Linux Audio

Check our new training course

Linux BSP upgrade and security maintenance

Need help to get security updates for your Linux BSP?
Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		The User Datagram Protocol (UDP).
   8 *
   9 * Authors:	Ross Biro
  10 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  12 *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
  13 *		Hirokazu Takahashi, <taka@valinux.co.jp>
  14 *
  15 * Fixes:
  16 *		Alan Cox	:	verify_area() calls
  17 *		Alan Cox	: 	stopped close while in use off icmp
  18 *					messages. Not a fix but a botch that
  19 *					for udp at least is 'valid'.
  20 *		Alan Cox	:	Fixed icmp handling properly
  21 *		Alan Cox	: 	Correct error for oversized datagrams
  22 *		Alan Cox	:	Tidied select() semantics.
  23 *		Alan Cox	:	udp_err() fixed properly, also now
  24 *					select and read wake correctly on errors
  25 *		Alan Cox	:	udp_send verify_area moved to avoid mem leak
  26 *		Alan Cox	:	UDP can count its memory
  27 *		Alan Cox	:	send to an unknown connection causes
  28 *					an ECONNREFUSED off the icmp, but
  29 *					does NOT close.
  30 *		Alan Cox	:	Switched to new sk_buff handlers. No more backlog!
  31 *		Alan Cox	:	Using generic datagram code. Even smaller and the PEEK
  32 *					bug no longer crashes it.
  33 *		Fred Van Kempen	: 	Net2e support for sk->broadcast.
  34 *		Alan Cox	:	Uses skb_free_datagram
  35 *		Alan Cox	:	Added get/set sockopt support.
  36 *		Alan Cox	:	Broadcasting without option set returns EACCES.
  37 *		Alan Cox	:	No wakeup calls. Instead we now use the callbacks.
  38 *		Alan Cox	:	Use ip_tos and ip_ttl
  39 *		Alan Cox	:	SNMP Mibs
  40 *		Alan Cox	:	MSG_DONTROUTE, and 0.0.0.0 support.
  41 *		Matt Dillon	:	UDP length checks.
  42 *		Alan Cox	:	Smarter af_inet used properly.
  43 *		Alan Cox	:	Use new kernel side addressing.
  44 *		Alan Cox	:	Incorrect return on truncated datagram receive.
  45 *	Arnt Gulbrandsen 	:	New udp_send and stuff
  46 *		Alan Cox	:	Cache last socket
  47 *		Alan Cox	:	Route cache
  48 *		Jon Peatfield	:	Minor efficiency fix to sendto().
  49 *		Mike Shaver	:	RFC1122 checks.
  50 *		Alan Cox	:	Nonblocking error fix.
  51 *	Willy Konynenberg	:	Transparent proxying support.
  52 *		Mike McLagan	:	Routing by source
  53 *		David S. Miller	:	New socket lookup architecture.
  54 *					Last socket cache retained as it
  55 *					does have a high hit rate.
  56 *		Olaf Kirch	:	Don't linearise iovec on sendmsg.
  57 *		Andi Kleen	:	Some cleanups, cache destination entry
  58 *					for connect.
  59 *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
  60 *		Melvin Smith	:	Check msg_name not msg_namelen in sendto(),
  61 *					return ENOTCONN for unconnected sockets (POSIX)
  62 *		Janos Farkas	:	don't deliver multi/broadcasts to a different
  63 *					bound-to-device socket
  64 *	Hirokazu Takahashi	:	HW checksumming for outgoing UDP
  65 *					datagrams.
  66 *	Hirokazu Takahashi	:	sendfile() on UDP works now.
  67 *		Arnaldo C. Melo :	convert /proc/net/udp to seq_file
  68 *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
  69 *	Alexey Kuznetsov:		allow both IPv4 and IPv6 sockets to bind
  70 *					a single port at the same time.
  71 *	Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
  72 *	James Chapman		:	Add L2TP encapsulation type.
  73 */
  74
  75#define pr_fmt(fmt) "UDP: " fmt
  76
 
  77#include <linux/uaccess.h>
  78#include <asm/ioctls.h>
  79#include <linux/memblock.h>
  80#include <linux/highmem.h>
  81#include <linux/swap.h>
  82#include <linux/types.h>
  83#include <linux/fcntl.h>
  84#include <linux/module.h>
  85#include <linux/socket.h>
  86#include <linux/sockios.h>
  87#include <linux/igmp.h>
  88#include <linux/inetdevice.h>
  89#include <linux/in.h>
  90#include <linux/errno.h>
  91#include <linux/timer.h>
  92#include <linux/mm.h>
  93#include <linux/inet.h>
  94#include <linux/netdevice.h>
  95#include <linux/slab.h>
  96#include <net/tcp_states.h>
  97#include <linux/skbuff.h>
  98#include <linux/proc_fs.h>
  99#include <linux/seq_file.h>
 100#include <net/net_namespace.h>
 101#include <net/icmp.h>
 102#include <net/inet_hashtables.h>
 103#include <net/ip_tunnels.h>
 104#include <net/route.h>
 105#include <net/checksum.h>
 
 106#include <net/xfrm.h>
 107#include <trace/events/udp.h>
 108#include <linux/static_key.h>
 
 109#include <trace/events/skb.h>
 110#include <net/busy_poll.h>
 111#include "udp_impl.h"
 112#include <net/sock_reuseport.h>
 113#include <net/addrconf.h>
 114#include <net/udp_tunnel.h>
 
 
 
 
 115
 116struct udp_table udp_table __read_mostly;
 117EXPORT_SYMBOL(udp_table);
 118
 119long sysctl_udp_mem[3] __read_mostly;
 120EXPORT_SYMBOL(sysctl_udp_mem);
 121
 122atomic_long_t udp_memory_allocated;
 123EXPORT_SYMBOL(udp_memory_allocated);
 
 
 124
 125#define MAX_UDP_PORTS 65536
 126#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
 
 
 
 
 
 127
 128static int udp_lib_lport_inuse(struct net *net, __u16 num,
 129			       const struct udp_hslot *hslot,
 130			       unsigned long *bitmap,
 131			       struct sock *sk, unsigned int log)
 132{
 133	struct sock *sk2;
 134	kuid_t uid = sock_i_uid(sk);
 135
 136	sk_for_each(sk2, &hslot->head) {
 137		if (net_eq(sock_net(sk2), net) &&
 138		    sk2 != sk &&
 139		    (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
 140		    (!sk2->sk_reuse || !sk->sk_reuse) &&
 141		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 142		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 143		    inet_rcv_saddr_equal(sk, sk2, true)) {
 144			if (sk2->sk_reuseport && sk->sk_reuseport &&
 145			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 146			    uid_eq(uid, sock_i_uid(sk2))) {
 147				if (!bitmap)
 148					return 0;
 149			} else {
 150				if (!bitmap)
 151					return 1;
 152				__set_bit(udp_sk(sk2)->udp_port_hash >> log,
 153					  bitmap);
 154			}
 155		}
 156	}
 157	return 0;
 158}
 159
 160/*
 161 * Note: we still hold spinlock of primary hash chain, so no other writer
 162 * can insert/delete a socket with local_port == num
 163 */
 164static int udp_lib_lport_inuse2(struct net *net, __u16 num,
 165				struct udp_hslot *hslot2,
 166				struct sock *sk)
 167{
 168	struct sock *sk2;
 169	kuid_t uid = sock_i_uid(sk);
 170	int res = 0;
 171
 172	spin_lock(&hslot2->lock);
 173	udp_portaddr_for_each_entry(sk2, &hslot2->head) {
 174		if (net_eq(sock_net(sk2), net) &&
 175		    sk2 != sk &&
 176		    (udp_sk(sk2)->udp_port_hash == num) &&
 177		    (!sk2->sk_reuse || !sk->sk_reuse) &&
 178		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 179		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 180		    inet_rcv_saddr_equal(sk, sk2, true)) {
 181			if (sk2->sk_reuseport && sk->sk_reuseport &&
 182			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 183			    uid_eq(uid, sock_i_uid(sk2))) {
 184				res = 0;
 185			} else {
 186				res = 1;
 187			}
 188			break;
 189		}
 190	}
 191	spin_unlock(&hslot2->lock);
 192	return res;
 193}
 194
 195static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
 196{
 197	struct net *net = sock_net(sk);
 198	kuid_t uid = sock_i_uid(sk);
 199	struct sock *sk2;
 200
 201	sk_for_each(sk2, &hslot->head) {
 202		if (net_eq(sock_net(sk2), net) &&
 203		    sk2 != sk &&
 204		    sk2->sk_family == sk->sk_family &&
 205		    ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
 206		    (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
 207		    (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 208		    sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
 209		    inet_rcv_saddr_equal(sk, sk2, false)) {
 210			return reuseport_add_sock(sk, sk2,
 211						  inet_rcv_saddr_any(sk));
 212		}
 213	}
 214
 215	return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
 216}
 217
 218/**
 219 *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
 220 *
 221 *  @sk:          socket struct in question
 222 *  @snum:        port number to look up
 223 *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
 224 *                   with NULL address
 225 */
 226int udp_lib_get_port(struct sock *sk, unsigned short snum,
 227		     unsigned int hash2_nulladdr)
 228{
 
 229	struct udp_hslot *hslot, *hslot2;
 230	struct udp_table *udptable = sk->sk_prot->h.udp_table;
 231	int    error = 1;
 232	struct net *net = sock_net(sk);
 
 233
 234	if (!snum) {
 
 
 235		int low, high, remaining;
 236		unsigned int rand;
 237		unsigned short first, last;
 238		DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
 239
 240		inet_get_local_port_range(net, &low, &high);
 241		remaining = (high - low) + 1;
 242
 243		rand = prandom_u32();
 244		first = reciprocal_scale(rand, remaining) + low;
 245		/*
 246		 * force rand to be an odd multiple of UDP_HTABLE_SIZE
 247		 */
 248		rand = (rand | 1) * (udptable->mask + 1);
 249		last = first + udptable->mask + 1;
 250		do {
 251			hslot = udp_hashslot(udptable, net, first);
 252			bitmap_zero(bitmap, PORTS_PER_CHAIN);
 253			spin_lock_bh(&hslot->lock);
 254			udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
 255					    udptable->log);
 256
 257			snum = first;
 258			/*
 259			 * Iterate on all possible values of snum for this hash.
 260			 * Using steps of an odd multiple of UDP_HTABLE_SIZE
 261			 * give us randomization and full range coverage.
 262			 */
 263			do {
 264				if (low <= snum && snum <= high &&
 265				    !test_bit(snum >> udptable->log, bitmap) &&
 266				    !inet_is_local_reserved_port(net, snum))
 267					goto found;
 268				snum += rand;
 269			} while (snum != first);
 270			spin_unlock_bh(&hslot->lock);
 271			cond_resched();
 272		} while (++first != last);
 273		goto fail;
 274	} else {
 275		hslot = udp_hashslot(udptable, net, snum);
 276		spin_lock_bh(&hslot->lock);
 277		if (hslot->count > 10) {
 278			int exist;
 279			unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
 280
 281			slot2          &= udptable->mask;
 282			hash2_nulladdr &= udptable->mask;
 283
 284			hslot2 = udp_hashslot2(udptable, slot2);
 285			if (hslot->count < hslot2->count)
 286				goto scan_primary_hash;
 287
 288			exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
 289			if (!exist && (hash2_nulladdr != slot2)) {
 290				hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
 291				exist = udp_lib_lport_inuse2(net, snum, hslot2,
 292							     sk);
 293			}
 294			if (exist)
 295				goto fail_unlock;
 296			else
 297				goto found;
 298		}
 299scan_primary_hash:
 300		if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
 301			goto fail_unlock;
 302	}
 303found:
 304	inet_sk(sk)->inet_num = snum;
 305	udp_sk(sk)->udp_port_hash = snum;
 306	udp_sk(sk)->udp_portaddr_hash ^= snum;
 307	if (sk_unhashed(sk)) {
 308		if (sk->sk_reuseport &&
 309		    udp_reuseport_add_sock(sk, hslot)) {
 310			inet_sk(sk)->inet_num = 0;
 311			udp_sk(sk)->udp_port_hash = 0;
 312			udp_sk(sk)->udp_portaddr_hash ^= snum;
 313			goto fail_unlock;
 314		}
 315
 316		sk_add_node_rcu(sk, &hslot->head);
 317		hslot->count++;
 318		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 319
 320		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 321		spin_lock(&hslot2->lock);
 322		if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 323		    sk->sk_family == AF_INET6)
 324			hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
 325					   &hslot2->head);
 326		else
 327			hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
 328					   &hslot2->head);
 329		hslot2->count++;
 330		spin_unlock(&hslot2->lock);
 331	}
 332	sock_set_flag(sk, SOCK_RCU_FREE);
 333	error = 0;
 334fail_unlock:
 335	spin_unlock_bh(&hslot->lock);
 336fail:
 337	return error;
 338}
 339EXPORT_SYMBOL(udp_lib_get_port);
 340
 341int udp_v4_get_port(struct sock *sk, unsigned short snum)
 342{
 343	unsigned int hash2_nulladdr =
 344		ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
 345	unsigned int hash2_partial =
 346		ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
 347
 348	/* precompute partial secondary hash */
 349	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
 350	return udp_lib_get_port(sk, snum, hash2_nulladdr);
 351}
 352
 353static int compute_score(struct sock *sk, struct net *net,
 354			 __be32 saddr, __be16 sport,
 355			 __be32 daddr, unsigned short hnum,
 356			 int dif, int sdif)
 357{
 358	int score;
 359	struct inet_sock *inet;
 360	bool dev_match;
 361
 362	if (!net_eq(sock_net(sk), net) ||
 363	    udp_sk(sk)->udp_port_hash != hnum ||
 364	    ipv6_only_sock(sk))
 365		return -1;
 366
 367	if (sk->sk_rcv_saddr != daddr)
 368		return -1;
 369
 370	score = (sk->sk_family == PF_INET) ? 2 : 1;
 371
 372	inet = inet_sk(sk);
 373	if (inet->inet_daddr) {
 374		if (inet->inet_daddr != saddr)
 375			return -1;
 376		score += 4;
 377	}
 378
 379	if (inet->inet_dport) {
 380		if (inet->inet_dport != sport)
 381			return -1;
 382		score += 4;
 383	}
 384
 385	dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
 386					dif, sdif);
 387	if (!dev_match)
 388		return -1;
 389	score += 4;
 
 390
 391	if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
 392		score++;
 393	return score;
 394}
 395
 396static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
 397		       const __u16 lport, const __be32 faddr,
 398		       const __be16 fport)
 399{
 400	static u32 udp_ehash_secret __read_mostly;
 401
 402	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
 403
 404	return __inet_ehashfn(laddr, lport, faddr, fport,
 405			      udp_ehash_secret + net_hash_mix(net));
 406}
 407
 408/* called with rcu_read_lock() */
 409static struct sock *udp4_lib_lookup2(struct net *net,
 410				     __be32 saddr, __be16 sport,
 411				     __be32 daddr, unsigned int hnum,
 412				     int dif, int sdif,
 413				     struct udp_hslot *hslot2,
 414				     struct sk_buff *skb)
 415{
 416	struct sock *sk, *result;
 417	int score, badness;
 418	u32 hash = 0;
 419
 420	result = NULL;
 421	badness = 0;
 422	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
 423		score = compute_score(sk, net, saddr, sport,
 424				      daddr, hnum, dif, sdif);
 425		if (score > badness) {
 426			if (sk->sk_reuseport &&
 427			    sk->sk_state != TCP_ESTABLISHED) {
 428				hash = udp_ehashfn(net, daddr, hnum,
 429						   saddr, sport);
 430				result = reuseport_select_sock(sk, hash, skb,
 431							sizeof(struct udphdr));
 432				if (result && !reuseport_has_conns(sk, false))
 433					return result;
 434			}
 435			badness = score;
 436			result = sk;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 437		}
 438	}
 439	return result;
 440}
 441
 442/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
 443 * harder than this. -DaveM
 444 */
 445struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
 446		__be16 sport, __be32 daddr, __be16 dport, int dif,
 447		int sdif, struct udp_table *udptable, struct sk_buff *skb)
 448{
 449	struct sock *result;
 450	unsigned short hnum = ntohs(dport);
 451	unsigned int hash2, slot2;
 452	struct udp_hslot *hslot2;
 
 453
 454	hash2 = ipv4_portaddr_hash(net, daddr, hnum);
 455	slot2 = hash2 & udptable->mask;
 456	hslot2 = &udptable->hash2[slot2];
 457
 
 458	result = udp4_lib_lookup2(net, saddr, sport,
 459				  daddr, hnum, dif, sdif,
 460				  hslot2, skb);
 461	if (!result) {
 462		hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
 463		slot2 = hash2 & udptable->mask;
 464		hslot2 = &udptable->hash2[slot2];
 465
 466		result = udp4_lib_lookup2(net, saddr, sport,
 467					  htonl(INADDR_ANY), hnum, dif, sdif,
 468					  hslot2, skb);
 
 
 
 
 
 469	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 470	if (IS_ERR(result))
 471		return NULL;
 472	return result;
 473}
 474EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
 475
 476static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
 477						 __be16 sport, __be16 dport,
 478						 struct udp_table *udptable)
 479{
 480	const struct iphdr *iph = ip_hdr(skb);
 481
 482	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
 483				 iph->daddr, dport, inet_iif(skb),
 484				 inet_sdif(skb), udptable, skb);
 485}
 486
 487struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
 488				 __be16 sport, __be16 dport)
 489{
 490	const struct iphdr *iph = ip_hdr(skb);
 
 
 491
 492	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
 493				 iph->daddr, dport, inet_iif(skb),
 494				 inet_sdif(skb), &udp_table, NULL);
 
 
 495}
 496EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
 497
 498/* Must be called under rcu_read_lock().
 499 * Does increment socket refcount.
 500 */
 501#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
 502struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 503			     __be32 daddr, __be16 dport, int dif)
 504{
 505	struct sock *sk;
 506
 507	sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
 508			       dif, 0, &udp_table, NULL);
 509	if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
 510		sk = NULL;
 511	return sk;
 512}
 513EXPORT_SYMBOL_GPL(udp4_lib_lookup);
 514#endif
 515
 516static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
 517				       __be16 loc_port, __be32 loc_addr,
 518				       __be16 rmt_port, __be32 rmt_addr,
 519				       int dif, int sdif, unsigned short hnum)
 520{
 521	struct inet_sock *inet = inet_sk(sk);
 522
 523	if (!net_eq(sock_net(sk), net) ||
 524	    udp_sk(sk)->udp_port_hash != hnum ||
 525	    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
 526	    (inet->inet_dport != rmt_port && inet->inet_dport) ||
 527	    (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
 528	    ipv6_only_sock(sk) ||
 529	    !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
 530		return false;
 531	if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
 532		return false;
 533	return true;
 534}
 535
 536DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
 537void udp_encap_enable(void)
 538{
 539	static_branch_inc(&udp_encap_needed_key);
 540}
 541EXPORT_SYMBOL(udp_encap_enable);
 542
 
 
 
 
 
 
 543/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
 544 * through error handlers in encapsulations looking for a match.
 545 */
 546static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
 547{
 548	int i;
 549
 550	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
 551		int (*handler)(struct sk_buff *skb, u32 info);
 552		const struct ip_tunnel_encap_ops *encap;
 553
 554		encap = rcu_dereference(iptun_encaps[i]);
 555		if (!encap)
 556			continue;
 557		handler = encap->err_handler;
 558		if (handler && !handler(skb, info))
 559			return 0;
 560	}
 561
 562	return -ENOENT;
 563}
 564
 565/* Try to match ICMP errors to UDP tunnels by looking up a socket without
 566 * reversing source and destination port: this will match tunnels that force the
 567 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
 568 * lwtunnels might actually break this assumption by being configured with
 569 * different destination ports on endpoints, in this case we won't be able to
 570 * trace ICMP messages back to them.
 571 *
 572 * If this doesn't match any socket, probe tunnels with arbitrary destination
 573 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
 574 * we've sent packets to won't necessarily match the local destination port.
 575 *
 576 * Then ask the tunnel implementation to match the error against a valid
 577 * association.
 578 *
 579 * Return an error if we can't find a match, the socket if we need further
 580 * processing, zero otherwise.
 581 */
 582static struct sock *__udp4_lib_err_encap(struct net *net,
 583					 const struct iphdr *iph,
 584					 struct udphdr *uh,
 585					 struct udp_table *udptable,
 
 586					 struct sk_buff *skb, u32 info)
 587{
 
 588	int network_offset, transport_offset;
 589	struct sock *sk;
 590
 591	network_offset = skb_network_offset(skb);
 592	transport_offset = skb_transport_offset(skb);
 593
 594	/* Network header needs to point to the outer IPv4 header inside ICMP */
 595	skb_reset_network_header(skb);
 596
 597	/* Transport header needs to point to the UDP header */
 598	skb_set_transport_header(skb, iph->ihl << 2);
 599
 
 
 
 
 
 
 
 
 
 
 600	sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
 601			       iph->saddr, uh->dest, skb->dev->ifindex, 0,
 602			       udptable, NULL);
 603	if (sk) {
 604		int (*lookup)(struct sock *sk, struct sk_buff *skb);
 605		struct udp_sock *up = udp_sk(sk);
 606
 607		lookup = READ_ONCE(up->encap_err_lookup);
 608		if (!lookup || lookup(sk, skb))
 609			sk = NULL;
 610	}
 611
 
 612	if (!sk)
 613		sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
 614
 615	skb_set_transport_header(skb, transport_offset);
 616	skb_set_network_header(skb, network_offset);
 617
 618	return sk;
 619}
 620
 621/*
 622 * This routine is called by the ICMP module when it gets some
 623 * sort of error condition.  If err < 0 then the socket should
 624 * be closed and the error returned to the user.  If err > 0
 625 * it's just the icmp type << 8 | icmp code.
 626 * Header points to the ip header of the error packet. We move
 627 * on past this. Then (as it used to claim before adjustment)
 628 * header points to the first 8 bytes of the udp header.  We need
 629 * to find the appropriate port.
 630 */
 631
 632int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
 633{
 634	struct inet_sock *inet;
 635	const struct iphdr *iph = (const struct iphdr *)skb->data;
 636	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
 637	const int type = icmp_hdr(skb)->type;
 638	const int code = icmp_hdr(skb)->code;
 639	bool tunnel = false;
 640	struct sock *sk;
 641	int harderr;
 642	int err;
 643	struct net *net = dev_net(skb->dev);
 644
 645	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
 646			       iph->saddr, uh->source, skb->dev->ifindex,
 647			       inet_sdif(skb), udptable, NULL);
 648	if (!sk) {
 
 649		/* No socket for error: try tunnels before discarding */
 650		sk = ERR_PTR(-ENOENT);
 651		if (static_branch_unlikely(&udp_encap_needed_key)) {
 652			sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
 653						  info);
 654			if (!sk)
 655				return 0;
 656		}
 
 657
 658		if (IS_ERR(sk)) {
 659			__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 660			return PTR_ERR(sk);
 661		}
 662
 663		tunnel = true;
 664	}
 665
 666	err = 0;
 667	harderr = 0;
 668	inet = inet_sk(sk);
 669
 670	switch (type) {
 671	default:
 672	case ICMP_TIME_EXCEEDED:
 673		err = EHOSTUNREACH;
 674		break;
 675	case ICMP_SOURCE_QUENCH:
 676		goto out;
 677	case ICMP_PARAMETERPROB:
 678		err = EPROTO;
 679		harderr = 1;
 680		break;
 681	case ICMP_DEST_UNREACH:
 682		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
 683			ipv4_sk_update_pmtu(skb, sk, info);
 684			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
 685				err = EMSGSIZE;
 686				harderr = 1;
 687				break;
 688			}
 689			goto out;
 690		}
 691		err = EHOSTUNREACH;
 692		if (code <= NR_ICMP_UNREACH) {
 693			harderr = icmp_err_convert[code].fatal;
 694			err = icmp_err_convert[code].errno;
 695		}
 696		break;
 697	case ICMP_REDIRECT:
 698		ipv4_sk_redirect(skb, sk);
 699		goto out;
 700	}
 701
 702	/*
 703	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
 704	 *	4.1.3.3.
 705	 */
 706	if (tunnel) {
 707		/* ...not for tunnels though: we don't have a sending socket */
 
 
 
