xref: /openbmc/linux/net/ipv4/udp.c (revision 2596e07a)
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/inetdevice.h>
94 #include <linux/in.h>
95 #include <linux/errno.h>
96 #include <linux/timer.h>
97 #include <linux/mm.h>
98 #include <linux/inet.h>
99 #include <linux/netdevice.h>
100 #include <linux/slab.h>
101 #include <net/tcp_states.h>
102 #include <linux/skbuff.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 #include <net/sock_reuseport.h>
117 
118 struct udp_table udp_table __read_mostly;
119 EXPORT_SYMBOL(udp_table);
120 
121 long sysctl_udp_mem[3] __read_mostly;
122 EXPORT_SYMBOL(sysctl_udp_mem);
123 
124 int sysctl_udp_rmem_min __read_mostly;
125 EXPORT_SYMBOL(sysctl_udp_rmem_min);
126 
127 int sysctl_udp_wmem_min __read_mostly;
128 EXPORT_SYMBOL(sysctl_udp_wmem_min);
129 
130 atomic_long_t udp_memory_allocated;
131 EXPORT_SYMBOL(udp_memory_allocated);
132 
133 #define MAX_UDP_PORTS 65536
134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
135 
136 static int udp_lib_lport_inuse(struct net *net, __u16 num,
137 			       const struct udp_hslot *hslot,
138 			       unsigned long *bitmap,
139 			       struct sock *sk,
140 			       int (*saddr_comp)(const struct sock *sk1,
141 						 const struct sock *sk2,
142 						 bool match_wildcard),
143 			       unsigned int log)
144 {
145 	struct sock *sk2;
146 	struct hlist_nulls_node *node;
147 	kuid_t uid = sock_i_uid(sk);
148 
149 	sk_nulls_for_each(sk2, node, &hslot->head) {
150 		if (net_eq(sock_net(sk2), net) &&
151 		    sk2 != sk &&
152 		    (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
153 		    (!sk2->sk_reuse || !sk->sk_reuse) &&
154 		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
155 		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
156 		    (!sk2->sk_reuseport || !sk->sk_reuseport ||
157 		     rcu_access_pointer(sk->sk_reuseport_cb) ||
158 		     !uid_eq(uid, sock_i_uid(sk2))) &&
159 		    saddr_comp(sk, sk2, true)) {
160 			if (!bitmap)
161 				return 1;
162 			__set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap);
163 		}
164 	}
165 	return 0;
166 }
167 
168 /*
169  * Note: we still hold spinlock of primary hash chain, so no other writer
170  * can insert/delete a socket with local_port == num
171  */
172 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
173 				struct udp_hslot *hslot2,
174 				struct sock *sk,
175 				int (*saddr_comp)(const struct sock *sk1,
176 						  const struct sock *sk2,
177 						  bool match_wildcard))
178 {
179 	struct sock *sk2;
180 	struct hlist_nulls_node *node;
181 	kuid_t uid = sock_i_uid(sk);
182 	int res = 0;
183 
184 	spin_lock(&hslot2->lock);
185 	udp_portaddr_for_each_entry(sk2, node, &hslot2->head) {
186 		if (net_eq(sock_net(sk2), net) &&
187 		    sk2 != sk &&
188 		    (udp_sk(sk2)->udp_port_hash == num) &&
189 		    (!sk2->sk_reuse || !sk->sk_reuse) &&
190 		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
191 		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
192 		    (!sk2->sk_reuseport || !sk->sk_reuseport ||
193 		     rcu_access_pointer(sk->sk_reuseport_cb) ||
194 		     !uid_eq(uid, sock_i_uid(sk2))) &&
195 		    saddr_comp(sk, sk2, true)) {
196 			res = 1;
197 			break;
198 		}
199 	}
200 	spin_unlock(&hslot2->lock);
201 	return res;
202 }
203 
204 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot,
205 				  int (*saddr_same)(const struct sock *sk1,
206 						    const struct sock *sk2,
207 						    bool match_wildcard))
208 {
209 	struct net *net = sock_net(sk);
210 	struct hlist_nulls_node *node;
211 	kuid_t uid = sock_i_uid(sk);
212 	struct sock *sk2;
213 
214 	sk_nulls_for_each(sk2, node, &hslot->head) {
215 		if (net_eq(sock_net(sk2), net) &&
216 		    sk2 != sk &&
217 		    sk2->sk_family == sk->sk_family &&
218 		    ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
219 		    (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
220 		    (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
221 		    sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
222 		    (*saddr_same)(sk, sk2, false)) {
223 			return reuseport_add_sock(sk, sk2);
224 		}
225 	}
226 
227 	/* Initial allocation may have already happened via setsockopt */
228 	if (!rcu_access_pointer(sk->sk_reuseport_cb))
229 		return reuseport_alloc(sk);
230 	return 0;
231 }
232 
233 /**
234  *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
235  *
236  *  @sk:          socket struct in question
237  *  @snum:        port number to look up
238  *  @saddr_comp:  AF-dependent comparison of bound local IP addresses
239  *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
240  *                   with NULL address
241  */
242 int udp_lib_get_port(struct sock *sk, unsigned short snum,
243 		     int (*saddr_comp)(const struct sock *sk1,
244 				       const struct sock *sk2,
245 				       bool match_wildcard),
246 		     unsigned int hash2_nulladdr)
247 {
248 	struct udp_hslot *hslot, *hslot2;
249 	struct udp_table *udptable = sk->sk_prot->h.udp_table;
250 	int    error = 1;
251 	struct net *net = sock_net(sk);
252 
253 	if (!snum) {
254 		int low, high, remaining;
255 		unsigned int rand;
256 		unsigned short first, last;
257 		DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
258 
259 		inet_get_local_port_range(net, &low, &high);
260 		remaining = (high - low) + 1;
261 
262 		rand = prandom_u32();
263 		first = reciprocal_scale(rand, remaining) + low;
264 		/*
265 		 * force rand to be an odd multiple of UDP_HTABLE_SIZE
266 		 */
267 		rand = (rand | 1) * (udptable->mask + 1);
268 		last = first + udptable->mask + 1;
269 		do {
270 			hslot = udp_hashslot(udptable, net, first);
271 			bitmap_zero(bitmap, PORTS_PER_CHAIN);
272 			spin_lock_bh(&hslot->lock);
273 			udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
274 					    saddr_comp, udptable->log);
275 
276 			snum = first;
277 			/*
278 			 * Iterate on all possible values of snum for this hash.
279 			 * Using steps of an odd multiple of UDP_HTABLE_SIZE
280 			 * give us randomization and full range coverage.
281 			 */
282 			do {
283 				if (low <= snum && snum <= high &&
284 				    !test_bit(snum >> udptable->log, bitmap) &&
285 				    !inet_is_local_reserved_port(net, snum))
286 					goto found;
287 				snum += rand;
288 			} while (snum != first);
289 			spin_unlock_bh(&hslot->lock);
290 		} while (++first != last);
291 		goto fail;
292 	} else {
293 		hslot = udp_hashslot(udptable, net, snum);
294 		spin_lock_bh(&hslot->lock);
295 		if (hslot->count > 10) {
296 			int exist;
297 			unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
298 
299 			slot2          &= udptable->mask;
300 			hash2_nulladdr &= udptable->mask;
301 
302 			hslot2 = udp_hashslot2(udptable, slot2);
303 			if (hslot->count < hslot2->count)
304 				goto scan_primary_hash;
305 
306 			exist = udp_lib_lport_inuse2(net, snum, hslot2,
307 						     sk, saddr_comp);
308 			if (!exist && (hash2_nulladdr != slot2)) {
309 				hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
310 				exist = udp_lib_lport_inuse2(net, snum, hslot2,
311 							     sk, saddr_comp);
312 			}
313 			if (exist)
314 				goto fail_unlock;
315 			else
316 				goto found;
317 		}
318 scan_primary_hash:
319 		if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
320 					saddr_comp, 0))
321 			goto fail_unlock;
322 	}
323 found:
324 	inet_sk(sk)->inet_num = snum;
325 	udp_sk(sk)->udp_port_hash = snum;
326 	udp_sk(sk)->udp_portaddr_hash ^= snum;
327 	if (sk_unhashed(sk)) {
328 		if (sk->sk_reuseport &&
329 		    udp_reuseport_add_sock(sk, hslot, saddr_comp)) {
330 			inet_sk(sk)->inet_num = 0;
331 			udp_sk(sk)->udp_port_hash = 0;
332 			udp_sk(sk)->udp_portaddr_hash ^= snum;
333 			goto fail_unlock;
334 		}
335 
336 		sk_nulls_add_node_rcu(sk, &hslot->head);
337 		hslot->count++;
338 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
339 
340 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
341 		spin_lock(&hslot2->lock);
342 		hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
343 					 &hslot2->head);
344 		hslot2->count++;
345 		spin_unlock(&hslot2->lock);
346 	}
347 	error = 0;
348 fail_unlock:
349 	spin_unlock_bh(&hslot->lock);
350 fail:
351 	return error;
352 }
353 EXPORT_SYMBOL(udp_lib_get_port);
354 
355 /* match_wildcard == true:  0.0.0.0 equals to any IPv4 addresses
356  * match_wildcard == false: addresses must be exactly the same, i.e.
