xref: /openbmc/linux/net/openvswitch/conntrack.c (revision 83146efc)
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 = ct ? nf_ct_labels_find(ct) : NULL;
172 
173 	if (cl)
174 		memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
175 	else
176 		memset(labels, 0, OVS_CT_LABELS_LEN);
177 }
178 
179 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
180 					const struct nf_conntrack_tuple *orig,
181 					u8 icmp_proto)
182 {
183 	key->ct_orig_proto = orig->dst.protonum;
184 	if (orig->dst.protonum == icmp_proto) {
185 		key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
186 		key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
187 	} else {
188 		key->ct.orig_tp.src = orig->src.u.all;
189 		key->ct.orig_tp.dst = orig->dst.u.all;
190 	}
191 }
192 
193 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
194 				const struct nf_conntrack_zone *zone,
195 				const struct nf_conn *ct)
196 {
197 	key->ct_state = state;
198 	key->ct_zone = zone->id;
199 	key->ct.mark = ovs_ct_get_mark(ct);
200 	ovs_ct_get_labels(ct, &key->ct.labels);
201 
202 	if (ct) {
203 		const struct nf_conntrack_tuple *orig;
204 
205 		/* Use the master if we have one. */
206 		if (ct->master)
207 			ct = ct->master;
208 		orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
209 
210 		/* IP version must match with the master connection. */
211 		if (key->eth.type == htons(ETH_P_IP) &&
212 		    nf_ct_l3num(ct) == NFPROTO_IPV4) {
213 			key->ipv4.ct_orig.src = orig->src.u3.ip;
214 			key->ipv4.ct_orig.dst = orig->dst.u3.ip;
215 			__ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
216 			return;
217 		} else if (key->eth.type == htons(ETH_P_IPV6) &&
218 			   !sw_flow_key_is_nd(key) &&
219 			   nf_ct_l3num(ct) == NFPROTO_IPV6) {
220 			key->ipv6.ct_orig.src = orig->src.u3.in6;
221 			key->ipv6.ct_orig.dst = orig->dst.u3.in6;
222 			__ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
223 			return;
224 		}
225 	}
226 	/* Clear 'ct_orig_proto' to mark the non-existence of conntrack
227 	 * original direction key fields.
228 	 */
229 	key->ct_orig_proto = 0;
230 }
231 
232 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
233  * previously sent the packet to conntrack via the ct action.  If
234  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
235  * initialized from the connection status.
236  */
237 static void ovs_ct_update_key(const struct sk_buff *skb,
238 			      const struct ovs_conntrack_info *info,
239 			      struct sw_flow_key *key, bool post_ct,
240 			      bool keep_nat_flags)
241 {
242 	const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
243 	enum ip_conntrack_info ctinfo;
244 	struct nf_conn *ct;
245 	u8 state = 0;
246 
247 	ct = nf_ct_get(skb, &ctinfo);
248 	if (ct) {
249 		state = ovs_ct_get_state(ctinfo);
250 		/* All unconfirmed entries are NEW connections. */
251 		if (!nf_ct_is_confirmed(ct))
252 			state |= OVS_CS_F_NEW;
253 		/* OVS persists the related flag for the duration of the
254 		 * connection.
255 		 */
256 		if (ct->master)
257 			state |= OVS_CS_F_RELATED;
258 		if (keep_nat_flags) {
259 			state |= key->ct_state & OVS_CS_F_NAT_MASK;
260 		} else {
261 			if (ct->status & IPS_SRC_NAT)
262 				state |= OVS_CS_F_SRC_NAT;
263 			if (ct->status & IPS_DST_NAT)
264 				state |= OVS_CS_F_DST_NAT;
265 		}
266 		zone = nf_ct_zone(ct);
267 	} else if (post_ct) {
268 		state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
269 		if (info)
270 			zone = &info->zone;
271 	}
272 	__ovs_ct_update_key(key, state, zone, ct);
273 }
274 
275 /* This is called to initialize CT key fields possibly coming in from the local
276  * stack.
277  */
278 void ovs_ct_fill_key(const struct sk_buff *skb,
279 		     struct sw_flow_key *key,
280 		     bool post_ct)
281 {
282 	ovs_ct_update_key(skb, NULL, key, post_ct, false);
283 }
284 
285 int ovs_ct_put_key(const struct sw_flow_key *swkey,
286 		   const struct sw_flow_key *output, struct sk_buff *skb)
287 {
288 	if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
289 		return -EMSGSIZE;
290 
291 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
292 	    nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
293 		return -EMSGSIZE;
294 
295 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
296 	    nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
297 		return -EMSGSIZE;
298 
299 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
300 	    nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
301 		    &output->ct.labels))
302 		return -EMSGSIZE;
303 
304 	if (swkey->ct_orig_proto) {
305 		if (swkey->eth.type == htons(ETH_P_IP)) {
306 			struct ovs_key_ct_tuple_ipv4 orig;
307 
308 			memset(&orig, 0, sizeof(orig));
309 			orig.ipv4_src = output->ipv4.ct_orig.src;
310 			orig.ipv4_dst = output->ipv4.ct_orig.dst;
311 			orig.src_port = output->ct.orig_tp.src;
312 			orig.dst_port = output->ct.orig_tp.dst;
313 			orig.ipv4_proto = output->ct_orig_proto;
314 
315 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
316 				    sizeof(orig), &orig))
317 				return -EMSGSIZE;
318 		} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
319 			struct ovs_key_ct_tuple_ipv6 orig;
320 
321 			memset(&orig, 0, sizeof(orig));
322 			memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32,
323 			       sizeof(orig.ipv6_src));
324 			memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32,
325 			       sizeof(orig.ipv6_dst));
326 			orig.src_port = output->ct.orig_tp.src;
327 			orig.dst_port = output->ct.orig_tp.dst;
328 			orig.ipv6_proto = output->ct_orig_proto;
329 
330 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
331 				    sizeof(orig), &orig))
332 				return -EMSGSIZE;
333 		}
334 	}
335 
336 	return 0;
337 }
338 
339 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
340 			   u32 ct_mark, u32 mask)
341 {
342 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
343 	u32 new_mark;
344 
345 	new_mark = ct_mark | (READ_ONCE(ct->mark) & ~(mask));
346 	if (READ_ONCE(ct->mark) != new_mark) {
347 		WRITE_ONCE(ct->mark, new_mark);
348 		if (nf_ct_is_confirmed(ct))
349 			nf_conntrack_event_cache(IPCT_MARK, ct);
350 		key->ct.mark = new_mark;
351 	}
352 
353 	return 0;
354 #else
355 	return -ENOTSUPP;
356 #endif
357 }
358 
359 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
360 {
361 	struct nf_conn_labels *cl;
362 
363 	cl = nf_ct_labels_find(ct);
364 	if (!cl) {
365 		nf_ct_labels_ext_add(ct);
366 		cl = nf_ct_labels_find(ct);
367 	}
368 
369 	return cl;
370 }
371 
372 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
373  * since the new connection is not yet confirmed, and thus no-one else has
374  * access to it's labels, we simply write them over.
