Linux Audio

Check our new training course

Loading...
v4.6
 
   1/*
   2 *	UDP over IPv6
   3 *	Linux INET6 implementation
   4 *
   5 *	Authors:
   6 *	Pedro Roque		<roque@di.fc.ul.pt>
   7 *
   8 *	Based on linux/ipv4/udp.c
   9 *
  10 *	Fixes:
  11 *	Hideaki YOSHIFUJI	:	sin6_scope_id support
  12 *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
  13 *	Alexey Kuznetsov		allow both IPv4 and IPv6 sockets to bind
  14 *					a single port at the same time.
  15 *      Kazunori MIYAZAWA @USAGI:       change process style to use ip6_append_data
  16 *      YOSHIFUJI Hideaki @USAGI:	convert /proc/net/udp6 to seq_file.
  17 *
  18 *	This program is free software; you can redistribute it and/or
  19 *      modify it under the terms of the GNU General Public License
  20 *      as published by the Free Software Foundation; either version
  21 *      2 of the License, or (at your option) any later version.
  22 */
  23
 
  24#include <linux/errno.h>
  25#include <linux/types.h>
  26#include <linux/socket.h>
  27#include <linux/sockios.h>
  28#include <linux/net.h>
  29#include <linux/in6.h>
  30#include <linux/netdevice.h>
  31#include <linux/if_arp.h>
  32#include <linux/ipv6.h>
  33#include <linux/icmpv6.h>
  34#include <linux/init.h>
  35#include <linux/module.h>
  36#include <linux/skbuff.h>
  37#include <linux/slab.h>
  38#include <asm/uaccess.h>
 
 
  39
  40#include <net/addrconf.h>
  41#include <net/ndisc.h>
  42#include <net/protocol.h>
  43#include <net/transp_v6.h>
  44#include <net/ip6_route.h>
  45#include <net/raw.h>
 
  46#include <net/tcp_states.h>
  47#include <net/ip6_checksum.h>
 
  48#include <net/xfrm.h>
 
  49#include <net/inet6_hashtables.h>
  50#include <net/busy_poll.h>
  51#include <net/sock_reuseport.h>
 
  52
  53#include <linux/proc_fs.h>
  54#include <linux/seq_file.h>
  55#include <trace/events/skb.h>
  56#include "udp_impl.h"
  57
  58static u32 udp6_ehashfn(const struct net *net,
  59			const struct in6_addr *laddr,
  60			const u16 lport,
  61			const struct in6_addr *faddr,
  62			const __be16 fport)
  63{
  64	static u32 udp6_ehash_secret __read_mostly;
  65	static u32 udp_ipv6_hash_secret __read_mostly;
 
  66
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  67	u32 lhash, fhash;
  68
  69	net_get_random_once(&udp6_ehash_secret,
  70			    sizeof(udp6_ehash_secret));
  71	net_get_random_once(&udp_ipv6_hash_secret,
  72			    sizeof(udp_ipv6_hash_secret));
  73
  74	lhash = (__force u32)laddr->s6_addr32[3];
  75	fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
  76
  77	return __inet6_ehashfn(lhash, lport, fhash, fport,
  78			       udp_ipv6_hash_secret + net_hash_mix(net));
  79}
  80
  81static u32 udp6_portaddr_hash(const struct net *net,
  82			      const struct in6_addr *addr6,
  83			      unsigned int port)
  84{
  85	unsigned int hash, mix = net_hash_mix(net);
  86
  87	if (ipv6_addr_any(addr6))
  88		hash = jhash_1word(0, mix);
  89	else if (ipv6_addr_v4mapped(addr6))
  90		hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
  91	else
  92		hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
  93
  94	return hash ^ port;
  95}
  96
  97int udp_v6_get_port(struct sock *sk, unsigned short snum)
  98{
  99	unsigned int hash2_nulladdr =
 100		udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
 101	unsigned int hash2_partial =
 102		udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
 103
 104	/* precompute partial secondary hash */
 105	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
 106	return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
 107}
 108
 109static void udp_v6_rehash(struct sock *sk)
 110{
 111	u16 new_hash = udp6_portaddr_hash(sock_net(sk),
 112					  &sk->sk_v6_rcv_saddr,
 113					  inet_sk(sk)->inet_num);
 
 114
 115	udp_lib_rehash(sk, new_hash);
 116}
 117
 118static inline int compute_score(struct sock *sk, struct net *net,
 119				unsigned short hnum,
 120				const struct in6_addr *saddr, __be16 sport,
 121				const struct in6_addr *daddr, __be16 dport,
 122				int dif)
 123{
 124	int score;
 125	struct inet_sock *inet;
 126
 127	if (!net_eq(sock_net(sk), net) ||
 128	    udp_sk(sk)->udp_port_hash != hnum ||
 129	    sk->sk_family != PF_INET6)
 130		return -1;
 131
 132	score = 0;
 133	inet = inet_sk(sk);
 134
 135	if (inet->inet_dport) {
 136		if (inet->inet_dport != sport)
 137			return -1;
 138		score++;
 139	}
 140
 141	if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
 142		if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
 143			return -1;
 144		score++;
 145	}
 146
 147	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
 148		if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
 149			return -1;
 150		score++;
 151	}
 152
 153	if (sk->sk_bound_dev_if) {
 154		if (sk->sk_bound_dev_if != dif)
 155			return -1;
 156		score++;
 157	}
 158
 159	if (sk->sk_incoming_cpu == raw_smp_processor_id())
 160		score++;
 161
 162	return score;
 163}
 164
 165static inline int compute_score2(struct sock *sk, struct net *net,
 166				 const struct in6_addr *saddr, __be16 sport,
 167				 const struct in6_addr *daddr,
 168				 unsigned short hnum, int dif)
 169{
 170	int score;
 171	struct inet_sock *inet;
 
 172
 173	if (!net_eq(sock_net(sk), net) ||
 174	    udp_sk(sk)->udp_port_hash != hnum ||
 175	    sk->sk_family != PF_INET6)
 176		return -1;
 177
 178	if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
 179		return -1;
 180
 181	score = 0;
 182	inet = inet_sk(sk);
 183
 184	if (inet->inet_dport) {
 185		if (inet->inet_dport != sport)
 186			return -1;
 187		score++;
 188	}
 189
 190	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
 191		if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
 192			return -1;
 193		score++;
 194	}
 195
 196	if (sk->sk_bound_dev_if) {
 197		if (sk->sk_bound_dev_if != dif)
 198			return -1;
 
 
 199		score++;
 200	}
 201
 202	if (sk->sk_incoming_cpu == raw_smp_processor_id())
 203		score++;
 204
 205	return score;
 206}
 207
 208/* called with read_rcu_lock() */
 209static struct sock *udp6_lib_lookup2(struct net *net,
 210		const struct in6_addr *saddr, __be16 sport,
 211		const struct in6_addr *daddr, unsigned int hnum, int dif,
 212		struct udp_hslot *hslot2, unsigned int slot2,
 213		struct sk_buff *skb)
 214{
 215	struct sock *sk, *result;
 216	struct hlist_nulls_node *node;
 217	int score, badness, matches = 0, reuseport = 0;
 218	bool select_ok = true;
 219	u32 hash = 0;
 220
 221begin:
 222	result = NULL;
 223	badness = -1;
 224	udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
 225		score = compute_score2(sk, net, saddr, sport,
 226				      daddr, hnum, dif);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 227		if (score > badness) {
 228			result = sk;
 229			badness = score;
 230			reuseport = sk->sk_reuseport;
 231			if (reuseport) {
 232				hash = udp6_ehashfn(net, daddr, hnum,
 233						    saddr, sport);
 234				if (select_ok) {
 235					struct sock *sk2;
 236
 237					sk2 = reuseport_select_sock(sk, hash, skb,
 238							sizeof(struct udphdr));
 239					if (sk2) {
 240						result = sk2;
 241						select_ok = false;
 242						goto found;
 243					}
 244				}
 245				matches = 1;
 246			}
 247		} else if (score == badness && reuseport) {
 248			matches++;
 249			if (reciprocal_scale(hash, matches) == 0)
 250				result = sk;
 251			hash = next_pseudo_random32(hash);
 252		}
 253	}
 254	/*
 255	 * if the nulls value we got at the end of this lookup is
 256	 * not the expected one, we must restart lookup.
 257	 * We probably met an item that was moved to another chain.
 258	 */
 259	if (get_nulls_value(node) != slot2)
 260		goto begin;
 261
 262	if (result) {
 263found:
 264		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
 265			result = NULL;
 266		else if (unlikely(compute_score2(result, net, saddr, sport,
 267				  daddr, hnum, dif) < badness)) {
 268			sock_put(result);
 269			goto begin;
 270		}
 271	}
 272	return result;
 273}
 274
 275struct sock *__udp6_lib_lookup(struct net *net,
 276				      const struct in6_addr *saddr, __be16 sport,
 277				      const struct in6_addr *daddr, __be16 dport,
 278				      int dif, struct udp_table *udptable,
 279				      struct sk_buff *skb)
 
 280{
 281	struct sock *sk, *result;
 282	struct hlist_nulls_node *node;
 283	unsigned short hnum = ntohs(dport);
 284	unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
 285	struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
 286	int score, badness, matches = 0, reuseport = 0;
 287	bool select_ok = true;
 288	u32 hash = 0;
 289
 290	rcu_read_lock();
 291	if (hslot->count > 10) {
 292		hash2 = udp6_portaddr_hash(net, daddr, hnum);
 293		slot2 = hash2 & udptable->mask;
 294		hslot2 = &udptable->hash2[slot2];
 295		if (hslot->count < hslot2->count)
 296			goto begin;
 297
 298		result = udp6_lib_lookup2(net, saddr, sport,
 299					  daddr, hnum, dif,
 300					  hslot2, slot2, skb);
 301		if (!result) {
 302			hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
 303			slot2 = hash2 & udptable->mask;
 304			hslot2 = &udptable->hash2[slot2];
 305			if (hslot->count < hslot2->count)
 306				goto begin;
 307
 308			result = udp6_lib_lookup2(net, saddr, sport,
 309						  &in6addr_any, hnum, dif,
 310						  hslot2, slot2, skb);
 311		}
 312		rcu_read_unlock();
 313		return result;
 314	}
 315begin:
 316	result = NULL;
 317	badness = -1;
 318	sk_nulls_for_each_rcu(sk, node, &hslot->head) {
 319		score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
 
 
 
 320		if (score > badness) {
 321			result = sk;
 322			badness = score;
 323			reuseport = sk->sk_reuseport;
 324			if (reuseport) {
 325				hash = udp6_ehashfn(net, daddr, hnum,
 326						    saddr, sport);
 327				if (select_ok) {
 328					struct sock *sk2;
 329
 330					sk2 = reuseport_select_sock(sk, hash, skb,
 331							sizeof(struct udphdr));
 332					if (sk2) {
 333						result = sk2;
 334						select_ok = false;
 335						goto found;
 336					}
 337				}
 338				matches = 1;
 339			}
 340		} else if (score == badness && reuseport) {
 341			matches++;
 342			if (reciprocal_scale(hash, matches) == 0)
 
 343				result = sk;
 344			hash = next_pseudo_random32(hash);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 345		}
 346	}
 347	/*
 348	 * if the nulls value we got at the end of this lookup is
 349	 * not the expected one, we must restart lookup.
 350	 * We probably met an item that was moved to another chain.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 351	 */
 352	if (get_nulls_value(node) != slot)
 353		goto begin;
 354
 355	if (result) {
 356found:
 357		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
 358			result = NULL;
 359		else if (unlikely(compute_score(result, net, hnum, saddr, sport,
 360					daddr, dport, dif) < badness)) {
 361			sock_put(result);
 362			goto begin;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 363		}
 364	}
 365	rcu_read_unlock();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 366	return result;
 367}
 368EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
 369
 370static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
 371					  __be16 sport, __be16 dport,
 372					  struct udp_table *udptable)
 373{
 374	struct sock *sk;
 375	const struct ipv6hdr *iph = ipv6_hdr(skb);
 376
 377	sk = skb_steal_sock(skb);
 378	if (unlikely(sk))
 379		return sk;
 380	return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
 381				 &iph->daddr, dport, inet6_iif(skb),
 382				 udptable, skb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 383}
 384
 385struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
 
 
 
 
 386			     const struct in6_addr *daddr, __be16 dport, int dif)
 387{
 388	return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table, NULL);
 
 
 
 
 
 
 389}
 390EXPORT_SYMBOL_GPL(udp6_lib_lookup);
 
 
 
 
 
 
 
 
 
 
 391
 392/*
 393 *	This should be easy, if there is something there we
 394 *	return it, otherwise we block.
 395 */
 396
 397int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 398		  int noblock, int flags, int *addr_len)
 399{
 400	struct ipv6_pinfo *np = inet6_sk(sk);
 401	struct inet_sock *inet = inet_sk(sk);
 402	struct sk_buff *skb;
 403	unsigned int ulen, copied;
 404	int peeked, off = 0;
 405	int err;
 406	int is_udplite = IS_UDPLITE(sk);
 
 407	bool checksum_valid = false;
 408	int is_udp4;
 409	bool slow;
 410
 411	if (flags & MSG_ERRQUEUE)
 412		return ipv6_recv_error(sk, msg, len, addr_len);
 413
 414	if (np->rxpmtu && np->rxopt.bits.rxpmtu)
 415		return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
 416
 417try_again:
 418	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
 419				  &peeked, &off, &err);
 420	if (!skb)
 421		goto out;
 422
 423	ulen = skb->len - sizeof(struct udphdr);
 424	copied = len;
 425	if (copied > ulen)
 426		copied = ulen;
 427	else if (copied < ulen)
 428		msg->msg_flags |= MSG_TRUNC;
 429
 430	is_udp4 = (skb->protocol == htons(ETH_P_IP));
 
 431
 432	/*
 433	 * If checksum is needed at all, try to do it while copying the
 434	 * data.  If the data is truncated, or if we only want a partial
 435	 * coverage checksum (UDP-Lite), do it before the copy.
 436	 */
 437
 438	if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
 439		checksum_valid = !udp_lib_checksum_complete(skb);
 
 
 440		if (!checksum_valid)
 441			goto csum_copy_err;
 442	}
 443
 444	if (checksum_valid || skb_csum_unnecessary(skb))
 445		err = skb_copy_datagram_msg(skb, sizeof(struct udphdr),
 446					    msg, copied);
 447	else {
 448		err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), msg);
 
