• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 #ifdef CONFIG_LOWPOWER_PROTOCOL
145 #include <net/lowpower_protocol.h>
146 #endif /* CONFIG_LOWPOWER_PROTOCOL */
147 
148 /*
149  *	Process Router Attention IP option (RFC 2113)
150  */
ip_call_ra_chain(struct sk_buff * skb)151 bool ip_call_ra_chain(struct sk_buff *skb)
152 {
153 	struct ip_ra_chain *ra;
154 	u8 protocol = ip_hdr(skb)->protocol;
155 	struct sock *last = NULL;
156 	struct net_device *dev = skb->dev;
157 	struct net *net = dev_net(dev);
158 
159 	for (ra = rcu_dereference(net->ipv4.ra_chain); ra; ra = rcu_dereference(ra->next)) {
160 		struct sock *sk = ra->sk;
161 
162 		/* If socket is bound to an interface, only report
163 		 * the packet if it came  from that interface.
164 		 */
165 		if (sk && inet_sk(sk)->inet_num == protocol &&
166 		    (!sk->sk_bound_dev_if ||
167 		     sk->sk_bound_dev_if == dev->ifindex)) {
168 			if (ip_is_fragment(ip_hdr(skb))) {
169 				if (ip_defrag(net, skb, IP_DEFRAG_CALL_RA_CHAIN))
170 					return true;
171 			}
172 			if (last) {
173 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
174 				if (skb2)
175 					raw_rcv(last, skb2);
176 			}
177 			last = sk;
178 		}
179 	}
180 
181 	if (last) {
182 		raw_rcv(last, skb);
183 		return true;
184 	}
185 	return false;
186 }
187 
188 INDIRECT_CALLABLE_DECLARE(int udp_rcv(struct sk_buff *));
189 INDIRECT_CALLABLE_DECLARE(int tcp_v4_rcv(struct sk_buff *));
ip_protocol_deliver_rcu(struct net * net,struct sk_buff * skb,int protocol)190 void ip_protocol_deliver_rcu(struct net *net, struct sk_buff *skb, int protocol)
191 {
192 	const struct net_protocol *ipprot;
193 	int raw, ret;
194 
195 resubmit:
196 	raw = raw_local_deliver(skb, protocol);
197 
198 	ipprot = rcu_dereference(inet_protos[protocol]);
199 	if (ipprot) {
200 		if (!ipprot->no_policy) {
201 			if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
202 				kfree_skb(skb);
203 				return;
204 			}
205 			nf_reset_ct(skb);
206 		}
207 		ret = INDIRECT_CALL_2(ipprot->handler, tcp_v4_rcv, udp_rcv,
208 				      skb);
209 		if (ret < 0) {
210 			protocol = -ret;
211 			goto resubmit;
212 		}
213 		__IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
214 	} else {
215 		if (!raw) {
216 			if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
217 				__IP_INC_STATS(net, IPSTATS_MIB_INUNKNOWNPROTOS);
218 				icmp_send(skb, ICMP_DEST_UNREACH,
219 					  ICMP_PROT_UNREACH, 0);
220 			}
221 			kfree_skb(skb);
222 		} else {
223 			__IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
224 			consume_skb(skb);
225 		}
226 	}
227 }
228 
ip_local_deliver_finish(struct net * net,struct sock * sk,struct sk_buff * skb)229 static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
230 {
231 	__skb_pull(skb, skb_network_header_len(skb));
232 
233 	rcu_read_lock();
234 	ip_protocol_deliver_rcu(net, skb, ip_hdr(skb)->protocol);
235 	rcu_read_unlock();
236 
237 	return 0;
238 }
239 
240 /*
241  * 	Deliver IP Packets to the higher protocol layers.
242  */
ip_local_deliver(struct sk_buff * skb)243 int ip_local_deliver(struct sk_buff *skb)
244 {
245 	/*
246 	 *	Reassemble IP fragments.
247 	 */
248 	struct net *net = dev_net(skb->dev);
249 #ifdef CONFIG_LOWPOWER_PROTOCOL
250 	int ret;
251 #endif /* CONFIG_LOWPOWER_PROTOCOL */
252 
253 	if (ip_is_fragment(ip_hdr(skb))) {
254 		if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
255 			return 0;
256 	}
257 
258 #ifdef CONFIG_LOWPOWER_PROTOCOL
259 	if (netfilter_bypass_enable(net, skb, ip_local_deliver_finish, &ret))
260 		return ret;
261 #endif /* CONFIG_LOWPOWER_PROTOCOL */
262 
263 	return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
264 		       net, NULL, skb, skb->dev, NULL,
265 		       ip_local_deliver_finish);
266 }
267 
ip_rcv_options(struct sk_buff * skb,struct net_device * dev)268 static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev)
269 {
270 	struct ip_options *opt;
271 	const struct iphdr *iph;
272 
273 	/* It looks as overkill, because not all
274 	   IP options require packet mangling.
275 	   But it is the easiest for now, especially taking
276 	   into account that combination of IP options
277 	   and running sniffer is extremely rare condition.
