xref: /openbmc/linux/net/openvswitch/conntrack.c (revision 8f8d5745bb520c76b81abef4a2cb3023d0313bfd)
1 /*
2  * Copyright (c) 2015 Nicira, Inc.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  */
13 
14 #include <linux/module.h>
15 #include <linux/openvswitch.h>
16 #include <linux/tcp.h>
17 #include <linux/udp.h>
18 #include <linux/sctp.h>
19 #include <linux/static_key.h>
20 #include <net/ip.h>
21 #include <net/genetlink.h>
22 #include <net/netfilter/nf_conntrack_core.h>
23 #include <net/netfilter/nf_conntrack_count.h>
24 #include <net/netfilter/nf_conntrack_helper.h>
25 #include <net/netfilter/nf_conntrack_labels.h>
26 #include <net/netfilter/nf_conntrack_seqadj.h>
27 #include <net/netfilter/nf_conntrack_zones.h>
28 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
29 #include <net/ipv6_frag.h>
30 
31 #ifdef CONFIG_NF_NAT_NEEDED
32 #include <net/netfilter/nf_nat.h>
33 #endif
34 
35 #include "datapath.h"
36 #include "conntrack.h"
37 #include "flow.h"
38 #include "flow_netlink.h"
39 
40 struct ovs_ct_len_tbl {
41 	int maxlen;
42 	int minlen;
43 };
44 
45 /* Metadata mark for masked write to conntrack mark */
46 struct md_mark {
47 	u32 value;
48 	u32 mask;
49 };
50 
51 /* Metadata label for masked write to conntrack label. */
52 struct md_labels {
53 	struct ovs_key_ct_labels value;
54 	struct ovs_key_ct_labels mask;
55 };
56 
57 enum ovs_ct_nat {
58 	OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
59 	OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
60 	OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
61 };
62 
63 /* Conntrack action context for execution. */
64 struct ovs_conntrack_info {
65 	struct nf_conntrack_helper *helper;
66 	struct nf_conntrack_zone zone;
67 	struct nf_conn *ct;
68 	u8 commit : 1;
69 	u8 nat : 3;                 /* enum ovs_ct_nat */
70 	u8 force : 1;
71 	u8 have_eventmask : 1;
72 	u16 family;
73 	u32 eventmask;              /* Mask of 1 << IPCT_*. */
74 	struct md_mark mark;
75 	struct md_labels labels;
76 #ifdef CONFIG_NF_NAT_NEEDED
77 	struct nf_nat_range2 range;  /* Only present for SRC NAT and DST NAT. */
78 #endif
79 };
80 
81 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
82 #define OVS_CT_LIMIT_UNLIMITED	0
83 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
84 #define CT_LIMIT_HASH_BUCKETS 512
85 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
86 
87 struct ovs_ct_limit {
88 	/* Elements in ovs_ct_limit_info->limits hash table */
89 	struct hlist_node hlist_node;
90 	struct rcu_head rcu;
91 	u16 zone;
92 	u32 limit;
93 };
94 
95 struct ovs_ct_limit_info {
96 	u32 default_limit;
97 	struct hlist_head *limits;
98 	struct nf_conncount_data *data;
99 };
100 
101 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
102 	[OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
103 };
104 #endif
105 
106 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
107 
108 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
109 
110 static u16 key_to_nfproto(const struct sw_flow_key *key)
111 {
112 	switch (ntohs(key->eth.type)) {
113 	case ETH_P_IP:
114 		return NFPROTO_IPV4;
115 	case ETH_P_IPV6:
116 		return NFPROTO_IPV6;
117 	default:
118 		return NFPROTO_UNSPEC;
119 	}
120 }
121 
122 /* Map SKB connection state into the values used by flow definition. */
123 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
124 {
125 	u8 ct_state = OVS_CS_F_TRACKED;
126 
127 	switch (ctinfo) {
128 	case IP_CT_ESTABLISHED_REPLY:
129 	case IP_CT_RELATED_REPLY:
130 		ct_state |= OVS_CS_F_REPLY_DIR;
131 		break;
132 	default:
133 		break;
134 	}
135 
136 	switch (ctinfo) {
137 	case IP_CT_ESTABLISHED:
138 	case IP_CT_ESTABLISHED_REPLY:
139 		ct_state |= OVS_CS_F_ESTABLISHED;
140 		break;
141 	case IP_CT_RELATED:
142 	case IP_CT_RELATED_REPLY:
143 		ct_state |= OVS_CS_F_RELATED;
144 		break;
145 	case IP_CT_NEW:
146 		ct_state |= OVS_CS_F_NEW;
147 		break;
148 	default:
149 		break;
150 	}
151 
152 	return ct_state;
153 }
154 
155 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
156 {
157 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
158 	return ct ? ct->mark : 0;
159 #else
160 	return 0;
161 #endif
162 }
163 
164 /* Guard against conntrack labels max size shrinking below 128 bits. */
165 #if NF_CT_LABELS_MAX_SIZE < 16
166 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
167 #endif
168 
169 static void ovs_ct_get_labels(const struct nf_conn *ct,
170 			      struct ovs_key_ct_labels *labels)
171 {
172 	struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
173 
174 	if (cl)
175 		memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
176 	else
177 		memset(labels, 0, OVS_CT_LABELS_LEN);
178 }
179 
180 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
181 					const struct nf_conntrack_tuple *orig,
182 					u8 icmp_proto)
183 {
184 	key->ct_orig_proto = orig->dst.protonum;
185 	if (orig->dst.protonum == icmp_proto) {
186 		key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
187 		key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
188 	} else {
189 		key->ct.orig_tp.src = orig->src.u.all;
190 		key->ct.orig_tp.dst = orig->dst.u.all;
191 	}
192 }
193 
194 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
195 				const struct nf_conntrack_zone *zone,
196 				const struct nf_conn *ct)
197 {
198 	key->ct_state = state;
199 	key->ct_zone = zone->id;
200 	key->ct.mark = ovs_ct_get_mark(ct);
201 	ovs_ct_get_labels(ct, &key->ct.labels);
202 
203 	if (ct) {
204 		const struct nf_conntrack_tuple *orig;
205 
206 		/* Use the master if we have one. */
207 		if (ct->master)
208 			ct = ct->master;
209 		orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
210 
211 		/* IP version must match with the master connection. */
212 		if (key->eth.type == htons(ETH_P_IP) &&
213 		    nf_ct_l3num(ct) == NFPROTO_IPV4) {
214 			key->ipv4.ct_orig.src = orig->src.u3.ip;
215 			key->ipv4.ct_orig.dst = orig->dst.u3.ip;
216 			__ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
217 			return;
218 		} else if (key->eth.type == htons(ETH_P_IPV6) &&
219 			   !sw_flow_key_is_nd(key) &&
220 			   nf_ct_l3num(ct) == NFPROTO_IPV6) {
221 			key->ipv6.ct_orig.src = orig->src.u3.in6;
222 			key->ipv6.ct_orig.dst = orig->dst.u3.in6;
223 			__ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
224 			return;
225 		}
226 	}
227 	/* Clear 'ct_orig_proto' to mark the non-existence of conntrack
228 	 * original direction key fields.
229 	 */
230 	key->ct_orig_proto = 0;
231 }
232 
233 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
234  * previously sent the packet to conntrack via the ct action.  If
235  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
236  * initialized from the connection status.
237  */
238 static void ovs_ct_update_key(const struct sk_buff *skb,
239 			      const struct ovs_conntrack_info *info,
240 			      struct sw_flow_key *key, bool post_ct,
241 			      bool keep_nat_flags)
242 {
243 	const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
244 	enum ip_conntrack_info ctinfo;
245 	struct nf_conn *ct;
246 	u8 state = 0;
247 
248 	ct = nf_ct_get(skb, &ctinfo);
249 	if (ct) {
250 		state = ovs_ct_get_state(ctinfo);
251 		/* All unconfirmed entries are NEW connections. */
252 		if (!nf_ct_is_confirmed(ct))
253 			state |= OVS_CS_F_NEW;
254 		/* OVS persists the related flag for the duration of the
255 		 * connection.
256 		 */
257 		if (ct->master)
258 			state |= OVS_CS_F_RELATED;
259 		if (keep_nat_flags) {
260 			state |= key->ct_state & OVS_CS_F_NAT_MASK;
261 		} else {
262 			if (ct->status & IPS_SRC_NAT)
263 				state |= OVS_CS_F_SRC_NAT;
264 			if (ct->status & IPS_DST_NAT)
265 				state |= OVS_CS_F_DST_NAT;
266 		}
267 		zone = nf_ct_zone(ct);
268 	} else if (post_ct) {
269 		state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
270 		if (info)
271 			zone = &info->zone;
272 	}
273 	__ovs_ct_update_key(key, state, zone, ct);
274 }
275 
276 /* This is called to initialize CT key fields possibly coming in from the local
277  * stack.
