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.
6 * The Internet Protocol (IP) module.
8 * Version: $Id: ip_input.c,v 1.55 2002/01/12 07:39:45 davem Exp $
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Donald Becker, <becker@super.org>
13 * Alan Cox, <Alan.Cox@linux.org>
15 * Stefan Becker, <stefanb@yello.ping.de>
16 * Jorge Cwik, <jorge@laser.satlink.net>
17 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
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
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
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.
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.
118 #include <asm/system.h>
119 #include <linux/module.h>
120 #include <linux/types.h>
121 #include <linux/kernel.h>
122 #include <linux/string.h>
123 #include <linux/errno.h>
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>
134 #include <net/snmp.h>
136 #include <net/protocol.h>
137 #include <net/route.h>
138 #include <linux/skbuff.h>
139 #include <net/sock.h>
141 #include <net/icmp.h>
143 #include <net/checksum.h>
144 #include <linux/netfilter_ipv4.h>
145 #include <net/xfrm.h>
146 #include <linux/mroute.h>
147 #include <linux/netlink.h>
150 * SNMP management statistics
153 DEFINE_SNMP_STAT(struct ipstats_mib, ip_statistics) __read_mostly;
156 * Process Router Attention IP option
158 int ip_call_ra_chain(struct sk_buff *skb)
160 struct ip_ra_chain *ra;
161 u8 protocol = ip_hdr(skb)->protocol;
162 struct sock *last = NULL;
163 struct net_device *dev = skb->dev;
165 read_lock(&ip_ra_lock);
166 for (ra = ip_ra_chain; ra; ra = ra->next) {
167 struct sock *sk = ra->sk;
169 /* If socket is bound to an interface, only report
170 * the packet if it came from that interface.
172 if (sk && inet_sk(sk)->num == protocol &&
173 (!sk->sk_bound_dev_if ||
174 sk->sk_bound_dev_if == dev->ifindex) &&
175 sock_net(sk) == dev_net(dev)) {
176 if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
177 if (ip_defrag(skb, IP_DEFRAG_CALL_RA_CHAIN)) {
178 read_unlock(&ip_ra_lock);
183 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
193 read_unlock(&ip_ra_lock);
196 read_unlock(&ip_ra_lock);
200 static int ip_local_deliver_finish(struct sk_buff *skb)
202 struct net *net = dev_net(skb->dev);
204 __skb_pull(skb, ip_hdrlen(skb));
206 /* Point into the IP datagram, just past the header. */
207 skb_reset_transport_header(skb);
211 int protocol = ip_hdr(skb)->protocol;
213 struct net_protocol *ipprot;
216 raw = raw_local_deliver(skb, protocol);
218 hash = protocol & (MAX_INET_PROTOS - 1);
219 ipprot = rcu_dereference(inet_protos[hash]);
220 if (ipprot != NULL && (net == &init_net || ipprot->netns_ok)) {
223 if (!ipprot->no_policy) {
224 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
230 ret = ipprot->handler(skb);
235 IP_INC_STATS_BH(IPSTATS_MIB_INDELIVERS);
238 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
239 IP_INC_STATS_BH(IPSTATS_MIB_INUNKNOWNPROTOS);
240 icmp_send(skb, ICMP_DEST_UNREACH,
241 ICMP_PROT_UNREACH, 0);
244 IP_INC_STATS_BH(IPSTATS_MIB_INDELIVERS);
255 * Deliver IP Packets to the higher protocol layers.
257 int ip_local_deliver(struct sk_buff *skb)
260 * Reassemble IP fragments.
263 if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
264 if (ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER))
268 return NF_HOOK(PF_INET, NF_INET_LOCAL_IN, skb, skb->dev, NULL,
269 ip_local_deliver_finish);
272 static inline int ip_rcv_options(struct sk_buff *skb)
274 struct ip_options *opt;
276 struct net_device *dev = skb->dev;
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.
285 if (skb_cow(skb, skb_headroom(skb))) {
286 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
291 opt = &(IPCB(skb)->opt);
292 opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
294 if (ip_options_compile(dev_net(dev), opt, skb)) {
295 IP_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
299 if (unlikely(opt->srr)) {
300 struct in_device *in_dev = in_dev_get(dev);
302 if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
303 if (IN_DEV_LOG_MARTIANS(in_dev) &&
305 printk(KERN_INFO "source route option "
306 "%u.%u.%u.%u -> %u.%u.%u.%u\n",
308 NIPQUAD(iph->daddr));
316 if (ip_options_rcv_srr(skb))
325 static int ip_rcv_finish(struct sk_buff *skb)
327 const struct iphdr *iph = ip_hdr(skb);
331 * Initialise the virtual path cache for the packet. It describes
332 * how the packet travels inside Linux networking.
334 if (skb->dst == NULL) {
335 int err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos,
338 if (err == -EHOSTUNREACH)
339 IP_INC_STATS_BH(IPSTATS_MIB_INADDRERRORS);
340 else if (err == -ENETUNREACH)
341 IP_INC_STATS_BH(IPSTATS_MIB_INNOROUTES);
346 #ifdef CONFIG_NET_CLS_ROUTE
347 if (unlikely(skb->dst->tclassid)) {
348 struct ip_rt_acct *st = per_cpu_ptr(ip_rt_acct, smp_processor_id());
349 u32 idx = skb->dst->tclassid;
350 st[idx&0xFF].o_packets++;
351 st[idx&0xFF].o_bytes+=skb->len;
352 st[(idx>>16)&0xFF].i_packets++;
353 st[(idx>>16)&0xFF].i_bytes+=skb->len;
357 if (iph->ihl > 5 && ip_rcv_options(skb))
361 if (rt->rt_type == RTN_MULTICAST)
362 IP_INC_STATS_BH(IPSTATS_MIB_INMCASTPKTS);
363 else if (rt->rt_type == RTN_BROADCAST)
364 IP_INC_STATS_BH(IPSTATS_MIB_INBCASTPKTS);
366 return dst_input(skb);
374 * Main IP Receive routine.
376 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
381 /* When the interface is in promisc. mode, drop all the crap
382 * that it receives, do not try to analyse it.
384 if (skb->pkt_type == PACKET_OTHERHOST)
387 IP_INC_STATS_BH(IPSTATS_MIB_INRECEIVES);
389 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) {
390 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
394 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
400 * RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the checksum.
402 * Is the datagram acceptable?
404 * 1. Length at least the size of an ip header
406 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums]
407 * 4. Doesn't have a bogus length
410 if (iph->ihl < 5 || iph->version != 4)
413 if (!pskb_may_pull(skb, iph->ihl*4))
418 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
421 len = ntohs(iph->tot_len);
422 if (skb->len < len) {
423 IP_INC_STATS_BH(IPSTATS_MIB_INTRUNCATEDPKTS);
425 } else if (len < (iph->ihl*4))
428 /* Our transport medium may have padded the buffer out. Now we know it
429 * is IP we can trim to the true length of the frame.
430 * Note this now means skb->len holds ntohs(iph->tot_len).
432 if (pskb_trim_rcsum(skb, len)) {
433 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
437 /* Remove any debris in the socket control block */
438 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
440 return NF_HOOK(PF_INET, NF_INET_PRE_ROUTING, skb, dev, NULL,
444 IP_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
451 EXPORT_SYMBOL(ip_statistics);