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 Encapsulation 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 <linux/sock_diag.h>
97	#include <net/tcp_states.h>
98	#include <linux/skbuff.h>
99	#include <linux/proc_fs.h>
100	#include <linux/seq_file.h>
101	#include <net/net_namespace.h>
102	#include <net/icmp.h>
103	#include <net/inet_hashtables.h>
104	#include <net/ip.h>
105	#include <net/ip_tunnels.h>
106	#include <net/route.h>
107	#include <net/checksum.h>
108	#include <net/gso.h>
109	#include <net/xfrm.h>
110	#include <trace/events/udp.h>
111	#include <linux/static_key.h>
112	#include <linux/btf_ids.h>
113	#include <trace/events/skb.h>
114	#include <net/busy_poll.h>
115	#include "udp_impl.h"
116	#include <net/sock_reuseport.h>
117	#include <net/addrconf.h>
118	#include <net/udp_tunnel.h>
119	#include <net/gro.h>
120	#if IS_ENABLED(CONFIG_IPV6)
121	#include <net/ipv6_stubs.h>
122	#endif
123	#include <net/rps.h>
124
125	struct udp_table udp_table __read_mostly;
126
127	long sysctl_udp_mem[3] __read_mostly;
128	EXPORT_IPV6_MOD(sysctl_udp_mem);
129
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	kuid_t uid = sk_uid(sk);
147	struct sock *sk2;
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: sk_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	kuid_t uid = sk_uid(sk);
182	struct sock *sk2;
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: sk_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 = sk_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: sk_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	sock_set_flag(sk, flag: SOCK_RCU_FREE);
330
331	sk_add_node_rcu(sk, list: &hslot->head);
332	hslot->count++;
333	sock_prot_inuse_add(net: sock_net(sk), prot: sk->sk_prot, val: 1);
334
335	hslot2 = udp_hashslot2(table: udptable, udp_sk(sk)->udp_portaddr_hash);
336	spin_lock(lock: &hslot2->lock);
337	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
338	sk->sk_family == AF_INET6)
339	hlist_add_tail_rcu(n: &udp_sk(sk)->udp_portaddr_node,
340	h: &hslot2->head);
341	else
342	hlist_add_head_rcu(n: &udp_sk(sk)->udp_portaddr_node,
343	h: &hslot2->head);
344	hslot2->count++;
345	spin_unlock(lock: &hslot2->lock);
346	}
347
348	error = 0;
349	fail_unlock:
350	spin_unlock_bh(lock: &hslot->lock);
351	fail:
352	return error;
353	}
354	EXPORT_IPV6_MOD(udp_lib_get_port);
355
356	int udp_v4_get_port(struct sock *sk, unsigned short snum)
357	{
358	unsigned int hash2_nulladdr =
359	ipv4_portaddr_hash(net: sock_net(sk), htonl(INADDR_ANY), port: snum);
360	unsigned int hash2_partial =
361	ipv4_portaddr_hash(net: sock_net(sk), inet_sk(sk)->inet_rcv_saddr, port: 0);
362
363	/* precompute partial secondary hash */
364	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
365	return udp_lib_get_port(sk, snum, hash2_nulladdr);
366	}
367
368	static int compute_score(struct sock *sk, const struct net *net,
369	__be32 saddr, __be16 sport,
370	__be32 daddr, unsigned short hnum,
371	int dif, int sdif)
372	{
373	int score;
374	struct inet_sock *inet;
375	bool dev_match;
376
377	if (!net_eq(net1: sock_net(sk), net2: net) ||
378	udp_sk(sk)->udp_port_hash != hnum ||
379	ipv6_only_sock(sk))
380	return -1;
381
382	if (sk->sk_rcv_saddr != daddr)
383	return -1;
384
385	score = (sk->sk_family == PF_INET) ? 2 : 1;
386
387	inet = inet_sk(sk);
388	if (inet->inet_daddr) {
389	if (inet->inet_daddr != saddr)
390	return -1;
391	score += 4;
392	}
393
394	if (inet->inet_dport) {
395	if (inet->inet_dport != sport)
396	return -1;
397	score += 4;
398	}
399
400	dev_match = udp_sk_bound_dev_eq(net, bound_dev_if: sk->sk_bound_dev_if,
401	dif, sdif);
402	if (!dev_match)
403	return -1;
404	if (sk->sk_bound_dev_if)
405	score += 4;
406
407	if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
408	score++;
409	return score;
410	}
411
412	u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
413	const __be32 faddr, const __be16 fport)
414	{
415	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
416
417	return __inet_ehashfn(laddr, lport, faddr, fport,
418	udp_ehash_secret + net_hash_mix(net));
419	}
420	EXPORT_IPV6_MOD(udp_ehashfn);
421
422	/**
423	* udp4_lib_lookup1() - Simplified lookup using primary hash (destination port)
424	* @net: Network namespace
425	* @saddr: Source address, network order
426	* @sport: Source port, network order
427	* @daddr: Destination address, network order
428	* @hnum: Destination port, host order
429	* @dif: Destination interface index
430	* @sdif: Destination bridge port index, if relevant
431	* @udptable: Set of UDP hash tables
432	*
433	* Simplified lookup to be used as fallback if no sockets are found due to a
434	* potential race between (receive) address change, and lookup happening before
435	* the rehash operation. This function ignores SO_REUSEPORT groups while scoring
436	* result sockets, because if we have one, we don't need the fallback at all.
437	*
438	* Called under rcu_read_lock().
439	*
440	* Return: socket with highest matching score if any, NULL if none
441	*/
442	static struct sock *udp4_lib_lookup1(const struct net *net,
443	__be32 saddr, __be16 sport,
444	__be32 daddr, unsigned int hnum,
445	int dif, int sdif,
446	const struct udp_table *udptable)
447	{
448	unsigned int slot = udp_hashfn(net, num: hnum, mask: udptable->mask);
449	struct udp_hslot *hslot = &udptable->hash[slot];
450	struct sock *sk, *result = NULL;
451	int score, badness = 0;
452
453	sk_for_each_rcu(sk, &hslot->head) {
454	score = compute_score(sk, net,
455	saddr, sport, daddr, hnum, dif, sdif);
456	if (score > badness) {
457	result = sk;
458	badness = score;
459	}
460	}
461
462	return result;
463	}
464
465	/* called with rcu_read_lock() */
466	static struct sock *udp4_lib_lookup2(const struct net *net,
467	__be32 saddr, __be16 sport,
468	__be32 daddr, unsigned int hnum,
469	int dif, int sdif,
470	struct udp_hslot *hslot2,
471	struct sk_buff *skb)
472	{
473	struct sock *sk, *result;
474	int score, badness;
475	bool need_rescore;
476
477	result = NULL;
478	badness = 0;
479	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
480	need_rescore = false;
481	rescore:
482	score = compute_score(sk: need_rescore ? result : sk, net, saddr,
483	sport, daddr, hnum, dif, sdif);
484	if (score > badness) {
485	badness = score;
486
487	if (need_rescore)
488	continue;
489
490	if (sk->sk_state == TCP_ESTABLISHED) {
491	result = sk;
492	continue;
493	}
494
495	result = inet_lookup_reuseport(net, sk, skb, doff: sizeof(struct udphdr),
496	saddr, sport, daddr, hnum, ehashfn: udp_ehashfn);
497	if (!result) {
498	result = sk;
499	continue;
500	}
501
502	/* Fall back to scoring if group has connections */
503	if (!reuseport_has_conns(sk))
504	return result;
505
506	/* Reuseport logic returned an error, keep original score. */
507	if (IS_ERR(ptr: result))
508	continue;
509
510	/* compute_score is too long of a function to be
511	* inlined, and calling it again here yields
512	* measurable overhead for some
513	* workloads. Work around it by jumping
514	* backwards to rescore 'result'.
515	*/
516	need_rescore = true;
517	goto rescore;
518	}
519	}
520	return result;
521	}
522
523	#if IS_ENABLED(CONFIG_BASE_SMALL)
524	static struct sock *udp4_lib_lookup4(const struct net *net,
525	__be32 saddr, __be16 sport,
526	__be32 daddr, unsigned int hnum,
527	int dif, int sdif,
528	struct udp_table *udptable)
529	{
530	return NULL;
531	}
532
533	static void udp_rehash4(struct udp_table *udptable, struct sock *sk,
534	u16 newhash4)
535	{
536	}
537
538	static void udp_unhash4(struct udp_table *udptable, struct sock *sk)
539	{
540	}
541	#else /* !CONFIG_BASE_SMALL */
542	static struct sock *udp4_lib_lookup4(const struct net *net,
543	__be32 saddr, __be16 sport,
544	__be32 daddr, unsigned int hnum,
545	int dif, int sdif,
546	struct udp_table *udptable)
547	{
548	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
549	const struct hlist_nulls_node *node;
550	struct udp_hslot *hslot4;
551	unsigned int hash4, slot;
552	struct udp_sock *up;
553	struct sock *sk;
554
555	hash4 = udp_ehashfn(net, daddr, hnum, saddr, sport);
556	slot = hash4 & udptable->mask;
557	hslot4 = &udptable->hash4[slot];
558	INET_ADDR_COOKIE(acookie, saddr, daddr);
559
560	begin:
561	/* SLAB_TYPESAFE_BY_RCU not used, so we don't need to touch sk_refcnt */
562	udp_lrpa_for_each_entry_rcu(up, node, &hslot4->nulls_head) {
563	sk = (struct sock *)up;
564	if (inet_match(net, sk, acookie, ports, dif, sdif))
565	return sk;
566	}
567
568	/* if the nulls value we got at the end of this lookup is not the
569	* expected one, we must restart lookup. We probably met an item that
570	* was moved to another chain due to rehash.
571	*/
572	if (get_nulls_value(node) != slot)
573	goto begin;
574
575	return NULL;
576	}
577
578	/* udp_rehash4() only checks hslot4, and hash4_cnt is not processed. */
579	static void udp_rehash4(struct udp_table *udptable, struct sock *sk,
580	u16 newhash4)
581	{
582	struct udp_hslot *hslot4, *nhslot4;
583
584	hslot4 = udp_hashslot4(udptable, udp_sk(sk)->udp_lrpa_hash);
585	nhslot4 = udp_hashslot4(udptable, newhash4);
586	udp_sk(sk)->udp_lrpa_hash = newhash4;
587
588	if (hslot4 != nhslot4) {
589	spin_lock_bh(&hslot4->lock);
590	hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_lrpa_node);
591	hslot4->count--;
592	spin_unlock_bh(&hslot4->lock);
593
594	spin_lock_bh(&nhslot4->lock);
595	hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_lrpa_node,
596	&nhslot4->nulls_head);
597	nhslot4->count++;
598	spin_unlock_bh(&nhslot4->lock);
599	}
600	}
601
602	static void udp_unhash4(struct udp_table *udptable, struct sock *sk)
603	{
604	struct udp_hslot *hslot2, *hslot4;
605
606	if (udp_hashed4(sk)) {
607	hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
608	hslot4 = udp_hashslot4(udptable, udp_sk(sk)->udp_lrpa_hash);
609
610	spin_lock(&hslot4->lock);
611	hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_lrpa_node);
612	hslot4->count--;
613	spin_unlock(&hslot4->lock);
614
615	spin_lock(&hslot2->lock);
616	udp_hash4_dec(hslot2);
617	spin_unlock(&hslot2->lock);
618	}
619	}
620
621	void udp_lib_hash4(struct sock *sk, u16 hash)
622	{
623	struct udp_hslot *hslot, *hslot2, *hslot4;
624	struct net *net = sock_net(sk);
625	struct udp_table *udptable;
626
627	/* Connected udp socket can re-connect to another remote address, which
628	* will be handled by rehash. Thus no need to redo hash4 here.
629	*/
630	if (udp_hashed4(sk))
631	return;
632
633	udptable = net->ipv4.udp_table;
634	hslot = udp_hashslot(udptable, net, udp_sk(sk)->udp_port_hash);
635	hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
636	hslot4 = udp_hashslot4(udptable, hash);
637	udp_sk(sk)->udp_lrpa_hash = hash;
638
639	spin_lock_bh(&hslot->lock);
640	if (rcu_access_pointer(sk->sk_reuseport_cb))
641	reuseport_detach_sock(sk);
642
643	spin_lock(&hslot4->lock);
644	hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_lrpa_node,
645	&hslot4->nulls_head);
646	hslot4->count++;
647	spin_unlock(&hslot4->lock);
648
649	spin_lock(&hslot2->lock);
650	udp_hash4_inc(hslot2);
651	spin_unlock(&hslot2->lock);
652
653	spin_unlock_bh(&hslot->lock);
654	}
655	EXPORT_IPV6_MOD(udp_lib_hash4);
656
657	/* call with sock lock */
658	void udp4_hash4(struct sock *sk)
659	{
660	struct net *net = sock_net(sk);
661	unsigned int hash;
662
663	if (sk_unhashed(sk) || sk->sk_rcv_saddr == htonl(INADDR_ANY))
664	return;
665
666	hash = udp_ehashfn(net, sk->sk_rcv_saddr, sk->sk_num,
667	sk->sk_daddr, sk->sk_dport);
668
669	udp_lib_hash4(sk, hash);
670	}
671	EXPORT_IPV6_MOD(udp4_hash4);
672	#endif /* CONFIG_BASE_SMALL */
673
674	/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
675	* harder than this. -DaveM
676	*/
677	struct sock *__udp4_lib_lookup(const struct net *net, __be32 saddr,
678	__be16 sport, __be32 daddr, __be16 dport, int dif,
679	int sdif, struct udp_table *udptable, struct sk_buff *skb)
680	{
681	unsigned short hnum = ntohs(dport);
682	struct udp_hslot *hslot2;
683	struct sock *result, *sk;
684	unsigned int hash2;
685
686	hash2 = ipv4_portaddr_hash(net, saddr: daddr, port: hnum);
687	hslot2 = udp_hashslot2(table: udptable, hash: hash2);
688
689	if (udp_has_hash4(hslot2)) {
690	result = udp4_lib_lookup4(net, saddr, sport, daddr, hnum,
691	dif, sdif, udptable);
692	if (result) /* udp4_lib_lookup4 return sk or NULL */
693	return result;
694	}
695
696	/* Lookup connected or non-wildcard socket */
697	result = udp4_lib_lookup2(net, saddr, sport,
698	daddr, hnum, dif, sdif,
699	hslot2, skb);
700	if (!IS_ERR_OR_NULL(ptr: result) && result->sk_state == TCP_ESTABLISHED)
701	goto done;
702
703	/* Lookup redirect from BPF */
704	if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
705	udptable == net->ipv4.udp_table) {
706	sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, doff: sizeof(struct udphdr),
707	saddr, sport, daddr, hnum, dif,
708	ehashfn: udp_ehashfn);
709	if (sk) {
710	result = sk;
711	goto done;
712	}
713	}
714
715	/* Got non-wildcard socket or error on first lookup */
716	if (result)
717	goto done;
718
719	/* Lookup wildcard sockets */
720	hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), port: hnum);
721	hslot2 = udp_hashslot2(table: udptable, hash: hash2);
722
723	result = udp4_lib_lookup2(net, saddr, sport,
724	htonl(INADDR_ANY), hnum, dif, sdif,
725	hslot2, skb);
726	if (!IS_ERR_OR_NULL(ptr: result))
727	goto done;
728
729	/* Primary hash (destination port) lookup as fallback for this race:
730	* 1. __ip4_datagram_connect() sets sk_rcv_saddr
731	* 2. lookup (this function): new sk_rcv_saddr, hashes not updated yet
732	* 3. rehash operation updating _secondary and four-tuple_ hashes
733	* The primary hash doesn't need an update after 1., so, thanks to this
734	* further step, 1. and 3. don't need to be atomic against the lookup.
735	*/
736	result = udp4_lib_lookup1(net, saddr, sport, daddr, hnum, dif, sdif,
737	udptable);
738
739	done:
740	if (IS_ERR(ptr: result))
741	return NULL;
742	return result;
743	}
744	EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
745
746	static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
747	__be16 sport, __be16 dport,
748	struct udp_table *udptable)
749	{
750	const struct iphdr *iph = ip_hdr(skb);
751
752	return __udp4_lib_lookup(dev_net(dev: skb->dev), iph->saddr, sport,
753	iph->daddr, dport, inet_iif(skb),
754	inet_sdif(skb), udptable, skb);
755	}
756
757	struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
758	__be16 sport, __be16 dport)
759	{
760	const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
761	const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
762	struct net *net = dev_net(dev: skb->dev);
763	int iif, sdif;
764
765	inet_get_iif_sdif(skb, iif: &iif, sdif: &sdif);
766
767	return __udp4_lib_lookup(net, iph->saddr, sport,
768	iph->daddr, dport, iif,
769	sdif, net->ipv4.udp_table, NULL);
770	}
771
772	/* Must be called under rcu_read_lock().