278  */
279 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
280 {
281 	ovs_ct_update_key(skb, NULL, key, false, false);
282 }
283 
284 #define IN6_ADDR_INITIALIZER(ADDR) \
285 	{ (ADDR).s6_addr32[0], (ADDR).s6_addr32[1], \
286 	  (ADDR).s6_addr32[2], (ADDR).s6_addr32[3] }
287 
288 int ovs_ct_put_key(const struct sw_flow_key *swkey,
289 		   const struct sw_flow_key *output, struct sk_buff *skb)
290 {
291 	if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
292 		return -EMSGSIZE;
293 
294 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
295 	    nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
296 		return -EMSGSIZE;
297 
298 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
299 	    nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
300 		return -EMSGSIZE;
301 
302 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
303 	    nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
304 		    &output->ct.labels))
305 		return -EMSGSIZE;
306 
307 	if (swkey->ct_orig_proto) {
308 		if (swkey->eth.type == htons(ETH_P_IP)) {
309 			struct ovs_key_ct_tuple_ipv4 orig = {
310 				output->ipv4.ct_orig.src,
311 				output->ipv4.ct_orig.dst,
312 				output->ct.orig_tp.src,
313 				output->ct.orig_tp.dst,
314 				output->ct_orig_proto,
315 			};
316 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
317 				    sizeof(orig), &orig))
318 				return -EMSGSIZE;
319 		} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
320 			struct ovs_key_ct_tuple_ipv6 orig = {
321 				IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.src),
322 				IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.dst),
323 				output->ct.orig_tp.src,
324 				output->ct.orig_tp.dst,
325 				output->ct_orig_proto,
326 			};
327 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
328 				    sizeof(orig), &orig))
329 				return -EMSGSIZE;
330 		}
331 	}
332 
333 	return 0;
334 }
335 
336 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
337 			   u32 ct_mark, u32 mask)
338 {
339 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
340 	u32 new_mark;
341 
342 	new_mark = ct_mark | (ct->mark & ~(mask));
343 	if (ct->mark != new_mark) {
344 		ct->mark = new_mark;
345 		if (nf_ct_is_confirmed(ct))
346 			nf_conntrack_event_cache(IPCT_MARK, ct);
347 		key->ct.mark = new_mark;
348 	}
349 
350 	return 0;
351 #else
352 	return -ENOTSUPP;
353 #endif
354 }
355 
356 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
357 {
358 	struct nf_conn_labels *cl;
359 
360 	cl = nf_ct_labels_find(ct);
361 	if (!cl) {
362 		nf_ct_labels_ext_add(ct);
363 		cl = nf_ct_labels_find(ct);
364 	}
365 
366 	return cl;
367 }
368 
369 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
370  * since the new connection is not yet confirmed, and thus no-one else has
371  * access to it's labels, we simply write them over.
372  */
373 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
374 			      const struct ovs_key_ct_labels *labels,
375 			      const struct ovs_key_ct_labels *mask)
376 {
377 	struct nf_conn_labels *cl, *master_cl;
378 	bool have_mask = labels_nonzero(mask);
379 
380 	/* Inherit master's labels to the related connection? */
381 	master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
382 
383 	if (!master_cl && !have_mask)
384 		return 0;   /* Nothing to do. */
385 
386 	cl = ovs_ct_get_conn_labels(ct);
387 	if (!cl)
388 		return -ENOSPC;
389 
390 	/* Inherit the master's labels, if any. */
391 	if (master_cl)
392 		*cl = *master_cl;
393 
394 	if (have_mask) {
395 		u32 *dst = (u32 *)cl->bits;
396 		int i;
397 
398 		for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
399 			dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
400 				(labels->ct_labels_32[i]
401 				 & mask->ct_labels_32[i]);
402 	}
403 
404 	/* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
405 	 * IPCT_LABEL bit is set in the event cache.
406 	 */
407 	nf_conntrack_event_cache(IPCT_LABEL, ct);
408 
409 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
410 
411 	return 0;
412 }
413 
414 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
415 			     const struct ovs_key_ct_labels *labels,
416 			     const struct ovs_key_ct_labels *mask)
417 {
418 	struct nf_conn_labels *cl;
419 	int err;
420 
421 	cl = ovs_ct_get_conn_labels(ct);
422 	if (!cl)
423 		return -ENOSPC;
424 
425 	err = nf_connlabels_replace(ct, labels->ct_labels_32,
426 				    mask->ct_labels_32,
427 				    OVS_CT_LABELS_LEN_32);
428 	if (err)
429 		return err;
430 
431 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
432 
433 	return 0;
434 }
435 
436 /* 'skb' should already be pulled to nh_ofs. */
437 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
438 {
439 	const struct nf_conntrack_helper *helper;
440 	const struct nf_conn_help *help;
441 	enum ip_conntrack_info ctinfo;
442 	unsigned int protoff;
443 	struct nf_conn *ct;
444 	int err;
445 
446 	ct = nf_ct_get(skb, &ctinfo);
447 	if (!ct || ctinfo == IP_CT_RELATED_REPLY)
448 		return NF_ACCEPT;
449 
450 	help = nfct_help(ct);
451 	if (!help)
452 		return NF_ACCEPT;
453 
454 	helper = rcu_dereference(help->helper);
455 	if (!helper)
456 		return NF_ACCEPT;
457 
458 	switch (proto) {
459 	case NFPROTO_IPV4:
460 		protoff = ip_hdrlen(skb);
461 		break;
462 	case NFPROTO_IPV6: {
463 		u8 nexthdr = ipv6_hdr(skb)->nexthdr;
464 		__be16 frag_off;
465 		int ofs;
466 
467 		ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
468 				       &frag_off);
469 		if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
470 			pr_debug("proto header not found\n");
471 			return NF_ACCEPT;
472 		}
473 		protoff = ofs;
474 		break;
475 	}
476 	default:
477 		WARN_ONCE(1, "helper invoked on non-IP family!");
478 		return NF_DROP;
479 	}
480 
481 	err = helper->help(skb, protoff, ct, ctinfo);
482 	if (err != NF_ACCEPT)
483 		return err;
484 
485 	/* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
486 	 * FTP with NAT) adusting the TCP payload size when mangling IP
487 	 * addresses and/or port numbers in the text-based control connection.
488 	 */
489 	if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
490 	    !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
491 		return NF_DROP;
492 	return NF_ACCEPT;
493 }
494 
495 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
496  * value if 'skb' is freed.
497  */
498 static int handle_fragments(struct net *net, struct sw_flow_key *key,
499 			    u16 zone, struct sk_buff *skb)
500 {
501 	struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
502 	int err;
503 
504 	if (key->eth.type == htons(ETH_P_IP)) {
505 		enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
506 
507 		memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
508 		err = ip_defrag(net, skb, user);
509 		if (err)
510 			return err;
511 
512 		ovs_cb.mru = IPCB(skb)->frag_max_size;
513 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
514 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
515 		enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
516 
517 		memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
518 		err = nf_ct_frag6_gather(net, skb, user);
519 		if (err) {
520 			if (err != -EINPROGRESS)
521 				kfree_skb(skb);
522 			return err;
523 		}
524 
525 		key->ip.proto = ipv6_hdr(skb)->nexthdr;
526 		ovs_cb.mru = IP6CB(skb)->frag_max_size;
527 #endif
528 	} else {
529 		kfree_skb(skb);
530 		return -EPFNOSUPPORT;
531 	}
532 
533 	key->ip.frag = OVS_FRAG_TYPE_NONE;
534 	skb_clear_hash(skb);
535 	skb->ignore_df = 1;
536 	*OVS_CB(skb) = ovs_cb;
537 
538 	return 0;
539 }
540 
541 static struct nf_conntrack_expect *
542 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
543 		   u16 proto, const struct sk_buff *skb)
544 {
545 	struct nf_conntrack_tuple tuple;
546 	struct nf_conntrack_expect *exp;
547 
548 	if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
549 		return NULL;
550 
551 	exp = __nf_ct_expect_find(net, zone, &tuple);
552 	if (exp) {
553 		struct nf_conntrack_tuple_hash *h;
554 
555 		/* Delete existing conntrack entry, if it clashes with the
556 		 * expectation.  This can happen since conntrack ALGs do not
557 		 * check for clashes between (new) expectations and existing
558 		 * conntrack entries.  nf_conntrack_in() will check the
559 		 * expectations only if a conntrack entry can not be found,
560 		 * which can lead to OVS finding the expectation (here) in the
561 		 * init direction, but which will not be removed by the
562 		 * nf_conntrack_in() call, if a matching conntrack entry is
563 		 * found instead.  In this case all init direction packets
564 		 * would be reported as new related packets, while reply
565 		 * direction packets would be reported as un-related
566 		 * established packets.
