xref: /openbmc/linux/net/openvswitch/conntrack.c (revision b38269ec)
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_rcsum(skb, nh_off);
813 
814 	return err;
815 }
816 
817 static void ovs_nat_update_key(struct sw_flow_key *key,
818 			       const struct sk_buff *skb,
819 			       enum nf_nat_manip_type maniptype)
820 {
821 	if (maniptype == NF_NAT_MANIP_SRC) {
822 		__be16 src;
823 
824 		key->ct_state |= OVS_CS_F_SRC_NAT;
825 		if (key->eth.type == htons(ETH_P_IP))
826 			key->ipv4.addr.src = ip_hdr(skb)->saddr;
827 		else if (key->eth.type == htons(ETH_P_IPV6))
828 			memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
829 			       sizeof(key->ipv6.addr.src));
830 		else
831 			return;
832 
833 		if (key->ip.proto == IPPROTO_UDP)
834 			src = udp_hdr(skb)->source;
835 		else if (key->ip.proto == IPPROTO_TCP)
836 			src = tcp_hdr(skb)->source;
837 		else if (key->ip.proto == IPPROTO_SCTP)
838 			src = sctp_hdr(skb)->source;
839 		else
840 			return;
841 
842 		key->tp.src = src;
843 	} else {
844 		__be16 dst;
845 
846 		key->ct_state |= OVS_CS_F_DST_NAT;
847 		if (key->eth.type == htons(ETH_P_IP))
848 			key->ipv4.addr.dst = ip_hdr(skb)->daddr;
849 		else if (key->eth.type == htons(ETH_P_IPV6))
850 			memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
851 			       sizeof(key->ipv6.addr.dst));
852 		else
853 			return;
854 
855 		if (key->ip.proto == IPPROTO_UDP)
856 			dst = udp_hdr(skb)->dest;
857 		else if (key->ip.proto == IPPROTO_TCP)
858 			dst = tcp_hdr(skb)->dest;
859 		else if (key->ip.proto == IPPROTO_SCTP)
860 			dst = sctp_hdr(skb)->dest;
861 		else
862 			return;
863 
864 		key->tp.dst = dst;
865 	}
866 }
867 
868 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
869 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
870 		      const struct ovs_conntrack_info *info,
871 		      struct sk_buff *skb, struct nf_conn *ct,
872 		      enum ip_conntrack_info ctinfo)
873 {
874 	enum nf_nat_manip_type maniptype;
875 	int err;
876 
877 	/* Add NAT extension if not confirmed yet. */
878 	if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
879 		return NF_ACCEPT;   /* Can't NAT. */
880 
881 	/* Determine NAT type.
882 	 * Check if the NAT type can be deduced from the tracked connection.
883 	 * Make sure new expected connections (IP_CT_RELATED) are NATted only
884 	 * when committing.
885 	 */
886 	if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
887 	    ct->status & IPS_NAT_MASK &&
888 	    (ctinfo != IP_CT_RELATED || info->commit)) {
889 		/* NAT an established or related connection like before. */
890 		if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
891 			/* This is the REPLY direction for a connection
892 			 * for which NAT was applied in the forward
893 			 * direction.  Do the reverse NAT.
894 			 */
895 			maniptype = ct->status & IPS_SRC_NAT
896 				? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
897 		else
898 			maniptype = ct->status & IPS_SRC_NAT
899 				? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
900 	} else if (info->nat & OVS_CT_SRC_NAT) {
901 		maniptype = NF_NAT_MANIP_SRC;
902 	} else if (info->nat & OVS_CT_DST_NAT) {
903 		maniptype = NF_NAT_MANIP_DST;
904 	} else {
905 		return NF_ACCEPT; /* Connection is not NATed. */
906 	}
907 	err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
908 
909 	if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) {
910 		if (ct->status & IPS_SRC_NAT) {
911 			if (maniptype == NF_NAT_MANIP_SRC)
912 				maniptype = NF_NAT_MANIP_DST;
913 			else
914 				maniptype = NF_NAT_MANIP_SRC;
915 
916 			err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range,
917 						 maniptype);
918 		} else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
919 			err = ovs_ct_nat_execute(skb, ct, ctinfo, NULL,
920 						 NF_NAT_MANIP_SRC);
921 		}
922 	}
923 
924 	/* Mark NAT done if successful and update the flow key. */
925 	if (err == NF_ACCEPT)
926 		ovs_nat_update_key(key, skb, maniptype);
927 
928 	return err;
929 }
930 #else /* !CONFIG_NF_NAT */
931 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
932 		      const struct ovs_conntrack_info *info,
933 		      struct sk_buff *skb, struct nf_conn *ct,
934 		      enum ip_conntrack_info ctinfo)
935 {
936 	return NF_ACCEPT;
937 }
938 #endif
939 
940 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
941  * not done already.  Update key with new CT state after passing the packet
942  * through conntrack.
943  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
944  * set to NULL and 0 will be returned.
945  */
946 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
947 			   const struct ovs_conntrack_info *info,
948 			   struct sk_buff *skb)
949 {
950 	/* If we are recirculating packets to match on conntrack fields and
951 	 * committing with a separate conntrack action,  then we don't need to
952 	 * actually run the packet through conntrack twice unless it's for a
953 	 * different zone.
954 	 */
955 	bool cached = skb_nfct_cached(net, key, info, skb);
956 	enum ip_conntrack_info ctinfo;
957 	struct nf_conn *ct;
958 
959 	if (!cached) {
960 		struct nf_hook_state state = {
961 			.hook = NF_INET_PRE_ROUTING,
962 			.pf = info->family,
963 			.net = net,
964 		};
965 		struct nf_conn *tmpl = info->ct;
966 		int err;
967 
968 		/* Associate skb with specified zone. */
969 		if (tmpl) {
970 			nf_conntrack_put(skb_nfct(skb));
971 			nf_conntrack_get(&tmpl->ct_general);
972 			nf_ct_set(skb, tmpl, IP_CT_NEW);
973 		}
974 
975 		err = nf_conntrack_in(skb, &state);
976 		if (err != NF_ACCEPT)
977 			return -ENOENT;
978 
979 		/* Clear CT state NAT flags to mark that we have not yet done
980 		 * NAT after the nf_conntrack_in() call.  We can actually clear
981 		 * the whole state, as it will be re-initialized below.
