xref: /linux/net/openvswitch/conntrack.c (revision 1fc31357ad194fb98691f3d122bcd47e59239e83)
1 /*
2  * Copyright (c) 2015 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 
14 #include <linux/module.h>
15 #include <linux/openvswitch.h>
16 #include <linux/tcp.h>
17 #include <linux/udp.h>
18 #include <linux/sctp.h>
19 #include <net/ip.h>
20 #include <net/netfilter/nf_conntrack_core.h>
21 #include <net/netfilter/nf_conntrack_helper.h>
22 #include <net/netfilter/nf_conntrack_labels.h>
23 #include <net/netfilter/nf_conntrack_seqadj.h>
24 #include <net/netfilter/nf_conntrack_zones.h>
25 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
26 
27 #ifdef CONFIG_NF_NAT_NEEDED
28 #include <linux/netfilter/nf_nat.h>
29 #include <net/netfilter/nf_nat_core.h>
30 #include <net/netfilter/nf_nat_l3proto.h>
31 #endif
32 
33 #include "datapath.h"
34 #include "conntrack.h"
35 #include "flow.h"
36 #include "flow_netlink.h"
37 
38 struct ovs_ct_len_tbl {
39 	int maxlen;
40 	int minlen;
41 };
42 
43 /* Metadata mark for masked write to conntrack mark */
44 struct md_mark {
45 	u32 value;
46 	u32 mask;
47 };
48 
49 /* Metadata label for masked write to conntrack label. */
50 struct md_labels {
51 	struct ovs_key_ct_labels value;
52 	struct ovs_key_ct_labels mask;
53 };
54 
55 enum ovs_ct_nat {
56 	OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
57 	OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
58 	OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
59 };
60 
61 /* Conntrack action context for execution. */
62 struct ovs_conntrack_info {
63 	struct nf_conntrack_helper *helper;
64 	struct nf_conntrack_zone zone;
65 	struct nf_conn *ct;
66 	u8 commit : 1;
67 	u8 nat : 3;                 /* enum ovs_ct_nat */
68 	u16 family;
69 	struct md_mark mark;
70 	struct md_labels labels;
71 #ifdef CONFIG_NF_NAT_NEEDED
72 	struct nf_nat_range range;  /* Only present for SRC NAT and DST NAT. */
73 #endif
74 };
75 
76 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
77 
78 static u16 key_to_nfproto(const struct sw_flow_key *key)
79 {
80 	switch (ntohs(key->eth.type)) {
81 	case ETH_P_IP:
82 		return NFPROTO_IPV4;
83 	case ETH_P_IPV6:
84 		return NFPROTO_IPV6;
85 	default:
86 		return NFPROTO_UNSPEC;
87 	}
88 }
89 
90 /* Map SKB connection state into the values used by flow definition. */
91 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
92 {
93 	u8 ct_state = OVS_CS_F_TRACKED;
94 
95 	switch (ctinfo) {
96 	case IP_CT_ESTABLISHED_REPLY:
97 	case IP_CT_RELATED_REPLY:
98 		ct_state |= OVS_CS_F_REPLY_DIR;
99 		break;
100 	default:
101 		break;
102 	}
103 
104 	switch (ctinfo) {
105 	case IP_CT_ESTABLISHED:
106 	case IP_CT_ESTABLISHED_REPLY:
107 		ct_state |= OVS_CS_F_ESTABLISHED;
108 		break;
109 	case IP_CT_RELATED:
110 	case IP_CT_RELATED_REPLY:
111 		ct_state |= OVS_CS_F_RELATED;
112 		break;
113 	case IP_CT_NEW:
114 		ct_state |= OVS_CS_F_NEW;
115 		break;
116 	default:
117 		break;
118 	}
119 
120 	return ct_state;
121 }
122 
123 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
124 {
125 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
126 	return ct ? ct->mark : 0;
127 #else
128 	return 0;
129 #endif
130 }
131 
132 static void ovs_ct_get_labels(const struct nf_conn *ct,
133 			      struct ovs_key_ct_labels *labels)
134 {
135 	struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
136 
137 	if (cl) {
138 		size_t len = sizeof(cl->bits);
139 
140 		if (len > OVS_CT_LABELS_LEN)
141 			len = OVS_CT_LABELS_LEN;
142 		else if (len < OVS_CT_LABELS_LEN)
143 			memset(labels, 0, OVS_CT_LABELS_LEN);
144 		memcpy(labels, cl->bits, len);
145 	} else {
146 		memset(labels, 0, OVS_CT_LABELS_LEN);
147 	}
148 }
149 
150 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
151 				const struct nf_conntrack_zone *zone,
152 				const struct nf_conn *ct)
153 {
154 	key->ct.state = state;
155 	key->ct.zone = zone->id;
156 	key->ct.mark = ovs_ct_get_mark(ct);
157 	ovs_ct_get_labels(ct, &key->ct.labels);
158 }
159 
160 /* Update 'key' based on skb->nfct.  If 'post_ct' is true, then OVS has
161  * previously sent the packet to conntrack via the ct action.  If
162  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
163  * initialized from the connection status.
164  */
165 static void ovs_ct_update_key(const struct sk_buff *skb,
166 			      const struct ovs_conntrack_info *info,
167 			      struct sw_flow_key *key, bool post_ct,
168 			      bool keep_nat_flags)
169 {
170 	const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
171 	enum ip_conntrack_info ctinfo;
172 	struct nf_conn *ct;
173 	u8 state = 0;
174 
175 	ct = nf_ct_get(skb, &ctinfo);
176 	if (ct) {
177 		state = ovs_ct_get_state(ctinfo);
178 		/* All unconfirmed entries are NEW connections. */
179 		if (!nf_ct_is_confirmed(ct))
180 			state |= OVS_CS_F_NEW;
181 		/* OVS persists the related flag for the duration of the
182 		 * connection.
