xref: /openbmc/linux/net/ipv4/udp.c (revision 2daf7fab)
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/bpf-cgroup.h>
78 #include <linux/uaccess.h>
79 #include <asm/ioctls.h>
80 #include <linux/memblock.h>
81 #include <linux/highmem.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 ____cacheline_aligned_in_smp;
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 	if (sk->sk_bound_dev_if)
394 		score += 4;
395 
396 	if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
397 		score++;
398 	return score;
399 }
400 
401 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
402 		       const __u16 lport, const __be32 faddr,
403 		       const __be16 fport)
404 {
405 	static u32 udp_ehash_secret __read_mostly;
406 
407 	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
408 
409 	return __inet_ehashfn(laddr, lport, faddr, fport,
410 			      udp_ehash_secret + net_hash_mix(net));
411 }
412 
413 static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
414 				     struct sk_buff *skb,
415 				     __be32 saddr, __be16 sport,
416 				     __be32 daddr, unsigned short hnum)
417 {
418 	struct sock *reuse_sk = NULL;
419 	u32 hash;
420 
421 	if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
422 		hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
423 		reuse_sk = reuseport_select_sock(sk, hash, skb,
424 						 sizeof(struct udphdr));
425 	}
426 	return reuse_sk;
427 }
428 
429 /* called with rcu_read_lock() */
430 static struct sock *udp4_lib_lookup2(struct net *net,
431 				     __be32 saddr, __be16 sport,
432 				     __be32 daddr, unsigned int hnum,
433 				     int dif, int sdif,
434 				     struct udp_hslot *hslot2,
435 				     struct sk_buff *skb)
436 {
437 	struct sock *sk, *result;
438 	int score, badness;
439 
440 	result = NULL;
441 	badness = 0;
442 	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
443 		score = compute_score(sk, net, saddr, sport,
444 				      daddr, hnum, dif, sdif);
445 		if (score > badness) {
446 			result = lookup_reuseport(net, sk, skb,
447 						  saddr, sport, daddr, hnum);
448 			/* Fall back to scoring if group has connections */
449 			if (result && !reuseport_has_conns(sk, false))
450 				return result;
451 
452 			result = result ? : sk;
453 			badness = score;
454 		}
455 	}
456 	return result;
457 }
458 
459 static struct sock *udp4_lookup_run_bpf(struct net *net,
460 					struct udp_table *udptable,
461 					struct sk_buff *skb,
462 					__be32 saddr, __be16 sport,
463 					__be32 daddr, u16 hnum, const int dif)
464 {
465 	struct sock *sk, *reuse_sk;
466 	bool no_reuseport;
467 
468 	if (udptable != &udp_table)
469 		return NULL; /* only UDP is supported */
470 
471 	no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport,
472 					    daddr, hnum, dif, &sk);
473 	if (no_reuseport || IS_ERR_OR_NULL(sk))
474 		return sk;
475 
476 	reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
477 	if (reuse_sk)
478 		sk = reuse_sk;
479 	return sk;
480 }
481 
482 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
483  * harder than this. -DaveM
484  */
485 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
486 		__be16 sport, __be32 daddr, __be16 dport, int dif,
487 		int sdif, struct udp_table *udptable, struct sk_buff *skb)
488 {
489 	unsigned short hnum = ntohs(dport);
490 	unsigned int hash2, slot2;
491 	struct udp_hslot *hslot2;
492 	struct sock *result, *sk;
493 
494 	hash2 = ipv4_portaddr_hash(net, daddr, hnum);
495 	slot2 = hash2 & udptable->mask;
496 	hslot2 = &udptable->hash2[slot2];
497 
498 	/* Lookup connected or non-wildcard socket */
499 	result = udp4_lib_lookup2(net, saddr, sport,
500 				  daddr, hnum, dif, sdif,
501 				  hslot2, skb);
502 	if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
503 		goto done;
504 
505 	/* Lookup redirect from BPF */
506 	if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
507 		sk = udp4_lookup_run_bpf(net, udptable, skb,
508 					 saddr, sport, daddr, hnum, dif);
509 		if (sk) {
510 			result = sk;
511 			goto done;
512 		}
513 	}
514 
515 	/* Got non-wildcard socket or error on first lookup */
516 	if (result)
517 		goto done;
518 
519 	/* Lookup wildcard sockets */
520 	hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
521 	slot2 = hash2 & udptable->mask;
522 	hslot2 = &udptable->hash2[slot2];
523 
524 	result = udp4_lib_lookup2(net, saddr, sport,
525 				  htonl(INADDR_ANY), hnum, dif, sdif,
526 				  hslot2, skb);
527 done:
528 	if (IS_ERR(result))
529 		return NULL;
530 	return result;
531 }
532 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
533 
534 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
535 						 __be16 sport, __be16 dport,
536 						 struct udp_table *udptable)
537 {
538 	const struct iphdr *iph = ip_hdr(skb);
539 
540 	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
541 				 iph->daddr, dport, inet_iif(skb),
542 				 inet_sdif(skb), udptable, skb);
543 }
544 
545 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
546 				 __be16 sport, __be16 dport)
547 {
548 	const struct iphdr *iph = ip_hdr(skb);
549 
550 	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
551 				 iph->daddr, dport, inet_iif(skb),
552 				 inet_sdif(skb), &udp_table, NULL);
553 }
554 
555 /* Must be called under rcu_read_lock().
556  * Does increment socket refcount.
557  */
558 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
559 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
560 			     __be32 daddr, __be16 dport, int dif)
561 {
562 	struct sock *sk;
563 
564 	sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
565 			       dif, 0, &udp_table, NULL);
566 	if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
567 		sk = NULL;
568 	return sk;
569 }
570 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
571 #endif
572 
573 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
574 				       __be16 loc_port, __be32 loc_addr,
575 				       __be16 rmt_port, __be32 rmt_addr,
576 				       int dif, int sdif, unsigned short hnum)
577 {
578 	struct inet_sock *inet = inet_sk(sk);
579 
580 	if (!net_eq(sock_net(sk), net) ||
581 	    udp_sk(sk)->udp_port_hash != hnum ||
582 	    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
583 	    (inet->inet_dport != rmt_port && inet->inet_dport) ||
584 	    (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
585 	    ipv6_only_sock(sk) ||
586 	    !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
587 		return false;
588 	if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
589 		return false;
590 	return true;
591 }
592 
593 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
594 void udp_encap_enable(void)
595 {
596 	static_branch_inc(&udp_encap_needed_key);
597 }
598 EXPORT_SYMBOL(udp_encap_enable);
599 
600 void udp_encap_disable(void)
601 {
602 	static_branch_dec(&udp_encap_needed_key);
603 }
604 EXPORT_SYMBOL(udp_encap_disable);
605 
606 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
607  * through error handlers in encapsulations looking for a match.
608  */
609 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
610 {
611 	int i;
612 
613 	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
614 		int (*handler)(struct sk_buff *skb, u32 info);
615 		const struct ip_tunnel_encap_ops *encap;
616 
617 		encap = rcu_dereference(iptun_encaps[i]);
618 		if (!encap)
619 			continue;
620 		handler = encap->err_handler;
621 		if (handler && !handler(skb, info))
622 			return 0;
623 	}
624 
625 	return -ENOENT;
626 }
627 
628 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
629  * reversing source and destination port: this will match tunnels that force the
630  * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
631  * lwtunnels might actually break this assumption by being configured with
632  * different destination ports on endpoints, in this case we won't be able to
633  * trace ICMP messages back to them.
634  *
635  * If this doesn't match any socket, probe tunnels with arbitrary destination
636  * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
637  * we've sent packets to won't necessarily match the local destination port.
638  *
639  * Then ask the tunnel implementation to match the error against a valid
640  * association.
641  *
642  * Return an error if we can't find a match, the socket if we need further
643  * processing, zero otherwise.
644  */
645 static struct sock *__udp4_lib_err_encap(struct net *net,
646 					 const struct iphdr *iph,
647 					 struct udphdr *uh,
648 					 struct udp_table *udptable,
649 					 struct sock *sk,
650 					 struct sk_buff *skb, u32 info)
651 {
652 	int (*lookup)(struct sock *sk, struct sk_buff *skb);
653 	int network_offset, transport_offset;
654 	struct udp_sock *up;
655 
656 	network_offset = skb_network_offset(skb);
657 	transport_offset = skb_transport_offset(skb);
658 
659 	/* Network header needs to point to the outer IPv4 header inside ICMP */
660 	skb_reset_network_header(skb);
661 
662 	/* Transport header needs to point to the UDP header */
663 	skb_set_transport_header(skb, iph->ihl << 2);
664 
665 	if (sk) {
666 		up = udp_sk(sk);
667 
668 		lookup = READ_ONCE(up->encap_err_lookup);
669 		if (lookup && lookup(sk, skb))
670 			sk = NULL;
671 
672 		goto out;
673 	}
674 
675 	sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
676 			       iph->saddr, uh->dest, skb->dev->ifindex, 0,
677 			       udptable, NULL);
678 	if (sk) {
679 		up = udp_sk(sk);
680 
681 		lookup = READ_ONCE(up->encap_err_lookup);
682 		if (!lookup || lookup(sk, skb))
683 			sk = NULL;
684 	}
685 
686 out:
687 	if (!sk)
688 		sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
689 
690 	skb_set_transport_header(skb, transport_offset);
691 	skb_set_network_header(skb, network_offset);
692 
693 	return sk;
694 }
695 
696 /*
697  * This routine is called by the ICMP module when it gets some
698  * sort of error condition.  If err < 0 then the socket should
699  * be closed and the error returned to the user.  If err > 0
700  * it's just the icmp type << 8 | icmp code.
701  * Header points to the ip header of the error packet. We move
702  * on past this. Then (as it used to claim before adjustment)
703  * header points to the first 8 bytes of the udp header.  We need
704  * to find the appropriate port.
705  */
706 
707 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
708 {
709 	struct inet_sock *inet;
710 	const struct iphdr *iph = (const struct iphdr *)skb->data;
711 	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
712 	const int type = icmp_hdr(skb)->type;
713 	const int code = icmp_hdr(skb)->code;
714 	bool tunnel = false;
715 	struct sock *sk;
716 	int harderr;
717 	int err;
718 	struct net *net = dev_net(skb->dev);
719 
720 	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
721 			       iph->saddr, uh->source, skb->dev->ifindex,
722 			       inet_sdif(skb), udptable, NULL);
723 
724 	if (!sk || udp_sk(sk)->encap_type) {
725 		/* No socket for error: try tunnels before discarding */
726 		if (static_branch_unlikely(&udp_encap_needed_key)) {
727 			sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
728 						  info);
729 			if (!sk)
730 				return 0;
731 		} else
732 			sk = ERR_PTR(-ENOENT);
733 
734 		if (IS_ERR(sk)) {
735 			__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
736 			return PTR_ERR(sk);
737 		}
738 
739 		tunnel = true;
740 	}
741 
742 	err = 0;
743 	harderr = 0;
744 	inet = inet_sk(sk);
745 
746 	switch (type) {
747 	default:
748 	case ICMP_TIME_EXCEEDED:
749 		err = EHOSTUNREACH;
750 		break;
751 	case ICMP_SOURCE_QUENCH:
752 		goto out;
753 	case ICMP_PARAMETERPROB:
754 		err = EPROTO;
755 		harderr = 1;
756 		break;
757 	case ICMP_DEST_UNREACH:
758 		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
759 			ipv4_sk_update_pmtu(skb, sk, info);
760 			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
761 				err = EMSGSIZE;
762 				harderr = 1;
763 				break;
764 			}
765 			goto out;
766 		}
767 		err = EHOSTUNREACH;
768 		if (code <= NR_ICMP_UNREACH) {
769 			harderr = icmp_err_convert[code].fatal;
770 			err = icmp_err_convert[code].errno;
771 		}
772 		break;
773 	case ICMP_REDIRECT:
774 		ipv4_sk_redirect(skb, sk);
775 		goto out;
776 	}
777 
778 	/*
779 	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
780 	 *	4.1.3.3.
