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 User Datagram Protocol (UDP).
7 *
8 * Authors:	Ross Biro
9 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 *		Hirokazu Takahashi, <taka@valinux.co.jp>
13 *
14 * Fixes:
15 *		Alan Cox	:	verify_area() calls
16 *		Alan Cox	: 	stopped close while in use off icmp
17 *					messages. Not a fix but a botch that
18 *					for udp at least is 'valid'.
19 *		Alan Cox	:	Fixed icmp handling properly
20 *		Alan Cox	: 	Correct error for oversized datagrams
21 *		Alan Cox	:	Tidied select() semantics.
22 *		Alan Cox	:	udp_err() fixed properly, also now
23 *					select and read wake correctly on errors
24 *		Alan Cox	:	udp_send verify_area moved to avoid mem leak
25 *		Alan Cox	:	UDP can count its memory
26 *		Alan Cox	:	send to an unknown connection causes
27 *					an ECONNREFUSED off the icmp, but
28 *					does NOT close.
29 *		Alan Cox	:	Switched to new sk_buff handlers. No more backlog!
30 *		Alan Cox	:	Using generic datagram code. Even smaller and the PEEK
31 *					bug no longer crashes it.
32 *		Fred Van Kempen	: 	Net2e support for sk->broadcast.
33 *		Alan Cox	:	Uses skb_free_datagram
34 *		Alan Cox	:	Added get/set sockopt support.
35 *		Alan Cox	:	Broadcasting without option set returns EACCES.
36 *		Alan Cox	:	No wakeup calls. Instead we now use the callbacks.
37 *		Alan Cox	:	Use ip_tos and ip_ttl
38 *		Alan Cox	:	SNMP Mibs
39 *		Alan Cox	:	MSG_DONTROUTE, and 0.0.0.0 support.
40 *		Matt Dillon	:	UDP length checks.
41 *		Alan Cox	:	Smarter af_inet used properly.
42 *		Alan Cox	:	Use new kernel side addressing.
43 *		Alan Cox	:	Incorrect return on truncated datagram receive.
44 *	Arnt Gulbrandsen 	:	New udp_send and stuff
45 *		Alan Cox	:	Cache last socket
46 *		Alan Cox	:	Route cache
47 *		Jon Peatfield	:	Minor efficiency fix to sendto().
48 *		Mike Shaver	:	RFC1122 checks.
49 *		Alan Cox	:	Nonblocking error fix.
50 *	Willy Konynenberg	:	Transparent proxying support.
51 *		Mike McLagan	:	Routing by source
52 *		David S. Miller	:	New socket lookup architecture.
53 *					Last socket cache retained as it
54 *					does have a high hit rate.
55 *		Olaf Kirch	:	Don't linearise iovec on sendmsg.
56 *		Andi Kleen	:	Some cleanups, cache destination entry
57 *					for connect.
58 *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
59 *		Melvin Smith	:	Check msg_name not msg_namelen in sendto(),
60 *					return ENOTCONN for unconnected sockets (POSIX)
61 *		Janos Farkas	:	don't deliver multi/broadcasts to a different
62 *					bound-to-device socket
63 *	Hirokazu Takahashi	:	HW checksumming for outgoing UDP
64 *					datagrams.
65 *	Hirokazu Takahashi	:	sendfile() on UDP works now.
66 *		Arnaldo C. Melo :	convert /proc/net/udp to seq_file
67 *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
68 *	Alexey Kuznetsov:		allow both IPv4 and IPv6 sockets to bind
69 *					a single port at the same time.
70 *	Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 *	James Chapman		:	Add L2TP encapsulation type.
72 *
73 *
74 *		This program is free software; you can redistribute it and/or
75 *		modify it under the terms of the GNU General Public License
76 *		as published by the Free Software Foundation; either version
77 *		2 of the License, or (at your option) any later version.
78 */
79
80#define pr_fmt(fmt) "UDP: " fmt
81
82#include <asm/uaccess.h>
83#include <asm/ioctls.h>
84#include <linux/bootmem.h>
85#include <linux/highmem.h>
86#include <linux/swap.h>
87#include <linux/types.h>
88#include <linux/fcntl.h>
89#include <linux/module.h>
90#include <linux/socket.h>
91#include <linux/sockios.h>
92#include <linux/igmp.h>
93#include <linux/in.h>
94#include <linux/errno.h>
95#include <linux/timer.h>
96#include <linux/mm.h>
97#include <linux/inet.h>
98#include <linux/netdevice.h>
99#include <linux/slab.h>
100#include <net/tcp_states.h>
101#include <linux/skbuff.h>
102#include <linux/netdevice.h>
103#include <linux/proc_fs.h>
104#include <linux/seq_file.h>
105#include <net/net_namespace.h>
106#include <net/icmp.h>
107#include <net/inet_hashtables.h>
108#include <net/route.h>
109#include <net/checksum.h>
110#include <net/xfrm.h>
111#include <trace/events/udp.h>
112#include <linux/static_key.h>
113#include <trace/events/skb.h>
114#include <net/busy_poll.h>
115#include "udp_impl.h"
116
117struct udp_table udp_table __read_mostly;
118EXPORT_SYMBOL(udp_table);
119
120long sysctl_udp_mem[3] __read_mostly;
121EXPORT_SYMBOL(sysctl_udp_mem);
122
123int sysctl_udp_rmem_min __read_mostly;
124EXPORT_SYMBOL(sysctl_udp_rmem_min);
125
126int sysctl_udp_wmem_min __read_mostly;
127EXPORT_SYMBOL(sysctl_udp_wmem_min);
128
129atomic_long_t udp_memory_allocated;
130EXPORT_SYMBOL(udp_memory_allocated);
131
132#define MAX_UDP_PORTS 65536
133#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
134
135static int udp_lib_lport_inuse(struct net *net, __u16 num,
136			       const struct udp_hslot *hslot,
137			       unsigned long *bitmap,
138			       struct sock *sk,
139			       int (*saddr_comp)(const struct sock *sk1,
140						 const struct sock *sk2),
141			       unsigned int log)
142{
143	struct sock *sk2;
144	struct hlist_nulls_node *node;
145	kuid_t uid = sock_i_uid(sk);
146
147	sk_nulls_for_each(sk2, node, &hslot->head)
148		if (net_eq(sock_net(sk2), net) &&
149		    sk2 != sk &&
150		    (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
151		    (!sk2->sk_reuse || !sk->sk_reuse) &&
152		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
153		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
154		    (!sk2->sk_reuseport || !sk->sk_reuseport ||
155		      !uid_eq(uid, sock_i_uid(sk2))) &&
156		    (*saddr_comp)(sk, sk2)) {
157			if (bitmap)
158				__set_bit(udp_sk(sk2)->udp_port_hash >> log,
159					  bitmap);
160			else
161				return 1;
162		}
163	return 0;
164}
165
166/*
167 * Note: we still hold spinlock of primary hash chain, so no other writer
168 * can insert/delete a socket with local_port == num
169 */
170static int udp_lib_lport_inuse2(struct net *net, __u16 num,
171			       struct udp_hslot *hslot2,
172			       struct sock *sk,
173			       int (*saddr_comp)(const struct sock *sk1,
174						 const struct sock *sk2))
175{
176	struct sock *sk2;
177	struct hlist_nulls_node *node;
178	kuid_t uid = sock_i_uid(sk);
179	int res = 0;
180
181	spin_lock(&hslot2->lock);
182	udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
183		if (net_eq(sock_net(sk2), net) &&
184		    sk2 != sk &&
185		    (udp_sk(sk2)->udp_port_hash == num) &&
186		    (!sk2->sk_reuse || !sk->sk_reuse) &&
187		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
188		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
189		    (!sk2->sk_reuseport || !sk->sk_reuseport ||
190		      !uid_eq(uid, sock_i_uid(sk2))) &&
191		    (*saddr_comp)(sk, sk2)) {
192			res = 1;
193			break;
194		}
195	spin_unlock(&hslot2->lock);
196	return res;
197}
198
199/**
200 *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
201 *
202 *  @sk:          socket struct in question
203 *  @snum:        port number to look up
204 *  @saddr_comp:  AF-dependent comparison of bound local IP addresses
205 *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
206 *                   with NULL address
207 */
208int udp_lib_get_port(struct sock *sk, unsigned short snum,
209		       int (*saddr_comp)(const struct sock *sk1,
210					 const struct sock *sk2),
211		     unsigned int hash2_nulladdr)
212{
213	struct udp_hslot *hslot, *hslot2;
214	struct udp_table *udptable = sk->sk_prot->h.udp_table;
215	int    error = 1;
216	struct net *net = sock_net(sk);
217
218	if (!snum) {
219		int low, high, remaining;
220		unsigned int rand;
221		unsigned short first, last;
222		DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
223
224		inet_get_local_port_range(net, &low, &high);
225		remaining = (high - low) + 1;
226
227		rand = prandom_u32();
228		first = reciprocal_scale(rand, remaining) + low;
229		/*
230		 * force rand to be an odd multiple of UDP_HTABLE_SIZE
231		 */
232		rand = (rand | 1) * (udptable->mask + 1);
233		last = first + udptable->mask + 1;
234		do {
235			hslot = udp_hashslot(udptable, net, first);
236			bitmap_zero(bitmap, PORTS_PER_CHAIN);
237			spin_lock_bh(&hslot->lock);
238			udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
239					    saddr_comp, udptable->log);
240
241			snum = first;
242			/*
243			 * Iterate on all possible values of snum for this hash.
244			 * Using steps of an odd multiple of UDP_HTABLE_SIZE
245			 * give us randomization and full range coverage.