 
 449		if (err == -EINVAL)
 450			goto csum_copy_err;
 451	}
 452	if (unlikely(err)) {
 453		trace_kfree_skb(skb, udpv6_recvmsg);
 454		if (!peeked) {
 455			atomic_inc(&sk->sk_drops);
 456			if (is_udp4)
 457				UDP_INC_STATS_USER(sock_net(sk),
 458						   UDP_MIB_INERRORS,
 459						   is_udplite);
 460			else
 461				UDP6_INC_STATS_USER(sock_net(sk),
 462						    UDP_MIB_INERRORS,
 463						    is_udplite);
 464		}
 465		goto out_free;
 466	}
 467	if (!peeked) {
 468		if (is_udp4)
 469			UDP_INC_STATS_USER(sock_net(sk),
 470					UDP_MIB_INDATAGRAMS, is_udplite);
 471		else
 472			UDP6_INC_STATS_USER(sock_net(sk),
 473					UDP_MIB_INDATAGRAMS, is_udplite);
 474	}
 
 
 475
 476	sock_recv_ts_and_drops(msg, sk, skb);
 477
 478	/* Copy the address. */
 479	if (msg->msg_name) {
 480		DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
 481		sin6->sin6_family = AF_INET6;
 482		sin6->sin6_port = udp_hdr(skb)->source;
 483		sin6->sin6_flowinfo = 0;
 484
 485		if (is_udp4) {
 486			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
 487					       &sin6->sin6_addr);
 488			sin6->sin6_scope_id = 0;
 489		} else {
 490			sin6->sin6_addr = ipv6_hdr(skb)->saddr;
 491			sin6->sin6_scope_id =
 492				ipv6_iface_scope_id(&sin6->sin6_addr,
 493						    inet6_iif(skb));
 494		}
 495		*addr_len = sizeof(*sin6);
 
 
 
 
 496	}
 497
 
 
 
 498	if (np->rxopt.all)
 499		ip6_datagram_recv_common_ctl(sk, msg, skb);
 500
 501	if (is_udp4) {
 502		if (inet->cmsg_flags)
 503			ip_cmsg_recv(msg, skb);
 
 504	} else {
 505		if (np->rxopt.all)
 506			ip6_datagram_recv_specific_ctl(sk, msg, skb);
 507	}
 508
 509	err = copied;
 510	if (flags & MSG_TRUNC)
 511		err = ulen;
 512
 513out_free:
 514	skb_free_datagram_locked(sk, skb);
 515out:
 516	return err;
 517
 518csum_copy_err:
 519	slow = lock_sock_fast(sk);
 520	if (!skb_kill_datagram(sk, skb, flags)) {
 521		if (is_udp4) {
 522			UDP_INC_STATS_USER(sock_net(sk),
 523					UDP_MIB_CSUMERRORS, is_udplite);
 524			UDP_INC_STATS_USER(sock_net(sk),
 525					UDP_MIB_INERRORS, is_udplite);
 526		} else {
 527			UDP6_INC_STATS_USER(sock_net(sk),
 528					UDP_MIB_CSUMERRORS, is_udplite);
 529			UDP6_INC_STATS_USER(sock_net(sk),
 530					UDP_MIB_INERRORS, is_udplite);
 531		}
 532	}
 533	unlock_sock_fast(sk, slow);
 534
 535	/* starting over for a new packet, but check if we need to yield */
 536	cond_resched();
 537	msg->msg_flags &= ~MSG_TRUNC;
 538	goto try_again;
 539}
 540
 541void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 542		    u8 type, u8 code, int offset, __be32 info,
 543		    struct udp_table *udptable)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 544{
 545	struct ipv6_pinfo *np;
 546	const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
 547	const struct in6_addr *saddr = &hdr->saddr;
 548	const struct in6_addr *daddr = &hdr->daddr;
 549	struct udphdr *uh = (struct udphdr *)(skb->data+offset);
 
 550	struct sock *sk;
 551	int harderr;
 552	int err;
 553	struct net *net = dev_net(skb->dev);
 554
 555	sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
 556			       inet6_iif(skb), udptable, skb);
 557	if (!sk) {
 558		ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
 559				   ICMP6_MIB_INERRORS);
 560		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 561	}
 562
 563	harderr = icmpv6_err_convert(type, code, &err);
 564	np = inet6_sk(sk);
 565
 566	if (type == ICMPV6_PKT_TOOBIG) {
 567		if (!ip6_sk_accept_pmtu(sk))
 568			goto out;
 569		ip6_sk_update_pmtu(skb, sk, info);
 570		if (np->pmtudisc != IPV6_PMTUDISC_DONT)
 571			harderr = 1;
 572	}
 573	if (type == NDISC_REDIRECT) {
 574		ip6_sk_redirect(skb, sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 575		goto out;
 576	}
 577
 578	if (!np->recverr) {
 579		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
 580			goto out;
 581	} else {
 582		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
 583	}
 584
 585	sk->sk_err = err;
 586	sk->sk_error_report(sk);
 587out:
 588	sock_put(sk);
 589}
 590
 591static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 592{
 593	int rc;
 594
 595	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
 596		sock_rps_save_rxhash(sk, skb);
 597		sk_mark_napi_id(sk, skb);
 598		sk_incoming_cpu_update(sk);
 
 
 599	}
 600
 601	rc = sock_queue_rcv_skb(sk, skb);
 602	if (rc < 0) {
 603		int is_udplite = IS_UDPLITE(sk);
 
 604
 605		/* Note that an ENOMEM error is charged twice */
 606		if (rc == -ENOMEM)
 607			UDP6_INC_STATS_BH(sock_net(sk),
 608					UDP_MIB_RCVBUFERRORS, is_udplite);
 609		UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
 610		kfree_skb(skb);
 
 
 
 
 
 
 
 611		return -1;
 612	}
 613	return 0;
 614}
 615
 616static __inline__ void udpv6_err(struct sk_buff *skb,
 617				 struct inet6_skb_parm *opt, u8 type,
 618				 u8 code, int offset, __be32 info)
 619{
 620	__udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
 621}
 622
 623static struct static_key udpv6_encap_needed __read_mostly;
 624void udpv6_encap_enable(void)
 
 625{
 626	if (!static_key_enabled(&udpv6_encap_needed))
 627		static_key_slow_inc(&udpv6_encap_needed);
 628}
 629EXPORT_SYMBOL(udpv6_encap_enable);
 630
 631int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 632{
 
 633	struct udp_sock *up = udp_sk(sk);
 634	int rc;
 635	int is_udplite = IS_UDPLITE(sk);
 636
 637	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
 
 638		goto drop;
 
 
 639
 640	if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
 
 641		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
 642
 643		/*
 644		 * This is an encapsulation socket so pass the skb to
 645		 * the socket's udp_encap_rcv() hook. Otherwise, just
 646		 * fall through and pass this up the UDP socket.
 647		 * up->encap_rcv() returns the following value:
 648		 * =0 if skb was successfully passed to the encap
 649		 *    handler or was discarded by it.
 650		 * >0 if skb should be passed on to UDP.
 651		 * <0 if skb should be resubmitted as proto -N
 652		 */
 653
 654		/* if we're overly short, let UDP handle it */
 655		encap_rcv = ACCESS_ONCE(up->encap_rcv);
 656		if (skb->len > sizeof(struct udphdr) && encap_rcv) {
 657			int ret;
 658
 659			/* Verify checksum before giving to encap */
 660			if (udp_lib_checksum_complete(skb))
 661				goto csum_error;
 662
 663			ret = encap_rcv(sk, skb);
 664			if (ret <= 0) {
 665				UDP_INC_STATS_BH(sock_net(sk),
 666						 UDP_MIB_INDATAGRAMS,
 667						 is_udplite);
 668				return -ret;
 669			}
 670		}
 671
 672		/* FALLTHROUGH -- it's a UDP Packet */
 673	}
 674
 675	/*
 676	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
 677	 */
 678	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
 
 679
 680		if (up->pcrlen == 0) {          /* full coverage was set  */
 681			net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
 682					    UDP_SKB_CB(skb)->cscov, skb->len);
 683			goto drop;
 684		}
 685		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
 686			net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
 687					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
 688			goto drop;
 689		}
 690	}
 691
 692	if (rcu_access_pointer(sk->sk_filter)) {
 693		if (udp_lib_checksum_complete(skb))
 694			goto csum_error;
 695	}
 696
 697	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
 698		UDP6_INC_STATS_BH(sock_net(sk),
 699				  UDP_MIB_RCVBUFERRORS, is_udplite);
 700		goto drop;
 701	}
 702
 703	skb_dst_drop(skb);
 704
 705	bh_lock_sock(sk);
 706	rc = 0;
 707	if (!sock_owned_by_user(sk))
 708		rc = __udpv6_queue_rcv_skb(sk, skb);
 709	else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
 710		bh_unlock_sock(sk);
 711		goto drop;
 712	}
 713	bh_unlock_sock(sk);
 714
 715	return rc;
 716
 717csum_error:
 718	UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
 
 719drop:
 720	UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
 721	atomic_inc(&sk->sk_drops);
 722	kfree_skb(skb);
 723	return -1;
 724}
 725
 726static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 727				   __be16 loc_port, const struct in6_addr *loc_addr,
 728				   __be16 rmt_port, const struct in6_addr *rmt_addr,
 729				   int dif, unsigned short hnum)
 730{
 731	struct inet_sock *inet = inet_sk(sk);
 732
 733	if (!net_eq(sock_net(sk), net))
 734		return false;
 735
 736	if (udp_sk(sk)->udp_port_hash != hnum ||
 737	    sk->sk_family != PF_INET6 ||
 738	    (inet->inet_dport && inet->inet_dport != rmt_port) ||
 739	    (!ipv6_addr_any(&sk->sk_v6_daddr) &&
 740		    !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
 741	    (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
 742	    (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
 743		    !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
 744		return false;
 745	if (!inet6_mc_check(sk, loc_addr, rmt_addr))
 746		return false;
 747	return true;
 748}
 749
 750static void flush_stack(struct sock **stack, unsigned int count,
 751			struct sk_buff *skb, unsigned int final)
 752{
 753	struct sk_buff *skb1 = NULL;
 754	struct sock *sk;
 755	unsigned int i;
 756
 757	for (i = 0; i < count; i++) {
 758		sk = stack[i];
 759		if (likely(!skb1))
 760			skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
 761		if (!skb1) {
 762			atomic_inc(&sk->sk_drops);
 763			UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
 764					  IS_UDPLITE(sk));
 765			UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
 766					  IS_UDPLITE(sk));
 767		}
 768
 769		if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
 770			skb1 = NULL;
 771		sock_put(sk);
 772	}
 773	if (unlikely(skb1))
 774		kfree_skb(skb1);
 775}
 776
 777static void udp6_csum_zero_error(struct sk_buff *skb)
 778{
 779	/* RFC 2460 section 8.1 says that we SHOULD log
 780	 * this error. Well, it is reasonable.
 781	 */
 782	net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
 783			    &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
 784			    &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
 785}
 786
 787/*
 788 * Note: called only from the BH handler context,
 789 * so we don't need to lock the hashes.
 790 */
 791static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
 792		const struct in6_addr *saddr, const struct in6_addr *daddr,
 793		struct udp_table *udptable, int proto)
 794{
 795	struct sock *sk, *stack[256 / sizeof(struct sock *)];
 796	const struct udphdr *uh = udp_hdr(skb);
 797	struct hlist_nulls_node *node;
 798	unsigned short hnum = ntohs(uh->dest);
 799	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
 800	int dif = inet6_iif(skb);
 801	unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
 802	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
 803	bool inner_flushed = false;
 
 
 
 804
 805	if (use_hash2) {
 806		hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
 807			    udp_table.mask;
 808		hash2 = udp6_portaddr_hash(net, daddr, hnum) & udp_table.mask;
 809start_lookup:
 810		hslot = &udp_table.hash2[hash2];
 811		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
 812	}
 813
 814	spin_lock(&hslot->lock);
 815	sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
 816		if (__udp_v6_is_mcast_sock(net, sk,
 817					   uh->dest, daddr,
 818					   uh->source, saddr,
 819					   dif, hnum) &&
 820		    /* If zero checksum and no_check is not on for
 821		     * the socket then skip it.
 822		     */
 823		    (uh->check || udp_sk(sk)->no_check6_rx)) {
 824			if (unlikely(count == ARRAY_SIZE(stack))) {
 825				flush_stack(stack, count, skb, ~0);
 826				inner_flushed = true;
 827				count = 0;
 828			}
 829			stack[count++] = sk;
 830			sock_hold(sk);
 
 
 
 
 
 831		}
 832	}
 833
 834	spin_unlock(&hslot->lock);
 
 
 835
 836	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
 837	if (use_hash2 && hash2 != hash2_any) {
 838		hash2 = hash2_any;
 839		goto start_lookup;
 840	}
 841
 842	if (count) {
 843		flush_stack(stack, count, skb, count - 1);
 
 844	} else {
 845		if (!inner_flushed)
 846			UDP6_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
 847					  proto == IPPROTO_UDPLITE);
 848		consume_skb(skb);
 849	}
 850	return 0;
 851}
 852
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 853int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
 854		   int proto)
 855{
 
 
 856	struct net *net = dev_net(skb->dev);
 857	struct sock *sk;
 858	struct udphdr *uh;
 859	const struct in6_addr *saddr, *daddr;
 860	u32 ulen = 0;
 861
 862	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
 863		goto discard;
 864
 865	saddr = &ipv6_hdr(skb)->saddr;
 866	daddr = &ipv6_hdr(skb)->daddr;
 867	uh = udp_hdr(skb);
 868
 869	ulen = ntohs(uh->len);
 870	if (ulen > skb->len)
 871		goto short_packet;
 872
 873	if (proto == IPPROTO_UDP) {
 874		/* UDP validates ulen. */
 875
 876		/* Check for jumbo payload */
 877		if (ulen == 0)
 878			ulen = skb->len;
 879
 880		if (ulen < sizeof(*uh))
 881			goto short_packet;
 882
 883		if (ulen < skb->len) {
 884			if (pskb_trim_rcsum(skb, ulen))
 885				goto short_packet;
 886			saddr = &ipv6_hdr(skb)->saddr;
 887			daddr = &ipv6_hdr(skb)->daddr;
 888			uh = udp_hdr(skb);
 889		}
 890	}
 891
 892	if (udp6_csum_init(skb, uh, proto))
 893		goto csum_error;
 894
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 895	/*
 896	 *	Multicast receive code
 897	 */
 898	if (ipv6_addr_is_multicast(daddr))
 899		return __udp6_lib_mcast_deliver(net, skb,
 900				saddr, daddr, udptable, proto);
 901
 902	/* Unicast */
 903
 904	/*
 905	 * check socket cache ... must talk to Alan about his plans
 906	 * for sock caches... i'll skip this for now.
 907	 */
 908	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
 909	if (sk) {
 910		int ret;
 911
 912		if (!uh->check && !udp_sk(sk)->no_check6_rx) {
 913			sock_put(sk);
 914			udp6_csum_zero_error(skb);
 915			goto csum_error;
 916		}
 917
 918		if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
 919			skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
 920						 ip6_compute_pseudo);
 921
 922		ret = udpv6_queue_rcv_skb(sk, skb);
 923		sock_put(sk);
 924
 925		/* a return value > 0 means to resubmit the input */
 926		if (ret > 0)
 927			return ret;
 928
 929		return 0;
 930	}
 
