Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
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.
6 *
7 * The Internet Protocol (IP) module.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Donald Becker, <becker@super.org>
12 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
13 * Richard Underwood
14 * Stefan Becker, <stefanb@yello.ping.de>
15 * Jorge Cwik, <jorge@laser.satlink.net>
16 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
17 *
18 * Fixes:
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
46 * queue.
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
98 *
99 * To Fix:
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.
107 */
108
109#define pr_fmt(fmt) "IPv4: " fmt
110
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>
117
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>
127
128#include <net/snmp.h>
129#include <net/ip.h>
130#include <net/protocol.h>
131#include <net/route.h>
132#include <linux/skbuff.h>
133#include <net/sock.h>
134#include <net/arp.h>
135#include <net/icmp.h>
136#include <net/raw.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>
144
145/*
146 * Process Router Attention IP option (RFC 2113)
147 */
148bool ip_call_ra_chain(struct sk_buff *skb)
149{
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);
155
156 for (ra = rcu_dereference(net->ipv4.ra_chain); ra; ra = rcu_dereference(ra->next)) {
157 struct sock *sk = ra->sk;
158
159 /* If socket is bound to an interface, only report
160 * the packet if it came from that interface.
161 */
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))
167 return true;
168 }
169 if (last) {
170 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
171 if (skb2)
172 raw_rcv(last, skb2);
173 }
174 last = sk;
175 }
176 }
177
178 if (last) {
179 raw_rcv(last, skb);
180 return true;
181 }
182 return false;
183}
184
185INDIRECT_CALLABLE_DECLARE(int udp_rcv(struct sk_buff *));
186INDIRECT_CALLABLE_DECLARE(int tcp_v4_rcv(struct sk_buff *));
187void ip_protocol_deliver_rcu(struct net *net, struct sk_buff *skb, int protocol)
188{
189 const struct net_protocol *ipprot;
190 int raw, ret;
191
192resubmit:
193 raw = raw_local_deliver(skb, protocol);
194
195 ipprot = rcu_dereference(inet_protos[protocol]);
196 if (ipprot) {
197 if (!ipprot->no_policy) {
198 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
199 kfree_skb(skb);
200 return;
201 }
202 nf_reset_ct(skb);
203 }
204 ret = INDIRECT_CALL_2(ipprot->handler, tcp_v4_rcv, udp_rcv,
205 skb);
206 if (ret < 0) {
207 protocol = -ret;
208 goto resubmit;
209 }
210 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
211 } else {
212 if (!raw) {
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);
217 }
218 kfree_skb(skb);
219 } else {
220 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
221 consume_skb(skb);
222 }
223 }
224}
225
226static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
227{
228 __skb_pull(skb, skb_network_header_len(skb));
229
230 rcu_read_lock();
231 ip_protocol_deliver_rcu(net, skb, ip_hdr(skb)->protocol);
232 rcu_read_unlock();
233
234 return 0;
235}
236
237/*
238 * Deliver IP Packets to the higher protocol layers.
239 */
240int ip_local_deliver(struct sk_buff *skb)
241{
242 /*
243 * Reassemble IP fragments.
244 */
245 struct net *net = dev_net(skb->dev);
246
247 if (ip_is_fragment(ip_hdr(skb))) {
248 if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
249 return 0;
250 }
251
252 return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
253 net, NULL, skb, skb->dev, NULL,
254 ip_local_deliver_finish);
255}
256
257static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev)
258{
259 struct ip_options *opt;
260 const struct iphdr *iph;
261
262 /* It looks as overkill, because not all
263 IP options require packet mangling.
264 But it is the easiest for now, especially taking
265 into account that combination of IP options
266 and running sniffer is extremely rare condition.
