1	// SPDX-License-Identifier: GPL-2.0-or-later
2	/*
3	* INET An implementation of the TCP/IP protocol suite for the LINUX
4	* operating system. INET is implemented using the BSD Socket
5	* interface as the means of communication with the user level.
6	*
7	* The User Datagram Protocol (UDP).
8	*
9	* Authors: Ross Biro
10	* Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11	* Arnt Gulbrandsen, <agulbra@nvg.unit.no>
12	* Alan Cox, <alan@lxorguk.ukuu.org.uk>
13	* Hirokazu Takahashi, <taka@valinux.co.jp>
14	*
15	* Fixes:
16	* Alan Cox : verify_area() calls
17	* Alan Cox : stopped close while in use off icmp
18	* messages. Not a fix but a botch that
19	* for udp at least is 'valid'.
20	* Alan Cox : Fixed icmp handling properly
21	* Alan Cox : Correct error for oversized datagrams
22	* Alan Cox : Tidied select() semantics.
23	* Alan Cox : udp_err() fixed properly, also now
24	* select and read wake correctly on errors
25	* Alan Cox : udp_send verify_area moved to avoid mem leak
26	* Alan Cox : UDP can count its memory
27	* Alan Cox : send to an unknown connection causes
28	* an ECONNREFUSED off the icmp, but
29	* does NOT close.
30	* Alan Cox : Switched to new sk_buff handlers. No more backlog!
31	* Alan Cox : Using generic datagram code. Even smaller and the PEEK
32	* bug no longer crashes it.
33	* Fred Van Kempen : Net2e support for sk->broadcast.
34	* Alan Cox : Uses skb_free_datagram
35	* Alan Cox : Added get/set sockopt support.
36	* Alan Cox : Broadcasting without option set returns EACCES.
37	* Alan Cox : No wakeup calls. Instead we now use the callbacks.
38	* Alan Cox : Use ip_tos and ip_ttl
39	* Alan Cox : SNMP Mibs
40	* Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
41	* Matt Dillon : UDP length checks.
42	* Alan Cox : Smarter af_inet used properly.
43	* Alan Cox : Use new kernel side addressing.
44	* Alan Cox : Incorrect return on truncated datagram receive.
45	* Arnt Gulbrandsen : New udp_send and stuff
46	* Alan Cox : Cache last socket
47	* Alan Cox : Route cache
48	* Jon Peatfield : Minor efficiency fix to sendto().
49	* Mike Shaver : RFC1122 checks.
50	* Alan Cox : Nonblocking error fix.
51	* Willy Konynenberg : Transparent proxying support.
52	* Mike McLagan : Routing by source
53	* David S. Miller : New socket lookup architecture.
54	* Last socket cache retained as it
55	* does have a high hit rate.
56	* Olaf Kirch : Don't linearise iovec on sendmsg.
57	* Andi Kleen : Some cleanups, cache destination entry
58	* for connect.
59	* Vitaly E. Lavrov : Transparent proxy revived after year coma.
60	* Melvin Smith : Check msg_name not msg_namelen in sendto(),
61	* return ENOTCONN for unconnected sockets (POSIX)
62	* Janos Farkas : don't deliver multi/broadcasts to a different
63	* bound-to-device socket
64	* Hirokazu Takahashi : HW checksumming for outgoing UDP
65	* datagrams.
66	* Hirokazu Takahashi : sendfile() on UDP works now.
67	* Arnaldo C. Melo : convert /proc/net/udp to seq_file
68	* YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
69	* Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
70	* a single port at the same time.
71	* Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
72	* James Chapman : Add L2TP encapsulation type.
73	*/
74
75	#define pr_fmt(fmt) "UDP: " fmt
76
77	#include <linux/bpf-cgroup.h>
78	#include <linux/uaccess.h>
79	#include <asm/ioctls.h>
80	#include <linux/memblock.h>
81	#include <linux/highmem.h>
82	#include <linux/types.h>
83	#include <linux/fcntl.h>
84	#include <linux/module.h>
85	#include <linux/socket.h>
86	#include <linux/sockios.h>
87	#include <linux/igmp.h>
88	#include <linux/inetdevice.h>
89	#include <linux/in.h>
90	#include <linux/errno.h>
91	#include <linux/timer.h>
92	#include <linux/mm.h>
93	#include <linux/inet.h>
94	#include <linux/netdevice.h>
95	#include <linux/slab.h>
96	#include <net/tcp_states.h>
97	#include <linux/skbuff.h>
98	#include <linux/proc_fs.h>
99	#include <linux/seq_file.h>
100	#include <net/net_namespace.h>
101	#include <net/icmp.h>
102	#include <net/inet_hashtables.h>
103	#include <net/ip_tunnels.h>
104	#include <net/route.h>
105	#include <net/checksum.h>
106	#include <net/gso.h>
107	#include <net/xfrm.h>
108	#include <trace/events/udp.h>
109	#include <linux/static_key.h>
110	#include <linux/btf_ids.h>
111	#include <trace/events/skb.h>
112	#include <net/busy_poll.h>
113	#include "udp_impl.h"
114	#include <net/sock_reuseport.h>
115	#include <net/addrconf.h>
116	#include <net/udp_tunnel.h>
117	#include <net/gro.h>
118	#if IS_ENABLED(CONFIG_IPV6)
119	#include <net/ipv6_stubs.h>
120	#endif
121
122	struct udp_table udp_table __read_mostly;
123	EXPORT_SYMBOL(udp_table);
124
125	long sysctl_udp_mem[3] __read_mostly;
126	EXPORT_SYMBOL(sysctl_udp_mem);
127
128	atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
129	EXPORT_SYMBOL(udp_memory_allocated);
130	DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
131	EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
132
133	#define MAX_UDP_PORTS 65536
134	#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET)
135
136	static struct udp_table *udp_get_table_prot(struct sock *sk)
137	{
138	return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
139	}
140
141	static int udp_lib_lport_inuse(struct net *net, __u16 num,
142	const struct udp_hslot *hslot,
143	unsigned long *bitmap,
144	struct sock *sk, unsigned int log)
145	{
146	struct sock *sk2;
147	kuid_t uid = sock_i_uid(sk);
148
149	sk_for_each(sk2, &hslot->head) {
150	if (net_eq(net1: sock_net(sk: sk2), net2: net) &&
151	sk2 != sk &&
152	(bitmap || udp_sk(sk2)->udp_port_hash == num) &&
153	(!sk2->sk_reuse || !sk->sk_reuse) &&
154	(!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
155	sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
156	inet_rcv_saddr_equal(sk, sk2, match_wildcard: true)) {
157	if (sk2->sk_reuseport && sk->sk_reuseport &&
158	!rcu_access_pointer(sk->sk_reuseport_cb) &&
159	uid_eq(left: uid, right: sock_i_uid(sk: sk2))) {
160	if (!bitmap)
161	return 0;
162	} else {
163	if (!bitmap)
164	return 1;
165	__set_bit(udp_sk(sk2)->udp_port_hash >> log,
166	bitmap);
167	}
168	}
169	}
170	return 0;
171	}
172
173	/*
174	* Note: we still hold spinlock of primary hash chain, so no other writer
175	* can insert/delete a socket with local_port == num
176	*/
177	static int udp_lib_lport_inuse2(struct net *net, __u16 num,
178	struct udp_hslot *hslot2,
179	struct sock *sk)
180	{
181	struct sock *sk2;
182	kuid_t uid = sock_i_uid(sk);
183	int res = 0;
184
185	spin_lock(lock: &hslot2->lock);
186	udp_portaddr_for_each_entry(sk2, &hslot2->head) {
187	if (net_eq(net1: sock_net(sk: sk2), net2: net) &&
188	sk2 != sk &&
189	(udp_sk(sk2)->udp_port_hash == num) &&
190	(!sk2->sk_reuse || !sk->sk_reuse) &&
191	(!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
192	sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
193	inet_rcv_saddr_equal(sk, sk2, match_wildcard: true)) {
194	if (sk2->sk_reuseport && sk->sk_reuseport &&
195	!rcu_access_pointer(sk->sk_reuseport_cb) &&
196	uid_eq(left: uid, right: sock_i_uid(sk: sk2))) {
197	res = 0;
198	} else {
199	res = 1;
200	}
201	break;
202	}
203	}
204	spin_unlock(lock: &hslot2->lock);
205	return res;
206	}
207
208	static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
209	{
210	struct net *net = sock_net(sk);
211	kuid_t uid = sock_i_uid(sk);
212	struct sock *sk2;
213
214	sk_for_each(sk2, &hslot->head) {
215	if (net_eq(net1: sock_net(sk: sk2), net2: net) &&
216	sk2 != sk &&
217	sk2->sk_family == sk->sk_family &&
218	ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
219	(udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
220	(sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
221	sk2->sk_reuseport && uid_eq(left: uid, right: sock_i_uid(sk: sk2)) &&
222	inet_rcv_saddr_equal(sk, sk2, match_wildcard: false)) {
223	return reuseport_add_sock(sk, sk2,
224	bind_inany: inet_rcv_saddr_any(sk));
225	}
226	}
227
228	return reuseport_alloc(sk, bind_inany: inet_rcv_saddr_any(sk));
229	}
230
231	/**
232	* udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
233	*
234	* @sk: socket struct in question
235	* @snum: port number to look up
236	* @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
237	* with NULL address
238	*/
239	int udp_lib_get_port(struct sock *sk, unsigned short snum,
240	unsigned int hash2_nulladdr)
241	{
242	struct udp_table *udptable = udp_get_table_prot(sk);
243	struct udp_hslot *hslot, *hslot2;
244	struct net *net = sock_net(sk);
245	int error = -EADDRINUSE;
246
247	if (!snum) {
248	DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
249	unsigned short first, last;
250	int low, high, remaining;
251	unsigned int rand;
252
253	inet_sk_get_local_port_range(sk, low: &low, high: &high);
254	remaining = (high - low) + 1;
255
256	rand = get_random_u32();
257	first = reciprocal_scale(val: rand, ep_ro: remaining) + low;
258	/*
259	* force rand to be an odd multiple of UDP_HTABLE_SIZE
260	*/
261	rand = (rand | 1) * (udptable->mask + 1);
262	last = first + udptable->mask + 1;
263	do {
264	hslot = udp_hashslot(table: udptable, net, num: first);
265	bitmap_zero(dst: bitmap, PORTS_PER_CHAIN);
266	spin_lock_bh(lock: &hslot->lock);
267	udp_lib_lport_inuse(net, num: snum, hslot, bitmap, sk,
268	log: udptable->log);
269
270	snum = first;
271	/*
272	* Iterate on all possible values of snum for this hash.
273	* Using steps of an odd multiple of UDP_HTABLE_SIZE
274	* give us randomization and full range coverage.
275	*/
276	do {
277	if (low <= snum && snum <= high &&
278	!test_bit(snum >> udptable->log, bitmap) &&
279	!inet_is_local_reserved_port(net, port: snum))
280	goto found;
281	snum += rand;
282	} while (snum != first);
283	spin_unlock_bh(lock: &hslot->lock);
284	cond_resched();
285	} while (++first != last);
286	goto fail;
287	} else {
288	hslot = udp_hashslot(table: udptable, net, num: snum);
289	spin_lock_bh(lock: &hslot->lock);
290	if (hslot->count > 10) {
291	int exist;
292	unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
293
294	slot2 &= udptable->mask;
295	hash2_nulladdr &= udptable->mask;
296
297	hslot2 = udp_hashslot2(table: udptable, hash: slot2);
298	if (hslot->count < hslot2->count)
299	goto scan_primary_hash;
300
301	exist = udp_lib_lport_inuse2(net, num: snum, hslot2, sk);
302	if (!exist && (hash2_nulladdr != slot2)) {
303	hslot2 = udp_hashslot2(table: udptable, hash: hash2_nulladdr);
304	exist = udp_lib_lport_inuse2(net, num: snum, hslot2,
305	sk);
306	}
307	if (exist)
308	goto fail_unlock;
309	else
310	goto found;
311	}
312	scan_primary_hash:
313	if (udp_lib_lport_inuse(net, num: snum, hslot, NULL, sk, log: 0))
314	goto fail_unlock;
315	}
316	found:
317	inet_sk(sk)->inet_num = snum;
318	udp_sk(sk)->udp_port_hash = snum;
319	udp_sk(sk)->udp_portaddr_hash ^= snum;
320	if (sk_unhashed(sk)) {
321	if (sk->sk_reuseport &&
322	udp_reuseport_add_sock(sk, hslot)) {
323	inet_sk(sk)->inet_num = 0;
324	udp_sk(sk)->udp_port_hash = 0;
325	udp_sk(sk)->udp_portaddr_hash ^= snum;
326	goto fail_unlock;
327	}
328
329	sk_add_node_rcu(sk, list: &hslot->head);
330	hslot->count++;
331	sock_prot_inuse_add(net: sock_net(sk), prot: sk->sk_prot, val: 1);
332
333	hslot2 = udp_hashslot2(table: udptable, udp_sk(sk)->udp_portaddr_hash);
334	spin_lock(lock: &hslot2->lock);
335	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
336	sk->sk_family == AF_INET6)
337	hlist_add_tail_rcu(n: &udp_sk(sk)->udp_portaddr_node,
338	h: &hslot2->head);
339	else
340	hlist_add_head_rcu(n: &udp_sk(sk)->udp_portaddr_node,
341	h: &hslot2->head);
342	hslot2->count++;
343	spin_unlock(lock: &hslot2->lock);
344	}
345	sock_set_flag(sk, flag: SOCK_RCU_FREE);
346	error = 0;
347	fail_unlock:
348	spin_unlock_bh(lock: &hslot->lock);
349	fail:
350	return error;
351	}
352	EXPORT_SYMBOL(udp_lib_get_port);
353
354	int udp_v4_get_port(struct sock *sk, unsigned short snum)
355	{
356	unsigned int hash2_nulladdr =
357	ipv4_portaddr_hash(net: sock_net(sk), htonl(INADDR_ANY), port: snum);
358	unsigned int hash2_partial =
359	ipv4_portaddr_hash(net: sock_net(sk), inet_sk(sk)->inet_rcv_saddr, port: 0);
360
361	/* precompute partial secondary hash */
362	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
363	return udp_lib_get_port(sk, snum, hash2_nulladdr);
364	}
365
366	static int compute_score(struct sock *sk, struct net *net,
367	__be32 saddr, __be16 sport,
368	__be32 daddr, unsigned short hnum,
369	int dif, int sdif)
370	{
371	int score;
372	struct inet_sock *inet;
373	bool dev_match;
374
375	if (!net_eq(net1: sock_net(sk), net2: net) ||
376	udp_sk(sk)->udp_port_hash != hnum ||
377	ipv6_only_sock(sk))
378	return -1;
379
380	if (sk->sk_rcv_saddr != daddr)
381	return -1;
382
383	score = (sk->sk_family == PF_INET) ? 2 : 1;
384
385	inet = inet_sk(sk);
386	if (inet->inet_daddr) {
387	if (inet->inet_daddr != saddr)
388	return -1;
389	score += 4;
390	}
391
392	if (inet->inet_dport) {
393	if (inet->inet_dport != sport)
394	return -1;
395	score += 4;
396	}
397
398	dev_match = udp_sk_bound_dev_eq(net, bound_dev_if: sk->sk_bound_dev_if,
399	dif, sdif);
400	if (!dev_match)
401	return -1;
402	if (sk->sk_bound_dev_if)
403	score += 4;
404
405	if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
406	score++;
407	return score;
408	}
409
410	INDIRECT_CALLABLE_SCOPE
411	u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
412	const __be32 faddr, const __be16 fport)
413	{
414	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
415
416	return __inet_ehashfn(laddr, lport, faddr, fport,
417	udp_ehash_secret + net_hash_mix(net));
418	}
419
420	/* called with rcu_read_lock() */
421	static struct sock *udp4_lib_lookup2(struct net *net,
422	__be32 saddr, __be16 sport,
423	__be32 daddr, unsigned int hnum,
424	int dif, int sdif,
425	struct udp_hslot *hslot2,
426	struct sk_buff *skb)
427	{
428	struct sock *sk, *result;
429	int score, badness;
430
431	result = NULL;
432	badness = 0;
433	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
434	score = compute_score(sk, net, saddr, sport,
435	daddr, hnum, dif, sdif);
436	if (score > badness) {
437	badness = score;
438
439	if (sk->sk_state == TCP_ESTABLISHED) {
440	result = sk;
441	continue;
442	}
443
444	result = inet_lookup_reuseport(net, sk, skb, doff: sizeof(struct udphdr),
445	saddr, sport, daddr, hnum, ehashfn: udp_ehashfn);
446	if (!result) {
447	result = sk;
448	continue;
449	}
450
451	/* Fall back to scoring if group has connections */
452	if (!reuseport_has_conns(sk))
453	return result;
454
455	/* Reuseport logic returned an error, keep original score. */
456	if (IS_ERR(ptr: result))
457	continue;
458
459	badness = compute_score(sk: result, net, saddr, sport,
460	daddr, hnum, dif, sdif);
461
462	}
463	}
464	return result;
465	}
466
467	/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
468	* harder than this. -DaveM
469	*/
470	struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
471	__be16 sport, __be32 daddr, __be16 dport, int dif,
472	int sdif, struct udp_table *udptable, struct sk_buff *skb)
473	{
474	unsigned short hnum = ntohs(dport);
475	unsigned int hash2, slot2;
476	struct udp_hslot *hslot2;
477	struct sock *result, *sk;
478
479	hash2 = ipv4_portaddr_hash(net, saddr: daddr, port: hnum);
480	slot2 = hash2 & udptable->mask;
481	hslot2 = &udptable->hash2[slot2];
482
483	/* Lookup connected or non-wildcard socket */
484	result = udp4_lib_lookup2(net, saddr, sport,
485	daddr, hnum, dif, sdif,
486	hslot2, skb);
487	if (!IS_ERR_OR_NULL(ptr: result) && result->sk_state == TCP_ESTABLISHED)
488	goto done;
489
490	/* Lookup redirect from BPF */
491	if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
492	udptable == net->ipv4.udp_table) {
493	sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, doff: sizeof(struct udphdr),
494	saddr, sport, daddr, hnum, dif,
495	ehashfn: udp_ehashfn);
496	if (sk) {
497	result = sk;
498	goto done;
499	}
500	}
501
502	/* Got non-wildcard socket or error on first lookup */
503	if (result)
504	goto done;
505
506	/* Lookup wildcard sockets */
507	hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), port: hnum);
508	slot2 = hash2 & udptable->mask;
509	hslot2 = &udptable->hash2[slot2];
510
511	result = udp4_lib_lookup2(net, saddr, sport,
512	htonl(INADDR_ANY), hnum, dif, sdif,
513	hslot2, skb);
514	done:
515	if (IS_ERR(ptr: result))
516	return NULL;
517	return result;
518	}
519	EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
520
521	static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
522	__be16 sport, __be16 dport,
523	struct udp_table *udptable)
524	{
525	const struct iphdr *iph = ip_hdr(skb);
526
527	return __udp4_lib_lookup(dev_net(dev: skb->dev), iph->saddr, sport,
528	iph->daddr, dport, inet_iif(skb),
529	inet_sdif(skb), udptable, skb);
530	}
531
532	struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
533	__be16 sport, __be16 dport)
534	{
535	const struct iphdr *iph = ip_hdr(skb);
536	struct net *net = dev_net(dev: skb->dev);
537	int iif, sdif;
538
539	inet_get_iif_sdif(skb, iif: &iif, sdif: &sdif);
540
541	return __udp4_lib_lookup(net, iph->saddr, sport,
542	iph->daddr, dport, iif,
543	sdif, net->ipv4.udp_table, NULL);
544	}
545
546	/* Must be called under rcu_read_lock().
