xref: /openbmc/linux/net/openvswitch/conntrack.c (revision e3d786a3)
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 <linux/netfilter/nf_nat.h>
33 #include <net/netfilter/nf_nat_core.h>
34 #include <net/netfilter/nf_nat_l3proto.h>
35 #endif
36 
37 #include "datapath.h"
38 #include "conntrack.h"
39 #include "flow.h"
40 #include "flow_netlink.h"
41 
42 struct ovs_ct_len_tbl {
43 	int maxlen;
44 	int minlen;
45 };
46 
47 /* Metadata mark for masked write to conntrack mark */
48 struct md_mark {
49 	u32 value;
50 	u32 mask;
51 };
52 
53 /* Metadata label for masked write to conntrack label. */
54 struct md_labels {
55 	struct ovs_key_ct_labels value;
56 	struct ovs_key_ct_labels mask;
57 };
58 
59 enum ovs_ct_nat {
60 	OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
61 	OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
62 	OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
63 };
64 
65 /* Conntrack action context for execution. */
66 struct ovs_conntrack_info {
67 	struct nf_conntrack_helper *helper;
68 	struct nf_conntrack_zone zone;
69 	struct nf_conn *ct;
70 	u8 commit : 1;
71 	u8 nat : 3;                 /* enum ovs_ct_nat */
72 	u8 force : 1;
73 	u8 have_eventmask : 1;
74 	u16 family;
75 	u32 eventmask;              /* Mask of 1 << IPCT_*. */
76 	struct md_mark mark;
77 	struct md_labels labels;
78 #ifdef CONFIG_NF_NAT_NEEDED
79 	struct nf_nat_range2 range;  /* Only present for SRC NAT and DST NAT. */
80 #endif
81 };
82 
83 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
84 #define OVS_CT_LIMIT_UNLIMITED	0
85 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
86 #define CT_LIMIT_HASH_BUCKETS 512
87 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
88 
89 struct ovs_ct_limit {
90 	/* Elements in ovs_ct_limit_info->limits hash table */
91 	struct hlist_node hlist_node;
92 	struct rcu_head rcu;
93 	u16 zone;
94 	u32 limit;
95 };
96 
97 struct ovs_ct_limit_info {
98 	u32 default_limit;
99 	struct hlist_head *limits;
100 	struct nf_conncount_data *data;
101 };
102 
103 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
104 	[OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
105 };
106 #endif
107 
108 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
109 
110 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
111 
112 static u16 key_to_nfproto(const struct sw_flow_key *key)
113 {
114 	switch (ntohs(key->eth.type)) {
115 	case ETH_P_IP:
116 		return NFPROTO_IPV4;
117 	case ETH_P_IPV6:
118 		return NFPROTO_IPV6;
119 	default:
120 		return NFPROTO_UNSPEC;
121 	}
122 }
123 
124 /* Map SKB connection state into the values used by flow definition. */
125 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
126 {
127 	u8 ct_state = OVS_CS_F_TRACKED;
128 
129 	switch (ctinfo) {
130 	case IP_CT_ESTABLISHED_REPLY:
131 	case IP_CT_RELATED_REPLY:
132 		ct_state |= OVS_CS_F_REPLY_DIR;
133 		break;
134 	default:
135 		break;
136 	}
137 
138 	switch (ctinfo) {
139 	case IP_CT_ESTABLISHED:
140 	case IP_CT_ESTABLISHED_REPLY:
141 		ct_state |= OVS_CS_F_ESTABLISHED;
142 		break;
143 	case IP_CT_RELATED:
144 	case IP_CT_RELATED_REPLY:
145 		ct_state |= OVS_CS_F_RELATED;
146 		break;
147 	case IP_CT_NEW:
148 		ct_state |= OVS_CS_F_NEW;
149 		break;
150 	default:
151 		break;
152 	}
153 
154 	return ct_state;
155 }
156 
157 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
158 {
159 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
160 	return ct ? ct->mark : 0;
161 #else
162 	return 0;
163 #endif
164 }
165 
166 /* Guard against conntrack labels max size shrinking below 128 bits. */
167 #if NF_CT_LABELS_MAX_SIZE < 16
168 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
169 #endif
170 
171 static void ovs_ct_get_labels(const struct nf_conn *ct,
172 			      struct ovs_key_ct_labels *labels)
173 {
174 	struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
175 
176 	if (cl)
177 		memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
178 	else
179 		memset(labels, 0, OVS_CT_LABELS_LEN);
180 }
181 
182 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
183 					const struct nf_conntrack_tuple *orig,
184 					u8 icmp_proto)
185 {
186 	key->ct_orig_proto = orig->dst.protonum;
187 	if (orig->dst.protonum == icmp_proto) {
188 		key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
189 		key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
190 	} else {
191 		key->ct.orig_tp.src = orig->src.u.all;
192 		key->ct.orig_tp.dst = orig->dst.u.all;
193 	}
194 }
195 
196 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
197 				const struct nf_conntrack_zone *zone,
198 				const struct nf_conn *ct)
199 {
200 	key->ct_state = state;
201 	key->ct_zone = zone->id;
202 	key->ct.mark = ovs_ct_get_mark(ct);
203 	ovs_ct_get_labels(ct, &key->ct.labels);
204 
205 	if (ct) {
206 		const struct nf_conntrack_tuple *orig;
207 
208 		/* Use the master if we have one. */
209 		if (ct->master)
210 			ct = ct->master;
211 		orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
212 
213 		/* IP version must match with the master connection. */
214 		if (key->eth.type == htons(ETH_P_IP) &&
215 		    nf_ct_l3num(ct) == NFPROTO_IPV4) {
216 			key->ipv4.ct_orig.src = orig->src.u3.ip;
217 			key->ipv4.ct_orig.dst = orig->dst.u3.ip;
218 			__ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
219 			return;
220 		} else if (key->eth.type == htons(ETH_P_IPV6) &&
221 			   !sw_flow_key_is_nd(key) &&
222 			   nf_ct_l3num(ct) == NFPROTO_IPV6) {
223 			key->ipv6.ct_orig.src = orig->src.u3.in6;
224 			key->ipv6.ct_orig.dst = orig->dst.u3.in6;
225 			__ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
226 			return;
227 		}
228 	}
229 	/* Clear 'ct_orig_proto' to mark the non-existence of conntrack
230 	 * original direction key fields.
231 	 */
232 	key->ct_orig_proto = 0;
233 }
234 
235 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
236  * previously sent the packet to conntrack via the ct action.  If
237  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
238  * initialized from the connection status.
239  */
240 static void ovs_ct_update_key(const struct sk_buff *skb,
241 			      const struct ovs_conntrack_info *info,
242 			      struct sw_flow_key *key, bool post_ct,
243 			      bool keep_nat_flags)
244 {
245 	const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
246 	enum ip_conntrack_info ctinfo;
247 	struct nf_conn *ct;
248 	u8 state = 0;
249 
250 	ct = nf_ct_get(skb, &ctinfo);
251 	if (ct) {
252 		state = ovs_ct_get_state(ctinfo);
253 		/* All unconfirmed entries are NEW connections. */
254 		if (!nf_ct_is_confirmed(ct))
255 			state |= OVS_CS_F_NEW;
256 		/* OVS persists the related flag for the duration of the
257 		 * connection.
258 		 */
259 		if (ct->master)
260 			state |= OVS_CS_F_RELATED;
261 		if (keep_nat_flags) {
262 			state |= key->ct_state & OVS_CS_F_NAT_MASK;
263 		} else {
264 			if (ct->status & IPS_SRC_NAT)
265 				state |= OVS_CS_F_SRC_NAT;
266 			if (ct->status & IPS_DST_NAT)
267 				state |= OVS_CS_F_DST_NAT;
268 		}
269 		zone = nf_ct_zone(ct);
270 	} else if (post_ct) {
271 		state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
272 		if (info)
273 			zone = &info->zone;
274 	}
275 	__ovs_ct_update_key(key, state, zone, ct);
276 }
277 
278 /* This is called to initialize CT key fields possibly coming in from the local
279  * stack.