183 		 */
184 		if (ct->master)
185 			state |= OVS_CS_F_RELATED;
186 		if (keep_nat_flags) {
187 			state |= key->ct.state & OVS_CS_F_NAT_MASK;
188 		} else {
189 			if (ct->status & IPS_SRC_NAT)
190 				state |= OVS_CS_F_SRC_NAT;
191 			if (ct->status & IPS_DST_NAT)
192 				state |= OVS_CS_F_DST_NAT;
193 		}
194 		zone = nf_ct_zone(ct);
195 	} else if (post_ct) {
196 		state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
197 		if (info)
198 			zone = &info->zone;
199 	}
200 	__ovs_ct_update_key(key, state, zone, ct);
201 }
202 
203 /* This is called to initialize CT key fields possibly coming in from the local
204  * stack.
205  */
206 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
207 {
208 	ovs_ct_update_key(skb, NULL, key, false, false);
209 }
210 
211 int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb)
212 {
213 	if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state))
214 		return -EMSGSIZE;
215 
216 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
217 	    nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone))
218 		return -EMSGSIZE;
219 
220 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
221 	    nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark))
222 		return -EMSGSIZE;
223 
224 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
225 	    nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels),
226 		    &key->ct.labels))
227 		return -EMSGSIZE;
228 
229 	return 0;
230 }
231 
232 static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key,
233 			   u32 ct_mark, u32 mask)
234 {
235 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
236 	enum ip_conntrack_info ctinfo;
237 	struct nf_conn *ct;
238 	u32 new_mark;
239 
240 	/* The connection could be invalid, in which case set_mark is no-op. */
241 	ct = nf_ct_get(skb, &ctinfo);
242 	if (!ct)
243 		return 0;
244 
245 	new_mark = ct_mark | (ct->mark & ~(mask));
246 	if (ct->mark != new_mark) {
247 		ct->mark = new_mark;
248 		nf_conntrack_event_cache(IPCT_MARK, ct);
249 		key->ct.mark = new_mark;
250 	}
251 
252 	return 0;
253 #else
254 	return -ENOTSUPP;
255 #endif
256 }
257 
258 static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key,
259 			     const struct ovs_key_ct_labels *labels,
260 			     const struct ovs_key_ct_labels *mask)
261 {
262 	enum ip_conntrack_info ctinfo;
263 	struct nf_conn_labels *cl;
264 	struct nf_conn *ct;
265 	int err;
266 
267 	/* The connection could be invalid, in which case set_label is no-op.*/
268 	ct = nf_ct_get(skb, &ctinfo);
269 	if (!ct)
270 		return 0;
271 
272 	cl = nf_ct_labels_find(ct);
273 	if (!cl) {
274 		nf_ct_labels_ext_add(ct);
275 		cl = nf_ct_labels_find(ct);
276 	}
277 	if (!cl || sizeof(cl->bits) < OVS_CT_LABELS_LEN)
278 		return -ENOSPC;
279 
280 	err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask,
281 				    OVS_CT_LABELS_LEN / sizeof(u32));
282 	if (err)
283 		return err;
284 
285 	ovs_ct_get_labels(ct, &key->ct.labels);
286 	return 0;
287 }
288 
289 /* 'skb' should already be pulled to nh_ofs. */
290 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
291 {
292 	const struct nf_conntrack_helper *helper;
293 	const struct nf_conn_help *help;
294 	enum ip_conntrack_info ctinfo;
295 	unsigned int protoff;
296 	struct nf_conn *ct;
297 	int err;
298 
299 	ct = nf_ct_get(skb, &ctinfo);
300 	if (!ct || ctinfo == IP_CT_RELATED_REPLY)
301 		return NF_ACCEPT;
302 
303 	help = nfct_help(ct);
304 	if (!help)
305 		return NF_ACCEPT;
306 
307 	helper = rcu_dereference(help->helper);
308 	if (!helper)
309 		return NF_ACCEPT;
310 
311 	switch (proto) {
312 	case NFPROTO_IPV4:
313 		protoff = ip_hdrlen(skb);
314 		break;
315 	case NFPROTO_IPV6: {
316 		u8 nexthdr = ipv6_hdr(skb)->nexthdr;
317 		__be16 frag_off;
318 		int ofs;
319 
320 		ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
321 				       &frag_off);
322 		if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
323 			pr_debug("proto header not found\n");
324 			return NF_ACCEPT;
325 		}
326 		protoff = ofs;
327 		break;
328 	}
329 	default:
330 		WARN_ONCE(1, "helper invoked on non-IP family!");
331 		return NF_DROP;
332 	}
333 
334 	err = helper->help(skb, protoff, ct, ctinfo);
335 	if (err != NF_ACCEPT)
336 		return err;
337 
338 	/* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
339 	 * FTP with NAT) adusting the TCP payload size when mangling IP
340 	 * addresses and/or port numbers in the text-based control connection.
341 	 */
342 	if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
343 	    !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
344 		return NF_DROP;
345 	return NF_ACCEPT;
346 }
347 
348 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
349  * value if 'skb' is freed.