357  *                          0.0.0.0 only equals to 0.0.0.0
358  */
359 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2,
360 				bool match_wildcard)
361 {
362 	struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
363 
364 	if (!ipv6_only_sock(sk2)) {
365 		if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)
366 			return 1;
367 		if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr)
368 			return match_wildcard;
369 	}
370 	return 0;
371 }
372 
373 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
374 			      unsigned int port)
375 {
376 	return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
377 }
378 
379 int udp_v4_get_port(struct sock *sk, unsigned short snum)
380 {
381 	unsigned int hash2_nulladdr =
382 		udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
383 	unsigned int hash2_partial =
384 		udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
385 
386 	/* precompute partial secondary hash */
387 	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
388 	return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
389 }
390 
391 static inline int compute_score(struct sock *sk, struct net *net,
392 				__be32 saddr, unsigned short hnum, __be16 sport,
393 				__be32 daddr, __be16 dport, int dif)
394 {
395 	int score;
396 	struct inet_sock *inet;
397 
398 	if (!net_eq(sock_net(sk), net) ||
399 	    udp_sk(sk)->udp_port_hash != hnum ||
400 	    ipv6_only_sock(sk))
401 		return -1;
402 
403 	score = (sk->sk_family == PF_INET) ? 2 : 1;
404 	inet = inet_sk(sk);
405 
406 	if (inet->inet_rcv_saddr) {
407 		if (inet->inet_rcv_saddr != daddr)
408 			return -1;
409 		score += 4;
410 	}
411 
412 	if (inet->inet_daddr) {
413 		if (inet->inet_daddr != saddr)
414 			return -1;
415 		score += 4;
416 	}
417 
418 	if (inet->inet_dport) {
419 		if (inet->inet_dport != sport)
420 			return -1;
421 		score += 4;
422 	}
423 
424 	if (sk->sk_bound_dev_if) {
425 		if (sk->sk_bound_dev_if != dif)
426 			return -1;
427 		score += 4;
428 	}
429 	if (sk->sk_incoming_cpu == raw_smp_processor_id())
430 		score++;
431 	return score;
432 }
433 
434 /*
435  * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
436  */
437 static inline int compute_score2(struct sock *sk, struct net *net,
438 				 __be32 saddr, __be16 sport,
439 				 __be32 daddr, unsigned int hnum, int dif)
440 {
441 	int score;
442 	struct inet_sock *inet;
443 
444 	if (!net_eq(sock_net(sk), net) ||
445 	    ipv6_only_sock(sk))
446 		return -1;
447 
448 	inet = inet_sk(sk);
449 
450 	if (inet->inet_rcv_saddr != daddr ||
451 	    inet->inet_num != hnum)
452 		return -1;
453 
454 	score = (sk->sk_family == PF_INET) ? 2 : 1;
455 
456 	if (inet->inet_daddr) {
457 		if (inet->inet_daddr != saddr)
458 			return -1;
459 		score += 4;
460 	}
461 
462 	if (inet->inet_dport) {
463 		if (inet->inet_dport != sport)
464 			return -1;
465 		score += 4;
466 	}
467 
468 	if (sk->sk_bound_dev_if) {
469 		if (sk->sk_bound_dev_if != dif)
470 			return -1;
471 		score += 4;
472 	}
473 
474 	if (sk->sk_incoming_cpu == raw_smp_processor_id())
475 		score++;
476 
477 	return score;
478 }
479 
480 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
481 		       const __u16 lport, const __be32 faddr,
482 		       const __be16 fport)
483 {
484 	static u32 udp_ehash_secret __read_mostly;
485 
486 	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
487 
488 	return __inet_ehashfn(laddr, lport, faddr, fport,
489 			      udp_ehash_secret + net_hash_mix(net));
490 }
491 
492 /* called with read_rcu_lock() */
493 static struct sock *udp4_lib_lookup2(struct net *net,
494 		__be32 saddr, __be16 sport,
495 		__be32 daddr, unsigned int hnum, int dif,
496 		struct udp_hslot *hslot2, unsigned int slot2,
497 		struct sk_buff *skb)
498 {
499 	struct sock *sk, *result;
500 	struct hlist_nulls_node *node;
501 	int score, badness, matches = 0, reuseport = 0;
502 	bool select_ok = true;
503 	u32 hash = 0;
504 
505 begin:
506 	result = NULL;
507 	badness = 0;
508 	udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
509 		score = compute_score2(sk, net, saddr, sport,
510 				      daddr, hnum, dif);
511 		if (score > badness) {
512 			result = sk;
513 			badness = score;
514 			reuseport = sk->sk_reuseport;
515 			if (reuseport) {
516 				hash = udp_ehashfn(net, daddr, hnum,
517 						   saddr, sport);
518 				if (select_ok) {
519 					struct sock *sk2;
520 
521 					sk2 = reuseport_select_sock(sk, hash, skb,
522 							sizeof(struct udphdr));
523 					if (sk2) {
524 						result = sk2;
525 						select_ok = false;
526 						goto found;
527 					}
528 				}
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) != slot2)
544 		goto begin;
545 	if (result) {
546 found:
547 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
548 			result = NULL;
549 		else if (unlikely(compute_score2(result, net, saddr, sport,
550 				  daddr, hnum, dif) < badness)) {
551 			sock_put(result);
552 			goto begin;
553 		}
554 	}
555 	return result;
556 }
557 
558 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
559  * harder than this. -DaveM
560  */
561 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
562 		__be16 sport, __be32 daddr, __be16 dport,
563 		int dif, struct udp_table *udptable, struct sk_buff *skb)
564 {
565 	struct sock *sk, *result;
566 	struct hlist_nulls_node *node;
567 	unsigned short hnum = ntohs(dport);
568 	unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
569 	struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
570 	int score, badness, matches = 0, reuseport = 0;
571 	bool select_ok = true;
572 	u32 hash = 0;
573 
574 	rcu_read_lock();
575 	if (hslot->count > 10) {
576 		hash2 = udp4_portaddr_hash(net, daddr, hnum);
577 		slot2 = hash2 & udptable->mask;
578 		hslot2 = &udptable->hash2[slot2];
579 		if (hslot->count < hslot2->count)
580 			goto begin;
581 
582 		result = udp4_lib_lookup2(net, saddr, sport,
583 					  daddr, hnum, dif,
584 					  hslot2, slot2, skb);
585 		if (!result) {
586 			hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
587 			slot2 = hash2 & udptable->mask;
588 			hslot2 = &udptable->hash2[slot2];
589 			if (hslot->count < hslot2->count)
590 				goto begin;
591 
592 			result = udp4_lib_lookup2(net, saddr, sport,
593 						  htonl(INADDR_ANY), hnum, dif,
594 						  hslot2, slot2, skb);
595 		}
596 		rcu_read_unlock();
597 		return result;
598 	}
599 begin:
600 	result = NULL;
601 	badness = 0;
602 	sk_nulls_for_each_rcu(sk, node, &hslot->head) {
603 		score = compute_score(sk, net, saddr, hnum, sport,
604 				      daddr, dport, dif);
605 		if (score > badness) {
606 			result = sk;
607 			badness = score;
608 			reuseport = sk->sk_reuseport;
609 			if (reuseport) {
610 				hash = udp_ehashfn(net, daddr, hnum,
611 						   saddr, sport);
612 				if (select_ok) {
613 					struct sock *sk2;
614 
615 					sk2 = reuseport_select_sock(sk, hash, skb,
616 							sizeof(struct udphdr));
617 					if (sk2) {
618 						result = sk2;
619 						select_ok = false;
620 						goto found;
621 					}
622 				}
623 				matches = 1;
624 			}
625 		} else if (score == badness && reuseport) {
626 			matches++;
627 			if (reciprocal_scale(hash, matches) == 0)
628 				result = sk;
629 			hash = next_pseudo_random32(hash);
630 		}
631 	}
632 	/*
633 	 * if the nulls value we got at the end of this lookup is
634 	 * not the expected one, we must restart lookup.
635 	 * We probably met an item that was moved to another chain.
636 	 */
637 	if (get_nulls_value(node) != slot)
638 		goto begin;
639 
640 	if (result) {
641 found:
642 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
643 			result = NULL;
644 		else if (unlikely(compute_score(result, net, saddr, hnum, sport,
645 				  daddr, dport, dif) < badness)) {
646 			sock_put(result);
647 			goto begin;
648 		}
649 	}
650 	rcu_read_unlock();
651 	return result;
652 }
653 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
654 
655 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
656 						 __be16 sport, __be16 dport,
657 						 struct udp_table *udptable)
658 {
659 	const struct iphdr *iph = ip_hdr(skb);
660 
661 	return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
662 				 iph->daddr, dport, inet_iif(skb),
663 				 udptable, skb);
664 }
665 
666 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
667 			     __be32 daddr, __be16 dport, int dif)
668 {
669 	return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif,
670 				 &udp_table, NULL);
671 }
672 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
673 
674 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
675 				       __be16 loc_port, __be32 loc_addr,
676 				       __be16 rmt_port, __be32 rmt_addr,
677 				       int dif, unsigned short hnum)
678 {
679 	struct inet_sock *inet = inet_sk(sk);
680 
681 	if (!net_eq(sock_net(sk), net) ||
682 	    udp_sk(sk)->udp_port_hash != hnum ||
683 	    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
684 	    (inet->inet_dport != rmt_port && inet->inet_dport) ||
685 	    (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
686 	    ipv6_only_sock(sk) ||
687 	    (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
688 		return false;
689 	if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
690 		return false;
691 	return true;
692 }
693 
694 /*
695  * This routine is called by the ICMP module when it gets some
696  * sort of error condition.  If err < 0 then the socket should
697  * be closed and the error returned to the user.  If err > 0
698  * it's just the icmp type << 8 | icmp code.
699  * Header points to the ip header of the error packet. We move
700  * on past this. Then (as it used to claim before adjustment)
701  * header points to the first 8 bytes of the udp header.  We need
702  * to find the appropriate port.
703  */
704 
705 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
706 {
707 	struct inet_sock *inet;
708 	const struct iphdr *iph = (const struct iphdr *)skb->data;
709 	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
710 	const int type = icmp_hdr(skb)->type;
711 	const int code = icmp_hdr(skb)->code;
712 	struct sock *sk;
713 	int harderr;
714 	int err;
715 	struct net *net = dev_net(skb->dev);
716 
717 	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
718 			iph->saddr, uh->source, skb->dev->ifindex, udptable,
719 			NULL);
720 	if (!sk) {
721 		ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
722 		return;	/* No socket for error */
723 	}
724 
725 	err = 0;
726 	harderr = 0;
727 	inet = inet_sk(sk);
728 
729 	switch (type) {
730 	default:
731 	case ICMP_TIME_EXCEEDED:
732 		err = EHOSTUNREACH;
733 		break;
734 	case ICMP_SOURCE_QUENCH:
735 		goto out;
736 	case ICMP_PARAMETERPROB:
737 		err = EPROTO;
738 		harderr = 1;
739 		break;
740 	case ICMP_DEST_UNREACH:
741 		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
742 			ipv4_sk_update_pmtu(skb, sk, info);
743 			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
744 				err = EMSGSIZE;
745 				harderr = 1;
746 				break;
747 			}
748 			goto out;
749 		}
750 		err = EHOSTUNREACH;
751 		if (code <= NR_ICMP_UNREACH) {
752 			harderr = icmp_err_convert[code].fatal;
753 			err = icmp_err_convert[code].errno;
754 		}
755 		break;
756 	case ICMP_REDIRECT:
757 		ipv4_sk_redirect(skb, sk);
758 		goto out;
759 	}
760 
761 	/*
762 	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
763 	 *	4.1.3.3.
