xref: /linux/net/ipv4/ip_input.c (revision 8be4d31cb8aaeea27bde4b7ddb26e28a89062ebf)
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 	if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
230 		__IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
231 		kfree_skb_reason(skb, SKB_DROP_REASON_NOMEM);
232 		return 0;
233 	}
234 
235 	skb_clear_delivery_time(skb);
236 	__skb_pull(skb, skb_network_header_len(skb));
237 
238 	rcu_read_lock();
239 	ip_protocol_deliver_rcu(net, skb, ip_hdr(skb)->protocol);
240 	rcu_read_unlock();
241 
242 	return 0;
243 }
244 
245 /*
246  * 	Deliver IP Packets to the higher protocol layers.
247  */
ip_local_deliver(struct sk_buff * skb)248 int ip_local_deliver(struct sk_buff *skb)
249 {
250 	/*
251 	 *	Reassemble IP fragments.
252 	 */
253 	struct net *net = dev_net(skb->dev);
254 
255 	if (ip_is_fragment(ip_hdr(skb))) {
256 		if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
257 			return 0;
258 	}
259 
260 	return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
261 		       net, NULL, skb, skb->dev, NULL,
262 		       ip_local_deliver_finish);
263 }
264 EXPORT_SYMBOL(ip_local_deliver);
265 
ip_rcv_options(struct sk_buff * skb,struct net_device * dev)266 static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev)
267 {
268 	struct ip_options *opt;
269 	const struct iphdr *iph;
270 
271 	/* It looks as overkill, because not all
272 	   IP options require packet mangling.
273 	   But it is the easiest for now, especially taking
274 	   into account that combination of IP options
275 	   and running sniffer is extremely rare condition.
276 					      --ANK (980813)
277 	*/
278 	if (skb_cow(skb, skb_headroom(skb))) {
279 		__IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
280 		goto drop;
281 	}
282 
283 	iph = ip_hdr(skb);
284 	opt = &(IPCB(skb)->opt);
285 	opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
286 
287 	if (ip_options_compile(dev_net(dev), opt, skb)) {
288 		__IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
289 		goto drop;
290 	}
291 
292 	if (unlikely(opt->srr)) {
293 		struct in_device *in_dev = __in_dev_get_rcu(dev);
294 
295 		if (in_dev) {
296 			if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
297 				if (IN_DEV_LOG_MARTIANS(in_dev))
298 					net_info_ratelimited("source route option %pI4 -> %pI4\n",
299 							     &iph->saddr,
300 							     &iph->daddr);
301 				goto drop;
302 			}
303 		}
304 
305 		if (ip_options_rcv_srr(skb, dev))
306 			goto drop;
307 	}
308 
309 	return false;
310 drop:
311 	return true;
312 }
313 
ip_can_use_hint(const struct sk_buff * skb,const struct iphdr * iph,const struct sk_buff * hint)314 static bool ip_can_use_hint(const struct sk_buff *skb, const struct iphdr *iph,
315 			    const struct sk_buff *hint)
316 {
317 	return hint && !skb_dst(skb) && ip_hdr(hint)->daddr == iph->daddr &&
318 	       ip_hdr(hint)->tos == iph->tos;
319 }
320 
321 int tcp_v4_early_demux(struct sk_buff *skb);
322 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)323 static int ip_rcv_finish_core(struct net *net,
324 			      struct sk_buff *skb, struct net_device *dev,
325 			      const struct sk_buff *hint)
326 {
327 	const struct iphdr *iph = ip_hdr(skb);
328 	struct rtable *rt;
329 	int drop_reason;
330 
331 	if (ip_can_use_hint(skb, iph, hint)) {
332 		drop_reason = ip_route_use_hint(skb, iph->daddr, iph->saddr,
333 						ip4h_dscp(iph), dev, hint);
334 		if (unlikely(drop_reason))
335 			goto drop_error;
336 	}
337 
338 	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
339 	if (READ_ONCE(net->ipv4.sysctl_ip_early_demux) &&
340 	    !skb_dst(skb) &&
341 	    !skb->sk &&
342 	    !ip_is_fragment(iph)) {
343 		switch (iph->protocol) {
344 		case IPPROTO_TCP:
345 			if (READ_ONCE(net->ipv4.sysctl_tcp_early_demux)) {
346 				tcp_v4_early_demux(skb);
347 
348 				/* must reload iph, skb->head might have changed */
349 				iph = ip_hdr(skb);
350 			}
351 			break;
352 		case IPPROTO_UDP:
353 			if (READ_ONCE(net->ipv4.sysctl_udp_early_demux)) {
354 				drop_reason = udp_v4_early_demux(skb);
355 				if (unlikely(drop_reason))
356 					goto drop_error;
357 				drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
358 
359 				/* must reload iph, skb->head might have changed */
360 				iph = ip_hdr(skb);
361 			}
362 			break;
363 		}
364 	}
365 
366 	/*
367 	 *	Initialise the virtual path cache for the packet. It describes
368 	 *	how the packet travels inside Linux networking.
