Linux Audio

Check our new training course

Loading...
v5.14.15
   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 <linux/btf_ids.h>
 110#include <trace/events/skb.h>
 111#include <net/busy_poll.h>
 112#include "udp_impl.h"
 113#include <net/sock_reuseport.h>
 114#include <net/addrconf.h>
 115#include <net/udp_tunnel.h>
 116#if IS_ENABLED(CONFIG_IPV6)
 117#include <net/ipv6_stubs.h>
 118#endif
 119
 120struct udp_table udp_table __read_mostly;
 121EXPORT_SYMBOL(udp_table);
 122
 123long sysctl_udp_mem[3] __read_mostly;
 124EXPORT_SYMBOL(sysctl_udp_mem);
 125
 126atomic_long_t udp_memory_allocated;
 127EXPORT_SYMBOL(udp_memory_allocated);
 128
 129#define MAX_UDP_PORTS 65536
 130#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
 131
 
 
 
 
 
 
 
 
 
 
 
 132static int udp_lib_lport_inuse(struct net *net, __u16 num,
 133			       const struct udp_hslot *hslot,
 134			       unsigned long *bitmap,
 135			       struct sock *sk, unsigned int log)
 136{
 137	struct sock *sk2;
 138	kuid_t uid = sock_i_uid(sk);
 139
 140	sk_for_each(sk2, &hslot->head) {
 141		if (net_eq(sock_net(sk2), net) &&
 142		    sk2 != sk &&
 143		    (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
 144		    (!sk2->sk_reuse || !sk->sk_reuse) &&
 145		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 146		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 147		    inet_rcv_saddr_equal(sk, sk2, true)) {
 148			if (sk2->sk_reuseport && sk->sk_reuseport &&
 149			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 150			    uid_eq(uid, sock_i_uid(sk2))) {
 151				if (!bitmap)
 152					return 0;
 153			} else {
 154				if (!bitmap)
 155					return 1;
 156				__set_bit(udp_sk(sk2)->udp_port_hash >> log,
 157					  bitmap);
 158			}
 159		}
 160	}
 161	return 0;
 162}
 163
 164/*
 165 * Note: we still hold spinlock of primary hash chain, so no other writer
 166 * can insert/delete a socket with local_port == num
 167 */
 168static int udp_lib_lport_inuse2(struct net *net, __u16 num,
 169				struct udp_hslot *hslot2,
 170				struct sock *sk)
 171{
 172	struct sock *sk2;
 173	kuid_t uid = sock_i_uid(sk);
 174	int res = 0;
 175
 176	spin_lock(&hslot2->lock);
 177	udp_portaddr_for_each_entry(sk2, &hslot2->head) {
 178		if (net_eq(sock_net(sk2), net) &&
 179		    sk2 != sk &&
 180		    (udp_sk(sk2)->udp_port_hash == num) &&
 181		    (!sk2->sk_reuse || !sk->sk_reuse) &&
 182		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 183		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 184		    inet_rcv_saddr_equal(sk, sk2, true)) {
 185			if (sk2->sk_reuseport && sk->sk_reuseport &&
 186			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 187			    uid_eq(uid, sock_i_uid(sk2))) {
 188				res = 0;
 189			} else {
 190				res = 1;
 191			}
 192			break;
 193		}
 194	}
 195	spin_unlock(&hslot2->lock);
 196	return res;
 197}
 198
 199static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
 200{
 201	struct net *net = sock_net(sk);
 202	kuid_t uid = sock_i_uid(sk);
 203	struct sock *sk2;
 204
 205	sk_for_each(sk2, &hslot->head) {
 206		if (net_eq(sock_net(sk2), net) &&
 207		    sk2 != sk &&
 208		    sk2->sk_family == sk->sk_family &&
 209		    ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
 210		    (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
 211		    (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 212		    sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
 213		    inet_rcv_saddr_equal(sk, sk2, false)) {
 214			return reuseport_add_sock(sk, sk2,
 215						  inet_rcv_saddr_any(sk));
 216		}
 217	}
 218
 219	return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
 220}
 221
 222/**
 223 *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
 224 *
 225 *  @sk:          socket struct in question
 226 *  @snum:        port number to look up
 227 *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
 228 *                   with NULL address
 229 */
 230int udp_lib_get_port(struct sock *sk, unsigned short snum,
 231		     unsigned int hash2_nulladdr)
 232{
 233	struct udp_hslot *hslot, *hslot2;
 234	struct udp_table *udptable = sk->sk_prot->h.udp_table;
 235	int    error = 1;
 236	struct net *net = sock_net(sk);
 237
 238	if (!snum) {
 239		int low, high, remaining;
 240		unsigned int rand;
 241		unsigned short first, last;
 242		DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
 243
 244		inet_get_local_port_range(net, &low, &high);
 245		remaining = (high - low) + 1;
 246
 247		rand = prandom_u32();
 248		first = reciprocal_scale(rand, remaining) + low;
 249		/*
 250		 * force rand to be an odd multiple of UDP_HTABLE_SIZE
 251		 */
 252		rand = (rand | 1) * (udptable->mask + 1);
 253		last = first + udptable->mask + 1;
 254		do {
 255			hslot = udp_hashslot(udptable, net, first);
 256			bitmap_zero(bitmap, PORTS_PER_CHAIN);
 257			spin_lock_bh(&hslot->lock);
 258			udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
 259					    udptable->log);
 260
 261			snum = first;
 262			/*
 263			 * Iterate on all possible values of snum for this hash.
 264			 * Using steps of an odd multiple of UDP_HTABLE_SIZE
 265			 * give us randomization and full range coverage.
 266			 */
 267			do {
 268				if (low <= snum && snum <= high &&
 269				    !test_bit(snum >> udptable->log, bitmap) &&
 270				    !inet_is_local_reserved_port(net, snum))
 271					goto found;
 272				snum += rand;
 273			} while (snum != first);
 274			spin_unlock_bh(&hslot->lock);
 275			cond_resched();
 276		} while (++first != last);
 277		goto fail;
 278	} else {
 279		hslot = udp_hashslot(udptable, net, snum);
 280		spin_lock_bh(&hslot->lock);
 281		if (hslot->count > 10) {
 282			int exist;
 283			unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
 284
 285			slot2          &= udptable->mask;
 286			hash2_nulladdr &= udptable->mask;
 287
 288			hslot2 = udp_hashslot2(udptable, slot2);
 289			if (hslot->count < hslot2->count)
 290				goto scan_primary_hash;
 291
 292			exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
 293			if (!exist && (hash2_nulladdr != slot2)) {
 294				hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
 295				exist = udp_lib_lport_inuse2(net, snum, hslot2,
 296							     sk);
 297			}
 298			if (exist)
 299				goto fail_unlock;
 300			else
 301				goto found;
 302		}
 303scan_primary_hash:
 304		if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
 305			goto fail_unlock;
 306	}
 307found:
 308	inet_sk(sk)->inet_num = snum;
 309	udp_sk(sk)->udp_port_hash = snum;
 310	udp_sk(sk)->udp_portaddr_hash ^= snum;
 311	if (sk_unhashed(sk)) {
 312		if (sk->sk_reuseport &&
 313		    udp_reuseport_add_sock(sk, hslot)) {
 314			inet_sk(sk)->inet_num = 0;
 315			udp_sk(sk)->udp_port_hash = 0;
 316			udp_sk(sk)->udp_portaddr_hash ^= snum;
 317			goto fail_unlock;
 318		}
 319
 320		sk_add_node_rcu(sk, &hslot->head);
 321		hslot->count++;
 322		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 323
 324		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 325		spin_lock(&hslot2->lock);
 326		if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 327		    sk->sk_family == AF_INET6)
 328			hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
 329					   &hslot2->head);
 330		else
 331			hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
 332					   &hslot2->head);
 333		hslot2->count++;
 334		spin_unlock(&hslot2->lock);
 335	}
 336	sock_set_flag(sk, SOCK_RCU_FREE);
 337	error = 0;
 338fail_unlock:
 339	spin_unlock_bh(&hslot->lock);
 340fail:
 341	return error;
 342}
 343EXPORT_SYMBOL(udp_lib_get_port);
 344
 345int udp_v4_get_port(struct sock *sk, unsigned short snum)
 346{
 347	unsigned int hash2_nulladdr =
 348		ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
 349	unsigned int hash2_partial =
 350		ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
 351
 352	/* precompute partial secondary hash */
 353	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
 354	return udp_lib_get_port(sk, snum, hash2_nulladdr);
 355}
 356
 357static int compute_score(struct sock *sk, struct net *net,
 358			 __be32 saddr, __be16 sport,
 359			 __be32 daddr, unsigned short hnum,
 360			 int dif, int sdif)
 361{
 362	int score;
 363	struct inet_sock *inet;
 364	bool dev_match;
 365
 366	if (!net_eq(sock_net(sk), net) ||
 367	    udp_sk(sk)->udp_port_hash != hnum ||
 368	    ipv6_only_sock(sk))
 369		return -1;
 370
 371	if (sk->sk_rcv_saddr != daddr)
 372		return -1;
 373
 374	score = (sk->sk_family == PF_INET) ? 2 : 1;
 375
 376	inet = inet_sk(sk);
 
 
 
 
 
 
 
 377	if (inet->inet_daddr) {
 378		if (inet->inet_daddr != saddr)
 379			return -1;
 380		score += 4;
 381	}
 382
 383	if (inet->inet_dport) {
 384		if (inet->inet_dport != sport)
 385			return -1;
 386		score += 4;
 387	}
 388
 389	dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
 390					dif, sdif);
 391	if (!dev_match)
 392		return -1;
 393	if (sk->sk_bound_dev_if)
 394		score += 4;
 395
 396	if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
 
 
 
 
 
 
 397		score++;
 398	return score;
 399}
 400
 401static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
 402		       const __u16 lport, const __be32 faddr,
 403		       const __be16 fport)
 404{
 405	static u32 udp_ehash_secret __read_mostly;
 406
 407	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
 408
 409	return __inet_ehashfn(laddr, lport, faddr, fport,
 410			      udp_ehash_secret + net_hash_mix(net));
 411}
 412
 413static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
 414				     struct sk_buff *skb,
 415				     __be32 saddr, __be16 sport,
 416				     __be32 daddr, unsigned short hnum)
 417{
 418	struct sock *reuse_sk = NULL;
 419	u32 hash;
 420
 421	if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
 422		hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
 423		reuse_sk = reuseport_select_sock(sk, hash, skb,
 424						 sizeof(struct udphdr));
 425	}
 426	return reuse_sk;
 427}
 428
 429/* called with rcu_read_lock() */
 430static struct sock *udp4_lib_lookup2(struct net *net,
 431				     __be32 saddr, __be16 sport,
 432				     __be32 daddr, unsigned int hnum,
 433				     int dif, int sdif,
 434				     struct udp_hslot *hslot2,
 435				     struct sk_buff *skb)
 436{
 437	struct sock *sk, *result;
 438	int score, badness;
 
 439
 440	result = NULL;
 441	badness = 0;
 442	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
 443		score = compute_score(sk, net, saddr, sport,
 444				      daddr, hnum, dif, sdif);
 445		if (score > badness) {
 446			result = lookup_reuseport(net, sk, skb,
 447						  saddr, sport, daddr, hnum);
 448			/* Fall back to scoring if group has connections */
 449			if (result && !reuseport_has_conns(sk, false))
 450				return result;
 451
 452			result = result ? : sk;
 
 453			badness = score;
 
 454		}
 455	}
 456	return result;
 457}
 458
 459static struct sock *udp4_lookup_run_bpf(struct net *net,
 460					struct udp_table *udptable,
 461					struct sk_buff *skb,
 462					__be32 saddr, __be16 sport,
 463					__be32 daddr, u16 hnum)
 464{
 465	struct sock *sk, *reuse_sk;
 466	bool no_reuseport;
 467
 468	if (udptable != &udp_table)
 469		return NULL; /* only UDP is supported */
 470
 471	no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP,
 472					    saddr, sport, daddr, hnum, &sk);
 473	if (no_reuseport || IS_ERR_OR_NULL(sk))
 474		return sk;
 475
 476	reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
 477	if (reuse_sk)
 478		sk = reuse_sk;
 479	return sk;
 480}
 481
 482/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
 483 * harder than this. -DaveM
 484 */
 485struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
 486		__be16 sport, __be32 daddr, __be16 dport, int dif,
 487		int sdif, struct udp_table *udptable, struct sk_buff *skb)
 488{
 
 489	unsigned short hnum = ntohs(dport);
 490	unsigned int hash2, slot2;
 491	struct udp_hslot *hslot2;
 492	struct sock *result, *sk;
 493
 494	hash2 = ipv4_portaddr_hash(net, daddr, hnum);
 495	slot2 = hash2 & udptable->mask;
 496	hslot2 = &udptable->hash2[slot2];
 497
 498	/* Lookup connected or non-wildcard socket */
 499	result = udp4_lib_lookup2(net, saddr, sport,
 500				  daddr, hnum, dif, sdif,
 501				  hslot2, skb);
 502	if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
 503		goto done;
 504
 505	/* Lookup redirect from BPF */
 506	if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
 507		sk = udp4_lookup_run_bpf(net, udptable, skb,
 508					 saddr, sport, daddr, hnum);
 509		if (sk) {
 510			result = sk;
 511			goto done;
 512		}
 513	}
 514
 515	/* Got non-wildcard socket or error on first lookup */
 516	if (result)
 517		goto done;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 518
 519	/* Lookup wildcard sockets */
 520	hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
 521	slot2 = hash2 & udptable->mask;
 522	hslot2 = &udptable->hash2[slot2];
 523
 524	result = udp4_lib_lookup2(net, saddr, sport,
 525				  htonl(INADDR_ANY), hnum, dif, sdif,
 526				  hslot2, skb);
 527done:
 528	if (IS_ERR(result))
 529		return NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 530	return result;
 531}
 532EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
 533
 534static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
 535						 __be16 sport, __be16 dport,
 536						 struct udp_table *udptable)
 537{
 538	const struct iphdr *iph = ip_hdr(skb);
 539
 540	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
 541				 iph->daddr, dport, inet_iif(skb),
 542				 inet_sdif(skb), udptable, skb);
 543}
 544
 545struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
 546				 __be16 sport, __be16 dport)
 547{
 548	const struct iphdr *iph = ip_hdr(skb);
 549
 550	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
 551				 iph->daddr, dport, inet_iif(skb),
 552				 inet_sdif(skb), &udp_table, NULL);
 553}
 
 554
 555/* Must be called under rcu_read_lock().
 556 * Does increment socket refcount.
 557 */
 558#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
 
 
 559struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 560			     __be32 daddr, __be16 dport, int dif)
 561{
 562	struct sock *sk;
 563
 564	sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
 565			       dif, 0, &udp_table, NULL);
 566	if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
 567		sk = NULL;
 568	return sk;
 569}
 570EXPORT_SYMBOL_GPL(udp4_lib_lookup);
 571#endif
 572
 573static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
 574				       __be16 loc_port, __be32 loc_addr,
 575				       __be16 rmt_port, __be32 rmt_addr,
 576				       int dif, int sdif, unsigned short hnum)
 577{
 578	struct inet_sock *inet = inet_sk(sk);
 579
 580	if (!net_eq(sock_net(sk), net) ||
 581	    udp_sk(sk)->udp_port_hash != hnum ||
 582	    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
 583	    (inet->inet_dport != rmt_port && inet->inet_dport) ||
 584	    (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
 585	    ipv6_only_sock(sk) ||
 586	    !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
 
 587		return false;
 588	if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
 589		return false;
 590	return true;
 591}
 592
 593DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
 594void udp_encap_enable(void)
 595{
 596	static_branch_inc(&udp_encap_needed_key);
 597}
 598EXPORT_SYMBOL(udp_encap_enable);
 599
 600void udp_encap_disable(void)
 601{
 602	static_branch_dec(&udp_encap_needed_key);
 603}
 604EXPORT_SYMBOL(udp_encap_disable);
 605
 606/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
 607 * through error handlers in encapsulations looking for a match.
 608 */
 609static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
 610{
 611	int i;
 612
 613	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
 614		int (*handler)(struct sk_buff *skb, u32 info);
 615		const struct ip_tunnel_encap_ops *encap;
 616
 617		encap = rcu_dereference(iptun_encaps[i]);
 618		if (!encap)
 619			continue;
 620		handler = encap->err_handler;
 621		if (handler && !handler(skb, info))
 622			return 0;
 623	}
 624
 625	return -ENOENT;
 626}
 627
 628/* Try to match ICMP errors to UDP tunnels by looking up a socket without
 629 * reversing source and destination port: this will match tunnels that force the
 630 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
 631 * lwtunnels might actually break this assumption by being configured with
 632 * different destination ports on endpoints, in this case we won't be able to
 633 * trace ICMP messages back to them.
 634 *
 635 * If this doesn't match any socket, probe tunnels with arbitrary destination
 636 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
 637 * we've sent packets to won't necessarily match the local destination port.
 638 *
 639 * Then ask the tunnel implementation to match the error against a valid
 640 * association.
 641 *
 642 * Return an error if we can't find a match, the socket if we need further
 643 * processing, zero otherwise.
 644 */
 645static struct sock *__udp4_lib_err_encap(struct net *net,
 646					 const struct iphdr *iph,
 647					 struct udphdr *uh,
 648					 struct udp_table *udptable,
 649					 struct sock *sk,
 650					 struct sk_buff *skb, u32 info)
 651{
 652	int (*lookup)(struct sock *sk, struct sk_buff *skb);
 653	int network_offset, transport_offset;
 654	struct udp_sock *up;
 655
 656	network_offset = skb_network_offset(skb);
 657	transport_offset = skb_transport_offset(skb);
 658
 659	/* Network header needs to point to the outer IPv4 header inside ICMP */
 660	skb_reset_network_header(skb);
 661
 662	/* Transport header needs to point to the UDP header */
 663	skb_set_transport_header(skb, iph->ihl << 2);
 664
 665	if (sk) {
 666		up = udp_sk(sk);
 667
 668		lookup = READ_ONCE(up->encap_err_lookup);
 669		if (lookup && lookup(sk, skb))
 670			sk = NULL;
 671
 672		goto out;
 673	}
 674
 675	sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
 676			       iph->saddr, uh->dest, skb->dev->ifindex, 0,
 677			       udptable, NULL);
 678	if (sk) {
 679		up = udp_sk(sk);
 680
 681		lookup = READ_ONCE(up->encap_err_lookup);
 682		if (!lookup || lookup(sk, skb))
 683			sk = NULL;
 684	}
 685
 686out:
 687	if (!sk)
 688		sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
 689
 690	skb_set_transport_header(skb, transport_offset);
 691	skb_set_network_header(skb, network_offset);
 692
 693	return sk;
 694}
 695
 696/*
 697 * This routine is called by the ICMP module when it gets some
 698 * sort of error condition.  If err < 0 then the socket should
 699 * be closed and the error returned to the user.  If err > 0
 700 * it's just the icmp type << 8 | icmp code.
 701 * Header points to the ip header of the error packet. We move
 702 * on past this. Then (as it used to claim before adjustment)
 703 * header points to the first 8 bytes of the udp header.  We need
 704 * to find the appropriate port.
 705 */
 706
 707int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
 708{
 709	struct inet_sock *inet;
 710	const struct iphdr *iph = (const struct iphdr *)skb->data;
 711	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
 712	const int type = icmp_hdr(skb)->type;
 713	const int code = icmp_hdr(skb)->code;
 714	bool tunnel = false;
 715	struct sock *sk;
 716	int harderr;
 717	int err;
 718	struct net *net = dev_net(skb->dev);
 719
 720	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
 721			       iph->saddr, uh->source, skb->dev->ifindex,
 722			       inet_sdif(skb), udptable, NULL);
 723
 724	if (!sk || udp_sk(sk)->encap_type) {
 725		/* No socket for error: try tunnels before discarding */
 726		if (static_branch_unlikely(&udp_encap_needed_key)) {
 727			sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
 728						  info);
 729			if (!sk)
 730				return 0;
 731		} else
 732			sk = ERR_PTR(-ENOENT);
 733
 734		if (IS_ERR(sk)) {
 735			__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 736			return PTR_ERR(sk);
 737		}
 738
 739		tunnel = true;
 740	}
 741
 742	err = 0;
 743	harderr = 0;
 744	inet = inet_sk(sk);
 745
 746	switch (type) {
 747	default:
 748	case ICMP_TIME_EXCEEDED:
 749		err = EHOSTUNREACH;
 750		break;
 751	case ICMP_SOURCE_QUENCH:
 752		goto out;
 753	case ICMP_PARAMETERPROB:
 754		err = EPROTO;
 755		harderr = 1;
 756		break;
 757	case ICMP_DEST_UNREACH:
 758		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
 759			ipv4_sk_update_pmtu(skb, sk, info);
 760			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
 761				err = EMSGSIZE;
 762				harderr = 1;
 763				break;
 764			}
 765			goto out;
 766		}
 767		err = EHOSTUNREACH;
 768		if (code <= NR_ICMP_UNREACH) {
 769			harderr = icmp_err_convert[code].fatal;
 770			err = icmp_err_convert[code].errno;
 771		}
 772		break;
 773	case ICMP_REDIRECT:
 774		ipv4_sk_redirect(skb, sk);
 775		goto out;
 776	}
 777
 778	/*
 779	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
 780	 *	4.1.3.3.
 781	 */
 782	if (tunnel) {
 783		/* ...not for tunnels though: we don't have a sending socket */
 784		goto out;
 785	}
 786	if (!inet->recverr) {
 787		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
 788			goto out;
 789	} else
 790		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
 791
 792	sk->sk_err = err;
 793	sk_error_report(sk);
 794out:
 795	return 0;
 796}
 797
 798int udp_err(struct sk_buff *skb, u32 info)
 799{
 800	return __udp4_lib_err(skb, info, &udp_table);
 801}
 802
 803/*
 804 * Throw away all pending data and cancel the corking. Socket is locked.
 805 */
 806void udp_flush_pending_frames(struct sock *sk)
 807{
 808	struct udp_sock *up = udp_sk(sk);
 809
 810	if (up->pending) {
 811		up->len = 0;
 812		up->pending = 0;
 813		ip_flush_pending_frames(sk);
 814	}
 815}
 816EXPORT_SYMBOL(udp_flush_pending_frames);
 817
 818/**
 819 * 	udp4_hwcsum  -  handle outgoing HW checksumming
 820 * 	@skb: 	sk_buff containing the filled-in UDP header
 821 * 	        (checksum field must be zeroed out)
 822 *	@src:	source IP address
 823 *	@dst:	destination IP address
 824 */
 825void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
 826{
 827	struct udphdr *uh = udp_hdr(skb);
 828	int offset = skb_transport_offset(skb);
 829	int len = skb->len - offset;
 830	int hlen = len;
 831	__wsum csum = 0;
 832
 833	if (!skb_has_frag_list(skb)) {
 834		/*
 835		 * Only one fragment on the socket.
 836		 */
 837		skb->csum_start = skb_transport_header(skb) - skb->head;
 838		skb->csum_offset = offsetof(struct udphdr, check);
 839		uh->check = ~csum_tcpudp_magic(src, dst, len,
 840					       IPPROTO_UDP, 0);
 841	} else {
 842		struct sk_buff *frags;
 843
 844		/*
 845		 * HW-checksum won't work as there are two or more
 846		 * fragments on the socket so that all csums of sk_buffs
 847		 * should be together
 848		 */
 849		skb_walk_frags(skb, frags) {
 850			csum = csum_add(csum, frags->csum);
 851			hlen -= frags->len;
 852		}
 853
 854		csum = skb_checksum(skb, offset, hlen, csum);
 855		skb->ip_summed = CHECKSUM_NONE;
 856
 857		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
 858		if (uh->check == 0)
 859			uh->check = CSUM_MANGLED_0;
 860	}
 861}
 862EXPORT_SYMBOL_GPL(udp4_hwcsum);
 863
 864/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
 865 * for the simple case like when setting the checksum for a UDP tunnel.
 866 */
 867void udp_set_csum(bool nocheck, struct sk_buff *skb,
 868		  __be32 saddr, __be32 daddr, int len)
 869{
 870	struct udphdr *uh = udp_hdr(skb);
 871
 872	if (nocheck) {
 873		uh->check = 0;
 874	} else if (skb_is_gso(skb)) {
 875		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 876	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 877		uh->check = 0;
 878		uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
 879		if (uh->check == 0)
 880			uh->check = CSUM_MANGLED_0;
 881	} else {
 882		skb->ip_summed = CHECKSUM_PARTIAL;
 883		skb->csum_start = skb_transport_header(skb) - skb->head;
 884		skb->csum_offset = offsetof(struct udphdr, check);
 885		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 886	}
 887}
 888EXPORT_SYMBOL(udp_set_csum);
 889
 890static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
 891			struct inet_cork *cork)
 892{
 893	struct sock *sk = skb->sk;
 894	struct inet_sock *inet = inet_sk(sk);
 895	struct udphdr *uh;
 896	int err;
 897	int is_udplite = IS_UDPLITE(sk);
 898	int offset = skb_transport_offset(skb);
 899	int len = skb->len - offset;
 900	int datalen = len - sizeof(*uh);
 901	__wsum csum = 0;
 902
 903	/*
 904	 * Create a UDP header
 905	 */
 906	uh = udp_hdr(skb);
 907	uh->source = inet->inet_sport;
 908	uh->dest = fl4->fl4_dport;
 909	uh->len = htons(len);
 910	uh->check = 0;
 911
 912	if (cork->gso_size) {
 913		const int hlen = skb_network_header_len(skb) +
 914				 sizeof(struct udphdr);
 915
 916		if (hlen + cork->gso_size > cork->fragsize) {
 917			kfree_skb(skb);
 918			return -EINVAL;
 919		}
 920		if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
 921			kfree_skb(skb);
 922			return -EINVAL;
 923		}
 924		if (sk->sk_no_check_tx) {
 925			kfree_skb(skb);
 926			return -EINVAL;
 927		}
 928		if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
 929		    dst_xfrm(skb_dst(skb))) {
 930			kfree_skb(skb);
 931			return -EIO;
 932		}
 933
 934		if (datalen > cork->gso_size) {
 935			skb_shinfo(skb)->gso_size = cork->gso_size;
 936			skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
 937			skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
 938								 cork->gso_size);
 939		}
 940		goto csum_partial;
 941	}
 942
 943	if (is_udplite)  				 /*     UDP-Lite      */
 944		csum = udplite_csum(skb);
 945
 946	else if (sk->sk_no_check_tx) {			 /* UDP csum off */
 947
 948		skb->ip_summed = CHECKSUM_NONE;
 949		goto send;
 950
 951	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
 952csum_partial:
 953
 954		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
 955		goto send;
 956
 957	} else
 958		csum = udp_csum(skb);
 959
 960	/* add protocol-dependent pseudo-header */
 961	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
 962				      sk->sk_protocol, csum);
 963	if (uh->check == 0)
 964		uh->check = CSUM_MANGLED_0;
 965
 966send:
 967	err = ip_send_skb(sock_net(sk), skb);
 968	if (err) {
 969		if (err == -ENOBUFS && !inet->recverr) {
 970			UDP_INC_STATS(sock_net(sk),
 971				      UDP_MIB_SNDBUFERRORS, is_udplite);
 972			err = 0;
 973		}
 974	} else
 975		UDP_INC_STATS(sock_net(sk),
 976			      UDP_MIB_OUTDATAGRAMS, is_udplite);
 977	return err;
 978}
 979
 980/*
 981 * Push out all pending data as one UDP datagram. Socket is locked.
 982 */
 983int udp_push_pending_frames(struct sock *sk)
 984{
 985	struct udp_sock  *up = udp_sk(sk);
 986	struct inet_sock *inet = inet_sk(sk);
 987	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
 988	struct sk_buff *skb;
 989	int err = 0;
 990
 991	skb = ip_finish_skb(sk, fl4);
 992	if (!skb)
 993		goto out;
 994
 995	err = udp_send_skb(skb, fl4, &inet->cork.base);
 996
 997out:
 998	up->len = 0;
 999	up->pending = 0;
1000	return err;
1001}
1002EXPORT_SYMBOL(udp_push_pending_frames);
1003
1004static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1005{
1006	switch (cmsg->cmsg_type) {
1007	case UDP_SEGMENT:
1008		if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1009			return -EINVAL;
1010		*gso_size = *(__u16 *)CMSG_DATA(cmsg);
1011		return 0;
1012	default:
1013		return -EINVAL;
1014	}
1015}
1016
1017int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1018{
1019	struct cmsghdr *cmsg;
1020	bool need_ip = false;
1021	int err;
1022
1023	for_each_cmsghdr(cmsg, msg) {
1024		if (!CMSG_OK(msg, cmsg))
1025			return -EINVAL;
1026
1027		if (cmsg->cmsg_level != SOL_UDP) {
1028			need_ip = true;
1029			continue;
1030		}
1031
1032		err = __udp_cmsg_send(cmsg, gso_size);
1033		if (err)
1034			return err;
1035	}
1036
1037	return need_ip;
1038}
1039EXPORT_SYMBOL_GPL(udp_cmsg_send);
1040
1041int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1042{
1043	struct inet_sock *inet = inet_sk(sk);
1044	struct udp_sock *up = udp_sk(sk);
1045	DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1046	struct flowi4 fl4_stack;
1047	struct flowi4 *fl4;
1048	int ulen = len;
1049	struct ipcm_cookie ipc;
1050	struct rtable *rt = NULL;
1051	int free = 0;
1052	int connected = 0;
1053	__be32 daddr, faddr, saddr;
1054	__be16 dport;
1055	u8  tos;
1056	int err, is_udplite = IS_UDPLITE(sk);
1057	int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1058	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1059	struct sk_buff *skb;
1060	struct ip_options_data opt_copy;
1061
1062	if (len > 0xFFFF)
1063		return -EMSGSIZE;
1064
1065	/*
1066	 *	Check the flags.
1067	 */
1068
1069	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1070		return -EOPNOTSUPP;
1071
 
