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