xref: /openbmc/linux/net/openvswitch/conntrack.c (revision abb84c46)
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 
683 	if (action & BIT(NF_NAT_MANIP_SRC))
684 		ovs_nat_update_key(key, skb, NF_NAT_MANIP_SRC);
685 	if (action & BIT(NF_NAT_MANIP_DST))
686 		ovs_nat_update_key(key, skb, NF_NAT_MANIP_DST);
687 
688 	return err;
689 }
690 #else /* !CONFIG_NF_NAT */
691 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
692 		      const struct ovs_conntrack_info *info,
693 		      struct sk_buff *skb, struct nf_conn *ct,
694 		      enum ip_conntrack_info ctinfo)
695 {
696 	return NF_ACCEPT;
697 }
698 #endif
699 
700 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
701  * not done already.  Update key with new CT state after passing the packet
702  * through conntrack.
703  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
704  * set to NULL and 0 will be returned.
705  */
706 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
707 			   const struct ovs_conntrack_info *info,
708 			   struct sk_buff *skb)
709 {
710 	/* If we are recirculating packets to match on conntrack fields and
711 	 * committing with a separate conntrack action,  then we don't need to
712 	 * actually run the packet through conntrack twice unless it's for a
713 	 * different zone.
714 	 */
715 	bool cached = skb_nfct_cached(net, key, info, skb);
716 	enum ip_conntrack_info ctinfo;
717 	struct nf_conn *ct;
718 
719 	if (!cached) {
720 		struct nf_hook_state state = {
721 			.hook = NF_INET_PRE_ROUTING,
722 			.pf = info->family,
723 			.net = net,
724 		};
725 		struct nf_conn *tmpl = info->ct;
726 		int err;
727 
728 		/* Associate skb with specified zone. */
729 		if (tmpl) {
730 			ct = nf_ct_get(skb, &ctinfo);
731 			nf_ct_put(ct);
732 			nf_conntrack_get(&tmpl->ct_general);
733 			nf_ct_set(skb, tmpl, IP_CT_NEW);
734 		}
735 
736 		err = nf_conntrack_in(skb, &state);
737 		if (err != NF_ACCEPT)
738 			return -ENOENT;
739 
740 		/* Clear CT state NAT flags to mark that we have not yet done
741 		 * NAT after the nf_conntrack_in() call.  We can actually clear
742 		 * the whole state, as it will be re-initialized below.
743 		 */
744 		key->ct_state = 0;
745 
746 		/* Update the key, but keep the NAT flags. */
747 		ovs_ct_update_key(skb, info, key, true, true);
748 	}
749 
750 	ct = nf_ct_get(skb, &ctinfo);
751 	if (ct) {
752 		bool add_helper = false;
753 
754 		/* Packets starting a new connection must be NATted before the
755 		 * helper, so that the helper knows about the NAT.  We enforce
756 		 * this by delaying both NAT and helper calls for unconfirmed
757 		 * connections until the committing CT action.  For later
758 		 * packets NAT and Helper may be called in either order.
759 		 *
760 		 * NAT will be done only if the CT action has NAT, and only
761 		 * once per packet (per zone), as guarded by the NAT bits in
762 		 * the key->ct_state.
763 		 */
764 		if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
765 		    (nf_ct_is_confirmed(ct) || info->commit) &&
766 		    ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
767 			return -EINVAL;
768 		}
769 
770 		/* Userspace may decide to perform a ct lookup without a helper
771 		 * specified followed by a (recirculate and) commit with one,
772 		 * or attach a helper in a later commit.  Therefore, for
773 		 * connections which we will commit, we may need to attach
774 		 * the helper here.
775 		 */
776 		if (!nf_ct_is_confirmed(ct) && info->commit &&
777 		    info->helper && !nfct_help(ct)) {
778 			int err = __nf_ct_try_assign_helper(ct, info->ct,
779 							    GFP_ATOMIC);
780 			if (err)
781 				return err;
782 			add_helper = true;
783 
784 			/* helper installed, add seqadj if NAT is required */
785 			if (info->nat && !nfct_seqadj(ct)) {
786 				if (!nfct_seqadj_ext_add(ct))
787 					return -EINVAL;
788 			}
789 		}
790 
791 		/* Call the helper only if:
792 		 * - nf_conntrack_in() was executed above ("!cached") or a
793 		 *   helper was just attached ("add_helper") for a confirmed
794 		 *   connection, or
795 		 * - When committing an unconfirmed connection.
796 		 */
797 		if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
798 					      info->commit) &&
799 		    nf_ct_helper(skb, ct, ctinfo, info->family) != NF_ACCEPT) {
800 			return -EINVAL;
801 		}
802 
803 		if (nf_ct_protonum(ct) == IPPROTO_TCP &&
804 		    nf_ct_is_confirmed(ct) && nf_conntrack_tcp_established(ct)) {
805 			/* Be liberal for tcp packets so that out-of-window
806 			 * packets are not marked invalid.