375  */
376 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
377 			      const struct ovs_key_ct_labels *labels,
378 			      const struct ovs_key_ct_labels *mask)
379 {
380 	struct nf_conn_labels *cl, *master_cl;
381 	bool have_mask = labels_nonzero(mask);
382 
383 	/* Inherit master's labels to the related connection? */
384 	master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
385 
386 	if (!master_cl && !have_mask)
387 		return 0;   /* Nothing to do. */
388 
389 	cl = ovs_ct_get_conn_labels(ct);
390 	if (!cl)
391 		return -ENOSPC;
392 
393 	/* Inherit the master's labels, if any. */
394 	if (master_cl)
395 		*cl = *master_cl;
396 
397 	if (have_mask) {
398 		u32 *dst = (u32 *)cl->bits;
399 		int i;
400 
401 		for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
402 			dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
403 				(labels->ct_labels_32[i]
404 				 & mask->ct_labels_32[i]);
405 	}
406 
407 	/* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
408 	 * IPCT_LABEL bit is set in the event cache.
409 	 */
410 	nf_conntrack_event_cache(IPCT_LABEL, ct);
411 
412 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
413 
414 	return 0;
415 }
416 
417 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
418 			     const struct ovs_key_ct_labels *labels,
419 			     const struct ovs_key_ct_labels *mask)
420 {
421 	struct nf_conn_labels *cl;
422 	int err;
423 
424 	cl = ovs_ct_get_conn_labels(ct);
425 	if (!cl)
426 		return -ENOSPC;
427 
428 	err = nf_connlabels_replace(ct, labels->ct_labels_32,
429 				    mask->ct_labels_32,
430 				    OVS_CT_LABELS_LEN_32);
431 	if (err)
432 		return err;
433 
434 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
435 
436 	return 0;
437 }
438 
439 static int ovs_ct_handle_fragments(struct net *net, struct sw_flow_key *key,
440 				   u16 zone, int family, struct sk_buff *skb)
441 {
442 	struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
443 	int err;
444 
445 	err = nf_ct_handle_fragments(net, skb, zone, family, &key->ip.proto, &ovs_cb.mru);
446 	if (err)
447 		return err;
448 
449 	/* The key extracted from the fragment that completed this datagram
450 	 * likely didn't have an L4 header, so regenerate it.
451 	 */
452 	ovs_flow_key_update_l3l4(skb, key);
453 	key->ip.frag = OVS_FRAG_TYPE_NONE;
454 	*OVS_CB(skb) = ovs_cb;
455 
456 	return 0;
457 }
458 
459 /* This replicates logic from nf_conntrack_core.c that is not exported. */
460 static enum ip_conntrack_info
461 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
462 {
463 	const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
464 
465 	if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
466 		return IP_CT_ESTABLISHED_REPLY;
467 	/* Once we've had two way comms, always ESTABLISHED. */
468 	if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
469 		return IP_CT_ESTABLISHED;
470 	if (test_bit(IPS_EXPECTED_BIT, &ct->status))
471 		return IP_CT_RELATED;
472 	return IP_CT_NEW;
473 }
474 
475 /* Find an existing connection which this packet belongs to without
476  * re-attributing statistics or modifying the connection state.  This allows an
477  * skb->_nfct lost due to an upcall to be recovered during actions execution.
478  *
479  * Must be called with rcu_read_lock.
480  *
481  * On success, populates skb->_nfct and returns the connection.  Returns NULL
482  * if there is no existing entry.
483  */
484 static struct nf_conn *
485 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
486 		     u8 l3num, struct sk_buff *skb, bool natted)
487 {
488 	struct nf_conntrack_tuple tuple;
489 	struct nf_conntrack_tuple_hash *h;
490 	struct nf_conn *ct;
491 
492 	if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
493 			       net, &tuple)) {
494 		pr_debug("ovs_ct_find_existing: Can't get tuple\n");
495 		return NULL;
496 	}
497 
498 	/* Must invert the tuple if skb has been transformed by NAT. */
499 	if (natted) {
500 		struct nf_conntrack_tuple inverse;
501 
502 		if (!nf_ct_invert_tuple(&inverse, &tuple)) {
503 			pr_debug("ovs_ct_find_existing: Inversion failed!\n");
504 			return NULL;
505 		}
506 		tuple = inverse;
507 	}
508 
509 	/* look for tuple match */
510 	h = nf_conntrack_find_get(net, zone, &tuple);
511 	if (!h)
512 		return NULL;   /* Not found. */
513 
514 	ct = nf_ct_tuplehash_to_ctrack(h);
515 
516 	/* Inverted packet tuple matches the reverse direction conntrack tuple,
517 	 * select the other tuplehash to get the right 'ctinfo' bits for this
518 	 * packet.
519 	 */
520 	if (natted)
521 		h = &ct->tuplehash[!h->tuple.dst.dir];
522 
523 	nf_ct_set(skb, ct, ovs_ct_get_info(h));
524 	return ct;
525 }
526 
527 static
528 struct nf_conn *ovs_ct_executed(struct net *net,
529 				const struct sw_flow_key *key,
530 				const struct ovs_conntrack_info *info,
531 				struct sk_buff *skb,
532 				bool *ct_executed)
533 {
534 	struct nf_conn *ct = NULL;
535 
536 	/* If no ct, check if we have evidence that an existing conntrack entry
537 	 * might be found for this skb.  This happens when we lose a skb->_nfct
538 	 * due to an upcall, or if the direction is being forced.  If the
539 	 * connection was not confirmed, it is not cached and needs to be run
540 	 * through conntrack again.
541 	 */
542 	*ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
543 		       !(key->ct_state & OVS_CS_F_INVALID) &&
544 		       (key->ct_zone == info->zone.id);
545 
546 	if (*ct_executed || (!key->ct_state && info->force)) {
547 		ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
548 					  !!(key->ct_state &
549 					  OVS_CS_F_NAT_MASK));
550 	}
551 
552 	return ct;
553 }
554 
555 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
556 static bool skb_nfct_cached(struct net *net,
557 			    const struct sw_flow_key *key,
558 			    const struct ovs_conntrack_info *info,
559 			    struct sk_buff *skb)
560 {
561 	enum ip_conntrack_info ctinfo;
562 	struct nf_conn *ct;
563 	bool ct_executed = true;
564 
565 	ct = nf_ct_get(skb, &ctinfo);
566 	if (!ct)
567 		ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
568 
569 	if (ct)
570 		nf_ct_get(skb, &ctinfo);
571 	else
572 		return false;
573 
574 	if (!net_eq(net, read_pnet(&ct->ct_net)))
575 		return false;
576 	if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
577 		return false;
578 	if (info->helper) {
579 		struct nf_conn_help *help;
580 
581 		help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
582 		if (help && rcu_access_pointer(help->helper) != info->helper)
583 			return false;
584 	}
585 	if (info->nf_ct_timeout) {
586 		struct nf_conn_timeout *timeout_ext;
587 
588 		timeout_ext = nf_ct_timeout_find(ct);
589 		if (!timeout_ext || info->nf_ct_timeout !=
590 		    rcu_dereference(timeout_ext->timeout))
591 			return false;
592 	}
593 	/* Force conntrack entry direction to the current packet? */
594 	if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
595 		/* Delete the conntrack entry if confirmed, else just release
596 		 * the reference.
