Linux Audio

Check our new training course

Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		The User Datagram Protocol (UDP).
   8 *
   9 * Authors:	Ross Biro
  10 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  12 *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
  13 *		Hirokazu Takahashi, <taka@valinux.co.jp>
  14 *
  15 * Fixes:
  16 *		Alan Cox	:	verify_area() calls
  17 *		Alan Cox	: 	stopped close while in use off icmp
  18 *					messages. Not a fix but a botch that
  19 *					for udp at least is 'valid'.
  20 *		Alan Cox	:	Fixed icmp handling properly
  21 *		Alan Cox	: 	Correct error for oversized datagrams
  22 *		Alan Cox	:	Tidied select() semantics.
  23 *		Alan Cox	:	udp_err() fixed properly, also now
  24 *					select and read wake correctly on errors
  25 *		Alan Cox	:	udp_send verify_area moved to avoid mem leak
  26 *		Alan Cox	:	UDP can count its memory
  27 *		Alan Cox	:	send to an unknown connection causes
  28 *					an ECONNREFUSED off the icmp, but
  29 *					does NOT close.
  30 *		Alan Cox	:	Switched to new sk_buff handlers. No more backlog!
  31 *		Alan Cox	:	Using generic datagram code. Even smaller and the PEEK
  32 *					bug no longer crashes it.
  33 *		Fred Van Kempen	: 	Net2e support for sk->broadcast.
  34 *		Alan Cox	:	Uses skb_free_datagram
  35 *		Alan Cox	:	Added get/set sockopt support.
  36 *		Alan Cox	:	Broadcasting without option set returns EACCES.
  37 *		Alan Cox	:	No wakeup calls. Instead we now use the callbacks.
  38 *		Alan Cox	:	Use ip_tos and ip_ttl
  39 *		Alan Cox	:	SNMP Mibs
  40 *		Alan Cox	:	MSG_DONTROUTE, and 0.0.0.0 support.
  41 *		Matt Dillon	:	UDP length checks.
  42 *		Alan Cox	:	Smarter af_inet used properly.
  43 *		Alan Cox	:	Use new kernel side addressing.
  44 *		Alan Cox	:	Incorrect return on truncated datagram receive.
  45 *	Arnt Gulbrandsen 	:	New udp_send and stuff
  46 *		Alan Cox	:	Cache last socket
  47 *		Alan Cox	:	Route cache
  48 *		Jon Peatfield	:	Minor efficiency fix to sendto().
  49 *		Mike Shaver	:	RFC1122 checks.
  50 *		Alan Cox	:	Nonblocking error fix.
  51 *	Willy Konynenberg	:	Transparent proxying support.
  52 *		Mike McLagan	:	Routing by source
  53 *		David S. Miller	:	New socket lookup architecture.
  54 *					Last socket cache retained as it
  55 *					does have a high hit rate.
  56 *		Olaf Kirch	:	Don't linearise iovec on sendmsg.
  57 *		Andi Kleen	:	Some cleanups, cache destination entry
  58 *					for connect.
  59 *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
  60 *		Melvin Smith	:	Check msg_name not msg_namelen in sendto(),
  61 *					return ENOTCONN for unconnected sockets (POSIX)
  62 *		Janos Farkas	:	don't deliver multi/broadcasts to a different
  63 *					bound-to-device socket
  64 *	Hirokazu Takahashi	:	HW checksumming for outgoing UDP
  65 *					datagrams.
  66 *	Hirokazu Takahashi	:	sendfile() on UDP works now.
  67 *		Arnaldo C. Melo :	convert /proc/net/udp to seq_file
  68 *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
  69 *	Alexey Kuznetsov:		allow both IPv4 and IPv6 sockets to bind
  70 *					a single port at the same time.
  71 *	Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
  72 *	James Chapman		:	Add L2TP encapsulation type.
 
 
 
 
 
 
  73 */
  74
  75#define pr_fmt(fmt) "UDP: " fmt
  76
  77#include <linux/uaccess.h>
  78#include <asm/ioctls.h>
  79#include <linux/memblock.h>
  80#include <linux/highmem.h>
  81#include <linux/swap.h>
  82#include <linux/types.h>
  83#include <linux/fcntl.h>
  84#include <linux/module.h>
  85#include <linux/socket.h>
  86#include <linux/sockios.h>
  87#include <linux/igmp.h>
  88#include <linux/inetdevice.h>
  89#include <linux/in.h>
  90#include <linux/errno.h>
  91#include <linux/timer.h>
  92#include <linux/mm.h>
  93#include <linux/inet.h>
  94#include <linux/netdevice.h>
  95#include <linux/slab.h>
  96#include <net/tcp_states.h>
  97#include <linux/skbuff.h>
  98#include <linux/proc_fs.h>
  99#include <linux/seq_file.h>
 100#include <net/net_namespace.h>
 101#include <net/icmp.h>
 102#include <net/inet_hashtables.h>
 103#include <net/ip_tunnels.h>
 104#include <net/route.h>
 105#include <net/checksum.h>
 106#include <net/xfrm.h>
 107#include <trace/events/udp.h>
 108#include <linux/static_key.h>
 109#include <trace/events/skb.h>
 110#include <net/busy_poll.h>
 111#include "udp_impl.h"
 112#include <net/sock_reuseport.h>
 113#include <net/addrconf.h>
 114#include <net/udp_tunnel.h>
 115
 116struct udp_table udp_table __read_mostly;
 117EXPORT_SYMBOL(udp_table);
 118
 119long sysctl_udp_mem[3] __read_mostly;
 120EXPORT_SYMBOL(sysctl_udp_mem);
 121
 
 
 
 
 
 
 122atomic_long_t udp_memory_allocated;
 123EXPORT_SYMBOL(udp_memory_allocated);
 124
 125#define MAX_UDP_PORTS 65536
 126#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
 127
 128static int udp_lib_lport_inuse(struct net *net, __u16 num,
 129			       const struct udp_hslot *hslot,
 130			       unsigned long *bitmap,
 131			       struct sock *sk, unsigned int log)
 
 
 
 132{
 133	struct sock *sk2;
 134	kuid_t uid = sock_i_uid(sk);
 135
 136	sk_for_each(sk2, &hslot->head) {
 137		if (net_eq(sock_net(sk2), net) &&
 138		    sk2 != sk &&
 139		    (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
 140		    (!sk2->sk_reuse || !sk->sk_reuse) &&
 141		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 142		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 143		    inet_rcv_saddr_equal(sk, sk2, true)) {
 144			if (sk2->sk_reuseport && sk->sk_reuseport &&
 145			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 146			    uid_eq(uid, sock_i_uid(sk2))) {
 147				if (!bitmap)
 148					return 0;
 149			} else {
 150				if (!bitmap)
 151					return 1;
 152				__set_bit(udp_sk(sk2)->udp_port_hash >> log,
 153					  bitmap);
 154			}
 
 155		}
 156	}
 157	return 0;
 158}
 159
 160/*
 161 * Note: we still hold spinlock of primary hash chain, so no other writer
 162 * can insert/delete a socket with local_port == num
 163 */
 164static int udp_lib_lport_inuse2(struct net *net, __u16 num,
 165				struct udp_hslot *hslot2,
 166				struct sock *sk)
 
 
 167{
 168	struct sock *sk2;
 169	kuid_t uid = sock_i_uid(sk);
 170	int res = 0;
 171
 172	spin_lock(&hslot2->lock);
 173	udp_portaddr_for_each_entry(sk2, &hslot2->head) {
 174		if (net_eq(sock_net(sk2), net) &&
 175		    sk2 != sk &&
 176		    (udp_sk(sk2)->udp_port_hash == num) &&
 177		    (!sk2->sk_reuse || !sk->sk_reuse) &&
 178		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 179		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 180		    inet_rcv_saddr_equal(sk, sk2, true)) {
 181			if (sk2->sk_reuseport && sk->sk_reuseport &&
 182			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 183			    uid_eq(uid, sock_i_uid(sk2))) {
 184				res = 0;
 185			} else {
 186				res = 1;
 187			}
 188			break;
 189		}
 190	}
 191	spin_unlock(&hslot2->lock);
 192	return res;
 193}
 194
 195static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
 196{
 197	struct net *net = sock_net(sk);
 198	kuid_t uid = sock_i_uid(sk);
 199	struct sock *sk2;
 200
 201	sk_for_each(sk2, &hslot->head) {
 202		if (net_eq(sock_net(sk2), net) &&
 203		    sk2 != sk &&
 204		    sk2->sk_family == sk->sk_family &&
 205		    ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
 206		    (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
 207		    (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 208		    sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
 209		    inet_rcv_saddr_equal(sk, sk2, false)) {
 210			return reuseport_add_sock(sk, sk2,
 211						  inet_rcv_saddr_any(sk));
 212		}
 213	}
 214
 215	return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
 216}
 217
 218/**
 219 *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
 220 *
 221 *  @sk:          socket struct in question
 222 *  @snum:        port number to look up
 
 223 *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
 224 *                   with NULL address
 225 */
 226int udp_lib_get_port(struct sock *sk, unsigned short snum,
 
 
 227		     unsigned int hash2_nulladdr)
 228{
 229	struct udp_hslot *hslot, *hslot2;
 230	struct udp_table *udptable = sk->sk_prot->h.udp_table;
 231	int    error = 1;
 232	struct net *net = sock_net(sk);
 233
 234	if (!snum) {
 235		int low, high, remaining;
 236		unsigned int rand;
 237		unsigned short first, last;
 238		DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
 239
 240		inet_get_local_port_range(net, &low, &high);
 241		remaining = (high - low) + 1;
 242
 243		rand = prandom_u32();
 244		first = reciprocal_scale(rand, remaining) + low;
 245		/*
 246		 * force rand to be an odd multiple of UDP_HTABLE_SIZE
 247		 */
 248		rand = (rand | 1) * (udptable->mask + 1);
 249		last = first + udptable->mask + 1;
 250		do {
 251			hslot = udp_hashslot(udptable, net, first);
 252			bitmap_zero(bitmap, PORTS_PER_CHAIN);
 253			spin_lock_bh(&hslot->lock);
 254			udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
 255					    udptable->log);
 256
 257			snum = first;
 258			/*
 259			 * Iterate on all possible values of snum for this hash.
 260			 * Using steps of an odd multiple of UDP_HTABLE_SIZE
 261			 * give us randomization and full range coverage.
 262			 */
 263			do {
 264				if (low <= snum && snum <= high &&
 265				    !test_bit(snum >> udptable->log, bitmap) &&
 266				    !inet_is_local_reserved_port(net, snum))
 267					goto found;
 268				snum += rand;
 269			} while (snum != first);
 270			spin_unlock_bh(&hslot->lock);
 271			cond_resched();
 272		} while (++first != last);
 273		goto fail;
 274	} else {
 275		hslot = udp_hashslot(udptable, net, snum);
 276		spin_lock_bh(&hslot->lock);
 277		if (hslot->count > 10) {
 278			int exist;
 279			unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
 280
 281			slot2          &= udptable->mask;
 282			hash2_nulladdr &= udptable->mask;
 283
 284			hslot2 = udp_hashslot2(udptable, slot2);
 285			if (hslot->count < hslot2->count)
 286				goto scan_primary_hash;
 287
 288			exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
 
 289			if (!exist && (hash2_nulladdr != slot2)) {
 290				hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
 291				exist = udp_lib_lport_inuse2(net, snum, hslot2,
 292							     sk);
 293			}
 294			if (exist)
 295				goto fail_unlock;
 296			else
 297				goto found;
 298		}
 299scan_primary_hash:
 300		if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
 
 301			goto fail_unlock;
 302	}
 303found:
 304	inet_sk(sk)->inet_num = snum;
 305	udp_sk(sk)->udp_port_hash = snum;
 306	udp_sk(sk)->udp_portaddr_hash ^= snum;
 307	if (sk_unhashed(sk)) {
 308		if (sk->sk_reuseport &&
 309		    udp_reuseport_add_sock(sk, hslot)) {
 310			inet_sk(sk)->inet_num = 0;
 311			udp_sk(sk)->udp_port_hash = 0;
 312			udp_sk(sk)->udp_portaddr_hash ^= snum;
 313			goto fail_unlock;
 314		}
 315
 316		sk_add_node_rcu(sk, &hslot->head);
 317		hslot->count++;
 318		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 319
 320		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 321		spin_lock(&hslot2->lock);
 322		if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 323		    sk->sk_family == AF_INET6)
 324			hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
 325					   &hslot2->head);
 326		else
 327			hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
 328					   &hslot2->head);
 329		hslot2->count++;
 330		spin_unlock(&hslot2->lock);
 331	}
 332	sock_set_flag(sk, SOCK_RCU_FREE);
 333	error = 0;
 334fail_unlock:
 335	spin_unlock_bh(&hslot->lock);
 336fail:
 337	return error;
 338}
 339EXPORT_SYMBOL(udp_lib_get_port);
 340
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 341int udp_v4_get_port(struct sock *sk, unsigned short snum)
 342{
 343	unsigned int hash2_nulladdr =
 344		ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
 345	unsigned int hash2_partial =
 346		ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
 347
 348	/* precompute partial secondary hash */
 349	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
 350	return udp_lib_get_port(sk, snum, hash2_nulladdr);
 351}
 352
 353static int compute_score(struct sock *sk, struct net *net,
 354			 __be32 saddr, __be16 sport,
 355			 __be32 daddr, unsigned short hnum,
 356			 int dif, int sdif)
 357{
 358	int score;
 359	struct inet_sock *inet;
 360	bool dev_match;
 361
 362	if (!net_eq(sock_net(sk), net) ||
 363	    udp_sk(sk)->udp_port_hash != hnum ||
 364	    ipv6_only_sock(sk))
 365		return -1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 366
 367	if (sk->sk_rcv_saddr != daddr)
 368		return -1;
 
 
 
 
 
 
 
 369
 370	score = (sk->sk_family == PF_INET) ? 2 : 1;
 
 371
 372	inet = inet_sk(sk);
 373	if (inet->inet_daddr) {
 374		if (inet->inet_daddr != saddr)
 375			return -1;
 376		score += 4;
 377	}
 378
 379	if (inet->inet_dport) {
 380		if (inet->inet_dport != sport)
 381			return -1;
 382		score += 4;
 383	}
 384
 385	dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
 386					dif, sdif);
 387	if (!dev_match)
 388		return -1;
 389	score += 4;
 390
 391	if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
 392		score++;
 
 
 
 
 
 
 
 
 
 393	return score;
 394}
 395
 396static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
 397		       const __u16 lport, const __be32 faddr,
 398		       const __be16 fport)
 399{
 400	static u32 udp_ehash_secret __read_mostly;
 401
 402	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
 403
 404	return __inet_ehashfn(laddr, lport, faddr, fport,
 405			      udp_ehash_secret + net_hash_mix(net));
 406}
 407
 408/* called with rcu_read_lock() */
 409static struct sock *udp4_lib_lookup2(struct net *net,
 410				     __be32 saddr, __be16 sport,
 411				     __be32 daddr, unsigned int hnum,
 412				     int dif, int sdif,
 413				     struct udp_hslot *hslot2,
 414				     struct sk_buff *skb)
 415{
 416	struct sock *sk, *result;
 
 417	int score, badness;
 418	u32 hash = 0;
 419
 
 420	result = NULL;
 421	badness = 0;
 422	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
 423		score = compute_score(sk, net, saddr, sport,
 424				      daddr, hnum, dif, sdif);
 425		if (score > badness) {
 426			if (sk->sk_reuseport &&
 427			    sk->sk_state != TCP_ESTABLISHED) {
 428				hash = udp_ehashfn(net, daddr, hnum,
 429						   saddr, sport);
 430				result = reuseport_select_sock(sk, hash, skb,
 431							sizeof(struct udphdr));
 432				if (result && !reuseport_has_conns(sk, false))
 433					return result;
 434			}
 435			badness = score;
 436			result = sk;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 437		}
 438	}
 439	return result;
 440}
 441
 442/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
 443 * harder than this. -DaveM
 444 */
 445struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
 446		__be16 sport, __be32 daddr, __be16 dport, int dif,
 447		int sdif, struct udp_table *udptable, struct sk_buff *skb)
 448{
 449	struct sock *result;
 
 450	unsigned short hnum = ntohs(dport);
 451	unsigned int hash2, slot2;
 452	struct udp_hslot *hslot2;
 
 453
 454	hash2 = ipv4_portaddr_hash(net, daddr, hnum);
 455	slot2 = hash2 & udptable->mask;
 456	hslot2 = &udptable->hash2[slot2];
 457
 458	result = udp4_lib_lookup2(net, saddr, sport,
 459				  daddr, hnum, dif, sdif,
 460				  hslot2, skb);
 461	if (!result) {
 462		hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
 463		slot2 = hash2 & udptable->mask;
 464		hslot2 = &udptable->hash2[slot2];
 
 
 465
 466		result = udp4_lib_lookup2(net, saddr, sport,
 467					  htonl(INADDR_ANY), hnum, dif, sdif,
 468					  hslot2, skb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 469	}
 470	if (IS_ERR(result))
 471		return NULL;
 472	return result;
 473}
 474EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
 475
 476static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
 477						 __be16 sport, __be16 dport,
 478						 struct udp_table *udptable)
 479{
 
 480	const struct iphdr *iph = ip_hdr(skb);
 481
 482	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
 483				 iph->daddr, dport, inet_iif(skb),
 484				 inet_sdif(skb), udptable, skb);
 485}
 486
 487struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
 488				 __be16 sport, __be16 dport)
 489{
 490	const struct iphdr *iph = ip_hdr(skb);
 491
 492	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
 493				 iph->daddr, dport, inet_iif(skb),
 494				 inet_sdif(skb), &udp_table, NULL);
 495}
 496EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
 497
 498/* Must be called under rcu_read_lock().
 499 * Does increment socket refcount.
 500 */
 501#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
 502struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 503			     __be32 daddr, __be16 dport, int dif)
 504{
 505	struct sock *sk;
 506
 507	sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
 508			       dif, 0, &udp_table, NULL);
 509	if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
 510		sk = NULL;
 511	return sk;
 512}
 513EXPORT_SYMBOL_GPL(udp4_lib_lookup);
 514#endif
 515
 516static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
 517				       __be16 loc_port, __be32 loc_addr,
 518				       __be16 rmt_port, __be32 rmt_addr,
 519				       int dif, int sdif, unsigned short hnum)
 520{
 521	struct inet_sock *inet = inet_sk(sk);
 522
 523	if (!net_eq(sock_net(sk), net) ||
 524	    udp_sk(sk)->udp_port_hash != hnum ||
 525	    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
 526	    (inet->inet_dport != rmt_port && inet->inet_dport) ||
 527	    (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
 528	    ipv6_only_sock(sk) ||
 529	    !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
 530		return false;
 531	if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
 532		return false;
 533	return true;
 534}
 535
 536DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
 537void udp_encap_enable(void)
 538{
 539	static_branch_inc(&udp_encap_needed_key);
 540}
 541EXPORT_SYMBOL(udp_encap_enable);
 542
 543/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
 544 * through error handlers in encapsulations looking for a match.
 545 */
 546static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
 547{
 548	int i;
 
 
 549
 550	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
 551		int (*handler)(struct sk_buff *skb, u32 info);
 552		const struct ip_tunnel_encap_ops *encap;
 553
 554		encap = rcu_dereference(iptun_encaps[i]);
 555		if (!encap)
 
 
 
 
 
