xref: /openbmc/linux/net/openvswitch/conntrack.c (revision e8614313)
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 "datapath.h"
29 #include "conntrack.h"
30 #include "flow.h"
31 #include "flow_netlink.h"
32 
33 struct ovs_ct_len_tbl {
34 	int maxlen;
35 	int minlen;
36 };
37 
38 /* Metadata mark for masked write to conntrack mark */
39 struct md_mark {
40 	u32 value;
41 	u32 mask;
42 };
43 
44 /* Metadata label for masked write to conntrack label. */
45 struct md_labels {
46 	struct ovs_key_ct_labels value;
47 	struct ovs_key_ct_labels mask;
48 };
49 
50 enum ovs_ct_nat {
51 	OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
52 	OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
53 	OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
54 };
55 
56 /* Conntrack action context for execution. */
57 struct ovs_conntrack_info {
58 	struct nf_conntrack_helper *helper;
59 	struct nf_conntrack_zone zone;
60 	struct nf_conn *ct;
61 	u8 commit : 1;
62 	u8 nat : 3;                 /* enum ovs_ct_nat */
63 	u8 force : 1;
64 	u8 have_eventmask : 1;
65 	u16 family;
66 	u32 eventmask;              /* Mask of 1 << IPCT_*. */
67 	struct md_mark mark;
68 	struct md_labels labels;
69 	char timeout[CTNL_TIMEOUT_NAME_MAX];
70 	struct nf_ct_timeout *nf_ct_timeout;
71 #if IS_ENABLED(CONFIG_NF_NAT)
72 	struct nf_nat_range2 range;  /* Only present for SRC NAT and DST NAT. */
73 #endif
74 };
75 
76 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
77 #define OVS_CT_LIMIT_UNLIMITED	0
78 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
79 #define CT_LIMIT_HASH_BUCKETS 512
80 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
81 
82 struct ovs_ct_limit {
83 	/* Elements in ovs_ct_limit_info->limits hash table */
84 	struct hlist_node hlist_node;
85 	struct rcu_head rcu;
86 	u16 zone;
87 	u32 limit;
88 };
89 
90 struct ovs_ct_limit_info {
91 	u32 default_limit;
92 	struct hlist_head *limits;
93 	struct nf_conncount_data *data;
94 };
95 
96 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
97 	[OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
98 };
99 #endif
100 
101 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
102 
103 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
104 
105 static u16 key_to_nfproto(const struct sw_flow_key *key)
106 {
107 	switch (ntohs(key->eth.type)) {
108 	case ETH_P_IP:
109 		return NFPROTO_IPV4;
110 	case ETH_P_IPV6:
111 		return NFPROTO_IPV6;
112 	default:
113 		return NFPROTO_UNSPEC;
114 	}
115 }
116 
117 /* Map SKB connection state into the values used by flow definition. */
118 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
119 {
120 	u8 ct_state = OVS_CS_F_TRACKED;
121 
122 	switch (ctinfo) {
123 	case IP_CT_ESTABLISHED_REPLY:
124 	case IP_CT_RELATED_REPLY:
125 		ct_state |= OVS_CS_F_REPLY_DIR;
126 		break;
127 	default:
128 		break;
129 	}
130 
131 	switch (ctinfo) {
132 	case IP_CT_ESTABLISHED:
133 	case IP_CT_ESTABLISHED_REPLY:
134 		ct_state |= OVS_CS_F_ESTABLISHED;
135 		break;
136 	case IP_CT_RELATED:
137 	case IP_CT_RELATED_REPLY:
138 		ct_state |= OVS_CS_F_RELATED;
139 		break;
140 	case IP_CT_NEW:
141 		ct_state |= OVS_CS_F_NEW;
142 		break;
143 	default:
144 		break;
145 	}
146 
147 	return ct_state;
148 }
149 
150 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
151 {
152 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
153 	return ct ? ct->mark : 0;
154 #else
155 	return 0;
156 #endif
157 }
158 
159 /* Guard against conntrack labels max size shrinking below 128 bits. */
160 #if NF_CT_LABELS_MAX_SIZE < 16
161 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
162 #endif
163 
164 static void ovs_ct_get_labels(const struct nf_conn *ct,
165 			      struct ovs_key_ct_labels *labels)
166 {
167 	struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
168 
169 	if (cl)
170 		memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
171 	else
172 		memset(labels, 0, OVS_CT_LABELS_LEN);
173 }
174 
175 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
176 					const struct nf_conntrack_tuple *orig,
177 					u8 icmp_proto)
178 {
179 	key->ct_orig_proto = orig->dst.protonum;
180 	if (orig->dst.protonum == icmp_proto) {
181 		key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
182 		key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
183 	} else {
184 		key->ct.orig_tp.src = orig->src.u.all;
185 		key->ct.orig_tp.dst = orig->dst.u.all;
186 	}
187 }
188 
189 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
190 				const struct nf_conntrack_zone *zone,
191 				const struct nf_conn *ct)
192 {
193 	key->ct_state = state;
194 	key->ct_zone = zone->id;
195 	key->ct.mark = ovs_ct_get_mark(ct);
196 	ovs_ct_get_labels(ct, &key->ct.labels);
197 
198 	if (ct) {
199 		const struct nf_conntrack_tuple *orig;
200 
201 		/* Use the master if we have one. */
202 		if (ct->master)
203 			ct = ct->master;
204 		orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
205 
206 		/* IP version must match with the master connection. */
207 		if (key->eth.type == htons(ETH_P_IP) &&
208 		    nf_ct_l3num(ct) == NFPROTO_IPV4) {
209 			key->ipv4.ct_orig.src = orig->src.u3.ip;
210 			key->ipv4.ct_orig.dst = orig->dst.u3.ip;
211 			__ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
212 			return;
213 		} else if (key->eth.type == htons(ETH_P_IPV6) &&
214 			   !sw_flow_key_is_nd(key) &&
215 			   nf_ct_l3num(ct) == NFPROTO_IPV6) {
216 			key->ipv6.ct_orig.src = orig->src.u3.in6;
217 			key->ipv6.ct_orig.dst = orig->dst.u3.in6;
218 			__ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
219 			return;
220 		}
221 	}
222 	/* Clear 'ct_orig_proto' to mark the non-existence of conntrack
223 	 * original direction key fields.
224 	 */
225 	key->ct_orig_proto = 0;
226 }
227 
228 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
229  * previously sent the packet to conntrack via the ct action.  If
230  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
231  * initialized from the connection status.
232  */
233 static void ovs_ct_update_key(const struct sk_buff *skb,
234 			      const struct ovs_conntrack_info *info,
235 			      struct sw_flow_key *key, bool post_ct,
236 			      bool keep_nat_flags)
237 {
238 	const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
239 	enum ip_conntrack_info ctinfo;
240 	struct nf_conn *ct;
241 	u8 state = 0;
242 
243 	ct = nf_ct_get(skb, &ctinfo);
244 	if (ct) {
245 		state = ovs_ct_get_state(ctinfo);
246 		/* All unconfirmed entries are NEW connections. */
247 		if (!nf_ct_is_confirmed(ct))
248 			state |= OVS_CS_F_NEW;
249 		/* OVS persists the related flag for the duration of the
250 		 * connection.
251 		 */
252 		if (ct->master)
253 			state |= OVS_CS_F_RELATED;
254 		if (keep_nat_flags) {
255 			state |= key->ct_state & OVS_CS_F_NAT_MASK;
256 		} else {
257 			if (ct->status & IPS_SRC_NAT)
258 				state |= OVS_CS_F_SRC_NAT;
259 			if (ct->status & IPS_DST_NAT)
260 				state |= OVS_CS_F_DST_NAT;
261 		}
262 		zone = nf_ct_zone(ct);
263 	} else if (post_ct) {
264 		state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
265 		if (info)
266 			zone = &info->zone;
267 	}
268 	__ovs_ct_update_key(key, state, zone, ct);
269 }
270 
271 /* This is called to initialize CT key fields possibly coming in from the local
272  * stack.