350  */
351 static int handle_fragments(struct net *net, struct sw_flow_key *key,
352 			    u16 zone, struct sk_buff *skb)
353 {
354 	struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
355 	int err;
356 
357 	if (key->eth.type == htons(ETH_P_IP)) {
358 		enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
359 
360 		memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
361 		err = ip_defrag(net, skb, user);
362 		if (err)
363 			return err;
364 
365 		ovs_cb.mru = IPCB(skb)->frag_max_size;
366 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
367 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
368 		enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
369 
370 		skb_orphan(skb);
371 		memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
372 		err = nf_ct_frag6_gather(net, skb, user);
373 		if (err)
374 			return err;
375 
376 		key->ip.proto = ipv6_hdr(skb)->nexthdr;
377 		ovs_cb.mru = IP6CB(skb)->frag_max_size;
378 #endif
379 	} else {
380 		kfree_skb(skb);
381 		return -EPFNOSUPPORT;
382 	}
383 
384 	key->ip.frag = OVS_FRAG_TYPE_NONE;
385 	skb_clear_hash(skb);
386 	skb->ignore_df = 1;
387 	*OVS_CB(skb) = ovs_cb;
388 
389 	return 0;
390 }
391 
392 static struct nf_conntrack_expect *
393 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
394 		   u16 proto, const struct sk_buff *skb)
395 {
396 	struct nf_conntrack_tuple tuple;
397 
398 	if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
399 		return NULL;
400 	return __nf_ct_expect_find(net, zone, &tuple);
401 }
402 
403 /* This replicates logic from nf_conntrack_core.c that is not exported. */
404 static enum ip_conntrack_info
405 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
406 {
407 	const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
408 
409 	if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
410 		return IP_CT_ESTABLISHED_REPLY;
411 	/* Once we've had two way comms, always ESTABLISHED. */
412 	if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
413 		return IP_CT_ESTABLISHED;
414 	if (test_bit(IPS_EXPECTED_BIT, &ct->status))
415 		return IP_CT_RELATED;
416 	return IP_CT_NEW;
417 }
418 
419 /* Find an existing connection which this packet belongs to without
420  * re-attributing statistics or modifying the connection state.  This allows an
421  * skb->nfct lost due to an upcall to be recovered during actions execution.
422  *
423  * Must be called with rcu_read_lock.
424  *
425  * On success, populates skb->nfct and skb->nfctinfo, and returns the
426  * connection.  Returns NULL if there is no existing entry.
427  */
428 static struct nf_conn *
429 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
430 		     u8 l3num, struct sk_buff *skb)
431 {
432 	struct nf_conntrack_l3proto *l3proto;
433 	struct nf_conntrack_l4proto *l4proto;
434 	struct nf_conntrack_tuple tuple;
435 	struct nf_conntrack_tuple_hash *h;
436 	struct nf_conn *ct;
437 	unsigned int dataoff;
438 	u8 protonum;
439 
440 	l3proto = __nf_ct_l3proto_find(l3num);
441 	if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
442 				 &protonum) <= 0) {
443 		pr_debug("ovs_ct_find_existing: Can't get protonum\n");
444 		return NULL;
445 	}
446 	l4proto = __nf_ct_l4proto_find(l3num, protonum);
447 	if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
448 			     protonum, net, &tuple, l3proto, l4proto)) {
449 		pr_debug("ovs_ct_find_existing: Can't get tuple\n");
450 		return NULL;
451 	}
452 
453 	/* look for tuple match */
454 	h = nf_conntrack_find_get(net, zone, &tuple);
455 	if (!h)
456 		return NULL;   /* Not found. */
457 
458 	ct = nf_ct_tuplehash_to_ctrack(h);
459 
460 	skb->nfct = &ct->ct_general;
461 	skb->nfctinfo = ovs_ct_get_info(h);
462 	return ct;
463 }
464 
465 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
466 static bool skb_nfct_cached(struct net *net,
467 			    const struct sw_flow_key *key,
468 			    const struct ovs_conntrack_info *info,
469 			    struct sk_buff *skb)
470 {
471 	enum ip_conntrack_info ctinfo;
472 	struct nf_conn *ct;
473 
474 	ct = nf_ct_get(skb, &ctinfo);
475 	/* If no ct, check if we have evidence that an existing conntrack entry
476 	 * might be found for this skb.  This happens when we lose a skb->nfct
477 	 * due to an upcall.  If the connection was not confirmed, it is not
478 	 * cached and needs to be run through conntrack again.
479 	 */
480 	if (!ct && key->ct.state & OVS_CS_F_TRACKED &&
481 	    !(key->ct.state & OVS_CS_F_INVALID) &&
482 	    key->ct.zone == info->zone.id)
483 		ct = ovs_ct_find_existing(net, &info->zone, info->family, skb);
484 	if (!ct)
485 		return false;
486 	if (!net_eq(net, read_pnet(&ct->ct_net)))
487 		return false;
488 	if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
489 		return false;
490 	if (info->helper) {
491 		struct nf_conn_help *help;
492 
493 		help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
494 		if (help && rcu_access_pointer(help->helper) != info->helper)
495 			return false;
496 	}
497 
498 	return true;
499 }
500 
501 #ifdef CONFIG_NF_NAT_NEEDED
502 /* Modelled after nf_nat_ipv[46]_fn().
503  * range is only used for new, uninitialized NAT state.
504  * Returns either NF_ACCEPT or NF_DROP.
