xref: /linux/net/ipv4/udp.c (revision 067012974c8ae31a8886046df082aeba93592972)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		The User Datagram Protocol (UDP).
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
12  *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
13  *		Hirokazu Takahashi, <taka@valinux.co.jp>
14  *
15  * Fixes:
16  *		Alan Cox	:	verify_area() calls
17  *		Alan Cox	: 	stopped close while in use off icmp
18  *					messages. Not a fix but a botch that
19  *					for udp at least is 'valid'.
20  *		Alan Cox	:	Fixed icmp handling properly
21  *		Alan Cox	: 	Correct error for oversized datagrams
22  *		Alan Cox	:	Tidied select() semantics.
23  *		Alan Cox	:	udp_err() fixed properly, also now
24  *					select and read wake correctly on errors
25  *		Alan Cox	:	udp_send verify_area moved to avoid mem leak
26  *		Alan Cox	:	UDP can count its memory
27  *		Alan Cox	:	send to an unknown connection causes
28  *					an ECONNREFUSED off the icmp, but
29  *					does NOT close.
30  *		Alan Cox	:	Switched to new sk_buff handlers. No more backlog!
31  *		Alan Cox	:	Using generic datagram code. Even smaller and the PEEK
32  *					bug no longer crashes it.
33  *		Fred Van Kempen	: 	Net2e support for sk->broadcast.
34  *		Alan Cox	:	Uses skb_free_datagram
35  *		Alan Cox	:	Added get/set sockopt support.
36  *		Alan Cox	:	Broadcasting without option set returns EACCES.
37  *		Alan Cox	:	No wakeup calls. Instead we now use the callbacks.
38  *		Alan Cox	:	Use ip_tos and ip_ttl
39  *		Alan Cox	:	SNMP Mibs
40  *		Alan Cox	:	MSG_DONTROUTE, and 0.0.0.0 support.
41  *		Matt Dillon	:	UDP length checks.
42  *		Alan Cox	:	Smarter af_inet used properly.
43  *		Alan Cox	:	Use new kernel side addressing.
44  *		Alan Cox	:	Incorrect return on truncated datagram receive.
45  *	Arnt Gulbrandsen 	:	New udp_send and stuff
46  *		Alan Cox	:	Cache last socket
47  *		Alan Cox	:	Route cache
48  *		Jon Peatfield	:	Minor efficiency fix to sendto().
49  *		Mike Shaver	:	RFC1122 checks.
50  *		Alan Cox	:	Nonblocking error fix.
51  *	Willy Konynenberg	:	Transparent proxying support.
52  *		Mike McLagan	:	Routing by source
53  *		David S. Miller	:	New socket lookup architecture.
54  *					Last socket cache retained as it
55  *					does have a high hit rate.
56  *		Olaf Kirch	:	Don't linearise iovec on sendmsg.
57  *		Andi Kleen	:	Some cleanups, cache destination entry
58  *					for connect.
59  *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
60  *		Melvin Smith	:	Check msg_name not msg_namelen in sendto(),
61  *					return ENOTCONN for unconnected sockets (POSIX)
62  *		Janos Farkas	:	don't deliver multi/broadcasts to a different
63  *					bound-to-device socket
64  *	Hirokazu Takahashi	:	HW checksumming for outgoing UDP
65  *					datagrams.
66  *	Hirokazu Takahashi	:	sendfile() on UDP works now.
67  *		Arnaldo C. Melo :	convert /proc/net/udp to seq_file
68  *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
69  *	Alexey Kuznetsov:		allow both IPv4 and IPv6 sockets to bind
70  *					a single port at the same time.
71  *	Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
72  *	James Chapman		:	Add L2TP encapsulation type.
73  */
74 
75 #define pr_fmt(fmt) "UDP: " fmt
76 
77 #include <linux/uaccess.h>
78 #include <asm/ioctls.h>
79 #include <linux/memblock.h>
80 #include <linux/highmem.h>
81 #include <linux/swap.h>
82 #include <linux/types.h>
83 #include <linux/fcntl.h>
84 #include <linux/module.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/igmp.h>
88 #include <linux/inetdevice.h>
89 #include <linux/in.h>
90 #include <linux/errno.h>
91 #include <linux/timer.h>
92 #include <linux/mm.h>
93 #include <linux/inet.h>
94 #include <linux/netdevice.h>
95 #include <linux/slab.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/net_namespace.h>
101 #include <net/icmp.h>
102 #include <net/inet_hashtables.h>
103 #include <net/ip_tunnels.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <trace/events/udp.h>
108 #include <linux/static_key.h>
109 #include <linux/btf_ids.h>
110 #include <trace/events/skb.h>
111 #include <net/busy_poll.h>
112 #include "udp_impl.h"
113 #include <net/sock_reuseport.h>
114 #include <net/addrconf.h>
115 #include <net/udp_tunnel.h>
116 #if IS_ENABLED(CONFIG_IPV6)
117 #include <net/ipv6_stubs.h>
118 #endif
119 
120 struct udp_table udp_table __read_mostly;
121 EXPORT_SYMBOL(udp_table);
122 
123 long sysctl_udp_mem[3] __read_mostly;
124 EXPORT_SYMBOL(sysctl_udp_mem);
125 
126 atomic_long_t udp_memory_allocated;
127 EXPORT_SYMBOL(udp_memory_allocated);
128 
129 #define MAX_UDP_PORTS 65536
130 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
131 
132 static int udp_lib_lport_inuse(struct net *net, __u16 num,
133 			       const struct udp_hslot *hslot,
134 			       unsigned long *bitmap,
135 			       struct sock *sk, unsigned int log)
136 {
137 	struct sock *sk2;
138 	kuid_t uid = sock_i_uid(sk);
139 
140 	sk_for_each(sk2, &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 		    inet_rcv_saddr_equal(sk, sk2, true)) {
148 			if (sk2->sk_reuseport && sk->sk_reuseport &&
149 			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
150 			    uid_eq(uid, sock_i_uid(sk2))) {
151 				if (!bitmap)
152 					return 0;
153 			} else {
154 				if (!bitmap)
155 					return 1;
156 				__set_bit(udp_sk(sk2)->udp_port_hash >> log,
157 					  bitmap);
158 			}
159 		}
160 	}
161 	return 0;
162 }
163 
164 /*
165  * Note: we still hold spinlock of primary hash chain, so no other writer
166  * can insert/delete a socket with local_port == num
167  */
168 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
169 				struct udp_hslot *hslot2,
170 				struct sock *sk)
171 {
172 	struct sock *sk2;
173 	kuid_t uid = sock_i_uid(sk);
174 	int res = 0;
175 
176 	spin_lock(&hslot2->lock);
177 	udp_portaddr_for_each_entry(sk2, &hslot2->head) {
178 		if (net_eq(sock_net(sk2), net) &&
179 		    sk2 != sk &&
180 		    (udp_sk(sk2)->udp_port_hash == num) &&
181 		    (!sk2->sk_reuse || !sk->sk_reuse) &&
182 		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
183 		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
184 		    inet_rcv_saddr_equal(sk, sk2, true)) {
185 			if (sk2->sk_reuseport && sk->sk_reuseport &&
186 			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
187 			    uid_eq(uid, sock_i_uid(sk2))) {
188 				res = 0;
189 			} else {
190 				res = 1;
191 			}
192 			break;
193 		}
194 	}
195 	spin_unlock(&hslot2->lock);
196 	return res;
197 }
198 
199 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
200 {
201 	struct net *net = sock_net(sk);
202 	kuid_t uid = sock_i_uid(sk);
203 	struct sock *sk2;
204 
205 	sk_for_each(sk2, &hslot->head) {
206 		if (net_eq(sock_net(sk2), net) &&
207 		    sk2 != sk &&
208 		    sk2->sk_family == sk->sk_family &&
209 		    ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
210 		    (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
211 		    (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
212 		    sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
213 		    inet_rcv_saddr_equal(sk, sk2, false)) {
214 			return reuseport_add_sock(sk, sk2,
215 						  inet_rcv_saddr_any(sk));
216 		}
217 	}
218 
219 	return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
220 }
221 
222 /**
223  *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
224  *
225  *  @sk:          socket struct in question
226  *  @snum:        port number to look up
227  *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
228  *                   with NULL address
229  */
230 int udp_lib_get_port(struct sock *sk, unsigned short snum,
231 		     unsigned int hash2_nulladdr)
232 {
233 	struct udp_hslot *hslot, *hslot2;
234 	struct udp_table *udptable = sk->sk_prot->h.udp_table;
235 	int    error = 1;
236 	struct net *net = sock_net(sk);
237 
238 	if (!snum) {
239 		int low, high, remaining;
240 		unsigned int rand;
241 		unsigned short first, last;
242 		DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
243 
244 		inet_get_local_port_range(net, &low, &high);
245 		remaining = (high - low) + 1;
246 
247 		rand = prandom_u32();
248 		first = reciprocal_scale(rand, remaining) + low;
249 		/*
250 		 * force rand to be an odd multiple of UDP_HTABLE_SIZE
251 		 */
252 		rand = (rand | 1) * (udptable->mask + 1);
253 		last = first + udptable->mask + 1;
254 		do {
255 			hslot = udp_hashslot(udptable, net, first);
256 			bitmap_zero(bitmap, PORTS_PER_CHAIN);
257 			spin_lock_bh(&hslot->lock);
258 			udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
259 					    udptable->log);
260 
261 			snum = first;
262 			/*
263 			 * Iterate on all possible values of snum for this hash.
264 			 * Using steps of an odd multiple of UDP_HTABLE_SIZE
265 			 * give us randomization and full range coverage.
