Linux Audio

Check our new training course

Loading...
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		The User Datagram Protocol (UDP).
   8 *
   9 * Authors:	Ross Biro
  10 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  12 *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
  13 *		Hirokazu Takahashi, <taka@valinux.co.jp>
  14 *
  15 * Fixes:
  16 *		Alan Cox	:	verify_area() calls
  17 *		Alan Cox	: 	stopped close while in use off icmp
  18 *					messages. Not a fix but a botch that
  19 *					for udp at least is 'valid'.
  20 *		Alan Cox	:	Fixed icmp handling properly
  21 *		Alan Cox	: 	Correct error for oversized datagrams
  22 *		Alan Cox	:	Tidied select() semantics.
  23 *		Alan Cox	:	udp_err() fixed properly, also now
  24 *					select and read wake correctly on errors
  25 *		Alan Cox	:	udp_send verify_area moved to avoid mem leak
  26 *		Alan Cox	:	UDP can count its memory
  27 *		Alan Cox	:	send to an unknown connection causes
  28 *					an ECONNREFUSED off the icmp, but
  29 *					does NOT close.
  30 *		Alan Cox	:	Switched to new sk_buff handlers. No more backlog!
  31 *		Alan Cox	:	Using generic datagram code. Even smaller and the PEEK
  32 *					bug no longer crashes it.
  33 *		Fred Van Kempen	: 	Net2e support for sk->broadcast.
  34 *		Alan Cox	:	Uses skb_free_datagram
  35 *		Alan Cox	:	Added get/set sockopt support.
  36 *		Alan Cox	:	Broadcasting without option set returns EACCES.
  37 *		Alan Cox	:	No wakeup calls. Instead we now use the callbacks.
  38 *		Alan Cox	:	Use ip_tos and ip_ttl
  39 *		Alan Cox	:	SNMP Mibs
  40 *		Alan Cox	:	MSG_DONTROUTE, and 0.0.0.0 support.
  41 *		Matt Dillon	:	UDP length checks.
  42 *		Alan Cox	:	Smarter af_inet used properly.
  43 *		Alan Cox	:	Use new kernel side addressing.
  44 *		Alan Cox	:	Incorrect return on truncated datagram receive.
  45 *	Arnt Gulbrandsen 	:	New udp_send and stuff
  46 *		Alan Cox	:	Cache last socket
  47 *		Alan Cox	:	Route cache
  48 *		Jon Peatfield	:	Minor efficiency fix to sendto().
  49 *		Mike Shaver	:	RFC1122 checks.
  50 *		Alan Cox	:	Nonblocking error fix.
  51 *	Willy Konynenberg	:	Transparent proxying support.
  52 *		Mike McLagan	:	Routing by source
  53 *		David S. Miller	:	New socket lookup architecture.
  54 *					Last socket cache retained as it
  55 *					does have a high hit rate.
  56 *		Olaf Kirch	:	Don't linearise iovec on sendmsg.
  57 *		Andi Kleen	:	Some cleanups, cache destination entry
  58 *					for connect.
  59 *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
  60 *		Melvin Smith	:	Check msg_name not msg_namelen in sendto(),
  61 *					return ENOTCONN for unconnected sockets (POSIX)
  62 *		Janos Farkas	:	don't deliver multi/broadcasts to a different
  63 *					bound-to-device socket
  64 *	Hirokazu Takahashi	:	HW checksumming for outgoing UDP
  65 *					datagrams.
  66 *	Hirokazu Takahashi	:	sendfile() on UDP works now.
  67 *		Arnaldo C. Melo :	convert /proc/net/udp to seq_file
  68 *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
  69 *	Alexey Kuznetsov:		allow both IPv4 and IPv6 sockets to bind
  70 *					a single port at the same time.
  71 *	Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
  72 *	James Chapman		:	Add L2TP encapsulation type.
  73 */
  74
  75#define pr_fmt(fmt) "UDP: " fmt
  76
  77#include <linux/bpf-cgroup.h>
  78#include <linux/uaccess.h>
  79#include <asm/ioctls.h>
  80#include <linux/memblock.h>
  81#include <linux/highmem.h>
  82#include <linux/types.h>
  83#include <linux/fcntl.h>
  84#include <linux/module.h>
  85#include <linux/socket.h>
  86#include <linux/sockios.h>
  87#include <linux/igmp.h>
  88#include <linux/inetdevice.h>
  89#include <linux/in.h>
  90#include <linux/errno.h>
  91#include <linux/timer.h>
  92#include <linux/mm.h>
  93#include <linux/inet.h>
  94#include <linux/netdevice.h>
  95#include <linux/slab.h>
  96#include <net/tcp_states.h>
  97#include <linux/skbuff.h>
  98#include <linux/proc_fs.h>
  99#include <linux/seq_file.h>
 100#include <net/net_namespace.h>
 101#include <net/icmp.h>
 102#include <net/inet_hashtables.h>
 103#include <net/ip.h>
 104#include <net/ip_tunnels.h>
 105#include <net/route.h>
 106#include <net/checksum.h>
 107#include <net/gso.h>
 108#include <net/xfrm.h>
 109#include <trace/events/udp.h>
 110#include <linux/static_key.h>
 111#include <linux/btf_ids.h>
 112#include <trace/events/skb.h>
 113#include <net/busy_poll.h>
 114#include "udp_impl.h"
 115#include <net/sock_reuseport.h>
 116#include <net/addrconf.h>
 117#include <net/udp_tunnel.h>
 118#include <net/gro.h>
 119#if IS_ENABLED(CONFIG_IPV6)
 120#include <net/ipv6_stubs.h>
 121#endif
 122
 123struct udp_table udp_table __read_mostly;
 124EXPORT_SYMBOL(udp_table);
 125
 126long sysctl_udp_mem[3] __read_mostly;
 127EXPORT_SYMBOL(sysctl_udp_mem);
 128
 129atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
 130EXPORT_SYMBOL(udp_memory_allocated);
 131DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
 132EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
 133
 134#define MAX_UDP_PORTS 65536
 135#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET)
 136
 137static struct udp_table *udp_get_table_prot(struct sock *sk)
 138{
 139	return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
 140}
 141
 142static int udp_lib_lport_inuse(struct net *net, __u16 num,
 143			       const struct udp_hslot *hslot,
 144			       unsigned long *bitmap,
 145			       struct sock *sk, unsigned int log)
 146{
 147	struct sock *sk2;
 148	kuid_t uid = sock_i_uid(sk);
 149
 150	sk_for_each(sk2, &hslot->head) {
 151		if (net_eq(sock_net(sk2), net) &&
 152		    sk2 != sk &&
 153		    (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
 154		    (!sk2->sk_reuse || !sk->sk_reuse) &&
 155		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 156		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 157		    inet_rcv_saddr_equal(sk, sk2, true)) {
 158			if (sk2->sk_reuseport && sk->sk_reuseport &&
 159			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 160			    uid_eq(uid, sock_i_uid(sk2))) {
 161				if (!bitmap)
 162					return 0;
 163			} else {
 164				if (!bitmap)
 165					return 1;
 166				__set_bit(udp_sk(sk2)->udp_port_hash >> log,
 167					  bitmap);
 168			}
 169		}
 170	}
 171	return 0;
 172}
 173
 174/*
 175 * Note: we still hold spinlock of primary hash chain, so no other writer
 176 * can insert/delete a socket with local_port == num
 177 */
 178static int udp_lib_lport_inuse2(struct net *net, __u16 num,
 179				struct udp_hslot *hslot2,
 180				struct sock *sk)
 181{
 182	struct sock *sk2;
 183	kuid_t uid = sock_i_uid(sk);
 184	int res = 0;
 185
 186	spin_lock(&hslot2->lock);
 187	udp_portaddr_for_each_entry(sk2, &hslot2->head) {
 188		if (net_eq(sock_net(sk2), net) &&
 189		    sk2 != sk &&
 190		    (udp_sk(sk2)->udp_port_hash == num) &&
 191		    (!sk2->sk_reuse || !sk->sk_reuse) &&
 192		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 193		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 194		    inet_rcv_saddr_equal(sk, sk2, true)) {
 195			if (sk2->sk_reuseport && sk->sk_reuseport &&
 196			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 197			    uid_eq(uid, sock_i_uid(sk2))) {
 198				res = 0;
 199			} else {
 200				res = 1;
 201			}
 202			break;
 203		}
 204	}
 205	spin_unlock(&hslot2->lock);
 206	return res;
 207}
 208
 209static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
 210{
 211	struct net *net = sock_net(sk);
 212	kuid_t uid = sock_i_uid(sk);
 213	struct sock *sk2;
 214
 215	sk_for_each(sk2, &hslot->head) {
 216		if (net_eq(sock_net(sk2), net) &&
 217		    sk2 != sk &&
 218		    sk2->sk_family == sk->sk_family &&
 219		    ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
 220		    (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
 221		    (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 222		    sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
 223		    inet_rcv_saddr_equal(sk, sk2, false)) {
 224			return reuseport_add_sock(sk, sk2,
 225						  inet_rcv_saddr_any(sk));
 226		}
 227	}
 228
 229	return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
 230}
 231
 232/**
 233 *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
 234 *
 235 *  @sk:          socket struct in question
 236 *  @snum:        port number to look up
 237 *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
 238 *                   with NULL address
 239 */
 240int udp_lib_get_port(struct sock *sk, unsigned short snum,
 241		     unsigned int hash2_nulladdr)
 242{
 243	struct udp_table *udptable = udp_get_table_prot(sk);
 244	struct udp_hslot *hslot, *hslot2;
 245	struct net *net = sock_net(sk);
 246	int error = -EADDRINUSE;
 247
 248	if (!snum) {
 249		DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
 250		unsigned short first, last;
 251		int low, high, remaining;
 252		unsigned int rand;
 253
 254		inet_sk_get_local_port_range(sk, &low, &high);
 255		remaining = (high - low) + 1;
 256
 257		rand = get_random_u32();
 258		first = reciprocal_scale(rand, remaining) + low;
 259		/*
 260		 * force rand to be an odd multiple of UDP_HTABLE_SIZE
 261		 */
 262		rand = (rand | 1) * (udptable->mask + 1);
 263		last = first + udptable->mask + 1;
 264		do {
 265			hslot = udp_hashslot(udptable, net, first);
 266			bitmap_zero(bitmap, PORTS_PER_CHAIN);
 267			spin_lock_bh(&hslot->lock);
 268			udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
 269					    udptable->log);
 270
 271			snum = first;
 272			/*
 273			 * Iterate on all possible values of snum for this hash.
 274			 * Using steps of an odd multiple of UDP_HTABLE_SIZE
 275			 * give us randomization and full range coverage.
 276			 */
 277			do {
 278				if (low <= snum && snum <= high &&
 279				    !test_bit(snum >> udptable->log, bitmap) &&
 280				    !inet_is_local_reserved_port(net, snum))
 281					goto found;
 282				snum += rand;
 283			} while (snum != first);
 284			spin_unlock_bh(&hslot->lock);
 285			cond_resched();
 286		} while (++first != last);
 287		goto fail;
 288	} else {
 289		hslot = udp_hashslot(udptable, net, snum);
 290		spin_lock_bh(&hslot->lock);
 291		if (hslot->count > 10) {
 292			int exist;
 293			unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
 294
 295			slot2          &= udptable->mask;
 296			hash2_nulladdr &= udptable->mask;
 297
 298			hslot2 = udp_hashslot2(udptable, slot2);
 299			if (hslot->count < hslot2->count)
 300				goto scan_primary_hash;
 301
 302			exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
 303			if (!exist && (hash2_nulladdr != slot2)) {
 304				hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
 305				exist = udp_lib_lport_inuse2(net, snum, hslot2,
 306							     sk);
 307			}
 308			if (exist)
 309				goto fail_unlock;
 310			else
 311				goto found;
 312		}
 313scan_primary_hash:
 314		if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
 315			goto fail_unlock;
 316	}
 317found:
 318	inet_sk(sk)->inet_num = snum;
 319	udp_sk(sk)->udp_port_hash = snum;
 320	udp_sk(sk)->udp_portaddr_hash ^= snum;
 321	if (sk_unhashed(sk)) {
 322		if (sk->sk_reuseport &&
 323		    udp_reuseport_add_sock(sk, hslot)) {
 324			inet_sk(sk)->inet_num = 0;
 325			udp_sk(sk)->udp_port_hash = 0;
 326			udp_sk(sk)->udp_portaddr_hash ^= snum;
 327			goto fail_unlock;
 328		}
 329
 330		sock_set_flag(sk, SOCK_RCU_FREE);
 331
 332		sk_add_node_rcu(sk, &hslot->head);
 333		hslot->count++;
 334		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 335
 336		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 337		spin_lock(&hslot2->lock);
 338		if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 339		    sk->sk_family == AF_INET6)
 340			hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
 341					   &hslot2->head);
 342		else
 343			hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
 344					   &hslot2->head);
 345		hslot2->count++;
 346		spin_unlock(&hslot2->lock);
 347	}
 348
 349	error = 0;
 350fail_unlock:
 351	spin_unlock_bh(&hslot->lock);
 352fail:
 353	return error;
 354}
 355EXPORT_SYMBOL(udp_lib_get_port);
 356
 357int udp_v4_get_port(struct sock *sk, unsigned short snum)
 358{
 359	unsigned int hash2_nulladdr =
 360		ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
 361	unsigned int hash2_partial =
 362		ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
 363
 364	/* precompute partial secondary hash */
 365	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
 366	return udp_lib_get_port(sk, snum, hash2_nulladdr);
 367}
 368
 369static int compute_score(struct sock *sk, const struct net *net,
 370			 __be32 saddr, __be16 sport,
 371			 __be32 daddr, unsigned short hnum,
 372			 int dif, int sdif)
 373{
 374	int score;
 375	struct inet_sock *inet;
 376	bool dev_match;
 377
 378	if (!net_eq(sock_net(sk), net) ||
 379	    udp_sk(sk)->udp_port_hash != hnum ||
 380	    ipv6_only_sock(sk))
 381		return -1;
 382
 383	if (sk->sk_rcv_saddr != daddr)
 384		return -1;
 385
 386	score = (sk->sk_family == PF_INET) ? 2 : 1;
 387
 388	inet = inet_sk(sk);
 389	if (inet->inet_daddr) {
 390		if (inet->inet_daddr != saddr)
 391			return -1;
 392		score += 4;
 393	}
 394
 395	if (inet->inet_dport) {
 396		if (inet->inet_dport != sport)
 397			return -1;
 398		score += 4;
 399	}
 400
 401	dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
 402					dif, sdif);
 403	if (!dev_match)
 404		return -1;
 405	if (sk->sk_bound_dev_if)
 406		score += 4;
 407
 408	if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
 409		score++;
 410	return score;
 411}
 412
 413u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
 414		const __be32 faddr, const __be16 fport)
 
 415{
 
 
 416	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
 417
 418	return __inet_ehashfn(laddr, lport, faddr, fport,
 419			      udp_ehash_secret + net_hash_mix(net));
 420}
 421EXPORT_SYMBOL(udp_ehashfn);
 422
 423/**
 424 * udp4_lib_lookup1() - Simplified lookup using primary hash (destination port)
 425 * @net:	Network namespace
 426 * @saddr:	Source address, network order
 427 * @sport:	Source port, network order
 428 * @daddr:	Destination address, network order
 429 * @hnum:	Destination port, host order
 430 * @dif:	Destination interface index
 431 * @sdif:	Destination bridge port index, if relevant
 432 * @udptable:	Set of UDP hash tables
 433 *
 434 * Simplified lookup to be used as fallback if no sockets are found due to a
 435 * potential race between (receive) address change, and lookup happening before
 436 * the rehash operation. This function ignores SO_REUSEPORT groups while scoring
 437 * result sockets, because if we have one, we don't need the fallback at all.
 438 *
 439 * Called under rcu_read_lock().
 440 *
 441 * Return: socket with highest matching score if any, NULL if none
 442 */
 443static struct sock *udp4_lib_lookup1(const struct net *net,
 444				     __be32 saddr, __be16 sport,
 445				     __be32 daddr, unsigned int hnum,
 446				     int dif, int sdif,
 447				     const struct udp_table *udptable)
 448{
 449	unsigned int slot = udp_hashfn(net, hnum, udptable->mask);
 450	struct udp_hslot *hslot = &udptable->hash[slot];
 451	struct sock *sk, *result = NULL;
 452	int score, badness = 0;
 453
 454	sk_for_each_rcu(sk, &hslot->head) {
 455		score = compute_score(sk, net,
 456				      saddr, sport, daddr, hnum, dif, sdif);
 457		if (score > badness) {
 458			result = sk;
 459			badness = score;
 460		}
 461	}
 462
 463	return result;
 464}
 465
 466/* called with rcu_read_lock() */
 467static struct sock *udp4_lib_lookup2(const struct net *net,
 468				     __be32 saddr, __be16 sport,
 469				     __be32 daddr, unsigned int hnum,
 470				     int dif, int sdif,
 471				     struct udp_hslot *hslot2,
 472				     struct sk_buff *skb)
 473{
 474	struct sock *sk, *result;
 475	int score, badness;
 476	bool need_rescore;
 477
 478	result = NULL;
 479	badness = 0;
 480	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
 481		need_rescore = false;
 482rescore:
 483		score = compute_score(need_rescore ? result : sk, net, saddr,
 484				      sport, daddr, hnum, dif, sdif);
 485		if (score > badness) {
 486			badness = score;
 487
 488			if (need_rescore)
 489				continue;
 490
 491			if (sk->sk_state == TCP_ESTABLISHED) {
 492				result = sk;
 493				continue;
 494			}
 495
 496			result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
 497						       saddr, sport, daddr, hnum, udp_ehashfn);
 498			if (!result) {
 499				result = sk;
 500				continue;
 501			}
 502
 503			/* Fall back to scoring if group has connections */
 504			if (!reuseport_has_conns(sk))
 505				return result;
 506
 507			/* Reuseport logic returned an error, keep original score. */
 508			if (IS_ERR(result))
 509				continue;
 510
 511			/* compute_score is too long of a function to be
 512			 * inlined, and calling it again here yields
 513			 * measureable overhead for some
 514			 * workloads. Work around it by jumping
 515			 * backwards to rescore 'result'.
 516			 */
 517			need_rescore = true;
 518			goto rescore;
 519		}
 520	}
 521	return result;
 522}
 523
 524#if IS_ENABLED(CONFIG_BASE_SMALL)
 525static struct sock *udp4_lib_lookup4(const struct net *net,
 526				     __be32 saddr, __be16 sport,
 527				     __be32 daddr, unsigned int hnum,
 528				     int dif, int sdif,
 529				     struct udp_table *udptable)
 530{
 531	return NULL;
 532}
 533
 534static void udp_rehash4(struct udp_table *udptable, struct sock *sk,
 535			u16 newhash4)
 536{
 537}
 538
 539static void udp_unhash4(struct udp_table *udptable, struct sock *sk)
 540{
 541}
 542#else /* !CONFIG_BASE_SMALL */
 543static struct sock *udp4_lib_lookup4(const struct net *net,
 544				     __be32 saddr, __be16 sport,
 545				     __be32 daddr, unsigned int hnum,
 546				     int dif, int sdif,
 547				     struct udp_table *udptable)
 548{
 549	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
 550	const struct hlist_nulls_node *node;
 551	struct udp_hslot *hslot4;
 552	unsigned int hash4, slot;
 553	struct udp_sock *up;
 554	struct sock *sk;
 555
 556	hash4 = udp_ehashfn(net, daddr, hnum, saddr, sport);
 557	slot = hash4 & udptable->mask;
 558	hslot4 = &udptable->hash4[slot];
 559	INET_ADDR_COOKIE(acookie, saddr, daddr);
 560
 561begin:
 562	/* SLAB_TYPESAFE_BY_RCU not used, so we don't need to touch sk_refcnt */
 563	udp_lrpa_for_each_entry_rcu(up, node, &hslot4->nulls_head) {
 564		sk = (struct sock *)up;
 565		if (inet_match(net, sk, acookie, ports, dif, sdif))
 566			return sk;
 567	}
 568
 569	/* if the nulls value we got at the end of this lookup is not the
 570	 * expected one, we must restart lookup. We probably met an item that
 571	 * was moved to another chain due to rehash.
 572	 */
 573	if (get_nulls_value(node) != slot)
 574		goto begin;
 575
 576	return NULL;
 577}
 578
 579/* udp_rehash4() only checks hslot4, and hash4_cnt is not processed. */
 580static void udp_rehash4(struct udp_table *udptable, struct sock *sk,
 581			u16 newhash4)
 582{
 583	struct udp_hslot *hslot4, *nhslot4;
 584
 585	hslot4 = udp_hashslot4(udptable, udp_sk(sk)->udp_lrpa_hash);
 586	nhslot4 = udp_hashslot4(udptable, newhash4);
 587	udp_sk(sk)->udp_lrpa_hash = newhash4;
 588
 589	if (hslot4 != nhslot4) {
 590		spin_lock_bh(&hslot4->lock);
 591		hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_lrpa_node);
 592		hslot4->count--;
 593		spin_unlock_bh(&hslot4->lock);
 594
 595		spin_lock_bh(&nhslot4->lock);
 596		hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_lrpa_node,
 597					 &nhslot4->nulls_head);
 598		nhslot4->count++;
 599		spin_unlock_bh(&nhslot4->lock);
 600	}
 601}
 602
 603static void udp_unhash4(struct udp_table *udptable, struct sock *sk)
 604{
 605	struct udp_hslot *hslot2, *hslot4;
 606
 607	if (udp_hashed4(sk)) {
 608		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 609		hslot4 = udp_hashslot4(udptable, udp_sk(sk)->udp_lrpa_hash);
 610
 611		spin_lock(&hslot4->lock);
 612		hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_lrpa_node);
 613		hslot4->count--;
 614		spin_unlock(&hslot4->lock);
 615
 616		spin_lock(&hslot2->lock);
 617		udp_hash4_dec(hslot2);
 618		spin_unlock(&hslot2->lock);
 619	}
 620}
 621
 622void udp_lib_hash4(struct sock *sk, u16 hash)
 623{
 624	struct udp_hslot *hslot, *hslot2, *hslot4;
 625	struct net *net = sock_net(sk);
 626	struct udp_table *udptable;
 627
 628	/* Connected udp socket can re-connect to another remote address, which
 629	 * will be handled by rehash. Thus no need to redo hash4 here.
 630	 */
 631	if (udp_hashed4(sk))
 632		return;
 633
 634	udptable = net->ipv4.udp_table;
 635	hslot = udp_hashslot(udptable, net, udp_sk(sk)->udp_port_hash);
 636	hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 637	hslot4 = udp_hashslot4(udptable, hash);
 638	udp_sk(sk)->udp_lrpa_hash = hash;
 639
 640	spin_lock_bh(&hslot->lock);
 641	if (rcu_access_pointer(sk->sk_reuseport_cb))
 642		reuseport_detach_sock(sk);
 643
 644	spin_lock(&hslot4->lock);
 645	hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_lrpa_node,
 646				 &hslot4->nulls_head);
 647	hslot4->count++;
 648	spin_unlock(&hslot4->lock);
 649
 650	spin_lock(&hslot2->lock);
 651	udp_hash4_inc(hslot2);
 652	spin_unlock(&hslot2->lock);
 653
 654	spin_unlock_bh(&hslot->lock);
 655}
 656EXPORT_SYMBOL(udp_lib_hash4);
 657
 658/* call with sock lock */
 659void udp4_hash4(struct sock *sk)
 660{
 661	struct net *net = sock_net(sk);
 662	unsigned int hash;
 663
 664	if (sk_unhashed(sk) || sk->sk_rcv_saddr == htonl(INADDR_ANY))
 665		return;
 666
 667	hash = udp_ehashfn(net, sk->sk_rcv_saddr, sk->sk_num,
 668			   sk->sk_daddr, sk->sk_dport);
 669
 670	udp_lib_hash4(sk, hash);
 671}
 672EXPORT_SYMBOL(udp4_hash4);
 673#endif /* CONFIG_BASE_SMALL */
 674
 675/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
 676 * harder than this. -DaveM
 677 */
 678struct sock *__udp4_lib_lookup(const struct net *net, __be32 saddr,
 679		__be16 sport, __be32 daddr, __be16 dport, int dif,
 680		int sdif, struct udp_table *udptable, struct sk_buff *skb)
 681{
 682	unsigned short hnum = ntohs(dport);
 
 683	struct udp_hslot *hslot2;
 684	struct sock *result, *sk;
 685	unsigned int hash2;
 686
 687	hash2 = ipv4_portaddr_hash(net, daddr, hnum);
 688	hslot2 = udp_hashslot2(udptable, hash2);
 689
 690	if (udp_has_hash4(hslot2)) {
 691		result = udp4_lib_lookup4(net, saddr, sport, daddr, hnum,
 692					  dif, sdif, udptable);
 693		if (result) /* udp4_lib_lookup4 return sk or NULL */
 694			return result;
 695	}
 696
 697	/* Lookup connected or non-wildcard socket */
 698	result = udp4_lib_lookup2(net, saddr, sport,
 699				  daddr, hnum, dif, sdif,
 700				  hslot2, skb);
 701	if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
 702		goto done;
 703
 704	/* Lookup redirect from BPF */
 705	if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
 706	    udptable == net->ipv4.udp_table) {
 707		sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
 708					       saddr, sport, daddr, hnum, dif,
 709					       udp_ehashfn);
 710		if (sk) {
 711			result = sk;
 712			goto done;
 713		}
 714	}
 715
 716	/* Got non-wildcard socket or error on first lookup */
 717	if (result)
 718		goto done;
 719
 720	/* Lookup wildcard sockets */
 721	hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
 722	hslot2 = udp_hashslot2(udptable, hash2);
 
