xref: /openbmc/linux/net/unix/af_unix.c (revision ee89bd6b)
1 /*
2  * NET4:	Implementation of BSD Unix domain sockets.
3  *
4  * Authors:	Alan Cox, <alan@lxorguk.ukuu.org.uk>
5  *
6  *		This program is free software; you can redistribute it and/or
7  *		modify it under the terms of the GNU General Public License
8  *		as published by the Free Software Foundation; either version
9  *		2 of the License, or (at your option) any later version.
10  *
11  * Fixes:
12  *		Linus Torvalds	:	Assorted bug cures.
13  *		Niibe Yutaka	:	async I/O support.
14  *		Carsten Paeth	:	PF_UNIX check, address fixes.
15  *		Alan Cox	:	Limit size of allocated blocks.
16  *		Alan Cox	:	Fixed the stupid socketpair bug.
17  *		Alan Cox	:	BSD compatibility fine tuning.
18  *		Alan Cox	:	Fixed a bug in connect when interrupted.
19  *		Alan Cox	:	Sorted out a proper draft version of
20  *					file descriptor passing hacked up from
21  *					Mike Shaver's work.
22  *		Marty Leisner	:	Fixes to fd passing
23  *		Nick Nevin	:	recvmsg bugfix.
24  *		Alan Cox	:	Started proper garbage collector
25  *		Heiko EiBfeldt	:	Missing verify_area check
26  *		Alan Cox	:	Started POSIXisms
27  *		Andreas Schwab	:	Replace inode by dentry for proper
28  *					reference counting
29  *		Kirk Petersen	:	Made this a module
30  *	    Christoph Rohland	:	Elegant non-blocking accept/connect algorithm.
31  *					Lots of bug fixes.
32  *	     Alexey Kuznetosv	:	Repaired (I hope) bugs introduces
33  *					by above two patches.
34  *	     Andrea Arcangeli	:	If possible we block in connect(2)
35  *					if the max backlog of the listen socket
36  *					is been reached. This won't break
37  *					old apps and it will avoid huge amount
38  *					of socks hashed (this for unix_gc()
39  *					performances reasons).
40  *					Security fix that limits the max
41  *					number of socks to 2*max_files and
42  *					the number of skb queueable in the
43  *					dgram receiver.
44  *		Artur Skawina   :	Hash function optimizations
45  *	     Alexey Kuznetsov   :	Full scale SMP. Lot of bugs are introduced 8)
46  *	      Malcolm Beattie   :	Set peercred for socketpair
47  *	     Michal Ostrowski   :       Module initialization cleanup.
48  *	     Arnaldo C. Melo	:	Remove MOD_{INC,DEC}_USE_COUNT,
49  *	     				the core infrastructure is doing that
50  *	     				for all net proto families now (2.5.69+)
51  *
52  *
53  * Known differences from reference BSD that was tested:
54  *
55  *	[TO FIX]
56  *	ECONNREFUSED is not returned from one end of a connected() socket to the
57  *		other the moment one end closes.
58  *	fstat() doesn't return st_dev=0, and give the blksize as high water mark
59  *		and a fake inode identifier (nor the BSD first socket fstat twice bug).
60  *	[NOT TO FIX]
61  *	accept() returns a path name even if the connecting socket has closed
62  *		in the meantime (BSD loses the path and gives up).
63  *	accept() returns 0 length path for an unbound connector. BSD returns 16
64  *		and a null first byte in the path (but not for gethost/peername - BSD bug ??)
65  *	socketpair(...SOCK_RAW..) doesn't panic the kernel.
66  *	BSD af_unix apparently has connect forgetting to block properly.
67  *		(need to check this with the POSIX spec in detail)
68  *
69  * Differences from 2.0.0-11-... (ANK)
70  *	Bug fixes and improvements.
71  *		- client shutdown killed server socket.
72  *		- removed all useless cli/sti pairs.
73  *
74  *	Semantic changes/extensions.
75  *		- generic control message passing.
76  *		- SCM_CREDENTIALS control message.
77  *		- "Abstract" (not FS based) socket bindings.
78  *		  Abstract names are sequences of bytes (not zero terminated)
79  *		  started by 0, so that this name space does not intersect
80  *		  with BSD names.
81  */
82 
83 #include <linux/module.h>
84 #include <linux/kernel.h>
85 #include <linux/signal.h>
86 #include <linux/sched.h>
87 #include <linux/errno.h>
88 #include <linux/string.h>
89 #include <linux/stat.h>
90 #include <linux/dcache.h>
91 #include <linux/namei.h>
92 #include <linux/socket.h>
93 #include <linux/un.h>
94 #include <linux/fcntl.h>
95 #include <linux/termios.h>
96 #include <linux/sockios.h>
97 #include <linux/net.h>
98 #include <linux/in.h>
99 #include <linux/fs.h>
100 #include <linux/slab.h>
101 #include <asm/uaccess.h>
102 #include <linux/skbuff.h>
103 #include <linux/netdevice.h>
104 #include <net/net_namespace.h>
105 #include <net/sock.h>
106 #include <net/tcp_states.h>
107 #include <net/af_unix.h>
108 #include <linux/proc_fs.h>
109 #include <linux/seq_file.h>
110 #include <net/scm.h>
111 #include <linux/init.h>
112 #include <linux/poll.h>
113 #include <linux/rtnetlink.h>
114 #include <linux/mount.h>
115 #include <net/checksum.h>
116 #include <linux/security.h>
117 
118 struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
119 EXPORT_SYMBOL_GPL(unix_socket_table);
120 DEFINE_SPINLOCK(unix_table_lock);
121 EXPORT_SYMBOL_GPL(unix_table_lock);
122 static atomic_long_t unix_nr_socks;
123 
124 
125 static struct hlist_head *unix_sockets_unbound(void *addr)
126 {
127 	unsigned long hash = (unsigned long)addr;
128 
129 	hash ^= hash >> 16;
130 	hash ^= hash >> 8;
131 	hash %= UNIX_HASH_SIZE;
132 	return &unix_socket_table[UNIX_HASH_SIZE + hash];
133 }
134 
135 #define UNIX_ABSTRACT(sk)	(unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
136 
137 #ifdef CONFIG_SECURITY_NETWORK
138 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
139 {
140 	memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
141 }
142 
143 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
144 {
145 	scm->secid = *UNIXSID(skb);
146 }
147 #else
148 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
149 { }
150 
151 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
152 { }
153 #endif /* CONFIG_SECURITY_NETWORK */
154 
155 /*
156  *  SMP locking strategy:
157  *    hash table is protected with spinlock unix_table_lock
158  *    each socket state is protected by separate spin lock.
