}
2898 		break;
2899 	}
2900 	case OVS_KEY_ATTR_IPV6: {
2901 		const struct ovs_key_ipv6 *ipv6_key;
2902 
2903 		if (eth_type != htons(ETH_P_IPV6))
2904 			return -EINVAL;
2905 
2906 		ipv6_key = nla_data(ovs_key);
2907 
2908 		if (masked) {
2909 			const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2910 
2911 			/* Non-writeable fields. */
2912 			if (mask->ipv6_proto || mask->ipv6_frag)
2913 				return -EINVAL;
2914 
2915 			/* Invalid bits in the flow label mask? */
2916 			if (ntohl(mask->ipv6_label) & 0xFFF00000)
2917 				return -EINVAL;
2918 		} else {
2919 			if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2920 				return -EINVAL;
2921 
2922 			if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2923 				return -EINVAL;
2924 		}
2925 		if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2926 			return -EINVAL;
2927 
2928 		break;
2929 	}
2930 	case OVS_KEY_ATTR_TCP:
2931 		if ((eth_type != htons(ETH_P_IP) &&
2932 		     eth_type != htons(ETH_P_IPV6)) ||
2933 		    flow_key->ip.proto != IPPROTO_TCP)
2934 			return -EINVAL;
2935 
2936 		break;
2937 
2938 	case OVS_KEY_ATTR_UDP:
2939 		if ((eth_type != htons(ETH_P_IP) &&
2940 		     eth_type != htons(ETH_P_IPV6)) ||
2941 		    flow_key->ip.proto != IPPROTO_UDP)
2942 			return -EINVAL;
2943 
2944 		break;
2945 
2946 	case OVS_KEY_ATTR_MPLS:
2947 		if (!eth_p_mpls(eth_type))
2948 			return -EINVAL;
2949 		break;
2950 
2951 	case OVS_KEY_ATTR_SCTP:
2952 		if ((eth_type != htons(ETH_P_IP) &&
2953 		     eth_type != htons(ETH_P_IPV6)) ||
2954 		    flow_key->ip.proto != IPPROTO_SCTP)
2955 			return -EINVAL;
2956 
2957 		break;
2958 
2959 	default:
2960 		return -EINVAL;
2961 	}
2962 
2963 	/* Convert non-masked non-tunnel set actions to masked set actions. */
2964 	if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2965 		int start, len = key_len * 2;
2966 		struct nlattr *at;
2967 
2968 		*skip_copy = true;
2969 
2970 		start = add_nested_action_start(sfa,
2971 						OVS_ACTION_ATTR_SET_TO_MASKED,
2972 						log);
2973 		if (start < 0)
2974 			return start;
2975 
2976 		at = __add_action(sfa, key_type, NULL, len, log);
2977 		if (IS_ERR(at))
2978 			return PTR_ERR(at);
2979 
2980 		memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2981 		memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2982 		/* Clear non-writeable bits from otherwise writeable fields. */
2983 		if (key_type == OVS_KEY_ATTR_IPV6) {
2984 			struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2985 
2986 			mask->ipv6_label &= htonl(0x000FFFFF);
2987 		}
2988 		add_nested_action_end(*sfa, start);
2989 	}
2990 
2991 	return 0;
2992 }
2993 
2994 static int validate_userspace(const struct nlattr *attr)
2995 {
2996 	static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2997 		[OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2998 		[OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2999 		[OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
3000 	};
3001 	struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
3002 	int error;
3003 
3004 	error = nla_parse_deprecated_strict(a, OVS_USERSPACE_ATTR_MAX,
3005 					    nla_data(attr), nla_len(attr),
3006 					    userspace_policy, NULL);
3007 	if (error)
3008 		return error;
3009 
3010 	if (!a[OVS_USERSPACE_ATTR_PID] ||
3011 	    !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