547	* Does increment socket refcount.
548	*/
549	#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
550	struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
551	__be32 daddr, __be16 dport, int dif)
552	{
553	struct sock *sk;
554
555	sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
556	dif, 0, net->ipv4.udp_table, NULL);
557	if (sk && !refcount_inc_not_zero(r: &sk->sk_refcnt))
558	sk = NULL;
559	return sk;
560	}
561	EXPORT_SYMBOL_GPL(udp4_lib_lookup);
562	#endif
563
564	static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
565	__be16 loc_port, __be32 loc_addr,
566	__be16 rmt_port, __be32 rmt_addr,
567	int dif, int sdif, unsigned short hnum)
568	{
569	const struct inet_sock *inet = inet_sk(sk);
570
571	if (!net_eq(net1: sock_net(sk), net2: net) ||
572	udp_sk(sk)->udp_port_hash != hnum ||
573	(inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
574	(inet->inet_dport != rmt_port && inet->inet_dport) ||
575	(inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
576	ipv6_only_sock(sk) ||
577	!udp_sk_bound_dev_eq(net, bound_dev_if: sk->sk_bound_dev_if, dif, sdif))
578	return false;
579	if (!ip_mc_sf_allow(sk, local: loc_addr, rmt: rmt_addr, dif, sdif))
580	return false;
581	return true;
582	}
583
584	DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
585	EXPORT_SYMBOL(udp_encap_needed_key);
586
587	#if IS_ENABLED(CONFIG_IPV6)
588	DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
589	EXPORT_SYMBOL(udpv6_encap_needed_key);
590	#endif
591
592	void udp_encap_enable(void)
593	{
594	static_branch_inc(&udp_encap_needed_key);
595	}
596	EXPORT_SYMBOL(udp_encap_enable);
597
598	void udp_encap_disable(void)
599	{
600	static_branch_dec(&udp_encap_needed_key);
601	}
602	EXPORT_SYMBOL(udp_encap_disable);
603
604	/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
605	* through error handlers in encapsulations looking for a match.
606	*/
607	static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
608	{
609	int i;
610
611	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
612	int (*handler)(struct sk_buff *skb, u32 info);
613	const struct ip_tunnel_encap_ops *encap;
614
615	encap = rcu_dereference(iptun_encaps[i]);
616	if (!encap)
617	continue;
618	handler = encap->err_handler;
619	if (handler && !handler(skb, info))
620	return 0;
621	}
622
623	return -ENOENT;
624	}
625
626	/* Try to match ICMP errors to UDP tunnels by looking up a socket without
627	* reversing source and destination port: this will match tunnels that force the
628	* same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
629	* lwtunnels might actually break this assumption by being configured with
630	* different destination ports on endpoints, in this case we won't be able to
631	* trace ICMP messages back to them.
632	*
633	* If this doesn't match any socket, probe tunnels with arbitrary destination
634	* ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
635	* we've sent packets to won't necessarily match the local destination port.
636	*
637	* Then ask the tunnel implementation to match the error against a valid
638	* association.
639	*
640	* Return an error if we can't find a match, the socket if we need further
641	* processing, zero otherwise.
642	*/
643	static struct sock *__udp4_lib_err_encap(struct net *net,
644	const struct iphdr *iph,
645	struct udphdr *uh,
646	struct udp_table *udptable,
647	struct sock *sk,
648	struct sk_buff *skb, u32 info)
649	{
650	int (*lookup)(struct sock *sk, struct sk_buff *skb);
651	int network_offset, transport_offset;
652	struct udp_sock *up;
653
654	network_offset = skb_network_offset(skb);
655	transport_offset = skb_transport_offset(skb);
656
657	/* Network header needs to point to the outer IPv4 header inside ICMP */
658	skb_reset_network_header(skb);
659
660	/* Transport header needs to point to the UDP header */
661	skb_set_transport_header(skb, offset: iph->ihl << 2);
662
663	if (sk) {
664	up = udp_sk(sk);
665
666	lookup = READ_ONCE(up->encap_err_lookup);
667	if (lookup && lookup(sk, skb))
668	sk = NULL;
669
670	goto out;
671	}
672
673	sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
674	iph->saddr, uh->dest, skb->dev->ifindex, 0,
675	udptable, NULL);
676	if (sk) {
677	up = udp_sk(sk);
678
679	lookup = READ_ONCE(up->encap_err_lookup);
680	if (!lookup || lookup(sk, skb))
681	sk = NULL;
682	}
683
684	out:
685	if (!sk)
686	sk = ERR_PTR(error: __udp4_lib_err_encap_no_sk(skb, info));
687
688	skb_set_transport_header(skb, offset: transport_offset);
689	skb_set_network_header(skb, offset: network_offset);
690
691	return sk;
692	}
693
694	/*
695	* This routine is called by the ICMP module when it gets some
696	* sort of error condition. If err < 0 then the socket should
697	* be closed and the error returned to the user. If err > 0
698	* it's just the icmp type << 8 | icmp code.
699	* Header points to the ip header of the error packet. We move
700	* on past this. Then (as it used to claim before adjustment)
701	* header points to the first 8 bytes of the udp header. We need
702	* to find the appropriate port.
703	*/
704
705	int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
706	{
707	struct inet_sock *inet;
708	const struct iphdr *iph = (const struct iphdr *)skb->data;
709	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
710	const int type = icmp_hdr(skb)->type;
711	const int code = icmp_hdr(skb)->code;
712	bool tunnel = false;
713	struct sock *sk;
714	int harderr;
715	int err;
716	struct net *net = dev_net(dev: skb->dev);
717
718	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
719	iph->saddr, uh->source, skb->dev->ifindex,
720	inet_sdif(skb), udptable, NULL);
721
722	if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
723	/* No socket for error: try tunnels before discarding */
724	if (static_branch_unlikely(&udp_encap_needed_key)) {
725	sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
726	info);
727	if (!sk)
728	return 0;
729	} else
730	sk = ERR_PTR(error: -ENOENT);
731
732	if (IS_ERR(ptr: sk)) {
733	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
734	return PTR_ERR(ptr: sk);
735	}
736
737	tunnel = true;
738	}
739
740	err = 0;
741	harderr = 0;
742	inet = inet_sk(sk);
743
744	switch (type) {
745	default:
746	case ICMP_TIME_EXCEEDED:
747	err = EHOSTUNREACH;
748	break;
749	case ICMP_SOURCE_QUENCH:
750	goto out;
751	case ICMP_PARAMETERPROB:
752	err = EPROTO;
753	harderr = 1;
754	break;
755	case ICMP_DEST_UNREACH:
756	if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
757	ipv4_sk_update_pmtu(skb, sk, mtu: info);
758	if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
759	err = EMSGSIZE;
760	harderr = 1;
761	break;
762	}
763	goto out;
764	}
765	err = EHOSTUNREACH;
766	if (code <= NR_ICMP_UNREACH) {
767	harderr = icmp_err_convert[code].fatal;
768	err = icmp_err_convert[code].errno;
769	}
770	break;
771	case ICMP_REDIRECT:
772	ipv4_sk_redirect(skb, sk);
773	goto out;
774	}
775
776	/*
777	* RFC1122: OK. Passes ICMP errors back to application, as per
778	* 4.1.3.3.
779	*/
780	if (tunnel) {
781	/* ...not for tunnels though: we don't have a sending socket */
782	if (udp_sk(sk)->encap_err_rcv)
783	udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
784	(u8 *)(uh+1));
785	goto out;
786	}
787	if (!inet_test_bit(RECVERR, sk)) {
788	if (!harderr || sk->sk_state != TCP_ESTABLISHED)
789	goto out;
790	} else
791	ip_icmp_error(sk, skb, err, port: uh->dest, info, payload: (u8 *)(uh+1));
792
793	sk->sk_err = err;
794	sk_error_report(sk);
795	out:
796	return 0;
797	}
798
799	int udp_err(struct sk_buff *skb, u32 info)
800	{
801	return __udp4_lib_err(skb, info, udptable: dev_net(dev: skb->dev)->ipv4.udp_table);
802	}
803
804	/*
805	* Throw away all pending data and cancel the corking. Socket is locked.
806	*/
807	void udp_flush_pending_frames(struct sock *sk)
808	{
809	struct udp_sock *up = udp_sk(sk);
810
811	if (up->pending) {
812	up->len = 0;
813	WRITE_ONCE(up->pending, 0);
814	ip_flush_pending_frames(sk);
815	}
816	}
817	EXPORT_SYMBOL(udp_flush_pending_frames);
818
819	/**
820	* udp4_hwcsum - handle outgoing HW checksumming
821	* @skb: sk_buff containing the filled-in UDP header
822	* (checksum field must be zeroed out)
823	* @src: source IP address
824	* @dst: destination IP address
825	*/
826	void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
827	{
828	struct udphdr *uh = udp_hdr(skb);
829	int offset = skb_transport_offset(skb);
830	int len = skb->len - offset;
831	int hlen = len;
832	__wsum csum = 0;
833
834	if (!skb_has_frag_list(skb)) {
835	/*
836	* Only one fragment on the socket.
837	*/
838	skb->csum_start = skb_transport_header(skb) - skb->head;
839	skb->csum_offset = offsetof(struct udphdr, check);
840	uh->check = ~csum_tcpudp_magic(saddr: src, daddr: dst, len,
841	IPPROTO_UDP, sum: 0);
842	} else {
843	struct sk_buff *frags;
844
845	/*
846	* HW-checksum won't work as there are two or more
847	* fragments on the socket so that all csums of sk_buffs
848	* should be together
849	*/
850	skb_walk_frags(skb, frags) {
851	csum = csum_add(csum, addend: frags->csum);
852	hlen -= frags->len;
853	}
854
855	csum = skb_checksum(skb, offset, len: hlen, csum);
856	skb->ip_summed = CHECKSUM_NONE;
857
858	uh->check = csum_tcpudp_magic(saddr: src, daddr: dst, len, IPPROTO_UDP, sum: csum);
859	if (uh->check == 0)
860	uh->check = CSUM_MANGLED_0;
861	}
862	}
863	EXPORT_SYMBOL_GPL(udp4_hwcsum);
864
865	/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
866	* for the simple case like when setting the checksum for a UDP tunnel.