 
 
 
 
1072	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1073
1074	fl4 = &inet->cork.fl.u.ip4;
1075	if (up->pending) {
1076		/*
1077		 * There are pending frames.
1078		 * The socket lock must be held while it's corked.
1079		 */
1080		lock_sock(sk);
1081		if (likely(up->pending)) {
1082			if (unlikely(up->pending != AF_INET)) {
1083				release_sock(sk);
1084				return -EINVAL;
1085			}
1086			goto do_append_data;
1087		}
1088		release_sock(sk);
1089	}
1090	ulen += sizeof(struct udphdr);
1091
1092	/*
1093	 *	Get and verify the address.
1094	 */
1095	if (usin) {
 
1096		if (msg->msg_namelen < sizeof(*usin))
1097			return -EINVAL;
1098		if (usin->sin_family != AF_INET) {
1099			if (usin->sin_family != AF_UNSPEC)
1100				return -EAFNOSUPPORT;
1101		}
1102
1103		daddr = usin->sin_addr.s_addr;
1104		dport = usin->sin_port;
1105		if (dport == 0)
1106			return -EINVAL;
1107	} else {
1108		if (sk->sk_state != TCP_ESTABLISHED)
1109			return -EDESTADDRREQ;
1110		daddr = inet->inet_daddr;
1111		dport = inet->inet_dport;
1112		/* Open fast path for connected socket.
1113		   Route will not be used, if at least one option is set.
1114		 */
1115		connected = 1;
1116	}
1117
1118	ipcm_init_sk(&ipc, inet);
1119	ipc.gso_size = READ_ONCE(up->gso_size);
 
1120
1121	if (msg->msg_controllen) {
1122		err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1123		if (err > 0)
1124			err = ip_cmsg_send(sk, msg, &ipc,
1125					   sk->sk_family == AF_INET6);
1126		if (unlikely(err < 0)) {
1127			kfree(ipc.opt);
1128			return err;
1129		}
1130		if (ipc.opt)
1131			free = 1;
1132		connected = 0;
1133	}
1134	if (!ipc.opt) {
1135		struct ip_options_rcu *inet_opt;
1136
1137		rcu_read_lock();
1138		inet_opt = rcu_dereference(inet->inet_opt);
1139		if (inet_opt) {
1140			memcpy(&opt_copy, inet_opt,
1141			       sizeof(*inet_opt) + inet_opt->opt.optlen);
1142			ipc.opt = &opt_copy.opt;
1143		}
1144		rcu_read_unlock();
1145	}
1146
1147	if (cgroup_bpf_enabled(BPF_CGROUP_UDP4_SENDMSG) && !connected) {
1148		err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1149					    (struct sockaddr *)usin, &ipc.addr);
1150		if (err)
1151			goto out_free;
1152		if (usin) {
1153			if (usin->sin_port == 0) {
1154				/* BPF program set invalid port. Reject it. */
1155				err = -EINVAL;
1156				goto out_free;
1157			}
1158			daddr = usin->sin_addr.s_addr;
1159			dport = usin->sin_port;
1160		}
1161	}
1162
1163	saddr = ipc.addr;
1164	ipc.addr = faddr = daddr;
1165
 
 
1166	if (ipc.opt && ipc.opt->opt.srr) {
1167		if (!daddr) {
1168			err = -EINVAL;
1169			goto out_free;
1170		}
1171		faddr = ipc.opt->opt.faddr;
1172		connected = 0;
1173	}
1174	tos = get_rttos(&ipc, inet);
1175	if (sock_flag(sk, SOCK_LOCALROUTE) ||
1176	    (msg->msg_flags & MSG_DONTROUTE) ||
1177	    (ipc.opt && ipc.opt->opt.is_strictroute)) {
1178		tos |= RTO_ONLINK;
1179		connected = 0;
1180	}
1181
1182	if (ipv4_is_multicast(daddr)) {
1183		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1184			ipc.oif = inet->mc_index;
1185		if (!saddr)
1186			saddr = inet->mc_addr;
1187		connected = 0;
1188	} else if (!ipc.oif) {
1189		ipc.oif = inet->uc_index;
1190	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1191		/* oif is set, packet is to local broadcast and
1192		 * uc_index is set. oif is most likely set
1193		 * by sk_bound_dev_if. If uc_index != oif check if the
1194		 * oif is an L3 master and uc_index is an L3 slave.
1195		 * If so, we want to allow the send using the uc_index.
1196		 */
1197		if (ipc.oif != inet->uc_index &&
1198		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1199							      inet->uc_index)) {
1200			ipc.oif = inet->uc_index;
1201		}
1202	}
1203
1204	if (connected)
1205		rt = (struct rtable *)sk_dst_check(sk, 0);
1206
1207	if (!rt) {
1208		struct net *net = sock_net(sk);
1209		__u8 flow_flags = inet_sk_flowi_flags(sk);
1210
1211		fl4 = &fl4_stack;
1212
1213		flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1214				   RT_SCOPE_UNIVERSE, sk->sk_protocol,
1215				   flow_flags,
1216				   faddr, saddr, dport, inet->inet_sport,
1217				   sk->sk_uid);
1218
1219		security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1220		rt = ip_route_output_flow(net, fl4, sk);
1221		if (IS_ERR(rt)) {
1222			err = PTR_ERR(rt);
1223			rt = NULL;
1224			if (err == -ENETUNREACH)
1225				IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1226			goto out;
1227		}
1228
1229		err = -EACCES;
1230		if ((rt->rt_flags & RTCF_BROADCAST) &&
1231		    !sock_flag(sk, SOCK_BROADCAST))
1232			goto out;
1233		if (connected)
1234			sk_dst_set(sk, dst_clone(&rt->dst));
1235	}
1236
1237	if (msg->msg_flags&MSG_CONFIRM)
1238		goto do_confirm;
1239back_from_confirm:
1240
1241	saddr = fl4->saddr;
1242	if (!ipc.addr)
1243		daddr = ipc.addr = fl4->daddr;
1244
1245	/* Lockless fast path for the non-corking case. */
1246	if (!corkreq) {
1247		struct inet_cork cork;
1248
1249		skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1250				  sizeof(struct udphdr), &ipc, &rt,
1251				  &cork, msg->msg_flags);
1252		err = PTR_ERR(skb);
1253		if (!IS_ERR_OR_NULL(skb))
1254			err = udp_send_skb(skb, fl4, &cork);
1255		goto out;
1256	}
1257
1258	lock_sock(sk);
1259	if (unlikely(up->pending)) {
1260		/* The socket is already corked while preparing it. */
1261		/* ... which is an evident application bug. --ANK */
1262		release_sock(sk);
1263
1264		net_dbg_ratelimited("socket already corked\n");
1265		err = -EINVAL;
1266		goto out;
1267	}
1268	/*
1269	 *	Now cork the socket to pend data.
1270	 */
1271	fl4 = &inet->cork.fl.u.ip4;
1272	fl4->daddr = daddr;
1273	fl4->saddr = saddr;
1274	fl4->fl4_dport = dport;
1275	fl4->fl4_sport = inet->inet_sport;
1276	up->pending = AF_INET;
1277
1278do_append_data:
1279	up->len += ulen;
1280	err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1281			     sizeof(struct udphdr), &ipc, &rt,
1282			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1283	if (err)
1284		udp_flush_pending_frames(sk);
1285	else if (!corkreq)
1286		err = udp_push_pending_frames(sk);
1287	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1288		up->pending = 0;
1289	release_sock(sk);
1290
1291out:
1292	ip_rt_put(rt);
1293out_free:
1294	if (free)
1295		kfree(ipc.opt);
1296	if (!err)
1297		return len;
1298	/*
1299	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1300	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1301	 * we don't have a good statistic (IpOutDiscards but it can be too many
1302	 * things).  We could add another new stat but at least for now that
1303	 * seems like overkill.
1304	 */
1305	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1306		UDP_INC_STATS(sock_net(sk),
1307			      UDP_MIB_SNDBUFERRORS, is_udplite);
1308	}
1309	return err;
1310
1311do_confirm:
1312	if (msg->msg_flags & MSG_PROBE)
1313		dst_confirm_neigh(&rt->dst, &fl4->daddr);
1314	if (!(msg->msg_flags&MSG_PROBE) || len)
1315		goto back_from_confirm;
1316	err = 0;
1317	goto out;
1318}
1319EXPORT_SYMBOL(udp_sendmsg);
1320
1321int udp_sendpage(struct sock *sk, struct page *page, int offset,
1322		 size_t size, int flags)
1323{
1324	struct inet_sock *inet = inet_sk(sk);
1325	struct udp_sock *up = udp_sk(sk);
1326	int ret;
1327
1328	if (flags & MSG_SENDPAGE_NOTLAST)
1329		flags |= MSG_MORE;
1330
1331	if (!up->pending) {
1332		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
1333
1334		/* Call udp_sendmsg to specify destination address which
1335		 * sendpage interface can't pass.
1336		 * This will succeed only when the socket is connected.
1337		 */
1338		ret = udp_sendmsg(sk, &msg, 0);
1339		if (ret < 0)
1340			return ret;
1341	}
1342
1343	lock_sock(sk);
1344
1345	if (unlikely(!up->pending)) {
1346		release_sock(sk);
1347
1348		net_dbg_ratelimited("cork failed\n");
1349		return -EINVAL;
1350	}
1351
1352	ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1353			     page, offset, size, flags);
1354	if (ret == -EOPNOTSUPP) {
1355		release_sock(sk);
1356		return sock_no_sendpage(sk->sk_socket, page, offset,
1357					size, flags);
1358	}
1359	if (ret < 0) {
1360		udp_flush_pending_frames(sk);
1361		goto out;
1362	}
1363
1364	up->len += size;
1365	if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE)))
1366		ret = udp_push_pending_frames(sk);
1367	if (!ret)
1368		ret = size;
1369out:
1370	release_sock(sk);
1371	return ret;
1372}
1373
1374#define UDP_SKB_IS_STATELESS 0x80000000
1375
1376/* all head states (dst, sk, nf conntrack) except skb extensions are
1377 * cleared by udp_rcv().
1378 *
1379 * We need to preserve secpath, if present, to eventually process
1380 * IP_CMSG_PASSSEC at recvmsg() time.
1381 *
1382 * Other extensions can be cleared.
1383 */
1384static bool udp_try_make_stateless(struct sk_buff *skb)
1385{
1386	if (!skb_has_extensions(skb))
1387		return true;
1388
1389	if (!secpath_exists(skb)) {
1390		skb_ext_reset(skb);
1391		return true;
1392	}
1393
1394	return false;
1395}
1396
1397static void udp_set_dev_scratch(struct sk_buff *skb)
1398{
1399	struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1400
1401	BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1402	scratch->_tsize_state = skb->truesize;
1403#if BITS_PER_LONG == 64
1404	scratch->len = skb->len;
1405	scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1406	scratch->is_linear = !skb_is_nonlinear(skb);
1407#endif
1408	if (udp_try_make_stateless(skb))
1409		scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1410}
1411
1412static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1413{
1414	/* We come here after udp_lib_checksum_complete() returned 0.
1415	 * This means that __skb_checksum_complete() might have
1416	 * set skb->csum_valid to 1.
1417	 * On 64bit platforms, we can set csum_unnecessary
1418	 * to true, but only if the skb is not shared.
1419	 */
1420#if BITS_PER_LONG == 64
1421	if (!skb_shared(skb))
1422		udp_skb_scratch(skb)->csum_unnecessary = true;
1423#endif
1424}
1425
1426static int udp_skb_truesize(struct sk_buff *skb)
1427{
1428	return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1429}
1430
1431static bool udp_skb_has_head_state(struct sk_buff *skb)
1432{
1433	return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1434}
1435
1436/* fully reclaim rmem/fwd memory allocated for skb */
1437static void udp_rmem_release(struct sock *sk, int size, int partial,
1438			     bool rx_queue_lock_held)
1439{
1440	struct udp_sock *up = udp_sk(sk);
1441	struct sk_buff_head *sk_queue;
1442	int amt;
1443
1444	if (likely(partial)) {
1445		up->forward_deficit += size;
1446		size = up->forward_deficit;
1447		if (size < (sk->sk_rcvbuf >> 2) &&
1448		    !skb_queue_empty(&up->reader_queue))
1449			return;
1450	} else {
1451		size += up->forward_deficit;
1452	}
1453	up->forward_deficit = 0;
1454
1455	/* acquire the sk_receive_queue for fwd allocated memory scheduling,
1456	 * if the called don't held it already
1457	 */
1458	sk_queue = &sk->sk_receive_queue;
1459	if (!rx_queue_lock_held)
1460		spin_lock(&sk_queue->lock);
1461
1462
1463	sk->sk_forward_alloc += size;
1464	amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1465	sk->sk_forward_alloc -= amt;
1466
1467	if (amt)
1468		__sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1469
1470	atomic_sub(size, &sk->sk_rmem_alloc);
1471
1472	/* this can save us from acquiring the rx queue lock on next receive */
1473	skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1474
1475	if (!rx_queue_lock_held)
1476		spin_unlock(&sk_queue->lock);
1477}
1478
1479/* Note: called with reader_queue.lock held.
1480 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1481 * This avoids a cache line miss while receive_queue lock is held.
1482 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1483 */
1484void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1485{
1486	prefetch(&skb->data);
1487	udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1488}
1489EXPORT_SYMBOL(udp_skb_destructor);
1490
1491/* as above, but the caller held the rx queue lock, too */
1492static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1493{
1494	prefetch(&skb->data);
1495	udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1496}
1497
1498/* Idea of busylocks is to let producers grab an extra spinlock
1499 * to relieve pressure on the receive_queue spinlock shared by consumer.
1500 * Under flood, this means that only one producer can be in line
1501 * trying to acquire the receive_queue spinlock.
1502 * These busylock can be allocated on a per cpu manner, instead of a
1503 * per socket one (that would consume a cache line per socket)
1504 */
1505static int udp_busylocks_log __read_mostly;
1506static spinlock_t *udp_busylocks __read_mostly;
1507
1508static spinlock_t *busylock_acquire(void *ptr)
1509{
1510	spinlock_t *busy;
1511
1512	busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1513	spin_lock(busy);
1514	return busy;
1515}
1516
1517static void busylock_release(spinlock_t *busy)
1518{
1519	if (busy)
1520		spin_unlock(busy);
1521}
1522
1523int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1524{
1525	struct sk_buff_head *list = &sk->sk_receive_queue;
1526	int rmem, delta, amt, err = -ENOMEM;
1527	spinlock_t *busy = NULL;
1528	int size;
1529
1530	/* try to avoid the costly atomic add/sub pair when the receive
1531	 * queue is full; always allow at least a packet
1532	 */
1533	rmem = atomic_read(&sk->sk_rmem_alloc);
1534	if (rmem > sk->sk_rcvbuf)
1535		goto drop;
1536
1537	/* Under mem pressure, it might be helpful to help udp_recvmsg()
1538	 * having linear skbs :
1539	 * - Reduce memory overhead and thus increase receive queue capacity
1540	 * - Less cache line misses at copyout() time
1541	 * - Less work at consume_skb() (less alien page frag freeing)
1542	 */
1543	if (rmem > (sk->sk_rcvbuf >> 1)) {
1544		skb_condense(skb);
1545
1546		busy = busylock_acquire(sk);
1547	}
1548	size = skb->truesize;
1549	udp_set_dev_scratch(skb);
1550
1551	/* we drop only if the receive buf is full and the receive
1552	 * queue contains some other skb
1553	 */
1554	rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1555	if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1556		goto uncharge_drop;
1557
1558	spin_lock(&list->lock);
1559	if (size >= sk->sk_forward_alloc) {
1560		amt = sk_mem_pages(size);
1561		delta = amt << SK_MEM_QUANTUM_SHIFT;
1562		if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1563			err = -ENOBUFS;
1564			spin_unlock(&list->lock);
1565			goto uncharge_drop;
1566		}
1567
1568		sk->sk_forward_alloc += delta;
1569	}
1570
1571	sk->sk_forward_alloc -= size;
1572
1573	/* no need to setup a destructor, we will explicitly release the
1574	 * forward allocated memory on dequeue
1575	 */
1576	sock_skb_set_dropcount(sk, skb);
1577
1578	__skb_queue_tail(list, skb);
1579	spin_unlock(&list->lock);
1580
1581	if (!sock_flag(sk, SOCK_DEAD))
1582		sk->sk_data_ready(sk);
1583
1584	busylock_release(busy);
1585	return 0;
1586
1587uncharge_drop:
1588	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1589
1590drop:
1591	atomic_inc(&sk->sk_drops);
1592	busylock_release(busy);
1593	return err;
1594}
1595EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1596
1597void udp_destruct_sock(struct sock *sk)
1598{
1599	/* reclaim completely the forward allocated memory */
1600	struct udp_sock *up = udp_sk(sk);
1601	unsigned int total = 0;
1602	struct sk_buff *skb;
1603
1604	skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1605	while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1606		total += skb->truesize;
1607		kfree_skb(skb);
1608	}
1609	udp_rmem_release(sk, total, 0, true);
1610
1611	inet_sock_destruct(sk);
1612}
1613EXPORT_SYMBOL_GPL(udp_destruct_sock);
1614
1615int udp_init_sock(struct sock *sk)
1616{
1617	skb_queue_head_init(&udp_sk(sk)->reader_queue);
1618	sk->sk_destruct = udp_destruct_sock;
1619	return 0;
1620}
1621EXPORT_SYMBOL_GPL(udp_init_sock);
1622
1623void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1624{
1625	if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1626		bool slow = lock_sock_fast(sk);
1627
1628		sk_peek_offset_bwd(sk, len);
1629		unlock_sock_fast(sk, slow);
1630	}
1631
1632	if (!skb_unref(skb))
1633		return;
1634
1635	/* In the more common cases we cleared the head states previously,
1636	 * see __udp_queue_rcv_skb().
1637	 */
1638	if (unlikely(udp_skb_has_head_state(skb)))
1639		skb_release_head_state(skb);
1640	__consume_stateless_skb(skb);
1641}
1642EXPORT_SYMBOL_GPL(skb_consume_udp);
1643
1644static struct sk_buff *__first_packet_length(struct sock *sk,
1645					     struct sk_buff_head *rcvq,
1646					     int *total)
1647{
1648	struct sk_buff *skb;
1649
1650	while ((skb = skb_peek(rcvq)) != NULL) {
1651		if (udp_lib_checksum_complete(skb)) {
1652			__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1653					IS_UDPLITE(sk));
1654			__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1655					IS_UDPLITE(sk));
1656			atomic_inc(&sk->sk_drops);
1657			__skb_unlink(skb, rcvq);
1658			*total += skb->truesize;
1659			kfree_skb(skb);
1660		} else {
1661			udp_skb_csum_unnecessary_set(skb);
 
