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 <[email protected]>
7  * (C) 2007 by Patrick McHardy <[email protected]>
8  *
9  * Based on the old ipv4-only ip_queue.c:
10  * (C) 2000-2002 James Morris <[email protected]>
11  * (C) 2003-2005 Netfilter Core Team <[email protected]>
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 u32 nfqnl_get_sk_secctx(struct sk_buff *skb, char **secdata)
474 {
475 	u32 seclen = 0;
476 #if IS_ENABLED(CONFIG_NETWORK_SECMARK)
477 	if (!skb || !sk_fullsock(skb->sk))
478 		return 0;
479 
480 	read_lock_bh(&skb->sk->sk_callback_lock);
481 
482 	if (skb->secmark)
483 		security_secid_to_secctx(skb->secmark, secdata, &seclen);
484 
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 	char *secdata = NULL;
571 	u32 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, &secdata);
646 		if (seclen)
647 			size += nla_total_size(seclen);
648 	}
649 
650 	skb = alloc_skb(size, GFP_ATOMIC);
651 	if (!skb) {
652 		skb_tx_error(entskb);
653 		goto nlmsg_failure;
654 	}
655 
656 	nlh = nfnl_msg_put(skb, 0, 0,
657 			   nfnl_msg_type(NFNL_SUBSYS_QUEUE, NFQNL_MSG_PACKET),
658 			   0, entry->state.pf, NFNETLINK_V0,
659 			   htons(queue->queue_num));
660 	if (!nlh) {
661 		skb_tx_error(entskb);
662 		kfree_skb(skb);
663 		goto nlmsg_failure;
664 	}
665 
666 	nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
667 	pmsg = nla_data(nla);
668 	pmsg->hw_protocol	= entskb->protocol;
669 	pmsg->hook		= entry->state.hook;
670 	*packet_id_ptr		= &pmsg->packet_id;
671 
672 	indev = entry->state.in;
673 	if (indev) {
674 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
675 		if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
676 			goto nla_put_failure;
677 #else
678 		if (entry->state.pf == PF_BRIDGE) {
679 			/* Case 1: indev is physical input device, we need to
680 			 * look for bridge group (when called from
681 			 * netfilter_bridge) */
682 			if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
683 					 htonl(indev->ifindex)) ||
684 			/* this is the bridge group "brX" */
685 			/* rcu_read_lock()ed by __nf_queue */
686 			    nla_put_be32(skb, NFQA_IFINDEX_INDEV,
687 					 htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
688 				goto nla_put_failure;
689 		} else {
690 			int physinif;
691 
692 			/* Case 2: indev is bridge group, we need to look for
693 			 * physical device (when called from ipv4) */
694 			if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
695 					 htonl(indev->ifindex)))
696 				goto nla_put_failure;
697 
698 			physinif = nf_bridge_get_physinif(entskb);
699 			if (physinif &&
700 			    nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
701 					 htonl(physinif)))
702 				goto nla_put_failure;
703 		}
704 #endif
705 	}
706 
707 	if (outdev) {
708 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
709 		if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
710 			goto nla_put_failure;
711 #else
712 		if (entry->state.pf == PF_BRIDGE) {
713 			/* Case 1: outdev is physical output device, we need to
714 			 * look for bridge group (when called from
715 			 * netfilter_bridge) */
716 			if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
717 					 htonl(outdev->ifindex)) ||
718 			/* this is the bridge group "brX" */
719 			/* rcu_read_lock()ed by __nf_queue */
720 			    nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
721 					 htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
722 				goto nla_put_failure;
723 		} else {
724 			int physoutif;
725 
726 			/* Case 2: outdev is bridge group, we need to look for
727 			 * physical output device (when called from ipv4) */
728 			if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
729 					 htonl(outdev->ifindex)))
730 				goto nla_put_failure;
731 
732 			physoutif = nf_bridge_get_physoutif(entskb);
733 			if (physoutif &&
734 			    nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
735 					 htonl(physoutif)))
736 				goto nla_put_failure;
737 		}
738 #endif
739 	}
740 
741 	if (entskb->mark &&
742 	    nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
743 		goto nla_put_failure;
744 
745 	if (entskb->priority &&
746 	    nla_put_be32(skb, NFQA_PRIORITY, htonl(entskb->priority)))
747 		goto nla_put_failure;
748 
749 	if (indev && entskb->dev &&
750 	    skb_mac_header_was_set(entskb) &&
751 	    skb_mac_header_len(entskb) != 0) {
752 		struct nfqnl_msg_packet_hw phw;
753 		int len;
754 
755 		memset(&phw, 0, sizeof(phw));
756 		len = dev_parse_header(entskb, phw.hw_addr);
757 		if (len) {
758 			phw.hw_addrlen = htons(len);
759 			if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
760 				goto nla_put_failure;
761 		}
762 	}
763 
764 	if (nfqnl_put_bridge(entry, skb) < 0)
765 		goto nla_put_failure;
766 
767 	if (entry->state.hook <= NF_INET_FORWARD && tstamp) {
768 		struct nfqnl_msg_packet_timestamp ts;
769 		struct timespec64 kts = ktime_to_timespec64(tstamp);
770 
771 		ts.sec = cpu_to_be64(kts.tv_sec);