280  */
281 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
282 {
283 	ovs_ct_update_key(skb, NULL, key, false, false);
284 }
285 
286 #define IN6_ADDR_INITIALIZER(ADDR) \
287 	{ (ADDR).s6_addr32[0], (ADDR).s6_addr32[1], \
288 	  (ADDR).s6_addr32[2], (ADDR).s6_addr32[3] }
289 
290 int ovs_ct_put_key(const struct sw_flow_key *swkey,
291 		   const struct sw_flow_key *output, struct sk_buff *skb)
292 {
293 	if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
294 		return -EMSGSIZE;
295 
296 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
297 	    nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
298 		return -EMSGSIZE;
299 
300 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
301 	    nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
302 		return -EMSGSIZE;
303 
304 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
305 	    nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
306 		    &output->ct.labels))
307 		return -EMSGSIZE;
308 
309 	if (swkey->ct_orig_proto) {
310 		if (swkey->eth.type == htons(ETH_P_IP)) {
311 			struct ovs_key_ct_tuple_ipv4 orig = {
312 				output->ipv4.ct_orig.src,
313 				output->ipv4.ct_orig.dst,
314 				output->ct.orig_tp.src,
315 				output->ct.orig_tp.dst,
316 				output->ct_orig_proto,
317 			};
318 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
319 				    sizeof(orig), &orig))
320 				return -EMSGSIZE;
321 		} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
322 			struct ovs_key_ct_tuple_ipv6 orig = {
323 				IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.src),
324 				IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.dst),
325 				output->ct.orig_tp.src,
326 				output->ct.orig_tp.dst,
327 				output->ct_orig_proto,
328 			};
329 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
330 				    sizeof(orig), &orig))
331 				return -EMSGSIZE;
332 		}
333 	}
334 
335 	return 0;
336 }
337 
338 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
339 			   u32 ct_mark, u32 mask)
340 {
341 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
342 	u32 new_mark;
343 
344 	new_mark = ct_mark | (ct->mark & ~(mask));
345 	if (ct->mark != new_mark) {
346 		ct->mark = new_mark;
347 		if (nf_ct_is_confirmed(ct))
348 			nf_conntrack_event_cache(IPCT_MARK, ct);
349 		key->ct.mark = new_mark;
350 	}
351 
352 	return 0;
353 #else
354 	return -ENOTSUPP;
355 #endif
356 }
357 
358 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
359 {
360 	struct nf_conn_labels *cl;
361 
362 	cl = nf_ct_labels_find(ct);
363 	if (!cl) {
364 		nf_ct_labels_ext_add(ct);
365 		cl = nf_ct_labels_find(ct);
366 	}
367 
368 	return cl;
369 }
370 
371 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
372  * since the new connection is not yet confirmed, and thus no-one else has
373  * access to it's labels, we simply write them over.
374  */
375 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
376 			      const struct ovs_key_ct_labels *labels,
377 			      const struct ovs_key_ct_labels *mask)
378 {
379 	struct nf_conn_labels *cl, *master_cl;
380 	bool have_mask = labels_nonzero(mask);
381 
382 	/* Inherit master's labels to the related connection? */
383 	master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
384 
385 	if (!master_cl && !have_mask)
386 		return 0;   /* Nothing to do. */
387 
388 	cl = ovs_ct_get_conn_labels(ct);
389 	if (!cl)
390 		return -ENOSPC;
391 
392 	/* Inherit the master's labels, if any. */
393 	if (master_cl)
394 		*cl = *master_cl;
395 
396 	if (have_mask) {
397 		u32 *dst = (u32 *)cl->bits;
398 		int i;
399 
400 		for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
401 			dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
402 				(labels->ct_labels_32[i]
403 				 & mask->ct_labels_32[i]);
404 	}
405 
406 	/* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
407 	 * IPCT_LABEL bit is set in the event cache.
408 	 */
409 	nf_conntrack_event_cache(IPCT_LABEL, ct);
410 
411 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
412 
413 	return 0;
414 }
415 
416 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
417 			     const struct ovs_key_ct_labels *labels,
418 			     const struct ovs_key_ct_labels *mask)
419 {
420 	struct nf_conn_labels *cl;
421 	int err;
422 
423 	cl = ovs_ct_get_conn_labels(ct);
424 	if (!cl)
425 		return -ENOSPC;
426 
427 	err = nf_connlabels_replace(ct, labels->ct_labels_32,
428 				    mask->ct_labels_32,
429 				    OVS_CT_LABELS_LEN_32);
430 	if (err)
431 		return err;
432 
433 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
434 
435 	return 0;
436 }
437 
438 /* 'skb' should already be pulled to nh_ofs. */
439 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
440 {
441 	const struct nf_conntrack_helper *helper;
442 	const struct nf_conn_help *help;
443 	enum ip_conntrack_info ctinfo;
444 	unsigned int protoff;
445 	struct nf_conn *ct;
446 	int err;
447 
448 	ct = nf_ct_get(skb, &ctinfo);
449 	if (!ct || ctinfo == IP_CT_RELATED_REPLY)
450 		return NF_ACCEPT;
451 
452 	help = nfct_help(ct);
453 	if (!help)
454 		return NF_ACCEPT;
455 
456 	helper = rcu_dereference(help->helper);
457 	if (!helper)
458 		return NF_ACCEPT;
459 
460 	switch (proto) {
461 	case NFPROTO_IPV4:
462 		protoff = ip_hdrlen(skb);
463 		break;
464 	case NFPROTO_IPV6: {
465 		u8 nexthdr = ipv6_hdr(skb)->nexthdr;
466 		__be16 frag_off;
467 		int ofs;
468 
469 		ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
470 				       &frag_off);
471 		if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
472 			pr_debug("proto header not found\n");
473 			return NF_ACCEPT;
474 		}
475 		protoff = ofs;
476 		break;
477 	}
478 	default:
479 		WARN_ONCE(1, "helper invoked on non-IP family!");
480 		return NF_DROP;
481 	}
482 
483 	err = helper->help(skb, protoff, ct, ctinfo);
484 	if (err != NF_ACCEPT)
485 		return err;
486 
487 	/* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
488 	 * FTP with NAT) adusting the TCP payload size when mangling IP
489 	 * addresses and/or port numbers in the text-based control connection.
490 	 */
491 	if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
492 	    !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
493 		return NF_DROP;
494 	return NF_ACCEPT;
495 }
496 
497 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
498  * value if 'skb' is freed.
499  */
500 static int handle_fragments(struct net *net, struct sw_flow_key *key,
501 			    u16 zone, struct sk_buff *skb)
502 {
503 	struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
504 	int err;
505 
506 	if (key->eth.type == htons(ETH_P_IP)) {
507 		enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
508 
509 		memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
510 		err = ip_defrag(net, skb, user);
511 		if (err)
512 			return err;
513 
514 		ovs_cb.mru = IPCB(skb)->frag_max_size;
515 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
516 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
517 		enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
518 
519 		memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
520 		err = nf_ct_frag6_gather(net, skb, user);
521 		if (err) {
522 			if (err != -EINPROGRESS)
523 				kfree_skb(skb);
524 			return err;
525 		}
526 
527 		key->ip.proto = ipv6_hdr(skb)->nexthdr;
528 		ovs_cb.mru = IP6CB(skb)->frag_max_size;
529 #endif
530 	} else {
531 		kfree_skb(skb);
532 		return -EPFNOSUPPORT;
533 	}
534 
535 	key->ip.frag = OVS_FRAG_TYPE_NONE;
536 	skb_clear_hash(skb);
537 	skb->ignore_df = 1;
538 	*OVS_CB(skb) = ovs_cb;
539 
540 	return 0;
541 }
542 
543 static struct nf_conntrack_expect *
544 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
545 		   u16 proto, const struct sk_buff *skb)
546 {
547 	struct nf_conntrack_tuple tuple;
548 	struct nf_conntrack_expect *exp;
549 
550 	if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
551 		return NULL;
552 
553 	exp = __nf_ct_expect_find(net, zone, &tuple);
554 	if (exp) {
555 		struct nf_conntrack_tuple_hash *h;
556 
557 		/* Delete existing conntrack entry, if it clashes with the
558 		 * expectation.  This can happen since conntrack ALGs do not
559 		 * check for clashes between (new) expectations and existing
560 		 * conntrack entries.  nf_conntrack_in() will check the
561 		 * expectations only if a conntrack entry can not be found,
562 		 * which can lead to OVS finding the expectation (here) in the
563 		 * init direction, but which will not be removed by the
564 		 * nf_conntrack_in() call, if a matching conntrack entry is
565 		 * found instead.  In this case all init direction packets
566 		 * would be reported as new related packets, while reply
567 		 * direction packets would be reported as un-related
568 		 * established packets.
569 		 */
570 		h = nf_conntrack_find_get(net, zone, &tuple);
571 		if (h) {
572 			struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
573 
574 			nf_ct_delete(ct, 0, 0);
575 			nf_conntrack_put(&ct->ct_general);
576 		}
577 	}
578 
579 	return exp;
580 }
581 
582 /* This replicates logic from nf_conntrack_core.c that is not exported. */
583 static enum ip_conntrack_info
584 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
585 {
586 	const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
587 
588 	if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
589 		return IP_CT_ESTABLISHED_REPLY;
590 	/* Once we've had two way comms, always ESTABLISHED. */
591 	if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
592 		return IP_CT_ESTABLISHED;
593 	if (test_bit(IPS_EXPECTED_BIT, &ct->status))
594 		return IP_CT_RELATED;
595 	return IP_CT_NEW;
596 }
597 
598 /* Find an existing connection which this packet belongs to without
599  * re-attributing statistics or modifying the connection state.  This allows an
600  * skb->_nfct lost due to an upcall to be recovered during actions execution.
