xref: /openbmc/linux/net/openvswitch/conntrack.c (revision 7cffcade)
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 <net/ip.h>
13 #include <net/genetlink.h>
14 #include <net/netfilter/nf_conntrack_core.h>
15 #include <net/netfilter/nf_conntrack_count.h>
16 #include <net/netfilter/nf_conntrack_helper.h>
17 #include <net/netfilter/nf_conntrack_labels.h>
18 #include <net/netfilter/nf_conntrack_seqadj.h>
19 #include <net/netfilter/nf_conntrack_timeout.h>
20 #include <net/netfilter/nf_conntrack_zones.h>
21 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
22 #include <net/ipv6_frag.h>
23 
24 #if IS_ENABLED(CONFIG_NF_NAT)
25 #include <net/netfilter/nf_nat.h>
26 #endif
27 
28 #include <net/netfilter/nf_conntrack_act_ct.h>
29 
30 #include "datapath.h"
31 #include "conntrack.h"
32 #include "flow.h"
33 #include "flow_netlink.h"
34 
35 struct ovs_ct_len_tbl {
36 	int maxlen;
37 	int minlen;
38 };
39 
40 /* Metadata mark for masked write to conntrack mark */
41 struct md_mark {
42 	u32 value;
43 	u32 mask;
44 };
45 
46 /* Metadata label for masked write to conntrack label. */
47 struct md_labels {
48 	struct ovs_key_ct_labels value;
49 	struct ovs_key_ct_labels mask;
50 };
51 
52 enum ovs_ct_nat {
53 	OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
54 	OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
55 	OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
56 };
57 
58 /* Conntrack action context for execution. */
59 struct ovs_conntrack_info {
60 	struct nf_conntrack_helper *helper;
61 	struct nf_conntrack_zone zone;
62 	struct nf_conn *ct;
63 	u8 commit : 1;
64 	u8 nat : 3;                 /* enum ovs_ct_nat */
65 	u8 force : 1;
66 	u8 have_eventmask : 1;
67 	u16 family;
68 	u32 eventmask;              /* Mask of 1 << IPCT_*. */
69 	struct md_mark mark;
70 	struct md_labels labels;
71 	char timeout[CTNL_TIMEOUT_NAME_MAX];
72 	struct nf_ct_timeout *nf_ct_timeout;
73 #if IS_ENABLED(CONFIG_NF_NAT)
74 	struct nf_nat_range2 range;  /* Only present for SRC NAT and DST NAT. */
75 #endif
76 };
77 
78 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
79 #define OVS_CT_LIMIT_UNLIMITED	0
80 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
81 #define CT_LIMIT_HASH_BUCKETS 512
82 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
83 
84 struct ovs_ct_limit {
85 	/* Elements in ovs_ct_limit_info->limits hash table */
86 	struct hlist_node hlist_node;
87 	struct rcu_head rcu;
88 	u16 zone;
89 	u32 limit;
90 };
91 
92 struct ovs_ct_limit_info {
93 	u32 default_limit;
94 	struct hlist_head *limits;
95 	struct nf_conncount_data *data;
96 };
97 
98 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
99 	[OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
100 };
101 #endif
102 
103 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
104 
105 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
106 
107 static u16 key_to_nfproto(const struct sw_flow_key *key)
108 {
109 	switch (ntohs(key->eth.type)) {
110 	case ETH_P_IP:
111 		return NFPROTO_IPV4;
112 	case ETH_P_IPV6:
113 		return NFPROTO_IPV6;
114 	default:
115 		return NFPROTO_UNSPEC;
116 	}
117 }
118 
119 /* Map SKB connection state into the values used by flow definition. */
120 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
121 {
122 	u8 ct_state = OVS_CS_F_TRACKED;
123 
124 	switch (ctinfo) {
125 	case IP_CT_ESTABLISHED_REPLY:
126 	case IP_CT_RELATED_REPLY:
127 		ct_state |= OVS_CS_F_REPLY_DIR;
128 		break;
129 	default:
130 		break;
131 	}
132 
133 	switch (ctinfo) {
134 	case IP_CT_ESTABLISHED:
135 	case IP_CT_ESTABLISHED_REPLY:
136 		ct_state |= OVS_CS_F_ESTABLISHED;
137 		break;
138 	case IP_CT_RELATED:
139 	case IP_CT_RELATED_REPLY:
140 		ct_state |= OVS_CS_F_RELATED;
141 		break;
142 	case IP_CT_NEW:
143 		ct_state |= OVS_CS_F_NEW;
144 		break;
145 	default:
146 		break;
147 	}
148 
149 	return ct_state;
150 }
151 
152 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
153 {
154 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
155 	return ct ? READ_ONCE(ct->mark) : 0;
156 #else
157 	return 0;
158 #endif
159 }
160 
161 /* Guard against conntrack labels max size shrinking below 128 bits. */
162 #if NF_CT_LABELS_MAX_SIZE < 16
163 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
164 #endif
165 
166 static void ovs_ct_get_labels(const struct nf_conn *ct,
167 			      struct ovs_key_ct_labels *labels)
168 {
169 	struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
170 
171 	if (cl)
172 		memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
173 	else
174 		memset(labels, 0, OVS_CT_LABELS_LEN);
175 }
176 
177 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
178 					const struct nf_conntrack_tuple *orig,
179 					u8 icmp_proto)
180 {
181 	key->ct_orig_proto = orig->dst.protonum;
182 	if (orig->dst.protonum == icmp_proto) {
183 		key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
184 		key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
185 	} else {
186 		key->ct.orig_tp.src = orig->src.u.all;
187 		key->ct.orig_tp.dst = orig->dst.u.all;
188 	}
189 }
190 
191 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
192 				const struct nf_conntrack_zone *zone,
193 				const struct nf_conn *ct)
194 {
195 	key->ct_state = state;
196 	key->ct_zone = zone->id;
197 	key->ct.mark = ovs_ct_get_mark(ct);
198 	ovs_ct_get_labels(ct, &key->ct.labels);
199 
200 	if (ct) {
201 		const struct nf_conntrack_tuple *orig;
202 
203 		/* Use the master if we have one. */
204 		if (ct->master)
205 			ct = ct->master;
206 		orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
207 
208 		/* IP version must match with the master connection. */
209 		if (key->eth.type == htons(ETH_P_IP) &&
210 		    nf_ct_l3num(ct) == NFPROTO_IPV4) {
211 			key->ipv4.ct_orig.src = orig->src.u3.ip;
212 			key->ipv4.ct_orig.dst = orig->dst.u3.ip;
213 			__ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
214 			return;
215 		} else if (key->eth.type == htons(ETH_P_IPV6) &&
216 			   !sw_flow_key_is_nd(key) &&
217 			   nf_ct_l3num(ct) == NFPROTO_IPV6) {
218 			key->ipv6.ct_orig.src = orig->src.u3.in6;
219 			key->ipv6.ct_orig.dst = orig->dst.u3.in6;
220 			__ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
221 			return;
222 		}
223 	}
224 	/* Clear 'ct_orig_proto' to mark the non-existence of conntrack
225 	 * original direction key fields.
226 	 */
227 	key->ct_orig_proto = 0;
228 }
229 
230 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
231  * previously sent the packet to conntrack via the ct action.  If
232  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
233  * initialized from the connection status.
234  */
235 static void ovs_ct_update_key(const struct sk_buff *skb,
236 			      const struct ovs_conntrack_info *info,
237 			      struct sw_flow_key *key, bool post_ct,
238 			      bool keep_nat_flags)
239 {
240 	const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
241 	enum ip_conntrack_info ctinfo;
242 	struct nf_conn *ct;
243 	u8 state = 0;
244 
245 	ct = nf_ct_get(skb, &ctinfo);
246 	if (ct) {
247 		state = ovs_ct_get_state(ctinfo);
248 		/* All unconfirmed entries are NEW connections. */
249 		if (!nf_ct_is_confirmed(ct))
250 			state |= OVS_CS_F_NEW;
251 		/* OVS persists the related flag for the duration of the
252 		 * connection.
253 		 */
254 		if (ct->master)
255 			state |= OVS_CS_F_RELATED;
256 		if (keep_nat_flags) {
257 			state |= key->ct_state & OVS_CS_F_NAT_MASK;
258 		} else {
259 			if (ct->status & IPS_SRC_NAT)
260 				state |= OVS_CS_F_SRC_NAT;
261 			if (ct->status & IPS_DST_NAT)
262 				state |= OVS_CS_F_DST_NAT;
263 		}
264 		zone = nf_ct_zone(ct);
265 	} else if (post_ct) {
266 		state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
267 		if (info)
268 			zone = &info->zone;
269 	}
270 	__ovs_ct_update_key(key, state, zone, ct);
271 }
272 
273 /* This is called to initialize CT key fields possibly coming in from the local
274  * stack.
