xref: /linux/net/ipv4/ip_input.c (revision 17bbde2e1716e2ee4b997d476b48ae85c5a47671)
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  */
ip_call_ra_chain(struct sk_buff * skb)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 *));
ip_protocol_deliver_rcu(struct net * net,struct sk_buff * skb,int protocol)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_reason(skb,
200 						 SKB_DROP_REASON_XFRM_POLICY);
201 				return;
202 			}
203 			nf_reset_ct(skb);
204 		}
205 		ret = INDIRECT_CALL_2(ipprot->handler, tcp_v4_rcv, udp_rcv,
206 				      skb);
207 		if (ret < 0) {
208 			protocol = -ret;
209 			goto resubmit;
210 		}
211 		__IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
212 	} else {
213 		if (!raw) {
214 			if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
215 				__IP_INC_STATS(net, IPSTATS_MIB_INUNKNOWNPROTOS);
216 				icmp_send(skb, ICMP_DEST_UNREACH,
217 					  ICMP_PROT_UNREACH, 0);
218 			}
219 			kfree_skb_reason(skb, SKB_DROP_REASON_IP_NOPROTO);
220 		} else {
221 			__IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
222 			consume_skb(skb);
223 		}
224 	}
225 }
226 
ip_local_deliver_finish(struct net * net,struct sock * sk,struct sk_buff * skb)227 static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
228 {
229 	skb_clear_delivery_time(skb);
230 	__skb_pull(skb, skb_network_header_len(skb));
231 
232 	rcu_read_lock();
233 	ip_protocol_deliver_rcu(net, skb, ip_hdr(skb)->protocol);
234 	rcu_read_unlock();
235 
236 	return 0;
237 }
238 
239 /*
240  * 	Deliver IP Packets to the higher protocol layers.
241  */
ip_local_deliver(struct sk_buff * skb)242 int ip_local_deliver(struct sk_buff *skb)
243 {
244 	/*
245 	 *	Reassemble IP fragments.
246 	 */
247 	struct net *net = dev_net(skb->dev);
248 
249 	if (ip_is_fragment(ip_hdr(skb))) {
250 		if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
251 			return 0;
252 	}
253 
254 	return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
255 		       net, NULL, skb, skb->dev, NULL,
256 		       ip_local_deliver_finish);
257 }
258 EXPORT_SYMBOL(ip_local_deliver);
259 
ip_rcv_options(struct sk_buff * skb,struct net_device * dev)260 static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev)
261 {
262 	struct ip_options *opt;
263 	const struct iphdr *iph;
264 
265 	/* It looks as overkill, because not all
266 	   IP options require packet mangling.
267 	   But it is the easiest for now, especially taking
268 	   into account that combination of IP options
269 	   and running sniffer is extremely rare condition.
270 					      --ANK (980813)
271 	*/
272 	if (skb_cow(skb, skb_headroom(skb))) {
273 		__IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
274 		goto drop;
275 	}
276 
277 	iph = ip_hdr(skb);
278 	opt = &(IPCB(skb)->opt);
279 	opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
280 
281 	if (ip_options_compile(dev_net(dev), opt, skb)) {
282 		__IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
283 		goto drop;
284 	}
285 
286 	if (unlikely(opt->srr)) {
287 		struct in_device *in_dev = __in_dev_get_rcu(dev);
288 
289 		if (in_dev) {
290 			if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
291 				if (IN_DEV_LOG_MARTIANS(in_dev))
292 					net_info_ratelimited("source route option %pI4 -> %pI4\n",
293 							     &iph->saddr,
294 							     &iph->daddr);
295 				goto drop;
296 			}
297 		}
298 
299 		if (ip_options_rcv_srr(skb, dev))
300 			goto drop;
301 	}
302 
303 	return false;
304 drop:
305 	return true;
306 }
307 
ip_can_use_hint(const struct sk_buff * skb,const struct iphdr * iph,const struct sk_buff * hint)308 static bool ip_can_use_hint(const struct sk_buff *skb, const struct iphdr *iph,
309 			    const struct sk_buff *hint)
310 {