867	*/
868	void udp_set_csum(bool nocheck, struct sk_buff *skb,
869	__be32 saddr, __be32 daddr, int len)
870	{
871	struct udphdr *uh = udp_hdr(skb);
872
873	if (nocheck) {
874	uh->check = 0;
875	} else if (skb_is_gso(skb)) {
876	uh->check = ~udp_v4_check(len, saddr, daddr, base: 0);
877	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
878	uh->check = 0;
879	uh->check = udp_v4_check(len, saddr, daddr, base: lco_csum(skb));
880	if (uh->check == 0)
881	uh->check = CSUM_MANGLED_0;
882	} else {
883	skb->ip_summed = CHECKSUM_PARTIAL;
884	skb->csum_start = skb_transport_header(skb) - skb->head;
885	skb->csum_offset = offsetof(struct udphdr, check);
886	uh->check = ~udp_v4_check(len, saddr, daddr, base: 0);
887	}
888	}
889	EXPORT_SYMBOL(udp_set_csum);
890
891	static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
892	struct inet_cork *cork)
893	{
894	struct sock *sk = skb->sk;
895	struct inet_sock *inet = inet_sk(sk);
896	struct udphdr *uh;
897	int err;
898	int is_udplite = IS_UDPLITE(sk);
899	int offset = skb_transport_offset(skb);
900	int len = skb->len - offset;
901	int datalen = len - sizeof(*uh);
902	__wsum csum = 0;
903
904	/*
905	* Create a UDP header
906	*/
907	uh = udp_hdr(skb);
908	uh->source = inet->inet_sport;
909	uh->dest = fl4->fl4_dport;
910	uh->len = htons(len);
911	uh->check = 0;
912
913	if (cork->gso_size) {
914	const int hlen = skb_network_header_len(skb) +
915	sizeof(struct udphdr);
916
917	if (hlen + cork->gso_size > cork->fragsize) {
918	kfree_skb(skb);
919	return -EINVAL;
920	}
921	if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
922	kfree_skb(skb);
923	return -EINVAL;
924	}
925	if (sk->sk_no_check_tx) {
926	kfree_skb(skb);
927	return -EINVAL;
928	}
929	if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
930	dst_xfrm(dst: skb_dst(skb))) {
931	kfree_skb(skb);
932	return -EIO;
933	}
934
935	if (datalen > cork->gso_size) {
936	skb_shinfo(skb)->gso_size = cork->gso_size;
937	skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
938	skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
939	cork->gso_size);
940	}
941	goto csum_partial;
942	}
943
944	if (is_udplite) /* UDP-Lite */
945	csum = udplite_csum(skb);
946
947	else if (sk->sk_no_check_tx) { /* UDP csum off */
948
949	skb->ip_summed = CHECKSUM_NONE;
950	goto send;
951
952	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
953	csum_partial:
954
955	udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
956	goto send;
957
958	} else
959	csum = udp_csum(skb);
960
961	/* add protocol-dependent pseudo-header */
962	uh->check = csum_tcpudp_magic(saddr: fl4->saddr, daddr: fl4->daddr, len,
963	proto: sk->sk_protocol, sum: csum);
964	if (uh->check == 0)
965	uh->check = CSUM_MANGLED_0;
966
967	send:
968	err = ip_send_skb(net: sock_net(sk), skb);
969	if (err) {
970	if (err == -ENOBUFS &&
971	!inet_test_bit(RECVERR, sk)) {
972	UDP_INC_STATS(sock_net(sk),
973	UDP_MIB_SNDBUFERRORS, is_udplite);
974	err = 0;
975	}
976	} else
977	UDP_INC_STATS(sock_net(sk),
978	UDP_MIB_OUTDATAGRAMS, is_udplite);
979	return err;
980	}
981
982	/*
983	* Push out all pending data as one UDP datagram. Socket is locked.
984	*/
985	int udp_push_pending_frames(struct sock *sk)
986	{
987	struct udp_sock *up = udp_sk(sk);
988	struct inet_sock *inet = inet_sk(sk);
989	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
990	struct sk_buff *skb;
991	int err = 0;
992
993	skb = ip_finish_skb(sk, fl4);
994	if (!skb)
995	goto out;
996
997	err = udp_send_skb(skb, fl4, cork: &inet->cork.base);
998
999	out:
1000	up->len = 0;
1001	WRITE_ONCE(up->pending, 0);
1002	return err;
1003	}
1004	EXPORT_SYMBOL(udp_push_pending_frames);
1005
1006	static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1007	{
1008	switch (cmsg->cmsg_type) {
1009	case UDP_SEGMENT:
1010	if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1011	return -EINVAL;
1012	*gso_size = *(__u16 *)CMSG_DATA(cmsg);
1013	return 0;
1014	default:
1015	return -EINVAL;
1016	}
1017	}
1018
1019	int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1020	{
1021	struct cmsghdr *cmsg;
1022	bool need_ip = false;
1023	int err;
1024
1025	for_each_cmsghdr(cmsg, msg) {
1026	if (!CMSG_OK(msg, cmsg))
1027	return -EINVAL;
1028
1029	if (cmsg->cmsg_level != SOL_UDP) {
1030	need_ip = true;
1031	continue;
1032	}
1033
1034	err = __udp_cmsg_send(cmsg, gso_size);
1035	if (err)
1036	return err;
1037	}
1038
1039	return need_ip;
1040	}
1041	EXPORT_SYMBOL_GPL(udp_cmsg_send);
1042
1043	int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1044	{
1045	struct inet_sock *inet = inet_sk(sk);
1046	struct udp_sock *up = udp_sk(sk);
1047	DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1048	struct flowi4 fl4_stack;
1049	struct flowi4 *fl4;
1050	int ulen = len;
1051	struct ipcm_cookie ipc;
1052	struct rtable *rt = NULL;
1053	int free = 0;
1054	int connected = 0;
1055	__be32 daddr, faddr, saddr;
1056	u8 tos, scope;
1057	__be16 dport;
1058	int err, is_udplite = IS_UDPLITE(sk);
1059	int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1060	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1061	struct sk_buff *skb;
1062	struct ip_options_data opt_copy;
1063	int uc_index;
1064
1065	if (len > 0xFFFF)
1066	return -EMSGSIZE;
1067
1068	/*
1069	* Check the flags.
1070	*/
1071
1072	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1073	return -EOPNOTSUPP;
1074
1075	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1076
1077	fl4 = &inet->cork.fl.u.ip4;
1078	if (READ_ONCE(up->pending)) {
1079	/*
1080	* There are pending frames.
1081	* The socket lock must be held while it's corked.
1082	*/
1083	lock_sock(sk);
1084	if (likely(up->pending)) {
1085	if (unlikely(up->pending != AF_INET)) {
1086	release_sock(sk);
1087	return -EINVAL;
1088	}
1089	goto do_append_data;
1090	}
1091	release_sock(sk);
1092	}
1093	ulen += sizeof(struct udphdr);
1094
1095	/*
1096	* Get and verify the address.
1097	*/
1098	if (usin) {
1099	if (msg->msg_namelen < sizeof(*usin))
1100	return -EINVAL;
1101	if (usin->sin_family != AF_INET) {
1102	if (usin->sin_family != AF_UNSPEC)
1103	return -EAFNOSUPPORT;
1104	}
1105
1106	daddr = usin->sin_addr.s_addr;
1107	dport = usin->sin_port;
1108	if (dport == 0)
1109	return -EINVAL;
1110	} else {
1111	if (sk->sk_state != TCP_ESTABLISHED)
1112	return -EDESTADDRREQ;
1113	daddr = inet->inet_daddr;
1114	dport = inet->inet_dport;
1115	/* Open fast path for connected socket.
1116	Route will not be used, if at least one option is set.
1117	*/
1118	connected = 1;
1119	}
1120
1121	ipcm_init_sk(ipcm: &ipc, inet);
1122	ipc.gso_size = READ_ONCE(up->gso_size);
1123
1124	if (msg->msg_controllen) {
1125	err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1126	if (err > 0)
1127	err = ip_cmsg_send(sk, msg, ipc: &ipc,
1128	allow_ipv6: sk->sk_family == AF_INET6);
1129	if (unlikely(err < 0)) {
1130	kfree(objp: ipc.opt);
1131	return err;
1132	}
1133	if (ipc.opt)
1134	free = 1;
1135	connected = 0;
1136	}
1137	if (!ipc.opt) {
1138	struct ip_options_rcu *inet_opt;
1139
1140	rcu_read_lock();
1141	inet_opt = rcu_dereference(inet->inet_opt);
1142	if (inet_opt) {
1143	memcpy(&opt_copy, inet_opt,
1144	sizeof(*inet_opt) + inet_opt->opt.optlen);
1145	ipc.opt = &opt_copy.opt;
1146	}
1147	rcu_read_unlock();
1148	}
1149
1150	if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1151	err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1152	(struct sockaddr *)usin,
1153	&msg->msg_namelen,
1154	&ipc.addr);
1155	if (err)
1156	goto out_free;
1157	if (usin) {
1158	if (usin->sin_port == 0) {
1159	/* BPF program set invalid port. Reject it. */
1160	err = -EINVAL;
1161	goto out_free;
1162	}
1163	daddr = usin->sin_addr.s_addr;
1164	dport = usin->sin_port;
1165	}
1166	}
1167
1168	saddr = ipc.addr;
1169	ipc.addr = faddr = daddr;
1170
1171	if (ipc.opt && ipc.opt->opt.srr) {
1172	if (!daddr) {
1173	err = -EINVAL;
1174	goto out_free;
1175	}
1176	faddr = ipc.opt->opt.faddr;
1177	connected = 0;
1178	}
1179	tos = get_rttos(ipc: &ipc, inet);
1180	scope = ip_sendmsg_scope(inet, ipc: &ipc, msg);
1181	if (scope == RT_SCOPE_LINK)
1182	connected = 0;
1183
1184	uc_index = READ_ONCE(inet->uc_index);
1185	if (ipv4_is_multicast(addr: daddr)) {
1186	if (!ipc.oif || netif_index_is_l3_master(net: sock_net(sk), ifindex: ipc.oif))
1187	ipc.oif = READ_ONCE(inet->mc_index);
1188	if (!saddr)
1189	saddr = READ_ONCE(inet->mc_addr);
1190	connected = 0;
1191	} else if (!ipc.oif) {
1192	ipc.oif = uc_index;
1193	} else if (ipv4_is_lbcast(addr: daddr) && uc_index) {
1194	/* oif is set, packet is to local broadcast and
1195	* uc_index is set. oif is most likely set
1196	* by sk_bound_dev_if. If uc_index != oif check if the
1197	* oif is an L3 master and uc_index is an L3 slave.
1198	* If so, we want to allow the send using the uc_index.
1199	*/
1200	if (ipc.oif != uc_index &&
1201	ipc.oif == l3mdev_master_ifindex_by_index(net: sock_net(sk),
1202	ifindex: uc_index)) {
1203	ipc.oif = uc_index;
1204	}
1205	}
1206
1207	if (connected)
1208	rt = (struct rtable *)sk_dst_check(sk, cookie: 0);
1209
1210	if (!rt) {
1211	struct net *net = sock_net(sk);
1212	__u8 flow_flags = inet_sk_flowi_flags(sk);
1213
1214	fl4 = &fl4_stack;
1215
1216	flowi4_init_output(fl4, oif: ipc.oif, mark: ipc.sockc.mark, tos, scope,
1217	proto: sk->sk_protocol, flags: flow_flags, daddr: faddr, saddr,
1218	dport, sport: inet->inet_sport, uid: sk->sk_uid);
1219
1220	security_sk_classify_flow(sk, flic: flowi4_to_flowi_common(fl4));
1221	rt = ip_route_output_flow(net, flp: fl4, sk);
1222	if (IS_ERR(ptr: rt)) {
1223	err = PTR_ERR(ptr: rt);
1224	rt = NULL;
1225	if (err == -ENETUNREACH)
1226	IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1227	goto out;
1228	}
1229
1230	err = -EACCES;
1231	if ((rt->rt_flags & RTCF_BROADCAST) &&
1232	!sock_flag(sk, flag: SOCK_BROADCAST))
1233	goto out;
1234	if (connected)
1235	sk_dst_set(sk, dst: dst_clone(dst: &rt->dst));
1236	}
1237
1238	if (msg->msg_flags&MSG_CONFIRM)
1239	goto do_confirm;
1240	back_from_confirm:
1241
1242	saddr = fl4->saddr;
1243	if (!ipc.addr)
1244	daddr = ipc.addr = fl4->daddr;
1245
1246	/* Lockless fast path for the non-corking case. */
1247	if (!corkreq) {
1248	struct inet_cork cork;
1249
1250	skb = ip_make_skb(sk, fl4, getfrag, from: msg, length: ulen,
1251	transhdrlen: sizeof(struct udphdr), ipc: &ipc, rtp: &rt,
1252	cork: &cork, flags: msg->msg_flags);
1253	err = PTR_ERR(ptr: skb);
1254	if (!IS_ERR_OR_NULL(ptr: skb))
1255	err = udp_send_skb(skb, fl4, cork: &cork);
1256	goto out;
1257	}
1258
1259	lock_sock(sk);
1260	if (unlikely(up->pending)) {
1261	/* The socket is already corked while preparing it. */
1262	/* ... which is an evident application bug. --ANK */
1263	release_sock(sk);
1264
1265	net_dbg_ratelimited("socket already corked\n");
1266	err = -EINVAL;
1267	goto out;
1268	}
1269	/*
1270	* Now cork the socket to pend data.
1271	*/
1272	fl4 = &inet->cork.fl.u.ip4;
1273	fl4->daddr = daddr;
1274	fl4->saddr = saddr;
1275	fl4->fl4_dport = dport;
1276	fl4->fl4_sport = inet->inet_sport;
1277	WRITE_ONCE(up->pending, AF_INET);
1278
1279	do_append_data:
1280	up->len += ulen;
1281	err = ip_append_data(sk, fl4, getfrag, from: msg, len: ulen,
1282	protolen: sizeof(struct udphdr), ipc: &ipc, rt: &rt,
1283	flags: corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1284	if (err)
1285	udp_flush_pending_frames(sk);
1286	else if (!corkreq)
1287	err = udp_push_pending_frames(sk);
1288	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1289	WRITE_ONCE(up->pending, 0);
1290	release_sock(sk);
1291
1292	out:
1293	ip_rt_put(rt);
1294	out_free:
1295	if (free)
1296	kfree(objp: ipc.opt);
1297	if (!err)
1298	return len;
1299	/*
1300	* ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1301	* ENOBUFS might not be good (it's not tunable per se), but otherwise
1302	* we don't have a good statistic (IpOutDiscards but it can be too many
1303	* things). We could add another new stat but at least for now that
1304	* seems like overkill.
1305	*/
1306	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1307	UDP_INC_STATS(sock_net(sk),
1308	UDP_MIB_SNDBUFERRORS, is_udplite);
1309	}
1310	return err;
1311
1312	do_confirm:
1313	if (msg->msg_flags & MSG_PROBE)
1314	dst_confirm_neigh(dst: &rt->dst, daddr: &fl4->daddr);
1315	if (!(msg->msg_flags&MSG_PROBE) || len)
1316	goto back_from_confirm;
1317	err = 0;
1318	goto out;
1319	}
1320	EXPORT_SYMBOL(udp_sendmsg);
1321
1322	void udp_splice_eof(struct socket *sock)
1323	{
1324	struct sock *sk = sock->sk;
1325	struct udp_sock *up = udp_sk(sk);
1326
1327	if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1328	return;
1329
1330	lock_sock(sk);
1331	if (up->pending && !udp_test_bit(CORK, sk))
1332	udp_push_pending_frames(sk);
1333	release_sock(sk);
1334	}
1335	EXPORT_SYMBOL_GPL(udp_splice_eof);
1336
1337	#define UDP_SKB_IS_STATELESS 0x80000000
1338
1339	/* all head states (dst, sk, nf conntrack) except skb extensions are
1340	* cleared by udp_rcv().
1341	*
1342	* We need to preserve secpath, if present, to eventually process
1343	* IP_CMSG_PASSSEC at recvmsg() time.
1344	*
1345	* Other extensions can be cleared.
1346	*/
1347	static bool udp_try_make_stateless(struct sk_buff *skb)
1348	{
1349	if (!skb_has_extensions(skb))
1350	return true;
1351
1352	if (!secpath_exists(skb)) {
1353	skb_ext_reset(skb);
1354	return true;
1355	}
1356
1357	return false;
1358	}
1359
1360	static void udp_set_dev_scratch(struct sk_buff *skb)
1361	{
1362	struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1363
1364	BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1365	scratch->_tsize_state = skb->truesize;
1366	#if BITS_PER_LONG == 64
1367	scratch->len = skb->len;
1368	scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1369	scratch->is_linear = !skb_is_nonlinear(skb);
1370	#endif
1371	if (udp_try_make_stateless(skb))
1372	scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1373	}
1374
1375	static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1376	{
1377	/* We come here after udp_lib_checksum_complete() returned 0.
