xref: /openbmc/linux/net/ipv4/udp.c (revision b737eecd)
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 void udp_encap_disable(void)
600 {
601 	static_branch_dec(&udp_encap_needed_key);
602 }
603 EXPORT_SYMBOL(udp_encap_disable);
604 
605 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
606  * through error handlers in encapsulations looking for a match.
607  */
608 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
609 {
610 	int i;
611 
612 	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
613 		int (*handler)(struct sk_buff *skb, u32 info);
614 		const struct ip_tunnel_encap_ops *encap;
615 
616 		encap = rcu_dereference(iptun_encaps[i]);
617 		if (!encap)
618 			continue;
619 		handler = encap->err_handler;
620 		if (handler && !handler(skb, info))
621 			return 0;
622 	}
623 
624 	return -ENOENT;
625 }
626 
627 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
628  * reversing source and destination port: this will match tunnels that force the
629  * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
630  * lwtunnels might actually break this assumption by being configured with
631  * different destination ports on endpoints, in this case we won't be able to
632  * trace ICMP messages back to them.
633  *
634  * If this doesn't match any socket, probe tunnels with arbitrary destination
635  * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
636  * we've sent packets to won't necessarily match the local destination port.
637  *
638  * Then ask the tunnel implementation to match the error against a valid
639  * association.
640  *
641  * Return an error if we can't find a match, the socket if we need further
642  * processing, zero otherwise.
643  */
644 static struct sock *__udp4_lib_err_encap(struct net *net,
645 					 const struct iphdr *iph,
646 					 struct udphdr *uh,
647 					 struct udp_table *udptable,
648 					 struct sk_buff *skb, u32 info)
649 {
650 	int network_offset, transport_offset;
651 	struct sock *sk;
652 
653 	network_offset = skb_network_offset(skb);
654 	transport_offset = skb_transport_offset(skb);
655 
656 	/* Network header needs to point to the outer IPv4 header inside ICMP */
657 	skb_reset_network_header(skb);
658 
659 	/* Transport header needs to point to the UDP header */
660 	skb_set_transport_header(skb, iph->ihl << 2);
661 
662 	sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
663 			       iph->saddr, uh->dest, skb->dev->ifindex, 0,
664 			       udptable, NULL);
665 	if (sk) {
666 		int (*lookup)(struct sock *sk, struct sk_buff *skb);
667 		struct udp_sock *up = udp_sk(sk);
668 
669 		lookup = READ_ONCE(up->encap_err_lookup);
670 		if (!lookup || lookup(sk, skb))
671 			sk = NULL;
672 	}
673 
674 	if (!sk)
675 		sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
676 
677 	skb_set_transport_header(skb, transport_offset);
678 	skb_set_network_header(skb, network_offset);
679 
680 	return sk;
681 }
682 
683 /*
684  * This routine is called by the ICMP module when it gets some
685  * sort of error condition.  If err < 0 then the socket should
686  * be closed and the error returned to the user.  If err > 0
687  * it's just the icmp type << 8 | icmp code.
688  * Header points to the ip header of the error packet. We move
689  * on past this. Then (as it used to claim before adjustment)
690  * header points to the first 8 bytes of the udp header.  We need
691  * to find the appropriate port.
692  */
693 
694 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
695 {
696 	struct inet_sock *inet;
697 	const struct iphdr *iph = (const struct iphdr *)skb->data;
698 	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
699 	const int type = icmp_hdr(skb)->type;
700 	const int code = icmp_hdr(skb)->code;
701 	bool tunnel = false;
702 	struct sock *sk;
703 	int harderr;
704 	int err;
705 	struct net *net = dev_net(skb->dev);
706 
707 	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
708 			       iph->saddr, uh->source, skb->dev->ifindex,
709 			       inet_sdif(skb), udptable, NULL);
710 	if (!sk || udp_sk(sk)->encap_type) {
711 		/* No socket for error: try tunnels before discarding */
712 		sk = ERR_PTR(-ENOENT);
713 		if (static_branch_unlikely(&udp_encap_needed_key)) {
714 			sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
715 						  info);
716 			if (!sk)
717 				return 0;
718 		}
719 
720 		if (IS_ERR(sk)) {
721 			__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
722 			return PTR_ERR(sk);
723 		}
724 
725 		tunnel = true;
726 	}
727 
728 	err = 0;
729 	harderr = 0;
730 	inet = inet_sk(sk);
731 
732 	switch (type) {
733 	default:
734 	case ICMP_TIME_EXCEEDED:
735 		err = EHOSTUNREACH;
736 		break;
737 	case ICMP_SOURCE_QUENCH:
738 		goto out;
739 	case ICMP_PARAMETERPROB:
740 		err = EPROTO;
741 		harderr = 1;
742 		break;
743 	case ICMP_DEST_UNREACH:
744 		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
745 			ipv4_sk_update_pmtu(skb, sk, info);
746 			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
747 				err = EMSGSIZE;
748 				harderr = 1;
749 				break;
750 			}
751 			goto out;
752 		}
753 		err = EHOSTUNREACH;
754 		if (code <= NR_ICMP_UNREACH) {
755 			harderr = icmp_err_convert[code].fatal;
756 			err = icmp_err_convert[code].errno;
757 		}
758 		break;
759 	case ICMP_REDIRECT:
760 		ipv4_sk_redirect(skb, sk);
761 		goto out;
762 	}
763 
764 	/*
765 	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
766 	 *	4.1.3.3.
767 	 */
768 	if (tunnel) {
769 		/* ...not for tunnels though: we don't have a sending socket */
770 		goto out;
771 	}
772 	if (!inet->recverr) {
773 		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
774 			goto out;
775 	} else
776 		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
777 
778 	sk->sk_err = err;
779 	sk->sk_error_report(sk);
780 out:
781 	return 0;
782 }
783 
784 int udp_err(struct sk_buff *skb, u32 info)
785 {
786 	return __udp4_lib_err(skb, info, &udp_table);
787 }
788 
789 /*
790  * Throw away all pending data and cancel the corking. Socket is locked.
791  */
792 void udp_flush_pending_frames(struct sock *sk)
793 {
794 	struct udp_sock *up = udp_sk(sk);
795 
796 	if (up->pending) {
797 		up->len = 0;
798 		up->pending = 0;
799 		ip_flush_pending_frames(sk);
800 	}
801 }
802 EXPORT_SYMBOL(udp_flush_pending_frames);
803 
804 /**
805  * 	udp4_hwcsum  -  handle outgoing HW checksumming
806  * 	@skb: 	sk_buff containing the filled-in UDP header
807  * 	        (checksum field must be zeroed out)
808  *	@src:	source IP address
809  *	@dst:	destination IP address
810  */
811 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
812 {
813 	struct udphdr *uh = udp_hdr(skb);
814 	int offset = skb_transport_offset(skb);
815 	int len = skb->len - offset;
816 	int hlen = len;
817 	__wsum csum = 0;
818 
819 	if (!skb_has_frag_list(skb)) {
820 		/*
821 		 * Only one fragment on the socket.
822 		 */
823 		skb->csum_start = skb_transport_header(skb) - skb->head;
824 		skb->csum_offset = offsetof(struct udphdr, check);
825 		uh->check = ~csum_tcpudp_magic(src, dst, len,
826 					       IPPROTO_UDP, 0);
827 	} else {
828 		struct sk_buff *frags;
829 
830 		/*
831 		 * HW-checksum won't work as there are two or more
832 		 * fragments on the socket so that all csums of sk_buffs
833 		 * should be together
834 		 */
835 		skb_walk_frags(skb, frags) {
836 			csum = csum_add(csum, frags->csum);
837 			hlen -= frags->len;
838 		}
839 
840 		csum = skb_checksum(skb, offset, hlen, csum);
841 		skb->ip_summed = CHECKSUM_NONE;
842 
843 		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
844 		if (uh->check == 0)
845 			uh->check = CSUM_MANGLED_0;
846 	}
847 }
848 EXPORT_SYMBOL_GPL(udp4_hwcsum);
849 
850 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
851  * for the simple case like when setting the checksum for a UDP tunnel.
852  */
853 void udp_set_csum(bool nocheck, struct sk_buff *skb,
854 		  __be32 saddr, __be32 daddr, int len)
855 {
856 	struct udphdr *uh = udp_hdr(skb);
857 
858 	if (nocheck) {
859 		uh->check = 0;
860 	} else if (skb_is_gso(skb)) {
861 		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
862 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
863 		uh->check = 0;
864 		uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
865 		if (uh->check == 0)
866 			uh->check = CSUM_MANGLED_0;
867 	} else {
868 		skb->ip_summed = CHECKSUM_PARTIAL;
869 		skb->csum_start = skb_transport_header(skb) - skb->head;
870 		skb->csum_offset = offsetof(struct udphdr, check);
871 		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
872 	}
873 }
874 EXPORT_SYMBOL(udp_set_csum);
875 
876 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
877 			struct inet_cork *cork)
878 {
879 	struct sock *sk = skb->sk;
880 	struct inet_sock *inet = inet_sk(sk);
881 	struct udphdr *uh;
882 	int err;
883 	int is_udplite = IS_UDPLITE(sk);
884 	int offset = skb_transport_offset(skb);
885 	int len = skb->len - offset;
886 	int datalen = len - sizeof(*uh);
887 	__wsum csum = 0;
888 
889 	/*
890 	 * Create a UDP header
891 	 */
892 	uh = udp_hdr(skb);
893 	uh->source = inet->inet_sport;
894 	uh->dest = fl4->fl4_dport;
895 	uh->len = htons(len);
896 	uh->check = 0;
897 
898 	if (cork->gso_size) {
899 		const int hlen = skb_network_header_len(skb) +
900 				 sizeof(struct udphdr);
901 
902 		if (hlen + cork->gso_size > cork->fragsize) {
903 			kfree_skb(skb);
904 			return -EINVAL;
905 		}
906 		if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
907 			kfree_skb(skb);
908 			return -EINVAL;
909 		}
910 		if (sk->sk_no_check_tx) {
911 			kfree_skb(skb);
912 			return -EINVAL;
913 		}
914 		if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
915 		    dst_xfrm(skb_dst(skb))) {
916 			kfree_skb(skb);
917 			return -EIO;
918 		}
919 
920 		if (datalen > cork->gso_size) {
921 			skb_shinfo(skb)->gso_size = cork->gso_size;
922 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
923 			skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
924 								 cork->gso_size);
925 		}
926 		goto csum_partial;
927 	}
928 
929 	if (is_udplite)  				 /*     UDP-Lite      */
930 		csum = udplite_csum(skb);
931 
932 	else if (sk->sk_no_check_tx) {			 /* UDP csum off */
933 
934 		skb->ip_summed = CHECKSUM_NONE;
935 		goto send;
936 
937 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
938 csum_partial:
939 
940 		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
941 		goto send;
942 
943 	} else
944 		csum = udp_csum(skb);
945 
946 	/* add protocol-dependent pseudo-header */
947 	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
948 				      sk->sk_protocol, csum);
949 	if (uh->check == 0)
950 		uh->check = CSUM_MANGLED_0;
951 
952 send:
953 	err = ip_send_skb(sock_net(sk), skb);
954 	if (err) {
955 		if (err == -ENOBUFS && !inet->recverr) {
956 			UDP_INC_STATS(sock_net(sk),
957 				      UDP_MIB_SNDBUFERRORS, is_udplite);
958 			err = 0;
959 		}
960 	} else
961 		UDP_INC_STATS(sock_net(sk),
962 			      UDP_MIB_OUTDATAGRAMS, is_udplite);
963 	return err;
964 }
965 
966 /*
967  * Push out all pending data as one UDP datagram. Socket is locked.