764 	 */
765 	if (!inet->recverr) {
766 		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
767 			goto out;
768 	} else
769 		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
770 
771 	sk->sk_err = err;
772 	sk->sk_error_report(sk);
773 out:
774 	sock_put(sk);
775 }
776 
777 void udp_err(struct sk_buff *skb, u32 info)
778 {
779 	__udp4_lib_err(skb, info, &udp_table);
780 }
781 
782 /*
783  * Throw away all pending data and cancel the corking. Socket is locked.
784  */
785 void udp_flush_pending_frames(struct sock *sk)
786 {
787 	struct udp_sock *up = udp_sk(sk);
788 
789 	if (up->pending) {
790 		up->len = 0;
791 		up->pending = 0;
792 		ip_flush_pending_frames(sk);
793 	}
794 }
795 EXPORT_SYMBOL(udp_flush_pending_frames);
796 
797 /**
798  * 	udp4_hwcsum  -  handle outgoing HW checksumming
799  * 	@skb: 	sk_buff containing the filled-in UDP header
800  * 	        (checksum field must be zeroed out)
801  *	@src:	source IP address
802  *	@dst:	destination IP address
803  */
804 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
805 {
806 	struct udphdr *uh = udp_hdr(skb);
807 	int offset = skb_transport_offset(skb);
808 	int len = skb->len - offset;
809 	int hlen = len;
810 	__wsum csum = 0;
811 
812 	if (!skb_has_frag_list(skb)) {
813 		/*
814 		 * Only one fragment on the socket.
815 		 */
816 		skb->csum_start = skb_transport_header(skb) - skb->head;
817 		skb->csum_offset = offsetof(struct udphdr, check);
818 		uh->check = ~csum_tcpudp_magic(src, dst, len,
819 					       IPPROTO_UDP, 0);
820 	} else {
821 		struct sk_buff *frags;
822 
823 		/*
824 		 * HW-checksum won't work as there are two or more
825 		 * fragments on the socket so that all csums of sk_buffs
826 		 * should be together
827 		 */
828 		skb_walk_frags(skb, frags) {
829 			csum = csum_add(csum, frags->csum);
830 			hlen -= frags->len;
831 		}
832 
833 		csum = skb_checksum(skb, offset, hlen, csum);
834 		skb->ip_summed = CHECKSUM_NONE;
835 
836 		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
837 		if (uh->check == 0)
838 			uh->check = CSUM_MANGLED_0;
839 	}
840 }
841 EXPORT_SYMBOL_GPL(udp4_hwcsum);
842 
843 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
844  * for the simple case like when setting the checksum for a UDP tunnel.
845  */
846 void udp_set_csum(bool nocheck, struct sk_buff *skb,
847 		  __be32 saddr, __be32 daddr, int len)
848 {
849 	struct udphdr *uh = udp_hdr(skb);
850 
851 	if (nocheck)
852 		uh->check = 0;
853 	else if (skb_is_gso(skb))
854 		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
855 	else if (skb_dst(skb) && skb_dst(skb)->dev &&
856 		 (skb_dst(skb)->dev->features &
857 		  (NETIF_F_IP_CSUM | NETIF_F_HW_CSUM))) {
858 
859 		BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL);
860 
861 		skb->ip_summed = CHECKSUM_PARTIAL;
862 		skb->csum_start = skb_transport_header(skb) - skb->head;
863 		skb->csum_offset = offsetof(struct udphdr, check);
864 		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
865 	} else {
866 		__wsum csum;
867 
868 		BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL);
869 
870 		uh->check = 0;
871 		csum = skb_checksum(skb, 0, len, 0);
872 		uh->check = udp_v4_check(len, saddr, daddr, csum);
873 		if (uh->check == 0)
874 			uh->check = CSUM_MANGLED_0;
875 
876 		skb->ip_summed = CHECKSUM_UNNECESSARY;
877 	}
878 }
879 EXPORT_SYMBOL(udp_set_csum);
880 
881 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
882 {
883 	struct sock *sk = skb->sk;
884 	struct inet_sock *inet = inet_sk(sk);
885 	struct udphdr *uh;
886 	int err = 0;
887 	int is_udplite = IS_UDPLITE(sk);
888 	int offset = skb_transport_offset(skb);
889 	int len = skb->len - offset;
890 	__wsum csum = 0;
891 
892 	/*
893 	 * Create a UDP header
894 	 */
895 	uh = udp_hdr(skb);
896 	uh->source = inet->inet_sport;
897 	uh->dest = fl4->fl4_dport;
898 	uh->len = htons(len);
899 	uh->check = 0;
900 
901 	if (is_udplite)  				 /*     UDP-Lite      */
902 		csum = udplite_csum(skb);
903 
904 	else if (sk->sk_no_check_tx) {   /* UDP csum disabled */
905 
906 		skb->ip_summed = CHECKSUM_NONE;
907 		goto send;
908 
909 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
910 
911 		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
912 		goto send;
913 
914 	} else
915 		csum = udp_csum(skb);
916 
917 	/* add protocol-dependent pseudo-header */
918 	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
919 				      sk->sk_protocol, csum);
920 	if (uh->check == 0)
921 		uh->check = CSUM_MANGLED_0;
922 
923 send:
924 	err = ip_send_skb(sock_net(sk), skb);
925 	if (err) {
926 		if (err == -ENOBUFS && !inet->recverr) {
927 			UDP_INC_STATS_USER(sock_net(sk),
928 					   UDP_MIB_SNDBUFERRORS, is_udplite);
929 			err = 0;
930 		}
931 	} else
932 		UDP_INC_STATS_USER(sock_net(sk),
933 				   UDP_MIB_OUTDATAGRAMS, is_udplite);
934 	return err;
935 }
936 
937 /*
938  * Push out all pending data as one UDP datagram. Socket is locked.
939  */
940 int udp_push_pending_frames(struct sock *sk)
941 {
942 	struct udp_sock  *up = udp_sk(sk);
943 	struct inet_sock *inet = inet_sk(sk);
944 	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
945 	struct sk_buff *skb;
946 	int err = 0;
947 
948 	skb = ip_finish_skb(sk, fl4);
949 	if (!skb)
950 		goto out;
951 
952 	err = udp_send_skb(skb, fl4);
953 
954 out:
955 	up->len = 0;
956 	up->pending = 0;
957 	return err;
958 }
959 EXPORT_SYMBOL(udp_push_pending_frames);
960 
961 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
962 {
963 	struct inet_sock *inet = inet_sk(sk);
964 	struct udp_sock *up = udp_sk(sk);
965 	struct flowi4 fl4_stack;
966 	struct flowi4 *fl4;
967 	int ulen = len;
968 	struct ipcm_cookie ipc;
969 	struct rtable *rt = NULL;
970 	int free = 0;
971 	int connected = 0;
972 	__be32 daddr, faddr, saddr;
973 	__be16 dport;
974 	u8  tos;
975 	int err, is_udplite = IS_UDPLITE(sk);
976 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
977 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
978 	struct sk_buff *skb;
979 	struct ip_options_data opt_copy;
980 
981 	if (len > 0xFFFF)
982 		return -EMSGSIZE;
983 
984 	/*
985 	 *	Check the flags.
986 	 */
987 
988 	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
989 		return -EOPNOTSUPP;
990 
991 	ipc.opt = NULL;
992 	ipc.tx_flags = 0;
993 	ipc.ttl = 0;
994 	ipc.tos = -1;
995 
996 	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
997 
998 	fl4 = &inet->cork.fl.u.ip4;
999 	if (up->pending) {
1000 		/*
1001 		 * There are pending frames.
1002 		 * The socket lock must be held while it's corked.
1003 		 */
1004 		lock_sock(sk);
1005 		if (likely(up->pending)) {
1006 			if (unlikely(up->pending != AF_INET)) {
1007 				release_sock(sk);
1008 				return -EINVAL;
1009 			}
1010 			goto do_append_data;
1011 		}
1012 		release_sock(sk);
1013 	}
1014 	ulen += sizeof(struct udphdr);
1015 
1016 	/*
1017 	 *	Get and verify the address.
1018 	 */
1019 	if (msg->msg_name) {
1020 		DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1021 		if (msg->msg_namelen < sizeof(*usin))
1022 			return -EINVAL;
1023 		if (usin->sin_family != AF_INET) {
1024 			if (usin->sin_family != AF_UNSPEC)
1025 				return -EAFNOSUPPORT;
1026 		}
1027 
1028 		daddr = usin->sin_addr.s_addr;
1029 		dport = usin->sin_port;
1030 		if (dport == 0)
1031 			return -EINVAL;
1032 	} else {
1033 		if (sk->sk_state != TCP_ESTABLISHED)
1034 			return -EDESTADDRREQ;
1035 		daddr = inet->inet_daddr;
1036 		dport = inet->inet_dport;
1037 		/* Open fast path for connected socket.
