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 struct sk_buff *
544 nfqnl_build_packet_message(struct net *net, struct nfqnl_instance *queue,
545 			   struct nf_queue_entry *entry,
546 			   __be32 **packet_id_ptr)
547 {
548 	size_t size;
549 	size_t data_len = 0, cap_len = 0;
550 	unsigned int hlen = 0;
551 	struct sk_buff *skb;
552 	struct nlattr *nla;
553 	struct nfqnl_msg_packet_hdr *pmsg;
554 	struct nlmsghdr *nlh;
555 	struct sk_buff *entskb = entry->skb;
556 	struct net_device *indev;
557 	struct net_device *outdev;
558 	struct nf_conn *ct = NULL;
559 	enum ip_conntrack_info ctinfo = 0;
560 	const struct nfnl_ct_hook *nfnl_ct;
561 	bool csum_verify;
562 	char *secdata = NULL;
563 	u32 seclen = 0;
564 	ktime_t tstamp;
565 
566 	size = nlmsg_total_size(sizeof(struct nfgenmsg))
567 		+ nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
568 		+ nla_total_size(sizeof(u_int32_t))	/* ifindex */
569 		+ nla_total_size(sizeof(u_int32_t))	/* ifindex */
570 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
571 		+ nla_total_size(sizeof(u_int32_t))	/* ifindex */
572 		+ nla_total_size(sizeof(u_int32_t))	/* ifindex */
573 #endif
574 		+ nla_total_size(sizeof(u_int32_t))	/* mark */
575 		+ nla_total_size(sizeof(u_int32_t))	/* priority */
576 		+ nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
577 		+ nla_total_size(sizeof(u_int32_t))	/* skbinfo */
578 #if IS_ENABLED(CONFIG_CGROUP_NET_CLASSID)
579 		+ nla_total_size(sizeof(u_int32_t))	/* classid */
580 #endif
581 		+ nla_total_size(sizeof(u_int32_t));	/* cap_len */
582 
583 	tstamp = skb_tstamp_cond(entskb, false);
584 	if (tstamp)
585 		size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
586 
587 	size += nfqnl_get_bridge_size(entry);
588 
589 	if (entry->state.hook <= NF_INET_FORWARD ||
590 	   (entry->state.hook == NF_INET_POST_ROUTING && entskb->sk == NULL))
591 		csum_verify = !skb_csum_unnecessary(entskb);
592 	else
593 		csum_verify = false;
594 
595 	outdev = entry->state.out;
596 
597 	switch ((enum nfqnl_config_mode)READ_ONCE(queue->copy_mode)) {
598 	case NFQNL_COPY_META:
599 	case NFQNL_COPY_NONE:
600 		break;
601 
602 	case NFQNL_COPY_PACKET:
603 		if (!(queue->flags & NFQA_CFG_F_GSO) &&
604 		    entskb->ip_summed == CHECKSUM_PARTIAL &&
605 		    skb_checksum_help(entskb))
606 			return NULL;
607 
608 		data_len = READ_ONCE(queue->copy_range);
609 		if (data_len > entskb->len)
610 			data_len = entskb->len;
611 
612 		hlen = skb_zerocopy_headlen(entskb);
613 		hlen = min_t(unsigned int, hlen, data_len);
614 		size += sizeof(struct nlattr) + hlen;
615 		cap_len = entskb->len;
616 		break;
617 	}
618 
619 	nfnl_ct = rcu_dereference(nfnl_ct_hook);
620 
621 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
622 	if (queue->flags & NFQA_CFG_F_CONNTRACK) {
623 		if (nfnl_ct != NULL) {
624 			ct = nf_ct_get(entskb, &ctinfo);
625 			if (ct != NULL)
626 				size += nfnl_ct->build_size(ct);
627 		}
628 	}
629 #endif
630 
631 	if (queue->flags & NFQA_CFG_F_UID_GID) {
632 		size += (nla_total_size(sizeof(u_int32_t))	/* uid */
633 			+ nla_total_size(sizeof(u_int32_t)));	/* gid */
634 	}
635 
636 	if ((queue->flags & NFQA_CFG_F_SECCTX) && entskb->sk) {
637 		seclen = nfqnl_get_sk_secctx(entskb, &secdata);
638 		if (seclen)
639 			size += nla_total_size(seclen);
640 	}
641 
642 	skb = alloc_skb(size, GFP_ATOMIC);
643 	if (!skb) {
644 		skb_tx_error(entskb);
645 		goto nlmsg_failure;
646 	}
647 
648 	nlh = nfnl_msg_put(skb, 0, 0,
649 			   nfnl_msg_type(NFNL_SUBSYS_QUEUE, NFQNL_MSG_PACKET),
650 			   0, entry->state.pf, NFNETLINK_V0,
651 			   htons(queue->queue_num));
652 	if (!nlh) {
653 		skb_tx_error(entskb);
654 		kfree_skb(skb);
655 		goto nlmsg_failure;
656 	}
657 
658 	nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
659 	pmsg = nla_data(nla);
660 	pmsg->hw_protocol	= entskb->protocol;
661 	pmsg->hook		= entry->state.hook;
662 	*packet_id_ptr		= &pmsg->packet_id;
663 
664 	indev = entry->state.in;
665 	if (indev) {
666 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
667 		if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
668 			goto nla_put_failure;
669 #else
670 		if (entry->state.pf == PF_BRIDGE) {
671 			/* Case 1: indev is physical input device, we need to
672 			 * look for bridge group (when called from
673 			 * netfilter_bridge) */
674 			if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
675 					 htonl(indev->ifindex)) ||
676 			/* this is the bridge group "brX" */
677 			/* rcu_read_lock()ed by __nf_queue */
678 			    nla_put_be32(skb, NFQA_IFINDEX_INDEV,
679 					 htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
680 				goto nla_put_failure;
681 		} else {
682 			int physinif;
683 
684 			/* Case 2: indev is bridge group, we need to look for
685 			 * physical device (when called from ipv4) */
686 			if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
687 					 htonl(indev->ifindex)))
688 				goto nla_put_failure;
689 
690 			physinif = nf_bridge_get_physinif(entskb);
691 			if (physinif &&
692 			    nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
