xref: /linux/net/openvswitch/conntrack.c (revision cf85f810f911234a06a4ef2439e8694b93b717fc)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2015 Nicira, Inc.
4  */
5 
6 #include <linux/module.h>
7 #include <linux/openvswitch.h>
8 #include <linux/tcp.h>
9 #include <linux/udp.h>
10 #include <linux/sctp.h>
11 #include <linux/static_key.h>
12 #include <linux/string_helpers.h>
13 #include <net/ip.h>
14 #include <net/genetlink.h>
15 #include <net/netfilter/nf_conntrack_core.h>
16 #include <net/netfilter/nf_conntrack_count.h>
17 #include <net/netfilter/nf_conntrack_helper.h>
18 #include <net/netfilter/nf_conntrack_labels.h>
19 #include <net/netfilter/nf_conntrack_seqadj.h>
20 #include <net/netfilter/nf_conntrack_timeout.h>
21 #include <net/netfilter/nf_conntrack_zones.h>
22 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
23 #include <net/ipv6_frag.h>
24 
25 #if IS_ENABLED(CONFIG_NF_NAT)
26 #include <net/netfilter/nf_nat.h>
27 #endif
28 
29 #include <net/netfilter/nf_conntrack_act_ct.h>
30 
31 #include "datapath.h"
32 #include "drop.h"
33 #include "conntrack.h"
34 #include "flow.h"
35 #include "flow_netlink.h"
36 
37 struct ovs_ct_len_tbl {
38 	int maxlen;
39 	int minlen;
40 };
41 
42 /* Metadata mark for masked write to conntrack mark */
43 struct md_mark {
44 	u32 value;
45 	u32 mask;
46 };
47 
48 /* Metadata label for masked write to conntrack label. */
49 struct md_labels {
50 	struct ovs_key_ct_labels value;
51 	struct ovs_key_ct_labels mask;
52 };
53 
54 enum ovs_ct_nat {
55 	OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
56 	OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
57 	OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
58 };
59 
60 /* Conntrack action context for execution. */
61 struct ovs_conntrack_info {
62 	struct nf_conntrack_helper *helper;
63 	struct nf_conntrack_zone zone;
64 	struct nf_conn *ct;
65 	u8 commit : 1;
66 	u8 nat : 3;                 /* enum ovs_ct_nat */
67 	u8 force : 1;
68 	u8 have_eventmask : 1;
69 	u16 family;
70 	u32 eventmask;              /* Mask of 1 << IPCT_*. */
71 	struct md_mark mark;
72 	struct md_labels labels;
73 	char timeout[CTNL_TIMEOUT_NAME_MAX];
74 	struct nf_ct_timeout *nf_ct_timeout;
75 #if IS_ENABLED(CONFIG_NF_NAT)
76 	struct nf_nat_range2 range;  /* Only present for SRC NAT and DST NAT. */
77 #endif
78 };
79 
80 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
81 #define OVS_CT_LIMIT_UNLIMITED	0
82 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
83 #define CT_LIMIT_HASH_BUCKETS 512
84 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
85 
86 struct ovs_ct_limit {
87 	/* Elements in ovs_ct_limit_info->limits hash table */
88 	struct hlist_node hlist_node;
89 	struct rcu_head rcu;
90 	u16 zone;
91 	u32 limit;
92 };
93 
94 struct ovs_ct_limit_info {
95 	u32 default_limit;
96 	struct hlist_head *limits;
97 	struct nf_conncount_data *data;
98 };
99 
100 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
101 	[OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
102 };
103 #endif
104 
105 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
106 
107 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
108 
109 static u16 key_to_nfproto(const struct sw_flow_key *key)
110 {
111 	switch (ntohs(key->eth.type)) {
112 	case ETH_P_IP:
113 		return NFPROTO_IPV4;
114 	case ETH_P_IPV6:
115 		return NFPROTO_IPV6;
116 	default:
117 		return NFPROTO_UNSPEC;
118 	}
119 }
120 
121 /* Map SKB connection state into the values used by flow definition. */
122 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
123 {
124 	u8 ct_state = OVS_CS_F_TRACKED;
125 
126 	switch (ctinfo) {
127 	case IP_CT_ESTABLISHED_REPLY:
128 	case IP_CT_RELATED_REPLY:
129 		ct_state |= OVS_CS_F_REPLY_DIR;
130 		break;
131 	default:
132 		break;
133 	}
134 
135 	switch (ctinfo) {
136 	case IP_CT_ESTABLISHED:
137 	case IP_CT_ESTABLISHED_REPLY:
138 		ct_state |= OVS_CS_F_ESTABLISHED;
139 		break;
140 	case IP_CT_RELATED:
141 	case IP_CT_RELATED_REPLY:
142 		ct_state |= OVS_CS_F_RELATED;
143 		break;
144 	case IP_CT_NEW:
145 		ct_state |= OVS_CS_F_NEW;
146 		break;
147 	default:
148 		break;
149 	}
150 
151 	return ct_state;
152 }
153 
154 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
155 {
156 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
157 	return ct ? READ_ONCE(ct->mark) : 0;
158 #else
159 	return 0;
160 #endif
161 }
162 
163 /* Guard against conntrack labels max size shrinking below 128 bits. */
164 #if NF_CT_LABELS_MAX_SIZE < 16
165 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
166 #endif
167 
168 static void ovs_ct_get_labels(const struct nf_conn *ct,
169 			      struct ovs_key_ct_labels *labels)
170 {
171 	struct nf_conn_labels *cl = NULL;
172 
173 	if (ct) {
174 		if (ct->master && !nf_ct_is_confirmed(ct))
175 			ct = ct->master;
176 		cl = nf_ct_labels_find(ct);
177 	}
178 	if (cl)
179 		memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
180 	else
181 		memset(labels, 0, OVS_CT_LABELS_LEN);
182 }
183 
184 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
185 					const struct nf_conntrack_tuple *orig,
186 					u8 icmp_proto)
187 {
188 	key->ct_orig_proto = orig->dst.protonum;
189 	if (orig->dst.protonum == icmp_proto) {
190 		key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
191 		key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
192 	} else {
193 		key->ct.orig_tp.src = orig->src.u.all;
194 		key->ct.orig_tp.dst = orig->dst.u.all;
195 	}
196 }
197 
198 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
199 				const struct nf_conntrack_zone *zone,
200 				const struct nf_conn *ct)
201 {
202 	key->ct_state = state;
203 	key->ct_zone = zone->id;
204 	key->ct.mark = ovs_ct_get_mark(ct);
205 	ovs_ct_get_labels(ct, &key->ct.labels);
206 
207 	if (ct) {
208 		const struct nf_conntrack_tuple *orig;
209 
210 		/* Use the master if we have one. */
211 		if (ct->master)
212 			ct = ct->master;
213 		orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
214 
215 		/* IP version must match with the master connection. */
216 		if (key->eth.type == htons(ETH_P_IP) &&
217 		    nf_ct_l3num(ct) == NFPROTO_IPV4) {
218 			key->ipv4.ct_orig.src = orig->src.u3.ip;
219 			key->ipv4.ct_orig.dst = orig->dst.u3.ip;
220 			__ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
221 			return;
222 		} else if (key->eth.type == htons(ETH_P_IPV6) &&
223 			   !sw_flow_key_is_nd(key) &&
224 			   nf_ct_l3num(ct) == NFPROTO_IPV6) {
225 			key->ipv6.ct_orig.src = orig->src.u3.in6;
226 			key->ipv6.ct_orig.dst = orig->dst.u3.in6;
227 			__ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
228 			return;
229 		}
230 	}
231 	/* Clear 'ct_orig_proto' to mark the non-existence of conntrack
232 	 * original direction key fields.
233 	 */
234 	key->ct_orig_proto = 0;
235 }
236 
237 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
238  * previously sent the packet to conntrack via the ct action.  If
239  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
240  * initialized from the connection status.
241  */
242 static void ovs_ct_update_key(const struct sk_buff *skb,
243 			      const struct ovs_conntrack_info *info,
244 			      struct sw_flow_key *key, bool post_ct,
245 			      bool keep_nat_flags)
246 {
247 	const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
248 	enum ip_conntrack_info ctinfo;
249 	struct nf_conn *ct;
250 	u8 state = 0;
251 
252 	ct = nf_ct_get(skb, &ctinfo);
253 	if (ct) {
254 		state = ovs_ct_get_state(ctinfo);
255 		/* All unconfirmed entries are NEW connections. */
256 		if (!nf_ct_is_confirmed(ct))
257 			state |= OVS_CS_F_NEW;
258 		/* OVS persists the related flag for the duration of the
259 		 * connection.