505  */
506 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
507 			      enum ip_conntrack_info ctinfo,
508 			      const struct nf_nat_range *range,
509 			      enum nf_nat_manip_type maniptype)
510 {
511 	int hooknum, nh_off, err = NF_ACCEPT;
512 
513 	nh_off = skb_network_offset(skb);
514 	skb_pull(skb, nh_off);
515 
516 	/* See HOOK2MANIP(). */
517 	if (maniptype == NF_NAT_MANIP_SRC)
518 		hooknum = NF_INET_LOCAL_IN; /* Source NAT */
519 	else
520 		hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
521 
522 	switch (ctinfo) {
523 	case IP_CT_RELATED:
524 	case IP_CT_RELATED_REPLY:
525 		if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
526 		    skb->protocol == htons(ETH_P_IP) &&
527 		    ip_hdr(skb)->protocol == IPPROTO_ICMP) {
528 			if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
529 							   hooknum))
530 				err = NF_DROP;
531 			goto push;
532 		} else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
533 			   skb->protocol == htons(ETH_P_IPV6)) {
534 			__be16 frag_off;
535 			u8 nexthdr = ipv6_hdr(skb)->nexthdr;
536 			int hdrlen = ipv6_skip_exthdr(skb,
537 						      sizeof(struct ipv6hdr),
538 						      &nexthdr, &frag_off);
539 
540 			if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
541 				if (!nf_nat_icmpv6_reply_translation(skb, ct,
542 								     ctinfo,
543 								     hooknum,
544 								     hdrlen))
545 					err = NF_DROP;
546 				goto push;
547 			}
548 		}
549 		/* Non-ICMP, fall thru to initialize if needed. */
550 	case IP_CT_NEW:
551 		/* Seen it before?  This can happen for loopback, retrans,
552 		 * or local packets.
553 		 */
554 		if (!nf_nat_initialized(ct, maniptype)) {
555 			/* Initialize according to the NAT action. */
556 			err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
557 				/* Action is set up to establish a new
558 				 * mapping.
559 				 */
560 				? nf_nat_setup_info(ct, range, maniptype)
561 				: nf_nat_alloc_null_binding(ct, hooknum);
562 			if (err != NF_ACCEPT)
563 				goto push;
564 		}
565 		break;
566 
567 	case IP_CT_ESTABLISHED:
568 	case IP_CT_ESTABLISHED_REPLY:
569 		break;
570 
571 	default:
572 		err = NF_DROP;
573 		goto push;
574 	}
575 
576 	err = nf_nat_packet(ct, ctinfo, hooknum, skb);
577 push:
578 	skb_push(skb, nh_off);
579 
580 	return err;
581 }
582 
583 static void ovs_nat_update_key(struct sw_flow_key *key,
584 			       const struct sk_buff *skb,
585 			       enum nf_nat_manip_type maniptype)
586 {
587 	if (maniptype == NF_NAT_MANIP_SRC) {
588 		__be16 src;
589 
590 		key->ct.state |= OVS_CS_F_SRC_NAT;
591 		if (key->eth.type == htons(ETH_P_IP))
592 			key->ipv4.addr.src = ip_hdr(skb)->saddr;
593 		else if (key->eth.type == htons(ETH_P_IPV6))
594 			memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
595 			       sizeof(key->ipv6.addr.src));
596 		else
597 			return;
598 
599 		if (key->ip.proto == IPPROTO_UDP)
600 			src = udp_hdr(skb)->source;
601 		else if (key->ip.proto == IPPROTO_TCP)
602 			src = tcp_hdr(skb)->source;
603 		else if (key->ip.proto == IPPROTO_SCTP)
604 			src = sctp_hdr(skb)->source;
605 		else
606 			return;
607 
608 		key->tp.src = src;
609 	} else {
610 		__be16 dst;
611 
612 		key->ct.state |= OVS_CS_F_DST_NAT;
613 		if (key->eth.type == htons(ETH_P_IP))
614 			key->ipv4.addr.dst = ip_hdr(skb)->daddr;
615 		else if (key->eth.type == htons(ETH_P_IPV6))
616 			memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
617 			       sizeof(key->ipv6.addr.dst));
618 		else
619 			return;
620 
621 		if (key->ip.proto == IPPROTO_UDP)
622 			dst = udp_hdr(skb)->dest;
623 		else if (key->ip.proto == IPPROTO_TCP)
624 			dst = tcp_hdr(skb)->dest;
625 		else if (key->ip.proto == IPPROTO_SCTP)
626 			dst = sctp_hdr(skb)->dest;
627 		else
628 			return;
629 
630 		key->tp.dst = dst;
631 	}
632 }
633 
634 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
635 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
636 		      const struct ovs_conntrack_info *info,
637 		      struct sk_buff *skb, struct nf_conn *ct,
638 		      enum ip_conntrack_info ctinfo)
639 {
640 	enum nf_nat_manip_type maniptype;
641 	int err;
642 
643 	if (nf_ct_is_untracked(ct)) {
644 		/* A NAT action may only be performed on tracked packets. */
645 		return NF_ACCEPT;
646 	}
647 
648 	/* Add NAT extension if not confirmed yet. */
649 	if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
650 		return NF_ACCEPT;   /* Can't NAT. */
651 
652 	/* Determine NAT type.
653 	 * Check if the NAT type can be deduced from the tracked connection.
654 	 * Make sure new expected connections (IP_CT_RELATED) are NATted only
655 	 * when committing.
656 	 */
657 	if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
658 	    ct->status & IPS_NAT_MASK &&
659 	    (ctinfo != IP_CT_RELATED || info->commit)) {
660 		/* NAT an established or related connection like before. */
661 		if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
662 			/* This is the REPLY direction for a connection
663 			 * for which NAT was applied in the forward
664 			 * direction.  Do the reverse NAT.
665 			 */
666 			maniptype = ct->status & IPS_SRC_NAT
667 				? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
668 		else
669 			maniptype = ct->status & IPS_SRC_NAT
670 				? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
671 	} else if (info->nat & OVS_CT_SRC_NAT) {
672 		maniptype = NF_NAT_MANIP_SRC;
673 	} else if (info->nat & OVS_CT_DST_NAT) {
674 		maniptype = NF_NAT_MANIP_DST;
675 	} else {
676 		return NF_ACCEPT; /* Connection is not NATed. */
677 	}
678 	err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
679 
680 	/* Mark NAT done if successful and update the flow key. */
681 	if (err == NF_ACCEPT)
682 		ovs_nat_update_key(key, skb, maniptype);
683 
684 	return err;
685 }
686 #else /* !CONFIG_NF_NAT_NEEDED */
687 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
688 		      const struct ovs_conntrack_info *info,
689 		      struct sk_buff *skb, struct nf_conn *ct,
690 		      enum ip_conntrack_info ctinfo)
691 {
692 	return NF_ACCEPT;
693 }
694 #endif
695 
696 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
697  * not done already.  Update key with new CT state after passing the packet
698  * through conntrack.