781 	 */
782 	if (tunnel) {
783 		/* ...not for tunnels though: we don't have a sending socket */
784 		goto out;
785 	}
786 	if (!inet->recverr) {
787 		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
788 			goto out;
789 	} else
790 		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
791 
792 	sk->sk_err = err;
793 	sk_error_report(sk);
794 out:
795 	return 0;
796 }
797 
798 int udp_err(struct sk_buff *skb, u32 info)
799 {
800 	return __udp4_lib_err(skb, info, &udp_table);
801 }
802 
803 /*
804  * Throw away all pending data and cancel the corking. Socket is locked.
805  */
806 void udp_flush_pending_frames(struct sock *sk)
807 {
808 	struct udp_sock *up = udp_sk(sk);
809 
810 	if (up->pending) {
811 		up->len = 0;
812 		up->pending = 0;
813 		ip_flush_pending_frames(sk);
814 	}
815 }
816 EXPORT_SYMBOL(udp_flush_pending_frames);
817 
818 /**
819  * 	udp4_hwcsum  -  handle outgoing HW checksumming
820  * 	@skb: 	sk_buff containing the filled-in UDP header
821  * 	        (checksum field must be zeroed out)
822  *	@src:	source IP address
823  *	@dst:	destination IP address
824  */
825 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
826 {
827 	struct udphdr *uh = udp_hdr(skb);
828 	int offset = skb_transport_offset(skb);
829 	int len = skb->len - offset;
830 	int hlen = len;
831 	__wsum csum = 0;
832 
833 	if (!skb_has_frag_list(skb)) {
834 		/*
835 		 * Only one fragment on the socket.
836 		 */
837 		skb->csum_start = skb_transport_header(skb) - skb->head;
838 		skb->csum_offset = offsetof(struct udphdr, check);
839 		uh->check = ~csum_tcpudp_magic(src, dst, len,
840 					       IPPROTO_UDP, 0);
841 	} else {
842 		struct sk_buff *frags;
843 
844 		/*
845 		 * HW-checksum won't work as there are two or more
846 		 * fragments on the socket so that all csums of sk_buffs
847 		 * should be together
848 		 */
849 		skb_walk_frags(skb, frags) {
850 			csum = csum_add(csum, frags->csum);
851 			hlen -= frags->len;
852 		}
853 
854 		csum = skb_checksum(skb, offset, hlen, csum);
855 		skb->ip_summed = CHECKSUM_NONE;
856 
857 		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
858 		if (uh->check == 0)
859 			uh->check = CSUM_MANGLED_0;
860 	}
861 }
862 EXPORT_SYMBOL_GPL(udp4_hwcsum);
863 
864 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
865  * for the simple case like when setting the checksum for a UDP tunnel.
866  */
867 void udp_set_csum(bool nocheck, struct sk_buff *skb,
868 		  __be32 saddr, __be32 daddr, int len)
869 {
870 	struct udphdr *uh = udp_hdr(skb);
871 
872 	if (nocheck) {
873 		uh->check = 0;
874 	} else if (skb_is_gso(skb)) {
875 		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
876 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
877 		uh->check = 0;
878 		uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
879 		if (uh->check == 0)
880 			uh->check = CSUM_MANGLED_0;
881 	} else {
882 		skb->ip_summed = CHECKSUM_PARTIAL;
883 		skb->csum_start = skb_transport_header(skb) - skb->head;
884 		skb->csum_offset = offsetof(struct udphdr, check);
885 		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
886 	}
887 }
888 EXPORT_SYMBOL(udp_set_csum);
889 
890 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
891 			struct inet_cork *cork)
892 {
893 	struct sock *sk = skb->sk;
894 	struct inet_sock *inet = inet_sk(sk);
895 	struct udphdr *uh;
896 	int err;
897 	int is_udplite = IS_UDPLITE(sk);
898 	int offset = skb_transport_offset(skb);
899 	int len = skb->len - offset;
900 	int datalen = len - sizeof(*uh);
901 	__wsum csum = 0;
902 
903 	/*
904 	 * Create a UDP header
905 	 */
906 	uh = udp_hdr(skb);
907 	uh->source = inet->inet_sport;
908 	uh->dest = fl4->fl4_dport;
909 	uh->len = htons(len);
910 	uh->check = 0;
911 
912 	if (cork->gso_size) {
913 		const int hlen = skb_network_header_len(skb) +
914 				 sizeof(struct udphdr);
915 
916 		if (hlen + cork->gso_size > cork->fragsize) {
917 			kfree_skb(skb);
918 			return -EINVAL;
919 		}
920 		if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
921 			kfree_skb(skb);
922 			return -EINVAL;
923 		}
924 		if (sk->sk_no_check_tx) {
925 			kfree_skb(skb);
926 			return -EINVAL;
927 		}
928 		if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
929 		    dst_xfrm(skb_dst(skb))) {
930 			kfree_skb(skb);
931 			return -EIO;
932 		}
933 
934 		if (datalen > cork->gso_size) {
935 			skb_shinfo(skb)->gso_size = cork->gso_size;
936 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
937 			skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
938 								 cork->gso_size);
939 		}
940 		goto csum_partial;
941 	}
942 
943 	if (is_udplite)  				 /*     UDP-Lite      */
944 		csum = udplite_csum(skb);
945 
946 	else if (sk->sk_no_check_tx) {			 /* UDP csum off */
947 
948 		skb->ip_summed = CHECKSUM_NONE;
949 		goto send;
950 
951 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
952 csum_partial:
953 
954 		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
955 		goto send;
956 
957 	} else
958 		csum = udp_csum(skb);
959 
960 	/* add protocol-dependent pseudo-header */
961 	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
962 				      sk->sk_protocol, csum);
963 	if (uh->check == 0)
964 		uh->check = CSUM_MANGLED_0;
965 
966 send:
967 	err = ip_send_skb(sock_net(sk), skb);
968 	if (err) {
969 		if (err == -ENOBUFS && !inet->recverr) {
970 			UDP_INC_STATS(sock_net(sk),
971 				      UDP_MIB_SNDBUFERRORS, is_udplite);
972 			err = 0;
973 		}
974 	} else
975 		UDP_INC_STATS(sock_net(sk),
976 			      UDP_MIB_OUTDATAGRAMS, is_udplite);
977 	return err;
978 }
979 
980 /*
981  * Push out all pending data as one UDP datagram. Socket is locked.
982  */
983 int udp_push_pending_frames(struct sock *sk)
984 {
985 	struct udp_sock  *up = udp_sk(sk);
986 	struct inet_sock *inet = inet_sk(sk);
987 	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
988 	struct sk_buff *skb;
989 	int err = 0;
990 
991 	skb = ip_finish_skb(sk, fl4);
992 	if (!skb)
993 		goto out;
994 
995 	err = udp_send_skb(skb, fl4, &inet->cork.base);
996 
997 out:
998 	up->len = 0;
999 	up->pending = 0;
1000 	return err;
1001 }
1002 EXPORT_SYMBOL(udp_push_pending_frames);
1003 
1004 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1005 {
1006 	switch (cmsg->cmsg_type) {
1007 	case UDP_SEGMENT:
1008 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1009 			return -EINVAL;
1010 		*gso_size = *(__u16 *)CMSG_DATA(cmsg);
1011 		return 0;
1012 	default:
1013 		return -EINVAL;
1014 	}
1015 }
1016 
1017 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1018 {
1019 	struct cmsghdr *cmsg;
1020 	bool need_ip = false;
1021 	int err;
1022 
1023 	for_each_cmsghdr(cmsg, msg) {
1024 		if (!CMSG_OK(msg, cmsg))
1025 			return -EINVAL;
1026 
1027 		if (cmsg->cmsg_level != SOL_UDP) {
1028 			need_ip = true;
1029 			continue;
1030 		}
1031 
1032 		err = __udp_cmsg_send(cmsg, gso_size);
1033 		if (err)
1034 			return err;
1035 	}
1036 
1037 	return need_ip;
1038 }
1039 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1040 
1041 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1042 {
1043 	struct inet_sock *inet = inet_sk(sk);
1044 	struct udp_sock *up = udp_sk(sk);
1045 	DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1046 	struct flowi4 fl4_stack;
1047 	struct flowi4 *fl4;
1048 	int ulen = len;
1049 	struct ipcm_cookie ipc;
1050 	struct rtable *rt = NULL;
1051 	int free = 0;
1052 	int connected = 0;
1053 	__be32 daddr, faddr, saddr;
1054 	__be16 dport;
1055 	u8  tos;
1056 	int err, is_udplite = IS_UDPLITE(sk);
1057 	int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1058 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1059 	struct sk_buff *skb;
1060 	struct ip_options_data opt_copy;
1061 
1062 	if (len > 0xFFFF)
1063 		return -EMSGSIZE;
1064 
1065 	/*
1066 	 *	Check the flags.
1067 	 */
1068 
1069 	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1070 		return -EOPNOTSUPP;
1071 
1072 	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1073 
1074 	fl4 = &inet->cork.fl.u.ip4;
1075 	if (up->pending) {
1076 		/*
1077 		 * There are pending frames.
1078 		 * The socket lock must be held while it's corked.
1079 		 */
1080 		lock_sock(sk);
1081 		if (likely(up->pending)) {
1082 			if (unlikely(up->pending != AF_INET)) {
1083 				release_sock(sk);
1084 				return -EINVAL;
1085 			}
1086 			goto do_append_data;
1087 		}
1088 		release_sock(sk);
1089 	}
1090 	ulen += sizeof(struct udphdr);
1091 
1092 	/*
1093 	 *	Get and verify the address.
1094 	 */
1095 	if (usin) {
1096 		if (msg->msg_namelen < sizeof(*usin))
1097 			return -EINVAL;
1098 		if (usin->sin_family != AF_INET) {
1099 			if (usin->sin_family != AF_UNSPEC)
1100 				return -EAFNOSUPPORT;
1101 		}
1102 
1103 		daddr = usin->sin_addr.s_addr;
1104 		dport = usin->sin_port;
1105 		if (dport == 0)
1106 			return -EINVAL;
1107 	} else {
1108 		if (sk->sk_state != TCP_ESTABLISHED)
1109 			return -EDESTADDRREQ;
1110 		daddr = inet->inet_daddr;
1111 		dport = inet->inet_dport;
1112 		/* Open fast path for connected socket.
1113 		   Route will not be used, if at least one option is set.
1114 		 */
1115 		connected = 1;
1116 	}
1117 
1118 	ipcm_init_sk(&ipc, inet);
1119 	ipc.gso_size = READ_ONCE(up->gso_size);
1120 
1121 	if (msg->msg_controllen) {
1122 		err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1123 		if (err > 0)
1124 			err = ip_cmsg_send(sk, msg, &ipc,
1125 					   sk->sk_family == AF_INET6);
1126 		if (unlikely(err < 0)) {
1127 			kfree(ipc.opt);
1128 			return err;
1129 		}
1130 		if (ipc.opt)
1131 			free = 1;
1132 		connected = 0;
1133 	}
1134 	if (!ipc.opt) {
1135 		struct ip_options_rcu *inet_opt;
1136 
1137 		rcu_read_lock();
1138 		inet_opt = rcu_dereference(inet->inet_opt);
1139 		if (inet_opt) {
1140 			memcpy(&opt_copy, inet_opt,
1141 			       sizeof(*inet_opt) + inet_opt->opt.optlen);
1142 			ipc.opt = &opt_copy.opt;
1143 		}
1144 		rcu_read_unlock();
1145 	}
1146 
1147 	if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1148 		err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1149 					    (struct sockaddr *)usin, &ipc.addr);
1150 		if (err)
1151 			goto out_free;
1152 		if (usin) {
1153 			if (usin->sin_port == 0) {
1154 				/* BPF program set invalid port. Reject it. */
1155 				err = -EINVAL;
1156 				goto out_free;
1157 			}
1158 			daddr = usin->sin_addr.s_addr;
1159 			dport = usin->sin_port;
1160 		}
1161 	}
1162 
1163 	saddr = ipc.addr;
1164 	ipc.addr = faddr = daddr;
1165 
1166 	if (ipc.opt && ipc.opt->opt.srr) {
1167 		if (!daddr) {
1168 			err = -EINVAL;
1169 			goto out_free;
1170 		}
1171 		faddr = ipc.opt->opt.faddr;
1172 		connected = 0;
1173 	}
1174 	tos = get_rttos(&ipc, inet);
1175 	if (sock_flag(sk, SOCK_LOCALROUTE) ||
1176 	    (msg->msg_flags & MSG_DONTROUTE) ||
1177 	    (ipc.opt && ipc.opt->opt.is_strictroute)) {
1178 		tos |= RTO_ONLINK;
1179 		connected = 0;
1180 	}
1181 
1182 	if (ipv4_is_multicast(daddr)) {
1183 		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1184 			ipc.oif = inet->mc_index;
1185 		if (!saddr)
1186 			saddr = inet->mc_addr;
1187 		connected = 0;
1188 	} else if (!ipc.oif) {
1189 		ipc.oif = inet->uc_index;
1190 	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1191 		/* oif is set, packet is to local broadcast and
1192 		 * uc_index is set. oif is most likely set
1193 		 * by sk_bound_dev_if. If uc_index != oif check if the
1194 		 * oif is an L3 master and uc_index is an L3 slave.