275  */
276 void ovs_ct_fill_key(const struct sk_buff *skb,
277 		     struct sw_flow_key *key,
278 		     bool post_ct)
279 {
280 	ovs_ct_update_key(skb, NULL, key, post_ct, false);
281 }
282 
283 int ovs_ct_put_key(const struct sw_flow_key *swkey,
284 		   const struct sw_flow_key *output, struct sk_buff *skb)
285 {
286 	if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
287 		return -EMSGSIZE;
288 
289 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
290 	    nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
291 		return -EMSGSIZE;
292 
293 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
294 	    nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
295 		return -EMSGSIZE;
296 
297 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
298 	    nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
299 		    &output->ct.labels))
300 		return -EMSGSIZE;
301 
302 	if (swkey->ct_orig_proto) {
303 		if (swkey->eth.type == htons(ETH_P_IP)) {
304 			struct ovs_key_ct_tuple_ipv4 orig;
305 
306 			memset(&orig, 0, sizeof(orig));
307 			orig.ipv4_src = output->ipv4.ct_orig.src;
308 			orig.ipv4_dst = output->ipv4.ct_orig.dst;
309 			orig.src_port = output->ct.orig_tp.src;
310 			orig.dst_port = output->ct.orig_tp.dst;
311 			orig.ipv4_proto = output->ct_orig_proto;
312 
313 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
314 				    sizeof(orig), &orig))
315 				return -EMSGSIZE;
316 		} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
317 			struct ovs_key_ct_tuple_ipv6 orig;
318 
319 			memset(&orig, 0, sizeof(orig));
320 			memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32,
321 			       sizeof(orig.ipv6_src));
322 			memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32,
323 			       sizeof(orig.ipv6_dst));
324 			orig.src_port = output->ct.orig_tp.src;
325 			orig.dst_port = output->ct.orig_tp.dst;
326 			orig.ipv6_proto = output->ct_orig_proto;
327 
328 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
329 				    sizeof(orig), &orig))
330 				return -EMSGSIZE;
331 		}
332 	}
333 
334 	return 0;
335 }
336 
337 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
338 			   u32 ct_mark, u32 mask)
339 {
340 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
341 	u32 new_mark;
342 
343 	new_mark = ct_mark | (READ_ONCE(ct->mark) & ~(mask));
344 	if (READ_ONCE(ct->mark) != new_mark) {
345 		WRITE_ONCE(ct->mark, new_mark);
346 		if (nf_ct_is_confirmed(ct))
347 			nf_conntrack_event_cache(IPCT_MARK, ct);
348 		key->ct.mark = new_mark;
349 	}
350 
351 	return 0;
352 #else
353 	return -ENOTSUPP;
354 #endif
355 }
356 
357 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
358 {
359 	struct nf_conn_labels *cl;
360 
361 	cl = nf_ct_labels_find(ct);
362 	if (!cl) {
363 		nf_ct_labels_ext_add(ct);
364 		cl = nf_ct_labels_find(ct);
365 	}
366 
367 	return cl;
368 }
369 
370 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
371  * since the new connection is not yet confirmed, and thus no-one else has
372  * access to it's labels, we simply write them over.
373  */
374 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
375 			      const struct ovs_key_ct_labels *labels,
376 			      const struct ovs_key_ct_labels *mask)
377 {
378 	struct nf_conn_labels *cl, *master_cl;
379 	bool have_mask = labels_nonzero(mask);
380 
381 	/* Inherit master's labels to the related connection? */
382 	master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
383 
384 	if (!master_cl && !have_mask)
385 		return 0;   /* Nothing to do. */
386 
387 	cl = ovs_ct_get_conn_labels(ct);
388 	if (!cl)
389 		return -ENOSPC;
390 
391 	/* Inherit the master's labels, if any. */
392 	if (master_cl)
393 		*cl = *master_cl;
394 
395 	if (have_mask) {
396 		u32 *dst = (u32 *)cl->bits;
397 		int i;
398 
399 		for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
400 			dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
401 				(labels->ct_labels_32[i]
402 				 & mask->ct_labels_32[i]);
403 	}
404 
405 	/* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
406 	 * IPCT_LABEL bit is set in the event cache.
407 	 */
408 	nf_conntrack_event_cache(IPCT_LABEL, ct);
409 
410 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
411 
412 	return 0;
413 }
414 
415 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
416 			     const struct ovs_key_ct_labels *labels,
417 			     const struct ovs_key_ct_labels *mask)
418 {
419 	struct nf_conn_labels *cl;
420 	int err;
421 
422 	cl = ovs_ct_get_conn_labels(ct);
423 	if (!cl)
424 		return -ENOSPC;
425 
426 	err = nf_connlabels_replace(ct, labels->ct_labels_32,
427 				    mask->ct_labels_32,
428 				    OVS_CT_LABELS_LEN_32);
429 	if (err)
430 		return err;
431 
432 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
433 
434 	return 0;
435 }
436 
437 /* 'skb' should already be pulled to nh_ofs. */
438 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
439 {
440 	const struct nf_conntrack_helper *helper;
441 	const struct nf_conn_help *help;
442 	enum ip_conntrack_info ctinfo;
443 	unsigned int protoff;
444 	struct nf_conn *ct;
445 	int err;
446 
447 	ct = nf_ct_get(skb, &ctinfo);
448 	if (!ct || ctinfo == IP_CT_RELATED_REPLY)
449 		return NF_ACCEPT;
450 
451 	help = nfct_help(ct);
452 	if (!help)
453 		return NF_ACCEPT;
454 
455 	helper = rcu_dereference(help->helper);
456 	if (!helper)
457 		return NF_ACCEPT;
458 
459 	switch (proto) {
460 	case NFPROTO_IPV4:
461 		protoff = ip_hdrlen(skb);
462 		break;
463 	case NFPROTO_IPV6: {
464 		u8 nexthdr = ipv6_hdr(skb)->nexthdr;
465 		__be16 frag_off;
466 		int ofs;
467 
468 		ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
469 				       &frag_off);
470 		if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
471 			pr_debug("proto header not found\n");
472 			return NF_ACCEPT;
473 		}
474 		protoff = ofs;
475 		break;
476 	}
477 	default:
478 		WARN_ONCE(1, "helper invoked on non-IP family!");
479 		return NF_DROP;
480 	}
481 
482 	err = helper->help(skb, protoff, ct, ctinfo);
483 	if (err != NF_ACCEPT)
484 		return err;
485 
486 	/* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
487 	 * FTP with NAT) adusting the TCP payload size when mangling IP
488 	 * addresses and/or port numbers in the text-based control connection.
489 	 */
490 	if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
491 	    !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
492 		return NF_DROP;
493 	return NF_ACCEPT;
494 }
495 
496 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
497  * value if 'skb' is freed.
498  */
499 static int handle_fragments(struct net *net, struct sw_flow_key *key,
500 			    u16 zone, struct sk_buff *skb)
501 {
502 	struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
503 	int err;
504 
505 	if (key->eth.type == htons(ETH_P_IP)) {
506 		enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
507 
508 		memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
509 		err = ip_defrag(net, skb, user);
510 		if (err)
511 			return err;
512 
513 		ovs_cb.mru = IPCB(skb)->frag_max_size;
514 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
515 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
516 		enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
517 
518 		memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
519 		err = nf_ct_frag6_gather(net, skb, user);
520 		if (err) {
521 			if (err != -EINPROGRESS)
522 				kfree_skb(skb);
523 			return err;
524 		}
525 
526 		key->ip.proto = ipv6_hdr(skb)->nexthdr;
527 		ovs_cb.mru = IP6CB(skb)->frag_max_size;
528 #endif
529 	} else {
530 		kfree_skb(skb);
531 		return -EPFNOSUPPORT;
532 	}
533 
534 	/* The key extracted from the fragment that completed this datagram
535 	 * likely didn't have an L4 header, so regenerate it.
536 	 */
537 	ovs_flow_key_update_l3l4(skb, key);
538 
539 	key->ip.frag = OVS_FRAG_TYPE_NONE;
540 	skb_clear_hash(skb);
541 	skb->ignore_df = 1;
542 	*OVS_CB(skb) = ovs_cb;
543 
544 	return 0;
545 }
546 
547 static struct nf_conntrack_expect *
548 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
549 		   u16 proto, const struct sk_buff *skb)
550 {
551 	struct nf_conntrack_tuple tuple;
552 	struct nf_conntrack_expect *exp;
553 
554 	if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
555 		return NULL;
556 
557 	exp = __nf_ct_expect_find(net, zone, &tuple);
558 	if (exp) {
559 		struct nf_conntrack_tuple_hash *h;
560 
561 		/* Delete existing conntrack entry, if it clashes with the
562 		 * expectation.  This can happen since conntrack ALGs do not
563 		 * check for clashes between (new) expectations and existing
564 		 * conntrack entries.  nf_conntrack_in() will check the
565 		 * expectations only if a conntrack entry can not be found,
566 		 * which can lead to OVS finding the expectation (here) in the
567 		 * init direction, but which will not be removed by the
568 		 * nf_conntrack_in() call, if a matching conntrack entry is
569 		 * found instead.  In this case all init direction packets
570 		 * would be reported as new related packets, while reply
571 		 * direction packets would be reported as un-related
572 		 * established packets.
573 		 */
574 		h = nf_conntrack_find_get(net, zone, &tuple);
575 		if (h) {
576 			struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
577 
578 			nf_ct_delete(ct, 0, 0);
579 			nf_ct_put(ct);
580 		}
581 	}
582 
583 	return exp;
584 }
585 
586 /* This replicates logic from nf_conntrack_core.c that is not exported. */
587 static enum ip_conntrack_info
588 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
589 {
590 	const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
591 
592 	if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
593 		return IP_CT_ESTABLISHED_REPLY;
594 	/* Once we've had two way comms, always ESTABLISHED. */
595 	if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
596 		return IP_CT_ESTABLISHED;
597 	if (test_bit(IPS_EXPECTED_BIT, &ct->status))
598 		return IP_CT_RELATED;
599 	return IP_CT_NEW;
600 }
601 
602 /* Find an existing connection which this packet belongs to without
603  * re-attributing statistics or modifying the connection state.  This allows an
604  * skb->_nfct lost due to an upcall to be recovered during actions execution.
