1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * The Internet Protocol (IP) module.
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Donald Becker, <becker@super.org>
12 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
14 * Stefan Becker, <stefanb@yello.ping.de>
15 * Jorge Cwik, <jorge@laser.satlink.net>
16 * 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
100 * IP fragmentation wants rewriting cleanly. The RFC815 algorithm is much more efficient
101 * and could be made very efficient with the addition of some virtual memory hacks to permit
102 * the allocation of a buffer that can then be 'grown' by twiddling page tables.
103 * Output fragmentation wants updating along with the buffer management to use a single
104 * interleaved copy algorithm so that fragmenting has a one copy overhead. Actual packet
105 * output should probably do its own fragmentation at the UDP/RAW layer. TCP shouldn't cause
106 * fragmentation anyway.
109 #define pr_fmt(fmt) "IPv4: " fmt
111 #include <linux/module.h>
112 #include <linux/types.h>
113 #include <linux/kernel.h>
114 #include <linux/string.h>
115 #include <linux/errno.h>
116 #include <linux/slab.h>
118 #include <linux/net.h>
119 #include <linux/socket.h>
120 #include <linux/sockios.h>
121 #include <linux/in.h>
122 #include <linux/inet.h>
123 #include <linux/inetdevice.h>
124 #include <linux/netdevice.h>
125 #include <linux/etherdevice.h>
126 #include <linux/indirect_call_wrapper.h>
128 #include <net/snmp.h>
130 #include <net/protocol.h>
131 #include <net/route.h>
132 #include <linux/skbuff.h>
133 #include <net/sock.h>
135 #include <net/icmp.h>
137 #include <net/checksum.h>
138 #include <net/inet_ecn.h>
139 #include <linux/netfilter_ipv4.h>
140 #include <net/xfrm.h>
141 #include <linux/mroute.h>
142 #include <linux/netlink.h>
143 #include <net/dst_metadata.h>
146 * Process Router Attention IP option (RFC 2113)
148 bool ip_call_ra_chain(struct sk_buff *skb)
150 struct ip_ra_chain *ra;
151 u8 protocol = ip_hdr(skb)->protocol;
152 struct sock *last = NULL;
153 struct net_device *dev = skb->dev;
154 struct net *net = dev_net(dev);
156 for (ra = rcu_dereference(net->ipv4.ra_chain); ra; ra = rcu_dereference(ra->next)) {
157 struct sock *sk = ra->sk;
159 /* If socket is bound to an interface, only report
160 * the packet if it came from that interface.
162 if (sk && inet_sk(sk)->inet_num == protocol &&
163 (!sk->sk_bound_dev_if ||
164 sk->sk_bound_dev_if == dev->ifindex)) {
165 if (ip_is_fragment(ip_hdr(skb))) {
166 if (ip_defrag(net, skb, IP_DEFRAG_CALL_RA_CHAIN))
170 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
185 INDIRECT_CALLABLE_DECLARE(int udp_rcv(struct sk_buff *));
186 INDIRECT_CALLABLE_DECLARE(int tcp_v4_rcv(struct sk_buff *));
187 void ip_protocol_deliver_rcu(struct net *net, struct sk_buff *skb, int protocol)
189 const struct net_protocol *ipprot;
193 raw = raw_local_deliver(skb, protocol);
195 ipprot = rcu_dereference(inet_protos[protocol]);
197 if (!ipprot->no_policy) {
198 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
204 ret = INDIRECT_CALL_2(ipprot->handler, tcp_v4_rcv, udp_rcv,
210 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
213 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
214 __IP_INC_STATS(net, IPSTATS_MIB_INUNKNOWNPROTOS);
215 icmp_send(skb, ICMP_DEST_UNREACH,
216 ICMP_PROT_UNREACH, 0);
220 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
226 static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
228 __skb_pull(skb, skb_network_header_len(skb));
231 ip_protocol_deliver_rcu(net, skb, ip_hdr(skb)->protocol);
238 * Deliver IP Packets to the higher protocol layers.
240 int ip_local_deliver(struct sk_buff *skb)
243 * Reassemble IP fragments.
245 struct net *net = dev_net(skb->dev);
247 if (ip_is_fragment(ip_hdr(skb))) {
248 if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
252 return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
253 net, NULL, skb, skb->dev, NULL,
254 ip_local_deliver_finish);
256 EXPORT_SYMBOL(ip_local_deliver);
258 static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev)
260 struct ip_options *opt;
261 const struct iphdr *iph;
263 /* It looks as overkill, because not all
264 IP options require packet mangling.
265 But it is the easiest for now, especially taking
266 into account that combination of IP options
267 and running sniffer is extremely rare condition.