693 					 htonl(physinif)))
694 				goto nla_put_failure;
695 		}
696 #endif
697 	}
698 
699 	if (outdev) {
700 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
701 		if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
702 			goto nla_put_failure;
703 #else
704 		if (entry->state.pf == PF_BRIDGE) {
705 			/* Case 1: outdev is physical output device, we need to
706 			 * look for bridge group (when called from
707 			 * netfilter_bridge) */
708 			if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
709 					 htonl(outdev->ifindex)) ||
710 			/* this is the bridge group "brX" */
711 			/* rcu_read_lock()ed by __nf_queue */
712 			    nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
713 					 htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
714 				goto nla_put_failure;
715 		} else {
716 			int physoutif;
717 
718 			/* Case 2: outdev is bridge group, we need to look for
719 			 * physical output device (when called from ipv4) */
720 			if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
721 					 htonl(outdev->ifindex)))
722 				goto nla_put_failure;
723 
724 			physoutif = nf_bridge_get_physoutif(entskb);
725 			if (physoutif &&
726 			    nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
727 					 htonl(physoutif)))
728 				goto nla_put_failure;
729 		}
730 #endif
731 	}
732 
733 	if (entskb->mark &&
734 	    nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
735 		goto nla_put_failure;
736 
737 	if (entskb->priority &&
738 	    nla_put_be32(skb, NFQA_PRIORITY, htonl(entskb->priority)))
739 		goto nla_put_failure;
740 
741 	if (indev && entskb->dev &&
742 	    skb_mac_header_was_set(entskb) &&
743 	    skb_mac_header_len(entskb) != 0) {
744 		struct nfqnl_msg_packet_hw phw;
745 		int len;
746 
747 		memset(&phw, 0, sizeof(phw));
748 		len = dev_parse_header(entskb, phw.hw_addr);
749 		if (len) {
750 			phw.hw_addrlen = htons(len);
751 			if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
752 				goto nla_put_failure;
753 		}
754 	}
755 
756 	if (nfqnl_put_bridge(entry, skb) < 0)
757 		goto nla_put_failure;
758 
759 	if (entry->state.hook <= NF_INET_FORWARD && tstamp) {
760 		struct nfqnl_msg_packet_timestamp ts;
761 		struct timespec64 kts = ktime_to_timespec64(tstamp);
762 
763 		ts.sec = cpu_to_be64(kts.tv_sec);
764 		ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
765 
766 		if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
767 			goto nla_put_failure;
768 	}
769 
770 	if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
771 	    nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
772 		goto nla_put_failure;
773 
774 	if (nfqnl_put_sk_classid(skb, entskb->sk) < 0)
775 		goto nla_put_failure;
776 
777 	if (seclen && nla_put(skb, NFQA_SECCTX, seclen, secdata))
778 		goto nla_put_failure;
779 
780 	if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
781 		goto nla_put_failure;
782 
783 	if (cap_len > data_len &&
784 	    nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
785 		goto nla_put_failure;
786 
787 	if (nfqnl_put_packet_info(skb, entskb, csum_verify))
788 		goto nla_put_failure;
789 
790 	if (data_len) {
791 		struct nlattr *nla;
792 
793 		if (skb_tailroom(skb) < sizeof(*nla) + hlen)
794 			goto nla_put_failure;
795 
796 		nla = skb_put(skb, sizeof(*nla));
797 		nla->nla_type = NFQA_PAYLOAD;
798 		nla->nla_len = nla_attr_size(data_len);
799 
800 		if (skb_zerocopy(skb, entskb, data_len, hlen))
801 			goto nla_put_failure;
802 	}
803 
804 	nlh->nlmsg_len = skb->len;
805 	if (seclen)
806 		security_release_secctx(secdata, seclen);
807 	return skb;
808 
809 nla_put_failure:
810 	skb_tx_error(entskb);
811 	kfree_skb(skb);
812 	net_err_ratelimited("nf_queue: error creating packet message\n");
813 nlmsg_failure:
814 	if (seclen)
815 		security_release_secctx(secdata, seclen);
816 	return NULL;
817 }
818 
819 static bool nf_ct_drop_unconfirmed(const struct nf_queue_entry *entry)
820 {
821 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
822 	static const unsigned long flags = IPS_CONFIRMED | IPS_DYING;
823 	struct nf_conn *ct = (void *)skb_nfct(entry->skb);
824 	unsigned long status;
825 	unsigned int use;
826 
827 	if (!ct)
828 		return false;
829 
830 	status = READ_ONCE(ct->status);
831 	if ((status & flags) == IPS_DYING)
832 		return true;
833 
834 	if (status & IPS_CONFIRMED)
835 		return false;
836 
837 	/* in some cases skb_clone() can occur after initial conntrack
838 	 * pickup, but conntrack assumes exclusive skb->_nfct ownership for
839 	 * unconfirmed entries.
840 	 *
841 	 * This happens for br_netfilter and with ip multicast routing.
842 	 * We can't be solved with serialization here because one clone could
843 	 * have been queued for local delivery.
844 	 */
845 	use = refcount_read(&ct->ct_general.use);
846 	if (likely(use == 1))
847 		return false;
848 
849 	/* Can't decrement further? Exclusive ownership. */
850 	if (!refcount_dec_not_one(&ct->ct_general.use))
851 		return false;
852 
853 	skb_set_nfct(entry->skb, 0);
854 	/* No nf_ct_put(): we already decremented .use and it cannot
855 	 * drop down to 0.