 
 931
 932	if (!uh->check) {
 933		udp6_csum_zero_error(skb);
 934		goto csum_error;
 935	}
 936
 937	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 938		goto discard;
 
 939
 940	if (udp_lib_checksum_complete(skb))
 941		goto csum_error;
 942
 943	UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
 944	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 945
 946	kfree_skb(skb);
 947	return 0;
 948
 949short_packet:
 
 
 950	net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
 951			    proto == IPPROTO_UDPLITE ? "-Lite" : "",
 952			    saddr, ntohs(uh->source),
 953			    ulen, skb->len,
 954			    daddr, ntohs(uh->dest));
 955	goto discard;
 
 
 
 956csum_error:
 957	UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
 
 
 958discard:
 959	UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
 960	kfree_skb(skb);
 961	return 0;
 962}
 963
 964static __inline__ int udpv6_rcv(struct sk_buff *skb)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 965{
 966	return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 967}
 968
 969/*
 970 * Throw away all pending data and cancel the corking. Socket is locked.
 971 */
 972static void udp_v6_flush_pending_frames(struct sock *sk)
 973{
 974	struct udp_sock *up = udp_sk(sk);
 975
 976	if (up->pending == AF_INET)
 977		udp_flush_pending_frames(sk);
 978	else if (up->pending) {
 979		up->len = 0;
 980		up->pending = 0;
 981		ip6_flush_pending_frames(sk);
 982	}
 983}
 984
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 985/**
 986 *	udp6_hwcsum_outgoing  -  handle outgoing HW checksumming
 987 *	@sk:	socket we are sending on
 988 *	@skb:	sk_buff containing the filled-in UDP header
 989 *		(checksum field must be zeroed out)
 
 
 
 990 */
 991static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
 992				 const struct in6_addr *saddr,
 993				 const struct in6_addr *daddr, int len)
 994{
 995	unsigned int offset;
 996	struct udphdr *uh = udp_hdr(skb);
 997	struct sk_buff *frags = skb_shinfo(skb)->frag_list;
 998	__wsum csum = 0;
 999
1000	if (!frags) {
1001		/* Only one fragment on the socket.  */
1002		skb->csum_start = skb_transport_header(skb) - skb->head;
1003		skb->csum_offset = offsetof(struct udphdr, check);
1004		uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1005	} else {
1006		/*
1007		 * HW-checksum won't work as there are two or more
1008		 * fragments on the socket so that all csums of sk_buffs
1009		 * should be together
1010		 */
1011		offset = skb_transport_offset(skb);
1012		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
 
1013
1014		skb->ip_summed = CHECKSUM_NONE;
1015
1016		do {
1017			csum = csum_add(csum, frags->csum);
1018		} while ((frags = frags->next));
1019
1020		uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1021					    csum);
1022		if (uh->check == 0)
1023			uh->check = CSUM_MANGLED_0;
1024	}
1025}
1026
1027/*
1028 *	Sending
1029 */
1030
1031static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
 
1032{
1033	struct sock *sk = skb->sk;
1034	struct udphdr *uh;
1035	int err = 0;
1036	int is_udplite = IS_UDPLITE(sk);
1037	__wsum csum = 0;
1038	int offset = skb_transport_offset(skb);
1039	int len = skb->len - offset;
 
1040
1041	/*
1042	 * Create a UDP header
1043	 */
1044	uh = udp_hdr(skb);
1045	uh->source = fl6->fl6_sport;
1046	uh->dest = fl6->fl6_dport;
1047	uh->len = htons(len);
1048	uh->check = 0;
1049
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1050	if (is_udplite)
1051		csum = udplite_csum(skb);
1052	else if (udp_sk(sk)->no_check6_tx) {   /* UDP csum disabled */
1053		skb->ip_summed = CHECKSUM_NONE;
1054		goto send;
1055	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
 
1056		udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1057		goto send;
1058	} else
1059		csum = udp_csum(skb);
1060
1061	/* add protocol-dependent pseudo-header */
1062	uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1063				    len, fl6->flowi6_proto, csum);
1064	if (uh->check == 0)
1065		uh->check = CSUM_MANGLED_0;
1066
1067send:
1068	err = ip6_send_skb(skb);
1069	if (err) {
1070		if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1071			UDP6_INC_STATS_USER(sock_net(sk),
1072					    UDP_MIB_SNDBUFERRORS, is_udplite);
1073			err = 0;
1074		}
1075	} else
1076		UDP6_INC_STATS_USER(sock_net(sk),
1077				    UDP_MIB_OUTDATAGRAMS, is_udplite);
 
1078	return err;
1079}
1080
1081static int udp_v6_push_pending_frames(struct sock *sk)
1082{
1083	struct sk_buff *skb;
1084	struct udp_sock  *up = udp_sk(sk);
1085	struct flowi6 fl6;
1086	int err = 0;
1087
1088	if (up->pending == AF_INET)
1089		return udp_push_pending_frames(sk);
1090
1091	/* ip6_finish_skb will release the cork, so make a copy of
1092	 * fl6 here.
1093	 */
1094	fl6 = inet_sk(sk)->cork.fl.u.ip6;
1095
1096	skb = ip6_finish_skb(sk);
1097	if (!skb)
1098		goto out;
1099
1100	err = udp_v6_send_skb(skb, &fl6);
1101
1102out:
1103	up->len = 0;
1104	up->pending = 0;
1105	return err;
1106}
1107
1108int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1109{
1110	struct ipv6_txoptions opt_space;
1111	struct udp_sock *up = udp_sk(sk);
1112	struct inet_sock *inet = inet_sk(sk);
1113	struct ipv6_pinfo *np = inet6_sk(sk);
1114	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1115	struct in6_addr *daddr, *final_p, final;
1116	struct ipv6_txoptions *opt = NULL;
1117	struct ipv6_txoptions *opt_to_free = NULL;
1118	struct ip6_flowlabel *flowlabel = NULL;
1119	struct flowi6 fl6;
 
1120	struct dst_entry *dst;
 
1121	int addr_len = msg->msg_namelen;
 
1122	int ulen = len;
1123	int hlimit = -1;
1124	int tclass = -1;
1125	int dontfrag = -1;
1126	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1127	int err;
1128	int connected = 0;
1129	int is_udplite = IS_UDPLITE(sk);
1130	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1131
 
 
 
 
 
1132	/* destination address check */
1133	if (sin6) {
1134		if (addr_len < offsetof(struct sockaddr, sa_data))
1135			return -EINVAL;
1136
1137		switch (sin6->sin6_family) {
1138		case AF_INET6:
1139			if (addr_len < SIN6_LEN_RFC2133)
1140				return -EINVAL;
1141			daddr = &sin6->sin6_addr;
 
 
 
 
1142			break;
1143		case AF_INET:
1144			goto do_udp_sendmsg;
1145		case AF_UNSPEC:
1146			msg->msg_name = sin6 = NULL;
1147			msg->msg_namelen = addr_len = 0;
1148			daddr = NULL;
1149			break;
1150		default:
1151			return -EINVAL;
1152		}
1153	} else if (!up->pending) {
1154		if (sk->sk_state != TCP_ESTABLISHED)
1155			return -EDESTADDRREQ;
1156		daddr = &sk->sk_v6_daddr;
1157	} else
1158		daddr = NULL;
1159
1160	if (daddr) {
1161		if (ipv6_addr_v4mapped(daddr)) {
1162			struct sockaddr_in sin;
1163			sin.sin_family = AF_INET;
1164			sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1165			sin.sin_addr.s_addr = daddr->s6_addr32[3];
1166			msg->msg_name = &sin;
1167			msg->msg_namelen = sizeof(sin);
1168do_udp_sendmsg:
1169			if (__ipv6_only_sock(sk))
1170				return -ENETUNREACH;
1171			return udp_sendmsg(sk, msg, len);
 
 
1172		}
1173	}
1174
1175	if (up->pending == AF_INET)
1176		return udp_sendmsg(sk, msg, len);
1177
1178	/* Rough check on arithmetic overflow,
1179	   better check is made in ip6_append_data().
1180	   */
1181	if (len > INT_MAX - sizeof(struct udphdr))
1182		return -EMSGSIZE;
1183
1184	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
1185	if (up->pending) {
 
 
1186		/*
1187		 * There are pending frames.
1188		 * The socket lock must be held while it's corked.
1189		 */
1190		lock_sock(sk);
1191		if (likely(up->pending)) {
1192			if (unlikely(up->pending != AF_INET6)) {
1193				release_sock(sk);
1194				return -EAFNOSUPPORT;
1195			}
1196			dst = NULL;
1197			goto do_append_data;
1198		}
1199		release_sock(sk);
1200	}
1201	ulen += sizeof(struct udphdr);
1202
1203	memset(&fl6, 0, sizeof(fl6));
1204
1205	if (sin6) {
1206		if (sin6->sin6_port == 0)
1207			return -EINVAL;
1208
1209		fl6.fl6_dport = sin6->sin6_port;
1210		daddr = &sin6->sin6_addr;
1211
1212		if (np->sndflow) {
1213			fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1214			if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1215				flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1216				if (!flowlabel)
1217					return -EINVAL;
1218			}
1219		}
1220
1221		/*
1222		 * Otherwise it will be difficult to maintain
1223		 * sk->sk_dst_cache.
1224		 */
1225		if (sk->sk_state == TCP_ESTABLISHED &&
1226		    ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1227			daddr = &sk->sk_v6_daddr;
1228
1229		if (addr_len >= sizeof(struct sockaddr_in6) &&
1230		    sin6->sin6_scope_id &&
1231		    __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1232			fl6.flowi6_oif = sin6->sin6_scope_id;
1233	} else {
1234		if (sk->sk_state != TCP_ESTABLISHED)
1235			return -EDESTADDRREQ;
1236
1237		fl6.fl6_dport = inet->inet_dport;
1238		daddr = &sk->sk_v6_daddr;
1239		fl6.flowlabel = np->flow_label;
1240		connected = 1;
1241	}
1242
1243	if (!fl6.flowi6_oif)
1244		fl6.flowi6_oif = sk->sk_bound_dev_if;
1245
1246	if (!fl6.flowi6_oif)
1247		fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1248
1249	fl6.flowi6_mark = sk->sk_mark;
1250
1251	if (msg->msg_controllen) {
1252		opt = &opt_space;
1253		memset(opt, 0, sizeof(struct ipv6_txoptions));
1254		opt->tot_len = sizeof(*opt);
 
1255
1256		err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1257					    &hlimit, &tclass, &dontfrag);
 
 
 
 
1258		if (err < 0) {
1259			fl6_sock_release(flowlabel);
1260			return err;
1261		}
1262		if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1263			flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1264			if (!flowlabel)
1265				return -EINVAL;
1266		}
1267		if (!(opt->opt_nflen|opt->opt_flen))
1268			opt = NULL;
1269		connected = 0;
1270	}
1271	if (!opt) {
1272		opt = txopt_get(np);
1273		opt_to_free = opt;
1274	}
1275	if (flowlabel)
1276		opt = fl6_merge_options(&opt_space, flowlabel, opt);
1277	opt = ipv6_fixup_options(&opt_space, opt);
 
1278
1279	fl6.flowi6_proto = sk->sk_protocol;
1280	if (!ipv6_addr_any(daddr))
1281		fl6.daddr = *daddr;
1282	else
1283		fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1284	if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1285		fl6.saddr = np->saddr;
1286	fl6.fl6_sport = inet->inet_sport;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1287
1288	final_p = fl6_update_dst(&fl6, opt, &final);
1289	if (final_p)
1290		connected = 0;
1291
1292	if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1293		fl6.flowi6_oif = np->mcast_oif;
1294		connected = 0;
1295	} else if (!fl6.flowi6_oif)
1296		fl6.flowi6_oif = np->ucast_oif;
1297
1298	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1299
1300	dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
 
 
 
 
 
1301	if (IS_ERR(dst)) {
1302		err = PTR_ERR(dst);
1303		dst = NULL;
1304		goto out;
1305	}
1306
1307	if (hlimit < 0)
1308		hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1309
1310	if (tclass < 0)
1311		tclass = np->tclass;
1312
1313	if (msg->msg_flags&MSG_CONFIRM)
1314		goto do_confirm;
1315back_from_confirm:
1316
1317	/* Lockless fast path for the non-corking case */
1318	if (!corkreq) {
1319		struct sk_buff *skb;
1320
1321		skb = ip6_make_skb(sk, getfrag, msg, ulen,
1322				   sizeof(struct udphdr), hlimit, tclass, opt,
1323				   &fl6, (struct rt6_info *)dst,
1324				   msg->msg_flags, dontfrag);
1325		err = PTR_ERR(skb);
1326		if (!IS_ERR_OR_NULL(skb))
1327			err = udp_v6_send_skb(skb, &fl6);
1328		goto release_dst;
 
1329	}
1330
1331	lock_sock(sk);
1332	if (unlikely(up->pending)) {
1333		/* The socket is already corked while preparing it. */
1334		/* ... which is an evident application bug. --ANK */
1335		release_sock(sk);
1336
1337		net_dbg_ratelimited("udp cork app bug 2\n");
1338		err = -EINVAL;
1339		goto out;
1340	}
1341
1342	up->pending = AF_INET6;
1343
1344do_append_data:
1345	if (dontfrag < 0)
1346		dontfrag = np->dontfrag;
1347	up->len += ulen;
1348	err = ip6_append_data(sk, getfrag, msg, ulen,
1349		sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1350		(struct rt6_info *)dst,
1351		corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1352	if (err)
1353		udp_v6_flush_pending_frames(sk);
1354	else if (!corkreq)
1355		err = udp_v6_push_pending_frames(sk);
1356	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1357		up->pending = 0;
1358
1359	if (err > 0)
1360		err = np->recverr ? net_xmit_errno(err) : 0;
1361	release_sock(sk);
1362
1363release_dst:
1364	if (dst) {
1365		if (connected) {
1366			ip6_dst_store(sk, dst,
1367				      ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1368				      &sk->sk_v6_daddr : NULL,
1369#ifdef CONFIG_IPV6_SUBTREES
1370				      ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1371				      &np->saddr :
1372#endif
1373				      NULL);
1374		} else {
1375			dst_release(dst);
1376		}
1377		dst = NULL;
1378	}
1379
1380out:
1381	dst_release(dst);
 
