Linux Audio

Check our new training course

Loading...
v5.14.15
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * NET4:	Implementation of BSD Unix domain sockets.
   4 *
   5 * Authors:	Alan Cox, <alan@lxorguk.ukuu.org.uk>
   6 *
   7 * Fixes:
   8 *		Linus Torvalds	:	Assorted bug cures.
   9 *		Niibe Yutaka	:	async I/O support.
  10 *		Carsten Paeth	:	PF_UNIX check, address fixes.
  11 *		Alan Cox	:	Limit size of allocated blocks.
  12 *		Alan Cox	:	Fixed the stupid socketpair bug.
  13 *		Alan Cox	:	BSD compatibility fine tuning.
  14 *		Alan Cox	:	Fixed a bug in connect when interrupted.
  15 *		Alan Cox	:	Sorted out a proper draft version of
  16 *					file descriptor passing hacked up from
  17 *					Mike Shaver's work.
  18 *		Marty Leisner	:	Fixes to fd passing
  19 *		Nick Nevin	:	recvmsg bugfix.
  20 *		Alan Cox	:	Started proper garbage collector
  21 *		Heiko EiBfeldt	:	Missing verify_area check
  22 *		Alan Cox	:	Started POSIXisms
  23 *		Andreas Schwab	:	Replace inode by dentry for proper
  24 *					reference counting
  25 *		Kirk Petersen	:	Made this a module
  26 *	    Christoph Rohland	:	Elegant non-blocking accept/connect algorithm.
  27 *					Lots of bug fixes.
  28 *	     Alexey Kuznetosv	:	Repaired (I hope) bugs introduces
  29 *					by above two patches.
  30 *	     Andrea Arcangeli	:	If possible we block in connect(2)
  31 *					if the max backlog of the listen socket
  32 *					is been reached. This won't break
  33 *					old apps and it will avoid huge amount
  34 *					of socks hashed (this for unix_gc()
  35 *					performances reasons).
  36 *					Security fix that limits the max
  37 *					number of socks to 2*max_files and
  38 *					the number of skb queueable in the
  39 *					dgram receiver.
  40 *		Artur Skawina   :	Hash function optimizations
  41 *	     Alexey Kuznetsov   :	Full scale SMP. Lot of bugs are introduced 8)
  42 *	      Malcolm Beattie   :	Set peercred for socketpair
  43 *	     Michal Ostrowski   :       Module initialization cleanup.
  44 *	     Arnaldo C. Melo	:	Remove MOD_{INC,DEC}_USE_COUNT,
  45 *	     				the core infrastructure is doing that
  46 *	     				for all net proto families now (2.5.69+)
  47 *
  48 * Known differences from reference BSD that was tested:
  49 *
  50 *	[TO FIX]
  51 *	ECONNREFUSED is not returned from one end of a connected() socket to the
  52 *		other the moment one end closes.
  53 *	fstat() doesn't return st_dev=0, and give the blksize as high water mark
  54 *		and a fake inode identifier (nor the BSD first socket fstat twice bug).
  55 *	[NOT TO FIX]
  56 *	accept() returns a path name even if the connecting socket has closed
  57 *		in the meantime (BSD loses the path and gives up).
  58 *	accept() returns 0 length path for an unbound connector. BSD returns 16
  59 *		and a null first byte in the path (but not for gethost/peername - BSD bug ??)
  60 *	socketpair(...SOCK_RAW..) doesn't panic the kernel.
  61 *	BSD af_unix apparently has connect forgetting to block properly.
  62 *		(need to check this with the POSIX spec in detail)
  63 *
  64 * Differences from 2.0.0-11-... (ANK)
  65 *	Bug fixes and improvements.
  66 *		- client shutdown killed server socket.
  67 *		- removed all useless cli/sti pairs.
  68 *
  69 *	Semantic changes/extensions.
  70 *		- generic control message passing.
  71 *		- SCM_CREDENTIALS control message.
  72 *		- "Abstract" (not FS based) socket bindings.
  73 *		  Abstract names are sequences of bytes (not zero terminated)
  74 *		  started by 0, so that this name space does not intersect
  75 *		  with BSD names.
  76 */
  77
  78#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  79
  80#include <linux/module.h>
  81#include <linux/kernel.h>
  82#include <linux/signal.h>
  83#include <linux/sched/signal.h>
  84#include <linux/errno.h>
  85#include <linux/string.h>
  86#include <linux/stat.h>
  87#include <linux/dcache.h>
  88#include <linux/namei.h>
  89#include <linux/socket.h>
  90#include <linux/un.h>
  91#include <linux/fcntl.h>
  92#include <linux/termios.h>
  93#include <linux/sockios.h>
  94#include <linux/net.h>
  95#include <linux/in.h>
  96#include <linux/fs.h>
  97#include <linux/slab.h>
  98#include <linux/uaccess.h>
  99#include <linux/skbuff.h>
 100#include <linux/netdevice.h>
 101#include <net/net_namespace.h>
 102#include <net/sock.h>
 103#include <net/tcp_states.h>
 104#include <net/af_unix.h>
 105#include <linux/proc_fs.h>
 106#include <linux/seq_file.h>
 107#include <net/scm.h>
 108#include <linux/init.h>
 109#include <linux/poll.h>
 110#include <linux/rtnetlink.h>
 111#include <linux/mount.h>
 112#include <net/checksum.h>
 113#include <linux/security.h>
 114#include <linux/freezer.h>
 115#include <linux/file.h>
 116
 117#include "scm.h"
 118
 119struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
 120EXPORT_SYMBOL_GPL(unix_socket_table);
 121DEFINE_SPINLOCK(unix_table_lock);
 122EXPORT_SYMBOL_GPL(unix_table_lock);
 123static atomic_long_t unix_nr_socks;
 124
 125
 126static struct hlist_head *unix_sockets_unbound(void *addr)
 127{
 128	unsigned long hash = (unsigned long)addr;
 129
 130	hash ^= hash >> 16;
 131	hash ^= hash >> 8;
 132	hash %= UNIX_HASH_SIZE;
 133	return &unix_socket_table[UNIX_HASH_SIZE + hash];
 134}
 135
 136#define UNIX_ABSTRACT(sk)	(unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
 137
 138#ifdef CONFIG_SECURITY_NETWORK
 139static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 140{
 141	UNIXCB(skb).secid = scm->secid;
 142}
 143
 144static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 145{
 146	scm->secid = UNIXCB(skb).secid;
 147}
 148
 149static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
 150{
 151	return (scm->secid == UNIXCB(skb).secid);
 152}
 153#else
 154static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 155{ }
 156
 157static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 158{ }
 159
 160static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
 161{
 162	return true;
 163}
 164#endif /* CONFIG_SECURITY_NETWORK */
 165
 166/*
 167 *  SMP locking strategy:
 168 *    hash table is protected with spinlock unix_table_lock
 169 *    each socket state is protected by separate spin lock.
 170 */
 171
 172static inline unsigned int unix_hash_fold(__wsum n)
 173{
 174	unsigned int hash = (__force unsigned int)csum_fold(n);
 175
 176	hash ^= hash>>8;
 177	return hash&(UNIX_HASH_SIZE-1);
 178}
 179
 180#define unix_peer(sk) (unix_sk(sk)->peer)
 181
 182static inline int unix_our_peer(struct sock *sk, struct sock *osk)
 183{
 184	return unix_peer(osk) == sk;
 185}
 186
 187static inline int unix_may_send(struct sock *sk, struct sock *osk)
 188{
 189	return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
 190}
 191
 192static inline int unix_recvq_full(const struct sock *sk)
 193{
 194	return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
 195}
 196
 197static inline int unix_recvq_full_lockless(const struct sock *sk)
 198{
 199	return skb_queue_len_lockless(&sk->sk_receive_queue) >
 200		READ_ONCE(sk->sk_max_ack_backlog);
 201}
 202
 203struct sock *unix_peer_get(struct sock *s)
 204{
 205	struct sock *peer;
 206
 207	unix_state_lock(s);
 208	peer = unix_peer(s);
 209	if (peer)
 210		sock_hold(peer);
 211	unix_state_unlock(s);
 212	return peer;
 213}
 214EXPORT_SYMBOL_GPL(unix_peer_get);
 215
 216static inline void unix_release_addr(struct unix_address *addr)
 217{
 218	if (refcount_dec_and_test(&addr->refcnt))
 219		kfree(addr);
 220}
 221
 222/*
 223 *	Check unix socket name:
 224 *		- should be not zero length.
 225 *	        - if started by not zero, should be NULL terminated (FS object)
 226 *		- if started by zero, it is abstract name.
 227 */
 228
 229static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
 230{
 231	*hashp = 0;
 232
 233	if (len <= sizeof(short) || len > sizeof(*sunaddr))
 234		return -EINVAL;
 235	if (!sunaddr || sunaddr->sun_family != AF_UNIX)
 236		return -EINVAL;
 237	if (sunaddr->sun_path[0]) {
 238		/*
 239		 * This may look like an off by one error but it is a bit more
 240		 * subtle. 108 is the longest valid AF_UNIX path for a binding.
 241		 * sun_path[108] doesn't as such exist.  However in kernel space
 242		 * we are guaranteed that it is a valid memory location in our
 243		 * kernel address buffer.
 244		 */
 245		((char *)sunaddr)[len] = 0;
 246		len = strlen(sunaddr->sun_path)+1+sizeof(short);
 247		return len;
 248	}
 249
 250	*hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
 251	return len;
 252}
 253
 254static void __unix_remove_socket(struct sock *sk)
 255{
 256	sk_del_node_init(sk);
 257}
 258
 259static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
 260{
 261	WARN_ON(!sk_unhashed(sk));
 262	sk_add_node(sk, list);
 263}
 264
 265static void __unix_set_addr(struct sock *sk, struct unix_address *addr,
 266			    unsigned hash)
 267{
 268	__unix_remove_socket(sk);
 269	smp_store_release(&unix_sk(sk)->addr, addr);
 270	__unix_insert_socket(&unix_socket_table[hash], sk);
 271}
 272
 273static inline void unix_remove_socket(struct sock *sk)
 274{
 275	spin_lock(&unix_table_lock);
 276	__unix_remove_socket(sk);
 277	spin_unlock(&unix_table_lock);
 278}
 279
 280static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
 281{
 282	spin_lock(&unix_table_lock);
 283	__unix_insert_socket(list, sk);
 284	spin_unlock(&unix_table_lock);
 285}
 286
 287static struct sock *__unix_find_socket_byname(struct net *net,
 288					      struct sockaddr_un *sunname,
 289					      int len, unsigned int hash)
 290{
 291	struct sock *s;
 292
 293	sk_for_each(s, &unix_socket_table[hash]) {
 294		struct unix_sock *u = unix_sk(s);
 295
 296		if (!net_eq(sock_net(s), net))
 297			continue;
 298
 299		if (u->addr->len == len &&
 300		    !memcmp(u->addr->name, sunname, len))
 301			return s;
 302	}
 303	return NULL;
 304}
 305
 306static inline struct sock *unix_find_socket_byname(struct net *net,
 307						   struct sockaddr_un *sunname,
 308						   int len, unsigned int hash)
 
 309{
 310	struct sock *s;
 311
 312	spin_lock(&unix_table_lock);
 313	s = __unix_find_socket_byname(net, sunname, len, hash);
 314	if (s)
 315		sock_hold(s);
 316	spin_unlock(&unix_table_lock);
 317	return s;
 318}
 319
 320static struct sock *unix_find_socket_byinode(struct inode *i)
 321{
 322	struct sock *s;
 323
 324	spin_lock(&unix_table_lock);
 325	sk_for_each(s,
 326		    &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
 327		struct dentry *dentry = unix_sk(s)->path.dentry;
 328
 329		if (dentry && d_backing_inode(dentry) == i) {
 330			sock_hold(s);
 331			goto found;
 332		}
 333	}
 334	s = NULL;
 335found:
 336	spin_unlock(&unix_table_lock);
 337	return s;
 338}
 339
 340/* Support code for asymmetrically connected dgram sockets
 341 *
 342 * If a datagram socket is connected to a socket not itself connected
 343 * to the first socket (eg, /dev/log), clients may only enqueue more
 344 * messages if the present receive queue of the server socket is not
 345 * "too large". This means there's a second writeability condition
 346 * poll and sendmsg need to test. The dgram recv code will do a wake
 347 * up on the peer_wait wait queue of a socket upon reception of a
 348 * datagram which needs to be propagated to sleeping would-be writers
 349 * since these might not have sent anything so far. This can't be
 350 * accomplished via poll_wait because the lifetime of the server
 351 * socket might be less than that of its clients if these break their
 352 * association with it or if the server socket is closed while clients
 353 * are still connected to it and there's no way to inform "a polling
 354 * implementation" that it should let go of a certain wait queue
 355 *
 356 * In order to propagate a wake up, a wait_queue_entry_t of the client
 357 * socket is enqueued on the peer_wait queue of the server socket
 358 * whose wake function does a wake_up on the ordinary client socket
 359 * wait queue. This connection is established whenever a write (or
 360 * poll for write) hit the flow control condition and broken when the
 361 * association to the server socket is dissolved or after a wake up
 362 * was relayed.
 363 */
 364
 365static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
 366				      void *key)
 367{
 368	struct unix_sock *u;
 369	wait_queue_head_t *u_sleep;
 370
 371	u = container_of(q, struct unix_sock, peer_wake);
 372
 373	__remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
 374			    q);
 375	u->peer_wake.private = NULL;
 376
 377	/* relaying can only happen while the wq still exists */
 378	u_sleep = sk_sleep(&u->sk);
 379	if (u_sleep)
 380		wake_up_interruptible_poll(u_sleep, key_to_poll(key));
 381
 382	return 0;
 383}
 384
 385static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
 386{
 387	struct unix_sock *u, *u_other;
 388	int rc;
 389
 390	u = unix_sk(sk);
 391	u_other = unix_sk(other);
 392	rc = 0;
 393	spin_lock(&u_other->peer_wait.lock);
 394
 395	if (!u->peer_wake.private) {
 396		u->peer_wake.private = other;
 397		__add_wait_queue(&u_other->peer_wait, &u->peer_wake);
 398
 399		rc = 1;
 400	}
 401
 402	spin_unlock(&u_other->peer_wait.lock);
 403	return rc;
 404}
 405
 406static void unix_dgram_peer_wake_disconnect(struct sock *sk,
 407					    struct sock *other)
 408{
 409	struct unix_sock *u, *u_other;
 410
 411	u = unix_sk(sk);
 412	u_other = unix_sk(other);
 413	spin_lock(&u_other->peer_wait.lock);
 414
 415	if (u->peer_wake.private == other) {
 416		__remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
 417		u->peer_wake.private = NULL;
 418	}
 419
 420	spin_unlock(&u_other->peer_wait.lock);
 421}
 422
 423static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
 424						   struct sock *other)
 425{
 426	unix_dgram_peer_wake_disconnect(sk, other);
 427	wake_up_interruptible_poll(sk_sleep(sk),
 428				   EPOLLOUT |
 429				   EPOLLWRNORM |
 430				   EPOLLWRBAND);
 431}
 432
 433/* preconditions:
 434 *	- unix_peer(sk) == other
 435 *	- association is stable
 436 */
 437static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
 438{
 439	int connected;
 440
 441	connected = unix_dgram_peer_wake_connect(sk, other);
 442
 443	/* If other is SOCK_DEAD, we want to make sure we signal
 444	 * POLLOUT, such that a subsequent write() can get a
 445	 * -ECONNREFUSED. Otherwise, if we haven't queued any skbs
 446	 * to other and its full, we will hang waiting for POLLOUT.
 447	 */
 448	if (unix_recvq_full(other) && !sock_flag(other, SOCK_DEAD))
 449		return 1;
 450
 451	if (connected)
 452		unix_dgram_peer_wake_disconnect(sk, other);
 453
 454	return 0;
 455}
 456
 457static int unix_writable(const struct sock *sk)
 458{
 459	return sk->sk_state != TCP_LISTEN &&
 460	       (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
 461}
 462
 463static void unix_write_space(struct sock *sk)
 464{
 465	struct socket_wq *wq;
 466
 467	rcu_read_lock();
 468	if (unix_writable(sk)) {
 469		wq = rcu_dereference(sk->sk_wq);
 470		if (skwq_has_sleeper(wq))
 471			wake_up_interruptible_sync_poll(&wq->wait,
 472				EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
 473		sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
 474	}
 475	rcu_read_unlock();
 476}
 477
 478/* When dgram socket disconnects (or changes its peer), we clear its receive
 479 * queue of packets arrived from previous peer. First, it allows to do
 480 * flow control based only on wmem_alloc; second, sk connected to peer
 481 * may receive messages only from that peer. */
 482static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
 483{
 484	if (!skb_queue_empty(&sk->sk_receive_queue)) {
 485		skb_queue_purge(&sk->sk_receive_queue);
 486		wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
 487
 488		/* If one link of bidirectional dgram pipe is disconnected,
 489		 * we signal error. Messages are lost. Do not make this,
 490		 * when peer was not connected to us.
 491		 */
 492		if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
 493			other->sk_err = ECONNRESET;
 494			sk_error_report(other);
 495		}
 496	}
 497}
 498
 499static void unix_sock_destructor(struct sock *sk)
 500{
 501	struct unix_sock *u = unix_sk(sk);
 502
 503	skb_queue_purge(&sk->sk_receive_queue);
 504
 505	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
 506	WARN_ON(!sk_unhashed(sk));
 507	WARN_ON(sk->sk_socket);
 508	if (!sock_flag(sk, SOCK_DEAD)) {
 509		pr_info("Attempt to release alive unix socket: %p\n", sk);
 510		return;
 511	}
 512
 513	if (u->addr)
 514		unix_release_addr(u->addr);
 515
 516	atomic_long_dec(&unix_nr_socks);
 517	local_bh_disable();
 518	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 519	local_bh_enable();
 520#ifdef UNIX_REFCNT_DEBUG
 521	pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
 522		atomic_long_read(&unix_nr_socks));
 523#endif
 524}
 525
 526static void unix_release_sock(struct sock *sk, int embrion)
 527{
 528	struct unix_sock *u = unix_sk(sk);
 529	struct path path;
 530	struct sock *skpair;
 531	struct sk_buff *skb;
 532	int state;
 533
 534	unix_remove_socket(sk);
 535
 536	/* Clear state */
 537	unix_state_lock(sk);
 538	sock_orphan(sk);
 539	sk->sk_shutdown = SHUTDOWN_MASK;
 540	path	     = u->path;
 541	u->path.dentry = NULL;
 542	u->path.mnt = NULL;
 543	state = sk->sk_state;
 544	sk->sk_state = TCP_CLOSE;
 545
 546	skpair = unix_peer(sk);
 547	unix_peer(sk) = NULL;
 548
 549	unix_state_unlock(sk);
 550
 551	wake_up_interruptible_all(&u->peer_wait);
 552
 
 
 553	if (skpair != NULL) {
 554		if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
 555			unix_state_lock(skpair);
 556			/* No more writes */
 557			skpair->sk_shutdown = SHUTDOWN_MASK;
 558			if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
 559				skpair->sk_err = ECONNRESET;
 560			unix_state_unlock(skpair);
 561			skpair->sk_state_change(skpair);
 562			sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
 563		}
 564
 565		unix_dgram_peer_wake_disconnect(sk, skpair);
 566		sock_put(skpair); /* It may now die */
 
 567	}
 568
 569	/* Try to flush out this socket. Throw out buffers at least */
 570
 571	while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 572		if (state == TCP_LISTEN)
 573			unix_release_sock(skb->sk, 1);
 574		/* passed fds are erased in the kfree_skb hook	      */
 575		UNIXCB(skb).consumed = skb->len;
 576		kfree_skb(skb);
 577	}
 578
 579	if (path.dentry)
 580		path_put(&path);
 581
 582	sock_put(sk);
 583
 584	/* ---- Socket is dead now and most probably destroyed ---- */
 585
 586	/*
 587	 * Fixme: BSD difference: In BSD all sockets connected to us get
 588	 *	  ECONNRESET and we die on the spot. In Linux we behave
 589	 *	  like files and pipes do and wait for the last
 590	 *	  dereference.
 591	 *
 592	 * Can't we simply set sock->err?
 593	 *
 594	 *	  What the above comment does talk about? --ANK(980817)
 595	 */
 596
 597	if (unix_tot_inflight)
 598		unix_gc();		/* Garbage collect fds */
 599}
 600
 601static void init_peercred(struct sock *sk)
 602{
 603	const struct cred *old_cred;
 604	struct pid *old_pid;
 605
 606	spin_lock(&sk->sk_peer_lock);
 607	old_pid = sk->sk_peer_pid;
 608	old_cred = sk->sk_peer_cred;
 609	sk->sk_peer_pid  = get_pid(task_tgid(current));
 610	sk->sk_peer_cred = get_current_cred();
 611	spin_unlock(&sk->sk_peer_lock);
 612
 613	put_pid(old_pid);
 614	put_cred(old_cred);
 615}
 616
 617static void copy_peercred(struct sock *sk, struct sock *peersk)
 618{
 619	const struct cred *old_cred;
 620	struct pid *old_pid;
 621
 622	if (sk < peersk) {
 623		spin_lock(&sk->sk_peer_lock);
 624		spin_lock_nested(&peersk->sk_peer_lock, SINGLE_DEPTH_NESTING);
 625	} else {
 626		spin_lock(&peersk->sk_peer_lock);
 627		spin_lock_nested(&sk->sk_peer_lock, SINGLE_DEPTH_NESTING);
 628	}
 629	old_pid = sk->sk_peer_pid;
 630	old_cred = sk->sk_peer_cred;
 631	sk->sk_peer_pid  = get_pid(peersk->sk_peer_pid);
 632	sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
 633
 634	spin_unlock(&sk->sk_peer_lock);
 635	spin_unlock(&peersk->sk_peer_lock);
 636
 637	put_pid(old_pid);
 638	put_cred(old_cred);
 639}
 640
 641static int unix_listen(struct socket *sock, int backlog)
 642{
 643	int err;
 644	struct sock *sk = sock->sk;
 645	struct unix_sock *u = unix_sk(sk);
 
 646
 647	err = -EOPNOTSUPP;
 648	if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
 649		goto out;	/* Only stream/seqpacket sockets accept */
 650	err = -EINVAL;
 651	if (!u->addr)
 652		goto out;	/* No listens on an unbound socket */
 653	unix_state_lock(sk);
 654	if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
 655		goto out_unlock;
 656	if (backlog > sk->sk_max_ack_backlog)
 657		wake_up_interruptible_all(&u->peer_wait);
 658	sk->sk_max_ack_backlog	= backlog;
 659	sk->sk_state		= TCP_LISTEN;
 660	/* set credentials so connect can copy them */
 661	init_peercred(sk);
 662	err = 0;
 663
 664out_unlock:
 665	unix_state_unlock(sk);
 