773	* Does increment socket refcount.
774	*/
775	#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
776	struct sock *udp4_lib_lookup(const struct net *net, __be32 saddr, __be16 sport,
777	__be32 daddr, __be16 dport, int dif)
778	{
779	struct sock *sk;
780
781	sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
782	dif, 0, net->ipv4.udp_table, NULL);
783	if (sk && !refcount_inc_not_zero(r: &sk->sk_refcnt))
784	sk = NULL;
785	return sk;
786	}
787	EXPORT_SYMBOL_GPL(udp4_lib_lookup);
788	#endif
789
790	static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
791	__be16 loc_port, __be32 loc_addr,
792	__be16 rmt_port, __be32 rmt_addr,
793	int dif, int sdif, unsigned short hnum)
794	{
795	const struct inet_sock *inet = inet_sk(sk);
796
797	if (!net_eq(net1: sock_net(sk), net2: net) ||
798	udp_sk(sk)->udp_port_hash != hnum ||
799	(inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
800	(inet->inet_dport != rmt_port && inet->inet_dport) ||
801	(inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
802	ipv6_only_sock(sk) ||
803	!udp_sk_bound_dev_eq(net, bound_dev_if: sk->sk_bound_dev_if, dif, sdif))
804	return false;
805	if (!ip_mc_sf_allow(sk, local: loc_addr, rmt: rmt_addr, dif, sdif))
806	return false;
807	return true;
808	}
809
810	DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
811	EXPORT_IPV6_MOD(udp_encap_needed_key);
812
813	#if IS_ENABLED(CONFIG_IPV6)
814	DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
815	EXPORT_IPV6_MOD(udpv6_encap_needed_key);
816	#endif
817
818	void udp_encap_enable(void)
819	{
820	static_branch_inc(&udp_encap_needed_key);
821	}
822	EXPORT_SYMBOL(udp_encap_enable);
823
824	void udp_encap_disable(void)
825	{
826	static_branch_dec(&udp_encap_needed_key);
827	}
828	EXPORT_SYMBOL(udp_encap_disable);
829
830	/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
831	* through error handlers in encapsulations looking for a match.
832	*/
833	static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
834	{
835	int i;
836
837	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
838	int (*handler)(struct sk_buff *skb, u32 info);
839	const struct ip_tunnel_encap_ops *encap;
840
841	encap = rcu_dereference(iptun_encaps[i]);
842	if (!encap)
843	continue;
844	handler = encap->err_handler;
845	if (handler && !handler(skb, info))
846	return 0;
847	}
848
849	return -ENOENT;
850	}
851
852	/* Try to match ICMP errors to UDP tunnels by looking up a socket without
853	* reversing source and destination port: this will match tunnels that force the
854	* same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
855	* lwtunnels might actually break this assumption by being configured with
856	* different destination ports on endpoints, in this case we won't be able to
857	* trace ICMP messages back to them.
858	*
859	* If this doesn't match any socket, probe tunnels with arbitrary destination
860	* ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
861	* we've sent packets to won't necessarily match the local destination port.
862	*
863	* Then ask the tunnel implementation to match the error against a valid
864	* association.
865	*
866	* Return an error if we can't find a match, the socket if we need further
867	* processing, zero otherwise.
868	*/
869	static struct sock *__udp4_lib_err_encap(struct net *net,
870	const struct iphdr *iph,
871	struct udphdr *uh,
872	struct udp_table *udptable,
873	struct sock *sk,
874	struct sk_buff *skb, u32 info)
875	{
876	int (*lookup)(struct sock *sk, struct sk_buff *skb);
877	int network_offset, transport_offset;
878	struct udp_sock *up;
879
880	network_offset = skb_network_offset(skb);
881	transport_offset = skb_transport_offset(skb);
882
883	/* Network header needs to point to the outer IPv4 header inside ICMP */
884	skb_reset_network_header(skb);
885
886	/* Transport header needs to point to the UDP header */
887	skb_set_transport_header(skb, offset: iph->ihl << 2);
888
889	if (sk) {
890	up = udp_sk(sk);
891
892	lookup = READ_ONCE(up->encap_err_lookup);
893	if (lookup && lookup(sk, skb))
894	sk = NULL;
895
896	goto out;
897	}
898
899	sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
900	iph->saddr, uh->dest, skb->dev->ifindex, 0,
901	udptable, NULL);
902	if (sk) {
903	up = udp_sk(sk);
904
905	lookup = READ_ONCE(up->encap_err_lookup);
906	if (!lookup || lookup(sk, skb))
907	sk = NULL;
908	}
909
910	out:
911	if (!sk)
912	sk = ERR_PTR(error: __udp4_lib_err_encap_no_sk(skb, info));
913
914	skb_set_transport_header(skb, offset: transport_offset);
915	skb_set_network_header(skb, offset: network_offset);
916
917	return sk;
918	}
919
920	/*
921	* This routine is called by the ICMP module when it gets some
922	* sort of error condition. If err < 0 then the socket should
923	* be closed and the error returned to the user. If err > 0
924	* it's just the icmp type << 8 | icmp code.
925	* Header points to the ip header of the error packet. We move
926	* on past this. Then (as it used to claim before adjustment)
927	* header points to the first 8 bytes of the udp header. We need
928	* to find the appropriate port.
929	*/
930
931	int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
932	{
933	struct inet_sock *inet;
934	const struct iphdr *iph = (const struct iphdr *)skb->data;
935	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
936	const int type = icmp_hdr(skb)->type;
937	const int code = icmp_hdr(skb)->code;
938	bool tunnel = false;
939	struct sock *sk;
940	int harderr;
941	int err;
942	struct net *net = dev_net(dev: skb->dev);
943
944	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
945	iph->saddr, uh->source, skb->dev->ifindex,
946	inet_sdif(skb), udptable, NULL);
947
948	if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
949	/* No socket for error: try tunnels before discarding */
950	if (static_branch_unlikely(&udp_encap_needed_key)) {
951	sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
952	info);
953	if (!sk)
954	return 0;
955	} else
956	sk = ERR_PTR(error: -ENOENT);
957
958	if (IS_ERR(ptr: sk)) {
959	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
960	return PTR_ERR(ptr: sk);
961	}
962
963	tunnel = true;
964	}
965
966	err = 0;
967	harderr = 0;
968	inet = inet_sk(sk);
969
970	switch (type) {
971	default:
972	case ICMP_TIME_EXCEEDED:
973	err = EHOSTUNREACH;
974	break;
975	case ICMP_SOURCE_QUENCH:
976	goto out;
977	case ICMP_PARAMETERPROB:
978	err = EPROTO;
979	harderr = 1;
980	break;
981	case ICMP_DEST_UNREACH:
982	if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
983	ipv4_sk_update_pmtu(skb, sk, mtu: info);
984	if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
985	err = EMSGSIZE;
986	harderr = 1;
987	break;
988	}
989	goto out;
990	}
991	err = EHOSTUNREACH;
992	if (code <= NR_ICMP_UNREACH) {
993	harderr = icmp_err_convert[code].fatal;
994	err = icmp_err_convert[code].errno;
995	}
996	break;
997	case ICMP_REDIRECT:
998	ipv4_sk_redirect(skb, sk);
999	goto out;
1000	}
1001
1002	/*
1003	* RFC1122: OK. Passes ICMP errors back to application, as per
1004	* 4.1.3.3.
1005	*/
1006	if (tunnel) {
1007	/* ...not for tunnels though: we don't have a sending socket */
1008	if (udp_sk(sk)->encap_err_rcv)
1009	udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
1010	(u8 *)(uh+1));
1011	goto out;
1012	}
1013	if (!inet_test_bit(RECVERR, sk)) {
1014	if (!harderr || sk->sk_state != TCP_ESTABLISHED)
1015	goto out;
1016	} else
1017	ip_icmp_error(sk, skb, err, port: uh->dest, info, payload: (u8 *)(uh+1));
1018
1019	sk->sk_err = err;
1020	sk_error_report(sk);
1021	out:
1022	return 0;
1023	}
1024
1025	int udp_err(struct sk_buff *skb, u32 info)
1026	{
1027	return __udp4_lib_err(skb, info, udptable: dev_net(dev: skb->dev)->ipv4.udp_table);
1028	}
1029
1030	/*
1031	* Throw away all pending data and cancel the corking. Socket is locked.
1032	*/
1033	void udp_flush_pending_frames(struct sock *sk)
1034	{
1035	struct udp_sock *up = udp_sk(sk);
1036
1037	if (up->pending) {
1038	up->len = 0;
1039	WRITE_ONCE(up->pending, 0);
1040	ip_flush_pending_frames(sk);
1041	}
1042	}
1043	EXPORT_IPV6_MOD(udp_flush_pending_frames);
1044
1045	/**
1046	* udp4_hwcsum - handle outgoing HW checksumming
1047	* @skb: sk_buff containing the filled-in UDP header
1048	* (checksum field must be zeroed out)
1049	* @src: source IP address
1050	* @dst: destination IP address
1051	*/
1052	void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
1053	{
1054	struct udphdr *uh = udp_hdr(skb);
1055	int offset = skb_transport_offset(skb);
1056	int len = skb->len - offset;
1057	int hlen = len;
1058	__wsum csum = 0;
1059
1060	if (!skb_has_frag_list(skb)) {
1061	/*
1062	* Only one fragment on the socket.
1063	*/
1064	skb->csum_start = skb_transport_header(skb) - skb->head;
1065	skb->csum_offset = offsetof(struct udphdr, check);
1066	uh->check = ~csum_tcpudp_magic(saddr: src, daddr: dst, len,
1067	IPPROTO_UDP, sum: 0);
1068	} else {
1069	struct sk_buff *frags;
1070
1071	/*
1072	* HW-checksum won't work as there are two or more
1073	* fragments on the socket so that all csums of sk_buffs
1074	* should be together
1075	*/
1076	skb_walk_frags(skb, frags) {
1077	csum = csum_add(csum, addend: frags->csum);
1078	hlen -= frags->len;
1079	}
1080
1081	csum = skb_checksum(skb, offset, len: hlen, csum);
1082	skb->ip_summed = CHECKSUM_NONE;
1083
1084	uh->check = csum_tcpudp_magic(saddr: src, daddr: dst, len, IPPROTO_UDP, sum: csum);
1085	if (uh->check == 0)
1086	uh->check = CSUM_MANGLED_0;
1087	}
1088	}
1089	EXPORT_SYMBOL_GPL(udp4_hwcsum);
1090
1091	/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
1092	* for the simple case like when setting the checksum for a UDP tunnel.
1093	*/
1094	void udp_set_csum(bool nocheck, struct sk_buff *skb,
1095	__be32 saddr, __be32 daddr, int len)
1096	{
1097	struct udphdr *uh = udp_hdr(skb);
1098
1099	if (nocheck) {
1100	uh->check = 0;
1101	} else if (skb_is_gso(skb)) {
1102	uh->check = ~udp_v4_check(len, saddr, daddr, base: 0);
1103	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
1104	uh->check = 0;
1105	uh->check = udp_v4_check(len, saddr, daddr, base: lco_csum(skb));
1106	if (uh->check == 0)
1107	uh->check = CSUM_MANGLED_0;
1108	} else {
1109	skb->ip_summed = CHECKSUM_PARTIAL;
1110	skb->csum_start = skb_transport_header(skb) - skb->head;
1111	skb->csum_offset = offsetof(struct udphdr, check);
1112	uh->check = ~udp_v4_check(len, saddr, daddr, base: 0);
1113	}
1114	}
1115	EXPORT_SYMBOL(udp_set_csum);
1116
1117	static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
1118	struct inet_cork *cork)
1119	{
1120	struct sock *sk = skb->sk;
1121	struct inet_sock *inet = inet_sk(sk);
1122	struct udphdr *uh;
1123	int err;
1124	int is_udplite = IS_UDPLITE(sk);
1125	int offset = skb_transport_offset(skb);
1126	int len = skb->len - offset;
1127	int datalen = len - sizeof(*uh);
1128	__wsum csum = 0;
1129
1130	/*
1131	* Create a UDP header
1132	*/
1133	uh = udp_hdr(skb);
1134	uh->source = inet->inet_sport;
1135	uh->dest = fl4->fl4_dport;
1136	uh->len = htons(len);
1137	uh->check = 0;
1138
1139	if (cork->gso_size) {
1140	const int hlen = skb_network_header_len(skb) +
1141	sizeof(struct udphdr);
1142
1143	if (hlen + min(datalen, cork->gso_size) > cork->fragsize) {
1144	kfree_skb(skb);
1145	return -EMSGSIZE;
1146	}
1147	if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1148	kfree_skb(skb);
1149	return -EINVAL;
1150	}
1151	if (sk->sk_no_check_tx) {
1152	kfree_skb(skb);
1153	return -EINVAL;
1154	}
1155	if (is_udplite || dst_xfrm(dst: skb_dst(skb))) {
1156	kfree_skb(skb);
1157	return -EIO;
1158	}
1159
1160	if (datalen > cork->gso_size) {
1161	skb_shinfo(skb)->gso_size = cork->gso_size;
1162	skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1163	skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1164	cork->gso_size);
1165
1166	/* Don't checksum the payload, skb will get segmented */
1167	goto csum_partial;
1168	}
1169	}
1170
1171	if (is_udplite) /* UDP-Lite */
1172	csum = udplite_csum(skb);
1173
1174	else if (sk->sk_no_check_tx) { /* UDP csum off */
1175
1176	skb->ip_summed = CHECKSUM_NONE;
1177	goto send;
1178
1179	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1180	csum_partial:
1181
1182	udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
1183	goto send;
1184
1185	} else
1186	csum = udp_csum(skb);
1187
1188	/* add protocol-dependent pseudo-header */
1189	uh->check = csum_tcpudp_magic(saddr: fl4->saddr, daddr: fl4->daddr, len,
1190	proto: sk->sk_protocol, sum: csum);
1191	if (uh->check == 0)
1192	uh->check = CSUM_MANGLED_0;
1193
1194	send:
1195	err = ip_send_skb(net: sock_net(sk), skb);
1196	if (err) {
1197	if (err == -ENOBUFS &&
1198	!inet_test_bit(RECVERR, sk)) {
1199	UDP_INC_STATS(sock_net(sk),
1200	UDP_MIB_SNDBUFERRORS, is_udplite);
1201	err = 0;
1202	}
1203	} else
1204	UDP_INC_STATS(sock_net(sk),
1205	UDP_MIB_OUTDATAGRAMS, is_udplite);
1206	return err;
1207	}
1208
1209	/*
1210	* Push out all pending data as one UDP datagram. Socket is locked.
