xref: /linux/net/ipv4/ip_input.c (revision a5c4300389bb33ade2515c082709217f0614cf15)
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 #include <linux/slab.h>
123 
124 #include <linux/net.h>
125 #include <linux/socket.h>
126 #include <linux/sockios.h>
127 #include <linux/in.h>
128 #include <linux/inet.h>
129 #include <linux/inetdevice.h>
130 #include <linux/netdevice.h>
131 #include <linux/etherdevice.h>
132 
133 #include <net/snmp.h>
134 #include <net/ip.h>
135 #include <net/protocol.h>
136 #include <net/route.h>
137 #include <linux/skbuff.h>
138 #include <net/sock.h>
139 #include <net/arp.h>
140 #include <net/icmp.h>
141 #include <net/raw.h>
142 #include <net/checksum.h>
143 #include <linux/netfilter_ipv4.h>
144 #include <net/xfrm.h>
145 #include <linux/mroute.h>
146 #include <linux/netlink.h>
147 
148 /*
149  *	Process Router Attention IP option
150  */
151 int ip_call_ra_chain(struct sk_buff *skb)
152 {
153 	struct ip_ra_chain *ra;
154 	u8 protocol = ip_hdr(skb)->protocol;
155 	struct sock *last = NULL;
156 	struct net_device *dev = skb->dev;
157 
158 	read_lock(&ip_ra_lock);
159 	for (ra = ip_ra_chain; ra; ra = ra->next) {
160 		struct sock *sk = ra->sk;
161 
162 		/* If socket is bound to an interface, only report
163 		 * the packet if it came  from that interface.
164 		 */
165 		if (sk && inet_sk(sk)->inet_num == protocol &&
166 		    (!sk->sk_bound_dev_if ||
167 		     sk->sk_bound_dev_if == dev->ifindex) &&
168 		    net_eq(sock_net(sk), dev_net(dev))) {
169 			if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
170 				if (ip_defrag(skb, IP_DEFRAG_CALL_RA_CHAIN)) {
171 					read_unlock(&ip_ra_lock);
172 					return 1;
173 				}
174 			}
175 			if (last) {
176 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
177 				if (skb2)
178 					raw_rcv(last, skb2);
179 			}
180 			last = sk;
181 		}
182 	}
183 
184 	if (last) {
185 		raw_rcv(last, skb);
186 		read_unlock(&ip_ra_lock);
187 		return 1;
188 	}
189 	read_unlock(&ip_ra_lock);
190 	return 0;
191 }
192 
193 static int ip_local_deliver_finish(struct sk_buff *skb)
194 {
195 	struct net *net = dev_net(skb->dev);
196 
197 	__skb_pull(skb, ip_hdrlen(skb));
198 
199 	/* Point into the IP datagram, just past the header. */
200 	skb_reset_transport_header(skb);
201 
202 	rcu_read_lock();
203 	{
204 		int protocol = ip_hdr(skb)->protocol;
205 		int hash, raw;
206 		const struct net_protocol *ipprot;
207 
208 	resubmit:
209 		raw = raw_local_deliver(skb, protocol);
210 
211 		hash = protocol & (MAX_INET_PROTOS - 1);
212 		ipprot = rcu_dereference(inet_protos[hash]);
213 		if (ipprot != NULL) {
214 			int ret;
215 
216 			if (!net_eq(net, &init_net) && !ipprot->netns_ok) {
217 				if (net_ratelimit())
218 					printk("%s: proto %d isn't netns-ready\n",
219 						__func__, protocol);
220 				kfree_skb(skb);
221 				goto out;
222 			}
223 
224 			if (!ipprot->no_policy) {
225 				if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
226 					kfree_skb(skb);
227 					goto out;
228 				}
229 				nf_reset(skb);
230 			}
231 			ret = ipprot->handler(skb);
232 			if (ret < 0) {
233 				protocol = -ret;
234 				goto resubmit;
235 			}
236 			IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS);
237 		} else {
238 			if (!raw) {
239 				if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
240 					IP_INC_STATS_BH(net, IPSTATS_MIB_INUNKNOWNPROTOS);
241 					icmp_send(skb, ICMP_DEST_UNREACH,
242 						  ICMP_PROT_UNREACH, 0);
243 				}
244 			} else
245 				IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS);
246 			kfree_skb(skb);
247 		}
248 	}
249  out:
250 	rcu_read_unlock();
251 
252 	return 0;
253 }
254 
255 /*
256  * 	Deliver IP Packets to the higher protocol layers.
257  */
258 int ip_local_deliver(struct sk_buff *skb)
259 {
260 	/*
261 	 *	Reassemble IP fragments.
262 	 */
263 
264 	if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
265 		if (ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER))
266 			return 0;
267 	}
268 
269 	return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN, skb, skb->dev, NULL,
270 		       ip_local_deliver_finish);
271 }
272 
273 static inline int ip_rcv_options(struct sk_buff *skb)
274 {
275 	struct ip_options *opt;
276 	struct iphdr *iph;
277 	struct net_device *dev = skb->dev;
278 
279 	/* It looks as overkill, because not all
280 	   IP options require packet mangling.
281 	   But it is the easiest for now, especially taking
282 	   into account that combination of IP options
283 	   and running sniffer is extremely rare condition.