1382	fl6_sock_release(flowlabel);
1383	txopt_put(opt_to_free);
1384	if (!err)
1385		return len;
1386	/*
1387	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1388	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1389	 * we don't have a good statistic (IpOutDiscards but it can be too many
1390	 * things).  We could add another new stat but at least for now that
1391	 * seems like overkill.
1392	 */
1393	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1394		UDP6_INC_STATS_USER(sock_net(sk),
1395				UDP_MIB_SNDBUFERRORS, is_udplite);
1396	}
1397	return err;
1398
1399do_confirm:
1400	dst_confirm(dst);
 
1401	if (!(msg->msg_flags&MSG_PROBE) || len)
1402		goto back_from_confirm;
1403	err = 0;
1404	goto out;
1405}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1406
1407void udpv6_destroy_sock(struct sock *sk)
1408{
1409	struct udp_sock *up = udp_sk(sk);
1410	lock_sock(sk);
 
 
 
1411	udp_v6_flush_pending_frames(sk);
1412	release_sock(sk);
1413
1414	if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1415		void (*encap_destroy)(struct sock *sk);
1416		encap_destroy = ACCESS_ONCE(up->encap_destroy);
1417		if (encap_destroy)
1418			encap_destroy(sk);
 
 
 
 
 
 
1419	}
1420
1421	inet6_destroy_sock(sk);
1422}
1423
1424/*
1425 *	Socket option code for UDP
1426 */
1427int udpv6_setsockopt(struct sock *sk, int level, int optname,
1428		     char __user *optval, unsigned int optlen)
1429{
1430	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1431		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
 
1432					  udp_v6_push_pending_frames);
1433	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1434}
1435
1436#ifdef CONFIG_COMPAT
1437int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1438			    char __user *optval, unsigned int optlen)
1439{
1440	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1441		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1442					  udp_v6_push_pending_frames);
1443	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1444}
1445#endif
1446
1447int udpv6_getsockopt(struct sock *sk, int level, int optname,
1448		     char __user *optval, int __user *optlen)
1449{
1450	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1451		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1452	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1453}
1454
1455#ifdef CONFIG_COMPAT
1456int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1457			    char __user *optval, int __user *optlen)
1458{
1459	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1460		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1461	return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1462}
1463#endif
1464
1465static const struct inet6_protocol udpv6_protocol = {
1466	.handler	=	udpv6_rcv,
1467	.err_handler	=	udpv6_err,
1468	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1469};
1470
1471/* ------------------------------------------------------------------------ */
1472#ifdef CONFIG_PROC_FS
1473int udp6_seq_show(struct seq_file *seq, void *v)
1474{
1475	if (v == SEQ_START_TOKEN) {
1476		seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1477	} else {
1478		int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1479		struct inet_sock *inet = inet_sk(v);
1480		__u16 srcp = ntohs(inet->inet_sport);
1481		__u16 destp = ntohs(inet->inet_dport);
1482		ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
 
1483	}
1484	return 0;
1485}
1486
1487static const struct file_operations udp6_afinfo_seq_fops = {
1488	.owner    = THIS_MODULE,
1489	.open     = udp_seq_open,
1490	.read     = seq_read,
1491	.llseek   = seq_lseek,
1492	.release  = seq_release_net
1493};
 
1494
1495static struct udp_seq_afinfo udp6_seq_afinfo = {
1496	.name		= "udp6",
1497	.family		= AF_INET6,
1498	.udp_table	= &udp_table,
1499	.seq_fops	= &udp6_afinfo_seq_fops,
1500	.seq_ops	= {
1501		.show		= udp6_seq_show,
1502	},
1503};
1504
1505int __net_init udp6_proc_init(struct net *net)
1506{
1507	return udp_proc_register(net, &udp6_seq_afinfo);
 
 
 
1508}
1509
1510void udp6_proc_exit(struct net *net)
1511{
1512	udp_proc_unregister(net, &udp6_seq_afinfo);
1513}
1514#endif /* CONFIG_PROC_FS */
1515
1516void udp_v6_clear_sk(struct sock *sk, int size)
1517{
1518	struct inet_sock *inet = inet_sk(sk);
1519
1520	/* we do not want to clear pinet6 field, because of RCU lookups */
1521	sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1522
1523	size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1524	memset(&inet->pinet6 + 1, 0, size);
1525}
1526
1527/* ------------------------------------------------------------------------ */
1528
1529struct proto udpv6_prot = {
1530	.name		   = "UDPv6",
1531	.owner		   = THIS_MODULE,
1532	.close		   = udp_lib_close,
1533	.connect	   = ip6_datagram_connect,
1534	.disconnect	   = udp_disconnect,
1535	.ioctl		   = udp_ioctl,
1536	.destroy	   = udpv6_destroy_sock,
1537	.setsockopt	   = udpv6_setsockopt,
1538	.getsockopt	   = udpv6_getsockopt,
1539	.sendmsg	   = udpv6_sendmsg,
1540	.recvmsg	   = udpv6_recvmsg,
1541	.backlog_rcv	   = __udpv6_queue_rcv_skb,
1542	.release_cb	   = ip6_datagram_release_cb,
1543	.hash		   = udp_lib_hash,
1544	.unhash		   = udp_lib_unhash,
1545	.rehash		   = udp_v6_rehash,
1546	.get_port	   = udp_v6_get_port,
1547	.memory_allocated  = &udp_memory_allocated,
1548	.sysctl_mem	   = sysctl_udp_mem,
1549	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1550	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1551	.obj_size	   = sizeof(struct udp6_sock),
1552	.slab_flags	   = SLAB_DESTROY_BY_RCU,
1553	.h.udp_table	   = &udp_table,
1554#ifdef CONFIG_COMPAT
1555	.compat_setsockopt = compat_udpv6_setsockopt,
1556	.compat_getsockopt = compat_udpv6_getsockopt,
1557#endif
1558	.clear_sk	   = udp_v6_clear_sk,
 
 
 
 
 
 
 
 
 
 
1559};
1560
1561static struct inet_protosw udpv6_protosw = {
1562	.type =      SOCK_DGRAM,
1563	.protocol =  IPPROTO_UDP,
1564	.prot =      &udpv6_prot,
1565	.ops =       &inet6_dgram_ops,
1566	.flags =     INET_PROTOSW_PERMANENT,
1567};
1568
1569int __init udpv6_init(void)
1570{
1571	int ret;
1572
1573	ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
 
 
 
 
 
1574	if (ret)
1575		goto out;
1576
1577	ret = inet6_register_protosw(&udpv6_protosw);
1578	if (ret)
1579		goto out_udpv6_protocol;
1580out:
1581	return ret;
1582
1583out_udpv6_protocol:
1584	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1585	goto out;
1586}
1587
1588void udpv6_exit(void)
1589{
1590	inet6_unregister_protosw(&udpv6_protosw);
1591	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1592}
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 *	UDP over IPv6
   4 *	Linux INET6 implementation
   5 *
   6 *	Authors:
   7 *	Pedro Roque		<roque@di.fc.ul.pt>
   8 *
   9 *	Based on linux/ipv4/udp.c
  10 *
  11 *	Fixes:
  12 *	Hideaki YOSHIFUJI	:	sin6_scope_id support
  13 *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
  14 *	Alexey Kuznetsov		allow both IPv4 and IPv6 sockets to bind
  15 *					a single port at the same time.
  16 *      Kazunori MIYAZAWA @USAGI:       change process style to use ip6_append_data
  17 *      YOSHIFUJI Hideaki @USAGI:	convert /proc/net/udp6 to seq_file.
 
 
 
 
 
  18 */
  19
  20#include <linux/bpf-cgroup.h>
  21#include <linux/errno.h>
  22#include <linux/types.h>
  23#include <linux/socket.h>
  24#include <linux/sockios.h>
  25#include <linux/net.h>
  26#include <linux/in6.h>
  27#include <linux/netdevice.h>
  28#include <linux/if_arp.h>
  29#include <linux/ipv6.h>
  30#include <linux/icmpv6.h>
  31#include <linux/init.h>
  32#include <linux/module.h>
  33#include <linux/skbuff.h>
  34#include <linux/slab.h>
  35#include <linux/uaccess.h>
  36#include <linux/indirect_call_wrapper.h>
  37#include <trace/events/udp.h>
  38
  39#include <net/addrconf.h>
  40#include <net/ndisc.h>
  41#include <net/protocol.h>
  42#include <net/transp_v6.h>
  43#include <net/ip6_route.h>
  44#include <net/raw.h>
  45#include <net/seg6.h>
  46#include <net/tcp_states.h>
  47#include <net/ip6_checksum.h>
  48#include <net/ip6_tunnel.h>
  49#include <net/xfrm.h>
  50#include <net/inet_hashtables.h>
  51#include <net/inet6_hashtables.h>
  52#include <net/busy_poll.h>
  53#include <net/sock_reuseport.h>
  54#include <net/gro.h>
  55
  56#include <linux/proc_fs.h>
  57#include <linux/seq_file.h>
  58#include <trace/events/skb.h>
  59#include "udp_impl.h"
  60
  61static void udpv6_destruct_sock(struct sock *sk)
 
 
 
 
  62{
  63	udp_destruct_common(sk);
  64	inet6_sock_destruct(sk);
  65}
  66
  67int udpv6_init_sock(struct sock *sk)
  68{
  69	udp_lib_init_sock(sk);
  70	sk->sk_destruct = udpv6_destruct_sock;
  71	set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
  72	return 0;
  73}
  74
  75INDIRECT_CALLABLE_SCOPE
  76u32 udp6_ehashfn(const struct net *net,
  77		 const struct in6_addr *laddr,
  78		 const u16 lport,
  79		 const struct in6_addr *faddr,
  80		 const __be16 fport)
  81{
  82	u32 lhash, fhash;
  83
  84	net_get_random_once(&udp6_ehash_secret,
  85			    sizeof(udp6_ehash_secret));
  86	net_get_random_once(&udp_ipv6_hash_secret,
  87			    sizeof(udp_ipv6_hash_secret));
  88
  89	lhash = (__force u32)laddr->s6_addr32[3];
  90	fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
  91
  92	return __inet6_ehashfn(lhash, lport, fhash, fport,
  93			       udp6_ehash_secret + net_hash_mix(net));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  94}
  95
  96int udp_v6_get_port(struct sock *sk, unsigned short snum)
  97{
  98	unsigned int hash2_nulladdr =
  99		ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
 100	unsigned int hash2_partial =
 101		ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
 102
 103	/* precompute partial secondary hash */
 104	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
 105	return udp_lib_get_port(sk, snum, hash2_nulladdr);
 106}
 107
 108void udp_v6_rehash(struct sock *sk)
 109{
 110	u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
 111					  &sk->sk_v6_rcv_saddr,
 112					  inet_sk(sk)->inet_num);
 113	u16 new_hash4;
 114
 115	if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) {
 116		new_hash4 = udp_ehashfn(sock_net(sk),
 117					sk->sk_rcv_saddr, sk->sk_num,
 118					sk->sk_daddr, sk->sk_dport);
 119	} else {
 120		new_hash4 = udp6_ehashfn(sock_net(sk),
 121					 &sk->sk_v6_rcv_saddr, sk->sk_num,
 122					 &sk->sk_v6_daddr, sk->sk_dport);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 123	}
 124
 125	udp_lib_rehash(sk, new_hash, new_hash4);
 
 
 
 126}
 127
 128static int compute_score(struct sock *sk, const struct net *net,
 129			 const struct in6_addr *saddr, __be16 sport,
 130			 const struct in6_addr *daddr, unsigned short hnum,
 131			 int dif, int sdif)
 132{
 133	int bound_dev_if, score;
 134	struct inet_sock *inet;
 135	bool dev_match;
 136
 137	if (!net_eq(sock_net(sk), net) ||
 138	    udp_sk(sk)->udp_port_hash != hnum ||
 139	    sk->sk_family != PF_INET6)
 140		return -1;
 141
 142	if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
 143		return -1;
 144
 145	score = 0;
 146	inet = inet_sk(sk);
 147
 148	if (inet->inet_dport) {
 149		if (inet->inet_dport != sport)
 150			return -1;
 151		score++;
 152	}
 153
 154	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
 155		if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
 156			return -1;
 157		score++;
 158	}
 159
 160	bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
 161	dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif);
 162	if (!dev_match)
 163		return -1;
 164	if (bound_dev_if)
 165		score++;
 
 166
 167	if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
 168		score++;
 169
 170	return score;
 171}
 172
 173/**
 174 * udp6_lib_lookup1() - Simplified lookup using primary hash (destination port)
 175 * @net:	Network namespace
 176 * @saddr:	Source address, network order
 177 * @sport:	Source port, network order
 178 * @daddr:	Destination address, network order
 179 * @hnum:	Destination port, host order
 180 * @dif:	Destination interface index
 181 * @sdif:	Destination bridge port index, if relevant
 182 * @udptable:	Set of UDP hash tables
 183 *
 184 * Simplified lookup to be used as fallback if no sockets are found due to a
 185 * potential race between (receive) address change, and lookup happening before
 186 * the rehash operation. This function ignores SO_REUSEPORT groups while scoring
 187 * result sockets, because if we have one, we don't need the fallback at all.
 188 *
 189 * Called under rcu_read_lock().
 190 *
 191 * Return: socket with highest matching score if any, NULL if none
 192 */
 193static struct sock *udp6_lib_lookup1(const struct net *net,
 194				     const struct in6_addr *saddr, __be16 sport,
 195				     const struct in6_addr *daddr,
 196				     unsigned int hnum, int dif, int sdif,
 197				     const struct udp_table *udptable)
 198{
 199	unsigned int slot = udp_hashfn(net, hnum, udptable->mask);
 200	struct udp_hslot *hslot = &udptable->hash[slot];
 201	struct sock *sk, *result = NULL;
 202	int score, badness = 0;
 203
 204	sk_for_each_rcu(sk, &hslot->head) {
 205		score = compute_score(sk, net,
 206				      saddr, sport, daddr, hnum, dif, sdif);
 207		if (score > badness) {
 208			result = sk;
 209			badness = score;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 210		}
 211	}
 