1211	*/
1212	int udp_push_pending_frames(struct sock *sk)
1213	{
1214	struct udp_sock *up = udp_sk(sk);
1215	struct inet_sock *inet = inet_sk(sk);
1216	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1217	struct sk_buff *skb;
1218	int err = 0;
1219
1220	skb = ip_finish_skb(sk, fl4);
1221	if (!skb)
1222	goto out;
1223
1224	err = udp_send_skb(skb, fl4, cork: &inet->cork.base);
1225
1226	out:
1227	up->len = 0;
1228	WRITE_ONCE(up->pending, 0);
1229	return err;
1230	}
1231	EXPORT_IPV6_MOD(udp_push_pending_frames);
1232
1233	static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1234	{
1235	switch (cmsg->cmsg_type) {
1236	case UDP_SEGMENT:
1237	if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1238	return -EINVAL;
1239	*gso_size = *(__u16 *)CMSG_DATA(cmsg);
1240	return 0;
1241	default:
1242	return -EINVAL;
1243	}
1244	}
1245
1246	int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1247	{
1248	struct cmsghdr *cmsg;
1249	bool need_ip = false;
1250	int err;
1251
1252	for_each_cmsghdr(cmsg, msg) {
1253	if (!CMSG_OK(msg, cmsg))
1254	return -EINVAL;
1255
1256	if (cmsg->cmsg_level != SOL_UDP) {
1257	need_ip = true;
1258	continue;
1259	}
1260
1261	err = __udp_cmsg_send(cmsg, gso_size);
1262	if (err)
1263	return err;
1264	}
1265
1266	return need_ip;
1267	}
1268	EXPORT_IPV6_MOD_GPL(udp_cmsg_send);
1269
1270	int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1271	{
1272	struct inet_sock *inet = inet_sk(sk);
1273	struct udp_sock *up = udp_sk(sk);
1274	DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1275	struct flowi4 fl4_stack;
1276	struct flowi4 *fl4;
1277	int ulen = len;
1278	struct ipcm_cookie ipc;
1279	struct rtable *rt = NULL;
1280	int free = 0;
1281	int connected = 0;
1282	__be32 daddr, faddr, saddr;
1283	u8 scope;
1284	__be16 dport;
1285	int err, is_udplite = IS_UDPLITE(sk);
1286	int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1287	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1288	struct sk_buff *skb;
1289	struct ip_options_data opt_copy;
1290	int uc_index;
1291
1292	if (len > 0xFFFF)
1293	return -EMSGSIZE;
1294
1295	/*
1296	* Check the flags.
1297	*/
1298
1299	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1300	return -EOPNOTSUPP;
1301
1302	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1303
1304	fl4 = &inet->cork.fl.u.ip4;
1305	if (READ_ONCE(up->pending)) {
1306	/*
1307	* There are pending frames.
1308	* The socket lock must be held while it's corked.
1309	*/
1310	lock_sock(sk);
1311	if (likely(up->pending)) {
1312	if (unlikely(up->pending != AF_INET)) {
1313	release_sock(sk);
1314	return -EINVAL;
1315	}
1316	goto do_append_data;
1317	}
1318	release_sock(sk);
1319	}
1320	ulen += sizeof(struct udphdr);
1321
1322	/*
1323	* Get and verify the address.
1324	*/
1325	if (usin) {
1326	if (msg->msg_namelen < sizeof(*usin))
1327	return -EINVAL;
1328	if (usin->sin_family != AF_INET) {
1329	if (usin->sin_family != AF_UNSPEC)
1330	return -EAFNOSUPPORT;
1331	}
1332
1333	daddr = usin->sin_addr.s_addr;
1334	dport = usin->sin_port;
1335	if (dport == 0)
1336	return -EINVAL;
1337	} else {
1338	if (sk->sk_state != TCP_ESTABLISHED)
1339	return -EDESTADDRREQ;
1340	daddr = inet->inet_daddr;
1341	dport = inet->inet_dport;
1342	/* Open fast path for connected socket.
1343	Route will not be used, if at least one option is set.
1344	*/
1345	connected = 1;
1346	}
1347
1348	ipcm_init_sk(ipcm: &ipc, inet);
1349	ipc.gso_size = READ_ONCE(up->gso_size);
1350
1351	if (msg->msg_controllen) {
1352	err = udp_cmsg_send(sk, msg, gso_size: &ipc.gso_size);
1353	if (err > 0) {
1354	err = ip_cmsg_send(sk, msg, ipc: &ipc,
1355	allow_ipv6: sk->sk_family == AF_INET6);
1356	connected = 0;
1357	}
1358	if (unlikely(err < 0)) {
1359	kfree(objp: ipc.opt);
1360	return err;
1361	}
1362	if (ipc.opt)
1363	free = 1;
1364	}
1365	if (!ipc.opt) {
1366	struct ip_options_rcu *inet_opt;
1367
1368	rcu_read_lock();
1369	inet_opt = rcu_dereference(inet->inet_opt);
1370	if (inet_opt) {
1371	memcpy(&opt_copy, inet_opt,
1372	sizeof(*inet_opt) + inet_opt->opt.optlen);
1373	ipc.opt = &opt_copy.opt;
1374	}
1375	rcu_read_unlock();
1376	}
1377
1378	if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1379	err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1380	(struct sockaddr *)usin,
1381	&msg->msg_namelen,
1382	&ipc.addr);
1383	if (err)
1384	goto out_free;
1385	if (usin) {
1386	if (usin->sin_port == 0) {
1387	/* BPF program set invalid port. Reject it. */
1388	err = -EINVAL;
1389	goto out_free;
1390	}
1391	daddr = usin->sin_addr.s_addr;
1392	dport = usin->sin_port;
1393	}
1394	}
1395
1396	saddr = ipc.addr;
1397	ipc.addr = faddr = daddr;
1398
1399	if (ipc.opt && ipc.opt->opt.srr) {
1400	if (!daddr) {
1401	err = -EINVAL;
1402	goto out_free;
1403	}
1404	faddr = ipc.opt->opt.faddr;
1405	connected = 0;
1406	}
1407	scope = ip_sendmsg_scope(inet, ipc: &ipc, msg);
1408	if (scope == RT_SCOPE_LINK)
1409	connected = 0;
1410
1411	uc_index = READ_ONCE(inet->uc_index);
1412	if (ipv4_is_multicast(addr: daddr)) {
1413	if (!ipc.oif || netif_index_is_l3_master(net: sock_net(sk), ifindex: ipc.oif))
1414	ipc.oif = READ_ONCE(inet->mc_index);
1415	if (!saddr)
1416	saddr = READ_ONCE(inet->mc_addr);
1417	connected = 0;
1418	} else if (!ipc.oif) {
1419	ipc.oif = uc_index;
1420	} else if (ipv4_is_lbcast(addr: daddr) && uc_index) {
1421	/* oif is set, packet is to local broadcast and
1422	* uc_index is set. oif is most likely set
1423	* by sk_bound_dev_if. If uc_index != oif check if the
1424	* oif is an L3 master and uc_index is an L3 slave.
1425	* If so, we want to allow the send using the uc_index.
1426	*/
1427	if (ipc.oif != uc_index &&
1428	ipc.oif == l3mdev_master_ifindex_by_index(net: sock_net(sk),
1429	ifindex: uc_index)) {
1430	ipc.oif = uc_index;
1431	}
1432	}
1433
1434	if (connected)
1435	rt = dst_rtable(sk_dst_check(sk, 0));
1436
1437	if (!rt) {
1438	struct net *net = sock_net(sk);
1439	__u8 flow_flags = inet_sk_flowi_flags(sk);
1440
1441	fl4 = &fl4_stack;
1442
1443	flowi4_init_output(fl4, oif: ipc.oif, mark: ipc.sockc.mark,
1444	tos: ipc.tos & INET_DSCP_MASK, scope,
1445	proto: sk->sk_protocol, flags: flow_flags, daddr: faddr, saddr,
1446	dport, sport: inet->inet_sport,
1447	uid: sk_uid(sk));
1448
1449	security_sk_classify_flow(sk, flic: flowi4_to_flowi_common(fl4));
1450	rt = ip_route_output_flow(net, flp: fl4, sk);
1451	if (IS_ERR(ptr: rt)) {
1452	err = PTR_ERR(ptr: rt);
1453	rt = NULL;
1454	if (err == -ENETUNREACH)
1455	IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1456	goto out;
1457	}
1458
1459	err = -EACCES;
1460	if ((rt->rt_flags & RTCF_BROADCAST) &&
1461	!sock_flag(sk, flag: SOCK_BROADCAST))
1462	goto out;
1463	if (connected)
1464	sk_dst_set(sk, dst: dst_clone(dst: &rt->dst));
1465	}
1466
1467	if (msg->msg_flags&MSG_CONFIRM)
1468	goto do_confirm;
1469	back_from_confirm:
1470
1471	saddr = fl4->saddr;
1472	if (!ipc.addr)
1473	daddr = ipc.addr = fl4->daddr;
1474
1475	/* Lockless fast path for the non-corking case. */
1476	if (!corkreq) {
1477	struct inet_cork cork;
1478
1479	skb = ip_make_skb(sk, fl4, getfrag, from: msg, length: ulen,
1480	transhdrlen: sizeof(struct udphdr), ipc: &ipc, rtp: &rt,
1481	cork: &cork, flags: msg->msg_flags);
1482	err = PTR_ERR(ptr: skb);
1483	if (!IS_ERR_OR_NULL(ptr: skb))
1484	err = udp_send_skb(skb, fl4, cork: &cork);
1485	goto out;
1486	}
1487
1488	lock_sock(sk);
1489	if (unlikely(up->pending)) {
1490	/* The socket is already corked while preparing it. */
1491	/* ... which is an evident application bug. --ANK */
1492	release_sock(sk);
1493
1494	net_dbg_ratelimited("socket already corked\n");
1495	err = -EINVAL;
1496	goto out;
1497	}
1498	/*
1499	* Now cork the socket to pend data.
1500	*/
1501	fl4 = &inet->cork.fl.u.ip4;
1502	fl4->daddr = daddr;
1503	fl4->saddr = saddr;
1504	fl4->fl4_dport = dport;
1505	fl4->fl4_sport = inet->inet_sport;
1506	WRITE_ONCE(up->pending, AF_INET);
1507
1508	do_append_data:
1509	up->len += ulen;
1510	err = ip_append_data(sk, fl4, getfrag, from: msg, len: ulen,
1511	protolen: sizeof(struct udphdr), ipc: &ipc, rt: &rt,
1512	flags: corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1513	if (err)
1514	udp_flush_pending_frames(sk);
1515	else if (!corkreq)
1516	err = udp_push_pending_frames(sk);
1517	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1518	WRITE_ONCE(up->pending, 0);
1519	release_sock(sk);
1520
1521	out:
1522	ip_rt_put(rt);
1523	out_free:
1524	if (free)
1525	kfree(objp: ipc.opt);
1526	if (!err)
1527	return len;
1528	/*
1529	* ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1530	* ENOBUFS might not be good (it's not tunable per se), but otherwise
1531	* we don't have a good statistic (IpOutDiscards but it can be too many
1532	* things). We could add another new stat but at least for now that
1533	* seems like overkill.
1534	*/
1535	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1536	UDP_INC_STATS(sock_net(sk),
1537	UDP_MIB_SNDBUFERRORS, is_udplite);
1538	}
1539	return err;
1540
1541	do_confirm:
1542	if (msg->msg_flags & MSG_PROBE)
1543	dst_confirm_neigh(dst: &rt->dst, daddr: &fl4->daddr);
1544	if (!(msg->msg_flags&MSG_PROBE) || len)
1545	goto back_from_confirm;
1546	err = 0;
1547	goto out;
1548	}
1549	EXPORT_SYMBOL(udp_sendmsg);
1550
1551	void udp_splice_eof(struct socket *sock)
1552	{
1553	struct sock *sk = sock->sk;
1554	struct udp_sock *up = udp_sk(sk);
1555
1556	if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1557	return;
1558
1559	lock_sock(sk);
1560	if (up->pending && !udp_test_bit(CORK, sk))
1561	udp_push_pending_frames(sk);
1562	release_sock(sk);
1563	}
1564	EXPORT_IPV6_MOD_GPL(udp_splice_eof);
1565
1566	#define UDP_SKB_IS_STATELESS 0x80000000
1567
1568	/* all head states (dst, sk, nf conntrack) except skb extensions are
1569	* cleared by udp_rcv().
1570	*
1571	* We need to preserve secpath, if present, to eventually process
1572	* IP_CMSG_PASSSEC at recvmsg() time.
1573	*
1574	* Other extensions can be cleared.
1575	*/
1576	static bool udp_try_make_stateless(struct sk_buff *skb)
1577	{
1578	if (!skb_has_extensions(skb))
1579	return true;
1580
1581	if (!secpath_exists(skb)) {
1582	skb_ext_reset(skb);
1583	return true;
1584	}
1585
1586	return false;
1587	}
1588
1589	static void udp_set_dev_scratch(struct sk_buff *skb)
1590	{
1591	struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1592
1593	BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1594	scratch->_tsize_state = skb->truesize;
1595	#if BITS_PER_LONG == 64
1596	scratch->len = skb->len;
1597	scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1598	scratch->is_linear = !skb_is_nonlinear(skb);
1599	#endif
1600	if (udp_try_make_stateless(skb))
1601	scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1602	}
1603
1604	static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1605	{
1606	/* We come here after udp_lib_checksum_complete() returned 0.
1607	* This means that __skb_checksum_complete() might have
1608	* set skb->csum_valid to 1.
1609	* On 64bit platforms, we can set csum_unnecessary
1610	* to true, but only if the skb is not shared.
1611	*/
1612	#if BITS_PER_LONG == 64
1613	if (!skb_shared(skb))
1614	udp_skb_scratch(skb)->csum_unnecessary = true;
1615	#endif
1616	}
1617
1618	static int udp_skb_truesize(struct sk_buff *skb)
1619	{
1620	return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1621	}
1622
1623	static bool udp_skb_has_head_state(struct sk_buff *skb)
1624	{
1625	return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1626	}
1627
1628	/* fully reclaim rmem/fwd memory allocated for skb */
1629	static void udp_rmem_release(struct sock *sk, unsigned int size,
1630	int partial, bool rx_queue_lock_held)
1631	{
1632	struct udp_sock *up = udp_sk(sk);
1633	struct sk_buff_head *sk_queue;
1634	unsigned int amt;
1635
1636	if (likely(partial)) {
1637	up->forward_deficit += size;
1638	size = up->forward_deficit;
1639	if (size < READ_ONCE(up->forward_threshold) &&
1640	!skb_queue_empty(list: &up->reader_queue))
1641	return;
1642	} else {
1643	size += up->forward_deficit;
1644	}
1645	up->forward_deficit = 0;
1646
1647	/* acquire the sk_receive_queue for fwd allocated memory scheduling,
1648	* if the called don't held it already
1649	*/
1650	sk_queue = &sk->sk_receive_queue;
1651	if (!rx_queue_lock_held)
1652	spin_lock(lock: &sk_queue->lock);
1653
1654	amt = (size + sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1655	sk_forward_alloc_add(sk, val: size - amt);
1656
1657	if (amt)
1658	__sk_mem_reduce_allocated(sk, amount: amt >> PAGE_SHIFT);
1659
1660	atomic_sub(i: size, v: &sk->sk_rmem_alloc);
1661
1662	/* this can save us from acquiring the rx queue lock on next receive */
1663	skb_queue_splice_tail_init(list: sk_queue, head: &up->reader_queue);
1664
1665	if (!rx_queue_lock_held)
1666	spin_unlock(lock: &sk_queue->lock);
1667	}
1668
1669	/* Note: called with reader_queue.lock held.
1670	* Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1671	* This avoids a cache line miss while receive_queue lock is held.
1672	* Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1673	*/
1674	void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1675	{
1676	prefetch(&skb->data);
1677	udp_rmem_release(sk, size: udp_skb_truesize(skb), partial: 1, rx_queue_lock_held: false);
1678	}
1679	EXPORT_IPV6_MOD(udp_skb_destructor);
1680
1681	/* as above, but the caller held the rx queue lock, too */
1682	static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1683	{
1684	prefetch(&skb->data);
1685	udp_rmem_release(sk, size: udp_skb_truesize(skb), partial: 1, rx_queue_lock_held: true);
1686	}
1687
1688	static int udp_rmem_schedule(struct sock *sk, int size)
1689	{
1690	int delta;
1691
1692	delta = size - sk->sk_forward_alloc;
1693	if (delta > 0 && !__sk_mem_schedule(sk, size: delta, SK_MEM_RECV))
1694	return -ENOBUFS;
1695
1696	return 0;
1697	}
1698
1699	int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1700	{
1701	struct sk_buff_head *list = &sk->sk_receive_queue;
1702	struct udp_prod_queue *udp_prod_queue;
1703	struct sk_buff *next, *to_drop = NULL;
1704	struct llist_node *ll_list;
1705	unsigned int rmem, rcvbuf;
1706	int size, err = -ENOMEM;
1707	int total_size = 0;
1708	int q_size = 0;
1709	int dropcount;
1710	int nb = 0;
1711
1712	rmem = atomic_read(v: &sk->sk_rmem_alloc);
1713	rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1714	size = skb->truesize;
1715
1716	udp_prod_queue = &udp_sk(sk)->udp_prod_queue[numa_node_id()];
1717
1718	rmem += atomic_read(v: &udp_prod_queue->rmem_alloc);
1719
1720	/* Immediately drop when the receive queue is full.