605  *
606  * Must be called with rcu_read_lock.
607  *
608  * On success, populates skb->_nfct and returns the connection.  Returns NULL
609  * if there is no existing entry.
610  */
611 static struct nf_conn *
612 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
613 		     u8 l3num, struct sk_buff *skb, bool natted)
614 {
615 	struct nf_conntrack_tuple tuple;
616 	struct nf_conntrack_tuple_hash *h;
617 	struct nf_conn *ct;
618 
619 	if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
620 			       net, &tuple)) {
621 		pr_debug("ovs_ct_find_existing: Can't get tuple\n");
622 		return NULL;
623 	}
624 
625 	/* Must invert the tuple if skb has been transformed by NAT. */
626 	if (natted) {
627 		struct nf_conntrack_tuple inverse;
628 
629 		if (!nf_ct_invert_tuple(&inverse, &tuple)) {
630 			pr_debug("ovs_ct_find_existing: Inversion failed!\n");
631 			return NULL;
632 		}
633 		tuple = inverse;
634 	}
635 
636 	/* look for tuple match */
637 	h = nf_conntrack_find_get(net, zone, &tuple);
638 	if (!h)
639 		return NULL;   /* Not found. */
640 
641 	ct = nf_ct_tuplehash_to_ctrack(h);
642 
643 	/* Inverted packet tuple matches the reverse direction conntrack tuple,
644 	 * select the other tuplehash to get the right 'ctinfo' bits for this
645 	 * packet.
646 	 */
647 	if (natted)
648 		h = &ct->tuplehash[!h->tuple.dst.dir];
649 
650 	nf_ct_set(skb, ct, ovs_ct_get_info(h));
651 	return ct;
652 }
653 
654 static
655 struct nf_conn *ovs_ct_executed(struct net *net,
656 				const struct sw_flow_key *key,
657 				const struct ovs_conntrack_info *info,
658 				struct sk_buff *skb,
659 				bool *ct_executed)
660 {
661 	struct nf_conn *ct = NULL;
662 
663 	/* If no ct, check if we have evidence that an existing conntrack entry
664 	 * might be found for this skb.  This happens when we lose a skb->_nfct
665 	 * due to an upcall, or if the direction is being forced.  If the
666 	 * connection was not confirmed, it is not cached and needs to be run
667 	 * through conntrack again.
668 	 */
669 	*ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
670 		       !(key->ct_state & OVS_CS_F_INVALID) &&
671 		       (key->ct_zone == info->zone.id);
672 
673 	if (*ct_executed || (!key->ct_state && info->force)) {
674 		ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
675 					  !!(key->ct_state &
676 					  OVS_CS_F_NAT_MASK));
677 	}
678 
679 	return ct;
680 }
681 
682 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
683 static bool skb_nfct_cached(struct net *net,
684 			    const struct sw_flow_key *key,
685 			    const struct ovs_conntrack_info *info,
686 			    struct sk_buff *skb)
687 {
688 	enum ip_conntrack_info ctinfo;
689 	struct nf_conn *ct;
690 	bool ct_executed = true;
691 
692 	ct = nf_ct_get(skb, &ctinfo);
693 	if (!ct)
694 		ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
695 
696 	if (ct)
697 		nf_ct_get(skb, &ctinfo);
698 	else
699 		return false;
700 
701 	if (!net_eq(net, read_pnet(&ct->ct_net)))
702 		return false;
703 	if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
704 		return false;
705 	if (info->helper) {
706 		struct nf_conn_help *help;
707 
708 		help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
709 		if (help && rcu_access_pointer(help->helper) != info->helper)
710 			return false;
711 	}
712 	if (info->nf_ct_timeout) {
713 		struct nf_conn_timeout *timeout_ext;
714 
715 		timeout_ext = nf_ct_timeout_find(ct);
716 		if (!timeout_ext || info->nf_ct_timeout !=
717 		    rcu_dereference(timeout_ext->timeout))
718 			return false;
719 	}
720 	/* Force conntrack entry direction to the current packet? */
721 	if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
722 		/* Delete the conntrack entry if confirmed, else just release
723 		 * the reference.
724 		 */
725 		if (nf_ct_is_confirmed(ct))
726 			nf_ct_delete(ct, 0, 0);
727 
728 		nf_ct_put(ct);
729 		nf_ct_set(skb, NULL, 0);
730 		return false;
731 	}
732 
733 	return ct_executed;
734 }
735 
736 #if IS_ENABLED(CONFIG_NF_NAT)
737 static void ovs_nat_update_key(struct sw_flow_key *key,
738 			       const struct sk_buff *skb,
739 			       enum nf_nat_manip_type maniptype)
740 {
741 	if (maniptype == NF_NAT_MANIP_SRC) {
742 		__be16 src;
743 
744 		key->ct_state |= OVS_CS_F_SRC_NAT;
745 		if (key->eth.type == htons(ETH_P_IP))
746 			key->ipv4.addr.src = ip_hdr(skb)->saddr;
747 		else if (key->eth.type == htons(ETH_P_IPV6))
748 			memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
749 			       sizeof(key->ipv6.addr.src));
750 		else
751 			return;
752 
753 		if (key->ip.proto == IPPROTO_UDP)
754 			src = udp_hdr(skb)->source;
755 		else if (key->ip.proto == IPPROTO_TCP)
756 			src = tcp_hdr(skb)->source;
757 		else if (key->ip.proto == IPPROTO_SCTP)
758 			src = sctp_hdr(skb)->source;
759 		else
760 			return;
761 
762 		key->tp.src = src;
763 	} else {
764 		__be16 dst;
765 
766 		key->ct_state |= OVS_CS_F_DST_NAT;
767 		if (key->eth.type == htons(ETH_P_IP))
768 			key->ipv4.addr.dst = ip_hdr(skb)->daddr;
769 		else if (key->eth.type == htons(ETH_P_IPV6))
770 			memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
771 			       sizeof(key->ipv6.addr.dst));
772 		else
773 			return;
774 
775 		if (key->ip.proto == IPPROTO_UDP)
776 			dst = udp_hdr(skb)->dest;
777 		else if (key->ip.proto == IPPROTO_TCP)
778 			dst = tcp_hdr(skb)->dest;
779 		else if (key->ip.proto == IPPROTO_SCTP)
780 			dst = sctp_hdr(skb)->dest;
781 		else
782 			return;
783 
784 		key->tp.dst = dst;
785 	}
786 }
787 
788 /* Modelled after nf_nat_ipv[46]_fn().
789  * range is only used for new, uninitialized NAT state.
790  * Returns either NF_ACCEPT or NF_DROP.
791  */
792 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
793 			      enum ip_conntrack_info ctinfo,
794 			      const struct nf_nat_range2 *range,
795 			      enum nf_nat_manip_type maniptype, struct sw_flow_key *key)
796 {
797 	int hooknum, nh_off, err = NF_ACCEPT;
798 
799 	nh_off = skb_network_offset(skb);
800 	skb_pull_rcsum(skb, nh_off);
801 
802 	/* See HOOK2MANIP(). */
803 	if (maniptype == NF_NAT_MANIP_SRC)
804 		hooknum = NF_INET_LOCAL_IN; /* Source NAT */
805 	else
806 		hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
807 
808 	switch (ctinfo) {
809 	case IP_CT_RELATED:
810 	case IP_CT_RELATED_REPLY:
811 		if (IS_ENABLED(CONFIG_NF_NAT) &&
812 		    skb->protocol == htons(ETH_P_IP) &&
813 		    ip_hdr(skb)->protocol == IPPROTO_ICMP) {
814 			if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
815 							   hooknum))
816 				err = NF_DROP;
817 			goto push;
818 		} else if (IS_ENABLED(CONFIG_IPV6) &&
819 			   skb->protocol == htons(ETH_P_IPV6)) {
820 			__be16 frag_off;
821 			u8 nexthdr = ipv6_hdr(skb)->nexthdr;
822 			int hdrlen = ipv6_skip_exthdr(skb,
823 						      sizeof(struct ipv6hdr),
824 						      &nexthdr, &frag_off);
825 
826 			if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
827 				if (!nf_nat_icmpv6_reply_translation(skb, ct,
828 								     ctinfo,
829 								     hooknum,
830 								     hdrlen))
831 					err = NF_DROP;
832 				goto push;
833 			}
834 		}
835 		/* Non-ICMP, fall thru to initialize if needed. */
836 		fallthrough;
837 	case IP_CT_NEW:
838 		/* Seen it before?  This can happen for loopback, retrans,
839 		 * or local packets.
840 		 */
841 		if (!nf_nat_initialized(ct, maniptype)) {
842 			/* Initialize according to the NAT action. */
843 			err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
844 				/* Action is set up to establish a new
845 				 * mapping.