246			 */
247			do {
248				if (low <= snum && snum <= high &&
249				    !test_bit(snum >> udptable->log, bitmap) &&
250				    !inet_is_local_reserved_port(net, snum))
251					goto found;
252				snum += rand;
253			} while (snum != first);
254			spin_unlock_bh(&hslot->lock);
255		} while (++first != last);
256		goto fail;
257	} else {
258		hslot = udp_hashslot(udptable, net, snum);
259		spin_lock_bh(&hslot->lock);
260		if (hslot->count > 10) {
261			int exist;
262			unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
263
264			slot2          &= udptable->mask;
265			hash2_nulladdr &= udptable->mask;
266
267			hslot2 = udp_hashslot2(udptable, slot2);
268			if (hslot->count < hslot2->count)
269				goto scan_primary_hash;
270
271			exist = udp_lib_lport_inuse2(net, snum, hslot2,
272						     sk, saddr_comp);
273			if (!exist && (hash2_nulladdr != slot2)) {
274				hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
275				exist = udp_lib_lport_inuse2(net, snum, hslot2,
276							     sk, saddr_comp);
277			}
278			if (exist)
279				goto fail_unlock;
280			else
281				goto found;
282		}
283scan_primary_hash:
284		if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
285					saddr_comp, 0))
286			goto fail_unlock;
287	}
288found:
289	inet_sk(sk)->inet_num = snum;
290	udp_sk(sk)->udp_port_hash = snum;
291	udp_sk(sk)->udp_portaddr_hash ^= snum;
292	if (sk_unhashed(sk)) {
293		sk_nulls_add_node_rcu(sk, &hslot->head);
294		hslot->count++;
295		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
296
297		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
298		spin_lock(&hslot2->lock);
299		hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
300					 &hslot2->head);
301		hslot2->count++;
302		spin_unlock(&hslot2->lock);
303	}
304	error = 0;
305fail_unlock:
306	spin_unlock_bh(&hslot->lock);
307fail:
308	return error;
309}
310EXPORT_SYMBOL(udp_lib_get_port);
311
312static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
313{
314	struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
315
316	return 	(!ipv6_only_sock(sk2)  &&
317		 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
318		   inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
319}
320
321static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
322				       unsigned int port)
323{
324	return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
325}
326
327int udp_v4_get_port(struct sock *sk, unsigned short snum)
328{
329	unsigned int hash2_nulladdr =
330		udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
331	unsigned int hash2_partial =
332		udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
333
334	/* precompute partial secondary hash */
335	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
336	return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
337}
338
339static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
340			 unsigned short hnum,
341			 __be16 sport, __be32 daddr, __be16 dport, int dif)
342{
343	int score = -1;
344
345	if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
346			!ipv6_only_sock(sk)) {
347		struct inet_sock *inet = inet_sk(sk);
348
349		score = (sk->sk_family == PF_INET ? 2 : 1);
350		if (inet->inet_rcv_saddr) {
351			if (inet->inet_rcv_saddr != daddr)
352				return -1;
353			score += 4;
354		}
355		if (inet->inet_daddr) {
356			if (inet->inet_daddr != saddr)
357				return -1;
358			score += 4;
359		}
360		if (inet->inet_dport) {
361			if (inet->inet_dport != sport)
362				return -1;
363			score += 4;
364		}
365		if (sk->sk_bound_dev_if) {
366			if (sk->sk_bound_dev_if != dif)
367				return -1;
368			score += 4;
369		}
370	}
371	return score;
372}
373
374/*
375 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
376 */
377static inline int compute_score2(struct sock *sk, struct net *net,
378				 __be32 saddr, __be16 sport,
379				 __be32 daddr, unsigned int hnum, int dif)
380{
381	int score = -1;
382
383	if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
384		struct inet_sock *inet = inet_sk(sk);
385
386		if (inet->inet_rcv_saddr != daddr)
387			return -1;
388		if (inet->inet_num != hnum)
389			return -1;
390
391		score = (sk->sk_family == PF_INET ? 2 : 1);
392		if (inet->inet_daddr) {
393			if (inet->inet_daddr != saddr)
394				return -1;
395			score += 4;
396		}
397		if (inet->inet_dport) {
398			if (inet->inet_dport != sport)
399				return -1;
400			score += 4;
401		}
402		if (sk->sk_bound_dev_if) {
403			if (sk->sk_bound_dev_if != dif)
404				return -1;
405			score += 4;
406		}
407	}
408	return score;
409}
410
411static unsigned int udp_ehashfn(struct net *net, const __be32 laddr,
412				 const __u16 lport, const __be32 faddr,
413				 const __be16 fport)
414{
415	static u32 udp_ehash_secret __read_mostly;
416
417	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
418
419	return __inet_ehashfn(laddr, lport, faddr, fport,
420			      udp_ehash_secret + net_hash_mix(net));
421}
422
423
424/* called with read_rcu_lock() */
425static struct sock *udp4_lib_lookup2(struct net *net,
426		__be32 saddr, __be16 sport,
427		__be32 daddr, unsigned int hnum, int dif,
428		struct udp_hslot *hslot2, unsigned int slot2)
429{
430	struct sock *sk, *result;
431	struct hlist_nulls_node *node;
432	int score, badness, matches = 0, reuseport = 0;
433	u32 hash = 0;
434
435begin:
436	result = NULL;
437	badness = 0;
438	udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
439		score = compute_score2(sk, net, saddr, sport,
440				      daddr, hnum, dif);
441		if (score > badness) {
442			result = sk;
443			badness = score;
444			reuseport = sk->sk_reuseport;
445			if (reuseport) {
446				hash = udp_ehashfn(net, daddr, hnum,
447						   saddr, sport);
448				matches = 1;
449			}
450		} else if (score == badness && reuseport) {
451			matches++;
452			if (reciprocal_scale(hash, matches) == 0)
453				result = sk;
454			hash = next_pseudo_random32(hash);
455		}
456	}
457	/*
458	 * if the nulls value we got at the end of this lookup is
459	 * not the expected one, we must restart lookup.
460	 * We probably met an item that was moved to another chain.
461	 */
462	if (get_nulls_value(node) != slot2)
463		goto begin;
464	if (result) {
465		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
466			result = NULL;
467		else if (unlikely(compute_score2(result, net, saddr, sport,
468				  daddr, hnum, dif) < badness)) {
469			sock_put(result);
470			goto begin;
471		}
472	}
473	return result;
474}
475
476/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
477 * harder than this. -DaveM
478 */
479struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
480		__be16 sport, __be32 daddr, __be16 dport,
481		int dif, struct udp_table *udptable)
482{
483	struct sock *sk, *result;
484	struct hlist_nulls_node *node;
485	unsigned short hnum = ntohs(dport);
486	unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
487	struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
488	int score, badness, matches = 0, reuseport = 0;
489	u32 hash = 0;
490
491	rcu_read_lock();
492	if (hslot->count > 10) {
493		hash2 = udp4_portaddr_hash(net, daddr, hnum);
494		slot2 = hash2 & udptable->mask;
495		hslot2 = &udptable->hash2[slot2];
496		if (hslot->count < hslot2->count)
497			goto begin;
498
499		result = udp4_lib_lookup2(net, saddr, sport,
500					  daddr, hnum, dif,
501					  hslot2, slot2);
502		if (!result) {
503			hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
504			slot2 = hash2 & udptable->mask;
505			hslot2 = &udptable->hash2[slot2];
506			if (hslot->count < hslot2->count)
507				goto begin;
508
509			result = udp4_lib_lookup2(net, saddr, sport,
510						  htonl(INADDR_ANY), hnum, dif,
511						  hslot2, slot2);
512		}
513		rcu_read_unlock();
514		return result;
515	}
516begin:
517	result = NULL;
518	badness = 0;
519	sk_nulls_for_each_rcu(sk, node, &hslot->head) {
520		score = compute_score(sk, net, saddr, hnum, sport,
521				      daddr, dport, dif);
522		if (score > badness) {
523			result = sk;
524			badness = score;
525			reuseport = sk->sk_reuseport;
526			if (reuseport) {
527				hash = udp_ehashfn(net, daddr, hnum,
528						   saddr, sport);
529				matches = 1;
530			}
531		} else if (score == badness && reuseport) {
532			matches++;
533			if (reciprocal_scale(hash, matches) == 0)
534				result = sk;
535			hash = next_pseudo_random32(hash);
536		}
537	}
538	/*
539	 * if the nulls value we got at the end of this lookup is
540	 * not the expected one, we must restart lookup.
541	 * We probably met an item that was moved to another chain.
542	 */
543	if (get_nulls_value(node) != slot)
544		goto begin;
545
546	if (result) {
547		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
548			result = NULL;
549		else if (unlikely(compute_score(result, net, saddr, hnum, sport,
550				  daddr, dport, dif) < badness)) {
551			sock_put(result);
552			goto begin;
553		}
554	}
555	rcu_read_unlock();
556	return result;
557}
558EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
559
560static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
561						 __be16 sport, __be16 dport,
562						 struct udp_table *udptable)
563{
564	const struct iphdr *iph = ip_hdr(skb);
565
566	return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
567				 iph->daddr, dport, inet_iif(skb),
568				 udptable);
569}
570
571struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
572			     __be32 daddr, __be16 dport, int dif)
573{
574	return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
575}
576EXPORT_SYMBOL_GPL(udp4_lib_lookup);
577
578static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
579				       __be16 loc_port, __be32 loc_addr,
580				       __be16 rmt_port, __be32 rmt_addr,
581				       int dif, unsigned short hnum)
582{
583	struct inet_sock *inet = inet_sk(sk);
584
585	if (!net_eq(sock_net(sk), net) ||
586	    udp_sk(sk)->udp_port_hash != hnum ||
587	    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
588	    (inet->inet_dport != rmt_port && inet->inet_dport) ||
589	    (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
590	    ipv6_only_sock(sk) ||
591	    (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
592		return false;
593	if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
594		return false;
595	return true;
596}
597
598/*
599 * This routine is called by the ICMP module when it gets some
600 * sort of error condition.  If err < 0 then the socket should
601 * be closed and the error returned to the user.  If err > 0
602 * it's just the icmp type << 8 | icmp code.