567 		 */
568 		h = nf_conntrack_find_get(net, zone, &tuple);
569 		if (h) {
570 			struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
571 
572 			nf_ct_delete(ct, 0, 0);
573 			nf_conntrack_put(&ct->ct_general);
574 		}
575 	}
576 
577 	return exp;
578 }
579 
580 /* This replicates logic from nf_conntrack_core.c that is not exported. */
581 static enum ip_conntrack_info
582 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
583 {
584 	const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
585 
586 	if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
587 		return IP_CT_ESTABLISHED_REPLY;
588 	/* Once we've had two way comms, always ESTABLISHED. */
589 	if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
590 		return IP_CT_ESTABLISHED;
591 	if (test_bit(IPS_EXPECTED_BIT, &ct->status))
592 		return IP_CT_RELATED;
593 	return IP_CT_NEW;
594 }
595 
596 /* Find an existing connection which this packet belongs to without
597  * re-attributing statistics or modifying the connection state.  This allows an
598  * skb->_nfct lost due to an upcall to be recovered during actions execution.
599  *
600  * Must be called with rcu_read_lock.
601  *
602  * On success, populates skb->_nfct and returns the connection.  Returns NULL
603  * if there is no existing entry.
604  */
605 static struct nf_conn *
606 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
607 		     u8 l3num, struct sk_buff *skb, bool natted)
608 {
609 	struct nf_conntrack_tuple tuple;
610 	struct nf_conntrack_tuple_hash *h;
611 	struct nf_conn *ct;
612 
613 	if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
614 			       net, &tuple)) {
615 		pr_debug("ovs_ct_find_existing: Can't get tuple\n");
616 		return NULL;
617 	}
618 
619 	/* Must invert the tuple if skb has been transformed by NAT. */
620 	if (natted) {
621 		struct nf_conntrack_tuple inverse;
622 
623 		if (!nf_ct_invert_tuple(&inverse, &tuple)) {
624 			pr_debug("ovs_ct_find_existing: Inversion failed!\n");
625 			return NULL;
626 		}
627 		tuple = inverse;
628 	}
629 
630 	/* look for tuple match */
631 	h = nf_conntrack_find_get(net, zone, &tuple);
632 	if (!h)
633 		return NULL;   /* Not found. */
634 
635 	ct = nf_ct_tuplehash_to_ctrack(h);
636 
637 	/* Inverted packet tuple matches the reverse direction conntrack tuple,
638 	 * select the other tuplehash to get the right 'ctinfo' bits for this
639 	 * packet.
640 	 */
641 	if (natted)
642 		h = &ct->tuplehash[!h->tuple.dst.dir];
643 
644 	nf_ct_set(skb, ct, ovs_ct_get_info(h));
645 	return ct;
646 }
647 
648 static
649 struct nf_conn *ovs_ct_executed(struct net *net,
650 				const struct sw_flow_key *key,
651 				const struct ovs_conntrack_info *info,
652 				struct sk_buff *skb,
653 				bool *ct_executed)
654 {
655 	struct nf_conn *ct = NULL;
656 
657 	/* If no ct, check if we have evidence that an existing conntrack entry
658 	 * might be found for this skb.  This happens when we lose a skb->_nfct
659 	 * due to an upcall, or if the direction is being forced.  If the
660 	 * connection was not confirmed, it is not cached and needs to be run
661 	 * through conntrack again.
662 	 */
663 	*ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
664 		       !(key->ct_state & OVS_CS_F_INVALID) &&
665 		       (key->ct_zone == info->zone.id);
666 
667 	if (*ct_executed || (!key->ct_state && info->force)) {
668 		ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
669 					  !!(key->ct_state &
670 					  OVS_CS_F_NAT_MASK));
671 	}
672 
673 	return ct;
674 }
675 
676 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
677 static bool skb_nfct_cached(struct net *net,
678 			    const struct sw_flow_key *key,
679 			    const struct ovs_conntrack_info *info,
680 			    struct sk_buff *skb)
681 {
682 	enum ip_conntrack_info ctinfo;
683 	struct nf_conn *ct;
684 	bool ct_executed = true;
685 
686 	ct = nf_ct_get(skb, &ctinfo);
687 	if (!ct)
688 		ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
689 
690 	if (ct)
691 		nf_ct_get(skb, &ctinfo);
692 	else
693 		return false;
694 
695 	if (!net_eq(net, read_pnet(&ct->ct_net)))
696 		return false;
697 	if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
698 		return false;
699 	if (info->helper) {
700 		struct nf_conn_help *help;
701 
702 		help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
703 		if (help && rcu_access_pointer(help->helper) != info->helper)
704 			return false;
705 	}
706 	/* Force conntrack entry direction to the current packet? */
707 	if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
708 		/* Delete the conntrack entry if confirmed, else just release
709 		 * the reference.
710 		 */
711 		if (nf_ct_is_confirmed(ct))
712 			nf_ct_delete(ct, 0, 0);
713 
714 		nf_conntrack_put(&ct->ct_general);
715 		nf_ct_set(skb, NULL, 0);
716 		return false;
717 	}
718 
719 	return ct_executed;
720 }
721 
722 #ifdef CONFIG_NF_NAT_NEEDED
723 /* Modelled after nf_nat_ipv[46]_fn().
724  * range is only used for new, uninitialized NAT state.
725  * Returns either NF_ACCEPT or NF_DROP.
726  */
727 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
728 			      enum ip_conntrack_info ctinfo,
729 			      const struct nf_nat_range2 *range,
730 			      enum nf_nat_manip_type maniptype)
731 {
732 	int hooknum, nh_off, err = NF_ACCEPT;
733 
734 	nh_off = skb_network_offset(skb);
735 	skb_pull_rcsum(skb, nh_off);
736 
737 	/* See HOOK2MANIP(). */
738 	if (maniptype == NF_NAT_MANIP_SRC)
739 		hooknum = NF_INET_LOCAL_IN; /* Source NAT */
740 	else
741 		hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
742 
743 	switch (ctinfo) {
744 	case IP_CT_RELATED:
745 	case IP_CT_RELATED_REPLY:
746 		if (IS_ENABLED(CONFIG_NF_NAT) &&
747 		    skb->protocol == htons(ETH_P_IP) &&
748 		    ip_hdr(skb)->protocol == IPPROTO_ICMP) {
749 			if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
750 							   hooknum))
751 				err = NF_DROP;
752 			goto push;
753 		} else if (IS_ENABLED(CONFIG_IPV6) &&
754 			   skb->protocol == htons(ETH_P_IPV6)) {
755 			__be16 frag_off;
756 			u8 nexthdr = ipv6_hdr(skb)->nexthdr;
757 			int hdrlen = ipv6_skip_exthdr(skb,
758 						      sizeof(struct ipv6hdr),
759 						      &nexthdr, &frag_off);
760 
761 			if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
762 				if (!nf_nat_icmpv6_reply_translation(skb, ct,
763 								     ctinfo,
764 								     hooknum,
765 								     hdrlen))
766 					err = NF_DROP;
767 				goto push;
768 			}
769 		}
770 		/* Non-ICMP, fall thru to initialize if needed. */
771 		/* fall through */
772 	case IP_CT_NEW:
773 		/* Seen it before?  This can happen for loopback, retrans,
774 		 * or local packets.
775 		 */
776 		if (!nf_nat_initialized(ct, maniptype)) {
777 			/* Initialize according to the NAT action. */
778 			err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
779 				/* Action is set up to establish a new
780 				 * mapping.
781 				 */
782 				? nf_nat_setup_info(ct, range, maniptype)
783 				: nf_nat_alloc_null_binding(ct, hooknum);
784 			if (err != NF_ACCEPT)
785 				goto push;
786 		}
787 		break;
788 
789 	case IP_CT_ESTABLISHED:
790 	case IP_CT_ESTABLISHED_REPLY:
791 		break;
792 
793 	default:
794 		err = NF_DROP;
795 		goto push;
796 	}
797 
798 	err = nf_nat_packet(ct, ctinfo, hooknum, skb);
799 push:
800 	skb_push(skb, nh_off);
801 	skb_postpush_rcsum(skb, skb->data, nh_off);
802 
803 	return err;
804 }
805 
806 static void ovs_nat_update_key(struct sw_flow_key *key,
807 			       const struct sk_buff *skb,
808 			       enum nf_nat_manip_type maniptype)
809 {
810 	if (maniptype == NF_NAT_MANIP_SRC) {
811 		__be16 src;
812 
813 		key->ct_state |= OVS_CS_F_SRC_NAT;
814 		if (key->eth.type == htons(ETH_P_IP))
815 			key->ipv4.addr.src = ip_hdr(skb)->saddr;
816 		else if (key->eth.type == htons(ETH_P_IPV6))
817 			memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
818 			       sizeof(key->ipv6.addr.src));
819 		else
820 			return;
821 
822 		if (key->ip.proto == IPPROTO_UDP)
823 			src = udp_hdr(skb)->source;
824 		else if (key->ip.proto == IPPROTO_TCP)
825 			src = tcp_hdr(skb)->source;
826 		else if (key->ip.proto == IPPROTO_SCTP)
827 			src = sctp_hdr(skb)->source;
828 		else
829 			return;
830 
831 		key->tp.src = src;
832 	} else {
833 		__be16 dst;
834 
835 		key->ct_state |= OVS_CS_F_DST_NAT;
836 		if (key->eth.type == htons(ETH_P_IP))
837 			key->ipv4.addr.dst = ip_hdr(skb)->daddr;
838 		else if (key->eth.type == htons(ETH_P_IPV6))
839 			memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
840 			       sizeof(key->ipv6.addr.dst));
841 		else
842 			return;
843 
844 		if (key->ip.proto == IPPROTO_UDP)
845 			dst = udp_hdr(skb)->dest;
846 		else if (key->ip.proto == IPPROTO_TCP)
847 			dst = tcp_hdr(skb)->dest;
848 		else if (key->ip.proto == IPPROTO_SCTP)
849 			dst = sctp_hdr(skb)->dest;
850 		else
851 			return;
852 
853 		key->tp.dst = dst;
854 	}
855 }
856 
857 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
858 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
859 		      const struct ovs_conntrack_info *info,
860 		      struct sk_buff *skb, struct nf_conn *ct,
861 		      enum ip_conntrack_info ctinfo)
862 {
863 	enum nf_nat_manip_type maniptype;
864 	int err;
865 
866 	/* Add NAT extension if not confirmed yet. */
867 	if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
868 		return NF_ACCEPT;   /* Can't NAT. */
869 
870 	/* Determine NAT type.