267 --ANK (980813)
268 */
269 if (skb_cow(skb, skb_headroom(skb))) {
270 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
271 goto drop;
272 }
273
274 iph = ip_hdr(skb);
275 opt = &(IPCB(skb)->opt);
276 opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
277
278 if (ip_options_compile(dev_net(dev), opt, skb)) {
279 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
280 goto drop;
281 }
282
283 if (unlikely(opt->srr)) {
284 struct in_device *in_dev = __in_dev_get_rcu(dev);
285
286 if (in_dev) {
287 if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
288 if (IN_DEV_LOG_MARTIANS(in_dev))
289 net_info_ratelimited("source route option %pI4 -> %pI4\n",
290 &iph->saddr,
291 &iph->daddr);
292 goto drop;
293 }
294 }
295
296 if (ip_options_rcv_srr(skb, dev))
297 goto drop;
298 }
299
300 return false;
301drop:
302 return true;
303}
304
305static bool ip_can_use_hint(const struct sk_buff *skb, const struct iphdr *iph,
306 const struct sk_buff *hint)
307{
308 return hint && !skb_dst(skb) && ip_hdr(hint)->daddr == iph->daddr &&
309 ip_hdr(hint)->tos == iph->tos;
310}
311
312INDIRECT_CALLABLE_DECLARE(int udp_v4_early_demux(struct sk_buff *));
313INDIRECT_CALLABLE_DECLARE(int tcp_v4_early_demux(struct sk_buff *));
314static int ip_rcv_finish_core(struct net *net, struct sock *sk,
315 struct sk_buff *skb, struct net_device *dev,
316 const struct sk_buff *hint)
317{
318 const struct iphdr *iph = ip_hdr(skb);
319 int (*edemux)(struct sk_buff *skb);
320 struct rtable *rt;
321 int err;
322
323 if (ip_can_use_hint(skb, iph, hint)) {
324 err = ip_route_use_hint(skb, iph->daddr, iph->saddr, iph->tos,
325 dev, hint);
326 if (unlikely(err))
327 goto drop_error;
328 }
329
330 if (net->ipv4.sysctl_ip_early_demux &&
331 !skb_dst(skb) &&
332 !skb->sk &&
333 !ip_is_fragment(iph)) {
334 const struct net_protocol *ipprot;
335 int protocol = iph->protocol;
336
337 ipprot = rcu_dereference(inet_protos[protocol]);
338 if (ipprot && (edemux = READ_ONCE(ipprot->early_demux))) {
339 err = INDIRECT_CALL_2(edemux, tcp_v4_early_demux,
340 udp_v4_early_demux, skb);
341 if (unlikely(err))
342 goto drop_error;
343 /* must reload iph, skb->head might have changed */
344 iph = ip_hdr(skb);
345 }
346 }
347
348 /*
349 * Initialise the virtual path cache for the packet. It describes
350 * how the packet travels inside Linux networking.
351 */
352 if (!skb_valid_dst(skb)) {
353 err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
354 iph->tos, dev);
355 if (unlikely(err))
356 goto drop_error;
357 }
358
359#ifdef CONFIG_IP_ROUTE_CLASSID
360 if (unlikely(skb_dst(skb)->tclassid)) {
361 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
362 u32 idx = skb_dst(skb)->tclassid;
363 st[idx&0xFF].o_packets++;
364 st[idx&0xFF].o_bytes += skb->len;
365 st[(idx>>16)&0xFF].i_packets++;
366 st[(idx>>16)&0xFF].i_bytes += skb->len;
367 }
368#endif
369
370 if (iph->ihl > 5 && ip_rcv_options(skb, dev))
371 goto drop;
372
373 rt = skb_rtable(skb);
374 if (rt->rt_type == RTN_MULTICAST) {
375 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
376 } else if (rt->rt_type == RTN_BROADCAST) {
377 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
378 } else if (skb->pkt_type == PACKET_BROADCAST ||
379 skb->pkt_type == PACKET_MULTICAST) {
380 struct in_device *in_dev = __in_dev_get_rcu(dev);
381
382 /* RFC 1122 3.3.6:
383 *
384 * When a host sends a datagram to a link-layer broadcast
385 * address, the IP destination address MUST be a legal IP
386 * broadcast or IP multicast address.
387 *
388 * A host SHOULD silently discard a datagram that is received
389 * via a link-layer broadcast (see Section 2.4) but does not
390 * specify an IP multicast or broadcast destination address.
391 *
392 * This doesn't explicitly say L2 *broadcast*, but broadcast is
393 * in a way a form of multicast and the most common use case for
394 * this is 802.11 protecting against cross-station spoofing (the
395 * so-called "hole-196" attack) so do it for both.
