Loading...
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
150/*
151 * Process Router Attention IP option (RFC 2113)
152 */
153bool ip_call_ra_chain(struct sk_buff *skb)
154{
155 struct ip_ra_chain *ra;
156 u8 protocol = ip_hdr(skb)->protocol;
157 struct sock *last = NULL;
158 struct net_device *dev = skb->dev;
159
160 for (ra = rcu_dereference(ip_ra_chain); ra; ra = rcu_dereference(ra->next)) {
161 struct sock *sk = ra->sk;
162
163 /* If socket is bound to an interface, only report
164 * the packet if it came from that interface.
165 */
166 if (sk && inet_sk(sk)->inet_num == protocol &&
167 (!sk->sk_bound_dev_if ||
168 sk->sk_bound_dev_if == dev->ifindex) &&
169 net_eq(sock_net(sk), dev_net(dev))) {
170 if (ip_is_fragment(ip_hdr(skb))) {
171 if (ip_defrag(skb, IP_DEFRAG_CALL_RA_CHAIN))
172 return true;
173 }
174 if (last) {
175 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
176 if (skb2)
177 raw_rcv(last, skb2);
178 }
179 last = sk;
180 }
181 }
182
183 if (last) {
184 raw_rcv(last, skb);
185 return true;
186 }
187 return false;
188}
189
190static int ip_local_deliver_finish(struct sk_buff *skb)
191{
192 struct net *net = dev_net(skb->dev);
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 != NULL) {
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_BH(net, IPSTATS_MIB_INDELIVERS);
222 } else {
223 if (!raw) {
224 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
225 IP_INC_STATS_BH(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_BH(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
251 if (ip_is_fragment(ip_hdr(skb))) {
252 if (ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER))
253 return 0;
254 }
255
256 return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN, skb, skb->dev, NULL,
257 ip_local_deliver_finish);
258}
259
260static inline bool ip_rcv_options(struct sk_buff *skb)
261{
262 struct ip_options *opt;
263 const struct iphdr *iph;
264 struct net_device *dev = skb->dev;
265
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.
271 --ANK (980813)
272 */
273 if (skb_cow(skb, skb_headroom(skb))) {
274 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
275 goto drop;
276 }
277
278 iph = ip_hdr(skb);
279 opt = &(IPCB(skb)->opt);
280 opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
281
282 if (ip_options_compile(dev_net(dev), opt, skb)) {
283 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
284 goto drop;
285 }
286
287 if (unlikely(opt->srr)) {
288 struct in_device *in_dev = __in_dev_get_rcu(dev);
289
290 if (in_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",
294 &iph->saddr,
295 &iph->daddr);
296 goto drop;
297 }
298 }
299
300 if (ip_options_rcv_srr(skb))
301 goto drop;
302 }
303
304 return false;
305drop:
306 return true;
307}
308
309int sysctl_ip_early_demux __read_mostly = 1;
310EXPORT_SYMBOL(sysctl_ip_early_demux);
311
312static int ip_rcv_finish(struct sk_buff *skb)
313{
314 const struct iphdr *iph = ip_hdr(skb);
315 struct rtable *rt;
316
317 if (sysctl_ip_early_demux && !skb_dst(skb) && skb->sk == NULL) {
318 const struct net_protocol *ipprot;
319 int protocol = iph->protocol;
320
321 ipprot = rcu_dereference(inet_protos[protocol]);
322 if (ipprot && ipprot->early_demux) {
323 ipprot->early_demux(skb);
324 /* must reload iph, skb->head might have changed */
325 iph = ip_hdr(skb);
326 }
327 }
328
329 /*
330 * Initialise the virtual path cache for the packet. It describes
331 * how the packet travels inside Linux networking.