871 	 * Check if the NAT type can be deduced from the tracked connection.
872 	 * Make sure new expected connections (IP_CT_RELATED) are NATted only
873 	 * when committing.
874 	 */
875 	if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
876 	    ct->status & IPS_NAT_MASK &&
877 	    (ctinfo != IP_CT_RELATED || info->commit)) {
878 		/* NAT an established or related connection like before. */
879 		if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
880 			/* This is the REPLY direction for a connection
881 			 * for which NAT was applied in the forward
882 			 * direction.  Do the reverse NAT.
883 			 */
884 			maniptype = ct->status & IPS_SRC_NAT
885 				? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
886 		else
887 			maniptype = ct->status & IPS_SRC_NAT
888 				? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
889 	} else if (info->nat & OVS_CT_SRC_NAT) {
890 		maniptype = NF_NAT_MANIP_SRC;
891 	} else if (info->nat & OVS_CT_DST_NAT) {
892 		maniptype = NF_NAT_MANIP_DST;
893 	} else {
894 		return NF_ACCEPT; /* Connection is not NATed. */
895 	}
896 	err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
897 
898 	/* Mark NAT done if successful and update the flow key. */
899 	if (err == NF_ACCEPT)
900 		ovs_nat_update_key(key, skb, maniptype);
901 
902 	return err;
903 }
904 #else /* !CONFIG_NF_NAT_NEEDED */
905 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
906 		      const struct ovs_conntrack_info *info,
907 		      struct sk_buff *skb, struct nf_conn *ct,
908 		      enum ip_conntrack_info ctinfo)
909 {
910 	return NF_ACCEPT;
911 }
912 #endif
913 
914 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
915  * not done already.  Update key with new CT state after passing the packet
916  * through conntrack.
917  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
918  * set to NULL and 0 will be returned.
919  */
920 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
921 			   const struct ovs_conntrack_info *info,
922 			   struct sk_buff *skb)
923 {
924 	/* If we are recirculating packets to match on conntrack fields and
925 	 * committing with a separate conntrack action,  then we don't need to
926 	 * actually run the packet through conntrack twice unless it's for a
927 	 * different zone.
928 	 */
929 	bool cached = skb_nfct_cached(net, key, info, skb);
930 	enum ip_conntrack_info ctinfo;
931 	struct nf_conn *ct;
932 
933 	if (!cached) {
934 		struct nf_hook_state state = {
935 			.hook = NF_INET_PRE_ROUTING,
936 			.pf = info->family,
937 			.net = net,
938 		};
939 		struct nf_conn *tmpl = info->ct;
940 		int err;
941 
942 		/* Associate skb with specified zone. */
943 		if (tmpl) {
944 			if (skb_nfct(skb))
945 				nf_conntrack_put(skb_nfct(skb));
946 			nf_conntrack_get(&tmpl->ct_general);
947 			nf_ct_set(skb, tmpl, IP_CT_NEW);
948 		}
949 
950 		err = nf_conntrack_in(skb, &state);
951 		if (err != NF_ACCEPT)
952 			return -ENOENT;
953 
954 		/* Clear CT state NAT flags to mark that we have not yet done
955 		 * NAT after the nf_conntrack_in() call.  We can actually clear
956 		 * the whole state, as it will be re-initialized below.
957 		 */
958 		key->ct_state = 0;
959 
960 		/* Update the key, but keep the NAT flags. */
961 		ovs_ct_update_key(skb, info, key, true, true);
962 	}
963 
964 	ct = nf_ct_get(skb, &ctinfo);
965 	if (ct) {
966 		/* Packets starting a new connection must be NATted before the
967 		 * helper, so that the helper knows about the NAT.  We enforce
968 		 * this by delaying both NAT and helper calls for unconfirmed
969 		 * connections until the committing CT action.  For later
970 		 * packets NAT and Helper may be called in either order.
971 		 *
972 		 * NAT will be done only if the CT action has NAT, and only
973 		 * once per packet (per zone), as guarded by the NAT bits in
974 		 * the key->ct_state.
975 		 */
976 		if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
977 		    (nf_ct_is_confirmed(ct) || info->commit) &&
978 		    ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
979 			return -EINVAL;
980 		}
981 
982 		/* Userspace may decide to perform a ct lookup without a helper
983 		 * specified followed by a (recirculate and) commit with one.
984 		 * Therefore, for unconfirmed connections which we will commit,
985 		 * we need to attach the helper here.
986 		 */
987 		if (!nf_ct_is_confirmed(ct) && info->commit &&
988 		    info->helper && !nfct_help(ct)) {
989 			int err = __nf_ct_try_assign_helper(ct, info->ct,
990 							    GFP_ATOMIC);
991 			if (err)
992 				return err;
993 		}
994 
995 		/* Call the helper only if:
996 		 * - nf_conntrack_in() was executed above ("!cached") for a
997 		 *   confirmed connection, or
998 		 * - When committing an unconfirmed connection.
999 		 */
1000 		if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
1001 		    ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1002 			return -EINVAL;
1003 		}
1004 	}
1005 
1006 	return 0;
1007 }
1008 
1009 /* Lookup connection and read fields into key. */
1010 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1011 			 const struct ovs_conntrack_info *info,
1012 			 struct sk_buff *skb)
1013 {
1014 	struct nf_conntrack_expect *exp;
1015 
1016 	/* If we pass an expected packet through nf_conntrack_in() the
1017 	 * expectation is typically removed, but the packet could still be
1018 	 * lost in upcall processing.  To prevent this from happening we
1019 	 * perform an explicit expectation lookup.  Expected connections are
1020 	 * always new, and will be passed through conntrack only when they are
1021 	 * committed, as it is OK to remove the expectation at that time.
1022 	 */
1023 	exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1024 	if (exp) {
1025 		u8 state;
1026 
1027 		/* NOTE: New connections are NATted and Helped only when
1028 		 * committed, so we are not calling into NAT here.