396 */
397 if (in_dev &&
398 IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST))
399 goto drop;
400 }
401
402 return NET_RX_SUCCESS;
403
404drop:
405 kfree_skb(skb);
406 return NET_RX_DROP;
407
408drop_error:
409 if (err == -EXDEV)
410 __NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
411 goto drop;
412}
413
414static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
415{
416 struct net_device *dev = skb->dev;
417 int ret;
418
419 /* if ingress device is enslaved to an L3 master device pass the
420 * skb to its handler for processing
421 */
422 skb = l3mdev_ip_rcv(skb);
423 if (!skb)
424 return NET_RX_SUCCESS;
425
426 ret = ip_rcv_finish_core(net, sk, skb, dev, NULL);
427 if (ret != NET_RX_DROP)
428 ret = dst_input(skb);
429 return ret;
430}
431
432/*
433 * Main IP Receive routine.
434 */
435static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net)
436{
437 const struct iphdr *iph;
438 u32 len;
439
440 /* When the interface is in promisc. mode, drop all the crap
441 * that it receives, do not try to analyse it.
442 */
443 if (skb->pkt_type == PACKET_OTHERHOST)
444 goto drop;
445
446 __IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
447
448 skb = skb_share_check(skb, GFP_ATOMIC);
449 if (!skb) {
450 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
451 goto out;
452 }
453
454 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
455 goto inhdr_error;
456
457 iph = ip_hdr(skb);
458
459 /*
460 * RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
461 *
462 * Is the datagram acceptable?
463 *
464 * 1. Length at least the size of an ip header
465 * 2. Version of 4
466 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums]
467 * 4. Doesn't have a bogus length
468 */
469
470 if (iph->ihl < 5 || iph->version != 4)
471 goto inhdr_error;
472
473 BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
474 BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
475 BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
476 __IP_ADD_STATS(net,
477 IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
478 max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
479
480 if (!pskb_may_pull(skb, iph->ihl*4))
481 goto inhdr_error;
482
483 iph = ip_hdr(skb);
484
485 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
486 goto csum_error;
487
488 len = ntohs(iph->tot_len);
489 if (skb->len < len) {
490 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
491 goto drop;
492 } else if (len < (iph->ihl*4))
493 goto inhdr_error;
494
495 /* Our transport medium may have padded the buffer out. Now we know it
496 * is IP we can trim to the true length of the frame.
497 * Note this now means skb->len holds ntohs(iph->tot_len).
498 */
499 if (pskb_trim_rcsum(skb, len)) {
500 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
501 goto drop;
502 }
503
504 iph = ip_hdr(skb);
505 skb->transport_header = skb->network_header + iph->ihl*4;
506
507 /* Remove any debris in the socket control block */
508 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
509 IPCB(skb)->iif = skb->skb_iif;
510
511 /* Must drop socket now because of tproxy. */
512 if (!skb_sk_is_prefetched(skb))
513 skb_orphan(skb);
514
515 return skb;
516
517csum_error:
518 __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
519inhdr_error:
520 __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
521drop:
522 kfree_skb(skb);
523out:
524 return NULL;
525}
526
527/*
528 * IP receive entry point
529 */
530int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
531 struct net_device *orig_dev)
532{
533 struct net *net = dev_net(dev);
534
535 skb = ip_rcv_core(skb, net);
536 if (skb == NULL)
537 return NET_RX_DROP;
538
539 return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
540 net, NULL, skb, dev, NULL,
541 ip_rcv_finish);
542}
543
544static void ip_sublist_rcv_finish(struct list_head *head)
545{
546 struct sk_buff *skb, *next;
547
548 list_for_each_entry_safe(skb, next, head, list) {
549 skb_list_del_init(skb);
550 dst_input(skb);
551 }
552}
553
554static struct sk_buff *ip_extract_route_hint(const struct net *net,