1195 		 * If so, we want to allow the send using the uc_index.
1196 		 */
1197 		if (ipc.oif != inet->uc_index &&
1198 		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1199 							      inet->uc_index)) {
1200 			ipc.oif = inet->uc_index;
1201 		}
1202 	}
1203 
1204 	if (connected)
1205 		rt = (struct rtable *)sk_dst_check(sk, 0);
1206 
1207 	if (!rt) {
1208 		struct net *net = sock_net(sk);
1209 		__u8 flow_flags = inet_sk_flowi_flags(sk);
1210 
1211 		fl4 = &fl4_stack;
1212 
1213 		flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1214 				   RT_SCOPE_UNIVERSE, sk->sk_protocol,
1215 				   flow_flags,
1216 				   faddr, saddr, dport, inet->inet_sport,
1217 				   sk->sk_uid);
1218 
1219 		security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1220 		rt = ip_route_output_flow(net, fl4, sk);
1221 		if (IS_ERR(rt)) {
1222 			err = PTR_ERR(rt);
1223 			rt = NULL;
1224 			if (err == -ENETUNREACH)
1225 				IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1226 			goto out;
1227 		}
1228 
1229 		err = -EACCES;
1230 		if ((rt->rt_flags & RTCF_BROADCAST) &&
1231 		    !sock_flag(sk, SOCK_BROADCAST))
1232 			goto out;
1233 		if (connected)
1234 			sk_dst_set(sk, dst_clone(&rt->dst));
1235 	}
1236 
1237 	if (msg->msg_flags&MSG_CONFIRM)
1238 		goto do_confirm;
1239 back_from_confirm:
1240 
1241 	saddr = fl4->saddr;
1242 	if (!ipc.addr)
1243 		daddr = ipc.addr = fl4->daddr;
1244 
1245 	/* Lockless fast path for the non-corking case. */
1246 	if (!corkreq) {
1247 		struct inet_cork cork;
1248 
1249 		skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1250 				  sizeof(struct udphdr), &ipc, &rt,
1251 				  &cork, msg->msg_flags);
1252 		err = PTR_ERR(skb);
1253 		if (!IS_ERR_OR_NULL(skb))
1254 			err = udp_send_skb(skb, fl4, &cork);
1255 		goto out;
1256 	}
1257 
1258 	lock_sock(sk);
1259 	if (unlikely(up->pending)) {
1260 		/* The socket is already corked while preparing it. */
1261 		/* ... which is an evident application bug. --ANK */
1262 		release_sock(sk);
1263 
1264 		net_dbg_ratelimited("socket already corked\n");
1265 		err = -EINVAL;
1266 		goto out;
1267 	}
1268 	/*
1269 	 *	Now cork the socket to pend data.
1270 	 */
1271 	fl4 = &inet->cork.fl.u.ip4;
1272 	fl4->daddr = daddr;
1273 	fl4->saddr = saddr;
1274 	fl4->fl4_dport = dport;
1275 	fl4->fl4_sport = inet->inet_sport;
1276 	up->pending = AF_INET;
1277 
1278 do_append_data:
1279 	up->len += ulen;
1280 	err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1281 			     sizeof(struct udphdr), &ipc, &rt,
1282 			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1283 	if (err)
1284 		udp_flush_pending_frames(sk);
1285 	else if (!corkreq)
1286 		err = udp_push_pending_frames(sk);
1287 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1288 		up->pending = 0;
1289 	release_sock(sk);
1290 
1291 out:
1292 	ip_rt_put(rt);
1293 out_free:
1294 	if (free)
1295 		kfree(ipc.opt);
1296 	if (!err)
1297 		return len;
1298 	/*
1299 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1300 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1301 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1302 	 * things).  We could add another new stat but at least for now that
1303 	 * seems like overkill.
1304 	 */
1305 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1306 		UDP_INC_STATS(sock_net(sk),
1307 			      UDP_MIB_SNDBUFERRORS, is_udplite);
1308 	}
1309 	return err;
1310 
1311 do_confirm:
1312 	if (msg->msg_flags & MSG_PROBE)
1313 		dst_confirm_neigh(&rt->dst, &fl4->daddr);
1314 	if (!(msg->msg_flags&MSG_PROBE) || len)
1315 		goto back_from_confirm;
1316 	err = 0;
1317 	goto out;
1318 }
1319 EXPORT_SYMBOL(udp_sendmsg);
1320 
1321 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1322 		 size_t size, int flags)
1323 {
1324 	struct inet_sock *inet = inet_sk(sk);
1325 	struct udp_sock *up = udp_sk(sk);
1326 	int ret;
1327 
1328 	if (flags & MSG_SENDPAGE_NOTLAST)
1329 		flags |= MSG_MORE;
1330 
1331 	if (!up->pending) {
1332 		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
1333 
1334 		/* Call udp_sendmsg to specify destination address which
1335 		 * sendpage interface can't pass.
1336 		 * This will succeed only when the socket is connected.
1337 		 */
1338 		ret = udp_sendmsg(sk, &msg, 0);
1339 		if (ret < 0)
1340 			return ret;
1341 	}
1342 
1343 	lock_sock(sk);
1344 
1345 	if (unlikely(!up->pending)) {
1346 		release_sock(sk);
1347 
1348 		net_dbg_ratelimited("cork failed\n");
1349 		return -EINVAL;
1350 	}
1351 
1352 	ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1353 			     page, offset, size, flags);
1354 	if (ret == -EOPNOTSUPP) {
1355 		release_sock(sk);
1356 		return sock_no_sendpage(sk->sk_socket, page, offset,
1357 					size, flags);
1358 	}
1359 	if (ret < 0) {
1360 		udp_flush_pending_frames(sk);
1361 		goto out;
1362 	}
1363 
1364 	up->len += size;
1365 	if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE)))
1366 		ret = udp_push_pending_frames(sk);
1367 	if (!ret)
1368 		ret = size;
1369 out:
1370 	release_sock(sk);
1371 	return ret;
1372 }
1373 
1374 #define UDP_SKB_IS_STATELESS 0x80000000
1375 
1376 /* all head states (dst, sk, nf conntrack) except skb extensions are
1377  * cleared by udp_rcv().
1378  *
1379  * We need to preserve secpath, if present, to eventually process
1380  * IP_CMSG_PASSSEC at recvmsg() time.
1381  *
1382  * Other extensions can be cleared.
1383  */
1384 static bool udp_try_make_stateless(struct sk_buff *skb)
1385 {
1386 	if (!skb_has_extensions(skb))
1387 		return true;
1388 
1389 	if (!secpath_exists(skb)) {
1390 		skb_ext_reset(skb);
1391 		return true;
1392 	}
1393 
1394 	return false;
1395 }
1396 
1397 static void udp_set_dev_scratch(struct sk_buff *skb)
1398 {
1399 	struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1400 
1401 	BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1402 	scratch->_tsize_state = skb->truesize;
1403 #if BITS_PER_LONG == 64
1404 	scratch->len = skb->len;
1405 	scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1406 	scratch->is_linear = !skb_is_nonlinear(skb);
1407 #endif
1408 	if (udp_try_make_stateless(skb))
1409 		scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1410 }
1411 
1412 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1413 {
1414 	/* We come here after udp_lib_checksum_complete() returned 0.
1415 	 * This means that __skb_checksum_complete() might have
1416 	 * set skb->csum_valid to 1.
1417 	 * On 64bit platforms, we can set csum_unnecessary
1418 	 * to true, but only if the skb is not shared.
1419 	 */
1420 #if BITS_PER_LONG == 64
1421 	if (!skb_shared(skb))
1422 		udp_skb_scratch(skb)->csum_unnecessary = true;
1423 #endif
1424 }
1425 
1426 static int udp_skb_truesize(struct sk_buff *skb)
1427 {
1428 	return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1429 }
1430 
1431 static bool udp_skb_has_head_state(struct sk_buff *skb)
1432 {
1433 	return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1434 }
1435 
1436 /* fully reclaim rmem/fwd memory allocated for skb */
1437 static void udp_rmem_release(struct sock *sk, int size, int partial,
1438 			     bool rx_queue_lock_held)
1439 {
1440 	struct udp_sock *up = udp_sk(sk);
1441 	struct sk_buff_head *sk_queue;
1442 	int amt;
1443 
1444 	if (likely(partial)) {
1445 		up->forward_deficit += size;
1446 		size = up->forward_deficit;
1447 		if (size < (sk->sk_rcvbuf >> 2) &&
1448 		    !skb_queue_empty(&up->reader_queue))
1449 			return;
1450 	} else {
1451 		size += up->forward_deficit;
1452 	}
1453 	up->forward_deficit = 0;
1454 
1455 	/* acquire the sk_receive_queue for fwd allocated memory scheduling,
1456 	 * if the called don't held it already
1457 	 */
1458 	sk_queue = &sk->sk_receive_queue;
1459 	if (!rx_queue_lock_held)
1460 		spin_lock(&sk_queue->lock);
1461 
1462 
1463 	sk->sk_forward_alloc += size;
1464 	amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1465 	sk->sk_forward_alloc -= amt;
1466 
1467 	if (amt)
1468 		__sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1469 
1470 	atomic_sub(size, &sk->sk_rmem_alloc);
1471 
1472 	/* this can save us from acquiring the rx queue lock on next receive */
1473 	skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1474 
1475 	if (!rx_queue_lock_held)
1476 		spin_unlock(&sk_queue->lock);
1477 }
1478 
1479 /* Note: called with reader_queue.lock held.
1480  * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1481  * This avoids a cache line miss while receive_queue lock is held.
1482  * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1483  */
1484 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1485 {
1486 	prefetch(&skb->data);
1487 	udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1488 }
1489 EXPORT_SYMBOL(udp_skb_destructor);
1490 
1491 /* as above, but the caller held the rx queue lock, too */
1492 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1493 {
1494 	prefetch(&skb->data);
1495 	udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1496 }
1497 
1498 /* Idea of busylocks is to let producers grab an extra spinlock
1499  * to relieve pressure on the receive_queue spinlock shared by consumer.
1500  * Under flood, this means that only one producer can be in line
1501  * trying to acquire the receive_queue spinlock.
