1 /*
2  * Copyright (c) 2007-2014 Nicira, Inc.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16  * 02110-1301, USA
17  */
18 
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 
21 #include "flow.h"
22 #include "datapath.h"
23 #include <linux/uaccess.h>
24 #include <linux/netdevice.h>
25 #include <linux/etherdevice.h>
26 #include <linux/if_ether.h>
27 #include <linux/if_vlan.h>
28 #include <net/llc_pdu.h>
29 #include <linux/kernel.h>
30 #include <linux/jhash.h>
31 #include <linux/jiffies.h>
32 #include <linux/llc.h>
33 #include <linux/module.h>
34 #include <linux/in.h>
35 #include <linux/rcupdate.h>
36 #include <linux/if_arp.h>
37 #include <linux/ip.h>
38 #include <linux/ipv6.h>
39 #include <linux/sctp.h>
40 #include <linux/tcp.h>
41 #include <linux/udp.h>
42 #include <linux/icmp.h>
43 #include <linux/icmpv6.h>
44 #include <linux/rculist.h>
45 #include <net/geneve.h>
46 #include <net/ip.h>
47 #include <net/ipv6.h>
48 #include <net/ndisc.h>
49 #include <net/mpls.h>
50 #include <net/vxlan.h>
51 
52 #include "flow_netlink.h"
53 
54 struct ovs_len_tbl {
55 	int len;
56 	const struct ovs_len_tbl *next;
57 };
58 
59 #define OVS_ATTR_NESTED -1
60 
61 static void update_range(struct sw_flow_match *match,
62 			 size_t offset, size_t size, bool is_mask)
63 {
64 	struct sw_flow_key_range *range;
65 	size_t start = rounddown(offset, sizeof(long));
66 	size_t end = roundup(offset + size, sizeof(long));
67 
68 	if (!is_mask)
69 		range = &match->range;
70 	else
71 		range = &match->mask->range;
72 
73 	if (range->start == range->end) {
74 		range->start = start;
75 		range->end = end;
76 		return;
77 	}
78 
79 	if (range->start > start)
80 		range->start = start;
81 
82 	if (range->end < end)
83 		range->end = end;
84 }
85 
86 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
87 	do { \
88 		update_range(match, offsetof(struct sw_flow_key, field),    \
89 			     sizeof((match)->key->field), is_mask);	    \
90 		if (is_mask)						    \
91 			(match)->mask->key.field = value;		    \
92 		else							    \
93 			(match)->key->field = value;		            \
94 	} while (0)
95 
96 #define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask)	    \
97 	do {								    \
98 		update_range(match, offset, len, is_mask);		    \
99 		if (is_mask)						    \
100 			memcpy((u8 *)&(match)->mask->key + offset, value_p, \
101 			       len);					   \
102 		else							    \
103 			memcpy((u8 *)(match)->key + offset, value_p, len);  \
104 	} while (0)
105 
106 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask)		      \
107 	SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
108 				  value_p, len, is_mask)
109 
110 #define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask)		    \
111 	do {								    \
112 		update_range(match, offsetof(struct sw_flow_key, field),    \
113 			     sizeof((match)->key->field), is_mask);	    \
114 		if (is_mask)						    \
115 			memset((u8 *)&(match)->mask->key.field, value,      \
116 			       sizeof((match)->mask->key.field));	    \
117 		else							    \
118 			memset((u8 *)&(match)->key->field, value,           \
119 			       sizeof((match)->key->field));                \
120 	} while (0)
121 
122 static bool match_validate(const struct sw_flow_match *match,
123 			   u64 key_attrs, u64 mask_attrs, bool log)
124 {
125 	u64 key_expected = 1 << OVS_KEY_ATTR_ETHERNET;
126 	u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
127 
128 	/* The following mask attributes allowed only if they
129 	 * pass the validation tests. */
130 	mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
131 			| (1 << OVS_KEY_ATTR_IPV6)
132 			| (1 << OVS_KEY_ATTR_TCP)
133 			| (1 << OVS_KEY_ATTR_TCP_FLAGS)
134 			| (1 << OVS_KEY_ATTR_UDP)
135 			| (1 << OVS_KEY_ATTR_SCTP)
136 			| (1 << OVS_KEY_ATTR_ICMP)
137 			| (1 << OVS_KEY_ATTR_ICMPV6)
138 			| (1 << OVS_KEY_ATTR_ARP)
139 			| (1 << OVS_KEY_ATTR_ND)
140 			| (1 << OVS_KEY_ATTR_MPLS));
141 
142 	/* Always allowed mask fields. */
143 	mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
144 		       | (1 << OVS_KEY_ATTR_IN_PORT)
145 		       | (1 << OVS_KEY_ATTR_ETHERTYPE));
146 
147 	/* Check key attributes. */
148 	if (match->key->eth.type == htons(ETH_P_ARP)
149 			|| match->key->eth.type == htons(ETH_P_RARP)) {
150 		key_expected |= 1 << OVS_KEY_ATTR_ARP;
151 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
152 			mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
153 	}
154 
155 	if (eth_p_mpls(match->key->eth.type)) {
156 		key_expected |= 1 << OVS_KEY_ATTR_MPLS;
157 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
158 			mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
159 	}
160 
161 	if (match->key->eth.type == htons(ETH_P_IP)) {
162 		key_expected |= 1 << OVS_KEY_ATTR_IPV4;
163 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
164 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
165 
166 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
167 			if (match->key->ip.proto == IPPROTO_UDP) {
168 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
169 				if (match->mask && (match->mask->key.ip.proto == 0xff))
170 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
171 			}
172 
173 			if (match->key->ip.proto == IPPROTO_SCTP) {
174 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
175 				if (match->mask && (match->mask->key.ip.proto == 0xff))
176 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
177 			}
178 
179 			if (match->key->ip.proto == IPPROTO_TCP) {
180 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
181 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
182 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
183 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
184 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
185 				}
186 			}
187 
188 			if (match->key->ip.proto == IPPROTO_ICMP) {
189 				key_expected |= 1 << OVS_KEY_ATTR_ICMP;
190 				if (match->mask && (match->mask->key.ip.proto == 0xff))
191 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
192 			}
193 		}
194 	}
195 
196 	if (match->key->eth.type == htons(ETH_P_IPV6)) {
197 		key_expected |= 1 << OVS_KEY_ATTR_IPV6;
198 		if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
199 			mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
200 
201 		if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
202 			if (match->key->ip.proto == IPPROTO_UDP) {
203 				key_expected |= 1 << OVS_KEY_ATTR_UDP;
204 				if (match->mask && (match->mask->key.ip.proto == 0xff))
205 					mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
206 			}
207 
208 			if (match->key->ip.proto == IPPROTO_SCTP) {
209 				key_expected |= 1 << OVS_KEY_ATTR_SCTP;
210 				if (match->mask && (match->mask->key.ip.proto == 0xff))
211 					mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
212 			}
213 
214 			if (match->key->ip.proto == IPPROTO_TCP) {
215 				key_expected |= 1 << OVS_KEY_ATTR_TCP;
216 				key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
217 				if (match->mask && (match->mask->key.ip.proto == 0xff)) {
218 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
219 					mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
220 				}
221 			}
222 
223 			if (match->key->ip.proto == IPPROTO_ICMPV6) {
224 				key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
225 				if (match->mask && (match->mask->key.ip.proto == 0xff))
226 					mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
227 
228 				if (match->key->tp.src ==
229 						htons(NDISC_NEIGHBOUR_SOLICITATION) ||
230 				    match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
231 					key_expected |= 1 << OVS_KEY_ATTR_ND;
232 					if (match->mask && (match->mask->key.tp.src == htons(0xff)))
233 						mask_allowed |= 1 << OVS_KEY_ATTR_ND;
234 				}
235 			}
236 		}
237 	}
238 
239 	if ((key_attrs & key_expected) != key_expected) {
240 		/* Key attributes check failed. */
241 		OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
242 			  (unsigned long long)key_attrs,
243 			  (unsigned long long)key_expected);
244 		return false;
245 	}
246 
247 	if ((mask_attrs & mask_allowed) != mask_attrs) {
248 		/* Mask attributes check failed. */
249 		OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
250 			  (unsigned long long)mask_attrs,
251 			  (unsigned long long)mask_allowed);
252 		return false;
253 	}
254 
255 	return true;
256 }
257 
258 size_t ovs_tun_key_attr_size(void)
259 {
260 	/* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
261 	 * updating this function.
262 	 */
263 	return    nla_total_size(8)    /* OVS_TUNNEL_KEY_ATTR_ID */
264 		+ nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
265 		+ nla_total_size(4)    /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
266 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TOS */
267 		+ nla_total_size(1)    /* OVS_TUNNEL_KEY_ATTR_TTL */
268 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
269 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_CSUM */
270 		+ nla_total_size(0)    /* OVS_TUNNEL_KEY_ATTR_OAM */
271 		+ nla_total_size(256)  /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
272 		/* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS is mutually exclusive with
273 		 * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
274 		 */
275 		+ nla_total_size(2)    /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
276 		+ nla_total_size(2);   /* OVS_TUNNEL_KEY_ATTR_TP_DST */
277 }
278 
279 size_t ovs_key_attr_size(void)
280 {
281 	/* Whenever adding new OVS_KEY_ FIELDS, we should consider
282 	 * updating this function.