369 	 */
370 	if (!skb_valid_dst(skb)) {
371 		drop_reason = ip_route_input_noref(skb, iph->daddr, iph->saddr,
372 						   ip4h_dscp(iph), dev);
373 		if (unlikely(drop_reason))
374 			goto drop_error;
375 		drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
376 	} else {
377 		struct in_device *in_dev = __in_dev_get_rcu(dev);
378 
379 		if (in_dev && IN_DEV_ORCONF(in_dev, NOPOLICY))
380 			IPCB(skb)->flags |= IPSKB_NOPOLICY;
381 	}
382 
383 #ifdef CONFIG_IP_ROUTE_CLASSID
384 	if (unlikely(skb_dst(skb)->tclassid)) {
385 		struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
386 		u32 idx = skb_dst(skb)->tclassid;
387 		st[idx&0xFF].o_packets++;
388 		st[idx&0xFF].o_bytes += skb->len;
389 		st[(idx>>16)&0xFF].i_packets++;
390 		st[(idx>>16)&0xFF].i_bytes += skb->len;
391 	}
392 #endif
393 
394 	if (iph->ihl > 5 && ip_rcv_options(skb, dev))
395 		goto drop;
396 
397 	rt = skb_rtable(skb);
398 	if (rt->rt_type == RTN_MULTICAST) {
399 		__IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
400 	} else if (rt->rt_type == RTN_BROADCAST) {
401 		__IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
402 	} else if (skb->pkt_type == PACKET_BROADCAST ||
403 		   skb->pkt_type == PACKET_MULTICAST) {
404 		struct in_device *in_dev = __in_dev_get_rcu(dev);
405 
406 		/* RFC 1122 3.3.6:
407 		 *
408 		 *   When a host sends a datagram to a link-layer broadcast
409 		 *   address, the IP destination address MUST be a legal IP
410 		 *   broadcast or IP multicast address.
411 		 *
412 		 *   A host SHOULD silently discard a datagram that is received
413 		 *   via a link-layer broadcast (see Section 2.4) but does not
414 		 *   specify an IP multicast or broadcast destination address.
415 		 *
416 		 * This doesn't explicitly say L2 *broadcast*, but broadcast is
417 		 * in a way a form of multicast and the most common use case for
418 		 * this is 802.11 protecting against cross-station spoofing (the
419 		 * so-called "hole-196" attack) so do it for both.
420 		 */
421 		if (in_dev &&
422 		    IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST)) {
423 			drop_reason = SKB_DROP_REASON_UNICAST_IN_L2_MULTICAST;
424 			goto drop;
425 		}
426 	}
427 
428 	return NET_RX_SUCCESS;
429 
430 drop:
431 	kfree_skb_reason(skb, drop_reason);
432 	return NET_RX_DROP;
433 
434 drop_error:
435 	if (drop_reason == SKB_DROP_REASON_IP_RPFILTER)
436 		__NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
437 	goto drop;
438 }
439 
ip_rcv_finish(struct net * net,struct sock * sk,struct sk_buff * skb)440 static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
441 {
442 	struct net_device *dev = skb->dev;
443 	int ret;
444 
445 	/* if ingress device is enslaved to an L3 master device pass the
446 	 * skb to its handler for processing
447 	 */
448 	skb = l3mdev_ip_rcv(skb);
449 	if (!skb)
450 		return NET_RX_SUCCESS;
451 
452 	ret = ip_rcv_finish_core(net, skb, dev, NULL);
453 	if (ret != NET_RX_DROP)
454 		ret = dst_input(skb);
455 	return ret;
456 }
457 
458 /*
459  * 	Main IP Receive routine.