982 		 */
983 		key->ct_state = 0;
984 
985 		/* Update the key, but keep the NAT flags. */
986 		ovs_ct_update_key(skb, info, key, true, true);
987 	}
988 
989 	ct = nf_ct_get(skb, &ctinfo);
990 	if (ct) {
991 		bool add_helper = false;
992 
993 		/* Packets starting a new connection must be NATted before the
994 		 * helper, so that the helper knows about the NAT.  We enforce
995 		 * this by delaying both NAT and helper calls for unconfirmed
996 		 * connections until the committing CT action.  For later
997 		 * packets NAT and Helper may be called in either order.
998 		 *
999 		 * NAT will be done only if the CT action has NAT, and only
1000 		 * once per packet (per zone), as guarded by the NAT bits in
1001 		 * the key->ct_state.
1002 		 */
1003 		if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
1004 		    (nf_ct_is_confirmed(ct) || info->commit) &&
1005 		    ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
1006 			return -EINVAL;
1007 		}
1008 
1009 		/* Userspace may decide to perform a ct lookup without a helper
1010 		 * specified followed by a (recirculate and) commit with one,
1011 		 * or attach a helper in a later commit.  Therefore, for
1012 		 * connections which we will commit, we may need to attach
1013 		 * the helper here.
1014 		 */
1015 		if (info->commit && info->helper && !nfct_help(ct)) {
1016 			int err = __nf_ct_try_assign_helper(ct, info->ct,
1017 							    GFP_ATOMIC);
1018 			if (err)
1019 				return err;
1020 			add_helper = true;
1021 
1022 			/* helper installed, add seqadj if NAT is required */
1023 			if (info->nat && !nfct_seqadj(ct)) {
1024 				if (!nfct_seqadj_ext_add(ct))
1025 					return -EINVAL;
1026 			}
1027 		}
1028 
1029 		/* Call the helper only if:
1030 		 * - nf_conntrack_in() was executed above ("!cached") or a
1031 		 *   helper was just attached ("add_helper") for a confirmed
1032 		 *   connection, or
1033 		 * - When committing an unconfirmed connection.
1034 		 */
1035 		if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
1036 					      info->commit) &&
1037 		    ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1038 			return -EINVAL;
1039 		}
1040 
1041 		if (nf_ct_protonum(ct) == IPPROTO_TCP &&
1042 		    nf_ct_is_confirmed(ct) && nf_conntrack_tcp_established(ct)) {
1043 			/* Be liberal for tcp packets so that out-of-window
1044 			 * packets are not marked invalid.
1045 			 */
1046 			nf_ct_set_tcp_be_liberal(ct);
1047 		}
1048 	}
1049 
1050 	return 0;
1051 }
1052 
1053 /* Lookup connection and read fields into key. */
1054 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1055 			 const struct ovs_conntrack_info *info,
1056 			 struct sk_buff *skb)
1057 {
1058 	struct nf_conntrack_expect *exp;
1059 
1060 	/* If we pass an expected packet through nf_conntrack_in() the
1061 	 * expectation is typically removed, but the packet could still be
1062 	 * lost in upcall processing.  To prevent this from happening we
1063 	 * perform an explicit expectation lookup.  Expected connections are
1064 	 * always new, and will be passed through conntrack only when they are
1065 	 * committed, as it is OK to remove the expectation at that time.
1066 	 */
1067 	exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1068 	if (exp) {
1069 		u8 state;
1070 
1071 		/* NOTE: New connections are NATted and Helped only when
1072 		 * committed, so we are not calling into NAT here.
1073 		 */
1074 		state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1075 		__ovs_ct_update_key(key, state, &info->zone, exp->master);
1076 	} else {
1077 		struct nf_conn *ct;
1078 		int err;
1079 
1080 		err = __ovs_ct_lookup(net, key, info, skb);
1081 		if (err)
1082 			return err;
1083 
1084 		ct = (struct nf_conn *)skb_nfct(skb);
1085 		if (ct)
1086 			nf_ct_deliver_cached_events(ct);
1087 	}
1088 
1089 	return 0;
1090 }
1091 
1092 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1093 {
1094 	size_t i;
1095 
1096 	for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1097 		if (labels->ct_labels_32[i])
1098 			return true;
1099 
1100 	return false;
1101 }
1102 
1103 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1104 static struct hlist_head *ct_limit_hash_bucket(
1105 	const struct ovs_ct_limit_info *info, u16 zone)
1106 {
1107 	return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1108 }
1109 
1110 /* Call with ovs_mutex */
1111 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1112 			 struct ovs_ct_limit *new_ct_limit)
1113 {
1114 	struct ovs_ct_limit *ct_limit;
1115 	struct hlist_head *head;
1116 
1117 	head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1118 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1119 		if (ct_limit->zone == new_ct_limit->zone) {
1120 			hlist_replace_rcu(&ct_limit->hlist_node,
1121 					  &new_ct_limit->hlist_node);
1122 			kfree_rcu(ct_limit, rcu);
1123 			return;
1124 		}
1125 	}
1126 
1127 	hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1128 }
1129 
1130 /* Call with ovs_mutex */
1131 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1132 {
1133 	struct ovs_ct_limit *ct_limit;
