1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * This is a module which is used for queueing packets and communicating with
4  * userspace via nfnetlink.
5  *
6  * (C) 2005 by Harald Welte <laforge@netfilter.org>
7  * (C) 2007 by Patrick McHardy <kaber@trash.net>
8  *
9  * Based on the old ipv4-only ip_queue.c:
10  * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
11  * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
12  */
13 
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 
16 #include <linux/module.h>
17 #include <linux/skbuff.h>
18 #include <linux/init.h>
19 #include <linux/spinlock.h>
20 #include <linux/slab.h>
21 #include <linux/notifier.h>
22 #include <linux/netdevice.h>
23 #include <linux/netfilter.h>
24 #include <linux/proc_fs.h>
25 #include <linux/netfilter_ipv4.h>
26 #include <linux/netfilter_ipv6.h>
27 #include <linux/netfilter_bridge.h>
28 #include <linux/netfilter/nfnetlink.h>
29 #include <linux/netfilter/nfnetlink_queue.h>
30 #include <linux/netfilter/nf_conntrack_common.h>
31 #include <linux/list.h>
32 #include <linux/cgroup-defs.h>
33 #include <net/gso.h>
34 #include <net/sock.h>
35 #include <net/tcp_states.h>
36 #include <net/netfilter/nf_queue.h>
37 #include <net/netns/generic.h>
38 
39 #include <linux/atomic.h>
40 
41 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
42 #include "../bridge/br_private.h"
43 #endif
44 
45 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
46 #include <net/netfilter/nf_conntrack.h>
47 #endif
48 
49 #define NFQNL_QMAX_DEFAULT 1024
50 
51 /* We're using struct nlattr which has 16bit nla_len. Note that nla_len
52  * includes the header length. Thus, the maximum packet length that we
53  * support is 65531 bytes. We send truncated packets if the specified length
54  * is larger than that.  Userspace can check for presence of NFQA_CAP_LEN
55  * attribute to detect truncation.
56  */
57 #define NFQNL_MAX_COPY_RANGE (0xffff - NLA_HDRLEN)
58 
59 struct nfqnl_instance {
60 	struct hlist_node hlist;		/* global list of queues */
61 	struct rcu_head rcu;
62 
63 	u32 peer_portid;
64 	unsigned int queue_maxlen;
65 	unsigned int copy_range;
66 	unsigned int queue_dropped;
67 	unsigned int queue_user_dropped;
68 
69 
70 	u_int16_t queue_num;			/* number of this queue */
71 	u_int8_t copy_mode;
72 	u_int32_t flags;			/* Set using NFQA_CFG_FLAGS */
73 /*
74  * Following fields are dirtied for each queued packet,
75  * keep them in same cache line if possible.
76  */
77 	spinlock_t	lock	____cacheline_aligned_in_smp;
78 	unsigned int	queue_total;
79 	unsigned int	id_sequence;		/* 'sequence' of pkt ids */
80 	struct list_head queue_list;		/* packets in queue */
81 };
82 
83 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
84 
85 static unsigned int nfnl_queue_net_id __read_mostly;
86 
87 #define INSTANCE_BUCKETS	16
88 struct nfnl_queue_net {
89 	spinlock_t instances_lock;
90 	struct hlist_head instance_table[INSTANCE_BUCKETS];
91 };
92 
93 static struct nfnl_queue_net *nfnl_queue_pernet(struct net *net)
94 {
95 	return net_generic(net, nfnl_queue_net_id);
96 }
97 
98 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
99 {
100 	return ((queue_num >> 8) ^ queue_num) % INSTANCE_BUCKETS;
101 }
102 
103 static struct nfqnl_instance *
104 instance_lookup(struct nfnl_queue_net *q, u_int16_t queue_num)
105 {
106 	struct hlist_head *head;
107 	struct nfqnl_instance *inst;
108 
109 	head = &q->instance_table[instance_hashfn(queue_num)];
110 	hlist_for_each_entry_rcu(inst, head, hlist) {
111 		if (inst->queue_num == queue_num)
112 			return inst;
113 	}
114 	return NULL;
115 }
116 
117 static struct nfqnl_instance *
118 instance_create(struct nfnl_queue_net *q, u_int16_t queue_num, u32 portid)
119 {
120 	struct nfqnl_instance *inst;
121 	unsigned int h;
122 	int err;
123 
124 	spin_lock(&q->instances_lock);
125 	if (instance_lookup(q, queue_num)) {
126 		err = -EEXIST;
127 		goto out_unlock;
128 	}
129 
130 	inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
131 	if (!inst) {
132 		err = -ENOMEM;
133 		goto out_unlock;
134 	}
135 
136 	inst->queue_num = queue_num;
137 	inst->peer_portid = portid;
138 	inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
139 	inst->copy_range = NFQNL_MAX_COPY_RANGE;
140 	inst->copy_mode = NFQNL_COPY_NONE;
141 	spin_lock_init(&inst->lock);
142 	INIT_LIST_HEAD(&inst->queue_list);
143 
144 	if (!try_module_get(THIS_MODULE)) {
145 		err = -EAGAIN;
146 		goto out_free;
147 	}
148 
149 	h = instance_hashfn(queue_num);
150 	hlist_add_head_rcu(&inst->hlist, &q->instance_table[h]);
151 
152 	spin_unlock(&q->instances_lock);
153 
154 	return inst;
155 
156 out_free:
157 	kfree(inst);
158 out_unlock:
159 	spin_unlock(&q->instances_lock);
160 	return ERR_PTR(err);
161 }
162 
163 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
164 			unsigned long data);
165 
166 static void
167 instance_destroy_rcu(struct rcu_head *head)
168 {
169 	struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
170 						   rcu);
171 
172 	rcu_read_lock();
173 	nfqnl_flush(inst, NULL, 0);
174 	rcu_read_unlock();
175 	kfree(inst);
176 	module_put(THIS_MODULE);
177 }
178 
179 static void
180 __instance_destroy(struct nfqnl_instance *inst)
181 {
182 	hlist_del_rcu(&inst->hlist);
183 	call_rcu(&inst->rcu, instance_destroy_rcu);
184 }
185 
186 static void
187 instance_destroy(struct nfnl_queue_net *q, struct nfqnl_instance *inst)
188 {
189 	spin_lock(&q->instances_lock);
190 	__instance_destroy(inst);
191 	spin_unlock(&q->instances_lock);
192 }
193 
194 static inline void
195 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
196 {
197        list_add_tail(&entry->list, &queue->queue_list);
198        queue->queue_total++;
199 }
200 
201 static void
202 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
203 {
204 	list_del(&entry->list);
205 	queue->queue_total--;
206 }
207 
208 static struct nf_queue_entry *
209 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
210 {
211 	struct nf_queue_entry *entry = NULL, *i;
212 
213 	spin_lock_bh(&queue->lock);
214 
215 	list_for_each_entry(i, &queue->queue_list, list) {
216 		if (i->id == id) {
217 			entry = i;
218 			break;
219 		}
220 	}
221 
222 	if (entry)
223 		__dequeue_entry(queue, entry);
224 
225 	spin_unlock_bh(&queue->lock);
226 
227 	return entry;
228 }
229 
230 static unsigned int nf_iterate(struct sk_buff *skb,
231 			       struct nf_hook_state *state,
232 			       const struct nf_hook_entries *hooks,
233 			       unsigned int *index)
234 {
235 	const struct nf_hook_entry *hook;
236 	unsigned int verdict, i = *index;
237 
238 	while (i < hooks->num_hook_entries) {
239 		hook = &hooks->hooks[i];
240 repeat:
241 		verdict = nf_hook_entry_hookfn(hook, skb, state);
242 		if (verdict != NF_ACCEPT) {
243 			*index = i;
244 			if (verdict != NF_REPEAT)
245 				return verdict;
