xref: /linux/net/ipv4/ip_input.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		The Internet Protocol (IP) module.
7  *
8  * Authors:	Ross Biro
9  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *		Donald Becker, <becker@super.org>
11  *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
12  *		Richard Underwood
13  *		Stefan Becker, <stefanb@yello.ping.de>
14  *		Jorge Cwik, <jorge@laser.satlink.net>
15  *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
16  *
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  *
100  *
101  * To Fix:
102  *		IP fragmentation wants rewriting cleanly. The RFC815 algorithm is much more efficient
103  *		and could be made very efficient with the addition of some virtual memory hacks to permit
104  *		the allocation of a buffer that can then be 'grown' by twiddling page tables.
105  *		Output fragmentation wants updating along with the buffer management to use a single
106  *		interleaved copy algorithm so that fragmenting has a one copy overhead. Actual packet
107  *		output should probably do its own fragmentation at the UDP/RAW layer. TCP shouldn't cause
108  *		fragmentation anyway.
109  *
110  *		This program is free software; you can redistribute it and/or
111  *		modify it under the terms of the GNU General Public License
112  *		as published by the Free Software Foundation; either version
113  *		2 of the License, or (at your option) any later version.
114  */
115 
116 #define pr_fmt(fmt) "IPv4: " fmt
117 
118 #include <linux/module.h>
119 #include <linux/types.h>
120 #include <linux/kernel.h>
121 #include <linux/string.h>
122 #include <linux/errno.h>
123 #include <linux/slab.h>
124 
125 #include <linux/net.h>
126 #include <linux/socket.h>
127 #include <linux/sockios.h>
128 #include <linux/in.h>
129 #include <linux/inet.h>
130 #include <linux/inetdevice.h>
131 #include <linux/netdevice.h>
132 #include <linux/etherdevice.h>
133 
134 #include <net/snmp.h>
135 #include <net/ip.h>
136 #include <net/protocol.h>
137 #include <net/route.h>
138 #include <linux/skbuff.h>
139 #include <net/sock.h>
140 #include <net/arp.h>
141 #include <net/icmp.h>
142 #include <net/raw.h>
143 #include <net/checksum.h>
144 #include <net/inet_ecn.h>
145 #include <linux/netfilter_ipv4.h>
146 #include <net/xfrm.h>
147 #include <linux/mroute.h>
148 #include <linux/netlink.h>
149 #include <net/dst_metadata.h>
150 
151 /*
152  *	Process Router Attention IP option (RFC 2113)
153  */
154 bool ip_call_ra_chain(struct sk_buff *skb)
155 {
156 	struct ip_ra_chain *ra;
157 	u8 protocol = ip_hdr(skb)->protocol;
158 	struct sock *last = NULL;
159 	struct net_device *dev = skb->dev;
160 	struct net *net = dev_net(dev);
161 
162 	for (ra = rcu_dereference(ip_ra_chain); ra; ra = rcu_dereference(ra->next)) {
163 		struct sock *sk = ra->sk;
164 
165 		/* If socket is bound to an interface, only report
166 		 * the packet if it came  from that interface.
167 		 */
168 		if (sk && inet_sk(sk)->inet_num == protocol &&
169 		    (!sk->sk_bound_dev_if ||
170 		     sk->sk_bound_dev_if == dev->ifindex) &&
171 		    net_eq(sock_net(sk), net)) {
172 			if (ip_is_fragment(ip_hdr(skb))) {
173 				if (ip_defrag(net, skb, IP_DEFRAG_CALL_RA_CHAIN))
174 					return true;
175 			}
176 			if (last) {
177 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
178 				if (skb2)
179 					raw_rcv(last, skb2);
180 			}
181 			last = sk;
182 		}
183 	}
184 
185 	if (last) {
186 		raw_rcv(last, skb);
187 		return true;
188 	}
189 	return false;
190 }
191 
192 static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
193 {
194 	__skb_pull(skb, skb_network_header_len(skb));
195 
196 	rcu_read_lock();
197 	{
198 		int protocol = ip_hdr(skb)->protocol;
199 		const struct net_protocol *ipprot;
200 		int raw;
201 
202 	resubmit:
203 		raw = raw_local_deliver(skb, protocol);
204 
205 		ipprot = rcu_dereference(inet_protos[protocol]);
206 		if (ipprot) {
207 			int ret;
208 
209 			if (!ipprot->no_policy) {
210 				if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
211 					kfree_skb(skb);
212 					goto out;
213 				}
214 				nf_reset(skb);
215 			}
216 			ret = ipprot->handler(skb);
217 			if (ret < 0) {
218 				protocol = -ret;
219 				goto resubmit;
220 			}
221 			__IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
222 		} else {
223 			if (!raw) {
224 				if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
225 					__IP_INC_STATS(net, IPSTATS_MIB_INUNKNOWNPROTOS);
226 					icmp_send(skb, ICMP_DEST_UNREACH,
227 						  ICMP_PROT_UNREACH, 0);
228 				}
229 				kfree_skb(skb);
230 			} else {
231 				__IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
232 				consume_skb(skb);
233 			}
234 		}
235 	}
236  out:
237 	rcu_read_unlock();
238 
239 	return 0;
240 }
241 
242 /*
243  * 	Deliver IP Packets to the higher protocol layers.
