Linux Audio

Check our new training course

Loading...
v4.17
 
   1/*
   2 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   3 *		operating system.  INET is implemented using the  BSD Socket
   4 *		interface as the means of communication with the user level.
   5 *
   6 *		Support for INET connection oriented protocols.
   7 *
   8 * Authors:	See the TCP sources
   9 *
  10 *		This program is free software; you can redistribute it and/or
  11 *		modify it under the terms of the GNU General Public License
  12 *		as published by the Free Software Foundation; either version
  13 *		2 of the License, or(at your option) any later version.
  14 */
  15
  16#include <linux/module.h>
  17#include <linux/jhash.h>
  18
  19#include <net/inet_connection_sock.h>
  20#include <net/inet_hashtables.h>
  21#include <net/inet_timewait_sock.h>
  22#include <net/ip.h>
  23#include <net/route.h>
  24#include <net/tcp_states.h>
  25#include <net/xfrm.h>
  26#include <net/tcp.h>
  27#include <net/sock_reuseport.h>
  28#include <net/addrconf.h>
  29
  30#ifdef INET_CSK_DEBUG
  31const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
  32EXPORT_SYMBOL(inet_csk_timer_bug_msg);
  33#endif
  34
  35#if IS_ENABLED(CONFIG_IPV6)
  36/* match_wildcard == true:  IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6
  37 *                          only, and any IPv4 addresses if not IPv6 only
  38 * match_wildcard == false: addresses must be exactly the same, i.e.
  39 *                          IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
  40 *                          and 0.0.0.0 equals to 0.0.0.0 only
 
  41 */
  42static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
  43				 const struct in6_addr *sk2_rcv_saddr6,
  44				 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
  45				 bool sk1_ipv6only, bool sk2_ipv6only,
  46				 bool match_wildcard)
 
  47{
  48	int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
  49	int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
  50
  51	/* if both are mapped, treat as IPv4 */
  52	if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
  53		if (!sk2_ipv6only) {
  54			if (sk1_rcv_saddr == sk2_rcv_saddr)
  55				return true;
  56			if (!sk1_rcv_saddr || !sk2_rcv_saddr)
  57				return match_wildcard;
  58		}
  59		return false;
  60	}
  61
  62	if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
  63		return true;
  64
  65	if (addr_type2 == IPV6_ADDR_ANY && match_wildcard &&
  66	    !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
  67		return true;
  68
  69	if (addr_type == IPV6_ADDR_ANY && match_wildcard &&
  70	    !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
  71		return true;
  72
  73	if (sk2_rcv_saddr6 &&
  74	    ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
  75		return true;
  76
  77	return false;
  78}
  79#endif
  80
  81/* match_wildcard == true:  0.0.0.0 equals to any IPv4 addresses
  82 * match_wildcard == false: addresses must be exactly the same, i.e.
  83 *                          0.0.0.0 only equals to 0.0.0.0
  84 */
  85static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
  86				 bool sk2_ipv6only, bool match_wildcard)
 
  87{
  88	if (!sk2_ipv6only) {
  89		if (sk1_rcv_saddr == sk2_rcv_saddr)
  90			return true;
  91		if (!sk1_rcv_saddr || !sk2_rcv_saddr)
  92			return match_wildcard;
  93	}
  94	return false;
  95}
  96
  97bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
  98			  bool match_wildcard)
  99{
 100#if IS_ENABLED(CONFIG_IPV6)
 101	if (sk->sk_family == AF_INET6)
 102		return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
 103					    inet6_rcv_saddr(sk2),
 104					    sk->sk_rcv_saddr,
 105					    sk2->sk_rcv_saddr,
 106					    ipv6_only_sock(sk),
 107					    ipv6_only_sock(sk2),
 
 108					    match_wildcard);
 109#endif
 110	return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr,
 111				    ipv6_only_sock(sk2), match_wildcard);
 
 112}
 113EXPORT_SYMBOL(inet_rcv_saddr_equal);
 114
 115void inet_get_local_port_range(struct net *net, int *low, int *high)
 116{
 117	unsigned int seq;
 
 
 
 
 
 118
 119	do {
 120		seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 121
 122		*low = net->ipv4.ip_local_ports.range[0];
 123		*high = net->ipv4.ip_local_ports.range[1];
 124	} while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
 
 
 
 
 
 
 
 125}
 126EXPORT_SYMBOL(inet_get_local_port_range);
 127
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 128static int inet_csk_bind_conflict(const struct sock *sk,
 129				  const struct inet_bind_bucket *tb,
 
 130				  bool relax, bool reuseport_ok)
 131{
 132	struct sock *sk2;
 133	bool reuse = sk->sk_reuse;
 134	bool reuseport = !!sk->sk_reuseport && reuseport_ok;
 135	kuid_t uid = sock_i_uid((struct sock *)sk);
 
 
 
 136
 137	/*
 138	 * Unlike other sk lookup places we do not check
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 139	 * for sk_net here, since _all_ the socks listed
 140	 * in tb->owners list belong to the same net - the
 141	 * one this bucket belongs to.
 142	 */
 
 
 
 143
 144	sk_for_each_bound(sk2, &tb->owners) {
 145		if (sk != sk2 &&
 146		    (!sk->sk_bound_dev_if ||
 147		     !sk2->sk_bound_dev_if ||
 148		     sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
 149			if ((!reuse || !sk2->sk_reuse ||
 150			    sk2->sk_state == TCP_LISTEN) &&
 151			    (!reuseport || !sk2->sk_reuseport ||
 152			     rcu_access_pointer(sk->sk_reuseport_cb) ||
 153			     (sk2->sk_state != TCP_TIME_WAIT &&
 154			     !uid_eq(uid, sock_i_uid(sk2))))) {
 155				if (inet_rcv_saddr_equal(sk, sk2, true))
 156					break;
 157			}
 158			if (!relax && reuse && sk2->sk_reuse &&
 159			    sk2->sk_state != TCP_LISTEN) {
 160				if (inet_rcv_saddr_equal(sk, sk2, true))
 161					break;
 162			}
 163		}
 164	}
 165	return sk2 != NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 166}
 167
 168/*
 169 * Find an open port number for the socket.  Returns with the
 170 * inet_bind_hashbucket lock held.
 171 */
 172static struct inet_bind_hashbucket *
 173inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret)
 174{
 175	struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
 176	int port = 0;
 177	struct inet_bind_hashbucket *head;
 
 
 178	struct net *net = sock_net(sk);
 179	int i, low, high, attempt_half;
 180	struct inet_bind_bucket *tb;
 181	u32 remaining, offset;
 
 182
 
 
 183	attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
 184other_half_scan:
 185	inet_get_local_port_range(net, &low, &high);
 186	high++; /* [32768, 60999] -> [32768, 61000[ */
 187	if (high - low < 4)
 188		attempt_half = 0;
 189	if (attempt_half) {
 190		int half = low + (((high - low) >> 2) << 1);
 191
 192		if (attempt_half == 1)
 193			high = half;
 194		else
 195			low = half;
 196	}
 197	remaining = high - low;
 198	if (likely(remaining > 1))
 199		remaining &= ~1U;
 200
 201	offset = prandom_u32() % remaining;
 202	/* __inet_hash_connect() favors ports having @low parity
 203	 * We do the opposite to not pollute connect() users.
 204	 */
 205	offset |= 1U;
 206
 207other_parity_scan:
 208	port = low + offset;
 209	for (i = 0; i < remaining; i += 2, port += 2) {
 210		if (unlikely(port >= high))
 211			port -= remaining;
 212		if (inet_is_local_reserved_port(net, port))
 213			continue;
 214		head = &hinfo->bhash[inet_bhashfn(net, port,
 215						  hinfo->bhash_size)];
 216		spin_lock_bh(&head->lock);
 
 
 
 
 
 
 
 
 217		inet_bind_bucket_for_each(tb, &head->chain)
 218			if (net_eq(ib_net(tb), net) && tb->port == port) {
 219				if (!inet_csk_bind_conflict(sk, tb, false, false))
 
 220					goto success;
 
 221				goto next_port;
 222			}
 223		tb = NULL;
 224		goto success;
 225next_port:
 226		spin_unlock_bh(&head->lock);
 227		cond_resched();
 228	}
 229
 230	offset--;
 231	if (!(offset & 1))
 232		goto other_parity_scan;
 233
 234	if (attempt_half == 1) {
 235		/* OK we now try the upper half of the range */
 236		attempt_half = 2;
 237		goto other_half_scan;
 238	}
 
 
 
 
 
 
 239	return NULL;
 240success:
 241	*port_ret = port;
 242	*tb_ret = tb;
 
 
 243	return head;
 244}
 245
 246static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
 247				     struct sock *sk)
 248{
 249	kuid_t uid = sock_i_uid(sk);
 250
 251	if (tb->fastreuseport <= 0)
 252		return 0;
 253	if (!sk->sk_reuseport)
 254		return 0;
 255	if (rcu_access_pointer(sk->sk_reuseport_cb))
 256		return 0;
 257	if (!uid_eq(tb->fastuid, uid))
 258		return 0;
 259	/* We only need to check the rcv_saddr if this tb was once marked
 260	 * without fastreuseport and then was reset, as we can only know that
 261	 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
 262	 * owners list.
 263	 */
 264	if (tb->fastreuseport == FASTREUSEPORT_ANY)
 265		return 1;
 266#if IS_ENABLED(CONFIG_IPV6)
 267	if (tb->fast_sk_family == AF_INET6)
 268		return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
 269					    inet6_rcv_saddr(sk),
 270					    tb->fast_rcv_saddr,
 271					    sk->sk_rcv_saddr,
 272					    tb->fast_ipv6_only,
 273					    ipv6_only_sock(sk), true);
 274#endif
 275	return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
 276				    ipv6_only_sock(sk), true);
 277}
 278
 279/* Obtain a reference to a local port for the given sock,
 280 * if snum is zero it means select any available local port.
 281 * We try to allocate an odd port (and leave even ports for connect())
 282 */
 283int inet_csk_get_port(struct sock *sk, unsigned short snum)
 284{
 285	bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
 286	struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
 287	int ret = 1, port = snum;
 288	struct inet_bind_hashbucket *head;
 289	struct net *net = sock_net(sk);
 290	struct inet_bind_bucket *tb = NULL;
 291	kuid_t uid = sock_i_uid(sk);
 