 723
 724	result = udp4_lib_lookup2(net, saddr, sport,
 725				  htonl(INADDR_ANY), hnum, dif, sdif,
 726				  hslot2, skb);
 727	if (!IS_ERR_OR_NULL(result))
 728		goto done;
 729
 730	/* Primary hash (destination port) lookup as fallback for this race:
 731	 *   1. __ip4_datagram_connect() sets sk_rcv_saddr
 732	 *   2. lookup (this function): new sk_rcv_saddr, hashes not updated yet
 733	 *   3. rehash operation updating _secondary and four-tuple_ hashes
 734	 * The primary hash doesn't need an update after 1., so, thanks to this
 735	 * further step, 1. and 3. don't need to be atomic against the lookup.
 736	 */
 737	result = udp4_lib_lookup1(net, saddr, sport, daddr, hnum, dif, sdif,
 738				  udptable);
 739
 740done:
 741	if (IS_ERR(result))
 742		return NULL;
 743	return result;
 744}
 745EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
 746
 747static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
 748						 __be16 sport, __be16 dport,
 749						 struct udp_table *udptable)
 750{
 751	const struct iphdr *iph = ip_hdr(skb);
 752
 753	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
 754				 iph->daddr, dport, inet_iif(skb),
 755				 inet_sdif(skb), udptable, skb);
 756}
 757
 758struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
 759				 __be16 sport, __be16 dport)
 760{
 761	const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
 762	const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
 763	struct net *net = dev_net(skb->dev);
 764	int iif, sdif;
 765
 766	inet_get_iif_sdif(skb, &iif, &sdif);
 767
 768	return __udp4_lib_lookup(net, iph->saddr, sport,
 769				 iph->daddr, dport, iif,
 770				 sdif, net->ipv4.udp_table, NULL);
 771}
 772
 773/* Must be called under rcu_read_lock().
 774 * Does increment socket refcount.
 775 */
 776#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
 777struct sock *udp4_lib_lookup(const struct net *net, __be32 saddr, __be16 sport,
 778			     __be32 daddr, __be16 dport, int dif)
 779{
 780	struct sock *sk;
 781
 782	sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
 783			       dif, 0, net->ipv4.udp_table, NULL);
 784	if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
 785		sk = NULL;
 786	return sk;
 787}
 788EXPORT_SYMBOL_GPL(udp4_lib_lookup);
 789#endif
 790
 791static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
 792				       __be16 loc_port, __be32 loc_addr,
 793				       __be16 rmt_port, __be32 rmt_addr,
 794				       int dif, int sdif, unsigned short hnum)
 795{
 796	const struct inet_sock *inet = inet_sk(sk);
 797
 798	if (!net_eq(sock_net(sk), net) ||
 799	    udp_sk(sk)->udp_port_hash != hnum ||
 800	    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
 801	    (inet->inet_dport != rmt_port && inet->inet_dport) ||
 802	    (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
 803	    ipv6_only_sock(sk) ||
 804	    !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
 805		return false;
 806	if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
 807		return false;
 808	return true;
 809}
 810
 811DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
 812EXPORT_SYMBOL(udp_encap_needed_key);
 813
 814#if IS_ENABLED(CONFIG_IPV6)
 815DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
 816EXPORT_SYMBOL(udpv6_encap_needed_key);
 817#endif
 818
 819void udp_encap_enable(void)
 820{
 821	static_branch_inc(&udp_encap_needed_key);
 822}
 823EXPORT_SYMBOL(udp_encap_enable);
 824
 825void udp_encap_disable(void)
 826{
 827	static_branch_dec(&udp_encap_needed_key);
 828}
 829EXPORT_SYMBOL(udp_encap_disable);
 830
 831/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
 832 * through error handlers in encapsulations looking for a match.
 833 */
 834static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
 835{
 836	int i;
 837
 838	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
 839		int (*handler)(struct sk_buff *skb, u32 info);
 840		const struct ip_tunnel_encap_ops *encap;
 841
 842		encap = rcu_dereference(iptun_encaps[i]);
 843		if (!encap)
 844			continue;
 845		handler = encap->err_handler;
 846		if (handler && !handler(skb, info))
 847			return 0;
 848	}
 849
 850	return -ENOENT;
 851}
 852
 853/* Try to match ICMP errors to UDP tunnels by looking up a socket without
 854 * reversing source and destination port: this will match tunnels that force the
 855 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
 856 * lwtunnels might actually break this assumption by being configured with
 857 * different destination ports on endpoints, in this case we won't be able to
 858 * trace ICMP messages back to them.
 859 *
 860 * If this doesn't match any socket, probe tunnels with arbitrary destination
 861 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
 862 * we've sent packets to won't necessarily match the local destination port.
 863 *
 864 * Then ask the tunnel implementation to match the error against a valid
 865 * association.
 866 *
 867 * Return an error if we can't find a match, the socket if we need further
 868 * processing, zero otherwise.
 869 */
 870static struct sock *__udp4_lib_err_encap(struct net *net,
 871					 const struct iphdr *iph,
 872					 struct udphdr *uh,
 873					 struct udp_table *udptable,
 874					 struct sock *sk,
 875					 struct sk_buff *skb, u32 info)
 876{
 877	int (*lookup)(struct sock *sk, struct sk_buff *skb);
 878	int network_offset, transport_offset;
 879	struct udp_sock *up;
 880
 881	network_offset = skb_network_offset(skb);
 882	transport_offset = skb_transport_offset(skb);
 883
 884	/* Network header needs to point to the outer IPv4 header inside ICMP */
 885	skb_reset_network_header(skb);
 886
 887	/* Transport header needs to point to the UDP header */
 888	skb_set_transport_header(skb, iph->ihl << 2);
 889
 890	if (sk) {
 891		up = udp_sk(sk);
 892
 893		lookup = READ_ONCE(up->encap_err_lookup);
 894		if (lookup && lookup(sk, skb))
 895			sk = NULL;
 896
 897		goto out;
 898	}
 899
 900	sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
 901			       iph->saddr, uh->dest, skb->dev->ifindex, 0,
 902			       udptable, NULL);
 903	if (sk) {
 904		up = udp_sk(sk);
 905
 906		lookup = READ_ONCE(up->encap_err_lookup);
 907		if (!lookup || lookup(sk, skb))
 908			sk = NULL;
 909	}
 910
 911out:
 912	if (!sk)
 913		sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
 914
 915	skb_set_transport_header(skb, transport_offset);
 916	skb_set_network_header(skb, network_offset);
 917
 918	return sk;
 919}
 920
 921/*
 922 * This routine is called by the ICMP module when it gets some
 923 * sort of error condition.  If err < 0 then the socket should
 924 * be closed and the error returned to the user.  If err > 0
 925 * it's just the icmp type << 8 | icmp code.
 926 * Header points to the ip header of the error packet. We move
 927 * on past this. Then (as it used to claim before adjustment)
 928 * header points to the first 8 bytes of the udp header.  We need
 929 * to find the appropriate port.
 930 */
 931
 932int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
 933{
 934	struct inet_sock *inet;
 935	const struct iphdr *iph = (const struct iphdr *)skb->data;
 936	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
 937	const int type = icmp_hdr(skb)->type;
 938	const int code = icmp_hdr(skb)->code;
 939	bool tunnel = false;
 940	struct sock *sk;
 941	int harderr;
 942	int err;
 943	struct net *net = dev_net(skb->dev);
 944
 945	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
 946			       iph->saddr, uh->source, skb->dev->ifindex,
 947			       inet_sdif(skb), udptable, NULL);
 948
 949	if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
 950		/* No socket for error: try tunnels before discarding */
 951		if (static_branch_unlikely(&udp_encap_needed_key)) {
 952			sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
 953						  info);
 954			if (!sk)
 955				return 0;
 956		} else
 957			sk = ERR_PTR(-ENOENT);
 958
 959		if (IS_ERR(sk)) {
 960			__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 961			return PTR_ERR(sk);
 962		}
 963
 964		tunnel = true;
 965	}
 966
 967	err = 0;
 968	harderr = 0;
 969	inet = inet_sk(sk);
 970
 971	switch (type) {
 972	default:
 973	case ICMP_TIME_EXCEEDED:
 974		err = EHOSTUNREACH;
 975		break;
 976	case ICMP_SOURCE_QUENCH:
 977		goto out;
 978	case ICMP_PARAMETERPROB:
 979		err = EPROTO;
 980		harderr = 1;
 981		break;
 982	case ICMP_DEST_UNREACH:
 983		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
 984			ipv4_sk_update_pmtu(skb, sk, info);
 985			if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
 986				err = EMSGSIZE;
 987				harderr = 1;
 988				break;
 989			}
 990			goto out;
 991		}
 992		err = EHOSTUNREACH;
 993		if (code <= NR_ICMP_UNREACH) {
 994			harderr = icmp_err_convert[code].fatal;
 995			err = icmp_err_convert[code].errno;
 996		}
 997		break;
 998	case ICMP_REDIRECT:
 999		ipv4_sk_redirect(skb, sk);
1000		goto out;
1001	}
1002
1003	/*
1004	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
1005	 *	4.1.3.3.
1006	 */
1007	if (tunnel) {
1008		/* ...not for tunnels though: we don't have a sending socket */
1009		if (udp_sk(sk)->encap_err_rcv)
1010			udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
1011						  (u8 *)(uh+1));
1012		goto out;
1013	}
1014	if (!inet_test_bit(RECVERR, sk)) {
1015		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
1016			goto out;
1017	} else
1018		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
1019
1020	sk->sk_err = err;
1021	sk_error_report(sk);
1022out:
1023	return 0;
1024}
1025
1026int udp_err(struct sk_buff *skb, u32 info)
1027{
1028	return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
1029}
1030
1031/*
1032 * Throw away all pending data and cancel the corking. Socket is locked.
1033 */
1034void udp_flush_pending_frames(struct sock *sk)
1035{
1036	struct udp_sock *up = udp_sk(sk);
1037
1038	if (up->pending) {
1039		up->len = 0;
1040		WRITE_ONCE(up->pending, 0);
1041		ip_flush_pending_frames(sk);
1042	}
1043}
1044EXPORT_SYMBOL(udp_flush_pending_frames);
1045
1046/**
1047 * 	udp4_hwcsum  -  handle outgoing HW checksumming
1048 * 	@skb: 	sk_buff containing the filled-in UDP header
1049 * 	        (checksum field must be zeroed out)
1050 *	@src:	source IP address
1051 *	@dst:	destination IP address
1052 */
1053void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
1054{
1055	struct udphdr *uh = udp_hdr(skb);
1056	int offset = skb_transport_offset(skb);
1057	int len = skb->len - offset;
1058	int hlen = len;
1059	__wsum csum = 0;
1060
1061	if (!skb_has_frag_list(skb)) {
1062		/*
1063		 * Only one fragment on the socket.
1064		 */
1065		skb->csum_start = skb_transport_header(skb) - skb->head;
1066		skb->csum_offset = offsetof(struct udphdr, check);
1067		uh->check = ~csum_tcpudp_magic(src, dst, len,
1068					       IPPROTO_UDP, 0);
1069	} else {
1070		struct sk_buff *frags;
1071
1072		/*
1073		 * HW-checksum won't work as there are two or more
1074		 * fragments on the socket so that all csums of sk_buffs
1075		 * should be together
1076		 */
1077		skb_walk_frags(skb, frags) {
1078			csum = csum_add(csum, frags->csum);
1079			hlen -= frags->len;
1080		}
1081
1082		csum = skb_checksum(skb, offset, hlen, csum);
1083		skb->ip_summed = CHECKSUM_NONE;
1084
1085		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
1086		if (uh->check == 0)
1087			uh->check = CSUM_MANGLED_0;
1088	}
1089}
1090EXPORT_SYMBOL_GPL(udp4_hwcsum);
1091
1092/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
1093 * for the simple case like when setting the checksum for a UDP tunnel.
1094 */
1095void udp_set_csum(bool nocheck, struct sk_buff *skb,
1096		  __be32 saddr, __be32 daddr, int len)
1097{
1098	struct udphdr *uh = udp_hdr(skb);
1099
1100	if (nocheck) {
1101		uh->check = 0;
1102	} else if (skb_is_gso(skb)) {
1103		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
1104	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
1105		uh->check = 0;
1106		uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
1107		if (uh->check == 0)
1108			uh->check = CSUM_MANGLED_0;
1109	} else {
1110		skb->ip_summed = CHECKSUM_PARTIAL;
1111		skb->csum_start = skb_transport_header(skb) - skb->head;
1112		skb->csum_offset = offsetof(struct udphdr, check);
1113		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
1114	}
1115}
1116EXPORT_SYMBOL(udp_set_csum);
1117
1118static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
1119			struct inet_cork *cork)
1120{
1121	struct sock *sk = skb->sk;
1122	struct inet_sock *inet = inet_sk(sk);
1123	struct udphdr *uh;
1124	int err;
1125	int is_udplite = IS_UDPLITE(sk);
1126	int offset = skb_transport_offset(skb);
1127	int len = skb->len - offset;
1128	int datalen = len - sizeof(*uh);
1129	__wsum csum = 0;
1130
1131	/*
1132	 * Create a UDP header
1133	 */
1134	uh = udp_hdr(skb);
1135	uh->source = inet->inet_sport;
1136	uh->dest = fl4->fl4_dport;
1137	uh->len = htons(len);
1138	uh->check = 0;
1139
1140	if (cork->gso_size) {
1141		const int hlen = skb_network_header_len(skb) +
1142				 sizeof(struct udphdr);
1143
1144		if (hlen + min(datalen, cork->gso_size) > cork->fragsize) {
1145			kfree_skb(skb);
1146			return -EMSGSIZE;
1147		}
1148		if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1149			kfree_skb(skb);
1150			return -EINVAL;
1151		}
1152		if (sk->sk_no_check_tx) {
1153			kfree_skb(skb);
1154			return -EINVAL;
1155		}
1156		if (is_udplite || dst_xfrm(skb_dst(skb))) {
 
1157			kfree_skb(skb);
1158			return -EIO;
1159		}
1160
1161		if (datalen > cork->gso_size) {
1162			skb_shinfo(skb)->gso_size = cork->gso_size;
1163			skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1164			skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1165								 cork->gso_size);
1166
1167			/* Don't checksum the payload, skb will get segmented */
1168			goto csum_partial;
1169		}
 
1170	}
1171
1172	if (is_udplite)  				 /*     UDP-Lite      */
1173		csum = udplite_csum(skb);
1174
1175	else if (sk->sk_no_check_tx) {			 /* UDP csum off */
1176
1177		skb->ip_summed = CHECKSUM_NONE;
1178		goto send;
1179
1180	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1181csum_partial:
1182
1183		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
1184		goto send;
1185
1186	} else
1187		csum = udp_csum(skb);
1188
1189	/* add protocol-dependent pseudo-header */
1190	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
1191				      sk->sk_protocol, csum);
1192	if (uh->check == 0)
1193		uh->check = CSUM_MANGLED_0;
1194
1195send:
1196	err = ip_send_skb(sock_net(sk), skb);
1197	if (err) {
1198		if (err == -ENOBUFS &&
1199		    !inet_test_bit(RECVERR, sk)) {
1200			UDP_INC_STATS(sock_net(sk),
1201				      UDP_MIB_SNDBUFERRORS, is_udplite);
1202			err = 0;
1203		}
1204	} else
1205		UDP_INC_STATS(sock_net(sk),
1206			      UDP_MIB_OUTDATAGRAMS, is_udplite);
1207	return err;
1208}
1209
1210/*
1211 * Push out all pending data as one UDP datagram. Socket is locked.
1212 */
1213int udp_push_pending_frames(struct sock *sk)
1214{
1215	struct udp_sock  *up = udp_sk(sk);
1216	struct inet_sock *inet = inet_sk(sk);
1217	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1218	struct sk_buff *skb;
1219	int err = 0;
1220
1221	skb = ip_finish_skb(sk, fl4);
1222	if (!skb)
1223		goto out;
1224
1225	err = udp_send_skb(skb, fl4, &inet->cork.base);
1226
1227out:
1228	up->len = 0;
1229	WRITE_ONCE(up->pending, 0);
1230	return err;
1231}
1232EXPORT_SYMBOL(udp_push_pending_frames);
1233
1234static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1235{
1236	switch (cmsg->cmsg_type) {
1237	case UDP_SEGMENT:
1238		if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1239			return -EINVAL;
1240		*gso_size = *(__u16 *)CMSG_DATA(cmsg);
1241		return 0;
1242	default:
1243		return -EINVAL;
1244	}
1245}
1246
1247int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1248{
1249	struct cmsghdr *cmsg;
1250	bool need_ip = false;
1251	int err;
1252
1253	for_each_cmsghdr(cmsg, msg) {
1254		if (!CMSG_OK(msg, cmsg))
1255			return -EINVAL;
1256
1257		if (cmsg->cmsg_level != SOL_UDP) {
1258			need_ip = true;
1259			continue;
1260		}
1261
1262		err = __udp_cmsg_send(cmsg, gso_size);
1263		if (err)
1264			return err;
1265	}
1266
1267	return need_ip;
1268}
1269EXPORT_SYMBOL_GPL(udp_cmsg_send);
1270
1271int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1272{
1273	struct inet_sock *inet = inet_sk(sk);
1274	struct udp_sock *up = udp_sk(sk);
1275	DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1276	struct flowi4 fl4_stack;
1277	struct flowi4 *fl4;
1278	int ulen = len;
1279	struct ipcm_cookie ipc;
1280	struct rtable *rt = NULL;
1281	int free = 0;
1282	int connected = 0;
1283	__be32 daddr, faddr, saddr;
1284	u8 tos, scope;
1285	__be16 dport;
 
1286	int err, is_udplite = IS_UDPLITE(sk);
1287	int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1288	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1289	struct sk_buff *skb;
1290	struct ip_options_data opt_copy;
1291	int uc_index;
1292
1293	if (len > 0xFFFF)
1294		return -EMSGSIZE;
1295
1296	/*
1297	 *	Check the flags.
1298	 */
1299
1300	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1301		return -EOPNOTSUPP;
1302
1303	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1304
1305	fl4 = &inet->cork.fl.u.ip4;
1306	if (READ_ONCE(up->pending)) {
1307		/*
1308		 * There are pending frames.
1309		 * The socket lock must be held while it's corked.
1310		 */
1311		lock_sock(sk);
1312		if (likely(up->pending)) {
1313			if (unlikely(up->pending != AF_INET)) {
1314				release_sock(sk);
1315				return -EINVAL;
1316			}
1317			goto do_append_data;
1318		}
1319		release_sock(sk);
1320	}
1321	ulen += sizeof(struct udphdr);
1322
1323	/*
1324	 *	Get and verify the address.
1325	 */
1326	if (usin) {
1327		if (msg->msg_namelen < sizeof(*usin))
1328			return -EINVAL;
1329		if (usin->sin_family != AF_INET) {
1330			if (usin->sin_family != AF_UNSPEC)
1331				return -EAFNOSUPPORT;
1332		}
1333
1334		daddr = usin->sin_addr.s_addr;
1335		dport = usin->sin_port;
1336		if (dport == 0)
1337			return -EINVAL;
1338	} else {
1339		if (sk->sk_state != TCP_ESTABLISHED)
1340			return -EDESTADDRREQ;
1341		daddr = inet->inet_daddr;
1342		dport = inet->inet_dport;
1343		/* Open fast path for connected socket.
1344		   Route will not be used, if at least one option is set.
1345		 */
1346		connected = 1;
1347	}
1348
1349	ipcm_init_sk(&ipc, inet);
1350	ipc.gso_size = READ_ONCE(up->gso_size);
1351
1352	if (msg->msg_controllen) {
1353		err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1354		if (err > 0) {
1355			err = ip_cmsg_send(sk, msg, &ipc,
1356					   sk->sk_family == AF_INET6);
1357			connected = 0;
1358		}
1359		if (unlikely(err < 0)) {
1360			kfree(ipc.opt);
1361			return err;
1362		}
1363		if (ipc.opt)
1364			free = 1;
 
1365	}
1366	if (!ipc.opt) {
1367		struct ip_options_rcu *inet_opt;
1368
1369		rcu_read_lock();
1370		inet_opt = rcu_dereference(inet->inet_opt);
1371		if (inet_opt) {
1372			memcpy(&opt_copy, inet_opt,
1373			       sizeof(*inet_opt) + inet_opt->opt.optlen);
1374			ipc.opt = &opt_copy.opt;
1375		}
1376		rcu_read_unlock();
1377	}
1378
1379	if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1380		err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1381					    (struct sockaddr *)usin,
1382					    &msg->msg_namelen,
1383					    &ipc.addr);
1384		if (err)
1385			goto out_free;
1386		if (usin) {
1387			if (usin->sin_port == 0) {
1388				/* BPF program set invalid port. Reject it. */
1389				err = -EINVAL;
1390				goto out_free;
1391			}
1392			daddr = usin->sin_addr.s_addr;
1393			dport = usin->sin_port;
1394		}
1395	}
1396
1397	saddr = ipc.addr;
1398	ipc.addr = faddr = daddr;
1399
1400	if (ipc.opt && ipc.opt->opt.srr) {
1401		if (!daddr) {
1402			err = -EINVAL;
1403			goto out_free;
1404		}
1405		faddr = ipc.opt->opt.faddr;
1406		connected = 0;
1407	}
1408	tos = get_rttos(&ipc, inet);
1409	scope = ip_sendmsg_scope(inet, &ipc, msg);
1410	if (scope == RT_SCOPE_LINK)
 
 
1411		connected = 0;
 
1412
1413	uc_index = READ_ONCE(inet->uc_index);
1414	if (ipv4_is_multicast(daddr)) {
1415		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1416			ipc.oif = READ_ONCE(inet->mc_index);
1417		if (!saddr)
1418			saddr = READ_ONCE(inet->mc_addr);
1419		connected = 0;
1420	} else if (!ipc.oif) {
1421		ipc.oif = uc_index;
1422	} else if (ipv4_is_lbcast(daddr) && uc_index) {
1423		/* oif is set, packet is to local broadcast and
1424		 * uc_index is set. oif is most likely set
1425		 * by sk_bound_dev_if. If uc_index != oif check if the
1426		 * oif is an L3 master and uc_index is an L3 slave.
1427		 * If so, we want to allow the send using the uc_index.
1428		 */
1429		if (ipc.oif != uc_index &&
1430		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1431							      uc_index)) {
1432			ipc.oif = uc_index;
1433		}
1434	}
1435
1436	if (connected)
1437		rt = dst_rtable(sk_dst_check(sk, 0));
1438
1439	if (!rt) {
1440		struct net *net = sock_net(sk);
1441		__u8 flow_flags = inet_sk_flowi_flags(sk);
1442
1443		fl4 = &fl4_stack;
1444
1445		flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1446				   sk->sk_protocol, flow_flags, faddr, saddr,
1447				   dport, inet->inet_sport, sk->sk_uid);
 
 
1448
1449		security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1450		rt = ip_route_output_flow(net, fl4, sk);
1451		if (IS_ERR(rt)) {
1452			err = PTR_ERR(rt);
1453			rt = NULL;
1454			if (err == -ENETUNREACH)
1455				IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1456			goto out;
1457		}
1458
1459		err = -EACCES;
1460		if ((rt->rt_flags & RTCF_BROADCAST) &&
1461		    !sock_flag(sk, SOCK_BROADCAST))
1462			goto out;
1463		if (connected)
1464			sk_dst_set(sk, dst_clone(&rt->dst));
1465	}
1466
1467	if (msg->msg_flags&MSG_CONFIRM)
1468		goto do_confirm;
1469back_from_confirm:
1470
1471	saddr = fl4->saddr;
1472	if (!ipc.addr)
1473		daddr = ipc.addr = fl4->daddr;
1474
1475	/* Lockless fast path for the non-corking case. */
1476	if (!corkreq) {
1477		struct inet_cork cork;
1478
1479		skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1480				  sizeof(struct udphdr), &ipc, &rt,
1481				  &cork, msg->msg_flags);
1482		err = PTR_ERR(skb);
1483		if (!IS_ERR_OR_NULL(skb))
1484			err = udp_send_skb(skb, fl4, &cork);
1485		goto out;
1486	}
1487
1488	lock_sock(sk);
1489	if (unlikely(up->pending)) {
1490		/* The socket is already corked while preparing it. */
1491		/* ... which is an evident application bug. --ANK */
1492		release_sock(sk);
1493
1494		net_dbg_ratelimited("socket already corked\n");
1495		err = -EINVAL;
1496		goto out;
1497	}
1498	/*
1499	 *	Now cork the socket to pend data.
1500	 */
1501	fl4 = &inet->cork.fl.u.ip4;
1502	fl4->daddr = daddr;
1503	fl4->saddr = saddr;
1504	fl4->fl4_dport = dport;
1505	fl4->fl4_sport = inet->inet_sport;
1506	WRITE_ONCE(up->pending, AF_INET);
1507
1508do_append_data:
1509	up->len += ulen;
1510	err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1511			     sizeof(struct udphdr), &ipc, &rt,
1512			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1513	if (err)
1514		udp_flush_pending_frames(sk);
1515	else if (!corkreq)
1516		err = udp_push_pending_frames(sk);
1517	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1518		WRITE_ONCE(up->pending, 0);
1519	release_sock(sk);
1520
1521out:
1522	ip_rt_put(rt);
1523out_free:
1524	if (free)
1525		kfree(ipc.opt);
1526	if (!err)
1527		return len;
1528	/*
1529	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1530	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1531	 * we don't have a good statistic (IpOutDiscards but it can be too many
1532	 * things).  We could add another new stat but at least for now that
1533	 * seems like overkill.
1534	 */
1535	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1536		UDP_INC_STATS(sock_net(sk),
1537			      UDP_MIB_SNDBUFERRORS, is_udplite);
1538	}
1539	return err;
1540
1541do_confirm:
1542	if (msg->msg_flags & MSG_PROBE)
1543		dst_confirm_neigh(&rt->dst, &fl4->daddr);
1544	if (!(msg->msg_flags&MSG_PROBE) || len)
1545		goto back_from_confirm;
1546	err = 0;
1547	goto out;
1548}
1549EXPORT_SYMBOL(udp_sendmsg);
1550
1551void udp_splice_eof(struct socket *sock)
 
1552{
1553	struct sock *sk = sock->sk;
1554	struct udp_sock *up = udp_sk(sk);
 
1555
1556	if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1557		return;
 
 
 
 
 
 
 
 
 
 
 
 
1558
1559	lock_sock(sk);
1560	if (up->pending && !udp_test_bit(CORK, sk))
1561		udp_push_pending_frames(sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1562	release_sock(sk);
 
1563}
1564EXPORT_SYMBOL_GPL(udp_splice_eof);
1565
1566#define UDP_SKB_IS_STATELESS 0x80000000
1567
1568/* all head states (dst, sk, nf conntrack) except skb extensions are
1569 * cleared by udp_rcv().
1570 *
1571 * We need to preserve secpath, if present, to eventually process
1572 * IP_CMSG_PASSSEC at recvmsg() time.
1573 *
1574 * Other extensions can be cleared.
1575 */
1576static bool udp_try_make_stateless(struct sk_buff *skb)
1577{
1578	if (!skb_has_extensions(skb))
1579		return true;
1580
1581	if (!secpath_exists(skb)) {
1582		skb_ext_reset(skb);
1583		return true;
1584	}
1585
1586	return false;
1587}
1588
1589static void udp_set_dev_scratch(struct sk_buff *skb)
1590{
1591	struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1592
1593	BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1594	scratch->_tsize_state = skb->truesize;
1595#if BITS_PER_LONG == 64
1596	scratch->len = skb->len;
1597	scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1598	scratch->is_linear = !skb_is_nonlinear(skb);
1599#endif
1600	if (udp_try_make_stateless(skb))
1601		scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1602}
1603
1604static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1605{
1606	/* We come here after udp_lib_checksum_complete() returned 0.
1607	 * This means that __skb_checksum_complete() might have
1608	 * set skb->csum_valid to 1.
1609	 * On 64bit platforms, we can set csum_unnecessary
1610	 * to true, but only if the skb is not shared.
1611	 */
1612#if BITS_PER_LONG == 64
1613	if (!skb_shared(skb))
1614		udp_skb_scratch(skb)->csum_unnecessary = true;
1615#endif
1616}
1617
1618static int udp_skb_truesize(struct sk_buff *skb)
1619{
1620	return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1621}
1622
1623static bool udp_skb_has_head_state(struct sk_buff *skb)
1624{
1625	return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1626}
1627
1628/* fully reclaim rmem/fwd memory allocated for skb */
1629static void udp_rmem_release(struct sock *sk, int size, int partial,
1630			     bool rx_queue_lock_held)
1631{
1632	struct udp_sock *up = udp_sk(sk);
1633	struct sk_buff_head *sk_queue;
1634	int amt;
1635
1636	if (likely(partial)) {
1637		up->forward_deficit += size;
1638		size = up->forward_deficit;
1639		if (size < READ_ONCE(up->forward_threshold) &&
1640		    !skb_queue_empty(&up->reader_queue))
1641			return;
1642	} else {
1643		size += up->forward_deficit;
1644	}
1645	up->forward_deficit = 0;
1646
1647	/* acquire the sk_receive_queue for fwd allocated memory scheduling,
1648	 * if the called don't held it already
1649	 */
1650	sk_queue = &sk->sk_receive_queue;
1651	if (!rx_queue_lock_held)
1652		spin_lock(&sk_queue->lock);
1653
1654
1655	sk_forward_alloc_add(sk, size);
1656	amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1657	sk_forward_alloc_add(sk, -amt);
1658
1659	if (amt)
1660		__sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1661
1662	atomic_sub(size, &sk->sk_rmem_alloc);
1663
1664	/* this can save us from acquiring the rx queue lock on next receive */
1665	skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1666
1667	if (!rx_queue_lock_held)
1668		spin_unlock(&sk_queue->lock);
1669}
1670
1671/* Note: called with reader_queue.lock held.
1672 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1673 * This avoids a cache line miss while receive_queue lock is held.
1674 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1675 */
1676void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1677{
1678	prefetch(&skb->data);
1679	udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1680}
1681EXPORT_SYMBOL(udp_skb_destructor);
1682
1683/* as above, but the caller held the rx queue lock, too */
1684static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1685{
1686	prefetch(&skb->data);
1687	udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1688}
1689
1690/* Idea of busylocks is to let producers grab an extra spinlock
1691 * to relieve pressure on the receive_queue spinlock shared by consumer.
1692 * Under flood, this means that only one producer can be in line
1693 * trying to acquire the receive_queue spinlock.
1694 * These busylock can be allocated on a per cpu manner, instead of a
1695 * per socket one (that would consume a cache line per socket)
1696 */
1697static int udp_busylocks_log __read_mostly;
1698static spinlock_t *udp_busylocks __read_mostly;
1699
1700static spinlock_t *busylock_acquire(void *ptr)
1701{
1702	spinlock_t *busy;
1703
1704	busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1705	spin_lock(busy);
1706	return busy;
1707}
1708
1709static void busylock_release(spinlock_t *busy)
1710{
1711	if (busy)
1712		spin_unlock(busy);
1713}
1714
1715static int udp_rmem_schedule(struct sock *sk, int size)
1716{
1717	int delta;
1718
1719	delta = size - sk->sk_forward_alloc;
1720	if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1721		return -ENOBUFS;
1722
1723	return 0;
1724}
1725
1726int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1727{
1728	struct sk_buff_head *list = &sk->sk_receive_queue;
1729	int rmem, err = -ENOMEM;
1730	spinlock_t *busy = NULL;
1731	int size, rcvbuf;
1732
1733	/* Immediately drop when the receive queue is full.
1734	 * Always allow at least one packet.
1735	 */
1736	rmem = atomic_read(&sk->sk_rmem_alloc);
1737	rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1738	if (rmem > rcvbuf)
1739		goto drop;
1740
1741	/* Under mem pressure, it might be helpful to help udp_recvmsg()
1742	 * having linear skbs :
1743	 * - Reduce memory overhead and thus increase receive queue capacity
1744	 * - Less cache line misses at copyout() time
1745	 * - Less work at consume_skb() (less alien page frag freeing)
1746	 */
1747	if (rmem > (rcvbuf >> 1)) {
1748		skb_condense(skb);
1749
1750		busy = busylock_acquire(sk);
1751	}
1752	size = skb->truesize;
1753	udp_set_dev_scratch(skb);
1754
1755	atomic_add(size, &sk->sk_rmem_alloc);
 
 
 
 
 
1756
1757	spin_lock(&list->lock);
1758	err = udp_rmem_schedule(sk, size);
1759	if (err) {
1760		spin_unlock(&list->lock);
1761		goto uncharge_drop;
 
 
 
 
 