807 			 */
808 			nf_ct_set_tcp_be_liberal(ct);
809 		}
810 
811 		nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
812 	}
813 
814 	return 0;
815 }
816 
817 /* Lookup connection and read fields into key. */
818 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
819 			 const struct ovs_conntrack_info *info,
820 			 struct sk_buff *skb)
821 {
822 	struct nf_conn *ct;
823 	int err;
824 
825 	err = __ovs_ct_lookup(net, key, info, skb);
826 	if (err)
827 		return err;
828 
829 	ct = (struct nf_conn *)skb_nfct(skb);
830 	if (ct)
831 		nf_ct_deliver_cached_events(ct);
832 
833 	return 0;
834 }
835 
836 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
837 {
838 	size_t i;
839 
840 	for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
841 		if (labels->ct_labels_32[i])
842 			return true;
843 
844 	return false;
845 }
846 
847 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
848 static struct hlist_head *ct_limit_hash_bucket(
849 	const struct ovs_ct_limit_info *info, u16 zone)
850 {
851 	return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
852 }
853 
854 /* Call with ovs_mutex */
855 static void ct_limit_set(const struct ovs_ct_limit_info *info,
856 			 struct ovs_ct_limit *new_ct_limit)
857 {
858 	struct ovs_ct_limit *ct_limit;
859 	struct hlist_head *head;
860 
861 	head = ct_limit_hash_bucket(info, new_ct_limit->zone);
862 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
863 		if (ct_limit->zone == new_ct_limit->zone) {
864 			hlist_replace_rcu(&ct_limit->hlist_node,
865 					  &new_ct_limit->hlist_node);
866 			kfree_rcu(ct_limit, rcu);
867 			return;
868 		}
869 	}
870 
871 	hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
872 }
873 
874 /* Call with ovs_mutex */
875 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
876 {
877 	struct ovs_ct_limit *ct_limit;
878 	struct hlist_head *head;
879 	struct hlist_node *n;
880 
881 	head = ct_limit_hash_bucket(info, zone);
882 	hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
883 		if (ct_limit->zone == zone) {
884 			hlist_del_rcu(&ct_limit->hlist_node);
885 			kfree_rcu(ct_limit, rcu);
886 			return;
887 		}
888 	}
889 }
890 
891 /* Call with RCU read lock */
892 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
893 {
894 	struct ovs_ct_limit *ct_limit;
895 	struct hlist_head *head;
896 
897 	head = ct_limit_hash_bucket(info, zone);
898 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
899 		if (ct_limit->zone == zone)
900 			return ct_limit->limit;
901 	}
902 
903 	return info->default_limit;
904 }
905 
906 static int ovs_ct_check_limit(struct net *net,
907 			      const struct ovs_conntrack_info *info,
908 			      const struct nf_conntrack_tuple *tuple)
909 {
910 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
911 	const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
912 	u32 per_zone_limit, connections;
913 	u32 conncount_key;
914 
915 	conncount_key = info->zone.id;
916 
917 	per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
918 	if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
919 		return 0;
920 
921 	connections = nf_conncount_count(net, ct_limit_info->data,
922 					 &conncount_key, tuple, &info->zone);
923 	if (connections > per_zone_limit)
924 		return -ENOMEM;
925 
926 	return 0;
927 }
928 #endif
929 
930 /* Lookup connection and confirm if unconfirmed. */
931 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
932 			 const struct ovs_conntrack_info *info,
933 			 struct sk_buff *skb)
934 {
935 	enum ip_conntrack_info ctinfo;
936 	struct nf_conn *ct;
937 	int err;
938 
939 	err = __ovs_ct_lookup(net, key, info, skb);
940 	if (err)
941 		return err;
942 
943 	/* The connection could be invalid, in which case this is a no-op.*/
944 	ct = nf_ct_get(skb, &ctinfo);
945 	if (!ct)
946 		return 0;
947 
948 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
949 	if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
950 		if (!nf_ct_is_confirmed(ct)) {
951 			err = ovs_ct_check_limit(net, info,
952 				&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
953 			if (err) {
954 				net_warn_ratelimited("openvswitch: zone: %u "
955 					"exceeds conntrack limit\n",
956 					info->zone.id);
957 				return err;
958 			}
959 		}
960 	}
961 #endif
962 
963 	/* Set the conntrack event mask if given.  NEW and DELETE events have
964 	 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
965 	 * typically would receive many kinds of updates.  Setting the event
966 	 * mask allows those events to be filtered.  The set event mask will
967 	 * remain in effect for the lifetime of the connection unless changed
968 	 * by a further CT action with both the commit flag and the eventmask
969 	 * option. */
970 	if (info->have_eventmask) {
971 		struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
972 
973 		if (cache)
974 			cache->ctmask = info->eventmask;
975 	}
976 
977 	/* Apply changes before confirming the connection so that the initial
978 	 * conntrack NEW netlink event carries the values given in the CT
979 	 * action.
980 	 */
981 	if (info->mark.mask) {
982 		err = ovs_ct_set_mark(ct, key, info->mark.value,
983 				      info->mark.mask);
984 		if (err)
985 			return err;
986 	}
987 	if (!nf_ct_is_confirmed(ct)) {
988 		err = ovs_ct_init_labels(ct, key, &info->labels.value,
989 					 &info->labels.mask);
990 		if (err)
991 			return err;
992 
993 		nf_conn_act_ct_ext_add(skb, ct, ctinfo);
994 	} else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
995 		   labels_nonzero(&info->labels.mask)) {
996 		err = ovs_ct_set_labels(ct, key, &info->labels.value,
997 					&info->labels.mask);
998 		if (err)
999 			return err;
1000 	}
1001 	/* This will take care of sending queued events even if the connection
1002 	 * is already confirmed.
