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.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Donald Becker, <becker@super.org>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
13 * Stefan Becker, <stefanb@yello.ping.de>
14 * Jorge Cwik, <jorge@laser.satlink.net>
15 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
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
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
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.
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.
116 #define pr_fmt(fmt) "IPv4: " fmt
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>
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 <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>
152 * Process Router Attention IP option (RFC 2113)
154 bool ip_call_ra_chain(struct sk_buff *skb)
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);
162 for (ra = rcu_dereference(net->ipv4.ra_chain); ra; ra = rcu_dereference(ra->next)) {
163 struct sock *sk = ra->sk;
165 /* If socket is bound to an interface, only report
166 * the packet if it came from that interface.
168 if (sk && inet_sk(sk)->inet_num == protocol &&
169 (!sk->sk_bound_dev_if ||
170 sk->sk_bound_dev_if == dev->ifindex)) {
171 if (ip_is_fragment(ip_hdr(skb))) {
172 if (ip_defrag(net, skb, IP_DEFRAG_CALL_RA_CHAIN))
176 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
191 static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
193 __skb_pull(skb, skb_network_header_len(skb));
197 int protocol = ip_hdr(skb)->protocol;
198 const struct net_protocol *ipprot;
202 raw = raw_local_deliver(skb, protocol);
204 ipprot = rcu_dereference(inet_protos[protocol]);
208 if (!ipprot->no_policy) {
209 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
215 ret = ipprot->handler(skb);
220 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
223 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
224 __IP_INC_STATS(net, IPSTATS_MIB_INUNKNOWNPROTOS);
225 icmp_send(skb, ICMP_DEST_UNREACH,
226 ICMP_PROT_UNREACH, 0);
230 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
242 * Deliver IP Packets to the higher protocol layers.
244 int ip_local_deliver(struct sk_buff *skb)
247 * Reassemble IP fragments.
249 struct net *net = dev_net(skb->dev);
251 if (ip_is_fragment(ip_hdr(skb))) {
252 if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
256 return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
257 net, NULL, skb, skb->dev, NULL,
258 ip_local_deliver_finish);
261 static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev)
263 struct ip_options *opt;
264 const struct iphdr *iph;
266 /* It looks as overkill, because not all
267 IP options require packet mangling.
268 But it is the easiest for now, especially taking
269 into account that combination of IP options
270 and running sniffer is extremely rare condition.
273 if (skb_cow(skb, skb_headroom(skb))) {
274 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
279 opt = &(IPCB(skb)->opt);
280 opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
282 if (ip_options_compile(dev_net(dev), opt, skb)) {
283 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
287 if (unlikely(opt->srr)) {
288 struct in_device *in_dev = __in_dev_get_rcu(dev);
291 if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
292 if (IN_DEV_LOG_MARTIANS(in_dev))
293 net_info_ratelimited("source route option %pI4 -> %pI4\n",
300 if (ip_options_rcv_srr(skb, dev))
309 int udp_v4_early_demux(struct sk_buff *);
310 int tcp_v4_early_demux(struct sk_buff *);
311 static int ip_rcv_finish_core(struct net *net, struct sock *sk,
312 struct sk_buff *skb, struct net_device *dev)
314 const struct iphdr *iph = ip_hdr(skb);
318 if (READ_ONCE(net->ipv4.sysctl_ip_early_demux) &&
321 !ip_is_fragment(iph)) {
322 switch (iph->protocol) {
324 if (READ_ONCE(net->ipv4.sysctl_tcp_early_demux)) {
325 tcp_v4_early_demux(skb);
327 /* must reload iph, skb->head might have changed */
332 if (READ_ONCE(net->ipv4.sysctl_udp_early_demux)) {
333 err = udp_v4_early_demux(skb);
337 /* must reload iph, skb->head might have changed */
345 * Initialise the virtual path cache for the packet. It describes
346 * how the packet travels inside Linux networking.
348 if (!skb_valid_dst(skb)) {
349 err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
355 #ifdef CONFIG_IP_ROUTE_CLASSID
356 if (unlikely(skb_dst(skb)->tclassid)) {
357 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
358 u32 idx = skb_dst(skb)->tclassid;
359 st[idx&0xFF].o_packets++;
360 st[idx&0xFF].o_bytes += skb->len;
361 st[(idx>>16)&0xFF].i_packets++;
362 st[(idx>>16)&0xFF].i_bytes += skb->len;
366 if (iph->ihl > 5 && ip_rcv_options(skb, dev))
369 rt = skb_rtable(skb);
370 if (rt->rt_type == RTN_MULTICAST) {
371 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
372 } else if (rt->rt_type == RTN_BROADCAST) {
373 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
374 } else if (skb->pkt_type == PACKET_BROADCAST ||
375 skb->pkt_type == PACKET_MULTICAST) {
376 struct in_device *in_dev = __in_dev_get_rcu(dev);
380 * When a host sends a datagram to a link-layer broadcast
381 * address, the IP destination address MUST be a legal IP
382 * broadcast or IP multicast address.
384 * A host SHOULD silently discard a datagram that is received
385 * via a link-layer broadcast (see Section 2.4) but does not
386 * specify an IP multicast or broadcast destination address.
