1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2015 Nicira, Inc.
4 */
5
6#include <linux/module.h>
7#include <linux/openvswitch.h>
8#include <linux/tcp.h>
9#include <linux/udp.h>
10#include <linux/sctp.h>
11#include <linux/static_key.h>
12#include <net/ip.h>
13#include <net/genetlink.h>
14#include <net/netfilter/nf_conntrack_core.h>
15#include <net/netfilter/nf_conntrack_count.h>
16#include <net/netfilter/nf_conntrack_helper.h>
17#include <net/netfilter/nf_conntrack_labels.h>
18#include <net/netfilter/nf_conntrack_seqadj.h>
19#include <net/netfilter/nf_conntrack_timeout.h>
20#include <net/netfilter/nf_conntrack_zones.h>
21#include <net/netfilter/ipv6/nf_defrag_ipv6.h>
22#include <net/ipv6_frag.h>
23
24#if IS_ENABLED(CONFIG_NF_NAT)
25#include <net/netfilter/nf_nat.h>
26#endif
27
28#include "datapath.h"
29#include "conntrack.h"
30#include "flow.h"
31#include "flow_netlink.h"
32
33struct ovs_ct_len_tbl {
34	int maxlen;
35	int minlen;
36};
37
38/* Metadata mark for masked write to conntrack mark */
39struct md_mark {
40	u32 value;
41	u32 mask;
42};
43
44/* Metadata label for masked write to conntrack label. */
45struct md_labels {
46	struct ovs_key_ct_labels value;
47	struct ovs_key_ct_labels mask;
48};
49
50enum ovs_ct_nat {
51	OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
52	OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
53	OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
54};
55
56/* Conntrack action context for execution. */
57struct ovs_conntrack_info {
58	struct nf_conntrack_helper *helper;
59	struct nf_conntrack_zone zone;
60	struct nf_conn *ct;
61	u8 commit : 1;
62	u8 nat : 3;                 /* enum ovs_ct_nat */
63	u8 force : 1;
64	u8 have_eventmask : 1;
65	u16 family;
66	u32 eventmask;              /* Mask of 1 << IPCT_*. */
67	struct md_mark mark;
68	struct md_labels labels;
69	char timeout[CTNL_TIMEOUT_NAME_MAX];
70	struct nf_ct_timeout *nf_ct_timeout;
71#if IS_ENABLED(CONFIG_NF_NAT)
72	struct nf_nat_range2 range;  /* Only present for SRC NAT and DST NAT. */
73#endif
74};
75
76#if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
77#define OVS_CT_LIMIT_UNLIMITED	0
78#define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
79#define CT_LIMIT_HASH_BUCKETS 512
80static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
81
82struct ovs_ct_limit {
83	/* Elements in ovs_ct_limit_info->limits hash table */
84	struct hlist_node hlist_node;
85	struct rcu_head rcu;
86	u16 zone;
87	u32 limit;
88};
89
90struct ovs_ct_limit_info {
91	u32 default_limit;
92	struct hlist_head *limits;
93	struct nf_conncount_data *data;
94};
95
96static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
97	[OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
98};
99#endif
100
101static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
102
103static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
104
105static u16 key_to_nfproto(const struct sw_flow_key *key)
106{
107	switch (ntohs(key->eth.type)) {
108	case ETH_P_IP:
109		return NFPROTO_IPV4;
110	case ETH_P_IPV6:
111		return NFPROTO_IPV6;
112	default:
113		return NFPROTO_UNSPEC;
114	}
115}
116
117/* Map SKB connection state into the values used by flow definition. */
118static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
119{
120	u8 ct_state = OVS_CS_F_TRACKED;
121
122	switch (ctinfo) {
123	case IP_CT_ESTABLISHED_REPLY:
124	case IP_CT_RELATED_REPLY:
125		ct_state |= OVS_CS_F_REPLY_DIR;
126		break;
127	default:
128		break;
129	}
130
131	switch (ctinfo) {
132	case IP_CT_ESTABLISHED:
133	case IP_CT_ESTABLISHED_REPLY:
134		ct_state |= OVS_CS_F_ESTABLISHED;
135		break;
136	case IP_CT_RELATED:
137	case IP_CT_RELATED_REPLY:
138		ct_state |= OVS_CS_F_RELATED;
139		break;
140	case IP_CT_NEW:
141		ct_state |= OVS_CS_F_NEW;
142		break;
143	default:
144		break;
145	}
146
147	return ct_state;
148}
149
150static u32 ovs_ct_get_mark(const struct nf_conn *ct)
151{
152#if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
153	return ct ? READ_ONCE(ct->mark) : 0;
154#else
155	return 0;
156#endif
157}
158
159/* Guard against conntrack labels max size shrinking below 128 bits. */
160#if NF_CT_LABELS_MAX_SIZE < 16
161#error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
162#endif
163
164static void ovs_ct_get_labels(const struct nf_conn *ct,
165			      struct ovs_key_ct_labels *labels)
166{
167	struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
168
169	if (cl)
170		memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
171	else
172		memset(labels, 0, OVS_CT_LABELS_LEN);
173}
174
175static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
176					const struct nf_conntrack_tuple *orig,
177					u8 icmp_proto)
178{
179	key->ct_orig_proto = orig->dst.protonum;
180	if (orig->dst.protonum == icmp_proto) {
181		key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
182		key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
183	} else {
184		key->ct.orig_tp.src = orig->src.u.all;
185		key->ct.orig_tp.dst = orig->dst.u.all;
186	}
187}
188
189static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
190				const struct nf_conntrack_zone *zone,
191				const struct nf_conn *ct)
192{
193	key->ct_state = state;
194	key->ct_zone = zone->id;
195	key->ct.mark = ovs_ct_get_mark(ct);
196	ovs_ct_get_labels(ct, &key->ct.labels);
197
198	if (ct) {
199		const struct nf_conntrack_tuple *orig;
200
201		/* Use the master if we have one. */
202		if (ct->master)
203			ct = ct->master;
204		orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
205
206		/* IP version must match with the master connection. */
207		if (key->eth.type == htons(ETH_P_IP) &&
208		    nf_ct_l3num(ct) == NFPROTO_IPV4) {
209			key->ipv4.ct_orig.src = orig->src.u3.ip;
210			key->ipv4.ct_orig.dst = orig->dst.u3.ip;
211			__ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
212			return;
213		} else if (key->eth.type == htons(ETH_P_IPV6) &&
214			   !sw_flow_key_is_nd(key) &&
215			   nf_ct_l3num(ct) == NFPROTO_IPV6) {
216			key->ipv6.ct_orig.src = orig->src.u3.in6;
217			key->ipv6.ct_orig.dst = orig->dst.u3.in6;
218			__ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
219			return;
220		}
221	}
222	/* Clear 'ct_orig_proto' to mark the non-existence of conntrack
223	 * original direction key fields.
224	 */
225	key->ct_orig_proto = 0;
226}
227
228/* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
229 * previously sent the packet to conntrack via the ct action.  If
230 * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
231 * initialized from the connection status.
232 */
233static void ovs_ct_update_key(const struct sk_buff *skb,
234			      const struct ovs_conntrack_info *info,
235			      struct sw_flow_key *key, bool post_ct,
236			      bool keep_nat_flags)
237{
238	const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
239	enum ip_conntrack_info ctinfo;
240	struct nf_conn *ct;
241	u8 state = 0;
242
243	ct = nf_ct_get(skb, &ctinfo);
244	if (ct) {
245		state = ovs_ct_get_state(ctinfo);
246		/* All unconfirmed entries are NEW connections. */
247		if (!nf_ct_is_confirmed(ct))
248			state |= OVS_CS_F_NEW;
249		/* OVS persists the related flag for the duration of the
250		 * connection.
251		 */
252		if (ct->master)
253			state |= OVS_CS_F_RELATED;
254		if (keep_nat_flags) {
255			state |= key->ct_state & OVS_CS_F_NAT_MASK;
256		} else {
257			if (ct->status & IPS_SRC_NAT)
258				state |= OVS_CS_F_SRC_NAT;
259			if (ct->status & IPS_DST_NAT)
260				state |= OVS_CS_F_DST_NAT;
261		}
262		zone = nf_ct_zone(ct);
263	} else if (post_ct) {
264		state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
265		if (info)
266			zone = &info->zone;
267	}
268	__ovs_ct_update_key(key, state, zone, ct);
269}
270
271/* This is called to initialize CT key fields possibly coming in from the local
272 * stack.