603 * Header points to the ip header of the error packet. We move
604 * on past this. Then (as it used to claim before adjustment)
605 * header points to the first 8 bytes of the udp header.  We need
606 * to find the appropriate port.
607 */
608
609void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
610{
611	struct inet_sock *inet;
612	const struct iphdr *iph = (const struct iphdr *)skb->data;
613	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
614	const int type = icmp_hdr(skb)->type;
615	const int code = icmp_hdr(skb)->code;
616	struct sock *sk;
617	int harderr;
618	int err;
619	struct net *net = dev_net(skb->dev);
620
621	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
622			iph->saddr, uh->source, skb->dev->ifindex, udptable);
623	if (sk == NULL) {
624		ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
625		return;	/* No socket for error */
626	}
627
628	err = 0;
629	harderr = 0;
630	inet = inet_sk(sk);
631
632	switch (type) {
633	default:
634	case ICMP_TIME_EXCEEDED:
635		err = EHOSTUNREACH;
636		break;
637	case ICMP_SOURCE_QUENCH:
638		goto out;
639	case ICMP_PARAMETERPROB:
640		err = EPROTO;
641		harderr = 1;
642		break;
643	case ICMP_DEST_UNREACH:
644		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
645			ipv4_sk_update_pmtu(skb, sk, info);
646			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
647				err = EMSGSIZE;
648				harderr = 1;
649				break;
650			}
651			goto out;
652		}
653		err = EHOSTUNREACH;
654		if (code <= NR_ICMP_UNREACH) {
655			harderr = icmp_err_convert[code].fatal;
656			err = icmp_err_convert[code].errno;
657		}
658		break;
659	case ICMP_REDIRECT:
660		ipv4_sk_redirect(skb, sk);
661		goto out;
662	}
663
664	/*
665	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
666	 *	4.1.3.3.
667	 */
668	if (!inet->recverr) {
669		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
670			goto out;
671	} else
672		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
673
674	sk->sk_err = err;
675	sk->sk_error_report(sk);
676out:
677	sock_put(sk);
678}
679
680void udp_err(struct sk_buff *skb, u32 info)
681{
682	__udp4_lib_err(skb, info, &udp_table);
683}
684
685/*
686 * Throw away all pending data and cancel the corking. Socket is locked.
687 */
688void udp_flush_pending_frames(struct sock *sk)
689{
690	struct udp_sock *up = udp_sk(sk);
691
692	if (up->pending) {
693		up->len = 0;
694		up->pending = 0;
695		ip_flush_pending_frames(sk);
696	}
697}
698EXPORT_SYMBOL(udp_flush_pending_frames);
699
700/**
701 * 	udp4_hwcsum  -  handle outgoing HW checksumming
702 * 	@skb: 	sk_buff containing the filled-in UDP header
703 * 	        (checksum field must be zeroed out)
704 *	@src:	source IP address
705 *	@dst:	destination IP address
706 */
707void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
708{
709	struct udphdr *uh = udp_hdr(skb);
710	int offset = skb_transport_offset(skb);
711	int len = skb->len - offset;
712	int hlen = len;
713	__wsum csum = 0;
714
715	if (!skb_has_frag_list(skb)) {
716		/*
717		 * Only one fragment on the socket.
718		 */
719		skb->csum_start = skb_transport_header(skb) - skb->head;
720		skb->csum_offset = offsetof(struct udphdr, check);
721		uh->check = ~csum_tcpudp_magic(src, dst, len,
722					       IPPROTO_UDP, 0);
723	} else {
724		struct sk_buff *frags;
725
726		/*
727		 * HW-checksum won't work as there are two or more
728		 * fragments on the socket so that all csums of sk_buffs
729		 * should be together
730		 */
731		skb_walk_frags(skb, frags) {
732			csum = csum_add(csum, frags->csum);
733			hlen -= frags->len;
734		}
735
736		csum = skb_checksum(skb, offset, hlen, csum);
737		skb->ip_summed = CHECKSUM_NONE;
738
739		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
740		if (uh->check == 0)
741			uh->check = CSUM_MANGLED_0;
742	}
743}
744EXPORT_SYMBOL_GPL(udp4_hwcsum);
745
746/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
747 * for the simple case like when setting the checksum for a UDP tunnel.
748 */
749void udp_set_csum(bool nocheck, struct sk_buff *skb,
750		  __be32 saddr, __be32 daddr, int len)
751{
752	struct udphdr *uh = udp_hdr(skb);
753
754	if (nocheck)
755		uh->check = 0;
756	else if (skb_is_gso(skb))
757		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
758	else if (skb_dst(skb) && skb_dst(skb)->dev &&
759		 (skb_dst(skb)->dev->features & NETIF_F_V4_CSUM)) {
760
761		BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL);
762
763		skb->ip_summed = CHECKSUM_PARTIAL;
764		skb->csum_start = skb_transport_header(skb) - skb->head;
765		skb->csum_offset = offsetof(struct udphdr, check);
766		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
767	} else {
768		__wsum csum;
769
770		BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL);
771
772		uh->check = 0;
773		csum = skb_checksum(skb, 0, len, 0);
774		uh->check = udp_v4_check(len, saddr, daddr, csum);
775		if (uh->check == 0)
776			uh->check = CSUM_MANGLED_0;
777
778		skb->ip_summed = CHECKSUM_UNNECESSARY;
779	}
780}
781EXPORT_SYMBOL(udp_set_csum);
782
783static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
784{
785	struct sock *sk = skb->sk;
786	struct inet_sock *inet = inet_sk(sk);
787	struct udphdr *uh;
788	int err = 0;
789	int is_udplite = IS_UDPLITE(sk);
790	int offset = skb_transport_offset(skb);
791	int len = skb->len - offset;
792	__wsum csum = 0;
793
794	/*
795	 * Create a UDP header
796	 */
797	uh = udp_hdr(skb);
798	uh->source = inet->inet_sport;
799	uh->dest = fl4->fl4_dport;
800	uh->len = htons(len);
801	uh->check = 0;
802
803	if (is_udplite)  				 /*     UDP-Lite      */
804		csum = udplite_csum(skb);
805
806	else if (sk->sk_no_check_tx) {   /* UDP csum disabled */
807
808		skb->ip_summed = CHECKSUM_NONE;
809		goto send;
810
811	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
812
813		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
814		goto send;
815
816	} else
817		csum = udp_csum(skb);
818
819	/* add protocol-dependent pseudo-header */
820	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
821				      sk->sk_protocol, csum);
822	if (uh->check == 0)
823		uh->check = CSUM_MANGLED_0;
824
825send:
826	err = ip_send_skb(sock_net(sk), skb);
827	if (err) {
828		if (err == -ENOBUFS && !inet->recverr) {
829			UDP_INC_STATS_USER(sock_net(sk),
830					   UDP_MIB_SNDBUFERRORS, is_udplite);
831			err = 0;
832		}
833	} else
834		UDP_INC_STATS_USER(sock_net(sk),
835				   UDP_MIB_OUTDATAGRAMS, is_udplite);
836	return err;
837}
838
839/*
840 * Push out all pending data as one UDP datagram. Socket is locked.
841 */
842int udp_push_pending_frames(struct sock *sk)
843{
844	struct udp_sock  *up = udp_sk(sk);
845	struct inet_sock *inet = inet_sk(sk);
846	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
847	struct sk_buff *skb;
848	int err = 0;
849
850	skb = ip_finish_skb(sk, fl4);
851	if (!skb)
852		goto out;
853
854	err = udp_send_skb(skb, fl4);
855
856out:
857	up->len = 0;
858	up->pending = 0;
859	return err;
860}
861EXPORT_SYMBOL(udp_push_pending_frames);
862
863int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
864		size_t len)
865{
866	struct inet_sock *inet = inet_sk(sk);
867	struct udp_sock *up = udp_sk(sk);
868	struct flowi4 fl4_stack;
869	struct flowi4 *fl4;
870	int ulen = len;
871	struct ipcm_cookie ipc;
872	struct rtable *rt = NULL;
873	int free = 0;
874	int connected = 0;
875	__be32 daddr, faddr, saddr;
876	__be16 dport;
877	u8  tos;
878	int err, is_udplite = IS_UDPLITE(sk);
879	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
880	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
881	struct sk_buff *skb;
882	struct ip_options_data opt_copy;
883
884	if (len > 0xFFFF)
885		return -EMSGSIZE;
886
887	/*
888	 *	Check the flags.
889	 */
890
891	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
892		return -EOPNOTSUPP;
893
894	ipc.opt = NULL;
895	ipc.tx_flags = 0;
896	ipc.ttl = 0;
897	ipc.tos = -1;
898
899	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
900
901	fl4 = &inet->cork.fl.u.ip4;
902	if (up->pending) {
903		/*
904		 * There are pending frames.
905		 * The socket lock must be held while it's corked.
906		 */
907		lock_sock(sk);
908		if (likely(up->pending)) {
909			if (unlikely(up->pending != AF_INET)) {
910				release_sock(sk);
911				return -EINVAL;
912			}
913			goto do_append_data;
914		}
915		release_sock(sk);
916	}
917	ulen += sizeof(struct udphdr);
918
919	/*
920	 *	Get and verify the address.
921	 */
922	if (msg->msg_name) {
923		DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
924		if (msg->msg_namelen < sizeof(*usin))
925			return -EINVAL;
926		if (usin->sin_family != AF_INET) {
927			if (usin->sin_family != AF_UNSPEC)
928				return -EAFNOSUPPORT;
929		}
930
931		daddr = usin->sin_addr.s_addr;
932		dport = usin->sin_port;
933		if (dport == 0)
934			return -EINVAL;
935	} else {
936		if (sk->sk_state != TCP_ESTABLISHED)
937			return -EDESTADDRREQ;
938		daddr = inet->inet_daddr;
939		dport = inet->inet_dport;
940		/* Open fast path for connected socket.