1721	* Cast to unsigned int performs the boundary check for INT_MAX.
1722	*/
1723	if (rmem + size > rcvbuf) {
1724	if (rcvbuf > INT_MAX >> 1)
1725	goto drop;
1726
1727	/* Accept the packet if queue is empty. */
1728	if (rmem)
1729	goto drop;
1730	}
1731
1732	/* Under mem pressure, it might be helpful to help udp_recvmsg()
1733	* having linear skbs :
1734	* - Reduce memory overhead and thus increase receive queue capacity
1735	* - Less cache line misses at copyout() time
1736	* - Less work at consume_skb() (less alien page frag freeing)
1737	*/
1738	if (rmem > (rcvbuf >> 1)) {
1739	skb_condense(skb);
1740	size = skb->truesize;
1741	}
1742
1743	udp_set_dev_scratch(skb);
1744
1745	atomic_add(i: size, v: &udp_prod_queue->rmem_alloc);
1746
1747	if (!llist_add(new: &skb->ll_node, head: &udp_prod_queue->ll_root))
1748	return 0;
1749
1750	dropcount = sock_flag(sk, flag: SOCK_RXQ_OVFL) ? sk_drops_read(sk) : 0;
1751
1752	spin_lock(lock: &list->lock);
1753
1754	ll_list = llist_del_all(head: &udp_prod_queue->ll_root);
1755
1756	ll_list = llist_reverse_order(head: ll_list);
1757
1758	llist_for_each_entry_safe(skb, next, ll_list, ll_node) {
1759	size = udp_skb_truesize(skb);
1760	total_size += size;
1761	err = udp_rmem_schedule(sk, size);
1762	if (unlikely(err)) {
1763	/* Free the skbs outside of locked section. */
1764	skb->next = to_drop;
1765	to_drop = skb;
1766	continue;
1767	}
1768
1769	q_size += size;
1770	sk_forward_alloc_add(sk, val: -size);
1771
1772	/* no need to setup a destructor, we will explicitly release the
1773	* forward allocated memory on dequeue
1774	*/
1775	SOCK_SKB_CB(skb)->dropcount = dropcount;
1776	nb++;
1777	__skb_queue_tail(list, newsk: skb);
1778	}
1779
1780	atomic_add(i: q_size, v: &sk->sk_rmem_alloc);
1781
1782	spin_unlock(lock: &list->lock);
1783
1784	if (!sock_flag(sk, flag: SOCK_DEAD)) {
1785	/* Multiple threads might be blocked in recvmsg(),
1786	* using prepare_to_wait_exclusive().
1787	*/
1788	while (nb) {
1789	INDIRECT_CALL_1(sk->sk_data_ready,
1790	sock_def_readable, sk);
1791	nb--;
1792	}
1793	}
1794
1795	if (unlikely(to_drop)) {
1796	for (nb = 0; to_drop != NULL; nb++) {
1797	skb = to_drop;
1798	to_drop = skb->next;
1799	skb_mark_not_on_list(skb);
1800	/* TODO: update SNMP values. */
1801	sk_skb_reason_drop(sk, skb, reason: SKB_DROP_REASON_PROTO_MEM);
1802	}
1803	numa_drop_add(ndc: &udp_sk(sk)->drop_counters, val: nb);
1804	}
1805
1806	atomic_sub(i: total_size, v: &udp_prod_queue->rmem_alloc);
1807
1808	return 0;
1809
1810	drop:
1811	udp_drops_inc(sk);
1812	return err;
1813	}
1814	EXPORT_IPV6_MOD_GPL(__udp_enqueue_schedule_skb);
1815
1816	void udp_destruct_common(struct sock *sk)
1817	{
1818	/* reclaim completely the forward allocated memory */
1819	struct udp_sock *up = udp_sk(sk);
1820	unsigned int total = 0;
1821	struct sk_buff *skb;
1822
1823	skb_queue_splice_tail_init(list: &sk->sk_receive_queue, head: &up->reader_queue);
1824	while ((skb = __skb_dequeue(list: &up->reader_queue)) != NULL) {
1825	total += skb->truesize;
1826	kfree_skb(skb);
1827	}
1828	udp_rmem_release(sk, size: total, partial: 0, rx_queue_lock_held: true);
1829	kfree(objp: up->udp_prod_queue);
1830	}
1831	EXPORT_IPV6_MOD_GPL(udp_destruct_common);
1832
1833	static void udp_destruct_sock(struct sock *sk)
1834	{
1835	udp_destruct_common(sk);
1836	inet_sock_destruct(sk);
1837	}
1838
1839	int udp_init_sock(struct sock *sk)
1840	{
1841	int res = udp_lib_init_sock(sk);
1842
1843	sk->sk_destruct = udp_destruct_sock;
1844	set_bit(nr: SOCK_SUPPORT_ZC, addr: &sk->sk_socket->flags);
1845	return res;
1846	}
1847
1848	void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1849	{
1850	if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset)))
1851	sk_peek_offset_bwd(sk, val: len);
1852
1853	if (!skb_shared(skb)) {
1854	skb_orphan(skb);
1855	skb_attempt_defer_free(skb);
1856	return;
1857	}
1858
1859	if (!skb_unref(skb))
1860	return;
1861
1862	/* In the more common cases we cleared the head states previously,
1863	* see __udp_queue_rcv_skb().
1864	*/
1865	if (unlikely(udp_skb_has_head_state(skb)))
1866	skb_release_head_state(skb);
1867	__consume_stateless_skb(skb);
1868	}
1869	EXPORT_IPV6_MOD_GPL(skb_consume_udp);
1870
1871	static struct sk_buff *__first_packet_length(struct sock *sk,
1872	struct sk_buff_head *rcvq,
1873	unsigned int *total)
1874	{
1875	struct sk_buff *skb;
1876
1877	while ((skb = skb_peek(list_: rcvq)) != NULL) {
1878	if (udp_lib_checksum_complete(skb)) {
1879	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1880	IS_UDPLITE(sk));
1881	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1882	IS_UDPLITE(sk));
1883	udp_drops_inc(sk);
1884	__skb_unlink(skb, list: rcvq);
1885	*total += skb->truesize;
1886	kfree_skb_reason(skb, reason: SKB_DROP_REASON_UDP_CSUM);
1887	} else {
1888	udp_skb_csum_unnecessary_set(skb);
1889	break;
1890	}
1891	}
1892	return skb;
1893	}
1894
1895	/**
1896	* first_packet_length - return length of first packet in receive queue
1897	* @sk: socket
1898	*
1899	* Drops all bad checksum frames, until a valid one is found.
1900	* Returns the length of found skb, or -1 if none is found.
1901	*/
1902	static int first_packet_length(struct sock *sk)
1903	{
1904	struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1905	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1906	unsigned int total = 0;
1907	struct sk_buff *skb;
1908	int res;
1909
1910	spin_lock_bh(lock: &rcvq->lock);
1911	skb = __first_packet_length(sk, rcvq, total: &total);
1912	if (!skb && !skb_queue_empty_lockless(list: sk_queue)) {
1913	spin_lock(lock: &sk_queue->lock);
1914	skb_queue_splice_tail_init(list: sk_queue, head: rcvq);
1915	spin_unlock(lock: &sk_queue->lock);
1916
1917	skb = __first_packet_length(sk, rcvq, total: &total);
1918	}
1919	res = skb ? skb->len : -1;
1920	if (total)
1921	udp_rmem_release(sk, size: total, partial: 1, rx_queue_lock_held: false);
1922	spin_unlock_bh(lock: &rcvq->lock);
1923	return res;
1924	}
1925
1926	/*
1927	* IOCTL requests applicable to the UDP protocol
1928	*/
1929
1930	int udp_ioctl(struct sock *sk, int cmd, int *karg)
1931	{
1932	switch (cmd) {
1933	case SIOCOUTQ:
1934	{
1935	*karg = sk_wmem_alloc_get(sk);
1936	return 0;
1937	}
1938
1939	case SIOCINQ:
1940	{
1941	*karg = max_t(int, 0, first_packet_length(sk));
1942	return 0;
1943	}
1944
1945	default:
1946	return -ENOIOCTLCMD;
1947	}
1948
1949	return 0;
1950	}
1951	EXPORT_IPV6_MOD(udp_ioctl);
1952
1953	struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1954	int *off, int *err)
1955	{
1956	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1957	struct sk_buff_head *queue;
1958	struct sk_buff *last;
1959	long timeo;
1960	int error;
1961
1962	queue = &udp_sk(sk)->reader_queue;
1963	timeo = sock_rcvtimeo(sk, noblock: flags & MSG_DONTWAIT);
1964	do {
1965	struct sk_buff *skb;
1966
1967	error = sock_error(sk);
1968	if (error)
1969	break;
1970
1971	error = -EAGAIN;
1972	do {
1973	spin_lock_bh(lock: &queue->lock);
1974	skb = __skb_try_recv_from_queue(queue, flags, off, err,
1975	last: &last);
1976	if (skb) {
1977	if (!(flags & MSG_PEEK))
1978	udp_skb_destructor(sk, skb);
1979	spin_unlock_bh(lock: &queue->lock);
1980	return skb;
1981	}
1982
1983	if (skb_queue_empty_lockless(list: sk_queue)) {
1984	spin_unlock_bh(lock: &queue->lock);
1985	goto busy_check;
1986	}
1987
1988	/* refill the reader queue and walk it again
1989	* keep both queues locked to avoid re-acquiring
1990	* the sk_receive_queue lock if fwd memory scheduling
1991	* is needed.
1992	*/
1993	spin_lock(lock: &sk_queue->lock);
1994	skb_queue_splice_tail_init(list: sk_queue, head: queue);
1995
1996	skb = __skb_try_recv_from_queue(queue, flags, off, err,
1997	last: &last);
1998	if (skb && !(flags & MSG_PEEK))
1999	udp_skb_dtor_locked(sk, skb);
2000	spin_unlock(lock: &sk_queue->lock);
2001	spin_unlock_bh(lock: &queue->lock);
2002	if (skb)
2003	return skb;
2004
2005	busy_check:
2006	if (!sk_can_busy_loop(sk))
2007	break;
2008
2009	sk_busy_loop(sk, nonblock: flags & MSG_DONTWAIT);
2010	} while (!skb_queue_empty_lockless(list: sk_queue));
2011
2012	/* sk_queue is empty, reader_queue may contain peeked packets */
2013	} while (timeo &&
2014	!__skb_wait_for_more_packets(sk, queue: &sk->sk_receive_queue,
2015	err: &error, timeo_p: &timeo,
2016	skb: (struct sk_buff *)sk_queue));
2017
2018	*err = error;
2019	return NULL;
2020	}
2021	EXPORT_SYMBOL(__skb_recv_udp);
2022
2023	int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
2024	{
2025	struct sk_buff *skb;
2026	int err;
2027
2028	try_again:
2029	skb = skb_recv_udp(sk, MSG_DONTWAIT, err: &err);
2030	if (!skb)
2031	return err;
2032
2033	if (udp_lib_checksum_complete(skb)) {
2034	int is_udplite = IS_UDPLITE(sk);
2035	struct net *net = sock_net(sk);
2036
2037	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
2038	__UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
2039	udp_drops_inc(sk);
2040	kfree_skb_reason(skb, reason: SKB_DROP_REASON_UDP_CSUM);
2041	goto try_again;
2042	}
2043
2044	WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
2045	return recv_actor(sk, skb);
2046	}
2047	EXPORT_IPV6_MOD(udp_read_skb);
2048
2049	/*
2050	* This should be easy, if there is something there we
2051	* return it, otherwise we block.
2052	*/
2053
2054	int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
2055	int *addr_len)
2056	{
2057	struct inet_sock *inet = inet_sk(sk);
2058	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
2059	struct sk_buff *skb;
2060	unsigned int ulen, copied;
2061	int off, err, peeking = flags & MSG_PEEK;
2062	int is_udplite = IS_UDPLITE(sk);
2063	bool checksum_valid = false;
2064
2065	if (flags & MSG_ERRQUEUE)
2066	return ip_recv_error(sk, msg, len, addr_len);
2067
2068	try_again:
2069	off = sk_peek_offset(sk, flags);
2070	skb = __skb_recv_udp(sk, flags, &off, &err);
2071	if (!skb)
2072	return err;
2073
2074	ulen = udp_skb_len(skb);
2075	copied = len;
2076	if (copied > ulen - off)
2077	copied = ulen - off;
2078	else if (copied < ulen)
2079	msg->msg_flags |= MSG_TRUNC;
2080
2081	/*
2082	* If checksum is needed at all, try to do it while copying the
2083	* data. If the data is truncated, or if we only want a partial
2084	* coverage checksum (UDP-Lite), do it before the copy.
2085	*/
2086
2087	if (copied < ulen || peeking ||
2088	(is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
2089	checksum_valid = udp_skb_csum_unnecessary(skb) ||
2090	!__udp_lib_checksum_complete(skb);
2091	if (!checksum_valid)
2092	goto csum_copy_err;
2093	}
2094
2095	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
2096	if (udp_skb_is_linear(skb))
2097	err = copy_linear_skb(skb, len: copied, off, to: &msg->msg_iter);
2098	else
2099	err = skb_copy_datagram_msg(from: skb, offset: off, msg, size: copied);
2100	} else {
2101	err = skb_copy_and_csum_datagram_msg(skb, hlen: off, msg);
2102
2103	if (err == -EINVAL)
2104	goto csum_copy_err;
2105	}
2106
2107	if (unlikely(err)) {
2108	if (!peeking) {
2109	udp_drops_inc(sk);
2110	UDP_INC_STATS(sock_net(sk),
2111	UDP_MIB_INERRORS, is_udplite);
2112	}
2113	kfree_skb(skb);
2114	return err;
2115	}
2116
2117	if (!peeking)
2118	UDP_INC_STATS(sock_net(sk),
2119	UDP_MIB_INDATAGRAMS, is_udplite);
2120
2121	sock_recv_cmsgs(msg, sk, skb);
2122
2123	/* Copy the address. */
2124	if (sin) {
2125	sin->sin_family = AF_INET;
2126	sin->sin_port = udp_hdr(skb)->source;
2127	sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
2128	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
2129	*addr_len = sizeof(*sin);
2130
2131	BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
2132	(struct sockaddr *)sin,
2133	addr_len);
2134	}
2135
2136	if (udp_test_bit(GRO_ENABLED, sk))
2137	udp_cmsg_recv(msg, sk, skb);
2138
2139	if (inet_cmsg_flags(inet))
2140	ip_cmsg_recv_offset(msg, sk, skb, tlen: sizeof(struct udphdr), offset: off);
2141
2142	err = copied;
2143	if (flags & MSG_TRUNC)
2144	err = ulen;
2145
2146	skb_consume_udp(sk, skb, len: peeking ? -err : err);
2147	return err;
2148
2149	csum_copy_err:
2150	if (!__sk_queue_drop_skb(sk, sk_queue: &udp_sk(sk)->reader_queue, skb, flags,
2151	destructor: udp_skb_destructor)) {
2152	UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2153	UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2154	}
2155	kfree_skb_reason(skb, reason: SKB_DROP_REASON_UDP_CSUM);
2156
2157	/* starting over for a new packet, but check if we need to yield */
2158	cond_resched();
2159	msg->msg_flags &= ~MSG_TRUNC;
2160	goto try_again;
2161	}
2162
2163	int udp_pre_connect(struct sock *sk, struct sockaddr_unsized *uaddr,
2164	int addr_len)
2165	{
2166	/* This check is replicated from __ip4_datagram_connect() and
2167	* intended to prevent BPF program called below from accessing bytes
2168	* that are out of the bound specified by user in addr_len.
2169	*/
2170	if (addr_len < sizeof(struct sockaddr_in))
2171	return -EINVAL;
2172
2173	return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
2174	}
2175	EXPORT_IPV6_MOD(udp_pre_connect);
2176
2177	static int udp_connect(struct sock *sk, struct sockaddr_unsized *uaddr,
2178	int addr_len)
2179	{
2180	int res;
2181
2182	lock_sock(sk);
2183	res = __ip4_datagram_connect(sk, uaddr, addr_len);
2184	if (!res)
2185	udp4_hash4(sk);
2186	release_sock(sk);
2187	return res;
2188	}
2189
2190	int __udp_disconnect(struct sock *sk, int flags)
2191	{
2192	struct inet_sock *inet = inet_sk(sk);
2193	/*
2194	* 1003.1g - break association.