266 			 */
267 			do {
268 				if (low <= snum && snum <= high &&
269 				    !test_bit(snum >> udptable->log, bitmap) &&
270 				    !inet_is_local_reserved_port(net, snum))
271 					goto found;
272 				snum += rand;
273 			} while (snum != first);
274 			spin_unlock_bh(&hslot->lock);
275 			cond_resched();
276 		} while (++first != last);
277 		goto fail;
278 	} else {
279 		hslot = udp_hashslot(udptable, net, snum);
280 		spin_lock_bh(&hslot->lock);
281 		if (hslot->count > 10) {
282 			int exist;
283 			unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
284 
285 			slot2          &= udptable->mask;
286 			hash2_nulladdr &= udptable->mask;
287 
288 			hslot2 = udp_hashslot2(udptable, slot2);
289 			if (hslot->count < hslot2->count)
290 				goto scan_primary_hash;
291 
292 			exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
293 			if (!exist && (hash2_nulladdr != slot2)) {
294 				hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
295 				exist = udp_lib_lport_inuse2(net, snum, hslot2,
296 							     sk);
297 			}
298 			if (exist)
299 				goto fail_unlock;
300 			else
301 				goto found;
302 		}
303 scan_primary_hash:
304 		if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
305 			goto fail_unlock;
306 	}
307 found:
308 	inet_sk(sk)->inet_num = snum;
309 	udp_sk(sk)->udp_port_hash = snum;
310 	udp_sk(sk)->udp_portaddr_hash ^= snum;
311 	if (sk_unhashed(sk)) {
312 		if (sk->sk_reuseport &&
313 		    udp_reuseport_add_sock(sk, hslot)) {
314 			inet_sk(sk)->inet_num = 0;
315 			udp_sk(sk)->udp_port_hash = 0;
316 			udp_sk(sk)->udp_portaddr_hash ^= snum;
317 			goto fail_unlock;
318 		}
319 
320 		sk_add_node_rcu(sk, &hslot->head);
321 		hslot->count++;
322 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
323 
324 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
325 		spin_lock(&hslot2->lock);
326 		if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
327 		    sk->sk_family == AF_INET6)
328 			hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
329 					   &hslot2->head);
330 		else
331 			hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
332 					   &hslot2->head);
333 		hslot2->count++;
334 		spin_unlock(&hslot2->lock);
335 	}
336 	sock_set_flag(sk, SOCK_RCU_FREE);
337 	error = 0;
338 fail_unlock:
339 	spin_unlock_bh(&hslot->lock);
340 fail:
341 	return error;
342 }
343 EXPORT_SYMBOL(udp_lib_get_port);
344 
345 int udp_v4_get_port(struct sock *sk, unsigned short snum)
346 {
347 	unsigned int hash2_nulladdr =
348 		ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
349 	unsigned int hash2_partial =
350 		ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
351 
352 	/* precompute partial secondary hash */
353 	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
354 	return udp_lib_get_port(sk, snum, hash2_nulladdr);
355 }
356 
357 static int compute_score(struct sock *sk, struct net *net,
358 			 __be32 saddr, __be16 sport,
359 			 __be32 daddr, unsigned short hnum,
360 			 int dif, int sdif)
361 {
362 	int score;
363 	struct inet_sock *inet;
364 	bool dev_match;
365 
366 	if (!net_eq(sock_net(sk), net) ||
367 	    udp_sk(sk)->udp_port_hash != hnum ||
368 	    ipv6_only_sock(sk))
369 		return -1;
370 
371 	if (sk->sk_rcv_saddr != daddr)
372 		return -1;
373 
374 	score = (sk->sk_family == PF_INET) ? 2 : 1;
375 
376 	inet = inet_sk(sk);
377 	if (inet->inet_daddr) {
378 		if (inet->inet_daddr != saddr)
379 			return -1;
380 		score += 4;
381 	}
382 
383 	if (inet->inet_dport) {
384 		if (inet->inet_dport != sport)
385 			return -1;
386 		score += 4;
387 	}
388 
389 	dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
390 					dif, sdif);
391 	if (!dev_match)
392 		return -1;
393 	score += 4;
394 
395 	if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
396 		score++;
397 	return score;
398 }
399 
400 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
401 		       const __u16 lport, const __be32 faddr,
402 		       const __be16 fport)
403 {
404 	static u32 udp_ehash_secret __read_mostly;
405 
406 	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
407 
408 	return __inet_ehashfn(laddr, lport, faddr, fport,
409 			      udp_ehash_secret + net_hash_mix(net));
410 }
411 
412 static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
413 				     struct sk_buff *skb,
414 				     __be32 saddr, __be16 sport,
415 				     __be32 daddr, unsigned short hnum)
416 {
417 	struct sock *reuse_sk = NULL;
418 	u32 hash;
419 
420 	if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
421 		hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
422 		reuse_sk = reuseport_select_sock(sk, hash, skb,
423 						 sizeof(struct udphdr));
424 	}
425 	return reuse_sk;
426 }
427 
428 /* called with rcu_read_lock() */
429 static struct sock *udp4_lib_lookup2(struct net *net,
430 				     __be32 saddr, __be16 sport,
431 				     __be32 daddr, unsigned int hnum,
432 				     int dif, int sdif,
433 				     struct udp_hslot *hslot2,
434 				     struct sk_buff *skb)
435 {
436 	struct sock *sk, *result;
437 	int score, badness;
438 
439 	result = NULL;
440 	badness = 0;
441 	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
442 		score = compute_score(sk, net, saddr, sport,
443 				      daddr, hnum, dif, sdif);
444 		if (score > badness) {
445 			result = lookup_reuseport(net, sk, skb,
446 						  saddr, sport, daddr, hnum);
447 			/* Fall back to scoring if group has connections */
448 			if (result && !reuseport_has_conns(sk, false))
449 				return result;
450 
451 			result = result ? : sk;
452 			badness = score;
453 		}
454 	}
455 	return result;
456 }
457 
458 static struct sock *udp4_lookup_run_bpf(struct net *net,
459 					struct udp_table *udptable,
460 					struct sk_buff *skb,
461 					__be32 saddr, __be16 sport,
462 					__be32 daddr, u16 hnum)
463 {
464 	struct sock *sk, *reuse_sk;
465 	bool no_reuseport;
466 
467 	if (udptable != &udp_table)
468 		return NULL; /* only UDP is supported */
469 
470 	no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP,
471 					    saddr, sport, daddr, hnum, &sk);
472 	if (no_reuseport || IS_ERR_OR_NULL(sk))
473 		return sk;
474 
475 	reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
476 	if (reuse_sk)
477 		sk = reuse_sk;
478 	return sk;
479 }
480 
481 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
482  * harder than this. -DaveM
483  */
484 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
485 		__be16 sport, __be32 daddr, __be16 dport, int dif,
486 		int sdif, struct udp_table *udptable, struct sk_buff *skb)
487 {
488 	unsigned short hnum = ntohs(dport);
489 	unsigned int hash2, slot2;
490 	struct udp_hslot *hslot2;
491 	struct sock *result, *sk;
492 
493 	hash2 = ipv4_portaddr_hash(net, daddr, hnum);
494 	slot2 = hash2 & udptable->mask;
495 	hslot2 = &udptable->hash2[slot2];
496 
497 	/* Lookup connected or non-wildcard socket */
498 	result = udp4_lib_lookup2(net, saddr, sport,
499 				  daddr, hnum, dif, sdif,
500 				  hslot2, skb);
501 	if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
502 		goto done;
503 
504 	/* Lookup redirect from BPF */
505 	if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
506 		sk = udp4_lookup_run_bpf(net, udptable, skb,
507 					 saddr, sport, daddr, hnum);
508 		if (sk) {
509 			result = sk;
510 			goto done;
511 		}
512 	}
513 
514 	/* Got non-wildcard socket or error on first lookup */
515 	if (result)
516 		goto done;
517 
518 	/* Lookup wildcard sockets */
519 	hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
520 	slot2 = hash2 & udptable->mask;
521 	hslot2 = &udptable->hash2[slot2];
522 
523 	result = udp4_lib_lookup2(net, saddr, sport,
524 				  htonl(INADDR_ANY), hnum, dif, sdif,
525 				  hslot2, skb);
526 done:
527 	if (IS_ERR(result))
528 		return NULL;
529 	return result;
530 }
531 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
532 
533 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
534 						 __be16 sport, __be16 dport,
535 						 struct udp_table *udptable)
536 {
537 	const struct iphdr *iph = ip_hdr(skb);
538 
539 	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
540 				 iph->daddr, dport, inet_iif(skb),
541 				 inet_sdif(skb), udptable, skb);
542 }
543 
544 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
545 				 __be16 sport, __be16 dport)
546 {
547 	const struct iphdr *iph = ip_hdr(skb);
548 
549 	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
550 				 iph->daddr, dport, inet_iif(skb),
551 				 inet_sdif(skb), &udp_table, NULL);
552 }
553 
554 /* Must be called under rcu_read_lock().
555  * Does increment socket refcount.
556  */
557 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
558 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
559 			     __be32 daddr, __be16 dport, int dif)
560 {
561 	struct sock *sk;
562 
563 	sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
564 			       dif, 0, &udp_table, NULL);
565 	if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
566 		sk = NULL;
567 	return sk;
568 }
569 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
570 #endif
571 
572 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
573 				       __be16 loc_port, __be32 loc_addr,
574 				       __be16 rmt_port, __be32 rmt_addr,
575 				       int dif, int sdif, unsigned short hnum)
576 {
577 	struct inet_sock *inet = inet_sk(sk);
578 
579 	if (!net_eq(sock_net(sk), net) ||
580 	    udp_sk(sk)->udp_port_hash != hnum ||
581 	    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
582 	    (inet->inet_dport != rmt_port && inet->inet_dport) ||
583 	    (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
584 	    ipv6_only_sock(sk) ||
585 	    !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
586 		return false;
587 	if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
588 		return false;
589 	return true;
590 }
591 
592 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
593 void udp_encap_enable(void)
594 {
595 	static_branch_inc(&udp_encap_needed_key);
596 }
597 EXPORT_SYMBOL(udp_encap_enable);
598 
599 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
600  * through error handlers in encapsulations looking for a match.
601  */
602 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
603 {
604 	int i;
605 
606 	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
607 		int (*handler)(struct sk_buff *skb, u32 info);
608 		const struct ip_tunnel_encap_ops *encap;
609 
610 		encap = rcu_dereference(iptun_encaps[i]);
611 		if (!encap)
612 			continue;
613 		handler = encap->err_handler;
614 		if (handler && !handler(skb, info))
615 			return 0;
616 	}
617 
618 	return -ENOENT;
619 }
620 
621 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
622  * reversing source and destination port: this will match tunnels that force the
623  * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
624  * lwtunnels might actually break this assumption by being configured with
625  * different destination ports on endpoints, in this case we won't be able to
626  * trace ICMP messages back to them.
627  *
628  * If this doesn't match any socket, probe tunnels with arbitrary destination
629  * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
630  * we've sent packets to won't necessarily match the local destination port.
631  *
632  * Then ask the tunnel implementation to match the error against a valid
633  * association.
634  *
635  * Return an error if we can't find a match, the socket if we need further
636  * processing, zero otherwise.
637  */
638 static struct sock *__udp4_lib_err_encap(struct net *net,
639 					 const struct iphdr *iph,
640 					 struct udphdr *uh,
641 					 struct udp_table *udptable,
642 					 struct sk_buff *skb, u32 info)
643 {
644 	int network_offset, transport_offset;
645 	struct sock *sk;
646 
647 	network_offset = skb_network_offset(skb);
648 	transport_offset = skb_transport_offset(skb);
649 
650 	/* Network header needs to point to the outer IPv4 header inside ICMP */
651 	skb_reset_network_header(skb);
652 
653 	/* Transport header needs to point to the UDP header */
654 	skb_set_transport_header(skb, iph->ihl << 2);
655 
656 	sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
657 			       iph->saddr, uh->dest, skb->dev->ifindex, 0,
658 			       udptable, NULL);
659 	if (sk) {
660 		int (*lookup)(struct sock *sk, struct sk_buff *skb);
661 		struct udp_sock *up = udp_sk(sk);
662 
663 		lookup = READ_ONCE(up->encap_err_lookup);
664 		if (!lookup || lookup(sk, skb))
665 			sk = NULL;
666 	}
667 
668 	if (!sk)
669 		sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
670 
671 	skb_set_transport_header(skb, transport_offset);
672 	skb_set_network_header(skb, network_offset);
673 
674 	return sk;
675 }
676 
677 /*
678  * This routine is called by the ICMP module when it gets some
679  * sort of error condition.  If err < 0 then the socket should
680  * be closed and the error returned to the user.  If err > 0
681  * it's just the icmp type << 8 | icmp code.
682  * Header points to the ip header of the error packet. We move
683  * on past this. Then (as it used to claim before adjustment)
684  * header points to the first 8 bytes of the udp header.  We need
685  * to find the appropriate port.
686  */
687 
688 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
689 {
690 	struct inet_sock *inet;
691 	const struct iphdr *iph = (const struct iphdr *)skb->data;
692 	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
693 	const int type = icmp_hdr(skb)->type;
694 	const int code = icmp_hdr(skb)->code;
695 	bool tunnel = false;
696 	struct sock *sk;
697 	int harderr;
698 	int err;
699 	struct net *net = dev_net(skb->dev);
700 
701 	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
702 			       iph->saddr, uh->source, skb->dev->ifindex,
703 			       inet_sdif(skb), udptable, NULL);
704 	if (!sk || udp_sk(sk)->encap_type) {
705 		/* No socket for error: try tunnels before discarding */
706 		sk = ERR_PTR(-ENOENT);
707 		if (static_branch_unlikely(&udp_encap_needed_key)) {
708 			sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
709 						  info);
710 			if (!sk)
711 				return 0;
712 		}
713 
714 		if (IS_ERR(sk)) {
715 			__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
716 			return PTR_ERR(sk);
717 		}
718 
719 		tunnel = true;
720 	}
721 
722 	err = 0;
723 	harderr = 0;
724 	inet = inet_sk(sk);
725 
726 	switch (type) {
727 	default:
728 	case ICMP_TIME_EXCEEDED:
729 		err = EHOSTUNREACH;
730 		break;
731 	case ICMP_SOURCE_QUENCH:
732 		goto out;
733 	case ICMP_PARAMETERPROB:
734 		err = EPROTO;
735 		harderr = 1;
736 		break;
737 	case ICMP_DEST_UNREACH:
738 		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
739 			ipv4_sk_update_pmtu(skb, sk, info);
740 			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
741 				err = EMSGSIZE;
742 				harderr = 1;
743 				break;
744 			}
745 			goto out;
746 		}
747 		err = EHOSTUNREACH;
748 		if (code <= NR_ICMP_UNREACH) {
749 			harderr = icmp_err_convert[code].fatal;
750 			err = icmp_err_convert[code].errno;
751 		}
752 		break;
753 	case ICMP_REDIRECT:
754 		ipv4_sk_redirect(skb, sk);
755 		goto out;
756 	}
757 
758 	/*
759 	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
760 	 *	4.1.3.3.
761 	 */
762 	if (tunnel) {
763 		/* ...not for tunnels though: we don't have a sending socket */
764 		goto out;
765 	}
766 	if (!inet->recverr) {
767 		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
768 			goto out;
769 	} else
770 		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
771 
772 	sk->sk_err = err;
773 	sk->sk_error_report(sk);
774 out:
775 	return 0;
776 }
777 
778 int udp_err(struct sk_buff *skb, u32 info)
779 {
780 	return __udp4_lib_err(skb, info, &udp_table);
781 }
782 
783 /*
784  * Throw away all pending data and cancel the corking. Socket is locked.
785  */
786 void udp_flush_pending_frames(struct sock *sk)
787 {
788 	struct udp_sock *up = udp_sk(sk);
789 
790 	if (up->pending) {
791 		up->len = 0;
792 		up->pending = 0;
793 		ip_flush_pending_frames(sk);
794 	}
795 }
796 EXPORT_SYMBOL(udp_flush_pending_frames);
797 
798 /**
799  * 	udp4_hwcsum  -  handle outgoing HW checksumming
800  * 	@skb: 	sk_buff containing the filled-in UDP header
801  * 	        (checksum field must be zeroed out)
802  *	@src:	source IP address
803  *	@dst:	destination IP address
804  */
805 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
806 {
807 	struct udphdr *uh = udp_hdr(skb);
808 	int offset = skb_transport_offset(skb);
809 	int len = skb->len - offset;
810 	int hlen = len;
811 	__wsum csum = 0;
812 
813 	if (!skb_has_frag_list(skb)) {
814 		/*
815 		 * Only one fragment on the socket.