1134 	struct hlist_head *head;
1135 	struct hlist_node *n;
1136 
1137 	head = ct_limit_hash_bucket(info, zone);
1138 	hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1139 		if (ct_limit->zone == zone) {
1140 			hlist_del_rcu(&ct_limit->hlist_node);
1141 			kfree_rcu(ct_limit, rcu);
1142 			return;
1143 		}
1144 	}
1145 }
1146 
1147 /* Call with RCU read lock */
1148 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1149 {
1150 	struct ovs_ct_limit *ct_limit;
1151 	struct hlist_head *head;
1152 
1153 	head = ct_limit_hash_bucket(info, zone);
1154 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1155 		if (ct_limit->zone == zone)
1156 			return ct_limit->limit;
1157 	}
1158 
1159 	return info->default_limit;
1160 }
1161 
1162 static int ovs_ct_check_limit(struct net *net,
1163 			      const struct ovs_conntrack_info *info,
1164 			      const struct nf_conntrack_tuple *tuple)
1165 {
1166 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1167 	const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1168 	u32 per_zone_limit, connections;
1169 	u32 conncount_key;
1170 
1171 	conncount_key = info->zone.id;
1172 
1173 	per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1174 	if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1175 		return 0;
1176 
1177 	connections = nf_conncount_count(net, ct_limit_info->data,
1178 					 &conncount_key, tuple, &info->zone);
1179 	if (connections > per_zone_limit)
1180 		return -ENOMEM;
1181 
1182 	return 0;
1183 }
1184 #endif
1185 
1186 /* Lookup connection and confirm if unconfirmed. */
1187 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1188 			 const struct ovs_conntrack_info *info,
1189 			 struct sk_buff *skb)
1190 {
1191 	enum ip_conntrack_info ctinfo;
1192 	struct nf_conn *ct;
1193 	int err;
1194 
1195 	err = __ovs_ct_lookup(net, key, info, skb);
1196 	if (err)
1197 		return err;
1198 
1199 	/* The connection could be invalid, in which case this is a no-op.*/
1200 	ct = nf_ct_get(skb, &ctinfo);
1201 	if (!ct)
1202 		return 0;
1203 
1204 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1205 	if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1206 		if (!nf_ct_is_confirmed(ct)) {
1207 			err = ovs_ct_check_limit(net, info,
1208 				&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1209 			if (err) {
1210 				net_warn_ratelimited("openvswitch: zone: %u "
1211 					"exceeds conntrack limit\n",
1212 					info->zone.id);
1213 				return err;
1214 			}
1215 		}
1216 	}
1217 #endif
1218 
1219 	/* Set the conntrack event mask if given.  NEW and DELETE events have
1220 	 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1221 	 * typically would receive many kinds of updates.  Setting the event
1222 	 * mask allows those events to be filtered.  The set event mask will
1223 	 * remain in effect for the lifetime of the connection unless changed
1224 	 * by a further CT action with both the commit flag and the eventmask
1225 	 * option. */
1226 	if (info->have_eventmask) {
1227 		struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1228 
1229 		if (cache)
1230 			cache->ctmask = info->eventmask;
1231 	}
1232 
1233 	/* Apply changes before confirming the connection so that the initial
1234 	 * conntrack NEW netlink event carries the values given in the CT
1235 	 * action.
1236 	 */
1237 	if (info->mark.mask) {
1238 		err = ovs_ct_set_mark(ct, key, info->mark.value,
1239 				      info->mark.mask);
1240 		if (err)
1241 			return err;
1242 	}
1243 	if (!nf_ct_is_confirmed(ct)) {
1244 		err = ovs_ct_init_labels(ct, key, &info->labels.value,
1245 					 &info->labels.mask);
1246 		if (err)
1247 			return err;
1248 	} else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1249 		   labels_nonzero(&info->labels.mask)) {
1250 		err = ovs_ct_set_labels(ct, key, &info->labels.value,
1251 					&info->labels.mask);
1252 		if (err)
1253 			return err;
1254 	}
1255 	/* This will take care of sending queued events even if the connection
1256 	 * is already confirmed.
1257 	 */
1258 	if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1259 		return -EINVAL;
1260 
1261 	return 0;
1262 }
1263 
1264 /* Trim the skb to the length specified by the IP/IPv6 header,
1265  * removing any trailing lower-layer padding. This prepares the skb
1266  * for higher-layer processing that assumes skb->len excludes padding
1267  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1268  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1269  */
1270 static int ovs_skb_network_trim(struct sk_buff *skb)
1271 {
1272 	unsigned int len;
1273 	int err;
1274 
1275 	switch (skb->protocol) {
1276 	case htons(ETH_P_IP):
1277 		len = ntohs(ip_hdr(skb)->tot_len);
1278 		break;
1279 	case htons(ETH_P_IPV6):
1280 		len = sizeof(struct ipv6hdr)
1281 			+ ntohs(ipv6_hdr(skb)->payload_len);
1282 		break;
1283 	default:
1284 		len = skb->len;
1285 	}
1286 
1287 	err = pskb_trim_rcsum(skb, len);
1288 	if (err)
1289 		kfree_skb(skb);
1290 
1291 	return err;
1292 }
1293 
1294 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1295  * value if 'skb' is freed.