1029 		 */
1030 		state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1031 		__ovs_ct_update_key(key, state, &info->zone, exp->master);
1032 	} else {
1033 		struct nf_conn *ct;
1034 		int err;
1035 
1036 		err = __ovs_ct_lookup(net, key, info, skb);
1037 		if (err)
1038 			return err;
1039 
1040 		ct = (struct nf_conn *)skb_nfct(skb);
1041 		if (ct)
1042 			nf_ct_deliver_cached_events(ct);
1043 	}
1044 
1045 	return 0;
1046 }
1047 
1048 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1049 {
1050 	size_t i;
1051 
1052 	for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1053 		if (labels->ct_labels_32[i])
1054 			return true;
1055 
1056 	return false;
1057 }
1058 
1059 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1060 static struct hlist_head *ct_limit_hash_bucket(
1061 	const struct ovs_ct_limit_info *info, u16 zone)
1062 {
1063 	return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1064 }
1065 
1066 /* Call with ovs_mutex */
1067 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1068 			 struct ovs_ct_limit *new_ct_limit)
1069 {
1070 	struct ovs_ct_limit *ct_limit;
1071 	struct hlist_head *head;
1072 
1073 	head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1074 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1075 		if (ct_limit->zone == new_ct_limit->zone) {
1076 			hlist_replace_rcu(&ct_limit->hlist_node,
1077 					  &new_ct_limit->hlist_node);
1078 			kfree_rcu(ct_limit, rcu);
1079 			return;
1080 		}
1081 	}
1082 
1083 	hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1084 }
1085 
1086 /* Call with ovs_mutex */
1087 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1088 {
1089 	struct ovs_ct_limit *ct_limit;
1090 	struct hlist_head *head;
1091 	struct hlist_node *n;
1092 
1093 	head = ct_limit_hash_bucket(info, zone);
1094 	hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1095 		if (ct_limit->zone == zone) {
1096 			hlist_del_rcu(&ct_limit->hlist_node);
1097 			kfree_rcu(ct_limit, rcu);
1098 			return;
1099 		}
1100 	}
1101 }
1102 
1103 /* Call with RCU read lock */
1104 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1105 {
1106 	struct ovs_ct_limit *ct_limit;
1107 	struct hlist_head *head;
1108 
1109 	head = ct_limit_hash_bucket(info, zone);
1110 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1111 		if (ct_limit->zone == zone)
1112 			return ct_limit->limit;
1113 	}
1114 
1115 	return info->default_limit;
1116 }
1117 
1118 static int ovs_ct_check_limit(struct net *net,
1119 			      const struct ovs_conntrack_info *info,
1120 			      const struct nf_conntrack_tuple *tuple)
1121 {
1122 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1123 	const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1124 	u32 per_zone_limit, connections;
1125 	u32 conncount_key;
1126 
1127 	conncount_key = info->zone.id;
1128 
1129 	per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1130 	if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1131 		return 0;
1132 
1133 	connections = nf_conncount_count(net, ct_limit_info->data,
1134 					 &conncount_key, tuple, &info->zone);
1135 	if (connections > per_zone_limit)
1136 		return -ENOMEM;
1137 
1138 	return 0;
1139 }
1140 #endif
1141 
1142 /* Lookup connection and confirm if unconfirmed. */
1143 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1144 			 const struct ovs_conntrack_info *info,
1145 			 struct sk_buff *skb)
1146 {
1147 	enum ip_conntrack_info ctinfo;
1148 	struct nf_conn *ct;
1149 	int err;
1150 
1151 	err = __ovs_ct_lookup(net, key, info, skb);
1152 	if (err)
1153 		return err;
1154 
1155 	/* The connection could be invalid, in which case this is a no-op.*/
1156 	ct = nf_ct_get(skb, &ctinfo);
1157 	if (!ct)
1158 		return 0;
1159 
1160 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1161 	if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1162 		if (!nf_ct_is_confirmed(ct)) {
1163 			err = ovs_ct_check_limit(net, info,
1164 				&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1165 			if (err) {
1166 				net_warn_ratelimited("openvswitch: zone: %u "
1167 					"exceeds conntrack limit\n",
1168 					info->zone.id);
1169 				return err;
1170 			}
1171 		}
1172 	}
1173 #endif
1174 
1175 	/* Set the conntrack event mask if given.  NEW and DELETE events have
1176 	 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1177 	 * typically would receive many kinds of updates.  Setting the event
1178 	 * mask allows those events to be filtered.  The set event mask will
1179 	 * remain in effect for the lifetime of the connection unless changed
1180 	 * by a further CT action with both the commit flag and the eventmask
1181 	 * option. */
1182 	if (info->have_eventmask) {
1183 		struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1184 
1185 		if (cache)
1186 			cache->ctmask = info->eventmask;
1187 	}
1188 
1189 	/* Apply changes before confirming the connection so that the initial
1190 	 * conntrack NEW netlink event carries the values given in the CT
1191 	 * action.
1192 	 */
1193 	if (info->mark.mask) {
1194 		err = ovs_ct_set_mark(ct, key, info->mark.value,
1195 				      info->mark.mask);
1196 		if (err)
1197 			return err;
1198 	}
1199 	if (!nf_ct_is_confirmed(ct)) {
1200 		err = ovs_ct_init_labels(ct, key, &info->labels.value,
1201 					 &info->labels.mask);
1202 		if (err)
1203 			return err;
1204 	} else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1205 		   labels_nonzero(&info->labels.mask)) {
1206 		err = ovs_ct_set_labels(ct, key, &info->labels.value,
1207 					&info->labels.mask);
1208 		if (err)
1209 			return err;
1210 	}
1211 	/* This will take care of sending queued events even if the connection
1212 	 * is already confirmed.
1213 	 */
1214 	if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1215 		return -EINVAL;
1216 
1217 	return 0;
1218 }
1219 
1220 /* Trim the skb to the length specified by the IP/IPv6 header,
1221  * removing any trailing lower-layer padding. This prepares the skb
1222  * for higher-layer processing that assumes skb->len excludes padding
1223  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1224  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1225  */
1226 static int ovs_skb_network_trim(struct sk_buff *skb)
1227 {
1228 	unsigned int len;
1229 	int err;
1230 
1231 	switch (skb->protocol) {
1232 	case htons(ETH_P_IP):
1233 		len = ntohs(ip_hdr(skb)->tot_len);
1234 		break;
1235 	case htons(ETH_P_IPV6):
1236 		len = sizeof(struct ipv6hdr)
1237 			+ ntohs(ipv6_hdr(skb)->payload_len);
1238 		break;
1239 	default:
1240 		len = skb->len;
1241 	}
1242 
1243 	err = pskb_trim_rcsum(skb, len);
1244 	if (err)
1245 		kfree_skb(skb);
1246 
1247 	return err;
1248 }
1249 
1250 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1251  * value if 'skb' is freed.
1252  */
1253 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1254 		   struct sw_flow_key *key,
1255 		   const struct ovs_conntrack_info *info)
1256 {
1257 	int nh_ofs;
1258 	int err;
1259 
1260 	/* The conntrack module expects to be working at L3. */
1261 	nh_ofs = skb_network_offset(skb);
1262 	skb_pull_rcsum(skb, nh_ofs);
1263 
1264 	err = ovs_skb_network_trim(skb);
1265 	if (err)
1266 		return err;
1267 
1268 	if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1269 		err = handle_fragments(net, key, info->zone.id, skb);
1270 		if (err)
1271 			return err;
1272 	}
1273 
1274 	if (info->commit)
1275 		err = ovs_ct_commit(net, key, info, skb);
1276 	else
1277 		err = ovs_ct_lookup(net, key, info, skb);
1278 
1279 	skb_push(skb, nh_ofs);
1280 	skb_postpush_rcsum(skb, skb->data, nh_ofs);
1281 	if (err)
1282 		kfree_skb(skb);
1283 	return err;
1284 }
1285 
1286 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1287 {
1288 	if (skb_nfct(skb)) {
1289 		nf_conntrack_put(skb_nfct(skb));
1290 		nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1291 		ovs_ct_fill_key(skb, key);
1292 	}
1293 
1294 	return 0;
1295 }
1296 
1297 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1298 			     const struct sw_flow_key *key, bool log)
1299 {
1300 	struct nf_conntrack_helper *helper;
1301 	struct nf_conn_help *help;
1302 
1303 	helper = nf_conntrack_helper_try_module_get(name, info->family,
1304 						    key->ip.proto);
1305 	if (!helper) {
1306 		OVS_NLERR(log, "Unknown helper \"%s\"", name);
1307 		return -EINVAL;
1308 	}
1309 
1310 	help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1311 	if (!help) {
1312 		nf_conntrack_helper_put(helper);
1313 		return -ENOMEM;
1314 	}
1315 
1316 	rcu_assign_pointer(help->helper, helper);
1317 	info->helper = helper;
1318 
1319 	if (info->nat)
1320 		request_module("ip_nat_%s", name);
1321 
1322 	return 0;
1323 }
1324 
1325 #ifdef CONFIG_NF_NAT_NEEDED
1326 static int parse_nat(const struct nlattr *attr,
1327 		     struct ovs_conntrack_info *info, bool log)
1328 {
1329 	struct nlattr *a;
1330 	int rem;
1331 	bool have_ip_max = false;
1332 	bool have_proto_max = false;
1333 	bool ip_vers = (info->family == NFPROTO_IPV6);
1334 
1335 	nla_for_each_nested(a, attr, rem) {
1336 		static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1337 			[OVS_NAT_ATTR_SRC] = {0, 0},
1338 			[OVS_NAT_ATTR_DST] = {0, 0},
1339 			[OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1340 						 sizeof(struct in6_addr)},
1341 			[OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1342 						 sizeof(struct in6_addr)},
1343 			[OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1344 			[OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1345 			[OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1346 			[OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1347 			[OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1348 		};
1349 		int type = nla_type(a);
1350 
1351 		if (type > OVS_NAT_ATTR_MAX) {
1352 			OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1353 				  type, OVS_NAT_ATTR_MAX);
1354 			return -EINVAL;
1355 		}
1356 
1357 		if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1358 			OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1359 				  type, nla_len(a),
1360 				  ovs_nat_attr_lens[type][ip_vers]);
1361 			return -EINVAL;
1362 		}
1363 
1364 		switch (type) {
1365 		case OVS_NAT_ATTR_SRC:
1366 		case OVS_NAT_ATTR_DST:
1367 			if (info->nat) {
1368 				OVS_NLERR(log, "Only one type of NAT may be specified");
1369 				return -ERANGE;
1370 			}
1371 			info->nat |= OVS_CT_NAT;
1372 			info->nat |= ((type == OVS_NAT_ATTR_SRC)
1373 					? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1374 			break;
1375 
1376 		case OVS_NAT_ATTR_IP_MIN:
1377 			nla_memcpy(&info->range.min_addr, a,
1378 				   sizeof(info->range.min_addr));
1379 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1380 			break;
1381 
1382 		case OVS_NAT_ATTR_IP_MAX:
1383 			have_ip_max = true;
1384 			nla_memcpy(&info->range.max_addr, a,
1385 				   sizeof(info->range.max_addr));
1386 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1387 			break;
1388 
1389 		case OVS_NAT_ATTR_PROTO_MIN:
1390 			info->range.min_proto.all = htons(nla_get_u16(a));
1391 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1392 			break;
1393 
1394 		case OVS_NAT_ATTR_PROTO_MAX:
1395 			have_proto_max = true;
1396 			info->range.max_proto.all = htons(nla_get_u16(a));
1397 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1398 			break;
1399 
1400 		case OVS_NAT_ATTR_PERSISTENT:
1401 			info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1402 			break;
1403 
1404 		case OVS_NAT_ATTR_PROTO_HASH:
1405 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1406 			break;
1407 
1408 		case OVS_NAT_ATTR_PROTO_RANDOM:
1409 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1410 			break;
1411 
1412 		default:
1413 			OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1414 			return -EINVAL;
1415 		}
1416 	}
1417 
1418 	if (rem > 0) {
1419 		OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1420 		return -EINVAL;
1421 	}
1422 	if (!info->nat) {
1423 		/* Do not allow flags if no type is given. */
1424 		if (info->range.flags) {
1425 			OVS_NLERR(log,
1426 				  "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1427 				  );
1428 			return -EINVAL;
1429 		}
1430 		info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1431 	} else if (!info->commit) {
1432 		OVS_NLERR(log,
1433 			  "NAT attributes may be specified only when CT COMMIT flag is also specified."