699  * Note that if the packet is deemed invalid by conntrack, skb->nfct will be
700  * set to NULL and 0 will be returned.
701  */
702 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
703 			   const struct ovs_conntrack_info *info,
704 			   struct sk_buff *skb)
705 {
706 	/* If we are recirculating packets to match on conntrack fields and
707 	 * committing with a separate conntrack action,  then we don't need to
708 	 * actually run the packet through conntrack twice unless it's for a
709 	 * different zone.
710 	 */
711 	bool cached = skb_nfct_cached(net, key, info, skb);
712 	enum ip_conntrack_info ctinfo;
713 	struct nf_conn *ct;
714 
715 	if (!cached) {
716 		struct nf_conn *tmpl = info->ct;
717 		int err;
718 
719 		/* Associate skb with specified zone. */
720 		if (tmpl) {
721 			if (skb->nfct)
722 				nf_conntrack_put(skb->nfct);
723 			nf_conntrack_get(&tmpl->ct_general);
724 			skb->nfct = &tmpl->ct_general;
725 			skb->nfctinfo = IP_CT_NEW;
726 		}
727 
728 		err = nf_conntrack_in(net, info->family,
729 				      NF_INET_PRE_ROUTING, skb);
730 		if (err != NF_ACCEPT)
731 			return -ENOENT;
732 
733 		/* Clear CT state NAT flags to mark that we have not yet done
734 		 * NAT after the nf_conntrack_in() call.  We can actually clear
735 		 * the whole state, as it will be re-initialized below.
736 		 */
737 		key->ct.state = 0;
738 
739 		/* Update the key, but keep the NAT flags. */
740 		ovs_ct_update_key(skb, info, key, true, true);
741 	}
742 
743 	ct = nf_ct_get(skb, &ctinfo);
744 	if (ct) {
745 		/* Packets starting a new connection must be NATted before the
746 		 * helper, so that the helper knows about the NAT.  We enforce
747 		 * this by delaying both NAT and helper calls for unconfirmed
748 		 * connections until the committing CT action.  For later
749 		 * packets NAT and Helper may be called in either order.
750 		 *
751 		 * NAT will be done only if the CT action has NAT, and only
752 		 * once per packet (per zone), as guarded by the NAT bits in
753 		 * the key->ct.state.
754 		 */
755 		if (info->nat && !(key->ct.state & OVS_CS_F_NAT_MASK) &&
756 		    (nf_ct_is_confirmed(ct) || info->commit) &&
757 		    ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
758 			return -EINVAL;
759 		}
760 
761 		/* Userspace may decide to perform a ct lookup without a helper
762 		 * specified followed by a (recirculate and) commit with one.
763 		 * Therefore, for unconfirmed connections which we will commit,
764 		 * we need to attach the helper here.
765 		 */
766 		if (!nf_ct_is_confirmed(ct) && info->commit &&
767 		    info->helper && !nfct_help(ct)) {
768 			int err = __nf_ct_try_assign_helper(ct, info->ct,
769 							    GFP_ATOMIC);
770 			if (err)
771 				return err;
772 		}
773 
774 		/* Call the helper only if:
775 		 * - nf_conntrack_in() was executed above ("!cached") for a
776 		 *   confirmed connection, or
777 		 * - When committing an unconfirmed connection.
778 		 */
779 		if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
780 		    ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
781 			return -EINVAL;
782 		}
783 	}
784 
785 	return 0;
786 }
787 
788 /* Lookup connection and read fields into key. */
789 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
790 			 const struct ovs_conntrack_info *info,
791 			 struct sk_buff *skb)
792 {
793 	struct nf_conntrack_expect *exp;
794 
795 	/* If we pass an expected packet through nf_conntrack_in() the
796 	 * expectation is typically removed, but the packet could still be
797 	 * lost in upcall processing.  To prevent this from happening we
798 	 * perform an explicit expectation lookup.  Expected connections are
799 	 * always new, and will be passed through conntrack only when they are
800 	 * committed, as it is OK to remove the expectation at that time.
801 	 */
802 	exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
803 	if (exp) {
804 		u8 state;
805 
806 		/* NOTE: New connections are NATted and Helped only when
807 		 * committed, so we are not calling into NAT here.
808 		 */
809 		state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
810 		__ovs_ct_update_key(key, state, &info->zone, exp->master);
811 	} else {
812 		struct nf_conn *ct;
813 		int err;
814 
815 		err = __ovs_ct_lookup(net, key, info, skb);
816 		if (err)
817 			return err;
818 
819 		ct = (struct nf_conn *)skb->nfct;
820 		if (ct)
821 			nf_ct_deliver_cached_events(ct);
822 	}
823 
824 	return 0;
825 }
826 
827 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
828 {
829 	size_t i;
830 
831 	for (i = 0; i < sizeof(*labels); i++)
832 		if (labels->ct_labels[i])
833 			return true;
834 
835 	return false;
836 }
837 
838 /* Lookup connection and confirm if unconfirmed. */
839 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
840 			 const struct ovs_conntrack_info *info,
841 			 struct sk_buff *skb)
842 {
843 	int err;
844 
845 	err = __ovs_ct_lookup(net, key, info, skb);
846 	if (err)
847 		return err;
848 
849 	/* Apply changes before confirming the connection so that the initial
850 	 * conntrack NEW netlink event carries the values given in the CT
851 	 * action.