 
 556			continue;
 557		handler = encap->err_handler;
 558		if (handler && !handler(skb, info))
 559			return 0;
 560	}
 561
 562	return -ENOENT;
 563}
 564
 565/* Try to match ICMP errors to UDP tunnels by looking up a socket without
 566 * reversing source and destination port: this will match tunnels that force the
 567 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
 568 * lwtunnels might actually break this assumption by being configured with
 569 * different destination ports on endpoints, in this case we won't be able to
 570 * trace ICMP messages back to them.
 571 *
 572 * If this doesn't match any socket, probe tunnels with arbitrary destination
 573 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
 574 * we've sent packets to won't necessarily match the local destination port.
 575 *
 576 * Then ask the tunnel implementation to match the error against a valid
 577 * association.
 578 *
 579 * Return an error if we can't find a match, the socket if we need further
 580 * processing, zero otherwise.
 581 */
 582static struct sock *__udp4_lib_err_encap(struct net *net,
 583					 const struct iphdr *iph,
 584					 struct udphdr *uh,
 585					 struct udp_table *udptable,
 586					 struct sk_buff *skb, u32 info)
 587{
 588	int network_offset, transport_offset;
 589	struct sock *sk;
 590
 591	network_offset = skb_network_offset(skb);
 592	transport_offset = skb_transport_offset(skb);
 593
 594	/* Network header needs to point to the outer IPv4 header inside ICMP */
 595	skb_reset_network_header(skb);
 596
 597	/* Transport header needs to point to the UDP header */
 598	skb_set_transport_header(skb, iph->ihl << 2);
 599
 600	sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
 601			       iph->saddr, uh->dest, skb->dev->ifindex, 0,
 602			       udptable, NULL);
 603	if (sk) {
 604		int (*lookup)(struct sock *sk, struct sk_buff *skb);
 605		struct udp_sock *up = udp_sk(sk);
 606
 607		lookup = READ_ONCE(up->encap_err_lookup);
 608		if (!lookup || lookup(sk, skb))
 609			sk = NULL;
 610	}
 611
 612	if (!sk)
 613		sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
 614
 615	skb_set_transport_header(skb, transport_offset);
 616	skb_set_network_header(skb, network_offset);
 617
 618	return sk;
 619}
 620
 621/*
 622 * This routine is called by the ICMP module when it gets some
 623 * sort of error condition.  If err < 0 then the socket should
 624 * be closed and the error returned to the user.  If err > 0
 625 * it's just the icmp type << 8 | icmp code.
 626 * Header points to the ip header of the error packet. We move
 627 * on past this. Then (as it used to claim before adjustment)
 628 * header points to the first 8 bytes of the udp header.  We need
 629 * to find the appropriate port.
 630 */
 631
 632int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
 633{
 634	struct inet_sock *inet;
 635	const struct iphdr *iph = (const struct iphdr *)skb->data;
 636	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
 637	const int type = icmp_hdr(skb)->type;
 638	const int code = icmp_hdr(skb)->code;
 639	bool tunnel = false;
 640	struct sock *sk;
 641	int harderr;
 642	int err;
 643	struct net *net = dev_net(skb->dev);
 644
 645	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
 646			       iph->saddr, uh->source, skb->dev->ifindex,
 647			       inet_sdif(skb), udptable, NULL);
 648	if (!sk) {
 649		/* No socket for error: try tunnels before discarding */
 650		sk = ERR_PTR(-ENOENT);
 651		if (static_branch_unlikely(&udp_encap_needed_key)) {
 652			sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
 653						  info);
 654			if (!sk)
 655				return 0;
 656		}
 657
 658		if (IS_ERR(sk)) {
 659			__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 660			return PTR_ERR(sk);
 661		}
 662
 663		tunnel = true;
 664	}
 665
 666	err = 0;
 667	harderr = 0;
 668	inet = inet_sk(sk);
 669
 670	switch (type) {
 671	default:
 672	case ICMP_TIME_EXCEEDED:
 673		err = EHOSTUNREACH;
 674		break;
 675	case ICMP_SOURCE_QUENCH:
 676		goto out;
 677	case ICMP_PARAMETERPROB:
 678		err = EPROTO;
 679		harderr = 1;
 680		break;
 681	case ICMP_DEST_UNREACH:
 682		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
 683			ipv4_sk_update_pmtu(skb, sk, info);
 684			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
 685				err = EMSGSIZE;
 686				harderr = 1;
 687				break;
 688			}
 689			goto out;
 690		}
 691		err = EHOSTUNREACH;
 692		if (code <= NR_ICMP_UNREACH) {
 693			harderr = icmp_err_convert[code].fatal;
 694			err = icmp_err_convert[code].errno;
 695		}
 696		break;
 697	case ICMP_REDIRECT:
 698		ipv4_sk_redirect(skb, sk);
 699		goto out;
 700	}
 701
 702	/*
 703	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
 704	 *	4.1.3.3.
 705	 */
 706	if (tunnel) {
 707		/* ...not for tunnels though: we don't have a sending socket */
 708		goto out;
 709	}
 710	if (!inet->recverr) {
 711		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
 712			goto out;
 713	} else
 714		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
 715
 716	sk->sk_err = err;
 717	sk->sk_error_report(sk);
 718out:
 719	return 0;
 720}
 721
 722int udp_err(struct sk_buff *skb, u32 info)
 723{
 724	return __udp4_lib_err(skb, info, &udp_table);
 725}
 726
 727/*
 728 * Throw away all pending data and cancel the corking. Socket is locked.
 729 */
 730void udp_flush_pending_frames(struct sock *sk)
 731{
 732	struct udp_sock *up = udp_sk(sk);
 733
 734	if (up->pending) {
 735		up->len = 0;
 736		up->pending = 0;
 737		ip_flush_pending_frames(sk);
 738	}
 739}
 740EXPORT_SYMBOL(udp_flush_pending_frames);
 741
 742/**
 743 * 	udp4_hwcsum  -  handle outgoing HW checksumming
 744 * 	@skb: 	sk_buff containing the filled-in UDP header
 745 * 	        (checksum field must be zeroed out)
 746 *	@src:	source IP address
 747 *	@dst:	destination IP address
 748 */
 749void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
 750{
 751	struct udphdr *uh = udp_hdr(skb);
 
 752	int offset = skb_transport_offset(skb);
 753	int len = skb->len - offset;
 754	int hlen = len;
 755	__wsum csum = 0;
 756
 757	if (!skb_has_frag_list(skb)) {
 758		/*
 759		 * Only one fragment on the socket.
 760		 */
 761		skb->csum_start = skb_transport_header(skb) - skb->head;
 762		skb->csum_offset = offsetof(struct udphdr, check);
 763		uh->check = ~csum_tcpudp_magic(src, dst, len,
 764					       IPPROTO_UDP, 0);
 765	} else {
 766		struct sk_buff *frags;
 767
 768		/*
 769		 * HW-checksum won't work as there are two or more
 770		 * fragments on the socket so that all csums of sk_buffs
 771		 * should be together
 772		 */
 773		skb_walk_frags(skb, frags) {
 774			csum = csum_add(csum, frags->csum);
 775			hlen -= frags->len;
 776		}
 777
 778		csum = skb_checksum(skb, offset, hlen, csum);
 779		skb->ip_summed = CHECKSUM_NONE;
 780
 781		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
 782		if (uh->check == 0)
 783			uh->check = CSUM_MANGLED_0;
 784	}
 785}
 786EXPORT_SYMBOL_GPL(udp4_hwcsum);
 787
 788/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
 789 * for the simple case like when setting the checksum for a UDP tunnel.
 790 */
 791void udp_set_csum(bool nocheck, struct sk_buff *skb,
 792		  __be32 saddr, __be32 daddr, int len)
 793{
 794	struct udphdr *uh = udp_hdr(skb);
 795
 796	if (nocheck) {
 797		uh->check = 0;
 798	} else if (skb_is_gso(skb)) {
 799		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 800	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 801		uh->check = 0;
 802		uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
 803		if (uh->check == 0)
 804			uh->check = CSUM_MANGLED_0;
 805	} else {
 806		skb->ip_summed = CHECKSUM_PARTIAL;
 807		skb->csum_start = skb_transport_header(skb) - skb->head;
 808		skb->csum_offset = offsetof(struct udphdr, check);
 809		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 810	}
 811}
 812EXPORT_SYMBOL(udp_set_csum);
 813
 814static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
 815			struct inet_cork *cork)
 816{
 817	struct sock *sk = skb->sk;
 818	struct inet_sock *inet = inet_sk(sk);
 819	struct udphdr *uh;
 820	int err = 0;
 821	int is_udplite = IS_UDPLITE(sk);
 822	int offset = skb_transport_offset(skb);
 823	int len = skb->len - offset;
 824	int datalen = len - sizeof(*uh);
 825	__wsum csum = 0;
 826
 827	/*
 828	 * Create a UDP header
 829	 */
 830	uh = udp_hdr(skb);
 831	uh->source = inet->inet_sport;
 832	uh->dest = fl4->fl4_dport;
 833	uh->len = htons(len);
 834	uh->check = 0;
 835
 836	if (cork->gso_size) {
 837		const int hlen = skb_network_header_len(skb) +
 838				 sizeof(struct udphdr);
 839
 840		if (hlen + cork->gso_size > cork->fragsize) {
 841			kfree_skb(skb);
 842			return -EINVAL;
 843		}
 844		if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
 845			kfree_skb(skb);
 846			return -EINVAL;
 847		}
 848		if (sk->sk_no_check_tx) {
 849			kfree_skb(skb);
 850			return -EINVAL;
 851		}
 852		if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
 853		    dst_xfrm(skb_dst(skb))) {
 854			kfree_skb(skb);
 855			return -EIO;
 856		}
 857
 858		if (datalen > cork->gso_size) {
 859			skb_shinfo(skb)->gso_size = cork->gso_size;
 860			skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
 861			skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
 862								 cork->gso_size);
 863		}
 864		goto csum_partial;
 865	}
 866
 867	if (is_udplite)  				 /*     UDP-Lite      */
 868		csum = udplite_csum(skb);
 869
 870	else if (sk->sk_no_check_tx) {			 /* UDP csum off */
 871
 872		skb->ip_summed = CHECKSUM_NONE;
 873		goto send;
 874
 875	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
 876csum_partial:
 877
 878		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
 879		goto send;
 880
 881	} else
 882		csum = udp_csum(skb);
 883
 884	/* add protocol-dependent pseudo-header */
 885	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
 886				      sk->sk_protocol, csum);
 887	if (uh->check == 0)
 888		uh->check = CSUM_MANGLED_0;
 889
 890send:
 891	err = ip_send_skb(sock_net(sk), skb);
 892	if (err) {
 893		if (err == -ENOBUFS && !inet->recverr) {
 894			UDP_INC_STATS(sock_net(sk),
 895				      UDP_MIB_SNDBUFERRORS, is_udplite);
 896			err = 0;
 897		}
 898	} else
 899		UDP_INC_STATS(sock_net(sk),
 900			      UDP_MIB_OUTDATAGRAMS, is_udplite);
 901	return err;
 902}
 903
 904/*
 905 * Push out all pending data as one UDP datagram. Socket is locked.
 906 */
 907int udp_push_pending_frames(struct sock *sk)
 908{
 909	struct udp_sock  *up = udp_sk(sk);
 910	struct inet_sock *inet = inet_sk(sk);
 911	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
 912	struct sk_buff *skb;
 913	int err = 0;
 914
 915	skb = ip_finish_skb(sk, fl4);
 916	if (!skb)
 917		goto out;
 918
 919	err = udp_send_skb(skb, fl4, &inet->cork.base);
 920
 921out:
 922	up->len = 0;
 923	up->pending = 0;
 924	return err;
 925}
 926EXPORT_SYMBOL(udp_push_pending_frames);
 927
 928static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
 929{
 930	switch (cmsg->cmsg_type) {
 931	case UDP_SEGMENT:
 932		if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
 933			return -EINVAL;
 934		*gso_size = *(__u16 *)CMSG_DATA(cmsg);
 935		return 0;
 936	default:
 937		return -EINVAL;
 938	}
 939}
 940
 941int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
 942{
 943	struct cmsghdr *cmsg;
 944	bool need_ip = false;
 945	int err;
 946
 947	for_each_cmsghdr(cmsg, msg) {
 948		if (!CMSG_OK(msg, cmsg))
 949			return -EINVAL;
 950
 951		if (cmsg->cmsg_level != SOL_UDP) {
 952			need_ip = true;
 953			continue;
 954		}
 955
 956		err = __udp_cmsg_send(cmsg, gso_size);
 957		if (err)
 958			return err;
 959	}
 960
 961	return need_ip;
 962}
 963EXPORT_SYMBOL_GPL(udp_cmsg_send);
 964
 965int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 
 966{
 967	struct inet_sock *inet = inet_sk(sk);
 968	struct udp_sock *up = udp_sk(sk);
 969	DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
 970	struct flowi4 fl4_stack;
 971	struct flowi4 *fl4;
 972	int ulen = len;
 973	struct ipcm_cookie ipc;
 974	struct rtable *rt = NULL;
 975	int free = 0;
 976	int connected = 0;
 977	__be32 daddr, faddr, saddr;
 978	__be16 dport;
 979	u8  tos;
 980	int err, is_udplite = IS_UDPLITE(sk);
 981	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
 982	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
 983	struct sk_buff *skb;
 984	struct ip_options_data opt_copy;
 985
 986	if (len > 0xFFFF)
 987		return -EMSGSIZE;
 988
 989	/*
 990	 *	Check the flags.
 991	 */
 992
 993	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
 994		return -EOPNOTSUPP;
 995
 
 
 
 996	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
 997
 998	fl4 = &inet->cork.fl.u.ip4;
 999	if (up->pending) {
1000		/*
1001		 * There are pending frames.
1002		 * The socket lock must be held while it's corked.
1003		 */
1004		lock_sock(sk);
1005		if (likely(up->pending)) {
1006			if (unlikely(up->pending != AF_INET)) {
1007				release_sock(sk);
1008				return -EINVAL;
1009			}
1010			goto do_append_data;
1011		}
1012		release_sock(sk);
1013	}
1014	ulen += sizeof(struct udphdr);
1015
1016	/*
1017	 *	Get and verify the address.
1018	 */
1019	if (usin) {
 
1020		if (msg->msg_namelen < sizeof(*usin))
1021			return -EINVAL;
1022		if (usin->sin_family != AF_INET) {
1023			if (usin->sin_family != AF_UNSPEC)
1024				return -EAFNOSUPPORT;
1025		}
1026
1027		daddr = usin->sin_addr.s_addr;
1028		dport = usin->sin_port;
1029		if (dport == 0)
1030			return -EINVAL;
1031	} else {
1032		if (sk->sk_state != TCP_ESTABLISHED)
1033			return -EDESTADDRREQ;
1034		daddr = inet->inet_daddr;
1035		dport = inet->inet_dport;
1036		/* Open fast path for connected socket.
1037		   Route will not be used, if at least one option is set.
1038		 */
1039		connected = 1;
1040	}
 
