xref: /linux/net/openvswitch/conntrack.c (revision 0e901ee5c6f9e1a382099cd7dbee1360c80c441c)
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_ct_put(ct);
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 			ct = nf_ct_get(skb, &ctinfo);
749 			nf_ct_put(ct);
750 			nf_conntrack_get(&tmpl->ct_general);
751 			nf_ct_set(skb, tmpl, IP_CT_NEW);
752 		}
753 
754 		err = nf_conntrack_in(skb, &state);
755 		if (err != NF_ACCEPT)
756 			return verdict_to_errno(err);
757 
758 		/* Clear CT state NAT flags to mark that we have not yet done
759 		 * NAT after the nf_conntrack_in() call.  We can actually clear
760 		 * the whole state, as it will be re-initialized below.
761 		 */
762 		key->ct_state = 0;
763 
764 		/* Update the key, but keep the NAT flags. */
765 		ovs_ct_update_key(skb, info, key, true, true);
766 	}
767 
768 	ct = nf_ct_get(skb, &ctinfo);
769 	if (ct) {
770 		bool add_helper = false;
771 
772 		/* Packets starting a new connection must be NATted before the
773 		 * helper, so that the helper knows about the NAT.  We enforce
774 		 * this by delaying both NAT and helper calls for unconfirmed
775 		 * connections until the committing CT action.  For later
776 		 * packets NAT and Helper may be called in either order.
777 		 *
778 		 * NAT will be done only if the CT action has NAT, and only
779 		 * once per packet (per zone), as guarded by the NAT bits in
780 		 * the key->ct_state.
781 		 */
782 		if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
783 		    (nf_ct_is_confirmed(ct) || info->commit)) {
784 			int err = ovs_ct_nat(net, key, info, skb, ct, ctinfo);
785 
786 			err = verdict_to_errno(err);
787 			if (err)
788 				return err;
789 		}
790 
791 		/* Userspace may decide to perform a ct lookup without a helper
792 		 * specified followed by a (recirculate and) commit with one,
793 		 * or attach a helper in a later commit.  Therefore, for
794 		 * connections which we will commit, we may need to attach
795 		 * the helper here.
796 		 */
797 		if (!nf_ct_is_confirmed(ct) && info->commit &&
798 		    info->helper && !nfct_help(ct)) {
799 			int err = __nf_ct_try_assign_helper(ct, info->ct,
800 							    GFP_ATOMIC);
801 			if (err)
802 				return err;
803 			add_helper = true;
804 
805 			/* helper installed, add seqadj if NAT is required */
806 			if (info->nat && !nfct_seqadj(ct)) {
807 				if (!nfct_seqadj_ext_add(ct))
808 					return -EINVAL;
809 			}
810 		}
811 
812 		/* Call the helper only if:
813 		 * - nf_conntrack_in() was executed above ("!cached") or a
814 		 *   helper was just attached ("add_helper") for a confirmed
815 		 *   connection, or
816 		 * - When committing an unconfirmed connection.
817 		 */
818 		if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
819 					      info->commit)) {
820 			int err = nf_ct_helper(skb, ct, ctinfo, info->family);
821 
822 			err = verdict_to_errno(err);
823 			if (err)
824 				return err;
825 		}
826 
827 		if (nf_ct_protonum(ct) == IPPROTO_TCP &&
828 		    nf_ct_is_confirmed(ct) && nf_conntrack_tcp_established(ct)) {
829 			/* Be liberal for tcp packets so that out-of-window
830 			 * packets are not marked invalid.
831 			 */
832 			nf_ct_set_tcp_be_liberal(ct);
833 		}
834 
835 		nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
836 	}
837 
838 	return 0;
839 }
840 
841 /* Lookup connection and read fields into key. */
842 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
843 			 const struct ovs_conntrack_info *info,
844 			 struct sk_buff *skb)
845 {
846 	struct nf_conn *ct;
847 	int err;
848 
849 	err = __ovs_ct_lookup(net, key, info, skb);
850 	if (err)
851 		return err;
852 
853 	ct = (struct nf_conn *)skb_nfct(skb);
854 	if (ct)
855 		nf_ct_deliver_cached_events(ct);
856 
857 	return 0;
858 }
859 
860 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
861 {
862 	size_t i;
863 
864 	for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
865 		if (labels->ct_labels_32[i])
866 			return true;
867 
868 	return false;
869 }
870 
871 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
872 static struct hlist_head *ct_limit_hash_bucket(
873 	const struct ovs_ct_limit_info *info, u16 zone)
874 {
875 	return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
876 }
877 
878 /* Call with ovs_mutex */
879 static void ct_limit_set(const struct ovs_ct_limit_info *info,
880 			 struct ovs_ct_limit *new_ct_limit)
881 {
882 	struct ovs_ct_limit *ct_limit;
883 	struct hlist_head *head;
884 
885 	head = ct_limit_hash_bucket(info, new_ct_limit->zone);
886 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node,