 666out:
 667	return err;
 668}
 669
 670static int unix_release(struct socket *);
 671static int unix_bind(struct socket *, struct sockaddr *, int);
 672static int unix_stream_connect(struct socket *, struct sockaddr *,
 673			       int addr_len, int flags);
 674static int unix_socketpair(struct socket *, struct socket *);
 675static int unix_accept(struct socket *, struct socket *, int, bool);
 676static int unix_getname(struct socket *, struct sockaddr *, int);
 677static __poll_t unix_poll(struct file *, struct socket *, poll_table *);
 678static __poll_t unix_dgram_poll(struct file *, struct socket *,
 679				    poll_table *);
 680static int unix_ioctl(struct socket *, unsigned int, unsigned long);
 681#ifdef CONFIG_COMPAT
 682static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg);
 683#endif
 684static int unix_shutdown(struct socket *, int);
 685static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
 686static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
 687static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
 688				    size_t size, int flags);
 689static ssize_t unix_stream_splice_read(struct socket *,  loff_t *ppos,
 690				       struct pipe_inode_info *, size_t size,
 691				       unsigned int flags);
 692static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
 693static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
 694static int unix_dgram_connect(struct socket *, struct sockaddr *,
 695			      int, int);
 696static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
 697static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
 698				  int);
 699
 700static int unix_set_peek_off(struct sock *sk, int val)
 701{
 702	struct unix_sock *u = unix_sk(sk);
 703
 704	if (mutex_lock_interruptible(&u->iolock))
 705		return -EINTR;
 706
 707	sk->sk_peek_off = val;
 708	mutex_unlock(&u->iolock);
 709
 710	return 0;
 711}
 712
 713#ifdef CONFIG_PROC_FS
 714static void unix_show_fdinfo(struct seq_file *m, struct socket *sock)
 715{
 716	struct sock *sk = sock->sk;
 717	struct unix_sock *u;
 718
 719	if (sk) {
 720		u = unix_sk(sock->sk);
 721		seq_printf(m, "scm_fds: %u\n",
 722			   atomic_read(&u->scm_stat.nr_fds));
 723	}
 724}
 725#else
 726#define unix_show_fdinfo NULL
 727#endif
 728
 729static const struct proto_ops unix_stream_ops = {
 730	.family =	PF_UNIX,
 731	.owner =	THIS_MODULE,
 732	.release =	unix_release,
 733	.bind =		unix_bind,
 734	.connect =	unix_stream_connect,
 735	.socketpair =	unix_socketpair,
 736	.accept =	unix_accept,
 737	.getname =	unix_getname,
 738	.poll =		unix_poll,
 739	.ioctl =	unix_ioctl,
 740#ifdef CONFIG_COMPAT
 741	.compat_ioctl =	unix_compat_ioctl,
 742#endif
 743	.listen =	unix_listen,
 744	.shutdown =	unix_shutdown,
 745	.sendmsg =	unix_stream_sendmsg,
 746	.recvmsg =	unix_stream_recvmsg,
 747	.mmap =		sock_no_mmap,
 748	.sendpage =	unix_stream_sendpage,
 749	.splice_read =	unix_stream_splice_read,
 750	.set_peek_off =	unix_set_peek_off,
 751	.show_fdinfo =	unix_show_fdinfo,
 752};
 753
 754static const struct proto_ops unix_dgram_ops = {
 755	.family =	PF_UNIX,
 756	.owner =	THIS_MODULE,
 757	.release =	unix_release,
 758	.bind =		unix_bind,
 759	.connect =	unix_dgram_connect,
 760	.socketpair =	unix_socketpair,
 761	.accept =	sock_no_accept,
 762	.getname =	unix_getname,
 763	.poll =		unix_dgram_poll,
 764	.ioctl =	unix_ioctl,
 765#ifdef CONFIG_COMPAT
 766	.compat_ioctl =	unix_compat_ioctl,
 767#endif
 768	.listen =	sock_no_listen,
 769	.shutdown =	unix_shutdown,
 770	.sendmsg =	unix_dgram_sendmsg,
 771	.recvmsg =	unix_dgram_recvmsg,
 772	.mmap =		sock_no_mmap,
 773	.sendpage =	sock_no_sendpage,
 774	.set_peek_off =	unix_set_peek_off,
 775	.show_fdinfo =	unix_show_fdinfo,
 776};
 777
 778static const struct proto_ops unix_seqpacket_ops = {
 779	.family =	PF_UNIX,
 780	.owner =	THIS_MODULE,
 781	.release =	unix_release,
 782	.bind =		unix_bind,
 783	.connect =	unix_stream_connect,
 784	.socketpair =	unix_socketpair,
 785	.accept =	unix_accept,
 786	.getname =	unix_getname,
 787	.poll =		unix_dgram_poll,
 788	.ioctl =	unix_ioctl,
 789#ifdef CONFIG_COMPAT
 790	.compat_ioctl =	unix_compat_ioctl,
 791#endif
 792	.listen =	unix_listen,
 793	.shutdown =	unix_shutdown,
 794	.sendmsg =	unix_seqpacket_sendmsg,
 795	.recvmsg =	unix_seqpacket_recvmsg,
 796	.mmap =		sock_no_mmap,
 797	.sendpage =	sock_no_sendpage,
 798	.set_peek_off =	unix_set_peek_off,
 799	.show_fdinfo =	unix_show_fdinfo,
 800};
 801
 802static struct proto unix_proto = {
 803	.name			= "UNIX",
 804	.owner			= THIS_MODULE,
 805	.obj_size		= sizeof(struct unix_sock),
 806};
 807
 808static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
 809{
 810	struct sock *sk = NULL;
 811	struct unix_sock *u;
 812
 813	atomic_long_inc(&unix_nr_socks);
 814	if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
 815		goto out;
 816
 817	sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
 818	if (!sk)
 819		goto out;
 820
 821	sock_init_data(sock, sk);
 822
 823	sk->sk_allocation	= GFP_KERNEL_ACCOUNT;
 824	sk->sk_write_space	= unix_write_space;
 825	sk->sk_max_ack_backlog	= net->unx.sysctl_max_dgram_qlen;
 826	sk->sk_destruct		= unix_sock_destructor;
 827	u	  = unix_sk(sk);
 828	u->path.dentry = NULL;
 829	u->path.mnt = NULL;
 830	spin_lock_init(&u->lock);
 831	atomic_long_set(&u->inflight, 0);
 832	INIT_LIST_HEAD(&u->link);
 833	mutex_init(&u->iolock); /* single task reading lock */
 834	mutex_init(&u->bindlock); /* single task binding lock */
 835	init_waitqueue_head(&u->peer_wait);
 836	init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
 837	memset(&u->scm_stat, 0, sizeof(struct scm_stat));
 838	unix_insert_socket(unix_sockets_unbound(sk), sk);
 839out:
 840	if (sk == NULL)
 841		atomic_long_dec(&unix_nr_socks);
 842	else {
 843		local_bh_disable();
 844		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 845		local_bh_enable();
 846	}
 847	return sk;
 848}
 849
 850static int unix_create(struct net *net, struct socket *sock, int protocol,
 851		       int kern)
 852{
 853	if (protocol && protocol != PF_UNIX)
 854		return -EPROTONOSUPPORT;
 855
 856	sock->state = SS_UNCONNECTED;
 857
 858	switch (sock->type) {
 859	case SOCK_STREAM:
 860		sock->ops = &unix_stream_ops;
 861		break;
 862		/*
 863		 *	Believe it or not BSD has AF_UNIX, SOCK_RAW though
 864		 *	nothing uses it.
 865		 */
 866	case SOCK_RAW:
 867		sock->type = SOCK_DGRAM;
 868		fallthrough;
 869	case SOCK_DGRAM:
 870		sock->ops = &unix_dgram_ops;
 871		break;
 872	case SOCK_SEQPACKET:
 873		sock->ops = &unix_seqpacket_ops;
 874		break;
 875	default:
 876		return -ESOCKTNOSUPPORT;
 877	}
 878
 879	return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
 880}
 881
 882static int unix_release(struct socket *sock)
 883{
 884	struct sock *sk = sock->sk;
 885
 886	if (!sk)
 887		return 0;
 888
 889	unix_release_sock(sk, 0);
 890	sock->sk = NULL;
 891
 892	return 0;
 893}
 894
 895static int unix_autobind(struct socket *sock)
 896{
 897	struct sock *sk = sock->sk;
 898	struct net *net = sock_net(sk);
 899	struct unix_sock *u = unix_sk(sk);
 900	static u32 ordernum = 1;
 901	struct unix_address *addr;
 902	int err;
 903	unsigned int retries = 0;
 904
 905	err = mutex_lock_interruptible(&u->bindlock);
 906	if (err)
 907		return err;
 908
 
 909	if (u->addr)
 910		goto out;
 911
 912	err = -ENOMEM;
 913	addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 914	if (!addr)
 915		goto out;
 916
 917	addr->name->sun_family = AF_UNIX;
 918	refcount_set(&addr->refcnt, 1);
 919
 920retry:
 921	addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 922	addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
 923	addr->hash ^= sk->sk_type;
 924
 925	spin_lock(&unix_table_lock);
 926	ordernum = (ordernum+1)&0xFFFFF;
 927
 928	if (__unix_find_socket_byname(net, addr->name, addr->len, addr->hash)) {
 
 929		spin_unlock(&unix_table_lock);
 930		/*
 931		 * __unix_find_socket_byname() may take long time if many names
 932		 * are already in use.
 933		 */
 934		cond_resched();
 935		/* Give up if all names seems to be in use. */
 936		if (retries++ == 0xFFFFF) {
 937			err = -ENOSPC;
 938			kfree(addr);
 939			goto out;
 940		}
 941		goto retry;
 942	}
 
 943
 944	__unix_set_addr(sk, addr, addr->hash);
 
 
 945	spin_unlock(&unix_table_lock);
 946	err = 0;
 947
 948out:	mutex_unlock(&u->bindlock);
 949	return err;
 950}
 951
 952static struct sock *unix_find_other(struct net *net,
 953				    struct sockaddr_un *sunname, int len,
 954				    int type, unsigned int hash, int *error)
 955{
 956	struct sock *u;
 957	struct path path;
 958	int err = 0;
 959
 960	if (sunname->sun_path[0]) {
 961		struct inode *inode;
 962		err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
 963		if (err)
 964			goto fail;
 965		inode = d_backing_inode(path.dentry);
 966		err = path_permission(&path, MAY_WRITE);
 967		if (err)
 968			goto put_fail;
 969
 970		err = -ECONNREFUSED;
 971		if (!S_ISSOCK(inode->i_mode))
 972			goto put_fail;
 973		u = unix_find_socket_byinode(inode);
 974		if (!u)
 975			goto put_fail;
 976
 977		if (u->sk_type == type)
 978			touch_atime(&path);
 979
 980		path_put(&path);
 981
 982		err = -EPROTOTYPE;
 983		if (u->sk_type != type) {
 984			sock_put(u);
 985			goto fail;
 986		}
 987	} else {
 988		err = -ECONNREFUSED;
 989		u = unix_find_socket_byname(net, sunname, len, type ^ hash);
 990		if (u) {
 991			struct dentry *dentry;
 992			dentry = unix_sk(u)->path.dentry;
 993			if (dentry)
 994				touch_atime(&unix_sk(u)->path);
 995		} else
 996			goto fail;
 997	}
 998	return u;
 999
1000put_fail:
1001	path_put(&path);
1002fail:
1003	*error = err;
1004	return NULL;
1005}
1006
1007static int unix_bind_bsd(struct sock *sk, struct unix_address *addr)
1008{
1009	struct unix_sock *u = unix_sk(sk);
1010	umode_t mode = S_IFSOCK |
1011	       (SOCK_INODE(sk->sk_socket)->i_mode & ~current_umask());
1012	struct user_namespace *ns; // barf...
1013	struct path parent;
1014	struct dentry *dentry;
1015	unsigned int hash;
1016	int err;
1017
1018	/*
1019	 * Get the parent directory, calculate the hash for last
1020	 * component.
1021	 */
1022	dentry = kern_path_create(AT_FDCWD, addr->name->sun_path, &parent, 0);
 
1023	if (IS_ERR(dentry))
1024		return PTR_ERR(dentry);
1025	ns = mnt_user_ns(parent.mnt);
1026
1027	/*
1028	 * All right, let's create it.
1029	 */
1030	err = security_path_mknod(&parent, dentry, mode, 0);
1031	if (!err)
1032		err = vfs_mknod(ns, d_inode(parent.dentry), dentry, mode, 0);
1033	if (err)
1034		goto out;
1035	err = mutex_lock_interruptible(&u->bindlock);
1036	if (err)
1037		goto out_unlink;
1038	if (u->addr)
1039		goto out_unlock;
1040
1041	addr->hash = UNIX_HASH_SIZE;
1042	hash = d_backing_inode(dentry)->i_ino & (UNIX_HASH_SIZE - 1);
1043	spin_lock(&unix_table_lock);
1044	u->path.mnt = mntget(parent.mnt);
1045	u->path.dentry = dget(dentry);
1046	__unix_set_addr(sk, addr, hash);
1047	spin_unlock(&unix_table_lock);
1048	mutex_unlock(&u->bindlock);
1049	done_path_create(&parent, dentry);
1050	return 0;
1051
1052out_unlock:
1053	mutex_unlock(&u->bindlock);
1054	err = -EINVAL;
1055out_unlink:
1056	/* failed after successful mknod?  unlink what we'd created... */
1057	vfs_unlink(ns, d_inode(parent.dentry), dentry, NULL);
1058out:
1059	done_path_create(&parent, dentry);
1060	return err;
1061}
1062
1063static int unix_bind_abstract(struct sock *sk, struct unix_address *addr)
1064{
1065	struct unix_sock *u = unix_sk(sk);
1066	int err;
1067
1068	err = mutex_lock_interruptible(&u->bindlock);
1069	if (err)
1070		return err;
1071
1072	if (u->addr) {
1073		mutex_unlock(&u->bindlock);
1074		return -EINVAL;
1075	}
1076
1077	spin_lock(&unix_table_lock);
1078	if (__unix_find_socket_byname(sock_net(sk), addr->name, addr->len,
1079				      addr->hash)) {
1080		spin_unlock(&unix_table_lock);
1081		mutex_unlock(&u->bindlock);
1082		return -EADDRINUSE;
1083	}
1084	__unix_set_addr(sk, addr, addr->hash);
1085	spin_unlock(&unix_table_lock);
1086	mutex_unlock(&u->bindlock);
1087	return 0;
1088}
1089
1090static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1091{
1092	struct sock *sk = sock->sk;
 
 
1093	struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1094	char *sun_path = sunaddr->sun_path;
1095	int err;
1096	unsigned int hash;
1097	struct unix_address *addr;
 
 
1098
 
1099	if (addr_len < offsetofend(struct sockaddr_un, sun_family) ||
1100	    sunaddr->sun_family != AF_UNIX)
1101		return -EINVAL;
1102
1103	if (addr_len == sizeof(short))
1104		return unix_autobind(sock);
 
 
1105
1106	err = unix_mkname(sunaddr, addr_len, &hash);
1107	if (err < 0)
1108		return err;
1109	addr_len = err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1110	addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1111	if (!addr)
1112		return -ENOMEM;
1113
1114	memcpy(addr->name, sunaddr, addr_len);
1115	addr->len = addr_len;
1116	addr->hash = hash ^ sk->sk_type;
1117	refcount_set(&addr->refcnt, 1);
1118
1119	if (sun_path[0])
1120		err = unix_bind_bsd(sk, addr);
1121	else
1122		err = unix_bind_abstract(sk, addr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1123	if (err)
1124		unix_release_addr(addr);
1125	return err == -EEXIST ? -EADDRINUSE : err;
 