 
 
 
 
 
 
 212
 
 
 
 
 
 
 
 
 
 
 213	return result;
 214}
 215
 216/* called with rcu_read_lock() */
 217static struct sock *udp6_lib_lookup2(const struct net *net,
 218		const struct in6_addr *saddr, __be16 sport,
 219		const struct in6_addr *daddr, unsigned int hnum,
 220		int dif, int sdif, struct udp_hslot *hslot2,
 221		struct sk_buff *skb)
 222{
 223	struct sock *sk, *result;
 224	int score, badness;
 225	bool need_rescore;
 226
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 227	result = NULL;
 228	badness = -1;
 229	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
 230		need_rescore = false;
 231rescore:
 232		score = compute_score(need_rescore ? result : sk, net, saddr,
 233				      sport, daddr, hnum, dif, sdif);
 234		if (score > badness) {
 
 235			badness = score;
 236
 237			if (need_rescore)
 238				continue;
 239
 240			if (sk->sk_state == TCP_ESTABLISHED) {
 241				result = sk;
 242				continue;
 
 
 
 
 
 
 
 
 
 243			}
 244
 245			result = inet6_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
 246							saddr, sport, daddr, hnum, udp6_ehashfn);
 247			if (!result) {
 248				result = sk;
 249				continue;
 250			}
 251
 252			/* Fall back to scoring if group has connections */
 253			if (!reuseport_has_conns(sk))
 254				return result;
 255
 256			/* Reuseport logic returned an error, keep original score. */
 257			if (IS_ERR(result))
 258				continue;
 259
 260			/* compute_score is too long of a function to be
 261			 * inlined, and calling it again here yields
 262			 * measureable overhead for some
 263			 * workloads. Work around it by jumping
 264			 * backwards to rescore 'result'.
 265			 */
 266			need_rescore = true;
 267			goto rescore;
 268		}
 269	}
 270	return result;
 271}
 272
 273#if IS_ENABLED(CONFIG_BASE_SMALL)
 274static struct sock *udp6_lib_lookup4(const struct net *net,
 275				     const struct in6_addr *saddr, __be16 sport,
 276				     const struct in6_addr *daddr,
 277				     unsigned int hnum, int dif, int sdif,
 278				     struct udp_table *udptable)
 279{
 280	return NULL;
 281}
 282
 283static void udp6_hash4(struct sock *sk)
 284{
 285}
 286#else /* !CONFIG_BASE_SMALL */
 287static struct sock *udp6_lib_lookup4(const struct net *net,
 288				     const struct in6_addr *saddr, __be16 sport,
 289				     const struct in6_addr *daddr,
 290				     unsigned int hnum, int dif, int sdif,
 291				     struct udp_table *udptable)
 292{
 293	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
 294	const struct hlist_nulls_node *node;
 295	struct udp_hslot *hslot4;
 296	unsigned int hash4, slot;
 297	struct udp_sock *up;
 298	struct sock *sk;
 299
 300	hash4 = udp6_ehashfn(net, daddr, hnum, saddr, sport);
 301	slot = hash4 & udptable->mask;
 302	hslot4 = &udptable->hash4[slot];
 303
 304begin:
 305	udp_lrpa_for_each_entry_rcu(up, node, &hslot4->nulls_head) {
 306		sk = (struct sock *)up;
 307		if (inet6_match(net, sk, saddr, daddr, ports, dif, sdif))
 308			return sk;
 309	}
 310
 311	/* if the nulls value we got at the end of this lookup is not the
 312	 * expected one, we must restart lookup. We probably met an item that
 313	 * was moved to another chain due to rehash.
 314	 */
 315	if (get_nulls_value(node) != slot)
 316		goto begin;
 317
 318	return NULL;
 319}
 320
 321static void udp6_hash4(struct sock *sk)
 322{
 323	struct net *net = sock_net(sk);
 324	unsigned int hash;
 325
 326	if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) {
 327		udp4_hash4(sk);
 328		return;
 329	}
 330
 331	if (sk_unhashed(sk) || ipv6_addr_any(&sk->sk_v6_rcv_saddr))
 332		return;
 333
 334	hash = udp6_ehashfn(net, &sk->sk_v6_rcv_saddr, sk->sk_num,
 335			    &sk->sk_v6_daddr, sk->sk_dport);
 336
 337	udp_lib_hash4(sk, hash);
 338}
 339#endif /* CONFIG_BASE_SMALL */
 340
 341/* rcu_read_lock() must be held */
 342struct sock *__udp6_lib_lookup(const struct net *net,
 343			       const struct in6_addr *saddr, __be16 sport,
 344			       const struct in6_addr *daddr, __be16 dport,
 345			       int dif, int sdif, struct udp_table *udptable,
 346			       struct sk_buff *skb)
 347{
 348	unsigned short hnum = ntohs(dport);
 349	struct udp_hslot *hslot2;
 350	struct sock *result, *sk;
 351	unsigned int hash2;
 352
 353	hash2 = ipv6_portaddr_hash(net, daddr, hnum);
 354	hslot2 = udp_hashslot2(udptable, hash2);
 355
 356	if (udp_has_hash4(hslot2)) {
 357		result = udp6_lib_lookup4(net, saddr, sport, daddr, hnum,
 358					  dif, sdif, udptable);
 359		if (result) /* udp6_lib_lookup4 return sk or NULL */
 360			return result;
 361	}
 362
 363	/* Lookup connected or non-wildcard sockets */
 364	result = udp6_lib_lookup2(net, saddr, sport,
 365				  daddr, hnum, dif, sdif,
 366				  hslot2, skb);
 367	if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
 368		goto done;
 369
 370	/* Lookup redirect from BPF */
 371	if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
 372	    udptable == net->ipv4.udp_table) {
 373		sk = inet6_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
 374						saddr, sport, daddr, hnum, dif,
 375						udp6_ehashfn);
 376		if (sk) {
 377			result = sk;
 378			goto done;
 379		}
 380	}
 381
 382	/* Got non-wildcard socket or error on first lookup */
 383	if (result)
 384		goto done;
 385
 386	/* Lookup wildcard sockets */
 387	hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
 388	hslot2 = udp_hashslot2(udptable, hash2);
 389
 390	result = udp6_lib_lookup2(net, saddr, sport,
 391				  &in6addr_any, hnum, dif, sdif,
 392				  hslot2, skb);
 393	if (!IS_ERR_OR_NULL(result))
 394		goto done;
 395
 396	/* Cover address change/lookup/rehash race: see __udp4_lib_lookup() */
 397	result = udp6_lib_lookup1(net, saddr, sport, daddr, hnum, dif, sdif,
 398				  udptable);
 399
 400done:
 401	if (IS_ERR(result))
 402		return NULL;
 403	return result;
 404}
 405EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
 406
 407static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
 408					  __be16 sport, __be16 dport,
 409					  struct udp_table *udptable)
 410{
 
 411	const struct ipv6hdr *iph = ipv6_hdr(skb);
 412
 413	return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
 
 
 
 414				 &iph->daddr, dport, inet6_iif(skb),
 415				 inet6_sdif(skb), udptable, skb);
 416}
 417
 418struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
 419				 __be16 sport, __be16 dport)
 420{
 421	const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
 422	const struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + offset);
 423	struct net *net = dev_net(skb->dev);
 424	int iif, sdif;
 425
 426	inet6_get_iif_sdif(skb, &iif, &sdif);
 427
 428	return __udp6_lib_lookup(net, &iph->saddr, sport,
 429				 &iph->daddr, dport, iif,
 430				 sdif, net->ipv4.udp_table, NULL);
 431}
 432
 433/* Must be called under rcu_read_lock().
 434 * Does increment socket refcount.
 435 */
 436#if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
 437struct sock *udp6_lib_lookup(const struct net *net, const struct in6_addr *saddr, __be16 sport,
 438			     const struct in6_addr *daddr, __be16 dport, int dif)
 439{
 440	struct sock *sk;
 441
 442	sk =  __udp6_lib_lookup(net, saddr, sport, daddr, dport,
 443				dif, 0, net->ipv4.udp_table, NULL);
 444	if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
 445		sk = NULL;
 446	return sk;
 447}
 448EXPORT_SYMBOL_GPL(udp6_lib_lookup);
 449#endif
 450
 451/* do not use the scratch area len for jumbogram: their length execeeds the
 452 * scratch area space; note that the IP6CB flags is still in the first
 453 * cacheline, so checking for jumbograms is cheap
 454 */
 455static int udp6_skb_len(struct sk_buff *skb)
 456{
 457	return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
 458}
 459
 460/*
 461 *	This should be easy, if there is something there we
 462 *	return it, otherwise we block.
 463 */
 464
 465int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 466		  int flags, int *addr_len)
 467{
 468	struct ipv6_pinfo *np = inet6_sk(sk);
 469	struct inet_sock *inet = inet_sk(sk);
 470	struct sk_buff *skb;
 471	unsigned int ulen, copied;
 472	int off, err, peeking = flags & MSG_PEEK;
 
 473	int is_udplite = IS_UDPLITE(sk);
 474	struct udp_mib __percpu *mib;
 475	bool checksum_valid = false;
 476	int is_udp4;
 
 477
 478	if (flags & MSG_ERRQUEUE)
 479		return ipv6_recv_error(sk, msg, len, addr_len);
 480
 481	if (np->rxpmtu && np->rxopt.bits.rxpmtu)
 482		return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
 483
 484try_again:
 485	off = sk_peek_offset(sk, flags);
 486	skb = __skb_recv_udp(sk, flags, &off, &err);
 487	if (!skb)
 488		return err;
 489
 490	ulen = udp6_skb_len(skb);
 491	copied = len;
 492	if (copied > ulen - off)
 493		copied = ulen - off;
 494	else if (copied < ulen)
 495		msg->msg_flags |= MSG_TRUNC;
 496
 497	is_udp4 = (skb->protocol == htons(ETH_P_IP));
 498	mib = __UDPX_MIB(sk, is_udp4);
 499
 500	/*
 501	 * If checksum is needed at all, try to do it while copying the
 502	 * data.  If the data is truncated, or if we only want a partial
 503	 * coverage checksum (UDP-Lite), do it before the copy.
 504	 */
 505
 506	if (copied < ulen || peeking ||
 507	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
 508		checksum_valid = udp_skb_csum_unnecessary(skb) ||
 509				!__udp_lib_checksum_complete(skb);
 510		if (!checksum_valid)
 511			goto csum_copy_err;
 512	}
 513
 514	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
 515		if (udp_skb_is_linear(skb))
 516			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
 517		else
 518			err = skb_copy_datagram_msg(skb, off, msg, copied);
 519	} else {
 520		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
 521		if (err == -EINVAL)
 522			goto csum_copy_err;
 523	}
 524	if (unlikely(err)) {
 525		if (!peeking) {
 
 526			atomic_inc(&sk->sk_drops);
 527			SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
 528		}
 529		kfree_skb(skb);
 530		return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 531	}
 532	if (!peeking)
 533		SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
 534
 535	sock_recv_cmsgs(msg, sk, skb);
 536
 537	/* Copy the address. */
 538	if (msg->msg_name) {
 539		DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
 540		sin6->sin6_family = AF_INET6;
 541		sin6->sin6_port = udp_hdr(skb)->source;
 542		sin6->sin6_flowinfo = 0;
 543
 544		if (is_udp4) {
 545			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
 546					       &sin6->sin6_addr);
 547			sin6->sin6_scope_id = 0;
 548		} else {
 549			sin6->sin6_addr = ipv6_hdr(skb)->saddr;
 550			sin6->sin6_scope_id =
 551				ipv6_iface_scope_id(&sin6->sin6_addr,
 552						    inet6_iif(skb));
 553		}
 554		*addr_len = sizeof(*sin6);
 555
 556		BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
 557						      (struct sockaddr *)sin6,
 558						      addr_len);
 559	}
 560
 561	if (udp_test_bit(GRO_ENABLED, sk))
 562		udp_cmsg_recv(msg, sk, skb);
 563
 564	if (np->rxopt.all)
 565		ip6_datagram_recv_common_ctl(sk, msg, skb);
 566
 567	if (is_udp4) {
 568		if (inet_cmsg_flags(inet))
 569			ip_cmsg_recv_offset(msg, sk, skb,
 570					    sizeof(struct udphdr), off);
 571	} else {
 572		if (np->rxopt.all)
 573			ip6_datagram_recv_specific_ctl(sk, msg, skb);
 574	}
 575
 576	err = copied;
 577	if (flags & MSG_TRUNC)
 578		err = ulen;
 579
 580	skb_consume_udp(sk, skb, peeking ? -err : err);
 
 
 581	return err;
 582
 583csum_copy_err:
 584	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
 585				 udp_skb_destructor)) {
 586		SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
 587		SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
 
 
 
 
 
 
 
 
 