260 		 */
261 		if (ct->master)
262 			state |= OVS_CS_F_RELATED;
263 		if (keep_nat_flags) {
264 			state |= key->ct_state & OVS_CS_F_NAT_MASK;
265 		} else {
266 			if (ct->status & IPS_SRC_NAT)
267 				state |= OVS_CS_F_SRC_NAT;
268 			if (ct->status & IPS_DST_NAT)
269 				state |= OVS_CS_F_DST_NAT;
270 		}
271 		zone = nf_ct_zone(ct);
272 	} else if (post_ct) {
273 		state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
274 		if (info)
275 			zone = &info->zone;
276 	}
277 	__ovs_ct_update_key(key, state, zone, ct);
278 }
279 
280 /* This is called to initialize CT key fields possibly coming in from the local
281  * stack.
282  */
283 void ovs_ct_fill_key(const struct sk_buff *skb,
284 		     struct sw_flow_key *key,
285 		     bool post_ct)
286 {
287 	ovs_ct_update_key(skb, NULL, key, post_ct, false);
288 }
289 
290 int ovs_ct_put_key(const struct sw_flow_key *swkey,
291 		   const struct sw_flow_key *output, struct sk_buff *skb)
292 {
293 	if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
294 		return -EMSGSIZE;
295 
296 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
297 	    nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
298 		return -EMSGSIZE;
299 
300 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
301 	    nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
302 		return -EMSGSIZE;
303 
304 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
305 	    nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
306 		    &output->ct.labels))
307 		return -EMSGSIZE;
308 
309 	if (swkey->ct_orig_proto) {
310 		if (swkey->eth.type == htons(ETH_P_IP)) {
311 			struct ovs_key_ct_tuple_ipv4 orig;
312 
313 			memset(&orig, 0, sizeof(orig));
314 			orig.ipv4_src = output->ipv4.ct_orig.src;
315 			orig.ipv4_dst = output->ipv4.ct_orig.dst;
316 			orig.src_port = output->ct.orig_tp.src;
317 			orig.dst_port = output->ct.orig_tp.dst;
318 			orig.ipv4_proto = output->ct_orig_proto;
319 
320 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
321 				    sizeof(orig), &orig))
322 				return -EMSGSIZE;
323 		} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
324 			struct ovs_key_ct_tuple_ipv6 orig;
325 
326 			memset(&orig, 0, sizeof(orig));
327 			memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32,
328 			       sizeof(orig.ipv6_src));
329 			memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32,
330 			       sizeof(orig.ipv6_dst));
331 			orig.src_port = output->ct.orig_tp.src;
332 			orig.dst_port = output->ct.orig_tp.dst;
333 			orig.ipv6_proto = output->ct_orig_proto;
334 
335 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
336 				    sizeof(orig), &orig))
337 				return -EMSGSIZE;
338 		}
339 	}
340 
341 	return 0;
342 }
343 
344 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
345 			   u32 ct_mark, u32 mask)
346 {
347 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
348 	u32 new_mark;
349 
350 	new_mark = ct_mark | (READ_ONCE(ct->mark) & ~(mask));
351 	if (READ_ONCE(ct->mark) != new_mark) {
352 		WRITE_ONCE(ct->mark, new_mark);
353 		if (nf_ct_is_confirmed(ct))
354 			nf_conntrack_event_cache(IPCT_MARK, ct);
355 		key->ct.mark = new_mark;
356 	}
357 
358 	return 0;
359 #else
360 	return -ENOTSUPP;
361 #endif
362 }
363 
364 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
365 {
366 	struct nf_conn_labels *cl;
367 
368 	cl = nf_ct_labels_find(ct);
369 	if (!cl) {
370 		nf_ct_labels_ext_add(ct);
371 		cl = nf_ct_labels_find(ct);
372 	}
373 
374 	return cl;
375 }
376 
377 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
378  * since the new connection is not yet confirmed, and thus no-one else has
379  * access to it's labels, we simply write them over.
380  */
381 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
382 			      const struct ovs_key_ct_labels *labels,
383 			      const struct ovs_key_ct_labels *mask)
384 {
385 	struct nf_conn_labels *cl, *master_cl;
386 	bool have_mask = labels_nonzero(mask);
387 
388 	/* Inherit master's labels to the related connection? */
389 	master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
390 
391 	if (!master_cl && !have_mask)
392 		return 0;   /* Nothing to do. */
393 
394 	cl = ovs_ct_get_conn_labels(ct);
395 	if (!cl)
396 		return -ENOSPC;
397 
398 	/* Inherit the master's labels, if any. */
399 	if (master_cl)
400 		*cl = *master_cl;
401 
402 	if (have_mask) {
403 		u32 *dst = (u32 *)cl->bits;
404 		int i;
405 
406 		for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
407 			dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
408 				(labels->ct_labels_32[i]
409 				 & mask->ct_labels_32[i]);
410 	}
411 
412 	/* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
413 	 * IPCT_LABEL bit is set in the event cache.
414 	 */
415 	nf_conntrack_event_cache(IPCT_LABEL, ct);
416 
417 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
418 
419 	return 0;
420 }
421 
422 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
423 			     const struct ovs_key_ct_labels *labels,
424 			     const struct ovs_key_ct_labels *mask)
425 {
426 	struct nf_conn_labels *cl;
427 	int err;
428 
429 	cl = ovs_ct_get_conn_labels(ct);
430 	if (!cl)
431 		return -ENOSPC;
432 
433 	err = nf_connlabels_replace(ct, labels->ct_labels_32,
434 				    mask->ct_labels_32,
435 				    OVS_CT_LABELS_LEN_32);
436 	if (err)
437 		return err;
438 
439 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
440 
441 	return 0;
442 }
443 
444 static int ovs_ct_handle_fragments(struct net *net, struct sw_flow_key *key,
445 				   u16 zone, int family, struct sk_buff *skb)
446 {
447 	struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
448 	int err;
449 
450 	err = nf_ct_handle_fragments(net, skb, zone, family, &key->ip.proto, &ovs_cb.mru);
451 	if (err)
452 		return err;
453 
454 	/* The key extracted from the fragment that completed this datagram
455 	 * likely didn't have an L4 header, so regenerate it.
456 	 */
457 	ovs_flow_key_update_l3l4(skb, key);
458 	key->ip.frag = OVS_FRAG_TYPE_NONE;
459 	*OVS_CB(skb) = ovs_cb;
460 
461 	return 0;
462 }
463 
464 /* This replicates logic from nf_conntrack_core.c that is not exported. */
465 static enum ip_conntrack_info
466 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
467 {
468 	const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
469 
470 	if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
471 		return IP_CT_ESTABLISHED_REPLY;
472 	/* Once we've had two way comms, always ESTABLISHED. */
473 	if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
474 		return IP_CT_ESTABLISHED;
475 	if (test_bit(IPS_EXPECTED_BIT, &ct->status))
476 		return IP_CT_RELATED;
477 	return IP_CT_NEW;
478 }
479 
480 /* Find an existing connection which this packet belongs to without
481  * re-attributing statistics or modifying the connection state.  This allows an
482  * skb->_nfct lost due to an upcall to be recovered during actions execution.
483  *
484  * Must be called with rcu_read_lock.
485  *
486  * On success, populates skb->_nfct and returns the connection.  Returns NULL
487  * if there is no existing entry.