273  */
274 void ovs_ct_fill_key(const struct sk_buff *skb,
275 		     struct sw_flow_key *key,
276 		     bool post_ct)
277 {
278 	ovs_ct_update_key(skb, NULL, key, post_ct, false);
279 }
280 
281 int ovs_ct_put_key(const struct sw_flow_key *swkey,
282 		   const struct sw_flow_key *output, struct sk_buff *skb)
283 {
284 	if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
285 		return -EMSGSIZE;
286 
287 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
288 	    nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
289 		return -EMSGSIZE;
290 
291 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
292 	    nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
293 		return -EMSGSIZE;
294 
295 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
296 	    nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
297 		    &output->ct.labels))
298 		return -EMSGSIZE;
299 
300 	if (swkey->ct_orig_proto) {
301 		if (swkey->eth.type == htons(ETH_P_IP)) {
302 			struct ovs_key_ct_tuple_ipv4 orig;
303 
304 			memset(&orig, 0, sizeof(orig));
305 			orig.ipv4_src = output->ipv4.ct_orig.src;
306 			orig.ipv4_dst = output->ipv4.ct_orig.dst;
307 			orig.src_port = output->ct.orig_tp.src;
308 			orig.dst_port = output->ct.orig_tp.dst;
309 			orig.ipv4_proto = output->ct_orig_proto;
310 
311 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
312 				    sizeof(orig), &orig))
313 				return -EMSGSIZE;
314 		} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
315 			struct ovs_key_ct_tuple_ipv6 orig;
316 
317 			memset(&orig, 0, sizeof(orig));
318 			memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32,
319 			       sizeof(orig.ipv6_src));
320 			memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32,
321 			       sizeof(orig.ipv6_dst));
322 			orig.src_port = output->ct.orig_tp.src;
323 			orig.dst_port = output->ct.orig_tp.dst;
324 			orig.ipv6_proto = output->ct_orig_proto;
325 
326 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
327 				    sizeof(orig), &orig))
328 				return -EMSGSIZE;
329 		}
330 	}
331 
332 	return 0;
333 }
334 
335 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
336 			   u32 ct_mark, u32 mask)
337 {
338 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
339 	u32 new_mark;
340 
341 	new_mark = ct_mark | (ct->mark & ~(mask));
342 	if (ct->mark != new_mark) {
343 		ct->mark = new_mark;
344 		if (nf_ct_is_confirmed(ct))
345 			nf_conntrack_event_cache(IPCT_MARK, ct);
346 		key->ct.mark = new_mark;
347 	}
348 
349 	return 0;
350 #else
351 	return -ENOTSUPP;
352 #endif
353 }
354 
355 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
356 {
357 	struct nf_conn_labels *cl;
358 
359 	cl = nf_ct_labels_find(ct);
360 	if (!cl) {
361 		nf_ct_labels_ext_add(ct);
362 		cl = nf_ct_labels_find(ct);
363 	}
364 
365 	return cl;
366 }
367 
368 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
369  * since the new connection is not yet confirmed, and thus no-one else has
370  * access to it's labels, we simply write them over.
371  */
372 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
373 			      const struct ovs_key_ct_labels *labels,
374 			      const struct ovs_key_ct_labels *mask)
375 {
376 	struct nf_conn_labels *cl, *master_cl;
377 	bool have_mask = labels_nonzero(mask);
378 
379 	/* Inherit master's labels to the related connection? */
380 	master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
381 
382 	if (!master_cl && !have_mask)
383 		return 0;   /* Nothing to do. */
384 
385 	cl = ovs_ct_get_conn_labels(ct);
386 	if (!cl)
387 		return -ENOSPC;
388 
389 	/* Inherit the master's labels, if any. */
390 	if (master_cl)
391 		*cl = *master_cl;
392 
393 	if (have_mask) {
394 		u32 *dst = (u32 *)cl->bits;
395 		int i;
396 
397 		for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
398 			dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
399 				(labels->ct_labels_32[i]
400 				 & mask->ct_labels_32[i]);
401 	}
402 
403 	/* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
404 	 * IPCT_LABEL bit is set in the event cache.
405 	 */
406 	nf_conntrack_event_cache(IPCT_LABEL, ct);
407 
408 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
409 
410 	return 0;
411 }
412 
413 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
414 			     const struct ovs_key_ct_labels *labels,
415 			     const struct ovs_key_ct_labels *mask)
416 {
417 	struct nf_conn_labels *cl;
418 	int err;
419 
420 	cl = ovs_ct_get_conn_labels(ct);
421 	if (!cl)
422 		return -ENOSPC;
423 
424 	err = nf_connlabels_replace(ct, labels->ct_labels_32,
425 				    mask->ct_labels_32,
426 				    OVS_CT_LABELS_LEN_32);
427 	if (err)
428 		return err;
429 
430 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
431 
432 	return 0;
433 }
434 
435 /* 'skb' should already be pulled to nh_ofs. */
436 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
437 {
438 	const struct nf_conntrack_helper *helper;
439 	const struct nf_conn_help *help;
440 	enum ip_conntrack_info ctinfo;
441 	unsigned int protoff;
442 	struct nf_conn *ct;
443 	int err;
444 
445 	ct = nf_ct_get(skb, &ctinfo);
446 	if (!ct || ctinfo == IP_CT_RELATED_REPLY)
447 		return NF_ACCEPT;
448 
449 	help = nfct_help(ct);
450 	if (!help)
451 		return NF_ACCEPT;
452 
453 	helper = rcu_dereference(help->helper);
454 	if (!helper)
455 		return NF_ACCEPT;
456 
457 	switch (proto) {
458 	case NFPROTO_IPV4:
459 		protoff = ip_hdrlen(skb);
460 		break;
461 	case NFPROTO_IPV6: {
462 		u8 nexthdr = ipv6_hdr(skb)->nexthdr;
463 		__be16 frag_off;
464 		int ofs;
465 
466 		ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
467 				       &frag_off);
468 		if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
469 			pr_debug("proto header not found\n");
470 			return NF_ACCEPT;
471 		}
472 		protoff = ofs;
473 		break;
474 	}
475 	default:
476 		WARN_ONCE(1, "helper invoked on non-IP family!");
477 		return NF_DROP;
478 	}
479 
480 	err = helper->help(skb, protoff, ct, ctinfo);
481 	if (err != NF_ACCEPT)
482 		return err;
483 
484 	/* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
485 	 * FTP with NAT) adusting the TCP payload size when mangling IP
486 	 * addresses and/or port numbers in the text-based control connection.
487 	 */
488 	if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
489 	    !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
490 		return NF_DROP;
491 	return NF_ACCEPT;
492 }
493 
494 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
495  * value if 'skb' is freed.
496  */
497 static int handle_fragments(struct net *net, struct sw_flow_key *key,
498 			    u16 zone, struct sk_buff *skb)
499 {
500 	struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
501 	int err;
502 
503 	if (key->eth.type == htons(ETH_P_IP)) {
504 		enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
505 
506 		memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
507 		err = ip_defrag(net, skb, user);
508 		if (err)
509 			return err;
510 
511 		ovs_cb.mru = IPCB(skb)->frag_max_size;
512 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
513 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
514 		enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
515 
516 		memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
517 		err = nf_ct_frag6_gather(net, skb, user);
518 		if (err) {
519 			if (err != -EINPROGRESS)
520 				kfree_skb(skb);
521 			return err;
522 		}
523 
524 		key->ip.proto = ipv6_hdr(skb)->nexthdr;
525 		ovs_cb.mru = IP6CB(skb)->frag_max_size;
526 #endif
527 	} else {
528 		kfree_skb(skb);
529 		return -EPFNOSUPPORT;
530 	}
531 
532 	/* The key extracted from the fragment that completed this datagram
533 	 * likely didn't have an L4 header, so regenerate it.
534 	 */
535 	ovs_flow_key_update_l3l4(skb, key);
536 
537 	key->ip.frag = OVS_FRAG_TYPE_NONE;
538 	skb_clear_hash(skb);
539 	skb->ignore_df = 1;
540 	*OVS_CB(skb) = ovs_cb;
541 
542 	return 0;
543 }
544 
545 static struct nf_conntrack_expect *
546 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
547 		   u16 proto, const struct sk_buff *skb)
548 {
549 	struct nf_conntrack_tuple tuple;
550 	struct nf_conntrack_expect *exp;
551 
552 	if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
553 		return NULL;
554 
555 	exp = __nf_ct_expect_find(net, zone, &tuple);
556 	if (exp) {
557 		struct nf_conntrack_tuple_hash *h;
558 
559 		/* Delete existing conntrack entry, if it clashes with the
560 		 * expectation.  This can happen since conntrack ALGs do not
561 		 * check for clashes between (new) expectations and existing
562 		 * conntrack entries.  nf_conntrack_in() will check the
563 		 * expectations only if a conntrack entry can not be found,
564 		 * which can lead to OVS finding the expectation (here) in the
565 		 * init direction, but which will not be removed by the
566 		 * nf_conntrack_in() call, if a matching conntrack entry is
567 		 * found instead.  In this case all init direction packets
568 		 * would be reported as new related packets, while reply
569 		 * direction packets would be reported as un-related
570 		 * established packets.
571 		 */
572 		h = nf_conntrack_find_get(net, zone, &tuple);
573 		if (h) {
574 			struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
575 
576 			nf_ct_delete(ct, 0, 0);
577 			nf_conntrack_put(&ct->ct_general);
578 		}
579 	}
580 
581 	return exp;
582 }
583 
584 /* This replicates logic from nf_conntrack_core.c that is not exported. */
585 static enum ip_conntrack_info
586 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
587 {
588 	const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
589 
590 	if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
591 		return IP_CT_ESTABLISHED_REPLY;
592 	/* Once we've had two way comms, always ESTABLISHED. */
593 	if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
594 		return IP_CT_ESTABLISHED;
595 	if (test_bit(IPS_EXPECTED_BIT, &ct->status))
596 		return IP_CT_RELATED;
597 	return IP_CT_NEW;
598 }
599 
600 /* Find an existing connection which this packet belongs to without
601  * re-attributing statistics or modifying the connection state.  This allows an
602  * skb->_nfct lost due to an upcall to be recovered during actions execution.
603  *
604  * Must be called with rcu_read_lock.
605  *
606  * On success, populates skb->_nfct and returns the connection.  Returns NULL
607  * if there is no existing entry.