772 		ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
773 
774 		if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
775 			goto nla_put_failure;
776 	}
777 
778 	if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
779 	    nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
780 		goto nla_put_failure;
781 
782 	if (nfqnl_put_sk_classid(skb, entskb->sk) < 0)
783 		goto nla_put_failure;
784 
785 	if (seclen && nla_put(skb, NFQA_SECCTX, seclen, secdata))
786 		goto nla_put_failure;
787 
788 	if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
789 		goto nla_put_failure;
790 
791 	if (cap_len > data_len &&
792 	    nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
793 		goto nla_put_failure;
794 
795 	if (nfqnl_put_packet_info(skb, entskb, csum_verify))
796 		goto nla_put_failure;
797 
798 	if (data_len) {
799 		struct nlattr *nla;
800 
801 		if (skb_tailroom(skb) < sizeof(*nla) + hlen)
802 			goto nla_put_failure;
803 
804 		nla = skb_put(skb, sizeof(*nla));
805 		nla->nla_type = NFQA_PAYLOAD;
806 		nla->nla_len = nla_attr_size(data_len);
807 
808 		if (skb_zerocopy(skb, entskb, data_len, hlen))
809 			goto nla_put_failure;
810 	}
811 
812 	nlh->nlmsg_len = skb->len;
813 	if (seclen)
814 		security_release_secctx(secdata, seclen);
815 	return skb;
816 
817 nla_put_failure:
818 	skb_tx_error(entskb);
819 	kfree_skb(skb);
820 	net_err_ratelimited("nf_queue: error creating packet message\n");
821 nlmsg_failure:
822 	if (seclen)
823 		security_release_secctx(secdata, seclen);
824 	return NULL;
825 }
826 
827 static bool nf_ct_drop_unconfirmed(const struct nf_queue_entry *entry)
828 {
829 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
830 	static const unsigned long flags = IPS_CONFIRMED | IPS_DYING;
831 	struct nf_conn *ct = (void *)skb_nfct(entry->skb);
832 	unsigned long status;
833 	unsigned int use;
834 
835 	if (!ct)
836 		return false;
837 
838 	status = READ_ONCE(ct->status);
839 	if ((status & flags) == IPS_DYING)
840 		return true;
841 
842 	if (status & IPS_CONFIRMED)
843 		return false;
844 
845 	/* in some cases skb_clone() can occur after initial conntrack
846 	 * pickup, but conntrack assumes exclusive skb->_nfct ownership for
847 	 * unconfirmed entries.
848 	 *
849 	 * This happens for br_netfilter and with ip multicast routing.
850 	 * We can't be solved with serialization here because one clone could
851 	 * have been queued for local delivery.
852 	 */
853 	use = refcount_read(&ct->ct_general.use);
854 	if (likely(use == 1))
855 		return false;
856 
857 	/* Can't decrement further? Exclusive ownership. */
858 	if (!refcount_dec_not_one(&ct->ct_general.use))
859 		return false;
860 
861 	skb_set_nfct(entry->skb, 0);
862 	/* No nf_ct_put(): we already decremented .use and it cannot
863 	 * drop down to 0.
864 	 */
865 	return true;
866 #endif
867 	return false;
868 }
869 
870 static int
871 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
872 			struct nf_queue_entry *entry)
873 {
874 	struct sk_buff *nskb;
875 	int err = -ENOBUFS;
876 	__be32 *packet_id_ptr;
877 	int failopen = 0;
878 
879 	nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
880 	if (nskb == NULL) {
881 		err = -ENOMEM;
882 		goto err_out;
883 	}
884 	spin_lock_bh(&queue->lock);
885 
886 	if (nf_ct_drop_unconfirmed(entry))
887 		goto err_out_free_nskb;
888 
889 	if (queue->queue_total >= queue->queue_maxlen) {
890 		if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
891 			failopen = 1;
892 			err = 0;
893 		} else {
894 			queue->queue_dropped++;
895 			net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
896 					     queue->queue_total);
897 		}
898 		goto err_out_free_nskb;
899 	}
900 	entry->id = ++queue->id_sequence;
901 	*packet_id_ptr = htonl(entry->id);
902 
903 	/* nfnetlink_unicast will either free the nskb or add it to a socket */
904 	err = nfnetlink_unicast(nskb, net, queue->peer_portid);
905 	if (err < 0) {
906 		if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
907 			failopen = 1;
908 			err = 0;
909 		} else {
910 			queue->queue_user_dropped++;
911 		}
912 		goto err_out_unlock;
913 	}
914 
915 	__enqueue_entry(queue, entry);
916 
917 	spin_unlock_bh(&queue->lock);
918 	return 0;
919 
920 err_out_free_nskb:
921 	kfree_skb(nskb);
922 err_out_unlock:
923 	spin_unlock_bh(&queue->lock);
924 	if (failopen)
925 		nfqnl_reinject(entry, NF_ACCEPT);
926 err_out:
927 	return err;
928 }
929 
930 static struct nf_queue_entry *
931 nf_queue_entry_dup(struct nf_queue_entry *e)
932 {
933 	struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
934 
935 	if (!entry)
936 		return NULL;
937 
938 	if (nf_queue_entry_get_refs(entry))
939 		return entry;
940 
941 	kfree(entry);
942 	return NULL;
943 }
944 
945 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
946 /* When called from bridge netfilter, skb->data must point to MAC header
947  * before calling skb_gso_segment(). Else, original MAC header is lost
948  * and segmented skbs will be sent to wrong destination.