846 				 */
847 				? nf_nat_setup_info(ct, range, maniptype)
848 				: nf_nat_alloc_null_binding(ct, hooknum);
849 			if (err != NF_ACCEPT)
850 				goto push;
851 		}
852 		break;
853 
854 	case IP_CT_ESTABLISHED:
855 	case IP_CT_ESTABLISHED_REPLY:
856 		break;
857 
858 	default:
859 		err = NF_DROP;
860 		goto push;
861 	}
862 
863 	err = nf_nat_packet(ct, ctinfo, hooknum, skb);
864 push:
865 	skb_push_rcsum(skb, nh_off);
866 
867 	/* Update the flow key if NAT successful. */
868 	if (err == NF_ACCEPT)
869 		ovs_nat_update_key(key, skb, maniptype);
870 
871 	return err;
872 }
873 
874 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
875 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
876 		      const struct ovs_conntrack_info *info,
877 		      struct sk_buff *skb, struct nf_conn *ct,
878 		      enum ip_conntrack_info ctinfo)
879 {
880 	enum nf_nat_manip_type maniptype;
881 	int err;
882 
883 	/* Add NAT extension if not confirmed yet. */
884 	if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
885 		return NF_ACCEPT;   /* Can't NAT. */
886 
887 	/* Determine NAT type.
888 	 * Check if the NAT type can be deduced from the tracked connection.
889 	 * Make sure new expected connections (IP_CT_RELATED) are NATted only
890 	 * when committing.
891 	 */
892 	if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
893 	    ct->status & IPS_NAT_MASK &&
894 	    (ctinfo != IP_CT_RELATED || info->commit)) {
895 		/* NAT an established or related connection like before. */
896 		if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
897 			/* This is the REPLY direction for a connection
898 			 * for which NAT was applied in the forward
899 			 * direction.  Do the reverse NAT.
900 			 */
901 			maniptype = ct->status & IPS_SRC_NAT
902 				? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
903 		else
904 			maniptype = ct->status & IPS_SRC_NAT
905 				? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
906 	} else if (info->nat & OVS_CT_SRC_NAT) {
907 		maniptype = NF_NAT_MANIP_SRC;
908 	} else if (info->nat & OVS_CT_DST_NAT) {
909 		maniptype = NF_NAT_MANIP_DST;
910 	} else {
911 		return NF_ACCEPT; /* Connection is not NATed. */
912 	}
913 	err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype, key);
914 
915 	if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) {
916 		if (ct->status & IPS_SRC_NAT) {
917 			if (maniptype == NF_NAT_MANIP_SRC)
918 				maniptype = NF_NAT_MANIP_DST;
919 			else
920 				maniptype = NF_NAT_MANIP_SRC;
921 
922 			err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range,
923 						 maniptype, key);
924 		} else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
925 			err = ovs_ct_nat_execute(skb, ct, ctinfo, NULL,
926 						 NF_NAT_MANIP_SRC, key);
927 		}
928 	}
929 
930 	return err;
931 }
932 #else /* !CONFIG_NF_NAT */
933 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
934 		      const struct ovs_conntrack_info *info,
935 		      struct sk_buff *skb, struct nf_conn *ct,
936 		      enum ip_conntrack_info ctinfo)
937 {
938 	return NF_ACCEPT;
939 }
940 #endif
941 
942 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
943  * not done already.  Update key with new CT state after passing the packet
944  * through conntrack.
945  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
946  * set to NULL and 0 will be returned.
947  */
948 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
949 			   const struct ovs_conntrack_info *info,
950 			   struct sk_buff *skb)
951 {
952 	/* If we are recirculating packets to match on conntrack fields and
953 	 * committing with a separate conntrack action,  then we don't need to
954 	 * actually run the packet through conntrack twice unless it's for a
955 	 * different zone.
956 	 */
957 	bool cached = skb_nfct_cached(net, key, info, skb);
958 	enum ip_conntrack_info ctinfo;
959 	struct nf_conn *ct;
960 
961 	if (!cached) {
962 		struct nf_hook_state state = {
963 			.hook = NF_INET_PRE_ROUTING,
964 			.pf = info->family,
965 			.net = net,
966 		};
967 		struct nf_conn *tmpl = info->ct;
968 		int err;
969 
970 		/* Associate skb with specified zone. */
971 		if (tmpl) {
972 			ct = nf_ct_get(skb, &ctinfo);
973 			nf_ct_put(ct);
974 			nf_conntrack_get(&tmpl->ct_general);
975 			nf_ct_set(skb, tmpl, IP_CT_NEW);
976 		}
977 
978 		err = nf_conntrack_in(skb, &state);
979 		if (err != NF_ACCEPT)
980 			return -ENOENT;
981 
982 		/* Clear CT state NAT flags to mark that we have not yet done
983 		 * NAT after the nf_conntrack_in() call.  We can actually clear
984 		 * the whole state, as it will be re-initialized below.
985 		 */
986 		key->ct_state = 0;
987 
988 		/* Update the key, but keep the NAT flags. */
989 		ovs_ct_update_key(skb, info, key, true, true);
990 	}
991 
992 	ct = nf_ct_get(skb, &ctinfo);
993 	if (ct) {
994 		bool add_helper = false;
995 
996 		/* Packets starting a new connection must be NATted before the
997 		 * helper, so that the helper knows about the NAT.  We enforce
998 		 * this by delaying both NAT and helper calls for unconfirmed
999 		 * connections until the committing CT action.  For later
1000 		 * packets NAT and Helper may be called in either order.
1001 		 *
1002 		 * NAT will be done only if the CT action has NAT, and only
1003 		 * once per packet (per zone), as guarded by the NAT bits in
1004 		 * the key->ct_state.
1005 		 */
1006 		if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
1007 		    (nf_ct_is_confirmed(ct) || info->commit) &&
1008 		    ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
1009 			return -EINVAL;
1010 		}
1011 
1012 		/* Userspace may decide to perform a ct lookup without a helper
1013 		 * specified followed by a (recirculate and) commit with one,
1014 		 * or attach a helper in a later commit.  Therefore, for
1015 		 * connections which we will commit, we may need to attach
1016 		 * the helper here.
1017 		 */
1018 		if (!nf_ct_is_confirmed(ct) && info->commit &&
1019 		    info->helper && !nfct_help(ct)) {
1020 			int err = __nf_ct_try_assign_helper(ct, info->ct,
1021 							    GFP_ATOMIC);
1022 			if (err)
1023 				return err;
1024 			add_helper = true;
1025 
1026 			/* helper installed, add seqadj if NAT is required */
1027 			if (info->nat && !nfct_seqadj(ct)) {
1028 				if (!nfct_seqadj_ext_add(ct))
1029 					return -EINVAL;
1030 			}
1031 		}
1032 
1033 		/* Call the helper only if:
1034 		 * - nf_conntrack_in() was executed above ("!cached") or a
1035 		 *   helper was just attached ("add_helper") for a confirmed
1036 		 *   connection, or
1037 		 * - When committing an unconfirmed connection.
1038 		 */
1039 		if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
1040 					      info->commit) &&
1041 		    ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1042 			return -EINVAL;
1043 		}
1044 
1045 		if (nf_ct_protonum(ct) == IPPROTO_TCP &&
1046 		    nf_ct_is_confirmed(ct) && nf_conntrack_tcp_established(ct)) {
1047 			/* Be liberal for tcp packets so that out-of-window
1048 			 * packets are not marked invalid.
1049 			 */
1050 			nf_ct_set_tcp_be_liberal(ct);
1051 		}
1052 
1053 		nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
1054 	}
1055 
1056 	return 0;
1057 }
1058 
1059 /* Lookup connection and read fields into key. */
1060 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1061 			 const struct ovs_conntrack_info *info,
1062 			 struct sk_buff *skb)
1063 {
1064 	struct nf_conntrack_expect *exp;
1065 
1066 	/* If we pass an expected packet through nf_conntrack_in() the
1067 	 * expectation is typically removed, but the packet could still be
1068 	 * lost in upcall processing.  To prevent this from happening we
1069 	 * perform an explicit expectation lookup.  Expected connections are
1070 	 * always new, and will be passed through conntrack only when they are
1071 	 * committed, as it is OK to remove the expectation at that time.
1072 	 */
1073 	exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1074 	if (exp) {
1075 		u8 state;
1076 
1077 		/* NOTE: New connections are NATted and Helped only when
1078 		 * committed, so we are not calling into NAT here.
