xref: /linux/net/ipv4/ip_input.c (revision 6fdcba32711044c35c0e1b094cbd8f3f0b4472c9)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		The Internet Protocol (IP) module.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Donald Becker, <becker@super.org>
12  *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
13  *		Richard Underwood
14  *		Stefan Becker, <stefanb@yello.ping.de>
15  *		Jorge Cwik, <jorge@laser.satlink.net>
16  *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
17  *
18  * Fixes:
19  *		Alan Cox	:	Commented a couple of minor bits of surplus code
20  *		Alan Cox	:	Undefining IP_FORWARD doesn't include the code
21  *					(just stops a compiler warning).
22  *		Alan Cox	:	Frames with >=MAX_ROUTE record routes, strict routes or loose routes
23  *					are junked rather than corrupting things.
24  *		Alan Cox	:	Frames to bad broadcast subnets are dumped
25  *					We used to process them non broadcast and
26  *					boy could that cause havoc.
27  *		Alan Cox	:	ip_forward sets the free flag on the
28  *					new frame it queues. Still crap because
29  *					it copies the frame but at least it
30  *					doesn't eat memory too.
31  *		Alan Cox	:	Generic queue code and memory fixes.
32  *		Fred Van Kempen :	IP fragment support (borrowed from NET2E)
33  *		Gerhard Koerting:	Forward fragmented frames correctly.
34  *		Gerhard Koerting: 	Fixes to my fix of the above 8-).
35  *		Gerhard Koerting:	IP interface addressing fix.
36  *		Linus Torvalds	:	More robustness checks
37  *		Alan Cox	:	Even more checks: Still not as robust as it ought to be
38  *		Alan Cox	:	Save IP header pointer for later
39  *		Alan Cox	:	ip option setting
40  *		Alan Cox	:	Use ip_tos/ip_ttl settings
41  *		Alan Cox	:	Fragmentation bogosity removed
42  *					(Thanks to Mark.Bush@prg.ox.ac.uk)
43  *		Dmitry Gorodchanin :	Send of a raw packet crash fix.
44  *		Alan Cox	:	Silly ip bug when an overlength
45  *					fragment turns up. Now frees the
46  *					queue.
47  *		Linus Torvalds/ :	Memory leakage on fragmentation
48  *		Alan Cox	:	handling.
49  *		Gerhard Koerting:	Forwarding uses IP priority hints
50  *		Teemu Rantanen	:	Fragment problems.
51  *		Alan Cox	:	General cleanup, comments and reformat
52  *		Alan Cox	:	SNMP statistics
53  *		Alan Cox	:	BSD address rule semantics. Also see
54  *					UDP as there is a nasty checksum issue
55  *					if you do things the wrong way.
56  *		Alan Cox	:	Always defrag, moved IP_FORWARD to the config.in file
57  *		Alan Cox	: 	IP options adjust sk->priority.
58  *		Pedro Roque	:	Fix mtu/length error in ip_forward.
59  *		Alan Cox	:	Avoid ip_chk_addr when possible.
60  *	Richard Underwood	:	IP multicasting.
61  *		Alan Cox	:	Cleaned up multicast handlers.
62  *		Alan Cox	:	RAW sockets demultiplex in the BSD style.
63  *		Gunther Mayer	:	Fix the SNMP reporting typo
64  *		Alan Cox	:	Always in group 224.0.0.1
65  *	Pauline Middelink	:	Fast ip_checksum update when forwarding
66  *					Masquerading support.
67  *		Alan Cox	:	Multicast loopback error for 224.0.0.1
68  *		Alan Cox	:	IP_MULTICAST_LOOP option.
69  *		Alan Cox	:	Use notifiers.
70  *		Bjorn Ekwall	:	Removed ip_csum (from slhc.c too)
71  *		Bjorn Ekwall	:	Moved ip_fast_csum to ip.h (inline!)
72  *		Stefan Becker   :       Send out ICMP HOST REDIRECT
73  *	Arnt Gulbrandsen	:	ip_build_xmit
74  *		Alan Cox	:	Per socket routing cache
75  *		Alan Cox	:	Fixed routing cache, added header cache.