 
 
1662			break;
1663		}
1664	}
1665	return skb;
1666}
1667
1668/**
1669 *	first_packet_length	- return length of first packet in receive queue
1670 *	@sk: socket
1671 *
1672 *	Drops all bad checksum frames, until a valid one is found.
1673 *	Returns the length of found skb, or -1 if none is found.
1674 */
1675static int first_packet_length(struct sock *sk)
1676{
1677	struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1678	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1679	struct sk_buff *skb;
1680	int total = 0;
1681	int res;
1682
1683	spin_lock_bh(&rcvq->lock);
1684	skb = __first_packet_length(sk, rcvq, &total);
1685	if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1686		spin_lock(&sk_queue->lock);
1687		skb_queue_splice_tail_init(sk_queue, rcvq);
1688		spin_unlock(&sk_queue->lock);
1689
1690		skb = __first_packet_length(sk, rcvq, &total);
1691	}
1692	res = skb ? skb->len : -1;
1693	if (total)
1694		udp_rmem_release(sk, total, 1, false);
1695	spin_unlock_bh(&rcvq->lock);
1696	return res;
1697}
1698
1699/*
1700 *	IOCTL requests applicable to the UDP protocol
1701 */
1702
1703int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1704{
1705	switch (cmd) {
1706	case SIOCOUTQ:
1707	{
1708		int amount = sk_wmem_alloc_get(sk);
1709
1710		return put_user(amount, (int __user *)arg);
1711	}
1712
1713	case SIOCINQ:
1714	{
1715		int amount = max_t(int, 0, first_packet_length(sk));
1716
1717		return put_user(amount, (int __user *)arg);
1718	}
1719
1720	default:
1721		return -ENOIOCTLCMD;
1722	}
1723
1724	return 0;
1725}
1726EXPORT_SYMBOL(udp_ioctl);
1727
1728struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1729			       int noblock, int *off, int *err)
1730{
1731	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1732	struct sk_buff_head *queue;
1733	struct sk_buff *last;
1734	long timeo;
1735	int error;
1736
1737	queue = &udp_sk(sk)->reader_queue;
1738	flags |= noblock ? MSG_DONTWAIT : 0;
1739	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1740	do {
1741		struct sk_buff *skb;
1742
1743		error = sock_error(sk);
1744		if (error)
1745			break;
1746
1747		error = -EAGAIN;
 
1748		do {
1749			spin_lock_bh(&queue->lock);
1750			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1751							err, &last);
 
 
1752			if (skb) {
1753				if (!(flags & MSG_PEEK))
1754					udp_skb_destructor(sk, skb);
1755				spin_unlock_bh(&queue->lock);
1756				return skb;
1757			}
1758
1759			if (skb_queue_empty_lockless(sk_queue)) {
1760				spin_unlock_bh(&queue->lock);
1761				goto busy_check;
1762			}
1763
1764			/* refill the reader queue and walk it again
1765			 * keep both queues locked to avoid re-acquiring
1766			 * the sk_receive_queue lock if fwd memory scheduling
1767			 * is needed.
1768			 */
1769			spin_lock(&sk_queue->lock);
1770			skb_queue_splice_tail_init(sk_queue, queue);
1771
1772			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1773							err, &last);
1774			if (skb && !(flags & MSG_PEEK))
1775				udp_skb_dtor_locked(sk, skb);
1776			spin_unlock(&sk_queue->lock);
1777			spin_unlock_bh(&queue->lock);
1778			if (skb)
1779				return skb;
1780
1781busy_check:
1782			if (!sk_can_busy_loop(sk))
1783				break;
1784
1785			sk_busy_loop(sk, flags & MSG_DONTWAIT);
1786		} while (!skb_queue_empty_lockless(sk_queue));
1787
1788		/* sk_queue is empty, reader_queue may contain peeked packets */
1789	} while (timeo &&
1790		 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1791					      &error, &timeo,
1792					      (struct sk_buff *)sk_queue));
1793
1794	*err = error;
1795	return NULL;
1796}
1797EXPORT_SYMBOL(__skb_recv_udp);
1798
1799int udp_read_sock(struct sock *sk, read_descriptor_t *desc,
1800		  sk_read_actor_t recv_actor)
1801{
1802	int copied = 0;
1803
1804	while (1) {
1805		struct sk_buff *skb;
1806		int err, used;
1807
1808		skb = skb_recv_udp(sk, 0, 1, &err);
1809		if (!skb)
1810			return err;
1811		used = recv_actor(desc, skb, 0, skb->len);
1812		if (used <= 0) {
1813			if (!copied)
1814				copied = used;
1815			kfree_skb(skb);
1816			break;
1817		} else if (used <= skb->len) {
1818			copied += used;
1819		}
1820
1821		kfree_skb(skb);
1822		if (!desc->count)
1823			break;
1824	}
1825
1826	return copied;
1827}
1828EXPORT_SYMBOL(udp_read_sock);
1829
1830/*
1831 * 	This should be easy, if there is something there we
1832 * 	return it, otherwise we block.
1833 */
1834
1835int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1836		int flags, int *addr_len)
1837{
1838	struct inet_sock *inet = inet_sk(sk);
1839	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1840	struct sk_buff *skb;
1841	unsigned int ulen, copied;
1842	int off, err, peeking = flags & MSG_PEEK;
 
1843	int is_udplite = IS_UDPLITE(sk);
1844	bool checksum_valid = false;
1845
1846	if (flags & MSG_ERRQUEUE)
1847		return ip_recv_error(sk, msg, len, addr_len);
1848
1849try_again:
 
1850	off = sk_peek_offset(sk, flags);
1851	skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1852	if (!skb)
1853		return err;
1854
1855	ulen = udp_skb_len(skb);
1856	copied = len;
1857	if (copied > ulen - off)
1858		copied = ulen - off;
1859	else if (copied < ulen)
1860		msg->msg_flags |= MSG_TRUNC;
1861
1862	/*
1863	 * If checksum is needed at all, try to do it while copying the
1864	 * data.  If the data is truncated, or if we only want a partial
1865	 * coverage checksum (UDP-Lite), do it before the copy.
1866	 */
1867
1868	if (copied < ulen || peeking ||
1869	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1870		checksum_valid = udp_skb_csum_unnecessary(skb) ||
1871				!__udp_lib_checksum_complete(skb);
1872		if (!checksum_valid)
1873			goto csum_copy_err;
1874	}
1875
1876	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1877		if (udp_skb_is_linear(skb))
1878			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1879		else
1880			err = skb_copy_datagram_msg(skb, off, msg, copied);
1881	} else {
1882		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1883
1884		if (err == -EINVAL)
1885			goto csum_copy_err;
1886	}
1887
1888	if (unlikely(err)) {
1889		if (!peeking) {
1890			atomic_inc(&sk->sk_drops);
1891			UDP_INC_STATS(sock_net(sk),
1892				      UDP_MIB_INERRORS, is_udplite);
1893		}
1894		kfree_skb(skb);
1895		return err;
1896	}
1897
1898	if (!peeking)
1899		UDP_INC_STATS(sock_net(sk),
1900			      UDP_MIB_INDATAGRAMS, is_udplite);
1901
1902	sock_recv_ts_and_drops(msg, sk, skb);
1903
1904	/* Copy the address. */
1905	if (sin) {
1906		sin->sin_family = AF_INET;
1907		sin->sin_port = udp_hdr(skb)->source;
1908		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1909		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1910		*addr_len = sizeof(*sin);
1911
1912		BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1913						      (struct sockaddr *)sin);
1914	}
1915
1916	if (udp_sk(sk)->gro_enabled)
1917		udp_cmsg_recv(msg, sk, skb);
1918
1919	if (inet->cmsg_flags)
1920		ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1921
1922	err = copied;
1923	if (flags & MSG_TRUNC)
1924		err = ulen;
1925
1926	skb_consume_udp(sk, skb, peeking ? -err : err);
1927	return err;
1928
1929csum_copy_err:
1930	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1931				 udp_skb_destructor)) {
1932		UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1933		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1934	}
1935	kfree_skb(skb);
1936
1937	/* starting over for a new packet, but check if we need to yield */
1938	cond_resched();
1939	msg->msg_flags &= ~MSG_TRUNC;
1940	goto try_again;
1941}
1942
1943int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1944{
1945	/* This check is replicated from __ip4_datagram_connect() and
1946	 * intended to prevent BPF program called below from accessing bytes
1947	 * that are out of the bound specified by user in addr_len.
1948	 */
1949	if (addr_len < sizeof(struct sockaddr_in))
1950		return -EINVAL;
1951
1952	return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1953}
1954EXPORT_SYMBOL(udp_pre_connect);
1955
1956int __udp_disconnect(struct sock *sk, int flags)
1957{
1958	struct inet_sock *inet = inet_sk(sk);
1959	/*
1960	 *	1003.1g - break association.
1961	 */
1962
1963	sk->sk_state = TCP_CLOSE;
1964	inet->inet_daddr = 0;
1965	inet->inet_dport = 0;
1966	sock_rps_reset_rxhash(sk);
1967	sk->sk_bound_dev_if = 0;
1968	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1969		inet_reset_saddr(sk);
1970		if (sk->sk_prot->rehash &&
1971		    (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1972			sk->sk_prot->rehash(sk);
1973	}
1974
1975	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1976		sk->sk_prot->unhash(sk);
1977		inet->inet_sport = 0;
1978	}
1979	sk_dst_reset(sk);
1980	return 0;
1981}
1982EXPORT_SYMBOL(__udp_disconnect);
1983
1984int udp_disconnect(struct sock *sk, int flags)
1985{
1986	lock_sock(sk);
1987	__udp_disconnect(sk, flags);
1988	release_sock(sk);
1989	return 0;
1990}
1991EXPORT_SYMBOL(udp_disconnect);
1992
1993void udp_lib_unhash(struct sock *sk)
1994{
1995	if (sk_hashed(sk)) {
1996		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1997		struct udp_hslot *hslot, *hslot2;
1998
1999		hslot  = udp_hashslot(udptable, sock_net(sk),
2000				      udp_sk(sk)->udp_port_hash);
2001		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2002
2003		spin_lock_bh(&hslot->lock);
2004		if (rcu_access_pointer(sk->sk_reuseport_cb))
2005			reuseport_detach_sock(sk);
2006		if (sk_del_node_init_rcu(sk)) {
2007			hslot->count--;
2008			inet_sk(sk)->inet_num = 0;
2009			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2010
2011			spin_lock(&hslot2->lock);
2012			hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2013			hslot2->count--;
2014			spin_unlock(&hslot2->lock);
2015		}
2016		spin_unlock_bh(&hslot->lock);
2017	}
2018}
2019EXPORT_SYMBOL(udp_lib_unhash);
2020
2021/*
2022 * inet_rcv_saddr was changed, we must rehash secondary hash
2023 */
2024void udp_lib_rehash(struct sock *sk, u16 newhash)
2025{
2026	if (sk_hashed(sk)) {
2027		struct udp_table *udptable = sk->sk_prot->h.udp_table;
2028		struct udp_hslot *hslot, *hslot2, *nhslot2;
2029
2030		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2031		nhslot2 = udp_hashslot2(udptable, newhash);
2032		udp_sk(sk)->udp_portaddr_hash = newhash;
2033
2034		if (hslot2 != nhslot2 ||
2035		    rcu_access_pointer(sk->sk_reuseport_cb)) {
2036			hslot = udp_hashslot(udptable, sock_net(sk),
2037					     udp_sk(sk)->udp_port_hash);
2038			/* we must lock primary chain too */
2039			spin_lock_bh(&hslot->lock);
2040			if (rcu_access_pointer(sk->sk_reuseport_cb))
2041				reuseport_detach_sock(sk);
2042
2043			if (hslot2 != nhslot2) {
2044				spin_lock(&hslot2->lock);
2045				hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2046				hslot2->count--;
2047				spin_unlock(&hslot2->lock);
2048
2049				spin_lock(&nhslot2->lock);
2050				hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2051							 &nhslot2->head);
2052				nhslot2->count++;
2053				spin_unlock(&nhslot2->lock);
2054			}
2055
2056			spin_unlock_bh(&hslot->lock);
2057		}
2058	}
2059}
2060EXPORT_SYMBOL(udp_lib_rehash);
2061
2062void udp_v4_rehash(struct sock *sk)
2063{
2064	u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2065					  inet_sk(sk)->inet_rcv_saddr,
2066					  inet_sk(sk)->inet_num);
2067	udp_lib_rehash(sk, new_hash);
2068}
2069
2070static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2071{
2072	int rc;
2073
2074	if (inet_sk(sk)->inet_daddr) {
2075		sock_rps_save_rxhash(sk, skb);
2076		sk_mark_napi_id(sk, skb);
2077		sk_incoming_cpu_update(sk);
2078	} else {
2079		sk_mark_napi_id_once(sk, skb);
2080	}
2081
2082	rc = __udp_enqueue_schedule_skb(sk, skb);
2083	if (rc < 0) {
2084		int is_udplite = IS_UDPLITE(sk);
2085
2086		/* Note that an ENOMEM error is charged twice */
2087		if (rc == -ENOMEM)
2088			UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2089					is_udplite);
2090		else
2091			UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2092				      is_udplite);
2093		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2094		kfree_skb(skb);
2095		trace_udp_fail_queue_rcv_skb(rc, sk);
2096		return -1;
2097	}
2098
2099	return 0;
2100}
2101
 
 
 
 
 
 
 
2102/* returns:
2103 *  -1: error
2104 *   0: success
2105 *  >0: "udp encap" protocol resubmission
2106 *
2107 * Note that in the success and error cases, the skb is assumed to
2108 * have either been requeued or freed.
2109 */
2110static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2111{
2112	struct udp_sock *up = udp_sk(sk);
2113	int is_udplite = IS_UDPLITE(sk);
2114
2115	/*
2116	 *	Charge it to the socket, dropping if the queue is full.
2117	 */
2118	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2119		goto drop;
2120	nf_reset_ct(skb);
2121
2122	if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2123		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2124
2125		/*
2126		 * This is an encapsulation socket so pass the skb to
2127		 * the socket's udp_encap_rcv() hook. Otherwise, just
2128		 * fall through and pass this up the UDP socket.
2129		 * up->encap_rcv() returns the following value:
2130		 * =0 if skb was successfully passed to the encap
2131		 *    handler or was discarded by it.
2132		 * >0 if skb should be passed on to UDP.
2133		 * <0 if skb should be resubmitted as proto -N
2134		 */
2135
2136		/* if we're overly short, let UDP handle it */
2137		encap_rcv = READ_ONCE(up->encap_rcv);
2138		if (encap_rcv) {
2139			int ret;
2140
2141			/* Verify checksum before giving to encap */
2142			if (udp_lib_checksum_complete(skb))
2143				goto csum_error;
2144
2145			ret = encap_rcv(sk, skb);
2146			if (ret <= 0) {
2147				__UDP_INC_STATS(sock_net(sk),
2148						UDP_MIB_INDATAGRAMS,
2149						is_udplite);
2150				return -ret;
2151			}
2152		}
2153
2154		/* FALLTHROUGH -- it's a UDP Packet */
2155	}
2156
2157	/*
2158	 * 	UDP-Lite specific tests, ignored on UDP sockets
2159	 */
2160	if ((up->pcflag & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
2161
2162		/*
2163		 * MIB statistics other than incrementing the error count are
2164		 * disabled for the following two types of errors: these depend
2165		 * on the application settings, not on the functioning of the
2166		 * protocol stack as such.
2167		 *
2168		 * RFC 3828 here recommends (sec 3.3): "There should also be a
2169		 * way ... to ... at least let the receiving application block
2170		 * delivery of packets with coverage values less than a value
2171		 * provided by the application."
2172		 */
2173		if (up->pcrlen == 0) {          /* full coverage was set  */
2174			net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2175					    UDP_SKB_CB(skb)->cscov, skb->len);
2176			goto drop;
2177		}
2178		/* The next case involves violating the min. coverage requested
2179		 * by the receiver. This is subtle: if receiver wants x and x is
2180		 * greater than the buffersize/MTU then receiver will complain
2181		 * that it wants x while sender emits packets of smaller size y.
2182		 * Therefore the above ...()->partial_cov statement is essential.
2183		 */
2184		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
2185			net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2186					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
2187			goto drop;
2188		}
2189	}
2190
2191	prefetch(&sk->sk_rmem_alloc);
2192	if (rcu_access_pointer(sk->sk_filter) &&
2193	    udp_lib_checksum_complete(skb))
2194			goto csum_error;
2195
2196	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2197		goto drop;
2198
2199	udp_csum_pull_header(skb);
2200
2201	ipv4_pktinfo_prepare(sk, skb);
2202	return __udp_queue_rcv_skb(sk, skb);
2203
2204csum_error:
2205	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2206drop:
2207	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2208	atomic_inc(&sk->sk_drops);
2209	kfree_skb(skb);
2210	return -1;
2211}
2212
2213static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2214{
2215	struct sk_buff *next, *segs;
2216	int ret;
2217
2218	if (likely(!udp_unexpected_gso(sk, skb)))
2219		return udp_queue_rcv_one_skb(sk, skb);
2220
2221	BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2222	__skb_push(skb, -skb_mac_offset(skb));
2223	segs = udp_rcv_segment(sk, skb, true);
2224	skb_list_walk_safe(segs, skb, next) {
2225		__skb_pull(skb, skb_transport_offset(skb));
2226
2227		udp_post_segment_fix_csum(skb);
2228		ret = udp_queue_rcv_one_skb(sk, skb);
2229		if (ret > 0)
2230			ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2231	}
2232	return 0;
2233}
2234
2235/* For TCP sockets, sk_rx_dst is protected by socket lock
2236 * For UDP, we use xchg() to guard against concurrent changes.
2237 */
2238bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2239{
2240	struct dst_entry *old;
2241
2242	if (dst_hold_safe(dst)) {
2243		old = xchg(&sk->sk_rx_dst, dst);
2244		dst_release(old);
2245		return old != dst;
2246	}
2247	return false;
2248}
2249EXPORT_SYMBOL(udp_sk_rx_dst_set);
2250
2251/*
2252 *	Multicasts and broadcasts go to each listener.
2253 *
2254 *	Note: called only from the BH handler context.
2255 */
2256static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2257				    struct udphdr  *uh,
2258				    __be32 saddr, __be32 daddr,
2259				    struct udp_table *udptable,
2260				    int proto)
2261{
2262	struct sock *sk, *first = NULL;
2263	unsigned short hnum = ntohs(uh->dest);
2264	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2265	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2266	unsigned int offset = offsetof(typeof(*sk), sk_node);
2267	int dif = skb->dev->ifindex;
2268	int sdif = inet_sdif(skb);
2269	struct hlist_node *node;
2270	struct sk_buff *nskb;
2271
2272	if (use_hash2) {
2273		hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2274			    udptable->mask;
2275		hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2276start_lookup:
2277		hslot = &udptable->hash2[hash2];
2278		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2279	}
2280
2281	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2282		if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2283					 uh->source, saddr, dif, sdif, hnum))
2284			continue;
2285
2286		if (!first) {
2287			first = sk;
2288			continue;
2289		}
2290		nskb = skb_clone(skb, GFP_ATOMIC);
2291
2292		if (unlikely(!nskb)) {
2293			atomic_inc(&sk->sk_drops);
2294			__UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2295					IS_UDPLITE(sk));
2296			__UDP_INC_STATS(net, UDP_MIB_INERRORS,
2297					IS_UDPLITE(sk));
2298			continue;
2299		}
2300		if (udp_queue_rcv_skb(sk, nskb) > 0)
2301			consume_skb(nskb);
2302	}
2303
2304	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
2305	if (use_hash2 && hash2 != hash2_any) {
2306		hash2 = hash2_any;
2307		goto start_lookup;
2308	}
2309
2310	if (first) {
2311		if (udp_queue_rcv_skb(first, skb) > 0)
2312			consume_skb(skb);
2313	} else {
2314		kfree_skb(skb);
2315		__UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2316				proto == IPPROTO_UDPLITE);
2317	}
2318	return 0;
2319}
2320
2321/* Initialize UDP checksum. If exited with zero value (success),
2322 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2323 * Otherwise, csum completion requires checksumming packet body,
2324 * including udp header and folding it to skb->csum.
2325 */
2326static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2327				 int proto)
2328{
2329	int err;
2330
2331	UDP_SKB_CB(skb)->partial_cov = 0;
2332	UDP_SKB_CB(skb)->cscov = skb->len;
2333
2334	if (proto == IPPROTO_UDPLITE) {
2335		err = udplite_checksum_init(skb, uh);
2336		if (err)
2337			return err;
2338
2339		if (UDP_SKB_CB(skb)->partial_cov) {
2340			skb->csum = inet_compute_pseudo(skb, proto);
2341			return 0;
2342		}
2343	}
2344
2345	/* Note, we are only interested in != 0 or == 0, thus the
2346	 * force to int.
2347	 */
2348	err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2349							inet_compute_pseudo);
2350	if (err)
2351		return err;
2352
2353	if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2354		/* If SW calculated the value, we know it's bad */
2355		if (skb->csum_complete_sw)
2356			return 1;
2357
2358		/* HW says the value is bad. Let's validate that.
2359		 * skb->csum is no longer the full packet checksum,
2360		 * so don't treat it as such.
2361		 */
2362		skb_checksum_complete_unset(skb);
2363	}
2364
2365	return 0;
2366}
2367
2368/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2369 * return code conversion for ip layer consumption
2370 */
2371static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2372			       struct udphdr *uh)
2373{
2374	int ret;
2375
2376	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2377		skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2378
2379	ret = udp_queue_rcv_skb(sk, skb);
2380
2381	/* a return value > 0 means to resubmit the input, but
2382	 * it wants the return to be -protocol, or 0
2383	 */
2384	if (ret > 0)
2385		return -ret;
2386	return 0;
2387}
2388
2389/*
2390 *	All we need to do is get the socket, and then do a checksum.
2391 */
2392
2393int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2394		   int proto)
2395{
2396	struct sock *sk;
2397	struct udphdr *uh;
2398	unsigned short ulen;
2399	struct rtable *rt = skb_rtable(skb);
2400	__be32 saddr, daddr;
2401	struct net *net = dev_net(skb->dev);
2402	bool refcounted;
2403
2404	/*
2405	 *  Validate the packet.
2406	 */
2407	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2408		goto drop;		/* No space for header. */
2409
2410	uh   = udp_hdr(skb);
2411	ulen = ntohs(uh->len);
2412	saddr = ip_hdr(skb)->saddr;
2413	daddr = ip_hdr(skb)->daddr;
2414
2415	if (ulen > skb->len)
2416		goto short_packet;
2417
2418	if (proto == IPPROTO_UDP) {
2419		/* UDP validates ulen. */
2420		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2421			goto short_packet;
2422		uh = udp_hdr(skb);
2423	}
2424
2425	if (udp4_csum_init(skb, uh, proto))
2426		goto csum_error;
2427
2428	sk = skb_steal_sock(skb, &refcounted);
2429	if (sk) {
2430		struct dst_entry *dst = skb_dst(skb);
2431		int ret;
2432
2433		if (unlikely(sk->sk_rx_dst != dst))
2434			udp_sk_rx_dst_set(sk, dst);
2435
2436		ret = udp_unicast_rcv_skb(sk, skb, uh);
2437		if (refcounted)
2438			sock_put(sk);
2439		return ret;
 