488  */
489 static struct nf_conn *
490 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
491 		     u8 l3num, struct sk_buff *skb, bool natted)
492 {
493 	struct nf_conntrack_tuple tuple;
494 	struct nf_conntrack_tuple_hash *h;
495 	struct nf_conn *ct;
496 
497 	if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
498 			       net, &tuple)) {
499 		pr_debug("ovs_ct_find_existing: Can't get tuple\n");
500 		return NULL;
501 	}
502 
503 	/* Must invert the tuple if skb has been transformed by NAT. */
504 	if (natted) {
505 		struct nf_conntrack_tuple inverse;
506 
507 		if (!nf_ct_invert_tuple(&inverse, &tuple)) {
508 			pr_debug("ovs_ct_find_existing: Inversion failed!\n");
509 			return NULL;
510 		}
511 		tuple = inverse;
512 	}
513 
514 	/* look for tuple match */
515 	h = nf_conntrack_find_get(net, zone, &tuple);
516 	if (!h)
517 		return NULL;   /* Not found. */
518 
519 	ct = nf_ct_tuplehash_to_ctrack(h);
520 
521 	/* Inverted packet tuple matches the reverse direction conntrack tuple,
522 	 * select the other tuplehash to get the right 'ctinfo' bits for this
523 	 * packet.
524 	 */
525 	if (natted)
526 		h = &ct->tuplehash[!h->tuple.dst.dir];
527 
528 	nf_ct_set(skb, ct, ovs_ct_get_info(h));
529 	return ct;
530 }
531 
532 static
533 struct nf_conn *ovs_ct_executed(struct net *net,
534 				const struct sw_flow_key *key,
535 				const struct ovs_conntrack_info *info,
536 				struct sk_buff *skb,
537 				bool *ct_executed)
538 {
539 	struct nf_conn *ct = NULL;
540 
541 	/* If no ct, check if we have evidence that an existing conntrack entry
542 	 * might be found for this skb.  This happens when we lose a skb->_nfct
543 	 * due to an upcall, or if the direction is being forced.  If the
544 	 * connection was not confirmed, it is not cached and needs to be run
545 	 * through conntrack again.
546 	 */
547 	*ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
548 		       !(key->ct_state & OVS_CS_F_INVALID) &&
549 		       (key->ct_zone == info->zone.id);
550 
551 	if (*ct_executed || (!key->ct_state && info->force)) {
552 		ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
553 					  !!(key->ct_state &
554 					  OVS_CS_F_NAT_MASK));
555 	}
556 
557 	return ct;
558 }
559 
560 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
561 static bool skb_nfct_cached(struct net *net,
562 			    const struct sw_flow_key *key,
563 			    const struct ovs_conntrack_info *info,
564 			    struct sk_buff *skb)
565 {
566 	enum ip_conntrack_info ctinfo;
567 	struct nf_conn *ct;
568 	bool ct_executed = true;
569 
570 	ct = nf_ct_get(skb, &ctinfo);
571 	if (!ct)
572 		ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
573 
574 	if (ct)
575 		nf_ct_get(skb, &ctinfo);
576 	else
577 		return false;
578 
579 	if (!net_eq(net, read_pnet(&ct->ct_net)))
580 		return false;
581 	if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
582 		return false;
583 	if (info->helper) {
584 		struct nf_conn_help *help;
585 
586 		help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
587 		if (help && rcu_access_pointer(help->helper) != info->helper)
588 			return false;
589 	}
590 	if (info->nf_ct_timeout) {
591 		struct nf_conn_timeout *timeout_ext;
592 
593 		timeout_ext = nf_ct_timeout_find(ct);
594 		if (!timeout_ext || info->nf_ct_timeout !=
595 		    rcu_dereference(timeout_ext->timeout))
596 			return false;
597 	}
598 	/* Force conntrack entry direction to the current packet? */
599 	if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
600 		/* Delete the conntrack entry if confirmed, else just release
601 		 * the reference.
602 		 */
603 		if (nf_ct_is_confirmed(ct))
604 			nf_ct_delete(ct, 0, 0);
605 
606 		nf_reset_ct(skb);
607 		nf_ct_set(skb, NULL, 0);
608 		return false;
609 	}
610 
611 	return ct_executed;
612 }
613 
614 #if IS_ENABLED(CONFIG_NF_NAT)
615 static void ovs_nat_update_key(struct sw_flow_key *key,
616 			       const struct sk_buff *skb,
617 			       enum nf_nat_manip_type maniptype)
618 {
619 	if (maniptype == NF_NAT_MANIP_SRC) {
620 		__be16 src;
621 
622 		key->ct_state |= OVS_CS_F_SRC_NAT;
623 		if (key->eth.type == htons(ETH_P_IP))
624 			key->ipv4.addr.src = ip_hdr(skb)->saddr;
625 		else if (key->eth.type == htons(ETH_P_IPV6))
626 			memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
627 			       sizeof(key->ipv6.addr.src));
628 		else
629 			return;
630 
631 		if (key->ip.proto == IPPROTO_UDP)
632 			src = udp_hdr(skb)->source;
633 		else if (key->ip.proto == IPPROTO_TCP)
634 			src = tcp_hdr(skb)->source;
635 		else if (key->ip.proto == IPPROTO_SCTP)
636 			src = sctp_hdr(skb)->source;
637 		else
638 			return;
639 
640 		key->tp.src = src;
641 	} else {
642 		__be16 dst;
643 
644 		key->ct_state |= OVS_CS_F_DST_NAT;
645 		if (key->eth.type == htons(ETH_P_IP))
646 			key->ipv4.addr.dst = ip_hdr(skb)->daddr;
647 		else if (key->eth.type == htons(ETH_P_IPV6))
648 			memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
649 			       sizeof(key->ipv6.addr.dst));
650 		else
651 			return;
652 
653 		if (key->ip.proto == IPPROTO_UDP)
654 			dst = udp_hdr(skb)->dest;
655 		else if (key->ip.proto == IPPROTO_TCP)
656 			dst = tcp_hdr(skb)->dest;
657 		else if (key->ip.proto == IPPROTO_SCTP)
658 			dst = sctp_hdr(skb)->dest;
659 		else
660 			return;
661 
662 		key->tp.dst = dst;
663 	}
664 }
665 
666 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
667 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
668 		      const struct ovs_conntrack_info *info,
669 		      struct sk_buff *skb, struct nf_conn *ct,
670 		      enum ip_conntrack_info ctinfo)
671 {
672 	int err, action = 0;
673 
674 	if (!(info->nat & OVS_CT_NAT))
675 		return NF_ACCEPT;
676 	if (info->nat & OVS_CT_SRC_NAT)
677 		action |= BIT(NF_NAT_MANIP_SRC);
678 	if (info->nat & OVS_CT_DST_NAT)
679 		action |= BIT(NF_NAT_MANIP_DST);
680 
681 	err = nf_ct_nat(skb, ct, ctinfo, &action, &info->range, info->commit);
682 	if (err != NF_ACCEPT)
683 		return err;
684 
685 	if (action & BIT(NF_NAT_MANIP_SRC))
686 		ovs_nat_update_key(key, skb, NF_NAT_MANIP_SRC);
687 	if (action & BIT(NF_NAT_MANIP_DST))
688 		ovs_nat_update_key(key, skb, NF_NAT_MANIP_DST);
689 
690 	return err;
691 }
692 #else /* !CONFIG_NF_NAT */
693 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
694 		      const struct ovs_conntrack_info *info,
695 		      struct sk_buff *skb, struct nf_conn *ct,
696 		      enum ip_conntrack_info ctinfo)
697 {
698 	return NF_ACCEPT;
699 }
700 #endif
701 
702 static int verdict_to_errno(unsigned int verdict)
703 {
704 	switch (verdict & NF_VERDICT_MASK) {
705 	case NF_ACCEPT:
706 		return 0;
707 	case NF_DROP:
708 		return -EINVAL;
709 	case NF_STOLEN:
710 		return -EINPROGRESS;
711 	default:
712 		break;
713 	}
714 
715 	return -EINVAL;
716 }
717 
718 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
719  * not done already.  Update key with new CT state after passing the packet
720  * through conntrack.
721  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
722  * set to NULL and 0 will be returned.
723  */
724 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
725 			   const struct ovs_conntrack_info *info,
726 			   struct sk_buff *skb)
727 {
728 	/* If we are recirculating packets to match on conntrack fields and
729 	 * committing with a separate conntrack action,  then we don't need to
730 	 * actually run the packet through conntrack twice unless it's for a
731 	 * different zone.