 708		goto out;
 709	}
 710	if (!inet->recverr) {
 711		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
 712			goto out;
 713	} else
 714		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
 715
 716	sk->sk_err = err;
 717	sk->sk_error_report(sk);
 718out:
 719	return 0;
 720}
 721
 722int udp_err(struct sk_buff *skb, u32 info)
 723{
 724	return __udp4_lib_err(skb, info, &udp_table);
 725}
 726
 727/*
 728 * Throw away all pending data and cancel the corking. Socket is locked.
 729 */
 730void udp_flush_pending_frames(struct sock *sk)
 731{
 732	struct udp_sock *up = udp_sk(sk);
 733
 734	if (up->pending) {
 735		up->len = 0;
 736		up->pending = 0;
 737		ip_flush_pending_frames(sk);
 738	}
 739}
 740EXPORT_SYMBOL(udp_flush_pending_frames);
 741
 742/**
 743 * 	udp4_hwcsum  -  handle outgoing HW checksumming
 744 * 	@skb: 	sk_buff containing the filled-in UDP header
 745 * 	        (checksum field must be zeroed out)
 746 *	@src:	source IP address
 747 *	@dst:	destination IP address
 748 */
 749void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
 750{
 751	struct udphdr *uh = udp_hdr(skb);
 752	int offset = skb_transport_offset(skb);
 753	int len = skb->len - offset;
 754	int hlen = len;
 755	__wsum csum = 0;
 756
 757	if (!skb_has_frag_list(skb)) {
 758		/*
 759		 * Only one fragment on the socket.
 760		 */
 761		skb->csum_start = skb_transport_header(skb) - skb->head;
 762		skb->csum_offset = offsetof(struct udphdr, check);
 763		uh->check = ~csum_tcpudp_magic(src, dst, len,
 764					       IPPROTO_UDP, 0);
 765	} else {
 766		struct sk_buff *frags;
 767
 768		/*
 769		 * HW-checksum won't work as there are two or more
 770		 * fragments on the socket so that all csums of sk_buffs
 771		 * should be together
 772		 */
 773		skb_walk_frags(skb, frags) {
 774			csum = csum_add(csum, frags->csum);
 775			hlen -= frags->len;
 776		}
 777
 778		csum = skb_checksum(skb, offset, hlen, csum);
 779		skb->ip_summed = CHECKSUM_NONE;
 780
 781		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
 782		if (uh->check == 0)
 783			uh->check = CSUM_MANGLED_0;
 784	}
 785}
 786EXPORT_SYMBOL_GPL(udp4_hwcsum);
 787
 788/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
 789 * for the simple case like when setting the checksum for a UDP tunnel.
 790 */
 791void udp_set_csum(bool nocheck, struct sk_buff *skb,
 792		  __be32 saddr, __be32 daddr, int len)
 793{
 794	struct udphdr *uh = udp_hdr(skb);
 795
 796	if (nocheck) {
 797		uh->check = 0;
 798	} else if (skb_is_gso(skb)) {
 799		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 800	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 801		uh->check = 0;
 802		uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
 803		if (uh->check == 0)
 804			uh->check = CSUM_MANGLED_0;
 805	} else {
 806		skb->ip_summed = CHECKSUM_PARTIAL;
 807		skb->csum_start = skb_transport_header(skb) - skb->head;
 808		skb->csum_offset = offsetof(struct udphdr, check);
 809		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 810	}
 811}
 812EXPORT_SYMBOL(udp_set_csum);
 813
 814static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
 815			struct inet_cork *cork)
 816{
 817	struct sock *sk = skb->sk;
 818	struct inet_sock *inet = inet_sk(sk);
 819	struct udphdr *uh;
 820	int err = 0;
 821	int is_udplite = IS_UDPLITE(sk);
 822	int offset = skb_transport_offset(skb);
 823	int len = skb->len - offset;
 824	int datalen = len - sizeof(*uh);
 825	__wsum csum = 0;
 826
 827	/*
 828	 * Create a UDP header
 829	 */
 830	uh = udp_hdr(skb);
 831	uh->source = inet->inet_sport;
 832	uh->dest = fl4->fl4_dport;
 833	uh->len = htons(len);
 834	uh->check = 0;
 835
 836	if (cork->gso_size) {
 837		const int hlen = skb_network_header_len(skb) +
 838				 sizeof(struct udphdr);
 839
 840		if (hlen + cork->gso_size > cork->fragsize) {
 841			kfree_skb(skb);
 842			return -EINVAL;
 843		}
 844		if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
 845			kfree_skb(skb);
 846			return -EINVAL;
 847		}
 848		if (sk->sk_no_check_tx) {
 849			kfree_skb(skb);
 850			return -EINVAL;
 851		}
 852		if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
 853		    dst_xfrm(skb_dst(skb))) {
 854			kfree_skb(skb);
 855			return -EIO;
 856		}
 857
 858		if (datalen > cork->gso_size) {
 859			skb_shinfo(skb)->gso_size = cork->gso_size;
 860			skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
 861			skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
 862								 cork->gso_size);
 863		}
 864		goto csum_partial;
 865	}
 866
 867	if (is_udplite)  				 /*     UDP-Lite      */
 868		csum = udplite_csum(skb);
 869
 870	else if (sk->sk_no_check_tx) {			 /* UDP csum off */
 871
 872		skb->ip_summed = CHECKSUM_NONE;
 873		goto send;
 874
 875	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
 876csum_partial:
 877
 878		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
 879		goto send;
 880
 881	} else
 882		csum = udp_csum(skb);
 883
 884	/* add protocol-dependent pseudo-header */
 885	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
 886				      sk->sk_protocol, csum);
 887	if (uh->check == 0)
 888		uh->check = CSUM_MANGLED_0;
 889
 890send:
 891	err = ip_send_skb(sock_net(sk), skb);
 892	if (err) {
 893		if (err == -ENOBUFS && !inet->recverr) {
 
 894			UDP_INC_STATS(sock_net(sk),
 895				      UDP_MIB_SNDBUFERRORS, is_udplite);
 896			err = 0;
 897		}
 898	} else
 899		UDP_INC_STATS(sock_net(sk),
 900			      UDP_MIB_OUTDATAGRAMS, is_udplite);
 901	return err;
 902}
 903
 904/*
 905 * Push out all pending data as one UDP datagram. Socket is locked.
 906 */
 907int udp_push_pending_frames(struct sock *sk)
 908{
 909	struct udp_sock  *up = udp_sk(sk);
 910	struct inet_sock *inet = inet_sk(sk);
 911	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
 912	struct sk_buff *skb;
 913	int err = 0;
 914
 915	skb = ip_finish_skb(sk, fl4);
 916	if (!skb)
 917		goto out;
 918
 919	err = udp_send_skb(skb, fl4, &inet->cork.base);
 920
 921out:
 922	up->len = 0;
 923	up->pending = 0;
 924	return err;
 925}
 926EXPORT_SYMBOL(udp_push_pending_frames);
 927
 928static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
 929{
 930	switch (cmsg->cmsg_type) {
 931	case UDP_SEGMENT:
 932		if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
 933			return -EINVAL;
 934		*gso_size = *(__u16 *)CMSG_DATA(cmsg);
 935		return 0;
 936	default:
 937		return -EINVAL;
 938	}
 939}
 940
 941int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
 942{
 943	struct cmsghdr *cmsg;
 944	bool need_ip = false;
 945	int err;
 946
 947	for_each_cmsghdr(cmsg, msg) {
 948		if (!CMSG_OK(msg, cmsg))
 949			return -EINVAL;
 950
 951		if (cmsg->cmsg_level != SOL_UDP) {
 952			need_ip = true;
 953			continue;
 954		}
 955
 956		err = __udp_cmsg_send(cmsg, gso_size);
 957		if (err)
 958			return err;
 959	}
 960
 961	return need_ip;
 962}
 963EXPORT_SYMBOL_GPL(udp_cmsg_send);
 964
 965int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 966{
 967	struct inet_sock *inet = inet_sk(sk);
 968	struct udp_sock *up = udp_sk(sk);
 969	DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
 970	struct flowi4 fl4_stack;
 971	struct flowi4 *fl4;
 972	int ulen = len;
 973	struct ipcm_cookie ipc;
 974	struct rtable *rt = NULL;
 975	int free = 0;
 976	int connected = 0;
 977	__be32 daddr, faddr, saddr;
 
 978	__be16 dport;
 979	u8  tos;
 980	int err, is_udplite = IS_UDPLITE(sk);
 981	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
 982	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
 983	struct sk_buff *skb;
 984	struct ip_options_data opt_copy;
 
 985
 986	if (len > 0xFFFF)
 987		return -EMSGSIZE;
 988
 989	/*
 990	 *	Check the flags.
 991	 */
 992
 993	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
 994		return -EOPNOTSUPP;
 995
 996	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
 997
 998	fl4 = &inet->cork.fl.u.ip4;
 999	if (up->pending) {
1000		/*
1001		 * There are pending frames.
1002		 * The socket lock must be held while it's corked.
1003		 */
1004		lock_sock(sk);
1005		if (likely(up->pending)) {
1006			if (unlikely(up->pending != AF_INET)) {
1007				release_sock(sk);
1008				return -EINVAL;
1009			}
1010			goto do_append_data;
1011		}
1012		release_sock(sk);
1013	}
1014	ulen += sizeof(struct udphdr);
1015
1016	/*
1017	 *	Get and verify the address.
1018	 */
1019	if (usin) {
1020		if (msg->msg_namelen < sizeof(*usin))
1021			return -EINVAL;
1022		if (usin->sin_family != AF_INET) {
1023			if (usin->sin_family != AF_UNSPEC)
1024				return -EAFNOSUPPORT;
1025		}
1026
1027		daddr = usin->sin_addr.s_addr;
1028		dport = usin->sin_port;
1029		if (dport == 0)
1030			return -EINVAL;
1031	} else {
1032		if (sk->sk_state != TCP_ESTABLISHED)
1033			return -EDESTADDRREQ;
1034		daddr = inet->inet_daddr;
1035		dport = inet->inet_dport;
1036		/* Open fast path for connected socket.
1037		   Route will not be used, if at least one option is set.
1038		 */
1039		connected = 1;
1040	}
1041
1042	ipcm_init_sk(&ipc, inet);
1043	ipc.gso_size = up->gso_size;
1044
1045	if (msg->msg_controllen) {
1046		err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1047		if (err > 0)
1048			err = ip_cmsg_send(sk, msg, &ipc,
1049					   sk->sk_family == AF_INET6);
1050		if (unlikely(err < 0)) {
1051			kfree(ipc.opt);
1052			return err;
1053		}
1054		if (ipc.opt)
1055			free = 1;
1056		connected = 0;
1057	}
1058	if (!ipc.opt) {
1059		struct ip_options_rcu *inet_opt;
1060
1061		rcu_read_lock();
1062		inet_opt = rcu_dereference(inet->inet_opt);
1063		if (inet_opt) {
1064			memcpy(&opt_copy, inet_opt,
1065			       sizeof(*inet_opt) + inet_opt->opt.optlen);
1066			ipc.opt = &opt_copy.opt;
1067		}
1068		rcu_read_unlock();
1069	}
1070
1071	if (cgroup_bpf_enabled && !connected) {
1072		err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1073					    (struct sockaddr *)usin, &ipc.addr);
 
 
1074		if (err)
1075			goto out_free;
1076		if (usin) {
1077			if (usin->sin_port == 0) {
1078				/* BPF program set invalid port. Reject it. */
1079				err = -EINVAL;
1080				goto out_free;
1081			}
1082			daddr = usin->sin_addr.s_addr;
1083			dport = usin->sin_port;
1084		}
1085	}
1086
1087	saddr = ipc.addr;
1088	ipc.addr = faddr = daddr;
1089
1090	if (ipc.opt && ipc.opt->opt.srr) {
1091		if (!daddr) {
1092			err = -EINVAL;
1093			goto out_free;
1094		}
1095		faddr = ipc.opt->opt.faddr;
1096		connected = 0;
1097	}
1098	tos = get_rttos(&ipc, inet);
1099	if (sock_flag(sk, SOCK_LOCALROUTE) ||
1100	    (msg->msg_flags & MSG_DONTROUTE) ||
1101	    (ipc.opt && ipc.opt->opt.is_strictroute)) {
1102		tos |= RTO_ONLINK;
1103		connected = 0;
1104	}
1105
 
1106	if (ipv4_is_multicast(daddr)) {
1107		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1108			ipc.oif = inet->mc_index;
1109		if (!saddr)
1110			saddr = inet->mc_addr;
1111		connected = 0;
1112	} else if (!ipc.oif) {
1113		ipc.oif = inet->uc_index;
1114	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1115		/* oif is set, packet is to local broadcast and
1116		 * and uc_index is set. oif is most likely set
1117		 * by sk_bound_dev_if. If uc_index != oif check if the
1118		 * oif is an L3 master and uc_index is an L3 slave.
1119		 * If so, we want to allow the send using the uc_index.
1120		 */
1121		if (ipc.oif != inet->uc_index &&
1122		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1123							      inet->uc_index)) {
1124			ipc.oif = inet->uc_index;
1125		}
1126	}
1127
1128	if (connected)
1129		rt = (struct rtable *)sk_dst_check(sk, 0);
1130
1131	if (!rt) {
1132		struct net *net = sock_net(sk);
1133		__u8 flow_flags = inet_sk_flowi_flags(sk);
1134
1135		fl4 = &fl4_stack;
1136
1137		flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1138				   RT_SCOPE_UNIVERSE, sk->sk_protocol,
1139				   flow_flags,
1140				   faddr, saddr, dport, inet->inet_sport,
1141				   sk->sk_uid);
1142
1143		security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1144		rt = ip_route_output_flow(net, fl4, sk);
1145		if (IS_ERR(rt)) {
1146			err = PTR_ERR(rt);
1147			rt = NULL;
1148			if (err == -ENETUNREACH)
1149				IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1150			goto out;
1151		}
1152
1153		err = -EACCES;
1154		if ((rt->rt_flags & RTCF_BROADCAST) &&
1155		    !sock_flag(sk, SOCK_BROADCAST))
1156			goto out;
1157		if (connected)
1158			sk_dst_set(sk, dst_clone(&rt->dst));
1159	}
1160
1161	if (msg->msg_flags&MSG_CONFIRM)
1162		goto do_confirm;
1163back_from_confirm:
1164
1165	saddr = fl4->saddr;
1166	if (!ipc.addr)
1167		daddr = ipc.addr = fl4->daddr;
1168
1169	/* Lockless fast path for the non-corking case. */
1170	if (!corkreq) {
1171		struct inet_cork cork;
1172
1173		skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1174				  sizeof(struct udphdr), &ipc, &rt,
1175				  &cork, msg->msg_flags);
1176		err = PTR_ERR(skb);
1177		if (!IS_ERR_OR_NULL(skb))
1178			err = udp_send_skb(skb, fl4, &cork);
1179		goto out;
1180	}
1181
1182	lock_sock(sk);
1183	if (unlikely(up->pending)) {
1184		/* The socket is already corked while preparing it. */
1185		/* ... which is an evident application bug. --ANK */
1186		release_sock(sk);
1187
1188		net_dbg_ratelimited("socket already corked\n");
1189		err = -EINVAL;
1190		goto out;
1191	}
1192	/*
1193	 *	Now cork the socket to pend data.
1194	 */
1195	fl4 = &inet->cork.fl.u.ip4;
1196	fl4->daddr = daddr;
1197	fl4->saddr = saddr;
1198	fl4->fl4_dport = dport;
1199	fl4->fl4_sport = inet->inet_sport;
1200	up->pending = AF_INET;
1201
1202do_append_data:
1203	up->len += ulen;
1204	err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1205			     sizeof(struct udphdr), &ipc, &rt,
1206			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1207	if (err)
1208		udp_flush_pending_frames(sk);
1209	else if (!corkreq)
1210		err = udp_push_pending_frames(sk);
1211	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1212		up->pending = 0;
1213	release_sock(sk);
1214
1215out:
1216	ip_rt_put(rt);
1217out_free:
1218	if (free)
1219		kfree(ipc.opt);
1220	if (!err)
1221		return len;
1222	/*
1223	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1224	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1225	 * we don't have a good statistic (IpOutDiscards but it can be too many
1226	 * things).  We could add another new stat but at least for now that
1227	 * seems like overkill.
1228	 */
1229	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1230		UDP_INC_STATS(sock_net(sk),
1231			      UDP_MIB_SNDBUFERRORS, is_udplite);
1232	}
1233	return err;
1234
1235do_confirm:
1236	if (msg->msg_flags & MSG_PROBE)
1237		dst_confirm_neigh(&rt->dst, &fl4->daddr);
1238	if (!(msg->msg_flags&MSG_PROBE) || len)
1239		goto back_from_confirm;
1240	err = 0;
1241	goto out;
1242}
1243EXPORT_SYMBOL(udp_sendmsg);
1244
1245int udp_sendpage(struct sock *sk, struct page *page, int offset,
1246		 size_t size, int flags)
1247{
1248	struct inet_sock *inet = inet_sk(sk);
1249	struct udp_sock *up = udp_sk(sk);
1250	int ret;
1251
1252	if (flags & MSG_SENDPAGE_NOTLAST)
1253		flags |= MSG_MORE;
1254
1255	if (!up->pending) {
1256		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
1257
1258		/* Call udp_sendmsg to specify destination address which
1259		 * sendpage interface can't pass.
1260		 * This will succeed only when the socket is connected.
1261		 */
1262		ret = udp_sendmsg(sk, &msg, 0);
1263		if (ret < 0)
1264			return ret;
1265	}
1266
1267	lock_sock(sk);
1268
1269	if (unlikely(!up->pending)) {
1270		release_sock(sk);
1271
1272		net_dbg_ratelimited("cork failed\n");
1273		return -EINVAL;
1274	}
1275
1276	ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1277			     page, offset, size, flags);
1278	if (ret == -EOPNOTSUPP) {
1279		release_sock(sk);
1280		return sock_no_sendpage(sk->sk_socket, page, offset,
1281					size, flags);
1282	}
1283	if (ret < 0) {
1284		udp_flush_pending_frames(sk);
1285		goto out;
1286	}
1287
1288	up->len += size;
1289	if (!(up->corkflag || (flags&MSG_MORE)))
1290		ret = udp_push_pending_frames(sk);
1291	if (!ret)
1292		ret = size;
1293out:
1294	release_sock(sk);
1295	return ret;
1296}
 
1297
1298#define UDP_SKB_IS_STATELESS 0x80000000
1299
1300/* all head states (dst, sk, nf conntrack) except skb extensions are
1301 * cleared by udp_rcv().
1302 *
1303 * We need to preserve secpath, if present, to eventually process
1304 * IP_CMSG_PASSSEC at recvmsg() time.
1305 *
1306 * Other extensions can be cleared.
1307 */
1308static bool udp_try_make_stateless(struct sk_buff *skb)
1309{
1310	if (!skb_has_extensions(skb))
1311		return true;
1312
1313	if (!secpath_exists(skb)) {
1314		skb_ext_reset(skb);
1315		return true;
1316	}
1317
1318	return false;
1319}
1320
1321static void udp_set_dev_scratch(struct sk_buff *skb)
1322{
1323	struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1324
1325	BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1326	scratch->_tsize_state = skb->truesize;
1327#if BITS_PER_LONG == 64
1328	scratch->len = skb->len;
1329	scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1330	scratch->is_linear = !skb_is_nonlinear(skb);
1331#endif
1332	if (udp_try_make_stateless(skb))
1333		scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1334}
1335
1336static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1337{
1338	/* We come here after udp_lib_checksum_complete() returned 0.
1339	 * This means that __skb_checksum_complete() might have
1340	 * set skb->csum_valid to 1.
1341	 * On 64bit platforms, we can set csum_unnecessary
1342	 * to true, but only if the skb is not shared.
1343	 */
1344#if BITS_PER_LONG == 64
1345	if (!skb_shared(skb))
1346		udp_skb_scratch(skb)->csum_unnecessary = true;
1347#endif
1348}
1349
1350static int udp_skb_truesize(struct sk_buff *skb)
1351{
1352	return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1353}
1354
1355static bool udp_skb_has_head_state(struct sk_buff *skb)
1356{
1357	return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1358}
1359
1360/* fully reclaim rmem/fwd memory allocated for skb */
1361static void udp_rmem_release(struct sock *sk, int size, int partial,
1362			     bool rx_queue_lock_held)
1363{
1364	struct udp_sock *up = udp_sk(sk);
1365	struct sk_buff_head *sk_queue;
1366	int amt;
1367
1368	if (likely(partial)) {
1369		up->forward_deficit += size;
1370		size = up->forward_deficit;
1371		if (size < (sk->sk_rcvbuf >> 2))
 
1372			return;
1373	} else {
1374		size += up->forward_deficit;
1375	}
1376	up->forward_deficit = 0;
1377
1378	/* acquire the sk_receive_queue for fwd allocated memory scheduling,
1379	 * if the called don't held it already
1380	 */
1381	sk_queue = &sk->sk_receive_queue;
1382	if (!rx_queue_lock_held)
1383		spin_lock(&sk_queue->lock);
1384
1385
1386	sk->sk_forward_alloc += size;
1387	amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1388	sk->sk_forward_alloc -= amt;
1389
1390	if (amt)
1391		__sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1392
1393	atomic_sub(size, &sk->sk_rmem_alloc);
1394
1395	/* this can save us from acquiring the rx queue lock on next receive */
1396	skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1397
1398	if (!rx_queue_lock_held)
1399		spin_unlock(&sk_queue->lock);
1400}
1401
1402/* Note: called with reader_queue.lock held.
1403 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1404 * This avoids a cache line miss while receive_queue lock is held.
1405 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1406 */
1407void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1408{
1409	prefetch(&skb->data);
1410	udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1411}
1412EXPORT_SYMBOL(udp_skb_destructor);
1413
1414/* as above, but the caller held the rx queue lock, too */
1415static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1416{
1417	prefetch(&skb->data);
1418	udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1419}
1420
1421/* Idea of busylocks is to let producers grab an extra spinlock
1422 * to relieve pressure on the receive_queue spinlock shared by consumer.
1423 * Under flood, this means that only one producer can be in line
1424 * trying to acquire the receive_queue spinlock.
1425 * These busylock can be allocated on a per cpu manner, instead of a
1426 * per socket one (that would consume a cache line per socket)
1427 */
1428static int udp_busylocks_log __read_mostly;
1429static spinlock_t *udp_busylocks __read_mostly;
1430
1431static spinlock_t *busylock_acquire(void *ptr)
1432{
1433	spinlock_t *busy;
1434
1435	busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1436	spin_lock(busy);
1437	return busy;
1438}
1439
1440static void busylock_release(spinlock_t *busy)
1441{
1442	if (busy)
1443		spin_unlock(busy);
1444}
1445
 
 
 
 
 
 
 
 
 
 
 
1446int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1447{
1448	struct sk_buff_head *list = &sk->sk_receive_queue;
1449	int rmem, delta, amt, err = -ENOMEM;
1450	spinlock_t *busy = NULL;
1451	int size;
1452
1453	/* try to avoid the costly atomic add/sub pair when the receive
1454	 * queue is full; always allow at least a packet
1455	 */
1456	rmem = atomic_read(&sk->sk_rmem_alloc);
1457	if (rmem > sk->sk_rcvbuf)
1458		goto drop;
1459
1460	/* Under mem pressure, it might be helpful to help udp_recvmsg()
1461	 * having linear skbs :
1462	 * - Reduce memory overhead and thus increase receive queue capacity
1463	 * - Less cache line misses at copyout() time
1464	 * - Less work at consume_skb() (less alien page frag freeing)
1465	 */
1466	if (rmem > (sk->sk_rcvbuf >> 1)) {
1467		skb_condense(skb);
1468
1469		busy = busylock_acquire(sk);
1470	}
1471	size = skb->truesize;
1472	udp_set_dev_scratch(skb);
1473
1474	/* we drop only if the receive buf is full and the receive
1475	 * queue contains some other skb
1476	 */
1477	rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1478	if (rmem > (size + sk->sk_rcvbuf))
1479		goto uncharge_drop;
1480
1481	spin_lock(&list->lock);
1482	if (size >= sk->sk_forward_alloc) {
1483		amt = sk_mem_pages(size);
1484		delta = amt << SK_MEM_QUANTUM_SHIFT;
1485		if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1486			err = -ENOBUFS;
1487			spin_unlock(&list->lock);
1488			goto uncharge_drop;
1489		}
1490
1491		sk->sk_forward_alloc += delta;
1492	}
1493
1494	sk->sk_forward_alloc -= size;
1495
1496	/* no need to setup a destructor, we will explicitly release the
1497	 * forward allocated memory on dequeue
1498	 */
1499	sock_skb_set_dropcount(sk, skb);
1500
1501	__skb_queue_tail(list, skb);
1502	spin_unlock(&list->lock);
1503
1504	if (!sock_flag(sk, SOCK_DEAD))
1505		sk->sk_data_ready(sk);
1506
1507	busylock_release(busy);
1508	return 0;
1509
1510uncharge_drop:
1511	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1512
1513drop:
1514	atomic_inc(&sk->sk_drops);
1515	busylock_release(busy);
1516	return err;
1517}
1518EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1519
1520void udp_destruct_sock(struct sock *sk)
1521{
1522	/* reclaim completely the forward allocated memory */
1523	struct udp_sock *up = udp_sk(sk);
1524	unsigned int total = 0;
1525	struct sk_buff *skb;
1526
1527	skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1528	while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1529		total += skb->truesize;
1530		kfree_skb(skb);
1531	}
1532	udp_rmem_release(sk, total, 0, true);
 
 
1533
 
 
 
1534	inet_sock_destruct(sk);
1535}
1536EXPORT_SYMBOL_GPL(udp_destruct_sock);
1537
1538int udp_init_sock(struct sock *sk)
1539{
1540	skb_queue_head_init(&udp_sk(sk)->reader_queue);
1541	sk->sk_destruct = udp_destruct_sock;
 
1542	return 0;
1543}
1544EXPORT_SYMBOL_GPL(udp_init_sock);
1545
1546void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1547{
1548	if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1549		bool slow = lock_sock_fast(sk);
1550
1551		sk_peek_offset_bwd(sk, len);
1552		unlock_sock_fast(sk, slow);
1553	}
1554
1555	if (!skb_unref(skb))
1556		return;
1557
1558	/* In the more common cases we cleared the head states previously,
1559	 * see __udp_queue_rcv_skb().
1560	 */
1561	if (unlikely(udp_skb_has_head_state(skb)))
1562		skb_release_head_state(skb);
1563	__consume_stateless_skb(skb);
1564}
1565EXPORT_SYMBOL_GPL(skb_consume_udp);
1566
1567static struct sk_buff *__first_packet_length(struct sock *sk,
1568					     struct sk_buff_head *rcvq,
1569					     int *total)
1570{
1571	struct sk_buff *skb;
1572
1573	while ((skb = skb_peek(rcvq)) != NULL) {
1574		if (udp_lib_checksum_complete(skb)) {
1575			__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1576					IS_UDPLITE(sk));
1577			__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1578					IS_UDPLITE(sk));
1579			atomic_inc(&sk->sk_drops);
1580			__skb_unlink(skb, rcvq);
1581			*total += skb->truesize;
1582			kfree_skb(skb);
1583		} else {
1584			udp_skb_csum_unnecessary_set(skb);
1585			break;
1586		}
1587	}
1588	return skb;
1589}
1590
1591/**
1592 *	first_packet_length	- return length of first packet in receive queue
1593 *	@sk: socket
1594 *
1595 *	Drops all bad checksum frames, until a valid one is found.
1596 *	Returns the length of found skb, or -1 if none is found.
1597 */
1598static int first_packet_length(struct sock *sk)
1599{
1600	struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1601	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1602	struct sk_buff *skb;
1603	int total = 0;
1604	int res;
1605
1606	spin_lock_bh(&rcvq->lock);
1607	skb = __first_packet_length(sk, rcvq, &total);
1608	if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1609		spin_lock(&sk_queue->lock);
1610		skb_queue_splice_tail_init(sk_queue, rcvq);
1611		spin_unlock(&sk_queue->lock);
1612
1613		skb = __first_packet_length(sk, rcvq, &total);
1614	}
1615	res = skb ? skb->len : -1;
1616	if (total)
1617		udp_rmem_release(sk, total, 1, false);
1618	spin_unlock_bh(&rcvq->lock);
1619	return res;
1620}
1621
1622/*
1623 *	IOCTL requests applicable to the UDP protocol
1624 */
1625
1626int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1627{
1628	switch (cmd) {
1629	case SIOCOUTQ:
1630	{
1631		int amount = sk_wmem_alloc_get(sk);
1632
1633		return put_user(amount, (int __user *)arg);
1634	}
1635
1636	case SIOCINQ:
1637	{
1638		int amount = max_t(int, 0, first_packet_length(sk));
1639
1640		return put_user(amount, (int __user *)arg);
1641	}
1642
1643	default:
1644		return -ENOIOCTLCMD;
1645	}
1646
1647	return 0;
1648}
1649EXPORT_SYMBOL(udp_ioctl);
1650
1651struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1652			       int noblock, int *off, int *err)
1653{
1654	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1655	struct sk_buff_head *queue;
1656	struct sk_buff *last;
1657	long timeo;
1658	int error;
1659
1660	queue = &udp_sk(sk)->reader_queue;
1661	flags |= noblock ? MSG_DONTWAIT : 0;
1662	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1663	do {
1664		struct sk_buff *skb;
1665
1666		error = sock_error(sk);
1667		if (error)
1668			break;
1669
1670		error = -EAGAIN;
1671		do {
1672			spin_lock_bh(&queue->lock);
1673			skb = __skb_try_recv_from_queue(sk, queue, flags,
1674							udp_skb_destructor,
1675							off, err, &last);
1676			if (skb) {
 
