xref: /linux/net/ipv4/udp.c (revision 800c5eb7b5eba6cb2a32738d763fd59f0fbcdde4)
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 #include <asm/system.h>
81 #include <asm/uaccess.h>
82 #include <asm/ioctls.h>
83 #include <linux/bootmem.h>
84 #include <linux/highmem.h>
85 #include <linux/swap.h>
86 #include <linux/types.h>
87 #include <linux/fcntl.h>
88 #include <linux/module.h>
89 #include <linux/socket.h>
90 #include <linux/sockios.h>
91 #include <linux/igmp.h>
92 #include <linux/in.h>
93 #include <linux/errno.h>
94 #include <linux/timer.h>
95 #include <linux/mm.h>
96 #include <linux/inet.h>
97 #include <linux/netdevice.h>
98 #include <linux/slab.h>
99 #include <net/tcp_states.h>
100 #include <linux/skbuff.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <net/net_namespace.h>
104 #include <net/icmp.h>
105 #include <net/route.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <trace/events/udp.h>
109 #include "udp_impl.h"
110 
111 struct udp_table udp_table __read_mostly;
112 EXPORT_SYMBOL(udp_table);
113 
114 long sysctl_udp_mem[3] __read_mostly;
115 EXPORT_SYMBOL(sysctl_udp_mem);
116 
117 int sysctl_udp_rmem_min __read_mostly;
118 EXPORT_SYMBOL(sysctl_udp_rmem_min);
119 
120 int sysctl_udp_wmem_min __read_mostly;
121 EXPORT_SYMBOL(sysctl_udp_wmem_min);
122 
123 atomic_long_t udp_memory_allocated;
124 EXPORT_SYMBOL(udp_memory_allocated);
125 
126 #define MAX_UDP_PORTS 65536
127 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
128 
129 static int udp_lib_lport_inuse(struct net *net, __u16 num,
130 			       const struct udp_hslot *hslot,
131 			       unsigned long *bitmap,
132 			       struct sock *sk,
133 			       int (*saddr_comp)(const struct sock *sk1,
134 						 const struct sock *sk2),
135 			       unsigned int log)
136 {
137 	struct sock *sk2;
138 	struct hlist_nulls_node *node;
139 
140 	sk_nulls_for_each(sk2, node, &hslot->head)
141 		if (net_eq(sock_net(sk2), net) &&
142 		    sk2 != sk &&
143 		    (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
144 		    (!sk2->sk_reuse || !sk->sk_reuse) &&
145 		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
146 		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
147 		    (*saddr_comp)(sk, sk2)) {
148 			if (bitmap)
149 				__set_bit(udp_sk(sk2)->udp_port_hash >> log,
150 					  bitmap);
151 			else
152 				return 1;
153 		}
154 	return 0;
155 }
156 
157 /*
158  * Note: we still hold spinlock of primary hash chain, so no other writer
159  * can insert/delete a socket with local_port == num
160  */
161 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
162 			       struct udp_hslot *hslot2,
163 			       struct sock *sk,
164 			       int (*saddr_comp)(const struct sock *sk1,
165 						 const struct sock *sk2))
166 {
167 	struct sock *sk2;
168 	struct hlist_nulls_node *node;
169 	int res = 0;
170 
171 	spin_lock(&hslot2->lock);
172 	udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
173 		if (net_eq(sock_net(sk2), net) &&
174 		    sk2 != sk &&
175 		    (udp_sk(sk2)->udp_port_hash == num) &&
176 		    (!sk2->sk_reuse || !sk->sk_reuse) &&
177 		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
178 		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
179 		    (*saddr_comp)(sk, sk2)) {
180 			res = 1;
181 			break;
182 		}
183 	spin_unlock(&hslot2->lock);
184 	return res;
185 }
186 
187 /**
188  *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
189  *
190  *  @sk:          socket struct in question
191  *  @snum:        port number to look up
192  *  @saddr_comp:  AF-dependent comparison of bound local IP addresses
193  *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
194  *                   with NULL address
195  */
196 int udp_lib_get_port(struct sock *sk, unsigned short snum,
197 		       int (*saddr_comp)(const struct sock *sk1,
198 					 const struct sock *sk2),
199 		     unsigned int hash2_nulladdr)
200 {
201 	struct udp_hslot *hslot, *hslot2;
202 	struct udp_table *udptable = sk->sk_prot->h.udp_table;
203 	int    error = 1;
204 	struct net *net = sock_net(sk);
205 
206 	if (!snum) {
207 		int low, high, remaining;
208 		unsigned rand;
209 		unsigned short first, last;
210 		DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
211 
212 		inet_get_local_port_range(&low, &high);
213 		remaining = (high - low) + 1;
214 
215 		rand = net_random();
216 		first = (((u64)rand * remaining) >> 32) + low;
217 		/*
218 		 * force rand to be an odd multiple of UDP_HTABLE_SIZE
219 		 */
220 		rand = (rand | 1) * (udptable->mask + 1);
221 		last = first + udptable->mask + 1;
222 		do {
223 			hslot = udp_hashslot(udptable, net, first);
224 			bitmap_zero(bitmap, PORTS_PER_CHAIN);
225 			spin_lock_bh(&hslot->lock);
226 			udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
227 					    saddr_comp, udptable->log);
228 
229 			snum = first;
230 			/*
231 			 * Iterate on all possible values of snum for this hash.
232 			 * Using steps of an odd multiple of UDP_HTABLE_SIZE
233 			 * give us randomization and full range coverage.
234 			 */
235 			do {
236 				if (low <= snum && snum <= high &&
237 				    !test_bit(snum >> udptable->log, bitmap) &&
238 				    !inet_is_reserved_local_port(snum))
239 					goto found;
240 				snum += rand;
241 			} while (snum != first);
242 			spin_unlock_bh(&hslot->lock);
243 		} while (++first != last);
244 		goto fail;
245 	} else {
246 		hslot = udp_hashslot(udptable, net, snum);
247 		spin_lock_bh(&hslot->lock);
248 		if (hslot->count > 10) {
249 			int exist;
250 			unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
251 
252 			slot2          &= udptable->mask;
253 			hash2_nulladdr &= udptable->mask;
254 
255 			hslot2 = udp_hashslot2(udptable, slot2);
256 			if (hslot->count < hslot2->count)
257 				goto scan_primary_hash;
258 
259 			exist = udp_lib_lport_inuse2(net, snum, hslot2,
260 						     sk, saddr_comp);
261 			if (!exist && (hash2_nulladdr != slot2)) {
262 				hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
263 				exist = udp_lib_lport_inuse2(net, snum, hslot2,
264 							     sk, saddr_comp);
265 			}
266 			if (exist)
267 				goto fail_unlock;
268 			else
269 				goto found;
270 		}
271 scan_primary_hash:
272 		if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
273 					saddr_comp, 0))
274 			goto fail_unlock;
275 	}
276 found:
277 	inet_sk(sk)->inet_num = snum;
278 	udp_sk(sk)->udp_port_hash = snum;
279 	udp_sk(sk)->udp_portaddr_hash ^= snum;
280 	if (sk_unhashed(sk)) {
281 		sk_nulls_add_node_rcu(sk, &hslot->head);
282 		hslot->count++;
283 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
284 
285 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
286 		spin_lock(&hslot2->lock);
287 		hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
288 					 &hslot2->head);
289 		hslot2->count++;
290 		spin_unlock(&hslot2->lock);
291 	}
292 	error = 0;
293 fail_unlock:
294 	spin_unlock_bh(&hslot->lock);
295 fail:
296 	return error;
297 }
298 EXPORT_SYMBOL(udp_lib_get_port);
299 
300 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
301 {
302 	struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
303 
304 	return 	(!ipv6_only_sock(sk2)  &&
305 		 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
306 		   inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
307 }
308 
309 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
310 				       unsigned int port)
311 {
312 	return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
313 }
314 
315 int udp_v4_get_port(struct sock *sk, unsigned short snum)
316 {
317 	unsigned int hash2_nulladdr =
318 		udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
319 	unsigned int hash2_partial =
320 		udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
321 
322 	/* precompute partial secondary hash */
323 	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
324 	return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
325 }
326 
327 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
328 			 unsigned short hnum,
329 			 __be16 sport, __be32 daddr, __be16 dport, int dif)
330 {
331 	int score = -1;
332 
333 	if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
334 			!ipv6_only_sock(sk)) {
335 		struct inet_sock *inet = inet_sk(sk);
336 
337 		score = (sk->sk_family == PF_INET ? 1 : 0);
338 		if (inet->inet_rcv_saddr) {
339 			if (inet->inet_rcv_saddr != daddr)
340 				return -1;
341 			score += 2;
342 		}
343 		if (inet->inet_daddr) {
344 			if (inet->inet_daddr != saddr)
345 				return -1;
346 			score += 2;
347 		}
348 		if (inet->inet_dport) {
349 			if (inet->inet_dport != sport)
350 				return -1;
351 			score += 2;
352 		}
353 		if (sk->sk_bound_dev_if) {
354 			if (sk->sk_bound_dev_if != dif)
355 				return -1;
356 			score += 2;
357 		}
358 	}
359 	return score;
360 }
361 
362 /*
363  * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
364  */
365 #define SCORE2_MAX (1 + 2 + 2 + 2)
366 static inline int compute_score2(struct sock *sk, struct net *net,
367 				 __be32 saddr, __be16 sport,
368 				 __be32 daddr, unsigned int hnum, int dif)
369 {
370 	int score = -1;
371 
372 	if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
373 		struct inet_sock *inet = inet_sk(sk);
374 
375 		if (inet->inet_rcv_saddr != daddr)
376 			return -1;
377 		if (inet->inet_num != hnum)
378 			return -1;
379 
380 		score = (sk->sk_family == PF_INET ? 1 : 0);
381 		if (inet->inet_daddr) {
382 			if (inet->inet_daddr != saddr)
383 				return -1;
384 			score += 2;
385 		}
386 		if (inet->inet_dport) {
387 			if (inet->inet_dport != sport)
388 				return -1;
389 			score += 2;
390 		}
391 		if (sk->sk_bound_dev_if) {
392 			if (sk->sk_bound_dev_if != dif)
393 				return -1;
394 			score += 2;
395 		}
396 	}
397 	return score;
398 }
399 
400 
401 /* called with read_rcu_lock() */
402 static struct sock *udp4_lib_lookup2(struct net *net,
403 		__be32 saddr, __be16 sport,
404 		__be32 daddr, unsigned int hnum, int dif,
405 		struct udp_hslot *hslot2, unsigned int slot2)
406 {
407 	struct sock *sk, *result;
408 	struct hlist_nulls_node *node;
409 	int score, badness;
410 
411 begin:
412 	result = NULL;
413 	badness = -1;
414 	udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
415 		score = compute_score2(sk, net, saddr, sport,
416 				      daddr, hnum, dif);
417 		if (score > badness) {
418 			result = sk;
419 			badness = score;
420 			if (score == SCORE2_MAX)
421 				goto exact_match;
422 		}
423 	}
424 	/*
425 	 * if the nulls value we got at the end of this lookup is
426 	 * not the expected one, we must restart lookup.