852 	 */
853 	if (info->mark.mask) {
854 		err = ovs_ct_set_mark(skb, key, info->mark.value,
855 				      info->mark.mask);
856 		if (err)
857 			return err;
858 	}
859 	if (labels_nonzero(&info->labels.mask)) {
860 		err = ovs_ct_set_labels(skb, key, &info->labels.value,
861 					&info->labels.mask);
862 		if (err)
863 			return err;
864 	}
865 	/* This will take care of sending queued events even if the connection
866 	 * is already confirmed.
867 	 */
868 	if (nf_conntrack_confirm(skb) != NF_ACCEPT)
869 		return -EINVAL;
870 
871 	return 0;
872 }
873 
874 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
875  * value if 'skb' is freed.
876  */
877 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
878 		   struct sw_flow_key *key,
879 		   const struct ovs_conntrack_info *info)
880 {
881 	int nh_ofs;
882 	int err;
883 
884 	/* The conntrack module expects to be working at L3. */
885 	nh_ofs = skb_network_offset(skb);
886 	skb_pull(skb, nh_ofs);
887 
888 	if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
889 		err = handle_fragments(net, key, info->zone.id, skb);
890 		if (err)
891 			return err;
892 	}
893 
894 	if (info->commit)
895 		err = ovs_ct_commit(net, key, info, skb);
896 	else
897 		err = ovs_ct_lookup(net, key, info, skb);
898 
899 	skb_push(skb, nh_ofs);
900 	if (err)
901 		kfree_skb(skb);
902 	return err;
903 }
904 
905 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
906 			     const struct sw_flow_key *key, bool log)
907 {
908 	struct nf_conntrack_helper *helper;
909 	struct nf_conn_help *help;
910 
911 	helper = nf_conntrack_helper_try_module_get(name, info->family,
912 						    key->ip.proto);
913 	if (!helper) {
914 		OVS_NLERR(log, "Unknown helper \"%s\"", name);
915 		return -EINVAL;
916 	}
917 
918 	help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
919 	if (!help) {
920 		module_put(helper->me);
921 		return -ENOMEM;
922 	}
923 
924 	rcu_assign_pointer(help->helper, helper);
925 	info->helper = helper;
926 	return 0;
927 }
928 
929 #ifdef CONFIG_NF_NAT_NEEDED
930 static int parse_nat(const struct nlattr *attr,
931 		     struct ovs_conntrack_info *info, bool log)
932 {
933 	struct nlattr *a;
934 	int rem;
935 	bool have_ip_max = false;
936 	bool have_proto_max = false;
937 	bool ip_vers = (info->family == NFPROTO_IPV6);
938 
939 	nla_for_each_nested(a, attr, rem) {
940 		static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
941 			[OVS_NAT_ATTR_SRC] = {0, 0},
942 			[OVS_NAT_ATTR_DST] = {0, 0},
943 			[OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
944 						 sizeof(struct in6_addr)},
945 			[OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
946 						 sizeof(struct in6_addr)},
947 			[OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
948 			[OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
949 			[OVS_NAT_ATTR_PERSISTENT] = {0, 0},
950 			[OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
951 			[OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
952 		};
953 		int type = nla_type(a);
954 
955 		if (type > OVS_NAT_ATTR_MAX) {
956 			OVS_NLERR(log,
957 				  "Unknown NAT attribute (type=%d, max=%d).\n",
958 				  type, OVS_NAT_ATTR_MAX);
959 			return -EINVAL;
960 		}
961 
962 		if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
963 			OVS_NLERR(log,
964 				  "NAT attribute type %d has unexpected length (%d != %d).\n",
965 				  type, nla_len(a),
966 				  ovs_nat_attr_lens[type][ip_vers]);
967 			return -EINVAL;
968 		}
969 
970 		switch (type) {
971 		case OVS_NAT_ATTR_SRC:
972 		case OVS_NAT_ATTR_DST:
973 			if (info->nat) {
974 				OVS_NLERR(log,
975 					  "Only one type of NAT may be specified.\n"
976 					  );
977 				return -ERANGE;
978 			}
979 			info->nat |= OVS_CT_NAT;
980 			info->nat |= ((type == OVS_NAT_ATTR_SRC)
981 					? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
982 			break;
983 
984 		case OVS_NAT_ATTR_IP_MIN:
985 			nla_memcpy(&info->range.min_addr, a,
986 				   sizeof(info->range.min_addr));
987 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
988 			break;
989 
990 		case OVS_NAT_ATTR_IP_MAX:
991 			have_ip_max = true;
992 			nla_memcpy(&info->range.max_addr, a,
993 				   sizeof(info->range.max_addr));
994 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
995 			break;
996 
997 		case OVS_NAT_ATTR_PROTO_MIN:
998 			info->range.min_proto.all = htons(nla_get_u16(a));
999 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1000 			break;
1001 
1002 		case OVS_NAT_ATTR_PROTO_MAX:
1003 			have_proto_max = true;
1004 			info->range.max_proto.all = htons(nla_get_u16(a));
1005 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1006 			break;
1007 
1008 		case OVS_NAT_ATTR_PERSISTENT:
1009 			info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1010 			break;
1011 
1012 		case OVS_NAT_ATTR_PROTO_HASH:
1013 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1014 			break;
1015 
1016 		case OVS_NAT_ATTR_PROTO_RANDOM:
1017 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1018 			break;
1019 
1020 		default:
1021 			OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1022 			return -EINVAL;
1023 		}
1024 	}
1025 
1026 	if (rem > 0) {
1027 		OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1028 		return -EINVAL;
1029 	}
1030 	if (!info->nat) {
1031 		/* Do not allow flags if no type is given. */
1032 		if (info->range.flags) {
1033 			OVS_NLERR(log,
1034 				  "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1035 				  );
