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 * Version: $Id: ip_input.c,v 1.54.2.1 2002/01/12 07:39:23 davem Exp $
9 *
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Donald Becker, <becker@super.org>
13 * Alan Cox, <Alan.Cox@linux.org>
14 * Richard Underwood
15 * Stefan Becker, <stefanb@yello.ping.de>
16 * Jorge Cwik, <jorge@laser.satlink.net>
17 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
18 *
19 *
20 * Fixes:
21 * Alan Cox : Commented a couple of minor bits of surplus code
22 * Alan Cox : Undefining IP_FORWARD doesn't include the code
23 * (just stops a compiler warning).
24 * Alan Cox : Frames with >=MAX_ROUTE record routes, strict routes or loose routes
25 * are junked rather than corrupting things.
26 * Alan Cox : Frames to bad broadcast subnets are dumped
27 * We used to process them non broadcast and
28 * boy could that cause havoc.
29 * Alan Cox : ip_forward sets the free flag on the
30 * new frame it queues. Still crap because
31 * it copies the frame but at least it
32 * doesn't eat memory too.
33 * Alan Cox : Generic queue code and memory fixes.
34 * Fred Van Kempen : IP fragment support (borrowed from NET2E)
35 * Gerhard Koerting: Forward fragmented frames correctly.
36 * Gerhard Koerting: Fixes to my fix of the above 8-).
37 * Gerhard Koerting: IP interface addressing fix.
38 * Linus Torvalds : More robustness checks
39 * Alan Cox : Even more checks: Still not as robust as it ought to be
40 * Alan Cox : Save IP header pointer for later
41 * Alan Cox : ip option setting
42 * Alan Cox : Use ip_tos/ip_ttl settings
43 * Alan Cox : Fragmentation bogosity removed
44 * (Thanks to Mark.Bush@prg.ox.ac.uk)
45 * Dmitry Gorodchanin : Send of a raw packet crash fix.
46 * Alan Cox : Silly ip bug when an overlength
47 * fragment turns up. Now frees the
48 * queue.
49 * Linus Torvalds/ : Memory leakage on fragmentation
50 * Alan Cox : handling.
51 * Gerhard Koerting: Forwarding uses IP priority hints
52 * Teemu Rantanen : Fragment problems.
53 * Alan Cox : General cleanup, comments and reformat
54 * Alan Cox : SNMP statistics
55 * Alan Cox : BSD address rule semantics. Also see
56 * UDP as there is a nasty checksum issue
57 * if you do things the wrong way.
58 * Alan Cox : Always defrag, moved IP_FORWARD to the config.in file
59 * Alan Cox : IP options adjust sk->priority.
60 * Pedro Roque : Fix mtu/length error in ip_forward.
61 * Alan Cox : Avoid ip_chk_addr when possible.
62 * Richard Underwood : IP multicasting.
63 * Alan Cox : Cleaned up multicast handlers.
64 * Alan Cox : RAW sockets demultiplex in the BSD style.
65 * Gunther Mayer : Fix the SNMP reporting typo
66 * Alan Cox : Always in group 224.0.0.1
67 * Pauline Middelink : Fast ip_checksum update when forwarding
68 * Masquerading support.
69 * Alan Cox : Multicast loopback error for 224.0.0.1
70 * Alan Cox : IP_MULTICAST_LOOP option.
71 * Alan Cox : Use notifiers.
72 * Bjorn Ekwall : Removed ip_csum (from slhc.c too)
73 * Bjorn Ekwall : Moved ip_fast_csum to ip.h (inline!)
74 * Stefan Becker : Send out ICMP HOST REDIRECT
75 * Arnt Gulbrandsen : ip_build_xmit
76 * Alan Cox : Per socket routing cache
77 * Alan Cox : Fixed routing cache, added header cache.
78 * Alan Cox : Loopback didn't work right in original ip_build_xmit - fixed it.
79 * Alan Cox : Only send ICMP_REDIRECT if src/dest are the same net.