1126}
1127
1128static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1129{
1130	if (unlikely(sk1 == sk2) || !sk2) {
1131		unix_state_lock(sk1);
1132		return;
1133	}
1134	if (sk1 < sk2) {
1135		unix_state_lock(sk1);
1136		unix_state_lock_nested(sk2);
1137	} else {
1138		unix_state_lock(sk2);
1139		unix_state_lock_nested(sk1);
1140	}
1141}
1142
1143static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1144{
1145	if (unlikely(sk1 == sk2) || !sk2) {
1146		unix_state_unlock(sk1);
1147		return;
1148	}
1149	unix_state_unlock(sk1);
1150	unix_state_unlock(sk2);
1151}
1152
1153static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1154			      int alen, int flags)
1155{
1156	struct sock *sk = sock->sk;
1157	struct net *net = sock_net(sk);
1158	struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1159	struct sock *other;
1160	unsigned int hash;
1161	int err;
1162
1163	err = -EINVAL;
1164	if (alen < offsetofend(struct sockaddr, sa_family))
1165		goto out;
1166
1167	if (addr->sa_family != AF_UNSPEC) {
1168		err = unix_mkname(sunaddr, alen, &hash);
1169		if (err < 0)
1170			goto out;
1171		alen = err;
1172
1173		if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1174		    !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1175			goto out;
1176
1177restart:
1178		other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1179		if (!other)
1180			goto out;
1181
1182		unix_state_double_lock(sk, other);
1183
1184		/* Apparently VFS overslept socket death. Retry. */
1185		if (sock_flag(other, SOCK_DEAD)) {
1186			unix_state_double_unlock(sk, other);
1187			sock_put(other);
1188			goto restart;
1189		}
1190
1191		err = -EPERM;
1192		if (!unix_may_send(sk, other))
1193			goto out_unlock;
1194
1195		err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1196		if (err)
1197			goto out_unlock;
1198
1199	} else {
1200		/*
1201		 *	1003.1g breaking connected state with AF_UNSPEC
1202		 */
1203		other = NULL;
1204		unix_state_double_lock(sk, other);
1205	}
1206
1207	/*
1208	 * If it was connected, reconnect.
1209	 */
1210	if (unix_peer(sk)) {
1211		struct sock *old_peer = unix_peer(sk);
1212		unix_peer(sk) = other;
1213		unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1214
1215		unix_state_double_unlock(sk, other);
1216
1217		if (other != old_peer)
1218			unix_dgram_disconnected(sk, old_peer);
1219		sock_put(old_peer);
1220	} else {
1221		unix_peer(sk) = other;
1222		unix_state_double_unlock(sk, other);
1223	}
1224	return 0;
1225
1226out_unlock:
1227	unix_state_double_unlock(sk, other);
1228	sock_put(other);
1229out:
1230	return err;
1231}
1232
1233static long unix_wait_for_peer(struct sock *other, long timeo)
1234	__releases(&unix_sk(other)->lock)
1235{
1236	struct unix_sock *u = unix_sk(other);
1237	int sched;
1238	DEFINE_WAIT(wait);
1239
1240	prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1241
1242	sched = !sock_flag(other, SOCK_DEAD) &&
1243		!(other->sk_shutdown & RCV_SHUTDOWN) &&
1244		unix_recvq_full(other);
1245
1246	unix_state_unlock(other);
1247
1248	if (sched)
1249		timeo = schedule_timeout(timeo);
1250
1251	finish_wait(&u->peer_wait, &wait);
1252	return timeo;
1253}
1254
1255static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1256			       int addr_len, int flags)
1257{
1258	struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1259	struct sock *sk = sock->sk;
1260	struct net *net = sock_net(sk);
1261	struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1262	struct sock *newsk = NULL;
1263	struct sock *other = NULL;
1264	struct sk_buff *skb = NULL;
1265	unsigned int hash;
1266	int st;
1267	int err;
1268	long timeo;
1269
1270	err = unix_mkname(sunaddr, addr_len, &hash);
1271	if (err < 0)
1272		goto out;
1273	addr_len = err;
1274
1275	if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1276	    (err = unix_autobind(sock)) != 0)
1277		goto out;
1278
1279	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1280
1281	/* First of all allocate resources.
1282	   If we will make it after state is locked,
1283	   we will have to recheck all again in any case.
1284	 */
1285
1286	err = -ENOMEM;
1287
1288	/* create new sock for complete connection */
1289	newsk = unix_create1(sock_net(sk), NULL, 0);
1290	if (newsk == NULL)
1291		goto out;
1292
1293	/* Allocate skb for sending to listening sock */
1294	skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1295	if (skb == NULL)
1296		goto out;
1297
1298restart:
1299	/*  Find listening sock. */
1300	other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1301	if (!other)
1302		goto out;
1303
1304	/* Latch state of peer */
1305	unix_state_lock(other);
1306
1307	/* Apparently VFS overslept socket death. Retry. */
1308	if (sock_flag(other, SOCK_DEAD)) {
1309		unix_state_unlock(other);
1310		sock_put(other);
1311		goto restart;
1312	}
1313
1314	err = -ECONNREFUSED;
1315	if (other->sk_state != TCP_LISTEN)
1316		goto out_unlock;
1317	if (other->sk_shutdown & RCV_SHUTDOWN)
1318		goto out_unlock;
1319
1320	if (unix_recvq_full(other)) {
1321		err = -EAGAIN;
1322		if (!timeo)
1323			goto out_unlock;
1324
1325		timeo = unix_wait_for_peer(other, timeo);
1326
1327		err = sock_intr_errno(timeo);
1328		if (signal_pending(current))
1329			goto out;
1330		sock_put(other);
1331		goto restart;
1332	}
1333
1334	/* Latch our state.
1335
1336	   It is tricky place. We need to grab our state lock and cannot
1337	   drop lock on peer. It is dangerous because deadlock is
1338	   possible. Connect to self case and simultaneous
1339	   attempt to connect are eliminated by checking socket
1340	   state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1341	   check this before attempt to grab lock.
1342
1343	   Well, and we have to recheck the state after socket locked.
1344	 */
1345	st = sk->sk_state;
1346
1347	switch (st) {
1348	case TCP_CLOSE:
1349		/* This is ok... continue with connect */
1350		break;
1351	case TCP_ESTABLISHED:
1352		/* Socket is already connected */
1353		err = -EISCONN;
1354		goto out_unlock;
1355	default:
1356		err = -EINVAL;
1357		goto out_unlock;
1358	}
1359
1360	unix_state_lock_nested(sk);
1361
1362	if (sk->sk_state != st) {
1363		unix_state_unlock(sk);
1364		unix_state_unlock(other);
1365		sock_put(other);
1366		goto restart;
1367	}
1368
1369	err = security_unix_stream_connect(sk, other, newsk);
1370	if (err) {
1371		unix_state_unlock(sk);
1372		goto out_unlock;
1373	}
1374
1375	/* The way is open! Fastly set all the necessary fields... */
1376
1377	sock_hold(sk);
1378	unix_peer(newsk)	= sk;
1379	newsk->sk_state		= TCP_ESTABLISHED;
1380	newsk->sk_type		= sk->sk_type;
1381	init_peercred(newsk);
1382	newu = unix_sk(newsk);
1383	RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1384	otheru = unix_sk(other);
1385
1386	/* copy address information from listening to new sock
1387	 *
1388	 * The contents of *(otheru->addr) and otheru->path
1389	 * are seen fully set up here, since we have found
1390	 * otheru in hash under unix_table_lock.  Insertion
1391	 * into the hash chain we'd found it in had been done
1392	 * in an earlier critical area protected by unix_table_lock,
1393	 * the same one where we'd set *(otheru->addr) contents,
1394	 * as well as otheru->path and otheru->addr itself.
1395	 *
1396	 * Using smp_store_release() here to set newu->addr
1397	 * is enough to make those stores, as well as stores
1398	 * to newu->path visible to anyone who gets newu->addr
1399	 * by smp_load_acquire().  IOW, the same warranties
1400	 * as for unix_sock instances bound in unix_bind() or
1401	 * in unix_autobind().
1402	 */
1403	if (otheru->path.dentry) {
1404		path_get(&otheru->path);
1405		newu->path = otheru->path;
1406	}
1407	refcount_inc(&otheru->addr->refcnt);
1408	smp_store_release(&newu->addr, otheru->addr);
1409
1410	/* Set credentials */
1411	copy_peercred(sk, other);
1412
1413	sock->state	= SS_CONNECTED;
1414	sk->sk_state	= TCP_ESTABLISHED;
1415	sock_hold(newsk);
1416
1417	smp_mb__after_atomic();	/* sock_hold() does an atomic_inc() */
1418	unix_peer(sk)	= newsk;
1419
1420	unix_state_unlock(sk);
1421
1422	/* take ten and send info to listening sock */
1423	spin_lock(&other->sk_receive_queue.lock);
1424	__skb_queue_tail(&other->sk_receive_queue, skb);
1425	spin_unlock(&other->sk_receive_queue.lock);
1426	unix_state_unlock(other);
1427	other->sk_data_ready(other);
1428	sock_put(other);
1429	return 0;
1430
1431out_unlock:
1432	if (other)
1433		unix_state_unlock(other);
1434
1435out:
1436	kfree_skb(skb);
1437	if (newsk)
1438		unix_release_sock(newsk, 0);
1439	if (other)
1440		sock_put(other);
1441	return err;
1442}
1443
1444static int unix_socketpair(struct socket *socka, struct socket *sockb)
1445{
1446	struct sock *ska = socka->sk, *skb = sockb->sk;
1447
1448	/* Join our sockets back to back */
1449	sock_hold(ska);
1450	sock_hold(skb);
1451	unix_peer(ska) = skb;
1452	unix_peer(skb) = ska;
1453	init_peercred(ska);
1454	init_peercred(skb);
1455
1456	if (ska->sk_type != SOCK_DGRAM) {
1457		ska->sk_state = TCP_ESTABLISHED;
1458		skb->sk_state = TCP_ESTABLISHED;
1459		socka->state  = SS_CONNECTED;
1460		sockb->state  = SS_CONNECTED;
1461	}
1462	return 0;
1463}
1464
1465static void unix_sock_inherit_flags(const struct socket *old,
1466				    struct socket *new)
1467{
1468	if (test_bit(SOCK_PASSCRED, &old->flags))
1469		set_bit(SOCK_PASSCRED, &new->flags);
1470	if (test_bit(SOCK_PASSSEC, &old->flags))
1471		set_bit(SOCK_PASSSEC, &new->flags);
1472}
1473
1474static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1475		       bool kern)
1476{
1477	struct sock *sk = sock->sk;
1478	struct sock *tsk;
1479	struct sk_buff *skb;
1480	int err;
1481
1482	err = -EOPNOTSUPP;
1483	if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1484		goto out;
1485
1486	err = -EINVAL;
1487	if (sk->sk_state != TCP_LISTEN)
1488		goto out;
1489
1490	/* If socket state is TCP_LISTEN it cannot change (for now...),
1491	 * so that no locks are necessary.
1492	 */
1493
1494	skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1495	if (!skb) {
1496		/* This means receive shutdown. */
1497		if (err == 0)
1498			err = -EINVAL;
1499		goto out;
1500	}
1501
1502	tsk = skb->sk;
1503	skb_free_datagram(sk, skb);
1504	wake_up_interruptible(&unix_sk(sk)->peer_wait);
1505
1506	/* attach accepted sock to socket */
1507	unix_state_lock(tsk);
1508	newsock->state = SS_CONNECTED;
1509	unix_sock_inherit_flags(sock, newsock);
1510	sock_graft(tsk, newsock);
1511	unix_state_unlock(tsk);
1512	return 0;
1513
1514out:
1515	return err;
1516}
1517
1518
1519static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
1520{
1521	struct sock *sk = sock->sk;
1522	struct unix_address *addr;
1523	DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1524	int err = 0;
1525
1526	if (peer) {
1527		sk = unix_peer_get(sk);
1528
1529		err = -ENOTCONN;
1530		if (!sk)
1531			goto out;
1532		err = 0;
1533	} else {
1534		sock_hold(sk);
1535	}
1536
1537	addr = smp_load_acquire(&unix_sk(sk)->addr);
1538	if (!addr) {
1539		sunaddr->sun_family = AF_UNIX;
1540		sunaddr->sun_path[0] = 0;
1541		err = sizeof(short);
1542	} else {
1543		err = addr->len;
1544		memcpy(sunaddr, addr->name, addr->len);
1545	}
1546	sock_put(sk);
1547out:
1548	return err;
1549}
1550
1551static void unix_peek_fds(struct scm_cookie *scm, struct sk_buff *skb)
1552{
1553	scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1554
1555	/*
1556	 * Garbage collection of unix sockets starts by selecting a set of
1557	 * candidate sockets which have reference only from being in flight
1558	 * (total_refs == inflight_refs).  This condition is checked once during
1559	 * the candidate collection phase, and candidates are marked as such, so
1560	 * that non-candidates can later be ignored.  While inflight_refs is
1561	 * protected by unix_gc_lock, total_refs (file count) is not, hence this
1562	 * is an instantaneous decision.
1563	 *
1564	 * Once a candidate, however, the socket must not be reinstalled into a
1565	 * file descriptor while the garbage collection is in progress.
1566	 *
1567	 * If the above conditions are met, then the directed graph of
1568	 * candidates (*) does not change while unix_gc_lock is held.
1569	 *
1570	 * Any operations that changes the file count through file descriptors
1571	 * (dup, close, sendmsg) does not change the graph since candidates are
1572	 * not installed in fds.
1573	 *
1574	 * Dequeing a candidate via recvmsg would install it into an fd, but
1575	 * that takes unix_gc_lock to decrement the inflight count, so it's
1576	 * serialized with garbage collection.
1577	 *
1578	 * MSG_PEEK is special in that it does not change the inflight count,
1579	 * yet does install the socket into an fd.  The following lock/unlock
1580	 * pair is to ensure serialization with garbage collection.  It must be
1581	 * done between incrementing the file count and installing the file into
1582	 * an fd.
1583	 *
1584	 * If garbage collection starts after the barrier provided by the
1585	 * lock/unlock, then it will see the elevated refcount and not mark this
1586	 * as a candidate.  If a garbage collection is already in progress
1587	 * before the file count was incremented, then the lock/unlock pair will
1588	 * ensure that garbage collection is finished before progressing to
1589	 * installing the fd.
1590	 *
1591	 * (*) A -> B where B is on the queue of A or B is on the queue of C
1592	 * which is on the queue of listening socket A.
1593	 */
1594	spin_lock(&unix_gc_lock);
1595	spin_unlock(&unix_gc_lock);
1596}
1597
1598static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1599{
1600	int err = 0;
1601
1602	UNIXCB(skb).pid  = get_pid(scm->pid);
1603	UNIXCB(skb).uid = scm->creds.uid;
1604	UNIXCB(skb).gid = scm->creds.gid;
1605	UNIXCB(skb).fp = NULL;
1606	unix_get_secdata(scm, skb);
1607	if (scm->fp && send_fds)
1608		err = unix_attach_fds(scm, skb);
1609
1610	skb->destructor = unix_destruct_scm;
1611	return err;
1612}
1613
1614static bool unix_passcred_enabled(const struct socket *sock,
1615				  const struct sock *other)
1616{
1617	return test_bit(SOCK_PASSCRED, &sock->flags) ||
1618	       !other->sk_socket ||
1619	       test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1620}
1621
1622/*
1623 * Some apps rely on write() giving SCM_CREDENTIALS
1624 * We include credentials if source or destination socket
1625 * asserted SOCK_PASSCRED.
1626 */
1627static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1628			    const struct sock *other)
1629{
1630	if (UNIXCB(skb).pid)
1631		return;
1632	if (unix_passcred_enabled(sock, other)) {
1633		UNIXCB(skb).pid  = get_pid(task_tgid(current));
1634		current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1635	}
1636}
1637
1638static int maybe_init_creds(struct scm_cookie *scm,
1639			    struct socket *socket,
1640			    const struct sock *other)
1641{
1642	int err;
1643	struct msghdr msg = { .msg_controllen = 0 };
1644
1645	err = scm_send(socket, &msg, scm, false);
1646	if (err)
1647		return err;
1648
1649	if (unix_passcred_enabled(socket, other)) {
1650		scm->pid = get_pid(task_tgid(current));
1651		current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1652	}
1653	return err;
1654}
1655
1656static bool unix_skb_scm_eq(struct sk_buff *skb,
1657			    struct scm_cookie *scm)
1658{
1659	const struct unix_skb_parms *u = &UNIXCB(skb);
1660
1661	return u->pid == scm->pid &&
1662	       uid_eq(u->uid, scm->creds.uid) &&
1663	       gid_eq(u->gid, scm->creds.gid) &&
1664	       unix_secdata_eq(scm, skb);
1665}
1666
1667static void scm_stat_add(struct sock *sk, struct sk_buff *skb)
1668{
1669	struct scm_fp_list *fp = UNIXCB(skb).fp;
1670	struct unix_sock *u = unix_sk(sk);
1671
1672	if (unlikely(fp && fp->count))
1673		atomic_add(fp->count, &u->scm_stat.nr_fds);
1674}
1675
1676static void scm_stat_del(struct sock *sk, struct sk_buff *skb)
1677{
1678	struct scm_fp_list *fp = UNIXCB(skb).fp;
1679	struct unix_sock *u = unix_sk(sk);
1680
1681	if (unlikely(fp && fp->count))
1682		atomic_sub(fp->count, &u->scm_stat.nr_fds);
1683}
1684
1685/*
1686 *	Send AF_UNIX data.
1687 */
1688
1689static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1690			      size_t len)
1691{
1692	struct sock *sk = sock->sk;
1693	struct net *net = sock_net(sk);
1694	struct unix_sock *u = unix_sk(sk);
1695	DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1696	struct sock *other = NULL;
1697	int namelen = 0; /* fake GCC */
1698	int err;
1699	unsigned int hash;
1700	struct sk_buff *skb;
1701	long timeo;
1702	struct scm_cookie scm;
1703	int data_len = 0;
1704	int sk_locked;
1705
1706	wait_for_unix_gc();
1707	err = scm_send(sock, msg, &scm, false);
1708	if (err < 0)
1709		return err;
1710
1711	err = -EOPNOTSUPP;
1712	if (msg->msg_flags&MSG_OOB)
1713		goto out;
1714
1715	if (msg->msg_namelen) {
1716		err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1717		if (err < 0)
1718			goto out;
1719		namelen = err;
1720	} else {
1721		sunaddr = NULL;
1722		err = -ENOTCONN;
1723		other = unix_peer_get(sk);
1724		if (!other)
1725			goto out;
1726	}
1727
1728	if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1729	    && (err = unix_autobind(sock)) != 0)
1730		goto out;
1731
1732	err = -EMSGSIZE;
1733	if (len > sk->sk_sndbuf - 32)
1734		goto out;
1735
1736	if (len > SKB_MAX_ALLOC) {
1737		data_len = min_t(size_t,
1738				 len - SKB_MAX_ALLOC,
1739				 MAX_SKB_FRAGS * PAGE_SIZE);
1740		data_len = PAGE_ALIGN(data_len);
1741
1742		BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1743	}
1744
1745	skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1746				   msg->msg_flags & MSG_DONTWAIT, &err,
1747				   PAGE_ALLOC_COSTLY_ORDER);
1748	if (skb == NULL)
1749		goto out;
1750
1751	err = unix_scm_to_skb(&scm, skb, true);
1752	if (err < 0)
1753		goto out_free;
1754
1755	skb_put(skb, len - data_len);
1756	skb->data_len = data_len;
1757	skb->len = len;
1758	err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1759	if (err)
1760		goto out_free;
1761
1762	timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1763
1764restart:
1765	if (!other) {
1766		err = -ECONNRESET;
1767		if (sunaddr == NULL)
1768			goto out_free;
1769
1770		other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1771					hash, &err);
1772		if (other == NULL)
1773			goto out_free;
1774	}
1775
1776	if (sk_filter(other, skb) < 0) {
1777		/* Toss the packet but do not return any error to the sender */
1778		err = len;
1779		goto out_free;
1780	}
1781
1782	sk_locked = 0;
1783	unix_state_lock(other);
1784restart_locked:
1785	err = -EPERM;
1786	if (!unix_may_send(sk, other))
1787		goto out_unlock;
1788
1789	if (unlikely(sock_flag(other, SOCK_DEAD))) {
1790		/*
1791		 *	Check with 1003.1g - what should
1792		 *	datagram error
1793		 */
1794		unix_state_unlock(other);
1795		sock_put(other);
1796
1797		if (!sk_locked)
1798			unix_state_lock(sk);
1799
1800		err = 0;
1801		if (unix_peer(sk) == other) {
1802			unix_peer(sk) = NULL;
1803			unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1804
1805			unix_state_unlock(sk);
1806
1807			unix_dgram_disconnected(sk, other);
1808			sock_put(other);
1809			err = -ECONNREFUSED;
1810		} else {
1811			unix_state_unlock(sk);
1812		}
1813
1814		other = NULL;
1815		if (err)
1816			goto out_free;
1817		goto restart;
1818	}
1819
1820	err = -EPIPE;
1821	if (other->sk_shutdown & RCV_SHUTDOWN)
1822		goto out_unlock;
1823
1824	if (sk->sk_type != SOCK_SEQPACKET) {
1825		err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1826		if (err)
1827			goto out_unlock;
1828	}
1829
1830	/* other == sk && unix_peer(other) != sk if
1831	 * - unix_peer(sk) == NULL, destination address bound to sk
1832	 * - unix_peer(sk) == sk by time of get but disconnected before lock
1833	 */
1834	if (other != sk &&
1835	    unlikely(unix_peer(other) != sk &&
1836	    unix_recvq_full_lockless(other))) {
1837		if (timeo) {
1838			timeo = unix_wait_for_peer(other, timeo);
1839
1840			err = sock_intr_errno(timeo);
1841			if (signal_pending(current))
1842				goto out_free;
1843
1844			goto restart;
1845		}
1846
1847		if (!sk_locked) {
1848			unix_state_unlock(other);
1849			unix_state_double_lock(sk, other);
1850		}
1851
1852		if (unix_peer(sk) != other ||
1853		    unix_dgram_peer_wake_me(sk, other)) {
1854			err = -EAGAIN;
1855			sk_locked = 1;
1856			goto out_unlock;
1857		}
1858
1859		if (!sk_locked) {
1860			sk_locked = 1;
1861			goto restart_locked;
1862		}
1863	}
1864
1865	if (unlikely(sk_locked))
1866		unix_state_unlock(sk);
1867
1868	if (sock_flag(other, SOCK_RCVTSTAMP))
1869		__net_timestamp(skb);
1870	maybe_add_creds(skb, sock, other);
1871	scm_stat_add(other, skb);
1872	skb_queue_tail(&other->sk_receive_queue, skb);
1873	unix_state_unlock(other);
1874	other->sk_data_ready(other);
1875	sock_put(other);
1876	scm_destroy(&scm);
1877	return len;
1878
1879out_unlock:
1880	if (sk_locked)
1881		unix_state_unlock(sk);
1882	unix_state_unlock(other);
1883out_free:
1884	kfree_skb(skb);
1885out:
1886	if (other)
1887		sock_put(other);
1888	scm_destroy(&scm);
1889	return err;
1890}
1891
1892/* We use paged skbs for stream sockets, and limit occupancy to 32768
1893 * bytes, and a minimum of a full page.
1894 */
1895#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1896
1897static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1898			       size_t len)
1899{
1900	struct sock *sk = sock->sk;
1901	struct sock *other = NULL;
1902	int err, size;
1903	struct sk_buff *skb;
1904	int sent = 0;
1905	struct scm_cookie scm;
1906	bool fds_sent = false;
1907	int data_len;
1908
1909	wait_for_unix_gc();
1910	err = scm_send(sock, msg, &scm, false);
1911	if (err < 0)
1912		return err;
1913
1914	err = -EOPNOTSUPP;
1915	if (msg->msg_flags&MSG_OOB)
1916		goto out_err;
1917
1918	if (msg->msg_namelen) {
1919		err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1920		goto out_err;
1921	} else {
1922		err = -ENOTCONN;
1923		other = unix_peer(sk);
1924		if (!other)
1925			goto out_err;
1926	}
1927
1928	if (sk->sk_shutdown & SEND_SHUTDOWN)
1929		goto pipe_err;
1930
1931	while (sent < len) {
1932		size = len - sent;
1933
1934		/* Keep two messages in the pipe so it schedules better */
1935		size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1936
1937		/* allow fallback to order-0 allocations */
1938		size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1939
1940		data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1941
1942		data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1943
1944		skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1945					   msg->msg_flags & MSG_DONTWAIT, &err,
1946					   get_order(UNIX_SKB_FRAGS_SZ));
1947		if (!skb)
1948			goto out_err;
1949
1950		/* Only send the fds in the first buffer */
1951		err = unix_scm_to_skb(&scm, skb, !fds_sent);
1952		if (err < 0) {
1953			kfree_skb(skb);
1954			goto out_err;
1955		}
1956		fds_sent = true;
1957
1958		skb_put(skb, size - data_len);
1959		skb->data_len = data_len;
1960		skb->len = size;
1961		err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
1962		if (err) {
1963			kfree_skb(skb);
1964			goto out_err;
1965		}
1966
1967		unix_state_lock(other);
1968
1969		if (sock_flag(other, SOCK_DEAD) ||
1970		    (other->sk_shutdown & RCV_SHUTDOWN))
1971			goto pipe_err_free;
1972
1973		maybe_add_creds(skb, sock, other);
1974		scm_stat_add(other, skb);
1975		skb_queue_tail(&other->sk_receive_queue, skb);
1976		unix_state_unlock(other);
1977		other->sk_data_ready(other);
1978		sent += size;
1979	}
1980
1981	scm_destroy(&scm);
1982
1983	return sent;
1984
1985pipe_err_free:
1986	unix_state_unlock(other);
1987	kfree_skb(skb);
1988pipe_err:
1989	if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1990		send_sig(SIGPIPE, current, 0);
1991	err = -EPIPE;
1992out_err:
1993	scm_destroy(&scm);
1994	return sent ? : err;
1995}
1996
1997static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1998				    int offset, size_t size, int flags)
1999{
2000	int err;
2001	bool send_sigpipe = false;
2002	bool init_scm = true;
2003	struct scm_cookie scm;
2004	struct sock *other, *sk = socket->sk;
2005	struct sk_buff *skb, *newskb = NULL, *tail = NULL;
2006
2007	if (flags & MSG_OOB)
2008		return -EOPNOTSUPP;
2009
2010	other = unix_peer(sk);
2011	if (!other || sk->sk_state != TCP_ESTABLISHED)
2012		return -ENOTCONN;
2013
2014	if (false) {
2015alloc_skb:
2016		unix_state_unlock(other);
2017		mutex_unlock(&unix_sk(other)->iolock);
2018		newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
2019					      &err, 0);
2020		if (!newskb)
2021			goto err;
2022	}
2023
2024	/* we must acquire iolock as we modify already present
2025	 * skbs in the sk_receive_queue and mess with skb->len
2026	 */
2027	err = mutex_lock_interruptible(&unix_sk(other)->iolock);
2028	if (err) {
2029		err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
2030		goto err;
2031	}
2032
2033	if (sk->sk_shutdown & SEND_SHUTDOWN) {
2034		err = -EPIPE;
2035		send_sigpipe = true;
2036		goto err_unlock;
2037	}
2038
2039	unix_state_lock(other);
2040
2041	if (sock_flag(other, SOCK_DEAD) ||
2042	    other->sk_shutdown & RCV_SHUTDOWN) {
2043		err = -EPIPE;
2044		send_sigpipe = true;
2045		goto err_state_unlock;
2046	}
2047
2048	if (init_scm) {
2049		err = maybe_init_creds(&scm, socket, other);
2050		if (err)
2051			goto err_state_unlock;
2052		init_scm = false;
2053	}
2054
2055	skb = skb_peek_tail(&other->sk_receive_queue);
2056	if (tail && tail == skb) {
2057		skb = newskb;
2058	} else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
2059		if (newskb) {
2060			skb = newskb;
2061		} else {
2062			tail = skb;
2063			goto alloc_skb;
2064		}
2065	} else if (newskb) {
2066		/* this is fast path, we don't necessarily need to
2067		 * call to kfree_skb even though with newskb == NULL
2068		 * this - does no harm
2069		 */
2070		consume_skb(newskb);
2071		newskb = NULL;
2072	}
2073
2074	if (skb_append_pagefrags(skb, page, offset, size)) {
2075		tail = skb;
2076		goto alloc_skb;
2077	}
2078
2079	skb->len += size;
2080	skb->data_len += size;
2081	skb->truesize += size;
2082	refcount_add(size, &sk->sk_wmem_alloc);
2083
2084	if (newskb) {
2085		err = unix_scm_to_skb(&scm, skb, false);
2086		if (err)
2087			goto err_state_unlock;
2088		spin_lock(&other->sk_receive_queue.lock);
2089		__skb_queue_tail(&other->sk_receive_queue, newskb);
2090		spin_unlock(&other->sk_receive_queue.lock);
2091	}
2092
2093	unix_state_unlock(other);
2094	mutex_unlock(&unix_sk(other)->iolock);
2095
2096	other->sk_data_ready(other);
2097	scm_destroy(&scm);
2098	return size;
2099
2100err_state_unlock:
2101	unix_state_unlock(other);
2102err_unlock:
2103	mutex_unlock(&unix_sk(other)->iolock);
2104err:
2105	kfree_skb(newskb);
2106	if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2107		send_sig(SIGPIPE, current, 0);
2108	if (!init_scm)
2109		scm_destroy(&scm);
2110	return err;
2111}
2112
2113static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2114				  size_t len)
2115{
2116	int err;
2117	struct sock *sk = sock->sk;
2118
2119	err = sock_error(sk);
2120	if (err)
2121		return err;
2122
2123	if (sk->sk_state != TCP_ESTABLISHED)
2124		return -ENOTCONN;
2125
2126	if (msg->msg_namelen)
2127		msg->msg_namelen = 0;
2128
2129	return unix_dgram_sendmsg(sock, msg, len);
2130}
2131
2132static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2133				  size_t size, int flags)
2134{
2135	struct sock *sk = sock->sk;
2136
2137	if (sk->sk_state != TCP_ESTABLISHED)
2138		return -ENOTCONN;
2139
2140	return unix_dgram_recvmsg(sock, msg, size, flags);
2141}
2142
2143static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2144{
2145	struct unix_address *addr = smp_load_acquire(&unix_sk(sk)->addr);
2146
2147	if (addr) {
2148		msg->msg_namelen = addr->len;
2149		memcpy(msg->msg_name, addr->name, addr->len);
2150	}
2151}
2152
2153static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2154			      size_t size, int flags)
2155{
2156	struct scm_cookie scm;
2157	struct sock *sk = sock->sk;
2158	struct unix_sock *u = unix_sk(sk);
2159	struct sk_buff *skb, *last;
2160	long timeo;
2161	int skip;
2162	int err;
2163
2164	err = -EOPNOTSUPP;
2165	if (flags&MSG_OOB)
2166		goto out;
2167
2168	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2169
2170	do {
2171		mutex_lock(&u->iolock);
2172
2173		skip = sk_peek_offset(sk, flags);
2174		skb = __skb_try_recv_datagram(sk, &sk->sk_receive_queue, flags,
2175					      &skip, &err, &last);
2176		if (skb) {
2177			if (!(flags & MSG_PEEK))
2178				scm_stat_del(sk, skb);
2179			break;
2180		}
2181
2182		mutex_unlock(&u->iolock);
2183
2184		if (err != -EAGAIN)
2185			break;
2186	} while (timeo &&
2187		 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
2188					      &err, &timeo, last));
2189
2190	if (!skb) { /* implies iolock unlocked */
2191		unix_state_lock(sk);
2192		/* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2193		if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2194		    (sk->sk_shutdown & RCV_SHUTDOWN))
2195			err = 0;
2196		unix_state_unlock(sk);
2197		goto out;
2198	}
2199
2200	if (wq_has_sleeper(&u->peer_wait))
2201		wake_up_interruptible_sync_poll(&u->peer_wait,
2202						EPOLLOUT | EPOLLWRNORM |
2203						EPOLLWRBAND);
2204
2205	if (msg->msg_name)
2206		unix_copy_addr(msg, skb->sk);
2207
2208	if (size > skb->len - skip)
2209		size = skb->len - skip;
2210	else if (size < skb->len - skip)
2211		msg->msg_flags |= MSG_TRUNC;