283 	 */
284 	BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 26);
285 
286 	return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
287 		+ nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
288 		  + ovs_tun_key_attr_size()
289 		+ nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
290 		+ nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
291 		+ nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
292 		+ nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
293 		+ nla_total_size(1)   /* OVS_KEY_ATTR_CT_STATE */
294 		+ nla_total_size(2)   /* OVS_KEY_ATTR_CT_ZONE */
295 		+ nla_total_size(4)   /* OVS_KEY_ATTR_CT_MARK */
296 		+ nla_total_size(16)  /* OVS_KEY_ATTR_CT_LABEL */
297 		+ nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
298 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
299 		+ nla_total_size(4)   /* OVS_KEY_ATTR_VLAN */
300 		+ nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
301 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
302 		+ nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
303 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
304 		+ nla_total_size(28); /* OVS_KEY_ATTR_ND */
305 }
306 
307 static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
308 	[OVS_TUNNEL_KEY_ATTR_ID]	    = { .len = sizeof(u64) },
309 	[OVS_TUNNEL_KEY_ATTR_IPV4_SRC]	    = { .len = sizeof(u32) },
310 	[OVS_TUNNEL_KEY_ATTR_IPV4_DST]	    = { .len = sizeof(u32) },
311 	[OVS_TUNNEL_KEY_ATTR_TOS]	    = { .len = 1 },
312 	[OVS_TUNNEL_KEY_ATTR_TTL]	    = { .len = 1 },
313 	[OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
314 	[OVS_TUNNEL_KEY_ATTR_CSUM]	    = { .len = 0 },
315 	[OVS_TUNNEL_KEY_ATTR_TP_SRC]	    = { .len = sizeof(u16) },
316 	[OVS_TUNNEL_KEY_ATTR_TP_DST]	    = { .len = sizeof(u16) },
317 	[OVS_TUNNEL_KEY_ATTR_OAM]	    = { .len = 0 },
318 	[OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS]   = { .len = OVS_ATTR_NESTED },
319 	[OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS]    = { .len = OVS_ATTR_NESTED },
320 };
321 
322 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
323 static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
324 	[OVS_KEY_ATTR_ENCAP]	 = { .len = OVS_ATTR_NESTED },
325 	[OVS_KEY_ATTR_PRIORITY]	 = { .len = sizeof(u32) },
326 	[OVS_KEY_ATTR_IN_PORT]	 = { .len = sizeof(u32) },
327 	[OVS_KEY_ATTR_SKB_MARK]	 = { .len = sizeof(u32) },
328 	[OVS_KEY_ATTR_ETHERNET]	 = { .len = sizeof(struct ovs_key_ethernet) },
329 	[OVS_KEY_ATTR_VLAN]	 = { .len = sizeof(__be16) },
330 	[OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
331 	[OVS_KEY_ATTR_IPV4]	 = { .len = sizeof(struct ovs_key_ipv4) },
332 	[OVS_KEY_ATTR_IPV6]	 = { .len = sizeof(struct ovs_key_ipv6) },
333 	[OVS_KEY_ATTR_TCP]	 = { .len = sizeof(struct ovs_key_tcp) },
334 	[OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
335 	[OVS_KEY_ATTR_UDP]	 = { .len = sizeof(struct ovs_key_udp) },
336 	[OVS_KEY_ATTR_SCTP]	 = { .len = sizeof(struct ovs_key_sctp) },
337 	[OVS_KEY_ATTR_ICMP]	 = { .len = sizeof(struct ovs_key_icmp) },
338 	[OVS_KEY_ATTR_ICMPV6]	 = { .len = sizeof(struct ovs_key_icmpv6) },
339 	[OVS_KEY_ATTR_ARP]	 = { .len = sizeof(struct ovs_key_arp) },
340 	[OVS_KEY_ATTR_ND]	 = { .len = sizeof(struct ovs_key_nd) },
341 	[OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
342 	[OVS_KEY_ATTR_DP_HASH]	 = { .len = sizeof(u32) },
343 	[OVS_KEY_ATTR_TUNNEL]	 = { .len = OVS_ATTR_NESTED,
344 				     .next = ovs_tunnel_key_lens, },
345 	[OVS_KEY_ATTR_MPLS]	 = { .len = sizeof(struct ovs_key_mpls) },
346 	[OVS_KEY_ATTR_CT_STATE]	 = { .len = sizeof(u8) },
347 	[OVS_KEY_ATTR_CT_ZONE]	 = { .len = sizeof(u16) },
348 	[OVS_KEY_ATTR_CT_MARK]	 = { .len = sizeof(u32) },
349 	[OVS_KEY_ATTR_CT_LABEL]	 = { .len = sizeof(struct ovs_key_ct_label) },
350 };
351 
352 static bool is_all_zero(const u8 *fp, size_t size)
353 {
354 	int i;
355 
356 	if (!fp)
357 		return false;
358 
359 	for (i = 0; i < size; i++)
360 		if (fp[i])
361 			return false;
362 
363 	return true;
364 }
365 
366 static int __parse_flow_nlattrs(const struct nlattr *attr,
367 				const struct nlattr *a[],
368 				u64 *attrsp, bool log, bool nz)
369 {
370 	const struct nlattr *nla;
371 	u64 attrs;
372 	int rem;
373 
374 	attrs = *attrsp;
375 	nla_for_each_nested(nla, attr, rem) {
376 		u16 type = nla_type(nla);
377 		int expected_len;
378 
379 		if (type > OVS_KEY_ATTR_MAX) {
380 			OVS_NLERR(log, "Key type %d is out of range max %d",
381 				  type, OVS_KEY_ATTR_MAX);
382 			return -EINVAL;
383 		}
384 
385 		if (attrs & (1 << type)) {
386 			OVS_NLERR(log, "Duplicate key (type %d).", type);
387 			return -EINVAL;
388 		}
389 
390 		expected_len = ovs_key_lens[type].len;
391 		if (nla_len(nla) != expected_len && expected_len != OVS_ATTR_NESTED) {
392 			OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
393 				  type, nla_len(nla), expected_len);
394 			return -EINVAL;
395 		}
396 
397 		if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
398 			attrs |= 1 << type;
399 			a[type] = nla;
400 		}
401 	}
402 	if (rem) {
403 		OVS_NLERR(log, "Message has %d unknown bytes.", rem);
404 		return -EINVAL;
405 	}
406 
407 	*attrsp = attrs;
408 	return 0;
409 }
410 
411 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
412 				   const struct nlattr *a[], u64 *attrsp,
413 				   bool log)
414 {
415 	return __parse_flow_nlattrs(attr, a, attrsp, log, true);
416 }
417 
418 static int parse_flow_nlattrs(const struct nlattr *attr,
419 			      const struct nlattr *a[], u64 *attrsp,
420 			      bool log)
421 {
422 	return __parse_flow_nlattrs(attr, a, attrsp, log, false);
423 }
424 
425 static int genev_tun_opt_from_nlattr(const struct nlattr *a,
426 				     struct sw_flow_match *match, bool is_mask,
427 				     bool log)
428 {
429 	unsigned long opt_key_offset;
430 
431 	if (nla_len(a) > sizeof(match->key->tun_opts)) {
432 		OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
433 			  nla_len(a), sizeof(match->key->tun_opts));
434 		return -EINVAL;
435 	}
436 
437 	if (nla_len(a) % 4 != 0) {
438 		OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
439 			  nla_len(a));
440 		return -EINVAL;
441 	}
442 
443 	/* We need to record the length of the options passed
444 	 * down, otherwise packets with the same format but
445 	 * additional options will be silently matched.
446 	 */
447 	if (!is_mask) {
448 		SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
449 				false);
450 	} else {
451 		/* This is somewhat unusual because it looks at
452 		 * both the key and mask while parsing the
453 		 * attributes (and by extension assumes the key
454 		 * is parsed first). Normally, we would verify
455 		 * that each is the correct length and that the
456 		 * attributes line up in the validate function.
457 		 * However, that is difficult because this is
458 		 * variable length and we won't have the
459 		 * information later.