76  *		Alan Cox	:	Loopback didn't work right in original ip_build_xmit - fixed it.
77  *		Alan Cox	:	Only send ICMP_REDIRECT if src/dest are the same net.
78  *		Alan Cox	:	Incoming IP option handling.
79  *		Alan Cox	:	Set saddr on raw output frames as per BSD.
80  *		Alan Cox	:	Stopped broadcast source route explosions.
81  *		Alan Cox	:	Can disable source routing
82  *		Takeshi Sone    :	Masquerading didn't work.
83  *	Dave Bonn,Alan Cox	:	Faster IP forwarding whenever possible.
84  *		Alan Cox	:	Memory leaks, tramples, misc debugging.
85  *		Alan Cox	:	Fixed multicast (by popular demand 8))
86  *		Alan Cox	:	Fixed forwarding (by even more popular demand 8))
87  *		Alan Cox	:	Fixed SNMP statistics [I think]
88  *	Gerhard Koerting	:	IP fragmentation forwarding fix
89  *		Alan Cox	:	Device lock against page fault.
90  *		Alan Cox	:	IP_HDRINCL facility.
91  *	Werner Almesberger	:	Zero fragment bug
92  *		Alan Cox	:	RAW IP frame length bug
93  *		Alan Cox	:	Outgoing firewall on build_xmit
94  *		A.N.Kuznetsov	:	IP_OPTIONS support throughout the kernel
95  *		Alan Cox	:	Multicast routing hooks
96  *		Jos Vos		:	Do accounting *before* call_in_firewall
97  *	Willy Konynenberg	:	Transparent proxying support
98  *
99  * To Fix:
100  *		IP fragmentation wants rewriting cleanly. The RFC815 algorithm is much more efficient
101  *		and could be made very efficient with the addition of some virtual memory hacks to permit
102  *		the allocation of a buffer that can then be 'grown' by twiddling page tables.
103  *		Output fragmentation wants updating along with the buffer management to use a single
104  *		interleaved copy algorithm so that fragmenting has a one copy overhead. Actual packet
105  *		output should probably do its own fragmentation at the UDP/RAW layer. TCP shouldn't cause
106  *		fragmentation anyway.
107  */
108 
109 #define pr_fmt(fmt) "IPv4: " fmt
110 
111 #include <linux/module.h>
112 #include <linux/types.h>
113 #include <linux/kernel.h>
114 #include <linux/string.h>
115 #include <linux/errno.h>
116 #include <linux/slab.h>
117 
118 #include <linux/net.h>
119 #include <linux/socket.h>
120 #include <linux/sockios.h>
121 #include <linux/in.h>
122 #include <linux/inet.h>
123 #include <linux/inetdevice.h>
124 #include <linux/netdevice.h>
125 #include <linux/etherdevice.h>
126 #include <linux/indirect_call_wrapper.h>
127 
128 #include <net/snmp.h>
129 #include <net/ip.h>
130 #include <net/protocol.h>
131 #include <net/route.h>
132 #include <linux/skbuff.h>
133 #include <net/sock.h>
134 #include <net/arp.h>
135 #include <net/icmp.h>
136 #include <net/raw.h>
137 #include <net/checksum.h>
138 #include <net/inet_ecn.h>
139 #include <linux/netfilter_ipv4.h>
140 #include <net/xfrm.h>
141 #include <linux/mroute.h>
142 #include <linux/netlink.h>
143 #include <net/dst_metadata.h>
144 
145 /*
146  *	Process Router Attention IP option (RFC 2113)
147  */
148 bool ip_call_ra_chain(struct sk_buff *skb)
149 {
150 	struct ip_ra_chain *ra;
151 	u8 protocol = ip_hdr(skb)->protocol;
152 	struct sock *last = NULL;
153 	struct net_device *dev = skb->dev;
154 	struct net *net = dev_net(dev);
155 
156 	for (ra = rcu_dereference(net->ipv4.ra_chain); ra; ra = rcu_dereference(ra->next)) {
157 		struct sock *sk = ra->sk;
158 
159 		/* If socket is bound to an interface, only report
160 		 * the packet if it came  from that interface.