1003 	 */
1004 	if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1005 		return -EINVAL;
1006 
1007 	return 0;
1008 }
1009 
1010 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1011  * value if 'skb' is freed.
1012  */
1013 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1014 		   struct sw_flow_key *key,
1015 		   const struct ovs_conntrack_info *info)
1016 {
1017 	int nh_ofs;
1018 	int err;
1019 
1020 	/* The conntrack module expects to be working at L3. */
1021 	nh_ofs = skb_network_offset(skb);
1022 	skb_pull_rcsum(skb, nh_ofs);
1023 
1024 	err = nf_ct_skb_network_trim(skb, info->family);
1025 	if (err) {
1026 		kfree_skb(skb);
1027 		return err;
1028 	}
1029 
1030 	if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1031 		err = ovs_ct_handle_fragments(net, key, info->zone.id,
1032 					      info->family, skb);
1033 		if (err)
1034 			return err;
1035 	}
1036 
1037 	if (info->commit)
1038 		err = ovs_ct_commit(net, key, info, skb);
1039 	else
1040 		err = ovs_ct_lookup(net, key, info, skb);
1041 
1042 	skb_push_rcsum(skb, nh_ofs);
1043 	if (err)
1044 		ovs_kfree_skb_reason(skb, OVS_DROP_CONNTRACK);
1045 	return err;
1046 }
1047 
1048 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1049 {
1050 	enum ip_conntrack_info ctinfo;
1051 	struct nf_conn *ct;
1052 
1053 	ct = nf_ct_get(skb, &ctinfo);
1054 
1055 	nf_ct_put(ct);
1056 	nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1057 
1058 	if (key)
1059 		ovs_ct_fill_key(skb, key, false);
1060 
1061 	return 0;
1062 }
1063 
1064 #if IS_ENABLED(CONFIG_NF_NAT)
1065 static int parse_nat(const struct nlattr *attr,
1066 		     struct ovs_conntrack_info *info, bool log)
1067 {
1068 	struct nlattr *a;
1069 	int rem;
1070 	bool have_ip_max = false;
1071 	bool have_proto_max = false;
1072 	bool ip_vers = (info->family == NFPROTO_IPV6);
1073 
1074 	nla_for_each_nested(a, attr, rem) {
1075 		static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1076 			[OVS_NAT_ATTR_SRC] = {0, 0},
1077 			[OVS_NAT_ATTR_DST] = {0, 0},
1078 			[OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1079 						 sizeof(struct in6_addr)},
1080 			[OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1081 						 sizeof(struct in6_addr)},
1082 			[OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1083 			[OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1084 			[OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1085 			[OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1086 			[OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1087 		};
1088 		int type = nla_type(a);
1089 
1090 		if (type > OVS_NAT_ATTR_MAX) {
1091 			OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1092 				  type, OVS_NAT_ATTR_MAX);
1093 			return -EINVAL;
1094 		}
1095 
1096 		if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1097 			OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1098 				  type, nla_len(a),
1099 				  ovs_nat_attr_lens[type][ip_vers]);
1100 			return -EINVAL;
1101 		}
1102 
1103 		switch (type) {
1104 		case OVS_NAT_ATTR_SRC:
1105 		case OVS_NAT_ATTR_DST:
1106 			if (info->nat) {
1107 				OVS_NLERR(log, "Only one type of NAT may be specified");
1108 				return -ERANGE;
1109 			}
1110 			info->nat |= OVS_CT_NAT;
1111 			info->nat |= ((type == OVS_NAT_ATTR_SRC)
1112 					? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1113 			break;
1114 
1115 		case OVS_NAT_ATTR_IP_MIN:
1116 			nla_memcpy(&info->range.min_addr, a,
1117 				   sizeof(info->range.min_addr));
1118 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1119 			break;
1120 
1121 		case OVS_NAT_ATTR_IP_MAX:
1122 			have_ip_max = true;
1123 			nla_memcpy(&info->range.max_addr, a,
1124 				   sizeof(info->range.max_addr));
1125 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1126 			break;
1127 
1128 		case OVS_NAT_ATTR_PROTO_MIN:
1129 			info->range.min_proto.all = htons(nla_get_u16(a));
1130 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1131 			break;
1132 
1133 		case OVS_NAT_ATTR_PROTO_MAX:
1134 			have_proto_max = true;
1135 			info->range.max_proto.all = htons(nla_get_u16(a));
1136 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1137 			break;
1138 
1139 		case OVS_NAT_ATTR_PERSISTENT:
1140 			info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1141 			break;
1142 
1143 		case OVS_NAT_ATTR_PROTO_HASH:
1144 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1145 			break;
1146 
1147 		case OVS_NAT_ATTR_PROTO_RANDOM:
1148 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1149 			break;
1150 
1151 		default:
1152 			OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1153 			return -EINVAL;
1154 		}
1155 	}
1156 
1157 	if (rem > 0) {
1158 		OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1159 		return -EINVAL;
1160 	}
1161 	if (!info->nat) {
1162 		/* Do not allow flags if no type is given. */
1163 		if (info->range.flags) {
1164 			OVS_NLERR(log,
1165 				  "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1166 				  );
1167 			return -EINVAL;
1168 		}
1169 		info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1170 	} else if (!info->commit) {
1171 		OVS_NLERR(log,
1172 			  "NAT attributes may be specified only when CT COMMIT flag is also specified."