608  */
609 static struct nf_conn *
610 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
611 		     u8 l3num, struct sk_buff *skb, bool natted)
612 {
613 	struct nf_conntrack_tuple tuple;
614 	struct nf_conntrack_tuple_hash *h;
615 	struct nf_conn *ct;
616 
617 	if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
618 			       net, &tuple)) {
619 		pr_debug("ovs_ct_find_existing: Can't get tuple\n");
620 		return NULL;
621 	}
622 
623 	/* Must invert the tuple if skb has been transformed by NAT. */
624 	if (natted) {
625 		struct nf_conntrack_tuple inverse;
626 
627 		if (!nf_ct_invert_tuple(&inverse, &tuple)) {
628 			pr_debug("ovs_ct_find_existing: Inversion failed!\n");
629 			return NULL;
630 		}
631 		tuple = inverse;
632 	}
633 
634 	/* look for tuple match */
635 	h = nf_conntrack_find_get(net, zone, &tuple);
636 	if (!h)
637 		return NULL;   /* Not found. */
638 
639 	ct = nf_ct_tuplehash_to_ctrack(h);
640 
641 	/* Inverted packet tuple matches the reverse direction conntrack tuple,
642 	 * select the other tuplehash to get the right 'ctinfo' bits for this
643 	 * packet.
644 	 */
645 	if (natted)
646 		h = &ct->tuplehash[!h->tuple.dst.dir];
647 
648 	nf_ct_set(skb, ct, ovs_ct_get_info(h));
649 	return ct;
650 }
651 
652 static
653 struct nf_conn *ovs_ct_executed(struct net *net,
654 				const struct sw_flow_key *key,
655 				const struct ovs_conntrack_info *info,
656 				struct sk_buff *skb,
657 				bool *ct_executed)
658 {
659 	struct nf_conn *ct = NULL;
660 
661 	/* If no ct, check if we have evidence that an existing conntrack entry
662 	 * might be found for this skb.  This happens when we lose a skb->_nfct
663 	 * due to an upcall, or if the direction is being forced.  If the
664 	 * connection was not confirmed, it is not cached and needs to be run
665 	 * through conntrack again.
666 	 */
667 	*ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
668 		       !(key->ct_state & OVS_CS_F_INVALID) &&
669 		       (key->ct_zone == info->zone.id);
670 
671 	if (*ct_executed || (!key->ct_state && info->force)) {
672 		ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
673 					  !!(key->ct_state &
674 					  OVS_CS_F_NAT_MASK));
675 	}
676 
677 	return ct;
678 }
679 
680 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
681 static bool skb_nfct_cached(struct net *net,
682 			    const struct sw_flow_key *key,
683 			    const struct ovs_conntrack_info *info,
684 			    struct sk_buff *skb)
685 {
686 	enum ip_conntrack_info ctinfo;
687 	struct nf_conn *ct;
688 	bool ct_executed = true;
689 
690 	ct = nf_ct_get(skb, &ctinfo);
691 	if (!ct)
692 		ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
693 
694 	if (ct)
695 		nf_ct_get(skb, &ctinfo);
696 	else
697 		return false;
698 
699 	if (!net_eq(net, read_pnet(&ct->ct_net)))
700 		return false;
701 	if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
702 		return false;
703 	if (info->helper) {
704 		struct nf_conn_help *help;
705 
706 		help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
707 		if (help && rcu_access_pointer(help->helper) != info->helper)
708 			return false;
709 	}
710 	if (info->nf_ct_timeout) {
711 		struct nf_conn_timeout *timeout_ext;
712 
713 		timeout_ext = nf_ct_timeout_find(ct);
714 		if (!timeout_ext || info->nf_ct_timeout !=
715 		    rcu_dereference(timeout_ext->timeout))
716 			return false;
717 	}
718 	/* Force conntrack entry direction to the current packet? */
719 	if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
720 		/* Delete the conntrack entry if confirmed, else just release
721 		 * the reference.
722 		 */
723 		if (nf_ct_is_confirmed(ct))
724 			nf_ct_delete(ct, 0, 0);
725 
726 		nf_conntrack_put(&ct->ct_general);
727 		nf_ct_set(skb, NULL, 0);
728 		return false;
729 	}
730 
731 	return ct_executed;
732 }
733 
734 #if IS_ENABLED(CONFIG_NF_NAT)
735 /* Modelled after nf_nat_ipv[46]_fn().
736  * range is only used for new, uninitialized NAT state.
737  * Returns either NF_ACCEPT or NF_DROP.
738  */
739 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
740 			      enum ip_conntrack_info ctinfo,
741 			      const struct nf_nat_range2 *range,
742 			      enum nf_nat_manip_type maniptype)
743 {
744 	int hooknum, nh_off, err = NF_ACCEPT;
745 
746 	nh_off = skb_network_offset(skb);
747 	skb_pull_rcsum(skb, nh_off);
748 
749 	/* See HOOK2MANIP(). */
750 	if (maniptype == NF_NAT_MANIP_SRC)
751 		hooknum = NF_INET_LOCAL_IN; /* Source NAT */
752 	else
753 		hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
754 
755 	switch (ctinfo) {
756 	case IP_CT_RELATED:
757 	case IP_CT_RELATED_REPLY:
758 		if (IS_ENABLED(CONFIG_NF_NAT) &&
759 		    skb->protocol == htons(ETH_P_IP) &&
760 		    ip_hdr(skb)->protocol == IPPROTO_ICMP) {
761 			if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
762 							   hooknum))
763 				err = NF_DROP;
764 			goto push;
765 		} else if (IS_ENABLED(CONFIG_IPV6) &&
766 			   skb->protocol == htons(ETH_P_IPV6)) {
767 			__be16 frag_off;
768 			u8 nexthdr = ipv6_hdr(skb)->nexthdr;
769 			int hdrlen = ipv6_skip_exthdr(skb,
770 						      sizeof(struct ipv6hdr),
771 						      &nexthdr, &frag_off);
772 
773 			if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
774 				if (!nf_nat_icmpv6_reply_translation(skb, ct,
775 								     ctinfo,
776 								     hooknum,
777 								     hdrlen))
778 					err = NF_DROP;
779 				goto push;
780 			}
781 		}
782 		/* Non-ICMP, fall thru to initialize if needed. */
783 		fallthrough;
784 	case IP_CT_NEW:
785 		/* Seen it before?  This can happen for loopback, retrans,
786 		 * or local packets.
787 		 */
788 		if (!nf_nat_initialized(ct, maniptype)) {
789 			/* Initialize according to the NAT action. */
790 			err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
791 				/* Action is set up to establish a new
792 				 * mapping.
793 				 */
794 				? nf_nat_setup_info(ct, range, maniptype)
795 				: nf_nat_alloc_null_binding(ct, hooknum);
796 			if (err != NF_ACCEPT)
797 				goto push;
798 		}
799 		break;
800 
801 	case IP_CT_ESTABLISHED:
802 	case IP_CT_ESTABLISHED_REPLY:
803 		break;
804 
805 	default:
806 		err = NF_DROP;
807 		goto push;
808 	}
809 
810 	err = nf_nat_packet(ct, ctinfo, hooknum, skb);
811 push:
812 	skb_push(skb, nh_off);
813 	skb_postpush_rcsum(skb, skb->data, nh_off);
814 
815 	return err;
816 }
817 
818 static void ovs_nat_update_key(struct sw_flow_key *key,
819 			       const struct sk_buff *skb,
820 			       enum nf_nat_manip_type maniptype)
821 {
822 	if (maniptype == NF_NAT_MANIP_SRC) {
823 		__be16 src;
824 
825 		key->ct_state |= OVS_CS_F_SRC_NAT;
826 		if (key->eth.type == htons(ETH_P_IP))
827 			key->ipv4.addr.src = ip_hdr(skb)->saddr;
828 		else if (key->eth.type == htons(ETH_P_IPV6))
829 			memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
830 			       sizeof(key->ipv6.addr.src));
831 		else
832 			return;
833 
834 		if (key->ip.proto == IPPROTO_UDP)
835 			src = udp_hdr(skb)->source;
836 		else if (key->ip.proto == IPPROTO_TCP)
837 			src = tcp_hdr(skb)->source;
838 		else if (key->ip.proto == IPPROTO_SCTP)
839 			src = sctp_hdr(skb)->source;
840 		else
841 			return;
842 
843 		key->tp.src = src;
844 	} else {
845 		__be16 dst;
846 
847 		key->ct_state |= OVS_CS_F_DST_NAT;
848 		if (key->eth.type == htons(ETH_P_IP))
849 			key->ipv4.addr.dst = ip_hdr(skb)->daddr;
850 		else if (key->eth.type == htons(ETH_P_IPV6))
851 			memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
852 			       sizeof(key->ipv6.addr.dst));
853 		else
854 			return;
855 
856 		if (key->ip.proto == IPPROTO_UDP)
857 			dst = udp_hdr(skb)->dest;
858 		else if (key->ip.proto == IPPROTO_TCP)
859 			dst = tcp_hdr(skb)->dest;
860 		else if (key->ip.proto == IPPROTO_SCTP)
861 			dst = sctp_hdr(skb)->dest;
862 		else
863 			return;
864 
865 		key->tp.dst = dst;
866 	}
867 }
868 
869 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
870 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
871 		      const struct ovs_conntrack_info *info,
872 		      struct sk_buff *skb, struct nf_conn *ct,
873 		      enum ip_conntrack_info ctinfo)
874 {
875 	enum nf_nat_manip_type maniptype;
876 	int err;
877 
878 	/* Add NAT extension if not confirmed yet. */
879 	if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
880 		return NF_ACCEPT;   /* Can't NAT. */
881 
882 	/* Determine NAT type.