1502  * These busylock can be allocated on a per cpu manner, instead of a
1503  * per socket one (that would consume a cache line per socket)
1504  */
1505 static int udp_busylocks_log __read_mostly;
1506 static spinlock_t *udp_busylocks __read_mostly;
1507 
1508 static spinlock_t *busylock_acquire(void *ptr)
1509 {
1510 	spinlock_t *busy;
1511 
1512 	busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1513 	spin_lock(busy);
1514 	return busy;
1515 }
1516 
1517 static void busylock_release(spinlock_t *busy)
1518 {
1519 	if (busy)
1520 		spin_unlock(busy);
1521 }
1522 
1523 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1524 {
1525 	struct sk_buff_head *list = &sk->sk_receive_queue;
1526 	int rmem, delta, amt, err = -ENOMEM;
1527 	spinlock_t *busy = NULL;
1528 	int size;
1529 
1530 	/* try to avoid the costly atomic add/sub pair when the receive
1531 	 * queue is full; always allow at least a packet
1532 	 */
1533 	rmem = atomic_read(&sk->sk_rmem_alloc);
1534 	if (rmem > sk->sk_rcvbuf)
1535 		goto drop;
1536 
1537 	/* Under mem pressure, it might be helpful to help udp_recvmsg()
1538 	 * having linear skbs :
1539 	 * - Reduce memory overhead and thus increase receive queue capacity
1540 	 * - Less cache line misses at copyout() time
1541 	 * - Less work at consume_skb() (less alien page frag freeing)
1542 	 */
1543 	if (rmem > (sk->sk_rcvbuf >> 1)) {
1544 		skb_condense(skb);
1545 
1546 		busy = busylock_acquire(sk);
1547 	}
1548 	size = skb->truesize;
1549 	udp_set_dev_scratch(skb);
1550 
1551 	/* we drop only if the receive buf is full and the receive
1552 	 * queue contains some other skb
1553 	 */
1554 	rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1555 	if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1556 		goto uncharge_drop;
1557 
1558 	spin_lock(&list->lock);
1559 	if (size >= sk->sk_forward_alloc) {
1560 		amt = sk_mem_pages(size);
1561 		delta = amt << SK_MEM_QUANTUM_SHIFT;
1562 		if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1563 			err = -ENOBUFS;
1564 			spin_unlock(&list->lock);
1565 			goto uncharge_drop;
1566 		}
1567 
1568 		sk->sk_forward_alloc += delta;
1569 	}
1570 
1571 	sk->sk_forward_alloc -= size;
1572 
1573 	/* no need to setup a destructor, we will explicitly release the
1574 	 * forward allocated memory on dequeue
1575 	 */
1576 	sock_skb_set_dropcount(sk, skb);
1577 
1578 	__skb_queue_tail(list, skb);
1579 	spin_unlock(&list->lock);
1580 
1581 	if (!sock_flag(sk, SOCK_DEAD))
1582 		sk->sk_data_ready(sk);
1583 
1584 	busylock_release(busy);
1585 	return 0;
1586 
1587 uncharge_drop:
1588 	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1589 
1590 drop:
1591 	atomic_inc(&sk->sk_drops);
1592 	busylock_release(busy);
1593 	return err;
1594 }
1595 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1596 
1597 void udp_destruct_sock(struct sock *sk)
1598 {
1599 	/* reclaim completely the forward allocated memory */
1600 	struct udp_sock *up = udp_sk(sk);
1601 	unsigned int total = 0;
1602 	struct sk_buff *skb;
1603 
1604 	skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1605 	while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1606 		total += skb->truesize;
1607 		kfree_skb(skb);
1608 	}
1609 	udp_rmem_release(sk, total, 0, true);
1610 
1611 	inet_sock_destruct(sk);
1612 }
1613 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1614 
1615 int udp_init_sock(struct sock *sk)
1616 {
1617 	skb_queue_head_init(&udp_sk(sk)->reader_queue);
1618 	sk->sk_destruct = udp_destruct_sock;
1619 	return 0;
1620 }
1621 EXPORT_SYMBOL_GPL(udp_init_sock);
1622 
1623 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1624 {
1625 	if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1626 		bool slow = lock_sock_fast(sk);
1627 
1628 		sk_peek_offset_bwd(sk, len);
1629 		unlock_sock_fast(sk, slow);
1630 	}
1631 
1632 	if (!skb_unref(skb))
1633 		return;
1634 
1635 	/* In the more common cases we cleared the head states previously,
1636 	 * see __udp_queue_rcv_skb().
1637 	 */
1638 	if (unlikely(udp_skb_has_head_state(skb)))
1639 		skb_release_head_state(skb);
1640 	__consume_stateless_skb(skb);
1641 }
1642 EXPORT_SYMBOL_GPL(skb_consume_udp);
1643 
1644 static struct sk_buff *__first_packet_length(struct sock *sk,
1645 					     struct sk_buff_head *rcvq,
1646 					     int *total)
1647 {
1648 	struct sk_buff *skb;
1649 
1650 	while ((skb = skb_peek(rcvq)) != NULL) {
1651 		if (udp_lib_checksum_complete(skb)) {
1652 			__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1653 					IS_UDPLITE(sk));
1654 			__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1655 					IS_UDPLITE(sk));
1656 			atomic_inc(&sk->sk_drops);
1657 			__skb_unlink(skb, rcvq);
1658 			*total += skb->truesize;
1659 			kfree_skb(skb);
1660 		} else {
1661 			udp_skb_csum_unnecessary_set(skb);
1662 			break;
1663 		}
1664 	}
1665 	return skb;
1666 }
1667 
1668 /**
1669  *	first_packet_length	- return length of first packet in receive queue
1670  *	@sk: socket
1671  *
1672  *	Drops all bad checksum frames, until a valid one is found.
1673  *	Returns the length of found skb, or -1 if none is found.
1674  */
1675 static int first_packet_length(struct sock *sk)
1676 {
1677 	struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1678 	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1679 	struct sk_buff *skb;
1680 	int total = 0;
1681 	int res;
1682 
1683 	spin_lock_bh(&rcvq->lock);
1684 	skb = __first_packet_length(sk, rcvq, &total);
1685 	if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1686 		spin_lock(&sk_queue->lock);
1687 		skb_queue_splice_tail_init(sk_queue, rcvq);
1688 		spin_unlock(&sk_queue->lock);
1689 
1690 		skb = __first_packet_length(sk, rcvq, &total);
1691 	}
1692 	res = skb ? skb->len : -1;
1693 	if (total)
1694 		udp_rmem_release(sk, total, 1, false);
1695 	spin_unlock_bh(&rcvq->lock);
1696 	return res;
1697 }
1698 
1699 /*
1700  *	IOCTL requests applicable to the UDP protocol
1701  */
1702 
1703 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1704 {
1705 	switch (cmd) {
1706 	case SIOCOUTQ:
1707 	{
1708 		int amount = sk_wmem_alloc_get(sk);
1709 
1710 		return put_user(amount, (int __user *)arg);
1711 	}
1712 
1713 	case SIOCINQ:
1714 	{
1715 		int amount = max_t(int, 0, first_packet_length(sk));
1716 
1717 		return put_user(amount, (int __user *)arg);
1718 	}
1719 
1720 	default:
1721 		return -ENOIOCTLCMD;
1722 	}
1723 
1724 	return 0;
1725 }
1726 EXPORT_SYMBOL(udp_ioctl);
1727 
1728 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1729 			       int noblock, int *off, int *err)
1730 {
1731 	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1732 	struct sk_buff_head *queue;
1733 	struct sk_buff *last;
1734 	long timeo;
1735 	int error;
1736 
1737 	queue = &udp_sk(sk)->reader_queue;
1738 	flags |= noblock ? MSG_DONTWAIT : 0;
1739 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1740 	do {
1741 		struct sk_buff *skb;
1742 
1743 		error = sock_error(sk);
1744 		if (error)
1745 			break;
1746 
1747 		error = -EAGAIN;
1748 		do {
1749 			spin_lock_bh(&queue->lock);
1750 			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1751 							err, &last);
1752 			if (skb) {
1753 				if (!(flags & MSG_PEEK))
1754 					udp_skb_destructor(sk, skb);
1755 				spin_unlock_bh(&queue->lock);
1756 				return skb;
1757 			}
1758 
1759 			if (skb_queue_empty_lockless(sk_queue)) {
1760 				spin_unlock_bh(&queue->lock);
1761 				goto busy_check;
1762 			}
1763 
1764 			/* refill the reader queue and walk it again
1765 			 * keep both queues locked to avoid re-acquiring
1766 			 * the sk_receive_queue lock if fwd memory scheduling
1767 			 * is needed.
1768 			 */
1769 			spin_lock(&sk_queue->lock);
1770 			skb_queue_splice_tail_init(sk_queue, queue);
1771 
1772 			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1773 							err, &last);
1774 			if (skb && !(flags & MSG_PEEK))
1775 				udp_skb_dtor_locked(sk, skb);
1776 			spin_unlock(&sk_queue->lock);
1777 			spin_unlock_bh(&queue->lock);
1778 			if (skb)
1779 				return skb;
1780 
1781 busy_check:
1782 			if (!sk_can_busy_loop(sk))
1783 				break;
1784 
1785 			sk_busy_loop(sk, flags & MSG_DONTWAIT);
1786 		} while (!skb_queue_empty_lockless(sk_queue));
1787 
1788 		/* sk_queue is empty, reader_queue may contain peeked packets */
1789 	} while (timeo &&
1790 		 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1791 					      &error, &timeo,
1792 					      (struct sk_buff *)sk_queue));
1793 
1794 	*err = error;
1795 	return NULL;
1796 }
1797 EXPORT_SYMBOL(__skb_recv_udp);
1798 
1799 int udp_read_sock(struct sock *sk, read_descriptor_t *desc,
1800 		  sk_read_actor_t recv_actor)
1801 {
1802 	int copied = 0;
1803 
1804 	while (1) {
1805 		struct sk_buff *skb;
1806 		int err, used;
1807 
1808 		skb = skb_recv_udp(sk, 0, 1, &err);
1809 		if (!skb)
1810 			return err;
1811 
1812 		if (udp_lib_checksum_complete(skb)) {
1813 			__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1814 					IS_UDPLITE(sk));
1815 			__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1816 					IS_UDPLITE(sk));
1817 			atomic_inc(&sk->sk_drops);
1818 			kfree_skb(skb);
1819 			continue;
1820 		}
1821 
1822 		used = recv_actor(desc, skb, 0, skb->len);
1823 		if (used <= 0) {
1824 			if (!copied)
1825 				copied = used;
1826 			kfree_skb(skb);
1827 			break;
1828 		} else if (used <= skb->len) {
1829 			copied += used;
1830 		}
1831 
1832 		kfree_skb(skb);
1833 		if (!desc->count)
1834 			break;
1835 	}
1836 
1837 	return copied;
1838 }
1839 EXPORT_SYMBOL(udp_read_sock);
1840 
1841 /*
1842  * 	This should be easy, if there is something there we
1843  * 	return it, otherwise we block.
1844  */
1845 
1846 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1847 		int flags, int *addr_len)
1848 {
1849 	struct inet_sock *inet = inet_sk(sk);
1850 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1851 	struct sk_buff *skb;
1852 	unsigned int ulen, copied;
1853 	int off, err, peeking = flags & MSG_PEEK;
1854 	int is_udplite = IS_UDPLITE(sk);
1855 	bool checksum_valid = false;
1856 
1857 	if (flags & MSG_ERRQUEUE)
1858 		return ip_recv_error(sk, msg, len, addr_len);
1859 
1860 try_again:
1861 	off = sk_peek_offset(sk, flags);
1862 	skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1863 	if (!skb)
1864 		return err;
1865 
1866 	ulen = udp_skb_len(skb);
1867 	copied = len;
1868 	if (copied > ulen - off)
1869 		copied = ulen - off;
1870 	else if (copied < ulen)
1871 		msg->msg_flags |= MSG_TRUNC;
1872 
1873 	/*
1874 	 * If checksum is needed at all, try to do it while copying the
1875 	 * data.  If the data is truncated, or if we only want a partial
1876 	 * coverage checksum (UDP-Lite), do it before the copy.