968  */
969 int udp_push_pending_frames(struct sock *sk)
970 {
971 	struct udp_sock  *up = udp_sk(sk);
972 	struct inet_sock *inet = inet_sk(sk);
973 	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
974 	struct sk_buff *skb;
975 	int err = 0;
976 
977 	skb = ip_finish_skb(sk, fl4);
978 	if (!skb)
979 		goto out;
980 
981 	err = udp_send_skb(skb, fl4, &inet->cork.base);
982 
983 out:
984 	up->len = 0;
985 	up->pending = 0;
986 	return err;
987 }
988 EXPORT_SYMBOL(udp_push_pending_frames);
989 
990 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
991 {
992 	switch (cmsg->cmsg_type) {
993 	case UDP_SEGMENT:
994 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
995 			return -EINVAL;
996 		*gso_size = *(__u16 *)CMSG_DATA(cmsg);
997 		return 0;
998 	default:
999 		return -EINVAL;
1000 	}
1001 }
1002 
1003 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1004 {
1005 	struct cmsghdr *cmsg;
1006 	bool need_ip = false;
1007 	int err;
1008 
1009 	for_each_cmsghdr(cmsg, msg) {
1010 		if (!CMSG_OK(msg, cmsg))
1011 			return -EINVAL;
1012 
1013 		if (cmsg->cmsg_level != SOL_UDP) {
1014 			need_ip = true;
1015 			continue;
1016 		}
1017 
1018 		err = __udp_cmsg_send(cmsg, gso_size);
1019 		if (err)
1020 			return err;
1021 	}
1022 
1023 	return need_ip;
1024 }
1025 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1026 
1027 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1028 {
1029 	struct inet_sock *inet = inet_sk(sk);
1030 	struct udp_sock *up = udp_sk(sk);
1031 	DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1032 	struct flowi4 fl4_stack;
1033 	struct flowi4 *fl4;
1034 	int ulen = len;
1035 	struct ipcm_cookie ipc;
1036 	struct rtable *rt = NULL;
1037 	int free = 0;
1038 	int connected = 0;
1039 	__be32 daddr, faddr, saddr;
1040 	__be16 dport;
1041 	u8  tos;
1042 	int err, is_udplite = IS_UDPLITE(sk);
1043 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1044 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1045 	struct sk_buff *skb;
1046 	struct ip_options_data opt_copy;
1047 
1048 	if (len > 0xFFFF)
1049 		return -EMSGSIZE;
1050 
1051 	/*
1052 	 *	Check the flags.
1053 	 */
1054 
1055 	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1056 		return -EOPNOTSUPP;
1057 
1058 	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1059 
1060 	fl4 = &inet->cork.fl.u.ip4;
1061 	if (up->pending) {
1062 		/*
1063 		 * There are pending frames.
1064 		 * The socket lock must be held while it's corked.
1065 		 */
1066 		lock_sock(sk);
1067 		if (likely(up->pending)) {
1068 			if (unlikely(up->pending != AF_INET)) {
1069 				release_sock(sk);
1070 				return -EINVAL;
1071 			}
1072 			goto do_append_data;
1073 		}
1074 		release_sock(sk);
1075 	}
1076 	ulen += sizeof(struct udphdr);
1077 
1078 	/*
1079 	 *	Get and verify the address.
1080 	 */
1081 	if (usin) {
1082 		if (msg->msg_namelen < sizeof(*usin))
1083 			return -EINVAL;
1084 		if (usin->sin_family != AF_INET) {
1085 			if (usin->sin_family != AF_UNSPEC)
1086 				return -EAFNOSUPPORT;
1087 		}
1088 
1089 		daddr = usin->sin_addr.s_addr;
1090 		dport = usin->sin_port;
1091 		if (dport == 0)
1092 			return -EINVAL;
1093 	} else {
1094 		if (sk->sk_state != TCP_ESTABLISHED)
1095 			return -EDESTADDRREQ;
1096 		daddr = inet->inet_daddr;
1097 		dport = inet->inet_dport;
1098 		/* Open fast path for connected socket.
1099 		   Route will not be used, if at least one option is set.
1100 		 */
1101 		connected = 1;
1102 	}
1103 
1104 	ipcm_init_sk(&ipc, inet);
1105 	ipc.gso_size = up->gso_size;
1106 
1107 	if (msg->msg_controllen) {
1108 		err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1109 		if (err > 0)
1110 			err = ip_cmsg_send(sk, msg, &ipc,
1111 					   sk->sk_family == AF_INET6);
1112 		if (unlikely(err < 0)) {
1113 			kfree(ipc.opt);
1114 			return err;
1115 		}
1116 		if (ipc.opt)
1117 			free = 1;
1118 		connected = 0;
1119 	}
1120 	if (!ipc.opt) {
1121 		struct ip_options_rcu *inet_opt;
1122 
1123 		rcu_read_lock();
1124 		inet_opt = rcu_dereference(inet->inet_opt);
1125 		if (inet_opt) {
1126 			memcpy(&opt_copy, inet_opt,
1127 			       sizeof(*inet_opt) + inet_opt->opt.optlen);
1128 			ipc.opt = &opt_copy.opt;
1129 		}
1130 		rcu_read_unlock();
1131 	}
1132 
1133 	if (cgroup_bpf_enabled(BPF_CGROUP_UDP4_SENDMSG) && !connected) {
1134 		err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1135 					    (struct sockaddr *)usin, &ipc.addr);
1136 		if (err)
1137 			goto out_free;
1138 		if (usin) {
1139 			if (usin->sin_port == 0) {
1140 				/* BPF program set invalid port. Reject it. */
1141 				err = -EINVAL;
1142 				goto out_free;
1143 			}
1144 			daddr = usin->sin_addr.s_addr;
1145 			dport = usin->sin_port;
1146 		}
1147 	}
1148 
1149 	saddr = ipc.addr;
1150 	ipc.addr = faddr = daddr;
1151 
1152 	if (ipc.opt && ipc.opt->opt.srr) {
1153 		if (!daddr) {
1154 			err = -EINVAL;
1155 			goto out_free;
1156 		}
1157 		faddr = ipc.opt->opt.faddr;
1158 		connected = 0;
1159 	}
1160 	tos = get_rttos(&ipc, inet);
1161 	if (sock_flag(sk, SOCK_LOCALROUTE) ||
1162 	    (msg->msg_flags & MSG_DONTROUTE) ||
1163 	    (ipc.opt && ipc.opt->opt.is_strictroute)) {
1164 		tos |= RTO_ONLINK;
1165 		connected = 0;
1166 	}
1167 
1168 	if (ipv4_is_multicast(daddr)) {
1169 		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1170 			ipc.oif = inet->mc_index;
1171 		if (!saddr)
1172 			saddr = inet->mc_addr;
1173 		connected = 0;
1174 	} else if (!ipc.oif) {
1175 		ipc.oif = inet->uc_index;
1176 	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1177 		/* oif is set, packet is to local broadcast and
1178 		 * uc_index is set. oif is most likely set
1179 		 * by sk_bound_dev_if. If uc_index != oif check if the
1180 		 * oif is an L3 master and uc_index is an L3 slave.
1181 		 * If so, we want to allow the send using the uc_index.
1182 		 */
1183 		if (ipc.oif != inet->uc_index &&
1184 		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1185 							      inet->uc_index)) {
1186 			ipc.oif = inet->uc_index;
1187 		}
1188 	}
1189 
1190 	if (connected)
1191 		rt = (struct rtable *)sk_dst_check(sk, 0);
1192 
1193 	if (!rt) {
1194 		struct net *net = sock_net(sk);
1195 		__u8 flow_flags = inet_sk_flowi_flags(sk);
1196 
1197 		fl4 = &fl4_stack;
1198 
1199 		flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1200 				   RT_SCOPE_UNIVERSE, sk->sk_protocol,
1201 				   flow_flags,
1202 				   faddr, saddr, dport, inet->inet_sport,
1203 				   sk->sk_uid);
1204 
1205 		security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1206 		rt = ip_route_output_flow(net, fl4, sk);
1207 		if (IS_ERR(rt)) {
1208 			err = PTR_ERR(rt);
1209 			rt = NULL;
1210 			if (err == -ENETUNREACH)
1211 				IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1212 			goto out;
1213 		}
1214 
1215 		err = -EACCES;
1216 		if ((rt->rt_flags & RTCF_BROADCAST) &&
1217 		    !sock_flag(sk, SOCK_BROADCAST))
1218 			goto out;
1219 		if (connected)
1220 			sk_dst_set(sk, dst_clone(&rt->dst));
1221 	}
1222 
1223 	if (msg->msg_flags&MSG_CONFIRM)
1224 		goto do_confirm;
1225 back_from_confirm:
1226 
1227 	saddr = fl4->saddr;
1228 	if (!ipc.addr)
1229 		daddr = ipc.addr = fl4->daddr;
1230 
1231 	/* Lockless fast path for the non-corking case. */
1232 	if (!corkreq) {
1233 		struct inet_cork cork;
1234 
1235 		skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1236 				  sizeof(struct udphdr), &ipc, &rt,
1237 				  &cork, msg->msg_flags);
1238 		err = PTR_ERR(skb);
1239 		if (!IS_ERR_OR_NULL(skb))
1240 			err = udp_send_skb(skb, fl4, &cork);
1241 		goto out;
1242 	}
1243 
1244 	lock_sock(sk);
1245 	if (unlikely(up->pending)) {
1246 		/* The socket is already corked while preparing it. */
1247 		/* ... which is an evident application bug. --ANK */
1248 		release_sock(sk);
1249 
1250 		net_dbg_ratelimited("socket already corked\n");
1251 		err = -EINVAL;
1252 		goto out;
1253 	}
1254 	/*
1255 	 *	Now cork the socket to pend data.
1256 	 */
1257 	fl4 = &inet->cork.fl.u.ip4;
1258 	fl4->daddr = daddr;
1259 	fl4->saddr = saddr;
1260 	fl4->fl4_dport = dport;
1261 	fl4->fl4_sport = inet->inet_sport;
1262 	up->pending = AF_INET;
1263 
1264 do_append_data:
1265 	up->len += ulen;
1266 	err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1267 			     sizeof(struct udphdr), &ipc, &rt,
1268 			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1269 	if (err)
1270 		udp_flush_pending_frames(sk);
1271 	else if (!corkreq)
1272 		err = udp_push_pending_frames(sk);
1273 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1274 		up->pending = 0;
1275 	release_sock(sk);
1276 
1277 out:
1278 	ip_rt_put(rt);
1279 out_free:
1280 	if (free)
1281 		kfree(ipc.opt);
1282 	if (!err)
1283 		return len;
1284 	/*
1285 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1286 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1287 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1288 	 * things).  We could add another new stat but at least for now that
1289 	 * seems like overkill.