941		   Route will not be used, if at least one option is set.
942		 */
943		connected = 1;
944	}
945	ipc.addr = inet->inet_saddr;
946
947	ipc.oif = sk->sk_bound_dev_if;
948
949	sock_tx_timestamp(sk, &ipc.tx_flags);
950
951	if (msg->msg_controllen) {
952		err = ip_cmsg_send(sock_net(sk), msg, &ipc,
953				   sk->sk_family == AF_INET6);
954		if (err)
955			return err;
956		if (ipc.opt)
957			free = 1;
958		connected = 0;
959	}
960	if (!ipc.opt) {
961		struct ip_options_rcu *inet_opt;
962
963		rcu_read_lock();
964		inet_opt = rcu_dereference(inet->inet_opt);
965		if (inet_opt) {
966			memcpy(&opt_copy, inet_opt,
967			       sizeof(*inet_opt) + inet_opt->opt.optlen);
968			ipc.opt = &opt_copy.opt;
969		}
970		rcu_read_unlock();
971	}
972
973	saddr = ipc.addr;
974	ipc.addr = faddr = daddr;
975
976	if (ipc.opt && ipc.opt->opt.srr) {
977		if (!daddr)
978			return -EINVAL;
979		faddr = ipc.opt->opt.faddr;
980		connected = 0;
981	}
982	tos = get_rttos(&ipc, inet);
983	if (sock_flag(sk, SOCK_LOCALROUTE) ||
984	    (msg->msg_flags & MSG_DONTROUTE) ||
985	    (ipc.opt && ipc.opt->opt.is_strictroute)) {
986		tos |= RTO_ONLINK;
987		connected = 0;
988	}
989
990	if (ipv4_is_multicast(daddr)) {
991		if (!ipc.oif)
992			ipc.oif = inet->mc_index;
993		if (!saddr)
994			saddr = inet->mc_addr;
995		connected = 0;
996	} else if (!ipc.oif)
997		ipc.oif = inet->uc_index;
998
999	if (connected)
1000		rt = (struct rtable *)sk_dst_check(sk, 0);
1001
1002	if (rt == NULL) {
1003		struct net *net = sock_net(sk);
1004
1005		fl4 = &fl4_stack;
1006		flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1007				   RT_SCOPE_UNIVERSE, sk->sk_protocol,
1008				   inet_sk_flowi_flags(sk),
1009				   faddr, saddr, dport, inet->inet_sport,
1010				   sock_i_uid(sk));
1011
1012		security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1013		rt = ip_route_output_flow(net, fl4, sk);
1014		if (IS_ERR(rt)) {
1015			err = PTR_ERR(rt);
1016			rt = NULL;
1017			if (err == -ENETUNREACH)
1018				IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1019			goto out;
1020		}
1021
1022		err = -EACCES;
1023		if ((rt->rt_flags & RTCF_BROADCAST) &&
1024		    !sock_flag(sk, SOCK_BROADCAST))
1025			goto out;
1026		if (connected)
1027			sk_dst_set(sk, dst_clone(&rt->dst));
1028	}
1029
1030	if (msg->msg_flags&MSG_CONFIRM)
1031		goto do_confirm;
1032back_from_confirm:
1033
1034	saddr = fl4->saddr;
1035	if (!ipc.addr)
1036		daddr = ipc.addr = fl4->daddr;
1037
1038	/* Lockless fast path for the non-corking case. */
1039	if (!corkreq) {
1040		skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
1041				  sizeof(struct udphdr), &ipc, &rt,
1042				  msg->msg_flags);
1043		err = PTR_ERR(skb);
1044		if (!IS_ERR_OR_NULL(skb))
1045			err = udp_send_skb(skb, fl4);
1046		goto out;
1047	}
1048
1049	lock_sock(sk);
1050	if (unlikely(up->pending)) {
1051		/* The socket is already corked while preparing it. */
1052		/* ... which is an evident application bug. --ANK */
1053		release_sock(sk);
1054
1055		LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
1056		err = -EINVAL;
1057		goto out;
1058	}
1059	/*
1060	 *	Now cork the socket to pend data.
1061	 */
1062	fl4 = &inet->cork.fl.u.ip4;
1063	fl4->daddr = daddr;
1064	fl4->saddr = saddr;
1065	fl4->fl4_dport = dport;
1066	fl4->fl4_sport = inet->inet_sport;
1067	up->pending = AF_INET;
1068
1069do_append_data:
1070	up->len += ulen;
1071	err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1072			     sizeof(struct udphdr), &ipc, &rt,
1073			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1074	if (err)
1075		udp_flush_pending_frames(sk);
1076	else if (!corkreq)
1077		err = udp_push_pending_frames(sk);
1078	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1079		up->pending = 0;
1080	release_sock(sk);
1081
1082out:
1083	ip_rt_put(rt);
1084	if (free)
1085		kfree(ipc.opt);
1086	if (!err)
1087		return len;
1088	/*
1089	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1090	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1091	 * we don't have a good statistic (IpOutDiscards but it can be too many
1092	 * things).  We could add another new stat but at least for now that
1093	 * seems like overkill.
1094	 */
1095	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1096		UDP_INC_STATS_USER(sock_net(sk),
1097				UDP_MIB_SNDBUFERRORS, is_udplite);
1098	}
1099	return err;
1100
1101do_confirm:
1102	dst_confirm(&rt->dst);
1103	if (!(msg->msg_flags&MSG_PROBE) || len)
1104		goto back_from_confirm;
1105	err = 0;
1106	goto out;
1107}
1108EXPORT_SYMBOL(udp_sendmsg);
1109
1110int udp_sendpage(struct sock *sk, struct page *page, int offset,
1111		 size_t size, int flags)
1112{
1113	struct inet_sock *inet = inet_sk(sk);
1114	struct udp_sock *up = udp_sk(sk);
1115	int ret;
1116
1117	if (flags & MSG_SENDPAGE_NOTLAST)
1118		flags |= MSG_MORE;
1119
1120	if (!up->pending) {
1121		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
1122
1123		/* Call udp_sendmsg to specify destination address which
1124		 * sendpage interface can't pass.
1125		 * This will succeed only when the socket is connected.
1126		 */
1127		ret = udp_sendmsg(NULL, sk, &msg, 0);
1128		if (ret < 0)
1129			return ret;
1130	}
1131
1132	lock_sock(sk);
1133
1134	if (unlikely(!up->pending)) {
1135		release_sock(sk);
1136
1137		LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1138		return -EINVAL;
1139	}
1140
1141	ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1142			     page, offset, size, flags);
1143	if (ret == -EOPNOTSUPP) {
1144		release_sock(sk);
1145		return sock_no_sendpage(sk->sk_socket, page, offset,
1146					size, flags);
1147	}
1148	if (ret < 0) {
1149		udp_flush_pending_frames(sk);
1150		goto out;
1151	}
1152
1153	up->len += size;
1154	if (!(up->corkflag || (flags&MSG_MORE)))
1155		ret = udp_push_pending_frames(sk);
1156	if (!ret)
1157		ret = size;
1158out:
1159	release_sock(sk);
1160	return ret;
1161}
1162
1163
1164/**
1165 *	first_packet_length	- return length of first packet in receive queue
1166 *	@sk: socket
1167 *
1168 *	Drops all bad checksum frames, until a valid one is found.
1169 *	Returns the length of found skb, or 0 if none is found.
1170 */
1171static unsigned int first_packet_length(struct sock *sk)
1172{
1173	struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1174	struct sk_buff *skb;
1175	unsigned int res;
1176
1177	__skb_queue_head_init(&list_kill);
1178
1179	spin_lock_bh(&rcvq->lock);
1180	while ((skb = skb_peek(rcvq)) != NULL &&
1181		udp_lib_checksum_complete(skb)) {
1182		UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1183				 IS_UDPLITE(sk));
1184		UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1185				 IS_UDPLITE(sk));
1186		atomic_inc(&sk->sk_drops);
1187		__skb_unlink(skb, rcvq);
1188		__skb_queue_tail(&list_kill, skb);
1189	}
1190	res = skb ? skb->len : 0;
1191	spin_unlock_bh(&rcvq->lock);
1192
1193	if (!skb_queue_empty(&list_kill)) {
1194		bool slow = lock_sock_fast(sk);
1195
1196		__skb_queue_purge(&list_kill);
1197		sk_mem_reclaim_partial(sk);
1198		unlock_sock_fast(sk, slow);
1199	}
1200	return res;
1201}
1202
1203/*
1204 *	IOCTL requests applicable to the UDP protocol
1205 */
1206
1207int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1208{
1209	switch (cmd) {
1210	case SIOCOUTQ:
1211	{
1212		int amount = sk_wmem_alloc_get(sk);
1213
1214		return put_user(amount, (int __user *)arg);
1215	}
1216
1217	case SIOCINQ:
1218	{
1219		unsigned int amount = first_packet_length(sk);
1220
1221		if (amount)
1222			/*
1223			 * We will only return the amount
1224			 * of this packet since that is all
1225			 * that will be read.
1226			 */
1227			amount -= sizeof(struct udphdr);
1228
1229		return put_user(amount, (int __user *)arg);
1230	}
1231
1232	default:
1233		return -ENOIOCTLCMD;
1234	}
1235
1236	return 0;
1237}
1238EXPORT_SYMBOL(udp_ioctl);
1239
1240/*
1241 * 	This should be easy, if there is something there we
1242 * 	return it, otherwise we block.