887 				 lockdep_ovsl_is_held()) {
888 		if (ct_limit->zone == new_ct_limit->zone) {
889 			hlist_replace_rcu(&ct_limit->hlist_node,
890 					  &new_ct_limit->hlist_node);
891 			kfree_rcu(ct_limit, rcu);
892 			return;
893 		}
894 	}
895 
896 	hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
897 }
898 
899 /* Call with ovs_mutex */
900 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
901 {
902 	struct ovs_ct_limit *ct_limit;
903 	struct hlist_head *head;
904 	struct hlist_node *n;
905 
906 	head = ct_limit_hash_bucket(info, zone);
907 	hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
908 		if (ct_limit->zone == zone) {
909 			hlist_del_rcu(&ct_limit->hlist_node);
910 			kfree_rcu(ct_limit, rcu);
911 			return;
912 		}
913 	}
914 }
915 
916 /* Call with RCU read lock */
917 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
918 {
919 	struct ovs_ct_limit *ct_limit;
920 	struct hlist_head *head;
921 
922 	head = ct_limit_hash_bucket(info, zone);
923 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
924 		if (ct_limit->zone == zone)
925 			return ct_limit->limit;
926 	}
927 
928 	return info->default_limit;
929 }
930 
931 static int ovs_ct_check_limit(struct net *net,
932 			      const struct sk_buff *skb,
933 			      const struct ovs_conntrack_info *info)
934 {
935 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
936 	const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
937 	u32 per_zone_limit, connections;
938 	u32 conncount_key;
939 
940 	conncount_key = info->zone.id;
941 
942 	per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
943 	if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
944 		return 0;
945 
946 	connections = nf_conncount_count_skb(net, skb, info->family,
947 					     ct_limit_info->data,
948 					     &conncount_key);
949 	if (connections > per_zone_limit)
950 		return -ENOMEM;
951 
952 	return 0;
953 }
954 #endif
955 
956 /* Lookup connection and confirm if unconfirmed. */
957 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
958 			 const struct ovs_conntrack_info *info,
959 			 struct sk_buff *skb)
960 {
961 	enum ip_conntrack_info ctinfo;
962 	struct nf_conn *ct;
963 	int err;
964 
965 	err = __ovs_ct_lookup(net, key, info, skb);
966 	if (err)
967 		return err;
968 
969 	/* The connection could be invalid, in which case this is a no-op.*/
970 	ct = nf_ct_get(skb, &ctinfo);
971 	if (!ct)
972 		return 0;
973 
974 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
975 	if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
976 		if (!nf_ct_is_confirmed(ct)) {
977 			err = ovs_ct_check_limit(net, skb, info);
978 			if (err) {
979 				net_warn_ratelimited("openvswitch: zone: %u "
980 					"exceeds conntrack limit\n",
981 					info->zone.id);
982 				return err;
983 			}
984 		}
985 	}
986 #endif
987 
988 	/* Set the conntrack event mask if given.  NEW and DELETE events have
989 	 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
990 	 * typically would receive many kinds of updates.  Setting the event
991 	 * mask allows those events to be filtered.  The set event mask will
992 	 * remain in effect for the lifetime of the connection unless changed
993 	 * by a further CT action with both the commit flag and the eventmask
994 	 * option. */
995 	if (info->have_eventmask) {
996 		struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
997 
998 		if (cache)
999 			cache->ctmask = info->eventmask;
1000 	}
1001 
1002 	/* Apply changes before confirming the connection so that the initial
1003 	 * conntrack NEW netlink event carries the values given in the CT
1004 	 * action.
1005 	 */
1006 	if (info->mark.mask) {
1007 		err = ovs_ct_set_mark(ct, key, info->mark.value,
1008 				      info->mark.mask);
1009 		if (err)
1010 			return err;
1011 	}
1012 	if (!nf_ct_is_confirmed(ct)) {
1013 		err = ovs_ct_init_labels(ct, key, &info->labels.value,
1014 					 &info->labels.mask);
1015 		if (err)
1016 			return err;
1017 
1018 		nf_conn_act_ct_ext_add(skb, ct, ctinfo);
1019 	} else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1020 		   labels_nonzero(&info->labels.mask)) {
1021 		err = ovs_ct_set_labels(ct, key, &info->labels.value,
1022 					&info->labels.mask);
1023 		if (err)
1024 			return err;
1025 	}
1026 	/* This will take care of sending queued events even if the connection
1027 	 * is already confirmed.
1028 	 */
1029 	err = nf_conntrack_confirm(skb);
1030 
1031 	return verdict_to_errno(err);
1032 }
1033 
1034 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1035  * value if 'skb' is freed.