816 		 */
817 		skb->csum_start = skb_transport_header(skb) - skb->head;
818 		skb->csum_offset = offsetof(struct udphdr, check);
819 		uh->check = ~csum_tcpudp_magic(src, dst, len,
820 					       IPPROTO_UDP, 0);
821 	} else {
822 		struct sk_buff *frags;
823 
824 		/*
825 		 * HW-checksum won't work as there are two or more
826 		 * fragments on the socket so that all csums of sk_buffs
827 		 * should be together
828 		 */
829 		skb_walk_frags(skb, frags) {
830 			csum = csum_add(csum, frags->csum);
831 			hlen -= frags->len;
832 		}
833 
834 		csum = skb_checksum(skb, offset, hlen, csum);
835 		skb->ip_summed = CHECKSUM_NONE;
836 
837 		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
838 		if (uh->check == 0)
839 			uh->check = CSUM_MANGLED_0;
840 	}
841 }
842 EXPORT_SYMBOL_GPL(udp4_hwcsum);
843 
844 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
845  * for the simple case like when setting the checksum for a UDP tunnel.
846  */
847 void udp_set_csum(bool nocheck, struct sk_buff *skb,
848 		  __be32 saddr, __be32 daddr, int len)
849 {
850 	struct udphdr *uh = udp_hdr(skb);
851 
852 	if (nocheck) {
853 		uh->check = 0;
854 	} else if (skb_is_gso(skb)) {
855 		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
856 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
857 		uh->check = 0;
858 		uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
859 		if (uh->check == 0)
860 			uh->check = CSUM_MANGLED_0;
861 	} else {
862 		skb->ip_summed = CHECKSUM_PARTIAL;
863 		skb->csum_start = skb_transport_header(skb) - skb->head;
864 		skb->csum_offset = offsetof(struct udphdr, check);
865 		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
866 	}
867 }
868 EXPORT_SYMBOL(udp_set_csum);
869 
870 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
871 			struct inet_cork *cork)
872 {
873 	struct sock *sk = skb->sk;
874 	struct inet_sock *inet = inet_sk(sk);
875 	struct udphdr *uh;
876 	int err;
877 	int is_udplite = IS_UDPLITE(sk);
878 	int offset = skb_transport_offset(skb);
879 	int len = skb->len - offset;
880 	int datalen = len - sizeof(*uh);
881 	__wsum csum = 0;
882 
883 	/*
884 	 * Create a UDP header
885 	 */
886 	uh = udp_hdr(skb);
887 	uh->source = inet->inet_sport;
888 	uh->dest = fl4->fl4_dport;
889 	uh->len = htons(len);
890 	uh->check = 0;
891 
892 	if (cork->gso_size) {
893 		const int hlen = skb_network_header_len(skb) +
894 				 sizeof(struct udphdr);
895 
896 		if (hlen + cork->gso_size > cork->fragsize) {
897 			kfree_skb(skb);
898 			return -EINVAL;
899 		}
900 		if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
901 			kfree_skb(skb);
902 			return -EINVAL;
903 		}
904 		if (sk->sk_no_check_tx) {
905 			kfree_skb(skb);
906 			return -EINVAL;
907 		}
908 		if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
909 		    dst_xfrm(skb_dst(skb))) {
910 			kfree_skb(skb);
911 			return -EIO;
912 		}
913 
914 		if (datalen > cork->gso_size) {
915 			skb_shinfo(skb)->gso_size = cork->gso_size;
916 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
917 			skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
918 								 cork->gso_size);
919 		}
920 		goto csum_partial;
921 	}
922 
923 	if (is_udplite)  				 /*     UDP-Lite      */
924 		csum = udplite_csum(skb);
925 
926 	else if (sk->sk_no_check_tx) {			 /* UDP csum off */
927 
928 		skb->ip_summed = CHECKSUM_NONE;
929 		goto send;
930 
931 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
932 csum_partial:
933 
934 		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
935 		goto send;
936 
937 	} else
938 		csum = udp_csum(skb);
939 
940 	/* add protocol-dependent pseudo-header */
941 	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
942 				      sk->sk_protocol, csum);
943 	if (uh->check == 0)
944 		uh->check = CSUM_MANGLED_0;
945 
946 send:
947 	err = ip_send_skb(sock_net(sk), skb);
948 	if (err) {
949 		if (err == -ENOBUFS && !inet->recverr) {
950 			UDP_INC_STATS(sock_net(sk),
951 				      UDP_MIB_SNDBUFERRORS, is_udplite);
952 			err = 0;
953 		}
954 	} else
955 		UDP_INC_STATS(sock_net(sk),
956 			      UDP_MIB_OUTDATAGRAMS, is_udplite);
957 	return err;
958 }
959 
960 /*
961  * Push out all pending data as one UDP datagram. Socket is locked.
962  */
963 int udp_push_pending_frames(struct sock *sk)
964 {
965 	struct udp_sock  *up = udp_sk(sk);
966 	struct inet_sock *inet = inet_sk(sk);
967 	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
968 	struct sk_buff *skb;
969 	int err = 0;
970 
971 	skb = ip_finish_skb(sk, fl4);
972 	if (!skb)
973 		goto out;
974 
975 	err = udp_send_skb(skb, fl4, &inet->cork.base);
976 
977 out:
978 	up->len = 0;
979 	up->pending = 0;
980 	return err;
981 }
982 EXPORT_SYMBOL(udp_push_pending_frames);
983 
984 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
985 {
986 	switch (cmsg->cmsg_type) {
987 	case UDP_SEGMENT:
988 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
989 			return -EINVAL;
990 		*gso_size = *(__u16 *)CMSG_DATA(cmsg);
991 		return 0;
992 	default:
993 		return -EINVAL;
994 	}
995 }
996 
997 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
998 {
999 	struct cmsghdr *cmsg;
1000 	bool need_ip = false;
1001 	int err;
1002 
1003 	for_each_cmsghdr(cmsg, msg) {
1004 		if (!CMSG_OK(msg, cmsg))
1005 			return -EINVAL;
1006 
1007 		if (cmsg->cmsg_level != SOL_UDP) {
1008 			need_ip = true;
1009 			continue;
1010 		}
1011 
1012 		err = __udp_cmsg_send(cmsg, gso_size);
1013 		if (err)
1014 			return err;
1015 	}
1016 
1017 	return need_ip;
1018 }
1019 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1020 
1021 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1022 {
1023 	struct inet_sock *inet = inet_sk(sk);
1024 	struct udp_sock *up = udp_sk(sk);
1025 	DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1026 	struct flowi4 fl4_stack;
1027 	struct flowi4 *fl4;
1028 	int ulen = len;
1029 	struct ipcm_cookie ipc;
1030 	struct rtable *rt = NULL;
1031 	int free = 0;
1032 	int connected = 0;
1033 	__be32 daddr, faddr, saddr;
1034 	__be16 dport;
1035 	u8  tos;
1036 	int err, is_udplite = IS_UDPLITE(sk);
1037 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1038 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1039 	struct sk_buff *skb;
1040 	struct ip_options_data opt_copy;
1041 
1042 	if (len > 0xFFFF)
1043 		return -EMSGSIZE;
1044 
1045 	/*
1046 	 *	Check the flags.
1047 	 */
1048 
1049 	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1050 		return -EOPNOTSUPP;
1051 
1052 	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1053 
1054 	fl4 = &inet->cork.fl.u.ip4;
1055 	if (up->pending) {
1056 		/*
1057 		 * There are pending frames.
1058 		 * The socket lock must be held while it's corked.
1059 		 */
1060 		lock_sock(sk);
1061 		if (likely(up->pending)) {
1062 			if (unlikely(up->pending != AF_INET)) {
1063 				release_sock(sk);
1064 				return -EINVAL;
1065 			}
1066 			goto do_append_data;
1067 		}
1068 		release_sock(sk);
1069 	}
1070 	ulen += sizeof(struct udphdr);
1071 
1072 	/*
1073 	 *	Get and verify the address.
1074 	 */
1075 	if (usin) {
1076 		if (msg->msg_namelen < sizeof(*usin))
1077 			return -EINVAL;
1078 		if (usin->sin_family != AF_INET) {
1079 			if (usin->sin_family != AF_UNSPEC)
1080 				return -EAFNOSUPPORT;
1081 		}
1082 
1083 		daddr = usin->sin_addr.s_addr;
1084 		dport = usin->sin_port;
1085 		if (dport == 0)
1086 			return -EINVAL;
1087 	} else {
1088 		if (sk->sk_state != TCP_ESTABLISHED)
1089 			return -EDESTADDRREQ;
1090 		daddr = inet->inet_daddr;
1091 		dport = inet->inet_dport;
1092 		/* Open fast path for connected socket.
1093 		   Route will not be used, if at least one option is set.
1094 		 */
1095 		connected = 1;
1096 	}
1097 
1098 	ipcm_init_sk(&ipc, inet);
1099 	ipc.gso_size = up->gso_size;
1100 
1101 	if (msg->msg_controllen) {
1102 		err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1103 		if (err > 0)
1104 			err = ip_cmsg_send(sk, msg, &ipc,
1105 					   sk->sk_family == AF_INET6);
1106 		if (unlikely(err < 0)) {
1107 			kfree(ipc.opt);
1108 			return err;
1109 		}
1110 		if (ipc.opt)
1111 			free = 1;
1112 		connected = 0;
1113 	}
1114 	if (!ipc.opt) {
1115 		struct ip_options_rcu *inet_opt;
1116 
1117 		rcu_read_lock();
1118 		inet_opt = rcu_dereference(inet->inet_opt);
1119 		if (inet_opt) {
1120 			memcpy(&opt_copy, inet_opt,
1121 			       sizeof(*inet_opt) + inet_opt->opt.optlen);
1122 			ipc.opt = &opt_copy.opt;
1123 		}
1124 		rcu_read_unlock();
1125 	}
1126 
1127 	if (cgroup_bpf_enabled && !connected) {
1128 		err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1129 					    (struct sockaddr *)usin, &ipc.addr);
1130 		if (err)
1131 			goto out_free;
1132 		if (usin) {
1133 			if (usin->sin_port == 0) {
1134 				/* BPF program set invalid port. Reject it. */
1135 				err = -EINVAL;
1136 				goto out_free;
1137 			}
1138 			daddr = usin->sin_addr.s_addr;
1139 			dport = usin->sin_port;
1140 		}
1141 	}
1142 
1143 	saddr = ipc.addr;
1144 	ipc.addr = faddr = daddr;
1145 
1146 	if (ipc.opt && ipc.opt->opt.srr) {
1147 		if (!daddr) {
1148 			err = -EINVAL;
1149 			goto out_free;
1150 		}
1151 		faddr = ipc.opt->opt.faddr;
1152 		connected = 0;
1153 	}
1154 	tos = get_rttos(&ipc, inet);
1155 	if (sock_flag(sk, SOCK_LOCALROUTE) ||
1156 	    (msg->msg_flags & MSG_DONTROUTE) ||
1157 	    (ipc.opt && ipc.opt->opt.is_strictroute)) {
1158 		tos |= RTO_ONLINK;
1159 		connected = 0;
1160 	}
1161 
1162 	if (ipv4_is_multicast(daddr)) {
1163 		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1164 			ipc.oif = inet->mc_index;
1165 		if (!saddr)
1166 			saddr = inet->mc_addr;
1167 		connected = 0;
1168 	} else if (!ipc.oif) {
1169 		ipc.oif = inet->uc_index;
1170 	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1171 		/* oif is set, packet is to local broadcast and
1172 		 * uc_index is set. oif is most likely set
1173 		 * by sk_bound_dev_if. If uc_index != oif check if the
1174 		 * oif is an L3 master and uc_index is an L3 slave.
1175 		 * If so, we want to allow the send using the uc_index.
1176 		 */
1177 		if (ipc.oif != inet->uc_index &&
1178 		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1179 							      inet->uc_index)) {
1180 			ipc.oif = inet->uc_index;
1181 		}
1182 	}
1183 
1184 	if (connected)
1185 		rt = (struct rtable *)sk_dst_check(sk, 0);
1186 
1187 	if (!rt) {
1188 		struct net *net = sock_net(sk);
1189 		__u8 flow_flags = inet_sk_flowi_flags(sk);
1190 
1191 		fl4 = &fl4_stack;
1192 
1193 		flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1194 				   RT_SCOPE_UNIVERSE, sk->sk_protocol,
1195 				   flow_flags,
1196 				   faddr, saddr, dport, inet->inet_sport,
1197 				   sk->sk_uid);
1198 
1199 		security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1200 		rt = ip_route_output_flow(net, fl4, sk);
1201 		if (IS_ERR(rt)) {
1202 			err = PTR_ERR(rt);
1203 			rt = NULL;
1204 			if (err == -ENETUNREACH)
1205 				IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1206 			goto out;
1207 		}
1208 
1209 		err = -EACCES;
1210 		if ((rt->rt_flags & RTCF_BROADCAST) &&
1211 		    !sock_flag(sk, SOCK_BROADCAST))
1212 			goto out;
1213 		if (connected)
1214 			sk_dst_set(sk, dst_clone(&rt->dst));
1215 	}
1216 
1217 	if (msg->msg_flags&MSG_CONFIRM)
1218 		goto do_confirm;
1219 back_from_confirm:
1220 
1221 	saddr = fl4->saddr;
1222 	if (!ipc.addr)
1223 		daddr = ipc.addr = fl4->daddr;
1224 
1225 	/* Lockless fast path for the non-corking case. */
1226 	if (!corkreq) {
1227 		struct inet_cork cork;
1228 
1229 		skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1230 				  sizeof(struct udphdr), &ipc, &rt,
1231 				  &cork, msg->msg_flags);
1232 		err = PTR_ERR(skb);
1233 		if (!IS_ERR_OR_NULL(skb))
1234 			err = udp_send_skb(skb, fl4, &cork);
1235 		goto out;
1236 	}
1237 
1238 	lock_sock(sk);
1239 	if (unlikely(up->pending)) {
1240 		/* The socket is already corked while preparing it. */
1241 		/* ... which is an evident application bug. --ANK */
1242 		release_sock(sk);
1243 
1244 		net_dbg_ratelimited("socket already corked\n");
1245 		err = -EINVAL;
1246 		goto out;
1247 	}
1248 	/*
1249 	 *	Now cork the socket to pend data.