1290 	 */
1291 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1292 		UDP_INC_STATS(sock_net(sk),
1293 			      UDP_MIB_SNDBUFERRORS, is_udplite);
1294 	}
1295 	return err;
1296 
1297 do_confirm:
1298 	if (msg->msg_flags & MSG_PROBE)
1299 		dst_confirm_neigh(&rt->dst, &fl4->daddr);
1300 	if (!(msg->msg_flags&MSG_PROBE) || len)
1301 		goto back_from_confirm;
1302 	err = 0;
1303 	goto out;
1304 }
1305 EXPORT_SYMBOL(udp_sendmsg);
1306 
1307 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1308 		 size_t size, int flags)
1309 {
1310 	struct inet_sock *inet = inet_sk(sk);
1311 	struct udp_sock *up = udp_sk(sk);
1312 	int ret;
1313 
1314 	if (flags & MSG_SENDPAGE_NOTLAST)
1315 		flags |= MSG_MORE;
1316 
1317 	if (!up->pending) {
1318 		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
1319 
1320 		/* Call udp_sendmsg to specify destination address which
1321 		 * sendpage interface can't pass.
1322 		 * This will succeed only when the socket is connected.
1323 		 */
1324 		ret = udp_sendmsg(sk, &msg, 0);
1325 		if (ret < 0)
1326 			return ret;
1327 	}
1328 
1329 	lock_sock(sk);
1330 
1331 	if (unlikely(!up->pending)) {
1332 		release_sock(sk);
1333 
1334 		net_dbg_ratelimited("cork failed\n");
1335 		return -EINVAL;
1336 	}
1337 
1338 	ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1339 			     page, offset, size, flags);
1340 	if (ret == -EOPNOTSUPP) {
1341 		release_sock(sk);
1342 		return sock_no_sendpage(sk->sk_socket, page, offset,
1343 					size, flags);
1344 	}
1345 	if (ret < 0) {
1346 		udp_flush_pending_frames(sk);
1347 		goto out;
1348 	}
1349 
1350 	up->len += size;
1351 	if (!(up->corkflag || (flags&MSG_MORE)))
1352 		ret = udp_push_pending_frames(sk);
1353 	if (!ret)
1354 		ret = size;
1355 out:
1356 	release_sock(sk);
1357 	return ret;
1358 }
1359 
1360 #define UDP_SKB_IS_STATELESS 0x80000000
1361 
1362 /* all head states (dst, sk, nf conntrack) except skb extensions are
1363  * cleared by udp_rcv().
1364  *
1365  * We need to preserve secpath, if present, to eventually process
1366  * IP_CMSG_PASSSEC at recvmsg() time.
1367  *
1368  * Other extensions can be cleared.
1369  */
1370 static bool udp_try_make_stateless(struct sk_buff *skb)
1371 {
1372 	if (!skb_has_extensions(skb))
1373 		return true;
1374 
1375 	if (!secpath_exists(skb)) {
1376 		skb_ext_reset(skb);
1377 		return true;
1378 	}
1379 
1380 	return false;
1381 }
1382 
1383 static void udp_set_dev_scratch(struct sk_buff *skb)
1384 {
1385 	struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1386 
1387 	BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1388 	scratch->_tsize_state = skb->truesize;
1389 #if BITS_PER_LONG == 64
1390 	scratch->len = skb->len;
1391 	scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1392 	scratch->is_linear = !skb_is_nonlinear(skb);
1393 #endif
1394 	if (udp_try_make_stateless(skb))
1395 		scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1396 }
1397 
1398 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1399 {
1400 	/* We come here after udp_lib_checksum_complete() returned 0.
1401 	 * This means that __skb_checksum_complete() might have
1402 	 * set skb->csum_valid to 1.
1403 	 * On 64bit platforms, we can set csum_unnecessary
1404 	 * to true, but only if the skb is not shared.
1405 	 */
1406 #if BITS_PER_LONG == 64
1407 	if (!skb_shared(skb))
1408 		udp_skb_scratch(skb)->csum_unnecessary = true;
1409 #endif
1410 }
1411 
1412 static int udp_skb_truesize(struct sk_buff *skb)
1413 {
1414 	return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1415 }
1416 
1417 static bool udp_skb_has_head_state(struct sk_buff *skb)
1418 {
1419 	return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1420 }
1421 
1422 /* fully reclaim rmem/fwd memory allocated for skb */
1423 static void udp_rmem_release(struct sock *sk, int size, int partial,
1424 			     bool rx_queue_lock_held)
1425 {
1426 	struct udp_sock *up = udp_sk(sk);
1427 	struct sk_buff_head *sk_queue;
1428 	int amt;
1429 
1430 	if (likely(partial)) {
1431 		up->forward_deficit += size;
1432 		size = up->forward_deficit;
1433 		if (size < (sk->sk_rcvbuf >> 2) &&
1434 		    !skb_queue_empty(&up->reader_queue))
1435 			return;
1436 	} else {
1437 		size += up->forward_deficit;
1438 	}
1439 	up->forward_deficit = 0;
1440 
1441 	/* acquire the sk_receive_queue for fwd allocated memory scheduling,
1442 	 * if the called don't held it already
1443 	 */
1444 	sk_queue = &sk->sk_receive_queue;
1445 	if (!rx_queue_lock_held)
1446 		spin_lock(&sk_queue->lock);
1447 
1448 
1449 	sk->sk_forward_alloc += size;
1450 	amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1451 	sk->sk_forward_alloc -= amt;
1452 
1453 	if (amt)
1454 		__sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1455 
1456 	atomic_sub(size, &sk->sk_rmem_alloc);
1457 
1458 	/* this can save us from acquiring the rx queue lock on next receive */
1459 	skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1460 
1461 	if (!rx_queue_lock_held)
1462 		spin_unlock(&sk_queue->lock);
1463 }
1464 
1465 /* Note: called with reader_queue.lock held.
1466  * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1467  * This avoids a cache line miss while receive_queue lock is held.
1468  * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1469  */
1470 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1471 {
1472 	prefetch(&skb->data);
1473 	udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1474 }
1475 EXPORT_SYMBOL(udp_skb_destructor);
1476 
1477 /* as above, but the caller held the rx queue lock, too */
1478 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1479 {
1480 	prefetch(&skb->data);
1481 	udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1482 }
1483 
1484 /* Idea of busylocks is to let producers grab an extra spinlock
1485  * to relieve pressure on the receive_queue spinlock shared by consumer.
1486  * Under flood, this means that only one producer can be in line
1487  * trying to acquire the receive_queue spinlock.
1488  * These busylock can be allocated on a per cpu manner, instead of a
1489  * per socket one (that would consume a cache line per socket)
1490  */
1491 static int udp_busylocks_log __read_mostly;
1492 static spinlock_t *udp_busylocks __read_mostly;
1493 
1494 static spinlock_t *busylock_acquire(void *ptr)
1495 {
1496 	spinlock_t *busy;
1497 
1498 	busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1499 	spin_lock(busy);
1500 	return busy;
1501 }
1502 
1503 static void busylock_release(spinlock_t *busy)
1504 {
1505 	if (busy)
1506 		spin_unlock(busy);
1507 }
1508 
1509 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1510 {
1511 	struct sk_buff_head *list = &sk->sk_receive_queue;
1512 	int rmem, delta, amt, err = -ENOMEM;
1513 	spinlock_t *busy = NULL;
1514 	int size;
1515 
1516 	/* try to avoid the costly atomic add/sub pair when the receive
1517 	 * queue is full; always allow at least a packet
1518 	 */
1519 	rmem = atomic_read(&sk->sk_rmem_alloc);
1520 	if (rmem > sk->sk_rcvbuf)
1521 		goto drop;
1522 
1523 	/* Under mem pressure, it might be helpful to help udp_recvmsg()
1524 	 * having linear skbs :
1525 	 * - Reduce memory overhead and thus increase receive queue capacity
1526 	 * - Less cache line misses at copyout() time
1527 	 * - Less work at consume_skb() (less alien page frag freeing)
1528 	 */
1529 	if (rmem > (sk->sk_rcvbuf >> 1)) {
1530 		skb_condense(skb);
1531 
1532 		busy = busylock_acquire(sk);
1533 	}
1534 	size = skb->truesize;
1535 	udp_set_dev_scratch(skb);
1536 
1537 	/* we drop only if the receive buf is full and the receive
1538 	 * queue contains some other skb
1539 	 */
1540 	rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1541 	if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1542 		goto uncharge_drop;
1543 
1544 	spin_lock(&list->lock);
1545 	if (size >= sk->sk_forward_alloc) {
1546 		amt = sk_mem_pages(size);
1547 		delta = amt << SK_MEM_QUANTUM_SHIFT;
1548 		if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1549 			err = -ENOBUFS;
1550 			spin_unlock(&list->lock);
1551 			goto uncharge_drop;
1552 		}
1553 
1554 		sk->sk_forward_alloc += delta;
1555 	}
1556 
1557 	sk->sk_forward_alloc -= size;
1558 
1559 	/* no need to setup a destructor, we will explicitly release the
1560 	 * forward allocated memory on dequeue
1561 	 */
1562 	sock_skb_set_dropcount(sk, skb);
1563 
1564 	__skb_queue_tail(list, skb);
1565 	spin_unlock(&list->lock);
1566 
1567 	if (!sock_flag(sk, SOCK_DEAD))
1568 		sk->sk_data_ready(sk);
1569 
1570 	busylock_release(busy);
1571 	return 0;
1572 
1573 uncharge_drop:
1574 	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1575 
1576 drop:
1577 	atomic_inc(&sk->sk_drops);
1578 	busylock_release(busy);
1579 	return err;
1580 }
1581 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1582 
1583 void udp_destruct_sock(struct sock *sk)
1584 {
1585 	/* reclaim completely the forward allocated memory */
1586 	struct udp_sock *up = udp_sk(sk);
1587 	unsigned int total = 0;
1588 	struct sk_buff *skb;
1589 
1590 	skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1591 	while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1592 		total += skb->truesize;
1593 		kfree_skb(skb);
1594 	}
1595 	udp_rmem_release(sk, total, 0, true);
1596 
1597 	inet_sock_destruct(sk);
1598 }
1599 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1600 
1601 int udp_init_sock(struct sock *sk)
1602 {
1603 	skb_queue_head_init(&udp_sk(sk)->reader_queue);
1604 	sk->sk_destruct = udp_destruct_sock;
1605 	return 0;
1606 }
1607 EXPORT_SYMBOL_GPL(udp_init_sock);
1608 
1609 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1610 {
1611 	if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1612 		bool slow = lock_sock_fast(sk);
1613 
1614 		sk_peek_offset_bwd(sk, len);
1615 		unlock_sock_fast(sk, slow);
1616 	}
1617 
1618 	if (!skb_unref(skb))
1619 		return;
1620 
1621 	/* In the more common cases we cleared the head states previously,
1622 	 * see __udp_queue_rcv_skb().