80 * Alan Cox : Incoming IP option handling.
81 * Alan Cox : Set saddr on raw output frames as per BSD.
82 * Alan Cox : Stopped broadcast source route explosions.
83 * Alan Cox : Can disable source routing
84 * Takeshi Sone : Masquerading didn't work.
85 * Dave Bonn,Alan Cox : Faster IP forwarding whenever possible.
86 * Alan Cox : Memory leaks, tramples, misc debugging.
87 * Alan Cox : Fixed multicast (by popular demand 8))
88 * Alan Cox : Fixed forwarding (by even more popular demand 8))
89 * Alan Cox : Fixed SNMP statistics [I think]
90 * Gerhard Koerting : IP fragmentation forwarding fix
91 * Alan Cox : Device lock against page fault.
92 * Alan Cox : IP_HDRINCL facility.
93 * Werner Almesberger : Zero fragment bug
94 * Alan Cox : RAW IP frame length bug
95 * Alan Cox : Outgoing firewall on build_xmit
96 * A.N.Kuznetsov : IP_OPTIONS support throughout the kernel
97 * Alan Cox : Multicast routing hooks
98 * Jos Vos : Do accounting *before* call_in_firewall
99 * Willy Konynenberg : Transparent proxying support
100 *
101 *
102 *
103 * To Fix:
104 * IP fragmentation wants rewriting cleanly. The RFC815 algorithm is much more efficient
105 * and could be made very efficient with the addition of some virtual memory hacks to permit
106 * the allocation of a buffer that can then be 'grown' by twiddling page tables.
107 * Output fragmentation wants updating along with the buffer management to use a single
108 * interleaved copy algorithm so that fragmenting has a one copy overhead. Actual packet
109 * output should probably do its own fragmentation at the UDP/RAW layer. TCP shouldn't cause
110 * fragmentation anyway.
111 *
112 * This program is free software; you can redistribute it and/or
113 * modify it under the terms of the GNU General Public License
114 * as published by the Free Software Foundation; either version
115 * 2 of the License, or (at your option) any later version.
116 */
117
118 #include <asm/system.h>
119 #include <linux/types.h>
120 #include <linux/kernel.h>
121 #include <linux/string.h>
122 #include <linux/errno.h>
123 #include <linux/config.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/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 <linux/mroute.h>
145 #include <linux/netlink.h>
146
147 /*
148 * SNMP management statistics
149 */
150
151 struct ip_mib ip_statistics[NR_CPUS*2];
152
153 /*
154 * Process Router Attention IP option
155 */
ip_call_ra_chain(struct sk_buff * skb)156 int ip_call_ra_chain(struct sk_buff *skb)
157 {
158 struct ip_ra_chain *ra;
159 u8 protocol = skb->nh.iph->protocol;
160 struct sock *last = NULL;
161
162 read_lock(&ip_ra_lock);
163 for (ra = ip_ra_chain; ra; ra = ra->next) {
164 struct sock *sk = ra->sk;
165
166 /* If socket is bound to an interface, only report
167 * the packet if it came from that interface.