1250 	 */
1251 	fl4 = &inet->cork.fl.u.ip4;
1252 	fl4->daddr = daddr;
1253 	fl4->saddr = saddr;
1254 	fl4->fl4_dport = dport;
1255 	fl4->fl4_sport = inet->inet_sport;
1256 	up->pending = AF_INET;
1257 
1258 do_append_data:
1259 	up->len += ulen;
1260 	err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1261 			     sizeof(struct udphdr), &ipc, &rt,
1262 			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1263 	if (err)
1264 		udp_flush_pending_frames(sk);
1265 	else if (!corkreq)
1266 		err = udp_push_pending_frames(sk);
1267 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1268 		up->pending = 0;
1269 	release_sock(sk);
1270 
1271 out:
1272 	ip_rt_put(rt);
1273 out_free:
1274 	if (free)
1275 		kfree(ipc.opt);
1276 	if (!err)
1277 		return len;
1278 	/*
1279 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1280 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1281 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1282 	 * things).  We could add another new stat but at least for now that
1283 	 * seems like overkill.
1284 	 */
1285 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1286 		UDP_INC_STATS(sock_net(sk),
1287 			      UDP_MIB_SNDBUFERRORS, is_udplite);
1288 	}
1289 	return err;
1290 
1291 do_confirm:
1292 	if (msg->msg_flags & MSG_PROBE)
1293 		dst_confirm_neigh(&rt->dst, &fl4->daddr);
1294 	if (!(msg->msg_flags&MSG_PROBE) || len)
1295 		goto back_from_confirm;
1296 	err = 0;
1297 	goto out;
1298 }
1299 EXPORT_SYMBOL(udp_sendmsg);
1300 
1301 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1302 		 size_t size, int flags)
1303 {
1304 	struct inet_sock *inet = inet_sk(sk);
1305 	struct udp_sock *up = udp_sk(sk);
1306 	int ret;
1307 
1308 	if (flags & MSG_SENDPAGE_NOTLAST)
1309 		flags |= MSG_MORE;
1310 
1311 	if (!up->pending) {
1312 		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
1313 
1314 		/* Call udp_sendmsg to specify destination address which
1315 		 * sendpage interface can't pass.
1316 		 * This will succeed only when the socket is connected.
1317 		 */
1318 		ret = udp_sendmsg(sk, &msg, 0);
1319 		if (ret < 0)
1320 			return ret;
1321 	}
1322 
1323 	lock_sock(sk);
1324 
1325 	if (unlikely(!up->pending)) {
1326 		release_sock(sk);
1327 
1328 		net_dbg_ratelimited("cork failed\n");
1329 		return -EINVAL;
1330 	}
1331 
1332 	ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1333 			     page, offset, size, flags);
1334 	if (ret == -EOPNOTSUPP) {
1335 		release_sock(sk);
1336 		return sock_no_sendpage(sk->sk_socket, page, offset,
1337 					size, flags);
1338 	}
1339 	if (ret < 0) {
1340 		udp_flush_pending_frames(sk);
1341 		goto out;
1342 	}
1343 
1344 	up->len += size;
1345 	if (!(up->corkflag || (flags&MSG_MORE)))
1346 		ret = udp_push_pending_frames(sk);
1347 	if (!ret)
1348 		ret = size;
1349 out:
1350 	release_sock(sk);
1351 	return ret;
1352 }
1353 
1354 #define UDP_SKB_IS_STATELESS 0x80000000
1355 
1356 /* all head states (dst, sk, nf conntrack) except skb extensions are
1357  * cleared by udp_rcv().
1358  *
1359  * We need to preserve secpath, if present, to eventually process
1360  * IP_CMSG_PASSSEC at recvmsg() time.
1361  *
1362  * Other extensions can be cleared.
1363  */
1364 static bool udp_try_make_stateless(struct sk_buff *skb)
1365 {
1366 	if (!skb_has_extensions(skb))
1367 		return true;
1368 
1369 	if (!secpath_exists(skb)) {
1370 		skb_ext_reset(skb);
1371 		return true;
1372 	}
1373 
1374 	return false;
1375 }
1376 
1377 static void udp_set_dev_scratch(struct sk_buff *skb)
1378 {
1379 	struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1380 
1381 	BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1382 	scratch->_tsize_state = skb->truesize;
1383 #if BITS_PER_LONG == 64
1384 	scratch->len = skb->len;
1385 	scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1386 	scratch->is_linear = !skb_is_nonlinear(skb);
1387 #endif
1388 	if (udp_try_make_stateless(skb))
1389 		scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1390 }
1391 
1392 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1393 {
1394 	/* We come here after udp_lib_checksum_complete() returned 0.
1395 	 * This means that __skb_checksum_complete() might have
1396 	 * set skb->csum_valid to 1.
1397 	 * On 64bit platforms, we can set csum_unnecessary
1398 	 * to true, but only if the skb is not shared.
1399 	 */
1400 #if BITS_PER_LONG == 64
1401 	if (!skb_shared(skb))
1402 		udp_skb_scratch(skb)->csum_unnecessary = true;
1403 #endif
1404 }
1405 
1406 static int udp_skb_truesize(struct sk_buff *skb)
1407 {
1408 	return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1409 }
1410 
1411 static bool udp_skb_has_head_state(struct sk_buff *skb)
1412 {
1413 	return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1414 }
1415 
1416 /* fully reclaim rmem/fwd memory allocated for skb */
1417 static void udp_rmem_release(struct sock *sk, int size, int partial,
1418 			     bool rx_queue_lock_held)
1419 {
1420 	struct udp_sock *up = udp_sk(sk);
1421 	struct sk_buff_head *sk_queue;
1422 	int amt;
1423 
1424 	if (likely(partial)) {
1425 		up->forward_deficit += size;
1426 		size = up->forward_deficit;
1427 		if (size < (sk->sk_rcvbuf >> 2) &&
1428 		    !skb_queue_empty(&up->reader_queue))
1429 			return;
1430 	} else {
1431 		size += up->forward_deficit;
1432 	}
1433 	up->forward_deficit = 0;
1434 
1435 	/* acquire the sk_receive_queue for fwd allocated memory scheduling,
1436 	 * if the called don't held it already
1437 	 */
1438 	sk_queue = &sk->sk_receive_queue;
1439 	if (!rx_queue_lock_held)
1440 		spin_lock(&sk_queue->lock);
1441 
1442 
1443 	sk->sk_forward_alloc += size;
1444 	amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1445 	sk->sk_forward_alloc -= amt;
1446 
1447 	if (amt)
1448 		__sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1449 
1450 	atomic_sub(size, &sk->sk_rmem_alloc);
1451 
1452 	/* this can save us from acquiring the rx queue lock on next receive */
1453 	skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1454 
1455 	if (!rx_queue_lock_held)
1456 		spin_unlock(&sk_queue->lock);
1457 }
1458 
1459 /* Note: called with reader_queue.lock held.
1460  * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1461  * This avoids a cache line miss while receive_queue lock is held.
1462  * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1463  */
1464 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1465 {
1466 	prefetch(&skb->data);
1467 	udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1468 }
1469 EXPORT_SYMBOL(udp_skb_destructor);
1470 
1471 /* as above, but the caller held the rx queue lock, too */
1472 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1473 {
1474 	prefetch(&skb->data);
1475 	udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1476 }
1477 
1478 /* Idea of busylocks is to let producers grab an extra spinlock
1479  * to relieve pressure on the receive_queue spinlock shared by consumer.
1480  * Under flood, this means that only one producer can be in line
1481  * trying to acquire the receive_queue spinlock.
1482  * These busylock can be allocated on a per cpu manner, instead of a
1483  * per socket one (that would consume a cache line per socket)
1484  */
1485 static int udp_busylocks_log __read_mostly;
1486 static spinlock_t *udp_busylocks __read_mostly;
1487 
1488 static spinlock_t *busylock_acquire(void *ptr)
1489 {
1490 	spinlock_t *busy;
1491 
1492 	busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1493 	spin_lock(busy);
1494 	return busy;
1495 }
1496 
1497 static void busylock_release(spinlock_t *busy)
1498 {
1499 	if (busy)
1500 		spin_unlock(busy);
1501 }
1502 
1503 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1504 {
1505 	struct sk_buff_head *list = &sk->sk_receive_queue;
1506 	int rmem, delta, amt, err = -ENOMEM;
1507 	spinlock_t *busy = NULL;
1508 	int size;
1509 
1510 	/* try to avoid the costly atomic add/sub pair when the receive
1511 	 * queue is full; always allow at least a packet
1512 	 */
1513 	rmem = atomic_read(&sk->sk_rmem_alloc);
1514 	if (rmem > sk->sk_rcvbuf)
1515 		goto drop;
1516 
1517 	/* Under mem pressure, it might be helpful to help udp_recvmsg()
1518 	 * having linear skbs :
1519 	 * - Reduce memory overhead and thus increase receive queue capacity
1520 	 * - Less cache line misses at copyout() time
1521 	 * - Less work at consume_skb() (less alien page frag freeing)
1522 	 */
1523 	if (rmem > (sk->sk_rcvbuf >> 1)) {
1524 		skb_condense(skb);
1525 
1526 		busy = busylock_acquire(sk);
1527 	}
1528 	size = skb->truesize;
1529 	udp_set_dev_scratch(skb);
1530 
1531 	/* we drop only if the receive buf is full and the receive
1532 	 * queue contains some other skb
1533 	 */
1534 	rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1535 	if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1536 		goto uncharge_drop;
1537 
1538 	spin_lock(&list->lock);
1539 	if (size >= sk->sk_forward_alloc) {
1540 		amt = sk_mem_pages(size);
1541 		delta = amt << SK_MEM_QUANTUM_SHIFT;
1542 		if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1543 			err = -ENOBUFS;
1544 			spin_unlock(&list->lock);
1545 			goto uncharge_drop;
1546 		}
1547 
1548 		sk->sk_forward_alloc += delta;
1549 	}
1550 
1551 	sk->sk_forward_alloc -= size;
1552 
1553 	/* no need to setup a destructor, we will explicitly release the
1554 	 * forward allocated memory on dequeue
1555 	 */
1556 	sock_skb_set_dropcount(sk, skb);
1557 
1558 	__skb_queue_tail(list, skb);
1559 	spin_unlock(&list->lock);
1560 
1561 	if (!sock_flag(sk, SOCK_DEAD))
1562 		sk->sk_data_ready(sk);
1563 
1564 	busylock_release(busy);
1565 	return 0;
1566 
1567 uncharge_drop:
1568 	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1569 
1570 drop:
1571 	atomic_inc(&sk->sk_drops);
1572 	busylock_release(busy);
1573 	return err;
1574 }
1575 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1576 
1577 void udp_destruct_sock(struct sock *sk)
1578 {
1579 	/* reclaim completely the forward allocated memory */
1580 	struct udp_sock *up = udp_sk(sk);
1581 	unsigned int total = 0;
1582 	struct sk_buff *skb;
1583 
1584 	skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1585 	while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1586 		total += skb->truesize;
1587 		kfree_skb(skb);
1588 	}
1589 	udp_rmem_release(sk, total, 0, true);
1590 
1591 	inet_sock_destruct(sk);
1592 }
1593 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1594 
1595 int udp_init_sock(struct sock *sk)
1596 {
1597 	skb_queue_head_init(&udp_sk(sk)->reader_queue);
1598 	sk->sk_destruct = udp_destruct_sock;
1599 	return 0;
1600 }
1601 EXPORT_SYMBOL_GPL(udp_init_sock);
1602 
1603 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1604 {
1605 	if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1606 		bool slow = lock_sock_fast(sk);
1607 
1608 		sk_peek_offset_bwd(sk, len);
1609 		unlock_sock_fast(sk, slow);
1610 	}
1611 
1612 	if (!skb_unref(skb))
1613 		return;
1614 
1615 	/* In the more common cases we cleared the head states previously,
1616 	 * see __udp_queue_rcv_skb().
1617 	 */
1618 	if (unlikely(udp_skb_has_head_state(skb)))
1619 		skb_release_head_state(skb);
1620 	__consume_stateless_skb(skb);
1621 }
1622 EXPORT_SYMBOL_GPL(skb_consume_udp);
1623 
1624 static struct sk_buff *__first_packet_length(struct sock *sk,
1625 					     struct sk_buff_head *rcvq,
1626 					     int *total)
1627 {
1628 	struct sk_buff *skb;
1629 
1630 	while ((skb = skb_peek(rcvq)) != NULL) {
1631 		if (udp_lib_checksum_complete(skb)) {
1632 			__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1633 					IS_UDPLITE(sk));
1634 			__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1635 					IS_UDPLITE(sk));
1636 			atomic_inc(&sk->sk_drops);
1637 			__skb_unlink(skb, rcvq);
1638 			*total += skb->truesize;
1639 			kfree_skb(skb);
1640 		} else {
1641 			udp_skb_csum_unnecessary_set(skb);
1642 			break;
1643 		}
1644 	}
1645 	return skb;
1646 }
1647 
1648 /**
1649  *	first_packet_length	- return length of first packet in receive queue
1650  *	@sk: socket
1651  *
1652  *	Drops all bad checksum frames, until a valid one is found.