 
1677				spin_unlock_bh(&queue->lock);
1678				return skb;
1679			}
1680
1681			if (skb_queue_empty_lockless(sk_queue)) {
1682				spin_unlock_bh(&queue->lock);
1683				goto busy_check;
1684			}
1685
1686			/* refill the reader queue and walk it again
1687			 * keep both queues locked to avoid re-acquiring
1688			 * the sk_receive_queue lock if fwd memory scheduling
1689			 * is needed.
1690			 */
1691			spin_lock(&sk_queue->lock);
1692			skb_queue_splice_tail_init(sk_queue, queue);
1693
1694			skb = __skb_try_recv_from_queue(sk, queue, flags,
1695							udp_skb_dtor_locked,
1696							off, err, &last);
 
1697			spin_unlock(&sk_queue->lock);
1698			spin_unlock_bh(&queue->lock);
1699			if (skb)
1700				return skb;
1701
1702busy_check:
1703			if (!sk_can_busy_loop(sk))
1704				break;
1705
1706			sk_busy_loop(sk, flags & MSG_DONTWAIT);
1707		} while (!skb_queue_empty_lockless(sk_queue));
1708
1709		/* sk_queue is empty, reader_queue may contain peeked packets */
1710	} while (timeo &&
1711		 !__skb_wait_for_more_packets(sk, &error, &timeo,
 
1712					      (struct sk_buff *)sk_queue));
1713
1714	*err = error;
1715	return NULL;
1716}
1717EXPORT_SYMBOL(__skb_recv_udp);
1718
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1719/*
1720 * 	This should be easy, if there is something there we
1721 * 	return it, otherwise we block.
1722 */
1723
1724int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1725		int flags, int *addr_len)
1726{
1727	struct inet_sock *inet = inet_sk(sk);
1728	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1729	struct sk_buff *skb;
1730	unsigned int ulen, copied;
1731	int off, err, peeking = flags & MSG_PEEK;
1732	int is_udplite = IS_UDPLITE(sk);
1733	bool checksum_valid = false;
1734
1735	if (flags & MSG_ERRQUEUE)
1736		return ip_recv_error(sk, msg, len, addr_len);
1737
1738try_again:
1739	off = sk_peek_offset(sk, flags);
1740	skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1741	if (!skb)
1742		return err;
1743
1744	ulen = udp_skb_len(skb);
1745	copied = len;
1746	if (copied > ulen - off)
1747		copied = ulen - off;
1748	else if (copied < ulen)
1749		msg->msg_flags |= MSG_TRUNC;
1750
1751	/*
1752	 * If checksum is needed at all, try to do it while copying the
1753	 * data.  If the data is truncated, or if we only want a partial
1754	 * coverage checksum (UDP-Lite), do it before the copy.
1755	 */
1756
1757	if (copied < ulen || peeking ||
1758	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1759		checksum_valid = udp_skb_csum_unnecessary(skb) ||
1760				!__udp_lib_checksum_complete(skb);
1761		if (!checksum_valid)
1762			goto csum_copy_err;
1763	}
1764
1765	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1766		if (udp_skb_is_linear(skb))
1767			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1768		else
1769			err = skb_copy_datagram_msg(skb, off, msg, copied);
1770	} else {
1771		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1772
1773		if (err == -EINVAL)
1774			goto csum_copy_err;
1775	}
1776
1777	if (unlikely(err)) {
1778		if (!peeking) {
1779			atomic_inc(&sk->sk_drops);
1780			UDP_INC_STATS(sock_net(sk),
1781				      UDP_MIB_INERRORS, is_udplite);
1782		}
1783		kfree_skb(skb);
1784		return err;
1785	}
1786
1787	if (!peeking)
1788		UDP_INC_STATS(sock_net(sk),
1789			      UDP_MIB_INDATAGRAMS, is_udplite);
1790
1791	sock_recv_ts_and_drops(msg, sk, skb);
1792
1793	/* Copy the address. */
1794	if (sin) {
1795		sin->sin_family = AF_INET;
1796		sin->sin_port = udp_hdr(skb)->source;
1797		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1798		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1799		*addr_len = sizeof(*sin);
1800
1801		if (cgroup_bpf_enabled)
1802			BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1803							(struct sockaddr *)sin);
1804	}
1805
1806	if (udp_sk(sk)->gro_enabled)
1807		udp_cmsg_recv(msg, sk, skb);
1808
1809	if (inet->cmsg_flags)
1810		ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1811
1812	err = copied;
1813	if (flags & MSG_TRUNC)
1814		err = ulen;
1815
1816	skb_consume_udp(sk, skb, peeking ? -err : err);
1817	return err;
1818
1819csum_copy_err:
1820	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1821				 udp_skb_destructor)) {
1822		UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1823		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1824	}
1825	kfree_skb(skb);
1826
1827	/* starting over for a new packet, but check if we need to yield */
1828	cond_resched();
1829	msg->msg_flags &= ~MSG_TRUNC;
1830	goto try_again;
1831}
1832
1833int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1834{
1835	/* This check is replicated from __ip4_datagram_connect() and
1836	 * intended to prevent BPF program called below from accessing bytes
1837	 * that are out of the bound specified by user in addr_len.
1838	 */
1839	if (addr_len < sizeof(struct sockaddr_in))
1840		return -EINVAL;
1841
1842	return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1843}
1844EXPORT_SYMBOL(udp_pre_connect);
1845
1846int __udp_disconnect(struct sock *sk, int flags)
1847{
1848	struct inet_sock *inet = inet_sk(sk);
1849	/*
1850	 *	1003.1g - break association.
1851	 */
1852
1853	sk->sk_state = TCP_CLOSE;
1854	inet->inet_daddr = 0;
1855	inet->inet_dport = 0;
1856	sock_rps_reset_rxhash(sk);
1857	sk->sk_bound_dev_if = 0;
1858	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1859		inet_reset_saddr(sk);
 
 
 
 
1860
1861	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1862		sk->sk_prot->unhash(sk);
1863		inet->inet_sport = 0;
1864	}
1865	sk_dst_reset(sk);
1866	return 0;
1867}
1868EXPORT_SYMBOL(__udp_disconnect);
1869
1870int udp_disconnect(struct sock *sk, int flags)
1871{
1872	lock_sock(sk);
1873	__udp_disconnect(sk, flags);
1874	release_sock(sk);
1875	return 0;
1876}
1877EXPORT_SYMBOL(udp_disconnect);
1878
1879void udp_lib_unhash(struct sock *sk)
1880{
1881	if (sk_hashed(sk)) {
1882		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1883		struct udp_hslot *hslot, *hslot2;
1884
1885		hslot  = udp_hashslot(udptable, sock_net(sk),
1886				      udp_sk(sk)->udp_port_hash);
1887		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1888
1889		spin_lock_bh(&hslot->lock);
1890		if (rcu_access_pointer(sk->sk_reuseport_cb))
1891			reuseport_detach_sock(sk);
1892		if (sk_del_node_init_rcu(sk)) {
1893			hslot->count--;
1894			inet_sk(sk)->inet_num = 0;
1895			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1896
1897			spin_lock(&hslot2->lock);
1898			hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1899			hslot2->count--;
1900			spin_unlock(&hslot2->lock);
1901		}
1902		spin_unlock_bh(&hslot->lock);
1903	}
1904}
1905EXPORT_SYMBOL(udp_lib_unhash);
1906
1907/*
1908 * inet_rcv_saddr was changed, we must rehash secondary hash
1909 */
1910void udp_lib_rehash(struct sock *sk, u16 newhash)
1911{
1912	if (sk_hashed(sk)) {
1913		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1914		struct udp_hslot *hslot, *hslot2, *nhslot2;
1915
1916		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1917		nhslot2 = udp_hashslot2(udptable, newhash);
1918		udp_sk(sk)->udp_portaddr_hash = newhash;
1919
1920		if (hslot2 != nhslot2 ||
1921		    rcu_access_pointer(sk->sk_reuseport_cb)) {
1922			hslot = udp_hashslot(udptable, sock_net(sk),
1923					     udp_sk(sk)->udp_port_hash);
1924			/* we must lock primary chain too */
1925			spin_lock_bh(&hslot->lock);
1926			if (rcu_access_pointer(sk->sk_reuseport_cb))
1927				reuseport_detach_sock(sk);
1928
1929			if (hslot2 != nhslot2) {
1930				spin_lock(&hslot2->lock);
1931				hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1932				hslot2->count--;
1933				spin_unlock(&hslot2->lock);
1934
1935				spin_lock(&nhslot2->lock);
1936				hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1937							 &nhslot2->head);
1938				nhslot2->count++;
1939				spin_unlock(&nhslot2->lock);
1940			}
1941
1942			spin_unlock_bh(&hslot->lock);
1943		}
1944	}
1945}
1946EXPORT_SYMBOL(udp_lib_rehash);
1947
1948void udp_v4_rehash(struct sock *sk)
1949{
1950	u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1951					  inet_sk(sk)->inet_rcv_saddr,
1952					  inet_sk(sk)->inet_num);
1953	udp_lib_rehash(sk, new_hash);
1954}
1955
1956static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1957{
1958	int rc;
1959
1960	if (inet_sk(sk)->inet_daddr) {
1961		sock_rps_save_rxhash(sk, skb);
1962		sk_mark_napi_id(sk, skb);
1963		sk_incoming_cpu_update(sk);
1964	} else {
1965		sk_mark_napi_id_once(sk, skb);
1966	}
1967
1968	rc = __udp_enqueue_schedule_skb(sk, skb);
1969	if (rc < 0) {
1970		int is_udplite = IS_UDPLITE(sk);
 
1971
1972		/* Note that an ENOMEM error is charged twice */
1973		if (rc == -ENOMEM)
1974			UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1975					is_udplite);
 
 
 
 
 
 
1976		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1977		kfree_skb(skb);
1978		trace_udp_fail_queue_rcv_skb(rc, sk);
1979		return -1;
1980	}
1981
1982	return 0;
1983}
1984
1985/* returns:
1986 *  -1: error
1987 *   0: success
1988 *  >0: "udp encap" protocol resubmission
1989 *
1990 * Note that in the success and error cases, the skb is assumed to
1991 * have either been requeued or freed.
1992 */
1993static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
1994{
 
1995	struct udp_sock *up = udp_sk(sk);
1996	int is_udplite = IS_UDPLITE(sk);
1997
1998	/*
1999	 *	Charge it to the socket, dropping if the queue is full.
2000	 */
2001	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 
2002		goto drop;
 
2003	nf_reset_ct(skb);
2004
2005	if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
 
2006		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2007
2008		/*
2009		 * This is an encapsulation socket so pass the skb to
2010		 * the socket's udp_encap_rcv() hook. Otherwise, just
2011		 * fall through and pass this up the UDP socket.
2012		 * up->encap_rcv() returns the following value:
2013		 * =0 if skb was successfully passed to the encap
2014		 *    handler or was discarded by it.
2015		 * >0 if skb should be passed on to UDP.
2016		 * <0 if skb should be resubmitted as proto -N
2017		 */
2018
2019		/* if we're overly short, let UDP handle it */
2020		encap_rcv = READ_ONCE(up->encap_rcv);
2021		if (encap_rcv) {
2022			int ret;
2023
2024			/* Verify checksum before giving to encap */
2025			if (udp_lib_checksum_complete(skb))
2026				goto csum_error;
2027
2028			ret = encap_rcv(sk, skb);
2029			if (ret <= 0) {
2030				__UDP_INC_STATS(sock_net(sk),
2031						UDP_MIB_INDATAGRAMS,
2032						is_udplite);
2033				return -ret;
2034			}
2035		}
2036
2037		/* FALLTHROUGH -- it's a UDP Packet */
2038	}
2039
2040	/*
2041	 * 	UDP-Lite specific tests, ignored on UDP sockets
2042	 */
2043	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
 
2044
2045		/*
2046		 * MIB statistics other than incrementing the error count are
2047		 * disabled for the following two types of errors: these depend
2048		 * on the application settings, not on the functioning of the
2049		 * protocol stack as such.
2050		 *
2051		 * RFC 3828 here recommends (sec 3.3): "There should also be a
2052		 * way ... to ... at least let the receiving application block
2053		 * delivery of packets with coverage values less than a value
2054		 * provided by the application."
2055		 */
2056		if (up->pcrlen == 0) {          /* full coverage was set  */
2057			net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2058					    UDP_SKB_CB(skb)->cscov, skb->len);
2059			goto drop;
2060		}
2061		/* The next case involves violating the min. coverage requested
2062		 * by the receiver. This is subtle: if receiver wants x and x is
2063		 * greater than the buffersize/MTU then receiver will complain
2064		 * that it wants x while sender emits packets of smaller size y.
2065		 * Therefore the above ...()->partial_cov statement is essential.
2066		 */
2067		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
2068			net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2069					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
2070			goto drop;
2071		}
2072	}
2073
2074	prefetch(&sk->sk_rmem_alloc);
2075	if (rcu_access_pointer(sk->sk_filter) &&
2076	    udp_lib_checksum_complete(skb))
2077			goto csum_error;
2078
2079	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
 
2080		goto drop;
 
2081
2082	udp_csum_pull_header(skb);
2083
2084	ipv4_pktinfo_prepare(sk, skb);
2085	return __udp_queue_rcv_skb(sk, skb);
2086
2087csum_error:
 
2088	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2089drop:
2090	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2091	atomic_inc(&sk->sk_drops);
2092	kfree_skb(skb);
2093	return -1;
2094}
2095
2096static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2097{
2098	struct sk_buff *next, *segs;
2099	int ret;
2100
2101	if (likely(!udp_unexpected_gso(sk, skb)))
2102		return udp_queue_rcv_one_skb(sk, skb);
2103
2104	BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET);
2105	__skb_push(skb, -skb_mac_offset(skb));
2106	segs = udp_rcv_segment(sk, skb, true);
2107	for (skb = segs; skb; skb = next) {
2108		next = skb->next;
2109		__skb_pull(skb, skb_transport_offset(skb));
 
 
2110		ret = udp_queue_rcv_one_skb(sk, skb);
2111		if (ret > 0)
2112			ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret);
2113	}
2114	return 0;
2115}
2116
2117/* For TCP sockets, sk_rx_dst is protected by socket lock
2118 * For UDP, we use xchg() to guard against concurrent changes.
2119 */
2120bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2121{
2122	struct dst_entry *old;
2123
2124	if (dst_hold_safe(dst)) {
2125		old = xchg(&sk->sk_rx_dst, dst);
2126		dst_release(old);
2127		return old != dst;
2128	}
2129	return false;
2130}
2131EXPORT_SYMBOL(udp_sk_rx_dst_set);
2132
2133/*
2134 *	Multicasts and broadcasts go to each listener.
2135 *
2136 *	Note: called only from the BH handler context.
2137 */
2138static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2139				    struct udphdr  *uh,
2140				    __be32 saddr, __be32 daddr,
2141				    struct udp_table *udptable,
2142				    int proto)
2143{
2144	struct sock *sk, *first = NULL;
2145	unsigned short hnum = ntohs(uh->dest);
2146	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2147	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2148	unsigned int offset = offsetof(typeof(*sk), sk_node);
2149	int dif = skb->dev->ifindex;
2150	int sdif = inet_sdif(skb);
2151	struct hlist_node *node;
2152	struct sk_buff *nskb;
2153
2154	if (use_hash2) {
2155		hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2156			    udptable->mask;
2157		hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2158start_lookup:
2159		hslot = &udptable->hash2[hash2];
2160		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2161	}
2162
2163	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2164		if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2165					 uh->source, saddr, dif, sdif, hnum))
2166			continue;
2167
2168		if (!first) {
2169			first = sk;
2170			continue;
2171		}
2172		nskb = skb_clone(skb, GFP_ATOMIC);
2173
2174		if (unlikely(!nskb)) {
2175			atomic_inc(&sk->sk_drops);
2176			__UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2177					IS_UDPLITE(sk));
2178			__UDP_INC_STATS(net, UDP_MIB_INERRORS,
2179					IS_UDPLITE(sk));
2180			continue;
2181		}
2182		if (udp_queue_rcv_skb(sk, nskb) > 0)
2183			consume_skb(nskb);
2184	}
2185
2186	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
2187	if (use_hash2 && hash2 != hash2_any) {
2188		hash2 = hash2_any;
2189		goto start_lookup;
2190	}
2191
2192	if (first) {
2193		if (udp_queue_rcv_skb(first, skb) > 0)
2194			consume_skb(skb);
2195	} else {
2196		kfree_skb(skb);
2197		__UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2198				proto == IPPROTO_UDPLITE);
2199	}
2200	return 0;
2201}
2202
2203/* Initialize UDP checksum. If exited with zero value (success),
2204 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2205 * Otherwise, csum completion requires checksumming packet body,
2206 * including udp header and folding it to skb->csum.
2207 */
2208static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2209				 int proto)
2210{
2211	int err;
2212
2213	UDP_SKB_CB(skb)->partial_cov = 0;
2214	UDP_SKB_CB(skb)->cscov = skb->len;
2215
2216	if (proto == IPPROTO_UDPLITE) {
2217		err = udplite_checksum_init(skb, uh);
2218		if (err)
2219			return err;
2220
2221		if (UDP_SKB_CB(skb)->partial_cov) {
2222			skb->csum = inet_compute_pseudo(skb, proto);
2223			return 0;
2224		}
2225	}
2226
2227	/* Note, we are only interested in != 0 or == 0, thus the
2228	 * force to int.
2229	 */
2230	err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2231							inet_compute_pseudo);
2232	if (err)
2233		return err;
2234
2235	if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2236		/* If SW calculated the value, we know it's bad */
2237		if (skb->csum_complete_sw)
2238			return 1;
2239
2240		/* HW says the value is bad. Let's validate that.
2241		 * skb->csum is no longer the full packet checksum,
2242		 * so don't treat it as such.
2243		 */
2244		skb_checksum_complete_unset(skb);
2245	}
2246
2247	return 0;
2248}
2249
2250/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2251 * return code conversion for ip layer consumption
2252 */
2253static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2254			       struct udphdr *uh)
2255{
2256	int ret;
2257
2258	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2259		skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2260
2261	ret = udp_queue_rcv_skb(sk, skb);
2262
2263	/* a return value > 0 means to resubmit the input, but
2264	 * it wants the return to be -protocol, or 0
2265	 */
2266	if (ret > 0)
2267		return -ret;
2268	return 0;
2269}
2270
2271/*
2272 *	All we need to do is get the socket, and then do a checksum.
2273 */
2274
2275int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2276		   int proto)
2277{
2278	struct sock *sk;
2279	struct udphdr *uh;
2280	unsigned short ulen;
2281	struct rtable *rt = skb_rtable(skb);
2282	__be32 saddr, daddr;
2283	struct net *net = dev_net(skb->dev);
 
 
 
 
2284
2285	/*
2286	 *  Validate the packet.
2287	 */
2288	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2289		goto drop;		/* No space for header. */
2290
2291	uh   = udp_hdr(skb);
2292	ulen = ntohs(uh->len);
2293	saddr = ip_hdr(skb)->saddr;
2294	daddr = ip_hdr(skb)->daddr;
2295
2296	if (ulen > skb->len)
2297		goto short_packet;
2298
2299	if (proto == IPPROTO_UDP) {
2300		/* UDP validates ulen. */
2301		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2302			goto short_packet;
2303		uh = udp_hdr(skb);
2304	}
2305
2306	if (udp4_csum_init(skb, uh, proto))
2307		goto csum_error;
2308
2309	sk = skb_steal_sock(skb);
 
 
 
 
2310	if (sk) {
2311		struct dst_entry *dst = skb_dst(skb);
2312		int ret;
2313
2314		if (unlikely(sk->sk_rx_dst != dst))
2315			udp_sk_rx_dst_set(sk, dst);
2316
2317		ret = udp_unicast_rcv_skb(sk, skb, uh);
2318		sock_put(sk);
 
2319		return ret;
2320	}
2321
2322	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2323		return __udp4_lib_mcast_deliver(net, skb, uh,
2324						saddr, daddr, udptable, proto);
2325
2326	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2327	if (sk)
2328		return udp_unicast_rcv_skb(sk, skb, uh);
2329
2330	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2331		goto drop;
2332	nf_reset_ct(skb);
2333
2334	/* No socket. Drop packet silently, if checksum is wrong */
2335	if (udp_lib_checksum_complete(skb))
2336		goto csum_error;
2337
 
2338	__UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2339	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2340
2341	/*
2342	 * Hmm.  We got an UDP packet to a port to which we
2343	 * don't wanna listen.  Ignore it.
2344	 */
2345	kfree_skb(skb);
2346	return 0;
2347
2348short_packet:
 
2349	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2350			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2351			    &saddr, ntohs(uh->source),
2352			    ulen, skb->len,
2353			    &daddr, ntohs(uh->dest));
2354	goto drop;
2355
2356csum_error:
2357	/*
2358	 * RFC1122: OK.  Discards the bad packet silently (as far as
2359	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2360	 */
 
2361	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2362			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2363			    &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2364			    ulen);
2365	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2366drop:
2367	__UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2368	kfree_skb(skb);
2369	return 0;
2370}
2371
2372/* We can only early demux multicast if there is a single matching socket.
2373 * If more than one socket found returns NULL
2374 */
2375static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2376						  __be16 loc_port, __be32 loc_addr,
2377						  __be16 rmt_port, __be32 rmt_addr,
2378						  int dif, int sdif)
2379{
2380	struct sock *sk, *result;
2381	unsigned short hnum = ntohs(loc_port);
2382	unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2383	struct udp_hslot *hslot = &udp_table.hash[slot];
 
 
 
 
2384
2385	/* Do not bother scanning a too big list */
2386	if (hslot->count > 10)
2387		return NULL;
2388
2389	result = NULL;
2390	sk_for_each_rcu(sk, &hslot->head) {
2391		if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2392					rmt_port, rmt_addr, dif, sdif, hnum)) {
2393			if (result)
2394				return NULL;
2395			result = sk;
2396		}
2397	}
2398
2399	return result;
2400}
2401
2402/* For unicast we should only early demux connected sockets or we can
2403 * break forwarding setups.  The chains here can be long so only check
2404 * if the first socket is an exact match and if not move on.
2405 */
2406static struct sock *__udp4_lib_demux_lookup(struct net *net,
2407					    __be16 loc_port, __be32 loc_addr,
2408					    __be16 rmt_port, __be32 rmt_addr,
2409					    int dif, int sdif)
2410{
2411	unsigned short hnum = ntohs(loc_port);
2412	unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2413	unsigned int slot2 = hash2 & udp_table.mask;
2414	struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2415	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2416	const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
 
 
 
2417	struct sock *sk;
2418
 
 
 
 
 
2419	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2420		if (INET_MATCH(sk, net, acookie, rmt_addr,
2421			       loc_addr, ports, dif, sdif))
2422			return sk;
2423		/* Only check first socket in chain */
2424		break;
2425	}
2426	return NULL;
2427}
2428
2429int udp_v4_early_demux(struct sk_buff *skb)
2430{
2431	struct net *net = dev_net(skb->dev);
2432	struct in_device *in_dev = NULL;
2433	const struct iphdr *iph;
2434	const struct udphdr *uh;
2435	struct sock *sk = NULL;
2436	struct dst_entry *dst;
2437	int dif = skb->dev->ifindex;
2438	int sdif = inet_sdif(skb);
2439	int ours;
2440
2441	/* validate the packet */
2442	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2443		return 0;
2444
2445	iph = ip_hdr(skb);
2446	uh = udp_hdr(skb);
2447
2448	if (skb->pkt_type == PACKET_MULTICAST) {
2449		in_dev = __in_dev_get_rcu(skb->dev);
2450
2451		if (!in_dev)
2452			return 0;
2453
2454		ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2455				       iph->protocol);
2456		if (!ours)
2457			return 0;
2458
2459		sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2460						   uh->source, iph->saddr,
2461						   dif, sdif);
2462	} else if (skb->pkt_type == PACKET_HOST) {
2463		sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2464					     uh->source, iph->saddr, dif, sdif);
2465	}
2466
2467	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2468		return 0;
2469
2470	skb->sk = sk;
2471	skb->destructor = sock_efree;
2472	dst = READ_ONCE(sk->sk_rx_dst);
2473
2474	if (dst)
2475		dst = dst_check(dst, 0);
2476	if (dst) {
2477		u32 itag = 0;
2478
2479		/* set noref for now.
2480		 * any place which wants to hold dst has to call
2481		 * dst_hold_safe()
2482		 */
2483		skb_dst_set_noref(skb, dst);
2484
2485		/* for unconnected multicast sockets we need to validate
2486		 * the source on each packet
2487		 */
2488		if (!inet_sk(sk)->inet_daddr && in_dev)
2489			return ip_mc_validate_source(skb, iph->daddr,
2490						     iph->saddr, iph->tos,
 