 
1762	}
1763
1764	sk_forward_alloc_add(sk, -size);
1765
1766	/* no need to setup a destructor, we will explicitly release the
1767	 * forward allocated memory on dequeue
1768	 */
1769	sock_skb_set_dropcount(sk, skb);
1770
1771	__skb_queue_tail(list, skb);
1772	spin_unlock(&list->lock);
1773
1774	if (!sock_flag(sk, SOCK_DEAD))
1775		INDIRECT_CALL_1(sk->sk_data_ready, sock_def_readable, sk);
1776
1777	busylock_release(busy);
1778	return 0;
1779
1780uncharge_drop:
1781	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1782
1783drop:
1784	atomic_inc(&sk->sk_drops);
1785	busylock_release(busy);
1786	return err;
1787}
1788EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1789
1790void udp_destruct_common(struct sock *sk)
1791{
1792	/* reclaim completely the forward allocated memory */
1793	struct udp_sock *up = udp_sk(sk);
1794	unsigned int total = 0;
1795	struct sk_buff *skb;
1796
1797	skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1798	while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1799		total += skb->truesize;
1800		kfree_skb(skb);
1801	}
1802	udp_rmem_release(sk, total, 0, true);
1803}
1804EXPORT_SYMBOL_GPL(udp_destruct_common);
1805
1806static void udp_destruct_sock(struct sock *sk)
1807{
1808	udp_destruct_common(sk);
1809	inet_sock_destruct(sk);
1810}
1811
1812int udp_init_sock(struct sock *sk)
1813{
1814	udp_lib_init_sock(sk);
1815	sk->sk_destruct = udp_destruct_sock;
1816	set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1817	return 0;
1818}
1819
1820void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1821{
1822	if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset)))
 
 
1823		sk_peek_offset_bwd(sk, len);
 
 
1824
1825	if (!skb_unref(skb))
1826		return;
1827
1828	/* In the more common cases we cleared the head states previously,
1829	 * see __udp_queue_rcv_skb().
1830	 */
1831	if (unlikely(udp_skb_has_head_state(skb)))
1832		skb_release_head_state(skb);
1833	__consume_stateless_skb(skb);
1834}
1835EXPORT_SYMBOL_GPL(skb_consume_udp);
1836
1837static struct sk_buff *__first_packet_length(struct sock *sk,
1838					     struct sk_buff_head *rcvq,
1839					     int *total)
1840{
1841	struct sk_buff *skb;
1842
1843	while ((skb = skb_peek(rcvq)) != NULL) {
1844		if (udp_lib_checksum_complete(skb)) {
1845			__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1846					IS_UDPLITE(sk));
1847			__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1848					IS_UDPLITE(sk));
1849			atomic_inc(&sk->sk_drops);
1850			__skb_unlink(skb, rcvq);
1851			*total += skb->truesize;
1852			kfree_skb(skb);
1853		} else {
1854			udp_skb_csum_unnecessary_set(skb);
1855			break;
1856		}
1857	}
1858	return skb;
1859}
1860
1861/**
1862 *	first_packet_length	- return length of first packet in receive queue
1863 *	@sk: socket
1864 *
1865 *	Drops all bad checksum frames, until a valid one is found.
1866 *	Returns the length of found skb, or -1 if none is found.
1867 */
1868static int first_packet_length(struct sock *sk)
1869{
1870	struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1871	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1872	struct sk_buff *skb;
1873	int total = 0;
1874	int res;
1875
1876	spin_lock_bh(&rcvq->lock);
1877	skb = __first_packet_length(sk, rcvq, &total);
1878	if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1879		spin_lock(&sk_queue->lock);
1880		skb_queue_splice_tail_init(sk_queue, rcvq);
1881		spin_unlock(&sk_queue->lock);
1882
1883		skb = __first_packet_length(sk, rcvq, &total);
1884	}
1885	res = skb ? skb->len : -1;
1886	if (total)
1887		udp_rmem_release(sk, total, 1, false);
1888	spin_unlock_bh(&rcvq->lock);
1889	return res;
1890}
1891
1892/*
1893 *	IOCTL requests applicable to the UDP protocol
1894 */
1895
1896int udp_ioctl(struct sock *sk, int cmd, int *karg)
1897{
1898	switch (cmd) {
1899	case SIOCOUTQ:
1900	{
1901		*karg = sk_wmem_alloc_get(sk);
1902		return 0;
 
1903	}
1904
1905	case SIOCINQ:
1906	{
1907		*karg = max_t(int, 0, first_packet_length(sk));
1908		return 0;
 
1909	}
1910
1911	default:
1912		return -ENOIOCTLCMD;
1913	}
1914
1915	return 0;
1916}
1917EXPORT_SYMBOL(udp_ioctl);
1918
1919struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1920			       int *off, int *err)
1921{
1922	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1923	struct sk_buff_head *queue;
1924	struct sk_buff *last;
1925	long timeo;
1926	int error;
1927
1928	queue = &udp_sk(sk)->reader_queue;
1929	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1930	do {
1931		struct sk_buff *skb;
1932
1933		error = sock_error(sk);
1934		if (error)
1935			break;
1936
1937		error = -EAGAIN;
1938		do {
1939			spin_lock_bh(&queue->lock);
1940			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1941							err, &last);
1942			if (skb) {
1943				if (!(flags & MSG_PEEK))
1944					udp_skb_destructor(sk, skb);
1945				spin_unlock_bh(&queue->lock);
1946				return skb;
1947			}
1948
1949			if (skb_queue_empty_lockless(sk_queue)) {
1950				spin_unlock_bh(&queue->lock);
1951				goto busy_check;
1952			}
1953
1954			/* refill the reader queue and walk it again
1955			 * keep both queues locked to avoid re-acquiring
1956			 * the sk_receive_queue lock if fwd memory scheduling
1957			 * is needed.
1958			 */
1959			spin_lock(&sk_queue->lock);
1960			skb_queue_splice_tail_init(sk_queue, queue);
1961
1962			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1963							err, &last);
1964			if (skb && !(flags & MSG_PEEK))
1965				udp_skb_dtor_locked(sk, skb);
1966			spin_unlock(&sk_queue->lock);
1967			spin_unlock_bh(&queue->lock);
1968			if (skb)
1969				return skb;
1970
1971busy_check:
1972			if (!sk_can_busy_loop(sk))
1973				break;
1974
1975			sk_busy_loop(sk, flags & MSG_DONTWAIT);
1976		} while (!skb_queue_empty_lockless(sk_queue));
1977
1978		/* sk_queue is empty, reader_queue may contain peeked packets */
1979	} while (timeo &&
1980		 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1981					      &error, &timeo,
1982					      (struct sk_buff *)sk_queue));
1983
1984	*err = error;
1985	return NULL;
1986}
1987EXPORT_SYMBOL(__skb_recv_udp);
1988
1989int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1990{
1991	struct sk_buff *skb;
1992	int err;
1993
1994try_again:
1995	skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1996	if (!skb)
1997		return err;
1998
1999	if (udp_lib_checksum_complete(skb)) {
2000		int is_udplite = IS_UDPLITE(sk);
2001		struct net *net = sock_net(sk);
2002
2003		__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
2004		__UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
2005		atomic_inc(&sk->sk_drops);
2006		kfree_skb(skb);
2007		goto try_again;
2008	}
2009
2010	WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
2011	return recv_actor(sk, skb);
 
 
 
2012}
2013EXPORT_SYMBOL(udp_read_skb);
2014
2015/*
2016 * 	This should be easy, if there is something there we
2017 * 	return it, otherwise we block.
2018 */
2019
2020int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
2021		int *addr_len)
2022{
2023	struct inet_sock *inet = inet_sk(sk);
2024	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
2025	struct sk_buff *skb;
2026	unsigned int ulen, copied;
2027	int off, err, peeking = flags & MSG_PEEK;
2028	int is_udplite = IS_UDPLITE(sk);
2029	bool checksum_valid = false;
2030
2031	if (flags & MSG_ERRQUEUE)
2032		return ip_recv_error(sk, msg, len, addr_len);
2033
2034try_again:
2035	off = sk_peek_offset(sk, flags);
2036	skb = __skb_recv_udp(sk, flags, &off, &err);
2037	if (!skb)
2038		return err;
2039
2040	ulen = udp_skb_len(skb);
2041	copied = len;
2042	if (copied > ulen - off)
2043		copied = ulen - off;
2044	else if (copied < ulen)
2045		msg->msg_flags |= MSG_TRUNC;
2046
2047	/*
2048	 * If checksum is needed at all, try to do it while copying the
2049	 * data.  If the data is truncated, or if we only want a partial
2050	 * coverage checksum (UDP-Lite), do it before the copy.
2051	 */
2052
2053	if (copied < ulen || peeking ||
2054	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
2055		checksum_valid = udp_skb_csum_unnecessary(skb) ||
2056				!__udp_lib_checksum_complete(skb);
2057		if (!checksum_valid)
2058			goto csum_copy_err;
2059	}
2060
2061	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
2062		if (udp_skb_is_linear(skb))
2063			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
2064		else
2065			err = skb_copy_datagram_msg(skb, off, msg, copied);
2066	} else {
2067		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
2068
2069		if (err == -EINVAL)
2070			goto csum_copy_err;
2071	}
2072
2073	if (unlikely(err)) {
2074		if (!peeking) {
2075			atomic_inc(&sk->sk_drops);
2076			UDP_INC_STATS(sock_net(sk),
2077				      UDP_MIB_INERRORS, is_udplite);
2078		}
2079		kfree_skb(skb);
2080		return err;
2081	}
2082
2083	if (!peeking)
2084		UDP_INC_STATS(sock_net(sk),
2085			      UDP_MIB_INDATAGRAMS, is_udplite);
2086
2087	sock_recv_cmsgs(msg, sk, skb);
2088
2089	/* Copy the address. */
2090	if (sin) {
2091		sin->sin_family = AF_INET;
2092		sin->sin_port = udp_hdr(skb)->source;
2093		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
2094		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
2095		*addr_len = sizeof(*sin);
2096
2097		BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
2098						      (struct sockaddr *)sin,
2099						      addr_len);
2100	}
2101
2102	if (udp_test_bit(GRO_ENABLED, sk))
2103		udp_cmsg_recv(msg, sk, skb);
2104
2105	if (inet_cmsg_flags(inet))
2106		ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
2107
2108	err = copied;
2109	if (flags & MSG_TRUNC)
2110		err = ulen;
2111
2112	skb_consume_udp(sk, skb, peeking ? -err : err);
2113	return err;
2114
2115csum_copy_err:
2116	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
2117				 udp_skb_destructor)) {
2118		UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2119		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2120	}
2121	kfree_skb(skb);
2122
2123	/* starting over for a new packet, but check if we need to yield */
2124	cond_resched();
2125	msg->msg_flags &= ~MSG_TRUNC;
2126	goto try_again;
2127}
2128
2129int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
2130{
2131	/* This check is replicated from __ip4_datagram_connect() and
2132	 * intended to prevent BPF program called below from accessing bytes
2133	 * that are out of the bound specified by user in addr_len.
2134	 */
2135	if (addr_len < sizeof(struct sockaddr_in))
2136		return -EINVAL;
2137
2138	return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
2139}
2140EXPORT_SYMBOL(udp_pre_connect);
2141
2142static int udp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
2143{
2144	int res;
2145
2146	lock_sock(sk);
2147	res = __ip4_datagram_connect(sk, uaddr, addr_len);
2148	if (!res)
2149		udp4_hash4(sk);
2150	release_sock(sk);
2151	return res;
2152}
2153
2154int __udp_disconnect(struct sock *sk, int flags)
2155{
2156	struct inet_sock *inet = inet_sk(sk);
2157	/*
2158	 *	1003.1g - break association.
2159	 */
2160
2161	sk->sk_state = TCP_CLOSE;
2162	inet->inet_daddr = 0;
2163	inet->inet_dport = 0;
2164	sock_rps_reset_rxhash(sk);
2165	sk->sk_bound_dev_if = 0;
2166	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
2167		inet_reset_saddr(sk);
2168		if (sk->sk_prot->rehash &&
2169		    (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2170			sk->sk_prot->rehash(sk);
2171	}
2172
2173	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
2174		sk->sk_prot->unhash(sk);
2175		inet->inet_sport = 0;
2176	}
2177	sk_dst_reset(sk);
2178	return 0;
2179}
2180EXPORT_SYMBOL(__udp_disconnect);
2181
2182int udp_disconnect(struct sock *sk, int flags)
2183{
2184	lock_sock(sk);
2185	__udp_disconnect(sk, flags);
2186	release_sock(sk);
2187	return 0;
2188}
2189EXPORT_SYMBOL(udp_disconnect);
2190
2191void udp_lib_unhash(struct sock *sk)
2192{
2193	if (sk_hashed(sk)) {
2194		struct udp_table *udptable = udp_get_table_prot(sk);
2195		struct udp_hslot *hslot, *hslot2;
2196
2197		hslot  = udp_hashslot(udptable, sock_net(sk),
2198				      udp_sk(sk)->udp_port_hash);
2199		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2200
2201		spin_lock_bh(&hslot->lock);
2202		if (rcu_access_pointer(sk->sk_reuseport_cb))
2203			reuseport_detach_sock(sk);
2204		if (sk_del_node_init_rcu(sk)) {
2205			hslot->count--;
2206			inet_sk(sk)->inet_num = 0;
2207			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2208
2209			spin_lock(&hslot2->lock);
2210			hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2211			hslot2->count--;
2212			spin_unlock(&hslot2->lock);
2213
2214			udp_unhash4(udptable, sk);
2215		}
2216		spin_unlock_bh(&hslot->lock);
2217	}
2218}
2219EXPORT_SYMBOL(udp_lib_unhash);
2220
2221/*
2222 * inet_rcv_saddr was changed, we must rehash secondary hash
2223 */
2224void udp_lib_rehash(struct sock *sk, u16 newhash, u16 newhash4)
2225{
2226	if (sk_hashed(sk)) {
2227		struct udp_table *udptable = udp_get_table_prot(sk);
2228		struct udp_hslot *hslot, *hslot2, *nhslot2;
2229
2230		hslot = udp_hashslot(udptable, sock_net(sk),
2231				     udp_sk(sk)->udp_port_hash);
2232		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2233		nhslot2 = udp_hashslot2(udptable, newhash);
2234		udp_sk(sk)->udp_portaddr_hash = newhash;
2235
2236		if (hslot2 != nhslot2 ||
2237		    rcu_access_pointer(sk->sk_reuseport_cb)) {
 
 
2238			/* we must lock primary chain too */
2239			spin_lock_bh(&hslot->lock);
2240			if (rcu_access_pointer(sk->sk_reuseport_cb))
2241				reuseport_detach_sock(sk);
2242
2243			if (hslot2 != nhslot2) {
2244				spin_lock(&hslot2->lock);
2245				hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2246				hslot2->count--;
2247				spin_unlock(&hslot2->lock);
2248
2249				spin_lock(&nhslot2->lock);
2250				hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2251							 &nhslot2->head);
2252				nhslot2->count++;
2253				spin_unlock(&nhslot2->lock);
2254			}
2255
2256			spin_unlock_bh(&hslot->lock);
2257		}
2258
2259		/* Now process hash4 if necessary:
2260		 * (1) update hslot4;
2261		 * (2) update hslot2->hash4_cnt.
2262		 * Note that hslot2/hslot4 should be checked separately, as
2263		 * either of them may change with the other unchanged.
2264		 */
2265		if (udp_hashed4(sk)) {
2266			spin_lock_bh(&hslot->lock);
2267
2268			udp_rehash4(udptable, sk, newhash4);
2269			if (hslot2 != nhslot2) {
2270				spin_lock(&hslot2->lock);
2271				udp_hash4_dec(hslot2);
2272				spin_unlock(&hslot2->lock);
2273
2274				spin_lock(&nhslot2->lock);
2275				udp_hash4_inc(nhslot2);
2276				spin_unlock(&nhslot2->lock);
2277			}
2278
2279			spin_unlock_bh(&hslot->lock);
2280		}
2281	}
2282}
2283EXPORT_SYMBOL(udp_lib_rehash);
2284
2285void udp_v4_rehash(struct sock *sk)
2286{
2287	u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2288					  inet_sk(sk)->inet_rcv_saddr,
2289					  inet_sk(sk)->inet_num);
2290	u16 new_hash4 = udp_ehashfn(sock_net(sk),
2291				    sk->sk_rcv_saddr, sk->sk_num,
2292				    sk->sk_daddr, sk->sk_dport);
2293
2294	udp_lib_rehash(sk, new_hash, new_hash4);
2295}
2296
2297static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2298{
2299	int rc;
2300
2301	if (inet_sk(sk)->inet_daddr) {
2302		sock_rps_save_rxhash(sk, skb);
2303		sk_mark_napi_id(sk, skb);
2304		sk_incoming_cpu_update(sk);
2305	} else {
2306		sk_mark_napi_id_once(sk, skb);
2307	}
2308
2309	rc = __udp_enqueue_schedule_skb(sk, skb);
2310	if (rc < 0) {
2311		int is_udplite = IS_UDPLITE(sk);
2312		int drop_reason;
2313
2314		/* Note that an ENOMEM error is charged twice */
2315		if (rc == -ENOMEM) {
2316			UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2317					is_udplite);
2318			drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2319		} else {
2320			UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2321				      is_udplite);
2322			drop_reason = SKB_DROP_REASON_PROTO_MEM;
2323		}
2324		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2325		trace_udp_fail_queue_rcv_skb(rc, sk, skb);
2326		sk_skb_reason_drop(sk, skb, drop_reason);
2327		return -1;
2328	}
2329
2330	return 0;
2331}
2332
2333/* returns:
2334 *  -1: error
2335 *   0: success
2336 *  >0: "udp encap" protocol resubmission
2337 *
2338 * Note that in the success and error cases, the skb is assumed to
2339 * have either been requeued or freed.
2340 */
2341static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2342{
2343	int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2344	struct udp_sock *up = udp_sk(sk);
2345	int is_udplite = IS_UDPLITE(sk);
2346
2347	/*
2348	 *	Charge it to the socket, dropping if the queue is full.
2349	 */
2350	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2351		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2352		goto drop;
2353	}
2354	nf_reset_ct(skb);
2355
2356	if (static_branch_unlikely(&udp_encap_needed_key) &&
2357	    READ_ONCE(up->encap_type)) {
2358		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2359
2360		/*
2361		 * This is an encapsulation socket so pass the skb to
2362		 * the socket's udp_encap_rcv() hook. Otherwise, just
2363		 * fall through and pass this up the UDP socket.
2364		 * up->encap_rcv() returns the following value:
2365		 * =0 if skb was successfully passed to the encap
2366		 *    handler or was discarded by it.
2367		 * >0 if skb should be passed on to UDP.
2368		 * <0 if skb should be resubmitted as proto -N
2369		 */
2370
2371		/* if we're overly short, let UDP handle it */
2372		encap_rcv = READ_ONCE(up->encap_rcv);
2373		if (encap_rcv) {
2374			int ret;
2375
2376			/* Verify checksum before giving to encap */
2377			if (udp_lib_checksum_complete(skb))
2378				goto csum_error;
2379
2380			ret = encap_rcv(sk, skb);
2381			if (ret <= 0) {
2382				__UDP_INC_STATS(sock_net(sk),
2383						UDP_MIB_INDATAGRAMS,
2384						is_udplite);
2385				return -ret;
2386			}
2387		}
2388
2389		/* FALLTHROUGH -- it's a UDP Packet */
2390	}
2391
2392	/*
2393	 * 	UDP-Lite specific tests, ignored on UDP sockets
2394	 */
2395	if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2396		u16 pcrlen = READ_ONCE(up->pcrlen);
2397
2398		/*
2399		 * MIB statistics other than incrementing the error count are
2400		 * disabled for the following two types of errors: these depend
2401		 * on the application settings, not on the functioning of the
2402		 * protocol stack as such.
2403		 *
2404		 * RFC 3828 here recommends (sec 3.3): "There should also be a
2405		 * way ... to ... at least let the receiving application block
2406		 * delivery of packets with coverage values less than a value
2407		 * provided by the application."
2408		 */
2409		if (pcrlen == 0) {          /* full coverage was set  */
2410			net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2411					    UDP_SKB_CB(skb)->cscov, skb->len);
2412			goto drop;
2413		}
2414		/* The next case involves violating the min. coverage requested
2415		 * by the receiver. This is subtle: if receiver wants x and x is
2416		 * greater than the buffersize/MTU then receiver will complain
2417		 * that it wants x while sender emits packets of smaller size y.
2418		 * Therefore the above ...()->partial_cov statement is essential.
2419		 */
2420		if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2421			net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2422					    UDP_SKB_CB(skb)->cscov, pcrlen);
2423			goto drop;
2424		}
2425	}
2426
2427	prefetch(&sk->sk_rmem_alloc);
2428	if (rcu_access_pointer(sk->sk_filter) &&
2429	    udp_lib_checksum_complete(skb))
2430			goto csum_error;
2431
2432	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2433		drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2434		goto drop;
2435	}
2436
2437	udp_csum_pull_header(skb);
2438
2439	ipv4_pktinfo_prepare(sk, skb, true);
2440	return __udp_queue_rcv_skb(sk, skb);
2441
2442csum_error:
2443	drop_reason = SKB_DROP_REASON_UDP_CSUM;
2444	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2445drop:
2446	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2447	atomic_inc(&sk->sk_drops);
2448	sk_skb_reason_drop(sk, skb, drop_reason);
2449	return -1;
2450}
2451
2452static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2453{
2454	struct sk_buff *next, *segs;
2455	int ret;
2456
2457	if (likely(!udp_unexpected_gso(sk, skb)))
2458		return udp_queue_rcv_one_skb(sk, skb);
2459
2460	BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2461	__skb_push(skb, -skb_mac_offset(skb));
2462	segs = udp_rcv_segment(sk, skb, true);
2463	skb_list_walk_safe(segs, skb, next) {
2464		__skb_pull(skb, skb_transport_offset(skb));
2465
2466		udp_post_segment_fix_csum(skb);
2467		ret = udp_queue_rcv_one_skb(sk, skb);
2468		if (ret > 0)
2469			ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2470	}
2471	return 0;
2472}
2473
2474/* For TCP sockets, sk_rx_dst is protected by socket lock
2475 * For UDP, we use xchg() to guard against concurrent changes.
2476 */
2477bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2478{
2479	struct dst_entry *old;
2480
2481	if (dst_hold_safe(dst)) {
2482		old = unrcu_pointer(xchg(&sk->sk_rx_dst, RCU_INITIALIZER(dst)));
2483		dst_release(old);
2484		return old != dst;
2485	}
2486	return false;
2487}
2488EXPORT_SYMBOL(udp_sk_rx_dst_set);
2489
2490/*
2491 *	Multicasts and broadcasts go to each listener.
2492 *
2493 *	Note: called only from the BH handler context.
2494 */
2495static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2496				    struct udphdr  *uh,
2497				    __be32 saddr, __be32 daddr,
2498				    struct udp_table *udptable,
2499				    int proto)
2500{
2501	struct sock *sk, *first = NULL;
2502	unsigned short hnum = ntohs(uh->dest);
2503	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2504	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2505	unsigned int offset = offsetof(typeof(*sk), sk_node);
2506	int dif = skb->dev->ifindex;
2507	int sdif = inet_sdif(skb);
2508	struct hlist_node *node;
2509	struct sk_buff *nskb;
2510
2511	if (use_hash2) {
2512		hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2513			    udptable->mask;
2514		hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2515start_lookup:
2516		hslot = &udptable->hash2[hash2].hslot;
2517		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2518	}
2519
2520	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2521		if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2522					 uh->source, saddr, dif, sdif, hnum))
2523			continue;
2524
2525		if (!first) {
2526			first = sk;
2527			continue;
2528		}
2529		nskb = skb_clone(skb, GFP_ATOMIC);
2530
2531		if (unlikely(!nskb)) {
2532			atomic_inc(&sk->sk_drops);
2533			__UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2534					IS_UDPLITE(sk));
2535			__UDP_INC_STATS(net, UDP_MIB_INERRORS,
2536					IS_UDPLITE(sk));
2537			continue;
2538		}
2539		if (udp_queue_rcv_skb(sk, nskb) > 0)
2540			consume_skb(nskb);
2541	}
2542
2543	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
2544	if (use_hash2 && hash2 != hash2_any) {
2545		hash2 = hash2_any;
2546		goto start_lookup;
2547	}
2548
2549	if (first) {
2550		if (udp_queue_rcv_skb(first, skb) > 0)
2551			consume_skb(skb);
2552	} else {
2553		kfree_skb(skb);
2554		__UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2555				proto == IPPROTO_UDPLITE);
2556	}
2557	return 0;
2558}
2559
2560/* Initialize UDP checksum. If exited with zero value (success),
2561 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2562 * Otherwise, csum completion requires checksumming packet body,
2563 * including udp header and folding it to skb->csum.
2564 */
2565static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2566				 int proto)
2567{
2568	int err;
2569
2570	UDP_SKB_CB(skb)->partial_cov = 0;
2571	UDP_SKB_CB(skb)->cscov = skb->len;
2572
2573	if (proto == IPPROTO_UDPLITE) {
2574		err = udplite_checksum_init(skb, uh);
2575		if (err)
2576			return err;
2577
2578		if (UDP_SKB_CB(skb)->partial_cov) {
2579			skb->csum = inet_compute_pseudo(skb, proto);
2580			return 0;
2581		}
2582	}
2583
2584	/* Note, we are only interested in != 0 or == 0, thus the
2585	 * force to int.
2586	 */
2587	err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2588							inet_compute_pseudo);
2589	if (err)
2590		return err;
2591
2592	if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2593		/* If SW calculated the value, we know it's bad */
2594		if (skb->csum_complete_sw)
2595			return 1;
2596
2597		/* HW says the value is bad. Let's validate that.
2598		 * skb->csum is no longer the full packet checksum,
2599		 * so don't treat it as such.
2600		 */
2601		skb_checksum_complete_unset(skb);
2602	}
2603
2604	return 0;
2605}
2606
2607/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2608 * return code conversion for ip layer consumption
2609 */
2610static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2611			       struct udphdr *uh)
2612{
2613	int ret;
2614
2615	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2616		skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2617
2618	ret = udp_queue_rcv_skb(sk, skb);
2619
2620	/* a return value > 0 means to resubmit the input, but
2621	 * it wants the return to be -protocol, or 0
2622	 */
2623	if (ret > 0)
2624		return -ret;
2625	return 0;
2626}
2627
2628/*
2629 *	All we need to do is get the socket, and then do a checksum.
2630 */
2631
2632int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2633		   int proto)
2634{
2635	struct sock *sk = NULL;
2636	struct udphdr *uh;
2637	unsigned short ulen;
2638	struct rtable *rt = skb_rtable(skb);
2639	__be32 saddr, daddr;
2640	struct net *net = dev_net(skb->dev);
2641	bool refcounted;
2642	int drop_reason;
2643
2644	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2645
2646	/*
2647	 *  Validate the packet.
2648	 */
2649	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2650		goto drop;		/* No space for header. */
2651
2652	uh   = udp_hdr(skb);
2653	ulen = ntohs(uh->len);
2654	saddr = ip_hdr(skb)->saddr;
2655	daddr = ip_hdr(skb)->daddr;
2656
2657	if (ulen > skb->len)
2658		goto short_packet;
2659
2660	if (proto == IPPROTO_UDP) {
2661		/* UDP validates ulen. */
2662		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2663			goto short_packet;
2664		uh = udp_hdr(skb);
2665	}
2666
2667	if (udp4_csum_init(skb, uh, proto))
2668		goto csum_error;
2669
2670	sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
2671			     &refcounted, udp_ehashfn);
2672	if (IS_ERR(sk))
2673		goto no_sk;
2674
2675	if (sk) {
2676		struct dst_entry *dst = skb_dst(skb);
2677		int ret;
2678
2679		if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2680			udp_sk_rx_dst_set(sk, dst);
2681
2682		ret = udp_unicast_rcv_skb(sk, skb, uh);
2683		if (refcounted)
2684			sock_put(sk);
2685		return ret;
2686	}
2687
2688	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2689		return __udp4_lib_mcast_deliver(net, skb, uh,
2690						saddr, daddr, udptable, proto);
2691
2692	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2693	if (sk)
2694		return udp_unicast_rcv_skb(sk, skb, uh);
2695no_sk:
2696	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2697		goto drop;
2698	nf_reset_ct(skb);
2699
2700	/* No socket. Drop packet silently, if checksum is wrong */
2701	if (udp_lib_checksum_complete(skb))
2702		goto csum_error;
2703
2704	drop_reason = SKB_DROP_REASON_NO_SOCKET;
2705	__UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2706	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2707
2708	/*
2709	 * Hmm.  We got an UDP packet to a port to which we
2710	 * don't wanna listen.  Ignore it.
2711	 */
2712	sk_skb_reason_drop(sk, skb, drop_reason);
2713	return 0;
2714
2715short_packet:
2716	drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2717	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2718			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2719			    &saddr, ntohs(uh->source),
2720			    ulen, skb->len,
2721			    &daddr, ntohs(uh->dest));
2722	goto drop;
2723
2724csum_error:
2725	/*
2726	 * RFC1122: OK.  Discards the bad packet silently (as far as
2727	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2728	 */
2729	drop_reason = SKB_DROP_REASON_UDP_CSUM;
2730	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2731			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2732			    &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2733			    ulen);
2734	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2735drop:
2736	__UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2737	sk_skb_reason_drop(sk, skb, drop_reason);
2738	return 0;
2739}
2740
2741/* We can only early demux multicast if there is a single matching socket.
2742 * If more than one socket found returns NULL
2743 */
2744static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2745						  __be16 loc_port, __be32 loc_addr,
2746						  __be16 rmt_port, __be32 rmt_addr,
2747						  int dif, int sdif)
2748{
2749	struct udp_table *udptable = net->ipv4.udp_table;
2750	unsigned short hnum = ntohs(loc_port);
2751	struct sock *sk, *result;
2752	struct udp_hslot *hslot;
2753	unsigned int slot;
2754
2755	slot = udp_hashfn(net, hnum, udptable->mask);
2756	hslot = &udptable->hash[slot];
2757
2758	/* Do not bother scanning a too big list */
2759	if (hslot->count > 10)
2760		return NULL;
2761
2762	result = NULL;
2763	sk_for_each_rcu(sk, &hslot->head) {
2764		if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2765					rmt_port, rmt_addr, dif, sdif, hnum)) {
2766			if (result)
2767				return NULL;
2768			result = sk;
2769		}
2770	}
2771
2772	return result;
2773}
2774
2775/* For unicast we should only early demux connected sockets or we can
2776 * break forwarding setups.  The chains here can be long so only check
2777 * if the first socket is an exact match and if not move on.
2778 */
2779static struct sock *__udp4_lib_demux_lookup(struct net *net,
2780					    __be16 loc_port, __be32 loc_addr,
2781					    __be16 rmt_port, __be32 rmt_addr,
2782					    int dif, int sdif)
2783{
2784	struct udp_table *udptable = net->ipv4.udp_table;
2785	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2786	unsigned short hnum = ntohs(loc_port);
 