1041
1042	ipcm_init_sk(&ipc, inet);
1043	ipc.gso_size = up->gso_size;
1044
 
1045	if (msg->msg_controllen) {
1046		err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1047		if (err > 0)
1048			err = ip_cmsg_send(sk, msg, &ipc,
1049					   sk->sk_family == AF_INET6);
1050		if (unlikely(err < 0)) {
1051			kfree(ipc.opt);
1052			return err;
1053		}
1054		if (ipc.opt)
1055			free = 1;
1056		connected = 0;
1057	}
1058	if (!ipc.opt) {
1059		struct ip_options_rcu *inet_opt;
1060
1061		rcu_read_lock();
1062		inet_opt = rcu_dereference(inet->inet_opt);
1063		if (inet_opt) {
1064			memcpy(&opt_copy, inet_opt,
1065			       sizeof(*inet_opt) + inet_opt->opt.optlen);
1066			ipc.opt = &opt_copy.opt;
1067		}
1068		rcu_read_unlock();
1069	}
1070
1071	if (cgroup_bpf_enabled && !connected) {
1072		err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1073					    (struct sockaddr *)usin, &ipc.addr);
1074		if (err)
1075			goto out_free;
1076		if (usin) {
1077			if (usin->sin_port == 0) {
1078				/* BPF program set invalid port. Reject it. */
1079				err = -EINVAL;
1080				goto out_free;
1081			}
1082			daddr = usin->sin_addr.s_addr;
1083			dport = usin->sin_port;
1084		}
1085	}
1086
1087	saddr = ipc.addr;
1088	ipc.addr = faddr = daddr;
1089
1090	if (ipc.opt && ipc.opt->opt.srr) {
1091		if (!daddr) {
1092			err = -EINVAL;
1093			goto out_free;
1094		}
1095		faddr = ipc.opt->opt.faddr;
1096		connected = 0;
1097	}
1098	tos = get_rttos(&ipc, inet);
1099	if (sock_flag(sk, SOCK_LOCALROUTE) ||
1100	    (msg->msg_flags & MSG_DONTROUTE) ||
1101	    (ipc.opt && ipc.opt->opt.is_strictroute)) {
1102		tos |= RTO_ONLINK;
1103		connected = 0;
1104	}
1105
1106	if (ipv4_is_multicast(daddr)) {
1107		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1108			ipc.oif = inet->mc_index;
1109		if (!saddr)
1110			saddr = inet->mc_addr;
1111		connected = 0;
1112	} else if (!ipc.oif) {
1113		ipc.oif = inet->uc_index;
1114	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1115		/* oif is set, packet is to local broadcast and
1116		 * and uc_index is set. oif is most likely set
1117		 * by sk_bound_dev_if. If uc_index != oif check if the
1118		 * oif is an L3 master and uc_index is an L3 slave.
1119		 * If so, we want to allow the send using the uc_index.
1120		 */
1121		if (ipc.oif != inet->uc_index &&
1122		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1123							      inet->uc_index)) {
1124			ipc.oif = inet->uc_index;
1125		}
1126	}
1127
1128	if (connected)
1129		rt = (struct rtable *)sk_dst_check(sk, 0);
1130
1131	if (!rt) {
1132		struct net *net = sock_net(sk);
1133		__u8 flow_flags = inet_sk_flowi_flags(sk);
1134
1135		fl4 = &fl4_stack;
1136
1137		flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1138				   RT_SCOPE_UNIVERSE, sk->sk_protocol,
1139				   flow_flags,
1140				   faddr, saddr, dport, inet->inet_sport,
1141				   sk->sk_uid);
1142
1143		security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1144		rt = ip_route_output_flow(net, fl4, sk);
1145		if (IS_ERR(rt)) {
1146			err = PTR_ERR(rt);
1147			rt = NULL;
1148			if (err == -ENETUNREACH)
1149				IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1150			goto out;
1151		}
1152
1153		err = -EACCES;
1154		if ((rt->rt_flags & RTCF_BROADCAST) &&
1155		    !sock_flag(sk, SOCK_BROADCAST))
1156			goto out;
1157		if (connected)
1158			sk_dst_set(sk, dst_clone(&rt->dst));
1159	}
1160
1161	if (msg->msg_flags&MSG_CONFIRM)
1162		goto do_confirm;
1163back_from_confirm:
1164
1165	saddr = fl4->saddr;
1166	if (!ipc.addr)
1167		daddr = ipc.addr = fl4->daddr;
1168
1169	/* Lockless fast path for the non-corking case. */
1170	if (!corkreq) {
1171		struct inet_cork cork;
1172
1173		skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1174				  sizeof(struct udphdr), &ipc, &rt,
1175				  &cork, msg->msg_flags);
1176		err = PTR_ERR(skb);
1177		if (!IS_ERR_OR_NULL(skb))
1178			err = udp_send_skb(skb, fl4, &cork);
1179		goto out;
1180	}
1181
1182	lock_sock(sk);
1183	if (unlikely(up->pending)) {
1184		/* The socket is already corked while preparing it. */
1185		/* ... which is an evident application bug. --ANK */
1186		release_sock(sk);
1187
1188		net_dbg_ratelimited("socket already corked\n");
1189		err = -EINVAL;
1190		goto out;
1191	}
1192	/*
1193	 *	Now cork the socket to pend data.
1194	 */
1195	fl4 = &inet->cork.fl.u.ip4;
1196	fl4->daddr = daddr;
1197	fl4->saddr = saddr;
1198	fl4->fl4_dport = dport;
1199	fl4->fl4_sport = inet->inet_sport;
1200	up->pending = AF_INET;
1201
1202do_append_data:
1203	up->len += ulen;
1204	err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1205			     sizeof(struct udphdr), &ipc, &rt,
1206			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1207	if (err)
1208		udp_flush_pending_frames(sk);
1209	else if (!corkreq)
1210		err = udp_push_pending_frames(sk);
1211	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1212		up->pending = 0;
1213	release_sock(sk);
1214
1215out:
1216	ip_rt_put(rt);
1217out_free:
1218	if (free)
1219		kfree(ipc.opt);
1220	if (!err)
1221		return len;
1222	/*
1223	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1224	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1225	 * we don't have a good statistic (IpOutDiscards but it can be too many
1226	 * things).  We could add another new stat but at least for now that
1227	 * seems like overkill.
1228	 */
1229	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1230		UDP_INC_STATS(sock_net(sk),
1231			      UDP_MIB_SNDBUFERRORS, is_udplite);
1232	}
1233	return err;
1234
1235do_confirm:
1236	if (msg->msg_flags & MSG_PROBE)
1237		dst_confirm_neigh(&rt->dst, &fl4->daddr);
1238	if (!(msg->msg_flags&MSG_PROBE) || len)
1239		goto back_from_confirm;
1240	err = 0;
1241	goto out;
1242}
1243EXPORT_SYMBOL(udp_sendmsg);
1244
1245int udp_sendpage(struct sock *sk, struct page *page, int offset,
1246		 size_t size, int flags)
1247{
1248	struct inet_sock *inet = inet_sk(sk);
1249	struct udp_sock *up = udp_sk(sk);
1250	int ret;
1251
1252	if (flags & MSG_SENDPAGE_NOTLAST)
1253		flags |= MSG_MORE;
1254
1255	if (!up->pending) {
1256		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
1257
1258		/* Call udp_sendmsg to specify destination address which
1259		 * sendpage interface can't pass.
1260		 * This will succeed only when the socket is connected.
1261		 */
1262		ret = udp_sendmsg(sk, &msg, 0);
1263		if (ret < 0)
1264			return ret;
1265	}
1266
1267	lock_sock(sk);
1268
1269	if (unlikely(!up->pending)) {
1270		release_sock(sk);
1271
1272		net_dbg_ratelimited("cork failed\n");
1273		return -EINVAL;
1274	}
1275
1276	ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1277			     page, offset, size, flags);
1278	if (ret == -EOPNOTSUPP) {
1279		release_sock(sk);
1280		return sock_no_sendpage(sk->sk_socket, page, offset,
1281					size, flags);
1282	}
1283	if (ret < 0) {
1284		udp_flush_pending_frames(sk);
1285		goto out;
1286	}
1287
1288	up->len += size;
1289	if (!(up->corkflag || (flags&MSG_MORE)))
1290		ret = udp_push_pending_frames(sk);
1291	if (!ret)
1292		ret = size;
1293out:
1294	release_sock(sk);
1295	return ret;
1296}
1297
1298#define UDP_SKB_IS_STATELESS 0x80000000
1299
1300/* all head states (dst, sk, nf conntrack) except skb extensions are
1301 * cleared by udp_rcv().
1302 *
1303 * We need to preserve secpath, if present, to eventually process
1304 * IP_CMSG_PASSSEC at recvmsg() time.
1305 *
1306 * Other extensions can be cleared.
1307 */
1308static bool udp_try_make_stateless(struct sk_buff *skb)
1309{
1310	if (!skb_has_extensions(skb))
1311		return true;
1312
1313	if (!secpath_exists(skb)) {
1314		skb_ext_reset(skb);
1315		return true;
1316	}
1317
1318	return false;
1319}
1320
1321static void udp_set_dev_scratch(struct sk_buff *skb)
1322{
1323	struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1324
1325	BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1326	scratch->_tsize_state = skb->truesize;
1327#if BITS_PER_LONG == 64
1328	scratch->len = skb->len;
1329	scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1330	scratch->is_linear = !skb_is_nonlinear(skb);
1331#endif
1332	if (udp_try_make_stateless(skb))
1333		scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1334}
1335
1336static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1337{
1338	/* We come here after udp_lib_checksum_complete() returned 0.
1339	 * This means that __skb_checksum_complete() might have
1340	 * set skb->csum_valid to 1.
1341	 * On 64bit platforms, we can set csum_unnecessary
1342	 * to true, but only if the skb is not shared.
1343	 */
1344#if BITS_PER_LONG == 64
1345	if (!skb_shared(skb))
1346		udp_skb_scratch(skb)->csum_unnecessary = true;
1347#endif
1348}
1349
1350static int udp_skb_truesize(struct sk_buff *skb)
1351{
1352	return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1353}
1354
1355static bool udp_skb_has_head_state(struct sk_buff *skb)
1356{
1357	return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1358}
1359
1360/* fully reclaim rmem/fwd memory allocated for skb */
1361static void udp_rmem_release(struct sock *sk, int size, int partial,
1362			     bool rx_queue_lock_held)
1363{
1364	struct udp_sock *up = udp_sk(sk);
1365	struct sk_buff_head *sk_queue;
1366	int amt;
1367
1368	if (likely(partial)) {
1369		up->forward_deficit += size;
1370		size = up->forward_deficit;
1371		if (size < (sk->sk_rcvbuf >> 2))
1372			return;
1373	} else {
1374		size += up->forward_deficit;
1375	}
1376	up->forward_deficit = 0;
1377
1378	/* acquire the sk_receive_queue for fwd allocated memory scheduling,
1379	 * if the called don't held it already
1380	 */
1381	sk_queue = &sk->sk_receive_queue;
1382	if (!rx_queue_lock_held)
1383		spin_lock(&sk_queue->lock);
1384
1385
1386	sk->sk_forward_alloc += size;
1387	amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1388	sk->sk_forward_alloc -= amt;
1389
1390	if (amt)
1391		__sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1392
1393	atomic_sub(size, &sk->sk_rmem_alloc);
1394
1395	/* this can save us from acquiring the rx queue lock on next receive */
1396	skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1397
1398	if (!rx_queue_lock_held)
1399		spin_unlock(&sk_queue->lock);
1400}
1401
1402/* Note: called with reader_queue.lock held.
1403 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1404 * This avoids a cache line miss while receive_queue lock is held.
1405 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1406 */
1407void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1408{
1409	prefetch(&skb->data);
1410	udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1411}
1412EXPORT_SYMBOL(udp_skb_destructor);
1413
1414/* as above, but the caller held the rx queue lock, too */
1415static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1416{
1417	prefetch(&skb->data);
1418	udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1419}
1420
1421/* Idea of busylocks is to let producers grab an extra spinlock
1422 * to relieve pressure on the receive_queue spinlock shared by consumer.
1423 * Under flood, this means that only one producer can be in line
1424 * trying to acquire the receive_queue spinlock.
1425 * These busylock can be allocated on a per cpu manner, instead of a
1426 * per socket one (that would consume a cache line per socket)
1427 */
1428static int udp_busylocks_log __read_mostly;
1429static spinlock_t *udp_busylocks __read_mostly;
1430
1431static spinlock_t *busylock_acquire(void *ptr)
1432{
1433	spinlock_t *busy;
1434
1435	busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1436	spin_lock(busy);
1437	return busy;
1438}
1439
1440static void busylock_release(spinlock_t *busy)
1441{
1442	if (busy)
1443		spin_unlock(busy);
1444}
1445
1446int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1447{
1448	struct sk_buff_head *list = &sk->sk_receive_queue;
1449	int rmem, delta, amt, err = -ENOMEM;
1450	spinlock_t *busy = NULL;
1451	int size;
1452
1453	/* try to avoid the costly atomic add/sub pair when the receive
1454	 * queue is full; always allow at least a packet
1455	 */
1456	rmem = atomic_read(&sk->sk_rmem_alloc);
1457	if (rmem > sk->sk_rcvbuf)
1458		goto drop;
1459
1460	/* Under mem pressure, it might be helpful to help udp_recvmsg()
1461	 * having linear skbs :
1462	 * - Reduce memory overhead and thus increase receive queue capacity
1463	 * - Less cache line misses at copyout() time
1464	 * - Less work at consume_skb() (less alien page frag freeing)
1465	 */
1466	if (rmem > (sk->sk_rcvbuf >> 1)) {
1467		skb_condense(skb);
1468
1469		busy = busylock_acquire(sk);
1470	}
1471	size = skb->truesize;
1472	udp_set_dev_scratch(skb);
1473
1474	/* we drop only if the receive buf is full and the receive
1475	 * queue contains some other skb
1476	 */
1477	rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1478	if (rmem > (size + sk->sk_rcvbuf))
1479		goto uncharge_drop;
1480
1481	spin_lock(&list->lock);
1482	if (size >= sk->sk_forward_alloc) {
1483		amt = sk_mem_pages(size);
1484		delta = amt << SK_MEM_QUANTUM_SHIFT;
1485		if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1486			err = -ENOBUFS;
1487			spin_unlock(&list->lock);
1488			goto uncharge_drop;
1489		}
1490
1491		sk->sk_forward_alloc += delta;
1492	}
1493
1494	sk->sk_forward_alloc -= size;
1495
1496	/* no need to setup a destructor, we will explicitly release the
1497	 * forward allocated memory on dequeue
1498	 */
1499	sock_skb_set_dropcount(sk, skb);
1500
1501	__skb_queue_tail(list, skb);
1502	spin_unlock(&list->lock);
1503
1504	if (!sock_flag(sk, SOCK_DEAD))
1505		sk->sk_data_ready(sk);
1506
1507	busylock_release(busy);
1508	return 0;
1509
1510uncharge_drop:
1511	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1512
1513drop:
1514	atomic_inc(&sk->sk_drops);
1515	busylock_release(busy);
1516	return err;
1517}
1518EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1519
1520void udp_destruct_sock(struct sock *sk)
1521{
1522	/* reclaim completely the forward allocated memory */
1523	struct udp_sock *up = udp_sk(sk);
1524	unsigned int total = 0;
1525	struct sk_buff *skb;
1526
1527	skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1528	while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1529		total += skb->truesize;
1530		kfree_skb(skb);
1531	}
1532	udp_rmem_release(sk, total, 0, true);
1533
1534	inet_sock_destruct(sk);
1535}
1536EXPORT_SYMBOL_GPL(udp_destruct_sock);
1537
1538int udp_init_sock(struct sock *sk)
1539{
1540	skb_queue_head_init(&udp_sk(sk)->reader_queue);
1541	sk->sk_destruct = udp_destruct_sock;
1542	return 0;
1543}
1544EXPORT_SYMBOL_GPL(udp_init_sock);
1545
1546void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1547{
1548	if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1549		bool slow = lock_sock_fast(sk);
1550
1551		sk_peek_offset_bwd(sk, len);
1552		unlock_sock_fast(sk, slow);
1553	}
1554
1555	if (!skb_unref(skb))
1556		return;
1557
1558	/* In the more common cases we cleared the head states previously,
1559	 * see __udp_queue_rcv_skb().
1560	 */
1561	if (unlikely(udp_skb_has_head_state(skb)))
1562		skb_release_head_state(skb);
1563	__consume_stateless_skb(skb);
1564}
1565EXPORT_SYMBOL_GPL(skb_consume_udp);
1566
1567static struct sk_buff *__first_packet_length(struct sock *sk,
1568					     struct sk_buff_head *rcvq,
1569					     int *total)
1570{
1571	struct sk_buff *skb;
1572
1573	while ((skb = skb_peek(rcvq)) != NULL) {
1574		if (udp_lib_checksum_complete(skb)) {
1575			__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1576					IS_UDPLITE(sk));
1577			__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1578					IS_UDPLITE(sk));
1579			atomic_inc(&sk->sk_drops);
1580			__skb_unlink(skb, rcvq);
1581			*total += skb->truesize;
1582			kfree_skb(skb);
1583		} else {
1584			udp_skb_csum_unnecessary_set(skb);
1585			break;
1586		}
1587	}
1588	return skb;
1589}
1590
1591/**
1592 *	first_packet_length	- return length of first packet in receive queue
1593 *	@sk: socket
1594 *
1595 *	Drops all bad checksum frames, until a valid one is found.
1596 *	Returns the length of found skb, or -1 if none is found.
1597 */
1598static int first_packet_length(struct sock *sk)
1599{
1600	struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1601	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1602	struct sk_buff *skb;
1603	int total = 0;
1604	int res;
 
1605
1606	spin_lock_bh(&rcvq->lock);
1607	skb = __first_packet_length(sk, rcvq, &total);
1608	if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1609		spin_lock(&sk_queue->lock);
1610		skb_queue_splice_tail_init(sk_queue, rcvq);
1611		spin_unlock(&sk_queue->lock);
1612
1613		skb = __first_packet_length(sk, rcvq, &total);
1614	}
1615	res = skb ? skb->len : -1;
1616	if (total)
1617		udp_rmem_release(sk, total, 1, false);
1618	spin_unlock_bh(&rcvq->lock);
 
 
 
 
 
 
 
 
1619	return res;
1620}
1621
1622/*
1623 *	IOCTL requests applicable to the UDP protocol
1624 */
1625
1626int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1627{
1628	switch (cmd) {
1629	case SIOCOUTQ:
1630	{
1631		int amount = sk_wmem_alloc_get(sk);
1632
1633		return put_user(amount, (int __user *)arg);
1634	}
1635
1636	case SIOCINQ:
1637	{
1638		int amount = max_t(int, 0, first_packet_length(sk));
 
 
 
 
 
 
 
 
1639
1640		return put_user(amount, (int __user *)arg);
1641	}
1642
1643	default:
1644		return -ENOIOCTLCMD;
1645	}
1646
1647	return 0;
1648}
1649EXPORT_SYMBOL(udp_ioctl);
1650
1651struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1652			       int noblock, int *off, int *err)
1653{
1654	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1655	struct sk_buff_head *queue;
1656	struct sk_buff *last;
1657	long timeo;
1658	int error;
1659
1660	queue = &udp_sk(sk)->reader_queue;
1661	flags |= noblock ? MSG_DONTWAIT : 0;
1662	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1663	do {
1664		struct sk_buff *skb;
1665
1666		error = sock_error(sk);
1667		if (error)
1668			break;
1669
1670		error = -EAGAIN;
1671		do {
1672			spin_lock_bh(&queue->lock);
1673			skb = __skb_try_recv_from_queue(sk, queue, flags,
1674							udp_skb_destructor,
1675							off, err, &last);
1676			if (skb) {
1677				spin_unlock_bh(&queue->lock);
1678				return skb;
1679			}
1680
1681			if (skb_queue_empty_lockless(sk_queue)) {
1682				spin_unlock_bh(&queue->lock);
1683				goto busy_check;
1684			}
1685
1686			/* refill the reader queue and walk it again
1687			 * keep both queues locked to avoid re-acquiring
1688			 * the sk_receive_queue lock if fwd memory scheduling
1689			 * is needed.
1690			 */
1691			spin_lock(&sk_queue->lock);
1692			skb_queue_splice_tail_init(sk_queue, queue);
1693
1694			skb = __skb_try_recv_from_queue(sk, queue, flags,
1695							udp_skb_dtor_locked,
1696							off, err, &last);
1697			spin_unlock(&sk_queue->lock);
1698			spin_unlock_bh(&queue->lock);
1699			if (skb)
1700				return skb;
1701
1702busy_check:
1703			if (!sk_can_busy_loop(sk))
1704				break;
1705
1706			sk_busy_loop(sk, flags & MSG_DONTWAIT);
1707		} while (!skb_queue_empty_lockless(sk_queue));
1708
1709		/* sk_queue is empty, reader_queue may contain peeked packets */
1710	} while (timeo &&
1711		 !__skb_wait_for_more_packets(sk, &error, &timeo,
1712					      (struct sk_buff *)sk_queue));
1713
1714	*err = error;
1715	return NULL;
1716}
1717EXPORT_SYMBOL(__skb_recv_udp);
1718
1719/*
1720 * 	This should be easy, if there is something there we
1721 * 	return it, otherwise we block.
1722 */
1723
1724int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1725		int flags, int *addr_len)
1726{
1727	struct inet_sock *inet = inet_sk(sk);
1728	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1729	struct sk_buff *skb;
1730	unsigned int ulen, copied;
1731	int off, err, peeking = flags & MSG_PEEK;
 
1732	int is_udplite = IS_UDPLITE(sk);
1733	bool checksum_valid = false;
 
 
 
 
 
 
1734
1735	if (flags & MSG_ERRQUEUE)
1736		return ip_recv_error(sk, msg, len, addr_len);
1737
1738try_again:
1739	off = sk_peek_offset(sk, flags);
1740	skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1741	if (!skb)
1742		return err;
1743
1744	ulen = udp_skb_len(skb);
1745	copied = len;
1746	if (copied > ulen - off)
1747		copied = ulen - off;
1748	else if (copied < ulen)
1749		msg->msg_flags |= MSG_TRUNC;
1750
1751	/*
1752	 * If checksum is needed at all, try to do it while copying the
1753	 * data.  If the data is truncated, or if we only want a partial
1754	 * coverage checksum (UDP-Lite), do it before the copy.
1755	 */
1756
1757	if (copied < ulen || peeking ||
1758	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1759		checksum_valid = udp_skb_csum_unnecessary(skb) ||
1760				!__udp_lib_checksum_complete(skb);
1761		if (!checksum_valid)
1762			goto csum_copy_err;
1763	}
1764
1765	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1766		if (udp_skb_is_linear(skb))
1767			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1768		else
1769			err = skb_copy_datagram_msg(skb, off, msg, copied);
1770	} else {
1771		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1772
1773		if (err == -EINVAL)
1774			goto csum_copy_err;
1775	}
1776
1777	if (unlikely(err)) {
1778		if (!peeking) {
1779			atomic_inc(&sk->sk_drops);
1780			UDP_INC_STATS(sock_net(sk),
1781				      UDP_MIB_INERRORS, is_udplite);
1782		}
1783		kfree_skb(skb);
1784		return err;
1785	}
1786
1787	if (!peeking)
1788		UDP_INC_STATS(sock_net(sk),
1789			      UDP_MIB_INDATAGRAMS, is_udplite);
1790
1791	sock_recv_ts_and_drops(msg, sk, skb);
1792
1793	/* Copy the address. */
1794	if (sin) {
1795		sin->sin_family = AF_INET;
1796		sin->sin_port = udp_hdr(skb)->source;
1797		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1798		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1799		*addr_len = sizeof(*sin);
1800
1801		if (cgroup_bpf_enabled)
1802			BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1803							(struct sockaddr *)sin);
1804	}
1805
1806	if (udp_sk(sk)->gro_enabled)
1807		udp_cmsg_recv(msg, sk, skb);
1808
1809	if (inet->cmsg_flags)
1810		ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1811
1812	err = copied;
1813	if (flags & MSG_TRUNC)
1814		err = ulen;
1815
1816	skb_consume_udp(sk, skb, peeking ? -err : err);
 
 
1817	return err;
1818
1819csum_copy_err:
1820	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1821				 udp_skb_destructor)) {
1822		UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1823		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1824	}
1825	kfree_skb(skb);
1826
1827	/* starting over for a new packet, but check if we need to yield */
1828	cond_resched();
 
 
1829	msg->msg_flags &= ~MSG_TRUNC;
1830	goto try_again;
1831}
1832
1833int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1834{
1835	/* This check is replicated from __ip4_datagram_connect() and
1836	 * intended to prevent BPF program called below from accessing bytes
1837	 * that are out of the bound specified by user in addr_len.
1838	 */
1839	if (addr_len < sizeof(struct sockaddr_in))
1840		return -EINVAL;
1841
1842	return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1843}
1844EXPORT_SYMBOL(udp_pre_connect);
1845
1846int __udp_disconnect(struct sock *sk, int flags)
1847{
1848	struct inet_sock *inet = inet_sk(sk);
1849	/*
1850	 *	1003.1g - break association.
1851	 */
1852
1853	sk->sk_state = TCP_CLOSE;
1854	inet->inet_daddr = 0;
1855	inet->inet_dport = 0;
1856	sock_rps_reset_rxhash(sk);
1857	sk->sk_bound_dev_if = 0;
1858	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1859		inet_reset_saddr(sk);
1860
1861	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1862		sk->sk_prot->unhash(sk);
1863		inet->inet_sport = 0;
1864	}
1865	sk_dst_reset(sk);
1866	return 0;
1867}
1868EXPORT_SYMBOL(__udp_disconnect);
1869
1870int udp_disconnect(struct sock *sk, int flags)
1871{
1872	lock_sock(sk);
1873	__udp_disconnect(sk, flags);
1874	release_sock(sk);
1875	return 0;
1876}
1877EXPORT_SYMBOL(udp_disconnect);
1878
1879void udp_lib_unhash(struct sock *sk)
1880{
1881	if (sk_hashed(sk)) {
1882		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1883		struct udp_hslot *hslot, *hslot2;
1884
1885		hslot  = udp_hashslot(udptable, sock_net(sk),
1886				      udp_sk(sk)->udp_port_hash);
1887		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1888
1889		spin_lock_bh(&hslot->lock);
1890		if (rcu_access_pointer(sk->sk_reuseport_cb))
1891			reuseport_detach_sock(sk);
1892		if (sk_del_node_init_rcu(sk)) {
1893			hslot->count--;
1894			inet_sk(sk)->inet_num = 0;
1895			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1896
1897			spin_lock(&hslot2->lock);
1898			hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1899			hslot2->count--;
1900			spin_unlock(&hslot2->lock);
1901		}
1902		spin_unlock_bh(&hslot->lock);
1903	}
1904}
1905EXPORT_SYMBOL(udp_lib_unhash);
1906
1907/*
1908 * inet_rcv_saddr was changed, we must rehash secondary hash
1909 */
1910void udp_lib_rehash(struct sock *sk, u16 newhash)
1911{
1912	if (sk_hashed(sk)) {
1913		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1914		struct udp_hslot *hslot, *hslot2, *nhslot2;
1915
1916		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1917		nhslot2 = udp_hashslot2(udptable, newhash);
1918		udp_sk(sk)->udp_portaddr_hash = newhash;
1919
1920		if (hslot2 != nhslot2 ||
1921		    rcu_access_pointer(sk->sk_reuseport_cb)) {
1922			hslot = udp_hashslot(udptable, sock_net(sk),
1923					     udp_sk(sk)->udp_port_hash);
1924			/* we must lock primary chain too */
1925			spin_lock_bh(&hslot->lock);
1926			if (rcu_access_pointer(sk->sk_reuseport_cb))
1927				reuseport_detach_sock(sk);
1928
1929			if (hslot2 != nhslot2) {
1930				spin_lock(&hslot2->lock);
1931				hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1932				hslot2->count--;
1933				spin_unlock(&hslot2->lock);
1934
1935				spin_lock(&nhslot2->lock);
1936				hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1937							 &nhslot2->head);
1938				nhslot2->count++;
1939				spin_unlock(&nhslot2->lock);
1940			}
1941
1942			spin_unlock_bh(&hslot->lock);
1943		}
1944	}
1945}
1946EXPORT_SYMBOL(udp_lib_rehash);
1947
1948void udp_v4_rehash(struct sock *sk)
1949{
1950	u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1951					  inet_sk(sk)->inet_rcv_saddr,
1952					  inet_sk(sk)->inet_num);
1953	udp_lib_rehash(sk, new_hash);
1954}
1955
1956static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1957{
1958	int rc;
1959
1960	if (inet_sk(sk)->inet_daddr) {
1961		sock_rps_save_rxhash(sk, skb);
1962		sk_mark_napi_id(sk, skb);
1963		sk_incoming_cpu_update(sk);
1964	} else {
1965		sk_mark_napi_id_once(sk, skb);
1966	}
1967
1968	rc = __udp_enqueue_schedule_skb(sk, skb);
1969	if (rc < 0) {
1970		int is_udplite = IS_UDPLITE(sk);
1971
1972		/* Note that an ENOMEM error is charged twice */
1973		if (rc == -ENOMEM)
1974			UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1975					is_udplite);
1976		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1977		kfree_skb(skb);
1978		trace_udp_fail_queue_rcv_skb(rc, sk);
1979		return -1;
1980	}
1981
1982	return 0;
 