273 */
274void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
275{
276	ovs_ct_update_key(skb, NULL, key, false, false);
277}
278
279int ovs_ct_put_key(const struct sw_flow_key *swkey,
280		   const struct sw_flow_key *output, struct sk_buff *skb)
281{
282	if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
283		return -EMSGSIZE;
284
285	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
286	    nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
287		return -EMSGSIZE;
288
289	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
290	    nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
291		return -EMSGSIZE;
292
293	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
294	    nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
295		    &output->ct.labels))
296		return -EMSGSIZE;
297
298	if (swkey->ct_orig_proto) {
299		if (swkey->eth.type == htons(ETH_P_IP)) {
300			struct ovs_key_ct_tuple_ipv4 orig;
301
302			memset(&orig, 0, sizeof(orig));
303			orig.ipv4_src = output->ipv4.ct_orig.src;
304			orig.ipv4_dst = output->ipv4.ct_orig.dst;
305			orig.src_port = output->ct.orig_tp.src;
306			orig.dst_port = output->ct.orig_tp.dst;
307			orig.ipv4_proto = output->ct_orig_proto;
308
309			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
310				    sizeof(orig), &orig))
311				return -EMSGSIZE;
312		} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
313			struct ovs_key_ct_tuple_ipv6 orig;
314
315			memset(&orig, 0, sizeof(orig));
316			memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32,
317			       sizeof(orig.ipv6_src));
318			memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32,
319			       sizeof(orig.ipv6_dst));
320			orig.src_port = output->ct.orig_tp.src;
321			orig.dst_port = output->ct.orig_tp.dst;
322			orig.ipv6_proto = output->ct_orig_proto;
323
324			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
325				    sizeof(orig), &orig))
326				return -EMSGSIZE;
327		}
328	}
329
330	return 0;
331}
332
333static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
334			   u32 ct_mark, u32 mask)
335{
336#if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
337	u32 new_mark;
338
339	new_mark = ct_mark | (READ_ONCE(ct->mark) & ~(mask));
340	if (READ_ONCE(ct->mark) != new_mark) {
341		WRITE_ONCE(ct->mark, new_mark);
342		if (nf_ct_is_confirmed(ct))
343			nf_conntrack_event_cache(IPCT_MARK, ct);
344		key->ct.mark = new_mark;
345	}
346
347	return 0;
348#else
349	return -ENOTSUPP;
350#endif
351}
352
353static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
354{
355	struct nf_conn_labels *cl;
356
357	cl = nf_ct_labels_find(ct);
358	if (!cl) {
359		nf_ct_labels_ext_add(ct);
360		cl = nf_ct_labels_find(ct);
361	}
362
363	return cl;
364}
365
366/* Initialize labels for a new, yet to be committed conntrack entry.  Note that
367 * since the new connection is not yet confirmed, and thus no-one else has
368 * access to it's labels, we simply write them over.
369 */
370static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
371			      const struct ovs_key_ct_labels *labels,
372			      const struct ovs_key_ct_labels *mask)
373{
374	struct nf_conn_labels *cl, *master_cl;
375	bool have_mask = labels_nonzero(mask);
376
377	/* Inherit master's labels to the related connection? */
378	master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
379
380	if (!master_cl && !have_mask)
381		return 0;   /* Nothing to do. */
382
383	cl = ovs_ct_get_conn_labels(ct);
384	if (!cl)
385		return -ENOSPC;
386
387	/* Inherit the master's labels, if any. */
388	if (master_cl)
389		*cl = *master_cl;
390
391	if (have_mask) {
392		u32 *dst = (u32 *)cl->bits;
393		int i;
394
395		for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
396			dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
397				(labels->ct_labels_32[i]
398				 & mask->ct_labels_32[i]);
399	}
400
401	/* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
402	 * IPCT_LABEL bit is set in the event cache.
403	 */
404	nf_conntrack_event_cache(IPCT_LABEL, ct);
405
406	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
407
408	return 0;
409}
410
411static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
412			     const struct ovs_key_ct_labels *labels,
413			     const struct ovs_key_ct_labels *mask)
414{
415	struct nf_conn_labels *cl;
416	int err;
417
418	cl = ovs_ct_get_conn_labels(ct);
419	if (!cl)
420		return -ENOSPC;
421
422	err = nf_connlabels_replace(ct, labels->ct_labels_32,
423				    mask->ct_labels_32,
424				    OVS_CT_LABELS_LEN_32);
425	if (err)
426		return err;
427
428	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
429
430	return 0;
431}
432
433/* 'skb' should already be pulled to nh_ofs. */
434static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
435{
436	const struct nf_conntrack_helper *helper;
437	const struct nf_conn_help *help;
438	enum ip_conntrack_info ctinfo;
439	unsigned int protoff;
440	struct nf_conn *ct;
441	int err;
442
443	ct = nf_ct_get(skb, &ctinfo);
444	if (!ct || ctinfo == IP_CT_RELATED_REPLY)
445		return NF_ACCEPT;
446
447	help = nfct_help(ct);
448	if (!help)
449		return NF_ACCEPT;
450
451	helper = rcu_dereference(help->helper);
452	if (!helper)
453		return NF_ACCEPT;
454
455	switch (proto) {
456	case NFPROTO_IPV4:
457		protoff = ip_hdrlen(skb);
458		break;
459	case NFPROTO_IPV6: {
460		u8 nexthdr = ipv6_hdr(skb)->nexthdr;
461		__be16 frag_off;
462		int ofs;
463
464		ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
465				       &frag_off);
466		if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
467			pr_debug("proto header not found\n");
468			return NF_ACCEPT;
469		}
470		protoff = ofs;
471		break;
472	}
473	default:
474		WARN_ONCE(1, "helper invoked on non-IP family!");
475		return NF_DROP;
476	}
477
478	err = helper->help(skb, protoff, ct, ctinfo);
479	if (err != NF_ACCEPT)
480		return err;
481
482	/* Adjust seqs after helper.  This is needed due to some helpers (e.g.,
483	 * FTP with NAT) adusting the TCP payload size when mangling IP
484	 * addresses and/or port numbers in the text-based control connection.
485	 */
486	if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
487	    !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
488		return NF_DROP;
489	return NF_ACCEPT;
490}
491
492/* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
493 * value if 'skb' is freed.
494 */
495static int handle_fragments(struct net *net, struct sw_flow_key *key,
496			    u16 zone, struct sk_buff *skb)
497{
498	struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
499	int err;
500
501	if (key->eth.type == htons(ETH_P_IP)) {
502		enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
503
504		memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
505		err = ip_defrag(net, skb, user);
506		if (err)
507			return err;
508
509		ovs_cb.mru = IPCB(skb)->frag_max_size;
510#if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
511	} else if (key->eth.type == htons(ETH_P_IPV6)) {
512		enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
513
514		memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
515		err = nf_ct_frag6_gather(net, skb, user);
516		if (err) {
517			if (err != -EINPROGRESS)
518				kfree_skb(skb);
519			return err;
520		}
521
522		key->ip.proto = ipv6_hdr(skb)->nexthdr;
523		ovs_cb.mru = IP6CB(skb)->frag_max_size;
524#endif
525	} else {
526		kfree_skb(skb);
527		return -EPFNOSUPPORT;
528	}
529
530	/* The key extracted from the fragment that completed this datagram
531	 * likely didn't have an L4 header, so regenerate it.
532	 */
533	ovs_flow_key_update_l3l4(skb, key);
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
543static struct nf_conntrack_expect *
544ovs_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. */
583static enum ip_conntrack_info
584ovs_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 */
607static struct nf_conn *
608ovs_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_tuple(&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
650static
651struct 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. */
679static 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	if (info->nf_ct_timeout) {
709		struct nf_conn_timeout *timeout_ext;
710
711		timeout_ext = nf_ct_timeout_find(ct);
712		if (!timeout_ext || info->nf_ct_timeout !=
713		    rcu_dereference(timeout_ext->timeout))
714			return false;
715	}
716	/* Force conntrack entry direction to the current packet? */
717	if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
718		/* Delete the conntrack entry if confirmed, else just release
719		 * the reference.