883 	 * Check if the NAT type can be deduced from the tracked connection.
884 	 * Make sure new expected connections (IP_CT_RELATED) are NATted only
885 	 * when committing.
886 	 */
887 	if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
888 	    ct->status & IPS_NAT_MASK &&
889 	    (ctinfo != IP_CT_RELATED || info->commit)) {
890 		/* NAT an established or related connection like before. */
891 		if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
892 			/* This is the REPLY direction for a connection
893 			 * for which NAT was applied in the forward
894 			 * direction.  Do the reverse NAT.
895 			 */
896 			maniptype = ct->status & IPS_SRC_NAT
897 				? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
898 		else
899 			maniptype = ct->status & IPS_SRC_NAT
900 				? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
901 	} else if (info->nat & OVS_CT_SRC_NAT) {
902 		maniptype = NF_NAT_MANIP_SRC;
903 	} else if (info->nat & OVS_CT_DST_NAT) {
904 		maniptype = NF_NAT_MANIP_DST;
905 	} else {
906 		return NF_ACCEPT; /* Connection is not NATed. */
907 	}
908 	err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
909 
910 	if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) {
911 		if (ct->status & IPS_SRC_NAT) {
912 			if (maniptype == NF_NAT_MANIP_SRC)
913 				maniptype = NF_NAT_MANIP_DST;
914 			else
915 				maniptype = NF_NAT_MANIP_SRC;
916 
917 			err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range,
918 						 maniptype);
919 		} else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
920 			err = ovs_ct_nat_execute(skb, ct, ctinfo, NULL,
921 						 NF_NAT_MANIP_SRC);
922 		}
923 	}
924 
925 	/* Mark NAT done if successful and update the flow key. */
926 	if (err == NF_ACCEPT)
927 		ovs_nat_update_key(key, skb, maniptype);
928 
929 	return err;
930 }
931 #else /* !CONFIG_NF_NAT */
932 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
933 		      const struct ovs_conntrack_info *info,
934 		      struct sk_buff *skb, struct nf_conn *ct,
935 		      enum ip_conntrack_info ctinfo)
936 {
937 	return NF_ACCEPT;
938 }
939 #endif
940 
941 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
942  * not done already.  Update key with new CT state after passing the packet
943  * through conntrack.
944  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
945  * set to NULL and 0 will be returned.
946  */
947 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
948 			   const struct ovs_conntrack_info *info,
949 			   struct sk_buff *skb)
950 {
951 	/* If we are recirculating packets to match on conntrack fields and
952 	 * committing with a separate conntrack action,  then we don't need to
953 	 * actually run the packet through conntrack twice unless it's for a
954 	 * different zone.
955 	 */
956 	bool cached = skb_nfct_cached(net, key, info, skb);
957 	enum ip_conntrack_info ctinfo;
958 	struct nf_conn *ct;
959 
960 	if (!cached) {
961 		struct nf_hook_state state = {
962 			.hook = NF_INET_PRE_ROUTING,
963 			.pf = info->family,
964 			.net = net,
965 		};
966 		struct nf_conn *tmpl = info->ct;
967 		int err;
968 
969 		/* Associate skb with specified zone. */
970 		if (tmpl) {
971 			if (skb_nfct(skb))
972 				nf_conntrack_put(skb_nfct(skb));
973 			nf_conntrack_get(&tmpl->ct_general);
974 			nf_ct_set(skb, tmpl, IP_CT_NEW);
975 		}
976 
977 		err = nf_conntrack_in(skb, &state);
978 		if (err != NF_ACCEPT)
979 			return -ENOENT;
980 
981 		/* Clear CT state NAT flags to mark that we have not yet done
982 		 * NAT after the nf_conntrack_in() call.  We can actually clear
983 		 * the whole state, as it will be re-initialized below.
984 		 */
985 		key->ct_state = 0;
986 
987 		/* Update the key, but keep the NAT flags. */
988 		ovs_ct_update_key(skb, info, key, true, true);
989 	}
990 
991 	ct = nf_ct_get(skb, &ctinfo);
992 	if (ct) {
993 		bool add_helper = false;
994 
995 		/* Packets starting a new connection must be NATted before the
996 		 * helper, so that the helper knows about the NAT.  We enforce
997 		 * this by delaying both NAT and helper calls for unconfirmed
998 		 * connections until the committing CT action.  For later
999 		 * packets NAT and Helper may be called in either order.
1000 		 *
1001 		 * NAT will be done only if the CT action has NAT, and only
1002 		 * once per packet (per zone), as guarded by the NAT bits in
1003 		 * the key->ct_state.
1004 		 */
1005 		if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
1006 		    (nf_ct_is_confirmed(ct) || info->commit) &&
1007 		    ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
1008 			return -EINVAL;
1009 		}
1010 
1011 		/* Userspace may decide to perform a ct lookup without a helper
1012 		 * specified followed by a (recirculate and) commit with one,
1013 		 * or attach a helper in a later commit.  Therefore, for
1014 		 * connections which we will commit, we may need to attach
1015 		 * the helper here.
1016 		 */
1017 		if (info->commit && info->helper && !nfct_help(ct)) {
1018 			int err = __nf_ct_try_assign_helper(ct, info->ct,
1019 							    GFP_ATOMIC);
1020 			if (err)
1021 				return err;
1022 			add_helper = true;
1023 
1024 			/* helper installed, add seqadj if NAT is required */
1025 			if (info->nat && !nfct_seqadj(ct)) {
1026 				if (!nfct_seqadj_ext_add(ct))
1027 					return -EINVAL;
1028 			}
1029 		}
1030 
1031 		/* Call the helper only if:
1032 		 * - nf_conntrack_in() was executed above ("!cached") or a
1033 		 *   helper was just attached ("add_helper") for a confirmed
1034 		 *   connection, or
1035 		 * - When committing an unconfirmed connection.
1036 		 */
1037 		if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
1038 					      info->commit) &&
1039 		    ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1040 			return -EINVAL;
1041 		}
1042 
1043 		if (nf_ct_protonum(ct) == IPPROTO_TCP &&
1044 		    nf_ct_is_confirmed(ct) && nf_conntrack_tcp_established(ct)) {
1045 			/* Be liberal for tcp packets so that out-of-window
1046 			 * packets are not marked invalid.
1047 			 */
1048 			nf_ct_set_tcp_be_liberal(ct);
1049 		}
1050 	}
1051 
1052 	return 0;
1053 }
1054 
1055 /* Lookup connection and read fields into key. */
1056 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1057 			 const struct ovs_conntrack_info *info,
1058 			 struct sk_buff *skb)
1059 {
1060 	struct nf_conntrack_expect *exp;
1061 
1062 	/* If we pass an expected packet through nf_conntrack_in() the
1063 	 * expectation is typically removed, but the packet could still be
1064 	 * lost in upcall processing.  To prevent this from happening we
1065 	 * perform an explicit expectation lookup.  Expected connections are
1066 	 * always new, and will be passed through conntrack only when they are
1067 	 * committed, as it is OK to remove the expectation at that time.
1068 	 */
1069 	exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1070 	if (exp) {
1071 		u8 state;
1072 
1073 		/* NOTE: New connections are NATted and Helped only when
1074 		 * committed, so we are not calling into NAT here.
