xref: /openbmc/linux/net/ipv4/udp.c (revision b830f94f)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		The User Datagram Protocol (UDP).
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
12  *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
13  *		Hirokazu Takahashi, <taka@valinux.co.jp>
14  *
15  * Fixes:
16  *		Alan Cox	:	verify_area() calls
17  *		Alan Cox	: 	stopped close while in use off icmp
18  *					messages. Not a fix but a botch that
19  *					for udp at least is 'valid'.
20  *		Alan Cox	:	Fixed icmp handling properly
21  *		Alan Cox	: 	Correct error for oversized datagrams
22  *		Alan Cox	:	Tidied select() semantics.
23  *		Alan Cox	:	udp_err() fixed properly, also now
24  *					select and read wake correctly on errors
25  *		Alan Cox	:	udp_send verify_area moved to avoid mem leak
26  *		Alan Cox	:	UDP can count its memory
27  *		Alan Cox	:	send to an unknown connection causes
28  *					an ECONNREFUSED off the icmp, but
29  *					does NOT close.
30  *		Alan Cox	:	Switched to new sk_buff handlers. No more backlog!
31  *		Alan Cox	:	Using generic datagram code. Even smaller and the PEEK
32  *					bug no longer crashes it.
33  *		Fred Van Kempen	: 	Net2e support for sk->broadcast.
34  *		Alan Cox	:	Uses skb_free_datagram
35  *		Alan Cox	:	Added get/set sockopt support.
36  *		Alan Cox	:	Broadcasting without option set returns EACCES.
37  *		Alan Cox	:	No wakeup calls. Instead we now use the callbacks.
38  *		Alan Cox	:	Use ip_tos and ip_ttl
39  *		Alan Cox	:	SNMP Mibs
40  *		Alan Cox	:	MSG_DONTROUTE, and 0.0.0.0 support.
41  *		Matt Dillon	:	UDP length checks.
42  *		Alan Cox	:	Smarter af_inet used properly.
43  *		Alan Cox	:	Use new kernel side addressing.
44  *		Alan Cox	:	Incorrect return on truncated datagram receive.
45  *	Arnt Gulbrandsen 	:	New udp_send and stuff
46  *		Alan Cox	:	Cache last socket
47  *		Alan Cox	:	Route cache
48  *		Jon Peatfield	:	Minor efficiency fix to sendto().
49  *		Mike Shaver	:	RFC1122 checks.
50  *		Alan Cox	:	Nonblocking error fix.
51  *	Willy Konynenberg	:	Transparent proxying support.
52  *		Mike McLagan	:	Routing by source
53  *		David S. Miller	:	New socket lookup architecture.
54  *					Last socket cache retained as it
55  *					does have a high hit rate.
56  *		Olaf Kirch	:	Don't linearise iovec on sendmsg.
57  *		Andi Kleen	:	Some cleanups, cache destination entry
58  *					for connect.
59  *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
60  *		Melvin Smith	:	Check msg_name not msg_namelen in sendto(),
61  *					return ENOTCONN for unconnected sockets (POSIX)
62  *		Janos Farkas	:	don't deliver multi/broadcasts to a different
63  *					bound-to-device socket
64  *	Hirokazu Takahashi	:	HW checksumming for outgoing UDP
65  *					datagrams.
66  *	Hirokazu Takahashi	:	sendfile() on UDP works now.
67  *		Arnaldo C. Melo :	convert /proc/net/udp to seq_file
68  *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
69  *	Alexey Kuznetsov:		allow both IPv4 and IPv6 sockets to bind
70  *					a single port at the same time.
71  *	Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
72  *	James Chapman		:	Add L2TP encapsulation type.
73  */
74 
75 #define pr_fmt(fmt) "UDP: " fmt
76 
77 #include <linux/uaccess.h>
78 #include <asm/ioctls.h>
79 #include <linux/memblock.h>
80 #include <linux/highmem.h>
81 #include <linux/swap.h>
82 #include <linux/types.h>
83 #include <linux/fcntl.h>
84 #include <linux/module.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/igmp.h>
88 #include <linux/inetdevice.h>
89 #include <linux/in.h>
90 #include <linux/errno.h>
91 #include <linux/timer.h>
92 #include <linux/mm.h>
93 #include <linux/inet.h>
94 #include <linux/netdevice.h>
95 #include <linux/slab.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/net_namespace.h>
101 #include <net/icmp.h>
102 #include <net/inet_hashtables.h>
103 #include <net/ip_tunnels.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <trace/events/udp.h>
108 #include <linux/static_key.h>
109 #include <trace/events/skb.h>
110 #include <net/busy_poll.h>
111 #include "udp_impl.h"
112 #include <net/sock_reuseport.h>
113 #include <net/addrconf.h>
114 #include <net/udp_tunnel.h>
115 
116 struct udp_table udp_table __read_mostly;
117 EXPORT_SYMBOL(udp_table);
118 
119 long sysctl_udp_mem[3] __read_mostly;
120 EXPORT_SYMBOL(sysctl_udp_mem);
121 
122 atomic_long_t udp_memory_allocated;
123 EXPORT_SYMBOL(udp_memory_allocated);
124 
125 #define MAX_UDP_PORTS 65536
126 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
127 
128 static int udp_lib_lport_inuse(struct net *net, __u16 num,
129 			       const struct udp_hslot *hslot,
130 			       unsigned long *bitmap,
131 			       struct sock *sk, unsigned int log)
132 {
133 	struct sock *sk2;
134 	kuid_t uid = sock_i_uid(sk);
135 
136 	sk_for_each(sk2, &hslot->head) {
137 		if (net_eq(sock_net(sk2), net) &&
138 		    sk2 != sk &&
139 		    (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
140 		    (!sk2->sk_reuse || !sk->sk_reuse) &&
141 		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
142 		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
143 		    inet_rcv_saddr_equal(sk, sk2, true)) {
144 			if (sk2->sk_reuseport && sk->sk_reuseport &&
145 			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
146 			    uid_eq(uid, sock_i_uid(sk2))) {
147 				if (!bitmap)
148 					return 0;
149 			} else {
150 				if (!bitmap)
151 					return 1;
152 				__set_bit(udp_sk(sk2)->udp_port_hash >> log,
153 					  bitmap);
154 			}
155 		}
156 	}
157 	return 0;
158 }
159 
160 /*
161  * Note: we still hold spinlock of primary hash chain, so no other writer
162  * can insert/delete a socket with local_port == num
163  */
164 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
165 				struct udp_hslot *hslot2,
166 				struct sock *sk)
167 {
168 	struct sock *sk2;
169 	kuid_t uid = sock_i_uid(sk);
170 	int res = 0;
171 
172 	spin_lock(&hslot2->lock);
173 	udp_portaddr_for_each_entry(sk2, &hslot2->head) {
174 		if (net_eq(sock_net(sk2), net) &&
175 		    sk2 != sk &&
176 		    (udp_sk(sk2)->udp_port_hash == num) &&
177 		    (!sk2->sk_reuse || !sk->sk_reuse) &&
178 		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
179 		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
180 		    inet_rcv_saddr_equal(sk, sk2, true)) {
181 			if (sk2->sk_reuseport && sk->sk_reuseport &&
182 			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
183 			    uid_eq(uid, sock_i_uid(sk2))) {
184 				res = 0;
185 			} else {
186 				res = 1;
187 			}
188 			break;
189 		}
190 	}
191 	spin_unlock(&hslot2->lock);
192 	return res;
193 }
194 
195 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
196 {
197 	struct net *net = sock_net(sk);
198 	kuid_t uid = sock_i_uid(sk);
199 	struct sock *sk2;
200 
201 	sk_for_each(sk2, &hslot->head) {
202 		if (net_eq(sock_net(sk2), net) &&
203 		    sk2 != sk &&
204 		    sk2->sk_family == sk->sk_family &&
205 		    ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
206 		    (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
207 		    (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
208 		    sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
209 		    inet_rcv_saddr_equal(sk, sk2, false)) {
210 			return reuseport_add_sock(sk, sk2,
211 						  inet_rcv_saddr_any(sk));
212 		}
213 	}
214 
215 	return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
216 }
217 
218 /**
219  *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
220  *
221  *  @sk:          socket struct in question
222  *  @snum:        port number to look up
223  *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
224  *                   with NULL address
225  */
226 int udp_lib_get_port(struct sock *sk, unsigned short snum,
227 		     unsigned int hash2_nulladdr)
228 {
229 	struct udp_hslot *hslot, *hslot2;
230 	struct udp_table *udptable = sk->sk_prot->h.udp_table;
231 	int    error = 1;
232 	struct net *net = sock_net(sk);
233 
234 	if (!snum) {
235 		int low, high, remaining;
236 		unsigned int rand;
237 		unsigned short first, last;
238 		DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
239 
240 		inet_get_local_port_range(net, &low, &high);
241 		remaining = (high - low) + 1;
242 
243 		rand = prandom_u32();
244 		first = reciprocal_scale(rand, remaining) + low;
245 		/*
246 		 * force rand to be an odd multiple of UDP_HTABLE_SIZE
247 		 */
248 		rand = (rand | 1) * (udptable->mask + 1);
249 		last = first + udptable->mask + 1;
250 		do {
251 			hslot = udp_hashslot(udptable, net, first);
252 			bitmap_zero(bitmap, PORTS_PER_CHAIN);
253 			spin_lock_bh(&hslot->lock);
254 			udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
255 					    udptable->log);
256 
257 			snum = first;
258 			/*
259 			 * Iterate on all possible values of snum for this hash.
260 			 * Using steps of an odd multiple of UDP_HTABLE_SIZE
261 			 * give us randomization and full range coverage.