161 		 */
162 		if (sk && inet_sk(sk)->inet_num == protocol &&
163 		    (!sk->sk_bound_dev_if ||
164 		     sk->sk_bound_dev_if == dev->ifindex)) {
165 			if (ip_is_fragment(ip_hdr(skb))) {
166 				if (ip_defrag(net, skb, IP_DEFRAG_CALL_RA_CHAIN))
167 					return true;
168 			}
169 			if (last) {
170 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
171 				if (skb2)
172 					raw_rcv(last, skb2);
173 			}
174 			last = sk;
175 		}
176 	}
177 
178 	if (last) {
179 		raw_rcv(last, skb);
180 		return true;
181 	}
182 	return false;
183 }
184 
185 INDIRECT_CALLABLE_DECLARE(int udp_rcv(struct sk_buff *));
186 INDIRECT_CALLABLE_DECLARE(int tcp_v4_rcv(struct sk_buff *));
187 void ip_protocol_deliver_rcu(struct net *net, struct sk_buff *skb, int protocol)
188 {
189 	const struct net_protocol *ipprot;
190 	int raw, ret;
191 
192 resubmit:
193 	raw = raw_local_deliver(skb, protocol);
194 
195 	ipprot = rcu_dereference(inet_protos[protocol]);
196 	if (ipprot) {
197 		if (!ipprot->no_policy) {
198 			if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
199 				kfree_skb(skb);
200 				return;
201 			}
202 			nf_reset_ct(skb);
203 		}
204 		ret = INDIRECT_CALL_2(ipprot->handler, tcp_v4_rcv, udp_rcv,
205 				      skb);
206 		if (ret < 0) {
207 			protocol = -ret;
208 			goto resubmit;
209 		}
210 		__IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
211 	} else {
212 		if (!raw) {
213 			if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
214 				__IP_INC_STATS(net, IPSTATS_MIB_INUNKNOWNPROTOS);
215 				icmp_send(skb, ICMP_DEST_UNREACH,
216 					  ICMP_PROT_UNREACH, 0);
217 			}
218 			kfree_skb(skb);
219 		} else {
220 			__IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
221 			consume_skb(skb);
222 		}
223 	}
224 }
225 
226 static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
227 {
228 	__skb_pull(skb, skb_network_header_len(skb));
229 
230 	rcu_read_lock();
231 	ip_protocol_deliver_rcu(net, skb, ip_hdr(skb)->protocol);
232 	rcu_read_unlock();
233 
234 	return 0;
235 }
236 
237 /*
238  * 	Deliver IP Packets to the higher protocol layers.
239  */
240 int ip_local_deliver(struct sk_buff *skb)
241 {
242 	/*
243 	 *	Reassemble IP fragments.
244 	 */
245 	struct net *net = dev_net(skb->dev);
246 
247 	if (ip_is_fragment(ip_hdr(skb))) {
248 		if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
249 			return 0;
250 	}
251 
252 	return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
253 		       net, NULL, skb, skb->dev, NULL,
254 		       ip_local_deliver_finish);
255 }
256 
257 static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev)
258 {
259 	struct ip_options *opt;
260 	const struct iphdr *iph;
261 
262 	/* It looks as overkill, because not all
263 	   IP options require packet mangling.
264 	   But it is the easiest for now, especially taking
265 	   into account that combination of IP options
266 	   and running sniffer is extremely rare condition.