246 			goto repeat;
247 		}
248 		i++;
249 	}
250 
251 	*index = i;
252 	return NF_ACCEPT;
253 }
254 
255 static struct nf_hook_entries *nf_hook_entries_head(const struct net *net, u8 pf, u8 hooknum)
256 {
257 	switch (pf) {
258 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
259 	case NFPROTO_BRIDGE:
260 		return rcu_dereference(net->nf.hooks_bridge[hooknum]);
261 #endif
262 	case NFPROTO_IPV4:
263 		return rcu_dereference(net->nf.hooks_ipv4[hooknum]);
264 	case NFPROTO_IPV6:
265 		return rcu_dereference(net->nf.hooks_ipv6[hooknum]);
266 	default:
267 		WARN_ON_ONCE(1);
268 		return NULL;
269 	}
270 
271 	return NULL;
272 }
273 
274 static int nf_ip_reroute(struct sk_buff *skb, const struct nf_queue_entry *entry)
275 {
276 #ifdef CONFIG_INET
277 	const struct ip_rt_info *rt_info = nf_queue_entry_reroute(entry);
278 
279 	if (entry->state.hook == NF_INET_LOCAL_OUT) {
280 		const struct iphdr *iph = ip_hdr(skb);
281 
282 		if (!(iph->tos == rt_info->tos &&
283 		      skb->mark == rt_info->mark &&
284 		      iph->daddr == rt_info->daddr &&
285 		      iph->saddr == rt_info->saddr))
286 			return ip_route_me_harder(entry->state.net, entry->state.sk,
287 						  skb, RTN_UNSPEC);
288 	}
289 #endif
290 	return 0;
291 }
292 
293 static int nf_reroute(struct sk_buff *skb, struct nf_queue_entry *entry)
294 {
295 	const struct nf_ipv6_ops *v6ops;
296 	int ret = 0;
297 
298 	switch (entry->state.pf) {
299 	case AF_INET:
300 		ret = nf_ip_reroute(skb, entry);
301 		break;
302 	case AF_INET6:
303 		v6ops = rcu_dereference(nf_ipv6_ops);
304 		if (v6ops)
305 			ret = v6ops->reroute(skb, entry);
306 		break;
307 	}
308 	return ret;
309 }
310 
311 /* caller must hold rcu read-side lock */
312 static void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
313 {
314 	const struct nf_hook_entry *hook_entry;
315 	const struct nf_hook_entries *hooks;
316 	struct sk_buff *skb = entry->skb;
317 	const struct net *net;
318 	unsigned int i;
319 	int err;
320 	u8 pf;
321 
322 	net = entry->state.net;
323 	pf = entry->state.pf;
324 
325 	hooks = nf_hook_entries_head(net, pf, entry->state.hook);
326 
327 	i = entry->hook_index;
328 	if (!hooks || i >= hooks->num_hook_entries) {
329 		kfree_skb_reason(skb, SKB_DROP_REASON_NETFILTER_DROP);
330 		nf_queue_entry_free(entry);
331 		return;
332 	}
333 
334 	hook_entry = &hooks->hooks[i];
335 
336 	/* Continue traversal iff userspace said ok... */
337 	if (verdict == NF_REPEAT)
338 		verdict = nf_hook_entry_hookfn(hook_entry, skb, &entry->state);
339 
340 	if (verdict == NF_ACCEPT) {
341 		if (nf_reroute(skb, entry) < 0)
342 			verdict = NF_DROP;
343 	}
344 
345 	if (verdict == NF_ACCEPT) {
346 next_hook:
347 		++i;
348 		verdict = nf_iterate(skb, &entry->state, hooks, &i);
349 	}
350 
351 	switch (verdict & NF_VERDICT_MASK) {
352 	case NF_ACCEPT:
353 	case NF_STOP:
354 		local_bh_disable();
355 		entry->state.okfn(entry->state.net, entry->state.sk, skb);
356 		local_bh_enable();
357 		break;
358 	case NF_QUEUE:
359 		err = nf_queue(skb, &entry->state, i, verdict);
360 		if (err == 1)
361 			goto next_hook;
362 		break;
363 	case NF_STOLEN:
364 		break;
365 	default:
366 		kfree_skb(skb);
367 	}
368 
369 	nf_queue_entry_free(entry);
370 }
371 
372 static void nfqnl_reinject(struct nf_queue_entry *entry, unsigned int verdict)
373 {
374 	const struct nf_ct_hook *ct_hook;
375 
376 	if (verdict == NF_ACCEPT ||
377 	    verdict == NF_REPEAT ||
378 	    verdict == NF_STOP) {
379 		unsigned int ct_verdict = verdict;
380 
381 		rcu_read_lock();
382 		ct_hook = rcu_dereference(nf_ct_hook);
383 		if (ct_hook)
384 			ct_verdict = ct_hook->update(entry->state.net, entry->skb);
385 		rcu_read_unlock();
386 
387 		switch (ct_verdict & NF_VERDICT_MASK) {
388 		case NF_ACCEPT:
389 			/* follow userspace verdict, could be REPEAT */
390 			break;
391 		case NF_STOLEN:
392 			nf_queue_entry_free(entry);
393 			return;
394 		default:
395 			verdict = ct_verdict & NF_VERDICT_MASK;
396 			break;
397 		}
398 	}
399 	nf_reinject(entry, verdict);
400 }
401 
402 static void
403 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
404 {
405 	struct nf_queue_entry *entry, *next;
406 
407 	spin_lock_bh(&queue->lock);
408 	list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
409 		if (!cmpfn || cmpfn(entry, data)) {
410 			list_del(&entry->list);
411 			queue->queue_total--;
412 			nfqnl_reinject(entry, NF_DROP);
413 		}
414 	}
415 	spin_unlock_bh(&queue->lock);
416 }
417 
418 static int
419 nfqnl_put_packet_info(struct sk_buff *nlskb, struct sk_buff *packet,
420 		      bool csum_verify)
421 {
422 	__u32 flags = 0;
423 
424 	if (packet->ip_summed == CHECKSUM_PARTIAL)
425 		flags = NFQA_SKB_CSUMNOTREADY;
426 	else if (csum_verify)
427 		flags = NFQA_SKB_CSUM_NOTVERIFIED;
428 
429 	if (skb_is_gso(packet))
430 		flags |= NFQA_SKB_GSO;
431 
432 	return flags ? nla_put_be32(nlskb, NFQA_SKB_INFO, htonl(flags)) : 0;
433 }
434 
435 static int nfqnl_put_sk_uidgid(struct sk_buff *skb, struct sock *sk)
436 {
437 	const struct cred *cred;
438 
439 	if (!sk_fullsock(sk))
440 		return 0;
441 
442 	read_lock_bh(&sk->sk_callback_lock);
443 	if (sk->sk_socket && sk->sk_socket->file) {
444 		cred = sk->sk_socket->file->f_cred;
445 		if (nla_put_be32(skb, NFQA_UID,
446 		    htonl(from_kuid_munged(&init_user_ns, cred->fsuid))))
447 			goto nla_put_failure;
448 		if (nla_put_be32(skb, NFQA_GID,
449 		    htonl(from_kgid_munged(&init_user_ns, cred->fsgid))))
450 			goto nla_put_failure;
451 	}
452 	read_unlock_bh(&sk->sk_callback_lock);
453 	return 0;
454 
455 nla_put_failure:
456 	read_unlock_bh(&sk->sk_callback_lock);
457 	return -1;
458 }
459 
460 static int nfqnl_put_sk_classid(struct sk_buff *skb, struct sock *sk)
461 {
462 #if IS_ENABLED(CONFIG_CGROUP_NET_CLASSID)
463 	if (sk && sk_fullsock(sk)) {
464 		u32 classid = sock_cgroup_classid(&sk->sk_cgrp_data);
465 
466 		if (classid && nla_put_be32(skb, NFQA_CGROUP_CLASSID, htonl(classid)))
467 			return -1;
468 	}
469 #endif
470 	return 0;
471 }
472 
473 static int nfqnl_get_sk_secctx(struct sk_buff *skb, struct lsm_context *ctx)
474 {
475 	int seclen = 0;
476 #if IS_ENABLED(CONFIG_NETWORK_SECMARK)
477 
478 	if (!skb || !sk_fullsock(skb->sk))
479 		return 0;
480 
481 	read_lock_bh(&skb->sk->sk_callback_lock);
482 
483 	if (skb->secmark)
484 		seclen = security_secid_to_secctx(skb->secmark, ctx);
485 	read_unlock_bh(&skb->sk->sk_callback_lock);
486 #endif
487 	return seclen;
488 }
489 
490 static u32 nfqnl_get_bridge_size(struct nf_queue_entry *entry)
491 {
492 	struct sk_buff *entskb = entry->skb;
493 	u32 nlalen = 0;
494 
495 	if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
496 		return 0;
497 
498 	if (skb_vlan_tag_present(entskb))
499 		nlalen += nla_total_size(nla_total_size(sizeof(__be16)) +
500 					 nla_total_size(sizeof(__be16)));
501 