460 		 */
461 		if (match->key->tun_opts_len != nla_len(a)) {
462 			OVS_NLERR(log, "Geneve option len %d != mask len %d",
463 				  match->key->tun_opts_len, nla_len(a));
464 			return -EINVAL;
465 		}
466 
467 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
468 	}
469 
470 	opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
471 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
472 				  nla_len(a), is_mask);
473 	return 0;
474 }
475 
476 static const struct nla_policy vxlan_opt_policy[OVS_VXLAN_EXT_MAX + 1] = {
477 	[OVS_VXLAN_EXT_GBP]	= { .type = NLA_U32 },
478 };
479 
480 static int vxlan_tun_opt_from_nlattr(const struct nlattr *a,
481 				     struct sw_flow_match *match, bool is_mask,
482 				     bool log)
483 {
484 	struct nlattr *tb[OVS_VXLAN_EXT_MAX+1];
485 	unsigned long opt_key_offset;
486 	struct vxlan_metadata opts;
487 	int err;
488 
489 	BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
490 
491 	err = nla_parse_nested(tb, OVS_VXLAN_EXT_MAX, a, vxlan_opt_policy);
492 	if (err < 0)
493 		return err;
494 
495 	memset(&opts, 0, sizeof(opts));
496 
497 	if (tb[OVS_VXLAN_EXT_GBP])
498 		opts.gbp = nla_get_u32(tb[OVS_VXLAN_EXT_GBP]);
499 
500 	if (!is_mask)
501 		SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
502 	else
503 		SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
504 
505 	opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
506 	SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
507 				  is_mask);
508 	return 0;
509 }
510 
511 static int ipv4_tun_from_nlattr(const struct nlattr *attr,
512 				struct sw_flow_match *match, bool is_mask,
513 				bool log)
514 {
515 	struct nlattr *a;
516 	int rem;
517 	bool ttl = false;
518 	__be16 tun_flags = 0;
519 	int opts_type = 0;
520 
521 	nla_for_each_nested(a, attr, rem) {
522 		int type = nla_type(a);
523 		int err;
524 
525 		if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
526 			OVS_NLERR(log, "Tunnel attr %d out of range max %d",
527 				  type, OVS_TUNNEL_KEY_ATTR_MAX);
528 			return -EINVAL;
529 		}
530 
531 		if (ovs_tunnel_key_lens[type].len != nla_len(a) &&
532 		    ovs_tunnel_key_lens[type].len != OVS_ATTR_NESTED) {
533 			OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
534 				  type, nla_len(a), ovs_tunnel_key_lens[type].len);
535 			return -EINVAL;
536 		}
537 
538 		switch (type) {
539 		case OVS_TUNNEL_KEY_ATTR_ID:
540 			SW_FLOW_KEY_PUT(match, tun_key.tun_id,
541 					nla_get_be64(a), is_mask);
542 			tun_flags |= TUNNEL_KEY;
543 			break;
544 		case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
545 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
546 					nla_get_in_addr(a), is_mask);
547 			break;
548 		case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
549 			SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
550 					nla_get_in_addr(a), is_mask);
551 			break;
552 		case OVS_TUNNEL_KEY_ATTR_TOS:
553 			SW_FLOW_KEY_PUT(match, tun_key.tos,
554 					nla_get_u8(a), is_mask);
555 			break;
556 		case OVS_TUNNEL_KEY_ATTR_TTL:
557 			SW_FLOW_KEY_PUT(match, tun_key.ttl,
558 					nla_get_u8(a), is_mask);
559 			ttl = true;
560 			break;
561 		case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
562 			tun_flags |= TUNNEL_DONT_FRAGMENT;
563 			break;
564 		case OVS_TUNNEL_KEY_ATTR_CSUM:
565 			tun_flags |= TUNNEL_CSUM;
566 			break;
567 		case OVS_TUNNEL_KEY_ATTR_TP_SRC:
568 			SW_FLOW_KEY_PUT(match, tun_key.tp_src,
569 					nla_get_be16(a), is_mask);
570 			break;
571 		case OVS_TUNNEL_KEY_ATTR_TP_DST:
572 			SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
573 					nla_get_be16(a), is_mask);
574 			break;
575 		case OVS_TUNNEL_KEY_ATTR_OAM:
576 			tun_flags |= TUNNEL_OAM;
577 			break;
578 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
579 			if (opts_type) {
580 				OVS_NLERR(log, "Multiple metadata blocks provided");
581 				return -EINVAL;
582 			}
583 
584 			err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
585 			if (err)
586 				return err;
587 
588 			tun_flags |= TUNNEL_GENEVE_OPT;
589 			opts_type = type;
590 			break;
591 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
592 			if (opts_type) {
593 				OVS_NLERR(log, "Multiple metadata blocks provided");
594 				return -EINVAL;
595 			}
596 
597 			err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
598 			if (err)
599 				return err;
600 
601 			tun_flags |= TUNNEL_VXLAN_OPT;
602 			opts_type = type;
603 			break;
604 		default:
605 			OVS_NLERR(log, "Unknown IPv4 tunnel attribute %d",
606 				  type);
607 			return -EINVAL;
608 		}
609 	}
610 
611 	SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
612 
613 	if (rem > 0) {
614 		OVS_NLERR(log, "IPv4 tunnel attribute has %d unknown bytes.",
615 			  rem);
616 		return -EINVAL;
617 	}
618 
619 	if (!is_mask) {
620 		if (!match->key->tun_key.u.ipv4.dst) {
621 			OVS_NLERR(log, "IPv4 tunnel dst address is zero");
622 			return -EINVAL;
623 		}
624 
625 		if (!ttl) {
626 			OVS_NLERR(log, "IPv4 tunnel TTL not specified.");
627 			return -EINVAL;
628 		}
629 	}
630 
631 	return opts_type;
632 }
633 
634 static int vxlan_opt_to_nlattr(struct sk_buff *skb,
635 			       const void *tun_opts, int swkey_tun_opts_len)
636 {
637 	const struct vxlan_metadata *opts = tun_opts;
638 	struct nlattr *nla;
639 
640 	nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
641 	if (!nla)
642 		return -EMSGSIZE;
643 
644 	if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
645 		return -EMSGSIZE;
646 
647 	nla_nest_end(skb, nla);
648 	return 0;
649 }
650 
651 static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
652 				const struct ip_tunnel_key *output,
653 				const void *tun_opts, int swkey_tun_opts_len)
654 {
655 	if (output->tun_flags & TUNNEL_KEY &&
656 	    nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
657 		return -EMSGSIZE;
658 	if (output->u.ipv4.src &&
659 	    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
660 			    output->u.ipv4.src))
661 		return -EMSGSIZE;
662 	if (output->u.ipv4.dst &&
663 	    nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
664 			    output->u.ipv4.dst))
665 		return -EMSGSIZE;
666 	if (output->tos &&
667 	    nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
668 		return -EMSGSIZE;
669 	if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
670 		return -EMSGSIZE;
671 	if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
672 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
673 		return -EMSGSIZE;
674 	if ((output->tun_flags & TUNNEL_CSUM) &&
675 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
676 		return -EMSGSIZE;
677 	if (output->tp_src &&
678 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
679 		return -EMSGSIZE;
680 	if (output->tp_dst &&
681 	    nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
682 		return -EMSGSIZE;
683 	if ((output->tun_flags & TUNNEL_OAM) &&
684 	    nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
685 		return -EMSGSIZE;
686 	if (tun_opts) {
687 		if (output->tun_flags & TUNNEL_GENEVE_OPT &&
688 		    nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
689 			    swkey_tun_opts_len, tun_opts))
690 			return -EMSGSIZE;
691 		else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
692 			 vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
693 			return -EMSGSIZE;
694 	}
695 
696 	return 0;
697 }
698 
699 static int ipv4_tun_to_nlattr(struct sk_buff *skb,
700 			      const struct ip_tunnel_key *output,
701 			      const void *tun_opts, int swkey_tun_opts_len)
702 {
703 	struct nlattr *nla;
704 	int err;
705 
706 	nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
707 	if (!nla)
708 		return -EMSGSIZE;
709 
710 	err = __ipv4_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len);
711 	if (err)
712 		return err;
713 
714 	nla_nest_end(skb, nla);
715 	return 0;
716 }
717 
718 int ovs_nla_put_egress_tunnel_key(struct sk_buff *skb,
719 				  const struct ip_tunnel_info *egress_tun_info,
720 				  const void *egress_tun_opts)
721 {
722 	return __ipv4_tun_to_nlattr(skb, &egress_tun_info->key,
723 				    egress_tun_opts,
724 				    egress_tun_info->options_len);
725 }
726 
727 static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
728 				 u64 *attrs, const struct nlattr **a,
729 				 bool is_mask, bool log)
730 {
731 	if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
732 		u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
733 
734 		SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
735 		*attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
736 	}
737 
738 	if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
739 		u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
740 
741 		SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
742 		*attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
743 	}
744 
745 	if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
746 		SW_FLOW_KEY_PUT(match, phy.priority,
747 			  nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
748 		*attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
749 	}
750 
751 	if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
752 		u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
753 
754 		if (is_mask) {
755 			in_port = 0xffffffff; /* Always exact match in_port. */
756 		} else if (in_port >= DP_MAX_PORTS) {
757 			OVS_NLERR(log, "Port %d exceeds max allowable %d",
758 				  in_port, DP_MAX_PORTS);
759 			return -EINVAL;
760 		}
761 
762 		SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
763 		*attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
764 	} else if (!is_mask) {
765 		SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
766 	}
767 
768 	if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
769 		uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
770 
771 		SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
772 		*attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
773 	}
774 	if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
775 		if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
776 					 is_mask, log) < 0)
777 			return -EINVAL;
778 		*attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
779 	}
780 
781 	if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
782 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
783 		u8 ct_state = nla_get_u8(a[OVS_KEY_ATTR_CT_STATE]);
784 
785 		SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
786 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
787 	}
788 	if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
789 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
790 		u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
791 
792 		SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
793 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
794 	}
795 	if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
796 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
797 		u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
798 
799 		SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
800 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
801 	}
802 	if (*attrs & (1 << OVS_KEY_ATTR_CT_LABEL) &&
803 	    ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABEL)) {
804 		const struct ovs_key_ct_label *cl;
805 
806 		cl = nla_data(a[OVS_KEY_ATTR_CT_LABEL]);
807 		SW_FLOW_KEY_MEMCPY(match, ct.label, cl->ct_label,
808 				   sizeof(*cl), is_mask);
809 		*attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABEL);
810 	}
811 	return 0;
812 }
813 
814 static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
815 				u64 attrs, const struct nlattr **a,
816 				bool is_mask, bool log)
817 {
818 	int err;
819 
820 	err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
821 	if (err)
822 		return err;
823 
824 	if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
825 		const struct ovs_key_ethernet *eth_key;
826 
827 		eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
828 		SW_FLOW_KEY_MEMCPY(match, eth.src,