3012 		return -EINVAL;
3013 
3014 	return 0;
3015 }
3016 
3017 static const struct nla_policy cpl_policy[OVS_CHECK_PKT_LEN_ATTR_MAX + 1] = {
3018 	[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN] = {.type = NLA_U16 },
3019 	[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER] = {.type = NLA_NESTED },
3020 	[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL] = {.type = NLA_NESTED },
3021 };
3022 
3023 static int validate_and_copy_check_pkt_len(struct net *net,
3024 					   const struct nlattr *attr,
3025 					   const struct sw_flow_key *key,
3026 					   struct sw_flow_actions **sfa,
3027 					   __be16 eth_type, __be16 vlan_tci,
3028 					   u32 mpls_label_count,
3029 					   bool log, bool last, u32 depth)
3030 {
3031 	const struct nlattr *acts_if_greater, *acts_if_lesser_eq;
3032 	struct nlattr *a[OVS_CHECK_PKT_LEN_ATTR_MAX + 1];
3033 	struct check_pkt_len_arg arg;
3034 	int nested_acts_start;
3035 	int start, err;
3036 
3037 	err = nla_parse_deprecated_strict(a, OVS_CHECK_PKT_LEN_ATTR_MAX,
3038 					  nla_data(attr), nla_len(attr),
3039 					  cpl_policy, NULL);
3040 	if (err)
3041 		return err;
3042 
3043 	if (!a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN] ||
3044 	    !nla_get_u16(a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN]))
3045 		return -EINVAL;
3046 
3047 	acts_if_lesser_eq = a[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL];
3048 	acts_if_greater = a[OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER];
3049 
3050 	/* Both the nested action should be present. */
3051 	if (!acts_if_greater || !acts_if_lesser_eq)
3052 		return -EINVAL;
3053 
3054 	/* validation done, copy the nested actions. */
3055 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_CHECK_PKT_LEN,
3056 					log);
3057 	if (start < 0)
3058 		return start;
3059 
3060 	arg.pkt_len = nla_get_u16(a[OVS_CHECK_PKT_LEN_ATTR_PKT_LEN]);
3061 	arg.exec_for_lesser_equal =
3062 		last || !actions_may_change_flow(acts_if_lesser_eq);
3063 	arg.exec_for_greater =
3064 		last || !actions_may_change_flow(acts_if_greater);
3065 
3066 	err = ovs_nla_add_action(sfa, OVS_CHECK_PKT_LEN_ATTR_ARG, &arg,
3067 				 sizeof(arg), log);
3068 	if (err)
3069 		return err;
3070 
3071 	nested_acts_start = add_nested_action_start(sfa,
3072 		OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL, log);
3073 	if (nested_acts_start < 0)
3074 		return nested_acts_start;
3075 
3076 	err = __ovs_nla_copy_actions(net, acts_if_lesser_eq, key, sfa,
3077 				     eth_type, vlan_tci, mpls_label_count, log,
3078 				     depth + 1);
3079 
3080 	if (err)
3081 		return err;
3082 
3083 	add_nested_action_end(*sfa, nested_acts_start);
3084 
3085 	nested_acts_start = add_nested_action_start(sfa,
3086 		OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER, log);
3087 	if (nested_acts_start < 0)
3088 		return nested_acts_start;
3089 
3090 	err = __ovs_nla_copy_actions(net, acts_if_greater, key, sfa,
3091 				     eth_type, vlan_tci, mpls_label_count, log,
3092 				     depth + 1);
3093 
3094 	if (err)
3095 		return err;
3096 
3097 	add_nested_action_end(*sfa, nested_acts_start);
3098 	add_nested_action_end(*sfa, start);
3099 	return 0;
3100 }
3101 
3102 static int validate_psample(const struct nlattr *attr)
3103 {
3104 	static const struct nla_policy policy[OVS_PSAMPLE_ATTR_MAX + 1] = {