2195	*/
2196
2197	sk->sk_state = TCP_CLOSE;
2198	inet->inet_daddr = 0;
2199	inet->inet_dport = 0;
2200	sock_rps_reset_rxhash(sk);
2201	sk->sk_bound_dev_if = 0;
2202	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
2203	inet_reset_saddr(sk);
2204	if (sk->sk_prot->rehash &&
2205	(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2206	sk->sk_prot->rehash(sk);
2207	}
2208
2209	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
2210	sk->sk_prot->unhash(sk);
2211	inet->inet_sport = 0;
2212	}
2213	sk_dst_reset(sk);
2214	return 0;
2215	}
2216	EXPORT_SYMBOL(__udp_disconnect);
2217
2218	int udp_disconnect(struct sock *sk, int flags)
2219	{
2220	lock_sock(sk);
2221	__udp_disconnect(sk, flags);
2222	release_sock(sk);
2223	return 0;
2224	}
2225	EXPORT_IPV6_MOD(udp_disconnect);
2226
2227	void udp_lib_unhash(struct sock *sk)
2228	{
2229	if (sk_hashed(sk)) {
2230	struct udp_table *udptable = udp_get_table_prot(sk);
2231	struct udp_hslot *hslot, *hslot2;
2232
2233	sock_rps_delete_flow(sk);
2234	hslot = udp_hashslot(table: udptable, net: sock_net(sk),
2235	udp_sk(sk)->udp_port_hash);
2236	hslot2 = udp_hashslot2(table: udptable, udp_sk(sk)->udp_portaddr_hash);
2237
2238	spin_lock_bh(lock: &hslot->lock);
2239	if (rcu_access_pointer(sk->sk_reuseport_cb))
2240	reuseport_detach_sock(sk);
2241	if (sk_del_node_init_rcu(sk)) {
2242	hslot->count--;
2243	inet_sk(sk)->inet_num = 0;
2244	sock_prot_inuse_add(net: sock_net(sk), prot: sk->sk_prot, val: -1);
2245
2246	spin_lock(lock: &hslot2->lock);
2247	hlist_del_init_rcu(n: &udp_sk(sk)->udp_portaddr_node);
2248	hslot2->count--;
2249	spin_unlock(lock: &hslot2->lock);
2250
2251	udp_unhash4(udptable, sk);
2252	}
2253	spin_unlock_bh(lock: &hslot->lock);
2254	}
2255	}
2256	EXPORT_IPV6_MOD(udp_lib_unhash);
2257
2258	/*
2259	* inet_rcv_saddr was changed, we must rehash secondary hash
2260	*/
2261	void udp_lib_rehash(struct sock *sk, u16 newhash, u16 newhash4)
2262	{
2263	if (sk_hashed(sk)) {
2264	struct udp_table *udptable = udp_get_table_prot(sk);
2265	struct udp_hslot *hslot, *hslot2, *nhslot2;
2266
2267	hslot = udp_hashslot(table: udptable, net: sock_net(sk),
2268	udp_sk(sk)->udp_port_hash);
2269	hslot2 = udp_hashslot2(table: udptable, udp_sk(sk)->udp_portaddr_hash);
2270	nhslot2 = udp_hashslot2(table: udptable, hash: newhash);
2271	udp_sk(sk)->udp_portaddr_hash = newhash;
2272
2273	if (hslot2 != nhslot2 ||
2274	rcu_access_pointer(sk->sk_reuseport_cb)) {
2275	/* we must lock primary chain too */
2276	spin_lock_bh(lock: &hslot->lock);
2277	if (rcu_access_pointer(sk->sk_reuseport_cb))
2278	reuseport_detach_sock(sk);
2279
2280	if (hslot2 != nhslot2) {
2281	spin_lock(lock: &hslot2->lock);
2282	hlist_del_init_rcu(n: &udp_sk(sk)->udp_portaddr_node);
2283	hslot2->count--;
2284	spin_unlock(lock: &hslot2->lock);
2285
2286	spin_lock(lock: &nhslot2->lock);
2287	hlist_add_head_rcu(n: &udp_sk(sk)->udp_portaddr_node,
2288	h: &nhslot2->head);
2289	nhslot2->count++;
2290	spin_unlock(lock: &nhslot2->lock);
2291	}
2292
2293	spin_unlock_bh(lock: &hslot->lock);
2294	}
2295
2296	/* Now process hash4 if necessary:
2297	* (1) update hslot4;
2298	* (2) update hslot2->hash4_cnt.
2299	* Note that hslot2/hslot4 should be checked separately, as
2300	* either of them may change with the other unchanged.
2301	*/
2302	if (udp_hashed4(sk)) {
2303	spin_lock_bh(lock: &hslot->lock);
2304
2305	udp_rehash4(udptable, sk, newhash4);
2306	if (hslot2 != nhslot2) {
2307	spin_lock(lock: &hslot2->lock);
2308	udp_hash4_dec(hslot2);
2309	spin_unlock(lock: &hslot2->lock);
2310
2311	spin_lock(lock: &nhslot2->lock);
2312	udp_hash4_inc(hslot2: nhslot2);
2313	spin_unlock(lock: &nhslot2->lock);
2314	}
2315
2316	spin_unlock_bh(lock: &hslot->lock);
2317	}
2318	}
2319	}
2320	EXPORT_IPV6_MOD(udp_lib_rehash);
2321
2322	void udp_v4_rehash(struct sock *sk)
2323	{
2324	u16 new_hash = ipv4_portaddr_hash(net: sock_net(sk),
2325	inet_sk(sk)->inet_rcv_saddr,
2326	inet_sk(sk)->inet_num);
2327	u16 new_hash4 = udp_ehashfn(net: sock_net(sk),
2328	laddr: sk->sk_rcv_saddr, lport: sk->sk_num,
2329	faddr: sk->sk_daddr, fport: sk->sk_dport);
2330
2331	udp_lib_rehash(sk, newhash: new_hash, newhash4: new_hash4);
2332	}
2333
2334	static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2335	{
2336	int rc;
2337
2338	if (inet_sk(sk)->inet_daddr) {
2339	sock_rps_save_rxhash(sk, skb);
2340	sk_mark_napi_id(sk, skb);
2341	sk_incoming_cpu_update(sk);
2342	} else {
2343	sk_mark_napi_id_once(sk, skb);
2344	}
2345
2346	rc = __udp_enqueue_schedule_skb(sk, skb);
2347	if (rc < 0) {
2348	int is_udplite = IS_UDPLITE(sk);
2349	int drop_reason;
2350
2351	/* Note that an ENOMEM error is charged twice */
2352	if (rc == -ENOMEM) {
2353	UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2354	is_udplite);
2355	drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2356	} else {
2357	UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2358	is_udplite);
2359	drop_reason = SKB_DROP_REASON_PROTO_MEM;
2360	}
2361	UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2362	trace_udp_fail_queue_rcv_skb(rc, sk, skb);
2363	sk_skb_reason_drop(sk, skb, reason: drop_reason);
2364	return -1;
2365	}
2366
2367	return 0;
2368	}
2369
2370	/* returns:
2371	* -1: error
2372	* 0: success
2373	* >0: "udp encap" protocol resubmission
2374	*
2375	* Note that in the success and error cases, the skb is assumed to
2376	* have either been requeued or freed.
2377	*/
2378	static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2379	{
2380	enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2381	struct udp_sock *up = udp_sk(sk);
2382	int is_udplite = IS_UDPLITE(sk);
2383
2384	/*
2385	* Charge it to the socket, dropping if the queue is full.
2386	*/
2387	if (!xfrm4_policy_check(sk, dir: XFRM_POLICY_IN, skb)) {
2388	drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2389	goto drop;
2390	}
2391	nf_reset_ct(skb);
2392
2393	if (static_branch_unlikely(&udp_encap_needed_key) &&
2394	READ_ONCE(up->encap_type)) {
2395	int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2396
2397	/*
2398	* This is an encapsulation socket so pass the skb to
2399	* the socket's udp_encap_rcv() hook. Otherwise, just
2400	* fall through and pass this up the UDP socket.
2401	* up->encap_rcv() returns the following value:
2402	* =0 if skb was successfully passed to the encap
2403	* handler or was discarded by it.
2404	* >0 if skb should be passed on to UDP.
2405	* <0 if skb should be resubmitted as proto -N
2406	*/
2407
2408	/* if we're overly short, let UDP handle it */
2409	encap_rcv = READ_ONCE(up->encap_rcv);
2410	if (encap_rcv) {
2411	int ret;
2412
2413	/* Verify checksum before giving to encap */
2414	if (udp_lib_checksum_complete(skb))
2415	goto csum_error;
2416
2417	ret = encap_rcv(sk, skb);
2418	if (ret <= 0) {
2419	__UDP_INC_STATS(sock_net(sk),
2420	UDP_MIB_INDATAGRAMS,
2421	is_udplite);
2422	return -ret;
2423	}
2424	}
2425
2426	/* FALLTHROUGH -- it's a UDP Packet */
2427	}
2428
2429	/*
2430	* UDP-Lite specific tests, ignored on UDP sockets
2431	*/
2432	if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2433	u16 pcrlen = READ_ONCE(up->pcrlen);
2434
2435	/*
2436	* MIB statistics other than incrementing the error count are
2437	* disabled for the following two types of errors: these depend
2438	* on the application settings, not on the functioning of the
2439	* protocol stack as such.
2440	*
2441	* RFC 3828 here recommends (sec 3.3): "There should also be a
2442	* way ... to ... at least let the receiving application block
2443	* delivery of packets with coverage values less than a value
2444	* provided by the application."
2445	*/
2446	if (pcrlen == 0) { /* full coverage was set */
2447	net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2448	UDP_SKB_CB(skb)->cscov, skb->len);
2449	goto drop;
2450	}
2451	/* The next case involves violating the min. coverage requested
2452	* by the receiver. This is subtle: if receiver wants x and x is
2453	* greater than the buffersize/MTU then receiver will complain
2454	* that it wants x while sender emits packets of smaller size y.
2455	* Therefore the above ...()->partial_cov statement is essential.
2456	*/
2457	if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2458	net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2459	UDP_SKB_CB(skb)->cscov, pcrlen);
2460	goto drop;
2461	}
2462	}
2463
2464	prefetch(&sk->sk_rmem_alloc);
2465	if (rcu_access_pointer(sk->sk_filter) &&
2466	udp_lib_checksum_complete(skb))
2467	goto csum_error;
2468
2469	if (sk_filter_trim_cap(sk, skb, cap: sizeof(struct udphdr), reason: &drop_reason))
2470	goto drop;
2471
2472	udp_csum_pull_header(skb);
2473
2474	ipv4_pktinfo_prepare(sk, skb, drop_dst: true);
2475	return __udp_queue_rcv_skb(sk, skb);
2476
2477	csum_error:
2478	drop_reason = SKB_DROP_REASON_UDP_CSUM;
2479	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2480	drop:
2481	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2482	udp_drops_inc(sk);
2483	sk_skb_reason_drop(sk, skb, reason: drop_reason);
2484	return -1;
2485	}
2486
2487	static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2488	{
2489	struct sk_buff *next, *segs;
2490	int ret;
2491
2492	if (likely(!udp_unexpected_gso(sk, skb)))
2493	return udp_queue_rcv_one_skb(sk, skb);
2494
2495	BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2496	__skb_push(skb, len: -skb_mac_offset(skb));
2497	segs = udp_rcv_segment(sk, skb, ipv4: true);
2498	skb_list_walk_safe(segs, skb, next) {
2499	__skb_pull(skb, len: skb_transport_offset(skb));
2500
2501	udp_post_segment_fix_csum(skb);
2502	ret = udp_queue_rcv_one_skb(sk, skb);
2503	if (ret > 0)
2504	ip_protocol_deliver_rcu(net: dev_net(dev: skb->dev), skb, proto: ret);
2505	}
2506	return 0;
2507	}
2508
2509	/* For TCP sockets, sk_rx_dst is protected by socket lock
2510	* For UDP, we use xchg() to guard against concurrent changes.
2511	*/
2512	bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2513	{
2514	struct dst_entry *old;
2515
2516	if (dst_hold_safe(dst)) {
2517	old = unrcu_pointer(xchg(&sk->sk_rx_dst, RCU_INITIALIZER(dst)));
2518	dst_release(dst: old);
2519	return old != dst;
2520	}
2521	return false;
2522	}
2523	EXPORT_IPV6_MOD(udp_sk_rx_dst_set);
2524
2525	/*
2526	* Multicasts and broadcasts go to each listener.
2527	*
2528	* Note: called only from the BH handler context.
2529	*/
2530	static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2531	struct udphdr *uh,
2532	__be32 saddr, __be32 daddr,
2533	struct udp_table *udptable,
2534	int proto)
2535	{
2536	struct sock *sk, *first = NULL;
2537	unsigned short hnum = ntohs(uh->dest);
2538	struct udp_hslot *hslot = udp_hashslot(table: udptable, net, num: hnum);
2539	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2540	unsigned int offset = offsetof(typeof(*sk), sk_node);
2541	int dif = skb->dev->ifindex;
2542	int sdif = inet_sdif(skb);
2543	struct hlist_node *node;
2544	struct sk_buff *nskb;
2545
2546	if (use_hash2) {
2547	hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), port: hnum) &
2548	udptable->mask;
2549	hash2 = ipv4_portaddr_hash(net, saddr: daddr, port: hnum) & udptable->mask;
2550	start_lookup:
2551	hslot = &udptable->hash2[hash2].hslot;
2552	offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2553	}
2554
2555	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2556	if (!__udp_is_mcast_sock(net, sk, loc_port: uh->dest, loc_addr: daddr,
2557	rmt_port: uh->source, rmt_addr: saddr, dif, sdif, hnum))
2558	continue;
2559
2560	if (!first) {
2561	first = sk;
2562	continue;
2563	}
2564	nskb = skb_clone(skb, GFP_ATOMIC);
2565
2566	if (unlikely(!nskb)) {
2567	udp_drops_inc(sk);
2568	__UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2569	IS_UDPLITE(sk));
2570	__UDP_INC_STATS(net, UDP_MIB_INERRORS,
2571	IS_UDPLITE(sk));
2572	continue;
2573	}
2574	if (udp_queue_rcv_skb(sk, skb: nskb) > 0)
2575	consume_skb(skb: nskb);
2576	}
2577
2578	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
2579	if (use_hash2 && hash2 != hash2_any) {
2580	hash2 = hash2_any;
2581	goto start_lookup;
2582	}
2583
2584	if (first) {
2585	if (udp_queue_rcv_skb(sk: first, skb) > 0)
2586	consume_skb(skb);
2587	} else {
2588	kfree_skb(skb);
2589	__UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2590	proto == IPPROTO_UDPLITE);
2591	}
2592	return 0;
2593	}
2594
2595	/* Initialize UDP checksum. If exited with zero value (success),
2596	* CHECKSUM_UNNECESSARY means, that no more checks are required.
2597	* Otherwise, csum completion requires checksumming packet body,
2598	* including udp header and folding it to skb->csum.
2599	*/
2600	static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2601	int proto)
2602	{
2603	int err;
2604
2605	UDP_SKB_CB(skb)->partial_cov = 0;
2606	UDP_SKB_CB(skb)->cscov = skb->len;
2607
2608	if (proto == IPPROTO_UDPLITE) {
2609	err = udplite_checksum_init(skb, uh);
2610	if (err)
2611	return err;
2612
2613	if (UDP_SKB_CB(skb)->partial_cov) {
2614	skb->csum = inet_compute_pseudo(skb, proto);
2615	return 0;
2616	}
2617	}
2618
2619	/* Note, we are only interested in != 0 or == 0, thus the
2620	* force to int.
2621	*/
2622	err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2623	inet_compute_pseudo);
2624	if (err)
2625	return err;
2626
2627	if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2628	/* If SW calculated the value, we know it's bad */
2629	if (skb->csum_complete_sw)
2630	return 1;
2631
2632	/* HW says the value is bad. Let's validate that.
2633	* skb->csum is no longer the full packet checksum,
2634	* so don't treat it as such.