601  *
602  * Must be called with rcu_read_lock.
603  *
604  * On success, populates skb->_nfct and returns the connection.  Returns NULL
605  * if there is no existing entry.
606  */
607 static struct nf_conn *
608 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
609 		     u8 l3num, struct sk_buff *skb, bool natted)
610 {
611 	struct nf_conntrack_tuple tuple;
612 	struct nf_conntrack_tuple_hash *h;
613 	struct nf_conn *ct;
614 
615 	if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
616 			       net, &tuple)) {
617 		pr_debug("ovs_ct_find_existing: Can't get tuple\n");
618 		return NULL;
619 	}
620 
621 	/* Must invert the tuple if skb has been transformed by NAT. */
622 	if (natted) {
623 		struct nf_conntrack_tuple inverse;
624 
625 		if (!nf_ct_invert_tuplepr(&inverse, &tuple)) {
626 			pr_debug("ovs_ct_find_existing: Inversion failed!\n");
627 			return NULL;
628 		}
629 		tuple = inverse;
630 	}
631 
632 	/* look for tuple match */
633 	h = nf_conntrack_find_get(net, zone, &tuple);
634 	if (!h)
635 		return NULL;   /* Not found. */
636 
637 	ct = nf_ct_tuplehash_to_ctrack(h);
638 
639 	/* Inverted packet tuple matches the reverse direction conntrack tuple,
640 	 * select the other tuplehash to get the right 'ctinfo' bits for this
641 	 * packet.
642 	 */
643 	if (natted)
644 		h = &ct->tuplehash[!h->tuple.dst.dir];
645 
646 	nf_ct_set(skb, ct, ovs_ct_get_info(h));
647 	return ct;
648 }
649 
650 static
651 struct nf_conn *ovs_ct_executed(struct net *net,
652 				const struct sw_flow_key *key,
653 				const struct ovs_conntrack_info *info,
654 				struct sk_buff *skb,
655 				bool *ct_executed)
656 {
657 	struct nf_conn *ct = NULL;
658 
659 	/* If no ct, check if we have evidence that an existing conntrack entry
660 	 * might be found for this skb.  This happens when we lose a skb->_nfct
661 	 * due to an upcall, or if the direction is being forced.  If the
662 	 * connection was not confirmed, it is not cached and needs to be run
663 	 * through conntrack again.
664 	 */
665 	*ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
666 		       !(key->ct_state & OVS_CS_F_INVALID) &&
667 		       (key->ct_zone == info->zone.id);
668 
669 	if (*ct_executed || (!key->ct_state && info->force)) {
670 		ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
671 					  !!(key->ct_state &
672 					  OVS_CS_F_NAT_MASK));
673 	}
674 
675 	return ct;
676 }
677 
678 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
679 static bool skb_nfct_cached(struct net *net,
680 			    const struct sw_flow_key *key,
681 			    const struct ovs_conntrack_info *info,
682 			    struct sk_buff *skb)
683 {
684 	enum ip_conntrack_info ctinfo;
685 	struct nf_conn *ct;
686 	bool ct_executed = true;
687 
688 	ct = nf_ct_get(skb, &ctinfo);
689 	if (!ct)
690 		ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
691 
692 	if (ct)
693 		nf_ct_get(skb, &ctinfo);
694 	else
695 		return false;
696 
697 	if (!net_eq(net, read_pnet(&ct->ct_net)))
698 		return false;
699 	if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
700 		return false;
701 	if (info->helper) {
702 		struct nf_conn_help *help;
703 
704 		help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
705 		if (help && rcu_access_pointer(help->helper) != info->helper)
706 			return false;
707 	}
708 	/* Force conntrack entry direction to the current packet? */
709 	if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
710 		/* Delete the conntrack entry if confirmed, else just release
711 		 * the reference.
712 		 */
713 		if (nf_ct_is_confirmed(ct))
714 			nf_ct_delete(ct, 0, 0);
715 
716 		nf_conntrack_put(&ct->ct_general);
717 		nf_ct_set(skb, NULL, 0);
718 		return false;
719 	}
720 
721 	return ct_executed;
722 }
723 
724 #ifdef CONFIG_NF_NAT_NEEDED
725 /* Modelled after nf_nat_ipv[46]_fn().
726  * range is only used for new, uninitialized NAT state.
727  * Returns either NF_ACCEPT or NF_DROP.
728  */
729 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
730 			      enum ip_conntrack_info ctinfo,
731 			      const struct nf_nat_range2 *range,
732 			      enum nf_nat_manip_type maniptype)
733 {
734 	int hooknum, nh_off, err = NF_ACCEPT;
735 
736 	nh_off = skb_network_offset(skb);
737 	skb_pull_rcsum(skb, nh_off);
738 
739 	/* See HOOK2MANIP(). */
740 	if (maniptype == NF_NAT_MANIP_SRC)
741 		hooknum = NF_INET_LOCAL_IN; /* Source NAT */
742 	else
743 		hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
744 
745 	switch (ctinfo) {
746 	case IP_CT_RELATED:
747 	case IP_CT_RELATED_REPLY:
748 		if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
749 		    skb->protocol == htons(ETH_P_IP) &&
750 		    ip_hdr(skb)->protocol == IPPROTO_ICMP) {
751 			if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
752 							   hooknum))
753 				err = NF_DROP;
754 			goto push;
755 		} else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
756 			   skb->protocol == htons(ETH_P_IPV6)) {
757 			__be16 frag_off;
758 			u8 nexthdr = ipv6_hdr(skb)->nexthdr;
759 			int hdrlen = ipv6_skip_exthdr(skb,
760 						      sizeof(struct ipv6hdr),
761 						      &nexthdr, &frag_off);
762 
763 			if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
764 				if (!nf_nat_icmpv6_reply_translation(skb, ct,
765 								     ctinfo,
766 								     hooknum,
767 								     hdrlen))
768 					err = NF_DROP;
769 				goto push;
770 			}
771 		}
772 		/* Non-ICMP, fall thru to initialize if needed. */
773 		/* fall through */
774 	case IP_CT_NEW:
775 		/* Seen it before?  This can happen for loopback, retrans,
776 		 * or local packets.
777 		 */
778 		if (!nf_nat_initialized(ct, maniptype)) {
779 			/* Initialize according to the NAT action. */
780 			err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
781 				/* Action is set up to establish a new
782 				 * mapping.
783 				 */
784 				? nf_nat_setup_info(ct, range, maniptype)
785 				: nf_nat_alloc_null_binding(ct, hooknum);
786 			if (err != NF_ACCEPT)
787 				goto push;
788 		}
789 		break;
790 
791 	case IP_CT_ESTABLISHED:
792 	case IP_CT_ESTABLISHED_REPLY:
793 		break;
794 
795 	default:
796 		err = NF_DROP;
797 		goto push;
798 	}
799 
800 	err = nf_nat_packet(ct, ctinfo, hooknum, skb);
801 push:
802 	skb_push(skb, nh_off);
803 	skb_postpush_rcsum(skb, skb->data, nh_off);
804 
805 	return err;
806 }
807 
808 static void ovs_nat_update_key(struct sw_flow_key *key,
809 			       const struct sk_buff *skb,
810 			       enum nf_nat_manip_type maniptype)
811 {
812 	if (maniptype == NF_NAT_MANIP_SRC) {
813 		__be16 src;
814 
815 		key->ct_state |= OVS_CS_F_SRC_NAT;
816 		if (key->eth.type == htons(ETH_P_IP))
817 			key->ipv4.addr.src = ip_hdr(skb)->saddr;
818 		else if (key->eth.type == htons(ETH_P_IPV6))
819 			memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
820 			       sizeof(key->ipv6.addr.src));
821 		else
822 			return;
823 
824 		if (key->ip.proto == IPPROTO_UDP)
825 			src = udp_hdr(skb)->source;
826 		else if (key->ip.proto == IPPROTO_TCP)
827 			src = tcp_hdr(skb)->source;
828 		else if (key->ip.proto == IPPROTO_SCTP)
829 			src = sctp_hdr(skb)->source;
830 		else
831 			return;
832 
833 		key->tp.src = src;
834 	} else {
835 		__be16 dst;
836 
837 		key->ct_state |= OVS_CS_F_DST_NAT;
838 		if (key->eth.type == htons(ETH_P_IP))
839 			key->ipv4.addr.dst = ip_hdr(skb)->daddr;
840 		else if (key->eth.type == htons(ETH_P_IPV6))
841 			memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
842 			       sizeof(key->ipv6.addr.dst));
843 		else
844 			return;
845 
846 		if (key->ip.proto == IPPROTO_UDP)
847 			dst = udp_hdr(skb)->dest;
848 		else if (key->ip.proto == IPPROTO_TCP)
849 			dst = tcp_hdr(skb)->dest;
850 		else if (key->ip.proto == IPPROTO_SCTP)
851 			dst = sctp_hdr(skb)->dest;
852 		else
853 			return;
854 
855 		key->tp.dst = dst;
856 	}
857 }
858 
859 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
860 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
861 		      const struct ovs_conntrack_info *info,
862 		      struct sk_buff *skb, struct nf_conn *ct,
863 		      enum ip_conntrack_info ctinfo)
864 {
865 	enum nf_nat_manip_type maniptype;
866 	int err;
867 
868 	/* Add NAT extension if not confirmed yet. */
869 	if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
870 		return NF_ACCEPT;   /* Can't NAT. */
871 
872 	/* Determine NAT type.