427 	 * We probably met an item that was moved to another chain.
428 	 */
429 	if (get_nulls_value(node) != slot2)
430 		goto begin;
431 
432 	if (result) {
433 exact_match:
434 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
435 			result = NULL;
436 		else if (unlikely(compute_score2(result, net, saddr, sport,
437 				  daddr, hnum, dif) < badness)) {
438 			sock_put(result);
439 			goto begin;
440 		}
441 	}
442 	return result;
443 }
444 
445 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
446  * harder than this. -DaveM
447  */
448 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
449 		__be16 sport, __be32 daddr, __be16 dport,
450 		int dif, struct udp_table *udptable)
451 {
452 	struct sock *sk, *result;
453 	struct hlist_nulls_node *node;
454 	unsigned short hnum = ntohs(dport);
455 	unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
456 	struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
457 	int score, badness;
458 
459 	rcu_read_lock();
460 	if (hslot->count > 10) {
461 		hash2 = udp4_portaddr_hash(net, daddr, hnum);
462 		slot2 = hash2 & udptable->mask;
463 		hslot2 = &udptable->hash2[slot2];
464 		if (hslot->count < hslot2->count)
465 			goto begin;
466 
467 		result = udp4_lib_lookup2(net, saddr, sport,
468 					  daddr, hnum, dif,
469 					  hslot2, slot2);
470 		if (!result) {
471 			hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
472 			slot2 = hash2 & udptable->mask;
473 			hslot2 = &udptable->hash2[slot2];
474 			if (hslot->count < hslot2->count)
475 				goto begin;
476 
477 			result = udp4_lib_lookup2(net, saddr, sport,
478 						  htonl(INADDR_ANY), hnum, dif,
479 						  hslot2, slot2);
480 		}
481 		rcu_read_unlock();
482 		return result;
483 	}
484 begin:
485 	result = NULL;
486 	badness = -1;
487 	sk_nulls_for_each_rcu(sk, node, &hslot->head) {
488 		score = compute_score(sk, net, saddr, hnum, sport,
489 				      daddr, dport, dif);
490 		if (score > badness) {
491 			result = sk;
492 			badness = score;
493 		}
494 	}
495 	/*
496 	 * if the nulls value we got at the end of this lookup is
497 	 * not the expected one, we must restart lookup.
498 	 * We probably met an item that was moved to another chain.
499 	 */
500 	if (get_nulls_value(node) != slot)
501 		goto begin;
502 
503 	if (result) {
504 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
505 			result = NULL;
506 		else if (unlikely(compute_score(result, net, saddr, hnum, sport,
507 				  daddr, dport, dif) < badness)) {
508 			sock_put(result);
509 			goto begin;
510 		}
511 	}
512 	rcu_read_unlock();
513 	return result;
514 }
515 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
516 
517 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
518 						 __be16 sport, __be16 dport,
519 						 struct udp_table *udptable)
520 {
521 	struct sock *sk;
522 	const struct iphdr *iph = ip_hdr(skb);
523 
524 	if (unlikely(sk = skb_steal_sock(skb)))
525 		return sk;
526 	else
527 		return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
528 					 iph->daddr, dport, inet_iif(skb),
529 					 udptable);
530 }
531 
532 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
533 			     __be32 daddr, __be16 dport, int dif)
534 {
535 	return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
536 }
537 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
538 
539 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
540 					     __be16 loc_port, __be32 loc_addr,
541 					     __be16 rmt_port, __be32 rmt_addr,
542 					     int dif)
543 {
544 	struct hlist_nulls_node *node;
545 	struct sock *s = sk;
546 	unsigned short hnum = ntohs(loc_port);
547 
548 	sk_nulls_for_each_from(s, node) {
549 		struct inet_sock *inet = inet_sk(s);
550 
551 		if (!net_eq(sock_net(s), net) ||
552 		    udp_sk(s)->udp_port_hash != hnum ||
553 		    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
554 		    (inet->inet_dport != rmt_port && inet->inet_dport) ||
555 		    (inet->inet_rcv_saddr &&
556 		     inet->inet_rcv_saddr != loc_addr) ||
557 		    ipv6_only_sock(s) ||
558 		    (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
559 			continue;
560 		if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
561 			continue;
562 		goto found;
563 	}
564 	s = NULL;
565 found:
566 	return s;
567 }
568 
569 /*
570  * This routine is called by the ICMP module when it gets some
571  * sort of error condition.  If err < 0 then the socket should
572  * be closed and the error returned to the user.  If err > 0
573  * it's just the icmp type << 8 | icmp code.
574  * Header points to the ip header of the error packet. We move
575  * on past this. Then (as it used to claim before adjustment)
576  * header points to the first 8 bytes of the udp header.  We need
577  * to find the appropriate port.
578  */
579 
580 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
581 {
582 	struct inet_sock *inet;
583 	const struct iphdr *iph = (const struct iphdr *)skb->data;
584 	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
585 	const int type = icmp_hdr(skb)->type;
586 	const int code = icmp_hdr(skb)->code;
587 	struct sock *sk;
588 	int harderr;
589 	int err;
590 	struct net *net = dev_net(skb->dev);
591 
592 	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
593 			iph->saddr, uh->source, skb->dev->ifindex, udptable);
594 	if (sk == NULL) {
595 		ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
596 		return;	/* No socket for error */
597 	}
598 
599 	err = 0;
600 	harderr = 0;
601 	inet = inet_sk(sk);
602 
603 	switch (type) {
604 	default:
605 	case ICMP_TIME_EXCEEDED:
606 		err = EHOSTUNREACH;
607 		break;
608 	case ICMP_SOURCE_QUENCH:
609 		goto out;
610 	case ICMP_PARAMETERPROB:
611 		err = EPROTO;
612 		harderr = 1;
613 		break;
614 	case ICMP_DEST_UNREACH:
615 		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
616 			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
617 				err = EMSGSIZE;
618 				harderr = 1;
619 				break;
620 			}
621 			goto out;
622 		}
623 		err = EHOSTUNREACH;
624 		if (code <= NR_ICMP_UNREACH) {
625 			harderr = icmp_err_convert[code].fatal;
626 			err = icmp_err_convert[code].errno;
627 		}
628 		break;
629 	}
630 
631 	/*
632 	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
633 	 *	4.1.3.3.
634 	 */
635 	if (!inet->recverr) {
636 		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
637 			goto out;
638 	} else
639 		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
640 
641 	sk->sk_err = err;
642 	sk->sk_error_report(sk);
643 out:
644 	sock_put(sk);
645 }
646 
647 void udp_err(struct sk_buff *skb, u32 info)
648 {
649 	__udp4_lib_err(skb, info, &udp_table);
650 }
651 
652 /*
653  * Throw away all pending data and cancel the corking. Socket is locked.