159  */
160 
161 static inline unsigned int unix_hash_fold(__wsum n)
162 {
163 	unsigned int hash = (__force unsigned int)n;
164 
165 	hash ^= hash>>16;
166 	hash ^= hash>>8;
167 	return hash&(UNIX_HASH_SIZE-1);
168 }
169 
170 #define unix_peer(sk) (unix_sk(sk)->peer)
171 
172 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
173 {
174 	return unix_peer(osk) == sk;
175 }
176 
177 static inline int unix_may_send(struct sock *sk, struct sock *osk)
178 {
179 	return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
180 }
181 
182 static inline int unix_recvq_full(struct sock const *sk)
183 {
184 	return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
185 }
186 
187 struct sock *unix_peer_get(struct sock *s)
188 {
189 	struct sock *peer;
190 
191 	unix_state_lock(s);
192 	peer = unix_peer(s);
193 	if (peer)
194 		sock_hold(peer);
195 	unix_state_unlock(s);
196 	return peer;
197 }
198 EXPORT_SYMBOL_GPL(unix_peer_get);
199 
200 static inline void unix_release_addr(struct unix_address *addr)
201 {
202 	if (atomic_dec_and_test(&addr->refcnt))
203 		kfree(addr);
204 }
205 
206 /*
207  *	Check unix socket name:
208  *		- should be not zero length.
209  *	        - if started by not zero, should be NULL terminated (FS object)
210  *		- if started by zero, it is abstract name.
211  */
212 
213 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
214 {
215 	if (len <= sizeof(short) || len > sizeof(*sunaddr))
216 		return -EINVAL;
217 	if (!sunaddr || sunaddr->sun_family != AF_UNIX)
218 		return -EINVAL;
219 	if (sunaddr->sun_path[0]) {
220 		/*
221 		 * This may look like an off by one error but it is a bit more
222 		 * subtle. 108 is the longest valid AF_UNIX path for a binding.
223 		 * sun_path[108] doesn't as such exist.  However in kernel space
224 		 * we are guaranteed that it is a valid memory location in our
225 		 * kernel address buffer.
226 		 */
227 		((char *)sunaddr)[len] = 0;
228 		len = strlen(sunaddr->sun_path)+1+sizeof(short);
229 		return len;
230 	}
231 
232 	*hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
233 	return len;
234 }
235 
236 static void __unix_remove_socket(struct sock *sk)
237 {
238 	sk_del_node_init(sk);
239 }
240 
241 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
242 {
243 	WARN_ON(!sk_unhashed(sk));
244 	sk_add_node(sk, list);
245 }
246 
247 static inline void unix_remove_socket(struct sock *sk)
248 {
249 	spin_lock(&unix_table_lock);
250 	__unix_remove_socket(sk);
251 	spin_unlock(&unix_table_lock);
252 }
253 
254 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
255 {
256 	spin_lock(&unix_table_lock);
257 	__unix_insert_socket(list, sk);
258 	spin_unlock(&unix_table_lock);
259 }
260 
261 static struct sock *__unix_find_socket_byname(struct net *net,
262 					      struct sockaddr_un *sunname,
263 					      int len, int type, unsigned int hash)
264 {
265 	struct sock *s;
266 
267 	sk_for_each(s, &unix_socket_table[hash ^ type]) {
268 		struct unix_sock *u = unix_sk(s);
269 
270 		if (!net_eq(sock_net(s), net))
271 			continue;
272 
273 		if (u->addr->len == len &&
274 		    !memcmp(u->addr->name, sunname, len))
275 			goto found;
276 	}
277 	s = NULL;
278 found:
279 	return s;
280 }
281 
282 static inline struct sock *unix_find_socket_byname(struct net *net,
283 						   struct sockaddr_un *sunname,
284 						   int len, int type,
285 						   unsigned int hash)
286 {
287 	struct sock *s;
288 
289 	spin_lock(&unix_table_lock);
290 	s = __unix_find_socket_byname(net, sunname, len, type, hash);
291 	if (s)
292 		sock_hold(s);
293 	spin_unlock(&unix_table_lock);
294 	return s;
295 }
296 
297 static struct sock *unix_find_socket_byinode(struct inode *i)
298 {
299 	struct sock *s;
300 
301 	spin_lock(&unix_table_lock);
302 	sk_for_each(s,
303 		    &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
304 		struct dentry *dentry = unix_sk(s)->path.dentry;
305 
306 		if (dentry && dentry->d_inode == i) {
307 			sock_hold(s);
308 			goto found;
309 		}
310 	}
311 	s = NULL;
312 found:
313 	spin_unlock(&unix_table_lock);
314 	return s;
315 }
316 
317 static inline int unix_writable(struct sock *sk)
318 {
319 	return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
320 }
321 
322 static void unix_write_space(struct sock *sk)
323 {
324 	struct socket_wq *wq;
325 
326 	rcu_read_lock();
327 	if (unix_writable(sk)) {
328 		wq = rcu_dereference(sk->sk_wq);
329 		if (wq_has_sleeper(wq))
330 			wake_up_interruptible_sync_poll(&wq->wait,
331 				POLLOUT | POLLWRNORM | POLLWRBAND);
332 		sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
333 	}
334 	rcu_read_unlock();
335 }
336 
337 /* When dgram socket disconnects (or changes its peer), we clear its receive
338  * queue of packets arrived from previous peer. First, it allows to do
339  * flow control based only on wmem_alloc; second, sk connected to peer
340  * may receive messages only from that peer. */
341 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
342 {
343 	if (!skb_queue_empty(&sk->sk_receive_queue)) {
344 		skb_queue_purge(&sk->sk_receive_queue);
345 		wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
346 
347 		/* If one link of bidirectional dgram pipe is disconnected,
348 		 * we signal error. Messages are lost. Do not make this,
349 		 * when peer was not connected to us.