1036 			return -EINVAL;
1037 		}
1038 		info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1039 	} else if (!info->commit) {
1040 		OVS_NLERR(log,
1041 			  "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1042 			  );
1043 		return -EINVAL;
1044 	}
1045 	/* Allow missing IP_MAX. */
1046 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1047 		memcpy(&info->range.max_addr, &info->range.min_addr,
1048 		       sizeof(info->range.max_addr));
1049 	}
1050 	/* Allow missing PROTO_MAX. */
1051 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1052 	    !have_proto_max) {
1053 		info->range.max_proto.all = info->range.min_proto.all;
1054 	}
1055 	return 0;
1056 }
1057 #endif
1058 
1059 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1060 	[OVS_CT_ATTR_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1061 	[OVS_CT_ATTR_ZONE]	= { .minlen = sizeof(u16),
1062 				    .maxlen = sizeof(u16) },
1063 	[OVS_CT_ATTR_MARK]	= { .minlen = sizeof(struct md_mark),
1064 				    .maxlen = sizeof(struct md_mark) },
1065 	[OVS_CT_ATTR_LABELS]	= { .minlen = sizeof(struct md_labels),
1066 				    .maxlen = sizeof(struct md_labels) },
1067 	[OVS_CT_ATTR_HELPER]	= { .minlen = 1,
1068 				    .maxlen = NF_CT_HELPER_NAME_LEN },
1069 #ifdef CONFIG_NF_NAT_NEEDED
1070 	/* NAT length is checked when parsing the nested attributes. */
1071 	[OVS_CT_ATTR_NAT]	= { .minlen = 0, .maxlen = INT_MAX },
1072 #endif
1073 };
1074 
1075 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1076 		    const char **helper, bool log)
1077 {
1078 	struct nlattr *a;
1079 	int rem;
1080 
1081 	nla_for_each_nested(a, attr, rem) {
1082 		int type = nla_type(a);
1083 		int maxlen = ovs_ct_attr_lens[type].maxlen;
1084 		int minlen = ovs_ct_attr_lens[type].minlen;
1085 
1086 		if (type > OVS_CT_ATTR_MAX) {
1087 			OVS_NLERR(log,
1088 				  "Unknown conntrack attr (type=%d, max=%d)",
1089 				  type, OVS_CT_ATTR_MAX);
1090 			return -EINVAL;
1091 		}
1092 		if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1093 			OVS_NLERR(log,
1094 				  "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1095 				  type, nla_len(a), maxlen);
1096 			return -EINVAL;
1097 		}
1098 
1099 		switch (type) {
1100 		case OVS_CT_ATTR_COMMIT:
1101 			info->commit = true;
1102 			break;
1103 #ifdef CONFIG_NF_CONNTRACK_ZONES
1104 		case OVS_CT_ATTR_ZONE:
1105 			info->zone.id = nla_get_u16(a);
1106 			break;
1107 #endif
1108 #ifdef CONFIG_NF_CONNTRACK_MARK
1109 		case OVS_CT_ATTR_MARK: {
1110 			struct md_mark *mark = nla_data(a);
1111 
1112 			if (!mark->mask) {
1113 				OVS_NLERR(log, "ct_mark mask cannot be 0");
1114 				return -EINVAL;
1115 			}
1116 			info->mark = *mark;
1117 			break;
1118 		}
1119 #endif
1120 #ifdef CONFIG_NF_CONNTRACK_LABELS
1121 		case OVS_CT_ATTR_LABELS: {
1122 			struct md_labels *labels = nla_data(a);
1123 
1124 			if (!labels_nonzero(&labels->mask)) {
1125 				OVS_NLERR(log, "ct_labels mask cannot be 0");
1126 				return -EINVAL;
1127 			}
1128 			info->labels = *labels;
1129 			break;
1130 		}
1131 #endif
1132 		case OVS_CT_ATTR_HELPER:
1133 			*helper = nla_data(a);
1134 			if (!memchr(*helper, '\0', nla_len(a))) {
1135 				OVS_NLERR(log, "Invalid conntrack helper");
1136 				return -EINVAL;
1137 			}
1138 			break;
1139 #ifdef CONFIG_NF_NAT_NEEDED
1140 		case OVS_CT_ATTR_NAT: {
1141 			int err = parse_nat(a, info, log);
1142 
1143 			if (err)
1144 				return err;
1145 			break;
1146 		}
1147 #endif
1148 		default:
1149 			OVS_NLERR(log, "Unknown conntrack attr (%d)",
1150 				  type);
1151 			return -EINVAL;
1152 		}
1153 	}
1154 
1155 #ifdef CONFIG_NF_CONNTRACK_MARK
1156 	if (!info->commit && info->mark.mask) {
1157 		OVS_NLERR(log,
1158 			  "Setting conntrack mark requires 'commit' flag.");
1159 		return -EINVAL;
1160 	}
1161 #endif
1162 #ifdef CONFIG_NF_CONNTRACK_LABELS
1163 	if (!info->commit && labels_nonzero(&info->labels.mask)) {
1164 		OVS_NLERR(log,
1165 			  "Setting conntrack labels requires 'commit' flag.");
1166 		return -EINVAL;
1167 	}
1168 #endif
1169 	if (rem > 0) {
1170 		OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1171 		return -EINVAL;
1172 	}
1173 
1174 	return 0;
1175 }
1176 
1177 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1178 {
1179 	if (attr == OVS_KEY_ATTR_CT_STATE)
1180 		return true;
1181 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1182 	    attr == OVS_KEY_ATTR_CT_ZONE)
1183 		return true;
1184 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1185 	    attr == OVS_KEY_ATTR_CT_MARK)
1186 		return true;
1187 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1188 	    attr == OVS_KEY_ATTR_CT_LABELS) {
1189 		struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1190 
1191 		return ovs_net->xt_label;
1192 	}
1193 
1194 	return false;
1195 }
1196 
1197 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1198 		       const struct sw_flow_key *key,
1199 		       struct sw_flow_actions **sfa,  bool log)
1200 {
1201 	struct ovs_conntrack_info ct_info;
1202 	const char *helper = NULL;
1203 	u16 family;
1204 	int err;
1205 
1206 	family = key_to_nfproto(key);
1207 	if (family == NFPROTO_UNSPEC) {
1208 		OVS_NLERR(log, "ct family unspecified");
1209 		return -EINVAL;
1210 	}
1211 
1212 	memset(&ct_info, 0, sizeof(ct_info));
1213 	ct_info.family = family;
1214 
1215 	nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1216 			NF_CT_DEFAULT_ZONE_DIR, 0);
1217 
1218 	err = parse_ct(attr, &ct_info, &helper, log);