2787	struct udp_hslot *hslot2;
2788	unsigned int hash2;
2789	__portpair ports;
2790	struct sock *sk;
2791
2792	hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2793	hslot2 = udp_hashslot2(udptable, hash2);
 
2794	ports = INET_COMBINED_PORTS(rmt_port, hnum);
2795
2796	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2797		if (inet_match(net, sk, acookie, ports, dif, sdif))
2798			return sk;
2799		/* Only check first socket in chain */
2800		break;
2801	}
2802	return NULL;
2803}
2804
2805int udp_v4_early_demux(struct sk_buff *skb)
2806{
2807	struct net *net = dev_net(skb->dev);
2808	struct in_device *in_dev = NULL;
2809	const struct iphdr *iph;
2810	const struct udphdr *uh;
2811	struct sock *sk = NULL;
2812	struct dst_entry *dst;
2813	int dif = skb->dev->ifindex;
2814	int sdif = inet_sdif(skb);
2815	int ours;
2816
2817	/* validate the packet */
2818	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2819		return 0;
2820
2821	iph = ip_hdr(skb);
2822	uh = udp_hdr(skb);
2823
2824	if (skb->pkt_type == PACKET_MULTICAST) {
2825		in_dev = __in_dev_get_rcu(skb->dev);
2826
2827		if (!in_dev)
2828			return 0;
2829
2830		ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2831				       iph->protocol);
2832		if (!ours)
2833			return 0;
2834
2835		sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2836						   uh->source, iph->saddr,
2837						   dif, sdif);
2838	} else if (skb->pkt_type == PACKET_HOST) {
2839		sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2840					     uh->source, iph->saddr, dif, sdif);
2841	}
2842
2843	if (!sk)
2844		return 0;
2845
2846	skb->sk = sk;
2847	DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
2848	skb->destructor = sock_pfree;
2849	dst = rcu_dereference(sk->sk_rx_dst);
2850
2851	if (dst)
2852		dst = dst_check(dst, 0);
2853	if (dst) {
2854		u32 itag = 0;
2855
2856		/* set noref for now.
2857		 * any place which wants to hold dst has to call
2858		 * dst_hold_safe()
2859		 */
2860		skb_dst_set_noref(skb, dst);
2861
2862		/* for unconnected multicast sockets we need to validate
2863		 * the source on each packet
2864		 */
2865		if (!inet_sk(sk)->inet_daddr && in_dev)
2866			return ip_mc_validate_source(skb, iph->daddr,
2867						     iph->saddr,
2868						     ip4h_dscp(iph),
2869						     skb->dev, in_dev, &itag);
2870	}
2871	return 0;
2872}
2873
2874int udp_rcv(struct sk_buff *skb)
2875{
2876	return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2877}
2878
2879void udp_destroy_sock(struct sock *sk)
2880{
2881	struct udp_sock *up = udp_sk(sk);
2882	bool slow = lock_sock_fast(sk);
2883
2884	/* protects from races with udp_abort() */
2885	sock_set_flag(sk, SOCK_DEAD);
2886	udp_flush_pending_frames(sk);
2887	unlock_sock_fast(sk, slow);
2888	if (static_branch_unlikely(&udp_encap_needed_key)) {
2889		if (up->encap_type) {
2890			void (*encap_destroy)(struct sock *sk);
2891			encap_destroy = READ_ONCE(up->encap_destroy);
2892			if (encap_destroy)
2893				encap_destroy(sk);
2894		}
2895		if (udp_test_bit(ENCAP_ENABLED, sk))
2896			static_branch_dec(&udp_encap_needed_key);
2897	}
2898}
2899
2900static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2901				       struct sock *sk)
2902{
2903#ifdef CONFIG_XFRM
2904	if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2905		if (family == AF_INET)
2906			WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv);
2907		else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2908			WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv);
2909	}
2910#endif
2911}
2912
2913/*
2914 *	Socket option code for UDP
2915 */
2916int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2917		       sockptr_t optval, unsigned int optlen,
2918		       int (*push_pending_frames)(struct sock *))
2919{
2920	struct udp_sock *up = udp_sk(sk);
2921	int val, valbool;
2922	int err = 0;
2923	int is_udplite = IS_UDPLITE(sk);
2924
2925	if (level == SOL_SOCKET) {
2926		err = sk_setsockopt(sk, level, optname, optval, optlen);
2927
2928		if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2929			sockopt_lock_sock(sk);
2930			/* paired with READ_ONCE in udp_rmem_release() */
2931			WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2932			sockopt_release_sock(sk);
2933		}
2934		return err;
2935	}
2936
2937	if (optlen < sizeof(int))
2938		return -EINVAL;
2939
2940	if (copy_from_sockptr(&val, optval, sizeof(val)))
2941		return -EFAULT;
2942
2943	valbool = val ? 1 : 0;
2944
2945	switch (optname) {
2946	case UDP_CORK:
2947		if (val != 0) {
2948			udp_set_bit(CORK, sk);
2949		} else {
2950			udp_clear_bit(CORK, sk);
2951			lock_sock(sk);
2952			push_pending_frames(sk);
2953			release_sock(sk);
2954		}
2955		break;
2956
2957	case UDP_ENCAP:
2958		switch (val) {
2959		case 0:
2960#ifdef CONFIG_XFRM
2961		case UDP_ENCAP_ESPINUDP:
2962			set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk);
2963#if IS_ENABLED(CONFIG_IPV6)
2964			if (sk->sk_family == AF_INET6)
2965				WRITE_ONCE(up->encap_rcv,
2966					   ipv6_stub->xfrm6_udp_encap_rcv);
2967			else
2968#endif
2969				WRITE_ONCE(up->encap_rcv,
2970					   xfrm4_udp_encap_rcv);
2971#endif
2972			fallthrough;
2973		case UDP_ENCAP_L2TPINUDP:
2974			WRITE_ONCE(up->encap_type, val);
2975			udp_tunnel_encap_enable(sk);
 
 
2976			break;
2977		default:
2978			err = -ENOPROTOOPT;
2979			break;
2980		}
2981		break;
2982
2983	case UDP_NO_CHECK6_TX:
2984		udp_set_no_check6_tx(sk, valbool);
2985		break;
2986
2987	case UDP_NO_CHECK6_RX:
2988		udp_set_no_check6_rx(sk, valbool);
2989		break;
2990
2991	case UDP_SEGMENT:
2992		if (val < 0 || val > USHRT_MAX)
2993			return -EINVAL;
2994		WRITE_ONCE(up->gso_size, val);
2995		break;
2996
2997	case UDP_GRO:
 
2998
2999		/* when enabling GRO, accept the related GSO packet type */
3000		if (valbool)
3001			udp_tunnel_encap_enable(sk);
3002		udp_assign_bit(GRO_ENABLED, sk, valbool);
3003		udp_assign_bit(ACCEPT_L4, sk, valbool);
3004		set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk);
3005		break;
3006
3007	/*
3008	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
3009	 */
3010	/* The sender sets actual checksum coverage length via this option.
3011	 * The case coverage > packet length is handled by send module. */
3012	case UDPLITE_SEND_CSCOV:
3013		if (!is_udplite)         /* Disable the option on UDP sockets */
3014			return -ENOPROTOOPT;
3015		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
3016			val = 8;
3017		else if (val > USHRT_MAX)
3018			val = USHRT_MAX;
3019		WRITE_ONCE(up->pcslen, val);
3020		udp_set_bit(UDPLITE_SEND_CC, sk);
3021		break;
3022
3023	/* The receiver specifies a minimum checksum coverage value. To make
3024	 * sense, this should be set to at least 8 (as done below). If zero is
3025	 * used, this again means full checksum coverage.                     */
3026	case UDPLITE_RECV_CSCOV:
3027		if (!is_udplite)         /* Disable the option on UDP sockets */
3028			return -ENOPROTOOPT;
3029		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
3030			val = 8;
3031		else if (val > USHRT_MAX)
3032			val = USHRT_MAX;
3033		WRITE_ONCE(up->pcrlen, val);
3034		udp_set_bit(UDPLITE_RECV_CC, sk);
3035		break;
3036
3037	default:
3038		err = -ENOPROTOOPT;
3039		break;
3040	}
3041
3042	return err;
3043}
3044EXPORT_SYMBOL(udp_lib_setsockopt);
3045
3046int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
3047		   unsigned int optlen)
3048{
3049	if (level == SOL_UDP  ||  level == SOL_UDPLITE || level == SOL_SOCKET)
3050		return udp_lib_setsockopt(sk, level, optname,
3051					  optval, optlen,
3052					  udp_push_pending_frames);
3053	return ip_setsockopt(sk, level, optname, optval, optlen);
3054}
3055
3056int udp_lib_getsockopt(struct sock *sk, int level, int optname,
3057		       char __user *optval, int __user *optlen)
3058{
3059	struct udp_sock *up = udp_sk(sk);
3060	int val, len;
3061
3062	if (get_user(len, optlen))
3063		return -EFAULT;
3064
 
 
3065	if (len < 0)
3066		return -EINVAL;
3067
3068	len = min_t(unsigned int, len, sizeof(int));
3069
3070	switch (optname) {
3071	case UDP_CORK:
3072		val = udp_test_bit(CORK, sk);
3073		break;
3074
3075	case UDP_ENCAP:
3076		val = READ_ONCE(up->encap_type);
3077		break;
3078
3079	case UDP_NO_CHECK6_TX:
3080		val = udp_get_no_check6_tx(sk);
3081		break;
3082
3083	case UDP_NO_CHECK6_RX:
3084		val = udp_get_no_check6_rx(sk);
3085		break;
3086
3087	case UDP_SEGMENT:
3088		val = READ_ONCE(up->gso_size);
3089		break;
3090
3091	case UDP_GRO:
3092		val = udp_test_bit(GRO_ENABLED, sk);
3093		break;
3094
3095	/* The following two cannot be changed on UDP sockets, the return is
3096	 * always 0 (which corresponds to the full checksum coverage of UDP). */
3097	case UDPLITE_SEND_CSCOV:
3098		val = READ_ONCE(up->pcslen);
3099		break;
3100
3101	case UDPLITE_RECV_CSCOV:
3102		val = READ_ONCE(up->pcrlen);
3103		break;
3104
3105	default:
3106		return -ENOPROTOOPT;
3107	}
3108
3109	if (put_user(len, optlen))
3110		return -EFAULT;
3111	if (copy_to_user(optval, &val, len))
3112		return -EFAULT;
3113	return 0;
3114}
3115EXPORT_SYMBOL(udp_lib_getsockopt);
3116
3117int udp_getsockopt(struct sock *sk, int level, int optname,
3118		   char __user *optval, int __user *optlen)
3119{
3120	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
3121		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
3122	return ip_getsockopt(sk, level, optname, optval, optlen);
3123}
3124
3125/**
3126 * 	udp_poll - wait for a UDP event.
3127 *	@file: - file struct
3128 *	@sock: - socket
3129 *	@wait: - poll table
3130 *
3131 *	This is same as datagram poll, except for the special case of
3132 *	blocking sockets. If application is using a blocking fd
3133 *	and a packet with checksum error is in the queue;
3134 *	then it could get return from select indicating data available
3135 *	but then block when reading it. Add special case code
3136 *	to work around these arguably broken applications.
3137 */
3138__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
3139{
3140	__poll_t mask = datagram_poll(file, sock, wait);
3141	struct sock *sk = sock->sk;
3142
3143	if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
3144		mask |= EPOLLIN | EPOLLRDNORM;
3145
3146	/* Check for false positives due to checksum errors */
3147	if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
3148	    !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
3149		mask &= ~(EPOLLIN | EPOLLRDNORM);
3150
3151	/* psock ingress_msg queue should not contain any bad checksum frames */
3152	if (sk_is_readable(sk))
3153		mask |= EPOLLIN | EPOLLRDNORM;
3154	return mask;
3155
3156}
3157EXPORT_SYMBOL(udp_poll);
3158
3159int udp_abort(struct sock *sk, int err)
3160{
3161	if (!has_current_bpf_ctx())
3162		lock_sock(sk);
3163
3164	/* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
3165	 * with close()
3166	 */
3167	if (sock_flag(sk, SOCK_DEAD))
3168		goto out;
3169
3170	sk->sk_err = err;
3171	sk_error_report(sk);
3172	__udp_disconnect(sk, 0);
3173
3174out:
3175	if (!has_current_bpf_ctx())
3176		release_sock(sk);
3177
3178	return 0;
3179}
3180EXPORT_SYMBOL_GPL(udp_abort);
3181
3182struct proto udp_prot = {
3183	.name			= "UDP",
3184	.owner			= THIS_MODULE,
3185	.close			= udp_lib_close,
3186	.pre_connect		= udp_pre_connect,
3187	.connect		= udp_connect,
3188	.disconnect		= udp_disconnect,
3189	.ioctl			= udp_ioctl,
3190	.init			= udp_init_sock,
3191	.destroy		= udp_destroy_sock,
3192	.setsockopt		= udp_setsockopt,
3193	.getsockopt		= udp_getsockopt,
3194	.sendmsg		= udp_sendmsg,
3195	.recvmsg		= udp_recvmsg,
3196	.splice_eof		= udp_splice_eof,
3197	.release_cb		= ip4_datagram_release_cb,
3198	.hash			= udp_lib_hash,
3199	.unhash			= udp_lib_unhash,
3200	.rehash			= udp_v4_rehash,
3201	.get_port		= udp_v4_get_port,
3202	.put_port		= udp_lib_unhash,
3203#ifdef CONFIG_BPF_SYSCALL
3204	.psock_update_sk_prot	= udp_bpf_update_proto,
3205#endif
3206	.memory_allocated	= &udp_memory_allocated,
3207	.per_cpu_fw_alloc	= &udp_memory_per_cpu_fw_alloc,
3208
3209	.sysctl_mem		= sysctl_udp_mem,
3210	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_udp_wmem_min),
3211	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_udp_rmem_min),
3212	.obj_size		= sizeof(struct udp_sock),
3213	.h.udp_table		= NULL,
3214	.diag_destroy		= udp_abort,
3215};
3216EXPORT_SYMBOL(udp_prot);
3217
3218/* ------------------------------------------------------------------------ */
3219#ifdef CONFIG_PROC_FS
3220
3221static unsigned short seq_file_family(const struct seq_file *seq);
3222static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
3223{
3224	unsigned short family = seq_file_family(seq);
3225
3226	/* AF_UNSPEC is used as a match all */
3227	return ((family == AF_UNSPEC || family == sk->sk_family) &&
3228		net_eq(sock_net(sk), seq_file_net(seq)));
3229}
3230
3231#ifdef CONFIG_BPF_SYSCALL
3232static const struct seq_operations bpf_iter_udp_seq_ops;
3233#endif
3234static struct udp_table *udp_get_table_seq(struct seq_file *seq,
3235					   struct net *net)
3236{
3237	const struct udp_seq_afinfo *afinfo;
3238
3239#ifdef CONFIG_BPF_SYSCALL
3240	if (seq->op == &bpf_iter_udp_seq_ops)
3241		return net->ipv4.udp_table;
3242#endif
3243
3244	afinfo = pde_data(file_inode(seq->file));
3245	return afinfo->udp_table ? : net->ipv4.udp_table;
3246}
3247
3248static struct sock *udp_get_first(struct seq_file *seq, int start)
3249{
3250	struct udp_iter_state *state = seq->private;
3251	struct net *net = seq_file_net(seq);
 
3252	struct udp_table *udptable;
3253	struct sock *sk;
3254
3255	udptable = udp_get_table_seq(seq, net);
 
 
 
 
 
3256
3257	for (state->bucket = start; state->bucket <= udptable->mask;
3258	     ++state->bucket) {
3259		struct udp_hslot *hslot = &udptable->hash[state->bucket];
3260
3261		if (hlist_empty(&hslot->head))
3262			continue;
3263
3264		spin_lock_bh(&hslot->lock);
3265		sk_for_each(sk, &hslot->head) {
3266			if (seq_sk_match(seq, sk))
 
 
 
3267				goto found;
3268		}
3269		spin_unlock_bh(&hslot->lock);
3270	}
3271	sk = NULL;
3272found:
3273	return sk;
3274}
3275
3276static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3277{
3278	struct udp_iter_state *state = seq->private;
3279	struct net *net = seq_file_net(seq);
 
3280	struct udp_table *udptable;
3281
 
 
 
 
 
3282	do {
3283		sk = sk_next(sk);
3284	} while (sk && !seq_sk_match(seq, sk));
 
 
3285
3286	if (!sk) {
3287		udptable = udp_get_table_seq(seq, net);
3288
3289		if (state->bucket <= udptable->mask)
3290			spin_unlock_bh(&udptable->hash[state->bucket].lock);
3291
3292		return udp_get_first(seq, state->bucket + 1);
3293	}
3294	return sk;
3295}
3296
3297static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3298{
3299	struct sock *sk = udp_get_first(seq, 0);
3300
3301	if (sk)
3302		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3303			--pos;
3304	return pos ? NULL : sk;
3305}
3306
3307void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3308{
3309	struct udp_iter_state *state = seq->private;
3310	state->bucket = MAX_UDP_PORTS;
3311
3312	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3313}
3314EXPORT_SYMBOL(udp_seq_start);
3315
3316void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3317{
3318	struct sock *sk;
3319
3320	if (v == SEQ_START_TOKEN)
3321		sk = udp_get_idx(seq, 0);
3322	else
3323		sk = udp_get_next(seq, v);
3324
3325	++*pos;
3326	return sk;
3327}
3328EXPORT_SYMBOL(udp_seq_next);
3329
3330void udp_seq_stop(struct seq_file *seq, void *v)
3331{
3332	struct udp_iter_state *state = seq->private;
 
3333	struct udp_table *udptable;
3334
3335	udptable = udp_get_table_seq(seq, seq_file_net(seq));
 
 
 
 
 
3336
3337	if (state->bucket <= udptable->mask)
3338		spin_unlock_bh(&udptable->hash[state->bucket].lock);
3339}
3340EXPORT_SYMBOL(udp_seq_stop);
3341
3342/* ------------------------------------------------------------------------ */
3343static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3344		int bucket)
3345{
3346	struct inet_sock *inet = inet_sk(sp);
3347	__be32 dest = inet->inet_daddr;
3348	__be32 src  = inet->inet_rcv_saddr;
3349	__u16 destp	  = ntohs(inet->inet_dport);
3350	__u16 srcp	  = ntohs(inet->inet_sport);
3351
3352	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3353		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3354		bucket, src, srcp, dest, destp, sp->sk_state,
3355		sk_wmem_alloc_get(sp),
3356		udp_rqueue_get(sp),
3357		0, 0L, 0,
3358		from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3359		0, sock_i_ino(sp),
3360		refcount_read(&sp->sk_refcnt), sp,
3361		atomic_read(&sp->sk_drops));
3362}
3363
3364int udp4_seq_show(struct seq_file *seq, void *v)
3365{
3366	seq_setwidth(seq, 127);
3367	if (v == SEQ_START_TOKEN)
3368		seq_puts(seq, "   sl  local_address rem_address   st tx_queue "
3369			   "rx_queue tr tm->when retrnsmt   uid  timeout "
3370			   "inode ref pointer drops");
3371	else {
3372		struct udp_iter_state *state = seq->private;
3373
3374		udp4_format_sock(v, seq, state->bucket);
3375	}
3376	seq_pad(seq, '\n');
3377	return 0;
3378}
3379
3380#ifdef CONFIG_BPF_SYSCALL
3381struct bpf_iter__udp {
3382	__bpf_md_ptr(struct bpf_iter_meta *, meta);
3383	__bpf_md_ptr(struct udp_sock *, udp_sk);
3384	uid_t uid __aligned(8);
3385	int bucket __aligned(8);
3386};
3387
3388struct bpf_udp_iter_state {
3389	struct udp_iter_state state;
3390	unsigned int cur_sk;
3391	unsigned int end_sk;
3392	unsigned int max_sk;
3393	int offset;
3394	struct sock **batch;
3395	bool st_bucket_done;
3396};
3397
3398static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3399				      unsigned int new_batch_sz);
3400static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3401{
3402	struct bpf_udp_iter_state *iter = seq->private;
3403	struct udp_iter_state *state = &iter->state;
3404	struct net *net = seq_file_net(seq);
3405	int resume_bucket, resume_offset;
3406	struct udp_table *udptable;
3407	unsigned int batch_sks = 0;
3408	bool resized = false;
3409	struct sock *sk;
3410
3411	resume_bucket = state->bucket;
3412	resume_offset = iter->offset;
3413
3414	/* The current batch is done, so advance the bucket. */
3415	if (iter->st_bucket_done)
3416		state->bucket++;
3417
3418	udptable = udp_get_table_seq(seq, net);
3419
3420again:
3421	/* New batch for the next bucket.
3422	 * Iterate over the hash table to find a bucket with sockets matching
3423	 * the iterator attributes, and return the first matching socket from
3424	 * the bucket. The remaining matched sockets from the bucket are batched
3425	 * before releasing the bucket lock. This allows BPF programs that are
3426	 * called in seq_show to acquire the bucket lock if needed.
3427	 */
3428	iter->cur_sk = 0;
3429	iter->end_sk = 0;
3430	iter->st_bucket_done = false;
3431	batch_sks = 0;
3432
3433	for (; state->bucket <= udptable->mask; state->bucket++) {
3434		struct udp_hslot *hslot2 = &udptable->hash2[state->bucket].hslot;
3435
3436		if (hlist_empty(&hslot2->head))
3437			continue;
3438
3439		iter->offset = 0;
3440		spin_lock_bh(&hslot2->lock);
3441		udp_portaddr_for_each_entry(sk, &hslot2->head) {
3442			if (seq_sk_match(seq, sk)) {
3443				/* Resume from the last iterated socket at the
3444				 * offset in the bucket before iterator was stopped.
3445				 */
3446				if (state->bucket == resume_bucket &&
3447				    iter->offset < resume_offset) {
3448					++iter->offset;
3449					continue;
3450				}
3451				if (iter->end_sk < iter->max_sk) {
3452					sock_hold(sk);
3453					iter->batch[iter->end_sk++] = sk;
3454				}
3455				batch_sks++;
3456			}
3457		}
3458		spin_unlock_bh(&hslot2->lock);
3459
3460		if (iter->end_sk)
3461			break;
3462	}
3463
3464	/* All done: no batch made. */
3465	if (!iter->end_sk)
3466		return NULL;
3467
3468	if (iter->end_sk == batch_sks) {
3469		/* Batching is done for the current bucket; return the first
3470		 * socket to be iterated from the batch.
3471		 */
3472		iter->st_bucket_done = true;
3473		goto done;
3474	}
3475	if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3476		resized = true;
3477		/* After allocating a larger batch, retry one more time to grab
3478		 * the whole bucket.
3479		 */
3480		goto again;
3481	}
3482done:
3483	return iter->batch[0];
3484}
3485
3486static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3487{
3488	struct bpf_udp_iter_state *iter = seq->private;
3489	struct sock *sk;
3490
3491	/* Whenever seq_next() is called, the iter->cur_sk is
3492	 * done with seq_show(), so unref the iter->cur_sk.
3493	 */
3494	if (iter->cur_sk < iter->end_sk) {
3495		sock_put(iter->batch[iter->cur_sk++]);
3496		++iter->offset;
3497	}
3498
3499	/* After updating iter->cur_sk, check if there are more sockets
3500	 * available in the current bucket batch.
3501	 */
3502	if (iter->cur_sk < iter->end_sk)
3503		sk = iter->batch[iter->cur_sk];
3504	else
3505		/* Prepare a new batch. */
3506		sk = bpf_iter_udp_batch(seq);
3507
3508	++*pos;
3509	return sk;
3510}
3511
3512static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3513{
3514	/* bpf iter does not support lseek, so it always
3515	 * continue from where it was stop()-ped.
3516	 */
3517	if (*pos)
3518		return bpf_iter_udp_batch(seq);
3519
3520	return SEQ_START_TOKEN;
3521}
3522
3523static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3524			     struct udp_sock *udp_sk, uid_t uid, int bucket)
3525{
3526	struct bpf_iter__udp ctx;
3527
3528	meta->seq_num--;  /* skip SEQ_START_TOKEN */
3529	ctx.meta = meta;
3530	ctx.udp_sk = udp_sk;
3531	ctx.uid = uid;
3532	ctx.bucket = bucket;
3533	return bpf_iter_run_prog(prog, &ctx);
3534}
3535
3536static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3537{
3538	struct udp_iter_state *state = seq->private;
3539	struct bpf_iter_meta meta;
3540	struct bpf_prog *prog;
3541	struct sock *sk = v;
3542	uid_t uid;
3543	int ret;
3544
3545	if (v == SEQ_START_TOKEN)
3546		return 0;
3547
3548	lock_sock(sk);
3549
3550	if (unlikely(sk_unhashed(sk))) {
3551		ret = SEQ_SKIP;
3552		goto unlock;
3553	}
3554
3555	uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3556	meta.seq = seq;
3557	prog = bpf_iter_get_info(&meta, false);
3558	ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3559
3560unlock:
3561	release_sock(sk);
3562	return ret;
3563}
3564
3565static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3566{
3567	while (iter->cur_sk < iter->end_sk)
3568		sock_put(iter->batch[iter->cur_sk++]);
3569}
3570
3571static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3572{
3573	struct bpf_udp_iter_state *iter = seq->private;
3574	struct bpf_iter_meta meta;
3575	struct bpf_prog *prog;
3576
3577	if (!v) {
3578		meta.seq = seq;
3579		prog = bpf_iter_get_info(&meta, true);
3580		if (prog)
3581			(void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3582	}
3583
3584	if (iter->cur_sk < iter->end_sk) {
3585		bpf_iter_udp_put_batch(iter);
3586		iter->st_bucket_done = false;
3587	}
3588}
3589
3590static const struct seq_operations bpf_iter_udp_seq_ops = {
3591	.start		= bpf_iter_udp_seq_start,
3592	.next		= bpf_iter_udp_seq_next,
3593	.stop		= bpf_iter_udp_seq_stop,
3594	.show		= bpf_iter_udp_seq_show,
3595};
3596#endif
3597
3598static unsigned short seq_file_family(const struct seq_file *seq)
3599{
3600	const struct udp_seq_afinfo *afinfo;
3601
3602#ifdef CONFIG_BPF_SYSCALL
3603	/* BPF iterator: bpf programs to filter sockets. */
3604	if (seq->op == &bpf_iter_udp_seq_ops)
3605		return AF_UNSPEC;
3606#endif
3607
3608	/* Proc fs iterator */
3609	afinfo = pde_data(file_inode(seq->file));
3610	return afinfo->family;
3611}
3612
3613const struct seq_operations udp_seq_ops = {
3614	.start		= udp_seq_start,
3615	.next		= udp_seq_next,
3616	.stop		= udp_seq_stop,
3617	.show		= udp4_seq_show,
3618};
3619EXPORT_SYMBOL(udp_seq_ops);
3620
3621static struct udp_seq_afinfo udp4_seq_afinfo = {
3622	.family		= AF_INET,
3623	.udp_table	= NULL,
3624};
3625
3626static int __net_init udp4_proc_init_net(struct net *net)
3627{
3628	if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3629			sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3630		return -ENOMEM;
3631	return 0;
3632}
3633
3634static void __net_exit udp4_proc_exit_net(struct net *net)
3635{
3636	remove_proc_entry("udp", net->proc_net);
3637}
3638
3639static struct pernet_operations udp4_net_ops = {
3640	.init = udp4_proc_init_net,
3641	.exit = udp4_proc_exit_net,
3642};
3643
3644int __init udp4_proc_init(void)
3645{
3646	return register_pernet_subsys(&udp4_net_ops);
3647}
3648
3649void udp4_proc_exit(void)
3650{
3651	unregister_pernet_subsys(&udp4_net_ops);
3652}
3653#endif /* CONFIG_PROC_FS */
3654
3655static __initdata unsigned long uhash_entries;
3656static int __init set_uhash_entries(char *str)
3657{
3658	ssize_t ret;
3659
3660	if (!str)
3661		return 0;
3662
3663	ret = kstrtoul(str, 0, &uhash_entries);
3664	if (ret)
3665		return 0;
3666
3667	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3668		uhash_entries = UDP_HTABLE_SIZE_MIN;
3669	return 1;
3670}
3671__setup("uhash_entries=", set_uhash_entries);
3672
3673void __init udp_table_init(struct udp_table *table, const char *name)
3674{
3675	unsigned int i, slot_size;
3676
3677	slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) +
3678		    udp_hash4_slot_size();
3679	table->hash = alloc_large_system_hash(name,
3680					      slot_size,
3681					      uhash_entries,
3682					      21, /* one slot per 2 MB */
3683					      0,
3684					      &table->log,
3685					      &table->mask,
3686					      UDP_HTABLE_SIZE_MIN,
3687					      UDP_HTABLE_SIZE_MAX);
3688
3689	table->hash2 = (void *)(table->hash + (table->mask + 1));
3690	for (i = 0; i <= table->mask; i++) {
3691		INIT_HLIST_HEAD(&table->hash[i].head);
3692		table->hash[i].count = 0;
3693		spin_lock_init(&table->hash[i].lock);
3694	}
3695	for (i = 0; i <= table->mask; i++) {
3696		INIT_HLIST_HEAD(&table->hash2[i].hslot.head);
3697		table->hash2[i].hslot.count = 0;
3698		spin_lock_init(&table->hash2[i].hslot.lock);
3699	}
3700	udp_table_hash4_init(table);
3701}
3702
3703u32 udp_flow_hashrnd(void)
3704{
3705	static u32 hashrnd __read_mostly;
3706
3707	net_get_random_once(&hashrnd, sizeof(hashrnd));
3708
3709	return hashrnd;
3710}
3711EXPORT_SYMBOL(udp_flow_hashrnd);
3712
3713static void __net_init udp_sysctl_init(struct net *net)
3714{
3715	net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3716	net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3717
3718#ifdef CONFIG_NET_L3_MASTER_DEV
3719	net->ipv4.sysctl_udp_l3mdev_accept = 0;
3720#endif
3721}
3722
3723static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3724{
3725	struct udp_table *udptable;
3726	unsigned int slot_size;
3727	int i;
3728
3729	udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3730	if (!udptable)
3731		goto out;
3732
3733	slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) +
3734		    udp_hash4_slot_size();
3735	udptable->hash = vmalloc_huge(hash_entries * slot_size,
3736				      GFP_KERNEL_ACCOUNT);
3737	if (!udptable->hash)
3738		goto free_table;
3739
3740	udptable->hash2 = (void *)(udptable->hash + hash_entries);
3741	udptable->mask = hash_entries - 1;
3742	udptable->log = ilog2(hash_entries);
3743
3744	for (i = 0; i < hash_entries; i++) {
3745		INIT_HLIST_HEAD(&udptable->hash[i].head);
3746		udptable->hash[i].count = 0;
3747		spin_lock_init(&udptable->hash[i].lock);
3748
3749		INIT_HLIST_HEAD(&udptable->hash2[i].hslot.head);
3750		udptable->hash2[i].hslot.count = 0;
3751		spin_lock_init(&udptable->hash2[i].hslot.lock);
3752	}
3753	udp_table_hash4_init(udptable);
3754
3755	return udptable;
3756
3757free_table:
3758	kfree(udptable);
3759out:
3760	return NULL;
3761}
3762
3763static void __net_exit udp_pernet_table_free(struct net *net)
3764{
3765	struct udp_table *udptable = net->ipv4.udp_table;
3766
3767	if (udptable == &udp_table)
3768		return;
3769
3770	kvfree(udptable->hash);
3771	kfree(udptable);
3772}
3773
3774static void __net_init udp_set_table(struct net *net)
3775{
3776	struct udp_table *udptable;
3777	unsigned int hash_entries;
3778	struct net *old_net;
3779
3780	if (net_eq(net, &init_net))
3781		goto fallback;
3782
3783	old_net = current->nsproxy->net_ns;
3784	hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3785	if (!hash_entries)
3786		goto fallback;
3787
3788	/* Set min to keep the bitmap on stack in udp_lib_get_port() */
3789	if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3790		hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3791	else
3792		hash_entries = roundup_pow_of_two(hash_entries);
3793
3794	udptable = udp_pernet_table_alloc(hash_entries);
3795	if (udptable) {
3796		net->ipv4.udp_table = udptable;
3797	} else {
3798		pr_warn("Failed to allocate UDP hash table (entries: %u) "
3799			"for a netns, fallback to the global one\n",
3800			hash_entries);
3801fallback:
3802		net->ipv4.udp_table = &udp_table;
3803	}
3804}
3805
3806static int __net_init udp_pernet_init(struct net *net)
3807{
3808	udp_sysctl_init(net);
3809	udp_set_table(net);
3810
3811	return 0;
3812}
3813
3814static void __net_exit udp_pernet_exit(struct net *net)
3815{
3816	udp_pernet_table_free(net);
3817}
3818
3819static struct pernet_operations __net_initdata udp_sysctl_ops = {
3820	.init	= udp_pernet_init,
3821	.exit	= udp_pernet_exit,
3822};
3823
3824#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3825DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3826		     struct udp_sock *udp_sk, uid_t uid, int bucket)
3827
3828static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3829				      unsigned int new_batch_sz)
3830{
3831	struct sock **new_batch;
3832
3833	new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3834				   GFP_USER | __GFP_NOWARN);
3835	if (!new_batch)
3836		return -ENOMEM;
3837
3838	bpf_iter_udp_put_batch(iter);
3839	kvfree(iter->batch);
3840	iter->batch = new_batch;
3841	iter->max_sk = new_batch_sz;
3842
3843	return 0;
3844}
3845
3846#define INIT_BATCH_SZ 16
3847
3848static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3849{
3850	struct bpf_udp_iter_state *iter = priv_data;
 