873 	 * Check if the NAT type can be deduced from the tracked connection.
874 	 * Make sure new expected connections (IP_CT_RELATED) are NATted only
875 	 * when committing.
876 	 */
877 	if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
878 	    ct->status & IPS_NAT_MASK &&
879 	    (ctinfo != IP_CT_RELATED || info->commit)) {
880 		/* NAT an established or related connection like before. */
881 		if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
882 			/* This is the REPLY direction for a connection
883 			 * for which NAT was applied in the forward
884 			 * direction.  Do the reverse NAT.
885 			 */
886 			maniptype = ct->status & IPS_SRC_NAT
887 				? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
888 		else
889 			maniptype = ct->status & IPS_SRC_NAT
890 				? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
891 	} else if (info->nat & OVS_CT_SRC_NAT) {
892 		maniptype = NF_NAT_MANIP_SRC;
893 	} else if (info->nat & OVS_CT_DST_NAT) {
894 		maniptype = NF_NAT_MANIP_DST;
895 	} else {
896 		return NF_ACCEPT; /* Connection is not NATed. */
897 	}
898 	err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
899 
900 	/* Mark NAT done if successful and update the flow key. */
901 	if (err == NF_ACCEPT)
902 		ovs_nat_update_key(key, skb, maniptype);
903 
904 	return err;
905 }
906 #else /* !CONFIG_NF_NAT_NEEDED */
907 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
908 		      const struct ovs_conntrack_info *info,
909 		      struct sk_buff *skb, struct nf_conn *ct,
910 		      enum ip_conntrack_info ctinfo)
911 {
912 	return NF_ACCEPT;
913 }
914 #endif
915 
916 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
917  * not done already.  Update key with new CT state after passing the packet
918  * through conntrack.
919  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
920  * set to NULL and 0 will be returned.
921  */
922 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
923 			   const struct ovs_conntrack_info *info,
924 			   struct sk_buff *skb)
925 {
926 	/* If we are recirculating packets to match on conntrack fields and
927 	 * committing with a separate conntrack action,  then we don't need to
928 	 * actually run the packet through conntrack twice unless it's for a
929 	 * different zone.
930 	 */
931 	bool cached = skb_nfct_cached(net, key, info, skb);
932 	enum ip_conntrack_info ctinfo;
933 	struct nf_conn *ct;
934 
935 	if (!cached) {
936 		struct nf_hook_state state = {
937 			.hook = NF_INET_PRE_ROUTING,
938 			.pf = info->family,
939 			.net = net,
940 		};
941 		struct nf_conn *tmpl = info->ct;
942 		int err;
943 
944 		/* Associate skb with specified zone. */
945 		if (tmpl) {
946 			if (skb_nfct(skb))
947 				nf_conntrack_put(skb_nfct(skb));
948 			nf_conntrack_get(&tmpl->ct_general);
949 			nf_ct_set(skb, tmpl, IP_CT_NEW);
950 		}
951 
952 		err = nf_conntrack_in(skb, &state);
953 		if (err != NF_ACCEPT)
954 			return -ENOENT;
955 
956 		/* Clear CT state NAT flags to mark that we have not yet done
957 		 * NAT after the nf_conntrack_in() call.  We can actually clear
958 		 * the whole state, as it will be re-initialized below.
959 		 */
960 		key->ct_state = 0;
961 
962 		/* Update the key, but keep the NAT flags. */
963 		ovs_ct_update_key(skb, info, key, true, true);
964 	}
965 
966 	ct = nf_ct_get(skb, &ctinfo);
967 	if (ct) {
968 		/* Packets starting a new connection must be NATted before the
969 		 * helper, so that the helper knows about the NAT.  We enforce
970 		 * this by delaying both NAT and helper calls for unconfirmed
971 		 * connections until the committing CT action.  For later
972 		 * packets NAT and Helper may be called in either order.
973 		 *
974 		 * NAT will be done only if the CT action has NAT, and only
975 		 * once per packet (per zone), as guarded by the NAT bits in
976 		 * the key->ct_state.
977 		 */
978 		if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
979 		    (nf_ct_is_confirmed(ct) || info->commit) &&
980 		    ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
981 			return -EINVAL;
982 		}
983 
984 		/* Userspace may decide to perform a ct lookup without a helper
985 		 * specified followed by a (recirculate and) commit with one.
986 		 * Therefore, for unconfirmed connections which we will commit,
987 		 * we need to attach the helper here.
988 		 */
989 		if (!nf_ct_is_confirmed(ct) && info->commit &&
990 		    info->helper && !nfct_help(ct)) {
991 			int err = __nf_ct_try_assign_helper(ct, info->ct,
992 							    GFP_ATOMIC);
993 			if (err)
994 				return err;
995 		}
996 
997 		/* Call the helper only if:
998 		 * - nf_conntrack_in() was executed above ("!cached") for a
999 		 *   confirmed connection, or
1000 		 * - When committing an unconfirmed connection.
1001 		 */
1002 		if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
1003 		    ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1004 			return -EINVAL;
1005 		}
1006 	}
1007 
1008 	return 0;
1009 }
1010 
1011 /* Lookup connection and read fields into key. */
1012 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1013 			 const struct ovs_conntrack_info *info,
1014 			 struct sk_buff *skb)
1015 {
1016 	struct nf_conntrack_expect *exp;
1017 
1018 	/* If we pass an expected packet through nf_conntrack_in() the
1019 	 * expectation is typically removed, but the packet could still be
1020 	 * lost in upcall processing.  To prevent this from happening we
1021 	 * perform an explicit expectation lookup.  Expected connections are
1022 	 * always new, and will be passed through conntrack only when they are
1023 	 * committed, as it is OK to remove the expectation at that time.
1024 	 */
1025 	exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1026 	if (exp) {
1027 		u8 state;
1028 
1029 		/* NOTE: New connections are NATted and Helped only when
1030 		 * committed, so we are not calling into NAT here.