1378	* This means that __skb_checksum_complete() might have
1379	* set skb->csum_valid to 1.
1380	* On 64bit platforms, we can set csum_unnecessary
1381	* to true, but only if the skb is not shared.
1382	*/
1383	#if BITS_PER_LONG == 64
1384	if (!skb_shared(skb))
1385	udp_skb_scratch(skb)->csum_unnecessary = true;
1386	#endif
1387	}
1388
1389	static int udp_skb_truesize(struct sk_buff *skb)
1390	{
1391	return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1392	}
1393
1394	static bool udp_skb_has_head_state(struct sk_buff *skb)
1395	{
1396	return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1397	}
1398
1399	/* fully reclaim rmem/fwd memory allocated for skb */
1400	static void udp_rmem_release(struct sock *sk, int size, int partial,
1401	bool rx_queue_lock_held)
1402	{
1403	struct udp_sock *up = udp_sk(sk);
1404	struct sk_buff_head *sk_queue;
1405	int amt;
1406
1407	if (likely(partial)) {
1408	up->forward_deficit += size;
1409	size = up->forward_deficit;
1410	if (size < READ_ONCE(up->forward_threshold) &&
1411	!skb_queue_empty(list: &up->reader_queue))
1412	return;
1413	} else {
1414	size += up->forward_deficit;
1415	}
1416	up->forward_deficit = 0;
1417
1418	/* acquire the sk_receive_queue for fwd allocated memory scheduling,
1419	* if the called don't held it already
1420	*/
1421	sk_queue = &sk->sk_receive_queue;
1422	if (!rx_queue_lock_held)
1423	spin_lock(lock: &sk_queue->lock);
1424
1425
1426	sk_forward_alloc_add(sk, val: size);
1427	amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1428	sk_forward_alloc_add(sk, val: -amt);
1429
1430	if (amt)
1431	__sk_mem_reduce_allocated(sk, amount: amt >> PAGE_SHIFT);
1432
1433	atomic_sub(i: size, v: &sk->sk_rmem_alloc);
1434
1435	/* this can save us from acquiring the rx queue lock on next receive */
1436	skb_queue_splice_tail_init(list: sk_queue, head: &up->reader_queue);
1437
1438	if (!rx_queue_lock_held)
1439	spin_unlock(lock: &sk_queue->lock);
1440	}
1441
1442	/* Note: called with reader_queue.lock held.
1443	* Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1444	* This avoids a cache line miss while receive_queue lock is held.
1445	* Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1446	*/
1447	void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1448	{
1449	prefetch(&skb->data);
1450	udp_rmem_release(sk, size: udp_skb_truesize(skb), partial: 1, rx_queue_lock_held: false);
1451	}
1452	EXPORT_SYMBOL(udp_skb_destructor);
1453
1454	/* as above, but the caller held the rx queue lock, too */
1455	static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1456	{
1457	prefetch(&skb->data);
1458	udp_rmem_release(sk, size: udp_skb_truesize(skb), partial: 1, rx_queue_lock_held: true);
1459	}
1460
1461	/* Idea of busylocks is to let producers grab an extra spinlock
1462	* to relieve pressure on the receive_queue spinlock shared by consumer.
1463	* Under flood, this means that only one producer can be in line
1464	* trying to acquire the receive_queue spinlock.
1465	* These busylock can be allocated on a per cpu manner, instead of a
1466	* per socket one (that would consume a cache line per socket)
1467	*/
1468	static int udp_busylocks_log __read_mostly;
1469	static spinlock_t *udp_busylocks __read_mostly;
1470
1471	static spinlock_t *busylock_acquire(void *ptr)
1472	{
1473	spinlock_t *busy;
1474
1475	busy = udp_busylocks + hash_ptr(ptr, bits: udp_busylocks_log);
1476	spin_lock(lock: busy);
1477	return busy;
1478	}
1479
1480	static void busylock_release(spinlock_t *busy)
1481	{
1482	if (busy)
1483	spin_unlock(lock: busy);
1484	}
1485
1486	static int udp_rmem_schedule(struct sock *sk, int size)
1487	{
1488	int delta;
1489
1490	delta = size - sk->sk_forward_alloc;
1491	if (delta > 0 && !__sk_mem_schedule(sk, size: delta, SK_MEM_RECV))
1492	return -ENOBUFS;
1493
1494	return 0;
1495	}
1496
1497	int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1498	{
1499	struct sk_buff_head *list = &sk->sk_receive_queue;
1500	int rmem, err = -ENOMEM;
1501	spinlock_t *busy = NULL;
1502	int size;
1503
1504	/* try to avoid the costly atomic add/sub pair when the receive
1505	* queue is full; always allow at least a packet
1506	*/
1507	rmem = atomic_read(v: &sk->sk_rmem_alloc);
1508	if (rmem > sk->sk_rcvbuf)
1509	goto drop;
1510
1511	/* Under mem pressure, it might be helpful to help udp_recvmsg()
1512	* having linear skbs :
1513	* - Reduce memory overhead and thus increase receive queue capacity
1514	* - Less cache line misses at copyout() time
1515	* - Less work at consume_skb() (less alien page frag freeing)
1516	*/
1517	if (rmem > (sk->sk_rcvbuf >> 1)) {
1518	skb_condense(skb);
1519
1520	busy = busylock_acquire(ptr: sk);
1521	}
1522	size = skb->truesize;
1523	udp_set_dev_scratch(skb);
1524
1525	/* we drop only if the receive buf is full and the receive
1526	* queue contains some other skb
1527	*/
1528	rmem = atomic_add_return(i: size, v: &sk->sk_rmem_alloc);
1529	if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1530	goto uncharge_drop;
1531
1532	spin_lock(lock: &list->lock);
1533	err = udp_rmem_schedule(sk, size);
1534	if (err) {
1535	spin_unlock(lock: &list->lock);
1536	goto uncharge_drop;
1537	}
1538
1539	sk_forward_alloc_add(sk, val: -size);
1540
1541	/* no need to setup a destructor, we will explicitly release the
1542	* forward allocated memory on dequeue
1543	*/
1544	sock_skb_set_dropcount(sk, skb);
1545
1546	__skb_queue_tail(list, newsk: skb);
1547	spin_unlock(lock: &list->lock);
1548
1549	if (!sock_flag(sk, flag: SOCK_DEAD))
1550	INDIRECT_CALL_1(sk->sk_data_ready, sock_def_readable, sk);
1551
1552	busylock_release(busy);
1553	return 0;
1554
1555	uncharge_drop:
1556	atomic_sub(i: skb->truesize, v: &sk->sk_rmem_alloc);
1557
1558	drop:
1559	atomic_inc(v: &sk->sk_drops);
1560	busylock_release(busy);
1561	return err;
1562	}
1563	EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1564
1565	void udp_destruct_common(struct sock *sk)
1566	{
1567	/* reclaim completely the forward allocated memory */
1568	struct udp_sock *up = udp_sk(sk);
1569	unsigned int total = 0;
1570	struct sk_buff *skb;
1571
1572	skb_queue_splice_tail_init(list: &sk->sk_receive_queue, head: &up->reader_queue);
1573	while ((skb = __skb_dequeue(list: &up->reader_queue)) != NULL) {
1574	total += skb->truesize;
1575	kfree_skb(skb);
1576	}
1577	udp_rmem_release(sk, size: total, partial: 0, rx_queue_lock_held: true);
1578	}
1579	EXPORT_SYMBOL_GPL(udp_destruct_common);
1580
1581	static void udp_destruct_sock(struct sock *sk)
1582	{
1583	udp_destruct_common(sk);
1584	inet_sock_destruct(sk);
1585	}
1586
1587	int udp_init_sock(struct sock *sk)
1588	{
1589	udp_lib_init_sock(sk);
1590	sk->sk_destruct = udp_destruct_sock;
1591	set_bit(SOCK_SUPPORT_ZC, addr: &sk->sk_socket->flags);
1592	return 0;
1593	}
1594
1595	void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1596	{
1597	if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset)))
1598	sk_peek_offset_bwd(sk, val: len);
1599
1600	if (!skb_unref(skb))
1601	return;
1602
1603	/* In the more common cases we cleared the head states previously,
1604	* see __udp_queue_rcv_skb().
1605	*/
1606	if (unlikely(udp_skb_has_head_state(skb)))
1607	skb_release_head_state(skb);
1608	__consume_stateless_skb(skb);
1609	}
1610	EXPORT_SYMBOL_GPL(skb_consume_udp);
1611
1612	static struct sk_buff *__first_packet_length(struct sock *sk,
1613	struct sk_buff_head *rcvq,
1614	int *total)
1615	{
1616	struct sk_buff *skb;
1617
1618	while ((skb = skb_peek(list_: rcvq)) != NULL) {
1619	if (udp_lib_checksum_complete(skb)) {
1620	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1621	IS_UDPLITE(sk));
1622	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1623	IS_UDPLITE(sk));
1624	atomic_inc(v: &sk->sk_drops);
1625	__skb_unlink(skb, list: rcvq);
1626	*total += skb->truesize;
1627	kfree_skb(skb);
1628	} else {
1629	udp_skb_csum_unnecessary_set(skb);
1630	break;
1631	}
1632	}
1633	return skb;
1634	}
1635
1636	/**
1637	* first_packet_length - return length of first packet in receive queue
1638	* @sk: socket
1639	*
1640	* Drops all bad checksum frames, until a valid one is found.
1641	* Returns the length of found skb, or -1 if none is found.
1642	*/
1643	static int first_packet_length(struct sock *sk)
1644	{
1645	struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1646	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1647	struct sk_buff *skb;
1648	int total = 0;
1649	int res;
1650
1651	spin_lock_bh(lock: &rcvq->lock);
1652	skb = __first_packet_length(sk, rcvq, total: &total);
1653	if (!skb && !skb_queue_empty_lockless(list: sk_queue)) {
1654	spin_lock(lock: &sk_queue->lock);
1655	skb_queue_splice_tail_init(list: sk_queue, head: rcvq);
1656	spin_unlock(lock: &sk_queue->lock);
1657
1658	skb = __first_packet_length(sk, rcvq, total: &total);
1659	}
1660	res = skb ? skb->len : -1;
1661	if (total)
1662	udp_rmem_release(sk, size: total, partial: 1, rx_queue_lock_held: false);
1663	spin_unlock_bh(lock: &rcvq->lock);
1664	return res;
1665	}
1666
1667	/*
1668	* IOCTL requests applicable to the UDP protocol
1669	*/
1670
1671	int udp_ioctl(struct sock *sk, int cmd, int *karg)
1672	{
1673	switch (cmd) {
1674	case SIOCOUTQ:
1675	{
1676	*karg = sk_wmem_alloc_get(sk);
1677	return 0;
1678	}
1679
1680	case SIOCINQ:
1681	{
1682	*karg = max_t(int, 0, first_packet_length(sk));
1683	return 0;
1684	}
1685
1686	default:
1687	return -ENOIOCTLCMD;
1688	}
1689
1690	return 0;
1691	}
1692	EXPORT_SYMBOL(udp_ioctl);
1693
1694	struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1695	int *off, int *err)
1696	{
1697	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1698	struct sk_buff_head *queue;
1699	struct sk_buff *last;
1700	long timeo;
1701	int error;
1702
1703	queue = &udp_sk(sk)->reader_queue;
1704	timeo = sock_rcvtimeo(sk, noblock: flags & MSG_DONTWAIT);
1705	do {
1706	struct sk_buff *skb;
1707
1708	error = sock_error(sk);
1709	if (error)
1710	break;
1711
1712	error = -EAGAIN;
1713	do {
1714	spin_lock_bh(lock: &queue->lock);
1715	skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1716	err, last: &last);
1717	if (skb) {
1718	if (!(flags & MSG_PEEK))
1719	udp_skb_destructor(sk, skb);
1720	spin_unlock_bh(lock: &queue->lock);
1721	return skb;
1722	}
1723
1724	if (skb_queue_empty_lockless(list: sk_queue)) {
1725	spin_unlock_bh(lock: &queue->lock);
1726	goto busy_check;
1727	}
1728
1729	/* refill the reader queue and walk it again
1730	* keep both queues locked to avoid re-acquiring
1731	* the sk_receive_queue lock if fwd memory scheduling
1732	* is needed.
1733	*/
1734	spin_lock(lock: &sk_queue->lock);
1735	skb_queue_splice_tail_init(list: sk_queue, head: queue);
1736
1737	skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1738	err, last: &last);
1739	if (skb && !(flags & MSG_PEEK))
1740	udp_skb_dtor_locked(sk, skb);
1741	spin_unlock(lock: &sk_queue->lock);
1742	spin_unlock_bh(lock: &queue->lock);
1743	if (skb)
1744	return skb;
1745
1746	busy_check:
1747	if (!sk_can_busy_loop(sk))
1748	break;
1749
1750	sk_busy_loop(sk, nonblock: flags & MSG_DONTWAIT);
1751	} while (!skb_queue_empty_lockless(list: sk_queue));
1752
1753	/* sk_queue is empty, reader_queue may contain peeked packets */
1754	} while (timeo &&
1755	!__skb_wait_for_more_packets(sk, queue: &sk->sk_receive_queue,
1756	err: &error, timeo_p: &timeo,
1757	skb: (struct sk_buff *)sk_queue));
1758
1759	*err = error;
1760	return NULL;
1761	}
1762	EXPORT_SYMBOL(__skb_recv_udp);
1763
1764	int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1765	{
1766	struct sk_buff *skb;
1767	int err;
1768
1769	try_again:
1770	skb = skb_recv_udp(sk, MSG_DONTWAIT, err: &err);
1771	if (!skb)
1772	return err;
1773
1774	if (udp_lib_checksum_complete(skb)) {
1775	int is_udplite = IS_UDPLITE(sk);
1776	struct net *net = sock_net(sk);
1777
1778	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1779	__UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1780	atomic_inc(v: &sk->sk_drops);
1781	kfree_skb(skb);
1782	goto try_again;
1783	}
1784
1785	WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1786	return recv_actor(sk, skb);
1787	}
1788	EXPORT_SYMBOL(udp_read_skb);
1789
1790	/*
1791	* This should be easy, if there is something there we
1792	* return it, otherwise we block.
1793	*/
1794
1795	int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1796	int *addr_len)
1797	{
1798	struct inet_sock *inet = inet_sk(sk);
1799	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1800	struct sk_buff *skb;
1801	unsigned int ulen, copied;
1802	int off, err, peeking = flags & MSG_PEEK;
1803	int is_udplite = IS_UDPLITE(sk);
1804	bool checksum_valid = false;
1805
1806	if (flags & MSG_ERRQUEUE)
1807	return ip_recv_error(sk, msg, len, addr_len);
1808
1809	try_again:
1810	off = sk_peek_offset(sk, flags);
1811	skb = __skb_recv_udp(sk, flags, &off, &err);
1812	if (!skb)
1813	return err;
1814
1815	ulen = udp_skb_len(skb);
1816	copied = len;
1817	if (copied > ulen - off)
1818	copied = ulen - off;
1819	else if (copied < ulen)
1820	msg->msg_flags |= MSG_TRUNC;
1821
1822	/*
1823	* If checksum is needed at all, try to do it while copying the
1824	* data. If the data is truncated, or if we only want a partial
1825	* coverage checksum (UDP-Lite), do it before the copy.