1243 */
1244
1245int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1246		size_t len, int noblock, int flags, int *addr_len)
1247{
1248	struct inet_sock *inet = inet_sk(sk);
1249	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1250	struct sk_buff *skb;
1251	unsigned int ulen, copied;
1252	int peeked, off = 0;
1253	int err;
1254	int is_udplite = IS_UDPLITE(sk);
1255	bool slow;
1256
1257	if (flags & MSG_ERRQUEUE)
1258		return ip_recv_error(sk, msg, len, addr_len);
1259
1260try_again:
1261	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1262				  &peeked, &off, &err);
1263	if (!skb)
1264		goto out;
1265
1266	ulen = skb->len - sizeof(struct udphdr);
1267	copied = len;
1268	if (copied > ulen)
1269		copied = ulen;
1270	else if (copied < ulen)
1271		msg->msg_flags |= MSG_TRUNC;
1272
1273	/*
1274	 * If checksum is needed at all, try to do it while copying the
1275	 * data.  If the data is truncated, or if we only want a partial
1276	 * coverage checksum (UDP-Lite), do it before the copy.
1277	 */
1278
1279	if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1280		if (udp_lib_checksum_complete(skb))
1281			goto csum_copy_err;
1282	}
1283
1284	if (skb_csum_unnecessary(skb))
1285		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1286					      msg->msg_iov, copied);
1287	else {
1288		err = skb_copy_and_csum_datagram_iovec(skb,
1289						       sizeof(struct udphdr),
1290						       msg->msg_iov);
1291
1292		if (err == -EINVAL)
1293			goto csum_copy_err;
1294	}
1295
1296	if (unlikely(err)) {
1297		trace_kfree_skb(skb, udp_recvmsg);
1298		if (!peeked) {
1299			atomic_inc(&sk->sk_drops);
1300			UDP_INC_STATS_USER(sock_net(sk),
1301					   UDP_MIB_INERRORS, is_udplite);
1302		}
1303		goto out_free;
1304	}
1305
1306	if (!peeked)
1307		UDP_INC_STATS_USER(sock_net(sk),
1308				UDP_MIB_INDATAGRAMS, is_udplite);
1309
1310	sock_recv_ts_and_drops(msg, sk, skb);
1311
1312	/* Copy the address. */
1313	if (sin) {
1314		sin->sin_family = AF_INET;
1315		sin->sin_port = udp_hdr(skb)->source;
1316		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1317		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1318		*addr_len = sizeof(*sin);
1319	}
1320	if (inet->cmsg_flags)
1321		ip_cmsg_recv(msg, skb);
1322
1323	err = copied;
1324	if (flags & MSG_TRUNC)
1325		err = ulen;
1326
1327out_free:
1328	skb_free_datagram_locked(sk, skb);
1329out:
1330	return err;
1331
1332csum_copy_err:
1333	slow = lock_sock_fast(sk);
1334	if (!skb_kill_datagram(sk, skb, flags)) {
1335		UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1336		UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1337	}
1338	unlock_sock_fast(sk, slow);
1339
1340	if (noblock)
1341		return -EAGAIN;
1342
1343	/* starting over for a new packet */
1344	msg->msg_flags &= ~MSG_TRUNC;
1345	goto try_again;
1346}
1347
1348
1349int udp_disconnect(struct sock *sk, int flags)
1350{
1351	struct inet_sock *inet = inet_sk(sk);
1352	/*
1353	 *	1003.1g - break association.
1354	 */
1355
1356	sk->sk_state = TCP_CLOSE;
1357	inet->inet_daddr = 0;
1358	inet->inet_dport = 0;
1359	sock_rps_reset_rxhash(sk);
1360	sk->sk_bound_dev_if = 0;
1361	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1362		inet_reset_saddr(sk);
1363
1364	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1365		sk->sk_prot->unhash(sk);
1366		inet->inet_sport = 0;
1367	}
1368	sk_dst_reset(sk);
1369	return 0;
1370}
1371EXPORT_SYMBOL(udp_disconnect);
1372
1373void udp_lib_unhash(struct sock *sk)
1374{
1375	if (sk_hashed(sk)) {
1376		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1377		struct udp_hslot *hslot, *hslot2;
1378
1379		hslot  = udp_hashslot(udptable, sock_net(sk),
1380				      udp_sk(sk)->udp_port_hash);
1381		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1382
1383		spin_lock_bh(&hslot->lock);
1384		if (sk_nulls_del_node_init_rcu(sk)) {
1385			hslot->count--;
1386			inet_sk(sk)->inet_num = 0;
1387			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1388
1389			spin_lock(&hslot2->lock);
1390			hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1391			hslot2->count--;
1392			spin_unlock(&hslot2->lock);
1393		}
1394		spin_unlock_bh(&hslot->lock);
1395	}
1396}
1397EXPORT_SYMBOL(udp_lib_unhash);
1398
1399/*
1400 * inet_rcv_saddr was changed, we must rehash secondary hash
1401 */
1402void udp_lib_rehash(struct sock *sk, u16 newhash)
1403{
1404	if (sk_hashed(sk)) {
1405		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1406		struct udp_hslot *hslot, *hslot2, *nhslot2;
1407
1408		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1409		nhslot2 = udp_hashslot2(udptable, newhash);
1410		udp_sk(sk)->udp_portaddr_hash = newhash;
1411		if (hslot2 != nhslot2) {
1412			hslot = udp_hashslot(udptable, sock_net(sk),
1413					     udp_sk(sk)->udp_port_hash);
1414			/* we must lock primary chain too */
1415			spin_lock_bh(&hslot->lock);
1416
1417			spin_lock(&hslot2->lock);
1418			hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1419			hslot2->count--;
1420			spin_unlock(&hslot2->lock);
1421
1422			spin_lock(&nhslot2->lock);
1423			hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1424						 &nhslot2->head);
1425			nhslot2->count++;
1426			spin_unlock(&nhslot2->lock);
1427
1428			spin_unlock_bh(&hslot->lock);
1429		}
1430	}
1431}
1432EXPORT_SYMBOL(udp_lib_rehash);
1433
1434static void udp_v4_rehash(struct sock *sk)
1435{
1436	u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1437					  inet_sk(sk)->inet_rcv_saddr,
1438					  inet_sk(sk)->inet_num);
1439	udp_lib_rehash(sk, new_hash);
1440}
1441
1442static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1443{
1444	int rc;
1445
1446	if (inet_sk(sk)->inet_daddr) {
1447		sock_rps_save_rxhash(sk, skb);
1448		sk_mark_napi_id(sk, skb);
1449	}
1450
1451	rc = sock_queue_rcv_skb(sk, skb);
1452	if (rc < 0) {
1453		int is_udplite = IS_UDPLITE(sk);
1454
1455		/* Note that an ENOMEM error is charged twice */
1456		if (rc == -ENOMEM)
1457			UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1458					 is_udplite);
1459		UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1460		kfree_skb(skb);
1461		trace_udp_fail_queue_rcv_skb(rc, sk);
1462		return -1;
1463	}
1464
1465	return 0;
1466
1467}
1468
1469static struct static_key udp_encap_needed __read_mostly;
1470void udp_encap_enable(void)
1471{
1472	if (!static_key_enabled(&udp_encap_needed))
1473		static_key_slow_inc(&udp_encap_needed);
1474}
1475EXPORT_SYMBOL(udp_encap_enable);
1476
1477/* returns:
1478 *  -1: error
1479 *   0: success
1480 *  >0: "udp encap" protocol resubmission
1481 *
1482 * Note that in the success and error cases, the skb is assumed to
1483 * have either been requeued or freed.
1484 */
1485int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1486{
1487	struct udp_sock *up = udp_sk(sk);
1488	int rc;
1489	int is_udplite = IS_UDPLITE(sk);
1490
1491	/*
1492	 *	Charge it to the socket, dropping if the queue is full.
1493	 */
1494	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1495		goto drop;
1496	nf_reset(skb);
1497
1498	if (static_key_false(&udp_encap_needed) && up->encap_type) {
1499		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1500
1501		/*
1502		 * This is an encapsulation socket so pass the skb to
1503		 * the socket's udp_encap_rcv() hook. Otherwise, just
1504		 * fall through and pass this up the UDP socket.
1505		 * up->encap_rcv() returns the following value:
1506		 * =0 if skb was successfully passed to the encap
1507		 *    handler or was discarded by it.
1508		 * >0 if skb should be passed on to UDP.
1509		 * <0 if skb should be resubmitted as proto -N
1510		 */
1511
1512		/* if we're overly short, let UDP handle it */
1513		encap_rcv = ACCESS_ONCE(up->encap_rcv);
1514		if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1515			int ret;
1516
1517			/* Verify checksum before giving to encap */
1518			if (udp_lib_checksum_complete(skb))
1519				goto csum_error;
1520
1521			ret = encap_rcv(sk, skb);
1522			if (ret <= 0) {
1523				UDP_INC_STATS_BH(sock_net(sk),
1524						 UDP_MIB_INDATAGRAMS,
1525						 is_udplite);
1526				return -ret;
1527			}
1528		}
1529
1530		/* FALLTHROUGH -- it's a UDP Packet */
1531	}
1532
1533	/*
1534	 * 	UDP-Lite specific tests, ignored on UDP sockets
1535	 */
1536	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
1537
1538		/*
1539		 * MIB statistics other than incrementing the error count are
1540		 * disabled for the following two types of errors: these depend
1541		 * on the application settings, not on the functioning of the
1542		 * protocol stack as such.
1543		 *
1544		 * RFC 3828 here recommends (sec 3.3): "There should also be a
1545		 * way ... to ... at least let the receiving application block
1546		 * delivery of packets with coverage values less than a value
1547		 * provided by the application."
1548		 */
1549		if (up->pcrlen == 0) {          /* full coverage was set  */
1550			LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1551				       UDP_SKB_CB(skb)->cscov, skb->len);
1552			goto drop;
1553		}
1554		/* The next case involves violating the min. coverage requested
1555		 * by the receiver. This is subtle: if receiver wants x and x is
1556		 * greater than the buffersize/MTU then receiver will complain
1557		 * that it wants x while sender emits packets of smaller size y.