1038 		   Route will not be used, if at least one option is set.
1039 		 */
1040 		connected = 1;
1041 	}
1042 	ipc.addr = inet->inet_saddr;
1043 
1044 	ipc.oif = sk->sk_bound_dev_if;
1045 
1046 	sock_tx_timestamp(sk, &ipc.tx_flags);
1047 
1048 	if (msg->msg_controllen) {
1049 		err = ip_cmsg_send(sock_net(sk), msg, &ipc,
1050 				   sk->sk_family == AF_INET6);
1051 		if (unlikely(err)) {
1052 			kfree(ipc.opt);
1053 			return err;
1054 		}
1055 		if (ipc.opt)
1056 			free = 1;
1057 		connected = 0;
1058 	}
1059 	if (!ipc.opt) {
1060 		struct ip_options_rcu *inet_opt;
1061 
1062 		rcu_read_lock();
1063 		inet_opt = rcu_dereference(inet->inet_opt);
1064 		if (inet_opt) {
1065 			memcpy(&opt_copy, inet_opt,
1066 			       sizeof(*inet_opt) + inet_opt->opt.optlen);
1067 			ipc.opt = &opt_copy.opt;
1068 		}
1069 		rcu_read_unlock();
1070 	}
1071 
1072 	saddr = ipc.addr;
1073 	ipc.addr = faddr = daddr;
1074 
1075 	if (ipc.opt && ipc.opt->opt.srr) {
1076 		if (!daddr)
1077 			return -EINVAL;
1078 		faddr = ipc.opt->opt.faddr;
1079 		connected = 0;
1080 	}
1081 	tos = get_rttos(&ipc, inet);
1082 	if (sock_flag(sk, SOCK_LOCALROUTE) ||
1083 	    (msg->msg_flags & MSG_DONTROUTE) ||
1084 	    (ipc.opt && ipc.opt->opt.is_strictroute)) {
1085 		tos |= RTO_ONLINK;
1086 		connected = 0;
1087 	}
1088 
1089 	if (ipv4_is_multicast(daddr)) {
1090 		if (!ipc.oif)
1091 			ipc.oif = inet->mc_index;
1092 		if (!saddr)
1093 			saddr = inet->mc_addr;
1094 		connected = 0;
1095 	} else if (!ipc.oif)
1096 		ipc.oif = inet->uc_index;
1097 
1098 	if (connected)
1099 		rt = (struct rtable *)sk_dst_check(sk, 0);
1100 
1101 	if (!rt) {
1102 		struct net *net = sock_net(sk);
1103 		__u8 flow_flags = inet_sk_flowi_flags(sk);
1104 
1105 		fl4 = &fl4_stack;
1106 
1107 		flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1108 				   RT_SCOPE_UNIVERSE, sk->sk_protocol,
1109 				   flow_flags,
1110 				   faddr, saddr, dport, inet->inet_sport);
1111 
1112 		if (!saddr && ipc.oif) {
1113 			err = l3mdev_get_saddr(net, ipc.oif, fl4);
1114 			if (err < 0)
1115 				goto out;
1116 		}
1117 
1118 		security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1119 		rt = ip_route_output_flow(net, fl4, sk);
1120 		if (IS_ERR(rt)) {
1121 			err = PTR_ERR(rt);
1122 			rt = NULL;
1123 			if (err == -ENETUNREACH)
1124 				IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1125 			goto out;
1126 		}
1127 
1128 		err = -EACCES;
1129 		if ((rt->rt_flags & RTCF_BROADCAST) &&
1130 		    !sock_flag(sk, SOCK_BROADCAST))
1131 			goto out;
1132 		if (connected)
1133 			sk_dst_set(sk, dst_clone(&rt->dst));
1134 	}
1135 
1136 	if (msg->msg_flags&MSG_CONFIRM)
1137 		goto do_confirm;
1138 back_from_confirm:
1139 
1140 	saddr = fl4->saddr;
1141 	if (!ipc.addr)
1142 		daddr = ipc.addr = fl4->daddr;
1143 
1144 	/* Lockless fast path for the non-corking case. */
1145 	if (!corkreq) {
1146 		skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1147 				  sizeof(struct udphdr), &ipc, &rt,
1148 				  msg->msg_flags);
1149 		err = PTR_ERR(skb);
1150 		if (!IS_ERR_OR_NULL(skb))
1151 			err = udp_send_skb(skb, fl4);
1152 		goto out;
1153 	}
1154 
1155 	lock_sock(sk);
1156 	if (unlikely(up->pending)) {
1157 		/* The socket is already corked while preparing it. */
1158 		/* ... which is an evident application bug. --ANK */
1159 		release_sock(sk);
1160 
1161 		net_dbg_ratelimited("cork app bug 2\n");
1162 		err = -EINVAL;
1163 		goto out;
1164 	}
1165 	/*
1166 	 *	Now cork the socket to pend data.
1167 	 */
1168 	fl4 = &inet->cork.fl.u.ip4;
1169 	fl4->daddr = daddr;
1170 	fl4->saddr = saddr;
1171 	fl4->fl4_dport = dport;
1172 	fl4->fl4_sport = inet->inet_sport;
1173 	up->pending = AF_INET;
1174 
1175 do_append_data:
1176 	up->len += ulen;
1177 	err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1178 			     sizeof(struct udphdr), &ipc, &rt,
1179 			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1180 	if (err)
1181 		udp_flush_pending_frames(sk);
1182 	else if (!corkreq)
1183 		err = udp_push_pending_frames(sk);
1184 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1185 		up->pending = 0;
1186 	release_sock(sk);
1187 
1188 out:
1189 	ip_rt_put(rt);
1190 	if (free)
1191 		kfree(ipc.opt);
1192 	if (!err)
1193 		return len;
1194 	/*
1195 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1196 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1197 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1198 	 * things).  We could add another new stat but at least for now that
1199 	 * seems like overkill.
1200 	 */
1201 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1202 		UDP_INC_STATS_USER(sock_net(sk),
1203 				UDP_MIB_SNDBUFERRORS, is_udplite);
1204 	}
1205 	return err;
1206 
1207 do_confirm:
1208 	dst_confirm(&rt->dst);
1209 	if (!(msg->msg_flags&MSG_PROBE) || len)
1210 		goto back_from_confirm;
1211 	err = 0;
1212 	goto out;
1213 }
1214 EXPORT_SYMBOL(udp_sendmsg);
1215 
1216 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1217 		 size_t size, int flags)
1218 {
1219 	struct inet_sock *inet = inet_sk(sk);
1220 	struct udp_sock *up = udp_sk(sk);
1221 	int ret;
1222 
1223 	if (flags & MSG_SENDPAGE_NOTLAST)
1224 		flags |= MSG_MORE;
1225 
1226 	if (!up->pending) {
1227 		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
1228 
1229 		/* Call udp_sendmsg to specify destination address which
1230 		 * sendpage interface can't pass.
1231 		 * This will succeed only when the socket is connected.
1232 		 */
1233 		ret = udp_sendmsg(sk, &msg, 0);
1234 		if (ret < 0)
1235 			return ret;
1236 	}
1237 
1238 	lock_sock(sk);
1239 
1240 	if (unlikely(!up->pending)) {
1241 		release_sock(sk);
1242 
1243 		net_dbg_ratelimited("udp cork app bug 3\n");
1244 		return -EINVAL;
1245 	}
1246 
1247 	ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1248 			     page, offset, size, flags);
1249 	if (ret == -EOPNOTSUPP) {
1250 		release_sock(sk);
1251 		return sock_no_sendpage(sk->sk_socket, page, offset,
1252 					size, flags);
1253 	}
1254 	if (ret < 0) {
1255 		udp_flush_pending_frames(sk);
1256 		goto out;
1257 	}
1258 
1259 	up->len += size;
1260 	if (!(up->corkflag || (flags&MSG_MORE)))
1261 		ret = udp_push_pending_frames(sk);
1262 	if (!ret)
1263 		ret = size;
1264 out:
1265 	release_sock(sk);
1266 	return ret;
1267 }
1268 
1269 /**
1270  *	first_packet_length	- return length of first packet in receive queue
1271  *	@sk: socket
1272  *
1273  *	Drops all bad checksum frames, until a valid one is found.
1274  *	Returns the length of found skb, or 0 if none is found.
1275  */
1276 static unsigned int first_packet_length(struct sock *sk)
1277 {
1278 	struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1279 	struct sk_buff *skb;
1280 	unsigned int res;
1281 
1282 	__skb_queue_head_init(&list_kill);
1283 
1284 	spin_lock_bh(&rcvq->lock);
1285 	while ((skb = skb_peek(rcvq)) != NULL &&
1286 		udp_lib_checksum_complete(skb)) {
1287 		UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1288 				 IS_UDPLITE(sk));
1289 		UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1290 				 IS_UDPLITE(sk));
1291 		atomic_inc(&sk->sk_drops);
1292 		__skb_unlink(skb, rcvq);
1293 		__skb_queue_tail(&list_kill, skb);
1294 	}
1295 	res = skb ? skb->len : 0;
1296 	spin_unlock_bh(&rcvq->lock);
1297 
1298 	if (!skb_queue_empty(&list_kill)) {
1299 		bool slow = lock_sock_fast(sk);
1300 
1301 		__skb_queue_purge(&list_kill);
1302 		sk_mem_reclaim_partial(sk);
1303 		unlock_sock_fast(sk, slow);
1304 	}
1305 	return res;
1306 }
1307 
1308 /*
1309  *	IOCTL requests applicable to the UDP protocol
1310  */
1311 
1312 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1313 {
1314 	switch (cmd) {
1315 	case SIOCOUTQ:
1316 	{
1317 		int amount = sk_wmem_alloc_get(sk);
1318 
1319 		return put_user(amount, (int __user *)arg);
1320 	}
1321 
1322 	case SIOCINQ:
1323 	{
1324 		unsigned int amount = first_packet_length(sk);
1325 
1326 		if (amount)
1327 			/*
1328 			 * We will only return the amount
1329 			 * of this packet since that is all
1330 			 * that will be read.
1331 			 */
1332 			amount -= sizeof(struct udphdr);
1333 
1334 		return put_user(amount, (int __user *)arg);
1335 	}
1336 
1337 	default:
1338 		return -ENOIOCTLCMD;
1339 	}
1340 
1341 	return 0;
1342 }
1343 EXPORT_SYMBOL(udp_ioctl);
1344 
1345 /*
1346  * 	This should be easy, if there is something there we
1347  * 	return it, otherwise we block.
1348  */
1349 
1350 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1351 		int flags, int *addr_len)
1352 {
1353 	struct inet_sock *inet = inet_sk(sk);
1354 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1355 	struct sk_buff *skb;
1356 	unsigned int ulen, copied;
1357 	int peeked, off = 0;
1358 	int err;
1359 	int is_udplite = IS_UDPLITE(sk);
1360 	bool checksum_valid = false;
1361 	bool slow;
1362 
1363 	if (flags & MSG_ERRQUEUE)
1364 		return ip_recv_error(sk, msg, len, addr_len);
1365 
1366 try_again:
1367 	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1368 				  &peeked, &off, &err);
1369 	if (!skb)
1370 		goto out;
1371 
1372 	ulen = skb->len - sizeof(struct udphdr);
1373 	copied = len;
1374 	if (copied > ulen)
1375 		copied = ulen;
1376 	else if (copied < ulen)
1377 		msg->msg_flags |= MSG_TRUNC;
1378 
1379 	/*
1380 	 * If checksum is needed at all, try to do it while copying the
1381 	 * data.  If the data is truncated, or if we only want a partial
1382 	 * coverage checksum (UDP-Lite), do it before the copy.