856 	 */
857 	return true;
858 #endif
859 	return false;
860 }
861 
862 static int
863 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
864 			struct nf_queue_entry *entry)
865 {
866 	struct sk_buff *nskb;
867 	int err = -ENOBUFS;
868 	__be32 *packet_id_ptr;
869 	int failopen = 0;
870 
871 	nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
872 	if (nskb == NULL) {
873 		err = -ENOMEM;
874 		goto err_out;
875 	}
876 	spin_lock_bh(&queue->lock);
877 
878 	if (nf_ct_drop_unconfirmed(entry))
879 		goto err_out_free_nskb;
880 
881 	if (queue->queue_total >= queue->queue_maxlen) {
882 		if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
883 			failopen = 1;
884 			err = 0;
885 		} else {
886 			queue->queue_dropped++;
887 			net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
888 					     queue->queue_total);
889 		}
890 		goto err_out_free_nskb;
891 	}
892 	entry->id = ++queue->id_sequence;
893 	*packet_id_ptr = htonl(entry->id);
894 
895 	/* nfnetlink_unicast will either free the nskb or add it to a socket */
896 	err = nfnetlink_unicast(nskb, net, queue->peer_portid);
897 	if (err < 0) {
898 		if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
899 			failopen = 1;
900 			err = 0;
901 		} else {
902 			queue->queue_user_dropped++;
903 		}
904 		goto err_out_unlock;
905 	}
906 
907 	__enqueue_entry(queue, entry);
908 
909 	spin_unlock_bh(&queue->lock);
910 	return 0;
911 
912 err_out_free_nskb:
913 	kfree_skb(nskb);
914 err_out_unlock:
915 	spin_unlock_bh(&queue->lock);
916 	if (failopen)
917 		nfqnl_reinject(entry, NF_ACCEPT);
918 err_out:
919 	return err;
920 }
921 
922 static struct nf_queue_entry *
923 nf_queue_entry_dup(struct nf_queue_entry *e)
924 {
925 	struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
926 
927 	if (!entry)
928 		return NULL;
929 
930 	if (nf_queue_entry_get_refs(entry))
931 		return entry;
932 
933 	kfree(entry);
934 	return NULL;
935 }
936 
937 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
938 /* When called from bridge netfilter, skb->data must point to MAC header
939  * before calling skb_gso_segment(). Else, original MAC header is lost
940  * and segmented skbs will be sent to wrong destination.
941  */
942 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
943 {
944 	if (nf_bridge_info_get(skb))
945 		__skb_push(skb, skb->network_header - skb->mac_header);
946 }
947 
948 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
949 {
950 	if (nf_bridge_info_get(skb))
951 		__skb_pull(skb, skb->network_header - skb->mac_header);
952 }
953 #else
954 #define nf_bridge_adjust_skb_data(s) do {} while (0)
955 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
956 #endif
957 
958 static int
959 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
960 			   struct sk_buff *skb, struct nf_queue_entry *entry)
961 {
962 	int ret = -ENOMEM;
963 	struct nf_queue_entry *entry_seg;
964 
965 	nf_bridge_adjust_segmented_data(skb);
966 
967 	if (skb->next == NULL) { /* last packet, no need to copy entry */
968 		struct sk_buff *gso_skb = entry->skb;
969 		entry->skb = skb;
970 		ret = __nfqnl_enqueue_packet(net, queue, entry);
971 		if (ret)
972 			entry->skb = gso_skb;
973 		return ret;
974 	}
975 
976 	skb_mark_not_on_list(skb);
977 
978 	entry_seg = nf_queue_entry_dup(entry);
979 	if (entry_seg) {
980 		entry_seg->skb = skb;
981 		ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
982 		if (ret)
983 			nf_queue_entry_free(entry_seg);
984 	}
985 	return ret;
986 }
987 
988 static int
989 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
990 {
991 	unsigned int queued;
992 	struct nfqnl_instance *queue;
993 	struct sk_buff *skb, *segs, *nskb;
994 	int err = -ENOBUFS;
995 	struct net *net = entry->state.net;
996 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
997 
998 	/* rcu_read_lock()ed by nf_hook_thresh */
999 	queue = instance_lookup(q, queuenum);
1000 	if (!queue)
1001 		return -ESRCH;
1002 
1003 	if (queue->copy_mode == NFQNL_COPY_NONE)
1004 		return -EINVAL;
1005 
1006 	skb = entry->skb;
1007 
1008 	switch (entry->state.pf) {
1009 	case NFPROTO_IPV4:
1010 		skb->protocol = htons(ETH_P_IP);
1011 		break;
1012 	case NFPROTO_IPV6:
1013 		skb->protocol = htons(ETH_P_IPV6);
1014 		break;
1015 	}
1016 
1017 	if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
1018 		return __nfqnl_enqueue_packet(net, queue, entry);
1019 
1020 	nf_bridge_adjust_skb_data(skb);
1021 	segs = skb_gso_segment(skb, 0);
1022 	/* Does not use PTR_ERR to limit the number of error codes that can be
1023 	 * returned by nf_queue.  For instance, callers rely on -ESRCH to
1024 	 * mean 'ignore this hook'.