267 					      --ANK (980813)
268 	*/
269 	if (skb_cow(skb, skb_headroom(skb))) {
270 		__IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
271 		goto drop;
272 	}
273 
274 	iph = ip_hdr(skb);
275 	opt = &(IPCB(skb)->opt);
276 	opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
277 
278 	if (ip_options_compile(dev_net(dev), opt, skb)) {
279 		__IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
280 		goto drop;
281 	}
282 
283 	if (unlikely(opt->srr)) {
284 		struct in_device *in_dev = __in_dev_get_rcu(dev);
285 
286 		if (in_dev) {
287 			if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
288 				if (IN_DEV_LOG_MARTIANS(in_dev))
289 					net_info_ratelimited("source route option %pI4 -> %pI4\n",
290 							     &iph->saddr,
291 							     &iph->daddr);
292 				goto drop;
293 			}
294 		}
295 
296 		if (ip_options_rcv_srr(skb, dev))
297 			goto drop;
298 	}
299 
300 	return false;
301 drop:
302 	return true;
303 }
304 
305 static bool ip_can_use_hint(const struct sk_buff *skb, const struct iphdr *iph,
306 			    const struct sk_buff *hint)
307 {
308 	return hint && !skb_dst(skb) && ip_hdr(hint)->daddr == iph->daddr &&
309 	       ip_hdr(hint)->tos == iph->tos;
310 }
311 
312 INDIRECT_CALLABLE_DECLARE(int udp_v4_early_demux(struct sk_buff *));
313 INDIRECT_CALLABLE_DECLARE(int tcp_v4_early_demux(struct sk_buff *));
314 static int ip_rcv_finish_core(struct net *net, struct sock *sk,
315 			      struct sk_buff *skb, struct net_device *dev,
316 			      const struct sk_buff *hint)
317 {
318 	const struct iphdr *iph = ip_hdr(skb);
319 	int (*edemux)(struct sk_buff *skb);
320 	struct rtable *rt;
321 	int err;
322 
323 	if (ip_can_use_hint(skb, iph, hint)) {
324 		err = ip_route_use_hint(skb, iph->daddr, iph->saddr, iph->tos,
325 					dev, hint);
326 		if (unlikely(err))
327 			goto drop_error;
328 	}
329 
330 	if (net->ipv4.sysctl_ip_early_demux &&
331 	    !skb_dst(skb) &&
332 	    !skb->sk &&
333 	    !ip_is_fragment(iph)) {
334 		const struct net_protocol *ipprot;
335 		int protocol = iph->protocol;
336 
337 		ipprot = rcu_dereference(inet_protos[protocol]);
338 		if (ipprot && (edemux = READ_ONCE(ipprot->early_demux))) {
339 			err = INDIRECT_CALL_2(edemux, tcp_v4_early_demux,
340 					      udp_v4_early_demux, skb);
341 			if (unlikely(err))
342 				goto drop_error;
343 			/* must reload iph, skb->head might have changed */
344 			iph = ip_hdr(skb);
345 		}
346 	}
347 
348 	/*
349 	 *	Initialise the virtual path cache for the packet. It describes
350 	 *	how the packet travels inside Linux networking.
351 	 */
352 	if (!skb_valid_dst(skb)) {
353 		err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
354 					   iph->tos, dev);
355 		if (unlikely(err))
356 			goto drop_error;
357 	}
358 
359 #ifdef CONFIG_IP_ROUTE_CLASSID
360 	if (unlikely(skb_dst(skb)->tclassid)) {
361 		struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
362 		u32 idx = skb_dst(skb)->tclassid;
363 		st[idx&0xFF].o_packets++;
364 		st[idx&0xFF].o_bytes += skb->len;
365 		st[(idx>>16)&0xFF].i_packets++;
366 		st[(idx>>16)&0xFF].i_bytes += skb->len;
367 	}
368 #endif
369 
370 	if (iph->ihl > 5 && ip_rcv_options(skb, dev))
371 		goto drop;
372 
373 	rt = skb_rtable(skb);
374 	if (rt->rt_type == RTN_MULTICAST) {
375 		__IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
376 	} else if (rt->rt_type == RTN_BROADCAST) {
377 		__IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
378 	} else if (skb->pkt_type == PACKET_BROADCAST ||
379 		   skb->pkt_type == PACKET_MULTICAST) {
380 		struct in_device *in_dev = __in_dev_get_rcu(dev);
381 
382 		/* RFC 1122 3.3.6:
383 		 *
384 		 *   When a host sends a datagram to a link-layer broadcast
385 		 *   address, the IP destination address MUST be a legal IP
386 		 *   broadcast or IP multicast address.