278 					      --ANK (980813)
279 	*/
280 	if (skb_cow(skb, skb_headroom(skb))) {
281 		__IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
282 		goto drop;
283 	}
284 
285 	iph = ip_hdr(skb);
286 	opt = &(IPCB(skb)->opt);
287 	opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
288 
289 	if (ip_options_compile(dev_net(dev), opt, skb)) {
290 		__IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
291 		goto drop;
292 	}
293 
294 	if (unlikely(opt->srr)) {
295 		struct in_device *in_dev = __in_dev_get_rcu(dev);
296 
297 		if (in_dev) {
298 			if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
299 				if (IN_DEV_LOG_MARTIANS(in_dev))
300 					net_info_ratelimited("source route option %pI4 -> %pI4\n",
301 							     &iph->saddr,
302 							     &iph->daddr);
303 				goto drop;
304 			}
305 		}
306 
307 		if (ip_options_rcv_srr(skb, dev))
308 			goto drop;
309 	}
310 
311 	return false;
312 drop:
313 	return true;
314 }
315 
ip_can_use_hint(const struct sk_buff * skb,const struct iphdr * iph,const struct sk_buff * hint)316 static bool ip_can_use_hint(const struct sk_buff *skb, const struct iphdr *iph,
317 			    const struct sk_buff *hint)
318 {
319 	return hint && !skb_dst(skb) && ip_hdr(hint)->daddr == iph->daddr &&
320 	       ip_hdr(hint)->tos == iph->tos;
321 }
322 
323 int tcp_v4_early_demux(struct sk_buff *skb);
324 int udp_v4_early_demux(struct sk_buff *skb);
ip_rcv_finish_core(struct net * net,struct sock * sk,struct sk_buff * skb,struct net_device * dev,const struct sk_buff * hint)325 static int ip_rcv_finish_core(struct net *net, struct sock *sk,
326 			      struct sk_buff *skb, struct net_device *dev,
327 			      const struct sk_buff *hint)
328 {
329 	const struct iphdr *iph = ip_hdr(skb);
330 	struct rtable *rt;
331 	int err;
332 
333 	if (ip_can_use_hint(skb, iph, hint)) {
334 		err = ip_route_use_hint(skb, iph->daddr, iph->saddr, iph->tos,
335 					dev, hint);
336 		if (unlikely(err))
337 			goto drop_error;
338 	}
339 
340 	if (READ_ONCE(net->ipv4.sysctl_ip_early_demux) &&
341 	    !skb_dst(skb) &&
342 	    !skb->sk &&
343 	    !ip_is_fragment(iph)) {
344 		switch (iph->protocol) {
345 		case IPPROTO_TCP:
346 			if (READ_ONCE(net->ipv4.sysctl_tcp_early_demux)) {
347 				tcp_v4_early_demux(skb);
348 
349 				/* must reload iph, skb->head might have changed */
350 				iph = ip_hdr(skb);
351 			}
352 			break;
353 		case IPPROTO_UDP:
354 			if (READ_ONCE(net->ipv4.sysctl_udp_early_demux)) {
355 				err = udp_v4_early_demux(skb);
356 				if (unlikely(err))
357 					goto drop_error;
358 
359 				/* must reload iph, skb->head might have changed */
360 				iph = ip_hdr(skb);
361 			}
362 			break;
363 		}
364 	}
365 
366 	/*
367 	 *	Initialise the virtual path cache for the packet. It describes
368 	 *	how the packet travels inside Linux networking.
369 	 */
370 	if (!skb_valid_dst(skb)) {
371 		err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
372 					   iph->tos, dev);
373 		if (unlikely(err))
374 			goto drop_error;
375 	} else {
376 		struct in_device *in_dev = __in_dev_get_rcu(dev);
377 
378 		if (in_dev && IN_DEV_ORCONF(in_dev, NOPOLICY))
379 			IPCB(skb)->flags |= IPSKB_NOPOLICY;
380 	}
381 
382 #ifdef CONFIG_IP_ROUTE_CLASSID
383 	if (unlikely(skb_dst(skb)->tclassid)) {
384 		struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
385 		u32 idx = skb_dst(skb)->tclassid;
386 		st[idx&0xFF].o_packets++;
387 		st[idx&0xFF].o_bytes += skb->len;
388 		st[(idx>>16)&0xFF].i_packets++;
389 		st[(idx>>16)&0xFF].i_bytes += skb->len;
390 	}
391 #endif
392 
393 	if (iph->ihl > 5 && ip_rcv_options(skb, dev))
394 		goto drop;
395 
396 	rt = skb_rtable(skb);
397 	if (rt->rt_type == RTN_MULTICAST) {
398 		__IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
399 	} else if (rt->rt_type == RTN_BROADCAST) {
400 		__IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
401 	} else if (skb->pkt_type == PACKET_BROADCAST ||
402 		   skb->pkt_type == PACKET_MULTICAST) {
403 		struct in_device *in_dev = __in_dev_get_rcu(dev);
404 
405 		/* RFC 1122 3.3.6:
406 		 *
407 		 *   When a host sends a datagram to a link-layer broadcast
408 		 *   address, the IP destination address MUST be a legal IP
409 		 *   broadcast or IP multicast address.