1031 		 */
1032 		state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1033 		__ovs_ct_update_key(key, state, &info->zone, exp->master);
1034 	} else {
1035 		struct nf_conn *ct;
1036 		int err;
1037 
1038 		err = __ovs_ct_lookup(net, key, info, skb);
1039 		if (err)
1040 			return err;
1041 
1042 		ct = (struct nf_conn *)skb_nfct(skb);
1043 		if (ct)
1044 			nf_ct_deliver_cached_events(ct);
1045 	}
1046 
1047 	return 0;
1048 }
1049 
1050 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1051 {
1052 	size_t i;
1053 
1054 	for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1055 		if (labels->ct_labels_32[i])
1056 			return true;
1057 
1058 	return false;
1059 }
1060 
1061 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1062 static struct hlist_head *ct_limit_hash_bucket(
1063 	const struct ovs_ct_limit_info *info, u16 zone)
1064 {
1065 	return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1066 }
1067 
1068 /* Call with ovs_mutex */
1069 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1070 			 struct ovs_ct_limit *new_ct_limit)
1071 {
1072 	struct ovs_ct_limit *ct_limit;
1073 	struct hlist_head *head;
1074 
1075 	head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1076 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1077 		if (ct_limit->zone == new_ct_limit->zone) {
1078 			hlist_replace_rcu(&ct_limit->hlist_node,
1079 					  &new_ct_limit->hlist_node);
1080 			kfree_rcu(ct_limit, rcu);
1081 			return;
1082 		}
1083 	}
1084 
1085 	hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1086 }
1087 
1088 /* Call with ovs_mutex */
1089 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1090 {
1091 	struct ovs_ct_limit *ct_limit;
1092 	struct hlist_head *head;
1093 	struct hlist_node *n;
1094 
1095 	head = ct_limit_hash_bucket(info, zone);
1096 	hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1097 		if (ct_limit->zone == zone) {
1098 			hlist_del_rcu(&ct_limit->hlist_node);
1099 			kfree_rcu(ct_limit, rcu);
1100 			return;
1101 		}
1102 	}
1103 }
1104 
1105 /* Call with RCU read lock */
1106 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1107 {
1108 	struct ovs_ct_limit *ct_limit;
1109 	struct hlist_head *head;
1110 
1111 	head = ct_limit_hash_bucket(info, zone);
1112 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1113 		if (ct_limit->zone == zone)
1114 			return ct_limit->limit;
1115 	}
1116 
1117 	return info->default_limit;
1118 }
1119 
1120 static int ovs_ct_check_limit(struct net *net,
1121 			      const struct ovs_conntrack_info *info,
1122 			      const struct nf_conntrack_tuple *tuple)
1123 {
1124 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1125 	const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1126 	u32 per_zone_limit, connections;
1127 	u32 conncount_key;
1128 
1129 	conncount_key = info->zone.id;
1130 
1131 	per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1132 	if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1133 		return 0;
1134 
1135 	connections = nf_conncount_count(net, ct_limit_info->data,
1136 					 &conncount_key, tuple, &info->zone);
1137 	if (connections > per_zone_limit)
1138 		return -ENOMEM;
1139 
1140 	return 0;
1141 }
1142 #endif
1143 
1144 /* Lookup connection and confirm if unconfirmed. */
1145 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1146 			 const struct ovs_conntrack_info *info,
1147 			 struct sk_buff *skb)
1148 {
1149 	enum ip_conntrack_info ctinfo;
1150 	struct nf_conn *ct;
1151 	int err;
1152 
1153 	err = __ovs_ct_lookup(net, key, info, skb);
1154 	if (err)
1155 		return err;
1156 
1157 	/* The connection could be invalid, in which case this is a no-op.*/
1158 	ct = nf_ct_get(skb, &ctinfo);
1159 	if (!ct)
1160 		return 0;
1161 
1162 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1163 	if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1164 		if (!nf_ct_is_confirmed(ct)) {
1165 			err = ovs_ct_check_limit(net, info,
1166 				&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1167 			if (err) {
1168 				net_warn_ratelimited("openvswitch: zone: %u "
1169 					"execeeds conntrack limit\n",
1170 					info->zone.id);
1171 				return err;
1172 			}
1173 		}
1174 	}
1175 #endif
1176 
1177 	/* Set the conntrack event mask if given.  NEW and DELETE events have
1178 	 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1179 	 * typically would receive many kinds of updates.  Setting the event
1180 	 * mask allows those events to be filtered.  The set event mask will
1181 	 * remain in effect for the lifetime of the connection unless changed
1182 	 * by a further CT action with both the commit flag and the eventmask
1183 	 * option. */
1184 	if (info->have_eventmask) {
1185 		struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1186 
1187 		if (cache)
1188 			cache->ctmask = info->eventmask;
1189 	}
1190 
1191 	/* Apply changes before confirming the connection so that the initial
1192 	 * conntrack NEW netlink event carries the values given in the CT
1193 	 * action.
1194 	 */
1195 	if (info->mark.mask) {
1196 		err = ovs_ct_set_mark(ct, key, info->mark.value,
1197 				      info->mark.mask);
1198 		if (err)
1199 			return err;
1200 	}
1201 	if (!nf_ct_is_confirmed(ct)) {
1202 		err = ovs_ct_init_labels(ct, key, &info->labels.value,
1203 					 &info->labels.mask);
1204 		if (err)
1205 			return err;
1206 	} else if (labels_nonzero(&info->labels.mask)) {
1207 		err = ovs_ct_set_labels(ct, key, &info->labels.value,
1208 					&info->labels.mask);
1209 		if (err)
1210 			return err;
1211 	}
1212 	/* This will take care of sending queued events even if the connection
1213 	 * is already confirmed.
1214 	 */
1215 	if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1216 		return -EINVAL;
1217 
1218 	return 0;
1219 }
1220 
1221 /* Trim the skb to the length specified by the IP/IPv6 header,
1222  * removing any trailing lower-layer padding. This prepares the skb
1223  * for higher-layer processing that assumes skb->len excludes padding
1224  * (such as nf_ip_checksum). The caller needs to pull the skb to the
1225  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1226  */
1227 static int ovs_skb_network_trim(struct sk_buff *skb)
1228 {
1229 	unsigned int len;
1230 	int err;
1231 
1232 	switch (skb->protocol) {
1233 	case htons(ETH_P_IP):
1234 		len = ntohs(ip_hdr(skb)->tot_len);
1235 		break;
1236 	case htons(ETH_P_IPV6):
1237 		len = sizeof(struct ipv6hdr)
1238 			+ ntohs(ipv6_hdr(skb)->payload_len);
1239 		break;
1240 	default:
1241 		len = skb->len;
1242 	}
1243 
1244 	err = pskb_trim_rcsum(skb, len);
1245 	if (err)
1246 		kfree_skb(skb);
1247 
1248 	return err;
1249 }
1250 
1251 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1252  * value if 'skb' is freed.
1253  */
1254 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1255 		   struct sw_flow_key *key,
1256 		   const struct ovs_conntrack_info *info)
1257 {
1258 	int nh_ofs;
1259 	int err;
1260 
1261 	/* The conntrack module expects to be working at L3. */
1262 	nh_ofs = skb_network_offset(skb);
1263 	skb_pull_rcsum(skb, nh_ofs);
1264 
1265 	err = ovs_skb_network_trim(skb);
1266 	if (err)
1267 		return err;
1268 
1269 	if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1270 		err = handle_fragments(net, key, info->zone.id, skb);
1271 		if (err)
1272 			return err;
1273 	}
1274 
1275 	if (info->commit)
1276 		err = ovs_ct_commit(net, key, info, skb);
1277 	else
1278 		err = ovs_ct_lookup(net, key, info, skb);
1279 
1280 	skb_push(skb, nh_ofs);
1281 	skb_postpush_rcsum(skb, skb->data, nh_ofs);
1282 	if (err)
1283 		kfree_skb(skb);
1284 	return err;
1285 }
1286 
1287 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1288 {
1289 	if (skb_nfct(skb)) {
1290 		nf_conntrack_put(skb_nfct(skb));
1291 		nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1292 		ovs_ct_fill_key(skb, key);
1293 	}
1294 
1295 	return 0;
1296 }
1297 
1298 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1299 			     const struct sw_flow_key *key, bool log)
1300 {
1301 	struct nf_conntrack_helper *helper;
1302 	struct nf_conn_help *help;
1303 
1304 	helper = nf_conntrack_helper_try_module_get(name, info->family,
1305 						    key->ip.proto);
1306 	if (!helper) {
1307 		OVS_NLERR(log, "Unknown helper \"%s\"", name);
1308 		return -EINVAL;
1309 	}
1310 
1311 	help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1312 	if (!help) {
1313 		nf_conntrack_helper_put(helper);
1314 		return -ENOMEM;
1315 	}
1316 
1317 	rcu_assign_pointer(help->helper, helper);
1318 	info->helper = helper;
1319 
1320 	if (info->nat)
1321 		request_module("ip_nat_%s", name);
1322 
1323 	return 0;
1324 }
1325 
1326 #ifdef CONFIG_NF_NAT_NEEDED
1327 static int parse_nat(const struct nlattr *attr,
1328 		     struct ovs_conntrack_info *info, bool log)
1329 {
1330 	struct nlattr *a;
1331 	int rem;
1332 	bool have_ip_max = false;
1333 	bool have_proto_max = false;
1334 	bool ip_vers = (info->family == NFPROTO_IPV6);
1335 
1336 	nla_for_each_nested(a, attr, rem) {
1337 		static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1338 			[OVS_NAT_ATTR_SRC] = {0, 0},
1339 			[OVS_NAT_ATTR_DST] = {0, 0},
1340 			[OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1341 						 sizeof(struct in6_addr)},
1342 			[OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1343 						 sizeof(struct in6_addr)},
1344 			[OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1345 			[OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1346 			[OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1347 			[OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1348 			[OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1349 		};
1350 		int type = nla_type(a);
1351 
1352 		if (type > OVS_NAT_ATTR_MAX) {
1353 			OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1354 				  type, OVS_NAT_ATTR_MAX);
1355 			return -EINVAL;
1356 		}
1357 
1358 		if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1359 			OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1360 				  type, nla_len(a),
1361 				  ovs_nat_attr_lens[type][ip_vers]);
1362 			return -EINVAL;
1363 		}
1364 
1365 		switch (type) {
1366 		case OVS_NAT_ATTR_SRC:
1367 		case OVS_NAT_ATTR_DST:
1368 			if (info->nat) {
1369 				OVS_NLERR(log, "Only one type of NAT may be specified");
1370 				return -ERANGE;
1371 			}
1372 			info->nat |= OVS_CT_NAT;
1373 			info->nat |= ((type == OVS_NAT_ATTR_SRC)
1374 					? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1375 			break;
1376 
1377 		case OVS_NAT_ATTR_IP_MIN:
1378 			nla_memcpy(&info->range.min_addr, a,
1379 				   sizeof(info->range.min_addr));
1380 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1381 			break;
1382 
1383 		case OVS_NAT_ATTR_IP_MAX:
1384 			have_ip_max = true;
1385 			nla_memcpy(&info->range.max_addr, a,
1386 				   sizeof(info->range.max_addr));
1387 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1388 			break;
1389 
1390 		case OVS_NAT_ATTR_PROTO_MIN:
1391 			info->range.min_proto.all = htons(nla_get_u16(a));
1392 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1393 			break;
1394 
1395 		case OVS_NAT_ATTR_PROTO_MAX:
1396 			have_proto_max = true;
1397 			info->range.max_proto.all = htons(nla_get_u16(a));
1398 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1399 			break;
1400 
1401 		case OVS_NAT_ATTR_PERSISTENT:
1402 			info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1403 			break;
1404 
1405 		case OVS_NAT_ATTR_PROTO_HASH:
1406 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1407 			break;
1408 
1409 		case OVS_NAT_ATTR_PROTO_RANDOM:
1410 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1411 			break;
1412 
1413 		default:
1414 			OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1415 			return -EINVAL;
1416 		}
1417 	}
1418 
1419 	if (rem > 0) {
1420 		OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1421 		return -EINVAL;
1422 	}
1423 	if (!info->nat) {
1424 		/* Do not allow flags if no type is given. */
1425 		if (info->range.flags) {
1426 			OVS_NLERR(log,
1427 				  "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1428 				  );
1429 			return -EINVAL;
1430 		}
1431 		info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1432 	} else if (!info->commit) {
1433 		OVS_NLERR(log,
1434 			  "NAT attributes may be specified only when CT COMMIT flag is also specified."