1036  */
1037 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1038 		   struct sw_flow_key *key,
1039 		   const struct ovs_conntrack_info *info)
1040 {
1041 	int nh_ofs;
1042 	int err;
1043 
1044 	/* The conntrack module expects to be working at L3. */
1045 	nh_ofs = skb_network_offset(skb);
1046 	skb_pull_rcsum(skb, nh_ofs);
1047 
1048 	err = nf_ct_skb_network_trim(skb, info->family);
1049 	if (err) {
1050 		kfree_skb(skb);
1051 		return err;
1052 	}
1053 
1054 	if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1055 		err = ovs_ct_handle_fragments(net, key, info->zone.id,
1056 					      info->family, skb);
1057 		if (err)
1058 			return err;
1059 	}
1060 
1061 	if (info->commit)
1062 		err = ovs_ct_commit(net, key, info, skb);
1063 	else
1064 		err = ovs_ct_lookup(net, key, info, skb);
1065 
1066 	/* conntrack core returned NF_STOLEN */
1067 	if (err == -EINPROGRESS)
1068 		return err;
1069 
1070 	skb_push_rcsum(skb, nh_ofs);
1071 	if (err)
1072 		ovs_kfree_skb_reason(skb, OVS_DROP_CONNTRACK);
1073 	return err;
1074 }
1075 
1076 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1077 {
1078 	enum ip_conntrack_info ctinfo;
1079 	struct nf_conn *ct;
1080 
1081 	ct = nf_ct_get(skb, &ctinfo);
1082 
1083 	nf_ct_put(ct);
1084 	nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1085 
1086 	if (key)
1087 		ovs_ct_fill_key(skb, key, false);
1088 
1089 	return 0;
1090 }
1091 
1092 #if IS_ENABLED(CONFIG_NF_NAT)
1093 static int parse_nat(const struct nlattr *attr,
1094 		     struct ovs_conntrack_info *info, bool log)
1095 {
1096 	struct nlattr *a;
1097 	int rem;
1098 	bool have_ip_max = false;
1099 	bool have_proto_max = false;
1100 	bool ip_vers = (info->family == NFPROTO_IPV6);
1101 
1102 	nla_for_each_nested(a, attr, rem) {
1103 		static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1104 			[OVS_NAT_ATTR_SRC] = {0, 0},
1105 			[OVS_NAT_ATTR_DST] = {0, 0},
1106 			[OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1107 						 sizeof(struct in6_addr)},
1108 			[OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1109 						 sizeof(struct in6_addr)},
1110 			[OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1111 			[OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1112 			[OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1113 			[OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1114 			[OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1115 		};
1116 		int type = nla_type(a);
1117 
1118 		if (type > OVS_NAT_ATTR_MAX) {
1119 			OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1120 				  type, OVS_NAT_ATTR_MAX);
1121 			return -EINVAL;
1122 		}
1123 
1124 		if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1125 			OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1126 				  type, nla_len(a),
1127 				  ovs_nat_attr_lens[type][ip_vers]);
1128 			return -EINVAL;
1129 		}
1130 
1131 		switch (type) {
1132 		case OVS_NAT_ATTR_SRC:
1133 		case OVS_NAT_ATTR_DST:
1134 			if (info->nat) {
1135 				OVS_NLERR(log, "Only one type of NAT may be specified");
1136 				return -ERANGE;
1137 			}
1138 			info->nat |= OVS_CT_NAT;
1139 			info->nat |= ((type == OVS_NAT_ATTR_SRC)
1140 					? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1141 			break;
1142 
1143 		case OVS_NAT_ATTR_IP_MIN:
1144 			nla_memcpy(&info->range.min_addr, a,
1145 				   sizeof(info->range.min_addr));
1146 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1147 			break;
1148 
1149 		case OVS_NAT_ATTR_IP_MAX:
1150 			have_ip_max = true;
1151 			nla_memcpy(&info->range.max_addr, a,
1152 				   sizeof(info->range.max_addr));
1153 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1154 			break;
1155 
1156 		case OVS_NAT_ATTR_PROTO_MIN:
1157 			info->range.min_proto.all = htons(nla_get_u16(a));
1158 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1159 			break;
1160 
1161 		case OVS_NAT_ATTR_PROTO_MAX:
1162 			have_proto_max = true;
1163 			info->range.max_proto.all = htons(nla_get_u16(a));
1164 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1165 			break;
1166 
1167 		case OVS_NAT_ATTR_PERSISTENT:
1168 			info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1169 			break;
1170 
1171 		case OVS_NAT_ATTR_PROTO_HASH:
1172 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1173 			break;
1174 
1175 		case OVS_NAT_ATTR_PROTO_RANDOM:
1176 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1177 			break;
1178 
1179 		default:
1180 			OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1181 			return -EINVAL;
1182 		}
1183 	}
1184 
1185 	if (rem > 0) {
1186 		OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1187 		return -EINVAL;
1188 	}
1189 	if (!info->nat) {
1190 		/* Do not allow flags if no type is given. */
1191 		if (info->range.flags) {
1192 			OVS_NLERR(log,
1193 				  "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1194 				  );
1195 			return -EINVAL;
1196 		}
1197 		info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1198 	} else if (!info->commit) {
1199 		OVS_NLERR(log,
1200 			  "NAT attributes may be specified only when CT COMMIT flag is also specified."