597 		 */
598 		if (nf_ct_is_confirmed(ct))
599 			nf_ct_delete(ct, 0, 0);
600 
601 		nf_ct_put(ct);
602 		nf_ct_set(skb, NULL, 0);
603 		return false;
604 	}
605 
606 	return ct_executed;
607 }
608 
609 #if IS_ENABLED(CONFIG_NF_NAT)
610 static void ovs_nat_update_key(struct sw_flow_key *key,
611 			       const struct sk_buff *skb,
612 			       enum nf_nat_manip_type maniptype)
613 {
614 	if (maniptype == NF_NAT_MANIP_SRC) {
615 		__be16 src;
616 
617 		key->ct_state |= OVS_CS_F_SRC_NAT;
618 		if (key->eth.type == htons(ETH_P_IP))
619 			key->ipv4.addr.src = ip_hdr(skb)->saddr;
620 		else if (key->eth.type == htons(ETH_P_IPV6))
621 			memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
622 			       sizeof(key->ipv6.addr.src));
623 		else
624 			return;
625 
626 		if (key->ip.proto == IPPROTO_UDP)
627 			src = udp_hdr(skb)->source;
628 		else if (key->ip.proto == IPPROTO_TCP)
629 			src = tcp_hdr(skb)->source;
630 		else if (key->ip.proto == IPPROTO_SCTP)
631 			src = sctp_hdr(skb)->source;
632 		else
633 			return;
634 
635 		key->tp.src = src;
636 	} else {
637 		__be16 dst;
638 
639 		key->ct_state |= OVS_CS_F_DST_NAT;
640 		if (key->eth.type == htons(ETH_P_IP))
641 			key->ipv4.addr.dst = ip_hdr(skb)->daddr;
642 		else if (key->eth.type == htons(ETH_P_IPV6))
643 			memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
644 			       sizeof(key->ipv6.addr.dst));
645 		else
646 			return;
647 
648 		if (key->ip.proto == IPPROTO_UDP)
649 			dst = udp_hdr(skb)->dest;
650 		else if (key->ip.proto == IPPROTO_TCP)
651 			dst = tcp_hdr(skb)->dest;
652 		else if (key->ip.proto == IPPROTO_SCTP)
653 			dst = sctp_hdr(skb)->dest;
654 		else
655 			return;
656 
657 		key->tp.dst = dst;
658 	}
659 }
660 
661 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
662 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
663 		      const struct ovs_conntrack_info *info,
664 		      struct sk_buff *skb, struct nf_conn *ct,
665 		      enum ip_conntrack_info ctinfo)
666 {
667 	int err, action = 0;
668 
669 	if (!(info->nat & OVS_CT_NAT))
670 		return NF_ACCEPT;
671 	if (info->nat & OVS_CT_SRC_NAT)
672 		action |= BIT(NF_NAT_MANIP_SRC);
673 	if (info->nat & OVS_CT_DST_NAT)
674 		action |= BIT(NF_NAT_MANIP_DST);
675 
676 	err = nf_ct_nat(skb, ct, ctinfo, &action, &info->range, info->commit);
677 
678 	if (action & BIT(NF_NAT_MANIP_SRC))
679 		ovs_nat_update_key(key, skb, NF_NAT_MANIP_SRC);
680 	if (action & BIT(NF_NAT_MANIP_DST))
681 		ovs_nat_update_key(key, skb, NF_NAT_MANIP_DST);
682 
683 	return err;
684 }
685 #else /* !CONFIG_NF_NAT */
686 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
687 		      const struct ovs_conntrack_info *info,
688 		      struct sk_buff *skb, struct nf_conn *ct,
689 		      enum ip_conntrack_info ctinfo)
690 {
691 	return NF_ACCEPT;
692 }
693 #endif
694 
695 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
696  * not done already.  Update key with new CT state after passing the packet
697  * through conntrack.
698  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
699  * set to NULL and 0 will be returned.
700  */
701 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
702 			   const struct ovs_conntrack_info *info,
703 			   struct sk_buff *skb)
704 {
705 	/* If we are recirculating packets to match on conntrack fields and
706 	 * committing with a separate conntrack action,  then we don't need to
707 	 * actually run the packet through conntrack twice unless it's for a
708 	 * different zone.
709 	 */
710 	bool cached = skb_nfct_cached(net, key, info, skb);
711 	enum ip_conntrack_info ctinfo;
712 	struct nf_conn *ct;
713 
714 	if (!cached) {
715 		struct nf_hook_state state = {
716 			.hook = NF_INET_PRE_ROUTING,
717 			.pf = info->family,
718 			.net = net,
719 		};
720 		struct nf_conn *tmpl = info->ct;
721 		int err;
722 
723 		/* Associate skb with specified zone. */
724 		if (tmpl) {
725 			ct = nf_ct_get(skb, &ctinfo);
726 			nf_ct_put(ct);
727 			nf_conntrack_get(&tmpl->ct_general);
728 			nf_ct_set(skb, tmpl, IP_CT_NEW);
729 		}
730 
731 		err = nf_conntrack_in(skb, &state);
732 		if (err != NF_ACCEPT)
733 			return -ENOENT;
734 
735 		/* Clear CT state NAT flags to mark that we have not yet done
736 		 * NAT after the nf_conntrack_in() call.  We can actually clear
737 		 * the whole state, as it will be re-initialized below.
738 		 */
739 		key->ct_state = 0;
740 
741 		/* Update the key, but keep the NAT flags. */
742 		ovs_ct_update_key(skb, info, key, true, true);
743 	}
744 
745 	ct = nf_ct_get(skb, &ctinfo);
746 	if (ct) {
747 		bool add_helper = false;
748 
749 		/* Packets starting a new connection must be NATted before the
750 		 * helper, so that the helper knows about the NAT.  We enforce
751 		 * this by delaying both NAT and helper calls for unconfirmed
752 		 * connections until the committing CT action.  For later
753 		 * packets NAT and Helper may be called in either order.
754 		 *
755 		 * NAT will be done only if the CT action has NAT, and only
756 		 * once per packet (per zone), as guarded by the NAT bits in
757 		 * the key->ct_state.
758 		 */
759 		if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
760 		    (nf_ct_is_confirmed(ct) || info->commit) &&
761 		    ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
762 			return -EINVAL;
763 		}
764 
765 		/* Userspace may decide to perform a ct lookup without a helper
766 		 * specified followed by a (recirculate and) commit with one,
767 		 * or attach a helper in a later commit.  Therefore, for
768 		 * connections which we will commit, we may need to attach
769 		 * the helper here.
770 		 */
771 		if (!nf_ct_is_confirmed(ct) && info->commit &&
772 		    info->helper && !nfct_help(ct)) {
773 			int err = __nf_ct_try_assign_helper(ct, info->ct,
774 							    GFP_ATOMIC);
775 			if (err)
776 				return err;
777 			add_helper = true;
778 
779 			/* helper installed, add seqadj if NAT is required */
780 			if (info->nat && !nfct_seqadj(ct)) {
781 				if (!nfct_seqadj_ext_add(ct))
782 					return -EINVAL;
783 			}
784 		}
785 
786 		/* Call the helper only if:
787 		 * - nf_conntrack_in() was executed above ("!cached") or a
788 		 *   helper was just attached ("add_helper") for a confirmed
789 		 *   connection, or
790 		 * - When committing an unconfirmed connection.
791 		 */
792 		if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
793 					      info->commit) &&
794 		    nf_ct_helper(skb, ct, ctinfo, info->family) != NF_ACCEPT) {
795 			return -EINVAL;
796 		}
797 
798 		if (nf_ct_protonum(ct) == IPPROTO_TCP &&
799 		    nf_ct_is_confirmed(ct) && nf_conntrack_tcp_established(ct)) {
800 			/* Be liberal for tcp packets so that out-of-window
801 			 * packets are not marked invalid.
