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)) {
199 kfree_skb_reason(skb,
200 SKB_DROP_REASON_XFRM_POLICY);
205 ret = INDIRECT_CALL_2(ipprot->handler, tcp_v4_rcv, udp_rcv,
211 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
214 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
215 __IP_INC_STATS(net, IPSTATS_MIB_INUNKNOWNPROTOS);
216 icmp_send(skb, ICMP_DEST_UNREACH,
217 ICMP_PROT_UNREACH, 0);
219 kfree_skb_reason(skb, SKB_DROP_REASON_IP_NOPROTO);
221 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
227 static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
229 skb_clear_delivery_time(skb);
230 __skb_pull(skb, skb_network_header_len(skb));
233 ip_protocol_deliver_rcu(net, skb, ip_hdr(skb)->protocol);
240 * Deliver IP Packets to the higher protocol layers.
242 int ip_local_deliver(struct sk_buff *skb)
245 * Reassemble IP fragments.
247 struct net *net = dev_net(skb->dev);
249 if (ip_is_fragment(ip_hdr(skb))) {
250 if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
254 return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
255 net, NULL, skb, skb->dev, NULL,
256 ip_local_deliver_finish);
258 EXPORT_SYMBOL(ip_local_deliver);
260 static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev)
262 struct ip_options *opt;
263 const struct iphdr *iph;
265 /* It looks as overkill, because not all
266 IP options require packet mangling.
267 But it is the easiest for now, especially taking
268 into account that combination of IP options
269 and running sniffer is extremely rare condition.
272 if (skb_cow(skb, skb_headroom(skb))) {
273 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
278 opt = &(IPCB(skb)->opt);
279 opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
281 if (ip_options_compile(dev_net(dev), opt, skb)) {
282 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
286 if (unlikely(opt->srr)) {
287 struct in_device *in_dev = __in_dev_get_rcu(dev);
290 if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
291 if (IN_DEV_LOG_MARTIANS(in_dev))
292 net_info_ratelimited("source route option %pI4 -> %pI4\n",
299 if (ip_options_rcv_srr(skb, dev))
308 static bool ip_can_use_hint(const struct sk_buff *skb, const struct iphdr *iph,
309 const struct sk_buff *hint)
311 return hint && !skb_dst(skb) && ip_hdr(hint)->daddr == iph->daddr &&
312 ip_hdr(hint)->tos == iph->tos;
315 int tcp_v4_early_demux(struct sk_buff *skb);
316 int udp_v4_early_demux(struct sk_buff *skb);
317 static int ip_rcv_finish_core(struct net *net, struct sock *sk,
318 struct sk_buff *skb, struct net_device *dev,
319 const struct sk_buff *hint)
321 const struct iphdr *iph = ip_hdr(skb);
322 int err, drop_reason;
325 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
327 if (ip_can_use_hint(skb, iph, hint)) {
328 err = ip_route_use_hint(skb, iph->daddr, iph->saddr, iph->tos,
334 if (READ_ONCE(net->ipv4.sysctl_ip_early_demux) &&
337 !ip_is_fragment(iph)) {
338 switch (iph->protocol) {
340 if (READ_ONCE(net->ipv4.sysctl_tcp_early_demux)) {
341 tcp_v4_early_demux(skb);
343 /* must reload iph, skb->head might have changed */
348 if (READ_ONCE(net->ipv4.sysctl_udp_early_demux)) {
349 err = udp_v4_early_demux(skb);
353 /* must reload iph, skb->head might have changed */
361 * Initialise the virtual path cache for the packet. It describes
362 * how the packet travels inside Linux networking.
364 if (!skb_valid_dst(skb)) {
365 err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
371 #ifdef CONFIG_IP_ROUTE_CLASSID
372 if (unlikely(skb_dst(skb)->tclassid)) {
373 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
374 u32 idx = skb_dst(skb)->tclassid;
375 st[idx&0xFF].o_packets++;
376 st[idx&0xFF].o_bytes += skb->len;
377 st[(idx>>16)&0xFF].i_packets++;
378 st[(idx>>16)&0xFF].i_bytes += skb->len;
382 if (iph->ihl > 5 && ip_rcv_options(skb, dev))
385 rt = skb_rtable(skb);
386 if (rt->rt_type == RTN_MULTICAST) {
387 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
388 } else if (rt->rt_type == RTN_BROADCAST) {
389 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
390 } else if (skb->pkt_type == PACKET_BROADCAST ||
391 skb->pkt_type == PACKET_MULTICAST) {
392 struct in_device *in_dev = __in_dev_get_rcu(dev);
396 * When a host sends a datagram to a link-layer broadcast
397 * address, the IP destination address MUST be a legal IP
398 * broadcast or IP multicast address.
400 * A host SHOULD silently discard a datagram that is received
401 * via a link-layer broadcast (see Section 2.4) but does not
402 * specify an IP multicast or broadcast destination address.
404 * This doesn't explicitly say L2 *broadcast*, but broadcast is
405 * in a way a form of multicast and the most common use case for
406 * this is 802.11 protecting against cross-station spoofing (the
407 * so-called "hole-196" attack) so do it for both.