1173 			  );
1174 		return -EINVAL;
1175 	}
1176 	/* Allow missing IP_MAX. */
1177 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1178 		memcpy(&info->range.max_addr, &info->range.min_addr,
1179 		       sizeof(info->range.max_addr));
1180 	}
1181 	/* Allow missing PROTO_MAX. */
1182 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1183 	    !have_proto_max) {
1184 		info->range.max_proto.all = info->range.min_proto.all;
1185 	}
1186 	return 0;
1187 }
1188 #endif
1189 
1190 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1191 	[OVS_CT_ATTR_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1192 	[OVS_CT_ATTR_FORCE_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1193 	[OVS_CT_ATTR_ZONE]	= { .minlen = sizeof(u16),
1194 				    .maxlen = sizeof(u16) },
1195 	[OVS_CT_ATTR_MARK]	= { .minlen = sizeof(struct md_mark),
1196 				    .maxlen = sizeof(struct md_mark) },
1197 	[OVS_CT_ATTR_LABELS]	= { .minlen = sizeof(struct md_labels),
1198 				    .maxlen = sizeof(struct md_labels) },
1199 	[OVS_CT_ATTR_HELPER]	= { .minlen = 1,
1200 				    .maxlen = NF_CT_HELPER_NAME_LEN },
1201 #if IS_ENABLED(CONFIG_NF_NAT)
1202 	/* NAT length is checked when parsing the nested attributes. */
1203 	[OVS_CT_ATTR_NAT]	= { .minlen = 0, .maxlen = INT_MAX },
1204 #endif
1205 	[OVS_CT_ATTR_EVENTMASK]	= { .minlen = sizeof(u32),
1206 				    .maxlen = sizeof(u32) },
1207 	[OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1208 				  .maxlen = CTNL_TIMEOUT_NAME_MAX },
1209 };
1210 
1211 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1212 		    const char **helper, bool log)
1213 {
1214 	struct nlattr *a;
1215 	int rem;
1216 
1217 	nla_for_each_nested(a, attr, rem) {
1218 		int type = nla_type(a);
1219 		int maxlen;
1220 		int minlen;
1221 
1222 		if (type > OVS_CT_ATTR_MAX) {
1223 			OVS_NLERR(log,
1224 				  "Unknown conntrack attr (type=%d, max=%d)",
1225 				  type, OVS_CT_ATTR_MAX);
1226 			return -EINVAL;
1227 		}
1228 
1229 		maxlen = ovs_ct_attr_lens[type].maxlen;
1230 		minlen = ovs_ct_attr_lens[type].minlen;
1231 		if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1232 			OVS_NLERR(log,
1233 				  "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1234 				  type, nla_len(a), maxlen);
1235 			return -EINVAL;
1236 		}
1237 
1238 		switch (type) {
1239 		case OVS_CT_ATTR_FORCE_COMMIT:
1240 			info->force = true;
1241 			fallthrough;
1242 		case OVS_CT_ATTR_COMMIT:
1243 			info->commit = true;
1244 			break;
1245 #ifdef CONFIG_NF_CONNTRACK_ZONES
1246 		case OVS_CT_ATTR_ZONE:
1247 			info->zone.id = nla_get_u16(a);
1248 			break;
1249 #endif
1250 #ifdef CONFIG_NF_CONNTRACK_MARK
1251 		case OVS_CT_ATTR_MARK: {
1252 			struct md_mark *mark = nla_data(a);
1253 
1254 			if (!mark->mask) {
1255 				OVS_NLERR(log, "ct_mark mask cannot be 0");
1256 				return -EINVAL;
1257 			}
1258 			info->mark = *mark;
1259 			break;
1260 		}
1261 #endif
1262 #ifdef CONFIG_NF_CONNTRACK_LABELS
1263 		case OVS_CT_ATTR_LABELS: {
1264 			struct md_labels *labels = nla_data(a);
1265 
1266 			if (!labels_nonzero(&labels->mask)) {
1267 				OVS_NLERR(log, "ct_labels mask cannot be 0");
1268 				return -EINVAL;
1269 			}
1270 			info->labels = *labels;
1271 			break;
1272 		}
1273 #endif
1274 		case OVS_CT_ATTR_HELPER:
1275 			*helper = nla_data(a);
1276 			if (!string_is_terminated(*helper, nla_len(a))) {
1277 				OVS_NLERR(log, "Invalid conntrack helper");
1278 				return -EINVAL;
1279 			}
1280 			break;
1281 #if IS_ENABLED(CONFIG_NF_NAT)
1282 		case OVS_CT_ATTR_NAT: {
1283 			int err = parse_nat(a, info, log);
1284 
1285 			if (err)
1286 				return err;
1287 			break;
1288 		}
1289 #endif
1290 		case OVS_CT_ATTR_EVENTMASK:
1291 			info->have_eventmask = true;
1292 			info->eventmask = nla_get_u32(a);
1293 			break;
1294 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1295 		case OVS_CT_ATTR_TIMEOUT:
1296 			memcpy(info->timeout, nla_data(a), nla_len(a));
1297 			if (!string_is_terminated(info->timeout, nla_len(a))) {
1298 				OVS_NLERR(log, "Invalid conntrack timeout");
1299 				return -EINVAL;
1300 			}
1301 			break;
1302 #endif
1303 
1304 		default:
1305 			OVS_NLERR(log, "Unknown conntrack attr (%d)",
1306 				  type);
1307 			return -EINVAL;
1308 		}
1309 	}
1310 
1311 #ifdef CONFIG_NF_CONNTRACK_MARK
1312 	if (!info->commit && info->mark.mask) {
1313 		OVS_NLERR(log,
1314 			  "Setting conntrack mark requires 'commit' flag.");
1315 		return -EINVAL;
1316 	}
1317 #endif
1318 #ifdef CONFIG_NF_CONNTRACK_LABELS
1319 	if (!info->commit && labels_nonzero(&info->labels.mask)) {
1320 		OVS_NLERR(log,
1321 			  "Setting conntrack labels requires 'commit' flag.");
1322 		return -EINVAL;
1323 	}
1324 #endif
1325 	if (rem > 0) {
1326 		OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1327 		return -EINVAL;
1328 	}
1329 
1330 	return 0;
1331 }
1332 
1333 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1334 {
1335 	if (attr == OVS_KEY_ATTR_CT_STATE)
1336 		return true;
1337 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1338 	    attr == OVS_KEY_ATTR_CT_ZONE)