244  */
245 int ip_local_deliver(struct sk_buff *skb)
246 {
247 	/*
248 	 *	Reassemble IP fragments.
249 	 */
250 	struct net *net = dev_net(skb->dev);
251 
252 	if (ip_is_fragment(ip_hdr(skb))) {
253 		if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
254 			return 0;
255 	}
256 
257 	return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
258 		       net, NULL, skb, skb->dev, NULL,
259 		       ip_local_deliver_finish);
260 }
261 
262 static inline bool ip_rcv_options(struct sk_buff *skb)
263 {
264 	struct ip_options *opt;
265 	const struct iphdr *iph;
266 	struct net_device *dev = skb->dev;
267 
268 	/* It looks as overkill, because not all
269 	   IP options require packet mangling.
270 	   But it is the easiest for now, especially taking
271 	   into account that combination of IP options
272 	   and running sniffer is extremely rare condition.
273 					      --ANK (980813)
274 	*/
275 	if (skb_cow(skb, skb_headroom(skb))) {
276 		__IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
277 		goto drop;
278 	}
279 
280 	iph = ip_hdr(skb);
281 	opt = &(IPCB(skb)->opt);
282 	opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
283 
284 	if (ip_options_compile(dev_net(dev), opt, skb)) {
285 		__IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
286 		goto drop;
287 	}
288 
289 	if (unlikely(opt->srr)) {
290 		struct in_device *in_dev = __in_dev_get_rcu(dev);
291 
292 		if (in_dev) {
293 			if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
294 				if (IN_DEV_LOG_MARTIANS(in_dev))
295 					net_info_ratelimited("source route option %pI4 -> %pI4\n",
296 							     &iph->saddr,
297 							     &iph->daddr);
298 				goto drop;
299 			}
300 		}
301 
302 		if (ip_options_rcv_srr(skb))
303 			goto drop;
304 	}
305 
306 	return false;
307 drop:
308 	return true;
309 }
310 
311 static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
312 {
313 	const struct iphdr *iph = ip_hdr(skb);
314 	struct rtable *rt;
315 	struct net_device *dev = skb->dev;
316 
317 	/* if ingress device is enslaved to an L3 master device pass the
318 	 * skb to its handler for processing
319 	 */
320 	skb = l3mdev_ip_rcv(skb);
321 	if (!skb)
322 		return NET_RX_SUCCESS;
323 
324 	if (net->ipv4.sysctl_ip_early_demux &&
325 	    !skb_dst(skb) &&
326 	    !skb->sk &&
327 	    !ip_is_fragment(iph)) {
328 		const struct net_protocol *ipprot;
329 		int protocol = iph->protocol;
330 
331 		ipprot = rcu_dereference(inet_protos[protocol]);
332 		if (ipprot && ipprot->early_demux) {
333 			ipprot->early_demux(skb);
334 			/* must reload iph, skb->head might have changed */
335 			iph = ip_hdr(skb);
336 		}
337 	}
338 
339 	/*
340 	 *	Initialise the virtual path cache for the packet. It describes
341 	 *	how the packet travels inside Linux networking.