 
 
 
2440	}
2441
2442	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2443		return __udp4_lib_mcast_deliver(net, skb, uh,
2444						saddr, daddr, udptable, proto);
2445
2446	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2447	if (sk)
2448		return udp_unicast_rcv_skb(sk, skb, uh);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2449
2450	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2451		goto drop;
2452	nf_reset_ct(skb);
2453
2454	/* No socket. Drop packet silently, if checksum is wrong */
2455	if (udp_lib_checksum_complete(skb))
2456		goto csum_error;
2457
2458	__UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2459	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2460
2461	/*
2462	 * Hmm.  We got an UDP packet to a port to which we
2463	 * don't wanna listen.  Ignore it.
2464	 */
2465	kfree_skb(skb);
2466	return 0;
2467
2468short_packet:
2469	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2470			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2471			    &saddr, ntohs(uh->source),
2472			    ulen, skb->len,
2473			    &daddr, ntohs(uh->dest));
2474	goto drop;
2475
2476csum_error:
2477	/*
2478	 * RFC1122: OK.  Discards the bad packet silently (as far as
2479	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2480	 */
2481	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2482			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2483			    &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2484			    ulen);
2485	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2486drop:
2487	__UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2488	kfree_skb(skb);
2489	return 0;
2490}
2491
2492/* We can only early demux multicast if there is a single matching socket.
2493 * If more than one socket found returns NULL
2494 */
2495static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2496						  __be16 loc_port, __be32 loc_addr,
2497						  __be16 rmt_port, __be32 rmt_addr,
2498						  int dif, int sdif)
2499{
2500	struct sock *sk, *result;
2501	unsigned short hnum = ntohs(loc_port);
2502	unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2503	struct udp_hslot *hslot = &udp_table.hash[slot];
2504
2505	/* Do not bother scanning a too big list */
2506	if (hslot->count > 10)
2507		return NULL;
2508
2509	result = NULL;
2510	sk_for_each_rcu(sk, &hslot->head) {
2511		if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2512					rmt_port, rmt_addr, dif, sdif, hnum)) {
2513			if (result)
2514				return NULL;
2515			result = sk;
2516		}
2517	}
2518
2519	return result;
2520}
2521
2522/* For unicast we should only early demux connected sockets or we can
2523 * break forwarding setups.  The chains here can be long so only check
2524 * if the first socket is an exact match and if not move on.
2525 */
2526static struct sock *__udp4_lib_demux_lookup(struct net *net,
2527					    __be16 loc_port, __be32 loc_addr,
2528					    __be16 rmt_port, __be32 rmt_addr,
2529					    int dif, int sdif)
2530{
2531	unsigned short hnum = ntohs(loc_port);
2532	unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2533	unsigned int slot2 = hash2 & udp_table.mask;
2534	struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2535	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2536	const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2537	struct sock *sk;
2538
2539	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2540		if (INET_MATCH(sk, net, acookie, rmt_addr,
2541			       loc_addr, ports, dif, sdif))
2542			return sk;
2543		/* Only check first socket in chain */
2544		break;
2545	}
2546	return NULL;
2547}
2548
2549int udp_v4_early_demux(struct sk_buff *skb)
2550{
2551	struct net *net = dev_net(skb->dev);
2552	struct in_device *in_dev = NULL;
2553	const struct iphdr *iph;
2554	const struct udphdr *uh;
2555	struct sock *sk = NULL;
2556	struct dst_entry *dst;
2557	int dif = skb->dev->ifindex;
2558	int sdif = inet_sdif(skb);
2559	int ours;
2560
2561	/* validate the packet */
2562	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2563		return 0;
2564
2565	iph = ip_hdr(skb);
2566	uh = udp_hdr(skb);
2567
2568	if (skb->pkt_type == PACKET_MULTICAST) {
2569		in_dev = __in_dev_get_rcu(skb->dev);
2570
2571		if (!in_dev)
2572			return 0;
2573
2574		ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2575				       iph->protocol);
2576		if (!ours)
2577			return 0;
2578
2579		sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2580						   uh->source, iph->saddr,
2581						   dif, sdif);
2582	} else if (skb->pkt_type == PACKET_HOST) {
2583		sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2584					     uh->source, iph->saddr, dif, sdif);
2585	}
2586
2587	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2588		return 0;
2589
2590	skb->sk = sk;
2591	skb->destructor = sock_efree;
2592	dst = READ_ONCE(sk->sk_rx_dst);
2593
2594	if (dst)
2595		dst = dst_check(dst, 0);
2596	if (dst) {
2597		u32 itag = 0;
2598
2599		/* set noref for now.
2600		 * any place which wants to hold dst has to call
2601		 * dst_hold_safe()
2602		 */
2603		skb_dst_set_noref(skb, dst);
2604
2605		/* for unconnected multicast sockets we need to validate
2606		 * the source on each packet
2607		 */
2608		if (!inet_sk(sk)->inet_daddr && in_dev)
2609			return ip_mc_validate_source(skb, iph->daddr,
2610						     iph->saddr,
2611						     iph->tos & IPTOS_RT_MASK,
2612						     skb->dev, in_dev, &itag);
2613	}
2614	return 0;
2615}
2616
2617int udp_rcv(struct sk_buff *skb)
2618{
2619	return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2620}
2621
2622void udp_destroy_sock(struct sock *sk)
2623{
2624	struct udp_sock *up = udp_sk(sk);
2625	bool slow = lock_sock_fast(sk);
2626
2627	/* protects from races with udp_abort() */
2628	sock_set_flag(sk, SOCK_DEAD);
2629	udp_flush_pending_frames(sk);
2630	unlock_sock_fast(sk, slow);
2631	if (static_branch_unlikely(&udp_encap_needed_key)) {
2632		if (up->encap_type) {
2633			void (*encap_destroy)(struct sock *sk);
2634			encap_destroy = READ_ONCE(up->encap_destroy);
2635			if (encap_destroy)
2636				encap_destroy(sk);
2637		}
2638		if (up->encap_enabled)
2639			static_branch_dec(&udp_encap_needed_key);
2640	}
2641}
2642
2643/*
2644 *	Socket option code for UDP
2645 */
2646int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2647		       sockptr_t optval, unsigned int optlen,
2648		       int (*push_pending_frames)(struct sock *))
2649{
2650	struct udp_sock *up = udp_sk(sk);
2651	int val, valbool;
2652	int err = 0;
2653	int is_udplite = IS_UDPLITE(sk);
2654
2655	if (optlen < sizeof(int))
2656		return -EINVAL;
2657
2658	if (copy_from_sockptr(&val, optval, sizeof(val)))
2659		return -EFAULT;
2660
2661	valbool = val ? 1 : 0;
2662
2663	switch (optname) {
2664	case UDP_CORK:
2665		if (val != 0) {
2666			WRITE_ONCE(up->corkflag, 1);
2667		} else {
2668			WRITE_ONCE(up->corkflag, 0);
2669			lock_sock(sk);
2670			push_pending_frames(sk);
2671			release_sock(sk);
2672		}
2673		break;
2674
2675	case UDP_ENCAP:
2676		switch (val) {
2677		case 0:
2678#ifdef CONFIG_XFRM
2679		case UDP_ENCAP_ESPINUDP:
2680		case UDP_ENCAP_ESPINUDP_NON_IKE:
2681#if IS_ENABLED(CONFIG_IPV6)
2682			if (sk->sk_family == AF_INET6)
2683				up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
2684			else
2685#endif
2686				up->encap_rcv = xfrm4_udp_encap_rcv;
2687#endif
2688			fallthrough;
2689		case UDP_ENCAP_L2TPINUDP:
2690			up->encap_type = val;
2691			lock_sock(sk);
2692			udp_tunnel_encap_enable(sk->sk_socket);
2693			release_sock(sk);
2694			break;
2695		default:
2696			err = -ENOPROTOOPT;
2697			break;
2698		}
2699		break;
2700
2701	case UDP_NO_CHECK6_TX:
2702		up->no_check6_tx = valbool;
2703		break;
2704
2705	case UDP_NO_CHECK6_RX:
2706		up->no_check6_rx = valbool;
2707		break;
2708
2709	case UDP_SEGMENT:
2710		if (val < 0 || val > USHRT_MAX)
2711			return -EINVAL;
2712		WRITE_ONCE(up->gso_size, val);
2713		break;
2714
2715	case UDP_GRO:
2716		lock_sock(sk);
2717
2718		/* when enabling GRO, accept the related GSO packet type */
2719		if (valbool)
2720			udp_tunnel_encap_enable(sk->sk_socket);
2721		up->gro_enabled = valbool;
2722		up->accept_udp_l4 = valbool;
2723		release_sock(sk);
2724		break;
2725
2726	/*
2727	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
2728	 */
2729	/* The sender sets actual checksum coverage length via this option.
2730	 * The case coverage > packet length is handled by send module. */
2731	case UDPLITE_SEND_CSCOV:
2732		if (!is_udplite)         /* Disable the option on UDP sockets */
2733			return -ENOPROTOOPT;
2734		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2735			val = 8;
2736		else if (val > USHRT_MAX)
2737			val = USHRT_MAX;
2738		up->pcslen = val;
2739		up->pcflag |= UDPLITE_SEND_CC;
2740		break;
2741
2742	/* The receiver specifies a minimum checksum coverage value. To make
2743	 * sense, this should be set to at least 8 (as done below). If zero is
2744	 * used, this again means full checksum coverage.                     */
2745	case UDPLITE_RECV_CSCOV:
2746		if (!is_udplite)         /* Disable the option on UDP sockets */
2747			return -ENOPROTOOPT;
2748		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2749			val = 8;
2750		else if (val > USHRT_MAX)
2751			val = USHRT_MAX;
2752		up->pcrlen = val;
2753		up->pcflag |= UDPLITE_RECV_CC;
2754		break;
2755
2756	default:
2757		err = -ENOPROTOOPT;
2758		break;
2759	}
2760
2761	return err;
2762}
2763EXPORT_SYMBOL(udp_lib_setsockopt);
2764
2765int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2766		   unsigned int optlen)
2767{
2768	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2769		return udp_lib_setsockopt(sk, level, optname,
2770					  optval, optlen,
2771					  udp_push_pending_frames);
2772	return ip_setsockopt(sk, level, optname, optval, optlen);
2773}
2774
 
 
 
 
 
 
 
 
 
 
 
2775int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2776		       char __user *optval, int __user *optlen)
2777{
2778	struct udp_sock *up = udp_sk(sk);
2779	int val, len;
2780
2781	if (get_user(len, optlen))
2782		return -EFAULT;
2783
2784	len = min_t(unsigned int, len, sizeof(int));
2785
2786	if (len < 0)
2787		return -EINVAL;
2788
2789	switch (optname) {
2790	case UDP_CORK:
2791		val = READ_ONCE(up->corkflag);
2792		break;
2793
2794	case UDP_ENCAP:
2795		val = up->encap_type;
2796		break;
2797
2798	case UDP_NO_CHECK6_TX:
2799		val = up->no_check6_tx;
2800		break;
2801
2802	case UDP_NO_CHECK6_RX:
2803		val = up->no_check6_rx;
2804		break;
2805
2806	case UDP_SEGMENT:
2807		val = READ_ONCE(up->gso_size);
2808		break;
2809
2810	case UDP_GRO:
2811		val = up->gro_enabled;
2812		break;
2813
2814	/* The following two cannot be changed on UDP sockets, the return is
2815	 * always 0 (which corresponds to the full checksum coverage of UDP). */
2816	case UDPLITE_SEND_CSCOV:
2817		val = up->pcslen;
2818		break;
2819
2820	case UDPLITE_RECV_CSCOV:
2821		val = up->pcrlen;
2822		break;
2823
2824	default:
2825		return -ENOPROTOOPT;
2826	}
2827
2828	if (put_user(len, optlen))
2829		return -EFAULT;
2830	if (copy_to_user(optval, &val, len))
2831		return -EFAULT;
2832	return 0;
2833}
2834EXPORT_SYMBOL(udp_lib_getsockopt);
2835
2836int udp_getsockopt(struct sock *sk, int level, int optname,
2837		   char __user *optval, int __user *optlen)
2838{
2839	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2840		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2841	return ip_getsockopt(sk, level, optname, optval, optlen);
2842}
2843
 
 
 
 
 
 
 
 
 
2844/**
2845 * 	udp_poll - wait for a UDP event.
2846 *	@file: - file struct
2847 *	@sock: - socket
2848 *	@wait: - poll table
2849 *
2850 *	This is same as datagram poll, except for the special case of
2851 *	blocking sockets. If application is using a blocking fd
2852 *	and a packet with checksum error is in the queue;
2853 *	then it could get return from select indicating data available
2854 *	but then block when reading it. Add special case code
2855 *	to work around these arguably broken applications.
2856 */
2857__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2858{
2859	__poll_t mask = datagram_poll(file, sock, wait);
2860	struct sock *sk = sock->sk;
2861
2862	if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2863		mask |= EPOLLIN | EPOLLRDNORM;
2864
2865	/* Check for false positives due to checksum errors */
2866	if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2867	    !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2868		mask &= ~(EPOLLIN | EPOLLRDNORM);
2869
2870	return mask;
2871
2872}
2873EXPORT_SYMBOL(udp_poll);
2874
2875int udp_abort(struct sock *sk, int err)
2876{
2877	lock_sock(sk);
2878
2879	/* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2880	 * with close()
2881	 */
2882	if (sock_flag(sk, SOCK_DEAD))
2883		goto out;
2884
2885	sk->sk_err = err;
2886	sk_error_report(sk);
2887	__udp_disconnect(sk, 0);
2888
2889out:
2890	release_sock(sk);
2891
2892	return 0;
2893}
2894EXPORT_SYMBOL_GPL(udp_abort);
2895
2896struct proto udp_prot = {
2897	.name			= "UDP",
2898	.owner			= THIS_MODULE,
2899	.close			= udp_lib_close,
2900	.pre_connect		= udp_pre_connect,
2901	.connect		= ip4_datagram_connect,
2902	.disconnect		= udp_disconnect,
2903	.ioctl			= udp_ioctl,
2904	.init			= udp_init_sock,
2905	.destroy		= udp_destroy_sock,
2906	.setsockopt		= udp_setsockopt,
2907	.getsockopt		= udp_getsockopt,
2908	.sendmsg		= udp_sendmsg,
2909	.recvmsg		= udp_recvmsg,
2910	.sendpage		= udp_sendpage,
2911	.release_cb		= ip4_datagram_release_cb,
2912	.hash			= udp_lib_hash,
2913	.unhash			= udp_lib_unhash,
2914	.rehash			= udp_v4_rehash,
2915	.get_port		= udp_v4_get_port,
2916#ifdef CONFIG_BPF_SYSCALL
2917	.psock_update_sk_prot	= udp_bpf_update_proto,
2918#endif
2919	.memory_allocated	= &udp_memory_allocated,
2920	.sysctl_mem		= sysctl_udp_mem,
2921	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2922	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2923	.obj_size		= sizeof(struct udp_sock),
2924	.h.udp_table		= &udp_table,
 
 
 
 
2925	.diag_destroy		= udp_abort,
2926};
2927EXPORT_SYMBOL(udp_prot);
2928
2929/* ------------------------------------------------------------------------ */
2930#ifdef CONFIG_PROC_FS
2931
2932static struct sock *udp_get_first(struct seq_file *seq, int start)
2933{
2934	struct sock *sk;
2935	struct udp_seq_afinfo *afinfo;
2936	struct udp_iter_state *state = seq->private;
2937	struct net *net = seq_file_net(seq);
2938
2939	if (state->bpf_seq_afinfo)
2940		afinfo = state->bpf_seq_afinfo;
2941	else
2942		afinfo = PDE_DATA(file_inode(seq->file));
2943
2944	for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2945	     ++state->bucket) {
2946		struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2947
2948		if (hlist_empty(&hslot->head))
2949			continue;
2950
2951		spin_lock_bh(&hslot->lock);
2952		sk_for_each(sk, &hslot->head) {
2953			if (!net_eq(sock_net(sk), net))
2954				continue;
2955			if (afinfo->family == AF_UNSPEC ||
2956			    sk->sk_family == afinfo->family)
2957				goto found;
2958		}
2959		spin_unlock_bh(&hslot->lock);
2960	}
2961	sk = NULL;
2962found:
2963	return sk;
2964}
2965
2966static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2967{
2968	struct udp_seq_afinfo *afinfo;
2969	struct udp_iter_state *state = seq->private;
2970	struct net *net = seq_file_net(seq);
2971
2972	if (state->bpf_seq_afinfo)
2973		afinfo = state->bpf_seq_afinfo;
2974	else
2975		afinfo = PDE_DATA(file_inode(seq->file));
2976
2977	do {
2978		sk = sk_next(sk);
2979	} while (sk && (!net_eq(sock_net(sk), net) ||
2980			(afinfo->family != AF_UNSPEC &&
2981			 sk->sk_family != afinfo->family)));
2982
2983	if (!sk) {
2984		if (state->bucket <= afinfo->udp_table->mask)
2985			spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2986		return udp_get_first(seq, state->bucket + 1);
2987	}
2988	return sk;
2989}
2990
2991static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2992{
2993	struct sock *sk = udp_get_first(seq, 0);
2994
2995	if (sk)
2996		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2997			--pos;
2998	return pos ? NULL : sk;
2999}
3000
3001void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3002{
3003	struct udp_iter_state *state = seq->private;
3004	state->bucket = MAX_UDP_PORTS;
3005
3006	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3007}
3008EXPORT_SYMBOL(udp_seq_start);
3009
3010void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3011{
3012	struct sock *sk;
3013
3014	if (v == SEQ_START_TOKEN)
3015		sk = udp_get_idx(seq, 0);
3016	else
3017		sk = udp_get_next(seq, v);
3018
3019	++*pos;
3020	return sk;
3021}
3022EXPORT_SYMBOL(udp_seq_next);
3023
3024void udp_seq_stop(struct seq_file *seq, void *v)
3025{
3026	struct udp_seq_afinfo *afinfo;
3027	struct udp_iter_state *state = seq->private;
3028
3029	if (state->bpf_seq_afinfo)
3030		afinfo = state->bpf_seq_afinfo;
3031	else
3032		afinfo = PDE_DATA(file_inode(seq->file));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3033
3034	if (state->bucket <= afinfo->udp_table->mask)
3035		spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
 
 
 
 
 
 
 
 
 
3036}
3037EXPORT_SYMBOL(udp_seq_stop);
3038
3039/* ------------------------------------------------------------------------ */
3040static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3041		int bucket)
3042{
3043	struct inet_sock *inet = inet_sk(sp);
3044	__be32 dest = inet->inet_daddr;
3045	__be32 src  = inet->inet_rcv_saddr;
3046	__u16 destp	  = ntohs(inet->inet_dport);
3047	__u16 srcp	  = ntohs(inet->inet_sport);
3048
3049	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3050		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3051		bucket, src, srcp, dest, destp, sp->sk_state,
3052		sk_wmem_alloc_get(sp),
3053		udp_rqueue_get(sp),
3054		0, 0L, 0,
3055		from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3056		0, sock_i_ino(sp),
3057		refcount_read(&sp->sk_refcnt), sp,
3058		atomic_read(&sp->sk_drops));
3059}
3060
3061int udp4_seq_show(struct seq_file *seq, void *v)
3062{
3063	seq_setwidth(seq, 127);
3064	if (v == SEQ_START_TOKEN)
3065		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
3066			   "rx_queue tr tm->when retrnsmt   uid  timeout "
3067			   "inode ref pointer drops");
3068	else {
3069		struct udp_iter_state *state = seq->private;
3070
3071		udp4_format_sock(v, seq, state->bucket);
3072	}
3073	seq_pad(seq, '\n');
3074	return 0;
3075}
3076
3077#ifdef CONFIG_BPF_SYSCALL
3078struct bpf_iter__udp {
3079	__bpf_md_ptr(struct bpf_iter_meta *, meta);
3080	__bpf_md_ptr(struct udp_sock *, udp_sk);
3081	uid_t uid __aligned(8);
3082	int bucket __aligned(8);
3083};
3084
3085static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3086			     struct udp_sock *udp_sk, uid_t uid, int bucket)
3087{
3088	struct bpf_iter__udp ctx;
3089
3090	meta->seq_num--;  /* skip SEQ_START_TOKEN */
3091	ctx.meta = meta;
3092	ctx.udp_sk = udp_sk;
3093	ctx.uid = uid;
3094	ctx.bucket = bucket;
3095	return bpf_iter_run_prog(prog, &ctx);
3096}
3097
3098static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3099{
3100	struct udp_iter_state *state = seq->private;
3101	struct bpf_iter_meta meta;
3102	struct bpf_prog *prog;
3103	struct sock *sk = v;
3104	uid_t uid;
3105
3106	if (v == SEQ_START_TOKEN)
3107		return 0;
3108
3109	uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3110	meta.seq = seq;
3111	prog = bpf_iter_get_info(&meta, false);
3112	return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3113}
3114
3115static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3116{
3117	struct bpf_iter_meta meta;
3118	struct bpf_prog *prog;
3119
3120	if (!v) {
3121		meta.seq = seq;
3122		prog = bpf_iter_get_info(&meta, true);
3123		if (prog)
3124			(void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3125	}
3126
3127	udp_seq_stop(seq, v);
3128}
3129
3130static const struct seq_operations bpf_iter_udp_seq_ops = {
3131	.start		= udp_seq_start,
3132	.next		= udp_seq_next,
3133	.stop		= bpf_iter_udp_seq_stop,
3134	.show		= bpf_iter_udp_seq_show,
3135};
3136#endif
3137
3138const struct seq_operations udp_seq_ops = {
3139	.start		= udp_seq_start,
3140	.next		= udp_seq_next,
3141	.stop		= udp_seq_stop,
3142	.show		= udp4_seq_show,
3143};
3144EXPORT_SYMBOL(udp_seq_ops);
3145
 
