Linux Audio

Check our new training course

Loading...
v3.15
 
  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}
v5.14.15
  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}
256EXPORT_SYMBOL(ip_local_deliver);
257
258static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev)
259{
260	struct ip_options *opt;
261	const struct iphdr *iph;
 
262
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.
268					      --ANK (980813)
269	*/
270	if (skb_cow(skb, skb_headroom(skb))) {
271		__IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
272		goto drop;
273	}
274
275	iph = ip_hdr(skb);
276	opt = &(IPCB(skb)->opt);
277	opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
278
279	if (ip_options_compile(dev_net(dev), opt, skb)) {
280		__IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
281		goto drop;
282	}
283
284	if (unlikely(opt->srr)) {
285		struct in_device *in_dev = __in_dev_get_rcu(dev);
286
287		if (in_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",
291							     &iph->saddr,
292							     &iph->daddr);
293				goto drop;
294			}
295		}
296
297		if (ip_options_rcv_srr(skb, dev))
298			goto drop;
299	}
300
301	return false;
302drop:
303	return true;
304}
305
306static bool ip_can_use_hint(const struct sk_buff *skb, const struct iphdr *iph,
307			    const struct sk_buff *hint)
308{
309	return hint && !skb_dst(skb) && ip_hdr(hint)->daddr == iph->daddr &&
310	       ip_hdr(hint)->tos == iph->tos;
311}
312
313INDIRECT_CALLABLE_DECLARE(int udp_v4_early_demux(struct sk_buff *));
314INDIRECT_CALLABLE_DECLARE(int tcp_v4_early_demux(struct sk_buff *));
315static 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)
318{
319	const struct iphdr *iph = ip_hdr(skb);
320	int (*edemux)(struct sk_buff *skb);
321	struct rtable *rt;
322	int err;
323
324	if (ip_can_use_hint(skb, iph, hint)) {
325		err = ip_route_use_hint(skb, iph->daddr, iph->saddr, iph->tos,
326					dev, hint);
327		if (unlikely(err))
328			goto drop_error;
329	}
330
331	if (net->ipv4.sysctl_ip_early_demux &&
332	    !skb_dst(skb) &&
333	    !skb->sk &&
334	    !ip_is_fragment(iph)) {
335		const struct net_protocol *ipprot;
336		int protocol = iph->protocol;
337
338		ipprot = rcu_dereference(inet_protos[protocol]);
339		if (ipprot && (edemux = READ_ONCE(ipprot->early_demux))) {
340			err = INDIRECT_CALL_2(edemux, tcp_v4_early_demux,
341					      udp_v4_early_demux, skb);
342			if (unlikely(err))
343				goto drop_error;
344			/* must reload iph, skb->head might have changed */
345			iph = ip_hdr(skb);
346		}
347	}
348
349	/*
350	 *	Initialise the virtual path cache for the packet. It describes
351	 *	how the packet travels inside Linux networking.
352	 */
353	if (!skb_valid_dst(skb)) {
354		err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
355					   iph->tos, dev);
356		if (unlikely(err))
357			goto drop_error;
 
 
 