342 	 */
343 	if (!skb_valid_dst(skb)) {
344 		int err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
345 					       iph->tos, dev);
346 		if (unlikely(err)) {
347 			if (err == -EXDEV)
348 				__NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
349 			goto drop;
350 		}
351 	}
352 
353 #ifdef CONFIG_IP_ROUTE_CLASSID
354 	if (unlikely(skb_dst(skb)->tclassid)) {
355 		struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
356 		u32 idx = skb_dst(skb)->tclassid;
357 		st[idx&0xFF].o_packets++;
358 		st[idx&0xFF].o_bytes += skb->len;
359 		st[(idx>>16)&0xFF].i_packets++;
360 		st[(idx>>16)&0xFF].i_bytes += skb->len;
361 	}
362 #endif
363 
364 	if (iph->ihl > 5 && ip_rcv_options(skb))
365 		goto drop;
366 
367 	rt = skb_rtable(skb);
368 	if (rt->rt_type == RTN_MULTICAST) {
369 		__IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
370 	} else if (rt->rt_type == RTN_BROADCAST) {
371 		__IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
372 	} else if (skb->pkt_type == PACKET_BROADCAST ||
373 		   skb->pkt_type == PACKET_MULTICAST) {
374 		struct in_device *in_dev = __in_dev_get_rcu(dev);
375 
376 		/* RFC 1122 3.3.6:
377 		 *
378 		 *   When a host sends a datagram to a link-layer broadcast
379 		 *   address, the IP destination address MUST be a legal IP
380 		 *   broadcast or IP multicast address.
381 		 *
382 		 *   A host SHOULD silently discard a datagram that is received
383 		 *   via a link-layer broadcast (see Section 2.4) but does not
384 		 *   specify an IP multicast or broadcast destination address.
385 		 *
386 		 * This doesn't explicitly say L2 *broadcast*, but broadcast is
387 		 * in a way a form of multicast and the most common use case for
388 		 * this is 802.11 protecting against cross-station spoofing (the
389 		 * so-called "hole-196" attack) so do it for both.
390 		 */
391 		if (in_dev &&
392 		    IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST))
393 			goto drop;
394 	}
395 
396 	return dst_input(skb);
397 
398 drop:
399 	kfree_skb(skb);
400 	return NET_RX_DROP;
401 }
402 
403 /*
404  * 	Main IP Receive routine.
405  */
406 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
407 {
408 	const struct iphdr *iph;
409 	struct net *net;
410 	u32 len;
411 
412 	/* When the interface is in promisc. mode, drop all the crap
413 	 * that it receives, do not try to analyse it.
414 	 */
415 	if (skb->pkt_type == PACKET_OTHERHOST)
416 		goto drop;
417 
418 
419 	net = dev_net(dev);
420 	__IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
421 
422 	skb = skb_share_check(skb, GFP_ATOMIC);
423 	if (!skb) {
424 		__IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
425 		goto out;
426 	}
427 
428 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
429 		goto inhdr_error;
430 
431 	iph = ip_hdr(skb);
432 
433 	/*
434 	 *	RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
435 	 *
436 	 *	Is the datagram acceptable?
437 	 *
438 	 *	1.	Length at least the size of an ip header
439 	 *	2.	Version of 4
440 	 *	3.	Checksums correctly. [Speed optimisation for later, skip loopback checksums]
441 	 *	4.	Doesn't have a bogus length
442 	 */
443 
444 	if (iph->ihl < 5 || iph->version != 4)
445 		goto inhdr_error;
446 
447 	BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
448 	BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
449 	BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
450 	__IP_ADD_STATS(net,
451 		       IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
452 		       max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
453 
454 	if (!pskb_may_pull(skb, iph->ihl*4))
455 		goto inhdr_error;
456 
457 	iph = ip_hdr(skb);
458 
459 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
460 		goto csum_error;
461 
462 	len = ntohs(iph->tot_len);
463 	if (skb->len < len) {
464 		__IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
465 		goto drop;
466 	} else if (len < (iph->ihl*4))
467 		goto inhdr_error;
468 
469 	/* Our transport medium may have padded the buffer out. Now we know it
470 	 * is IP we can trim to the true length of the frame.
471 	 * Note this now means skb->len holds ntohs(iph->tot_len).
472 	 */
473 	if (pskb_trim_rcsum(skb, len)) {
474 		__IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
475 		goto drop;
476 	}
477 
478 	skb->transport_header = skb->network_header + iph->ihl*4;
479 
480 	/* Remove any debris in the socket control block */
481 	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
482 	IPCB(skb)->iif = skb->skb_iif;
483 
484 	/* Must drop socket now because of tproxy. */
485 	skb_orphan(skb);
486 
487 	return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
488 		       net, NULL, skb, dev, NULL,
489 		       ip_rcv_finish);
490 
491 csum_error:
492 	__IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
493 inhdr_error:
494 	__IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
495 drop:
496 	kfree_skb(skb);
497 out:
498 	return NET_RX_DROP;
499 }
500