1877 	 */
1878 
1879 	if (copied < ulen || peeking ||
1880 	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1881 		checksum_valid = udp_skb_csum_unnecessary(skb) ||
1882 				!__udp_lib_checksum_complete(skb);
1883 		if (!checksum_valid)
1884 			goto csum_copy_err;
1885 	}
1886 
1887 	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1888 		if (udp_skb_is_linear(skb))
1889 			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1890 		else
1891 			err = skb_copy_datagram_msg(skb, off, msg, copied);
1892 	} else {
1893 		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1894 
1895 		if (err == -EINVAL)
1896 			goto csum_copy_err;
1897 	}
1898 
1899 	if (unlikely(err)) {
1900 		if (!peeking) {
1901 			atomic_inc(&sk->sk_drops);
1902 			UDP_INC_STATS(sock_net(sk),
1903 				      UDP_MIB_INERRORS, is_udplite);
1904 		}
1905 		kfree_skb(skb);
1906 		return err;
1907 	}
1908 
1909 	if (!peeking)
1910 		UDP_INC_STATS(sock_net(sk),
1911 			      UDP_MIB_INDATAGRAMS, is_udplite);
1912 
1913 	sock_recv_ts_and_drops(msg, sk, skb);
1914 
1915 	/* Copy the address. */
1916 	if (sin) {
1917 		sin->sin_family = AF_INET;
1918 		sin->sin_port = udp_hdr(skb)->source;
1919 		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1920 		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1921 		*addr_len = sizeof(*sin);
1922 
1923 		BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1924 						      (struct sockaddr *)sin);
1925 	}
1926 
1927 	if (udp_sk(sk)->gro_enabled)
1928 		udp_cmsg_recv(msg, sk, skb);
1929 
1930 	if (inet->cmsg_flags)
1931 		ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1932 
1933 	err = copied;
1934 	if (flags & MSG_TRUNC)
1935 		err = ulen;
1936 
1937 	skb_consume_udp(sk, skb, peeking ? -err : err);
1938 	return err;
1939 
1940 csum_copy_err:
1941 	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1942 				 udp_skb_destructor)) {
1943 		UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1944 		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1945 	}
1946 	kfree_skb(skb);
1947 
1948 	/* starting over for a new packet, but check if we need to yield */
1949 	cond_resched();
1950 	msg->msg_flags &= ~MSG_TRUNC;
1951 	goto try_again;
1952 }
1953 
1954 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1955 {
1956 	/* This check is replicated from __ip4_datagram_connect() and
1957 	 * intended to prevent BPF program called below from accessing bytes
1958 	 * that are out of the bound specified by user in addr_len.
1959 	 */
1960 	if (addr_len < sizeof(struct sockaddr_in))
1961 		return -EINVAL;
1962 
1963 	return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1964 }
1965 EXPORT_SYMBOL(udp_pre_connect);
1966 
1967 int __udp_disconnect(struct sock *sk, int flags)
1968 {
1969 	struct inet_sock *inet = inet_sk(sk);
1970 	/*
1971 	 *	1003.1g - break association.
1972 	 */
1973 
1974 	sk->sk_state = TCP_CLOSE;
1975 	inet->inet_daddr = 0;
1976 	inet->inet_dport = 0;
1977 	sock_rps_reset_rxhash(sk);
1978 	sk->sk_bound_dev_if = 0;
1979 	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1980 		inet_reset_saddr(sk);
1981 		if (sk->sk_prot->rehash &&
1982 		    (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1983 			sk->sk_prot->rehash(sk);
1984 	}
1985 
1986 	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1987 		sk->sk_prot->unhash(sk);
1988 		inet->inet_sport = 0;
1989 	}
1990 	sk_dst_reset(sk);
1991 	return 0;
1992 }
1993 EXPORT_SYMBOL(__udp_disconnect);
1994 
1995 int udp_disconnect(struct sock *sk, int flags)
1996 {
1997 	lock_sock(sk);
1998 	__udp_disconnect(sk, flags);
1999 	release_sock(sk);
2000 	return 0;
2001 }
2002 EXPORT_SYMBOL(udp_disconnect);
2003 
2004 void udp_lib_unhash(struct sock *sk)
2005 {
2006 	if (sk_hashed(sk)) {
2007 		struct udp_table *udptable = sk->sk_prot->h.udp_table;
2008 		struct udp_hslot *hslot, *hslot2;
2009 
2010 		hslot  = udp_hashslot(udptable, sock_net(sk),
2011 				      udp_sk(sk)->udp_port_hash);
2012 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2013 
2014 		spin_lock_bh(&hslot->lock);
2015 		if (rcu_access_pointer(sk->sk_reuseport_cb))
2016 			reuseport_detach_sock(sk);
2017 		if (sk_del_node_init_rcu(sk)) {
2018 			hslot->count--;
2019 			inet_sk(sk)->inet_num = 0;
2020 			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2021 
2022 			spin_lock(&hslot2->lock);
2023 			hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2024 			hslot2->count--;
2025 			spin_unlock(&hslot2->lock);
2026 		}
2027 		spin_unlock_bh(&hslot->lock);
2028 	}
2029 }
2030 EXPORT_SYMBOL(udp_lib_unhash);
2031 
2032 /*
2033  * inet_rcv_saddr was changed, we must rehash secondary hash
2034  */
2035 void udp_lib_rehash(struct sock *sk, u16 newhash)
2036 {
2037 	if (sk_hashed(sk)) {
2038 		struct udp_table *udptable = sk->sk_prot->h.udp_table;
2039 		struct udp_hslot *hslot, *hslot2, *nhslot2;
2040 
2041 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2042 		nhslot2 = udp_hashslot2(udptable, newhash);
2043 		udp_sk(sk)->udp_portaddr_hash = newhash;
2044 
2045 		if (hslot2 != nhslot2 ||
2046 		    rcu_access_pointer(sk->sk_reuseport_cb)) {
2047 			hslot = udp_hashslot(udptable, sock_net(sk),
2048 					     udp_sk(sk)->udp_port_hash);
2049 			/* we must lock primary chain too */
2050 			spin_lock_bh(&hslot->lock);
2051 			if (rcu_access_pointer(sk->sk_reuseport_cb))
2052 				reuseport_detach_sock(sk);
2053 
2054 			if (hslot2 != nhslot2) {
2055 				spin_lock(&hslot2->lock);
2056 				hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2057 				hslot2->count--;
2058 				spin_unlock(&hslot2->lock);
2059 
2060 				spin_lock(&nhslot2->lock);
2061 				hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2062 							 &nhslot2->head);
2063 				nhslot2->count++;
2064 				spin_unlock(&nhslot2->lock);
2065 			}
2066 
2067 			spin_unlock_bh(&hslot->lock);
2068 		}
2069 	}
2070 }
2071 EXPORT_SYMBOL(udp_lib_rehash);
2072 
2073 void udp_v4_rehash(struct sock *sk)
2074 {
2075 	u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2076 					  inet_sk(sk)->inet_rcv_saddr,
2077 					  inet_sk(sk)->inet_num);
2078 	udp_lib_rehash(sk, new_hash);
2079 }
2080 
2081 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2082 {
2083 	int rc;
2084 
2085 	if (inet_sk(sk)->inet_daddr) {
2086 		sock_rps_save_rxhash(sk, skb);
2087 		sk_mark_napi_id(sk, skb);
2088 		sk_incoming_cpu_update(sk);
2089 	} else {
2090 		sk_mark_napi_id_once(sk, skb);
2091 	}
2092 
2093 	rc = __udp_enqueue_schedule_skb(sk, skb);
2094 	if (rc < 0) {
2095 		int is_udplite = IS_UDPLITE(sk);
2096 		int drop_reason;
2097 
2098 		/* Note that an ENOMEM error is charged twice */
2099 		if (rc == -ENOMEM) {
2100 			UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2101 					is_udplite);
2102 			drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2103 		} else {
2104 			UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2105 				      is_udplite);
2106 			drop_reason = SKB_DROP_REASON_PROTO_MEM;
2107 		}
2108 		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2109 		kfree_skb_reason(skb, drop_reason);
2110 		trace_udp_fail_queue_rcv_skb(rc, sk);
2111 		return -1;
2112 	}
2113 
2114 	return 0;
2115 }
2116 
2117 /* returns:
2118  *  -1: error
2119  *   0: success
2120  *  >0: "udp encap" protocol resubmission
2121  *
2122  * Note that in the success and error cases, the skb is assumed to
2123  * have either been requeued or freed.
2124  */
2125 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2126 {
2127 	int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2128 	struct udp_sock *up = udp_sk(sk);
2129 	int is_udplite = IS_UDPLITE(sk);
2130 
2131 	/*
2132 	 *	Charge it to the socket, dropping if the queue is full.
2133 	 */
2134 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2135 		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2136 		goto drop;
2137 	}
2138 	nf_reset_ct(skb);
2139 
2140 	if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2141 		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2142 
2143 		/*
2144 		 * This is an encapsulation socket so pass the skb to
2145 		 * the socket's udp_encap_rcv() hook. Otherwise, just
2146 		 * fall through and pass this up the UDP socket.
2147 		 * up->encap_rcv() returns the following value:
2148 		 * =0 if skb was successfully passed to the encap
2149 		 *    handler or was discarded by it.
2150 		 * >0 if skb should be passed on to UDP.
2151 		 * <0 if skb should be resubmitted as proto -N
2152 		 */
2153 
2154 		/* if we're overly short, let UDP handle it */
2155 		encap_rcv = READ_ONCE(up->encap_rcv);
2156 		if (encap_rcv) {
2157 			int ret;
2158 
2159 			/* Verify checksum before giving to encap */
2160 			if (udp_lib_checksum_complete(skb))
2161 				goto csum_error;
2162 
2163 			ret = encap_rcv(sk, skb);
2164 			if (ret <= 0) {
2165 				__UDP_INC_STATS(sock_net(sk),
2166 						UDP_MIB_INDATAGRAMS,
2167 						is_udplite);
2168 				return -ret;
2169 			}
2170 		}
2171 
2172 		/* FALLTHROUGH -- it's a UDP Packet */
2173 	}
2174 
2175 	/*
2176 	 * 	UDP-Lite specific tests, ignored on UDP sockets
2177 	 */
2178 	if ((up->pcflag & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
2179 
2180 		/*
2181 		 * MIB statistics other than incrementing the error count are
2182 		 * disabled for the following two types of errors: these depend
2183 		 * on the application settings, not on the functioning of the
2184 		 * protocol stack as such.
2185 		 *
2186 		 * RFC 3828 here recommends (sec 3.3): "There should also be a
2187 		 * way ... to ... at least let the receiving application block
2188 		 * delivery of packets with coverage values less than a value
2189 		 * provided by the application."
2190 		 */
2191 		if (up->pcrlen == 0) {          /* full coverage was set  */
2192 			net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2193 					    UDP_SKB_CB(skb)->cscov, skb->len);
2194 			goto drop;
2195 		}
2196 		/* The next case involves violating the min. coverage requested
2197 		 * by the receiver. This is subtle: if receiver wants x and x is
2198 		 * greater than the buffersize/MTU then receiver will complain
2199 		 * that it wants x while sender emits packets of smaller size y.
2200 		 * Therefore the above ...()->partial_cov statement is essential.
2201 		 */
2202 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
2203 			net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2204 					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
2205 			goto drop;
2206 		}
2207 	}
2208 
2209 	prefetch(&sk->sk_rmem_alloc);
2210 	if (rcu_access_pointer(sk->sk_filter) &&
2211 	    udp_lib_checksum_complete(skb))
2212 			goto csum_error;
2213 
2214 	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2215 		drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2216 		goto drop;
2217 	}
2218 
2219 	udp_csum_pull_header(skb);
2220 
2221 	ipv4_pktinfo_prepare(sk, skb);
2222 	return __udp_queue_rcv_skb(sk, skb);
2223 
2224 csum_error:
2225 	drop_reason = SKB_DROP_REASON_UDP_CSUM;
2226 	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2227 drop:
2228 	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2229 	atomic_inc(&sk->sk_drops);
2230 	kfree_skb_reason(skb, drop_reason);
2231 	return -1;
2232 }
2233 
2234 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2235 {
2236 	struct sk_buff *next, *segs;
2237 	int ret;
2238 
2239 	if (likely(!udp_unexpected_gso(sk, skb)))
2240 		return udp_queue_rcv_one_skb(sk, skb);
2241 
2242 	BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2243 	__skb_push(skb, -skb_mac_offset(skb));
2244 	segs = udp_rcv_segment(sk, skb, true);
2245 	skb_list_walk_safe(segs, skb, next) {
2246 		__skb_pull(skb, skb_transport_offset(skb));
2247 
2248 		udp_post_segment_fix_csum(skb);
2249 		ret = udp_queue_rcv_one_skb(sk, skb);
2250 		if (ret > 0)
2251 			ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2252 	}
2253 	return 0;
2254 }
2255 
2256 /* For TCP sockets, sk_rx_dst is protected by socket lock
2257  * For UDP, we use xchg() to guard against concurrent changes.
2258  */
2259 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2260 {
2261 	struct dst_entry *old;
2262 
2263 	if (dst_hold_safe(dst)) {
2264 		old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2265 		dst_release(old);
2266 		return old != dst;
2267 	}
2268 	return false;
2269 }
2270 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2271 
2272 /*
2273  *	Multicasts and broadcasts go to each listener.