1983}
1984
1985/* returns:
1986 *  -1: error
1987 *   0: success
1988 *  >0: "udp encap" protocol resubmission
1989 *
1990 * Note that in the success and error cases, the skb is assumed to
1991 * have either been requeued or freed.
1992 */
1993static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
1994{
1995	struct udp_sock *up = udp_sk(sk);
 
1996	int is_udplite = IS_UDPLITE(sk);
1997
1998	/*
1999	 *	Charge it to the socket, dropping if the queue is full.
2000	 */
2001	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2002		goto drop;
2003	nf_reset_ct(skb);
2004
2005	if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2006		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2007
 
2008		/*
2009		 * This is an encapsulation socket so pass the skb to
2010		 * the socket's udp_encap_rcv() hook. Otherwise, just
2011		 * fall through and pass this up the UDP socket.
2012		 * up->encap_rcv() returns the following value:
2013		 * =0 if skb was successfully passed to the encap
2014		 *    handler or was discarded by it.
2015		 * >0 if skb should be passed on to UDP.
2016		 * <0 if skb should be resubmitted as proto -N
2017		 */
2018
2019		/* if we're overly short, let UDP handle it */
2020		encap_rcv = READ_ONCE(up->encap_rcv);
2021		if (encap_rcv) {
2022			int ret;
2023
2024			/* Verify checksum before giving to encap */
2025			if (udp_lib_checksum_complete(skb))
2026				goto csum_error;
2027
2028			ret = encap_rcv(sk, skb);
2029			if (ret <= 0) {
2030				__UDP_INC_STATS(sock_net(sk),
2031						UDP_MIB_INDATAGRAMS,
2032						is_udplite);
2033				return -ret;
2034			}
2035		}
2036
2037		/* FALLTHROUGH -- it's a UDP Packet */
2038	}
2039
2040	/*
2041	 * 	UDP-Lite specific tests, ignored on UDP sockets
2042	 */
2043	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
2044
2045		/*
2046		 * MIB statistics other than incrementing the error count are
2047		 * disabled for the following two types of errors: these depend
2048		 * on the application settings, not on the functioning of the
2049		 * protocol stack as such.
2050		 *
2051		 * RFC 3828 here recommends (sec 3.3): "There should also be a
2052		 * way ... to ... at least let the receiving application block
2053		 * delivery of packets with coverage values less than a value
2054		 * provided by the application."
2055		 */
2056		if (up->pcrlen == 0) {          /* full coverage was set  */
2057			net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2058					    UDP_SKB_CB(skb)->cscov, skb->len);
 
2059			goto drop;
2060		}
2061		/* The next case involves violating the min. coverage requested
2062		 * by the receiver. This is subtle: if receiver wants x and x is
2063		 * greater than the buffersize/MTU then receiver will complain
2064		 * that it wants x while sender emits packets of smaller size y.
2065		 * Therefore the above ...()->partial_cov statement is essential.
2066		 */
2067		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
2068			net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2069					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
 
2070			goto drop;
2071		}
2072	}
2073
2074	prefetch(&sk->sk_rmem_alloc);
2075	if (rcu_access_pointer(sk->sk_filter) &&
2076	    udp_lib_checksum_complete(skb))
2077			goto csum_error;
 
2078
2079	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2080		goto drop;
2081
2082	udp_csum_pull_header(skb);
2083
2084	ipv4_pktinfo_prepare(sk, skb);
2085	return __udp_queue_rcv_skb(sk, skb);
 
 
 
 
 
 
 
 
2086
2087csum_error:
2088	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2089drop:
2090	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2091	atomic_inc(&sk->sk_drops);
2092	kfree_skb(skb);
2093	return -1;
2094}
2095
2096static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2097{
2098	struct sk_buff *next, *segs;
2099	int ret;
2100
2101	if (likely(!udp_unexpected_gso(sk, skb)))
2102		return udp_queue_rcv_one_skb(sk, skb);
 
 
 
 
2103
2104	BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET);
2105	__skb_push(skb, -skb_mac_offset(skb));
2106	segs = udp_rcv_segment(sk, skb, true);
2107	for (skb = segs; skb; skb = next) {
2108		next = skb->next;
2109		__skb_pull(skb, skb_transport_offset(skb));
2110		ret = udp_queue_rcv_one_skb(sk, skb);
2111		if (ret > 0)
2112			ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret);
2113	}
2114	return 0;
2115}
2116
2117/* For TCP sockets, sk_rx_dst is protected by socket lock
2118 * For UDP, we use xchg() to guard against concurrent changes.
2119 */
2120bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2121{
2122	struct dst_entry *old;
 
2123
2124	if (dst_hold_safe(dst)) {
2125		old = xchg(&sk->sk_rx_dst, dst);
2126		dst_release(old);
2127		return old != dst;
2128	}
2129	return false;
 
2130}
2131EXPORT_SYMBOL(udp_sk_rx_dst_set);
2132
2133/*
2134 *	Multicasts and broadcasts go to each listener.
2135 *
2136 *	Note: called only from the BH handler context.
2137 */
2138static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2139				    struct udphdr  *uh,
2140				    __be32 saddr, __be32 daddr,
2141				    struct udp_table *udptable,
2142				    int proto)
2143{
2144	struct sock *sk, *first = NULL;
2145	unsigned short hnum = ntohs(uh->dest);
2146	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2147	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2148	unsigned int offset = offsetof(typeof(*sk), sk_node);
2149	int dif = skb->dev->ifindex;
2150	int sdif = inet_sdif(skb);
2151	struct hlist_node *node;
2152	struct sk_buff *nskb;
2153
2154	if (use_hash2) {
2155		hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2156			    udptable->mask;
2157		hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2158start_lookup:
2159		hslot = &udptable->hash2[hash2];
2160		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2161	}
2162
2163	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2164		if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2165					 uh->source, saddr, dif, sdif, hnum))
2166			continue;
2167
2168		if (!first) {
2169			first = sk;
2170			continue;
2171		}
2172		nskb = skb_clone(skb, GFP_ATOMIC);
2173
2174		if (unlikely(!nskb)) {
2175			atomic_inc(&sk->sk_drops);
2176			__UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2177					IS_UDPLITE(sk));
2178			__UDP_INC_STATS(net, UDP_MIB_INERRORS,
2179					IS_UDPLITE(sk));
2180			continue;
2181		}
2182		if (udp_queue_rcv_skb(sk, nskb) > 0)
2183			consume_skb(nskb);
2184	}
 
 
 
 
 
2185
2186	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
2187	if (use_hash2 && hash2 != hash2_any) {
2188		hash2 = hash2_any;
2189		goto start_lookup;
2190	}
2191
2192	if (first) {
2193		if (udp_queue_rcv_skb(first, skb) > 0)
2194			consume_skb(skb);
 
 
 
 
 
2195	} else {
2196		kfree_skb(skb);
2197		__UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2198				proto == IPPROTO_UDPLITE);
2199	}
2200	return 0;
2201}
2202
2203/* Initialize UDP checksum. If exited with zero value (success),
2204 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2205 * Otherwise, csum completion requires checksumming packet body,
2206 * including udp header and folding it to skb->csum.
2207 */
2208static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2209				 int proto)
2210{
 
2211	int err;
2212
2213	UDP_SKB_CB(skb)->partial_cov = 0;
2214	UDP_SKB_CB(skb)->cscov = skb->len;
2215
2216	if (proto == IPPROTO_UDPLITE) {
2217		err = udplite_checksum_init(skb, uh);
2218		if (err)
2219			return err;
2220
2221		if (UDP_SKB_CB(skb)->partial_cov) {
2222			skb->csum = inet_compute_pseudo(skb, proto);
2223			return 0;
2224		}
2225	}
2226
2227	/* Note, we are only interested in != 0 or == 0, thus the
2228	 * force to int.
 
 
 
 
 
 
 
 
 
 
 
2229	 */
2230	err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2231							inet_compute_pseudo);
2232	if (err)
2233		return err;
2234
2235	if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2236		/* If SW calculated the value, we know it's bad */
2237		if (skb->csum_complete_sw)
2238			return 1;
2239
2240		/* HW says the value is bad. Let's validate that.
2241		 * skb->csum is no longer the full packet checksum,
2242		 * so don't treat it as such.
2243		 */
2244		skb_checksum_complete_unset(skb);
2245	}
2246
2247	return 0;
2248}
2249
2250/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2251 * return code conversion for ip layer consumption
2252 */
2253static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2254			       struct udphdr *uh)
2255{
2256	int ret;
2257
2258	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2259		skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2260
2261	ret = udp_queue_rcv_skb(sk, skb);
2262
2263	/* a return value > 0 means to resubmit the input, but
2264	 * it wants the return to be -protocol, or 0
2265	 */
2266	if (ret > 0)
2267		return -ret;
2268	return 0;
2269}
2270
2271/*
2272 *	All we need to do is get the socket, and then do a checksum.
2273 */
2274
2275int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2276		   int proto)
2277{
2278	struct sock *sk;
2279	struct udphdr *uh;
2280	unsigned short ulen;
2281	struct rtable *rt = skb_rtable(skb);
2282	__be32 saddr, daddr;
2283	struct net *net = dev_net(skb->dev);
2284
2285	/*
2286	 *  Validate the packet.
2287	 */
2288	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2289		goto drop;		/* No space for header. */
2290
2291	uh   = udp_hdr(skb);
2292	ulen = ntohs(uh->len);
2293	saddr = ip_hdr(skb)->saddr;
2294	daddr = ip_hdr(skb)->daddr;
2295
2296	if (ulen > skb->len)
2297		goto short_packet;
2298
2299	if (proto == IPPROTO_UDP) {
2300		/* UDP validates ulen. */
2301		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2302			goto short_packet;
2303		uh = udp_hdr(skb);
2304	}
2305
2306	if (udp4_csum_init(skb, uh, proto))
2307		goto csum_error;
2308
2309	sk = skb_steal_sock(skb);
2310	if (sk) {
2311		struct dst_entry *dst = skb_dst(skb);
2312		int ret;
2313
2314		if (unlikely(sk->sk_rx_dst != dst))
2315			udp_sk_rx_dst_set(sk, dst);
2316
2317		ret = udp_unicast_rcv_skb(sk, skb, uh);
2318		sock_put(sk);
2319		return ret;
2320	}
2321
2322	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2323		return __udp4_lib_mcast_deliver(net, skb, uh,
2324						saddr, daddr, udptable, proto);
2325
2326	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2327	if (sk)
2328		return udp_unicast_rcv_skb(sk, skb, uh);
 
 
 
 
 
 
 
 
 
 
2329
2330	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2331		goto drop;
2332	nf_reset_ct(skb);
2333
2334	/* No socket. Drop packet silently, if checksum is wrong */
2335	if (udp_lib_checksum_complete(skb))
2336		goto csum_error;
2337
2338	__UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2339	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2340
2341	/*
2342	 * Hmm.  We got an UDP packet to a port to which we
2343	 * don't wanna listen.  Ignore it.
2344	 */
2345	kfree_skb(skb);
2346	return 0;
2347
2348short_packet:
2349	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2350			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2351			    &saddr, ntohs(uh->source),
2352			    ulen, skb->len,
2353			    &daddr, ntohs(uh->dest));
 
 
 
2354	goto drop;
2355
2356csum_error:
2357	/*
2358	 * RFC1122: OK.  Discards the bad packet silently (as far as
2359	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2360	 */
2361	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2362			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2363			    &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2364			    ulen);
2365	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
 
 
2366drop:
2367	__UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2368	kfree_skb(skb);
2369	return 0;
2370}
2371
2372/* We can only early demux multicast if there is a single matching socket.
2373 * If more than one socket found returns NULL
2374 */
2375static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2376						  __be16 loc_port, __be32 loc_addr,
2377						  __be16 rmt_port, __be32 rmt_addr,
2378						  int dif, int sdif)
2379{
2380	struct sock *sk, *result;
2381	unsigned short hnum = ntohs(loc_port);
2382	unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2383	struct udp_hslot *hslot = &udp_table.hash[slot];
2384
2385	/* Do not bother scanning a too big list */
2386	if (hslot->count > 10)
2387		return NULL;
2388
2389	result = NULL;
2390	sk_for_each_rcu(sk, &hslot->head) {
2391		if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2392					rmt_port, rmt_addr, dif, sdif, hnum)) {
2393			if (result)
2394				return NULL;
2395			result = sk;
2396		}
2397	}
2398
2399	return result;
2400}
2401
2402/* For unicast we should only early demux connected sockets or we can
2403 * break forwarding setups.  The chains here can be long so only check
2404 * if the first socket is an exact match and if not move on.
2405 */
2406static struct sock *__udp4_lib_demux_lookup(struct net *net,
2407					    __be16 loc_port, __be32 loc_addr,
2408					    __be16 rmt_port, __be32 rmt_addr,
2409					    int dif, int sdif)
2410{
2411	unsigned short hnum = ntohs(loc_port);
2412	unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2413	unsigned int slot2 = hash2 & udp_table.mask;
2414	struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2415	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2416	const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2417	struct sock *sk;
2418
2419	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2420		if (INET_MATCH(sk, net, acookie, rmt_addr,
2421			       loc_addr, ports, dif, sdif))
2422			return sk;
2423		/* Only check first socket in chain */
2424		break;
2425	}
2426	return NULL;
2427}
2428
2429int udp_v4_early_demux(struct sk_buff *skb)
2430{
2431	struct net *net = dev_net(skb->dev);
2432	struct in_device *in_dev = NULL;
2433	const struct iphdr *iph;
2434	const struct udphdr *uh;
2435	struct sock *sk = NULL;
2436	struct dst_entry *dst;
2437	int dif = skb->dev->ifindex;
2438	int sdif = inet_sdif(skb);
2439	int ours;
2440
2441	/* validate the packet */
2442	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2443		return 0;
2444
2445	iph = ip_hdr(skb);
2446	uh = udp_hdr(skb);
2447
2448	if (skb->pkt_type == PACKET_MULTICAST) {
2449		in_dev = __in_dev_get_rcu(skb->dev);
2450
2451		if (!in_dev)
2452			return 0;
2453
2454		ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2455				       iph->protocol);
2456		if (!ours)
2457			return 0;
2458
2459		sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2460						   uh->source, iph->saddr,
2461						   dif, sdif);
2462	} else if (skb->pkt_type == PACKET_HOST) {
2463		sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2464					     uh->source, iph->saddr, dif, sdif);
2465	}
2466
2467	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2468		return 0;
2469
2470	skb->sk = sk;
2471	skb->destructor = sock_efree;
2472	dst = READ_ONCE(sk->sk_rx_dst);
2473
2474	if (dst)
2475		dst = dst_check(dst, 0);
2476	if (dst) {
2477		u32 itag = 0;
2478
2479		/* set noref for now.
2480		 * any place which wants to hold dst has to call
2481		 * dst_hold_safe()
2482		 */
2483		skb_dst_set_noref(skb, dst);
2484
2485		/* for unconnected multicast sockets we need to validate
2486		 * the source on each packet
2487		 */
2488		if (!inet_sk(sk)->inet_daddr && in_dev)
2489			return ip_mc_validate_source(skb, iph->daddr,
2490						     iph->saddr, iph->tos,
2491						     skb->dev, in_dev, &itag);
2492	}
2493	return 0;
2494}
2495
2496int udp_rcv(struct sk_buff *skb)
2497{
2498	return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2499}
2500
2501void udp_destroy_sock(struct sock *sk)
2502{
2503	struct udp_sock *up = udp_sk(sk);
2504	bool slow = lock_sock_fast(sk);
2505	udp_flush_pending_frames(sk);
2506	unlock_sock_fast(sk, slow);
2507	if (static_branch_unlikely(&udp_encap_needed_key)) {
2508		if (up->encap_type) {
2509			void (*encap_destroy)(struct sock *sk);
2510			encap_destroy = READ_ONCE(up->encap_destroy);
2511			if (encap_destroy)
2512				encap_destroy(sk);
2513		}
2514		if (up->encap_enabled)
2515			static_branch_dec(&udp_encap_needed_key);
2516	}
2517}
2518
2519/*
2520 *	Socket option code for UDP
2521 */
2522int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2523		       char __user *optval, unsigned int optlen,
2524		       int (*push_pending_frames)(struct sock *))
2525{
2526	struct udp_sock *up = udp_sk(sk);
2527	int val, valbool;
2528	int err = 0;
2529	int is_udplite = IS_UDPLITE(sk);
2530
2531	if (optlen < sizeof(int))
2532		return -EINVAL;
2533
2534	if (get_user(val, (int __user *)optval))
2535		return -EFAULT;
2536
2537	valbool = val ? 1 : 0;
2538
2539	switch (optname) {
2540	case UDP_CORK:
2541		if (val != 0) {
2542			up->corkflag = 1;
2543		} else {
2544			up->corkflag = 0;
2545			lock_sock(sk);
2546			push_pending_frames(sk);
2547			release_sock(sk);
2548		}
2549		break;
2550
2551	case UDP_ENCAP:
2552		switch (val) {
2553		case 0:
2554		case UDP_ENCAP_ESPINUDP:
2555		case UDP_ENCAP_ESPINUDP_NON_IKE:
2556			up->encap_rcv = xfrm4_udp_encap_rcv;
2557			/* FALLTHROUGH */
2558		case UDP_ENCAP_L2TPINUDP:
2559			up->encap_type = val;
2560			lock_sock(sk);
2561			udp_tunnel_encap_enable(sk->sk_socket);
2562			release_sock(sk);
2563			break;
2564		default:
2565			err = -ENOPROTOOPT;
2566			break;
2567		}
2568		break;
2569
2570	case UDP_NO_CHECK6_TX:
2571		up->no_check6_tx = valbool;
2572		break;
2573
2574	case UDP_NO_CHECK6_RX:
2575		up->no_check6_rx = valbool;
2576		break;
2577
2578	case UDP_SEGMENT:
2579		if (val < 0 || val > USHRT_MAX)
2580			return -EINVAL;
2581		up->gso_size = val;
2582		break;
2583
2584	case UDP_GRO:
2585		lock_sock(sk);
2586		if (valbool)
2587			udp_tunnel_encap_enable(sk->sk_socket);
2588		up->gro_enabled = valbool;
2589		release_sock(sk);
2590		break;
2591
2592	/*
2593	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
2594	 */
2595	/* The sender sets actual checksum coverage length via this option.
2596	 * The case coverage > packet length is handled by send module. */
2597	case UDPLITE_SEND_CSCOV:
2598		if (!is_udplite)         /* Disable the option on UDP sockets */
2599			return -ENOPROTOOPT;
2600		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2601			val = 8;
2602		else if (val > USHRT_MAX)
2603			val = USHRT_MAX;
2604		up->pcslen = val;
2605		up->pcflag |= UDPLITE_SEND_CC;
2606		break;
2607
2608	/* The receiver specifies a minimum checksum coverage value. To make
2609	 * sense, this should be set to at least 8 (as done below). If zero is
2610	 * used, this again means full checksum coverage.                     */
2611	case UDPLITE_RECV_CSCOV:
2612		if (!is_udplite)         /* Disable the option on UDP sockets */
2613			return -ENOPROTOOPT;
2614		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2615			val = 8;
2616		else if (val > USHRT_MAX)
2617			val = USHRT_MAX;
2618		up->pcrlen = val;
2619		up->pcflag |= UDPLITE_RECV_CC;
2620		break;
2621
2622	default:
2623		err = -ENOPROTOOPT;
2624		break;
2625	}
2626
2627	return err;
2628}
2629EXPORT_SYMBOL(udp_lib_setsockopt);
2630
2631int udp_setsockopt(struct sock *sk, int level, int optname,
2632		   char __user *optval, unsigned int optlen)
2633{
2634	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2635		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2636					  udp_push_pending_frames);
2637	return ip_setsockopt(sk, level, optname, optval, optlen);
2638}
2639
2640#ifdef CONFIG_COMPAT
2641int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2642			  char __user *optval, unsigned int optlen)
2643{
2644	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2645		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2646					  udp_push_pending_frames);
2647	return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2648}
2649#endif
2650
2651int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2652		       char __user *optval, int __user *optlen)
2653{
2654	struct udp_sock *up = udp_sk(sk);
2655	int val, len;
2656
2657	if (get_user(len, optlen))
2658		return -EFAULT;
2659
2660	len = min_t(unsigned int, len, sizeof(int));
2661
2662	if (len < 0)
2663		return -EINVAL;
2664
2665	switch (optname) {
2666	case UDP_CORK:
2667		val = up->corkflag;
2668		break;
2669
2670	case UDP_ENCAP:
2671		val = up->encap_type;
2672		break;
2673
2674	case UDP_NO_CHECK6_TX:
2675		val = up->no_check6_tx;
2676		break;
2677
2678	case UDP_NO_CHECK6_RX:
2679		val = up->no_check6_rx;
2680		break;
2681
2682	case UDP_SEGMENT:
2683		val = up->gso_size;
2684		break;
2685
2686	/* The following two cannot be changed on UDP sockets, the return is
2687	 * always 0 (which corresponds to the full checksum coverage of UDP). */
2688	case UDPLITE_SEND_CSCOV:
2689		val = up->pcslen;
2690		break;
2691
2692	case UDPLITE_RECV_CSCOV:
2693		val = up->pcrlen;
2694		break;
2695
2696	default:
2697		return -ENOPROTOOPT;
2698	}
2699
2700	if (put_user(len, optlen))
2701		return -EFAULT;
2702	if (copy_to_user(optval, &val, len))
2703		return -EFAULT;
2704	return 0;
2705}
2706EXPORT_SYMBOL(udp_lib_getsockopt);
2707
2708int udp_getsockopt(struct sock *sk, int level, int optname,
2709		   char __user *optval, int __user *optlen)
2710{
2711	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2712		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2713	return ip_getsockopt(sk, level, optname, optval, optlen);
2714}
2715
2716#ifdef CONFIG_COMPAT
2717int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2718				 char __user *optval, int __user *optlen)
2719{
2720	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2721		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2722	return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2723}
2724#endif
2725/**
2726 * 	udp_poll - wait for a UDP event.
2727 *	@file - file struct
2728 *	@sock - socket
2729 *	@wait - poll table
2730 *
2731 *	This is same as datagram poll, except for the special case of
2732 *	blocking sockets. If application is using a blocking fd
2733 *	and a packet with checksum error is in the queue;
2734 *	then it could get return from select indicating data available
2735 *	but then block when reading it. Add special case code
2736 *	to work around these arguably broken applications.
2737 */
2738__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2739{
2740	__poll_t mask = datagram_poll(file, sock, wait);
2741	struct sock *sk = sock->sk;
2742
2743	if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2744		mask |= EPOLLIN | EPOLLRDNORM;
2745
2746	/* Check for false positives due to checksum errors */
2747	if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2748	    !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2749		mask &= ~(EPOLLIN | EPOLLRDNORM);
2750
2751	return mask;
2752
2753}
2754EXPORT_SYMBOL(udp_poll);
2755
2756int udp_abort(struct sock *sk, int err)
2757{
2758	lock_sock(sk);
2759
2760	sk->sk_err = err;
2761	sk->sk_error_report(sk);
2762	__udp_disconnect(sk, 0);
2763
2764	release_sock(sk);
2765
2766	return 0;
2767}
2768EXPORT_SYMBOL_GPL(udp_abort);
2769
2770struct proto udp_prot = {
2771	.name			= "UDP",
2772	.owner			= THIS_MODULE,
2773	.close			= udp_lib_close,
2774	.pre_connect		= udp_pre_connect,
2775	.connect		= ip4_datagram_connect,
2776	.disconnect		= udp_disconnect,
2777	.ioctl			= udp_ioctl,
2778	.init			= udp_init_sock,
2779	.destroy		= udp_destroy_sock,
2780	.setsockopt		= udp_setsockopt,
2781	.getsockopt		= udp_getsockopt,
2782	.sendmsg		= udp_sendmsg,
2783	.recvmsg		= udp_recvmsg,
2784	.sendpage		= udp_sendpage,
2785	.release_cb		= ip4_datagram_release_cb,
2786	.hash			= udp_lib_hash,
2787	.unhash			= udp_lib_unhash,
2788	.rehash			= udp_v4_rehash,
2789	.get_port		= udp_v4_get_port,
2790	.memory_allocated	= &udp_memory_allocated,
2791	.sysctl_mem		= sysctl_udp_mem,
2792	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2793	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2794	.obj_size		= sizeof(struct udp_sock),
2795	.h.udp_table		= &udp_table,
2796#ifdef CONFIG_COMPAT
2797	.compat_setsockopt	= compat_udp_setsockopt,
2798	.compat_getsockopt	= compat_udp_getsockopt,
2799#endif
2800	.diag_destroy		= udp_abort,
2801};
2802EXPORT_SYMBOL(udp_prot);
2803
2804/* ------------------------------------------------------------------------ */
2805#ifdef CONFIG_PROC_FS
2806
2807static struct sock *udp_get_first(struct seq_file *seq, int start)
2808{
2809	struct sock *sk;
2810	struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2811	struct udp_iter_state *state = seq->private;
2812	struct net *net = seq_file_net(seq);
2813
2814	for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2815	     ++state->bucket) {
2816		struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
 