 292
 293	if (!port) {
 294		head = inet_csk_find_open_port(sk, &tb, &port);
 295		if (!head)
 296			return ret;
 297		if (!tb)
 298			goto tb_not_found;
 299		goto success;
 300	}
 301	head = &hinfo->bhash[inet_bhashfn(net, port,
 302					  hinfo->bhash_size)];
 303	spin_lock_bh(&head->lock);
 304	inet_bind_bucket_for_each(tb, &head->chain)
 305		if (net_eq(ib_net(tb), net) && tb->port == port)
 306			goto tb_found;
 307tb_not_found:
 308	tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
 309				     net, head, port);
 310	if (!tb)
 311		goto fail_unlock;
 312tb_found:
 313	if (!hlist_empty(&tb->owners)) {
 314		if (sk->sk_reuse == SK_FORCE_REUSE)
 315			goto success;
 316
 317		if ((tb->fastreuse > 0 && reuse) ||
 318		    sk_reuseport_match(tb, sk))
 319			goto success;
 320		if (inet_csk_bind_conflict(sk, tb, true, true))
 321			goto fail_unlock;
 322	}
 323success:
 324	if (hlist_empty(&tb->owners)) {
 325		tb->fastreuse = reuse;
 326		if (sk->sk_reuseport) {
 327			tb->fastreuseport = FASTREUSEPORT_ANY;
 328			tb->fastuid = uid;
 329			tb->fast_rcv_saddr = sk->sk_rcv_saddr;
 330			tb->fast_ipv6_only = ipv6_only_sock(sk);
 331			tb->fast_sk_family = sk->sk_family;
 332#if IS_ENABLED(CONFIG_IPV6)
 333			tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
 334#endif
 335		} else {
 336			tb->fastreuseport = 0;
 337		}
 338	} else {
 339		if (!reuse)
 340			tb->fastreuse = 0;
 341		if (sk->sk_reuseport) {
 342			/* We didn't match or we don't have fastreuseport set on
 343			 * the tb, but we have sk_reuseport set on this socket
 344			 * and we know that there are no bind conflicts with
 345			 * this socket in this tb, so reset our tb's reuseport
 346			 * settings so that any subsequent sockets that match
 347			 * our current socket will be put on the fast path.
 348			 *
 349			 * If we reset we need to set FASTREUSEPORT_STRICT so we
 350			 * do extra checking for all subsequent sk_reuseport
 351			 * socks.
 352			 */
 353			if (!sk_reuseport_match(tb, sk)) {
 354				tb->fastreuseport = FASTREUSEPORT_STRICT;
 355				tb->fastuid = uid;
 356				tb->fast_rcv_saddr = sk->sk_rcv_saddr;
 357				tb->fast_ipv6_only = ipv6_only_sock(sk);
 358				tb->fast_sk_family = sk->sk_family;
 359#if IS_ENABLED(CONFIG_IPV6)
 360				tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
 361#endif
 362			}
 363		} else {
 364			tb->fastreuseport = 0;
 365		}
 366	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 367	if (!inet_csk(sk)->icsk_bind_hash)
 368		inet_bind_hash(sk, tb, port);
 369	WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 
 370	ret = 0;
 371
 372fail_unlock:
 
 
 
 
 
 
 
 
 373	spin_unlock_bh(&head->lock);
 374	return ret;
 375}
 376EXPORT_SYMBOL_GPL(inet_csk_get_port);
 377
 378/*
 379 * Wait for an incoming connection, avoid race conditions. This must be called
 380 * with the socket locked.
 381 */
 382static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 383{
 384	struct inet_connection_sock *icsk = inet_csk(sk);
 385	DEFINE_WAIT(wait);
 386	int err;
 387
 388	/*
 389	 * True wake-one mechanism for incoming connections: only
 390	 * one process gets woken up, not the 'whole herd'.
 391	 * Since we do not 'race & poll' for established sockets
 392	 * anymore, the common case will execute the loop only once.
 393	 *
 394	 * Subtle issue: "add_wait_queue_exclusive()" will be added
 395	 * after any current non-exclusive waiters, and we know that
 396	 * it will always _stay_ after any new non-exclusive waiters
 397	 * because all non-exclusive waiters are added at the
 398	 * beginning of the wait-queue. As such, it's ok to "drop"
 399	 * our exclusiveness temporarily when we get woken up without
 400	 * having to remove and re-insert us on the wait queue.
 401	 */
 402	for (;;) {
 403		prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 404					  TASK_INTERRUPTIBLE);
 405		release_sock(sk);
 406		if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 407			timeo = schedule_timeout(timeo);
 408		sched_annotate_sleep();
 409		lock_sock(sk);
 410		err = 0;
 411		if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 412			break;
 413		err = -EINVAL;
 414		if (sk->sk_state != TCP_LISTEN)
 415			break;
 416		err = sock_intr_errno(timeo);
 417		if (signal_pending(current))
 418			break;
 419		err = -EAGAIN;
 420		if (!timeo)
 421			break;
 422	}
 423	finish_wait(sk_sleep(sk), &wait);
 424	return err;
 425}
 426
 427/*
 428 * This will accept the next outstanding connection.
 429 */
 430struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
 431{
 432	struct inet_connection_sock *icsk = inet_csk(sk);
 433	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 434	struct request_sock *req;
 435	struct sock *newsk;
 436	int error;
 437
 438	lock_sock(sk);
 439
 440	/* We need to make sure that this socket is listening,
 441	 * and that it has something pending.
 442	 */
 443	error = -EINVAL;
 444	if (sk->sk_state != TCP_LISTEN)
 445		goto out_err;
 446
 447	/* Find already established connection */
 448	if (reqsk_queue_empty(queue)) {
 449		long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 450
 451		/* If this is a non blocking socket don't sleep */
 452		error = -EAGAIN;
 453		if (!timeo)
 454			goto out_err;
 455
 456		error = inet_csk_wait_for_connect(sk, timeo);
 457		if (error)
 458			goto out_err;
 459	}
 460	req = reqsk_queue_remove(queue, sk);
 461	newsk = req->sk;
 462
 463	if (sk->sk_protocol == IPPROTO_TCP &&
 464	    tcp_rsk(req)->tfo_listener) {
 465		spin_lock_bh(&queue->fastopenq.lock);
 466		if (tcp_rsk(req)->tfo_listener) {
 467			/* We are still waiting for the final ACK from 3WHS
 468			 * so can't free req now. Instead, we set req->sk to
 469			 * NULL to signify that the child socket is taken
 470			 * so reqsk_fastopen_remove() will free the req
 471			 * when 3WHS finishes (or is aborted).
 472			 */
 473			req->sk = NULL;
 474			req = NULL;
 475		}
 476		spin_unlock_bh(&queue->fastopenq.lock);
 477	}
 
 478out:
 479	release_sock(sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 480	if (req)
 481		reqsk_put(req);
 
 
 
 
 482	return newsk;
 483out_err:
 484	newsk = NULL;
 485	req = NULL;
 486	*err = error;
 487	goto out;
 488}
 489EXPORT_SYMBOL(inet_csk_accept);
 490
 491/*
 492 * Using different timers for retransmit, delayed acks and probes
 493 * We may wish use just one timer maintaining a list of expire jiffies
 494 * to optimize.
 495 */
 496void inet_csk_init_xmit_timers(struct sock *sk,
 497			       void (*retransmit_handler)(struct timer_list *t),
 498			       void (*delack_handler)(struct timer_list *t),
 499			       void (*keepalive_handler)(struct timer_list *t))
 500{
 501	struct inet_connection_sock *icsk = inet_csk(sk);
 502
 503	timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
 504	timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
 505	timer_setup(&sk->sk_timer, keepalive_handler, 0);
 506	icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 507}
 508EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 509
 510void inet_csk_clear_xmit_timers(struct sock *sk)
 511{
 512	struct inet_connection_sock *icsk = inet_csk(sk);
 513
 514	icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 515
 516	sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 517	sk_stop_timer(sk, &icsk->icsk_delack_timer);
 518	sk_stop_timer(sk, &sk->sk_timer);
 519}
 520EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 521
 522void inet_csk_delete_keepalive_timer(struct sock *sk)
 523{
 524	sk_stop_timer(sk, &sk->sk_timer);
 525}
 526EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 527
 528void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 529{
 530	sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 531}
 532EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 533
 534struct dst_entry *inet_csk_route_req(const struct sock *sk,
 535				     struct flowi4 *fl4,
 536				     const struct request_sock *req)
 537{
 538	const struct inet_request_sock *ireq = inet_rsk(req);
 539	struct net *net = read_pnet(&ireq->ireq_net);
 540	struct ip_options_rcu *opt;
 541	struct rtable *rt;
 542
 543	opt = ireq_opt_deref(ireq);
 
 544
 545	flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 546			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 547			   sk->sk_protocol, inet_sk_flowi_flags(sk),
 548			   (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 549			   ireq->ir_loc_addr, ireq->ir_rmt_port,
 550			   htons(ireq->ir_num), sk->sk_uid);
 551	security_req_classify_flow(req, flowi4_to_flowi(fl4));
 552	rt = ip_route_output_flow(net, fl4, sk);
 553	if (IS_ERR(rt))
 554		goto no_route;
 555	if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 556		goto route_err;
 
 557	return &rt->dst;
 558
 559route_err:
 560	ip_rt_put(rt);
 561no_route:
 
 562	__IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 563	return NULL;
 564}
 565EXPORT_SYMBOL_GPL(inet_csk_route_req);
 566
 567struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
 568					    struct sock *newsk,
 569					    const struct request_sock *req)
 570{
 571	const struct inet_request_sock *ireq = inet_rsk(req);
 572	struct net *net = read_pnet(&ireq->ireq_net);
 573	struct inet_sock *newinet = inet_sk(newsk);
 574	struct ip_options_rcu *opt;
 575	struct flowi4 *fl4;
 576	struct rtable *rt;
 577
 578	opt = rcu_dereference(ireq->ireq_opt);
 579	fl4 = &newinet->cork.fl.u.ip4;
 580
 581	flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 582			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 583			   sk->sk_protocol, inet_sk_flowi_flags(sk),
 584			   (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 585			   ireq->ir_loc_addr, ireq->ir_rmt_port,
 586			   htons(ireq->ir_num), sk->sk_uid);
 587	security_req_classify_flow(req, flowi4_to_flowi(fl4));
 588	rt = ip_route_output_flow(net, fl4, sk);
 589	if (IS_ERR(rt))
 590		goto no_route;
 591	if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 592		goto route_err;
 593	return &rt->dst;
 594
 595route_err:
 596	ip_rt_put(rt);
 597no_route:
 598	__IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 599	return NULL;
 600}
 601EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
 602
 603#if IS_ENABLED(CONFIG_IPV6)
 604#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 605#else
 606#define AF_INET_FAMILY(fam) true
 607#endif
 608
 609/* Decide when to expire the request and when to resend SYN-ACK */
 610static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 611				  const int max_retries,
 612				  const u8 rskq_defer_accept,
 613				  int *expire, int *resend)
 614{
 615	if (!rskq_defer_accept) {
 616		*expire = req->num_timeout >= thresh;
 617		*resend = 1;
 618		return;
 619	}
 620	*expire = req->num_timeout >= thresh &&
 621		  (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
 622	/*
 623	 * Do not resend while waiting for data after ACK,
 624	 * start to resend on end of deferring period to give
 625	 * last chance for data or ACK to create established socket.
 626	 */
 627	*resend = !inet_rsk(req)->acked ||
 628		  req->num_timeout >= rskq_defer_accept - 1;
 629}
 630
 631int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
 632{
 633	int err = req->rsk_ops->rtx_syn_ack(parent, req);
 634
 635	if (!err)
 636		req->num_retrans++;
 637	return err;
 638}
 639EXPORT_SYMBOL(inet_rtx_syn_ack);
 640
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 641/* return true if req was found in the ehash table */
 642static bool reqsk_queue_unlink(struct request_sock_queue *queue,
 643			       struct request_sock *req)
 644{
 645	struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
 646	bool found = false;
 647
 648	if (sk_hashed(req_to_sk(req))) {
 