332 */
333 if (!skb_dst(skb)) {
334 int err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
335 iph->tos, skb->dev);
336 if (unlikely(err)) {
337 if (err == -EXDEV)
338 NET_INC_STATS_BH(dev_net(skb->dev),
339 LINUX_MIB_IPRPFILTER);
340 goto drop;
341 }
342 }
343
344#ifdef CONFIG_IP_ROUTE_CLASSID
345 if (unlikely(skb_dst(skb)->tclassid)) {
346 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
347 u32 idx = skb_dst(skb)->tclassid;
348 st[idx&0xFF].o_packets++;
349 st[idx&0xFF].o_bytes += skb->len;
350 st[(idx>>16)&0xFF].i_packets++;
351 st[(idx>>16)&0xFF].i_bytes += skb->len;
352 }
353#endif
354
355 if (iph->ihl > 5 && ip_rcv_options(skb))
356 goto drop;
357
358 rt = skb_rtable(skb);
359 if (rt->rt_type == RTN_MULTICAST) {
360 IP_UPD_PO_STATS_BH(dev_net(rt->dst.dev), IPSTATS_MIB_INMCAST,
361 skb->len);
362 } else if (rt->rt_type == RTN_BROADCAST)
363 IP_UPD_PO_STATS_BH(dev_net(rt->dst.dev), IPSTATS_MIB_INBCAST,
364 skb->len);
365
366 return dst_input(skb);
367
368drop:
369 kfree_skb(skb);
370 return NET_RX_DROP;
371}
372
373/*
374 * Main IP Receive routine.
375 */
376int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
377{
378 const struct iphdr *iph;
379 u32 len;
380
381 /* When the interface is in promisc. mode, drop all the crap
382 * that it receives, do not try to analyse it.
383 */
384 if (skb->pkt_type == PACKET_OTHERHOST)
385 goto drop;
386
387
388 IP_UPD_PO_STATS_BH(dev_net(dev), IPSTATS_MIB_IN, skb->len);
389
390 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) {
391 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
392 goto out;
393 }
394
395 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
396 goto inhdr_error;
397
398 iph = ip_hdr(skb);
399
400 /*
401 * RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
402 *
403 * Is the datagram acceptable?
404 *
405 * 1. Length at least the size of an ip header
406 * 2. Version of 4
407 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums]
408 * 4. Doesn't have a bogus length
409 */
410
411 if (iph->ihl < 5 || iph->version != 4)
412 goto inhdr_error;
413
414 BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
415 BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
416 BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
417 IP_ADD_STATS_BH(dev_net(dev),
418 IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
419 max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
420
421 if (!pskb_may_pull(skb, iph->ihl*4))
422 goto inhdr_error;
423
424 iph = ip_hdr(skb);
425
426 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
427 goto csum_error;
428
429 len = ntohs(iph->tot_len);
430 if (skb->len < len) {
431 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
432 goto drop;
433 } else if (len < (iph->ihl*4))
434 goto inhdr_error;
435
436 /* Our transport medium may have padded the buffer out. Now we know it
437 * is IP we can trim to the true length of the frame.
438 * Note this now means skb->len holds ntohs(iph->tot_len).
439 */
440 if (pskb_trim_rcsum(skb, len)) {
441 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
442 goto drop;
443 }
444
445 skb->transport_header = skb->network_header + iph->ihl*4;
446
447 /* Remove any debris in the socket control block */
448 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
449
450 /* Must drop socket now because of tproxy. */
451 skb_orphan(skb);
452
453 return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, dev, NULL,
454 ip_rcv_finish);
455
456csum_error:
457 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_CSUMERRORS);
458inhdr_error:
459 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
460drop:
461 kfree_skb(skb);
462out:
463 return NET_RX_DROP;
464}
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_reason(skb,
200 SKB_DROP_REASON_XFRM_POLICY);
201 return;
202 }
203 nf_reset_ct(skb);
204 }
205 ret = INDIRECT_CALL_2(ipprot->handler, tcp_v4_rcv, udp_rcv,
206 skb);
207 if (ret < 0) {
208 protocol = -ret;
209 goto resubmit;
210 }
211 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
212 } else {
213 if (!raw) {
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);
218 }
219 kfree_skb_reason(skb, SKB_DROP_REASON_IP_NOPROTO);
220 } else {
221 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
222 consume_skb(skb);
223 }
224 }
225}
226
227static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
228{
229 skb_clear_delivery_time(skb);
230 __skb_pull(skb, skb_network_header_len(skb));
231
232 rcu_read_lock();
233 ip_protocol_deliver_rcu(net, skb, ip_hdr(skb)->protocol);
234 rcu_read_unlock();
235
236 return 0;
237}
238
239/*
240 * Deliver IP Packets to the higher protocol layers.
241 */
242int ip_local_deliver(struct sk_buff *skb)
243{
244 /*
245 * Reassemble IP fragments.