2817
2818		if (hlist_empty(&hslot->head))
2819			continue;
2820
2821		spin_lock_bh(&hslot->lock);
2822		sk_for_each(sk, &hslot->head) {
2823			if (!net_eq(sock_net(sk), net))
2824				continue;
2825			if (sk->sk_family == afinfo->family)
2826				goto found;
2827		}
2828		spin_unlock_bh(&hslot->lock);
2829	}
2830	sk = NULL;
2831found:
2832	return sk;
2833}
2834
2835static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2836{
2837	struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2838	struct udp_iter_state *state = seq->private;
2839	struct net *net = seq_file_net(seq);
2840
2841	do {
2842		sk = sk_next(sk);
2843	} while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2844
2845	if (!sk) {
2846		if (state->bucket <= afinfo->udp_table->mask)
2847			spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2848		return udp_get_first(seq, state->bucket + 1);
2849	}
2850	return sk;
2851}
2852
2853static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2854{
2855	struct sock *sk = udp_get_first(seq, 0);
2856
2857	if (sk)
2858		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2859			--pos;
2860	return pos ? NULL : sk;
2861}
2862
2863void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2864{
2865	struct udp_iter_state *state = seq->private;
2866	state->bucket = MAX_UDP_PORTS;
2867
2868	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2869}
2870EXPORT_SYMBOL(udp_seq_start);
2871
2872void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2873{
2874	struct sock *sk;
2875
2876	if (v == SEQ_START_TOKEN)
2877		sk = udp_get_idx(seq, 0);
2878	else
2879		sk = udp_get_next(seq, v);
2880
2881	++*pos;
2882	return sk;
2883}
2884EXPORT_SYMBOL(udp_seq_next);
2885
2886void udp_seq_stop(struct seq_file *seq, void *v)
2887{
2888	struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2889	struct udp_iter_state *state = seq->private;
2890
2891	if (state->bucket <= afinfo->udp_table->mask)
2892		spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2893}
2894EXPORT_SYMBOL(udp_seq_stop);
2895
2896/* ------------------------------------------------------------------------ */
2897static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2898		int bucket)
2899{
2900	struct inet_sock *inet = inet_sk(sp);
2901	__be32 dest = inet->inet_daddr;
2902	__be32 src  = inet->inet_rcv_saddr;
2903	__u16 destp	  = ntohs(inet->inet_dport);
2904	__u16 srcp	  = ntohs(inet->inet_sport);
2905
2906	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2907		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
2908		bucket, src, srcp, dest, destp, sp->sk_state,
2909		sk_wmem_alloc_get(sp),
2910		udp_rqueue_get(sp),
2911		0, 0L, 0,
2912		from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2913		0, sock_i_ino(sp),
2914		refcount_read(&sp->sk_refcnt), sp,
2915		atomic_read(&sp->sk_drops));
2916}
2917
2918int udp4_seq_show(struct seq_file *seq, void *v)
2919{
2920	seq_setwidth(seq, 127);
2921	if (v == SEQ_START_TOKEN)
2922		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
 
2923			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2924			   "inode ref pointer drops");
2925	else {
2926		struct udp_iter_state *state = seq->private;
 
2927
2928		udp4_format_sock(v, seq, state->bucket);
 
2929	}
2930	seq_pad(seq, '\n');
2931	return 0;
2932}
2933
2934const struct seq_operations udp_seq_ops = {
2935	.start		= udp_seq_start,
2936	.next		= udp_seq_next,
2937	.stop		= udp_seq_stop,
2938	.show		= udp4_seq_show,
2939};
2940EXPORT_SYMBOL(udp_seq_ops);
2941
2942static struct udp_seq_afinfo udp4_seq_afinfo = {
 
2943	.family		= AF_INET,
2944	.udp_table	= &udp_table,
 
 
 
 
 
 
2945};
2946
2947static int __net_init udp4_proc_init_net(struct net *net)
2948{
2949	if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2950			sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2951		return -ENOMEM;
2952	return 0;
2953}
2954
2955static void __net_exit udp4_proc_exit_net(struct net *net)
2956{
2957	remove_proc_entry("udp", net->proc_net);
2958}
2959
2960static struct pernet_operations udp4_net_ops = {
2961	.init = udp4_proc_init_net,
2962	.exit = udp4_proc_exit_net,
2963};
2964
2965int __init udp4_proc_init(void)
2966{
2967	return register_pernet_subsys(&udp4_net_ops);
2968}
2969
2970void udp4_proc_exit(void)
2971{
2972	unregister_pernet_subsys(&udp4_net_ops);
2973}
2974#endif /* CONFIG_PROC_FS */
2975
2976static __initdata unsigned long uhash_entries;
2977static int __init set_uhash_entries(char *str)
2978{
2979	ssize_t ret;
2980
2981	if (!str)
2982		return 0;
2983
2984	ret = kstrtoul(str, 0, &uhash_entries);
2985	if (ret)
2986		return 0;
2987
2988	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2989		uhash_entries = UDP_HTABLE_SIZE_MIN;
2990	return 1;
2991}
2992__setup("uhash_entries=", set_uhash_entries);
2993
2994void __init udp_table_init(struct udp_table *table, const char *name)
2995{
2996	unsigned int i;
2997
2998	table->hash = alloc_large_system_hash(name,
2999					      2 * sizeof(struct udp_hslot),
3000					      uhash_entries,
3001					      21, /* one slot per 2 MB */
3002					      0,
3003					      &table->log,
3004					      &table->mask,
3005					      UDP_HTABLE_SIZE_MIN,
3006					      64 * 1024);
3007
 
 
 
 
 
 
 
 
 
 
3008	table->hash2 = table->hash + (table->mask + 1);
3009	for (i = 0; i <= table->mask; i++) {
3010		INIT_HLIST_HEAD(&table->hash[i].head);
3011		table->hash[i].count = 0;
3012		spin_lock_init(&table->hash[i].lock);
3013	}
3014	for (i = 0; i <= table->mask; i++) {
3015		INIT_HLIST_HEAD(&table->hash2[i].head);
3016		table->hash2[i].count = 0;
3017		spin_lock_init(&table->hash2[i].lock);
3018	}
3019}
3020
3021u32 udp_flow_hashrnd(void)
3022{
3023	static u32 hashrnd __read_mostly;
3024
3025	net_get_random_once(&hashrnd, sizeof(hashrnd));
 
 
 
 
 
3026
3027	return hashrnd;
 
3028}
3029EXPORT_SYMBOL(udp_flow_hashrnd);
3030
3031static void __udp_sysctl_init(struct net *net)
3032{
3033	net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3034	net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3035
3036#ifdef CONFIG_NET_L3_MASTER_DEV
3037	net->ipv4.sysctl_udp_l3mdev_accept = 0;
3038#endif
3039}
3040
3041static int __net_init udp_sysctl_init(struct net *net)
3042{
3043	__udp_sysctl_init(net);
 
 
 
 
 
3044	return 0;
3045}
3046
3047static struct pernet_operations __net_initdata udp_sysctl_ops = {
3048	.init	= udp_sysctl_init,
3049};
3050
3051void __init udp_init(void)
3052{
3053	unsigned long limit;
3054	unsigned int i;
 
 
3055
3056	udp_table_init(&udp_table, "UDP");
3057	limit = nr_free_buffer_pages() / 8;
3058	limit = max(limit, 128UL);
3059	sysctl_udp_mem[0] = limit / 4 * 3;
3060	sysctl_udp_mem[1] = limit;
3061	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3062
3063	__udp_sysctl_init(&init_net);
 
 
3064
3065	/* 16 spinlocks per cpu */
3066	udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3067	udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3068				GFP_KERNEL);
3069	if (!udp_busylocks)
3070		panic("UDP: failed to alloc udp_busylocks\n");
3071	for (i = 0; i < (1U << udp_busylocks_log); i++)
3072		spin_lock_init(udp_busylocks + i);
3073
3074	if (register_pernet_subsys(&udp_sysctl_ops))
3075		panic("UDP: failed to init sysctl parameters.\n");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3076}
v3.1
 
   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#include <asm/system.h>
  81#include <asm/uaccess.h>
 
  82#include <asm/ioctls.h>
  83#include <linux/bootmem.h>
  84#include <linux/highmem.h>
  85#include <linux/swap.h>
  86#include <linux/types.h>
  87#include <linux/fcntl.h>
  88#include <linux/module.h>
  89#include <linux/socket.h>
  90#include <linux/sockios.h>
  91#include <linux/igmp.h>
 
  92#include <linux/in.h>
  93#include <linux/errno.h>
  94#include <linux/timer.h>
  95#include <linux/mm.h>
  96#include <linux/inet.h>
  97#include <linux/netdevice.h>
  98#include <linux/slab.h>
  99#include <net/tcp_states.h>
 100#include <linux/skbuff.h>
 101#include <linux/proc_fs.h>
 102#include <linux/seq_file.h>
 103#include <net/net_namespace.h>
 104#include <net/icmp.h>
 
 
 105#include <net/route.h>
 106#include <net/checksum.h>
 107#include <net/xfrm.h>
 108#include <trace/events/udp.h>
 
 
 
 109#include "udp_impl.h"
 
 
 
 110
 111struct udp_table udp_table __read_mostly;
 112EXPORT_SYMBOL(udp_table);
 113
 114long sysctl_udp_mem[3] __read_mostly;
 115EXPORT_SYMBOL(sysctl_udp_mem);
 116
 117int sysctl_udp_rmem_min __read_mostly;
 118EXPORT_SYMBOL(sysctl_udp_rmem_min);
 119
 120int sysctl_udp_wmem_min __read_mostly;
 121EXPORT_SYMBOL(sysctl_udp_wmem_min);
 122
 123atomic_long_t udp_memory_allocated;
 124EXPORT_SYMBOL(udp_memory_allocated);
 125
 126#define MAX_UDP_PORTS 65536
 127#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
 128
 129static int udp_lib_lport_inuse(struct net *net, __u16 num,
 130			       const struct udp_hslot *hslot,
 131			       unsigned long *bitmap,
 132			       struct sock *sk,
 133			       int (*saddr_comp)(const struct sock *sk1,
 134						 const struct sock *sk2),
 135			       unsigned int log)
 136{
 137	struct sock *sk2;
 138	struct hlist_nulls_node *node;
 139
 140	sk_nulls_for_each(sk2, node, &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		    (*saddr_comp)(sk, sk2)) {
 148			if (bitmap)
 
 
 
 
 
 
 
 149				__set_bit(udp_sk(sk2)->udp_port_hash >> log,
 150					  bitmap);
 151			else
 152				return 1;
 153		}
 
 154	return 0;
 155}
 156
 157/*
 158 * Note: we still hold spinlock of primary hash chain, so no other writer
 159 * can insert/delete a socket with local_port == num
 160 */
 161static int udp_lib_lport_inuse2(struct net *net, __u16 num,
 162			       struct udp_hslot *hslot2,
 163			       struct sock *sk,
 164			       int (*saddr_comp)(const struct sock *sk1,
 165						 const struct sock *sk2))
 166{
 167	struct sock *sk2;
 168	struct hlist_nulls_node *node;
 169	int res = 0;
 170
 171	spin_lock(&hslot2->lock);
 172	udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
 173		if (net_eq(sock_net(sk2), net) &&
 174		    sk2 != sk &&
 175		    (udp_sk(sk2)->udp_port_hash == num) &&
 176		    (!sk2->sk_reuse || !sk->sk_reuse) &&
 177		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 178		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 179		    (*saddr_comp)(sk, sk2)) {
 180			res = 1;
 
 
 
 
 
 
 181			break;
 182		}
 
 183	spin_unlock(&hslot2->lock);
 184	return res;
 185}
 186
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 187/**
 188 *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
 189 *
 190 *  @sk:          socket struct in question
 191 *  @snum:        port number to look up
 192 *  @saddr_comp:  AF-dependent comparison of bound local IP addresses
 193 *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
 194 *                   with NULL address
 195 */
 196int udp_lib_get_port(struct sock *sk, unsigned short snum,
 197		       int (*saddr_comp)(const struct sock *sk1,
 198					 const struct sock *sk2),
 199		     unsigned int hash2_nulladdr)
 200{
 201	struct udp_hslot *hslot, *hslot2;
 202	struct udp_table *udptable = sk->sk_prot->h.udp_table;
 203	int    error = 1;
 204	struct net *net = sock_net(sk);
 205
 206	if (!snum) {
 207		int low, high, remaining;
 208		unsigned rand;
 209		unsigned short first, last;
 210		DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
 211
 212		inet_get_local_port_range(&low, &high);
 213		remaining = (high - low) + 1;
 214
 215		rand = net_random();
 216		first = (((u64)rand * remaining) >> 32) + low;
 217		/*
 218		 * force rand to be an odd multiple of UDP_HTABLE_SIZE
 219		 */
 220		rand = (rand | 1) * (udptable->mask + 1);
 221		last = first + udptable->mask + 1;
 222		do {
 223			hslot = udp_hashslot(udptable, net, first);
 224			bitmap_zero(bitmap, PORTS_PER_CHAIN);
 225			spin_lock_bh(&hslot->lock);
 226			udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
 227					    saddr_comp, udptable->log);
 228
 229			snum = first;
 230			/*
 231			 * Iterate on all possible values of snum for this hash.
 232			 * Using steps of an odd multiple of UDP_HTABLE_SIZE
 233			 * give us randomization and full range coverage.
 234			 */
 235			do {
 236				if (low <= snum && snum <= high &&
 237				    !test_bit(snum >> udptable->log, bitmap) &&
 238				    !inet_is_reserved_local_port(snum))
 239					goto found;
 240				snum += rand;
 241			} while (snum != first);
 242			spin_unlock_bh(&hslot->lock);
 
 243		} while (++first != last);
 244		goto fail;
 245	} else {
 246		hslot = udp_hashslot(udptable, net, snum);
 247		spin_lock_bh(&hslot->lock);
 248		if (hslot->count > 10) {
 249			int exist;
 250			unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
 251
 252			slot2          &= udptable->mask;
 253			hash2_nulladdr &= udptable->mask;
 254
 255			hslot2 = udp_hashslot2(udptable, slot2);
 256			if (hslot->count < hslot2->count)
 257				goto scan_primary_hash;
 258
 259			exist = udp_lib_lport_inuse2(net, snum, hslot2,
 260						     sk, saddr_comp);
 261			if (!exist && (hash2_nulladdr != slot2)) {
 262				hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
 263				exist = udp_lib_lport_inuse2(net, snum, hslot2,
 264							     sk, saddr_comp);
 265			}
 266			if (exist)
 267				goto fail_unlock;
 268			else
 269				goto found;
 270		}
 271scan_primary_hash:
 272		if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
 273					saddr_comp, 0))
 274			goto fail_unlock;
 275	}
 276found:
 277	inet_sk(sk)->inet_num = snum;
 278	udp_sk(sk)->udp_port_hash = snum;
 279	udp_sk(sk)->udp_portaddr_hash ^= snum;
 280	if (sk_unhashed(sk)) {
 281		sk_nulls_add_node_rcu(sk, &hslot->head);
 