2491						     skb->dev, in_dev, &itag);
2492	}
2493	return 0;
2494}
2495
2496int udp_rcv(struct sk_buff *skb)
2497{
2498	return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2499}
2500
2501void udp_destroy_sock(struct sock *sk)
2502{
2503	struct udp_sock *up = udp_sk(sk);
2504	bool slow = lock_sock_fast(sk);
 
 
 
2505	udp_flush_pending_frames(sk);
2506	unlock_sock_fast(sk, slow);
2507	if (static_branch_unlikely(&udp_encap_needed_key)) {
2508		if (up->encap_type) {
2509			void (*encap_destroy)(struct sock *sk);
2510			encap_destroy = READ_ONCE(up->encap_destroy);
2511			if (encap_destroy)
2512				encap_destroy(sk);
2513		}
2514		if (up->encap_enabled)
2515			static_branch_dec(&udp_encap_needed_key);
2516	}
2517}
2518
 
 
 
 
 
 
 
 
 
 
 
 
 
2519/*
2520 *	Socket option code for UDP
2521 */
2522int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2523		       char __user *optval, unsigned int optlen,
2524		       int (*push_pending_frames)(struct sock *))
2525{
2526	struct udp_sock *up = udp_sk(sk);
2527	int val, valbool;
2528	int err = 0;
2529	int is_udplite = IS_UDPLITE(sk);
2530
 
 
 
 
 
 
 
 
 
 
 
 
2531	if (optlen < sizeof(int))
2532		return -EINVAL;
2533
2534	if (get_user(val, (int __user *)optval))
2535		return -EFAULT;
2536
2537	valbool = val ? 1 : 0;
2538
2539	switch (optname) {
2540	case UDP_CORK:
2541		if (val != 0) {
2542			up->corkflag = 1;
2543		} else {
2544			up->corkflag = 0;
2545			lock_sock(sk);
2546			push_pending_frames(sk);
2547			release_sock(sk);
2548		}
2549		break;
2550
2551	case UDP_ENCAP:
2552		switch (val) {
2553		case 0:
 
2554		case UDP_ENCAP_ESPINUDP:
 
 
2555		case UDP_ENCAP_ESPINUDP_NON_IKE:
2556			up->encap_rcv = xfrm4_udp_encap_rcv;
2557			/* FALLTHROUGH */
 
 
 
 
 
 
 
 
2558		case UDP_ENCAP_L2TPINUDP:
2559			up->encap_type = val;
2560			lock_sock(sk);
2561			udp_tunnel_encap_enable(sk->sk_socket);
2562			release_sock(sk);
2563			break;
2564		default:
2565			err = -ENOPROTOOPT;
2566			break;
2567		}
2568		break;
2569
2570	case UDP_NO_CHECK6_TX:
2571		up->no_check6_tx = valbool;
2572		break;
2573
2574	case UDP_NO_CHECK6_RX:
2575		up->no_check6_rx = valbool;
2576		break;
2577
2578	case UDP_SEGMENT:
2579		if (val < 0 || val > USHRT_MAX)
2580			return -EINVAL;
2581		up->gso_size = val;
2582		break;
2583
2584	case UDP_GRO:
2585		lock_sock(sk);
 
2586		if (valbool)
2587			udp_tunnel_encap_enable(sk->sk_socket);
2588		up->gro_enabled = valbool;
2589		release_sock(sk);
 
2590		break;
2591
2592	/*
2593	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
2594	 */
2595	/* The sender sets actual checksum coverage length via this option.
2596	 * The case coverage > packet length is handled by send module. */
2597	case UDPLITE_SEND_CSCOV:
2598		if (!is_udplite)         /* Disable the option on UDP sockets */
2599			return -ENOPROTOOPT;
2600		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2601			val = 8;
2602		else if (val > USHRT_MAX)
2603			val = USHRT_MAX;
2604		up->pcslen = val;
2605		up->pcflag |= UDPLITE_SEND_CC;
2606		break;
2607
2608	/* The receiver specifies a minimum checksum coverage value. To make
2609	 * sense, this should be set to at least 8 (as done below). If zero is
2610	 * used, this again means full checksum coverage.                     */
2611	case UDPLITE_RECV_CSCOV:
2612		if (!is_udplite)         /* Disable the option on UDP sockets */
2613			return -ENOPROTOOPT;
2614		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2615			val = 8;
2616		else if (val > USHRT_MAX)
2617			val = USHRT_MAX;
2618		up->pcrlen = val;
2619		up->pcflag |= UDPLITE_RECV_CC;
2620		break;
2621
2622	default:
2623		err = -ENOPROTOOPT;
2624		break;
2625	}
2626
2627	return err;
2628}
2629EXPORT_SYMBOL(udp_lib_setsockopt);
2630
2631int udp_setsockopt(struct sock *sk, int level, int optname,
2632		   char __user *optval, unsigned int optlen)
2633{
2634	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2635		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
 
2636					  udp_push_pending_frames);
2637	return ip_setsockopt(sk, level, optname, optval, optlen);
2638}
2639
2640#ifdef CONFIG_COMPAT
2641int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2642			  char __user *optval, unsigned int optlen)
2643{
2644	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2645		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2646					  udp_push_pending_frames);
2647	return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2648}
2649#endif
2650
2651int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2652		       char __user *optval, int __user *optlen)
2653{
2654	struct udp_sock *up = udp_sk(sk);
2655	int val, len;
2656
2657	if (get_user(len, optlen))
2658		return -EFAULT;
2659
2660	len = min_t(unsigned int, len, sizeof(int));
2661
2662	if (len < 0)
2663		return -EINVAL;
2664
2665	switch (optname) {
2666	case UDP_CORK:
2667		val = up->corkflag;
2668		break;
2669
2670	case UDP_ENCAP:
2671		val = up->encap_type;
2672		break;
2673
2674	case UDP_NO_CHECK6_TX:
2675		val = up->no_check6_tx;
2676		break;
2677
2678	case UDP_NO_CHECK6_RX:
2679		val = up->no_check6_rx;
2680		break;
2681
2682	case UDP_SEGMENT:
2683		val = up->gso_size;
 
 
 
 
2684		break;
2685
2686	/* The following two cannot be changed on UDP sockets, the return is
2687	 * always 0 (which corresponds to the full checksum coverage of UDP). */
2688	case UDPLITE_SEND_CSCOV:
2689		val = up->pcslen;
2690		break;
2691
2692	case UDPLITE_RECV_CSCOV:
2693		val = up->pcrlen;
2694		break;
2695
2696	default:
2697		return -ENOPROTOOPT;
2698	}
2699
2700	if (put_user(len, optlen))
2701		return -EFAULT;
2702	if (copy_to_user(optval, &val, len))
2703		return -EFAULT;
2704	return 0;
2705}
2706EXPORT_SYMBOL(udp_lib_getsockopt);
2707
2708int udp_getsockopt(struct sock *sk, int level, int optname,
2709		   char __user *optval, int __user *optlen)
2710{
2711	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2712		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2713	return ip_getsockopt(sk, level, optname, optval, optlen);
2714}
2715
2716#ifdef CONFIG_COMPAT
2717int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2718				 char __user *optval, int __user *optlen)
2719{
2720	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2721		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2722	return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2723}
2724#endif
2725/**
2726 * 	udp_poll - wait for a UDP event.
2727 *	@file - file struct
2728 *	@sock - socket
2729 *	@wait - poll table
2730 *
2731 *	This is same as datagram poll, except for the special case of
2732 *	blocking sockets. If application is using a blocking fd
2733 *	and a packet with checksum error is in the queue;
2734 *	then it could get return from select indicating data available
2735 *	but then block when reading it. Add special case code
2736 *	to work around these arguably broken applications.
2737 */
2738__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2739{
2740	__poll_t mask = datagram_poll(file, sock, wait);
2741	struct sock *sk = sock->sk;
2742
2743	if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2744		mask |= EPOLLIN | EPOLLRDNORM;
2745
2746	/* Check for false positives due to checksum errors */
2747	if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2748	    !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2749		mask &= ~(EPOLLIN | EPOLLRDNORM);
2750
 
 
 
2751	return mask;
2752
2753}
2754EXPORT_SYMBOL(udp_poll);
2755
2756int udp_abort(struct sock *sk, int err)
2757{
2758	lock_sock(sk);
 
 
 
 
 
 
 
2759
2760	sk->sk_err = err;
2761	sk->sk_error_report(sk);
2762	__udp_disconnect(sk, 0);
2763
2764	release_sock(sk);
 
 
2765
2766	return 0;
2767}
2768EXPORT_SYMBOL_GPL(udp_abort);
2769
2770struct proto udp_prot = {
2771	.name			= "UDP",
2772	.owner			= THIS_MODULE,
2773	.close			= udp_lib_close,
2774	.pre_connect		= udp_pre_connect,
2775	.connect		= ip4_datagram_connect,
2776	.disconnect		= udp_disconnect,
2777	.ioctl			= udp_ioctl,
2778	.init			= udp_init_sock,
2779	.destroy		= udp_destroy_sock,
2780	.setsockopt		= udp_setsockopt,
2781	.getsockopt		= udp_getsockopt,
2782	.sendmsg		= udp_sendmsg,
2783	.recvmsg		= udp_recvmsg,
2784	.sendpage		= udp_sendpage,
2785	.release_cb		= ip4_datagram_release_cb,
2786	.hash			= udp_lib_hash,
2787	.unhash			= udp_lib_unhash,
2788	.rehash			= udp_v4_rehash,
2789	.get_port		= udp_v4_get_port,
 
 
 
 
2790	.memory_allocated	= &udp_memory_allocated,
 
 
2791	.sysctl_mem		= sysctl_udp_mem,
2792	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2793	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2794	.obj_size		= sizeof(struct udp_sock),
2795	.h.udp_table		= &udp_table,
2796#ifdef CONFIG_COMPAT
2797	.compat_setsockopt	= compat_udp_setsockopt,
2798	.compat_getsockopt	= compat_udp_getsockopt,
2799#endif
2800	.diag_destroy		= udp_abort,
2801};
2802EXPORT_SYMBOL(udp_prot);
2803
2804/* ------------------------------------------------------------------------ */
2805#ifdef CONFIG_PROC_FS
2806
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2807static struct sock *udp_get_first(struct seq_file *seq, int start)
2808{
2809	struct sock *sk;
2810	struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2811	struct udp_iter_state *state = seq->private;
2812	struct net *net = seq_file_net(seq);
 
 
 
 
2813
2814	for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2815	     ++state->bucket) {
2816		struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2817
2818		if (hlist_empty(&hslot->head))
2819			continue;
2820
2821		spin_lock_bh(&hslot->lock);
2822		sk_for_each(sk, &hslot->head) {
2823			if (!net_eq(sock_net(sk), net))
2824				continue;
2825			if (sk->sk_family == afinfo->family)
2826				goto found;
2827		}
2828		spin_unlock_bh(&hslot->lock);
2829	}
2830	sk = NULL;
2831found:
2832	return sk;
2833}
2834
2835static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2836{
2837	struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2838	struct udp_iter_state *state = seq->private;
2839	struct net *net = seq_file_net(seq);
 
2840
2841	do {
2842		sk = sk_next(sk);
2843	} while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2844
2845	if (!sk) {
2846		if (state->bucket <= afinfo->udp_table->mask)
2847			spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
 
 
 
2848		return udp_get_first(seq, state->bucket + 1);
2849	}
2850	return sk;
2851}
2852
2853static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2854{
2855	struct sock *sk = udp_get_first(seq, 0);
2856
2857	if (sk)
2858		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2859			--pos;
2860	return pos ? NULL : sk;
2861}
2862
2863void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2864{
2865	struct udp_iter_state *state = seq->private;
2866	state->bucket = MAX_UDP_PORTS;
2867
2868	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2869}
2870EXPORT_SYMBOL(udp_seq_start);
2871
2872void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2873{
2874	struct sock *sk;
2875
2876	if (v == SEQ_START_TOKEN)
2877		sk = udp_get_idx(seq, 0);
2878	else
2879		sk = udp_get_next(seq, v);
2880
2881	++*pos;
2882	return sk;
2883}
2884EXPORT_SYMBOL(udp_seq_next);
2885
2886void udp_seq_stop(struct seq_file *seq, void *v)
2887{
2888	struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2889	struct udp_iter_state *state = seq->private;
 
 
 
2890
2891	if (state->bucket <= afinfo->udp_table->mask)
2892		spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2893}
2894EXPORT_SYMBOL(udp_seq_stop);
2895
2896/* ------------------------------------------------------------------------ */
2897static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2898		int bucket)
2899{
2900	struct inet_sock *inet = inet_sk(sp);
2901	__be32 dest = inet->inet_daddr;
2902	__be32 src  = inet->inet_rcv_saddr;
2903	__u16 destp	  = ntohs(inet->inet_dport);
2904	__u16 srcp	  = ntohs(inet->inet_sport);
2905
2906	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2907		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
2908		bucket, src, srcp, dest, destp, sp->sk_state,
2909		sk_wmem_alloc_get(sp),
2910		udp_rqueue_get(sp),
2911		0, 0L, 0,
2912		from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2913		0, sock_i_ino(sp),
2914		refcount_read(&sp->sk_refcnt), sp,
2915		atomic_read(&sp->sk_drops));
2916}
2917
2918int udp4_seq_show(struct seq_file *seq, void *v)
2919{
2920	seq_setwidth(seq, 127);
2921	if (v == SEQ_START_TOKEN)
2922		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2923			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2924			   "inode ref pointer drops");
2925	else {
2926		struct udp_iter_state *state = seq->private;
2927
2928		udp4_format_sock(v, seq, state->bucket);
2929	}
2930	seq_pad(seq, '\n');
2931	return 0;
2932}
2933
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2934const struct seq_operations udp_seq_ops = {
2935	.start		= udp_seq_start,
2936	.next		= udp_seq_next,
2937	.stop		= udp_seq_stop,
2938	.show		= udp4_seq_show,
2939};
2940EXPORT_SYMBOL(udp_seq_ops);
2941
2942static struct udp_seq_afinfo udp4_seq_afinfo = {
2943	.family		= AF_INET,
2944	.udp_table	= &udp_table,
2945};
2946
2947static int __net_init udp4_proc_init_net(struct net *net)
2948{
2949	if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2950			sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2951		return -ENOMEM;
2952	return 0;
2953}
2954
2955static void __net_exit udp4_proc_exit_net(struct net *net)
2956{
2957	remove_proc_entry("udp", net->proc_net);
2958}
2959
2960static struct pernet_operations udp4_net_ops = {
2961	.init = udp4_proc_init_net,
2962	.exit = udp4_proc_exit_net,
2963};
2964
2965int __init udp4_proc_init(void)
2966{
2967	return register_pernet_subsys(&udp4_net_ops);
2968}
2969
2970void udp4_proc_exit(void)
2971{
2972	unregister_pernet_subsys(&udp4_net_ops);
2973}
2974#endif /* CONFIG_PROC_FS */
2975
2976static __initdata unsigned long uhash_entries;
2977static int __init set_uhash_entries(char *str)
2978{
2979	ssize_t ret;
2980
2981	if (!str)
2982		return 0;
2983
2984	ret = kstrtoul(str, 0, &uhash_entries);
2985	if (ret)
2986		return 0;
2987
2988	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2989		uhash_entries = UDP_HTABLE_SIZE_MIN;
2990	return 1;
2991}
2992__setup("uhash_entries=", set_uhash_entries);
2993
2994void __init udp_table_init(struct udp_table *table, const char *name)
2995{
2996	unsigned int i;
2997
2998	table->hash = alloc_large_system_hash(name,
2999					      2 * sizeof(struct udp_hslot),
3000					      uhash_entries,
3001					      21, /* one slot per 2 MB */
3002					      0,
3003					      &table->log,
3004					      &table->mask,
3005					      UDP_HTABLE_SIZE_MIN,
3006					      64 * 1024);
3007
3008	table->hash2 = table->hash + (table->mask + 1);
3009	for (i = 0; i <= table->mask; i++) {
3010		INIT_HLIST_HEAD(&table->hash[i].head);
3011		table->hash[i].count = 0;
3012		spin_lock_init(&table->hash[i].lock);
3013	}
3014	for (i = 0; i <= table->mask; i++) {
3015		INIT_HLIST_HEAD(&table->hash2[i].head);
3016		table->hash2[i].count = 0;
3017		spin_lock_init(&table->hash2[i].lock);
3018	}
3019}
3020
3021u32 udp_flow_hashrnd(void)
3022{
3023	static u32 hashrnd __read_mostly;
3024
3025	net_get_random_once(&hashrnd, sizeof(hashrnd));
3026
3027	return hashrnd;
3028}
3029EXPORT_SYMBOL(udp_flow_hashrnd);
3030
3031static void __udp_sysctl_init(struct net *net)
3032{
3033	net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3034	net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3035
3036#ifdef CONFIG_NET_L3_MASTER_DEV
3037	net->ipv4.sysctl_udp_l3mdev_accept = 0;
3038#endif
3039}
3040
3041static int __net_init udp_sysctl_init(struct net *net)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3042{
3043	__udp_sysctl_init(net);
 
 
3044	return 0;
3045}
3046
 
 
 
 
 
3047static struct pernet_operations __net_initdata udp_sysctl_ops = {
3048	.init	= udp_sysctl_init,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3049};
3050
 
 
 
 
 
 
 
 
3051void __init udp_init(void)
3052{
3053	unsigned long limit;
3054	unsigned int i;
3055
3056	udp_table_init(&udp_table, "UDP");
3057	limit = nr_free_buffer_pages() / 8;
3058	limit = max(limit, 128UL);
3059	sysctl_udp_mem[0] = limit / 4 * 3;
3060	sysctl_udp_mem[1] = limit;
3061	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3062
3063	__udp_sysctl_init(&init_net);
3064
3065	/* 16 spinlocks per cpu */
3066	udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3067	udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3068				GFP_KERNEL);
3069	if (!udp_busylocks)
3070		panic("UDP: failed to alloc udp_busylocks\n");
3071	for (i = 0; i < (1U << udp_busylocks_log); i++)
3072		spin_lock_init(udp_busylocks + i);
3073
3074	if (register_pernet_subsys(&udp_sysctl_ops))
3075		panic("UDP: failed to init sysctl parameters.\n");
 
 
 
 
3076}
v6.8
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		The User Datagram Protocol (UDP).
   8 *
   9 * Authors:	Ross Biro
  10 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  12 *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
  13 *		Hirokazu Takahashi, <taka@valinux.co.jp>
  14 *
  15 * Fixes:
  16 *		Alan Cox	:	verify_area() calls
  17 *		Alan Cox	: 	stopped close while in use off icmp
  18 *					messages. Not a fix but a botch that
  19 *					for udp at least is 'valid'.
  20 *		Alan Cox	:	Fixed icmp handling properly
  21 *		Alan Cox	: 	Correct error for oversized datagrams
  22 *		Alan Cox	:	Tidied select() semantics.
  23 *		Alan Cox	:	udp_err() fixed properly, also now
  24 *					select and read wake correctly on errors
  25 *		Alan Cox	:	udp_send verify_area moved to avoid mem leak
  26 *		Alan Cox	:	UDP can count its memory
  27 *		Alan Cox	:	send to an unknown connection causes
  28 *					an ECONNREFUSED off the icmp, but
  29 *					does NOT close.
  30 *		Alan Cox	:	Switched to new sk_buff handlers. No more backlog!
  31 *		Alan Cox	:	Using generic datagram code. Even smaller and the PEEK
  32 *					bug no longer crashes it.
  33 *		Fred Van Kempen	: 	Net2e support for sk->broadcast.
  34 *		Alan Cox	:	Uses skb_free_datagram
  35 *		Alan Cox	:	Added get/set sockopt support.
  36 *		Alan Cox	:	Broadcasting without option set returns EACCES.
  37 *		Alan Cox	:	No wakeup calls. Instead we now use the callbacks.
  38 *		Alan Cox	:	Use ip_tos and ip_ttl
  39 *		Alan Cox	:	SNMP Mibs
  40 *		Alan Cox	:	MSG_DONTROUTE, and 0.0.0.0 support.
  41 *		Matt Dillon	:	UDP length checks.
  42 *		Alan Cox	:	Smarter af_inet used properly.
  43 *		Alan Cox	:	Use new kernel side addressing.
  44 *		Alan Cox	:	Incorrect return on truncated datagram receive.
  45 *	Arnt Gulbrandsen 	:	New udp_send and stuff
  46 *		Alan Cox	:	Cache last socket
  47 *		Alan Cox	:	Route cache
  48 *		Jon Peatfield	:	Minor efficiency fix to sendto().
  49 *		Mike Shaver	:	RFC1122 checks.
  50 *		Alan Cox	:	Nonblocking error fix.
  51 *	Willy Konynenberg	:	Transparent proxying support.
  52 *		Mike McLagan	:	Routing by source
  53 *		David S. Miller	:	New socket lookup architecture.
  54 *					Last socket cache retained as it
  55 *					does have a high hit rate.
  56 *		Olaf Kirch	:	Don't linearise iovec on sendmsg.
  57 *		Andi Kleen	:	Some cleanups, cache destination entry
  58 *					for connect.
  59 *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
  60 *		Melvin Smith	:	Check msg_name not msg_namelen in sendto(),
  61 *					return ENOTCONN for unconnected sockets (POSIX)
  62 *		Janos Farkas	:	don't deliver multi/broadcasts to a different
  63 *					bound-to-device socket
  64 *	Hirokazu Takahashi	:	HW checksumming for outgoing UDP
  65 *					datagrams.
  66 *	Hirokazu Takahashi	:	sendfile() on UDP works now.
  67 *		Arnaldo C. Melo :	convert /proc/net/udp to seq_file
  68 *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
  69 *	Alexey Kuznetsov:		allow both IPv4 and IPv6 sockets to bind
  70 *					a single port at the same time.
  71 *	Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
  72 *	James Chapman		:	Add L2TP encapsulation type.
  73 */
  74
  75#define pr_fmt(fmt) "UDP: " fmt
  76
  77#include <linux/bpf-cgroup.h>
  78#include <linux/uaccess.h>
  79#include <asm/ioctls.h>
  80#include <linux/memblock.h>
  81#include <linux/highmem.h>
 