350 		 */
351 		if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
352 			other->sk_err = ECONNRESET;
353 			other->sk_error_report(other);
354 		}
355 	}
356 }
357 
358 static void unix_sock_destructor(struct sock *sk)
359 {
360 	struct unix_sock *u = unix_sk(sk);
361 
362 	skb_queue_purge(&sk->sk_receive_queue);
363 
364 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
365 	WARN_ON(!sk_unhashed(sk));
366 	WARN_ON(sk->sk_socket);
367 	if (!sock_flag(sk, SOCK_DEAD)) {
368 		printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
369 		return;
370 	}
371 
372 	if (u->addr)
373 		unix_release_addr(u->addr);
374 
375 	atomic_long_dec(&unix_nr_socks);
376 	local_bh_disable();
377 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
378 	local_bh_enable();
379 #ifdef UNIX_REFCNT_DEBUG
380 	printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk,
381 		atomic_long_read(&unix_nr_socks));
382 #endif
383 }
384 
385 static void unix_release_sock(struct sock *sk, int embrion)
386 {
387 	struct unix_sock *u = unix_sk(sk);
388 	struct path path;
389 	struct sock *skpair;
390 	struct sk_buff *skb;
391 	int state;
392 
393 	unix_remove_socket(sk);
394 
395 	/* Clear state */
396 	unix_state_lock(sk);
397 	sock_orphan(sk);
398 	sk->sk_shutdown = SHUTDOWN_MASK;
399 	path	     = u->path;
400 	u->path.dentry = NULL;
401 	u->path.mnt = NULL;
402 	state = sk->sk_state;
403 	sk->sk_state = TCP_CLOSE;
404 	unix_state_unlock(sk);
405 
406 	wake_up_interruptible_all(&u->peer_wait);
407 
408 	skpair = unix_peer(sk);
409 
410 	if (skpair != NULL) {
411 		if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
412 			unix_state_lock(skpair);
413 			/* No more writes */
414 			skpair->sk_shutdown = SHUTDOWN_MASK;
415 			if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
416 				skpair->sk_err = ECONNRESET;
417 			unix_state_unlock(skpair);
418 			skpair->sk_state_change(skpair);
419 			sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
420 		}
421 		sock_put(skpair); /* It may now die */
422 		unix_peer(sk) = NULL;
423 	}
424 
425 	/* Try to flush out this socket. Throw out buffers at least */
426 
427 	while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
428 		if (state == TCP_LISTEN)
429 			unix_release_sock(skb->sk, 1);
430 		/* passed fds are erased in the kfree_skb hook	      */
431 		kfree_skb(skb);
432 	}
433 
434 	if (path.dentry)
435 		path_put(&path);
436 
437 	sock_put(sk);
438 
439 	/* ---- Socket is dead now and most probably destroyed ---- */
440 
441 	/*
442 	 * Fixme: BSD difference: In BSD all sockets connected to us get
443 	 *	  ECONNRESET and we die on the spot. In Linux we behave
444 	 *	  like files and pipes do and wait for the last
445 	 *	  dereference.
446 	 *
447 	 * Can't we simply set sock->err?
448 	 *
449 	 *	  What the above comment does talk about? --ANK(980817)
450 	 */
451 
452 	if (unix_tot_inflight)
453 		unix_gc();		/* Garbage collect fds */
454 }
455 
456 static void init_peercred(struct sock *sk)
457 {
458 	put_pid(sk->sk_peer_pid);
459 	if (sk->sk_peer_cred)
460 		put_cred(sk->sk_peer_cred);
461 	sk->sk_peer_pid  = get_pid(task_tgid(current));
462 	sk->sk_peer_cred = get_current_cred();
463 }
464 
465 static void copy_peercred(struct sock *sk, struct sock *peersk)
466 {
467 	put_pid(sk->sk_peer_pid);
468 	if (sk->sk_peer_cred)
469 		put_cred(sk->sk_peer_cred);
470 	sk->sk_peer_pid  = get_pid(peersk->sk_peer_pid);
471 	sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
472 }
473 
474 static int unix_listen(struct socket *sock, int backlog)
475 {
476 	int err;
477 	struct sock *sk = sock->sk;
478 	struct unix_sock *u = unix_sk(sk);
479 	struct pid *old_pid = NULL;
480 
481 	err = -EOPNOTSUPP;
482 	if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
483 		goto out;	/* Only stream/seqpacket sockets accept */
484 	err = -EINVAL;
485 	if (!u->addr)
486 		goto out;	/* No listens on an unbound socket */
487 	unix_state_lock(sk);
488 	if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
489 		goto out_unlock;
490 	if (backlog > sk->sk_max_ack_backlog)
491 		wake_up_interruptible_all(&u->peer_wait);
492 	sk->sk_max_ack_backlog	= backlog;
493 	sk->sk_state		= TCP_LISTEN;
494 	/* set credentials so connect can copy them */
495 	init_peercred(sk);
496 	err = 0;
497 
498 out_unlock:
499 	unix_state_unlock(sk);
500 	put_pid(old_pid);
501 out:
502 	return err;
503 }
504 
505 static int unix_release(struct socket *);
506 static int unix_bind(struct socket *, struct sockaddr *, int);
507 static int unix_stream_connect(struct socket *, struct sockaddr *,
508 			       int addr_len, int flags);
509 static int unix_socketpair(struct socket *, struct socket *);
510 static int unix_accept(struct socket *, struct socket *, int);
511 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
512 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
513 static unsigned int unix_dgram_poll(struct file *, struct socket *,
514 				    poll_table *);
515 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
516 static int unix_shutdown(struct socket *, int);
517 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
518 			       struct msghdr *, size_t);
519 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
520 			       struct msghdr *, size_t, int);
521 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
522 			      struct msghdr *, size_t);
523 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
524 			      struct msghdr *, size_t, int);
525 static int unix_dgram_connect(struct socket *, struct sockaddr *,
526 			      int, int);
527 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
528 				  struct msghdr *, size_t);
529 static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *,
530 				  struct msghdr *, size_t, int);
531 
532 static void unix_set_peek_off(struct sock *sk, int val)
533 {
534 	struct unix_sock *u = unix_sk(sk);
535 
536 	mutex_lock(&u->readlock);
537 	sk->sk_peek_off = val;
538 	mutex_unlock(&u->readlock);
539 }
540 
541 
542 static const struct proto_ops unix_stream_ops = {
543 	.family =	PF_UNIX,
544 	.owner =	THIS_MODULE,
545 	.release =	unix_release,
546 	.bind =		unix_bind,
547 	.connect =	unix_stream_connect,
548 	.socketpair =	unix_socketpair,
549 	.accept =	unix_accept,
550 	.getname =	unix_getname,
551 	.poll =		unix_poll,
552 	.ioctl =	unix_ioctl,
553 	.listen =	unix_listen,
554 	.shutdown =	unix_shutdown,
555 	.setsockopt =	sock_no_setsockopt,
556 	.getsockopt =	sock_no_getsockopt,
557 	.sendmsg =	unix_stream_sendmsg,
558 	.recvmsg =	unix_stream_recvmsg,
559 	.mmap =		sock_no_mmap,
560 	.sendpage =	sock_no_sendpage,
561 	.set_peek_off =	unix_set_peek_off,
562 };
563 
564 static const struct proto_ops unix_dgram_ops = {
565 	.family =	PF_UNIX,
566 	.owner =	THIS_MODULE,
567 	.release =	unix_release,
568 	.bind =		unix_bind,