2212
2213	err = skb_copy_datagram_msg(skb, skip, msg, size);
2214	if (err)
2215		goto out_free;
2216
2217	if (sock_flag(sk, SOCK_RCVTSTAMP))
2218		__sock_recv_timestamp(msg, sk, skb);
2219
2220	memset(&scm, 0, sizeof(scm));
2221
2222	scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2223	unix_set_secdata(&scm, skb);
2224
2225	if (!(flags & MSG_PEEK)) {
2226		if (UNIXCB(skb).fp)
2227			unix_detach_fds(&scm, skb);
2228
2229		sk_peek_offset_bwd(sk, skb->len);
2230	} else {
2231		/* It is questionable: on PEEK we could:
2232		   - do not return fds - good, but too simple 8)
2233		   - return fds, and do not return them on read (old strategy,
2234		     apparently wrong)
2235		   - clone fds (I chose it for now, it is the most universal
2236		     solution)
2237
2238		   POSIX 1003.1g does not actually define this clearly
2239		   at all. POSIX 1003.1g doesn't define a lot of things
2240		   clearly however!
2241
2242		*/
2243
2244		sk_peek_offset_fwd(sk, size);
2245
2246		if (UNIXCB(skb).fp)
2247			unix_peek_fds(&scm, skb);
2248	}
2249	err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2250
2251	scm_recv(sock, msg, &scm, flags);
2252
2253out_free:
2254	skb_free_datagram(sk, skb);
2255	mutex_unlock(&u->iolock);
2256out:
2257	return err;
2258}
2259
2260/*
2261 *	Sleep until more data has arrived. But check for races..
2262 */
2263static long unix_stream_data_wait(struct sock *sk, long timeo,
2264				  struct sk_buff *last, unsigned int last_len,
2265				  bool freezable)
2266{
2267	struct sk_buff *tail;
2268	DEFINE_WAIT(wait);
2269
2270	unix_state_lock(sk);
2271
2272	for (;;) {
2273		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2274
2275		tail = skb_peek_tail(&sk->sk_receive_queue);
2276		if (tail != last ||
2277		    (tail && tail->len != last_len) ||
2278		    sk->sk_err ||
2279		    (sk->sk_shutdown & RCV_SHUTDOWN) ||
2280		    signal_pending(current) ||
2281		    !timeo)
2282			break;
2283
2284		sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2285		unix_state_unlock(sk);
2286		if (freezable)
2287			timeo = freezable_schedule_timeout(timeo);
2288		else
2289			timeo = schedule_timeout(timeo);
2290		unix_state_lock(sk);
2291
2292		if (sock_flag(sk, SOCK_DEAD))
2293			break;
2294
2295		sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2296	}
2297
2298	finish_wait(sk_sleep(sk), &wait);
2299	unix_state_unlock(sk);
2300	return timeo;
2301}
2302
2303static unsigned int unix_skb_len(const struct sk_buff *skb)
2304{
2305	return skb->len - UNIXCB(skb).consumed;
2306}
2307
2308struct unix_stream_read_state {
2309	int (*recv_actor)(struct sk_buff *, int, int,
2310			  struct unix_stream_read_state *);
2311	struct socket *socket;
2312	struct msghdr *msg;
2313	struct pipe_inode_info *pipe;
2314	size_t size;
2315	int flags;
2316	unsigned int splice_flags;
2317};
2318
2319static int unix_stream_read_generic(struct unix_stream_read_state *state,
2320				    bool freezable)
2321{
2322	struct scm_cookie scm;
2323	struct socket *sock = state->socket;
2324	struct sock *sk = sock->sk;
2325	struct unix_sock *u = unix_sk(sk);
2326	int copied = 0;
2327	int flags = state->flags;
2328	int noblock = flags & MSG_DONTWAIT;
2329	bool check_creds = false;
2330	int target;
2331	int err = 0;
2332	long timeo;
2333	int skip;
2334	size_t size = state->size;
2335	unsigned int last_len;
2336
2337	if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2338		err = -EINVAL;
2339		goto out;
2340	}
2341
2342	if (unlikely(flags & MSG_OOB)) {
2343		err = -EOPNOTSUPP;
2344		goto out;
2345	}
2346
2347	target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2348	timeo = sock_rcvtimeo(sk, noblock);
2349
2350	memset(&scm, 0, sizeof(scm));
2351
2352	/* Lock the socket to prevent queue disordering
2353	 * while sleeps in memcpy_tomsg
2354	 */
2355	mutex_lock(&u->iolock);
2356
2357	skip = max(sk_peek_offset(sk, flags), 0);
2358
2359	do {
2360		int chunk;
2361		bool drop_skb;
2362		struct sk_buff *skb, *last;
2363
2364redo:
2365		unix_state_lock(sk);
2366		if (sock_flag(sk, SOCK_DEAD)) {
2367			err = -ECONNRESET;
2368			goto unlock;
2369		}
2370		last = skb = skb_peek(&sk->sk_receive_queue);
2371		last_len = last ? last->len : 0;
2372again:
2373		if (skb == NULL) {
2374			if (copied >= target)
2375				goto unlock;
2376
2377			/*
2378			 *	POSIX 1003.1g mandates this order.
2379			 */
2380
2381			err = sock_error(sk);
2382			if (err)
2383				goto unlock;
2384			if (sk->sk_shutdown & RCV_SHUTDOWN)
2385				goto unlock;
2386
2387			unix_state_unlock(sk);
2388			if (!timeo) {
2389				err = -EAGAIN;
2390				break;
2391			}
2392
2393			mutex_unlock(&u->iolock);
2394
2395			timeo = unix_stream_data_wait(sk, timeo, last,
2396						      last_len, freezable);
2397
2398			if (signal_pending(current)) {
2399				err = sock_intr_errno(timeo);
2400				scm_destroy(&scm);
2401				goto out;
2402			}
2403
2404			mutex_lock(&u->iolock);
2405			goto redo;
2406unlock:
2407			unix_state_unlock(sk);
2408			break;
2409		}
2410
2411		while (skip >= unix_skb_len(skb)) {
2412			skip -= unix_skb_len(skb);
2413			last = skb;
2414			last_len = skb->len;
2415			skb = skb_peek_next(skb, &sk->sk_receive_queue);
2416			if (!skb)
2417				goto again;
2418		}
2419
2420		unix_state_unlock(sk);
2421
2422		if (check_creds) {
2423			/* Never glue messages from different writers */
2424			if (!unix_skb_scm_eq(skb, &scm))
2425				break;
2426		} else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2427			/* Copy credentials */
2428			scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2429			unix_set_secdata(&scm, skb);
2430			check_creds = true;
2431		}
2432
2433		/* Copy address just once */
2434		if (state->msg && state->msg->msg_name) {
2435			DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2436					 state->msg->msg_name);
2437			unix_copy_addr(state->msg, skb->sk);
2438			sunaddr = NULL;
2439		}
2440
2441		chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2442		skb_get(skb);
2443		chunk = state->recv_actor(skb, skip, chunk, state);
2444		drop_skb = !unix_skb_len(skb);
2445		/* skb is only safe to use if !drop_skb */
2446		consume_skb(skb);
2447		if (chunk < 0) {
2448			if (copied == 0)
2449				copied = -EFAULT;
2450			break;
2451		}
2452		copied += chunk;
2453		size -= chunk;
2454
2455		if (drop_skb) {
2456			/* the skb was touched by a concurrent reader;
2457			 * we should not expect anything from this skb
2458			 * anymore and assume it invalid - we can be
2459			 * sure it was dropped from the socket queue
2460			 *
2461			 * let's report a short read
2462			 */
2463			err = 0;
2464			break;
2465		}
2466
2467		/* Mark read part of skb as used */
2468		if (!(flags & MSG_PEEK)) {
2469			UNIXCB(skb).consumed += chunk;
2470
2471			sk_peek_offset_bwd(sk, chunk);
2472
2473			if (UNIXCB(skb).fp) {
2474				scm_stat_del(sk, skb);
2475				unix_detach_fds(&scm, skb);
2476			}
2477
2478			if (unix_skb_len(skb))
2479				break;
2480
2481			skb_unlink(skb, &sk->sk_receive_queue);
2482			consume_skb(skb);
2483
2484			if (scm.fp)
2485				break;
2486		} else {
2487			/* It is questionable, see note in unix_dgram_recvmsg.
2488			 */
2489			if (UNIXCB(skb).fp)
2490				unix_peek_fds(&scm, skb);
2491
2492			sk_peek_offset_fwd(sk, chunk);
2493
2494			if (UNIXCB(skb).fp)
2495				break;
2496
2497			skip = 0;
2498			last = skb;
2499			last_len = skb->len;
2500			unix_state_lock(sk);
2501			skb = skb_peek_next(skb, &sk->sk_receive_queue);
2502			if (skb)
2503				goto again;
2504			unix_state_unlock(sk);
2505			break;
2506		}
2507	} while (size);
2508
2509	mutex_unlock(&u->iolock);
2510	if (state->msg)
2511		scm_recv(sock, state->msg, &scm, flags);
2512	else
2513		scm_destroy(&scm);
2514out:
2515	return copied ? : err;
2516}
2517
2518static int unix_stream_read_actor(struct sk_buff *skb,
2519				  int skip, int chunk,
2520				  struct unix_stream_read_state *state)
2521{
2522	int ret;
2523
2524	ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2525				    state->msg, chunk);
2526	return ret ?: chunk;
2527}
2528
2529static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2530			       size_t size, int flags)
2531{
2532	struct unix_stream_read_state state = {
2533		.recv_actor = unix_stream_read_actor,
2534		.socket = sock,
2535		.msg = msg,
2536		.size = size,
2537		.flags = flags
2538	};
2539
2540	return unix_stream_read_generic(&state, true);
2541}
2542
2543static int unix_stream_splice_actor(struct sk_buff *skb,
2544				    int skip, int chunk,
2545				    struct unix_stream_read_state *state)
2546{
2547	return skb_splice_bits(skb, state->socket->sk,
2548			       UNIXCB(skb).consumed + skip,
2549			       state->pipe, chunk, state->splice_flags);
2550}
2551
2552static ssize_t unix_stream_splice_read(struct socket *sock,  loff_t *ppos,
2553				       struct pipe_inode_info *pipe,
2554				       size_t size, unsigned int flags)
2555{
2556	struct unix_stream_read_state state = {
2557		.recv_actor = unix_stream_splice_actor,
2558		.socket = sock,
2559		.pipe = pipe,
2560		.size = size,
2561		.splice_flags = flags,
2562	};
2563
2564	if (unlikely(*ppos))
2565		return -ESPIPE;
2566
2567	if (sock->file->f_flags & O_NONBLOCK ||
2568	    flags & SPLICE_F_NONBLOCK)
2569		state.flags = MSG_DONTWAIT;
2570
2571	return unix_stream_read_generic(&state, false);
2572}
2573
2574static int unix_shutdown(struct socket *sock, int mode)
2575{
2576	struct sock *sk = sock->sk;
2577	struct sock *other;
2578
2579	if (mode < SHUT_RD || mode > SHUT_RDWR)
2580		return -EINVAL;
2581	/* This maps:
2582	 * SHUT_RD   (0) -> RCV_SHUTDOWN  (1)
2583	 * SHUT_WR   (1) -> SEND_SHUTDOWN (2)
2584	 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2585	 */
2586	++mode;
2587
2588	unix_state_lock(sk);
2589	sk->sk_shutdown |= mode;
2590	other = unix_peer(sk);
2591	if (other)
2592		sock_hold(other);
2593	unix_state_unlock(sk);
2594	sk->sk_state_change(sk);
2595
2596	if (other &&
2597		(sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2598
2599		int peer_mode = 0;
2600
2601		if (mode&RCV_SHUTDOWN)
2602			peer_mode |= SEND_SHUTDOWN;
2603		if (mode&SEND_SHUTDOWN)
2604			peer_mode |= RCV_SHUTDOWN;
2605		unix_state_lock(other);
2606		other->sk_shutdown |= peer_mode;
2607		unix_state_unlock(other);
2608		other->sk_state_change(other);
2609		if (peer_mode == SHUTDOWN_MASK)
2610			sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2611		else if (peer_mode & RCV_SHUTDOWN)
2612			sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2613	}
2614	if (other)
2615		sock_put(other);
2616
2617	return 0;
2618}
2619
2620long unix_inq_len(struct sock *sk)
2621{
2622	struct sk_buff *skb;
2623	long amount = 0;
2624
2625	if (sk->sk_state == TCP_LISTEN)
2626		return -EINVAL;
2627
2628	spin_lock(&sk->sk_receive_queue.lock);
2629	if (sk->sk_type == SOCK_STREAM ||
2630	    sk->sk_type == SOCK_SEQPACKET) {
2631		skb_queue_walk(&sk->sk_receive_queue, skb)
2632			amount += unix_skb_len(skb);
2633	} else {
2634		skb = skb_peek(&sk->sk_receive_queue);
2635		if (skb)
2636			amount = skb->len;
2637	}
2638	spin_unlock(&sk->sk_receive_queue.lock);
2639
2640	return amount;
2641}
2642EXPORT_SYMBOL_GPL(unix_inq_len);
2643
2644long unix_outq_len(struct sock *sk)
2645{
2646	return sk_wmem_alloc_get(sk);
2647}
2648EXPORT_SYMBOL_GPL(unix_outq_len);
2649
2650static int unix_open_file(struct sock *sk)
2651{
2652	struct path path;
2653	struct file *f;
2654	int fd;
2655
2656	if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2657		return -EPERM;
2658
2659	if (!smp_load_acquire(&unix_sk(sk)->addr))
2660		return -ENOENT;
2661
2662	path = unix_sk(sk)->path;
2663	if (!path.dentry)
2664		return -ENOENT;
2665
2666	path_get(&path);
2667
2668	fd = get_unused_fd_flags(O_CLOEXEC);
2669	if (fd < 0)
2670		goto out;
2671
2672	f = dentry_open(&path, O_PATH, current_cred());
2673	if (IS_ERR(f)) {
2674		put_unused_fd(fd);
2675		fd = PTR_ERR(f);
2676		goto out;
2677	}
2678
2679	fd_install(fd, f);
2680out:
2681	path_put(&path);
2682
2683	return fd;
2684}
2685
2686static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2687{
2688	struct sock *sk = sock->sk;
2689	long amount = 0;
2690	int err;
2691
2692	switch (cmd) {
2693	case SIOCOUTQ:
2694		amount = unix_outq_len(sk);
2695		err = put_user(amount, (int __user *)arg);
2696		break;
2697	case SIOCINQ:
2698		amount = unix_inq_len(sk);
2699		if (amount < 0)
2700			err = amount;
2701		else
2702			err = put_user(amount, (int __user *)arg);
2703		break;
2704	case SIOCUNIXFILE:
2705		err = unix_open_file(sk);
2706		break;
2707	default:
2708		err = -ENOIOCTLCMD;
2709		break;
2710	}
2711	return err;
2712}
2713
2714#ifdef CONFIG_COMPAT
2715static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2716{
2717	return unix_ioctl(sock, cmd, (unsigned long)compat_ptr(arg));
2718}
2719#endif
2720
2721static __poll_t unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2722{
2723	struct sock *sk = sock->sk;
2724	__poll_t mask;
2725
2726	sock_poll_wait(file, sock, wait);
2727	mask = 0;
2728
2729	/* exceptional events? */
2730	if (sk->sk_err)
2731		mask |= EPOLLERR;
2732	if (sk->sk_shutdown == SHUTDOWN_MASK)
2733		mask |= EPOLLHUP;
2734	if (sk->sk_shutdown & RCV_SHUTDOWN)
2735		mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2736
2737	/* readable? */
2738	if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
2739		mask |= EPOLLIN | EPOLLRDNORM;
2740
2741	/* Connection-based need to check for termination and startup */
2742	if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2743	    sk->sk_state == TCP_CLOSE)
2744		mask |= EPOLLHUP;
2745
2746	/*
2747	 * we set writable also when the other side has shut down the
2748	 * connection. This prevents stuck sockets.
2749	 */
2750	if (unix_writable(sk))
2751		mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2752
2753	return mask;
2754}
2755
2756static __poll_t unix_dgram_poll(struct file *file, struct socket *sock,
2757				    poll_table *wait)
2758{
2759	struct sock *sk = sock->sk, *other;
2760	unsigned int writable;
2761	__poll_t mask;
2762
2763	sock_poll_wait(file, sock, wait);
2764	mask = 0;
2765
2766	/* exceptional events? */
2767	if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
2768		mask |= EPOLLERR |
2769			(sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
2770
2771	if (sk->sk_shutdown & RCV_SHUTDOWN)
2772		mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2773	if (sk->sk_shutdown == SHUTDOWN_MASK)
2774		mask |= EPOLLHUP;
2775
2776	/* readable? */
2777	if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
2778		mask |= EPOLLIN | EPOLLRDNORM;
2779
2780	/* Connection-based need to check for termination and startup */
2781	if (sk->sk_type == SOCK_SEQPACKET) {
2782		if (sk->sk_state == TCP_CLOSE)
2783			mask |= EPOLLHUP;
2784		/* connection hasn't started yet? */
2785		if (sk->sk_state == TCP_SYN_SENT)
2786			return mask;
2787	}
2788
2789	/* No write status requested, avoid expensive OUT tests. */
2790	if (!(poll_requested_events(wait) & (EPOLLWRBAND|EPOLLWRNORM|EPOLLOUT)))
2791		return mask;
2792
2793	writable = unix_writable(sk);
2794	if (writable) {
2795		unix_state_lock(sk);
2796
2797		other = unix_peer(sk);
2798		if (other && unix_peer(other) != sk &&
2799		    unix_recvq_full_lockless(other) &&
2800		    unix_dgram_peer_wake_me(sk, other))
2801			writable = 0;
2802
2803		unix_state_unlock(sk);
2804	}
2805
2806	if (writable)
2807		mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2808	else
2809		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2810
2811	return mask;
2812}
2813
2814#ifdef CONFIG_PROC_FS
2815
2816#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2817
2818#define get_bucket(x) ((x) >> BUCKET_SPACE)
2819#define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2820#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2821
2822static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2823{
2824	unsigned long offset = get_offset(*pos);
2825	unsigned long bucket = get_bucket(*pos);
2826	struct sock *sk;
2827	unsigned long count = 0;
2828
2829	for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2830		if (sock_net(sk) != seq_file_net(seq))
2831			continue;
2832		if (++count == offset)
2833			break;
2834	}
2835
2836	return sk;
2837}
2838
2839static struct sock *unix_next_socket(struct seq_file *seq,
2840				     struct sock *sk,
2841				     loff_t *pos)
2842{
2843	unsigned long bucket;
2844
2845	while (sk > (struct sock *)SEQ_START_TOKEN) {
2846		sk = sk_next(sk);
2847		if (!sk)
2848			goto next_bucket;
2849		if (sock_net(sk) == seq_file_net(seq))
2850			return sk;
2851	}
2852
2853	do {
2854		sk = unix_from_bucket(seq, pos);
2855		if (sk)
2856			return sk;
2857
2858next_bucket:
2859		bucket = get_bucket(*pos) + 1;
2860		*pos = set_bucket_offset(bucket, 1);
2861	} while (bucket < ARRAY_SIZE(unix_socket_table));
2862
2863	return NULL;
2864}
2865
2866static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2867	__acquires(unix_table_lock)
2868{
2869	spin_lock(&unix_table_lock);
2870
2871	if (!*pos)
2872		return SEQ_START_TOKEN;
2873
2874	if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2875		return NULL;
2876
2877	return unix_next_socket(seq, NULL, pos);
2878}
2879
2880static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2881{
2882	++*pos;
2883	return unix_next_socket(seq, v, pos);
2884}
2885
2886static void unix_seq_stop(struct seq_file *seq, void *v)
2887	__releases(unix_table_lock)
2888{
2889	spin_unlock(&unix_table_lock);
2890}
2891
2892static int unix_seq_show(struct seq_file *seq, void *v)
2893{
2894
2895	if (v == SEQ_START_TOKEN)
2896		seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2897			 "Inode Path\n");
2898	else {
2899		struct sock *s = v;
2900		struct unix_sock *u = unix_sk(s);
2901		unix_state_lock(s);
2902
2903		seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2904			s,
2905			refcount_read(&s->sk_refcnt),
2906			0,
2907			s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2908			s->sk_type,
2909			s->sk_socket ?
2910			(s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2911			(s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2912			sock_i_ino(s));
2913
2914		if (u->addr) {	// under unix_table_lock here
2915			int i, len;
2916			seq_putc(seq, ' ');
2917
2918			i = 0;
2919			len = u->addr->len - sizeof(short);
2920			if (!UNIX_ABSTRACT(s))
2921				len--;
2922			else {
2923				seq_putc(seq, '@');
2924				i++;
2925			}
2926			for ( ; i < len; i++)
2927				seq_putc(seq, u->addr->name->sun_path[i] ?:
2928					 '@');
2929		}
2930		unix_state_unlock(s);
2931		seq_putc(seq, '\n');
2932	}
2933
2934	return 0;
2935}
2936
2937static const struct seq_operations unix_seq_ops = {
2938	.start  = unix_seq_start,
2939	.next   = unix_seq_next,
2940	.stop   = unix_seq_stop,
2941	.show   = unix_seq_show,
2942};
2943#endif
2944
2945static const struct net_proto_family unix_family_ops = {
2946	.family = PF_UNIX,
2947	.create = unix_create,
2948	.owner	= THIS_MODULE,
2949};
2950
2951
2952static int __net_init unix_net_init(struct net *net)
2953{
2954	int error = -ENOMEM;
2955
2956	net->unx.sysctl_max_dgram_qlen = 10;
2957	if (unix_sysctl_register(net))
2958		goto out;
2959
2960#ifdef CONFIG_PROC_FS
2961	if (!proc_create_net("unix", 0, net->proc_net, &unix_seq_ops,
2962			sizeof(struct seq_net_private))) {
2963		unix_sysctl_unregister(net);
2964		goto out;
2965	}
2966#endif
2967	error = 0;
2968out:
2969	return error;
2970}
2971
2972static void __net_exit unix_net_exit(struct net *net)
2973{
2974	unix_sysctl_unregister(net);
2975	remove_proc_entry("unix", net->proc_net);
2976}
2977
2978static struct pernet_operations unix_net_ops = {
2979	.init = unix_net_init,
2980	.exit = unix_net_exit,
2981};
2982
2983static int __init af_unix_init(void)
2984{
2985	int rc = -1;
2986
2987	BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
2988
2989	rc = proto_register(&unix_proto, 1);
2990	if (rc != 0) {
2991		pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2992		goto out;
2993	}
2994
2995	sock_register(&unix_family_ops);
2996	register_pernet_subsys(&unix_net_ops);
2997out:
2998	return rc;
2999}
3000
3001static void __exit af_unix_exit(void)
3002{
3003	sock_unregister(PF_UNIX);
3004	proto_unregister(&unix_proto);
3005	unregister_pernet_subsys(&unix_net_ops);
3006}
3007
3008/* Earlier than device_initcall() so that other drivers invoking
3009   request_module() don't end up in a loop when modprobe tries
3010   to use a UNIX socket. But later than subsys_initcall() because
3011   we depend on stuff initialised there */
3012fs_initcall(af_unix_init);
3013module_exit(af_unix_exit);
3014
3015MODULE_LICENSE("GPL");
3016MODULE_ALIAS_NETPROTO(PF_UNIX);
v5.9
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * NET4:	Implementation of BSD Unix domain sockets.
   4 *
   5 * Authors:	Alan Cox, <alan@lxorguk.ukuu.org.uk>
   6 *
   7 * Fixes:
   8 *		Linus Torvalds	:	Assorted bug cures.
   9 *		Niibe Yutaka	:	async I/O support.
  10 *		Carsten Paeth	:	PF_UNIX check, address fixes.
  11 *		Alan Cox	:	Limit size of allocated blocks.
  12 *		Alan Cox	:	Fixed the stupid socketpair bug.
  13 *		Alan Cox	:	BSD compatibility fine tuning.
  14 *		Alan Cox	:	Fixed a bug in connect when interrupted.
  15 *		Alan Cox	:	Sorted out a proper draft version of
  16 *					file descriptor passing hacked up from
  17 *					Mike Shaver's work.
  18 *		Marty Leisner	:	Fixes to fd passing
  19 *		Nick Nevin	:	recvmsg bugfix.
  20 *		Alan Cox	:	Started proper garbage collector
  21 *		Heiko EiBfeldt	:	Missing verify_area check
  22 *		Alan Cox	:	Started POSIXisms
  23 *		Andreas Schwab	:	Replace inode by dentry for proper
  24 *					reference counting
  25 *		Kirk Petersen	:	Made this a module
  26 *	    Christoph Rohland	:	Elegant non-blocking accept/connect algorithm.
  27 *					Lots of bug fixes.
  28 *	     Alexey Kuznetosv	:	Repaired (I hope) bugs introduces
  29 *					by above two patches.
  30 *	     Andrea Arcangeli	:	If possible we block in connect(2)
  31 *					if the max backlog of the listen socket
  32 *					is been reached. This won't break
  33 *					old apps and it will avoid huge amount
  34 *					of socks hashed (this for unix_gc()
  35 *					performances reasons).
  36 *					Security fix that limits the max
  37 *					number of socks to 2*max_files and
  38 *					the number of skb queueable in the
  39 *					dgram receiver.
  40 *		Artur Skawina   :	Hash function optimizations
  41 *	     Alexey Kuznetsov   :	Full scale SMP. Lot of bugs are introduced 8)
  42 *	      Malcolm Beattie   :	Set peercred for socketpair
  43 *	     Michal Ostrowski   :       Module initialization cleanup.
  44 *	     Arnaldo C. Melo	:	Remove MOD_{INC,DEC}_USE_COUNT,
  45 *	     				the core infrastructure is doing that
  46 *	     				for all net proto families now (2.5.69+)
  47 *
  48 * Known differences from reference BSD that was tested:
  49 *
  50 *	[TO FIX]
  51 *	ECONNREFUSED is not returned from one end of a connected() socket to the
  52 *		other the moment one end closes.
  53 *	fstat() doesn't return st_dev=0, and give the blksize as high water mark
  54 *		and a fake inode identifier (nor the BSD first socket fstat twice bug).
  55 *	[NOT TO FIX]
  56 *	accept() returns a path name even if the connecting socket has closed
  57 *		in the meantime (BSD loses the path and gives up).
  58 *	accept() returns 0 length path for an unbound connector. BSD returns 16
  59 *		and a null first byte in the path (but not for gethost/peername - BSD bug ??)
  60 *	socketpair(...SOCK_RAW..) doesn't panic the kernel.
  61 *	BSD af_unix apparently has connect forgetting to block properly.
  62 *		(need to check this with the POSIX spec in detail)
  63 *
  64 * Differences from 2.0.0-11-... (ANK)
  65 *	Bug fixes and improvements.
  66 *		- client shutdown killed server socket.
  67 *		- removed all useless cli/sti pairs.
  68 *
  69 *	Semantic changes/extensions.
  70 *		- generic control message passing.
  71 *		- SCM_CREDENTIALS control message.
  72 *		- "Abstract" (not FS based) socket bindings.
  73 *		  Abstract names are sequences of bytes (not zero terminated)
  74 *		  started by 0, so that this name space does not intersect
  75 *		  with BSD names.
  76 */
  77
  78#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  79
  80#include <linux/module.h>
  81#include <linux/kernel.h>
  82#include <linux/signal.h>
  83#include <linux/sched/signal.h>
  84#include <linux/errno.h>
  85#include <linux/string.h>
  86#include <linux/stat.h>
  87#include <linux/dcache.h>
  88#include <linux/namei.h>
  89#include <linux/socket.h>
  90#include <linux/un.h>
  91#include <linux/fcntl.h>
  92#include <linux/termios.h>
  93#include <linux/sockios.h>
  94#include <linux/net.h>
  95#include <linux/in.h>
  96#include <linux/fs.h>
  97#include <linux/slab.h>
  98#include <linux/uaccess.h>
  99#include <linux/skbuff.h>
 100#include <linux/netdevice.h>
 101#include <net/net_namespace.h>
 102#include <net/sock.h>
 103#include <net/tcp_states.h>
 104#include <net/af_unix.h>
 105#include <linux/proc_fs.h>
 106#include <linux/seq_file.h>
 107#include <net/scm.h>
 108#include <linux/init.h>
 109#include <linux/poll.h>
 110#include <linux/rtnetlink.h>
 111#include <linux/mount.h>
 112#include <net/checksum.h>
 113#include <linux/security.h>
 114#include <linux/freezer.h>
 115#include <linux/file.h>
 116
 117#include "scm.h"
 118
 119struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
 120EXPORT_SYMBOL_GPL(unix_socket_table);
 121DEFINE_SPINLOCK(unix_table_lock);
 122EXPORT_SYMBOL_GPL(unix_table_lock);
 123static atomic_long_t unix_nr_socks;
 124
 125
 126static struct hlist_head *unix_sockets_unbound(void *addr)
 127{
 128	unsigned long hash = (unsigned long)addr;
 129
 130	hash ^= hash >> 16;
 131	hash ^= hash >> 8;
 132	hash %= UNIX_HASH_SIZE;
 133	return &unix_socket_table[UNIX_HASH_SIZE + hash];
 134}
 135
 136#define UNIX_ABSTRACT(sk)	(unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
 137
 138#ifdef CONFIG_SECURITY_NETWORK
 139static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 140{
 141	UNIXCB(skb).secid = scm->secid;
 142}
 143
 144static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 145{
 146	scm->secid = UNIXCB(skb).secid;
 147}
 148
 149static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
 150{
 151	return (scm->secid == UNIXCB(skb).secid);
 152}
 153#else
 154static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 155{ }
 156
 157static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 158{ }
 159
 160static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
 161{
 162	return true;
 163}
 164#endif /* CONFIG_SECURITY_NETWORK */
 165
 166/*
 167 *  SMP locking strategy:
 168 *    hash table is protected with spinlock unix_table_lock
 169 *    each socket state is protected by separate spin lock.
 170 */
 171
 172static inline unsigned int unix_hash_fold(__wsum n)
 173{
 174	unsigned int hash = (__force unsigned int)csum_fold(n);
 175
 176	hash ^= hash>>8;
 177	return hash&(UNIX_HASH_SIZE-1);
 178}
 179
 180#define unix_peer(sk) (unix_sk(sk)->peer)
 181
 182static inline int unix_our_peer(struct sock *sk, struct sock *osk)
 183{
 184	return unix_peer(osk) == sk;
 185}
 186
 187static inline int unix_may_send(struct sock *sk, struct sock *osk)
 188{
 189	return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
 190}
 191
 192static inline int unix_recvq_full(const struct sock *sk)
 193{
 194	return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
 195}
 196
 197static inline int unix_recvq_full_lockless(const struct sock *sk)
 198{
 199	return skb_queue_len_lockless(&sk->sk_receive_queue) >
 200		READ_ONCE(sk->sk_max_ack_backlog);
 201}
 202
 203struct sock *unix_peer_get(struct sock *s)
 204{
 205	struct sock *peer;
 206
 207	unix_state_lock(s);
 208	peer = unix_peer(s);
 209	if (peer)
 210		sock_hold(peer);
 211	unix_state_unlock(s);
 212	return peer;
 213}
 214EXPORT_SYMBOL_GPL(unix_peer_get);
 215
 216static inline void unix_release_addr(struct unix_address *addr)
 217{
 218	if (refcount_dec_and_test(&addr->refcnt))
 219		kfree(addr);
 220}
 221
 222/*
 223 *	Check unix socket name:
 224 *		- should be not zero length.
 225 *	        - if started by not zero, should be NULL terminated (FS object)
 226 *		- if started by zero, it is abstract name.
 227 */
 228
 229static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
 230{
 231	*hashp = 0;
 232
 233	if (len <= sizeof(short) || len > sizeof(*sunaddr))
 234		return -EINVAL;
 235	if (!sunaddr || sunaddr->sun_family != AF_UNIX)
 236		return -EINVAL;
 237	if (sunaddr->sun_path[0]) {
 238		/*
 239		 * This may look like an off by one error but it is a bit more
 240		 * subtle. 108 is the longest valid AF_UNIX path for a binding.
 241		 * sun_path[108] doesn't as such exist.  However in kernel space
 242		 * we are guaranteed that it is a valid memory location in our
 243		 * kernel address buffer.
 244		 */
 245		((char *)sunaddr)[len] = 0;
 246		len = strlen(sunaddr->sun_path)+1+sizeof(short);
 247		return len;
 248	}
 249
 250	*hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
 251	return len;
 252}
 253
 254static void __unix_remove_socket(struct sock *sk)
 255{
 256	sk_del_node_init(sk);
 257}
 258
 259static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
 260{
 261	WARN_ON(!sk_unhashed(sk));
 262	sk_add_node(sk, list);
 263}
 264
 