1623 	 */
1624 	if (unlikely(udp_skb_has_head_state(skb)))
1625 		skb_release_head_state(skb);
1626 	__consume_stateless_skb(skb);
1627 }
1628 EXPORT_SYMBOL_GPL(skb_consume_udp);
1629 
1630 static struct sk_buff *__first_packet_length(struct sock *sk,
1631 					     struct sk_buff_head *rcvq,
1632 					     int *total)
1633 {
1634 	struct sk_buff *skb;
1635 
1636 	while ((skb = skb_peek(rcvq)) != NULL) {
1637 		if (udp_lib_checksum_complete(skb)) {
1638 			__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1639 					IS_UDPLITE(sk));
1640 			__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1641 					IS_UDPLITE(sk));
1642 			atomic_inc(&sk->sk_drops);
1643 			__skb_unlink(skb, rcvq);
1644 			*total += skb->truesize;
1645 			kfree_skb(skb);
1646 		} else {
1647 			udp_skb_csum_unnecessary_set(skb);
1648 			break;
1649 		}
1650 	}
1651 	return skb;
1652 }
1653 
1654 /**
1655  *	first_packet_length	- return length of first packet in receive queue
1656  *	@sk: socket
1657  *
1658  *	Drops all bad checksum frames, until a valid one is found.
1659  *	Returns the length of found skb, or -1 if none is found.
1660  */
1661 static int first_packet_length(struct sock *sk)
1662 {
1663 	struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1664 	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1665 	struct sk_buff *skb;
1666 	int total = 0;
1667 	int res;
1668 
1669 	spin_lock_bh(&rcvq->lock);
1670 	skb = __first_packet_length(sk, rcvq, &total);
1671 	if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1672 		spin_lock(&sk_queue->lock);
1673 		skb_queue_splice_tail_init(sk_queue, rcvq);
1674 		spin_unlock(&sk_queue->lock);
1675 
1676 		skb = __first_packet_length(sk, rcvq, &total);
1677 	}
1678 	res = skb ? skb->len : -1;
1679 	if (total)
1680 		udp_rmem_release(sk, total, 1, false);
1681 	spin_unlock_bh(&rcvq->lock);
1682 	return res;
1683 }
1684 
1685 /*
1686  *	IOCTL requests applicable to the UDP protocol
1687  */
1688 
1689 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1690 {
1691 	switch (cmd) {
1692 	case SIOCOUTQ:
1693 	{
1694 		int amount = sk_wmem_alloc_get(sk);
1695 
1696 		return put_user(amount, (int __user *)arg);
1697 	}
1698 
1699 	case SIOCINQ:
1700 	{
1701 		int amount = max_t(int, 0, first_packet_length(sk));
1702 
1703 		return put_user(amount, (int __user *)arg);
1704 	}
1705 
1706 	default:
1707 		return -ENOIOCTLCMD;
1708 	}
1709 
1710 	return 0;
1711 }
1712 EXPORT_SYMBOL(udp_ioctl);
1713 
1714 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1715 			       int noblock, int *off, int *err)
1716 {
1717 	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1718 	struct sk_buff_head *queue;
1719 	struct sk_buff *last;
1720 	long timeo;
1721 	int error;
1722 
1723 	queue = &udp_sk(sk)->reader_queue;
1724 	flags |= noblock ? MSG_DONTWAIT : 0;
1725 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1726 	do {
1727 		struct sk_buff *skb;
1728 
1729 		error = sock_error(sk);
1730 		if (error)
1731 			break;
1732 
1733 		error = -EAGAIN;
1734 		do {
1735 			spin_lock_bh(&queue->lock);
1736 			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1737 							err, &last);
1738 			if (skb) {
1739 				if (!(flags & MSG_PEEK))
1740 					udp_skb_destructor(sk, skb);
1741 				spin_unlock_bh(&queue->lock);
1742 				return skb;
1743 			}
1744 
1745 			if (skb_queue_empty_lockless(sk_queue)) {
1746 				spin_unlock_bh(&queue->lock);
1747 				goto busy_check;
1748 			}
1749 
1750 			/* refill the reader queue and walk it again
1751 			 * keep both queues locked to avoid re-acquiring
1752 			 * the sk_receive_queue lock if fwd memory scheduling
1753 			 * is needed.
1754 			 */
1755 			spin_lock(&sk_queue->lock);
1756 			skb_queue_splice_tail_init(sk_queue, queue);
1757 
1758 			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1759 							err, &last);
1760 			if (skb && !(flags & MSG_PEEK))
1761 				udp_skb_dtor_locked(sk, skb);
1762 			spin_unlock(&sk_queue->lock);
1763 			spin_unlock_bh(&queue->lock);
1764 			if (skb)
1765 				return skb;
1766 
1767 busy_check:
1768 			if (!sk_can_busy_loop(sk))
1769 				break;
1770 
1771 			sk_busy_loop(sk, flags & MSG_DONTWAIT);
1772 		} while (!skb_queue_empty_lockless(sk_queue));
1773 
1774 		/* sk_queue is empty, reader_queue may contain peeked packets */
1775 	} while (timeo &&
1776 		 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1777 					      &error, &timeo,
1778 					      (struct sk_buff *)sk_queue));
1779 
1780 	*err = error;
1781 	return NULL;
1782 }
1783 EXPORT_SYMBOL(__skb_recv_udp);
1784 
1785 /*
1786  * 	This should be easy, if there is something there we
1787  * 	return it, otherwise we block.
1788  */
1789 
1790 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1791 		int flags, int *addr_len)
1792 {
1793 	struct inet_sock *inet = inet_sk(sk);
1794 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1795 	struct sk_buff *skb;
1796 	unsigned int ulen, copied;
1797 	int off, err, peeking = flags & MSG_PEEK;
1798 	int is_udplite = IS_UDPLITE(sk);
1799 	bool checksum_valid = false;
1800 
1801 	if (flags & MSG_ERRQUEUE)
1802 		return ip_recv_error(sk, msg, len, addr_len);
1803 
1804 try_again:
1805 	off = sk_peek_offset(sk, flags);
1806 	skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1807 	if (!skb)
1808 		return err;
1809 
1810 	ulen = udp_skb_len(skb);
1811 	copied = len;
1812 	if (copied > ulen - off)
1813 		copied = ulen - off;
1814 	else if (copied < ulen)
1815 		msg->msg_flags |= MSG_TRUNC;
1816 
1817 	/*
1818 	 * If checksum is needed at all, try to do it while copying the
1819 	 * data.  If the data is truncated, or if we only want a partial
1820 	 * coverage checksum (UDP-Lite), do it before the copy.
1821 	 */
1822 
1823 	if (copied < ulen || peeking ||
1824 	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1825 		checksum_valid = udp_skb_csum_unnecessary(skb) ||
1826 				!__udp_lib_checksum_complete(skb);
1827 		if (!checksum_valid)
1828 			goto csum_copy_err;
1829 	}
1830 
1831 	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1832 		if (udp_skb_is_linear(skb))
1833 			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1834 		else
1835 			err = skb_copy_datagram_msg(skb, off, msg, copied);
1836 	} else {
1837 		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1838 
1839 		if (err == -EINVAL)
1840 			goto csum_copy_err;
1841 	}
1842 
1843 	if (unlikely(err)) {
1844 		if (!peeking) {
1845 			atomic_inc(&sk->sk_drops);
1846 			UDP_INC_STATS(sock_net(sk),
1847 				      UDP_MIB_INERRORS, is_udplite);
1848 		}
1849 		kfree_skb(skb);
1850 		return err;
1851 	}
1852 
1853 	if (!peeking)
1854 		UDP_INC_STATS(sock_net(sk),
1855 			      UDP_MIB_INDATAGRAMS, is_udplite);
1856 
1857 	sock_recv_ts_and_drops(msg, sk, skb);
1858 
1859 	/* Copy the address. */
1860 	if (sin) {
1861 		sin->sin_family = AF_INET;
1862 		sin->sin_port = udp_hdr(skb)->source;
1863 		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1864 		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1865 		*addr_len = sizeof(*sin);
1866 
1867 		BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1868 						      (struct sockaddr *)sin);
1869 	}
1870 
1871 	if (udp_sk(sk)->gro_enabled)
1872 		udp_cmsg_recv(msg, sk, skb);
1873 
1874 	if (inet->cmsg_flags)
1875 		ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1876 
1877 	err = copied;
1878 	if (flags & MSG_TRUNC)
1879 		err = ulen;
1880 
1881 	skb_consume_udp(sk, skb, peeking ? -err : err);
1882 	return err;
1883 
1884 csum_copy_err:
1885 	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1886 				 udp_skb_destructor)) {
1887 		UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1888 		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1889 	}
1890 	kfree_skb(skb);
1891 
1892 	/* starting over for a new packet, but check if we need to yield */
1893 	cond_resched();
1894 	msg->msg_flags &= ~MSG_TRUNC;
1895 	goto try_again;
1896 }
1897 
1898 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1899 {
1900 	/* This check is replicated from __ip4_datagram_connect() and
1901 	 * intended to prevent BPF program called below from accessing bytes
1902 	 * that are out of the bound specified by user in addr_len.
1903 	 */
1904 	if (addr_len < sizeof(struct sockaddr_in))
1905 		return -EINVAL;
1906 
1907 	return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1908 }
1909 EXPORT_SYMBOL(udp_pre_connect);
1910 
1911 int __udp_disconnect(struct sock *sk, int flags)
1912 {
1913 	struct inet_sock *inet = inet_sk(sk);
1914 	/*
1915 	 *	1003.1g - break association.