 649		spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
 650
 651		spin_lock(lock);
 652		found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
 653		spin_unlock(lock);
 654	}
 655	if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
 656		reqsk_put(req);
 657	return found;
 658}
 659
 660void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
 661{
 662	if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
 
 
 663		reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 664		reqsk_put(req);
 665	}
 
 666}
 667EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
 668
 669void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
 670{
 671	inet_csk_reqsk_queue_drop(sk, req);
 672	reqsk_put(req);
 673}
 674EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
 675
 676static void reqsk_timer_handler(struct timer_list *t)
 677{
 678	struct request_sock *req = from_timer(req, t, rsk_timer);
 
 679	struct sock *sk_listener = req->rsk_listener;
 680	struct net *net = sock_net(sk_listener);
 681	struct inet_connection_sock *icsk = inet_csk(sk_listener);
 682	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 683	int qlen, expire = 0, resend = 0;
 684	int max_retries, thresh;
 685	u8 defer_accept;
 686
 687	if (inet_sk_state_load(sk_listener) != TCP_LISTEN)
 688		goto drop;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 689
 690	max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
 691	thresh = max_retries;
 
 
 692	/* Normally all the openreqs are young and become mature
 693	 * (i.e. converted to established socket) for first timeout.
 694	 * If synack was not acknowledged for 1 second, it means
 695	 * one of the following things: synack was lost, ack was lost,
 696	 * rtt is high or nobody planned to ack (i.e. synflood).
 697	 * When server is a bit loaded, queue is populated with old
 698	 * open requests, reducing effective size of queue.
 699	 * When server is well loaded, queue size reduces to zero
 700	 * after several minutes of work. It is not synflood,
 701	 * it is normal operation. The solution is pruning
 702	 * too old entries overriding normal timeout, when
 703	 * situation becomes dangerous.
 704	 *
 705	 * Essentially, we reserve half of room for young
 706	 * embrions; and abort old ones without pity, if old
 707	 * ones are about to clog our table.
 708	 */
 
 709	qlen = reqsk_queue_len(queue);
 710	if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
 711		int young = reqsk_queue_len_young(queue) << 1;
 712
 713		while (thresh > 2) {
 714			if (qlen < young)
 715				break;
 716			thresh--;
 717			young <<= 1;
 718		}
 719	}
 720	defer_accept = READ_ONCE(queue->rskq_defer_accept);
 721	if (defer_accept)
 722		max_retries = defer_accept;
 723	syn_ack_recalc(req, thresh, max_retries, defer_accept,
 724		       &expire, &resend);
 725	req->rsk_ops->syn_ack_timeout(req);
 726	if (!expire &&
 727	    (!resend ||
 728	     !inet_rtx_syn_ack(sk_listener, req) ||
 729	     inet_rsk(req)->acked)) {
 730		unsigned long timeo;
 731
 732		if (req->num_timeout++ == 0)
 733			atomic_dec(&queue->young);
 734		timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
 735		mod_timer(&req->rsk_timer, jiffies + timeo);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 736		return;
 737	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 738drop:
 739	inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
 740}
 741
 742static void reqsk_queue_hash_req(struct request_sock *req,
 743				 unsigned long timeout)
 744{
 745	req->num_retrans = 0;
 746	req->num_timeout = 0;
 747	req->sk = NULL;
 748
 749	timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
 750	mod_timer(&req->rsk_timer, jiffies + timeout);
 751
 752	inet_ehash_insert(req_to_sk(req), NULL);
 753	/* before letting lookups find us, make sure all req fields
 754	 * are committed to memory and refcnt initialized.
 755	 */
 756	smp_wmb();
 757	refcount_set(&req->rsk_refcnt, 2 + 1);
 758}
 759
 760void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 761				   unsigned long timeout)
 762{
 763	reqsk_queue_hash_req(req, timeout);
 764	inet_csk_reqsk_queue_added(sk);
 765}
 766EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 767
 
 
 
 
 
 
 
 
 
 
 
 768/**
 769 *	inet_csk_clone_lock - clone an inet socket, and lock its clone
 770 *	@sk: the socket to clone
 771 *	@req: request_sock
 772 *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 773 *
 774 *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
 775 */
 776struct sock *inet_csk_clone_lock(const struct sock *sk,
 777				 const struct request_sock *req,
 778				 const gfp_t priority)
 779{
 780	struct sock *newsk = sk_clone_lock(sk, priority);
 781
 782	if (newsk) {
 783		struct inet_connection_sock *newicsk = inet_csk(newsk);
 784
 785		inet_sk_set_state(newsk, TCP_SYN_RECV);
 786		newicsk->icsk_bind_hash = NULL;
 
 787
 788		inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
 789		inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
 790		inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
 791
 792		/* listeners have SOCK_RCU_FREE, not the children */
 793		sock_reset_flag(newsk, SOCK_RCU_FREE);
 794
 795		inet_sk(newsk)->mc_list = NULL;
 796
 797		newsk->sk_mark = inet_rsk(req)->ir_mark;
 798		atomic64_set(&newsk->sk_cookie,
 799			     atomic64_read(&inet_rsk(req)->ir_cookie));
 800
 801		newicsk->icsk_retransmits = 0;
 802		newicsk->icsk_backoff	  = 0;
 803		newicsk->icsk_probes_out  = 0;
 
 804
 805		/* Deinitialize accept_queue to trap illegal accesses. */
 806		memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 807
 
 
 808		security_inet_csk_clone(newsk, req);
 809	}
 810	return newsk;
 811}
 812EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
 813
 814/*
 815 * At this point, there should be no process reference to this
 816 * socket, and thus no user references at all.  Therefore we
 817 * can assume the socket waitqueue is inactive and nobody will
 818 * try to jump onto it.
 819 */
 820void inet_csk_destroy_sock(struct sock *sk)
 821{
 822	WARN_ON(sk->sk_state != TCP_CLOSE);
 823	WARN_ON(!sock_flag(sk, SOCK_DEAD));
 824
 825	/* It cannot be in hash table! */
 826	WARN_ON(!sk_unhashed(sk));
 827
 828	/* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 829	WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 830
 831	sk->sk_prot->destroy(sk);
 832
 833	sk_stream_kill_queues(sk);
 834
 835	xfrm_sk_free_policy(sk);
 836
 837	sk_refcnt_debug_release(sk);
 838
 839	percpu_counter_dec(sk->sk_prot->orphan_count);
 840
 841	sock_put(sk);
 842}
 843EXPORT_SYMBOL(inet_csk_destroy_sock);
 844
 845/* This function allows to force a closure of a socket after the call to
 846 * tcp/dccp_create_openreq_child().
 847 */
 848void inet_csk_prepare_forced_close(struct sock *sk)
 849	__releases(&sk->sk_lock.slock)
 850{
 851	/* sk_clone_lock locked the socket and set refcnt to 2 */
 852	bh_unlock_sock(sk);
 853	sock_put(sk);
 854
 855	/* The below has to be done to allow calling inet_csk_destroy_sock */
 856	sock_set_flag(sk, SOCK_DEAD);
 857	percpu_counter_inc(sk->sk_prot->orphan_count);
 858	inet_sk(sk)->inet_num = 0;
 859}
 860EXPORT_SYMBOL(inet_csk_prepare_forced_close);
 861
 862int inet_csk_listen_start(struct sock *sk, int backlog)
 
 
 
 
 
 
 
 
 
 
 863{
 864	struct inet_connection_sock *icsk = inet_csk(sk);
 865	struct inet_sock *inet = inet_sk(sk);
 866	int err = -EADDRINUSE;
 
 
 
 
 867
 868	reqsk_queue_alloc(&icsk->icsk_accept_queue);
 869
 870	sk->sk_max_ack_backlog = backlog;
 871	sk->sk_ack_backlog = 0;
 872	inet_csk_delack_init(sk);
 873
 874	/* There is race window here: we announce ourselves listening,
 875	 * but this transition is still not validated by get_port().
 876	 * It is OK, because this socket enters to hash table only
 877	 * after validation is complete.
 878	 */
 879	inet_sk_state_store(sk, TCP_LISTEN);
 880	if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 
 881		inet->inet_sport = htons(inet->inet_num);
 882
 883		sk_dst_reset(sk);
 884		err = sk->sk_prot->hash(sk);
 885
 886		if (likely(!err))
 887			return 0;
 888	}
 889
 890	inet_sk_set_state(sk, TCP_CLOSE);
 891	return err;
 892}
 893EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 894
 895static void inet_child_forget(struct sock *sk, struct request_sock *req,
 896			      struct sock *child)
 897{
 898	sk->sk_prot->disconnect(child, O_NONBLOCK);
 899
 900	sock_orphan(child);
 901
 902	percpu_counter_inc(sk->sk_prot->orphan_count);
 903
 904	if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
 905		BUG_ON(tcp_sk(child)->fastopen_rsk != req);
 906		BUG_ON(sk != req->rsk_listener);
 907
 908		/* Paranoid, to prevent race condition if
 909		 * an inbound pkt destined for child is
 910		 * blocked by sock lock in tcp_v4_rcv().
 911		 * Also to satisfy an assertion in
 912		 * tcp_v4_destroy_sock().
 913		 */
 914		tcp_sk(child)->fastopen_rsk = NULL;
 915	}
 916	inet_csk_destroy_sock(child);
 917}
 918
 919struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
 920				      struct request_sock *req,
 921				      struct sock *child)
 922{
 923	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
 924
 925	spin_lock(&queue->rskq_lock);
 926	if (unlikely(sk->sk_state != TCP_LISTEN)) {
 927		inet_child_forget(sk, req, child);
 928		child = NULL;
 929	} else {
 930		req->sk = child;
 931		req->dl_next = NULL;
 932		if (queue->rskq_accept_head == NULL)
 933			queue->rskq_accept_head = req;
 934		else
 935			queue->rskq_accept_tail->dl_next = req;
 936		queue->rskq_accept_tail = req;
 937		sk_acceptq_added(sk);
 938	}
 939	spin_unlock(&queue->rskq_lock);
 940	return child;
 941}
 942EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
 943
 944struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
 945					 struct request_sock *req, bool own_req)
 946{
 947	if (own_req) {
 948		inet_csk_reqsk_queue_drop(sk, req);
 949		reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 950		if (inet_csk_reqsk_queue_add(sk, req, child))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 951			return child;
 