 588	}
 589	kfree_skb(skb);
 590
 591	/* starting over for a new packet, but check if we need to yield */
 592	cond_resched();
 593	msg->msg_flags &= ~MSG_TRUNC;
 594	goto try_again;
 595}
 596
 597DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
 598void udpv6_encap_enable(void)
 599{
 600	static_branch_inc(&udpv6_encap_needed_key);
 601}
 602EXPORT_SYMBOL(udpv6_encap_enable);
 603
 604/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
 605 * through error handlers in encapsulations looking for a match.
 606 */
 607static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
 608				      struct inet6_skb_parm *opt,
 609				      u8 type, u8 code, int offset, __be32 info)
 610{
 611	int i;
 612
 613	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
 614		int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
 615			       u8 type, u8 code, int offset, __be32 info);
 616		const struct ip6_tnl_encap_ops *encap;
 617
 618		encap = rcu_dereference(ip6tun_encaps[i]);
 619		if (!encap)
 620			continue;
 621		handler = encap->err_handler;
 622		if (handler && !handler(skb, opt, type, code, offset, info))
 623			return 0;
 624	}
 625
 626	return -ENOENT;
 627}
 628
 629/* Try to match ICMP errors to UDP tunnels by looking up a socket without
 630 * reversing source and destination port: this will match tunnels that force the
 631 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
 632 * lwtunnels might actually break this assumption by being configured with
 633 * different destination ports on endpoints, in this case we won't be able to
 634 * trace ICMP messages back to them.
 635 *
 636 * If this doesn't match any socket, probe tunnels with arbitrary destination
 637 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
 638 * we've sent packets to won't necessarily match the local destination port.
 639 *
 640 * Then ask the tunnel implementation to match the error against a valid
 641 * association.
 642 *
 643 * Return an error if we can't find a match, the socket if we need further
 644 * processing, zero otherwise.
 645 */
 646static struct sock *__udp6_lib_err_encap(struct net *net,
 647					 const struct ipv6hdr *hdr, int offset,
 648					 struct udphdr *uh,
 649					 struct udp_table *udptable,
 650					 struct sock *sk,
 651					 struct sk_buff *skb,
 652					 struct inet6_skb_parm *opt,
 653					 u8 type, u8 code, __be32 info)
 654{
 655	int (*lookup)(struct sock *sk, struct sk_buff *skb);
 656	int network_offset, transport_offset;
 657	struct udp_sock *up;
 658
 659	network_offset = skb_network_offset(skb);
 660	transport_offset = skb_transport_offset(skb);
 661
 662	/* Network header needs to point to the outer IPv6 header inside ICMP */
 663	skb_reset_network_header(skb);
 664
 665	/* Transport header needs to point to the UDP header */
 666	skb_set_transport_header(skb, offset);
 667
 668	if (sk) {
 669		up = udp_sk(sk);
 670
 671		lookup = READ_ONCE(up->encap_err_lookup);
 672		if (lookup && lookup(sk, skb))
 673			sk = NULL;
 674
 675		goto out;
 676	}
 677
 678	sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
 679			       &hdr->saddr, uh->dest,
 680			       inet6_iif(skb), 0, udptable, skb);
 681	if (sk) {
 682		up = udp_sk(sk);
 683
 684		lookup = READ_ONCE(up->encap_err_lookup);
 685		if (!lookup || lookup(sk, skb))
 686			sk = NULL;
 687	}
 688
 689out:
 690	if (!sk) {
 691		sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
 692							offset, info));
 693	}
 694
 695	skb_set_transport_header(skb, transport_offset);
 696	skb_set_network_header(skb, network_offset);
 697
 698	return sk;
 699}
 700
 701int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 702		   u8 type, u8 code, int offset, __be32 info,
 703		   struct udp_table *udptable)
 704{
 705	struct ipv6_pinfo *np;
 706	const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
 707	const struct in6_addr *saddr = &hdr->saddr;
 708	const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr;
 709	struct udphdr *uh = (struct udphdr *)(skb->data+offset);
 710	bool tunnel = false;
 711	struct sock *sk;
 712	int harderr;
 713	int err;
 714	struct net *net = dev_net(skb->dev);
 715
 716	sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
 717			       inet6_iif(skb), inet6_sdif(skb), udptable, NULL);
 718
 719	if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
 720		/* No socket for error: try tunnels before discarding */
 721		if (static_branch_unlikely(&udpv6_encap_needed_key)) {
 722			sk = __udp6_lib_err_encap(net, hdr, offset, uh,
 723						  udptable, sk, skb,
 724						  opt, type, code, info);
 725			if (!sk)
 726				return 0;
 727		} else
 728			sk = ERR_PTR(-ENOENT);
 729
 730		if (IS_ERR(sk)) {
 731			__ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
 732					  ICMP6_MIB_INERRORS);
 733			return PTR_ERR(sk);
 734		}
 735
 736		tunnel = true;
 737	}
 738
 739	harderr = icmpv6_err_convert(type, code, &err);
 740	np = inet6_sk(sk);
 741
 742	if (type == ICMPV6_PKT_TOOBIG) {
 743		if (!ip6_sk_accept_pmtu(sk))
 744			goto out;
 745		ip6_sk_update_pmtu(skb, sk, info);
 746		if (READ_ONCE(np->pmtudisc) != IPV6_PMTUDISC_DONT)
 747			harderr = 1;
 748	}
 749	if (type == NDISC_REDIRECT) {
 750		if (tunnel) {
 751			ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
 752				     READ_ONCE(sk->sk_mark), sk->sk_uid);
 753		} else {
 754			ip6_sk_redirect(skb, sk);
 755		}
 756		goto out;
 757	}
 758
 759	/* Tunnels don't have an application socket: don't pass errors back */
 760	if (tunnel) {
 761		if (udp_sk(sk)->encap_err_rcv)
 762			udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest,
 763						  ntohl(info), (u8 *)(uh+1));
 764		goto out;
 765	}
 766
 767	if (!inet6_test_bit(RECVERR6, sk)) {
 768		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
 769			goto out;
 770	} else {
 771		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
 772	}
 773
 774	sk->sk_err = err;
 775	sk_error_report(sk);
 776out:
 777	return 0;
 778}
 779
 780static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 781{
 782	int rc;
 783
 784	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
 785		sock_rps_save_rxhash(sk, skb);
 786		sk_mark_napi_id(sk, skb);
 787		sk_incoming_cpu_update(sk);
 788	} else {
 789		sk_mark_napi_id_once(sk, skb);
 790	}
 791
 792	rc = __udp_enqueue_schedule_skb(sk, skb);
 793	if (rc < 0) {
 794		int is_udplite = IS_UDPLITE(sk);
 795		enum skb_drop_reason drop_reason;
 796
 797		/* Note that an ENOMEM error is charged twice */
 798		if (rc == -ENOMEM) {
 799			UDP6_INC_STATS(sock_net(sk),
 800					 UDP_MIB_RCVBUFERRORS, is_udplite);
 801			drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
 802		} else {
 803			UDP6_INC_STATS(sock_net(sk),
 804				       UDP_MIB_MEMERRORS, is_udplite);
 805			drop_reason = SKB_DROP_REASON_PROTO_MEM;
 806		}
 807		UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
 808		trace_udp_fail_queue_rcv_skb(rc, sk, skb);
 809		sk_skb_reason_drop(sk, skb, drop_reason);
 810		return -1;
 811	}
 
 
 812
 813	return 0;
 
 
 
 
 814}
 815
 816static __inline__ int udpv6_err(struct sk_buff *skb,
 817				struct inet6_skb_parm *opt, u8 type,
 818				u8 code, int offset, __be32 info)
 819{
 820	return __udp6_lib_err(skb, opt, type, code, offset, info,
 821			      dev_net(skb->dev)->ipv4.udp_table);
 822}
 
 823
 824static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
 825{
 826	enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
 827	struct udp_sock *up = udp_sk(sk);
 
 828	int is_udplite = IS_UDPLITE(sk);
 829
 830	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
 831		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
 832		goto drop;
 833	}
 834	nf_reset_ct(skb);
 835
 836	if (static_branch_unlikely(&udpv6_encap_needed_key) &&
 837	    READ_ONCE(up->encap_type)) {
 838		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
 839
 840		/*
 841		 * This is an encapsulation socket so pass the skb to
 842		 * the socket's udp_encap_rcv() hook. Otherwise, just
 843		 * fall through and pass this up the UDP socket.
 844		 * up->encap_rcv() returns the following value:
 845		 * =0 if skb was successfully passed to the encap
 846		 *    handler or was discarded by it.
 847		 * >0 if skb should be passed on to UDP.
 848		 * <0 if skb should be resubmitted as proto -N
 849		 */
 850
 851		/* if we're overly short, let UDP handle it */
 852		encap_rcv = READ_ONCE(up->encap_rcv);
 853		if (encap_rcv) {
 854			int ret;
 855
 856			/* Verify checksum before giving to encap */
 857			if (udp_lib_checksum_complete(skb))
 858				goto csum_error;
 859
 860			ret = encap_rcv(sk, skb);
 861			if (ret <= 0) {
 862				__UDP6_INC_STATS(sock_net(sk),
 863						 UDP_MIB_INDATAGRAMS,
 864						 is_udplite);
 865				return -ret;
 866			}
 867		}
 868
 869		/* FALLTHROUGH -- it's a UDP Packet */
 870	}
 871
 872	/*
 873	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
 874	 */
 875	if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
 876		u16 pcrlen = READ_ONCE(up->pcrlen);
 877
 878		if (pcrlen == 0) {          /* full coverage was set  */
 879			net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
 880					    UDP_SKB_CB(skb)->cscov, skb->len);
 881			goto drop;
 882		}
 883		if (UDP_SKB_CB(skb)->cscov < pcrlen) {
 884			net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
 885					    UDP_SKB_CB(skb)->cscov, pcrlen);
 886			goto drop;
 887		}
 888	}
 889
 890	prefetch(&sk->sk_rmem_alloc);
 891	if (rcu_access_pointer(sk->sk_filter) &&
 892	    udp_lib_checksum_complete(skb))
 893		goto csum_error;
 894
 895	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
 896		drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
 
 897		goto drop;
 898	}
 899
 900	udp_csum_pull_header(skb);
 901
 902	skb_dst_drop(skb);
 
 
 
 
 
 
 
 
 903
 904	return __udpv6_queue_rcv_skb(sk, skb);
 905
 906csum_error:
 907	drop_reason = SKB_DROP_REASON_UDP_CSUM;
 908	__UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
 909drop:
 910	__UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
 911	atomic_inc(&sk->sk_drops);
 912	sk_skb_reason_drop(sk, skb, drop_reason);
 913	return -1;
 914}
 915
 916static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 917{
 918	struct sk_buff *next, *segs;
 919	int ret;
 920
 921	if (likely(!udp_unexpected_gso(sk, skb)))
 922		return udpv6_queue_rcv_one_skb(sk, skb);
 923
 924	__skb_push(skb, -skb_mac_offset(skb));
 925	segs = udp_rcv_segment(sk, skb, false);
 926	skb_list_walk_safe(segs, skb, next) {
 927		__skb_pull(skb, skb_transport_offset(skb));
 928
 929		udp_post_segment_fix_csum(skb);
 930		ret = udpv6_queue_rcv_one_skb(sk, skb);
 931		if (ret > 0)
 932			ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
 933						 true);
 934	}
 935	return 0;
 936}
 937
 938static bool __udp_v6_is_mcast_sock(struct net *net, const struct sock *sk,
 939				   __be16 loc_port, const struct in6_addr *loc_addr,
 940				   __be16 rmt_port, const struct in6_addr *rmt_addr,
 941				   int dif, int sdif, unsigned short hnum)
 942{
 943	const struct inet_sock *inet = inet_sk(sk);
 944
 945	if (!net_eq(sock_net(sk), net))
 946		return false;
 947
 948	if (udp_sk(sk)->udp_port_hash != hnum ||
 949	    sk->sk_family != PF_INET6 ||
 950	    (inet->inet_dport && inet->inet_dport != rmt_port) ||
 951	    (!ipv6_addr_any(&sk->sk_v6_daddr) &&
 952		    !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
 953	    !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) ||
 954	    (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
 955		    !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
 956		return false;
 957	if (!inet6_mc_check(sk, loc_addr, rmt_addr))
 958		return false;
 959	return true;
 960}
 961
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 962static void udp6_csum_zero_error(struct sk_buff *skb)
 963{
 964	/* RFC 2460 section 8.1 says that we SHOULD log
 965	 * this error. Well, it is reasonable.
 966	 */
 967	net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
 968			    &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
 969			    &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
 970}
 971
 972/*
 973 * Note: called only from the BH handler context,
 974 * so we don't need to lock the hashes.
 975 */
 976static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
 977		const struct in6_addr *saddr, const struct in6_addr *daddr,
 978		struct udp_table *udptable, int proto)
 979{
 980	struct sock *sk, *first = NULL;
 981	const struct udphdr *uh = udp_hdr(skb);
 
 982	unsigned short hnum = ntohs(uh->dest);
 983	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
 984	unsigned int offset = offsetof(typeof(*sk), sk_node);
 
 985	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
 986	int dif = inet6_iif(skb);
 987	int sdif = inet6_sdif(skb);
 988	struct hlist_node *node;
 989	struct sk_buff *nskb;
 990
 991	if (use_hash2) {
 992		hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
 993			    udptable->mask;
 994		hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
 995start_lookup:
 996		hslot = &udptable->hash2[hash2].hslot;
 997		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
 998	}
 999
1000	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1001		if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
1002					    uh->source, saddr, dif, sdif,
1003					    hnum))
1004			continue;
1005		/* If zero checksum and no_check is not on for
1006		 * the socket then skip it.
1007		 */
1008		if (!uh->check && !udp_get_no_check6_rx(sk))
1009			continue;
1010		if (!first) {
1011			first = sk;
1012			continue;
1013		}
1014		nskb = skb_clone(skb, GFP_ATOMIC);
1015		if (unlikely(!nskb)) {
1016			atomic_inc(&sk->sk_drops);
1017			__UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1018					 IS_UDPLITE(sk));
1019			__UDP6_INC_STATS(net, UDP_MIB_INERRORS,
1020					 IS_UDPLITE(sk));
1021			continue;
1022		}
 
1023
1024		if (udpv6_queue_rcv_skb(sk, nskb) > 0)
1025			consume_skb(nskb);
1026	}
1027
1028	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
1029	if (use_hash2 && hash2 != hash2_any) {
1030		hash2 = hash2_any;
1031		goto start_lookup;
1032	}
1033
1034	if (first) {
1035		if (udpv6_queue_rcv_skb(first, skb) > 0)
1036			consume_skb(skb);
1037	} else {
1038		kfree_skb(skb);
1039		__UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
1040				 proto == IPPROTO_UDPLITE);
 