262 			 */
263 			do {
264 				if (low <= snum && snum <= high &&
265 				    !test_bit(snum >> udptable->log, bitmap) &&
266 				    !inet_is_local_reserved_port(net, snum))
267 					goto found;
268 				snum += rand;
269 			} while (snum != first);
270 			spin_unlock_bh(&hslot->lock);
271 			cond_resched();
272 		} while (++first != last);
273 		goto fail;
274 	} else {
275 		hslot = udp_hashslot(udptable, net, snum);
276 		spin_lock_bh(&hslot->lock);
277 		if (hslot->count > 10) {
278 			int exist;
279 			unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
280 
281 			slot2          &= udptable->mask;
282 			hash2_nulladdr &= udptable->mask;
283 
284 			hslot2 = udp_hashslot2(udptable, slot2);
285 			if (hslot->count < hslot2->count)
286 				goto scan_primary_hash;
287 
288 			exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
289 			if (!exist && (hash2_nulladdr != slot2)) {
290 				hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
291 				exist = udp_lib_lport_inuse2(net, snum, hslot2,
292 							     sk);
293 			}
294 			if (exist)
295 				goto fail_unlock;
296 			else
297 				goto found;
298 		}
299 scan_primary_hash:
300 		if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
301 			goto fail_unlock;
302 	}
303 found:
304 	inet_sk(sk)->inet_num = snum;
305 	udp_sk(sk)->udp_port_hash = snum;
306 	udp_sk(sk)->udp_portaddr_hash ^= snum;
307 	if (sk_unhashed(sk)) {
308 		if (sk->sk_reuseport &&
309 		    udp_reuseport_add_sock(sk, hslot)) {
310 			inet_sk(sk)->inet_num = 0;
311 			udp_sk(sk)->udp_port_hash = 0;
312 			udp_sk(sk)->udp_portaddr_hash ^= snum;
313 			goto fail_unlock;
314 		}
315 
316 		sk_add_node_rcu(sk, &hslot->head);
317 		hslot->count++;
318 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
319 
320 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
321 		spin_lock(&hslot2->lock);
322 		if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
323 		    sk->sk_family == AF_INET6)
324 			hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
325 					   &hslot2->head);
326 		else
327 			hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
328 					   &hslot2->head);
329 		hslot2->count++;
330 		spin_unlock(&hslot2->lock);
331 	}
332 	sock_set_flag(sk, SOCK_RCU_FREE);
333 	error = 0;
334 fail_unlock:
335 	spin_unlock_bh(&hslot->lock);
336 fail:
337 	return error;
338 }
339 EXPORT_SYMBOL(udp_lib_get_port);
340 
341 int udp_v4_get_port(struct sock *sk, unsigned short snum)
342 {
343 	unsigned int hash2_nulladdr =
344 		ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
345 	unsigned int hash2_partial =
346 		ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
347 
348 	/* precompute partial secondary hash */
349 	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
350 	return udp_lib_get_port(sk, snum, hash2_nulladdr);
351 }
352 
353 static int compute_score(struct sock *sk, struct net *net,
354 			 __be32 saddr, __be16 sport,
355 			 __be32 daddr, unsigned short hnum,
356 			 int dif, int sdif)
357 {
358 	int score;
359 	struct inet_sock *inet;
360 	bool dev_match;
361 
362 	if (!net_eq(sock_net(sk), net) ||
363 	    udp_sk(sk)->udp_port_hash != hnum ||
364 	    ipv6_only_sock(sk))
365 		return -1;
366 
367 	if (sk->sk_rcv_saddr != daddr)
368 		return -1;
369 
370 	score = (sk->sk_family == PF_INET) ? 2 : 1;
371 
372 	inet = inet_sk(sk);
373 	if (inet->inet_daddr) {
374 		if (inet->inet_daddr != saddr)
375 			return -1;
376 		score += 4;
377 	}
378 
379 	if (inet->inet_dport) {
380 		if (inet->inet_dport != sport)
381 			return -1;
382 		score += 4;
383 	}
384 
385 	dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
386 					dif, sdif);
387 	if (!dev_match)
388 		return -1;
389 	score += 4;
390 
391 	if (sk->sk_incoming_cpu == raw_smp_processor_id())
392 		score++;
393 	return score;
394 }
395 
396 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
397 		       const __u16 lport, const __be32 faddr,
398 		       const __be16 fport)
399 {
400 	static u32 udp_ehash_secret __read_mostly;
401 
402 	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
403 
404 	return __inet_ehashfn(laddr, lport, faddr, fport,
405 			      udp_ehash_secret + net_hash_mix(net));
406 }
407 
408 /* called with rcu_read_lock() */
409 static struct sock *udp4_lib_lookup2(struct net *net,
410 				     __be32 saddr, __be16 sport,
411 				     __be32 daddr, unsigned int hnum,
412 				     int dif, int sdif,
413 				     struct udp_hslot *hslot2,
414 				     struct sk_buff *skb)
415 {
416 	struct sock *sk, *result;
417 	int score, badness;
418 	u32 hash = 0;
419 
420 	result = NULL;
421 	badness = 0;
422 	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
423 		score = compute_score(sk, net, saddr, sport,
424 				      daddr, hnum, dif, sdif);
425 		if (score > badness) {
426 			if (sk->sk_reuseport) {
427 				hash = udp_ehashfn(net, daddr, hnum,
428 						   saddr, sport);
429 				result = reuseport_select_sock(sk, hash, skb,
430 							sizeof(struct udphdr));
431 				if (result)
432 					return result;
433 			}
434 			badness = score;
435 			result = sk;
436 		}
437 	}
438 	return result;
439 }
440 
441 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
442  * harder than this. -DaveM
443  */
444 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
445 		__be16 sport, __be32 daddr, __be16 dport, int dif,
446 		int sdif, struct udp_table *udptable, struct sk_buff *skb)
447 {
448 	struct sock *result;
449 	unsigned short hnum = ntohs(dport);
450 	unsigned int hash2, slot2;
451 	struct udp_hslot *hslot2;
452 
453 	hash2 = ipv4_portaddr_hash(net, daddr, hnum);
454 	slot2 = hash2 & udptable->mask;
455 	hslot2 = &udptable->hash2[slot2];
456 
457 	result = udp4_lib_lookup2(net, saddr, sport,
458 				  daddr, hnum, dif, sdif,
459 				  hslot2, skb);
460 	if (!result) {
461 		hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
462 		slot2 = hash2 & udptable->mask;
463 		hslot2 = &udptable->hash2[slot2];
464 
465 		result = udp4_lib_lookup2(net, saddr, sport,
466 					  htonl(INADDR_ANY), hnum, dif, sdif,
467 					  hslot2, skb);
468 	}
469 	if (IS_ERR(result))
470 		return NULL;
471 	return result;
472 }
473 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
474 
475 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
476 						 __be16 sport, __be16 dport,
477 						 struct udp_table *udptable)
478 {
479 	const struct iphdr *iph = ip_hdr(skb);
480 
481 	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
482 				 iph->daddr, dport, inet_iif(skb),
483 				 inet_sdif(skb), udptable, skb);
484 }
485 
486 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
487 				 __be16 sport, __be16 dport)
488 {
489 	const struct iphdr *iph = ip_hdr(skb);
490 
491 	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
492 				 iph->daddr, dport, inet_iif(skb),
493 				 inet_sdif(skb), &udp_table, NULL);
494 }
495 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
496 
497 /* Must be called under rcu_read_lock().
498  * Does increment socket refcount.
499  */
500 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
501 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
502 			     __be32 daddr, __be16 dport, int dif)
503 {
504 	struct sock *sk;
505 
506 	sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
507 			       dif, 0, &udp_table, NULL);
508 	if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
509 		sk = NULL;
510 	return sk;
511 }
512 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
513 #endif
514 
515 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
516 				       __be16 loc_port, __be32 loc_addr,
517 				       __be16 rmt_port, __be32 rmt_addr,
518 				       int dif, int sdif, unsigned short hnum)
519 {
520 	struct inet_sock *inet = inet_sk(sk);
521 
522 	if (!net_eq(sock_net(sk), net) ||
523 	    udp_sk(sk)->udp_port_hash != hnum ||
524 	    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
525 	    (inet->inet_dport != rmt_port && inet->inet_dport) ||
526 	    (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
527 	    ipv6_only_sock(sk) ||
528 	    !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
529 		return false;
530 	if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
531 		return false;
532 	return true;
533 }
534 
535 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
536 void udp_encap_enable(void)
537 {
538 	static_branch_inc(&udp_encap_needed_key);
539 }
540 EXPORT_SYMBOL(udp_encap_enable);
541 
542 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
543  * through error handlers in encapsulations looking for a match.
544  */
545 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
546 {
547 	int i;
548 
549 	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
550 		int (*handler)(struct sk_buff *skb, u32 info);
551 		const struct ip_tunnel_encap_ops *encap;
552 
553 		encap = rcu_dereference(iptun_encaps[i]);
554 		if (!encap)
555 			continue;
556 		handler = encap->err_handler;
557 		if (handler && !handler(skb, info))
558 			return 0;
559 	}
560 
561 	return -ENOENT;
562 }
563 
564 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
565  * reversing source and destination port: this will match tunnels that force the
566  * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
567  * lwtunnels might actually break this assumption by being configured with
568  * different destination ports on endpoints, in this case we won't be able to
569  * trace ICMP messages back to them.
570  *
571  * If this doesn't match any socket, probe tunnels with arbitrary destination
572  * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
573  * we've sent packets to won't necessarily match the local destination port.
574  *
575  * Then ask the tunnel implementation to match the error against a valid
576  * association.
577  *
578  * Return an error if we can't find a match, the socket if we need further
579  * processing, zero otherwise.
580  */
581 static struct sock *__udp4_lib_err_encap(struct net *net,
582 					 const struct iphdr *iph,
583 					 struct udphdr *uh,
584 					 struct udp_table *udptable,
585 					 struct sk_buff *skb, u32 info)
586 {
587 	int network_offset, transport_offset;
588 	struct sock *sk;
589 
590 	network_offset = skb_network_offset(skb);
591 	transport_offset = skb_transport_offset(skb);
592 
593 	/* Network header needs to point to the outer IPv4 header inside ICMP */
594 	skb_reset_network_header(skb);
595 
596 	/* Transport header needs to point to the UDP header */
597 	skb_set_transport_header(skb, iph->ihl << 2);
598 
599 	sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
600 			       iph->saddr, uh->dest, skb->dev->ifindex, 0,
601 			       udptable, NULL);
602 	if (sk) {
603 		int (*lookup)(struct sock *sk, struct sk_buff *skb);
604 		struct udp_sock *up = udp_sk(sk);
605 
606 		lookup = READ_ONCE(up->encap_err_lookup);
607 		if (!lookup || lookup(sk, skb))
608 			sk = NULL;
609 	}
610 
611 	if (!sk)
612 		sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
613 
614 	skb_set_transport_header(skb, transport_offset);
615 	skb_set_network_header(skb, network_offset);
616 
617 	return sk;
618 }
619 
620 /*
621  * This routine is called by the ICMP module when it gets some
622  * sort of error condition.  If err < 0 then the socket should
623  * be closed and the error returned to the user.  If err > 0
624  * it's just the icmp type << 8 | icmp code.
625  * Header points to the ip header of the error packet. We move
626  * on past this. Then (as it used to claim before adjustment)
627  * header points to the first 8 bytes of the udp header.  We need
628  * to find the appropriate port.
629  */
630 
631 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
632 {
633 	struct inet_sock *inet;
634 	const struct iphdr *iph = (const struct iphdr *)skb->data;
635 	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
636 	const int type = icmp_hdr(skb)->type;
637 	const int code = icmp_hdr(skb)->code;
638 	bool tunnel = false;
639 	struct sock *sk;
640 	int harderr;
641 	int err;
642 	struct net *net = dev_net(skb->dev);
643 
644 	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
645 			       iph->saddr, uh->source, skb->dev->ifindex,
646 			       inet_sdif(skb), udptable, NULL);
647 	if (!sk) {
648 		/* No socket for error: try tunnels before discarding */
649 		sk = ERR_PTR(-ENOENT);
650 		if (static_branch_unlikely(&udp_encap_needed_key)) {
651 			sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
652 						  info);
653 			if (!sk)
654 				return 0;
655 		}
656 
657 		if (IS_ERR(sk)) {
658 			__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
659 			return PTR_ERR(sk);
660 		}
661 
662 		tunnel = true;
663 	}
664 
665 	err = 0;
666 	harderr = 0;
667 	inet = inet_sk(sk);
668 
669 	switch (type) {
670 	default:
671 	case ICMP_TIME_EXCEEDED:
672 		err = EHOSTUNREACH;
673 		break;
674 	case ICMP_SOURCE_QUENCH:
675 		goto out;
676 	case ICMP_PARAMETERPROB:
677 		err = EPROTO;
678 		harderr = 1;
679 		break;
680 	case ICMP_DEST_UNREACH:
681 		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
682 			ipv4_sk_update_pmtu(skb, sk, info);
683 			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
684 				err = EMSGSIZE;
685 				harderr = 1;
686 				break;
687 			}
688 			goto out;
689 		}
690 		err = EHOSTUNREACH;
691 		if (code <= NR_ICMP_UNREACH) {
692 			harderr = icmp_err_convert[code].fatal;
693 			err = icmp_err_convert[code].errno;
694 		}
695 		break;
696 	case ICMP_REDIRECT:
697 		ipv4_sk_redirect(skb, sk);
698 		goto out;
699 	}
700 
701 	/*
702 	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
703 	 *	4.1.3.3.
704 	 */
705 	if (tunnel) {
706 		/* ...not for tunnels though: we don't have a sending socket */
707 		goto out;
708 	}
709 	if (!inet->recverr) {
710 		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
711 			goto out;
712 	} else
713 		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
714 
715 	sk->sk_err = err;
716 	sk->sk_error_report(sk);
717 out:
718 	return 0;
719 }
720 
721 int udp_err(struct sk_buff *skb, u32 info)
722 {
723 	return __udp4_lib_err(skb, info, &udp_table);
724 }
725 
726 /*
727  * Throw away all pending data and cancel the corking. Socket is locked.
728  */
729 void udp_flush_pending_frames(struct sock *sk)
730 {
731 	struct udp_sock *up = udp_sk(sk);
732 
733 	if (up->pending) {
734 		up->len = 0;
735 		up->pending = 0;
736 		ip_flush_pending_frames(sk);
737 	}
738 }
739 EXPORT_SYMBOL(udp_flush_pending_frames);
740 
741 /**
742  * 	udp4_hwcsum  -  handle outgoing HW checksumming
743  * 	@skb: 	sk_buff containing the filled-in UDP header
744  * 	        (checksum field must be zeroed out)
745  *	@src:	source IP address
746  *	@dst:	destination IP address
747  */
748 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
749 {
750 	struct udphdr *uh = udp_hdr(skb);
751 	int offset = skb_transport_offset(skb);
752 	int len = skb->len - offset;
753 	int hlen = len;
754 	__wsum csum = 0;
755 
756 	if (!skb_has_frag_list(skb)) {
757 		/*
758 		 * Only one fragment on the socket.