1201 			  );
1202 		return -EINVAL;
1203 	}
1204 	/* Allow missing IP_MAX. */
1205 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1206 		memcpy(&info->range.max_addr, &info->range.min_addr,
1207 		       sizeof(info->range.max_addr));
1208 	}
1209 	/* Allow missing PROTO_MAX. */
1210 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1211 	    !have_proto_max) {
1212 		info->range.max_proto.all = info->range.min_proto.all;
1213 	}
1214 	return 0;
1215 }
1216 #endif
1217 
1218 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1219 	[OVS_CT_ATTR_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1220 	[OVS_CT_ATTR_FORCE_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1221 	[OVS_CT_ATTR_ZONE]	= { .minlen = sizeof(u16),
1222 				    .maxlen = sizeof(u16) },
1223 	[OVS_CT_ATTR_MARK]	= { .minlen = sizeof(struct md_mark),
1224 				    .maxlen = sizeof(struct md_mark) },
1225 	[OVS_CT_ATTR_LABELS]	= { .minlen = sizeof(struct md_labels),
1226 				    .maxlen = sizeof(struct md_labels) },
1227 	[OVS_CT_ATTR_HELPER]	= { .minlen = 1,
1228 				    .maxlen = NF_CT_HELPER_NAME_LEN },
1229 #if IS_ENABLED(CONFIG_NF_NAT)
1230 	/* NAT length is checked when parsing the nested attributes. */
1231 	[OVS_CT_ATTR_NAT]	= { .minlen = 0, .maxlen = INT_MAX },
1232 #endif
1233 	[OVS_CT_ATTR_EVENTMASK]	= { .minlen = sizeof(u32),
1234 				    .maxlen = sizeof(u32) },
1235 	[OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1236 				  .maxlen = CTNL_TIMEOUT_NAME_MAX },
1237 };
1238 
1239 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1240 		    const char **helper, bool log)
1241 {
1242 	struct nlattr *a;
1243 	int rem;
1244 
1245 	nla_for_each_nested(a, attr, rem) {
1246 		int type = nla_type(a);
1247 		int maxlen;
1248 		int minlen;
1249 
1250 		if (type > OVS_CT_ATTR_MAX) {
1251 			OVS_NLERR(log,
1252 				  "Unknown conntrack attr (type=%d, max=%d)",
1253 				  type, OVS_CT_ATTR_MAX);
1254 			return -EINVAL;
1255 		}
1256 
1257 		maxlen = ovs_ct_attr_lens[type].maxlen;
1258 		minlen = ovs_ct_attr_lens[type].minlen;
1259 		if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1260 			OVS_NLERR(log,
1261 				  "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1262 				  type, nla_len(a), maxlen);
1263 			return -EINVAL;
1264 		}
1265 
1266 		switch (type) {
1267 		case OVS_CT_ATTR_FORCE_COMMIT:
1268 			info->force = true;
1269 			fallthrough;
1270 		case OVS_CT_ATTR_COMMIT:
1271 			info->commit = true;
1272 			break;
1273 #ifdef CONFIG_NF_CONNTRACK_ZONES
1274 		case OVS_CT_ATTR_ZONE:
1275 			info->zone.id = nla_get_u16(a);
1276 			break;
1277 #endif
1278 #ifdef CONFIG_NF_CONNTRACK_MARK
1279 		case OVS_CT_ATTR_MARK: {
1280 			struct md_mark *mark = nla_data(a);
1281 
1282 			if (!mark->mask) {
1283 				OVS_NLERR(log, "ct_mark mask cannot be 0");
1284 				return -EINVAL;
1285 			}
1286 			info->mark = *mark;
1287 			break;
1288 		}
1289 #endif
1290 #ifdef CONFIG_NF_CONNTRACK_LABELS
1291 		case OVS_CT_ATTR_LABELS: {
1292 			struct md_labels *labels = nla_data(a);
1293 
1294 			if (!labels_nonzero(&labels->mask)) {
1295 				OVS_NLERR(log, "ct_labels mask cannot be 0");
1296 				return -EINVAL;
1297 			}
1298 			info->labels = *labels;
1299 			break;
1300 		}
1301 #endif
1302 		case OVS_CT_ATTR_HELPER:
1303 			*helper = nla_data(a);
1304 			if (!string_is_terminated(*helper, nla_len(a))) {
1305 				OVS_NLERR(log, "Invalid conntrack helper");
1306 				return -EINVAL;
1307 			}
1308 			break;
1309 #if IS_ENABLED(CONFIG_NF_NAT)
1310 		case OVS_CT_ATTR_NAT: {
1311 			int err = parse_nat(a, info, log);
1312 
1313 			if (err)
1314 				return err;
1315 			break;
1316 		}
1317 #endif
1318 		case OVS_CT_ATTR_EVENTMASK:
1319 			info->have_eventmask = true;
1320 			info->eventmask = nla_get_u32(a);
1321 			break;
1322 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1323 		case OVS_CT_ATTR_TIMEOUT:
1324 			memcpy(info->timeout, nla_data(a), nla_len(a));
1325 			if (!string_is_terminated(info->timeout, nla_len(a))) {
1326 				OVS_NLERR(log, "Invalid conntrack timeout");
1327 				return -EINVAL;
1328 			}
1329 			break;
1330 #endif
1331 
1332 		default:
1333 			OVS_NLERR(log, "Unknown conntrack attr (%d)",
1334 				  type);
1335 			return -EINVAL;
1336 		}
1337 	}
1338 
1339 #ifdef CONFIG_NF_CONNTRACK_MARK
1340 	if (!info->commit && info->mark.mask) {
1341 		OVS_NLERR(log,
1342 			  "Setting conntrack mark requires 'commit' flag.");
1343 		return -EINVAL;
1344 	}
1345 #endif
1346 #ifdef CONFIG_NF_CONNTRACK_LABELS
1347 	if (!info->commit && labels_nonzero(&info->labels.mask)) {
1348 		OVS_NLERR(log,
1349 			  "Setting conntrack labels requires 'commit' flag.");
1350 		return -EINVAL;
1351 	}
1352 #endif
1353 	if (rem > 0) {
1354 		OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1355 		return -EINVAL;
1356 	}
1357 
1358 	return 0;
1359 }
1360 
1361 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1362 {
1363 	if (attr == OVS_KEY_ATTR_CT_STATE)
1364 		return true;
1365 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1366 	    attr == OVS_KEY_ATTR_CT_ZONE)
1367 		return true;