 
 
 
 
 
 
 
 282		hslot->count++;
 283		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 284
 285		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 286		spin_lock(&hslot2->lock);
 287		hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
 288					 &hslot2->head);
 
 
 
 
 
 289		hslot2->count++;
 290		spin_unlock(&hslot2->lock);
 291	}
 
 292	error = 0;
 293fail_unlock:
 294	spin_unlock_bh(&hslot->lock);
 295fail:
 296	return error;
 297}
 298EXPORT_SYMBOL(udp_lib_get_port);
 299
 300static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
 301{
 302	struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
 303
 304	return 	(!ipv6_only_sock(sk2)  &&
 305		 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
 306		   inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
 307}
 308
 309static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
 310				       unsigned int port)
 311{
 312	return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
 313}
 314
 315int udp_v4_get_port(struct sock *sk, unsigned short snum)
 316{
 317	unsigned int hash2_nulladdr =
 318		udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
 319	unsigned int hash2_partial =
 320		udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
 321
 322	/* precompute partial secondary hash */
 323	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
 324	return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
 325}
 326
 327static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
 328			 unsigned short hnum,
 329			 __be16 sport, __be32 daddr, __be16 dport, int dif)
 330{
 331	int score = -1;
 332
 333	if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
 334			!ipv6_only_sock(sk)) {
 335		struct inet_sock *inet = inet_sk(sk);
 336
 337		score = (sk->sk_family == PF_INET ? 1 : 0);
 338		if (inet->inet_rcv_saddr) {
 339			if (inet->inet_rcv_saddr != daddr)
 340				return -1;
 341			score += 2;
 342		}
 343		if (inet->inet_daddr) {
 344			if (inet->inet_daddr != saddr)
 345				return -1;
 346			score += 2;
 347		}
 348		if (inet->inet_dport) {
 349			if (inet->inet_dport != sport)
 350				return -1;
 351			score += 2;
 352		}
 353		if (sk->sk_bound_dev_if) {
 354			if (sk->sk_bound_dev_if != dif)
 355				return -1;
 356			score += 2;
 357		}
 358	}
 359	return score;
 360}
 361
 362/*
 363 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
 364 */
 365#define SCORE2_MAX (1 + 2 + 2 + 2)
 366static inline int compute_score2(struct sock *sk, struct net *net,
 367				 __be32 saddr, __be16 sport,
 368				 __be32 daddr, unsigned int hnum, int dif)
 369{
 370	int score = -1;
 371
 372	if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
 373		struct inet_sock *inet = inet_sk(sk);
 374
 375		if (inet->inet_rcv_saddr != daddr)
 
 
 376			return -1;
 377		if (inet->inet_num != hnum)
 
 
 
 
 378			return -1;
 
 
 379
 380		score = (sk->sk_family == PF_INET ? 1 : 0);
 381		if (inet->inet_daddr) {
 382			if (inet->inet_daddr != saddr)
 383				return -1;
 384			score += 2;
 385		}
 386		if (inet->inet_dport) {
 387			if (inet->inet_dport != sport)
 388				return -1;
 389			score += 2;
 390		}
 391		if (sk->sk_bound_dev_if) {
 392			if (sk->sk_bound_dev_if != dif)
 393				return -1;
 394			score += 2;
 395		}
 396	}
 397	return score;
 398}
 399
 
 
 
 
 
 400
 401/* called with read_rcu_lock() */
 
 
 
 
 
 
 402static struct sock *udp4_lib_lookup2(struct net *net,
 403		__be32 saddr, __be16 sport,
 404		__be32 daddr, unsigned int hnum, int dif,
 405		struct udp_hslot *hslot2, unsigned int slot2)
 
 
 406{
 407	struct sock *sk, *result;
 408	struct hlist_nulls_node *node;
 409	int score, badness;
 
 410
 411begin:
 412	result = NULL;
 413	badness = -1;
 414	udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
 415		score = compute_score2(sk, net, saddr, sport,
 416				      daddr, hnum, dif);
 417		if (score > badness) {
 
 
 
 
 
 
 
 
 
 
 418			result = sk;
 419			badness = score;
 420			if (score == SCORE2_MAX)
 421				goto exact_match;
 422		}
 423	}
 424	/*
 425	 * if the nulls value we got at the end of this lookup is
 426	 * not the expected one, we must restart lookup.
 427	 * We probably met an item that was moved to another chain.
 428	 */
 429	if (get_nulls_value(node) != slot2)
 430		goto begin;
 431
 432	if (result) {
 433exact_match:
 434		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
 435			result = NULL;
 436		else if (unlikely(compute_score2(result, net, saddr, sport,
 437				  daddr, hnum, dif) < badness)) {
 438			sock_put(result);
 439			goto begin;
 440		}
 441	}
 442	return result;
 443}
 444
 445/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
 446 * harder than this. -DaveM
 447 */
 448static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
 449		__be16 sport, __be32 daddr, __be16 dport,
 450		int dif, struct udp_table *udptable)
 451{
 452	struct sock *sk, *result;
 453	struct hlist_nulls_node *node;
 454	unsigned short hnum = ntohs(dport);
 455	unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
 456	struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
 457	int score, badness;
 458
 459	rcu_read_lock();
 460	if (hslot->count > 10) {
 461		hash2 = udp4_portaddr_hash(net, daddr, hnum);
 
 
 
 
 
 
 462		slot2 = hash2 & udptable->mask;
 463		hslot2 = &udptable->hash2[slot2];
 464		if (hslot->count < hslot2->count)
 465			goto begin;
 466
 467		result = udp4_lib_lookup2(net, saddr, sport,
 468					  daddr, hnum, dif,
 469					  hslot2, slot2);
 470		if (!result) {
 471			hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
 472			slot2 = hash2 & udptable->mask;
 473			hslot2 = &udptable->hash2[slot2];
 474			if (hslot->count < hslot2->count)
 475				goto begin;
 476
 477			result = udp4_lib_lookup2(net, saddr, sport,
 478						  htonl(INADDR_ANY), hnum, dif,
 479						  hslot2, slot2);
 480		}
 481		rcu_read_unlock();
 482		return result;
 483	}
 484begin:
 485	result = NULL;
 486	badness = -1;
 487	sk_nulls_for_each_rcu(sk, node, &hslot->head) {
 488		score = compute_score(sk, net, saddr, hnum, sport,
 489				      daddr, dport, dif);
 490		if (score > badness) {
 491			result = sk;
 492			badness = score;
 493		}
 494	}
 495	/*
 496	 * if the nulls value we got at the end of this lookup is
 497	 * not the expected one, we must restart lookup.
 498	 * We probably met an item that was moved to another chain.
 499	 */
 500	if (get_nulls_value(node) != slot)
 501		goto begin;
 502
 503	if (result) {
 504		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
 505			result = NULL;
 506		else if (unlikely(compute_score(result, net, saddr, hnum, sport,
 507				  daddr, dport, dif) < badness)) {
 508			sock_put(result);
 509			goto begin;
 510		}
 511	}
 512	rcu_read_unlock();
 
 513	return result;
 514}
 
 515
 516static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
 517						 __be16 sport, __be16 dport,
 518						 struct udp_table *udptable)
 519{
 520	struct sock *sk;
 521	const struct iphdr *iph = ip_hdr(skb);
 522
 523	if (unlikely(sk = skb_steal_sock(skb)))
 524		return sk;
 525	else
 526		return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
 527					 iph->daddr, dport, inet_iif(skb),
 528					 udptable);
 
 
 
 
 
 
 
 529}
 
 530
 
 
 
 
 531struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 532			     __be32 daddr, __be16 dport, int dif)
 533{
 534	return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
 
 
 
 
 
 
 535}
 536EXPORT_SYMBOL_GPL(udp4_lib_lookup);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 537
 538static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
 539					     __be16 loc_port, __be32 loc_addr,
 540					     __be16 rmt_port, __be32 rmt_addr,
 541					     int dif)
 542{
 543	struct hlist_nulls_node *node;
 544	struct sock *s = sk;
 545	unsigned short hnum = ntohs(loc_port);
 546
 547	sk_nulls_for_each_from(s, node) {
 548		struct inet_sock *inet = inet_sk(s);
 
 549
 550		if (!net_eq(sock_net(s), net) ||
 551		    udp_sk(s)->udp_port_hash != hnum ||
 552		    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
 553		    (inet->inet_dport != rmt_port && inet->inet_dport) ||
 554		    (inet->inet_rcv_saddr &&
 555		     inet->inet_rcv_saddr != loc_addr) ||
 556		    ipv6_only_sock(s) ||
 557		    (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
 558			continue;
 559		if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
 560			continue;
 561		goto found;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 562	}
 563	s = NULL;
 564found:
 565	return s;
 
 
 
 
 
 566}
 567
 568/*
 569 * This routine is called by the ICMP module when it gets some
 570 * sort of error condition.  If err < 0 then the socket should
 571 * be closed and the error returned to the user.  If err > 0
 572 * it's just the icmp type << 8 | icmp code.
 573 * Header points to the ip header of the error packet. We move
 574 * on past this. Then (as it used to claim before adjustment)
 575 * header points to the first 8 bytes of the udp header.  We need
 576 * to find the appropriate port.
 577 */
 578
 579void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
 580{
 581	struct inet_sock *inet;
 582	const struct iphdr *iph = (const struct iphdr *)skb->data;
 583	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
 584	const int type = icmp_hdr(skb)->type;
 585	const int code = icmp_hdr(skb)->code;
 
 586	struct sock *sk;
 587	int harderr;
 588	int err;
 589	struct net *net = dev_net(skb->dev);
 590
 591	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
 592			iph->saddr, uh->source, skb->dev->ifindex, udptable);
 593	if (sk == NULL) {
 594		ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 595		return;	/* No socket for error */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 596	}
 597
 598	err = 0;
 599	harderr = 0;
 600	inet = inet_sk(sk);
 601
 602	switch (type) {
 603	default:
 604	case ICMP_TIME_EXCEEDED:
 605		err = EHOSTUNREACH;
 606		break;
 607	case ICMP_SOURCE_QUENCH:
 608		goto out;
 609	case ICMP_PARAMETERPROB:
 610		err = EPROTO;
 611		harderr = 1;
 612		break;
 613	case ICMP_DEST_UNREACH:
 614		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
 
 615			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
 616				err = EMSGSIZE;
 617				harderr = 1;
 618				break;
 619			}
 620			goto out;
 621		}
 622		err = EHOSTUNREACH;
 623		if (code <= NR_ICMP_UNREACH) {
 624			harderr = icmp_err_convert[code].fatal;
 625			err = icmp_err_convert[code].errno;
 626		}
 627		break;
 
 
 
 628	}
 629
 630	/*
 631	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
 632	 *	4.1.3.3.
 633	 */
 
 
 
 
 634	if (!inet->recverr) {
 635		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
 636			goto out;
 637	} else
 638		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
 639
 640	sk->sk_err = err;
 641	sk->sk_error_report(sk);
 642out:
 643	sock_put(sk);
 644}
 645
 646void udp_err(struct sk_buff *skb, u32 info)
 647{
 648	__udp4_lib_err(skb, info, &udp_table);
 649}
 650
 651/*
 652 * Throw away all pending data and cancel the corking. Socket is locked.
 653 */
 654void udp_flush_pending_frames(struct sock *sk)
 655{
 656	struct udp_sock *up = udp_sk(sk);
 657
 658	if (up->pending) {
 659		up->len = 0;
 660		up->pending = 0;
 661		ip_flush_pending_frames(sk);
 662	}
 663}
 664EXPORT_SYMBOL(udp_flush_pending_frames);
 665
 666/**
 667 * 	udp4_hwcsum  -  handle outgoing HW checksumming
 668 * 	@skb: 	sk_buff containing the filled-in UDP header
 669 * 	        (checksum field must be zeroed out)
 670 *	@src:	source IP address
 671 *	@dst:	destination IP address
 672 */
 673static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
 674{
 675	struct udphdr *uh = udp_hdr(skb);
 676	struct sk_buff *frags = skb_shinfo(skb)->frag_list;
 677	int offset = skb_transport_offset(skb);
 678	int len = skb->len - offset;
 679	int hlen = len;
 680	__wsum csum = 0;
 681
 682	if (!frags) {
 683		/*
 684		 * Only one fragment on the socket.
 685		 */
 686		skb->csum_start = skb_transport_header(skb) - skb->head;
 687		skb->csum_offset = offsetof(struct udphdr, check);
 688		uh->check = ~csum_tcpudp_magic(src, dst, len,
 689					       IPPROTO_UDP, 0);
 690	} else {
 
 
 691		/*
 692		 * HW-checksum won't work as there are two or more
 693		 * fragments on the socket so that all csums of sk_buffs
 694		 * should be together
 695		 */
 696		do {
 697			csum = csum_add(csum, frags->csum);
 698			hlen -= frags->len;
 699		} while ((frags = frags->next));
 700
 701		csum = skb_checksum(skb, offset, hlen, csum);
 702		skb->ip_summed = CHECKSUM_NONE;
 703
 704		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
 705		if (uh->check == 0)
 706			uh->check = CSUM_MANGLED_0;
 707	}
 708}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 709
 710static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
 
 711{
 712	struct sock *sk = skb->sk;
 713	struct inet_sock *inet = inet_sk(sk);
 714	struct udphdr *uh;
 715	int err = 0;
 716	int is_udplite = IS_UDPLITE(sk);
 717	int offset = skb_transport_offset(skb);
 718	int len = skb->len - offset;
 
 719	__wsum csum = 0;
 720
 721	/*
 722	 * Create a UDP header
 723	 */
 724	uh = udp_hdr(skb);
 725	uh->source = inet->inet_sport;
 726	uh->dest = fl4->fl4_dport;
 727	uh->len = htons(len);
 728	uh->check = 0;
 729
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 730	if (is_udplite)  				 /*     UDP-Lite      */
 731		csum = udplite_csum(skb);
 732
 733	else if (sk->sk_no_check == UDP_CSUM_NOXMIT) {   /* UDP csum disabled */
 734
 735		skb->ip_summed = CHECKSUM_NONE;
 736		goto send;
 737
 738	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
 
 739
 740		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
 741		goto send;
 742
 743	} else
 744		csum = udp_csum(skb);
 745
 746	/* add protocol-dependent pseudo-header */
 747	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
 748				      sk->sk_protocol, csum);
 749	if (uh->check == 0)
 750		uh->check = CSUM_MANGLED_0;
 751
 752send:
 753	err = ip_send_skb(skb);
 754	if (err) {
 755		if (err == -ENOBUFS && !inet->recverr) {
 756			UDP_INC_STATS_USER(sock_net(sk),
 757					   UDP_MIB_SNDBUFERRORS, is_udplite);
 758			err = 0;
 759		}
 760	} else
 761		UDP_INC_STATS_USER(sock_net(sk),
 762				   UDP_MIB_OUTDATAGRAMS, is_udplite);
 763	return err;
 764}
 765
 766/*
 767 * Push out all pending data as one UDP datagram. Socket is locked.
 768 */
 769static int udp_push_pending_frames(struct sock *sk)
 770{
 771	struct udp_sock  *up = udp_sk(sk);
 772	struct inet_sock *inet = inet_sk(sk);
 773	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
 774	struct sk_buff *skb;
 775	int err = 0;
 776
 777	skb = ip_finish_skb(sk, fl4);
 778	if (!skb)
 779		goto out;
 780
 781	err = udp_send_skb(skb, fl4);
 782
 783out:
 784	up->len = 0;
 785	up->pending = 0;
 786	return err;
 787}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 788
 789int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
 790		size_t len)
 791{
 792	struct inet_sock *inet = inet_sk(sk);
 793	struct udp_sock *up = udp_sk(sk);
 
 794	struct flowi4 fl4_stack;
 795	struct flowi4 *fl4;
 796	int ulen = len;
 797	struct ipcm_cookie ipc;
 798	struct rtable *rt = NULL;
 799	int free = 0;
 800	int connected = 0;
 801	__be32 daddr, faddr, saddr;
 802	__be16 dport;
 803	u8  tos;
 804	int err, is_udplite = IS_UDPLITE(sk);
 805	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
 806	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
 807	struct sk_buff *skb;
 808	struct ip_options_data opt_copy;
 809
 810	if (len > 0xFFFF)
 811		return -EMSGSIZE;
 812
 813	/*
 814	 *	Check the flags.
 815	 */
 816
 817	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
 818		return -EOPNOTSUPP;
 819
 820	ipc.opt = NULL;
 821	ipc.tx_flags = 0;
 822
 823	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
 824
 825	fl4 = &inet->cork.fl.u.ip4;
 826	if (up->pending) {
 827		/*
 828		 * There are pending frames.
 829		 * The socket lock must be held while it's corked.
 830		 */
 831		lock_sock(sk);
 832		if (likely(up->pending)) {
 833			if (unlikely(up->pending != AF_INET)) {
 834				release_sock(sk);
 835				return -EINVAL;
 836			}
 837			goto do_append_data;
 838		}
 839		release_sock(sk);
 840	}
 841	ulen += sizeof(struct udphdr);
 842
 843	/*
 844	 *	Get and verify the address.
 845	 */
 846	if (msg->msg_name) {
 847		struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
 848		if (msg->msg_namelen < sizeof(*usin))
 849			return -EINVAL;
 850		if (usin->sin_family != AF_INET) {
 851			if (usin->sin_family != AF_UNSPEC)
 852				return -EAFNOSUPPORT;
 853		}
 854
 855		daddr = usin->sin_addr.s_addr;
 856		dport = usin->sin_port;
 857		if (dport == 0)
 858			return -EINVAL;
 859	} else {
 860		if (sk->sk_state != TCP_ESTABLISHED)
 861			return -EDESTADDRREQ;
 862		daddr = inet->inet_daddr;
 863		dport = inet->inet_dport;
 864		/* Open fast path for connected socket.
 865		   Route will not be used, if at least one option is set.
 866		 */
 867		connected = 1;
 868	}
 869	ipc.addr = inet->inet_saddr;
 870
 871	ipc.oif = sk->sk_bound_dev_if;
 872	err = sock_tx_timestamp(sk, &ipc.tx_flags);
 873	if (err)
 874		return err;
 875	if (msg->msg_controllen) {
 876		err = ip_cmsg_send(sock_net(sk), msg, &ipc);
 877		if (err)
 
 
 
 
 878			return err;
 
 879		if (ipc.opt)
 880			free = 1;
 881		connected = 0;
 882	}
 883	if (!ipc.opt) {
 884		struct ip_options_rcu *inet_opt;
 885
 886		rcu_read_lock();
 887		inet_opt = rcu_dereference(inet->inet_opt);
 888		if (inet_opt) {
 889			memcpy(&opt_copy, inet_opt,
 890			       sizeof(*inet_opt) + inet_opt->opt.optlen);
 891			ipc.opt = &opt_copy.opt;
 892		}
 893		rcu_read_unlock();
 894	}
 895
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 896	saddr = ipc.addr;
 897	ipc.addr = faddr = daddr;
 898
 899	if (ipc.opt && ipc.opt->opt.srr) {
 900		if (!daddr)
 901			return -EINVAL;
 
 
 902		faddr = ipc.opt->opt.faddr;
 903		connected = 0;
 904	}
 905	tos = RT_TOS(inet->tos);
 906	if (sock_flag(sk, SOCK_LOCALROUTE) ||
 907	    (msg->msg_flags & MSG_DONTROUTE) ||
 908	    (ipc.opt && ipc.opt->opt.is_strictroute)) {
 909		tos |= RTO_ONLINK;
 910		connected = 0;
 911	}
 912
 913	if (ipv4_is_multicast(daddr)) {
 914		if (!ipc.oif)
 915			ipc.oif = inet->mc_index;
 916		if (!saddr)
 917			saddr = inet->mc_addr;
 918		connected = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 919	}
 920
 921	if (connected)
 922		rt = (struct rtable *)sk_dst_check(sk, 0);
 923
 924	if (rt == NULL) {
 925		struct net *net = sock_net(sk);
 
 926
 927		fl4 = &fl4_stack;
 928		flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
 
 929				   RT_SCOPE_UNIVERSE, sk->sk_protocol,
 930				   inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
 931				   faddr, saddr, dport, inet->inet_sport);
 
 932
 933		security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
 934		rt = ip_route_output_flow(net, fl4, sk);
 935		if (IS_ERR(rt)) {
 936			err = PTR_ERR(rt);
 937			rt = NULL;
 938			if (err == -ENETUNREACH)
 939				IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 940			goto out;
 941		}
 942
 943		err = -EACCES;
 944		if ((rt->rt_flags & RTCF_BROADCAST) &&
 945		    !sock_flag(sk, SOCK_BROADCAST))
 946			goto out;
 947		if (connected)
 948			sk_dst_set(sk, dst_clone(&rt->dst));
 949	}
 950
 951	if (msg->msg_flags&MSG_CONFIRM)
 952		goto do_confirm;
 953back_from_confirm:
 954
 955	saddr = fl4->saddr;
 956	if (!ipc.addr)
 957		daddr = ipc.addr = fl4->daddr;
 958
 959	/* Lockless fast path for the non-corking case. */
 960	if (!corkreq) {
 961		skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
 