 
 
 
 
 
 
 
 265static inline void unix_remove_socket(struct sock *sk)
 266{
 267	spin_lock(&unix_table_lock);
 268	__unix_remove_socket(sk);
 269	spin_unlock(&unix_table_lock);
 270}
 271
 272static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
 273{
 274	spin_lock(&unix_table_lock);
 275	__unix_insert_socket(list, sk);
 276	spin_unlock(&unix_table_lock);
 277}
 278
 279static struct sock *__unix_find_socket_byname(struct net *net,
 280					      struct sockaddr_un *sunname,
 281					      int len, int type, unsigned int hash)
 282{
 283	struct sock *s;
 284
 285	sk_for_each(s, &unix_socket_table[hash ^ type]) {
 286		struct unix_sock *u = unix_sk(s);
 287
 288		if (!net_eq(sock_net(s), net))
 289			continue;
 290
 291		if (u->addr->len == len &&
 292		    !memcmp(u->addr->name, sunname, len))
 293			return s;
 294	}
 295	return NULL;
 296}
 297
 298static inline struct sock *unix_find_socket_byname(struct net *net,
 299						   struct sockaddr_un *sunname,
 300						   int len, int type,
 301						   unsigned int hash)
 302{
 303	struct sock *s;
 304
 305	spin_lock(&unix_table_lock);
 306	s = __unix_find_socket_byname(net, sunname, len, type, hash);
 307	if (s)
 308		sock_hold(s);
 309	spin_unlock(&unix_table_lock);
 310	return s;
 311}
 312
 313static struct sock *unix_find_socket_byinode(struct inode *i)
 314{
 315	struct sock *s;
 316
 317	spin_lock(&unix_table_lock);
 318	sk_for_each(s,
 319		    &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
 320		struct dentry *dentry = unix_sk(s)->path.dentry;
 321
 322		if (dentry && d_backing_inode(dentry) == i) {
 323			sock_hold(s);
 324			goto found;
 325		}
 326	}
 327	s = NULL;
 328found:
 329	spin_unlock(&unix_table_lock);
 330	return s;
 331}
 332
 333/* Support code for asymmetrically connected dgram sockets
 334 *
 335 * If a datagram socket is connected to a socket not itself connected
 336 * to the first socket (eg, /dev/log), clients may only enqueue more
 337 * messages if the present receive queue of the server socket is not
 338 * "too large". This means there's a second writeability condition
 339 * poll and sendmsg need to test. The dgram recv code will do a wake
 340 * up on the peer_wait wait queue of a socket upon reception of a
 341 * datagram which needs to be propagated to sleeping would-be writers
 342 * since these might not have sent anything so far. This can't be
 343 * accomplished via poll_wait because the lifetime of the server
 344 * socket might be less than that of its clients if these break their
 345 * association with it or if the server socket is closed while clients
 346 * are still connected to it and there's no way to inform "a polling
 347 * implementation" that it should let go of a certain wait queue
 348 *
 349 * In order to propagate a wake up, a wait_queue_entry_t of the client
 350 * socket is enqueued on the peer_wait queue of the server socket
 351 * whose wake function does a wake_up on the ordinary client socket
 352 * wait queue. This connection is established whenever a write (or
 353 * poll for write) hit the flow control condition and broken when the
 354 * association to the server socket is dissolved or after a wake up
 355 * was relayed.
 356 */
 357
 358static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
 359				      void *key)
 360{
 361	struct unix_sock *u;
 362	wait_queue_head_t *u_sleep;
 363
 364	u = container_of(q, struct unix_sock, peer_wake);
 365
 366	__remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
 367			    q);
 368	u->peer_wake.private = NULL;
 369
 370	/* relaying can only happen while the wq still exists */
 371	u_sleep = sk_sleep(&u->sk);
 372	if (u_sleep)
 373		wake_up_interruptible_poll(u_sleep, key_to_poll(key));
 374
 375	return 0;
 376}
 377
 378static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
 379{
 380	struct unix_sock *u, *u_other;
 381	int rc;
 382
 383	u = unix_sk(sk);
 384	u_other = unix_sk(other);
 385	rc = 0;
 386	spin_lock(&u_other->peer_wait.lock);
 387
 388	if (!u->peer_wake.private) {
 389		u->peer_wake.private = other;
 390		__add_wait_queue(&u_other->peer_wait, &u->peer_wake);
 391
 392		rc = 1;
 393	}
 394
 395	spin_unlock(&u_other->peer_wait.lock);
 396	return rc;
 397}
 398
 399static void unix_dgram_peer_wake_disconnect(struct sock *sk,
 400					    struct sock *other)
 401{
 402	struct unix_sock *u, *u_other;
 403
 404	u = unix_sk(sk);
 405	u_other = unix_sk(other);
 406	spin_lock(&u_other->peer_wait.lock);
 407
 408	if (u->peer_wake.private == other) {
 409		__remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
 410		u->peer_wake.private = NULL;
 411	}
 412
 413	spin_unlock(&u_other->peer_wait.lock);
 414}
 415
 416static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
 417						   struct sock *other)
 418{
 419	unix_dgram_peer_wake_disconnect(sk, other);
 420	wake_up_interruptible_poll(sk_sleep(sk),
 421				   EPOLLOUT |
 422				   EPOLLWRNORM |
 423				   EPOLLWRBAND);
 424}
 425
 426/* preconditions:
 427 *	- unix_peer(sk) == other
 428 *	- association is stable
 429 */
 430static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
 431{
 432	int connected;
 433
 434	connected = unix_dgram_peer_wake_connect(sk, other);
 435
 436	/* If other is SOCK_DEAD, we want to make sure we signal
 437	 * POLLOUT, such that a subsequent write() can get a
 438	 * -ECONNREFUSED. Otherwise, if we haven't queued any skbs
 439	 * to other and its full, we will hang waiting for POLLOUT.
 440	 */
 441	if (unix_recvq_full(other) && !sock_flag(other, SOCK_DEAD))
 442		return 1;
 443
 444	if (connected)
 445		unix_dgram_peer_wake_disconnect(sk, other);
 446
 447	return 0;
 448}
 449
 450static int unix_writable(const struct sock *sk)
 451{
 452	return sk->sk_state != TCP_LISTEN &&
 453	       (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
 454}
 455
 456static void unix_write_space(struct sock *sk)
 457{
 458	struct socket_wq *wq;
 459
 460	rcu_read_lock();
 461	if (unix_writable(sk)) {
 462		wq = rcu_dereference(sk->sk_wq);
 463		if (skwq_has_sleeper(wq))
 464			wake_up_interruptible_sync_poll(&wq->wait,
 465				EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
 466		sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
 467	}
 468	rcu_read_unlock();
 469}
 470
 471/* When dgram socket disconnects (or changes its peer), we clear its receive
 472 * queue of packets arrived from previous peer. First, it allows to do
 473 * flow control based only on wmem_alloc; second, sk connected to peer
 474 * may receive messages only from that peer. */
 475static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
 476{
 477	if (!skb_queue_empty(&sk->sk_receive_queue)) {
 478		skb_queue_purge(&sk->sk_receive_queue);
 479		wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
 480
 481		/* If one link of bidirectional dgram pipe is disconnected,
 482		 * we signal error. Messages are lost. Do not make this,
 483		 * when peer was not connected to us.
 484		 */
 485		if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
 486			other->sk_err = ECONNRESET;
 487			other->sk_error_report(other);
 488		}
 489	}
 490}
 491
 492static void unix_sock_destructor(struct sock *sk)
 493{
 494	struct unix_sock *u = unix_sk(sk);
 495
 496	skb_queue_purge(&sk->sk_receive_queue);
 497
 498	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
 499	WARN_ON(!sk_unhashed(sk));
 500	WARN_ON(sk->sk_socket);
 501	if (!sock_flag(sk, SOCK_DEAD)) {
 502		pr_info("Attempt to release alive unix socket: %p\n", sk);
 503		return;
 504	}
 505
 506	if (u->addr)
 507		unix_release_addr(u->addr);
 508
 509	atomic_long_dec(&unix_nr_socks);
 510	local_bh_disable();
 511	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 512	local_bh_enable();
 513#ifdef UNIX_REFCNT_DEBUG
 514	pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
 515		atomic_long_read(&unix_nr_socks));
 516#endif
 517}
 518
 519static void unix_release_sock(struct sock *sk, int embrion)
 520{
 521	struct unix_sock *u = unix_sk(sk);
 522	struct path path;
 523	struct sock *skpair;
 524	struct sk_buff *skb;
 525	int state;
 526
 527	unix_remove_socket(sk);
 528
 529	/* Clear state */
 530	unix_state_lock(sk);
 531	sock_orphan(sk);
 532	sk->sk_shutdown = SHUTDOWN_MASK;
 533	path	     = u->path;
 534	u->path.dentry = NULL;
 535	u->path.mnt = NULL;
 536	state = sk->sk_state;
 537	sk->sk_state = TCP_CLOSE;
 
 
 
 
 538	unix_state_unlock(sk);
 539
 540	wake_up_interruptible_all(&u->peer_wait);
 541
 542	skpair = unix_peer(sk);
 543
 544	if (skpair != NULL) {
 545		if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
 546			unix_state_lock(skpair);
 547			/* No more writes */
 548			skpair->sk_shutdown = SHUTDOWN_MASK;
 549			if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
 550				skpair->sk_err = ECONNRESET;
 551			unix_state_unlock(skpair);
 552			skpair->sk_state_change(skpair);
 553			sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
 554		}
 555
 556		unix_dgram_peer_wake_disconnect(sk, skpair);
 557		sock_put(skpair); /* It may now die */
 558		unix_peer(sk) = NULL;
 559	}
 560
 561	/* Try to flush out this socket. Throw out buffers at least */
 562
 563	while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 564		if (state == TCP_LISTEN)
 565			unix_release_sock(skb->sk, 1);
 566		/* passed fds are erased in the kfree_skb hook	      */
 567		UNIXCB(skb).consumed = skb->len;
 568		kfree_skb(skb);
 569	}
 570
 571	if (path.dentry)
 572		path_put(&path);
 573
 574	sock_put(sk);
 575
 576	/* ---- Socket is dead now and most probably destroyed ---- */
 577
 578	/*
 579	 * Fixme: BSD difference: In BSD all sockets connected to us get
 580	 *	  ECONNRESET and we die on the spot. In Linux we behave
 581	 *	  like files and pipes do and wait for the last
 582	 *	  dereference.
 583	 *
 584	 * Can't we simply set sock->err?
 585	 *
 586	 *	  What the above comment does talk about? --ANK(980817)
 587	 */
 588
 589	if (unix_tot_inflight)
 590		unix_gc();		/* Garbage collect fds */
 591}
 592
 593static void init_peercred(struct sock *sk)
 594{
 595	put_pid(sk->sk_peer_pid);
 596	if (sk->sk_peer_cred)
 597		put_cred(sk->sk_peer_cred);
 
 
 
 598	sk->sk_peer_pid  = get_pid(task_tgid(current));
 599	sk->sk_peer_cred = get_current_cred();
 
 
 
 
 600}
 601
 602static void copy_peercred(struct sock *sk, struct sock *peersk)
 603{
 604	put_pid(sk->sk_peer_pid);
 605	if (sk->sk_peer_cred)
 606		put_cred(sk->sk_peer_cred);
 
 
 
 
 
 
 
 
 
 607	sk->sk_peer_pid  = get_pid(peersk->sk_peer_pid);
 608	sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
 
 
 
 
 