949  */
950 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
951 {
952 	if (nf_bridge_info_get(skb))
953 		__skb_push(skb, skb->network_header - skb->mac_header);
954 }
955 
956 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
957 {
958 	if (nf_bridge_info_get(skb))
959 		__skb_pull(skb, skb->network_header - skb->mac_header);
960 }
961 #else
962 #define nf_bridge_adjust_skb_data(s) do {} while (0)
963 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
964 #endif
965 
966 static int
967 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
968 			   struct sk_buff *skb, struct nf_queue_entry *entry)
969 {
970 	int ret = -ENOMEM;
971 	struct nf_queue_entry *entry_seg;
972 
973 	nf_bridge_adjust_segmented_data(skb);
974 
975 	if (skb->next == NULL) { /* last packet, no need to copy entry */
976 		struct sk_buff *gso_skb = entry->skb;
977 		entry->skb = skb;
978 		ret = __nfqnl_enqueue_packet(net, queue, entry);
979 		if (ret)
980 			entry->skb = gso_skb;
981 		return ret;
982 	}
983 
984 	skb_mark_not_on_list(skb);
985 
986 	entry_seg = nf_queue_entry_dup(entry);
987 	if (entry_seg) {
988 		entry_seg->skb = skb;
989 		ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
990 		if (ret)
991 			nf_queue_entry_free(entry_seg);
992 	}
993 	return ret;
994 }
995 
996 static int
997 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
998 {
999 	unsigned int queued;
1000 	struct nfqnl_instance *queue;
1001 	struct sk_buff *skb, *segs, *nskb;
1002 	int err = -ENOBUFS;
1003 	struct net *net = entry->state.net;
1004 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1005 
1006 	/* rcu_read_lock()ed by nf_hook_thresh */
1007 	queue = instance_lookup(q, queuenum);
1008 	if (!queue)
1009 		return -ESRCH;
1010 
1011 	if (queue->copy_mode == NFQNL_COPY_NONE)
1012 		return -EINVAL;
1013 
1014 	skb = entry->skb;
1015 
1016 	switch (entry->state.pf) {
1017 	case NFPROTO_IPV4:
1018 		skb->protocol = htons(ETH_P_IP);
1019 		break;
1020 	case NFPROTO_IPV6:
1021 		skb->protocol = htons(ETH_P_IPV6);
1022 		break;
1023 	}
1024 
1025 	if (!skb_is_gso(skb) || ((queue->flags & NFQA_CFG_F_GSO) && !skb_is_gso_sctp(skb)))
1026 		return __nfqnl_enqueue_packet(net, queue, entry);
1027 
1028 	nf_bridge_adjust_skb_data(skb);
1029 	segs = skb_gso_segment(skb, 0);
1030 	/* Does not use PTR_ERR to limit the number of error codes that can be
1031 	 * returned by nf_queue.  For instance, callers rely on -ESRCH to
1032 	 * mean 'ignore this hook'.
1033 	 */
1034 	if (IS_ERR_OR_NULL(segs))
1035 		goto out_err;
1036 	queued = 0;
1037 	err = 0;
1038 	skb_list_walk_safe(segs, segs, nskb) {
1039 		if (err == 0)
1040 			err = __nfqnl_enqueue_packet_gso(net, queue,
1041 							segs, entry);
1042 		if (err == 0)
1043 			queued++;
1044 		else
1045 			kfree_skb(segs);
1046 	}
1047 
1048 	if (queued) {
1049 		if (err) /* some segments are already queued */
1050 			nf_queue_entry_free(entry);
1051 		kfree_skb(skb);
1052 		return 0;
1053 	}
1054  out_err:
1055 	nf_bridge_adjust_segmented_data(skb);
1056 	return err;
1057 }
1058 
1059 static int
1060 nfqnl_mangle(void *data, unsigned int data_len, struct nf_queue_entry *e, int diff)
1061 {
1062 	struct sk_buff *nskb;
1063 
1064 	if (diff < 0) {
1065 		unsigned int min_len = skb_transport_offset(e->skb);
1066 
1067 		if (data_len < min_len)
1068 			return -EINVAL;
1069 
1070 		if (pskb_trim(e->skb, data_len))
1071 			return -ENOMEM;
1072 	} else if (diff > 0) {
1073 		if (data_len > 0xFFFF)
1074 			return -EINVAL;
1075 		if (diff > skb_tailroom(e->skb)) {
1076 			nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
1077 					       diff, GFP_ATOMIC);
1078 			if (!nskb)
1079 				return -ENOMEM;
1080 			kfree_skb(e->skb);
1081 			e->skb = nskb;
1082 		}
1083 		skb_put(e->skb, diff);
1084 	}
1085 	if (skb_ensure_writable(e->skb, data_len))
1086 		return -ENOMEM;
1087 	skb_copy_to_linear_data(e->skb, data, data_len);
1088 	e->skb->ip_summed = CHECKSUM_NONE;
1089 	return 0;
1090 }
1091 
1092 static int
1093 nfqnl_set_mode(struct nfqnl_instance *queue,
1094 	       unsigned char mode, unsigned int range)
1095 {
1096 	int status = 0;
1097 
1098 	spin_lock_bh(&queue->lock);
1099 	switch (mode) {
1100 	case NFQNL_COPY_NONE:
1101 	case NFQNL_COPY_META:
1102 		queue->copy_mode = mode;
1103 		queue->copy_range = 0;
1104 		break;
1105 
1106 	case NFQNL_COPY_PACKET:
1107 		queue->copy_mode = mode;
1108 		if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
1109 			queue->copy_range = NFQNL_MAX_COPY_RANGE;
1110 		else
1111 			queue->copy_range = range;
1112 		break;
1113 
1114 	default:
1115 		status = -EINVAL;
1116 
1117 	}
1118 	spin_unlock_bh(&queue->lock);
1119 
1120 	return status;
1121 }
1122 
1123 static int
1124 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
1125 {
1126 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
1127 	int physinif, physoutif;
1128 
1129 	physinif = nf_bridge_get_physinif(entry->skb);
1130 	physoutif = nf_bridge_get_physoutif(entry->skb);
1131 
1132 	if (physinif == ifindex || physoutif == ifindex)
1133 		return 1;
1134 #endif
1135 	if (entry->state.in)
1136 		if (entry->state.in->ifindex == ifindex)
1137 			return 1;
1138 	if (entry->state.out)
1139 		if (entry->state.out->ifindex == ifindex)
1140 			return 1;
1141 
1142 	return 0;
1143 }
1144 
1145 /* drop all packets with either indev or outdev == ifindex from all queue
1146  * instances */
1147 static void
1148 nfqnl_dev_drop(struct net *net, int ifindex)
1149 {
1150 	int i;
1151 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1152 
1153 	rcu_read_lock();
1154 
1155 	for (i = 0; i < INSTANCE_BUCKETS; i++) {
1156 		struct nfqnl_instance *inst;
1157 		struct hlist_head *head = &q->instance_table[i];
1158 
1159 		hlist_for_each_entry_rcu(inst, head, hlist)
1160 			nfqnl_flush(inst, dev_cmp, ifindex);
1161 	}
1162 
1163 	rcu_read_unlock();
1164 }
1165 
1166 static int
1167 nfqnl_rcv_dev_event(struct notifier_block *this,
1168 		    unsigned long event, void *ptr)
1169 {
1170 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1171 
1172 	/* Drop any packets associated with the downed device */
1173 	if (event == NETDEV_DOWN)
1174 		nfqnl_dev_drop(dev_net(dev), dev->ifindex);
1175 	return NOTIFY_DONE;
1176 }
1177 
1178 static struct notifier_block nfqnl_dev_notifier = {
1179 	.notifier_call	= nfqnl_rcv_dev_event,
1180 };
1181 
1182 static void nfqnl_nf_hook_drop(struct net *net)
1183 {
1184 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1185 	int i;
1186 
1187 	/* This function is also called on net namespace error unwind,
1188 	 * when pernet_ops->init() failed and ->exit() functions of the
1189 	 * previous pernet_ops gets called.
1190 	 *
1191 	 * This may result in a call to nfqnl_nf_hook_drop() before
1192 	 * struct nfnl_queue_net was allocated.
1193 	 */
1194 	if (!q)
1195 		return;
1196 
1197 	for (i = 0; i < INSTANCE_BUCKETS; i++) {
1198 		struct nfqnl_instance *inst;
1199 		struct hlist_head *head = &q->instance_table[i];
1200 
1201 		hlist_for_each_entry_rcu(inst, head, hlist)
1202 			nfqnl_flush(inst, NULL, 0);
1203 	}
1204 }
1205 
1206 static int
1207 nfqnl_rcv_nl_event(struct notifier_block *this,
1208 		   unsigned long event, void *ptr)
1209 {
1210 	struct netlink_notify *n = ptr;
1211 	struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
1212 
1213 	if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
1214 		int i;
1215 
1216 		/* destroy all instances for this portid */
1217 		spin_lock(&q->instances_lock);
1218 		for (i = 0; i < INSTANCE_BUCKETS; i++) {
1219 			struct hlist_node *t2;
1220 			struct nfqnl_instance *inst;
1221 			struct hlist_head *head = &q->instance_table[i];
1222 
1223 			hlist_for_each_entry_safe(inst, t2, head, hlist) {
1224 				if (n->portid == inst->peer_portid)
1225 					__instance_destroy(inst);
1226 			}
1227 		}
1228 		spin_unlock(&q->instances_lock);
1229 	}
1230 	return NOTIFY_DONE;
1231 }
1232 
1233 static struct notifier_block nfqnl_rtnl_notifier = {
1234 	.notifier_call	= nfqnl_rcv_nl_event,
1235 };
1236 
1237 static const struct nla_policy nfqa_vlan_policy[NFQA_VLAN_MAX + 1] = {
1238 	[NFQA_VLAN_TCI]		= { .type = NLA_U16},
1239 	[NFQA_VLAN_PROTO]	= { .type = NLA_U16},
1240 };
1241 
1242 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
1243 	[NFQA_VERDICT_HDR]	= { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1244 	[NFQA_MARK]		= { .type = NLA_U32 },