1079 		 */
1080 		state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1081 		__ovs_ct_update_key(key, state, &info->zone, exp->master);
1082 	} else {
1083 		struct nf_conn *ct;
1084 		int err;
1085 
1086 		err = __ovs_ct_lookup(net, key, info, skb);
1087 		if (err)
1088 			return err;
1089 
1090 		ct = (struct nf_conn *)skb_nfct(skb);
1091 		if (ct)
1092 			nf_ct_deliver_cached_events(ct);
1093 	}
1094 
1095 	return 0;
1096 }
1097 
1098 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1099 {
1100 	size_t i;
1101 
1102 	for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1103 		if (labels->ct_labels_32[i])
1104 			return true;
1105 
1106 	return false;
1107 }
1108 
1109 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1110 static struct hlist_head *ct_limit_hash_bucket(
1111 	const struct ovs_ct_limit_info *info, u16 zone)
1112 {
1113 	return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1114 }
1115 
1116 /* Call with ovs_mutex */
1117 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1118 			 struct ovs_ct_limit *new_ct_limit)
1119 {
1120 	struct ovs_ct_limit *ct_limit;
1121 	struct hlist_head *head;
1122 
1123 	head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1124 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1125 		if (ct_limit->zone == new_ct_limit->zone) {
1126 			hlist_replace_rcu(&ct_limit->hlist_node,
1127 					  &new_ct_limit->hlist_node);
1128 			kfree_rcu(ct_limit, rcu);
1129 			return;
1130 		}
1131 	}
1132 
1133 	hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1134 }
1135 
1136 /* Call with ovs_mutex */
1137 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1138 {
1139 	struct ovs_ct_limit *ct_limit;
1140 	struct hlist_head *head;
1141 	struct hlist_node *n;
1142 
1143 	head = ct_limit_hash_bucket(info, zone);
1144 	hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1145 		if (ct_limit->zone == zone) {
1146 			hlist_del_rcu(&ct_limit->hlist_node);
1147 			kfree_rcu(ct_limit, rcu);
1148 			return;
1149 		}
1150 	}
1151 }
1152 
1153 /* Call with RCU read lock */
1154 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1155 {
1156 	struct ovs_ct_limit *ct_limit;
1157 	struct hlist_head *head;
1158 
1159 	head = ct_limit_hash_bucket(info, zone);
1160 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1161 		if (ct_limit->zone == zone)
1162 			return ct_limit->limit;
1163 	}
1164 
1165 	return info->default_limit;
1166 }
1167 
1168 static int ovs_ct_check_limit(struct net *net,
1169 			      const struct ovs_conntrack_info *info,
1170 			      const struct nf_conntrack_tuple *tuple)
1171 {
1172 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1173 	const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1174 	u32 per_zone_limit, connections;
1175 	u32 conncount_key;
1176 
1177 	conncount_key = info->zone.id;
1178 
1179 	per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1180 	if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1181 		return 0;
1182 
1183 	connections = nf_conncount_count(net, ct_limit_info->data,
1184 					 &conncount_key, tuple, &info->zone);
1185 	if (connections > per_zone_limit)
1186 		return -ENOMEM;
1187 
1188 	return 0;
1189 }
1190 #endif
1191 
1192 /* Lookup connection and confirm if unconfirmed. */
1193 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1194 			 const struct ovs_conntrack_info *info,
1195 			 struct sk_buff *skb)
1196 {
1197 	enum ip_conntrack_info ctinfo;
1198 	struct nf_conn *ct;
1199 	int err;
1200 
1201 	err = __ovs_ct_lookup(net, key, info, skb);
1202 	if (err)
1203 		return err;
1204 
1205 	/* The connection could be invalid, in which case this is a no-op.*/
1206 	ct = nf_ct_get(skb, &ctinfo);
1207 	if (!ct)
1208 		return 0;
1209 
1210 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1211 	if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1212 		if (!nf_ct_is_confirmed(ct)) {
1213 			err = ovs_ct_check_limit(net, info,
1214 				&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1215 			if (err) {
1216 				net_warn_ratelimited("openvswitch: zone: %u "
1217 					"exceeds conntrack limit\n",
1218 					info->zone.id);
1219 				return err;
1220 			}
1221 		}
1222 	}
1223 #endif
1224 
1225 	/* Set the conntrack event mask if given.  NEW and DELETE events have
1226 	 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1227 	 * typically would receive many kinds of updates.  Setting the event
1228 	 * mask allows those events to be filtered.  The set event mask will
1229 	 * remain in effect for the lifetime of the connection unless changed
1230 	 * by a further CT action with both the commit flag and the eventmask
1231 	 * option. */
1232 	if (info->have_eventmask) {
1233 		struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1234 
1235 		if (cache)
1236 			cache->ctmask = info->eventmask;
1237 	}
1238 
1239 	/* Apply changes before confirming the connection so that the initial
1240 	 * conntrack NEW netlink event carries the values given in the CT
1241 	 * action.
1242 	 */
1243 	if (info->mark.mask) {
1244 		err = ovs_ct_set_mark(ct, key, info->mark.value,
1245 				      info->mark.mask);
1246 		if (err)
1247 			return err;
1248 	}
1249 	if (!nf_ct_is_confirmed(ct)) {
1250 		err = ovs_ct_init_labels(ct, key, &info->labels.value,
1251 					 &info->labels.mask);
1252 		if (err)
1253 			return err;
1254 
1255 		nf_conn_act_ct_ext_add(ct);
1256 	} else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1257 		   labels_nonzero(&info->labels.mask)) {
1258 		err = ovs_ct_set_labels(ct, key, &info->labels.value,
1259 					&info->labels.mask);
1260 		if (err)
1261 			return err;
1262 	}
1263 	/* This will take care of sending queued events even if the connection
1264 	 * is already confirmed.
1265 	 */
1266 	if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1267 		return -EINVAL;
1268 
1269 	return 0;
1270 }
1271 
1272 /* Trim the skb to the length specified by the IP/IPv6 header,
1273  * removing any trailing lower-layer padding. This prepares the skb
1274  * for higher-layer processing that assumes skb->len excludes padding
1275  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1276  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1277  */
1278 static int ovs_skb_network_trim(struct sk_buff *skb)
1279 {
1280 	unsigned int len;
1281 	int err;
1282 
1283 	switch (skb->protocol) {
1284 	case htons(ETH_P_IP):
1285 		len = ntohs(ip_hdr(skb)->tot_len);
1286 		break;
1287 	case htons(ETH_P_IPV6):
1288 		len = sizeof(struct ipv6hdr)
1289 			+ ntohs(ipv6_hdr(skb)->payload_len);
1290 		break;
1291 	default:
1292 		len = skb->len;
1293 	}
1294 
1295 	err = pskb_trim_rcsum(skb, len);
1296 	if (err)
1297 		kfree_skb(skb);
1298 
1299 	return err;
1300 }
1301 
1302 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1303  * value if 'skb' is freed.
1304  */
1305 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1306 		   struct sw_flow_key *key,
1307 		   const struct ovs_conntrack_info *info)
1308 {
1309 	int nh_ofs;
1310 	int err;
1311 
1312 	/* The conntrack module expects to be working at L3. */
1313 	nh_ofs = skb_network_offset(skb);
1314 	skb_pull_rcsum(skb, nh_ofs);
1315 
1316 	err = ovs_skb_network_trim(skb);
1317 	if (err)
1318 		return err;
1319 
1320 	if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1321 		err = handle_fragments(net, key, info->zone.id, skb);
1322 		if (err)
1323 			return err;
1324 	}
1325 
1326 	if (info->commit)
1327 		err = ovs_ct_commit(net, key, info, skb);
1328 	else
1329 		err = ovs_ct_lookup(net, key, info, skb);
1330 
1331 	skb_push_rcsum(skb, nh_ofs);
1332 	if (err)
1333 		kfree_skb(skb);
1334 	return err;
1335 }
1336 
1337 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1338 {
1339 	enum ip_conntrack_info ctinfo;
1340 	struct nf_conn *ct;
1341 
1342 	ct = nf_ct_get(skb, &ctinfo);
1343 
1344 	nf_ct_put(ct);
1345 	nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1346 
1347 	if (key)
1348 		ovs_ct_fill_key(skb, key, false);
1349 
1350 	return 0;
1351 }
1352 
1353 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1354 			     const struct sw_flow_key *key, bool log)
1355 {
1356 	struct nf_conntrack_helper *helper;
1357 	struct nf_conn_help *help;
1358 	int ret = 0;
1359 
1360 	helper = nf_conntrack_helper_try_module_get(name, info->family,
1361 						    key->ip.proto);
1362 	if (!helper) {
1363 		OVS_NLERR(log, "Unknown helper \"%s\"", name);
1364 		return -EINVAL;
1365 	}
1366 
1367 	help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1368 	if (!help) {
1369 		nf_conntrack_helper_put(helper);
1370 		return -ENOMEM;
1371 	}
1372 
1373 #if IS_ENABLED(CONFIG_NF_NAT)
1374 	if (info->nat) {
1375 		ret = nf_nat_helper_try_module_get(name, info->family,
1376 						   key->ip.proto);
1377 		if (ret) {
1378 			nf_conntrack_helper_put(helper);
1379 			OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1380 				  name, ret);
1381 			return ret;
1382 		}
1383 	}
1384 #endif
1385 	rcu_assign_pointer(help->helper, helper);
1386 	info->helper = helper;
1387 	return ret;
1388 }
1389 
1390 #if IS_ENABLED(CONFIG_NF_NAT)
1391 static int parse_nat(const struct nlattr *attr,
1392 		     struct ovs_conntrack_info *info, bool log)
1393 {
1394 	struct nlattr *a;
1395 	int rem;
1396 	bool have_ip_max = false;
1397 	bool have_proto_max = false;
1398 	bool ip_vers = (info->family == NFPROTO_IPV6);
1399 
1400 	nla_for_each_nested(a, attr, rem) {
1401 		static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1402 			[OVS_NAT_ATTR_SRC] = {0, 0},
1403 			[OVS_NAT_ATTR_DST] = {0, 0},
1404 			[OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1405 						 sizeof(struct in6_addr)},
1406 			[OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1407 						 sizeof(struct in6_addr)},
1408 			[OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1409 			[OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1410 			[OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1411 			[OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1412 			[OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1413 		};
1414 		int type = nla_type(a);
1415 
1416 		if (type > OVS_NAT_ATTR_MAX) {
1417 			OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1418 				  type, OVS_NAT_ATTR_MAX);
1419 			return -EINVAL;
1420 		}
1421 
1422 		if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1423 			OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1424 				  type, nla_len(a),
1425 				  ovs_nat_attr_lens[type][ip_vers]);
1426 			return -EINVAL;
1427 		}
1428 
1429 		switch (type) {
1430 		case OVS_NAT_ATTR_SRC:
1431 		case OVS_NAT_ATTR_DST:
1432 			if (info->nat) {
1433 				OVS_NLERR(log, "Only one type of NAT may be specified");
1434 				return -ERANGE;
1435 			}
1436 			info->nat |= OVS_CT_NAT;
1437 			info->nat |= ((type == OVS_NAT_ATTR_SRC)
1438 					? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1439 			break;
1440 
1441 		case OVS_NAT_ATTR_IP_MIN:
1442 			nla_memcpy(&info->range.min_addr, a,
1443 				   sizeof(info->range.min_addr));
1444 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1445 			break;
1446 
1447 		case OVS_NAT_ATTR_IP_MAX:
1448 			have_ip_max = true;
1449 			nla_memcpy(&info->range.max_addr, a,
1450 				   sizeof(info->range.max_addr));
1451 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1452 			break;
1453 
1454 		case OVS_NAT_ATTR_PROTO_MIN:
1455 			info->range.min_proto.all = htons(nla_get_u16(a));
1456 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1457 			break;
1458 
1459 		case OVS_NAT_ATTR_PROTO_MAX:
1460 			have_proto_max = true;
1461 			info->range.max_proto.all = htons(nla_get_u16(a));
1462 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1463 			break;
1464 
1465 		case OVS_NAT_ATTR_PERSISTENT:
1466 			info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1467 			break;
1468 
1469 		case OVS_NAT_ATTR_PROTO_HASH:
1470 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1471 			break;
1472 
1473 		case OVS_NAT_ATTR_PROTO_RANDOM:
1474 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1475 			break;
1476 
1477 		default:
1478 			OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1479 			return -EINVAL;
1480 		}
1481 	}
1482 
1483 	if (rem > 0) {
1484 		OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1485 		return -EINVAL;
1486 	}
1487 	if (!info->nat) {
1488 		/* Do not allow flags if no type is given. */
1489 		if (info->range.flags) {
1490 			OVS_NLERR(log,
1491 				  "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1492 				  );
1493 			return -EINVAL;
1494 		}
1495 		info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1496 	} else if (!info->commit) {
1497 		OVS_NLERR(log,
1498 			  "NAT attributes may be specified only when CT COMMIT flag is also specified."