1339 		return true;
1340 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1341 	    attr == OVS_KEY_ATTR_CT_MARK)
1342 		return true;
1343 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1344 	    attr == OVS_KEY_ATTR_CT_LABELS) {
1345 		struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1346 
1347 		return ovs_net->xt_label;
1348 	}
1349 
1350 	return false;
1351 }
1352 
1353 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1354 		       const struct sw_flow_key *key,
1355 		       struct sw_flow_actions **sfa,  bool log)
1356 {
1357 	struct ovs_conntrack_info ct_info;
1358 	const char *helper = NULL;
1359 	u16 family;
1360 	int err;
1361 
1362 	family = key_to_nfproto(key);
1363 	if (family == NFPROTO_UNSPEC) {
1364 		OVS_NLERR(log, "ct family unspecified");
1365 		return -EINVAL;
1366 	}
1367 
1368 	memset(&ct_info, 0, sizeof(ct_info));
1369 	ct_info.family = family;
1370 
1371 	nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1372 			NF_CT_DEFAULT_ZONE_DIR, 0);
1373 
1374 	err = parse_ct(attr, &ct_info, &helper, log);
1375 	if (err)
1376 		return err;
1377 
1378 	/* Set up template for tracking connections in specific zones. */
1379 	ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1380 	if (!ct_info.ct) {
1381 		OVS_NLERR(log, "Failed to allocate conntrack template");
1382 		return -ENOMEM;
1383 	}
1384 
1385 	if (ct_info.timeout[0]) {
1386 		if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1387 				      ct_info.timeout))
1388 			OVS_NLERR(log,
1389 				  "Failed to associated timeout policy '%s'",
1390 				  ct_info.timeout);
1391 		else
1392 			ct_info.nf_ct_timeout = rcu_dereference(
1393 				nf_ct_timeout_find(ct_info.ct)->timeout);
1394 
1395 	}
1396 
1397 	if (helper) {
1398 		err = nf_ct_add_helper(ct_info.ct, helper, ct_info.family,
1399 				       key->ip.proto, ct_info.nat, &ct_info.helper);
1400 		if (err) {
1401 			OVS_NLERR(log, "Failed to add %s helper %d", helper, err);
1402 			goto err_free_ct;
1403 		}
1404 	}
1405 
1406 	err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1407 				 sizeof(ct_info), log);
1408 	if (err)
1409 		goto err_free_ct;
1410 
1411 	if (ct_info.commit)
1412 		__set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1413 	return 0;
1414 err_free_ct:
1415 	__ovs_ct_free_action(&ct_info);
1416 	return err;
1417 }
1418 
1419 #if IS_ENABLED(CONFIG_NF_NAT)
1420 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1421 			       struct sk_buff *skb)
1422 {
1423 	struct nlattr *start;
1424 
1425 	start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1426 	if (!start)
1427 		return false;
1428 
1429 	if (info->nat & OVS_CT_SRC_NAT) {
1430 		if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1431 			return false;
1432 	} else if (info->nat & OVS_CT_DST_NAT) {
1433 		if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1434 			return false;
1435 	} else {
1436 		goto out;
1437 	}
1438 
1439 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1440 		if (IS_ENABLED(CONFIG_NF_NAT) &&
1441 		    info->family == NFPROTO_IPV4) {
1442 			if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1443 					    info->range.min_addr.ip) ||
1444 			    (info->range.max_addr.ip
1445 			     != info->range.min_addr.ip &&
1446 			     (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1447 					      info->range.max_addr.ip))))
1448 				return false;
1449 		} else if (IS_ENABLED(CONFIG_IPV6) &&
1450 			   info->family == NFPROTO_IPV6) {
1451 			if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1452 					     &info->range.min_addr.in6) ||
1453 			    (memcmp(&info->range.max_addr.in6,
1454 				    &info->range.min_addr.in6,
1455 				    sizeof(info->range.max_addr.in6)) &&
1456 			     (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1457 					       &info->range.max_addr.in6))))
1458 				return false;
1459 		} else {
1460 			return false;
1461 		}
1462 	}
1463 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1464 	    (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1465 			 ntohs(info->range.min_proto.all)) ||
1466 	     (info->range.max_proto.all != info->range.min_proto.all &&
1467 	      nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1468 			  ntohs(info->range.max_proto.all)))))
1469 		return false;
1470 
1471 	if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1472 	    nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1473 		return false;
1474 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1475 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1476 		return false;
1477 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1478 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1479 		return false;
1480 out:
1481 	nla_nest_end(skb, start);
1482 
1483 	return true;
1484 }
1485 #endif
1486 
1487 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1488 			  struct sk_buff *skb)
1489 {
1490 	struct nlattr *start;
1491 
1492 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1493 	if (!start)
1494 		return -EMSGSIZE;
1495 
1496 	if (ct_info->commit && nla_put_flag(skb, ct_info->force