270 if (skb_cow(skb, skb_headroom(skb))) {
271 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
276 opt = &(IPCB(skb)->opt);
277 opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
279 if (ip_options_compile(dev_net(dev), opt, skb)) {
280 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
284 if (unlikely(opt->srr)) {
285 struct in_device *in_dev = __in_dev_get_rcu(dev);
288 if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
289 if (IN_DEV_LOG_MARTIANS(in_dev))
290 net_info_ratelimited("source route option %pI4 -> %pI4\n",
297 if (ip_options_rcv_srr(skb, dev))
306 static bool ip_can_use_hint(const struct sk_buff *skb, const struct iphdr *iph,
307 const struct sk_buff *hint)
309 return hint && !skb_dst(skb) && ip_hdr(hint)->daddr == iph->daddr &&
310 ip_hdr(hint)->tos == iph->tos;
313 int tcp_v4_early_demux(struct sk_buff *skb);
314 int udp_v4_early_demux(struct sk_buff *skb);
315 static int ip_rcv_finish_core(struct net *net, struct sock *sk,
316 struct sk_buff *skb, struct net_device *dev,
317 const struct sk_buff *hint)
319 const struct iphdr *iph = ip_hdr(skb);
320 int err, drop_reason;
323 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
325 if (ip_can_use_hint(skb, iph, hint)) {
326 err = ip_route_use_hint(skb, iph->daddr, iph->saddr, iph->tos,
332 if (READ_ONCE(net->ipv4.sysctl_ip_early_demux) &&
335 !ip_is_fragment(iph)) {
336 switch (iph->protocol) {
338 if (READ_ONCE(net->ipv4.sysctl_tcp_early_demux)) {
339 tcp_v4_early_demux(skb);
341 /* must reload iph, skb->head might have changed */
346 if (READ_ONCE(net->ipv4.sysctl_udp_early_demux)) {
347 err = udp_v4_early_demux(skb);
351 /* must reload iph, skb->head might have changed */
359 * Initialise the virtual path cache for the packet. It describes
360 * how the packet travels inside Linux networking.
362 if (!skb_valid_dst(skb)) {
363 err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
368 struct in_device *in_dev = __in_dev_get_rcu(dev);
370 if (in_dev && IN_DEV_ORCONF(in_dev, NOPOLICY))
371 IPCB(skb)->flags |= IPSKB_NOPOLICY;
374 #ifdef CONFIG_IP_ROUTE_CLASSID
375 if (unlikely(skb_dst(skb)->tclassid)) {
376 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
377 u32 idx = skb_dst(skb)->tclassid;
378 st[idx&0xFF].o_packets++;
379 st[idx&0xFF].o_bytes += skb->len;
380 st[(idx>>16)&0xFF].i_packets++;
381 st[(idx>>16)&0xFF].i_bytes += skb->len;
385 if (iph->ihl > 5 && ip_rcv_options(skb, dev))
388 rt = skb_rtable(skb);
389 if (rt->rt_type == RTN_MULTICAST) {
390 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
391 } else if (rt->rt_type == RTN_BROADCAST) {
392 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
393 } else if (skb->pkt_type == PACKET_BROADCAST ||
394 skb->pkt_type == PACKET_MULTICAST) {
395 struct in_device *in_dev = __in_dev_get_rcu(dev);
399 * When a host sends a datagram to a link-layer broadcast
400 * address, the IP destination address MUST be a legal IP
401 * broadcast or IP multicast address.
403 * A host SHOULD silently discard a datagram that is received
404 * via a link-layer broadcast (see Section 2.4) but does not
405 * specify an IP multicast or broadcast destination address.
407 * This doesn't explicitly say L2 *broadcast*, but broadcast is
408 * in a way a form of multicast and the most common use case for
409 * this is 802.11 protecting against cross-station spoofing (the
410 * so-called "hole-196" attack) so do it for both.
413 IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST)) {
414 drop_reason = SKB_DROP_REASON_UNICAST_IN_L2_MULTICAST;
419 return NET_RX_SUCCESS;
422 kfree_skb_reason(skb, drop_reason);
427 drop_reason = SKB_DROP_REASON_IP_RPFILTER;
428 __NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
433 static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
435 struct net_device *dev = skb->dev;
438 /* if ingress device is enslaved to an L3 master device pass the
439 * skb to its handler for processing
441 skb = l3mdev_ip_rcv(skb);
443 return NET_RX_SUCCESS;
445 ret = ip_rcv_finish_core(net, sk, skb, dev, NULL);
446 if (ret != NET_RX_DROP)
447 ret = dst_input(skb);
452 * Main IP Receive routine.
454 static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net)
456 const struct iphdr *iph;
460 /* When the interface is in promisc. mode, drop all the crap
461 * that it receives, do not try to analyse it.