1368 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1369 	    attr == OVS_KEY_ATTR_CT_MARK)
1370 		return true;
1371 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1372 	    attr == OVS_KEY_ATTR_CT_LABELS) {
1373 		struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1374 
1375 		return ovs_net->xt_label;
1376 	}
1377 
1378 	return false;
1379 }
1380 
1381 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1382 		       const struct sw_flow_key *key,
1383 		       struct sw_flow_actions **sfa,  bool log)
1384 {
1385 	struct ovs_conntrack_info ct_info;
1386 	const char *helper = NULL;
1387 	u16 family;
1388 	int err;
1389 
1390 	family = key_to_nfproto(key);
1391 	if (family == NFPROTO_UNSPEC) {
1392 		OVS_NLERR(log, "ct family unspecified");
1393 		return -EINVAL;
1394 	}
1395 
1396 	memset(&ct_info, 0, sizeof(ct_info));
1397 	ct_info.family = family;
1398 
1399 	nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1400 			NF_CT_DEFAULT_ZONE_DIR, 0);
1401 
1402 	err = parse_ct(attr, &ct_info, &helper, log);
1403 	if (err)
1404 		return err;
1405 
1406 	/* Set up template for tracking connections in specific zones. */
1407 	ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1408 	if (!ct_info.ct) {
1409 		OVS_NLERR(log, "Failed to allocate conntrack template");
1410 		return -ENOMEM;
1411 	}
1412 
1413 	if (ct_info.timeout[0]) {
1414 		if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1415 				      ct_info.timeout))
1416 			OVS_NLERR(log,
1417 				  "Failed to associated timeout policy '%s'",
1418 				  ct_info.timeout);
1419 		else
1420 			ct_info.nf_ct_timeout = rcu_dereference(
1421 				nf_ct_timeout_find(ct_info.ct)->timeout);
1422 
1423 	}
1424 
1425 	if (helper) {
1426 		err = nf_ct_add_helper(ct_info.ct, helper, ct_info.family,
1427 				       key->ip.proto, ct_info.nat, &ct_info.helper);
1428 		if (err) {
1429 			OVS_NLERR(log, "Failed to add %s helper %d", helper, err);
1430 			goto err_free_ct;
1431 		}
1432 	}
1433 
1434 	err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1435 				 sizeof(ct_info), log);
1436 	if (err)
1437 		goto err_free_ct;
1438 
1439 	if (ct_info.commit)
1440 		__set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1441 	return 0;
1442 err_free_ct:
1443 	__ovs_ct_free_action(&ct_info);
1444 	return err;
1445 }
1446 
1447 #if IS_ENABLED(CONFIG_NF_NAT)
1448 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1449 			       struct sk_buff *skb)
1450 {
1451 	struct nlattr *start;
1452 
1453 	start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1454 	if (!start)
1455 		return false;
1456 
1457 	if (info->nat & OVS_CT_SRC_NAT) {
1458 		if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1459 			return false;
1460 	} else if (info->nat & OVS_CT_DST_NAT) {
1461 		if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1462 			return false;
1463 	} else {
1464 		goto out;
1465 	}
1466 
1467 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1468 		if (IS_ENABLED(CONFIG_NF_NAT) &&
1469 		    info->family == NFPROTO_IPV4) {
1470 			if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1471 					    info->range.min_addr.ip) ||
1472 			    (info->range.max_addr.ip
1473 			     != info->range.min_addr.ip &&
1474 			     (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1475 					      info->range.max_addr.ip))))
1476 				return false;
1477 		} else if (IS_ENABLED(CONFIG_IPV6) &&
1478 			   info->family == NFPROTO_IPV6) {
1479 			if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1480 					     &info->range.min_addr.in6) ||
1481 			    (memcmp(&info->range.max_addr.in6,
1482 				    &info->range.min_addr.in6,
1483 				    sizeof(info->range.max_addr.in6)) &&
1484 			     (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1485 					       &info->range.max_addr.in6))))
1486 				return false;
1487 		} else {
1488 			return false;
1489 		}
1490 	}
1491 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1492 	    (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1493 			 ntohs(info->range.min_proto.all)) ||
1494 	     (info->range.max_proto.all != info->range.min_proto.all &&
1495 	      nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1496 			  ntohs(info->range.max_proto.all)))))
1497 		return false;
1498 
1499 	if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1500 	    nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1501 		return false;
1502 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1503 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1504 		return false;
1505 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1506 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1507 		return false;
1508 out:
1509 	nla_nest_end(skb, start);
1510 
1511 	return true;
1512 }
1513 #endif
1514 
1515 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1516 			  struct sk_buff *skb)
1517 {
1518 	struct nlattr *start;
1519 
1520 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1521 	if (!start)
1522 		return -EMSGSIZE;
1523 
1524 	if (ct_info->commit && nla_put_flag(skb, ct_info->force