 
 962				  sizeof(struct udphdr), &ipc, &rt,
 963				  msg->msg_flags);
 964		err = PTR_ERR(skb);
 965		if (skb && !IS_ERR(skb))
 966			err = udp_send_skb(skb, fl4);
 967		goto out;
 968	}
 969
 970	lock_sock(sk);
 971	if (unlikely(up->pending)) {
 972		/* The socket is already corked while preparing it. */
 973		/* ... which is an evident application bug. --ANK */
 974		release_sock(sk);
 975
 976		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
 977		err = -EINVAL;
 978		goto out;
 979	}
 980	/*
 981	 *	Now cork the socket to pend data.
 982	 */
 983	fl4 = &inet->cork.fl.u.ip4;
 984	fl4->daddr = daddr;
 985	fl4->saddr = saddr;
 986	fl4->fl4_dport = dport;
 987	fl4->fl4_sport = inet->inet_sport;
 988	up->pending = AF_INET;
 989
 990do_append_data:
 991	up->len += ulen;
 992	err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
 993			     sizeof(struct udphdr), &ipc, &rt,
 994			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
 995	if (err)
 996		udp_flush_pending_frames(sk);
 997	else if (!corkreq)
 998		err = udp_push_pending_frames(sk);
 999	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1000		up->pending = 0;
1001	release_sock(sk);
1002
1003out:
1004	ip_rt_put(rt);
 
1005	if (free)
1006		kfree(ipc.opt);
1007	if (!err)
1008		return len;
1009	/*
1010	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1011	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1012	 * we don't have a good statistic (IpOutDiscards but it can be too many
1013	 * things).  We could add another new stat but at least for now that
1014	 * seems like overkill.
1015	 */
1016	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1017		UDP_INC_STATS_USER(sock_net(sk),
1018				UDP_MIB_SNDBUFERRORS, is_udplite);
1019	}
1020	return err;
1021
1022do_confirm:
1023	dst_confirm(&rt->dst);
 
1024	if (!(msg->msg_flags&MSG_PROBE) || len)
1025		goto back_from_confirm;
1026	err = 0;
1027	goto out;
1028}
1029EXPORT_SYMBOL(udp_sendmsg);
1030
1031int udp_sendpage(struct sock *sk, struct page *page, int offset,
1032		 size_t size, int flags)
1033{
1034	struct inet_sock *inet = inet_sk(sk);
1035	struct udp_sock *up = udp_sk(sk);
1036	int ret;
1037
 
 
 
1038	if (!up->pending) {
1039		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
1040
1041		/* Call udp_sendmsg to specify destination address which
1042		 * sendpage interface can't pass.
1043		 * This will succeed only when the socket is connected.
1044		 */
1045		ret = udp_sendmsg(NULL, sk, &msg, 0);
1046		if (ret < 0)
1047			return ret;
1048	}
1049
1050	lock_sock(sk);
1051
1052	if (unlikely(!up->pending)) {
1053		release_sock(sk);
1054
1055		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
1056		return -EINVAL;
1057	}
1058
1059	ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1060			     page, offset, size, flags);
1061	if (ret == -EOPNOTSUPP) {
1062		release_sock(sk);
1063		return sock_no_sendpage(sk->sk_socket, page, offset,
1064					size, flags);
1065	}
1066	if (ret < 0) {
1067		udp_flush_pending_frames(sk);
1068		goto out;
1069	}
1070
1071	up->len += size;
1072	if (!(up->corkflag || (flags&MSG_MORE)))
1073		ret = udp_push_pending_frames(sk);
1074	if (!ret)
1075		ret = size;
1076out:
1077	release_sock(sk);
1078	return ret;
1079}
1080
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1081
1082/**
1083 *	first_packet_length	- return length of first packet in receive queue
1084 *	@sk: socket
1085 *
1086 *	Drops all bad checksum frames, until a valid one is found.
1087 *	Returns the length of found skb, or 0 if none is found.
1088 */
1089static unsigned int first_packet_length(struct sock *sk)
1090{
1091	struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
 
1092	struct sk_buff *skb;
1093	unsigned int res;
1094
1095	__skb_queue_head_init(&list_kill);
1096
1097	spin_lock_bh(&rcvq->lock);
1098	while ((skb = skb_peek(rcvq)) != NULL &&
1099		udp_lib_checksum_complete(skb)) {
1100		UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1101				 IS_UDPLITE(sk));
1102		atomic_inc(&sk->sk_drops);
1103		__skb_unlink(skb, rcvq);
1104		__skb_queue_tail(&list_kill, skb);
1105	}
1106	res = skb ? skb->len : 0;
 
 
1107	spin_unlock_bh(&rcvq->lock);
1108
1109	if (!skb_queue_empty(&list_kill)) {
1110		bool slow = lock_sock_fast(sk);
1111
1112		__skb_queue_purge(&list_kill);
1113		sk_mem_reclaim_partial(sk);
1114		unlock_sock_fast(sk, slow);
1115	}
1116	return res;
1117}
1118
1119/*
1120 *	IOCTL requests applicable to the UDP protocol
1121 */
1122
1123int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1124{
1125	switch (cmd) {
1126	case SIOCOUTQ:
1127	{
1128		int amount = sk_wmem_alloc_get(sk);
1129
1130		return put_user(amount, (int __user *)arg);
1131	}
1132
1133	case SIOCINQ:
1134	{
1135		unsigned int amount = first_packet_length(sk);
1136
1137		if (amount)
1138			/*
1139			 * We will only return the amount
1140			 * of this packet since that is all
1141			 * that will be read.
1142			 */
1143			amount -= sizeof(struct udphdr);
1144
1145		return put_user(amount, (int __user *)arg);
1146	}
1147
1148	default:
1149		return -ENOIOCTLCMD;
1150	}
1151
1152	return 0;
1153}
1154EXPORT_SYMBOL(udp_ioctl);
1155
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1156/*
1157 * 	This should be easy, if there is something there we
1158 * 	return it, otherwise we block.
1159 */
1160
1161int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1162		size_t len, int noblock, int flags, int *addr_len)
1163{
1164	struct inet_sock *inet = inet_sk(sk);
1165	struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1166	struct sk_buff *skb;
1167	unsigned int ulen;
1168	int peeked;
1169	int err;
1170	int is_udplite = IS_UDPLITE(sk);
1171	bool slow;
1172
1173	/*
1174	 *	Check any passed addresses
1175	 */
1176	if (addr_len)
1177		*addr_len = sizeof(*sin);
1178
1179	if (flags & MSG_ERRQUEUE)
1180		return ip_recv_error(sk, msg, len);
1181
1182try_again:
1183	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1184				  &peeked, &err);
1185	if (!skb)
1186		goto out;
1187
1188	ulen = skb->len - sizeof(struct udphdr);
1189	if (len > ulen)
1190		len = ulen;
1191	else if (len < ulen)
 
1192		msg->msg_flags |= MSG_TRUNC;
1193
1194	/*
1195	 * If checksum is needed at all, try to do it while copying the
1196	 * data.  If the data is truncated, or if we only want a partial
1197	 * coverage checksum (UDP-Lite), do it before the copy.
1198	 */
1199
1200	if (len < ulen || UDP_SKB_CB(skb)->partial_cov) {
1201		if (udp_lib_checksum_complete(skb))
 
 
 
1202			goto csum_copy_err;
1203	}
1204
1205	if (skb_csum_unnecessary(skb))
1206		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1207					      msg->msg_iov, len);
1208	else {
1209		err = skb_copy_and_csum_datagram_iovec(skb,
1210						       sizeof(struct udphdr),
1211						       msg->msg_iov);
1212
1213		if (err == -EINVAL)
1214			goto csum_copy_err;
1215	}
1216
1217	if (err)
1218		goto out_free;
 
 
 
 
 
 
 
1219
1220	if (!peeked)
1221		UDP_INC_STATS_USER(sock_net(sk),
1222				UDP_MIB_INDATAGRAMS, is_udplite);
1223
1224	sock_recv_ts_and_drops(msg, sk, skb);
1225
1226	/* Copy the address. */
1227	if (sin) {
1228		sin->sin_family = AF_INET;
1229		sin->sin_port = udp_hdr(skb)->source;
1230		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1231		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 
 
 
 
 
1232	}
 
 
 
 
1233	if (inet->cmsg_flags)
1234		ip_cmsg_recv(msg, skb);
1235
1236	err = len;
1237	if (flags & MSG_TRUNC)
1238		err = ulen;
1239
1240out_free:
1241	skb_free_datagram_locked(sk, skb);
1242out:
1243	return err;
1244
1245csum_copy_err:
1246	slow = lock_sock_fast(sk);
1247	if (!skb_kill_datagram(sk, skb, flags))
1248		UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1249	unlock_sock_fast(sk, slow);
 
 
1250
1251	if (noblock)
1252		return -EAGAIN;
1253
1254	/* starting over for a new packet */
1255	msg->msg_flags &= ~MSG_TRUNC;
1256	goto try_again;
1257}
1258
 
 
 
 
 
 
 
 
1259
1260int udp_disconnect(struct sock *sk, int flags)
 
 
 
 
1261{
1262	struct inet_sock *inet = inet_sk(sk);
1263	/*
1264	 *	1003.1g - break association.
1265	 */
1266
1267	sk->sk_state = TCP_CLOSE;
1268	inet->inet_daddr = 0;
1269	inet->inet_dport = 0;
1270	sock_rps_save_rxhash(sk, 0);
1271	sk->sk_bound_dev_if = 0;
1272	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1273		inet_reset_saddr(sk);
1274
1275	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1276		sk->sk_prot->unhash(sk);
1277		inet->inet_sport = 0;
1278	}
1279	sk_dst_reset(sk);
1280	return 0;
1281}
 
 
 
 
 
 
 
 
 
1282EXPORT_SYMBOL(udp_disconnect);
1283
1284void udp_lib_unhash(struct sock *sk)
1285{
1286	if (sk_hashed(sk)) {
1287		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1288		struct udp_hslot *hslot, *hslot2;
1289
1290		hslot  = udp_hashslot(udptable, sock_net(sk),
1291				      udp_sk(sk)->udp_port_hash);
1292		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1293
1294		spin_lock_bh(&hslot->lock);
1295		if (sk_nulls_del_node_init_rcu(sk)) {
 
 
1296			hslot->count--;
1297			inet_sk(sk)->inet_num = 0;
1298			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1299
1300			spin_lock(&hslot2->lock);
1301			hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1302			hslot2->count--;
1303			spin_unlock(&hslot2->lock);
1304		}
1305		spin_unlock_bh(&hslot->lock);
1306	}
1307}
1308EXPORT_SYMBOL(udp_lib_unhash);
1309
1310/*
1311 * inet_rcv_saddr was changed, we must rehash secondary hash
1312 */
1313void udp_lib_rehash(struct sock *sk, u16 newhash)
1314{
1315	if (sk_hashed(sk)) {
1316		struct udp_table *udptable = sk->sk_prot->h.udp_table;
1317		struct udp_hslot *hslot, *hslot2, *nhslot2;
1318
1319		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1320		nhslot2 = udp_hashslot2(udptable, newhash);
1321		udp_sk(sk)->udp_portaddr_hash = newhash;
1322		if (hslot2 != nhslot2) {
 
 
1323			hslot = udp_hashslot(udptable, sock_net(sk),
1324					     udp_sk(sk)->udp_port_hash);
1325			/* we must lock primary chain too */
1326			spin_lock_bh(&hslot->lock);
 
 
1327
1328			spin_lock(&hslot2->lock);
1329			hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1330			hslot2->count--;
1331			spin_unlock(&hslot2->lock);
1332
1333			spin_lock(&nhslot2->lock);
1334			hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1335						 &nhslot2->head);
1336			nhslot2->count++;
1337			spin_unlock(&nhslot2->lock);
 
 
1338
1339			spin_unlock_bh(&hslot->lock);
1340		}
1341	}
1342}
1343EXPORT_SYMBOL(udp_lib_rehash);
1344
1345static void udp_v4_rehash(struct sock *sk)
1346{
1347	u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1348					  inet_sk(sk)->inet_rcv_saddr,
1349					  inet_sk(sk)->inet_num);
1350	udp_lib_rehash(sk, new_hash);
1351}
1352
1353static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1354{
1355	int rc;
1356
1357	if (inet_sk(sk)->inet_daddr)
1358		sock_rps_save_rxhash(sk, skb->rxhash);
 
 
 
 
 
1359
1360	rc = ip_queue_rcv_skb(sk, skb);
1361	if (rc < 0) {
1362		int is_udplite = IS_UDPLITE(sk);
1363
1364		/* Note that an ENOMEM error is charged twice */
1365		if (rc == -ENOMEM)
1366			UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1367					 is_udplite);
1368		UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1369		kfree_skb(skb);
1370		trace_udp_fail_queue_rcv_skb(rc, sk);
1371		return -1;
1372	}
1373
1374	return 0;
1375
1376}
1377
1378/* returns:
1379 *  -1: error
1380 *   0: success
1381 *  >0: "udp encap" protocol resubmission
1382 *
1383 * Note that in the success and error cases, the skb is assumed to
1384 * have either been requeued or freed.
1385 */
1386int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1387{
1388	struct udp_sock *up = udp_sk(sk);
1389	int rc;
1390	int is_udplite = IS_UDPLITE(sk);
1391
1392	/*
1393	 *	Charge it to the socket, dropping if the queue is full.
1394	 */
1395	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1396		goto drop;
1397	nf_reset(skb);
 
 
 
1398
1399	if (up->encap_type) {
1400		/*
1401		 * This is an encapsulation socket so pass the skb to
1402		 * the socket's udp_encap_rcv() hook. Otherwise, just
1403		 * fall through and pass this up the UDP socket.
1404		 * up->encap_rcv() returns the following value:
1405		 * =0 if skb was successfully passed to the encap
1406		 *    handler or was discarded by it.
1407		 * >0 if skb should be passed on to UDP.
1408		 * <0 if skb should be resubmitted as proto -N
1409		 */
1410
1411		/* if we're overly short, let UDP handle it */
1412		if (skb->len > sizeof(struct udphdr) &&
1413		    up->encap_rcv != NULL) {
1414			int ret;
1415
1416			ret = (*up->encap_rcv)(sk, skb);
 
 
 
 
1417			if (ret <= 0) {
1418				UDP_INC_STATS_BH(sock_net(sk),
1419						 UDP_MIB_INDATAGRAMS,
1420						 is_udplite);
1421				return -ret;
1422			}
1423		}
1424
1425		/* FALLTHROUGH -- it's a UDP Packet */
1426	}
1427
1428	/*
1429	 * 	UDP-Lite specific tests, ignored on UDP sockets
1430	 */
1431	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
1432
1433		/*
1434		 * MIB statistics other than incrementing the error count are
1435		 * disabled for the following two types of errors: these depend
1436		 * on the application settings, not on the functioning of the
1437		 * protocol stack as such.
1438		 *
1439		 * RFC 3828 here recommends (sec 3.3): "There should also be a
1440		 * way ... to ... at least let the receiving application block
1441		 * delivery of packets with coverage values less than a value
1442		 * provided by the application."
1443		 */
1444		if (up->pcrlen == 0) {          /* full coverage was set  */
1445			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1446				"%d while full coverage %d requested\n",
1447				UDP_SKB_CB(skb)->cscov, skb->len);
1448			goto drop;
1449		}
1450		/* The next case involves violating the min. coverage requested
1451		 * by the receiver. This is subtle: if receiver wants x and x is
1452		 * greater than the buffersize/MTU then receiver will complain
1453		 * that it wants x while sender emits packets of smaller size y.
1454		 * Therefore the above ...()->partial_cov statement is essential.
1455		 */
1456		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
1457			LIMIT_NETDEBUG(KERN_WARNING
1458				"UDPLITE: coverage %d too small, need min %d\n",
1459				UDP_SKB_CB(skb)->cscov, up->pcrlen);
1460			goto drop;
1461		}
1462	}
1463
1464	if (rcu_dereference_raw(sk->sk_filter)) {
1465		if (udp_lib_checksum_complete(skb))
1466			goto drop;
1467	}
1468
1469
1470	if (sk_rcvqueues_full(sk, skb))
1471		goto drop;
1472
1473	rc = 0;
1474
1475	bh_lock_sock(sk);
1476	if (!sock_owned_by_user(sk))
1477		rc = __udp_queue_rcv_skb(sk, skb);
1478	else if (sk_add_backlog(sk, skb)) {
1479		bh_unlock_sock(sk);
1480		goto drop;
1481	}
1482	bh_unlock_sock(sk);
1483
1484	return rc;
1485
 
 
1486drop:
1487	UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1488	atomic_inc(&sk->sk_drops);
1489	kfree_skb(skb);
1490	return -1;
1491}
1492
 
 
 
 
1493
1494static void flush_stack(struct sock **stack, unsigned int count,
1495			struct sk_buff *skb, unsigned int final)
1496{
1497	unsigned int i;
1498	struct sk_buff *skb1 = NULL;
1499	struct sock *sk;
1500
1501	for (i = 0; i < count; i++) {
1502		sk = stack[i];
1503		if (likely(skb1 == NULL))
1504			skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
 
 
 
 
 
 
 
 
1505
1506		if (!skb1) {
1507			atomic_inc(&sk->sk_drops);
1508			UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1509					 IS_UDPLITE(sk));
1510			UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1511					 IS_UDPLITE(sk));
1512		}
1513
1514		if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1515			skb1 = NULL;
 
 
1516	}
1517	if (unlikely(skb1))
1518		kfree_skb(skb1);
1519}
 
1520
1521/*
1522 *	Multicasts and broadcasts go to each listener.
1523 *
1524 *	Note: called only from the BH handler context.
1525 */
1526static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1527				    struct udphdr  *uh,
1528				    __be32 saddr, __be32 daddr,
1529				    struct udp_table *udptable)
 
1530{
1531	struct sock *sk, *stack[256 / sizeof(struct sock *)];
1532	struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1533	int dif;
1534	unsigned int i, count = 0;
1535
1536	spin_lock(&hslot->lock);
1537	sk = sk_nulls_head(&hslot->head);
1538	dif = skb->dev->ifindex;
1539	sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1540	while (sk) {
1541		stack[count++] = sk;
1542		sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1543				       daddr, uh->source, saddr, dif);
1544		if (unlikely(count == ARRAY_SIZE(stack))) {
1545			if (!sk)
1546				break;
1547			flush_stack(stack, count, skb, ~0);
1548			count = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1549		}
 
 
1550	}
1551	/*
1552	 * before releasing chain lock, we must take a reference on sockets
1553	 */
1554	for (i = 0; i < count; i++)
1555		sock_hold(stack[i]);
1556
1557	spin_unlock(&hslot->lock);
 
 
 
 
1558
1559	/*
1560	 * do the slow work with no lock held
1561	 */
1562	if (count) {
1563		flush_stack(stack, count, skb, count - 1);
1564
1565		for (i = 0; i < count; i++)
1566			sock_put(stack[i]);
1567	} else {
1568		kfree_skb(skb);
 
 
1569	}
1570	return 0;
1571}
1572
1573/* Initialize UDP checksum. If exited with zero value (success),
1574 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1575 * Otherwise, csum completion requires chacksumming packet body,
1576 * including udp header and folding it to skb->csum.
1577 */
1578static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1579				 int proto)
1580{
1581	const struct iphdr *iph;
1582	int err;
1583
1584	UDP_SKB_CB(skb)->partial_cov = 0;
1585	UDP_SKB_CB(skb)->cscov = skb->len;
1586
1587	if (proto == IPPROTO_UDPLITE) {
1588		err = udplite_checksum_init(skb, uh);
1589		if (err)
1590			return err;
 
 
 
 
 
1591	}
1592
1593	iph = ip_hdr(skb);
1594	if (uh->check == 0) {
1595		skb->ip_summed = CHECKSUM_UNNECESSARY;
1596	} else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1597		if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1598				      proto, skb->csum))
1599			skb->ip_summed = CHECKSUM_UNNECESSARY;
1600	}
1601	if (!skb_csum_unnecessary(skb))
1602		skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1603					       skb->len, proto, 0);
1604	/* Probably, we should checksum udp header (it should be in cache
1605	 * in any case) and data in tiny packets (< rx copybreak).
1606	 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1607
 
 
 
 
 