732 	 */
733 	bool cached = skb_nfct_cached(net, key, info, skb);
734 	enum ip_conntrack_info ctinfo;
735 	struct nf_conn *ct;
736 
737 	if (!cached) {
738 		struct nf_hook_state state = {
739 			.hook = NF_INET_PRE_ROUTING,
740 			.pf = info->family,
741 			.net = net,
742 		};
743 		struct nf_conn *tmpl = info->ct;
744 		int err;
745 
746 		/* Associate skb with specified zone. */
747 		if (tmpl) {
748 			nf_reset_ct(skb);
749 			nf_conntrack_get(&tmpl->ct_general);
750 			nf_ct_set(skb, tmpl, IP_CT_NEW);
751 		}
752 
753 		err = nf_conntrack_in(skb, &state);
754 		if (err != NF_ACCEPT)
755 			return verdict_to_errno(err);
756 
757 		/* Clear CT state NAT flags to mark that we have not yet done
758 		 * NAT after the nf_conntrack_in() call.  We can actually clear
759 		 * the whole state, as it will be re-initialized below.
760 		 */
761 		key->ct_state = 0;
762 
763 		/* Update the key, but keep the NAT flags. */
764 		ovs_ct_update_key(skb, info, key, true, true);
765 	}
766 
767 	ct = nf_ct_get(skb, &ctinfo);
768 	if (ct) {
769 		bool add_helper = false;
770 
771 		/* Packets starting a new connection must be NATted before the
772 		 * helper, so that the helper knows about the NAT.  We enforce
773 		 * this by delaying both NAT and helper calls for unconfirmed
774 		 * connections until the committing CT action.  For later
775 		 * packets NAT and Helper may be called in either order.
776 		 *
777 		 * NAT will be done only if the CT action has NAT, and only
778 		 * once per packet (per zone), as guarded by the NAT bits in
779 		 * the key->ct_state.
780 		 */
781 		if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
782 		    (nf_ct_is_confirmed(ct) || info->commit)) {
783 			int err = ovs_ct_nat(net, key, info, skb, ct, ctinfo);
784 
785 			err = verdict_to_errno(err);
786 			if (err)
787 				return err;
788 		}
789 
790 		/* Userspace may decide to perform a ct lookup without a helper
791 		 * specified followed by a (recirculate and) commit with one,
792 		 * or attach a helper in a later commit.  Therefore, for
793 		 * connections which we will commit, we may need to attach
794 		 * the helper here.
795 		 */
796 		if (!nf_ct_is_confirmed(ct) && info->commit &&
797 		    info->helper && !nfct_help(ct)) {
798 			int err = __nf_ct_try_assign_helper(ct, info->ct,
799 							    GFP_ATOMIC);
800 			if (err)
801 				return err;
802 			add_helper = true;
803 
804 			/* helper installed, add seqadj if NAT is required */
805 			if (info->nat && !nfct_seqadj(ct)) {
806 				if (!nfct_seqadj_ext_add(ct))
807 					return -EINVAL;
808 			}
809 		}
810 
811 		/* Call the helper only if:
812 		 * - nf_conntrack_in() was executed above ("!cached") or a
813 		 *   helper was just attached ("add_helper") for a confirmed
814 		 *   connection, or
815 		 * - When committing an unconfirmed connection.
816 		 */
817 		if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
818 					      info->commit)) {
819 			int err = nf_ct_helper(skb, ct, ctinfo, info->family);
820 
821 			err = verdict_to_errno(err);
822 			if (err)
823 				return err;
824 		}
825 
826 		if (nf_ct_protonum(ct) == IPPROTO_TCP &&
827 		    nf_ct_is_confirmed(ct) && nf_conntrack_tcp_established(ct)) {
828 			/* Be liberal for tcp packets so that out-of-window
829 			 * packets are not marked invalid.
830 			 */
831 			nf_ct_set_tcp_be_liberal(ct);
832 		}
833 
834 		nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
835 	}
836 
837 	return 0;
838 }
839 
840 /* Lookup connection and read fields into key. */
841 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
842 			 const struct ovs_conntrack_info *info,
843 			 struct sk_buff *skb)
844 {
845 	struct nf_conn *ct;
846 	int err;
847 
848 	err = __ovs_ct_lookup(net, key, info, skb);
849 	if (err)
850 		return err;
851 
852 	ct = (struct nf_conn *)skb_nfct(skb);
853 	if (ct)
854 		nf_ct_deliver_cached_events(ct);
855 
856 	return 0;
857 }
858 
859 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
860 {
861 	size_t i;
862 
863 	for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
864 		if (labels->ct_labels_32[i])
865 			return true;
866 
867 	return false;
868 }
869 
870 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
871 static struct hlist_head *ct_limit_hash_bucket(
872 	const struct ovs_ct_limit_info *info, u16 zone)
873 {
874 	return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
875 }
876 
877 /* Call with ovs_mutex */
878 static void ct_limit_set(const struct ovs_ct_limit_info *info,
879 			 struct ovs_ct_limit *new_ct_limit)
880 {
881 	struct ovs_ct_limit *ct_limit;
882 	struct hlist_head *head;
883 
884 	head = ct_limit_hash_bucket(info, new_ct_limit->zone);
885 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node,
886 				 lockdep_ovsl_is_held()) {
887 		if (ct_limit->zone == new_ct_limit->zone) {
888 			hlist_replace_rcu(&ct_limit->hlist_node,
889 					  &new_ct_limit->hlist_node);
890 			kfree_rcu(ct_limit, rcu);
891 			return;
892 		}
893 	}
894 
895 	hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
896 }
897 
898 /* Call with ovs_mutex */
899 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
900 {
901 	struct ovs_ct_limit *ct_limit;
902 	struct hlist_head *head;
903 	struct hlist_node *n;
904 
905 	head = ct_limit_hash_bucket(info, zone);
906 	hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
907 		if (ct_limit->zone == zone) {
908 			hlist_del_rcu(&ct_limit->hlist_node);
909 			kfree_rcu(ct_limit, rcu);
910 			return;
911 		}
912 	}
913 }
914 
915 /* Call with RCU read lock */
916 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
917 {
918 	struct ovs_ct_limit *ct_limit;
919 	struct hlist_head *head;
920 
921 	head = ct_limit_hash_bucket(info, zone);
922 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
923 		if (ct_limit->zone == zone)
924 			return ct_limit->limit;
925 	}
926 
927 	return info->default_limit;
928 }
929 
930 static int ovs_ct_check_limit(struct net *net,
931 			      const struct sk_buff *skb,
932 			      const struct ovs_conntrack_info *info)
933 {
934 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
935 	const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
936 	u32 per_zone_limit, connections;
937 	u32 conncount_key;
938 
939 	conncount_key = info->zone.id;
940 
941 	per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
942 	if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
943 		return 0;
944 
945 	connections = nf_conncount_count_skb(net, skb, info->family,
946 					     ct_limit_info->data,
947 					     &conncount_key);
948 	if (connections > per_zone_limit)
949 		return -ENOMEM;
950 
951 	return 0;
952 }
953 #endif
954 
955 /* Lookup connection and confirm if unconfirmed. */
956 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
957 			 const struct ovs_conntrack_info *info,
958 			 struct sk_buff *skb)
959 {
960 	enum ip_conntrack_info ctinfo;
961 	struct nf_conn *ct;
962 	int err;
963 
964 	err = __ovs_ct_lookup(net, key, info, skb);
965 	if (err)
966 		return err;
967 
968 	/* The connection could be invalid, in which case this is a no-op.*/
969 	ct = nf_ct_get(skb, &ctinfo);
970 	if (!ct)
971 		return 0;
972 
973 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
974 	if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
975 		if (!nf_ct_is_confirmed(ct)) {
976 			err = ovs_ct_check_limit(net, skb, info);
977 			if (err) {
978 				net_warn_ratelimited("openvswitch: zone: %u "
979 					"exceeds conntrack limit\n",
980 					info->zone.id);
981 				return err;
982 			}
983 		}
984 	}
985 #endif
986 
987 	/* Set the conntrack event mask if given.  NEW and DELETE events have
988 	 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
989 	 * typically would receive many kinds of updates.  Setting the event
990 	 * mask allows those events to be filtered.  The set event mask will
991 	 * remain in effect for the lifetime of the connection unless changed
992 	 * by a further CT action with both the commit flag and the eventmask
993 	 * option. */
994 	if (info->have_eventmask) {
995 		struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
996 
997 		if (cache)
998 			cache->ctmask = info->eventmask;
999 	}
1000 
1001 	/* Apply changes before confirming the connection so that the initial
1002 	 * conntrack NEW netlink event carries the values given in the CT
1003 	 * action.