1497 					    ? OVS_CT_ATTR_FORCE_COMMIT
1498 					    : OVS_CT_ATTR_COMMIT))
1499 		return -EMSGSIZE;
1500 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1501 	    nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1502 		return -EMSGSIZE;
1503 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1504 	    nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1505 		    &ct_info->mark))
1506 		return -EMSGSIZE;
1507 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1508 	    labels_nonzero(&ct_info->labels.mask) &&
1509 	    nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1510 		    &ct_info->labels))
1511 		return -EMSGSIZE;
1512 	if (ct_info->helper) {
1513 		if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1514 				   ct_info->helper->name))
1515 			return -EMSGSIZE;
1516 	}
1517 	if (ct_info->have_eventmask &&
1518 	    nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1519 		return -EMSGSIZE;
1520 	if (ct_info->timeout[0]) {
1521 		if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1522 			return -EMSGSIZE;
1523 	}
1524 
1525 #if IS_ENABLED(CONFIG_NF_NAT)
1526 	if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1527 		return -EMSGSIZE;
1528 #endif
1529 	nla_nest_end(skb, start);
1530 
1531 	return 0;
1532 }
1533 
1534 void ovs_ct_free_action(const struct nlattr *a)
1535 {
1536 	struct ovs_conntrack_info *ct_info = nla_data(a);
1537 
1538 	__ovs_ct_free_action(ct_info);
1539 }
1540 
1541 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1542 {
1543 	if (ct_info->helper) {
1544 #if IS_ENABLED(CONFIG_NF_NAT)
1545 		if (ct_info->nat)
1546 			nf_nat_helper_put(ct_info->helper);
1547 #endif
1548 		nf_conntrack_helper_put(ct_info->helper);
1549 	}
1550 	if (ct_info->ct) {
1551 		if (ct_info->timeout[0])
1552 			nf_ct_destroy_timeout(ct_info->ct);
1553 		nf_ct_tmpl_free(ct_info->ct);
1554 	}
1555 }
1556 
1557 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1558 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1559 {
1560 	int i, err;
1561 
1562 	ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1563 					 GFP_KERNEL);
1564 	if (!ovs_net->ct_limit_info)
1565 		return -ENOMEM;
1566 
1567 	ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1568 	ovs_net->ct_limit_info->limits =
1569 		kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1570 			      GFP_KERNEL);
1571 	if (!ovs_net->ct_limit_info->limits) {
1572 		kfree(ovs_net->ct_limit_info);
1573 		return -ENOMEM;
1574 	}
1575 
1576 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1577 		INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1578 
1579 	ovs_net->ct_limit_info->data =
1580 		nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1581 
1582 	if (IS_ERR(ovs_net->ct_limit_info->data)) {
1583 		err = PTR_ERR(ovs_net->ct_limit_info->data);
1584 		kfree(ovs_net->ct_limit_info->limits);
1585 		kfree(ovs_net->ct_limit_info);
1586 		pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1587 		return err;
1588 	}
1589 	return 0;
1590 }
1591 
1592 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1593 {
1594 	const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1595 	int i;
1596 
1597 	nf_conncount_destroy(net, NFPROTO_INET, info->data);
1598 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1599 		struct hlist_head *head = &info->limits[i];
1600 		struct ovs_ct_limit *ct_limit;
1601 		struct hlist_node *next;
1602 
1603 		hlist_for_each_entry_safe(ct_limit, next, head, hlist_node)
1604 			kfree_rcu(ct_limit, rcu);
1605 	}
1606 	kfree(info->limits);
1607 	kfree(info);
1608 }
1609 
1610 static struct sk_buff *
1611 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1612 			     struct ovs_header **ovs_reply_header)
1613 {
1614 	struct ovs_header *ovs_header = genl_info_userhdr(info);
1615 	struct sk_buff *skb;
1616 
1617 	skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1618 	if (!skb)
1619 		return ERR_PTR(-ENOMEM);
1620 
1621 	*ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1622 					info->snd_seq,
1623 					&dp_ct_limit_genl_family, 0, cmd);
1624 
1625 	if (!*ovs_reply_header) {
1626 		nlmsg_free(skb);
1627 		return ERR_PTR(-EMSGSIZE);
1628 	}
1629 	(*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1630 
1631 	return skb;
1632 }
1633 
1634 static bool check_zone_id(int zone_id, u16 *pzone)
1635 {
1636 	if (zone_id >= 0 && zone_id <= 65535) {
1637 		*pzone = (u16)zone_id;
1638 		return true;
1639 	}
1640 	return false;
1641 }
1642 
1643 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1644 				       struct ovs_ct_limit_info *info)
1645 {
1646 	struct ovs_zone_limit *zone_limit;
1647 	int rem;
1648 	u16 zone;
1649 
1650 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1651 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1652 
1653 	while (rem >= sizeof(*zone_limit)) {
1654 		if (unlikely(zone_limit->zone_id ==
1655 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1656 			ovs_lock();
1657 			info->default_limit = zone_limit->limit;
1658 			ovs_unlock();
1659 		} else if (unlikely(!check_zone_id(