1525 					    ? OVS_CT_ATTR_FORCE_COMMIT
1526 					    : OVS_CT_ATTR_COMMIT))
1527 		return -EMSGSIZE;
1528 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1529 	    nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1530 		return -EMSGSIZE;
1531 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1532 	    nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1533 		    &ct_info->mark))
1534 		return -EMSGSIZE;
1535 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1536 	    labels_nonzero(&ct_info->labels.mask) &&
1537 	    nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1538 		    &ct_info->labels))
1539 		return -EMSGSIZE;
1540 	if (ct_info->helper) {
1541 		if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1542 				   ct_info->helper->name))
1543 			return -EMSGSIZE;
1544 	}
1545 	if (ct_info->have_eventmask &&
1546 	    nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1547 		return -EMSGSIZE;
1548 	if (ct_info->timeout[0]) {
1549 		if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1550 			return -EMSGSIZE;
1551 	}
1552 
1553 #if IS_ENABLED(CONFIG_NF_NAT)
1554 	if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1555 		return -EMSGSIZE;
1556 #endif
1557 	nla_nest_end(skb, start);
1558 
1559 	return 0;
1560 }
1561 
1562 void ovs_ct_free_action(const struct nlattr *a)
1563 {
1564 	struct ovs_conntrack_info *ct_info = nla_data(a);
1565 
1566 	__ovs_ct_free_action(ct_info);
1567 }
1568 
1569 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1570 {
1571 	if (ct_info->helper) {
1572 #if IS_ENABLED(CONFIG_NF_NAT)
1573 		if (ct_info->nat)
1574 			nf_nat_helper_put(ct_info->helper);
1575 #endif
1576 		nf_conntrack_helper_put(ct_info->helper);
1577 	}
1578 	if (ct_info->ct) {
1579 		if (ct_info->timeout[0])
1580 			nf_ct_destroy_timeout(ct_info->ct);
1581 		nf_ct_tmpl_free(ct_info->ct);
1582 	}
1583 }
1584 
1585 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1586 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1587 {
1588 	int i, err;
1589 
1590 	ovs_net->ct_limit_info = kmalloc_obj(*ovs_net->ct_limit_info);
1591 	if (!ovs_net->ct_limit_info)
1592 		return -ENOMEM;
1593 
1594 	ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1595 	ovs_net->ct_limit_info->limits =
1596 		kmalloc_objs(struct hlist_head, CT_LIMIT_HASH_BUCKETS);
1597 	if (!ovs_net->ct_limit_info->limits) {
1598 		kfree(ovs_net->ct_limit_info);
1599 		return -ENOMEM;
1600 	}
1601 
1602 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1603 		INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1604 
1605 	ovs_net->ct_limit_info->data = nf_conncount_init(net, sizeof(u32));
1606 
1607 	if (IS_ERR(ovs_net->ct_limit_info->data)) {
1608 		err = PTR_ERR(ovs_net->ct_limit_info->data);
1609 		kfree(ovs_net->ct_limit_info->limits);
1610 		kfree(ovs_net->ct_limit_info);
1611 		pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1612 		return err;
1613 	}
1614 	return 0;
1615 }
1616 
1617 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1618 {
1619 	const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1620 	int i;
1621 
1622 	nf_conncount_destroy(net, info->data);
1623 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1624 		struct hlist_head *head = &info->limits[i];
1625 		struct ovs_ct_limit *ct_limit;
1626 		struct hlist_node *next;
1627 
1628 		hlist_for_each_entry_safe(ct_limit, next, head, hlist_node)
1629 			kfree_rcu(ct_limit, rcu);
1630 	}
1631 	kfree(info->limits);
1632 	kfree(info);
1633 }
1634 
1635 static struct sk_buff *
1636 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1637 			     struct ovs_header **ovs_reply_header)
1638 {
1639 	struct ovs_header *ovs_header = genl_info_userhdr(info);
1640 	struct sk_buff *skb;
1641 
1642 	skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1643 	if (!skb)
1644 		return ERR_PTR(-ENOMEM);
1645 
1646 	*ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1647 					info->snd_seq,
1648 					&dp_ct_limit_genl_family, 0, cmd);
1649 
1650 	if (!*ovs_reply_header) {
1651 		nlmsg_free(skb);
1652 		return ERR_PTR(-EMSGSIZE);
1653 	}
1654 	(*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1655 
1656 	return skb;
1657 }
1658 
1659 static bool check_zone_id(int zone_id, u16 *pzone)
1660 {
1661 	if (zone_id >= 0 && zone_id <= 65535) {
1662 		*pzone = (u16)zone_id;
1663 		return true;
1664 	}
1665 	return false;
1666 }
1667 
1668 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1669 				       struct ovs_ct_limit_info *info)
1670 {
1671 	struct ovs_zone_limit *zone_limit;
1672 	int rem;
1673 	u16 zone;
1674 
1675 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1676 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1677 
1678 	while (rem >= sizeof(*zone_limit)) {
1679 		if (unlikely(zone_limit->zone_id ==
1680 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1681 			ovs_lock();
1682 			info->default_limit = zone_limit->limit;
1683 			ovs_unlock();
1684 		} else if (unlikely(!check_zone_id(
1685 				zone_limit->zone_id, &zone))) {
1686 			OVS_NLERR(true, "zone id is out of range");