1025 	 */
1026 	if (IS_ERR_OR_NULL(segs))
1027 		goto out_err;
1028 	queued = 0;
1029 	err = 0;
1030 	skb_list_walk_safe(segs, segs, nskb) {
1031 		if (err == 0)
1032 			err = __nfqnl_enqueue_packet_gso(net, queue,
1033 							segs, entry);
1034 		if (err == 0)
1035 			queued++;
1036 		else
1037 			kfree_skb(segs);
1038 	}
1039 
1040 	if (queued) {
1041 		if (err) /* some segments are already queued */
1042 			nf_queue_entry_free(entry);
1043 		kfree_skb(skb);
1044 		return 0;
1045 	}
1046  out_err:
1047 	nf_bridge_adjust_segmented_data(skb);
1048 	return err;
1049 }
1050 
1051 static int
1052 nfqnl_mangle(void *data, unsigned int data_len, struct nf_queue_entry *e, int diff)
1053 {
1054 	struct sk_buff *nskb;
1055 
1056 	if (diff < 0) {
1057 		unsigned int min_len = skb_transport_offset(e->skb);
1058 
1059 		if (data_len < min_len)
1060 			return -EINVAL;
1061 
1062 		if (pskb_trim(e->skb, data_len))
1063 			return -ENOMEM;
1064 	} else if (diff > 0) {
1065 		if (data_len > 0xFFFF)
1066 			return -EINVAL;
1067 		if (diff > skb_tailroom(e->skb)) {
1068 			nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
1069 					       diff, GFP_ATOMIC);
1070 			if (!nskb)
1071 				return -ENOMEM;
1072 			kfree_skb(e->skb);
1073 			e->skb = nskb;
1074 		}
1075 		skb_put(e->skb, diff);
1076 	}
1077 	if (skb_ensure_writable(e->skb, data_len))
1078 		return -ENOMEM;
1079 	skb_copy_to_linear_data(e->skb, data, data_len);
1080 	e->skb->ip_summed = CHECKSUM_NONE;
1081 	return 0;
1082 }
1083 
1084 static int
1085 nfqnl_set_mode(struct nfqnl_instance *queue,
1086 	       unsigned char mode, unsigned int range)
1087 {
1088 	int status = 0;
1089 
1090 	spin_lock_bh(&queue->lock);
1091 	switch (mode) {
1092 	case NFQNL_COPY_NONE:
1093 	case NFQNL_COPY_META:
1094 		queue->copy_mode = mode;
1095 		queue->copy_range = 0;
1096 		break;
1097 
1098 	case NFQNL_COPY_PACKET:
1099 		queue->copy_mode = mode;
1100 		if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
1101 			queue->copy_range = NFQNL_MAX_COPY_RANGE;
1102 		else
1103 			queue->copy_range = range;
1104 		break;
1105 
1106 	default:
1107 		status = -EINVAL;
1108 
1109 	}
1110 	spin_unlock_bh(&queue->lock);
1111 
1112 	return status;
1113 }
1114 
1115 static int
1116 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
1117 {
1118 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
1119 	int physinif, physoutif;
1120 
1121 	physinif = nf_bridge_get_physinif(entry->skb);
1122 	physoutif = nf_bridge_get_physoutif(entry->skb);
1123 
1124 	if (physinif == ifindex || physoutif == ifindex)
1125 		return 1;
1126 #endif
1127 	if (entry->state.in)
1128 		if (entry->state.in->ifindex == ifindex)
1129 			return 1;
1130 	if (entry->state.out)
1131 		if (entry->state.out->ifindex == ifindex)
1132 			return 1;
1133 
1134 	return 0;
1135 }
1136 
1137 /* drop all packets with either indev or outdev == ifindex from all queue
1138  * instances */
1139 static void
1140 nfqnl_dev_drop(struct net *net, int ifindex)
1141 {
1142 	int i;
1143 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1144 
1145 	rcu_read_lock();
1146 
1147 	for (i = 0; i < INSTANCE_BUCKETS; i++) {
1148 		struct nfqnl_instance *inst;
1149 		struct hlist_head *head = &q->instance_table[i];
1150 
1151 		hlist_for_each_entry_rcu(inst, head, hlist)
1152 			nfqnl_flush(inst, dev_cmp, ifindex);
1153 	}
1154 
1155 	rcu_read_unlock();
1156 }
1157 
1158 static int
1159 nfqnl_rcv_dev_event(struct notifier_block *this,
1160 		    unsigned long event, void *ptr)
1161 {
1162 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1163 
1164 	/* Drop any packets associated with the downed device */
1165 	if (event == NETDEV_DOWN)
1166 		nfqnl_dev_drop(dev_net(dev), dev->ifindex);
1167 	return NOTIFY_DONE;
1168 }
1169 
1170 static struct notifier_block nfqnl_dev_notifier = {
1171 	.notifier_call	= nfqnl_rcv_dev_event,
1172 };
1173 
1174 static void nfqnl_nf_hook_drop(struct net *net)
1175 {
1176 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1177 	int i;
1178 
1179 	/* This function is also called on net namespace error unwind,
1180 	 * when pernet_ops->init() failed and ->exit() functions of the
1181 	 * previous pernet_ops gets called.
1182 	 *
1183 	 * This may result in a call to nfqnl_nf_hook_drop() before
1184 	 * struct nfnl_queue_net was allocated.