1296  */
1297 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1298 		   struct sw_flow_key *key,
1299 		   const struct ovs_conntrack_info *info)
1300 {
1301 	int nh_ofs;
1302 	int err;
1303 
1304 	/* The conntrack module expects to be working at L3. */
1305 	nh_ofs = skb_network_offset(skb);
1306 	skb_pull_rcsum(skb, nh_ofs);
1307 
1308 	err = ovs_skb_network_trim(skb);
1309 	if (err)
1310 		return err;
1311 
1312 	if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1313 		err = handle_fragments(net, key, info->zone.id, skb);
1314 		if (err)
1315 			return err;
1316 	}
1317 
1318 	if (info->commit)
1319 		err = ovs_ct_commit(net, key, info, skb);
1320 	else
1321 		err = ovs_ct_lookup(net, key, info, skb);
1322 
1323 	skb_push_rcsum(skb, nh_ofs);
1324 	if (err)
1325 		kfree_skb(skb);
1326 	return err;
1327 }
1328 
1329 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1330 {
1331 	nf_conntrack_put(skb_nfct(skb));
1332 	nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1333 	ovs_ct_fill_key(skb, key, false);
1334 
1335 	return 0;
1336 }
1337 
1338 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1339 			     const struct sw_flow_key *key, bool log)
1340 {
1341 	struct nf_conntrack_helper *helper;
1342 	struct nf_conn_help *help;
1343 	int ret = 0;
1344 
1345 	helper = nf_conntrack_helper_try_module_get(name, info->family,
1346 						    key->ip.proto);
1347 	if (!helper) {
1348 		OVS_NLERR(log, "Unknown helper \"%s\"", name);
1349 		return -EINVAL;
1350 	}
1351 
1352 	help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1353 	if (!help) {
1354 		nf_conntrack_helper_put(helper);
1355 		return -ENOMEM;
1356 	}
1357 
1358 #if IS_ENABLED(CONFIG_NF_NAT)
1359 	if (info->nat) {
1360 		ret = nf_nat_helper_try_module_get(name, info->family,
1361 						   key->ip.proto);
1362 		if (ret) {
1363 			nf_conntrack_helper_put(helper);
1364 			OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1365 				  name, ret);
1366 			return ret;
1367 		}
1368 	}
1369 #endif
1370 	rcu_assign_pointer(help->helper, helper);
1371 	info->helper = helper;
1372 	return ret;
1373 }
1374 
1375 #if IS_ENABLED(CONFIG_NF_NAT)
1376 static int parse_nat(const struct nlattr *attr,
1377 		     struct ovs_conntrack_info *info, bool log)
1378 {
1379 	struct nlattr *a;
1380 	int rem;
1381 	bool have_ip_max = false;
1382 	bool have_proto_max = false;
1383 	bool ip_vers = (info->family == NFPROTO_IPV6);
1384 
1385 	nla_for_each_nested(a, attr, rem) {
1386 		static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1387 			[OVS_NAT_ATTR_SRC] = {0, 0},
1388 			[OVS_NAT_ATTR_DST] = {0, 0},
1389 			[OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1390 						 sizeof(struct in6_addr)},
1391 			[OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1392 						 sizeof(struct in6_addr)},
1393 			[OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1394 			[OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1395 			[OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1396 			[OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1397 			[OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1398 		};
1399 		int type = nla_type(a);
1400 
1401 		if (type > OVS_NAT_ATTR_MAX) {
1402 			OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1403 				  type, OVS_NAT_ATTR_MAX);
1404 			return -EINVAL;
1405 		}
1406 
1407 		if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1408 			OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1409 				  type, nla_len(a),
1410 				  ovs_nat_attr_lens[type][ip_vers]);
1411 			return -EINVAL;
1412 		}
1413 
1414 		switch (type) {
1415 		case OVS_NAT_ATTR_SRC:
1416 		case OVS_NAT_ATTR_DST:
1417 			if (info->nat) {
1418 				OVS_NLERR(log, "Only one type of NAT may be specified");
1419 				return -ERANGE;
1420 			}
1421 			info->nat |= OVS_CT_NAT;
1422 			info->nat |= ((type == OVS_NAT_ATTR_SRC)
1423 					? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1424 			break;
1425 
1426 		case OVS_NAT_ATTR_IP_MIN:
1427 			nla_memcpy(&info->range.min_addr, a,
1428 				   sizeof(info->range.min_addr));
1429 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1430 			break;
1431 
1432 		case OVS_NAT_ATTR_IP_MAX:
1433 			have_ip_max = true;
1434 			nla_memcpy(&info->range.max_addr, a,
1435 				   sizeof(info->range.max_addr));
1436 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1437 			break;
1438 
1439 		case OVS_NAT_ATTR_PROTO_MIN:
1440 			info->range.min_proto.all = htons(nla_get_u16(a));
1441 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1442 			break;
1443 
1444 		case OVS_NAT_ATTR_PROTO_MAX:
1445 			have_proto_max = true;
1446 			info->range.max_proto.all = htons(nla_get_u16(a));
1447 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1448 			break;
1449 
1450 		case OVS_NAT_ATTR_PERSISTENT:
1451 			info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1452 			break;
1453 
1454 		case OVS_NAT_ATTR_PROTO_HASH:
1455 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1456 			break;
1457 
1458 		case OVS_NAT_ATTR_PROTO_RANDOM:
1459 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1460 			break;
1461 
1462 		default:
1463 			OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1464 			return -EINVAL;
1465 		}
1466 	}
1467 
1468 	if (rem > 0) {
1469 		OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1470 		return -EINVAL;
1471 	}
1472 	if (!info->nat) {
1473 		/* Do not allow flags if no type is given. */
1474 		if (info->range.flags) {
1475 			OVS_NLERR(log,
1476 				  "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1477 				  );
1478 			return -EINVAL;
1479 		}
1480 		info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1481 	} else if (!info->commit) {
1482 		OVS_NLERR(log,
1483 			  "NAT attributes may be specified only when CT COMMIT flag is also specified."