 
 609}
 610
 611static int unix_listen(struct socket *sock, int backlog)
 612{
 613	int err;
 614	struct sock *sk = sock->sk;
 615	struct unix_sock *u = unix_sk(sk);
 616	struct pid *old_pid = NULL;
 617
 618	err = -EOPNOTSUPP;
 619	if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
 620		goto out;	/* Only stream/seqpacket sockets accept */
 621	err = -EINVAL;
 622	if (!u->addr)
 623		goto out;	/* No listens on an unbound socket */
 624	unix_state_lock(sk);
 625	if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
 626		goto out_unlock;
 627	if (backlog > sk->sk_max_ack_backlog)
 628		wake_up_interruptible_all(&u->peer_wait);
 629	sk->sk_max_ack_backlog	= backlog;
 630	sk->sk_state		= TCP_LISTEN;
 631	/* set credentials so connect can copy them */
 632	init_peercred(sk);
 633	err = 0;
 634
 635out_unlock:
 636	unix_state_unlock(sk);
 637	put_pid(old_pid);
 638out:
 639	return err;
 640}
 641
 642static int unix_release(struct socket *);
 643static int unix_bind(struct socket *, struct sockaddr *, int);
 644static int unix_stream_connect(struct socket *, struct sockaddr *,
 645			       int addr_len, int flags);
 646static int unix_socketpair(struct socket *, struct socket *);
 647static int unix_accept(struct socket *, struct socket *, int, bool);
 648static int unix_getname(struct socket *, struct sockaddr *, int);
 649static __poll_t unix_poll(struct file *, struct socket *, poll_table *);
 650static __poll_t unix_dgram_poll(struct file *, struct socket *,
 651				    poll_table *);
 652static int unix_ioctl(struct socket *, unsigned int, unsigned long);
 653#ifdef CONFIG_COMPAT
 654static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg);
 655#endif
 656static int unix_shutdown(struct socket *, int);
 657static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
 658static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
 659static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
 660				    size_t size, int flags);
 661static ssize_t unix_stream_splice_read(struct socket *,  loff_t *ppos,
 662				       struct pipe_inode_info *, size_t size,
 663				       unsigned int flags);
 664static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
 665static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
 666static int unix_dgram_connect(struct socket *, struct sockaddr *,
 667			      int, int);
 668static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
 669static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
 670				  int);
 671
 672static int unix_set_peek_off(struct sock *sk, int val)
 673{
 674	struct unix_sock *u = unix_sk(sk);
 675
 676	if (mutex_lock_interruptible(&u->iolock))
 677		return -EINTR;
 678
 679	sk->sk_peek_off = val;
 680	mutex_unlock(&u->iolock);
 681
 682	return 0;
 683}
 684
 685#ifdef CONFIG_PROC_FS
 686static void unix_show_fdinfo(struct seq_file *m, struct socket *sock)
 687{
 688	struct sock *sk = sock->sk;
 689	struct unix_sock *u;
 690
 691	if (sk) {
 692		u = unix_sk(sock->sk);
 693		seq_printf(m, "scm_fds: %u\n",
 694			   atomic_read(&u->scm_stat.nr_fds));
 695	}
 696}
 697#else
 698#define unix_show_fdinfo NULL
 699#endif
 700
 701static const struct proto_ops unix_stream_ops = {
 702	.family =	PF_UNIX,
 703	.owner =	THIS_MODULE,
 704	.release =	unix_release,
 705	.bind =		unix_bind,
 706	.connect =	unix_stream_connect,
 707	.socketpair =	unix_socketpair,
 708	.accept =	unix_accept,
 709	.getname =	unix_getname,
 710	.poll =		unix_poll,
 711	.ioctl =	unix_ioctl,
 712#ifdef CONFIG_COMPAT
 713	.compat_ioctl =	unix_compat_ioctl,
 714#endif
 715	.listen =	unix_listen,
 716	.shutdown =	unix_shutdown,
 717	.sendmsg =	unix_stream_sendmsg,
 718	.recvmsg =	unix_stream_recvmsg,
 719	.mmap =		sock_no_mmap,
 720	.sendpage =	unix_stream_sendpage,
 721	.splice_read =	unix_stream_splice_read,
 722	.set_peek_off =	unix_set_peek_off,
 723	.show_fdinfo =	unix_show_fdinfo,
 724};
 725
 726static const struct proto_ops unix_dgram_ops = {
 727	.family =	PF_UNIX,
 728	.owner =	THIS_MODULE,
 729	.release =	unix_release,
 730	.bind =		unix_bind,
 731	.connect =	unix_dgram_connect,
 732	.socketpair =	unix_socketpair,
 733	.accept =	sock_no_accept,
 734	.getname =	unix_getname,
 735	.poll =		unix_dgram_poll,
 736	.ioctl =	unix_ioctl,
 737#ifdef CONFIG_COMPAT
 738	.compat_ioctl =	unix_compat_ioctl,
 739#endif
 740	.listen =	sock_no_listen,
 741	.shutdown =	unix_shutdown,
 742	.sendmsg =	unix_dgram_sendmsg,
 743	.recvmsg =	unix_dgram_recvmsg,
 744	.mmap =		sock_no_mmap,
 745	.sendpage =	sock_no_sendpage,
 746	.set_peek_off =	unix_set_peek_off,
 747	.show_fdinfo =	unix_show_fdinfo,
 748};
 749
 750static const struct proto_ops unix_seqpacket_ops = {
 751	.family =	PF_UNIX,
 752	.owner =	THIS_MODULE,
 753	.release =	unix_release,
 754	.bind =		unix_bind,
 755	.connect =	unix_stream_connect,
 756	.socketpair =	unix_socketpair,
 757	.accept =	unix_accept,
 758	.getname =	unix_getname,
 759	.poll =		unix_dgram_poll,
 760	.ioctl =	unix_ioctl,
 761#ifdef CONFIG_COMPAT
 762	.compat_ioctl =	unix_compat_ioctl,
 763#endif
 764	.listen =	unix_listen,
 765	.shutdown =	unix_shutdown,
 766	.sendmsg =	unix_seqpacket_sendmsg,
 767	.recvmsg =	unix_seqpacket_recvmsg,
 768	.mmap =		sock_no_mmap,
 769	.sendpage =	sock_no_sendpage,
 770	.set_peek_off =	unix_set_peek_off,
 771	.show_fdinfo =	unix_show_fdinfo,
 772};
 773
 774static struct proto unix_proto = {
 775	.name			= "UNIX",
 776	.owner			= THIS_MODULE,
 777	.obj_size		= sizeof(struct unix_sock),
 778};
 779
 780static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
 781{
 782	struct sock *sk = NULL;
 783	struct unix_sock *u;
 784
 785	atomic_long_inc(&unix_nr_socks);
 786	if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
 787		goto out;
 788
 789	sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
 790	if (!sk)
 791		goto out;
 792
 793	sock_init_data(sock, sk);
 794
 795	sk->sk_allocation	= GFP_KERNEL_ACCOUNT;
 796	sk->sk_write_space	= unix_write_space;
 797	sk->sk_max_ack_backlog	= net->unx.sysctl_max_dgram_qlen;
 798	sk->sk_destruct		= unix_sock_destructor;
 799	u	  = unix_sk(sk);
 800	u->path.dentry = NULL;
 801	u->path.mnt = NULL;
 802	spin_lock_init(&u->lock);
 803	atomic_long_set(&u->inflight, 0);
 804	INIT_LIST_HEAD(&u->link);
 805	mutex_init(&u->iolock); /* single task reading lock */
 806	mutex_init(&u->bindlock); /* single task binding lock */
 807	init_waitqueue_head(&u->peer_wait);
 808	init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
 809	memset(&u->scm_stat, 0, sizeof(struct scm_stat));
 810	unix_insert_socket(unix_sockets_unbound(sk), sk);
 811out:
 812	if (sk == NULL)
 813		atomic_long_dec(&unix_nr_socks);
 814	else {
 815		local_bh_disable();
 816		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 817		local_bh_enable();
 818	}
 819	return sk;
 820}
 821
 822static int unix_create(struct net *net, struct socket *sock, int protocol,
 823		       int kern)
 824{
 825	if (protocol && protocol != PF_UNIX)
 826		return -EPROTONOSUPPORT;
 827
 828	sock->state = SS_UNCONNECTED;
 829
 830	switch (sock->type) {
 831	case SOCK_STREAM:
 832		sock->ops = &unix_stream_ops;
 833		break;
 834		/*
 835		 *	Believe it or not BSD has AF_UNIX, SOCK_RAW though
 836		 *	nothing uses it.
 837		 */
 838	case SOCK_RAW:
 839		sock->type = SOCK_DGRAM;
 840		fallthrough;
 841	case SOCK_DGRAM:
 842		sock->ops = &unix_dgram_ops;
 843		break;
 844	case SOCK_SEQPACKET:
 845		sock->ops = &unix_seqpacket_ops;
 846		break;
 847	default:
 848		return -ESOCKTNOSUPPORT;
 849	}
 850
 851	return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
 852}
 853
 854static int unix_release(struct socket *sock)
 855{
 856	struct sock *sk = sock->sk;
 857
 858	if (!sk)
 859		return 0;
 860
 861	unix_release_sock(sk, 0);
 862	sock->sk = NULL;
 863
 864	return 0;
 865}
 866
 867static int unix_autobind(struct socket *sock)
 868{
 869	struct sock *sk = sock->sk;
 870	struct net *net = sock_net(sk);
 871	struct unix_sock *u = unix_sk(sk);
 872	static u32 ordernum = 1;
 873	struct unix_address *addr;
 874	int err;
 875	unsigned int retries = 0;
 876
 877	err = mutex_lock_interruptible(&u->bindlock);
 878	if (err)
 879		return err;
 880
 881	err = 0;
 882	if (u->addr)
 883		goto out;
 884
 885	err = -ENOMEM;
 886	addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 887	if (!addr)
 888		goto out;
 889
 890	addr->name->sun_family = AF_UNIX;
 891	refcount_set(&addr->refcnt, 1);
 892
 893retry:
 894	addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 895	addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
 
 896
 897	spin_lock(&unix_table_lock);
 898	ordernum = (ordernum+1)&0xFFFFF;
 899
 900	if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
 901				      addr->hash)) {
 902		spin_unlock(&unix_table_lock);
 903		/*
 904		 * __unix_find_socket_byname() may take long time if many names
 905		 * are already in use.
 906		 */
 907		cond_resched();
 908		/* Give up if all names seems to be in use. */
 909		if (retries++ == 0xFFFFF) {
 910			err = -ENOSPC;
 911			kfree(addr);
 912			goto out;
 913		}
 914		goto retry;
 915	}
 916	addr->hash ^= sk->sk_type;
 917
 918	__unix_remove_socket(sk);
 919	smp_store_release(&u->addr, addr);
 920	__unix_insert_socket(&unix_socket_table[addr->hash], sk);
 921	spin_unlock(&unix_table_lock);
 922	err = 0;
 923
 924out:	mutex_unlock(&u->bindlock);
 925	return err;
 926}
 927
 928static struct sock *unix_find_other(struct net *net,
 929				    struct sockaddr_un *sunname, int len,
 930				    int type, unsigned int hash, int *error)
 931{
 932	struct sock *u;
 933	struct path path;
 934	int err = 0;
 935
 936	if (sunname->sun_path[0]) {
 937		struct inode *inode;
 938		err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
 939		if (err)
 940			goto fail;
 941		inode = d_backing_inode(path.dentry);
 942		err = inode_permission(inode, MAY_WRITE);
 943		if (err)
 944			goto put_fail;
 945
 946		err = -ECONNREFUSED;
 947		if (!S_ISSOCK(inode->i_mode))
 948			goto put_fail;
 949		u = unix_find_socket_byinode(inode);
 950		if (!u)
 951			goto put_fail;
 952
 953		if (u->sk_type == type)
 954			touch_atime(&path);
 955
 956		path_put(&path);
 957
 958		err = -EPROTOTYPE;
 959		if (u->sk_type != type) {
 960			sock_put(u);
 961			goto fail;
 962		}
 963	} else {
 964		err = -ECONNREFUSED;
 965		u = unix_find_socket_byname(net, sunname, len, type, hash);
 966		if (u) {
 967			struct dentry *dentry;
 968			dentry = unix_sk(u)->path.dentry;
 969			if (dentry)
 970				touch_atime(&unix_sk(u)->path);
 971		} else
 972			goto fail;
 973	}
 974	return u;
 975
 976put_fail:
 977	path_put(&path);
 978fail:
 979	*error = err;
 980	return NULL;
 981}
 982
 983static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
 984{
 
 
 
 
 
 985	struct dentry *dentry;
 986	struct path path;
 987	int err = 0;
 
 988	/*
 989	 * Get the parent directory, calculate the hash for last
 990	 * component.
 991	 */
 992	dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
 993	err = PTR_ERR(dentry);
 994	if (IS_ERR(dentry))
 995		return err;
 
 996
 997	/*
 998	 * All right, let's create it.
 999	 */
1000	err = security_path_mknod(&path, dentry, mode, 0);
1001	if (!err) {
1002		err = vfs_mknod(d_inode(path.dentry), dentry, mode, 0);
1003		if (!err) {
1004			res->mnt = mntget(path.mnt);
1005			res->dentry = dget(dentry);
1006		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1007	}
1008	done_path_create(&path, dentry);
1009	return err;
 