654  */
655 void udp_flush_pending_frames(struct sock *sk)
656 {
657 	struct udp_sock *up = udp_sk(sk);
658 
659 	if (up->pending) {
660 		up->len = 0;
661 		up->pending = 0;
662 		ip_flush_pending_frames(sk);
663 	}
664 }
665 EXPORT_SYMBOL(udp_flush_pending_frames);
666 
667 /**
668  * 	udp4_hwcsum  -  handle outgoing HW checksumming
669  * 	@skb: 	sk_buff containing the filled-in UDP header
670  * 	        (checksum field must be zeroed out)
671  *	@src:	source IP address
672  *	@dst:	destination IP address
673  */
674 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
675 {
676 	struct udphdr *uh = udp_hdr(skb);
677 	struct sk_buff *frags = skb_shinfo(skb)->frag_list;
678 	int offset = skb_transport_offset(skb);
679 	int len = skb->len - offset;
680 	int hlen = len;
681 	__wsum csum = 0;
682 
683 	if (!frags) {
684 		/*
685 		 * Only one fragment on the socket.
686 		 */
687 		skb->csum_start = skb_transport_header(skb) - skb->head;
688 		skb->csum_offset = offsetof(struct udphdr, check);
689 		uh->check = ~csum_tcpudp_magic(src, dst, len,
690 					       IPPROTO_UDP, 0);
691 	} else {
692 		/*
693 		 * HW-checksum won't work as there are two or more
694 		 * fragments on the socket so that all csums of sk_buffs
695 		 * should be together
696 		 */
697 		do {
698 			csum = csum_add(csum, frags->csum);
699 			hlen -= frags->len;
700 		} while ((frags = frags->next));
701 
702 		csum = skb_checksum(skb, offset, hlen, csum);
703 		skb->ip_summed = CHECKSUM_NONE;
704 
705 		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
706 		if (uh->check == 0)
707 			uh->check = CSUM_MANGLED_0;
708 	}
709 }
710 
711 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
712 {
713 	struct sock *sk = skb->sk;
714 	struct inet_sock *inet = inet_sk(sk);
715 	struct udphdr *uh;
716 	int err = 0;
717 	int is_udplite = IS_UDPLITE(sk);
718 	int offset = skb_transport_offset(skb);
719 	int len = skb->len - offset;
720 	__wsum csum = 0;
721 
722 	/*
723 	 * Create a UDP header
724 	 */
725 	uh = udp_hdr(skb);
726 	uh->source = inet->inet_sport;
727 	uh->dest = fl4->fl4_dport;
728 	uh->len = htons(len);
729 	uh->check = 0;
730 
731 	if (is_udplite)  				 /*     UDP-Lite      */
732 		csum = udplite_csum(skb);
733 
734 	else if (sk->sk_no_check == UDP_CSUM_NOXMIT) {   /* UDP csum disabled */
735 
736 		skb->ip_summed = CHECKSUM_NONE;
737 		goto send;
738 
739 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
740 
741 		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
742 		goto send;
743 
744 	} else
745 		csum = udp_csum(skb);
746 
747 	/* add protocol-dependent pseudo-header */
748 	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
749 				      sk->sk_protocol, csum);
750 	if (uh->check == 0)
751 		uh->check = CSUM_MANGLED_0;
752 
753 send:
754 	err = ip_send_skb(skb);
755 	if (err) {
756 		if (err == -ENOBUFS && !inet->recverr) {
757 			UDP_INC_STATS_USER(sock_net(sk),
758 					   UDP_MIB_SNDBUFERRORS, is_udplite);
759 			err = 0;
760 		}
761 	} else
762 		UDP_INC_STATS_USER(sock_net(sk),
763 				   UDP_MIB_OUTDATAGRAMS, is_udplite);
764 	return err;
765 }
766 
767 /*
768  * Push out all pending data as one UDP datagram. Socket is locked.
769  */
770 static int udp_push_pending_frames(struct sock *sk)
771 {
772 	struct udp_sock  *up = udp_sk(sk);
773 	struct inet_sock *inet = inet_sk(sk);
774 	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
775 	struct sk_buff *skb;
776 	int err = 0;
777 
778 	skb = ip_finish_skb(sk, fl4);
779 	if (!skb)
780 		goto out;
781 
782 	err = udp_send_skb(skb, fl4);
783 
784 out:
785 	up->len = 0;
786 	up->pending = 0;
787 	return err;
788 }
789 
790 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
791 		size_t len)
792 {
793 	struct inet_sock *inet = inet_sk(sk);
794 	struct udp_sock *up = udp_sk(sk);
795 	struct flowi4 fl4_stack;
796 	struct flowi4 *fl4;
797 	int ulen = len;
798 	struct ipcm_cookie ipc;
799 	struct rtable *rt = NULL;
800 	int free = 0;
801 	int connected = 0;
802 	__be32 daddr, faddr, saddr;
803 	__be16 dport;
804 	u8  tos;
805 	int err, is_udplite = IS_UDPLITE(sk);
806 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
807 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
808 	struct sk_buff *skb;
809 	struct ip_options_data opt_copy;
810 
811 	if (len > 0xFFFF)
812 		return -EMSGSIZE;
813 
814 	/*
815 	 *	Check the flags.
816 	 */
817 
818 	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
819 		return -EOPNOTSUPP;
820 
821 	ipc.opt = NULL;
822 	ipc.tx_flags = 0;
823 
824 	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
825 
826 	fl4 = &inet->cork.fl.u.ip4;
827 	if (up->pending) {
828 		/*
829 		 * There are pending frames.
830 		 * The socket lock must be held while it's corked.
831 		 */
832 		lock_sock(sk);
833 		if (likely(up->pending)) {
834 			if (unlikely(up->pending != AF_INET)) {
835 				release_sock(sk);
836 				return -EINVAL;
837 			}
838 			goto do_append_data;
839 		}
840 		release_sock(sk);
841 	}
842 	ulen += sizeof(struct udphdr);
843 
844 	/*
845 	 *	Get and verify the address.
846 	 */
847 	if (msg->msg_name) {
848 		struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
849 		if (msg->msg_namelen < sizeof(*usin))
850 			return -EINVAL;
851 		if (usin->sin_family != AF_INET) {
852 			if (usin->sin_family != AF_UNSPEC)
853 				return -EAFNOSUPPORT;
854 		}
855 
856 		daddr = usin->sin_addr.s_addr;
857 		dport = usin->sin_port;
858 		if (dport == 0)
859 			return -EINVAL;
860 	} else {
861 		if (sk->sk_state != TCP_ESTABLISHED)
862 			return -EDESTADDRREQ;
863 		daddr = inet->inet_daddr;
864 		dport = inet->inet_dport;
865 		/* Open fast path for connected socket.
866 		   Route will not be used, if at least one option is set.
867 		 */
868 		connected = 1;
869 	}
870 	ipc.addr = inet->inet_saddr;
871 
872 	ipc.oif = sk->sk_bound_dev_if;
873 	err = sock_tx_timestamp(sk, &ipc.tx_flags);
874 	if (err)
875 		return err;
876 	if (msg->msg_controllen) {
877 		err = ip_cmsg_send(sock_net(sk), msg, &ipc);
878 		if (err)
879 			return err;
880 		if (ipc.opt)
881 			free = 1;
882 		connected = 0;
883 	}
884 	if (!ipc.opt) {
885 		struct ip_options_rcu *inet_opt;
886 
887 		rcu_read_lock();
888 		inet_opt = rcu_dereference(inet->inet_opt);
889 		if (inet_opt) {
890 			memcpy(&opt_copy, inet_opt,
891 			       sizeof(*inet_opt) + inet_opt->opt.optlen);
892 			ipc.opt = &opt_copy.opt;
893 		}
894 		rcu_read_unlock();
895 	}
896 
897 	saddr = ipc.addr;
898 	ipc.addr = faddr = daddr;
899 
900 	if (ipc.opt && ipc.opt->opt.srr) {
901 		if (!daddr)
902 			return -EINVAL;
903 		faddr = ipc.opt->opt.faddr;
904 		connected = 0;
905 	}
906 	tos = RT_TOS(inet->tos);
907 	if (sock_flag(sk, SOCK_LOCALROUTE) ||
908 	    (msg->msg_flags & MSG_DONTROUTE) ||
909 	    (ipc.opt && ipc.opt->opt.is_strictroute)) {
910 		tos |= RTO_ONLINK;
911 		connected = 0;
912 	}
913 
914 	if (ipv4_is_multicast(daddr)) {
915 		if (!ipc.oif)
916 			ipc.oif = inet->mc_index;
917 		if (!saddr)
918 			saddr = inet->mc_addr;
919 		connected = 0;
920 	} else if (!ipc.oif)
921 		ipc.oif = inet->uc_index;
922 
923 	if (connected)
924 		rt = (struct rtable *)sk_dst_check(sk, 0);
925 
926 	if (rt == NULL) {
927 		struct net *net = sock_net(sk);
928 
929 		fl4 = &fl4_stack;
930 		flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
931 				   RT_SCOPE_UNIVERSE, sk->sk_protocol,
932 				   inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
933 				   faddr, saddr, dport, inet->inet_sport);
934 
935 		security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
936 		rt = ip_route_output_flow(net, fl4, sk);
937 		if (IS_ERR(rt)) {
938 			err = PTR_ERR(rt);
939 			rt = NULL;
940 			if (err == -ENETUNREACH)
941 				IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
942 			goto out;
943 		}
944 
945 		err = -EACCES;
946 		if ((rt->rt_flags & RTCF_BROADCAST) &&
947 		    !sock_flag(sk, SOCK_BROADCAST))
948 			goto out;
949 		if (connected)
950 			sk_dst_set(sk, dst_clone(&rt->dst));
951 	}
952 
953 	if (msg->msg_flags&MSG_CONFIRM)
954 		goto do_confirm;
955 back_from_confirm:
956 
957 	saddr = fl4->saddr;
958 	if (!ipc.addr)
959 		daddr = ipc.addr = fl4->daddr;
960 
961 	/* Lockless fast path for the non-corking case. */
962 	if (!corkreq) {
963 		skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
964 				  sizeof(struct udphdr), &ipc, &rt,
965 				  msg->msg_flags);
966 		err = PTR_ERR(skb);
967 		if (skb && !IS_ERR(skb))
968 			err = udp_send_skb(skb, fl4);
969 		goto out;
970 	}
971 
972 	lock_sock(sk);
973 	if (unlikely(up->pending)) {
974 		/* The socket is already corked while preparing it. */
975 		/* ... which is an evident application bug. --ANK */
976 		release_sock(sk);
977 
978 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
979 		err = -EINVAL;
980 		goto out;
981 	}
982 	/*
983 	 *	Now cork the socket to pend data.