759 		 */
760 		skb->csum_start = skb_transport_header(skb) - skb->head;
761 		skb->csum_offset = offsetof(struct udphdr, check);
762 		uh->check = ~csum_tcpudp_magic(src, dst, len,
763 					       IPPROTO_UDP, 0);
764 	} else {
765 		struct sk_buff *frags;
766 
767 		/*
768 		 * HW-checksum won't work as there are two or more
769 		 * fragments on the socket so that all csums of sk_buffs
770 		 * should be together
771 		 */
772 		skb_walk_frags(skb, frags) {
773 			csum = csum_add(csum, frags->csum);
774 			hlen -= frags->len;
775 		}
776 
777 		csum = skb_checksum(skb, offset, hlen, csum);
778 		skb->ip_summed = CHECKSUM_NONE;
779 
780 		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
781 		if (uh->check == 0)
782 			uh->check = CSUM_MANGLED_0;
783 	}
784 }
785 EXPORT_SYMBOL_GPL(udp4_hwcsum);
786 
787 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
788  * for the simple case like when setting the checksum for a UDP tunnel.
789  */
790 void udp_set_csum(bool nocheck, struct sk_buff *skb,
791 		  __be32 saddr, __be32 daddr, int len)
792 {
793 	struct udphdr *uh = udp_hdr(skb);
794 
795 	if (nocheck) {
796 		uh->check = 0;
797 	} else if (skb_is_gso(skb)) {
798 		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
799 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
800 		uh->check = 0;
801 		uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
802 		if (uh->check == 0)
803 			uh->check = CSUM_MANGLED_0;
804 	} else {
805 		skb->ip_summed = CHECKSUM_PARTIAL;
806 		skb->csum_start = skb_transport_header(skb) - skb->head;
807 		skb->csum_offset = offsetof(struct udphdr, check);
808 		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
809 	}
810 }
811 EXPORT_SYMBOL(udp_set_csum);
812 
813 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
814 			struct inet_cork *cork)
815 {
816 	struct sock *sk = skb->sk;
817 	struct inet_sock *inet = inet_sk(sk);
818 	struct udphdr *uh;
819 	int err = 0;
820 	int is_udplite = IS_UDPLITE(sk);
821 	int offset = skb_transport_offset(skb);
822 	int len = skb->len - offset;
823 	__wsum csum = 0;
824 
825 	/*
826 	 * Create a UDP header
827 	 */
828 	uh = udp_hdr(skb);
829 	uh->source = inet->inet_sport;
830 	uh->dest = fl4->fl4_dport;
831 	uh->len = htons(len);
832 	uh->check = 0;
833 
834 	if (cork->gso_size) {
835 		const int hlen = skb_network_header_len(skb) +
836 				 sizeof(struct udphdr);
837 
838 		if (hlen + cork->gso_size > cork->fragsize) {
839 			kfree_skb(skb);
840 			return -EINVAL;
841 		}
842 		if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
843 			kfree_skb(skb);
844 			return -EINVAL;
845 		}
846 		if (sk->sk_no_check_tx) {
847 			kfree_skb(skb);
848 			return -EINVAL;
849 		}
850 		if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
851 		    dst_xfrm(skb_dst(skb))) {
852 			kfree_skb(skb);
853 			return -EIO;
854 		}
855 
856 		skb_shinfo(skb)->gso_size = cork->gso_size;
857 		skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
858 		skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(len - sizeof(uh),
859 							 cork->gso_size);
860 		goto csum_partial;
861 	}
862 
863 	if (is_udplite)  				 /*     UDP-Lite      */
864 		csum = udplite_csum(skb);
865 
866 	else if (sk->sk_no_check_tx) {			 /* UDP csum off */
867 
868 		skb->ip_summed = CHECKSUM_NONE;
869 		goto send;
870 
871 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
872 csum_partial:
873 
874 		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
875 		goto send;
876 
877 	} else
878 		csum = udp_csum(skb);
879 
880 	/* add protocol-dependent pseudo-header */
881 	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
882 				      sk->sk_protocol, csum);
883 	if (uh->check == 0)
884 		uh->check = CSUM_MANGLED_0;
885 
886 send:
887 	err = ip_send_skb(sock_net(sk), skb);
888 	if (err) {
889 		if (err == -ENOBUFS && !inet->recverr) {
890 			UDP_INC_STATS(sock_net(sk),
891 				      UDP_MIB_SNDBUFERRORS, is_udplite);
892 			err = 0;
893 		}
894 	} else
895 		UDP_INC_STATS(sock_net(sk),
896 			      UDP_MIB_OUTDATAGRAMS, is_udplite);
897 	return err;
898 }
899 
900 /*
901  * Push out all pending data as one UDP datagram. Socket is locked.
902  */
903 int udp_push_pending_frames(struct sock *sk)
904 {
905 	struct udp_sock  *up = udp_sk(sk);
906 	struct inet_sock *inet = inet_sk(sk);
907 	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
908 	struct sk_buff *skb;
909 	int err = 0;
910 
911 	skb = ip_finish_skb(sk, fl4);
912 	if (!skb)
913 		goto out;
914 
915 	err = udp_send_skb(skb, fl4, &inet->cork.base);
916 
917 out:
918 	up->len = 0;
919 	up->pending = 0;
920 	return err;
921 }
922 EXPORT_SYMBOL(udp_push_pending_frames);
923 
924 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
925 {
926 	switch (cmsg->cmsg_type) {
927 	case UDP_SEGMENT:
928 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
929 			return -EINVAL;
930 		*gso_size = *(__u16 *)CMSG_DATA(cmsg);
931 		return 0;
932 	default:
933 		return -EINVAL;
934 	}
935 }
936 
937 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
938 {
939 	struct cmsghdr *cmsg;
940 	bool need_ip = false;
941 	int err;
942 
943 	for_each_cmsghdr(cmsg, msg) {
944 		if (!CMSG_OK(msg, cmsg))
945 			return -EINVAL;
946 
947 		if (cmsg->cmsg_level != SOL_UDP) {
948 			need_ip = true;
949 			continue;
950 		}
951 
952 		err = __udp_cmsg_send(cmsg, gso_size);
953 		if (err)
954 			return err;
955 	}
956 
957 	return need_ip;
958 }
959 EXPORT_SYMBOL_GPL(udp_cmsg_send);
960 
961 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
962 {
963 	struct inet_sock *inet = inet_sk(sk);
964 	struct udp_sock *up = udp_sk(sk);
965 	DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
966 	struct flowi4 fl4_stack;
967 	struct flowi4 *fl4;
968 	int ulen = len;
969 	struct ipcm_cookie ipc;
970 	struct rtable *rt = NULL;
971 	int free = 0;
972 	int connected = 0;
973 	__be32 daddr, faddr, saddr;
974 	__be16 dport;
975 	u8  tos;
976 	int err, is_udplite = IS_UDPLITE(sk);
977 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
978 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
979 	struct sk_buff *skb;
980 	struct ip_options_data opt_copy;
981 
982 	if (len > 0xFFFF)
983 		return -EMSGSIZE;
984 
985 	/*
986 	 *	Check the flags.
987 	 */
988 
989 	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
990 		return -EOPNOTSUPP;
991 
992 	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
993 
994 	fl4 = &inet->cork.fl.u.ip4;
995 	if (up->pending) {
996 		/*
997 		 * There are pending frames.
998 		 * The socket lock must be held while it's corked.
999 		 */
1000 		lock_sock(sk);
1001 		if (likely(up->pending)) {
1002 			if (unlikely(up->pending != AF_INET)) {
1003 				release_sock(sk);
1004 				return -EINVAL;
1005 			}
1006 			goto do_append_data;
1007 		}
1008 		release_sock(sk);
1009 	}
1010 	ulen += sizeof(struct udphdr);
1011 
1012 	/*
1013 	 *	Get and verify the address.
1014 	 */
1015 	if (usin) {
1016 		if (msg->msg_namelen < sizeof(*usin))
1017 			return -EINVAL;
1018 		if (usin->sin_family != AF_INET) {
1019 			if (usin->sin_family != AF_UNSPEC)
1020 				return -EAFNOSUPPORT;
1021 		}
1022 
1023 		daddr = usin->sin_addr.s_addr;
1024 		dport = usin->sin_port;
1025 		if (dport == 0)
1026 			return -EINVAL;
1027 	} else {
1028 		if (sk->sk_state != TCP_ESTABLISHED)
1029 			return -EDESTADDRREQ;
1030 		daddr = inet->inet_daddr;
1031 		dport = inet->inet_dport;
1032 		/* Open fast path for connected socket.
1033 		   Route will not be used, if at least one option is set.
1034 		 */
1035 		connected = 1;
1036 	}
1037 
1038 	ipcm_init_sk(&ipc, inet);
1039 	ipc.gso_size = up->gso_size;
1040 
1041 	if (msg->msg_controllen) {
1042 		err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1043 		if (err > 0)
1044 			err = ip_cmsg_send(sk, msg, &ipc,
1045 					   sk->sk_family == AF_INET6);
1046 		if (unlikely(err < 0)) {
1047 			kfree(ipc.opt);
1048 			return err;
1049 		}
1050 		if (ipc.opt)
1051 			free = 1;
1052 		connected = 0;
1053 	}
1054 	if (!ipc.opt) {
1055 		struct ip_options_rcu *inet_opt;
1056 
1057 		rcu_read_lock();
1058 		inet_opt = rcu_dereference(inet->inet_opt);
1059 		if (inet_opt) {
1060 			memcpy(&opt_copy, inet_opt,
1061 			       sizeof(*inet_opt) + inet_opt->opt.optlen);
1062 			ipc.opt = &opt_copy.opt;
1063 		}
1064 		rcu_read_unlock();
1065 	}
1066 
1067 	if (cgroup_bpf_enabled && !connected) {
1068 		err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1069 					    (struct sockaddr *)usin, &ipc.addr);
1070 		if (err)
1071 			goto out_free;
1072 		if (usin) {
1073 			if (usin->sin_port == 0) {
1074 				/* BPF program set invalid port. Reject it. */
1075 				err = -EINVAL;
1076 				goto out_free;
1077 			}
1078 			daddr = usin->sin_addr.s_addr;
1079 			dport = usin->sin_port;
1080 		}
1081 	}
1082 
1083 	saddr = ipc.addr;
1084 	ipc.addr = faddr = daddr;
1085 
1086 	if (ipc.opt && ipc.opt->opt.srr) {
1087 		if (!daddr) {
1088 			err = -EINVAL;
1089 			goto out_free;
1090 		}
1091 		faddr = ipc.opt->opt.faddr;
1092 		connected = 0;
1093 	}
1094 	tos = get_rttos(&ipc, inet);
1095 	if (sock_flag(sk, SOCK_LOCALROUTE) ||
1096 	    (msg->msg_flags & MSG_DONTROUTE) ||
1097 	    (ipc.opt && ipc.opt->opt.is_strictroute)) {
1098 		tos |= RTO_ONLINK;
1099 		connected = 0;
1100 	}
1101 
1102 	if (ipv4_is_multicast(daddr)) {
1103 		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1104 			ipc.oif = inet->mc_index;
1105 		if (!saddr)
1106 			saddr = inet->mc_addr;
1107 		connected = 0;
1108 	} else if (!ipc.oif) {
1109 		ipc.oif = inet->uc_index;
1110 	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1111 		/* oif is set, packet is to local broadcast and
1112 		 * and uc_index is set. oif is most likely set
1113 		 * by sk_bound_dev_if. If uc_index != oif check if the
1114 		 * oif is an L3 master and uc_index is an L3 slave.
1115 		 * If so, we want to allow the send using the uc_index.