829 				eth_key->eth_src, ETH_ALEN, is_mask);
830 		SW_FLOW_KEY_MEMCPY(match, eth.dst,
831 				eth_key->eth_dst, ETH_ALEN, is_mask);
832 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
833 	}
834 
835 	if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
836 		__be16 tci;
837 
838 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
839 		if (!(tci & htons(VLAN_TAG_PRESENT))) {
840 			if (is_mask)
841 				OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
842 			else
843 				OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
844 
845 			return -EINVAL;
846 		}
847 
848 		SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
849 		attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
850 	}
851 
852 	if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
853 		__be16 eth_type;
854 
855 		eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
856 		if (is_mask) {
857 			/* Always exact match EtherType. */
858 			eth_type = htons(0xffff);
859 		} else if (!eth_proto_is_802_3(eth_type)) {
860 			OVS_NLERR(log, "EtherType %x is less than min %x",
861 				  ntohs(eth_type), ETH_P_802_3_MIN);
862 			return -EINVAL;
863 		}
864 
865 		SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
866 		attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
867 	} else if (!is_mask) {
868 		SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
869 	}
870 
871 	if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
872 		const struct ovs_key_ipv4 *ipv4_key;
873 
874 		ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
875 		if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
876 			OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
877 				  ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
878 			return -EINVAL;
879 		}
880 		SW_FLOW_KEY_PUT(match, ip.proto,
881 				ipv4_key->ipv4_proto, is_mask);
882 		SW_FLOW_KEY_PUT(match, ip.tos,
883 				ipv4_key->ipv4_tos, is_mask);
884 		SW_FLOW_KEY_PUT(match, ip.ttl,
885 				ipv4_key->ipv4_ttl, is_mask);
886 		SW_FLOW_KEY_PUT(match, ip.frag,
887 				ipv4_key->ipv4_frag, is_mask);
888 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
889 				ipv4_key->ipv4_src, is_mask);
890 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
891 				ipv4_key->ipv4_dst, is_mask);
892 		attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
893 	}
894 
895 	if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
896 		const struct ovs_key_ipv6 *ipv6_key;
897 
898 		ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
899 		if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
900 			OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
901 				  ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
902 			return -EINVAL;
903 		}
904 
905 		if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
906 			OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
907 				  ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
908 			return -EINVAL;
909 		}
910 
911 		SW_FLOW_KEY_PUT(match, ipv6.label,
912 				ipv6_key->ipv6_label, is_mask);
913 		SW_FLOW_KEY_PUT(match, ip.proto,
914 				ipv6_key->ipv6_proto, is_mask);
915 		SW_FLOW_KEY_PUT(match, ip.tos,
916 				ipv6_key->ipv6_tclass, is_mask);
917 		SW_FLOW_KEY_PUT(match, ip.ttl,
918 				ipv6_key->ipv6_hlimit, is_mask);
919 		SW_FLOW_KEY_PUT(match, ip.frag,
920 				ipv6_key->ipv6_frag, is_mask);
921 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
922 				ipv6_key->ipv6_src,
923 				sizeof(match->key->ipv6.addr.src),
924 				is_mask);
925 		SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
926 				ipv6_key->ipv6_dst,
927 				sizeof(match->key->ipv6.addr.dst),
928 				is_mask);
929 
930 		attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
931 	}
932 
933 	if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
934 		const struct ovs_key_arp *arp_key;
935 
936 		arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
937 		if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
938 			OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
939 				  arp_key->arp_op);
940 			return -EINVAL;
941 		}
942 
943 		SW_FLOW_KEY_PUT(match, ipv4.addr.src,
944 				arp_key->arp_sip, is_mask);
945 		SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
946 			arp_key->arp_tip, is_mask);
947 		SW_FLOW_KEY_PUT(match, ip.proto,
948 				ntohs(arp_key->arp_op), is_mask);
949 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
950 				arp_key->arp_sha, ETH_ALEN, is_mask);
951 		SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
952 				arp_key->arp_tha, ETH_ALEN, is_mask);
953 
954 		attrs &= ~(1 << OVS_KEY_ATTR_ARP);
955 	}
956 
957 	if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
958 		const struct ovs_key_mpls *mpls_key;
959 
960 		mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
961 		SW_FLOW_KEY_PUT(match, mpls.top_lse,
962 				mpls_key->mpls_lse, is_mask);
963 
964 		attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
965 	 }
966 
967 	if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
968 		const struct ovs_key_tcp *tcp_key;
969 
970 		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
971 		SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
972 		SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
973 		attrs &= ~(1 << OVS_KEY_ATTR_TCP);
974 	}
975 
976 	if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
977 		SW_FLOW_KEY_PUT(match, tp.flags,
978 				nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
979 				is_mask);
980 		attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
981 	}
982 
983 	if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
984 		const struct ovs_key_udp *udp_key;
985 
986 		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
987 		SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
988 		SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
989 		attrs &= ~(1 << OVS_KEY_ATTR_UDP);
990 	}
991 
992 	if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
993 		const struct ovs_key_sctp *sctp_key;
994 
995 		sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
996 		SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
997 		SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
998 		attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
999 	}
1000 
1001 	if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1002 		const struct ovs_key_icmp *icmp_key;
1003 
1004 		icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1005 		SW_FLOW_KEY_PUT(match, tp.src,
1006 				htons(icmp_key->icmp_type), is_mask);
1007 		SW_FLOW_KEY_PUT(match, tp.dst,
1008 				htons(icmp_key->icmp_code), is_mask);
1009 		attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1010 	}
1011 
1012 	if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1013 		const struct ovs_key_icmpv6 *icmpv6_key;
1014 
1015 		icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1016 		SW_FLOW_KEY_PUT(match, tp.src,
1017 				htons(icmpv6_key->icmpv6_type), is_mask);
1018 		SW_FLOW_KEY_PUT(match, tp.dst,
1019 				htons(icmpv6_key->icmpv6_code), is_mask);
1020 		attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1021 	}
1022 
1023 	if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1024 		const struct ovs_key_nd *nd_key;
1025 
1026 		nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1027 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1028 			nd_key->nd_target,
1029 			sizeof(match->key->ipv6.nd.target),
1030 			is_mask);
1031 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1032 			nd_key->nd_sll, ETH_ALEN, is_mask);
1033 		SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1034 				nd_key->nd_tll, ETH_ALEN, is_mask);
1035 		attrs &= ~(1 << OVS_KEY_ATTR_ND);
1036 	}
1037 
1038 	if (attrs != 0) {
1039 		OVS_NLERR(log, "Unknown key attributes %llx",
1040 			  (unsigned long long)attrs);
1041 		return -EINVAL;
1042 	}
1043 
1044 	return 0;
1045 }
1046 
1047 static void nlattr_set(struct nlattr *attr, u8 val,
1048 		       const struct ovs_len_tbl *tbl)
1049 {
1050 	struct nlattr *nla;
1051 	int rem;
1052 
1053 	/* The nlattr stream should already have been validated */
1054 	nla_for_each_nested(nla, attr, rem) {
1055 		if (tbl && tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
1056 			nlattr_set(nla, val, tbl[nla_type(nla)].next);
1057 		else
1058 			memset(nla_data(nla), val, nla_len(nla));
1059 	}
1060 }
1061 
1062 static void mask_set_nlattr(struct nlattr *attr, u8 val)
1063 {
1064 	nlattr_set(attr, val, ovs_key_lens);
1065 }
1066 
1067 /**
1068  * ovs_nla_get_match - parses Netlink attributes into a flow key and
1069  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1070  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1071  * does not include any don't care bit.
1072  * @net: Used to determine per-namespace field support.
1073  * @match: receives the extracted flow match information.
1074  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1075  * sequence. The fields should of the packet that triggered the creation
1076  * of this flow.
1077  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1078  * attribute specifies the mask field of the wildcarded flow.
1079  * @log: Boolean to allow kernel error logging.  Normally true, but when
1080  * probing for feature compatibility this should be passed in as false to
1081  * suppress unnecessary error logging.
1082  */
1083 int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
1084 		      const struct nlattr *nla_key,
1085 		      const struct nlattr *nla_mask,
1086 		      bool log)
1087 {
1088 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1089 	const struct nlattr *encap;
1090 	struct nlattr *newmask = NULL;
1091 	u64 key_attrs = 0;
1092 	u64 mask_attrs = 0;
1093 	bool encap_valid = false;
1094 	int err;
1095 
1096 	err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
1097 	if (err)
1098 		return err;
1099 
1100 	if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1101 	    (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1102 	    (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1103 		__be16 tci;
1104 
1105 		if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1106 		      (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
1107 			OVS_NLERR(log, "Invalid Vlan frame.");
1108 			return -EINVAL;
1109 		}
1110 
1111 		key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1112 		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1113 		encap = a[OVS_KEY_ATTR_ENCAP];
1114 		key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1115 		encap_valid = true;
1116 
1117 		if (tci & htons(VLAN_TAG_PRESENT)) {
1118 			err = parse_flow_nlattrs(encap, a, &key_attrs, log);
1119 			if (err)
1120 				return err;
1121 		} else if (!tci) {
1122 			/* Corner case for truncated 802.1Q header. */
1123 			if (nla_len(encap)) {
1124 				OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
1125 				return -EINVAL;
1126 			}
1127 		} else {
1128 			OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
1129 			return  -EINVAL;
1130 		}
1131 	}
1132 
1133 	err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
1134 	if (err)
1135 		return err;
1136 
1137 	if (match->mask) {
1138 		if (!nla_mask) {
1139 			/* Create an exact match mask. We need to set to 0xff
1140 			 * all the 'match->mask' fields that have been touched
1141 			 * in 'match->key'. We cannot simply memset
1142 			 * 'match->mask', because padding bytes and fields not
1143 			 * specified in 'match->key' should be left to 0.
1144 			 * Instead, we use a stream of netlink attributes,
1145 			 * copied from 'key' and set to 0xff.
1146 			 * ovs_key_from_nlattrs() will take care of filling
1147 			 * 'match->mask' appropriately.
1148 			 */
1149 			newmask = kmemdup(nla_key,
1150 					  nla_total_size(nla_len(nla_key)),
1151 					  GFP_KERNEL);
1152 			if (!newmask)
1153 				return -ENOMEM;
1154 
1155 			mask_set_nlattr(newmask, 0xff);
1156 
1157 			/* The userspace does not send tunnel attributes that
1158 			 * are 0, but we should not wildcard them nonetheless.