802 			 */
803 			nf_ct_set_tcp_be_liberal(ct);
804 		}
805 
806 		nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
807 	}
808 
809 	return 0;
810 }
811 
812 /* Lookup connection and read fields into key. */
813 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
814 			 const struct ovs_conntrack_info *info,
815 			 struct sk_buff *skb)
816 {
817 	struct nf_conn *ct;
818 	int err;
819 
820 	err = __ovs_ct_lookup(net, key, info, skb);
821 	if (err)
822 		return err;
823 
824 	ct = (struct nf_conn *)skb_nfct(skb);
825 	if (ct)
826 		nf_ct_deliver_cached_events(ct);
827 
828 	return 0;
829 }
830 
831 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
832 {
833 	size_t i;
834 
835 	for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
836 		if (labels->ct_labels_32[i])
837 			return true;
838 
839 	return false;
840 }
841 
842 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
843 static struct hlist_head *ct_limit_hash_bucket(
844 	const struct ovs_ct_limit_info *info, u16 zone)
845 {
846 	return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
847 }
848 
849 /* Call with ovs_mutex */
850 static void ct_limit_set(const struct ovs_ct_limit_info *info,
851 			 struct ovs_ct_limit *new_ct_limit)
852 {
853 	struct ovs_ct_limit *ct_limit;
854 	struct hlist_head *head;
855 
856 	head = ct_limit_hash_bucket(info, new_ct_limit->zone);
857 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
858 		if (ct_limit->zone == new_ct_limit->zone) {
859 			hlist_replace_rcu(&ct_limit->hlist_node,
860 					  &new_ct_limit->hlist_node);
861 			kfree_rcu(ct_limit, rcu);
862 			return;
863 		}
864 	}
865 
866 	hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
867 }
868 
869 /* Call with ovs_mutex */
870 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
871 {
872 	struct ovs_ct_limit *ct_limit;
873 	struct hlist_head *head;
874 	struct hlist_node *n;
875 
876 	head = ct_limit_hash_bucket(info, zone);
877 	hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
878 		if (ct_limit->zone == zone) {
879 			hlist_del_rcu(&ct_limit->hlist_node);
880 			kfree_rcu(ct_limit, rcu);
881 			return;
882 		}
883 	}
884 }
885 
886 /* Call with RCU read lock */
887 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
888 {
889 	struct ovs_ct_limit *ct_limit;
890 	struct hlist_head *head;
891 
892 	head = ct_limit_hash_bucket(info, zone);
893 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
894 		if (ct_limit->zone == zone)
895 			return ct_limit->limit;
896 	}
897 
898 	return info->default_limit;
899 }
900 
901 static int ovs_ct_check_limit(struct net *net,
902 			      const struct ovs_conntrack_info *info,
903 			      const struct nf_conntrack_tuple *tuple)
904 {
905 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
906 	const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
907 	u32 per_zone_limit, connections;
908 	u32 conncount_key;
909 
910 	conncount_key = info->zone.id;
911 
912 	per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
913 	if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
914 		return 0;
915 
916 	connections = nf_conncount_count(net, ct_limit_info->data,
917 					 &conncount_key, tuple, &info->zone);
918 	if (connections > per_zone_limit)
919 		return -ENOMEM;
920 
921 	return 0;
922 }
923 #endif
924 
925 /* Lookup connection and confirm if unconfirmed. */
926 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
927 			 const struct ovs_conntrack_info *info,
928 			 struct sk_buff *skb)
929 {
930 	enum ip_conntrack_info ctinfo;
931 	struct nf_conn *ct;
932 	int err;
933 
934 	err = __ovs_ct_lookup(net, key, info, skb);
935 	if (err)
936 		return err;
937 
938 	/* The connection could be invalid, in which case this is a no-op.*/
939 	ct = nf_ct_get(skb, &ctinfo);
940 	if (!ct)
941 		return 0;
942 
943 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
944 	if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
945 		if (!nf_ct_is_confirmed(ct)) {
946 			err = ovs_ct_check_limit(net, info,
947 				&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
948 			if (err) {
949 				net_warn_ratelimited("openvswitch: zone: %u "
950 					"exceeds conntrack limit\n",
951 					info->zone.id);
952 				return err;
953 			}
954 		}
955 	}
956 #endif
957 
958 	/* Set the conntrack event mask if given.  NEW and DELETE events have
959 	 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
960 	 * typically would receive many kinds of updates.  Setting the event
961 	 * mask allows those events to be filtered.  The set event mask will
962 	 * remain in effect for the lifetime of the connection unless changed
963 	 * by a further CT action with both the commit flag and the eventmask
964 	 * option. */
965 	if (info->have_eventmask) {
966 		struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
967 
968 		if (cache)
969 			cache->ctmask = info->eventmask;
970 	}
971 
972 	/* Apply changes before confirming the connection so that the initial
973 	 * conntrack NEW netlink event carries the values given in the CT
974 	 * action.
975 	 */
976 	if (info->mark.mask) {
977 		err = ovs_ct_set_mark(ct, key, info->mark.value,
978 				      info->mark.mask);
979 		if (err)
980 			return err;
981 	}
982 	if (!nf_ct_is_confirmed(ct)) {
983 		err = ovs_ct_init_labels(ct, key, &info->labels.value,
984 					 &info->labels.mask);
985 		if (err)
986 			return err;
987 
988 		nf_conn_act_ct_ext_add(skb, ct, ctinfo);
989 	} else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
990 		   labels_nonzero(&info->labels.mask)) {
991 		err = ovs_ct_set_labels(ct, key, &info->labels.value,
992 					&info->labels.mask);
993 		if (err)
994 			return err;
995 	}
996 	/* This will take care of sending queued events even if the connection
997 	 * is already confirmed.
998 	 */
999 	if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1000 		return -EINVAL;
1001 
1002 	return 0;
1003 }
1004 
1005 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1006  * value if 'skb' is freed.