1434 			  );
1435 		return -EINVAL;
1436 	}
1437 	/* Allow missing IP_MAX. */
1438 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1439 		memcpy(&info->range.max_addr, &info->range.min_addr,
1440 		       sizeof(info->range.max_addr));
1441 	}
1442 	/* Allow missing PROTO_MAX. */
1443 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1444 	    !have_proto_max) {
1445 		info->range.max_proto.all = info->range.min_proto.all;
1446 	}
1447 	return 0;
1448 }
1449 #endif
1450 
1451 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1452 	[OVS_CT_ATTR_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1453 	[OVS_CT_ATTR_FORCE_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1454 	[OVS_CT_ATTR_ZONE]	= { .minlen = sizeof(u16),
1455 				    .maxlen = sizeof(u16) },
1456 	[OVS_CT_ATTR_MARK]	= { .minlen = sizeof(struct md_mark),
1457 				    .maxlen = sizeof(struct md_mark) },
1458 	[OVS_CT_ATTR_LABELS]	= { .minlen = sizeof(struct md_labels),
1459 				    .maxlen = sizeof(struct md_labels) },
1460 	[OVS_CT_ATTR_HELPER]	= { .minlen = 1,
1461 				    .maxlen = NF_CT_HELPER_NAME_LEN },
1462 #ifdef CONFIG_NF_NAT_NEEDED
1463 	/* NAT length is checked when parsing the nested attributes. */
1464 	[OVS_CT_ATTR_NAT]	= { .minlen = 0, .maxlen = INT_MAX },
1465 #endif
1466 	[OVS_CT_ATTR_EVENTMASK]	= { .minlen = sizeof(u32),
1467 				    .maxlen = sizeof(u32) },
1468 };
1469 
1470 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1471 		    const char **helper, bool log)
1472 {
1473 	struct nlattr *a;
1474 	int rem;
1475 
1476 	nla_for_each_nested(a, attr, rem) {
1477 		int type = nla_type(a);
1478 		int maxlen;
1479 		int minlen;
1480 
1481 		if (type > OVS_CT_ATTR_MAX) {
1482 			OVS_NLERR(log,
1483 				  "Unknown conntrack attr (type=%d, max=%d)",
1484 				  type, OVS_CT_ATTR_MAX);
1485 			return -EINVAL;
1486 		}
1487 
1488 		maxlen = ovs_ct_attr_lens[type].maxlen;
1489 		minlen = ovs_ct_attr_lens[type].minlen;
1490 		if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1491 			OVS_NLERR(log,
1492 				  "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1493 				  type, nla_len(a), maxlen);
1494 			return -EINVAL;
1495 		}
1496 
1497 		switch (type) {
1498 		case OVS_CT_ATTR_FORCE_COMMIT:
1499 			info->force = true;
1500 			/* fall through. */
1501 		case OVS_CT_ATTR_COMMIT:
1502 			info->commit = true;
1503 			break;
1504 #ifdef CONFIG_NF_CONNTRACK_ZONES
1505 		case OVS_CT_ATTR_ZONE:
1506 			info->zone.id = nla_get_u16(a);
1507 			break;
1508 #endif
1509 #ifdef CONFIG_NF_CONNTRACK_MARK
1510 		case OVS_CT_ATTR_MARK: {
1511 			struct md_mark *mark = nla_data(a);
1512 
1513 			if (!mark->mask) {
1514 				OVS_NLERR(log, "ct_mark mask cannot be 0");
1515 				return -EINVAL;
1516 			}
1517 			info->mark = *mark;
1518 			break;
1519 		}
1520 #endif
1521 #ifdef CONFIG_NF_CONNTRACK_LABELS
1522 		case OVS_CT_ATTR_LABELS: {
1523 			struct md_labels *labels = nla_data(a);
1524 
1525 			if (!labels_nonzero(&labels->mask)) {
1526 				OVS_NLERR(log, "ct_labels mask cannot be 0");
1527 				return -EINVAL;
1528 			}
1529 			info->labels = *labels;
1530 			break;
1531 		}
1532 #endif
1533 		case OVS_CT_ATTR_HELPER:
1534 			*helper = nla_data(a);
1535 			if (!memchr(*helper, '\0', nla_len(a))) {
1536 				OVS_NLERR(log, "Invalid conntrack helper");
1537 				return -EINVAL;
1538 			}
1539 			break;
1540 #ifdef CONFIG_NF_NAT_NEEDED
1541 		case OVS_CT_ATTR_NAT: {
1542 			int err = parse_nat(a, info, log);
1543 
1544 			if (err)
1545 				return err;
1546 			break;
1547 		}
1548 #endif
1549 		case OVS_CT_ATTR_EVENTMASK:
1550 			info->have_eventmask = true;
1551 			info->eventmask = nla_get_u32(a);
1552 			break;
1553 
1554 		default:
1555 			OVS_NLERR(log, "Unknown conntrack attr (%d)",
1556 				  type);
1557 			return -EINVAL;
1558 		}
1559 	}
1560 
1561 #ifdef CONFIG_NF_CONNTRACK_MARK
1562 	if (!info->commit && info->mark.mask) {
1563 		OVS_NLERR(log,
1564 			  "Setting conntrack mark requires 'commit' flag.");
1565 		return -EINVAL;
1566 	}
1567 #endif
1568 #ifdef CONFIG_NF_CONNTRACK_LABELS
1569 	if (!info->commit && labels_nonzero(&info->labels.mask)) {
1570 		OVS_NLERR(log,
1571 			  "Setting conntrack labels requires 'commit' flag.");
1572 		return -EINVAL;
1573 	}
1574 #endif
1575 	if (rem > 0) {
1576 		OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1577 		return -EINVAL;
1578 	}
1579 
1580 	return 0;
1581 }
1582 
1583 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1584 {
1585 	if (attr == OVS_KEY_ATTR_CT_STATE)
1586 		return true;
1587 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1588 	    attr == OVS_KEY_ATTR_CT_ZONE)
1589 		return true;
1590 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1591 	    attr == OVS_KEY_ATTR_CT_MARK)
1592 		return true;
1593 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1594 	    attr == OVS_KEY_ATTR_CT_LABELS) {
1595 		struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1596 
1597 		return ovs_net->xt_label;
1598 	}
1599 
1600 	return false;
1601 }
1602 
1603 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1604 		       const struct sw_flow_key *key,
1605 		       struct sw_flow_actions **sfa,  bool log)
1606 {
1607 	struct ovs_conntrack_info ct_info;
1608 	const char *helper = NULL;
1609 	u16 family;
1610 	int err;
1611 
1612 	family = key_to_nfproto(key);
1613 	if (family == NFPROTO_UNSPEC) {
1614 		OVS_NLERR(log, "ct family unspecified");
1615 		return -EINVAL;
1616 	}
1617 
1618 	memset(&ct_info, 0, sizeof(ct_info));
1619 	ct_info.family = family;
1620 
1621 	nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1622 			NF_CT_DEFAULT_ZONE_DIR, 0);
1623 
1624 	err = parse_ct(attr, &ct_info, &helper, log);
1625 	if (err)
1626 		return err;
1627 
1628 	/* Set up template for tracking connections in specific zones. */
1629 	ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1630 	if (!ct_info.ct) {
1631 		OVS_NLERR(log, "Failed to allocate conntrack template");
1632 		return -ENOMEM;
1633 	}
1634 	if (helper) {
1635 		err = ovs_ct_add_helper(&ct_info, helper, key, log);
1636 		if (err)
1637 			goto err_free_ct;
1638 	}
1639 
1640 	err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1641 				 sizeof(ct_info), log);
1642 	if (err)
1643 		goto err_free_ct;
1644 
1645 	__set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1646 	nf_conntrack_get(&ct_info.ct->ct_general);
1647 	return 0;
1648 err_free_ct:
1649 	__ovs_ct_free_action(&ct_info);
1650 	return err;
1651 }
1652 
1653 #ifdef CONFIG_NF_NAT_NEEDED
1654 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1655 			       struct sk_buff *skb)
1656 {
1657 	struct nlattr *start;
1658 
1659 	start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1660 	if (!start)
1661 		return false;
1662 
1663 	if (info->nat & OVS_CT_SRC_NAT) {
1664 		if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1665 			return false;
1666 	} else if (info->nat & OVS_CT_DST_NAT) {
1667 		if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1668 			return false;
1669 	} else {
1670 		goto out;
1671 	}
1672 
1673 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1674 		if (IS_ENABLED(CONFIG_NF_NAT) &&
1675 		    info->family == NFPROTO_IPV4) {
1676 			if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1677 					    info->range.min_addr.ip) ||
1678 			    (info->range.max_addr.ip
1679 			     != info->range.min_addr.ip &&
1680 			     (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1681 					      info->range.max_addr.ip))))
1682 				return false;
1683 		} else if (IS_ENABLED(CONFIG_IPV6) &&
1684 			   info->family == NFPROTO_IPV6) {
1685 			if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1686 					     &info->range.min_addr.in6) ||
1687 			    (memcmp(&info->range.max_addr.in6,
1688 				    &info->range.min_addr.in6,
1689 				    sizeof(info->range.max_addr.in6)) &&
1690 			     (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1691 					       &info->range.max_addr.in6))))
1692 				return false;