1159 			 */
1160 			if (match->key->tun_key.u.ipv4.dst)
1161 				SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1162 							 0xff, true);
1163 
1164 			nla_mask = newmask;
1165 		}
1166 
1167 		err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
1168 		if (err)
1169 			goto free_newmask;
1170 
1171 		/* Always match on tci. */
1172 		SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1173 
1174 		if (mask_attrs & 1 << OVS_KEY_ATTR_ENCAP) {
1175 			__be16 eth_type = 0;
1176 			__be16 tci = 0;
1177 
1178 			if (!encap_valid) {
1179 				OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
1180 				err = -EINVAL;
1181 				goto free_newmask;
1182 			}
1183 
1184 			mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1185 			if (a[OVS_KEY_ATTR_ETHERTYPE])
1186 				eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1187 
1188 			if (eth_type == htons(0xffff)) {
1189 				mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1190 				encap = a[OVS_KEY_ATTR_ENCAP];
1191 				err = parse_flow_mask_nlattrs(encap, a,
1192 							      &mask_attrs, log);
1193 				if (err)
1194 					goto free_newmask;
1195 			} else {
1196 				OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1197 					  ntohs(eth_type));
1198 				err = -EINVAL;
1199 				goto free_newmask;
1200 			}
1201 
1202 			if (a[OVS_KEY_ATTR_VLAN])
1203 				tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1204 
1205 			if (!(tci & htons(VLAN_TAG_PRESENT))) {
1206 				OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1207 					  ntohs(tci));
1208 				err = -EINVAL;
1209 				goto free_newmask;
1210 			}
1211 		}
1212 
1213 		err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1214 					   log);
1215 		if (err)
1216 			goto free_newmask;
1217 	}
1218 
1219 	if (!match_validate(match, key_attrs, mask_attrs, log))
1220 		err = -EINVAL;
1221 
1222 free_newmask:
1223 	kfree(newmask);
1224 	return err;
1225 }
1226 
1227 static size_t get_ufid_len(const struct nlattr *attr, bool log)
1228 {
1229 	size_t len;
1230 
1231 	if (!attr)
1232 		return 0;
1233 
1234 	len = nla_len(attr);
1235 	if (len < 1 || len > MAX_UFID_LENGTH) {
1236 		OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1237 			  nla_len(attr), MAX_UFID_LENGTH);
1238 		return 0;
1239 	}
1240 
1241 	return len;
1242 }
1243 
1244 /* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1245  * or false otherwise.
1246  */
1247 bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1248 		      bool log)
1249 {
1250 	sfid->ufid_len = get_ufid_len(attr, log);
1251 	if (sfid->ufid_len)
1252 		memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1253 
1254 	return sfid->ufid_len;
1255 }
1256 
1257 int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1258 			   const struct sw_flow_key *key, bool log)
1259 {
1260 	struct sw_flow_key *new_key;
1261 
1262 	if (ovs_nla_get_ufid(sfid, ufid, log))
1263 		return 0;
1264 
1265 	/* If UFID was not provided, use unmasked key. */
1266 	new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1267 	if (!new_key)
1268 		return -ENOMEM;
1269 	memcpy(new_key, key, sizeof(*key));
1270 	sfid->unmasked_key = new_key;
1271 
1272 	return 0;
1273 }
1274 
1275 u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1276 {
1277 	return attr ? nla_get_u32(attr) : 0;
1278 }
1279 
1280 /**
1281  * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
1282  * @key: Receives extracted in_port, priority, tun_key and skb_mark.
1283  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1284  * sequence.
1285  * @log: Boolean to allow kernel error logging.  Normally true, but when
1286  * probing for feature compatibility this should be passed in as false to
1287  * suppress unnecessary error logging.
1288  *
1289  * This parses a series of Netlink attributes that form a flow key, which must
1290  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1291  * get the metadata, that is, the parts of the flow key that cannot be
1292  * extracted from the packet itself.
1293  */
1294 
1295 int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
1296 			      struct sw_flow_key *key,
1297 			      bool log)
1298 {
1299 	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1300 	struct sw_flow_match match;
1301 	u64 attrs = 0;
1302 	int err;
1303 
1304 	err = parse_flow_nlattrs(attr, a, &attrs, log);
1305 	if (err)
1306 		return -EINVAL;
1307 
1308 	memset(&match, 0, sizeof(match));
1309 	match.key = key;
1310 
1311 	memset(&key->ct, 0, sizeof(key->ct));
1312 	key->phy.in_port = DP_MAX_PORTS;
1313 
1314 	return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
1315 }
1316 
1317 static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1318 			     const struct sw_flow_key *output, bool is_mask,
1319 			     struct sk_buff *skb)
1320 {
1321 	struct ovs_key_ethernet *eth_key;
1322 	struct nlattr *nla, *encap;
1323 
1324 	if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1325 		goto nla_put_failure;
1326 
1327 	if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1328 		goto nla_put_failure;
1329 
1330 	if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1331 		goto nla_put_failure;
1332 
1333 	if ((swkey->tun_key.u.ipv4.dst || is_mask)) {
1334 		const void *opts = NULL;
1335 
1336 		if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
1337 			opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
1338 
1339 		if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
1340 				       swkey->tun_opts_len))
1341 			goto nla_put_failure;
1342 	}
1343 
1344 	if (swkey->phy.in_port == DP_MAX_PORTS) {
1345 		if (is_mask && (output->phy.in_port == 0xffff))
1346 			if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1347 				goto nla_put_failure;
1348 	} else {
1349 		u16 upper_u16;
1350 		upper_u16 = !is_mask ? 0 : 0xffff;
1351 
1352 		if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1353 				(upper_u16 << 16) | output->phy.in_port))
1354 			goto nla_put_failure;
1355 	}
1356 
1357 	if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1358 		goto nla_put_failure;
1359 
1360 	if (ovs_ct_put_key(output, skb))
1361 		goto nla_put_failure;
1362 
1363 	nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1364 	if (!nla)
1365 		goto nla_put_failure;
1366 
1367 	eth_key = nla_data(nla);
1368 	ether_addr_copy(eth_key->eth_src, output->eth.src);
1369 	ether_addr_copy(eth_key->eth_dst, output->eth.dst);
1370 
1371 	if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1372 		__be16 eth_type;
1373 		eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1374 		if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1375 		    nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1376 			goto nla_put_failure;
1377 		encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1378 		if (!swkey->eth.tci)
1379 			goto unencap;
1380 	} else
1381 		encap = NULL;
1382 
1383 	if (swkey->eth.type == htons(ETH_P_802_2)) {
1384 		/*
1385 		 * Ethertype 802.2 is represented in the netlink with omitted
1386 		 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1387 		 * 0xffff in the mask attribute.  Ethertype can also
1388 		 * be wildcarded.
1389 		 */
1390 		if (is_mask && output->eth.type)
1391 			if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1392 						output->eth.type))
1393 				goto nla_put_failure;
1394 		goto unencap;
1395 	}
1396 
1397 	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1398 		goto nla_put_failure;
1399 
1400 	if (swkey->eth.type == htons(ETH_P_IP)) {
1401 		struct ovs_key_ipv4 *ipv4_key;
1402 
1403 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1404 		if (!nla)
1405 			goto nla_put_failure;
1406 		ipv4_key = nla_data(nla);
1407 		ipv4_key->ipv4_src = output->ipv4.addr.src;
1408 		ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1409 		ipv4_key->ipv4_proto = output->ip.proto;
1410 		ipv4_key->ipv4_tos = output->ip.tos;
1411 		ipv4_key->ipv4_ttl = output->ip.ttl;
1412 		ipv4_key->ipv4_frag = output->ip.frag;
1413 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1414 		struct ovs_key_ipv6 *ipv6_key;
1415 
1416 		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1417 		if (!nla)
1418 			goto nla_put_failure;
1419 		ipv6_key = nla_data(nla);
1420 		memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1421 				sizeof(ipv6_key->ipv6_src));
1422 		memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1423 				sizeof(ipv6_key->ipv6_dst));
1424 		ipv6_key->ipv6_label = output->ipv6.label;
1425 		ipv6_key->ipv6_proto = output->ip.proto;
1426 		ipv6_key->ipv6_tclass = output->ip.tos;
1427 		ipv6_key->ipv6_hlimit = output->ip.ttl;
1428 		ipv6_key->ipv6_frag = output->ip.frag;
1429 	} else if (swkey->eth.type == htons(ETH_P_ARP) ||
1430 		   swkey->eth.type == htons(ETH_P_RARP)) {
1431 		struct ovs_key_arp *arp_key;
1432 
1433 		nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1434 		if (!nla)
1435 			goto nla_put_failure;
1436 		arp_key = nla_data(nla);
1437 		memset(arp_key, 0, sizeof(struct ovs_key_arp));
1438 		arp_key->arp_sip = output->ipv4.addr.src;
1439 		arp_key->arp_tip = output->ipv4.addr.dst;
1440 		arp_key->arp_op = htons(output->ip.proto);
1441 		ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1442 		ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
1443 	} else if (eth_p_mpls(swkey->eth.type)) {
1444 		struct ovs_key_mpls *mpls_key;
1445 
1446 		nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1447 		if (!nla)
1448 			goto nla_put_failure;
1449 		mpls_key = nla_data(nla);
1450 		mpls_key->mpls_lse = output->mpls.top_lse;
1451 	}
1452 
1453 	if ((swkey->eth.type == htons(ETH_P_IP) ||
1454 	     swkey->eth.type == htons(ETH_P_IPV6)) &&
1455 	     swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1456 
1457 		if (swkey->ip.proto == IPPROTO_TCP) {
1458 			struct ovs_key_tcp *tcp_key;
1459 
1460 			nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1461 			if (!nla)
1462 				goto nla_put_failure;
1463 			tcp_key = nla_data(nla);
1464 			tcp_key->tcp_src = output->tp.src;
1465 			tcp_key->tcp_dst = output->tp.dst;
1466 			if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1467 					 output->tp.flags))
1468 				goto nla_put_failure;
1469 		} else if (swkey->ip.proto == IPPROTO_UDP) {
1470 			struct ovs_key_udp *udp_key;
1471 
1472 			nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1473 			if (!nla)
1474 				goto nla_put_failure;
1475 			udp_key = nla_data(nla);
1476 			udp_key->udp_src = output->tp.src;
1477 			udp_key->udp_dst = output->tp.dst;
1478 		} else if (swkey->ip.proto == IPPROTO_SCTP) {
1479 			struct ovs_key_sctp *sctp_key;
1480 
1481 			nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1482 			if (!nla)
1483 				goto nla_put_failure;
1484 			sctp_key = nla_data(nla);
1485 			sctp_key->sctp_src = output->tp.src;
1486 			sctp_key->sctp_dst = output->tp.dst;
1487 		} else if (swkey->eth.type == htons(ETH_P_IP) &&
1488 			   swkey->ip.proto == IPPROTO_ICMP) {
1489 			struct ovs_key_icmp *icmp_key;
1490 
1491 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1492 			if (!nla)
1493 				goto nla_put_failure;
1494 			icmp_key = nla_data(nla);
1495 			icmp_key->icmp_type = ntohs(output->tp.src);
1496 			icmp_key->icmp_code = ntohs(output->tp.dst);
1497 		} else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1498 			   swkey->ip.proto == IPPROTO_ICMPV6) {