1484 			  );
1485 		return -EINVAL;
1486 	}
1487 	/* Allow missing IP_MAX. */
1488 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1489 		memcpy(&info->range.max_addr, &info->range.min_addr,
1490 		       sizeof(info->range.max_addr));
1491 	}
1492 	/* Allow missing PROTO_MAX. */
1493 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1494 	    !have_proto_max) {
1495 		info->range.max_proto.all = info->range.min_proto.all;
1496 	}
1497 	return 0;
1498 }
1499 #endif
1500 
1501 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1502 	[OVS_CT_ATTR_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1503 	[OVS_CT_ATTR_FORCE_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1504 	[OVS_CT_ATTR_ZONE]	= { .minlen = sizeof(u16),
1505 				    .maxlen = sizeof(u16) },
1506 	[OVS_CT_ATTR_MARK]	= { .minlen = sizeof(struct md_mark),
1507 				    .maxlen = sizeof(struct md_mark) },
1508 	[OVS_CT_ATTR_LABELS]	= { .minlen = sizeof(struct md_labels),
1509 				    .maxlen = sizeof(struct md_labels) },
1510 	[OVS_CT_ATTR_HELPER]	= { .minlen = 1,
1511 				    .maxlen = NF_CT_HELPER_NAME_LEN },
1512 #if IS_ENABLED(CONFIG_NF_NAT)
1513 	/* NAT length is checked when parsing the nested attributes. */
1514 	[OVS_CT_ATTR_NAT]	= { .minlen = 0, .maxlen = INT_MAX },
1515 #endif
1516 	[OVS_CT_ATTR_EVENTMASK]	= { .minlen = sizeof(u32),
1517 				    .maxlen = sizeof(u32) },
1518 	[OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1519 				  .maxlen = CTNL_TIMEOUT_NAME_MAX },
1520 };
1521 
1522 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1523 		    const char **helper, bool log)
1524 {
1525 	struct nlattr *a;
1526 	int rem;
1527 
1528 	nla_for_each_nested(a, attr, rem) {
1529 		int type = nla_type(a);
1530 		int maxlen;
1531 		int minlen;
1532 
1533 		if (type > OVS_CT_ATTR_MAX) {
1534 			OVS_NLERR(log,
1535 				  "Unknown conntrack attr (type=%d, max=%d)",
1536 				  type, OVS_CT_ATTR_MAX);
1537 			return -EINVAL;
1538 		}
1539 
1540 		maxlen = ovs_ct_attr_lens[type].maxlen;
1541 		minlen = ovs_ct_attr_lens[type].minlen;
1542 		if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1543 			OVS_NLERR(log,
1544 				  "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1545 				  type, nla_len(a), maxlen);
1546 			return -EINVAL;
1547 		}
1548 
1549 		switch (type) {
1550 		case OVS_CT_ATTR_FORCE_COMMIT:
1551 			info->force = true;
1552 			fallthrough;
1553 		case OVS_CT_ATTR_COMMIT:
1554 			info->commit = true;
1555 			break;
1556 #ifdef CONFIG_NF_CONNTRACK_ZONES
1557 		case OVS_CT_ATTR_ZONE:
1558 			info->zone.id = nla_get_u16(a);
1559 			break;
1560 #endif
1561 #ifdef CONFIG_NF_CONNTRACK_MARK
1562 		case OVS_CT_ATTR_MARK: {
1563 			struct md_mark *mark = nla_data(a);
1564 
1565 			if (!mark->mask) {
1566 				OVS_NLERR(log, "ct_mark mask cannot be 0");
1567 				return -EINVAL;
1568 			}
1569 			info->mark = *mark;
1570 			break;
1571 		}
1572 #endif
1573 #ifdef CONFIG_NF_CONNTRACK_LABELS
1574 		case OVS_CT_ATTR_LABELS: {
1575 			struct md_labels *labels = nla_data(a);
1576 
1577 			if (!labels_nonzero(&labels->mask)) {
1578 				OVS_NLERR(log, "ct_labels mask cannot be 0");
1579 				return -EINVAL;
1580 			}
1581 			info->labels = *labels;
1582 			break;
1583 		}
1584 #endif
1585 		case OVS_CT_ATTR_HELPER:
1586 			*helper = nla_data(a);
1587 			if (!memchr(*helper, '\0', nla_len(a))) {
1588 				OVS_NLERR(log, "Invalid conntrack helper");
1589 				return -EINVAL;
1590 			}
1591 			break;
1592 #if IS_ENABLED(CONFIG_NF_NAT)
1593 		case OVS_CT_ATTR_NAT: {
1594 			int err = parse_nat(a, info, log);
1595 
1596 			if (err)
1597 				return err;
1598 			break;
1599 		}
1600 #endif
1601 		case OVS_CT_ATTR_EVENTMASK:
1602 			info->have_eventmask = true;
1603 			info->eventmask = nla_get_u32(a);
1604 			break;
1605 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1606 		case OVS_CT_ATTR_TIMEOUT:
1607 			memcpy(info->timeout, nla_data(a), nla_len(a));
1608 			if (!memchr(info->timeout, '\0', nla_len(a))) {
1609 				OVS_NLERR(log, "Invalid conntrack timeout");
1610 				return -EINVAL;
1611 			}
1612 			break;
1613 #endif
1614 
1615 		default:
1616 			OVS_NLERR(log, "Unknown conntrack attr (%d)",
1617 				  type);
1618 			return -EINVAL;
1619 		}
1620 	}
1621 
1622 #ifdef CONFIG_NF_CONNTRACK_MARK
1623 	if (!info->commit && info->mark.mask) {
1624 		OVS_NLERR(log,
1625 			  "Setting conntrack mark requires 'commit' flag.");
1626 		return -EINVAL;
1627 	}
1628 #endif
1629 #ifdef CONFIG_NF_CONNTRACK_LABELS
1630 	if (!info->commit && labels_nonzero(&info->labels.mask)) {
1631 		OVS_NLERR(log,
1632 			  "Setting conntrack labels requires 'commit' flag.");
1633 		return -EINVAL;
1634 	}
1635 #endif
1636 	if (rem > 0) {
1637 		OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1638 		return -EINVAL;
1639 	}
1640 
1641 	return 0;
1642 }
1643 
1644 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1645 {
1646 	if (attr == OVS_KEY_ATTR_CT_STATE)
1647 		return true;
1648 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1649 	    attr == OVS_KEY_ATTR_CT_ZONE)
1650 		return true;
1651 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1652 	    attr == OVS_KEY_ATTR_CT_MARK)
1653 		return true;
1654 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1655 	    attr == OVS_KEY_ATTR_CT_LABELS) {
1656 		struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1657 
1658 		return ovs_net->xt_label;
1659 	}
1660 
1661 	return false;
1662 }
1663 
1664 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1665 		       const struct sw_flow_key *key,
1666 		       struct sw_flow_actions **sfa,  bool log)
1667 {
1668 	struct ovs_conntrack_info ct_info;
1669 	const char *helper = NULL;
1670 	u16 family;
1671 	int err;
1672 
1673 	family = key_to_nfproto(key);
1674 	if (family == NFPROTO_UNSPEC) {
1675 		OVS_NLERR(log, "ct family unspecified");
1676 		return -EINVAL;
1677 	}
1678 
1679 	memset(&ct_info, 0, sizeof(ct_info));
1680 	ct_info.family = family;
1681 
1682 	nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1683 			NF_CT_DEFAULT_ZONE_DIR, 0);
1684 
1685 	err = parse_ct(attr, &ct_info, &helper, log);
1686 	if (err)
1687 		return err;
1688 
1689 	/* Set up template for tracking connections in specific zones. */
1690 	ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1691 	if (!ct_info.ct) {