311 	return hint && !skb_dst(skb) && ip_hdr(hint)->daddr == iph->daddr &&
312 	       ip_hdr(hint)->tos == iph->tos;
313 }
314 
315 int tcp_v4_early_demux(struct sk_buff *skb);
316 int udp_v4_early_demux(struct sk_buff *skb);
ip_rcv_finish_core(struct net * net,struct sk_buff * skb,struct net_device * dev,const struct sk_buff * hint)317 static int ip_rcv_finish_core(struct net *net,
318 			      struct sk_buff *skb, struct net_device *dev,
319 			      const struct sk_buff *hint)
320 {
321 	const struct iphdr *iph = ip_hdr(skb);
322 	struct rtable *rt;
323 	int drop_reason;
324 
325 	if (ip_can_use_hint(skb, iph, hint)) {
326 		drop_reason = ip_route_use_hint(skb, iph->daddr, iph->saddr,
327 						ip4h_dscp(iph), dev, hint);
328 		if (unlikely(drop_reason))
329 			goto drop_error;
330 	}
331 
332 	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
333 	if (READ_ONCE(net->ipv4.sysctl_ip_early_demux) &&
334 	    !skb_dst(skb) &&
335 	    !skb->sk &&
336 	    !ip_is_fragment(iph)) {
337 		switch (iph->protocol) {
338 		case IPPROTO_TCP:
339 			if (READ_ONCE(net->ipv4.sysctl_tcp_early_demux)) {
340 				tcp_v4_early_demux(skb);
341 
342 				/* must reload iph, skb->head might have changed */
343 				iph = ip_hdr(skb);
344 			}
345 			break;
346 		case IPPROTO_UDP:
347 			if (READ_ONCE(net->ipv4.sysctl_udp_early_demux)) {
348 				drop_reason = udp_v4_early_demux(skb);
349 				if (unlikely(drop_reason))
350 					goto drop_error;
351 				drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
352 
353 				/* must reload iph, skb->head might have changed */
354 				iph = ip_hdr(skb);
355 			}
356 			break;
357 		}
358 	}
359 
360 	/*
361 	 *	Initialise the virtual path cache for the packet. It describes
362 	 *	how the packet travels inside Linux networking.
363 	 */
364 	if (!skb_valid_dst(skb)) {
365 		drop_reason = ip_route_input_noref(skb, iph->daddr, iph->saddr,
366 						   ip4h_dscp(iph), dev);
367 		if (unlikely(drop_reason))
368 			goto drop_error;
369 		drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
370 	} else {
371 		struct in_device *in_dev = __in_dev_get_rcu(dev);
372 
373 		if (in_dev && IN_DEV_ORCONF(in_dev, NOPOLICY))
374 			IPCB(skb)->flags |= IPSKB_NOPOLICY;
375 	}
376 
377 #ifdef CONFIG_IP_ROUTE_CLASSID
378 	if (unlikely(skb_dst(skb)->tclassid)) {
379 		struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
380 		u32 idx = skb_dst(skb)->tclassid;
381 		st[idx&0xFF].o_packets++;
382 		st[idx&0xFF].o_bytes += skb->len;
383 		st[(idx>>16)&0xFF].i_packets++;
384 		st[(idx>>16)&0xFF].i_bytes += skb->len;
385 	}
386 #endif
387 
388 	if (iph->ihl > 5 && ip_rcv_options(skb, dev))
389 		goto drop;
390 
391 	rt = skb_rtable(skb);
392 	if (rt->rt_type == RTN_MULTICAST) {
393 		__IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
394 	} else if (rt->rt_type == RTN_BROADCAST) {
395 		__IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
396 	} else if (skb->pkt_type == PACKET_BROADCAST ||
397 		   skb->pkt_type == PACKET_MULTICAST) {
398 		struct in_device *in_dev = __in_dev_get_rcu(dev);
399 
400 		/* RFC 1122 3.3.6:
401 		 *
402 		 *   When a host sends a datagram to a link-layer broadcast
403 		 *   address, the IP destination address MUST be a legal IP
404 		 *   broadcast or IP multicast address.
405 		 *
406 		 *   A host SHOULD silently discard a datagram that is received
407 		 *   via a link-layer broadcast (see Section 2.4) but does not
408 		 *   specify an IP multicast or broadcast destination address.
409 		 *
410 		 * This doesn't explicitly say L2 *broadcast*, but broadcast is
411 		 * in a way a form of multicast and the most common use case for
412 		 * this is 802.11 protecting against cross-station spoofing (the
413 		 * so-called "hole-196" attack) so do it for both.