 952	}
 953	/* Too bad, another child took ownership of the request, undo. */
 
 954	bh_unlock_sock(child);
 955	sock_put(child);
 956	return NULL;
 957}
 958EXPORT_SYMBOL(inet_csk_complete_hashdance);
 959
 960/*
 961 *	This routine closes sockets which have been at least partially
 962 *	opened, but not yet accepted.
 963 */
 964void inet_csk_listen_stop(struct sock *sk)
 965{
 966	struct inet_connection_sock *icsk = inet_csk(sk);
 967	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 968	struct request_sock *next, *req;
 969
 970	/* Following specs, it would be better either to send FIN
 971	 * (and enter FIN-WAIT-1, it is normal close)
 972	 * or to send active reset (abort).
 973	 * Certainly, it is pretty dangerous while synflood, but it is
 974	 * bad justification for our negligence 8)
 975	 * To be honest, we are not able to make either
 976	 * of the variants now.			--ANK
 977	 */
 978	while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
 979		struct sock *child = req->sk;
 
 980
 981		local_bh_disable();
 982		bh_lock_sock(child);
 983		WARN_ON(sock_owned_by_user(child));
 984		sock_hold(child);
 985
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 986		inet_child_forget(sk, req, child);
 
 987		reqsk_put(req);
 988		bh_unlock_sock(child);
 989		local_bh_enable();
 990		sock_put(child);
 991
 992		cond_resched();
 993	}
 994	if (queue->fastopenq.rskq_rst_head) {
 995		/* Free all the reqs queued in rskq_rst_head. */
 996		spin_lock_bh(&queue->fastopenq.lock);
 997		req = queue->fastopenq.rskq_rst_head;
 998		queue->fastopenq.rskq_rst_head = NULL;
 999		spin_unlock_bh(&queue->fastopenq.lock);
1000		while (req != NULL) {
1001			next = req->dl_next;
1002			reqsk_put(req);
1003			req = next;
1004		}
1005	}
1006	WARN_ON_ONCE(sk->sk_ack_backlog);
1007}
1008EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1009
1010void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1011{
1012	struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1013	const struct inet_sock *inet = inet_sk(sk);
1014
1015	sin->sin_family		= AF_INET;
1016	sin->sin_addr.s_addr	= inet->inet_daddr;
1017	sin->sin_port		= inet->inet_dport;
1018}
1019EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1020
1021#ifdef CONFIG_COMPAT
1022int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
1023			       char __user *optval, int __user *optlen)
1024{
1025	const struct inet_connection_sock *icsk = inet_csk(sk);
1026
1027	if (icsk->icsk_af_ops->compat_getsockopt)
1028		return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
1029							    optval, optlen);
1030	return icsk->icsk_af_ops->getsockopt(sk, level, optname,
1031					     optval, optlen);
1032}
1033EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
1034
1035int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
1036			       char __user *optval, unsigned int optlen)
1037{
1038	const struct inet_connection_sock *icsk = inet_csk(sk);
1039
1040	if (icsk->icsk_af_ops->compat_setsockopt)
1041		return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
1042							    optval, optlen);
1043	return icsk->icsk_af_ops->setsockopt(sk, level, optname,
1044					     optval, optlen);
1045}
1046EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
1047#endif
1048
1049static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1050{
1051	const struct inet_sock *inet = inet_sk(sk);
1052	const struct ip_options_rcu *inet_opt;
1053	__be32 daddr = inet->inet_daddr;
1054	struct flowi4 *fl4;
1055	struct rtable *rt;
1056
1057	rcu_read_lock();
1058	inet_opt = rcu_dereference(inet->inet_opt);
1059	if (inet_opt && inet_opt->opt.srr)
1060		daddr = inet_opt->opt.faddr;
1061	fl4 = &fl->u.ip4;
1062	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1063				   inet->inet_saddr, inet->inet_dport,
1064				   inet->inet_sport, sk->sk_protocol,
1065				   RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1066	if (IS_ERR(rt))
1067		rt = NULL;
1068	if (rt)
1069		sk_setup_caps(sk, &rt->dst);
1070	rcu_read_unlock();
1071
1072	return &rt->dst;
1073}
1074
1075struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1076{
1077	struct dst_entry *dst = __sk_dst_check(sk, 0);
1078	struct inet_sock *inet = inet_sk(sk);
1079
1080	if (!dst) {
1081		dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1082		if (!dst)
1083			goto out;
1084	}
1085	dst->ops->update_pmtu(dst, sk, NULL, mtu);
1086
1087	dst = __sk_dst_check(sk, 0);
1088	if (!dst)
1089		dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1090out:
1091	return dst;
1092}
1093EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);
v6.8
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		Support for INET connection oriented protocols.
   8 *
   9 * Authors:	See the TCP sources
 
 
 
 
 
  10 */
  11
  12#include <linux/module.h>
  13#include <linux/jhash.h>
  14
  15#include <net/inet_connection_sock.h>
  16#include <net/inet_hashtables.h>
  17#include <net/inet_timewait_sock.h>
  18#include <net/ip.h>
  19#include <net/route.h>
  20#include <net/tcp_states.h>
  21#include <net/xfrm.h>
  22#include <net/tcp.h>
  23#include <net/sock_reuseport.h>
  24#include <net/addrconf.h>
  25
 
 
 
 
 
  26#if IS_ENABLED(CONFIG_IPV6)
  27/* match_sk*_wildcard == true:  IPV6_ADDR_ANY equals to any IPv6 addresses
  28 *				if IPv6 only, and any IPv4 addresses
  29 *				if not IPv6 only
  30 * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
  31 *				IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
  32 *				and 0.0.0.0 equals to 0.0.0.0 only
  33 */
  34static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
  35				 const struct in6_addr *sk2_rcv_saddr6,
  36				 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
  37				 bool sk1_ipv6only, bool sk2_ipv6only,
  38				 bool match_sk1_wildcard,
  39				 bool match_sk2_wildcard)
  40{
  41	int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
  42	int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
  43
  44	/* if both are mapped, treat as IPv4 */
  45	if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
  46		if (!sk2_ipv6only) {
  47			if (sk1_rcv_saddr == sk2_rcv_saddr)
  48				return true;
  49			return (match_sk1_wildcard && !sk1_rcv_saddr) ||
  50				(match_sk2_wildcard && !sk2_rcv_saddr);
  51		}
  52		return false;
  53	}
  54
  55	if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
  56		return true;
  57
  58	if (addr_type2 == IPV6_ADDR_ANY && match_sk2_wildcard &&
  59	    !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
  60		return true;
  61
  62	if (addr_type == IPV6_ADDR_ANY && match_sk1_wildcard &&
  63	    !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
  64		return true;
  65
  66	if (sk2_rcv_saddr6 &&
  67	    ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
  68		return true;
  69
  70	return false;
  71}
  72#endif
  73
  74/* match_sk*_wildcard == true:  0.0.0.0 equals to any IPv4 addresses
  75 * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
  76 *				0.0.0.0 only equals to 0.0.0.0
  77 */
  78static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
  79				 bool sk2_ipv6only, bool match_sk1_wildcard,
  80				 bool match_sk2_wildcard)
  81{
  82	if (!sk2_ipv6only) {
  83		if (sk1_rcv_saddr == sk2_rcv_saddr)
  84			return true;
  85		return (match_sk1_wildcard && !sk1_rcv_saddr) ||
  86			(match_sk2_wildcard && !sk2_rcv_saddr);
  87	}
  88	return false;
  89}
  90
  91bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
  92			  bool match_wildcard)
  93{
  94#if IS_ENABLED(CONFIG_IPV6)
  95	if (sk->sk_family == AF_INET6)
  96		return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
  97					    inet6_rcv_saddr(sk2),
  98					    sk->sk_rcv_saddr,
  99					    sk2->sk_rcv_saddr,
 100					    ipv6_only_sock(sk),
 101					    ipv6_only_sock(sk2),
 102					    match_wildcard,
 103					    match_wildcard);
 104#endif
 105	return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr,
 106				    ipv6_only_sock(sk2), match_wildcard,
 107				    match_wildcard);
 108}
 109EXPORT_SYMBOL(inet_rcv_saddr_equal);
 110
 111bool inet_rcv_saddr_any(const struct sock *sk)
 112{
 113#if IS_ENABLED(CONFIG_IPV6)
 114	if (sk->sk_family == AF_INET6)
 115		return ipv6_addr_any(&sk->sk_v6_rcv_saddr);
 116#endif
 117	return !sk->sk_rcv_saddr;
 118}
 119
 120/**
 121 *	inet_sk_get_local_port_range - fetch ephemeral ports range
 122 *	@sk: socket
 123 *	@low: pointer to low port
 124 *	@high: pointer to high port
 125 *
 126 *	Fetch netns port range (/proc/sys/net/ipv4/ip_local_port_range)
 127 *	Range can be overridden if socket got IP_LOCAL_PORT_RANGE option.
 128 *	Returns true if IP_LOCAL_PORT_RANGE was set on this socket.
 129 */
 130bool inet_sk_get_local_port_range(const struct sock *sk, int *low, int *high)
 131{
 132	int lo, hi, sk_lo, sk_hi;
 133	bool local_range = false;
 134	u32 sk_range;
 135
 136	inet_get_local_port_range(sock_net(sk), &lo, &hi);
 137
 138	sk_range = READ_ONCE(inet_sk(sk)->local_port_range);
 139	if (unlikely(sk_range)) {
 140		sk_lo = sk_range & 0xffff;
 141		sk_hi = sk_range >> 16;
 142
 143		if (lo <= sk_lo && sk_lo <= hi)
 144			lo = sk_lo;
 145		if (lo <= sk_hi && sk_hi <= hi)
 146			hi = sk_hi;
 147		local_range = true;
 148	}
 149
 150	*low = lo;
 151	*high = hi;
 152	return local_range;
 153}
 154EXPORT_SYMBOL(inet_sk_get_local_port_range);
 155
 156static bool inet_use_bhash2_on_bind(const struct sock *sk)
 157{
 158#if IS_ENABLED(CONFIG_IPV6)
 159	if (sk->sk_family == AF_INET6) {
 160		int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
 161
 162		if (addr_type == IPV6_ADDR_ANY)
 163			return false;
 164
 165		if (addr_type != IPV6_ADDR_MAPPED)
 166			return true;
 167	}
 168#endif
 169	return sk->sk_rcv_saddr != htonl(INADDR_ANY);
 170}
 171
 172static bool inet_bind_conflict(const struct sock *sk, struct sock *sk2,
 173			       kuid_t sk_uid, bool relax,
 174			       bool reuseport_cb_ok, bool reuseport_ok)
 175{
 176	int bound_dev_if2;
 177
 178	if (sk == sk2)
 179		return false;
 180
 181	bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
 182
 183	if (!sk->sk_bound_dev_if || !bound_dev_if2 ||
 184	    sk->sk_bound_dev_if == bound_dev_if2) {
 185		if (sk->sk_reuse && sk2->sk_reuse &&
 186		    sk2->sk_state != TCP_LISTEN) {
 187			if (!relax || (!reuseport_ok && sk->sk_reuseport &&
 188				       sk2->sk_reuseport && reuseport_cb_ok &&
 189				       (sk2->sk_state == TCP_TIME_WAIT ||
 190					uid_eq(sk_uid, sock_i_uid(sk2)))))
 191				return true;
 192		} else if (!reuseport_ok || !sk->sk_reuseport ||
 193			   !sk2->sk_reuseport || !reuseport_cb_ok ||
 194			   (sk2->sk_state != TCP_TIME_WAIT &&
 195			    !uid_eq(sk_uid, sock_i_uid(sk2)))) {
 196			return true;
 197		}
 198	}
 199	return false;
 200}
 201
 202static bool __inet_bhash2_conflict(const struct sock *sk, struct sock *sk2,
 203				   kuid_t sk_uid, bool relax,
 204				   bool reuseport_cb_ok, bool reuseport_ok)
 205{
 206	if (sk->sk_family == AF_INET && ipv6_only_sock(sk2))
 207		return false;
 208
 209	return inet_bind_conflict(sk, sk2, sk_uid, relax,
 210				  reuseport_cb_ok, reuseport_ok);
 211}
 212
 213static bool inet_bhash2_conflict(const struct sock *sk,
 214				 const struct inet_bind2_bucket *tb2,
 215				 kuid_t sk_uid,
 216				 bool relax, bool reuseport_cb_ok,
 217				 bool reuseport_ok)
 218{
 219	struct sock *sk2;
 220
 221	sk_for_each_bound(sk2, &tb2->owners) {
 222		if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax,
 223					   reuseport_cb_ok, reuseport_ok))
 224			return true;
 225	}
 226
 227	return false;
 228}
 229
 230#define sk_for_each_bound_bhash(__sk, __tb2, __tb)			\
 231	hlist_for_each_entry(__tb2, &(__tb)->bhash2, bhash_node)	\
 232		sk_for_each_bound(sk2, &(__tb2)->owners)
 233
 234/* This should be called only when the tb and tb2 hashbuckets' locks are held */
 235static int inet_csk_bind_conflict(const struct sock *sk,
 236				  const struct inet_bind_bucket *tb,
 237				  const struct inet_bind2_bucket *tb2, /* may be null */
 238				  bool relax, bool reuseport_ok)
 239{
 