1692 		OVS_NLERR(log, "Failed to allocate conntrack template");
1693 		return -ENOMEM;
1694 	}
1695 
1696 	if (ct_info.timeout[0]) {
1697 		if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1698 				      ct_info.timeout))
1699 			pr_info_ratelimited("Failed to associated timeout "
1700 					    "policy `%s'\n", ct_info.timeout);
1701 		else
1702 			ct_info.nf_ct_timeout = rcu_dereference(
1703 				nf_ct_timeout_find(ct_info.ct)->timeout);
1704 
1705 	}
1706 
1707 	if (helper) {
1708 		err = ovs_ct_add_helper(&ct_info, helper, key, log);
1709 		if (err)
1710 			goto err_free_ct;
1711 	}
1712 
1713 	err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1714 				 sizeof(ct_info), log);
1715 	if (err)
1716 		goto err_free_ct;
1717 
1718 	__set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1719 	nf_conntrack_get(&ct_info.ct->ct_general);
1720 	return 0;
1721 err_free_ct:
1722 	__ovs_ct_free_action(&ct_info);
1723 	return err;
1724 }
1725 
1726 #if IS_ENABLED(CONFIG_NF_NAT)
1727 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1728 			       struct sk_buff *skb)
1729 {
1730 	struct nlattr *start;
1731 
1732 	start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1733 	if (!start)
1734 		return false;
1735 
1736 	if (info->nat & OVS_CT_SRC_NAT) {
1737 		if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1738 			return false;
1739 	} else if (info->nat & OVS_CT_DST_NAT) {
1740 		if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1741 			return false;
1742 	} else {
1743 		goto out;
1744 	}
1745 
1746 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1747 		if (IS_ENABLED(CONFIG_NF_NAT) &&
1748 		    info->family == NFPROTO_IPV4) {
1749 			if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1750 					    info->range.min_addr.ip) ||
1751 			    (info->range.max_addr.ip
1752 			     != info->range.min_addr.ip &&
1753 			     (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1754 					      info->range.max_addr.ip))))
1755 				return false;
1756 		} else if (IS_ENABLED(CONFIG_IPV6) &&
1757 			   info->family == NFPROTO_IPV6) {
1758 			if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1759 					     &info->range.min_addr.in6) ||
1760 			    (memcmp(&info->range.max_addr.in6,
1761 				    &info->range.min_addr.in6,
1762 				    sizeof(info->range.max_addr.in6)) &&
1763 			     (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1764 					       &info->range.max_addr.in6))))
1765 				return false;
1766 		} else {
1767 			return false;
1768 		}
1769 	}
1770 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1771 	    (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1772 			 ntohs(info->range.min_proto.all)) ||
1773 	     (info->range.max_proto.all != info->range.min_proto.all &&
1774 	      nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1775 			  ntohs(info->range.max_proto.all)))))
1776 		return false;
1777 
1778 	if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1779 	    nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1780 		return false;
1781 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1782 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1783 		return false;
1784 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1785 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1786 		return false;
1787 out:
1788 	nla_nest_end(skb, start);
1789 
1790 	return true;
1791 }
1792 #endif
1793 
1794 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1795 			  struct sk_buff *skb)
1796 {
1797 	struct nlattr *start;
1798 
1799 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1800 	if (!start)
1801 		return -EMSGSIZE;
1802 
1803 	if (ct_info->commit && nla_put_flag(skb, ct_info->force
1804 					    ? OVS_CT_ATTR_FORCE_COMMIT
1805 					    : OVS_CT_ATTR_COMMIT))
1806 		return -EMSGSIZE;
1807 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1808 	    nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1809 		return -EMSGSIZE;
1810 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1811 	    nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1812 		    &ct_info->mark))
1813 		return -EMSGSIZE;
1814 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1815 	    labels_nonzero(&ct_info->labels.mask) &&
1816 	    nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1817 		    &ct_info->labels))
1818 		return -EMSGSIZE;
1819 	if (ct_info->helper) {
1820 		if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1821 				   ct_info->helper->name))
1822 			return -EMSGSIZE;
1823 	}
1824 	if (ct_info->have_eventmask &&
1825 	    nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1826 		return -EMSGSIZE;
1827 	if (ct_info->timeout[0]) {
1828 		if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1829 			return -EMSGSIZE;
1830 	}
1831 
1832 #if IS_ENABLED(CONFIG_NF_NAT)
1833 	if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1834 		return -EMSGSIZE;
1835 #endif
1836 	nla_nest_end(skb, start);
1837 
1838 	return 0;
1839 }
1840 
1841 void ovs_ct_free_action(const struct nlattr *a)
1842 {
1843 	struct ovs_conntrack_info *ct_info = nla_data(a);
1844 
1845 	__ovs_ct_free_action(ct_info);
1846 }
1847 
1848 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1849 {
1850 	if (ct_info->helper) {
1851 #if IS_ENABLED(CONFIG_NF_NAT)
1852 		if (ct_info->nat)
1853 			nf_nat_helper_put(ct_info->helper);
1854 #endif
1855 		nf_conntrack_helper_put(ct_info->helper);
1856 	}
1857 	if (ct_info->ct) {
1858 		if (ct_info->timeout[0])
1859 			nf_ct_destroy_timeout(ct_info->ct);
1860 		nf_ct_tmpl_free(ct_info->ct);
1861 	}
1862 }
1863 
1864 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1865 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1866 {
1867 	int i, err;
1868 
1869 	ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1870 					 GFP_KERNEL);
1871 	if (!ovs_net->ct_limit_info)
1872 		return -ENOMEM;
1873 
1874 	ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1875 	ovs_net->ct_limit_info->limits =
1876 		kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1877 			      GFP_KERNEL);
1878 	if (!ovs_net->ct_limit_info->limits) {
1879 		kfree(ovs_net->ct_limit_info);
1880 		return -ENOMEM;
1881 	}
1882 
1883 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1884 		INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1885 