1653  *	Returns the length of found skb, or -1 if none is found.
1654  */
1655 static int first_packet_length(struct sock *sk)
1656 {
1657 	struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1658 	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1659 	struct sk_buff *skb;
1660 	int total = 0;
1661 	int res;
1662 
1663 	spin_lock_bh(&rcvq->lock);
1664 	skb = __first_packet_length(sk, rcvq, &total);
1665 	if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1666 		spin_lock(&sk_queue->lock);
1667 		skb_queue_splice_tail_init(sk_queue, rcvq);
1668 		spin_unlock(&sk_queue->lock);
1669 
1670 		skb = __first_packet_length(sk, rcvq, &total);
1671 	}
1672 	res = skb ? skb->len : -1;
1673 	if (total)
1674 		udp_rmem_release(sk, total, 1, false);
1675 	spin_unlock_bh(&rcvq->lock);
1676 	return res;
1677 }
1678 
1679 /*
1680  *	IOCTL requests applicable to the UDP protocol
1681  */
1682 
1683 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1684 {
1685 	switch (cmd) {
1686 	case SIOCOUTQ:
1687 	{
1688 		int amount = sk_wmem_alloc_get(sk);
1689 
1690 		return put_user(amount, (int __user *)arg);
1691 	}
1692 
1693 	case SIOCINQ:
1694 	{
1695 		int amount = max_t(int, 0, first_packet_length(sk));
1696 
1697 		return put_user(amount, (int __user *)arg);
1698 	}
1699 
1700 	default:
1701 		return -ENOIOCTLCMD;
1702 	}
1703 
1704 	return 0;
1705 }
1706 EXPORT_SYMBOL(udp_ioctl);
1707 
1708 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1709 			       int noblock, int *off, int *err)
1710 {
1711 	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1712 	struct sk_buff_head *queue;
1713 	struct sk_buff *last;
1714 	long timeo;
1715 	int error;
1716 
1717 	queue = &udp_sk(sk)->reader_queue;
1718 	flags |= noblock ? MSG_DONTWAIT : 0;
1719 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1720 	do {
1721 		struct sk_buff *skb;
1722 
1723 		error = sock_error(sk);
1724 		if (error)
1725 			break;
1726 
1727 		error = -EAGAIN;
1728 		do {
1729 			spin_lock_bh(&queue->lock);
1730 			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1731 							err, &last);
1732 			if (skb) {
1733 				if (!(flags & MSG_PEEK))
1734 					udp_skb_destructor(sk, skb);
1735 				spin_unlock_bh(&queue->lock);
1736 				return skb;
1737 			}
1738 
1739 			if (skb_queue_empty_lockless(sk_queue)) {
1740 				spin_unlock_bh(&queue->lock);
1741 				goto busy_check;
1742 			}
1743 
1744 			/* refill the reader queue and walk it again
1745 			 * keep both queues locked to avoid re-acquiring
1746 			 * the sk_receive_queue lock if fwd memory scheduling
1747 			 * is needed.
1748 			 */
1749 			spin_lock(&sk_queue->lock);
1750 			skb_queue_splice_tail_init(sk_queue, queue);
1751 
1752 			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1753 							err, &last);
1754 			if (skb && !(flags & MSG_PEEK))
1755 				udp_skb_dtor_locked(sk, skb);
1756 			spin_unlock(&sk_queue->lock);
1757 			spin_unlock_bh(&queue->lock);
1758 			if (skb)
1759 				return skb;
1760 
1761 busy_check:
1762 			if (!sk_can_busy_loop(sk))
1763 				break;
1764 
1765 			sk_busy_loop(sk, flags & MSG_DONTWAIT);
1766 		} while (!skb_queue_empty_lockless(sk_queue));
1767 
1768 		/* sk_queue is empty, reader_queue may contain peeked packets */
1769 	} while (timeo &&
1770 		 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1771 					      &error, &timeo,
1772 					      (struct sk_buff *)sk_queue));
1773 
1774 	*err = error;
1775 	return NULL;
1776 }
1777 EXPORT_SYMBOL(__skb_recv_udp);
1778 
1779 /*
1780  * 	This should be easy, if there is something there we
1781  * 	return it, otherwise we block.
1782  */
1783 
1784 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1785 		int flags, int *addr_len)
1786 {
1787 	struct inet_sock *inet = inet_sk(sk);
1788 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1789 	struct sk_buff *skb;
1790 	unsigned int ulen, copied;
1791 	int off, err, peeking = flags & MSG_PEEK;
1792 	int is_udplite = IS_UDPLITE(sk);
1793 	bool checksum_valid = false;
1794 
1795 	if (flags & MSG_ERRQUEUE)
1796 		return ip_recv_error(sk, msg, len, addr_len);
1797 
1798 try_again:
1799 	off = sk_peek_offset(sk, flags);
1800 	skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1801 	if (!skb)
1802 		return err;
1803 
1804 	ulen = udp_skb_len(skb);
1805 	copied = len;
1806 	if (copied > ulen - off)
1807 		copied = ulen - off;
1808 	else if (copied < ulen)
1809 		msg->msg_flags |= MSG_TRUNC;
1810 
1811 	/*
1812 	 * If checksum is needed at all, try to do it while copying the
1813 	 * data.  If the data is truncated, or if we only want a partial
1814 	 * coverage checksum (UDP-Lite), do it before the copy.
1815 	 */
1816 
1817 	if (copied < ulen || peeking ||
1818 	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1819 		checksum_valid = udp_skb_csum_unnecessary(skb) ||
1820 				!__udp_lib_checksum_complete(skb);
1821 		if (!checksum_valid)
1822 			goto csum_copy_err;
1823 	}
1824 
1825 	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1826 		if (udp_skb_is_linear(skb))
1827 			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1828 		else
1829 			err = skb_copy_datagram_msg(skb, off, msg, copied);
1830 	} else {
1831 		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1832 
1833 		if (err == -EINVAL)
1834 			goto csum_copy_err;
1835 	}
1836 
1837 	if (unlikely(err)) {
1838 		if (!peeking) {
1839 			atomic_inc(&sk->sk_drops);
1840 			UDP_INC_STATS(sock_net(sk),
1841 				      UDP_MIB_INERRORS, is_udplite);
1842 		}
1843 		kfree_skb(skb);
1844 		return err;
1845 	}
1846 
1847 	if (!peeking)
1848 		UDP_INC_STATS(sock_net(sk),
1849 			      UDP_MIB_INDATAGRAMS, is_udplite);
1850 
1851 	sock_recv_ts_and_drops(msg, sk, skb);
1852 
1853 	/* Copy the address. */
1854 	if (sin) {
1855 		sin->sin_family = AF_INET;
1856 		sin->sin_port = udp_hdr(skb)->source;
1857 		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1858 		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1859 		*addr_len = sizeof(*sin);
1860 
1861 		if (cgroup_bpf_enabled)
1862 			BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1863 							(struct sockaddr *)sin);
1864 	}
1865 
1866 	if (udp_sk(sk)->gro_enabled)
1867 		udp_cmsg_recv(msg, sk, skb);
1868 
1869 	if (inet->cmsg_flags)
1870 		ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1871 
1872 	err = copied;
1873 	if (flags & MSG_TRUNC)
1874 		err = ulen;
1875 
1876 	skb_consume_udp(sk, skb, peeking ? -err : err);
1877 	return err;
1878 
1879 csum_copy_err:
1880 	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1881 				 udp_skb_destructor)) {
1882 		UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1883 		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1884 	}
1885 	kfree_skb(skb);
1886 
1887 	/* starting over for a new packet, but check if we need to yield */
1888 	cond_resched();
1889 	msg->msg_flags &= ~MSG_TRUNC;
1890 	goto try_again;
1891 }
1892 
1893 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1894 {
1895 	/* This check is replicated from __ip4_datagram_connect() and
1896 	 * intended to prevent BPF program called below from accessing bytes
1897 	 * that are out of the bound specified by user in addr_len.
1898 	 */
1899 	if (addr_len < sizeof(struct sockaddr_in))
1900 		return -EINVAL;
1901 
1902 	return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1903 }
1904 EXPORT_SYMBOL(udp_pre_connect);
1905 
1906 int __udp_disconnect(struct sock *sk, int flags)
1907 {
1908 	struct inet_sock *inet = inet_sk(sk);
1909 	/*
1910 	 *	1003.1g - break association.
1911 	 */
1912 
1913 	sk->sk_state = TCP_CLOSE;
1914 	inet->inet_daddr = 0;
1915 	inet->inet_dport = 0;
1916 	sock_rps_reset_rxhash(sk);
1917 	sk->sk_bound_dev_if = 0;
1918 	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1919 		inet_reset_saddr(sk);
1920 		if (sk->sk_prot->rehash &&
1921 		    (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1922 			sk->sk_prot->rehash(sk);
1923 	}
1924 
1925 	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1926 		sk->sk_prot->unhash(sk);
1927 		inet->inet_sport = 0;
1928 	}
1929 	sk_dst_reset(sk);
1930 	return 0;
1931 }
1932 EXPORT_SYMBOL(__udp_disconnect);
1933 
1934 int udp_disconnect(struct sock *sk, int flags)
1935 {
1936 	lock_sock(sk);
1937 	__udp_disconnect(sk, flags);
1938 	release_sock(sk);
1939 	return 0;
1940 }
1941 EXPORT_SYMBOL(udp_disconnect);
1942 
1943 void udp_lib_unhash(struct sock *sk)
1944 {
1945 	if (sk_hashed(sk)) {
1946 		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1947 		struct udp_hslot *hslot, *hslot2;
1948 
1949 		hslot  = udp_hashslot(udptable, sock_net(sk),
1950 				      udp_sk(sk)->udp_port_hash);
1951 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1952 
1953 		spin_lock_bh(&hslot->lock);
1954 		if (rcu_access_pointer(sk->sk_reuseport_cb))
1955 			reuseport_detach_sock(sk);
1956 		if (sk_del_node_init_rcu(sk)) {
1957 			hslot->count--;
1958 			inet_sk(sk)->inet_num = 0;
1959 			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1960 
1961 			spin_lock(&hslot2->lock);
1962 			hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1963 			hslot2->count--;
1964 			spin_unlock(&hslot2->lock);
1965 		}
1966 		spin_unlock_bh(&hslot->lock);
1967 	}
1968 }
1969 EXPORT_SYMBOL(udp_lib_unhash);
1970 
1971 /*
1972  * inet_rcv_saddr was changed, we must rehash secondary hash
1973  */
1974 void udp_lib_rehash(struct sock *sk, u16 newhash)
1975 {
1976 	if (sk_hashed(sk)) {
1977 		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1978 		struct udp_hslot *hslot, *hslot2, *nhslot2;
1979 
1980 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1981 		nhslot2 = udp_hashslot2(udptable, newhash);
1982 		udp_sk(sk)->udp_portaddr_hash = newhash;
1983 
1984 		if (hslot2 != nhslot2 ||
1985 		    rcu_access_pointer(sk->sk_reuseport_cb)) {
1986 			hslot = udp_hashslot(udptable, sock_net(sk),
1987 					     udp_sk(sk)->udp_port_hash);
1988 			/* we must lock primary chain too */
1989 			spin_lock_bh(&hslot->lock);
1990 			if (rcu_access_pointer(sk->sk_reuseport_cb))
1991 				reuseport_detach_sock(sk);
1992 
1993 			if (hslot2 != nhslot2) {
1994 				spin_lock(&hslot2->lock);
1995 				hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1996 				hslot2->count--;
1997 				spin_unlock(&hslot2->lock);
1998 
1999 				spin_lock(&nhslot2->lock);
2000 				hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2001 							 &nhslot2->head);
2002 				nhslot2->count++;
2003 				spin_unlock(&nhslot2->lock);
2004 			}
2005 
2006 			spin_unlock_bh(&hslot->lock);
2007 		}
2008 	}
2009 }
2010 EXPORT_SYMBOL(udp_lib_rehash);
2011 
2012 void udp_v4_rehash(struct sock *sk)
2013 {
2014 	u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2015 					  inet_sk(sk)->inet_rcv_saddr,
2016 					  inet_sk(sk)->inet_num);
2017 	udp_lib_rehash(sk, new_hash);
2018 }
2019 
2020 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2021 {
2022 	int rc;
2023 
2024 	if (inet_sk(sk)->inet_daddr) {
2025 		sock_rps_save_rxhash(sk, skb);
2026 		sk_mark_napi_id(sk, skb);
2027 		sk_incoming_cpu_update(sk);
2028 	} else {
2029 		sk_mark_napi_id_once(sk, skb);
2030 	}
2031 
2032 	rc = __udp_enqueue_schedule_skb(sk, skb);
2033 	if (rc < 0) {
2034 		int is_udplite = IS_UDPLITE(sk);
2035 
2036 		/* Note that an ENOMEM error is charged twice */
2037 		if (rc == -ENOMEM)
2038 			UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2039 					is_udplite);
2040 		else
2041 			UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2042 				      is_udplite);
2043 		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2044 		kfree_skb(skb);
2045 		trace_udp_fail_queue_rcv_skb(rc, sk);
2046 		return -1;
2047 	}
2048 
2049 	return 0;
2050 }
2051 
2052 /* returns:
2053  *  -1: error
2054  *   0: success
2055  *  >0: "udp encap" protocol resubmission
2056  *
2057  * Note that in the success and error cases, the skb is assumed to
2058  * have either been requeued or freed.