 
 
 240	kuid_t uid = sock_i_uid((struct sock *)sk);
 241	struct sock_reuseport *reuseport_cb;
 242	bool reuseport_cb_ok;
 243	struct sock *sk2;
 244
 245	rcu_read_lock();
 246	reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
 247	/* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
 248	reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
 249	rcu_read_unlock();
 250
 251	/* Conflicts with an existing IPV6_ADDR_ANY (if ipv6) or INADDR_ANY (if
 252	 * ipv4) should have been checked already. We need to do these two
 253	 * checks separately because their spinlocks have to be acquired/released
 254	 * independently of each other, to prevent possible deadlocks
 255	 */
 256	if (inet_use_bhash2_on_bind(sk))
 257		return tb2 && inet_bhash2_conflict(sk, tb2, uid, relax,
 258						   reuseport_cb_ok, reuseport_ok);
 259
 260	/* Unlike other sk lookup places we do not check
 261	 * for sk_net here, since _all_ the socks listed
 262	 * in tb->owners and tb2->owners list belong
 263	 * to the same net - the one this bucket belongs to.
 264	 */
 265	sk_for_each_bound_bhash(sk2, tb2, tb) {
 266		if (!inet_bind_conflict(sk, sk2, uid, relax, reuseport_cb_ok, reuseport_ok))
 267			continue;
 268
 269		if (inet_rcv_saddr_equal(sk, sk2, true))
 270			return true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 271	}
 272
 273	return false;
 274}
 275
 276/* Determine if there is a bind conflict with an existing IPV6_ADDR_ANY (if ipv6) or
 277 * INADDR_ANY (if ipv4) socket.
 278 *
 279 * Caller must hold bhash hashbucket lock with local bh disabled, to protect
 280 * against concurrent binds on the port for addr any
 281 */
 282static bool inet_bhash2_addr_any_conflict(const struct sock *sk, int port, int l3mdev,
 283					  bool relax, bool reuseport_ok)
 284{
 285	kuid_t uid = sock_i_uid((struct sock *)sk);
 286	const struct net *net = sock_net(sk);
 287	struct sock_reuseport *reuseport_cb;
 288	struct inet_bind_hashbucket *head2;
 289	struct inet_bind2_bucket *tb2;
 290	bool reuseport_cb_ok;
 291
 292	rcu_read_lock();
 293	reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
 294	/* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
 295	reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
 296	rcu_read_unlock();
 297
 298	head2 = inet_bhash2_addr_any_hashbucket(sk, net, port);
 299
 300	spin_lock(&head2->lock);
 301
 302	inet_bind_bucket_for_each(tb2, &head2->chain)
 303		if (inet_bind2_bucket_match_addr_any(tb2, net, port, l3mdev, sk))
 304			break;
 305
 306	if (tb2 && inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
 307					reuseport_ok)) {
 308		spin_unlock(&head2->lock);
 309		return true;
 310	}
 311
 312	spin_unlock(&head2->lock);
 313	return false;
 314}
 315
 316/*
 317 * Find an open port number for the socket.  Returns with the
 318 * inet_bind_hashbucket locks held if successful.
 319 */
 320static struct inet_bind_hashbucket *
 321inet_csk_find_open_port(const struct sock *sk, struct inet_bind_bucket **tb_ret,
 322			struct inet_bind2_bucket **tb2_ret,
 323			struct inet_bind_hashbucket **head2_ret, int *port_ret)
 324{
 325	struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
 326	int i, low, high, attempt_half, port, l3mdev;
 327	struct inet_bind_hashbucket *head, *head2;
 328	struct net *net = sock_net(sk);
 329	struct inet_bind2_bucket *tb2;
 330	struct inet_bind_bucket *tb;
 331	u32 remaining, offset;
 332	bool relax = false;
 333
 334	l3mdev = inet_sk_bound_l3mdev(sk);
 335ports_exhausted:
 336	attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
 337other_half_scan:
 338	inet_sk_get_local_port_range(sk, &low, &high);
 339	high++; /* [32768, 60999] -> [32768, 61000[ */
 340	if (high - low < 4)
 341		attempt_half = 0;
 342	if (attempt_half) {
 343		int half = low + (((high - low) >> 2) << 1);
 344
 345		if (attempt_half == 1)
 346			high = half;
 347		else
 348			low = half;
 349	}
 350	remaining = high - low;
 351	if (likely(remaining > 1))
 352		remaining &= ~1U;
 353
 354	offset = get_random_u32_below(remaining);
 355	/* __inet_hash_connect() favors ports having @low parity
 356	 * We do the opposite to not pollute connect() users.
 357	 */
 358	offset |= 1U;
 359
 360other_parity_scan:
 361	port = low + offset;
 362	for (i = 0; i < remaining; i += 2, port += 2) {
 363		if (unlikely(port >= high))
 364			port -= remaining;
 365		if (inet_is_local_reserved_port(net, port))
 366			continue;
 367		head = &hinfo->bhash[inet_bhashfn(net, port,
 368						  hinfo->bhash_size)];
 369		spin_lock_bh(&head->lock);
 370		if (inet_use_bhash2_on_bind(sk)) {
 371			if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, relax, false))
 372				goto next_port;
 373		}
 374
 375		head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
 376		spin_lock(&head2->lock);
 377		tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
 378		inet_bind_bucket_for_each(tb, &head->chain)
 379			if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
 380				if (!inet_csk_bind_conflict(sk, tb, tb2,
 381							    relax, false))
 382					goto success;
 383				spin_unlock(&head2->lock);
 384				goto next_port;
 385			}
 386		tb = NULL;
 387		goto success;
 388next_port:
 389		spin_unlock_bh(&head->lock);
 390		cond_resched();
 391	}
 392
 393	offset--;
 394	if (!(offset & 1))
 395		goto other_parity_scan;
 396
 397	if (attempt_half == 1) {
 398		/* OK we now try the upper half of the range */
 399		attempt_half = 2;
 400		goto other_half_scan;
 401	}
 402
 403	if (READ_ONCE(net->ipv4.sysctl_ip_autobind_reuse) && !relax) {
 404		/* We still have a chance to connect to different destinations */
 405		relax = true;
 406		goto ports_exhausted;
 407	}
 408	return NULL;
 409success:
 410	*port_ret = port;
 411	*tb_ret = tb;
 412	*tb2_ret = tb2;
 413	*head2_ret = head2;
 414	return head;
 415}
 416
 417static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
 418				     struct sock *sk)
 419{
 420	kuid_t uid = sock_i_uid(sk);
 421
 422	if (tb->fastreuseport <= 0)
 423		return 0;
 424	if (!sk->sk_reuseport)
 425		return 0;
 426	if (rcu_access_pointer(sk->sk_reuseport_cb))
 427		return 0;
 428	if (!uid_eq(tb->fastuid, uid))
 429		return 0;
 430	/* We only need to check the rcv_saddr if this tb was once marked
 431	 * without fastreuseport and then was reset, as we can only know that
 432	 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
 433	 * owners list.
 434	 */
 435	if (tb->fastreuseport == FASTREUSEPORT_ANY)
 436		return 1;
 437#if IS_ENABLED(CONFIG_IPV6)
 438	if (tb->fast_sk_family == AF_INET6)
 439		return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
 440					    inet6_rcv_saddr(sk),
 441					    tb->fast_rcv_saddr,
 442					    sk->sk_rcv_saddr,
 443					    tb->fast_ipv6_only,
 444					    ipv6_only_sock(sk), true, false);
 445#endif
 446	return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
 447				    ipv6_only_sock(sk), true, false);
 448}
 449
 450void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
 451			       struct sock *sk)
 
 
 
 452{
 
 
 
 
 