1004 	 */
1005 	if (info->mark.mask) {
1006 		err = ovs_ct_set_mark(ct, key, info->mark.value,
1007 				      info->mark.mask);
1008 		if (err)
1009 			return err;
1010 	}
1011 	if (!nf_ct_is_confirmed(ct)) {
1012 		err = ovs_ct_init_labels(ct, key, &info->labels.value,
1013 					 &info->labels.mask);
1014 		if (err)
1015 			return err;
1016 
1017 		nf_conn_act_ct_ext_add(skb, ct, ctinfo);
1018 	} else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1019 		   labels_nonzero(&info->labels.mask)) {
1020 		err = ovs_ct_set_labels(ct, key, &info->labels.value,
1021 					&info->labels.mask);
1022 		if (err)
1023 			return err;
1024 	}
1025 	/* This will take care of sending queued events even if the connection
1026 	 * is already confirmed.
1027 	 */
1028 	err = nf_conntrack_confirm(skb);
1029 
1030 	return verdict_to_errno(err);
1031 }
1032 
1033 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1034  * value if 'skb' is freed.
1035  */
1036 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1037 		   struct sw_flow_key *key,
1038 		   const struct ovs_conntrack_info *info)
1039 {
1040 	int nh_ofs;
1041 	int err;
1042 
1043 	/* The conntrack module expects to be working at L3. */
1044 	nh_ofs = skb_network_offset(skb);
1045 	skb_pull_rcsum(skb, nh_ofs);
1046 
1047 	err = nf_ct_skb_network_trim(skb, info->family);
1048 	if (err) {
1049 		kfree_skb(skb);
1050 		return err;
1051 	}
1052 
1053 	if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1054 		err = ovs_ct_handle_fragments(net, key, info->zone.id,
1055 					      info->family, skb);
1056 		if (err)
1057 			return err;
1058 	}
1059 
1060 	if (info->commit)
1061 		err = ovs_ct_commit(net, key, info, skb);
1062 	else
1063 		err = ovs_ct_lookup(net, key, info, skb);
1064 
1065 	/* conntrack core returned NF_STOLEN */
1066 	if (err == -EINPROGRESS)
1067 		return err;
1068 
1069 	skb_push_rcsum(skb, nh_ofs);
1070 	if (err)
1071 		ovs_kfree_skb_reason(skb, OVS_DROP_CONNTRACK);
1072 	return err;
1073 }
1074 
1075 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1076 {
1077 	nf_reset_ct(skb);
1078 	nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1079 
1080 	if (key)
1081 		ovs_ct_fill_key(skb, key, false);
1082 
1083 	return 0;
1084 }
1085 
1086 #if IS_ENABLED(CONFIG_NF_NAT)
1087 static int parse_nat(const struct nlattr *attr,
1088 		     struct ovs_conntrack_info *info, bool log)
1089 {
1090 	struct nlattr *a;
1091 	int rem;
1092 	bool have_ip_max = false;
1093 	bool have_proto_max = false;
1094 	bool ip_vers = (info->family == NFPROTO_IPV6);
1095 
1096 	nla_for_each_nested(a, attr, rem) {
1097 		static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1098 			[OVS_NAT_ATTR_SRC] = {0, 0},
1099 			[OVS_NAT_ATTR_DST] = {0, 0},
1100 			[OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1101 						 sizeof(struct in6_addr)},
1102 			[OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1103 						 sizeof(struct in6_addr)},
1104 			[OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1105 			[OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1106 			[OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1107 			[OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1108 			[OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1109 		};
1110 		int type = nla_type(a);
1111 
1112 		if (type > OVS_NAT_ATTR_MAX) {
1113 			OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1114 				  type, OVS_NAT_ATTR_MAX);
1115 			return -EINVAL;
1116 		}
1117 
1118 		if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1119 			OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1120 				  type, nla_len(a),
1121 				  ovs_nat_attr_lens[type][ip_vers]);
1122 			return -EINVAL;
1123 		}
1124 
1125 		switch (type) {
1126 		case OVS_NAT_ATTR_SRC:
1127 		case OVS_NAT_ATTR_DST:
1128 			if (info->nat) {
1129 				OVS_NLERR(log, "Only one type of NAT may be specified");
1130 				return -ERANGE;
1131 			}
1132 			info->nat |= OVS_CT_NAT;
1133 			info->nat |= ((type == OVS_NAT_ATTR_SRC)
1134 					? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1135 			break;
1136 
1137 		case OVS_NAT_ATTR_IP_MIN:
1138 			nla_memcpy(&info->range.min_addr, a,
1139 				   sizeof(info->range.min_addr));
1140 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1141 			break;
1142 
1143 		case OVS_NAT_ATTR_IP_MAX:
1144 			have_ip_max = true;
1145 			nla_memcpy(&info->range.max_addr, a,
1146 				   sizeof(info->range.max_addr));
1147 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1148 			break;
1149 
1150 		case OVS_NAT_ATTR_PROTO_MIN:
1151 			info->range.min_proto.all = htons(nla_get_u16(a));
1152 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1153 			break;
1154 
1155 		case OVS_NAT_ATTR_PROTO_MAX:
1156 			have_proto_max = true;
1157 			info->range.max_proto.all = htons(nla_get_u16(a));
1158 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1159 			break;
1160 
1161 		case OVS_NAT_ATTR_PERSISTENT:
1162 			info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1163 			break;
1164 
1165 		case OVS_NAT_ATTR_PROTO_HASH:
1166 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1167 			break;
1168 
1169 		case OVS_NAT_ATTR_PROTO_RANDOM:
1170 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1171 			break;
1172 
1173 		default:
1174 			OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1175 			return -EINVAL;
1176 		}
1177 	}
1178 
1179 	if (rem > 0) {
1180 		OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1181 		return -EINVAL;
1182 	}
1183 	if (!info->nat) {
1184 		/* Do not allow flags if no type is given. */
1185 		if (info->range.flags) {
1186 			OVS_NLERR(log,
1187 				  "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1188 				  );
1189 			return -EINVAL;
1190 		}
1191 		info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1192 	} else if (!info->commit) {
1193 		OVS_NLERR(log,
1194 			  "NAT attributes may be specified only when CT COMMIT flag is also specified."