1075 		 */
1076 		state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1077 		__ovs_ct_update_key(key, state, &info->zone, exp->master);
1078 	} else {
1079 		struct nf_conn *ct;
1080 		int err;
1081 
1082 		err = __ovs_ct_lookup(net, key, info, skb);
1083 		if (err)
1084 			return err;
1085 
1086 		ct = (struct nf_conn *)skb_nfct(skb);
1087 		if (ct)
1088 			nf_ct_deliver_cached_events(ct);
1089 	}
1090 
1091 	return 0;
1092 }
1093 
1094 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1095 {
1096 	size_t i;
1097 
1098 	for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1099 		if (labels->ct_labels_32[i])
1100 			return true;
1101 
1102 	return false;
1103 }
1104 
1105 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1106 static struct hlist_head *ct_limit_hash_bucket(
1107 	const struct ovs_ct_limit_info *info, u16 zone)
1108 {
1109 	return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1110 }
1111 
1112 /* Call with ovs_mutex */
1113 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1114 			 struct ovs_ct_limit *new_ct_limit)
1115 {
1116 	struct ovs_ct_limit *ct_limit;
1117 	struct hlist_head *head;
1118 
1119 	head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1120 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1121 		if (ct_limit->zone == new_ct_limit->zone) {
1122 			hlist_replace_rcu(&ct_limit->hlist_node,
1123 					  &new_ct_limit->hlist_node);
1124 			kfree_rcu(ct_limit, rcu);
1125 			return;
1126 		}
1127 	}
1128 
1129 	hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1130 }
1131 
1132 /* Call with ovs_mutex */
1133 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1134 {
1135 	struct ovs_ct_limit *ct_limit;
1136 	struct hlist_head *head;
1137 	struct hlist_node *n;
1138 
1139 	head = ct_limit_hash_bucket(info, zone);
1140 	hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1141 		if (ct_limit->zone == zone) {
1142 			hlist_del_rcu(&ct_limit->hlist_node);
1143 			kfree_rcu(ct_limit, rcu);
1144 			return;
1145 		}
1146 	}
1147 }
1148 
1149 /* Call with RCU read lock */
1150 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1151 {
1152 	struct ovs_ct_limit *ct_limit;
1153 	struct hlist_head *head;
1154 
1155 	head = ct_limit_hash_bucket(info, zone);
1156 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1157 		if (ct_limit->zone == zone)
1158 			return ct_limit->limit;
1159 	}
1160 
1161 	return info->default_limit;
1162 }
1163 
1164 static int ovs_ct_check_limit(struct net *net,
1165 			      const struct ovs_conntrack_info *info,
1166 			      const struct nf_conntrack_tuple *tuple)
1167 {
1168 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1169 	const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1170 	u32 per_zone_limit, connections;
1171 	u32 conncount_key;
1172 
1173 	conncount_key = info->zone.id;
1174 
1175 	per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1176 	if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1177 		return 0;
1178 
1179 	connections = nf_conncount_count(net, ct_limit_info->data,
1180 					 &conncount_key, tuple, &info->zone);
1181 	if (connections > per_zone_limit)
1182 		return -ENOMEM;
1183 
1184 	return 0;
1185 }
1186 #endif
1187 
1188 /* Lookup connection and confirm if unconfirmed. */
1189 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1190 			 const struct ovs_conntrack_info *info,
1191 			 struct sk_buff *skb)
1192 {
1193 	enum ip_conntrack_info ctinfo;
1194 	struct nf_conn *ct;
1195 	int err;
1196 
1197 	err = __ovs_ct_lookup(net, key, info, skb);
1198 	if (err)
1199 		return err;
1200 
1201 	/* The connection could be invalid, in which case this is a no-op.*/
1202 	ct = nf_ct_get(skb, &ctinfo);
1203 	if (!ct)
1204 		return 0;
1205 
1206 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1207 	if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1208 		if (!nf_ct_is_confirmed(ct)) {
1209 			err = ovs_ct_check_limit(net, info,
1210 				&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1211 			if (err) {
1212 				net_warn_ratelimited("openvswitch: zone: %u "
1213 					"exceeds conntrack limit\n",
1214 					info->zone.id);
1215 				return err;
1216 			}
1217 		}
1218 	}
1219 #endif
1220 
1221 	/* Set the conntrack event mask if given.  NEW and DELETE events have
1222 	 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1223 	 * typically would receive many kinds of updates.  Setting the event
1224 	 * mask allows those events to be filtered.  The set event mask will
1225 	 * remain in effect for the lifetime of the connection unless changed
1226 	 * by a further CT action with both the commit flag and the eventmask
1227 	 * option. */
1228 	if (info->have_eventmask) {
1229 		struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1230 
1231 		if (cache)
1232 			cache->ctmask = info->eventmask;
1233 	}
1234 
1235 	/* Apply changes before confirming the connection so that the initial
1236 	 * conntrack NEW netlink event carries the values given in the CT
1237 	 * action.
1238 	 */
1239 	if (info->mark.mask) {
1240 		err = ovs_ct_set_mark(ct, key, info->mark.value,
1241 				      info->mark.mask);
1242 		if (err)
1243 			return err;
1244 	}
1245 	if (!nf_ct_is_confirmed(ct)) {
1246 		err = ovs_ct_init_labels(ct, key, &info->labels.value,
1247 					 &info->labels.mask);
1248 		if (err)
1249 			return err;
1250 	} else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1251 		   labels_nonzero(&info->labels.mask)) {
1252 		err = ovs_ct_set_labels(ct, key, &info->labels.value,
1253 					&info->labels.mask);
1254 		if (err)
1255 			return err;
1256 	}
1257 	/* This will take care of sending queued events even if the connection
1258 	 * is already confirmed.
1259 	 */
1260 	if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1261 		return -EINVAL;
1262 
1263 	return 0;
1264 }
1265 
1266 /* Trim the skb to the length specified by the IP/IPv6 header,
1267  * removing any trailing lower-layer padding. This prepares the skb
1268  * for higher-layer processing that assumes skb->len excludes padding
1269  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1270  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1271  */
1272 static int ovs_skb_network_trim(struct sk_buff *skb)
1273 {
1274 	unsigned int len;
1275 	int err;
1276 
1277 	switch (skb->protocol) {
1278 	case htons(ETH_P_IP):
1279 		len = ntohs(ip_hdr(skb)->tot_len);
1280 		break;
1281 	case htons(ETH_P_IPV6):
1282 		len = sizeof(struct ipv6hdr)
1283 			+ ntohs(ipv6_hdr(skb)->payload_len);
1284 		break;
1285 	default:
1286 		len = skb->len;
1287 	}
1288 
1289 	err = pskb_trim_rcsum(skb, len);
1290 	if (err)
1291 		kfree_skb(skb);
1292 
1293 	return err;
1294 }
1295 
1296 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1297  * value if 'skb' is freed.
1298  */
1299 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1300 		   struct sw_flow_key *key,
1301 		   const struct ovs_conntrack_info *info)
1302 {
1303 	int nh_ofs;
1304 	int err;
1305 
1306 	/* The conntrack module expects to be working at L3. */
1307 	nh_ofs = skb_network_offset(skb);
1308 	skb_pull_rcsum(skb, nh_ofs);
1309 
1310 	err = ovs_skb_network_trim(skb);
1311 	if (err)
1312 		return err;
1313 
1314 	if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1315 		err = handle_fragments(net, key, info->zone.id, skb);
1316 		if (err)
1317 			return err;
1318 	}
1319 
1320 	if (info->commit)
1321 		err = ovs_ct_commit(net, key, info, skb);
1322 	else
1323 		err = ovs_ct_lookup(net, key, info, skb);
1324 
1325 	skb_push(skb, nh_ofs);
1326 	skb_postpush_rcsum(skb, skb->data, nh_ofs);
1327 	if (err)
1328 		kfree_skb(skb);
1329 	return err;
1330 }
1331 
1332 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1333 {
1334 	if (skb_nfct(skb)) {
1335 		nf_conntrack_put(skb_nfct(skb));
1336 		nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1337 		ovs_ct_fill_key(skb, key, false);
1338 	}
1339 
1340 	return 0;
1341 }
1342 
1343 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1344 			     const struct sw_flow_key *key, bool log)
1345 {
1346 	struct nf_conntrack_helper *helper;
1347 	struct nf_conn_help *help;
1348 	int ret = 0;
1349 
1350 	helper = nf_conntrack_helper_try_module_get(name, info->family,
1351 						    key->ip.proto);
1352 	if (!helper) {
1353 		OVS_NLERR(log, "Unknown helper \"%s\"", name);
1354 		return -EINVAL;
1355 	}
1356 
1357 	help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1358 	if (!help) {
1359 		nf_conntrack_helper_put(helper);
1360 		return -ENOMEM;
1361 	}
1362 
1363 #if IS_ENABLED(CONFIG_NF_NAT)
1364 	if (info->nat) {
1365 		ret = nf_nat_helper_try_module_get(name, info->family,
1366 						   key->ip.proto);
1367 		if (ret) {
1368 			nf_conntrack_helper_put(helper);
1369 			OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1370 				  name, ret);
1371 			return ret;
1372 		}
1373 	}
1374 #endif
1375 	rcu_assign_pointer(help->helper, helper);
1376 	info->helper = helper;
1377 	return ret;
1378 }
1379 
1380 #if IS_ENABLED(CONFIG_NF_NAT)
1381 static int parse_nat(const struct nlattr *attr,
1382 		     struct ovs_conntrack_info *info, bool log)
1383 {
1384 	struct nlattr *a;
1385 	int rem;
1386 	bool have_ip_max = false;
1387 	bool have_proto_max = false;
1388 	bool ip_vers = (info->family == NFPROTO_IPV6);
1389 
1390 	nla_for_each_nested(a, attr, rem) {
1391 		static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1392 			[OVS_NAT_ATTR_SRC] = {0, 0},
1393 			[OVS_NAT_ATTR_DST] = {0, 0},
1394 			[OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1395 						 sizeof(struct in6_addr)},
1396 			[OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1397 						 sizeof(struct in6_addr)},
1398 			[OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1399 			[OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1400 			[OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1401 			[OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1402 			[OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1403 		};
1404 		int type = nla_type(a);
1405 
1406 		if (type > OVS_NAT_ATTR_MAX) {
1407 			OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1408 				  type, OVS_NAT_ATTR_MAX);
1409 			return -EINVAL;
1410 		}
1411 
1412 		if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1413 			OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1414 				  type, nla_len(a),
1415 				  ovs_nat_attr_lens[type][ip_vers]);
1416 			return -EINVAL;
1417 		}
1418 
1419 		switch (type) {
1420 		case OVS_NAT_ATTR_SRC:
1421 		case OVS_NAT_ATTR_DST:
1422 			if (info->nat) {
1423 				OVS_NLERR(log, "Only one type of NAT may be specified");
1424 				return -ERANGE;
1425 			}
1426 			info->nat |= OVS_CT_NAT;
1427 			info->nat |= ((type == OVS_NAT_ATTR_SRC)
1428 					? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1429 			break;
1430 
1431 		case OVS_NAT_ATTR_IP_MIN:
1432 			nla_memcpy(&info->range.min_addr, a,
1433 				   sizeof(info->range.min_addr));
1434 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1435 			break;
1436 
1437 		case OVS_NAT_ATTR_IP_MAX:
1438 			have_ip_max = true;
1439 			nla_memcpy(&info->range.max_addr, a,
1440 				   sizeof(info->range.max_addr));
1441 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1442 			break;
1443 
1444 		case OVS_NAT_ATTR_PROTO_MIN:
1445 			info->range.min_proto.all = htons(nla_get_u16(a));
1446 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1447 			break;
1448 
1449 		case OVS_NAT_ATTR_PROTO_MAX:
1450 			have_proto_max = true;
1451 			info->range.max_proto.all = htons(nla_get_u16(a));
1452 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1453 			break;
1454 
1455 		case OVS_NAT_ATTR_PERSISTENT:
1456 			info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1457 			break;
1458 
1459 		case OVS_NAT_ATTR_PROTO_HASH:
1460 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1461 			break;
1462 
1463 		case OVS_NAT_ATTR_PROTO_RANDOM:
1464 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1465 			break;
1466 
1467 		default:
1468 			OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1469 			return -EINVAL;
1470 		}
1471 	}
1472 
1473 	if (rem > 0) {
1474 		OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1475 		return -EINVAL;
1476 	}
1477 	if (!info->nat) {
1478 		/* Do not allow flags if no type is given. */
1479 		if (info->range.flags) {
1480 			OVS_NLERR(log,
1481 				  "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1482 				  );
1483 			return -EINVAL;
1484 		}
1485 		info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1486 	} else if (!info->commit) {
1487 		OVS_NLERR(log,
1488 			  "NAT attributes may be specified only when CT COMMIT flag is also specified."