1245 	[NFQA_PAYLOAD]		= { .type = NLA_UNSPEC },
1246 	[NFQA_CT]		= { .type = NLA_UNSPEC },
1247 	[NFQA_EXP]		= { .type = NLA_UNSPEC },
1248 	[NFQA_VLAN]		= { .type = NLA_NESTED },
1249 	[NFQA_PRIORITY]		= { .type = NLA_U32 },
1250 };
1251 
1252 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
1253 	[NFQA_VERDICT_HDR]	= { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1254 	[NFQA_MARK]		= { .type = NLA_U32 },
1255 	[NFQA_PRIORITY]		= { .type = NLA_U32 },
1256 };
1257 
1258 static struct nfqnl_instance *
1259 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
1260 {
1261 	struct nfqnl_instance *queue;
1262 
1263 	queue = instance_lookup(q, queue_num);
1264 	if (!queue)
1265 		return ERR_PTR(-ENODEV);
1266 
1267 	if (queue->peer_portid != nlportid)
1268 		return ERR_PTR(-EPERM);
1269 
1270 	return queue;
1271 }
1272 
1273 static struct nfqnl_msg_verdict_hdr*
1274 verdicthdr_get(const struct nlattr * const nfqa[])
1275 {
1276 	struct nfqnl_msg_verdict_hdr *vhdr;
1277 	unsigned int verdict;
1278 
1279 	if (!nfqa[NFQA_VERDICT_HDR])
1280 		return NULL;
1281 
1282 	vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
1283 	verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
1284 	if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
1285 		return NULL;
1286 	return vhdr;
1287 }
1288 
1289 static int nfq_id_after(unsigned int id, unsigned int max)
1290 {
1291 	return (int)(id - max) > 0;
1292 }
1293 
1294 static int nfqnl_recv_verdict_batch(struct sk_buff *skb,
1295 				    const struct nfnl_info *info,
1296 				    const struct nlattr * const nfqa[])
1297 {
1298 	struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1299 	u16 queue_num = ntohs(info->nfmsg->res_id);
1300 	struct nf_queue_entry *entry, *tmp;
1301 	struct nfqnl_msg_verdict_hdr *vhdr;
1302 	struct nfqnl_instance *queue;
1303 	unsigned int verdict, maxid;
1304 	LIST_HEAD(batch_list);
1305 
1306 	queue = verdict_instance_lookup(q, queue_num,
1307 					NETLINK_CB(skb).portid);
1308 	if (IS_ERR(queue))
1309 		return PTR_ERR(queue);
1310 
1311 	vhdr = verdicthdr_get(nfqa);
1312 	if (!vhdr)
1313 		return -EINVAL;
1314 
1315 	verdict = ntohl(vhdr->verdict);
1316 	maxid = ntohl(vhdr->id);
1317 
1318 	spin_lock_bh(&queue->lock);
1319 
1320 	list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
1321 		if (nfq_id_after(entry->id, maxid))
1322 			break;
1323 		__dequeue_entry(queue, entry);
1324 		list_add_tail(&entry->list, &batch_list);
1325 	}
1326 
1327 	spin_unlock_bh(&queue->lock);
1328 
1329 	if (list_empty(&batch_list))
1330 		return -ENOENT;
1331 
1332 	list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1333 		if (nfqa[NFQA_MARK])
1334 			entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1335 
1336 		if (nfqa[NFQA_PRIORITY])
1337 			entry->skb->priority = ntohl(nla_get_be32(nfqa[NFQA_PRIORITY]));
1338 
1339 		nfqnl_reinject(entry, verdict);
1340 	}
1341 	return 0;
1342 }
1343 
1344 static struct nf_conn *nfqnl_ct_parse(const struct nfnl_ct_hook *nfnl_ct,
1345 				      const struct nlmsghdr *nlh,
1346 				      const struct nlattr * const nfqa[],
1347 				      struct nf_queue_entry *entry,
1348 				      enum ip_conntrack_info *ctinfo)
1349 {
1350 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1351 	struct nf_conn *ct;
1352 
1353 	ct = nf_ct_get(entry->skb, ctinfo);
1354 	if (ct == NULL)
1355 		return NULL;
1356 
1357 	if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1358 		return NULL;
1359 
1360 	if (nfqa[NFQA_EXP])
1361 		nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1362 				      NETLINK_CB(entry->skb).portid,
1363 				      nlmsg_report(nlh));
1364 	return ct;
1365 #else
1366 	return NULL;
1367 #endif
1368 }
1369 
1370 static int nfqa_parse_bridge(struct nf_queue_entry *entry,
1371 			     const struct nlattr * const nfqa[])
1372 {
1373 	if (nfqa[NFQA_VLAN]) {
1374 		struct nlattr *tb[NFQA_VLAN_MAX + 1];
1375 		int err;
1376 
1377 		err = nla_parse_nested_deprecated(tb, NFQA_VLAN_MAX,
1378 						  nfqa[NFQA_VLAN],
1379 						  nfqa_vlan_policy, NULL);
1380 		if (err < 0)
1381 			return err;
1382 
1383 		if (!tb[NFQA_VLAN_TCI] || !tb[NFQA_VLAN_PROTO])
1384 			return -EINVAL;
1385 
1386 		__vlan_hwaccel_put_tag(entry->skb,
