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