 
 453	kuid_t uid = sock_i_uid(sk);
 454	bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
 455
 456	if (hlist_empty(&tb->bhash2)) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 457		tb->fastreuse = reuse;
 458		if (sk->sk_reuseport) {
 459			tb->fastreuseport = FASTREUSEPORT_ANY;
 460			tb->fastuid = uid;
 461			tb->fast_rcv_saddr = sk->sk_rcv_saddr;
 462			tb->fast_ipv6_only = ipv6_only_sock(sk);
 463			tb->fast_sk_family = sk->sk_family;
 464#if IS_ENABLED(CONFIG_IPV6)
 465			tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
 466#endif
 467		} else {
 468			tb->fastreuseport = 0;
 469		}
 470	} else {
 471		if (!reuse)
 472			tb->fastreuse = 0;
 473		if (sk->sk_reuseport) {
 474			/* We didn't match or we don't have fastreuseport set on
 475			 * the tb, but we have sk_reuseport set on this socket
 476			 * and we know that there are no bind conflicts with
 477			 * this socket in this tb, so reset our tb's reuseport
 478			 * settings so that any subsequent sockets that match
 479			 * our current socket will be put on the fast path.
 480			 *
 481			 * If we reset we need to set FASTREUSEPORT_STRICT so we
 482			 * do extra checking for all subsequent sk_reuseport
 483			 * socks.
 484			 */
 485			if (!sk_reuseport_match(tb, sk)) {
 486				tb->fastreuseport = FASTREUSEPORT_STRICT;
 487				tb->fastuid = uid;
 488				tb->fast_rcv_saddr = sk->sk_rcv_saddr;
 489				tb->fast_ipv6_only = ipv6_only_sock(sk);
 490				tb->fast_sk_family = sk->sk_family;
 491#if IS_ENABLED(CONFIG_IPV6)
 492				tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
 493#endif
 494			}
 495		} else {
 496			tb->fastreuseport = 0;
 497		}
 498	}
 499}
 500
 501/* Obtain a reference to a local port for the given sock,
 502 * if snum is zero it means select any available local port.
 503 * We try to allocate an odd port (and leave even ports for connect())
 504 */
 505int inet_csk_get_port(struct sock *sk, unsigned short snum)
 506{
 507	struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
 508	bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
 509	bool found_port = false, check_bind_conflict = true;
 510	bool bhash_created = false, bhash2_created = false;
 511	int ret = -EADDRINUSE, port = snum, l3mdev;
 512	struct inet_bind_hashbucket *head, *head2;
 513	struct inet_bind2_bucket *tb2 = NULL;
 514	struct inet_bind_bucket *tb = NULL;
 515	bool head2_lock_acquired = false;
 516	struct net *net = sock_net(sk);
 517
 518	l3mdev = inet_sk_bound_l3mdev(sk);
 519
 520	if (!port) {
 521		head = inet_csk_find_open_port(sk, &tb, &tb2, &head2, &port);
 522		if (!head)
 523			return ret;
 524
 525		head2_lock_acquired = true;
 526
 527		if (tb && tb2)
 528			goto success;
 529		found_port = true;
 530	} else {
 531		head = &hinfo->bhash[inet_bhashfn(net, port,
 532						  hinfo->bhash_size)];
 533		spin_lock_bh(&head->lock);
 534		inet_bind_bucket_for_each(tb, &head->chain)
 535			if (inet_bind_bucket_match(tb, net, port, l3mdev))
 536				break;
 537	}
 538
 539	if (!tb) {
 540		tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, net,
 541					     head, port, l3mdev);
 542		if (!tb)
 543			goto fail_unlock;
 544		bhash_created = true;
 545	}
 546
 547	if (!found_port) {
 548		if (!hlist_empty(&tb->bhash2)) {
 549			if (sk->sk_reuse == SK_FORCE_REUSE ||
 550			    (tb->fastreuse > 0 && reuse) ||
 551			    sk_reuseport_match(tb, sk))
 552				check_bind_conflict = false;
 553		}
 554
 555		if (check_bind_conflict && inet_use_bhash2_on_bind(sk)) {
 556			if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, true, true))
 557				goto fail_unlock;
 558		}
 559
 560		head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
 561		spin_lock(&head2->lock);
 562		head2_lock_acquired = true;
 563		tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
 564	}
 565
 566	if (!tb2) {
 567		tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep,
 568					       net, head2, tb, sk);
 569		if (!tb2)
 570			goto fail_unlock;
 571		bhash2_created = true;
 572	}
 573
 574	if (!found_port && check_bind_conflict) {
 575		if (inet_csk_bind_conflict(sk, tb, tb2, true, true))
 576			goto fail_unlock;
 577	}
 578
 579success:
 580	inet_csk_update_fastreuse(tb, sk);
 581
 582	if (!inet_csk(sk)->icsk_bind_hash)
 583		inet_bind_hash(sk, tb, tb2, port);
 584	WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 585	WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);
 586	ret = 0;
 587
 588fail_unlock:
 589	if (ret) {
 590		if (bhash2_created)
 591			inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep, tb2);
 592		if (bhash_created)
 593			inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
 594	}
 595	if (head2_lock_acquired)
 596		spin_unlock(&head2->lock);
 597	spin_unlock_bh(&head->lock);
 598	return ret;
 599}
 600EXPORT_SYMBOL_GPL(inet_csk_get_port);
 601
 602/*
 603 * Wait for an incoming connection, avoid race conditions. This must be called
 604 * with the socket locked.
 605 */
 606static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 607{
 608	struct inet_connection_sock *icsk = inet_csk(sk);
 609	DEFINE_WAIT(wait);
 610	int err;
 611
 612	/*
 613	 * True wake-one mechanism for incoming connections: only
 614	 * one process gets woken up, not the 'whole herd'.
 615	 * Since we do not 'race & poll' for established sockets
 616	 * anymore, the common case will execute the loop only once.
 617	 *
 618	 * Subtle issue: "add_wait_queue_exclusive()" will be added
 619	 * after any current non-exclusive waiters, and we know that
 620	 * it will always _stay_ after any new non-exclusive waiters
 621	 * because all non-exclusive waiters are added at the
 622	 * beginning of the wait-queue. As such, it's ok to "drop"
 623	 * our exclusiveness temporarily when we get woken up without
 624	 * having to remove and re-insert us on the wait queue.
 625	 */
 626	for (;;) {
 627		prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 628					  TASK_INTERRUPTIBLE);
 629		release_sock(sk);
 630		if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 631			timeo = schedule_timeout(timeo);
 632		sched_annotate_sleep();
 633		lock_sock(sk);
 634		err = 0;
 635		if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 636			break;
 637		err = -EINVAL;
 638		if (sk->sk_state != TCP_LISTEN)
 639			break;
 640		err = sock_intr_errno(timeo);
 641		if (signal_pending(current))
 642			break;
 643		err = -EAGAIN;
 644		if (!timeo)
 645			break;
 646	}
 647	finish_wait(sk_sleep(sk), &wait);
 648	return err;
 649}
 650
 651/*
 652 * This will accept the next outstanding connection.
 653 */
 654struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
 655{
 656	struct inet_connection_sock *icsk = inet_csk(sk);
 657	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 658	struct request_sock *req;
 659	struct sock *newsk;
 660	int error;
 661
 662	lock_sock(sk);
 663
 664	/* We need to make sure that this socket is listening,
 665	 * and that it has something pending.
 666	 */
 667	error = -EINVAL;
 668	if (sk->sk_state != TCP_LISTEN)
 669		goto out_err;
 670
 671	/* Find already established connection */
 672	if (reqsk_queue_empty(queue)) {
 673		long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 674
 675		/* If this is a non blocking socket don't sleep */
 676		error = -EAGAIN;
 677		if (!timeo)
 678			goto out_err;
 679
 680		error = inet_csk_wait_for_connect(sk, timeo);
 681		if (error)
 682			goto out_err;
 683	}
 684	req = reqsk_queue_remove(queue, sk);
 685	newsk = req->sk;
 686
 687	if (sk->sk_protocol == IPPROTO_TCP &&
 688	    tcp_rsk(req)->tfo_listener) {
 689		spin_lock_bh(&queue->fastopenq.lock);
 690		if (tcp_rsk(req)->tfo_listener) {
 691			/* We are still waiting for the final ACK from 3WHS
 692			 * so can't free req now. Instead, we set req->sk to
 693			 * NULL to signify that the child socket is taken
 694			 * so reqsk_fastopen_remove() will free the req
 695			 * when 3WHS finishes (or is aborted).
 696			 */
 697			req->sk = NULL;
 698			req = NULL;
 699		}
 700		spin_unlock_bh(&queue->fastopenq.lock);
 701	}
 702
 703out:
 704	release_sock(sk);
 705	if (newsk && mem_cgroup_sockets_enabled) {
 706		int amt = 0;
 707
 708		/* atomically get the memory usage, set and charge the
 709		 * newsk->sk_memcg.
 710		 */
 711		lock_sock(newsk);
 712
 713		mem_cgroup_sk_alloc(newsk);
 714		if (newsk->sk_memcg) {
 715			/* The socket has not been accepted yet, no need
 716			 * to look at newsk->sk_wmem_queued.
 717			 */
 718			amt = sk_mem_pages(newsk->sk_forward_alloc +
 719					   atomic_read(&newsk->sk_rmem_alloc));
 720		}
 721
 722		if (amt)
 723			mem_cgroup_charge_skmem(newsk->sk_memcg, amt,
 724						GFP_KERNEL | __GFP_NOFAIL);
 725
 726		release_sock(newsk);
 727	}
 728	if (req)
 729		reqsk_put(req);
 730
 731	if (newsk)
 732		inet_init_csk_locks(newsk);
 733
 734	return newsk;
 735out_err:
 736	newsk = NULL;
 737	req = NULL;
 738	*err = error;
 739	goto out;
 740}
 741EXPORT_SYMBOL(inet_csk_accept);
 742
 743/*
 744 * Using different timers for retransmit, delayed acks and probes
 745 * We may wish use just one timer maintaining a list of expire jiffies
 746 * to optimize.
 747 */
 748void inet_csk_init_xmit_timers(struct sock *sk,
 749			       void (*retransmit_handler)(struct timer_list *t),
 750			       void (*delack_handler)(struct timer_list *t),
 751			       void (*keepalive_handler)(struct timer_list *t))
 752{
 753	struct inet_connection_sock *icsk = inet_csk(sk);
 754
 755	timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
 756	timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
 757	timer_setup(&sk->sk_timer, keepalive_handler, 0);
 758	icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 759}
 760EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 761
 762void inet_csk_clear_xmit_timers(struct sock *sk)
 763{
 764	struct inet_connection_sock *icsk = inet_csk(sk);
 765
 766	icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 767
 768	sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 769	sk_stop_timer(sk, &icsk->icsk_delack_timer);
 770	sk_stop_timer(sk, &sk->sk_timer);
 771}
 772EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 773
 774void inet_csk_delete_keepalive_timer(struct sock *sk)
 775{
 776	sk_stop_timer(sk, &sk->sk_timer);
 777}
 778EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 779
 780void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 781{
 782	sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 783}
 784EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 785
 786struct dst_entry *inet_csk_route_req(const struct sock *sk,
 787				     struct flowi4 *fl4,
 788				     const struct request_sock *req)
 789{
 790	const struct inet_request_sock *ireq = inet_rsk(req);
 791	struct net *net = read_pnet(&ireq->ireq_net);
 792	struct ip_options_rcu *opt;
 793	struct rtable *rt;
 794
 795	rcu_read_lock();
 796	opt = rcu_dereference(ireq->ireq_opt);
 797
 798	flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 799			   ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
 800			   sk->sk_protocol, inet_sk_flowi_flags(sk),
 801			   (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 802			   ireq->ir_loc_addr, ireq->ir_rmt_port,
 803			   htons(ireq->ir_num), sk->sk_uid);
 804	security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
 805	rt = ip_route_output_flow(net, fl4, sk);
 806	if (IS_ERR(rt))
 807		goto no_route;
 808	if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 809		goto route_err;
 810	rcu_read_unlock();
 811	return &rt->dst;
 812
 813route_err:
 814	ip_rt_put(rt);
 815no_route:
 816	rcu_read_unlock();
 817	__IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 818	return NULL;
 819}
 820EXPORT_SYMBOL_GPL(inet_csk_route_req);
 821
 822struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
 823					    struct sock *newsk,
 824					    const struct request_sock *req)
 825{
 826	const struct inet_request_sock *ireq = inet_rsk(req);
 827	struct net *net = read_pnet(&ireq->ireq_net);
 828	struct inet_sock *newinet = inet_sk(newsk);
 829	struct ip_options_rcu *opt;
 830	struct flowi4 *fl4;
 831	struct rtable *rt;
 832
 833	opt = rcu_dereference(ireq->ireq_opt);
 834	fl4 = &newinet->cork.fl.u.ip4;
 835
 836	flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 837			   ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
 838			   sk->sk_protocol, inet_sk_flowi_flags(sk),
 839			   (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 840			   ireq->ir_loc_addr, ireq->ir_rmt_port,
 841			   htons(ireq->ir_num), sk->sk_uid);
 842	security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
 843	rt = ip_route_output_flow(net, fl4, sk);
 844	if (IS_ERR(rt))
 845		goto no_route;
 846	if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 847		goto route_err;
 848	return &rt->dst;
 849
 850route_err:
 851	ip_rt_put(rt);
 852no_route:
 853	__IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 854	return NULL;
 855}
 856EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
 857
 