1387 			nla_get_be16(tb[NFQA_VLAN_PROTO]),
1388 			ntohs(nla_get_be16(tb[NFQA_VLAN_TCI])));
1389 	}
1390 
1391 	if (nfqa[NFQA_L2HDR]) {
1392 		int mac_header_len = entry->skb->network_header -
1393 			entry->skb->mac_header;
1394 
1395 		if (mac_header_len != nla_len(nfqa[NFQA_L2HDR]))
1396 			return -EINVAL;
1397 		else if (mac_header_len > 0)
1398 			memcpy(skb_mac_header(entry->skb),
1399 			       nla_data(nfqa[NFQA_L2HDR]),
1400 			       mac_header_len);
1401 	}
1402 
1403 	return 0;
1404 }
1405 
1406 static int nfqnl_recv_verdict(struct sk_buff *skb, const struct nfnl_info *info,
1407 			      const struct nlattr * const nfqa[])
1408 {
1409 	struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1410 	u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1411 	const struct nfnl_ct_hook *nfnl_ct;
1412 	struct nfqnl_msg_verdict_hdr *vhdr;
1413 	enum ip_conntrack_info ctinfo;
1414 	struct nfqnl_instance *queue;
1415 	struct nf_queue_entry *entry;
1416 	struct nf_conn *ct = NULL;
1417 	unsigned int verdict;
1418 	int err;
1419 
1420 	queue = verdict_instance_lookup(q, queue_num,
1421 					NETLINK_CB(skb).portid);
1422 	if (IS_ERR(queue))
1423 		return PTR_ERR(queue);
1424 
1425 	vhdr = verdicthdr_get(nfqa);
1426 	if (!vhdr)
1427 		return -EINVAL;
1428 
1429 	verdict = ntohl(vhdr->verdict);
1430 
1431 	entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1432 	if (entry == NULL)
1433 		return -ENOENT;
1434 
1435 	/* rcu lock already held from nfnl->call_rcu. */
1436 	nfnl_ct = rcu_dereference(nfnl_ct_hook);
1437 
1438 	if (nfqa[NFQA_CT]) {
1439 		if (nfnl_ct != NULL)
1440 			ct = nfqnl_ct_parse(nfnl_ct, info->nlh, nfqa, entry,
1441 					    &ctinfo);
1442 	}
1443 
1444 	if (entry->state.pf == PF_BRIDGE) {
1445 		err = nfqa_parse_bridge(entry, nfqa);
1446 		if (err < 0)
1447 			return err;
1448 	}
1449 
1450 	if (nfqa[NFQA_PAYLOAD]) {
1451 		u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1452 		int diff = payload_len - entry->skb->len;
1453 
1454 		if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1455 				 payload_len, entry, diff) < 0)
1456 			verdict = NF_DROP;
1457 
1458 		if (ct && diff)
1459 			nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1460 	}
1461 
1462 	if (nfqa[NFQA_MARK])
1463 		entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1464 
1465 	if (nfqa[NFQA_PRIORITY])
1466 		entry->skb->priority = ntohl(nla_get_be32(nfqa[NFQA_PRIORITY]));
1467 
1468 	nfqnl_reinject(entry, verdict);
1469 	return 0;
1470 }
1471 
1472 static int nfqnl_recv_unsupp(struct sk_buff *skb, const struct nfnl_info *info,
1473 			     const struct nlattr * const cda[])
1474 {
1475 	return -ENOTSUPP;
1476 }
1477 
1478 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1479 	[NFQA_CFG_CMD]		= { .len = sizeof(struct nfqnl_msg_config_cmd) },
1480 	[NFQA_CFG_PARAMS]	= { .len = sizeof(struct nfqnl_msg_config_params) },
1481 	[NFQA_CFG_QUEUE_MAXLEN]	= { .type = NLA_U32 },
1482 	[NFQA_CFG_MASK]		= { .type = NLA_U32 },
1483 	[NFQA_CFG_FLAGS]	= { .type = NLA_U32 },
1484 };
1485 
1486 static const struct nf_queue_handler nfqh = {
1487 	.outfn		= nfqnl_enqueue_packet,
1488 	.nf_hook_drop	= nfqnl_nf_hook_drop,
1489 };
1490 
1491 static int nfqnl_recv_config(struct sk_buff *skb, const struct nfnl_info *info,
1492 			     const struct nlattr * const nfqa[])
1493 {
1494 	struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1495 	u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1496 	struct nfqnl_msg_config_cmd *cmd = NULL;
1497 	struct nfqnl_instance *queue;
1498 	__u32 flags = 0, mask = 0;
1499 	int ret = 0;
1500 
1501 	if (nfqa[NFQA_CFG_CMD]) {
1502 		cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1503 
1504 		/* Obsolete commands without queue context */
1505 		switch (cmd->command) {
1506 		case NFQNL_CFG_CMD_PF_BIND: return 0;
1507 		case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1508 		}
1509 	}
1510 
1511 	/* Check if we support these flags in first place, dependencies should
1512 	 * be there too not to break atomicity.
1513 	 */
1514 	if (nfqa[NFQA_CFG_FLAGS]) {
1515 		if (!nfqa[NFQA_CFG_MASK]) {
1516 			/* A mask is needed to specify which flags are being
1517 			 * changed.