  82#include <linux/types.h>
  83#include <linux/fcntl.h>
  84#include <linux/module.h>
  85#include <linux/socket.h>
  86#include <linux/sockios.h>
  87#include <linux/igmp.h>
  88#include <linux/inetdevice.h>
  89#include <linux/in.h>
  90#include <linux/errno.h>
  91#include <linux/timer.h>
  92#include <linux/mm.h>
  93#include <linux/inet.h>
  94#include <linux/netdevice.h>
  95#include <linux/slab.h>
  96#include <net/tcp_states.h>
  97#include <linux/skbuff.h>
  98#include <linux/proc_fs.h>
  99#include <linux/seq_file.h>
 100#include <net/net_namespace.h>
 101#include <net/icmp.h>
 102#include <net/inet_hashtables.h>
 103#include <net/ip_tunnels.h>
 104#include <net/route.h>
 105#include <net/checksum.h>
 106#include <net/gso.h>
 107#include <net/xfrm.h>
 108#include <trace/events/udp.h>
 109#include <linux/static_key.h>
 110#include <linux/btf_ids.h>
 111#include <trace/events/skb.h>
 112#include <net/busy_poll.h>
 113#include "udp_impl.h"
 114#include <net/sock_reuseport.h>
 115#include <net/addrconf.h>
 116#include <net/udp_tunnel.h>
 117#include <net/gro.h>
 118#if IS_ENABLED(CONFIG_IPV6)
 119#include <net/ipv6_stubs.h>
 120#endif
 121
 122struct udp_table udp_table __read_mostly;
 123EXPORT_SYMBOL(udp_table);
 124
 125long sysctl_udp_mem[3] __read_mostly;
 126EXPORT_SYMBOL(sysctl_udp_mem);
 127
 128atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
 129EXPORT_SYMBOL(udp_memory_allocated);
 130DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
 131EXPORT_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
 136static 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
 141static int udp_lib_lport_inuse(struct net *net, __u16 num,
 142			       const struct udp_hslot *hslot,
 143			       unsigned long *bitmap,
 144			       struct sock *sk, unsigned int log)
 145{
 146	struct sock *sk2;
 147	kuid_t uid = sock_i_uid(sk);
 148
 149	sk_for_each(sk2, &hslot->head) {
 150		if (net_eq(sock_net(sk2), 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, true)) {
 157			if (sk2->sk_reuseport && sk->sk_reuseport &&
 158			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 159			    uid_eq(uid, sock_i_uid(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 */
 177static int udp_lib_lport_inuse2(struct net *net, __u16 num,
 178				struct udp_hslot *hslot2,
 179				struct sock *sk)
 180{
 181	struct sock *sk2;
 182	kuid_t uid = sock_i_uid(sk);
 183	int res = 0;
 184
 185	spin_lock(&hslot2->lock);
 186	udp_portaddr_for_each_entry(sk2, &hslot2->head) {
 187		if (net_eq(sock_net(sk2), 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, true)) {
 194			if (sk2->sk_reuseport && sk->sk_reuseport &&
 195			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 196			    uid_eq(uid, sock_i_uid(sk2))) {
 197				res = 0;
 198			} else {
 199				res = 1;
 200			}
 201			break;
 202		}
 203	}
 204	spin_unlock(&hslot2->lock);
 205	return res;
 206}
 207
 208static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
 209{
 210	struct net *net = sock_net(sk);
 211	kuid_t uid = sock_i_uid(sk);
 212	struct sock *sk2;
 213
 214	sk_for_each(sk2, &hslot->head) {
 215		if (net_eq(sock_net(sk2), 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(uid, sock_i_uid(sk2)) &&
 222		    inet_rcv_saddr_equal(sk, sk2, false)) {
 223			return reuseport_add_sock(sk, sk2,
 224						  inet_rcv_saddr_any(sk));
 225		}
 226	}
 227
 228	return reuseport_alloc(sk, 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 */
 239int 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, &high);
 254		remaining = (high - low) + 1;
 255
 256		rand = get_random_u32();
 257		first = reciprocal_scale(rand, 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(udptable, net, first);
 265			bitmap_zero(bitmap, PORTS_PER_CHAIN);
 266			spin_lock_bh(&hslot->lock);
 267			udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
 268					    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, snum))
 280					goto found;
 281				snum += rand;
 282			} while (snum != first);
 283			spin_unlock_bh(&hslot->lock);
 284			cond_resched();
 285		} while (++first != last);
 286		goto fail;
 287	} else {
 288		hslot = udp_hashslot(udptable, net, snum);
 289		spin_lock_bh(&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(udptable, slot2);
 298			if (hslot->count < hslot2->count)
 299				goto scan_primary_hash;
 300
 301			exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
 302			if (!exist && (hash2_nulladdr != slot2)) {
 303				hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
 304				exist = udp_lib_lport_inuse2(net, snum, hslot2,
 305							     sk);
 306			}
 307			if (exist)
 308				goto fail_unlock;
 309			else
 310				goto found;
 311		}
 312scan_primary_hash:
 313		if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
 314			goto fail_unlock;
 315	}
 316found:
 317	inet_sk(sk)->inet_num = snum;
 318	udp_sk(sk)->udp_port_hash = snum;
 319	udp_sk(sk)->udp_portaddr_hash ^= snum;
 320	if (sk_unhashed(sk)) {
 321		if (sk->sk_reuseport &&
 322		    udp_reuseport_add_sock(sk, hslot)) {
 323			inet_sk(sk)->inet_num = 0;
 324			udp_sk(sk)->udp_port_hash = 0;
 325			udp_sk(sk)->udp_portaddr_hash ^= snum;
 326			goto fail_unlock;
 327		}
 328
 329		sk_add_node_rcu(sk, &hslot->head);
 330		hslot->count++;
 331		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 332
 333		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 334		spin_lock(&hslot2->lock);
 335		if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 336		    sk->sk_family == AF_INET6)
 337			hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
 338					   &hslot2->head);
 339		else
 340			hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
 341					   &hslot2->head);
 342		hslot2->count++;
 343		spin_unlock(&hslot2->lock);
 344	}
 345	sock_set_flag(sk, SOCK_RCU_FREE);
 346	error = 0;
 347fail_unlock:
 348	spin_unlock_bh(&hslot->lock);
 349fail:
 350	return error;
 351}
 352EXPORT_SYMBOL(udp_lib_get_port);
 353
 354int udp_v4_get_port(struct sock *sk, unsigned short snum)
 355{
 356	unsigned int hash2_nulladdr =
 357		ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
 358	unsigned int hash2_partial =
 359		ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
 360
 361	/* precompute partial secondary hash */
 362	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
 363	return udp_lib_get_port(sk, snum, hash2_nulladdr);
 364}
 365
 366static int compute_score(struct sock *sk, struct net *net,
 367			 __be32 saddr, __be16 sport,
 368			 __be32 daddr, unsigned short hnum,
 369			 int dif, int sdif)
 370{
 371	int score;
 372	struct inet_sock *inet;
 373	bool dev_match;
 374
 375	if (!net_eq(sock_net(sk), net) ||
 376	    udp_sk(sk)->udp_port_hash != hnum ||
 377	    ipv6_only_sock(sk))
 378		return -1;
 379
 380	if (sk->sk_rcv_saddr != daddr)
 381		return -1;
 382
 383	score = (sk->sk_family == PF_INET) ? 2 : 1;
 384
 385	inet = inet_sk(sk);
 386	if (inet->inet_daddr) {
 387		if (inet->inet_daddr != saddr)
 388			return -1;
 389		score += 4;
 390	}
 391
 392	if (inet->inet_dport) {
 393		if (inet->inet_dport != sport)
 394			return -1;
 395		score += 4;
 396	}
 397
 398	dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
 399					dif, sdif);
 400	if (!dev_match)
 401		return -1;
 402	if (sk->sk_bound_dev_if)
 403		score += 4;
 404
 405	if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
 406		score++;
 407	return score;
 408}
 409
 410INDIRECT_CALLABLE_SCOPE
 411u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
 412		const __be32 faddr, const __be16 fport)
 413{
 414	static u32 udp_ehash_secret __read_mostly;
 415
 416	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
 417
 418	return __inet_ehashfn(laddr, lport, faddr, fport,
 419			      udp_ehash_secret + net_hash_mix(net));
 420}
 421
 422/* called with rcu_read_lock() */
 423static struct sock *udp4_lib_lookup2(struct net *net,
 424				     __be32 saddr, __be16 sport,
 425				     __be32 daddr, unsigned int hnum,
 426				     int dif, int sdif,
 427				     struct udp_hslot *hslot2,
 428				     struct sk_buff *skb)
 429{
 430	struct sock *sk, *result;
 431	int score, badness;
 
 432
 433	result = NULL;
 434	badness = 0;
 435	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
 436		score = compute_score(sk, net, saddr, sport,
 437				      daddr, hnum, dif, sdif);
 438		if (score > badness) {
 
 
 
 
 
 
 
 
 
 439			badness = score;
 440
 441			if (sk->sk_state == TCP_ESTABLISHED) {
 442				result = sk;
 443				continue;
 444			}
 445
 446			result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
 447						       saddr, sport, daddr, hnum, udp_ehashfn);
 448			if (!result) {
 449				result = sk;
 450				continue;
 451			}
 452
 453			/* Fall back to scoring if group has connections */
 454			if (!reuseport_has_conns(sk))
 455				return result;
 456
 457			/* Reuseport logic returned an error, keep original score. */
 458			if (IS_ERR(result))
 459				continue;
 460
 461			badness = compute_score(result, net, saddr, sport,
 462						daddr, hnum, dif, sdif);
 463
 464		}
 465	}
 466	return result;
 467}
 468
 469/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
 470 * harder than this. -DaveM
 471 */
 472struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
 473		__be16 sport, __be32 daddr, __be16 dport, int dif,
 474		int sdif, struct udp_table *udptable, struct sk_buff *skb)
 475{
 
 476	unsigned short hnum = ntohs(dport);
 477	unsigned int hash2, slot2;
 478	struct udp_hslot *hslot2;
 479	struct sock *result, *sk;
 480
 481	hash2 = ipv4_portaddr_hash(net, daddr, hnum);
 482	slot2 = hash2 & udptable->mask;
 483	hslot2 = &udptable->hash2[slot2];
 484
 485	/* Lookup connected or non-wildcard socket */
 486	result = udp4_lib_lookup2(net, saddr, sport,
 487				  daddr, hnum, dif, sdif,
 488				  hslot2, skb);
 489	if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
 490		goto done;
 491
 492	/* Lookup redirect from BPF */
 493	if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
 494	    udptable == net->ipv4.udp_table) {
 495		sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
 496					       saddr, sport, daddr, hnum, dif,
 497					       udp_ehashfn);
 498		if (sk) {
 499			result = sk;
 500			goto done;
 501		}
 502	}
 503
 504	/* Got non-wildcard socket or error on first lookup */
 505	if (result)
 506		goto done;
 507
 508	/* Lookup wildcard sockets */
 509	hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
 510	slot2 = hash2 & udptable->mask;
 511	hslot2 = &udptable->hash2[slot2];
 512
 513	result = udp4_lib_lookup2(net, saddr, sport,
 514				  htonl(INADDR_ANY), hnum, dif, sdif,
 515				  hslot2, skb);
 516done:
 517	if (IS_ERR(result))
 518		return NULL;
 519	return result;
 520}
 521EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
 522
 523static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
 524						 __be16 sport, __be16 dport,
 525						 struct udp_table *udptable)
 526{
 527	const struct iphdr *iph = ip_hdr(skb);
 528
 529	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
 530				 iph->daddr, dport, inet_iif(skb),
 531				 inet_sdif(skb), udptable, skb);
 532}
 533
 534struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
 535				 __be16 sport, __be16 dport)
 536{
 537	const struct iphdr *iph = ip_hdr(skb);
 538	struct net *net = dev_net(skb->dev);
 539	int iif, sdif;
 540
 541	inet_get_iif_sdif(skb, &iif, &sdif);
 542
 543	return __udp4_lib_lookup(net, iph->saddr, sport,
 544				 iph->daddr, dport, iif,
 545				 sdif, net->ipv4.udp_table, NULL);
 546}
 
 547
 548/* Must be called under rcu_read_lock().
 549 * Does increment socket refcount.
 550 */
 551#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
 552struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 553			     __be32 daddr, __be16 dport, int dif)
 554{
 555	struct sock *sk;
 556
 557	sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
 558			       dif, 0, net->ipv4.udp_table, NULL);
 559	if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
 560		sk = NULL;
 561	return sk;
 562}
 563EXPORT_SYMBOL_GPL(udp4_lib_lookup);
 564#endif
 565
 566static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
 567				       __be16 loc_port, __be32 loc_addr,
 568				       __be16 rmt_port, __be32 rmt_addr,
 569				       int dif, int sdif, unsigned short hnum)
 570{
 571	const struct inet_sock *inet = inet_sk(sk);
 572
 573	if (!net_eq(sock_net(sk), net) ||
 574	    udp_sk(sk)->udp_port_hash != hnum ||
 575	    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
 576	    (inet->inet_dport != rmt_port && inet->inet_dport) ||
 577	    (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
 578	    ipv6_only_sock(sk) ||
 579	    !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
 580		return false;
 581	if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
 582		return false;
 583	return true;
 584}
 585
 586DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
 587void udp_encap_enable(void)
 588{
 589	static_branch_inc(&udp_encap_needed_key);
 590}
 591EXPORT_SYMBOL(udp_encap_enable);
 592
 593void udp_encap_disable(void)
 594{
 595	static_branch_dec(&udp_encap_needed_key);
 596}
 597EXPORT_SYMBOL(udp_encap_disable);
 598
 599/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
 600 * through error handlers in encapsulations looking for a match.
 601 */
 602static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
 603{
 604	int i;
 605
 606	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
 607		int (*handler)(struct sk_buff *skb, u32 info);
 608		const struct ip_tunnel_encap_ops *encap;
 609
 610		encap = rcu_dereference(iptun_encaps[i]);
 611		if (!encap)
 612			continue;
 613		handler = encap->err_handler;
 614		if (handler && !handler(skb, info))
 615			return 0;
 616	}
 617
 618	return -ENOENT;
 619}
 620
 621/* Try to match ICMP errors to UDP tunnels by looking up a socket without
 622 * reversing source and destination port: this will match tunnels that force the
 623 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
 624 * lwtunnels might actually break this assumption by being configured with
 625 * different destination ports on endpoints, in this case we won't be able to
 626 * trace ICMP messages back to them.
 627 *
 628 * If this doesn't match any socket, probe tunnels with arbitrary destination
 629 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
 630 * we've sent packets to won't necessarily match the local destination port.
 631 *
 632 * Then ask the tunnel implementation to match the error against a valid
 633 * association.
 634 *
 635 * Return an error if we can't find a match, the socket if we need further
 636 * processing, zero otherwise.
 637 */
 638static struct sock *__udp4_lib_err_encap(struct net *net,
 639					 const struct iphdr *iph,
 640					 struct udphdr *uh,
 641					 struct udp_table *udptable,
 642					 struct sock *sk,
 643					 struct sk_buff *skb, u32 info)
 644{
 645	int (*lookup)(struct sock *sk, struct sk_buff *skb);
 646	int network_offset, transport_offset;
 647	struct udp_sock *up;
 648
 649	network_offset = skb_network_offset(skb);
 650	transport_offset = skb_transport_offset(skb);
 651
 652	/* Network header needs to point to the outer IPv4 header inside ICMP */
 653	skb_reset_network_header(skb);
 654
 655	/* Transport header needs to point to the UDP header */
 656	skb_set_transport_header(skb, iph->ihl << 2);
 657
 658	if (sk) {
 659		up = udp_sk(sk);
 660
 661		lookup = READ_ONCE(up->encap_err_lookup);
 662		if (lookup && lookup(sk, skb))
 663			sk = NULL;
 664
 665		goto out;
 666	}
 667
 668	sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
 669			       iph->saddr, uh->dest, skb->dev->ifindex, 0,
 670			       udptable, NULL);
 671	if (sk) {
 672		up = udp_sk(sk);
 
 673
 674		lookup = READ_ONCE(up->encap_err_lookup);
 675		if (!lookup || lookup(sk, skb))
 676			sk = NULL;
 677	}
 678
 679out:
 680	if (!sk)
 681		sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
 682
 683	skb_set_transport_header(skb, transport_offset);
 684	skb_set_network_header(skb, network_offset);
 685
 686	return sk;
 687}
 688
 689/*
 690 * This routine is called by the ICMP module when it gets some
 691 * sort of error condition.  If err < 0 then the socket should
 692 * be closed and the error returned to the user.  If err > 0
 693 * it's just the icmp type << 8 | icmp code.
 694 * Header points to the ip header of the error packet. We move
 695 * on past this. Then (as it used to claim before adjustment)
 696 * header points to the first 8 bytes of the udp header.  We need
 697 * to find the appropriate port.
 698 */
 699
 700int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
 701{
 702	struct inet_sock *inet;
 703	const struct iphdr *iph = (const struct iphdr *)skb->data;
 704	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
 705	const int type = icmp_hdr(skb)->type;
 706	const int code = icmp_hdr(skb)->code;
 707	bool tunnel = false;
 708	struct sock *sk;
 709	int harderr;
 710	int err;
 711	struct net *net = dev_net(skb->dev);
 712
 713	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
 714			       iph->saddr, uh->source, skb->dev->ifindex,
 715			       inet_sdif(skb), udptable, NULL);
 716
 717	if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
 718		/* No socket for error: try tunnels before discarding */
 
 719		if (static_branch_unlikely(&udp_encap_needed_key)) {
 720			sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
 721						  info);
 722			if (!sk)
 723				return 0;
 724		} else
 725			sk = ERR_PTR(-ENOENT);
 726
 727		if (IS_ERR(sk)) {
 728			__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 729			return PTR_ERR(sk);
 730		}
 731
 732		tunnel = true;
 733	}
 734
 735	err = 0;
 736	harderr = 0;
 737	inet = inet_sk(sk);
 738
 739	switch (type) {
 740	default:
 741	case ICMP_TIME_EXCEEDED:
 742		err = EHOSTUNREACH;
 743		break;
 744	case ICMP_SOURCE_QUENCH:
 745		goto out;
 746	case ICMP_PARAMETERPROB:
 747		err = EPROTO;
 748		harderr = 1;
 749		break;
 750	case ICMP_DEST_UNREACH:
 751		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
 752			ipv4_sk_update_pmtu(skb, sk, info);
 753			if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
 754				err = EMSGSIZE;
 755				harderr = 1;
 756				break;
 757			}
 758			goto out;
 759		}
 760		err = EHOSTUNREACH;
 761		if (code <= NR_ICMP_UNREACH) {
 762			harderr = icmp_err_convert[code].fatal;
 763			err = icmp_err_convert[code].errno;
 764		}
 765		break;
 766	case ICMP_REDIRECT:
 767		ipv4_sk_redirect(skb, sk);
 768		goto out;
 769	}
 770
 771	/*
 772	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
 773	 *	4.1.3.3.
 774	 */
 775	if (tunnel) {
 776		/* ...not for tunnels though: we don't have a sending socket */
 777		if (udp_sk(sk)->encap_err_rcv)
 778			udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
 779						  (u8 *)(uh+1));
 780		goto out;
 781	}
 782	if (!inet_test_bit(RECVERR, sk)) {
 783		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
 784			goto out;
 785	} else
 786		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
 787
 788	sk->sk_err = err;
 789	sk_error_report(sk);
 790out:
 791	return 0;
 792}
 793
 794int udp_err(struct sk_buff *skb, u32 info)
 795{
 796	return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
 797}
 798
 799/*
 800 * Throw away all pending data and cancel the corking. Socket is locked.
 801 */
 802void udp_flush_pending_frames(struct sock *sk)
 803{
 804	struct udp_sock *up = udp_sk(sk);
 805
 806	if (up->pending) {
 807		up->len = 0;
 808		WRITE_ONCE(up->pending, 0);
 809		ip_flush_pending_frames(sk);
 810	}
 811}
 812EXPORT_SYMBOL(udp_flush_pending_frames);
 813
 814/**
 815 * 	udp4_hwcsum  -  handle outgoing HW checksumming
 816 * 	@skb: 	sk_buff containing the filled-in UDP header
 817 * 	        (checksum field must be zeroed out)
 818 *	@src:	source IP address
 819 *	@dst:	destination IP address
 820 */
 821void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
 822{
 823	struct udphdr *uh = udp_hdr(skb);
 824	int offset = skb_transport_offset(skb);
 825	int len = skb->len - offset;
 826	int hlen = len;
 827	__wsum csum = 0;
 828
 829	if (!skb_has_frag_list(skb)) {
 830		/*
 831		 * Only one fragment on the socket.
 832		 */
 833		skb->csum_start = skb_transport_header(skb) - skb->head;
 834		skb->csum_offset = offsetof(struct udphdr, check);
 835		uh->check = ~csum_tcpudp_magic(src, dst, len,
 836					       IPPROTO_UDP, 0);
 837	} else {
 838		struct sk_buff *frags;
 839
 840		/*
 841		 * HW-checksum won't work as there are two or more
 842		 * fragments on the socket so that all csums of sk_buffs
 843		 * should be together
 844		 */
 845		skb_walk_frags(skb, frags) {
 846			csum = csum_add(csum, frags->csum);
 847			hlen -= frags->len;
 848		}
 849
 850		csum = skb_checksum(skb, offset, hlen, csum);
 851		skb->ip_summed = CHECKSUM_NONE;
 852
 853		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
 854		if (uh->check == 0)
 855			uh->check = CSUM_MANGLED_0;
 856	}
 857}
 858EXPORT_SYMBOL_GPL(udp4_hwcsum);
 859
 860/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
 861 * for the simple case like when setting the checksum for a UDP tunnel.
 862 */
 863void udp_set_csum(bool nocheck, struct sk_buff *skb,
 864		  __be32 saddr, __be32 daddr, int len)
 865{
 866	struct udphdr *uh = udp_hdr(skb);
 867
 868	if (nocheck) {
 869		uh->check = 0;
 870	} else if (skb_is_gso(skb)) {
 871		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 872	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 873		uh->check = 0;
 874		uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
 875		if (uh->check == 0)
 876			uh->check = CSUM_MANGLED_0;
 877	} else {
 878		skb->ip_summed = CHECKSUM_PARTIAL;
 879		skb->csum_start = skb_transport_header(skb) - skb->head;
 880		skb->csum_offset = offsetof(struct udphdr, check);
 881		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 882	}
 883}
 884EXPORT_SYMBOL(udp_set_csum);
 885
 886static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
 887			struct inet_cork *cork)
 888{
 889	struct sock *sk = skb->sk;
 890	struct inet_sock *inet = inet_sk(sk);
 891	struct udphdr *uh;
 892	int err;
 893	int is_udplite = IS_UDPLITE(sk);
 894	int offset = skb_transport_offset(skb);
 895	int len = skb->len - offset;
 896	int datalen = len - sizeof(*uh);
 897	__wsum csum = 0;
 898
 899	/*
 900	 * Create a UDP header
 901	 */
 902	uh = udp_hdr(skb);
 903	uh->source = inet->inet_sport;
 904	uh->dest = fl4->fl4_dport;
 905	uh->len = htons(len);
 906	uh->check = 0;
 907
 908	if (cork->gso_size) {
 909		const int hlen = skb_network_header_len(skb) +
 910				 sizeof(struct udphdr);
 911
 912		if (hlen + cork->gso_size > cork->fragsize) {
 913			kfree_skb(skb);
 914			return -EINVAL;
 915		}
 916		if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
 917			kfree_skb(skb);
 918			return -EINVAL;
 919		}
 920		if (sk->sk_no_check_tx) {
 921			kfree_skb(skb);
 922			return -EINVAL;
 923		}
 924		if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
 925		    dst_xfrm(skb_dst(skb))) {
 926			kfree_skb(skb);
 927			return -EIO;
 928		}
 929
 930		if (datalen > cork->gso_size) {
 931			skb_shinfo(skb)->gso_size = cork->gso_size;
 932			skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
 933			skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
 934								 cork->gso_size);
 935		}
 936		goto csum_partial;
 937	}
 938
 939	if (is_udplite)  				 /*     UDP-Lite      */
 940		csum = udplite_csum(skb);
 941
 942	else if (sk->sk_no_check_tx) {			 /* UDP csum off */
 943
 944		skb->ip_summed = CHECKSUM_NONE;
 945		goto send;
 946
 947	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
 948csum_partial:
 949
 950		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
 951		goto send;
 952
 953	} else
 954		csum = udp_csum(skb);
 955
 956	/* add protocol-dependent pseudo-header */
 957	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
 958				      sk->sk_protocol, csum);
 959	if (uh->check == 0)
 960		uh->check = CSUM_MANGLED_0;
 961
 962send:
 963	err = ip_send_skb(sock_net(sk), skb);
 964	if (err) {
 965		if (err == -ENOBUFS &&
 966		    !inet_test_bit(RECVERR, sk)) {
 967			UDP_INC_STATS(sock_net(sk),
 968				      UDP_MIB_SNDBUFERRORS, is_udplite);
 969			err = 0;
 970		}
 971	} else
 972		UDP_INC_STATS(sock_net(sk),
 973			      UDP_MIB_OUTDATAGRAMS, is_udplite);
 974	return err;
 975}
 976
 977/*
 978 * Push out all pending data as one UDP datagram. Socket is locked.
 979 */
 980int udp_push_pending_frames(struct sock *sk)
 981{
 982	struct udp_sock  *up = udp_sk(sk);
 983	struct inet_sock *inet = inet_sk(sk);
 984	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
 985	struct sk_buff *skb;
 986	int err = 0;
 987
 988	skb = ip_finish_skb(sk, fl4);
 989	if (!skb)
 990		goto out;
 991
 992	err = udp_send_skb(skb, fl4, &inet->cork.base);
 993
 994out:
 995	up->len = 0;
 996	WRITE_ONCE(up->pending, 0);
 997	return err;
 998}
 999EXPORT_SYMBOL(udp_push_pending_frames);
1000
1001static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1002{
1003	switch (cmsg->cmsg_type) {
1004	case UDP_SEGMENT:
1005		if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1006			return -EINVAL;
1007		*gso_size = *(__u16 *)CMSG_DATA(cmsg);
1008		return 0;
1009	default:
1010		return -EINVAL;
1011	}
1012}
1013
1014int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1015{
1016	struct cmsghdr *cmsg;
1017	bool need_ip = false;
1018	int err;
1019
1020	for_each_cmsghdr(cmsg, msg) {
1021		if (!CMSG_OK(msg, cmsg))
1022			return -EINVAL;
1023
1024		if (cmsg->cmsg_level != SOL_UDP) {
1025			need_ip = true;
1026			continue;
1027		}
1028
1029		err = __udp_cmsg_send(cmsg, gso_size);
1030		if (err)
1031			return err;
1032	}
1033
1034	return need_ip;
1035}
1036EXPORT_SYMBOL_GPL(udp_cmsg_send);
1037
1038int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1039{
1040	struct inet_sock *inet = inet_sk(sk);
1041	struct udp_sock *up = udp_sk(sk);
1042	DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1043	struct flowi4 fl4_stack;
1044	struct flowi4 *fl4;
1045	int ulen = len;
1046	struct ipcm_cookie ipc;
1047	struct rtable *rt = NULL;
1048	int free = 0;
1049	int connected = 0;
1050	__be32 daddr, faddr, saddr;
1051	u8 tos, scope;
1052	__be16 dport;
 
1053	int err, is_udplite = IS_UDPLITE(sk);
1054	int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1055	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1056	struct sk_buff *skb;
1057	struct ip_options_data opt_copy;
1058	int uc_index;
1059
1060	if (len > 0xFFFF)
1061		return -EMSGSIZE;
1062
1063	/*
1064	 *	Check the flags.
1065	 */
1066
1067	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1068		return -EOPNOTSUPP;
1069
1070	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1071
1072	fl4 = &inet->cork.fl.u.ip4;
1073	if (READ_ONCE(up->pending)) {
1074		/*
1075		 * There are pending frames.
1076		 * The socket lock must be held while it's corked.
1077		 */
1078		lock_sock(sk);
1079		if (likely(up->pending)) {
1080			if (unlikely(up->pending != AF_INET)) {
1081				release_sock(sk);
1082				return -EINVAL;
1083			}
1084			goto do_append_data;
1085		}
1086		release_sock(sk);
1087	}
1088	ulen += sizeof(struct udphdr);
1089
1090	/*
1091	 *	Get and verify the address.
1092	 */
1093	if (usin) {
1094		if (msg->msg_namelen < sizeof(*usin))
1095			return -EINVAL;
1096		if (usin->sin_family != AF_INET) {
1097			if (usin->sin_family != AF_UNSPEC)
1098				return -EAFNOSUPPORT;
1099		}
1100
1101		daddr = usin->sin_addr.s_addr;
1102		dport = usin->sin_port;
1103		if (dport == 0)
1104			return -EINVAL;
1105	} else {
1106		if (sk->sk_state != TCP_ESTABLISHED)
1107			return -EDESTADDRREQ;
1108		daddr = inet->inet_daddr;
1109		dport = inet->inet_dport;
1110		/* Open fast path for connected socket.
1111		   Route will not be used, if at least one option is set.
1112		 */
1113		connected = 1;
1114	}
1115
1116	ipcm_init_sk(&ipc, inet);
1117	ipc.gso_size = READ_ONCE(up->gso_size);
1118
1119	if (msg->msg_controllen) {
1120		err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1121		if (err > 0)
1122			err = ip_cmsg_send(sk, msg, &ipc,
1123					   sk->sk_family == AF_INET6);
1124		if (unlikely(err < 0)) {
1125			kfree(ipc.opt);
1126			return err;
1127		}
1128		if (ipc.opt)
1129			free = 1;
1130		connected = 0;
1131	}
1132	if (!ipc.opt) {
1133		struct ip_options_rcu *inet_opt;
1134
1135		rcu_read_lock();
1136		inet_opt = rcu_dereference(inet->inet_opt);
1137		if (inet_opt) {
1138			memcpy(&opt_copy, inet_opt,
1139			       sizeof(*inet_opt) + inet_opt->opt.optlen);
1140			ipc.opt = &opt_copy.opt;
1141		}
1142		rcu_read_unlock();
1143	}
1144
1145	if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1146		err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1147					    (struct sockaddr *)usin,
1148					    &msg->msg_namelen,
1149					    &ipc.addr);
1150		if (err)
1151			goto out_free;
1152		if (usin) {
1153			if (usin->sin_port == 0) {
1154				/* BPF program set invalid port. Reject it. */
1155				err = -EINVAL;
1156				goto out_free;
1157			}
1158			daddr = usin->sin_addr.s_addr;
1159			dport = usin->sin_port;
1160		}
1161	}
1162
1163	saddr = ipc.addr;
1164	ipc.addr = faddr = daddr;
1165
1166	if (ipc.opt && ipc.opt->opt.srr) {
1167		if (!daddr) {
1168			err = -EINVAL;
1169			goto out_free;
1170		}
1171		faddr = ipc.opt->opt.faddr;
1172		connected = 0;
1173	}
1174	tos = get_rttos(&ipc, inet);
1175	scope = ip_sendmsg_scope(inet, &ipc, msg);
1176	if (scope == RT_SCOPE_LINK)
 