1435 			  );
1436 		return -EINVAL;
1437 	}
1438 	/* Allow missing IP_MAX. */
1439 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1440 		memcpy(&info->range.max_addr, &info->range.min_addr,
1441 		       sizeof(info->range.max_addr));
1442 	}
1443 	/* Allow missing PROTO_MAX. */
1444 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1445 	    !have_proto_max) {
1446 		info->range.max_proto.all = info->range.min_proto.all;
1447 	}
1448 	return 0;
1449 }
1450 #endif
1451 
1452 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1453 	[OVS_CT_ATTR_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1454 	[OVS_CT_ATTR_FORCE_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1455 	[OVS_CT_ATTR_ZONE]	= { .minlen = sizeof(u16),
1456 				    .maxlen = sizeof(u16) },
1457 	[OVS_CT_ATTR_MARK]	= { .minlen = sizeof(struct md_mark),
1458 				    .maxlen = sizeof(struct md_mark) },
1459 	[OVS_CT_ATTR_LABELS]	= { .minlen = sizeof(struct md_labels),
1460 				    .maxlen = sizeof(struct md_labels) },
1461 	[OVS_CT_ATTR_HELPER]	= { .minlen = 1,
1462 				    .maxlen = NF_CT_HELPER_NAME_LEN },
1463 #ifdef CONFIG_NF_NAT_NEEDED
1464 	/* NAT length is checked when parsing the nested attributes. */
1465 	[OVS_CT_ATTR_NAT]	= { .minlen = 0, .maxlen = INT_MAX },
1466 #endif
1467 	[OVS_CT_ATTR_EVENTMASK]	= { .minlen = sizeof(u32),
1468 				    .maxlen = sizeof(u32) },
1469 };
1470 
1471 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1472 		    const char **helper, bool log)
1473 {
1474 	struct nlattr *a;
1475 	int rem;
1476 
1477 	nla_for_each_nested(a, attr, rem) {
1478 		int type = nla_type(a);
1479 		int maxlen;
1480 		int minlen;
1481 
1482 		if (type > OVS_CT_ATTR_MAX) {
1483 			OVS_NLERR(log,
1484 				  "Unknown conntrack attr (type=%d, max=%d)",
1485 				  type, OVS_CT_ATTR_MAX);
1486 			return -EINVAL;
1487 		}
1488 
1489 		maxlen = ovs_ct_attr_lens[type].maxlen;
1490 		minlen = ovs_ct_attr_lens[type].minlen;
1491 		if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1492 			OVS_NLERR(log,
1493 				  "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1494 				  type, nla_len(a), maxlen);
1495 			return -EINVAL;
1496 		}
1497 
1498 		switch (type) {
1499 		case OVS_CT_ATTR_FORCE_COMMIT:
1500 			info->force = true;
1501 			/* fall through. */
1502 		case OVS_CT_ATTR_COMMIT:
1503 			info->commit = true;
1504 			break;
1505 #ifdef CONFIG_NF_CONNTRACK_ZONES
1506 		case OVS_CT_ATTR_ZONE:
1507 			info->zone.id = nla_get_u16(a);
1508 			break;
1509 #endif
1510 #ifdef CONFIG_NF_CONNTRACK_MARK
1511 		case OVS_CT_ATTR_MARK: {
1512 			struct md_mark *mark = nla_data(a);
1513 
1514 			if (!mark->mask) {
1515 				OVS_NLERR(log, "ct_mark mask cannot be 0");
1516 				return -EINVAL;
1517 			}
1518 			info->mark = *mark;
1519 			break;
1520 		}
1521 #endif
1522 #ifdef CONFIG_NF_CONNTRACK_LABELS
1523 		case OVS_CT_ATTR_LABELS: {
1524 			struct md_labels *labels = nla_data(a);
1525 
1526 			if (!labels_nonzero(&labels->mask)) {
1527 				OVS_NLERR(log, "ct_labels mask cannot be 0");
1528 				return -EINVAL;
1529 			}
1530 			info->labels = *labels;
1531 			break;
1532 		}
1533 #endif
1534 		case OVS_CT_ATTR_HELPER:
1535 			*helper = nla_data(a);
1536 			if (!memchr(*helper, '\0', nla_len(a))) {
1537 				OVS_NLERR(log, "Invalid conntrack helper");
1538 				return -EINVAL;
1539 			}
1540 			break;
1541 #ifdef CONFIG_NF_NAT_NEEDED
1542 		case OVS_CT_ATTR_NAT: {
1543 			int err = parse_nat(a, info, log);
1544 
1545 			if (err)
1546 				return err;
1547 			break;
1548 		}
1549 #endif
1550 		case OVS_CT_ATTR_EVENTMASK:
1551 			info->have_eventmask = true;
1552 			info->eventmask = nla_get_u32(a);
1553 			break;
1554 
1555 		default:
1556 			OVS_NLERR(log, "Unknown conntrack attr (%d)",
1557 				  type);
1558 			return -EINVAL;
1559 		}
1560 	}
1561 
1562 #ifdef CONFIG_NF_CONNTRACK_MARK
1563 	if (!info->commit && info->mark.mask) {
1564 		OVS_NLERR(log,
1565 			  "Setting conntrack mark requires 'commit' flag.");
1566 		return -EINVAL;
1567 	}
1568 #endif
1569 #ifdef CONFIG_NF_CONNTRACK_LABELS
1570 	if (!info->commit && labels_nonzero(&info->labels.mask)) {
1571 		OVS_NLERR(log,
1572 			  "Setting conntrack labels requires 'commit' flag.");
1573 		return -EINVAL;
1574 	}
1575 #endif
1576 	if (rem > 0) {
1577 		OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1578 		return -EINVAL;
1579 	}
1580 
1581 	return 0;
1582 }
1583 
1584 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1585 {
1586 	if (attr == OVS_KEY_ATTR_CT_STATE)
1587 		return true;
1588 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1589 	    attr == OVS_KEY_ATTR_CT_ZONE)
1590 		return true;
1591 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1592 	    attr == OVS_KEY_ATTR_CT_MARK)
1593 		return true;
1594 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1595 	    attr == OVS_KEY_ATTR_CT_LABELS) {
1596 		struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1597 
1598 		return ovs_net->xt_label;
1599 	}
1600 
1601 	return false;
1602 }
1603 
1604 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1605 		       const struct sw_flow_key *key,
1606 		       struct sw_flow_actions **sfa,  bool log)
1607 {
1608 	struct ovs_conntrack_info ct_info;
1609 	const char *helper = NULL;
1610 	u16 family;
1611 	int err;
1612 
1613 	family = key_to_nfproto(key);
1614 	if (family == NFPROTO_UNSPEC) {
1615 		OVS_NLERR(log, "ct family unspecified");
1616 		return -EINVAL;
1617 	}
1618 
1619 	memset(&ct_info, 0, sizeof(ct_info));
1620 	ct_info.family = family;
1621 
1622 	nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1623 			NF_CT_DEFAULT_ZONE_DIR, 0);
1624 
1625 	err = parse_ct(attr, &ct_info, &helper, log);
1626 	if (err)
1627 		return err;
1628 
1629 	/* Set up template for tracking connections in specific zones. */
1630 	ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1631 	if (!ct_info.ct) {
1632 		OVS_NLERR(log, "Failed to allocate conntrack template");
1633 		return -ENOMEM;
1634 	}
1635 	if (helper) {
1636 		err = ovs_ct_add_helper(&ct_info, helper, key, log);
1637 		if (err)
1638 			goto err_free_ct;
1639 	}
1640 
1641 	err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1642 				 sizeof(ct_info), log);
1643 	if (err)
1644 		goto err_free_ct;
1645 
1646 	__set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1647 	nf_conntrack_get(&ct_info.ct->ct_general);
1648 	return 0;
1649 err_free_ct:
1650 	__ovs_ct_free_action(&ct_info);
1651 	return err;
1652 }
1653 
1654 #ifdef CONFIG_NF_NAT_NEEDED
1655 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1656 			       struct sk_buff *skb)
1657 {
1658 	struct nlattr *start;
1659 
1660 	start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1661 	if (!start)
1662 		return false;
1663 
1664 	if (info->nat & OVS_CT_SRC_NAT) {
1665 		if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1666 			return false;
1667 	} else if (info->nat & OVS_CT_DST_NAT) {
1668 		if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1669 			return false;
1670 	} else {
1671 		goto out;
1672 	}
1673 
1674 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1675 		if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1676 		    info->family == NFPROTO_IPV4) {
1677 			if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1678 					    info->range.min_addr.ip) ||
1679 			    (info->range.max_addr.ip
1680 			     != info->range.min_addr.ip &&
1681 			     (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1682 					      info->range.max_addr.ip))))
1683 				return false;
1684 		} else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1685 			   info->family == NFPROTO_IPV6) {
1686 			if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1687 					     &info->range.min_addr.in6) ||
1688 			    (memcmp(&info->range.max_addr.in6,
1689 				    &info->range.min_addr.in6,
1690 				    sizeof(info->range.max_addr.in6)) &&
1691 			     (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1692 					       &info->range.max_addr.in6))))