1041	}
1042	return 0;
1043}
1044
1045static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1046{
1047	if (udp_sk_rx_dst_set(sk, dst))
1048		sk->sk_rx_dst_cookie = rt6_get_cookie(dst_rt6_info(dst));
1049}
1050
1051/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
1052 * return code conversion for ip layer consumption
1053 */
1054static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
1055				struct udphdr *uh)
1056{
1057	int ret;
1058
1059	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1060		skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
1061
1062	ret = udpv6_queue_rcv_skb(sk, skb);
1063
1064	/* a return value > 0 means to resubmit the input */
1065	if (ret > 0)
1066		return ret;
1067	return 0;
1068}
1069
1070int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1071		   int proto)
1072{
1073	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1074	const struct in6_addr *saddr, *daddr;
1075	struct net *net = dev_net(skb->dev);
1076	struct sock *sk = NULL;
1077	struct udphdr *uh;
1078	bool refcounted;
1079	u32 ulen = 0;
1080
1081	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1082		goto discard;
1083
1084	saddr = &ipv6_hdr(skb)->saddr;
1085	daddr = &ipv6_hdr(skb)->daddr;
1086	uh = udp_hdr(skb);
1087
1088	ulen = ntohs(uh->len);
1089	if (ulen > skb->len)
1090		goto short_packet;
1091
1092	if (proto == IPPROTO_UDP) {
1093		/* UDP validates ulen. */
1094
1095		/* Check for jumbo payload */
1096		if (ulen == 0)
1097			ulen = skb->len;
1098
1099		if (ulen < sizeof(*uh))
1100			goto short_packet;
1101
1102		if (ulen < skb->len) {
1103			if (pskb_trim_rcsum(skb, ulen))
1104				goto short_packet;
1105			saddr = &ipv6_hdr(skb)->saddr;
1106			daddr = &ipv6_hdr(skb)->daddr;
1107			uh = udp_hdr(skb);
1108		}
1109	}
1110
1111	if (udp6_csum_init(skb, uh, proto))
1112		goto csum_error;
1113
1114	/* Check if the socket is already available, e.g. due to early demux */
1115	sk = inet6_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
1116			      &refcounted, udp6_ehashfn);
1117	if (IS_ERR(sk))
1118		goto no_sk;
1119
1120	if (sk) {
1121		struct dst_entry *dst = skb_dst(skb);
1122		int ret;
1123
1124		if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
1125			udp6_sk_rx_dst_set(sk, dst);
1126
1127		if (!uh->check && !udp_get_no_check6_rx(sk)) {
1128			if (refcounted)
1129				sock_put(sk);
1130			goto report_csum_error;
1131		}
1132
1133		ret = udp6_unicast_rcv_skb(sk, skb, uh);
1134		if (refcounted)
1135			sock_put(sk);
1136		return ret;
1137	}
1138
1139	/*
1140	 *	Multicast receive code
1141	 */
1142	if (ipv6_addr_is_multicast(daddr))
1143		return __udp6_lib_mcast_deliver(net, skb,
1144				saddr, daddr, udptable, proto);
1145
1146	/* Unicast */
 
 
 
 
 
1147	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1148	if (sk) {
1149		if (!uh->check && !udp_get_no_check6_rx(sk))
1150			goto report_csum_error;
1151		return udp6_unicast_rcv_skb(sk, skb, uh);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1152	}
1153no_sk:
1154	reason = SKB_DROP_REASON_NO_SOCKET;
1155
1156	if (!uh->check)
1157		goto report_csum_error;
 
 
1158
1159	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1160		goto discard;
1161	nf_reset_ct(skb);
1162
1163	if (udp_lib_checksum_complete(skb))
1164		goto csum_error;
1165
1166	__UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1167	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
1168
1169	sk_skb_reason_drop(sk, skb, reason);
1170	return 0;
1171
1172short_packet:
1173	if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1174		reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1175	net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
1176			    proto == IPPROTO_UDPLITE ? "-Lite" : "",
1177			    saddr, ntohs(uh->source),
1178			    ulen, skb->len,
1179			    daddr, ntohs(uh->dest));
1180	goto discard;
1181
1182report_csum_error:
1183	udp6_csum_zero_error(skb);
1184csum_error:
1185	if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1186		reason = SKB_DROP_REASON_UDP_CSUM;
1187	__UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1188discard:
1189	__UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1190	sk_skb_reason_drop(sk, skb, reason);
1191	return 0;
1192}
1193
1194
1195static struct sock *__udp6_lib_demux_lookup(struct net *net,
1196			__be16 loc_port, const struct in6_addr *loc_addr,
1197			__be16 rmt_port, const struct in6_addr *rmt_addr,
1198			int dif, int sdif)
1199{
1200	struct udp_table *udptable = net->ipv4.udp_table;
1201	unsigned short hnum = ntohs(loc_port);
1202	struct udp_hslot *hslot2;
1203	unsigned int hash2;
1204	__portpair ports;
1205	struct sock *sk;
1206
1207	hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
1208	hslot2 = udp_hashslot2(udptable, hash2);
1209	ports = INET_COMBINED_PORTS(rmt_port, hnum);
1210
1211	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1212		if (sk->sk_state == TCP_ESTABLISHED &&
1213		    inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif))
1214			return sk;
1215		/* Only check first socket in chain */
1216		break;
1217	}
1218	return NULL;
1219}
1220
1221void udp_v6_early_demux(struct sk_buff *skb)
1222{
1223	struct net *net = dev_net(skb->dev);
1224	const struct udphdr *uh;
1225	struct sock *sk;
1226	struct dst_entry *dst;
1227	int dif = skb->dev->ifindex;
1228	int sdif = inet6_sdif(skb);
1229
1230	if (!pskb_may_pull(skb, skb_transport_offset(skb) +
1231	    sizeof(struct udphdr)))
1232		return;
1233
1234	uh = udp_hdr(skb);
1235
1236	if (skb->pkt_type == PACKET_HOST)
1237		sk = __udp6_lib_demux_lookup(net, uh->dest,
1238					     &ipv6_hdr(skb)->daddr,
1239					     uh->source, &ipv6_hdr(skb)->saddr,
1240					     dif, sdif);
1241	else
1242		return;
1243
1244	if (!sk)
1245		return;
1246
1247	skb->sk = sk;
1248	DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
1249	skb->destructor = sock_pfree;
1250	dst = rcu_dereference(sk->sk_rx_dst);
1251
1252	if (dst)
1253		dst = dst_check(dst, sk->sk_rx_dst_cookie);
1254	if (dst) {
1255		/* set noref for now.
1256		 * any place which wants to hold dst has to call
1257		 * dst_hold_safe()
1258		 */
1259		skb_dst_set_noref(skb, dst);
1260	}
1261}
1262
1263INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1264{
1265	return __udp6_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
1266}
1267
1268/*
1269 * Throw away all pending data and cancel the corking. Socket is locked.
1270 */
1271static void udp_v6_flush_pending_frames(struct sock *sk)
1272{
1273	struct udp_sock *up = udp_sk(sk);
1274
1275	if (up->pending == AF_INET)
1276		udp_flush_pending_frames(sk);
1277	else if (up->pending) {
1278		up->len = 0;
1279		WRITE_ONCE(up->pending, 0);
1280		ip6_flush_pending_frames(sk);
1281	}
1282}
1283
1284static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
1285			     int addr_len)
1286{
1287	if (addr_len < offsetofend(struct sockaddr, sa_family))
1288		return -EINVAL;
1289	/* The following checks are replicated from __ip6_datagram_connect()
1290	 * and intended to prevent BPF program called below from accessing
1291	 * bytes that are out of the bound specified by user in addr_len.
1292	 */
1293	if (uaddr->sa_family == AF_INET) {
1294		if (ipv6_only_sock(sk))
1295			return -EAFNOSUPPORT;
1296		return udp_pre_connect(sk, uaddr, addr_len);
1297	}
1298
1299	if (addr_len < SIN6_LEN_RFC2133)
1300		return -EINVAL;
1301
1302	return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, &addr_len);
1303}
1304
1305static int udpv6_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1306{
1307	int res;
1308
1309	lock_sock(sk);
1310	res = __ip6_datagram_connect(sk, uaddr, addr_len);
1311	if (!res)
1312		udp6_hash4(sk);
1313	release_sock(sk);
1314	return res;
1315}
1316
1317/**
1318 *	udp6_hwcsum_outgoing  -  handle outgoing HW checksumming
1319 *	@sk:	socket we are sending on
1320 *	@skb:	sk_buff containing the filled-in UDP header
1321 *		(checksum field must be zeroed out)
1322 *	@saddr: source address
1323 *	@daddr: destination address
1324 *	@len:	length of packet
1325 */
1326static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1327				 const struct in6_addr *saddr,
1328				 const struct in6_addr *daddr, int len)
1329{
1330	unsigned int offset;
1331	struct udphdr *uh = udp_hdr(skb);
1332	struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1333	__wsum csum = 0;
1334
1335	if (!frags) {
1336		/* Only one fragment on the socket.  */
1337		skb->csum_start = skb_transport_header(skb) - skb->head;
1338		skb->csum_offset = offsetof(struct udphdr, check);
1339		uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1340	} else {
1341		/*
1342		 * HW-checksum won't work as there are two or more
1343		 * fragments on the socket so that all csums of sk_buffs
1344		 * should be together
1345		 */
1346		offset = skb_transport_offset(skb);
1347		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1348		csum = skb->csum;
1349
1350		skb->ip_summed = CHECKSUM_NONE;
1351
1352		do {
1353			csum = csum_add(csum, frags->csum);
1354		} while ((frags = frags->next));
1355
1356		uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1357					    csum);
1358		if (uh->check == 0)
1359			uh->check = CSUM_MANGLED_0;
1360	}
1361}
1362
1363/*
1364 *	Sending
1365 */
1366
1367static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1368			   struct inet_cork *cork)
1369{
1370	struct sock *sk = skb->sk;
1371	struct udphdr *uh;
1372	int err = 0;
1373	int is_udplite = IS_UDPLITE(sk);
1374	__wsum csum = 0;
1375	int offset = skb_transport_offset(skb);
1376	int len = skb->len - offset;
1377	int datalen = len - sizeof(*uh);
1378
1379	/*
1380	 * Create a UDP header
1381	 */
1382	uh = udp_hdr(skb);
1383	uh->source = fl6->fl6_sport;
1384	uh->dest = fl6->fl6_dport;
1385	uh->len = htons(len);
1386	uh->check = 0;
1387
1388	if (cork->gso_size) {
1389		const int hlen = skb_network_header_len(skb) +
1390				 sizeof(struct udphdr);
1391
1392		if (hlen + min(datalen, cork->gso_size) > cork->fragsize) {
1393			kfree_skb(skb);
1394			return -EMSGSIZE;
1395		}
1396		if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1397			kfree_skb(skb);
1398			return -EINVAL;
1399		}
1400		if (udp_get_no_check6_tx(sk)) {
1401			kfree_skb(skb);
1402			return -EINVAL;
1403		}
1404		if (is_udplite || dst_xfrm(skb_dst(skb))) {
1405			kfree_skb(skb);
1406			return -EIO;
1407		}
1408
1409		if (datalen > cork->gso_size) {
1410			skb_shinfo(skb)->gso_size = cork->gso_size;
1411			skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1412			skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1413								 cork->gso_size);
1414
1415			/* Don't checksum the payload, skb will get segmented */
1416			goto csum_partial;
1417		}
1418	}
1419
1420	if (is_udplite)
1421		csum = udplite_csum(skb);
1422	else if (udp_get_no_check6_tx(sk)) {   /* UDP csum disabled */
1423		skb->ip_summed = CHECKSUM_NONE;
1424		goto send;
1425	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1426csum_partial:
1427		udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1428		goto send;
1429	} else
1430		csum = udp_csum(skb);
1431
1432	/* add protocol-dependent pseudo-header */
1433	uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1434				    len, fl6->flowi6_proto, csum);
1435	if (uh->check == 0)
1436		uh->check = CSUM_MANGLED_0;
1437
1438send:
1439	err = ip6_send_skb(skb);
1440	if (err) {
1441		if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) {
1442			UDP6_INC_STATS(sock_net(sk),
1443				       UDP_MIB_SNDBUFERRORS, is_udplite);
1444			err = 0;
1445		}
1446	} else {
1447		UDP6_INC_STATS(sock_net(sk),
1448			       UDP_MIB_OUTDATAGRAMS, is_udplite);
1449	}
1450	return err;
1451}
1452
1453static int udp_v6_push_pending_frames(struct sock *sk)
1454{
1455	struct sk_buff *skb;
1456	struct udp_sock  *up = udp_sk(sk);
 
1457	int err = 0;
1458
1459	if (up->pending == AF_INET)
1460		return udp_push_pending_frames(sk);
1461
 
 
 
 
 
1462	skb = ip6_finish_skb(sk);
1463	if (!skb)
1464		goto out;
1465
1466	err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6,
1467			      &inet_sk(sk)->cork.base);
1468out:
1469	up->len = 0;
1470	WRITE_ONCE(up->pending, 0);
1471	return err;
1472}
1473
1474int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1475{
1476	struct ipv6_txoptions opt_space;
1477	struct udp_sock *up = udp_sk(sk);
1478	struct inet_sock *inet = inet_sk(sk);
1479	struct ipv6_pinfo *np = inet6_sk(sk);
1480	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1481	struct in6_addr *daddr, *final_p, final;
1482	struct ipv6_txoptions *opt = NULL;
1483	struct ipv6_txoptions *opt_to_free = NULL;
1484	struct ip6_flowlabel *flowlabel = NULL;
1485	struct inet_cork_full cork;
1486	struct flowi6 *fl6 = &cork.fl.u.ip6;
1487	struct dst_entry *dst;
1488	struct ipcm6_cookie ipc6;
1489	int addr_len = msg->msg_namelen;
1490	bool connected = false;
1491	int ulen = len;
1492	int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
 
 
 
1493	int err;
 
1494	int is_udplite = IS_UDPLITE(sk);
1495	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1496
1497	ipcm6_init(&ipc6);
1498	ipc6.gso_size = READ_ONCE(up->gso_size);
1499	ipc6.sockc.tsflags = READ_ONCE(sk->sk_tsflags);
1500	ipc6.sockc.mark = READ_ONCE(sk->sk_mark);
1501
1502	/* destination address check */
1503	if (sin6) {
1504		if (addr_len < offsetof(struct sockaddr, sa_data))
1505			return -EINVAL;
1506
1507		switch (sin6->sin6_family) {
1508		case AF_INET6:
1509			if (addr_len < SIN6_LEN_RFC2133)
1510				return -EINVAL;
1511			daddr = &sin6->sin6_addr;
1512			if (ipv6_addr_any(daddr) &&
1513			    ipv6_addr_v4mapped(&np->saddr))
1514				ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1515						       daddr);
1516			break;
1517		case AF_INET:
1518			goto do_udp_sendmsg;
1519		case AF_UNSPEC:
1520			msg->msg_name = sin6 = NULL;
1521			msg->msg_namelen = addr_len = 0;
1522			daddr = NULL;
1523			break;
1524		default:
1525			return -EINVAL;
1526		}
1527	} else if (!READ_ONCE(up->pending)) {
1528		if (sk->sk_state != TCP_ESTABLISHED)
1529			return -EDESTADDRREQ;
1530		daddr = &sk->sk_v6_daddr;
1531	} else
1532		daddr = NULL;
1533
1534	if (daddr) {
1535		if (ipv6_addr_v4mapped(daddr)) {
1536			struct sockaddr_in sin;
1537			sin.sin_family = AF_INET;
1538			sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1539			sin.sin_addr.s_addr = daddr->s6_addr32[3];
1540			msg->msg_name = &sin;
1541			msg->msg_namelen = sizeof(sin);
1542do_udp_sendmsg:
1543			err = ipv6_only_sock(sk) ?
1544				-ENETUNREACH : udp_sendmsg(sk, msg, len);
1545			msg->msg_name = sin6;
1546			msg->msg_namelen = addr_len;
1547			return err;
1548		}
1549	}
1550
 