3851	int ret;
3852
3853	ret = bpf_iter_init_seq_net(priv_data, aux);
3854	if (ret)
3855		return ret;
3856
3857	ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
 
 
 
3858	if (ret)
3859		bpf_iter_fini_seq_net(priv_data);
3860
3861	return ret;
3862}
3863
3864static void bpf_iter_fini_udp(void *priv_data)
3865{
3866	struct bpf_udp_iter_state *iter = priv_data;
3867
 
3868	bpf_iter_fini_seq_net(priv_data);
3869	kvfree(iter->batch);
3870}
3871
3872static const struct bpf_iter_seq_info udp_seq_info = {
3873	.seq_ops		= &bpf_iter_udp_seq_ops,
3874	.init_seq_private	= bpf_iter_init_udp,
3875	.fini_seq_private	= bpf_iter_fini_udp,
3876	.seq_priv_size		= sizeof(struct bpf_udp_iter_state),
3877};
3878
3879static struct bpf_iter_reg udp_reg_info = {
3880	.target			= "udp",
3881	.ctx_arg_info_size	= 1,
3882	.ctx_arg_info		= {
3883		{ offsetof(struct bpf_iter__udp, udp_sk),
3884		  PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3885	},
3886	.seq_info		= &udp_seq_info,
3887};
3888
3889static void __init bpf_iter_register(void)
3890{
3891	udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3892	if (bpf_iter_reg_target(&udp_reg_info))
3893		pr_warn("Warning: could not register bpf iterator udp\n");
3894}
3895#endif
3896
3897void __init udp_init(void)
3898{
3899	unsigned long limit;
3900	unsigned int i;
3901
3902	udp_table_init(&udp_table, "UDP");
3903	limit = nr_free_buffer_pages() / 8;
3904	limit = max(limit, 128UL);
3905	sysctl_udp_mem[0] = limit / 4 * 3;
3906	sysctl_udp_mem[1] = limit;
3907	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3908
3909	/* 16 spinlocks per cpu */
3910	udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3911	udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3912				GFP_KERNEL);
3913	if (!udp_busylocks)
3914		panic("UDP: failed to alloc udp_busylocks\n");
3915	for (i = 0; i < (1U << udp_busylocks_log); i++)
3916		spin_lock_init(udp_busylocks + i);
3917
3918	if (register_pernet_subsys(&udp_sysctl_ops))
3919		panic("UDP: failed to init sysctl parameters.\n");
3920
3921#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3922	bpf_iter_register();
3923#endif
3924}
v6.2
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		The User Datagram Protocol (UDP).
   8 *
   9 * Authors:	Ross Biro
  10 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  12 *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
  13 *		Hirokazu Takahashi, <taka@valinux.co.jp>
  14 *
  15 * Fixes:
  16 *		Alan Cox	:	verify_area() calls
  17 *		Alan Cox	: 	stopped close while in use off icmp
  18 *					messages. Not a fix but a botch that
  19 *					for udp at least is 'valid'.
  20 *		Alan Cox	:	Fixed icmp handling properly
  21 *		Alan Cox	: 	Correct error for oversized datagrams
  22 *		Alan Cox	:	Tidied select() semantics.
  23 *		Alan Cox	:	udp_err() fixed properly, also now
  24 *					select and read wake correctly on errors
  25 *		Alan Cox	:	udp_send verify_area moved to avoid mem leak
  26 *		Alan Cox	:	UDP can count its memory
  27 *		Alan Cox	:	send to an unknown connection causes
  28 *					an ECONNREFUSED off the icmp, but
  29 *					does NOT close.
  30 *		Alan Cox	:	Switched to new sk_buff handlers. No more backlog!
  31 *		Alan Cox	:	Using generic datagram code. Even smaller and the PEEK
  32 *					bug no longer crashes it.
  33 *		Fred Van Kempen	: 	Net2e support for sk->broadcast.
  34 *		Alan Cox	:	Uses skb_free_datagram
  35 *		Alan Cox	:	Added get/set sockopt support.
  36 *		Alan Cox	:	Broadcasting without option set returns EACCES.
  37 *		Alan Cox	:	No wakeup calls. Instead we now use the callbacks.
  38 *		Alan Cox	:	Use ip_tos and ip_ttl
  39 *		Alan Cox	:	SNMP Mibs
  40 *		Alan Cox	:	MSG_DONTROUTE, and 0.0.0.0 support.
  41 *		Matt Dillon	:	UDP length checks.
  42 *		Alan Cox	:	Smarter af_inet used properly.
  43 *		Alan Cox	:	Use new kernel side addressing.
  44 *		Alan Cox	:	Incorrect return on truncated datagram receive.
  45 *	Arnt Gulbrandsen 	:	New udp_send and stuff
  46 *		Alan Cox	:	Cache last socket
  47 *		Alan Cox	:	Route cache
  48 *		Jon Peatfield	:	Minor efficiency fix to sendto().
  49 *		Mike Shaver	:	RFC1122 checks.
  50 *		Alan Cox	:	Nonblocking error fix.
  51 *	Willy Konynenberg	:	Transparent proxying support.
  52 *		Mike McLagan	:	Routing by source
  53 *		David S. Miller	:	New socket lookup architecture.
  54 *					Last socket cache retained as it
  55 *					does have a high hit rate.
  56 *		Olaf Kirch	:	Don't linearise iovec on sendmsg.
  57 *		Andi Kleen	:	Some cleanups, cache destination entry
  58 *					for connect.
  59 *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
  60 *		Melvin Smith	:	Check msg_name not msg_namelen in sendto(),
  61 *					return ENOTCONN for unconnected sockets (POSIX)
  62 *		Janos Farkas	:	don't deliver multi/broadcasts to a different
  63 *					bound-to-device socket
  64 *	Hirokazu Takahashi	:	HW checksumming for outgoing UDP
  65 *					datagrams.
  66 *	Hirokazu Takahashi	:	sendfile() on UDP works now.
  67 *		Arnaldo C. Melo :	convert /proc/net/udp to seq_file
  68 *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
  69 *	Alexey Kuznetsov:		allow both IPv4 and IPv6 sockets to bind
  70 *					a single port at the same time.
  71 *	Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
  72 *	James Chapman		:	Add L2TP encapsulation type.
  73 */
  74
  75#define pr_fmt(fmt) "UDP: " fmt
  76
  77#include <linux/bpf-cgroup.h>
  78#include <linux/uaccess.h>
  79#include <asm/ioctls.h>
  80#include <linux/memblock.h>
  81#include <linux/highmem.h>
  82#include <linux/types.h>
  83#include <linux/fcntl.h>
  84#include <linux/module.h>
  85#include <linux/socket.h>
  86#include <linux/sockios.h>
  87#include <linux/igmp.h>
  88#include <linux/inetdevice.h>
  89#include <linux/in.h>
  90#include <linux/errno.h>
  91#include <linux/timer.h>
  92#include <linux/mm.h>
  93#include <linux/inet.h>
  94#include <linux/netdevice.h>
  95#include <linux/slab.h>
  96#include <net/tcp_states.h>
  97#include <linux/skbuff.h>
  98#include <linux/proc_fs.h>
  99#include <linux/seq_file.h>
 100#include <net/net_namespace.h>
 101#include <net/icmp.h>
 102#include <net/inet_hashtables.h>
 
 103#include <net/ip_tunnels.h>
 104#include <net/route.h>
 105#include <net/checksum.h>
 
 106#include <net/xfrm.h>
 107#include <trace/events/udp.h>
 108#include <linux/static_key.h>
 109#include <linux/btf_ids.h>
 110#include <trace/events/skb.h>
 111#include <net/busy_poll.h>
 112#include "udp_impl.h"
 113#include <net/sock_reuseport.h>
 114#include <net/addrconf.h>
 115#include <net/udp_tunnel.h>
 
 116#if IS_ENABLED(CONFIG_IPV6)
 117#include <net/ipv6_stubs.h>
 118#endif
 119
 120struct udp_table udp_table __read_mostly;
 121EXPORT_SYMBOL(udp_table);
 122
 123long sysctl_udp_mem[3] __read_mostly;
 124EXPORT_SYMBOL(sysctl_udp_mem);
 125
 126atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
 127EXPORT_SYMBOL(udp_memory_allocated);
 128DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
 129EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
 130
 131#define MAX_UDP_PORTS 65536
 132#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET)
 133
 134static struct udp_table *udp_get_table_prot(struct sock *sk)
 135{
 136	return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
 137}
 138
 139static int udp_lib_lport_inuse(struct net *net, __u16 num,
 140			       const struct udp_hslot *hslot,
 141			       unsigned long *bitmap,
 142			       struct sock *sk, unsigned int log)
 143{
 144	struct sock *sk2;
 145	kuid_t uid = sock_i_uid(sk);
 146
 147	sk_for_each(sk2, &hslot->head) {
 148		if (net_eq(sock_net(sk2), net) &&
 149		    sk2 != sk &&
 150		    (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
 151		    (!sk2->sk_reuse || !sk->sk_reuse) &&
 152		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 153		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 154		    inet_rcv_saddr_equal(sk, sk2, true)) {
 155			if (sk2->sk_reuseport && sk->sk_reuseport &&
 156			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 157			    uid_eq(uid, sock_i_uid(sk2))) {
 158				if (!bitmap)
 159					return 0;
 160			} else {
 161				if (!bitmap)
 162					return 1;
 163				__set_bit(udp_sk(sk2)->udp_port_hash >> log,
 164					  bitmap);
 165			}
 166		}
 167	}
 168	return 0;
 169}
 170
 171/*
 172 * Note: we still hold spinlock of primary hash chain, so no other writer
 173 * can insert/delete a socket with local_port == num
 174 */
 175static int udp_lib_lport_inuse2(struct net *net, __u16 num,
 176				struct udp_hslot *hslot2,
 177				struct sock *sk)
 178{
 179	struct sock *sk2;
 180	kuid_t uid = sock_i_uid(sk);
 181	int res = 0;
 182
 183	spin_lock(&hslot2->lock);
 184	udp_portaddr_for_each_entry(sk2, &hslot2->head) {
 185		if (net_eq(sock_net(sk2), net) &&
 186		    sk2 != sk &&
 187		    (udp_sk(sk2)->udp_port_hash == num) &&
 188		    (!sk2->sk_reuse || !sk->sk_reuse) &&
 189		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 190		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 191		    inet_rcv_saddr_equal(sk, sk2, true)) {
 192			if (sk2->sk_reuseport && sk->sk_reuseport &&
 193			    !rcu_access_pointer(sk->sk_reuseport_cb) &&
 194			    uid_eq(uid, sock_i_uid(sk2))) {
 195				res = 0;
 196			} else {
 197				res = 1;
 198			}
 199			break;
 200		}
 201	}
 202	spin_unlock(&hslot2->lock);
 203	return res;
 204}
 205
 206static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
 207{
 208	struct net *net = sock_net(sk);
 209	kuid_t uid = sock_i_uid(sk);
 210	struct sock *sk2;
 211
 212	sk_for_each(sk2, &hslot->head) {
 213		if (net_eq(sock_net(sk2), net) &&
 214		    sk2 != sk &&
 215		    sk2->sk_family == sk->sk_family &&
 216		    ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
 217		    (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
 218		    (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 219		    sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
 220		    inet_rcv_saddr_equal(sk, sk2, false)) {
 221			return reuseport_add_sock(sk, sk2,
 222						  inet_rcv_saddr_any(sk));
 223		}
 224	}
 225
 226	return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
 227}
 228
 229/**
 230 *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
 231 *
 232 *  @sk:          socket struct in question
 233 *  @snum:        port number to look up
 234 *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
 235 *                   with NULL address
 236 */
 237int udp_lib_get_port(struct sock *sk, unsigned short snum,
 238		     unsigned int hash2_nulladdr)
 239{
 240	struct udp_table *udptable = udp_get_table_prot(sk);
 241	struct udp_hslot *hslot, *hslot2;
 242	struct net *net = sock_net(sk);
 243	int error = -EADDRINUSE;
 244
 245	if (!snum) {
 246		DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
 247		unsigned short first, last;
 248		int low, high, remaining;
 249		unsigned int rand;
 250
 251		inet_get_local_port_range(net, &low, &high);
 252		remaining = (high - low) + 1;
 253
 254		rand = get_random_u32();
 255		first = reciprocal_scale(rand, remaining) + low;
 256		/*
 257		 * force rand to be an odd multiple of UDP_HTABLE_SIZE
 258		 */
 259		rand = (rand | 1) * (udptable->mask + 1);
 260		last = first + udptable->mask + 1;
 261		do {
 262			hslot = udp_hashslot(udptable, net, first);
 263			bitmap_zero(bitmap, PORTS_PER_CHAIN);
 264			spin_lock_bh(&hslot->lock);
 265			udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
 266					    udptable->log);
 267
 268			snum = first;
 269			/*
 270			 * Iterate on all possible values of snum for this hash.
 271			 * Using steps of an odd multiple of UDP_HTABLE_SIZE
 272			 * give us randomization and full range coverage.
 273			 */
 274			do {
 275				if (low <= snum && snum <= high &&
 276				    !test_bit(snum >> udptable->log, bitmap) &&
 277				    !inet_is_local_reserved_port(net, snum))
 278					goto found;
 279				snum += rand;
 280			} while (snum != first);
 281			spin_unlock_bh(&hslot->lock);
 282			cond_resched();
 283		} while (++first != last);
 284		goto fail;
 285	} else {
 286		hslot = udp_hashslot(udptable, net, snum);
 287		spin_lock_bh(&hslot->lock);
 288		if (hslot->count > 10) {
 289			int exist;
 290			unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
 291
 292			slot2          &= udptable->mask;
 293			hash2_nulladdr &= udptable->mask;
 294
 295			hslot2 = udp_hashslot2(udptable, slot2);
 296			if (hslot->count < hslot2->count)
 297				goto scan_primary_hash;
 298
 299			exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
 300			if (!exist && (hash2_nulladdr != slot2)) {
 301				hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
 302				exist = udp_lib_lport_inuse2(net, snum, hslot2,
 303							     sk);
 304			}
 305			if (exist)
 306				goto fail_unlock;
 307			else
 308				goto found;
 309		}
 310scan_primary_hash:
 311		if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
 312			goto fail_unlock;
 313	}
 314found:
 315	inet_sk(sk)->inet_num = snum;
 316	udp_sk(sk)->udp_port_hash = snum;
 317	udp_sk(sk)->udp_portaddr_hash ^= snum;
 318	if (sk_unhashed(sk)) {
 319		if (sk->sk_reuseport &&
 320		    udp_reuseport_add_sock(sk, hslot)) {
 321			inet_sk(sk)->inet_num = 0;
 322			udp_sk(sk)->udp_port_hash = 0;
 323			udp_sk(sk)->udp_portaddr_hash ^= snum;
 324			goto fail_unlock;
 325		}
 326
 
 
 327		sk_add_node_rcu(sk, &hslot->head);
 328		hslot->count++;
 329		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 330
 331		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 332		spin_lock(&hslot2->lock);
 333		if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 334		    sk->sk_family == AF_INET6)
 335			hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
 336					   &hslot2->head);
 337		else
 338			hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
 339					   &hslot2->head);
 340		hslot2->count++;
 341		spin_unlock(&hslot2->lock);
 342	}
 343	sock_set_flag(sk, SOCK_RCU_FREE);
 344	error = 0;
 345fail_unlock:
 346	spin_unlock_bh(&hslot->lock);
 347fail:
 348	return error;
 349}
 350EXPORT_SYMBOL(udp_lib_get_port);
 351
 352int udp_v4_get_port(struct sock *sk, unsigned short snum)
 353{
 354	unsigned int hash2_nulladdr =
 355		ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
 356	unsigned int hash2_partial =
 357		ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
 358
 359	/* precompute partial secondary hash */
 360	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
 361	return udp_lib_get_port(sk, snum, hash2_nulladdr);
 362}
 363
 364static int compute_score(struct sock *sk, struct net *net,
 365			 __be32 saddr, __be16 sport,
 366			 __be32 daddr, unsigned short hnum,
 367			 int dif, int sdif)
 368{
 369	int score;
 370	struct inet_sock *inet;
 371	bool dev_match;
 372
 373	if (!net_eq(sock_net(sk), net) ||
 374	    udp_sk(sk)->udp_port_hash != hnum ||
 375	    ipv6_only_sock(sk))
 376		return -1;
 377
 378	if (sk->sk_rcv_saddr != daddr)
 379		return -1;
 380
 381	score = (sk->sk_family == PF_INET) ? 2 : 1;
 382
 383	inet = inet_sk(sk);
 384	if (inet->inet_daddr) {
 385		if (inet->inet_daddr != saddr)
 386			return -1;
 387		score += 4;
 388	}
 389
 390	if (inet->inet_dport) {
 391		if (inet->inet_dport != sport)
 392			return -1;
 393		score += 4;
 394	}
 395
 396	dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
 397					dif, sdif);
 398	if (!dev_match)
 399		return -1;
 400	if (sk->sk_bound_dev_if)
 401		score += 4;
 402
 403	if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
 404		score++;
 405	return score;
 406}
 407
 408static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
 409		       const __u16 lport, const __be32 faddr,
 410		       const __be16 fport)
 411{
 412	static u32 udp_ehash_secret __read_mostly;
 413
 414	net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
 415
 416	return __inet_ehashfn(laddr, lport, faddr, fport,
 417			      udp_ehash_secret + net_hash_mix(net));
 418}
 
 419
 420static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
 421				     struct sk_buff *skb,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 422				     __be32 saddr, __be16 sport,
 423				     __be32 daddr, unsigned short hnum)
 
 
 424{
 425	struct sock *reuse_sk = NULL;
 426	u32 hash;
 
 
 427
 428	if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
 429		hash = udp_ehashfn(net, daddr, hnum, saddr, sport);
 430		reuse_sk = reuseport_select_sock(sk, hash, skb,
 431						 sizeof(struct udphdr));
 
 
 
 432	}
 433	return reuse_sk;
 
 434}
 435
 436/* called with rcu_read_lock() */
 437static struct sock *udp4_lib_lookup2(struct net *net,
 438				     __be32 saddr, __be16 sport,
 439				     __be32 daddr, unsigned int hnum,
 440				     int dif, int sdif,
 441				     struct udp_hslot *hslot2,
 442				     struct sk_buff *skb)
 443{
 444	struct sock *sk, *result;
 445	int score, badness;
 
 446
 447	result = NULL;
 448	badness = 0;
 449	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
 450		score = compute_score(sk, net, saddr, sport,
 451				      daddr, hnum, dif, sdif);
 
 
 452		if (score > badness) {
 453			result = lookup_reuseport(net, sk, skb,
 454						  saddr, sport, daddr, hnum);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 455			/* Fall back to scoring if group has connections */
 456			if (result && !reuseport_has_conns(sk))
 457				return result;
 458
 459			result = result ? : sk;
 460			badness = score;
 
 
 
 
 
 
 
 
 
 
 461		}
 462	}
 463	return result;
 464}
 465
 466static struct sock *udp4_lookup_run_bpf(struct net *net,
 467					struct udp_table *udptable,
 468					struct sk_buff *skb,
 469					__be32 saddr, __be16 sport,
 470					__be32 daddr, u16 hnum, const int dif)
 471{
 472	struct sock *sk, *reuse_sk;
 473	bool no_reuseport;
 474
 475	if (udptable != net->ipv4.udp_table)
 476		return NULL; /* only UDP is supported */
 477
 478	no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport,
 479					    daddr, hnum, dif, &sk);
 480	if (no_reuseport || IS_ERR_OR_NULL(sk))
 481		return sk;
 482
 483	reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
 484	if (reuse_sk)
 485		sk = reuse_sk;
 486	return sk;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 487}
 
 
 488
 489/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
 490 * harder than this. -DaveM
 491 */
 492struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
 493		__be16 sport, __be32 daddr, __be16 dport, int dif,
 494		int sdif, struct udp_table *udptable, struct sk_buff *skb)
 495{
 496	unsigned short hnum = ntohs(dport);
 497	unsigned int hash2, slot2;
 498	struct udp_hslot *hslot2;
 499	struct sock *result, *sk;
 
 500
 501	hash2 = ipv4_portaddr_hash(net, daddr, hnum);
 502	slot2 = hash2 & udptable->mask;
 503	hslot2 = &udptable->hash2[slot2];
 
 
 
 
 
 
 504
 505	/* Lookup connected or non-wildcard socket */
 506	result = udp4_lib_lookup2(net, saddr, sport,
 507				  daddr, hnum, dif, sdif,
 508				  hslot2, skb);
 509	if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
 510		goto done;
 511
 512	/* Lookup redirect from BPF */
 513	if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
 514		sk = udp4_lookup_run_bpf(net, udptable, skb,
 515					 saddr, sport, daddr, hnum, dif);
 
 
 516		if (sk) {
 517			result = sk;
 518			goto done;
 519		}
 520	}
 521
 522	/* Got non-wildcard socket or error on first lookup */
 523	if (result)
 524		goto done;
 525
 526	/* Lookup wildcard sockets */
 527	hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
 528	slot2 = hash2 & udptable->mask;
 529	hslot2 = &udptable->hash2[slot2];
 530
 531	result = udp4_lib_lookup2(net, saddr, sport,
 532				  htonl(INADDR_ANY), hnum, dif, sdif,
 533				  hslot2, skb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 534done:
 535	if (IS_ERR(result))
 536		return NULL;
 537	return result;
 538}
 539EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
 540
 541static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
 542						 __be16 sport, __be16 dport,
 543						 struct udp_table *udptable)
 544{
 545	const struct iphdr *iph = ip_hdr(skb);
 546
 547	return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
 548				 iph->daddr, dport, inet_iif(skb),
 549				 inet_sdif(skb), udptable, skb);
 550}
 551
 552struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
 553				 __be16 sport, __be16 dport)
 554{
 555	const struct iphdr *iph = ip_hdr(skb);
 
 556	struct net *net = dev_net(skb->dev);
 
 
 
 557
 558	return __udp4_lib_lookup(net, iph->saddr, sport,
 559				 iph->daddr, dport, inet_iif(skb),
 560				 inet_sdif(skb), net->ipv4.udp_table, NULL);
 561}
 562
 563/* Must be called under rcu_read_lock().
 564 * Does increment socket refcount.
 565 */
 566#if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
 567struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 568			     __be32 daddr, __be16 dport, int dif)
 569{
 570	struct sock *sk;
 571
 572	sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
 573			       dif, 0, net->ipv4.udp_table, NULL);
 574	if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
 575		sk = NULL;
 576	return sk;
 577}
 578EXPORT_SYMBOL_GPL(udp4_lib_lookup);
 579#endif
 580
 581static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
 582				       __be16 loc_port, __be32 loc_addr,
 583				       __be16 rmt_port, __be32 rmt_addr,
 584				       int dif, int sdif, unsigned short hnum)
 585{
 586	struct inet_sock *inet = inet_sk(sk);
 587
 588	if (!net_eq(sock_net(sk), net) ||
 589	    udp_sk(sk)->udp_port_hash != hnum ||
 590	    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
 591	    (inet->inet_dport != rmt_port && inet->inet_dport) ||
 592	    (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
 593	    ipv6_only_sock(sk) ||
 594	    !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
 595		return false;
 596	if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
 597		return false;
 598	return true;
 599}
 600
 601DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
 
 
 
 
 
 
 