1185 	 */
1186 	if (!q)
1187 		return;
1188 
1189 	for (i = 0; i < INSTANCE_BUCKETS; i++) {
1190 		struct nfqnl_instance *inst;
1191 		struct hlist_head *head = &q->instance_table[i];
1192 
1193 		hlist_for_each_entry_rcu(inst, head, hlist)
1194 			nfqnl_flush(inst, NULL, 0);
1195 	}
1196 }
1197 
1198 static int
1199 nfqnl_rcv_nl_event(struct notifier_block *this,
1200 		   unsigned long event, void *ptr)
1201 {
1202 	struct netlink_notify *n = ptr;
1203 	struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
1204 
1205 	if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
1206 		int i;
1207 
1208 		/* destroy all instances for this portid */
1209 		spin_lock(&q->instances_lock);
1210 		for (i = 0; i < INSTANCE_BUCKETS; i++) {
1211 			struct hlist_node *t2;
1212 			struct nfqnl_instance *inst;
1213 			struct hlist_head *head = &q->instance_table[i];
1214 
1215 			hlist_for_each_entry_safe(inst, t2, head, hlist) {
1216 				if (n->portid == inst->peer_portid)
1217 					__instance_destroy(inst);
1218 			}
1219 		}
1220 		spin_unlock(&q->instances_lock);
1221 	}
1222 	return NOTIFY_DONE;
1223 }
1224 
1225 static struct notifier_block nfqnl_rtnl_notifier = {
1226 	.notifier_call	= nfqnl_rcv_nl_event,
1227 };
1228 
1229 static const struct nla_policy nfqa_vlan_policy[NFQA_VLAN_MAX + 1] = {
1230 	[NFQA_VLAN_TCI]		= { .type = NLA_U16},
1231 	[NFQA_VLAN_PROTO]	= { .type = NLA_U16},
1232 };
1233 
1234 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
1235 	[NFQA_VERDICT_HDR]	= { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1236 	[NFQA_MARK]		= { .type = NLA_U32 },
1237 	[NFQA_PAYLOAD]		= { .type = NLA_UNSPEC },
1238 	[NFQA_CT]		= { .type = NLA_UNSPEC },
1239 	[NFQA_EXP]		= { .type = NLA_UNSPEC },
1240 	[NFQA_VLAN]		= { .type = NLA_NESTED },
1241 	[NFQA_PRIORITY]		= { .type = NLA_U32 },
1242 };
1243 
1244 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
1245 	[NFQA_VERDICT_HDR]	= { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1246 	[NFQA_MARK]		= { .type = NLA_U32 },
1247 	[NFQA_PRIORITY]		= { .type = NLA_U32 },
1248 };
1249 
1250 static struct nfqnl_instance *
1251 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
1252 {
1253 	struct nfqnl_instance *queue;
1254 
1255 	queue = instance_lookup(q, queue_num);
1256 	if (!queue)
1257 		return ERR_PTR(-ENODEV);
1258 
1259 	if (queue->peer_portid != nlportid)
1260 		return ERR_PTR(-EPERM);
1261 
1262 	return queue;
1263 }
1264 
1265 static struct nfqnl_msg_verdict_hdr*
1266 verdicthdr_get(const struct nlattr * const nfqa[])
1267 {
1268 	struct nfqnl_msg_verdict_hdr *vhdr;
1269 	unsigned int verdict;
1270 
1271 	if (!nfqa[NFQA_VERDICT_HDR])
1272 		return NULL;
1273 
1274 	vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
1275 	verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
1276 	if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
1277 		return NULL;
1278 	return vhdr;
1279 }
1280 
1281 static int nfq_id_after(unsigned int id, unsigned int max)
1282 {
1283 	return (int)(id - max) > 0;
1284 }
1285 
1286 static int nfqnl_recv_verdict_batch(struct sk_buff *skb,
1287 				    const struct nfnl_info *info,
1288 				    const struct nlattr * const nfqa[])
1289 {
1290 	struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1291 	u16 queue_num = ntohs(info->nfmsg->res_id);
1292 	struct nf_queue_entry *entry, *tmp;
1293 	struct nfqnl_msg_verdict_hdr *vhdr;
1294 	struct nfqnl_instance *queue;
1295 	unsigned int verdict, maxid;
1296 	LIST_HEAD(batch_list);
1297 
1298 	queue = verdict_instance_lookup(q, queue_num,
1299 					NETLINK_CB(skb).portid);
1300 	if (IS_ERR(queue))
1301 		return PTR_ERR(queue);
1302 
1303 	vhdr = verdicthdr_get(nfqa);
1304 	if (!vhdr)
1305 		return -EINVAL;
1306 
1307 	verdict = ntohl(vhdr->verdict);
1308 	maxid = ntohl(vhdr->id);
1309 
1310 	spin_lock_bh(&queue->lock);
1311 
1312 	list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
1313 		if (nfq_id_after(entry->id, maxid))
1314 			break;
1315 		__dequeue_entry(queue, entry);
1316 		list_add_tail(&entry->list, &batch_list);
1317 	}
1318 
1319 	spin_unlock_bh(&queue->lock);
1320 
1321 	if (list_empty(&batch_list))
1322 		return -ENOENT;
1323 
1324 	list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1325 		if (nfqa[NFQA_MARK])
1326 			entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1327 
1328 		if (nfqa[NFQA_PRIORITY])
1329 			entry->skb->priority = ntohl(nla_get_be32(nfqa[NFQA_PRIORITY]));
1330 
1331 		nfqnl_reinject(entry, verdict);
1332 	}
1333 	return 0;
1334 }
1335 
1336 static struct nf_conn *nfqnl_ct_parse(const struct nfnl_ct_hook *nfnl_ct,
1337 				      const struct nlmsghdr *nlh,
1338 				      const struct nlattr * const nfqa[],
1339 				      struct nf_queue_entry *entry,
1340 				      enum ip_conntrack_info *ctinfo)
1341 {
1342 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1343 	struct nf_conn *ct;
1344 
1345 	ct = nf_ct_get(entry->skb, ctinfo);