3105 		[OVS_PSAMPLE_ATTR_GROUP] = { .type = NLA_U32 },
3106 		[OVS_PSAMPLE_ATTR_COOKIE] = {
3107 			.type = NLA_BINARY,
3108 			.len = OVS_PSAMPLE_COOKIE_MAX_SIZE,
3109 		},
3110 	};
3111 	struct nlattr *a[OVS_PSAMPLE_ATTR_MAX + 1];
3112 	int err;
3113 
3114 	if (!IS_ENABLED(CONFIG_PSAMPLE))
3115 		return -EOPNOTSUPP;
3116 
3117 	err = nla_parse_nested(a, OVS_PSAMPLE_ATTR_MAX, attr, policy, NULL);
3118 	if (err)
3119 		return err;
3120 
3121 	return a[OVS_PSAMPLE_ATTR_GROUP] ? 0 : -EINVAL;
3122 }
3123 
3124 static int copy_action(const struct nlattr *from,
3125 		       struct sw_flow_actions **sfa, bool log)
3126 {
3127 	int totlen = NLA_ALIGN(from->nla_len);
3128 	struct nlattr *to;
3129 
3130 	to = reserve_sfa_size(sfa, from->nla_len, log);
3131 	if (IS_ERR(to))
3132 		return PTR_ERR(to);
3133 
3134 	memcpy(to, from, totlen);
3135 	return 0;
3136 }
3137 
3138 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
3139 				  const struct sw_flow_key *key,
3140 				  struct sw_flow_actions **sfa,
3141 				  __be16 eth_type, __be16 vlan_tci,
3142 				  u32 mpls_label_count, bool log,
3143 				  u32 depth)
3144 {
3145 	u8 mac_proto = ovs_key_mac_proto(key);
3146 	const struct nlattr *a;
3147 	int rem, err;
3148 
3149 	if (depth > OVS_COPY_ACTIONS_MAX_DEPTH)
3150 		return -EOVERFLOW;
3151 
3152 	nla_for_each_nested(a, attr, rem) {
3153 		/* Expected argument lengths, (u32)-1 for variable length. */
3154 		static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
3155 			[OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
3156 			[OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
3157 			[OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
3158 			[OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
3159 			[OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
3160 			[OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
3161 			[OVS_ACTION_ATTR_POP_VLAN] = 0,
3162 			[OVS_ACTION_ATTR_SET] = (u32)-1,
3163 			[OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
3164 			[OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
3165 			[OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
3166 			[OVS_ACTION_ATTR_CT] = (u32)-1,
3167 			[OVS_ACTION_ATTR_CT_CLEAR] = 0,
3168 			[OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
3169 			[OVS_ACTION_ATTR_PUSH_ETH] = sizeof(struct ovs_action_push_eth),
3170 			[OVS_ACTION_ATTR_POP_ETH] = 0,
3171 			[OVS_ACTION_ATTR_PUSH_NSH] = (u32)-1,
3172 			[OVS_ACTION_ATTR_POP_NSH] = 0,
3173 			[OVS_ACTION_ATTR_METER] = sizeof(u32),
3174 			[OVS_ACTION_ATTR_CLONE] = (u32)-1,
3175 			[OVS_ACTION_ATTR_CHECK_PKT_LEN] = (u32)-1,
3176 			[OVS_ACTION_ATTR_ADD_MPLS] = sizeof(struct ovs_action_add_mpls),
3177 			[OVS_ACTION_ATTR_DEC_TTL] = (u32)-1,
3178 			[OVS_ACTION_ATTR_DROP] = sizeof(u32),
3179 			[OVS_ACTION_ATTR_PSAMPLE] = (u32)-1,
3180 		};
3181 		const struct ovs_action_push_vlan *vlan;
3182 		int type = nla_type(a);
3183 		bool skip_copy;
3184 
3185 		if (type > OVS_ACTION_ATTR_MAX ||
3186 		    (action_lens[type] != nla_len(a) &&
3187 		     action_lens[type] != (u32)-1))
3188 			return -EINVAL;
3189 
3190 		skip_copy = false;
3191 		switch (type) {
3192 		case OVS_ACTION_ATTR_UNSPEC:
3193 			return -EINVAL;
3194 
3195 		case OVS_ACTION_ATTR_USERSPACE:
3196 			err = validate_userspace(a);