569 	.connect =	unix_dgram_connect,
570 	.socketpair =	unix_socketpair,
571 	.accept =	sock_no_accept,
572 	.getname =	unix_getname,
573 	.poll =		unix_dgram_poll,
574 	.ioctl =	unix_ioctl,
575 	.listen =	sock_no_listen,
576 	.shutdown =	unix_shutdown,
577 	.setsockopt =	sock_no_setsockopt,
578 	.getsockopt =	sock_no_getsockopt,
579 	.sendmsg =	unix_dgram_sendmsg,
580 	.recvmsg =	unix_dgram_recvmsg,
581 	.mmap =		sock_no_mmap,
582 	.sendpage =	sock_no_sendpage,
583 	.set_peek_off =	unix_set_peek_off,
584 };
585 
586 static const struct proto_ops unix_seqpacket_ops = {
587 	.family =	PF_UNIX,
588 	.owner =	THIS_MODULE,
589 	.release =	unix_release,
590 	.bind =		unix_bind,
591 	.connect =	unix_stream_connect,
592 	.socketpair =	unix_socketpair,
593 	.accept =	unix_accept,
594 	.getname =	unix_getname,
595 	.poll =		unix_dgram_poll,
596 	.ioctl =	unix_ioctl,
597 	.listen =	unix_listen,
598 	.shutdown =	unix_shutdown,
599 	.setsockopt =	sock_no_setsockopt,
600 	.getsockopt =	sock_no_getsockopt,
601 	.sendmsg =	unix_seqpacket_sendmsg,
602 	.recvmsg =	unix_seqpacket_recvmsg,
603 	.mmap =		sock_no_mmap,
604 	.sendpage =	sock_no_sendpage,
605 	.set_peek_off =	unix_set_peek_off,
606 };
607 
608 static struct proto unix_proto = {
609 	.name			= "UNIX",
610 	.owner			= THIS_MODULE,
611 	.obj_size		= sizeof(struct unix_sock),
612 };
613 
614 /*
615  * AF_UNIX sockets do not interact with hardware, hence they
616  * dont trigger interrupts - so it's safe for them to have
617  * bh-unsafe locking for their sk_receive_queue.lock. Split off
618  * this special lock-class by reinitializing the spinlock key:
619  */
620 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
621 
622 static struct sock *unix_create1(struct net *net, struct socket *sock)
623 {
624 	struct sock *sk = NULL;
625 	struct unix_sock *u;
626 
627 	atomic_long_inc(&unix_nr_socks);
628 	if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
629 		goto out;
630 
631 	sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
632 	if (!sk)
633 		goto out;
634 
635 	sock_init_data(sock, sk);
636 	lockdep_set_class(&sk->sk_receive_queue.lock,
637 				&af_unix_sk_receive_queue_lock_key);
638 
639 	sk->sk_write_space	= unix_write_space;
640 	sk->sk_max_ack_backlog	= net->unx.sysctl_max_dgram_qlen;
641 	sk->sk_destruct		= unix_sock_destructor;
642 	u	  = unix_sk(sk);
643 	u->path.dentry = NULL;
644 	u->path.mnt = NULL;
645 	spin_lock_init(&u->lock);
646 	atomic_long_set(&u->inflight, 0);
647 	INIT_LIST_HEAD(&u->link);
648 	mutex_init(&u->readlock); /* single task reading lock */
649 	init_waitqueue_head(&u->peer_wait);
650 	unix_insert_socket(unix_sockets_unbound(sk), sk);
651 out:
652 	if (sk == NULL)
653 		atomic_long_dec(&unix_nr_socks);
654 	else {
655 		local_bh_disable();
656 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
657 		local_bh_enable();
658 	}
659 	return sk;
660 }
661 
662 static int unix_create(struct net *net, struct socket *sock, int protocol,
663 		       int kern)
664 {
665 	if (protocol && protocol != PF_UNIX)
666 		return -EPROTONOSUPPORT;
667 
668 	sock->state = SS_UNCONNECTED;
669 
670 	switch (sock->type) {
671 	case SOCK_STREAM:
672 		sock->ops = &unix_stream_ops;
673 		break;
674 		/*
675 		 *	Believe it or not BSD has AF_UNIX, SOCK_RAW though
676 		 *	nothing uses it.
677 		 */
678 	case SOCK_RAW:
679 		sock->type = SOCK_DGRAM;
680 	case SOCK_DGRAM:
681 		sock->ops = &unix_dgram_ops;
682 		break;
683 	case SOCK_SEQPACKET:
684 		sock->ops = &unix_seqpacket_ops;
685 		break;
686 	default:
687 		return -ESOCKTNOSUPPORT;
688 	}
689 
690 	return unix_create1(net, sock) ? 0 : -ENOMEM;
691 }
692 
693 static int unix_release(struct socket *sock)
694 {
695 	struct sock *sk = sock->sk;
696 
697 	if (!sk)
698 		return 0;
699 
700 	unix_release_sock(sk, 0);
701 	sock->sk = NULL;
702 
703 	return 0;
704 }
705 
706 static int unix_autobind(struct socket *sock)
707 {
708 	struct sock *sk = sock->sk;
709 	struct net *net = sock_net(sk);
710 	struct unix_sock *u = unix_sk(sk);
711 	static u32 ordernum = 1;
712 	struct unix_address *addr;
713 	int err;
714 	unsigned int retries = 0;
715 
716 	mutex_lock(&u->readlock);
717 
718 	err = 0;
719 	if (u->addr)
720 		goto out;
721 
722 	err = -ENOMEM;
723 	addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
724 	if (!addr)
725 		goto out;
726 
727 	addr->name->sun_family = AF_UNIX;
728 	atomic_set(&addr->refcnt, 1);
729 
730 retry:
731 	addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
732 	addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
733 
734 	spin_lock(&unix_table_lock);
735 	ordernum = (ordernum+1)&0xFFFFF;
736 
737 	if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
738 				      addr->hash)) {
739 		spin_unlock(&unix_table_lock);
740 		/*
741 		 * __unix_find_socket_byname() may take long time if many names
742 		 * are already in use.
743 		 */
744 		cond_resched();
745 		/* Give up if all names seems to be in use. */
746 		if (retries++ == 0xFFFFF) {
747 			err = -ENOSPC;
748 			kfree(addr);
749 			goto out;
750 		}
751 		goto retry;
752 	}
753 	addr->hash ^= sk->sk_type;
754 
755 	__unix_remove_socket(sk);
756 	u->addr = addr;
757 	__unix_insert_socket(&unix_socket_table[addr->hash], sk);
758 	spin_unlock(&unix_table_lock);
759 	err = 0;
760 
761 out:	mutex_unlock(&u->readlock);
762 	return err;
763 }
764 
765 static struct sock *unix_find_other(struct net *net,
766 				    struct sockaddr_un *sunname, int len,
767 				    int type, unsigned int hash, int *error)
768 {
769 	struct sock *u;
770 	struct path path;
771 	int err = 0;
772 
773 	if (sunname->sun_path[0]) {
774 		struct inode *inode;
775 		err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
776 		if (err)
777 			goto fail;
778 		inode = path.dentry->d_inode;
779 		err = inode_permission(inode, MAY_WRITE);
780 		if (err)
781 			goto put_fail;
782 
783 		err = -ECONNREFUSED;
784 		if (!S_ISSOCK(inode->i_mode))
785 			goto put_fail;
786 		u = unix_find_socket_byinode(inode);
787 		if (!u)
788 			goto put_fail;
789 
790 		if (u->sk_type == type)
791 			touch_atime(&path);
792 
793 		path_put(&path);
794 
795 		err = -EPROTOTYPE;
796 		if (u->sk_type != type) {
797 			sock_put(u);
798 			goto fail;
799 		}
800 	} else {
801 		err = -ECONNREFUSED;
802 		u = unix_find_socket_byname(net, sunname, len, type, hash);
803 		if (u) {
804 			struct dentry *dentry;
805 			dentry = unix_sk(u)->path.dentry;
806 			if (dentry)
807 				touch_atime(&unix_sk(u)->path);
808 		} else
809 			goto fail;
810 	}
811 	return u;
812 
813 put_fail:
814 	path_put(&path);
815 fail:
816 	*error = err;
817 	return NULL;
818 }
819 
820 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
821 {
822 	struct dentry *dentry;
823 	struct path path;
824 	int err = 0;
825 	/*
826 	 * Get the parent directory, calculate the hash for last
827 	 * component.