984 	 */
985 	fl4 = &inet->cork.fl.u.ip4;
986 	fl4->daddr = daddr;
987 	fl4->saddr = saddr;
988 	fl4->fl4_dport = dport;
989 	fl4->fl4_sport = inet->inet_sport;
990 	up->pending = AF_INET;
991 
992 do_append_data:
993 	up->len += ulen;
994 	err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
995 			     sizeof(struct udphdr), &ipc, &rt,
996 			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
997 	if (err)
998 		udp_flush_pending_frames(sk);
999 	else if (!corkreq)
1000 		err = udp_push_pending_frames(sk);
1001 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1002 		up->pending = 0;
1003 	release_sock(sk);
1004 
1005 out:
1006 	ip_rt_put(rt);
1007 	if (free)
1008 		kfree(ipc.opt);
1009 	if (!err)
1010 		return len;
1011 	/*
1012 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1013 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1014 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1015 	 * things).  We could add another new stat but at least for now that
1016 	 * seems like overkill.
1017 	 */
1018 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1019 		UDP_INC_STATS_USER(sock_net(sk),
1020 				UDP_MIB_SNDBUFERRORS, is_udplite);
1021 	}
1022 	return err;
1023 
1024 do_confirm:
1025 	dst_confirm(&rt->dst);
1026 	if (!(msg->msg_flags&MSG_PROBE) || len)
1027 		goto back_from_confirm;
1028 	err = 0;
1029 	goto out;
1030 }
1031 EXPORT_SYMBOL(udp_sendmsg);
1032 
1033 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1034 		 size_t size, int flags)
1035 {
1036 	struct inet_sock *inet = inet_sk(sk);
1037 	struct udp_sock *up = udp_sk(sk);
1038 	int ret;
1039 
1040 	if (!up->pending) {
1041 		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
1042 
1043 		/* Call udp_sendmsg to specify destination address which
1044 		 * sendpage interface can't pass.
1045 		 * This will succeed only when the socket is connected.
1046 		 */
1047 		ret = udp_sendmsg(NULL, sk, &msg, 0);
1048 		if (ret < 0)
1049 			return ret;
1050 	}
1051 
1052 	lock_sock(sk);
1053 
1054 	if (unlikely(!up->pending)) {
1055 		release_sock(sk);
1056 
1057 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
1058 		return -EINVAL;
1059 	}
1060 
1061 	ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1062 			     page, offset, size, flags);
1063 	if (ret == -EOPNOTSUPP) {
1064 		release_sock(sk);
1065 		return sock_no_sendpage(sk->sk_socket, page, offset,
1066 					size, flags);
1067 	}
1068 	if (ret < 0) {
1069 		udp_flush_pending_frames(sk);
1070 		goto out;
1071 	}
1072 
1073 	up->len += size;
1074 	if (!(up->corkflag || (flags&MSG_MORE)))
1075 		ret = udp_push_pending_frames(sk);
1076 	if (!ret)
1077 		ret = size;
1078 out:
1079 	release_sock(sk);
1080 	return ret;
1081 }
1082 
1083 
1084 /**
1085  *	first_packet_length	- return length of first packet in receive queue
1086  *	@sk: socket
1087  *
1088  *	Drops all bad checksum frames, until a valid one is found.
1089  *	Returns the length of found skb, or 0 if none is found.
1090  */
1091 static unsigned int first_packet_length(struct sock *sk)
1092 {
1093 	struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1094 	struct sk_buff *skb;
1095 	unsigned int res;
1096 
1097 	__skb_queue_head_init(&list_kill);
1098 
1099 	spin_lock_bh(&rcvq->lock);
1100 	while ((skb = skb_peek(rcvq)) != NULL &&
1101 		udp_lib_checksum_complete(skb)) {
1102 		UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1103 				 IS_UDPLITE(sk));
1104 		atomic_inc(&sk->sk_drops);
1105 		__skb_unlink(skb, rcvq);
1106 		__skb_queue_tail(&list_kill, skb);
1107 	}
1108 	res = skb ? skb->len : 0;
1109 	spin_unlock_bh(&rcvq->lock);
1110 
1111 	if (!skb_queue_empty(&list_kill)) {
1112 		bool slow = lock_sock_fast(sk);
1113 
1114 		__skb_queue_purge(&list_kill);
1115 		sk_mem_reclaim_partial(sk);
1116 		unlock_sock_fast(sk, slow);
1117 	}
1118 	return res;
1119 }
1120 
1121 /*
1122  *	IOCTL requests applicable to the UDP protocol
1123  */
1124 
1125 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1126 {
1127 	switch (cmd) {
1128 	case SIOCOUTQ:
1129 	{
1130 		int amount = sk_wmem_alloc_get(sk);
1131 
1132 		return put_user(amount, (int __user *)arg);
1133 	}
1134 
1135 	case SIOCINQ:
1136 	{
1137 		unsigned int amount = first_packet_length(sk);
1138 
1139 		if (amount)
1140 			/*
1141 			 * We will only return the amount
1142 			 * of this packet since that is all
1143 			 * that will be read.
1144 			 */
1145 			amount -= sizeof(struct udphdr);
1146 
1147 		return put_user(amount, (int __user *)arg);
1148 	}
1149 
1150 	default:
1151 		return -ENOIOCTLCMD;
1152 	}
1153 
1154 	return 0;
1155 }
1156 EXPORT_SYMBOL(udp_ioctl);
1157 
1158 /*
1159  * 	This should be easy, if there is something there we
1160  * 	return it, otherwise we block.
1161  */
1162 
1163 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1164 		size_t len, int noblock, int flags, int *addr_len)
1165 {
1166 	struct inet_sock *inet = inet_sk(sk);
1167 	struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1168 	struct sk_buff *skb;
1169 	unsigned int ulen, copied;
1170 	int peeked;
1171 	int err;
1172 	int is_udplite = IS_UDPLITE(sk);
1173 	bool slow;
1174 
1175 	/*
1176 	 *	Check any passed addresses
1177 	 */
1178 	if (addr_len)
1179 		*addr_len = sizeof(*sin);
1180 
1181 	if (flags & MSG_ERRQUEUE)
1182 		return ip_recv_error(sk, msg, len);
1183 
1184 try_again:
1185 	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1186 				  &peeked, &err);
1187 	if (!skb)
1188 		goto out;
1189 
1190 	ulen = skb->len - sizeof(struct udphdr);
1191 	copied = len;
1192 	if (copied > ulen)
1193 		copied = ulen;
1194 	else if (copied < ulen)
1195 		msg->msg_flags |= MSG_TRUNC;
1196 
1197 	/*
1198 	 * If checksum is needed at all, try to do it while copying the
1199 	 * data.  If the data is truncated, or if we only want a partial
1200 	 * coverage checksum (UDP-Lite), do it before the copy.