1886 	ovs_net->ct_limit_info->data =
1887 		nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1888 
1889 	if (IS_ERR(ovs_net->ct_limit_info->data)) {
1890 		err = PTR_ERR(ovs_net->ct_limit_info->data);
1891 		kfree(ovs_net->ct_limit_info->limits);
1892 		kfree(ovs_net->ct_limit_info);
1893 		pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1894 		return err;
1895 	}
1896 	return 0;
1897 }
1898 
1899 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1900 {
1901 	const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1902 	int i;
1903 
1904 	nf_conncount_destroy(net, NFPROTO_INET, info->data);
1905 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1906 		struct hlist_head *head = &info->limits[i];
1907 		struct ovs_ct_limit *ct_limit;
1908 
1909 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node,
1910 					 lockdep_ovsl_is_held())
1911 			kfree_rcu(ct_limit, rcu);
1912 	}
1913 	kfree(info->limits);
1914 	kfree(info);
1915 }
1916 
1917 static struct sk_buff *
1918 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1919 			     struct ovs_header **ovs_reply_header)
1920 {
1921 	struct ovs_header *ovs_header = info->userhdr;
1922 	struct sk_buff *skb;
1923 
1924 	skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1925 	if (!skb)
1926 		return ERR_PTR(-ENOMEM);
1927 
1928 	*ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1929 					info->snd_seq,
1930 					&dp_ct_limit_genl_family, 0, cmd);
1931 
1932 	if (!*ovs_reply_header) {
1933 		nlmsg_free(skb);
1934 		return ERR_PTR(-EMSGSIZE);
1935 	}
1936 	(*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1937 
1938 	return skb;
1939 }
1940 
1941 static bool check_zone_id(int zone_id, u16 *pzone)
1942 {
1943 	if (zone_id >= 0 && zone_id <= 65535) {
1944 		*pzone = (u16)zone_id;
1945 		return true;
1946 	}
1947 	return false;
1948 }
1949 
1950 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1951 				       struct ovs_ct_limit_info *info)
1952 {
1953 	struct ovs_zone_limit *zone_limit;
1954 	int rem;
1955 	u16 zone;
1956 
1957 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1958 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1959 
1960 	while (rem >= sizeof(*zone_limit)) {
1961 		if (unlikely(zone_limit->zone_id ==
1962 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1963 			ovs_lock();
1964 			info->default_limit = zone_limit->limit;
1965 			ovs_unlock();
1966 		} else if (unlikely(!check_zone_id(
1967 				zone_limit->zone_id, &zone))) {
1968 			OVS_NLERR(true, "zone id is out of range");
1969 		} else {
1970 			struct ovs_ct_limit *ct_limit;
1971 
1972 			ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1973 			if (!ct_limit)
1974 				return -ENOMEM;
1975 
1976 			ct_limit->zone = zone;
1977 			ct_limit->limit = zone_limit->limit;
1978 
1979 			ovs_lock();
1980 			ct_limit_set(info, ct_limit);
1981 			ovs_unlock();
1982 		}
1983 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1984 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1985 				NLA_ALIGN(sizeof(*zone_limit)));
1986 	}
1987 
1988 	if (rem)
1989 		OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1990 
1991 	return 0;
1992 }
1993 
1994 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1995 				       struct ovs_ct_limit_info *info)
1996 {
1997 	struct ovs_zone_limit *zone_limit;
1998 	int rem;
1999 	u16 zone;
2000 
2001 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
2002 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2003 
2004 	while (rem >= sizeof(*zone_limit)) {
2005 		if (unlikely(zone_limit->zone_id ==
2006 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2007 			ovs_lock();
2008 			info->default_limit = OVS_CT_LIMIT_DEFAULT;
2009 			ovs_unlock();
2010 		} else if (unlikely(!check_zone_id(
2011 				zone_limit->zone_id, &zone))) {
2012 			OVS_NLERR(true, "zone id is out of range");
2013 		} else {
2014 			ovs_lock();
2015 			ct_limit_del(info, zone);
2016 			ovs_unlock();
2017 		}
2018 		rem -= NLA_ALIGN(sizeof(*zone_limit));
2019 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2020 				NLA_ALIGN(sizeof(*zone_limit)));
2021 	}
2022 
2023 	if (rem)
2024 		OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
2025 
2026 	return 0;
2027 }
2028 
2029 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2030 					  struct sk_buff *reply)
2031 {
2032 	struct ovs_zone_limit zone_limit = {
2033 		.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE,
2034 		.limit   = info->default_limit,
2035 	};
2036 
2037 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2038 }
2039 
2040 static int __ovs_ct_limit_get_zone_limit(struct net *net,
2041 					 struct nf_conncount_data *data,
2042 					 u16 zone_id, u32 limit,
2043 					 struct sk_buff *reply)
2044 {
2045 	struct nf_conntrack_zone ct_zone;
2046 	struct ovs_zone_limit zone_limit;
2047 	u32 conncount_key = zone_id;
2048 
2049 	zone_limit.zone_id = zone_id;
2050 	zone_limit.limit = limit;
2051 	nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2052 
2053 	zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2054 					      &ct_zone);
2055 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2056 }
2057 
2058 static int ovs_ct_limit_get_zone_limit(struct net *net,
2059 				       struct nlattr *nla_zone_limit,
2060 				       struct ovs_ct_limit_info *info,
2061 				       struct sk_buff *reply)
2062 {
2063 	struct ovs_zone_limit *zone_limit;
2064 	int rem, err;
2065 	u32 limit;
2066 	u16 zone;
2067 
2068 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
2069 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2070 
2071 	while (rem >= sizeof(*zone_limit)) {
2072 		if (unlikely(zone_limit->zone_id ==
2073 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2074 			err = ovs_ct_limit_get_default_limit(info, reply);
2075 			if (err)
2076 				return err;
2077 		} else if (unlikely(!check_zone_id(zone_limit->zone_id,
2078 							&zone))) {
2079 			OVS_NLERR(true, "zone id is out of range");
2080 		} else {
2081 			rcu_read_lock();
2082 			limit = ct_limit_get(info, zone);
2083 			rcu_read_unlock();
2084 
2085 			err = __ovs_ct_limit_get_zone_limit(
2086 				net, info->data, zone, limit, reply);
2087 			if (err)
2088 				return err;
2089 		}