1499 			struct ovs_key_icmpv6 *icmpv6_key;
1500 
1501 			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1502 						sizeof(*icmpv6_key));
1503 			if (!nla)
1504 				goto nla_put_failure;
1505 			icmpv6_key = nla_data(nla);
1506 			icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1507 			icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
1508 
1509 			if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1510 			    icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1511 				struct ovs_key_nd *nd_key;
1512 
1513 				nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1514 				if (!nla)
1515 					goto nla_put_failure;
1516 				nd_key = nla_data(nla);
1517 				memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1518 							sizeof(nd_key->nd_target));
1519 				ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1520 				ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
1521 			}
1522 		}
1523 	}
1524 
1525 unencap:
1526 	if (encap)
1527 		nla_nest_end(skb, encap);
1528 
1529 	return 0;
1530 
1531 nla_put_failure:
1532 	return -EMSGSIZE;
1533 }
1534 
1535 int ovs_nla_put_key(const struct sw_flow_key *swkey,
1536 		    const struct sw_flow_key *output, int attr, bool is_mask,
1537 		    struct sk_buff *skb)
1538 {
1539 	int err;
1540 	struct nlattr *nla;
1541 
1542 	nla = nla_nest_start(skb, attr);
1543 	if (!nla)
1544 		return -EMSGSIZE;
1545 	err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1546 	if (err)
1547 		return err;
1548 	nla_nest_end(skb, nla);
1549 
1550 	return 0;
1551 }
1552 
1553 /* Called with ovs_mutex or RCU read lock. */
1554 int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1555 {
1556 	if (ovs_identifier_is_ufid(&flow->id))
1557 		return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1558 			       flow->id.ufid);
1559 
1560 	return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1561 			       OVS_FLOW_ATTR_KEY, false, skb);
1562 }
1563 
1564 /* Called with ovs_mutex or RCU read lock. */
1565 int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
1566 {
1567 	return ovs_nla_put_key(&flow->key, &flow->key,
1568 				OVS_FLOW_ATTR_KEY, false, skb);
1569 }
1570 
1571 /* Called with ovs_mutex or RCU read lock. */
1572 int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1573 {
1574 	return ovs_nla_put_key(&flow->key, &flow->mask->key,
1575 				OVS_FLOW_ATTR_MASK, true, skb);
1576 }
1577 
1578 #define MAX_ACTIONS_BUFSIZE	(32 * 1024)
1579 
1580 static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
1581 {
1582 	struct sw_flow_actions *sfa;
1583 
1584 	if (size > MAX_ACTIONS_BUFSIZE) {
1585 		OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
1586 		return ERR_PTR(-EINVAL);
1587 	}
1588 
1589 	sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1590 	if (!sfa)
1591 		return ERR_PTR(-ENOMEM);
1592 
1593 	sfa->actions_len = 0;
1594 	return sfa;
1595 }
1596 
1597 static void ovs_nla_free_set_action(const struct nlattr *a)
1598 {
1599 	const struct nlattr *ovs_key = nla_data(a);
1600 	struct ovs_tunnel_info *ovs_tun;
1601 
1602 	switch (nla_type(ovs_key)) {
1603 	case OVS_KEY_ATTR_TUNNEL_INFO:
1604 		ovs_tun = nla_data(ovs_key);
1605 		dst_release((struct dst_entry *)ovs_tun->tun_dst);
1606 		break;
1607 	}
1608 }
1609 
1610 void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1611 {
1612 	const struct nlattr *a;
1613 	int rem;
1614 
1615 	if (!sf_acts)
1616 		return;
1617 
1618 	nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1619 		switch (nla_type(a)) {
1620 		case OVS_ACTION_ATTR_SET:
1621 			ovs_nla_free_set_action(a);
1622 			break;
1623 		case OVS_ACTION_ATTR_CT:
1624 			ovs_ct_free_action(a);
1625 			break;
1626 		}
1627 	}
1628 
1629 	kfree(sf_acts);
1630 }
1631 
1632 static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1633 {
1634 	ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1635 }
1636 
1637 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
1638  * The caller must hold rcu_read_lock for this to be sensible. */
1639 void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
1640 {
1641 	call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
1642 }
1643 
1644 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1645 				       int attr_len, bool log)
1646 {
1647 
1648 	struct sw_flow_actions *acts;
1649 	int new_acts_size;
1650 	int req_size = NLA_ALIGN(attr_len);
1651 	int next_offset = offsetof(struct sw_flow_actions, actions) +
1652 					(*sfa)->actions_len;
1653 
1654 	if (req_size <= (ksize(*sfa) - next_offset))
1655 		goto out;
1656 
1657 	new_acts_size = ksize(*sfa) * 2;
1658 
1659 	if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1660 		if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1661 			return ERR_PTR(-EMSGSIZE);
1662 		new_acts_size = MAX_ACTIONS_BUFSIZE;
1663 	}
1664 
1665 	acts = nla_alloc_flow_actions(new_acts_size, log);
1666 	if (IS_ERR(acts))
1667 		return (void *)acts;
1668 
1669 	memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1670 	acts->actions_len = (*sfa)->actions_len;
1671 	acts->orig_len = (*sfa)->orig_len;
1672 	kfree(*sfa);
1673 	*sfa = acts;
1674 
1675 out:
1676 	(*sfa)->actions_len += req_size;
1677 	return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1678 }
1679 
1680 static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1681 				   int attrtype, void *data, int len, bool log)
1682 {
1683 	struct nlattr *a;
1684 
1685 	a = reserve_sfa_size(sfa, nla_attr_size(len), log);
1686 	if (IS_ERR(a))
1687 		return a;
1688 
1689 	a->nla_type = attrtype;
1690 	a->nla_len = nla_attr_size(len);
1691 
1692 	if (data)
1693 		memcpy(nla_data(a), data, len);
1694 	memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1695 
1696 	return a;
1697 }
1698 
1699 int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1700 		       int len, bool log)
1701 {
1702 	struct nlattr *a;
1703 
1704 	a = __add_action(sfa, attrtype, data, len, log);
1705 
1706 	return PTR_ERR_OR_ZERO(a);
1707 }
1708 
1709 static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1710 					  int attrtype, bool log)
1711 {
1712 	int used = (*sfa)->actions_len;
1713 	int err;
1714 
1715 	err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
1716 	if (err)
1717 		return err;
1718 
1719 	return used;
1720 }
1721 
1722 static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1723 					 int st_offset)
1724 {
1725 	struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1726 							       st_offset);
1727 
1728 	a->nla_len = sfa->actions_len - st_offset;
1729 }
1730 
1731 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
1732 				  const struct sw_flow_key *key,
1733 				  int depth, struct sw_flow_actions **sfa,
1734 				  __be16 eth_type, __be16 vlan_tci, bool log);
1735 
1736 static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
1737 				    const struct sw_flow_key *key, int depth,
1738 				    struct sw_flow_actions **sfa,
1739 				    __be16 eth_type, __be16 vlan_tci, bool log)
1740 {
1741 	const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1742 	const struct nlattr *probability, *actions;
1743 	const struct nlattr *a;
1744 	int rem, start, err, st_acts;
1745 
1746 	memset(attrs, 0, sizeof(attrs));
1747 	nla_for_each_nested(a, attr, rem) {
1748 		int type = nla_type(a);
1749 		if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1750 			return -EINVAL;
1751 		attrs[type] = a;
1752 	}
1753 	if (rem)
1754 		return -EINVAL;
1755 
1756 	probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1757 	if (!probability || nla_len(probability) != sizeof(u32))
1758 		return -EINVAL;
1759 
1760 	actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1761 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1762 		return -EINVAL;
1763 
1764 	/* validation done, copy sample action. */
1765 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
1766 	if (start < 0)
1767 		return start;
1768 	err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1769 				 nla_data(probability), sizeof(u32), log);
1770 	if (err)
1771 		return err;
1772 	st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
1773 	if (st_acts < 0)
1774 		return st_acts;
1775 
1776 	err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
1777 				     eth_type, vlan_tci, log);
1778 	if (err)
1779 		return err;
1780 
1781 	add_nested_action_end(*sfa, st_acts);
1782 	add_nested_action_end(*sfa, start);
1783 
1784 	return 0;
1785 }
1786 
1787 void ovs_match_init(struct sw_flow_match *match,
1788 		    struct sw_flow_key *key,
1789 		    struct sw_flow_mask *mask)
1790 {
1791 	memset(match, 0, sizeof(*match));
1792 	match->key = key;
1793 	match->mask = mask;
1794 
1795 	memset(key, 0, sizeof(*key));
1796 
1797 	if (mask) {
1798 		memset(&mask->key, 0, sizeof(mask->key));
1799 		mask->range.start = mask->range.end = 0;
1800 	}
1801 }
1802 
1803 static int validate_geneve_opts(struct sw_flow_key *key)
1804 {
1805 	struct geneve_opt *option;
1806 	int opts_len = key->tun_opts_len;
1807 	bool crit_opt = false;
1808 
1809 	option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1810 	while (opts_len > 0) {
1811 		int len;
1812 
1813 		if (opts_len < sizeof(*option))
1814 			return -EINVAL;
1815 
1816 		len = sizeof(*option) + option->length * 4;
1817 		if (len > opts_len)
1818 			return -EINVAL;
1819 
1820 		crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1821 
1822 		option = (struct geneve_opt *)((u8 *)option + len);
1823 		opts_len -= len;
1824 	};
1825 
1826 	key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1827 
1828 	return 0;
1829 }
1830 
1831 static int validate_and_copy_set_tun(const struct nlattr *attr,
1832 				     struct sw_flow_actions **sfa, bool log)
1833 {
1834 	struct sw_flow_match match;
1835 	struct sw_flow_key key;
1836 	struct metadata_dst *tun_dst;
1837 	struct ip_tunnel_info *tun_info;
1838 	struct ovs_tunnel_info *ovs_tun;
1839 	struct nlattr *a;
1840 	int err = 0, start, opts_type;
1841 
1842 	ovs_match_init(&match, &key, NULL);
1843 	opts_type = ipv4_tun_from_nlattr(nla_data(attr), &match, false, log);
1844 	if (opts_type < 0)
1845 		return opts_type;
1846 
1847 	if (key.tun_opts_len) {
1848 		switch (opts_type) {
1849 		case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1850 			err = validate_geneve_opts(&key);
1851 			if (err < 0)
1852 				return err;
1853 			break;
1854 		case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1855 			break;
1856 		}
1857 	};
1858 
1859 	start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
1860 	if (start < 0)
1861 		return start;
1862 
1863 	tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1864 	if (!tun_dst)
1865 		return -ENOMEM;
1866 
1867 	a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1868 			 sizeof(*ovs_tun), log);
1869 	if (IS_ERR(a)) {
1870 		dst_release((struct dst_entry *)tun_dst);
1871 		return PTR_ERR(a);
1872 	}
1873 
1874 	ovs_tun = nla_data(a);
1875 	ovs_tun->tun_dst = tun_dst;
1876 
1877 	tun_info = &tun_dst->u.tun_info;
1878 	tun_info->mode = IP_TUNNEL_INFO_TX;
1879 	tun_info->key = key.tun_key;
1880 
1881 	/* We need to store the options in the action itself since
1882 	 * everything else will go away after flow setup. We can append
1883 	 * it to tun_info and then point there.