1693 				return false;
1694 		} else {
1695 			return false;
1696 		}
1697 	}
1698 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1699 	    (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1700 			 ntohs(info->range.min_proto.all)) ||
1701 	     (info->range.max_proto.all != info->range.min_proto.all &&
1702 	      nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1703 			  ntohs(info->range.max_proto.all)))))
1704 		return false;
1705 
1706 	if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1707 	    nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1708 		return false;
1709 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1710 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1711 		return false;
1712 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1713 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1714 		return false;
1715 out:
1716 	nla_nest_end(skb, start);
1717 
1718 	return true;
1719 }
1720 #endif
1721 
1722 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1723 			  struct sk_buff *skb)
1724 {
1725 	struct nlattr *start;
1726 
1727 	start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1728 	if (!start)
1729 		return -EMSGSIZE;
1730 
1731 	if (ct_info->commit && nla_put_flag(skb, ct_info->force
1732 					    ? OVS_CT_ATTR_FORCE_COMMIT
1733 					    : OVS_CT_ATTR_COMMIT))
1734 		return -EMSGSIZE;
1735 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1736 	    nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1737 		return -EMSGSIZE;
1738 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1739 	    nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1740 		    &ct_info->mark))
1741 		return -EMSGSIZE;
1742 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1743 	    labels_nonzero(&ct_info->labels.mask) &&
1744 	    nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1745 		    &ct_info->labels))
1746 		return -EMSGSIZE;
1747 	if (ct_info->helper) {
1748 		if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1749 				   ct_info->helper->name))
1750 			return -EMSGSIZE;
1751 	}
1752 	if (ct_info->have_eventmask &&
1753 	    nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1754 		return -EMSGSIZE;
1755 
1756 #ifdef CONFIG_NF_NAT_NEEDED
1757 	if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1758 		return -EMSGSIZE;
1759 #endif
1760 	nla_nest_end(skb, start);
1761 
1762 	return 0;
1763 }
1764 
1765 void ovs_ct_free_action(const struct nlattr *a)
1766 {
1767 	struct ovs_conntrack_info *ct_info = nla_data(a);
1768 
1769 	__ovs_ct_free_action(ct_info);
1770 }
1771 
1772 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1773 {
1774 	if (ct_info->helper)
1775 		nf_conntrack_helper_put(ct_info->helper);
1776 	if (ct_info->ct)
1777 		nf_ct_tmpl_free(ct_info->ct);
1778 }
1779 
1780 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1781 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1782 {
1783 	int i, err;
1784 
1785 	ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1786 					 GFP_KERNEL);
1787 	if (!ovs_net->ct_limit_info)
1788 		return -ENOMEM;
1789 
1790 	ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1791 	ovs_net->ct_limit_info->limits =
1792 		kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1793 			      GFP_KERNEL);
1794 	if (!ovs_net->ct_limit_info->limits) {
1795 		kfree(ovs_net->ct_limit_info);
1796 		return -ENOMEM;
1797 	}
1798 
1799 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1800 		INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1801 
1802 	ovs_net->ct_limit_info->data =
1803 		nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1804 
1805 	if (IS_ERR(ovs_net->ct_limit_info->data)) {
1806 		err = PTR_ERR(ovs_net->ct_limit_info->data);
1807 		kfree(ovs_net->ct_limit_info->limits);
1808 		kfree(ovs_net->ct_limit_info);
1809 		pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1810 		return err;
1811 	}
1812 	return 0;
1813 }
1814 
1815 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1816 {
1817 	const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1818 	int i;
1819 
1820 	nf_conncount_destroy(net, NFPROTO_INET, info->data);
1821 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1822 		struct hlist_head *head = &info->limits[i];
1823 		struct ovs_ct_limit *ct_limit;
1824 
1825 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node)
1826 			kfree_rcu(ct_limit, rcu);
1827 	}
1828 	kfree(ovs_net->ct_limit_info->limits);
1829 	kfree(ovs_net->ct_limit_info);
1830 }
1831 
1832 static struct sk_buff *
1833 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1834 			     struct ovs_header **ovs_reply_header)
1835 {
1836 	struct ovs_header *ovs_header = info->userhdr;
1837 	struct sk_buff *skb;
1838 
1839 	skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1840 	if (!skb)
1841 		return ERR_PTR(-ENOMEM);
1842 
1843 	*ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1844 					info->snd_seq,
1845 					&dp_ct_limit_genl_family, 0, cmd);
1846 
1847 	if (!*ovs_reply_header) {
1848 		nlmsg_free(skb);
1849 		return ERR_PTR(-EMSGSIZE);
1850 	}
1851 	(*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1852 
1853 	return skb;
1854 }
1855 
1856 static bool check_zone_id(int zone_id, u16 *pzone)
1857 {
1858 	if (zone_id >= 0 && zone_id <= 65535) {
1859 		*pzone = (u16)zone_id;
1860 		return true;
1861 	}
1862 	return false;
1863 }
1864 
1865 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1866 				       struct ovs_ct_limit_info *info)
1867 {
1868 	struct ovs_zone_limit *zone_limit;
1869 	int rem;
1870 	u16 zone;
1871 
1872 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1873 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1874 
1875 	while (rem >= sizeof(*zone_limit)) {
1876 		if (unlikely(zone_limit->zone_id ==
1877 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1878 			ovs_lock();
1879 			info->default_limit = zone_limit->limit;
1880 			ovs_unlock();
1881 		} else if (unlikely(!check_zone_id(
1882 				zone_limit->zone_id, &zone))) {
1883 			OVS_NLERR(true, "zone id is out of range");
1884 		} else {
1885 			struct ovs_ct_limit *ct_limit;
1886 
1887 			ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1888 			if (!ct_limit)
1889 				return -ENOMEM;
1890 
1891 			ct_limit->zone = zone;
1892 			ct_limit->limit = zone_limit->limit;
1893 
1894 			ovs_lock();
1895 			ct_limit_set(info, ct_limit);
1896 			ovs_unlock();
1897 		}
1898 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1899 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1900 				NLA_ALIGN(sizeof(*zone_limit)));
1901 	}
1902 
1903 	if (rem)
1904 		OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1905 
1906 	return 0;
1907 }
1908 
1909 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1910 				       struct ovs_ct_limit_info *info)
1911 {
1912 	struct ovs_zone_limit *zone_limit;
1913 	int rem;
1914 	u16 zone;
1915 
1916 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1917 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1918 
1919 	while (rem >= sizeof(*zone_limit)) {
1920 		if (unlikely(zone_limit->zone_id ==
1921 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1922 			ovs_lock();
1923 			info->default_limit = OVS_CT_LIMIT_DEFAULT;
1924 			ovs_unlock();
1925 		} else if (unlikely(!check_zone_id(
1926 				zone_limit->zone_id, &zone))) {
1927 			OVS_NLERR(true, "zone id is out of range");
1928 		} else {
1929 			ovs_lock();
1930 			ct_limit_del(info, zone);
1931 			ovs_unlock();
1932 		}
1933 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1934 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1935 				NLA_ALIGN(sizeof(*zone_limit)));
1936 	}
1937 
1938 	if (rem)
1939 		OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
1940 
1941 	return 0;
1942 }
1943 
1944 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
1945 					  struct sk_buff *reply)
1946 {
1947 	struct ovs_zone_limit zone_limit;
1948 	int err;
1949 
1950 	zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