720		 */
721		if (nf_ct_is_confirmed(ct))
722			nf_ct_delete(ct, 0, 0);
723
724		nf_conntrack_put(&ct->ct_general);
725		nf_ct_set(skb, NULL, 0);
726		return false;
727	}
728
729	return ct_executed;
730}
731
732#if IS_ENABLED(CONFIG_NF_NAT)
733static void ovs_nat_update_key(struct sw_flow_key *key,
734			       const struct sk_buff *skb,
735			       enum nf_nat_manip_type maniptype)
736{
737	if (maniptype == NF_NAT_MANIP_SRC) {
738		__be16 src;
739
740		key->ct_state |= OVS_CS_F_SRC_NAT;
741		if (key->eth.type == htons(ETH_P_IP))
742			key->ipv4.addr.src = ip_hdr(skb)->saddr;
743		else if (key->eth.type == htons(ETH_P_IPV6))
744			memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
745			       sizeof(key->ipv6.addr.src));
746		else
747			return;
748
749		if (key->ip.proto == IPPROTO_UDP)
750			src = udp_hdr(skb)->source;
751		else if (key->ip.proto == IPPROTO_TCP)
752			src = tcp_hdr(skb)->source;
753		else if (key->ip.proto == IPPROTO_SCTP)
754			src = sctp_hdr(skb)->source;
755		else
756			return;
757
758		key->tp.src = src;
759	} else {
760		__be16 dst;
761
762		key->ct_state |= OVS_CS_F_DST_NAT;
763		if (key->eth.type == htons(ETH_P_IP))
764			key->ipv4.addr.dst = ip_hdr(skb)->daddr;
765		else if (key->eth.type == htons(ETH_P_IPV6))
766			memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
767			       sizeof(key->ipv6.addr.dst));
768		else
769			return;
770
771		if (key->ip.proto == IPPROTO_UDP)
772			dst = udp_hdr(skb)->dest;
773		else if (key->ip.proto == IPPROTO_TCP)
774			dst = tcp_hdr(skb)->dest;
775		else if (key->ip.proto == IPPROTO_SCTP)
776			dst = sctp_hdr(skb)->dest;
777		else
778			return;
779
780		key->tp.dst = dst;
781	}
782}
783
784/* Modelled after nf_nat_ipv[46]_fn().
785 * range is only used for new, uninitialized NAT state.
786 * Returns either NF_ACCEPT or NF_DROP.
787 */
788static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
789			      enum ip_conntrack_info ctinfo,
790			      const struct nf_nat_range2 *range,
791			      enum nf_nat_manip_type maniptype, struct sw_flow_key *key)
792{
793	int hooknum, nh_off, err = NF_ACCEPT;
794
795	nh_off = skb_network_offset(skb);
796	skb_pull_rcsum(skb, nh_off);
797
798	/* See HOOK2MANIP(). */
799	if (maniptype == NF_NAT_MANIP_SRC)
800		hooknum = NF_INET_LOCAL_IN; /* Source NAT */
801	else
802		hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
803
804	switch (ctinfo) {
805	case IP_CT_RELATED:
806	case IP_CT_RELATED_REPLY:
807		if (IS_ENABLED(CONFIG_NF_NAT) &&
808		    skb->protocol == htons(ETH_P_IP) &&
809		    ip_hdr(skb)->protocol == IPPROTO_ICMP) {
810			if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
811							   hooknum))
812				err = NF_DROP;
813			goto push;
814		} else if (IS_ENABLED(CONFIG_IPV6) &&
815			   skb->protocol == htons(ETH_P_IPV6)) {
816			__be16 frag_off;
817			u8 nexthdr = ipv6_hdr(skb)->nexthdr;
818			int hdrlen = ipv6_skip_exthdr(skb,
819						      sizeof(struct ipv6hdr),
820						      &nexthdr, &frag_off);
821
822			if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
823				if (!nf_nat_icmpv6_reply_translation(skb, ct,
824								     ctinfo,
825								     hooknum,
826								     hdrlen))
827					err = NF_DROP;
828				goto push;
829			}
830		}
831		/* Non-ICMP, fall thru to initialize if needed. */
832		fallthrough;
833	case IP_CT_NEW:
834		/* Seen it before?  This can happen for loopback, retrans,
835		 * or local packets.
836		 */
837		if (!nf_nat_initialized(ct, maniptype)) {
838			/* Initialize according to the NAT action. */
839			err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
840				/* Action is set up to establish a new
841				 * mapping.
842				 */
843				? nf_nat_setup_info(ct, range, maniptype)
844				: nf_nat_alloc_null_binding(ct, hooknum);
845			if (err != NF_ACCEPT)
846				goto push;
847		}
848		break;
849
850	case IP_CT_ESTABLISHED:
851	case IP_CT_ESTABLISHED_REPLY:
852		break;
853
854	default:
855		err = NF_DROP;
856		goto push;
857	}
858
859	err = nf_nat_packet(ct, ctinfo, hooknum, skb);
860push:
861	skb_push(skb, nh_off);
862	skb_postpush_rcsum(skb, skb->data, nh_off);
863
864	/* Update the flow key if NAT successful. */
865	if (err == NF_ACCEPT)
866		ovs_nat_update_key(key, skb, maniptype);
867
868	return err;
869}
870
871/* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
872static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
873		      const struct ovs_conntrack_info *info,
874		      struct sk_buff *skb, struct nf_conn *ct,
875		      enum ip_conntrack_info ctinfo)
876{
877	enum nf_nat_manip_type maniptype;
878	int err;
879
880	/* Add NAT extension if not confirmed yet. */
881	if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
882		return NF_ACCEPT;   /* Can't NAT. */
883
884	/* Determine NAT type.
885	 * Check if the NAT type can be deduced from the tracked connection.
886	 * Make sure new expected connections (IP_CT_RELATED) are NATted only
887	 * when committing.
888	 */
889	if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
890	    ct->status & IPS_NAT_MASK &&
891	    (ctinfo != IP_CT_RELATED || info->commit)) {
892		/* NAT an established or related connection like before. */
893		if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
894			/* This is the REPLY direction for a connection
895			 * for which NAT was applied in the forward
896			 * direction.  Do the reverse NAT.
897			 */
898			maniptype = ct->status & IPS_SRC_NAT
899				? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
900		else
901			maniptype = ct->status & IPS_SRC_NAT
902				? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
903	} else if (info->nat & OVS_CT_SRC_NAT) {
904		maniptype = NF_NAT_MANIP_SRC;
905	} else if (info->nat & OVS_CT_DST_NAT) {
906		maniptype = NF_NAT_MANIP_DST;
907	} else {
908		return NF_ACCEPT; /* Connection is not NATed. */
909	}
910	err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype, key);
911
912	if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) {
913		if (ct->status & IPS_SRC_NAT) {
914			if (maniptype == NF_NAT_MANIP_SRC)
915				maniptype = NF_NAT_MANIP_DST;
916			else
917				maniptype = NF_NAT_MANIP_SRC;
918
919			err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range,
920						 maniptype, key);
921		} else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
922			err = ovs_ct_nat_execute(skb, ct, ctinfo, NULL,
923						 NF_NAT_MANIP_SRC, key);
924		}
925	}
926
927	return err;
928}
929#else /* !CONFIG_NF_NAT */
930static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
931		      const struct ovs_conntrack_info *info,
932		      struct sk_buff *skb, struct nf_conn *ct,
933		      enum ip_conntrack_info ctinfo)
934{
935	return NF_ACCEPT;
936}
937#endif
938
939/* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
940 * not done already.  Update key with new CT state after passing the packet
941 * through conntrack.
942 * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
943 * set to NULL and 0 will be returned.
944 */
945static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
946			   const struct ovs_conntrack_info *info,
947			   struct sk_buff *skb)
948{
949	/* If we are recirculating packets to match on conntrack fields and
950	 * committing with a separate conntrack action,  then we don't need to
951	 * actually run the packet through conntrack twice unless it's for a
952	 * different zone.