1499 			  );
1500 		return -EINVAL;
1501 	}
1502 	/* Allow missing IP_MAX. */
1503 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1504 		memcpy(&info->range.max_addr, &info->range.min_addr,
1505 		       sizeof(info->range.max_addr));
1506 	}
1507 	/* Allow missing PROTO_MAX. */
1508 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1509 	    !have_proto_max) {
1510 		info->range.max_proto.all = info->range.min_proto.all;
1511 	}
1512 	return 0;
1513 }
1514 #endif
1515 
1516 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1517 	[OVS_CT_ATTR_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1518 	[OVS_CT_ATTR_FORCE_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1519 	[OVS_CT_ATTR_ZONE]	= { .minlen = sizeof(u16),
1520 				    .maxlen = sizeof(u16) },
1521 	[OVS_CT_ATTR_MARK]	= { .minlen = sizeof(struct md_mark),
1522 				    .maxlen = sizeof(struct md_mark) },
1523 	[OVS_CT_ATTR_LABELS]	= { .minlen = sizeof(struct md_labels),
1524 				    .maxlen = sizeof(struct md_labels) },
1525 	[OVS_CT_ATTR_HELPER]	= { .minlen = 1,
1526 				    .maxlen = NF_CT_HELPER_NAME_LEN },
1527 #if IS_ENABLED(CONFIG_NF_NAT)
1528 	/* NAT length is checked when parsing the nested attributes. */
1529 	[OVS_CT_ATTR_NAT]	= { .minlen = 0, .maxlen = INT_MAX },
1530 #endif
1531 	[OVS_CT_ATTR_EVENTMASK]	= { .minlen = sizeof(u32),
1532 				    .maxlen = sizeof(u32) },
1533 	[OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1534 				  .maxlen = CTNL_TIMEOUT_NAME_MAX },
1535 };
1536 
1537 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1538 		    const char **helper, bool log)
1539 {
1540 	struct nlattr *a;
1541 	int rem;
1542 
1543 	nla_for_each_nested(a, attr, rem) {
1544 		int type = nla_type(a);
1545 		int maxlen;
1546 		int minlen;
1547 
1548 		if (type > OVS_CT_ATTR_MAX) {
1549 			OVS_NLERR(log,
1550 				  "Unknown conntrack attr (type=%d, max=%d)",
1551 				  type, OVS_CT_ATTR_MAX);
1552 			return -EINVAL;
1553 		}
1554 
1555 		maxlen = ovs_ct_attr_lens[type].maxlen;
1556 		minlen = ovs_ct_attr_lens[type].minlen;
1557 		if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1558 			OVS_NLERR(log,
1559 				  "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1560 				  type, nla_len(a), maxlen);
1561 			return -EINVAL;
1562 		}
1563 
1564 		switch (type) {
1565 		case OVS_CT_ATTR_FORCE_COMMIT:
1566 			info->force = true;
1567 			fallthrough;
1568 		case OVS_CT_ATTR_COMMIT:
1569 			info->commit = true;
1570 			break;
1571 #ifdef CONFIG_NF_CONNTRACK_ZONES
1572 		case OVS_CT_ATTR_ZONE:
1573 			info->zone.id = nla_get_u16(a);
1574 			break;
1575 #endif
1576 #ifdef CONFIG_NF_CONNTRACK_MARK
1577 		case OVS_CT_ATTR_MARK: {
1578 			struct md_mark *mark = nla_data(a);
1579 
1580 			if (!mark->mask) {
1581 				OVS_NLERR(log, "ct_mark mask cannot be 0");
1582 				return -EINVAL;
1583 			}
1584 			info->mark = *mark;
1585 			break;
1586 		}
1587 #endif
1588 #ifdef CONFIG_NF_CONNTRACK_LABELS
1589 		case OVS_CT_ATTR_LABELS: {
1590 			struct md_labels *labels = nla_data(a);
1591 
1592 			if (!labels_nonzero(&labels->mask)) {
1593 				OVS_NLERR(log, "ct_labels mask cannot be 0");
1594 				return -EINVAL;
1595 			}
1596 			info->labels = *labels;
1597 			break;
1598 		}
1599 #endif
1600 		case OVS_CT_ATTR_HELPER:
1601 			*helper = nla_data(a);
1602 			if (!memchr(*helper, '\0', nla_len(a))) {
1603 				OVS_NLERR(log, "Invalid conntrack helper");
1604 				return -EINVAL;
1605 			}
1606 			break;
1607 #if IS_ENABLED(CONFIG_NF_NAT)
1608 		case OVS_CT_ATTR_NAT: {
1609 			int err = parse_nat(a, info, log);
1610 
1611 			if (err)
1612 				return err;
1613 			break;
1614 		}
1615 #endif
1616 		case OVS_CT_ATTR_EVENTMASK:
1617 			info->have_eventmask = true;
1618 			info->eventmask = nla_get_u32(a);
1619 			break;
1620 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1621 		case OVS_CT_ATTR_TIMEOUT:
1622 			memcpy(info->timeout, nla_data(a), nla_len(a));
1623 			if (!memchr(info->timeout, '\0', nla_len(a))) {
1624 				OVS_NLERR(log, "Invalid conntrack timeout");
1625 				return -EINVAL;
1626 			}
1627 			break;
1628 #endif
1629 
1630 		default:
1631 			OVS_NLERR(log, "Unknown conntrack attr (%d)",
1632 				  type);
1633 			return -EINVAL;
1634 		}
1635 	}
1636 
1637 #ifdef CONFIG_NF_CONNTRACK_MARK
1638 	if (!info->commit && info->mark.mask) {
1639 		OVS_NLERR(log,
1640 			  "Setting conntrack mark requires 'commit' flag.");
1641 		return -EINVAL;
1642 	}
1643 #endif
1644 #ifdef CONFIG_NF_CONNTRACK_LABELS
1645 	if (!info->commit && labels_nonzero(&info->labels.mask)) {
1646 		OVS_NLERR(log,
1647 			  "Setting conntrack labels requires 'commit' flag.");
1648 		return -EINVAL;
1649 	}
1650 #endif
1651 	if (rem > 0) {
1652 		OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1653 		return -EINVAL;
1654 	}
1655 
1656 	return 0;
1657 }
1658 
1659 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1660 {
1661 	if (attr == OVS_KEY_ATTR_CT_STATE)
1662 		return true;
1663 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1664 	    attr == OVS_KEY_ATTR_CT_ZONE)
1665 		return true;
1666 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1667 	    attr == OVS_KEY_ATTR_CT_MARK)
1668 		return true;
1669 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1670 	    attr == OVS_KEY_ATTR_CT_LABELS) {
1671 		struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1672 
1673 		return ovs_net->xt_label;
1674 	}
1675 
1676 	return false;
1677 }
1678 
1679 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1680 		       const struct sw_flow_key *key,
1681 		       struct sw_flow_actions **sfa,  bool log)
1682 {
1683 	struct ovs_conntrack_info ct_info;
1684 	const char *helper = NULL;
1685 	u16 family;
1686 	int err;
1687 
1688 	family = key_to_nfproto(key);
1689 	if (family == NFPROTO_UNSPEC) {
1690 		OVS_NLERR(log, "ct family unspecified");
1691 		return -EINVAL;
1692 	}
1693 
1694 	memset(&ct_info, 0, sizeof(ct_info));
1695 	ct_info.family = family;
1696 
1697 	nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1698 			NF_CT_DEFAULT_ZONE_DIR, 0);
1699 
1700 	err = parse_ct(attr, &ct_info, &helper, log);
1701 	if (err)
1702 		return err;
1703 
1704 	/* Set up template for tracking connections in specific zones. */
1705 	ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1706 	if (!ct_info.ct) {
1707 		OVS_NLERR(log, "Failed to allocate conntrack template");
1708 		return -ENOMEM;
1709 	}
1710 
1711 	if (ct_info.timeout[0]) {
1712 		if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1713 				      ct_info.timeout))
1714 			pr_info_ratelimited("Failed to associated timeout "
1715 					    "policy `%s'\n", ct_info.timeout);
1716 		else
1717 			ct_info.nf_ct_timeout = rcu_dereference(
1718 				nf_ct_timeout_find(ct_info.ct)->timeout);
1719 
1720 	}
1721 
1722 	if (helper) {
1723 		err = ovs_ct_add_helper(&ct_info, helper, key, log);
1724 		if (err)
1725 			goto err_free_ct;
1726 	}
1727 
1728 	err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1729 				 sizeof(ct_info), log);
1730 	if (err)
1731 		goto err_free_ct;
1732 
1733 	__set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1734 	return 0;
1735 err_free_ct:
1736 	__ovs_ct_free_action(&ct_info);
1737 	return err;
1738 }
1739 
1740 #if IS_ENABLED(CONFIG_NF_NAT)
1741 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1742 			       struct sk_buff *skb)
1743 {
1744 	struct nlattr *start;
1745 
1746 	start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1747 	if (!start)
1748 		return false;
1749 
1750 	if (info->nat & OVS_CT_SRC_NAT) {
1751 		if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1752 			return false;
1753 	} else if (info->nat & OVS_CT_DST_NAT) {
1754 		if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1755 			return false;
1756 	} else {
1757 		goto out;
1758 	}
1759 
1760 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1761 		if (IS_ENABLED(CONFIG_NF_NAT) &&
1762 		    info->family == NFPROTO_IPV4) {
1763 			if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1764 					    info->range.min_addr.ip) ||
1765 			    (info->range.max_addr.ip
1766 			     != info->range.min_addr.ip &&
1767 			     (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1768 					      info->range.max_addr.ip))))
1769 				return false;
1770 		} else if (IS_ENABLED(CONFIG_IPV6) &&
1771 			   info->family == NFPROTO_IPV6) {
1772 			if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1773 					     &info->range.min_addr.in6) ||