 
1177		connected = 0;
 
1178
1179	uc_index = READ_ONCE(inet->uc_index);
1180	if (ipv4_is_multicast(daddr)) {
1181		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1182			ipc.oif = READ_ONCE(inet->mc_index);
1183		if (!saddr)
1184			saddr = READ_ONCE(inet->mc_addr);
1185		connected = 0;
1186	} else if (!ipc.oif) {
1187		ipc.oif = uc_index;
1188	} else if (ipv4_is_lbcast(daddr) && uc_index) {
1189		/* oif is set, packet is to local broadcast and
1190		 * uc_index is set. oif is most likely set
1191		 * by sk_bound_dev_if. If uc_index != oif check if the
1192		 * oif is an L3 master and uc_index is an L3 slave.
1193		 * If so, we want to allow the send using the uc_index.
1194		 */
1195		if (ipc.oif != uc_index &&
1196		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1197							      uc_index)) {
1198			ipc.oif = uc_index;
1199		}
1200	}
1201
1202	if (connected)
1203		rt = (struct rtable *)sk_dst_check(sk, 0);
1204
1205	if (!rt) {
1206		struct net *net = sock_net(sk);
1207		__u8 flow_flags = inet_sk_flowi_flags(sk);
1208
1209		fl4 = &fl4_stack;
1210
1211		flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1212				   sk->sk_protocol, flow_flags, faddr, saddr,
1213				   dport, inet->inet_sport, sk->sk_uid);
 
 
1214
1215		security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1216		rt = ip_route_output_flow(net, fl4, sk);
1217		if (IS_ERR(rt)) {
1218			err = PTR_ERR(rt);
1219			rt = NULL;
1220			if (err == -ENETUNREACH)
1221				IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1222			goto out;
1223		}
1224
1225		err = -EACCES;
1226		if ((rt->rt_flags & RTCF_BROADCAST) &&
1227		    !sock_flag(sk, SOCK_BROADCAST))
1228			goto out;
1229		if (connected)
1230			sk_dst_set(sk, dst_clone(&rt->dst));
1231	}
1232
1233	if (msg->msg_flags&MSG_CONFIRM)
1234		goto do_confirm;
1235back_from_confirm:
1236
1237	saddr = fl4->saddr;
1238	if (!ipc.addr)
1239		daddr = ipc.addr = fl4->daddr;
1240
1241	/* Lockless fast path for the non-corking case. */
1242	if (!corkreq) {
1243		struct inet_cork cork;
1244
1245		skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1246				  sizeof(struct udphdr), &ipc, &rt,
1247				  &cork, msg->msg_flags);
1248		err = PTR_ERR(skb);
1249		if (!IS_ERR_OR_NULL(skb))
1250			err = udp_send_skb(skb, fl4, &cork);
1251		goto out;
1252	}
1253
1254	lock_sock(sk);
1255	if (unlikely(up->pending)) {
1256		/* The socket is already corked while preparing it. */
1257		/* ... which is an evident application bug. --ANK */
1258		release_sock(sk);
1259
1260		net_dbg_ratelimited("socket already corked\n");
1261		err = -EINVAL;
1262		goto out;
1263	}
1264	/*
1265	 *	Now cork the socket to pend data.
1266	 */
1267	fl4 = &inet->cork.fl.u.ip4;
1268	fl4->daddr = daddr;
1269	fl4->saddr = saddr;
1270	fl4->fl4_dport = dport;
1271	fl4->fl4_sport = inet->inet_sport;
1272	WRITE_ONCE(up->pending, AF_INET);
1273
1274do_append_data:
1275	up->len += ulen;
1276	err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1277			     sizeof(struct udphdr), &ipc, &rt,
1278			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1279	if (err)
1280		udp_flush_pending_frames(sk);
1281	else if (!corkreq)
1282		err = udp_push_pending_frames(sk);
1283	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1284		WRITE_ONCE(up->pending, 0);
1285	release_sock(sk);
1286
1287out:
1288	ip_rt_put(rt);
1289out_free:
1290	if (free)
1291		kfree(ipc.opt);
1292	if (!err)
1293		return len;
1294	/*
1295	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1296	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1297	 * we don't have a good statistic (IpOutDiscards but it can be too many
1298	 * things).  We could add another new stat but at least for now that
1299	 * seems like overkill.
1300	 */
1301	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1302		UDP_INC_STATS(sock_net(sk),
1303			      UDP_MIB_SNDBUFERRORS, is_udplite);
1304	}
1305	return err;
1306
1307do_confirm:
1308	if (msg->msg_flags & MSG_PROBE)
1309		dst_confirm_neigh(&rt->dst, &fl4->daddr);
1310	if (!(msg->msg_flags&MSG_PROBE) || len)
1311		goto back_from_confirm;
1312	err = 0;
1313	goto out;
1314}
1315EXPORT_SYMBOL(udp_sendmsg);
1316
1317void udp_splice_eof(struct socket *sock)
 
1318{
1319	struct sock *sk = sock->sk;
1320	struct udp_sock *up = udp_sk(sk);
 
 
 
 
1321
1322	if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1323		return;
 
 
 
 
 
 
 
 
 
1324
1325	lock_sock(sk);
1326	if (up->pending && !udp_test_bit(CORK, sk))
1327		udp_push_pending_frames(sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1328	release_sock(sk);
 
1329}
1330EXPORT_SYMBOL_GPL(udp_splice_eof);
1331
1332#define UDP_SKB_IS_STATELESS 0x80000000
1333
1334/* all head states (dst, sk, nf conntrack) except skb extensions are
1335 * cleared by udp_rcv().
1336 *
1337 * We need to preserve secpath, if present, to eventually process
1338 * IP_CMSG_PASSSEC at recvmsg() time.
1339 *
1340 * Other extensions can be cleared.
1341 */
1342static bool udp_try_make_stateless(struct sk_buff *skb)
1343{
1344	if (!skb_has_extensions(skb))
1345		return true;
1346
1347	if (!secpath_exists(skb)) {
1348		skb_ext_reset(skb);
1349		return true;
1350	}
1351
1352	return false;
1353}
1354
1355static void udp_set_dev_scratch(struct sk_buff *skb)
1356{
1357	struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1358
1359	BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1360	scratch->_tsize_state = skb->truesize;
1361#if BITS_PER_LONG == 64
1362	scratch->len = skb->len;
1363	scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1364	scratch->is_linear = !skb_is_nonlinear(skb);
1365#endif
1366	if (udp_try_make_stateless(skb))
1367		scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1368}
1369
1370static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1371{
1372	/* We come here after udp_lib_checksum_complete() returned 0.
1373	 * This means that __skb_checksum_complete() might have
1374	 * set skb->csum_valid to 1.
1375	 * On 64bit platforms, we can set csum_unnecessary
1376	 * to true, but only if the skb is not shared.
1377	 */
1378#if BITS_PER_LONG == 64
1379	if (!skb_shared(skb))
1380		udp_skb_scratch(skb)->csum_unnecessary = true;
1381#endif
1382}
1383
1384static int udp_skb_truesize(struct sk_buff *skb)
1385{
1386	return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1387}
1388
1389static bool udp_skb_has_head_state(struct sk_buff *skb)
1390{
1391	return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1392}
1393
1394/* fully reclaim rmem/fwd memory allocated for skb */
1395static void udp_rmem_release(struct sock *sk, int size, int partial,
1396			     bool rx_queue_lock_held)
1397{
1398	struct udp_sock *up = udp_sk(sk);
1399	struct sk_buff_head *sk_queue;
1400	int amt;
1401
1402	if (likely(partial)) {
1403		up->forward_deficit += size;
1404		size = up->forward_deficit;
1405		if (size < READ_ONCE(up->forward_threshold) &&
1406		    !skb_queue_empty(&up->reader_queue))
1407			return;
1408	} else {
1409		size += up->forward_deficit;
1410	}
1411	up->forward_deficit = 0;
1412
1413	/* acquire the sk_receive_queue for fwd allocated memory scheduling,
1414	 * if the called don't held it already
1415	 */
1416	sk_queue = &sk->sk_receive_queue;
1417	if (!rx_queue_lock_held)
1418		spin_lock(&sk_queue->lock);
1419
1420
1421	sk_forward_alloc_add(sk, size);
1422	amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1423	sk_forward_alloc_add(sk, -amt);
1424
1425	if (amt)
1426		__sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1427
1428	atomic_sub(size, &sk->sk_rmem_alloc);
1429
1430	/* this can save us from acquiring the rx queue lock on next receive */
1431	skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1432
1433	if (!rx_queue_lock_held)
1434		spin_unlock(&sk_queue->lock);
1435}
1436
1437/* Note: called with reader_queue.lock held.
1438 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1439 * This avoids a cache line miss while receive_queue lock is held.
1440 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1441 */
1442void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1443{
1444	prefetch(&skb->data);
1445	udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1446}
1447EXPORT_SYMBOL(udp_skb_destructor);
1448
1449/* as above, but the caller held the rx queue lock, too */
1450static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1451{
1452	prefetch(&skb->data);
1453	udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1454}
1455
1456/* Idea of busylocks is to let producers grab an extra spinlock
1457 * to relieve pressure on the receive_queue spinlock shared by consumer.
1458 * Under flood, this means that only one producer can be in line
1459 * trying to acquire the receive_queue spinlock.
1460 * These busylock can be allocated on a per cpu manner, instead of a
1461 * per socket one (that would consume a cache line per socket)
1462 */
1463static int udp_busylocks_log __read_mostly;
1464static spinlock_t *udp_busylocks __read_mostly;
1465
1466static spinlock_t *busylock_acquire(void *ptr)
1467{
1468	spinlock_t *busy;
1469
1470	busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1471	spin_lock(busy);
1472	return busy;
1473}
1474
1475static void busylock_release(spinlock_t *busy)
1476{
1477	if (busy)
1478		spin_unlock(busy);
1479}
1480
1481static int udp_rmem_schedule(struct sock *sk, int size)
1482{
1483	int delta;
1484
1485	delta = size - sk->sk_forward_alloc;
1486	if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1487		return -ENOBUFS;
1488
1489	return 0;
1490}
1491
1492int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1493{
1494	struct sk_buff_head *list = &sk->sk_receive_queue;
1495	int rmem, err = -ENOMEM;
1496	spinlock_t *busy = NULL;
1497	int size;
1498
1499	/* try to avoid the costly atomic add/sub pair when the receive
1500	 * queue is full; always allow at least a packet
1501	 */
1502	rmem = atomic_read(&sk->sk_rmem_alloc);
1503	if (rmem > sk->sk_rcvbuf)
1504		goto drop;
1505
1506	/* Under mem pressure, it might be helpful to help udp_recvmsg()
1507	 * having linear skbs :
1508	 * - Reduce memory overhead and thus increase receive queue capacity
1509	 * - Less cache line misses at copyout() time
1510	 * - Less work at consume_skb() (less alien page frag freeing)
1511	 */
1512	if (rmem > (sk->sk_rcvbuf >> 1)) {
1513		skb_condense(skb);
1514
1515		busy = busylock_acquire(sk);
1516	}
1517	size = skb->truesize;
1518	udp_set_dev_scratch(skb);
1519
1520	/* we drop only if the receive buf is full and the receive
1521	 * queue contains some other skb
1522	 */
1523	rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1524	if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1525		goto uncharge_drop;
1526
1527	spin_lock(&list->lock);
1528	err = udp_rmem_schedule(sk, size);
1529	if (err) {
1530		spin_unlock(&list->lock);
1531		goto uncharge_drop;
 
 
 
 
 
 
1532	}
1533
1534	sk_forward_alloc_add(sk, -size);
1535
1536	/* no need to setup a destructor, we will explicitly release the
1537	 * forward allocated memory on dequeue
1538	 */
1539	sock_skb_set_dropcount(sk, skb);
1540
1541	__skb_queue_tail(list, skb);
1542	spin_unlock(&list->lock);
1543
1544	if (!sock_flag(sk, SOCK_DEAD))
1545		INDIRECT_CALL_1(sk->sk_data_ready, sock_def_readable, sk);
1546
1547	busylock_release(busy);
1548	return 0;
1549
1550uncharge_drop:
1551	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1552
1553drop:
1554	atomic_inc(&sk->sk_drops);
1555	busylock_release(busy);
1556	return err;
1557}
1558EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1559
1560void udp_destruct_common(struct sock *sk)
1561{
1562	/* reclaim completely the forward allocated memory */
1563	struct udp_sock *up = udp_sk(sk);
1564	unsigned int total = 0;
1565	struct sk_buff *skb;
1566
1567	skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1568	while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1569		total += skb->truesize;
1570		kfree_skb(skb);
1571	}
1572	udp_rmem_release(sk, total, 0, true);
1573}
1574EXPORT_SYMBOL_GPL(udp_destruct_common);
1575
1576static void udp_destruct_sock(struct sock *sk)
1577{
1578	udp_destruct_common(sk);
1579	inet_sock_destruct(sk);
1580}
 
1581
1582int udp_init_sock(struct sock *sk)
1583{
1584	udp_lib_init_sock(sk);
1585	sk->sk_destruct = udp_destruct_sock;
1586	set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1587	return 0;
1588}
 
1589
1590void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1591{
1592	sk_peek_offset_bwd(sk, len);
 
 
 
 
 
1593
1594	if (!skb_unref(skb))
1595		return;
1596
1597	/* In the more common cases we cleared the head states previously,
1598	 * see __udp_queue_rcv_skb().
1599	 */
1600	if (unlikely(udp_skb_has_head_state(skb)))
1601		skb_release_head_state(skb);
1602	__consume_stateless_skb(skb);
1603}
1604EXPORT_SYMBOL_GPL(skb_consume_udp);
1605
1606static struct sk_buff *__first_packet_length(struct sock *sk,
1607					     struct sk_buff_head *rcvq,
1608					     int *total)
1609{
1610	struct sk_buff *skb;
1611
1612	while ((skb = skb_peek(rcvq)) != NULL) {
1613		if (udp_lib_checksum_complete(skb)) {
1614			__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1615					IS_UDPLITE(sk));
1616			__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1617					IS_UDPLITE(sk));
1618			atomic_inc(&sk->sk_drops);
1619			__skb_unlink(skb, rcvq);
1620			*total += skb->truesize;
1621			kfree_skb(skb);
1622		} else {
1623			udp_skb_csum_unnecessary_set(skb);
1624			break;
1625		}
1626	}
1627	return skb;
1628}
1629
1630/**
1631 *	first_packet_length	- return length of first packet in receive queue
1632 *	@sk: socket
1633 *
1634 *	Drops all bad checksum frames, until a valid one is found.
1635 *	Returns the length of found skb, or -1 if none is found.
1636 */
1637static int first_packet_length(struct sock *sk)
1638{
1639	struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1640	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1641	struct sk_buff *skb;
1642	int total = 0;
1643	int res;
1644
1645	spin_lock_bh(&rcvq->lock);
1646	skb = __first_packet_length(sk, rcvq, &total);
1647	if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1648		spin_lock(&sk_queue->lock);
1649		skb_queue_splice_tail_init(sk_queue, rcvq);
1650		spin_unlock(&sk_queue->lock);
1651
1652		skb = __first_packet_length(sk, rcvq, &total);
1653	}
1654	res = skb ? skb->len : -1;
1655	if (total)
1656		udp_rmem_release(sk, total, 1, false);
1657	spin_unlock_bh(&rcvq->lock);
1658	return res;
1659}
1660
1661/*
1662 *	IOCTL requests applicable to the UDP protocol
1663 */
1664
1665int udp_ioctl(struct sock *sk, int cmd, int *karg)
1666{
1667	switch (cmd) {
1668	case SIOCOUTQ:
1669	{
1670		*karg = sk_wmem_alloc_get(sk);
1671		return 0;
 
1672	}
1673
1674	case SIOCINQ:
1675	{
1676		*karg = max_t(int, 0, first_packet_length(sk));
1677		return 0;
 
1678	}
1679
1680	default:
1681		return -ENOIOCTLCMD;
1682	}
1683
1684	return 0;
1685}
1686EXPORT_SYMBOL(udp_ioctl);
1687
1688struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1689			       int *off, int *err)
1690{
1691	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1692	struct sk_buff_head *queue;
1693	struct sk_buff *last;
1694	long timeo;
1695	int error;
1696
1697	queue = &udp_sk(sk)->reader_queue;
 
1698	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1699	do {
1700		struct sk_buff *skb;
1701
1702		error = sock_error(sk);
1703		if (error)
1704			break;
1705
1706		error = -EAGAIN;
1707		do {
1708			spin_lock_bh(&queue->lock);
1709			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1710							err, &last);
 
1711			if (skb) {
1712				if (!(flags & MSG_PEEK))
1713					udp_skb_destructor(sk, skb);
1714				spin_unlock_bh(&queue->lock);
1715				return skb;
1716			}
1717
1718			if (skb_queue_empty_lockless(sk_queue)) {
1719				spin_unlock_bh(&queue->lock);
1720				goto busy_check;
1721			}
1722
1723			/* refill the reader queue and walk it again
1724			 * keep both queues locked to avoid re-acquiring
1725			 * the sk_receive_queue lock if fwd memory scheduling
1726			 * is needed.
1727			 */
1728			spin_lock(&sk_queue->lock);
1729			skb_queue_splice_tail_init(sk_queue, queue);
1730
1731			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1732							err, &last);
1733			if (skb && !(flags & MSG_PEEK))
1734				udp_skb_dtor_locked(sk, skb);
1735			spin_unlock(&sk_queue->lock);
1736			spin_unlock_bh(&queue->lock);
1737			if (skb)
1738				return skb;
1739
1740busy_check:
1741			if (!sk_can_busy_loop(sk))
1742				break;
1743
1744			sk_busy_loop(sk, flags & MSG_DONTWAIT);
1745		} while (!skb_queue_empty_lockless(sk_queue));
1746
1747		/* sk_queue is empty, reader_queue may contain peeked packets */
1748	} while (timeo &&
1749		 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1750					      &error, &timeo,
1751					      (struct sk_buff *)sk_queue));
1752
1753	*err = error;
1754	return NULL;
1755}
1756EXPORT_SYMBOL(__skb_recv_udp);
1757
1758int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1759{
1760	struct sk_buff *skb;
1761	int err;
1762
1763try_again:
1764	skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1765	if (!skb)
1766		return err;
1767
1768	if (udp_lib_checksum_complete(skb)) {
1769		int is_udplite = IS_UDPLITE(sk);
1770		struct net *net = sock_net(sk);
1771
1772		__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1773		__UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1774		atomic_inc(&sk->sk_drops);
1775		kfree_skb(skb);
1776		goto try_again;
1777	}
1778
1779	WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1780	return recv_actor(sk, skb);
1781}
1782EXPORT_SYMBOL(udp_read_skb);
1783
1784/*
1785 * 	This should be easy, if there is something there we
1786 * 	return it, otherwise we block.
1787 */
1788
1789int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1790		int *addr_len)
1791{
1792	struct inet_sock *inet = inet_sk(sk);
1793	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1794	struct sk_buff *skb;
1795	unsigned int ulen, copied;
1796	int off, err, peeking = flags & MSG_PEEK;
1797	int is_udplite = IS_UDPLITE(sk);
1798	bool checksum_valid = false;
1799
1800	if (flags & MSG_ERRQUEUE)
1801		return ip_recv_error(sk, msg, len, addr_len);
1802
1803try_again:
1804	off = sk_peek_offset(sk, flags);
1805	skb = __skb_recv_udp(sk, flags, &off, &err);
1806	if (!skb)
1807		return err;
1808
1809	ulen = udp_skb_len(skb);
1810	copied = len;
1811	if (copied > ulen - off)
1812		copied = ulen - off;
1813	else if (copied < ulen)
1814		msg->msg_flags |= MSG_TRUNC;
1815
1816	/*
1817	 * If checksum is needed at all, try to do it while copying the
1818	 * data.  If the data is truncated, or if we only want a partial
1819	 * coverage checksum (UDP-Lite), do it before the copy.
1820	 */
1821
1822	if (copied < ulen || peeking ||
1823	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1824		checksum_valid = udp_skb_csum_unnecessary(skb) ||
1825				!__udp_lib_checksum_complete(skb);
1826		if (!checksum_valid)
1827			goto csum_copy_err;
1828	}
1829
1830	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1831		if (udp_skb_is_linear(skb))
1832			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1833		else
1834			err = skb_copy_datagram_msg(skb, off, msg, copied);
1835	} else {
1836		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1837
1838		if (err == -EINVAL)
1839			goto csum_copy_err;
1840	}
1841
1842	if (unlikely(err)) {
1843		if (!peeking) {
1844			atomic_inc(&sk->sk_drops);
1845			UDP_INC_STATS(sock_net(sk),
1846				      UDP_MIB_INERRORS, is_udplite);
1847		}
1848		kfree_skb(skb);
1849		return err;
1850	}
1851
1852	if (!peeking)
1853		UDP_INC_STATS(sock_net(sk),
1854			      UDP_MIB_INDATAGRAMS, is_udplite);
1855
1856	sock_recv_cmsgs(msg, sk, skb);
1857
1858	/* Copy the address. */
1859	if (sin) {
1860		sin->sin_family = AF_INET;
1861		sin->sin_port = udp_hdr(skb)->source;
1862		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1863		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1864		*addr_len = sizeof(*sin);
1865
1866		BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1867						      (struct sockaddr *)sin,
1868						      addr_len);
1869	}
1870
1871	if (udp_test_bit(GRO_ENABLED, sk))
1872		udp_cmsg_recv(msg, sk, skb);
1873
1874	if (inet_cmsg_flags(inet))
1875		ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1876
1877	err = copied;
1878	if (flags & MSG_TRUNC)
1879		err = ulen;
1880
1881	skb_consume_udp(sk, skb, peeking ? -err : err);
1882	return err;
1883
1884csum_copy_err:
1885	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1886				 udp_skb_destructor)) {
1887		UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1888		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1889	}
1890	kfree_skb(skb);
1891
1892	/* starting over for a new packet, but check if we need to yield */
1893	cond_resched();
1894	msg->msg_flags &= ~MSG_TRUNC;
1895	goto try_again;
1896}
1897
1898int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1899{
1900	/* This check is replicated from __ip4_datagram_connect() and
1901	 * intended to prevent BPF program called below from accessing bytes
1902	 * that are out of the bound specified by user in addr_len.
1903	 */
1904	if (addr_len < sizeof(struct sockaddr_in))
1905		return -EINVAL;
1906
1907	return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
1908}
1909EXPORT_SYMBOL(udp_pre_connect);
1910
1911int __udp_disconnect(struct sock *sk, int flags)
1912{
1913	struct inet_sock *inet = inet_sk(sk);
1914	/*
1915	 *	1003.1g - break association.
1916	 */
1917
1918	sk->sk_state = TCP_CLOSE;
1919	inet->inet_daddr = 0;
1920	inet->inet_dport = 0;
1921	sock_rps_reset_rxhash(sk);
1922	sk->sk_bound_dev_if = 0;
1923	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1924		inet_reset_saddr(sk);
1925		if (sk->sk_prot->rehash &&
1926		    (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1927			sk->sk_prot->rehash(sk);
1928	}
1929
1930	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1931		sk->sk_prot->unhash(sk);
1932		inet->inet_sport = 0;
1933	}
1934	sk_dst_reset(sk);
1935	return 0;
1936}
1937EXPORT_SYMBOL(__udp_disconnect);
1938
1939int udp_disconnect(struct sock *sk, int flags)
1940{
1941	lock_sock(sk);
1942	__udp_disconnect(sk, flags);
1943	release_sock(sk);
1944	return 0;
1945}
1946EXPORT_SYMBOL(udp_disconnect);
1947
1948void udp_lib_unhash(struct sock *sk)
1949{
1950	if (sk_hashed(sk)) {
1951		struct udp_table *udptable = udp_get_table_prot(sk);
1952		struct udp_hslot *hslot, *hslot2;
1953
1954		hslot  = udp_hashslot(udptable, sock_net(sk),
1955				      udp_sk(sk)->udp_port_hash);
1956		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1957
1958		spin_lock_bh(&hslot->lock);
1959		if (rcu_access_pointer(sk->sk_reuseport_cb))
1960			reuseport_detach_sock(sk);
1961		if (sk_del_node_init_rcu(sk)) {
1962			hslot->count--;
1963			inet_sk(sk)->inet_num = 0;
1964			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1965
1966			spin_lock(&hslot2->lock);
1967			hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1968			hslot2->count--;
1969			spin_unlock(&hslot2->lock);
1970		}
1971		spin_unlock_bh(&hslot->lock);
1972	}
1973}
1974EXPORT_SYMBOL(udp_lib_unhash);
1975
1976/*
1977 * inet_rcv_saddr was changed, we must rehash secondary hash
1978 */
1979void udp_lib_rehash(struct sock *sk, u16 newhash)
1980{
1981	if (sk_hashed(sk)) {
1982		struct udp_table *udptable = udp_get_table_prot(sk);
1983		struct udp_hslot *hslot, *hslot2, *nhslot2;
1984
1985		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1986		nhslot2 = udp_hashslot2(udptable, newhash);
1987		udp_sk(sk)->udp_portaddr_hash = newhash;
1988
1989		if (hslot2 != nhslot2 ||
1990		    rcu_access_pointer(sk->sk_reuseport_cb)) {
1991			hslot = udp_hashslot(udptable, sock_net(sk),
1992					     udp_sk(sk)->udp_port_hash);
1993			/* we must lock primary chain too */
1994			spin_lock_bh(&hslot->lock);
1995			if (rcu_access_pointer(sk->sk_reuseport_cb))
1996				reuseport_detach_sock(sk);
1997
1998			if (hslot2 != nhslot2) {
1999				spin_lock(&hslot2->lock);
2000				hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2001				hslot2->count--;
2002				spin_unlock(&hslot2->lock);
2003
2004				spin_lock(&nhslot2->lock);
2005				hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2006							 &nhslot2->head);
2007				nhslot2->count++;
2008				spin_unlock(&nhslot2->lock);
2009			}
2010
2011			spin_unlock_bh(&hslot->lock);
2012		}
2013	}
2014}
2015EXPORT_SYMBOL(udp_lib_rehash);
2016
2017void udp_v4_rehash(struct sock *sk)
2018{
2019	u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2020					  inet_sk(sk)->inet_rcv_saddr,
2021					  inet_sk(sk)->inet_num);
2022	udp_lib_rehash(sk, new_hash);
2023}
2024
2025static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2026{
2027	int rc;
2028
2029	if (inet_sk(sk)->inet_daddr) {
2030		sock_rps_save_rxhash(sk, skb);
2031		sk_mark_napi_id(sk, skb);
2032		sk_incoming_cpu_update(sk);
2033	} else {
2034		sk_mark_napi_id_once(sk, skb);
2035	}
2036
2037	rc = __udp_enqueue_schedule_skb(sk, skb);
2038	if (rc < 0) {
2039		int is_udplite = IS_UDPLITE(sk);
2040		int drop_reason;
2041
2042		/* Note that an ENOMEM error is charged twice */
2043		if (rc == -ENOMEM) {
2044			UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2045					is_udplite);
2046			drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2047		} else {
2048			UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2049				      is_udplite);
2050			drop_reason = SKB_DROP_REASON_PROTO_MEM;
2051		}
2052		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2053		kfree_skb_reason(skb, drop_reason);
2054		trace_udp_fail_queue_rcv_skb(rc, sk);
2055		return -1;
2056	}
2057
2058	return 0;
2059}
2060
2061/* returns:
2062 *  -1: error
2063 *   0: success
2064 *  >0: "udp encap" protocol resubmission
2065 *
2066 * Note that in the success and error cases, the skb is assumed to
2067 * have either been requeued or freed.
2068 */
2069static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2070{
2071	int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2072	struct udp_sock *up = udp_sk(sk);
2073	int is_udplite = IS_UDPLITE(sk);
2074
2075	/*
2076	 *	Charge it to the socket, dropping if the queue is full.
2077	 */
2078	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2079		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2080		goto drop;
2081	}
2082	nf_reset_ct(skb);
2083
2084	if (static_branch_unlikely(&udp_encap_needed_key) &&
2085	    READ_ONCE(up->encap_type)) {
2086		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2087
2088		/*
2089		 * This is an encapsulation socket so pass the skb to
2090		 * the socket's udp_encap_rcv() hook. Otherwise, just
2091		 * fall through and pass this up the UDP socket.
2092		 * up->encap_rcv() returns the following value:
2093		 * =0 if skb was successfully passed to the encap
2094		 *    handler or was discarded by it.
2095		 * >0 if skb should be passed on to UDP.
2096		 * <0 if skb should be resubmitted as proto -N
2097		 */
2098
2099		/* if we're overly short, let UDP handle it */
2100		encap_rcv = READ_ONCE(up->encap_rcv);
2101		if (encap_rcv) {
2102			int ret;
2103
2104			/* Verify checksum before giving to encap */
2105			if (udp_lib_checksum_complete(skb))
2106				goto csum_error;
2107
2108			ret = encap_rcv(sk, skb);
2109			if (ret <= 0) {
2110				__UDP_INC_STATS(sock_net(sk),
2111						UDP_MIB_INDATAGRAMS,
2112						is_udplite);
2113				return -ret;
2114			}
2115		}
2116
2117		/* FALLTHROUGH -- it's a UDP Packet */
2118	}
2119
2120	/*
2121	 * 	UDP-Lite specific tests, ignored on UDP sockets
2122	 */
2123	if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2124		u16 pcrlen = READ_ONCE(up->pcrlen);
2125
2126		/*
2127		 * MIB statistics other than incrementing the error count are
2128		 * disabled for the following two types of errors: these depend
2129		 * on the application settings, not on the functioning of the
2130		 * protocol stack as such.
2131		 *
2132		 * RFC 3828 here recommends (sec 3.3): "There should also be a
2133		 * way ... to ... at least let the receiving application block
2134		 * delivery of packets with coverage values less than a value
2135		 * provided by the application."
2136		 */
2137		if (pcrlen == 0) {          /* full coverage was set  */
2138			net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2139					    UDP_SKB_CB(skb)->cscov, skb->len);
2140			goto drop;
2141		}
2142		/* The next case involves violating the min. coverage requested
2143		 * by the receiver. This is subtle: if receiver wants x and x is
2144		 * greater than the buffersize/MTU then receiver will complain
2145		 * that it wants x while sender emits packets of smaller size y.
2146		 * Therefore the above ...()->partial_cov statement is essential.
2147		 */
2148		if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2149			net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2150					    UDP_SKB_CB(skb)->cscov, pcrlen);
2151			goto drop;
2152		}
2153	}
2154
2155	prefetch(&sk->sk_rmem_alloc);
2156	if (rcu_access_pointer(sk->sk_filter) &&
2157	    udp_lib_checksum_complete(skb))
2158			goto csum_error;
2159
2160	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2161		drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2162		goto drop;
2163	}
2164
2165	udp_csum_pull_header(skb);
2166
2167	ipv4_pktinfo_prepare(sk, skb, true);
2168	return __udp_queue_rcv_skb(sk, skb);
2169
2170csum_error:
2171	drop_reason = SKB_DROP_REASON_UDP_CSUM;
2172	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2173drop:
2174	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2175	atomic_inc(&sk->sk_drops);
2176	kfree_skb_reason(skb, drop_reason);
2177	return -1;
2178}
2179
2180static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2181{
2182	struct sk_buff *next, *segs;
2183	int ret;
2184
2185	if (likely(!udp_unexpected_gso(sk, skb)))
2186		return udp_queue_rcv_one_skb(sk, skb);
2187
2188	BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2189	__skb_push(skb, -skb_mac_offset(skb));
2190	segs = udp_rcv_segment(sk, skb, true);
2191	skb_list_walk_safe(segs, skb, next) {
 