1916 	 */
1917 
1918 	sk->sk_state = TCP_CLOSE;
1919 	inet->inet_daddr = 0;
1920 	inet->inet_dport = 0;
1921 	sock_rps_reset_rxhash(sk);
1922 	sk->sk_bound_dev_if = 0;
1923 	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1924 		inet_reset_saddr(sk);
1925 		if (sk->sk_prot->rehash &&
1926 		    (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1927 			sk->sk_prot->rehash(sk);
1928 	}
1929 
1930 	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1931 		sk->sk_prot->unhash(sk);
1932 		inet->inet_sport = 0;
1933 	}
1934 	sk_dst_reset(sk);
1935 	return 0;
1936 }
1937 EXPORT_SYMBOL(__udp_disconnect);
1938 
1939 int udp_disconnect(struct sock *sk, int flags)
1940 {
1941 	lock_sock(sk);
1942 	__udp_disconnect(sk, flags);
1943 	release_sock(sk);
1944 	return 0;
1945 }
1946 EXPORT_SYMBOL(udp_disconnect);
1947 
1948 void udp_lib_unhash(struct sock *sk)
1949 {
1950 	if (sk_hashed(sk)) {
1951 		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1952 		struct udp_hslot *hslot, *hslot2;
1953 
1954 		hslot  = udp_hashslot(udptable, sock_net(sk),
1955 				      udp_sk(sk)->udp_port_hash);
1956 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1957 
1958 		spin_lock_bh(&hslot->lock);
1959 		if (rcu_access_pointer(sk->sk_reuseport_cb))
1960 			reuseport_detach_sock(sk);
1961 		if (sk_del_node_init_rcu(sk)) {
1962 			hslot->count--;
1963 			inet_sk(sk)->inet_num = 0;
1964 			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1965 
1966 			spin_lock(&hslot2->lock);
1967 			hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1968 			hslot2->count--;
1969 			spin_unlock(&hslot2->lock);
1970 		}
1971 		spin_unlock_bh(&hslot->lock);
1972 	}
1973 }
1974 EXPORT_SYMBOL(udp_lib_unhash);
1975 
1976 /*
1977  * inet_rcv_saddr was changed, we must rehash secondary hash
1978  */
1979 void udp_lib_rehash(struct sock *sk, u16 newhash)
1980 {
1981 	if (sk_hashed(sk)) {
1982 		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1983 		struct udp_hslot *hslot, *hslot2, *nhslot2;
1984 
1985 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1986 		nhslot2 = udp_hashslot2(udptable, newhash);
1987 		udp_sk(sk)->udp_portaddr_hash = newhash;
1988 
1989 		if (hslot2 != nhslot2 ||
1990 		    rcu_access_pointer(sk->sk_reuseport_cb)) {
1991 			hslot = udp_hashslot(udptable, sock_net(sk),
1992 					     udp_sk(sk)->udp_port_hash);
1993 			/* we must lock primary chain too */
1994 			spin_lock_bh(&hslot->lock);
1995 			if (rcu_access_pointer(sk->sk_reuseport_cb))
1996 				reuseport_detach_sock(sk);
1997 
1998 			if (hslot2 != nhslot2) {
1999 				spin_lock(&hslot2->lock);
2000 				hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2001 				hslot2->count--;
2002 				spin_unlock(&hslot2->lock);
2003 
2004 				spin_lock(&nhslot2->lock);
2005 				hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2006 							 &nhslot2->head);
2007 				nhslot2->count++;
2008 				spin_unlock(&nhslot2->lock);
2009 			}
2010 
2011 			spin_unlock_bh(&hslot->lock);
2012 		}
2013 	}
2014 }
2015 EXPORT_SYMBOL(udp_lib_rehash);
2016 
2017 void udp_v4_rehash(struct sock *sk)
2018 {
2019 	u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2020 					  inet_sk(sk)->inet_rcv_saddr,
2021 					  inet_sk(sk)->inet_num);
2022 	udp_lib_rehash(sk, new_hash);
2023 }
2024 
2025 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2026 {
2027 	int rc;
2028 
2029 	if (inet_sk(sk)->inet_daddr) {
2030 		sock_rps_save_rxhash(sk, skb);
2031 		sk_mark_napi_id(sk, skb);
2032 		sk_incoming_cpu_update(sk);
2033 	} else {
2034 		sk_mark_napi_id_once(sk, skb);
2035 	}
2036 
2037 	rc = __udp_enqueue_schedule_skb(sk, skb);
2038 	if (rc < 0) {
2039 		int is_udplite = IS_UDPLITE(sk);
2040 
2041 		/* Note that an ENOMEM error is charged twice */
2042 		if (rc == -ENOMEM)
2043 			UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2044 					is_udplite);
2045 		else
2046 			UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2047 				      is_udplite);
2048 		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2049 		kfree_skb(skb);
2050 		trace_udp_fail_queue_rcv_skb(rc, sk);
2051 		return -1;
2052 	}
2053 
2054 	return 0;
2055 }
2056 
2057 /* returns:
2058  *  -1: error
2059  *   0: success
2060  *  >0: "udp encap" protocol resubmission
2061  *
2062  * Note that in the success and error cases, the skb is assumed to
2063  * have either been requeued or freed.
2064  */
2065 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2066 {
2067 	struct udp_sock *up = udp_sk(sk);
2068 	int is_udplite = IS_UDPLITE(sk);
2069 
2070 	/*
2071 	 *	Charge it to the socket, dropping if the queue is full.
2072 	 */
2073 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2074 		goto drop;
2075 	nf_reset_ct(skb);
2076 
2077 	if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2078 		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2079 
2080 		/*
2081 		 * This is an encapsulation socket so pass the skb to
2082 		 * the socket's udp_encap_rcv() hook. Otherwise, just
2083 		 * fall through and pass this up the UDP socket.
2084 		 * up->encap_rcv() returns the following value:
2085 		 * =0 if skb was successfully passed to the encap
2086 		 *    handler or was discarded by it.
2087 		 * >0 if skb should be passed on to UDP.
2088 		 * <0 if skb should be resubmitted as proto -N
2089 		 */
2090 
2091 		/* if we're overly short, let UDP handle it */
2092 		encap_rcv = READ_ONCE(up->encap_rcv);
2093 		if (encap_rcv) {
2094 			int ret;
2095 
2096 			/* Verify checksum before giving to encap */
2097 			if (udp_lib_checksum_complete(skb))
2098 				goto csum_error;
2099 
2100 			ret = encap_rcv(sk, skb);
2101 			if (ret <= 0) {
2102 				__UDP_INC_STATS(sock_net(sk),
2103 						UDP_MIB_INDATAGRAMS,
2104 						is_udplite);
2105 				return -ret;
2106 			}
2107 		}
2108 
2109 		/* FALLTHROUGH -- it's a UDP Packet */
2110 	}
2111 
2112 	/*
2113 	 * 	UDP-Lite specific tests, ignored on UDP sockets
2114 	 */
2115 	if ((up->pcflag & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
2116 
2117 		/*
2118 		 * MIB statistics other than incrementing the error count are
2119 		 * disabled for the following two types of errors: these depend
2120 		 * on the application settings, not on the functioning of the
2121 		 * protocol stack as such.
2122 		 *
2123 		 * RFC 3828 here recommends (sec 3.3): "There should also be a
2124 		 * way ... to ... at least let the receiving application block
2125 		 * delivery of packets with coverage values less than a value
2126 		 * provided by the application."
2127 		 */
2128 		if (up->pcrlen == 0) {          /* full coverage was set  */
2129 			net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2130 					    UDP_SKB_CB(skb)->cscov, skb->len);
2131 			goto drop;
2132 		}
2133 		/* The next case involves violating the min. coverage requested
2134 		 * by the receiver. This is subtle: if receiver wants x and x is
2135 		 * greater than the buffersize/MTU then receiver will complain
2136 		 * that it wants x while sender emits packets of smaller size y.
2137 		 * Therefore the above ...()->partial_cov statement is essential.
2138 		 */
2139 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
2140 			net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2141 					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
2142 			goto drop;
2143 		}
2144 	}
2145 
2146 	prefetch(&sk->sk_rmem_alloc);
2147 	if (rcu_access_pointer(sk->sk_filter) &&
2148 	    udp_lib_checksum_complete(skb))
2149 			goto csum_error;
2150 
2151 	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2152 		goto drop;
2153 
2154 	udp_csum_pull_header(skb);
2155 
2156 	ipv4_pktinfo_prepare(sk, skb);
2157 	return __udp_queue_rcv_skb(sk, skb);
2158 
2159 csum_error:
2160 	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2161 drop:
2162 	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2163 	atomic_inc(&sk->sk_drops);
2164 	kfree_skb(skb);
2165 	return -1;
2166 }
2167 
2168 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2169 {
2170 	struct sk_buff *next, *segs;
2171 	int ret;
2172 
2173 	if (likely(!udp_unexpected_gso(sk, skb)))
2174 		return udp_queue_rcv_one_skb(sk, skb);
2175 
2176 	BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2177 	__skb_push(skb, -skb_mac_offset(skb));
2178 	segs = udp_rcv_segment(sk, skb, true);
2179 	skb_list_walk_safe(segs, skb, next) {
2180 		__skb_pull(skb, skb_transport_offset(skb));
2181 		ret = udp_queue_rcv_one_skb(sk, skb);
2182 		if (ret > 0)
2183 			ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2184 	}
2185 	return 0;
2186 }
2187 
2188 /* For TCP sockets, sk_rx_dst is protected by socket lock
2189  * For UDP, we use xchg() to guard against concurrent changes.
2190  */
2191 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2192 {
2193 	struct dst_entry *old;
2194 
2195 	if (dst_hold_safe(dst)) {
2196 		old = xchg(&sk->sk_rx_dst, dst);
2197 		dst_release(old);
2198 		return old != dst;
2199 	}
2200 	return false;
2201 }
2202 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2203 
2204 /*
2205  *	Multicasts and broadcasts go to each listener.
2206  *
2207  *	Note: called only from the BH handler context.
2208  */
2209 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2210 				    struct udphdr  *uh,
2211 				    __be32 saddr, __be32 daddr,
2212 				    struct udp_table *udptable,
2213 				    int proto)
2214 {
2215 	struct sock *sk, *first = NULL;
2216 	unsigned short hnum = ntohs(uh->dest);
2217 	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2218 	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2219 	unsigned int offset = offsetof(typeof(*sk), sk_node);
2220 	int dif = skb->dev->ifindex;
2221 	int sdif = inet_sdif(skb);
2222 	struct hlist_node *node;
2223 	struct sk_buff *nskb;
2224 
2225 	if (use_hash2) {
2226 		hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2227 			    udptable->mask;
2228 		hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2229 start_lookup:
2230 		hslot = &udptable->hash2[hash2];
2231 		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2232 	}
2233 
2234 	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2235 		if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2236 					 uh->source, saddr, dif, sdif, hnum))
2237 			continue;
2238 
2239 		if (!first) {
2240 			first = sk;
2241 			continue;
2242 		}
2243 		nskb = skb_clone(skb, GFP_ATOMIC);
2244 
2245 		if (unlikely(!nskb)) {
2246 			atomic_inc(&sk->sk_drops);
2247 			__UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2248 					IS_UDPLITE(sk));
2249 			__UDP_INC_STATS(net, UDP_MIB_INERRORS,
2250 					IS_UDPLITE(sk));
2251 			continue;
2252 		}
2253 		if (udp_queue_rcv_skb(sk, nskb) > 0)
2254 			consume_skb(nskb);
2255 	}
2256 
2257 	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
2258 	if (use_hash2 && hash2 != hash2_any) {
2259 		hash2 = hash2_any;
2260 		goto start_lookup;
2261 	}
2262 
2263 	if (first) {
2264 		if (udp_queue_rcv_skb(first, skb) > 0)
2265 			consume_skb(skb);
2266 	} else {
2267 		kfree_skb(skb);
2268 		__UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2269 				proto == IPPROTO_UDPLITE);
2270 	}
2271 	return 0;
2272 }
2273 
2274 /* Initialize UDP checksum. If exited with zero value (success),
2275  * CHECKSUM_UNNECESSARY means, that no more checks are required.