246 */
247 struct net *net = dev_net(skb->dev);
248
249 if (ip_is_fragment(ip_hdr(skb))) {
250 if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
251 return 0;
252 }
253
254 return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
255 net, NULL, skb, skb->dev, NULL,
256 ip_local_deliver_finish);
257}
258EXPORT_SYMBOL(ip_local_deliver);
259
260static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev)
261{
262 struct ip_options *opt;
263 const struct iphdr *iph;
264
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.
270 --ANK (980813)
271 */
272 if (skb_cow(skb, skb_headroom(skb))) {
273 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
274 goto drop;
275 }
276
277 iph = ip_hdr(skb);
278 opt = &(IPCB(skb)->opt);
279 opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
280
281 if (ip_options_compile(dev_net(dev), opt, skb)) {
282 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
283 goto drop;
284 }
285
286 if (unlikely(opt->srr)) {
287 struct in_device *in_dev = __in_dev_get_rcu(dev);
288
289 if (in_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",
293 &iph->saddr,
294 &iph->daddr);
295 goto drop;
296 }
297 }
298
299 if (ip_options_rcv_srr(skb, dev))
300 goto drop;
301 }
302
303 return false;
304drop:
305 return true;
306}
307
308static bool ip_can_use_hint(const struct sk_buff *skb, const struct iphdr *iph,
309 const struct sk_buff *hint)
310{
311 return hint && !skb_dst(skb) && ip_hdr(hint)->daddr == iph->daddr &&
312 ip_hdr(hint)->tos == iph->tos;
313}
314
315int tcp_v4_early_demux(struct sk_buff *skb);
316int udp_v4_early_demux(struct sk_buff *skb);
317static 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)
320{
321 const struct iphdr *iph = ip_hdr(skb);
322 int err, drop_reason;
323 struct rtable *rt;
324
325 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
326
327 if (ip_can_use_hint(skb, iph, hint)) {
328 err = ip_route_use_hint(skb, iph->daddr, iph->saddr, iph->tos,
329 dev, hint);
330 if (unlikely(err))
331 goto drop_error;
332 }
333
334 if (READ_ONCE(net->ipv4.sysctl_ip_early_demux) &&
335 !skb_dst(skb) &&
336 !skb->sk &&
337 !ip_is_fragment(iph)) {
338 switch (iph->protocol) {
339 case IPPROTO_TCP:
340 if (READ_ONCE(net->ipv4.sysctl_tcp_early_demux)) {
341 tcp_v4_early_demux(skb);
342
343 /* must reload iph, skb->head might have changed */
344 iph = ip_hdr(skb);
345 }
346 break;
347 case IPPROTO_UDP:
348 if (READ_ONCE(net->ipv4.sysctl_udp_early_demux)) {
349 err = udp_v4_early_demux(skb);
350 if (unlikely(err))
351 goto drop_error;
352
353 /* must reload iph, skb->head might have changed */
354 iph = ip_hdr(skb);
355 }
356 break;
357 }
358 }
359
360 /*
361 * Initialise the virtual path cache for the packet. It describes
362 * how the packet travels inside Linux networking.
363 */
364 if (!skb_valid_dst(skb)) {
365 err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
366 iph->tos, dev);
367 if (unlikely(err))
368 goto drop_error;
369 } else {
370 struct in_device *in_dev = __in_dev_get_rcu(dev);
371
372 if (in_dev && IN_DEV_ORCONF(in_dev, NOPOLICY))
373 IPCB(skb)->flags |= IPSKB_NOPOLICY;
374 }
375
376#ifdef CONFIG_IP_ROUTE_CLASSID
377 if (unlikely(skb_dst(skb)->tclassid)) {
378 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
379 u32 idx = skb_dst(skb)->tclassid;
380 st[idx&0xFF].o_packets++;
381 st[idx&0xFF].o_bytes += skb->len;
382 st[(idx>>16)&0xFF].i_packets++;
383 st[(idx>>16)&0xFF].i_bytes += skb->len;
384 }
385#endif
386
387 if (iph->ihl > 5 && ip_rcv_options(skb, dev))
388 goto drop;
389
390 rt = skb_rtable(skb);
391 if (rt->rt_type == RTN_MULTICAST) {
392 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
393 } else if (rt->rt_type == RTN_BROADCAST) {
394 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
395 } else if (skb->pkt_type == PACKET_BROADCAST ||
396 skb->pkt_type == PACKET_MULTICAST) {
397 struct in_device *in_dev = __in_dev_get_rcu(dev);
398
399 /* RFC 1122 3.3.6:
400 *
401 * When a host sends a datagram to a link-layer broadcast
402 * address, the IP destination address MUST be a legal IP
403 * broadcast or IP multicast address.