 
358	}
359
360#ifdef CONFIG_IP_ROUTE_CLASSID
361	if (unlikely(skb_dst(skb)->tclassid)) {
362		struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
363		u32 idx = skb_dst(skb)->tclassid;
364		st[idx&0xFF].o_packets++;
365		st[idx&0xFF].o_bytes += skb->len;
366		st[(idx>>16)&0xFF].i_packets++;
367		st[(idx>>16)&0xFF].i_bytes += skb->len;
368	}
369#endif
370
371	if (iph->ihl > 5 && ip_rcv_options(skb, dev))
372		goto drop;
373
374	rt = skb_rtable(skb);
375	if (rt->rt_type == RTN_MULTICAST) {
376		__IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
377	} else if (rt->rt_type == RTN_BROADCAST) {
378		__IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
379	} else if (skb->pkt_type == PACKET_BROADCAST ||
380		   skb->pkt_type == PACKET_MULTICAST) {
381		struct in_device *in_dev = __in_dev_get_rcu(dev);
382
383		/* RFC 1122 3.3.6:
384		 *
385		 *   When a host sends a datagram to a link-layer broadcast
386		 *   address, the IP destination address MUST be a legal IP
387		 *   broadcast or IP multicast address.
388		 *
389		 *   A host SHOULD silently discard a datagram that is received
390		 *   via a link-layer broadcast (see Section 2.4) but does not
391		 *   specify an IP multicast or broadcast destination address.
392		 *
393		 * This doesn't explicitly say L2 *broadcast*, but broadcast is
394		 * in a way a form of multicast and the most common use case for
395		 * this is 802.11 protecting against cross-station spoofing (the
396		 * so-called "hole-196" attack) so do it for both.
397		 */
398		if (in_dev &&
399		    IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST))
400			goto drop;
401	}
402
403	return NET_RX_SUCCESS;
404
405drop:
406	kfree_skb(skb);
407	return NET_RX_DROP;
408
409drop_error:
410	if (err == -EXDEV)
411		__NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
412	goto drop;
413}
414
415static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
416{
417	struct net_device *dev = skb->dev;
418	int ret;
419
420	/* if ingress device is enslaved to an L3 master device pass the
421	 * skb to its handler for processing
422	 */
423	skb = l3mdev_ip_rcv(skb);
424	if (!skb)
425		return NET_RX_SUCCESS;
426
427	ret = ip_rcv_finish_core(net, sk, skb, dev, NULL);
428	if (ret != NET_RX_DROP)
429		ret = dst_input(skb);
430	return ret;
431}
432
433/*
434 * 	Main IP Receive routine.
435 */
436static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net)
437{
438	const struct iphdr *iph;
439	u32 len;
440
441	/* When the interface is in promisc. mode, drop all the crap
442	 * that it receives, do not try to analyse it.
443	 */
444	if (skb->pkt_type == PACKET_OTHERHOST)
445		goto drop;
446
447	__IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
448
449	skb = skb_share_check(skb, GFP_ATOMIC);
450	if (!skb) {
451		__IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
 
452		goto out;
453	}
454
455	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
456		goto inhdr_error;
457
458	iph = ip_hdr(skb);
459
460	/*
461	 *	RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
462	 *
463	 *	Is the datagram acceptable?
464	 *
465	 *	1.	Length at least the size of an ip header
466	 *	2.	Version of 4
467	 *	3.	Checksums correctly. [Speed optimisation for later, skip loopback checksums]
468	 *	4.	Doesn't have a bogus length
469	 */
470
471	if (iph->ihl < 5 || iph->version != 4)
472		goto inhdr_error;
473
474	BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
475	BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
476	BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
477	__IP_ADD_STATS(net,
478		       IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
479		       max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
480
481	if (!pskb_may_pull(skb, iph->ihl*4))
482		goto inhdr_error;
483
484	iph = ip_hdr(skb);
485
486	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
487		goto csum_error;
488
489	len = ntohs(iph->tot_len);
490	if (skb->len < len) {
491		__IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
492		goto drop;
493	} else if (len < (iph->ihl*4))
494		goto inhdr_error;
495
496	/* Our transport medium may have padded the buffer out. Now we know it
497	 * is IP we can trim to the true length of the frame.
498	 * Note this now means skb->len holds ntohs(iph->tot_len).
499	 */
500	if (pskb_trim_rcsum(skb, len)) {
501		__IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
502		goto drop;
503	}
504
505	iph = ip_hdr(skb);
506	skb->transport_header = skb->network_header + iph->ihl*4;
507
508	/* Remove any debris in the socket control block */
509	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
510	IPCB(skb)->iif = skb->skb_iif;
511
512	/* Must drop socket now because of tproxy. */
513	if (!skb_sk_is_prefetched(skb))
514		skb_orphan(skb);
515
516	return skb;
 