2276  * Otherwise, csum completion requires checksumming packet body,
2277  * including udp header and folding it to skb->csum.
2278  */
2279 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2280 				 int proto)
2281 {
2282 	int err;
2283 
2284 	UDP_SKB_CB(skb)->partial_cov = 0;
2285 	UDP_SKB_CB(skb)->cscov = skb->len;
2286 
2287 	if (proto == IPPROTO_UDPLITE) {
2288 		err = udplite_checksum_init(skb, uh);
2289 		if (err)
2290 			return err;
2291 
2292 		if (UDP_SKB_CB(skb)->partial_cov) {
2293 			skb->csum = inet_compute_pseudo(skb, proto);
2294 			return 0;
2295 		}
2296 	}
2297 
2298 	/* Note, we are only interested in != 0 or == 0, thus the
2299 	 * force to int.
2300 	 */
2301 	err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2302 							inet_compute_pseudo);
2303 	if (err)
2304 		return err;
2305 
2306 	if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2307 		/* If SW calculated the value, we know it's bad */
2308 		if (skb->csum_complete_sw)
2309 			return 1;
2310 
2311 		/* HW says the value is bad. Let's validate that.
2312 		 * skb->csum is no longer the full packet checksum,
2313 		 * so don't treat it as such.
2314 		 */
2315 		skb_checksum_complete_unset(skb);
2316 	}
2317 
2318 	return 0;
2319 }
2320 
2321 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2322  * return code conversion for ip layer consumption
2323  */
2324 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2325 			       struct udphdr *uh)
2326 {
2327 	int ret;
2328 
2329 	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2330 		skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2331 
2332 	ret = udp_queue_rcv_skb(sk, skb);
2333 
2334 	/* a return value > 0 means to resubmit the input, but
2335 	 * it wants the return to be -protocol, or 0
2336 	 */
2337 	if (ret > 0)
2338 		return -ret;
2339 	return 0;
2340 }
2341 
2342 /*
2343  *	All we need to do is get the socket, and then do a checksum.
2344  */
2345 
2346 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2347 		   int proto)
2348 {
2349 	struct sock *sk;
2350 	struct udphdr *uh;
2351 	unsigned short ulen;
2352 	struct rtable *rt = skb_rtable(skb);
2353 	__be32 saddr, daddr;
2354 	struct net *net = dev_net(skb->dev);
2355 	bool refcounted;
2356 
2357 	/*
2358 	 *  Validate the packet.
2359 	 */
2360 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2361 		goto drop;		/* No space for header. */
2362 
2363 	uh   = udp_hdr(skb);
2364 	ulen = ntohs(uh->len);
2365 	saddr = ip_hdr(skb)->saddr;
2366 	daddr = ip_hdr(skb)->daddr;
2367 
2368 	if (ulen > skb->len)
2369 		goto short_packet;
2370 
2371 	if (proto == IPPROTO_UDP) {
2372 		/* UDP validates ulen. */
2373 		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2374 			goto short_packet;
2375 		uh = udp_hdr(skb);
2376 	}
2377 
2378 	if (udp4_csum_init(skb, uh, proto))
2379 		goto csum_error;
2380 
2381 	sk = skb_steal_sock(skb, &refcounted);
2382 	if (sk) {
2383 		struct dst_entry *dst = skb_dst(skb);
2384 		int ret;
2385 
2386 		if (unlikely(sk->sk_rx_dst != dst))
2387 			udp_sk_rx_dst_set(sk, dst);
2388 
2389 		ret = udp_unicast_rcv_skb(sk, skb, uh);
2390 		if (refcounted)
2391 			sock_put(sk);
2392 		return ret;
2393 	}
2394 
2395 	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2396 		return __udp4_lib_mcast_deliver(net, skb, uh,
2397 						saddr, daddr, udptable, proto);
2398 
2399 	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2400 	if (sk)
2401 		return udp_unicast_rcv_skb(sk, skb, uh);
2402 
2403 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2404 		goto drop;
2405 	nf_reset_ct(skb);
2406 
2407 	/* No socket. Drop packet silently, if checksum is wrong */
2408 	if (udp_lib_checksum_complete(skb))
2409 		goto csum_error;
2410 
2411 	__UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2412 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2413 
2414 	/*
2415 	 * Hmm.  We got an UDP packet to a port to which we
2416 	 * don't wanna listen.  Ignore it.
2417 	 */
2418 	kfree_skb(skb);
2419 	return 0;
2420 
2421 short_packet:
2422 	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2423 			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2424 			    &saddr, ntohs(uh->source),
2425 			    ulen, skb->len,
2426 			    &daddr, ntohs(uh->dest));
2427 	goto drop;
2428 
2429 csum_error:
2430 	/*
2431 	 * RFC1122: OK.  Discards the bad packet silently (as far as
2432 	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2433 	 */
2434 	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2435 			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2436 			    &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2437 			    ulen);
2438 	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2439 drop:
2440 	__UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2441 	kfree_skb(skb);
2442 	return 0;
2443 }
2444 
2445 /* We can only early demux multicast if there is a single matching socket.
2446  * If more than one socket found returns NULL
2447  */
2448 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2449 						  __be16 loc_port, __be32 loc_addr,
2450 						  __be16 rmt_port, __be32 rmt_addr,
2451 						  int dif, int sdif)
2452 {
2453 	struct sock *sk, *result;
2454 	unsigned short hnum = ntohs(loc_port);
2455 	unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2456 	struct udp_hslot *hslot = &udp_table.hash[slot];
2457 
2458 	/* Do not bother scanning a too big list */
2459 	if (hslot->count > 10)
2460 		return NULL;
2461 
2462 	result = NULL;
2463 	sk_for_each_rcu(sk, &hslot->head) {
2464 		if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2465 					rmt_port, rmt_addr, dif, sdif, hnum)) {
2466 			if (result)
2467 				return NULL;
2468 			result = sk;
2469 		}
2470 	}
2471 
2472 	return result;
2473 }
2474 
2475 /* For unicast we should only early demux connected sockets or we can
2476  * break forwarding setups.  The chains here can be long so only check
2477  * if the first socket is an exact match and if not move on.
2478  */
2479 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2480 					    __be16 loc_port, __be32 loc_addr,
2481 					    __be16 rmt_port, __be32 rmt_addr,
2482 					    int dif, int sdif)
2483 {
2484 	unsigned short hnum = ntohs(loc_port);
2485 	unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2486 	unsigned int slot2 = hash2 & udp_table.mask;
2487 	struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2488 	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2489 	const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2490 	struct sock *sk;
2491 
2492 	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2493 		if (INET_MATCH(sk, net, acookie, rmt_addr,
2494 			       loc_addr, ports, dif, sdif))
2495 			return sk;
2496 		/* Only check first socket in chain */
2497 		break;
2498 	}
2499 	return NULL;
2500 }
2501 
2502 int udp_v4_early_demux(struct sk_buff *skb)
2503 {
2504 	struct net *net = dev_net(skb->dev);
2505 	struct in_device *in_dev = NULL;
2506 	const struct iphdr *iph;
2507 	const struct udphdr *uh;
2508 	struct sock *sk = NULL;
2509 	struct dst_entry *dst;
2510 	int dif = skb->dev->ifindex;
2511 	int sdif = inet_sdif(skb);
2512 	int ours;
2513 
2514 	/* validate the packet */
2515 	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2516 		return 0;
2517 
2518 	iph = ip_hdr(skb);
2519 	uh = udp_hdr(skb);
2520 
2521 	if (skb->pkt_type == PACKET_MULTICAST) {
2522 		in_dev = __in_dev_get_rcu(skb->dev);
2523 
2524 		if (!in_dev)
2525 			return 0;
2526 
2527 		ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2528 				       iph->protocol);
2529 		if (!ours)
2530 			return 0;
2531 
2532 		sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2533 						   uh->source, iph->saddr,
2534 						   dif, sdif);
2535 	} else if (skb->pkt_type == PACKET_HOST) {
2536 		sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2537 					     uh->source, iph->saddr, dif, sdif);
2538 	}
2539 
2540 	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2541 		return 0;
2542 
2543 	skb->sk = sk;
2544 	skb->destructor = sock_efree;
2545 	dst = READ_ONCE(sk->sk_rx_dst);
2546 
2547 	if (dst)
2548 		dst = dst_check(dst, 0);
2549 	if (dst) {
2550 		u32 itag = 0;
2551 
2552 		/* set noref for now.
2553 		 * any place which wants to hold dst has to call
2554 		 * dst_hold_safe()
2555 		 */
2556 		skb_dst_set_noref(skb, dst);
2557 
2558 		/* for unconnected multicast sockets we need to validate
2559 		 * the source on each packet
2560 		 */
2561 		if (!inet_sk(sk)->inet_daddr && in_dev)
2562 			return ip_mc_validate_source(skb, iph->daddr,
2563 						     iph->saddr,
2564 						     iph->tos & IPTOS_RT_MASK,
2565 						     skb->dev, in_dev, &itag);
2566 	}
2567 	return 0;
2568 }
2569 
2570 int udp_rcv(struct sk_buff *skb)
2571 {
2572 	return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2573 }
2574 
2575 void udp_destroy_sock(struct sock *sk)
2576 {
2577 	struct udp_sock *up = udp_sk(sk);
2578 	bool slow = lock_sock_fast(sk);
2579 	udp_flush_pending_frames(sk);
2580 	unlock_sock_fast(sk, slow);
2581 	if (static_branch_unlikely(&udp_encap_needed_key)) {
2582 		if (up->encap_type) {
2583 			void (*encap_destroy)(struct sock *sk);
2584 			encap_destroy = READ_ONCE(up->encap_destroy);
2585 			if (encap_destroy)
2586 				encap_destroy(sk);
2587 		}
2588 		if (up->encap_enabled)
2589 			static_branch_dec(&udp_encap_needed_key);
2590 	}
2591 }
2592 
2593 /*
2594  *	Socket option code for UDP
2595  */
2596 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2597 		       sockptr_t optval, unsigned int optlen,
2598 		       int (*push_pending_frames)(struct sock *))
2599 {
2600 	struct udp_sock *up = udp_sk(sk);
2601 	int val, valbool;
2602 	int err = 0;
2603 	int is_udplite = IS_UDPLITE(sk);
2604 
2605 	if (optlen < sizeof(int))
2606 		return -EINVAL;
2607 
2608 	if (copy_from_sockptr(&val, optval, sizeof(val)))
2609 		return -EFAULT;
2610 
2611 	valbool = val ? 1 : 0;
2612 
2613 	switch (optname) {
2614 	case UDP_CORK:
2615 		if (val != 0) {
2616 			up->corkflag = 1;
2617 		} else {
2618 			up->corkflag = 0;
2619 			lock_sock(sk);
2620 			push_pending_frames(sk);
2621 			release_sock(sk);
2622 		}
2623 		break;
2624 
2625 	case UDP_ENCAP:
2626 		switch (val) {
2627 		case 0:
2628 #ifdef CONFIG_XFRM
2629 		case UDP_ENCAP_ESPINUDP:
2630 		case UDP_ENCAP_ESPINUDP_NON_IKE:
2631 #if IS_ENABLED(CONFIG_IPV6)
2632 			if (sk->sk_family == AF_INET6)
2633 				up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2634 			else
2635 #endif
2636 				up->encap_rcv = xfrm4_udp_encap_rcv;
2637 #endif
2638 			fallthrough;
2639 		case UDP_ENCAP_L2TPINUDP:
2640 			up->encap_type = val;
2641 			lock_sock(sk);
2642 			udp_tunnel_encap_enable(sk->sk_socket);
2643 			release_sock(sk);
2644 			break;
2645 		default:
2646 			err = -ENOPROTOOPT;
2647 			break;
2648 		}
2649 		break;
2650 
2651 	case UDP_NO_CHECK6_TX:
2652 		up->no_check6_tx = valbool;
2653 		break;
2654 
2655 	case UDP_NO_CHECK6_RX:
2656 		up->no_check6_rx = valbool;
2657 		break;
2658 
2659 	case UDP_SEGMENT:
2660 		if (val < 0 || val > USHRT_MAX)
2661 			return -EINVAL;
2662 		up->gso_size = val;
2663 		break;
2664 
2665 	case UDP_GRO:
2666 		lock_sock(sk);
2667 		if (valbool)
2668 			udp_tunnel_encap_enable(sk->sk_socket);
2669 		up->gro_enabled = valbool;
2670 		release_sock(sk);
2671 		break;
2672 
2673 	/*
2674 	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
2675 	 */
2676 	/* The sender sets actual checksum coverage length via this option.