1951 	zone_limit.limit = info->default_limit;
1952 	err = nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1953 	if (err)
1954 		return err;
1955 
1956 	return 0;
1957 }
1958 
1959 static int __ovs_ct_limit_get_zone_limit(struct net *net,
1960 					 struct nf_conncount_data *data,
1961 					 u16 zone_id, u32 limit,
1962 					 struct sk_buff *reply)
1963 {
1964 	struct nf_conntrack_zone ct_zone;
1965 	struct ovs_zone_limit zone_limit;
1966 	u32 conncount_key = zone_id;
1967 
1968 	zone_limit.zone_id = zone_id;
1969 	zone_limit.limit = limit;
1970 	nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
1971 
1972 	zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
1973 					      &ct_zone);
1974 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1975 }
1976 
1977 static int ovs_ct_limit_get_zone_limit(struct net *net,
1978 				       struct nlattr *nla_zone_limit,
1979 				       struct ovs_ct_limit_info *info,
1980 				       struct sk_buff *reply)
1981 {
1982 	struct ovs_zone_limit *zone_limit;
1983 	int rem, err;
1984 	u32 limit;
1985 	u16 zone;
1986 
1987 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1988 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1989 
1990 	while (rem >= sizeof(*zone_limit)) {
1991 		if (unlikely(zone_limit->zone_id ==
1992 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1993 			err = ovs_ct_limit_get_default_limit(info, reply);
1994 			if (err)
1995 				return err;
1996 		} else if (unlikely(!check_zone_id(zone_limit->zone_id,
1997 							&zone))) {
1998 			OVS_NLERR(true, "zone id is out of range");
1999 		} else {
2000 			rcu_read_lock();
2001 			limit = ct_limit_get(info, zone);
2002 			rcu_read_unlock();
2003 
2004 			err = __ovs_ct_limit_get_zone_limit(
2005 				net, info->data, zone, limit, reply);
2006 			if (err)
2007 				return err;
2008 		}
2009 		rem -= NLA_ALIGN(sizeof(*zone_limit));
2010 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2011 				NLA_ALIGN(sizeof(*zone_limit)));
2012 	}
2013 
2014 	if (rem)
2015 		OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2016 
2017 	return 0;
2018 }
2019 
2020 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2021 					   struct ovs_ct_limit_info *info,
2022 					   struct sk_buff *reply)
2023 {
2024 	struct ovs_ct_limit *ct_limit;
2025 	struct hlist_head *head;
2026 	int i, err = 0;
2027 
2028 	err = ovs_ct_limit_get_default_limit(info, reply);
2029 	if (err)
2030 		return err;
2031 
2032 	rcu_read_lock();
2033 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2034 		head = &info->limits[i];
2035 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2036 			err = __ovs_ct_limit_get_zone_limit(net, info->data,
2037 				ct_limit->zone, ct_limit->limit, reply);
2038 			if (err)
2039 				goto exit_err;
2040 		}
2041 	}
2042 
2043 exit_err:
2044 	rcu_read_unlock();
2045 	return err;
2046 }
2047 
2048 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2049 {
2050 	struct nlattr **a = info->attrs;
2051 	struct sk_buff *reply;
2052 	struct ovs_header *ovs_reply_header;
2053 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2054 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2055 	int err;
2056 
2057 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2058 					     &ovs_reply_header);
2059 	if (IS_ERR(reply))
2060 		return PTR_ERR(reply);
2061 
2062 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2063 		err = -EINVAL;
2064 		goto exit_err;
2065 	}
2066 
2067 	err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2068 					  ct_limit_info);
2069 	if (err)
2070 		goto exit_err;
2071 
2072 	static_branch_enable(&ovs_ct_limit_enabled);
2073 
2074 	genlmsg_end(reply, ovs_reply_header);
2075 	return genlmsg_reply(reply, info);
2076 
2077 exit_err:
2078 	nlmsg_free(reply);
2079 	return err;
2080 }
2081 
2082 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2083 {
2084 	struct nlattr **a = info->attrs;
2085 	struct sk_buff *reply;
2086 	struct ovs_header *ovs_reply_header;
2087 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2088 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2089 	int err;
2090 
2091 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2092 					     &ovs_reply_header);
2093 	if (IS_ERR(reply))
2094 		return PTR_ERR(reply);
2095 
2096 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2097 		err = -EINVAL;
2098 		goto exit_err;
2099 	}
2100 
2101 	err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2102 					  ct_limit_info);
2103 	if (err)
2104 		goto exit_err;
2105 
2106 	genlmsg_end(reply, ovs_reply_header);
2107 	return genlmsg_reply(reply, info);
2108 
2109 exit_err:
2110 	nlmsg_free(reply);
2111 	return err;
2112 }
2113 
2114 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2115 {
2116 	struct nlattr **a = info->attrs;
2117 	struct nlattr *nla_reply;
2118 	struct sk_buff *reply;
2119 	struct ovs_header *ovs_reply_header;
2120 	struct net *net = sock_net(skb->sk);
2121 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2122 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2123 	int err;
2124 
2125 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2126 					     &ovs_reply_header);
2127 	if (IS_ERR(reply))
2128 		return PTR_ERR(reply);
2129 
2130 	nla_reply = nla_nest_start(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2131 
2132 	if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2133 		err = ovs_ct_limit_get_zone_limit(
2134 			net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2135 			reply);
2136 		if (err)
2137 			goto exit_err;
2138 	} else {
2139 		err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2140 						      reply);
2141 		if (err)
2142 			goto exit_err;
2143 	}
2144 
2145 	nla_nest_end(reply, nla_reply);
2146 	genlmsg_end(reply, ovs_reply_header);
2147 	return genlmsg_reply(reply, info);
2148 
2149 exit_err:
2150 	nlmsg_free(reply);
2151 	return err;
2152 }
2153 
2154 static struct genl_ops ct_limit_genl_ops[] = {
2155 	{ .cmd = OVS_CT_LIMIT_CMD_SET,
2156 		.flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2157 					   * privilege. */
2158 		.policy = ct_limit_policy,
2159 		.doit = ovs_ct_limit_cmd_set,
2160 	},
2161 	{ .cmd = OVS_CT_LIMIT_CMD_DEL,
2162 		.flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2163 					   * privilege. */
2164 		.policy = ct_limit_policy,
2165 		.doit = ovs_ct_limit_cmd_del,
2166 	},
2167 	{ .cmd = OVS_CT_LIMIT_CMD_GET,
2168 		.flags = 0,		  /* OK for unprivileged users. */
2169 		.policy = ct_limit_policy,
2170 		.doit = ovs_ct_limit_cmd_get,
2171 	},
2172 };
2173 
2174 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2175 	.name = OVS_CT_LIMIT_MCGROUP,
2176 };
2177 
2178 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2179 	.hdrsize = sizeof(struct ovs_header),
2180 	.name = OVS_CT_LIMIT_FAMILY,
2181 	.version = OVS_CT_LIMIT_VERSION,
2182 	.maxattr = OVS_CT_LIMIT_ATTR_MAX,
2183 	.netnsok = true,
2184 	.parallel_ops = true,
2185 	.ops = ct_limit_genl_ops,
2186 	.n_ops = ARRAY_SIZE(ct_limit_genl_ops),
2187 	.mcgrps = &ovs_ct_limit_multicast_group,
2188 	.n_mcgrps = 1,
2189 	.module = THIS_MODULE,
2190 };
2191 #endif
2192 
2193 int ovs_ct_init(struct net *net)
2194 {
2195 	unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2196 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2197 
2198 	if (nf_connlabels_get(net, n_bits - 1)) {
2199 		ovs_net->xt_label = false;
2200 		OVS_NLERR(true, "Failed to set connlabel length");
2201 	} else {
2202 		ovs_net->xt_label = true;
2203 	}
2204 
2205 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2206 	return ovs_ct_limit_init(net, ovs_net);
2207 #else
2208 	return 0;
2209 #endif
2210 }
2211 
2212 void ovs_ct_exit(struct net *net)
2213 {
2214 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2215 
2216 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2217 	ovs_ct_limit_exit(net, ovs_net);
2218 #endif
2219 
2220 	if (ovs_net->xt_label)
2221 		nf_connlabels_put(net);
2222 }
2223