1383 	 */
1384 
1385 	if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1386 		checksum_valid = !udp_lib_checksum_complete(skb);
1387 		if (!checksum_valid)
1388 			goto csum_copy_err;
1389 	}
1390 
1391 	if (checksum_valid || skb_csum_unnecessary(skb))
1392 		err = skb_copy_datagram_msg(skb, sizeof(struct udphdr),
1393 					    msg, copied);
1394 	else {
1395 		err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr),
1396 						     msg);
1397 
1398 		if (err == -EINVAL)
1399 			goto csum_copy_err;
1400 	}
1401 
1402 	if (unlikely(err)) {
1403 		trace_kfree_skb(skb, udp_recvmsg);
1404 		if (!peeked) {
1405 			atomic_inc(&sk->sk_drops);
1406 			UDP_INC_STATS_USER(sock_net(sk),
1407 					   UDP_MIB_INERRORS, is_udplite);
1408 		}
1409 		goto out_free;
1410 	}
1411 
1412 	if (!peeked)
1413 		UDP_INC_STATS_USER(sock_net(sk),
1414 				UDP_MIB_INDATAGRAMS, is_udplite);
1415 
1416 	sock_recv_ts_and_drops(msg, sk, skb);
1417 
1418 	/* Copy the address. */
1419 	if (sin) {
1420 		sin->sin_family = AF_INET;
1421 		sin->sin_port = udp_hdr(skb)->source;
1422 		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1423 		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1424 		*addr_len = sizeof(*sin);
1425 	}
1426 	if (inet->cmsg_flags)
1427 		ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr));
1428 
1429 	err = copied;
1430 	if (flags & MSG_TRUNC)
1431 		err = ulen;
1432 
1433 out_free:
1434 	skb_free_datagram_locked(sk, skb);
1435 out:
1436 	return err;
1437 
1438 csum_copy_err:
1439 	slow = lock_sock_fast(sk);
1440 	if (!skb_kill_datagram(sk, skb, flags)) {
1441 		UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1442 		UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1443 	}
1444 	unlock_sock_fast(sk, slow);
1445 
1446 	/* starting over for a new packet, but check if we need to yield */
1447 	cond_resched();
1448 	msg->msg_flags &= ~MSG_TRUNC;
1449 	goto try_again;
1450 }
1451 
1452 int udp_disconnect(struct sock *sk, int flags)
1453 {
1454 	struct inet_sock *inet = inet_sk(sk);
1455 	/*
1456 	 *	1003.1g - break association.
1457 	 */
1458 
1459 	sk->sk_state = TCP_CLOSE;
1460 	inet->inet_daddr = 0;
1461 	inet->inet_dport = 0;
1462 	sock_rps_reset_rxhash(sk);
1463 	sk->sk_bound_dev_if = 0;
1464 	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1465 		inet_reset_saddr(sk);
1466 
1467 	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1468 		sk->sk_prot->unhash(sk);
1469 		inet->inet_sport = 0;
1470 	}
1471 	sk_dst_reset(sk);
1472 	return 0;
1473 }
1474 EXPORT_SYMBOL(udp_disconnect);
1475 
1476 void udp_lib_unhash(struct sock *sk)
1477 {
1478 	if (sk_hashed(sk)) {
1479 		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1480 		struct udp_hslot *hslot, *hslot2;
1481 
1482 		hslot  = udp_hashslot(udptable, sock_net(sk),
1483 				      udp_sk(sk)->udp_port_hash);
1484 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1485 
1486 		spin_lock_bh(&hslot->lock);
1487 		if (rcu_access_pointer(sk->sk_reuseport_cb))
1488 			reuseport_detach_sock(sk);
1489 		if (sk_nulls_del_node_init_rcu(sk)) {
1490 			hslot->count--;
1491 			inet_sk(sk)->inet_num = 0;
1492 			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1493 
1494 			spin_lock(&hslot2->lock);
1495 			hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1496 			hslot2->count--;
1497 			spin_unlock(&hslot2->lock);
1498 		}
1499 		spin_unlock_bh(&hslot->lock);
1500 	}
1501 }
1502 EXPORT_SYMBOL(udp_lib_unhash);
1503 
1504 /*
1505  * inet_rcv_saddr was changed, we must rehash secondary hash
1506  */
1507 void udp_lib_rehash(struct sock *sk, u16 newhash)
1508 {
1509 	if (sk_hashed(sk)) {
1510 		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1511 		struct udp_hslot *hslot, *hslot2, *nhslot2;
1512 
1513 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1514 		nhslot2 = udp_hashslot2(udptable, newhash);
1515 		udp_sk(sk)->udp_portaddr_hash = newhash;
1516 
1517 		if (hslot2 != nhslot2 ||
1518 		    rcu_access_pointer(sk->sk_reuseport_cb)) {
1519 			hslot = udp_hashslot(udptable, sock_net(sk),
1520 					     udp_sk(sk)->udp_port_hash);
1521 			/* we must lock primary chain too */
1522 			spin_lock_bh(&hslot->lock);
1523 			if (rcu_access_pointer(sk->sk_reuseport_cb))
1524 				reuseport_detach_sock(sk);
1525 
1526 			if (hslot2 != nhslot2) {
1527 				spin_lock(&hslot2->lock);
1528 				hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1529 				hslot2->count--;
1530 				spin_unlock(&hslot2->lock);
1531 
1532 				spin_lock(&nhslot2->lock);
1533 				hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1534 							 &nhslot2->head);
1535 				nhslot2->count++;
1536 				spin_unlock(&nhslot2->lock);
1537 			}
1538 
1539 			spin_unlock_bh(&hslot->lock);
1540 		}
1541 	}
1542 }
1543 EXPORT_SYMBOL(udp_lib_rehash);
1544 
1545 static void udp_v4_rehash(struct sock *sk)
1546 {
1547 	u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1548 					  inet_sk(sk)->inet_rcv_saddr,
1549 					  inet_sk(sk)->inet_num);
1550 	udp_lib_rehash(sk, new_hash);
1551 }
1552 
1553 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1554 {
1555 	int rc;
1556 
1557 	if (inet_sk(sk)->inet_daddr) {
1558 		sock_rps_save_rxhash(sk, skb);
1559 		sk_mark_napi_id(sk, skb);
1560 		sk_incoming_cpu_update(sk);
1561 	}
1562 
1563 	rc = sock_queue_rcv_skb(sk, skb);
1564 	if (rc < 0) {
1565 		int is_udplite = IS_UDPLITE(sk);
1566 
1567 		/* Note that an ENOMEM error is charged twice */
1568 		if (rc == -ENOMEM)
1569 			UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1570 					 is_udplite);
1571 		UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1572 		kfree_skb(skb);
1573 		trace_udp_fail_queue_rcv_skb(rc, sk);
1574 		return -1;
1575 	}
1576 
1577 	return 0;
1578 
1579 }
1580 
1581 static struct static_key udp_encap_needed __read_mostly;
1582 void udp_encap_enable(void)
1583 {
1584 	if (!static_key_enabled(&udp_encap_needed))
1585 		static_key_slow_inc(&udp_encap_needed);
1586 }
1587 EXPORT_SYMBOL(udp_encap_enable);
1588 
1589 /* returns:
1590  *  -1: error
1591  *   0: success
1592  *  >0: "udp encap" protocol resubmission
1593  *
1594  * Note that in the success and error cases, the skb is assumed to
1595  * have either been requeued or freed.
1596  */
1597 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1598 {
1599 	struct udp_sock *up = udp_sk(sk);
1600 	int rc;
1601 	int is_udplite = IS_UDPLITE(sk);
1602 
1603 	/*
1604 	 *	Charge it to the socket, dropping if the queue is full.
1605 	 */
1606 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1607 		goto drop;
1608 	nf_reset(skb);
1609 
1610 	if (static_key_false(&udp_encap_needed) && up->encap_type) {
1611 		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1612 
1613 		/*
1614 		 * This is an encapsulation socket so pass the skb to
1615 		 * the socket's udp_encap_rcv() hook. Otherwise, just
1616 		 * fall through and pass this up the UDP socket.
1617 		 * up->encap_rcv() returns the following value:
1618 		 * =0 if skb was successfully passed to the encap
1619 		 *    handler or was discarded by it.
1620 		 * >0 if skb should be passed on to UDP.
1621 		 * <0 if skb should be resubmitted as proto -N
1622 		 */
1623 
1624 		/* if we're overly short, let UDP handle it */
1625 		encap_rcv = ACCESS_ONCE(up->encap_rcv);
1626 		if (skb->len > sizeof(struct udphdr) && encap_rcv) {
1627 			int ret;
1628 
1629 			/* Verify checksum before giving to encap */
1630 			if (udp_lib_checksum_complete(skb))
1631 				goto csum_error;
1632 
1633 			ret = encap_rcv(sk, skb);
1634 			if (ret <= 0) {
1635 				UDP_INC_STATS_BH(sock_net(sk),
1636 						 UDP_MIB_INDATAGRAMS,
1637 						 is_udplite);
1638 				return -ret;
1639 			}
1640 		}
1641 
1642 		/* FALLTHROUGH -- it's a UDP Packet */
1643 	}
1644 
1645 	/*
1646 	 * 	UDP-Lite specific tests, ignored on UDP sockets
1647 	 */
1648 	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
1649 
1650 		/*
1651 		 * MIB statistics other than incrementing the error count are
1652 		 * disabled for the following two types of errors: these depend
1653 		 * on the application settings, not on the functioning of the
1654 		 * protocol stack as such.
1655 		 *
1656 		 * RFC 3828 here recommends (sec 3.3): "There should also be a
1657 		 * way ... to ... at least let the receiving application block
1658 		 * delivery of packets with coverage values less than a value
1659 		 * provided by the application."
1660 		 */
1661 		if (up->pcrlen == 0) {          /* full coverage was set  */
1662 			net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1663 					    UDP_SKB_CB(skb)->cscov, skb->len);
1664 			goto drop;
1665 		}
1666 		/* The next case involves violating the min. coverage requested
1667 		 * by the receiver. This is subtle: if receiver wants x and x is
1668 		 * greater than the buffersize/MTU then receiver will complain
1669 		 * that it wants x while sender emits packets of smaller size y.
1670 		 * Therefore the above ...()->partial_cov statement is essential.