168 */
169 if (sk && sk->num == protocol
170 && ((sk->bound_dev_if == 0)
171 || (sk->bound_dev_if == skb->dev->ifindex))) {
172 if (skb->nh.iph->frag_off & htons(IP_MF|IP_OFFSET)) {
173 skb = ip_defrag(skb, IP_DEFRAG_CALL_RA_CHAIN);
174 if (skb == NULL) {
175 read_unlock(&ip_ra_lock);
176 return 1;
177 }
178 }
179 if (last) {
180 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
181 if (skb2)
182 raw_rcv(last, skb2);
183 }
184 last = sk;
185 }
186 }
187
188 if (last) {
189 raw_rcv(last, skb);
190 read_unlock(&ip_ra_lock);
191 return 1;
192 }
193 read_unlock(&ip_ra_lock);
194 return 0;
195 }
196
197 /* Handle this out of line, it is rare. */
ip_run_ipprot(struct sk_buff * skb,struct iphdr * iph,struct inet_protocol * ipprot,int force_copy)198 static int ip_run_ipprot(struct sk_buff *skb, struct iphdr *iph,
199 struct inet_protocol *ipprot, int force_copy)
200 {
201 int ret = 0;
202
203 do {
204 if (ipprot->protocol == iph->protocol) {
205 struct sk_buff *skb2 = skb;
206 if (ipprot->copy || force_copy)
207 skb2 = skb_clone(skb, GFP_ATOMIC);
208 if(skb2 != NULL) {
209 ret = 1;
210 ipprot->handler(skb2);
211 }
212 }
213 ipprot = (struct inet_protocol *) ipprot->next;
214 } while(ipprot != NULL);
215
216 return ret;
217 }
218
ip_local_deliver_finish(struct sk_buff * skb)219 static inline int ip_local_deliver_finish(struct sk_buff *skb)
220 {
221 int ihl = skb->nh.iph->ihl*4;
222
223 #ifdef CONFIG_NETFILTER_DEBUG
224 nf_debug_ip_local_deliver(skb);
225 #endif /*CONFIG_NETFILTER_DEBUG*/
226
227 __skb_pull(skb, ihl);
228
229 /* Free reference early: we don't need it any more, and it may
230 hold ip_conntrack module loaded indefinitely. */
231 nf_reset(skb);
232
233 /* Point into the IP datagram, just past the header. */
234 skb->h.raw = skb->data;
235
236 {
237 /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */
238 int protocol = skb->nh.iph->protocol;
239 int hash = protocol & (MAX_INET_PROTOS - 1);
240 struct sock *raw_sk = raw_v4_htable[hash];
241 struct inet_protocol *ipprot;
242 int flag;
243
244 /* If there maybe a raw socket we must check - if not we
245 * don't care less
246 */
247 if(raw_sk != NULL)
248 raw_sk = raw_v4_input(skb, skb->nh.iph, hash);
249
250 ipprot = (struct inet_protocol *) inet_protos[hash];
251 flag = 0;
252 if(ipprot != NULL) {
253 if(raw_sk == NULL &&
254 ipprot->next == NULL &&
255 ipprot->protocol == protocol) {
256 int ret;
257
258 /* Fast path... */
259 ret = ipprot->handler(skb);
260
261 return ret;
262 } else {
263 flag = ip_run_ipprot(skb, skb->nh.iph, ipprot, (raw_sk != NULL));
264 }
265 }
266
267 /* All protocols checked.
268 * If this packet was a broadcast, we may *not* reply to it, since that
269 * causes (proven, grin) ARP storms and a leakage of memory (i.e. all
270 * ICMP reply messages get queued up for transmission...)
271 */
272 if(raw_sk != NULL) { /* Shift to last raw user */
273 raw_rcv(raw_sk, skb);
274 sock_put(raw_sk);
275 } else if (!flag) { /* Free and report errors */
276 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PROT_UNREACH, 0);
277 kfree_skb(skb);
278 }
279 }
280
281 return 0;
282 }
283
284 /*
285 * Deliver IP Packets to the higher protocol layers.
286 */
ip_local_deliver(struct sk_buff * skb)287 int ip_local_deliver(struct sk_buff *skb)
288 {
289 /*
290 * Reassemble IP fragments.
291 */
292
293 if (skb->nh.iph->frag_off & htons(IP_MF|IP_OFFSET)) {
294 skb = ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER);
295 if (!skb)
296 return 0;
297 }
298
299 return NF_HOOK(PF_INET, NF_IP_LOCAL_IN, skb, skb->dev, NULL,
300 ip_local_deliver_finish);
301 }
302
ip_rcv_finish(struct sk_buff * skb)303 static inline int ip_rcv_finish(struct sk_buff *skb)
304 {
305 struct net_device *dev = skb->dev;
306 struct iphdr *iph = skb->nh.iph;
307
308 /*
309 * Initialise the virtual path cache for the packet. It describes
310 * how the packet travels inside Linux networking.