1518 			 */
1519 			return -EINVAL;
1520 		}
1521 
1522 		flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1523 		mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1524 
1525 		if (flags >= NFQA_CFG_F_MAX)
1526 			return -EOPNOTSUPP;
1527 
1528 #if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1529 		if (flags & mask & NFQA_CFG_F_SECCTX)
1530 			return -EOPNOTSUPP;
1531 #endif
1532 		if ((flags & mask & NFQA_CFG_F_CONNTRACK) &&
1533 		    !rcu_access_pointer(nfnl_ct_hook)) {
1534 #ifdef CONFIG_MODULES
1535 			nfnl_unlock(NFNL_SUBSYS_QUEUE);
1536 			request_module("ip_conntrack_netlink");
1537 			nfnl_lock(NFNL_SUBSYS_QUEUE);
1538 			if (rcu_access_pointer(nfnl_ct_hook))
1539 				return -EAGAIN;
1540 #endif
1541 			return -EOPNOTSUPP;
1542 		}
1543 	}
1544 
1545 	rcu_read_lock();
1546 	queue = instance_lookup(q, queue_num);
1547 	if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1548 		ret = -EPERM;
1549 		goto err_out_unlock;
1550 	}
1551 
1552 	if (cmd != NULL) {
1553 		switch (cmd->command) {
1554 		case NFQNL_CFG_CMD_BIND:
1555 			if (queue) {
1556 				ret = -EBUSY;
1557 				goto err_out_unlock;
1558 			}
1559 			queue = instance_create(q, queue_num,
1560 						NETLINK_CB(skb).portid);
1561 			if (IS_ERR(queue)) {
1562 				ret = PTR_ERR(queue);
1563 				goto err_out_unlock;
1564 			}
1565 			break;
1566 		case NFQNL_CFG_CMD_UNBIND:
1567 			if (!queue) {
1568 				ret = -ENODEV;
1569 				goto err_out_unlock;
1570 			}
1571 			instance_destroy(q, queue);
1572 			goto err_out_unlock;
1573 		case NFQNL_CFG_CMD_PF_BIND:
1574 		case NFQNL_CFG_CMD_PF_UNBIND:
1575 			break;
1576 		default:
1577 			ret = -ENOTSUPP;
1578 			goto err_out_unlock;
1579 		}
1580 	}
1581 
1582 	if (!queue) {
1583 		ret = -ENODEV;
1584 		goto err_out_unlock;
1585 	}
1586 
1587 	if (nfqa[NFQA_CFG_PARAMS]) {
1588 		struct nfqnl_msg_config_params *params =
1589 			nla_data(nfqa[NFQA_CFG_PARAMS]);
1590 
1591 		nfqnl_set_mode(queue, params->copy_mode,
1592 				ntohl(params->copy_range));
1593 	}
1594 
1595 	if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1596 		__be32 *queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1597 
1598 		spin_lock_bh(&queue->lock);
1599 		queue->queue_maxlen = ntohl(*queue_maxlen);
1600 		spin_unlock_bh(&queue->lock);
1601 	}
1602 
1603 	if (nfqa[NFQA_CFG_FLAGS]) {
1604 		spin_lock_bh(&queue->lock);
1605 		queue->flags &= ~mask;
1606 		queue->flags |= flags & mask;
1607 		spin_unlock_bh(&queue->lock);
1608 	}
1609 
1610 err_out_unlock:
1611 	rcu_read_unlock();
1612 	return ret;
1613 }
1614 
1615 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1616 	[NFQNL_MSG_PACKET]	= {
1617 		.call		= nfqnl_recv_unsupp,
1618 		.type		= NFNL_CB_RCU,
1619 		.attr_count	= NFQA_MAX,
1620 	},
1621 	[NFQNL_MSG_VERDICT]	= {
1622 		.call		= nfqnl_recv_verdict,
1623 		.type		= NFNL_CB_RCU,
1624 		.attr_count	= NFQA_MAX,
1625 		.policy		= nfqa_verdict_policy
1626 	},
1627 	[NFQNL_MSG_CONFIG]	= {
1628 		.call		= nfqnl_recv_config,
1629 		.type		= NFNL_CB_MUTEX,
1630 		.attr_count	= NFQA_CFG_MAX,
1631 		.policy		= nfqa_cfg_policy
1632 	},
1633 	[NFQNL_MSG_VERDICT_BATCH] = {
1634 		.call		= nfqnl_recv_verdict_batch,
1635 		.type		= NFNL_CB_RCU,
1636 		.attr_count	= NFQA_MAX,
1637 		.policy		= nfqa_verdict_batch_policy
1638 	},
1639 };
1640 
1641 static const struct nfnetlink_subsystem nfqnl_subsys = {
1642 	.name		= "nf_queue",
1643 	.subsys_id	= NFNL_SUBSYS_QUEUE,
1644 	.cb_count	= NFQNL_MSG_MAX,
1645 	.cb		= nfqnl_cb,
1646 };
1647 
1648 #ifdef CONFIG_PROC_FS
1649 struct iter_state {
1650 	struct seq_net_private p;
1651 	unsigned int bucket;
1652 };
1653 
1654 static struct hlist_node *get_first(struct seq_file *seq)
1655 {
1656 	struct iter_state *st = seq->private;
1657 	struct net *net;
1658 	struct nfnl_queue_net *q;
1659 
1660 	if (!st)
1661 		return NULL;
1662 
1663 	net = seq_file_net(seq);
1664 	q = nfnl_queue_pernet(net);
1665 	for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1666 		if (!hlist_empty(&q->instance_table[st->bucket]))
1667 			return q->instance_table[st->bucket].first;
1668 	}
1669 	return NULL;
1670 }
1671 
1672 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1673 {
1674 	struct iter_state *st = seq->private;
1675 	struct net *net = seq_file_net(seq);
1676 
1677 	h = h->next;
1678 	while (!h) {
1679 		struct nfnl_queue_net *q;
1680 
1681 		if (++st->bucket >= INSTANCE_BUCKETS)