 
1010}
1011
1012static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1013{
1014	struct sock *sk = sock->sk;
1015	struct net *net = sock_net(sk);
1016	struct unix_sock *u = unix_sk(sk);
1017	struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1018	char *sun_path = sunaddr->sun_path;
1019	int err;
1020	unsigned int hash;
1021	struct unix_address *addr;
1022	struct hlist_head *list;
1023	struct path path = { };
1024
1025	err = -EINVAL;
1026	if (addr_len < offsetofend(struct sockaddr_un, sun_family) ||
1027	    sunaddr->sun_family != AF_UNIX)
1028		goto out;
1029
1030	if (addr_len == sizeof(short)) {
1031		err = unix_autobind(sock);
1032		goto out;
1033	}
1034
1035	err = unix_mkname(sunaddr, addr_len, &hash);
1036	if (err < 0)
1037		goto out;
1038	addr_len = err;
1039
1040	if (sun_path[0]) {
1041		umode_t mode = S_IFSOCK |
1042		       (SOCK_INODE(sock)->i_mode & ~current_umask());
1043		err = unix_mknod(sun_path, mode, &path);
1044		if (err) {
1045			if (err == -EEXIST)
1046				err = -EADDRINUSE;
1047			goto out;
1048		}
1049	}
1050
1051	err = mutex_lock_interruptible(&u->bindlock);
1052	if (err)
1053		goto out_put;
1054
1055	err = -EINVAL;
1056	if (u->addr)
1057		goto out_up;
1058
1059	err = -ENOMEM;
1060	addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1061	if (!addr)
1062		goto out_up;
1063
1064	memcpy(addr->name, sunaddr, addr_len);
1065	addr->len = addr_len;
1066	addr->hash = hash ^ sk->sk_type;
1067	refcount_set(&addr->refcnt, 1);
1068
1069	if (sun_path[0]) {
1070		addr->hash = UNIX_HASH_SIZE;
1071		hash = d_backing_inode(path.dentry)->i_ino & (UNIX_HASH_SIZE - 1);
1072		spin_lock(&unix_table_lock);
1073		u->path = path;
1074		list = &unix_socket_table[hash];
1075	} else {
1076		spin_lock(&unix_table_lock);
1077		err = -EADDRINUSE;
1078		if (__unix_find_socket_byname(net, sunaddr, addr_len,
1079					      sk->sk_type, hash)) {
1080			unix_release_addr(addr);
1081			goto out_unlock;
1082		}
1083
1084		list = &unix_socket_table[addr->hash];
1085	}
1086
1087	err = 0;
1088	__unix_remove_socket(sk);
1089	smp_store_release(&u->addr, addr);
1090	__unix_insert_socket(list, sk);
1091
1092out_unlock:
1093	spin_unlock(&unix_table_lock);
1094out_up:
1095	mutex_unlock(&u->bindlock);
1096out_put:
1097	if (err)
1098		path_put(&path);
1099out:
1100	return err;
1101}
1102
1103static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1104{
1105	if (unlikely(sk1 == sk2) || !sk2) {
1106		unix_state_lock(sk1);
1107		return;
1108	}
1109	if (sk1 < sk2) {
1110		unix_state_lock(sk1);
1111		unix_state_lock_nested(sk2);
1112	} else {
1113		unix_state_lock(sk2);
1114		unix_state_lock_nested(sk1);
1115	}
1116}
1117
1118static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1119{
1120	if (unlikely(sk1 == sk2) || !sk2) {
1121		unix_state_unlock(sk1);
1122		return;
1123	}
1124	unix_state_unlock(sk1);
1125	unix_state_unlock(sk2);
1126}
1127
1128static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1129			      int alen, int flags)
1130{
1131	struct sock *sk = sock->sk;
1132	struct net *net = sock_net(sk);
1133	struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1134	struct sock *other;
1135	unsigned int hash;
1136	int err;
1137
1138	err = -EINVAL;
1139	if (alen < offsetofend(struct sockaddr, sa_family))
1140		goto out;
1141
1142	if (addr->sa_family != AF_UNSPEC) {
1143		err = unix_mkname(sunaddr, alen, &hash);
1144		if (err < 0)
1145			goto out;
1146		alen = err;
1147
1148		if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1149		    !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1150			goto out;
1151
1152restart:
1153		other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1154		if (!other)
1155			goto out;
1156
1157		unix_state_double_lock(sk, other);
1158
1159		/* Apparently VFS overslept socket death. Retry. */
1160		if (sock_flag(other, SOCK_DEAD)) {
1161			unix_state_double_unlock(sk, other);
1162			sock_put(other);
1163			goto restart;
1164		}
1165
1166		err = -EPERM;
1167		if (!unix_may_send(sk, other))
1168			goto out_unlock;
1169
1170		err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1171		if (err)
1172			goto out_unlock;
1173
1174	} else {
1175		/*
1176		 *	1003.1g breaking connected state with AF_UNSPEC
1177		 */
1178		other = NULL;
1179		unix_state_double_lock(sk, other);
1180	}
1181
1182	/*
1183	 * If it was connected, reconnect.
1184	 */
1185	if (unix_peer(sk)) {
1186		struct sock *old_peer = unix_peer(sk);
1187		unix_peer(sk) = other;
1188		unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1189
1190		unix_state_double_unlock(sk, other);
1191
1192		if (other != old_peer)
1193			unix_dgram_disconnected(sk, old_peer);
1194		sock_put(old_peer);
1195	} else {
1196		unix_peer(sk) = other;
1197		unix_state_double_unlock(sk, other);
1198	}
1199	return 0;
1200
1201out_unlock:
1202	unix_state_double_unlock(sk, other);
1203	sock_put(other);
1204out:
1205	return err;
1206}
1207
1208static long unix_wait_for_peer(struct sock *other, long timeo)
1209	__releases(&unix_sk(other)->lock)
1210{
1211	struct unix_sock *u = unix_sk(other);
1212	int sched;
1213	DEFINE_WAIT(wait);
1214
1215	prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1216
1217	sched = !sock_flag(other, SOCK_DEAD) &&
1218		!(other->sk_shutdown & RCV_SHUTDOWN) &&
1219		unix_recvq_full(other);
1220
1221	unix_state_unlock(other);
1222
1223	if (sched)
1224		timeo = schedule_timeout(timeo);
1225
1226	finish_wait(&u->peer_wait, &wait);
1227	return timeo;
1228}
1229
1230static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1231			       int addr_len, int flags)
1232{
1233	struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1234	struct sock *sk = sock->sk;
1235	struct net *net = sock_net(sk);
1236	struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1237	struct sock *newsk = NULL;
1238	struct sock *other = NULL;
1239	struct sk_buff *skb = NULL;
1240	unsigned int hash;
1241	int st;
1242	int err;
1243	long timeo;
1244
1245	err = unix_mkname(sunaddr, addr_len, &hash);
1246	if (err < 0)
1247		goto out;
1248	addr_len = err;
1249
1250	if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1251	    (err = unix_autobind(sock)) != 0)
1252		goto out;
1253
1254	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1255
1256	/* First of all allocate resources.
1257	   If we will make it after state is locked,
1258	   we will have to recheck all again in any case.
1259	 */
1260
1261	err = -ENOMEM;
1262
1263	/* create new sock for complete connection */
1264	newsk = unix_create1(sock_net(sk), NULL, 0);
1265	if (newsk == NULL)
1266		goto out;
1267
1268	/* Allocate skb for sending to listening sock */
1269	skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1270	if (skb == NULL)
1271		goto out;
1272
1273restart:
1274	/*  Find listening sock. */
1275	other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1276	if (!other)
1277		goto out;
1278
1279	/* Latch state of peer */
1280	unix_state_lock(other);
1281
1282	/* Apparently VFS overslept socket death. Retry. */
1283	if (sock_flag(other, SOCK_DEAD)) {
1284		unix_state_unlock(other);
1285		sock_put(other);
1286		goto restart;
1287	}
1288
1289	err = -ECONNREFUSED;
1290	if (other->sk_state != TCP_LISTEN)
1291		goto out_unlock;
1292	if (other->sk_shutdown & RCV_SHUTDOWN)
1293		goto out_unlock;
1294
1295	if (unix_recvq_full(other)) {
1296		err = -EAGAIN;
1297		if (!timeo)
1298			goto out_unlock;
1299
1300		timeo = unix_wait_for_peer(other, timeo);
1301
1302		err = sock_intr_errno(timeo);
1303		if (signal_pending(current))
1304			goto out;
1305		sock_put(other);
1306		goto restart;
1307	}
1308
1309	/* Latch our state.
1310
1311	   It is tricky place. We need to grab our state lock and cannot
1312	   drop lock on peer. It is dangerous because deadlock is
1313	   possible. Connect to self case and simultaneous
1314	   attempt to connect are eliminated by checking socket
1315	   state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1316	   check this before attempt to grab lock.
1317
1318	   Well, and we have to recheck the state after socket locked.
1319	 */
1320	st = sk->sk_state;
1321
1322	switch (st) {
1323	case TCP_CLOSE:
1324		/* This is ok... continue with connect */
1325		break;
1326	case TCP_ESTABLISHED:
1327		/* Socket is already connected */
1328		err = -EISCONN;
1329		goto out_unlock;
1330	default:
1331		err = -EINVAL;
1332		goto out_unlock;
1333	}
1334
1335	unix_state_lock_nested(sk);
1336
1337	if (sk->sk_state != st) {
1338		unix_state_unlock(sk);
1339		unix_state_unlock(other);
1340		sock_put(other);
1341		goto restart;
1342	}
1343
1344	err = security_unix_stream_connect(sk, other, newsk);
1345	if (err) {
1346		unix_state_unlock(sk);
1347		goto out_unlock;
1348	}
1349
1350	/* The way is open! Fastly set all the necessary fields... */
1351
1352	sock_hold(sk);
1353	unix_peer(newsk)	= sk;
1354	newsk->sk_state		= TCP_ESTABLISHED;
1355	newsk->sk_type		= sk->sk_type;
1356	init_peercred(newsk);
1357	newu = unix_sk(newsk);
1358	RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1359	otheru = unix_sk(other);
1360
1361	/* copy address information from listening to new sock
1362	 *
1363	 * The contents of *(otheru->addr) and otheru->path
1364	 * are seen fully set up here, since we have found
1365	 * otheru in hash under unix_table_lock.  Insertion
1366	 * into the hash chain we'd found it in had been done
1367	 * in an earlier critical area protected by unix_table_lock,
1368	 * the same one where we'd set *(otheru->addr) contents,
1369	 * as well as otheru->path and otheru->addr itself.
1370	 *
1371	 * Using smp_store_release() here to set newu->addr
1372	 * is enough to make those stores, as well as stores
1373	 * to newu->path visible to anyone who gets newu->addr
1374	 * by smp_load_acquire().  IOW, the same warranties
1375	 * as for unix_sock instances bound in unix_bind() or
1376	 * in unix_autobind().
1377	 */
1378	if (otheru->path.dentry) {
1379		path_get(&otheru->path);
1380		newu->path = otheru->path;
1381	}
1382	refcount_inc(&otheru->addr->refcnt);
1383	smp_store_release(&newu->addr, otheru->addr);
1384
1385	/* Set credentials */
1386	copy_peercred(sk, other);
1387
1388	sock->state	= SS_CONNECTED;
1389	sk->sk_state	= TCP_ESTABLISHED;
1390	sock_hold(newsk);
1391
1392	smp_mb__after_atomic();	/* sock_hold() does an atomic_inc() */
1393	unix_peer(sk)	= newsk;
1394
1395	unix_state_unlock(sk);
1396
1397	/* take ten and and send info to listening sock */
1398	spin_lock(&other->sk_receive_queue.lock);
1399	__skb_queue_tail(&other->sk_receive_queue, skb);
1400	spin_unlock(&other->sk_receive_queue.lock);
1401	unix_state_unlock(other);
1402	other->sk_data_ready(other);
1403	sock_put(other);
1404	return 0;
1405
1406out_unlock:
1407	if (other)
1408		unix_state_unlock(other);
1409
1410out:
1411	kfree_skb(skb);
1412	if (newsk)
1413		unix_release_sock(newsk, 0);
1414	if (other)
1415		sock_put(other);
1416	return err;
1417}
1418
1419static int unix_socketpair(struct socket *socka, struct socket *sockb)
1420{
1421	struct sock *ska = socka->sk, *skb = sockb->sk;
1422
1423	/* Join our sockets back to back */
1424	sock_hold(ska);
1425	sock_hold(skb);
1426	unix_peer(ska) = skb;
1427	unix_peer(skb) = ska;
1428	init_peercred(ska);
1429	init_peercred(skb);
1430
1431	if (ska->sk_type != SOCK_DGRAM) {
1432		ska->sk_state = TCP_ESTABLISHED;
1433		skb->sk_state = TCP_ESTABLISHED;
1434		socka->state  = SS_CONNECTED;
1435		sockb->state  = SS_CONNECTED;
1436	}
1437	return 0;
1438}
1439
1440static void unix_sock_inherit_flags(const struct socket *old,
1441				    struct socket *new)
1442{
1443	if (test_bit(SOCK_PASSCRED, &old->flags))
1444		set_bit(SOCK_PASSCRED, &new->flags);
1445	if (test_bit(SOCK_PASSSEC, &old->flags))
1446		set_bit(SOCK_PASSSEC, &new->flags);
1447}
1448
1449static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1450		       bool kern)
1451{
1452	struct sock *sk = sock->sk;
1453	struct sock *tsk;
1454	struct sk_buff *skb;
1455	int err;
1456
1457	err = -EOPNOTSUPP;
1458	if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1459		goto out;
1460
1461	err = -EINVAL;
1462	if (sk->sk_state != TCP_LISTEN)
1463		goto out;
1464
1465	/* If socket state is TCP_LISTEN it cannot change (for now...),
1466	 * so that no locks are necessary.
1467	 */
1468
1469	skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1470	if (!skb) {
1471		/* This means receive shutdown. */
1472		if (err == 0)
1473			err = -EINVAL;
1474		goto out;
1475	}
1476
1477	tsk = skb->sk;
1478	skb_free_datagram(sk, skb);
1479	wake_up_interruptible(&unix_sk(sk)->peer_wait);
1480
1481	/* attach accepted sock to socket */
1482	unix_state_lock(tsk);
1483	newsock->state = SS_CONNECTED;
1484	unix_sock_inherit_flags(sock, newsock);
1485	sock_graft(tsk, newsock);
1486	unix_state_unlock(tsk);
1487	return 0;
1488
1489out:
1490	return err;
1491}
1492
1493
1494static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
1495{
1496	struct sock *sk = sock->sk;
1497	struct unix_address *addr;
1498	DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1499	int err = 0;
1500
1501	if (peer) {
1502		sk = unix_peer_get(sk);
1503
1504		err = -ENOTCONN;
1505		if (!sk)
1506			goto out;
1507		err = 0;
1508	} else {
1509		sock_hold(sk);
1510	}
1511
1512	addr = smp_load_acquire(&unix_sk(sk)->addr);
1513	if (!addr) {
1514		sunaddr->sun_family = AF_UNIX;
1515		sunaddr->sun_path[0] = 0;
1516		err = sizeof(short);
1517	} else {
1518		err = addr->len;
1519		memcpy(sunaddr, addr->name, addr->len);
1520	}
1521	sock_put(sk);
1522out:
1523	return err;
1524}
1525
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1526static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1527{
1528	int err = 0;
1529
1530	UNIXCB(skb).pid  = get_pid(scm->pid);
1531	UNIXCB(skb).uid = scm->creds.uid;
1532	UNIXCB(skb).gid = scm->creds.gid;
1533	UNIXCB(skb).fp = NULL;
1534	unix_get_secdata(scm, skb);
1535	if (scm->fp && send_fds)
1536		err = unix_attach_fds(scm, skb);
1537
1538	skb->destructor = unix_destruct_scm;
1539	return err;
1540}
1541
1542static bool unix_passcred_enabled(const struct socket *sock,
1543				  const struct sock *other)
1544{
1545	return test_bit(SOCK_PASSCRED, &sock->flags) ||
1546	       !other->sk_socket ||
1547	       test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1548}
1549
1550/*
1551 * Some apps rely on write() giving SCM_CREDENTIALS
1552 * We include credentials if source or destination socket
1553 * asserted SOCK_PASSCRED.
1554 */
1555static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1556			    const struct sock *other)
1557{
1558	if (UNIXCB(skb).pid)
1559		return;
1560	if (unix_passcred_enabled(sock, other)) {
1561		UNIXCB(skb).pid  = get_pid(task_tgid(current));
1562		current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1563	}
1564}
1565
1566static int maybe_init_creds(struct scm_cookie *scm,
1567			    struct socket *socket,
1568			    const struct sock *other)
1569{
1570	int err;
1571	struct msghdr msg = { .msg_controllen = 0 };
1572
1573	err = scm_send(socket, &msg, scm, false);
1574	if (err)
1575		return err;
1576
1577	if (unix_passcred_enabled(socket, other)) {
1578		scm->pid = get_pid(task_tgid(current));
1579		current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1580	}
1581	return err;
1582}
1583
1584static bool unix_skb_scm_eq(struct sk_buff *skb,
1585			    struct scm_cookie *scm)
1586{
1587	const struct unix_skb_parms *u = &UNIXCB(skb);
1588
1589	return u->pid == scm->pid &&
1590	       uid_eq(u->uid, scm->creds.uid) &&
1591	       gid_eq(u->gid, scm->creds.gid) &&
1592	       unix_secdata_eq(scm, skb);
1593}
1594
1595static void scm_stat_add(struct sock *sk, struct sk_buff *skb)
1596{
1597	struct scm_fp_list *fp = UNIXCB(skb).fp;
1598	struct unix_sock *u = unix_sk(sk);
1599
1600	if (unlikely(fp && fp->count))
1601		atomic_add(fp->count, &u->scm_stat.nr_fds);
1602}
1603
1604static void scm_stat_del(struct sock *sk, struct sk_buff *skb)
1605{
1606	struct scm_fp_list *fp = UNIXCB(skb).fp;
1607	struct unix_sock *u = unix_sk(sk);
1608
1609	if (unlikely(fp && fp->count))
1610		atomic_sub(fp->count, &u->scm_stat.nr_fds);
1611}
1612
1613/*
1614 *	Send AF_UNIX data.
1615 */
1616
1617static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1618			      size_t len)
1619{
1620	struct sock *sk = sock->sk;
1621	struct net *net = sock_net(sk);
1622	struct unix_sock *u = unix_sk(sk);
1623	DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1624	struct sock *other = NULL;
1625	int namelen = 0; /* fake GCC */
1626	int err;
1627	unsigned int hash;
1628	struct sk_buff *skb;
1629	long timeo;
1630	struct scm_cookie scm;
1631	int data_len = 0;
1632	int sk_locked;
1633
1634	wait_for_unix_gc();
1635	err = scm_send(sock, msg, &scm, false);
1636	if (err < 0)
1637		return err;
1638
1639	err = -EOPNOTSUPP;
1640	if (msg->msg_flags&MSG_OOB)
1641		goto out;
1642
1643	if (msg->msg_namelen) {
1644		err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1645		if (err < 0)
1646			goto out;
1647		namelen = err;
1648	} else {
1649		sunaddr = NULL;
1650		err = -ENOTCONN;
1651		other = unix_peer_get(sk);
1652		if (!other)
1653			goto out;
1654	}
1655
1656	if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1657	    && (err = unix_autobind(sock)) != 0)
1658		goto out;
1659
1660	err = -EMSGSIZE;
1661	if (len > sk->sk_sndbuf - 32)
1662		goto out;
1663
1664	if (len > SKB_MAX_ALLOC) {
1665		data_len = min_t(size_t,
1666				 len - SKB_MAX_ALLOC,
1667				 MAX_SKB_FRAGS * PAGE_SIZE);
1668		data_len = PAGE_ALIGN(data_len);
1669
1670		BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1671	}
1672
1673	skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1674				   msg->msg_flags & MSG_DONTWAIT, &err,
1675				   PAGE_ALLOC_COSTLY_ORDER);
1676	if (skb == NULL)
1677		goto out;
1678
1679	err = unix_scm_to_skb(&scm, skb, true);
1680	if (err < 0)
1681		goto out_free;
1682
1683	skb_put(skb, len - data_len);
1684	skb->data_len = data_len;
1685	skb->len = len;
1686	err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1687	if (err)
1688		goto out_free;
1689
1690	timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1691
1692restart:
1693	if (!other) {
1694		err = -ECONNRESET;
1695		if (sunaddr == NULL)
1696			goto out_free;
1697
1698		other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1699					hash, &err);
1700		if (other == NULL)
1701			goto out_free;
1702	}
1703
1704	if (sk_filter(other, skb) < 0) {
1705		/* Toss the packet but do not return any error to the sender */
1706		err = len;
1707		goto out_free;
1708	}
1709
1710	sk_locked = 0;
1711	unix_state_lock(other);
1712restart_locked:
1713	err = -EPERM;
1714	if (!unix_may_send(sk, other))
1715		goto out_unlock;
1716
1717	if (unlikely(sock_flag(other, SOCK_DEAD))) {
1718		/*
1719		 *	Check with 1003.1g - what should
1720		 *	datagram error
1721		 */
1722		unix_state_unlock(other);
1723		sock_put(other);
1724
1725		if (!sk_locked)
1726			unix_state_lock(sk);
1727
1728		err = 0;
1729		if (unix_peer(sk) == other) {
1730			unix_peer(sk) = NULL;
1731			unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1732
1733			unix_state_unlock(sk);
1734
1735			unix_dgram_disconnected(sk, other);
1736			sock_put(other);
1737			err = -ECONNREFUSED;
1738		} else {
1739			unix_state_unlock(sk);
1740		}
1741
1742		other = NULL;
1743		if (err)
1744			goto out_free;
1745		goto restart;
1746	}
1747
1748	err = -EPIPE;
1749	if (other->sk_shutdown & RCV_SHUTDOWN)
1750		goto out_unlock;
1751
1752	if (sk->sk_type != SOCK_SEQPACKET) {
1753		err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1754		if (err)
1755			goto out_unlock;
1756	}
1757
1758	/* other == sk && unix_peer(other) != sk if
1759	 * - unix_peer(sk) == NULL, destination address bound to sk
1760	 * - unix_peer(sk) == sk by time of get but disconnected before lock
1761	 */
1762	if (other != sk &&
1763	    unlikely(unix_peer(other) != sk &&
1764	    unix_recvq_full_lockless(other))) {
1765		if (timeo) {
1766			timeo = unix_wait_for_peer(other, timeo);
1767
1768			err = sock_intr_errno(timeo);
1769			if (signal_pending(current))
1770				goto out_free;
1771
1772			goto restart;
1773		}
1774
1775		if (!sk_locked) {
1776			unix_state_unlock(other);
1777			unix_state_double_lock(sk, other);
1778		}
1779
1780		if (unix_peer(sk) != other ||
1781		    unix_dgram_peer_wake_me(sk, other)) {
1782			err = -EAGAIN;
1783			sk_locked = 1;
1784			goto out_unlock;
1785		}
1786
1787		if (!sk_locked) {
1788			sk_locked = 1;
1789			goto restart_locked;
1790		}
1791	}
1792
1793	if (unlikely(sk_locked))
1794		unix_state_unlock(sk);
1795
1796	if (sock_flag(other, SOCK_RCVTSTAMP))
1797		__net_timestamp(skb);
1798	maybe_add_creds(skb, sock, other);
1799	scm_stat_add(other, skb);
1800	skb_queue_tail(&other->sk_receive_queue, skb);
1801	unix_state_unlock(other);
1802	other->sk_data_ready(other);
1803	sock_put(other);
1804	scm_destroy(&scm);
1805	return len;
1806
1807out_unlock:
1808	if (sk_locked)
1809		unix_state_unlock(sk);
1810	unix_state_unlock(other);
1811out_free:
1812	kfree_skb(skb);
1813out:
1814	if (other)
1815		sock_put(other);
1816	scm_destroy(&scm);
1817	return err;
1818}
1819
1820/* We use paged skbs for stream sockets, and limit occupancy to 32768
1821 * bytes, and a minimum of a full page.
1822 */
1823#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1824
1825static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1826			       size_t len)
1827{
1828	struct sock *sk = sock->sk;
1829	struct sock *other = NULL;
1830	int err, size;
1831	struct sk_buff *skb;
1832	int sent = 0;
1833	struct scm_cookie scm;
1834	bool fds_sent = false;
1835	int data_len;
1836
1837	wait_for_unix_gc();
1838	err = scm_send(sock, msg, &scm, false);
1839	if (err < 0)
1840		return err;
1841
1842	err = -EOPNOTSUPP;
1843	if (msg->msg_flags&MSG_OOB)
1844		goto out_err;
1845
1846	if (msg->msg_namelen) {
1847		err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1848		goto out_err;
1849	} else {
1850		err = -ENOTCONN;
1851		other = unix_peer(sk);
1852		if (!other)
1853			goto out_err;
1854	}
1855
1856	if (sk->sk_shutdown & SEND_SHUTDOWN)
1857		goto pipe_err;
1858
1859	while (sent < len) {
1860		size = len - sent;
1861
1862		/* Keep two messages in the pipe so it schedules better */
1863		size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1864
1865		/* allow fallback to order-0 allocations */
1866		size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1867
1868		data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1869
1870		data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1871
1872		skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1873					   msg->msg_flags & MSG_DONTWAIT, &err,
1874					   get_order(UNIX_SKB_FRAGS_SZ));
1875		if (!skb)
1876			goto out_err;
1877
1878		/* Only send the fds in the first buffer */
1879		err = unix_scm_to_skb(&scm, skb, !fds_sent);
1880		if (err < 0) {
1881			kfree_skb(skb);
1882			goto out_err;
1883		}
1884		fds_sent = true;
1885
1886		skb_put(skb, size - data_len);
1887		skb->data_len = data_len;
1888		skb->len = size;
1889		err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
1890		if (err) {
1891			kfree_skb(skb);
1892			goto out_err;
1893		}
1894
1895		unix_state_lock(other);
1896
1897		if (sock_flag(other, SOCK_DEAD) ||
1898		    (other->sk_shutdown & RCV_SHUTDOWN))
1899			goto pipe_err_free;
1900
1901		maybe_add_creds(skb, sock, other);
1902		scm_stat_add(other, skb);
1903		skb_queue_tail(&other->sk_receive_queue, skb);
1904		unix_state_unlock(other);
1905		other->sk_data_ready(other);
1906		sent += size;
1907	}
1908
1909	scm_destroy(&scm);
1910
1911	return sent;
1912
1913pipe_err_free:
1914	unix_state_unlock(other);
1915	kfree_skb(skb);
1916pipe_err:
1917	if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1918		send_sig(SIGPIPE, current, 0);
1919	err = -EPIPE;
1920out_err:
1921	scm_destroy(&scm);
1922	return sent ? : err;
1923}
1924
1925static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1926				    int offset, size_t size, int flags)
1927{
1928	int err;
1929	bool send_sigpipe = false;
1930	bool init_scm = true;
1931	struct scm_cookie scm;
1932	struct sock *other, *sk = socket->sk;
1933	struct sk_buff *skb, *newskb = NULL, *tail = NULL;
1934
1935	if (flags & MSG_OOB)
1936		return -EOPNOTSUPP;
1937
1938	other = unix_peer(sk);
1939	if (!other || sk->sk_state != TCP_ESTABLISHED)
1940		return -ENOTCONN;
1941
1942	if (false) {
1943alloc_skb:
1944		unix_state_unlock(other);
1945		mutex_unlock(&unix_sk(other)->iolock);
1946		newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
1947					      &err, 0);
1948		if (!newskb)
1949			goto err;
1950	}
1951
1952	/* we must acquire iolock as we modify already present
1953	 * skbs in the sk_receive_queue and mess with skb->len
1954	 */
1955	err = mutex_lock_interruptible(&unix_sk(other)->iolock);
1956	if (err) {
1957		err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
1958		goto err;
1959	}
1960
1961	if (sk->sk_shutdown & SEND_SHUTDOWN) {
1962		err = -EPIPE;
1963		send_sigpipe = true;
1964		goto err_unlock;
1965	}
1966
1967	unix_state_lock(other);