460  */
ip_rcv_core(struct sk_buff * skb,struct net * net)461 static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net)
462 {
463 	const struct iphdr *iph;
464 	int drop_reason;
465 	u32 len;
466 
467 	/* When the interface is in promisc. mode, drop all the crap
468 	 * that it receives, do not try to analyse it.
469 	 */
470 	if (skb->pkt_type == PACKET_OTHERHOST) {
471 		dev_core_stats_rx_otherhost_dropped_inc(skb->dev);
472 		drop_reason = SKB_DROP_REASON_OTHERHOST;
473 		goto drop;
474 	}
475 
476 	__IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
477 
478 	skb = skb_share_check(skb, GFP_ATOMIC);
479 	if (!skb) {
480 		__IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
481 		goto out;
482 	}
483 
484 	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
485 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
486 		goto inhdr_error;
487 
488 	iph = ip_hdr(skb);
489 
490 	/*
491 	 *	RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
492 	 *
493 	 *	Is the datagram acceptable?
494 	 *
495 	 *	1.	Length at least the size of an ip header
496 	 *	2.	Version of 4
497 	 *	3.	Checksums correctly. [Speed optimisation for later, skip loopback checksums]
498 	 *	4.	Doesn't have a bogus length
499 	 */
500 
501 	if (iph->ihl < 5 || iph->version != 4)
502 		goto inhdr_error;
503 
504 	BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
505 	BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
506 	BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
507 	__IP_ADD_STATS(net,
508 		       IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
509 		       max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
510 
511 	if (!pskb_may_pull(skb, iph->ihl*4))
512 		goto inhdr_error;
513 
514 	iph = ip_hdr(skb);
515 
516 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
517 		goto csum_error;
518 
519 	len = iph_totlen(skb, iph);
520 	if (skb->len < len) {
521 		drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
522 		__IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
523 		goto drop;
524 	} else if (len < (iph->ihl*4))
525 		goto inhdr_error;
526 
527 	/* Our transport medium may have padded the buffer out. Now we know it
528 	 * is IP we can trim to the true length of the frame.
529 	 * Note this now means skb->len holds ntohs(iph->tot_len).
530 	 */
531 	if (pskb_trim_rcsum(skb, len)) {
532 		__IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
533 		goto drop;
534 	}
535 
536 	iph = ip_hdr(skb);
537 	skb->transport_header = skb->network_header + iph->ihl*4;
538 
539 	/* Remove any debris in the socket control block */
540 	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
541 	IPCB(skb)->iif = skb->skb_iif;
542 
543 	/* Must drop socket now because of tproxy. */
544 	if (!skb_sk_is_prefetched(skb))
545 		skb_orphan(skb);
546 
547 	return skb;
548 
549 csum_error:
550 	drop_reason = SKB_DROP_REASON_IP_CSUM;
551 	__IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
552 inhdr_error:
553 	if (drop_reason == SKB_DROP_REASON_NOT_SPECIFIED)
554 		drop_reason = SKB_DROP_REASON_IP_INHDR;
555 	__IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
556 drop:
557 	kfree_skb_reason(skb, drop_reason);
558 out:
559 	return NULL;
560 }
561 
562 /*
563  * IP receive entry point
564  */
ip_rcv(struct sk_buff * skb,struct net_device * dev,struct packet_type * pt,struct net_device * orig_dev)565 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
566 	   struct net_device *orig_dev)
567 {
568 	struct net *net = dev_net(dev);