1884 	 */
1885 	ip_tunnel_info_opts_set(tun_info,
1886 				TUN_METADATA_OPTS(&key, key.tun_opts_len),
1887 				key.tun_opts_len);
1888 	add_nested_action_end(*sfa, start);
1889 
1890 	return err;
1891 }
1892 
1893 /* Return false if there are any non-masked bits set.
1894  * Mask follows data immediately, before any netlink padding.
1895  */
1896 static bool validate_masked(u8 *data, int len)
1897 {
1898 	u8 *mask = data + len;
1899 
1900 	while (len--)
1901 		if (*data++ & ~*mask++)
1902 			return false;
1903 
1904 	return true;
1905 }
1906 
1907 static int validate_set(const struct nlattr *a,
1908 			const struct sw_flow_key *flow_key,
1909 			struct sw_flow_actions **sfa,
1910 			bool *skip_copy, __be16 eth_type, bool masked, bool log)
1911 {
1912 	const struct nlattr *ovs_key = nla_data(a);
1913 	int key_type = nla_type(ovs_key);
1914 	size_t key_len;
1915 
1916 	/* There can be only one key in a action */
1917 	if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1918 		return -EINVAL;
1919 
1920 	key_len = nla_len(ovs_key);
1921 	if (masked)
1922 		key_len /= 2;
1923 
1924 	if (key_type > OVS_KEY_ATTR_MAX ||
1925 	    (ovs_key_lens[key_type].len != key_len &&
1926 	     ovs_key_lens[key_type].len != OVS_ATTR_NESTED))
1927 		return -EINVAL;
1928 
1929 	if (masked && !validate_masked(nla_data(ovs_key), key_len))
1930 		return -EINVAL;
1931 
1932 	switch (key_type) {
1933 	const struct ovs_key_ipv4 *ipv4_key;
1934 	const struct ovs_key_ipv6 *ipv6_key;
1935 	int err;
1936 
1937 	case OVS_KEY_ATTR_PRIORITY:
1938 	case OVS_KEY_ATTR_SKB_MARK:
1939 	case OVS_KEY_ATTR_CT_MARK:
1940 	case OVS_KEY_ATTR_CT_LABEL:
1941 	case OVS_KEY_ATTR_ETHERNET:
1942 		break;
1943 
1944 	case OVS_KEY_ATTR_TUNNEL:
1945 		if (eth_p_mpls(eth_type))
1946 			return -EINVAL;
1947 
1948 		if (masked)
1949 			return -EINVAL; /* Masked tunnel set not supported. */
1950 
1951 		*skip_copy = true;
1952 		err = validate_and_copy_set_tun(a, sfa, log);
1953 		if (err)
1954 			return err;
1955 		break;
1956 
1957 	case OVS_KEY_ATTR_IPV4:
1958 		if (eth_type != htons(ETH_P_IP))
1959 			return -EINVAL;
1960 
1961 		ipv4_key = nla_data(ovs_key);
1962 
1963 		if (masked) {
1964 			const struct ovs_key_ipv4 *mask = ipv4_key + 1;
1965 
1966 			/* Non-writeable fields. */
1967 			if (mask->ipv4_proto || mask->ipv4_frag)
1968 				return -EINVAL;
1969 		} else {
1970 			if (ipv4_key->ipv4_proto != flow_key->ip.proto)
1971 				return -EINVAL;
1972 
1973 			if (ipv4_key->ipv4_frag != flow_key->ip.frag)
1974 				return -EINVAL;
1975 		}
1976 		break;
1977 
1978 	case OVS_KEY_ATTR_IPV6:
1979 		if (eth_type != htons(ETH_P_IPV6))
1980 			return -EINVAL;
1981 
1982 		ipv6_key = nla_data(ovs_key);
1983 
1984 		if (masked) {
1985 			const struct ovs_key_ipv6 *mask = ipv6_key + 1;
1986 
1987 			/* Non-writeable fields. */
1988 			if (mask->ipv6_proto || mask->ipv6_frag)
1989 				return -EINVAL;
1990 
1991 			/* Invalid bits in the flow label mask? */
1992 			if (ntohl(mask->ipv6_label) & 0xFFF00000)
1993 				return -EINVAL;
1994 		} else {
1995 			if (ipv6_key->ipv6_proto != flow_key->ip.proto)
1996 				return -EINVAL;
1997 
1998 			if (ipv6_key->ipv6_frag != flow_key->ip.frag)
1999 				return -EINVAL;
2000 		}
2001 		if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2002 			return -EINVAL;
2003 
2004 		break;
2005 
2006 	case OVS_KEY_ATTR_TCP:
2007 		if ((eth_type != htons(ETH_P_IP) &&
2008 		     eth_type != htons(ETH_P_IPV6)) ||
2009 		    flow_key->ip.proto != IPPROTO_TCP)
2010 			return -EINVAL;
2011 
2012 		break;
2013 
2014 	case OVS_KEY_ATTR_UDP:
2015 		if ((eth_type != htons(ETH_P_IP) &&
2016 		     eth_type != htons(ETH_P_IPV6)) ||
2017 		    flow_key->ip.proto != IPPROTO_UDP)
2018 			return -EINVAL;
2019 
2020 		break;
2021 
2022 	case OVS_KEY_ATTR_MPLS:
2023 		if (!eth_p_mpls(eth_type))
2024 			return -EINVAL;
2025 		break;
2026 
2027 	case OVS_KEY_ATTR_SCTP:
2028 		if ((eth_type != htons(ETH_P_IP) &&
2029 		     eth_type != htons(ETH_P_IPV6)) ||
2030 		    flow_key->ip.proto != IPPROTO_SCTP)
2031 			return -EINVAL;
2032 
2033 		break;
2034 
2035 	default:
2036 		return -EINVAL;
2037 	}
2038 
2039 	/* Convert non-masked non-tunnel set actions to masked set actions. */
2040 	if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2041 		int start, len = key_len * 2;
2042 		struct nlattr *at;
2043 
2044 		*skip_copy = true;
2045 
2046 		start = add_nested_action_start(sfa,
2047 						OVS_ACTION_ATTR_SET_TO_MASKED,
2048 						log);
2049 		if (start < 0)
2050 			return start;
2051 
2052 		at = __add_action(sfa, key_type, NULL, len, log);
2053 		if (IS_ERR(at))
2054 			return PTR_ERR(at);
2055 
2056 		memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2057 		memset(nla_data(at) + key_len, 0xff, key_len);    /* Mask. */
2058 		/* Clear non-writeable bits from otherwise writeable fields. */
2059 		if (key_type == OVS_KEY_ATTR_IPV6) {
2060 			struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2061 
2062 			mask->ipv6_label &= htonl(0x000FFFFF);
2063 		}
2064 		add_nested_action_end(*sfa, start);
2065 	}
2066 
2067 	return 0;
2068 }
2069 
2070 static int validate_userspace(const struct nlattr *attr)
2071 {
2072 	static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2073 		[OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2074 		[OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
2075 		[OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
2076 	};
2077 	struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2078 	int error;
2079 
2080 	error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2081 				 attr, userspace_policy);
2082 	if (error)
2083 		return error;
2084 
2085 	if (!a[OVS_USERSPACE_ATTR_PID] ||
2086 	    !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2087 		return -EINVAL;
2088 
2089 	return 0;
2090 }
2091 
2092 static int copy_action(const struct nlattr *from,
2093 		       struct sw_flow_actions **sfa, bool log)
2094 {
2095 	int totlen = NLA_ALIGN(from->nla_len);
2096 	struct nlattr *to;
2097 
2098 	to = reserve_sfa_size(sfa, from->nla_len, log);
2099 	if (IS_ERR(to))
2100 		return PTR_ERR(to);
2101 
2102 	memcpy(to, from, totlen);
2103 	return 0;
2104 }
2105 
2106 static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2107 				  const struct sw_flow_key *key,
2108 				  int depth, struct sw_flow_actions **sfa,
2109 				  __be16 eth_type, __be16 vlan_tci, bool log)
2110 {
2111 	const struct nlattr *a;
2112 	int rem, err;
2113 
2114 	if (depth >= SAMPLE_ACTION_DEPTH)
2115 		return -EOVERFLOW;
2116 
2117 	nla_for_each_nested(a, attr, rem) {
2118 		/* Expected argument lengths, (u32)-1 for variable length. */
2119 		static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2120 			[OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
2121 			[OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
2122 			[OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
2123 			[OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2124 			[OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
2125 			[OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2126 			[OVS_ACTION_ATTR_POP_VLAN] = 0,
2127 			[OVS_ACTION_ATTR_SET] = (u32)-1,
2128 			[OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
2129 			[OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2130 			[OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2131 			[OVS_ACTION_ATTR_CT] = (u32)-1,
2132 		};
2133 		const struct ovs_action_push_vlan *vlan;
2134 		int type = nla_type(a);
2135 		bool skip_copy;
2136 
2137 		if (type > OVS_ACTION_ATTR_MAX ||
2138 		    (action_lens[type] != nla_len(a) &&
2139 		     action_lens[type] != (u32)-1))
2140 			return -EINVAL;
2141 
2142 		skip_copy = false;
2143 		switch (type) {
2144 		case OVS_ACTION_ATTR_UNSPEC:
2145 			return -EINVAL;
2146 
2147 		case OVS_ACTION_ATTR_USERSPACE:
2148 			err = validate_userspace(a);
2149 			if (err)
2150 				return err;
2151 			break;
2152 
2153 		case OVS_ACTION_ATTR_OUTPUT:
2154 			if (nla_get_u32(a) >= DP_MAX_PORTS)
2155 				return -EINVAL;
2156 			break;
2157 
2158 		case OVS_ACTION_ATTR_HASH: {
2159 			const struct ovs_action_hash *act_hash = nla_data(a);
2160 
2161 			switch (act_hash->hash_alg) {
2162 			case OVS_HASH_ALG_L4:
2163 				break;
2164 			default:
2165 				return  -EINVAL;
2166 			}
2167 
2168 			break;
2169 		}
2170 
2171 		case OVS_ACTION_ATTR_POP_VLAN:
2172 			vlan_tci = htons(0);
2173 			break;
2174 
2175 		case OVS_ACTION_ATTR_PUSH_VLAN:
2176 			vlan = nla_data(a);
2177 			if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2178 				return -EINVAL;
2179 			if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2180 				return -EINVAL;
2181 			vlan_tci = vlan->vlan_tci;
2182 			break;
2183 
2184 		case OVS_ACTION_ATTR_RECIRC:
2185 			break;
2186 
2187 		case OVS_ACTION_ATTR_PUSH_MPLS: {
2188 			const struct ovs_action_push_mpls *mpls = nla_data(a);
2189 
2190 			if (!eth_p_mpls(mpls->mpls_ethertype))
2191 				return -EINVAL;
2192 			/* Prohibit push MPLS other than to a white list
2193 			 * for packets that have a known tag order.