953	 */
954	bool cached = skb_nfct_cached(net, key, info, skb);
955	enum ip_conntrack_info ctinfo;
956	struct nf_conn *ct;
957
958	if (!cached) {
959		struct nf_hook_state state = {
960			.hook = NF_INET_PRE_ROUTING,
961			.pf = info->family,
962			.net = net,
963		};
964		struct nf_conn *tmpl = info->ct;
965		int err;
966
967		/* Associate skb with specified zone. */
968		if (tmpl) {
969			if (skb_nfct(skb))
970				nf_conntrack_put(skb_nfct(skb));
971			nf_conntrack_get(&tmpl->ct_general);
972			nf_ct_set(skb, tmpl, IP_CT_NEW);
973		}
974
975		err = nf_conntrack_in(skb, &state);
976		if (err != NF_ACCEPT)
977			return -ENOENT;
978
979		/* Clear CT state NAT flags to mark that we have not yet done
980		 * NAT after the nf_conntrack_in() call.  We can actually clear
981		 * the whole state, as it will be re-initialized below.
982		 */
983		key->ct_state = 0;
984
985		/* Update the key, but keep the NAT flags. */
986		ovs_ct_update_key(skb, info, key, true, true);
987	}
988
989	ct = nf_ct_get(skb, &ctinfo);
990	if (ct) {
991		bool add_helper = false;
992
993		/* Packets starting a new connection must be NATted before the
994		 * helper, so that the helper knows about the NAT.  We enforce
995		 * this by delaying both NAT and helper calls for unconfirmed
996		 * connections until the committing CT action.  For later
997		 * packets NAT and Helper may be called in either order.
998		 *
999		 * NAT will be done only if the CT action has NAT, and only
1000		 * once per packet (per zone), as guarded by the NAT bits in
1001		 * the key->ct_state.
1002		 */
1003		if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
1004		    (nf_ct_is_confirmed(ct) || info->commit) &&
1005		    ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
1006			return -EINVAL;
1007		}
1008
1009		/* Userspace may decide to perform a ct lookup without a helper
1010		 * specified followed by a (recirculate and) commit with one,
1011		 * or attach a helper in a later commit.  Therefore, for
1012		 * connections which we will commit, we may need to attach
1013		 * the helper here.
1014		 */
1015		if (info->commit && info->helper && !nfct_help(ct)) {
1016			int err = __nf_ct_try_assign_helper(ct, info->ct,
1017							    GFP_ATOMIC);
1018			if (err)
1019				return err;
1020			add_helper = true;
1021
1022			/* helper installed, add seqadj if NAT is required */
1023			if (info->nat && !nfct_seqadj(ct)) {
1024				if (!nfct_seqadj_ext_add(ct))
1025					return -EINVAL;
1026			}
1027		}
1028
1029		/* Call the helper only if:
1030		 * - nf_conntrack_in() was executed above ("!cached") or a
1031		 *   helper was just attached ("add_helper") for a confirmed
1032		 *   connection, or
1033		 * - When committing an unconfirmed connection.
1034		 */
1035		if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
1036					      info->commit) &&
1037		    ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1038			return -EINVAL;
1039		}
1040	}
1041
1042	return 0;
1043}
1044
1045/* Lookup connection and read fields into key. */
1046static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1047			 const struct ovs_conntrack_info *info,
1048			 struct sk_buff *skb)
1049{
1050	struct nf_conntrack_expect *exp;
1051
1052	/* If we pass an expected packet through nf_conntrack_in() the
1053	 * expectation is typically removed, but the packet could still be
1054	 * lost in upcall processing.  To prevent this from happening we
1055	 * perform an explicit expectation lookup.  Expected connections are
1056	 * always new, and will be passed through conntrack only when they are
1057	 * committed, as it is OK to remove the expectation at that time.
1058	 */
1059	exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1060	if (exp) {
1061		u8 state;
1062
1063		/* NOTE: New connections are NATted and Helped only when
1064		 * committed, so we are not calling into NAT here.
1065		 */
1066		state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1067		__ovs_ct_update_key(key, state, &info->zone, exp->master);
1068	} else {
1069		struct nf_conn *ct;
1070		int err;
1071
1072		err = __ovs_ct_lookup(net, key, info, skb);
1073		if (err)
1074			return err;
1075
1076		ct = (struct nf_conn *)skb_nfct(skb);
1077		if (ct)
1078			nf_ct_deliver_cached_events(ct);
1079	}
1080
1081	return 0;
1082}
1083
1084static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1085{
1086	size_t i;
1087
1088	for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1089		if (labels->ct_labels_32[i])
1090			return true;
1091
1092	return false;
1093}
1094
1095#if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1096static struct hlist_head *ct_limit_hash_bucket(
1097	const struct ovs_ct_limit_info *info, u16 zone)
1098{
1099	return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1100}
1101
1102/* Call with ovs_mutex */
1103static void ct_limit_set(const struct ovs_ct_limit_info *info,
1104			 struct ovs_ct_limit *new_ct_limit)
1105{
1106	struct ovs_ct_limit *ct_limit;
1107	struct hlist_head *head;
1108
1109	head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1110	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1111		if (ct_limit->zone == new_ct_limit->zone) {
1112			hlist_replace_rcu(&ct_limit->hlist_node,
1113					  &new_ct_limit->hlist_node);
1114			kfree_rcu(ct_limit, rcu);
1115			return;
1116		}
1117	}
1118
1119	hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1120}
1121
1122/* Call with ovs_mutex */
1123static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1124{
1125	struct ovs_ct_limit *ct_limit;
1126	struct hlist_head *head;
1127	struct hlist_node *n;
1128
1129	head = ct_limit_hash_bucket(info, zone);
1130	hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1131		if (ct_limit->zone == zone) {
1132			hlist_del_rcu(&ct_limit->hlist_node);
1133			kfree_rcu(ct_limit, rcu);
1134			return;
1135		}
1136	}
1137}
1138
1139/* Call with RCU read lock */
1140static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1141{
1142	struct ovs_ct_limit *ct_limit;
1143	struct hlist_head *head;
1144
1145	head = ct_limit_hash_bucket(info, zone);
1146	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1147		if (ct_limit->zone == zone)
1148			return ct_limit->limit;
1149	}
1150
1151	return info->default_limit;
1152}
1153
1154static int ovs_ct_check_limit(struct net *net,
1155			      const struct ovs_conntrack_info *info,
1156			      const struct nf_conntrack_tuple *tuple)
1157{
1158	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1159	const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1160	u32 per_zone_limit, connections;
1161	u32 conncount_key;
1162
1163	conncount_key = info->zone.id;
1164
1165	per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1166	if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1167		return 0;
1168
1169	connections = nf_conncount_count(net, ct_limit_info->data,
1170					 &conncount_key, tuple, &info->zone);
1171	if (connections > per_zone_limit)
1172		return -ENOMEM;
1173
1174	return 0;
1175}
1176#endif
1177
1178/* Lookup connection and confirm if unconfirmed. */
1179static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1180			 const struct ovs_conntrack_info *info,
1181			 struct sk_buff *skb)
1182{
1183	enum ip_conntrack_info ctinfo;
1184	struct nf_conn *ct;
1185	int err;
1186
1187	err = __ovs_ct_lookup(net, key, info, skb);
1188	if (err)
1189		return err;
1190
1191	/* The connection could be invalid, in which case this is a no-op.*/
1192	ct = nf_ct_get(skb, &ctinfo);
1193	if (!ct)
1194		return 0;
1195
1196#if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1197	if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1198		if (!nf_ct_is_confirmed(ct)) {
1199			err = ovs_ct_check_limit(net, info,
1200				&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1201			if (err) {
1202				net_warn_ratelimited("openvswitch: zone: %u "
1203					"exceeds conntrack limit\n",
1204					info->zone.id);
1205				return err;
1206			}
1207		}
1208	}
1209#endif
1210
1211	/* Set the conntrack event mask if given.  NEW and DELETE events have
1212	 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1213	 * typically would receive many kinds of updates.  Setting the event
1214	 * mask allows those events to be filtered.  The set event mask will
1215	 * remain in effect for the lifetime of the connection unless changed
1216	 * by a further CT action with both the commit flag and the eventmask
1217	 * option. */
1218	if (info->have_eventmask) {
1219		struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1220
1221		if (cache)
1222			cache->ctmask = info->eventmask;
1223	}
1224
1225	/* Apply changes before confirming the connection so that the initial
1226	 * conntrack NEW netlink event carries the values given in the CT
1227	 * action.