1195 			  );
1196 		return -EINVAL;
1197 	}
1198 	/* Allow missing IP_MAX. */
1199 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1200 		memcpy(&info->range.max_addr, &info->range.min_addr,
1201 		       sizeof(info->range.max_addr));
1202 	}
1203 	/* Allow missing PROTO_MAX. */
1204 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1205 	    !have_proto_max) {
1206 		info->range.max_proto.all = info->range.min_proto.all;
1207 	}
1208 	return 0;
1209 }
1210 #endif
1211 
1212 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1213 	[OVS_CT_ATTR_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1214 	[OVS_CT_ATTR_FORCE_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1215 	[OVS_CT_ATTR_ZONE]	= { .minlen = sizeof(u16),
1216 				    .maxlen = sizeof(u16) },
1217 	[OVS_CT_ATTR_MARK]	= { .minlen = sizeof(struct md_mark),
1218 				    .maxlen = sizeof(struct md_mark) },
1219 	[OVS_CT_ATTR_LABELS]	= { .minlen = sizeof(struct md_labels),
1220 				    .maxlen = sizeof(struct md_labels) },
1221 	[OVS_CT_ATTR_HELPER]	= { .minlen = 1,
1222 				    .maxlen = NF_CT_HELPER_NAME_LEN },
1223 #if IS_ENABLED(CONFIG_NF_NAT)
1224 	/* NAT length is checked when parsing the nested attributes. */
1225 	[OVS_CT_ATTR_NAT]	= { .minlen = 0, .maxlen = INT_MAX },
1226 #endif
1227 	[OVS_CT_ATTR_EVENTMASK]	= { .minlen = sizeof(u32),
1228 				    .maxlen = sizeof(u32) },
1229 	[OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1230 				  .maxlen = CTNL_TIMEOUT_NAME_MAX },
1231 };
1232 
1233 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1234 		    const char **helper, bool log)
1235 {
1236 	struct nlattr *a;
1237 	int rem;
1238 
1239 	nla_for_each_nested(a, attr, rem) {
1240 		int type = nla_type(a);
1241 		int maxlen;
1242 		int minlen;
1243 
1244 		if (type > OVS_CT_ATTR_MAX) {
1245 			OVS_NLERR(log,
1246 				  "Unknown conntrack attr (type=%d, max=%d)",
1247 				  type, OVS_CT_ATTR_MAX);
1248 			return -EINVAL;
1249 		}
1250 
1251 		maxlen = ovs_ct_attr_lens[type].maxlen;
1252 		minlen = ovs_ct_attr_lens[type].minlen;
1253 		if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1254 			OVS_NLERR(log,
1255 				  "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1256 				  type, nla_len(a), maxlen);
1257 			return -EINVAL;
1258 		}
1259 
1260 		switch (type) {
1261 		case OVS_CT_ATTR_FORCE_COMMIT:
1262 			info->force = true;
1263 			fallthrough;
1264 		case OVS_CT_ATTR_COMMIT:
1265 			info->commit = true;
1266 			break;
1267 #ifdef CONFIG_NF_CONNTRACK_ZONES
1268 		case OVS_CT_ATTR_ZONE:
1269 			info->zone.id = nla_get_u16(a);
1270 			break;
1271 #endif
1272 #ifdef CONFIG_NF_CONNTRACK_MARK
1273 		case OVS_CT_ATTR_MARK: {
1274 			struct md_mark *mark = nla_data(a);
1275 
1276 			if (!mark->mask) {
1277 				OVS_NLERR(log, "ct_mark mask cannot be 0");
1278 				return -EINVAL;
1279 			}
1280 			info->mark = *mark;
1281 			break;
1282 		}
1283 #endif
1284 #ifdef CONFIG_NF_CONNTRACK_LABELS
1285 		case OVS_CT_ATTR_LABELS: {
1286 			struct md_labels *labels = nla_data(a);
1287 
1288 			if (!labels_nonzero(&labels->mask)) {
1289 				OVS_NLERR(log, "ct_labels mask cannot be 0");
1290 				return -EINVAL;
1291 			}
1292 			info->labels = *labels;
1293 			break;
1294 		}
1295 #endif
1296 		case OVS_CT_ATTR_HELPER:
1297 			*helper = nla_data(a);
1298 			if (!string_is_terminated(*helper, nla_len(a))) {
1299 				OVS_NLERR(log, "Invalid conntrack helper");
1300 				return -EINVAL;
1301 			}
1302 			break;
1303 #if IS_ENABLED(CONFIG_NF_NAT)
1304 		case OVS_CT_ATTR_NAT: {
1305 			int err = parse_nat(a, info, log);
1306 
1307 			if (err)
1308 				return err;
1309 			break;
1310 		}
1311 #endif
1312 		case OVS_CT_ATTR_EVENTMASK:
1313 			info->have_eventmask = true;
1314 			info->eventmask = nla_get_u32(a);
1315 			break;
1316 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1317 		case OVS_CT_ATTR_TIMEOUT:
1318 			memcpy(info->timeout, nla_data(a), nla_len(a));
1319 			if (!string_is_terminated(info->timeout, nla_len(a))) {
1320 				OVS_NLERR(log, "Invalid conntrack timeout");
1321 				return -EINVAL;
1322 			}
1323 			break;
1324 #endif
1325 
1326 		default:
1327 			OVS_NLERR(log, "Unknown conntrack attr (%d)",
1328 				  type);
1329 			return -EINVAL;
1330 		}
1331 	}
1332 
1333 #ifdef CONFIG_NF_CONNTRACK_MARK
1334 	if (!info->commit && info->mark.mask) {
1335 		OVS_NLERR(log,
1336 			  "Setting conntrack mark requires 'commit' flag.");
1337 		return -EINVAL;
1338 	}
1339 #endif
1340 #ifdef CONFIG_NF_CONNTRACK_LABELS
1341 	if (!info->commit && labels_nonzero(&info->labels.mask)) {
1342 		OVS_NLERR(log,
1343 			  "Setting conntrack labels requires 'commit' flag.");
1344 		return -EINVAL;
1345 	}
1346 #endif
1347 	if (rem > 0) {
1348 		OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1349 		return -EINVAL;
1350 	}
1351 
1352 	return 0;
1353 }
1354 
1355 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1356 {
1357 	if (attr == OVS_KEY_ATTR_CT_STATE)
1358 		return true;
1359 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1360 	    attr == OVS_KEY_ATTR_CT_ZONE)
1361 		return true;
1362 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1363 	    attr == OVS_KEY_ATTR_CT_MARK)
1364 		return true;
1365 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1366 	    attr == OVS_KEY_ATTR_CT_LABELS) {
1367 		struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1368 
1369 		return ovs_net->xt_label;
1370 	}
1371 
1372 	return false;
1373 }
1374 
1375 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1376 		       const struct sw_flow_key *key,
1377 		       struct sw_flow_actions **sfa,  bool log)
1378 {
1379 	struct ovs_conntrack_info ct_info;
1380 	const char *helper = NULL;
1381 	u16 family;
1382 	int err;
1383 
1384 	family = key_to_nfproto(key);
1385 	if (family == NFPROTO_UNSPEC) {
1386 		OVS_NLERR(log, "ct family unspecified");
1387 		return -EINVAL;
1388 	}
1389 
1390 	memset(&ct_info, 0, sizeof(ct_info));
1391 	ct_info.family = family;
1392 
1393 	nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1394 			NF_CT_DEFAULT_ZONE_DIR, 0);
1395 
1396 	err = parse_ct(attr, &ct_info, &helper, log);
1397 	if (err)
1398 		return err;
1399 
1400 	/* Set up template for tracking connections in specific zones. */
1401 	ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1402 	if (!ct_info.ct) {
1403 		OVS_NLERR(log, "Failed to allocate conntrack template");
1404 		return -ENOMEM;
1405 	}
1406 
1407 	if (ct_info.timeout[0]) {
1408 		if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1409 				      ct_info.timeout))
1410 			OVS_NLERR(log,
1411 				  "Failed to associated timeout policy '%s'",
1412 				  ct_info.timeout);
1413 		else
1414 			ct_info.nf_ct_timeout = rcu_dereference(
1415 				nf_ct_timeout_find(ct_info.ct)->timeout);
1416 
1417 	}
1418 
1419 	if (helper) {
1420 		err = nf_ct_add_helper(ct_info.ct, helper, ct_info.family,
1421 				       key->ip.proto, ct_info.nat, &ct_info.helper);
1422 		if (err) {
1423 			OVS_NLERR(log, "Failed to add %s helper %d", helper, err);
1424 			goto err_free_ct;
1425 		}
1426 	}
1427 
1428 	err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1429 				 sizeof(ct_info), log);
1430 	if (err)
1431 		goto err_free_ct;
1432 
1433 	if (ct_info.commit)
1434 		__set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1435 	return 0;
1436 err_free_ct:
1437 	__ovs_ct_free_action(&ct_info);
1438 	return err;
1439 }
1440 
1441 #if IS_ENABLED(CONFIG_NF_NAT)
1442 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1443 			       struct sk_buff *skb)
1444 {
1445 	struct nlattr *start;
1446 
1447 	start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1448 	if (!start)
1449 		return false;
1450 
1451 	if (info->nat & OVS_CT_SRC_NAT) {
1452 		if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1453 			return false;
1454 	} else if (info->nat & OVS_CT_DST_NAT) {
1455 		if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1456 			return false;
1457 	} else {
1458 		goto out;
1459 	}
1460 
1461 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1462 		if (IS_ENABLED(CONFIG_NF_NAT) &&
1463 		    info->family == NFPROTO_IPV4) {
1464 			if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1465 					    info->range.min_addr.ip) ||
1466 			    (info->range.max_addr.ip
1467 			     != info->range.min_addr.ip &&
1468 			     (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1469 					      info->range.max_addr.ip))))
1470 				return false;