555 struct sk_buff *skb, int rt_type)
556{
557 if (fib4_has_custom_rules(net) || rt_type == RTN_BROADCAST)
558 return NULL;
559
560 return skb;
561}
562
563static void ip_list_rcv_finish(struct net *net, struct sock *sk,
564 struct list_head *head)
565{
566 struct sk_buff *skb, *next, *hint = NULL;
567 struct dst_entry *curr_dst = NULL;
568 struct list_head sublist;
569
570 INIT_LIST_HEAD(&sublist);
571 list_for_each_entry_safe(skb, next, head, list) {
572 struct net_device *dev = skb->dev;
573 struct dst_entry *dst;
574
575 skb_list_del_init(skb);
576 /* if ingress device is enslaved to an L3 master device pass the
577 * skb to its handler for processing
578 */
579 skb = l3mdev_ip_rcv(skb);
580 if (!skb)
581 continue;
582 if (ip_rcv_finish_core(net, sk, skb, dev, hint) == NET_RX_DROP)
583 continue;
584
585 dst = skb_dst(skb);
586 if (curr_dst != dst) {
587 hint = ip_extract_route_hint(net, skb,
588 ((struct rtable *)dst)->rt_type);
589
590 /* dispatch old sublist */
591 if (!list_empty(&sublist))
592 ip_sublist_rcv_finish(&sublist);
593 /* start new sublist */
594 INIT_LIST_HEAD(&sublist);
595 curr_dst = dst;
596 }
597 list_add_tail(&skb->list, &sublist);
598 }
599 /* dispatch final sublist */
600 ip_sublist_rcv_finish(&sublist);
601}
602
603static void ip_sublist_rcv(struct list_head *head, struct net_device *dev,
604 struct net *net)
605{
606 NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL,
607 head, dev, NULL, ip_rcv_finish);
608 ip_list_rcv_finish(net, NULL, head);
609}
610
611/* Receive a list of IP packets */
612void ip_list_rcv(struct list_head *head, struct packet_type *pt,
613 struct net_device *orig_dev)
614{
615 struct net_device *curr_dev = NULL;
616 struct net *curr_net = NULL;
617 struct sk_buff *skb, *next;
618 struct list_head sublist;
619
620 INIT_LIST_HEAD(&sublist);
621 list_for_each_entry_safe(skb, next, head, list) {
622 struct net_device *dev = skb->dev;
623 struct net *net = dev_net(dev);
624
625 skb_list_del_init(skb);
626 skb = ip_rcv_core(skb, net);
627 if (skb == NULL)
628 continue;
629
630 if (curr_dev != dev || curr_net != net) {
631 /* dispatch old sublist */
632 if (!list_empty(&sublist))
633 ip_sublist_rcv(&sublist, curr_dev, curr_net);
634 /* start new sublist */
635 INIT_LIST_HEAD(&sublist);
636 curr_dev = dev;
637 curr_net = net;
638 }
639 list_add_tail(&skb->list, &sublist);
640 }
641 /* dispatch final sublist */
642 if (!list_empty(&sublist))
643 ip_sublist_rcv(&sublist, curr_dev, curr_net);
644}
1/*
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.
5 *
6 * The Internet Protocol (IP) module.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Donald Becker, <becker@super.org>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Richard Underwood
13 * Stefan Becker, <stefanb@yello.ping.de>
14 * Jorge Cwik, <jorge@laser.satlink.net>
15 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
16 *
17 *
18 * Fixes:
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
46 * queue.
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
98 *
99 *
100 *
101 * To Fix:
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.
109 *
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.
114 */
115
116#define pr_fmt(fmt) "IPv4: " fmt
117
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>
124
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
134#include <net/snmp.h>
135#include <net/ip.h>
136#include <net/protocol.h>
137#include <net/route.h>
138#include <linux/skbuff.h>
139#include <net/sock.h>
140#include <net/arp.h>
141#include <net/icmp.h>
142#include <net/raw.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>
150
151/*
152 * Process Router Attention IP option (RFC 2113)
153 */
154bool ip_call_ra_chain(struct sk_buff *skb)
155{
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);
161
162 for (ra = rcu_dereference(ip_ra_chain); ra; ra = rcu_dereference(ra->next)) {
163 struct sock *sk = ra->sk;
164
165 /* If socket is bound to an interface, only report
166 * the packet if it came from that interface.