1116 		 */
1117 		if (ipc.oif != inet->uc_index &&
1118 		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1119 							      inet->uc_index)) {
1120 			ipc.oif = inet->uc_index;
1121 		}
1122 	}
1123 
1124 	if (connected)
1125 		rt = (struct rtable *)sk_dst_check(sk, 0);
1126 
1127 	if (!rt) {
1128 		struct net *net = sock_net(sk);
1129 		__u8 flow_flags = inet_sk_flowi_flags(sk);
1130 
1131 		fl4 = &fl4_stack;
1132 
1133 		flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1134 				   RT_SCOPE_UNIVERSE, sk->sk_protocol,
1135 				   flow_flags,
1136 				   faddr, saddr, dport, inet->inet_sport,
1137 				   sk->sk_uid);
1138 
1139 		security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1140 		rt = ip_route_output_flow(net, fl4, sk);
1141 		if (IS_ERR(rt)) {
1142 			err = PTR_ERR(rt);
1143 			rt = NULL;
1144 			if (err == -ENETUNREACH)
1145 				IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1146 			goto out;
1147 		}
1148 
1149 		err = -EACCES;
1150 		if ((rt->rt_flags & RTCF_BROADCAST) &&
1151 		    !sock_flag(sk, SOCK_BROADCAST))
1152 			goto out;
1153 		if (connected)
1154 			sk_dst_set(sk, dst_clone(&rt->dst));
1155 	}
1156 
1157 	if (msg->msg_flags&MSG_CONFIRM)
1158 		goto do_confirm;
1159 back_from_confirm:
1160 
1161 	saddr = fl4->saddr;
1162 	if (!ipc.addr)
1163 		daddr = ipc.addr = fl4->daddr;
1164 
1165 	/* Lockless fast path for the non-corking case. */
1166 	if (!corkreq) {
1167 		struct inet_cork cork;
1168 
1169 		skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1170 				  sizeof(struct udphdr), &ipc, &rt,
1171 				  &cork, msg->msg_flags);
1172 		err = PTR_ERR(skb);
1173 		if (!IS_ERR_OR_NULL(skb))
1174 			err = udp_send_skb(skb, fl4, &cork);
1175 		goto out;
1176 	}
1177 
1178 	lock_sock(sk);
1179 	if (unlikely(up->pending)) {
1180 		/* The socket is already corked while preparing it. */
1181 		/* ... which is an evident application bug. --ANK */
1182 		release_sock(sk);
1183 
1184 		net_dbg_ratelimited("socket already corked\n");
1185 		err = -EINVAL;
1186 		goto out;
1187 	}
1188 	/*
1189 	 *	Now cork the socket to pend data.
1190 	 */
1191 	fl4 = &inet->cork.fl.u.ip4;
1192 	fl4->daddr = daddr;
1193 	fl4->saddr = saddr;
1194 	fl4->fl4_dport = dport;
1195 	fl4->fl4_sport = inet->inet_sport;
1196 	up->pending = AF_INET;
1197 
1198 do_append_data:
1199 	up->len += ulen;
1200 	err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1201 			     sizeof(struct udphdr), &ipc, &rt,
1202 			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1203 	if (err)
1204 		udp_flush_pending_frames(sk);
1205 	else if (!corkreq)
1206 		err = udp_push_pending_frames(sk);
1207 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1208 		up->pending = 0;
1209 	release_sock(sk);
1210 
1211 out:
1212 	ip_rt_put(rt);
1213 out_free:
1214 	if (free)
1215 		kfree(ipc.opt);
1216 	if (!err)
1217 		return len;
1218 	/*
1219 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1220 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1221 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1222 	 * things).  We could add another new stat but at least for now that
1223 	 * seems like overkill.
1224 	 */
1225 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1226 		UDP_INC_STATS(sock_net(sk),
1227 			      UDP_MIB_SNDBUFERRORS, is_udplite);
1228 	}
1229 	return err;
1230 
1231 do_confirm:
1232 	if (msg->msg_flags & MSG_PROBE)
1233 		dst_confirm_neigh(&rt->dst, &fl4->daddr);
1234 	if (!(msg->msg_flags&MSG_PROBE) || len)
1235 		goto back_from_confirm;
1236 	err = 0;
1237 	goto out;
1238 }
1239 EXPORT_SYMBOL(udp_sendmsg);
1240 
1241 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1242 		 size_t size, int flags)
1243 {
1244 	struct inet_sock *inet = inet_sk(sk);
1245 	struct udp_sock *up = udp_sk(sk);
1246 	int ret;
1247 
1248 	if (flags & MSG_SENDPAGE_NOTLAST)
1249 		flags |= MSG_MORE;
1250 
1251 	if (!up->pending) {
1252 		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
1253 
1254 		/* Call udp_sendmsg to specify destination address which
1255 		 * sendpage interface can't pass.
1256 		 * This will succeed only when the socket is connected.
1257 		 */
1258 		ret = udp_sendmsg(sk, &msg, 0);
1259 		if (ret < 0)
1260 			return ret;
1261 	}
1262 
1263 	lock_sock(sk);
1264 
1265 	if (unlikely(!up->pending)) {
1266 		release_sock(sk);
1267 
1268 		net_dbg_ratelimited("cork failed\n");
1269 		return -EINVAL;
1270 	}
1271 
1272 	ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1273 			     page, offset, size, flags);
1274 	if (ret == -EOPNOTSUPP) {
1275 		release_sock(sk);
1276 		return sock_no_sendpage(sk->sk_socket, page, offset,
1277 					size, flags);
1278 	}
1279 	if (ret < 0) {
1280 		udp_flush_pending_frames(sk);
1281 		goto out;
1282 	}
1283 
1284 	up->len += size;
1285 	if (!(up->corkflag || (flags&MSG_MORE)))
1286 		ret = udp_push_pending_frames(sk);
1287 	if (!ret)
1288 		ret = size;
1289 out:
1290 	release_sock(sk);
1291 	return ret;
1292 }
1293 
1294 #define UDP_SKB_IS_STATELESS 0x80000000
1295 
1296 static void udp_set_dev_scratch(struct sk_buff *skb)
1297 {
1298 	struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1299 
1300 	BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1301 	scratch->_tsize_state = skb->truesize;
1302 #if BITS_PER_LONG == 64
1303 	scratch->len = skb->len;
1304 	scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1305 	scratch->is_linear = !skb_is_nonlinear(skb);
1306 #endif
1307 	/* all head states execept sp (dst, sk, nf) are always cleared by
1308 	 * udp_rcv() and we need to preserve secpath, if present, to eventually
1309 	 * process IP_CMSG_PASSSEC at recvmsg() time
1310 	 */
1311 	if (likely(!skb_sec_path(skb)))
1312 		scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1313 }
1314 
1315 static int udp_skb_truesize(struct sk_buff *skb)
1316 {
1317 	return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1318 }
1319 
1320 static bool udp_skb_has_head_state(struct sk_buff *skb)
1321 {
1322 	return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1323 }
1324 
1325 /* fully reclaim rmem/fwd memory allocated for skb */
1326 static void udp_rmem_release(struct sock *sk, int size, int partial,
1327 			     bool rx_queue_lock_held)
1328 {
1329 	struct udp_sock *up = udp_sk(sk);
1330 	struct sk_buff_head *sk_queue;
1331 	int amt;
1332 
1333 	if (likely(partial)) {
1334 		up->forward_deficit += size;
1335 		size = up->forward_deficit;
1336 		if (size < (sk->sk_rcvbuf >> 2))
1337 			return;
1338 	} else {
1339 		size += up->forward_deficit;
1340 	}
1341 	up->forward_deficit = 0;
1342 
1343 	/* acquire the sk_receive_queue for fwd allocated memory scheduling,
1344 	 * if the called don't held it already
1345 	 */
1346 	sk_queue = &sk->sk_receive_queue;
1347 	if (!rx_queue_lock_held)
1348 		spin_lock(&sk_queue->lock);
1349 
1350 
1351 	sk->sk_forward_alloc += size;
1352 	amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1353 	sk->sk_forward_alloc -= amt;
1354 
1355 	if (amt)
1356 		__sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1357 
1358 	atomic_sub(size, &sk->sk_rmem_alloc);
1359 
1360 	/* this can save us from acquiring the rx queue lock on next receive */
1361 	skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1362 
1363 	if (!rx_queue_lock_held)
1364 		spin_unlock(&sk_queue->lock);
1365 }
1366 
1367 /* Note: called with reader_queue.lock held.
1368  * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1369  * This avoids a cache line miss while receive_queue lock is held.
1370  * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1371  */
1372 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1373 {
1374 	prefetch(&skb->data);
1375 	udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1376 }
1377 EXPORT_SYMBOL(udp_skb_destructor);
1378 
1379 /* as above, but the caller held the rx queue lock, too */
1380 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1381 {
1382 	prefetch(&skb->data);
1383 	udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1384 }
1385 
1386 /* Idea of busylocks is to let producers grab an extra spinlock
1387  * to relieve pressure on the receive_queue spinlock shared by consumer.
1388  * Under flood, this means that only one producer can be in line
1389  * trying to acquire the receive_queue spinlock.
1390  * These busylock can be allocated on a per cpu manner, instead of a
1391  * per socket one (that would consume a cache line per socket)
1392  */
1393 static int udp_busylocks_log __read_mostly;
1394 static spinlock_t *udp_busylocks __read_mostly;
1395 
1396 static spinlock_t *busylock_acquire(void *ptr)
1397 {
1398 	spinlock_t *busy;
1399 
1400 	busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1401 	spin_lock(busy);
1402 	return busy;
1403 }
1404 
1405 static void busylock_release(spinlock_t *busy)
1406 {
1407 	if (busy)
1408 		spin_unlock(busy);
1409 }
1410 
1411 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1412 {
1413 	struct sk_buff_head *list = &sk->sk_receive_queue;
1414 	int rmem, delta, amt, err = -ENOMEM;
1415 	spinlock_t *busy = NULL;
1416 	int size;
1417 
1418 	/* try to avoid the costly atomic add/sub pair when the receive
1419 	 * queue is full; always allow at least a packet
1420 	 */
1421 	rmem = atomic_read(&sk->sk_rmem_alloc);
1422 	if (rmem > sk->sk_rcvbuf)
1423 		goto drop;
1424 
1425 	/* Under mem pressure, it might be helpful to help udp_recvmsg()
1426 	 * having linear skbs :
1427 	 * - Reduce memory overhead and thus increase receive queue capacity
1428 	 * - Less cache line misses at copyout() time
1429 	 * - Less work at consume_skb() (less alien page frag freeing)
1430 	 */
1431 	if (rmem > (sk->sk_rcvbuf >> 1)) {
1432 		skb_condense(skb);
1433 
1434 		busy = busylock_acquire(sk);
1435 	}
1436 	size = skb->truesize;
1437 	udp_set_dev_scratch(skb);
1438 
1439 	/* we drop only if the receive buf is full and the receive
1440 	 * queue contains some other skb
1441 	 */
1442 	rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1443 	if (rmem > (size + sk->sk_rcvbuf))
1444 		goto uncharge_drop;
1445 
1446 	spin_lock(&list->lock);
1447 	if (size >= sk->sk_forward_alloc) {
1448 		amt = sk_mem_pages(size);
1449 		delta = amt << SK_MEM_QUANTUM_SHIFT;
1450 		if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1451 			err = -ENOBUFS;
1452 			spin_unlock(&list->lock);
1453 			goto uncharge_drop;
1454 		}
1455 
1456 		sk->sk_forward_alloc += delta;
1457 	}
1458 
1459 	sk->sk_forward_alloc -= size;
1460 
1461 	/* no need to setup a destructor, we will explicitly release the
1462 	 * forward allocated memory on dequeue
1463 	 */
1464 	sock_skb_set_dropcount(sk, skb);
1465 
1466 	__skb_queue_tail(list, skb);
1467 	spin_unlock(&list->lock);
1468 
1469 	if (!sock_flag(sk, SOCK_DEAD))
1470 		sk->sk_data_ready(sk);
1471 
1472 	busylock_release(busy);
1473 	return 0;
1474 
1475 uncharge_drop:
1476 	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1477 
1478 drop:
1479 	atomic_inc(&sk->sk_drops);
1480 	busylock_release(busy);
1481 	return err;
1482 }
1483 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1484 
1485 void udp_destruct_sock(struct sock *sk)
1486 {
1487 	/* reclaim completely the forward allocated memory */
1488 	struct udp_sock *up = udp_sk(sk);
1489 	unsigned int total = 0;
1490 	struct sk_buff *skb;
1491 
1492 	skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1493 	while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1494 		total += skb->truesize;
1495 		kfree_skb(skb);
1496 	}
1497 	udp_rmem_release(sk, total, 0, true);
1498 
1499 	inet_sock_destruct(sk);
1500 }
1501 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1502 
1503 int udp_init_sock(struct sock *sk)
1504 {
1505 	skb_queue_head_init(&udp_sk(sk)->reader_queue);
1506 	sk->sk_destruct = udp_destruct_sock;
1507 	return 0;
1508 }
1509 EXPORT_SYMBOL_GPL(udp_init_sock);
1510 
1511 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1512 {
1513 	if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1514 		bool slow = lock_sock_fast(sk);
1515 
1516 		sk_peek_offset_bwd(sk, len);
1517 		unlock_sock_fast(sk, slow);
1518 	}
1519 
1520 	if (!skb_unref(skb))
1521 		return;
1522 
1523 	/* In the more common cases we cleared the head states previously,
1524 	 * see __udp_queue_rcv_skb().