1660 				zone_limit->zone_id, &zone))) {
1661 			OVS_NLERR(true, "zone id is out of range");
1662 		} else {
1663 			struct ovs_ct_limit *ct_limit;
1664 
1665 			ct_limit = kmalloc(sizeof(*ct_limit),
1666 					   GFP_KERNEL_ACCOUNT);
1667 			if (!ct_limit)
1668 				return -ENOMEM;
1669 
1670 			ct_limit->zone = zone;
1671 			ct_limit->limit = zone_limit->limit;
1672 
1673 			ovs_lock();
1674 			ct_limit_set(info, ct_limit);
1675 			ovs_unlock();
1676 		}
1677 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1678 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1679 				NLA_ALIGN(sizeof(*zone_limit)));
1680 	}
1681 
1682 	if (rem)
1683 		OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1684 
1685 	return 0;
1686 }
1687 
1688 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1689 				       struct ovs_ct_limit_info *info)
1690 {
1691 	struct ovs_zone_limit *zone_limit;
1692 	int rem;
1693 	u16 zone;
1694 
1695 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1696 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1697 
1698 	while (rem >= sizeof(*zone_limit)) {
1699 		if (unlikely(zone_limit->zone_id ==
1700 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1701 			ovs_lock();
1702 			info->default_limit = OVS_CT_LIMIT_DEFAULT;
1703 			ovs_unlock();
1704 		} else if (unlikely(!check_zone_id(
1705 				zone_limit->zone_id, &zone))) {
1706 			OVS_NLERR(true, "zone id is out of range");
1707 		} else {
1708 			ovs_lock();
1709 			ct_limit_del(info, zone);
1710 			ovs_unlock();
1711 		}
1712 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1713 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1714 				NLA_ALIGN(sizeof(*zone_limit)));
1715 	}
1716 
1717 	if (rem)
1718 		OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
1719 
1720 	return 0;
1721 }
1722 
1723 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
1724 					  struct sk_buff *reply)
1725 {
1726 	struct ovs_zone_limit zone_limit = {
1727 		.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE,
1728 		.limit   = info->default_limit,
1729 	};
1730 
1731 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1732 }
1733 
1734 static int __ovs_ct_limit_get_zone_limit(struct net *net,
1735 					 struct nf_conncount_data *data,
1736 					 u16 zone_id, u32 limit,
1737 					 struct sk_buff *reply)
1738 {
1739 	struct nf_conntrack_zone ct_zone;
1740 	struct ovs_zone_limit zone_limit;
1741 	u32 conncount_key = zone_id;
1742 
1743 	zone_limit.zone_id = zone_id;
1744 	zone_limit.limit = limit;
1745 	nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
1746 
1747 	zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
1748 					      &ct_zone);
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 nlattr *nla_zone_limit,
1754 				       struct ovs_ct_limit_info *info,
1755 				       struct sk_buff *reply)
1756 {
1757 	struct ovs_zone_limit *zone_limit;
1758 	int rem, err;
1759 	u32 limit;
1760 	u16 zone;
1761 
1762 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1763 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1764 
1765 	while (rem >= sizeof(*zone_limit)) {
1766 		if (unlikely(zone_limit->zone_id ==
1767 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1768 			err = ovs_ct_limit_get_default_limit(info, reply);
1769 			if (err)
1770 				return err;
1771 		} else if (unlikely(!check_zone_id(zone_limit->zone_id,
1772 							&zone))) {
1773 			OVS_NLERR(true, "zone id is out of range");
1774 		} else {
1775 			rcu_read_lock();
1776 			limit = ct_limit_get(info, zone);
1777 			rcu_read_unlock();
1778 
1779 			err = __ovs_ct_limit_get_zone_limit(
1780 				net, info->data, zone, limit, reply);
1781 			if (err)
1782 				return err;
1783 		}
1784 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1785 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1786 				NLA_ALIGN(sizeof(*zone_limit)));
1787 	}
1788 
1789 	if (rem)
1790 		OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
1791 
1792 	return 0;
1793 }
1794 
1795 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
1796 					   struct ovs_ct_limit_info *info,
1797 					   struct sk_buff *reply)
1798 {
1799 	struct ovs_ct_limit *ct_limit;
1800 	struct hlist_head *head;
1801 	int i, err = 0;
1802 
1803 	err = ovs_ct_limit_get_default_limit(info, reply);
1804 	if (err)
1805 		return err;
1806 
1807 	rcu_read_lock();
1808 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1809 		head = &info->limits[i];
1810 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1811 			err = __ovs_ct_limit_get_zone_limit(net, info->data,
1812 				ct_limit->zone, ct_limit->limit, reply);
1813 			if (err)
1814 				goto exit_err;
1815 		}
1816 	}
1817 
1818 exit_err:
1819 	rcu_read_unlock();
1820 	return err;
1821 }
1822 
1823 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
1824 {
1825 	struct nlattr **a = info->attrs;
1826 	struct sk_buff *reply;
1827 	struct ovs_header *ovs_reply_header;
1828 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1829 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1830 	int err;