1687 		} else {
1688 			struct ovs_ct_limit *ct_limit;
1689 
1690 			ct_limit = kmalloc_obj(*ct_limit, GFP_KERNEL_ACCOUNT);
1691 			if (!ct_limit)
1692 				return -ENOMEM;
1693 
1694 			ct_limit->zone = zone;
1695 			ct_limit->limit = zone_limit->limit;
1696 
1697 			ovs_lock();
1698 			ct_limit_set(info, ct_limit);
1699 			ovs_unlock();
1700 		}
1701 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1702 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1703 				NLA_ALIGN(sizeof(*zone_limit)));
1704 	}
1705 
1706 	if (rem)
1707 		OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1708 
1709 	return 0;
1710 }
1711 
1712 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1713 				       struct ovs_ct_limit_info *info)
1714 {
1715 	struct ovs_zone_limit *zone_limit;
1716 	int rem;
1717 	u16 zone;
1718 
1719 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1720 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1721 
1722 	while (rem >= sizeof(*zone_limit)) {
1723 		if (unlikely(zone_limit->zone_id ==
1724 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1725 			ovs_lock();
1726 			info->default_limit = OVS_CT_LIMIT_DEFAULT;
1727 			ovs_unlock();
1728 		} else if (unlikely(!check_zone_id(
1729 				zone_limit->zone_id, &zone))) {
1730 			OVS_NLERR(true, "zone id is out of range");
1731 		} else {
1732 			ovs_lock();
1733 			ct_limit_del(info, zone);
1734 			ovs_unlock();
1735 		}
1736 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1737 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1738 				NLA_ALIGN(sizeof(*zone_limit)));
1739 	}
1740 
1741 	if (rem)
1742 		OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
1743 
1744 	return 0;
1745 }
1746 
1747 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
1748 					  struct sk_buff *reply)
1749 {
1750 	struct ovs_zone_limit zone_limit = {
1751 		.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE,
1752 		.limit   = info->default_limit,
1753 	};
1754 
1755 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1756 }
1757 
1758 static int __ovs_ct_limit_get_zone_limit(struct net *net,
1759 					 struct nf_conncount_data *data,
1760 					 u16 zone_id, u32 limit,
1761 					 struct sk_buff *reply)
1762 {
1763 	struct nf_conntrack_zone ct_zone;
1764 	struct ovs_zone_limit zone_limit;
1765 	u32 conncount_key = zone_id;
1766 
1767 	zone_limit.zone_id = zone_id;
1768 	zone_limit.limit = limit;
1769 	nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
1770 
1771 	zone_limit.count = nf_conncount_count_skb(net, NULL, 0, data,
1772 						  &conncount_key);
1773 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1774 }
1775 
1776 static int ovs_ct_limit_get_zone_limit(struct net *net,
1777 				       struct nlattr *nla_zone_limit,
1778 				       struct ovs_ct_limit_info *info,
1779 				       struct sk_buff *reply)
1780 {
1781 	struct ovs_zone_limit *zone_limit;
1782 	int rem, err;
1783 	u32 limit;
1784 	u16 zone;
1785 
1786 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1787 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1788 
1789 	while (rem >= sizeof(*zone_limit)) {
1790 		if (unlikely(zone_limit->zone_id ==
1791 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1792 			err = ovs_ct_limit_get_default_limit(info, reply);
1793 			if (err)
1794 				return err;
1795 		} else if (unlikely(!check_zone_id(zone_limit->zone_id,
1796 							&zone))) {
1797 			OVS_NLERR(true, "zone id is out of range");
1798 		} else {
1799 			rcu_read_lock();
1800 			limit = ct_limit_get(info, zone);
1801 
1802 			err = __ovs_ct_limit_get_zone_limit(
1803 				net, info->data, zone, limit, reply);
1804 			rcu_read_unlock();
1805 			if (err)
1806 				return err;
1807 		}
1808 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1809 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1810 				NLA_ALIGN(sizeof(*zone_limit)));
1811 	}
1812 
1813 	if (rem)
1814 		OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
1815 
1816 	return 0;
1817 }
1818 
1819 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
1820 					   struct ovs_ct_limit_info *info,
1821 					   struct sk_buff *reply)
1822 {
1823 	struct ovs_ct_limit *ct_limit;
1824 	struct hlist_head *head;
1825 	int i, err = 0;
1826 
1827 	err = ovs_ct_limit_get_default_limit(info, reply);
1828 	if (err)
1829 		return err;
1830 
1831 	rcu_read_lock();
1832 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1833 		head = &info->limits[i];
1834 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1835 			err = __ovs_ct_limit_get_zone_limit(net, info->data,
1836 				ct_limit->zone, ct_limit->limit, reply);
1837 			if (err)
1838 				goto exit_err;
1839 		}
1840 	}
1841 
1842 exit_err:
1843 	rcu_read_unlock();
1844 	return err;
1845 }
1846 
1847 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
1848 {
1849 	struct nlattr **a = info->attrs;
1850 	struct sk_buff *reply;
1851 	struct ovs_header *ovs_reply_header;
1852 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1853 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1854 	int err;