502 	if (entskb->network_header > entskb->mac_header)
503 		nlalen += nla_total_size((entskb->network_header -
504 					  entskb->mac_header));
505 
506 	return nlalen;
507 }
508 
509 static int nfqnl_put_bridge(struct nf_queue_entry *entry, struct sk_buff *skb)
510 {
511 	struct sk_buff *entskb = entry->skb;
512 
513 	if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
514 		return 0;
515 
516 	if (skb_vlan_tag_present(entskb)) {
517 		struct nlattr *nest;
518 
519 		nest = nla_nest_start(skb, NFQA_VLAN);
520 		if (!nest)
521 			goto nla_put_failure;
522 
523 		if (nla_put_be16(skb, NFQA_VLAN_TCI, htons(entskb->vlan_tci)) ||
524 		    nla_put_be16(skb, NFQA_VLAN_PROTO, entskb->vlan_proto))
525 			goto nla_put_failure;
526 
527 		nla_nest_end(skb, nest);
528 	}
529 
530 	if (entskb->mac_header < entskb->network_header) {
531 		int len = (int)(entskb->network_header - entskb->mac_header);
532 
533 		if (nla_put(skb, NFQA_L2HDR, len, skb_mac_header(entskb)))
534 			goto nla_put_failure;
535 	}
536 
537 	return 0;
538 
539 nla_put_failure:
540 	return -1;
541 }
542 
543 static int nf_queue_checksum_help(struct sk_buff *entskb)
544 {
545 	if (skb_csum_is_sctp(entskb))
546 		return skb_crc32c_csum_help(entskb);
547 
548 	return skb_checksum_help(entskb);
549 }
550 
551 static struct sk_buff *
552 nfqnl_build_packet_message(struct net *net, struct nfqnl_instance *queue,
553 			   struct nf_queue_entry *entry,
554 			   __be32 **packet_id_ptr)
555 {
556 	size_t size;
557 	size_t data_len = 0, cap_len = 0;
558 	unsigned int hlen = 0;
559 	struct sk_buff *skb;
560 	struct nlattr *nla;
561 	struct nfqnl_msg_packet_hdr *pmsg;
562 	struct nlmsghdr *nlh;
563 	struct sk_buff *entskb = entry->skb;
564 	struct net_device *indev;
565 	struct net_device *outdev;
566 	struct nf_conn *ct = NULL;
567 	enum ip_conntrack_info ctinfo = 0;
568 	const struct nfnl_ct_hook *nfnl_ct;
569 	bool csum_verify;
570 	struct lsm_context ctx = { NULL, 0, 0 };
571 	int seclen = 0;
572 	ktime_t tstamp;
573 
574 	size = nlmsg_total_size(sizeof(struct nfgenmsg))
575 		+ nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
576 		+ nla_total_size(sizeof(u_int32_t))	/* ifindex */
577 		+ nla_total_size(sizeof(u_int32_t))	/* ifindex */
578 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
579 		+ nla_total_size(sizeof(u_int32_t))	/* ifindex */
580 		+ nla_total_size(sizeof(u_int32_t))	/* ifindex */
581 #endif
582 		+ nla_total_size(sizeof(u_int32_t))	/* mark */
583 		+ nla_total_size(sizeof(u_int32_t))	/* priority */
584 		+ nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
585 		+ nla_total_size(sizeof(u_int32_t))	/* skbinfo */
586 #if IS_ENABLED(CONFIG_CGROUP_NET_CLASSID)
587 		+ nla_total_size(sizeof(u_int32_t))	/* classid */
588 #endif
589 		+ nla_total_size(sizeof(u_int32_t));	/* cap_len */
590 
591 	tstamp = skb_tstamp_cond(entskb, false);
592 	if (tstamp)
593 		size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
594 
595 	size += nfqnl_get_bridge_size(entry);
596 
597 	if (entry->state.hook <= NF_INET_FORWARD ||
598 	   (entry->state.hook == NF_INET_POST_ROUTING && entskb->sk == NULL))
599 		csum_verify = !skb_csum_unnecessary(entskb);
600 	else
601 		csum_verify = false;
602 
603 	outdev = entry->state.out;
604 
605 	switch ((enum nfqnl_config_mode)READ_ONCE(queue->copy_mode)) {
606 	case NFQNL_COPY_META:
607 	case NFQNL_COPY_NONE:
608 		break;
609 
610 	case NFQNL_COPY_PACKET:
611 		if (!(queue->flags & NFQA_CFG_F_GSO) &&
612 		    entskb->ip_summed == CHECKSUM_PARTIAL &&
613 		    nf_queue_checksum_help(entskb))
614 			return NULL;
615 
616 		data_len = READ_ONCE(queue->copy_range);
617 		if (data_len > entskb->len)
618 			data_len = entskb->len;
619 
620 		hlen = skb_zerocopy_headlen(entskb);
621 		hlen = min_t(unsigned int, hlen, data_len);
622 		size += sizeof(struct nlattr) + hlen;
623 		cap_len = entskb->len;
624 		break;
625 	}
626 
627 	nfnl_ct = rcu_dereference(nfnl_ct_hook);
628 
629 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
630 	if (queue->flags & NFQA_CFG_F_CONNTRACK) {
631 		if (nfnl_ct != NULL) {
632 			ct = nf_ct_get(entskb, &ctinfo);
633 			if (ct != NULL)
634 				size += nfnl_ct->build_size(ct);
635 		}
636 	}
637 #endif
638 
639 	if (queue->flags & NFQA_CFG_F_UID_GID) {
640 		size += (nla_total_size(sizeof(u_int32_t))	/* uid */
641 			+ nla_total_size(sizeof(u_int32_t)));	/* gid */
642 	}
643 
644 	if ((queue->flags & NFQA_CFG_F_SECCTX) && entskb->sk) {
645 		seclen = nfqnl_get_sk_secctx(entskb, &ctx);
646 		if (seclen < 0)
647 			return NULL;
648 		if (seclen)
649 			size += nla_total_size(seclen);
650 	}
651 
652 	skb = alloc_skb(size, GFP_ATOMIC);
653 	if (!skb) {
654 		skb_tx_error(entskb);
655 		goto nlmsg_failure;
656 	}
657 
658 	nlh = nfnl_msg_put(skb, 0, 0,
659 			   nfnl_msg_type(NFNL_SUBSYS_QUEUE, NFQNL_MSG_PACKET),
660 			   0, entry->state.pf, NFNETLINK_V0,
661 			   htons(queue->queue_num));
662 	if (!nlh) {
663 		skb_tx_error(entskb);
664 		kfree_skb(skb);
665 		goto nlmsg_failure;
666 	}
667 
668 	nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
669 	pmsg = nla_data(nla);
670 	pmsg->hw_protocol	= entskb->protocol;
671 	pmsg->hook		= entry->state.hook;
672 	*packet_id_ptr		= &pmsg->packet_id;
673 
674 	indev = entry->state.in;
675 	if (indev) {
676 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
677 		if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
678 			goto nla_put_failure;
679 #else
680 		if (entry->state.pf == PF_BRIDGE) {
681 			/* Case 1: indev is physical input device, we need to
682 			 * look for bridge group (when called from
683 			 * netfilter_bridge) */
684 			if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
685 					 htonl(indev->ifindex)) ||
686 			/* this is the bridge group "brX" */
687 			/* rcu_read_lock()ed by __nf_queue */
688 			    nla_put_be32(skb, NFQA_IFINDEX_INDEV,
689 					 htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
690 				goto nla_put_failure;
691 		} else {
692 			int physinif;
693 
694 			/* Case 2: indev is bridge group, we need to look for
695 			 * physical device (when called from ipv4) */
696 			if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
697 					 htonl(indev->ifindex)))
698 				goto nla_put_failure;
699 
700 			physinif = nf_bridge_get_physinif(entskb);
701 			if (physinif &&
702 			    nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
703 					 htonl(physinif)))
704 				goto nla_put_failure;
705 		}
706 #endif
707 	}
708 
709 	if (outdev) {
710 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
711 		if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
712 			goto nla_put_failure;
713 #else
714 		if (entry->state.pf == PF_BRIDGE) {
715 			/* Case 1: outdev is physical output device, we need to
716 			 * look for bridge group (when called from
717 			 * netfilter_bridge) */