3197 			if (err)
3198 				return err;
3199 			break;
3200 
3201 		case OVS_ACTION_ATTR_OUTPUT:
3202 			if (nla_get_u32(a) >= DP_MAX_PORTS)
3203 				return -EINVAL;
3204 			break;
3205 
3206 		case OVS_ACTION_ATTR_TRUNC: {
3207 			const struct ovs_action_trunc *trunc = nla_data(a);
3208 
3209 			if (trunc->max_len < ETH_HLEN)
3210 				return -EINVAL;
3211 			break;
3212 		}
3213 
3214 		case OVS_ACTION_ATTR_HASH: {
3215 			const struct ovs_action_hash *act_hash = nla_data(a);
3216 
3217 			switch (act_hash->hash_alg) {
3218 			case OVS_HASH_ALG_L4:
3219 				fallthrough;
3220 			case OVS_HASH_ALG_SYM_L4:
3221 				break;
3222 			default:
3223 				return  -EINVAL;
3224 			}
3225 
3226 			break;
3227 		}
3228 
3229 		case OVS_ACTION_ATTR_POP_VLAN:
3230 			if (mac_proto != MAC_PROTO_ETHERNET)
3231 				return -EINVAL;
3232 			vlan_tci = htons(0);
3233 			break;
3234 
3235 		case OVS_ACTION_ATTR_PUSH_VLAN:
3236 			if (mac_proto != MAC_PROTO_ETHERNET)
3237 				return -EINVAL;
3238 			vlan = nla_data(a);
3239 			if (!eth_type_vlan(vlan->vlan_tpid))
3240 				return -EINVAL;
3241 			if (!(vlan->vlan_tci & htons(VLAN_CFI_MASK)))
3242 				return -EINVAL;
3243 			vlan_tci = vlan->vlan_tci;
3244 			break;
3245 
3246 		case OVS_ACTION_ATTR_RECIRC:
3247 			break;
3248 
3249 		case OVS_ACTION_ATTR_ADD_MPLS: {
3250 			const struct ovs_action_add_mpls *mpls = nla_data(a);
3251 
3252 			if (!eth_p_mpls(mpls->mpls_ethertype))
3253 				return -EINVAL;
3254 
3255 			if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK) {
3256 				if (vlan_tci & htons(VLAN_CFI_MASK) ||
3257 				    (eth_type != htons(ETH_P_IP) &&
3258 				     eth_type != htons(ETH_P_IPV6) &&
3259 				     eth_type != htons(ETH_P_ARP) &&
3260 				     eth_type != htons(ETH_P_RARP) &&
3261 				     !eth_p_mpls(eth_type)))
3262 					return -EINVAL;
3263 				mpls_label_count++;
3264 			} else {
3265 				if (mac_proto == MAC_PROTO_ETHERNET) {
3266 					mpls_label_count = 1;
3267 					mac_proto = MAC_PROTO_NONE;
3268 				} else {
3269 					mpls_label_count++;
3270 				}
3271 			}
3272 			eth_type = mpls->mpls_ethertype;
3273 			break;
3274 		}
3275 
3276 		case OVS_ACTION_ATTR_PUSH_MPLS: {
3277 			const struct ovs_action_push_mpls *mpls = nla_data(a);
3278 
3279 			if (!eth_p_mpls(mpls->mpls_ethertype))
3280 				return -EINVAL;
3281 			/* Prohibit push MPLS other than to a white list
3282 			 * for packets that have a known tag order.
3283 			 */
3284 			if (vlan_tci & htons(VLAN_CFI_MASK) ||
3285 			    (eth_type != htons(ETH_P_IP) &&
3286 			     eth_type != htons(ETH_P_IPV6) &&
3287 			     eth_type != htons(ETH_P_ARP) &&
3288 			     eth_type != htons(ETH_P_RARP) &&
3289 			     !eth_p_mpls(eth_type)))
3290 				return -EINVAL;
3291 			eth_type = mpls->mpls_ethertype;
3292 			mpls_label_count++;
3293 			break;
3294 		}
3295 
3296 		case OVS_ACTION_ATTR_POP_MPLS: {
3297 			__be16  proto;
3298 			if (vlan_tci & htons(VLAN_CFI_MASK) ||
3299 			    !eth_p_mpls(eth_type))
3300 				return -EINVAL;
3301 
3302 			/* Disallow subsequent L2.5+ set actions and mpls_pop
3303 			 * actions once the last MPLS label in the packet is
3304 			 * popped as there is no check here to ensure that
3305 			 * the new eth type is valid and thus set actions could
3306 			 * write off the end of the packet or otherwise corrupt
3307 			 * it.