1346 	if (ct == NULL)
1347 		return NULL;
1348 
1349 	if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1350 		return NULL;
1351 
1352 	if (nfqa[NFQA_EXP])
1353 		nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1354 				      NETLINK_CB(entry->skb).portid,
1355 				      nlmsg_report(nlh));
1356 	return ct;
1357 #else
1358 	return NULL;
1359 #endif
1360 }
1361 
1362 static int nfqa_parse_bridge(struct nf_queue_entry *entry,
1363 			     const struct nlattr * const nfqa[])
1364 {
1365 	if (nfqa[NFQA_VLAN]) {
1366 		struct nlattr *tb[NFQA_VLAN_MAX + 1];
1367 		int err;
1368 
1369 		err = nla_parse_nested_deprecated(tb, NFQA_VLAN_MAX,
1370 						  nfqa[NFQA_VLAN],
1371 						  nfqa_vlan_policy, NULL);
1372 		if (err < 0)
1373 			return err;
1374 
1375 		if (!tb[NFQA_VLAN_TCI] || !tb[NFQA_VLAN_PROTO])
1376 			return -EINVAL;
1377 
1378 		__vlan_hwaccel_put_tag(entry->skb,
1379 			nla_get_be16(tb[NFQA_VLAN_PROTO]),
1380 			ntohs(nla_get_be16(tb[NFQA_VLAN_TCI])));
1381 	}
1382 
1383 	if (nfqa[NFQA_L2HDR]) {
1384 		int mac_header_len = entry->skb->network_header -
1385 			entry->skb->mac_header;
1386 
1387 		if (mac_header_len != nla_len(nfqa[NFQA_L2HDR]))
1388 			return -EINVAL;
1389 		else if (mac_header_len > 0)
1390 			memcpy(skb_mac_header(entry->skb),
1391 			       nla_data(nfqa[NFQA_L2HDR]),
1392 			       mac_header_len);
1393 	}
1394 
1395 	return 0;
1396 }
1397 
1398 static int nfqnl_recv_verdict(struct sk_buff *skb, const struct nfnl_info *info,
1399 			      const struct nlattr * const nfqa[])
1400 {
1401 	struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1402 	u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1403 	const struct nfnl_ct_hook *nfnl_ct;
1404 	struct nfqnl_msg_verdict_hdr *vhdr;
1405 	enum ip_conntrack_info ctinfo;
1406 	struct nfqnl_instance *queue;
1407 	struct nf_queue_entry *entry;
1408 	struct nf_conn *ct = NULL;
1409 	unsigned int verdict;
1410 	int err;
1411 
1412 	queue = verdict_instance_lookup(q, queue_num,
1413 					NETLINK_CB(skb).portid);
1414 	if (IS_ERR(queue))
1415 		return PTR_ERR(queue);
1416 
1417 	vhdr = verdicthdr_get(nfqa);
1418 	if (!vhdr)
1419 		return -EINVAL;
1420 
1421 	verdict = ntohl(vhdr->verdict);
1422 
1423 	entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1424 	if (entry == NULL)
1425 		return -ENOENT;
1426 
1427 	/* rcu lock already held from nfnl->call_rcu. */
1428 	nfnl_ct = rcu_dereference(nfnl_ct_hook);
1429 
1430 	if (nfqa[NFQA_CT]) {
1431 		if (nfnl_ct != NULL)
1432 			ct = nfqnl_ct_parse(nfnl_ct, info->nlh, nfqa, entry,
1433 					    &ctinfo);
1434 	}
1435 
1436 	if (entry->state.pf == PF_BRIDGE) {
1437 		err = nfqa_parse_bridge(entry, nfqa);
1438 		if (err < 0)
1439 			return err;
1440 	}
1441 
1442 	if (nfqa[NFQA_PAYLOAD]) {
1443 		u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1444 		int diff = payload_len - entry->skb->len;
1445 
1446 		if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1447 				 payload_len, entry, diff) < 0)
1448 			verdict = NF_DROP;
1449 
1450 		if (ct && diff)
1451 			nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1452 	}
1453 
1454 	if (nfqa[NFQA_MARK])
1455 		entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1456 
1457 	if (nfqa[NFQA_PRIORITY])
1458 		entry->skb->priority = ntohl(nla_get_be32(nfqa[NFQA_PRIORITY]));
1459 
1460 	nfqnl_reinject(entry, verdict);
1461 	return 0;
1462 }
1463 
1464 static int nfqnl_recv_unsupp(struct sk_buff *skb, const struct nfnl_info *info,
1465 			     const struct nlattr * const cda[])
1466 {
1467 	return -ENOTSUPP;
1468 }
1469 
1470 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1471 	[NFQA_CFG_CMD]		= { .len = sizeof(struct nfqnl_msg_config_cmd) },
1472 	[NFQA_CFG_PARAMS]	= { .len = sizeof(struct nfqnl_msg_config_params) },
1473 	[NFQA_CFG_QUEUE_MAXLEN]	= { .type = NLA_U32 },
1474 	[NFQA_CFG_MASK]		= { .type = NLA_U32 },
1475 	[NFQA_CFG_FLAGS]	= { .type = NLA_U32 },
1476 };
1477 
1478 static const struct nf_queue_handler nfqh = {
1479 	.outfn		= nfqnl_enqueue_packet,
1480 	.nf_hook_drop	= nfqnl_nf_hook_drop,
1481 };
1482 
1483 static int nfqnl_recv_config(struct sk_buff *skb, const struct nfnl_info *info,
1484 			     const struct nlattr * const nfqa[])
1485 {
1486 	struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1487 	u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1488 	struct nfqnl_msg_config_cmd *cmd = NULL;
1489 	struct nfqnl_instance *queue;
1490 	__u32 flags = 0, mask = 0;
1491 	int ret = 0;
1492 
1493 	if (nfqa[NFQA_CFG_CMD]) {
1494 		cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1495 
1496 		/* Obsolete commands without queue context */
1497 		switch (cmd->command) {
1498 		case NFQNL_CFG_CMD_PF_BIND: return 0;
1499 		case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1500 		}
1501 	}
1502 
1503 	/* Check if we support these flags in first place, dependencies should
1504 	 * be there too not to break atomicity.