828 	 */
829 	dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
830 	err = PTR_ERR(dentry);
831 	if (IS_ERR(dentry))
832 		return err;
833 
834 	/*
835 	 * All right, let's create it.
836 	 */
837 	err = security_path_mknod(&path, dentry, mode, 0);
838 	if (!err) {
839 		err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
840 		if (!err) {
841 			res->mnt = mntget(path.mnt);
842 			res->dentry = dget(dentry);
843 		}
844 	}
845 	done_path_create(&path, dentry);
846 	return err;
847 }
848 
849 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
850 {
851 	struct sock *sk = sock->sk;
852 	struct net *net = sock_net(sk);
853 	struct unix_sock *u = unix_sk(sk);
854 	struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
855 	char *sun_path = sunaddr->sun_path;
856 	int err;
857 	unsigned int hash;
858 	struct unix_address *addr;
859 	struct hlist_head *list;
860 
861 	err = -EINVAL;
862 	if (sunaddr->sun_family != AF_UNIX)
863 		goto out;
864 
865 	if (addr_len == sizeof(short)) {
866 		err = unix_autobind(sock);
867 		goto out;
868 	}
869 
870 	err = unix_mkname(sunaddr, addr_len, &hash);
871 	if (err < 0)
872 		goto out;
873 	addr_len = err;
874 
875 	mutex_lock(&u->readlock);
876 
877 	err = -EINVAL;
878 	if (u->addr)
879 		goto out_up;
880 
881 	err = -ENOMEM;
882 	addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
883 	if (!addr)
884 		goto out_up;
885 
886 	memcpy(addr->name, sunaddr, addr_len);
887 	addr->len = addr_len;
888 	addr->hash = hash ^ sk->sk_type;
889 	atomic_set(&addr->refcnt, 1);
890 
891 	if (sun_path[0]) {
892 		struct path path;
893 		umode_t mode = S_IFSOCK |
894 		       (SOCK_INODE(sock)->i_mode & ~current_umask());
895 		err = unix_mknod(sun_path, mode, &path);
896 		if (err) {
897 			if (err == -EEXIST)
898 				err = -EADDRINUSE;
899 			unix_release_addr(addr);
900 			goto out_up;
901 		}
902 		addr->hash = UNIX_HASH_SIZE;
903 		hash = path.dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1);
904 		spin_lock(&unix_table_lock);
905 		u->path = path;
906 		list = &unix_socket_table[hash];
907 	} else {
908 		spin_lock(&unix_table_lock);
909 		err = -EADDRINUSE;
910 		if (__unix_find_socket_byname(net, sunaddr, addr_len,
911 					      sk->sk_type, hash)) {
912 			unix_release_addr(addr);
913 			goto out_unlock;
914 		}
915 
916 		list = &unix_socket_table[addr->hash];
917 	}
918 
919 	err = 0;
920 	__unix_remove_socket(sk);
921 	u->addr = addr;
922 	__unix_insert_socket(list, sk);
923 
924 out_unlock:
925 	spin_unlock(&unix_table_lock);
926 out_up:
927 	mutex_unlock(&u->readlock);
928 out:
929 	return err;
930 }
931 
932 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
933 {
934 	if (unlikely(sk1 == sk2) || !sk2) {
935 		unix_state_lock(sk1);
936 		return;
937 	}
938 	if (sk1 < sk2) {
939 		unix_state_lock(sk1);
940 		unix_state_lock_nested(sk2);
941 	} else {
942 		unix_state_lock(sk2);
943 		unix_state_lock_nested(sk1);
944 	}
945 }
946 
947 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
948 {
949 	if (unlikely(sk1 == sk2) || !sk2) {
950 		unix_state_unlock(sk1);
951 		return;
952 	}
953 	unix_state_unlock(sk1);
954 	unix_state_unlock(sk2);
955 }
956 
957 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
958 			      int alen, int flags)
959 {
960 	struct sock *sk = sock->sk;
961 	struct net *net = sock_net(sk);
962 	struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
963 	struct sock *other;
964 	unsigned int hash;
965 	int err;
966 
967 	if (addr->sa_family != AF_UNSPEC) {
968 		err = unix_mkname(sunaddr, alen, &hash);
969 		if (err < 0)
970 			goto out;
971 		alen = err;
972 
973 		if (test_bit(SOCK_PASSCRED, &sock->flags) &&
974 		    !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
975 			goto out;
976 
977 restart:
978 		other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
979 		if (!other)
980 			goto out;
981 
982 		unix_state_double_lock(sk, other);
983 
984 		/* Apparently VFS overslept socket death. Retry. */
985 		if (sock_flag(other, SOCK_DEAD)) {
986 			unix_state_double_unlock(sk, other);
987 			sock_put(other);
988 			goto restart;
989 		}
990 
991 		err = -EPERM;
992 		if (!unix_may_send(sk, other))
993 			goto out_unlock;
994 
995 		err = security_unix_may_send(sk->sk_socket, other->sk_socket);
996 		if (err)
997 			goto out_unlock;
998 
999 	} else {
1000 		/*
1001 		 *	1003.1g breaking connected state with AF_UNSPEC
1002 		 */
1003 		other = NULL;
1004 		unix_state_double_lock(sk, other);
1005 	}
1006 
1007 	/*
1008 	 * If it was connected, reconnect.