3308 			 *
3309 			 * Support for these actions is planned using packet
3310 			 * recirculation.
3311 			 */
3312 			proto = nla_get_be16(a);
3313 
3314 			if (proto == htons(ETH_P_TEB) &&
3315 			    mac_proto != MAC_PROTO_NONE)
3316 				return -EINVAL;
3317 
3318 			mpls_label_count--;
3319 
3320 			if (!eth_p_mpls(proto) || !mpls_label_count)
3321 				eth_type = htons(0);
3322 			else
3323 				eth_type =  proto;
3324 
3325 			break;
3326 		}
3327 
3328 		case OVS_ACTION_ATTR_SET:
3329 			err = validate_set(a, key, sfa,
3330 					   &skip_copy, mac_proto, eth_type,
3331 					   false, log);
3332 			if (err)
3333 				return err;
3334 			break;
3335 
3336 		case OVS_ACTION_ATTR_SET_MASKED:
3337 			err = validate_set(a, key, sfa,
3338 					   &skip_copy, mac_proto, eth_type,
3339 					   true, log);
3340 			if (err)
3341 				return err;
3342 			break;
3343 
3344 		case OVS_ACTION_ATTR_SAMPLE: {
3345 			bool last = nla_is_last(a, rem);
3346 
3347 			err = validate_and_copy_sample(net, a, key, sfa,
3348 						       eth_type, vlan_tci,
3349 						       mpls_label_count,
3350 						       log, last, depth);
3351 			if (err)
3352 				return err;
3353 			skip_copy = true;
3354 			break;
3355 		}
3356 
3357 		case OVS_ACTION_ATTR_CT:
3358 			err = ovs_ct_copy_action(net, a, key, sfa, log);
3359 			if (err)
3360 				return err;
3361 			skip_copy = true;
3362 			break;
3363 
3364 		case OVS_ACTION_ATTR_CT_CLEAR:
3365 			break;
3366 
3367 		case OVS_ACTION_ATTR_PUSH_ETH:
3368 			/* Disallow pushing an Ethernet header if one
3369 			 * is already present */
3370 			if (mac_proto != MAC_PROTO_NONE)
3371 				return -EINVAL;
3372 			mac_proto = MAC_PROTO_ETHERNET;
3373 			break;
3374 
3375 		case OVS_ACTION_ATTR_POP_ETH:
3376 			if (mac_proto != MAC_PROTO_ETHERNET)
3377 				return -EINVAL;
3378 			if (vlan_tci & htons(VLAN_CFI_MASK))
3379 				return -EINVAL;
3380 			mac_proto = MAC_PROTO_NONE;
3381 			break;
3382 
3383 		case OVS_ACTION_ATTR_PUSH_NSH:
3384 			if (mac_proto != MAC_PROTO_ETHERNET) {
3385 				u8 next_proto;
3386 
3387 				next_proto = tun_p_from_eth_p(eth_type);
3388 				if (!next_proto)
3389 					return -EINVAL;
3390 			}
3391 			mac_proto = MAC_PROTO_NONE;
3392 			if (!validate_push_nsh(nla_data(a), log))
3393 				return -EINVAL;
3394 			break;
3395 
3396 		case OVS_ACTION_ATTR_POP_NSH: {
3397 			__be16 inner_proto;
3398 
3399 			if (eth_type != htons(ETH_P_NSH))
3400 				return -EINVAL;
3401 			inner_proto = tun_p_to_eth_p(key->nsh.base.np);
3402 			if (!inner_proto)
3403 				return -EINVAL;
3404 			if (key->nsh.base.np == TUN_P_ETHERNET)
3405 				mac_proto = MAC_PROTO_ETHERNET;
3406 			else
3407 				mac_proto = MAC_PROTO_NONE;
3408 			break;
3409 		}
3410 
3411 		case OVS_ACTION_ATTR_METER:
3412 			/* Non-existent meters are simply ignored.  */
3413 			break;
3414 
3415 		case OVS_ACTION_ATTR_CLONE: {
3416 			bool last = nla_is_last(a, rem);
3417 
3418 			err = validate_and_copy_clone(net, a, key, sfa,
3419 						      eth_type, vlan_tci,
3420 						      mpls_label_count,
3421 						      log, last, depth);
3422 			if (err)
3423 				return err;
3424 			skip_copy = true;
3425 			break;
3426 		}
3427 
3428 		case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