1774 			    (memcmp(&info->range.max_addr.in6,
1775 				    &info->range.min_addr.in6,
1776 				    sizeof(info->range.max_addr.in6)) &&
1777 			     (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1778 					       &info->range.max_addr.in6))))
1779 				return false;
1780 		} else {
1781 			return false;
1782 		}
1783 	}
1784 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1785 	    (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1786 			 ntohs(info->range.min_proto.all)) ||
1787 	     (info->range.max_proto.all != info->range.min_proto.all &&
1788 	      nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1789 			  ntohs(info->range.max_proto.all)))))
1790 		return false;
1791 
1792 	if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1793 	    nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1794 		return false;
1795 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1796 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1797 		return false;
1798 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1799 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1800 		return false;
1801 out:
1802 	nla_nest_end(skb, start);
1803 
1804 	return true;
1805 }
1806 #endif
1807 
1808 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1809 			  struct sk_buff *skb)
1810 {
1811 	struct nlattr *start;
1812 
1813 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1814 	if (!start)
1815 		return -EMSGSIZE;
1816 
1817 	if (ct_info->commit && nla_put_flag(skb, ct_info->force
1818 					    ? OVS_CT_ATTR_FORCE_COMMIT
1819 					    : OVS_CT_ATTR_COMMIT))
1820 		return -EMSGSIZE;
1821 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1822 	    nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1823 		return -EMSGSIZE;
1824 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1825 	    nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1826 		    &ct_info->mark))
1827 		return -EMSGSIZE;
1828 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1829 	    labels_nonzero(&ct_info->labels.mask) &&
1830 	    nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1831 		    &ct_info->labels))
1832 		return -EMSGSIZE;
1833 	if (ct_info->helper) {
1834 		if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1835 				   ct_info->helper->name))
1836 			return -EMSGSIZE;
1837 	}
1838 	if (ct_info->have_eventmask &&
1839 	    nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1840 		return -EMSGSIZE;
1841 	if (ct_info->timeout[0]) {
1842 		if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1843 			return -EMSGSIZE;
1844 	}
1845 
1846 #if IS_ENABLED(CONFIG_NF_NAT)
1847 	if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1848 		return -EMSGSIZE;
1849 #endif
1850 	nla_nest_end(skb, start);
1851 
1852 	return 0;
1853 }
1854 
1855 void ovs_ct_free_action(const struct nlattr *a)
1856 {
1857 	struct ovs_conntrack_info *ct_info = nla_data(a);
1858 
1859 	__ovs_ct_free_action(ct_info);
1860 }
1861 
1862 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1863 {
1864 	if (ct_info->helper) {
1865 #if IS_ENABLED(CONFIG_NF_NAT)
1866 		if (ct_info->nat)
1867 			nf_nat_helper_put(ct_info->helper);
1868 #endif
1869 		nf_conntrack_helper_put(ct_info->helper);
1870 	}
1871 	if (ct_info->ct) {
1872 		if (ct_info->timeout[0])
1873 			nf_ct_destroy_timeout(ct_info->ct);
1874 		nf_ct_tmpl_free(ct_info->ct);
1875 	}
1876 }
1877 
1878 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1879 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1880 {
1881 	int i, err;
1882 
1883 	ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1884 					 GFP_KERNEL);
1885 	if (!ovs_net->ct_limit_info)
1886 		return -ENOMEM;
1887 
1888 	ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1889 	ovs_net->ct_limit_info->limits =
1890 		kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1891 			      GFP_KERNEL);
1892 	if (!ovs_net->ct_limit_info->limits) {
1893 		kfree(ovs_net->ct_limit_info);
1894 		return -ENOMEM;
1895 	}
1896 
1897 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1898 		INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1899 
1900 	ovs_net->ct_limit_info->data =
1901 		nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1902 
1903 	if (IS_ERR(ovs_net->ct_limit_info->data)) {
1904 		err = PTR_ERR(ovs_net->ct_limit_info->data);
1905 		kfree(ovs_net->ct_limit_info->limits);
1906 		kfree(ovs_net->ct_limit_info);
1907 		pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1908 		return err;
1909 	}
1910 	return 0;
1911 }
1912 
1913 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1914 {
1915 	const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1916 	int i;
1917 
1918 	nf_conncount_destroy(net, NFPROTO_INET, info->data);
1919 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1920 		struct hlist_head *head = &info->limits[i];
1921 		struct ovs_ct_limit *ct_limit;
1922 
1923 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node,
1924 					 lockdep_ovsl_is_held())
1925 			kfree_rcu(ct_limit, rcu);
1926 	}
1927 	kfree(info->limits);
1928 	kfree(info);
1929 }
1930 
1931 static struct sk_buff *
1932 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1933 			     struct ovs_header **ovs_reply_header)
1934 {
1935 	struct ovs_header *ovs_header = info->userhdr;
1936 	struct sk_buff *skb;
1937 
1938 	skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1939 	if (!skb)
1940 		return ERR_PTR(-ENOMEM);
1941 
1942 	*ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1943 					info->snd_seq,
1944 					&dp_ct_limit_genl_family, 0, cmd);
1945 
1946 	if (!*ovs_reply_header) {
1947 		nlmsg_free(skb);
1948 		return ERR_PTR(-EMSGSIZE);
1949 	}
1950 	(*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1951 
1952 	return skb;
1953 }
1954 
1955 static bool check_zone_id(int zone_id, u16 *pzone)
1956 {
1957 	if (zone_id >= 0 && zone_id <= 65535) {
1958 		*pzone = (u16)zone_id;
1959 		return true;
1960 	}
1961 	return false;
1962 }
1963 
1964 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1965 				       struct ovs_ct_limit_info *info)
1966 {
1967 	struct ovs_zone_limit *zone_limit;
1968 	int rem;
1969 	u16 zone;
1970 
1971 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1972 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1973 
1974 	while (rem >= sizeof(*zone_limit)) {
1975 		if (unlikely(zone_limit->zone_id ==
1976 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1977 			ovs_lock();
1978 			info->default_limit = zone_limit->limit;
1979 			ovs_unlock();
1980 		} else if (unlikely(!check_zone_id(
1981 				zone_limit->zone_id, &zone))) {
1982 			OVS_NLERR(true, "zone id is out of range");
1983 		} else {
1984 			struct ovs_ct_limit *ct_limit;
1985 
1986 			ct_limit = kmalloc(sizeof(*ct_limit),
1987 					   GFP_KERNEL_ACCOUNT);
1988 			if (!ct_limit)
1989 				return -ENOMEM;
1990 
1991 			ct_limit->zone = zone;
1992 			ct_limit->limit = zone_limit->limit;
1993 
1994 			ovs_lock();
1995 			ct_limit_set(info, ct_limit);
1996 			ovs_unlock();
1997 		}
1998 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1999 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2000 				NLA_ALIGN(sizeof(*zone_limit)));
2001 	}
2002 
2003 	if (rem)
2004 		OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
2005 
2006 	return 0;
2007 }
2008 
2009 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
2010 				       struct ovs_ct_limit_info *info)
2011 {
2012 	struct ovs_zone_limit *zone_limit;
2013 	int rem;
2014 	u16 zone;
2015 
2016 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
2017 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2018 
2019 	while (rem >= sizeof(*zone_limit)) {
2020 		if (unlikely(zone_limit->zone_id ==
2021 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2022 			ovs_lock();
2023 			info->default_limit = OVS_CT_LIMIT_DEFAULT;
2024 			ovs_unlock();
2025 		} else if (unlikely(!check_zone_id(
2026 				zone_limit->zone_id, &zone))) {
2027 			OVS_NLERR(true, "zone id is out of range");
2028 		} else {
2029 			ovs_lock();
2030 			ct_limit_del(info, zone);
2031 			ovs_unlock();
2032 		}
2033 		rem -= NLA_ALIGN(sizeof(*zone_limit));
2034 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2035 				NLA_ALIGN(sizeof(*zone_limit)));
2036 	}
2037 
2038 	if (rem)
2039 		OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
2040 