1826	*/
1827
1828	if (copied < ulen || peeking ||
1829	(is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1830	checksum_valid = udp_skb_csum_unnecessary(skb) ||
1831	!__udp_lib_checksum_complete(skb);
1832	if (!checksum_valid)
1833	goto csum_copy_err;
1834	}
1835
1836	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1837	if (udp_skb_is_linear(skb))
1838	err = copy_linear_skb(skb, len: copied, off, to: &msg->msg_iter);
1839	else
1840	err = skb_copy_datagram_msg(from: skb, offset: off, msg, size: copied);
1841	} else {
1842	err = skb_copy_and_csum_datagram_msg(skb, hlen: off, msg);
1843
1844	if (err == -EINVAL)
1845	goto csum_copy_err;
1846	}
1847
1848	if (unlikely(err)) {
1849	if (!peeking) {
1850	atomic_inc(v: &sk->sk_drops);
1851	UDP_INC_STATS(sock_net(sk),
1852	UDP_MIB_INERRORS, is_udplite);
1853	}
1854	kfree_skb(skb);
1855	return err;
1856	}
1857
1858	if (!peeking)
1859	UDP_INC_STATS(sock_net(sk),
1860	UDP_MIB_INDATAGRAMS, is_udplite);
1861
1862	sock_recv_cmsgs(msg, sk, skb);
1863
1864	/* Copy the address. */
1865	if (sin) {
1866	sin->sin_family = AF_INET;
1867	sin->sin_port = udp_hdr(skb)->source;
1868	sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1869	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1870	*addr_len = sizeof(*sin);
1871
1872	BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1873	(struct sockaddr *)sin,
1874	addr_len);
1875	}
1876
1877	if (udp_test_bit(GRO_ENABLED, sk))
1878	udp_cmsg_recv(msg, sk, skb);
1879
1880	if (inet_cmsg_flags(inet))
1881	ip_cmsg_recv_offset(msg, sk, skb, tlen: sizeof(struct udphdr), offset: off);
1882
1883	err = copied;
1884	if (flags & MSG_TRUNC)
1885	err = ulen;
1886
1887	skb_consume_udp(sk, skb, peeking ? -err : err);
1888	return err;
1889
1890	csum_copy_err:
1891	if (!__sk_queue_drop_skb(sk, sk_queue: &udp_sk(sk)->reader_queue, skb, flags,
1892	destructor: udp_skb_destructor)) {
1893	UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1894	UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1895	}
1896	kfree_skb(skb);
1897
1898	/* starting over for a new packet, but check if we need to yield */
1899	cond_resched();
1900	msg->msg_flags &= ~MSG_TRUNC;
1901	goto try_again;
1902	}
1903
1904	int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1905	{
1906	/* This check is replicated from __ip4_datagram_connect() and
1907	* intended to prevent BPF program called below from accessing bytes
1908	* that are out of the bound specified by user in addr_len.
1909	*/
1910	if (addr_len < sizeof(struct sockaddr_in))
1911	return -EINVAL;
1912
1913	return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
1914	}
1915	EXPORT_SYMBOL(udp_pre_connect);
1916
1917	int __udp_disconnect(struct sock *sk, int flags)
1918	{
1919	struct inet_sock *inet = inet_sk(sk);
1920	/*
1921	* 1003.1g - break association.
1922	*/
1923
1924	sk->sk_state = TCP_CLOSE;
1925	inet->inet_daddr = 0;
1926	inet->inet_dport = 0;
1927	sock_rps_reset_rxhash(sk);
1928	sk->sk_bound_dev_if = 0;
1929	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1930	inet_reset_saddr(sk);
1931	if (sk->sk_prot->rehash &&
1932	(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1933	sk->sk_prot->rehash(sk);
1934	}
1935
1936	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1937	sk->sk_prot->unhash(sk);
1938	inet->inet_sport = 0;
1939	}
1940	sk_dst_reset(sk);
1941	return 0;
1942	}
1943	EXPORT_SYMBOL(__udp_disconnect);
1944
1945	int udp_disconnect(struct sock *sk, int flags)
1946	{
1947	lock_sock(sk);
1948	__udp_disconnect(sk, flags);
1949	release_sock(sk);
1950	return 0;
1951	}
1952	EXPORT_SYMBOL(udp_disconnect);
1953
1954	void udp_lib_unhash(struct sock *sk)
1955	{
1956	if (sk_hashed(sk)) {
1957	struct udp_table *udptable = udp_get_table_prot(sk);
1958	struct udp_hslot *hslot, *hslot2;
1959
1960	hslot = udp_hashslot(table: udptable, net: sock_net(sk),
1961	udp_sk(sk)->udp_port_hash);
1962	hslot2 = udp_hashslot2(table: udptable, udp_sk(sk)->udp_portaddr_hash);
1963
1964	spin_lock_bh(lock: &hslot->lock);
1965	if (rcu_access_pointer(sk->sk_reuseport_cb))
1966	reuseport_detach_sock(sk);
1967	if (sk_del_node_init_rcu(sk)) {
1968	hslot->count--;
1969	inet_sk(sk)->inet_num = 0;
1970	sock_prot_inuse_add(net: sock_net(sk), prot: sk->sk_prot, val: -1);
1971
1972	spin_lock(lock: &hslot2->lock);
1973	hlist_del_init_rcu(n: &udp_sk(sk)->udp_portaddr_node);
1974	hslot2->count--;
1975	spin_unlock(lock: &hslot2->lock);
1976	}
1977	spin_unlock_bh(lock: &hslot->lock);
1978	}
1979	}
1980	EXPORT_SYMBOL(udp_lib_unhash);
1981
1982	/*
1983	* inet_rcv_saddr was changed, we must rehash secondary hash
1984	*/
1985	void udp_lib_rehash(struct sock *sk, u16 newhash)
1986	{
1987	if (sk_hashed(sk)) {
1988	struct udp_table *udptable = udp_get_table_prot(sk);
1989	struct udp_hslot *hslot, *hslot2, *nhslot2;
1990
1991	hslot2 = udp_hashslot2(table: udptable, udp_sk(sk)->udp_portaddr_hash);
1992	nhslot2 = udp_hashslot2(table: udptable, hash: newhash);
1993	udp_sk(sk)->udp_portaddr_hash = newhash;
1994
1995	if (hslot2 != nhslot2 ||
1996	rcu_access_pointer(sk->sk_reuseport_cb)) {
1997	hslot = udp_hashslot(table: udptable, net: sock_net(sk),
1998	udp_sk(sk)->udp_port_hash);
1999	/* we must lock primary chain too */
2000	spin_lock_bh(lock: &hslot->lock);
2001	if (rcu_access_pointer(sk->sk_reuseport_cb))
2002	reuseport_detach_sock(sk);
2003
2004	if (hslot2 != nhslot2) {
2005	spin_lock(lock: &hslot2->lock);
2006	hlist_del_init_rcu(n: &udp_sk(sk)->udp_portaddr_node);
2007	hslot2->count--;
2008	spin_unlock(lock: &hslot2->lock);
2009
2010	spin_lock(lock: &nhslot2->lock);
2011	hlist_add_head_rcu(n: &udp_sk(sk)->udp_portaddr_node,
2012	h: &nhslot2->head);
2013	nhslot2->count++;
2014	spin_unlock(lock: &nhslot2->lock);
2015	}
2016
2017	spin_unlock_bh(lock: &hslot->lock);
2018	}
2019	}
2020	}
2021	EXPORT_SYMBOL(udp_lib_rehash);
2022
2023	void udp_v4_rehash(struct sock *sk)
2024	{
2025	u16 new_hash = ipv4_portaddr_hash(net: sock_net(sk),
2026	inet_sk(sk)->inet_rcv_saddr,
2027	inet_sk(sk)->inet_num);
2028	udp_lib_rehash(sk, new_hash);
2029	}
2030
2031	static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2032	{
2033	int rc;
2034
2035	if (inet_sk(sk)->inet_daddr) {
2036	sock_rps_save_rxhash(sk, skb);
2037	sk_mark_napi_id(sk, skb);
2038	sk_incoming_cpu_update(sk);
2039	} else {
2040	sk_mark_napi_id_once(sk, skb);
2041	}
2042
2043	rc = __udp_enqueue_schedule_skb(sk, skb);
2044	if (rc < 0) {
2045	int is_udplite = IS_UDPLITE(sk);
2046	int drop_reason;
2047
2048	/* Note that an ENOMEM error is charged twice */
2049	if (rc == -ENOMEM) {
2050	UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2051	is_udplite);
2052	drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2053	} else {
2054	UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2055	is_udplite);
2056	drop_reason = SKB_DROP_REASON_PROTO_MEM;
2057	}
2058	UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2059	kfree_skb_reason(skb, reason: drop_reason);
2060	trace_udp_fail_queue_rcv_skb(rc, sk);
2061	return -1;
2062	}
2063
2064	return 0;
2065	}
2066
2067	/* returns:
2068	* -1: error
2069	* 0: success
2070	* >0: "udp encap" protocol resubmission
2071	*
2072	* Note that in the success and error cases, the skb is assumed to
2073	* have either been requeued or freed.
2074	*/
2075	static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2076	{
2077	int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2078	struct udp_sock *up = udp_sk(sk);
2079	int is_udplite = IS_UDPLITE(sk);
2080
2081	/*
2082	* Charge it to the socket, dropping if the queue is full.
2083	*/
2084	if (!xfrm4_policy_check(sk, dir: XFRM_POLICY_IN, skb)) {
2085	drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2086	goto drop;
2087	}
2088	nf_reset_ct(skb);
2089
2090	if (static_branch_unlikely(&udp_encap_needed_key) &&
2091	READ_ONCE(up->encap_type)) {
2092	int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2093
2094	/*
2095	* This is an encapsulation socket so pass the skb to
2096	* the socket's udp_encap_rcv() hook. Otherwise, just
2097	* fall through and pass this up the UDP socket.
2098	* up->encap_rcv() returns the following value:
2099	* =0 if skb was successfully passed to the encap
2100	* handler or was discarded by it.
2101	* >0 if skb should be passed on to UDP.
2102	* <0 if skb should be resubmitted as proto -N
2103	*/
2104
2105	/* if we're overly short, let UDP handle it */
2106	encap_rcv = READ_ONCE(up->encap_rcv);
2107	if (encap_rcv) {
2108	int ret;
2109
2110	/* Verify checksum before giving to encap */
2111	if (udp_lib_checksum_complete(skb))
2112	goto csum_error;
2113
2114	ret = encap_rcv(sk, skb);
2115	if (ret <= 0) {
2116	__UDP_INC_STATS(sock_net(sk),
2117	UDP_MIB_INDATAGRAMS,
2118	is_udplite);
2119	return -ret;
2120	}
2121	}
2122
2123	/* FALLTHROUGH -- it's a UDP Packet */
2124	}
2125
2126	/*
2127	* UDP-Lite specific tests, ignored on UDP sockets
2128	*/
2129	if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2130	u16 pcrlen = READ_ONCE(up->pcrlen);
2131
2132	/*
2133	* MIB statistics other than incrementing the error count are
2134	* disabled for the following two types of errors: these depend
2135	* on the application settings, not on the functioning of the
2136	* protocol stack as such.
2137	*
2138	* RFC 3828 here recommends (sec 3.3): "There should also be a
2139	* way ... to ... at least let the receiving application block
2140	* delivery of packets with coverage values less than a value
2141	* provided by the application."
2142	*/
2143	if (pcrlen == 0) { /* full coverage was set */
2144	net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2145	UDP_SKB_CB(skb)->cscov, skb->len);
2146	goto drop;
2147	}
2148	/* The next case involves violating the min. coverage requested
2149	* by the receiver. This is subtle: if receiver wants x and x is
2150	* greater than the buffersize/MTU then receiver will complain
2151	* that it wants x while sender emits packets of smaller size y.
2152	* Therefore the above ...()->partial_cov statement is essential.
2153	*/
2154	if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2155	net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2156	UDP_SKB_CB(skb)->cscov, pcrlen);
2157	goto drop;
2158	}
2159	}
2160
2161	prefetch(&sk->sk_rmem_alloc);
2162	if (rcu_access_pointer(sk->sk_filter) &&
2163	udp_lib_checksum_complete(skb))
2164	goto csum_error;
2165
2166	if (sk_filter_trim_cap(sk, skb, cap: sizeof(struct udphdr))) {
2167	drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2168	goto drop;
2169	}
2170
2171	udp_csum_pull_header(skb);
2172
2173	ipv4_pktinfo_prepare(sk, skb, drop_dst: true);
2174	return __udp_queue_rcv_skb(sk, skb);
2175
2176	csum_error:
2177	drop_reason = SKB_DROP_REASON_UDP_CSUM;
2178	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2179	drop:
2180	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2181	atomic_inc(v: &sk->sk_drops);
2182	kfree_skb_reason(skb, reason: drop_reason);
2183	return -1;
2184	}
2185
2186	static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2187	{
2188	struct sk_buff *next, *segs;
2189	int ret;
2190
2191	if (likely(!udp_unexpected_gso(sk, skb)))
2192	return udp_queue_rcv_one_skb(sk, skb);
2193
2194	BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2195	__skb_push(skb, len: -skb_mac_offset(skb));
2196	segs = udp_rcv_segment(sk, skb, ipv4: true);
2197	skb_list_walk_safe(segs, skb, next) {
2198	__skb_pull(skb, len: skb_transport_offset(skb));
2199
2200	udp_post_segment_fix_csum(skb);
2201	ret = udp_queue_rcv_one_skb(sk, skb);
2202	if (ret > 0)
2203	ip_protocol_deliver_rcu(net: dev_net(dev: skb->dev), skb, proto: ret);
2204	}
2205	return 0;
2206	}
2207
2208	/* For TCP sockets, sk_rx_dst is protected by socket lock
2209	* For UDP, we use xchg() to guard against concurrent changes.
2210	*/
2211	bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2212	{
2213	struct dst_entry *old;
2214
2215	if (dst_hold_safe(dst)) {
2216	old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2217	dst_release(dst: old);
2218	return old != dst;
2219	}
2220	return false;
2221	}
2222	EXPORT_SYMBOL(udp_sk_rx_dst_set);
2223
2224	/*
2225	* Multicasts and broadcasts go to each listener.
2226	*
2227	* Note: called only from the BH handler context.
2228	*/
2229	static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2230	struct udphdr *uh,
2231	__be32 saddr, __be32 daddr,
2232	struct udp_table *udptable,
2233	int proto)
2234	{
2235	struct sock *sk, *first = NULL;
2236	unsigned short hnum = ntohs(uh->dest);
2237	struct udp_hslot *hslot = udp_hashslot(table: udptable, net, num: hnum);
2238	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2239	unsigned int offset = offsetof(typeof(*sk), sk_node);
2240	int dif = skb->dev->ifindex;
2241	int sdif = inet_sdif(skb);
2242	struct hlist_node *node;
2243	struct sk_buff *nskb;
2244
2245	if (use_hash2) {
2246	hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), port: hnum) &
2247	udptable->mask;
2248	hash2 = ipv4_portaddr_hash(net, saddr: daddr, port: hnum) & udptable->mask;
2249	start_lookup:
2250	hslot = &udptable->hash2[hash2];
2251	offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2252	}
2253
2254	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2255	if (!__udp_is_mcast_sock(net, sk, loc_port: uh->dest, loc_addr: daddr,
2256	rmt_port: uh->source, rmt_addr: saddr, dif, sdif, hnum))
2257	continue;
2258
2259	if (!first) {
2260	first = sk;
2261	continue;
2262	}
2263	nskb = skb_clone(skb, GFP_ATOMIC);
2264
2265	if (unlikely(!nskb)) {
2266	atomic_inc(v: &sk->sk_drops);
2267	__UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2268	IS_UDPLITE(sk));
2269	__UDP_INC_STATS(net, UDP_MIB_INERRORS,
2270	IS_UDPLITE(sk));
2271	continue;
2272	}
2273	if (udp_queue_rcv_skb(sk, skb: nskb) > 0)
2274	consume_skb(skb: nskb);
2275	}
2276
2277	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
2278	if (use_hash2 && hash2 != hash2_any) {
2279	hash2 = hash2_any;
2280	goto start_lookup;
2281	}
2282
2283	if (first) {
2284	if (udp_queue_rcv_skb(sk: first, skb) > 0)
2285	consume_skb(skb);
2286	} else {
2287	kfree_skb(skb);
2288	__UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2289	proto == IPPROTO_UDPLITE);
2290	}
2291	return 0;
2292	}
2293
2294	/* Initialize UDP checksum. If exited with zero value (success),
2295	* CHECKSUM_UNNECESSARY means, that no more checks are required.