1009 	 */
1010 	if (unix_peer(sk)) {
1011 		struct sock *old_peer = unix_peer(sk);
1012 		unix_peer(sk) = other;
1013 		unix_state_double_unlock(sk, other);
1014 
1015 		if (other != old_peer)
1016 			unix_dgram_disconnected(sk, old_peer);
1017 		sock_put(old_peer);
1018 	} else {
1019 		unix_peer(sk) = other;
1020 		unix_state_double_unlock(sk, other);
1021 	}
1022 	return 0;
1023 
1024 out_unlock:
1025 	unix_state_double_unlock(sk, other);
1026 	sock_put(other);
1027 out:
1028 	return err;
1029 }
1030 
1031 static long unix_wait_for_peer(struct sock *other, long timeo)
1032 {
1033 	struct unix_sock *u = unix_sk(other);
1034 	int sched;
1035 	DEFINE_WAIT(wait);
1036 
1037 	prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1038 
1039 	sched = !sock_flag(other, SOCK_DEAD) &&
1040 		!(other->sk_shutdown & RCV_SHUTDOWN) &&
1041 		unix_recvq_full(other);
1042 
1043 	unix_state_unlock(other);
1044 
1045 	if (sched)
1046 		timeo = schedule_timeout(timeo);
1047 
1048 	finish_wait(&u->peer_wait, &wait);
1049 	return timeo;
1050 }
1051 
1052 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1053 			       int addr_len, int flags)
1054 {
1055 	struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1056 	struct sock *sk = sock->sk;
1057 	struct net *net = sock_net(sk);
1058 	struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1059 	struct sock *newsk = NULL;
1060 	struct sock *other = NULL;
1061 	struct sk_buff *skb = NULL;
1062 	unsigned int hash;
1063 	int st;
1064 	int err;
1065 	long timeo;
1066 
1067 	err = unix_mkname(sunaddr, addr_len, &hash);
1068 	if (err < 0)
1069 		goto out;
1070 	addr_len = err;
1071 
1072 	if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1073 	    (err = unix_autobind(sock)) != 0)
1074 		goto out;
1075 
1076 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1077 
1078 	/* First of all allocate resources.
1079 	   If we will make it after state is locked,
1080 	   we will have to recheck all again in any case.
1081 	 */
1082 
1083 	err = -ENOMEM;
1084 
1085 	/* create new sock for complete connection */
1086 	newsk = unix_create1(sock_net(sk), NULL);
1087 	if (newsk == NULL)
1088 		goto out;
1089 
1090 	/* Allocate skb for sending to listening sock */
1091 	skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1092 	if (skb == NULL)
1093 		goto out;
1094 
1095 restart:
1096 	/*  Find listening sock. */
1097 	other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1098 	if (!other)
1099 		goto out;
1100 
1101 	/* Latch state of peer */
1102 	unix_state_lock(other);
1103 
1104 	/* Apparently VFS overslept socket death. Retry. */
1105 	if (sock_flag(other, SOCK_DEAD)) {
1106 		unix_state_unlock(other);
1107 		sock_put(other);
1108 		goto restart;
1109 	}
1110 
1111 	err = -ECONNREFUSED;
1112 	if (other->sk_state != TCP_LISTEN)
1113 		goto out_unlock;
1114 	if (other->sk_shutdown & RCV_SHUTDOWN)
1115 		goto out_unlock;
1116 
1117 	if (unix_recvq_full(other)) {
1118 		err = -EAGAIN;
1119 		if (!timeo)
1120 			goto out_unlock;
1121 
1122 		timeo = unix_wait_for_peer(other, timeo);
1123 
1124 		err = sock_intr_errno(timeo);
1125 		if (signal_pending(current))
1126 			goto out;
1127 		sock_put(other);
1128 		goto restart;
1129 	}
1130 
1131 	/* Latch our state.
1132 
1133 	   It is tricky place. We need to grab our state lock and cannot
1134 	   drop lock on peer. It is dangerous because deadlock is
1135 	   possible. Connect to self case and simultaneous
1136 	   attempt to connect are eliminated by checking socket
1137 	   state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1138 	   check this before attempt to grab lock.
1139 
1140 	   Well, and we have to recheck the state after socket locked.
1141 	 */
1142 	st = sk->sk_state;
1143 
1144 	switch (st) {
1145 	case TCP_CLOSE:
1146 		/* This is ok... continue with connect */
1147 		break;
1148 	case TCP_ESTABLISHED:
1149 		/* Socket is already connected */
1150 		err = -EISCONN;
1151 		goto out_unlock;
1152 	default:
1153 		err = -EINVAL;
1154 		goto out_unlock;
1155 	}
1156 
1157 	unix_state_lock_nested(sk);
1158 
1159 	if (sk->sk_state != st) {
1160 		unix_state_unlock(sk);
1161 		unix_state_unlock(other);
1162 		sock_put(other);
1163 		goto restart;
1164 	}
1165 
1166 	err = security_unix_stream_connect(sk, other, newsk);
1167 	if (err) {
1168 		unix_state_unlock(sk);
1169 		goto out_unlock;
1170 	}
1171 
1172 	/* The way is open! Fastly set all the necessary fields... */
1173 
1174 	sock_hold(sk);
1175 	unix_peer(newsk)	= sk;
1176 	newsk->sk_state		= TCP_ESTABLISHED;
1177 	newsk->sk_type		= sk->sk_type;
1178 	init_peercred(newsk);
1179 	newu = unix_sk(newsk);
1180 	RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1181 	otheru = unix_sk(other);
1182 
1183 	/* copy address information from listening to new sock*/
1184 	if (otheru->addr) {
1185 		atomic_inc(&otheru->addr->refcnt);
1186 		newu->addr = otheru->addr;
1187 	}
1188 	if (otheru->path.dentry) {
1189 		path_get(&otheru->path);
1190 		newu->path = otheru->path;
1191 	}
1192 
1193 	/* Set credentials */
1194 	copy_peercred(sk, other);
1195 
1196 	sock->state	= SS_CONNECTED;
1197 	sk->sk_state	= TCP_ESTABLISHED;
1198 	sock_hold(newsk);
1199 
1200 	smp_mb__after_atomic_inc();	/* sock_hold() does an atomic_inc() */
1201 	unix_peer(sk)	= newsk;
1202 
1203 	unix_state_unlock(sk);
1204 
1205 	/* take ten and and send info to listening sock */
1206 	spin_lock(&other->sk_receive_queue.lock);
1207 	__skb_queue_tail(&other->sk_receive_queue, skb);
1208 	spin_unlock(&other->sk_receive_queue.lock);
1209 	unix_state_unlock(other);
1210 	other->sk_data_ready(other, 0);
1211 	sock_put(other);
1212 	return 0;
1213 
1214 out_unlock:
1215 	if (other)
1216 		unix_state_unlock(other);
1217 
1218 out:
1219 	kfree_skb(skb);
1220 	if (newsk)
1221 		unix_release_sock(newsk, 0);
1222 	if (other)
1223 		sock_put(other);
1224 	return err;
1225 }
1226 
1227 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1228 {
1229 	struct sock *ska = socka->sk, *skb = sockb->sk;
1230 
1231 	/* Join our sockets back to back */
1232 	sock_hold(ska);
1233 	sock_hold(skb);
1234 	unix_peer(ska) = skb;
1235 	unix_peer(skb) = ska;
1236 	init_peercred(ska);
1237 	init_peercred(skb);
1238 
1239 	if (ska->sk_type != SOCK_DGRAM) {
1240 		ska->sk_state = TCP_ESTABLISHED;
1241 		skb->sk_state = TCP_ESTABLISHED;
1242 		socka->state  = SS_CONNECTED;
1243 		sockb->state  = SS_CONNECTED;
1244 	}
1245 	return 0;
1246 }
1247 
1248 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1249 {
1250 	struct sock *sk = sock->sk;
1251 	struct sock *tsk;
1252 	struct sk_buff *skb;
1253 	int err;
1254 
1255 	err = -EOPNOTSUPP;
1256 	if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1257 		goto out;
1258 
1259 	err = -EINVAL;
1260 	if (sk->sk_state != TCP_LISTEN)
1261 		goto out;
1262 
1263 	/* If socket state is TCP_LISTEN it cannot change (for now...),
1264 	 * so that no locks are necessary.