1505 	 */
1506 	if (nfqa[NFQA_CFG_FLAGS]) {
1507 		if (!nfqa[NFQA_CFG_MASK]) {
1508 			/* A mask is needed to specify which flags are being
1509 			 * changed.
1510 			 */
1511 			return -EINVAL;
1512 		}
1513 
1514 		flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1515 		mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1516 
1517 		if (flags >= NFQA_CFG_F_MAX)
1518 			return -EOPNOTSUPP;
1519 
1520 #if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1521 		if (flags & mask & NFQA_CFG_F_SECCTX)
1522 			return -EOPNOTSUPP;
1523 #endif
1524 		if ((flags & mask & NFQA_CFG_F_CONNTRACK) &&
1525 		    !rcu_access_pointer(nfnl_ct_hook)) {
1526 #ifdef CONFIG_MODULES
1527 			nfnl_unlock(NFNL_SUBSYS_QUEUE);
1528 			request_module("ip_conntrack_netlink");
1529 			nfnl_lock(NFNL_SUBSYS_QUEUE);
1530 			if (rcu_access_pointer(nfnl_ct_hook))
1531 				return -EAGAIN;
1532 #endif
1533 			return -EOPNOTSUPP;
1534 		}
1535 	}
1536 
1537 	rcu_read_lock();
1538 	queue = instance_lookup(q, queue_num);
1539 	if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1540 		ret = -EPERM;
1541 		goto err_out_unlock;
1542 	}
1543 
1544 	if (cmd != NULL) {
1545 		switch (cmd->command) {
1546 		case NFQNL_CFG_CMD_BIND:
1547 			if (queue) {
1548 				ret = -EBUSY;
1549 				goto err_out_unlock;
1550 			}
1551 			queue = instance_create(q, queue_num,
1552 						NETLINK_CB(skb).portid);
1553 			if (IS_ERR(queue)) {
1554 				ret = PTR_ERR(queue);
1555 				goto err_out_unlock;
1556 			}
1557 			break;
1558 		case NFQNL_CFG_CMD_UNBIND:
1559 			if (!queue) {
1560 				ret = -ENODEV;
1561 				goto err_out_unlock;
1562 			}
1563 			instance_destroy(q, queue);
1564 			goto err_out_unlock;
1565 		case NFQNL_CFG_CMD_PF_BIND:
1566 		case NFQNL_CFG_CMD_PF_UNBIND:
1567 			break;
1568 		default:
1569 			ret = -ENOTSUPP;
1570 			goto err_out_unlock;
1571 		}
1572 	}
1573 
1574 	if (!queue) {
1575 		ret = -ENODEV;
1576 		goto err_out_unlock;
1577 	}
1578 
1579 	if (nfqa[NFQA_CFG_PARAMS]) {
1580 		struct nfqnl_msg_config_params *params =
1581 			nla_data(nfqa[NFQA_CFG_PARAMS]);
1582 
1583 		nfqnl_set_mode(queue, params->copy_mode,
1584 				ntohl(params->copy_range));
1585 	}
1586 
1587 	if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1588 		__be32 *queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1589 
1590 		spin_lock_bh(&queue->lock);
1591 		queue->queue_maxlen = ntohl(*queue_maxlen);
1592 		spin_unlock_bh(&queue->lock);
1593 	}
1594 
1595 	if (nfqa[NFQA_CFG_FLAGS]) {
1596 		spin_lock_bh(&queue->lock);
1597 		queue->flags &= ~mask;
1598 		queue->flags |= flags & mask;
1599 		spin_unlock_bh(&queue->lock);
1600 	}
1601 
1602 err_out_unlock:
1603 	rcu_read_unlock();
1604 	return ret;
1605 }
1606 
1607 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1608 	[NFQNL_MSG_PACKET]	= {
1609 		.call		= nfqnl_recv_unsupp,
1610 		.type		= NFNL_CB_RCU,
1611 		.attr_count	= NFQA_MAX,
1612 	},
1613 	[NFQNL_MSG_VERDICT]	= {
1614 		.call		= nfqnl_recv_verdict,
1615 		.type		= NFNL_CB_RCU,
1616 		.attr_count	= NFQA_MAX,
1617 		.policy		= nfqa_verdict_policy
1618 	},
1619 	[NFQNL_MSG_CONFIG]	= {
1620 		.call		= nfqnl_recv_config,
1621 		.type		= NFNL_CB_MUTEX,
1622 		.attr_count	= NFQA_CFG_MAX,
1623 		.policy		= nfqa_cfg_policy
1624 	},
1625 	[NFQNL_MSG_VERDICT_BATCH] = {
1626 		.call		= nfqnl_recv_verdict_batch,
1627 		.type		= NFNL_CB_RCU,
1628 		.attr_count	= NFQA_MAX,
1629 		.policy		= nfqa_verdict_batch_policy
1630 	},
1631 };
1632 
1633 static const struct nfnetlink_subsystem nfqnl_subsys = {
1634 	.name		= "nf_queue",
1635 	.subsys_id	= NFNL_SUBSYS_QUEUE,
1636 	.cb_count	= NFQNL_MSG_MAX,
1637 	.cb		= nfqnl_cb,
1638 };
1639 
1640 #ifdef CONFIG_PROC_FS
1641 struct iter_state {
1642 	struct seq_net_private p;
1643 	unsigned int bucket;
1644 };
1645 
1646 static struct hlist_node *get_first(struct seq_file *seq)
1647 {
1648 	struct iter_state *st = seq->private;
1649 	struct net *net;
1650 	struct nfnl_queue_net *q;
1651 
1652 	if (!st)
1653 		return NULL;
1654 
1655 	net = seq_file_net(seq);
1656 	q = nfnl_queue_pernet(net);
1657 	for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1658 		if (!hlist_empty(&q->instance_table[st->bucket]))
1659 			return q->instance_table[st->bucket].first;
1660 	}
1661 	return NULL;
1662 }
1663 
1664 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1665 {
1666 	struct iter_state *st = seq->private;
1667 	struct net *net = seq_file_net(seq);
1668 
1669 	h = h->next;
1670 	while (!h) {