1007  */
1008 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1009 		   struct sw_flow_key *key,
1010 		   const struct ovs_conntrack_info *info)
1011 {
1012 	int nh_ofs;
1013 	int err;
1014 
1015 	/* The conntrack module expects to be working at L3. */
1016 	nh_ofs = skb_network_offset(skb);
1017 	skb_pull_rcsum(skb, nh_ofs);
1018 
1019 	err = nf_ct_skb_network_trim(skb, info->family);
1020 	if (err) {
1021 		kfree_skb(skb);
1022 		return err;
1023 	}
1024 
1025 	if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1026 		err = ovs_ct_handle_fragments(net, key, info->zone.id,
1027 					      info->family, skb);
1028 		if (err)
1029 			return err;
1030 	}
1031 
1032 	if (info->commit)
1033 		err = ovs_ct_commit(net, key, info, skb);
1034 	else
1035 		err = ovs_ct_lookup(net, key, info, skb);
1036 
1037 	skb_push_rcsum(skb, nh_ofs);
1038 	if (err)
1039 		ovs_kfree_skb_reason(skb, OVS_DROP_CONNTRACK);
1040 	return err;
1041 }
1042 
1043 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1044 {
1045 	enum ip_conntrack_info ctinfo;
1046 	struct nf_conn *ct;
1047 
1048 	ct = nf_ct_get(skb, &ctinfo);
1049 
1050 	nf_ct_put(ct);
1051 	nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1052 
1053 	if (key)
1054 		ovs_ct_fill_key(skb, key, false);
1055 
1056 	return 0;
1057 }
1058 
1059 #if IS_ENABLED(CONFIG_NF_NAT)
1060 static int parse_nat(const struct nlattr *attr,
1061 		     struct ovs_conntrack_info *info, bool log)
1062 {
1063 	struct nlattr *a;
1064 	int rem;
1065 	bool have_ip_max = false;
1066 	bool have_proto_max = false;
1067 	bool ip_vers = (info->family == NFPROTO_IPV6);
1068 
1069 	nla_for_each_nested(a, attr, rem) {
1070 		static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1071 			[OVS_NAT_ATTR_SRC] = {0, 0},
1072 			[OVS_NAT_ATTR_DST] = {0, 0},
1073 			[OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1074 						 sizeof(struct in6_addr)},
1075 			[OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1076 						 sizeof(struct in6_addr)},
1077 			[OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1078 			[OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1079 			[OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1080 			[OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1081 			[OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1082 		};
1083 		int type = nla_type(a);
1084 
1085 		if (type > OVS_NAT_ATTR_MAX) {
1086 			OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1087 				  type, OVS_NAT_ATTR_MAX);
1088 			return -EINVAL;
1089 		}
1090 
1091 		if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1092 			OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1093 				  type, nla_len(a),
1094 				  ovs_nat_attr_lens[type][ip_vers]);
1095 			return -EINVAL;
1096 		}
1097 
1098 		switch (type) {
1099 		case OVS_NAT_ATTR_SRC:
1100 		case OVS_NAT_ATTR_DST:
1101 			if (info->nat) {
1102 				OVS_NLERR(log, "Only one type of NAT may be specified");
1103 				return -ERANGE;
1104 			}
1105 			info->nat |= OVS_CT_NAT;
1106 			info->nat |= ((type == OVS_NAT_ATTR_SRC)
1107 					? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1108 			break;
1109 
1110 		case OVS_NAT_ATTR_IP_MIN:
1111 			nla_memcpy(&info->range.min_addr, a,
1112 				   sizeof(info->range.min_addr));
1113 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1114 			break;
1115 
1116 		case OVS_NAT_ATTR_IP_MAX:
1117 			have_ip_max = true;
1118 			nla_memcpy(&info->range.max_addr, a,
1119 				   sizeof(info->range.max_addr));
1120 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1121 			break;
1122 
1123 		case OVS_NAT_ATTR_PROTO_MIN:
1124 			info->range.min_proto.all = htons(nla_get_u16(a));
1125 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1126 			break;
1127 
1128 		case OVS_NAT_ATTR_PROTO_MAX:
1129 			have_proto_max = true;
1130 			info->range.max_proto.all = htons(nla_get_u16(a));
1131 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1132 			break;
1133 
1134 		case OVS_NAT_ATTR_PERSISTENT:
1135 			info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1136 			break;
1137 
1138 		case OVS_NAT_ATTR_PROTO_HASH:
1139 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1140 			break;
1141 
1142 		case OVS_NAT_ATTR_PROTO_RANDOM:
1143 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1144 			break;
1145 
1146 		default:
1147 			OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1148 			return -EINVAL;
1149 		}
1150 	}
1151 
1152 	if (rem > 0) {
1153 		OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1154 		return -EINVAL;
1155 	}
1156 	if (!info->nat) {
1157 		/* Do not allow flags if no type is given. */
1158 		if (info->range.flags) {
1159 			OVS_NLERR(log,
1160 				  "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1161 				  );
1162 			return -EINVAL;
1163 		}
1164 		info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1165 	} else if (!info->commit) {
1166 		OVS_NLERR(log,
1167 			  "NAT attributes may be specified only when CT COMMIT flag is also specified."
1168 			  );
1169 		return -EINVAL;
1170 	}
1171 	/* Allow missing IP_MAX. */
1172 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1173 		memcpy(&info->range.max_addr, &info->range.min_addr,
1174 		       sizeof(info->range.max_addr));
1175 	}
1176 	/* Allow missing PROTO_MAX. */
1177 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1178 	    !have_proto_max) {
1179 		info->range.max_proto.all = info->range.min_proto.all;
1180 	}
1181 	return 0;
1182 }
1183 #endif
1184 
1185 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1186 	[OVS_CT_ATTR_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1187 	[OVS_CT_ATTR_FORCE_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1188 	[OVS_CT_ATTR_ZONE]	= { .minlen = sizeof(u16),
1189 				    .maxlen = sizeof(u16) },
1190 	[OVS_CT_ATTR_MARK]	= { .minlen = sizeof(struct md_mark),
1191 				    .maxlen = sizeof(struct md_mark) },
1192 	[OVS_CT_ATTR_LABELS]	= { .minlen = sizeof(struct md_labels),
1193 				    .maxlen = sizeof(struct md_labels) },
1194 	[OVS_CT_ATTR_HELPER]	= { .minlen = 1,
1195 				    .maxlen = NF_CT_HELPER_NAME_LEN },
1196 #if IS_ENABLED(CONFIG_NF_NAT)
1197 	/* NAT length is checked when parsing the nested attributes. */
1198 	[OVS_CT_ATTR_NAT]	= { .minlen = 0, .maxlen = INT_MAX },
1199 #endif
1200 	[OVS_CT_ATTR_EVENTMASK]	= { .minlen = sizeof(u32),
1201 				    .maxlen = sizeof(u32) },
1202 	[OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1203 				  .maxlen = CTNL_TIMEOUT_NAME_MAX },
1204 };
1205 
1206 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1207 		    const char **helper, bool log)
1208 {
1209 	struct nlattr *a;
1210 	int rem;
1211 
1212 	nla_for_each_nested(a, attr, rem) {
1213 		int type = nla_type(a);
1214 		int maxlen;
1215 		int minlen;
1216 
1217 		if (type > OVS_CT_ATTR_MAX) {
1218 			OVS_NLERR(log,
1219 				  "Unknown conntrack attr (type=%d, max=%d)",
1220 				  type, OVS_CT_ATTR_MAX);
1221 			return -EINVAL;
1222 		}
1223 
1224 		maxlen = ovs_ct_attr_lens[type].maxlen;
1225 		minlen = ovs_ct_attr_lens[type].minlen;
1226 		if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1227 			OVS_NLERR(log,
1228 				  "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1229 				  type, nla_len(a), maxlen);
1230 			return -EINVAL;
1231 		}
1232 
1233 		switch (type) {
1234 		case OVS_CT_ATTR_FORCE_COMMIT:
1235 			info->force = true;
1236 			fallthrough;
1237 		case OVS_CT_ATTR_COMMIT:
1238 			info->commit = true;
1239 			break;
1240 #ifdef CONFIG_NF_CONNTRACK_ZONES
1241 		case OVS_CT_ATTR_ZONE:
1242 			info->zone.id = nla_get_u16(a);
1243 			break;
1244 #endif