1265 	 */
1266 
1267 	skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1268 	if (!skb) {
1269 		/* This means receive shutdown. */
1270 		if (err == 0)
1271 			err = -EINVAL;
1272 		goto out;
1273 	}
1274 
1275 	tsk = skb->sk;
1276 	skb_free_datagram(sk, skb);
1277 	wake_up_interruptible(&unix_sk(sk)->peer_wait);
1278 
1279 	/* attach accepted sock to socket */
1280 	unix_state_lock(tsk);
1281 	newsock->state = SS_CONNECTED;
1282 	sock_graft(tsk, newsock);
1283 	unix_state_unlock(tsk);
1284 	return 0;
1285 
1286 out:
1287 	return err;
1288 }
1289 
1290 
1291 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1292 {
1293 	struct sock *sk = sock->sk;
1294 	struct unix_sock *u;
1295 	DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1296 	int err = 0;
1297 
1298 	if (peer) {
1299 		sk = unix_peer_get(sk);
1300 
1301 		err = -ENOTCONN;
1302 		if (!sk)
1303 			goto out;
1304 		err = 0;
1305 	} else {
1306 		sock_hold(sk);
1307 	}
1308 
1309 	u = unix_sk(sk);
1310 	unix_state_lock(sk);
1311 	if (!u->addr) {
1312 		sunaddr->sun_family = AF_UNIX;
1313 		sunaddr->sun_path[0] = 0;
1314 		*uaddr_len = sizeof(short);
1315 	} else {
1316 		struct unix_address *addr = u->addr;
1317 
1318 		*uaddr_len = addr->len;
1319 		memcpy(sunaddr, addr->name, *uaddr_len);
1320 	}
1321 	unix_state_unlock(sk);
1322 	sock_put(sk);
1323 out:
1324 	return err;
1325 }
1326 
1327 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1328 {
1329 	int i;
1330 
1331 	scm->fp = UNIXCB(skb).fp;
1332 	UNIXCB(skb).fp = NULL;
1333 
1334 	for (i = scm->fp->count-1; i >= 0; i--)
1335 		unix_notinflight(scm->fp->fp[i]);
1336 }
1337 
1338 static void unix_destruct_scm(struct sk_buff *skb)
1339 {
1340 	struct scm_cookie scm;
1341 	memset(&scm, 0, sizeof(scm));
1342 	scm.pid  = UNIXCB(skb).pid;
1343 	if (UNIXCB(skb).fp)
1344 		unix_detach_fds(&scm, skb);
1345 
1346 	/* Alas, it calls VFS */
1347 	/* So fscking what? fput() had been SMP-safe since the last Summer */
1348 	scm_destroy(&scm);
1349 	sock_wfree(skb);
1350 }
1351 
1352 #define MAX_RECURSION_LEVEL 4
1353 
1354 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1355 {
1356 	int i;
1357 	unsigned char max_level = 0;
1358 	int unix_sock_count = 0;
1359 
1360 	for (i = scm->fp->count - 1; i >= 0; i--) {
1361 		struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1362 
1363 		if (sk) {
1364 			unix_sock_count++;
1365 			max_level = max(max_level,
1366 					unix_sk(sk)->recursion_level);
1367 		}
1368 	}
1369 	if (unlikely(max_level > MAX_RECURSION_LEVEL))
1370 		return -ETOOMANYREFS;
1371 
1372 	/*
1373 	 * Need to duplicate file references for the sake of garbage
1374 	 * collection.  Otherwise a socket in the fps might become a
1375 	 * candidate for GC while the skb is not yet queued.
1376 	 */
1377 	UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1378 	if (!UNIXCB(skb).fp)
1379 		return -ENOMEM;
1380 
1381 	if (unix_sock_count) {
1382 		for (i = scm->fp->count - 1; i >= 0; i--)
1383 			unix_inflight(scm->fp->fp[i]);
1384 	}
1385 	return max_level;
1386 }
1387 
1388 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1389 {
1390 	int err = 0;
1391 
1392 	UNIXCB(skb).pid  = get_pid(scm->pid);
1393 	UNIXCB(skb).uid = scm->creds.uid;
1394 	UNIXCB(skb).gid = scm->creds.gid;
1395 	UNIXCB(skb).fp = NULL;
1396 	if (scm->fp && send_fds)
1397 		err = unix_attach_fds(scm, skb);
1398 
1399 	skb->destructor = unix_destruct_scm;
1400 	return err;
1401 }
1402 
1403 /*
1404  * Some apps rely on write() giving SCM_CREDENTIALS
1405  * We include credentials if source or destination socket
1406  * asserted SOCK_PASSCRED.
1407  */
1408 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1409 			    const struct sock *other)
1410 {
1411 	if (UNIXCB(skb).pid)
1412 		return;
1413 	if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1414 	    !other->sk_socket ||
1415 	    test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1416 		UNIXCB(skb).pid  = get_pid(task_tgid(current));
1417 		current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1418 	}
1419 }
1420 
1421 /*
1422  *	Send AF_UNIX data.