388 * This doesn't explicitly say L2 *broadcast*, but broadcast is
389 * in a way a form of multicast and the most common use case for
390 * this is 802.11 protecting against cross-station spoofing (the
391 * so-called "hole-196" attack) so do it for both.
394 IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST))
398 return NET_RX_SUCCESS;
406 __NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
410 static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
412 struct net_device *dev = skb->dev;
415 /* if ingress device is enslaved to an L3 master device pass the
416 * skb to its handler for processing
418 skb = l3mdev_ip_rcv(skb);
420 return NET_RX_SUCCESS;
422 ret = ip_rcv_finish_core(net, sk, skb, dev);
423 if (ret != NET_RX_DROP)
424 ret = dst_input(skb);
429 * Main IP Receive routine.
431 static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net)
433 const struct iphdr *iph;
436 /* When the interface is in promisc. mode, drop all the crap
437 * that it receives, do not try to analyse it.
439 if (skb->pkt_type == PACKET_OTHERHOST)
443 __IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
445 skb = skb_share_check(skb, GFP_ATOMIC);
447 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
451 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
457 * RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
459 * Is the datagram acceptable?
461 * 1. Length at least the size of an ip header
463 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums]
464 * 4. Doesn't have a bogus length
467 if (iph->ihl < 5 || iph->version != 4)
470 BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
471 BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
472 BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
474 IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
475 max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
477 if (!pskb_may_pull(skb, iph->ihl*4))
482 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
485 len = ntohs(iph->tot_len);
486 if (skb->len < len) {
487 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
489 } else if (len < (iph->ihl*4))
492 /* Our transport medium may have padded the buffer out. Now we know it
493 * is IP we can trim to the true length of the frame.
494 * Note this now means skb->len holds ntohs(iph->tot_len).
496 if (pskb_trim_rcsum(skb, len)) {
497 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
502 skb->transport_header = skb->network_header + iph->ihl*4;
504 /* Remove any debris in the socket control block */
505 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
506 IPCB(skb)->iif = skb->skb_iif;
508 /* Must drop socket now because of tproxy. */
514 __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
516 __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
524 * IP receive entry point
526 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
527 struct net_device *orig_dev)
529 struct net *net = dev_net(dev);
531 skb = ip_rcv_core(skb, net);
534 return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
535 net, NULL, skb, dev, NULL,
539 static void ip_sublist_rcv_finish(struct list_head *head)
541 struct sk_buff *skb, *next;
543 list_for_each_entry_safe(skb, next, head, list) {
544 list_del(&skb->list);
545 /* Handle ip{6}_forward case, as sch_direct_xmit have
546 * another kind of SKB-list usage (see validate_xmit_skb_list)
553 static void ip_list_rcv_finish(struct net *net, struct sock *sk,
554 struct list_head *head)
556 struct dst_entry *curr_dst = NULL;
557 struct sk_buff *skb, *next;
558 struct list_head sublist;
560 INIT_LIST_HEAD(&sublist);
561 list_for_each_entry_safe(skb, next, head, list) {
562 struct net_device *dev = skb->dev;
563 struct dst_entry *dst;
565 skb_list_del_init(skb);
566 /* if ingress device is enslaved to an L3 master device pass the
567 * skb to its handler for processing
569 skb = l3mdev_ip_rcv(skb);
572 if (ip_rcv_finish_core(net, sk, skb, dev) == NET_RX_DROP)
576 if (curr_dst != dst) {
577 /* dispatch old sublist */
578 if (!list_empty(&sublist))
579 ip_sublist_rcv_finish(&sublist);
580 /* start new sublist */
581 INIT_LIST_HEAD(&sublist);
584 list_add_tail(&skb->list, &sublist);
586 /* dispatch final sublist */
587 ip_sublist_rcv_finish(&sublist);
590 static void ip_sublist_rcv(struct list_head *head, struct net_device *dev,
593 NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL,
594 head, dev, NULL, ip_rcv_finish);
595 ip_list_rcv_finish(net, NULL, head);
598 /* Receive a list of IP packets */
599 void ip_list_rcv(struct list_head *head, struct packet_type *pt,
600 struct net_device *orig_dev)
602 struct net_device *curr_dev = NULL;
603 struct net *curr_net = NULL;
604 struct sk_buff *skb, *next;
605 struct list_head sublist;
607 INIT_LIST_HEAD(&sublist);
608 list_for_each_entry_safe(skb, next, head, list) {
609 struct net_device *dev = skb->dev;
610 struct net *net = dev_net(dev);
612 skb_list_del_init(skb);
613 skb = ip_rcv_core(skb, net);
617 if (curr_dev != dev || curr_net != net) {
618 /* dispatch old sublist */
619 if (!list_empty(&sublist))
620 ip_sublist_rcv(&sublist, curr_dev, curr_net);
621 /* start new sublist */
622 INIT_LIST_HEAD(&sublist);
626 list_add_tail(&skb->list, &sublist);
628 /* dispatch final sublist */
629 ip_sublist_rcv(&sublist, curr_dev, curr_net);