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>
133 #include <linux/indirect_call_wrapper.h>
135 #include <net/snmp.h>
137 #include <net/protocol.h>
138 #include <net/route.h>
139 #include <linux/skbuff.h>
140 #include <net/sock.h>
142 #include <net/icmp.h>
144 #include <net/checksum.h>
145 #include <net/inet_ecn.h>
146 #include <linux/netfilter_ipv4.h>
147 #include <net/xfrm.h>
148 #include <linux/mroute.h>
149 #include <linux/netlink.h>
150 #include <net/dst_metadata.h>
153 * Process Router Attention IP option (RFC 2113)
155 bool ip_call_ra_chain(struct sk_buff *skb)
157 struct ip_ra_chain *ra;
158 u8 protocol = ip_hdr(skb)->protocol;
159 struct sock *last = NULL;
160 struct net_device *dev = skb->dev;
161 struct net *net = dev_net(dev);
163 for (ra = rcu_dereference(net->ipv4.ra_chain); ra; ra = rcu_dereference(ra->next)) {
164 struct sock *sk = ra->sk;
166 /* If socket is bound to an interface, only report
167 * the packet if it came from that interface.
169 if (sk && inet_sk(sk)->inet_num == protocol &&
170 (!sk->sk_bound_dev_if ||
171 sk->sk_bound_dev_if == dev->ifindex)) {
172 if (ip_is_fragment(ip_hdr(skb))) {
173 if (ip_defrag(net, skb, IP_DEFRAG_CALL_RA_CHAIN))
177 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
192 INDIRECT_CALLABLE_DECLARE(int udp_rcv(struct sk_buff *));
193 INDIRECT_CALLABLE_DECLARE(int tcp_v4_rcv(struct sk_buff *));
194 void ip_protocol_deliver_rcu(struct net *net, struct sk_buff *skb, int protocol)
196 const struct net_protocol *ipprot;
200 raw = raw_local_deliver(skb, protocol);
202 ipprot = rcu_dereference(inet_protos[protocol]);
204 if (!ipprot->no_policy) {
205 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
211 ret = INDIRECT_CALL_2(ipprot->handler, tcp_v4_rcv, udp_rcv,
217 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
220 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
221 __IP_INC_STATS(net, IPSTATS_MIB_INUNKNOWNPROTOS);
222 icmp_send(skb, ICMP_DEST_UNREACH,
223 ICMP_PROT_UNREACH, 0);
227 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
233 static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
235 __skb_pull(skb, skb_network_header_len(skb));
238 ip_protocol_deliver_rcu(net, skb, ip_hdr(skb)->protocol);
245 * Deliver IP Packets to the higher protocol layers.
247 int ip_local_deliver(struct sk_buff *skb)
250 * Reassemble IP fragments.
252 struct net *net = dev_net(skb->dev);
254 if (ip_is_fragment(ip_hdr(skb))) {
255 if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
259 return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
260 net, NULL, skb, skb->dev, NULL,
261 ip_local_deliver_finish);
264 static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev)
266 struct ip_options *opt;
267 const struct iphdr *iph;
269 /* It looks as overkill, because not all
270 IP options require packet mangling.
271 But it is the easiest for now, especially taking
272 into account that combination of IP options
273 and running sniffer is extremely rare condition.
276 if (skb_cow(skb, skb_headroom(skb))) {
277 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
282 opt = &(IPCB(skb)->opt);
283 opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
285 if (ip_options_compile(dev_net(dev), opt, skb)) {
286 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
290 if (unlikely(opt->srr)) {
291 struct in_device *in_dev = __in_dev_get_rcu(dev);
294 if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
295 if (IN_DEV_LOG_MARTIANS(in_dev))
296 net_info_ratelimited("source route option %pI4 -> %pI4\n",
303 if (ip_options_rcv_srr(skb, dev))
312 static int ip_rcv_finish_core(struct net *net, struct sock *sk,
313 struct sk_buff *skb, struct net_device *dev)
315 const struct iphdr *iph = ip_hdr(skb);
316 int (*edemux)(struct sk_buff *skb);
320 if (net->ipv4.sysctl_ip_early_demux &&
323 !ip_is_fragment(iph)) {
324 const struct net_protocol *ipprot;
325 int protocol = iph->protocol;
327 ipprot = rcu_dereference(inet_protos[protocol]);
328 if (ipprot && (edemux = READ_ONCE(ipprot->early_demux))) {
332 /* must reload iph, skb->head might have changed */
338 * Initialise the virtual path cache for the packet. It describes
339 * how the packet travels inside Linux networking.
341 if (!skb_valid_dst(skb)) {
342 err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
348 #ifdef CONFIG_IP_ROUTE_CLASSID
349 if (unlikely(skb_dst(skb)->tclassid)) {
350 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
351 u32 idx = skb_dst(skb)->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;
359 if (iph->ihl > 5 && ip_rcv_options(skb, dev))
362 rt = skb_rtable(skb);
363 if (rt->rt_type == RTN_MULTICAST) {
364 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
365 } else if (rt->rt_type == RTN_BROADCAST) {
366 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
367 } else if (skb->pkt_type == PACKET_BROADCAST ||
368 skb->pkt_type == PACKET_MULTICAST) {
369 struct in_device *in_dev = __in_dev_get_rcu(dev);
373 * When a host sends a datagram to a link-layer broadcast
374 * address, the IP destination address MUST be a legal IP
375 * broadcast or IP multicast address.