2192		__skb_pull(skb, skb_transport_offset(skb));
2193
2194		udp_post_segment_fix_csum(skb);
2195		ret = udp_queue_rcv_one_skb(sk, skb);
2196		if (ret > 0)
2197			ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2198	}
2199	return 0;
2200}
2201
2202/* For TCP sockets, sk_rx_dst is protected by socket lock
2203 * For UDP, we use xchg() to guard against concurrent changes.
2204 */
2205bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2206{
2207	struct dst_entry *old;
2208
2209	if (dst_hold_safe(dst)) {
2210		old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2211		dst_release(old);
2212		return old != dst;
2213	}
2214	return false;
2215}
2216EXPORT_SYMBOL(udp_sk_rx_dst_set);
2217
2218/*
2219 *	Multicasts and broadcasts go to each listener.
2220 *
2221 *	Note: called only from the BH handler context.
2222 */
2223static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2224				    struct udphdr  *uh,
2225				    __be32 saddr, __be32 daddr,
2226				    struct udp_table *udptable,
2227				    int proto)
2228{
2229	struct sock *sk, *first = NULL;
2230	unsigned short hnum = ntohs(uh->dest);
2231	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2232	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2233	unsigned int offset = offsetof(typeof(*sk), sk_node);
2234	int dif = skb->dev->ifindex;
2235	int sdif = inet_sdif(skb);
2236	struct hlist_node *node;
2237	struct sk_buff *nskb;
2238
2239	if (use_hash2) {
2240		hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2241			    udptable->mask;
2242		hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2243start_lookup:
2244		hslot = &udptable->hash2[hash2];
2245		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2246	}
2247
2248	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2249		if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2250					 uh->source, saddr, dif, sdif, hnum))
2251			continue;
2252
2253		if (!first) {
2254			first = sk;
2255			continue;
2256		}
2257		nskb = skb_clone(skb, GFP_ATOMIC);
2258
2259		if (unlikely(!nskb)) {
2260			atomic_inc(&sk->sk_drops);
2261			__UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2262					IS_UDPLITE(sk));
2263			__UDP_INC_STATS(net, UDP_MIB_INERRORS,
2264					IS_UDPLITE(sk));
2265			continue;
2266		}
2267		if (udp_queue_rcv_skb(sk, nskb) > 0)
2268			consume_skb(nskb);
2269	}
2270
2271	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
2272	if (use_hash2 && hash2 != hash2_any) {
2273		hash2 = hash2_any;
2274		goto start_lookup;
2275	}
2276
2277	if (first) {
2278		if (udp_queue_rcv_skb(first, skb) > 0)
2279			consume_skb(skb);
2280	} else {
2281		kfree_skb(skb);
2282		__UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2283				proto == IPPROTO_UDPLITE);
2284	}
2285	return 0;
2286}
2287
2288/* Initialize UDP checksum. If exited with zero value (success),
2289 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2290 * Otherwise, csum completion requires checksumming packet body,
2291 * including udp header and folding it to skb->csum.
2292 */
2293static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2294				 int proto)
2295{
2296	int err;
2297
2298	UDP_SKB_CB(skb)->partial_cov = 0;
2299	UDP_SKB_CB(skb)->cscov = skb->len;
2300
2301	if (proto == IPPROTO_UDPLITE) {
2302		err = udplite_checksum_init(skb, uh);
2303		if (err)
2304			return err;
2305
2306		if (UDP_SKB_CB(skb)->partial_cov) {
2307			skb->csum = inet_compute_pseudo(skb, proto);
2308			return 0;
2309		}
2310	}
2311
2312	/* Note, we are only interested in != 0 or == 0, thus the
2313	 * force to int.
2314	 */
2315	err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2316							inet_compute_pseudo);
2317	if (err)
2318		return err;
2319
2320	if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2321		/* If SW calculated the value, we know it's bad */
2322		if (skb->csum_complete_sw)
2323			return 1;
2324
2325		/* HW says the value is bad. Let's validate that.
2326		 * skb->csum is no longer the full packet checksum,
2327		 * so don't treat it as such.
2328		 */
2329		skb_checksum_complete_unset(skb);
2330	}
2331
2332	return 0;
2333}
2334
2335/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2336 * return code conversion for ip layer consumption
2337 */
2338static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2339			       struct udphdr *uh)
2340{
2341	int ret;
2342
2343	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2344		skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2345
2346	ret = udp_queue_rcv_skb(sk, skb);
2347
2348	/* a return value > 0 means to resubmit the input, but
2349	 * it wants the return to be -protocol, or 0
2350	 */
2351	if (ret > 0)
2352		return -ret;
2353	return 0;
2354}
2355
2356/*
2357 *	All we need to do is get the socket, and then do a checksum.
2358 */
2359
2360int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2361		   int proto)
2362{
2363	struct sock *sk;
2364	struct udphdr *uh;
2365	unsigned short ulen;
2366	struct rtable *rt = skb_rtable(skb);
2367	__be32 saddr, daddr;
2368	struct net *net = dev_net(skb->dev);
2369	bool refcounted;
2370	int drop_reason;
2371
2372	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2373
2374	/*
2375	 *  Validate the packet.
2376	 */
2377	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2378		goto drop;		/* No space for header. */
2379
2380	uh   = udp_hdr(skb);
2381	ulen = ntohs(uh->len);
2382	saddr = ip_hdr(skb)->saddr;
2383	daddr = ip_hdr(skb)->daddr;
2384
2385	if (ulen > skb->len)
2386		goto short_packet;
2387
2388	if (proto == IPPROTO_UDP) {
2389		/* UDP validates ulen. */
2390		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2391			goto short_packet;
2392		uh = udp_hdr(skb);
2393	}
2394
2395	if (udp4_csum_init(skb, uh, proto))
2396		goto csum_error;
2397
2398	sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
2399			     &refcounted, udp_ehashfn);
2400	if (IS_ERR(sk))
2401		goto no_sk;
2402
2403	if (sk) {
2404		struct dst_entry *dst = skb_dst(skb);
2405		int ret;
2406
2407		if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2408			udp_sk_rx_dst_set(sk, dst);
2409
2410		ret = udp_unicast_rcv_skb(sk, skb, uh);
2411		if (refcounted)
2412			sock_put(sk);
2413		return ret;
2414	}
2415
2416	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2417		return __udp4_lib_mcast_deliver(net, skb, uh,
2418						saddr, daddr, udptable, proto);
2419
2420	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2421	if (sk)
2422		return udp_unicast_rcv_skb(sk, skb, uh);
2423no_sk:
2424	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2425		goto drop;
2426	nf_reset_ct(skb);
2427
2428	/* No socket. Drop packet silently, if checksum is wrong */
2429	if (udp_lib_checksum_complete(skb))
2430		goto csum_error;
2431
2432	drop_reason = SKB_DROP_REASON_NO_SOCKET;
2433	__UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2434	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2435
2436	/*
2437	 * Hmm.  We got an UDP packet to a port to which we
2438	 * don't wanna listen.  Ignore it.
2439	 */
2440	kfree_skb_reason(skb, drop_reason);
2441	return 0;
2442
2443short_packet:
2444	drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2445	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2446			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2447			    &saddr, ntohs(uh->source),
2448			    ulen, skb->len,
2449			    &daddr, ntohs(uh->dest));
2450	goto drop;
2451
2452csum_error:
2453	/*
2454	 * RFC1122: OK.  Discards the bad packet silently (as far as
2455	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2456	 */
2457	drop_reason = SKB_DROP_REASON_UDP_CSUM;
2458	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2459			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2460			    &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2461			    ulen);
2462	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2463drop:
2464	__UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2465	kfree_skb_reason(skb, drop_reason);
2466	return 0;
2467}
2468
2469/* We can only early demux multicast if there is a single matching socket.
2470 * If more than one socket found returns NULL
2471 */
2472static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2473						  __be16 loc_port, __be32 loc_addr,
2474						  __be16 rmt_port, __be32 rmt_addr,
2475						  int dif, int sdif)
2476{
2477	struct udp_table *udptable = net->ipv4.udp_table;
2478	unsigned short hnum = ntohs(loc_port);
2479	struct sock *sk, *result;
2480	struct udp_hslot *hslot;
2481	unsigned int slot;
2482
2483	slot = udp_hashfn(net, hnum, udptable->mask);
2484	hslot = &udptable->hash[slot];
2485
2486	/* Do not bother scanning a too big list */
2487	if (hslot->count > 10)
2488		return NULL;
2489
2490	result = NULL;
2491	sk_for_each_rcu(sk, &hslot->head) {
2492		if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2493					rmt_port, rmt_addr, dif, sdif, hnum)) {
2494			if (result)
2495				return NULL;
2496			result = sk;
2497		}
2498	}
2499
2500	return result;
2501}
2502
2503/* For unicast we should only early demux connected sockets or we can
2504 * break forwarding setups.  The chains here can be long so only check
2505 * if the first socket is an exact match and if not move on.
2506 */
2507static struct sock *__udp4_lib_demux_lookup(struct net *net,
2508					    __be16 loc_port, __be32 loc_addr,
2509					    __be16 rmt_port, __be32 rmt_addr,
2510					    int dif, int sdif)
2511{
2512	struct udp_table *udptable = net->ipv4.udp_table;
 
 
 
2513	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2514	unsigned short hnum = ntohs(loc_port);
2515	unsigned int hash2, slot2;
2516	struct udp_hslot *hslot2;
2517	__portpair ports;
2518	struct sock *sk;
2519
2520	hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2521	slot2 = hash2 & udptable->mask;
2522	hslot2 = &udptable->hash2[slot2];
2523	ports = INET_COMBINED_PORTS(rmt_port, hnum);
2524
2525	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2526		if (inet_match(net, sk, acookie, ports, dif, sdif))
 
2527			return sk;
2528		/* Only check first socket in chain */
2529		break;
2530	}
2531	return NULL;
2532}
2533
2534int udp_v4_early_demux(struct sk_buff *skb)
2535{
2536	struct net *net = dev_net(skb->dev);
2537	struct in_device *in_dev = NULL;
2538	const struct iphdr *iph;
2539	const struct udphdr *uh;
2540	struct sock *sk = NULL;
2541	struct dst_entry *dst;
2542	int dif = skb->dev->ifindex;
2543	int sdif = inet_sdif(skb);
2544	int ours;
2545
2546	/* validate the packet */
2547	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2548		return 0;
2549
2550	iph = ip_hdr(skb);
2551	uh = udp_hdr(skb);
2552
2553	if (skb->pkt_type == PACKET_MULTICAST) {
2554		in_dev = __in_dev_get_rcu(skb->dev);
2555
2556		if (!in_dev)
2557			return 0;
2558
2559		ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2560				       iph->protocol);
2561		if (!ours)
2562			return 0;
2563
2564		sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2565						   uh->source, iph->saddr,
2566						   dif, sdif);
2567	} else if (skb->pkt_type == PACKET_HOST) {
2568		sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2569					     uh->source, iph->saddr, dif, sdif);
2570	}
2571
2572	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2573		return 0;
2574
2575	skb->sk = sk;
2576	skb->destructor = sock_efree;
2577	dst = rcu_dereference(sk->sk_rx_dst);
2578
2579	if (dst)
2580		dst = dst_check(dst, 0);
2581	if (dst) {
2582		u32 itag = 0;
2583
2584		/* set noref for now.
2585		 * any place which wants to hold dst has to call
2586		 * dst_hold_safe()
2587		 */
2588		skb_dst_set_noref(skb, dst);
2589
2590		/* for unconnected multicast sockets we need to validate
2591		 * the source on each packet
2592		 */
2593		if (!inet_sk(sk)->inet_daddr && in_dev)
2594			return ip_mc_validate_source(skb, iph->daddr,
2595						     iph->saddr,
2596						     iph->tos & IPTOS_RT_MASK,
2597						     skb->dev, in_dev, &itag);
2598	}
2599	return 0;
2600}
2601
2602int udp_rcv(struct sk_buff *skb)
2603{
2604	return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2605}
2606
2607void udp_destroy_sock(struct sock *sk)
2608{
2609	struct udp_sock *up = udp_sk(sk);
2610	bool slow = lock_sock_fast(sk);
2611
2612	/* protects from races with udp_abort() */
2613	sock_set_flag(sk, SOCK_DEAD);
2614	udp_flush_pending_frames(sk);
2615	unlock_sock_fast(sk, slow);
2616	if (static_branch_unlikely(&udp_encap_needed_key)) {
2617		if (up->encap_type) {
2618			void (*encap_destroy)(struct sock *sk);
2619			encap_destroy = READ_ONCE(up->encap_destroy);
2620			if (encap_destroy)
2621				encap_destroy(sk);
2622		}
2623		if (udp_test_bit(ENCAP_ENABLED, sk))
2624			static_branch_dec(&udp_encap_needed_key);
2625	}
2626}
2627
2628static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2629				       struct sock *sk)
2630{
2631#ifdef CONFIG_XFRM
2632	if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2633		if (family == AF_INET)
2634			WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv);
2635		else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2636			WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv);
2637	}
2638#endif
2639}
2640
2641/*
2642 *	Socket option code for UDP
2643 */
2644int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2645		       sockptr_t optval, unsigned int optlen,
2646		       int (*push_pending_frames)(struct sock *))
2647{
2648	struct udp_sock *up = udp_sk(sk);
2649	int val, valbool;
2650	int err = 0;
2651	int is_udplite = IS_UDPLITE(sk);
2652
2653	if (level == SOL_SOCKET) {
2654		err = sk_setsockopt(sk, level, optname, optval, optlen);
2655
2656		if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2657			sockopt_lock_sock(sk);
2658			/* paired with READ_ONCE in udp_rmem_release() */
2659			WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2660			sockopt_release_sock(sk);
2661		}
2662		return err;
2663	}
2664
2665	if (optlen < sizeof(int))
2666		return -EINVAL;
2667
2668	if (copy_from_sockptr(&val, optval, sizeof(val)))
2669		return -EFAULT;
2670
2671	valbool = val ? 1 : 0;
2672
2673	switch (optname) {
2674	case UDP_CORK:
2675		if (val != 0) {
2676			udp_set_bit(CORK, sk);
2677		} else {
2678			udp_clear_bit(CORK, sk);
2679			lock_sock(sk);
2680			push_pending_frames(sk);
2681			release_sock(sk);
2682		}
2683		break;
2684
2685	case UDP_ENCAP:
2686		switch (val) {
2687		case 0:
2688#ifdef CONFIG_XFRM
2689		case UDP_ENCAP_ESPINUDP:
2690			set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk);
2691			fallthrough;
2692		case UDP_ENCAP_ESPINUDP_NON_IKE:
2693#if IS_ENABLED(CONFIG_IPV6)
2694			if (sk->sk_family == AF_INET6)
2695				WRITE_ONCE(up->encap_rcv,
2696					   ipv6_stub->xfrm6_udp_encap_rcv);
2697			else
2698#endif
2699				WRITE_ONCE(up->encap_rcv,
2700					   xfrm4_udp_encap_rcv);
2701#endif
2702			fallthrough;
2703		case UDP_ENCAP_L2TPINUDP:
2704			WRITE_ONCE(up->encap_type, val);
2705			udp_tunnel_encap_enable(sk);
 
 
2706			break;
2707		default:
2708			err = -ENOPROTOOPT;
2709			break;
2710		}
2711		break;
2712
2713	case UDP_NO_CHECK6_TX:
2714		udp_set_no_check6_tx(sk, valbool);
2715		break;
2716
2717	case UDP_NO_CHECK6_RX:
2718		udp_set_no_check6_rx(sk, valbool);
2719		break;
2720
2721	case UDP_SEGMENT:
2722		if (val < 0 || val > USHRT_MAX)
2723			return -EINVAL;
2724		WRITE_ONCE(up->gso_size, val);
2725		break;
2726
2727	case UDP_GRO:
2728
2729		/* when enabling GRO, accept the related GSO packet type */
2730		if (valbool)
2731			udp_tunnel_encap_enable(sk);
2732		udp_assign_bit(GRO_ENABLED, sk, valbool);
2733		udp_assign_bit(ACCEPT_L4, sk, valbool);
2734		set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk);
2735		break;
2736
2737	/*
2738	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
2739	 */
2740	/* The sender sets actual checksum coverage length via this option.
2741	 * The case coverage > packet length is handled by send module. */
2742	case UDPLITE_SEND_CSCOV:
2743		if (!is_udplite)         /* Disable the option on UDP sockets */
2744			return -ENOPROTOOPT;
2745		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2746			val = 8;
2747		else if (val > USHRT_MAX)
2748			val = USHRT_MAX;
2749		WRITE_ONCE(up->pcslen, val);
2750		udp_set_bit(UDPLITE_SEND_CC, sk);
2751		break;
2752
2753	/* The receiver specifies a minimum checksum coverage value. To make
2754	 * sense, this should be set to at least 8 (as done below). If zero is
2755	 * used, this again means full checksum coverage.                     */
2756	case UDPLITE_RECV_CSCOV:
2757		if (!is_udplite)         /* Disable the option on UDP sockets */
2758			return -ENOPROTOOPT;
2759		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2760			val = 8;
2761		else if (val > USHRT_MAX)
2762			val = USHRT_MAX;
2763		WRITE_ONCE(up->pcrlen, val);
2764		udp_set_bit(UDPLITE_RECV_CC, sk);
2765		break;
2766
2767	default:
2768		err = -ENOPROTOOPT;
2769		break;
2770	}
2771
2772	return err;
2773}
2774EXPORT_SYMBOL(udp_lib_setsockopt);
2775
2776int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2777		   unsigned int optlen)
2778{
2779	if (level == SOL_UDP  ||  level == SOL_UDPLITE || level == SOL_SOCKET)
2780		return udp_lib_setsockopt(sk, level, optname,
2781					  optval, optlen,
2782					  udp_push_pending_frames);
2783	return ip_setsockopt(sk, level, optname, optval, optlen);
2784}
2785
 
 
 
 
 
 
 
 
 
 
 