1525 	 */
1526 	if (unlikely(udp_skb_has_head_state(skb)))
1527 		skb_release_head_state(skb);
1528 	__consume_stateless_skb(skb);
1529 }
1530 EXPORT_SYMBOL_GPL(skb_consume_udp);
1531 
1532 static struct sk_buff *__first_packet_length(struct sock *sk,
1533 					     struct sk_buff_head *rcvq,
1534 					     int *total)
1535 {
1536 	struct sk_buff *skb;
1537 
1538 	while ((skb = skb_peek(rcvq)) != NULL) {
1539 		if (udp_lib_checksum_complete(skb)) {
1540 			__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1541 					IS_UDPLITE(sk));
1542 			__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1543 					IS_UDPLITE(sk));
1544 			atomic_inc(&sk->sk_drops);
1545 			__skb_unlink(skb, rcvq);
1546 			*total += skb->truesize;
1547 			kfree_skb(skb);
1548 		} else {
1549 			/* the csum related bits could be changed, refresh
1550 			 * the scratch area
1551 			 */
1552 			udp_set_dev_scratch(skb);
1553 			break;
1554 		}
1555 	}
1556 	return skb;
1557 }
1558 
1559 /**
1560  *	first_packet_length	- return length of first packet in receive queue
1561  *	@sk: socket
1562  *
1563  *	Drops all bad checksum frames, until a valid one is found.
1564  *	Returns the length of found skb, or -1 if none is found.
1565  */
1566 static int first_packet_length(struct sock *sk)
1567 {
1568 	struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1569 	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1570 	struct sk_buff *skb;
1571 	int total = 0;
1572 	int res;
1573 
1574 	spin_lock_bh(&rcvq->lock);
1575 	skb = __first_packet_length(sk, rcvq, &total);
1576 	if (!skb && !skb_queue_empty(sk_queue)) {
1577 		spin_lock(&sk_queue->lock);
1578 		skb_queue_splice_tail_init(sk_queue, rcvq);
1579 		spin_unlock(&sk_queue->lock);
1580 
1581 		skb = __first_packet_length(sk, rcvq, &total);
1582 	}
1583 	res = skb ? skb->len : -1;
1584 	if (total)
1585 		udp_rmem_release(sk, total, 1, false);
1586 	spin_unlock_bh(&rcvq->lock);
1587 	return res;
1588 }
1589 
1590 /*
1591  *	IOCTL requests applicable to the UDP protocol
1592  */
1593 
1594 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1595 {
1596 	switch (cmd) {
1597 	case SIOCOUTQ:
1598 	{
1599 		int amount = sk_wmem_alloc_get(sk);
1600 
1601 		return put_user(amount, (int __user *)arg);
1602 	}
1603 
1604 	case SIOCINQ:
1605 	{
1606 		int amount = max_t(int, 0, first_packet_length(sk));
1607 
1608 		return put_user(amount, (int __user *)arg);
1609 	}
1610 
1611 	default:
1612 		return -ENOIOCTLCMD;
1613 	}
1614 
1615 	return 0;
1616 }
1617 EXPORT_SYMBOL(udp_ioctl);
1618 
1619 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1620 			       int noblock, int *off, int *err)
1621 {
1622 	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1623 	struct sk_buff_head *queue;
1624 	struct sk_buff *last;
1625 	long timeo;
1626 	int error;
1627 
1628 	queue = &udp_sk(sk)->reader_queue;
1629 	flags |= noblock ? MSG_DONTWAIT : 0;
1630 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1631 	do {
1632 		struct sk_buff *skb;
1633 
1634 		error = sock_error(sk);
1635 		if (error)
1636 			break;
1637 
1638 		error = -EAGAIN;
1639 		do {
1640 			spin_lock_bh(&queue->lock);
1641 			skb = __skb_try_recv_from_queue(sk, queue, flags,
1642 							udp_skb_destructor,
1643 							off, err, &last);
1644 			if (skb) {
1645 				spin_unlock_bh(&queue->lock);
1646 				return skb;
1647 			}
1648 
1649 			if (skb_queue_empty(sk_queue)) {
1650 				spin_unlock_bh(&queue->lock);
1651 				goto busy_check;
1652 			}
1653 
1654 			/* refill the reader queue and walk it again
1655 			 * keep both queues locked to avoid re-acquiring
1656 			 * the sk_receive_queue lock if fwd memory scheduling
1657 			 * is needed.
1658 			 */
1659 			spin_lock(&sk_queue->lock);
1660 			skb_queue_splice_tail_init(sk_queue, queue);
1661 
1662 			skb = __skb_try_recv_from_queue(sk, queue, flags,
1663 							udp_skb_dtor_locked,
1664 							off, err, &last);
1665 			spin_unlock(&sk_queue->lock);
1666 			spin_unlock_bh(&queue->lock);
1667 			if (skb)
1668 				return skb;
1669 
1670 busy_check:
1671 			if (!sk_can_busy_loop(sk))
1672 				break;
1673 
1674 			sk_busy_loop(sk, flags & MSG_DONTWAIT);
1675 		} while (!skb_queue_empty(sk_queue));
1676 
1677 		/* sk_queue is empty, reader_queue may contain peeked packets */
1678 	} while (timeo &&
1679 		 !__skb_wait_for_more_packets(sk, &error, &timeo,
1680 					      (struct sk_buff *)sk_queue));
1681 
1682 	*err = error;
1683 	return NULL;
1684 }
1685 EXPORT_SYMBOL(__skb_recv_udp);
1686 
1687 /*
1688  * 	This should be easy, if there is something there we
1689  * 	return it, otherwise we block.
1690  */
1691 
1692 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1693 		int flags, int *addr_len)
1694 {
1695 	struct inet_sock *inet = inet_sk(sk);
1696 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1697 	struct sk_buff *skb;
1698 	unsigned int ulen, copied;
1699 	int off, err, peeking = flags & MSG_PEEK;
1700 	int is_udplite = IS_UDPLITE(sk);
1701 	bool checksum_valid = false;
1702 
1703 	if (flags & MSG_ERRQUEUE)
1704 		return ip_recv_error(sk, msg, len, addr_len);
1705 
1706 try_again:
1707 	off = sk_peek_offset(sk, flags);
1708 	skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1709 	if (!skb)
1710 		return err;
1711 
1712 	ulen = udp_skb_len(skb);
1713 	copied = len;
1714 	if (copied > ulen - off)
1715 		copied = ulen - off;
1716 	else if (copied < ulen)
1717 		msg->msg_flags |= MSG_TRUNC;
1718 
1719 	/*
1720 	 * If checksum is needed at all, try to do it while copying the
1721 	 * data.  If the data is truncated, or if we only want a partial
1722 	 * coverage checksum (UDP-Lite), do it before the copy.
1723 	 */
1724 
1725 	if (copied < ulen || peeking ||
1726 	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1727 		checksum_valid = udp_skb_csum_unnecessary(skb) ||
1728 				!__udp_lib_checksum_complete(skb);
1729 		if (!checksum_valid)
1730 			goto csum_copy_err;
1731 	}
1732 
1733 	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1734 		if (udp_skb_is_linear(skb))
1735 			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1736 		else
1737 			err = skb_copy_datagram_msg(skb, off, msg, copied);
1738 	} else {
1739 		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1740 
1741 		if (err == -EINVAL)
1742 			goto csum_copy_err;
1743 	}
1744 
1745 	if (unlikely(err)) {
1746 		if (!peeking) {
1747 			atomic_inc(&sk->sk_drops);
1748 			UDP_INC_STATS(sock_net(sk),
1749 				      UDP_MIB_INERRORS, is_udplite);
1750 		}
1751 		kfree_skb(skb);
1752 		return err;
1753 	}
1754 
1755 	if (!peeking)
1756 		UDP_INC_STATS(sock_net(sk),
1757 			      UDP_MIB_INDATAGRAMS, is_udplite);
1758 
1759 	sock_recv_ts_and_drops(msg, sk, skb);
1760 
1761 	/* Copy the address. */
1762 	if (sin) {
1763 		sin->sin_family = AF_INET;
1764 		sin->sin_port = udp_hdr(skb)->source;
1765 		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1766 		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1767 		*addr_len = sizeof(*sin);
1768 
1769 		if (cgroup_bpf_enabled)
1770 			BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1771 							(struct sockaddr *)sin);
1772 	}
1773 
1774 	if (udp_sk(sk)->gro_enabled)
1775 		udp_cmsg_recv(msg, sk, skb);
1776 
1777 	if (inet->cmsg_flags)
1778 		ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1779 
1780 	err = copied;
1781 	if (flags & MSG_TRUNC)
1782 		err = ulen;
1783 
1784 	skb_consume_udp(sk, skb, peeking ? -err : err);
1785 	return err;
1786 
1787 csum_copy_err:
1788 	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1789 				 udp_skb_destructor)) {
1790 		UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1791 		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1792 	}
1793 	kfree_skb(skb);
1794 
1795 	/* starting over for a new packet, but check if we need to yield */
1796 	cond_resched();
1797 	msg->msg_flags &= ~MSG_TRUNC;
1798 	goto try_again;
1799 }
1800 
1801 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1802 {
1803 	/* This check is replicated from __ip4_datagram_connect() and
1804 	 * intended to prevent BPF program called below from accessing bytes
1805 	 * that are out of the bound specified by user in addr_len.
1806 	 */
1807 	if (addr_len < sizeof(struct sockaddr_in))
1808 		return -EINVAL;
1809 
1810 	return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1811 }
1812 EXPORT_SYMBOL(udp_pre_connect);
1813 
1814 int __udp_disconnect(struct sock *sk, int flags)
1815 {
1816 	struct inet_sock *inet = inet_sk(sk);
1817 	/*
1818 	 *	1003.1g - break association.
1819 	 */
1820 
1821 	sk->sk_state = TCP_CLOSE;
1822 	inet->inet_daddr = 0;
1823 	inet->inet_dport = 0;
1824 	sock_rps_reset_rxhash(sk);
1825 	sk->sk_bound_dev_if = 0;
1826 	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1827 		inet_reset_saddr(sk);
1828 
1829 	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1830 		sk->sk_prot->unhash(sk);
1831 		inet->inet_sport = 0;
1832 	}
1833 	sk_dst_reset(sk);
1834 	return 0;
1835 }
1836 EXPORT_SYMBOL(__udp_disconnect);
1837 
1838 int udp_disconnect(struct sock *sk, int flags)
1839 {
1840 	lock_sock(sk);
1841 	__udp_disconnect(sk, flags);
1842 	release_sock(sk);
1843 	return 0;
1844 }
1845 EXPORT_SYMBOL(udp_disconnect);
1846 
1847 void udp_lib_unhash(struct sock *sk)
1848 {
1849 	if (sk_hashed(sk)) {
1850 		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1851 		struct udp_hslot *hslot, *hslot2;
1852 
1853 		hslot  = udp_hashslot(udptable, sock_net(sk),
1854 				      udp_sk(sk)->udp_port_hash);
1855 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1856 
1857 		spin_lock_bh(&hslot->lock);
1858 		if (rcu_access_pointer(sk->sk_reuseport_cb))
1859 			reuseport_detach_sock(sk);
1860 		if (sk_del_node_init_rcu(sk)) {
1861 			hslot->count--;
1862 			inet_sk(sk)->inet_num = 0;
1863 			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1864 
1865 			spin_lock(&hslot2->lock);
1866 			hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1867 			hslot2->count--;
1868 			spin_unlock(&hslot2->lock);
1869 		}
1870 		spin_unlock_bh(&hslot->lock);
1871 	}
1872 }
1873 EXPORT_SYMBOL(udp_lib_unhash);
1874 
1875 /*
1876  * inet_rcv_saddr was changed, we must rehash secondary hash
1877  */
1878 void udp_lib_rehash(struct sock *sk, u16 newhash)
1879 {
1880 	if (sk_hashed(sk)) {
1881 		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1882 		struct udp_hslot *hslot, *hslot2, *nhslot2;
1883 
1884 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1885 		nhslot2 = udp_hashslot2(udptable, newhash);
1886 		udp_sk(sk)->udp_portaddr_hash = newhash;
1887 
1888 		if (hslot2 != nhslot2 ||
1889 		    rcu_access_pointer(sk->sk_reuseport_cb)) {
1890 			hslot = udp_hashslot(udptable, sock_net(sk),
1891 					     udp_sk(sk)->udp_port_hash);
1892 			/* we must lock primary chain too */
1893 			spin_lock_bh(&hslot->lock);
1894 			if (rcu_access_pointer(sk->sk_reuseport_cb))
1895 				reuseport_detach_sock(sk);
1896 
1897 			if (hslot2 != nhslot2) {
1898 				spin_lock(&hslot2->lock);
1899 				hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1900 				hslot2->count--;
1901 				spin_unlock(&hslot2->lock);
1902 
1903 				spin_lock(&nhslot2->lock);
1904 				hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1905 							 &nhslot2->head);
1906 				nhslot2->count++;
1907 				spin_unlock(&nhslot2->lock);
1908 			}
1909 
1910 			spin_unlock_bh(&hslot->lock);
1911 		}
1912 	}
1913 }
1914 EXPORT_SYMBOL(udp_lib_rehash);
1915 
1916 void udp_v4_rehash(struct sock *sk)
1917 {
1918 	u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1919 					  inet_sk(sk)->inet_rcv_saddr,
1920 					  inet_sk(sk)->inet_num);
1921 	udp_lib_rehash(sk, new_hash);
1922 }
1923 
1924 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1925 {
1926 	int rc;
1927 
1928 	if (inet_sk(sk)->inet_daddr) {
1929 		sock_rps_save_rxhash(sk, skb);
1930 		sk_mark_napi_id(sk, skb);
1931 		sk_incoming_cpu_update(sk);
1932 	} else {
1933 		sk_mark_napi_id_once(sk, skb);
1934 	}
1935 
1936 	rc = __udp_enqueue_schedule_skb(sk, skb);
1937 	if (rc < 0) {
1938 		int is_udplite = IS_UDPLITE(sk);
1939 
1940 		/* Note that an ENOMEM error is charged twice */
1941 		if (rc == -ENOMEM)
1942 			UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1943 					is_udplite);
1944 		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1945 		kfree_skb(skb);
1946 		trace_udp_fail_queue_rcv_skb(rc, sk);
1947 		return -1;
1948 	}
1949 
1950 	return 0;
1951 }
1952 
1953 /* returns:
1954  *  -1: error
1955  *   0: success
1956  *  >0: "udp encap" protocol resubmission
1957  *
1958  * Note that in the success and error cases, the skb is assumed to
1959  * have either been requeued or freed.