 
 
1551	/* Rough check on arithmetic overflow,
1552	   better check is made in ip6_append_data().
1553	   */
1554	if (len > INT_MAX - sizeof(struct udphdr))
1555		return -EMSGSIZE;
1556
1557	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
1558	if (READ_ONCE(up->pending)) {
1559		if (READ_ONCE(up->pending) == AF_INET)
1560			return udp_sendmsg(sk, msg, len);
1561		/*
1562		 * There are pending frames.
1563		 * The socket lock must be held while it's corked.
1564		 */
1565		lock_sock(sk);
1566		if (likely(up->pending)) {
1567			if (unlikely(up->pending != AF_INET6)) {
1568				release_sock(sk);
1569				return -EAFNOSUPPORT;
1570			}
1571			dst = NULL;
1572			goto do_append_data;
1573		}
1574		release_sock(sk);
1575	}
1576	ulen += sizeof(struct udphdr);
1577
1578	memset(fl6, 0, sizeof(*fl6));
1579
1580	if (sin6) {
1581		if (sin6->sin6_port == 0)
1582			return -EINVAL;
1583
1584		fl6->fl6_dport = sin6->sin6_port;
1585		daddr = &sin6->sin6_addr;
1586
1587		if (inet6_test_bit(SNDFLOW, sk)) {
1588			fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1589			if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) {
1590				flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1591				if (IS_ERR(flowlabel))
1592					return -EINVAL;
1593			}
1594		}
1595
1596		/*
1597		 * Otherwise it will be difficult to maintain
1598		 * sk->sk_dst_cache.
1599		 */
1600		if (sk->sk_state == TCP_ESTABLISHED &&
1601		    ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1602			daddr = &sk->sk_v6_daddr;
1603
1604		if (addr_len >= sizeof(struct sockaddr_in6) &&
1605		    sin6->sin6_scope_id &&
1606		    __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1607			fl6->flowi6_oif = sin6->sin6_scope_id;
1608	} else {
1609		if (sk->sk_state != TCP_ESTABLISHED)
1610			return -EDESTADDRREQ;
1611
1612		fl6->fl6_dport = inet->inet_dport;
1613		daddr = &sk->sk_v6_daddr;
1614		fl6->flowlabel = np->flow_label;
1615		connected = true;
1616	}
1617
1618	if (!fl6->flowi6_oif)
1619		fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if);
1620
1621	if (!fl6->flowi6_oif)
1622		fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1623
1624	fl6->flowi6_uid = sk->sk_uid;
1625
1626	if (msg->msg_controllen) {
1627		opt = &opt_space;
1628		memset(opt, 0, sizeof(struct ipv6_txoptions));
1629		opt->tot_len = sizeof(*opt);
1630		ipc6.opt = opt;
1631
1632		err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1633		if (err > 0) {
1634			err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6,
1635						    &ipc6);
1636			connected = false;
1637		}
1638		if (err < 0) {
1639			fl6_sock_release(flowlabel);
1640			return err;
1641		}
1642		if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1643			flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1644			if (IS_ERR(flowlabel))
1645				return -EINVAL;
1646		}
1647		if (!(opt->opt_nflen|opt->opt_flen))
1648			opt = NULL;
 
1649	}
1650	if (!opt) {
1651		opt = txopt_get(np);
1652		opt_to_free = opt;
1653	}
1654	if (flowlabel)
1655		opt = fl6_merge_options(&opt_space, flowlabel, opt);
1656	opt = ipv6_fixup_options(&opt_space, opt);
1657	ipc6.opt = opt;
1658
1659	fl6->flowi6_proto = sk->sk_protocol;
1660	fl6->flowi6_mark = ipc6.sockc.mark;
1661	fl6->daddr = *daddr;
1662	if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr))
1663		fl6->saddr = np->saddr;
1664	fl6->fl6_sport = inet->inet_sport;
1665
1666	if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) {
1667		err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1668					   (struct sockaddr *)sin6,
1669					   &addr_len,
1670					   &fl6->saddr);
1671		if (err)
1672			goto out_no_dst;
1673		if (sin6) {
1674			if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1675				/* BPF program rewrote IPv6-only by IPv4-mapped
1676				 * IPv6. It's currently unsupported.
1677				 */
1678				err = -ENOTSUPP;
1679				goto out_no_dst;
1680			}
1681			if (sin6->sin6_port == 0) {
1682				/* BPF program set invalid port. Reject it. */
1683				err = -EINVAL;
1684				goto out_no_dst;
1685			}
1686			fl6->fl6_dport = sin6->sin6_port;
1687			fl6->daddr = sin6->sin6_addr;
1688		}
1689	}
1690
1691	if (ipv6_addr_any(&fl6->daddr))
1692		fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1693
1694	final_p = fl6_update_dst(fl6, opt, &final);
1695	if (final_p)
1696		connected = false;
1697
1698	if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) {
1699		fl6->flowi6_oif = READ_ONCE(np->mcast_oif);
1700		connected = false;
1701	} else if (!fl6->flowi6_oif)
1702		fl6->flowi6_oif = READ_ONCE(np->ucast_oif);
1703
1704	security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6));
1705
1706	if (ipc6.tclass < 0)
1707		ipc6.tclass = np->tclass;
1708
1709	fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel);
1710
1711	dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected);
1712	if (IS_ERR(dst)) {
1713		err = PTR_ERR(dst);
1714		dst = NULL;
1715		goto out;
1716	}
1717
1718	if (ipc6.hlimit < 0)
1719		ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst);
 
 
 
1720
1721	if (msg->msg_flags&MSG_CONFIRM)
1722		goto do_confirm;
1723back_from_confirm:
1724
1725	/* Lockless fast path for the non-corking case */
1726	if (!corkreq) {
1727		struct sk_buff *skb;
1728
1729		skb = ip6_make_skb(sk, getfrag, msg, ulen,
1730				   sizeof(struct udphdr), &ipc6,
1731				   dst_rt6_info(dst),
1732				   msg->msg_flags, &cork);
1733		err = PTR_ERR(skb);
1734		if (!IS_ERR_OR_NULL(skb))
1735			err = udp_v6_send_skb(skb, fl6, &cork.base);
1736		/* ip6_make_skb steals dst reference */
1737		goto out_no_dst;
1738	}
1739
1740	lock_sock(sk);
1741	if (unlikely(up->pending)) {
1742		/* The socket is already corked while preparing it. */
1743		/* ... which is an evident application bug. --ANK */
1744		release_sock(sk);
1745
1746		net_dbg_ratelimited("udp cork app bug 2\n");
1747		err = -EINVAL;
1748		goto out;
1749	}
1750
1751	WRITE_ONCE(up->pending, AF_INET6);
1752
1753do_append_data:
1754	if (ipc6.dontfrag < 0)
1755		ipc6.dontfrag = inet6_test_bit(DONTFRAG, sk);
1756	up->len += ulen;
1757	err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1758			      &ipc6, fl6, dst_rt6_info(dst),
1759			      corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
 
1760	if (err)
1761		udp_v6_flush_pending_frames(sk);
1762	else if (!corkreq)
1763		err = udp_v6_push_pending_frames(sk);
1764	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1765		WRITE_ONCE(up->pending, 0);
1766
1767	if (err > 0)
1768		err = inet6_test_bit(RECVERR6, sk) ? net_xmit_errno(err) : 0;
1769	release_sock(sk);
1770
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1771out:
1772	dst_release(dst);
1773out_no_dst:
1774	fl6_sock_release(flowlabel);
1775	txopt_put(opt_to_free);
1776	if (!err)
1777		return len;
1778	/*
1779	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1780	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1781	 * we don't have a good statistic (IpOutDiscards but it can be too many
1782	 * things).  We could add another new stat but at least for now that
1783	 * seems like overkill.
1784	 */
1785	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1786		UDP6_INC_STATS(sock_net(sk),
1787			       UDP_MIB_SNDBUFERRORS, is_udplite);
1788	}
1789	return err;
1790
1791do_confirm:
1792	if (msg->msg_flags & MSG_PROBE)
1793		dst_confirm_neigh(dst, &fl6->daddr);
1794	if (!(msg->msg_flags&MSG_PROBE) || len)
1795		goto back_from_confirm;
1796	err = 0;
1797	goto out;
1798}
1799EXPORT_SYMBOL(udpv6_sendmsg);
1800
1801static void udpv6_splice_eof(struct socket *sock)
1802{
1803	struct sock *sk = sock->sk;
1804	struct udp_sock *up = udp_sk(sk);
1805
1806	if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1807		return;
1808
1809	lock_sock(sk);
1810	if (up->pending && !udp_test_bit(CORK, sk))
1811		udp_v6_push_pending_frames(sk);
1812	release_sock(sk);
1813}
1814
1815void udpv6_destroy_sock(struct sock *sk)
1816{
1817	struct udp_sock *up = udp_sk(sk);
1818	lock_sock(sk);
1819
1820	/* protects from races with udp_abort() */
1821	sock_set_flag(sk, SOCK_DEAD);
1822	udp_v6_flush_pending_frames(sk);
1823	release_sock(sk);
1824
1825	if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1826		if (up->encap_type) {
1827			void (*encap_destroy)(struct sock *sk);
1828			encap_destroy = READ_ONCE(up->encap_destroy);
1829			if (encap_destroy)
1830				encap_destroy(sk);
1831		}
1832		if (udp_test_bit(ENCAP_ENABLED, sk)) {
1833			static_branch_dec(&udpv6_encap_needed_key);
1834			udp_encap_disable();
1835		}
1836	}
 
 
1837}
1838
1839/*
1840 *	Socket option code for UDP
1841 */
1842int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
1843		     unsigned int optlen)
1844{
1845	if (level == SOL_UDP  ||  level == SOL_UDPLITE || level == SOL_SOCKET)
1846		return udp_lib_setsockopt(sk, level, optname,
1847					  optval, optlen,
1848					  udp_v6_push_pending_frames);
1849	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1850}
1851
 
 
 
 
 
 
 
 
 
 
 
1852int udpv6_getsockopt(struct sock *sk, int level, int optname,
1853		     char __user *optval, int __user *optlen)
1854{
1855	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1856		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1857	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1858}
1859
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1860
1861/* ------------------------------------------------------------------------ */
1862#ifdef CONFIG_PROC_FS
1863int udp6_seq_show(struct seq_file *seq, void *v)
1864{
1865	if (v == SEQ_START_TOKEN) {
1866		seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1867	} else {
1868		int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1869		const struct inet_sock *inet = inet_sk((const struct sock *)v);
1870		__u16 srcp = ntohs(inet->inet_sport);
1871		__u16 destp = ntohs(inet->inet_dport);
1872		__ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1873					  udp_rqueue_get(v), bucket);
1874	}
1875	return 0;
1876}
1877
1878const struct seq_operations udp6_seq_ops = {
1879	.start		= udp_seq_start,
1880	.next		= udp_seq_next,
1881	.stop		= udp_seq_stop,
1882	.show		= udp6_seq_show,
 
1883};
1884EXPORT_SYMBOL(udp6_seq_ops);
1885
1886static struct udp_seq_afinfo udp6_seq_afinfo = {
 
1887	.family		= AF_INET6,
1888	.udp_table	= NULL,
 
 
 
 
1889};
1890
1891int __net_init udp6_proc_init(struct net *net)
1892{
1893	if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1894			sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1895		return -ENOMEM;
1896	return 0;
1897}
1898
1899void udp6_proc_exit(struct net *net)
1900{
1901	remove_proc_entry("udp6", net->proc_net);
1902}
1903#endif /* CONFIG_PROC_FS */
1904
 
 
 
 
 
 
 
 
 
 
 
1905/* ------------------------------------------------------------------------ */
1906
1907struct proto udpv6_prot = {
1908	.name			= "UDPv6",
1909	.owner			= THIS_MODULE,
1910	.close			= udp_lib_close,
1911	.pre_connect		= udpv6_pre_connect,
1912	.connect		= udpv6_connect,
1913	.disconnect		= udp_disconnect,
1914	.ioctl			= udp_ioctl,
1915	.init			= udpv6_init_sock,
1916	.destroy		= udpv6_destroy_sock,
1917	.setsockopt		= udpv6_setsockopt,
1918	.getsockopt		= udpv6_getsockopt,
1919	.sendmsg		= udpv6_sendmsg,
1920	.recvmsg		= udpv6_recvmsg,
1921	.splice_eof		= udpv6_splice_eof,
1922	.release_cb		= ip6_datagram_release_cb,
1923	.hash			= udp_lib_hash,
1924	.unhash			= udp_lib_unhash,
1925	.rehash			= udp_v6_rehash,
1926	.get_port		= udp_v6_get_port,
1927	.put_port		= udp_lib_unhash,
1928#ifdef CONFIG_BPF_SYSCALL
1929	.psock_update_sk_prot	= udp_bpf_update_proto,
 
 
 
 
 
1930#endif
1931
1932	.memory_allocated	= &udp_memory_allocated,
1933	.per_cpu_fw_alloc	= &udp_memory_per_cpu_fw_alloc,
1934
1935	.sysctl_mem		= sysctl_udp_mem,
1936	.sysctl_wmem_offset     = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1937	.sysctl_rmem_offset     = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1938	.obj_size		= sizeof(struct udp6_sock),
1939	.ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6),
1940	.h.udp_table		= NULL,
1941	.diag_destroy		= udp_abort,
1942};
1943
1944static struct inet_protosw udpv6_protosw = {
1945	.type =      SOCK_DGRAM,
1946	.protocol =  IPPROTO_UDP,
1947	.prot =      &udpv6_prot,
1948	.ops =       &inet6_dgram_ops,
1949	.flags =     INET_PROTOSW_PERMANENT,
1950};
1951
1952int __init udpv6_init(void)
1953{
1954	int ret;
1955
1956	net_hotdata.udpv6_protocol = (struct inet6_protocol) {
1957		.handler     = udpv6_rcv,
1958		.err_handler = udpv6_err,
1959		.flags	     = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1960	};
1961	ret = inet6_add_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1962	if (ret)
1963		goto out;
1964
1965	ret = inet6_register_protosw(&udpv6_protosw);
1966	if (ret)
1967		goto out_udpv6_protocol;
1968out:
1969	return ret;
1970
1971out_udpv6_protocol:
1972	inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1973	goto out;
1974}
1975
1976void udpv6_exit(void)
1977{
1978	inet6_unregister_protosw(&udpv6_protosw);
1979	inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1980}