1558		 * Therefore the above ...()->partial_cov statement is essential.
1559		 */
1560		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
1561			LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1562				       UDP_SKB_CB(skb)->cscov, up->pcrlen);
1563			goto drop;
1564		}
1565	}
1566
1567	if (rcu_access_pointer(sk->sk_filter) &&
1568	    udp_lib_checksum_complete(skb))
1569		goto csum_error;
1570
1571
1572	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
1573		UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1574				 is_udplite);
1575		goto drop;
1576	}
1577
1578	rc = 0;
1579
1580	ipv4_pktinfo_prepare(sk, skb);
1581	bh_lock_sock(sk);
1582	if (!sock_owned_by_user(sk))
1583		rc = __udp_queue_rcv_skb(sk, skb);
1584	else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1585		bh_unlock_sock(sk);
1586		goto drop;
1587	}
1588	bh_unlock_sock(sk);
1589
1590	return rc;
1591
1592csum_error:
1593	UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1594drop:
1595	UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1596	atomic_inc(&sk->sk_drops);
1597	kfree_skb(skb);
1598	return -1;
1599}
1600
1601
1602static void flush_stack(struct sock **stack, unsigned int count,
1603			struct sk_buff *skb, unsigned int final)
1604{
1605	unsigned int i;
1606	struct sk_buff *skb1 = NULL;
1607	struct sock *sk;
1608
1609	for (i = 0; i < count; i++) {
1610		sk = stack[i];
1611		if (likely(skb1 == NULL))
1612			skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1613
1614		if (!skb1) {
1615			atomic_inc(&sk->sk_drops);
1616			UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1617					 IS_UDPLITE(sk));
1618			UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1619					 IS_UDPLITE(sk));
1620		}
1621
1622		if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1623			skb1 = NULL;
1624
1625		sock_put(sk);
1626	}
1627	if (unlikely(skb1))
1628		kfree_skb(skb1);
1629}
1630
1631/* For TCP sockets, sk_rx_dst is protected by socket lock
1632 * For UDP, we use xchg() to guard against concurrent changes.
1633 */
1634static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1635{
1636	struct dst_entry *old;
1637
1638	dst_hold(dst);
1639	old = xchg(&sk->sk_rx_dst, dst);
1640	dst_release(old);
1641}
1642
1643/*
1644 *	Multicasts and broadcasts go to each listener.
1645 *
1646 *	Note: called only from the BH handler context.
1647 */
1648static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1649				    struct udphdr  *uh,
1650				    __be32 saddr, __be32 daddr,
1651				    struct udp_table *udptable)
1652{
1653	struct sock *sk, *stack[256 / sizeof(struct sock *)];
1654	struct hlist_nulls_node *node;
1655	unsigned short hnum = ntohs(uh->dest);
1656	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1657	int dif = skb->dev->ifindex;
1658	unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
1659	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1660
1661	if (use_hash2) {
1662		hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1663			    udp_table.mask;
1664		hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask;
1665start_lookup:
1666		hslot = &udp_table.hash2[hash2];
1667		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1668	}
1669
1670	spin_lock(&hslot->lock);
1671	sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
1672		if (__udp_is_mcast_sock(net, sk,
1673					uh->dest, daddr,
1674					uh->source, saddr,
1675					dif, hnum)) {
1676			if (unlikely(count == ARRAY_SIZE(stack))) {
1677				flush_stack(stack, count, skb, ~0);
1678				count = 0;
1679			}
1680			stack[count++] = sk;
1681			sock_hold(sk);
1682		}
1683	}
1684
1685	spin_unlock(&hslot->lock);
1686
1687	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
1688	if (use_hash2 && hash2 != hash2_any) {
1689		hash2 = hash2_any;
1690		goto start_lookup;
1691	}
1692
1693	/*
1694	 * do the slow work with no lock held
1695	 */
1696	if (count) {
1697		flush_stack(stack, count, skb, count - 1);
1698	} else {
1699		kfree_skb(skb);
1700	}
1701	return 0;
1702}
1703
1704/* Initialize UDP checksum. If exited with zero value (success),
1705 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1706 * Otherwise, csum completion requires chacksumming packet body,
1707 * including udp header and folding it to skb->csum.
1708 */
1709static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1710				 int proto)
1711{
1712	int err;
1713
1714	UDP_SKB_CB(skb)->partial_cov = 0;
1715	UDP_SKB_CB(skb)->cscov = skb->len;
1716
1717	if (proto == IPPROTO_UDPLITE) {
1718		err = udplite_checksum_init(skb, uh);
1719		if (err)
1720			return err;
1721	}
1722
1723	return skb_checksum_init_zero_check(skb, proto, uh->check,
1724					    inet_compute_pseudo);
1725}
1726
1727/*
1728 *	All we need to do is get the socket, and then do a checksum.
1729 */
1730
1731int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1732		   int proto)
1733{
1734	struct sock *sk;
1735	struct udphdr *uh;
1736	unsigned short ulen;
1737	struct rtable *rt = skb_rtable(skb);
1738	__be32 saddr, daddr;
1739	struct net *net = dev_net(skb->dev);
1740
1741	/*
1742	 *  Validate the packet.
1743	 */
1744	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1745		goto drop;		/* No space for header. */
1746
1747	uh   = udp_hdr(skb);
1748	ulen = ntohs(uh->len);
1749	saddr = ip_hdr(skb)->saddr;
1750	daddr = ip_hdr(skb)->daddr;
1751
1752	if (ulen > skb->len)
1753		goto short_packet;
1754
1755	if (proto == IPPROTO_UDP) {
1756		/* UDP validates ulen. */
1757		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1758			goto short_packet;
1759		uh = udp_hdr(skb);
1760	}
1761
1762	if (udp4_csum_init(skb, uh, proto))
1763		goto csum_error;
1764
1765	sk = skb_steal_sock(skb);
1766	if (sk) {
1767		struct dst_entry *dst = skb_dst(skb);
1768		int ret;
1769
1770		if (unlikely(sk->sk_rx_dst != dst))
1771			udp_sk_rx_dst_set(sk, dst);
1772
1773		ret = udp_queue_rcv_skb(sk, skb);
1774		sock_put(sk);
1775		/* a return value > 0 means to resubmit the input, but
1776		 * it wants the return to be -protocol, or 0
1777		 */
1778		if (ret > 0)
1779			return -ret;
1780		return 0;
1781	} else {
1782		if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1783			return __udp4_lib_mcast_deliver(net, skb, uh,
1784					saddr, daddr, udptable);
1785
1786		sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1787	}
1788
1789	if (sk != NULL) {
1790		int ret;
1791
1792		if (udp_sk(sk)->convert_csum && uh->check && !IS_UDPLITE(sk))
1793			skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1794						 inet_compute_pseudo);
1795
1796		ret = udp_queue_rcv_skb(sk, skb);
1797		sock_put(sk);
1798
1799		/* a return value > 0 means to resubmit the input, but
1800		 * it wants the return to be -protocol, or 0
1801		 */
1802		if (ret > 0)
1803			return -ret;
1804		return 0;
1805	}
1806
1807	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1808		goto drop;
1809	nf_reset(skb);
1810
1811	/* No socket. Drop packet silently, if checksum is wrong */
1812	if (udp_lib_checksum_complete(skb))
1813		goto csum_error;
1814
1815	UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1816	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1817
1818	/*
1819	 * Hmm.  We got an UDP packet to a port to which we
1820	 * don't wanna listen.  Ignore it.
1821	 */
1822	kfree_skb(skb);
1823	return 0;
1824
1825short_packet:
1826	LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1827		       proto == IPPROTO_UDPLITE ? "Lite" : "",
1828		       &saddr, ntohs(uh->source),
1829		       ulen, skb->len,
1830		       &daddr, ntohs(uh->dest));
1831	goto drop;
1832
1833csum_error:
1834	/*
1835	 * RFC1122: OK.  Discards the bad packet silently (as far as
1836	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1837	 */
1838	LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1839		       proto == IPPROTO_UDPLITE ? "Lite" : "",
1840		       &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1841		       ulen);
1842	UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1843drop:
1844	UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1845	kfree_skb(skb);
1846	return 0;
1847}
1848
1849/* We can only early demux multicast if there is a single matching socket.
1850 * If more than one socket found returns NULL
1851 */
1852static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1853						  __be16 loc_port, __be32 loc_addr,
1854						  __be16 rmt_port, __be32 rmt_addr,
1855						  int dif)
1856{
1857	struct sock *sk, *result;
1858	struct hlist_nulls_node *node;
1859	unsigned short hnum = ntohs(loc_port);
1860	unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask);
1861	struct udp_hslot *hslot = &udp_table.hash[slot];
1862
1863	/* Do not bother scanning a too big list */
1864	if (hslot->count > 10)
1865		return NULL;
1866
1867	rcu_read_lock();
1868begin:
1869	count = 0;
1870	result = NULL;
1871	sk_nulls_for_each_rcu(sk, node, &hslot->head) {
1872		if (__udp_is_mcast_sock(net, sk,
1873					loc_port, loc_addr,
1874					rmt_port, rmt_addr,
1875					dif, hnum)) {
1876			result = sk;
1877			++count;
1878		}
1879	}
1880	/*
1881	 * if the nulls value we got at the end of this lookup is
1882	 * not the expected one, we must restart lookup.
1883	 * We probably met an item that was moved to another chain.