377 * A host SHOULD silently discard a datagram that is received
378 * via a link-layer broadcast (see Section 2.4) but does not
379 * specify an IP multicast or broadcast destination address.
381 * This doesn't explicitly say L2 *broadcast*, but broadcast is
382 * in a way a form of multicast and the most common use case for
383 * this is 802.11 protecting against cross-station spoofing (the
384 * so-called "hole-196" attack) so do it for both.
387 IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST))
391 return NET_RX_SUCCESS;
399 __NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
403 static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
405 struct net_device *dev = skb->dev;
408 /* if ingress device is enslaved to an L3 master device pass the
409 * skb to its handler for processing
411 skb = l3mdev_ip_rcv(skb);
413 return NET_RX_SUCCESS;
415 ret = ip_rcv_finish_core(net, sk, skb, dev);
416 if (ret != NET_RX_DROP)
417 ret = dst_input(skb);
422 * Main IP Receive routine.
424 static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net)
426 const struct iphdr *iph;
429 /* When the interface is in promisc. mode, drop all the crap
430 * that it receives, do not try to analyse it.
432 if (skb->pkt_type == PACKET_OTHERHOST)
435 __IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
437 skb = skb_share_check(skb, GFP_ATOMIC);
439 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
443 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
449 * RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
451 * Is the datagram acceptable?
453 * 1. Length at least the size of an ip header
455 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums]
456 * 4. Doesn't have a bogus length
459 if (iph->ihl < 5 || iph->version != 4)
462 BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
463 BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
464 BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
466 IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
467 max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
469 if (!pskb_may_pull(skb, iph->ihl*4))
474 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
477 len = ntohs(iph->tot_len);
478 if (skb->len < len) {
479 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
481 } else if (len < (iph->ihl*4))
484 /* Our transport medium may have padded the buffer out. Now we know it
485 * is IP we can trim to the true length of the frame.
486 * Note this now means skb->len holds ntohs(iph->tot_len).
488 if (pskb_trim_rcsum(skb, len)) {
489 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
494 skb->transport_header = skb->network_header + iph->ihl*4;
496 /* Remove any debris in the socket control block */
497 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
498 IPCB(skb)->iif = skb->skb_iif;
500 /* Must drop socket now because of tproxy. */
506 __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
508 __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
516 * IP receive entry point
518 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
519 struct net_device *orig_dev)
521 struct net *net = dev_net(dev);
523 skb = ip_rcv_core(skb, net);
527 return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
528 net, NULL, skb, dev, NULL,
532 static void ip_sublist_rcv_finish(struct list_head *head)
534 struct sk_buff *skb, *next;
536 list_for_each_entry_safe(skb, next, head, list) {
537 skb_list_del_init(skb);
542 static void ip_list_rcv_finish(struct net *net, struct sock *sk,
543 struct list_head *head)
545 struct dst_entry *curr_dst = NULL;
546 struct sk_buff *skb, *next;
547 struct list_head sublist;
549 INIT_LIST_HEAD(&sublist);
550 list_for_each_entry_safe(skb, next, head, list) {
551 struct net_device *dev = skb->dev;
552 struct dst_entry *dst;
554 skb_list_del_init(skb);
555 /* if ingress device is enslaved to an L3 master device pass the
556 * skb to its handler for processing
558 skb = l3mdev_ip_rcv(skb);
561 if (ip_rcv_finish_core(net, sk, skb, dev) == NET_RX_DROP)
565 if (curr_dst != dst) {
566 /* dispatch old sublist */
567 if (!list_empty(&sublist))
568 ip_sublist_rcv_finish(&sublist);
569 /* start new sublist */
570 INIT_LIST_HEAD(&sublist);
573 list_add_tail(&skb->list, &sublist);
575 /* dispatch final sublist */
576 ip_sublist_rcv_finish(&sublist);
579 static void ip_sublist_rcv(struct list_head *head, struct net_device *dev,
582 NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL,
583 head, dev, NULL, ip_rcv_finish);
584 ip_list_rcv_finish(net, NULL, head);
587 /* Receive a list of IP packets */
588 void ip_list_rcv(struct list_head *head, struct packet_type *pt,
589 struct net_device *orig_dev)
591 struct net_device *curr_dev = NULL;
592 struct net *curr_net = NULL;
593 struct sk_buff *skb, *next;
594 struct list_head sublist;
596 INIT_LIST_HEAD(&sublist);
597 list_for_each_entry_safe(skb, next, head, list) {
598 struct net_device *dev = skb->dev;
599 struct net *net = dev_net(dev);
601 skb_list_del_init(skb);
602 skb = ip_rcv_core(skb, net);
606 if (curr_dev != dev || curr_net != net) {
607 /* dispatch old sublist */
608 if (!list_empty(&sublist))
609 ip_sublist_rcv(&sublist, curr_dev, curr_net);
610 /* start new sublist */
611 INIT_LIST_HEAD(&sublist);
615 list_add_tail(&skb->list, &sublist);
617 /* dispatch final sublist */
618 ip_sublist_rcv(&sublist, curr_dev, curr_net);