3146static struct udp_seq_afinfo udp4_seq_afinfo = {
 
3147	.family		= AF_INET,
3148	.udp_table	= &udp_table,
 
 
 
 
3149};
3150
3151static int __net_init udp4_proc_init_net(struct net *net)
3152{
3153	if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3154			sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3155		return -ENOMEM;
3156	return 0;
3157}
3158
3159static void __net_exit udp4_proc_exit_net(struct net *net)
3160{
3161	remove_proc_entry("udp", net->proc_net);
3162}
3163
3164static struct pernet_operations udp4_net_ops = {
3165	.init = udp4_proc_init_net,
3166	.exit = udp4_proc_exit_net,
3167};
3168
3169int __init udp4_proc_init(void)
3170{
3171	return register_pernet_subsys(&udp4_net_ops);
3172}
3173
3174void udp4_proc_exit(void)
3175{
3176	unregister_pernet_subsys(&udp4_net_ops);
3177}
3178#endif /* CONFIG_PROC_FS */
3179
3180static __initdata unsigned long uhash_entries;
3181static int __init set_uhash_entries(char *str)
3182{
3183	ssize_t ret;
3184
3185	if (!str)
3186		return 0;
3187
3188	ret = kstrtoul(str, 0, &uhash_entries);
3189	if (ret)
3190		return 0;
3191
3192	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3193		uhash_entries = UDP_HTABLE_SIZE_MIN;
3194	return 1;
3195}
3196__setup("uhash_entries=", set_uhash_entries);
3197
3198void __init udp_table_init(struct udp_table *table, const char *name)
3199{
3200	unsigned int i;
3201
3202	table->hash = alloc_large_system_hash(name,
3203					      2 * sizeof(struct udp_hslot),
3204					      uhash_entries,
3205					      21, /* one slot per 2 MB */
3206					      0,
3207					      &table->log,
3208					      &table->mask,
3209					      UDP_HTABLE_SIZE_MIN,
3210					      64 * 1024);
3211
3212	table->hash2 = table->hash + (table->mask + 1);
3213	for (i = 0; i <= table->mask; i++) {
3214		INIT_HLIST_HEAD(&table->hash[i].head);
3215		table->hash[i].count = 0;
3216		spin_lock_init(&table->hash[i].lock);
3217	}
3218	for (i = 0; i <= table->mask; i++) {
3219		INIT_HLIST_HEAD(&table->hash2[i].head);
3220		table->hash2[i].count = 0;
3221		spin_lock_init(&table->hash2[i].lock);
3222	}
3223}
3224
3225u32 udp_flow_hashrnd(void)
3226{
3227	static u32 hashrnd __read_mostly;
3228
3229	net_get_random_once(&hashrnd, sizeof(hashrnd));
3230
3231	return hashrnd;
3232}
3233EXPORT_SYMBOL(udp_flow_hashrnd);
3234
3235static void __udp_sysctl_init(struct net *net)
3236{
3237	net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3238	net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3239
3240#ifdef CONFIG_NET_L3_MASTER_DEV
3241	net->ipv4.sysctl_udp_l3mdev_accept = 0;
3242#endif
3243}
3244
3245static int __net_init udp_sysctl_init(struct net *net)
3246{
3247	__udp_sysctl_init(net);
3248	return 0;
3249}
3250
3251static struct pernet_operations __net_initdata udp_sysctl_ops = {
3252	.init	= udp_sysctl_init,
3253};
3254
3255#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3256DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3257		     struct udp_sock *udp_sk, uid_t uid, int bucket)
3258
3259static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3260{
3261	struct udp_iter_state *st = priv_data;
3262	struct udp_seq_afinfo *afinfo;
3263	int ret;
3264
3265	afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3266	if (!afinfo)
3267		return -ENOMEM;
3268
3269	afinfo->family = AF_UNSPEC;
3270	afinfo->udp_table = &udp_table;
3271	st->bpf_seq_afinfo = afinfo;
3272	ret = bpf_iter_init_seq_net(priv_data, aux);
3273	if (ret)
3274		kfree(afinfo);
3275	return ret;
3276}
3277
3278static void bpf_iter_fini_udp(void *priv_data)
3279{
3280	struct udp_iter_state *st = priv_data;
3281
3282	kfree(st->bpf_seq_afinfo);
3283	bpf_iter_fini_seq_net(priv_data);
3284}
3285
3286static const struct bpf_iter_seq_info udp_seq_info = {
3287	.seq_ops		= &bpf_iter_udp_seq_ops,
3288	.init_seq_private	= bpf_iter_init_udp,
3289	.fini_seq_private	= bpf_iter_fini_udp,
3290	.seq_priv_size		= sizeof(struct udp_iter_state),
3291};
3292
3293static struct bpf_iter_reg udp_reg_info = {
3294	.target			= "udp",
3295	.ctx_arg_info_size	= 1,
3296	.ctx_arg_info		= {
3297		{ offsetof(struct bpf_iter__udp, udp_sk),
3298		  PTR_TO_BTF_ID_OR_NULL },
3299	},
3300	.seq_info		= &udp_seq_info,
3301};
3302
3303static void __init bpf_iter_register(void)
3304{
3305	udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3306	if (bpf_iter_reg_target(&udp_reg_info))
3307		pr_warn("Warning: could not register bpf iterator udp\n");
3308}
3309#endif
3310
3311void __init udp_init(void)
3312{
3313	unsigned long limit;
3314	unsigned int i;
3315
3316	udp_table_init(&udp_table, "UDP");
3317	limit = nr_free_buffer_pages() / 8;
3318	limit = max(limit, 128UL);
3319	sysctl_udp_mem[0] = limit / 4 * 3;
3320	sysctl_udp_mem[1] = limit;
3321	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3322
3323	__udp_sysctl_init(&init_net);
3324
3325	/* 16 spinlocks per cpu */
3326	udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3327	udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3328				GFP_KERNEL);
3329	if (!udp_busylocks)
3330		panic("UDP: failed to alloc udp_busylocks\n");
3331	for (i = 0; i < (1U << udp_busylocks_log); i++)
3332		spin_lock_init(udp_busylocks + i);
3333
3334	if (register_pernet_subsys(&udp_sysctl_ops))
3335		panic("UDP: failed to init sysctl parameters.\n");
3336
3337#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3338	bpf_iter_register();
3339#endif
3340}
v4.17
 
   1/*
   2 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   3 *		operating system.  INET is implemented using the  BSD Socket
   4 *		interface as the means of communication with the user level.
   5 *
   6 *		The User Datagram Protocol (UDP).
   7 *
   8 * Authors:	Ross Biro
   9 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  11 *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
  12 *		Hirokazu Takahashi, <taka@valinux.co.jp>
  13 *
  14 * Fixes:
  15 *		Alan Cox	:	verify_area() calls
  16 *		Alan Cox	: 	stopped close while in use off icmp
  17 *					messages. Not a fix but a botch that
  18 *					for udp at least is 'valid'.
  19 *		Alan Cox	:	Fixed icmp handling properly
  20 *		Alan Cox	: 	Correct error for oversized datagrams
  21 *		Alan Cox	:	Tidied select() semantics.
  22 *		Alan Cox	:	udp_err() fixed properly, also now
  23 *					select and read wake correctly on errors
  24 *		Alan Cox	:	udp_send verify_area moved to avoid mem leak
  25 *		Alan Cox	:	UDP can count its memory
  26 *		Alan Cox	:	send to an unknown connection causes
  27 *					an ECONNREFUSED off the icmp, but
  28 *					does NOT close.
  29 *		Alan Cox	:	Switched to new sk_buff handlers. No more backlog!
  30 *		Alan Cox	:	Using generic datagram code. Even smaller and the PEEK
  31 *					bug no longer crashes it.
  32 *		Fred Van Kempen	: 	Net2e support for sk->broadcast.
  33 *		Alan Cox	:	Uses skb_free_datagram
  34 *		Alan Cox	:	Added get/set sockopt support.
  35 *		Alan Cox	:	Broadcasting without option set returns EACCES.
  36 *		Alan Cox	:	No wakeup calls. Instead we now use the callbacks.
  37 *		Alan Cox	:	Use ip_tos and ip_ttl
  38 *		Alan Cox	:	SNMP Mibs
  39 *		Alan Cox	:	MSG_DONTROUTE, and 0.0.0.0 support.
  40 *		Matt Dillon	:	UDP length checks.
  41 *		Alan Cox	:	Smarter af_inet used properly.
  42 *		Alan Cox	:	Use new kernel side addressing.
  43 *		Alan Cox	:	Incorrect return on truncated datagram receive.
  44 *	Arnt Gulbrandsen 	:	New udp_send and stuff
  45 *		Alan Cox	:	Cache last socket
  46 *		Alan Cox	:	Route cache
  47 *		Jon Peatfield	:	Minor efficiency fix to sendto().
  48 *		Mike Shaver	:	RFC1122 checks.
  49 *		Alan Cox	:	Nonblocking error fix.
  50 *	Willy Konynenberg	:	Transparent proxying support.
  51 *		Mike McLagan	:	Routing by source
  52 *		David S. Miller	:	New socket lookup architecture.
  53 *					Last socket cache retained as it
  54 *					does have a high hit rate.
  55 *		Olaf Kirch	:	Don't linearise iovec on sendmsg.
  56 *		Andi Kleen	:	Some cleanups, cache destination entry
  57 *					for connect.
  58 *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
  59 *		Melvin Smith	:	Check msg_name not msg_namelen in sendto(),
  60 *					return ENOTCONN for unconnected sockets (POSIX)
  61 *		Janos Farkas	:	don't deliver multi/broadcasts to a different
  62 *					bound-to-device socket
  63 *	Hirokazu Takahashi	:	HW checksumming for outgoing UDP
  64 *					datagrams.
  65 *	Hirokazu Takahashi	:	sendfile() on UDP works now.
  66 *		Arnaldo C. Melo :	convert /proc/net/udp to seq_file
  67 *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
  68 *	Alexey Kuznetsov:		allow both IPv4 and IPv6 sockets to bind
  69 *					a single port at the same time.
  70 *	Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
  71 *	James Chapman		:	Add L2TP encapsulation type.
  72 *
  73 *
  74 *		This program is free software; you can redistribute it and/or
  75 *		modify it under the terms of the GNU General Public License
  76 *		as published by the Free Software Foundation; either version
  77 *		2 of the License, or (at your option) any later version.
  78 */
  79
  80#define pr_fmt(fmt) "UDP: " fmt
  81
  82#include <linux/uaccess.h>
  83#include <asm/ioctls.h>
  84#include <linux/bootmem.h>
  85#include <linux/highmem.h>
  86#include <linux/swap.h>
  87#include <linux/types.h>
  88#include <linux/fcntl.h>
  89#include <linux/module.h>
  90#include <linux/socket.h>
  91#include <linux/sockios.h>
  92#include <linux/igmp.h>
  93#include <linux/inetdevice.h>
  94#include <linux/in.h>
  95#include <linux/errno.h>
  96#include <linux/timer.h>
  97#include <linux/mm.h>
  98#include <linux/inet.h>
  99#include <linux/netdevice.h>
 100#include <linux/slab.h>
 101#include <net/tcp_states.h>
 102#include <linux/skbuff.h>
 103#include <linux/proc_fs.h>
 104#include <linux/seq_file.h>
 105#include <net/net_namespace.h>
 106#include <net/icmp.h>
 107#include <net/inet_hashtables.h>
 
 108#include <net/route.h>
 109#include <net/checksum.h>
 110#include <net/xfrm.h>
 111#include <trace/events/udp.h>
 112#include <linux/static_key.h>
 
 113#include <trace/events/skb.h>
 114#include <net/busy_poll.h>
 115#include "udp_impl.h"
 116#include <net/sock_reuseport.h>
 117#include <net/addrconf.h>
 
 
 
 
 118
 119struct udp_table udp_table __read_mostly;
 120EXPORT_SYMBOL(udp_table);
 121
 122long sysctl_udp_mem[3] __read_mostly;
 123EXPORT_SYMBOL(sysctl_udp_mem);
 124
 125atomic_long_t udp_memory_allocated;
 126EXPORT_SYMBOL(udp_memory_allocated);
 127
 128#define MAX_UDP_PORTS 65536
 129#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
 130
 131/* IPCB reference means this can not be used from early demux */
 132static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
 133{
 134#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
 135	if (!net->ipv4.sysctl_udp_l3mdev_accept &&
 136	    skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
 137		return true;
 138#endif
 139	return false;
 140}
 141
 142static int udp_lib_lport_inuse(struct net *net, __u16 num,
 143			       const struct udp_hslot *hslot,
 144			       unsigned long *bitmap,
 145			       struct sock *sk, unsigned int log)
 146{
 147	struct sock *sk2;
 148	kuid_t uid = sock_i_uid(sk);
 149
 150	sk_for_each(sk2, &hslot->head) {
 151		if (net_eq(sock_net(sk2), net) &&
 152		    sk2 != sk &&
 153		    (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
 154		    (!sk2->sk_reuse || !sk->sk_reuse) &&
 155		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 156		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 157		    inet_rcv_saddr_equal(sk, sk2, true)) {
 158			if (sk2->sk_reuseport && sk->sk_reuseport &&
 159			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 160			    uid_eq(uid, sock_i_uid(sk2))) {
 161				if (!bitmap)
 162					return 0;
 163			} else {
 164				if (!bitmap)
 165					return 1;
 166				__set_bit(udp_sk(sk2)->udp_port_hash >> log,
 167					  bitmap);
 168			}
 169		}
 170	}
 171	return 0;
 172}
 173
 174/*
 175 * Note: we still hold spinlock of primary hash chain, so no other writer
 176 * can insert/delete a socket with local_port == num
 177 */
 178static int udp_lib_lport_inuse2(struct net *net, __u16 num,
 179				struct udp_hslot *hslot2,
 180				struct sock *sk)
 181{
 182	struct sock *sk2;
 183	kuid_t uid = sock_i_uid(sk);
 184	int res = 0;
 185
 186	spin_lock(&hslot2->lock);
 187	udp_portaddr_for_each_entry(sk2, &hslot2->head) {
 188		if (net_eq(sock_net(sk2), net) &&
 189		    sk2 != sk &&
 190		    (udp_sk(sk2)->udp_port_hash == num) &&
 191		    (!sk2->sk_reuse || !sk->sk_reuse) &&
 192		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 193		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 194		    inet_rcv_saddr_equal(sk, sk2, true)) {
 195			if (sk2->sk_reuseport && sk->sk_reuseport &&
 196			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 197			    uid_eq(uid, sock_i_uid(sk2))) {
 198				res = 0;
 199			} else {
 200				res = 1;
 201			}
 202			break;
 203		}
 204	}
 205	spin_unlock(&hslot2->lock);
 206	return res;
 207}
 208
 209static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
 210{
 211	struct net *net = sock_net(sk);
 212	kuid_t uid = sock_i_uid(sk);
 213	struct sock *sk2;
 214
 215	sk_for_each(sk2, &hslot->head) {
 216		if (net_eq(sock_net(sk2), net) &&
 217		    sk2 != sk &&
 218		    sk2->sk_family == sk->sk_family &&
 219		    ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
 220		    (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
 221		    (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 222		    sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
 223		    inet_rcv_saddr_equal(sk, sk2, false)) {
 224			return reuseport_add_sock(sk, sk2);
 
 225		}
 226	}
 227
 228	return reuseport_alloc(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_hslot *hslot, *hslot2;
 243	struct udp_table *udptable = sk->sk_prot->h.udp_table;
 244	int    error = 1;
 245	struct net *net = sock_net(sk);
 246
 247	if (!snum) {
 248		int low, high, remaining;
 249		unsigned int rand;
 250		unsigned short first, last;
 251		DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
 252
 253		inet_get_local_port_range(net, &low, &high);
 254		remaining = (high - low) + 1;
 255
 256		rand = prandom_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, bool exact_dif)
 370{
 371	int score;
 372	struct inet_sock *inet;
 
 373
 374	if (!net_eq(sock_net(sk), net) ||
 375	    udp_sk(sk)->udp_port_hash != hnum ||
 376	    ipv6_only_sock(sk))
 377		return -1;
 378
 
 
 
 379	score = (sk->sk_family == PF_INET) ? 2 : 1;
 
 380	inet = inet_sk(sk);
 381
 382	if (inet->inet_rcv_saddr) {
 383		if (inet->inet_rcv_saddr != daddr)
 384			return -1;
 385		score += 4;
 386	}
 387
 388	if (inet->inet_daddr) {
 389		if (inet->inet_daddr != saddr)
 390			return -1;
 391		score += 4;
 392	}
 393
 394	if (inet->inet_dport) {
 395		if (inet->inet_dport != sport)
 396			return -1;
 397		score += 4;
 398	}
 399
 400	if (sk->sk_bound_dev_if || exact_dif) {
 401		bool dev_match = (sk->sk_bound_dev_if == dif ||
 402				  sk->sk_bound_dev_if == sdif);
 
 
 
 403
 404		if (!dev_match)
 405			return -1;
 406		if (sk->sk_bound_dev_if)
 407			score += 4;
 408	}
 409
 410	if (sk->sk_incoming_cpu == raw_smp_processor_id())
 411		score++;
 412	return score;
 413}
 414
 415static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
 416		       const __u16 lport, const __be32 faddr,
 417		       const __be16 fport)
 418{
 419	static u32 udp_ehash_secret __read_mostly;
 420
 421	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
 422
 423	return __inet_ehashfn(laddr, lport, faddr, fport,
 424			      udp_ehash_secret + net_hash_mix(net));
 425}
 426
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 427/* called with rcu_read_lock() */
 428static struct sock *udp4_lib_lookup2(struct net *net,
 429				     __be32 saddr, __be16 sport,
 430				     __be32 daddr, unsigned int hnum,
 431				     int dif, int sdif, bool exact_dif,
 432				     struct udp_hslot *hslot2,
 433				     struct sk_buff *skb)
 434{
 435	struct sock *sk, *result;
 436	int score, badness;
 437	u32 hash = 0;
 438
 439	result = NULL;
 440	badness = 0;
 441	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
 442		score = compute_score(sk, net, saddr, sport,
 443				      daddr, hnum, dif, sdif, exact_dif);
 444		if (score > badness) {
 445			if (sk->sk_reuseport) {
 446				hash = udp_ehashfn(net, daddr, hnum,
 447						   saddr, sport);
 448				result = reuseport_select_sock(sk, hash, skb,
 449							sizeof(struct udphdr));
 450				if (result)
 451					return result;
 452			}
 453			badness = score;
 454			result = sk;
 455		}
 456	}
 457	return result;
 458}
 459
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 460/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
 461 * harder than this. -DaveM
 462 */
 463struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
 464		__be16 sport, __be32 daddr, __be16 dport, int dif,
 465		int sdif, struct udp_table *udptable, struct sk_buff *skb)
 466{
 467	struct sock *sk, *result;
 468	unsigned short hnum = ntohs(dport);
 469	unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
 470	struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
 471	bool exact_dif = udp_lib_exact_dif_match(net, skb);
 472	int score, badness;
 473	u32 hash = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 474
 475	if (hslot->count > 10) {
 476		hash2 = ipv4_portaddr_hash(net, daddr, hnum);
 477		slot2 = hash2 & udptable->mask;
 478		hslot2 = &udptable->hash2[slot2];
 479		if (hslot->count < hslot2->count)
 480			goto begin;
 481
 482		result = udp4_lib_lookup2(net, saddr, sport,
 483					  daddr, hnum, dif, sdif,
 484					  exact_dif, hslot2, skb);
 485		if (!result) {
 486			unsigned int old_slot2 = slot2;
 487			hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
 488			slot2 = hash2 & udptable->mask;
 489			/* avoid searching the same slot again. */
 490			if (unlikely(slot2 == old_slot2))
 491				return result;
 492
 493			hslot2 = &udptable->hash2[slot2];
 494			if (hslot->count < hslot2->count)
 495				goto begin;
 
 496
 497			result = udp4_lib_lookup2(net, saddr, sport,
 498						  daddr, hnum, dif, sdif,
 499						  exact_dif, hslot2, skb);
 500		}
 501		return result;
 502	}
 503begin:
 504	result = NULL;
 505	badness = 0;
 506	sk_for_each_rcu(sk, &hslot->head) {
 507		score = compute_score(sk, net, saddr, sport,
 508				      daddr, hnum, dif, sdif, exact_dif);
 509		if (score > badness) {
 510			if (sk->sk_reuseport) {
 511				hash = udp_ehashfn(net, daddr, hnum,
 512						   saddr, sport);
 513				result = reuseport_select_sock(sk, hash, skb,
 514							sizeof(struct udphdr));
 515				if (result)
 516					return result;
 517			}
 518			result = sk;
 519			badness = score;
 520		}
 521	}
 522	return result;
 523}
 524EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
 525
 526static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
 527						 __be16 sport, __be16 dport,
 528						 struct udp_table *udptable)
 529{
 530	const struct iphdr *iph = ip_hdr(skb);
 531
 532	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
 533				 iph->daddr, dport, inet_iif(skb),
 534				 inet_sdif(skb), udptable, skb);
 535}
 536
 537struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
 538				 __be16 sport, __be16 dport)
 539{
 540	return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
 
 
 
 
 541}
 542EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
 543
 544/* Must be called under rcu_read_lock().
 545 * Does increment socket refcount.
 546 */
 547#if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
 548    IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
 549    IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
 550struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 551			     __be32 daddr, __be16 dport, int dif)
 552{
 553	struct sock *sk;
 554
 555	sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
 556			       dif, 0, &udp_table, NULL);
 557	if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
 558		sk = NULL;
 559	return sk;
 560}
 561EXPORT_SYMBOL_GPL(udp4_lib_lookup);
 562#endif
 563
 564static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
 565				       __be16 loc_port, __be32 loc_addr,
 566				       __be16 rmt_port, __be32 rmt_addr,
 567				       int dif, int sdif, unsigned short hnum)
 568{
 569	struct inet_sock *inet = inet_sk(sk);
 570
 571	if (!net_eq(sock_net(sk), net) ||
 572	    udp_sk(sk)->udp_port_hash != hnum ||
 573	    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
 574	    (inet->inet_dport != rmt_port && inet->inet_dport) ||
 575	    (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
 576	    ipv6_only_sock(sk) ||
 577	    (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
 578	     sk->sk_bound_dev_if != sdif))
 579		return false;
 580	if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
 581		return false;
 582	return true;
 583}
 584
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 585/*
 586 * This routine is called by the ICMP module when it gets some
 587 * sort of error condition.  If err < 0 then the socket should
 588 * be closed and the error returned to the user.  If err > 0
 589 * it's just the icmp type << 8 | icmp code.
 590 * Header points to the ip header of the error packet. We move
 591 * on past this. Then (as it used to claim before adjustment)
 592 * header points to the first 8 bytes of the udp header.  We need
 593 * to find the appropriate port.
 594 */
 595
 596void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
 597{
 598	struct inet_sock *inet;
 599	const struct iphdr *iph = (const struct iphdr *)skb->data;
 600	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
 601	const int type = icmp_hdr(skb)->type;
 602	const int code = icmp_hdr(skb)->code;
 