1245 #ifdef CONFIG_NF_CONNTRACK_MARK
1246 		case OVS_CT_ATTR_MARK: {
1247 			struct md_mark *mark = nla_data(a);
1248 
1249 			if (!mark->mask) {
1250 				OVS_NLERR(log, "ct_mark mask cannot be 0");
1251 				return -EINVAL;
1252 			}
1253 			info->mark = *mark;
1254 			break;
1255 		}
1256 #endif
1257 #ifdef CONFIG_NF_CONNTRACK_LABELS
1258 		case OVS_CT_ATTR_LABELS: {
1259 			struct md_labels *labels = nla_data(a);
1260 
1261 			if (!labels_nonzero(&labels->mask)) {
1262 				OVS_NLERR(log, "ct_labels mask cannot be 0");
1263 				return -EINVAL;
1264 			}
1265 			info->labels = *labels;
1266 			break;
1267 		}
1268 #endif
1269 		case OVS_CT_ATTR_HELPER:
1270 			*helper = nla_data(a);
1271 			if (!string_is_terminated(*helper, nla_len(a))) {
1272 				OVS_NLERR(log, "Invalid conntrack helper");
1273 				return -EINVAL;
1274 			}
1275 			break;
1276 #if IS_ENABLED(CONFIG_NF_NAT)
1277 		case OVS_CT_ATTR_NAT: {
1278 			int err = parse_nat(a, info, log);
1279 
1280 			if (err)
1281 				return err;
1282 			break;
1283 		}
1284 #endif
1285 		case OVS_CT_ATTR_EVENTMASK:
1286 			info->have_eventmask = true;
1287 			info->eventmask = nla_get_u32(a);
1288 			break;
1289 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1290 		case OVS_CT_ATTR_TIMEOUT:
1291 			memcpy(info->timeout, nla_data(a), nla_len(a));
1292 			if (!string_is_terminated(info->timeout, nla_len(a))) {
1293 				OVS_NLERR(log, "Invalid conntrack timeout");
1294 				return -EINVAL;
1295 			}
1296 			break;
1297 #endif
1298 
1299 		default:
1300 			OVS_NLERR(log, "Unknown conntrack attr (%d)",
1301 				  type);
1302 			return -EINVAL;
1303 		}
1304 	}
1305 
1306 #ifdef CONFIG_NF_CONNTRACK_MARK
1307 	if (!info->commit && info->mark.mask) {
1308 		OVS_NLERR(log,
1309 			  "Setting conntrack mark requires 'commit' flag.");
1310 		return -EINVAL;
1311 	}
1312 #endif
1313 #ifdef CONFIG_NF_CONNTRACK_LABELS
1314 	if (!info->commit && labels_nonzero(&info->labels.mask)) {
1315 		OVS_NLERR(log,
1316 			  "Setting conntrack labels requires 'commit' flag.");
1317 		return -EINVAL;
1318 	}
1319 #endif
1320 	if (rem > 0) {
1321 		OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1322 		return -EINVAL;
1323 	}
1324 
1325 	return 0;
1326 }
1327 
1328 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1329 {
1330 	if (attr == OVS_KEY_ATTR_CT_STATE)
1331 		return true;
1332 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1333 	    attr == OVS_KEY_ATTR_CT_ZONE)
1334 		return true;
1335 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1336 	    attr == OVS_KEY_ATTR_CT_MARK)
1337 		return true;
1338 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1339 	    attr == OVS_KEY_ATTR_CT_LABELS) {
1340 		struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1341 
1342 		return ovs_net->xt_label;
1343 	}
1344 
1345 	return false;
1346 }
1347 
1348 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1349 		       const struct sw_flow_key *key,
1350 		       struct sw_flow_actions **sfa,  bool log)
1351 {
1352 	struct ovs_conntrack_info ct_info;
1353 	const char *helper = NULL;
1354 	u16 family;
1355 	int err;
1356 
1357 	family = key_to_nfproto(key);
1358 	if (family == NFPROTO_UNSPEC) {
1359 		OVS_NLERR(log, "ct family unspecified");
1360 		return -EINVAL;
1361 	}
1362 
1363 	memset(&ct_info, 0, sizeof(ct_info));
1364 	ct_info.family = family;
1365 
1366 	nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1367 			NF_CT_DEFAULT_ZONE_DIR, 0);
1368 
1369 	err = parse_ct(attr, &ct_info, &helper, log);
1370 	if (err)
1371 		return err;
1372 
1373 	/* Set up template for tracking connections in specific zones. */
1374 	ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1375 	if (!ct_info.ct) {
1376 		OVS_NLERR(log, "Failed to allocate conntrack template");
1377 		return -ENOMEM;
1378 	}
1379 
1380 	if (ct_info.timeout[0]) {
1381 		if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1382 				      ct_info.timeout))
1383 			pr_info_ratelimited("Failed to associated timeout "
1384 					    "policy `%s'\n", ct_info.timeout);
1385 		else
1386 			ct_info.nf_ct_timeout = rcu_dereference(
1387 				nf_ct_timeout_find(ct_info.ct)->timeout);
1388 
1389 	}
1390 
1391 	if (helper) {
1392 		err = nf_ct_add_helper(ct_info.ct, helper, ct_info.family,
1393 				       key->ip.proto, ct_info.nat, &ct_info.helper);
1394 		if (err) {
1395 			OVS_NLERR(log, "Failed to add %s helper %d", helper, err);
1396 			goto err_free_ct;
1397 		}
1398 	}
1399 
1400 	err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1401 				 sizeof(ct_info), log);
1402 	if (err)
1403 		goto err_free_ct;
1404 
1405 	if (ct_info.commit)
1406 		__set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1407 	return 0;
1408 err_free_ct:
1409 	__ovs_ct_free_action(&ct_info);
1410 	return err;
1411 }
1412 
1413 #if IS_ENABLED(CONFIG_NF_NAT)
1414 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1415 			       struct sk_buff *skb)
1416 {
1417 	struct nlattr *start;
1418 
1419 	start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1420 	if (!start)
1421 		return false;
1422 
1423 	if (info->nat & OVS_CT_SRC_NAT) {
1424 		if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1425 			return false;
1426 	} else if (info->nat & OVS_CT_DST_NAT) {
1427 		if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1428 			return false;
1429 	} else {
1430 		goto out;
1431 	}
1432 
1433 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1434 		if (IS_ENABLED(CONFIG_NF_NAT) &&
1435 		    info->family == NFPROTO_IPV4) {
1436 			if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1437 					    info->range.min_addr.ip) ||
1438 			    (info->range.max_addr.ip
1439 			     != info->range.min_addr.ip &&
1440 			     (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1441 					      info->range.max_addr.ip))))
1442 				return false;
1443 		} else if (IS_ENABLED(CONFIG_IPV6) &&
1444 			   info->family == NFPROTO_IPV6) {
1445 			if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1446 					     &info->range.min_addr.in6) ||
1447 			    (memcmp(&info->range.max_addr.in6,
1448 				    &info->range.min_addr.in6,
1449 				    sizeof(info->range.max_addr.in6)) &&
1450 			     (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1451 					       &info->range.max_addr.in6))))
1452 				return false;
1453 		} else {
1454 			return false;
1455 		}
1456 	}
1457 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1458 	    (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1459 			 ntohs(info->range.min_proto.all)) ||
1460 	     (info->range.max_proto.all != info->range.min_proto.all &&
1461 	      nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1462 			  ntohs(info->range.max_proto.all)))))
1463 		return false;
1464 
1465 	if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1466 	    nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1467 		return false;
1468 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1469 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1470 		return false;
1471 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1472 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1473 		return false;
1474 out:
1475 	nla_nest_end(skb, start);
1476 
1477 	return true;
1478 }
1479 #endif
1480 
1481 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1482 			  struct sk_buff *skb)
1483 {
1484 	struct nlattr *start;
1485 
1486 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1487 	if (!start)
1488 		return -EMSGSIZE;
1489 
1490 	if (ct_info->commit && nla_put_flag(skb, ct_info->force
1491 					    ? OVS_CT_ATTR_FORCE_COMMIT