1682 			return NULL;
1683 
1684 		q = nfnl_queue_pernet(net);
1685 		h = q->instance_table[st->bucket].first;
1686 	}
1687 	return h;
1688 }
1689 
1690 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1691 {
1692 	struct hlist_node *head;
1693 	head = get_first(seq);
1694 
1695 	if (head)
1696 		while (pos && (head = get_next(seq, head)))
1697 			pos--;
1698 	return pos ? NULL : head;
1699 }
1700 
1701 static void *seq_start(struct seq_file *s, loff_t *pos)
1702 	__acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1703 {
1704 	spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1705 	return get_idx(s, *pos);
1706 }
1707 
1708 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1709 {
1710 	(*pos)++;
1711 	return get_next(s, v);
1712 }
1713 
1714 static void seq_stop(struct seq_file *s, void *v)
1715 	__releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1716 {
1717 	spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1718 }
1719 
1720 static int seq_show(struct seq_file *s, void *v)
1721 {
1722 	const struct nfqnl_instance *inst = v;
1723 
1724 	seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1725 		   inst->queue_num,
1726 		   inst->peer_portid, inst->queue_total,
1727 		   inst->copy_mode, inst->copy_range,
1728 		   inst->queue_dropped, inst->queue_user_dropped,
1729 		   inst->id_sequence, 1);
1730 	return 0;
1731 }
1732 
1733 static const struct seq_operations nfqnl_seq_ops = {
1734 	.start	= seq_start,
1735 	.next	= seq_next,
1736 	.stop	= seq_stop,
1737 	.show	= seq_show,
1738 };
1739 #endif /* PROC_FS */
1740 
1741 static int __net_init nfnl_queue_net_init(struct net *net)
1742 {
1743 	unsigned int i;
1744 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1745 
1746 	for (i = 0; i < INSTANCE_BUCKETS; i++)
1747 		INIT_HLIST_HEAD(&q->instance_table[i]);
1748 
1749 	spin_lock_init(&q->instances_lock);
1750 
1751 #ifdef CONFIG_PROC_FS
1752 	if (!proc_create_net("nfnetlink_queue", 0440, net->nf.proc_netfilter,
1753 			&nfqnl_seq_ops, sizeof(struct iter_state)))
1754 		return -ENOMEM;
1755 #endif
1756 	return 0;
1757 }
1758 
1759 static void __net_exit nfnl_queue_net_exit(struct net *net)
1760 {
1761 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1762 	unsigned int i;
1763 
1764 #ifdef CONFIG_PROC_FS
1765 	remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1766 #endif
1767 	for (i = 0; i < INSTANCE_BUCKETS; i++)
1768 		WARN_ON_ONCE(!hlist_empty(&q->instance_table[i]));
1769 }
1770 
1771 static struct pernet_operations nfnl_queue_net_ops = {
1772 	.init		= nfnl_queue_net_init,
1773 	.exit		= nfnl_queue_net_exit,
1774 	.id		= &nfnl_queue_net_id,
1775 	.size		= sizeof(struct nfnl_queue_net),
1776 };
1777 
1778 static int __init nfnetlink_queue_init(void)
1779 {
1780 	int status;
1781 
1782 	status = register_pernet_subsys(&nfnl_queue_net_ops);
1783 	if (status < 0) {
1784 		pr_err("failed to register pernet ops\n");
1785 		goto out;
1786 	}
1787 
1788 	netlink_register_notifier(&nfqnl_rtnl_notifier);
1789 	status = nfnetlink_subsys_register(&nfqnl_subsys);
1790 	if (status < 0) {
1791 		pr_err("failed to create netlink socket\n");
1792 		goto cleanup_netlink_notifier;
1793 	}
1794 
1795 	status = register_netdevice_notifier(&nfqnl_dev_notifier);
1796 	if (status < 0) {
1797 		pr_err("failed to register netdevice notifier\n");
1798 		goto cleanup_netlink_subsys;
1799 	}
1800 
1801 	nf_register_queue_handler(&nfqh);
1802 
1803 	return status;
1804 
1805 cleanup_netlink_subsys:
1806 	nfnetlink_subsys_unregister(&nfqnl_subsys);
1807 cleanup_netlink_notifier:
1808 	netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1809 	unregister_pernet_subsys(&nfnl_queue_net_ops);
1810 out:
1811 	return status;
1812 }
1813 
1814 static void __exit nfnetlink_queue_fini(void)
1815 {
1816 	nf_unregister_queue_handler();
1817 	unregister_netdevice_notifier(&nfqnl_dev_notifier);
1818 	nfnetlink_subsys_unregister(&nfqnl_subsys);
1819 	netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1820 	unregister_pernet_subsys(&nfnl_queue_net_ops);
1821 
1822 	rcu_barrier(); /* Wait for completion of call_rcu()'s */
1823 }
1824 
1825 MODULE_DESCRIPTION("netfilter packet queue handler");
1826 MODULE_AUTHOR("Harald Welte <[email protected]>");
1827 MODULE_LICENSE("GPL");
1828 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1829 
1830 module_init(nfnetlink_queue_init);
1831 module_exit(nfnetlink_queue_fini);
1832