2635	*/
2636	skb_checksum_complete_unset(skb);
2637	}
2638
2639	return 0;
2640	}
2641
2642	/* wrapper for udp_queue_rcv_skb taking care of csum conversion and
2643	* return code conversion for ip layer consumption
2644	*/
2645	static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2646	struct udphdr *uh)
2647	{
2648	int ret;
2649
2650	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2651	skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2652
2653	ret = udp_queue_rcv_skb(sk, skb);
2654
2655	/* a return value > 0 means to resubmit the input, but
2656	* it wants the return to be -protocol, or 0
2657	*/
2658	if (ret > 0)
2659	return -ret;
2660	return 0;
2661	}
2662
2663	/*
2664	* All we need to do is get the socket, and then do a checksum.
2665	*/
2666
2667	int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2668	int proto)
2669	{
2670	struct sock *sk = NULL;
2671	struct udphdr *uh;
2672	unsigned short ulen;
2673	struct rtable *rt = skb_rtable(skb);
2674	__be32 saddr, daddr;
2675	struct net *net = dev_net(dev: skb->dev);
2676	bool refcounted;
2677	int drop_reason;
2678
2679	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2680
2681	/*
2682	* Validate the packet.
2683	*/
2684	if (!pskb_may_pull(skb, len: sizeof(struct udphdr)))
2685	goto drop; /* No space for header. */
2686
2687	uh = udp_hdr(skb);
2688	ulen = ntohs(uh->len);
2689	saddr = ip_hdr(skb)->saddr;
2690	daddr = ip_hdr(skb)->daddr;
2691
2692	if (ulen > skb->len)
2693	goto short_packet;
2694
2695	if (proto == IPPROTO_UDP) {
2696	/* UDP validates ulen. */
2697	if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, len: ulen))
2698	goto short_packet;
2699	uh = udp_hdr(skb);
2700	}
2701
2702	if (udp4_csum_init(skb, uh, proto))
2703	goto csum_error;
2704
2705	sk = inet_steal_sock(net, skb, doff: sizeof(struct udphdr), saddr, sport: uh->source, daddr, dport: uh->dest,
2706	refcounted: &refcounted, ehashfn: udp_ehashfn);
2707	if (IS_ERR(ptr: sk))
2708	goto no_sk;
2709
2710	if (sk) {
2711	struct dst_entry *dst = skb_dst(skb);
2712	int ret;
2713
2714	if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2715	udp_sk_rx_dst_set(sk, dst);
2716
2717	ret = udp_unicast_rcv_skb(sk, skb, uh);
2718	if (refcounted)
2719	sock_put(sk);
2720	return ret;
2721	}
2722
2723	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2724	return __udp4_lib_mcast_deliver(net, skb, uh,
2725	saddr, daddr, udptable, proto);
2726
2727	sk = __udp4_lib_lookup_skb(skb, sport: uh->source, dport: uh->dest, udptable);
2728	if (sk)
2729	return udp_unicast_rcv_skb(sk, skb, uh);
2730	no_sk:
2731	if (!xfrm4_policy_check(NULL, dir: XFRM_POLICY_IN, skb))
2732	goto drop;
2733	nf_reset_ct(skb);
2734
2735	/* No socket. Drop packet silently, if checksum is wrong */
2736	if (udp_lib_checksum_complete(skb))
2737	goto csum_error;
2738
2739	drop_reason = SKB_DROP_REASON_NO_SOCKET;
2740	__UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2741	icmp_send(skb_in: skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, info: 0);
2742
2743	/*
2744	* Hmm. We got an UDP packet to a port to which we
2745	* don't wanna listen. Ignore it.
2746	*/
2747	sk_skb_reason_drop(sk, skb, reason: drop_reason);
2748	return 0;
2749
2750	short_packet:
2751	drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2752	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2753	proto == IPPROTO_UDPLITE ? "Lite" : "",
2754	&saddr, ntohs(uh->source),
2755	ulen, skb->len,
2756	&daddr, ntohs(uh->dest));
2757	goto drop;
2758
2759	csum_error:
2760	/*
2761	* RFC1122: OK. Discards the bad packet silently (as far as
2762	* the network is concerned, anyway) as per 4.1.3.4 (MUST).
2763	*/
2764	drop_reason = SKB_DROP_REASON_UDP_CSUM;
2765	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2766	proto == IPPROTO_UDPLITE ? "Lite" : "",
2767	&saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2768	ulen);
2769	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2770	drop:
2771	__UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2772	sk_skb_reason_drop(sk, skb, reason: drop_reason);
2773	return 0;
2774	}
2775
2776	/* We can only early demux multicast if there is a single matching socket.
2777	* If more than one socket found returns NULL
2778	*/
2779	static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2780	__be16 loc_port, __be32 loc_addr,
2781	__be16 rmt_port, __be32 rmt_addr,
2782	int dif, int sdif)
2783	{
2784	struct udp_table *udptable = net->ipv4.udp_table;
2785	unsigned short hnum = ntohs(loc_port);
2786	struct sock *sk, *result;
2787	struct udp_hslot *hslot;
2788	unsigned int slot;
2789
2790	slot = udp_hashfn(net, num: hnum, mask: udptable->mask);
2791	hslot = &udptable->hash[slot];
2792
2793	/* Do not bother scanning a too big list */
2794	if (hslot->count > 10)
2795	return NULL;
2796
2797	result = NULL;
2798	sk_for_each_rcu(sk, &hslot->head) {
2799	if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2800	rmt_port, rmt_addr, dif, sdif, hnum)) {
2801	if (result)
2802	return NULL;
2803	result = sk;
2804	}
2805	}
2806
2807	return result;
2808	}
2809
2810	/* For unicast we should only early demux connected sockets or we can
2811	* break forwarding setups. The chains here can be long so only check
2812	* if the first socket is an exact match and if not move on.
2813	*/
2814	static struct sock *__udp4_lib_demux_lookup(struct net *net,
2815	__be16 loc_port, __be32 loc_addr,
2816	__be16 rmt_port, __be32 rmt_addr,
2817	int dif, int sdif)
2818	{
2819	struct udp_table *udptable = net->ipv4.udp_table;
2820	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2821	unsigned short hnum = ntohs(loc_port);
2822	struct udp_hslot *hslot2;
2823	unsigned int hash2;
2824	__portpair ports;
2825	struct sock *sk;
2826
2827	hash2 = ipv4_portaddr_hash(net, saddr: loc_addr, port: hnum);
2828	hslot2 = udp_hashslot2(table: udptable, hash: hash2);
2829	ports = INET_COMBINED_PORTS(rmt_port, hnum);
2830
2831	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2832	if (inet_match(net, sk, cookie: acookie, ports, dif, sdif))
2833	return sk;
2834	/* Only check first socket in chain */
2835	break;
2836	}
2837	return NULL;
2838	}
2839
2840	enum skb_drop_reason udp_v4_early_demux(struct sk_buff *skb)
2841	{
2842	struct net *net = dev_net(dev: skb->dev);
2843	struct in_device *in_dev = NULL;
2844	const struct iphdr *iph;
2845	const struct udphdr *uh;
2846	struct sock *sk = NULL;
2847	struct dst_entry *dst;
2848	int dif = skb->dev->ifindex;
2849	int sdif = inet_sdif(skb);
2850	int ours;
2851
2852	/* validate the packet */
2853	if (!pskb_may_pull(skb, len: skb_transport_offset(skb) + sizeof(struct udphdr)))
2854	return SKB_NOT_DROPPED_YET;
2855
2856	iph = ip_hdr(skb);
2857	uh = udp_hdr(skb);
2858
2859	if (skb->pkt_type == PACKET_MULTICAST) {
2860	in_dev = __in_dev_get_rcu(dev: skb->dev);
2861
2862	if (!in_dev)
2863	return SKB_NOT_DROPPED_YET;
2864
2865	ours = ip_check_mc_rcu(dev: in_dev, mc_addr: iph->daddr, src_addr: iph->saddr,
2866	proto: iph->protocol);
2867	if (!ours)
2868	return SKB_NOT_DROPPED_YET;
2869
2870	sk = __udp4_lib_mcast_demux_lookup(net, loc_port: uh->dest, loc_addr: iph->daddr,
2871	rmt_port: uh->source, rmt_addr: iph->saddr,
2872	dif, sdif);
2873	} else if (skb->pkt_type == PACKET_HOST) {
2874	sk = __udp4_lib_demux_lookup(net, loc_port: uh->dest, loc_addr: iph->daddr,
2875	rmt_port: uh->source, rmt_addr: iph->saddr, dif, sdif);
2876	}
2877
2878	if (!sk)
2879	return SKB_NOT_DROPPED_YET;
2880
2881	skb->sk = sk;
2882	DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
2883	skb->destructor = sock_pfree;
2884	dst = rcu_dereference(sk->sk_rx_dst);
2885
2886	if (dst)
2887	dst = dst_check(dst, cookie: 0);
2888	if (dst) {
2889	u32 itag = 0;
2890
2891	/* set noref for now.
2892	* any place which wants to hold dst has to call
2893	* dst_hold_safe()
2894	*/
2895	skb_dst_set_noref(skb, dst);
2896
2897	/* for unconnected multicast sockets we need to validate
2898	* the source on each packet
2899	*/
2900	if (!inet_sk(sk)->inet_daddr && in_dev)
2901	return ip_mc_validate_source(skb, daddr: iph->daddr,
2902	saddr: iph->saddr,
2903	dscp: ip4h_dscp(ip4h: iph),
2904	dev: skb->dev, in_dev, itag: &itag);
2905	}
2906	return SKB_NOT_DROPPED_YET;
2907	}
2908
2909	int udp_rcv(struct sk_buff *skb)
2910	{
2911	return __udp4_lib_rcv(skb, udptable: dev_net(dev: skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2912	}
2913
2914	void udp_destroy_sock(struct sock *sk)
2915	{
2916	struct udp_sock *up = udp_sk(sk);
2917	bool slow = lock_sock_fast(sk);
2918
2919	/* protects from races with udp_abort() */
2920	sock_set_flag(sk, flag: SOCK_DEAD);
2921	udp_flush_pending_frames(sk);
2922	unlock_sock_fast(sk, slow);
2923	if (static_branch_unlikely(&udp_encap_needed_key)) {
2924	if (up->encap_type) {
2925	void (*encap_destroy)(struct sock *sk);
2926	encap_destroy = READ_ONCE(up->encap_destroy);
2927	if (encap_destroy)
2928	encap_destroy(sk);
2929	}
2930	if (udp_test_bit(ENCAP_ENABLED, sk)) {
2931	static_branch_dec(&udp_encap_needed_key);
2932	udp_tunnel_cleanup_gro(sk);
2933	}
2934	}
2935	}
2936
2937	typedef struct sk_buff *(*udp_gro_receive_t)(struct sock *sk,
2938	struct list_head *head,
2939	struct sk_buff *skb);
2940
2941	static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2942	struct sock *sk)
2943	{
2944	#ifdef CONFIG_XFRM
2945	udp_gro_receive_t new_gro_receive;
2946
2947	if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2948	if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2949	new_gro_receive = ipv6_stub->xfrm6_gro_udp_encap_rcv;
2950	else
2951	new_gro_receive = xfrm4_gro_udp_encap_rcv;
2952
2953	if (udp_sk(sk)->gro_receive != new_gro_receive) {
2954	/*
2955	* With IPV6_ADDRFORM the gro callback could change
2956	* after being set, unregister the old one, if valid.
2957	*/
2958	if (udp_sk(sk)->gro_receive)
2959	udp_tunnel_update_gro_rcv(sk, add: false);
2960
2961	WRITE_ONCE(udp_sk(sk)->gro_receive, new_gro_receive);
2962	udp_tunnel_update_gro_rcv(sk, add: true);
2963	}
2964	}
2965	#endif
2966	}
2967
2968	/*
2969	* Socket option code for UDP
2970	*/
2971	int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2972	sockptr_t optval, unsigned int optlen,
2973	int (*push_pending_frames)(struct sock *))
2974	{
2975	struct udp_sock *up = udp_sk(sk);
2976	int val, valbool;
2977	int err = 0;
2978	int is_udplite = IS_UDPLITE(sk);
2979
2980	if (level == SOL_SOCKET) {
2981	err = sk_setsockopt(sk, level, optname, optval, optlen);
2982
2983	if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2984	sockopt_lock_sock(sk);
2985	/* paired with READ_ONCE in udp_rmem_release() */
2986	WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2987	sockopt_release_sock(sk);
2988	}
2989	return err;
2990	}
2991
2992	if (optlen < sizeof(int))
2993	return -EINVAL;
2994
2995	if (copy_from_sockptr(dst: &val, src: optval, size: sizeof(val)))
2996	return -EFAULT;
2997
2998	valbool = val ? 1 : 0;
2999
3000	switch (optname) {
3001	case UDP_CORK:
3002	if (val != 0) {
3003	udp_set_bit(CORK, sk);
3004	} else {
3005	udp_clear_bit(CORK, sk);
3006	lock_sock(sk);
3007	push_pending_frames(sk);
3008	release_sock(sk);
3009	}
3010	break;
3011
3012	case UDP_ENCAP:
3013	sockopt_lock_sock(sk);
3014	switch (val) {
3015	case 0:
3016	#ifdef CONFIG_XFRM
3017	case UDP_ENCAP_ESPINUDP:
3018	set_xfrm_gro_udp_encap_rcv(encap_type: val, family: sk->sk_family, sk);
3019	#if IS_ENABLED(CONFIG_IPV6)
3020	if (sk->sk_family == AF_INET6)
3021	WRITE_ONCE(up->encap_rcv,
3022	ipv6_stub->xfrm6_udp_encap_rcv);
3023	else
3024	#endif
3025	WRITE_ONCE(up->encap_rcv,
3026	xfrm4_udp_encap_rcv);
3027	#endif
3028	fallthrough;
3029	case UDP_ENCAP_L2TPINUDP:
3030	WRITE_ONCE(up->encap_type, val);
3031	udp_tunnel_encap_enable(sk);
3032	break;
3033	default:
3034	err = -ENOPROTOOPT;
3035	break;
3036	}
3037	sockopt_release_sock(sk);
3038	break;
3039
3040	case UDP_NO_CHECK6_TX:
3041	udp_set_no_check6_tx(sk, val: valbool);
3042	break;
3043
3044	case UDP_NO_CHECK6_RX:
3045	udp_set_no_check6_rx(sk, val: valbool);
3046	break;
3047
3048	case UDP_SEGMENT:
3049	if (val < 0 || val > USHRT_MAX)
3050	return -EINVAL;
3051	WRITE_ONCE(up->gso_size, val);
3052	break;
3053
3054	case UDP_GRO:
3055	sockopt_lock_sock(sk);
3056	/* when enabling GRO, accept the related GSO packet type */
3057	if (valbool)
3058	udp_tunnel_encap_enable(sk);
3059	udp_assign_bit(GRO_ENABLED, sk, valbool);
3060	udp_assign_bit(ACCEPT_L4, sk, valbool);
3061	set_xfrm_gro_udp_encap_rcv(encap_type: up->encap_type, family: sk->sk_family, sk);
3062	sockopt_release_sock(sk);
3063	break;
3064
3065	/*
3066	* UDP-Lite's partial checksum coverage (RFC 3828).
3067	*/
3068	/* The sender sets actual checksum coverage length via this option.