1608	return 0;
1609}
1610
1611/*
1612 *	All we need to do is get the socket, and then do a checksum.
1613 */
1614
1615int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1616		   int proto)
1617{
1618	struct sock *sk;
1619	struct udphdr *uh;
1620	unsigned short ulen;
1621	struct rtable *rt = skb_rtable(skb);
1622	__be32 saddr, daddr;
1623	struct net *net = dev_net(skb->dev);
1624
1625	/*
1626	 *  Validate the packet.
1627	 */
1628	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1629		goto drop;		/* No space for header. */
1630
1631	uh   = udp_hdr(skb);
1632	ulen = ntohs(uh->len);
1633	saddr = ip_hdr(skb)->saddr;
1634	daddr = ip_hdr(skb)->daddr;
1635
1636	if (ulen > skb->len)
1637		goto short_packet;
1638
1639	if (proto == IPPROTO_UDP) {
1640		/* UDP validates ulen. */
1641		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1642			goto short_packet;
1643		uh = udp_hdr(skb);
1644	}
1645
1646	if (udp4_csum_init(skb, uh, proto))
1647		goto csum_error;
1648
 
 
 
 
 
 
 
 
 
 
 
 
 
1649	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1650		return __udp4_lib_mcast_deliver(net, skb, uh,
1651				saddr, daddr, udptable);
1652
1653	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1654
1655	if (sk != NULL) {
1656		int ret = udp_queue_rcv_skb(sk, skb);
1657		sock_put(sk);
1658
1659		/* a return value > 0 means to resubmit the input, but
1660		 * it wants the return to be -protocol, or 0
1661		 */
1662		if (ret > 0)
1663			return -ret;
1664		return 0;
1665	}
1666
1667	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1668		goto drop;
1669	nf_reset(skb);
1670
1671	/* No socket. Drop packet silently, if checksum is wrong */
1672	if (udp_lib_checksum_complete(skb))
1673		goto csum_error;
1674
1675	UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1676	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1677
1678	/*
1679	 * Hmm.  We got an UDP packet to a port to which we
1680	 * don't wanna listen.  Ignore it.
1681	 */
1682	kfree_skb(skb);
1683	return 0;
1684
1685short_packet:
1686	LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1687		       proto == IPPROTO_UDPLITE ? "-Lite" : "",
1688		       &saddr,
1689		       ntohs(uh->source),
1690		       ulen,
1691		       skb->len,
1692		       &daddr,
1693		       ntohs(uh->dest));
1694	goto drop;
1695
1696csum_error:
1697	/*
1698	 * RFC1122: OK.  Discards the bad packet silently (as far as
1699	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1700	 */
1701	LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1702		       proto == IPPROTO_UDPLITE ? "-Lite" : "",
1703		       &saddr,
1704		       ntohs(uh->source),
1705		       &daddr,
1706		       ntohs(uh->dest),
1707		       ulen);
1708drop:
1709	UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1710	kfree_skb(skb);
1711	return 0;
1712}
1713
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1714int udp_rcv(struct sk_buff *skb)
1715{
1716	return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1717}
1718
1719void udp_destroy_sock(struct sock *sk)
1720{
 
1721	bool slow = lock_sock_fast(sk);
1722	udp_flush_pending_frames(sk);
1723	unlock_sock_fast(sk, slow);
 
 
 
 
 
 
 
 
 
 
1724}
1725
1726/*
1727 *	Socket option code for UDP
1728 */
1729int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1730		       char __user *optval, unsigned int optlen,
1731		       int (*push_pending_frames)(struct sock *))
1732{
1733	struct udp_sock *up = udp_sk(sk);
1734	int val;
1735	int err = 0;
1736	int is_udplite = IS_UDPLITE(sk);
1737
1738	if (optlen < sizeof(int))
1739		return -EINVAL;
1740
1741	if (get_user(val, (int __user *)optval))
1742		return -EFAULT;
1743
 
 
1744	switch (optname) {
1745	case UDP_CORK:
1746		if (val != 0) {
1747			up->corkflag = 1;
1748		} else {
1749			up->corkflag = 0;
1750			lock_sock(sk);
1751			(*push_pending_frames)(sk);
1752			release_sock(sk);
1753		}
1754		break;
1755
1756	case UDP_ENCAP:
1757		switch (val) {
1758		case 0:
1759		case UDP_ENCAP_ESPINUDP:
1760		case UDP_ENCAP_ESPINUDP_NON_IKE:
1761			up->encap_rcv = xfrm4_udp_encap_rcv;
1762			/* FALLTHROUGH */
1763		case UDP_ENCAP_L2TPINUDP:
1764			up->encap_type = val;
 
 
 
1765			break;
1766		default:
1767			err = -ENOPROTOOPT;
1768			break;
1769		}
1770		break;
1771
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1772	/*
1773	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
1774	 */
1775	/* The sender sets actual checksum coverage length via this option.
1776	 * The case coverage > packet length is handled by send module. */
1777	case UDPLITE_SEND_CSCOV:
1778		if (!is_udplite)         /* Disable the option on UDP sockets */
1779			return -ENOPROTOOPT;
1780		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1781			val = 8;
1782		else if (val > USHRT_MAX)
1783			val = USHRT_MAX;
1784		up->pcslen = val;
1785		up->pcflag |= UDPLITE_SEND_CC;
1786		break;
1787
1788	/* The receiver specifies a minimum checksum coverage value. To make
1789	 * sense, this should be set to at least 8 (as done below). If zero is
1790	 * used, this again means full checksum coverage.                     */
1791	case UDPLITE_RECV_CSCOV:
1792		if (!is_udplite)         /* Disable the option on UDP sockets */
1793			return -ENOPROTOOPT;
1794		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
1795			val = 8;
1796		else if (val > USHRT_MAX)
1797			val = USHRT_MAX;
1798		up->pcrlen = val;
1799		up->pcflag |= UDPLITE_RECV_CC;
1800		break;
1801
1802	default:
1803		err = -ENOPROTOOPT;
1804		break;
1805	}
1806
1807	return err;
1808}
1809EXPORT_SYMBOL(udp_lib_setsockopt);
1810
1811int udp_setsockopt(struct sock *sk, int level, int optname,
1812		   char __user *optval, unsigned int optlen)
1813{
1814	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1815		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1816					  udp_push_pending_frames);
1817	return ip_setsockopt(sk, level, optname, optval, optlen);
1818}
1819
1820#ifdef CONFIG_COMPAT
1821int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1822			  char __user *optval, unsigned int optlen)
1823{
1824	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1825		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1826					  udp_push_pending_frames);
1827	return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1828}
1829#endif
1830
1831int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1832		       char __user *optval, int __user *optlen)
1833{
1834	struct udp_sock *up = udp_sk(sk);
1835	int val, len;
1836
1837	if (get_user(len, optlen))
1838		return -EFAULT;
1839
1840	len = min_t(unsigned int, len, sizeof(int));
1841
1842	if (len < 0)
1843		return -EINVAL;
1844
1845	switch (optname) {
1846	case UDP_CORK:
1847		val = up->corkflag;
1848		break;
1849
1850	case UDP_ENCAP:
1851		val = up->encap_type;
1852		break;
1853
 
 
 
 
 
 
 
 
 
 
 
 
1854	/* The following two cannot be changed on UDP sockets, the return is
1855	 * always 0 (which corresponds to the full checksum coverage of UDP). */
1856	case UDPLITE_SEND_CSCOV:
1857		val = up->pcslen;
1858		break;
1859
1860	case UDPLITE_RECV_CSCOV:
1861		val = up->pcrlen;
1862		break;
1863
1864	default:
1865		return -ENOPROTOOPT;
1866	}
1867
1868	if (put_user(len, optlen))
1869		return -EFAULT;
1870	if (copy_to_user(optval, &val, len))
1871		return -EFAULT;
1872	return 0;
1873}
1874EXPORT_SYMBOL(udp_lib_getsockopt);
1875
1876int udp_getsockopt(struct sock *sk, int level, int optname,
1877		   char __user *optval, int __user *optlen)
1878{
1879	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1880		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1881	return ip_getsockopt(sk, level, optname, optval, optlen);
1882}
1883
1884#ifdef CONFIG_COMPAT
1885int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1886				 char __user *optval, int __user *optlen)
1887{
1888	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1889		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1890	return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1891}
1892#endif
1893/**
1894 * 	udp_poll - wait for a UDP event.
1895 *	@file - file struct
1896 *	@sock - socket
1897 *	@wait - poll table
1898 *
1899 *	This is same as datagram poll, except for the special case of
1900 *	blocking sockets. If application is using a blocking fd
1901 *	and a packet with checksum error is in the queue;
1902 *	then it could get return from select indicating data available
1903 *	but then block when reading it. Add special case code
1904 *	to work around these arguably broken applications.
1905 */
1906unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1907{
1908	unsigned int mask = datagram_poll(file, sock, wait);
1909	struct sock *sk = sock->sk;
1910
 
 
 
1911	/* Check for false positives due to checksum errors */
1912	if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1913	    !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1914		mask &= ~(POLLIN | POLLRDNORM);
1915
1916	return mask;
1917
1918}
1919EXPORT_SYMBOL(udp_poll);
1920
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1921struct proto udp_prot = {
1922	.name		   = "UDP",
1923	.owner		   = THIS_MODULE,
1924	.close		   = udp_lib_close,
1925	.connect	   = ip4_datagram_connect,
1926	.disconnect	   = udp_disconnect,
1927	.ioctl		   = udp_ioctl,
1928	.destroy	   = udp_destroy_sock,
1929	.setsockopt	   = udp_setsockopt,
1930	.getsockopt	   = udp_getsockopt,
1931	.sendmsg	   = udp_sendmsg,
1932	.recvmsg	   = udp_recvmsg,
1933	.sendpage	   = udp_sendpage,
1934	.backlog_rcv	   = __udp_queue_rcv_skb,
1935	.hash		   = udp_lib_hash,
1936	.unhash		   = udp_lib_unhash,
1937	.rehash		   = udp_v4_rehash,
1938	.get_port	   = udp_v4_get_port,
1939	.memory_allocated  = &udp_memory_allocated,
1940	.sysctl_mem	   = sysctl_udp_mem,
1941	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1942	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1943	.obj_size	   = sizeof(struct udp_sock),
1944	.slab_flags	   = SLAB_DESTROY_BY_RCU,
1945	.h.udp_table	   = &udp_table,
 
1946#ifdef CONFIG_COMPAT
1947	.compat_setsockopt = compat_udp_setsockopt,
1948	.compat_getsockopt = compat_udp_getsockopt,
1949#endif
1950	.clear_sk	   = sk_prot_clear_portaddr_nulls,
1951};
1952EXPORT_SYMBOL(udp_prot);
1953
1954/* ------------------------------------------------------------------------ */
1955#ifdef CONFIG_PROC_FS
1956
1957static struct sock *udp_get_first(struct seq_file *seq, int start)
1958{
1959	struct sock *sk;
 
1960	struct udp_iter_state *state = seq->private;
1961	struct net *net = seq_file_net(seq);
1962
1963	for (state->bucket = start; state->bucket <= state->udp_table->mask;
1964	     ++state->bucket) {
1965		struct hlist_nulls_node *node;
1966		struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1967
1968		if (hlist_nulls_empty(&hslot->head))
1969			continue;
1970
1971		spin_lock_bh(&hslot->lock);
1972		sk_nulls_for_each(sk, node, &hslot->head) {
1973			if (!net_eq(sock_net(sk), net))
1974				continue;
1975			if (sk->sk_family == state->family)
1976				goto found;
1977		}
1978		spin_unlock_bh(&hslot->lock);
1979	}
1980	sk = NULL;
1981found:
1982	return sk;
1983}
1984
1985static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1986{
 
1987	struct udp_iter_state *state = seq->private;
1988	struct net *net = seq_file_net(seq);
1989
1990	do {
1991		sk = sk_nulls_next(sk);
1992	} while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1993
1994	if (!sk) {
1995		if (state->bucket <= state->udp_table->mask)
1996			spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1997		return udp_get_first(seq, state->bucket + 1);
1998	}
1999	return sk;
2000}
2001
2002static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2003{
2004	struct sock *sk = udp_get_first(seq, 0);
2005
2006	if (sk)
2007		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2008			--pos;
2009	return pos ? NULL : sk;
2010}
2011
2012static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2013{
2014	struct udp_iter_state *state = seq->private;
2015	state->bucket = MAX_UDP_PORTS;
2016
2017	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2018}
 
2019
2020static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2021{
2022	struct sock *sk;
2023
2024	if (v == SEQ_START_TOKEN)
2025		sk = udp_get_idx(seq, 0);
2026	else
2027		sk = udp_get_next(seq, v);
2028
2029	++*pos;
2030	return sk;
2031}
 
2032
2033static void udp_seq_stop(struct seq_file *seq, void *v)
2034{
 
2035	struct udp_iter_state *state = seq->private;
2036
2037	if (state->bucket <= state->udp_table->mask)
2038		spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2039}
2040
2041static int udp_seq_open(struct inode *inode, struct file *file)
2042{
2043	struct udp_seq_afinfo *afinfo = PDE(inode)->data;
2044	struct udp_iter_state *s;
2045	int err;
2046
2047	err = seq_open_net(inode, file, &afinfo->seq_ops,
2048			   sizeof(struct udp_iter_state));
2049	if (err < 0)
2050		return err;
2051
2052	s = ((struct seq_file *)file->private_data)->private;
2053	s->family		= afinfo->family;
2054	s->udp_table		= afinfo->udp_table;
2055	return err;
2056}
2057
2058/* ------------------------------------------------------------------------ */
2059int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2060{
2061	struct proc_dir_entry *p;
2062	int rc = 0;
2063
2064	afinfo->seq_fops.open		= udp_seq_open;
2065	afinfo->seq_fops.read		= seq_read;
2066	afinfo->seq_fops.llseek		= seq_lseek;
2067	afinfo->seq_fops.release	= seq_release_net;
2068
2069	afinfo->seq_ops.start		= udp_seq_start;
2070	afinfo->seq_ops.next		= udp_seq_next;
2071	afinfo->seq_ops.stop		= udp_seq_stop;
2072
2073	p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2074			     &afinfo->seq_fops, afinfo);
2075	if (!p)
2076		rc = -ENOMEM;
2077	return rc;
2078}
2079EXPORT_SYMBOL(udp_proc_register);
2080
2081void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2082{
2083	proc_net_remove(net, afinfo->name);
2084}
2085EXPORT_SYMBOL(udp_proc_unregister);
2086
2087/* ------------------------------------------------------------------------ */
2088static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2089		int bucket, int *len)
2090{
2091	struct inet_sock *inet = inet_sk(sp);
2092	__be32 dest = inet->inet_daddr;
2093	__be32 src  = inet->inet_rcv_saddr;
2094	__u16 destp	  = ntohs(inet->inet_dport);
2095	__u16 srcp	  = ntohs(inet->inet_sport);
2096
2097	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2098		" %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n",
2099		bucket, src, srcp, dest, destp, sp->sk_state,
2100		sk_wmem_alloc_get(sp),
2101		sk_rmem_alloc_get(sp),
2102		0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
2103		atomic_read(&sp->sk_refcnt), sp,
2104		atomic_read(&sp->sk_drops), len);
 
 
2105}
2106
2107int udp4_seq_show(struct seq_file *seq, void *v)
2108{
 
2109	if (v == SEQ_START_TOKEN)
2110		seq_printf(seq, "%-127s\n",
2111			   "  sl  local_address rem_address   st tx_queue "
2112			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2113			   "inode ref pointer drops");
2114	else {
2115		struct udp_iter_state *state = seq->private;
2116		int len;
2117
2118		udp4_format_sock(v, seq, state->bucket, &len);
2119		seq_printf(seq, "%*s\n", 127 - len, "");
2120	}
 
2121	return 0;
2122}
2123
2124/* ------------------------------------------------------------------------ */
 
 
 
 
 
 
 
2125static struct udp_seq_afinfo udp4_seq_afinfo = {
2126	.name		= "udp",
2127	.family		= AF_INET,
2128	.udp_table	= &udp_table,
2129	.seq_fops	= {
2130		.owner	=	THIS_MODULE,
2131	},
2132	.seq_ops	= {
2133		.show		= udp4_seq_show,
2134	},
2135};
2136
2137static int __net_init udp4_proc_init_net(struct net *net)
2138{
2139	return udp_proc_register(net, &udp4_seq_afinfo);
 
 
 
2140}
2141
2142static void __net_exit udp4_proc_exit_net(struct net *net)
2143{
2144	udp_proc_unregister(net, &udp4_seq_afinfo);
2145}
2146
2147static struct pernet_operations udp4_net_ops = {
2148	.init = udp4_proc_init_net,
2149	.exit = udp4_proc_exit_net,
2150};
2151
2152int __init udp4_proc_init(void)
2153{
2154	return register_pernet_subsys(&udp4_net_ops);
2155}
2156
2157void udp4_proc_exit(void)
2158{
2159	unregister_pernet_subsys(&udp4_net_ops);
2160}
2161#endif /* CONFIG_PROC_FS */
2162
2163static __initdata unsigned long uhash_entries;
2164static int __init set_uhash_entries(char *str)
2165{
 
 
2166	if (!str)
2167		return 0;
2168	uhash_entries = simple_strtoul(str, &str, 0);
 
 
 
 
2169	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2170		uhash_entries = UDP_HTABLE_SIZE_MIN;
2171	return 1;
2172}
2173__setup("uhash_entries=", set_uhash_entries);
2174
2175void __init udp_table_init(struct udp_table *table, const char *name)
2176{
2177	unsigned int i;
2178
2179	if (!CONFIG_BASE_SMALL)
2180		table->hash = alloc_large_system_hash(name,
2181			2 * sizeof(struct udp_hslot),
2182			uhash_entries,
2183			21, /* one slot per 2 MB */
2184			0,
2185			&table->log,
2186			&table->mask,
2187			64 * 1024);
2188	/*
2189	 * Make sure hash table has the minimum size
2190	 */
2191	if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
2192		table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
2193				      2 * sizeof(struct udp_hslot), GFP_KERNEL);
2194		if (!table->hash)
2195			panic(name);
2196		table->log = ilog2(UDP_HTABLE_SIZE_MIN);
2197		table->mask = UDP_HTABLE_SIZE_MIN - 1;
2198	}
2199	table->hash2 = table->hash + (table->mask + 1);
2200	for (i = 0; i <= table->mask; i++) {
2201		INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2202		table->hash[i].count = 0;
2203		spin_lock_init(&table->hash[i].lock);
2204	}
2205	for (i = 0; i <= table->mask; i++) {
2206		INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2207		table->hash2[i].count = 0;
2208		spin_lock_init(&table->hash2[i].lock);
2209	}
2210}
2211
2212void __init udp_init(void)
2213{
2214	unsigned long limit;
2215
2216	udp_table_init(&udp_table, "UDP");
2217	limit = nr_free_buffer_pages() / 8;
2218	limit = max(limit, 128UL);
2219	sysctl_udp_mem[0] = limit / 4 * 3;
2220	sysctl_udp_mem[1] = limit;
2221	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2222
2223	sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2224	sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2225}
 
2226
2227int udp4_ufo_send_check(struct sk_buff *skb)
2228{
2229	const struct iphdr *iph;
2230	struct udphdr *uh;
2231
2232	if (!pskb_may_pull(skb, sizeof(*uh)))
2233		return -EINVAL;
 
 
2234
2235	iph = ip_hdr(skb);
2236	uh = udp_hdr(skb);
2237
2238	uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2239				       IPPROTO_UDP, 0);
2240	skb->csum_start = skb_transport_header(skb) - skb->head;
2241	skb->csum_offset = offsetof(struct udphdr, check);
2242	skb->ip_summed = CHECKSUM_PARTIAL;
2243	return 0;
2244}
2245
2246struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, u32 features)
 
 
 
 
2247{
2248	struct sk_buff *segs = ERR_PTR(-EINVAL);
2249	unsigned int mss;
2250	int offset;
2251	__wsum csum;
2252
2253	mss = skb_shinfo(skb)->gso_size;
2254	if (unlikely(skb->len <= mss))
2255		goto out;
 
 
 
2256
2257	if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2258		/* Packet is from an untrusted source, reset gso_segs. */
2259		int type = skb_shinfo(skb)->gso_type;
2260
2261		if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2262			     !(type & (SKB_GSO_UDP))))
2263			goto out;
 
 
 
 
 
2264
2265		skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2266
2267		segs = NULL;
2268		goto out;
2269	}
2270
2271	/* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2272	 * do checksum of UDP packets sent as multiple IP fragments.
2273	 */
2274	offset = skb_checksum_start_offset(skb);
2275	csum = skb_checksum(skb, offset, skb->len - offset, 0);
2276	offset += skb->csum_offset;
2277	*(__sum16 *)(skb->data + offset) = csum_fold(csum);
2278	skb->ip_summed = CHECKSUM_NONE;
2279
2280	/* Fragment the skb. IP headers of the fragments are updated in
2281	 * inet_gso_segment()
2282	 */
2283	segs = skb_segment(skb, features);
2284out:
2285	return segs;
2286}
2287