1960  */
1961 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
1962 {
1963 	struct udp_sock *up = udp_sk(sk);
1964 	int is_udplite = IS_UDPLITE(sk);
1965 
1966 	/*
1967 	 *	Charge it to the socket, dropping if the queue is full.
1968 	 */
1969 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1970 		goto drop;
1971 	nf_reset(skb);
1972 
1973 	if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
1974 		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1975 
1976 		/*
1977 		 * This is an encapsulation socket so pass the skb to
1978 		 * the socket's udp_encap_rcv() hook. Otherwise, just
1979 		 * fall through and pass this up the UDP socket.
1980 		 * up->encap_rcv() returns the following value:
1981 		 * =0 if skb was successfully passed to the encap
1982 		 *    handler or was discarded by it.
1983 		 * >0 if skb should be passed on to UDP.
1984 		 * <0 if skb should be resubmitted as proto -N
1985 		 */
1986 
1987 		/* if we're overly short, let UDP handle it */
1988 		encap_rcv = READ_ONCE(up->encap_rcv);
1989 		if (encap_rcv) {
1990 			int ret;
1991 
1992 			/* Verify checksum before giving to encap */
1993 			if (udp_lib_checksum_complete(skb))
1994 				goto csum_error;
1995 
1996 			ret = encap_rcv(sk, skb);
1997 			if (ret <= 0) {
1998 				__UDP_INC_STATS(sock_net(sk),
1999 						UDP_MIB_INDATAGRAMS,
2000 						is_udplite);
2001 				return -ret;
2002 			}
2003 		}
2004 
2005 		/* FALLTHROUGH -- it's a UDP Packet */
2006 	}
2007 
2008 	/*
2009 	 * 	UDP-Lite specific tests, ignored on UDP sockets
2010 	 */
2011 	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
2012 
2013 		/*
2014 		 * MIB statistics other than incrementing the error count are
2015 		 * disabled for the following two types of errors: these depend
2016 		 * on the application settings, not on the functioning of the
2017 		 * protocol stack as such.
2018 		 *
2019 		 * RFC 3828 here recommends (sec 3.3): "There should also be a
2020 		 * way ... to ... at least let the receiving application block
2021 		 * delivery of packets with coverage values less than a value
2022 		 * provided by the application."
2023 		 */
2024 		if (up->pcrlen == 0) {          /* full coverage was set  */
2025 			net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2026 					    UDP_SKB_CB(skb)->cscov, skb->len);
2027 			goto drop;
2028 		}
2029 		/* The next case involves violating the min. coverage requested
2030 		 * by the receiver. This is subtle: if receiver wants x and x is
2031 		 * greater than the buffersize/MTU then receiver will complain
2032 		 * that it wants x while sender emits packets of smaller size y.
2033 		 * Therefore the above ...()->partial_cov statement is essential.
2034 		 */
2035 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
2036 			net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2037 					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
2038 			goto drop;
2039 		}
2040 	}
2041 
2042 	prefetch(&sk->sk_rmem_alloc);
2043 	if (rcu_access_pointer(sk->sk_filter) &&
2044 	    udp_lib_checksum_complete(skb))
2045 			goto csum_error;
2046 
2047 	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2048 		goto drop;
2049 
2050 	udp_csum_pull_header(skb);
2051 
2052 	ipv4_pktinfo_prepare(sk, skb);
2053 	return __udp_queue_rcv_skb(sk, skb);
2054 
2055 csum_error:
2056 	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2057 drop:
2058 	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2059 	atomic_inc(&sk->sk_drops);
2060 	kfree_skb(skb);
2061 	return -1;
2062 }
2063 
2064 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2065 {
2066 	struct sk_buff *next, *segs;
2067 	int ret;
2068 
2069 	if (likely(!udp_unexpected_gso(sk, skb)))
2070 		return udp_queue_rcv_one_skb(sk, skb);
2071 
2072 	BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET);
2073 	__skb_push(skb, -skb_mac_offset(skb));
2074 	segs = udp_rcv_segment(sk, skb, true);
2075 	for (skb = segs; skb; skb = next) {
2076 		next = skb->next;
2077 		__skb_pull(skb, skb_transport_offset(skb));
2078 		ret = udp_queue_rcv_one_skb(sk, skb);
2079 		if (ret > 0)
2080 			ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret);
2081 	}
2082 	return 0;
2083 }
2084 
2085 /* For TCP sockets, sk_rx_dst is protected by socket lock
2086  * For UDP, we use xchg() to guard against concurrent changes.
2087  */
2088 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2089 {
2090 	struct dst_entry *old;
2091 
2092 	if (dst_hold_safe(dst)) {
2093 		old = xchg(&sk->sk_rx_dst, dst);
2094 		dst_release(old);
2095 		return old != dst;
2096 	}
2097 	return false;
2098 }
2099 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2100 
2101 /*
2102  *	Multicasts and broadcasts go to each listener.
2103  *
2104  *	Note: called only from the BH handler context.
2105  */
2106 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2107 				    struct udphdr  *uh,
2108 				    __be32 saddr, __be32 daddr,
2109 				    struct udp_table *udptable,
2110 				    int proto)
2111 {
2112 	struct sock *sk, *first = NULL;
2113 	unsigned short hnum = ntohs(uh->dest);
2114 	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2115 	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2116 	unsigned int offset = offsetof(typeof(*sk), sk_node);
2117 	int dif = skb->dev->ifindex;
2118 	int sdif = inet_sdif(skb);
2119 	struct hlist_node *node;
2120 	struct sk_buff *nskb;
2121 
2122 	if (use_hash2) {
2123 		hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2124 			    udptable->mask;
2125 		hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2126 start_lookup:
2127 		hslot = &udptable->hash2[hash2];
2128 		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2129 	}
2130 
2131 	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2132 		if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2133 					 uh->source, saddr, dif, sdif, hnum))
2134 			continue;
2135 
2136 		if (!first) {
2137 			first = sk;
2138 			continue;
2139 		}
2140 		nskb = skb_clone(skb, GFP_ATOMIC);
2141 
2142 		if (unlikely(!nskb)) {
2143 			atomic_inc(&sk->sk_drops);
2144 			__UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2145 					IS_UDPLITE(sk));
2146 			__UDP_INC_STATS(net, UDP_MIB_INERRORS,
2147 					IS_UDPLITE(sk));
2148 			continue;
2149 		}
2150 		if (udp_queue_rcv_skb(sk, nskb) > 0)
2151 			consume_skb(nskb);
2152 	}
2153 
2154 	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
2155 	if (use_hash2 && hash2 != hash2_any) {
2156 		hash2 = hash2_any;
2157 		goto start_lookup;
2158 	}
2159 
2160 	if (first) {
2161 		if (udp_queue_rcv_skb(first, skb) > 0)
2162 			consume_skb(skb);
2163 	} else {
2164 		kfree_skb(skb);
2165 		__UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2166 				proto == IPPROTO_UDPLITE);
2167 	}
2168 	return 0;
2169 }
2170 
2171 /* Initialize UDP checksum. If exited with zero value (success),
2172  * CHECKSUM_UNNECESSARY means, that no more checks are required.
2173  * Otherwise, csum completion requires checksumming packet body,
2174  * including udp header and folding it to skb->csum.
2175  */
2176 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2177 				 int proto)
2178 {
2179 	int err;
2180 
2181 	UDP_SKB_CB(skb)->partial_cov = 0;
2182 	UDP_SKB_CB(skb)->cscov = skb->len;
2183 
2184 	if (proto == IPPROTO_UDPLITE) {
2185 		err = udplite_checksum_init(skb, uh);
2186 		if (err)
2187 			return err;
2188 
2189 		if (UDP_SKB_CB(skb)->partial_cov) {
2190 			skb->csum = inet_compute_pseudo(skb, proto);
2191 			return 0;
2192 		}
2193 	}
2194 
2195 	/* Note, we are only interested in != 0 or == 0, thus the
2196 	 * force to int.
2197 	 */
2198 	err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2199 							inet_compute_pseudo);
2200 	if (err)
2201 		return err;
2202 
2203 	if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2204 		/* If SW calculated the value, we know it's bad */
2205 		if (skb->csum_complete_sw)
2206 			return 1;
2207 
2208 		/* HW says the value is bad. Let's validate that.
2209 		 * skb->csum is no longer the full packet checksum,
2210 		 * so don't treat it as such.
2211 		 */
2212 		skb_checksum_complete_unset(skb);
2213 	}
2214 
2215 	return 0;
2216 }
2217 
2218 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2219  * return code conversion for ip layer consumption
2220  */
2221 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2222 			       struct udphdr *uh)
2223 {
2224 	int ret;
2225 
2226 	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2227 		skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2228 
2229 	ret = udp_queue_rcv_skb(sk, skb);
2230 
2231 	/* a return value > 0 means to resubmit the input, but
2232 	 * it wants the return to be -protocol, or 0
2233 	 */
2234 	if (ret > 0)
2235 		return -ret;
2236 	return 0;
2237 }
2238 
2239 /*
2240  *	All we need to do is get the socket, and then do a checksum.
2241  */
2242 
2243 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2244 		   int proto)
2245 {
2246 	struct sock *sk;
2247 	struct udphdr *uh;
2248 	unsigned short ulen;
2249 	struct rtable *rt = skb_rtable(skb);
2250 	__be32 saddr, daddr;
2251 	struct net *net = dev_net(skb->dev);
2252 
2253 	/*
2254 	 *  Validate the packet.