1671 		 */
1672 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
1673 			net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1674 					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
1675 			goto drop;
1676 		}
1677 	}
1678 
1679 	if (rcu_access_pointer(sk->sk_filter) &&
1680 	    udp_lib_checksum_complete(skb))
1681 		goto csum_error;
1682 
1683 	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
1684 		UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1685 				 is_udplite);
1686 		goto drop;
1687 	}
1688 
1689 	rc = 0;
1690 
1691 	ipv4_pktinfo_prepare(sk, skb);
1692 	bh_lock_sock(sk);
1693 	if (!sock_owned_by_user(sk))
1694 		rc = __udp_queue_rcv_skb(sk, skb);
1695 	else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1696 		bh_unlock_sock(sk);
1697 		goto drop;
1698 	}
1699 	bh_unlock_sock(sk);
1700 
1701 	return rc;
1702 
1703 csum_error:
1704 	UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1705 drop:
1706 	UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1707 	atomic_inc(&sk->sk_drops);
1708 	kfree_skb(skb);
1709 	return -1;
1710 }
1711 
1712 static void flush_stack(struct sock **stack, unsigned int count,
1713 			struct sk_buff *skb, unsigned int final)
1714 {
1715 	unsigned int i;
1716 	struct sk_buff *skb1 = NULL;
1717 	struct sock *sk;
1718 
1719 	for (i = 0; i < count; i++) {
1720 		sk = stack[i];
1721 		if (likely(!skb1))
1722 			skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1723 
1724 		if (!skb1) {
1725 			atomic_inc(&sk->sk_drops);
1726 			UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1727 					 IS_UDPLITE(sk));
1728 			UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1729 					 IS_UDPLITE(sk));
1730 		}
1731 
1732 		if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1733 			skb1 = NULL;
1734 
1735 		sock_put(sk);
1736 	}
1737 	if (unlikely(skb1))
1738 		kfree_skb(skb1);
1739 }
1740 
1741 /* For TCP sockets, sk_rx_dst is protected by socket lock
1742  * For UDP, we use xchg() to guard against concurrent changes.
1743  */
1744 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1745 {
1746 	struct dst_entry *old;
1747 
1748 	dst_hold(dst);
1749 	old = xchg(&sk->sk_rx_dst, dst);
1750 	dst_release(old);
1751 }
1752 
1753 /*
1754  *	Multicasts and broadcasts go to each listener.
1755  *
1756  *	Note: called only from the BH handler context.
1757  */
1758 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1759 				    struct udphdr  *uh,
1760 				    __be32 saddr, __be32 daddr,
1761 				    struct udp_table *udptable,
1762 				    int proto)
1763 {
1764 	struct sock *sk, *stack[256 / sizeof(struct sock *)];
1765 	struct hlist_nulls_node *node;
1766 	unsigned short hnum = ntohs(uh->dest);
1767 	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1768 	int dif = skb->dev->ifindex;
1769 	unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
1770 	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1771 	bool inner_flushed = false;
1772 
1773 	if (use_hash2) {
1774 		hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1775 			    udp_table.mask;
1776 		hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask;
1777 start_lookup:
1778 		hslot = &udp_table.hash2[hash2];
1779 		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1780 	}
1781 
1782 	spin_lock(&hslot->lock);
1783 	sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
1784 		if (__udp_is_mcast_sock(net, sk,
1785 					uh->dest, daddr,
1786 					uh->source, saddr,
1787 					dif, hnum)) {
1788 			if (unlikely(count == ARRAY_SIZE(stack))) {
1789 				flush_stack(stack, count, skb, ~0);
1790 				inner_flushed = true;
1791 				count = 0;
1792 			}
1793 			stack[count++] = sk;
1794 			sock_hold(sk);
1795 		}
1796 	}
1797 
1798 	spin_unlock(&hslot->lock);
1799 
1800 	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
1801 	if (use_hash2 && hash2 != hash2_any) {
1802 		hash2 = hash2_any;
1803 		goto start_lookup;
1804 	}
1805 
1806 	/*
1807 	 * do the slow work with no lock held
1808 	 */
1809 	if (count) {
1810 		flush_stack(stack, count, skb, count - 1);
1811 	} else {
1812 		if (!inner_flushed)
1813 			UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
1814 					 proto == IPPROTO_UDPLITE);
1815 		consume_skb(skb);
1816 	}
1817 	return 0;
1818 }
1819 
1820 /* Initialize UDP checksum. If exited with zero value (success),
1821  * CHECKSUM_UNNECESSARY means, that no more checks are required.
1822  * Otherwise, csum completion requires chacksumming packet body,
1823  * including udp header and folding it to skb->csum.
1824  */
1825 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1826 				 int proto)
1827 {
1828 	int err;
1829 
1830 	UDP_SKB_CB(skb)->partial_cov = 0;
1831 	UDP_SKB_CB(skb)->cscov = skb->len;
1832 
1833 	if (proto == IPPROTO_UDPLITE) {
1834 		err = udplite_checksum_init(skb, uh);
1835 		if (err)
1836 			return err;
1837 	}
1838 
1839 	return skb_checksum_init_zero_check(skb, proto, uh->check,
1840 					    inet_compute_pseudo);
1841 }
1842 
1843 /*
1844  *	All we need to do is get the socket, and then do a checksum.
1845  */
1846 
1847 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1848 		   int proto)
1849 {
1850 	struct sock *sk;
1851 	struct udphdr *uh;
1852 	unsigned short ulen;
1853 	struct rtable *rt = skb_rtable(skb);
1854 	__be32 saddr, daddr;
1855 	struct net *net = dev_net(skb->dev);
1856 
1857 	/*
1858 	 *  Validate the packet.
1859 	 */
1860 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1861 		goto drop;		/* No space for header. */
1862 
1863 	uh   = udp_hdr(skb);
1864 	ulen = ntohs(uh->len);
1865 	saddr = ip_hdr(skb)->saddr;
1866 	daddr = ip_hdr(skb)->daddr;
1867 
1868 	if (ulen > skb->len)
1869 		goto short_packet;
1870 
1871 	if (proto == IPPROTO_UDP) {
1872 		/* UDP validates ulen. */
1873 		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1874 			goto short_packet;
1875 		uh = udp_hdr(skb);
1876 	}
1877 
1878 	if (udp4_csum_init(skb, uh, proto))
1879 		goto csum_error;
1880 
1881 	sk = skb_steal_sock(skb);
1882 	if (sk) {
1883 		struct dst_entry *dst = skb_dst(skb);
1884 		int ret;
1885 
1886 		if (unlikely(sk->sk_rx_dst != dst))
1887 			udp_sk_rx_dst_set(sk, dst);
1888 
1889 		ret = udp_queue_rcv_skb(sk, skb);
1890 		sock_put(sk);
1891 		/* a return value > 0 means to resubmit the input, but
1892 		 * it wants the return to be -protocol, or 0
1893 		 */
1894 		if (ret > 0)
1895 			return -ret;
1896 		return 0;
1897 	}
1898 
1899 	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1900 		return __udp4_lib_mcast_deliver(net, skb, uh,
1901 						saddr, daddr, udptable, proto);
1902 
1903 	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1904 	if (sk) {
1905 		int ret;
1906 
1907 		if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1908 			skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1909 						 inet_compute_pseudo);
1910 
1911 		ret = udp_queue_rcv_skb(sk, skb);
1912 		sock_put(sk);
1913 
1914 		/* a return value > 0 means to resubmit the input, but
1915 		 * it wants the return to be -protocol, or 0
1916 		 */
1917 		if (ret > 0)
1918 			return -ret;
1919 		return 0;
1920 	}
1921 
1922 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1923 		goto drop;
1924 	nf_reset(skb);
1925 
1926 	/* No socket. Drop packet silently, if checksum is wrong */
1927 	if (udp_lib_checksum_complete(skb))
1928 		goto csum_error;
1929 
1930 	UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1931 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1932 
1933 	/*
1934 	 * Hmm.  We got an UDP packet to a port to which we
1935 	 * don't wanna listen.  Ignore it.
1936 	 */
1937 	kfree_skb(skb);
1938 	return 0;
1939 
1940 short_packet:
1941 	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1942 			    proto == IPPROTO_UDPLITE ? "Lite" : "",
1943 			    &saddr, ntohs(uh->source),
1944 			    ulen, skb->len,
1945 			    &daddr, ntohs(uh->dest));
1946 	goto drop;
1947 
1948 csum_error:
1949 	/*
1950 	 * RFC1122: OK.  Discards the bad packet silently (as far as
1951 	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1952 	 */
1953 	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1954 			    proto == IPPROTO_UDPLITE ? "Lite" : "",
1955 			    &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1956 			    ulen);
1957 	UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1958 drop:
1959 	UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1960 	kfree_skb(skb);
1961 	return 0;
1962 }
1963 
1964 /* We can only early demux multicast if there is a single matching socket.
1965  * If more than one socket found returns NULL
1966  */
1967 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1968 						  __be16 loc_port, __be32 loc_addr,
1969 						  __be16 rmt_port, __be32 rmt_addr,
1970 						  int dif)
1971 {
1972 	struct sock *sk, *result;
1973 	struct hlist_nulls_node *node;
1974 	unsigned short hnum = ntohs(loc_port);
1975 	unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask);
1976 	struct udp_hslot *hslot = &udp_table.hash[slot];
1977 
1978 	/* Do not bother scanning a too big list */
1979 	if (hslot->count > 10)
1980 		return NULL;
1981 
1982 	rcu_read_lock();
1983 begin:
1984 	count = 0;
1985 	result = NULL;
1986 	sk_nulls_for_each_rcu(sk, node, &hslot->head) {
1987 		if (__udp_is_mcast_sock(net, sk,
1988 					loc_port, loc_addr,
1989 					rmt_port, rmt_addr,
1990 					dif, hnum)) {
1991 			result = sk;
1992 			++count;
1993 		}
1994 	}
1995 	/*
1996 	 * if the nulls value we got at the end of this lookup is
1997 	 * not the expected one, we must restart lookup.
1998 	 * We probably met an item that was moved to another chain.
1999 	 */
2000 	if (get_nulls_value(node) != slot)
2001 		goto begin;
2002 
2003 	if (result) {
2004 		if (count != 1 ||
2005 		    unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
2006 			result = NULL;
2007 		else if (unlikely(!__udp_is_mcast_sock(net, result,
2008 						       loc_port, loc_addr,
2009 						       rmt_port, rmt_addr,
2010 						       dif, hnum))) {
2011 			sock_put(result);
2012 			result = NULL;
2013 		}
2014 	}
2015 	rcu_read_unlock();
2016 	return result;
2017 }
2018 
2019 /* For unicast we should only early demux connected sockets or we can
2020  * break forwarding setups.  The chains here can be long so only check
2021  * if the first socket is an exact match and if not move on.