1693 		} else {
1694 			return false;
1695 		}
1696 	}
1697 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1698 	    (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1699 			 ntohs(info->range.min_proto.all)) ||
1700 	     (info->range.max_proto.all != info->range.min_proto.all &&
1701 	      nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1702 			  ntohs(info->range.max_proto.all)))))
1703 		return false;
1704 
1705 	if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1706 	    nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1707 		return false;
1708 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1709 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1710 		return false;
1711 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1712 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1713 		return false;
1714 out:
1715 	nla_nest_end(skb, start);
1716 
1717 	return true;
1718 }
1719 #endif
1720 
1721 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1722 			  struct sk_buff *skb)
1723 {
1724 	struct nlattr *start;
1725 
1726 	start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1727 	if (!start)
1728 		return -EMSGSIZE;
1729 
1730 	if (ct_info->commit && nla_put_flag(skb, ct_info->force
1731 					    ? OVS_CT_ATTR_FORCE_COMMIT
1732 					    : OVS_CT_ATTR_COMMIT))
1733 		return -EMSGSIZE;
1734 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1735 	    nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1736 		return -EMSGSIZE;
1737 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1738 	    nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1739 		    &ct_info->mark))
1740 		return -EMSGSIZE;
1741 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1742 	    labels_nonzero(&ct_info->labels.mask) &&
1743 	    nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1744 		    &ct_info->labels))
1745 		return -EMSGSIZE;
1746 	if (ct_info->helper) {
1747 		if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1748 				   ct_info->helper->name))
1749 			return -EMSGSIZE;
1750 	}
1751 	if (ct_info->have_eventmask &&
1752 	    nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1753 		return -EMSGSIZE;
1754 
1755 #ifdef CONFIG_NF_NAT_NEEDED
1756 	if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1757 		return -EMSGSIZE;
1758 #endif
1759 	nla_nest_end(skb, start);
1760 
1761 	return 0;
1762 }
1763 
1764 void ovs_ct_free_action(const struct nlattr *a)
1765 {
1766 	struct ovs_conntrack_info *ct_info = nla_data(a);
1767 
1768 	__ovs_ct_free_action(ct_info);
1769 }
1770 
1771 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1772 {
1773 	if (ct_info->helper)
1774 		nf_conntrack_helper_put(ct_info->helper);
1775 	if (ct_info->ct)
1776 		nf_ct_tmpl_free(ct_info->ct);
1777 }
1778 
1779 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1780 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1781 {
1782 	int i, err;
1783 
1784 	ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1785 					 GFP_KERNEL);
1786 	if (!ovs_net->ct_limit_info)
1787 		return -ENOMEM;
1788 
1789 	ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1790 	ovs_net->ct_limit_info->limits =
1791 		kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1792 			      GFP_KERNEL);
1793 	if (!ovs_net->ct_limit_info->limits) {
1794 		kfree(ovs_net->ct_limit_info);
1795 		return -ENOMEM;
1796 	}
1797 
1798 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1799 		INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1800 
1801 	ovs_net->ct_limit_info->data =
1802 		nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1803 
1804 	if (IS_ERR(ovs_net->ct_limit_info->data)) {
1805 		err = PTR_ERR(ovs_net->ct_limit_info->data);
1806 		kfree(ovs_net->ct_limit_info->limits);
1807 		kfree(ovs_net->ct_limit_info);
1808 		pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1809 		return err;
1810 	}
1811 	return 0;
1812 }
1813 
1814 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1815 {
1816 	const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1817 	int i;
1818 
1819 	nf_conncount_destroy(net, NFPROTO_INET, info->data);
1820 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1821 		struct hlist_head *head = &info->limits[i];
1822 		struct ovs_ct_limit *ct_limit;
1823 
1824 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node)
1825 			kfree_rcu(ct_limit, rcu);
1826 	}
1827 	kfree(ovs_net->ct_limit_info->limits);
1828 	kfree(ovs_net->ct_limit_info);
1829 }
1830 
1831 static struct sk_buff *
1832 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1833 			     struct ovs_header **ovs_reply_header)
1834 {
1835 	struct ovs_header *ovs_header = info->userhdr;
1836 	struct sk_buff *skb;
1837 
1838 	skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1839 	if (!skb)
1840 		return ERR_PTR(-ENOMEM);
1841 
1842 	*ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1843 					info->snd_seq,
1844 					&dp_ct_limit_genl_family, 0, cmd);
1845 
1846 	if (!*ovs_reply_header) {
1847 		nlmsg_free(skb);
1848 		return ERR_PTR(-EMSGSIZE);
1849 	}
1850 	(*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1851 
1852 	return skb;
1853 }
1854 
1855 static bool check_zone_id(int zone_id, u16 *pzone)
1856 {
1857 	if (zone_id >= 0 && zone_id <= 65535) {
1858 		*pzone = (u16)zone_id;
1859 		return true;
1860 	}
1861 	return false;
1862 }
1863 
1864 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1865 				       struct ovs_ct_limit_info *info)
1866 {
1867 	struct ovs_zone_limit *zone_limit;
1868 	int rem;
1869 	u16 zone;
1870 
1871 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1872 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1873 
1874 	while (rem >= sizeof(*zone_limit)) {
1875 		if (unlikely(zone_limit->zone_id ==
1876 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1877 			ovs_lock();
1878 			info->default_limit = zone_limit->limit;
1879 			ovs_unlock();
1880 		} else if (unlikely(!check_zone_id(
1881 				zone_limit->zone_id, &zone))) {
1882 			OVS_NLERR(true, "zone id is out of range");
1883 		} else {
1884 			struct ovs_ct_limit *ct_limit;
1885 
1886 			ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1887 			if (!ct_limit)
1888 				return -ENOMEM;
1889 
1890 			ct_limit->zone = zone;
1891 			ct_limit->limit = zone_limit->limit;
1892 
1893 			ovs_lock();
1894 			ct_limit_set(info, ct_limit);
1895 			ovs_unlock();
1896 		}
1897 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1898 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1899 				NLA_ALIGN(sizeof(*zone_limit)));
1900 	}
1901 
1902 	if (rem)
1903 		OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1904 
1905 	return 0;
1906 }
1907 
1908 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1909 				       struct ovs_ct_limit_info *info)
1910 {
1911 	struct ovs_zone_limit *zone_limit;
1912 	int rem;
1913 	u16 zone;
1914 
1915 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1916 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1917 
1918 	while (rem >= sizeof(*zone_limit)) {
1919 		if (unlikely(zone_limit->zone_id ==
1920 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1921 			ovs_lock();
1922 			info->default_limit = OVS_CT_LIMIT_DEFAULT;
1923 			ovs_unlock();
1924 		} else if (unlikely(!check_zone_id(
1925 				zone_limit->zone_id, &zone))) {
1926 			OVS_NLERR(true, "zone id is out of range");
1927 		} else {
1928 			ovs_lock();
1929 			ct_limit_del(info, zone);
1930 			ovs_unlock();
1931 		}
1932 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1933 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1934 				NLA_ALIGN(sizeof(*zone_limit)));
1935 	}
1936 
1937 	if (rem)
1938 		OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
1939 
1940 	return 0;
1941 }
1942 
1943 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
1944 					  struct sk_buff *reply)
1945 {
1946 	struct ovs_zone_limit zone_limit;
1947 	int err;
1948 
1949 	zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