1884	 */
1885	if (get_nulls_value(node) != slot)
1886		goto begin;
1887
1888	if (result) {
1889		if (count != 1 ||
1890		    unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1891			result = NULL;
1892		else if (unlikely(!__udp_is_mcast_sock(net, result,
1893						       loc_port, loc_addr,
1894						       rmt_port, rmt_addr,
1895						       dif, hnum))) {
1896			sock_put(result);
1897			result = NULL;
1898		}
1899	}
1900	rcu_read_unlock();
1901	return result;
1902}
1903
1904/* For unicast we should only early demux connected sockets or we can
1905 * break forwarding setups.  The chains here can be long so only check
1906 * if the first socket is an exact match and if not move on.
1907 */
1908static struct sock *__udp4_lib_demux_lookup(struct net *net,
1909					    __be16 loc_port, __be32 loc_addr,
1910					    __be16 rmt_port, __be32 rmt_addr,
1911					    int dif)
1912{
1913	struct sock *sk, *result;
1914	struct hlist_nulls_node *node;
1915	unsigned short hnum = ntohs(loc_port);
1916	unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
1917	unsigned int slot2 = hash2 & udp_table.mask;
1918	struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1919	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
1920	const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1921
1922	rcu_read_lock();
1923	result = NULL;
1924	udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
1925		if (INET_MATCH(sk, net, acookie,
1926			       rmt_addr, loc_addr, ports, dif))
1927			result = sk;
1928		/* Only check first socket in chain */
1929		break;
1930	}
1931
1932	if (result) {
1933		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1934			result = NULL;
1935		else if (unlikely(!INET_MATCH(sk, net, acookie,
1936					      rmt_addr, loc_addr,
1937					      ports, dif))) {
1938			sock_put(result);
1939			result = NULL;
1940		}
1941	}
1942	rcu_read_unlock();
1943	return result;
1944}
1945
1946void udp_v4_early_demux(struct sk_buff *skb)
1947{
1948	struct net *net = dev_net(skb->dev);
1949	const struct iphdr *iph;
1950	const struct udphdr *uh;
1951	struct sock *sk;
1952	struct dst_entry *dst;
1953	int dif = skb->dev->ifindex;
1954
1955	/* validate the packet */
1956	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
1957		return;
1958
1959	iph = ip_hdr(skb);
1960	uh = udp_hdr(skb);
1961
1962	if (skb->pkt_type == PACKET_BROADCAST ||
1963	    skb->pkt_type == PACKET_MULTICAST)
1964		sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
1965						   uh->source, iph->saddr, dif);
1966	else if (skb->pkt_type == PACKET_HOST)
1967		sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
1968					     uh->source, iph->saddr, dif);
1969	else
1970		return;
1971
1972	if (!sk)
1973		return;
1974
1975	skb->sk = sk;
1976	skb->destructor = sock_efree;
1977	dst = sk->sk_rx_dst;
1978
1979	if (dst)
1980		dst = dst_check(dst, 0);
1981	if (dst)
1982		skb_dst_set_noref(skb, dst);
1983}
1984
1985int udp_rcv(struct sk_buff *skb)
1986{
1987	return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1988}
1989
1990void udp_destroy_sock(struct sock *sk)
1991{
1992	struct udp_sock *up = udp_sk(sk);
1993	bool slow = lock_sock_fast(sk);
1994	udp_flush_pending_frames(sk);
1995	unlock_sock_fast(sk, slow);
1996	if (static_key_false(&udp_encap_needed) && up->encap_type) {
1997		void (*encap_destroy)(struct sock *sk);
1998		encap_destroy = ACCESS_ONCE(up->encap_destroy);
1999		if (encap_destroy)
2000			encap_destroy(sk);
2001	}
2002}
2003
2004/*
2005 *	Socket option code for UDP
2006 */
2007int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2008		       char __user *optval, unsigned int optlen,
2009		       int (*push_pending_frames)(struct sock *))
2010{
2011	struct udp_sock *up = udp_sk(sk);
2012	int val, valbool;
2013	int err = 0;
2014	int is_udplite = IS_UDPLITE(sk);
2015
2016	if (optlen < sizeof(int))
2017		return -EINVAL;
2018
2019	if (get_user(val, (int __user *)optval))
2020		return -EFAULT;
2021
2022	valbool = val ? 1 : 0;
2023
2024	switch (optname) {
2025	case UDP_CORK:
2026		if (val != 0) {
2027			up->corkflag = 1;
2028		} else {
2029			up->corkflag = 0;
2030			lock_sock(sk);
2031			(*push_pending_frames)(sk);
2032			release_sock(sk);
2033		}
2034		break;
2035
2036	case UDP_ENCAP:
2037		switch (val) {
2038		case 0:
2039		case UDP_ENCAP_ESPINUDP:
2040		case UDP_ENCAP_ESPINUDP_NON_IKE:
2041			up->encap_rcv = xfrm4_udp_encap_rcv;
2042			/* FALLTHROUGH */
2043		case UDP_ENCAP_L2TPINUDP:
2044			up->encap_type = val;
2045			udp_encap_enable();
2046			break;
2047		default:
2048			err = -ENOPROTOOPT;
2049			break;
2050		}
2051		break;
2052
2053	case UDP_NO_CHECK6_TX:
2054		up->no_check6_tx = valbool;
2055		break;
2056
2057	case UDP_NO_CHECK6_RX:
2058		up->no_check6_rx = valbool;
2059		break;
2060
2061	/*
2062	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
2063	 */
2064	/* The sender sets actual checksum coverage length via this option.
2065	 * The case coverage > packet length is handled by send module. */
2066	case UDPLITE_SEND_CSCOV:
2067		if (!is_udplite)         /* Disable the option on UDP sockets */
2068			return -ENOPROTOOPT;
2069		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2070			val = 8;
2071		else if (val > USHRT_MAX)
2072			val = USHRT_MAX;
2073		up->pcslen = val;
2074		up->pcflag |= UDPLITE_SEND_CC;
2075		break;
2076
2077	/* The receiver specifies a minimum checksum coverage value. To make
2078	 * sense, this should be set to at least 8 (as done below). If zero is
2079	 * used, this again means full checksum coverage.                     */
2080	case UDPLITE_RECV_CSCOV:
2081		if (!is_udplite)         /* Disable the option on UDP sockets */
2082			return -ENOPROTOOPT;
2083		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2084			val = 8;
2085		else if (val > USHRT_MAX)
2086			val = USHRT_MAX;
2087		up->pcrlen = val;
2088		up->pcflag |= UDPLITE_RECV_CC;
2089		break;
2090
2091	default:
2092		err = -ENOPROTOOPT;
2093		break;
2094	}
2095
2096	return err;
2097}
2098EXPORT_SYMBOL(udp_lib_setsockopt);
2099
2100int udp_setsockopt(struct sock *sk, int level, int optname,
2101		   char __user *optval, unsigned int optlen)
2102{
2103	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2104		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2105					  udp_push_pending_frames);
2106	return ip_setsockopt(sk, level, optname, optval, optlen);
2107}
2108
2109#ifdef CONFIG_COMPAT
2110int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2111			  char __user *optval, unsigned int optlen)
2112{
2113	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2114		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2115					  udp_push_pending_frames);
2116	return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2117}
2118#endif
2119
2120int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2121		       char __user *optval, int __user *optlen)
2122{
2123	struct udp_sock *up = udp_sk(sk);
2124	int val, len;
2125
2126	if (get_user(len, optlen))
2127		return -EFAULT;
2128
2129	len = min_t(unsigned int, len, sizeof(int));
2130
2131	if (len < 0)
2132		return -EINVAL;
2133
2134	switch (optname) {
2135	case UDP_CORK:
2136		val = up->corkflag;
2137		break;
2138
2139	case UDP_ENCAP:
2140		val = up->encap_type;
2141		break;
2142
2143	case UDP_NO_CHECK6_TX:
2144		val = up->no_check6_tx;
2145		break;
2146
2147	case UDP_NO_CHECK6_RX:
2148		val = up->no_check6_rx;
2149		break;
2150
2151	/* The following two cannot be changed on UDP sockets, the return is
2152	 * always 0 (which corresponds to the full checksum coverage of UDP). */
2153	case UDPLITE_SEND_CSCOV:
2154		val = up->pcslen;
2155		break;
2156
2157	case UDPLITE_RECV_CSCOV:
2158		val = up->pcrlen;
2159		break;
2160
2161	default:
2162		return -ENOPROTOOPT;
2163	}
2164
2165	if (put_user(len, optlen))
2166		return -EFAULT;
2167	if (copy_to_user(optval, &val, len))
2168		return -EFAULT;
2169	return 0;
2170}
2171EXPORT_SYMBOL(udp_lib_getsockopt);
2172
2173int udp_getsockopt(struct sock *sk, int level, int optname,
2174		   char __user *optval, int __user *optlen)
2175{
2176	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2177		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2178	return ip_getsockopt(sk, level, optname, optval, optlen);
2179}
2180
2181#ifdef CONFIG_COMPAT
2182int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2183				 char __user *optval, int __user *optlen)
2184{
2185	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2186		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2187	return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2188}
2189#endif
2190/**
2191 * 	udp_poll - wait for a UDP event.
2192 *	@file - file struct
2193 *	@sock - socket
2194 *	@wait - poll table
2195 *
2196 *	This is same as datagram poll, except for the special case of
2197 *	blocking sockets. If application is using a blocking fd
2198 *	and a packet with checksum error is in the queue;
2199 *	then it could get return from select indicating data available
2200 *	but then block when reading it. Add special case code
2201 *	to work around these arguably broken applications.