410 		 *
411 		 *   A host SHOULD silently discard a datagram that is received
412 		 *   via a link-layer broadcast (see Section 2.4) but does not
413 		 *   specify an IP multicast or broadcast destination address.
414 		 *
415 		 * This doesn't explicitly say L2 *broadcast*, but broadcast is
416 		 * in a way a form of multicast and the most common use case for
417 		 * this is 802.11 protecting against cross-station spoofing (the
418 		 * so-called "hole-196" attack) so do it for both.
419 		 */
420 		if (in_dev &&
421 		    IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST))
422 			goto drop;
423 	}
424 
425 	return NET_RX_SUCCESS;
426 
427 drop:
428 	kfree_skb(skb);
429 	return NET_RX_DROP;
430 
431 drop_error:
432 	if (err == -EXDEV)
433 		__NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
434 	goto drop;
435 }
436 
ip_rcv_finish(struct net * net,struct sock * sk,struct sk_buff * skb)437 static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
438 {
439 	struct net_device *dev = skb->dev;
440 	int ret;
441 
442 	/* if ingress device is enslaved to an L3 master device pass the
443 	 * skb to its handler for processing
444 	 */
445 	skb = l3mdev_ip_rcv(skb);
446 	if (!skb)
447 		return NET_RX_SUCCESS;
448 
449 	ret = ip_rcv_finish_core(net, sk, skb, dev, NULL);
450 	if (ret != NET_RX_DROP)
451 		ret = dst_input(skb);
452 	return ret;
453 }
454 
455 /*
456  * 	Main IP Receive routine.
457  */
ip_rcv_core(struct sk_buff * skb,struct net * net)458 static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net)
459 {
460 	const struct iphdr *iph;
461 	u32 len;
462 
463 	/* When the interface is in promisc. mode, drop all the crap
464 	 * that it receives, do not try to analyse it.
465 	 */
466 	if (skb->pkt_type == PACKET_OTHERHOST)
467 		goto drop;
468 
469 	__IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
470 
471 	skb = skb_share_check(skb, GFP_ATOMIC);
472 	if (!skb) {
473 		__IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
474 		goto out;
475 	}
476 
477 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
478 		goto inhdr_error;
479 
480 	iph = ip_hdr(skb);
481 
482 	/*
483 	 *	RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
484 	 *
485 	 *	Is the datagram acceptable?
486 	 *
487 	 *	1.	Length at least the size of an ip header
488 	 *	2.	Version of 4
489 	 *	3.	Checksums correctly. [Speed optimisation for later, skip loopback checksums]
490 	 *	4.	Doesn't have a bogus length
491 	 */
492 
493 	if (iph->ihl < 5 || iph->version != 4)
494 		goto inhdr_error;
495 
496 	BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
497 	BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
498 	BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
499 	__IP_ADD_STATS(net,
500 		       IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
501 		       max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
502 
503 	if (!pskb_may_pull(skb, iph->ihl*4))
504 		goto inhdr_error;
505 
506 	iph = ip_hdr(skb);
507 
508 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
509 		goto csum_error;
510 
511 	len = ntohs(iph->tot_len);
512 	if (skb->len < len) {
513 		__IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
514 		goto drop;
515 	} else if (len < (iph->ihl*4))
516 		goto inhdr_error;
517 
518 	/* Our transport medium may have padded the buffer out. Now we know it
519 	 * is IP we can trim to the true length of the frame.
520 	 * Note this now means skb->len holds ntohs(iph->tot_len).