3069	* The case coverage > packet length is handled by send module. */
3070	case UDPLITE_SEND_CSCOV:
3071	if (!is_udplite) /* Disable the option on UDP sockets */
3072	return -ENOPROTOOPT;
3073	if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
3074	val = 8;
3075	else if (val > USHRT_MAX)
3076	val = USHRT_MAX;
3077	WRITE_ONCE(up->pcslen, val);
3078	udp_set_bit(UDPLITE_SEND_CC, sk);
3079	break;
3080
3081	/* The receiver specifies a minimum checksum coverage value. To make
3082	* sense, this should be set to at least 8 (as done below). If zero is
3083	* used, this again means full checksum coverage. */
3084	case UDPLITE_RECV_CSCOV:
3085	if (!is_udplite) /* Disable the option on UDP sockets */
3086	return -ENOPROTOOPT;
3087	if (val != 0 && val < 8) /* Avoid silly minimal values. */
3088	val = 8;
3089	else if (val > USHRT_MAX)
3090	val = USHRT_MAX;
3091	WRITE_ONCE(up->pcrlen, val);
3092	udp_set_bit(UDPLITE_RECV_CC, sk);
3093	break;
3094
3095	default:
3096	err = -ENOPROTOOPT;
3097	break;
3098	}
3099
3100	return err;
3101	}
3102	EXPORT_IPV6_MOD(udp_lib_setsockopt);
3103
3104	int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
3105	unsigned int optlen)
3106	{
3107	if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
3108	return udp_lib_setsockopt(sk, level, optname,
3109	optval, optlen,
3110	push_pending_frames: udp_push_pending_frames);
3111	return ip_setsockopt(sk, level, optname, optval, optlen);
3112	}
3113
3114	int udp_lib_getsockopt(struct sock *sk, int level, int optname,
3115	char __user *optval, int __user *optlen)
3116	{
3117	struct udp_sock *up = udp_sk(sk);
3118	int val, len;
3119
3120	if (get_user(len, optlen))
3121	return -EFAULT;
3122
3123	if (len < 0)
3124	return -EINVAL;
3125
3126	len = min_t(unsigned int, len, sizeof(int));
3127
3128	switch (optname) {
3129	case UDP_CORK:
3130	val = udp_test_bit(CORK, sk);
3131	break;
3132
3133	case UDP_ENCAP:
3134	val = READ_ONCE(up->encap_type);
3135	break;
3136
3137	case UDP_NO_CHECK6_TX:
3138	val = udp_get_no_check6_tx(sk);
3139	break;
3140
3141	case UDP_NO_CHECK6_RX:
3142	val = udp_get_no_check6_rx(sk);
3143	break;
3144
3145	case UDP_SEGMENT:
3146	val = READ_ONCE(up->gso_size);
3147	break;
3148
3149	case UDP_GRO:
3150	val = udp_test_bit(GRO_ENABLED, sk);
3151	break;
3152
3153	/* The following two cannot be changed on UDP sockets, the return is
3154	* always 0 (which corresponds to the full checksum coverage of UDP). */
3155	case UDPLITE_SEND_CSCOV:
3156	val = READ_ONCE(up->pcslen);
3157	break;
3158
3159	case UDPLITE_RECV_CSCOV:
3160	val = READ_ONCE(up->pcrlen);
3161	break;
3162
3163	default:
3164	return -ENOPROTOOPT;
3165	}
3166
3167	if (put_user(len, optlen))
3168	return -EFAULT;
3169	if (copy_to_user(to: optval, from: &val, n: len))
3170	return -EFAULT;
3171	return 0;
3172	}
3173	EXPORT_IPV6_MOD(udp_lib_getsockopt);
3174
3175	int udp_getsockopt(struct sock *sk, int level, int optname,
3176	char __user *optval, int __user *optlen)
3177	{
3178	if (level == SOL_UDP || level == SOL_UDPLITE)
3179	return udp_lib_getsockopt(sk, level, optname, optval, optlen);
3180	return ip_getsockopt(sk, level, optname, optval, optlen);
3181	}
3182
3183	/**
3184	* udp_poll - wait for a UDP event.
3185	* @file: - file struct
3186	* @sock: - socket
3187	* @wait: - poll table
3188	*
3189	* This is same as datagram poll, except for the special case of
3190	* blocking sockets. If application is using a blocking fd
3191	* and a packet with checksum error is in the queue;
3192	* then it could get return from select indicating data available
3193	* but then block when reading it. Add special case code
3194	* to work around these arguably broken applications.
3195	*/
3196	__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
3197	{
3198	__poll_t mask = datagram_poll(file, sock, wait);
3199	struct sock *sk = sock->sk;
3200
3201	if (!skb_queue_empty_lockless(list: &udp_sk(sk)->reader_queue))
3202	mask |= EPOLLIN | EPOLLRDNORM;
3203
3204	/* Check for false positives due to checksum errors */
3205	if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
3206	!(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
3207	mask &= ~(EPOLLIN | EPOLLRDNORM);
3208
3209	/* psock ingress_msg queue should not contain any bad checksum frames */
3210	if (sk_is_readable(sk))
3211	mask |= EPOLLIN | EPOLLRDNORM;
3212	return mask;
3213
3214	}
3215	EXPORT_IPV6_MOD(udp_poll);
3216
3217	int udp_abort(struct sock *sk, int err)
3218	{
3219	if (!has_current_bpf_ctx())
3220	lock_sock(sk);
3221
3222	/* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
3223	* with close()
3224	*/
3225	if (sock_flag(sk, flag: SOCK_DEAD))
3226	goto out;
3227
3228	sk->sk_err = err;
3229	sk_error_report(sk);
3230	__udp_disconnect(sk, 0);
3231
3232	out:
3233	if (!has_current_bpf_ctx())
3234	release_sock(sk);
3235
3236	return 0;
3237	}
3238	EXPORT_IPV6_MOD_GPL(udp_abort);
3239
3240	struct proto udp_prot = {
3241	.name = "UDP",
3242	.owner = THIS_MODULE,
3243	.close = udp_lib_close,
3244	.pre_connect = udp_pre_connect,
3245	.connect = udp_connect,
3246	.disconnect = udp_disconnect,
3247	.ioctl = udp_ioctl,
3248	.init = udp_init_sock,
3249	.destroy = udp_destroy_sock,
3250	.setsockopt = udp_setsockopt,
3251	.getsockopt = udp_getsockopt,
3252	.sendmsg = udp_sendmsg,
3253	.recvmsg = udp_recvmsg,
3254	.splice_eof = udp_splice_eof,
3255	.release_cb = ip4_datagram_release_cb,
3256	.hash = udp_lib_hash,
3257	.unhash = udp_lib_unhash,
3258	.rehash = udp_v4_rehash,
3259	.get_port = udp_v4_get_port,
3260	.put_port = udp_lib_unhash,
3261	#ifdef CONFIG_BPF_SYSCALL
3262	.psock_update_sk_prot = udp_bpf_update_proto,
3263	#endif
3264	.memory_allocated = &net_aligned_data.udp_memory_allocated,
3265	.per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
3266
3267	.sysctl_mem = sysctl_udp_mem,
3268	.sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
3269	.sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
3270	.obj_size = sizeof(struct udp_sock),
3271	.h.udp_table = NULL,
3272	.diag_destroy = udp_abort,
3273	};
3274	EXPORT_SYMBOL(udp_prot);
3275
3276	/* ------------------------------------------------------------------------ */
3277	#ifdef CONFIG_PROC_FS
3278
3279	static unsigned short seq_file_family(const struct seq_file *seq);
3280	static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
3281	{
3282	unsigned short family = seq_file_family(seq);
3283
3284	/* AF_UNSPEC is used as a match all */
3285	return ((family == AF_UNSPEC || family == sk->sk_family) &&
3286	net_eq(net1: sock_net(sk), net2: seq_file_net(seq)));
3287	}
3288
3289	#ifdef CONFIG_BPF_SYSCALL
3290	static const struct seq_operations bpf_iter_udp_seq_ops;
3291	#endif
3292	static struct udp_table *udp_get_table_seq(struct seq_file *seq,
3293	struct net *net)
3294	{
3295	const struct udp_seq_afinfo *afinfo;
3296
3297	#ifdef CONFIG_BPF_SYSCALL
3298	if (seq->op == &bpf_iter_udp_seq_ops)
3299	return net->ipv4.udp_table;
3300	#endif
3301
3302	afinfo = pde_data(inode: file_inode(f: seq->file));
3303	return afinfo->udp_table ? : net->ipv4.udp_table;
3304	}
3305
3306	static struct sock *udp_get_first(struct seq_file *seq, int start)
3307	{
3308	struct udp_iter_state *state = seq->private;
3309	struct net *net = seq_file_net(seq);
3310	struct udp_table *udptable;
3311	struct sock *sk;
3312
3313	udptable = udp_get_table_seq(seq, net);
3314
3315	for (state->bucket = start; state->bucket <= udptable->mask;
3316	++state->bucket) {
3317	struct udp_hslot *hslot = &udptable->hash[state->bucket];
3318
3319	if (hlist_empty(h: &hslot->head))
3320	continue;
3321
3322	spin_lock_bh(lock: &hslot->lock);
3323	sk_for_each(sk, &hslot->head) {
3324	if (seq_sk_match(seq, sk))
3325	goto found;
3326	}
3327	spin_unlock_bh(lock: &hslot->lock);
3328	}
3329	sk = NULL;
3330	found:
3331	return sk;
3332	}
3333
3334	static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3335	{
3336	struct udp_iter_state *state = seq->private;
3337	struct net *net = seq_file_net(seq);
3338	struct udp_table *udptable;
3339
3340	do {
3341	sk = sk_next(sk);
3342	} while (sk && !seq_sk_match(seq, sk));
3343
3344	if (!sk) {
3345	udptable = udp_get_table_seq(seq, net);
3346
3347	if (state->bucket <= udptable->mask)
3348	spin_unlock_bh(lock: &udptable->hash[state->bucket].lock);
3349
3350	return udp_get_first(seq, start: state->bucket + 1);
3351	}
3352	return sk;
3353	}
3354
3355	static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3356	{
3357	struct sock *sk = udp_get_first(seq, start: 0);
3358
3359	if (sk)
3360	while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3361	--pos;
3362	return pos ? NULL : sk;
3363	}
3364
3365	void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3366	{
3367	struct udp_iter_state *state = seq->private;
3368	state->bucket = MAX_UDP_PORTS;
3369
3370	return *pos ? udp_get_idx(seq, pos: *pos-1) : SEQ_START_TOKEN;
3371	}
3372	EXPORT_IPV6_MOD(udp_seq_start);
3373
3374	void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3375	{
3376	struct sock *sk;
3377
3378	if (v == SEQ_START_TOKEN)
3379	sk = udp_get_idx(seq, pos: 0);
3380	else
3381	sk = udp_get_next(seq, sk: v);
3382
3383	++*pos;
3384	return sk;
3385	}
3386	EXPORT_IPV6_MOD(udp_seq_next);
3387
3388	void udp_seq_stop(struct seq_file *seq, void *v)
3389	{
3390	struct udp_iter_state *state = seq->private;
3391	struct udp_table *udptable;
3392
3393	udptable = udp_get_table_seq(seq, net: seq_file_net(seq));
3394
3395	if (state->bucket <= udptable->mask)
3396	spin_unlock_bh(lock: &udptable->hash[state->bucket].lock);
3397	}
3398	EXPORT_IPV6_MOD(udp_seq_stop);
3399
3400	/* ------------------------------------------------------------------------ */
3401	static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3402	int bucket)
3403	{
3404	struct inet_sock *inet = inet_sk(sp);
3405	__be32 dest = inet->inet_daddr;
3406	__be32 src = inet->inet_rcv_saddr;
3407	__u16 destp = ntohs(inet->inet_dport);
3408	__u16 srcp = ntohs(inet->inet_sport);
3409
3410	seq_printf(m: f, fmt: "%5d: %08X:%04X %08X:%04X"
3411	" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3412	bucket, src, srcp, dest, destp, sp->sk_state,
3413	sk_wmem_alloc_get(sk: sp),
3414	udp_rqueue_get(sk: sp),
3415	0, 0L, 0,
3416	from_kuid_munged(to: seq_user_ns(seq: f), uid: sk_uid(sk: sp)),
3417	0, sock_i_ino(sk: sp),
3418	refcount_read(r: &sp->sk_refcnt), sp,
3419	sk_drops_read(sk: sp));
3420	}
3421
3422	int udp4_seq_show(struct seq_file *seq, void *v)
3423	{
3424	seq_setwidth(m: seq, size: 127);
3425	if (v == SEQ_START_TOKEN)
3426	seq_puts(m: seq, s: " sl local_address rem_address st tx_queue "
3427	"rx_queue tr tm->when retrnsmt uid timeout "
3428	"inode ref pointer drops");
3429	else {
3430	struct udp_iter_state *state = seq->private;
3431
3432	udp4_format_sock(sp: v, f: seq, bucket: state->bucket);
3433	}
3434	seq_pad(m: seq, c: '\n');
3435	return 0;
3436	}
3437
3438	#ifdef CONFIG_BPF_SYSCALL
3439	struct bpf_iter__udp {
3440	__bpf_md_ptr(struct bpf_iter_meta *, meta);
3441	__bpf_md_ptr(struct udp_sock *, udp_sk);
3442	uid_t uid __aligned(8);
3443	int bucket __aligned(8);
3444	};
3445
3446	union bpf_udp_iter_batch_item {
3447	struct sock *sk;
3448	__u64 cookie;
3449	};
3450
3451	struct bpf_udp_iter_state {
3452	struct udp_iter_state state;
3453	unsigned int cur_sk;
3454	unsigned int end_sk;
3455	unsigned int max_sk;
3456	union bpf_udp_iter_batch_item *batch;
3457	};
3458
3459	static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3460	unsigned int new_batch_sz, gfp_t flags);
3461	static struct sock *bpf_iter_udp_resume(struct sock *first_sk,
3462	union bpf_udp_iter_batch_item *cookies,
3463	int n_cookies)
3464	{
3465	struct sock *sk = NULL;
3466	int i;
3467
3468	for (i = 0; i < n_cookies; i++) {
3469	sk = first_sk;
3470	udp_portaddr_for_each_entry_from(sk)
3471	if (cookies[i].cookie == atomic64_read(v: &sk->sk_cookie))
3472	goto done;
3473	}
3474	done:
3475	return sk;
3476	}
3477
3478	static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3479	{
3480	struct bpf_udp_iter_state *iter = seq->private;
3481	struct udp_iter_state *state = &iter->state;
3482	unsigned int find_cookie, end_cookie;
3483	struct net *net = seq_file_net(seq);
3484	struct udp_table *udptable;
3485	unsigned int batch_sks = 0;
3486	int resume_bucket;
3487	int resizes = 0;
3488	struct sock *sk;
3489	int err = 0;
3490
3491	resume_bucket = state->bucket;
3492
3493	/* The current batch is done, so advance the bucket. */
3494	if (iter->cur_sk == iter->end_sk)
3495	state->bucket++;
3496
3497	udptable = udp_get_table_seq(seq, net);
3498
3499	again:
3500	/* New batch for the next bucket.
3501	* Iterate over the hash table to find a bucket with sockets matching
3502	* the iterator attributes, and return the first matching socket from
3503	* the bucket. The remaining matched sockets from the bucket are batched
3504	* before releasing the bucket lock. This allows BPF programs that are
3505	* called in seq_show to acquire the bucket lock if needed.
3506	*/
3507	find_cookie = iter->cur_sk;
3508	end_cookie = iter->end_sk;
3509	iter->cur_sk = 0;
3510	iter->end_sk = 0;
3511	batch_sks = 0;
3512
3513	for (; state->bucket <= udptable->mask; state->bucket++) {
3514	struct udp_hslot *hslot2 = &udptable->hash2[state->bucket].hslot;
3515
3516	if (hlist_empty(h: &hslot2->head))
3517	goto next_bucket;
3518
3519	spin_lock_bh(lock: &hslot2->lock);
3520	sk = hlist_entry_safe(hslot2->head.first, struct sock,
3521	__sk_common.skc_portaddr_node);
3522	/* Resume from the first (in iteration order) unseen socket from
3523	* the last batch that still exists in resume_bucket. Most of
3524	* the time this will just be where the last iteration left off
3525	* in resume_bucket unless that socket disappeared between
3526	* reads.
3527	*/
3528	if (state->bucket == resume_bucket)
3529	sk = bpf_iter_udp_resume(first_sk: sk, cookies: &iter->batch[find_cookie],
3530	n_cookies: end_cookie - find_cookie);
3531	fill_batch:
3532	udp_portaddr_for_each_entry_from(sk) {
3533	if (seq_sk_match(seq, sk)) {
3534	if (iter->end_sk < iter->max_sk) {
3535	sock_hold(sk);
3536	iter->batch[iter->end_sk++].sk = sk;
3537	}
3538	batch_sks++;
3539	}
3540	}
3541
3542	/* Allocate a larger batch and try again. */
3543	if (unlikely(resizes <= 1 && iter->end_sk &&
3544	iter->end_sk != batch_sks)) {
3545	resizes++;
3546
3547	/* First, try with GFP_USER to maximize the chances of
3548	* grabbing more memory.
3549	*/
3550	if (resizes == 1) {
3551	spin_unlock_bh(lock: &hslot2->lock);
3552	err = bpf_iter_udp_realloc_batch(iter,
3553	new_batch_sz: batch_sks * 3 / 2,
3554	GFP_USER);
3555	if (err)
3556	return ERR_PTR(error: err);
3557	/* Start over. */
3558	goto again;
3559	}
3560
3561	/* Next, hold onto the lock, so the bucket doesn't
3562	* change while we get the rest of the sockets.