2274  *
2275  *	Note: called only from the BH handler context.
2276  */
2277 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2278 				    struct udphdr  *uh,
2279 				    __be32 saddr, __be32 daddr,
2280 				    struct udp_table *udptable,
2281 				    int proto)
2282 {
2283 	struct sock *sk, *first = NULL;
2284 	unsigned short hnum = ntohs(uh->dest);
2285 	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2286 	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2287 	unsigned int offset = offsetof(typeof(*sk), sk_node);
2288 	int dif = skb->dev->ifindex;
2289 	int sdif = inet_sdif(skb);
2290 	struct hlist_node *node;
2291 	struct sk_buff *nskb;
2292 
2293 	if (use_hash2) {
2294 		hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2295 			    udptable->mask;
2296 		hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2297 start_lookup:
2298 		hslot = &udptable->hash2[hash2];
2299 		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2300 	}
2301 
2302 	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2303 		if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2304 					 uh->source, saddr, dif, sdif, hnum))
2305 			continue;
2306 
2307 		if (!first) {
2308 			first = sk;
2309 			continue;
2310 		}
2311 		nskb = skb_clone(skb, GFP_ATOMIC);
2312 
2313 		if (unlikely(!nskb)) {
2314 			atomic_inc(&sk->sk_drops);
2315 			__UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2316 					IS_UDPLITE(sk));
2317 			__UDP_INC_STATS(net, UDP_MIB_INERRORS,
2318 					IS_UDPLITE(sk));
2319 			continue;
2320 		}
2321 		if (udp_queue_rcv_skb(sk, nskb) > 0)
2322 			consume_skb(nskb);
2323 	}
2324 
2325 	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
2326 	if (use_hash2 && hash2 != hash2_any) {
2327 		hash2 = hash2_any;
2328 		goto start_lookup;
2329 	}
2330 
2331 	if (first) {
2332 		if (udp_queue_rcv_skb(first, skb) > 0)
2333 			consume_skb(skb);
2334 	} else {
2335 		kfree_skb(skb);
2336 		__UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2337 				proto == IPPROTO_UDPLITE);
2338 	}
2339 	return 0;
2340 }
2341 
2342 /* Initialize UDP checksum. If exited with zero value (success),
2343  * CHECKSUM_UNNECESSARY means, that no more checks are required.
2344  * Otherwise, csum completion requires checksumming packet body,
2345  * including udp header and folding it to skb->csum.
2346  */
2347 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2348 				 int proto)
2349 {
2350 	int err;
2351 
2352 	UDP_SKB_CB(skb)->partial_cov = 0;
2353 	UDP_SKB_CB(skb)->cscov = skb->len;
2354 
2355 	if (proto == IPPROTO_UDPLITE) {
2356 		err = udplite_checksum_init(skb, uh);
2357 		if (err)
2358 			return err;
2359 
2360 		if (UDP_SKB_CB(skb)->partial_cov) {
2361 			skb->csum = inet_compute_pseudo(skb, proto);
2362 			return 0;
2363 		}
2364 	}
2365 
2366 	/* Note, we are only interested in != 0 or == 0, thus the
2367 	 * force to int.
2368 	 */
2369 	err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2370 							inet_compute_pseudo);
2371 	if (err)
2372 		return err;
2373 
2374 	if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2375 		/* If SW calculated the value, we know it's bad */
2376 		if (skb->csum_complete_sw)
2377 			return 1;
2378 
2379 		/* HW says the value is bad. Let's validate that.
2380 		 * skb->csum is no longer the full packet checksum,
2381 		 * so don't treat it as such.
2382 		 */
2383 		skb_checksum_complete_unset(skb);
2384 	}
2385 
2386 	return 0;
2387 }
2388 
2389 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2390  * return code conversion for ip layer consumption
2391  */
2392 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2393 			       struct udphdr *uh)
2394 {
2395 	int ret;
2396 
2397 	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2398 		skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2399 
2400 	ret = udp_queue_rcv_skb(sk, skb);
2401 
2402 	/* a return value > 0 means to resubmit the input, but
2403 	 * it wants the return to be -protocol, or 0
2404 	 */
2405 	if (ret > 0)
2406 		return -ret;
2407 	return 0;
2408 }
2409 
2410 /*
2411  *	All we need to do is get the socket, and then do a checksum.
2412  */
2413 
2414 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2415 		   int proto)
2416 {
2417 	struct sock *sk;
2418 	struct udphdr *uh;
2419 	unsigned short ulen;
2420 	struct rtable *rt = skb_rtable(skb);
2421 	__be32 saddr, daddr;
2422 	struct net *net = dev_net(skb->dev);
2423 	bool refcounted;
2424 	int drop_reason;
2425 
2426 	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2427 
2428 	/*
2429 	 *  Validate the packet.
2430 	 */
2431 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2432 		goto drop;		/* No space for header. */
2433 
2434 	uh   = udp_hdr(skb);
2435 	ulen = ntohs(uh->len);
2436 	saddr = ip_hdr(skb)->saddr;
2437 	daddr = ip_hdr(skb)->daddr;
2438 
2439 	if (ulen > skb->len)
2440 		goto short_packet;
2441 
2442 	if (proto == IPPROTO_UDP) {
2443 		/* UDP validates ulen. */
2444 		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2445 			goto short_packet;
2446 		uh = udp_hdr(skb);
2447 	}
2448 
2449 	if (udp4_csum_init(skb, uh, proto))
2450 		goto csum_error;
2451 
2452 	sk = skb_steal_sock(skb, &refcounted);
2453 	if (sk) {
2454 		struct dst_entry *dst = skb_dst(skb);
2455 		int ret;
2456 
2457 		if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2458 			udp_sk_rx_dst_set(sk, dst);
2459 
2460 		ret = udp_unicast_rcv_skb(sk, skb, uh);
2461 		if (refcounted)
2462 			sock_put(sk);
2463 		return ret;
2464 	}
2465 
2466 	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2467 		return __udp4_lib_mcast_deliver(net, skb, uh,
2468 						saddr, daddr, udptable, proto);
2469 
2470 	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2471 	if (sk)
2472 		return udp_unicast_rcv_skb(sk, skb, uh);
2473 
2474 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2475 		goto drop;
2476 	nf_reset_ct(skb);
2477 
2478 	/* No socket. Drop packet silently, if checksum is wrong */
2479 	if (udp_lib_checksum_complete(skb))
2480 		goto csum_error;
2481 
2482 	drop_reason = SKB_DROP_REASON_NO_SOCKET;
2483 	__UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2484 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2485 
2486 	/*
2487 	 * Hmm.  We got an UDP packet to a port to which we
2488 	 * don't wanna listen.  Ignore it.
2489 	 */
2490 	kfree_skb_reason(skb, drop_reason);
2491 	return 0;
2492 
2493 short_packet:
2494 	drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2495 	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2496 			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2497 			    &saddr, ntohs(uh->source),
2498 			    ulen, skb->len,
2499 			    &daddr, ntohs(uh->dest));
2500 	goto drop;
2501 
2502 csum_error:
2503 	/*
2504 	 * RFC1122: OK.  Discards the bad packet silently (as far as
2505 	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2506 	 */
2507 	drop_reason = SKB_DROP_REASON_UDP_CSUM;
2508 	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2509 			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2510 			    &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2511 			    ulen);
2512 	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2513 drop:
2514 	__UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2515 	kfree_skb_reason(skb, drop_reason);
2516 	return 0;
2517 }
2518 
2519 /* We can only early demux multicast if there is a single matching socket.
2520  * If more than one socket found returns NULL
2521  */
2522 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2523 						  __be16 loc_port, __be32 loc_addr,
2524 						  __be16 rmt_port, __be32 rmt_addr,
2525 						  int dif, int sdif)
2526 {
2527 	struct sock *sk, *result;
2528 	unsigned short hnum = ntohs(loc_port);
2529 	unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2530 	struct udp_hslot *hslot = &udp_table.hash[slot];
2531 
2532 	/* Do not bother scanning a too big list */
2533 	if (hslot->count > 10)
2534 		return NULL;
2535 
2536 	result = NULL;
2537 	sk_for_each_rcu(sk, &hslot->head) {
2538 		if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2539 					rmt_port, rmt_addr, dif, sdif, hnum)) {
2540 			if (result)
2541 				return NULL;
2542 			result = sk;
2543 		}
2544 	}
2545 
2546 	return result;
2547 }
2548 
2549 /* For unicast we should only early demux connected sockets or we can
2550  * break forwarding setups.  The chains here can be long so only check
2551  * if the first socket is an exact match and if not move on.
2552  */
2553 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2554 					    __be16 loc_port, __be32 loc_addr,
2555 					    __be16 rmt_port, __be32 rmt_addr,
2556 					    int dif, int sdif)
2557 {
2558 	unsigned short hnum = ntohs(loc_port);
2559 	unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2560 	unsigned int slot2 = hash2 & udp_table.mask;
2561 	struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2562 	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2563 	const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2564 	struct sock *sk;
2565 
2566 	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2567 		if (INET_MATCH(sk, net, acookie, rmt_addr,
2568 			       loc_addr, ports, dif, sdif))
2569 			return sk;
2570 		/* Only check first socket in chain */
2571 		break;
2572 	}
2573 	return NULL;
2574 }
2575 
2576 int udp_v4_early_demux(struct sk_buff *skb)
2577 {
2578 	struct net *net = dev_net(skb->dev);
2579 	struct in_device *in_dev = NULL;
2580 	const struct iphdr *iph;
2581 	const struct udphdr *uh;
2582 	struct sock *sk = NULL;
2583 	struct dst_entry *dst;
2584 	int dif = skb->dev->ifindex;
2585 	int sdif = inet_sdif(skb);
2586 	int ours;
2587 
2588 	/* validate the packet */
2589 	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2590 		return 0;
2591 
2592 	iph = ip_hdr(skb);
2593 	uh = udp_hdr(skb);
2594 
2595 	if (skb->pkt_type == PACKET_MULTICAST) {
2596 		in_dev = __in_dev_get_rcu(skb->dev);
2597 
2598 		if (!in_dev)
2599 			return 0;
2600 
2601 		ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2602 				       iph->protocol);
2603 		if (!ours)
2604 			return 0;
2605 
2606 		sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2607 						   uh->source, iph->saddr,
2608 						   dif, sdif);
2609 	} else if (skb->pkt_type == PACKET_HOST) {
2610 		sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2611 					     uh->source, iph->saddr, dif, sdif);
2612 	}
2613 
2614 	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2615 		return 0;
2616 
2617 	skb->sk = sk;
2618 	skb->destructor = sock_efree;
2619 	dst = rcu_dereference(sk->sk_rx_dst);
2620 
2621 	if (dst)
2622 		dst = dst_check(dst, 0);
2623 	if (dst) {
2624 		u32 itag = 0;
2625 
2626 		/* set noref for now.