1201 	 */
1202 
1203 	if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1204 		if (udp_lib_checksum_complete(skb))
1205 			goto csum_copy_err;
1206 	}
1207 
1208 	if (skb_csum_unnecessary(skb))
1209 		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1210 					      msg->msg_iov, copied);
1211 	else {
1212 		err = skb_copy_and_csum_datagram_iovec(skb,
1213 						       sizeof(struct udphdr),
1214 						       msg->msg_iov);
1215 
1216 		if (err == -EINVAL)
1217 			goto csum_copy_err;
1218 	}
1219 
1220 	if (err)
1221 		goto out_free;
1222 
1223 	if (!peeked)
1224 		UDP_INC_STATS_USER(sock_net(sk),
1225 				UDP_MIB_INDATAGRAMS, is_udplite);
1226 
1227 	sock_recv_ts_and_drops(msg, sk, skb);
1228 
1229 	/* Copy the address. */
1230 	if (sin) {
1231 		sin->sin_family = AF_INET;
1232 		sin->sin_port = udp_hdr(skb)->source;
1233 		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1234 		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1235 	}
1236 	if (inet->cmsg_flags)
1237 		ip_cmsg_recv(msg, skb);
1238 
1239 	err = copied;
1240 	if (flags & MSG_TRUNC)
1241 		err = ulen;
1242 
1243 out_free:
1244 	skb_free_datagram_locked(sk, skb);
1245 out:
1246 	return err;
1247 
1248 csum_copy_err:
1249 	slow = lock_sock_fast(sk);
1250 	if (!skb_kill_datagram(sk, skb, flags))
1251 		UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1252 	unlock_sock_fast(sk, slow);
1253 
1254 	if (noblock)
1255 		return -EAGAIN;
1256 
1257 	/* starting over for a new packet */
1258 	msg->msg_flags &= ~MSG_TRUNC;
1259 	goto try_again;
1260 }
1261 
1262 
1263 int udp_disconnect(struct sock *sk, int flags)
1264 {
1265 	struct inet_sock *inet = inet_sk(sk);
1266 	/*
1267 	 *	1003.1g - break association.
1268 	 */
1269 
1270 	sk->sk_state = TCP_CLOSE;
1271 	inet->inet_daddr = 0;
1272 	inet->inet_dport = 0;
1273 	sock_rps_reset_rxhash(sk);
1274 	sk->sk_bound_dev_if = 0;
1275 	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1276 		inet_reset_saddr(sk);
1277 
1278 	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1279 		sk->sk_prot->unhash(sk);
1280 		inet->inet_sport = 0;
1281 	}
1282 	sk_dst_reset(sk);
1283 	return 0;
1284 }
1285 EXPORT_SYMBOL(udp_disconnect);
1286 
1287 void udp_lib_unhash(struct sock *sk)
1288 {
1289 	if (sk_hashed(sk)) {
1290 		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1291 		struct udp_hslot *hslot, *hslot2;
1292 
1293 		hslot  = udp_hashslot(udptable, sock_net(sk),
1294 				      udp_sk(sk)->udp_port_hash);
1295 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1296 
1297 		spin_lock_bh(&hslot->lock);
1298 		if (sk_nulls_del_node_init_rcu(sk)) {
1299 			hslot->count--;
1300 			inet_sk(sk)->inet_num = 0;
1301 			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1302 
1303 			spin_lock(&hslot2->lock);
1304 			hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1305 			hslot2->count--;
1306 			spin_unlock(&hslot2->lock);
1307 		}
1308 		spin_unlock_bh(&hslot->lock);
1309 	}
1310 }
1311 EXPORT_SYMBOL(udp_lib_unhash);
1312 
1313 /*
1314  * inet_rcv_saddr was changed, we must rehash secondary hash
1315  */
1316 void udp_lib_rehash(struct sock *sk, u16 newhash)
1317 {
1318 	if (sk_hashed(sk)) {
1319 		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1320 		struct udp_hslot *hslot, *hslot2, *nhslot2;
1321 
1322 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1323 		nhslot2 = udp_hashslot2(udptable, newhash);
1324 		udp_sk(sk)->udp_portaddr_hash = newhash;
1325 		if (hslot2 != nhslot2) {
1326 			hslot = udp_hashslot(udptable, sock_net(sk),
1327 					     udp_sk(sk)->udp_port_hash);
1328 			/* we must lock primary chain too */
1329 			spin_lock_bh(&hslot->lock);
1330 
1331 			spin_lock(&hslot2->lock);
1332 			hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1333 			hslot2->count--;
1334 			spin_unlock(&hslot2->lock);
1335 
1336 			spin_lock(&nhslot2->lock);
1337 			hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1338 						 &nhslot2->head);
1339 			nhslot2->count++;
1340 			spin_unlock(&nhslot2->lock);
1341 
1342 			spin_unlock_bh(&hslot->lock);
1343 		}
1344 	}
1345 }
1346 EXPORT_SYMBOL(udp_lib_rehash);
1347 
1348 static void udp_v4_rehash(struct sock *sk)
1349 {
1350 	u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1351 					  inet_sk(sk)->inet_rcv_saddr,
1352 					  inet_sk(sk)->inet_num);
1353 	udp_lib_rehash(sk, new_hash);
1354 }
1355 
1356 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1357 {
1358 	int rc;
1359 
1360 	if (inet_sk(sk)->inet_daddr)
1361 		sock_rps_save_rxhash(sk, skb);
1362 
1363 	rc = sock_queue_rcv_skb(sk, skb);
1364 	if (rc < 0) {
1365 		int is_udplite = IS_UDPLITE(sk);
1366 
1367 		/* Note that an ENOMEM error is charged twice */
1368 		if (rc == -ENOMEM)
1369 			UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1370 					 is_udplite);
1371 		UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1372 		kfree_skb(skb);
1373 		trace_udp_fail_queue_rcv_skb(rc, sk);
1374 		return -1;
1375 	}
1376 
1377 	return 0;
1378 
1379 }
1380 
1381 /* returns:
1382  *  -1: error
1383  *   0: success
1384  *  >0: "udp encap" protocol resubmission
1385  *
1386  * Note that in the success and error cases, the skb is assumed to
1387  * have either been requeued or freed.
1388  */
1389 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1390 {
1391 	struct udp_sock *up = udp_sk(sk);
1392 	int rc;
1393 	int is_udplite = IS_UDPLITE(sk);
1394 
1395 	/*
1396 	 *	Charge it to the socket, dropping if the queue is full.
1397 	 */
1398 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1399 		goto drop;
1400 	nf_reset(skb);
1401 
1402 	if (up->encap_type) {
1403 		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1404 
1405 		/*
1406 		 * This is an encapsulation socket so pass the skb to
1407 		 * the socket's udp_encap_rcv() hook. Otherwise, just
1408 		 * fall through and pass this up the UDP socket.
1409 		 * up->encap_rcv() returns the following value:
1410 		 * =0 if skb was successfully passed to the encap
1411 		 *    handler or was discarded by it.
1412 		 * >0 if skb should be passed on to UDP.
1413 		 * <0 if skb should be resubmitted as proto -N
1414 		 */
1415 
1416 		/* if we're overly short, let UDP handle it */
1417 		encap_rcv = ACCESS_ONCE(up->encap_rcv);
1418 		if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1419 			int ret;
1420 
1421 			ret = encap_rcv(sk, skb);
1422 			if (ret <= 0) {
1423 				UDP_INC_STATS_BH(sock_net(sk),
1424 						 UDP_MIB_INDATAGRAMS,
1425 						 is_udplite);
1426 				return -ret;
1427 			}
1428 		}
1429 
1430 		/* FALLTHROUGH -- it's a UDP Packet */
1431 	}
1432 
1433 	/*
1434 	 * 	UDP-Lite specific tests, ignored on UDP sockets
1435 	 */
1436 	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
1437 
1438 		/*
1439 		 * MIB statistics other than incrementing the error count are
1440 		 * disabled for the following two types of errors: these depend
1441 		 * on the application settings, not on the functioning of the
1442 		 * protocol stack as such.
1443 		 *
1444 		 * RFC 3828 here recommends (sec 3.3): "There should also be a
1445 		 * way ... to ... at least let the receiving application block
1446 		 * delivery of packets with coverage values less than a value
1447 		 * provided by the application."
1448 		 */
1449 		if (up->pcrlen == 0) {          /* full coverage was set  */
1450 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1451 				"%d while full coverage %d requested\n",
1452 				UDP_SKB_CB(skb)->cscov, skb->len);
1453 			goto drop;
1454 		}
1455 		/* The next case involves violating the min. coverage requested
1456 		 * by the receiver. This is subtle: if receiver wants x and x is
1457 		 * greater than the buffersize/MTU then receiver will complain
1458 		 * that it wants x while sender emits packets of smaller size y.
1459 		 * Therefore the above ...()->partial_cov statement is essential.