1471 		} else if (IS_ENABLED(CONFIG_IPV6) &&
1472 			   info->family == NFPROTO_IPV6) {
1473 			if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1474 					     &info->range.min_addr.in6) ||
1475 			    (memcmp(&info->range.max_addr.in6,
1476 				    &info->range.min_addr.in6,
1477 				    sizeof(info->range.max_addr.in6)) &&
1478 			     (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1479 					       &info->range.max_addr.in6))))
1480 				return false;
1481 		} else {
1482 			return false;
1483 		}
1484 	}
1485 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1486 	    (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1487 			 ntohs(info->range.min_proto.all)) ||
1488 	     (info->range.max_proto.all != info->range.min_proto.all &&
1489 	      nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1490 			  ntohs(info->range.max_proto.all)))))
1491 		return false;
1492 
1493 	if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1494 	    nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1495 		return false;
1496 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1497 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1498 		return false;
1499 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1500 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1501 		return false;
1502 out:
1503 	nla_nest_end(skb, start);
1504 
1505 	return true;
1506 }
1507 #endif
1508 
1509 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1510 			  struct sk_buff *skb)
1511 {
1512 	struct nlattr *start;
1513 
1514 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1515 	if (!start)
1516 		return -EMSGSIZE;
1517 
1518 	if (ct_info->commit && nla_put_flag(skb, ct_info->force
1519 					    ? OVS_CT_ATTR_FORCE_COMMIT
1520 					    : OVS_CT_ATTR_COMMIT))
1521 		return -EMSGSIZE;
1522 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1523 	    nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1524 		return -EMSGSIZE;
1525 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1526 	    nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1527 		    &ct_info->mark))
1528 		return -EMSGSIZE;
1529 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1530 	    labels_nonzero(&ct_info->labels.mask) &&
1531 	    nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1532 		    &ct_info->labels))
1533 		return -EMSGSIZE;
1534 	if (ct_info->helper) {
1535 		if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1536 				   ct_info->helper->name))
1537 			return -EMSGSIZE;
1538 	}
1539 	if (ct_info->have_eventmask &&
1540 	    nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1541 		return -EMSGSIZE;
1542 	if (ct_info->timeout[0]) {
1543 		if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1544 			return -EMSGSIZE;
1545 	}
1546 
1547 #if IS_ENABLED(CONFIG_NF_NAT)
1548 	if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1549 		return -EMSGSIZE;
1550 #endif
1551 	nla_nest_end(skb, start);
1552 
1553 	return 0;
1554 }
1555 
1556 void ovs_ct_free_action(const struct nlattr *a)
1557 {
1558 	struct ovs_conntrack_info *ct_info = nla_data(a);
1559 
1560 	__ovs_ct_free_action(ct_info);
1561 }
1562 
1563 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1564 {
1565 	if (ct_info->helper) {
1566 #if IS_ENABLED(CONFIG_NF_NAT)
1567 		if (ct_info->nat)
1568 			nf_nat_helper_put(ct_info->helper);
1569 #endif
1570 		nf_conntrack_helper_put(ct_info->helper);
1571 	}
1572 	if (ct_info->ct) {
1573 		if (ct_info->timeout[0])
1574 			nf_ct_destroy_timeout(ct_info->ct);
1575 		nf_ct_tmpl_free(ct_info->ct);
1576 	}
1577 }
1578 
1579 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1580 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1581 {
1582 	int i, err;
1583 
1584 	ovs_net->ct_limit_info = kmalloc_obj(*ovs_net->ct_limit_info);
1585 	if (!ovs_net->ct_limit_info)
1586 		return -ENOMEM;
1587 
1588 	ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1589 	ovs_net->ct_limit_info->limits =
1590 		kmalloc_objs(struct hlist_head, CT_LIMIT_HASH_BUCKETS);
1591 	if (!ovs_net->ct_limit_info->limits) {
1592 		kfree(ovs_net->ct_limit_info);
1593 		return -ENOMEM;
1594 	}
1595 
1596 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1597 		INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1598 
1599 	ovs_net->ct_limit_info->data = nf_conncount_init(net, sizeof(u32));
1600 
1601 	if (IS_ERR(ovs_net->ct_limit_info->data)) {
1602 		err = PTR_ERR(ovs_net->ct_limit_info->data);
1603 		kfree(ovs_net->ct_limit_info->limits);
1604 		kfree(ovs_net->ct_limit_info);
1605 		pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1606 		return err;
1607 	}
1608 	return 0;
1609 }
1610 
1611 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1612 {
1613 	const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1614 	int i;
1615 
1616 	nf_conncount_destroy(net, info->data);
1617 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1618 		struct hlist_head *head = &info->limits[i];
1619 		struct ovs_ct_limit *ct_limit;
1620 		struct hlist_node *next;
1621 
1622 		hlist_for_each_entry_safe(ct_limit, next, head, hlist_node)
1623 			kfree_rcu(ct_limit, rcu);
1624 	}
1625 	kfree(info->limits);
1626 	kfree(info);
1627 }
1628 
1629 static struct sk_buff *
1630 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1631 			     struct ovs_header **ovs_reply_header)
1632 {
1633 	struct ovs_header *ovs_header = genl_info_userhdr(info);
1634 	struct sk_buff *skb;
1635 
1636 	skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1637 	if (!skb)
1638 		return ERR_PTR(-ENOMEM);
1639 
1640 	*ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1641 					info->snd_seq,
1642 					&dp_ct_limit_genl_family, 0, cmd);
1643 
1644 	if (!*ovs_reply_header) {
1645 		nlmsg_free(skb);
1646 		return ERR_PTR(-EMSGSIZE);
1647 	}
1648 	(*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1649 
1650 	return skb;
1651 }
1652 
1653 static bool check_zone_id(int zone_id, u16 *pzone)
1654 {
1655 	if (zone_id >= 0 && zone_id <= 65535) {
1656 		*pzone = (u16)zone_id;
1657 		return true;
1658 	}
1659 	return false;
1660 }
1661 
1662 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1663 				       struct ovs_ct_limit_info *info)
1664 {
1665 	struct ovs_zone_limit *zone_limit;
1666 	int rem;
1667 	u16 zone;
1668 
1669 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1670 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1671 
1672 	while (rem >= sizeof(*zone_limit)) {
1673 		if (unlikely(zone_limit->zone_id ==
1674 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1675 			ovs_lock();
1676 			info->default_limit = zone_limit->limit;
1677 			ovs_unlock();
1678 		} else if (unlikely(!check_zone_id(
1679 				zone_limit->zone_id, &zone))) {
1680 			OVS_NLERR(true, "zone id is out of range");
1681 		} else {
1682 			struct ovs_ct_limit *ct_limit;
1683 
1684 			ct_limit = kmalloc_obj(*ct_limit, GFP_KERNEL_ACCOUNT);
1685 			if (!ct_limit)
1686 				return -ENOMEM;
1687 
1688 			ct_limit->zone = zone;
1689 			ct_limit->limit = zone_limit->limit;
1690 
1691 			ovs_lock();
1692 			ct_limit_set(info, ct_limit);
1693 			ovs_unlock();
1694 		}
1695 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1696 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1697 				NLA_ALIGN(sizeof(*zone_limit)));
1698 	}
1699 
1700 	if (rem)
1701 		OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1702 
1703 	return 0;
1704 }
1705 
1706 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1707 				       struct ovs_ct_limit_info *info)
1708 {
1709 	struct ovs_zone_limit *zone_limit;
1710 	int rem;
1711 	u16 zone;
1712 
1713 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1714 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1715 
1716 	while (rem >= sizeof(*zone_limit)) {
1717 		if (unlikely(zone_limit->zone_id ==
1718 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1719 			ovs_lock();
1720 			info->default_limit = OVS_CT_LIMIT_DEFAULT;
1721 			ovs_unlock();
1722 		} else if (unlikely(!check_zone_id(
1723 				zone_limit->zone_id, &zone))) {
1724 			OVS_NLERR(true, "zone id is out of range");
1725 		} else {
1726 			ovs_lock();
1727 			ct_limit_del(info, zone);
1728 			ovs_unlock();
1729 		}
1730 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1731 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1732 				NLA_ALIGN(sizeof(*zone_limit)));
1733 	}
1734 
1735 	if (rem)
1736 		OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