 603	struct sock *sk;
 604	int harderr;
 605	int err;
 606	struct net *net = dev_net(skb->dev);
 607
 608	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
 609			       iph->saddr, uh->source, skb->dev->ifindex, 0,
 610			       udptable, NULL);
 611	if (!sk) {
 612		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 613		return;	/* No socket for error */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 614	}
 615
 616	err = 0;
 617	harderr = 0;
 618	inet = inet_sk(sk);
 619
 620	switch (type) {
 621	default:
 622	case ICMP_TIME_EXCEEDED:
 623		err = EHOSTUNREACH;
 624		break;
 625	case ICMP_SOURCE_QUENCH:
 626		goto out;
 627	case ICMP_PARAMETERPROB:
 628		err = EPROTO;
 629		harderr = 1;
 630		break;
 631	case ICMP_DEST_UNREACH:
 632		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
 633			ipv4_sk_update_pmtu(skb, sk, info);
 634			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
 635				err = EMSGSIZE;
 636				harderr = 1;
 637				break;
 638			}
 639			goto out;
 640		}
 641		err = EHOSTUNREACH;
 642		if (code <= NR_ICMP_UNREACH) {
 643			harderr = icmp_err_convert[code].fatal;
 644			err = icmp_err_convert[code].errno;
 645		}
 646		break;
 647	case ICMP_REDIRECT:
 648		ipv4_sk_redirect(skb, sk);
 649		goto out;
 650	}
 651
 652	/*
 653	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
 654	 *	4.1.3.3.
 655	 */
 
 
 
 
 656	if (!inet->recverr) {
 657		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
 658			goto out;
 659	} else
 660		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
 661
 662	sk->sk_err = err;
 663	sk->sk_error_report(sk);
 664out:
 665	return;
 666}
 667
 668void udp_err(struct sk_buff *skb, u32 info)
 669{
 670	__udp4_lib_err(skb, info, &udp_table);
 671}
 672
 673/*
 674 * Throw away all pending data and cancel the corking. Socket is locked.
 675 */
 676void udp_flush_pending_frames(struct sock *sk)
 677{
 678	struct udp_sock *up = udp_sk(sk);
 679
 680	if (up->pending) {
 681		up->len = 0;
 682		up->pending = 0;
 683		ip_flush_pending_frames(sk);
 684	}
 685}
 686EXPORT_SYMBOL(udp_flush_pending_frames);
 687
 688/**
 689 * 	udp4_hwcsum  -  handle outgoing HW checksumming
 690 * 	@skb: 	sk_buff containing the filled-in UDP header
 691 * 	        (checksum field must be zeroed out)
 692 *	@src:	source IP address
 693 *	@dst:	destination IP address
 694 */
 695void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
 696{
 697	struct udphdr *uh = udp_hdr(skb);
 698	int offset = skb_transport_offset(skb);
 699	int len = skb->len - offset;
 700	int hlen = len;
 701	__wsum csum = 0;
 702
 703	if (!skb_has_frag_list(skb)) {
 704		/*
 705		 * Only one fragment on the socket.
 706		 */
 707		skb->csum_start = skb_transport_header(skb) - skb->head;
 708		skb->csum_offset = offsetof(struct udphdr, check);
 709		uh->check = ~csum_tcpudp_magic(src, dst, len,
 710					       IPPROTO_UDP, 0);
 711	} else {
 712		struct sk_buff *frags;
 713
 714		/*
 715		 * HW-checksum won't work as there are two or more
 716		 * fragments on the socket so that all csums of sk_buffs
 717		 * should be together
 718		 */
 719		skb_walk_frags(skb, frags) {
 720			csum = csum_add(csum, frags->csum);
 721			hlen -= frags->len;
 722		}
 723
 724		csum = skb_checksum(skb, offset, hlen, csum);
 725		skb->ip_summed = CHECKSUM_NONE;
 726
 727		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
 728		if (uh->check == 0)
 729			uh->check = CSUM_MANGLED_0;
 730	}
 731}
 732EXPORT_SYMBOL_GPL(udp4_hwcsum);
 733
 734/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
 735 * for the simple case like when setting the checksum for a UDP tunnel.
 736 */
 737void udp_set_csum(bool nocheck, struct sk_buff *skb,
 738		  __be32 saddr, __be32 daddr, int len)
 739{
 740	struct udphdr *uh = udp_hdr(skb);
 741
 742	if (nocheck) {
 743		uh->check = 0;
 744	} else if (skb_is_gso(skb)) {
 745		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 746	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 747		uh->check = 0;
 748		uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
 749		if (uh->check == 0)
 750			uh->check = CSUM_MANGLED_0;
 751	} else {
 752		skb->ip_summed = CHECKSUM_PARTIAL;
 753		skb->csum_start = skb_transport_header(skb) - skb->head;
 754		skb->csum_offset = offsetof(struct udphdr, check);
 755		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 756	}
 757}
 758EXPORT_SYMBOL(udp_set_csum);
 759
 760static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
 
 761{
 762	struct sock *sk = skb->sk;
 763	struct inet_sock *inet = inet_sk(sk);
 764	struct udphdr *uh;
 765	int err = 0;
 766	int is_udplite = IS_UDPLITE(sk);
 767	int offset = skb_transport_offset(skb);
 768	int len = skb->len - offset;
 
 769	__wsum csum = 0;
 770
 771	/*
 772	 * Create a UDP header
 773	 */
 774	uh = udp_hdr(skb);
 775	uh->source = inet->inet_sport;
 776	uh->dest = fl4->fl4_dport;
 777	uh->len = htons(len);
 778	uh->check = 0;
 779
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 780	if (is_udplite)  				 /*     UDP-Lite      */
 781		csum = udplite_csum(skb);
 782
 783	else if (sk->sk_no_check_tx) {			 /* UDP csum off */
 784
 785		skb->ip_summed = CHECKSUM_NONE;
 786		goto send;
 787
 788	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
 
 789
 790		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
 791		goto send;
 792
 793	} else
 794		csum = udp_csum(skb);
 795
 796	/* add protocol-dependent pseudo-header */
 797	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
 798				      sk->sk_protocol, csum);
 799	if (uh->check == 0)
 800		uh->check = CSUM_MANGLED_0;
 801
 802send:
 803	err = ip_send_skb(sock_net(sk), skb);
 804	if (err) {
 805		if (err == -ENOBUFS && !inet->recverr) {
 806			UDP_INC_STATS(sock_net(sk),
 807				      UDP_MIB_SNDBUFERRORS, is_udplite);
 808			err = 0;
 809		}
 810	} else
 811		UDP_INC_STATS(sock_net(sk),
 812			      UDP_MIB_OUTDATAGRAMS, is_udplite);
 813	return err;
 814}
 815
 816/*
 817 * Push out all pending data as one UDP datagram. Socket is locked.
 818 */
 819int udp_push_pending_frames(struct sock *sk)
 820{
 821	struct udp_sock  *up = udp_sk(sk);
 822	struct inet_sock *inet = inet_sk(sk);
 823	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
 824	struct sk_buff *skb;
 825	int err = 0;
 826
 827	skb = ip_finish_skb(sk, fl4);
 828	if (!skb)
 829		goto out;
 830
 831	err = udp_send_skb(skb, fl4);
 832
 833out:
 834	up->len = 0;
 835	up->pending = 0;
 836	return err;
 837}
 838EXPORT_SYMBOL(udp_push_pending_frames);
 839
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 840int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 841{
 842	struct inet_sock *inet = inet_sk(sk);
 843	struct udp_sock *up = udp_sk(sk);
 
 844	struct flowi4 fl4_stack;
 845	struct flowi4 *fl4;
 846	int ulen = len;
 847	struct ipcm_cookie ipc;
 848	struct rtable *rt = NULL;
 849	int free = 0;
 850	int connected = 0;
 851	__be32 daddr, faddr, saddr;
 852	__be16 dport;
 853	u8  tos;
 854	int err, is_udplite = IS_UDPLITE(sk);
 855	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
 856	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
 857	struct sk_buff *skb;
 858	struct ip_options_data opt_copy;
 859
 860	if (len > 0xFFFF)
 861		return -EMSGSIZE;
 862
 863	/*
 864	 *	Check the flags.
 865	 */
 866
 867	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
 868		return -EOPNOTSUPP;
 869
 870	ipc.opt = NULL;
 871	ipc.tx_flags = 0;
 872	ipc.ttl = 0;
 873	ipc.tos = -1;
 874
 875	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
 876
 877	fl4 = &inet->cork.fl.u.ip4;
 878	if (up->pending) {
 879		/*
 880		 * There are pending frames.
 881		 * The socket lock must be held while it's corked.
 882		 */
 883		lock_sock(sk);
 884		if (likely(up->pending)) {
 885			if (unlikely(up->pending != AF_INET)) {
 886				release_sock(sk);
 887				return -EINVAL;
 888			}
 889			goto do_append_data;
 890		}
 891		release_sock(sk);
 892	}
 893	ulen += sizeof(struct udphdr);
 894
 895	/*
 896	 *	Get and verify the address.
 897	 */
 898	if (msg->msg_name) {
 899		DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
 900		if (msg->msg_namelen < sizeof(*usin))
 901			return -EINVAL;
 902		if (usin->sin_family != AF_INET) {
 903			if (usin->sin_family != AF_UNSPEC)
 904				return -EAFNOSUPPORT;
 905		}
 906
 907		daddr = usin->sin_addr.s_addr;
 908		dport = usin->sin_port;
 909		if (dport == 0)
 910			return -EINVAL;
 911	} else {
 912		if (sk->sk_state != TCP_ESTABLISHED)
 913			return -EDESTADDRREQ;
 914		daddr = inet->inet_daddr;
 915		dport = inet->inet_dport;
 916		/* Open fast path for connected socket.
 917		   Route will not be used, if at least one option is set.
 918		 */
 919		connected = 1;
 920	}
 921
 922	ipc.sockc.tsflags = sk->sk_tsflags;
 923	ipc.addr = inet->inet_saddr;
 924	ipc.oif = sk->sk_bound_dev_if;
 925
 926	if (msg->msg_controllen) {
 927		err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
 928		if (unlikely(err)) {
 
 
 
 929			kfree(ipc.opt);
 930			return err;
 931		}
 932		if (ipc.opt)
 933			free = 1;
 934		connected = 0;
 935	}
 936	if (!ipc.opt) {
 937		struct ip_options_rcu *inet_opt;
 938
 939		rcu_read_lock();
 940		inet_opt = rcu_dereference(inet->inet_opt);
 941		if (inet_opt) {
 942			memcpy(&opt_copy, inet_opt,
 943			       sizeof(*inet_opt) + inet_opt->opt.optlen);
 944			ipc.opt = &opt_copy.opt;
 945		}
 946		rcu_read_unlock();
 947	}
 948
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 949	saddr = ipc.addr;
 950	ipc.addr = faddr = daddr;
 951
 952	sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
 953
 954	if (ipc.opt && ipc.opt->opt.srr) {
 955		if (!daddr) {
 956			err = -EINVAL;
 957			goto out_free;
 958		}
 959		faddr = ipc.opt->opt.faddr;
 960		connected = 0;
 961	}
 962	tos = get_rttos(&ipc, inet);
 963	if (sock_flag(sk, SOCK_LOCALROUTE) ||
 964	    (msg->msg_flags & MSG_DONTROUTE) ||
 965	    (ipc.opt && ipc.opt->opt.is_strictroute)) {
 966		tos |= RTO_ONLINK;
 967		connected = 0;
 968	}
 969
 970	if (ipv4_is_multicast(daddr)) {
 971		if (!ipc.oif)
 972			ipc.oif = inet->mc_index;
 973		if (!saddr)
 974			saddr = inet->mc_addr;
 975		connected = 0;
 976	} else if (!ipc.oif) {
 977		ipc.oif = inet->uc_index;
 978	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
 979		/* oif is set, packet is to local broadcast and
 980		 * and uc_index is set. oif is most likely set
 981		 * by sk_bound_dev_if. If uc_index != oif check if the
 982		 * oif is an L3 master and uc_index is an L3 slave.
 983		 * If so, we want to allow the send using the uc_index.
 984		 */
 985		if (ipc.oif != inet->uc_index &&
 986		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
 987							      inet->uc_index)) {
 988			ipc.oif = inet->uc_index;
 989		}
 990	}
 991
 992	if (connected)
 993		rt = (struct rtable *)sk_dst_check(sk, 0);
 994
 995	if (!rt) {
 996		struct net *net = sock_net(sk);
 997		__u8 flow_flags = inet_sk_flowi_flags(sk);
 998
 999		fl4 = &fl4_stack;
1000
1001		flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1002				   RT_SCOPE_UNIVERSE, sk->sk_protocol,
1003				   flow_flags,
1004				   faddr, saddr, dport, inet->inet_sport,
1005				   sk->sk_uid);
1006
1007		security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1008		rt = ip_route_output_flow(net, fl4, sk);
1009		if (IS_ERR(rt)) {
1010			err = PTR_ERR(rt);
1011			rt = NULL;
1012			if (err == -ENETUNREACH)
1013				IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1014			goto out;
1015		}
1016
1017		err = -EACCES;
1018		if ((rt->rt_flags & RTCF_BROADCAST) &&
1019		    !sock_flag(sk, SOCK_BROADCAST))
1020			goto out;
1021		if (connected)
1022			sk_dst_set(sk, dst_clone(&rt->dst));
1023	}
1024
1025	if (msg->msg_flags&MSG_CONFIRM)
1026		goto do_confirm;
1027back_from_confirm:
1028
1029	saddr = fl4->saddr;
1030	if (!ipc.addr)
1031		daddr = ipc.addr = fl4->daddr;
1032
1033	/* Lockless fast path for the non-corking case. */
1034	if (!corkreq) {
 
 
1035		skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1036				  sizeof(struct udphdr), &ipc, &rt,
1037				  msg->msg_flags);
1038		err = PTR_ERR(skb);
1039		if (!IS_ERR_OR_NULL(skb))
1040			err = udp_send_skb(skb, fl4);
1041		goto out;
1042	}
1043
1044	lock_sock(sk);
1045	if (unlikely(up->pending)) {
1046		/* The socket is already corked while preparing it. */
1047		/* ... which is an evident application bug. --ANK */
1048		release_sock(sk);
1049
1050		net_dbg_ratelimited("socket already corked\n");
1051		err = -EINVAL;
1052		goto out;
1053	}
1054	/*
1055	 *	Now cork the socket to pend data.
1056	 */
1057	fl4 = &inet->cork.fl.u.ip4;
1058	fl4->daddr = daddr;
1059	fl4->saddr = saddr;
1060	fl4->fl4_dport = dport;
1061	fl4->fl4_sport = inet->inet_sport;
1062	up->pending = AF_INET;
1063
1064do_append_data:
1065	up->len += ulen;
1066	err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1067			     sizeof(struct udphdr), &ipc, &rt,
1068			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1069	if (err)
1070		udp_flush_pending_frames(sk);
1071	else if (!corkreq)
1072		err = udp_push_pending_frames(sk);
1073	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1074		up->pending = 0;
1075	release_sock(sk);
1076
1077out:
1078	ip_rt_put(rt);
1079out_free:
1080	if (free)
1081		kfree(ipc.opt);
1082	if (!err)
1083		return len;
1084	/*
1085	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1086	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1087	 * we don't have a good statistic (IpOutDiscards but it can be too many
1088	 * things).  We could add another new stat but at least for now that
1089	 * seems like overkill.
1090	 */
1091	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1092		UDP_INC_STATS(sock_net(sk),
1093			      UDP_MIB_SNDBUFERRORS, is_udplite);
1094	}
1095	return err;
1096
1097do_confirm:
1098	if (msg->msg_flags & MSG_PROBE)
1099		dst_confirm_neigh(&rt->dst, &fl4->daddr);
1100	if (!(msg->msg_flags&MSG_PROBE) || len)
1101		goto back_from_confirm;
1102	err = 0;
1103	goto out;
1104}
1105EXPORT_SYMBOL(udp_sendmsg);
1106
1107int udp_sendpage(struct sock *sk, struct page *page, int offset,
1108		 size_t size, int flags)
1109{
1110	struct inet_sock *inet = inet_sk(sk);
1111	struct udp_sock *up = udp_sk(sk);
1112	int ret;
1113
1114	if (flags & MSG_SENDPAGE_NOTLAST)
1115		flags |= MSG_MORE;
1116
1117	if (!up->pending) {
1118		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
1119
1120		/* Call udp_sendmsg to specify destination address which
1121		 * sendpage interface can't pass.
1122		 * This will succeed only when the socket is connected.
1123		 */
1124		ret = udp_sendmsg(sk, &msg, 0);
1125		if (ret < 0)
1126			return ret;
1127	}
1128
1129	lock_sock(sk);
1130
1131	if (unlikely(!up->pending)) {
1132		release_sock(sk);
1133
1134		net_dbg_ratelimited("cork failed\n");
1135		return -EINVAL;
1136	}
1137
1138	ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1139			     page, offset, size, flags);
1140	if (ret == -EOPNOTSUPP) {
1141		release_sock(sk);
1142		return sock_no_sendpage(sk->sk_socket, page, offset,
1143					size, flags);
1144	}
1145	if (ret < 0) {
1146		udp_flush_pending_frames(sk);
1147		goto out;
1148	}
1149
1150	up->len += size;
1151	if (!(up->corkflag || (flags&MSG_MORE)))
1152		ret = udp_push_pending_frames(sk);
1153	if (!ret)
1154		ret = size;
1155out:
1156	release_sock(sk);
1157	return ret;
1158}
1159
1160#define UDP_SKB_IS_STATELESS 0x80000000
1161
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1162static void udp_set_dev_scratch(struct sk_buff *skb)
1163{
1164	struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1165
1166	BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1167	scratch->_tsize_state = skb->truesize;
1168#if BITS_PER_LONG == 64
1169	scratch->len = skb->len;
1170	scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1171	scratch->is_linear = !skb_is_nonlinear(skb);
1172#endif
1173	/* all head states execept sp (dst, sk, nf) are always cleared by
1174	 * udp_rcv() and we need to preserve secpath, if present, to eventually
1175	 * process IP_CMSG_PASSSEC at recvmsg() time
 
 
 
 
 
 
 
 
1176	 */
1177	if (likely(!skb_sec_path(skb)))
1178		scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
 
 
1179}
1180
1181static int udp_skb_truesize(struct sk_buff *skb)
1182{
1183	return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1184}
1185
1186static bool udp_skb_has_head_state(struct sk_buff *skb)
1187{
1188	return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1189}
1190
1191/* fully reclaim rmem/fwd memory allocated for skb */
1192static void udp_rmem_release(struct sock *sk, int size, int partial,
1193			     bool rx_queue_lock_held)
1194{
1195	struct udp_sock *up = udp_sk(sk);
1196	struct sk_buff_head *sk_queue;
1197	int amt;
1198
1199	if (likely(partial)) {
1200		up->forward_deficit += size;
1201		size = up->forward_deficit;
1202		if (size < (sk->sk_rcvbuf >> 2))
 
1203			return;
1204	} else {
1205		size += up->forward_deficit;
1206	}
1207	up->forward_deficit = 0;
1208
1209	/* acquire the sk_receive_queue for fwd allocated memory scheduling,
1210	 * if the called don't held it already
1211	 */
1212	sk_queue = &sk->sk_receive_queue;
1213	if (!rx_queue_lock_held)
1214		spin_lock(&sk_queue->lock);
1215
1216
1217	sk->sk_forward_alloc += size;
1218	amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1219	sk->sk_forward_alloc -= amt;
1220
1221	if (amt)
1222		__sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1223
1224	atomic_sub(size, &sk->sk_rmem_alloc);
1225
1226	/* this can save us from acquiring the rx queue lock on next receive */
1227	skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1228
1229	if (!rx_queue_lock_held)
1230		spin_unlock(&sk_queue->lock);
1231}
1232
1233/* Note: called with reader_queue.lock held.
1234 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1235 * This avoids a cache line miss while receive_queue lock is held.
1236 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1237 */
1238void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1239{
1240	prefetch(&skb->data);
1241	udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1242}
1243EXPORT_SYMBOL(udp_skb_destructor);
1244
1245/* as above, but the caller held the rx queue lock, too */
1246static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1247{
1248	prefetch(&skb->data);
1249	udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1250}
1251
1252/* Idea of busylocks is to let producers grab an extra spinlock
1253 * to relieve pressure on the receive_queue spinlock shared by consumer.
1254 * Under flood, this means that only one producer can be in line
1255 * trying to acquire the receive_queue spinlock.
1256 * These busylock can be allocated on a per cpu manner, instead of a
1257 * per socket one (that would consume a cache line per socket)
1258 */
1259static int udp_busylocks_log __read_mostly;
1260static spinlock_t *udp_busylocks __read_mostly;
1261
1262static spinlock_t *busylock_acquire(void *ptr)
1263{
1264	spinlock_t *busy;
1265
1266	busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1267	spin_lock(busy);
1268	return busy;
1269}
1270
1271static void busylock_release(spinlock_t *busy)
1272{
1273	if (busy)
1274		spin_unlock(busy);
1275}
1276
1277int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1278{
1279	struct sk_buff_head *list = &sk->sk_receive_queue;
1280	int rmem, delta, amt, err = -ENOMEM;
1281	spinlock_t *busy = NULL;
1282	int size;
1283
1284	/* try to avoid the costly atomic add/sub pair when the receive
1285	 * queue is full; always allow at least a packet
1286	 */
1287	rmem = atomic_read(&sk->sk_rmem_alloc);
1288	if (rmem > sk->sk_rcvbuf)
1289		goto drop;
1290
1291	/* Under mem pressure, it might be helpful to help udp_recvmsg()
1292	 * having linear skbs :
1293	 * - Reduce memory overhead and thus increase receive queue capacity
1294	 * - Less cache line misses at copyout() time
1295	 * - Less work at consume_skb() (less alien page frag freeing)
1296	 */
1297	if (rmem > (sk->sk_rcvbuf >> 1)) {
1298		skb_condense(skb);
1299
1300		busy = busylock_acquire(sk);
1301	}
1302	size = skb->truesize;
1303	udp_set_dev_scratch(skb);
1304
1305	/* we drop only if the receive buf is full and the receive
1306	 * queue contains some other skb
1307	 */
1308	rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1309	if (rmem > (size + sk->sk_rcvbuf))
1310		goto uncharge_drop;
1311
1312	spin_lock(&list->lock);
1313	if (size >= sk->sk_forward_alloc) {
1314		amt = sk_mem_pages(size);
1315		delta = amt << SK_MEM_QUANTUM_SHIFT;
1316		if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1317			err = -ENOBUFS;
1318			spin_unlock(&list->lock);
1319			goto uncharge_drop;
1320		}
1321
1322		sk->sk_forward_alloc += delta;
1323	}
1324
1325	sk->sk_forward_alloc -= size;
1326
1327	/* no need to setup a destructor, we will explicitly release the
1328	 * forward allocated memory on dequeue
1329	 */
1330	sock_skb_set_dropcount(sk, skb);
1331
1332	__skb_queue_tail(list, skb);
1333	spin_unlock(&list->lock);
1334
1335	if (!sock_flag(sk, SOCK_DEAD))
1336		sk->sk_data_ready(sk);
1337
1338	busylock_release(busy);
1339	return 0;
1340
1341uncharge_drop:
1342	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1343
1344drop:
1345	atomic_inc(&sk->sk_drops);
1346	busylock_release(busy);
1347	return err;
1348}
1349EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1350
1351void udp_destruct_sock(struct sock *sk)
1352{
1353	/* reclaim completely the forward allocated memory */
1354	struct udp_sock *up = udp_sk(sk);
1355	unsigned int total = 0;
1356	struct sk_buff *skb;
1357
1358	skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1359	while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1360		total += skb->truesize;
1361		kfree_skb(skb);
1362	}
1363	udp_rmem_release(sk, total, 0, true);
1364
1365	inet_sock_destruct(sk);
1366}
1367EXPORT_SYMBOL_GPL(udp_destruct_sock);
1368
1369int udp_init_sock(struct sock *sk)
1370{
1371	skb_queue_head_init(&udp_sk(sk)->reader_queue);
1372	sk->sk_destruct = udp_destruct_sock;
1373	return 0;
1374}
1375EXPORT_SYMBOL_GPL(udp_init_sock);
1376
1377void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1378{
1379	if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1380		bool slow = lock_sock_fast(sk);
1381
1382		sk_peek_offset_bwd(sk, len);
1383		unlock_sock_fast(sk, slow);
1384	}
1385
1386	if (!skb_unref(skb))
1387		return;
1388
1389	/* In the more common cases we cleared the head states previously,
1390	 * see __udp_queue_rcv_skb().
1391	 */
1392	if (unlikely(udp_skb_has_head_state(skb)))
1393		skb_release_head_state(skb);
1394	__consume_stateless_skb(skb);
1395}
1396EXPORT_SYMBOL_GPL(skb_consume_udp);
1397
1398static struct sk_buff *__first_packet_length(struct sock *sk,
1399					     struct sk_buff_head *rcvq,
1400					     int *total)
1401{
1402	struct sk_buff *skb;
1403
1404	while ((skb = skb_peek(rcvq)) != NULL) {
1405		if (udp_lib_checksum_complete(skb)) {
1406			__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1407					IS_UDPLITE(sk));
1408			__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1409					IS_UDPLITE(sk));
1410			atomic_inc(&sk->sk_drops);
1411			__skb_unlink(skb, rcvq);
1412			*total += skb->truesize;
1413			kfree_skb(skb);
1414		} else {
1415			/* the csum related bits could be changed, refresh
1416			 * the scratch area
1417			 */
1418			udp_set_dev_scratch(skb);
1419			break;
1420		}
1421	}
1422	return skb;
1423}
1424
1425/**
1426 *	first_packet_length	- return length of first packet in receive queue
1427 *	@sk: socket
1428 *
1429 *	Drops all bad checksum frames, until a valid one is found.
1430 *	Returns the length of found skb, or -1 if none is found.
1431 */
1432static int first_packet_length(struct sock *sk)
1433{
1434	struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1435	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1436	struct sk_buff *skb;
1437	int total = 0;
1438	int res;
1439
1440	spin_lock_bh(&rcvq->lock);
1441	skb = __first_packet_length(sk, rcvq, &total);
1442	if (!skb && !skb_queue_empty(sk_queue)) {
1443		spin_lock(&sk_queue->lock);
1444		skb_queue_splice_tail_init(sk_queue, rcvq);
1445		spin_unlock(&sk_queue->lock);
1446
1447		skb = __first_packet_length(sk, rcvq, &total);
1448	}
1449	res = skb ? skb->len : -1;
1450	if (total)
1451		udp_rmem_release(sk, total, 1, false);
1452	spin_unlock_bh(&rcvq->lock);
1453	return res;
1454}
1455
1456/*
1457 *	IOCTL requests applicable to the UDP protocol
1458 */
1459
1460int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1461{
1462	switch (cmd) {
1463	case SIOCOUTQ:
1464	{
1465		int amount = sk_wmem_alloc_get(sk);
1466
1467		return put_user(amount, (int __user *)arg);
1468	}
1469
1470	case SIOCINQ:
1471	{
1472		int amount = max_t(int, 0, first_packet_length(sk));
1473
1474		return put_user(amount, (int __user *)arg);
1475	}
1476
1477	default:
1478		return -ENOIOCTLCMD;
1479	}
1480
1481	return 0;
1482}
1483EXPORT_SYMBOL(udp_ioctl);
1484
1485struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1486			       int noblock, int *peeked, int *off, int *err)
1487{
1488	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1489	struct sk_buff_head *queue;
1490	struct sk_buff *last;
1491	long timeo;
1492	int error;
1493
1494	queue = &udp_sk(sk)->reader_queue;
1495	flags |= noblock ? MSG_DONTWAIT : 0;
1496	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1497	do {
1498		struct sk_buff *skb;
1499
1500		error = sock_error(sk);
1501		if (error)
1502			break;
1503
1504		error = -EAGAIN;
1505		*peeked = 0;
1506		do {
1507			spin_lock_bh(&queue->lock);
1508			skb = __skb_try_recv_from_queue(sk, queue, flags,
1509							udp_skb_destructor,
1510							peeked, off, err,
1511							&last);
1512			if (skb) {
 