718 			if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
719 					 htonl(outdev->ifindex)) ||
720 			/* this is the bridge group "brX" */
721 			/* rcu_read_lock()ed by __nf_queue */
722 			    nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
723 					 htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
724 				goto nla_put_failure;
725 		} else {
726 			int physoutif;
727 
728 			/* Case 2: outdev is bridge group, we need to look for
729 			 * physical output device (when called from ipv4) */
730 			if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
731 					 htonl(outdev->ifindex)))
732 				goto nla_put_failure;
733 
734 			physoutif = nf_bridge_get_physoutif(entskb);
735 			if (physoutif &&
736 			    nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
737 					 htonl(physoutif)))
738 				goto nla_put_failure;
739 		}
740 #endif
741 	}
742 
743 	if (entskb->mark &&
744 	    nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
745 		goto nla_put_failure;
746 
747 	if (entskb->priority &&
748 	    nla_put_be32(skb, NFQA_PRIORITY, htonl(entskb->priority)))
749 		goto nla_put_failure;
750 
751 	if (indev && entskb->dev &&
752 	    skb_mac_header_was_set(entskb) &&
753 	    skb_mac_header_len(entskb) != 0) {
754 		struct nfqnl_msg_packet_hw phw;
755 		int len;
756 
757 		memset(&phw, 0, sizeof(phw));
758 		len = dev_parse_header(entskb, phw.hw_addr);
759 		if (len) {
760 			phw.hw_addrlen = htons(len);
761 			if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
762 				goto nla_put_failure;
763 		}
764 	}
765 
766 	if (nfqnl_put_bridge(entry, skb) < 0)
767 		goto nla_put_failure;
768 
769 	if (entry->state.hook <= NF_INET_FORWARD && tstamp) {
770 		struct nfqnl_msg_packet_timestamp ts;
771 		struct timespec64 kts = ktime_to_timespec64(tstamp);
772 
773 		ts.sec = cpu_to_be64(kts.tv_sec);
774 		ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
775 
776 		if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
777 			goto nla_put_failure;
778 	}
779 
780 	if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
781 	    nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
782 		goto nla_put_failure;
783 
784 	if (nfqnl_put_sk_classid(skb, entskb->sk) < 0)
785 		goto nla_put_failure;
786 
787 	if (seclen > 0 && nla_put(skb, NFQA_SECCTX, ctx.len, ctx.context))
788 		goto nla_put_failure;
789 
790 	if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
791 		goto nla_put_failure;
792 
793 	if (cap_len > data_len &&
794 	    nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
795 		goto nla_put_failure;
796 
797 	if (nfqnl_put_packet_info(skb, entskb, csum_verify))
798 		goto nla_put_failure;
799 
800 	if (data_len) {
801 		struct nlattr *nla;
802 
803 		if (skb_tailroom(skb) < sizeof(*nla) + hlen)
804 			goto nla_put_failure;
805 
806 		nla = skb_put(skb, sizeof(*nla));
807 		nla->nla_type = NFQA_PAYLOAD;
808 		nla->nla_len = nla_attr_size(data_len);
809 
810 		if (skb_zerocopy(skb, entskb, data_len, hlen))
811 			goto nla_put_failure;
812 	}
813 
814 	nlh->nlmsg_len = skb->len;
815 	if (seclen >= 0)
816 		security_release_secctx(&ctx);
817 	return skb;
818 
819 nla_put_failure:
820 	skb_tx_error(entskb);
821 	kfree_skb(skb);
822 	net_err_ratelimited("nf_queue: error creating packet message\n");
823 nlmsg_failure:
824 	if (seclen >= 0)
825 		security_release_secctx(&ctx);
826 	return NULL;
827 }
828 
829 static bool nf_ct_drop_unconfirmed(const struct nf_queue_entry *entry)
830 {
831 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
832 	static const unsigned long flags = IPS_CONFIRMED | IPS_DYING;
833 	struct nf_conn *ct = (void *)skb_nfct(entry->skb);
834 	unsigned long status;
835 	unsigned int use;
836 
837 	if (!ct)
838 		return false;
839 
840 	status = READ_ONCE(ct->status);
841 	if ((status & flags) == IPS_DYING)
842 		return true;
843 
844 	if (status & IPS_CONFIRMED)
845 		return false;
846 
847 	/* in some cases skb_clone() can occur after initial conntrack
848 	 * pickup, but conntrack assumes exclusive skb->_nfct ownership for
849 	 * unconfirmed entries.
850 	 *
851 	 * This happens for br_netfilter and with ip multicast routing.
852 	 * We can't be solved with serialization here because one clone could
853 	 * have been queued for local delivery.
854 	 */
855 	use = refcount_read(&ct->ct_general.use);
856 	if (likely(use == 1))
857 		return false;
858 
859 	/* Can't decrement further? Exclusive ownership. */
860 	if (!refcount_dec_not_one(&ct->ct_general.use))
861 		return false;
862 
863 	skb_set_nfct(entry->skb, 0);
864 	/* No nf_ct_put(): we already decremented .use and it cannot
865 	 * drop down to 0.
866 	 */
867 	return true;
868 #endif
869 	return false;
870 }
871 
872 static int
873 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
874 			struct nf_queue_entry *entry)
875 {
876 	struct sk_buff *nskb;
877 	int err = -ENOBUFS;
878 	__be32 *packet_id_ptr;
879 	int failopen = 0;
880 
881 	nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
882 	if (nskb == NULL) {
883 		err = -ENOMEM;
884 		goto err_out;
885 	}
886 	spin_lock_bh(&queue->lock);
887 
888 	if (nf_ct_drop_unconfirmed(entry))
889 		goto err_out_free_nskb;
890 
891 	if (queue->queue_total >= queue->queue_maxlen) {
892 		if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
893 			failopen = 1;
894 			err = 0;
895 		} else {
896 			queue->queue_dropped++;
897 			net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
898 					     queue->queue_total);
899 		}
900 		goto err_out_free_nskb;
901 	}
902 	entry->id = ++queue->id_sequence;
903 	*packet_id_ptr = htonl(entry->id);
904 
905 	/* nfnetlink_unicast will either free the nskb or add it to a socket */
906 	err = nfnetlink_unicast(nskb, net, queue->peer_portid);
907 	if (err < 0) {
908 		if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
909 			failopen = 1;
910 			err = 0;
911 		} else {
912 			queue->queue_user_dropped++;
913 		}
914 		goto err_out_unlock;
915 	}
916 
917 	__enqueue_entry(queue, entry);
918 
919 	spin_unlock_bh(&queue->lock);
920 	return 0;
921 
922 err_out_free_nskb:
923 	kfree_skb(nskb);
924 err_out_unlock:
925 	spin_unlock_bh(&queue->lock);
926 	if (failopen)
927 		nfqnl_reinject(entry, NF_ACCEPT);
928 err_out:
929 	return err;
930 }
931 
932 static struct nf_queue_entry *
933 nf_queue_entry_dup(struct nf_queue_entry *e)
934 {
935 	struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
936 
937 	if (!entry)
938 		return NULL;
939 
940 	if (nf_queue_entry_get_refs(entry))
941 		return entry;
942 
943 	kfree(entry);
944 	return NULL;
945 }
946 
947 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
948 /* When called from bridge netfilter, skb->data must point to MAC header
949  * before calling skb_gso_segment(). Else, original MAC header is lost
950  * and segmented skbs will be sent to wrong destination.