1950 	zone_limit.limit = info->default_limit;
1951 	err = nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1952 	if (err)
1953 		return err;
1954 
1955 	return 0;
1956 }
1957 
1958 static int __ovs_ct_limit_get_zone_limit(struct net *net,
1959 					 struct nf_conncount_data *data,
1960 					 u16 zone_id, u32 limit,
1961 					 struct sk_buff *reply)
1962 {
1963 	struct nf_conntrack_zone ct_zone;
1964 	struct ovs_zone_limit zone_limit;
1965 	u32 conncount_key = zone_id;
1966 
1967 	zone_limit.zone_id = zone_id;
1968 	zone_limit.limit = limit;
1969 	nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
1970 
1971 	zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
1972 					      &ct_zone);
1973 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1974 }
1975 
1976 static int ovs_ct_limit_get_zone_limit(struct net *net,
1977 				       struct nlattr *nla_zone_limit,
1978 				       struct ovs_ct_limit_info *info,
1979 				       struct sk_buff *reply)
1980 {
1981 	struct ovs_zone_limit *zone_limit;
1982 	int rem, err;
1983 	u32 limit;
1984 	u16 zone;
1985 
1986 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1987 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1988 
1989 	while (rem >= sizeof(*zone_limit)) {
1990 		if (unlikely(zone_limit->zone_id ==
1991 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1992 			err = ovs_ct_limit_get_default_limit(info, reply);
1993 			if (err)
1994 				return err;
1995 		} else if (unlikely(!check_zone_id(zone_limit->zone_id,
1996 							&zone))) {
1997 			OVS_NLERR(true, "zone id is out of range");
1998 		} else {
1999 			rcu_read_lock();
2000 			limit = ct_limit_get(info, zone);
2001 			rcu_read_unlock();
2002 
2003 			err = __ovs_ct_limit_get_zone_limit(
2004 				net, info->data, zone, limit, reply);
2005 			if (err)
2006 				return err;
2007 		}
2008 		rem -= NLA_ALIGN(sizeof(*zone_limit));
2009 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2010 				NLA_ALIGN(sizeof(*zone_limit)));
2011 	}
2012 
2013 	if (rem)
2014 		OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2015 
2016 	return 0;
2017 }
2018 
2019 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2020 					   struct ovs_ct_limit_info *info,
2021 					   struct sk_buff *reply)
2022 {
2023 	struct ovs_ct_limit *ct_limit;
2024 	struct hlist_head *head;
2025 	int i, err = 0;
2026 
2027 	err = ovs_ct_limit_get_default_limit(info, reply);
2028 	if (err)
2029 		return err;
2030 
2031 	rcu_read_lock();
2032 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2033 		head = &info->limits[i];
2034 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2035 			err = __ovs_ct_limit_get_zone_limit(net, info->data,
2036 				ct_limit->zone, ct_limit->limit, reply);
2037 			if (err)
2038 				goto exit_err;
2039 		}
2040 	}
2041 
2042 exit_err:
2043 	rcu_read_unlock();
2044 	return err;
2045 }
2046 
2047 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2048 {
2049 	struct nlattr **a = info->attrs;
2050 	struct sk_buff *reply;
2051 	struct ovs_header *ovs_reply_header;
2052 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2053 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2054 	int err;
2055 
2056 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2057 					     &ovs_reply_header);
2058 	if (IS_ERR(reply))
2059 		return PTR_ERR(reply);
2060 
2061 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2062 		err = -EINVAL;
2063 		goto exit_err;
2064 	}
2065 
2066 	err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2067 					  ct_limit_info);
2068 	if (err)
2069 		goto exit_err;
2070 
2071 	static_branch_enable(&ovs_ct_limit_enabled);
2072 
2073 	genlmsg_end(reply, ovs_reply_header);
2074 	return genlmsg_reply(reply, info);
2075 
2076 exit_err:
2077 	nlmsg_free(reply);
2078 	return err;
2079 }
2080 
2081 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2082 {
2083 	struct nlattr **a = info->attrs;
2084 	struct sk_buff *reply;
2085 	struct ovs_header *ovs_reply_header;
2086 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2087 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2088 	int err;
2089 
2090 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2091 					     &ovs_reply_header);
2092 	if (IS_ERR(reply))
2093 		return PTR_ERR(reply);
2094 
2095 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2096 		err = -EINVAL;
2097 		goto exit_err;
2098 	}
2099 
2100 	err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2101 					  ct_limit_info);
2102 	if (err)
2103 		goto exit_err;
2104 
2105 	genlmsg_end(reply, ovs_reply_header);
2106 	return genlmsg_reply(reply, info);
2107 
2108 exit_err:
2109 	nlmsg_free(reply);
2110 	return err;
2111 }
2112 
2113 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2114 {
2115 	struct nlattr **a = info->attrs;
2116 	struct nlattr *nla_reply;
2117 	struct sk_buff *reply;
2118 	struct ovs_header *ovs_reply_header;
2119 	struct net *net = sock_net(skb->sk);
2120 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2121 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2122 	int err;
2123 
2124 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2125 					     &ovs_reply_header);
2126 	if (IS_ERR(reply))
2127 		return PTR_ERR(reply);
2128 
2129 	nla_reply = nla_nest_start(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2130 
2131 	if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2132 		err = ovs_ct_limit_get_zone_limit(
2133 			net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2134 			reply);
2135 		if (err)
2136 			goto exit_err;
2137 	} else {
2138 		err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2139 						      reply);
2140 		if (err)
2141 			goto exit_err;
2142 	}
2143 
2144 	nla_nest_end(reply, nla_reply);
2145 	genlmsg_end(reply, ovs_reply_header);
2146 	return genlmsg_reply(reply, info);
2147 
2148 exit_err:
2149 	nlmsg_free(reply);
2150 	return err;
2151 }
2152 
2153 static struct genl_ops ct_limit_genl_ops[] = {
2154 	{ .cmd = OVS_CT_LIMIT_CMD_SET,
2155 		.flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2156 					   * privilege. */
2157 		.policy = ct_limit_policy,
2158 		.doit = ovs_ct_limit_cmd_set,
2159 	},
2160 	{ .cmd = OVS_CT_LIMIT_CMD_DEL,
2161 		.flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2162 					   * privilege. */
2163 		.policy = ct_limit_policy,
2164 		.doit = ovs_ct_limit_cmd_del,
2165 	},
2166 	{ .cmd = OVS_CT_LIMIT_CMD_GET,
2167 		.flags = 0,		  /* OK for unprivileged users. */
2168 		.policy = ct_limit_policy,
2169 		.doit = ovs_ct_limit_cmd_get,
2170 	},
2171 };
2172 
2173 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2174 	.name = OVS_CT_LIMIT_MCGROUP,
2175 };
2176 
2177 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2178 	.hdrsize = sizeof(struct ovs_header),
2179 	.name = OVS_CT_LIMIT_FAMILY,
2180 	.version = OVS_CT_LIMIT_VERSION,
2181 	.maxattr = OVS_CT_LIMIT_ATTR_MAX,
2182 	.netnsok = true,
2183 	.parallel_ops = true,
2184 	.ops = ct_limit_genl_ops,
2185 	.n_ops = ARRAY_SIZE(ct_limit_genl_ops),
2186 	.mcgrps = &ovs_ct_limit_multicast_group,
2187 	.n_mcgrps = 1,
2188 	.module = THIS_MODULE,
2189 };
2190 #endif
2191 
2192 int ovs_ct_init(struct net *net)
2193 {
2194 	unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2195 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2196 
2197 	if (nf_connlabels_get(net, n_bits - 1)) {
2198 		ovs_net->xt_label = false;
2199 		OVS_NLERR(true, "Failed to set connlabel length");
2200 	} else {
2201 		ovs_net->xt_label = true;
2202 	}
2203 
2204 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2205 	return ovs_ct_limit_init(net, ovs_net);
2206 #else
2207 	return 0;
2208 #endif
2209 }
2210 
2211 void ovs_ct_exit(struct net *net)
2212 {
2213 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2214 
2215 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2216 	ovs_ct_limit_exit(net, ovs_net);
2217 #endif
2218 
2219 	if (ovs_net->xt_label)
2220 		nf_connlabels_put(net);
2221 }
2222