 602void udp_encap_enable(void)
 603{
 604	static_branch_inc(&udp_encap_needed_key);
 605}
 606EXPORT_SYMBOL(udp_encap_enable);
 607
 608void udp_encap_disable(void)
 609{
 610	static_branch_dec(&udp_encap_needed_key);
 611}
 612EXPORT_SYMBOL(udp_encap_disable);
 613
 614/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
 615 * through error handlers in encapsulations looking for a match.
 616 */
 617static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
 618{
 619	int i;
 620
 621	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
 622		int (*handler)(struct sk_buff *skb, u32 info);
 623		const struct ip_tunnel_encap_ops *encap;
 624
 625		encap = rcu_dereference(iptun_encaps[i]);
 626		if (!encap)
 627			continue;
 628		handler = encap->err_handler;
 629		if (handler && !handler(skb, info))
 630			return 0;
 631	}
 632
 633	return -ENOENT;
 634}
 635
 636/* Try to match ICMP errors to UDP tunnels by looking up a socket without
 637 * reversing source and destination port: this will match tunnels that force the
 638 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
 639 * lwtunnels might actually break this assumption by being configured with
 640 * different destination ports on endpoints, in this case we won't be able to
 641 * trace ICMP messages back to them.
 642 *
 643 * If this doesn't match any socket, probe tunnels with arbitrary destination
 644 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
 645 * we've sent packets to won't necessarily match the local destination port.
 646 *
 647 * Then ask the tunnel implementation to match the error against a valid
 648 * association.
 649 *
 650 * Return an error if we can't find a match, the socket if we need further
 651 * processing, zero otherwise.
 652 */
 653static struct sock *__udp4_lib_err_encap(struct net *net,
 654					 const struct iphdr *iph,
 655					 struct udphdr *uh,
 656					 struct udp_table *udptable,
 657					 struct sock *sk,
 658					 struct sk_buff *skb, u32 info)
 659{
 660	int (*lookup)(struct sock *sk, struct sk_buff *skb);
 661	int network_offset, transport_offset;
 662	struct udp_sock *up;
 663
 664	network_offset = skb_network_offset(skb);
 665	transport_offset = skb_transport_offset(skb);
 666
 667	/* Network header needs to point to the outer IPv4 header inside ICMP */
 668	skb_reset_network_header(skb);
 669
 670	/* Transport header needs to point to the UDP header */
 671	skb_set_transport_header(skb, iph->ihl << 2);
 672
 673	if (sk) {
 674		up = udp_sk(sk);
 675
 676		lookup = READ_ONCE(up->encap_err_lookup);
 677		if (lookup && lookup(sk, skb))
 678			sk = NULL;
 679
 680		goto out;
 681	}
 682
 683	sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
 684			       iph->saddr, uh->dest, skb->dev->ifindex, 0,
 685			       udptable, NULL);
 686	if (sk) {
 687		up = udp_sk(sk);
 688
 689		lookup = READ_ONCE(up->encap_err_lookup);
 690		if (!lookup || lookup(sk, skb))
 691			sk = NULL;
 692	}
 693
 694out:
 695	if (!sk)
 696		sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
 697
 698	skb_set_transport_header(skb, transport_offset);
 699	skb_set_network_header(skb, network_offset);
 700
 701	return sk;
 702}
 703
 704/*
 705 * This routine is called by the ICMP module when it gets some
 706 * sort of error condition.  If err < 0 then the socket should
 707 * be closed and the error returned to the user.  If err > 0
 708 * it's just the icmp type << 8 | icmp code.
 709 * Header points to the ip header of the error packet. We move
 710 * on past this. Then (as it used to claim before adjustment)
 711 * header points to the first 8 bytes of the udp header.  We need
 712 * to find the appropriate port.
 713 */
 714
 715int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
 716{
 717	struct inet_sock *inet;
 718	const struct iphdr *iph = (const struct iphdr *)skb->data;
 719	struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
 720	const int type = icmp_hdr(skb)->type;
 721	const int code = icmp_hdr(skb)->code;
 722	bool tunnel = false;
 723	struct sock *sk;
 724	int harderr;
 725	int err;
 726	struct net *net = dev_net(skb->dev);
 727
 728	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
 729			       iph->saddr, uh->source, skb->dev->ifindex,
 730			       inet_sdif(skb), udptable, NULL);
 731
 732	if (!sk || udp_sk(sk)->encap_type) {
 733		/* No socket for error: try tunnels before discarding */
 734		if (static_branch_unlikely(&udp_encap_needed_key)) {
 735			sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
 736						  info);
 737			if (!sk)
 738				return 0;
 739		} else
 740			sk = ERR_PTR(-ENOENT);
 741
 742		if (IS_ERR(sk)) {
 743			__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 744			return PTR_ERR(sk);
 745		}
 746
 747		tunnel = true;
 748	}
 749
 750	err = 0;
 751	harderr = 0;
 752	inet = inet_sk(sk);
 753
 754	switch (type) {
 755	default:
 756	case ICMP_TIME_EXCEEDED:
 757		err = EHOSTUNREACH;
 758		break;
 759	case ICMP_SOURCE_QUENCH:
 760		goto out;
 761	case ICMP_PARAMETERPROB:
 762		err = EPROTO;
 763		harderr = 1;
 764		break;
 765	case ICMP_DEST_UNREACH:
 766		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
 767			ipv4_sk_update_pmtu(skb, sk, info);
 768			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
 769				err = EMSGSIZE;
 770				harderr = 1;
 771				break;
 772			}
 773			goto out;
 774		}
 775		err = EHOSTUNREACH;
 776		if (code <= NR_ICMP_UNREACH) {
 777			harderr = icmp_err_convert[code].fatal;
 778			err = icmp_err_convert[code].errno;
 779		}
 780		break;
 781	case ICMP_REDIRECT:
 782		ipv4_sk_redirect(skb, sk);
 783		goto out;
 784	}
 785
 786	/*
 787	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
 788	 *	4.1.3.3.
 789	 */
 790	if (tunnel) {
 791		/* ...not for tunnels though: we don't have a sending socket */
 792		if (udp_sk(sk)->encap_err_rcv)
 793			udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
 794						  (u8 *)(uh+1));
 795		goto out;
 796	}
 797	if (!inet->recverr) {
 798		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
 799			goto out;
 800	} else
 801		ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
 802
 803	sk->sk_err = err;
 804	sk_error_report(sk);
 805out:
 806	return 0;
 807}
 808
 809int udp_err(struct sk_buff *skb, u32 info)
 810{
 811	return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
 812}
 813
 814/*
 815 * Throw away all pending data and cancel the corking. Socket is locked.
 816 */
 817void udp_flush_pending_frames(struct sock *sk)
 818{
 819	struct udp_sock *up = udp_sk(sk);
 820
 821	if (up->pending) {
 822		up->len = 0;
 823		up->pending = 0;
 824		ip_flush_pending_frames(sk);
 825	}
 826}
 827EXPORT_SYMBOL(udp_flush_pending_frames);
 828
 829/**
 830 * 	udp4_hwcsum  -  handle outgoing HW checksumming
 831 * 	@skb: 	sk_buff containing the filled-in UDP header
 832 * 	        (checksum field must be zeroed out)
 833 *	@src:	source IP address
 834 *	@dst:	destination IP address
 835 */
 836void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
 837{
 838	struct udphdr *uh = udp_hdr(skb);
 839	int offset = skb_transport_offset(skb);
 840	int len = skb->len - offset;
 841	int hlen = len;
 842	__wsum csum = 0;
 843
 844	if (!skb_has_frag_list(skb)) {
 845		/*
 846		 * Only one fragment on the socket.
 847		 */
 848		skb->csum_start = skb_transport_header(skb) - skb->head;
 849		skb->csum_offset = offsetof(struct udphdr, check);
 850		uh->check = ~csum_tcpudp_magic(src, dst, len,
 851					       IPPROTO_UDP, 0);
 852	} else {
 853		struct sk_buff *frags;
 854
 855		/*
 856		 * HW-checksum won't work as there are two or more
 857		 * fragments on the socket so that all csums of sk_buffs
 858		 * should be together
 859		 */
 860		skb_walk_frags(skb, frags) {
 861			csum = csum_add(csum, frags->csum);
 862			hlen -= frags->len;
 863		}
 864
 865		csum = skb_checksum(skb, offset, hlen, csum);
 866		skb->ip_summed = CHECKSUM_NONE;
 867
 868		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
 869		if (uh->check == 0)
 870			uh->check = CSUM_MANGLED_0;
 871	}
 872}
 873EXPORT_SYMBOL_GPL(udp4_hwcsum);
 874
 875/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
 876 * for the simple case like when setting the checksum for a UDP tunnel.
 877 */
 878void udp_set_csum(bool nocheck, struct sk_buff *skb,
 879		  __be32 saddr, __be32 daddr, int len)
 880{
 881	struct udphdr *uh = udp_hdr(skb);
 882
 883	if (nocheck) {
 884		uh->check = 0;
 885	} else if (skb_is_gso(skb)) {
 886		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 887	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 888		uh->check = 0;
 889		uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
 890		if (uh->check == 0)
 891			uh->check = CSUM_MANGLED_0;
 892	} else {
 893		skb->ip_summed = CHECKSUM_PARTIAL;
 894		skb->csum_start = skb_transport_header(skb) - skb->head;
 895		skb->csum_offset = offsetof(struct udphdr, check);
 896		uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 897	}
 898}
 899EXPORT_SYMBOL(udp_set_csum);
 900
 901static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
 902			struct inet_cork *cork)
 903{
 904	struct sock *sk = skb->sk;
 905	struct inet_sock *inet = inet_sk(sk);
 906	struct udphdr *uh;
 907	int err;
 908	int is_udplite = IS_UDPLITE(sk);
 909	int offset = skb_transport_offset(skb);
 910	int len = skb->len - offset;
 911	int datalen = len - sizeof(*uh);
 912	__wsum csum = 0;
 913
 914	/*
 915	 * Create a UDP header
 916	 */
 917	uh = udp_hdr(skb);
 918	uh->source = inet->inet_sport;
 919	uh->dest = fl4->fl4_dport;
 920	uh->len = htons(len);
 921	uh->check = 0;
 922
 923	if (cork->gso_size) {
 924		const int hlen = skb_network_header_len(skb) +
 925				 sizeof(struct udphdr);
 926
 927		if (hlen + cork->gso_size > cork->fragsize) {
 928			kfree_skb(skb);
 929			return -EINVAL;
 930		}
 931		if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
 932			kfree_skb(skb);
 933			return -EINVAL;
 934		}
 935		if (sk->sk_no_check_tx) {
 936			kfree_skb(skb);
 937			return -EINVAL;
 938		}
 939		if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
 940		    dst_xfrm(skb_dst(skb))) {
 941			kfree_skb(skb);
 942			return -EIO;
 943		}
 944
 945		if (datalen > cork->gso_size) {
 946			skb_shinfo(skb)->gso_size = cork->gso_size;
 947			skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
 948			skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
 949								 cork->gso_size);
 
 
 
 950		}
 951		goto csum_partial;
 952	}
 953
 954	if (is_udplite)  				 /*     UDP-Lite      */
 955		csum = udplite_csum(skb);
 956
 957	else if (sk->sk_no_check_tx) {			 /* UDP csum off */
 958
 959		skb->ip_summed = CHECKSUM_NONE;
 960		goto send;
 961
 962	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
 963csum_partial:
 964
 965		udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
 966		goto send;
 967
 968	} else
 969		csum = udp_csum(skb);
 970
 971	/* add protocol-dependent pseudo-header */
 972	uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
 973				      sk->sk_protocol, csum);
 974	if (uh->check == 0)
 975		uh->check = CSUM_MANGLED_0;
 976
 977send:
 978	err = ip_send_skb(sock_net(sk), skb);
 979	if (err) {
 980		if (err == -ENOBUFS && !inet->recverr) {
 
 981			UDP_INC_STATS(sock_net(sk),
 982				      UDP_MIB_SNDBUFERRORS, is_udplite);
 983			err = 0;
 984		}
 985	} else
 986		UDP_INC_STATS(sock_net(sk),
 987			      UDP_MIB_OUTDATAGRAMS, is_udplite);
 988	return err;
 989}
 990
 991/*
 992 * Push out all pending data as one UDP datagram. Socket is locked.
 993 */
 994int udp_push_pending_frames(struct sock *sk)
 995{
 996	struct udp_sock  *up = udp_sk(sk);
 997	struct inet_sock *inet = inet_sk(sk);
 998	struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
 999	struct sk_buff *skb;
1000	int err = 0;
1001
1002	skb = ip_finish_skb(sk, fl4);
1003	if (!skb)
1004		goto out;
1005
1006	err = udp_send_skb(skb, fl4, &inet->cork.base);
1007
1008out:
1009	up->len = 0;
1010	up->pending = 0;
1011	return err;
1012}
1013EXPORT_SYMBOL(udp_push_pending_frames);
1014
1015static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1016{
1017	switch (cmsg->cmsg_type) {
1018	case UDP_SEGMENT:
1019		if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1020			return -EINVAL;
1021		*gso_size = *(__u16 *)CMSG_DATA(cmsg);
1022		return 0;
1023	default:
1024		return -EINVAL;
1025	}
1026}
1027
1028int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1029{
1030	struct cmsghdr *cmsg;
1031	bool need_ip = false;
1032	int err;
1033
1034	for_each_cmsghdr(cmsg, msg) {
1035		if (!CMSG_OK(msg, cmsg))
1036			return -EINVAL;
1037
1038		if (cmsg->cmsg_level != SOL_UDP) {
1039			need_ip = true;
1040			continue;
1041		}
1042
1043		err = __udp_cmsg_send(cmsg, gso_size);
1044		if (err)
1045			return err;
1046	}
1047
1048	return need_ip;
1049}
1050EXPORT_SYMBOL_GPL(udp_cmsg_send);
1051
1052int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1053{
1054	struct inet_sock *inet = inet_sk(sk);
1055	struct udp_sock *up = udp_sk(sk);
1056	DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1057	struct flowi4 fl4_stack;
1058	struct flowi4 *fl4;
1059	int ulen = len;
1060	struct ipcm_cookie ipc;
1061	struct rtable *rt = NULL;
1062	int free = 0;
1063	int connected = 0;
1064	__be32 daddr, faddr, saddr;
 
1065	__be16 dport;
1066	u8  tos;
1067	int err, is_udplite = IS_UDPLITE(sk);
1068	int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1069	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1070	struct sk_buff *skb;
1071	struct ip_options_data opt_copy;
 
1072
1073	if (len > 0xFFFF)
1074		return -EMSGSIZE;
1075
1076	/*
1077	 *	Check the flags.
1078	 */
1079
1080	if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1081		return -EOPNOTSUPP;
1082
1083	getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1084
1085	fl4 = &inet->cork.fl.u.ip4;
1086	if (up->pending) {
1087		/*
1088		 * There are pending frames.
1089		 * The socket lock must be held while it's corked.
1090		 */
1091		lock_sock(sk);
1092		if (likely(up->pending)) {
1093			if (unlikely(up->pending != AF_INET)) {
1094				release_sock(sk);
1095				return -EINVAL;
1096			}
1097			goto do_append_data;
1098		}
1099		release_sock(sk);
1100	}
1101	ulen += sizeof(struct udphdr);
1102
1103	/*
1104	 *	Get and verify the address.
1105	 */
1106	if (usin) {
1107		if (msg->msg_namelen < sizeof(*usin))
1108			return -EINVAL;
1109		if (usin->sin_family != AF_INET) {
1110			if (usin->sin_family != AF_UNSPEC)
1111				return -EAFNOSUPPORT;
1112		}
1113
1114		daddr = usin->sin_addr.s_addr;
1115		dport = usin->sin_port;
1116		if (dport == 0)
1117			return -EINVAL;
1118	} else {
1119		if (sk->sk_state != TCP_ESTABLISHED)
1120			return -EDESTADDRREQ;
1121		daddr = inet->inet_daddr;
1122		dport = inet->inet_dport;
1123		/* Open fast path for connected socket.
1124		   Route will not be used, if at least one option is set.
1125		 */
1126		connected = 1;
1127	}
1128
1129	ipcm_init_sk(&ipc, inet);
1130	ipc.gso_size = READ_ONCE(up->gso_size);
1131
1132	if (msg->msg_controllen) {
1133		err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1134		if (err > 0)
1135			err = ip_cmsg_send(sk, msg, &ipc,
1136					   sk->sk_family == AF_INET6);
 
 
1137		if (unlikely(err < 0)) {
1138			kfree(ipc.opt);
1139			return err;
1140		}
1141		if (ipc.opt)
1142			free = 1;
1143		connected = 0;
1144	}
1145	if (!ipc.opt) {
1146		struct ip_options_rcu *inet_opt;
1147
1148		rcu_read_lock();
1149		inet_opt = rcu_dereference(inet->inet_opt);
1150		if (inet_opt) {
1151			memcpy(&opt_copy, inet_opt,
1152			       sizeof(*inet_opt) + inet_opt->opt.optlen);
1153			ipc.opt = &opt_copy.opt;
1154		}
1155		rcu_read_unlock();
1156	}
1157
1158	if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1159		err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1160					    (struct sockaddr *)usin, &ipc.addr);
 
 
1161		if (err)
1162			goto out_free;
1163		if (usin) {
1164			if (usin->sin_port == 0) {
1165				/* BPF program set invalid port. Reject it. */
1166				err = -EINVAL;
1167				goto out_free;
1168			}
1169			daddr = usin->sin_addr.s_addr;
1170			dport = usin->sin_port;
1171		}
1172	}
1173
1174	saddr = ipc.addr;
1175	ipc.addr = faddr = daddr;
1176
1177	if (ipc.opt && ipc.opt->opt.srr) {
1178		if (!daddr) {
1179			err = -EINVAL;
1180			goto out_free;
1181		}
1182		faddr = ipc.opt->opt.faddr;
1183		connected = 0;
1184	}
1185	tos = get_rttos(&ipc, inet);
1186	if (sock_flag(sk, SOCK_LOCALROUTE) ||
1187	    (msg->msg_flags & MSG_DONTROUTE) ||
1188	    (ipc.opt && ipc.opt->opt.is_strictroute)) {
1189		tos |= RTO_ONLINK;
1190		connected = 0;
1191	}
1192
 
1193	if (ipv4_is_multicast(daddr)) {
1194		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1195			ipc.oif = inet->mc_index;
1196		if (!saddr)
1197			saddr = inet->mc_addr;
1198		connected = 0;
1199	} else if (!ipc.oif) {
1200		ipc.oif = inet->uc_index;
1201	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1202		/* oif is set, packet is to local broadcast and
1203		 * uc_index is set. oif is most likely set
1204		 * by sk_bound_dev_if. If uc_index != oif check if the
1205		 * oif is an L3 master and uc_index is an L3 slave.
1206		 * If so, we want to allow the send using the uc_index.
1207		 */
1208		if (ipc.oif != inet->uc_index &&
1209		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1210							      inet->uc_index)) {
1211			ipc.oif = inet->uc_index;
1212		}
1213	}
1214
1215	if (connected)
1216		rt = (struct rtable *)sk_dst_check(sk, 0);
1217
1218	if (!rt) {
1219		struct net *net = sock_net(sk);
1220		__u8 flow_flags = inet_sk_flowi_flags(sk);
1221
1222		fl4 = &fl4_stack;
1223
1224		flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1225				   RT_SCOPE_UNIVERSE, sk->sk_protocol,
1226				   flow_flags,
1227				   faddr, saddr, dport, inet->inet_sport,
1228				   sk->sk_uid);
1229
1230		security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1231		rt = ip_route_output_flow(net, fl4, sk);
1232		if (IS_ERR(rt)) {
1233			err = PTR_ERR(rt);
1234			rt = NULL;
1235			if (err == -ENETUNREACH)
1236				IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1237			goto out;
1238		}
1239
1240		err = -EACCES;
1241		if ((rt->rt_flags & RTCF_BROADCAST) &&
1242		    !sock_flag(sk, SOCK_BROADCAST))
1243			goto out;
1244		if (connected)
1245			sk_dst_set(sk, dst_clone(&rt->dst));
1246	}
1247
1248	if (msg->msg_flags&MSG_CONFIRM)
1249		goto do_confirm;
1250back_from_confirm:
1251
1252	saddr = fl4->saddr;
1253	if (!ipc.addr)
1254		daddr = ipc.addr = fl4->daddr;
1255
1256	/* Lockless fast path for the non-corking case. */
1257	if (!corkreq) {
1258		struct inet_cork cork;
1259
1260		skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1261				  sizeof(struct udphdr), &ipc, &rt,
1262				  &cork, msg->msg_flags);
1263		err = PTR_ERR(skb);
1264		if (!IS_ERR_OR_NULL(skb))
1265			err = udp_send_skb(skb, fl4, &cork);
1266		goto out;
1267	}
1268
1269	lock_sock(sk);
1270	if (unlikely(up->pending)) {
1271		/* The socket is already corked while preparing it. */
1272		/* ... which is an evident application bug. --ANK */
1273		release_sock(sk);
1274
1275		net_dbg_ratelimited("socket already corked\n");
1276		err = -EINVAL;
1277		goto out;
1278	}
1279	/*
1280	 *	Now cork the socket to pend data.
1281	 */
1282	fl4 = &inet->cork.fl.u.ip4;
1283	fl4->daddr = daddr;
1284	fl4->saddr = saddr;
1285	fl4->fl4_dport = dport;
1286	fl4->fl4_sport = inet->inet_sport;
1287	up->pending = AF_INET;
1288
1289do_append_data:
1290	up->len += ulen;
1291	err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1292			     sizeof(struct udphdr), &ipc, &rt,
1293			     corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1294	if (err)
1295		udp_flush_pending_frames(sk);
1296	else if (!corkreq)
1297		err = udp_push_pending_frames(sk);
1298	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1299		up->pending = 0;
1300	release_sock(sk);
1301
1302out:
1303	ip_rt_put(rt);
1304out_free:
1305	if (free)
1306		kfree(ipc.opt);
1307	if (!err)
1308		return len;
1309	/*
1310	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1311	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1312	 * we don't have a good statistic (IpOutDiscards but it can be too many
1313	 * things).  We could add another new stat but at least for now that
1314	 * seems like overkill.
1315	 */
1316	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1317		UDP_INC_STATS(sock_net(sk),
1318			      UDP_MIB_SNDBUFERRORS, is_udplite);
1319	}
1320	return err;
1321
1322do_confirm:
1323	if (msg->msg_flags & MSG_PROBE)
1324		dst_confirm_neigh(&rt->dst, &fl4->daddr);
1325	if (!(msg->msg_flags&MSG_PROBE) || len)
1326		goto back_from_confirm;
1327	err = 0;
1328	goto out;
1329}
1330EXPORT_SYMBOL(udp_sendmsg);
1331
1332int udp_sendpage(struct sock *sk, struct page *page, int offset,
1333		 size_t size, int flags)
1334{
1335	struct inet_sock *inet = inet_sk(sk);
1336	struct udp_sock *up = udp_sk(sk);
1337	int ret;
1338
1339	if (flags & MSG_SENDPAGE_NOTLAST)
1340		flags |= MSG_MORE;
1341
1342	if (!up->pending) {
1343		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
1344
1345		/* Call udp_sendmsg to specify destination address which
1346		 * sendpage interface can't pass.
1347		 * This will succeed only when the socket is connected.
1348		 */
1349		ret = udp_sendmsg(sk, &msg, 0);
1350		if (ret < 0)
1351			return ret;
1352	}
1353
1354	lock_sock(sk);
1355
1356	if (unlikely(!up->pending)) {
1357		release_sock(sk);
1358
1359		net_dbg_ratelimited("cork failed\n");
1360		return -EINVAL;
1361	}
1362
1363	ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1364			     page, offset, size, flags);
1365	if (ret == -EOPNOTSUPP) {
1366		release_sock(sk);
1367		return sock_no_sendpage(sk->sk_socket, page, offset,
1368					size, flags);
1369	}
1370	if (ret < 0) {
1371		udp_flush_pending_frames(sk);
1372		goto out;
1373	}
1374
1375	up->len += size;
1376	if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE)))
1377		ret = udp_push_pending_frames(sk);
1378	if (!ret)
1379		ret = size;
1380out:
1381	release_sock(sk);
1382	return ret;
1383}
 
1384
1385#define UDP_SKB_IS_STATELESS 0x80000000
1386
1387/* all head states (dst, sk, nf conntrack) except skb extensions are
1388 * cleared by udp_rcv().
1389 *
1390 * We need to preserve secpath, if present, to eventually process
1391 * IP_CMSG_PASSSEC at recvmsg() time.
1392 *
1393 * Other extensions can be cleared.
1394 */
1395static bool udp_try_make_stateless(struct sk_buff *skb)
1396{
1397	if (!skb_has_extensions(skb))
1398		return true;
1399
1400	if (!secpath_exists(skb)) {
1401		skb_ext_reset(skb);
1402		return true;
1403	}
1404
1405	return false;
1406}
1407
1408static void udp_set_dev_scratch(struct sk_buff *skb)
1409{
1410	struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1411
1412	BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1413	scratch->_tsize_state = skb->truesize;
1414#if BITS_PER_LONG == 64
1415	scratch->len = skb->len;
1416	scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1417	scratch->is_linear = !skb_is_nonlinear(skb);
1418#endif
1419	if (udp_try_make_stateless(skb))
1420		scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1421}
1422
1423static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1424{
1425	/* We come here after udp_lib_checksum_complete() returned 0.
1426	 * This means that __skb_checksum_complete() might have
1427	 * set skb->csum_valid to 1.
1428	 * On 64bit platforms, we can set csum_unnecessary
1429	 * to true, but only if the skb is not shared.
1430	 */
1431#if BITS_PER_LONG == 64
1432	if (!skb_shared(skb))
1433		udp_skb_scratch(skb)->csum_unnecessary = true;
1434#endif
1435}
1436
1437static int udp_skb_truesize(struct sk_buff *skb)
1438{
1439	return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1440}
1441
1442static bool udp_skb_has_head_state(struct sk_buff *skb)
1443{
1444	return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1445}
1446
1447/* fully reclaim rmem/fwd memory allocated for skb */
1448static void udp_rmem_release(struct sock *sk, int size, int partial,
1449			     bool rx_queue_lock_held)
1450{
1451	struct udp_sock *up = udp_sk(sk);
1452	struct sk_buff_head *sk_queue;
1453	int amt;
1454
1455	if (likely(partial)) {
1456		up->forward_deficit += size;
1457		size = up->forward_deficit;
1458		if (size < READ_ONCE(up->forward_threshold) &&
1459		    !skb_queue_empty(&up->reader_queue))
1460			return;
1461	} else {
1462		size += up->forward_deficit;
1463	}
1464	up->forward_deficit = 0;
1465
1466	/* acquire the sk_receive_queue for fwd allocated memory scheduling,
1467	 * if the called don't held it already
1468	 */
1469	sk_queue = &sk->sk_receive_queue;
1470	if (!rx_queue_lock_held)
1471		spin_lock(&sk_queue->lock);
1472
1473
1474	sk->sk_forward_alloc += size;
1475	amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1476	sk->sk_forward_alloc -= amt;
1477
1478	if (amt)
1479		__sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1480
1481	atomic_sub(size, &sk->sk_rmem_alloc);
1482
1483	/* this can save us from acquiring the rx queue lock on next receive */
1484	skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1485
1486	if (!rx_queue_lock_held)
1487		spin_unlock(&sk_queue->lock);
1488}
1489
1490/* Note: called with reader_queue.lock held.
1491 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1492 * This avoids a cache line miss while receive_queue lock is held.
1493 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1494 */
1495void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1496{
1497	prefetch(&skb->data);
1498	udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1499}
1500EXPORT_SYMBOL(udp_skb_destructor);
1501
1502/* as above, but the caller held the rx queue lock, too */
1503static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1504{
1505	prefetch(&skb->data);
1506	udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1507}
1508
1509/* Idea of busylocks is to let producers grab an extra spinlock
1510 * to relieve pressure on the receive_queue spinlock shared by consumer.
1511 * Under flood, this means that only one producer can be in line
1512 * trying to acquire the receive_queue spinlock.
1513 * These busylock can be allocated on a per cpu manner, instead of a
1514 * per socket one (that would consume a cache line per socket)
1515 */
1516static int udp_busylocks_log __read_mostly;
1517static spinlock_t *udp_busylocks __read_mostly;
1518
1519static spinlock_t *busylock_acquire(void *ptr)
1520{
1521	spinlock_t *busy;
1522
1523	busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1524	spin_lock(busy);
1525	return busy;
1526}
1527
1528static void busylock_release(spinlock_t *busy)
1529{
1530	if (busy)
1531		spin_unlock(busy);
1532}
1533
 
 
 
 
 
 
 
 
 
 
 
1534int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1535{
1536	struct sk_buff_head *list = &sk->sk_receive_queue;
1537	int rmem, delta, amt, err = -ENOMEM;
1538	spinlock_t *busy = NULL;
1539	int size;
1540
1541	/* try to avoid the costly atomic add/sub pair when the receive
1542	 * queue is full; always allow at least a packet
1543	 */
1544	rmem = atomic_read(&sk->sk_rmem_alloc);
1545	if (rmem > sk->sk_rcvbuf)
 