1460 		 */
1461 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
1462 			LIMIT_NETDEBUG(KERN_WARNING
1463 				"UDPLITE: coverage %d too small, need min %d\n",
1464 				UDP_SKB_CB(skb)->cscov, up->pcrlen);
1465 			goto drop;
1466 		}
1467 	}
1468 
1469 	if (rcu_access_pointer(sk->sk_filter) &&
1470 	    udp_lib_checksum_complete(skb))
1471 		goto drop;
1472 
1473 
1474 	if (sk_rcvqueues_full(sk, skb))
1475 		goto drop;
1476 
1477 	rc = 0;
1478 
1479 	ipv4_pktinfo_prepare(skb);
1480 	bh_lock_sock(sk);
1481 	if (!sock_owned_by_user(sk))
1482 		rc = __udp_queue_rcv_skb(sk, skb);
1483 	else if (sk_add_backlog(sk, skb)) {
1484 		bh_unlock_sock(sk);
1485 		goto drop;
1486 	}
1487 	bh_unlock_sock(sk);
1488 
1489 	return rc;
1490 
1491 drop:
1492 	UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1493 	atomic_inc(&sk->sk_drops);
1494 	kfree_skb(skb);
1495 	return -1;
1496 }
1497 
1498 
1499 static void flush_stack(struct sock **stack, unsigned int count,
1500 			struct sk_buff *skb, unsigned int final)
1501 {
1502 	unsigned int i;
1503 	struct sk_buff *skb1 = NULL;
1504 	struct sock *sk;
1505 
1506 	for (i = 0; i < count; i++) {
1507 		sk = stack[i];
1508 		if (likely(skb1 == NULL))
1509 			skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1510 
1511 		if (!skb1) {
1512 			atomic_inc(&sk->sk_drops);
1513 			UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1514 					 IS_UDPLITE(sk));
1515 			UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1516 					 IS_UDPLITE(sk));
1517 		}
1518 
1519 		if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1520 			skb1 = NULL;
1521 	}
1522 	if (unlikely(skb1))
1523 		kfree_skb(skb1);
1524 }
1525 
1526 /*
1527  *	Multicasts and broadcasts go to each listener.
1528  *
1529  *	Note: called only from the BH handler context.
1530  */
1531 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1532 				    struct udphdr  *uh,
1533 				    __be32 saddr, __be32 daddr,
1534 				    struct udp_table *udptable)
1535 {
1536 	struct sock *sk, *stack[256 / sizeof(struct sock *)];
1537 	struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1538 	int dif;
1539 	unsigned int i, count = 0;
1540 
1541 	spin_lock(&hslot->lock);
1542 	sk = sk_nulls_head(&hslot->head);
1543 	dif = skb->dev->ifindex;
1544 	sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1545 	while (sk) {
1546 		stack[count++] = sk;
1547 		sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1548 				       daddr, uh->source, saddr, dif);
1549 		if (unlikely(count == ARRAY_SIZE(stack))) {
1550 			if (!sk)
1551 				break;
1552 			flush_stack(stack, count, skb, ~0);
1553 			count = 0;
1554 		}
1555 	}
1556 	/*
1557 	 * before releasing chain lock, we must take a reference on sockets
1558 	 */
1559 	for (i = 0; i < count; i++)
1560 		sock_hold(stack[i]);
1561 
1562 	spin_unlock(&hslot->lock);
1563 
1564 	/*
1565 	 * do the slow work with no lock held
1566 	 */
1567 	if (count) {
1568 		flush_stack(stack, count, skb, count - 1);
1569 
1570 		for (i = 0; i < count; i++)
1571 			sock_put(stack[i]);
1572 	} else {
1573 		kfree_skb(skb);
1574 	}
1575 	return 0;
1576 }
1577 
1578 /* Initialize UDP checksum. If exited with zero value (success),
1579  * CHECKSUM_UNNECESSARY means, that no more checks are required.
1580  * Otherwise, csum completion requires chacksumming packet body,
1581  * including udp header and folding it to skb->csum.
1582  */
1583 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1584 				 int proto)
1585 {
1586 	const struct iphdr *iph;
1587 	int err;
1588 
1589 	UDP_SKB_CB(skb)->partial_cov = 0;
1590 	UDP_SKB_CB(skb)->cscov = skb->len;
1591 
1592 	if (proto == IPPROTO_UDPLITE) {
1593 		err = udplite_checksum_init(skb, uh);
1594 		if (err)
1595 			return err;
1596 	}
1597 
1598 	iph = ip_hdr(skb);
1599 	if (uh->check == 0) {
1600 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1601 	} else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1602 		if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1603 				      proto, skb->csum))
1604 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1605 	}
1606 	if (!skb_csum_unnecessary(skb))
1607 		skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1608 					       skb->len, proto, 0);
1609 	/* Probably, we should checksum udp header (it should be in cache
1610 	 * in any case) and data in tiny packets (< rx copybreak).
1611 	 */
1612 
1613 	return 0;
1614 }
1615 
1616 /*
1617  *	All we need to do is get the socket, and then do a checksum.
1618  */
1619 
1620 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1621 		   int proto)
1622 {
1623 	struct sock *sk;
1624 	struct udphdr *uh;
1625 	unsigned short ulen;
1626 	struct rtable *rt = skb_rtable(skb);
1627 	__be32 saddr, daddr;
1628 	struct net *net = dev_net(skb->dev);
1629 
1630 	/*
1631 	 *  Validate the packet.
1632 	 */
1633 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1634 		goto drop;		/* No space for header. */
1635 
1636 	uh   = udp_hdr(skb);
1637 	ulen = ntohs(uh->len);
1638 	saddr = ip_hdr(skb)->saddr;
1639 	daddr = ip_hdr(skb)->daddr;
1640 
1641 	if (ulen > skb->len)
1642 		goto short_packet;
1643 
1644 	if (proto == IPPROTO_UDP) {
1645 		/* UDP validates ulen. */
1646 		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1647 			goto short_packet;
1648 		uh = udp_hdr(skb);
1649 	}
1650 
1651 	if (udp4_csum_init(skb, uh, proto))
1652 		goto csum_error;
1653 
1654 	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1655 		return __udp4_lib_mcast_deliver(net, skb, uh,
1656 				saddr, daddr, udptable);
1657 
1658 	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1659 
1660 	if (sk != NULL) {
1661 		int ret = udp_queue_rcv_skb(sk, skb);
1662 		sock_put(sk);
1663 
1664 		/* a return value > 0 means to resubmit the input, but
1665 		 * it wants the return to be -protocol, or 0
1666 		 */
1667 		if (ret > 0)
1668 			return -ret;
1669 		return 0;
1670 	}
1671 
1672 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1673 		goto drop;
1674 	nf_reset(skb);
1675 
1676 	/* No socket. Drop packet silently, if checksum is wrong */
1677 	if (udp_lib_checksum_complete(skb))
1678 		goto csum_error;
1679 
1680 	UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1681 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1682 
1683 	/*
1684 	 * Hmm.  We got an UDP packet to a port to which we
1685 	 * don't wanna listen.  Ignore it.
1686 	 */
1687 	kfree_skb(skb);
1688 	return 0;
1689 
1690 short_packet:
1691 	LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1692 		       proto == IPPROTO_UDPLITE ? "-Lite" : "",
1693 		       &saddr,
1694 		       ntohs(uh->source),
1695 		       ulen,
1696 		       skb->len,
1697 		       &daddr,
1698 		       ntohs(uh->dest));
1699 	goto drop;
1700 
1701 csum_error:
1702 	/*
1703 	 * RFC1122: OK.  Discards the bad packet silently (as far as
1704 	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1705 	 */
1706 	LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1707 		       proto == IPPROTO_UDPLITE ? "-Lite" : "",
1708 		       &saddr,
1709 		       ntohs(uh->source),
1710 		       &daddr,
1711 		       ntohs(uh->dest),
1712 		       ulen);
1713 drop:
1714 	UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1715 	kfree_skb(skb);
1716 	return 0;
1717 }
1718 
1719 int udp_rcv(struct sk_buff *skb)
1720 {
1721 	return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1722 }
1723 
1724 void udp_destroy_sock(struct sock *sk)
1725 {
1726 	bool slow = lock_sock_fast(sk);
1727 	udp_flush_pending_frames(sk);
1728 	unlock_sock_fast(sk, slow);
1729 }
1730 
1731 /*
1732  *	Socket option code for UDP
1733  */
1734 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1735 		       char __user *optval, unsigned int optlen,
1736 		       int (*push_pending_frames)(struct sock *))
1737 {
1738 	struct udp_sock *up = udp_sk(sk);
1739 	int val;
1740 	int err = 0;
1741 	int is_udplite = IS_UDPLITE(sk);
1742 
1743 	if (optlen < sizeof(int))
1744 		return -EINVAL;
1745 
1746 	if (get_user(val, (int __user *)optval))
1747 		return -EFAULT;
1748 
1749 	switch (optname) {
1750 	case UDP_CORK:
1751 		if (val != 0) {
1752 			up->corkflag = 1;
1753 		} else {
1754 			up->corkflag = 0;
1755 			lock_sock(sk);
1756 			(*push_pending_frames)(sk);
1757 			release_sock(sk);
1758 		}
1759 		break;
1760 
1761 	case UDP_ENCAP:
1762 		switch (val) {
1763 		case 0:
1764 		case UDP_ENCAP_ESPINUDP:
1765 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1766 			up->encap_rcv = xfrm4_udp_encap_rcv;
1767 			/* FALLTHROUGH */
1768 		case UDP_ENCAP_L2TPINUDP:
1769 			up->encap_type = val;
1770 			break;
1771 		default:
1772 			err = -ENOPROTOOPT;
1773 			break;
1774 		}
1775 		break;
1776 
1777 	/*
1778 	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
1779 	 */
1780 	/* The sender sets actual checksum coverage length via this option.