2090 		rem -= NLA_ALIGN(sizeof(*zone_limit));
2091 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2092 				NLA_ALIGN(sizeof(*zone_limit)));
2093 	}
2094 
2095 	if (rem)
2096 		OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2097 
2098 	return 0;
2099 }
2100 
2101 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2102 					   struct ovs_ct_limit_info *info,
2103 					   struct sk_buff *reply)
2104 {
2105 	struct ovs_ct_limit *ct_limit;
2106 	struct hlist_head *head;
2107 	int i, err = 0;
2108 
2109 	err = ovs_ct_limit_get_default_limit(info, reply);
2110 	if (err)
2111 		return err;
2112 
2113 	rcu_read_lock();
2114 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2115 		head = &info->limits[i];
2116 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2117 			err = __ovs_ct_limit_get_zone_limit(net, info->data,
2118 				ct_limit->zone, ct_limit->limit, reply);
2119 			if (err)
2120 				goto exit_err;
2121 		}
2122 	}
2123 
2124 exit_err:
2125 	rcu_read_unlock();
2126 	return err;
2127 }
2128 
2129 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2130 {
2131 	struct nlattr **a = info->attrs;
2132 	struct sk_buff *reply;
2133 	struct ovs_header *ovs_reply_header;
2134 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2135 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2136 	int err;
2137 
2138 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2139 					     &ovs_reply_header);
2140 	if (IS_ERR(reply))
2141 		return PTR_ERR(reply);
2142 
2143 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2144 		err = -EINVAL;
2145 		goto exit_err;
2146 	}
2147 
2148 	err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2149 					  ct_limit_info);
2150 	if (err)
2151 		goto exit_err;
2152 
2153 	static_branch_enable(&ovs_ct_limit_enabled);
2154 
2155 	genlmsg_end(reply, ovs_reply_header);
2156 	return genlmsg_reply(reply, info);
2157 
2158 exit_err:
2159 	nlmsg_free(reply);
2160 	return err;
2161 }
2162 
2163 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2164 {
2165 	struct nlattr **a = info->attrs;
2166 	struct sk_buff *reply;
2167 	struct ovs_header *ovs_reply_header;
2168 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2169 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2170 	int err;
2171 
2172 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2173 					     &ovs_reply_header);
2174 	if (IS_ERR(reply))
2175 		return PTR_ERR(reply);
2176 
2177 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2178 		err = -EINVAL;
2179 		goto exit_err;
2180 	}
2181 
2182 	err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2183 					  ct_limit_info);
2184 	if (err)
2185 		goto exit_err;
2186 
2187 	genlmsg_end(reply, ovs_reply_header);
2188 	return genlmsg_reply(reply, info);
2189 
2190 exit_err:
2191 	nlmsg_free(reply);
2192 	return err;
2193 }
2194 
2195 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2196 {
2197 	struct nlattr **a = info->attrs;
2198 	struct nlattr *nla_reply;
2199 	struct sk_buff *reply;
2200 	struct ovs_header *ovs_reply_header;
2201 	struct net *net = sock_net(skb->sk);
2202 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2203 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2204 	int err;
2205 
2206 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2207 					     &ovs_reply_header);
2208 	if (IS_ERR(reply))
2209 		return PTR_ERR(reply);
2210 
2211 	nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2212 	if (!nla_reply) {
2213 		err = -EMSGSIZE;
2214 		goto exit_err;
2215 	}
2216 
2217 	if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2218 		err = ovs_ct_limit_get_zone_limit(
2219 			net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2220 			reply);
2221 		if (err)
2222 			goto exit_err;
2223 	} else {
2224 		err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2225 						      reply);
2226 		if (err)
2227 			goto exit_err;
2228 	}
2229 
2230 	nla_nest_end(reply, nla_reply);
2231 	genlmsg_end(reply, ovs_reply_header);
2232 	return genlmsg_reply(reply, info);
2233 
2234 exit_err:
2235 	nlmsg_free(reply);
2236 	return err;
2237 }
2238 
2239 static const struct genl_small_ops ct_limit_genl_ops[] = {
2240 	{ .cmd = OVS_CT_LIMIT_CMD_SET,
2241 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2242 		.flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2243 					   * privilege. */
2244 		.doit = ovs_ct_limit_cmd_set,
2245 	},
2246 	{ .cmd = OVS_CT_LIMIT_CMD_DEL,
2247 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2248 		.flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2249 					   * privilege. */
2250 		.doit = ovs_ct_limit_cmd_del,
2251 	},
2252 	{ .cmd = OVS_CT_LIMIT_CMD_GET,
2253 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2254 		.flags = 0,		  /* OK for unprivileged users. */
2255 		.doit = ovs_ct_limit_cmd_get,
2256 	},
2257 };
2258 
2259 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2260 	.name = OVS_CT_LIMIT_MCGROUP,
2261 };
2262 
2263 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2264 	.hdrsize = sizeof(struct ovs_header),
2265 	.name = OVS_CT_LIMIT_FAMILY,
2266 	.version = OVS_CT_LIMIT_VERSION,
2267 	.maxattr = OVS_CT_LIMIT_ATTR_MAX,
2268 	.policy = ct_limit_policy,
2269 	.netnsok = true,
2270 	.parallel_ops = true,
2271 	.small_ops = ct_limit_genl_ops,
2272 	.n_small_ops = ARRAY_SIZE(ct_limit_genl_ops),
2273 	.mcgrps = &ovs_ct_limit_multicast_group,
2274 	.n_mcgrps = 1,
2275 	.module = THIS_MODULE,
2276 };
2277 #endif
2278 
2279 int ovs_ct_init(struct net *net)
2280 {
2281 	unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2282 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2283 
2284 	if (nf_connlabels_get(net, n_bits - 1)) {
2285 		ovs_net->xt_label = false;
2286 		OVS_NLERR(true, "Failed to set connlabel length");
2287 	} else {
2288 		ovs_net->xt_label = true;
2289 	}
2290 
2291 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2292 	return ovs_ct_limit_init(net, ovs_net);
2293 #else
2294 	return 0;
2295 #endif
2296 }
2297 
2298 void ovs_ct_exit(struct net *net)
2299 {
2300 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2301 
2302 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2303 	ovs_ct_limit_exit(net, ovs_net);
2304 #endif
2305 
2306 	if (ovs_net->xt_label)
2307 		nf_connlabels_put(net);
2308 }
2309