284 					      --ANK (980813)
285 	*/
286 	if (skb_cow(skb, skb_headroom(skb))) {
287 		IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
288 		goto drop;
289 	}
290 
291 	iph = ip_hdr(skb);
292 	opt = &(IPCB(skb)->opt);
293 	opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
294 
295 	if (ip_options_compile(dev_net(dev), opt, skb)) {
296 		IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
297 		goto drop;
298 	}
299 
300 	if (unlikely(opt->srr)) {
301 		struct in_device *in_dev = in_dev_get(dev);
302 		if (in_dev) {
303 			if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
304 				if (IN_DEV_LOG_MARTIANS(in_dev) &&
305 				    net_ratelimit())
306 					printk(KERN_INFO "source route option %pI4 -> %pI4\n",
307 					       &iph->saddr, &iph->daddr);
308 				in_dev_put(in_dev);
309 				goto drop;
310 			}
311 
312 			in_dev_put(in_dev);
313 		}
314 
315 		if (ip_options_rcv_srr(skb))
316 			goto drop;
317 	}
318 
319 	return 0;
320 drop:
321 	return -1;
322 }
323 
324 static int ip_rcv_finish(struct sk_buff *skb)
325 {
326 	const struct iphdr *iph = ip_hdr(skb);
327 	struct rtable *rt;
328 
329 	/*
330 	 *	Initialise the virtual path cache for the packet. It describes
331 	 *	how the packet travels inside Linux networking.
332 	 */
333 	if (skb_dst(skb) == NULL) {
334 		int err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
335 					       iph->tos, skb->dev);
336 		if (unlikely(err)) {
337 			if (err == -EHOSTUNREACH)
338 				IP_INC_STATS_BH(dev_net(skb->dev),
339 						IPSTATS_MIB_INADDRERRORS);
340 			else if (err == -ENETUNREACH)
341 				IP_INC_STATS_BH(dev_net(skb->dev),
342 						IPSTATS_MIB_INNOROUTES);
343 			goto drop;
344 		}
345 	}
346 
347 #ifdef CONFIG_NET_CLS_ROUTE
348 	if (unlikely(skb_dst(skb)->tclassid)) {
349 		struct ip_rt_acct *st = per_cpu_ptr(ip_rt_acct, smp_processor_id());
350 		u32 idx = skb_dst(skb)->tclassid;
351 		st[idx&0xFF].o_packets++;
352 		st[idx&0xFF].o_bytes += skb->len;
353 		st[(idx>>16)&0xFF].i_packets++;
354 		st[(idx>>16)&0xFF].i_bytes += skb->len;
355 	}
356 #endif
357 
358 	if (iph->ihl > 5 && ip_rcv_options(skb))
359 		goto drop;
360 
361 	rt = skb_rtable(skb);
362 	if (rt->rt_type == RTN_MULTICAST) {
363 		IP_UPD_PO_STATS_BH(dev_net(rt->u.dst.dev), IPSTATS_MIB_INMCAST,
364 				skb->len);
365 	} else if (rt->rt_type == RTN_BROADCAST)
366 		IP_UPD_PO_STATS_BH(dev_net(rt->u.dst.dev), IPSTATS_MIB_INBCAST,
367 				skb->len);
368 
369 	return dst_input(skb);
370 
371 drop:
372 	kfree_skb(skb);
373 	return NET_RX_DROP;
374 }
375 
376 /*
377  * 	Main IP Receive routine.
378  */
379 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
380 {
381 	struct iphdr *iph;
382 	u32 len;
383 
384 	/* When the interface is in promisc. mode, drop all the crap
385 	 * that it receives, do not try to analyse it.
386 	 */
387 	if (skb->pkt_type == PACKET_OTHERHOST)
388 		goto drop;
389 
390 
391 	IP_UPD_PO_STATS_BH(dev_net(dev), IPSTATS_MIB_IN, skb->len);
392 
393 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) {
394 		IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
395 		goto out;
396 	}
397 
398 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
399 		goto inhdr_error;
400 
401 	iph = ip_hdr(skb);
402 
403 	/*
404 	 *	RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
405 	 *
406 	 *	Is the datagram acceptable?
407 	 *
408 	 *	1.	Length at least the size of an ip header
409 	 *	2.	Version of 4
410 	 *	3.	Checksums correctly. [Speed optimisation for later, skip loopback checksums]
411 	 *	4.	Doesn't have a bogus length
412 	 */
413 
414 	if (iph->ihl < 5 || iph->version != 4)
415 		goto inhdr_error;
416 
417 	if (!pskb_may_pull(skb, iph->ihl*4))
418 		goto inhdr_error;
419 
420 	iph = ip_hdr(skb);
421 
422 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
423 		goto inhdr_error;
424 
425 	len = ntohs(iph->tot_len);
426 	if (skb->len < len) {
427 		IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
428 		goto drop;
429 	} else if (len < (iph->ihl*4))
430 		goto inhdr_error;
431 
432 	/* Our transport medium may have padded the buffer out. Now we know it
433 	 * is IP we can trim to the true length of the frame.
434 	 * Note this now means skb->len holds ntohs(iph->tot_len).
435 	 */
436 	if (pskb_trim_rcsum(skb, len)) {
437 		IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
438 		goto drop;
439 	}
440 
441 	/* Remove any debris in the socket control block */
442 	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
443 
444 	/* Must drop socket now because of tproxy. */
445 	skb_orphan(skb);
446 
447 	return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, dev, NULL,
448 		       ip_rcv_finish);
449 
450 inhdr_error:
451 	IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
452 drop:
453 	kfree_skb(skb);
454 out:
455 	return NET_RX_DROP;
456 }
457