3563	*/
3564	err = bpf_iter_udp_realloc_batch(iter, new_batch_sz: batch_sks,
3565	GFP_NOWAIT);
3566	if (err) {
3567	spin_unlock_bh(lock: &hslot2->lock);
3568	return ERR_PTR(error: err);
3569	}
3570
3571	/* Pick up where we left off. */
3572	sk = iter->batch[iter->end_sk - 1].sk;
3573	sk = hlist_entry_safe(sk->__sk_common.skc_portaddr_node.next,
3574	struct sock,
3575	__sk_common.skc_portaddr_node);
3576	batch_sks = iter->end_sk;
3577	goto fill_batch;
3578	}
3579
3580	spin_unlock_bh(lock: &hslot2->lock);
3581
3582	if (iter->end_sk)
3583	break;
3584	next_bucket:
3585	resizes = 0;
3586	}
3587
3588	WARN_ON_ONCE(iter->end_sk != batch_sks);
3589	return iter->end_sk ? iter->batch[0].sk : NULL;
3590	}
3591
3592	static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3593	{
3594	struct bpf_udp_iter_state *iter = seq->private;
3595	struct sock *sk;
3596
3597	/* Whenever seq_next() is called, the iter->cur_sk is
3598	* done with seq_show(), so unref the iter->cur_sk.
3599	*/
3600	if (iter->cur_sk < iter->end_sk)
3601	sock_put(sk: iter->batch[iter->cur_sk++].sk);
3602
3603	/* After updating iter->cur_sk, check if there are more sockets
3604	* available in the current bucket batch.
3605	*/
3606	if (iter->cur_sk < iter->end_sk)
3607	sk = iter->batch[iter->cur_sk].sk;
3608	else
3609	/* Prepare a new batch. */
3610	sk = bpf_iter_udp_batch(seq);
3611
3612	++*pos;
3613	return sk;
3614	}
3615
3616	static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3617	{
3618	/* bpf iter does not support lseek, so it always
3619	* continue from where it was stop()-ped.
3620	*/
3621	if (*pos)
3622	return bpf_iter_udp_batch(seq);
3623
3624	return SEQ_START_TOKEN;
3625	}
3626
3627	static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3628	struct udp_sock *udp_sk, uid_t uid, int bucket)
3629	{
3630	struct bpf_iter__udp ctx;
3631
3632	meta->seq_num--; /* skip SEQ_START_TOKEN */
3633	ctx.meta = meta;
3634	ctx.udp_sk = udp_sk;
3635	ctx.uid = uid;
3636	ctx.bucket = bucket;
3637	return bpf_iter_run_prog(prog, ctx: &ctx);
3638	}
3639
3640	static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3641	{
3642	struct udp_iter_state *state = seq->private;
3643	struct bpf_iter_meta meta;
3644	struct bpf_prog *prog;
3645	struct sock *sk = v;
3646	uid_t uid;
3647	int ret;
3648
3649	if (v == SEQ_START_TOKEN)
3650	return 0;
3651
3652	lock_sock(sk);
3653
3654	if (unlikely(sk_unhashed(sk))) {
3655	ret = SEQ_SKIP;
3656	goto unlock;
3657	}
3658
3659	uid = from_kuid_munged(to: seq_user_ns(seq), uid: sk_uid(sk));
3660	meta.seq = seq;
3661	prog = bpf_iter_get_info(meta: &meta, in_stop: false);
3662	ret = udp_prog_seq_show(prog, meta: &meta, udp_sk: v, uid, bucket: state->bucket);
3663
3664	unlock:
3665	release_sock(sk);
3666	return ret;
3667	}
3668
3669	static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3670	{
3671	union bpf_udp_iter_batch_item *item;
3672	unsigned int cur_sk = iter->cur_sk;
3673	__u64 cookie;
3674
3675	/* Remember the cookies of the sockets we haven't seen yet, so we can
3676	* pick up where we left off next time around.
3677	*/
3678	while (cur_sk < iter->end_sk) {
3679	item = &iter->batch[cur_sk++];
3680	cookie = sock_gen_cookie(sk: item->sk);
3681	sock_put(sk: item->sk);
3682	item->cookie = cookie;
3683	}
3684	}
3685
3686	static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3687	{
3688	struct bpf_udp_iter_state *iter = seq->private;
3689	struct bpf_iter_meta meta;
3690	struct bpf_prog *prog;
3691
3692	if (!v) {
3693	meta.seq = seq;
3694	prog = bpf_iter_get_info(meta: &meta, in_stop: true);
3695	if (prog)
3696	(void)udp_prog_seq_show(prog, meta: &meta, udp_sk: v, uid: 0, bucket: 0);
3697	}
3698
3699	if (iter->cur_sk < iter->end_sk)
3700	bpf_iter_udp_put_batch(iter);
3701	}
3702
3703	static const struct seq_operations bpf_iter_udp_seq_ops = {
3704	.start = bpf_iter_udp_seq_start,
3705	.next = bpf_iter_udp_seq_next,
3706	.stop = bpf_iter_udp_seq_stop,
3707	.show = bpf_iter_udp_seq_show,
3708	};
3709	#endif
3710
3711	static unsigned short seq_file_family(const struct seq_file *seq)
3712	{
3713	const struct udp_seq_afinfo *afinfo;
3714
3715	#ifdef CONFIG_BPF_SYSCALL
3716	/* BPF iterator: bpf programs to filter sockets. */
3717	if (seq->op == &bpf_iter_udp_seq_ops)
3718	return AF_UNSPEC;
3719	#endif
3720
3721	/* Proc fs iterator */
3722	afinfo = pde_data(inode: file_inode(f: seq->file));
3723	return afinfo->family;
3724	}
3725
3726	const struct seq_operations udp_seq_ops = {
3727	.start = udp_seq_start,
3728	.next = udp_seq_next,
3729	.stop = udp_seq_stop,
3730	.show = udp4_seq_show,
3731	};
3732	EXPORT_IPV6_MOD(udp_seq_ops);
3733
3734	static struct udp_seq_afinfo udp4_seq_afinfo = {
3735	.family = AF_INET,
3736	.udp_table = NULL,
3737	};
3738
3739	static int __net_init udp4_proc_init_net(struct net *net)
3740	{
3741	if (!proc_create_net_data(name: "udp", mode: 0444, parent: net->proc_net, ops: &udp_seq_ops,
3742	state_size: sizeof(struct udp_iter_state), data: &udp4_seq_afinfo))
3743	return -ENOMEM;
3744	return 0;
3745	}
3746
3747	static void __net_exit udp4_proc_exit_net(struct net *net)
3748	{
3749	remove_proc_entry("udp", net->proc_net);
3750	}
3751
3752	static struct pernet_operations udp4_net_ops = {
3753	.init = udp4_proc_init_net,
3754	.exit = udp4_proc_exit_net,
3755	};
3756
3757	int __init udp4_proc_init(void)
3758	{
3759	return register_pernet_subsys(&udp4_net_ops);
3760	}
3761
3762	void udp4_proc_exit(void)
3763	{
3764	unregister_pernet_subsys(&udp4_net_ops);
3765	}
3766	#endif /* CONFIG_PROC_FS */
3767
3768	static __initdata unsigned long uhash_entries;
3769	static int __init set_uhash_entries(char *str)
3770	{
3771	ssize_t ret;
3772
3773	if (!str)
3774	return 0;
3775
3776	ret = kstrtoul(s: str, base: 0, res: &uhash_entries);
3777	if (ret)
3778	return 0;
3779
3780	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3781	uhash_entries = UDP_HTABLE_SIZE_MIN;
3782	return 1;
3783	}
3784	__setup("uhash_entries=", set_uhash_entries);
3785
3786	void __init udp_table_init(struct udp_table *table, const char *name)
3787	{
3788	unsigned int i, slot_size;
3789
3790	slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) +
3791	udp_hash4_slot_size();
3792	table->hash = alloc_large_system_hash(tablename: name,
3793	bucketsize: slot_size,
3794	numentries: uhash_entries,
3795	scale: 21, /* one slot per 2 MB */
3796	flags: 0,
3797	hash_shift: &table->log,
3798	hash_mask: &table->mask,
3799	UDP_HTABLE_SIZE_MIN,
3800	UDP_HTABLE_SIZE_MAX);
3801
3802	table->hash2 = (void *)(table->hash + (table->mask + 1));
3803	for (i = 0; i <= table->mask; i++) {
3804	INIT_HLIST_HEAD(&table->hash[i].head);
3805	table->hash[i].count = 0;
3806	spin_lock_init(&table->hash[i].lock);
3807	}
3808	for (i = 0; i <= table->mask; i++) {
3809	INIT_HLIST_HEAD(&table->hash2[i].hslot.head);
3810	table->hash2[i].hslot.count = 0;
3811	spin_lock_init(&table->hash2[i].hslot.lock);
3812	}
3813	udp_table_hash4_init(table);
3814	}
3815
3816	u32 udp_flow_hashrnd(void)
3817	{
3818	static u32 hashrnd __read_mostly;
3819
3820	net_get_random_once(&hashrnd, sizeof(hashrnd));
3821
3822	return hashrnd;
3823	}
3824	EXPORT_SYMBOL(udp_flow_hashrnd);
3825
3826	static void __net_init udp_sysctl_init(struct net *net)
3827	{
3828	net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3829	net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3830
3831	#ifdef CONFIG_NET_L3_MASTER_DEV
3832	net->ipv4.sysctl_udp_l3mdev_accept = 0;
3833	#endif
3834	}
3835
3836	static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3837	{
3838	struct udp_table *udptable;
3839	unsigned int slot_size;
3840	int i;
3841
3842	udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3843	if (!udptable)
3844	goto out;
3845
3846	slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) +
3847	udp_hash4_slot_size();
3848	udptable->hash = vmalloc_huge(size: hash_entries * slot_size,
3849	GFP_KERNEL_ACCOUNT);
3850	if (!udptable->hash)
3851	goto free_table;
3852
3853	udptable->hash2 = (void *)(udptable->hash + hash_entries);
3854	udptable->mask = hash_entries - 1;
3855	udptable->log = ilog2(hash_entries);
3856
3857	for (i = 0; i < hash_entries; i++) {
3858	INIT_HLIST_HEAD(&udptable->hash[i].head);
3859	udptable->hash[i].count = 0;
3860	spin_lock_init(&udptable->hash[i].lock);
3861
3862	INIT_HLIST_HEAD(&udptable->hash2[i].hslot.head);
3863	udptable->hash2[i].hslot.count = 0;
3864	spin_lock_init(&udptable->hash2[i].hslot.lock);
3865	}
3866	udp_table_hash4_init(table: udptable);
3867
3868	return udptable;
3869
3870	free_table:
3871	kfree(objp: udptable);
3872	out:
3873	return NULL;
3874	}
3875
3876	static void __net_exit udp_pernet_table_free(struct net *net)
3877	{
3878	struct udp_table *udptable = net->ipv4.udp_table;
3879
3880	if (udptable == &udp_table)
3881	return;
3882
3883	kvfree(addr: udptable->hash);
3884	kfree(objp: udptable);
3885	}
3886
3887	static void __net_init udp_set_table(struct net *net)
3888	{
3889	struct udp_table *udptable;
3890	unsigned int hash_entries;
3891	struct net *old_net;
3892
3893	if (net_eq(net1: net, net2: &init_net))
3894	goto fallback;
3895
3896	old_net = current->nsproxy->net_ns;
3897	hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3898	if (!hash_entries)
3899	goto fallback;
3900
3901	/* Set min to keep the bitmap on stack in udp_lib_get_port() */
3902	if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3903	hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3904	else
3905	hash_entries = roundup_pow_of_two(hash_entries);
3906
3907	udptable = udp_pernet_table_alloc(hash_entries);
3908	if (udptable) {
3909	net->ipv4.udp_table = udptable;
3910	} else {
3911	pr_warn("Failed to allocate UDP hash table (entries: %u) "
3912	"for a netns, fallback to the global one\n",
3913	hash_entries);
3914	fallback:
3915	net->ipv4.udp_table = &udp_table;
3916	}
3917	}
3918
3919	static int __net_init udp_pernet_init(struct net *net)
3920	{
3921	#if IS_ENABLED(CONFIG_NET_UDP_TUNNEL)
3922	int i;
3923
3924	/* No tunnel is configured */
3925	for (i = 0; i < ARRAY_SIZE(net->ipv4.udp_tunnel_gro); ++i) {
3926	INIT_HLIST_HEAD(&net->ipv4.udp_tunnel_gro[i].list);
3927	RCU_INIT_POINTER(net->ipv4.udp_tunnel_gro[i].sk, NULL);
3928	}
3929	#endif
3930	udp_sysctl_init(net);
3931	udp_set_table(net);
3932
3933	return 0;
3934	}
3935
3936	static void __net_exit udp_pernet_exit(struct net *net)
3937	{
3938	udp_pernet_table_free(net);
3939	}
3940
3941	static struct pernet_operations __net_initdata udp_sysctl_ops = {
3942	.init = udp_pernet_init,
3943	.exit = udp_pernet_exit,
3944	};
3945
3946	#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3947	DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3948	struct udp_sock *udp_sk, uid_t uid, int bucket)
3949
3950	static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3951	unsigned int new_batch_sz, gfp_t flags)
3952	{
3953	union bpf_udp_iter_batch_item *new_batch;
3954
3955	new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3956	flags | __GFP_NOWARN);
3957	if (!new_batch)
3958	return -ENOMEM;
3959
3960	if (flags != GFP_NOWAIT)
3961	bpf_iter_udp_put_batch(iter);
3962
3963	memcpy(new_batch, iter->batch, sizeof(*iter->batch) * iter->end_sk);
3964	kvfree(addr: iter->batch);
3965	iter->batch = new_batch;
3966	iter->max_sk = new_batch_sz;
3967
3968	return 0;
3969	}
3970
3971	#define INIT_BATCH_SZ 16
3972
3973	static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3974	{
3975	struct bpf_udp_iter_state *iter = priv_data;
3976	int ret;
3977
3978	ret = bpf_iter_init_seq_net(priv_data, aux);
3979	if (ret)
3980	return ret;
3981
3982	ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ, GFP_USER);
3983	if (ret)
3984	bpf_iter_fini_seq_net(priv_data);
3985
3986	iter->state.bucket = -1;
3987
3988	return ret;
3989	}
3990
3991	static void bpf_iter_fini_udp(void *priv_data)
3992	{
3993	struct bpf_udp_iter_state *iter = priv_data;
3994
3995	bpf_iter_fini_seq_net(priv_data);
3996	kvfree(addr: iter->batch);
3997	}
3998
3999	static const struct bpf_iter_seq_info udp_seq_info = {
4000	.seq_ops = &bpf_iter_udp_seq_ops,
4001	.init_seq_private = bpf_iter_init_udp,
4002	.fini_seq_private = bpf_iter_fini_udp,
4003	.seq_priv_size = sizeof(struct bpf_udp_iter_state),
4004	};
4005
4006	static struct bpf_iter_reg udp_reg_info = {
4007	.target = "udp",
4008	.ctx_arg_info_size = 1,
4009	.ctx_arg_info = {
4010	{ offsetof(struct bpf_iter__udp, udp_sk),
4011	PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
4012	},
4013	.seq_info = &udp_seq_info,
4014	};
4015
4016	static void __init bpf_iter_register(void)
4017	{
4018	udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
4019	if (bpf_iter_reg_target(reg_info: &udp_reg_info))
4020	pr_warn("Warning: could not register bpf iterator udp\n");
4021	}
4022	#endif
4023
4024	void __init udp_init(void)
4025	{
4026	unsigned long limit;
4027
4028	udp_table_init(table: &udp_table, name: "UDP");
4029	limit = nr_free_buffer_pages() / 8;
4030	limit = max(limit, 128UL);
4031	sysctl_udp_mem[0] = limit / 4 * 3;
4032	sysctl_udp_mem[1] = limit;
4033	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
4034
4035	if (register_pernet_subsys(&udp_sysctl_ops))
4036	panic(fmt: "UDP: failed to init sysctl parameters.\n");
4037
4038	#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
4039	bpf_iter_register();
4040	#endif
4041	}
4042