1228	 */
1229	if (info->mark.mask) {
1230		err = ovs_ct_set_mark(ct, key, info->mark.value,
1231				      info->mark.mask);
1232		if (err)
1233			return err;
1234	}
1235	if (!nf_ct_is_confirmed(ct)) {
1236		err = ovs_ct_init_labels(ct, key, &info->labels.value,
1237					 &info->labels.mask);
1238		if (err)
1239			return err;
1240	} else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1241		   labels_nonzero(&info->labels.mask)) {
1242		err = ovs_ct_set_labels(ct, key, &info->labels.value,
1243					&info->labels.mask);
1244		if (err)
1245			return err;
1246	}
1247	/* This will take care of sending queued events even if the connection
1248	 * is already confirmed.
1249	 */
1250	if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1251		return -EINVAL;
1252
1253	return 0;
1254}
1255
1256/* Trim the skb to the length specified by the IP/IPv6 header,
1257 * removing any trailing lower-layer padding. This prepares the skb
1258 * for higher-layer processing that assumes skb->len excludes padding
1259 * (such as nf_ip_checksum). The caller needs to pull the skb to the
1260 * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1261 */
1262static int ovs_skb_network_trim(struct sk_buff *skb)
1263{
1264	unsigned int len;
1265	int err;
1266
1267	switch (skb->protocol) {
1268	case htons(ETH_P_IP):
1269		len = ntohs(ip_hdr(skb)->tot_len);
1270		break;
1271	case htons(ETH_P_IPV6):
1272		len = sizeof(struct ipv6hdr)
1273			+ ntohs(ipv6_hdr(skb)->payload_len);
1274		break;
1275	default:
1276		len = skb->len;
1277	}
1278
1279	err = pskb_trim_rcsum(skb, len);
1280	if (err)
1281		kfree_skb(skb);
1282
1283	return err;
1284}
1285
1286/* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1287 * value if 'skb' is freed.
1288 */
1289int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1290		   struct sw_flow_key *key,
1291		   const struct ovs_conntrack_info *info)
1292{
1293	int nh_ofs;
1294	int err;
1295
1296	/* The conntrack module expects to be working at L3. */
1297	nh_ofs = skb_network_offset(skb);
1298	skb_pull_rcsum(skb, nh_ofs);
1299
1300	err = ovs_skb_network_trim(skb);
1301	if (err)
1302		return err;
1303
1304	if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1305		err = handle_fragments(net, key, info->zone.id, skb);
1306		if (err)
1307			return err;
1308	}
1309
1310	if (info->commit)
1311		err = ovs_ct_commit(net, key, info, skb);
1312	else
1313		err = ovs_ct_lookup(net, key, info, skb);
1314
1315	skb_push(skb, nh_ofs);
1316	skb_postpush_rcsum(skb, skb->data, nh_ofs);
1317	if (err)
1318		kfree_skb(skb);
1319	return err;
1320}
1321
1322int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1323{
1324	if (skb_nfct(skb)) {
1325		nf_conntrack_put(skb_nfct(skb));
1326		nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1327		if (key)
1328			ovs_ct_fill_key(skb, key);
1329	}
1330
1331	return 0;
1332}
1333
1334static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1335			     const struct sw_flow_key *key, bool log)
1336{
1337	struct nf_conntrack_helper *helper;
1338	struct nf_conn_help *help;
1339	int ret = 0;
1340
1341	helper = nf_conntrack_helper_try_module_get(name, info->family,
1342						    key->ip.proto);
1343	if (!helper) {
1344		OVS_NLERR(log, "Unknown helper \"%s\"", name);
1345		return -EINVAL;
1346	}
1347
1348	help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1349	if (!help) {
1350		nf_conntrack_helper_put(helper);
1351		return -ENOMEM;
1352	}
1353
1354#if IS_ENABLED(CONFIG_NF_NAT)
1355	if (info->nat) {
1356		ret = nf_nat_helper_try_module_get(name, info->family,
1357						   key->ip.proto);
1358		if (ret) {
1359			nf_conntrack_helper_put(helper);
1360			OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1361				  name, ret);
1362			return ret;
1363		}
1364	}
1365#endif
1366	rcu_assign_pointer(help->helper, helper);
1367	info->helper = helper;
1368	return ret;
1369}
1370
1371#if IS_ENABLED(CONFIG_NF_NAT)
1372static int parse_nat(const struct nlattr *attr,
1373		     struct ovs_conntrack_info *info, bool log)
1374{
1375	struct nlattr *a;
1376	int rem;
1377	bool have_ip_max = false;
1378	bool have_proto_max = false;
1379	bool ip_vers = (info->family == NFPROTO_IPV6);
1380
1381	nla_for_each_nested(a, attr, rem) {
1382		static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1383			[OVS_NAT_ATTR_SRC] = {0, 0},
1384			[OVS_NAT_ATTR_DST] = {0, 0},
1385			[OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1386						 sizeof(struct in6_addr)},
1387			[OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1388						 sizeof(struct in6_addr)},
1389			[OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1390			[OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1391			[OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1392			[OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1393			[OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1394		};
1395		int type = nla_type(a);
1396
1397		if (type > OVS_NAT_ATTR_MAX) {
1398			OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1399				  type, OVS_NAT_ATTR_MAX);
1400			return -EINVAL;
1401		}
1402
1403		if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1404			OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1405				  type, nla_len(a),
1406				  ovs_nat_attr_lens[type][ip_vers]);
1407			return -EINVAL;
1408		}
1409
1410		switch (type) {
1411		case OVS_NAT_ATTR_SRC:
1412		case OVS_NAT_ATTR_DST:
1413			if (info->nat) {
1414				OVS_NLERR(log, "Only one type of NAT may be specified");
1415				return -ERANGE;
1416			}
1417			info->nat |= OVS_CT_NAT;
1418			info->nat |= ((type == OVS_NAT_ATTR_SRC)
1419					? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1420			break;
1421
1422		case OVS_NAT_ATTR_IP_MIN:
1423			nla_memcpy(&info->range.min_addr, a,
1424				   sizeof(info->range.min_addr));
1425			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1426			break;
1427
1428		case OVS_NAT_ATTR_IP_MAX:
1429			have_ip_max = true;
1430			nla_memcpy(&info->range.max_addr, a,
1431				   sizeof(info->range.max_addr));
1432			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1433			break;
1434
1435		case OVS_NAT_ATTR_PROTO_MIN:
1436			info->range.min_proto.all = htons(nla_get_u16(a));
1437			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1438			break;
1439
1440		case OVS_NAT_ATTR_PROTO_MAX:
1441			have_proto_max = true;
1442			info->range.max_proto.all = htons(nla_get_u16(a));
1443			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1444			break;
1445
1446		case OVS_NAT_ATTR_PERSISTENT:
1447			info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1448			break;
1449
1450		case OVS_NAT_ATTR_PROTO_HASH:
1451			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1452			break;
1453
1454		case OVS_NAT_ATTR_PROTO_RANDOM:
1455			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1456			break;
1457
1458		default:
1459			OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1460			return -EINVAL;
1461		}
1462	}
1463
1464	if (rem > 0) {
1465		OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1466		return -EINVAL;
1467	}
1468	if (!info->nat) {
1469		/* Do not allow flags if no type is given. */
1470		if (info->range.flags) {
1471			OVS_NLERR(log,
1472				  "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1473				  );
1474			return -EINVAL;
1475		}
1476		info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1477	} else if (!info->commit) {
1478		OVS_NLERR(log,
1479			  "NAT attributes may be specified only when CT COMMIT flag is also specified."