1781 	 * The case coverage > packet length is handled by send module. */
1782 	case UDPLITE_SEND_CSCOV:
1783 		if (!is_udplite)         /* Disable the option on UDP sockets */
1784 			return -ENOPROTOOPT;
1785 		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1786 			val = 8;
1787 		else if (val > USHRT_MAX)
1788 			val = USHRT_MAX;
1789 		up->pcslen = val;
1790 		up->pcflag |= UDPLITE_SEND_CC;
1791 		break;
1792 
1793 	/* The receiver specifies a minimum checksum coverage value. To make
1794 	 * sense, this should be set to at least 8 (as done below). If zero is
1795 	 * used, this again means full checksum coverage.                     */
1796 	case UDPLITE_RECV_CSCOV:
1797 		if (!is_udplite)         /* Disable the option on UDP sockets */
1798 			return -ENOPROTOOPT;
1799 		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
1800 			val = 8;
1801 		else if (val > USHRT_MAX)
1802 			val = USHRT_MAX;
1803 		up->pcrlen = val;
1804 		up->pcflag |= UDPLITE_RECV_CC;
1805 		break;
1806 
1807 	default:
1808 		err = -ENOPROTOOPT;
1809 		break;
1810 	}
1811 
1812 	return err;
1813 }
1814 EXPORT_SYMBOL(udp_lib_setsockopt);
1815 
1816 int udp_setsockopt(struct sock *sk, int level, int optname,
1817 		   char __user *optval, unsigned int optlen)
1818 {
1819 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1820 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1821 					  udp_push_pending_frames);
1822 	return ip_setsockopt(sk, level, optname, optval, optlen);
1823 }
1824 
1825 #ifdef CONFIG_COMPAT
1826 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1827 			  char __user *optval, unsigned int optlen)
1828 {
1829 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1830 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1831 					  udp_push_pending_frames);
1832 	return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1833 }
1834 #endif
1835 
1836 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1837 		       char __user *optval, int __user *optlen)
1838 {
1839 	struct udp_sock *up = udp_sk(sk);
1840 	int val, len;
1841 
1842 	if (get_user(len, optlen))
1843 		return -EFAULT;
1844 
1845 	len = min_t(unsigned int, len, sizeof(int));
1846 
1847 	if (len < 0)
1848 		return -EINVAL;
1849 
1850 	switch (optname) {
1851 	case UDP_CORK:
1852 		val = up->corkflag;
1853 		break;
1854 
1855 	case UDP_ENCAP:
1856 		val = up->encap_type;
1857 		break;
1858 
1859 	/* The following two cannot be changed on UDP sockets, the return is
1860 	 * always 0 (which corresponds to the full checksum coverage of UDP). */
1861 	case UDPLITE_SEND_CSCOV:
1862 		val = up->pcslen;
1863 		break;
1864 
1865 	case UDPLITE_RECV_CSCOV:
1866 		val = up->pcrlen;
1867 		break;
1868 
1869 	default:
1870 		return -ENOPROTOOPT;
1871 	}
1872 
1873 	if (put_user(len, optlen))
1874 		return -EFAULT;
1875 	if (copy_to_user(optval, &val, len))
1876 		return -EFAULT;
1877 	return 0;
1878 }
1879 EXPORT_SYMBOL(udp_lib_getsockopt);
1880 
1881 int udp_getsockopt(struct sock *sk, int level, int optname,
1882 		   char __user *optval, int __user *optlen)
1883 {
1884 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1885 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1886 	return ip_getsockopt(sk, level, optname, optval, optlen);
1887 }
1888 
1889 #ifdef CONFIG_COMPAT
1890 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1891 				 char __user *optval, int __user *optlen)
1892 {
1893 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1894 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1895 	return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1896 }
1897 #endif
1898 /**
1899  * 	udp_poll - wait for a UDP event.
1900  *	@file - file struct
1901  *	@sock - socket
1902  *	@wait - poll table
1903  *
1904  *	This is same as datagram poll, except for the special case of
1905  *	blocking sockets. If application is using a blocking fd
1906  *	and a packet with checksum error is in the queue;
1907  *	then it could get return from select indicating data available
1908  *	but then block when reading it. Add special case code
1909  *	to work around these arguably broken applications.
1910  */
1911 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1912 {
1913 	unsigned int mask = datagram_poll(file, sock, wait);
1914 	struct sock *sk = sock->sk;
1915 
1916 	/* Check for false positives due to checksum errors */
1917 	if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1918 	    !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1919 		mask &= ~(POLLIN | POLLRDNORM);
1920 
1921 	return mask;
1922 
1923 }
1924 EXPORT_SYMBOL(udp_poll);
1925 
1926 struct proto udp_prot = {
1927 	.name		   = "UDP",
1928 	.owner		   = THIS_MODULE,
1929 	.close		   = udp_lib_close,
1930 	.connect	   = ip4_datagram_connect,
1931 	.disconnect	   = udp_disconnect,
1932 	.ioctl		   = udp_ioctl,
1933 	.destroy	   = udp_destroy_sock,
1934 	.setsockopt	   = udp_setsockopt,
1935 	.getsockopt	   = udp_getsockopt,
1936 	.sendmsg	   = udp_sendmsg,
1937 	.recvmsg	   = udp_recvmsg,
1938 	.sendpage	   = udp_sendpage,
1939 	.backlog_rcv	   = __udp_queue_rcv_skb,
1940 	.hash		   = udp_lib_hash,
1941 	.unhash		   = udp_lib_unhash,
1942 	.rehash		   = udp_v4_rehash,
1943 	.get_port	   = udp_v4_get_port,
1944 	.memory_allocated  = &udp_memory_allocated,
1945 	.sysctl_mem	   = sysctl_udp_mem,
1946 	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1947 	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1948 	.obj_size	   = sizeof(struct udp_sock),
1949 	.slab_flags	   = SLAB_DESTROY_BY_RCU,
1950 	.h.udp_table	   = &udp_table,
1951 #ifdef CONFIG_COMPAT
1952 	.compat_setsockopt = compat_udp_setsockopt,
1953 	.compat_getsockopt = compat_udp_getsockopt,
1954 #endif
1955 	.clear_sk	   = sk_prot_clear_portaddr_nulls,
1956 };
1957 EXPORT_SYMBOL(udp_prot);
1958 
1959 /* ------------------------------------------------------------------------ */
1960 #ifdef CONFIG_PROC_FS
1961 
1962 static struct sock *udp_get_first(struct seq_file *seq, int start)
1963 {
1964 	struct sock *sk;
1965 	struct udp_iter_state *state = seq->private;
1966 	struct net *net = seq_file_net(seq);
1967 
1968 	for (state->bucket = start; state->bucket <= state->udp_table->mask;
1969 	     ++state->bucket) {
1970 		struct hlist_nulls_node *node;
1971 		struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1972 
1973 		if (hlist_nulls_empty(&hslot->head))
1974 			continue;
1975 
1976 		spin_lock_bh(&hslot->lock);
1977 		sk_nulls_for_each(sk, node, &hslot->head) {
1978 			if (!net_eq(sock_net(sk), net))
1979 				continue;
1980 			if (sk->sk_family == state->family)
1981 				goto found;
1982 		}
1983 		spin_unlock_bh(&hslot->lock);
1984 	}
1985 	sk = NULL;
1986 found:
1987 	return sk;
1988 }
1989 
1990 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1991 {
1992 	struct udp_iter_state *state = seq->private;
1993 	struct net *net = seq_file_net(seq);
1994 
1995 	do {
1996 		sk = sk_nulls_next(sk);
1997 	} while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1998 
1999 	if (!sk) {
2000 		if (state->bucket <= state->udp_table->mask)
2001 			spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2002 		return udp_get_first(seq, state->bucket + 1);
2003 	}
2004 	return sk;
2005 }
2006 
2007 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2008 {
2009 	struct sock *sk = udp_get_first(seq, 0);
2010 
2011 	if (sk)
2012 		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2013 			--pos;
2014 	return pos ? NULL : sk;
2015 }
2016 
2017 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2018 {
2019 	struct udp_iter_state *state = seq->private;
2020 	state->bucket = MAX_UDP_PORTS;
2021 
2022 	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2023 }
2024 
2025 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2026 {
2027 	struct sock *sk;
2028 
2029 	if (v == SEQ_START_TOKEN)
2030 		sk = udp_get_idx(seq, 0);
2031 	else
2032 		sk = udp_get_next(seq, v);
2033 
2034 	++*pos;