951  */
952 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
953 {
954 	if (nf_bridge_info_get(skb))
955 		__skb_push(skb, skb->network_header - skb->mac_header);
956 }
957 
958 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
959 {
960 	if (nf_bridge_info_get(skb))
961 		__skb_pull(skb, skb->network_header - skb->mac_header);
962 }
963 #else
964 #define nf_bridge_adjust_skb_data(s) do {} while (0)
965 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
966 #endif
967 
968 static int
969 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
970 			   struct sk_buff *skb, struct nf_queue_entry *entry)
971 {
972 	int ret = -ENOMEM;
973 	struct nf_queue_entry *entry_seg;
974 
975 	nf_bridge_adjust_segmented_data(skb);
976 
977 	if (skb->next == NULL) { /* last packet, no need to copy entry */
978 		struct sk_buff *gso_skb = entry->skb;
979 		entry->skb = skb;
980 		ret = __nfqnl_enqueue_packet(net, queue, entry);
981 		if (ret)
982 			entry->skb = gso_skb;
983 		return ret;
984 	}
985 
986 	skb_mark_not_on_list(skb);
987 
988 	entry_seg = nf_queue_entry_dup(entry);
989 	if (entry_seg) {
990 		entry_seg->skb = skb;
991 		ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
992 		if (ret)
993 			nf_queue_entry_free(entry_seg);
994 	}
995 	return ret;
996 }
997 
998 static int
999 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
1000 {
1001 	unsigned int queued;
1002 	struct nfqnl_instance *queue;
1003 	struct sk_buff *skb, *segs, *nskb;
1004 	int err = -ENOBUFS;
1005 	struct net *net = entry->state.net;
1006 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1007 
1008 	/* rcu_read_lock()ed by nf_hook_thresh */
1009 	queue = instance_lookup(q, queuenum);
1010 	if (!queue)
1011 		return -ESRCH;
1012 
1013 	if (queue->copy_mode == NFQNL_COPY_NONE)
1014 		return -EINVAL;
1015 
1016 	skb = entry->skb;
1017 
1018 	switch (entry->state.pf) {
1019 	case NFPROTO_IPV4:
1020 		skb->protocol = htons(ETH_P_IP);
1021 		break;
1022 	case NFPROTO_IPV6:
1023 		skb->protocol = htons(ETH_P_IPV6);
1024 		break;
1025 	}
1026 
1027 	if (!skb_is_gso(skb) || ((queue->flags & NFQA_CFG_F_GSO) && !skb_is_gso_sctp(skb)))
1028 		return __nfqnl_enqueue_packet(net, queue, entry);
1029 
1030 	nf_bridge_adjust_skb_data(skb);
1031 	segs = skb_gso_segment(skb, 0);
1032 	/* Does not use PTR_ERR to limit the number of error codes that can be
1033 	 * returned by nf_queue.  For instance, callers rely on -ESRCH to
1034 	 * mean 'ignore this hook'.
1035 	 */
1036 	if (IS_ERR_OR_NULL(segs))
1037 		goto out_err;
1038 	queued = 0;
1039 	err = 0;
1040 	skb_list_walk_safe(segs, segs, nskb) {
1041 		if (err == 0)
1042 			err = __nfqnl_enqueue_packet_gso(net, queue,
1043 							segs, entry);
1044 		if (err == 0)
1045 			queued++;
1046 		else
1047 			kfree_skb(segs);
1048 	}
1049 
1050 	if (queued) {
1051 		if (err) /* some segments are already queued */
1052 			nf_queue_entry_free(entry);
1053 		kfree_skb(skb);
1054 		return 0;
1055 	}
1056  out_err:
1057 	nf_bridge_adjust_segmented_data(skb);
1058 	return err;
1059 }
1060 
1061 static int
1062 nfqnl_mangle(void *data, unsigned int data_len, struct nf_queue_entry *e, int diff)
1063 {
1064 	struct sk_buff *nskb;
1065 
1066 	if (diff < 0) {
1067 		unsigned int min_len = skb_transport_offset(e->skb);
1068 
1069 		if (data_len < min_len)
1070 			return -EINVAL;
1071 
1072 		if (pskb_trim(e->skb, data_len))
1073 			return -ENOMEM;
1074 	} else if (diff > 0) {
1075 		if (data_len > 0xFFFF)
1076 			return -EINVAL;
1077 		if (diff > skb_tailroom(e->skb)) {
1078 			nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
1079 					       diff, GFP_ATOMIC);
1080 			if (!nskb)
1081 				return -ENOMEM;
1082 			kfree_skb(e->skb);
1083 			e->skb = nskb;
1084 		}
1085 		skb_put(e->skb, diff);
1086 	}
1087 	if (skb_ensure_writable(e->skb, data_len))
1088 		return -ENOMEM;
1089 	skb_copy_to_linear_data(e->skb, data, data_len);
1090 	e->skb->ip_summed = CHECKSUM_NONE;
1091 	return 0;
1092 }
1093 
1094 static int
1095 nfqnl_set_mode(struct nfqnl_instance *queue,
1096 	       unsigned char mode, unsigned int range)
1097 {
1098 	int status = 0;
1099 
1100 	spin_lock_bh(&queue->lock);
1101 	switch (mode) {
1102 	case NFQNL_COPY_NONE:
1103 	case NFQNL_COPY_META:
1104 		queue->copy_mode = mode;
1105 		queue->copy_range = 0;
1106 		break;
1107 
1108 	case NFQNL_COPY_PACKET:
1109 		queue->copy_mode = mode;
1110 		if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
1111 			queue->copy_range = NFQNL_MAX_COPY_RANGE;
1112 		else
1113 			queue->copy_range = range;
1114 		break;
1115 
1116 	default:
1117 		status = -EINVAL;
1118 
1119 	}
1120 	spin_unlock_bh(&queue->lock);
1121 
1122 	return status;
1123 }
1124 
1125 static int
1126 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
1127 {
1128 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
1129 	int physinif, physoutif;
1130 
1131 	physinif = nf_bridge_get_physinif(entry->skb);
1132 	physoutif = nf_bridge_get_physoutif(entry->skb);
1133 
1134 	if (physinif == ifindex || physoutif == ifindex)
1135 		return 1;
1136 #endif
1137 	if (entry->state.in)
1138 		if (entry->state.in->ifindex == ifindex)
1139 			return 1;
1140 	if (entry->state.out)
1141 		if (entry->state.out->ifindex == ifindex)
1142 			return 1;
1143 
1144 	return 0;
1145 }
1146 
1147 /* drop all packets with either indev or outdev == ifindex from all queue
1148  * instances */
1149 static void
1150 nfqnl_dev_drop(struct net *net, int ifindex)
1151 {
1152 	int i;
1153 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1154 
1155 	rcu_read_lock();
1156 
1157 	for (i = 0; i < INSTANCE_BUCKETS; i++) {
1158 		struct nfqnl_instance *inst;
1159 		struct hlist_head *head = &q->instance_table[i];
1160 
1161 		hlist_for_each_entry_rcu(inst, head, hlist)
1162 			nfqnl_flush(inst, dev_cmp, ifindex);
1163 	}
1164 
1165 	rcu_read_unlock();
1166 }
1167 
1168 static int
1169 nfqnl_rcv_dev_event(struct notifier_block *this,
1170 		    unsigned long event, void *ptr)
1171 {
1172 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1173 
1174 	/* Drop any packets associated with the downed device */
1175 	if (event == NETDEV_DOWN)
1176 		nfqnl_dev_drop(dev_net(dev), dev->ifindex);
1177 	return NOTIFY_DONE;
1178 }
1179 
1180 static struct notifier_block nfqnl_dev_notifier = {
1181 	.notifier_call	= nfqnl_rcv_dev_event,
1182 };
1183 
1184 static void nfqnl_nf_hook_drop(struct net *net)
1185 {
1186 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1187 	int i;
1188 
1189 	/* This function is also called on net namespace error unwind,
1190 	 * when pernet_ops->init() failed and ->exit() functions of the
1191 	 * previous pernet_ops gets called.