 
1513				spin_unlock_bh(&queue->lock);
1514				return skb;
1515			}
1516
1517			if (skb_queue_empty(sk_queue)) {
1518				spin_unlock_bh(&queue->lock);
1519				goto busy_check;
1520			}
1521
1522			/* refill the reader queue and walk it again
1523			 * keep both queues locked to avoid re-acquiring
1524			 * the sk_receive_queue lock if fwd memory scheduling
1525			 * is needed.
1526			 */
1527			spin_lock(&sk_queue->lock);
1528			skb_queue_splice_tail_init(sk_queue, queue);
1529
1530			skb = __skb_try_recv_from_queue(sk, queue, flags,
1531							udp_skb_dtor_locked,
1532							peeked, off, err,
1533							&last);
1534			spin_unlock(&sk_queue->lock);
1535			spin_unlock_bh(&queue->lock);
1536			if (skb)
1537				return skb;
1538
1539busy_check:
1540			if (!sk_can_busy_loop(sk))
1541				break;
1542
1543			sk_busy_loop(sk, flags & MSG_DONTWAIT);
1544		} while (!skb_queue_empty(sk_queue));
1545
1546		/* sk_queue is empty, reader_queue may contain peeked packets */
1547	} while (timeo &&
1548		 !__skb_wait_for_more_packets(sk, &error, &timeo,
 
1549					      (struct sk_buff *)sk_queue));
1550
1551	*err = error;
1552	return NULL;
1553}
1554EXPORT_SYMBOL_GPL(__skb_recv_udp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1555
1556/*
1557 * 	This should be easy, if there is something there we
1558 * 	return it, otherwise we block.
1559 */
1560
1561int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1562		int flags, int *addr_len)
1563{
1564	struct inet_sock *inet = inet_sk(sk);
1565	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1566	struct sk_buff *skb;
1567	unsigned int ulen, copied;
1568	int peeked, peeking, off;
1569	int err;
1570	int is_udplite = IS_UDPLITE(sk);
1571	bool checksum_valid = false;
1572
1573	if (flags & MSG_ERRQUEUE)
1574		return ip_recv_error(sk, msg, len, addr_len);
1575
1576try_again:
1577	peeking = flags & MSG_PEEK;
1578	off = sk_peek_offset(sk, flags);
1579	skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1580	if (!skb)
1581		return err;
1582
1583	ulen = udp_skb_len(skb);
1584	copied = len;
1585	if (copied > ulen - off)
1586		copied = ulen - off;
1587	else if (copied < ulen)
1588		msg->msg_flags |= MSG_TRUNC;
1589
1590	/*
1591	 * If checksum is needed at all, try to do it while copying the
1592	 * data.  If the data is truncated, or if we only want a partial
1593	 * coverage checksum (UDP-Lite), do it before the copy.
1594	 */
1595
1596	if (copied < ulen || peeking ||
1597	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1598		checksum_valid = udp_skb_csum_unnecessary(skb) ||
1599				!__udp_lib_checksum_complete(skb);
1600		if (!checksum_valid)
1601			goto csum_copy_err;
1602	}
1603
1604	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1605		if (udp_skb_is_linear(skb))
1606			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1607		else
1608			err = skb_copy_datagram_msg(skb, off, msg, copied);
1609	} else {
1610		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1611
1612		if (err == -EINVAL)
1613			goto csum_copy_err;
1614	}
1615
1616	if (unlikely(err)) {
1617		if (!peeked) {
1618			atomic_inc(&sk->sk_drops);
1619			UDP_INC_STATS(sock_net(sk),
1620				      UDP_MIB_INERRORS, is_udplite);
1621		}
1622		kfree_skb(skb);
1623		return err;
1624	}
1625
1626	if (!peeked)
1627		UDP_INC_STATS(sock_net(sk),
1628			      UDP_MIB_INDATAGRAMS, is_udplite);
1629
1630	sock_recv_ts_and_drops(msg, sk, skb);
1631
1632	/* Copy the address. */
1633	if (sin) {
1634		sin->sin_family = AF_INET;
1635		sin->sin_port = udp_hdr(skb)->source;
1636		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1637		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1638		*addr_len = sizeof(*sin);
 
 
 
1639	}
 
 
 
 
1640	if (inet->cmsg_flags)
1641		ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1642
1643	err = copied;
1644	if (flags & MSG_TRUNC)
1645		err = ulen;
1646
1647	skb_consume_udp(sk, skb, peeking ? -err : err);
1648	return err;
1649
1650csum_copy_err:
1651	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1652				 udp_skb_destructor)) {
1653		UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1654		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1655	}
1656	kfree_skb(skb);
1657
1658	/* starting over for a new packet, but check if we need to yield */
1659	cond_resched();
1660	msg->msg_flags &= ~MSG_TRUNC;
1661	goto try_again;
1662}
1663
1664int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1665{
1666	/* This check is replicated from __ip4_datagram_connect() and
1667	 * intended to prevent BPF program called below from accessing bytes
1668	 * that are out of the bound specified by user in addr_len.
1669	 */
1670	if (addr_len < sizeof(struct sockaddr_in))
1671		return -EINVAL;
1672
1673	return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1674}
1675EXPORT_SYMBOL(udp_pre_connect);
1676
1677int __udp_disconnect(struct sock *sk, int flags)
1678{
1679	struct inet_sock *inet = inet_sk(sk);
1680	/*
1681	 *	1003.1g - break association.
1682	 */
1683
1684	sk->sk_state = TCP_CLOSE;
1685	inet->inet_daddr = 0;
1686	inet->inet_dport = 0;
1687	sock_rps_reset_rxhash(sk);
1688	sk->sk_bound_dev_if = 0;
1689	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1690		inet_reset_saddr(sk);
 
 
 
 
1691
1692	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1693		sk->sk_prot->unhash(sk);
1694		inet->inet_sport = 0;
1695	}
1696	sk_dst_reset(sk);
1697	return 0;
1698}
1699EXPORT_SYMBOL(__udp_disconnect);
1700
1701int udp_disconnect(struct sock *sk, int flags)
1702{
1703	lock_sock(sk);
1704	__udp_disconnect(sk, flags);
1705	release_sock(sk);
1706	return 0;
1707}
1708EXPORT_SYMBOL(udp_disconnect);
1709
1710void udp_lib_unhash(struct sock *sk)
1711{
1712	if (sk_hashed(sk)) {
1713		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1714		struct udp_hslot *hslot, *hslot2;
1715
1716		hslot  = udp_hashslot(udptable, sock_net(sk),
1717				      udp_sk(sk)->udp_port_hash);
1718		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1719
1720		spin_lock_bh(&hslot->lock);
1721		if (rcu_access_pointer(sk->sk_reuseport_cb))
1722			reuseport_detach_sock(sk);
1723		if (sk_del_node_init_rcu(sk)) {
1724			hslot->count--;
1725			inet_sk(sk)->inet_num = 0;
1726			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1727
1728			spin_lock(&hslot2->lock);
1729			hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1730			hslot2->count--;
1731			spin_unlock(&hslot2->lock);
1732		}
1733		spin_unlock_bh(&hslot->lock);
1734	}
1735}
1736EXPORT_SYMBOL(udp_lib_unhash);
1737
1738/*
1739 * inet_rcv_saddr was changed, we must rehash secondary hash
1740 */
1741void udp_lib_rehash(struct sock *sk, u16 newhash)
1742{
1743	if (sk_hashed(sk)) {
1744		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1745		struct udp_hslot *hslot, *hslot2, *nhslot2;
1746
1747		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1748		nhslot2 = udp_hashslot2(udptable, newhash);
1749		udp_sk(sk)->udp_portaddr_hash = newhash;
1750
1751		if (hslot2 != nhslot2 ||
1752		    rcu_access_pointer(sk->sk_reuseport_cb)) {
1753			hslot = udp_hashslot(udptable, sock_net(sk),
1754					     udp_sk(sk)->udp_port_hash);
1755			/* we must lock primary chain too */
1756			spin_lock_bh(&hslot->lock);
1757			if (rcu_access_pointer(sk->sk_reuseport_cb))
1758				reuseport_detach_sock(sk);
1759
1760			if (hslot2 != nhslot2) {
1761				spin_lock(&hslot2->lock);
1762				hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1763				hslot2->count--;
1764				spin_unlock(&hslot2->lock);
1765
1766				spin_lock(&nhslot2->lock);
1767				hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1768							 &nhslot2->head);
1769				nhslot2->count++;
1770				spin_unlock(&nhslot2->lock);
1771			}
1772
1773			spin_unlock_bh(&hslot->lock);
1774		}
1775	}
1776}
1777EXPORT_SYMBOL(udp_lib_rehash);
1778
1779static void udp_v4_rehash(struct sock *sk)
1780{
1781	u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1782					  inet_sk(sk)->inet_rcv_saddr,
1783					  inet_sk(sk)->inet_num);
1784	udp_lib_rehash(sk, new_hash);
1785}
1786
1787static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1788{
1789	int rc;
1790
1791	if (inet_sk(sk)->inet_daddr) {
1792		sock_rps_save_rxhash(sk, skb);
1793		sk_mark_napi_id(sk, skb);
1794		sk_incoming_cpu_update(sk);
1795	} else {
1796		sk_mark_napi_id_once(sk, skb);
1797	}
1798
1799	rc = __udp_enqueue_schedule_skb(sk, skb);
1800	if (rc < 0) {
1801		int is_udplite = IS_UDPLITE(sk);
1802
1803		/* Note that an ENOMEM error is charged twice */
1804		if (rc == -ENOMEM)
1805			UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1806					is_udplite);
 
 
 
1807		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1808		kfree_skb(skb);
1809		trace_udp_fail_queue_rcv_skb(rc, sk);
1810		return -1;
1811	}
1812
1813	return 0;
1814}
1815
1816static struct static_key udp_encap_needed __read_mostly;
1817void udp_encap_enable(void)
1818{
1819	static_key_enable(&udp_encap_needed);
1820}
1821EXPORT_SYMBOL(udp_encap_enable);
1822
1823/* returns:
1824 *  -1: error
1825 *   0: success
1826 *  >0: "udp encap" protocol resubmission
1827 *
1828 * Note that in the success and error cases, the skb is assumed to
1829 * have either been requeued or freed.
1830 */
1831static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1832{
1833	struct udp_sock *up = udp_sk(sk);
1834	int is_udplite = IS_UDPLITE(sk);
1835
1836	/*
1837	 *	Charge it to the socket, dropping if the queue is full.
1838	 */
1839	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1840		goto drop;
1841	nf_reset(skb);
1842
1843	if (static_key_false(&udp_encap_needed) && up->encap_type) {
1844		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1845
1846		/*
1847		 * This is an encapsulation socket so pass the skb to
1848		 * the socket's udp_encap_rcv() hook. Otherwise, just
1849		 * fall through and pass this up the UDP socket.
1850		 * up->encap_rcv() returns the following value:
1851		 * =0 if skb was successfully passed to the encap
1852		 *    handler or was discarded by it.
1853		 * >0 if skb should be passed on to UDP.
1854		 * <0 if skb should be resubmitted as proto -N
1855		 */
1856
1857		/* if we're overly short, let UDP handle it */
1858		encap_rcv = READ_ONCE(up->encap_rcv);
1859		if (encap_rcv) {
1860			int ret;
1861
1862			/* Verify checksum before giving to encap */
1863			if (udp_lib_checksum_complete(skb))
1864				goto csum_error;
1865
1866			ret = encap_rcv(sk, skb);
1867			if (ret <= 0) {
1868				__UDP_INC_STATS(sock_net(sk),
1869						UDP_MIB_INDATAGRAMS,
1870						is_udplite);
1871				return -ret;
1872			}
1873		}
1874
1875		/* FALLTHROUGH -- it's a UDP Packet */
1876	}
1877
1878	/*
1879	 * 	UDP-Lite specific tests, ignored on UDP sockets
1880	 */
1881	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
1882
1883		/*
1884		 * MIB statistics other than incrementing the error count are
1885		 * disabled for the following two types of errors: these depend
1886		 * on the application settings, not on the functioning of the
1887		 * protocol stack as such.
1888		 *
1889		 * RFC 3828 here recommends (sec 3.3): "There should also be a
1890		 * way ... to ... at least let the receiving application block
1891		 * delivery of packets with coverage values less than a value
1892		 * provided by the application."
1893		 */
1894		if (up->pcrlen == 0) {          /* full coverage was set  */
1895			net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1896					    UDP_SKB_CB(skb)->cscov, skb->len);
1897			goto drop;
1898		}
1899		/* The next case involves violating the min. coverage requested
1900		 * by the receiver. This is subtle: if receiver wants x and x is
1901		 * greater than the buffersize/MTU then receiver will complain
1902		 * that it wants x while sender emits packets of smaller size y.
1903		 * Therefore the above ...()->partial_cov statement is essential.
1904		 */
1905		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
1906			net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1907					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
1908			goto drop;
1909		}
1910	}
1911
1912	prefetch(&sk->sk_rmem_alloc);
1913	if (rcu_access_pointer(sk->sk_filter) &&
1914	    udp_lib_checksum_complete(skb))
1915			goto csum_error;
1916
1917	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1918		goto drop;
1919
1920	udp_csum_pull_header(skb);
1921
1922	ipv4_pktinfo_prepare(sk, skb);
1923	return __udp_queue_rcv_skb(sk, skb);
1924
1925csum_error:
1926	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1927drop:
1928	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1929	atomic_inc(&sk->sk_drops);
1930	kfree_skb(skb);
1931	return -1;
1932}
1933
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1934/* For TCP sockets, sk_rx_dst is protected by socket lock
1935 * For UDP, we use xchg() to guard against concurrent changes.
1936 */
1937bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1938{
1939	struct dst_entry *old;
1940
1941	if (dst_hold_safe(dst)) {
1942		old = xchg(&sk->sk_rx_dst, dst);
1943		dst_release(old);
1944		return old != dst;
1945	}
1946	return false;
1947}
1948EXPORT_SYMBOL(udp_sk_rx_dst_set);
1949
1950/*
1951 *	Multicasts and broadcasts go to each listener.
1952 *
1953 *	Note: called only from the BH handler context.
1954 */
1955static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1956				    struct udphdr  *uh,
1957				    __be32 saddr, __be32 daddr,
1958				    struct udp_table *udptable,
1959				    int proto)
1960{
1961	struct sock *sk, *first = NULL;
1962	unsigned short hnum = ntohs(uh->dest);
1963	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1964	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1965	unsigned int offset = offsetof(typeof(*sk), sk_node);
1966	int dif = skb->dev->ifindex;
1967	int sdif = inet_sdif(skb);
1968	struct hlist_node *node;
1969	struct sk_buff *nskb;
1970
1971	if (use_hash2) {
1972		hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1973			    udptable->mask;
1974		hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
1975start_lookup:
1976		hslot = &udptable->hash2[hash2];
1977		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1978	}
1979
1980	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1981		if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1982					 uh->source, saddr, dif, sdif, hnum))
1983			continue;
1984
1985		if (!first) {
1986			first = sk;
1987			continue;
1988		}
1989		nskb = skb_clone(skb, GFP_ATOMIC);
1990
1991		if (unlikely(!nskb)) {
1992			atomic_inc(&sk->sk_drops);
1993			__UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1994					IS_UDPLITE(sk));
1995			__UDP_INC_STATS(net, UDP_MIB_INERRORS,
1996					IS_UDPLITE(sk));
1997			continue;
1998		}
1999		if (udp_queue_rcv_skb(sk, nskb) > 0)
2000			consume_skb(nskb);
2001	}
2002
2003	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
2004	if (use_hash2 && hash2 != hash2_any) {
2005		hash2 = hash2_any;
2006		goto start_lookup;
2007	}
2008
2009	if (first) {
2010		if (udp_queue_rcv_skb(first, skb) > 0)
2011			consume_skb(skb);
2012	} else {
2013		kfree_skb(skb);
2014		__UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2015				proto == IPPROTO_UDPLITE);
2016	}
2017	return 0;
2018}
2019
2020/* Initialize UDP checksum. If exited with zero value (success),
2021 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2022 * Otherwise, csum completion requires chacksumming packet body,
2023 * including udp header and folding it to skb->csum.
2024 */
2025static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2026				 int proto)
2027{
2028	int err;
2029
2030	UDP_SKB_CB(skb)->partial_cov = 0;
2031	UDP_SKB_CB(skb)->cscov = skb->len;
2032
2033	if (proto == IPPROTO_UDPLITE) {
2034		err = udplite_checksum_init(skb, uh);
2035		if (err)
2036			return err;
2037
2038		if (UDP_SKB_CB(skb)->partial_cov) {
2039			skb->csum = inet_compute_pseudo(skb, proto);
2040			return 0;
2041		}
2042	}
2043
2044	/* Note, we are only interested in != 0 or == 0, thus the
2045	 * force to int.
2046	 */
2047	return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2048							 inet_compute_pseudo);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2049}
2050
2051/*
2052 *	All we need to do is get the socket, and then do a checksum.
2053 */
2054
2055int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2056		   int proto)
2057{
2058	struct sock *sk;
2059	struct udphdr *uh;
2060	unsigned short ulen;
2061	struct rtable *rt = skb_rtable(skb);
2062	__be32 saddr, daddr;
2063	struct net *net = dev_net(skb->dev);
 