2194 			 */
2195 			if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2196 			    (eth_type != htons(ETH_P_IP) &&
2197 			     eth_type != htons(ETH_P_IPV6) &&
2198 			     eth_type != htons(ETH_P_ARP) &&
2199 			     eth_type != htons(ETH_P_RARP) &&
2200 			     !eth_p_mpls(eth_type)))
2201 				return -EINVAL;
2202 			eth_type = mpls->mpls_ethertype;
2203 			break;
2204 		}
2205 
2206 		case OVS_ACTION_ATTR_POP_MPLS:
2207 			if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2208 			    !eth_p_mpls(eth_type))
2209 				return -EINVAL;
2210 
2211 			/* Disallow subsequent L2.5+ set and mpls_pop actions
2212 			 * as there is no check here to ensure that the new
2213 			 * eth_type is valid and thus set actions could
2214 			 * write off the end of the packet or otherwise
2215 			 * corrupt it.
2216 			 *
2217 			 * Support for these actions is planned using packet
2218 			 * recirculation.
2219 			 */
2220 			eth_type = htons(0);
2221 			break;
2222 
2223 		case OVS_ACTION_ATTR_SET:
2224 			err = validate_set(a, key, sfa,
2225 					   &skip_copy, eth_type, false, log);
2226 			if (err)
2227 				return err;
2228 			break;
2229 
2230 		case OVS_ACTION_ATTR_SET_MASKED:
2231 			err = validate_set(a, key, sfa,
2232 					   &skip_copy, eth_type, true, log);
2233 			if (err)
2234 				return err;
2235 			break;
2236 
2237 		case OVS_ACTION_ATTR_SAMPLE:
2238 			err = validate_and_copy_sample(net, a, key, depth, sfa,
2239 						       eth_type, vlan_tci, log);
2240 			if (err)
2241 				return err;
2242 			skip_copy = true;
2243 			break;
2244 
2245 		case OVS_ACTION_ATTR_CT:
2246 			err = ovs_ct_copy_action(net, a, key, sfa, log);
2247 			if (err)
2248 				return err;
2249 			skip_copy = true;
2250 			break;
2251 
2252 		default:
2253 			OVS_NLERR(log, "Unknown Action type %d", type);
2254 			return -EINVAL;
2255 		}
2256 		if (!skip_copy) {
2257 			err = copy_action(a, sfa, log);
2258 			if (err)
2259 				return err;
2260 		}
2261 	}
2262 
2263 	if (rem > 0)
2264 		return -EINVAL;
2265 
2266 	return 0;
2267 }
2268 
2269 /* 'key' must be the masked key. */
2270 int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
2271 			 const struct sw_flow_key *key,
2272 			 struct sw_flow_actions **sfa, bool log)
2273 {
2274 	int err;
2275 
2276 	*sfa = nla_alloc_flow_actions(nla_len(attr), log);
2277 	if (IS_ERR(*sfa))
2278 		return PTR_ERR(*sfa);
2279 
2280 	(*sfa)->orig_len = nla_len(attr);
2281 	err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
2282 				     key->eth.tci, log);
2283 	if (err)
2284 		ovs_nla_free_flow_actions(*sfa);
2285 
2286 	return err;
2287 }
2288 
2289 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2290 {
2291 	const struct nlattr *a;
2292 	struct nlattr *start;
2293 	int err = 0, rem;
2294 
2295 	start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2296 	if (!start)
2297 		return -EMSGSIZE;
2298 
2299 	nla_for_each_nested(a, attr, rem) {
2300 		int type = nla_type(a);
2301 		struct nlattr *st_sample;
2302 
2303 		switch (type) {
2304 		case OVS_SAMPLE_ATTR_PROBABILITY:
2305 			if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2306 				    sizeof(u32), nla_data(a)))
2307 				return -EMSGSIZE;
2308 			break;
2309 		case OVS_SAMPLE_ATTR_ACTIONS:
2310 			st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2311 			if (!st_sample)
2312 				return -EMSGSIZE;
2313 			err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2314 			if (err)
2315 				return err;
2316 			nla_nest_end(skb, st_sample);
2317 			break;
2318 		}
2319 	}
2320 
2321 	nla_nest_end(skb, start);
2322 	return err;
2323 }
2324 
2325 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2326 {
2327 	const struct nlattr *ovs_key = nla_data(a);
2328 	int key_type = nla_type(ovs_key);
2329 	struct nlattr *start;
2330 	int err;
2331 
2332 	switch (key_type) {
2333 	case OVS_KEY_ATTR_TUNNEL_INFO: {
2334 		struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2335 		struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
2336 
2337 		start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2338 		if (!start)
2339 			return -EMSGSIZE;
2340 
2341 		err = ipv4_tun_to_nlattr(skb, &tun_info->key,
2342 					 tun_info->options_len ?
2343 					     ip_tunnel_info_opts(tun_info) : NULL,
2344 					 tun_info->options_len);
2345 		if (err)
2346 			return err;
2347 		nla_nest_end(skb, start);
2348 		break;
2349 	}
2350 	default:
2351 		if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2352 			return -EMSGSIZE;
2353 		break;
2354 	}
2355 
2356 	return 0;
2357 }
2358 
2359 static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2360 						struct sk_buff *skb)
2361 {
2362 	const struct nlattr *ovs_key = nla_data(a);
2363 	struct nlattr *nla;
2364 	size_t key_len = nla_len(ovs_key) / 2;
2365 
2366 	/* Revert the conversion we did from a non-masked set action to
2367 	 * masked set action.
2368 	 */
2369 	nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2370 	if (!nla)
2371 		return -EMSGSIZE;
2372 
2373 	if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2374 		return -EMSGSIZE;
2375 
2376 	nla_nest_end(skb, nla);
2377 	return 0;
2378 }
2379 
2380 int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2381 {
2382 	const struct nlattr *a;
2383 	int rem, err;
2384 
2385 	nla_for_each_attr(a, attr, len, rem) {
2386 		int type = nla_type(a);
2387 
2388 		switch (type) {
2389 		case OVS_ACTION_ATTR_SET:
2390 			err = set_action_to_attr(a, skb);
2391 			if (err)
2392 				return err;
2393 			break;
2394 
2395 		case OVS_ACTION_ATTR_SET_TO_MASKED:
2396 			err = masked_set_action_to_set_action_attr(a, skb);
2397 			if (err)
2398 				return err;
2399 			break;
2400 
2401 		case OVS_ACTION_ATTR_SAMPLE:
2402 			err = sample_action_to_attr(a, skb);
2403 			if (err)
2404 				return err;
2405 			break;
2406 
2407 		case OVS_ACTION_ATTR_CT:
2408 			err = ovs_ct_action_to_attr(nla_data(a), skb);
2409 			if (err)
2410 				return err;
2411 			break;
2412 
2413 		default:
2414 			if (nla_put(skb, type, nla_len(a), nla_data(a)))
2415 				return -EMSGSIZE;
2416 			break;
2417 		}
2418 	}
2419 
2420 	return 0;
2421 }
2422