2255 	 */
2256 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2257 		goto drop;		/* No space for header. */
2258 
2259 	uh   = udp_hdr(skb);
2260 	ulen = ntohs(uh->len);
2261 	saddr = ip_hdr(skb)->saddr;
2262 	daddr = ip_hdr(skb)->daddr;
2263 
2264 	if (ulen > skb->len)
2265 		goto short_packet;
2266 
2267 	if (proto == IPPROTO_UDP) {
2268 		/* UDP validates ulen. */
2269 		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2270 			goto short_packet;
2271 		uh = udp_hdr(skb);
2272 	}
2273 
2274 	if (udp4_csum_init(skb, uh, proto))
2275 		goto csum_error;
2276 
2277 	sk = skb_steal_sock(skb);
2278 	if (sk) {
2279 		struct dst_entry *dst = skb_dst(skb);
2280 		int ret;
2281 
2282 		if (unlikely(sk->sk_rx_dst != dst))
2283 			udp_sk_rx_dst_set(sk, dst);
2284 
2285 		ret = udp_unicast_rcv_skb(sk, skb, uh);
2286 		sock_put(sk);
2287 		return ret;
2288 	}
2289 
2290 	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2291 		return __udp4_lib_mcast_deliver(net, skb, uh,
2292 						saddr, daddr, udptable, proto);
2293 
2294 	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2295 	if (sk)
2296 		return udp_unicast_rcv_skb(sk, skb, uh);
2297 
2298 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2299 		goto drop;
2300 	nf_reset(skb);
2301 
2302 	/* No socket. Drop packet silently, if checksum is wrong */
2303 	if (udp_lib_checksum_complete(skb))
2304 		goto csum_error;
2305 
2306 	__UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2307 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2308 
2309 	/*
2310 	 * Hmm.  We got an UDP packet to a port to which we
2311 	 * don't wanna listen.  Ignore it.
2312 	 */
2313 	kfree_skb(skb);
2314 	return 0;
2315 
2316 short_packet:
2317 	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2318 			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2319 			    &saddr, ntohs(uh->source),
2320 			    ulen, skb->len,
2321 			    &daddr, ntohs(uh->dest));
2322 	goto drop;
2323 
2324 csum_error:
2325 	/*
2326 	 * RFC1122: OK.  Discards the bad packet silently (as far as
2327 	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2328 	 */
2329 	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2330 			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2331 			    &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2332 			    ulen);
2333 	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2334 drop:
2335 	__UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2336 	kfree_skb(skb);
2337 	return 0;
2338 }
2339 
2340 /* We can only early demux multicast if there is a single matching socket.
2341  * If more than one socket found returns NULL
2342  */
2343 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2344 						  __be16 loc_port, __be32 loc_addr,
2345 						  __be16 rmt_port, __be32 rmt_addr,
2346 						  int dif, int sdif)
2347 {
2348 	struct sock *sk, *result;
2349 	unsigned short hnum = ntohs(loc_port);
2350 	unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2351 	struct udp_hslot *hslot = &udp_table.hash[slot];
2352 
2353 	/* Do not bother scanning a too big list */
2354 	if (hslot->count > 10)
2355 		return NULL;
2356 
2357 	result = NULL;
2358 	sk_for_each_rcu(sk, &hslot->head) {
2359 		if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2360 					rmt_port, rmt_addr, dif, sdif, hnum)) {
2361 			if (result)
2362 				return NULL;
2363 			result = sk;
2364 		}
2365 	}
2366 
2367 	return result;
2368 }
2369 
2370 /* For unicast we should only early demux connected sockets or we can
2371  * break forwarding setups.  The chains here can be long so only check
2372  * if the first socket is an exact match and if not move on.
2373  */
2374 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2375 					    __be16 loc_port, __be32 loc_addr,
2376 					    __be16 rmt_port, __be32 rmt_addr,
2377 					    int dif, int sdif)
2378 {
2379 	unsigned short hnum = ntohs(loc_port);
2380 	unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2381 	unsigned int slot2 = hash2 & udp_table.mask;
2382 	struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2383 	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2384 	const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2385 	struct sock *sk;
2386 
2387 	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2388 		if (INET_MATCH(sk, net, acookie, rmt_addr,
2389 			       loc_addr, ports, dif, sdif))
2390 			return sk;
2391 		/* Only check first socket in chain */
2392 		break;
2393 	}
2394 	return NULL;
2395 }
2396 
2397 int udp_v4_early_demux(struct sk_buff *skb)
2398 {
2399 	struct net *net = dev_net(skb->dev);
2400 	struct in_device *in_dev = NULL;
2401 	const struct iphdr *iph;
2402 	const struct udphdr *uh;
2403 	struct sock *sk = NULL;
2404 	struct dst_entry *dst;
2405 	int dif = skb->dev->ifindex;
2406 	int sdif = inet_sdif(skb);
2407 	int ours;
2408 
2409 	/* validate the packet */
2410 	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2411 		return 0;
2412 
2413 	iph = ip_hdr(skb);
2414 	uh = udp_hdr(skb);
2415 
2416 	if (skb->pkt_type == PACKET_MULTICAST) {
2417 		in_dev = __in_dev_get_rcu(skb->dev);
2418 
2419 		if (!in_dev)
2420 			return 0;
2421 
2422 		ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2423 				       iph->protocol);
2424 		if (!ours)
2425 			return 0;
2426 
2427 		sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2428 						   uh->source, iph->saddr,
2429 						   dif, sdif);
2430 	} else if (skb->pkt_type == PACKET_HOST) {
2431 		sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2432 					     uh->source, iph->saddr, dif, sdif);
2433 	}
2434 
2435 	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2436 		return 0;
2437 
2438 	skb->sk = sk;
2439 	skb->destructor = sock_efree;
2440 	dst = READ_ONCE(sk->sk_rx_dst);
2441 
2442 	if (dst)
2443 		dst = dst_check(dst, 0);
2444 	if (dst) {
2445 		u32 itag = 0;
2446 
2447 		/* set noref for now.
2448 		 * any place which wants to hold dst has to call
2449 		 * dst_hold_safe()
2450 		 */
2451 		skb_dst_set_noref(skb, dst);
2452 
2453 		/* for unconnected multicast sockets we need to validate
2454 		 * the source on each packet
2455 		 */
2456 		if (!inet_sk(sk)->inet_daddr && in_dev)
2457 			return ip_mc_validate_source(skb, iph->daddr,
2458 						     iph->saddr, iph->tos,
2459 						     skb->dev, in_dev, &itag);
2460 	}
2461 	return 0;
2462 }
2463 
2464 int udp_rcv(struct sk_buff *skb)
2465 {
2466 	return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2467 }
2468 
2469 void udp_destroy_sock(struct sock *sk)
2470 {
2471 	struct udp_sock *up = udp_sk(sk);
2472 	bool slow = lock_sock_fast(sk);
2473 	udp_flush_pending_frames(sk);
2474 	unlock_sock_fast(sk, slow);
2475 	if (static_branch_unlikely(&udp_encap_needed_key)) {
2476 		if (up->encap_type) {
2477 			void (*encap_destroy)(struct sock *sk);
2478 			encap_destroy = READ_ONCE(up->encap_destroy);
2479 			if (encap_destroy)
2480 				encap_destroy(sk);
2481 		}
2482 		if (up->encap_enabled)
2483 			static_branch_dec(&udp_encap_needed_key);
2484 	}
2485 }
2486 
2487 /*
2488  *	Socket option code for UDP
2489  */
2490 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2491 		       char __user *optval, unsigned int optlen,
2492 		       int (*push_pending_frames)(struct sock *))
2493 {
2494 	struct udp_sock *up = udp_sk(sk);
2495 	int val, valbool;
2496 	int err = 0;
2497 	int is_udplite = IS_UDPLITE(sk);
2498 
2499 	if (optlen < sizeof(int))
2500 		return -EINVAL;
2501 
2502 	if (get_user(val, (int __user *)optval))
2503 		return -EFAULT;
2504 
2505 	valbool = val ? 1 : 0;
2506 
2507 	switch (optname) {
2508 	case UDP_CORK:
2509 		if (val != 0) {
2510 			up->corkflag = 1;
2511 		} else {
2512 			up->corkflag = 0;
2513 			lock_sock(sk);
2514 			push_pending_frames(sk);
2515 			release_sock(sk);
2516 		}
2517 		break;
2518 
2519 	case UDP_ENCAP:
2520 		switch (val) {
2521 		case 0:
2522 		case UDP_ENCAP_ESPINUDP:
2523 		case UDP_ENCAP_ESPINUDP_NON_IKE:
2524 			up->encap_rcv = xfrm4_udp_encap_rcv;
2525 			/* FALLTHROUGH */
2526 		case UDP_ENCAP_L2TPINUDP:
2527 			up->encap_type = val;
2528 			lock_sock(sk);
2529 			udp_tunnel_encap_enable(sk->sk_socket);
2530 			release_sock(sk);
2531 			break;
2532 		default:
2533 			err = -ENOPROTOOPT;
2534 			break;
2535 		}
2536 		break;
2537 
2538 	case UDP_NO_CHECK6_TX:
2539 		up->no_check6_tx = valbool;
2540 		break;
2541 
2542 	case UDP_NO_CHECK6_RX:
2543 		up->no_check6_rx = valbool;
2544 		break;
2545 
2546 	case UDP_SEGMENT:
2547 		if (val < 0 || val > USHRT_MAX)
2548 			return -EINVAL;
2549 		up->gso_size = val;
2550 		break;
2551 
2552 	case UDP_GRO:
2553 		lock_sock(sk);
2554 		if (valbool)
2555 			udp_tunnel_encap_enable(sk->sk_socket);
2556 		up->gro_enabled = valbool;
2557 		release_sock(sk);
2558 		break;
2559 
2560 	/*
2561 	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
2562 	 */
2563 	/* The sender sets actual checksum coverage length via this option.
2564 	 * The case coverage > packet length is handled by send module. */
2565 	case UDPLITE_SEND_CSCOV:
2566 		if (!is_udplite)         /* Disable the option on UDP sockets */
2567 			return -ENOPROTOOPT;
2568 		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2569 			val = 8;
2570 		else if (val > USHRT_MAX)
2571 			val = USHRT_MAX;
2572 		up->pcslen = val;
2573 		up->pcflag |= UDPLITE_SEND_CC;
2574 		break;
2575 
2576 	/* The receiver specifies a minimum checksum coverage value. To make
2577 	 * sense, this should be set to at least 8 (as done below). If zero is
2578 	 * used, this again means full checksum coverage.                     */
2579 	case UDPLITE_RECV_CSCOV:
2580 		if (!is_udplite)         /* Disable the option on UDP sockets */
2581 			return -ENOPROTOOPT;
2582 		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2583 			val = 8;
2584 		else if (val > USHRT_MAX)
2585 			val = USHRT_MAX;
2586 		up->pcrlen = val;
2587 		up->pcflag |= UDPLITE_RECV_CC;
2588 		break;
2589 
2590 	default:
2591 		err = -ENOPROTOOPT;
2592 		break;
2593 	}
2594 
2595 	return err;
2596 }
2597 EXPORT_SYMBOL(udp_lib_setsockopt);
2598 
2599 int udp_setsockopt(struct sock *sk, int level, int optname,
2600 		   char __user *optval, unsigned int optlen)
2601 {
2602 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2603 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2604 					  udp_push_pending_frames);
2605 	return ip_setsockopt(sk, level, optname, optval, optlen);
2606 }
2607 
2608 #ifdef CONFIG_COMPAT
2609 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2610 			  char __user *optval, unsigned int optlen)
2611 {
2612 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2613 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2614 					  udp_push_pending_frames);
2615 	return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2616 }
2617 #endif
2618 
2619 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2620 		       char __user *optval, int __user *optlen)
2621 {
2622 	struct udp_sock *up = udp_sk(sk);
2623 	int val, len;
2624 
2625 	if (get_user(len, optlen))
2626 		return -EFAULT;
2627 
2628 	len = min_t(unsigned int, len, sizeof(int));
2629 
2630 	if (len < 0)
2631 		return -EINVAL;
2632 
2633 	switch (optname) {
2634 	case UDP_CORK:
2635 		val = up->corkflag;
2636 		break;
2637 
2638 	case UDP_ENCAP:
2639 		val = up->encap_type;
2640 		break;
2641 
2642 	case UDP_NO_CHECK6_TX:
2643 		val = up->no_check6_tx;
2644 		break;
2645 
2646 	case UDP_NO_CHECK6_RX:
2647 		val = up->no_check6_rx;
2648 		break;
2649 
2650 	case UDP_SEGMENT:
2651 		val = up->gso_size;
2652 		break;
2653 
2654 	/* The following two cannot be changed on UDP sockets, the return is
2655 	 * always 0 (which corresponds to the full checksum coverage of UDP). */
2656 	case UDPLITE_SEND_CSCOV:
2657 		val = up->pcslen;
2658 		break;
2659 
2660 	case UDPLITE_RECV_CSCOV:
2661 		val = up->pcrlen;
2662 		break;
2663 
2664 	default:
2665 		return -ENOPROTOOPT;
2666 	}
2667 
2668 	if (put_user(len, optlen))
2669 		return -EFAULT;
2670 	if (copy_to_user(optval, &val, len))
2671 		return -EFAULT;
2672 	return 0;
2673 }
2674 EXPORT_SYMBOL(udp_lib_getsockopt);
2675 
2676 int udp_getsockopt(struct sock *sk, int level, int optname,
2677 		   char __user *optval, int __user *optlen)
2678 {
2679 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2680 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2681 	return ip_getsockopt(sk, level, optname, optval, optlen);
2682 }
2683 
2684 #ifdef CONFIG_COMPAT
2685 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2686 				 char __user *optval, int __user *optlen)
2687 {
2688 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2689 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2690 	return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2691 }
2692 #endif
2693 /**
2694  * 	udp_poll - wait for a UDP event.