1492 					    : OVS_CT_ATTR_COMMIT))
1493 		return -EMSGSIZE;
1494 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1495 	    nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1496 		return -EMSGSIZE;
1497 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1498 	    nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1499 		    &ct_info->mark))
1500 		return -EMSGSIZE;
1501 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1502 	    labels_nonzero(&ct_info->labels.mask) &&
1503 	    nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1504 		    &ct_info->labels))
1505 		return -EMSGSIZE;
1506 	if (ct_info->helper) {
1507 		if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1508 				   ct_info->helper->name))
1509 			return -EMSGSIZE;
1510 	}
1511 	if (ct_info->have_eventmask &&
1512 	    nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1513 		return -EMSGSIZE;
1514 	if (ct_info->timeout[0]) {
1515 		if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1516 			return -EMSGSIZE;
1517 	}
1518 
1519 #if IS_ENABLED(CONFIG_NF_NAT)
1520 	if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1521 		return -EMSGSIZE;
1522 #endif
1523 	nla_nest_end(skb, start);
1524 
1525 	return 0;
1526 }
1527 
1528 void ovs_ct_free_action(const struct nlattr *a)
1529 {
1530 	struct ovs_conntrack_info *ct_info = nla_data(a);
1531 
1532 	__ovs_ct_free_action(ct_info);
1533 }
1534 
1535 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1536 {
1537 	if (ct_info->helper) {
1538 #if IS_ENABLED(CONFIG_NF_NAT)
1539 		if (ct_info->nat)
1540 			nf_nat_helper_put(ct_info->helper);
1541 #endif
1542 		nf_conntrack_helper_put(ct_info->helper);
1543 	}
1544 	if (ct_info->ct) {
1545 		if (ct_info->timeout[0])
1546 			nf_ct_destroy_timeout(ct_info->ct);
1547 		nf_ct_tmpl_free(ct_info->ct);
1548 	}
1549 }
1550 
1551 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1552 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1553 {
1554 	int i, err;
1555 
1556 	ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1557 					 GFP_KERNEL);
1558 	if (!ovs_net->ct_limit_info)
1559 		return -ENOMEM;
1560 
1561 	ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1562 	ovs_net->ct_limit_info->limits =
1563 		kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1564 			      GFP_KERNEL);
1565 	if (!ovs_net->ct_limit_info->limits) {
1566 		kfree(ovs_net->ct_limit_info);
1567 		return -ENOMEM;
1568 	}
1569 
1570 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1571 		INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1572 
1573 	ovs_net->ct_limit_info->data =
1574 		nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1575 
1576 	if (IS_ERR(ovs_net->ct_limit_info->data)) {
1577 		err = PTR_ERR(ovs_net->ct_limit_info->data);
1578 		kfree(ovs_net->ct_limit_info->limits);
1579 		kfree(ovs_net->ct_limit_info);
1580 		pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1581 		return err;
1582 	}
1583 	return 0;
1584 }
1585 
1586 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1587 {
1588 	const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1589 	int i;
1590 
1591 	nf_conncount_destroy(net, NFPROTO_INET, info->data);
1592 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1593 		struct hlist_head *head = &info->limits[i];
1594 		struct ovs_ct_limit *ct_limit;
1595 
1596 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node,
1597 					 lockdep_ovsl_is_held())
1598 			kfree_rcu(ct_limit, rcu);
1599 	}
1600 	kfree(info->limits);
1601 	kfree(info);
1602 }
1603 
1604 static struct sk_buff *
1605 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1606 			     struct ovs_header **ovs_reply_header)
1607 {
1608 	struct ovs_header *ovs_header = genl_info_userhdr(info);
1609 	struct sk_buff *skb;
1610 
1611 	skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1612 	if (!skb)
1613 		return ERR_PTR(-ENOMEM);
1614 
1615 	*ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1616 					info->snd_seq,
1617 					&dp_ct_limit_genl_family, 0, cmd);
1618 
1619 	if (!*ovs_reply_header) {
1620 		nlmsg_free(skb);
1621 		return ERR_PTR(-EMSGSIZE);
1622 	}
1623 	(*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1624 
1625 	return skb;
1626 }
1627 
1628 static bool check_zone_id(int zone_id, u16 *pzone)
1629 {
1630 	if (zone_id >= 0 && zone_id <= 65535) {
1631 		*pzone = (u16)zone_id;
1632 		return true;
1633 	}
1634 	return false;
1635 }
1636 
1637 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1638 				       struct ovs_ct_limit_info *info)
1639 {
1640 	struct ovs_zone_limit *zone_limit;
1641 	int rem;
1642 	u16 zone;
1643 
1644 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1645 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1646 
1647 	while (rem >= sizeof(*zone_limit)) {
1648 		if (unlikely(zone_limit->zone_id ==
1649 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1650 			ovs_lock();
1651 			info->default_limit = zone_limit->limit;
1652 			ovs_unlock();
1653 		} else if (unlikely(!check_zone_id(
1654 				zone_limit->zone_id, &zone))) {
1655 			OVS_NLERR(true, "zone id is out of range");
1656 		} else {
1657 			struct ovs_ct_limit *ct_limit;
1658 
1659 			ct_limit = kmalloc(sizeof(*ct_limit),
1660 					   GFP_KERNEL_ACCOUNT);
1661 			if (!ct_limit)
1662 				return -ENOMEM;
1663 
1664 			ct_limit->zone = zone;
1665 			ct_limit->limit = zone_limit->limit;
1666 
1667 			ovs_lock();
1668 			ct_limit_set(info, ct_limit);
1669 			ovs_unlock();
1670 		}
1671 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1672 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1673 				NLA_ALIGN(sizeof(*zone_limit)));
1674 	}
1675 
1676 	if (rem)
1677 		OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1678 
1679 	return 0;
1680 }
1681 
1682 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1683 				       struct ovs_ct_limit_info *info)
1684 {
1685 	struct ovs_zone_limit *zone_limit;
1686 	int rem;
1687 	u16 zone;
1688 
1689 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1690 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1691 
1692 	while (rem >= sizeof(*zone_limit)) {
1693 		if (unlikely(zone_limit->zone_id ==
1694 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1695 			ovs_lock();
1696 			info->default_limit = OVS_CT_LIMIT_DEFAULT;
1697 			ovs_unlock();
1698 		} else if (unlikely(!check_zone_id(
1699 				zone_limit->zone_id, &zone))) {
1700 			OVS_NLERR(true, "zone id is out of range");
1701 		} else {
1702 			ovs_lock();
1703 			ct_limit_del(info, zone);
1704 			ovs_unlock();
1705 		}
1706 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1707 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1708 				NLA_ALIGN(sizeof(*zone_limit)));
1709 	}
1710 
1711 	if (rem)
1712 		OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
1713 
1714 	return 0;
1715 }
1716 
1717 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
1718 					  struct sk_buff *reply)
1719 {
1720 	struct ovs_zone_limit zone_limit = {
1721 		.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE,
1722 		.limit   = info->default_limit,
1723 	};
1724 
1725 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1726 }
1727 
1728 static int __ovs_ct_limit_get_zone_limit(struct net *net,
1729 					 struct nf_conncount_data *data,
1730 					 u16 zone_id, u32 limit,
1731 					 struct sk_buff *reply)
1732 {
1733 	struct nf_conntrack_zone ct_zone;
1734 	struct ovs_zone_limit zone_limit;
1735 	u32 conncount_key = zone_id;
1736 
1737 	zone_limit.zone_id = zone_id;
1738 	zone_limit.limit = limit;
1739 	nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
1740 
