xref: /linux/net/ipv4/ip_input.c (revision 93d546399c2b7d66a54d5fbd5eee17de19246bf6)
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 #include <asm/system.h>
117 #include <linux/module.h>
118 #include <linux/types.h>
119 #include <linux/kernel.h>
120 #include <linux/string.h>
121 #include <linux/errno.h>
122 
123 #include <linux/net.h>
124 #include <linux/socket.h>
125 #include <linux/sockios.h>
126 #include <linux/in.h>
127 #include <linux/inet.h>
128 #include <linux/inetdevice.h>
129 #include <linux/netdevice.h>
130 #include <linux/etherdevice.h>
131 
132 #include <net/snmp.h>
133 #include <net/ip.h>
134 #include <net/protocol.h>
135 #include <net/route.h>
136 #include <linux/skbuff.h>
137 #include <net/sock.h>
138 #include <net/arp.h>
139 #include <net/icmp.h>
140 #include <net/raw.h>
141 #include <net/checksum.h>
142 #include <linux/netfilter_ipv4.h>
143 #include <net/xfrm.h>
144 #include <linux/mroute.h>
145 #include <linux/netlink.h>
146 
147 /*
148  *	Process Router Attention IP option
149  */
150 int ip_call_ra_chain(struct sk_buff *skb)
151 {
152 	struct ip_ra_chain *ra;
153 	u8 protocol = ip_hdr(skb)->protocol;
154 	struct sock *last = NULL;
155 	struct net_device *dev = skb->dev;
156 
157 	read_lock(&ip_ra_lock);
158 	for (ra = ip_ra_chain; ra; ra = ra->next) {
159 		struct sock *sk = ra->sk;
160 
161 		/* If socket is bound to an interface, only report
162 		 * the packet if it came  from that interface.
163 		 */
164 		if (sk && inet_sk(sk)->num == protocol &&
165 		    (!sk->sk_bound_dev_if ||
166 		     sk->sk_bound_dev_if == dev->ifindex) &&
167 		    sock_net(sk) == dev_net(dev)) {
168 			if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
169 				if (ip_defrag(skb, IP_DEFRAG_CALL_RA_CHAIN)) {
170 					read_unlock(&ip_ra_lock);
171 					return 1;
172 				}
173 			}
174 			if (last) {
175 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
176 				if (skb2)
177 					raw_rcv(last, skb2);
178 			}
179 			last = sk;
180 		}
181 	}
182 
183 	if (last) {
184 		raw_rcv(last, skb);
185 		read_unlock(&ip_ra_lock);
186 		return 1;
187 	}
188 	read_unlock(&ip_ra_lock);
189 	return 0;
190 }
191 
192 static int ip_local_deliver_finish(struct sk_buff *skb)
193 {
194 	struct net *net = dev_net(skb->dev);
195 
196 	__skb_pull(skb, ip_hdrlen(skb));
197 
198 	/* Point into the IP datagram, just past the header. */
199 	skb_reset_transport_header(skb);
200 
201 	rcu_read_lock();
202 	{
203 		int protocol = ip_hdr(skb)->protocol;
204 		int hash, raw;
205 		struct net_protocol *ipprot;
206 
207 	resubmit:
208 		raw = raw_local_deliver(skb, protocol);
209 
210 		hash = protocol & (MAX_INET_PROTOS - 1);
211 		ipprot = rcu_dereference(inet_protos[hash]);
212 		if (ipprot != NULL) {
213 			int ret;
214 
215 			if (!net_eq(net, &init_net) && !ipprot->netns_ok) {
216 				if (net_ratelimit())
217 					printk("%s: proto %d isn't netns-ready\n",
218 						__func__, protocol);
219 				kfree_skb(skb);
220 				goto out;
221 			}
222 
223 			if (!ipprot->no_policy) {
224 				if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
225 					kfree_skb(skb);
226 					goto out;
227 				}
228 				nf_reset(skb);
229 			}
230 			ret = ipprot->handler(skb);
231 			if (ret < 0) {
232 				protocol = -ret;
233 				goto resubmit;
234 			}
235 			IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS);
236 		} else {
237 			if (!raw) {
238 				if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
239 					IP_INC_STATS_BH(net, IPSTATS_MIB_INUNKNOWNPROTOS);
240 					icmp_send(skb, ICMP_DEST_UNREACH,
241 						  ICMP_PROT_UNREACH, 0);
242 				}
243 			} else
244 				IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS);
245 			kfree_skb(skb);
246 		}
247 	}
248  out:
249 	rcu_read_unlock();
250 
251 	return 0;
252 }
253 
254 /*
255  * 	Deliver IP Packets to the higher protocol layers.
256  */
257 int ip_local_deliver(struct sk_buff *skb)
258 {
259 	/*
260 	 *	Reassemble IP fragments.
261 	 */
262 
263 	if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
264 		if (ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER))
265 			return 0;
266 	}
267 
268 	return NF_HOOK(PF_INET, NF_INET_LOCAL_IN, skb, skb->dev, NULL,
269 		       ip_local_deliver_finish);
270 }
271 
272 static inline int ip_rcv_options(struct sk_buff *skb)
273 {
274 	struct ip_options *opt;
275 	struct iphdr *iph;
276 	struct net_device *dev = skb->dev;
277 
278 	/* It looks as overkill, because not all
279 	   IP options require packet mangling.
280 	   But it is the easiest for now, especially taking
281 	   into account that combination of IP options
282 	   and running sniffer is extremely rare condition.