414 		 */
415 		if (in_dev &&
416 		    IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST)) {
417 			drop_reason = SKB_DROP_REASON_UNICAST_IN_L2_MULTICAST;
418 			goto drop;
419 		}
420 	}
421 
422 	return NET_RX_SUCCESS;
423 
424 drop:
425 	kfree_skb_reason(skb, drop_reason);
426 	return NET_RX_DROP;
427 
428 drop_error:
429 	if (drop_reason == SKB_DROP_REASON_IP_RPFILTER)
430 		__NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
431 	goto drop;
432 }
433 
ip_rcv_finish(struct net * net,struct sock * sk,struct sk_buff * skb)434 static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
435 {
436 	struct net_device *dev = skb->dev;
437 	int ret;
438 
439 	/* if ingress device is enslaved to an L3 master device pass the
440 	 * skb to its handler for processing
441 	 */
442 	skb = l3mdev_ip_rcv(skb);
443 	if (!skb)
444 		return NET_RX_SUCCESS;
445 
446 	ret = ip_rcv_finish_core(net, skb, dev, NULL);
447 	if (ret != NET_RX_DROP)
448 		ret = dst_input(skb);
449 	return ret;
450 }
451 
452 /*
453  * 	Main IP Receive routine.
454  */
ip_rcv_core(struct sk_buff * skb,struct net * net)455 static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net)
456 {
457 	const struct iphdr *iph;
458 	int drop_reason;
459 	u32 len;
460 
461 	/* When the interface is in promisc. mode, drop all the crap
462 	 * that it receives, do not try to analyse it.
463 	 */
464 	if (skb->pkt_type == PACKET_OTHERHOST) {
465 		dev_core_stats_rx_otherhost_dropped_inc(skb->dev);
466 		drop_reason = SKB_DROP_REASON_OTHERHOST;
467 		goto drop;
468 	}
469 
470 	__IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
471 
472 	skb = skb_share_check(skb, GFP_ATOMIC);
473 	if (!skb) {
474 		__IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
475 		goto out;
476 	}
477 
478 	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
479 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
480 		goto inhdr_error;
481 
482 	iph = ip_hdr(skb);
483 
484 	/*
485 	 *	RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
486 	 *
487 	 *	Is the datagram acceptable?
488 	 *
489 	 *	1.	Length at least the size of an ip header
490 	 *	2.	Version of 4
491 	 *	3.	Checksums correctly. [Speed optimisation for later, skip loopback checksums]
492 	 *	4.	Doesn't have a bogus length
493 	 */
494 
495 	if (iph->ihl < 5 || iph->version != 4)
496 		goto inhdr_error;
497 
498 	BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
499 	BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
500 	BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
501 	__IP_ADD_STATS(net,
502 		       IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
503 		       max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
504 
505 	if (!pskb_may_pull(skb, iph->ihl*4))
506 		goto inhdr_error;
507 
508 	iph = ip_hdr(skb);
509 
510 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
511 		goto csum_error;
512 
513 	len = iph_totlen(skb, iph);
514 	if (skb->len < len) {
515 		drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
516 		__IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
517 		goto drop;
518 	} else if (len < (iph->ihl*4))
519 		goto inhdr_error;
520 
521 	/* Our transport medium may have padded the buffer out. Now we know it
522 	 * is IP we can trim to the true length of the frame.
523 	 * Note this now means skb->len holds ntohs(iph->tot_len).