1741 	zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
1742 					      &ct_zone);
1743 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1744 }
1745 
1746 static int ovs_ct_limit_get_zone_limit(struct net *net,
1747 				       struct nlattr *nla_zone_limit,
1748 				       struct ovs_ct_limit_info *info,
1749 				       struct sk_buff *reply)
1750 {
1751 	struct ovs_zone_limit *zone_limit;
1752 	int rem, err;
1753 	u32 limit;
1754 	u16 zone;
1755 
1756 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1757 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1758 
1759 	while (rem >= sizeof(*zone_limit)) {
1760 		if (unlikely(zone_limit->zone_id ==
1761 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1762 			err = ovs_ct_limit_get_default_limit(info, reply);
1763 			if (err)
1764 				return err;
1765 		} else if (unlikely(!check_zone_id(zone_limit->zone_id,
1766 							&zone))) {
1767 			OVS_NLERR(true, "zone id is out of range");
1768 		} else {
1769 			rcu_read_lock();
1770 			limit = ct_limit_get(info, zone);
1771 			rcu_read_unlock();
1772 
1773 			err = __ovs_ct_limit_get_zone_limit(
1774 				net, info->data, zone, limit, reply);
1775 			if (err)
1776 				return err;
1777 		}
1778 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1779 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1780 				NLA_ALIGN(sizeof(*zone_limit)));
1781 	}
1782 
1783 	if (rem)
1784 		OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
1785 
1786 	return 0;
1787 }
1788 
1789 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
1790 					   struct ovs_ct_limit_info *info,
1791 					   struct sk_buff *reply)
1792 {
1793 	struct ovs_ct_limit *ct_limit;
1794 	struct hlist_head *head;
1795 	int i, err = 0;
1796 
1797 	err = ovs_ct_limit_get_default_limit(info, reply);
1798 	if (err)
1799 		return err;
1800 
1801 	rcu_read_lock();
1802 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1803 		head = &info->limits[i];
1804 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1805 			err = __ovs_ct_limit_get_zone_limit(net, info->data,
1806 				ct_limit->zone, ct_limit->limit, reply);
1807 			if (err)
1808 				goto exit_err;
1809 		}
1810 	}
1811 
1812 exit_err:
1813 	rcu_read_unlock();
1814 	return err;
1815 }
1816 
1817 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
1818 {
1819 	struct nlattr **a = info->attrs;
1820 	struct sk_buff *reply;
1821 	struct ovs_header *ovs_reply_header;
1822 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1823 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1824 	int err;
1825 
1826 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
1827 					     &ovs_reply_header);
1828 	if (IS_ERR(reply))
1829 		return PTR_ERR(reply);
1830 
1831 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1832 		err = -EINVAL;
1833 		goto exit_err;
1834 	}
1835 
1836 	err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
1837 					  ct_limit_info);
1838 	if (err)
1839 		goto exit_err;
1840 
1841 	static_branch_enable(&ovs_ct_limit_enabled);
1842 
1843 	genlmsg_end(reply, ovs_reply_header);
1844 	return genlmsg_reply(reply, info);
1845 
1846 exit_err:
1847 	nlmsg_free(reply);
1848 	return err;
1849 }
1850 
1851 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
1852 {
1853 	struct nlattr **a = info->attrs;
1854 	struct sk_buff *reply;
1855 	struct ovs_header *ovs_reply_header;
1856 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1857 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1858 	int err;
1859 
1860 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
1861 					     &ovs_reply_header);
1862 	if (IS_ERR(reply))
1863 		return PTR_ERR(reply);
1864 
1865 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1866 		err = -EINVAL;
1867 		goto exit_err;
1868 	}
1869 
1870 	err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
1871 					  ct_limit_info);
1872 	if (err)
1873 		goto exit_err;
1874 
1875 	genlmsg_end(reply, ovs_reply_header);
1876 	return genlmsg_reply(reply, info);
1877 
1878 exit_err:
1879 	nlmsg_free(reply);
1880 	return err;
1881 }
1882 
1883 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
1884 {
1885 	struct nlattr **a = info->attrs;
1886 	struct nlattr *nla_reply;
1887 	struct sk_buff *reply;
1888 	struct ovs_header *ovs_reply_header;
1889 	struct net *net = sock_net(skb->sk);
1890 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1891 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1892 	int err;
1893 
1894 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
1895 					     &ovs_reply_header);
1896 	if (IS_ERR(reply))
1897 		return PTR_ERR(reply);
1898 
1899 	nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
1900 	if (!nla_reply) {
1901 		err = -EMSGSIZE;
1902 		goto exit_err;
1903 	}
1904 
1905 	if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1906 		err = ovs_ct_limit_get_zone_limit(
1907 			net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
1908 			reply);
1909 		if (err)
1910 			goto exit_err;
1911 	} else {
1912 		err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
1913 						      reply);
1914 		if (err)
1915 			goto exit_err;
1916 	}
1917 
1918 	nla_nest_end(reply, nla_reply);
1919 	genlmsg_end(reply, ovs_reply_header);
1920 	return genlmsg_reply(reply, info);
1921 
1922 exit_err:
1923 	nlmsg_free(reply);
1924 	return err;
1925 }
1926 
1927 static const struct genl_small_ops ct_limit_genl_ops[] = {
1928 	{ .cmd = OVS_CT_LIMIT_CMD_SET,
1929 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1930 		.flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
1931 					       * privilege.
1932 					       */
1933 		.doit = ovs_ct_limit_cmd_set,
1934 	},
1935 	{ .cmd = OVS_CT_LIMIT_CMD_DEL,
1936 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1937 		.flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
1938 					       * privilege.
1939 					       */
1940 		.doit = ovs_ct_limit_cmd_del,
1941 	},
1942 	{ .cmd = OVS_CT_LIMIT_CMD_GET,
1943 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1944 		.flags = 0,		  /* OK for unprivileged users. */
1945 		.doit = ovs_ct_limit_cmd_get,
1946 	},
1947 };
1948 
1949 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
1950 	.name = OVS_CT_LIMIT_MCGROUP,
1951 };
1952 
1953 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
1954 	.hdrsize = sizeof(struct ovs_header),
1955 	.name = OVS_CT_LIMIT_FAMILY,
1956 	.version = OVS_CT_LIMIT_VERSION,
1957 	.maxattr = OVS_CT_LIMIT_ATTR_MAX,
1958 	.policy = ct_limit_policy,
1959 	.netnsok = true,
1960 	.parallel_ops = true,
1961 	.small_ops = ct_limit_genl_ops,
1962 	.n_small_ops = ARRAY_SIZE(ct_limit_genl_ops),
1963 	.resv_start_op = OVS_CT_LIMIT_CMD_GET + 1,
1964 	.mcgrps = &ovs_ct_limit_multicast_group,
1965 	.n_mcgrps = 1,
1966 	.module = THIS_MODULE,
1967 };
1968 #endif
1969 
1970 int ovs_ct_init(struct net *net)
1971 {
1972 	unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1973 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1974 
1975 	if (nf_connlabels_get(net, n_bits - 1)) {
1976 		ovs_net->xt_label = false;
1977 		OVS_NLERR(true, "Failed to set connlabel length");
1978 	} else {
1979 		ovs_net->xt_label = true;
1980 	}
1981 
1982 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1983 	return ovs_ct_limit_init(net, ovs_net);
1984 #else
1985 	return 0;
1986 #endif
1987 }
1988 
1989 void ovs_ct_exit(struct net *net)
1990 {
1991 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1992 
1993 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1994 	ovs_ct_limit_exit(net, ovs_net);
1995 #endif
1996 
1997 	if (ovs_net->xt_label)
1998 		nf_connlabels_put(net);
1999 }
2000