1480			  );
1481		return -EINVAL;
1482	}
1483	/* Allow missing IP_MAX. */
1484	if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1485		memcpy(&info->range.max_addr, &info->range.min_addr,
1486		       sizeof(info->range.max_addr));
1487	}
1488	/* Allow missing PROTO_MAX. */
1489	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1490	    !have_proto_max) {
1491		info->range.max_proto.all = info->range.min_proto.all;
1492	}
1493	return 0;
1494}
1495#endif
1496
1497static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1498	[OVS_CT_ATTR_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1499	[OVS_CT_ATTR_FORCE_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1500	[OVS_CT_ATTR_ZONE]	= { .minlen = sizeof(u16),
1501				    .maxlen = sizeof(u16) },
1502	[OVS_CT_ATTR_MARK]	= { .minlen = sizeof(struct md_mark),
1503				    .maxlen = sizeof(struct md_mark) },
1504	[OVS_CT_ATTR_LABELS]	= { .minlen = sizeof(struct md_labels),
1505				    .maxlen = sizeof(struct md_labels) },
1506	[OVS_CT_ATTR_HELPER]	= { .minlen = 1,
1507				    .maxlen = NF_CT_HELPER_NAME_LEN },
1508#if IS_ENABLED(CONFIG_NF_NAT)
1509	/* NAT length is checked when parsing the nested attributes. */
1510	[OVS_CT_ATTR_NAT]	= { .minlen = 0, .maxlen = INT_MAX },
1511#endif
1512	[OVS_CT_ATTR_EVENTMASK]	= { .minlen = sizeof(u32),
1513				    .maxlen = sizeof(u32) },
1514	[OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1515				  .maxlen = CTNL_TIMEOUT_NAME_MAX },
1516};
1517
1518static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1519		    const char **helper, bool log)
1520{
1521	struct nlattr *a;
1522	int rem;
1523
1524	nla_for_each_nested(a, attr, rem) {
1525		int type = nla_type(a);
1526		int maxlen;
1527		int minlen;
1528
1529		if (type > OVS_CT_ATTR_MAX) {
1530			OVS_NLERR(log,
1531				  "Unknown conntrack attr (type=%d, max=%d)",
1532				  type, OVS_CT_ATTR_MAX);
1533			return -EINVAL;
1534		}
1535
1536		maxlen = ovs_ct_attr_lens[type].maxlen;
1537		minlen = ovs_ct_attr_lens[type].minlen;
1538		if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1539			OVS_NLERR(log,
1540				  "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1541				  type, nla_len(a), maxlen);
1542			return -EINVAL;
1543		}
1544
1545		switch (type) {
1546		case OVS_CT_ATTR_FORCE_COMMIT:
1547			info->force = true;
1548			fallthrough;
1549		case OVS_CT_ATTR_COMMIT:
1550			info->commit = true;
1551			break;
1552#ifdef CONFIG_NF_CONNTRACK_ZONES
1553		case OVS_CT_ATTR_ZONE:
1554			info->zone.id = nla_get_u16(a);
1555			break;
1556#endif
1557#ifdef CONFIG_NF_CONNTRACK_MARK
1558		case OVS_CT_ATTR_MARK: {
1559			struct md_mark *mark = nla_data(a);
1560
1561			if (!mark->mask) {
1562				OVS_NLERR(log, "ct_mark mask cannot be 0");
1563				return -EINVAL;
1564			}
1565			info->mark = *mark;
1566			break;
1567		}
1568#endif
1569#ifdef CONFIG_NF_CONNTRACK_LABELS
1570		case OVS_CT_ATTR_LABELS: {
1571			struct md_labels *labels = nla_data(a);
1572
1573			if (!labels_nonzero(&labels->mask)) {
1574				OVS_NLERR(log, "ct_labels mask cannot be 0");
1575				return -EINVAL;
1576			}
1577			info->labels = *labels;
1578			break;
1579		}
1580#endif
1581		case OVS_CT_ATTR_HELPER:
1582			*helper = nla_data(a);
1583			if (!memchr(*helper, '\0', nla_len(a))) {
1584				OVS_NLERR(log, "Invalid conntrack helper");
1585				return -EINVAL;
1586			}
1587			break;
1588#if IS_ENABLED(CONFIG_NF_NAT)
1589		case OVS_CT_ATTR_NAT: {
1590			int err = parse_nat(a, info, log);
1591
1592			if (err)
1593				return err;
1594			break;
1595		}
1596#endif
1597		case OVS_CT_ATTR_EVENTMASK:
1598			info->have_eventmask = true;
1599			info->eventmask = nla_get_u32(a);
1600			break;
1601#ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1602		case OVS_CT_ATTR_TIMEOUT:
1603			memcpy(info->timeout, nla_data(a), nla_len(a));
1604			if (!memchr(info->timeout, '\0', nla_len(a))) {
1605				OVS_NLERR(log, "Invalid conntrack timeout");
1606				return -EINVAL;
1607			}
1608			break;
1609#endif
1610
1611		default:
1612			OVS_NLERR(log, "Unknown conntrack attr (%d)",
1613				  type);
1614			return -EINVAL;
1615		}
1616	}
1617
1618#ifdef CONFIG_NF_CONNTRACK_MARK
1619	if (!info->commit && info->mark.mask) {
1620		OVS_NLERR(log,
1621			  "Setting conntrack mark requires 'commit' flag.");
1622		return -EINVAL;
1623	}
1624#endif
1625#ifdef CONFIG_NF_CONNTRACK_LABELS
1626	if (!info->commit && labels_nonzero(&info->labels.mask)) {
1627		OVS_NLERR(log,
1628			  "Setting conntrack labels requires 'commit' flag.");
1629		return -EINVAL;
1630	}
1631#endif
1632	if (rem > 0) {
1633		OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1634		return -EINVAL;
1635	}
1636
1637	return 0;
1638}
1639
1640bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1641{
1642	if (attr == OVS_KEY_ATTR_CT_STATE)
1643		return true;
1644	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1645	    attr == OVS_KEY_ATTR_CT_ZONE)
1646		return true;
1647	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1648	    attr == OVS_KEY_ATTR_CT_MARK)
1649		return true;
1650	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1651	    attr == OVS_KEY_ATTR_CT_LABELS) {
1652		struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1653
1654		return ovs_net->xt_label;
1655	}
1656
1657	return false;
1658}
1659
1660int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1661		       const struct sw_flow_key *key,
1662		       struct sw_flow_actions **sfa,  bool log)
1663{
1664	struct ovs_conntrack_info ct_info;
1665	const char *helper = NULL;
1666	u16 family;
1667	int err;
1668
1669	family = key_to_nfproto(key);
1670	if (family == NFPROTO_UNSPEC) {
1671		OVS_NLERR(log, "ct family unspecified");
1672		return -EINVAL;
1673	}
1674
1675	memset(&ct_info, 0, sizeof(ct_info));
1676	ct_info.family = family;
1677
1678	nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1679			NF_CT_DEFAULT_ZONE_DIR, 0);
1680
1681	err = parse_ct(attr, &ct_info, &helper, log);
1682	if (err)
1683		return err;
1684
1685	/* Set up template for tracking connections in specific zones. */
1686	ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1687	if (!ct_info.ct) {
1688		OVS_NLERR(log, "Failed to allocate conntrack template");
1689		return -ENOMEM;
1690	}
1691
1692	if (ct_info.timeout[0]) {
1693		if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1694				      ct_info.timeout))
1695			pr_info_ratelimited("Failed to associated timeout "
1696					    "policy `%s'\n", ct_info.timeout);
1697		else
1698			ct_info.nf_ct_timeout = rcu_dereference(
1699				nf_ct_timeout_find(ct_info.ct)->timeout);
1700
1701	}
1702
1703	if (helper) {
1704		err = ovs_ct_add_helper(&ct_info, helper, key, log);
1705		if (err)
1706			goto err_free_ct;
1707	}
1708
1709	err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1710				 sizeof(ct_info), log);
1711	if (err)
1712		goto err_free_ct;
1713
1714	__set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1715	nf_conntrack_get(&ct_info.ct->ct_general);
1716	return 0;
1717err_free_ct:
1718	__ovs_ct_free_action(&ct_info);
1719	return err;
1720}
1721
1722#if IS_ENABLED(CONFIG_NF_NAT)
1723static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1724			       struct sk_buff *skb)
1725{
1726	struct nlattr *start;
1727
1728	start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1729	if (!start)
1730		return false;
1731
1732	if (info->nat & OVS_CT_SRC_NAT) {
1733		if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1734			return false;
1735	} else if (info->nat & OVS_CT_DST_NAT) {
1736		if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1737			return false;
1738	} else {
1739		goto out;
1740	}
1741
1742	if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1743		if (IS_ENABLED(CONFIG_NF_NAT) &&
1744		    info->family == NFPROTO_IPV4) {
1745			if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1746					    info->range.min_addr.ip) ||
1747			    (info->range.max_addr.ip
1748			     != info->range.min_addr.ip &&
1749			     (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1750					      info->range.max_addr.ip))))
1751				return false;
1752		} else if (IS_ENABLED(CONFIG_IPV6) &&
1753			   info->family == NFPROTO_IPV6) {