2059  */
2060 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2061 {
2062 	struct udp_sock *up = udp_sk(sk);
2063 	int is_udplite = IS_UDPLITE(sk);
2064 
2065 	/*
2066 	 *	Charge it to the socket, dropping if the queue is full.
2067 	 */
2068 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2069 		goto drop;
2070 	nf_reset_ct(skb);
2071 
2072 	if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2073 		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2074 
2075 		/*
2076 		 * This is an encapsulation socket so pass the skb to
2077 		 * the socket's udp_encap_rcv() hook. Otherwise, just
2078 		 * fall through and pass this up the UDP socket.
2079 		 * up->encap_rcv() returns the following value:
2080 		 * =0 if skb was successfully passed to the encap
2081 		 *    handler or was discarded by it.
2082 		 * >0 if skb should be passed on to UDP.
2083 		 * <0 if skb should be resubmitted as proto -N
2084 		 */
2085 
2086 		/* if we're overly short, let UDP handle it */
2087 		encap_rcv = READ_ONCE(up->encap_rcv);
2088 		if (encap_rcv) {
2089 			int ret;
2090 
2091 			/* Verify checksum before giving to encap */
2092 			if (udp_lib_checksum_complete(skb))
2093 				goto csum_error;
2094 
2095 			ret = encap_rcv(sk, skb);
2096 			if (ret <= 0) {
2097 				__UDP_INC_STATS(sock_net(sk),
2098 						UDP_MIB_INDATAGRAMS,
2099 						is_udplite);
2100 				return -ret;
2101 			}
2102 		}
2103 
2104 		/* FALLTHROUGH -- it's a UDP Packet */
2105 	}
2106 
2107 	/*
2108 	 * 	UDP-Lite specific tests, ignored on UDP sockets
2109 	 */
2110 	if ((up->pcflag & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
2111 
2112 		/*
2113 		 * MIB statistics other than incrementing the error count are
2114 		 * disabled for the following two types of errors: these depend
2115 		 * on the application settings, not on the functioning of the
2116 		 * protocol stack as such.
2117 		 *
2118 		 * RFC 3828 here recommends (sec 3.3): "There should also be a
2119 		 * way ... to ... at least let the receiving application block
2120 		 * delivery of packets with coverage values less than a value
2121 		 * provided by the application."
2122 		 */
2123 		if (up->pcrlen == 0) {          /* full coverage was set  */
2124 			net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2125 					    UDP_SKB_CB(skb)->cscov, skb->len);
2126 			goto drop;
2127 		}
2128 		/* The next case involves violating the min. coverage requested
2129 		 * by the receiver. This is subtle: if receiver wants x and x is
2130 		 * greater than the buffersize/MTU then receiver will complain
2131 		 * that it wants x while sender emits packets of smaller size y.
2132 		 * Therefore the above ...()->partial_cov statement is essential.
2133 		 */
2134 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
2135 			net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2136 					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
2137 			goto drop;
2138 		}
2139 	}
2140 
2141 	prefetch(&sk->sk_rmem_alloc);
2142 	if (rcu_access_pointer(sk->sk_filter) &&
2143 	    udp_lib_checksum_complete(skb))
2144 			goto csum_error;
2145 
2146 	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2147 		goto drop;
2148 
2149 	udp_csum_pull_header(skb);
2150 
2151 	ipv4_pktinfo_prepare(sk, skb);
2152 	return __udp_queue_rcv_skb(sk, skb);
2153 
2154 csum_error:
2155 	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2156 drop:
2157 	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2158 	atomic_inc(&sk->sk_drops);
2159 	kfree_skb(skb);
2160 	return -1;
2161 }
2162 
2163 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2164 {
2165 	struct sk_buff *next, *segs;
2166 	int ret;
2167 
2168 	if (likely(!udp_unexpected_gso(sk, skb)))
2169 		return udp_queue_rcv_one_skb(sk, skb);
2170 
2171 	BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2172 	__skb_push(skb, -skb_mac_offset(skb));
2173 	segs = udp_rcv_segment(sk, skb, true);
2174 	skb_list_walk_safe(segs, skb, next) {
2175 		__skb_pull(skb, skb_transport_offset(skb));
2176 		ret = udp_queue_rcv_one_skb(sk, skb);
2177 		if (ret > 0)
2178 			ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2179 	}
2180 	return 0;
2181 }
2182 
2183 /* For TCP sockets, sk_rx_dst is protected by socket lock
2184  * For UDP, we use xchg() to guard against concurrent changes.
2185  */
2186 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2187 {
2188 	struct dst_entry *old;
2189 
2190 	if (dst_hold_safe(dst)) {
2191 		old = xchg(&sk->sk_rx_dst, dst);
2192 		dst_release(old);
2193 		return old != dst;
2194 	}
2195 	return false;
2196 }
2197 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2198 
2199 /*
2200  *	Multicasts and broadcasts go to each listener.
2201  *
2202  *	Note: called only from the BH handler context.
2203  */
2204 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2205 				    struct udphdr  *uh,
2206 				    __be32 saddr, __be32 daddr,
2207 				    struct udp_table *udptable,
2208 				    int proto)
2209 {
2210 	struct sock *sk, *first = NULL;
2211 	unsigned short hnum = ntohs(uh->dest);
2212 	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2213 	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2214 	unsigned int offset = offsetof(typeof(*sk), sk_node);
2215 	int dif = skb->dev->ifindex;
2216 	int sdif = inet_sdif(skb);
2217 	struct hlist_node *node;
2218 	struct sk_buff *nskb;
2219 
2220 	if (use_hash2) {
2221 		hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2222 			    udptable->mask;
2223 		hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2224 start_lookup:
2225 		hslot = &udptable->hash2[hash2];
2226 		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2227 	}
2228 
2229 	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2230 		if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2231 					 uh->source, saddr, dif, sdif, hnum))
2232 			continue;
2233 
2234 		if (!first) {
2235 			first = sk;
2236 			continue;
2237 		}
2238 		nskb = skb_clone(skb, GFP_ATOMIC);
2239 
2240 		if (unlikely(!nskb)) {
2241 			atomic_inc(&sk->sk_drops);
2242 			__UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2243 					IS_UDPLITE(sk));
2244 			__UDP_INC_STATS(net, UDP_MIB_INERRORS,
2245 					IS_UDPLITE(sk));
2246 			continue;
2247 		}
2248 		if (udp_queue_rcv_skb(sk, nskb) > 0)
2249 			consume_skb(nskb);
2250 	}
2251 
2252 	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
2253 	if (use_hash2 && hash2 != hash2_any) {
2254 		hash2 = hash2_any;
2255 		goto start_lookup;
2256 	}
2257 
2258 	if (first) {
2259 		if (udp_queue_rcv_skb(first, skb) > 0)
2260 			consume_skb(skb);
2261 	} else {
2262 		kfree_skb(skb);
2263 		__UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2264 				proto == IPPROTO_UDPLITE);
2265 	}
2266 	return 0;
2267 }
2268 
2269 /* Initialize UDP checksum. If exited with zero value (success),
2270  * CHECKSUM_UNNECESSARY means, that no more checks are required.
2271  * Otherwise, csum completion requires checksumming packet body,
2272  * including udp header and folding it to skb->csum.
2273  */
2274 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2275 				 int proto)
2276 {
2277 	int err;
2278 
2279 	UDP_SKB_CB(skb)->partial_cov = 0;
2280 	UDP_SKB_CB(skb)->cscov = skb->len;
2281 
2282 	if (proto == IPPROTO_UDPLITE) {
2283 		err = udplite_checksum_init(skb, uh);
2284 		if (err)
2285 			return err;
2286 
2287 		if (UDP_SKB_CB(skb)->partial_cov) {
2288 			skb->csum = inet_compute_pseudo(skb, proto);
2289 			return 0;
2290 		}
2291 	}
2292 
2293 	/* Note, we are only interested in != 0 or == 0, thus the
2294 	 * force to int.
2295 	 */
2296 	err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2297 							inet_compute_pseudo);
2298 	if (err)
2299 		return err;
2300 
2301 	if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2302 		/* If SW calculated the value, we know it's bad */
2303 		if (skb->csum_complete_sw)
2304 			return 1;
2305 
2306 		/* HW says the value is bad. Let's validate that.
2307 		 * skb->csum is no longer the full packet checksum,
2308 		 * so don't treat it as such.
2309 		 */
2310 		skb_checksum_complete_unset(skb);
2311 	}
2312 
2313 	return 0;
2314 }
2315 
2316 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2317  * return code conversion for ip layer consumption
2318  */
2319 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2320 			       struct udphdr *uh)
2321 {
2322 	int ret;
2323 
2324 	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2325 		skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2326 
2327 	ret = udp_queue_rcv_skb(sk, skb);
2328 
2329 	/* a return value > 0 means to resubmit the input, but
2330 	 * it wants the return to be -protocol, or 0
2331 	 */
2332 	if (ret > 0)
2333 		return -ret;
2334 	return 0;
2335 }
2336 
2337 /*
2338  *	All we need to do is get the socket, and then do a checksum.
2339  */
2340 
2341 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2342 		   int proto)
2343 {
2344 	struct sock *sk;
2345 	struct udphdr *uh;
2346 	unsigned short ulen;
2347 	struct rtable *rt = skb_rtable(skb);
2348 	__be32 saddr, daddr;
2349 	struct net *net = dev_net(skb->dev);
2350 	bool refcounted;
2351 
2352 	/*
2353 	 *  Validate the packet.
2354 	 */
2355 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2356 		goto drop;		/* No space for header. */
2357 
2358 	uh   = udp_hdr(skb);
2359 	ulen = ntohs(uh->len);
2360 	saddr = ip_hdr(skb)->saddr;
2361 	daddr = ip_hdr(skb)->daddr;
2362 
2363 	if (ulen > skb->len)
2364 		goto short_packet;
2365 
2366 	if (proto == IPPROTO_UDP) {
2367 		/* UDP validates ulen. */
2368 		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2369 			goto short_packet;
2370 		uh = udp_hdr(skb);
2371 	}
2372 
2373 	if (udp4_csum_init(skb, uh, proto))
2374 		goto csum_error;
2375 
2376 	sk = skb_steal_sock(skb, &refcounted);
2377 	if (sk) {
2378 		struct dst_entry *dst = skb_dst(skb);
2379 		int ret;
2380 
2381 		if (unlikely(sk->sk_rx_dst != dst))
2382 			udp_sk_rx_dst_set(sk, dst);
2383 
2384 		ret = udp_unicast_rcv_skb(sk, skb, uh);
2385 		if (refcounted)
2386 			sock_put(sk);
2387 		return ret;
2388 	}
2389 
2390 	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2391 		return __udp4_lib_mcast_deliver(net, skb, uh,
2392 						saddr, daddr, udptable, proto);
2393 
2394 	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2395 	if (sk)
2396 		return udp_unicast_rcv_skb(sk, skb, uh);
2397 
2398 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2399 		goto drop;
2400 	nf_reset_ct(skb);
2401 
2402 	/* No socket. Drop packet silently, if checksum is wrong */
2403 	if (udp_lib_checksum_complete(skb))
2404 		goto csum_error;
2405 
2406 	__UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2407 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2408 
2409 	/*
2410 	 * Hmm.  We got an UDP packet to a port to which we
2411 	 * don't wanna listen.  Ignore it.
2412 	 */
2413 	kfree_skb(skb);
2414 	return 0;
2415 
2416 short_packet:
2417 	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2418 			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2419 			    &saddr, ntohs(uh->source),
2420 			    ulen, skb->len,
2421 			    &daddr, ntohs(uh->dest));
2422 	goto drop;
2423 
2424 csum_error:
2425 	/*
2426 	 * RFC1122: OK.  Discards the bad packet silently (as far as
2427 	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2428 	 */
2429 	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2430 			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2431 			    &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2432 			    ulen);
2433 	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2434 drop:
2435 	__UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2436 	kfree_skb(skb);
2437 	return 0;
2438 }
2439 
2440 /* We can only early demux multicast if there is a single matching socket.
2441  * If more than one socket found returns NULL
2442  */
2443 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2444 						  __be16 loc_port, __be32 loc_addr,
2445 						  __be16 rmt_port, __be32 rmt_addr,
2446 						  int dif, int sdif)
2447 {
2448 	struct sock *sk, *result;
2449 	unsigned short hnum = ntohs(loc_port);
2450 	unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2451 	struct udp_hslot *hslot = &udp_table.hash[slot];
2452 
2453 	/* Do not bother scanning a too big list */
2454 	if (hslot->count > 10)
2455 		return NULL;
2456 
2457 	result = NULL;
2458 	sk_for_each_rcu(sk, &hslot->head) {
2459 		if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2460 					rmt_port, rmt_addr, dif, sdif, hnum)) {
2461 			if (result)
2462 				return NULL;
2463 			result = sk;
2464 		}
2465 	}
2466 
2467 	return result;
2468 }
2469 
2470 /* For unicast we should only early demux connected sockets or we can
2471  * break forwarding setups.  The chains here can be long so only check
2472  * if the first socket is an exact match and if not move on.