2064
2065	/*
2066	 *  Validate the packet.
2067	 */
2068	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2069		goto drop;		/* No space for header. */
2070
2071	uh   = udp_hdr(skb);
2072	ulen = ntohs(uh->len);
2073	saddr = ip_hdr(skb)->saddr;
2074	daddr = ip_hdr(skb)->daddr;
2075
2076	if (ulen > skb->len)
2077		goto short_packet;
2078
2079	if (proto == IPPROTO_UDP) {
2080		/* UDP validates ulen. */
2081		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2082			goto short_packet;
2083		uh = udp_hdr(skb);
2084	}
2085
2086	if (udp4_csum_init(skb, uh, proto))
2087		goto csum_error;
2088
2089	sk = skb_steal_sock(skb);
2090	if (sk) {
2091		struct dst_entry *dst = skb_dst(skb);
2092		int ret;
2093
2094		if (unlikely(sk->sk_rx_dst != dst))
2095			udp_sk_rx_dst_set(sk, dst);
2096
2097		ret = udp_queue_rcv_skb(sk, skb);
2098		sock_put(sk);
2099		/* a return value > 0 means to resubmit the input, but
2100		 * it wants the return to be -protocol, or 0
2101		 */
2102		if (ret > 0)
2103			return -ret;
2104		return 0;
2105	}
2106
2107	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2108		return __udp4_lib_mcast_deliver(net, skb, uh,
2109						saddr, daddr, udptable, proto);
2110
2111	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2112	if (sk) {
2113		int ret;
2114
2115		if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2116			skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2117						 inet_compute_pseudo);
2118
2119		ret = udp_queue_rcv_skb(sk, skb);
2120
2121		/* a return value > 0 means to resubmit the input, but
2122		 * it wants the return to be -protocol, or 0
2123		 */
2124		if (ret > 0)
2125			return -ret;
2126		return 0;
2127	}
2128
2129	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2130		goto drop;
2131	nf_reset(skb);
2132
2133	/* No socket. Drop packet silently, if checksum is wrong */
2134	if (udp_lib_checksum_complete(skb))
2135		goto csum_error;
2136
2137	__UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2138	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2139
2140	/*
2141	 * Hmm.  We got an UDP packet to a port to which we
2142	 * don't wanna listen.  Ignore it.
2143	 */
2144	kfree_skb(skb);
2145	return 0;
2146
2147short_packet:
2148	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2149			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2150			    &saddr, ntohs(uh->source),
2151			    ulen, skb->len,
2152			    &daddr, ntohs(uh->dest));
2153	goto drop;
2154
2155csum_error:
2156	/*
2157	 * RFC1122: OK.  Discards the bad packet silently (as far as
2158	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2159	 */
2160	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2161			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2162			    &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2163			    ulen);
2164	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2165drop:
2166	__UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2167	kfree_skb(skb);
2168	return 0;
2169}
2170
2171/* We can only early demux multicast if there is a single matching socket.
2172 * If more than one socket found returns NULL
2173 */
2174static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2175						  __be16 loc_port, __be32 loc_addr,
2176						  __be16 rmt_port, __be32 rmt_addr,
2177						  int dif, int sdif)
2178{
2179	struct sock *sk, *result;
2180	unsigned short hnum = ntohs(loc_port);
2181	unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2182	struct udp_hslot *hslot = &udp_table.hash[slot];
2183
2184	/* Do not bother scanning a too big list */
2185	if (hslot->count > 10)
2186		return NULL;
2187
2188	result = NULL;
2189	sk_for_each_rcu(sk, &hslot->head) {
2190		if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2191					rmt_port, rmt_addr, dif, sdif, hnum)) {
2192			if (result)
2193				return NULL;
2194			result = sk;
2195		}
2196	}
2197
2198	return result;
2199}
2200
2201/* For unicast we should only early demux connected sockets or we can
2202 * break forwarding setups.  The chains here can be long so only check
2203 * if the first socket is an exact match and if not move on.
2204 */
2205static struct sock *__udp4_lib_demux_lookup(struct net *net,
2206					    __be16 loc_port, __be32 loc_addr,
2207					    __be16 rmt_port, __be32 rmt_addr,
2208					    int dif, int sdif)
2209{
2210	unsigned short hnum = ntohs(loc_port);
2211	unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2212	unsigned int slot2 = hash2 & udp_table.mask;
2213	struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2214	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2215	const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2216	struct sock *sk;
2217
2218	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2219		if (INET_MATCH(sk, net, acookie, rmt_addr,
2220			       loc_addr, ports, dif, sdif))
2221			return sk;
2222		/* Only check first socket in chain */
2223		break;
2224	}
2225	return NULL;
2226}
2227
2228int udp_v4_early_demux(struct sk_buff *skb)
2229{
2230	struct net *net = dev_net(skb->dev);
2231	struct in_device *in_dev = NULL;
2232	const struct iphdr *iph;
2233	const struct udphdr *uh;
2234	struct sock *sk = NULL;
2235	struct dst_entry *dst;
2236	int dif = skb->dev->ifindex;
2237	int sdif = inet_sdif(skb);
2238	int ours;
2239
2240	/* validate the packet */
2241	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2242		return 0;
2243
2244	iph = ip_hdr(skb);
2245	uh = udp_hdr(skb);
2246
2247	if (skb->pkt_type == PACKET_MULTICAST) {
2248		in_dev = __in_dev_get_rcu(skb->dev);
2249
2250		if (!in_dev)
2251			return 0;
2252
2253		ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2254				       iph->protocol);
2255		if (!ours)
2256			return 0;
2257
2258		sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2259						   uh->source, iph->saddr,
2260						   dif, sdif);
2261	} else if (skb->pkt_type == PACKET_HOST) {
2262		sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2263					     uh->source, iph->saddr, dif, sdif);
2264	}
2265
2266	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2267		return 0;
2268
2269	skb->sk = sk;
2270	skb->destructor = sock_efree;
2271	dst = READ_ONCE(sk->sk_rx_dst);
2272
2273	if (dst)
2274		dst = dst_check(dst, 0);
2275	if (dst) {
2276		u32 itag = 0;
2277
2278		/* set noref for now.
2279		 * any place which wants to hold dst has to call
2280		 * dst_hold_safe()
2281		 */
2282		skb_dst_set_noref(skb, dst);
2283
2284		/* for unconnected multicast sockets we need to validate
2285		 * the source on each packet
2286		 */
2287		if (!inet_sk(sk)->inet_daddr && in_dev)
2288			return ip_mc_validate_source(skb, iph->daddr,
2289						     iph->saddr, iph->tos,
 
2290						     skb->dev, in_dev, &itag);
2291	}
2292	return 0;
2293}
2294
2295int udp_rcv(struct sk_buff *skb)
2296{
2297	return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2298}
2299
2300void udp_destroy_sock(struct sock *sk)
2301{
2302	struct udp_sock *up = udp_sk(sk);
2303	bool slow = lock_sock_fast(sk);
 
 
 
2304	udp_flush_pending_frames(sk);
2305	unlock_sock_fast(sk, slow);
2306	if (static_key_false(&udp_encap_needed) && up->encap_type) {
2307		void (*encap_destroy)(struct sock *sk);
2308		encap_destroy = READ_ONCE(up->encap_destroy);
2309		if (encap_destroy)
2310			encap_destroy(sk);
 
 
 
 
2311	}
2312}
2313
2314/*
2315 *	Socket option code for UDP
2316 */
2317int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2318		       char __user *optval, unsigned int optlen,
2319		       int (*push_pending_frames)(struct sock *))
2320{
2321	struct udp_sock *up = udp_sk(sk);
2322	int val, valbool;
2323	int err = 0;
2324	int is_udplite = IS_UDPLITE(sk);
2325
2326	if (optlen < sizeof(int))
2327		return -EINVAL;
2328
2329	if (get_user(val, (int __user *)optval))
2330		return -EFAULT;
2331
2332	valbool = val ? 1 : 0;
2333
2334	switch (optname) {
2335	case UDP_CORK:
2336		if (val != 0) {
2337			up->corkflag = 1;
2338		} else {
2339			up->corkflag = 0;
2340			lock_sock(sk);
2341			push_pending_frames(sk);
2342			release_sock(sk);
2343		}
2344		break;
2345
2346	case UDP_ENCAP:
2347		switch (val) {
2348		case 0:
 
2349		case UDP_ENCAP_ESPINUDP:
2350		case UDP_ENCAP_ESPINUDP_NON_IKE:
2351			up->encap_rcv = xfrm4_udp_encap_rcv;
2352			/* FALLTHROUGH */
 
 
 
 
 
 
2353		case UDP_ENCAP_L2TPINUDP:
2354			up->encap_type = val;
2355			udp_encap_enable();
 
 
2356			break;
2357		default:
2358			err = -ENOPROTOOPT;
2359			break;
2360		}
2361		break;
2362
2363	case UDP_NO_CHECK6_TX:
2364		up->no_check6_tx = valbool;
2365		break;
2366
2367	case UDP_NO_CHECK6_RX:
2368		up->no_check6_rx = valbool;
2369		break;
2370
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2371	/*
2372	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
2373	 */
2374	/* The sender sets actual checksum coverage length via this option.
2375	 * The case coverage > packet length is handled by send module. */
2376	case UDPLITE_SEND_CSCOV:
2377		if (!is_udplite)         /* Disable the option on UDP sockets */
2378			return -ENOPROTOOPT;
2379		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2380			val = 8;
2381		else if (val > USHRT_MAX)
2382			val = USHRT_MAX;
2383		up->pcslen = val;
2384		up->pcflag |= UDPLITE_SEND_CC;
2385		break;
2386
2387	/* The receiver specifies a minimum checksum coverage value. To make
2388	 * sense, this should be set to at least 8 (as done below). If zero is
2389	 * used, this again means full checksum coverage.                     */
2390	case UDPLITE_RECV_CSCOV:
2391		if (!is_udplite)         /* Disable the option on UDP sockets */
2392			return -ENOPROTOOPT;
2393		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2394			val = 8;
2395		else if (val > USHRT_MAX)
2396			val = USHRT_MAX;
2397		up->pcrlen = val;
2398		up->pcflag |= UDPLITE_RECV_CC;
2399		break;
2400
2401	default:
2402		err = -ENOPROTOOPT;
2403		break;
2404	}
2405
2406	return err;
2407}
2408EXPORT_SYMBOL(udp_lib_setsockopt);
2409
2410int udp_setsockopt(struct sock *sk, int level, int optname,
2411		   char __user *optval, unsigned int optlen)
2412{
2413	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2414		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
 
2415					  udp_push_pending_frames);
2416	return ip_setsockopt(sk, level, optname, optval, optlen);
2417}
2418
2419#ifdef CONFIG_COMPAT
2420int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2421			  char __user *optval, unsigned int optlen)
2422{
2423	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2424		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2425					  udp_push_pending_frames);
2426	return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2427}
2428#endif
2429
2430int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2431		       char __user *optval, int __user *optlen)
2432{
2433	struct udp_sock *up = udp_sk(sk);
2434	int val, len;
2435
2436	if (get_user(len, optlen))
2437		return -EFAULT;
2438
2439	len = min_t(unsigned int, len, sizeof(int));
2440
2441	if (len < 0)
2442		return -EINVAL;
2443
2444	switch (optname) {
2445	case UDP_CORK:
2446		val = up->corkflag;
2447		break;
2448
2449	case UDP_ENCAP:
2450		val = up->encap_type;
2451		break;
2452
2453	case UDP_NO_CHECK6_TX:
2454		val = up->no_check6_tx;
2455		break;
2456
2457	case UDP_NO_CHECK6_RX:
2458		val = up->no_check6_rx;
2459		break;
2460
 
 
 
 
 
 
 
 
2461	/* The following two cannot be changed on UDP sockets, the return is
2462	 * always 0 (which corresponds to the full checksum coverage of UDP). */
2463	case UDPLITE_SEND_CSCOV:
2464		val = up->pcslen;
2465		break;
2466
2467	case UDPLITE_RECV_CSCOV:
2468		val = up->pcrlen;
2469		break;
2470
2471	default:
2472		return -ENOPROTOOPT;
2473	}
2474
2475	if (put_user(len, optlen))
2476		return -EFAULT;
2477	if (copy_to_user(optval, &val, len))
2478		return -EFAULT;
2479	return 0;
2480}
2481EXPORT_SYMBOL(udp_lib_getsockopt);
2482
2483int udp_getsockopt(struct sock *sk, int level, int optname,
2484		   char __user *optval, int __user *optlen)
2485{
2486	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2487		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2488	return ip_getsockopt(sk, level, optname, optval, optlen);
2489}
2490
2491#ifdef CONFIG_COMPAT
2492int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2493				 char __user *optval, int __user *optlen)
2494{
2495	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2496		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2497	return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2498}
2499#endif
2500/**
2501 * 	udp_poll - wait for a UDP event.
2502 *	@file - file struct
2503 *	@sock - socket
2504 *	@wait - poll table
2505 *
2506 *	This is same as datagram poll, except for the special case of
2507 *	blocking sockets. If application is using a blocking fd
2508 *	and a packet with checksum error is in the queue;
2509 *	then it could get return from select indicating data available
2510 *	but then block when reading it. Add special case code
2511 *	to work around these arguably broken applications.
2512 */
2513__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2514{
2515	__poll_t mask = datagram_poll(file, sock, wait);
2516	struct sock *sk = sock->sk;
2517
2518	if (!skb_queue_empty(&udp_sk(sk)->reader_queue))
2519		mask |= EPOLLIN | EPOLLRDNORM;
2520
2521	/* Check for false positives due to checksum errors */
2522	if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2523	    !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2524		mask &= ~(EPOLLIN | EPOLLRDNORM);
2525
2526	return mask;
2527
2528}
2529EXPORT_SYMBOL(udp_poll);
2530
2531int udp_abort(struct sock *sk, int err)
2532{
2533	lock_sock(sk);
2534
 
 
 
 
 
 
2535	sk->sk_err = err;
2536	sk->sk_error_report(sk);
2537	__udp_disconnect(sk, 0);
2538
 
2539	release_sock(sk);
2540
2541	return 0;
2542}
2543EXPORT_SYMBOL_GPL(udp_abort);
2544
2545struct proto udp_prot = {
2546	.name			= "UDP",
2547	.owner			= THIS_MODULE,
2548	.close			= udp_lib_close,
2549	.pre_connect		= udp_pre_connect,
2550	.connect		= ip4_datagram_connect,
2551	.disconnect		= udp_disconnect,
2552	.ioctl			= udp_ioctl,
2553	.init			= udp_init_sock,
2554	.destroy		= udp_destroy_sock,
2555	.setsockopt		= udp_setsockopt,
2556	.getsockopt		= udp_getsockopt,
2557	.sendmsg		= udp_sendmsg,
2558	.recvmsg		= udp_recvmsg,
2559	.sendpage		= udp_sendpage,
2560	.release_cb		= ip4_datagram_release_cb,
2561	.hash			= udp_lib_hash,
2562	.unhash			= udp_lib_unhash,
2563	.rehash			= udp_v4_rehash,
2564	.get_port		= udp_v4_get_port,
 
 
 
2565	.memory_allocated	= &udp_memory_allocated,
2566	.sysctl_mem		= sysctl_udp_mem,
2567	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2568	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2569	.obj_size		= sizeof(struct udp_sock),
2570	.h.udp_table		= &udp_table,
2571#ifdef CONFIG_COMPAT
2572	.compat_setsockopt	= compat_udp_setsockopt,
2573	.compat_getsockopt	= compat_udp_getsockopt,
2574#endif
2575	.diag_destroy		= udp_abort,
2576};
2577EXPORT_SYMBOL(udp_prot);
2578
2579/* ------------------------------------------------------------------------ */
2580#ifdef CONFIG_PROC_FS
2581
2582static struct sock *udp_get_first(struct seq_file *seq, int start)
2583{
2584	struct sock *sk;
 
2585	struct udp_iter_state *state = seq->private;
2586	struct net *net = seq_file_net(seq);
2587
2588	for (state->bucket = start; state->bucket <= state->udp_table->mask;
 
 
 
 
 
2589	     ++state->bucket) {
2590		struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2591
2592		if (hlist_empty(&hslot->head))
2593			continue;
2594
2595		spin_lock_bh(&hslot->lock);
2596		sk_for_each(sk, &hslot->head) {
2597			if (!net_eq(sock_net(sk), net))
2598				continue;
2599			if (sk->sk_family == state->family)
 
2600				goto found;
2601		}
2602		spin_unlock_bh(&hslot->lock);
2603	}
2604	sk = NULL;
2605found:
2606	return sk;
2607}
2608
2609static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2610{
 
2611	struct udp_iter_state *state = seq->private;
2612	struct net *net = seq_file_net(seq);
2613
 
 
 
 
 
2614	do {
2615		sk = sk_next(sk);
2616	} while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
 
 
2617
2618	if (!sk) {
2619		if (state->bucket <= state->udp_table->mask)
2620			spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2621		return udp_get_first(seq, state->bucket + 1);
2622	}
2623	return sk;
2624}
2625
2626static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2627{
2628	struct sock *sk = udp_get_first(seq, 0);
2629
2630	if (sk)
2631		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2632			--pos;
2633	return pos ? NULL : sk;
2634}
2635
2636static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2637{
2638	struct udp_iter_state *state = seq->private;
2639	state->bucket = MAX_UDP_PORTS;
2640
2641	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2642}
 
2643
2644static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2645{
2646	struct sock *sk;
2647
2648	if (v == SEQ_START_TOKEN)
2649		sk = udp_get_idx(seq, 0);
2650	else
2651		sk = udp_get_next(seq, v);
2652
2653	++*pos;
2654	return sk;
2655}
 
2656
2657static void udp_seq_stop(struct seq_file *seq, void *v)
2658{
 
2659	struct udp_iter_state *state = seq->private;
2660
2661	if (state->bucket <= state->udp_table->mask)
2662		spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2663}
2664
2665int udp_seq_open(struct inode *inode, struct file *file)
2666{
2667	struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2668	struct udp_iter_state *s;
2669	int err;
2670
2671	err = seq_open_net(inode, file, &afinfo->seq_ops,
2672			   sizeof(struct udp_iter_state));
2673	if (err < 0)
2674		return err;
2675
2676	s = ((struct seq_file *)file->private_data)->private;
2677	s->family		= afinfo->family;
2678	s->udp_table		= afinfo->udp_table;
2679	return err;
2680}
2681EXPORT_SYMBOL(udp_seq_open);
2682
2683/* ------------------------------------------------------------------------ */
2684int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2685{
2686	struct proc_dir_entry *p;
2687	int rc = 0;
2688
2689	afinfo->seq_ops.start		= udp_seq_start;
2690	afinfo->seq_ops.next		= udp_seq_next;
2691	afinfo->seq_ops.stop		= udp_seq_stop;
2692
2693	p = proc_create_data(afinfo->name, 0444, net->proc_net,
2694			     afinfo->seq_fops, afinfo);
2695	if (!p)
2696		rc = -ENOMEM;
2697	return rc;
2698}
2699EXPORT_SYMBOL(udp_proc_register);
2700
2701void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2702{
2703	remove_proc_entry(afinfo->name, net->proc_net);
2704}
2705EXPORT_SYMBOL(udp_proc_unregister);
2706
2707/* ------------------------------------------------------------------------ */
2708static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2709		int bucket)
2710{
2711	struct inet_sock *inet = inet_sk(sp);
2712	__be32 dest = inet->inet_daddr;
2713	__be32 src  = inet->inet_rcv_saddr;
2714	__u16 destp	  = ntohs(inet->inet_dport);
2715	__u16 srcp	  = ntohs(inet->inet_sport);
2716
2717	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2718		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2719		bucket, src, srcp, dest, destp, sp->sk_state,
2720		sk_wmem_alloc_get(sp),
2721		sk_rmem_alloc_get(sp),
2722		0, 0L, 0,
2723		from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2724		0, sock_i_ino(sp),
2725		refcount_read(&sp->sk_refcnt), sp,
2726		atomic_read(&sp->sk_drops));
2727}
2728
2729int udp4_seq_show(struct seq_file *seq, void *v)
2730{
2731	seq_setwidth(seq, 127);
2732	if (v == SEQ_START_TOKEN)
2733		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2734			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2735			   "inode ref pointer drops");
2736	else {
2737		struct udp_iter_state *state = seq->private;
2738
2739		udp4_format_sock(v, seq, state->bucket);
2740	}
2741	seq_pad(seq, '\n');
2742	return 0;
2743}
2744
2745static const struct file_operations udp_afinfo_seq_fops = {
2746	.open     = udp_seq_open,
2747	.read     = seq_read,
2748	.llseek   = seq_lseek,
2749	.release  = seq_release_net
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2750};
 
 
 
 
 
 
 
 
 
2751
2752/* ------------------------------------------------------------------------ */
2753static struct udp_seq_afinfo udp4_seq_afinfo = {
2754	.name		= "udp",
2755	.family		= AF_INET,
2756	.udp_table	= &udp_table,
2757	.seq_fops	= &udp_afinfo_seq_fops,
2758	.seq_ops	= {
2759		.show		= udp4_seq_show,
2760	},
2761};
2762
2763static int __net_init udp4_proc_init_net(struct net *net)
2764{
2765	return udp_proc_register(net, &udp4_seq_afinfo);
 
 
 
2766}
2767
2768static void __net_exit udp4_proc_exit_net(struct net *net)
2769{
2770	udp_proc_unregister(net, &udp4_seq_afinfo);
2771}
2772
2773static struct pernet_operations udp4_net_ops = {
2774	.init = udp4_proc_init_net,
2775	.exit = udp4_proc_exit_net,
2776};
2777
2778int __init udp4_proc_init(void)
2779{
2780	return register_pernet_subsys(&udp4_net_ops);
2781}
2782
2783void udp4_proc_exit(void)
2784{
2785	unregister_pernet_subsys(&udp4_net_ops);
2786}
2787#endif /* CONFIG_PROC_FS */
2788
2789static __initdata unsigned long uhash_entries;
2790static int __init set_uhash_entries(char *str)
2791{
2792	ssize_t ret;
2793
2794	if (!str)
2795		return 0;
2796
2797	ret = kstrtoul(str, 0, &uhash_entries);
2798	if (ret)
2799		return 0;
2800
2801	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2802		uhash_entries = UDP_HTABLE_SIZE_MIN;
2803	return 1;
2804}
2805__setup("uhash_entries=", set_uhash_entries);
2806
2807void __init udp_table_init(struct udp_table *table, const char *name)
2808{
2809	unsigned int i;
2810
2811	table->hash = alloc_large_system_hash(name,
2812					      2 * sizeof(struct udp_hslot),
2813					      uhash_entries,
2814					      21, /* one slot per 2 MB */
2815					      0,
2816					      &table->log,
2817					      &table->mask,
2818					      UDP_HTABLE_SIZE_MIN,
2819					      64 * 1024);
2820
2821	table->hash2 = table->hash + (table->mask + 1);
2822	for (i = 0; i <= table->mask; i++) {
2823		INIT_HLIST_HEAD(&table->hash[i].head);
2824		table->hash[i].count = 0;
2825		spin_lock_init(&table->hash[i].lock);
2826	}
2827	for (i = 0; i <= table->mask; i++) {
2828		INIT_HLIST_HEAD(&table->hash2[i].head);
2829		table->hash2[i].count = 0;
2830		spin_lock_init(&table->hash2[i].lock);
2831	}
2832}
2833
2834u32 udp_flow_hashrnd(void)
2835{
2836	static u32 hashrnd __read_mostly;
2837
2838	net_get_random_once(&hashrnd, sizeof(hashrnd));
2839
2840	return hashrnd;
2841}
2842EXPORT_SYMBOL(udp_flow_hashrnd);
2843
2844static void __udp_sysctl_init(struct net *net)
2845{
2846	net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2847	net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2848
2849#ifdef CONFIG_NET_L3_MASTER_DEV
2850	net->ipv4.sysctl_udp_l3mdev_accept = 0;
2851#endif
2852}
2853
2854static int __net_init udp_sysctl_init(struct net *net)
2855{
2856	__udp_sysctl_init(net);
2857	return 0;
2858}
2859
2860static struct pernet_operations __net_initdata udp_sysctl_ops = {
2861	.init	= udp_sysctl_init,
2862};
2863
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2864void __init udp_init(void)
2865{
2866	unsigned long limit;
2867	unsigned int i;
2868
2869	udp_table_init(&udp_table, "UDP");
2870	limit = nr_free_buffer_pages() / 8;
2871	limit = max(limit, 128UL);
2872	sysctl_udp_mem[0] = limit / 4 * 3;
2873	sysctl_udp_mem[1] = limit;
2874	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2875
2876	__udp_sysctl_init(&init_net);
2877
2878	/* 16 spinlocks per cpu */
2879	udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2880	udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2881				GFP_KERNEL);
2882	if (!udp_busylocks)
2883		panic("UDP: failed to alloc udp_busylocks\n");
2884	for (i = 0; i < (1U << udp_busylocks_log); i++)
2885		spin_lock_init(udp_busylocks + i);
2886
2887	if (register_pernet_subsys(&udp_sysctl_ops))
2888		panic("UDP: failed to init sysctl parameters.\n");
 
 
 
 
2889}