2202 */
2203unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2204{
2205	unsigned int mask = datagram_poll(file, sock, wait);
2206	struct sock *sk = sock->sk;
2207
2208	sock_rps_record_flow(sk);
2209
2210	/* Check for false positives due to checksum errors */
2211	if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2212	    !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2213		mask &= ~(POLLIN | POLLRDNORM);
2214
2215	return mask;
2216
2217}
2218EXPORT_SYMBOL(udp_poll);
2219
2220struct proto udp_prot = {
2221	.name		   = "UDP",
2222	.owner		   = THIS_MODULE,
2223	.close		   = udp_lib_close,
2224	.connect	   = ip4_datagram_connect,
2225	.disconnect	   = udp_disconnect,
2226	.ioctl		   = udp_ioctl,
2227	.destroy	   = udp_destroy_sock,
2228	.setsockopt	   = udp_setsockopt,
2229	.getsockopt	   = udp_getsockopt,
2230	.sendmsg	   = udp_sendmsg,
2231	.recvmsg	   = udp_recvmsg,
2232	.sendpage	   = udp_sendpage,
2233	.backlog_rcv	   = __udp_queue_rcv_skb,
2234	.release_cb	   = ip4_datagram_release_cb,
2235	.hash		   = udp_lib_hash,
2236	.unhash		   = udp_lib_unhash,
2237	.rehash		   = udp_v4_rehash,
2238	.get_port	   = udp_v4_get_port,
2239	.memory_allocated  = &udp_memory_allocated,
2240	.sysctl_mem	   = sysctl_udp_mem,
2241	.sysctl_wmem	   = &sysctl_udp_wmem_min,
2242	.sysctl_rmem	   = &sysctl_udp_rmem_min,
2243	.obj_size	   = sizeof(struct udp_sock),
2244	.slab_flags	   = SLAB_DESTROY_BY_RCU,
2245	.h.udp_table	   = &udp_table,
2246#ifdef CONFIG_COMPAT
2247	.compat_setsockopt = compat_udp_setsockopt,
2248	.compat_getsockopt = compat_udp_getsockopt,
2249#endif
2250	.clear_sk	   = sk_prot_clear_portaddr_nulls,
2251};
2252EXPORT_SYMBOL(udp_prot);
2253
2254/* ------------------------------------------------------------------------ */
2255#ifdef CONFIG_PROC_FS
2256
2257static struct sock *udp_get_first(struct seq_file *seq, int start)
2258{
2259	struct sock *sk;
2260	struct udp_iter_state *state = seq->private;
2261	struct net *net = seq_file_net(seq);
2262
2263	for (state->bucket = start; state->bucket <= state->udp_table->mask;
2264	     ++state->bucket) {
2265		struct hlist_nulls_node *node;
2266		struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2267
2268		if (hlist_nulls_empty(&hslot->head))
2269			continue;
2270
2271		spin_lock_bh(&hslot->lock);
2272		sk_nulls_for_each(sk, node, &hslot->head) {
2273			if (!net_eq(sock_net(sk), net))
2274				continue;
2275			if (sk->sk_family == state->family)
2276				goto found;
2277		}
2278		spin_unlock_bh(&hslot->lock);
2279	}
2280	sk = NULL;
2281found:
2282	return sk;
2283}
2284
2285static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2286{
2287	struct udp_iter_state *state = seq->private;
2288	struct net *net = seq_file_net(seq);
2289
2290	do {
2291		sk = sk_nulls_next(sk);
2292	} while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2293
2294	if (!sk) {
2295		if (state->bucket <= state->udp_table->mask)
2296			spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2297		return udp_get_first(seq, state->bucket + 1);
2298	}
2299	return sk;
2300}
2301
2302static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2303{
2304	struct sock *sk = udp_get_first(seq, 0);
2305
2306	if (sk)
2307		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2308			--pos;
2309	return pos ? NULL : sk;
2310}
2311
2312static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2313{
2314	struct udp_iter_state *state = seq->private;
2315	state->bucket = MAX_UDP_PORTS;
2316
2317	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2318}
2319
2320static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2321{
2322	struct sock *sk;
2323
2324	if (v == SEQ_START_TOKEN)
2325		sk = udp_get_idx(seq, 0);
2326	else
2327		sk = udp_get_next(seq, v);
2328
2329	++*pos;
2330	return sk;
2331}
2332
2333static void udp_seq_stop(struct seq_file *seq, void *v)
2334{
2335	struct udp_iter_state *state = seq->private;
2336
2337	if (state->bucket <= state->udp_table->mask)
2338		spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2339}
2340
2341int udp_seq_open(struct inode *inode, struct file *file)
2342{
2343	struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2344	struct udp_iter_state *s;
2345	int err;
2346
2347	err = seq_open_net(inode, file, &afinfo->seq_ops,
2348			   sizeof(struct udp_iter_state));
2349	if (err < 0)
2350		return err;
2351
2352	s = ((struct seq_file *)file->private_data)->private;
2353	s->family		= afinfo->family;
2354	s->udp_table		= afinfo->udp_table;
2355	return err;
2356}
2357EXPORT_SYMBOL(udp_seq_open);
2358
2359/* ------------------------------------------------------------------------ */
2360int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2361{
2362	struct proc_dir_entry *p;
2363	int rc = 0;
2364
2365	afinfo->seq_ops.start		= udp_seq_start;
2366	afinfo->seq_ops.next		= udp_seq_next;
2367	afinfo->seq_ops.stop		= udp_seq_stop;
2368
2369	p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2370			     afinfo->seq_fops, afinfo);
2371	if (!p)
2372		rc = -ENOMEM;
2373	return rc;
2374}
2375EXPORT_SYMBOL(udp_proc_register);
2376
2377void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2378{
2379	remove_proc_entry(afinfo->name, net->proc_net);
2380}
2381EXPORT_SYMBOL(udp_proc_unregister);
2382
2383/* ------------------------------------------------------------------------ */
2384static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2385		int bucket)
2386{
2387	struct inet_sock *inet = inet_sk(sp);
2388	__be32 dest = inet->inet_daddr;
2389	__be32 src  = inet->inet_rcv_saddr;
2390	__u16 destp	  = ntohs(inet->inet_dport);
2391	__u16 srcp	  = ntohs(inet->inet_sport);
2392
2393	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2394		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2395		bucket, src, srcp, dest, destp, sp->sk_state,
2396		sk_wmem_alloc_get(sp),
2397		sk_rmem_alloc_get(sp),
2398		0, 0L, 0,
2399		from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2400		0, sock_i_ino(sp),
2401		atomic_read(&sp->sk_refcnt), sp,
2402		atomic_read(&sp->sk_drops));
2403}
2404
2405int udp4_seq_show(struct seq_file *seq, void *v)
2406{
2407	seq_setwidth(seq, 127);
2408	if (v == SEQ_START_TOKEN)
2409		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2410			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2411			   "inode ref pointer drops");
2412	else {
2413		struct udp_iter_state *state = seq->private;
2414
2415		udp4_format_sock(v, seq, state->bucket);
2416	}
2417	seq_pad(seq, '\n');
2418	return 0;
2419}
2420
2421static const struct file_operations udp_afinfo_seq_fops = {
2422	.owner    = THIS_MODULE,
2423	.open     = udp_seq_open,
2424	.read     = seq_read,
2425	.llseek   = seq_lseek,
2426	.release  = seq_release_net
2427};
2428
2429/* ------------------------------------------------------------------------ */
2430static struct udp_seq_afinfo udp4_seq_afinfo = {
2431	.name		= "udp",
2432	.family		= AF_INET,
2433	.udp_table	= &udp_table,
2434	.seq_fops	= &udp_afinfo_seq_fops,
2435	.seq_ops	= {
2436		.show		= udp4_seq_show,
2437	},
2438};
2439
2440static int __net_init udp4_proc_init_net(struct net *net)
2441{
2442	return udp_proc_register(net, &udp4_seq_afinfo);
2443}
2444
2445static void __net_exit udp4_proc_exit_net(struct net *net)
2446{
2447	udp_proc_unregister(net, &udp4_seq_afinfo);
2448}
2449
2450static struct pernet_operations udp4_net_ops = {
2451	.init = udp4_proc_init_net,
2452	.exit = udp4_proc_exit_net,
2453};
2454
2455int __init udp4_proc_init(void)
2456{
2457	return register_pernet_subsys(&udp4_net_ops);
2458}
2459
2460void udp4_proc_exit(void)
2461{
2462	unregister_pernet_subsys(&udp4_net_ops);
2463}
2464#endif /* CONFIG_PROC_FS */
2465
2466static __initdata unsigned long uhash_entries;
2467static int __init set_uhash_entries(char *str)
2468{
2469	ssize_t ret;
2470
2471	if (!str)
2472		return 0;
2473
2474	ret = kstrtoul(str, 0, &uhash_entries);
2475	if (ret)
2476		return 0;
2477
2478	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2479		uhash_entries = UDP_HTABLE_SIZE_MIN;
2480	return 1;
2481}
2482__setup("uhash_entries=", set_uhash_entries);
2483
2484void __init udp_table_init(struct udp_table *table, const char *name)
2485{
2486	unsigned int i;
2487
2488	table->hash = alloc_large_system_hash(name,
2489					      2 * sizeof(struct udp_hslot),
2490					      uhash_entries,
2491					      21, /* one slot per 2 MB */
2492					      0,
2493					      &table->log,
2494					      &table->mask,
2495					      UDP_HTABLE_SIZE_MIN,
2496					      64 * 1024);
2497
2498	table->hash2 = table->hash + (table->mask + 1);
2499	for (i = 0; i <= table->mask; i++) {
2500		INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2501		table->hash[i].count = 0;
2502		spin_lock_init(&table->hash[i].lock);
2503	}
2504	for (i = 0; i <= table->mask; i++) {
2505		INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2506		table->hash2[i].count = 0;
2507		spin_lock_init(&table->hash2[i].lock);
2508	}
2509}
2510
2511void __init udp_init(void)
2512{
2513	unsigned long limit;
2514
2515	udp_table_init(&udp_table, "UDP");
2516	limit = nr_free_buffer_pages() / 8;
2517	limit = max(limit, 128UL);
2518	sysctl_udp_mem[0] = limit / 4 * 3;
2519	sysctl_udp_mem[1] = limit;
2520	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2521
2522	sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2523	sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2524}
2525