2695  *	@file - file struct
2696  *	@sock - socket
2697  *	@wait - poll table
2698  *
2699  *	This is same as datagram poll, except for the special case of
2700  *	blocking sockets. If application is using a blocking fd
2701  *	and a packet with checksum error is in the queue;
2702  *	then it could get return from select indicating data available
2703  *	but then block when reading it. Add special case code
2704  *	to work around these arguably broken applications.
2705  */
2706 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2707 {
2708 	__poll_t mask = datagram_poll(file, sock, wait);
2709 	struct sock *sk = sock->sk;
2710 
2711 	if (!skb_queue_empty(&udp_sk(sk)->reader_queue))
2712 		mask |= EPOLLIN | EPOLLRDNORM;
2713 
2714 	/* Check for false positives due to checksum errors */
2715 	if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2716 	    !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2717 		mask &= ~(EPOLLIN | EPOLLRDNORM);
2718 
2719 	return mask;
2720 
2721 }
2722 EXPORT_SYMBOL(udp_poll);
2723 
2724 int udp_abort(struct sock *sk, int err)
2725 {
2726 	lock_sock(sk);
2727 
2728 	sk->sk_err = err;
2729 	sk->sk_error_report(sk);
2730 	__udp_disconnect(sk, 0);
2731 
2732 	release_sock(sk);
2733 
2734 	return 0;
2735 }
2736 EXPORT_SYMBOL_GPL(udp_abort);
2737 
2738 struct proto udp_prot = {
2739 	.name			= "UDP",
2740 	.owner			= THIS_MODULE,
2741 	.close			= udp_lib_close,
2742 	.pre_connect		= udp_pre_connect,
2743 	.connect		= ip4_datagram_connect,
2744 	.disconnect		= udp_disconnect,
2745 	.ioctl			= udp_ioctl,
2746 	.init			= udp_init_sock,
2747 	.destroy		= udp_destroy_sock,
2748 	.setsockopt		= udp_setsockopt,
2749 	.getsockopt		= udp_getsockopt,
2750 	.sendmsg		= udp_sendmsg,
2751 	.recvmsg		= udp_recvmsg,
2752 	.sendpage		= udp_sendpage,
2753 	.release_cb		= ip4_datagram_release_cb,
2754 	.hash			= udp_lib_hash,
2755 	.unhash			= udp_lib_unhash,
2756 	.rehash			= udp_v4_rehash,
2757 	.get_port		= udp_v4_get_port,
2758 	.memory_allocated	= &udp_memory_allocated,
2759 	.sysctl_mem		= sysctl_udp_mem,
2760 	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2761 	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2762 	.obj_size		= sizeof(struct udp_sock),
2763 	.h.udp_table		= &udp_table,
2764 #ifdef CONFIG_COMPAT
2765 	.compat_setsockopt	= compat_udp_setsockopt,
2766 	.compat_getsockopt	= compat_udp_getsockopt,
2767 #endif
2768 	.diag_destroy		= udp_abort,
2769 };
2770 EXPORT_SYMBOL(udp_prot);
2771 
2772 /* ------------------------------------------------------------------------ */
2773 #ifdef CONFIG_PROC_FS
2774 
2775 static struct sock *udp_get_first(struct seq_file *seq, int start)
2776 {
2777 	struct sock *sk;
2778 	struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2779 	struct udp_iter_state *state = seq->private;
2780 	struct net *net = seq_file_net(seq);
2781 
2782 	for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2783 	     ++state->bucket) {
2784 		struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2785 
2786 		if (hlist_empty(&hslot->head))
2787 			continue;
2788 
2789 		spin_lock_bh(&hslot->lock);
2790 		sk_for_each(sk, &hslot->head) {
2791 			if (!net_eq(sock_net(sk), net))
2792 				continue;
2793 			if (sk->sk_family == afinfo->family)
2794 				goto found;
2795 		}
2796 		spin_unlock_bh(&hslot->lock);
2797 	}
2798 	sk = NULL;
2799 found:
2800 	return sk;
2801 }
2802 
2803 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2804 {
2805 	struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2806 	struct udp_iter_state *state = seq->private;
2807 	struct net *net = seq_file_net(seq);
2808 
2809 	do {
2810 		sk = sk_next(sk);
2811 	} while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2812 
2813 	if (!sk) {
2814 		if (state->bucket <= afinfo->udp_table->mask)
2815 			spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2816 		return udp_get_first(seq, state->bucket + 1);
2817 	}
2818 	return sk;
2819 }
2820 
2821 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2822 {
2823 	struct sock *sk = udp_get_first(seq, 0);
2824 
2825 	if (sk)
2826 		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2827 			--pos;
2828 	return pos ? NULL : sk;
2829 }
2830 
2831 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2832 {
2833 	struct udp_iter_state *state = seq->private;
2834 	state->bucket = MAX_UDP_PORTS;
2835 
2836 	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2837 }
2838 EXPORT_SYMBOL(udp_seq_start);
2839 
2840 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2841 {
2842 	struct sock *sk;
2843 
2844 	if (v == SEQ_START_TOKEN)
2845 		sk = udp_get_idx(seq, 0);
2846 	else
2847 		sk = udp_get_next(seq, v);
2848 
2849 	++*pos;
2850 	return sk;
2851 }
2852 EXPORT_SYMBOL(udp_seq_next);
2853 
2854 void udp_seq_stop(struct seq_file *seq, void *v)
2855 {
2856 	struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2857 	struct udp_iter_state *state = seq->private;
2858 
2859 	if (state->bucket <= afinfo->udp_table->mask)
2860 		spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2861 }
2862 EXPORT_SYMBOL(udp_seq_stop);
2863 
2864 /* ------------------------------------------------------------------------ */
2865 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2866 		int bucket)
2867 {
2868 	struct inet_sock *inet = inet_sk(sp);
2869 	__be32 dest = inet->inet_daddr;
2870 	__be32 src  = inet->inet_rcv_saddr;
2871 	__u16 destp	  = ntohs(inet->inet_dport);
2872 	__u16 srcp	  = ntohs(inet->inet_sport);
2873 
2874 	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2875 		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
2876 		bucket, src, srcp, dest, destp, sp->sk_state,
2877 		sk_wmem_alloc_get(sp),
2878 		udp_rqueue_get(sp),
2879 		0, 0L, 0,
2880 		from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2881 		0, sock_i_ino(sp),
2882 		refcount_read(&sp->sk_refcnt), sp,
2883 		atomic_read(&sp->sk_drops));
2884 }
2885 
2886 int udp4_seq_show(struct seq_file *seq, void *v)
2887 {
2888 	seq_setwidth(seq, 127);
2889 	if (v == SEQ_START_TOKEN)
2890 		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2891 			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2892 			   "inode ref pointer drops");
2893 	else {
2894 		struct udp_iter_state *state = seq->private;
2895 
2896 		udp4_format_sock(v, seq, state->bucket);
2897 	}
2898 	seq_pad(seq, '\n');
2899 	return 0;
2900 }
2901 
2902 const struct seq_operations udp_seq_ops = {
2903 	.start		= udp_seq_start,
2904 	.next		= udp_seq_next,
2905 	.stop		= udp_seq_stop,
2906 	.show		= udp4_seq_show,
2907 };
2908 EXPORT_SYMBOL(udp_seq_ops);
2909 
2910 static struct udp_seq_afinfo udp4_seq_afinfo = {
2911 	.family		= AF_INET,
2912 	.udp_table	= &udp_table,
2913 };
2914 
2915 static int __net_init udp4_proc_init_net(struct net *net)
2916 {
2917 	if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2918 			sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2919 		return -ENOMEM;
2920 	return 0;
2921 }
2922 
2923 static void __net_exit udp4_proc_exit_net(struct net *net)
2924 {
2925 	remove_proc_entry("udp", net->proc_net);
2926 }
2927 
2928 static struct pernet_operations udp4_net_ops = {
2929 	.init = udp4_proc_init_net,
2930 	.exit = udp4_proc_exit_net,
2931 };
2932 
2933 int __init udp4_proc_init(void)
2934 {
2935 	return register_pernet_subsys(&udp4_net_ops);
2936 }
2937 
2938 void udp4_proc_exit(void)
2939 {
2940 	unregister_pernet_subsys(&udp4_net_ops);
2941 }
2942 #endif /* CONFIG_PROC_FS */
2943 
2944 static __initdata unsigned long uhash_entries;
2945 static int __init set_uhash_entries(char *str)
2946 {
2947 	ssize_t ret;
2948 
2949 	if (!str)
2950 		return 0;
2951 
2952 	ret = kstrtoul(str, 0, &uhash_entries);
2953 	if (ret)
2954 		return 0;
2955 
2956 	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2957 		uhash_entries = UDP_HTABLE_SIZE_MIN;
2958 	return 1;
2959 }
2960 __setup("uhash_entries=", set_uhash_entries);
2961 
2962 void __init udp_table_init(struct udp_table *table, const char *name)
2963 {
2964 	unsigned int i;
2965 
2966 	table->hash = alloc_large_system_hash(name,
2967 					      2 * sizeof(struct udp_hslot),
2968 					      uhash_entries,
2969 					      21, /* one slot per 2 MB */
2970 					      0,
2971 					      &table->log,
2972 					      &table->mask,
2973 					      UDP_HTABLE_SIZE_MIN,
2974 					      64 * 1024);
2975 
2976 	table->hash2 = table->hash + (table->mask + 1);
2977 	for (i = 0; i <= table->mask; i++) {
2978 		INIT_HLIST_HEAD(&table->hash[i].head);
2979 		table->hash[i].count = 0;
2980 		spin_lock_init(&table->hash[i].lock);
2981 	}
2982 	for (i = 0; i <= table->mask; i++) {
2983 		INIT_HLIST_HEAD(&table->hash2[i].head);
2984 		table->hash2[i].count = 0;
2985 		spin_lock_init(&table->hash2[i].lock);
2986 	}
2987 }
2988 
2989 u32 udp_flow_hashrnd(void)
2990 {
2991 	static u32 hashrnd __read_mostly;
2992 
2993 	net_get_random_once(&hashrnd, sizeof(hashrnd));
2994 
2995 	return hashrnd;
2996 }
2997 EXPORT_SYMBOL(udp_flow_hashrnd);
2998 
2999 static void __udp_sysctl_init(struct net *net)
3000 {
3001 	net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3002 	net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3003 
3004 #ifdef CONFIG_NET_L3_MASTER_DEV
3005 	net->ipv4.sysctl_udp_l3mdev_accept = 0;
3006 #endif
3007 }
3008 
3009 static int __net_init udp_sysctl_init(struct net *net)
3010 {
3011 	__udp_sysctl_init(net);
3012 	return 0;
3013 }
3014 
3015 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3016 	.init	= udp_sysctl_init,
3017 };
3018 
3019 void __init udp_init(void)
3020 {
3021 	unsigned long limit;
3022 	unsigned int i;
3023 
3024 	udp_table_init(&udp_table, "UDP");
3025 	limit = nr_free_buffer_pages() / 8;
3026 	limit = max(limit, 128UL);
3027 	sysctl_udp_mem[0] = limit / 4 * 3;
3028 	sysctl_udp_mem[1] = limit;
3029 	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3030 
3031 	__udp_sysctl_init(&init_net);
3032 
3033 	/* 16 spinlocks per cpu */
3034 	udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3035 	udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3036 				GFP_KERNEL);
3037 	if (!udp_busylocks)
3038 		panic("UDP: failed to alloc udp_busylocks\n");
3039 	for (i = 0; i < (1U << udp_busylocks_log); i++)
3040 		spin_lock_init(udp_busylocks + i);
3041 
3042 	if (register_pernet_subsys(&udp_sysctl_ops))
3043 		panic("UDP: failed to init sysctl parameters.\n");
3044 }
3045