463 if (skb->pkt_type == PACKET_OTHERHOST) {
464 drop_reason = SKB_DROP_REASON_OTHERHOST;
468 __IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
470 skb = skb_share_check(skb, GFP_ATOMIC);
472 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
476 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
477 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
483 * RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
485 * Is the datagram acceptable?
487 * 1. Length at least the size of an ip header
489 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums]
490 * 4. Doesn't have a bogus length
493 if (iph->ihl < 5 || iph->version != 4)
496 BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
497 BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
498 BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
500 IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
501 max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
503 if (!pskb_may_pull(skb, iph->ihl*4))
508 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
511 len = ntohs(iph->tot_len);
512 if (skb->len < len) {
513 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
514 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
516 } else if (len < (iph->ihl*4))
519 /* Our transport medium may have padded the buffer out. Now we know it
520 * is IP we can trim to the true length of the frame.
521 * Note this now means skb->len holds ntohs(iph->tot_len).
523 if (pskb_trim_rcsum(skb, len)) {
524 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
529 skb->transport_header = skb->network_header + iph->ihl*4;
531 /* Remove any debris in the socket control block */
532 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
533 IPCB(skb)->iif = skb->skb_iif;
535 /* Must drop socket now because of tproxy. */
536 if (!skb_sk_is_prefetched(skb))
542 drop_reason = SKB_DROP_REASON_IP_CSUM;
543 __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
545 if (drop_reason == SKB_DROP_REASON_NOT_SPECIFIED)
546 drop_reason = SKB_DROP_REASON_IP_INHDR;
547 __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
549 kfree_skb_reason(skb, drop_reason);
555 * IP receive entry point
557 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
558 struct net_device *orig_dev)
560 struct net *net = dev_net(dev);
562 skb = ip_rcv_core(skb, net);
566 return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
567 net, NULL, skb, dev, NULL,
571 static void ip_sublist_rcv_finish(struct list_head *head)
573 struct sk_buff *skb, *next;
575 list_for_each_entry_safe(skb, next, head, list) {
576 skb_list_del_init(skb);
581 static struct sk_buff *ip_extract_route_hint(const struct net *net,
582 struct sk_buff *skb, int rt_type)
584 if (fib4_has_custom_rules(net) || rt_type == RTN_BROADCAST ||
585 IPCB(skb)->flags & IPSKB_MULTIPATH)
591 static void ip_list_rcv_finish(struct net *net, struct sock *sk,
592 struct list_head *head)
594 struct sk_buff *skb, *next, *hint = NULL;
595 struct dst_entry *curr_dst = NULL;
596 struct list_head sublist;
598 INIT_LIST_HEAD(&sublist);
599 list_for_each_entry_safe(skb, next, head, list) {
600 struct net_device *dev = skb->dev;
601 struct dst_entry *dst;
603 skb_list_del_init(skb);
604 /* if ingress device is enslaved to an L3 master device pass the
605 * skb to its handler for processing
607 skb = l3mdev_ip_rcv(skb);
610 if (ip_rcv_finish_core(net, sk, skb, dev, hint) == NET_RX_DROP)
614 if (curr_dst != dst) {
615 hint = ip_extract_route_hint(net, skb,
616 ((struct rtable *)dst)->rt_type);
618 /* dispatch old sublist */
619 if (!list_empty(&sublist))
620 ip_sublist_rcv_finish(&sublist);
621 /* start new sublist */
622 INIT_LIST_HEAD(&sublist);
625 list_add_tail(&skb->list, &sublist);
627 /* dispatch final sublist */
628 ip_sublist_rcv_finish(&sublist);
631 static void ip_sublist_rcv(struct list_head *head, struct net_device *dev,
634 NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL,
635 head, dev, NULL, ip_rcv_finish);
636 ip_list_rcv_finish(net, NULL, head);
639 /* Receive a list of IP packets */
640 void ip_list_rcv(struct list_head *head, struct packet_type *pt,
641 struct net_device *orig_dev)
643 struct net_device *curr_dev = NULL;
644 struct net *curr_net = NULL;
645 struct sk_buff *skb, *next;
646 struct list_head sublist;
648 INIT_LIST_HEAD(&sublist);
649 list_for_each_entry_safe(skb, next, head, list) {
650 struct net_device *dev = skb->dev;
651 struct net *net = dev_net(dev);
653 skb_list_del_init(skb);
654 skb = ip_rcv_core(skb, net);
658 if (curr_dev != dev || curr_net != net) {
659 /* dispatch old sublist */
660 if (!list_empty(&sublist))
661 ip_sublist_rcv(&sublist, curr_dev, curr_net);
662 /* start new sublist */
663 INIT_LIST_HEAD(&sublist);
667 list_add_tail(&skb->list, &sublist);
669 /* dispatch final sublist */
670 if (!list_empty(&sublist))
671 ip_sublist_rcv(&sublist, curr_dev, curr_net);