2035 	return sk;
2036 }
2037 
2038 static void udp_seq_stop(struct seq_file *seq, void *v)
2039 {
2040 	struct udp_iter_state *state = seq->private;
2041 
2042 	if (state->bucket <= state->udp_table->mask)
2043 		spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2044 }
2045 
2046 int udp_seq_open(struct inode *inode, struct file *file)
2047 {
2048 	struct udp_seq_afinfo *afinfo = PDE(inode)->data;
2049 	struct udp_iter_state *s;
2050 	int err;
2051 
2052 	err = seq_open_net(inode, file, &afinfo->seq_ops,
2053 			   sizeof(struct udp_iter_state));
2054 	if (err < 0)
2055 		return err;
2056 
2057 	s = ((struct seq_file *)file->private_data)->private;
2058 	s->family		= afinfo->family;
2059 	s->udp_table		= afinfo->udp_table;
2060 	return err;
2061 }
2062 EXPORT_SYMBOL(udp_seq_open);
2063 
2064 /* ------------------------------------------------------------------------ */
2065 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2066 {
2067 	struct proc_dir_entry *p;
2068 	int rc = 0;
2069 
2070 	afinfo->seq_ops.start		= udp_seq_start;
2071 	afinfo->seq_ops.next		= udp_seq_next;
2072 	afinfo->seq_ops.stop		= udp_seq_stop;
2073 
2074 	p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2075 			     afinfo->seq_fops, afinfo);
2076 	if (!p)
2077 		rc = -ENOMEM;
2078 	return rc;
2079 }
2080 EXPORT_SYMBOL(udp_proc_register);
2081 
2082 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2083 {
2084 	proc_net_remove(net, afinfo->name);
2085 }
2086 EXPORT_SYMBOL(udp_proc_unregister);
2087 
2088 /* ------------------------------------------------------------------------ */
2089 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2090 		int bucket, int *len)
2091 {
2092 	struct inet_sock *inet = inet_sk(sp);
2093 	__be32 dest = inet->inet_daddr;
2094 	__be32 src  = inet->inet_rcv_saddr;
2095 	__u16 destp	  = ntohs(inet->inet_dport);
2096 	__u16 srcp	  = ntohs(inet->inet_sport);
2097 
2098 	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2099 		" %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n",
2100 		bucket, src, srcp, dest, destp, sp->sk_state,
2101 		sk_wmem_alloc_get(sp),
2102 		sk_rmem_alloc_get(sp),
2103 		0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
2104 		atomic_read(&sp->sk_refcnt), sp,
2105 		atomic_read(&sp->sk_drops), len);
2106 }
2107 
2108 int udp4_seq_show(struct seq_file *seq, void *v)
2109 {
2110 	if (v == SEQ_START_TOKEN)
2111 		seq_printf(seq, "%-127s\n",
2112 			   "  sl  local_address rem_address   st tx_queue "
2113 			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2114 			   "inode ref pointer drops");
2115 	else {
2116 		struct udp_iter_state *state = seq->private;
2117 		int len;
2118 
2119 		udp4_format_sock(v, seq, state->bucket, &len);
2120 		seq_printf(seq, "%*s\n", 127 - len, "");
2121 	}
2122 	return 0;
2123 }
2124 
2125 static const struct file_operations udp_afinfo_seq_fops = {
2126 	.owner    = THIS_MODULE,
2127 	.open     = udp_seq_open,
2128 	.read     = seq_read,
2129 	.llseek   = seq_lseek,
2130 	.release  = seq_release_net
2131 };
2132 
2133 /* ------------------------------------------------------------------------ */
2134 static struct udp_seq_afinfo udp4_seq_afinfo = {
2135 	.name		= "udp",
2136 	.family		= AF_INET,
2137 	.udp_table	= &udp_table,
2138 	.seq_fops	= &udp_afinfo_seq_fops,
2139 	.seq_ops	= {
2140 		.show		= udp4_seq_show,
2141 	},
2142 };
2143 
2144 static int __net_init udp4_proc_init_net(struct net *net)
2145 {
2146 	return udp_proc_register(net, &udp4_seq_afinfo);
2147 }
2148 
2149 static void __net_exit udp4_proc_exit_net(struct net *net)
2150 {
2151 	udp_proc_unregister(net, &udp4_seq_afinfo);
2152 }
2153 
2154 static struct pernet_operations udp4_net_ops = {
2155 	.init = udp4_proc_init_net,
2156 	.exit = udp4_proc_exit_net,
2157 };
2158 
2159 int __init udp4_proc_init(void)
2160 {
2161 	return register_pernet_subsys(&udp4_net_ops);
2162 }
2163 
2164 void udp4_proc_exit(void)
2165 {
2166 	unregister_pernet_subsys(&udp4_net_ops);
2167 }
2168 #endif /* CONFIG_PROC_FS */
2169 
2170 static __initdata unsigned long uhash_entries;
2171 static int __init set_uhash_entries(char *str)
2172 {
2173 	if (!str)
2174 		return 0;
2175 	uhash_entries = simple_strtoul(str, &str, 0);
2176 	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2177 		uhash_entries = UDP_HTABLE_SIZE_MIN;
2178 	return 1;
2179 }
2180 __setup("uhash_entries=", set_uhash_entries);
2181 
2182 void __init udp_table_init(struct udp_table *table, const char *name)
2183 {
2184 	unsigned int i;
2185 
2186 	if (!CONFIG_BASE_SMALL)
2187 		table->hash = alloc_large_system_hash(name,
2188 			2 * sizeof(struct udp_hslot),
2189 			uhash_entries,
2190 			21, /* one slot per 2 MB */
2191 			0,
2192 			&table->log,
2193 			&table->mask,
2194 			64 * 1024);
2195 	/*
2196 	 * Make sure hash table has the minimum size
2197 	 */
2198 	if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
2199 		table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
2200 				      2 * sizeof(struct udp_hslot), GFP_KERNEL);
2201 		if (!table->hash)
2202 			panic(name);
2203 		table->log = ilog2(UDP_HTABLE_SIZE_MIN);
2204 		table->mask = UDP_HTABLE_SIZE_MIN - 1;
2205 	}
2206 	table->hash2 = table->hash + (table->mask + 1);
2207 	for (i = 0; i <= table->mask; i++) {
2208 		INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2209 		table->hash[i].count = 0;
2210 		spin_lock_init(&table->hash[i].lock);
2211 	}
2212 	for (i = 0; i <= table->mask; i++) {
2213 		INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2214 		table->hash2[i].count = 0;
2215 		spin_lock_init(&table->hash2[i].lock);
2216 	}
2217 }
2218 
2219 void __init udp_init(void)
2220 {
2221 	unsigned long limit;
2222 
2223 	udp_table_init(&udp_table, "UDP");
2224 	limit = nr_free_buffer_pages() / 8;
2225 	limit = max(limit, 128UL);
2226 	sysctl_udp_mem[0] = limit / 4 * 3;
2227 	sysctl_udp_mem[1] = limit;
2228 	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2229 
2230 	sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2231 	sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2232 }
2233 
2234 int udp4_ufo_send_check(struct sk_buff *skb)
2235 {
2236 	const struct iphdr *iph;
2237 	struct udphdr *uh;
2238 
2239 	if (!pskb_may_pull(skb, sizeof(*uh)))
2240 		return -EINVAL;
2241 
2242 	iph = ip_hdr(skb);
2243 	uh = udp_hdr(skb);
2244 
2245 	uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2246 				       IPPROTO_UDP, 0);
2247 	skb->csum_start = skb_transport_header(skb) - skb->head;
2248 	skb->csum_offset = offsetof(struct udphdr, check);
2249 	skb->ip_summed = CHECKSUM_PARTIAL;
2250 	return 0;
2251 }
2252 
2253 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
2254 	netdev_features_t features)
2255 {
2256 	struct sk_buff *segs = ERR_PTR(-EINVAL);
2257 	unsigned int mss;
2258 	int offset;
2259 	__wsum csum;
2260 
2261 	mss = skb_shinfo(skb)->gso_size;
2262 	if (unlikely(skb->len <= mss))
2263 		goto out;
2264 
2265 	if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2266 		/* Packet is from an untrusted source, reset gso_segs. */
2267 		int type = skb_shinfo(skb)->gso_type;
2268 
2269 		if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2270 			     !(type & (SKB_GSO_UDP))))
2271 			goto out;
2272 
2273 		skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2274 
2275 		segs = NULL;
2276 		goto out;
2277 	}
2278 
2279 	/* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2280 	 * do checksum of UDP packets sent as multiple IP fragments.
2281 	 */
2282 	offset = skb_checksum_start_offset(skb);
2283 	csum = skb_checksum(skb, offset, skb->len - offset, 0);
2284 	offset += skb->csum_offset;
2285 	*(__sum16 *)(skb->data + offset) = csum_fold(csum);
2286 	skb->ip_summed = CHECKSUM_NONE;
2287 
2288 	/* Fragment the skb. IP headers of the fragments are updated in
2289 	 * inet_gso_segment()
2290 	 */
2291 	segs = skb_segment(skb, features);
2292 out:
2293 	return segs;
2294 }
2295 
2296