1671 		struct nfnl_queue_net *q;
1672 
1673 		if (++st->bucket >= INSTANCE_BUCKETS)
1674 			return NULL;
1675 
1676 		q = nfnl_queue_pernet(net);
1677 		h = q->instance_table[st->bucket].first;
1678 	}
1679 	return h;
1680 }
1681 
1682 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1683 {
1684 	struct hlist_node *head;
1685 	head = get_first(seq);
1686 
1687 	if (head)
1688 		while (pos && (head = get_next(seq, head)))
1689 			pos--;
1690 	return pos ? NULL : head;
1691 }
1692 
1693 static void *seq_start(struct seq_file *s, loff_t *pos)
1694 	__acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1695 {
1696 	spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1697 	return get_idx(s, *pos);
1698 }
1699 
1700 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1701 {
1702 	(*pos)++;
1703 	return get_next(s, v);
1704 }
1705 
1706 static void seq_stop(struct seq_file *s, void *v)
1707 	__releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1708 {
1709 	spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1710 }
1711 
1712 static int seq_show(struct seq_file *s, void *v)
1713 {
1714 	const struct nfqnl_instance *inst = v;
1715 
1716 	seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1717 		   inst->queue_num,
1718 		   inst->peer_portid, inst->queue_total,
1719 		   inst->copy_mode, inst->copy_range,
1720 		   inst->queue_dropped, inst->queue_user_dropped,
1721 		   inst->id_sequence, 1);
1722 	return 0;
1723 }
1724 
1725 static const struct seq_operations nfqnl_seq_ops = {
1726 	.start	= seq_start,
1727 	.next	= seq_next,
1728 	.stop	= seq_stop,
1729 	.show	= seq_show,
1730 };
1731 #endif /* PROC_FS */
1732 
1733 static int __net_init nfnl_queue_net_init(struct net *net)
1734 {
1735 	unsigned int i;
1736 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1737 
1738 	for (i = 0; i < INSTANCE_BUCKETS; i++)
1739 		INIT_HLIST_HEAD(&q->instance_table[i]);
1740 
1741 	spin_lock_init(&q->instances_lock);
1742 
1743 #ifdef CONFIG_PROC_FS
1744 	if (!proc_create_net("nfnetlink_queue", 0440, net->nf.proc_netfilter,
1745 			&nfqnl_seq_ops, sizeof(struct iter_state)))
1746 		return -ENOMEM;
1747 #endif
1748 	return 0;
1749 }
1750 
1751 static void __net_exit nfnl_queue_net_exit(struct net *net)
1752 {
1753 	struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1754 	unsigned int i;
1755 
1756 #ifdef CONFIG_PROC_FS
1757 	remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1758 #endif
1759 	for (i = 0; i < INSTANCE_BUCKETS; i++)
1760 		WARN_ON_ONCE(!hlist_empty(&q->instance_table[i]));
1761 }
1762 
1763 static struct pernet_operations nfnl_queue_net_ops = {
1764 	.init		= nfnl_queue_net_init,
1765 	.exit		= nfnl_queue_net_exit,
1766 	.id		= &nfnl_queue_net_id,
1767 	.size		= sizeof(struct nfnl_queue_net),
1768 };
1769 
1770 static int __init nfnetlink_queue_init(void)
1771 {
1772 	int status;
1773 
1774 	status = register_pernet_subsys(&nfnl_queue_net_ops);
1775 	if (status < 0) {
1776 		pr_err("failed to register pernet ops\n");
1777 		goto out;
1778 	}
1779 
1780 	netlink_register_notifier(&nfqnl_rtnl_notifier);
1781 	status = nfnetlink_subsys_register(&nfqnl_subsys);
1782 	if (status < 0) {
1783 		pr_err("failed to create netlink socket\n");
1784 		goto cleanup_netlink_notifier;
1785 	}
1786 
1787 	status = register_netdevice_notifier(&nfqnl_dev_notifier);
1788 	if (status < 0) {
1789 		pr_err("failed to register netdevice notifier\n");
1790 		goto cleanup_netlink_subsys;
1791 	}
1792 
1793 	nf_register_queue_handler(&nfqh);
1794 
1795 	return status;
1796 
1797 cleanup_netlink_subsys:
1798 	nfnetlink_subsys_unregister(&nfqnl_subsys);
1799 cleanup_netlink_notifier:
1800 	netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1801 	unregister_pernet_subsys(&nfnl_queue_net_ops);
1802 out:
1803 	return status;
1804 }
1805 
1806 static void __exit nfnetlink_queue_fini(void)
1807 {
1808 	nf_unregister_queue_handler();
1809 	unregister_netdevice_notifier(&nfqnl_dev_notifier);
1810 	nfnetlink_subsys_unregister(&nfqnl_subsys);
1811 	netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1812 	unregister_pernet_subsys(&nfnl_queue_net_ops);
1813 
1814 	rcu_barrier(); /* Wait for completion of call_rcu()'s */
1815 }
1816 
1817 MODULE_DESCRIPTION("netfilter packet queue handler");
1818 MODULE_AUTHOR("Harald Welte <[email protected]>");
1819 MODULE_LICENSE("GPL");
1820 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1821 
1822 module_init(nfnetlink_queue_init);
1823 module_exit(nfnetlink_queue_fini);
1824