 
 
 
 
 
 858/* Decide when to expire the request and when to resend SYN-ACK */
 859static void syn_ack_recalc(struct request_sock *req,
 860			   const int max_syn_ack_retries,
 861			   const u8 rskq_defer_accept,
 862			   int *expire, int *resend)
 863{
 864	if (!rskq_defer_accept) {
 865		*expire = req->num_timeout >= max_syn_ack_retries;
 866		*resend = 1;
 867		return;
 868	}
 869	*expire = req->num_timeout >= max_syn_ack_retries &&
 870		  (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
 871	/* Do not resend while waiting for data after ACK,
 
 872	 * start to resend on end of deferring period to give
 873	 * last chance for data or ACK to create established socket.
 874	 */
 875	*resend = !inet_rsk(req)->acked ||
 876		  req->num_timeout >= rskq_defer_accept - 1;
 877}
 878
 879int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
 880{
 881	int err = req->rsk_ops->rtx_syn_ack(parent, req);
 882
 883	if (!err)
 884		req->num_retrans++;
 885	return err;
 886}
 887EXPORT_SYMBOL(inet_rtx_syn_ack);
 888
 889static struct request_sock *inet_reqsk_clone(struct request_sock *req,
 890					     struct sock *sk)
 891{
 892	struct sock *req_sk, *nreq_sk;
 893	struct request_sock *nreq;
 894
 895	nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
 896	if (!nreq) {
 897		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
 898
 899		/* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
 900		sock_put(sk);
 901		return NULL;
 902	}
 903
 904	req_sk = req_to_sk(req);
 905	nreq_sk = req_to_sk(nreq);
 906
 907	memcpy(nreq_sk, req_sk,
 908	       offsetof(struct sock, sk_dontcopy_begin));
 909	memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
 910	       req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end));
 911
 912	sk_node_init(&nreq_sk->sk_node);
 913	nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
 914#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
 915	nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
 916#endif
 917	nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
 918
 919	nreq->rsk_listener = sk;
 920
 921	/* We need not acquire fastopenq->lock
 922	 * because the child socket is locked in inet_csk_listen_stop().
 923	 */
 924	if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
 925		rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
 926
 927	return nreq;
 928}
 929
 930static void reqsk_queue_migrated(struct request_sock_queue *queue,
 931				 const struct request_sock *req)
 932{
 933	if (req->num_timeout == 0)
 934		atomic_inc(&queue->young);
 935	atomic_inc(&queue->qlen);
 936}
 937
 938static void reqsk_migrate_reset(struct request_sock *req)
 939{
 940	req->saved_syn = NULL;
 941#if IS_ENABLED(CONFIG_IPV6)
 942	inet_rsk(req)->ipv6_opt = NULL;
 943	inet_rsk(req)->pktopts = NULL;
 944#else
 945	inet_rsk(req)->ireq_opt = NULL;
 946#endif
 947}
 948
 949/* return true if req was found in the ehash table */
 950static bool reqsk_queue_unlink(struct request_sock *req)
 
 951{
 952	struct sock *sk = req_to_sk(req);
 953	bool found = false;
 954
 955	if (sk_hashed(sk)) {
 956		struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
 957		spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
 958
 959		spin_lock(lock);
 960		found = __sk_nulls_del_node_init_rcu(sk);
 961		spin_unlock(lock);
 962	}
 963	if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
 964		reqsk_put(req);
 965	return found;
 966}
 967
 968bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
 969{
 970	bool unlinked = reqsk_queue_unlink(req);
 971
 972	if (unlinked) {
 973		reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 974		reqsk_put(req);
 975	}
 976	return unlinked;
 977}
 978EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
 979
 980void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
 981{
 982	inet_csk_reqsk_queue_drop(sk, req);
 983	reqsk_put(req);
 984}
 985EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
 986
 987static void reqsk_timer_handler(struct timer_list *t)
 988{
 989	struct request_sock *req = from_timer(req, t, rsk_timer);
 990	struct request_sock *nreq = NULL, *oreq = req;
 991	struct sock *sk_listener = req->rsk_listener;
 992	struct inet_connection_sock *icsk;
 993	struct request_sock_queue *queue;
 994	struct net *net;
 995	int max_syn_ack_retries, qlen, expire = 0, resend = 0;
 996
 997	if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
 998		struct sock *nsk;
 999
1000		nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
1001		if (!nsk)
1002			goto drop;
1003
1004		nreq = inet_reqsk_clone(req, nsk);
1005		if (!nreq)
1006			goto drop;
1007
1008		/* The new timer for the cloned req can decrease the 2
1009		 * by calling inet_csk_reqsk_queue_drop_and_put(), so
1010		 * hold another count to prevent use-after-free and
1011		 * call reqsk_put() just before return.
1012		 */
1013		refcount_set(&nreq->rsk_refcnt, 2 + 1);
1014		timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1015		reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
1016
1017		req = nreq;
1018		sk_listener = nsk;
1019	}
1020
1021	icsk = inet_csk(sk_listener);
1022	net = sock_net(sk_listener);
1023	max_syn_ack_retries = READ_ONCE(icsk->icsk_syn_retries) ? :
1024		READ_ONCE(net->ipv4.sysctl_tcp_synack_retries);
1025	/* Normally all the openreqs are young and become mature
1026	 * (i.e. converted to established socket) for first timeout.
1027	 * If synack was not acknowledged for 1 second, it means
1028	 * one of the following things: synack was lost, ack was lost,
1029	 * rtt is high or nobody planned to ack (i.e. synflood).
1030	 * When server is a bit loaded, queue is populated with old
1031	 * open requests, reducing effective size of queue.
1032	 * When server is well loaded, queue size reduces to zero
1033	 * after several minutes of work. It is not synflood,
1034	 * it is normal operation. The solution is pruning
1035	 * too old entries overriding normal timeout, when
1036	 * situation becomes dangerous.
1037	 *
1038	 * Essentially, we reserve half of room for young
1039	 * embrions; and abort old ones without pity, if old
1040	 * ones are about to clog our table.
1041	 */
1042	queue = &icsk->icsk_accept_queue;
1043	qlen = reqsk_queue_len(queue);
1044	if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
1045		int young = reqsk_queue_len_young(queue) << 1;
1046
1047		while (max_syn_ack_retries > 2) {
1048			if (qlen < young)
1049				break;
1050			max_syn_ack_retries--;
1051			young <<= 1;
1052		}
1053	}
1054	syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
 
 
 
1055		       &expire, &resend);
1056	req->rsk_ops->syn_ack_timeout(req);
1057	if (!expire &&
1058	    (!resend ||
1059	     !inet_rtx_syn_ack(sk_listener, req) ||
1060	     inet_rsk(req)->acked)) {
 
 
1061		if (req->num_timeout++ == 0)
1062			atomic_dec(&queue->young);
1063		mod_timer(&req->rsk_timer, jiffies + reqsk_timeout(req, TCP_RTO_MAX));
1064
1065		if (!nreq)
1066			return;
1067
1068		if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
1069			/* delete timer */
1070			inet_csk_reqsk_queue_drop(sk_listener, nreq);
1071			goto no_ownership;
1072		}
1073
1074		__NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
1075		reqsk_migrate_reset(oreq);
1076		reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
1077		reqsk_put(oreq);
1078
1079		reqsk_put(nreq);
1080		return;
1081	}
1082
1083	/* Even if we can clone the req, we may need not retransmit any more
1084	 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
1085	 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
1086	 */
1087	if (nreq) {
1088		__NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
1089no_ownership:
1090		reqsk_migrate_reset(nreq);
1091		reqsk_queue_removed(queue, nreq);
1092		__reqsk_free(nreq);
1093	}
1094
1095drop:
1096	inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq);
1097}
1098
1099static void reqsk_queue_hash_req(struct request_sock *req,
1100				 unsigned long timeout)
1101{
 
 
 
 
1102	timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1103	mod_timer(&req->rsk_timer, jiffies + timeout);
1104
1105	inet_ehash_insert(req_to_sk(req), NULL, NULL);
1106	/* before letting lookups find us, make sure all req fields
1107	 * are committed to memory and refcnt initialized.
1108	 */
1109	smp_wmb();
1110	refcount_set(&req->rsk_refcnt, 2 + 1);
1111}
1112
1113void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
1114				   unsigned long timeout)
1115{
1116	reqsk_queue_hash_req(req, timeout);
1117	inet_csk_reqsk_queue_added(sk);
1118}
1119EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
1120
1121static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
1122			   const gfp_t priority)
1123{
1124	struct inet_connection_sock *icsk = inet_csk(newsk);
1125
1126	if (!icsk->icsk_ulp_ops)
1127		return;
1128
1129	icsk->icsk_ulp_ops->clone(req, newsk, priority);
1130}
1131
1132/**
1133 *	inet_csk_clone_lock - clone an inet socket, and lock its clone
1134 *	@sk: the socket to clone
1135 *	@req: request_sock
1136 *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1137 *
1138 *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1139 */
1140struct sock *inet_csk_clone_lock(const struct sock *sk,
1141				 const struct request_sock *req,
1142				 const gfp_t priority)
1143{
1144	struct sock *newsk = sk_clone_lock(sk, priority);
1145
1146	if (newsk) {
1147		struct inet_connection_sock *newicsk = inet_csk(newsk);
1148
1149		inet_sk_set_state(newsk, TCP_SYN_RECV);
1150		newicsk->icsk_bind_hash = NULL;
1151		newicsk->icsk_bind2_hash = NULL;
1152
1153		inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
1154		inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
1155		inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
1156
1157		/* listeners have SOCK_RCU_FREE, not the children */
1158		sock_reset_flag(newsk, SOCK_RCU_FREE);
1159
1160		inet_sk(newsk)->mc_list = NULL;
1161
1162		newsk->sk_mark = inet_rsk(req)->ir_mark;
1163		atomic64_set(&newsk->sk_cookie,
1164			     atomic64_read(&inet_rsk(req)->ir_cookie));
1165
1166		newicsk->icsk_retransmits = 0;
1167		newicsk->icsk_backoff	  = 0;
1168		newicsk->icsk_probes_out  = 0;
1169		newicsk->icsk_probes_tstamp = 0;
1170
1171		/* Deinitialize accept_queue to trap illegal accesses. */
1172		memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
1173
1174		inet_clone_ulp(req, newsk, priority);
1175
1176		security_inet_csk_clone(newsk, req);
1177	}
1178	return newsk;
1179}
1180EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
1181
1182/*
1183 * At this point, there should be no process reference to this
1184 * socket, and thus no user references at all.  Therefore we
1185 * can assume the socket waitqueue is inactive and nobody will
1186 * try to jump onto it.
1187 */
1188void inet_csk_destroy_sock(struct sock *sk)
1189{
1190	WARN_ON(sk->sk_state != TCP_CLOSE);
1191	WARN_ON(!sock_flag(sk, SOCK_DEAD));
1192
1193	/* It cannot be in hash table! */
1194	WARN_ON(!sk_unhashed(sk));
1195
1196	/* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1197	WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1198
1199	sk->sk_prot->destroy(sk);
1200
1201	sk_stream_kill_queues(sk);
1202
1203	xfrm_sk_free_policy(sk);
1204
1205	this_cpu_dec(*sk->sk_prot->orphan_count);
 