167 */
168 if (sk && inet_sk(sk)->inet_num == protocol &&
169 (!sk->sk_bound_dev_if ||
170 sk->sk_bound_dev_if == dev->ifindex) &&
171 net_eq(sock_net(sk), net)) {
172 if (ip_is_fragment(ip_hdr(skb))) {
173 if (ip_defrag(net, skb, IP_DEFRAG_CALL_RA_CHAIN))
174 return true;
175 }
176 if (last) {
177 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
178 if (skb2)
179 raw_rcv(last, skb2);
180 }
181 last = sk;
182 }
183 }
184
185 if (last) {
186 raw_rcv(last, skb);
187 return true;
188 }
189 return false;
190}
191
192static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
193{
194 __skb_pull(skb, skb_network_header_len(skb));
195
196 rcu_read_lock();
197 {
198 int protocol = ip_hdr(skb)->protocol;
199 const struct net_protocol *ipprot;
200 int raw;
201
202 resubmit:
203 raw = raw_local_deliver(skb, protocol);
204
205 ipprot = rcu_dereference(inet_protos[protocol]);
206 if (ipprot) {
207 int ret;
208
209 if (!ipprot->no_policy) {
210 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
211 kfree_skb(skb);
212 goto out;
213 }
214 nf_reset(skb);
215 }
216 ret = ipprot->handler(skb);
217 if (ret < 0) {
218 protocol = -ret;
219 goto resubmit;
220 }
221 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
222 } else {
223 if (!raw) {
224 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
225 __IP_INC_STATS(net, IPSTATS_MIB_INUNKNOWNPROTOS);
226 icmp_send(skb, ICMP_DEST_UNREACH,
227 ICMP_PROT_UNREACH, 0);
228 }
229 kfree_skb(skb);
230 } else {
231 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
232 consume_skb(skb);
233 }
234 }
235 }
236 out:
237 rcu_read_unlock();
238
239 return 0;
240}
241
242/*
243 * Deliver IP Packets to the higher protocol layers.
244 */
245int ip_local_deliver(struct sk_buff *skb)
246{
247 /*
248 * Reassemble IP fragments.
249 */
250 struct net *net = dev_net(skb->dev);
251
252 if (ip_is_fragment(ip_hdr(skb))) {
253 if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
254 return 0;
255 }
256
257 return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
258 net, NULL, skb, skb->dev, NULL,
259 ip_local_deliver_finish);
260}
261
262static inline bool ip_rcv_options(struct sk_buff *skb)
263{
264 struct ip_options *opt;
265 const struct iphdr *iph;
266 struct net_device *dev = skb->dev;
267
268 /* It looks as overkill, because not all
269 IP options require packet mangling.
270 But it is the easiest for now, especially taking
271 into account that combination of IP options
272 and running sniffer is extremely rare condition.
273 --ANK (980813)
274 */
275 if (skb_cow(skb, skb_headroom(skb))) {
276 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
277 goto drop;
278 }
279
280 iph = ip_hdr(skb);
281 opt = &(IPCB(skb)->opt);
282 opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
283
284 if (ip_options_compile(dev_net(dev), opt, skb)) {
285 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
286 goto drop;
287 }
288
289 if (unlikely(opt->srr)) {
290 struct in_device *in_dev = __in_dev_get_rcu(dev);
291
292 if (in_dev) {
293 if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
294 if (IN_DEV_LOG_MARTIANS(in_dev))
295 net_info_ratelimited("source route option %pI4 -> %pI4\n",
296 &iph->saddr,
297 &iph->daddr);
298 goto drop;
299 }
300 }
301
302 if (ip_options_rcv_srr(skb))
303 goto drop;
304 }
305
306 return false;
307drop:
308 return true;
309}
310
311static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
312{
313 const struct iphdr *iph = ip_hdr(skb);
314 struct rtable *rt;
315 struct net_device *dev = skb->dev;
316
317 /* if ingress device is enslaved to an L3 master device pass the
318 * skb to its handler for processing
319 */
320 skb = l3mdev_ip_rcv(skb);
321 if (!skb)
322 return NET_RX_SUCCESS;
323
324 if (net->ipv4.sysctl_ip_early_demux &&
325 !skb_dst(skb) &&
326 !skb->sk &&
327 !ip_is_fragment(iph)) {
328 const struct net_protocol *ipprot;
329 int protocol = iph->protocol;
330
331 ipprot = rcu_dereference(inet_protos[protocol]);
332 if (ipprot && ipprot->early_demux) {
333 ipprot->early_demux(skb);
334 /* must reload iph, skb->head might have changed */
335 iph = ip_hdr(skb);
336 }
337 }
338
339 /*
340 * Initialise the virtual path cache for the packet. It describes
341 * how the packet travels inside Linux networking.