1219 	if (err)
1220 		return err;
1221 
1222 	/* Set up template for tracking connections in specific zones. */
1223 	ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1224 	if (!ct_info.ct) {
1225 		OVS_NLERR(log, "Failed to allocate conntrack template");
1226 		return -ENOMEM;
1227 	}
1228 
1229 	__set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1230 	nf_conntrack_get(&ct_info.ct->ct_general);
1231 
1232 	if (helper) {
1233 		err = ovs_ct_add_helper(&ct_info, helper, key, log);
1234 		if (err)
1235 			goto err_free_ct;
1236 	}
1237 
1238 	err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1239 				 sizeof(ct_info), log);
1240 	if (err)
1241 		goto err_free_ct;
1242 
1243 	return 0;
1244 err_free_ct:
1245 	__ovs_ct_free_action(&ct_info);
1246 	return err;
1247 }
1248 
1249 #ifdef CONFIG_NF_NAT_NEEDED
1250 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1251 			       struct sk_buff *skb)
1252 {
1253 	struct nlattr *start;
1254 
1255 	start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1256 	if (!start)
1257 		return false;
1258 
1259 	if (info->nat & OVS_CT_SRC_NAT) {
1260 		if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1261 			return false;
1262 	} else if (info->nat & OVS_CT_DST_NAT) {
1263 		if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1264 			return false;
1265 	} else {
1266 		goto out;
1267 	}
1268 
1269 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1270 		if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1271 		    info->family == NFPROTO_IPV4) {
1272 			if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1273 					    info->range.min_addr.ip) ||
1274 			    (info->range.max_addr.ip
1275 			     != info->range.min_addr.ip &&
1276 			     (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1277 					      info->range.max_addr.ip))))
1278 				return false;
1279 		} else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1280 			   info->family == NFPROTO_IPV6) {
1281 			if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1282 					     &info->range.min_addr.in6) ||
1283 			    (memcmp(&info->range.max_addr.in6,
1284 				    &info->range.min_addr.in6,
1285 				    sizeof(info->range.max_addr.in6)) &&
1286 			     (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1287 					       &info->range.max_addr.in6))))
1288 				return false;
1289 		} else {
1290 			return false;
1291 		}
1292 	}
1293 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1294 	    (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1295 			 ntohs(info->range.min_proto.all)) ||
1296 	     (info->range.max_proto.all != info->range.min_proto.all &&
1297 	      nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1298 			  ntohs(info->range.max_proto.all)))))
1299 		return false;
1300 
1301 	if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1302 	    nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1303 		return false;
1304 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1305 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1306 		return false;
1307 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1308 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1309 		return false;
1310 out:
1311 	nla_nest_end(skb, start);
1312 
1313 	return true;
1314 }
1315 #endif
1316 
1317 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1318 			  struct sk_buff *skb)
1319 {
1320 	struct nlattr *start;
1321 
1322 	start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1323 	if (!start)
1324 		return -EMSGSIZE;
1325 
1326 	if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
1327 		return -EMSGSIZE;
1328 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1329 	    nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1330 		return -EMSGSIZE;
1331 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1332 	    nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1333 		    &ct_info->mark))
1334 		return -EMSGSIZE;
1335 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1336 	    labels_nonzero(&ct_info->labels.mask) &&
1337 	    nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1338 		    &ct_info->labels))
1339 		return -EMSGSIZE;
1340 	if (ct_info->helper) {
1341 		if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1342 				   ct_info->helper->name))
1343 			return -EMSGSIZE;
1344 	}
1345 #ifdef CONFIG_NF_NAT_NEEDED
1346 	if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1347 		return -EMSGSIZE;
1348 #endif
1349 	nla_nest_end(skb, start);
1350 
1351 	return 0;
1352 }
1353 
1354 void ovs_ct_free_action(const struct nlattr *a)
1355 {
1356 	struct ovs_conntrack_info *ct_info = nla_data(a);
1357 
1358 	__ovs_ct_free_action(ct_info);
1359 }
1360 
1361 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1362 {
1363 	if (ct_info->helper)
1364 		module_put(ct_info->helper->me);
1365 	if (ct_info->ct)
1366 		nf_ct_tmpl_free(ct_info->ct);
1367 }
1368 
1369 void ovs_ct_init(struct net *net)
1370 {
1371 	unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1372 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1373 
1374 	if (nf_connlabels_get(net, n_bits - 1)) {
1375 		ovs_net->xt_label = false;
1376 		OVS_NLERR(true, "Failed to set connlabel length");
1377 	} else {
1378 		ovs_net->xt_label = true;
1379 	}
1380 }
1381 
1382 void ovs_ct_exit(struct net *net)
1383 {
1384 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1385 
1386 	if (ovs_net->xt_label)
1387 		nf_connlabels_put(net);
1388 }
1389