1754			if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1755					     &info->range.min_addr.in6) ||
1756			    (memcmp(&info->range.max_addr.in6,
1757				    &info->range.min_addr.in6,
1758				    sizeof(info->range.max_addr.in6)) &&
1759			     (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1760					       &info->range.max_addr.in6))))
1761				return false;
1762		} else {
1763			return false;
1764		}
1765	}
1766	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1767	    (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1768			 ntohs(info->range.min_proto.all)) ||
1769	     (info->range.max_proto.all != info->range.min_proto.all &&
1770	      nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1771			  ntohs(info->range.max_proto.all)))))
1772		return false;
1773
1774	if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1775	    nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1776		return false;
1777	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1778	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1779		return false;
1780	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1781	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1782		return false;
1783out:
1784	nla_nest_end(skb, start);
1785
1786	return true;
1787}
1788#endif
1789
1790int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1791			  struct sk_buff *skb)
1792{
1793	struct nlattr *start;
1794
1795	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1796	if (!start)
1797		return -EMSGSIZE;
1798
1799	if (ct_info->commit && nla_put_flag(skb, ct_info->force
1800					    ? OVS_CT_ATTR_FORCE_COMMIT
1801					    : OVS_CT_ATTR_COMMIT))
1802		return -EMSGSIZE;
1803	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1804	    nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1805		return -EMSGSIZE;
1806	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1807	    nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1808		    &ct_info->mark))
1809		return -EMSGSIZE;
1810	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1811	    labels_nonzero(&ct_info->labels.mask) &&
1812	    nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1813		    &ct_info->labels))
1814		return -EMSGSIZE;
1815	if (ct_info->helper) {
1816		if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1817				   ct_info->helper->name))
1818			return -EMSGSIZE;
1819	}
1820	if (ct_info->have_eventmask &&
1821	    nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1822		return -EMSGSIZE;
1823	if (ct_info->timeout[0]) {
1824		if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1825			return -EMSGSIZE;
1826	}
1827
1828#if IS_ENABLED(CONFIG_NF_NAT)
1829	if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1830		return -EMSGSIZE;
1831#endif
1832	nla_nest_end(skb, start);
1833
1834	return 0;
1835}
1836
1837void ovs_ct_free_action(const struct nlattr *a)
1838{
1839	struct ovs_conntrack_info *ct_info = nla_data(a);
1840
1841	__ovs_ct_free_action(ct_info);
1842}
1843
1844static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1845{
1846	if (ct_info->helper) {
1847#if IS_ENABLED(CONFIG_NF_NAT)
1848		if (ct_info->nat)
1849			nf_nat_helper_put(ct_info->helper);
1850#endif
1851		nf_conntrack_helper_put(ct_info->helper);
1852	}
1853	if (ct_info->ct) {
1854		if (ct_info->timeout[0])
1855			nf_ct_destroy_timeout(ct_info->ct);
1856		nf_ct_tmpl_free(ct_info->ct);
1857	}
1858}
1859
1860#if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1861static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1862{
1863	int i, err;
1864
1865	ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1866					 GFP_KERNEL);
1867	if (!ovs_net->ct_limit_info)
1868		return -ENOMEM;
1869
1870	ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1871	ovs_net->ct_limit_info->limits =
1872		kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1873			      GFP_KERNEL);
1874	if (!ovs_net->ct_limit_info->limits) {
1875		kfree(ovs_net->ct_limit_info);
1876		return -ENOMEM;
1877	}
1878
1879	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1880		INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1881
1882	ovs_net->ct_limit_info->data =
1883		nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1884
1885	if (IS_ERR(ovs_net->ct_limit_info->data)) {
1886		err = PTR_ERR(ovs_net->ct_limit_info->data);
1887		kfree(ovs_net->ct_limit_info->limits);
1888		kfree(ovs_net->ct_limit_info);
1889		pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1890		return err;
1891	}
1892	return 0;
1893}
1894
1895static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1896{
1897	const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1898	int i;
1899
1900	nf_conncount_destroy(net, NFPROTO_INET, info->data);
1901	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1902		struct hlist_head *head = &info->limits[i];
1903		struct ovs_ct_limit *ct_limit;
1904
1905		hlist_for_each_entry_rcu(ct_limit, head, hlist_node,
1906					 lockdep_ovsl_is_held())
1907			kfree_rcu(ct_limit, rcu);
1908	}
1909	kfree(info->limits);
1910	kfree(info);
1911}
1912
1913static struct sk_buff *
1914ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1915			     struct ovs_header **ovs_reply_header)
1916{
1917	struct ovs_header *ovs_header = info->userhdr;
1918	struct sk_buff *skb;
1919
1920	skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1921	if (!skb)
1922		return ERR_PTR(-ENOMEM);
1923
1924	*ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1925					info->snd_seq,
1926					&dp_ct_limit_genl_family, 0, cmd);
1927
1928	if (!*ovs_reply_header) {
1929		nlmsg_free(skb);
1930		return ERR_PTR(-EMSGSIZE);
1931	}
1932	(*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1933
1934	return skb;
1935}
1936
1937static bool check_zone_id(int zone_id, u16 *pzone)
1938{
1939	if (zone_id >= 0 && zone_id <= 65535) {
1940		*pzone = (u16)zone_id;
1941		return true;
1942	}
1943	return false;
1944}
1945
1946static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1947				       struct ovs_ct_limit_info *info)
1948{
1949	struct ovs_zone_limit *zone_limit;
1950	int rem;
1951	u16 zone;
1952
1953	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1954	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1955
1956	while (rem >= sizeof(*zone_limit)) {
1957		if (unlikely(zone_limit->zone_id ==
1958				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1959			ovs_lock();
1960			info->default_limit = zone_limit->limit;
1961			ovs_unlock();
1962		} else if (unlikely(!check_zone_id(
1963				zone_limit->zone_id, &zone))) {
1964			OVS_NLERR(true, "zone id is out of range");
1965		} else {
1966			struct ovs_ct_limit *ct_limit;
1967
1968			ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1969			if (!ct_limit)
1970				return -ENOMEM;
1971
1972			ct_limit->zone = zone;
1973			ct_limit->limit = zone_limit->limit;
1974
1975			ovs_lock();
1976			ct_limit_set(info, ct_limit);
1977			ovs_unlock();
1978		}
1979		rem -= NLA_ALIGN(sizeof(*zone_limit));
1980		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1981				NLA_ALIGN(sizeof(*zone_limit)));
1982	}
1983
1984	if (rem)
1985		OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1986
1987	return 0;
1988}
1989
1990static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1991				       struct ovs_ct_limit_info *info)
1992{
1993	struct ovs_zone_limit *zone_limit;
1994	int rem;
1995	u16 zone;
1996
1997	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1998	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1999
2000	while (rem >= sizeof(*zone_limit)) {
2001		if (unlikely(zone_limit->zone_id ==
2002				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2003			ovs_lock();
2004			info->default_limit = OVS_CT_LIMIT_DEFAULT;
2005			ovs_unlock();
2006		} else if (unlikely(!check_zone_id(
2007				zone_limit->zone_id, &zone))) {
2008			OVS_NLERR(true, "zone id is out of range");
2009		} else {
2010			ovs_lock();
2011			ct_limit_del(info, zone);
2012			ovs_unlock();
2013		}
2014		rem -= NLA_ALIGN(sizeof(*zone_limit));
2015		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2016				NLA_ALIGN(sizeof(*zone_limit)));
2017	}
2018
2019	if (rem)