2627 		 * any place which wants to hold dst has to call
2628 		 * dst_hold_safe()
2629 		 */
2630 		skb_dst_set_noref(skb, dst);
2631 
2632 		/* for unconnected multicast sockets we need to validate
2633 		 * the source on each packet
2634 		 */
2635 		if (!inet_sk(sk)->inet_daddr && in_dev)
2636 			return ip_mc_validate_source(skb, iph->daddr,
2637 						     iph->saddr,
2638 						     iph->tos & IPTOS_RT_MASK,
2639 						     skb->dev, in_dev, &itag);
2640 	}
2641 	return 0;
2642 }
2643 
2644 int udp_rcv(struct sk_buff *skb)
2645 {
2646 	return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2647 }
2648 
2649 void udp_destroy_sock(struct sock *sk)
2650 {
2651 	struct udp_sock *up = udp_sk(sk);
2652 	bool slow = lock_sock_fast(sk);
2653 
2654 	/* protects from races with udp_abort() */
2655 	sock_set_flag(sk, SOCK_DEAD);
2656 	udp_flush_pending_frames(sk);
2657 	unlock_sock_fast(sk, slow);
2658 	if (static_branch_unlikely(&udp_encap_needed_key)) {
2659 		if (up->encap_type) {
2660 			void (*encap_destroy)(struct sock *sk);
2661 			encap_destroy = READ_ONCE(up->encap_destroy);
2662 			if (encap_destroy)
2663 				encap_destroy(sk);
2664 		}
2665 		if (up->encap_enabled)
2666 			static_branch_dec(&udp_encap_needed_key);
2667 	}
2668 }
2669 
2670 /*
2671  *	Socket option code for UDP
2672  */
2673 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2674 		       sockptr_t optval, unsigned int optlen,
2675 		       int (*push_pending_frames)(struct sock *))
2676 {
2677 	struct udp_sock *up = udp_sk(sk);
2678 	int val, valbool;
2679 	int err = 0;
2680 	int is_udplite = IS_UDPLITE(sk);
2681 
2682 	if (optlen < sizeof(int))
2683 		return -EINVAL;
2684 
2685 	if (copy_from_sockptr(&val, optval, sizeof(val)))
2686 		return -EFAULT;
2687 
2688 	valbool = val ? 1 : 0;
2689 
2690 	switch (optname) {
2691 	case UDP_CORK:
2692 		if (val != 0) {
2693 			WRITE_ONCE(up->corkflag, 1);
2694 		} else {
2695 			WRITE_ONCE(up->corkflag, 0);
2696 			lock_sock(sk);
2697 			push_pending_frames(sk);
2698 			release_sock(sk);
2699 		}
2700 		break;
2701 
2702 	case UDP_ENCAP:
2703 		switch (val) {
2704 		case 0:
2705 #ifdef CONFIG_XFRM
2706 		case UDP_ENCAP_ESPINUDP:
2707 		case UDP_ENCAP_ESPINUDP_NON_IKE:
2708 #if IS_ENABLED(CONFIG_IPV6)
2709 			if (sk->sk_family == AF_INET6)
2710 				up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2711 			else
2712 #endif
2713 				up->encap_rcv = xfrm4_udp_encap_rcv;
2714 #endif
2715 			fallthrough;
2716 		case UDP_ENCAP_L2TPINUDP:
2717 			up->encap_type = val;
2718 			lock_sock(sk);
2719 			udp_tunnel_encap_enable(sk->sk_socket);
2720 			release_sock(sk);
2721 			break;
2722 		default:
2723 			err = -ENOPROTOOPT;
2724 			break;
2725 		}
2726 		break;
2727 
2728 	case UDP_NO_CHECK6_TX:
2729 		up->no_check6_tx = valbool;
2730 		break;
2731 
2732 	case UDP_NO_CHECK6_RX:
2733 		up->no_check6_rx = valbool;
2734 		break;
2735 
2736 	case UDP_SEGMENT:
2737 		if (val < 0 || val > USHRT_MAX)
2738 			return -EINVAL;
2739 		WRITE_ONCE(up->gso_size, val);
2740 		break;
2741 
2742 	case UDP_GRO:
2743 		lock_sock(sk);
2744 
2745 		/* when enabling GRO, accept the related GSO packet type */
2746 		if (valbool)
2747 			udp_tunnel_encap_enable(sk->sk_socket);
2748 		up->gro_enabled = valbool;
2749 		up->accept_udp_l4 = valbool;
2750 		release_sock(sk);
2751 		break;
2752 
2753 	/*
2754 	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
2755 	 */
2756 	/* The sender sets actual checksum coverage length via this option.
2757 	 * The case coverage > packet length is handled by send module. */
2758 	case UDPLITE_SEND_CSCOV:
2759 		if (!is_udplite)         /* Disable the option on UDP sockets */
2760 			return -ENOPROTOOPT;
2761 		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2762 			val = 8;
2763 		else if (val > USHRT_MAX)
2764 			val = USHRT_MAX;
2765 		up->pcslen = val;
2766 		up->pcflag |= UDPLITE_SEND_CC;
2767 		break;
2768 
2769 	/* The receiver specifies a minimum checksum coverage value. To make
2770 	 * sense, this should be set to at least 8 (as done below). If zero is
2771 	 * used, this again means full checksum coverage.                     */
2772 	case UDPLITE_RECV_CSCOV:
2773 		if (!is_udplite)         /* Disable the option on UDP sockets */
2774 			return -ENOPROTOOPT;
2775 		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2776 			val = 8;
2777 		else if (val > USHRT_MAX)
2778 			val = USHRT_MAX;
2779 		up->pcrlen = val;
2780 		up->pcflag |= UDPLITE_RECV_CC;
2781 		break;
2782 
2783 	default:
2784 		err = -ENOPROTOOPT;
2785 		break;
2786 	}
2787 
2788 	return err;
2789 }
2790 EXPORT_SYMBOL(udp_lib_setsockopt);
2791 
2792 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2793 		   unsigned int optlen)
2794 {
2795 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2796 		return udp_lib_setsockopt(sk, level, optname,
2797 					  optval, optlen,
2798 					  udp_push_pending_frames);
2799 	return ip_setsockopt(sk, level, optname, optval, optlen);
2800 }
2801 
2802 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2803 		       char __user *optval, int __user *optlen)
2804 {
2805 	struct udp_sock *up = udp_sk(sk);
2806 	int val, len;
2807 
2808 	if (get_user(len, optlen))
2809 		return -EFAULT;
2810 
2811 	len = min_t(unsigned int, len, sizeof(int));
2812 
2813 	if (len < 0)
2814 		return -EINVAL;
2815 
2816 	switch (optname) {
2817 	case UDP_CORK:
2818 		val = READ_ONCE(up->corkflag);
2819 		break;
2820 
2821 	case UDP_ENCAP:
2822 		val = up->encap_type;
2823 		break;
2824 
2825 	case UDP_NO_CHECK6_TX:
2826 		val = up->no_check6_tx;
2827 		break;
2828 
2829 	case UDP_NO_CHECK6_RX:
2830 		val = up->no_check6_rx;
2831 		break;
2832 
2833 	case UDP_SEGMENT:
2834 		val = READ_ONCE(up->gso_size);
2835 		break;
2836 
2837 	case UDP_GRO:
2838 		val = up->gro_enabled;
2839 		break;
2840 
2841 	/* The following two cannot be changed on UDP sockets, the return is
2842 	 * always 0 (which corresponds to the full checksum coverage of UDP). */
2843 	case UDPLITE_SEND_CSCOV:
2844 		val = up->pcslen;
2845 		break;
2846 
2847 	case UDPLITE_RECV_CSCOV:
2848 		val = up->pcrlen;
2849 		break;
2850 
2851 	default:
2852 		return -ENOPROTOOPT;
2853 	}
2854 
2855 	if (put_user(len, optlen))
2856 		return -EFAULT;
2857 	if (copy_to_user(optval, &val, len))
2858 		return -EFAULT;
2859 	return 0;
2860 }
2861 EXPORT_SYMBOL(udp_lib_getsockopt);
2862 
2863 int udp_getsockopt(struct sock *sk, int level, int optname,
2864 		   char __user *optval, int __user *optlen)
2865 {
2866 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2867 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2868 	return ip_getsockopt(sk, level, optname, optval, optlen);
2869 }
2870 
2871 /**
2872  * 	udp_poll - wait for a UDP event.
2873  *	@file: - file struct
2874  *	@sock: - socket
2875  *	@wait: - poll table
2876  *
2877  *	This is same as datagram poll, except for the special case of
2878  *	blocking sockets. If application is using a blocking fd
2879  *	and a packet with checksum error is in the queue;
2880  *	then it could get return from select indicating data available
2881  *	but then block when reading it. Add special case code
2882  *	to work around these arguably broken applications.
2883  */
2884 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2885 {
2886 	__poll_t mask = datagram_poll(file, sock, wait);
2887 	struct sock *sk = sock->sk;
2888 
2889 	if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2890 		mask |= EPOLLIN | EPOLLRDNORM;
2891 
2892 	/* Check for false positives due to checksum errors */
2893 	if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2894 	    !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2895 		mask &= ~(EPOLLIN | EPOLLRDNORM);
2896 
2897 	/* psock ingress_msg queue should not contain any bad checksum frames */
2898 	if (sk_is_readable(sk))
2899 		mask |= EPOLLIN | EPOLLRDNORM;
2900 	return mask;
2901 
2902 }
2903 EXPORT_SYMBOL(udp_poll);
2904 
2905 int udp_abort(struct sock *sk, int err)
2906 {
2907 	lock_sock(sk);
2908 
2909 	/* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2910 	 * with close()
2911 	 */
2912 	if (sock_flag(sk, SOCK_DEAD))
2913 		goto out;
2914 
2915 	sk->sk_err = err;
2916 	sk_error_report(sk);
2917 	__udp_disconnect(sk, 0);
2918 
2919 out:
2920 	release_sock(sk);
2921 
2922 	return 0;
2923 }
2924 EXPORT_SYMBOL_GPL(udp_abort);
2925 
2926 struct proto udp_prot = {
2927 	.name			= "UDP",
2928 	.owner			= THIS_MODULE,
2929 	.close			= udp_lib_close,
2930 	.pre_connect		= udp_pre_connect,
2931 	.connect		= ip4_datagram_connect,
2932 	.disconnect		= udp_disconnect,
2933 	.ioctl			= udp_ioctl,
2934 	.init			= udp_init_sock,
2935 	.destroy		= udp_destroy_sock,
2936 	.setsockopt		= udp_setsockopt,
2937 	.getsockopt		= udp_getsockopt,
2938 	.sendmsg		= udp_sendmsg,
2939 	.recvmsg		= udp_recvmsg,
2940 	.sendpage		= udp_sendpage,
2941 	.release_cb		= ip4_datagram_release_cb,
2942 	.hash			= udp_lib_hash,
2943 	.unhash			= udp_lib_unhash,
2944 	.rehash			= udp_v4_rehash,
2945 	.get_port		= udp_v4_get_port,
2946 	.put_port		= udp_lib_unhash,
2947 #ifdef CONFIG_BPF_SYSCALL
2948 	.psock_update_sk_prot	= udp_bpf_update_proto,
2949 #endif
2950 	.memory_allocated	= &udp_memory_allocated,
2951 	.sysctl_mem		= sysctl_udp_mem,
2952 	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2953 	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2954 	.obj_size		= sizeof(struct udp_sock),
2955 	.h.udp_table		= &udp_table,
2956 	.diag_destroy		= udp_abort,
2957 };
2958 EXPORT_SYMBOL(udp_prot);
2959 
2960 /* ------------------------------------------------------------------------ */
2961 #ifdef CONFIG_PROC_FS
2962 
2963 static struct sock *udp_get_first(struct seq_file *seq, int start)
2964 {
2965 	struct sock *sk;
2966 	struct udp_seq_afinfo *afinfo;
2967 	struct udp_iter_state *state = seq->private;
2968 	struct net *net = seq_file_net(seq);
2969 
2970 	if (state->bpf_seq_afinfo)
2971 		afinfo = state->bpf_seq_afinfo;
2972 	else
2973 		afinfo = pde_data(file_inode(seq->file));
2974 
2975 	for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2976 	     ++state->bucket) {
2977 		struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2978 
2979 		if (hlist_empty(&hslot->head))
2980 			continue;
2981 
2982 		spin_lock_bh(&hslot->lock);
2983 		sk_for_each(sk, &hslot->head) {
2984 			if (!net_eq(sock_net(sk), net))
2985 				continue;
2986 			if (afinfo->family == AF_UNSPEC ||
2987 			    sk->sk_family == afinfo->family)
2988 				goto found;
2989 		}
2990 		spin_unlock_bh(&hslot->lock);
2991 	}
2992 	sk = NULL;
2993 found:
2994 	return sk;
2995 }
2996 
2997 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2998 {