3429 			bool last = nla_is_last(a, rem);
3430 
3431 			err = validate_and_copy_check_pkt_len(net, a, key, sfa,
3432 							      eth_type,
3433 							      vlan_tci,
3434 							      mpls_label_count,
3435 							      log, last,
3436 							      depth);
3437 			if (err)
3438 				return err;
3439 			skip_copy = true;
3440 			break;
3441 		}
3442 
3443 		case OVS_ACTION_ATTR_DEC_TTL:
3444 			err = validate_and_copy_dec_ttl(net, a, key, sfa,
3445 							eth_type, vlan_tci,
3446 							mpls_label_count, log,
3447 							depth);
3448 			if (err)
3449 				return err;
3450 			skip_copy = true;
3451 			break;
3452 
3453 		case OVS_ACTION_ATTR_DROP:
3454 			if (!nla_is_last(a, rem))
3455 				return -EINVAL;
3456 			break;
3457 
3458 		case OVS_ACTION_ATTR_PSAMPLE:
3459 			err = validate_psample(a);
3460 			if (err)
3461 				return err;
3462 			break;
3463 
3464 		default:
3465 			OVS_NLERR(log, "Unknown Action type %d", type);
3466 			return -EINVAL;
3467 		}
3468 		if (!skip_copy) {
3469 			err = copy_action(a, sfa, log);
3470 			if (err)
3471 				return err;
3472 		}
3473 	}
3474 
3475 	if (rem > 0)
3476 		return -EINVAL;
3477 
3478 	return 0;
3479 }
3480 
3481 /* 'key' must be the masked key. */
3482 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
3483 			 const struct sw_flow_key *key,
3484 			 struct sw_flow_actions **sfa, bool log)
3485 {
3486 	int err;
3487 	u32 mpls_label_count = 0;
3488 
3489 	*sfa = nla_alloc_flow_actions(nla_len(attr));
3490 	if (IS_ERR(*sfa))
3491 		return PTR_ERR(*sfa);
3492 
3493 	if (eth_p_mpls(key->eth.type))
3494 		mpls_label_count = hweight_long(key->mpls.num_labels_mask);
3495 
3496 	(*sfa)->orig_len = nla_len(attr);
3497 	err = __ovs_nla_copy_actions(net, attr, key, sfa, key->eth.type,
3498 				     key->eth.vlan.tci, mpls_label_count, log,
3499 				     0);
3500 	if (err)
3501 		ovs_nla_free_flow_actions(*sfa);
3502 
3503 	return err;
3504 }
3505 
3506 static int sample_action_to_attr(const struct nlattr *attr,
3507 				 struct sk_buff *skb)
3508 {
3509 	struct nlattr *start, *ac_start = NULL, *sample_arg;
3510 	int err = 0, rem = nla_len(attr);
3511 	const struct sample_arg *arg;
3512 	struct nlattr *actions;
3513 
3514 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SAMPLE);
3515 	if (!start)
3516 		return -EMSGSIZE;
3517 
3518 	sample_arg = nla_data(attr);
3519 	arg = nla_data(sample_arg);
3520 	actions = nla_next(sample_arg, &rem);
3521 
3522 	if (nla_put_u32(skb, OVS_SAMPLE_ATTR_PROBABILITY, arg->probability)) {
3523 		err = -EMSGSIZE;
3524 		goto out;
3525 	}
3526 
3527 	ac_start = nla_nest_start_noflag(skb, OVS_SAMPLE_ATTR_ACTIONS);
3528 	if (!ac_start) {
3529 		err = -EMSGSIZE;
3530 		goto out;
3531 	}
3532 
3533 	err = ovs_nla_put_actions(actions, rem, skb);
3534 
3535 out:
3536 	if (err) {
3537 		nla_nest_cancel(skb, ac_start);
3538 		nla_nest_cancel(skb, start);
3539 	} else {
3540 		nla_nest_end(skb, ac_start);
3541 		nla_nest_end(skb, start);
3542 	}
3543 
3544 	return err;
3545 }
3546 
3547 static int clone_action_to_attr(const struct nlattr *attr,
3548 				struct sk_buff *skb)
3549 {
3550 	struct nlattr *start;
3551 	int err = 0, rem = nla_len(attr);
3552 