2473  */
2474 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2475 					    __be16 loc_port, __be32 loc_addr,
2476 					    __be16 rmt_port, __be32 rmt_addr,
2477 					    int dif, int sdif)
2478 {
2479 	unsigned short hnum = ntohs(loc_port);
2480 	unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2481 	unsigned int slot2 = hash2 & udp_table.mask;
2482 	struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2483 	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2484 	const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2485 	struct sock *sk;
2486 
2487 	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2488 		if (INET_MATCH(sk, net, acookie, rmt_addr,
2489 			       loc_addr, ports, dif, sdif))
2490 			return sk;
2491 		/* Only check first socket in chain */
2492 		break;
2493 	}
2494 	return NULL;
2495 }
2496 
2497 int udp_v4_early_demux(struct sk_buff *skb)
2498 {
2499 	struct net *net = dev_net(skb->dev);
2500 	struct in_device *in_dev = NULL;
2501 	const struct iphdr *iph;
2502 	const struct udphdr *uh;
2503 	struct sock *sk = NULL;
2504 	struct dst_entry *dst;
2505 	int dif = skb->dev->ifindex;
2506 	int sdif = inet_sdif(skb);
2507 	int ours;
2508 
2509 	/* validate the packet */
2510 	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2511 		return 0;
2512 
2513 	iph = ip_hdr(skb);
2514 	uh = udp_hdr(skb);
2515 
2516 	if (skb->pkt_type == PACKET_MULTICAST) {
2517 		in_dev = __in_dev_get_rcu(skb->dev);
2518 
2519 		if (!in_dev)
2520 			return 0;
2521 
2522 		ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2523 				       iph->protocol);
2524 		if (!ours)
2525 			return 0;
2526 
2527 		sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2528 						   uh->source, iph->saddr,
2529 						   dif, sdif);
2530 	} else if (skb->pkt_type == PACKET_HOST) {
2531 		sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2532 					     uh->source, iph->saddr, dif, sdif);
2533 	}
2534 
2535 	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2536 		return 0;
2537 
2538 	skb->sk = sk;
2539 	skb->destructor = sock_efree;
2540 	dst = READ_ONCE(sk->sk_rx_dst);
2541 
2542 	if (dst)
2543 		dst = dst_check(dst, 0);
2544 	if (dst) {
2545 		u32 itag = 0;
2546 
2547 		/* set noref for now.
2548 		 * any place which wants to hold dst has to call
2549 		 * dst_hold_safe()
2550 		 */
2551 		skb_dst_set_noref(skb, dst);
2552 
2553 		/* for unconnected multicast sockets we need to validate
2554 		 * the source on each packet
2555 		 */
2556 		if (!inet_sk(sk)->inet_daddr && in_dev)
2557 			return ip_mc_validate_source(skb, iph->daddr,
2558 						     iph->saddr,
2559 						     iph->tos & IPTOS_RT_MASK,
2560 						     skb->dev, in_dev, &itag);
2561 	}
2562 	return 0;
2563 }
2564 
2565 int udp_rcv(struct sk_buff *skb)
2566 {
2567 	return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2568 }
2569 
2570 void udp_destroy_sock(struct sock *sk)
2571 {
2572 	struct udp_sock *up = udp_sk(sk);
2573 	bool slow = lock_sock_fast(sk);
2574 	udp_flush_pending_frames(sk);
2575 	unlock_sock_fast(sk, slow);
2576 	if (static_branch_unlikely(&udp_encap_needed_key)) {
2577 		if (up->encap_type) {
2578 			void (*encap_destroy)(struct sock *sk);
2579 			encap_destroy = READ_ONCE(up->encap_destroy);
2580 			if (encap_destroy)
2581 				encap_destroy(sk);
2582 		}
2583 		if (up->encap_enabled)
2584 			static_branch_dec(&udp_encap_needed_key);
2585 	}
2586 }
2587 
2588 /*
2589  *	Socket option code for UDP
2590  */
2591 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2592 		       sockptr_t optval, unsigned int optlen,
2593 		       int (*push_pending_frames)(struct sock *))
2594 {
2595 	struct udp_sock *up = udp_sk(sk);
2596 	int val, valbool;
2597 	int err = 0;
2598 	int is_udplite = IS_UDPLITE(sk);
2599 
2600 	if (optlen < sizeof(int))
2601 		return -EINVAL;
2602 
2603 	if (copy_from_sockptr(&val, optval, sizeof(val)))
2604 		return -EFAULT;
2605 
2606 	valbool = val ? 1 : 0;
2607 
2608 	switch (optname) {
2609 	case UDP_CORK:
2610 		if (val != 0) {
2611 			up->corkflag = 1;
2612 		} else {
2613 			up->corkflag = 0;
2614 			lock_sock(sk);
2615 			push_pending_frames(sk);
2616 			release_sock(sk);
2617 		}
2618 		break;
2619 
2620 	case UDP_ENCAP:
2621 		switch (val) {
2622 		case 0:
2623 #ifdef CONFIG_XFRM
2624 		case UDP_ENCAP_ESPINUDP:
2625 		case UDP_ENCAP_ESPINUDP_NON_IKE:
2626 #if IS_ENABLED(CONFIG_IPV6)
2627 			if (sk->sk_family == AF_INET6)
2628 				up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2629 			else
2630 #endif
2631 				up->encap_rcv = xfrm4_udp_encap_rcv;
2632 #endif
2633 			fallthrough;
2634 		case UDP_ENCAP_L2TPINUDP:
2635 			up->encap_type = val;
2636 			lock_sock(sk);
2637 			udp_tunnel_encap_enable(sk->sk_socket);
2638 			release_sock(sk);
2639 			break;
2640 		default:
2641 			err = -ENOPROTOOPT;
2642 			break;
2643 		}
2644 		break;
2645 
2646 	case UDP_NO_CHECK6_TX:
2647 		up->no_check6_tx = valbool;
2648 		break;
2649 
2650 	case UDP_NO_CHECK6_RX:
2651 		up->no_check6_rx = valbool;
2652 		break;
2653 
2654 	case UDP_SEGMENT:
2655 		if (val < 0 || val > USHRT_MAX)
2656 			return -EINVAL;
2657 		up->gso_size = val;
2658 		break;
2659 
2660 	case UDP_GRO:
2661 		lock_sock(sk);
2662 		if (valbool)
2663 			udp_tunnel_encap_enable(sk->sk_socket);
2664 		up->gro_enabled = valbool;
2665 		release_sock(sk);
2666 		break;
2667 
2668 	/*
2669 	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
2670 	 */
2671 	/* The sender sets actual checksum coverage length via this option.
2672 	 * The case coverage > packet length is handled by send module. */
2673 	case UDPLITE_SEND_CSCOV:
2674 		if (!is_udplite)         /* Disable the option on UDP sockets */
2675 			return -ENOPROTOOPT;
2676 		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2677 			val = 8;
2678 		else if (val > USHRT_MAX)
2679 			val = USHRT_MAX;
2680 		up->pcslen = val;
2681 		up->pcflag |= UDPLITE_SEND_CC;
2682 		break;
2683 
2684 	/* The receiver specifies a minimum checksum coverage value. To make
2685 	 * sense, this should be set to at least 8 (as done below). If zero is
2686 	 * used, this again means full checksum coverage.                     */
2687 	case UDPLITE_RECV_CSCOV:
2688 		if (!is_udplite)         /* Disable the option on UDP sockets */
2689 			return -ENOPROTOOPT;
2690 		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2691 			val = 8;
2692 		else if (val > USHRT_MAX)
2693 			val = USHRT_MAX;
2694 		up->pcrlen = val;
2695 		up->pcflag |= UDPLITE_RECV_CC;
2696 		break;
2697 
2698 	default:
2699 		err = -ENOPROTOOPT;
2700 		break;
2701 	}
2702 
2703 	return err;
2704 }
2705 EXPORT_SYMBOL(udp_lib_setsockopt);
2706 
2707 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2708 		   unsigned int optlen)
2709 {
2710 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2711 		return udp_lib_setsockopt(sk, level, optname,
2712 					  optval, optlen,
2713 					  udp_push_pending_frames);
2714 	return ip_setsockopt(sk, level, optname, optval, optlen);
2715 }
2716 
2717 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2718 		       char __user *optval, int __user *optlen)
2719 {
2720 	struct udp_sock *up = udp_sk(sk);
2721 	int val, len;
2722 
2723 	if (get_user(len, optlen))
2724 		return -EFAULT;
2725 
2726 	len = min_t(unsigned int, len, sizeof(int));
2727 
2728 	if (len < 0)
2729 		return -EINVAL;
2730 
2731 	switch (optname) {
2732 	case UDP_CORK:
2733 		val = up->corkflag;
2734 		break;
2735 
2736 	case UDP_ENCAP:
2737 		val = up->encap_type;
2738 		break;
2739 
2740 	case UDP_NO_CHECK6_TX:
2741 		val = up->no_check6_tx;
2742 		break;
2743 
2744 	case UDP_NO_CHECK6_RX:
2745 		val = up->no_check6_rx;
2746 		break;
2747 
2748 	case UDP_SEGMENT:
2749 		val = up->gso_size;
2750 		break;
2751 
2752 	/* The following two cannot be changed on UDP sockets, the return is
2753 	 * always 0 (which corresponds to the full checksum coverage of UDP). */
2754 	case UDPLITE_SEND_CSCOV:
2755 		val = up->pcslen;
2756 		break;
2757 
2758 	case UDPLITE_RECV_CSCOV:
2759 		val = up->pcrlen;
2760 		break;
2761 
2762 	default:
2763 		return -ENOPROTOOPT;
2764 	}
2765 
2766 	if (put_user(len, optlen))
2767 		return -EFAULT;
2768 	if (copy_to_user(optval, &val, len))
2769 		return -EFAULT;
2770 	return 0;
2771 }
2772 EXPORT_SYMBOL(udp_lib_getsockopt);
2773 
2774 int udp_getsockopt(struct sock *sk, int level, int optname,
2775 		   char __user *optval, int __user *optlen)
2776 {
2777 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2778 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2779 	return ip_getsockopt(sk, level, optname, optval, optlen);
2780 }
2781 
2782 /**
2783  * 	udp_poll - wait for a UDP event.
2784  *	@file: - file struct
2785  *	@sock: - socket
2786  *	@wait: - poll table
2787  *
2788  *	This is same as datagram poll, except for the special case of
2789  *	blocking sockets. If application is using a blocking fd
2790  *	and a packet with checksum error is in the queue;
2791  *	then it could get return from select indicating data available
2792  *	but then block when reading it. Add special case code
2793  *	to work around these arguably broken applications.