517
518csum_error:
519	__IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
520inhdr_error:
521	__IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
522drop:
523	kfree_skb(skb);
524out:
525	return NULL;
526}
527
528/*
529 * IP receive entry point
530 */
531int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
532	   struct net_device *orig_dev)
533{
534	struct net *net = dev_net(dev);
535
536	skb = ip_rcv_core(skb, net);
537	if (skb == NULL)
538		return NET_RX_DROP;
539
540	return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
541		       net, NULL, skb, dev, NULL,
542		       ip_rcv_finish);
543}
544
545static void ip_sublist_rcv_finish(struct list_head *head)
546{
547	struct sk_buff *skb, *next;
548
549	list_for_each_entry_safe(skb, next, head, list) {
550		skb_list_del_init(skb);
551		dst_input(skb);
552	}
553}
554
555static struct sk_buff *ip_extract_route_hint(const struct net *net,
556					     struct sk_buff *skb, int rt_type)
557{
558	if (fib4_has_custom_rules(net) || rt_type == RTN_BROADCAST)
559		return NULL;
560
561	return skb;
562}
563
564static void ip_list_rcv_finish(struct net *net, struct sock *sk,
565			       struct list_head *head)
566{
567	struct sk_buff *skb, *next, *hint = NULL;
568	struct dst_entry *curr_dst = NULL;
569	struct list_head sublist;
570
571	INIT_LIST_HEAD(&sublist);
572	list_for_each_entry_safe(skb, next, head, list) {
573		struct net_device *dev = skb->dev;
574		struct dst_entry *dst;
575
576		skb_list_del_init(skb);
577		/* if ingress device is enslaved to an L3 master device pass the
578		 * skb to its handler for processing
579		 */
580		skb = l3mdev_ip_rcv(skb);
581		if (!skb)
582			continue;
583		if (ip_rcv_finish_core(net, sk, skb, dev, hint) == NET_RX_DROP)
584			continue;
585
586		dst = skb_dst(skb);
587		if (curr_dst != dst) {
588			hint = ip_extract_route_hint(net, skb,
589					       ((struct rtable *)dst)->rt_type);
590
591			/* dispatch old sublist */
592			if (!list_empty(&sublist))
593				ip_sublist_rcv_finish(&sublist);
594			/* start new sublist */
595			INIT_LIST_HEAD(&sublist);
596			curr_dst = dst;
597		}
598		list_add_tail(&skb->list, &sublist);
599	}
600	/* dispatch final sublist */
601	ip_sublist_rcv_finish(&sublist);
602}
603
604static void ip_sublist_rcv(struct list_head *head, struct net_device *dev,
605			   struct net *net)
606{
607	NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL,
608		     head, dev, NULL, ip_rcv_finish);
609	ip_list_rcv_finish(net, NULL, head);
610}
611
612/* Receive a list of IP packets */
613void ip_list_rcv(struct list_head *head, struct packet_type *pt,
614		 struct net_device *orig_dev)
615{
616	struct net_device *curr_dev = NULL;
617	struct net *curr_net = NULL;
618	struct sk_buff *skb, *next;
619	struct list_head sublist;
620
621	INIT_LIST_HEAD(&sublist);
622	list_for_each_entry_safe(skb, next, head, list) {
623		struct net_device *dev = skb->dev;
624		struct net *net = dev_net(dev);
625
626		skb_list_del_init(skb);
627		skb = ip_rcv_core(skb, net);
628		if (skb == NULL)
629			continue;
630
631		if (curr_dev != dev || curr_net != net) {
632			/* dispatch old sublist */
633			if (!list_empty(&sublist))
634				ip_sublist_rcv(&sublist, curr_dev, curr_net);
635			/* start new sublist */
636			INIT_LIST_HEAD(&sublist);
637			curr_dev = dev;
638			curr_net = net;
639		}
640		list_add_tail(&skb->list, &sublist);
641	}
642	/* dispatch final sublist */
643	if (!list_empty(&sublist))
644		ip_sublist_rcv(&sublist, curr_dev, curr_net);
645}