2041 	return 0;
2042 }
2043 
2044 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2045 					  struct sk_buff *reply)
2046 {
2047 	struct ovs_zone_limit zone_limit = {
2048 		.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE,
2049 		.limit   = info->default_limit,
2050 	};
2051 
2052 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2053 }
2054 
2055 static int __ovs_ct_limit_get_zone_limit(struct net *net,
2056 					 struct nf_conncount_data *data,
2057 					 u16 zone_id, u32 limit,
2058 					 struct sk_buff *reply)
2059 {
2060 	struct nf_conntrack_zone ct_zone;
2061 	struct ovs_zone_limit zone_limit;
2062 	u32 conncount_key = zone_id;
2063 
2064 	zone_limit.zone_id = zone_id;
2065 	zone_limit.limit = limit;
2066 	nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2067 
2068 	zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2069 					      &ct_zone);
2070 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2071 }
2072 
2073 static int ovs_ct_limit_get_zone_limit(struct net *net,
2074 				       struct nlattr *nla_zone_limit,
2075 				       struct ovs_ct_limit_info *info,
2076 				       struct sk_buff *reply)
2077 {
2078 	struct ovs_zone_limit *zone_limit;
2079 	int rem, err;
2080 	u32 limit;
2081 	u16 zone;
2082 
2083 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
2084 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2085 
2086 	while (rem >= sizeof(*zone_limit)) {
2087 		if (unlikely(zone_limit->zone_id ==
2088 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2089 			err = ovs_ct_limit_get_default_limit(info, reply);
2090 			if (err)
2091 				return err;
2092 		} else if (unlikely(!check_zone_id(zone_limit->zone_id,
2093 							&zone))) {
2094 			OVS_NLERR(true, "zone id is out of range");
2095 		} else {
2096 			rcu_read_lock();
2097 			limit = ct_limit_get(info, zone);
2098 			rcu_read_unlock();
2099 
2100 			err = __ovs_ct_limit_get_zone_limit(
2101 				net, info->data, zone, limit, reply);
2102 			if (err)
2103 				return err;
2104 		}
2105 		rem -= NLA_ALIGN(sizeof(*zone_limit));
2106 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2107 				NLA_ALIGN(sizeof(*zone_limit)));
2108 	}
2109 
2110 	if (rem)
2111 		OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2112 
2113 	return 0;
2114 }
2115 
2116 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2117 					   struct ovs_ct_limit_info *info,
2118 					   struct sk_buff *reply)
2119 {
2120 	struct ovs_ct_limit *ct_limit;
2121 	struct hlist_head *head;
2122 	int i, err = 0;
2123 
2124 	err = ovs_ct_limit_get_default_limit(info, reply);
2125 	if (err)
2126 		return err;
2127 
2128 	rcu_read_lock();
2129 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2130 		head = &info->limits[i];
2131 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2132 			err = __ovs_ct_limit_get_zone_limit(net, info->data,
2133 				ct_limit->zone, ct_limit->limit, reply);
2134 			if (err)
2135 				goto exit_err;
2136 		}
2137 	}
2138 
2139 exit_err:
2140 	rcu_read_unlock();
2141 	return err;
2142 }
2143 
2144 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2145 {
2146 	struct nlattr **a = info->attrs;
2147 	struct sk_buff *reply;
2148 	struct ovs_header *ovs_reply_header;
2149 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2150 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2151 	int err;
2152 
2153 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2154 					     &ovs_reply_header);
2155 	if (IS_ERR(reply))
2156 		return PTR_ERR(reply);
2157 
2158 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2159 		err = -EINVAL;
2160 		goto exit_err;
2161 	}
2162 
2163 	err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2164 					  ct_limit_info);
2165 	if (err)
2166 		goto exit_err;
2167 
2168 	static_branch_enable(&ovs_ct_limit_enabled);
2169 
2170 	genlmsg_end(reply, ovs_reply_header);
2171 	return genlmsg_reply(reply, info);
2172 
2173 exit_err:
2174 	nlmsg_free(reply);
2175 	return err;
2176 }
2177 
2178 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2179 {
2180 	struct nlattr **a = info->attrs;
2181 	struct sk_buff *reply;
2182 	struct ovs_header *ovs_reply_header;
2183 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2184 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2185 	int err;
2186 
2187 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2188 					     &ovs_reply_header);
2189 	if (IS_ERR(reply))
2190 		return PTR_ERR(reply);
2191 
2192 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2193 		err = -EINVAL;
2194 		goto exit_err;
2195 	}
2196 
2197 	err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2198 					  ct_limit_info);
2199 	if (err)
2200 		goto exit_err;
2201 
2202 	genlmsg_end(reply, ovs_reply_header);
2203 	return genlmsg_reply(reply, info);
2204 
2205 exit_err:
2206 	nlmsg_free(reply);
2207 	return err;
2208 }
2209 
2210 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2211 {
2212 	struct nlattr **a = info->attrs;
2213 	struct nlattr *nla_reply;
2214 	struct sk_buff *reply;
2215 	struct ovs_header *ovs_reply_header;
2216 	struct net *net = sock_net(skb->sk);
2217 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2218 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2219 	int err;
2220 
2221 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2222 					     &ovs_reply_header);
2223 	if (IS_ERR(reply))
2224 		return PTR_ERR(reply);
2225 
2226 	nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2227 	if (!nla_reply) {
2228 		err = -EMSGSIZE;
2229 		goto exit_err;
2230 	}
2231 
2232 	if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2233 		err = ovs_ct_limit_get_zone_limit(
2234 			net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2235 			reply);
2236 		if (err)
2237 			goto exit_err;
2238 	} else {
2239 		err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2240 						      reply);
2241 		if (err)
2242 			goto exit_err;
2243 	}
2244 
2245 	nla_nest_end(reply, nla_reply);
2246 	genlmsg_end(reply, ovs_reply_header);
2247 	return genlmsg_reply(reply, info);
2248 
2249 exit_err:
2250 	nlmsg_free(reply);
2251 	return err;
2252 }
2253 
2254 static const struct genl_small_ops ct_limit_genl_ops[] = {
2255 	{ .cmd = OVS_CT_LIMIT_CMD_SET,
2256 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2257 		.flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2258 					       * privilege.
2259 					       */
2260 		.doit = ovs_ct_limit_cmd_set,
2261 	},
2262 	{ .cmd = OVS_CT_LIMIT_CMD_DEL,
2263 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2264 		.flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2265 					       * privilege.
2266 					       */
2267 		.doit = ovs_ct_limit_cmd_del,
2268 	},
2269 	{ .cmd = OVS_CT_LIMIT_CMD_GET,
2270 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2271 		.flags = 0,		  /* OK for unprivileged users. */
2272 		.doit = ovs_ct_limit_cmd_get,
2273 	},
2274 };
2275 
2276 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2277 	.name = OVS_CT_LIMIT_MCGROUP,
2278 };
2279 
2280 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2281 	.hdrsize = sizeof(struct ovs_header),
2282 	.name = OVS_CT_LIMIT_FAMILY,
2283 	.version = OVS_CT_LIMIT_VERSION,
2284 	.maxattr = OVS_CT_LIMIT_ATTR_MAX,
2285 	.policy = ct_limit_policy,
2286 	.netnsok = true,
2287 	.parallel_ops = true,
2288 	.small_ops = ct_limit_genl_ops,
2289 	.n_small_ops = ARRAY_SIZE(ct_limit_genl_ops),
2290 	.resv_start_op = OVS_CT_LIMIT_CMD_GET + 1,
2291 	.mcgrps = &ovs_ct_limit_multicast_group,
2292 	.n_mcgrps = 1,
2293 	.module = THIS_MODULE,
2294 };
2295 #endif
2296 
2297 int ovs_ct_init(struct net *net)
2298 {
2299 	unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2300 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2301 
2302 	if (nf_connlabels_get(net, n_bits - 1)) {
2303 		ovs_net->xt_label = false;
2304 		OVS_NLERR(true, "Failed to set connlabel length");
2305 	} else {
2306 		ovs_net->xt_label = true;
2307 	}
2308 
2309 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2310 	return ovs_ct_limit_init(net, ovs_net);
2311 #else
2312 	return 0;
2313 #endif
2314 }
2315 
2316 void ovs_ct_exit(struct net *net)
2317 {
2318 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2319 
2320 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2321 	ovs_ct_limit_exit(net, ovs_net);
2322 #endif
2323 
2324 	if (ovs_net->xt_label)
2325 		nf_connlabels_put(net);
2326 }
2327