1855 
1856 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
1857 					     &ovs_reply_header);
1858 	if (IS_ERR(reply))
1859 		return PTR_ERR(reply);
1860 
1861 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1862 		err = -EINVAL;
1863 		goto exit_err;
1864 	}
1865 
1866 	err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
1867 					  ct_limit_info);
1868 	if (err)
1869 		goto exit_err;
1870 
1871 	static_branch_enable(&ovs_ct_limit_enabled);
1872 
1873 	genlmsg_end(reply, ovs_reply_header);
1874 	return genlmsg_reply(reply, info);
1875 
1876 exit_err:
1877 	nlmsg_free(reply);
1878 	return err;
1879 }
1880 
1881 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
1882 {
1883 	struct nlattr **a = info->attrs;
1884 	struct sk_buff *reply;
1885 	struct ovs_header *ovs_reply_header;
1886 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1887 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1888 	int err;
1889 
1890 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
1891 					     &ovs_reply_header);
1892 	if (IS_ERR(reply))
1893 		return PTR_ERR(reply);
1894 
1895 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1896 		err = -EINVAL;
1897 		goto exit_err;
1898 	}
1899 
1900 	err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
1901 					  ct_limit_info);
1902 	if (err)
1903 		goto exit_err;
1904 
1905 	genlmsg_end(reply, ovs_reply_header);
1906 	return genlmsg_reply(reply, info);
1907 
1908 exit_err:
1909 	nlmsg_free(reply);
1910 	return err;
1911 }
1912 
1913 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
1914 {
1915 	struct nlattr **a = info->attrs;
1916 	struct nlattr *nla_reply;
1917 	struct sk_buff *reply;
1918 	struct ovs_header *ovs_reply_header;
1919 	struct net *net = sock_net(skb->sk);
1920 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1921 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1922 	int err;
1923 
1924 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
1925 					     &ovs_reply_header);
1926 	if (IS_ERR(reply))
1927 		return PTR_ERR(reply);
1928 
1929 	nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
1930 	if (!nla_reply) {
1931 		err = -EMSGSIZE;
1932 		goto exit_err;
1933 	}
1934 
1935 	if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1936 		err = ovs_ct_limit_get_zone_limit(
1937 			net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
1938 			reply);
1939 		if (err)
1940 			goto exit_err;
1941 	} else {
1942 		err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
1943 						      reply);
1944 		if (err)
1945 			goto exit_err;
1946 	}
1947 
1948 	nla_nest_end(reply, nla_reply);
1949 	genlmsg_end(reply, ovs_reply_header);
1950 	return genlmsg_reply(reply, info);
1951 
1952 exit_err:
1953 	nlmsg_free(reply);
1954 	return err;
1955 }
1956 
1957 static const struct genl_small_ops ct_limit_genl_ops[] = {
1958 	{ .cmd = OVS_CT_LIMIT_CMD_SET,
1959 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1960 		.flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
1961 					       * privilege.
1962 					       */
1963 		.doit = ovs_ct_limit_cmd_set,
1964 	},
1965 	{ .cmd = OVS_CT_LIMIT_CMD_DEL,
1966 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1967 		.flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
1968 					       * privilege.
1969 					       */
1970 		.doit = ovs_ct_limit_cmd_del,
1971 	},
1972 	{ .cmd = OVS_CT_LIMIT_CMD_GET,
1973 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1974 		.flags = 0,		  /* OK for unprivileged users. */
1975 		.doit = ovs_ct_limit_cmd_get,
1976 	},
1977 };
1978 
1979 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
1980 	.name = OVS_CT_LIMIT_MCGROUP,
1981 };
1982 
1983 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
1984 	.hdrsize = sizeof(struct ovs_header),
1985 	.name = OVS_CT_LIMIT_FAMILY,
1986 	.version = OVS_CT_LIMIT_VERSION,
1987 	.maxattr = OVS_CT_LIMIT_ATTR_MAX,
1988 	.policy = ct_limit_policy,
1989 	.netnsok = true,
1990 	.parallel_ops = true,
1991 	.small_ops = ct_limit_genl_ops,
1992 	.n_small_ops = ARRAY_SIZE(ct_limit_genl_ops),
1993 	.resv_start_op = OVS_CT_LIMIT_CMD_GET + 1,
1994 	.mcgrps = &ovs_ct_limit_multicast_group,
1995 	.n_mcgrps = 1,
1996 	.module = THIS_MODULE,
1997 };
1998 #endif
1999 
2000 int ovs_ct_init(struct net *net)
2001 {
2002 	unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2003 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2004 
2005 	if (nf_connlabels_get(net, n_bits - 1)) {
2006 		ovs_net->xt_label = false;
2007 		OVS_NLERR(true, "Failed to set connlabel length");
2008 	} else {
2009 		ovs_net->xt_label = true;
2010 	}
2011 
2012 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2013 	return ovs_ct_limit_init(net, ovs_net);
2014 #else
2015 	return 0;
2016 #endif
2017 }
2018 
2019 void ovs_ct_exit(struct net *net)
2020 {
2021 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2022 
2023 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2024 	ovs_ct_limit_exit(net, ovs_net);
2025 #endif
2026 
2027 	if (ovs_net->xt_label)
2028 		nf_connlabels_put(net);
2029 }
2030