 
1206
1207	sock_put(sk);
1208}
1209EXPORT_SYMBOL(inet_csk_destroy_sock);
1210
1211/* This function allows to force a closure of a socket after the call to
1212 * tcp/dccp_create_openreq_child().
1213 */
1214void inet_csk_prepare_forced_close(struct sock *sk)
1215	__releases(&sk->sk_lock.slock)
1216{
1217	/* sk_clone_lock locked the socket and set refcnt to 2 */
1218	bh_unlock_sock(sk);
1219	sock_put(sk);
1220	inet_csk_prepare_for_destroy_sock(sk);
 
 
 
1221	inet_sk(sk)->inet_num = 0;
1222}
1223EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1224
1225static int inet_ulp_can_listen(const struct sock *sk)
1226{
1227	const struct inet_connection_sock *icsk = inet_csk(sk);
1228
1229	if (icsk->icsk_ulp_ops && !icsk->icsk_ulp_ops->clone)
1230		return -EINVAL;
1231
1232	return 0;
1233}
1234
1235int inet_csk_listen_start(struct sock *sk)
1236{
1237	struct inet_connection_sock *icsk = inet_csk(sk);
1238	struct inet_sock *inet = inet_sk(sk);
1239	int err;
1240
1241	err = inet_ulp_can_listen(sk);
1242	if (unlikely(err))
1243		return err;
1244
1245	reqsk_queue_alloc(&icsk->icsk_accept_queue);
1246
 
1247	sk->sk_ack_backlog = 0;
1248	inet_csk_delack_init(sk);
1249
1250	/* There is race window here: we announce ourselves listening,
1251	 * but this transition is still not validated by get_port().
1252	 * It is OK, because this socket enters to hash table only
1253	 * after validation is complete.
1254	 */
1255	inet_sk_state_store(sk, TCP_LISTEN);
1256	err = sk->sk_prot->get_port(sk, inet->inet_num);
1257	if (!err) {
1258		inet->inet_sport = htons(inet->inet_num);
1259
1260		sk_dst_reset(sk);
1261		err = sk->sk_prot->hash(sk);
1262
1263		if (likely(!err))
1264			return 0;
1265	}
1266
1267	inet_sk_set_state(sk, TCP_CLOSE);
1268	return err;
1269}
1270EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1271
1272static void inet_child_forget(struct sock *sk, struct request_sock *req,
1273			      struct sock *child)
1274{
1275	sk->sk_prot->disconnect(child, O_NONBLOCK);
1276
1277	sock_orphan(child);
1278
1279	this_cpu_inc(*sk->sk_prot->orphan_count);
1280
1281	if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1282		BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1283		BUG_ON(sk != req->rsk_listener);
1284
1285		/* Paranoid, to prevent race condition if
1286		 * an inbound pkt destined for child is
1287		 * blocked by sock lock in tcp_v4_rcv().
1288		 * Also to satisfy an assertion in
1289		 * tcp_v4_destroy_sock().
1290		 */
1291		RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1292	}
1293	inet_csk_destroy_sock(child);
1294}
1295
1296struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1297				      struct request_sock *req,
1298				      struct sock *child)
1299{
1300	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1301
1302	spin_lock(&queue->rskq_lock);
1303	if (unlikely(sk->sk_state != TCP_LISTEN)) {
1304		inet_child_forget(sk, req, child);
1305		child = NULL;
1306	} else {
1307		req->sk = child;
1308		req->dl_next = NULL;
1309		if (queue->rskq_accept_head == NULL)
1310			WRITE_ONCE(queue->rskq_accept_head, req);
1311		else
1312			queue->rskq_accept_tail->dl_next = req;
1313		queue->rskq_accept_tail = req;
1314		sk_acceptq_added(sk);
1315	}
1316	spin_unlock(&queue->rskq_lock);
1317	return child;
1318}
1319EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1320
1321struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1322					 struct request_sock *req, bool own_req)
1323{
1324	if (own_req) {
1325		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1326		reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1327
1328		if (sk != req->rsk_listener) {
1329			/* another listening sk has been selected,
1330			 * migrate the req to it.
1331			 */
1332			struct request_sock *nreq;
1333
1334			/* hold a refcnt for the nreq->rsk_listener
1335			 * which is assigned in inet_reqsk_clone()
1336			 */
1337			sock_hold(sk);
1338			nreq = inet_reqsk_clone(req, sk);
1339			if (!nreq) {
1340				inet_child_forget(sk, req, child);
1341				goto child_put;
1342			}
1343
1344			refcount_set(&nreq->rsk_refcnt, 1);
1345			if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1346				__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1347				reqsk_migrate_reset(req);
1348				reqsk_put(req);
1349				return child;
1350			}
1351
1352			__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1353			reqsk_migrate_reset(nreq);
1354			__reqsk_free(nreq);
1355		} else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1356			return child;
1357		}
1358	}
1359	/* Too bad, another child took ownership of the request, undo. */
1360child_put:
1361	bh_unlock_sock(child);
1362	sock_put(child);
1363	return NULL;
1364}
1365EXPORT_SYMBOL(inet_csk_complete_hashdance);
1366
1367/*
1368 *	This routine closes sockets which have been at least partially
1369 *	opened, but not yet accepted.
1370 */
1371void inet_csk_listen_stop(struct sock *sk)
1372{
1373	struct inet_connection_sock *icsk = inet_csk(sk);
1374	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1375	struct request_sock *next, *req;
1376
1377	/* Following specs, it would be better either to send FIN
1378	 * (and enter FIN-WAIT-1, it is normal close)
1379	 * or to send active reset (abort).
1380	 * Certainly, it is pretty dangerous while synflood, but it is
1381	 * bad justification for our negligence 8)
1382	 * To be honest, we are not able to make either
1383	 * of the variants now.			--ANK
1384	 */
1385	while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1386		struct sock *child = req->sk, *nsk;
1387		struct request_sock *nreq;
1388
1389		local_bh_disable();
1390		bh_lock_sock(child);
1391		WARN_ON(sock_owned_by_user(child));
1392		sock_hold(child);
1393
1394		nsk = reuseport_migrate_sock(sk, child, NULL);
1395		if (nsk) {
1396			nreq = inet_reqsk_clone(req, nsk);
1397			if (nreq) {
1398				refcount_set(&nreq->rsk_refcnt, 1);
1399
1400				if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1401					__NET_INC_STATS(sock_net(nsk),
1402							LINUX_MIB_TCPMIGRATEREQSUCCESS);
1403					reqsk_migrate_reset(req);
1404				} else {
1405					__NET_INC_STATS(sock_net(nsk),
1406							LINUX_MIB_TCPMIGRATEREQFAILURE);
1407					reqsk_migrate_reset(nreq);
1408					__reqsk_free(nreq);
1409				}
1410
1411				/* inet_csk_reqsk_queue_add() has already
1412				 * called inet_child_forget() on failure case.
1413				 */
1414				goto skip_child_forget;
1415			}
1416		}
1417
1418		inet_child_forget(sk, req, child);
1419skip_child_forget:
1420		reqsk_put(req);
1421		bh_unlock_sock(child);
1422		local_bh_enable();
1423		sock_put(child);
1424
1425		cond_resched();
1426	}
1427	if (queue->fastopenq.rskq_rst_head) {
1428		/* Free all the reqs queued in rskq_rst_head. */
1429		spin_lock_bh(&queue->fastopenq.lock);
1430		req = queue->fastopenq.rskq_rst_head;
1431		queue->fastopenq.rskq_rst_head = NULL;
1432		spin_unlock_bh(&queue->fastopenq.lock);
1433		while (req != NULL) {
1434			next = req->dl_next;
1435			reqsk_put(req);
1436			req = next;
1437		}
1438	}
1439	WARN_ON_ONCE(sk->sk_ack_backlog);
1440}
1441EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1442
1443void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1444{
1445	struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1446	const struct inet_sock *inet = inet_sk(sk);
1447
1448	sin->sin_family		= AF_INET;
1449	sin->sin_addr.s_addr	= inet->inet_daddr;
1450	sin->sin_port		= inet->inet_dport;
1451}
1452EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1453
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1454static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1455{
1456	const struct inet_sock *inet = inet_sk(sk);
1457	const struct ip_options_rcu *inet_opt;
1458	__be32 daddr = inet->inet_daddr;
1459	struct flowi4 *fl4;
1460	struct rtable *rt;
1461
1462	rcu_read_lock();
1463	inet_opt = rcu_dereference(inet->inet_opt);
1464	if (inet_opt && inet_opt->opt.srr)
1465		daddr = inet_opt->opt.faddr;
1466	fl4 = &fl->u.ip4;
1467	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1468				   inet->inet_saddr, inet->inet_dport,
1469				   inet->inet_sport, sk->sk_protocol,
1470				   RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1471	if (IS_ERR(rt))
1472		rt = NULL;
1473	if (rt)
1474		sk_setup_caps(sk, &rt->dst);
1475	rcu_read_unlock();
1476
1477	return &rt->dst;
1478}
1479
1480struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1481{
1482	struct dst_entry *dst = __sk_dst_check(sk, 0);
1483	struct inet_sock *inet = inet_sk(sk);
1484
1485	if (!dst) {
1486		dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1487		if (!dst)
1488			goto out;
1489	}
1490	dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1491
1492	dst = __sk_dst_check(sk, 0);
1493	if (!dst)
1494		dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1495out:
1496	return dst;
1497}
1498EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);