404 *
405 * A host SHOULD silently discard a datagram that is received
406 * via a link-layer broadcast (see Section 2.4) but does not
407 * specify an IP multicast or broadcast destination address.
408 *
409 * This doesn't explicitly say L2 *broadcast*, but broadcast is
410 * in a way a form of multicast and the most common use case for
411 * this is 802.11 protecting against cross-station spoofing (the
412 * so-called "hole-196" attack) so do it for both.
413 */
414 if (in_dev &&
415 IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST)) {
416 drop_reason = SKB_DROP_REASON_UNICAST_IN_L2_MULTICAST;
417 goto drop;
418 }
419 }
420
421 return NET_RX_SUCCESS;
422
423drop:
424 kfree_skb_reason(skb, drop_reason);
425 return NET_RX_DROP;
426
427drop_error:
428 if (err == -EXDEV) {
429 drop_reason = SKB_DROP_REASON_IP_RPFILTER;
430 __NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
431 }
432 goto drop;
433}
434
435static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
436{
437 struct net_device *dev = skb->dev;
438 int ret;
439
440 /* if ingress device is enslaved to an L3 master device pass the
441 * skb to its handler for processing
442 */
443 skb = l3mdev_ip_rcv(skb);
444 if (!skb)
445 return NET_RX_SUCCESS;
446
447 ret = ip_rcv_finish_core(net, sk, skb, dev, NULL);
448 if (ret != NET_RX_DROP)
449 ret = dst_input(skb);
450 return ret;
451}
452
453/*
454 * Main IP Receive routine.
455 */
456static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net)
457{
458 const struct iphdr *iph;
459 int drop_reason;
460 u32 len;
461
462 /* When the interface is in promisc. mode, drop all the crap
463 * that it receives, do not try to analyse it.
464 */
465 if (skb->pkt_type == PACKET_OTHERHOST) {
466 dev_core_stats_rx_otherhost_dropped_inc(skb->dev);
467 drop_reason = SKB_DROP_REASON_OTHERHOST;
468 goto drop;
469 }
470
471 __IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
472
473 skb = skb_share_check(skb, GFP_ATOMIC);
474 if (!skb) {
475 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
476 goto out;
477 }
478
479 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
480 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
481 goto inhdr_error;
482
483 iph = ip_hdr(skb);
484
485 /*
486 * RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
487 *
488 * Is the datagram acceptable?
489 *
490 * 1. Length at least the size of an ip header
491 * 2. Version of 4
492 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums]
493 * 4. Doesn't have a bogus length
494 */
495
496 if (iph->ihl < 5 || iph->version != 4)
497 goto inhdr_error;
498
499 BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
500 BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
501 BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
502 __IP_ADD_STATS(net,
503 IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
504 max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
505
506 if (!pskb_may_pull(skb, iph->ihl*4))
507 goto inhdr_error;
508
509 iph = ip_hdr(skb);
510
511 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
512 goto csum_error;
513
514 len = iph_totlen(skb, iph);
515 if (skb->len < len) {
516 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
517 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
518 goto drop;
519 } else if (len < (iph->ihl*4))
520 goto inhdr_error;
521
522 /* Our transport medium may have padded the buffer out. Now we know it
523 * is IP we can trim to the true length of the frame.
524 * Note this now means skb->len holds ntohs(iph->tot_len).