2677 	 * The case coverage > packet length is handled by send module. */
2678 	case UDPLITE_SEND_CSCOV:
2679 		if (!is_udplite)         /* Disable the option on UDP sockets */
2680 			return -ENOPROTOOPT;
2681 		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2682 			val = 8;
2683 		else if (val > USHRT_MAX)
2684 			val = USHRT_MAX;
2685 		up->pcslen = val;
2686 		up->pcflag |= UDPLITE_SEND_CC;
2687 		break;
2688 
2689 	/* The receiver specifies a minimum checksum coverage value. To make
2690 	 * sense, this should be set to at least 8 (as done below). If zero is
2691 	 * used, this again means full checksum coverage.                     */
2692 	case UDPLITE_RECV_CSCOV:
2693 		if (!is_udplite)         /* Disable the option on UDP sockets */
2694 			return -ENOPROTOOPT;
2695 		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2696 			val = 8;
2697 		else if (val > USHRT_MAX)
2698 			val = USHRT_MAX;
2699 		up->pcrlen = val;
2700 		up->pcflag |= UDPLITE_RECV_CC;
2701 		break;
2702 
2703 	default:
2704 		err = -ENOPROTOOPT;
2705 		break;
2706 	}
2707 
2708 	return err;
2709 }
2710 EXPORT_SYMBOL(udp_lib_setsockopt);
2711 
2712 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2713 		   unsigned int optlen)
2714 {
2715 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2716 		return udp_lib_setsockopt(sk, level, optname,
2717 					  optval, optlen,
2718 					  udp_push_pending_frames);
2719 	return ip_setsockopt(sk, level, optname, optval, optlen);
2720 }
2721 
2722 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2723 		       char __user *optval, int __user *optlen)
2724 {
2725 	struct udp_sock *up = udp_sk(sk);
2726 	int val, len;
2727 
2728 	if (get_user(len, optlen))
2729 		return -EFAULT;
2730 
2731 	len = min_t(unsigned int, len, sizeof(int));
2732 
2733 	if (len < 0)
2734 		return -EINVAL;
2735 
2736 	switch (optname) {
2737 	case UDP_CORK:
2738 		val = up->corkflag;
2739 		break;
2740 
2741 	case UDP_ENCAP:
2742 		val = up->encap_type;
2743 		break;
2744 
2745 	case UDP_NO_CHECK6_TX:
2746 		val = up->no_check6_tx;
2747 		break;
2748 
2749 	case UDP_NO_CHECK6_RX:
2750 		val = up->no_check6_rx;
2751 		break;
2752 
2753 	case UDP_SEGMENT:
2754 		val = up->gso_size;
2755 		break;
2756 
2757 	/* The following two cannot be changed on UDP sockets, the return is
2758 	 * always 0 (which corresponds to the full checksum coverage of UDP). */
2759 	case UDPLITE_SEND_CSCOV:
2760 		val = up->pcslen;
2761 		break;
2762 
2763 	case UDPLITE_RECV_CSCOV:
2764 		val = up->pcrlen;
2765 		break;
2766 
2767 	default:
2768 		return -ENOPROTOOPT;
2769 	}
2770 
2771 	if (put_user(len, optlen))
2772 		return -EFAULT;
2773 	if (copy_to_user(optval, &val, len))
2774 		return -EFAULT;
2775 	return 0;
2776 }
2777 EXPORT_SYMBOL(udp_lib_getsockopt);
2778 
2779 int udp_getsockopt(struct sock *sk, int level, int optname,
2780 		   char __user *optval, int __user *optlen)
2781 {
2782 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2783 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2784 	return ip_getsockopt(sk, level, optname, optval, optlen);
2785 }
2786 
2787 /**
2788  * 	udp_poll - wait for a UDP event.
2789  *	@file: - file struct
2790  *	@sock: - socket
2791  *	@wait: - poll table
2792  *
2793  *	This is same as datagram poll, except for the special case of
2794  *	blocking sockets. If application is using a blocking fd
2795  *	and a packet with checksum error is in the queue;
2796  *	then it could get return from select indicating data available
2797  *	but then block when reading it. Add special case code
2798  *	to work around these arguably broken applications.
2799  */
2800 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2801 {
2802 	__poll_t mask = datagram_poll(file, sock, wait);
2803 	struct sock *sk = sock->sk;
2804 
2805 	if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2806 		mask |= EPOLLIN | EPOLLRDNORM;
2807 
2808 	/* Check for false positives due to checksum errors */
2809 	if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2810 	    !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2811 		mask &= ~(EPOLLIN | EPOLLRDNORM);
2812 
2813 	return mask;
2814 
2815 }
2816 EXPORT_SYMBOL(udp_poll);
2817 
2818 int udp_abort(struct sock *sk, int err)
2819 {
2820 	lock_sock(sk);
2821 
2822 	sk->sk_err = err;
2823 	sk->sk_error_report(sk);
2824 	__udp_disconnect(sk, 0);
2825 
2826 	release_sock(sk);
2827 
2828 	return 0;
2829 }
2830 EXPORT_SYMBOL_GPL(udp_abort);
2831 
2832 struct proto udp_prot = {
2833 	.name			= "UDP",
2834 	.owner			= THIS_MODULE,
2835 	.close			= udp_lib_close,
2836 	.pre_connect		= udp_pre_connect,
2837 	.connect		= ip4_datagram_connect,
2838 	.disconnect		= udp_disconnect,
2839 	.ioctl			= udp_ioctl,
2840 	.init			= udp_init_sock,
2841 	.destroy		= udp_destroy_sock,
2842 	.setsockopt		= udp_setsockopt,
2843 	.getsockopt		= udp_getsockopt,
2844 	.sendmsg		= udp_sendmsg,
2845 	.recvmsg		= udp_recvmsg,
2846 	.sendpage		= udp_sendpage,
2847 	.release_cb		= ip4_datagram_release_cb,
2848 	.hash			= udp_lib_hash,
2849 	.unhash			= udp_lib_unhash,
2850 	.rehash			= udp_v4_rehash,
2851 	.get_port		= udp_v4_get_port,
2852 	.memory_allocated	= &udp_memory_allocated,
2853 	.sysctl_mem		= sysctl_udp_mem,
2854 	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2855 	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2856 	.obj_size		= sizeof(struct udp_sock),
2857 	.h.udp_table		= &udp_table,
2858 	.diag_destroy		= udp_abort,
2859 };
2860 EXPORT_SYMBOL(udp_prot);
2861 
2862 /* ------------------------------------------------------------------------ */
2863 #ifdef CONFIG_PROC_FS
2864 
2865 static struct sock *udp_get_first(struct seq_file *seq, int start)
2866 {
2867 	struct sock *sk;
2868 	struct udp_seq_afinfo *afinfo;
2869 	struct udp_iter_state *state = seq->private;
2870 	struct net *net = seq_file_net(seq);
2871 
2872 	if (state->bpf_seq_afinfo)
2873 		afinfo = state->bpf_seq_afinfo;
2874 	else
2875 		afinfo = PDE_DATA(file_inode(seq->file));
2876 
2877 	for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2878 	     ++state->bucket) {
2879 		struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2880 
2881 		if (hlist_empty(&hslot->head))
2882 			continue;
2883 
2884 		spin_lock_bh(&hslot->lock);
2885 		sk_for_each(sk, &hslot->head) {
2886 			if (!net_eq(sock_net(sk), net))
2887 				continue;
2888 			if (afinfo->family == AF_UNSPEC ||
2889 			    sk->sk_family == afinfo->family)
2890 				goto found;
2891 		}
2892 		spin_unlock_bh(&hslot->lock);
2893 	}
2894 	sk = NULL;
2895 found:
2896 	return sk;
2897 }
2898 
2899 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2900 {
2901 	struct udp_seq_afinfo *afinfo;
2902 	struct udp_iter_state *state = seq->private;
2903 	struct net *net = seq_file_net(seq);
2904 
2905 	if (state->bpf_seq_afinfo)
2906 		afinfo = state->bpf_seq_afinfo;
2907 	else
2908 		afinfo = PDE_DATA(file_inode(seq->file));
2909 
2910 	do {
2911 		sk = sk_next(sk);
2912 	} while (sk && (!net_eq(sock_net(sk), net) ||
2913 			(afinfo->family != AF_UNSPEC &&
2914 			 sk->sk_family != afinfo->family)));
2915 
2916 	if (!sk) {
2917 		if (state->bucket <= afinfo->udp_table->mask)
2918 			spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2919 		return udp_get_first(seq, state->bucket + 1);
2920 	}
2921 	return sk;
2922 }
2923 
2924 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2925 {
2926 	struct sock *sk = udp_get_first(seq, 0);
2927 
2928 	if (sk)
2929 		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2930 			--pos;
2931 	return pos ? NULL : sk;
2932 }
2933 
2934 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2935 {
2936 	struct udp_iter_state *state = seq->private;
2937 	state->bucket = MAX_UDP_PORTS;
2938 
2939 	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2940 }
2941 EXPORT_SYMBOL(udp_seq_start);
2942 
2943 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2944 {
2945 	struct sock *sk;
2946 
2947 	if (v == SEQ_START_TOKEN)
2948 		sk = udp_get_idx(seq, 0);
2949 	else
2950 		sk = udp_get_next(seq, v);
2951 
2952 	++*pos;
2953 	return sk;
2954 }
2955 EXPORT_SYMBOL(udp_seq_next);
2956 
2957 void udp_seq_stop(struct seq_file *seq, void *v)
2958 {
2959 	struct udp_seq_afinfo *afinfo;
2960 	struct udp_iter_state *state = seq->private;
2961 
2962 	if (state->bpf_seq_afinfo)
2963 		afinfo = state->bpf_seq_afinfo;
2964 	else
2965 		afinfo = PDE_DATA(file_inode(seq->file));
2966 
2967 	if (state->bucket <= afinfo->udp_table->mask)
2968 		spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2969 }
2970 EXPORT_SYMBOL(udp_seq_stop);
2971 
2972 /* ------------------------------------------------------------------------ */