1489 			  );
1490 		return -EINVAL;
1491 	}
1492 	/* Allow missing IP_MAX. */
1493 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1494 		memcpy(&info->range.max_addr, &info->range.min_addr,
1495 		       sizeof(info->range.max_addr));
1496 	}
1497 	/* Allow missing PROTO_MAX. */
1498 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1499 	    !have_proto_max) {
1500 		info->range.max_proto.all = info->range.min_proto.all;
1501 	}
1502 	return 0;
1503 }
1504 #endif
1505 
1506 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1507 	[OVS_CT_ATTR_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1508 	[OVS_CT_ATTR_FORCE_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1509 	[OVS_CT_ATTR_ZONE]	= { .minlen = sizeof(u16),
1510 				    .maxlen = sizeof(u16) },
1511 	[OVS_CT_ATTR_MARK]	= { .minlen = sizeof(struct md_mark),
1512 				    .maxlen = sizeof(struct md_mark) },
1513 	[OVS_CT_ATTR_LABELS]	= { .minlen = sizeof(struct md_labels),
1514 				    .maxlen = sizeof(struct md_labels) },
1515 	[OVS_CT_ATTR_HELPER]	= { .minlen = 1,
1516 				    .maxlen = NF_CT_HELPER_NAME_LEN },
1517 #if IS_ENABLED(CONFIG_NF_NAT)
1518 	/* NAT length is checked when parsing the nested attributes. */
1519 	[OVS_CT_ATTR_NAT]	= { .minlen = 0, .maxlen = INT_MAX },
1520 #endif
1521 	[OVS_CT_ATTR_EVENTMASK]	= { .minlen = sizeof(u32),
1522 				    .maxlen = sizeof(u32) },
1523 	[OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1524 				  .maxlen = CTNL_TIMEOUT_NAME_MAX },
1525 };
1526 
1527 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1528 		    const char **helper, bool log)
1529 {
1530 	struct nlattr *a;
1531 	int rem;
1532 
1533 	nla_for_each_nested(a, attr, rem) {
1534 		int type = nla_type(a);
1535 		int maxlen;
1536 		int minlen;
1537 
1538 		if (type > OVS_CT_ATTR_MAX) {
1539 			OVS_NLERR(log,
1540 				  "Unknown conntrack attr (type=%d, max=%d)",
1541 				  type, OVS_CT_ATTR_MAX);
1542 			return -EINVAL;
1543 		}
1544 
1545 		maxlen = ovs_ct_attr_lens[type].maxlen;
1546 		minlen = ovs_ct_attr_lens[type].minlen;
1547 		if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1548 			OVS_NLERR(log,
1549 				  "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1550 				  type, nla_len(a), maxlen);
1551 			return -EINVAL;
1552 		}
1553 
1554 		switch (type) {
1555 		case OVS_CT_ATTR_FORCE_COMMIT:
1556 			info->force = true;
1557 			fallthrough;
1558 		case OVS_CT_ATTR_COMMIT:
1559 			info->commit = true;
1560 			break;
1561 #ifdef CONFIG_NF_CONNTRACK_ZONES
1562 		case OVS_CT_ATTR_ZONE:
1563 			info->zone.id = nla_get_u16(a);
1564 			break;
1565 #endif
1566 #ifdef CONFIG_NF_CONNTRACK_MARK
1567 		case OVS_CT_ATTR_MARK: {
1568 			struct md_mark *mark = nla_data(a);
1569 
1570 			if (!mark->mask) {
1571 				OVS_NLERR(log, "ct_mark mask cannot be 0");
1572 				return -EINVAL;
1573 			}
1574 			info->mark = *mark;
1575 			break;
1576 		}
1577 #endif
1578 #ifdef CONFIG_NF_CONNTRACK_LABELS
1579 		case OVS_CT_ATTR_LABELS: {
1580 			struct md_labels *labels = nla_data(a);
1581 
1582 			if (!labels_nonzero(&labels->mask)) {
1583 				OVS_NLERR(log, "ct_labels mask cannot be 0");
1584 				return -EINVAL;
1585 			}
1586 			info->labels = *labels;
1587 			break;
1588 		}
1589 #endif
1590 		case OVS_CT_ATTR_HELPER:
1591 			*helper = nla_data(a);
1592 			if (!memchr(*helper, '\0', nla_len(a))) {
1593 				OVS_NLERR(log, "Invalid conntrack helper");
1594 				return -EINVAL;
1595 			}
1596 			break;
1597 #if IS_ENABLED(CONFIG_NF_NAT)
1598 		case OVS_CT_ATTR_NAT: {
1599 			int err = parse_nat(a, info, log);
1600 
1601 			if (err)
1602 				return err;
1603 			break;
1604 		}
1605 #endif
1606 		case OVS_CT_ATTR_EVENTMASK:
1607 			info->have_eventmask = true;
1608 			info->eventmask = nla_get_u32(a);
1609 			break;
1610 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1611 		case OVS_CT_ATTR_TIMEOUT:
1612 			memcpy(info->timeout, nla_data(a), nla_len(a));
1613 			if (!memchr(info->timeout, '\0', nla_len(a))) {
1614 				OVS_NLERR(log, "Invalid conntrack timeout");
1615 				return -EINVAL;
1616 			}
1617 			break;
1618 #endif
1619 
1620 		default:
1621 			OVS_NLERR(log, "Unknown conntrack attr (%d)",
1622 				  type);
1623 			return -EINVAL;
1624 		}
1625 	}
1626 
1627 #ifdef CONFIG_NF_CONNTRACK_MARK
1628 	if (!info->commit && info->mark.mask) {
1629 		OVS_NLERR(log,
1630 			  "Setting conntrack mark requires 'commit' flag.");
1631 		return -EINVAL;
1632 	}
1633 #endif
1634 #ifdef CONFIG_NF_CONNTRACK_LABELS
1635 	if (!info->commit && labels_nonzero(&info->labels.mask)) {
1636 		OVS_NLERR(log,
1637 			  "Setting conntrack labels requires 'commit' flag.");
1638 		return -EINVAL;
1639 	}
1640 #endif
1641 	if (rem > 0) {
1642 		OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1643 		return -EINVAL;
1644 	}
1645 
1646 	return 0;
1647 }
1648 
1649 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1650 {
1651 	if (attr == OVS_KEY_ATTR_CT_STATE)
1652 		return true;
1653 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1654 	    attr == OVS_KEY_ATTR_CT_ZONE)
1655 		return true;
1656 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1657 	    attr == OVS_KEY_ATTR_CT_MARK)
1658 		return true;
1659 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1660 	    attr == OVS_KEY_ATTR_CT_LABELS) {
1661 		struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1662 
1663 		return ovs_net->xt_label;
1664 	}
1665 
1666 	return false;
1667 }
1668 
1669 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1670 		       const struct sw_flow_key *key,
1671 		       struct sw_flow_actions **sfa,  bool log)
1672 {
1673 	struct ovs_conntrack_info ct_info;
1674 	const char *helper = NULL;
1675 	u16 family;
1676 	int err;
1677 
1678 	family = key_to_nfproto(key);
1679 	if (family == NFPROTO_UNSPEC) {
1680 		OVS_NLERR(log, "ct family unspecified");
1681 		return -EINVAL;
1682 	}
1683 
1684 	memset(&ct_info, 0, sizeof(ct_info));
1685 	ct_info.family = family;
1686 
1687 	nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1688 			NF_CT_DEFAULT_ZONE_DIR, 0);
1689 
1690 	err = parse_ct(attr, &ct_info, &helper, log);
1691 	if (err)
1692 		return err;
1693 
1694 	/* Set up template for tracking connections in specific zones. */
1695 	ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1696 	if (!ct_info.ct) {
1697 		OVS_NLERR(log, "Failed to allocate conntrack template");
1698 		return -ENOMEM;
1699 	}
1700 
1701 	if (ct_info.timeout[0]) {
1702 		if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1703 				      ct_info.timeout))
1704 			pr_info_ratelimited("Failed to associated timeout "
1705 					    "policy `%s'\n", ct_info.timeout);
1706 		else
1707 			ct_info.nf_ct_timeout = rcu_dereference(
1708 				nf_ct_timeout_find(ct_info.ct)->timeout);
1709 
1710 	}
1711 
1712 	if (helper) {
1713 		err = ovs_ct_add_helper(&ct_info, helper, key, log);
1714 		if (err)
1715 			goto err_free_ct;
1716 	}
1717 
1718 	err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1719 				 sizeof(ct_info), log);
1720 	if (err)
1721 		goto err_free_ct;
1722 
1723 	__set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1724 	nf_conntrack_get(&ct_info.ct->ct_general);
1725 	return 0;
1726 err_free_ct:
1727 	__ovs_ct_free_action(&ct_info);
1728 	return err;
1729 }
1730 
1731 #if IS_ENABLED(CONFIG_NF_NAT)
1732 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1733 			       struct sk_buff *skb)
1734 {
1735 	struct nlattr *start;
1736 
1737 	start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1738 	if (!start)
1739 		return false;
1740 
1741 	if (info->nat & OVS_CT_SRC_NAT) {
1742 		if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1743 			return false;
1744 	} else if (info->nat & OVS_CT_DST_NAT) {
1745 		if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1746 			return false;
1747 	} else {
1748 		goto out;
1749 	}
1750 
1751 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1752 		if (IS_ENABLED(CONFIG_NF_NAT) &&
1753 		    info->family == NFPROTO_IPV4) {
1754 			if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1755 					    info->range.min_addr.ip) ||
1756 			    (info->range.max_addr.ip
1757 			     != info->range.min_addr.ip &&
1758 			     (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1759 					      info->range.max_addr.ip))))
1760 				return false;
1761 		} else if (IS_ENABLED(CONFIG_IPV6) &&
1762 			   info->family == NFPROTO_IPV6) {