2022  */
2023 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2024 					    __be16 loc_port, __be32 loc_addr,
2025 					    __be16 rmt_port, __be32 rmt_addr,
2026 					    int dif)
2027 {
2028 	struct sock *sk, *result;
2029 	struct hlist_nulls_node *node;
2030 	unsigned short hnum = ntohs(loc_port);
2031 	unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
2032 	unsigned int slot2 = hash2 & udp_table.mask;
2033 	struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2034 	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2035 	const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2036 
2037 	rcu_read_lock();
2038 	result = NULL;
2039 	udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
2040 		if (INET_MATCH(sk, net, acookie,
2041 			       rmt_addr, loc_addr, ports, dif))
2042 			result = sk;
2043 		/* Only check first socket in chain */
2044 		break;
2045 	}
2046 
2047 	if (result) {
2048 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
2049 			result = NULL;
2050 		else if (unlikely(!INET_MATCH(sk, net, acookie,
2051 					      rmt_addr, loc_addr,
2052 					      ports, dif))) {
2053 			sock_put(result);
2054 			result = NULL;
2055 		}
2056 	}
2057 	rcu_read_unlock();
2058 	return result;
2059 }
2060 
2061 void udp_v4_early_demux(struct sk_buff *skb)
2062 {
2063 	struct net *net = dev_net(skb->dev);
2064 	const struct iphdr *iph;
2065 	const struct udphdr *uh;
2066 	struct sock *sk;
2067 	struct dst_entry *dst;
2068 	int dif = skb->dev->ifindex;
2069 	int ours;
2070 
2071 	/* validate the packet */
2072 	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2073 		return;
2074 
2075 	iph = ip_hdr(skb);
2076 	uh = udp_hdr(skb);
2077 
2078 	if (skb->pkt_type == PACKET_BROADCAST ||
2079 	    skb->pkt_type == PACKET_MULTICAST) {
2080 		struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
2081 
2082 		if (!in_dev)
2083 			return;
2084 
2085 		ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2086 				       iph->protocol);
2087 		if (!ours)
2088 			return;
2089 		sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2090 						   uh->source, iph->saddr, dif);
2091 	} else if (skb->pkt_type == PACKET_HOST) {
2092 		sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2093 					     uh->source, iph->saddr, dif);
2094 	} else {
2095 		return;
2096 	}
2097 
2098 	if (!sk)
2099 		return;
2100 
2101 	skb->sk = sk;
2102 	skb->destructor = sock_efree;
2103 	dst = READ_ONCE(sk->sk_rx_dst);
2104 
2105 	if (dst)
2106 		dst = dst_check(dst, 0);
2107 	if (dst) {
2108 		/* DST_NOCACHE can not be used without taking a reference */
2109 		if (dst->flags & DST_NOCACHE) {
2110 			if (likely(atomic_inc_not_zero(&dst->__refcnt)))
2111 				skb_dst_set(skb, dst);
2112 		} else {
2113 			skb_dst_set_noref(skb, dst);
2114 		}
2115 	}
2116 }
2117 
2118 int udp_rcv(struct sk_buff *skb)
2119 {
2120 	return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2121 }
2122 
2123 void udp_destroy_sock(struct sock *sk)
2124 {
2125 	struct udp_sock *up = udp_sk(sk);
2126 	bool slow = lock_sock_fast(sk);
2127 	udp_flush_pending_frames(sk);
2128 	unlock_sock_fast(sk, slow);
2129 	if (static_key_false(&udp_encap_needed) && up->encap_type) {
2130 		void (*encap_destroy)(struct sock *sk);
2131 		encap_destroy = ACCESS_ONCE(up->encap_destroy);
2132 		if (encap_destroy)
2133 			encap_destroy(sk);
2134 	}
2135 }
2136 
2137 /*
2138  *	Socket option code for UDP
2139  */
2140 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2141 		       char __user *optval, unsigned int optlen,
2142 		       int (*push_pending_frames)(struct sock *))
2143 {
2144 	struct udp_sock *up = udp_sk(sk);
2145 	int val, valbool;
2146 	int err = 0;
2147 	int is_udplite = IS_UDPLITE(sk);
2148 
2149 	if (optlen < sizeof(int))
2150 		return -EINVAL;
2151 
2152 	if (get_user(val, (int __user *)optval))
2153 		return -EFAULT;
2154 
2155 	valbool = val ? 1 : 0;
2156 
2157 	switch (optname) {
2158 	case UDP_CORK:
2159 		if (val != 0) {
2160 			up->corkflag = 1;
2161 		} else {
2162 			up->corkflag = 0;
2163 			lock_sock(sk);
2164 			push_pending_frames(sk);
2165 			release_sock(sk);
2166 		}
2167 		break;
2168 
2169 	case UDP_ENCAP:
2170 		switch (val) {
2171 		case 0:
2172 		case UDP_ENCAP_ESPINUDP:
2173 		case UDP_ENCAP_ESPINUDP_NON_IKE:
2174 			up->encap_rcv = xfrm4_udp_encap_rcv;
2175 			/* FALLTHROUGH */
2176 		case UDP_ENCAP_L2TPINUDP:
2177 			up->encap_type = val;
2178 			udp_encap_enable();
2179 			break;
2180 		default:
2181 			err = -ENOPROTOOPT;
2182 			break;
2183 		}
2184 		break;
2185 
2186 	case UDP_NO_CHECK6_TX:
2187 		up->no_check6_tx = valbool;
2188 		break;
2189 
2190 	case UDP_NO_CHECK6_RX:
2191 		up->no_check6_rx = valbool;
2192 		break;
2193 
2194 	/*
2195 	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
2196 	 */
2197 	/* The sender sets actual checksum coverage length via this option.
2198 	 * The case coverage > packet length is handled by send module. */
2199 	case UDPLITE_SEND_CSCOV:
2200 		if (!is_udplite)         /* Disable the option on UDP sockets */
2201 			return -ENOPROTOOPT;
2202 		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2203 			val = 8;
2204 		else if (val > USHRT_MAX)
2205 			val = USHRT_MAX;
2206 		up->pcslen = val;
2207 		up->pcflag |= UDPLITE_SEND_CC;
2208 		break;
2209 
2210 	/* The receiver specifies a minimum checksum coverage value. To make
2211 	 * sense, this should be set to at least 8 (as done below). If zero is
2212 	 * used, this again means full checksum coverage.                     */
2213 	case UDPLITE_RECV_CSCOV:
2214 		if (!is_udplite)         /* Disable the option on UDP sockets */
2215 			return -ENOPROTOOPT;
2216 		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2217 			val = 8;
2218 		else if (val > USHRT_MAX)
2219 			val = USHRT_MAX;
2220 		up->pcrlen = val;
2221 		up->pcflag |= UDPLITE_RECV_CC;
2222 		break;
2223 
2224 	default:
2225 		err = -ENOPROTOOPT;
2226 		break;
2227 	}
2228 
2229 	return err;
2230 }
2231 EXPORT_SYMBOL(udp_lib_setsockopt);
2232 
2233 int udp_setsockopt(struct sock *sk, int level, int optname,
2234 		   char __user *optval, unsigned int optlen)
2235 {
2236 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2237 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2238 					  udp_push_pending_frames);
2239 	return ip_setsockopt(sk, level, optname, optval, optlen);
2240 }
2241 
2242 #ifdef CONFIG_COMPAT
2243 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2244 			  char __user *optval, unsigned int optlen)
2245 {
2246 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2247 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2248 					  udp_push_pending_frames);
2249 	return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2250 }
2251 #endif
2252 
2253 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2254 		       char __user *optval, int __user *optlen)
2255 {
2256 	struct udp_sock *up = udp_sk(sk);
2257 	int val, len;
2258 
2259 	if (get_user(len, optlen))
2260 		return -EFAULT;
2261 
2262 	len = min_t(unsigned int, len, sizeof(int));
2263 
2264 	if (len < 0)
2265 		return -EINVAL;
2266 
2267 	switch (optname) {
2268 	case UDP_CORK:
2269 		val = up->corkflag;
2270 		break;
2271 
2272 	case UDP_ENCAP:
2273 		val = up->encap_type;
2274 		break;
2275 
2276 	case UDP_NO_CHECK6_TX:
2277 		val = up->no_check6_tx;
2278 		break;
2279 
2280 	case UDP_NO_CHECK6_RX:
2281 		val = up->no_check6_rx;
2282 		break;
2283 
2284 	/* The following two cannot be changed on UDP sockets, the return is
2285 	 * always 0 (which corresponds to the full checksum coverage of UDP). */
2286 	case UDPLITE_SEND_CSCOV:
2287 		val = up->pcslen;
2288 		break;
2289 
2290 	case UDPLITE_RECV_CSCOV:
2291 		val = up->pcrlen;
2292 		break;
2293 
2294 	default:
2295 		return -ENOPROTOOPT;
2296 	}
2297 
2298 	if (put_user(len, optlen))
2299 		return -EFAULT;
2300 	if (copy_to_user(optval, &val, len))
2301 		return -EFAULT;
2302 	return 0;
2303 }
2304 EXPORT_SYMBOL(udp_lib_getsockopt);
2305 
2306 int udp_getsockopt(struct sock *sk, int level, int optname,
2307 		   char __user *optval, int __user *optlen)
2308 {
2309 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2310 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2311 	return ip_getsockopt(sk, level, optname, optval, optlen);
2312 }
2313 
2314 #ifdef CONFIG_COMPAT
2315 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2316 				 char __user *optval, int __user *optlen)
2317 {
2318 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2319 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2320 	return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2321 }
2322 #endif
2323 /**
2324  * 	udp_poll - wait for a UDP event.
2325  *	@file - file struct
2326  *	@sock - socket
2327  *	@wait - poll table
2328  *
2329  *	This is same as datagram poll, except for the special case of
2330  *	blocking sockets. If application is using a blocking fd
2331  *	and a packet with checksum error is in the queue;
2332  *	then it could get return from select indicating data available
2333  *	but then block when reading it. Add special case code
2334  *	to work around these arguably broken applications.