569 
570 	skb = ip_rcv_core(skb, net);
571 	if (skb == NULL)
572 		return NET_RX_DROP;
573 
574 	return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
575 		       net, NULL, skb, dev, NULL,
576 		       ip_rcv_finish);
577 }
578 
ip_sublist_rcv_finish(struct list_head * head)579 static void ip_sublist_rcv_finish(struct list_head *head)
580 {
581 	struct sk_buff *skb, *next;
582 
583 	list_for_each_entry_safe(skb, next, head, list) {
584 		skb_list_del_init(skb);
585 		dst_input(skb);
586 	}
587 }
588 
ip_extract_route_hint(const struct net * net,struct sk_buff * skb,int rt_type)589 static struct sk_buff *ip_extract_route_hint(const struct net *net,
590 					     struct sk_buff *skb, int rt_type)
591 {
592 	if (fib4_has_custom_rules(net) || rt_type == RTN_BROADCAST ||
593 	    IPCB(skb)->flags & IPSKB_MULTIPATH)
594 		return NULL;
595 
596 	return skb;
597 }
598 
ip_list_rcv_finish(struct net * net,struct list_head * head)599 static void ip_list_rcv_finish(struct net *net, struct list_head *head)
600 {
601 	struct sk_buff *skb, *next, *hint = NULL;
602 	struct dst_entry *curr_dst = NULL;
603 	LIST_HEAD(sublist);
604 
605 	list_for_each_entry_safe(skb, next, head, list) {
606 		struct net_device *dev = skb->dev;
607 		struct dst_entry *dst;
608 
609 		skb_list_del_init(skb);
610 		/* if ingress device is enslaved to an L3 master device pass the
611 		 * skb to its handler for processing
612 		 */
613 		skb = l3mdev_ip_rcv(skb);
614 		if (!skb)
615 			continue;
616 		if (ip_rcv_finish_core(net, skb, dev, hint) == NET_RX_DROP)
617 			continue;
618 
619 		dst = skb_dst(skb);
620 		if (curr_dst != dst) {
621 			hint = ip_extract_route_hint(net, skb,
622 						     dst_rtable(dst)->rt_type);
623 
624 			/* dispatch old sublist */
625 			if (!list_empty(&sublist))
626 				ip_sublist_rcv_finish(&sublist);
627 			/* start new sublist */
628 			INIT_LIST_HEAD(&sublist);
629 			curr_dst = dst;
630 		}
631 		list_add_tail(&skb->list, &sublist);
632 	}
633 	/* dispatch final sublist */
634 	ip_sublist_rcv_finish(&sublist);
635 }
636 
ip_sublist_rcv(struct list_head * head,struct net_device * dev,struct net * net)637 static void ip_sublist_rcv(struct list_head *head, struct net_device *dev,
638 			   struct net *net)
639 {
640 	NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL,
641 		     head, dev, NULL, ip_rcv_finish);
642 	ip_list_rcv_finish(net, head);
643 }
644 
645 /* Receive a list of IP packets */
ip_list_rcv(struct list_head * head,struct packet_type * pt,struct net_device * orig_dev)646 void ip_list_rcv(struct list_head *head, struct packet_type *pt,
647 		 struct net_device *orig_dev)
648 {
649 	struct net_device *curr_dev = NULL;
650 	struct net *curr_net = NULL;
651 	struct sk_buff *skb, *next;
652 	LIST_HEAD(sublist);
653 
654 	list_for_each_entry_safe(skb, next, head, list) {
655 		struct net_device *dev = skb->dev;
656 		struct net *net = dev_net(dev);
657 
658 		skb_list_del_init(skb);
659 		skb = ip_rcv_core(skb, net);
660 		if (skb == NULL)
661 			continue;
662 
663 		if (curr_dev != dev || curr_net != net) {
664 			/* dispatch old sublist */
665 			if (!list_empty(&sublist))
666 				ip_sublist_rcv(&sublist, curr_dev, curr_net);
667 			/* start new sublist */
668 			INIT_LIST_HEAD(&sublist);
669 			curr_dev = dev;
670 			curr_net = net;
671 		}
672 		list_add_tail(&skb->list, &sublist);
673 	}
674 	/* dispatch final sublist */
675 	if (!list_empty(&sublist))
676 		ip_sublist_rcv(&sublist, curr_dev, curr_net);
677 }
678