3553 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CLONE);
3554 	if (!start)
3555 		return -EMSGSIZE;
3556 
3557 	/* Skipping the OVS_CLONE_ATTR_EXEC that is always the first attribute. */
3558 	attr = nla_next(nla_data(attr), &rem);
3559 	err = ovs_nla_put_actions(attr, rem, skb);
3560 
3561 	if (err)
3562 		nla_nest_cancel(skb, start);
3563 	else
3564 		nla_nest_end(skb, start);
3565 
3566 	return err;
3567 }
3568 
3569 static int check_pkt_len_action_to_attr(const struct nlattr *attr,
3570 					struct sk_buff *skb)
3571 {
3572 	struct nlattr *start, *ac_start = NULL;
3573 	const struct check_pkt_len_arg *arg;
3574 	const struct nlattr *a, *cpl_arg;
3575 	int err = 0, rem = nla_len(attr);
3576 
3577 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CHECK_PKT_LEN);
3578 	if (!start)
3579 		return -EMSGSIZE;
3580 
3581 	/* The first nested attribute in 'attr' is always
3582 	 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
3583 	 */
3584 	cpl_arg = nla_data(attr);
3585 	arg = nla_data(cpl_arg);
3586 
3587 	if (nla_put_u16(skb, OVS_CHECK_PKT_LEN_ATTR_PKT_LEN, arg->pkt_len)) {
3588 		err = -EMSGSIZE;
3589 		goto out;
3590 	}
3591 
3592 	/* Second nested attribute in 'attr' is always
3593 	 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
3594 	 */
3595 	a = nla_next(cpl_arg, &rem);
3596 	ac_start =  nla_nest_start_noflag(skb,
3597 					  OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL);
3598 	if (!ac_start) {
3599 		err = -EMSGSIZE;
3600 		goto out;
3601 	}
3602 
3603 	err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3604 	if (err) {
3605 		nla_nest_cancel(skb, ac_start);
3606 		goto out;
3607 	} else {
3608 		nla_nest_end(skb, ac_start);
3609 	}
3610 
3611 	/* Third nested attribute in 'attr' is always
3612 	 * OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER.
3613 	 */
3614 	a = nla_next(a, &rem);
3615 	ac_start =  nla_nest_start_noflag(skb,
3616 					  OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER);
3617 	if (!ac_start) {
3618 		err = -EMSGSIZE;
3619 		goto out;
3620 	}
3621 
3622 	err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3623 	if (err) {
3624 		nla_nest_cancel(skb, ac_start);
3625 		goto out;
3626 	} else {
3627 		nla_nest_end(skb, ac_start);
3628 	}
3629 
3630 	nla_nest_end(skb, start);
3631 	return 0;
3632 
3633 out:
3634 	nla_nest_cancel(skb, start);
3635 	return err;
3636 }
3637 
3638 static int dec_ttl_action_to_attr(const struct nlattr *attr,
3639 				  struct sk_buff *skb)
3640 {
3641 	struct nlattr *start, *action_start;
3642 	const struct nlattr *a;
3643 	int err = 0, rem;
3644 
3645 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_DEC_TTL);
3646 	if (!start)
3647 		return -EMSGSIZE;
3648 
3649 	nla_for_each_attr(a, nla_data(attr), nla_len(attr), rem) {
3650 		switch (nla_type(a)) {
3651 		case OVS_DEC_TTL_ATTR_ACTION:
3652 
3653 			action_start = nla_nest_start_noflag(skb, OVS_DEC_TTL_ATTR_ACTION);
3654 			if (!action_start) {
3655 				err = -EMSGSIZE;
3656 				goto out;
3657 			}
3658 
3659 			err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
3660 			if (err)
3661 				goto out;
3662 
3663 			nla_nest_end(skb, action_start);
3664 			break;
3665 
3666 		default:
3667 			/* Ignore all other option to be future compatible */
3668 			break;
3669 		}
3670 	}
3671 
3672 	nla_nest_end(skb, start);
3673 	return 0;
3674 
3675 out:
3676 	nla_nest_cancel(skb, start);
3677 	return err;
3678 }
3679 
3680 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
3681 {
3682 	const struct nlattr *ovs_key = nla_data(a);
3683 	int key_type = nla_type(ovs_key);
3684 	struct nlattr *start;
3685 	int err;
3686 
3687 	switch (key_type) {
3688 	case OVS_KEY_ATTR_TUNNEL_INFO: {
3689 		struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
3690 		struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
3691 
3692 		start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SET);
3693 		if (!start)
3694 			return -EMSGSIZE;
3695 
3696 		err =  ip_tun_to_nlattr(skb, &tun_info->key,
3697 					ip_tunnel_info_opts(tun_info),
3698 					tun_info->options_len,
3699 					ip_tunnel_info_af(tun_info), tun_info->mode);
3700 		if (err)
3701 			return err;
3702 		nla_nest_end(skb, start);
3703 		break;
3704 	}
3705 	default:
3706 		if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
3707 			return -EMSGSIZE;
3708 		break;
3709 	}
3710 
3711 	return 0;
3712 }
3713 
3714 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
3715 						struct sk_buff *skb)
3716 {
3717 	const struct nlattr *ovs_key = nla_data(a);
3718 	struct nlattr *nla;
3719 	size_t key_len = nla_len(ovs_key) / 2;
3720 
3721 	/* Revert the conversion we did from a non-masked set action to
3722 	 * masked set action.
3723 	 */
3724 	nla = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_SET);
3725 	if (!nla)
3726 		return -EMSGSIZE;
3727 
3728 	if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
3729 		return -EMSGSIZE;
3730 
3731 	nla_nest_end(skb, nla);
3732 	return 0;
3733 }
3734 
3735 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
3736 {
3737 	const struct nlattr *a;
3738 	int rem, err;
3739 
3740 	nla_for_each_attr(a, attr, len, rem) {
3741 		int type = nla_type(a);
3742 
3743 		switch (type) {
3744 		case OVS_ACTION_ATTR_SET:
3745 			err = set_action_to_attr(a, skb);
3746 			if (err)
3747 				return err;
3748 			break;
3749 
3750 		case OVS_ACTION_ATTR_SET_TO_MASKED:
3751 			err = masked_set_action_to_set_action_attr(a, skb);
3752 			if (err)
3753 				return err;
3754 			break;
3755 
3756 		case OVS_ACTION_ATTR_SAMPLE:
3757 			err = sample_action_to_attr(a, skb);
3758 			if (err)
3759 				return err;
3760 			break;
3761 
3762 		case OVS_ACTION_ATTR_CT:
3763 			err = ovs_ct_action_to_attr(nla_data(a), skb);
3764 			if (err)
3765 				return err;
3766 			break;
3767 
3768 		case OVS_ACTION_ATTR_CLONE:
3769 			err = clone_action_to_attr(a, skb);
3770 			if (err)
3771 				return err;
3772 			break;
3773 
3774 		case OVS_ACTION_ATTR_CHECK_PKT_LEN:
3775 			err = check_pkt_len_action_to_attr(a, skb);
3776 			if (err)
3777 				return err;
3778 			break;
3779 
3780 		case OVS_ACTION_ATTR_DEC_TTL:
3781 			err = dec_ttl_action_to_attr(a, skb);
3782 			if (err)
3783 				return err;
3784 			break;
3785 
3786 		default:
3787 			if (nla_put(skb, type, nla_len(a), nla_data(a)))
3788 				return -EMSGSIZE;
3789 			break;
3790 		}
3791 	}
3792 
3793 	return 0;
3794 }
3795