387 		 *
388 		 *   A host SHOULD silently discard a datagram that is received
389 		 *   via a link-layer broadcast (see Section 2.4) but does not
390 		 *   specify an IP multicast or broadcast destination address.
391 		 *
392 		 * This doesn't explicitly say L2 *broadcast*, but broadcast is
393 		 * in a way a form of multicast and the most common use case for
394 		 * this is 802.11 protecting against cross-station spoofing (the
395 		 * so-called "hole-196" attack) so do it for both.
396 		 */
397 		if (in_dev &&
398 		    IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST))
399 			goto drop;
400 	}
401 
402 	return NET_RX_SUCCESS;
403 
404 drop:
405 	kfree_skb(skb);
406 	return NET_RX_DROP;
407 
408 drop_error:
409 	if (err == -EXDEV)
410 		__NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
411 	goto drop;
412 }
413 
414 static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
415 {
416 	struct net_device *dev = skb->dev;
417 	int ret;
418 
419 	/* if ingress device is enslaved to an L3 master device pass the
420 	 * skb to its handler for processing
421 	 */
422 	skb = l3mdev_ip_rcv(skb);
423 	if (!skb)
424 		return NET_RX_SUCCESS;
425 
426 	ret = ip_rcv_finish_core(net, sk, skb, dev, NULL);
427 	if (ret != NET_RX_DROP)
428 		ret = dst_input(skb);
429 	return ret;
430 }
431 
432 /*
433  * 	Main IP Receive routine.
434  */
435 static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net)
436 {
437 	const struct iphdr *iph;
438 	u32 len;
439 
440 	/* When the interface is in promisc. mode, drop all the crap
441 	 * that it receives, do not try to analyse it.
442 	 */
443 	if (skb->pkt_type == PACKET_OTHERHOST)
444 		goto drop;
445 
446 	__IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
447 
448 	skb = skb_share_check(skb, GFP_ATOMIC);
449 	if (!skb) {
450 		__IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
451 		goto out;
452 	}
453 
454 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
455 		goto inhdr_error;
456 
457 	iph = ip_hdr(skb);
458 
459 	/*
460 	 *	RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
461 	 *
462 	 *	Is the datagram acceptable?
463 	 *
464 	 *	1.	Length at least the size of an ip header
465 	 *	2.	Version of 4
466 	 *	3.	Checksums correctly. [Speed optimisation for later, skip loopback checksums]
467 	 *	4.	Doesn't have a bogus length
468 	 */
469 
470 	if (iph->ihl < 5 || iph->version != 4)
471 		goto inhdr_error;
472 
473 	BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
474 	BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
475 	BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
476 	__IP_ADD_STATS(net,
477 		       IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
478 		       max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
479 
480 	if (!pskb_may_pull(skb, iph->ihl*4))
481 		goto inhdr_error;
482 
483 	iph = ip_hdr(skb);
484 
485 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
486 		goto csum_error;
487 
488 	len = ntohs(iph->tot_len);
489 	if (skb->len < len) {
490 		__IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
491 		goto drop;
492 	} else if (len < (iph->ihl*4))
493 		goto inhdr_error;
494 
495 	/* Our transport medium may have padded the buffer out. Now we know it
496 	 * is IP we can trim to the true length of the frame.
497 	 * Note this now means skb->len holds ntohs(iph->tot_len).