1546		goto drop;
1547
1548	/* Under mem pressure, it might be helpful to help udp_recvmsg()
1549	 * having linear skbs :
1550	 * - Reduce memory overhead and thus increase receive queue capacity
1551	 * - Less cache line misses at copyout() time
1552	 * - Less work at consume_skb() (less alien page frag freeing)
1553	 */
1554	if (rmem > (sk->sk_rcvbuf >> 1)) {
1555		skb_condense(skb);
1556
1557		busy = busylock_acquire(sk);
1558	}
1559	size = skb->truesize;
1560	udp_set_dev_scratch(skb);
1561
1562	/* we drop only if the receive buf is full and the receive
1563	 * queue contains some other skb
1564	 */
1565	rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1566	if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1567		goto uncharge_drop;
1568
1569	spin_lock(&list->lock);
1570	if (size >= sk->sk_forward_alloc) {
1571		amt = sk_mem_pages(size);
1572		delta = amt << PAGE_SHIFT;
1573		if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1574			err = -ENOBUFS;
1575			spin_unlock(&list->lock);
1576			goto uncharge_drop;
1577		}
1578
1579		sk->sk_forward_alloc += delta;
1580	}
1581
1582	sk->sk_forward_alloc -= size;
1583
1584	/* no need to setup a destructor, we will explicitly release the
1585	 * forward allocated memory on dequeue
1586	 */
1587	sock_skb_set_dropcount(sk, skb);
1588
1589	__skb_queue_tail(list, skb);
1590	spin_unlock(&list->lock);
1591
1592	if (!sock_flag(sk, SOCK_DEAD))
1593		sk->sk_data_ready(sk);
1594
1595	busylock_release(busy);
1596	return 0;
1597
1598uncharge_drop:
1599	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1600
1601drop:
1602	atomic_inc(&sk->sk_drops);
1603	busylock_release(busy);
1604	return err;
1605}
1606EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1607
1608void udp_destruct_common(struct sock *sk)
1609{
1610	/* reclaim completely the forward allocated memory */
1611	struct udp_sock *up = udp_sk(sk);
1612	unsigned int total = 0;
1613	struct sk_buff *skb;
1614
1615	skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1616	while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1617		total += skb->truesize;
1618		kfree_skb(skb);
1619	}
1620	udp_rmem_release(sk, total, 0, true);
1621}
1622EXPORT_SYMBOL_GPL(udp_destruct_common);
1623
1624static void udp_destruct_sock(struct sock *sk)
1625{
1626	udp_destruct_common(sk);
1627	inet_sock_destruct(sk);
1628}
1629
1630int udp_init_sock(struct sock *sk)
1631{
1632	udp_lib_init_sock(sk);
1633	sk->sk_destruct = udp_destruct_sock;
1634	set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1635	return 0;
1636}
1637
1638void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1639{
1640	if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1641		bool slow = lock_sock_fast(sk);
1642
1643		sk_peek_offset_bwd(sk, len);
1644		unlock_sock_fast(sk, slow);
1645	}
1646
1647	if (!skb_unref(skb))
1648		return;
1649
1650	/* In the more common cases we cleared the head states previously,
1651	 * see __udp_queue_rcv_skb().
1652	 */
1653	if (unlikely(udp_skb_has_head_state(skb)))
1654		skb_release_head_state(skb);
1655	__consume_stateless_skb(skb);
1656}
1657EXPORT_SYMBOL_GPL(skb_consume_udp);
1658
1659static struct sk_buff *__first_packet_length(struct sock *sk,
1660					     struct sk_buff_head *rcvq,
1661					     int *total)
1662{
1663	struct sk_buff *skb;
1664
1665	while ((skb = skb_peek(rcvq)) != NULL) {
1666		if (udp_lib_checksum_complete(skb)) {
1667			__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1668					IS_UDPLITE(sk));
1669			__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1670					IS_UDPLITE(sk));
1671			atomic_inc(&sk->sk_drops);
1672			__skb_unlink(skb, rcvq);
1673			*total += skb->truesize;
1674			kfree_skb(skb);
1675		} else {
1676			udp_skb_csum_unnecessary_set(skb);
1677			break;
1678		}
1679	}
1680	return skb;
1681}
1682
1683/**
1684 *	first_packet_length	- return length of first packet in receive queue
1685 *	@sk: socket
1686 *
1687 *	Drops all bad checksum frames, until a valid one is found.
1688 *	Returns the length of found skb, or -1 if none is found.
1689 */
1690static int first_packet_length(struct sock *sk)
1691{
1692	struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1693	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1694	struct sk_buff *skb;
1695	int total = 0;
1696	int res;
1697
1698	spin_lock_bh(&rcvq->lock);
1699	skb = __first_packet_length(sk, rcvq, &total);
1700	if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1701		spin_lock(&sk_queue->lock);
1702		skb_queue_splice_tail_init(sk_queue, rcvq);
1703		spin_unlock(&sk_queue->lock);
1704
1705		skb = __first_packet_length(sk, rcvq, &total);
1706	}
1707	res = skb ? skb->len : -1;
1708	if (total)
1709		udp_rmem_release(sk, total, 1, false);
1710	spin_unlock_bh(&rcvq->lock);
1711	return res;
1712}
1713
1714/*
1715 *	IOCTL requests applicable to the UDP protocol
1716 */
1717
1718int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1719{
1720	switch (cmd) {
1721	case SIOCOUTQ:
1722	{
1723		int amount = sk_wmem_alloc_get(sk);
1724
1725		return put_user(amount, (int __user *)arg);
1726	}
1727
1728	case SIOCINQ:
1729	{
1730		int amount = max_t(int, 0, first_packet_length(sk));
1731
1732		return put_user(amount, (int __user *)arg);
1733	}
1734
1735	default:
1736		return -ENOIOCTLCMD;
1737	}
1738
1739	return 0;
1740}
1741EXPORT_SYMBOL(udp_ioctl);
1742
1743struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1744			       int *off, int *err)
1745{
1746	struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1747	struct sk_buff_head *queue;
1748	struct sk_buff *last;
1749	long timeo;
1750	int error;
1751
1752	queue = &udp_sk(sk)->reader_queue;
1753	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1754	do {
1755		struct sk_buff *skb;
1756
1757		error = sock_error(sk);
1758		if (error)
1759			break;
1760
1761		error = -EAGAIN;
1762		do {
1763			spin_lock_bh(&queue->lock);
1764			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1765							err, &last);
1766			if (skb) {
1767				if (!(flags & MSG_PEEK))
1768					udp_skb_destructor(sk, skb);
1769				spin_unlock_bh(&queue->lock);
1770				return skb;
1771			}
1772
1773			if (skb_queue_empty_lockless(sk_queue)) {
1774				spin_unlock_bh(&queue->lock);
1775				goto busy_check;
1776			}
1777
1778			/* refill the reader queue and walk it again
1779			 * keep both queues locked to avoid re-acquiring
1780			 * the sk_receive_queue lock if fwd memory scheduling
1781			 * is needed.
1782			 */
1783			spin_lock(&sk_queue->lock);
1784			skb_queue_splice_tail_init(sk_queue, queue);
1785
1786			skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1787							err, &last);
1788			if (skb && !(flags & MSG_PEEK))
1789				udp_skb_dtor_locked(sk, skb);
1790			spin_unlock(&sk_queue->lock);
1791			spin_unlock_bh(&queue->lock);
1792			if (skb)
1793				return skb;
1794
1795busy_check:
1796			if (!sk_can_busy_loop(sk))
1797				break;
1798
1799			sk_busy_loop(sk, flags & MSG_DONTWAIT);
1800		} while (!skb_queue_empty_lockless(sk_queue));
1801
1802		/* sk_queue is empty, reader_queue may contain peeked packets */
1803	} while (timeo &&
1804		 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1805					      &error, &timeo,
1806					      (struct sk_buff *)sk_queue));
1807
1808	*err = error;
1809	return NULL;
1810}
1811EXPORT_SYMBOL(__skb_recv_udp);
1812
1813int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1814{
1815	struct sk_buff *skb;
1816	int err, copied;
1817
1818try_again:
1819	skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1820	if (!skb)
1821		return err;
1822
1823	if (udp_lib_checksum_complete(skb)) {
1824		int is_udplite = IS_UDPLITE(sk);
1825		struct net *net = sock_net(sk);
1826
1827		__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1828		__UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1829		atomic_inc(&sk->sk_drops);
1830		kfree_skb(skb);
1831		goto try_again;
1832	}
1833
1834	WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1835	copied = recv_actor(sk, skb);
1836	kfree_skb(skb);
1837
1838	return copied;
1839}
1840EXPORT_SYMBOL(udp_read_skb);
1841
1842/*
1843 * 	This should be easy, if there is something there we
1844 * 	return it, otherwise we block.
1845 */
1846
1847int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1848		int *addr_len)
1849{
1850	struct inet_sock *inet = inet_sk(sk);
1851	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1852	struct sk_buff *skb;
1853	unsigned int ulen, copied;
1854	int off, err, peeking = flags & MSG_PEEK;
1855	int is_udplite = IS_UDPLITE(sk);
1856	bool checksum_valid = false;
1857
1858	if (flags & MSG_ERRQUEUE)
1859		return ip_recv_error(sk, msg, len, addr_len);
1860
1861try_again:
1862	off = sk_peek_offset(sk, flags);
1863	skb = __skb_recv_udp(sk, flags, &off, &err);
1864	if (!skb)
1865		return err;
1866
1867	ulen = udp_skb_len(skb);
1868	copied = len;
1869	if (copied > ulen - off)
1870		copied = ulen - off;
1871	else if (copied < ulen)
1872		msg->msg_flags |= MSG_TRUNC;
1873
1874	/*
1875	 * If checksum is needed at all, try to do it while copying the
1876	 * data.  If the data is truncated, or if we only want a partial
1877	 * coverage checksum (UDP-Lite), do it before the copy.
1878	 */
1879
1880	if (copied < ulen || peeking ||
1881	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1882		checksum_valid = udp_skb_csum_unnecessary(skb) ||
1883				!__udp_lib_checksum_complete(skb);
1884		if (!checksum_valid)
1885			goto csum_copy_err;
1886	}
1887
1888	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1889		if (udp_skb_is_linear(skb))
1890			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1891		else
1892			err = skb_copy_datagram_msg(skb, off, msg, copied);
1893	} else {
1894		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1895
1896		if (err == -EINVAL)
1897			goto csum_copy_err;
1898	}
1899
1900	if (unlikely(err)) {
1901		if (!peeking) {
1902			atomic_inc(&sk->sk_drops);
1903			UDP_INC_STATS(sock_net(sk),
1904				      UDP_MIB_INERRORS, is_udplite);
1905		}
1906		kfree_skb(skb);
1907		return err;
1908	}
1909
1910	if (!peeking)
1911		UDP_INC_STATS(sock_net(sk),
1912			      UDP_MIB_INDATAGRAMS, is_udplite);
1913
1914	sock_recv_cmsgs(msg, sk, skb);
1915
1916	/* Copy the address. */
1917	if (sin) {
1918		sin->sin_family = AF_INET;
1919		sin->sin_port = udp_hdr(skb)->source;
1920		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1921		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1922		*addr_len = sizeof(*sin);
1923
1924		BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1925						      (struct sockaddr *)sin);
 
1926	}
1927
1928	if (udp_sk(sk)->gro_enabled)
1929		udp_cmsg_recv(msg, sk, skb);
1930
1931	if (inet->cmsg_flags)
1932		ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1933
1934	err = copied;
1935	if (flags & MSG_TRUNC)
1936		err = ulen;
1937
1938	skb_consume_udp(sk, skb, peeking ? -err : err);
1939	return err;
1940
1941csum_copy_err:
1942	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1943				 udp_skb_destructor)) {
1944		UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1945		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1946	}
1947	kfree_skb(skb);
1948
1949	/* starting over for a new packet, but check if we need to yield */
1950	cond_resched();
1951	msg->msg_flags &= ~MSG_TRUNC;
1952	goto try_again;
1953}
1954
1955int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1956{
1957	/* This check is replicated from __ip4_datagram_connect() and
1958	 * intended to prevent BPF program called below from accessing bytes
1959	 * that are out of the bound specified by user in addr_len.
1960	 */
1961	if (addr_len < sizeof(struct sockaddr_in))
1962		return -EINVAL;
1963
1964	return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1965}
1966EXPORT_SYMBOL(udp_pre_connect);
1967
 
 
 
 
 
 
 
 
 
 
 
 
1968int __udp_disconnect(struct sock *sk, int flags)
1969{
1970	struct inet_sock *inet = inet_sk(sk);
1971	/*
1972	 *	1003.1g - break association.
1973	 */
1974
1975	sk->sk_state = TCP_CLOSE;
1976	inet->inet_daddr = 0;
1977	inet->inet_dport = 0;
1978	sock_rps_reset_rxhash(sk);
1979	sk->sk_bound_dev_if = 0;
1980	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1981		inet_reset_saddr(sk);
1982		if (sk->sk_prot->rehash &&
1983		    (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1984			sk->sk_prot->rehash(sk);
1985	}
1986
1987	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1988		sk->sk_prot->unhash(sk);
1989		inet->inet_sport = 0;
1990	}
1991	sk_dst_reset(sk);
1992	return 0;
1993}
1994EXPORT_SYMBOL(__udp_disconnect);
1995
1996int udp_disconnect(struct sock *sk, int flags)
1997{
1998	lock_sock(sk);
1999	__udp_disconnect(sk, flags);
2000	release_sock(sk);
2001	return 0;
2002}
2003EXPORT_SYMBOL(udp_disconnect);
2004
2005void udp_lib_unhash(struct sock *sk)
2006{
2007	if (sk_hashed(sk)) {
2008		struct udp_table *udptable = udp_get_table_prot(sk);
2009		struct udp_hslot *hslot, *hslot2;
2010
2011		hslot  = udp_hashslot(udptable, sock_net(sk),
2012				      udp_sk(sk)->udp_port_hash);
2013		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2014
2015		spin_lock_bh(&hslot->lock);
2016		if (rcu_access_pointer(sk->sk_reuseport_cb))
2017			reuseport_detach_sock(sk);
2018		if (sk_del_node_init_rcu(sk)) {
2019			hslot->count--;
2020			inet_sk(sk)->inet_num = 0;
2021			sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2022
2023			spin_lock(&hslot2->lock);
2024			hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2025			hslot2->count--;
2026			spin_unlock(&hslot2->lock);
 
 
2027		}
2028		spin_unlock_bh(&hslot->lock);
2029	}
2030}
2031EXPORT_SYMBOL(udp_lib_unhash);
2032
2033/*
2034 * inet_rcv_saddr was changed, we must rehash secondary hash
2035 */
2036void udp_lib_rehash(struct sock *sk, u16 newhash)
2037{
2038	if (sk_hashed(sk)) {
2039		struct udp_table *udptable = udp_get_table_prot(sk);
2040		struct udp_hslot *hslot, *hslot2, *nhslot2;
2041
 
 
2042		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2043		nhslot2 = udp_hashslot2(udptable, newhash);
2044		udp_sk(sk)->udp_portaddr_hash = newhash;
2045
2046		if (hslot2 != nhslot2 ||
2047		    rcu_access_pointer(sk->sk_reuseport_cb)) {
2048			hslot = udp_hashslot(udptable, sock_net(sk),
2049					     udp_sk(sk)->udp_port_hash);
2050			/* we must lock primary chain too */
2051			spin_lock_bh(&hslot->lock);
2052			if (rcu_access_pointer(sk->sk_reuseport_cb))
2053				reuseport_detach_sock(sk);
2054
2055			if (hslot2 != nhslot2) {
2056				spin_lock(&hslot2->lock);
2057				hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2058				hslot2->count--;
2059				spin_unlock(&hslot2->lock);
2060
2061				spin_lock(&nhslot2->lock);
2062				hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2063							 &nhslot2->head);
2064				nhslot2->count++;
2065				spin_unlock(&nhslot2->lock);
2066			}
2067
2068			spin_unlock_bh(&hslot->lock);
2069		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2070	}
2071}
2072EXPORT_SYMBOL(udp_lib_rehash);
2073
2074void udp_v4_rehash(struct sock *sk)
2075{
2076	u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2077					  inet_sk(sk)->inet_rcv_saddr,
2078					  inet_sk(sk)->inet_num);
2079	udp_lib_rehash(sk, new_hash);
 
 
 
 
2080}
2081
2082static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2083{
2084	int rc;
2085
2086	if (inet_sk(sk)->inet_daddr) {
2087		sock_rps_save_rxhash(sk, skb);
2088		sk_mark_napi_id(sk, skb);
2089		sk_incoming_cpu_update(sk);
2090	} else {
2091		sk_mark_napi_id_once(sk, skb);
2092	}
2093
2094	rc = __udp_enqueue_schedule_skb(sk, skb);
2095	if (rc < 0) {
2096		int is_udplite = IS_UDPLITE(sk);
2097		int drop_reason;
2098
2099		/* Note that an ENOMEM error is charged twice */
2100		if (rc == -ENOMEM) {
2101			UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2102					is_udplite);
2103			drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2104		} else {
2105			UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2106				      is_udplite);
2107			drop_reason = SKB_DROP_REASON_PROTO_MEM;
2108		}
2109		UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2110		kfree_skb_reason(skb, drop_reason);
2111		trace_udp_fail_queue_rcv_skb(rc, sk);
2112		return -1;
2113	}
2114
2115	return 0;
2116}
2117
2118/* returns:
2119 *  -1: error
2120 *   0: success
2121 *  >0: "udp encap" protocol resubmission
2122 *
2123 * Note that in the success and error cases, the skb is assumed to
2124 * have either been requeued or freed.
2125 */
2126static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2127{
2128	int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2129	struct udp_sock *up = udp_sk(sk);
2130	int is_udplite = IS_UDPLITE(sk);
2131
2132	/*
2133	 *	Charge it to the socket, dropping if the queue is full.
2134	 */
2135	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2136		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2137		goto drop;
2138	}
2139	nf_reset_ct(skb);
2140
2141	if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
 
2142		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2143
2144		/*
2145		 * This is an encapsulation socket so pass the skb to
2146		 * the socket's udp_encap_rcv() hook. Otherwise, just
2147		 * fall through and pass this up the UDP socket.
2148		 * up->encap_rcv() returns the following value:
2149		 * =0 if skb was successfully passed to the encap
2150		 *    handler or was discarded by it.
2151		 * >0 if skb should be passed on to UDP.
2152		 * <0 if skb should be resubmitted as proto -N
2153		 */
2154
2155		/* if we're overly short, let UDP handle it */
2156		encap_rcv = READ_ONCE(up->encap_rcv);
2157		if (encap_rcv) {
2158			int ret;
2159
2160			/* Verify checksum before giving to encap */
2161			if (udp_lib_checksum_complete(skb))
2162				goto csum_error;
2163
2164			ret = encap_rcv(sk, skb);
2165			if (ret <= 0) {
2166				__UDP_INC_STATS(sock_net(sk),
2167						UDP_MIB_INDATAGRAMS,
2168						is_udplite);
2169				return -ret;
2170			}
2171		}
2172
2173		/* FALLTHROUGH -- it's a UDP Packet */
2174	}
2175
2176	/*
2177	 * 	UDP-Lite specific tests, ignored on UDP sockets
2178	 */
2179	if ((up->pcflag & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
 
2180
2181		/*
2182		 * MIB statistics other than incrementing the error count are
2183		 * disabled for the following two types of errors: these depend
2184		 * on the application settings, not on the functioning of the
2185		 * protocol stack as such.
2186		 *
2187		 * RFC 3828 here recommends (sec 3.3): "There should also be a
2188		 * way ... to ... at least let the receiving application block
2189		 * delivery of packets with coverage values less than a value
2190		 * provided by the application."
2191		 */
2192		if (up->pcrlen == 0) {          /* full coverage was set  */
2193			net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2194					    UDP_SKB_CB(skb)->cscov, skb->len);
2195			goto drop;
2196		}
2197		/* The next case involves violating the min. coverage requested
2198		 * by the receiver. This is subtle: if receiver wants x and x is
2199		 * greater than the buffersize/MTU then receiver will complain
2200		 * that it wants x while sender emits packets of smaller size y.
2201		 * Therefore the above ...()->partial_cov statement is essential.
2202		 */
2203		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
2204			net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2205					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
2206			goto drop;
2207		}
2208	}
2209
2210	prefetch(&sk->sk_rmem_alloc);
2211	if (rcu_access_pointer(sk->sk_filter) &&
2212	    udp_lib_checksum_complete(skb))
2213			goto csum_error;
2214
2215	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2216		drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2217		goto drop;
2218	}
2219
2220	udp_csum_pull_header(skb);
2221
2222	ipv4_pktinfo_prepare(sk, skb);
2223	return __udp_queue_rcv_skb(sk, skb);
2224
2225csum_error:
2226	drop_reason = SKB_DROP_REASON_UDP_CSUM;
2227	__UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2228drop:
2229	__UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2230	atomic_inc(&sk->sk_drops);
2231	kfree_skb_reason(skb, drop_reason);
2232	return -1;
2233}
2234
2235static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2236{
2237	struct sk_buff *next, *segs;
2238	int ret;
2239
2240	if (likely(!udp_unexpected_gso(sk, skb)))
2241		return udp_queue_rcv_one_skb(sk, skb);
2242
2243	BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2244	__skb_push(skb, -skb_mac_offset(skb));
2245	segs = udp_rcv_segment(sk, skb, true);
2246	skb_list_walk_safe(segs, skb, next) {
2247		__skb_pull(skb, skb_transport_offset(skb));
2248
2249		udp_post_segment_fix_csum(skb);
2250		ret = udp_queue_rcv_one_skb(sk, skb);
2251		if (ret > 0)
2252			ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2253	}
2254	return 0;
2255}
2256
2257/* For TCP sockets, sk_rx_dst is protected by socket lock
2258 * For UDP, we use xchg() to guard against concurrent changes.
2259 */
2260bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2261{
2262	struct dst_entry *old;
2263
2264	if (dst_hold_safe(dst)) {
2265		old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst);
2266		dst_release(old);
2267		return old != dst;
2268	}
2269	return false;
2270}
2271EXPORT_SYMBOL(udp_sk_rx_dst_set);
2272
2273/*
2274 *	Multicasts and broadcasts go to each listener.
2275 *
2276 *	Note: called only from the BH handler context.
2277 */
2278static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2279				    struct udphdr  *uh,
2280				    __be32 saddr, __be32 daddr,
2281				    struct udp_table *udptable,
2282				    int proto)
2283{
2284	struct sock *sk, *first = NULL;
2285	unsigned short hnum = ntohs(uh->dest);
2286	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2287	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2288	unsigned int offset = offsetof(typeof(*sk), sk_node);
2289	int dif = skb->dev->ifindex;
2290	int sdif = inet_sdif(skb);
2291	struct hlist_node *node;
2292	struct sk_buff *nskb;
2293
2294	if (use_hash2) {
2295		hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2296			    udptable->mask;
2297		hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2298start_lookup:
2299		hslot = &udptable->hash2[hash2];
2300		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2301	}
2302
2303	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2304		if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2305					 uh->source, saddr, dif, sdif, hnum))
2306			continue;
2307
2308		if (!first) {
2309			first = sk;
2310			continue;
2311		}
2312		nskb = skb_clone(skb, GFP_ATOMIC);
2313
2314		if (unlikely(!nskb)) {
2315			atomic_inc(&sk->sk_drops);
2316			__UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2317					IS_UDPLITE(sk));
2318			__UDP_INC_STATS(net, UDP_MIB_INERRORS,
2319					IS_UDPLITE(sk));
2320			continue;
2321		}
2322		if (udp_queue_rcv_skb(sk, nskb) > 0)
2323			consume_skb(nskb);
2324	}
2325
2326	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
2327	if (use_hash2 && hash2 != hash2_any) {
2328		hash2 = hash2_any;
2329		goto start_lookup;
2330	}
2331
2332	if (first) {
2333		if (udp_queue_rcv_skb(first, skb) > 0)
2334			consume_skb(skb);
2335	} else {
2336		kfree_skb(skb);
2337		__UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2338				proto == IPPROTO_UDPLITE);
2339	}
2340	return 0;
2341}
2342
2343/* Initialize UDP checksum. If exited with zero value (success),
2344 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2345 * Otherwise, csum completion requires checksumming packet body,
2346 * including udp header and folding it to skb->csum.
2347 */
2348static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2349				 int proto)
2350{
2351	int err;
2352
2353	UDP_SKB_CB(skb)->partial_cov = 0;
2354	UDP_SKB_CB(skb)->cscov = skb->len;
2355
2356	if (proto == IPPROTO_UDPLITE) {
2357		err = udplite_checksum_init(skb, uh);
2358		if (err)
2359			return err;
2360
2361		if (UDP_SKB_CB(skb)->partial_cov) {
2362			skb->csum = inet_compute_pseudo(skb, proto);
2363			return 0;
2364		}
2365	}
2366
2367	/* Note, we are only interested in != 0 or == 0, thus the
2368	 * force to int.
2369	 */
2370	err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2371							inet_compute_pseudo);
2372	if (err)
2373		return err;
2374
2375	if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2376		/* If SW calculated the value, we know it's bad */
2377		if (skb->csum_complete_sw)
2378			return 1;
2379
2380		/* HW says the value is bad. Let's validate that.
2381		 * skb->csum is no longer the full packet checksum,
2382		 * so don't treat it as such.
2383		 */
2384		skb_checksum_complete_unset(skb);
2385	}
2386
2387	return 0;
2388}
2389
2390/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2391 * return code conversion for ip layer consumption
2392 */
2393static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2394			       struct udphdr *uh)
2395{
2396	int ret;
2397
2398	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2399		skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2400
2401	ret = udp_queue_rcv_skb(sk, skb);
2402
2403	/* a return value > 0 means to resubmit the input, but
2404	 * it wants the return to be -protocol, or 0
2405	 */
2406	if (ret > 0)
2407		return -ret;
2408	return 0;
2409}
2410
2411/*
2412 *	All we need to do is get the socket, and then do a checksum.
2413 */
2414
2415int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2416		   int proto)
2417{
2418	struct sock *sk;
2419	struct udphdr *uh;
2420	unsigned short ulen;
2421	struct rtable *rt = skb_rtable(skb);
2422	__be32 saddr, daddr;
2423	struct net *net = dev_net(skb->dev);
2424	bool refcounted;
2425	int drop_reason;
2426
2427	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2428
2429	/*
2430	 *  Validate the packet.
2431	 */
2432	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2433		goto drop;		/* No space for header. */
2434
2435	uh   = udp_hdr(skb);
2436	ulen = ntohs(uh->len);
2437	saddr = ip_hdr(skb)->saddr;
2438	daddr = ip_hdr(skb)->daddr;
2439
2440	if (ulen > skb->len)
2441		goto short_packet;
2442
2443	if (proto == IPPROTO_UDP) {
2444		/* UDP validates ulen. */
2445		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2446			goto short_packet;
2447		uh = udp_hdr(skb);
2448	}
2449
2450	if (udp4_csum_init(skb, uh, proto))
2451		goto csum_error;
2452
2453	sk = skb_steal_sock(skb, &refcounted);
 
 
 
 
2454	if (sk) {
2455		struct dst_entry *dst = skb_dst(skb);
2456		int ret;
2457
2458		if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2459			udp_sk_rx_dst_set(sk, dst);
2460
2461		ret = udp_unicast_rcv_skb(sk, skb, uh);
2462		if (refcounted)
2463			sock_put(sk);
2464		return ret;
2465	}
2466
2467	if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2468		return __udp4_lib_mcast_deliver(net, skb, uh,
2469						saddr, daddr, udptable, proto);
2470
2471	sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2472	if (sk)
2473		return udp_unicast_rcv_skb(sk, skb, uh);
2474
2475	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2476		goto drop;
2477	nf_reset_ct(skb);
2478
2479	/* No socket. Drop packet silently, if checksum is wrong */
2480	if (udp_lib_checksum_complete(skb))
2481		goto csum_error;
2482
2483	drop_reason = SKB_DROP_REASON_NO_SOCKET;
2484	__UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2485	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2486
2487	/*
2488	 * Hmm.  We got an UDP packet to a port to which we
2489	 * don't wanna listen.  Ignore it.
2490	 */
2491	kfree_skb_reason(skb, drop_reason);
2492	return 0;
2493
2494short_packet:
2495	drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2496	net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2497			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2498			    &saddr, ntohs(uh->source),
2499			    ulen, skb->len,
2500			    &daddr, ntohs(uh->dest));
2501	goto drop;
2502
2503csum_error:
2504	/*
2505	 * RFC1122: OK.  Discards the bad packet silently (as far as
2506	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2507	 */
2508	drop_reason = SKB_DROP_REASON_UDP_CSUM;
2509	net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2510			    proto == IPPROTO_UDPLITE ? "Lite" : "",
2511			    &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2512			    ulen);
2513	__UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2514drop:
2515	__UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2516	kfree_skb_reason(skb, drop_reason);
2517	return 0;
2518}
2519
2520/* We can only early demux multicast if there is a single matching socket.
2521 * If more than one socket found returns NULL
2522 */
2523static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2524						  __be16 loc_port, __be32 loc_addr,
2525						  __be16 rmt_port, __be32 rmt_addr,
2526						  int dif, int sdif)
2527{
2528	struct udp_table *udptable = net->ipv4.udp_table;
2529	unsigned short hnum = ntohs(loc_port);
2530	struct sock *sk, *result;
2531	struct udp_hslot *hslot;
2532	unsigned int slot;
2533
2534	slot = udp_hashfn(net, hnum, udptable->mask);
2535	hslot = &udptable->hash[slot];
2536
2537	/* Do not bother scanning a too big list */
2538	if (hslot->count > 10)
2539		return NULL;
2540
2541	result = NULL;
2542	sk_for_each_rcu(sk, &hslot->head) {
2543		if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2544					rmt_port, rmt_addr, dif, sdif, hnum)) {
2545			if (result)
2546				return NULL;
2547			result = sk;
2548		}
2549	}
2550
2551	return result;
2552}
2553
2554/* For unicast we should only early demux connected sockets or we can
2555 * break forwarding setups.  The chains here can be long so only check
2556 * if the first socket is an exact match and if not move on.
2557 */
2558static struct sock *__udp4_lib_demux_lookup(struct net *net,
2559					    __be16 loc_port, __be32 loc_addr,
2560					    __be16 rmt_port, __be32 rmt_addr,
2561					    int dif, int sdif)
2562{
2563	struct udp_table *udptable = net->ipv4.udp_table;
2564	INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2565	unsigned short hnum = ntohs(loc_port);
2566	unsigned int hash2, slot2;
2567	struct udp_hslot *hslot2;
 