1968
1969	if (sock_flag(other, SOCK_DEAD) ||
1970	    other->sk_shutdown & RCV_SHUTDOWN) {
1971		err = -EPIPE;
1972		send_sigpipe = true;
1973		goto err_state_unlock;
1974	}
1975
1976	if (init_scm) {
1977		err = maybe_init_creds(&scm, socket, other);
1978		if (err)
1979			goto err_state_unlock;
1980		init_scm = false;
1981	}
1982
1983	skb = skb_peek_tail(&other->sk_receive_queue);
1984	if (tail && tail == skb) {
1985		skb = newskb;
1986	} else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
1987		if (newskb) {
1988			skb = newskb;
1989		} else {
1990			tail = skb;
1991			goto alloc_skb;
1992		}
1993	} else if (newskb) {
1994		/* this is fast path, we don't necessarily need to
1995		 * call to kfree_skb even though with newskb == NULL
1996		 * this - does no harm
1997		 */
1998		consume_skb(newskb);
1999		newskb = NULL;
2000	}
2001
2002	if (skb_append_pagefrags(skb, page, offset, size)) {
2003		tail = skb;
2004		goto alloc_skb;
2005	}
2006
2007	skb->len += size;
2008	skb->data_len += size;
2009	skb->truesize += size;
2010	refcount_add(size, &sk->sk_wmem_alloc);
2011
2012	if (newskb) {
2013		err = unix_scm_to_skb(&scm, skb, false);
2014		if (err)
2015			goto err_state_unlock;
2016		spin_lock(&other->sk_receive_queue.lock);
2017		__skb_queue_tail(&other->sk_receive_queue, newskb);
2018		spin_unlock(&other->sk_receive_queue.lock);
2019	}
2020
2021	unix_state_unlock(other);
2022	mutex_unlock(&unix_sk(other)->iolock);
2023
2024	other->sk_data_ready(other);
2025	scm_destroy(&scm);
2026	return size;
2027
2028err_state_unlock:
2029	unix_state_unlock(other);
2030err_unlock:
2031	mutex_unlock(&unix_sk(other)->iolock);
2032err:
2033	kfree_skb(newskb);
2034	if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2035		send_sig(SIGPIPE, current, 0);
2036	if (!init_scm)
2037		scm_destroy(&scm);
2038	return err;
2039}
2040
2041static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2042				  size_t len)
2043{
2044	int err;
2045	struct sock *sk = sock->sk;
2046
2047	err = sock_error(sk);
2048	if (err)
2049		return err;
2050
2051	if (sk->sk_state != TCP_ESTABLISHED)
2052		return -ENOTCONN;
2053
2054	if (msg->msg_namelen)
2055		msg->msg_namelen = 0;
2056
2057	return unix_dgram_sendmsg(sock, msg, len);
2058}
2059
2060static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2061				  size_t size, int flags)
2062{
2063	struct sock *sk = sock->sk;
2064
2065	if (sk->sk_state != TCP_ESTABLISHED)
2066		return -ENOTCONN;
2067
2068	return unix_dgram_recvmsg(sock, msg, size, flags);
2069}
2070
2071static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2072{
2073	struct unix_address *addr = smp_load_acquire(&unix_sk(sk)->addr);
2074
2075	if (addr) {
2076		msg->msg_namelen = addr->len;
2077		memcpy(msg->msg_name, addr->name, addr->len);
2078	}
2079}
2080
2081static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2082			      size_t size, int flags)
2083{
2084	struct scm_cookie scm;
2085	struct sock *sk = sock->sk;
2086	struct unix_sock *u = unix_sk(sk);
2087	struct sk_buff *skb, *last;
2088	long timeo;
2089	int skip;
2090	int err;
2091
2092	err = -EOPNOTSUPP;
2093	if (flags&MSG_OOB)
2094		goto out;
2095
2096	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2097
2098	do {
2099		mutex_lock(&u->iolock);
2100
2101		skip = sk_peek_offset(sk, flags);
2102		skb = __skb_try_recv_datagram(sk, &sk->sk_receive_queue, flags,
2103					      &skip, &err, &last);
2104		if (skb) {
2105			if (!(flags & MSG_PEEK))
2106				scm_stat_del(sk, skb);
2107			break;
2108		}
2109
2110		mutex_unlock(&u->iolock);
2111
2112		if (err != -EAGAIN)
2113			break;
2114	} while (timeo &&
2115		 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
2116					      &err, &timeo, last));
2117
2118	if (!skb) { /* implies iolock unlocked */
2119		unix_state_lock(sk);
2120		/* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2121		if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2122		    (sk->sk_shutdown & RCV_SHUTDOWN))
2123			err = 0;
2124		unix_state_unlock(sk);
2125		goto out;
2126	}
2127
2128	if (wq_has_sleeper(&u->peer_wait))
2129		wake_up_interruptible_sync_poll(&u->peer_wait,
2130						EPOLLOUT | EPOLLWRNORM |
2131						EPOLLWRBAND);
2132
2133	if (msg->msg_name)
2134		unix_copy_addr(msg, skb->sk);
2135
2136	if (size > skb->len - skip)
2137		size = skb->len - skip;
2138	else if (size < skb->len - skip)
2139		msg->msg_flags |= MSG_TRUNC;
2140
2141	err = skb_copy_datagram_msg(skb, skip, msg, size);
2142	if (err)
2143		goto out_free;
2144
2145	if (sock_flag(sk, SOCK_RCVTSTAMP))
2146		__sock_recv_timestamp(msg, sk, skb);
2147
2148	memset(&scm, 0, sizeof(scm));
2149
2150	scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2151	unix_set_secdata(&scm, skb);
2152
2153	if (!(flags & MSG_PEEK)) {
2154		if (UNIXCB(skb).fp)
2155			unix_detach_fds(&scm, skb);
2156
2157		sk_peek_offset_bwd(sk, skb->len);
2158	} else {
2159		/* It is questionable: on PEEK we could:
2160		   - do not return fds - good, but too simple 8)
2161		   - return fds, and do not return them on read (old strategy,
2162		     apparently wrong)
2163		   - clone fds (I chose it for now, it is the most universal
2164		     solution)
2165
2166		   POSIX 1003.1g does not actually define this clearly
2167		   at all. POSIX 1003.1g doesn't define a lot of things
2168		   clearly however!
2169
2170		*/
2171
2172		sk_peek_offset_fwd(sk, size);
2173
2174		if (UNIXCB(skb).fp)
2175			scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2176	}
2177	err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2178
2179	scm_recv(sock, msg, &scm, flags);
2180
2181out_free:
2182	skb_free_datagram(sk, skb);
2183	mutex_unlock(&u->iolock);
2184out:
2185	return err;
2186}
2187
2188/*
2189 *	Sleep until more data has arrived. But check for races..
2190 */
2191static long unix_stream_data_wait(struct sock *sk, long timeo,
2192				  struct sk_buff *last, unsigned int last_len,
2193				  bool freezable)
2194{
2195	struct sk_buff *tail;
2196	DEFINE_WAIT(wait);
2197
2198	unix_state_lock(sk);
2199
2200	for (;;) {
2201		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2202
2203		tail = skb_peek_tail(&sk->sk_receive_queue);
2204		if (tail != last ||
2205		    (tail && tail->len != last_len) ||
2206		    sk->sk_err ||
2207		    (sk->sk_shutdown & RCV_SHUTDOWN) ||
2208		    signal_pending(current) ||
2209		    !timeo)
2210			break;
2211
2212		sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2213		unix_state_unlock(sk);
2214		if (freezable)
2215			timeo = freezable_schedule_timeout(timeo);
2216		else
2217			timeo = schedule_timeout(timeo);
2218		unix_state_lock(sk);
2219
2220		if (sock_flag(sk, SOCK_DEAD))
2221			break;
2222
2223		sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2224	}
2225
2226	finish_wait(sk_sleep(sk), &wait);
2227	unix_state_unlock(sk);
2228	return timeo;
2229}
2230
2231static unsigned int unix_skb_len(const struct sk_buff *skb)
2232{
2233	return skb->len - UNIXCB(skb).consumed;
2234}
2235
2236struct unix_stream_read_state {
2237	int (*recv_actor)(struct sk_buff *, int, int,
2238			  struct unix_stream_read_state *);
2239	struct socket *socket;
2240	struct msghdr *msg;
2241	struct pipe_inode_info *pipe;
2242	size_t size;
2243	int flags;
2244	unsigned int splice_flags;
2245};
2246
2247static int unix_stream_read_generic(struct unix_stream_read_state *state,
2248				    bool freezable)
2249{
2250	struct scm_cookie scm;
2251	struct socket *sock = state->socket;
2252	struct sock *sk = sock->sk;
2253	struct unix_sock *u = unix_sk(sk);
2254	int copied = 0;
2255	int flags = state->flags;
2256	int noblock = flags & MSG_DONTWAIT;
2257	bool check_creds = false;
2258	int target;
2259	int err = 0;
2260	long timeo;
2261	int skip;
2262	size_t size = state->size;
2263	unsigned int last_len;
2264
2265	if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2266		err = -EINVAL;
2267		goto out;
2268	}
2269
2270	if (unlikely(flags & MSG_OOB)) {
2271		err = -EOPNOTSUPP;
2272		goto out;
2273	}
2274
2275	target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2276	timeo = sock_rcvtimeo(sk, noblock);
2277
2278	memset(&scm, 0, sizeof(scm));
2279
2280	/* Lock the socket to prevent queue disordering
2281	 * while sleeps in memcpy_tomsg
2282	 */
2283	mutex_lock(&u->iolock);
2284
2285	skip = max(sk_peek_offset(sk, flags), 0);
2286
2287	do {
2288		int chunk;
2289		bool drop_skb;
2290		struct sk_buff *skb, *last;
2291
2292redo:
2293		unix_state_lock(sk);
2294		if (sock_flag(sk, SOCK_DEAD)) {
2295			err = -ECONNRESET;
2296			goto unlock;
2297		}
2298		last = skb = skb_peek(&sk->sk_receive_queue);
2299		last_len = last ? last->len : 0;
2300again:
2301		if (skb == NULL) {
2302			if (copied >= target)
2303				goto unlock;
2304
2305			/*
2306			 *	POSIX 1003.1g mandates this order.
2307			 */
2308
2309			err = sock_error(sk);
2310			if (err)
2311				goto unlock;
2312			if (sk->sk_shutdown & RCV_SHUTDOWN)
2313				goto unlock;
2314
2315			unix_state_unlock(sk);
2316			if (!timeo) {
2317				err = -EAGAIN;
2318				break;
2319			}
2320
2321			mutex_unlock(&u->iolock);
2322
2323			timeo = unix_stream_data_wait(sk, timeo, last,
2324						      last_len, freezable);
2325
2326			if (signal_pending(current)) {
2327				err = sock_intr_errno(timeo);
2328				scm_destroy(&scm);
2329				goto out;
2330			}
2331
2332			mutex_lock(&u->iolock);
2333			goto redo;
2334unlock:
2335			unix_state_unlock(sk);
2336			break;
2337		}
2338
2339		while (skip >= unix_skb_len(skb)) {
2340			skip -= unix_skb_len(skb);
2341			last = skb;
2342			last_len = skb->len;
2343			skb = skb_peek_next(skb, &sk->sk_receive_queue);
2344			if (!skb)
2345				goto again;
2346		}
2347
2348		unix_state_unlock(sk);
2349
2350		if (check_creds) {
2351			/* Never glue messages from different writers */
2352			if (!unix_skb_scm_eq(skb, &scm))
2353				break;
2354		} else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2355			/* Copy credentials */
2356			scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2357			unix_set_secdata(&scm, skb);
2358			check_creds = true;
2359		}
2360
2361		/* Copy address just once */
2362		if (state->msg && state->msg->msg_name) {
2363			DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2364					 state->msg->msg_name);
2365			unix_copy_addr(state->msg, skb->sk);
2366			sunaddr = NULL;
2367		}
2368
2369		chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2370		skb_get(skb);
2371		chunk = state->recv_actor(skb, skip, chunk, state);
2372		drop_skb = !unix_skb_len(skb);
2373		/* skb is only safe to use if !drop_skb */
2374		consume_skb(skb);
2375		if (chunk < 0) {
2376			if (copied == 0)
2377				copied = -EFAULT;
2378			break;
2379		}
2380		copied += chunk;
2381		size -= chunk;
2382
2383		if (drop_skb) {
2384			/* the skb was touched by a concurrent reader;
2385			 * we should not expect anything from this skb
2386			 * anymore and assume it invalid - we can be
2387			 * sure it was dropped from the socket queue
2388			 *
2389			 * let's report a short read
2390			 */
2391			err = 0;
2392			break;
2393		}
2394
2395		/* Mark read part of skb as used */
2396		if (!(flags & MSG_PEEK)) {
2397			UNIXCB(skb).consumed += chunk;
2398
2399			sk_peek_offset_bwd(sk, chunk);
2400
2401			if (UNIXCB(skb).fp) {
2402				scm_stat_del(sk, skb);
2403				unix_detach_fds(&scm, skb);
2404			}
2405
2406			if (unix_skb_len(skb))
2407				break;
2408
2409			skb_unlink(skb, &sk->sk_receive_queue);
2410			consume_skb(skb);
2411
2412			if (scm.fp)
2413				break;
2414		} else {
2415			/* It is questionable, see note in unix_dgram_recvmsg.
2416			 */
2417			if (UNIXCB(skb).fp)
2418				scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2419
2420			sk_peek_offset_fwd(sk, chunk);
2421
2422			if (UNIXCB(skb).fp)
2423				break;
2424
2425			skip = 0;
2426			last = skb;
2427			last_len = skb->len;
2428			unix_state_lock(sk);
2429			skb = skb_peek_next(skb, &sk->sk_receive_queue);
2430			if (skb)
2431				goto again;
2432			unix_state_unlock(sk);
2433			break;
2434		}
2435	} while (size);
2436
2437	mutex_unlock(&u->iolock);
2438	if (state->msg)
2439		scm_recv(sock, state->msg, &scm, flags);
2440	else
2441		scm_destroy(&scm);
2442out:
2443	return copied ? : err;
2444}
2445
2446static int unix_stream_read_actor(struct sk_buff *skb,
2447				  int skip, int chunk,
2448				  struct unix_stream_read_state *state)
2449{
2450	int ret;
2451
2452	ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2453				    state->msg, chunk);
2454	return ret ?: chunk;
2455}
2456
2457static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2458			       size_t size, int flags)
2459{
2460	struct unix_stream_read_state state = {
2461		.recv_actor = unix_stream_read_actor,
2462		.socket = sock,
2463		.msg = msg,
2464		.size = size,
2465		.flags = flags
2466	};
2467
2468	return unix_stream_read_generic(&state, true);
2469}
2470
2471static int unix_stream_splice_actor(struct sk_buff *skb,
2472				    int skip, int chunk,
2473				    struct unix_stream_read_state *state)
2474{
2475	return skb_splice_bits(skb, state->socket->sk,
2476			       UNIXCB(skb).consumed + skip,
2477			       state->pipe, chunk, state->splice_flags);
2478}
2479
2480static ssize_t unix_stream_splice_read(struct socket *sock,  loff_t *ppos,
2481				       struct pipe_inode_info *pipe,
2482				       size_t size, unsigned int flags)
2483{
2484	struct unix_stream_read_state state = {
2485		.recv_actor = unix_stream_splice_actor,
2486		.socket = sock,
2487		.pipe = pipe,
2488		.size = size,
2489		.splice_flags = flags,
2490	};
2491
2492	if (unlikely(*ppos))
2493		return -ESPIPE;
2494
2495	if (sock->file->f_flags & O_NONBLOCK ||
2496	    flags & SPLICE_F_NONBLOCK)
2497		state.flags = MSG_DONTWAIT;
2498
2499	return unix_stream_read_generic(&state, false);
2500}
2501
2502static int unix_shutdown(struct socket *sock, int mode)
2503{
2504	struct sock *sk = sock->sk;
2505	struct sock *other;
2506
2507	if (mode < SHUT_RD || mode > SHUT_RDWR)
2508		return -EINVAL;
2509	/* This maps:
2510	 * SHUT_RD   (0) -> RCV_SHUTDOWN  (1)
2511	 * SHUT_WR   (1) -> SEND_SHUTDOWN (2)
2512	 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2513	 */
2514	++mode;
2515
2516	unix_state_lock(sk);
2517	sk->sk_shutdown |= mode;
2518	other = unix_peer(sk);
2519	if (other)
2520		sock_hold(other);
2521	unix_state_unlock(sk);
2522	sk->sk_state_change(sk);
2523
2524	if (other &&
2525		(sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2526
2527		int peer_mode = 0;
2528
2529		if (mode&RCV_SHUTDOWN)
2530			peer_mode |= SEND_SHUTDOWN;
2531		if (mode&SEND_SHUTDOWN)
2532			peer_mode |= RCV_SHUTDOWN;
2533		unix_state_lock(other);
2534		other->sk_shutdown |= peer_mode;
2535		unix_state_unlock(other);
2536		other->sk_state_change(other);
2537		if (peer_mode == SHUTDOWN_MASK)
2538			sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2539		else if (peer_mode & RCV_SHUTDOWN)
2540			sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2541	}
2542	if (other)
2543		sock_put(other);
2544
2545	return 0;
2546}
2547
2548long unix_inq_len(struct sock *sk)
2549{
2550	struct sk_buff *skb;
2551	long amount = 0;
2552
2553	if (sk->sk_state == TCP_LISTEN)
2554		return -EINVAL;
2555
2556	spin_lock(&sk->sk_receive_queue.lock);
2557	if (sk->sk_type == SOCK_STREAM ||
2558	    sk->sk_type == SOCK_SEQPACKET) {
2559		skb_queue_walk(&sk->sk_receive_queue, skb)
2560			amount += unix_skb_len(skb);
2561	} else {
2562		skb = skb_peek(&sk->sk_receive_queue);
2563		if (skb)
2564			amount = skb->len;
2565	}
2566	spin_unlock(&sk->sk_receive_queue.lock);
2567
2568	return amount;
2569}
2570EXPORT_SYMBOL_GPL(unix_inq_len);
2571
2572long unix_outq_len(struct sock *sk)
2573{
2574	return sk_wmem_alloc_get(sk);
2575}
2576EXPORT_SYMBOL_GPL(unix_outq_len);
2577
2578static int unix_open_file(struct sock *sk)
2579{
2580	struct path path;
2581	struct file *f;
2582	int fd;
2583
2584	if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2585		return -EPERM;
2586
2587	if (!smp_load_acquire(&unix_sk(sk)->addr))
2588		return -ENOENT;
2589
2590	path = unix_sk(sk)->path;
2591	if (!path.dentry)
2592		return -ENOENT;
2593
2594	path_get(&path);
2595
2596	fd = get_unused_fd_flags(O_CLOEXEC);
2597	if (fd < 0)
2598		goto out;
2599
2600	f = dentry_open(&path, O_PATH, current_cred());
2601	if (IS_ERR(f)) {
2602		put_unused_fd(fd);
2603		fd = PTR_ERR(f);
2604		goto out;
2605	}
2606
2607	fd_install(fd, f);
2608out:
2609	path_put(&path);
2610
2611	return fd;
2612}
2613
2614static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2615{
2616	struct sock *sk = sock->sk;
2617	long amount = 0;
2618	int err;
2619
2620	switch (cmd) {
2621	case SIOCOUTQ:
2622		amount = unix_outq_len(sk);
2623		err = put_user(amount, (int __user *)arg);
2624		break;
2625	case SIOCINQ:
2626		amount = unix_inq_len(sk);
2627		if (amount < 0)
2628			err = amount;
2629		else
2630			err = put_user(amount, (int __user *)arg);
2631		break;
2632	case SIOCUNIXFILE:
2633		err = unix_open_file(sk);
2634		break;
2635	default:
2636		err = -ENOIOCTLCMD;
2637		break;
2638	}
2639	return err;
2640}
2641
2642#ifdef CONFIG_COMPAT
2643static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2644{
2645	return unix_ioctl(sock, cmd, (unsigned long)compat_ptr(arg));
2646}
2647#endif
2648
2649static __poll_t unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2650{
2651	struct sock *sk = sock->sk;
2652	__poll_t mask;
2653
2654	sock_poll_wait(file, sock, wait);
2655	mask = 0;
2656
2657	/* exceptional events? */
2658	if (sk->sk_err)
2659		mask |= EPOLLERR;
2660	if (sk->sk_shutdown == SHUTDOWN_MASK)
2661		mask |= EPOLLHUP;
2662	if (sk->sk_shutdown & RCV_SHUTDOWN)
2663		mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2664
2665	/* readable? */
2666	if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
2667		mask |= EPOLLIN | EPOLLRDNORM;
2668
2669	/* Connection-based need to check for termination and startup */
2670	if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2671	    sk->sk_state == TCP_CLOSE)
2672		mask |= EPOLLHUP;
2673
2674	/*
2675	 * we set writable also when the other side has shut down the
2676	 * connection. This prevents stuck sockets.
2677	 */
2678	if (unix_writable(sk))
2679		mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2680
2681	return mask;
2682}
2683
2684static __poll_t unix_dgram_poll(struct file *file, struct socket *sock,
2685				    poll_table *wait)
2686{
2687	struct sock *sk = sock->sk, *other;
2688	unsigned int writable;
2689	__poll_t mask;
2690
2691	sock_poll_wait(file, sock, wait);
2692	mask = 0;
2693
2694	/* exceptional events? */
2695	if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
2696		mask |= EPOLLERR |
2697			(sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
2698
2699	if (sk->sk_shutdown & RCV_SHUTDOWN)
2700		mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2701	if (sk->sk_shutdown == SHUTDOWN_MASK)
2702		mask |= EPOLLHUP;
2703
2704	/* readable? */
2705	if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
2706		mask |= EPOLLIN | EPOLLRDNORM;
2707
2708	/* Connection-based need to check for termination and startup */
2709	if (sk->sk_type == SOCK_SEQPACKET) {
2710		if (sk->sk_state == TCP_CLOSE)
2711			mask |= EPOLLHUP;
2712		/* connection hasn't started yet? */
2713		if (sk->sk_state == TCP_SYN_SENT)
2714			return mask;
2715	}
2716
2717	/* No write status requested, avoid expensive OUT tests. */
2718	if (!(poll_requested_events(wait) & (EPOLLWRBAND|EPOLLWRNORM|EPOLLOUT)))
2719		return mask;
2720
2721	writable = unix_writable(sk);
2722	if (writable) {
2723		unix_state_lock(sk);
2724
2725		other = unix_peer(sk);
2726		if (other && unix_peer(other) != sk &&
2727		    unix_recvq_full(other) &&
2728		    unix_dgram_peer_wake_me(sk, other))
2729			writable = 0;
2730
2731		unix_state_unlock(sk);
2732	}
2733
2734	if (writable)
2735		mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2736	else
2737		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2738
2739	return mask;
2740}
2741
2742#ifdef CONFIG_PROC_FS
2743
2744#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2745
2746#define get_bucket(x) ((x) >> BUCKET_SPACE)
2747#define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2748#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2749
2750static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2751{
2752	unsigned long offset = get_offset(*pos);
2753	unsigned long bucket = get_bucket(*pos);
2754	struct sock *sk;
2755	unsigned long count = 0;
2756
2757	for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2758		if (sock_net(sk) != seq_file_net(seq))
2759			continue;
2760		if (++count == offset)
2761			break;
2762	}
2763
2764	return sk;
2765}
2766
2767static struct sock *unix_next_socket(struct seq_file *seq,
2768				     struct sock *sk,
2769				     loff_t *pos)
2770{
2771	unsigned long bucket;
2772
2773	while (sk > (struct sock *)SEQ_START_TOKEN) {
2774		sk = sk_next(sk);
2775		if (!sk)
2776			goto next_bucket;
2777		if (sock_net(sk) == seq_file_net(seq))
2778			return sk;
2779	}
2780
2781	do {
2782		sk = unix_from_bucket(seq, pos);
2783		if (sk)
2784			return sk;
2785
2786next_bucket:
2787		bucket = get_bucket(*pos) + 1;
2788		*pos = set_bucket_offset(bucket, 1);
2789	} while (bucket < ARRAY_SIZE(unix_socket_table));
2790
2791	return NULL;
2792}
2793
2794static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2795	__acquires(unix_table_lock)
2796{
2797	spin_lock(&unix_table_lock);
2798
2799	if (!*pos)
2800		return SEQ_START_TOKEN;
2801
2802	if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2803		return NULL;
2804
2805	return unix_next_socket(seq, NULL, pos);
2806}
2807
2808static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2809{
2810	++*pos;
2811	return unix_next_socket(seq, v, pos);
2812}
2813
2814static void unix_seq_stop(struct seq_file *seq, void *v)
2815	__releases(unix_table_lock)
2816{
2817	spin_unlock(&unix_table_lock);
2818}
2819
2820static int unix_seq_show(struct seq_file *seq, void *v)
2821{
2822
2823	if (v == SEQ_START_TOKEN)
2824		seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2825			 "Inode Path\n");
2826	else {
2827		struct sock *s = v;
2828		struct unix_sock *u = unix_sk(s);
2829		unix_state_lock(s);
2830
2831		seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2832			s,
2833			refcount_read(&s->sk_refcnt),
2834			0,
2835			s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2836			s->sk_type,
2837			s->sk_socket ?
2838			(s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2839			(s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2840			sock_i_ino(s));
2841
2842		if (u->addr) {	// under unix_table_lock here
2843			int i, len;
2844			seq_putc(seq, ' ');
2845
2846			i = 0;
2847			len = u->addr->len - sizeof(short);
2848			if (!UNIX_ABSTRACT(s))
2849				len--;
2850			else {
2851				seq_putc(seq, '@');
2852				i++;
2853			}
2854			for ( ; i < len; i++)
2855				seq_putc(seq, u->addr->name->sun_path[i] ?:
2856					 '@');
2857		}
2858		unix_state_unlock(s);
2859		seq_putc(seq, '\n');
2860	}
2861
2862	return 0;
2863}
2864
2865static const struct seq_operations unix_seq_ops = {
2866	.start  = unix_seq_start,
2867	.next   = unix_seq_next,
2868	.stop   = unix_seq_stop,
2869	.show   = unix_seq_show,
2870};
2871#endif
2872
2873static const struct net_proto_family unix_family_ops = {
2874	.family = PF_UNIX,
2875	.create = unix_create,
2876	.owner	= THIS_MODULE,
2877};
2878
2879
2880static int __net_init unix_net_init(struct net *net)
2881{
2882	int error = -ENOMEM;
2883
2884	net->unx.sysctl_max_dgram_qlen = 10;
2885	if (unix_sysctl_register(net))
2886		goto out;
2887
2888#ifdef CONFIG_PROC_FS
2889	if (!proc_create_net("unix", 0, net->proc_net, &unix_seq_ops,
2890			sizeof(struct seq_net_private))) {
2891		unix_sysctl_unregister(net);
2892		goto out;
2893	}
2894#endif
2895	error = 0;
2896out:
2897	return error;
2898}
2899
2900static void __net_exit unix_net_exit(struct net *net)
2901{
2902	unix_sysctl_unregister(net);
2903	remove_proc_entry("unix", net->proc_net);
2904}
2905
2906static struct pernet_operations unix_net_ops = {
2907	.init = unix_net_init,
2908	.exit = unix_net_exit,
2909};
2910
2911static int __init af_unix_init(void)
2912{
2913	int rc = -1;
2914
2915	BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
2916
2917	rc = proto_register(&unix_proto, 1);
2918	if (rc != 0) {
2919		pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2920		goto out;
2921	}
2922
2923	sock_register(&unix_family_ops);
2924	register_pernet_subsys(&unix_net_ops);
2925out:
2926	return rc;
2927}
2928
2929static void __exit af_unix_exit(void)
2930{
2931	sock_unregister(PF_UNIX);
2932	proto_unregister(&unix_proto);
2933	unregister_pernet_subsys(&unix_net_ops);
2934}
2935
2936/* Earlier than device_initcall() so that other drivers invoking
2937   request_module() don't end up in a loop when modprobe tries
2938   to use a UNIX socket. But later than subsys_initcall() because
2939   we depend on stuff initialised there */
2940fs_initcall(af_unix_init);
2941module_exit(af_unix_exit);
2942
2943MODULE_LICENSE("GPL");
2944MODULE_ALIAS_NETPROTO(PF_UNIX);