2020		OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
2021
2022	return 0;
2023}
2024
2025static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2026					  struct sk_buff *reply)
2027{
2028	struct ovs_zone_limit zone_limit = {
2029		.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE,
2030		.limit   = info->default_limit,
2031	};
2032
2033	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2034}
2035
2036static int __ovs_ct_limit_get_zone_limit(struct net *net,
2037					 struct nf_conncount_data *data,
2038					 u16 zone_id, u32 limit,
2039					 struct sk_buff *reply)
2040{
2041	struct nf_conntrack_zone ct_zone;
2042	struct ovs_zone_limit zone_limit;
2043	u32 conncount_key = zone_id;
2044
2045	zone_limit.zone_id = zone_id;
2046	zone_limit.limit = limit;
2047	nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2048
2049	zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2050					      &ct_zone);
2051	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2052}
2053
2054static int ovs_ct_limit_get_zone_limit(struct net *net,
2055				       struct nlattr *nla_zone_limit,
2056				       struct ovs_ct_limit_info *info,
2057				       struct sk_buff *reply)
2058{
2059	struct ovs_zone_limit *zone_limit;
2060	int rem, err;
2061	u32 limit;
2062	u16 zone;
2063
2064	rem = NLA_ALIGN(nla_len(nla_zone_limit));
2065	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2066
2067	while (rem >= sizeof(*zone_limit)) {
2068		if (unlikely(zone_limit->zone_id ==
2069				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2070			err = ovs_ct_limit_get_default_limit(info, reply);
2071			if (err)
2072				return err;
2073		} else if (unlikely(!check_zone_id(zone_limit->zone_id,
2074							&zone))) {
2075			OVS_NLERR(true, "zone id is out of range");
2076		} else {
2077			rcu_read_lock();
2078			limit = ct_limit_get(info, zone);
2079			rcu_read_unlock();
2080
2081			err = __ovs_ct_limit_get_zone_limit(
2082				net, info->data, zone, limit, reply);
2083			if (err)
2084				return err;
2085		}
2086		rem -= NLA_ALIGN(sizeof(*zone_limit));
2087		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2088				NLA_ALIGN(sizeof(*zone_limit)));
2089	}
2090
2091	if (rem)
2092		OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2093
2094	return 0;
2095}
2096
2097static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2098					   struct ovs_ct_limit_info *info,
2099					   struct sk_buff *reply)
2100{
2101	struct ovs_ct_limit *ct_limit;
2102	struct hlist_head *head;
2103	int i, err = 0;
2104
2105	err = ovs_ct_limit_get_default_limit(info, reply);
2106	if (err)
2107		return err;
2108
2109	rcu_read_lock();
2110	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2111		head = &info->limits[i];
2112		hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2113			err = __ovs_ct_limit_get_zone_limit(net, info->data,
2114				ct_limit->zone, ct_limit->limit, reply);
2115			if (err)
2116				goto exit_err;
2117		}
2118	}
2119
2120exit_err:
2121	rcu_read_unlock();
2122	return err;
2123}
2124
2125static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2126{
2127	struct nlattr **a = info->attrs;
2128	struct sk_buff *reply;
2129	struct ovs_header *ovs_reply_header;
2130	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2131	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2132	int err;
2133
2134	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2135					     &ovs_reply_header);
2136	if (IS_ERR(reply))
2137		return PTR_ERR(reply);
2138
2139	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2140		err = -EINVAL;
2141		goto exit_err;
2142	}
2143
2144	err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2145					  ct_limit_info);
2146	if (err)
2147		goto exit_err;
2148
2149	static_branch_enable(&ovs_ct_limit_enabled);
2150
2151	genlmsg_end(reply, ovs_reply_header);
2152	return genlmsg_reply(reply, info);
2153
2154exit_err:
2155	nlmsg_free(reply);
2156	return err;
2157}
2158
2159static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2160{
2161	struct nlattr **a = info->attrs;
2162	struct sk_buff *reply;
2163	struct ovs_header *ovs_reply_header;
2164	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2165	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2166	int err;
2167
2168	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2169					     &ovs_reply_header);
2170	if (IS_ERR(reply))
2171		return PTR_ERR(reply);
2172
2173	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2174		err = -EINVAL;
2175		goto exit_err;
2176	}
2177
2178	err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2179					  ct_limit_info);
2180	if (err)
2181		goto exit_err;
2182
2183	genlmsg_end(reply, ovs_reply_header);
2184	return genlmsg_reply(reply, info);
2185
2186exit_err:
2187	nlmsg_free(reply);
2188	return err;
2189}
2190
2191static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2192{
2193	struct nlattr **a = info->attrs;
2194	struct nlattr *nla_reply;
2195	struct sk_buff *reply;
2196	struct ovs_header *ovs_reply_header;
2197	struct net *net = sock_net(skb->sk);
2198	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2199	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2200	int err;
2201
2202	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2203					     &ovs_reply_header);
2204	if (IS_ERR(reply))
2205		return PTR_ERR(reply);
2206
2207	nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2208	if (!nla_reply) {
2209		err = -EMSGSIZE;
2210		goto exit_err;
2211	}
2212
2213	if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2214		err = ovs_ct_limit_get_zone_limit(
2215			net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2216			reply);
2217		if (err)
2218			goto exit_err;
2219	} else {
2220		err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2221						      reply);
2222		if (err)
2223			goto exit_err;
2224	}
2225
2226	nla_nest_end(reply, nla_reply);
2227	genlmsg_end(reply, ovs_reply_header);
2228	return genlmsg_reply(reply, info);
2229
2230exit_err:
2231	nlmsg_free(reply);
2232	return err;
2233}
2234
2235static const struct genl_small_ops ct_limit_genl_ops[] = {
2236	{ .cmd = OVS_CT_LIMIT_CMD_SET,
2237		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2238		.flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2239					   * privilege. */
2240		.doit = ovs_ct_limit_cmd_set,
2241	},
2242	{ .cmd = OVS_CT_LIMIT_CMD_DEL,
2243		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2244		.flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2245					   * privilege. */
2246		.doit = ovs_ct_limit_cmd_del,
2247	},
2248	{ .cmd = OVS_CT_LIMIT_CMD_GET,
2249		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2250		.flags = 0,		  /* OK for unprivileged users. */
2251		.doit = ovs_ct_limit_cmd_get,
2252	},
2253};
2254
2255static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2256	.name = OVS_CT_LIMIT_MCGROUP,
2257};
2258
2259struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2260	.hdrsize = sizeof(struct ovs_header),
2261	.name = OVS_CT_LIMIT_FAMILY,
2262	.version = OVS_CT_LIMIT_VERSION,
2263	.maxattr = OVS_CT_LIMIT_ATTR_MAX,
2264	.policy = ct_limit_policy,
2265	.netnsok = true,
2266	.parallel_ops = true,
2267	.small_ops = ct_limit_genl_ops,
2268	.n_small_ops = ARRAY_SIZE(ct_limit_genl_ops),
2269	.mcgrps = &ovs_ct_limit_multicast_group,
2270	.n_mcgrps = 1,
2271	.module = THIS_MODULE,
2272};
2273#endif
2274
2275int ovs_ct_init(struct net *net)
2276{
2277	unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2278	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2279
2280	if (nf_connlabels_get(net, n_bits - 1)) {
2281		ovs_net->xt_label = false;
2282		OVS_NLERR(true, "Failed to set connlabel length");
2283	} else {
2284		ovs_net->xt_label = true;
2285	}
2286
2287#if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2288	return ovs_ct_limit_init(net, ovs_net);
2289#else
2290	return 0;
2291#endif
2292}
2293
2294void ovs_ct_exit(struct net *net)
2295{
2296	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2297
2298#if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2299	ovs_ct_limit_exit(net, ovs_net);
2300#endif
2301
2302	if (ovs_net->xt_label)
2303		nf_connlabels_put(net);
2304}
2305