2568	__portpair ports;
2569	struct sock *sk;
2570
2571	hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2572	slot2 = hash2 & udptable->mask;
2573	hslot2 = &udptable->hash2[slot2];
2574	ports = INET_COMBINED_PORTS(rmt_port, hnum);
2575
2576	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2577		if (inet_match(net, sk, acookie, ports, dif, sdif))
2578			return sk;
2579		/* Only check first socket in chain */
2580		break;
2581	}
2582	return NULL;
2583}
2584
2585int udp_v4_early_demux(struct sk_buff *skb)
2586{
2587	struct net *net = dev_net(skb->dev);
2588	struct in_device *in_dev = NULL;
2589	const struct iphdr *iph;
2590	const struct udphdr *uh;
2591	struct sock *sk = NULL;
2592	struct dst_entry *dst;
2593	int dif = skb->dev->ifindex;
2594	int sdif = inet_sdif(skb);
2595	int ours;
2596
2597	/* validate the packet */
2598	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2599		return 0;
2600
2601	iph = ip_hdr(skb);
2602	uh = udp_hdr(skb);
2603
2604	if (skb->pkt_type == PACKET_MULTICAST) {
2605		in_dev = __in_dev_get_rcu(skb->dev);
2606
2607		if (!in_dev)
2608			return 0;
2609
2610		ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2611				       iph->protocol);
2612		if (!ours)
2613			return 0;
2614
2615		sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2616						   uh->source, iph->saddr,
2617						   dif, sdif);
2618	} else if (skb->pkt_type == PACKET_HOST) {
2619		sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2620					     uh->source, iph->saddr, dif, sdif);
2621	}
2622
2623	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2624		return 0;
2625
2626	skb->sk = sk;
2627	skb->destructor = sock_efree;
 
2628	dst = rcu_dereference(sk->sk_rx_dst);
2629
2630	if (dst)
2631		dst = dst_check(dst, 0);
2632	if (dst) {
2633		u32 itag = 0;
2634
2635		/* set noref for now.
2636		 * any place which wants to hold dst has to call
2637		 * dst_hold_safe()
2638		 */
2639		skb_dst_set_noref(skb, dst);
2640
2641		/* for unconnected multicast sockets we need to validate
2642		 * the source on each packet
2643		 */
2644		if (!inet_sk(sk)->inet_daddr && in_dev)
2645			return ip_mc_validate_source(skb, iph->daddr,
2646						     iph->saddr,
2647						     iph->tos & IPTOS_RT_MASK,
2648						     skb->dev, in_dev, &itag);
2649	}
2650	return 0;
2651}
2652
2653int udp_rcv(struct sk_buff *skb)
2654{
2655	return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2656}
2657
2658void udp_destroy_sock(struct sock *sk)
2659{
2660	struct udp_sock *up = udp_sk(sk);
2661	bool slow = lock_sock_fast(sk);
2662
2663	/* protects from races with udp_abort() */
2664	sock_set_flag(sk, SOCK_DEAD);
2665	udp_flush_pending_frames(sk);
2666	unlock_sock_fast(sk, slow);
2667	if (static_branch_unlikely(&udp_encap_needed_key)) {
2668		if (up->encap_type) {
2669			void (*encap_destroy)(struct sock *sk);
2670			encap_destroy = READ_ONCE(up->encap_destroy);
2671			if (encap_destroy)
2672				encap_destroy(sk);
2673		}
2674		if (up->encap_enabled)
2675			static_branch_dec(&udp_encap_needed_key);
2676	}
2677}
2678
 
 
 
 
 
 
 
 
 
 
 
 
 
2679/*
2680 *	Socket option code for UDP
2681 */
2682int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2683		       sockptr_t optval, unsigned int optlen,
2684		       int (*push_pending_frames)(struct sock *))
2685{
2686	struct udp_sock *up = udp_sk(sk);
2687	int val, valbool;
2688	int err = 0;
2689	int is_udplite = IS_UDPLITE(sk);
2690
2691	if (level == SOL_SOCKET) {
2692		err = sk_setsockopt(sk, level, optname, optval, optlen);
2693
2694		if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2695			sockopt_lock_sock(sk);
2696			/* paired with READ_ONCE in udp_rmem_release() */
2697			WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2698			sockopt_release_sock(sk);
2699		}
2700		return err;
2701	}
2702
2703	if (optlen < sizeof(int))
2704		return -EINVAL;
2705
2706	if (copy_from_sockptr(&val, optval, sizeof(val)))
2707		return -EFAULT;
2708
2709	valbool = val ? 1 : 0;
2710
2711	switch (optname) {
2712	case UDP_CORK:
2713		if (val != 0) {
2714			WRITE_ONCE(up->corkflag, 1);
2715		} else {
2716			WRITE_ONCE(up->corkflag, 0);
2717			lock_sock(sk);
2718			push_pending_frames(sk);
2719			release_sock(sk);
2720		}
2721		break;
2722
2723	case UDP_ENCAP:
2724		switch (val) {
2725		case 0:
2726#ifdef CONFIG_XFRM
2727		case UDP_ENCAP_ESPINUDP:
2728		case UDP_ENCAP_ESPINUDP_NON_IKE:
2729#if IS_ENABLED(CONFIG_IPV6)
2730			if (sk->sk_family == AF_INET6)
2731				up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv;
 
2732			else
2733#endif
2734				up->encap_rcv = xfrm4_udp_encap_rcv;
 
2735#endif
2736			fallthrough;
2737		case UDP_ENCAP_L2TPINUDP:
2738			up->encap_type = val;
2739			lock_sock(sk);
2740			udp_tunnel_encap_enable(sk->sk_socket);
2741			release_sock(sk);
2742			break;
2743		default:
2744			err = -ENOPROTOOPT;
2745			break;
2746		}
2747		break;
2748
2749	case UDP_NO_CHECK6_TX:
2750		up->no_check6_tx = valbool;
2751		break;
2752
2753	case UDP_NO_CHECK6_RX:
2754		up->no_check6_rx = valbool;
2755		break;
2756
2757	case UDP_SEGMENT:
2758		if (val < 0 || val > USHRT_MAX)
2759			return -EINVAL;
2760		WRITE_ONCE(up->gso_size, val);
2761		break;
2762
2763	case UDP_GRO:
2764		lock_sock(sk);
2765
2766		/* when enabling GRO, accept the related GSO packet type */
2767		if (valbool)
2768			udp_tunnel_encap_enable(sk->sk_socket);
2769		up->gro_enabled = valbool;
2770		up->accept_udp_l4 = valbool;
2771		release_sock(sk);
2772		break;
2773
2774	/*
2775	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
2776	 */
2777	/* The sender sets actual checksum coverage length via this option.
2778	 * The case coverage > packet length is handled by send module. */
2779	case UDPLITE_SEND_CSCOV:
2780		if (!is_udplite)         /* Disable the option on UDP sockets */
2781			return -ENOPROTOOPT;
2782		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2783			val = 8;
2784		else if (val > USHRT_MAX)
2785			val = USHRT_MAX;
2786		up->pcslen = val;
2787		up->pcflag |= UDPLITE_SEND_CC;
2788		break;
2789
2790	/* The receiver specifies a minimum checksum coverage value. To make
2791	 * sense, this should be set to at least 8 (as done below). If zero is
2792	 * used, this again means full checksum coverage.                     */
2793	case UDPLITE_RECV_CSCOV:
2794		if (!is_udplite)         /* Disable the option on UDP sockets */
2795			return -ENOPROTOOPT;
2796		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2797			val = 8;
2798		else if (val > USHRT_MAX)
2799			val = USHRT_MAX;
2800		up->pcrlen = val;
2801		up->pcflag |= UDPLITE_RECV_CC;
2802		break;
2803
2804	default:
2805		err = -ENOPROTOOPT;
2806		break;
2807	}
2808
2809	return err;
2810}
2811EXPORT_SYMBOL(udp_lib_setsockopt);
2812
2813int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2814		   unsigned int optlen)
2815{
2816	if (level == SOL_UDP  ||  level == SOL_UDPLITE || level == SOL_SOCKET)
2817		return udp_lib_setsockopt(sk, level, optname,
2818					  optval, optlen,
2819					  udp_push_pending_frames);
2820	return ip_setsockopt(sk, level, optname, optval, optlen);
2821}
2822
2823int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2824		       char __user *optval, int __user *optlen)
2825{
2826	struct udp_sock *up = udp_sk(sk);
2827	int val, len;
2828
2829	if (get_user(len, optlen))
2830		return -EFAULT;
2831
2832	len = min_t(unsigned int, len, sizeof(int));
2833
2834	if (len < 0)
2835		return -EINVAL;
2836
 
 
2837	switch (optname) {
2838	case UDP_CORK:
2839		val = READ_ONCE(up->corkflag);
2840		break;
2841
2842	case UDP_ENCAP:
2843		val = up->encap_type;
2844		break;
2845
2846	case UDP_NO_CHECK6_TX:
2847		val = up->no_check6_tx;
2848		break;
2849
2850	case UDP_NO_CHECK6_RX:
2851		val = up->no_check6_rx;
2852		break;
2853
2854	case UDP_SEGMENT:
2855		val = READ_ONCE(up->gso_size);
2856		break;
2857
2858	case UDP_GRO:
2859		val = up->gro_enabled;
2860		break;
2861
2862	/* The following two cannot be changed on UDP sockets, the return is
2863	 * always 0 (which corresponds to the full checksum coverage of UDP). */
2864	case UDPLITE_SEND_CSCOV:
2865		val = up->pcslen;
2866		break;
2867
2868	case UDPLITE_RECV_CSCOV:
2869		val = up->pcrlen;
2870		break;
2871
2872	default:
2873		return -ENOPROTOOPT;
2874	}
2875
2876	if (put_user(len, optlen))
2877		return -EFAULT;
2878	if (copy_to_user(optval, &val, len))
2879		return -EFAULT;
2880	return 0;
2881}
2882EXPORT_SYMBOL(udp_lib_getsockopt);
2883
2884int udp_getsockopt(struct sock *sk, int level, int optname,
2885		   char __user *optval, int __user *optlen)
2886{
2887	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2888		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2889	return ip_getsockopt(sk, level, optname, optval, optlen);
2890}
2891
2892/**
2893 * 	udp_poll - wait for a UDP event.
2894 *	@file: - file struct
2895 *	@sock: - socket
2896 *	@wait: - poll table
2897 *
2898 *	This is same as datagram poll, except for the special case of
2899 *	blocking sockets. If application is using a blocking fd
2900 *	and a packet with checksum error is in the queue;
2901 *	then it could get return from select indicating data available
2902 *	but then block when reading it. Add special case code
2903 *	to work around these arguably broken applications.
2904 */
2905__poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2906{
2907	__poll_t mask = datagram_poll(file, sock, wait);
2908	struct sock *sk = sock->sk;
2909
2910	if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2911		mask |= EPOLLIN | EPOLLRDNORM;
2912
2913	/* Check for false positives due to checksum errors */
2914	if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2915	    !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2916		mask &= ~(EPOLLIN | EPOLLRDNORM);
2917
2918	/* psock ingress_msg queue should not contain any bad checksum frames */
2919	if (sk_is_readable(sk))
2920		mask |= EPOLLIN | EPOLLRDNORM;
2921	return mask;
2922
2923}
2924EXPORT_SYMBOL(udp_poll);
2925
2926int udp_abort(struct sock *sk, int err)
2927{
2928	lock_sock(sk);
 
2929
2930	/* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2931	 * with close()
2932	 */
2933	if (sock_flag(sk, SOCK_DEAD))
2934		goto out;
2935
2936	sk->sk_err = err;
2937	sk_error_report(sk);
2938	__udp_disconnect(sk, 0);
2939
2940out:
2941	release_sock(sk);
 
2942
2943	return 0;
2944}
2945EXPORT_SYMBOL_GPL(udp_abort);
2946
2947struct proto udp_prot = {
2948	.name			= "UDP",
2949	.owner			= THIS_MODULE,
2950	.close			= udp_lib_close,
2951	.pre_connect		= udp_pre_connect,
2952	.connect		= ip4_datagram_connect,
2953	.disconnect		= udp_disconnect,
2954	.ioctl			= udp_ioctl,
2955	.init			= udp_init_sock,
2956	.destroy		= udp_destroy_sock,
2957	.setsockopt		= udp_setsockopt,
2958	.getsockopt		= udp_getsockopt,
2959	.sendmsg		= udp_sendmsg,
2960	.recvmsg		= udp_recvmsg,
2961	.sendpage		= udp_sendpage,
2962	.release_cb		= ip4_datagram_release_cb,
2963	.hash			= udp_lib_hash,
2964	.unhash			= udp_lib_unhash,
2965	.rehash			= udp_v4_rehash,
2966	.get_port		= udp_v4_get_port,
2967	.put_port		= udp_lib_unhash,
2968#ifdef CONFIG_BPF_SYSCALL
2969	.psock_update_sk_prot	= udp_bpf_update_proto,
2970#endif
2971	.memory_allocated	= &udp_memory_allocated,
2972	.per_cpu_fw_alloc	= &udp_memory_per_cpu_fw_alloc,
2973
2974	.sysctl_mem		= sysctl_udp_mem,
2975	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2976	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2977	.obj_size		= sizeof(struct udp_sock),
2978	.h.udp_table		= NULL,
2979	.diag_destroy		= udp_abort,
2980};
2981EXPORT_SYMBOL(udp_prot);
2982
2983/* ------------------------------------------------------------------------ */
2984#ifdef CONFIG_PROC_FS
2985
2986static struct udp_table *udp_get_table_afinfo(struct udp_seq_afinfo *afinfo,
2987					      struct net *net)
 
 
 
 
 
 
 
 
 
 
 
 
 
2988{
 
 
 
 
 
 
 
 
2989	return afinfo->udp_table ? : net->ipv4.udp_table;
2990}
2991
2992static struct sock *udp_get_first(struct seq_file *seq, int start)
2993{
2994	struct udp_iter_state *state = seq->private;
2995	struct net *net = seq_file_net(seq);
2996	struct udp_seq_afinfo *afinfo;
2997	struct udp_table *udptable;
2998	struct sock *sk;
2999
3000	if (state->bpf_seq_afinfo)
3001		afinfo = state->bpf_seq_afinfo;
3002	else
3003		afinfo = pde_data(file_inode(seq->file));
3004
3005	udptable = udp_get_table_afinfo(afinfo, net);
3006
3007	for (state->bucket = start; state->bucket <= udptable->mask;
3008	     ++state->bucket) {
3009		struct udp_hslot *hslot = &udptable->hash[state->bucket];
3010
3011		if (hlist_empty(&hslot->head))
3012			continue;
3013
3014		spin_lock_bh(&hslot->lock);
3015		sk_for_each(sk, &hslot->head) {
3016			if (!net_eq(sock_net(sk), net))
3017				continue;
3018			if (afinfo->family == AF_UNSPEC ||
3019			    sk->sk_family == afinfo->family)
3020				goto found;
3021		}
3022		spin_unlock_bh(&hslot->lock);
3023	}
3024	sk = NULL;
3025found:
3026	return sk;
3027}
3028
3029static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3030{
3031	struct udp_iter_state *state = seq->private;
3032	struct net *net = seq_file_net(seq);
3033	struct udp_seq_afinfo *afinfo;
3034	struct udp_table *udptable;
3035
3036	if (state->bpf_seq_afinfo)
3037		afinfo = state->bpf_seq_afinfo;
3038	else
3039		afinfo = pde_data(file_inode(seq->file));
3040
3041	do {
3042		sk = sk_next(sk);
3043	} while (sk && (!net_eq(sock_net(sk), net) ||
3044			(afinfo->family != AF_UNSPEC &&
3045			 sk->sk_family != afinfo->family)));
3046
3047	if (!sk) {
3048		udptable = udp_get_table_afinfo(afinfo, net);
3049
3050		if (state->bucket <= udptable->mask)
3051			spin_unlock_bh(&udptable->hash[state->bucket].lock);
3052
3053		return udp_get_first(seq, state->bucket + 1);
3054	}
3055	return sk;
3056}
3057
3058static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3059{
3060	struct sock *sk = udp_get_first(seq, 0);
3061
3062	if (sk)
3063		while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3064			--pos;
3065	return pos ? NULL : sk;
3066}
3067
3068void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3069{
3070	struct udp_iter_state *state = seq->private;
3071	state->bucket = MAX_UDP_PORTS;
3072
3073	return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3074}
3075EXPORT_SYMBOL(udp_seq_start);
3076
3077void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3078{
3079	struct sock *sk;
3080
3081	if (v == SEQ_START_TOKEN)
3082		sk = udp_get_idx(seq, 0);
3083	else
3084		sk = udp_get_next(seq, v);
3085
3086	++*pos;
3087	return sk;
3088}
3089EXPORT_SYMBOL(udp_seq_next);
3090
3091void udp_seq_stop(struct seq_file *seq, void *v)
3092{
3093	struct udp_iter_state *state = seq->private;
3094	struct udp_seq_afinfo *afinfo;
3095	struct udp_table *udptable;
3096
3097	if (state->bpf_seq_afinfo)
3098		afinfo = state->bpf_seq_afinfo;
3099	else
3100		afinfo = pde_data(file_inode(seq->file));
3101
3102	udptable = udp_get_table_afinfo(afinfo, seq_file_net(seq));
3103
3104	if (state->bucket <= udptable->mask)
3105		spin_unlock_bh(&udptable->hash[state->bucket].lock);
3106}
3107EXPORT_SYMBOL(udp_seq_stop);
3108
3109/* ------------------------------------------------------------------------ */
3110static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3111		int bucket)
3112{
3113	struct inet_sock *inet = inet_sk(sp);
3114	__be32 dest = inet->inet_daddr;
3115	__be32 src  = inet->inet_rcv_saddr;
3116	__u16 destp	  = ntohs(inet->inet_dport);
3117	__u16 srcp	  = ntohs(inet->inet_sport);
3118
3119	seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3120		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3121		bucket, src, srcp, dest, destp, sp->sk_state,
3122		sk_wmem_alloc_get(sp),
3123		udp_rqueue_get(sp),
3124		0, 0L, 0,
3125		from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3126		0, sock_i_ino(sp),
3127		refcount_read(&sp->sk_refcnt), sp,
3128		atomic_read(&sp->sk_drops));
3129}
3130
3131int udp4_seq_show(struct seq_file *seq, void *v)
3132{
3133	seq_setwidth(seq, 127);
3134	if (v == SEQ_START_TOKEN)
3135		seq_puts(seq, "   sl  local_address rem_address   st tx_queue "
3136			   "rx_queue tr tm->when retrnsmt   uid  timeout "
3137			   "inode ref pointer drops");
3138	else {
3139		struct udp_iter_state *state = seq->private;
3140
3141		udp4_format_sock(v, seq, state->bucket);
3142	}
3143	seq_pad(seq, '\n');
3144	return 0;
3145}
3146
3147#ifdef CONFIG_BPF_SYSCALL
3148struct bpf_iter__udp {
3149	__bpf_md_ptr(struct bpf_iter_meta *, meta);
3150	__bpf_md_ptr(struct udp_sock *, udp_sk);
3151	uid_t uid __aligned(8);
3152	int bucket __aligned(8);
3153};
3154
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3155static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3156			     struct udp_sock *udp_sk, uid_t uid, int bucket)
3157{
3158	struct bpf_iter__udp ctx;
3159
3160	meta->seq_num--;  /* skip SEQ_START_TOKEN */
3161	ctx.meta = meta;
3162	ctx.udp_sk = udp_sk;
3163	ctx.uid = uid;
3164	ctx.bucket = bucket;
3165	return bpf_iter_run_prog(prog, &ctx);
3166}
3167
3168static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3169{
3170	struct udp_iter_state *state = seq->private;
3171	struct bpf_iter_meta meta;
3172	struct bpf_prog *prog;
3173	struct sock *sk = v;
3174	uid_t uid;
 
3175
3176	if (v == SEQ_START_TOKEN)
3177		return 0;
3178
 
 
 
 
 
 
 
3179	uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3180	meta.seq = seq;
3181	prog = bpf_iter_get_info(&meta, false);
3182	return udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
 
 
 
 
 
 
 
 
 
 
3183}
3184
3185static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3186{
 
3187	struct bpf_iter_meta meta;
3188	struct bpf_prog *prog;
3189
3190	if (!v) {
3191		meta.seq = seq;
3192		prog = bpf_iter_get_info(&meta, true);
3193		if (prog)
3194			(void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3195	}
3196
3197	udp_seq_stop(seq, v);
 
 
 
3198}
3199
3200static const struct seq_operations bpf_iter_udp_seq_ops = {
3201	.start		= udp_seq_start,
3202	.next		= udp_seq_next,
3203	.stop		= bpf_iter_udp_seq_stop,
3204	.show		= bpf_iter_udp_seq_show,
3205};
3206#endif
3207
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3208const struct seq_operations udp_seq_ops = {
3209	.start		= udp_seq_start,
3210	.next		= udp_seq_next,
3211	.stop		= udp_seq_stop,
3212	.show		= udp4_seq_show,
3213};
3214EXPORT_SYMBOL(udp_seq_ops);
3215
3216static struct udp_seq_afinfo udp4_seq_afinfo = {
3217	.family		= AF_INET,
3218	.udp_table	= NULL,
3219};
3220
3221static int __net_init udp4_proc_init_net(struct net *net)
3222{
3223	if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3224			sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3225		return -ENOMEM;
3226	return 0;
3227}
3228
3229static void __net_exit udp4_proc_exit_net(struct net *net)
3230{
3231	remove_proc_entry("udp", net->proc_net);
3232}
3233
3234static struct pernet_operations udp4_net_ops = {
3235	.init = udp4_proc_init_net,
3236	.exit = udp4_proc_exit_net,
3237};
3238
3239int __init udp4_proc_init(void)
3240{
3241	return register_pernet_subsys(&udp4_net_ops);
3242}
3243
3244void udp4_proc_exit(void)
3245{
3246	unregister_pernet_subsys(&udp4_net_ops);
3247}
3248#endif /* CONFIG_PROC_FS */
3249
3250static __initdata unsigned long uhash_entries;
3251static int __init set_uhash_entries(char *str)
3252{
3253	ssize_t ret;
3254
3255	if (!str)
3256		return 0;
3257
3258	ret = kstrtoul(str, 0, &uhash_entries);
3259	if (ret)
3260		return 0;
3261
3262	if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3263		uhash_entries = UDP_HTABLE_SIZE_MIN;
3264	return 1;
3265}
3266__setup("uhash_entries=", set_uhash_entries);
3267
3268void __init udp_table_init(struct udp_table *table, const char *name)
3269{
3270	unsigned int i;
3271
 
 
3272	table->hash = alloc_large_system_hash(name,
3273					      2 * sizeof(struct udp_hslot),
3274					      uhash_entries,
3275					      21, /* one slot per 2 MB */
3276					      0,
3277					      &table->log,
3278					      &table->mask,
3279					      UDP_HTABLE_SIZE_MIN,
3280					      UDP_HTABLE_SIZE_MAX);
3281
3282	table->hash2 = table->hash + (table->mask + 1);
3283	for (i = 0; i <= table->mask; i++) {
3284		INIT_HLIST_HEAD(&table->hash[i].head);
3285		table->hash[i].count = 0;
3286		spin_lock_init(&table->hash[i].lock);
3287	}
3288	for (i = 0; i <= table->mask; i++) {
3289		INIT_HLIST_HEAD(&table->hash2[i].head);
3290		table->hash2[i].count = 0;
3291		spin_lock_init(&table->hash2[i].lock);
3292	}
 
3293}
3294
3295u32 udp_flow_hashrnd(void)
3296{
3297	static u32 hashrnd __read_mostly;
3298
3299	net_get_random_once(&hashrnd, sizeof(hashrnd));
3300
3301	return hashrnd;
3302}
3303EXPORT_SYMBOL(udp_flow_hashrnd);
3304
3305static void __net_init udp_sysctl_init(struct net *net)
3306{
3307	net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3308	net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3309
3310#ifdef CONFIG_NET_L3_MASTER_DEV
3311	net->ipv4.sysctl_udp_l3mdev_accept = 0;
3312#endif
3313}
3314
3315static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3316{
3317	struct udp_table *udptable;
 
3318	int i;
3319
3320	udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3321	if (!udptable)
3322		goto out;
3323
3324	udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
 
 
3325				      GFP_KERNEL_ACCOUNT);
3326	if (!udptable->hash)
3327		goto free_table;
3328
3329	udptable->hash2 = udptable->hash + hash_entries;
3330	udptable->mask = hash_entries - 1;
3331	udptable->log = ilog2(hash_entries);
3332
3333	for (i = 0; i < hash_entries; i++) {
3334		INIT_HLIST_HEAD(&udptable->hash[i].head);
3335		udptable->hash[i].count = 0;
3336		spin_lock_init(&udptable->hash[i].lock);
3337
3338		INIT_HLIST_HEAD(&udptable->hash2[i].head);
3339		udptable->hash2[i].count = 0;
3340		spin_lock_init(&udptable->hash2[i].lock);
3341	}
 
3342
3343	return udptable;
3344
3345free_table:
3346	kfree(udptable);
3347out:
3348	return NULL;
3349}
3350
3351static void __net_exit udp_pernet_table_free(struct net *net)
3352{
3353	struct udp_table *udptable = net->ipv4.udp_table;
3354
3355	if (udptable == &udp_table)
3356		return;
3357
3358	kvfree(udptable->hash);
3359	kfree(udptable);
3360}
3361
3362static void __net_init udp_set_table(struct net *net)
3363{
3364	struct udp_table *udptable;
3365	unsigned int hash_entries;
3366	struct net *old_net;
3367
3368	if (net_eq(net, &init_net))
3369		goto fallback;
3370
3371	old_net = current->nsproxy->net_ns;
3372	hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3373	if (!hash_entries)
3374		goto fallback;
3375
3376	/* Set min to keep the bitmap on stack in udp_lib_get_port() */
3377	if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3378		hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3379	else
3380		hash_entries = roundup_pow_of_two(hash_entries);
3381
3382	udptable = udp_pernet_table_alloc(hash_entries);
3383	if (udptable) {
3384		net->ipv4.udp_table = udptable;
3385	} else {
3386		pr_warn("Failed to allocate UDP hash table (entries: %u) "
3387			"for a netns, fallback to the global one\n",
3388			hash_entries);
3389fallback:
3390		net->ipv4.udp_table = &udp_table;
3391	}
3392}
3393
3394static int __net_init udp_pernet_init(struct net *net)
3395{
3396	udp_sysctl_init(net);
3397	udp_set_table(net);
3398
3399	return 0;
3400}
3401
3402static void __net_exit udp_pernet_exit(struct net *net)
3403{
3404	udp_pernet_table_free(net);
3405}
3406
3407static struct pernet_operations __net_initdata udp_sysctl_ops = {
3408	.init	= udp_pernet_init,
3409	.exit	= udp_pernet_exit,
3410};
3411
3412#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3413DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3414		     struct udp_sock *udp_sk, uid_t uid, int bucket)
3415
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3416static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3417{
3418	struct udp_iter_state *st = priv_data;
3419	struct udp_seq_afinfo *afinfo;
3420	int ret;
3421
3422	afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3423	if (!afinfo)
3424		return -ENOMEM;
3425
3426	afinfo->family = AF_UNSPEC;
3427	afinfo->udp_table = NULL;
3428	st->bpf_seq_afinfo = afinfo;
3429	ret = bpf_iter_init_seq_net(priv_data, aux);
3430	if (ret)
3431		kfree(afinfo);
 
3432	return ret;
3433}
3434
3435static void bpf_iter_fini_udp(void *priv_data)
3436{
3437	struct udp_iter_state *st = priv_data;
3438
3439	kfree(st->bpf_seq_afinfo);
3440	bpf_iter_fini_seq_net(priv_data);
 
3441}
3442
3443static const struct bpf_iter_seq_info udp_seq_info = {
3444	.seq_ops		= &bpf_iter_udp_seq_ops,
3445	.init_seq_private	= bpf_iter_init_udp,
3446	.fini_seq_private	= bpf_iter_fini_udp,
3447	.seq_priv_size		= sizeof(struct udp_iter_state),
3448};
3449
3450static struct bpf_iter_reg udp_reg_info = {
3451	.target			= "udp",
3452	.ctx_arg_info_size	= 1,
3453	.ctx_arg_info		= {
3454		{ offsetof(struct bpf_iter__udp, udp_sk),
3455		  PTR_TO_BTF_ID_OR_NULL },
3456	},
3457	.seq_info		= &udp_seq_info,
3458};
3459
3460static void __init bpf_iter_register(void)
3461{
3462	udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3463	if (bpf_iter_reg_target(&udp_reg_info))
3464		pr_warn("Warning: could not register bpf iterator udp\n");
3465}
3466#endif
3467
3468void __init udp_init(void)
3469{
3470	unsigned long limit;
3471	unsigned int i;
3472
3473	udp_table_init(&udp_table, "UDP");
3474	limit = nr_free_buffer_pages() / 8;
3475	limit = max(limit, 128UL);
3476	sysctl_udp_mem[0] = limit / 4 * 3;
3477	sysctl_udp_mem[1] = limit;
3478	sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3479
3480	/* 16 spinlocks per cpu */
3481	udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3482	udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3483				GFP_KERNEL);
3484	if (!udp_busylocks)
3485		panic("UDP: failed to alloc udp_busylocks\n");
3486	for (i = 0; i < (1U << udp_busylocks_log); i++)
3487		spin_lock_init(udp_busylocks + i);
3488
3489	if (register_pernet_subsys(&udp_sysctl_ops))
3490		panic("UDP: failed to init sysctl parameters.\n");
3491
3492#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3493	bpf_iter_register();
3494#endif
3495}