1192 	 *
1193 	 * This may result in a call to nfqnl_nf_hook_drop() before
1194 	 * struct nfnl_queue_net was allocated.
1195 	 */
1196 	if (!q)
1197 		return;
1198 
1199 	for (i = 0; i < INSTANCE_BUCKETS; i++) {
1200 		struct nfqnl_instance *inst;
1201 		struct hlist_head *head = &q->instance_table[i];
1202 
1203 		hlist_for_each_entry_rcu(inst, head, hlist)
1204 			nfqnl_flush(inst, NULL, 0);
1205 	}
1206 }
1207 
1208 static int
1209 nfqnl_rcv_nl_event(struct notifier_block *this,
1210 		   unsigned long event, void *ptr)
1211 {
1212 	struct netlink_notify *n = ptr;
1213 	struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
1214 
1215 	if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
1216 		int i;
1217 
1218 		/* destroy all instances for this portid */
1219 		spin_lock(&q->instances_lock);
1220 		for (i = 0; i < INSTANCE_BUCKETS; i++) {
1221 			struct hlist_node *t2;
1222 			struct nfqnl_instance *inst;
1223 			struct hlist_head *head = &q->instance_table[i];
1224 
1225 			hlist_for_each_entry_safe(inst, t2, head, hlist) {
1226 				if (n->portid == inst->peer_portid)
1227 					__instance_destroy(inst);
1228 			}
1229 		}
1230 		spin_unlock(&q->instances_lock);
1231 	}
1232 	return NOTIFY_DONE;
1233 }
1234 
1235 static struct notifier_block nfqnl_rtnl_notifier = {
1236 	.notifier_call	= nfqnl_rcv_nl_event,
1237 };
1238 
1239 static const struct nla_policy nfqa_vlan_policy[NFQA_VLAN_MAX + 1] = {
1240 	[NFQA_VLAN_TCI]		= { .type = NLA_U16},
1241 	[NFQA_VLAN_PROTO]	= { .type = NLA_U16},
1242 };
1243 
1244 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
1245 	[NFQA_VERDICT_HDR]	= { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1246 	[NFQA_MARK]		= { .type = NLA_U32 },
1247 	[NFQA_PAYLOAD]		= { .type = NLA_UNSPEC },
1248 	[NFQA_CT]		= { .type = NLA_UNSPEC },
1249 	[NFQA_EXP]		= { .type = NLA_UNSPEC },
1250 	[NFQA_VLAN]		= { .type = NLA_NESTED },
1251 	[NFQA_PRIORITY]		= { .type = NLA_U32 },
1252 };
1253 
1254 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
1255 	[NFQA_VERDICT_HDR]	= { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1256 	[NFQA_MARK]		= { .type = NLA_U32 },
1257 	[NFQA_PRIORITY]		= { .type = NLA_U32 },
1258 };
1259 
1260 static struct nfqnl_instance *
1261 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
1262 {
1263 	struct nfqnl_instance *queue;
1264 
1265 	queue = instance_lookup(q, queue_num);
1266 	if (!queue)
1267 		return ERR_PTR(-ENODEV);
1268 
1269 	if (queue->peer_portid != nlportid)
1270 		return ERR_PTR(-EPERM);
1271 
1272 	return queue;
1273 }
1274 
1275 static struct nfqnl_msg_verdict_hdr*
1276 verdicthdr_get(const struct nlattr * const nfqa[])
1277 {
1278 	struct nfqnl_msg_verdict_hdr *vhdr;
1279 	unsigned int verdict;
1280 
1281 	if (!nfqa[NFQA_VERDICT_HDR])
1282 		return NULL;
1283 
1284 	vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
1285 	verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
1286 	if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
1287 		return NULL;
1288 	return vhdr;
1289 }
1290 
1291 static int nfq_id_after(unsigned int id, unsigned int max)
1292 {
1293 	return (int)(id - max) > 0;
1294 }
1295 
1296 static int nfqnl_recv_verdict_batch(struct sk_buff *skb,
1297 				    const struct nfnl_info *info,
1298 				    const struct nlattr * const nfqa[])
1299 {
1300 	struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1301 	u16 queue_num = ntohs(info->nfmsg->res_id);
1302 	struct nf_queue_entry *entry, *tmp;
1303 	struct nfqnl_msg_verdict_hdr *vhdr;
1304 	struct nfqnl_instance *queue;
1305 	unsigned int verdict, maxid;
1306 	LIST_HEAD(batch_list);
1307 
1308 	queue = verdict_instance_lookup(q, queue_num,
1309 					NETLINK_CB(skb).portid);
1310 	if (IS_ERR(queue))
1311 		return PTR_ERR(queue);
1312 
1313 	vhdr = verdicthdr_get(nfqa);
1314 	if (!vhdr)
1315 		return -EINVAL;
1316 
1317 	verdict = ntohl(vhdr->verdict);
1318 	maxid = ntohl(vhdr->id);
1319 
1320 	spin_lock_bh(&queue->lock);
1321 
1322 	list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
1323 		if (nfq_id_after(entry->id, maxid))
1324 			break;
1325 		__dequeue_entry(queue, entry);
1326 		list_add_tail(&entry->list, &batch_list);
1327 	}
1328 
1329 	spin_unlock_bh(&queue->lock);
1330 
1331 	if (list_empty(&batch_list))
1332 		return -ENOENT;
1333 
1334 	list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1335 		if (nfqa[NFQA_MARK])
1336 			entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1337 
1338 		if (nfqa[NFQA_PRIORITY])
1339 			entry->skb->priority = ntohl(nla_get_be32(nfqa[NFQA_PRIORITY]));
1340 
1341 		nfqnl_reinject(entry, verdict);
1342 	}
1343 	return 0;
1344 }
1345 
1346 static struct nf_conn *nfqnl_ct_parse(const struct nfnl_ct_hook *nfnl_ct,
1347 				      const struct nlmsghdr *nlh,
1348 				      const struct nlattr * const nfqa[],
1349 				      struct nf_queue_entry *entry,
1350 				      enum ip_conntrack_info *ctinfo)
1351 {
1352 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1353 	struct nf_conn *ct;
1354 
1355 	ct = nf_ct_get(entry->skb, ctinfo);
1356 	if (ct == NULL)
1357 		return NULL;
1358 
1359 	if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1360 		return NULL;
1361 
1362 	if (nfqa[NFQA_EXP])
1363 		nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1364 				      NETLINK_CB(entry->skb).portid,
1365 				      nlmsg_report(nlh));
1366 	return ct;
1367 #else
1368 	return NULL;
1369 #endif
1370 }
1371 
1372 static int nfqa_parse_bridge(struct nf_queue_entry *entry,
1373 			     const struct nlattr * const nfqa[])
1374 {
1375 	if (nfqa[NFQA_VLAN]) {
1376 		struct nlattr *tb[NFQA_VLAN_MAX + 1];
1377 		int err;
1378 
1379 		err = nla_parse_nested_deprecated(tb, NFQA_VLAN_MAX,
1380 						  nfqa[NFQA_VLAN],
1381 						  nfqa_vlan_policy, NULL);
1382 		if (err < 0)
1383 			return err;
1384 
1385 		if (!tb[NFQA_VLAN_TCI] || !tb[NFQA_VLAN_PROTO])
1386 			return -EINVAL;
1387 
1388 		__vlan_hwaccel_put_tag(entry->skb,
1389 			nla_get_be16(tb[NFQA_VLAN_PROTO]),
1390 			ntohs(nla_get_be16(tb[NFQA_VLAN_TCI])));
1391 	}
1392 
1393 	if (nfqa[NFQA_L2HDR]) {
1394 		int mac_header_len = entry->skb->network_header -
1395 			entry->skb->mac_header;
1396 
1397 		if (mac_header_len != nla_len(nfqa[NFQA_L2HDR]))
1398 			return -EINVAL;
1399 		else if (mac_header_len > 0)
1400 			memcpy(skb_mac_header(entry->skb),