1831 
1832 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
1833 					     &ovs_reply_header);
1834 	if (IS_ERR(reply))
1835 		return PTR_ERR(reply);
1836 
1837 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1838 		err = -EINVAL;
1839 		goto exit_err;
1840 	}
1841 
1842 	err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
1843 					  ct_limit_info);
1844 	if (err)
1845 		goto exit_err;
1846 
1847 	static_branch_enable(&ovs_ct_limit_enabled);
1848 
1849 	genlmsg_end(reply, ovs_reply_header);
1850 	return genlmsg_reply(reply, info);
1851 
1852 exit_err:
1853 	nlmsg_free(reply);
1854 	return err;
1855 }
1856 
1857 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
1858 {
1859 	struct nlattr **a = info->attrs;
1860 	struct sk_buff *reply;
1861 	struct ovs_header *ovs_reply_header;
1862 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1863 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1864 	int err;
1865 
1866 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
1867 					     &ovs_reply_header);
1868 	if (IS_ERR(reply))
1869 		return PTR_ERR(reply);
1870 
1871 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1872 		err = -EINVAL;
1873 		goto exit_err;
1874 	}
1875 
1876 	err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
1877 					  ct_limit_info);
1878 	if (err)
1879 		goto exit_err;
1880 
1881 	genlmsg_end(reply, ovs_reply_header);
1882 	return genlmsg_reply(reply, info);
1883 
1884 exit_err:
1885 	nlmsg_free(reply);
1886 	return err;
1887 }
1888 
1889 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
1890 {
1891 	struct nlattr **a = info->attrs;
1892 	struct nlattr *nla_reply;
1893 	struct sk_buff *reply;
1894 	struct ovs_header *ovs_reply_header;
1895 	struct net *net = sock_net(skb->sk);
1896 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1897 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1898 	int err;
1899 
1900 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
1901 					     &ovs_reply_header);
1902 	if (IS_ERR(reply))
1903 		return PTR_ERR(reply);
1904 
1905 	nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
1906 	if (!nla_reply) {
1907 		err = -EMSGSIZE;
1908 		goto exit_err;
1909 	}
1910 
1911 	if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1912 		err = ovs_ct_limit_get_zone_limit(
1913 			net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
1914 			reply);
1915 		if (err)
1916 			goto exit_err;
1917 	} else {
1918 		err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
1919 						      reply);
1920 		if (err)
1921 			goto exit_err;
1922 	}
1923 
1924 	nla_nest_end(reply, nla_reply);
1925 	genlmsg_end(reply, ovs_reply_header);
1926 	return genlmsg_reply(reply, info);
1927 
1928 exit_err:
1929 	nlmsg_free(reply);
1930 	return err;
1931 }
1932 
1933 static const struct genl_small_ops ct_limit_genl_ops[] = {
1934 	{ .cmd = OVS_CT_LIMIT_CMD_SET,
1935 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1936 		.flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
1937 					       * privilege.
1938 					       */
1939 		.doit = ovs_ct_limit_cmd_set,
1940 	},
1941 	{ .cmd = OVS_CT_LIMIT_CMD_DEL,
1942 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1943 		.flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
1944 					       * privilege.
1945 					       */
1946 		.doit = ovs_ct_limit_cmd_del,
1947 	},
1948 	{ .cmd = OVS_CT_LIMIT_CMD_GET,
1949 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1950 		.flags = 0,		  /* OK for unprivileged users. */
1951 		.doit = ovs_ct_limit_cmd_get,
1952 	},
1953 };
1954 
1955 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
1956 	.name = OVS_CT_LIMIT_MCGROUP,
1957 };
1958 
1959 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
1960 	.hdrsize = sizeof(struct ovs_header),
1961 	.name = OVS_CT_LIMIT_FAMILY,
1962 	.version = OVS_CT_LIMIT_VERSION,
1963 	.maxattr = OVS_CT_LIMIT_ATTR_MAX,
1964 	.policy = ct_limit_policy,
1965 	.netnsok = true,
1966 	.parallel_ops = true,
1967 	.small_ops = ct_limit_genl_ops,
1968 	.n_small_ops = ARRAY_SIZE(ct_limit_genl_ops),
1969 	.resv_start_op = OVS_CT_LIMIT_CMD_GET + 1,
1970 	.mcgrps = &ovs_ct_limit_multicast_group,
1971 	.n_mcgrps = 1,
1972 	.module = THIS_MODULE,
1973 };
1974 #endif
1975 
1976 int ovs_ct_init(struct net *net)
1977 {
1978 	unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1979 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1980 
1981 	if (nf_connlabels_get(net, n_bits - 1)) {
1982 		ovs_net->xt_label = false;
1983 		OVS_NLERR(true, "Failed to set connlabel length");
1984 	} else {
1985 		ovs_net->xt_label = true;
1986 	}
1987 
1988 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1989 	return ovs_ct_limit_init(net, ovs_net);
1990 #else
1991 	return 0;
1992 #endif
1993 }
1994 
1995 void ovs_ct_exit(struct net *net)
1996 {
1997 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1998 
1999 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2000 	ovs_ct_limit_exit(net, ovs_net);
2001 #endif
2002 
2003 	if (ovs_net->xt_label)
2004 		nf_connlabels_put(net);
2005 }
2006