2794  */
2795 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2796 {
2797 	__poll_t mask = datagram_poll(file, sock, wait);
2798 	struct sock *sk = sock->sk;
2799 
2800 	if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2801 		mask |= EPOLLIN | EPOLLRDNORM;
2802 
2803 	/* Check for false positives due to checksum errors */
2804 	if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2805 	    !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2806 		mask &= ~(EPOLLIN | EPOLLRDNORM);
2807 
2808 	return mask;
2809 
2810 }
2811 EXPORT_SYMBOL(udp_poll);
2812 
2813 int udp_abort(struct sock *sk, int err)
2814 {
2815 	lock_sock(sk);
2816 
2817 	sk->sk_err = err;
2818 	sk->sk_error_report(sk);
2819 	__udp_disconnect(sk, 0);
2820 
2821 	release_sock(sk);
2822 
2823 	return 0;
2824 }
2825 EXPORT_SYMBOL_GPL(udp_abort);
2826 
2827 struct proto udp_prot = {
2828 	.name			= "UDP",
2829 	.owner			= THIS_MODULE,
2830 	.close			= udp_lib_close,
2831 	.pre_connect		= udp_pre_connect,
2832 	.connect		= ip4_datagram_connect,
2833 	.disconnect		= udp_disconnect,
2834 	.ioctl			= udp_ioctl,
2835 	.init			= udp_init_sock,
2836 	.destroy		= udp_destroy_sock,
2837 	.setsockopt		= udp_setsockopt,
2838 	.getsockopt		= udp_getsockopt,
2839 	.sendmsg		= udp_sendmsg,
2840 	.recvmsg		= udp_recvmsg,
2841 	.sendpage		= udp_sendpage,
2842 	.release_cb		= ip4_datagram_release_cb,
2843 	.hash			= udp_lib_hash,
2844 	.unhash			= udp_lib_unhash,
2845 	.rehash			= udp_v4_rehash,
2846 	.get_port		= udp_v4_get_port,
2847 	.memory_allocated	= &udp_memory_allocated,
2848 	.sysctl_mem		= sysctl_udp_mem,
2849 	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2850 	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2851 	.obj_size		= sizeof(struct udp_sock),
2852 	.h.udp_table		= &udp_table,
2853 	.diag_destroy		= udp_abort,
2854 };
2855 EXPORT_SYMBOL(udp_prot);
2856 
2857 /* ------------------------------------------------------------------------ */
2858 #ifdef CONFIG_PROC_FS
2859 
2860 static struct sock *udp_get_first(struct seq_file *seq, int start)
2861 {
2862 	struct sock *sk;
2863 	struct udp_seq_afinfo *afinfo;
2864 	struct udp_iter_state *state = seq->private;
2865 	struct net *net = seq_file_net(seq);
2866 
2867 	if (state->bpf_seq_afinfo)
2868 		afinfo = state->bpf_seq_afinfo;
2869 	else
2870 		afinfo = PDE_DATA(file_inode(seq->file));
2871 
2872 	for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2873 	     ++state->bucket) {
2874 		struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2875 
2876 		if (hlist_empty(&hslot->head))
2877 			continue;
2878 
2879 		spin_lock_bh(&hslot->lock);
2880 		sk_for_each(sk, &hslot->head) {
2881 			if (!net_eq(sock_net(sk), net))
2882 				continue;
2883 			if (afinfo->family == AF_UNSPEC ||
2884 			    sk->sk_family == afinfo->family)
2885 				goto found;
2886 		}
2887 		spin_unlock_bh(&hslot->lock);
2888 	}
2889 	sk = NULL;
2890 found:
2891 	return sk;
2892 }
2893 
2894 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2895 {
2896 	struct udp_seq_afinfo *afinfo;
2897 	struct udp_iter_state *state = seq->private;
2898 	struct net *net = seq_file_net(seq);
2899 
2900 	if (state->bpf_seq_afinfo)
2901 		afinfo = state->bpf_seq_afinfo;
2902 	else
2903 		afinfo = PDE_DATA(file_inode(seq->file));
2904 
2905 	do {
2906 		sk = sk_next(sk);
2907 	} while (sk && (!net_eq(sock_net(sk), net) ||
2908 			(afinfo->family != AF_UNSPEC &&
2909 			 sk->sk_family != afinfo->family)));
2910 
2911 	if (!sk) {
2912 		if (state->bucket <= afinfo->udp_table->mask)
2913 			spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2914 		return udp_get_first(seq, state->bucket + 1);
2915 	}
2916 	return sk;
2917 }
2918 
2919 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2920 {
2921 	struct sock *sk = udp_get_first(seq, 0);
2922 
2923 	if (sk)
2924 		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2925 			--pos;
2926 	return pos ? NULL : sk;
2927 }
2928 
2929 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2930 {
2931 	struct udp_iter_state *state = seq->private;
2932 	state->bucket = MAX_UDP_PORTS;
2933 
2934 	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2935 }
2936 EXPORT_SYMBOL(udp_seq_start);
2937 
2938 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2939 {
2940 	struct sock *sk;
2941 
2942 	if (v == SEQ_START_TOKEN)
2943 		sk = udp_get_idx(seq, 0);
2944 	else
2945 		sk = udp_get_next(seq, v);
2946 
2947 	++*pos;
2948 	return sk;
2949 }
2950 EXPORT_SYMBOL(udp_seq_next);
2951 
2952 void udp_seq_stop(struct seq_file *seq, void *v)
2953 {
2954 	struct udp_seq_afinfo *afinfo;
2955 	struct udp_iter_state *state = seq->private;
2956 
2957 	if (state->bpf_seq_afinfo)
2958 		afinfo = state->bpf_seq_afinfo;
2959 	else
2960 		afinfo = PDE_DATA(file_inode(seq->file));
2961 
2962 	if (state->bucket <= afinfo->udp_table->mask)
2963 		spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2964 }
2965 EXPORT_SYMBOL(udp_seq_stop);
2966 
2967 /* ------------------------------------------------------------------------ */
2968 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2969 		int bucket)
2970 {
2971 	struct inet_sock *inet = inet_sk(sp);
2972 	__be32 dest = inet->inet_daddr;
2973 	__be32 src  = inet->inet_rcv_saddr;
2974 	__u16 destp	  = ntohs(inet->inet_dport);
2975 	__u16 srcp	  = ntohs(inet->inet_sport);
2976 
2977 	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2978 		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
2979 		bucket, src, srcp, dest, destp, sp->sk_state,
2980 		sk_wmem_alloc_get(sp),
2981 		udp_rqueue_get(sp),
2982 		0, 0L, 0,
2983 		from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2984 		0, sock_i_ino(sp),
2985 		refcount_read(&sp->sk_refcnt), sp,
2986 		atomic_read(&sp->sk_drops));
2987 }
2988 
2989 int udp4_seq_show(struct seq_file *seq, void *v)
2990 {
2991 	seq_setwidth(seq, 127);
2992 	if (v == SEQ_START_TOKEN)
2993 		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2994 			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2995 			   "inode ref pointer drops");
2996 	else {
2997 		struct udp_iter_state *state = seq->private;
2998 
2999 		udp4_format_sock(v, seq, state->bucket);
3000 	}
3001 	seq_pad(seq, '\n');
3002 	return 0;
3003 }
3004 
3005 #ifdef CONFIG_BPF_SYSCALL
3006 struct bpf_iter__udp {
3007 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
3008 	__bpf_md_ptr(struct udp_sock *, udp_sk);
3009 	uid_t uid __aligned(8);
3010 	int bucket __aligned(8);
3011 };
3012 
3013 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3014 			     struct udp_sock *udp_sk, uid_t uid, int bucket)
3015 {
3016 	struct bpf_iter__udp ctx;
3017 
3018 	meta->seq_num--;  /* skip SEQ_START_TOKEN */
3019 	ctx.meta = meta;
3020 	ctx.udp_sk = udp_sk;
3021 	ctx.uid = uid;
3022 	ctx.bucket = bucket;
3023 	return bpf_iter_run_prog(prog, &ctx);
3024 }
3025 
3026 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3027 {
3028 	struct udp_iter_state *state = seq->private;
3029 	struct bpf_iter_meta meta;
3030 	struct bpf_prog *prog;
3031 	struct sock *sk = v;
3032 	uid_t uid;
3033 
3034 	if (v == SEQ_START_TOKEN)
3035 		return 0;
3036 
3037 	uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3038 	meta.seq = seq;
3039 	prog = bpf_iter_get_info(&meta, false);
3040 	return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3041 }
3042 
3043 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3044 {
3045 	struct bpf_iter_meta meta;
3046 	struct bpf_prog *prog;
3047 
3048 	if (!v) {
3049 		meta.seq = seq;
3050 		prog = bpf_iter_get_info(&meta, true);
3051 		if (prog)
3052 			(void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3053 	}
3054 
3055 	udp_seq_stop(seq, v);
3056 }
3057 
3058 static const struct seq_operations bpf_iter_udp_seq_ops = {
3059 	.start		= udp_seq_start,
3060 	.next		= udp_seq_next,
3061 	.stop		= bpf_iter_udp_seq_stop,
3062 	.show		= bpf_iter_udp_seq_show,
3063 };
3064 #endif
3065 
3066 const struct seq_operations udp_seq_ops = {
3067 	.start		= udp_seq_start,
3068 	.next		= udp_seq_next,
3069 	.stop		= udp_seq_stop,
3070 	.show		= udp4_seq_show,
3071 };
3072 EXPORT_SYMBOL(udp_seq_ops);
3073 
3074 static struct udp_seq_afinfo udp4_seq_afinfo = {
3075 	.family		= AF_INET,
3076 	.udp_table	= &udp_table,
3077 };
3078 
3079 static int __net_init udp4_proc_init_net(struct net *net)
3080 {
3081 	if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3082 			sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3083 		return -ENOMEM;
3084 	return 0;
3085 }
3086 
3087 static void __net_exit udp4_proc_exit_net(struct net *net)
3088 {
3089 	remove_proc_entry("udp", net->proc_net);
3090 }
3091 
3092 static struct pernet_operations udp4_net_ops = {
3093 	.init = udp4_proc_init_net,
3094 	.exit = udp4_proc_exit_net,
3095 };
3096 
3097 int __init udp4_proc_init(void)
3098 {
3099 	return register_pernet_subsys(&udp4_net_ops);
3100 }
3101 
3102 void udp4_proc_exit(void)
3103 {
3104 	unregister_pernet_subsys(&udp4_net_ops);
3105 }
3106 #endif /* CONFIG_PROC_FS */
3107 
3108 static __initdata unsigned long uhash_entries;
3109 static int __init set_uhash_entries(char *str)
3110 {
3111 	ssize_t ret;
3112 
3113 	if (!str)
3114 		return 0;
3115 
3116 	ret = kstrtoul(str, 0, &uhash_entries);
3117 	if (ret)
3118 		return 0;
3119 
3120 	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3121 		uhash_entries = UDP_HTABLE_SIZE_MIN;
3122 	return 1;
3123 }
3124 __setup("uhash_entries=", set_uhash_entries);
3125 
3126 void __init udp_table_init(struct udp_table *table, const char *name)
3127 {
3128 	unsigned int i;
3129 
3130 	table->hash = alloc_large_system_hash(name,
3131 					      2 * sizeof(struct udp_hslot),
3132 					      uhash_entries,
3133 					      21, /* one slot per 2 MB */
3134 					      0,
3135 					      &table->log,
3136 					      &table->mask,
3137 					      UDP_HTABLE_SIZE_MIN,
3138 					      64 * 1024);
3139 
3140 	table->hash2 = table->hash + (table->mask + 1);
3141 	for (i = 0; i <= table->mask; i++) {
3142 		INIT_HLIST_HEAD(&table->hash[i].head);
3143 		table->hash[i].count = 0;
3144 		spin_lock_init(&table->hash[i].lock);
3145 	}
3146 	for (i = 0; i <= table->mask; i++) {
3147 		INIT_HLIST_HEAD(&table->hash2[i].head);
3148 		table->hash2[i].count = 0;
3149 		spin_lock_init(&table->hash2[i].lock);
3150 	}
3151 }
3152 
3153 u32 udp_flow_hashrnd(void)
3154 {
3155 	static u32 hashrnd __read_mostly;
3156 
3157 	net_get_random_once(&hashrnd, sizeof(hashrnd));
3158 
3159 	return hashrnd;
3160 }
3161 EXPORT_SYMBOL(udp_flow_hashrnd);
3162 
3163 static void __udp_sysctl_init(struct net *net)
3164 {
3165 	net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3166 	net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3167 
3168 #ifdef CONFIG_NET_L3_MASTER_DEV
3169 	net->ipv4.sysctl_udp_l3mdev_accept = 0;
3170 #endif
3171 }
3172 
3173 static int __net_init udp_sysctl_init(struct net *net)
3174 {
3175 	__udp_sysctl_init(net);
3176 	return 0;
3177 }
3178 
3179 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3180 	.init	= udp_sysctl_init,
3181 };
3182 
3183 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3184 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3185 		     struct udp_sock *udp_sk, uid_t uid, int bucket)
3186 
3187 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3188 {
3189 	struct udp_iter_state *st = priv_data;
3190 	struct udp_seq_afinfo *afinfo;
3191 	int ret;
3192 
3193 	afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3194 	if (!afinfo)
3195 		return -ENOMEM;
3196 
3197 	afinfo->family = AF_UNSPEC;
3198 	afinfo->udp_table = &udp_table;
3199 	st->bpf_seq_afinfo = afinfo;
3200 	ret = bpf_iter_init_seq_net(priv_data, aux);
3201 	if (ret)
3202 		kfree(afinfo);
3203 	return ret;
3204 }
3205 
3206 static void bpf_iter_fini_udp(void *priv_data)
3207 {
3208 	struct udp_iter_state *st = priv_data;
3209 
3210 	kfree(st->bpf_seq_afinfo);
3211 	bpf_iter_fini_seq_net(priv_data);
3212 }
3213 
3214 static const struct bpf_iter_seq_info udp_seq_info = {
3215 	.seq_ops		= &bpf_iter_udp_seq_ops,
3216 	.init_seq_private	= bpf_iter_init_udp,
3217 	.fini_seq_private	= bpf_iter_fini_udp,
3218 	.seq_priv_size		= sizeof(struct udp_iter_state),
3219 };
3220 
3221 static struct bpf_iter_reg udp_reg_info = {
3222 	.target			= "udp",
3223 	.ctx_arg_info_size	= 1,
3224 	.ctx_arg_info		= {
3225 		{ offsetof(struct bpf_iter__udp, udp_sk),
3226 		  PTR_TO_BTF_ID_OR_NULL },
3227 	},
3228 	.seq_info		= &udp_seq_info,
3229 };
3230 
3231 static void __init bpf_iter_register(void)
3232 {
3233 	udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3234 	if (bpf_iter_reg_target(&udp_reg_info))
3235 		pr_warn("Warning: could not register bpf iterator udp\n");
3236 }
3237 #endif
3238 
3239 void __init udp_init(void)
3240 {
3241 	unsigned long limit;
3242 	unsigned int i;
3243 
3244 	udp_table_init(&udp_table, "UDP");
3245 	limit = nr_free_buffer_pages() / 8;
3246 	limit = max(limit, 128UL);
3247 	sysctl_udp_mem[0] = limit / 4 * 3;
3248 	sysctl_udp_mem[1] = limit;
3249 	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3250 
3251 	__udp_sysctl_init(&init_net);
3252 
3253 	/* 16 spinlocks per cpu */
3254 	udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3255 	udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3256 				GFP_KERNEL);
3257 	if (!udp_busylocks)
3258 		panic("UDP: failed to alloc udp_busylocks\n");
3259 	for (i = 0; i < (1U << udp_busylocks_log); i++)
3260 		spin_lock_init(udp_busylocks + i);
3261 
3262 	if (register_pernet_subsys(&udp_sysctl_ops))
3263 		panic("UDP: failed to init sysctl parameters.\n");
3264 
3265 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3266 	bpf_iter_register();
3267 #endif
3268 }
3269