1401 			       nla_data(nfqa[NFQA_L2HDR]),
1402 			       mac_header_len);
1403 	}
1404 
1405 	return 0;
1406 }
1407 
1408 static int nfqnl_recv_verdict(struct sk_buff *skb, const struct nfnl_info *info,
1409 			      const struct nlattr * const nfqa[])
1410 {
1411 	struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1412 	u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1413 	const struct nfnl_ct_hook *nfnl_ct;
1414 	struct nfqnl_msg_verdict_hdr *vhdr;
1415 	enum ip_conntrack_info ctinfo;
1416 	struct nfqnl_instance *queue;
1417 	struct nf_queue_entry *entry;
1418 	struct nf_conn *ct = NULL;
1419 	unsigned int verdict;
1420 	int err;
1421 
1422 	queue = verdict_instance_lookup(q, queue_num,
1423 					NETLINK_CB(skb).portid);
1424 	if (IS_ERR(queue))
1425 		return PTR_ERR(queue);
1426 
1427 	vhdr = verdicthdr_get(nfqa);
1428 	if (!vhdr)
1429 		return -EINVAL;
1430 
1431 	verdict = ntohl(vhdr->verdict);
1432 
1433 	entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1434 	if (entry == NULL)
1435 		return -ENOENT;
1436 
1437 	/* rcu lock already held from nfnl->call_rcu. */
1438 	nfnl_ct = rcu_dereference(nfnl_ct_hook);
1439 
1440 	if (nfqa[NFQA_CT]) {
1441 		if (nfnl_ct != NULL)
1442 			ct = nfqnl_ct_parse(nfnl_ct, info->nlh, nfqa, entry,
1443 					    &ctinfo);
1444 	}
1445 
1446 	if (entry->state.pf == PF_BRIDGE) {
1447 		err = nfqa_parse_bridge(entry, nfqa);
1448 		if (err < 0)
1449 			return err;
1450 	}
1451 
1452 	if (nfqa[NFQA_PAYLOAD]) {
1453 		u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1454 		int diff = payload_len - entry->skb->len;
1455 
1456 		if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1457 				 payload_len, entry, diff) < 0)
1458 			verdict = NF_DROP;
1459 
1460 		if (ct && diff)
1461 			nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1462 	}
1463 
1464 	if (nfqa[NFQA_MARK])
1465 		entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1466 
1467 	if (nfqa[NFQA_PRIORITY])
1468 		entry->skb->priority = ntohl(nla_get_be32(nfqa[NFQA_PRIORITY]));
1469 
1470 	nfqnl_reinject(entry, verdict);
1471 	return 0;
1472 }
1473 
1474 static int nfqnl_recv_unsupp(struct sk_buff *skb, const struct nfnl_info *info,
1475 			     const struct nlattr * const cda[])
1476 {
1477 	return -ENOTSUPP;
1478 }
1479 
1480 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1481 	[NFQA_CFG_CMD]		= { .len = sizeof(struct nfqnl_msg_config_cmd) },
1482 	[NFQA_CFG_PARAMS]	= { .len = sizeof(struct nfqnl_msg_config_params) },
1483 	[NFQA_CFG_QUEUE_MAXLEN]	= { .type = NLA_U32 },
1484 	[NFQA_CFG_MASK]		= { .type = NLA_U32 },
1485 	[NFQA_CFG_FLAGS]	= { .type = NLA_U32 },
1486 };
1487 
1488 static const struct nf_queue_handler nfqh = {
1489 	.outfn		= nfqnl_enqueue_packet,
1490 	.nf_hook_drop	= nfqnl_nf_hook_drop,
1491 };
1492 
1493 static int nfqnl_recv_config(struct sk_buff *skb, const struct nfnl_info *info,
1494 			     const struct nlattr * const nfqa[])
1495 {
1496 	struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1497 	u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1498 	struct nfqnl_msg_config_cmd *cmd = NULL;
1499 	struct nfqnl_instance *queue;
1500 	__u32 flags = 0, mask = 0;
1501 	int ret = 0;
1502 
1503 	if (nfqa[NFQA_CFG_CMD]) {
1504 		cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1505 
1506 		/* Obsolete commands without queue context */
1507 		switch (cmd->command) {
1508 		case NFQNL_CFG_CMD_PF_BIND: return 0;
1509 		case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1510 		}
1511 	}
1512 
1513 	/* Check if we support these flags in first place, dependencies should
1514 	 * be there too not to break atomicity.
1515 	 */
1516 	if (nfqa[NFQA_CFG_FLAGS]) {
1517 		if (!nfqa[NFQA_CFG_MASK]) {
1518 			/* A mask is needed to specify which flags are being
1519 			 * changed.
1520 			 */
1521 			return -EINVAL;
1522 		}
1523 
1524 		flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1525 		mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1526 
1527 		if (flags >= NFQA_CFG_F_MAX)
1528 			return -EOPNOTSUPP;
1529 
1530 #if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1531 		if (flags & mask & NFQA_CFG_F_SECCTX)
1532 			return -EOPNOTSUPP;
1533 #endif
1534 		if ((flags & mask & NFQA_CFG_F_CONNTRACK) &&
1535 		    !rcu_access_pointer(nfnl_ct_hook)) {
1536 #ifdef CONFIG_MODULES
1537 			nfnl_unlock(NFNL_SUBSYS_QUEUE);
1538 			request_module("ip_conntrack_netlink");
1539 			nfnl_lock(NFNL_SUBSYS_QUEUE);
1540 			if (rcu_access_pointer(nfnl_ct_hook))
1541 				return -EAGAIN;
1542 #endif
1543 			return -EOPNOTSUPP;
1544 		}
1545 	}
1546 
1547 	rcu_read_lock();
1548 	queue = instance_lookup(q, queue_num);
1549 	if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1550 		ret = -EPERM;
1551 		goto err_out_unlock;
1552 	}
1553 
1554 	if (cmd != NULL) {
1555 		switch (cmd->command) {
1556 		case NFQNL_CFG_CMD_BIND:
1557 			if (queue) {
1558 				ret = -EBUSY;
1559 				goto err_out_unlock;
1560 			}
1561 			queue = instance_create(q, queue_num,
1562 						NETLINK_CB(skb).portid);
1563 			if (IS_ERR(queue)) {
1564 				ret = PTR_ERR(queue);
1565 				goto err_out_unlock;
1566 			}
1567 			break;
1568 		case NFQNL_CFG_CMD_UNBIND:
1569 			if (!queue) {
1570 				ret = -ENODEV;
1571 				goto err_out_unlock;
1572 			}
1573 			instance_destroy(q, queue);
1574 			goto err_out_unlock;
1575 		case NFQNL_CFG_CMD_PF_BIND:
1576 		case NFQNL_CFG_CMD_PF_UNBIND:
1577 			break;
1578 		default:
1579 			ret = -ENOTSUPP;
1580 			goto err_out_unlock;
1581 		}
1582 	}
1583 
1584 	if (!queue) {
1585 		ret = -ENODEV;
1586 		goto err_out_unlock;
1587 	}
1588 
1589 	if (nfqa[NFQA_CFG_PARAMS]) {
1590 		struct nfqnl_msg_config_params *params =
1591 			nla_data(nfqa[NFQA_CFG_PARAMS]);
1592 
1593 		nfqnl_set_mode(queue, params->copy_mode,
1594 				ntohl(params->copy_range));
1595 	}
1596 
1597 	if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1598 		__be32 *queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1599 
1600 		spin_lock_bh(&queue->lock);
1601 		queue->queue_maxlen = ntohl(*queue_maxlen);
1602 		spin_unlock_bh(&queue->lock);
1603 	}
1604 
1605 	if (nfqa[NFQA_CFG_FLAGS]) {
1606 		spin_lock_bh(&queue->lock);
1607 		queue->flags &= ~mask;
1608 		queue->flags |= flags & mask;
1609 		spin_unlock_bh(&queue->lock);
1610 	}
1611 
1612 err_out_unlock:
1613 	rcu_read_unlock();
1614 	return ret;
1615 }
1616 
1617 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1618 	[NFQNL_MSG_PACKET]	= {
1619 		.call		= nfqnl_recv_unsupp,
1620 		.type		= NFNL_CB_RCU,
1621 		.attr_count	= NFQA_MAX,
1622 	},
1623 	[NFQNL_MSG_VERDICT]	= {
1624 		.call		= nfqnl_recv_verdict,
1625 		.type		= NFNL_CB_RCU,
1626 		.attr_count	= NFQA_MAX,
1627 		.policy		= nfqa_verdict_policy
1628 	},
1629 	[NFQNL_MSG_CONFIG]	= {
1630 		.call		= nfqnl_recv_config,
1631 		.type		= NFNL_CB_MUTEX,
1632 		.attr_count	= NFQA_CFG_MAX,
1633 		.policy		= nfqa_cfg_policy
1634 	},
1635 	[NFQNL_MSG_VERDICT_BATCH] = {
1636 		.call		= nfqnl_recv_verdict_batch,
1637 		.type		= NFNL_CB_RCU,
1638 		.attr_count	= NFQA_MAX,
1639 		.policy		= nfqa_verdict_batch_policy
1640 	},
1641 };
1642 
1643 static const struct nfnetlink_subsystem nfqnl_subsys = {
1644 	.name		= "nf_queue",
1645 	.subsys_id	= NFNL_SUBSYS_QUEUE,
1646 	.cb_count	= NFQNL_MSG_MAX,
1647 	.cb		= nfqnl_cb,
1648 };
1649 
1650 #ifdef CONFIG_PROC_FS
1651 struct iter_state {
1652 	struct seq_net_private p;
1653 	unsigned int bucket;
1654 };
1655 
1656 static struct hlist_node *get_first(struct seq_file *seq)
1657 {
1658 	struct iter_state *st = seq->private;
1659 	struct net *net;
1660 	struct nfnl_queue_net *q;
1661 
1662 	if (!st)
1663 		return NULL;
1664 
1665 	net = seq_file_net(seq);
1666 	q = nfnl_queue_pernet(net);
1667 	for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1668 		if (!hlist_empty(&q->instance_table[st->bucket]))
1669 			return q->instance_table[st->bucket].first;
1670 	}
1671 	return NULL;
1672 }
1673 
1674 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1675 {
1676 	struct iter_state *st = seq->private;
1677 	struct net *net = seq_file_net(seq);
1678 
1679 	h = h->next;
1680 	while (!h) {
1681 		struct nfnl_queue_net *q;
1682 
1683 		if (++st->bucket >= INSTANCE_BUCKETS)
1684 			return NULL;
1685 
1686 		q = nfnl_queue_pernet(net);
1687 		h = q->instance_table[st->bucket].first;
1688 	}
1689 	return h;
1690 }
1691 
1692 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1693 {
1694 	struct hlist_node *head;
1695 	head = get_first(seq);
1696 
1697 	if (head)
1698 		while (pos && (head = get_next(seq, head)))
1699 			pos--;
1700 	return pos ? NULL : head;
1701 }
1702 
1703 static void *seq_start(struct seq_file *s, loff_t *pos)
1704 	__acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1705 {
1706 	spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1707 	return get_idx(s, *pos);
1708 }
1709 
1710 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1711 {
1712 	(*pos)++;
1713 	return get_next(s, v);
1714 }
1715 
1716 static void seq_stop(struct seq_file *s, void *v)
1717 	__releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1718 {
1719 	spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1720 }
1721 
1722 static int seq_show(struct seq_file *s, void *v)
1723 {
1724 	const struct nfqnl_instance *inst = v;
1725 
1726 	seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1727 		   inst->queue_num,
1728 		   inst->peer_portid, inst->queue_total,
1729 		   inst->copy_mode, inst->copy_range,
1730 		   inst->queue_dropped, inst->queue_user_dropped,
1731 		   inst->id_sequence, 1);
1732 	return 0;
1733 }
1734 
1735 static const struct seq_operations nfqnl_seq_ops = {
1736 	.start	= seq_start,
1737 	.next	= seq_next,
1738 	.stop	= seq_stop,
1739 	.show	= seq_show,
1740 };
1741 #endif /* PROC_FS */
1742 
1743 static int __net_init nfnl_queue_net_init(struct net *net)
1744 {
1745 	unsigned int i;
1746 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1747 
1748 	for (i = 0; i < INSTANCE_BUCKETS; i++)
1749 		INIT_HLIST_HEAD(&q->instance_table[i]);
1750 
1751 	spin_lock_init(&q->instances_lock);
1752 
1753 #ifdef CONFIG_PROC_FS
1754 	if (!proc_create_net("nfnetlink_queue", 0440, net->nf.proc_netfilter,
1755 			&nfqnl_seq_ops, sizeof(struct iter_state)))
1756 		return -ENOMEM;
1757 #endif
1758 	return 0;
1759 }
1760 
1761 static void __net_exit nfnl_queue_net_exit(struct net *net)
1762 {
1763 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1764 	unsigned int i;
1765 
1766 #ifdef CONFIG_PROC_FS
1767 	remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1768 #endif
1769 	for (i = 0; i < INSTANCE_BUCKETS; i++)
1770 		WARN_ON_ONCE(!hlist_empty(&q->instance_table[i]));
1771 }
1772 
1773 static struct pernet_operations nfnl_queue_net_ops = {
1774 	.init		= nfnl_queue_net_init,
1775 	.exit		= nfnl_queue_net_exit,
1776 	.id		= &nfnl_queue_net_id,
1777 	.size		= sizeof(struct nfnl_queue_net),
1778 };
1779 
1780 static int __init nfnetlink_queue_init(void)
1781 {
1782 	int status;
1783 
1784 	status = register_pernet_subsys(&nfnl_queue_net_ops);
1785 	if (status < 0) {
1786 		pr_err("failed to register pernet ops\n");
1787 		goto out;
1788 	}
1789 
1790 	netlink_register_notifier(&nfqnl_rtnl_notifier);
1791 	status = nfnetlink_subsys_register(&nfqnl_subsys);
1792 	if (status < 0) {
1793 		pr_err("failed to create netlink socket\n");
1794 		goto cleanup_netlink_notifier;
1795 	}
1796 
1797 	status = register_netdevice_notifier(&nfqnl_dev_notifier);
1798 	if (status < 0) {
1799 		pr_err("failed to register netdevice notifier\n");
1800 		goto cleanup_netlink_subsys;
1801 	}
1802 
1803 	nf_register_queue_handler(&nfqh);
1804 
1805 	return status;
1806 
1807 cleanup_netlink_subsys:
1808 	nfnetlink_subsys_unregister(&nfqnl_subsys);
1809 cleanup_netlink_notifier:
1810 	netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1811 	unregister_pernet_subsys(&nfnl_queue_net_ops);
1812 out:
1813 	return status;
1814 }
1815 
1816 static void __exit nfnetlink_queue_fini(void)
1817 {
1818 	nf_unregister_queue_handler();
1819 	unregister_netdevice_notifier(&nfqnl_dev_notifier);
1820 	nfnetlink_subsys_unregister(&nfqnl_subsys);
1821 	netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1822 	unregister_pernet_subsys(&nfnl_queue_net_ops);
1823 
1824 	rcu_barrier(); /* Wait for completion of call_rcu()'s */
1825 }
1826 
1827 MODULE_DESCRIPTION("netfilter packet queue handler");
1828 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1829 MODULE_LICENSE("GPL");
1830 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1831 
1832 module_init(nfnetlink_queue_init);
1833 module_exit(nfnetlink_queue_fini);
1834