2999 	struct udp_seq_afinfo *afinfo;
3000 	struct udp_iter_state *state = seq->private;
3001 	struct net *net = seq_file_net(seq);
3002 
3003 	if (state->bpf_seq_afinfo)
3004 		afinfo = state->bpf_seq_afinfo;
3005 	else
3006 		afinfo = pde_data(file_inode(seq->file));
3007 
3008 	do {
3009 		sk = sk_next(sk);
3010 	} while (sk && (!net_eq(sock_net(sk), net) ||
3011 			(afinfo->family != AF_UNSPEC &&
3012 			 sk->sk_family != afinfo->family)));
3013 
3014 	if (!sk) {
3015 		if (state->bucket <= afinfo->udp_table->mask)
3016 			spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
3017 		return udp_get_first(seq, state->bucket + 1);
3018 	}
3019 	return sk;
3020 }
3021 
3022 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3023 {
3024 	struct sock *sk = udp_get_first(seq, 0);
3025 
3026 	if (sk)
3027 		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3028 			--pos;
3029 	return pos ? NULL : sk;
3030 }
3031 
3032 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3033 {
3034 	struct udp_iter_state *state = seq->private;
3035 	state->bucket = MAX_UDP_PORTS;
3036 
3037 	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3038 }
3039 EXPORT_SYMBOL(udp_seq_start);
3040 
3041 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3042 {
3043 	struct sock *sk;
3044 
3045 	if (v == SEQ_START_TOKEN)
3046 		sk = udp_get_idx(seq, 0);
3047 	else
3048 		sk = udp_get_next(seq, v);
3049 
3050 	++*pos;
3051 	return sk;
3052 }
3053 EXPORT_SYMBOL(udp_seq_next);
3054 
3055 void udp_seq_stop(struct seq_file *seq, void *v)
3056 {
3057 	struct udp_seq_afinfo *afinfo;
3058 	struct udp_iter_state *state = seq->private;
3059 
3060 	if (state->bpf_seq_afinfo)
3061 		afinfo = state->bpf_seq_afinfo;
3062 	else
3063 		afinfo = pde_data(file_inode(seq->file));
3064 
3065 	if (state->bucket <= afinfo->udp_table->mask)
3066 		spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
3067 }
3068 EXPORT_SYMBOL(udp_seq_stop);
3069 
3070 /* ------------------------------------------------------------------------ */
3071 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3072 		int bucket)
3073 {
3074 	struct inet_sock *inet = inet_sk(sp);
3075 	__be32 dest = inet->inet_daddr;
3076 	__be32 src  = inet->inet_rcv_saddr;
3077 	__u16 destp	  = ntohs(inet->inet_dport);
3078 	__u16 srcp	  = ntohs(inet->inet_sport);
3079 
3080 	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3081 		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3082 		bucket, src, srcp, dest, destp, sp->sk_state,
3083 		sk_wmem_alloc_get(sp),
3084 		udp_rqueue_get(sp),
3085 		0, 0L, 0,
3086 		from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3087 		0, sock_i_ino(sp),
3088 		refcount_read(&sp->sk_refcnt), sp,
3089 		atomic_read(&sp->sk_drops));
3090 }
3091 
3092 int udp4_seq_show(struct seq_file *seq, void *v)
3093 {
3094 	seq_setwidth(seq, 127);
3095 	if (v == SEQ_START_TOKEN)
3096 		seq_puts(seq, "   sl  local_address rem_address   st tx_queue "
3097 			   "rx_queue tr tm->when retrnsmt   uid  timeout "
3098 			   "inode ref pointer drops");
3099 	else {
3100 		struct udp_iter_state *state = seq->private;
3101 
3102 		udp4_format_sock(v, seq, state->bucket);
3103 	}
3104 	seq_pad(seq, '\n');
3105 	return 0;
3106 }
3107 
3108 #ifdef CONFIG_BPF_SYSCALL
3109 struct bpf_iter__udp {
3110 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
3111 	__bpf_md_ptr(struct udp_sock *, udp_sk);
3112 	uid_t uid __aligned(8);
3113 	int bucket __aligned(8);
3114 };
3115 
3116 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3117 			     struct udp_sock *udp_sk, uid_t uid, int bucket)
3118 {
3119 	struct bpf_iter__udp ctx;
3120 
3121 	meta->seq_num--;  /* skip SEQ_START_TOKEN */
3122 	ctx.meta = meta;
3123 	ctx.udp_sk = udp_sk;
3124 	ctx.uid = uid;
3125 	ctx.bucket = bucket;
3126 	return bpf_iter_run_prog(prog, &ctx);
3127 }
3128 
3129 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3130 {
3131 	struct udp_iter_state *state = seq->private;
3132 	struct bpf_iter_meta meta;
3133 	struct bpf_prog *prog;
3134 	struct sock *sk = v;
3135 	uid_t uid;
3136 
3137 	if (v == SEQ_START_TOKEN)
3138 		return 0;
3139 
3140 	uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3141 	meta.seq = seq;
3142 	prog = bpf_iter_get_info(&meta, false);
3143 	return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3144 }
3145 
3146 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3147 {
3148 	struct bpf_iter_meta meta;
3149 	struct bpf_prog *prog;
3150 
3151 	if (!v) {
3152 		meta.seq = seq;
3153 		prog = bpf_iter_get_info(&meta, true);
3154 		if (prog)
3155 			(void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3156 	}
3157 
3158 	udp_seq_stop(seq, v);
3159 }
3160 
3161 static const struct seq_operations bpf_iter_udp_seq_ops = {
3162 	.start		= udp_seq_start,
3163 	.next		= udp_seq_next,
3164 	.stop		= bpf_iter_udp_seq_stop,
3165 	.show		= bpf_iter_udp_seq_show,
3166 };
3167 #endif
3168 
3169 const struct seq_operations udp_seq_ops = {
3170 	.start		= udp_seq_start,
3171 	.next		= udp_seq_next,
3172 	.stop		= udp_seq_stop,
3173 	.show		= udp4_seq_show,
3174 };
3175 EXPORT_SYMBOL(udp_seq_ops);
3176 
3177 static struct udp_seq_afinfo udp4_seq_afinfo = {
3178 	.family		= AF_INET,
3179 	.udp_table	= &udp_table,
3180 };
3181 
3182 static int __net_init udp4_proc_init_net(struct net *net)
3183 {
3184 	if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3185 			sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3186 		return -ENOMEM;
3187 	return 0;
3188 }
3189 
3190 static void __net_exit udp4_proc_exit_net(struct net *net)
3191 {
3192 	remove_proc_entry("udp", net->proc_net);
3193 }
3194 
3195 static struct pernet_operations udp4_net_ops = {
3196 	.init = udp4_proc_init_net,
3197 	.exit = udp4_proc_exit_net,
3198 };
3199 
3200 int __init udp4_proc_init(void)
3201 {
3202 	return register_pernet_subsys(&udp4_net_ops);
3203 }
3204 
3205 void udp4_proc_exit(void)
3206 {
3207 	unregister_pernet_subsys(&udp4_net_ops);
3208 }
3209 #endif /* CONFIG_PROC_FS */
3210 
3211 static __initdata unsigned long uhash_entries;
3212 static int __init set_uhash_entries(char *str)
3213 {
3214 	ssize_t ret;
3215 
3216 	if (!str)
3217 		return 0;
3218 
3219 	ret = kstrtoul(str, 0, &uhash_entries);
3220 	if (ret)
3221 		return 0;
3222 
3223 	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3224 		uhash_entries = UDP_HTABLE_SIZE_MIN;
3225 	return 1;
3226 }
3227 __setup("uhash_entries=", set_uhash_entries);
3228 
3229 void __init udp_table_init(struct udp_table *table, const char *name)
3230 {
3231 	unsigned int i;
3232 
3233 	table->hash = alloc_large_system_hash(name,
3234 					      2 * sizeof(struct udp_hslot),
3235 					      uhash_entries,
3236 					      21, /* one slot per 2 MB */
3237 					      0,
3238 					      &table->log,
3239 					      &table->mask,
3240 					      UDP_HTABLE_SIZE_MIN,
3241 					      64 * 1024);
3242 
3243 	table->hash2 = table->hash + (table->mask + 1);
3244 	for (i = 0; i <= table->mask; i++) {
3245 		INIT_HLIST_HEAD(&table->hash[i].head);
3246 		table->hash[i].count = 0;
3247 		spin_lock_init(&table->hash[i].lock);
3248 	}
3249 	for (i = 0; i <= table->mask; i++) {
3250 		INIT_HLIST_HEAD(&table->hash2[i].head);
3251 		table->hash2[i].count = 0;
3252 		spin_lock_init(&table->hash2[i].lock);
3253 	}
3254 }
3255 
3256 u32 udp_flow_hashrnd(void)
3257 {
3258 	static u32 hashrnd __read_mostly;
3259 
3260 	net_get_random_once(&hashrnd, sizeof(hashrnd));
3261 
3262 	return hashrnd;
3263 }
3264 EXPORT_SYMBOL(udp_flow_hashrnd);
3265 
3266 static void __udp_sysctl_init(struct net *net)
3267 {
3268 	net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3269 	net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3270 
3271 #ifdef CONFIG_NET_L3_MASTER_DEV
3272 	net->ipv4.sysctl_udp_l3mdev_accept = 0;
3273 #endif
3274 }
3275 
3276 static int __net_init udp_sysctl_init(struct net *net)
3277 {
3278 	__udp_sysctl_init(net);
3279 	return 0;
3280 }
3281 
3282 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3283 	.init	= udp_sysctl_init,
3284 };
3285 
3286 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3287 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3288 		     struct udp_sock *udp_sk, uid_t uid, int bucket)
3289 
3290 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3291 {
3292 	struct udp_iter_state *st = priv_data;
3293 	struct udp_seq_afinfo *afinfo;
3294 	int ret;
3295 
3296 	afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3297 	if (!afinfo)
3298 		return -ENOMEM;
3299 
3300 	afinfo->family = AF_UNSPEC;
3301 	afinfo->udp_table = &udp_table;
3302 	st->bpf_seq_afinfo = afinfo;
3303 	ret = bpf_iter_init_seq_net(priv_data, aux);
3304 	if (ret)
3305 		kfree(afinfo);
3306 	return ret;
3307 }
3308 
3309 static void bpf_iter_fini_udp(void *priv_data)
3310 {
3311 	struct udp_iter_state *st = priv_data;
3312 
3313 	kfree(st->bpf_seq_afinfo);
3314 	bpf_iter_fini_seq_net(priv_data);
3315 }
3316 
3317 static const struct bpf_iter_seq_info udp_seq_info = {
3318 	.seq_ops		= &bpf_iter_udp_seq_ops,
3319 	.init_seq_private	= bpf_iter_init_udp,
3320 	.fini_seq_private	= bpf_iter_fini_udp,
3321 	.seq_priv_size		= sizeof(struct udp_iter_state),
3322 };
3323 
3324 static struct bpf_iter_reg udp_reg_info = {
3325 	.target			= "udp",
3326 	.ctx_arg_info_size	= 1,
3327 	.ctx_arg_info		= {
3328 		{ offsetof(struct bpf_iter__udp, udp_sk),
3329 		  PTR_TO_BTF_ID_OR_NULL },
3330 	},
3331 	.seq_info		= &udp_seq_info,
3332 };
3333 
3334 static void __init bpf_iter_register(void)
3335 {
3336 	udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3337 	if (bpf_iter_reg_target(&udp_reg_info))
3338 		pr_warn("Warning: could not register bpf iterator udp\n");
3339 }
3340 #endif
3341 
3342 void __init udp_init(void)
3343 {
3344 	unsigned long limit;
3345 	unsigned int i;
3346 
3347 	udp_table_init(&udp_table, "UDP");
3348 	limit = nr_free_buffer_pages() / 8;
3349 	limit = max(limit, 128UL);
3350 	sysctl_udp_mem[0] = limit / 4 * 3;
3351 	sysctl_udp_mem[1] = limit;
3352 	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3353 
3354 	__udp_sysctl_init(&init_net);
3355 
3356 	/* 16 spinlocks per cpu */
3357 	udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3358 	udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3359 				GFP_KERNEL);
3360 	if (!udp_busylocks)
3361 		panic("UDP: failed to alloc udp_busylocks\n");
3362 	for (i = 0; i < (1U << udp_busylocks_log); i++)
3363 		spin_lock_init(udp_busylocks + i);
3364 
3365 	if (register_pernet_subsys(&udp_sysctl_ops))
3366 		panic("UDP: failed to init sysctl parameters.\n");
3367 
3368 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3369 	bpf_iter_register();
3370 #endif
3371 }
3372