1763 			if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1764 					     &info->range.min_addr.in6) ||
1765 			    (memcmp(&info->range.max_addr.in6,
1766 				    &info->range.min_addr.in6,
1767 				    sizeof(info->range.max_addr.in6)) &&
1768 			     (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1769 					       &info->range.max_addr.in6))))
1770 				return false;
1771 		} else {
1772 			return false;
1773 		}
1774 	}
1775 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1776 	    (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1777 			 ntohs(info->range.min_proto.all)) ||
1778 	     (info->range.max_proto.all != info->range.min_proto.all &&
1779 	      nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1780 			  ntohs(info->range.max_proto.all)))))
1781 		return false;
1782 
1783 	if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1784 	    nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1785 		return false;
1786 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1787 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1788 		return false;
1789 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1790 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1791 		return false;
1792 out:
1793 	nla_nest_end(skb, start);
1794 
1795 	return true;
1796 }
1797 #endif
1798 
1799 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1800 			  struct sk_buff *skb)
1801 {
1802 	struct nlattr *start;
1803 
1804 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1805 	if (!start)
1806 		return -EMSGSIZE;
1807 
1808 	if (ct_info->commit && nla_put_flag(skb, ct_info->force
1809 					    ? OVS_CT_ATTR_FORCE_COMMIT
1810 					    : OVS_CT_ATTR_COMMIT))
1811 		return -EMSGSIZE;
1812 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1813 	    nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1814 		return -EMSGSIZE;
1815 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1816 	    nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1817 		    &ct_info->mark))
1818 		return -EMSGSIZE;
1819 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1820 	    labels_nonzero(&ct_info->labels.mask) &&
1821 	    nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1822 		    &ct_info->labels))
1823 		return -EMSGSIZE;
1824 	if (ct_info->helper) {
1825 		if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1826 				   ct_info->helper->name))
1827 			return -EMSGSIZE;
1828 	}
1829 	if (ct_info->have_eventmask &&
1830 	    nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1831 		return -EMSGSIZE;
1832 	if (ct_info->timeout[0]) {
1833 		if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1834 			return -EMSGSIZE;
1835 	}
1836 
1837 #if IS_ENABLED(CONFIG_NF_NAT)
1838 	if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1839 		return -EMSGSIZE;
1840 #endif
1841 	nla_nest_end(skb, start);
1842 
1843 	return 0;
1844 }
1845 
1846 void ovs_ct_free_action(const struct nlattr *a)
1847 {
1848 	struct ovs_conntrack_info *ct_info = nla_data(a);
1849 
1850 	__ovs_ct_free_action(ct_info);
1851 }
1852 
1853 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1854 {
1855 	if (ct_info->helper) {
1856 #if IS_ENABLED(CONFIG_NF_NAT)
1857 		if (ct_info->nat)
1858 			nf_nat_helper_put(ct_info->helper);
1859 #endif
1860 		nf_conntrack_helper_put(ct_info->helper);
1861 	}
1862 	if (ct_info->ct) {
1863 		if (ct_info->timeout[0])
1864 			nf_ct_destroy_timeout(ct_info->ct);
1865 		nf_ct_tmpl_free(ct_info->ct);
1866 	}
1867 }
1868 
1869 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1870 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1871 {
1872 	int i, err;
1873 
1874 	ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1875 					 GFP_KERNEL);
1876 	if (!ovs_net->ct_limit_info)
1877 		return -ENOMEM;
1878 
1879 	ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1880 	ovs_net->ct_limit_info->limits =
1881 		kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1882 			      GFP_KERNEL);
1883 	if (!ovs_net->ct_limit_info->limits) {
1884 		kfree(ovs_net->ct_limit_info);
1885 		return -ENOMEM;
1886 	}
1887 
1888 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1889 		INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1890 
1891 	ovs_net->ct_limit_info->data =
1892 		nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1893 
1894 	if (IS_ERR(ovs_net->ct_limit_info->data)) {
1895 		err = PTR_ERR(ovs_net->ct_limit_info->data);
1896 		kfree(ovs_net->ct_limit_info->limits);
1897 		kfree(ovs_net->ct_limit_info);
1898 		pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1899 		return err;
1900 	}
1901 	return 0;
1902 }
1903 
1904 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1905 {
1906 	const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1907 	int i;
1908 
1909 	nf_conncount_destroy(net, NFPROTO_INET, info->data);
1910 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1911 		struct hlist_head *head = &info->limits[i];
1912 		struct ovs_ct_limit *ct_limit;
1913 
1914 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node,
1915 					 lockdep_ovsl_is_held())
1916 			kfree_rcu(ct_limit, rcu);
1917 	}
1918 	kfree(info->limits);
1919 	kfree(info);
1920 }
1921 
1922 static struct sk_buff *
1923 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1924 			     struct ovs_header **ovs_reply_header)
1925 {
1926 	struct ovs_header *ovs_header = info->userhdr;
1927 	struct sk_buff *skb;
1928 
1929 	skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1930 	if (!skb)
1931 		return ERR_PTR(-ENOMEM);
1932 
1933 	*ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1934 					info->snd_seq,
1935 					&dp_ct_limit_genl_family, 0, cmd);
1936 
1937 	if (!*ovs_reply_header) {
1938 		nlmsg_free(skb);
1939 		return ERR_PTR(-EMSGSIZE);
1940 	}
1941 	(*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1942 
1943 	return skb;
1944 }
1945 
1946 static bool check_zone_id(int zone_id, u16 *pzone)
1947 {
1948 	if (zone_id >= 0 && zone_id <= 65535) {
1949 		*pzone = (u16)zone_id;
1950 		return true;
1951 	}
1952 	return false;
1953 }
1954 
1955 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1956 				       struct ovs_ct_limit_info *info)
1957 {
1958 	struct ovs_zone_limit *zone_limit;
1959 	int rem;
1960 	u16 zone;
1961 
1962 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1963 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1964 
1965 	while (rem >= sizeof(*zone_limit)) {
1966 		if (unlikely(zone_limit->zone_id ==
1967 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1968 			ovs_lock();
1969 			info->default_limit = zone_limit->limit;
1970 			ovs_unlock();
1971 		} else if (unlikely(!check_zone_id(
1972 				zone_limit->zone_id, &zone))) {
1973 			OVS_NLERR(true, "zone id is out of range");
1974 		} else {
1975 			struct ovs_ct_limit *ct_limit;
1976 
1977 			ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1978 			if (!ct_limit)
1979 				return -ENOMEM;
1980 
1981 			ct_limit->zone = zone;
1982 			ct_limit->limit = zone_limit->limit;
1983 
1984 			ovs_lock();
1985 			ct_limit_set(info, ct_limit);
1986 			ovs_unlock();
1987 		}
1988 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1989 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1990 				NLA_ALIGN(sizeof(*zone_limit)));
1991 	}
1992 
1993 	if (rem)
1994 		OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1995 
1996 	return 0;
1997 }
1998 
1999 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
2000 				       struct ovs_ct_limit_info *info)
2001 {
2002 	struct ovs_zone_limit *zone_limit;
2003 	int rem;
2004 	u16 zone;
2005 
2006 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
2007 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2008 
2009 	while (rem >= sizeof(*zone_limit)) {
2010 		if (unlikely(zone_limit->zone_id ==
2011 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2012 			ovs_lock();
2013 			info->default_limit = OVS_CT_LIMIT_DEFAULT;
2014 			ovs_unlock();
2015 		} else if (unlikely(!check_zone_id(
2016 				zone_limit->zone_id, &zone))) {
2017 			OVS_NLERR(true, "zone id is out of range");
2018 		} else {
2019 			ovs_lock();
2020 			ct_limit_del(info, zone);
2021 			ovs_unlock();
2022 		}
2023 		rem -= NLA_ALIGN(sizeof(*zone_limit));
2024 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2025 				NLA_ALIGN(sizeof(*zone_limit)));
2026 	}
2027 
2028 	if (rem)