410 IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST)) {
411 drop_reason = SKB_DROP_REASON_UNICAST_IN_L2_MULTICAST;
416 return NET_RX_SUCCESS;
419 kfree_skb_reason(skb, drop_reason);
424 drop_reason = SKB_DROP_REASON_IP_RPFILTER;
425 __NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
430 static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
432 struct net_device *dev = skb->dev;
435 /* if ingress device is enslaved to an L3 master device pass the
436 * skb to its handler for processing
438 skb = l3mdev_ip_rcv(skb);
440 return NET_RX_SUCCESS;
442 ret = ip_rcv_finish_core(net, sk, skb, dev, NULL);
443 if (ret != NET_RX_DROP)
444 ret = dst_input(skb);
449 * Main IP Receive routine.
451 static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net)
453 const struct iphdr *iph;
457 /* When the interface is in promisc. mode, drop all the crap
458 * that it receives, do not try to analyse it.
460 if (skb->pkt_type == PACKET_OTHERHOST) {
461 dev_core_stats_rx_otherhost_dropped_inc(skb->dev);
462 drop_reason = SKB_DROP_REASON_OTHERHOST;
466 __IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
468 skb = skb_share_check(skb, GFP_ATOMIC);
470 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
474 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
475 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
481 * RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
483 * Is the datagram acceptable?
485 * 1. Length at least the size of an ip header
487 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums]
488 * 4. Doesn't have a bogus length
491 if (iph->ihl < 5 || iph->version != 4)
494 BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
495 BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
496 BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
498 IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
499 max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
501 if (!pskb_may_pull(skb, iph->ihl*4))
506 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
509 len = ntohs(iph->tot_len);
510 if (skb->len < len) {
511 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
512 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
514 } else if (len < (iph->ihl*4))
517 /* Our transport medium may have padded the buffer out. Now we know it
518 * is IP we can trim to the true length of the frame.
519 * Note this now means skb->len holds ntohs(iph->tot_len).
521 if (pskb_trim_rcsum(skb, len)) {
522 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
527 skb->transport_header = skb->network_header + iph->ihl*4;
529 /* Remove any debris in the socket control block */
530 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
531 IPCB(skb)->iif = skb->skb_iif;
533 /* Must drop socket now because of tproxy. */
534 if (!skb_sk_is_prefetched(skb))
540 drop_reason = SKB_DROP_REASON_IP_CSUM;
541 __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
543 if (drop_reason == SKB_DROP_REASON_NOT_SPECIFIED)
544 drop_reason = SKB_DROP_REASON_IP_INHDR;
545 __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
547 kfree_skb_reason(skb, drop_reason);
553 * IP receive entry point
555 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
556 struct net_device *orig_dev)
558 struct net *net = dev_net(dev);
560 skb = ip_rcv_core(skb, net);
564 return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
565 net, NULL, skb, dev, NULL,
569 static void ip_sublist_rcv_finish(struct list_head *head)
571 struct sk_buff *skb, *next;
573 list_for_each_entry_safe(skb, next, head, list) {
574 skb_list_del_init(skb);
579 static struct sk_buff *ip_extract_route_hint(const struct net *net,
580 struct sk_buff *skb, int rt_type)
582 if (fib4_has_custom_rules(net) || rt_type == RTN_BROADCAST)
588 static void ip_list_rcv_finish(struct net *net, struct sock *sk,
589 struct list_head *head)
591 struct sk_buff *skb, *next, *hint = NULL;
592 struct dst_entry *curr_dst = NULL;
593 struct list_head sublist;
595 INIT_LIST_HEAD(&sublist);
596 list_for_each_entry_safe(skb, next, head, list) {
597 struct net_device *dev = skb->dev;
598 struct dst_entry *dst;
600 skb_list_del_init(skb);
601 /* if ingress device is enslaved to an L3 master device pass the
602 * skb to its handler for processing
604 skb = l3mdev_ip_rcv(skb);
607 if (ip_rcv_finish_core(net, sk, skb, dev, hint) == NET_RX_DROP)
611 if (curr_dst != dst) {
612 hint = ip_extract_route_hint(net, skb,
613 ((struct rtable *)dst)->rt_type);
615 /* dispatch old sublist */
616 if (!list_empty(&sublist))
617 ip_sublist_rcv_finish(&sublist);
618 /* start new sublist */
619 INIT_LIST_HEAD(&sublist);
622 list_add_tail(&skb->list, &sublist);
624 /* dispatch final sublist */
625 ip_sublist_rcv_finish(&sublist);
628 static void ip_sublist_rcv(struct list_head *head, struct net_device *dev,
631 NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL,
632 head, dev, NULL, ip_rcv_finish);
633 ip_list_rcv_finish(net, NULL, head);
636 /* Receive a list of IP packets */
637 void ip_list_rcv(struct list_head *head, struct packet_type *pt,
638 struct net_device *orig_dev)
640 struct net_device *curr_dev = NULL;
641 struct net *curr_net = NULL;
642 struct sk_buff *skb, *next;
643 struct list_head sublist;
645 INIT_LIST_HEAD(&sublist);
646 list_for_each_entry_safe(skb, next, head, list) {
647 struct net_device *dev = skb->dev;
648 struct net *net = dev_net(dev);
650 skb_list_del_init(skb);
651 skb = ip_rcv_core(skb, net);
655 if (curr_dev != dev || curr_net != net) {
656 /* dispatch old sublist */
657 if (!list_empty(&sublist))
658 ip_sublist_rcv(&sublist, curr_dev, curr_net);
659 /* start new sublist */
660 INIT_LIST_HEAD(&sublist);
664 list_add_tail(&skb->list, &sublist);
666 /* dispatch final sublist */
667 if (!list_empty(&sublist))
668 ip_sublist_rcv(&sublist, curr_dev, curr_net);