342 */
343 if (!skb_valid_dst(skb)) {
344 int err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
345 iph->tos, dev);
346 if (unlikely(err)) {
347 if (err == -EXDEV)
348 __NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
349 goto drop;
350 }
351 }
352
353#ifdef CONFIG_IP_ROUTE_CLASSID
354 if (unlikely(skb_dst(skb)->tclassid)) {
355 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
356 u32 idx = skb_dst(skb)->tclassid;
357 st[idx&0xFF].o_packets++;
358 st[idx&0xFF].o_bytes += skb->len;
359 st[(idx>>16)&0xFF].i_packets++;
360 st[(idx>>16)&0xFF].i_bytes += skb->len;
361 }
362#endif
363
364 if (iph->ihl > 5 && ip_rcv_options(skb))
365 goto drop;
366
367 rt = skb_rtable(skb);
368 if (rt->rt_type == RTN_MULTICAST) {
369 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
370 } else if (rt->rt_type == RTN_BROADCAST) {
371 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
372 } else if (skb->pkt_type == PACKET_BROADCAST ||
373 skb->pkt_type == PACKET_MULTICAST) {
374 struct in_device *in_dev = __in_dev_get_rcu(dev);
375
376 /* RFC 1122 3.3.6:
377 *
378 * When a host sends a datagram to a link-layer broadcast
379 * address, the IP destination address MUST be a legal IP
380 * broadcast or IP multicast address.
381 *
382 * A host SHOULD silently discard a datagram that is received
383 * via a link-layer broadcast (see Section 2.4) but does not
384 * specify an IP multicast or broadcast destination address.
385 *
386 * This doesn't explicitly say L2 *broadcast*, but broadcast is
387 * in a way a form of multicast and the most common use case for
388 * this is 802.11 protecting against cross-station spoofing (the
389 * so-called "hole-196" attack) so do it for both.
390 */
391 if (in_dev &&
392 IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST))
393 goto drop;
394 }
395
396 return dst_input(skb);
397
398drop:
399 kfree_skb(skb);
400 return NET_RX_DROP;
401}
402
403/*
404 * Main IP Receive routine.
405 */
406int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
407{
408 const struct iphdr *iph;
409 struct net *net;
410 u32 len;
411
412 /* When the interface is in promisc. mode, drop all the crap
413 * that it receives, do not try to analyse it.
414 */
415 if (skb->pkt_type == PACKET_OTHERHOST)
416 goto drop;
417
418
419 net = dev_net(dev);
420 __IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
421
422 skb = skb_share_check(skb, GFP_ATOMIC);
423 if (!skb) {
424 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
425 goto out;
426 }
427
428 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
429 goto inhdr_error;
430
431 iph = ip_hdr(skb);
432
433 /*
434 * RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
435 *
436 * Is the datagram acceptable?
437 *
438 * 1. Length at least the size of an ip header
439 * 2. Version of 4
440 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums]
441 * 4. Doesn't have a bogus length
442 */
443
444 if (iph->ihl < 5 || iph->version != 4)
445 goto inhdr_error;
446
447 BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
448 BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
449 BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
450 __IP_ADD_STATS(net,
451 IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
452 max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
453
454 if (!pskb_may_pull(skb, iph->ihl*4))
455 goto inhdr_error;
456
457 iph = ip_hdr(skb);
458
459 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
460 goto csum_error;
461
462 len = ntohs(iph->tot_len);
463 if (skb->len < len) {
464 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
465 goto drop;
466 } else if (len < (iph->ihl*4))
467 goto inhdr_error;
468
469 /* Our transport medium may have padded the buffer out. Now we know it
470 * is IP we can trim to the true length of the frame.
471 * Note this now means skb->len holds ntohs(iph->tot_len).
472 */
473 if (pskb_trim_rcsum(skb, len)) {
474 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
475 goto drop;
476 }
477
478 skb->transport_header = skb->network_header + iph->ihl*4;
479
480 /* Remove any debris in the socket control block */
481 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
482 IPCB(skb)->iif = skb->skb_iif;
483
484 /* Must drop socket now because of tproxy. */
485 skb_orphan(skb);
486
487 return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
488 net, NULL, skb, dev, NULL,
489 ip_rcv_finish);
490
491csum_error:
492 __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
493inhdr_error:
494 __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
495drop:
496 kfree_skb(skb);
497out:
498 return NET_RX_DROP;
499}