2335  */
2336 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2337 {
2338 	unsigned int mask = datagram_poll(file, sock, wait);
2339 	struct sock *sk = sock->sk;
2340 
2341 	sock_rps_record_flow(sk);
2342 
2343 	/* Check for false positives due to checksum errors */
2344 	if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2345 	    !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2346 		mask &= ~(POLLIN | POLLRDNORM);
2347 
2348 	return mask;
2349 
2350 }
2351 EXPORT_SYMBOL(udp_poll);
2352 
2353 struct proto udp_prot = {
2354 	.name		   = "UDP",
2355 	.owner		   = THIS_MODULE,
2356 	.close		   = udp_lib_close,
2357 	.connect	   = ip4_datagram_connect,
2358 	.disconnect	   = udp_disconnect,
2359 	.ioctl		   = udp_ioctl,
2360 	.destroy	   = udp_destroy_sock,
2361 	.setsockopt	   = udp_setsockopt,
2362 	.getsockopt	   = udp_getsockopt,
2363 	.sendmsg	   = udp_sendmsg,
2364 	.recvmsg	   = udp_recvmsg,
2365 	.sendpage	   = udp_sendpage,
2366 	.backlog_rcv	   = __udp_queue_rcv_skb,
2367 	.release_cb	   = ip4_datagram_release_cb,
2368 	.hash		   = udp_lib_hash,
2369 	.unhash		   = udp_lib_unhash,
2370 	.rehash		   = udp_v4_rehash,
2371 	.get_port	   = udp_v4_get_port,
2372 	.memory_allocated  = &udp_memory_allocated,
2373 	.sysctl_mem	   = sysctl_udp_mem,
2374 	.sysctl_wmem	   = &sysctl_udp_wmem_min,
2375 	.sysctl_rmem	   = &sysctl_udp_rmem_min,
2376 	.obj_size	   = sizeof(struct udp_sock),
2377 	.slab_flags	   = SLAB_DESTROY_BY_RCU,
2378 	.h.udp_table	   = &udp_table,
2379 #ifdef CONFIG_COMPAT
2380 	.compat_setsockopt = compat_udp_setsockopt,
2381 	.compat_getsockopt = compat_udp_getsockopt,
2382 #endif
2383 	.clear_sk	   = sk_prot_clear_portaddr_nulls,
2384 };
2385 EXPORT_SYMBOL(udp_prot);
2386 
2387 /* ------------------------------------------------------------------------ */
2388 #ifdef CONFIG_PROC_FS
2389 
2390 static struct sock *udp_get_first(struct seq_file *seq, int start)
2391 {
2392 	struct sock *sk;
2393 	struct udp_iter_state *state = seq->private;
2394 	struct net *net = seq_file_net(seq);
2395 
2396 	for (state->bucket = start; state->bucket <= state->udp_table->mask;
2397 	     ++state->bucket) {
2398 		struct hlist_nulls_node *node;
2399 		struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2400 
2401 		if (hlist_nulls_empty(&hslot->head))
2402 			continue;
2403 
2404 		spin_lock_bh(&hslot->lock);
2405 		sk_nulls_for_each(sk, node, &hslot->head) {
2406 			if (!net_eq(sock_net(sk), net))
2407 				continue;
2408 			if (sk->sk_family == state->family)
2409 				goto found;
2410 		}
2411 		spin_unlock_bh(&hslot->lock);
2412 	}
2413 	sk = NULL;
2414 found:
2415 	return sk;
2416 }
2417 
2418 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2419 {
2420 	struct udp_iter_state *state = seq->private;
2421 	struct net *net = seq_file_net(seq);
2422 
2423 	do {
2424 		sk = sk_nulls_next(sk);
2425 	} while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2426 
2427 	if (!sk) {
2428 		if (state->bucket <= state->udp_table->mask)
2429 			spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2430 		return udp_get_first(seq, state->bucket + 1);
2431 	}
2432 	return sk;
2433 }
2434 
2435 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2436 {
2437 	struct sock *sk = udp_get_first(seq, 0);
2438 
2439 	if (sk)
2440 		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2441 			--pos;
2442 	return pos ? NULL : sk;
2443 }
2444 
2445 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2446 {
2447 	struct udp_iter_state *state = seq->private;
2448 	state->bucket = MAX_UDP_PORTS;
2449 
2450 	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2451 }
2452 
2453 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2454 {
2455 	struct sock *sk;
2456 
2457 	if (v == SEQ_START_TOKEN)
2458 		sk = udp_get_idx(seq, 0);
2459 	else
2460 		sk = udp_get_next(seq, v);
2461 
2462 	++*pos;
2463 	return sk;
2464 }
2465 
2466 static void udp_seq_stop(struct seq_file *seq, void *v)
2467 {
2468 	struct udp_iter_state *state = seq->private;
2469 
2470 	if (state->bucket <= state->udp_table->mask)
2471 		spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2472 }
2473 
2474 int udp_seq_open(struct inode *inode, struct file *file)
2475 {
2476 	struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2477 	struct udp_iter_state *s;
2478 	int err;
2479 
2480 	err = seq_open_net(inode, file, &afinfo->seq_ops,
2481 			   sizeof(struct udp_iter_state));
2482 	if (err < 0)
2483 		return err;
2484 
2485 	s = ((struct seq_file *)file->private_data)->private;
2486 	s->family		= afinfo->family;
2487 	s->udp_table		= afinfo->udp_table;
2488 	return err;
2489 }
2490 EXPORT_SYMBOL(udp_seq_open);
2491 
2492 /* ------------------------------------------------------------------------ */
2493 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2494 {
2495 	struct proc_dir_entry *p;
2496 	int rc = 0;
2497 
2498 	afinfo->seq_ops.start		= udp_seq_start;
2499 	afinfo->seq_ops.next		= udp_seq_next;
2500 	afinfo->seq_ops.stop		= udp_seq_stop;
2501 
2502 	p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2503 			     afinfo->seq_fops, afinfo);
2504 	if (!p)
2505 		rc = -ENOMEM;
2506 	return rc;
2507 }
2508 EXPORT_SYMBOL(udp_proc_register);
2509 
2510 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2511 {
2512 	remove_proc_entry(afinfo->name, net->proc_net);
2513 }
2514 EXPORT_SYMBOL(udp_proc_unregister);
2515 
2516 /* ------------------------------------------------------------------------ */
2517 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2518 		int bucket)
2519 {
2520 	struct inet_sock *inet = inet_sk(sp);
2521 	__be32 dest = inet->inet_daddr;
2522 	__be32 src  = inet->inet_rcv_saddr;
2523 	__u16 destp	  = ntohs(inet->inet_dport);
2524 	__u16 srcp	  = ntohs(inet->inet_sport);
2525 
2526 	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2527 		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2528 		bucket, src, srcp, dest, destp, sp->sk_state,
2529 		sk_wmem_alloc_get(sp),
2530 		sk_rmem_alloc_get(sp),
2531 		0, 0L, 0,
2532 		from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2533 		0, sock_i_ino(sp),
2534 		atomic_read(&sp->sk_refcnt), sp,
2535 		atomic_read(&sp->sk_drops));
2536 }
2537 
2538 int udp4_seq_show(struct seq_file *seq, void *v)
2539 {
2540 	seq_setwidth(seq, 127);
2541 	if (v == SEQ_START_TOKEN)
2542 		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2543 			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2544 			   "inode ref pointer drops");
2545 	else {
2546 		struct udp_iter_state *state = seq->private;
2547 
2548 		udp4_format_sock(v, seq, state->bucket);
2549 	}
2550 	seq_pad(seq, '\n');
2551 	return 0;
2552 }
2553 
2554 static const struct file_operations udp_afinfo_seq_fops = {
2555 	.owner    = THIS_MODULE,
2556 	.open     = udp_seq_open,
2557 	.read     = seq_read,
2558 	.llseek   = seq_lseek,
2559 	.release  = seq_release_net
2560 };
2561 
2562 /* ------------------------------------------------------------------------ */
2563 static struct udp_seq_afinfo udp4_seq_afinfo = {
2564 	.name		= "udp",
2565 	.family		= AF_INET,
2566 	.udp_table	= &udp_table,
2567 	.seq_fops	= &udp_afinfo_seq_fops,
2568 	.seq_ops	= {
2569 		.show		= udp4_seq_show,
2570 	},
2571 };
2572 
2573 static int __net_init udp4_proc_init_net(struct net *net)
2574 {
2575 	return udp_proc_register(net, &udp4_seq_afinfo);
2576 }
2577 
2578 static void __net_exit udp4_proc_exit_net(struct net *net)
2579 {
2580 	udp_proc_unregister(net, &udp4_seq_afinfo);
2581 }
2582 
2583 static struct pernet_operations udp4_net_ops = {
2584 	.init = udp4_proc_init_net,
2585 	.exit = udp4_proc_exit_net,
2586 };
2587 
2588 int __init udp4_proc_init(void)
2589 {
2590 	return register_pernet_subsys(&udp4_net_ops);
2591 }
2592 
2593 void udp4_proc_exit(void)
2594 {
2595 	unregister_pernet_subsys(&udp4_net_ops);
2596 }
2597 #endif /* CONFIG_PROC_FS */
2598 
2599 static __initdata unsigned long uhash_entries;
2600 static int __init set_uhash_entries(char *str)
2601 {
2602 	ssize_t ret;
2603 
2604 	if (!str)
2605 		return 0;
2606 
2607 	ret = kstrtoul(str, 0, &uhash_entries);
2608 	if (ret)
2609 		return 0;
2610 
2611 	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2612 		uhash_entries = UDP_HTABLE_SIZE_MIN;
2613 	return 1;
2614 }
2615 __setup("uhash_entries=", set_uhash_entries);
2616 
2617 void __init udp_table_init(struct udp_table *table, const char *name)
2618 {
2619 	unsigned int i;
2620 
2621 	table->hash = alloc_large_system_hash(name,
2622 					      2 * sizeof(struct udp_hslot),
2623 					      uhash_entries,
2624 					      21, /* one slot per 2 MB */
2625 					      0,
2626 					      &table->log,
2627 					      &table->mask,
2628 					      UDP_HTABLE_SIZE_MIN,
2629 					      64 * 1024);
2630 
2631 	table->hash2 = table->hash + (table->mask + 1);
2632 	for (i = 0; i <= table->mask; i++) {
2633 		INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2634 		table->hash[i].count = 0;
2635 		spin_lock_init(&table->hash[i].lock);
2636 	}
2637 	for (i = 0; i <= table->mask; i++) {
2638 		INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2639 		table->hash2[i].count = 0;
2640 		spin_lock_init(&table->hash2[i].lock);
2641 	}
2642 }
2643 
2644 u32 udp_flow_hashrnd(void)
2645 {
2646 	static u32 hashrnd __read_mostly;
2647 
2648 	net_get_random_once(&hashrnd, sizeof(hashrnd));
2649 
2650 	return hashrnd;
2651 }
2652 EXPORT_SYMBOL(udp_flow_hashrnd);
2653 
2654 void __init udp_init(void)
2655 {
2656 	unsigned long limit;
2657 
2658 	udp_table_init(&udp_table, "UDP");
2659 	limit = nr_free_buffer_pages() / 8;
2660 	limit = max(limit, 128UL);
2661 	sysctl_udp_mem[0] = limit / 4 * 3;
2662 	sysctl_udp_mem[1] = limit;
2663 	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2664 
2665 	sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2666 	sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2667 }
2668