498 	 */
499 	if (pskb_trim_rcsum(skb, len)) {
500 		__IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
501 		goto drop;
502 	}
503 
504 	iph = ip_hdr(skb);
505 	skb->transport_header = skb->network_header + iph->ihl*4;
506 
507 	/* Remove any debris in the socket control block */
508 	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
509 	IPCB(skb)->iif = skb->skb_iif;
510 
511 	/* Must drop socket now because of tproxy. */
512 	skb_orphan(skb);
513 
514 	return skb;
515 
516 csum_error:
517 	__IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
518 inhdr_error:
519 	__IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
520 drop:
521 	kfree_skb(skb);
522 out:
523 	return NULL;
524 }
525 
526 /*
527  * IP receive entry point
528  */
529 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
530 	   struct net_device *orig_dev)
531 {
532 	struct net *net = dev_net(dev);
533 
534 	skb = ip_rcv_core(skb, net);
535 	if (skb == NULL)
536 		return NET_RX_DROP;
537 
538 	return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
539 		       net, NULL, skb, dev, NULL,
540 		       ip_rcv_finish);
541 }
542 
543 static void ip_sublist_rcv_finish(struct list_head *head)
544 {
545 	struct sk_buff *skb, *next;
546 
547 	list_for_each_entry_safe(skb, next, head, list) {
548 		skb_list_del_init(skb);
549 		dst_input(skb);
550 	}
551 }
552 
553 static struct sk_buff *ip_extract_route_hint(const struct net *net,
554 					     struct sk_buff *skb, int rt_type)
555 {
556 	if (fib4_has_custom_rules(net) || rt_type == RTN_BROADCAST)
557 		return NULL;
558 
559 	return skb;
560 }
561 
562 static void ip_list_rcv_finish(struct net *net, struct sock *sk,
563 			       struct list_head *head)
564 {
565 	struct sk_buff *skb, *next, *hint = NULL;
566 	struct dst_entry *curr_dst = NULL;
567 	struct list_head sublist;
568 
569 	INIT_LIST_HEAD(&sublist);
570 	list_for_each_entry_safe(skb, next, head, list) {
571 		struct net_device *dev = skb->dev;
572 		struct dst_entry *dst;
573 
574 		skb_list_del_init(skb);
575 		/* if ingress device is enslaved to an L3 master device pass the
576 		 * skb to its handler for processing
577 		 */
578 		skb = l3mdev_ip_rcv(skb);
579 		if (!skb)
580 			continue;
581 		if (ip_rcv_finish_core(net, sk, skb, dev, hint) == NET_RX_DROP)
582 			continue;
583 
584 		dst = skb_dst(skb);
585 		if (curr_dst != dst) {
586 			hint = ip_extract_route_hint(net, skb,
587 					       ((struct rtable *)dst)->rt_type);
588 
589 			/* dispatch old sublist */
590 			if (!list_empty(&sublist))
591 				ip_sublist_rcv_finish(&sublist);
592 			/* start new sublist */
593 			INIT_LIST_HEAD(&sublist);
594 			curr_dst = dst;
595 		}
596 		list_add_tail(&skb->list, &sublist);
597 	}
598 	/* dispatch final sublist */
599 	ip_sublist_rcv_finish(&sublist);
600 }
601 
602 static void ip_sublist_rcv(struct list_head *head, struct net_device *dev,
603 			   struct net *net)
604 {
605 	NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL,
606 		     head, dev, NULL, ip_rcv_finish);
607 	ip_list_rcv_finish(net, NULL, head);
608 }
609 
610 /* Receive a list of IP packets */
611 void ip_list_rcv(struct list_head *head, struct packet_type *pt,
612 		 struct net_device *orig_dev)
613 {
614 	struct net_device *curr_dev = NULL;
615 	struct net *curr_net = NULL;
616 	struct sk_buff *skb, *next;
617 	struct list_head sublist;
618 
619 	INIT_LIST_HEAD(&sublist);
620 	list_for_each_entry_safe(skb, next, head, list) {
621 		struct net_device *dev = skb->dev;
622 		struct net *net = dev_net(dev);
623 
624 		skb_list_del_init(skb);
625 		skb = ip_rcv_core(skb, net);
626 		if (skb == NULL)
627 			continue;
628 
629 		if (curr_dev != dev || curr_net != net) {
630 			/* dispatch old sublist */
631 			if (!list_empty(&sublist))
632 				ip_sublist_rcv(&sublist, curr_dev, curr_net);
633 			/* start new sublist */
634 			INIT_LIST_HEAD(&sublist);
635 			curr_dev = dev;
636 			curr_net = net;
637 		}
638 		list_add_tail(&skb->list, &sublist);
639 	}
640 	/* dispatch final sublist */
641 	if (!list_empty(&sublist))
642 		ip_sublist_rcv(&sublist, curr_dev, curr_net);
643 }
644