2973 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2974 		int bucket)
2975 {
2976 	struct inet_sock *inet = inet_sk(sp);
2977 	__be32 dest = inet->inet_daddr;
2978 	__be32 src  = inet->inet_rcv_saddr;
2979 	__u16 destp	  = ntohs(inet->inet_dport);
2980 	__u16 srcp	  = ntohs(inet->inet_sport);
2981 
2982 	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2983 		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
2984 		bucket, src, srcp, dest, destp, sp->sk_state,
2985 		sk_wmem_alloc_get(sp),
2986 		udp_rqueue_get(sp),
2987 		0, 0L, 0,
2988 		from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2989 		0, sock_i_ino(sp),
2990 		refcount_read(&sp->sk_refcnt), sp,
2991 		atomic_read(&sp->sk_drops));
2992 }
2993 
2994 int udp4_seq_show(struct seq_file *seq, void *v)
2995 {
2996 	seq_setwidth(seq, 127);
2997 	if (v == SEQ_START_TOKEN)
2998 		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2999 			   "rx_queue tr tm->when retrnsmt   uid  timeout "
3000 			   "inode ref pointer drops");
3001 	else {
3002 		struct udp_iter_state *state = seq->private;
3003 
3004 		udp4_format_sock(v, seq, state->bucket);
3005 	}
3006 	seq_pad(seq, '\n');
3007 	return 0;
3008 }
3009 
3010 #ifdef CONFIG_BPF_SYSCALL
3011 struct bpf_iter__udp {
3012 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
3013 	__bpf_md_ptr(struct udp_sock *, udp_sk);
3014 	uid_t uid __aligned(8);
3015 	int bucket __aligned(8);
3016 };
3017 
3018 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3019 			     struct udp_sock *udp_sk, uid_t uid, int bucket)
3020 {
3021 	struct bpf_iter__udp ctx;
3022 
3023 	meta->seq_num--;  /* skip SEQ_START_TOKEN */
3024 	ctx.meta = meta;
3025 	ctx.udp_sk = udp_sk;
3026 	ctx.uid = uid;
3027 	ctx.bucket = bucket;
3028 	return bpf_iter_run_prog(prog, &ctx);
3029 }
3030 
3031 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3032 {
3033 	struct udp_iter_state *state = seq->private;
3034 	struct bpf_iter_meta meta;
3035 	struct bpf_prog *prog;
3036 	struct sock *sk = v;
3037 	uid_t uid;
3038 
3039 	if (v == SEQ_START_TOKEN)
3040 		return 0;
3041 
3042 	uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3043 	meta.seq = seq;
3044 	prog = bpf_iter_get_info(&meta, false);
3045 	return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3046 }
3047 
3048 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3049 {
3050 	struct bpf_iter_meta meta;
3051 	struct bpf_prog *prog;
3052 
3053 	if (!v) {
3054 		meta.seq = seq;
3055 		prog = bpf_iter_get_info(&meta, true);
3056 		if (prog)
3057 			(void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3058 	}
3059 
3060 	udp_seq_stop(seq, v);
3061 }
3062 
3063 static const struct seq_operations bpf_iter_udp_seq_ops = {
3064 	.start		= udp_seq_start,
3065 	.next		= udp_seq_next,
3066 	.stop		= bpf_iter_udp_seq_stop,
3067 	.show		= bpf_iter_udp_seq_show,
3068 };
3069 #endif
3070 
3071 const struct seq_operations udp_seq_ops = {
3072 	.start		= udp_seq_start,
3073 	.next		= udp_seq_next,
3074 	.stop		= udp_seq_stop,
3075 	.show		= udp4_seq_show,
3076 };
3077 EXPORT_SYMBOL(udp_seq_ops);
3078 
3079 static struct udp_seq_afinfo udp4_seq_afinfo = {
3080 	.family		= AF_INET,
3081 	.udp_table	= &udp_table,
3082 };
3083 
3084 static int __net_init udp4_proc_init_net(struct net *net)
3085 {
3086 	if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3087 			sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3088 		return -ENOMEM;
3089 	return 0;
3090 }
3091 
3092 static void __net_exit udp4_proc_exit_net(struct net *net)
3093 {
3094 	remove_proc_entry("udp", net->proc_net);
3095 }
3096 
3097 static struct pernet_operations udp4_net_ops = {
3098 	.init = udp4_proc_init_net,
3099 	.exit = udp4_proc_exit_net,
3100 };
3101 
3102 int __init udp4_proc_init(void)
3103 {
3104 	return register_pernet_subsys(&udp4_net_ops);
3105 }
3106 
3107 void udp4_proc_exit(void)
3108 {
3109 	unregister_pernet_subsys(&udp4_net_ops);
3110 }
3111 #endif /* CONFIG_PROC_FS */
3112 
3113 static __initdata unsigned long uhash_entries;
3114 static int __init set_uhash_entries(char *str)
3115 {
3116 	ssize_t ret;
3117 
3118 	if (!str)
3119 		return 0;
3120 
3121 	ret = kstrtoul(str, 0, &uhash_entries);
3122 	if (ret)
3123 		return 0;
3124 
3125 	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3126 		uhash_entries = UDP_HTABLE_SIZE_MIN;
3127 	return 1;
3128 }
3129 __setup("uhash_entries=", set_uhash_entries);
3130 
3131 void __init udp_table_init(struct udp_table *table, const char *name)
3132 {
3133 	unsigned int i;
3134 
3135 	table->hash = alloc_large_system_hash(name,
3136 					      2 * sizeof(struct udp_hslot),
3137 					      uhash_entries,
3138 					      21, /* one slot per 2 MB */
3139 					      0,
3140 					      &table->log,
3141 					      &table->mask,
3142 					      UDP_HTABLE_SIZE_MIN,
3143 					      64 * 1024);
3144 
3145 	table->hash2 = table->hash + (table->mask + 1);
3146 	for (i = 0; i <= table->mask; i++) {
3147 		INIT_HLIST_HEAD(&table->hash[i].head);
3148 		table->hash[i].count = 0;
3149 		spin_lock_init(&table->hash[i].lock);
3150 	}
3151 	for (i = 0; i <= table->mask; i++) {
3152 		INIT_HLIST_HEAD(&table->hash2[i].head);
3153 		table->hash2[i].count = 0;
3154 		spin_lock_init(&table->hash2[i].lock);
3155 	}
3156 }
3157 
3158 u32 udp_flow_hashrnd(void)
3159 {
3160 	static u32 hashrnd __read_mostly;
3161 
3162 	net_get_random_once(&hashrnd, sizeof(hashrnd));
3163 
3164 	return hashrnd;
3165 }
3166 EXPORT_SYMBOL(udp_flow_hashrnd);
3167 
3168 static void __udp_sysctl_init(struct net *net)
3169 {
3170 	net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3171 	net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3172 
3173 #ifdef CONFIG_NET_L3_MASTER_DEV
3174 	net->ipv4.sysctl_udp_l3mdev_accept = 0;
3175 #endif
3176 }
3177 
3178 static int __net_init udp_sysctl_init(struct net *net)
3179 {
3180 	__udp_sysctl_init(net);
3181 	return 0;
3182 }
3183 
3184 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3185 	.init	= udp_sysctl_init,
3186 };
3187 
3188 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3189 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3190 		     struct udp_sock *udp_sk, uid_t uid, int bucket)
3191 
3192 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3193 {
3194 	struct udp_iter_state *st = priv_data;
3195 	struct udp_seq_afinfo *afinfo;
3196 	int ret;
3197 
3198 	afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3199 	if (!afinfo)
3200 		return -ENOMEM;
3201 
3202 	afinfo->family = AF_UNSPEC;
3203 	afinfo->udp_table = &udp_table;
3204 	st->bpf_seq_afinfo = afinfo;
3205 	ret = bpf_iter_init_seq_net(priv_data, aux);
3206 	if (ret)
3207 		kfree(afinfo);
3208 	return ret;
3209 }
3210 
3211 static void bpf_iter_fini_udp(void *priv_data)
3212 {
3213 	struct udp_iter_state *st = priv_data;
3214 
3215 	kfree(st->bpf_seq_afinfo);
3216 	bpf_iter_fini_seq_net(priv_data);
3217 }
3218 
3219 static const struct bpf_iter_seq_info udp_seq_info = {
3220 	.seq_ops		= &bpf_iter_udp_seq_ops,
3221 	.init_seq_private	= bpf_iter_init_udp,
3222 	.fini_seq_private	= bpf_iter_fini_udp,
3223 	.seq_priv_size		= sizeof(struct udp_iter_state),
3224 };
3225 
3226 static struct bpf_iter_reg udp_reg_info = {
3227 	.target			= "udp",
3228 	.ctx_arg_info_size	= 1,
3229 	.ctx_arg_info		= {
3230 		{ offsetof(struct bpf_iter__udp, udp_sk),
3231 		  PTR_TO_BTF_ID_OR_NULL },
3232 	},
3233 	.seq_info		= &udp_seq_info,
3234 };
3235 
3236 static void __init bpf_iter_register(void)
3237 {
3238 	udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3239 	if (bpf_iter_reg_target(&udp_reg_info))
3240 		pr_warn("Warning: could not register bpf iterator udp\n");
3241 }
3242 #endif
3243 
3244 void __init udp_init(void)
3245 {
3246 	unsigned long limit;
3247 	unsigned int i;
3248 
3249 	udp_table_init(&udp_table, "UDP");
3250 	limit = nr_free_buffer_pages() / 8;
3251 	limit = max(limit, 128UL);
3252 	sysctl_udp_mem[0] = limit / 4 * 3;
3253 	sysctl_udp_mem[1] = limit;
3254 	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3255 
3256 	__udp_sysctl_init(&init_net);
3257 
3258 	/* 16 spinlocks per cpu */
3259 	udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3260 	udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3261 				GFP_KERNEL);
3262 	if (!udp_busylocks)
3263 		panic("UDP: failed to alloc udp_busylocks\n");
3264 	for (i = 0; i < (1U << udp_busylocks_log); i++)
3265 		spin_lock_init(udp_busylocks + i);
3266 
3267 	if (register_pernet_subsys(&udp_sysctl_ops))
3268 		panic("UDP: failed to init sysctl parameters.\n");
3269 
3270 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3271 	bpf_iter_register();
3272 #endif
3273 }
3274