525 */
526 if (pskb_trim_rcsum(skb, len)) {
527 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
528 goto drop;
529 }
530
531 iph = ip_hdr(skb);
532 skb->transport_header = skb->network_header + iph->ihl*4;
533
534 /* Remove any debris in the socket control block */
535 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
536 IPCB(skb)->iif = skb->skb_iif;
537
538 /* Must drop socket now because of tproxy. */
539 if (!skb_sk_is_prefetched(skb))
540 skb_orphan(skb);
541
542 return skb;
543
544csum_error:
545 drop_reason = SKB_DROP_REASON_IP_CSUM;
546 __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
547inhdr_error:
548 if (drop_reason == SKB_DROP_REASON_NOT_SPECIFIED)
549 drop_reason = SKB_DROP_REASON_IP_INHDR;
550 __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
551drop:
552 kfree_skb_reason(skb, drop_reason);
553out:
554 return NULL;
555}
556
557/*
558 * IP receive entry point
559 */
560int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
561 struct net_device *orig_dev)
562{
563 struct net *net = dev_net(dev);
564
565 skb = ip_rcv_core(skb, net);
566 if (skb == NULL)
567 return NET_RX_DROP;
568
569 return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
570 net, NULL, skb, dev, NULL,
571 ip_rcv_finish);
572}
573
574static void ip_sublist_rcv_finish(struct list_head *head)
575{
576 struct sk_buff *skb, *next;
577
578 list_for_each_entry_safe(skb, next, head, list) {
579 skb_list_del_init(skb);
580 dst_input(skb);
581 }
582}
583
584static struct sk_buff *ip_extract_route_hint(const struct net *net,
585 struct sk_buff *skb, int rt_type)
586{
587 if (fib4_has_custom_rules(net) || rt_type == RTN_BROADCAST ||
588 IPCB(skb)->flags & IPSKB_MULTIPATH)
589 return NULL;
590
591 return skb;
592}
593
594static void ip_list_rcv_finish(struct net *net, struct sock *sk,
595 struct list_head *head)
596{
597 struct sk_buff *skb, *next, *hint = NULL;
598 struct dst_entry *curr_dst = NULL;
599 struct list_head sublist;
600
601 INIT_LIST_HEAD(&sublist);
602 list_for_each_entry_safe(skb, next, head, list) {
603 struct net_device *dev = skb->dev;
604 struct dst_entry *dst;
605
606 skb_list_del_init(skb);
607 /* if ingress device is enslaved to an L3 master device pass the
608 * skb to its handler for processing
609 */
610 skb = l3mdev_ip_rcv(skb);
611 if (!skb)
612 continue;
613 if (ip_rcv_finish_core(net, sk, skb, dev, hint) == NET_RX_DROP)
614 continue;
615
616 dst = skb_dst(skb);
617 if (curr_dst != dst) {
618 hint = ip_extract_route_hint(net, skb,
619 ((struct rtable *)dst)->rt_type);
620
621 /* dispatch old sublist */
622 if (!list_empty(&sublist))
623 ip_sublist_rcv_finish(&sublist);
624 /* start new sublist */
625 INIT_LIST_HEAD(&sublist);
626 curr_dst = dst;
627 }
628 list_add_tail(&skb->list, &sublist);
629 }
630 /* dispatch final sublist */
631 ip_sublist_rcv_finish(&sublist);
632}
633
634static void ip_sublist_rcv(struct list_head *head, struct net_device *dev,
635 struct net *net)
636{
637 NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL,
638 head, dev, NULL, ip_rcv_finish);
639 ip_list_rcv_finish(net, NULL, head);
640}
641
642/* Receive a list of IP packets */
643void ip_list_rcv(struct list_head *head, struct packet_type *pt,
644 struct net_device *orig_dev)
645{
646 struct net_device *curr_dev = NULL;
647 struct net *curr_net = NULL;
648 struct sk_buff *skb, *next;
649 struct list_head sublist;
650
651 INIT_LIST_HEAD(&sublist);
652 list_for_each_entry_safe(skb, next, head, list) {
653 struct net_device *dev = skb->dev;
654 struct net *net = dev_net(dev);
655
656 skb_list_del_init(skb);
657 skb = ip_rcv_core(skb, net);
658 if (skb == NULL)
659 continue;
660
661 if (curr_dev != dev || curr_net != net) {
662 /* dispatch old sublist */
663 if (!list_empty(&sublist))
664 ip_sublist_rcv(&sublist, curr_dev, curr_net);
665 /* start new sublist */
666 INIT_LIST_HEAD(&sublist);
667 curr_dev = dev;
668 curr_net = net;
669 }
670 list_add_tail(&skb->list, &sublist);
671 }
672 /* dispatch final sublist */
673 if (!list_empty(&sublist))
674 ip_sublist_rcv(&sublist, curr_dev, curr_net);
675}