283 					      --ANK (980813)
284 	*/
285 	if (skb_cow(skb, skb_headroom(skb))) {
286 		IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
287 		goto drop;
288 	}
289 
290 	iph = ip_hdr(skb);
291 	opt = &(IPCB(skb)->opt);
292 	opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
293 
294 	if (ip_options_compile(dev_net(dev), opt, skb)) {
295 		IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
296 		goto drop;
297 	}
298 
299 	if (unlikely(opt->srr)) {
300 		struct in_device *in_dev = in_dev_get(dev);
301 		if (in_dev) {
302 			if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
303 				if (IN_DEV_LOG_MARTIANS(in_dev) &&
304 				    net_ratelimit())
305 					printk(KERN_INFO "source route option "
306 					       NIPQUAD_FMT " -> " NIPQUAD_FMT "\n",
307 					       NIPQUAD(iph->saddr),
308 					       NIPQUAD(iph->daddr));
309 				in_dev_put(in_dev);
310 				goto drop;
311 			}
312 
313 			in_dev_put(in_dev);
314 		}
315 
316 		if (ip_options_rcv_srr(skb))
317 			goto drop;
318 	}
319 
320 	return 0;
321 drop:
322 	return -1;
323 }
324 
325 static int ip_rcv_finish(struct sk_buff *skb)
326 {
327 	const struct iphdr *iph = ip_hdr(skb);
328 	struct rtable *rt;
329 
330 	/*
331 	 *	Initialise the virtual path cache for the packet. It describes
332 	 *	how the packet travels inside Linux networking.
333 	 */
334 	if (skb->dst == NULL) {
335 		int err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos,
336 					 skb->dev);
337 		if (unlikely(err)) {
338 			if (err == -EHOSTUNREACH)
339 				IP_INC_STATS_BH(dev_net(skb->dev),
340 						IPSTATS_MIB_INADDRERRORS);
341 			else if (err == -ENETUNREACH)
342 				IP_INC_STATS_BH(dev_net(skb->dev),
343 						IPSTATS_MIB_INNOROUTES);
344 			goto drop;
345 		}
346 	}
347 
348 #ifdef CONFIG_NET_CLS_ROUTE
349 	if (unlikely(skb->dst->tclassid)) {
350 		struct ip_rt_acct *st = per_cpu_ptr(ip_rt_acct, smp_processor_id());
351 		u32 idx = skb->dst->tclassid;
352 		st[idx&0xFF].o_packets++;
353 		st[idx&0xFF].o_bytes+=skb->len;
354 		st[(idx>>16)&0xFF].i_packets++;
355 		st[(idx>>16)&0xFF].i_bytes+=skb->len;
356 	}
357 #endif
358 
359 	if (iph->ihl > 5 && ip_rcv_options(skb))
360 		goto drop;
361 
362 	rt = skb->rtable;
363 	if (rt->rt_type == RTN_MULTICAST)
364 		IP_INC_STATS_BH(dev_net(rt->u.dst.dev), IPSTATS_MIB_INMCASTPKTS);
365 	else if (rt->rt_type == RTN_BROADCAST)
366 		IP_INC_STATS_BH(dev_net(rt->u.dst.dev), IPSTATS_MIB_INBCASTPKTS);
367 
368 	return dst_input(skb);
369 
370 drop:
371 	kfree_skb(skb);
372 	return NET_RX_DROP;
373 }
374 
375 /*
376  * 	Main IP Receive routine.
377  */
378 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
379 {
380 	struct iphdr *iph;
381 	u32 len;
382 
383 	/* When the interface is in promisc. mode, drop all the crap
384 	 * that it receives, do not try to analyse it.
385 	 */
386 	if (skb->pkt_type == PACKET_OTHERHOST)
387 		goto drop;
388 
389 	IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INRECEIVES);
390 
391 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) {
392 		IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
393 		goto out;
394 	}
395 
396 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
397 		goto inhdr_error;
398 
399 	iph = ip_hdr(skb);
400 
401 	/*
402 	 *	RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
403 	 *
404 	 *	Is the datagram acceptable?
405 	 *
406 	 *	1.	Length at least the size of an ip header
407 	 *	2.	Version of 4
408 	 *	3.	Checksums correctly. [Speed optimisation for later, skip loopback checksums]
409 	 *	4.	Doesn't have a bogus length
410 	 */
411 
412 	if (iph->ihl < 5 || iph->version != 4)
413 		goto inhdr_error;
414 
415 	if (!pskb_may_pull(skb, iph->ihl*4))
416 		goto inhdr_error;
417 
418 	iph = ip_hdr(skb);
419 
420 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
421 		goto inhdr_error;
422 
423 	len = ntohs(iph->tot_len);
424 	if (skb->len < len) {
425 		IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
426 		goto drop;
427 	} else if (len < (iph->ihl*4))
428 		goto inhdr_error;
429 
430 	/* Our transport medium may have padded the buffer out. Now we know it
431 	 * is IP we can trim to the true length of the frame.
432 	 * Note this now means skb->len holds ntohs(iph->tot_len).
433 	 */
434 	if (pskb_trim_rcsum(skb, len)) {
435 		IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
436 		goto drop;
437 	}
438 
439 	/* Remove any debris in the socket control block */
440 	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
441 
442 	return NF_HOOK(PF_INET, NF_INET_PRE_ROUTING, skb, dev, NULL,
443 		       ip_rcv_finish);
444 
445 inhdr_error:
446 	IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
447 drop:
448 	kfree_skb(skb);
449 out:
450 	return NET_RX_DROP;
451 }
452