2296	* Otherwise, csum completion requires checksumming packet body,
2297	* including udp header and folding it to skb->csum.
2298	*/
2299	static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2300	int proto)
2301	{
2302	int err;
2303
2304	UDP_SKB_CB(skb)->partial_cov = 0;
2305	UDP_SKB_CB(skb)->cscov = skb->len;
2306
2307	if (proto == IPPROTO_UDPLITE) {
2308	err = udplite_checksum_init(skb, uh);
2309	if (err)
2310	return err;
2311
2312	if (UDP_SKB_CB(skb)->partial_cov) {
2313	skb->csum = inet_compute_pseudo(skb, proto);
2314	return 0;
2315	}
2316	}
2317
2318	/* Note, we are only interested in != 0 or == 0, thus the
2319	* force to int.
2320	*/
2321	err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2322	inet_compute_pseudo);
2323	if (err)
2324	return err;
2325
2326	if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2327	/* If SW calculated the value, we know it's bad */
2328	if (skb->csum_complete_sw)
2329	return 1;
2330
2331	/* HW says the value is bad. Let's validate that.
2332	* skb->csum is no longer the full packet checksum,
2333	* so don't treat it as such.
2334	*/
2335	skb_checksum_complete_unset(skb);
2336	}
2337
2338	return 0;
2339	}
2340
2341	/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2342	* return code conversion for ip layer consumption
2343	*/
2344	static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2345	struct udphdr *uh)
2346	{
2347	int ret;
2348
2349	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2350	skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2351
2352	ret = udp_queue_rcv_skb(sk, skb);
2353
2354	/* a return value > 0 means to resubmit the input, but
2355	* it wants the return to be -protocol, or 0
2356	*/
2357	if (ret > 0)
2358	return -ret;
2359	return 0;
2360	}
2361
2362	/*
2363	* All we need to do is get the socket, and then do a checksum.
2364	*/
2365
2366	int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2367	int proto)
2368	{
2369	struct sock *sk;
2370	struct udphdr *uh;
2371	unsigned short ulen;
2372	struct rtable *rt = skb_rtable(skb);
2373	__be32 saddr, daddr;
2374	struct net *net = dev_net(dev: skb->dev);
2375	bool refcounted;
2376	int drop_reason;
2377
2378	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2379
2380	/*
2381	* Validate the packet.
2382	*/
2383	if (!pskb_may_pull(skb, len: sizeof(struct udphdr)))
2384	goto drop; /* No space for header. */
2385
2386	uh = udp_hdr(skb);
2387	ulen = ntohs(uh->len);
2388	saddr = ip_hdr(skb)->saddr;
2389	daddr = ip_hdr(skb)->daddr;
2390
2391	if (ulen > skb->len)
2392	goto short_packet;
2393
2394	if (proto == IPPROTO_UDP) {
2395	/* UDP validates ulen. */
2396	if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, len: ulen))
2397	goto short_packet;
2398	uh = udp_hdr(skb);
2399	}
2400
2401	if (udp4_csum_init(skb, uh, proto))
2402	goto csum_error;
2403
2404	sk = inet_steal_sock(net, skb, doff: sizeof(struct udphdr), saddr, sport: uh->source, daddr, dport: uh->dest,
2405	refcounted: &refcounted, ehashfn: udp_ehashfn);
2406	if (IS_ERR(ptr: sk))
2407	goto no_sk;
2408
2409	if (sk) {
2410	struct dst_entry *dst = skb_dst(skb);
2411	int ret;
2412
2413	if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2414	udp_sk_rx_dst_set(sk, dst);
2415
2416	ret = udp_unicast_rcv_skb(sk, skb, uh);
2417	if (refcounted)
2418	sock_put(sk);
2419	return ret;
2420	}
2421
2422	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2423	return __udp4_lib_mcast_deliver(net, skb, uh,
2424	saddr, daddr, udptable, proto);
2425
2426	sk = __udp4_lib_lookup_skb(skb, sport: uh->source, dport: uh->dest, udptable);
2427	if (sk)
2428	return udp_unicast_rcv_skb(sk, skb, uh);
2429	no_sk:
2430	if (!xfrm4_policy_check(NULL, dir: XFRM_POLICY_IN, skb))
2431	goto drop;
2432	nf_reset_ct(skb);
2433
2434	/* No socket. Drop packet silently, if checksum is wrong */
2435	if (udp_lib_checksum_complete(skb))
2436	goto csum_error;
2437
2438	drop_reason = SKB_DROP_REASON_NO_SOCKET;
2439	__UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2440	icmp_send(skb_in: skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, info: 0);
2441
2442	/*
2443	* Hmm. We got an UDP packet to a port to which we
2444	* don't wanna listen. Ignore it.
2445	*/
2446	kfree_skb_reason(skb, reason: drop_reason);
2447	return 0;
2448
2449	short_packet:
2450	drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2451	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2452	proto == IPPROTO_UDPLITE ? "Lite" : "",
2453	&saddr, ntohs(uh->source),
2454	ulen, skb->len,
2455	&daddr, ntohs(uh->dest));
2456	goto drop;
2457
2458	csum_error:
2459	/*
2460	* RFC1122: OK. Discards the bad packet silently (as far as
2461	* the network is concerned, anyway) as per 4.1.3.4 (MUST).
2462	*/
2463	drop_reason = SKB_DROP_REASON_UDP_CSUM;
2464	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2465	proto == IPPROTO_UDPLITE ? "Lite" : "",
2466	&saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2467	ulen);
2468	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2469	drop:
2470	__UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2471	kfree_skb_reason(skb, reason: drop_reason);
2472	return 0;
2473	}
2474
2475	/* We can only early demux multicast if there is a single matching socket.
2476	* If more than one socket found returns NULL
2477	*/
2478	static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2479	__be16 loc_port, __be32 loc_addr,
2480	__be16 rmt_port, __be32 rmt_addr,
2481	int dif, int sdif)
2482	{
2483	struct udp_table *udptable = net->ipv4.udp_table;
2484	unsigned short hnum = ntohs(loc_port);
2485	struct sock *sk, *result;
2486	struct udp_hslot *hslot;
2487	unsigned int slot;
2488
2489	slot = udp_hashfn(net, num: hnum, mask: udptable->mask);
2490	hslot = &udptable->hash[slot];
2491
2492	/* Do not bother scanning a too big list */
2493	if (hslot->count > 10)
2494	return NULL;
2495
2496	result = NULL;
2497	sk_for_each_rcu(sk, &hslot->head) {
2498	if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2499	rmt_port, rmt_addr, dif, sdif, hnum)) {
2500	if (result)
2501	return NULL;
2502	result = sk;
2503	}
2504	}
2505
2506	return result;
2507	}
2508
2509	/* For unicast we should only early demux connected sockets or we can
2510	* break forwarding setups. The chains here can be long so only check
2511	* if the first socket is an exact match and if not move on.
2512	*/
2513	static struct sock *__udp4_lib_demux_lookup(struct net *net,
2514	__be16 loc_port, __be32 loc_addr,
2515	__be16 rmt_port, __be32 rmt_addr,
2516	int dif, int sdif)
2517	{
2518	struct udp_table *udptable = net->ipv4.udp_table;
2519	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2520	unsigned short hnum = ntohs(loc_port);
2521	unsigned int hash2, slot2;
2522	struct udp_hslot *hslot2;
2523	__portpair ports;
2524	struct sock *sk;
2525
2526	hash2 = ipv4_portaddr_hash(net, saddr: loc_addr, port: hnum);
2527	slot2 = hash2 & udptable->mask;
2528	hslot2 = &udptable->hash2[slot2];
2529	ports = INET_COMBINED_PORTS(rmt_port, hnum);
2530
2531	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2532	if (inet_match(net, sk, cookie: acookie, ports, dif, sdif))
2533	return sk;
2534	/* Only check first socket in chain */
2535	break;
2536	}
2537	return NULL;
2538	}
2539
2540	int udp_v4_early_demux(struct sk_buff *skb)
2541	{
2542	struct net *net = dev_net(dev: skb->dev);
2543	struct in_device *in_dev = NULL;
2544	const struct iphdr *iph;
2545	const struct udphdr *uh;
2546	struct sock *sk = NULL;
2547	struct dst_entry *dst;
2548	int dif = skb->dev->ifindex;
2549	int sdif = inet_sdif(skb);
2550	int ours;
2551
2552	/* validate the packet */
2553	if (!pskb_may_pull(skb, len: skb_transport_offset(skb) + sizeof(struct udphdr)))
2554	return 0;
2555
2556	iph = ip_hdr(skb);
2557	uh = udp_hdr(skb);
2558
2559	if (skb->pkt_type == PACKET_MULTICAST) {
2560	in_dev = __in_dev_get_rcu(dev: skb->dev);
2561
2562	if (!in_dev)
2563	return 0;
2564
2565	ours = ip_check_mc_rcu(dev: in_dev, mc_addr: iph->daddr, src_addr: iph->saddr,
2566	proto: iph->protocol);
2567	if (!ours)
2568	return 0;
2569
2570	sk = __udp4_lib_mcast_demux_lookup(net, loc_port: uh->dest, loc_addr: iph->daddr,
2571	rmt_port: uh->source, rmt_addr: iph->saddr,
2572	dif, sdif);
2573	} else if (skb->pkt_type == PACKET_HOST) {
2574	sk = __udp4_lib_demux_lookup(net, loc_port: uh->dest, loc_addr: iph->daddr,
2575	rmt_port: uh->source, rmt_addr: iph->saddr, dif, sdif);
2576	}
2577
2578	if (!sk)
2579	return 0;
2580
2581	skb->sk = sk;
2582	DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
2583	skb->destructor = sock_pfree;
2584	dst = rcu_dereference(sk->sk_rx_dst);
2585
2586	if (dst)
2587	dst = dst_check(dst, cookie: 0);
2588	if (dst) {
2589	u32 itag = 0;
2590
2591	/* set noref for now.
2592	* any place which wants to hold dst has to call
2593	* dst_hold_safe()
2594	*/
2595	skb_dst_set_noref(skb, dst);
2596
2597	/* for unconnected multicast sockets we need to validate
2598	* the source on each packet
2599	*/
2600	if (!inet_sk(sk)->inet_daddr && in_dev)
2601	return ip_mc_validate_source(skb, daddr: iph->daddr,
2602	saddr: iph->saddr,
2603	tos: iph->tos & IPTOS_RT_MASK,
2604	dev: skb->dev, in_dev, itag: &itag);
2605	}
2606	return 0;
2607	}
2608
2609	int udp_rcv(struct sk_buff *skb)
2610	{
2611	return __udp4_lib_rcv(skb, udptable: dev_net(dev: skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2612	}
2613
2614	void udp_destroy_sock(struct sock *sk)
2615	{
2616	struct udp_sock *up = udp_sk(sk);
2617	bool slow = lock_sock_fast(sk);
2618
2619	/* protects from races with udp_abort() */
2620	sock_set_flag(sk, flag: SOCK_DEAD);
2621	udp_flush_pending_frames(sk);
2622	unlock_sock_fast(sk, slow);
2623	if (static_branch_unlikely(&udp_encap_needed_key)) {
2624	if (up->encap_type) {
2625	void (*encap_destroy)(struct sock *sk);
2626	encap_destroy = READ_ONCE(up->encap_destroy);
2627	if (encap_destroy)
2628	encap_destroy(sk);
2629	}
2630	if (udp_test_bit(ENCAP_ENABLED, sk))
2631	static_branch_dec(&udp_encap_needed_key);
2632	}
2633	}
2634
2635	static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2636	struct sock *sk)
2637	{
2638	#ifdef CONFIG_XFRM
2639	if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2640	if (family == AF_INET)
2641	WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv);
2642	else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2643	WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv);
2644	}
2645	#endif
2646	}
2647
2648	/*
2649	* Socket option code for UDP
2650	*/
2651	int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2652	sockptr_t optval, unsigned int optlen,
2653	int (*push_pending_frames)(struct sock *))
2654	{
2655	struct udp_sock *up = udp_sk(sk);
2656	int val, valbool;
2657	int err = 0;
2658	int is_udplite = IS_UDPLITE(sk);
2659
2660	if (level == SOL_SOCKET) {
2661	err = sk_setsockopt(sk, level, optname, optval, optlen);
2662
2663	if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2664	sockopt_lock_sock(sk);
2665	/* paired with READ_ONCE in udp_rmem_release() */
2666	WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2667	sockopt_release_sock(sk);
2668	}
2669	return err;
2670	}
2671
2672	if (optlen < sizeof(int))
2673	return -EINVAL;
2674
2675	if (copy_from_sockptr(dst: &val, src: optval, size: sizeof(val)))
2676	return -EFAULT;
2677
2678	valbool = val ? 1 : 0;
2679
2680	switch (optname) {
2681	case UDP_CORK:
2682	if (val != 0) {
2683	udp_set_bit(CORK, sk);
2684	} else {
2685	udp_clear_bit(CORK, sk);
2686	lock_sock(sk);
2687	push_pending_frames(sk);
2688	release_sock(sk);
2689	}
2690	break;
2691
2692	case UDP_ENCAP:
2693	switch (val) {
2694	case 0:
2695	#ifdef CONFIG_XFRM
2696	case UDP_ENCAP_ESPINUDP:
2697	set_xfrm_gro_udp_encap_rcv(encap_type: val, family: sk->sk_family, sk);
2698	fallthrough;
2699	case UDP_ENCAP_ESPINUDP_NON_IKE:
2700	#if IS_ENABLED(CONFIG_IPV6)
2701	if (sk->sk_family == AF_INET6)
2702	WRITE_ONCE(up->encap_rcv,
2703	ipv6_stub->xfrm6_udp_encap_rcv);
2704	else
2705	#endif
2706	WRITE_ONCE(up->encap_rcv,
2707	xfrm4_udp_encap_rcv);
2708	#endif
2709	fallthrough;
2710	case UDP_ENCAP_L2TPINUDP:
2711	WRITE_ONCE(up->encap_type, val);
2712	udp_tunnel_encap_enable(sk);
2713	break;
2714	default:
2715	err = -ENOPROTOOPT;
2716	break;
2717	}
2718	break;
2719
2720	case UDP_NO_CHECK6_TX:
2721	udp_set_no_check6_tx(sk, val: valbool);
2722	break;
2723
2724	case UDP_NO_CHECK6_RX:
2725	udp_set_no_check6_rx(sk, val: valbool);
2726	break;
2727
2728	case UDP_SEGMENT:
2729	if (val < 0 || val > USHRT_MAX)
2730	return -EINVAL;
2731	WRITE_ONCE(up->gso_size, val);
2732	break;
2733
2734	case UDP_GRO:
2735
2736	/* when enabling GRO, accept the related GSO packet type */
2737	if (valbool)
2738	udp_tunnel_encap_enable(sk);
2739	udp_assign_bit(GRO_ENABLED, sk, valbool);
2740	udp_assign_bit(ACCEPT_L4, sk, valbool);
2741	set_xfrm_gro_udp_encap_rcv(encap_type: up->encap_type, family: sk->sk_family, sk);
2742	break;
2743
2744	/*
2745	* UDP-Lite's partial checksum coverage (RFC 3828).