521 	 */
522 	if (pskb_trim_rcsum(skb, len)) {
523 		__IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
524 		goto drop;
525 	}
526 
527 	iph = ip_hdr(skb);
528 	skb->transport_header = skb->network_header + iph->ihl*4;
529 
530 	/* Remove any debris in the socket control block */
531 	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
532 	IPCB(skb)->iif = skb->skb_iif;
533 
534 	/* Must drop socket now because of tproxy. */
535 	if (!skb_sk_is_prefetched(skb))
536 		skb_orphan(skb);
537 
538 	return skb;
539 
540 csum_error:
541 	__IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
542 inhdr_error:
543 	__IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
544 drop:
545 	kfree_skb(skb);
546 out:
547 	return NULL;
548 }
549 
550 /*
551  * IP receive entry point
552  */
ip_rcv(struct sk_buff * skb,struct net_device * dev,struct packet_type * pt,struct net_device * orig_dev)553 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
554 	   struct net_device *orig_dev)
555 {
556 	struct net *net = dev_net(dev);
557 
558 	skb = ip_rcv_core(skb, net);
559 	if (skb == NULL)
560 		return NET_RX_DROP;
561 
562 	return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
563 		       net, NULL, skb, dev, NULL,
564 		       ip_rcv_finish);
565 }
566 
ip_sublist_rcv_finish(struct list_head * head)567 static void ip_sublist_rcv_finish(struct list_head *head)
568 {
569 	struct sk_buff *skb, *next;
570 
571 	list_for_each_entry_safe(skb, next, head, list) {
572 		skb_list_del_init(skb);
573 		dst_input(skb);
574 	}
575 }
576 
ip_extract_route_hint(const struct net * net,struct sk_buff * skb,int rt_type)577 static struct sk_buff *ip_extract_route_hint(const struct net *net,
578 					     struct sk_buff *skb, int rt_type)
579 {
580 	if (fib4_has_custom_rules(net) || rt_type == RTN_BROADCAST)
581 		return NULL;
582 
583 	return skb;
584 }
585 
ip_list_rcv_finish(struct net * net,struct sock * sk,struct list_head * head)586 static void ip_list_rcv_finish(struct net *net, struct sock *sk,
587 			       struct list_head *head)
588 {
589 	struct sk_buff *skb, *next, *hint = NULL;
590 	struct dst_entry *curr_dst = NULL;
591 	struct list_head sublist;
592 
593 	INIT_LIST_HEAD(&sublist);
594 	list_for_each_entry_safe(skb, next, head, list) {
595 		struct net_device *dev = skb->dev;
596 		struct dst_entry *dst;
597 
598 		skb_list_del_init(skb);
599 		/* if ingress device is enslaved to an L3 master device pass the
600 		 * skb to its handler for processing
601 		 */
602 		skb = l3mdev_ip_rcv(skb);
603 		if (!skb)
604 			continue;
605 		if (ip_rcv_finish_core(net, sk, skb, dev, hint) == NET_RX_DROP)
606 			continue;
607 
608 		dst = skb_dst(skb);
609 		if (curr_dst != dst) {
610 			hint = ip_extract_route_hint(net, skb,
611 					       ((struct rtable *)dst)->rt_type);
612 
613 			/* dispatch old sublist */
614 			if (!list_empty(&sublist))
615 				ip_sublist_rcv_finish(&sublist);
616 			/* start new sublist */
617 			INIT_LIST_HEAD(&sublist);
618 			curr_dst = dst;
619 		}
620 		list_add_tail(&skb->list, &sublist);
621 	}
622 	/* dispatch final sublist */
623 	ip_sublist_rcv_finish(&sublist);
624 }
625 
ip_sublist_rcv(struct list_head * head,struct net_device * dev,struct net * net)626 static void ip_sublist_rcv(struct list_head *head, struct net_device *dev,
627 			   struct net *net)
628 {
629 	NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL,
630 		     head, dev, NULL, ip_rcv_finish);
631 	ip_list_rcv_finish(net, NULL, head);
632 }
633 
634 /* Receive a list of IP packets */
ip_list_rcv(struct list_head * head,struct packet_type * pt,struct net_device * orig_dev)635 void ip_list_rcv(struct list_head *head, struct packet_type *pt,
636 		 struct net_device *orig_dev)
637 {
638 	struct net_device *curr_dev = NULL;
639 	struct net *curr_net = NULL;
640 	struct sk_buff *skb, *next;
641 	struct list_head sublist;
642 
643 	INIT_LIST_HEAD(&sublist);
644 	list_for_each_entry_safe(skb, next, head, list) {
645 		struct net_device *dev = skb->dev;
646 		struct net *net = dev_net(dev);
647 
648 		skb_list_del_init(skb);
649 		skb = ip_rcv_core(skb, net);
650 		if (skb == NULL)
651 			continue;
652 
653 		if (curr_dev != dev || curr_net != net) {
654 			/* dispatch old sublist */
655 			if (!list_empty(&sublist))
656 				ip_sublist_rcv(&sublist, curr_dev, curr_net);
657 			/* start new sublist */
658 			INIT_LIST_HEAD(&sublist);
659 			curr_dev = dev;
660 			curr_net = net;
661 		}
662 		list_add_tail(&skb->list, &sublist);
663 	}
664 	/* dispatch final sublist */
665 	if (!list_empty(&sublist))
666 		ip_sublist_rcv(&sublist, curr_dev, curr_net);
667 }
668