311 */
312 if (skb->dst == NULL) {
313 if (ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))
314 goto drop;
315 }
316
317 #ifdef CONFIG_NET_CLS_ROUTE
318 if (skb->dst->tclassid) {
319 struct ip_rt_acct *st = ip_rt_acct + 256*smp_processor_id();
320 u32 idx = skb->dst->tclassid;
321 st[idx&0xFF].o_packets++;
322 st[idx&0xFF].o_bytes+=skb->len;
323 st[(idx>>16)&0xFF].i_packets++;
324 st[(idx>>16)&0xFF].i_bytes+=skb->len;
325 }
326 #endif
327
328 if (iph->ihl > 5) {
329 struct ip_options *opt;
330
331 /* It looks as overkill, because not all
332 IP options require packet mangling.
333 But it is the easiest for now, especially taking
334 into account that combination of IP options
335 and running sniffer is extremely rare condition.
336 --ANK (980813)
337 */
338
339 if (skb_cow(skb, skb_headroom(skb)))
340 goto drop;
341 iph = skb->nh.iph;
342
343 if (ip_options_compile(NULL, skb))
344 goto inhdr_error;
345
346 opt = &(IPCB(skb)->opt);
347 if (opt->srr) {
348 struct in_device *in_dev = in_dev_get(dev);
349 if (in_dev) {
350 if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
351 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
352 printk(KERN_INFO "source route option %u.%u.%u.%u -> %u.%u.%u.%u\n",
353 NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
354 in_dev_put(in_dev);
355 goto drop;
356 }
357 in_dev_put(in_dev);
358 }
359 if (ip_options_rcv_srr(skb))
360 goto drop;
361 }
362 }
363
364 return skb->dst->input(skb);
365
366 inhdr_error:
367 IP_INC_STATS_BH(IpInHdrErrors);
368 drop:
369 kfree_skb(skb);
370 return NET_RX_DROP;
371 }
372
373 /*
374 * Main IP Receive routine.
375 */
ip_rcv(struct sk_buff * skb,struct net_device * dev,struct packet_type * pt)376 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt)
377 {
378 struct iphdr *iph;
379
380 /* When the interface is in promisc. mode, drop all the crap
381 * that it receives, do not try to analyse it.
382 */
383 if (skb->pkt_type == PACKET_OTHERHOST)
384 goto drop;
385
386 IP_INC_STATS_BH(IpInReceives);
387
388 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
389 goto out;
390
391 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
392 goto inhdr_error;
393
394 iph = skb->nh.iph;
395
396 /*
397 * RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the checksum.
398 *
399 * Is the datagram acceptable?
400 *
401 * 1. Length at least the size of an ip header
402 * 2. Version of 4
403 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums]
404 * 4. Doesn't have a bogus length
405 */
406
407 if (iph->ihl < 5 || iph->version != 4)
408 goto inhdr_error;
409
410 if (!pskb_may_pull(skb, iph->ihl*4))
411 goto inhdr_error;
412
413 iph = skb->nh.iph;
414
415 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
416 goto inhdr_error;
417
418 {
419 __u32 len = ntohs(iph->tot_len);
420 if (skb->len < len || len < (iph->ihl<<2))
421 goto inhdr_error;
422
423 /* Our transport medium may have padded the buffer out. Now we know it
424 * is IP we can trim to the true length of the frame.
425 * Note this now means skb->len holds ntohs(iph->tot_len).
426 */
427 if (skb->len > len) {
428 __pskb_trim(skb, len);
429 if (skb->ip_summed == CHECKSUM_HW)
430 skb->ip_summed = CHECKSUM_NONE;
431 }
432 }
433
434 return NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, dev, NULL,
435 ip_rcv_finish);
436
437 inhdr_error:
438 IP_INC_STATS_BH(IpInHdrErrors);
439 drop:
440 kfree_skb(skb);
441 out:
442 return NET_RX_DROP;
443 }
444
445