524 	 */
525 	if (pskb_trim_rcsum(skb, len)) {
526 		__IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
527 		goto drop;
528 	}
529 
530 	iph = ip_hdr(skb);
531 	skb->transport_header = skb->network_header + iph->ihl*4;
532 
533 	/* Remove any debris in the socket control block */
534 	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
535 	IPCB(skb)->iif = skb->skb_iif;
536 
537 	/* Must drop socket now because of tproxy. */
538 	if (!skb_sk_is_prefetched(skb))
539 		skb_orphan(skb);
540 
541 	return skb;
542 
543 csum_error:
544 	drop_reason = SKB_DROP_REASON_IP_CSUM;
545 	__IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
546 inhdr_error:
547 	if (drop_reason == SKB_DROP_REASON_NOT_SPECIFIED)
548 		drop_reason = SKB_DROP_REASON_IP_INHDR;
549 	__IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
550 drop:
551 	kfree_skb_reason(skb, drop_reason);
552 out:
553 	return NULL;
554 }
555 
556 /*
557  * IP receive entry point
558  */
ip_rcv(struct sk_buff * skb,struct net_device * dev,struct packet_type * pt,struct net_device * orig_dev)559 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
560 	   struct net_device *orig_dev)
561 {
562 	struct net *net = dev_net(dev);
563 
564 	skb = ip_rcv_core(skb, net);
565 	if (skb == NULL)
566 		return NET_RX_DROP;
567 
568 	return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
569 		       net, NULL, skb, dev, NULL,
570 		       ip_rcv_finish);
571 }
572 
ip_sublist_rcv_finish(struct list_head * head)573 static void ip_sublist_rcv_finish(struct list_head *head)
574 {
575 	struct sk_buff *skb, *next;
576 
577 	list_for_each_entry_safe(skb, next, head, list) {
578 		skb_list_del_init(skb);
579 		dst_input(skb);
580 	}
581 }
582 
ip_extract_route_hint(const struct net * net,struct sk_buff * skb,int rt_type)583 static struct sk_buff *ip_extract_route_hint(const struct net *net,
584 					     struct sk_buff *skb, int rt_type)
585 {
586 	if (fib4_has_custom_rules(net) || rt_type == RTN_BROADCAST ||
587 	    IPCB(skb)->flags & IPSKB_MULTIPATH)
588 		return NULL;
589 
590 	return skb;
591 }
592 
ip_list_rcv_finish(struct net * net,struct list_head * head)593 static void ip_list_rcv_finish(struct net *net, struct list_head *head)
594 {
595 	struct sk_buff *skb, *next, *hint = NULL;
596 	struct dst_entry *curr_dst = NULL;
597 	LIST_HEAD(sublist);
598 
599 	list_for_each_entry_safe(skb, next, head, list) {
600 		struct net_device *dev = skb->dev;
601 		struct dst_entry *dst;
602 
603 		skb_list_del_init(skb);
604 		/* if ingress device is enslaved to an L3 master device pass the
605 		 * skb to its handler for processing
606 		 */
607 		skb = l3mdev_ip_rcv(skb);
608 		if (!skb)
609 			continue;
610 		if (ip_rcv_finish_core(net, skb, dev, hint) == NET_RX_DROP)
611 			continue;
612 
613 		dst = skb_dst(skb);
614 		if (curr_dst != dst) {
615 			hint = ip_extract_route_hint(net, skb,
616 						     dst_rtable(dst)->rt_type);
617 
618 			/* dispatch old sublist */
619 			if (!list_empty(&sublist))
620 				ip_sublist_rcv_finish(&sublist);
621 			/* start new sublist */
622 			INIT_LIST_HEAD(&sublist);
623 			curr_dst = dst;
624 		}
625 		list_add_tail(&skb->list, &sublist);
626 	}
627 	/* dispatch final sublist */
628 	ip_sublist_rcv_finish(&sublist);
629 }
630 
ip_sublist_rcv(struct list_head * head,struct net_device * dev,struct net * net)631 static void ip_sublist_rcv(struct list_head *head, struct net_device *dev,
632 			   struct net *net)
633 {
634 	NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL,
635 		     head, dev, NULL, ip_rcv_finish);
636 	ip_list_rcv_finish(net, head);
637 }
638 
639 /* Receive a list of IP packets */
ip_list_rcv(struct list_head * head,struct packet_type * pt,struct net_device * orig_dev)640 void ip_list_rcv(struct list_head *head, struct packet_type *pt,
641 		 struct net_device *orig_dev)
642 {
643 	struct net_device *curr_dev = NULL;
644 	struct net *curr_net = NULL;
645 	struct sk_buff *skb, *next;
646 	LIST_HEAD(sublist);
647 
648 	list_for_each_entry_safe(skb, next, head, list) {
649 		struct net_device *dev = skb->dev;
650 		struct net *net = dev_net(dev);
651 
652 		skb_list_del_init(skb);
653 		skb = ip_rcv_core(skb, net);
654 		if (skb == NULL)
655 			continue;
656 
657 		if (curr_dev != dev || curr_net != net) {
658 			/* dispatch old sublist */
659 			if (!list_empty(&sublist))
660 				ip_sublist_rcv(&sublist, curr_dev, curr_net);
661 			/* start new sublist */
662 			INIT_LIST_HEAD(&sublist);
663 			curr_dev = dev;
664 			curr_net = net;
665 		}
666 		list_add_tail(&skb->list, &sublist);
667 	}
668 	/* dispatch final sublist */
669 	if (!list_empty(&sublist))
670 		ip_sublist_rcv(&sublist, curr_dev, curr_net);
671 }
672