2029 		OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
2030 
2031 	return 0;
2032 }
2033 
2034 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2035 					  struct sk_buff *reply)
2036 {
2037 	struct ovs_zone_limit zone_limit = {
2038 		.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE,
2039 		.limit   = info->default_limit,
2040 	};
2041 
2042 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2043 }
2044 
2045 static int __ovs_ct_limit_get_zone_limit(struct net *net,
2046 					 struct nf_conncount_data *data,
2047 					 u16 zone_id, u32 limit,
2048 					 struct sk_buff *reply)
2049 {
2050 	struct nf_conntrack_zone ct_zone;
2051 	struct ovs_zone_limit zone_limit;
2052 	u32 conncount_key = zone_id;
2053 
2054 	zone_limit.zone_id = zone_id;
2055 	zone_limit.limit = limit;
2056 	nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2057 
2058 	zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2059 					      &ct_zone);
2060 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2061 }
2062 
2063 static int ovs_ct_limit_get_zone_limit(struct net *net,
2064 				       struct nlattr *nla_zone_limit,
2065 				       struct ovs_ct_limit_info *info,
2066 				       struct sk_buff *reply)
2067 {
2068 	struct ovs_zone_limit *zone_limit;
2069 	int rem, err;
2070 	u32 limit;
2071 	u16 zone;
2072 
2073 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
2074 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2075 
2076 	while (rem >= sizeof(*zone_limit)) {
2077 		if (unlikely(zone_limit->zone_id ==
2078 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2079 			err = ovs_ct_limit_get_default_limit(info, reply);
2080 			if (err)
2081 				return err;
2082 		} else if (unlikely(!check_zone_id(zone_limit->zone_id,
2083 							&zone))) {
2084 			OVS_NLERR(true, "zone id is out of range");
2085 		} else {
2086 			rcu_read_lock();
2087 			limit = ct_limit_get(info, zone);
2088 			rcu_read_unlock();
2089 
2090 			err = __ovs_ct_limit_get_zone_limit(
2091 				net, info->data, zone, limit, reply);
2092 			if (err)
2093 				return err;
2094 		}
2095 		rem -= NLA_ALIGN(sizeof(*zone_limit));
2096 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2097 				NLA_ALIGN(sizeof(*zone_limit)));
2098 	}
2099 
2100 	if (rem)
2101 		OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2102 
2103 	return 0;
2104 }
2105 
2106 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2107 					   struct ovs_ct_limit_info *info,
2108 					   struct sk_buff *reply)
2109 {
2110 	struct ovs_ct_limit *ct_limit;
2111 	struct hlist_head *head;
2112 	int i, err = 0;
2113 
2114 	err = ovs_ct_limit_get_default_limit(info, reply);
2115 	if (err)
2116 		return err;
2117 
2118 	rcu_read_lock();
2119 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2120 		head = &info->limits[i];
2121 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2122 			err = __ovs_ct_limit_get_zone_limit(net, info->data,
2123 				ct_limit->zone, ct_limit->limit, reply);
2124 			if (err)
2125 				goto exit_err;
2126 		}
2127 	}
2128 
2129 exit_err:
2130 	rcu_read_unlock();
2131 	return err;
2132 }
2133 
2134 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2135 {
2136 	struct nlattr **a = info->attrs;
2137 	struct sk_buff *reply;
2138 	struct ovs_header *ovs_reply_header;
2139 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2140 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2141 	int err;
2142 
2143 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2144 					     &ovs_reply_header);
2145 	if (IS_ERR(reply))
2146 		return PTR_ERR(reply);
2147 
2148 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2149 		err = -EINVAL;
2150 		goto exit_err;
2151 	}
2152 
2153 	err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2154 					  ct_limit_info);
2155 	if (err)
2156 		goto exit_err;
2157 
2158 	static_branch_enable(&ovs_ct_limit_enabled);
2159 
2160 	genlmsg_end(reply, ovs_reply_header);
2161 	return genlmsg_reply(reply, info);
2162 
2163 exit_err:
2164 	nlmsg_free(reply);
2165 	return err;
2166 }
2167 
2168 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2169 {
2170 	struct nlattr **a = info->attrs;
2171 	struct sk_buff *reply;
2172 	struct ovs_header *ovs_reply_header;
2173 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2174 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2175 	int err;
2176 
2177 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2178 					     &ovs_reply_header);
2179 	if (IS_ERR(reply))
2180 		return PTR_ERR(reply);
2181 
2182 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2183 		err = -EINVAL;
2184 		goto exit_err;
2185 	}
2186 
2187 	err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2188 					  ct_limit_info);
2189 	if (err)
2190 		goto exit_err;
2191 
2192 	genlmsg_end(reply, ovs_reply_header);
2193 	return genlmsg_reply(reply, info);
2194 
2195 exit_err:
2196 	nlmsg_free(reply);
2197 	return err;
2198 }
2199 
2200 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2201 {
2202 	struct nlattr **a = info->attrs;
2203 	struct nlattr *nla_reply;
2204 	struct sk_buff *reply;
2205 	struct ovs_header *ovs_reply_header;
2206 	struct net *net = sock_net(skb->sk);
2207 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2208 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2209 	int err;
2210 
2211 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2212 					     &ovs_reply_header);
2213 	if (IS_ERR(reply))
2214 		return PTR_ERR(reply);
2215 
2216 	nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2217 	if (!nla_reply) {
2218 		err = -EMSGSIZE;
2219 		goto exit_err;
2220 	}
2221 
2222 	if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2223 		err = ovs_ct_limit_get_zone_limit(
2224 			net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2225 			reply);
2226 		if (err)
2227 			goto exit_err;
2228 	} else {
2229 		err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2230 						      reply);
2231 		if (err)
2232 			goto exit_err;
2233 	}
2234 
2235 	nla_nest_end(reply, nla_reply);
2236 	genlmsg_end(reply, ovs_reply_header);
2237 	return genlmsg_reply(reply, info);
2238 
2239 exit_err:
2240 	nlmsg_free(reply);
2241 	return err;
2242 }
2243 
2244 static const struct genl_small_ops ct_limit_genl_ops[] = {
2245 	{ .cmd = OVS_CT_LIMIT_CMD_SET,
2246 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2247 		.flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2248 					   * privilege. */
2249 		.doit = ovs_ct_limit_cmd_set,
2250 	},
2251 	{ .cmd = OVS_CT_LIMIT_CMD_DEL,
2252 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2253 		.flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2254 					   * privilege. */
2255 		.doit = ovs_ct_limit_cmd_del,
2256 	},
2257 	{ .cmd = OVS_CT_LIMIT_CMD_GET,
2258 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2259 		.flags = 0,		  /* OK for unprivileged users. */
2260 		.doit = ovs_ct_limit_cmd_get,
2261 	},
2262 };
2263 
2264 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2265 	.name = OVS_CT_LIMIT_MCGROUP,
2266 };
2267 
2268 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2269 	.hdrsize = sizeof(struct ovs_header),
2270 	.name = OVS_CT_LIMIT_FAMILY,
2271 	.version = OVS_CT_LIMIT_VERSION,
2272 	.maxattr = OVS_CT_LIMIT_ATTR_MAX,
2273 	.policy = ct_limit_policy,
2274 	.netnsok = true,
2275 	.parallel_ops = true,
2276 	.small_ops = ct_limit_genl_ops,
2277 	.n_small_ops = ARRAY_SIZE(ct_limit_genl_ops),
2278 	.mcgrps = &ovs_ct_limit_multicast_group,
2279 	.n_mcgrps = 1,
2280 	.module = THIS_MODULE,
2281 };
2282 #endif
2283 
2284 int ovs_ct_init(struct net *net)
2285 {
2286 	unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2287 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2288 
2289 	if (nf_connlabels_get(net, n_bits - 1)) {
2290 		ovs_net->xt_label = false;
2291 		OVS_NLERR(true, "Failed to set connlabel length");
2292 	} else {
2293 		ovs_net->xt_label = true;
2294 	}
2295 
2296 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2297 	return ovs_ct_limit_init(net, ovs_net);
2298 #else
2299 	return 0;
2300 #endif
2301 }
2302 
2303 void ovs_ct_exit(struct net *net)
2304 {
2305 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2306 
2307 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2308 	ovs_ct_limit_exit(net, ovs_net);
2309 #endif
2310 
2311 	if (ovs_net->xt_label)
2312 		nf_connlabels_put(net);
2313 }
2314