1737 
1738 	return 0;
1739 }
1740 
1741 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
1742 					  struct sk_buff *reply)
1743 {
1744 	struct ovs_zone_limit zone_limit = {
1745 		.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE,
1746 		.limit   = info->default_limit,
1747 	};
1748 
1749 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1750 }
1751 
1752 static int __ovs_ct_limit_get_zone_limit(struct net *net,
1753 					 struct nf_conncount_data *data,
1754 					 u16 zone_id, u32 limit,
1755 					 struct sk_buff *reply)
1756 {
1757 	struct nf_conntrack_zone ct_zone;
1758 	struct ovs_zone_limit zone_limit;
1759 	u32 conncount_key = zone_id;
1760 
1761 	zone_limit.zone_id = zone_id;
1762 	zone_limit.limit = limit;
1763 	nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
1764 
1765 	zone_limit.count = nf_conncount_count_skb(net, NULL, 0, data,
1766 						  &conncount_key);
1767 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1768 }
1769 
1770 static int ovs_ct_limit_get_zone_limit(struct net *net,
1771 				       struct nlattr *nla_zone_limit,
1772 				       struct ovs_ct_limit_info *info,
1773 				       struct sk_buff *reply)
1774 {
1775 	struct ovs_zone_limit *zone_limit;
1776 	int rem, err;
1777 	u32 limit;
1778 	u16 zone;
1779 
1780 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1781 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1782 
1783 	while (rem >= sizeof(*zone_limit)) {
1784 		if (unlikely(zone_limit->zone_id ==
1785 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1786 			err = ovs_ct_limit_get_default_limit(info, reply);
1787 			if (err)
1788 				return err;
1789 		} else if (unlikely(!check_zone_id(zone_limit->zone_id,
1790 							&zone))) {
1791 			OVS_NLERR(true, "zone id is out of range");
1792 		} else {
1793 			rcu_read_lock();
1794 			limit = ct_limit_get(info, zone);
1795 
1796 			err = __ovs_ct_limit_get_zone_limit(
1797 				net, info->data, zone, limit, reply);
1798 			rcu_read_unlock();
1799 			if (err)
1800 				return err;
1801 		}
1802 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1803 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1804 				NLA_ALIGN(sizeof(*zone_limit)));
1805 	}
1806 
1807 	if (rem)
1808 		OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
1809 
1810 	return 0;
1811 }
1812 
1813 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
1814 					   struct ovs_ct_limit_info *info,
1815 					   struct sk_buff *reply)
1816 {
1817 	struct ovs_ct_limit *ct_limit;
1818 	struct hlist_head *head;
1819 	int i, err = 0;
1820 
1821 	err = ovs_ct_limit_get_default_limit(info, reply);
1822 	if (err)
1823 		return err;
1824 
1825 	rcu_read_lock();
1826 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1827 		head = &info->limits[i];
1828 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1829 			err = __ovs_ct_limit_get_zone_limit(net, info->data,
1830 				ct_limit->zone, ct_limit->limit, reply);
1831 			if (err)
1832 				goto exit_err;
1833 		}
1834 	}
1835 
1836 exit_err:
1837 	rcu_read_unlock();
1838 	return err;
1839 }
1840 
1841 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
1842 {
1843 	struct nlattr **a = info->attrs;
1844 	struct sk_buff *reply;
1845 	struct ovs_header *ovs_reply_header;
1846 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1847 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1848 	int err;
1849 
1850 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
1851 					     &ovs_reply_header);
1852 	if (IS_ERR(reply))
1853 		return PTR_ERR(reply);
1854 
1855 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1856 		err = -EINVAL;
1857 		goto exit_err;
1858 	}
1859 
1860 	err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
1861 					  ct_limit_info);
1862 	if (err)
1863 		goto exit_err;
1864 
1865 	static_branch_enable(&ovs_ct_limit_enabled);
1866 
1867 	genlmsg_end(reply, ovs_reply_header);
1868 	return genlmsg_reply(reply, info);
1869 
1870 exit_err:
1871 	nlmsg_free(reply);
1872 	return err;
1873 }
1874 
1875 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
1876 {
1877 	struct nlattr **a = info->attrs;
1878 	struct sk_buff *reply;
1879 	struct ovs_header *ovs_reply_header;
1880 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1881 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1882 	int err;
1883 
1884 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
1885 					     &ovs_reply_header);
1886 	if (IS_ERR(reply))
1887 		return PTR_ERR(reply);
1888 
1889 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1890 		err = -EINVAL;
1891 		goto exit_err;
1892 	}
1893 
1894 	err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
1895 					  ct_limit_info);
1896 	if (err)
1897 		goto exit_err;
1898 
1899 	genlmsg_end(reply, ovs_reply_header);
1900 	return genlmsg_reply(reply, info);
1901 
1902 exit_err:
1903 	nlmsg_free(reply);
1904 	return err;
1905 }
1906 
1907 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
1908 {
1909 	struct nlattr **a = info->attrs;
1910 	struct nlattr *nla_reply;
1911 	struct sk_buff *reply;
1912 	struct ovs_header *ovs_reply_header;
1913 	struct net *net = sock_net(skb->sk);
1914 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1915 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1916 	int err;
1917 
1918 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
1919 					     &ovs_reply_header);
1920 	if (IS_ERR(reply))
1921 		return PTR_ERR(reply);
1922 
1923 	nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
1924 	if (!nla_reply) {
1925 		err = -EMSGSIZE;
1926 		goto exit_err;
1927 	}
1928 
1929 	if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1930 		err = ovs_ct_limit_get_zone_limit(
1931 			net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
1932 			reply);
1933 		if (err)
1934 			goto exit_err;
1935 	} else {
1936 		err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
1937 						      reply);
1938 		if (err)
1939 			goto exit_err;
1940 	}
1941 
1942 	nla_nest_end(reply, nla_reply);
1943 	genlmsg_end(reply, ovs_reply_header);
1944 	return genlmsg_reply(reply, info);
1945 
1946 exit_err:
1947 	nlmsg_free(reply);
1948 	return err;
1949 }
1950 
1951 static const struct genl_small_ops ct_limit_genl_ops[] = {
1952 	{ .cmd = OVS_CT_LIMIT_CMD_SET,
1953 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1954 		.flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
1955 					       * privilege.
1956 					       */
1957 		.doit = ovs_ct_limit_cmd_set,
1958 	},
1959 	{ .cmd = OVS_CT_LIMIT_CMD_DEL,
1960 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1961 		.flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
1962 					       * privilege.
1963 					       */
1964 		.doit = ovs_ct_limit_cmd_del,
1965 	},
1966 	{ .cmd = OVS_CT_LIMIT_CMD_GET,
1967 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1968 		.flags = 0,		  /* OK for unprivileged users. */
1969 		.doit = ovs_ct_limit_cmd_get,
1970 	},
1971 };
1972 
1973 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
1974 	.name = OVS_CT_LIMIT_MCGROUP,
1975 };
1976 
1977 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
1978 	.hdrsize = sizeof(struct ovs_header),
1979 	.name = OVS_CT_LIMIT_FAMILY,
1980 	.version = OVS_CT_LIMIT_VERSION,
1981 	.maxattr = OVS_CT_LIMIT_ATTR_MAX,
1982 	.policy = ct_limit_policy,
1983 	.netnsok = true,
1984 	.parallel_ops = true,
1985 	.small_ops = ct_limit_genl_ops,
1986 	.n_small_ops = ARRAY_SIZE(ct_limit_genl_ops),
1987 	.resv_start_op = OVS_CT_LIMIT_CMD_GET + 1,
1988 	.mcgrps = &ovs_ct_limit_multicast_group,
1989 	.n_mcgrps = 1,
1990 	.module = THIS_MODULE,
1991 };
1992 #endif
1993 
1994 int ovs_ct_init(struct net *net)
1995 {
1996 	unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1997 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1998 
1999 	if (nf_connlabels_get(net, n_bits - 1)) {
2000 		ovs_net->xt_label = false;
2001 		OVS_NLERR(true, "Failed to set connlabel length");
2002 	} else {
2003 		ovs_net->xt_label = true;
2004 	}
2005 
2006 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2007 	return ovs_ct_limit_init(net, ovs_net);
2008 #else
2009 	return 0;
2010 #endif
2011 }
2012 
2013 void ovs_ct_exit(struct net *net)
2014 {
2015 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2016 
2017 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2018 	ovs_ct_limit_exit(net, ovs_net);
2019 #endif
2020 
2021 	if (ovs_net->xt_label)
2022 		nf_connlabels_put(net);
2023 }
2024