2786int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2787		       char __user *optval, int __user *optlen)
2788{
2789	struct udp_sock *up = udp_sk(sk);
2790	int val, len;
2791
2792	if (get_user(len, optlen))
2793		return -EFAULT;
2794
2795	len = min_t(unsigned int, len, sizeof(int));
2796
2797	if (len < 0)
2798		return -EINVAL;
2799
2800	switch (optname) {
2801	case UDP_CORK:
2802		val = udp_test_bit(CORK, sk);
2803		break;
2804
2805	case UDP_ENCAP:
2806		val = READ_ONCE(up->encap_type);
2807		break;
2808
2809	case UDP_NO_CHECK6_TX:
2810		val = udp_get_no_check6_tx(sk);
2811		break;
2812
2813	case UDP_NO_CHECK6_RX:
2814		val = udp_get_no_check6_rx(sk);
2815		break;
2816
2817	case UDP_SEGMENT:
2818		val = READ_ONCE(up->gso_size);
2819		break;
2820
2821	case UDP_GRO:
2822		val = udp_test_bit(GRO_ENABLED, sk);
2823		break;
2824
2825	/* The following two cannot be changed on UDP sockets, the return is
2826	 * always 0 (which corresponds to the full checksum coverage of UDP). */
2827	case UDPLITE_SEND_CSCOV:
2828		val = READ_ONCE(up->pcslen);
2829		break;
2830
2831	case UDPLITE_RECV_CSCOV:
2832		val = READ_ONCE(up->pcrlen);
2833		break;
2834
2835	default:
2836		return -ENOPROTOOPT;
2837	}
2838
2839	if (put_user(len, optlen))
2840		return -EFAULT;
2841	if (copy_to_user(optval, &val, len))
2842		return -EFAULT;
2843	return 0;
2844}
2845EXPORT_SYMBOL(udp_lib_getsockopt);
2846
2847int udp_getsockopt(struct sock *sk, int level, int optname,
2848		   char __user *optval, int __user *optlen)
2849{
2850	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2851		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2852	return ip_getsockopt(sk, level, optname, optval, optlen);
2853}
2854
 
 
 
 
 
 
 
 
 
2855/**
2856 * 	udp_poll - wait for a UDP event.
2857 *	@file: - file struct
2858 *	@sock: - socket
2859 *	@wait: - poll table
2860 *
2861 *	This is same as datagram poll, except for the special case of
2862 *	blocking sockets. If application is using a blocking fd
2863 *	and a packet with checksum error is in the queue;
2864 *	then it could get return from select indicating data available
2865 *	but then block when reading it. Add special case code
2866 *	to work around these arguably broken applications.
2867 */
2868__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2869{
2870	__poll_t mask = datagram_poll(file, sock, wait);
2871	struct sock *sk = sock->sk;
2872
2873	if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2874		mask |= EPOLLIN | EPOLLRDNORM;
2875
2876	/* Check for false positives due to checksum errors */
2877	if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2878	    !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2879		mask &= ~(EPOLLIN | EPOLLRDNORM);
2880
2881	/* psock ingress_msg queue should not contain any bad checksum frames */
2882	if (sk_is_readable(sk))
2883		mask |= EPOLLIN | EPOLLRDNORM;
2884	return mask;
2885
2886}
2887EXPORT_SYMBOL(udp_poll);
2888
2889int udp_abort(struct sock *sk, int err)
2890{
2891	if (!has_current_bpf_ctx())
2892		lock_sock(sk);
2893
2894	/* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2895	 * with close()
2896	 */
2897	if (sock_flag(sk, SOCK_DEAD))
2898		goto out;
2899
2900	sk->sk_err = err;
2901	sk_error_report(sk);
2902	__udp_disconnect(sk, 0);
2903
2904out:
2905	if (!has_current_bpf_ctx())
2906		release_sock(sk);
2907
2908	return 0;
2909}
2910EXPORT_SYMBOL_GPL(udp_abort);
2911
2912struct proto udp_prot = {
2913	.name			= "UDP",
2914	.owner			= THIS_MODULE,
2915	.close			= udp_lib_close,
2916	.pre_connect		= udp_pre_connect,
2917	.connect		= ip4_datagram_connect,
2918	.disconnect		= udp_disconnect,
2919	.ioctl			= udp_ioctl,
2920	.init			= udp_init_sock,
2921	.destroy		= udp_destroy_sock,
2922	.setsockopt		= udp_setsockopt,
2923	.getsockopt		= udp_getsockopt,
2924	.sendmsg		= udp_sendmsg,
2925	.recvmsg		= udp_recvmsg,
2926	.splice_eof		= udp_splice_eof,
2927	.release_cb		= ip4_datagram_release_cb,
2928	.hash			= udp_lib_hash,
2929	.unhash			= udp_lib_unhash,
2930	.rehash			= udp_v4_rehash,
2931	.get_port		= udp_v4_get_port,
2932	.put_port		= udp_lib_unhash,
2933#ifdef CONFIG_BPF_SYSCALL
2934	.psock_update_sk_prot	= udp_bpf_update_proto,
2935#endif
2936	.memory_allocated	= &udp_memory_allocated,
2937	.per_cpu_fw_alloc	= &udp_memory_per_cpu_fw_alloc,
2938
2939	.sysctl_mem		= sysctl_udp_mem,
2940	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2941	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2942	.obj_size		= sizeof(struct udp_sock),
2943	.h.udp_table		= NULL,
 
 
 
 
2944	.diag_destroy		= udp_abort,
2945};
2946EXPORT_SYMBOL(udp_prot);
2947
2948/* ------------------------------------------------------------------------ */
2949#ifdef CONFIG_PROC_FS
2950
2951static unsigned short seq_file_family(const struct seq_file *seq);
2952static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2953{
2954	unsigned short family = seq_file_family(seq);
2955
2956	/* AF_UNSPEC is used as a match all */
2957	return ((family == AF_UNSPEC || family == sk->sk_family) &&
2958		net_eq(sock_net(sk), seq_file_net(seq)));
2959}
2960
2961#ifdef CONFIG_BPF_SYSCALL
2962static const struct seq_operations bpf_iter_udp_seq_ops;
2963#endif
2964static struct udp_table *udp_get_table_seq(struct seq_file *seq,
2965					   struct net *net)
2966{
2967	const struct udp_seq_afinfo *afinfo;
2968
2969#ifdef CONFIG_BPF_SYSCALL
2970	if (seq->op == &bpf_iter_udp_seq_ops)
2971		return net->ipv4.udp_table;
2972#endif
2973
2974	afinfo = pde_data(file_inode(seq->file));
2975	return afinfo->udp_table ? : net->ipv4.udp_table;
2976}
2977
2978static struct sock *udp_get_first(struct seq_file *seq, int start)
2979{
 
 
2980	struct udp_iter_state *state = seq->private;
2981	struct net *net = seq_file_net(seq);
2982	struct udp_table *udptable;
2983	struct sock *sk;
2984
2985	udptable = udp_get_table_seq(seq, net);
2986
2987	for (state->bucket = start; state->bucket <= udptable->mask;
2988	     ++state->bucket) {
2989		struct udp_hslot *hslot = &udptable->hash[state->bucket];
2990
2991		if (hlist_empty(&hslot->head))
2992			continue;
2993
2994		spin_lock_bh(&hslot->lock);
2995		sk_for_each(sk, &hslot->head) {
2996			if (seq_sk_match(seq, sk))
 
 
2997				goto found;
2998		}
2999		spin_unlock_bh(&hslot->lock);
3000	}
3001	sk = NULL;
3002found:
3003	return sk;
3004}
3005
3006static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3007{
 
3008	struct udp_iter_state *state = seq->private;
3009	struct net *net = seq_file_net(seq);
3010	struct udp_table *udptable;
3011
3012	do {
3013		sk = sk_next(sk);
3014	} while (sk && !seq_sk_match(seq, sk));
3015
3016	if (!sk) {
3017		udptable = udp_get_table_seq(seq, net);
3018
3019		if (state->bucket <= udptable->mask)
3020			spin_unlock_bh(&udptable->hash[state->bucket].lock);
3021
3022		return udp_get_first(seq, state->bucket + 1);
3023	}
3024	return sk;
3025}
3026
3027static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3028{
3029	struct sock *sk = udp_get_first(seq, 0);
3030
3031	if (sk)
3032		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3033			--pos;
3034	return pos ? NULL : sk;
3035}
3036
3037void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3038{
3039	struct udp_iter_state *state = seq->private;
3040	state->bucket = MAX_UDP_PORTS;
3041
3042	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3043}
3044EXPORT_SYMBOL(udp_seq_start);
3045
3046void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3047{
3048	struct sock *sk;
3049
3050	if (v == SEQ_START_TOKEN)
3051		sk = udp_get_idx(seq, 0);
3052	else
3053		sk = udp_get_next(seq, v);
3054
3055	++*pos;
3056	return sk;
3057}
3058EXPORT_SYMBOL(udp_seq_next);
3059
3060void udp_seq_stop(struct seq_file *seq, void *v)
3061{
 
3062	struct udp_iter_state *state = seq->private;
3063	struct udp_table *udptable;
3064
3065	udptable = udp_get_table_seq(seq, seq_file_net(seq));
3066
3067	if (state->bucket <= udptable->mask)
3068		spin_unlock_bh(&udptable->hash[state->bucket].lock);
3069}
3070EXPORT_SYMBOL(udp_seq_stop);
3071
3072/* ------------------------------------------------------------------------ */
3073static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3074		int bucket)
3075{
3076	struct inet_sock *inet = inet_sk(sp);
3077	__be32 dest = inet->inet_daddr;
3078	__be32 src  = inet->inet_rcv_saddr;
3079	__u16 destp	  = ntohs(inet->inet_dport);
3080	__u16 srcp	  = ntohs(inet->inet_sport);
3081
3082	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3083		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3084		bucket, src, srcp, dest, destp, sp->sk_state,
3085		sk_wmem_alloc_get(sp),
3086		udp_rqueue_get(sp),
3087		0, 0L, 0,
3088		from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3089		0, sock_i_ino(sp),
3090		refcount_read(&sp->sk_refcnt), sp,
3091		atomic_read(&sp->sk_drops));
3092}
3093
3094int udp4_seq_show(struct seq_file *seq, void *v)
3095{
3096	seq_setwidth(seq, 127);
3097	if (v == SEQ_START_TOKEN)
3098		seq_puts(seq, "   sl  local_address rem_address   st tx_queue "
3099			   "rx_queue tr tm->when retrnsmt   uid  timeout "
3100			   "inode ref pointer drops");
3101	else {
3102		struct udp_iter_state *state = seq->private;
3103
3104		udp4_format_sock(v, seq, state->bucket);
3105	}
3106	seq_pad(seq, '\n');
3107	return 0;
3108}
3109
3110#ifdef CONFIG_BPF_SYSCALL
3111struct bpf_iter__udp {
3112	__bpf_md_ptr(struct bpf_iter_meta *, meta);
3113	__bpf_md_ptr(struct udp_sock *, udp_sk);
3114	uid_t uid __aligned(8);
3115	int bucket __aligned(8);
3116};
3117
3118struct bpf_udp_iter_state {
3119	struct udp_iter_state state;
3120	unsigned int cur_sk;
3121	unsigned int end_sk;
3122	unsigned int max_sk;
3123	int offset;
3124	struct sock **batch;
3125	bool st_bucket_done;
3126};
3127
3128static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3129				      unsigned int new_batch_sz);
3130static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3131{
3132	struct bpf_udp_iter_state *iter = seq->private;
3133	struct udp_iter_state *state = &iter->state;
3134	struct net *net = seq_file_net(seq);
3135	int resume_bucket, resume_offset;
3136	struct udp_table *udptable;
3137	unsigned int batch_sks = 0;
3138	bool resized = false;
3139	struct sock *sk;
3140
3141	resume_bucket = state->bucket;
3142	resume_offset = iter->offset;
3143
3144	/* The current batch is done, so advance the bucket. */
3145	if (iter->st_bucket_done)
3146		state->bucket++;
3147
3148	udptable = udp_get_table_seq(seq, net);
3149
3150again:
3151	/* New batch for the next bucket.
3152	 * Iterate over the hash table to find a bucket with sockets matching
3153	 * the iterator attributes, and return the first matching socket from
3154	 * the bucket. The remaining matched sockets from the bucket are batched
3155	 * before releasing the bucket lock. This allows BPF programs that are
3156	 * called in seq_show to acquire the bucket lock if needed.
3157	 */
3158	iter->cur_sk = 0;
3159	iter->end_sk = 0;
3160	iter->st_bucket_done = false;
3161	batch_sks = 0;
3162
3163	for (; state->bucket <= udptable->mask; state->bucket++) {
3164		struct udp_hslot *hslot2 = &udptable->hash2[state->bucket];
3165
3166		if (hlist_empty(&hslot2->head))
3167			continue;
3168
3169		iter->offset = 0;
3170		spin_lock_bh(&hslot2->lock);
3171		udp_portaddr_for_each_entry(sk, &hslot2->head) {
3172			if (seq_sk_match(seq, sk)) {
3173				/* Resume from the last iterated socket at the
3174				 * offset in the bucket before iterator was stopped.
3175				 */
3176				if (state->bucket == resume_bucket &&
3177				    iter->offset < resume_offset) {
3178					++iter->offset;
3179					continue;
3180				}
3181				if (iter->end_sk < iter->max_sk) {
3182					sock_hold(sk);
3183					iter->batch[iter->end_sk++] = sk;
3184				}
3185				batch_sks++;
3186			}
3187		}
3188		spin_unlock_bh(&hslot2->lock);
3189
3190		if (iter->end_sk)
3191			break;
3192	}
3193
3194	/* All done: no batch made. */
3195	if (!iter->end_sk)
3196		return NULL;
3197
3198	if (iter->end_sk == batch_sks) {
3199		/* Batching is done for the current bucket; return the first
3200		 * socket to be iterated from the batch.
3201		 */
3202		iter->st_bucket_done = true;
3203		goto done;
3204	}
3205	if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3206		resized = true;
3207		/* After allocating a larger batch, retry one more time to grab
3208		 * the whole bucket.
3209		 */
3210		goto again;
3211	}
3212done:
3213	return iter->batch[0];
3214}
3215
3216static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3217{
3218	struct bpf_udp_iter_state *iter = seq->private;
3219	struct sock *sk;
3220
3221	/* Whenever seq_next() is called, the iter->cur_sk is
3222	 * done with seq_show(), so unref the iter->cur_sk.
3223	 */
3224	if (iter->cur_sk < iter->end_sk) {
3225		sock_put(iter->batch[iter->cur_sk++]);
3226		++iter->offset;
3227	}
3228
3229	/* After updating iter->cur_sk, check if there are more sockets
3230	 * available in the current bucket batch.
3231	 */
3232	if (iter->cur_sk < iter->end_sk)
3233		sk = iter->batch[iter->cur_sk];
3234	else
3235		/* Prepare a new batch. */
3236		sk = bpf_iter_udp_batch(seq);
3237
3238	++*pos;
3239	return sk;
3240}
3241
3242static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3243{
3244	/* bpf iter does not support lseek, so it always
3245	 * continue from where it was stop()-ped.
3246	 */
3247	if (*pos)
3248		return bpf_iter_udp_batch(seq);
3249
3250	return SEQ_START_TOKEN;
3251}
3252
3253static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3254			     struct udp_sock *udp_sk, uid_t uid, int bucket)
3255{
3256	struct bpf_iter__udp ctx;
3257
3258	meta->seq_num--;  /* skip SEQ_START_TOKEN */
3259	ctx.meta = meta;
3260	ctx.udp_sk = udp_sk;
3261	ctx.uid = uid;
3262	ctx.bucket = bucket;
3263	return bpf_iter_run_prog(prog, &ctx);
3264}
3265
3266static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3267{
3268	struct udp_iter_state *state = seq->private;
3269	struct bpf_iter_meta meta;
3270	struct bpf_prog *prog;
3271	struct sock *sk = v;
3272	uid_t uid;
3273	int ret;
3274
3275	if (v == SEQ_START_TOKEN)
3276		return 0;
3277
3278	lock_sock(sk);
3279
3280	if (unlikely(sk_unhashed(sk))) {
3281		ret = SEQ_SKIP;
3282		goto unlock;
3283	}
3284
3285	uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3286	meta.seq = seq;
3287	prog = bpf_iter_get_info(&meta, false);
3288	ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3289
3290unlock:
3291	release_sock(sk);
3292	return ret;
3293}
3294
3295static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3296{
3297	while (iter->cur_sk < iter->end_sk)
3298		sock_put(iter->batch[iter->cur_sk++]);
3299}
3300
3301static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3302{
3303	struct bpf_udp_iter_state *iter = seq->private;
3304	struct bpf_iter_meta meta;
3305	struct bpf_prog *prog;
3306
3307	if (!v) {
3308		meta.seq = seq;
3309		prog = bpf_iter_get_info(&meta, true);
3310		if (prog)
3311			(void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3312	}
3313
3314	if (iter->cur_sk < iter->end_sk) {
3315		bpf_iter_udp_put_batch(iter);
3316		iter->st_bucket_done = false;
3317	}
3318}
3319
3320static const struct seq_operations bpf_iter_udp_seq_ops = {
3321	.start		= bpf_iter_udp_seq_start,
3322	.next		= bpf_iter_udp_seq_next,
3323	.stop		= bpf_iter_udp_seq_stop,
3324	.show		= bpf_iter_udp_seq_show,
3325};
3326#endif
3327
3328static unsigned short seq_file_family(const struct seq_file *seq)
3329{
3330	const struct udp_seq_afinfo *afinfo;
3331
3332#ifdef CONFIG_BPF_SYSCALL
3333	/* BPF iterator: bpf programs to filter sockets. */
3334	if (seq->op == &bpf_iter_udp_seq_ops)
3335		return AF_UNSPEC;
3336#endif
3337
3338	/* Proc fs iterator */
3339	afinfo = pde_data(file_inode(seq->file));
3340	return afinfo->family;
3341}
3342
3343const struct seq_operations udp_seq_ops = {
3344	.start		= udp_seq_start,
3345	.next		= udp_seq_next,
3346	.stop		= udp_seq_stop,
3347	.show		= udp4_seq_show,
3348};
3349EXPORT_SYMBOL(udp_seq_ops);
3350
3351static struct udp_seq_afinfo udp4_seq_afinfo = {
3352	.family		= AF_INET,
3353	.udp_table	= NULL,
3354};
3355
3356static int __net_init udp4_proc_init_net(struct net *net)
3357{
3358	if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3359			sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3360		return -ENOMEM;
3361	return 0;
3362}
3363
3364static void __net_exit udp4_proc_exit_net(struct net *net)
3365{
3366	remove_proc_entry("udp", net->proc_net);
3367}
3368
3369static struct pernet_operations udp4_net_ops = {
3370	.init = udp4_proc_init_net,
3371	.exit = udp4_proc_exit_net,
3372};
3373
3374int __init udp4_proc_init(void)
3375{
3376	return register_pernet_subsys(&udp4_net_ops);
3377}
3378
3379void udp4_proc_exit(void)
3380{
3381	unregister_pernet_subsys(&udp4_net_ops);
3382}
3383#endif /* CONFIG_PROC_FS */
3384
3385static __initdata unsigned long uhash_entries;
3386static int __init set_uhash_entries(char *str)
3387{
3388	ssize_t ret;
3389
3390	if (!str)
3391		return 0;
3392
3393	ret = kstrtoul(str, 0, &uhash_entries);
3394	if (ret)
3395		return 0;
3396
3397	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3398		uhash_entries = UDP_HTABLE_SIZE_MIN;
3399	return 1;
3400}
3401__setup("uhash_entries=", set_uhash_entries);
3402
3403void __init udp_table_init(struct udp_table *table, const char *name)
3404{
3405	unsigned int i;
3406
3407	table->hash = alloc_large_system_hash(name,
3408					      2 * sizeof(struct udp_hslot),
3409					      uhash_entries,
3410					      21, /* one slot per 2 MB */
3411					      0,
3412					      &table->log,
3413					      &table->mask,
3414					      UDP_HTABLE_SIZE_MIN,
3415					      UDP_HTABLE_SIZE_MAX);
3416
3417	table->hash2 = table->hash + (table->mask + 1);
3418	for (i = 0; i <= table->mask; i++) {
3419		INIT_HLIST_HEAD(&table->hash[i].head);
3420		table->hash[i].count = 0;
3421		spin_lock_init(&table->hash[i].lock);
3422	}
3423	for (i = 0; i <= table->mask; i++) {
3424		INIT_HLIST_HEAD(&table->hash2[i].head);
3425		table->hash2[i].count = 0;
3426		spin_lock_init(&table->hash2[i].lock);
3427	}
3428}
3429
3430u32 udp_flow_hashrnd(void)
3431{
3432	static u32 hashrnd __read_mostly;
3433
3434	net_get_random_once(&hashrnd, sizeof(hashrnd));
3435
3436	return hashrnd;
3437}
3438EXPORT_SYMBOL(udp_flow_hashrnd);
3439
3440static void __net_init udp_sysctl_init(struct net *net)
3441{
3442	net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3443	net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3444
3445#ifdef CONFIG_NET_L3_MASTER_DEV
3446	net->ipv4.sysctl_udp_l3mdev_accept = 0;
3447#endif
3448}
3449
3450static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3451{
3452	struct udp_table *udptable;
3453	int i;
3454
3455	udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3456	if (!udptable)
3457		goto out;
3458
3459	udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
3460				      GFP_KERNEL_ACCOUNT);
3461	if (!udptable->hash)
3462		goto free_table;
3463
3464	udptable->hash2 = udptable->hash + hash_entries;
3465	udptable->mask = hash_entries - 1;
3466	udptable->log = ilog2(hash_entries);
3467
3468	for (i = 0; i < hash_entries; i++) {
3469		INIT_HLIST_HEAD(&udptable->hash[i].head);
3470		udptable->hash[i].count = 0;
3471		spin_lock_init(&udptable->hash[i].lock);
3472
3473		INIT_HLIST_HEAD(&udptable->hash2[i].head);
3474		udptable->hash2[i].count = 0;
3475		spin_lock_init(&udptable->hash2[i].lock);
3476	}
3477
3478	return udptable;
3479
3480free_table:
3481	kfree(udptable);
3482out:
3483	return NULL;
3484}
3485
3486static void __net_exit udp_pernet_table_free(struct net *net)
3487{
3488	struct udp_table *udptable = net->ipv4.udp_table;
3489
3490	if (udptable == &udp_table)
3491		return;
3492
3493	kvfree(udptable->hash);
3494	kfree(udptable);
3495}
3496
3497static void __net_init udp_set_table(struct net *net)
3498{
3499	struct udp_table *udptable;
3500	unsigned int hash_entries;
3501	struct net *old_net;
3502
3503	if (net_eq(net, &init_net))
3504		goto fallback;
3505
3506	old_net = current->nsproxy->net_ns;
3507	hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3508	if (!hash_entries)
3509		goto fallback;
3510
3511	/* Set min to keep the bitmap on stack in udp_lib_get_port() */
3512	if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3513		hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3514	else
3515		hash_entries = roundup_pow_of_two(hash_entries);
3516
3517	udptable = udp_pernet_table_alloc(hash_entries);
3518	if (udptable) {
3519		net->ipv4.udp_table = udptable;
3520	} else {
3521		pr_warn("Failed to allocate UDP hash table (entries: %u) "
3522			"for a netns, fallback to the global one\n",
3523			hash_entries);
3524fallback:
3525		net->ipv4.udp_table = &udp_table;
3526	}
3527}
3528
3529static int __net_init udp_pernet_init(struct net *net)
3530{
3531	udp_sysctl_init(net);
3532	udp_set_table(net);
3533
3534	return 0;
3535}
3536
3537static void __net_exit udp_pernet_exit(struct net *net)
3538{
3539	udp_pernet_table_free(net);
3540}
3541
3542static struct pernet_operations __net_initdata udp_sysctl_ops = {
3543	.init	= udp_pernet_init,
3544	.exit	= udp_pernet_exit,
3545};
3546
3547#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3548DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3549		     struct udp_sock *udp_sk, uid_t uid, int bucket)
3550
3551static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3552				      unsigned int new_batch_sz)
3553{
3554	struct sock **new_batch;
3555
3556	new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3557				   GFP_USER | __GFP_NOWARN);
3558	if (!new_batch)
3559		return -ENOMEM;
3560
3561	bpf_iter_udp_put_batch(iter);
3562	kvfree(iter->batch);
3563	iter->batch = new_batch;
3564	iter->max_sk = new_batch_sz;
3565
3566	return 0;
3567}
3568
3569#define INIT_BATCH_SZ 16
3570
3571static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3572{
3573	struct bpf_udp_iter_state *iter = priv_data;
3574	int ret;
3575
3576	ret = bpf_iter_init_seq_net(priv_data, aux);
3577	if (ret)
3578		return ret;
3579
3580	ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3581	if (ret)
3582		bpf_iter_fini_seq_net(priv_data);
3583
3584	return ret;
3585}
3586
3587static void bpf_iter_fini_udp(void *priv_data)
3588{
3589	struct bpf_udp_iter_state *iter = priv_data;
3590
3591	bpf_iter_fini_seq_net(priv_data);
3592	kvfree(iter->batch);
3593}
3594
3595static const struct bpf_iter_seq_info udp_seq_info = {
3596	.seq_ops		= &bpf_iter_udp_seq_ops,
3597	.init_seq_private	= bpf_iter_init_udp,
3598	.fini_seq_private	= bpf_iter_fini_udp,
3599	.seq_priv_size		= sizeof(struct bpf_udp_iter_state),
3600};
3601
3602static struct bpf_iter_reg udp_reg_info = {
3603	.target			= "udp",
3604	.ctx_arg_info_size	= 1,
3605	.ctx_arg_info		= {
3606		{ offsetof(struct bpf_iter__udp, udp_sk),
3607		  PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3608	},
3609	.seq_info		= &udp_seq_info,
3610};
3611
3612static void __init bpf_iter_register(void)
3613{
3614	udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3615	if (bpf_iter_reg_target(&udp_reg_info))
3616		pr_warn("Warning: could not register bpf iterator udp\n");
3617}
3618#endif
3619
3620void __init udp_init(void)
3621{
3622	unsigned long limit;
3623	unsigned int i;
3624
3625	udp_table_init(&udp_table, "UDP");
3626	limit = nr_free_buffer_pages() / 8;
3627	limit = max(limit, 128UL);
3628	sysctl_udp_mem[0] = limit / 4 * 3;
3629	sysctl_udp_mem[1] = limit;
3630	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3631
 
 
3632	/* 16 spinlocks per cpu */
3633	udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3634	udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3635				GFP_KERNEL);
3636	if (!udp_busylocks)
3637		panic("UDP: failed to alloc udp_busylocks\n");
3638	for (i = 0; i < (1U << udp_busylocks_log); i++)
3639		spin_lock_init(udp_busylocks + i);
3640
3641	if (register_pernet_subsys(&udp_sysctl_ops))
3642		panic("UDP: failed to init sysctl parameters.\n");
3643
3644#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3645	bpf_iter_register();
3646#endif
3647}