2746	*/
2747	/* The sender sets actual checksum coverage length via this option.
2748	* The case coverage > packet length is handled by send module. */
2749	case UDPLITE_SEND_CSCOV:
2750	if (!is_udplite) /* Disable the option on UDP sockets */
2751	return -ENOPROTOOPT;
2752	if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2753	val = 8;
2754	else if (val > USHRT_MAX)
2755	val = USHRT_MAX;
2756	WRITE_ONCE(up->pcslen, val);
2757	udp_set_bit(UDPLITE_SEND_CC, sk);
2758	break;
2759
2760	/* The receiver specifies a minimum checksum coverage value. To make
2761	* sense, this should be set to at least 8 (as done below). If zero is
2762	* used, this again means full checksum coverage. */
2763	case UDPLITE_RECV_CSCOV:
2764	if (!is_udplite) /* Disable the option on UDP sockets */
2765	return -ENOPROTOOPT;
2766	if (val != 0 && val < 8) /* Avoid silly minimal values. */
2767	val = 8;
2768	else if (val > USHRT_MAX)
2769	val = USHRT_MAX;
2770	WRITE_ONCE(up->pcrlen, val);
2771	udp_set_bit(UDPLITE_RECV_CC, sk);
2772	break;
2773
2774	default:
2775	err = -ENOPROTOOPT;
2776	break;
2777	}
2778
2779	return err;
2780	}
2781	EXPORT_SYMBOL(udp_lib_setsockopt);
2782
2783	int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2784	unsigned int optlen)
2785	{
2786	if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2787	return udp_lib_setsockopt(sk, level, optname,
2788	optval, optlen,
2789	udp_push_pending_frames);
2790	return ip_setsockopt(sk, level, optname, optval, optlen);
2791	}
2792
2793	int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2794	char __user *optval, int __user *optlen)
2795	{
2796	struct udp_sock *up = udp_sk(sk);
2797	int val, len;
2798
2799	if (get_user(len, optlen))
2800	return -EFAULT;
2801
2802	if (len < 0)
2803	return -EINVAL;
2804
2805	len = min_t(unsigned int, len, sizeof(int));
2806
2807	switch (optname) {
2808	case UDP_CORK:
2809	val = udp_test_bit(CORK, sk);
2810	break;
2811
2812	case UDP_ENCAP:
2813	val = READ_ONCE(up->encap_type);
2814	break;
2815
2816	case UDP_NO_CHECK6_TX:
2817	val = udp_get_no_check6_tx(sk);
2818	break;
2819
2820	case UDP_NO_CHECK6_RX:
2821	val = udp_get_no_check6_rx(sk);
2822	break;
2823
2824	case UDP_SEGMENT:
2825	val = READ_ONCE(up->gso_size);
2826	break;
2827
2828	case UDP_GRO:
2829	val = udp_test_bit(GRO_ENABLED, sk);
2830	break;
2831
2832	/* The following two cannot be changed on UDP sockets, the return is
2833	* always 0 (which corresponds to the full checksum coverage of UDP). */
2834	case UDPLITE_SEND_CSCOV:
2835	val = READ_ONCE(up->pcslen);
2836	break;
2837
2838	case UDPLITE_RECV_CSCOV:
2839	val = READ_ONCE(up->pcrlen);
2840	break;
2841
2842	default:
2843	return -ENOPROTOOPT;
2844	}
2845
2846	if (put_user(len, optlen))
2847	return -EFAULT;
2848	if (copy_to_user(to: optval, from: &val, n: len))
2849	return -EFAULT;
2850	return 0;
2851	}
2852	EXPORT_SYMBOL(udp_lib_getsockopt);
2853
2854	int udp_getsockopt(struct sock *sk, int level, int optname,
2855	char __user *optval, int __user *optlen)
2856	{
2857	if (level == SOL_UDP || level == SOL_UDPLITE)
2858	return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2859	return ip_getsockopt(sk, level, optname, optval, optlen);
2860	}
2861
2862	/**
2863	* udp_poll - wait for a UDP event.
2864	* @file: - file struct
2865	* @sock: - socket
2866	* @wait: - poll table
2867	*
2868	* This is same as datagram poll, except for the special case of
2869	* blocking sockets. If application is using a blocking fd
2870	* and a packet with checksum error is in the queue;
2871	* then it could get return from select indicating data available
2872	* but then block when reading it. Add special case code
2873	* to work around these arguably broken applications.
2874	*/
2875	__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2876	{
2877	__poll_t mask = datagram_poll(file, sock, wait);
2878	struct sock *sk = sock->sk;
2879
2880	if (!skb_queue_empty_lockless(list: &udp_sk(sk)->reader_queue))
2881	mask |= EPOLLIN | EPOLLRDNORM;
2882
2883	/* Check for false positives due to checksum errors */
2884	if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2885	!(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2886	mask &= ~(EPOLLIN | EPOLLRDNORM);
2887
2888	/* psock ingress_msg queue should not contain any bad checksum frames */
2889	if (sk_is_readable(sk))
2890	mask |= EPOLLIN | EPOLLRDNORM;
2891	return mask;
2892
2893	}
2894	EXPORT_SYMBOL(udp_poll);
2895
2896	int udp_abort(struct sock *sk, int err)
2897	{
2898	if (!has_current_bpf_ctx())
2899	lock_sock(sk);
2900
2901	/* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2902	* with close()
2903	*/
2904	if (sock_flag(sk, flag: SOCK_DEAD))
2905	goto out;
2906
2907	sk->sk_err = err;
2908	sk_error_report(sk);
2909	__udp_disconnect(sk, 0);
2910
2911	out:
2912	if (!has_current_bpf_ctx())
2913	release_sock(sk);
2914
2915	return 0;
2916	}
2917	EXPORT_SYMBOL_GPL(udp_abort);
2918
2919	struct proto udp_prot = {
2920	.name = "UDP",
2921	.owner = THIS_MODULE,
2922	.close = udp_lib_close,
2923	.pre_connect = udp_pre_connect,
2924	.connect = ip4_datagram_connect,
2925	.disconnect = udp_disconnect,
2926	.ioctl = udp_ioctl,
2927	.init = udp_init_sock,
2928	.destroy = udp_destroy_sock,
2929	.setsockopt = udp_setsockopt,
2930	.getsockopt = udp_getsockopt,
2931	.sendmsg = udp_sendmsg,
2932	.recvmsg = udp_recvmsg,
2933	.splice_eof = udp_splice_eof,
2934	.release_cb = ip4_datagram_release_cb,
2935	.hash = udp_lib_hash,
2936	.unhash = udp_lib_unhash,
2937	.rehash = udp_v4_rehash,
2938	.get_port = udp_v4_get_port,
2939	.put_port = udp_lib_unhash,
2940	#ifdef CONFIG_BPF_SYSCALL
2941	.psock_update_sk_prot = udp_bpf_update_proto,
2942	#endif
2943	.memory_allocated = &udp_memory_allocated,
2944	.per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2945
2946	.sysctl_mem = sysctl_udp_mem,
2947	.sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2948	.sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2949	.obj_size = sizeof(struct udp_sock),
2950	.h.udp_table = NULL,
2951	.diag_destroy = udp_abort,
2952	};
2953	EXPORT_SYMBOL(udp_prot);
2954
2955	/* ------------------------------------------------------------------------ */
2956	#ifdef CONFIG_PROC_FS
2957
2958	static unsigned short seq_file_family(const struct seq_file *seq);
2959	static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2960	{
2961	unsigned short family = seq_file_family(seq);
2962
2963	/* AF_UNSPEC is used as a match all */
2964	return ((family == AF_UNSPEC || family == sk->sk_family) &&
2965	net_eq(net1: sock_net(sk), net2: seq_file_net(seq)));
2966	}
2967
2968	#ifdef CONFIG_BPF_SYSCALL
2969	static const struct seq_operations bpf_iter_udp_seq_ops;
2970	#endif
2971	static struct udp_table *udp_get_table_seq(struct seq_file *seq,
2972	struct net *net)
2973	{
2974	const struct udp_seq_afinfo *afinfo;
2975
2976	#ifdef CONFIG_BPF_SYSCALL
2977	if (seq->op == &bpf_iter_udp_seq_ops)
2978	return net->ipv4.udp_table;
2979	#endif
2980
2981	afinfo = pde_data(inode: file_inode(f: seq->file));
2982	return afinfo->udp_table ? : net->ipv4.udp_table;
2983	}
2984
2985	static struct sock *udp_get_first(struct seq_file *seq, int start)
2986	{
2987	struct udp_iter_state *state = seq->private;
2988	struct net *net = seq_file_net(seq);
2989	struct udp_table *udptable;
2990	struct sock *sk;
2991
2992	udptable = udp_get_table_seq(seq, net);
2993
2994	for (state->bucket = start; state->bucket <= udptable->mask;
2995	++state->bucket) {
2996	struct udp_hslot *hslot = &udptable->hash[state->bucket];
2997
2998	if (hlist_empty(h: &hslot->head))
2999	continue;
3000
3001	spin_lock_bh(lock: &hslot->lock);
3002	sk_for_each(sk, &hslot->head) {
3003	if (seq_sk_match(seq, sk))
3004	goto found;
3005	}
3006	spin_unlock_bh(lock: &hslot->lock);
3007	}
3008	sk = NULL;
3009	found:
3010	return sk;
3011	}
3012
3013	static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3014	{
3015	struct udp_iter_state *state = seq->private;
3016	struct net *net = seq_file_net(seq);
3017	struct udp_table *udptable;
3018
3019	do {
3020	sk = sk_next(sk);
3021	} while (sk && !seq_sk_match(seq, sk));
3022
3023	if (!sk) {
3024	udptable = udp_get_table_seq(seq, net);
3025
3026	if (state->bucket <= udptable->mask)
3027	spin_unlock_bh(lock: &udptable->hash[state->bucket].lock);
3028
3029	return udp_get_first(seq, start: state->bucket + 1);
3030	}
3031	return sk;
3032	}
3033
3034	static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3035	{
3036	struct sock *sk = udp_get_first(seq, start: 0);
3037
3038	if (sk)
3039	while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3040	--pos;
3041	return pos ? NULL : sk;
3042	}
3043
3044	void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3045	{
3046	struct udp_iter_state *state = seq->private;
3047	state->bucket = MAX_UDP_PORTS;
3048
3049	return *pos ? udp_get_idx(seq, pos: *pos-1) : SEQ_START_TOKEN;
3050	}
3051	EXPORT_SYMBOL(udp_seq_start);
3052
3053	void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3054	{
3055	struct sock *sk;
3056
3057	if (v == SEQ_START_TOKEN)
3058	sk = udp_get_idx(seq, pos: 0);
3059	else
3060	sk = udp_get_next(seq, sk: v);
3061
3062	++*pos;
3063	return sk;
3064	}
3065	EXPORT_SYMBOL(udp_seq_next);
3066
3067	void udp_seq_stop(struct seq_file *seq, void *v)
3068	{
3069	struct udp_iter_state *state = seq->private;
3070	struct udp_table *udptable;
3071
3072	udptable = udp_get_table_seq(seq, net: seq_file_net(seq));
3073
3074	if (state->bucket <= udptable->mask)
3075	spin_unlock_bh(lock: &udptable->hash[state->bucket].lock);
3076	}
3077	EXPORT_SYMBOL(udp_seq_stop);
3078
3079	/* ------------------------------------------------------------------------ */
3080	static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3081	int bucket)
3082	{
3083	struct inet_sock *inet = inet_sk(sp);
3084	__be32 dest = inet->inet_daddr;
3085	__be32 src = inet->inet_rcv_saddr;
3086	__u16 destp = ntohs(inet->inet_dport);
3087	__u16 srcp = ntohs(inet->inet_sport);
3088
3089	seq_printf(m: f, fmt: "%5d: %08X:%04X %08X:%04X"
3090	" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3091	bucket, src, srcp, dest, destp, sp->sk_state,
3092	sk_wmem_alloc_get(sk: sp),
3093	udp_rqueue_get(sk: sp),
3094	0, 0L, 0,
3095	from_kuid_munged(to: seq_user_ns(seq: f), uid: sock_i_uid(sk: sp)),
3096	0, sock_i_ino(sk: sp),
3097	refcount_read(r: &sp->sk_refcnt), sp,
3098	atomic_read(v: &sp->sk_drops));
3099	}
3100
3101	int udp4_seq_show(struct seq_file *seq, void *v)
3102	{
3103	seq_setwidth(m: seq, size: 127);
3104	if (v == SEQ_START_TOKEN)
3105	seq_puts(m: seq, s: " sl local_address rem_address st tx_queue "
3106	"rx_queue tr tm->when retrnsmt uid timeout "
3107	"inode ref pointer drops");
3108	else {
3109	struct udp_iter_state *state = seq->private;
3110
3111	udp4_format_sock(sp: v, f: seq, bucket: state->bucket);
3112	}
3113	seq_pad(m: seq, c: '\n');
3114	return 0;
3115	}
3116
3117	#ifdef CONFIG_BPF_SYSCALL
3118	struct bpf_iter__udp {
3119	__bpf_md_ptr(struct bpf_iter_meta *, meta);
3120	__bpf_md_ptr(struct udp_sock *, udp_sk);
3121	uid_t uid __aligned(8);
3122	int bucket __aligned(8);
3123	};
3124
3125	struct bpf_udp_iter_state {
3126	struct udp_iter_state state;
3127	unsigned int cur_sk;
3128	unsigned int end_sk;
3129	unsigned int max_sk;
3130	int offset;
3131	struct sock **batch;
3132	bool st_bucket_done;
3133	};
3134
3135	static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3136	unsigned int new_batch_sz);
3137	static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3138	{
3139	struct bpf_udp_iter_state *iter = seq->private;
3140	struct udp_iter_state *state = &iter->state;
3141	struct net *net = seq_file_net(seq);
3142	int resume_bucket, resume_offset;
3143	struct udp_table *udptable;
3144	unsigned int batch_sks = 0;
3145	bool resized = false;
3146	struct sock *sk;
3147
3148	resume_bucket = state->bucket;
3149	resume_offset = iter->offset;
3150
3151	/* The current batch is done, so advance the bucket. */
3152	if (iter->st_bucket_done)
3153	state->bucket++;
3154
3155	udptable = udp_get_table_seq(seq, net);
3156
3157	again:
3158	/* New batch for the next bucket.
3159	* Iterate over the hash table to find a bucket with sockets matching
3160	* the iterator attributes, and return the first matching socket from
3161	* the bucket. The remaining matched sockets from the bucket are batched
3162	* before releasing the bucket lock. This allows BPF programs that are
3163	* called in seq_show to acquire the bucket lock if needed.
3164	*/
3165	iter->cur_sk = 0;
3166	iter->end_sk = 0;
3167	iter->st_bucket_done = false;
3168	batch_sks = 0;
3169
3170	for (; state->bucket <= udptable->mask; state->bucket++) {
3171	struct udp_hslot *hslot2 = &udptable->hash2[state->bucket];
3172
3173	if (hlist_empty(h: &hslot2->head))
3174	continue;
3175
3176	iter->offset = 0;
3177	spin_lock_bh(lock: &hslot2->lock);
3178	udp_portaddr_for_each_entry(sk, &hslot2->head) {
3179	if (seq_sk_match(seq, sk)) {
3180	/* Resume from the last iterated socket at the
3181	* offset in the bucket before iterator was stopped.
3182	*/
3183	if (state->bucket == resume_bucket &&
3184	iter->offset < resume_offset) {
3185	++iter->offset;
3186	continue;
3187	}
3188	if (iter->end_sk < iter->max_sk) {
3189	sock_hold(sk);
3190	iter->batch[iter->end_sk++] = sk;
3191	}
3192	batch_sks++;
3193	}
3194	}
3195	spin_unlock_bh(lock: &hslot2->lock);
3196
3197	if (iter->end_sk)
3198	break;
3199	}
3200
3201	/* All done: no batch made. */
3202	if (!iter->end_sk)
3203	return NULL;
3204
3205	if (iter->end_sk == batch_sks) {
3206	/* Batching is done for the current bucket; return the first
3207	* socket to be iterated from the batch.
3208	*/
3209	iter->st_bucket_done = true;
3210	goto done;
3211	}
3212	if (!resized && !bpf_iter_udp_realloc_batch(iter, new_batch_sz: batch_sks * 3 / 2)) {
3213	resized = true;
3214	/* After allocating a larger batch, retry one more time to grab
3215	* the whole bucket.
3216	*/
3217	goto again;
3218	}
3219	done:
3220	return iter->batch[0];
3221	}
3222
3223	static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3224	{
3225	struct bpf_udp_iter_state *iter = seq->private;
3226	struct sock *sk;
3227
3228	/* Whenever seq_next() is called, the iter->cur_sk is
3229	* done with seq_show(), so unref the iter->cur_sk.
3230	*/
3231	if (iter->cur_sk < iter->end_sk) {
3232	sock_put(sk: iter->batch[iter->cur_sk++]);
3233	++iter->offset;
3234	}
3235
3236	/* After updating iter->cur_sk, check if there are more sockets
3237	* available in the current bucket batch.
3238	*/
3239	if (iter->cur_sk < iter->end_sk)
3240	sk = iter->batch[iter->cur_sk];
3241	else
3242	/* Prepare a new batch. */
3243	sk = bpf_iter_udp_batch(seq);
3244
3245	++*pos;
3246	return sk;
3247	}
3248
3249	static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3250	{
3251	/* bpf iter does not support lseek, so it always
3252	* continue from where it was stop()-ped.
3253	*/
3254	if (*pos)
3255	return bpf_iter_udp_batch(seq);
3256
3257	return SEQ_START_TOKEN;
3258	}
3259
3260	static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3261	struct udp_sock *udp_sk, uid_t uid, int bucket)
3262	{
3263	struct bpf_iter__udp ctx;
3264
3265	meta->seq_num--; /* skip SEQ_START_TOKEN */
3266	ctx.meta = meta;
3267	ctx.udp_sk = udp_sk;
3268	ctx.uid = uid;
3269	ctx.bucket = bucket;
3270	return bpf_iter_run_prog(prog, ctx: &ctx);
3271	}
3272
3273	static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3274	{
3275	struct udp_iter_state *state = seq->private;
3276	struct bpf_iter_meta meta;
3277	struct bpf_prog *prog;
3278	struct sock *sk = v;
3279	uid_t uid;
3280	int ret;
3281
3282	if (v == SEQ_START_TOKEN)
3283	return 0;
3284
3285	lock_sock(sk);
3286
3287	if (unlikely(sk_unhashed(sk))) {
3288	ret = SEQ_SKIP;
3289	goto unlock;
3290	}
3291
3292	uid = from_kuid_munged(to: seq_user_ns(seq), uid: sock_i_uid(sk));
3293	meta.seq = seq;
3294	prog = bpf_iter_get_info(meta: &meta, in_stop: false);
3295	ret = udp_prog_seq_show(prog, meta: &meta, udp_sk: v, uid, bucket: state->bucket);
3296
3297	unlock:
3298	release_sock(sk);
3299	return ret;
3300	}
3301
3302	static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3303	{
3304	while (iter->cur_sk < iter->end_sk)
3305	sock_put(sk: iter->batch[iter->cur_sk++]);
3306	}
3307
3308	static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3309	{
3310	struct bpf_udp_iter_state *iter = seq->private;
3311	struct bpf_iter_meta meta;
3312	struct bpf_prog *prog;
3313
3314	if (!v) {
3315	meta.seq = seq;
3316	prog = bpf_iter_get_info(meta: &meta, in_stop: true);
3317	if (prog)
3318	(void)udp_prog_seq_show(prog, meta: &meta, udp_sk: v, uid: 0, bucket: 0);
3319	}
3320
3321	if (iter->cur_sk < iter->end_sk) {
3322	bpf_iter_udp_put_batch(iter);
3323	iter->st_bucket_done = false;
3324	}
3325	}
3326
3327	static const struct seq_operations bpf_iter_udp_seq_ops = {
3328	.start = bpf_iter_udp_seq_start,
3329	.next = bpf_iter_udp_seq_next,
3330	.stop = bpf_iter_udp_seq_stop,
3331	.show = bpf_iter_udp_seq_show,
3332	};
3333	#endif
3334
3335	static unsigned short seq_file_family(const struct seq_file *seq)
3336	{
3337	const struct udp_seq_afinfo *afinfo;
3338
3339	#ifdef CONFIG_BPF_SYSCALL
3340	/* BPF iterator: bpf programs to filter sockets. */
3341	if (seq->op == &bpf_iter_udp_seq_ops)
3342	return AF_UNSPEC;
3343	#endif
3344
3345	/* Proc fs iterator */
3346	afinfo = pde_data(inode: file_inode(f: seq->file));
3347	return afinfo->family;
3348	}
3349
3350	const struct seq_operations udp_seq_ops = {
3351	.start = udp_seq_start,
3352	.next = udp_seq_next,
3353	.stop = udp_seq_stop,
3354	.show = udp4_seq_show,
3355	};
3356	EXPORT_SYMBOL(udp_seq_ops);
3357
3358	static struct udp_seq_afinfo udp4_seq_afinfo = {
3359	.family = AF_INET,
3360	.udp_table = NULL,
3361	};
3362
3363	static int __net_init udp4_proc_init_net(struct net *net)
3364	{
3365	if (!proc_create_net_data(name: "udp", mode: 0444, parent: net->proc_net, ops: &udp_seq_ops,
3366	state_size: sizeof(struct udp_iter_state), data: &udp4_seq_afinfo))
3367	return -ENOMEM;
3368	return 0;
3369	}
3370
3371	static void __net_exit udp4_proc_exit_net(struct net *net)
3372	{
3373	remove_proc_entry("udp", net->proc_net);
3374	}
3375
3376	static struct pernet_operations udp4_net_ops = {
3377	.init = udp4_proc_init_net,
3378	.exit = udp4_proc_exit_net,
3379	};
3380
3381	int __init udp4_proc_init(void)
3382	{
3383	return register_pernet_subsys(&udp4_net_ops);
3384	}
3385
3386	void udp4_proc_exit(void)
3387	{
3388	unregister_pernet_subsys(&udp4_net_ops);
3389	}
3390	#endif /* CONFIG_PROC_FS */
3391
3392	static __initdata unsigned long uhash_entries;
3393	static int __init set_uhash_entries(char *str)
3394	{
3395	ssize_t ret;
3396
3397	if (!str)
3398	return 0;
3399
3400	ret = kstrtoul(s: str, base: 0, res: &uhash_entries);
3401	if (ret)
3402	return 0;
3403
3404	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3405	uhash_entries = UDP_HTABLE_SIZE_MIN;
3406	return 1;
3407	}
3408	__setup("uhash_entries=", set_uhash_entries);
3409
3410	void __init udp_table_init(struct udp_table *table, const char *name)
3411	{
3412	unsigned int i;
3413
3414	table->hash = alloc_large_system_hash(tablename: name,
3415	bucketsize: 2 * sizeof(struct udp_hslot),
3416	numentries: uhash_entries,
3417	scale: 21, /* one slot per 2 MB */
3418	flags: 0,
3419	hash_shift: &table->log,
3420	hash_mask: &table->mask,
3421	UDP_HTABLE_SIZE_MIN,
3422	UDP_HTABLE_SIZE_MAX);
3423
3424	table->hash2 = table->hash + (table->mask + 1);
3425	for (i = 0; i <= table->mask; i++) {
3426	INIT_HLIST_HEAD(&table->hash[i].head);
3427	table->hash[i].count = 0;
3428	spin_lock_init(&table->hash[i].lock);
3429	}
3430	for (i = 0; i <= table->mask; i++) {
3431	INIT_HLIST_HEAD(&table->hash2[i].head);
3432	table->hash2[i].count = 0;
3433	spin_lock_init(&table->hash2[i].lock);
3434	}
3435	}
3436
3437	u32 udp_flow_hashrnd(void)
3438	{
3439	static u32 hashrnd __read_mostly;
3440
3441	net_get_random_once(&hashrnd, sizeof(hashrnd));
3442
3443	return hashrnd;
3444	}
3445	EXPORT_SYMBOL(udp_flow_hashrnd);
3446
3447	static void __net_init udp_sysctl_init(struct net *net)
3448	{
3449	net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3450	net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3451
3452	#ifdef CONFIG_NET_L3_MASTER_DEV
3453	net->ipv4.sysctl_udp_l3mdev_accept = 0;
3454	#endif
3455	}
3456
3457	static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3458	{
3459	struct udp_table *udptable;
3460	int i;
3461
3462	udptable = kmalloc(size: sizeof(*udptable), GFP_KERNEL);
3463	if (!udptable)
3464	goto out;
3465
3466	udptable->hash = vmalloc_huge(size: hash_entries * 2 * sizeof(struct udp_hslot),
3467	GFP_KERNEL_ACCOUNT);
3468	if (!udptable->hash)
3469	goto free_table;
3470
3471	udptable->hash2 = udptable->hash + hash_entries;
3472	udptable->mask = hash_entries - 1;
3473	udptable->log = ilog2(hash_entries);
3474
3475	for (i = 0; i < hash_entries; i++) {
3476	INIT_HLIST_HEAD(&udptable->hash[i].head);
3477	udptable->hash[i].count = 0;
3478	spin_lock_init(&udptable->hash[i].lock);
3479
3480	INIT_HLIST_HEAD(&udptable->hash2[i].head);
3481	udptable->hash2[i].count = 0;
3482	spin_lock_init(&udptable->hash2[i].lock);
3483	}
3484
3485	return udptable;
3486
3487	free_table:
3488	kfree(objp: udptable);
3489	out:
3490	return NULL;
3491	}
3492
3493	static void __net_exit udp_pernet_table_free(struct net *net)
3494	{
3495	struct udp_table *udptable = net->ipv4.udp_table;
3496
3497	if (udptable == &udp_table)
3498	return;
3499
3500	kvfree(addr: udptable->hash);
3501	kfree(objp: udptable);
3502	}
3503
3504	static void __net_init udp_set_table(struct net *net)
3505	{
3506	struct udp_table *udptable;
3507	unsigned int hash_entries;
3508	struct net *old_net;
3509
3510	if (net_eq(net1: net, net2: &init_net))
3511	goto fallback;
3512
3513	old_net = current->nsproxy->net_ns;
3514	hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3515	if (!hash_entries)
3516	goto fallback;
3517
3518	/* Set min to keep the bitmap on stack in udp_lib_get_port() */
3519	if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3520	hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3521	else
3522	hash_entries = roundup_pow_of_two(hash_entries);
3523
3524	udptable = udp_pernet_table_alloc(hash_entries);
3525	if (udptable) {
3526	net->ipv4.udp_table = udptable;
3527	} else {
3528	pr_warn("Failed to allocate UDP hash table (entries: %u) "
3529	"for a netns, fallback to the global one\n",
3530	hash_entries);
3531	fallback:
3532	net->ipv4.udp_table = &udp_table;
3533	}
3534	}
3535
3536	static int __net_init udp_pernet_init(struct net *net)
3537	{
3538	udp_sysctl_init(net);
3539	udp_set_table(net);
3540
3541	return 0;
3542	}
3543
3544	static void __net_exit udp_pernet_exit(struct net *net)
3545	{
3546	udp_pernet_table_free(net);
3547	}
3548
3549	static struct pernet_operations __net_initdata udp_sysctl_ops = {
3550	.init = udp_pernet_init,
3551	.exit = udp_pernet_exit,
3552	};
3553
3554	#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3555	DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3556	struct udp_sock *udp_sk, uid_t uid, int bucket)
3557
3558	static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3559	unsigned int new_batch_sz)
3560	{
3561	struct sock **new_batch;
3562
3563	new_batch = kvmalloc_array(n: new_batch_sz, size: sizeof(*new_batch),
3564	GFP_USER | __GFP_NOWARN);
3565	if (!new_batch)
3566	return -ENOMEM;
3567
3568	bpf_iter_udp_put_batch(iter);
3569	kvfree(addr: iter->batch);
3570	iter->batch = new_batch;
3571	iter->max_sk = new_batch_sz;
3572
3573	return 0;
3574	}
3575
3576	#define INIT_BATCH_SZ 16
3577
3578	static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3579	{
3580	struct bpf_udp_iter_state *iter = priv_data;
3581	int ret;
3582
3583	ret = bpf_iter_init_seq_net(priv_data, aux);
3584	if (ret)
3585	return ret;
3586
3587	ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3588	if (ret)
3589	bpf_iter_fini_seq_net(priv_data);
3590
3591	return ret;
3592	}
3593
3594	static void bpf_iter_fini_udp(void *priv_data)
3595	{
3596	struct bpf_udp_iter_state *iter = priv_data;
3597
3598	bpf_iter_fini_seq_net(priv_data);
3599	kvfree(addr: iter->batch);
3600	}
3601
3602	static const struct bpf_iter_seq_info udp_seq_info = {
3603	.seq_ops = &bpf_iter_udp_seq_ops,
3604	.init_seq_private = bpf_iter_init_udp,
3605	.fini_seq_private = bpf_iter_fini_udp,
3606	.seq_priv_size = sizeof(struct bpf_udp_iter_state),
3607	};
3608
3609	static struct bpf_iter_reg udp_reg_info = {
3610	.target = "udp",
3611	.ctx_arg_info_size = 1,
3612	.ctx_arg_info = {
3613	{ offsetof(struct bpf_iter__udp, udp_sk),
3614	.reg_type: PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3615	},
3616	.seq_info = &udp_seq_info,
3617	};
3618
3619	static void __init bpf_iter_register(void)
3620	{
3621	udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3622	if (bpf_iter_reg_target(reg_info: &udp_reg_info))
3623	pr_warn("Warning: could not register bpf iterator udp\n");
3624	}
3625	#endif
3626
3627	void __init udp_init(void)
3628	{
3629	unsigned long limit;
3630	unsigned int i;
3631
3632	udp_table_init(table: &udp_table, name: "UDP");
3633	limit = nr_free_buffer_pages() / 8;
3634	limit = max(limit, 128UL);
3635	sysctl_udp_mem[0] = limit / 4 * 3;
3636	sysctl_udp_mem[1] = limit;
3637	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3638
3639	/* 16 spinlocks per cpu */
3640	udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3641	udp_busylocks = kmalloc(size: sizeof(spinlock_t) << udp_busylocks_log,
3642	GFP_KERNEL);
3643	if (!udp_busylocks)
3644	panic(fmt: "UDP: failed to alloc udp_busylocks\n");
3645	for (i = 0; i < (1U << udp_busylocks_log); i++)
3646	spin_lock_init(udp_busylocks + i);
3647
3648	if (register_pernet_subsys(&udp_sysctl_ops))
3649	panic(fmt: "UDP: failed to init sysctl parameters.\n");
3650
3651	#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3652	bpf_iter_register();
3653	#endif
3654	}
3655