1423  */
1424 
1425 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1426 			      struct msghdr *msg, size_t len)
1427 {
1428 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1429 	struct sock *sk = sock->sk;
1430 	struct net *net = sock_net(sk);
1431 	struct unix_sock *u = unix_sk(sk);
1432 	struct sockaddr_un *sunaddr = msg->msg_name;
1433 	struct sock *other = NULL;
1434 	int namelen = 0; /* fake GCC */
1435 	int err;
1436 	unsigned int hash;
1437 	struct sk_buff *skb;
1438 	long timeo;
1439 	struct scm_cookie tmp_scm;
1440 	int max_level;
1441 	int data_len = 0;
1442 
1443 	if (NULL == siocb->scm)
1444 		siocb->scm = &tmp_scm;
1445 	wait_for_unix_gc();
1446 	err = scm_send(sock, msg, siocb->scm, false);
1447 	if (err < 0)
1448 		return err;
1449 
1450 	err = -EOPNOTSUPP;
1451 	if (msg->msg_flags&MSG_OOB)
1452 		goto out;
1453 
1454 	if (msg->msg_namelen) {
1455 		err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1456 		if (err < 0)
1457 			goto out;
1458 		namelen = err;
1459 	} else {
1460 		sunaddr = NULL;
1461 		err = -ENOTCONN;
1462 		other = unix_peer_get(sk);
1463 		if (!other)
1464 			goto out;
1465 	}
1466 
1467 	if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1468 	    && (err = unix_autobind(sock)) != 0)
1469 		goto out;
1470 
1471 	err = -EMSGSIZE;
1472 	if (len > sk->sk_sndbuf - 32)
1473 		goto out;
1474 
1475 	if (len > SKB_MAX_ALLOC)
1476 		data_len = min_t(size_t,
1477 				 len - SKB_MAX_ALLOC,
1478 				 MAX_SKB_FRAGS * PAGE_SIZE);
1479 
1480 	skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1481 				   msg->msg_flags & MSG_DONTWAIT, &err);
1482 	if (skb == NULL)
1483 		goto out;
1484 
1485 	err = unix_scm_to_skb(siocb->scm, skb, true);
1486 	if (err < 0)
1487 		goto out_free;
1488 	max_level = err + 1;
1489 	unix_get_secdata(siocb->scm, skb);
1490 
1491 	skb_put(skb, len - data_len);
1492 	skb->data_len = data_len;
1493 	skb->len = len;
1494 	err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len);
1495 	if (err)
1496 		goto out_free;
1497 
1498 	timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1499 
1500 restart:
1501 	if (!other) {
1502 		err = -ECONNRESET;
1503 		if (sunaddr == NULL)
1504 			goto out_free;
1505 
1506 		other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1507 					hash, &err);
1508 		if (other == NULL)
1509 			goto out_free;
1510 	}
1511 
1512 	if (sk_filter(other, skb) < 0) {
1513 		/* Toss the packet but do not return any error to the sender */
1514 		err = len;
1515 		goto out_free;
1516 	}
1517 
1518 	unix_state_lock(other);
1519 	err = -EPERM;
1520 	if (!unix_may_send(sk, other))
1521 		goto out_unlock;
1522 
1523 	if (sock_flag(other, SOCK_DEAD)) {
1524 		/*
1525 		 *	Check with 1003.1g - what should
1526 		 *	datagram error
1527 		 */
1528 		unix_state_unlock(other);
1529 		sock_put(other);
1530 
1531 		err = 0;
1532 		unix_state_lock(sk);
1533 		if (unix_peer(sk) == other) {
1534 			unix_peer(sk) = NULL;
1535 			unix_state_unlock(sk);
1536 
1537 			unix_dgram_disconnected(sk, other);
1538 			sock_put(other);
1539 			err = -ECONNREFUSED;
1540 		} else {
1541 			unix_state_unlock(sk);
1542 		}
1543 
1544 		other = NULL;
1545 		if (err)
1546 			goto out_free;
1547 		goto restart;
1548 	}
1549 
1550 	err = -EPIPE;
1551 	if (other->sk_shutdown & RCV_SHUTDOWN)
1552 		goto out_unlock;
1553 
1554 	if (sk->sk_type != SOCK_SEQPACKET) {
1555 		err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1556 		if (err)
1557 			goto out_unlock;
1558 	}
1559 
1560 	if (unix_peer(other) != sk && unix_recvq_full(other)) {
1561 		if (!timeo) {
1562 			err = -EAGAIN;
1563 			goto out_unlock;
1564 		}
1565 
1566 		timeo = unix_wait_for_peer(other, timeo);
1567 
1568 		err = sock_intr_errno(timeo);
1569 		if (signal_pending(current))
1570 			goto out_free;
1571 
1572 		goto restart;
1573 	}
1574 
1575 	if (sock_flag(other, SOCK_RCVTSTAMP))
1576 		__net_timestamp(skb);
1577 	maybe_add_creds(skb, sock, other);
1578 	skb_queue_tail(&other->sk_receive_queue, skb);
1579 	if (max_level > unix_sk(other)->recursion_level)
1580 		unix_sk(other)->recursion_level = max_level;
1581 	unix_state_unlock(other);
1582 	other->sk_data_ready(other, len);
1583 	sock_put(other);
1584 	scm_destroy(siocb->scm);
1585 	return len;
1586 
1587 out_unlock:
1588 	unix_state_unlock(other);
1589 out_free:
1590 	kfree_skb(skb);
1591 out:
1592 	if (other)
1593 		sock_put(other);
1594 	scm_destroy(siocb->scm);
1595 	return err;
1596 }
1597 
1598 
1599 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1600 			       struct msghdr *msg, size_t len)
1601 {
1602 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1603 	struct sock *sk = sock->sk;
1604 	struct sock *other = NULL;
1605 	int err, size;
1606 	struct sk_buff *skb;
1607 	int sent = 0;
1608 	struct scm_cookie tmp_scm;
1609 	bool fds_sent = false;
1610 	int max_level;
1611 
1612 	if (NULL == siocb->scm)
1613 		siocb->scm = &tmp_scm;
1614 	wait_for_unix_gc();
1615 	err = scm_send(sock, msg, siocb->scm, false);
1616 	if (err < 0)
1617 		return err;
1618 
1619 	err = -EOPNOTSUPP;
1620 	if (msg->msg_flags&MSG_OOB)
1621 		goto out_err;
1622 
1623 	if (msg->msg_namelen) {
1624 		err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1625 		goto out_err;
1626 	} else {
1627 		err = -ENOTCONN;
1628 		other = unix_peer(sk);
1629 		if (!other)
1630 			goto out_err;
1631 	}
1632 
1633 	if (sk->sk_shutdown & SEND_SHUTDOWN)
1634 		goto pipe_err;
1635 
1636 	while (sent < len) {
1637 		/*
1638 		 *	Optimisation for the fact that under 0.01% of X
1639 		 *	messages typically need breaking up.
1640 		 */
1641 
1642 		size = len-sent;
1643 
1644 		/* Keep two messages in the pipe so it schedules better */
1645 		if (size > ((sk->sk_sndbuf >> 1) - 64))
1646 			size = (sk->sk_sndbuf >> 1) - 64;
1647 
1648 		if (size > SKB_MAX_ALLOC)
1649 			size = SKB_MAX_ALLOC;
1650 
1651 		/*
1652 		 *	Grab a buffer
1653 		 */
1654 
1655 		skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1656 					  &err);
1657 
1658 		if (skb == NULL)
1659 			goto out_err;
1660 
1661 		/*
1662 		 *	If you pass two values to the sock_alloc_send_skb
1663 		 *	it tries to grab the large buffer with GFP_NOFS
1664 		 *	(which can fail easily), and if it fails grab the
1665 		 *	fallback size buffer which is under a page and will
1666 		 *	succeed. [Alan]
1667 		 */
1668 		size = min_t(int, size, skb_tailroom(skb));
1669 
1670 
1671 		/* Only send the fds in the first buffer */
1672 		err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1673 		if (err < 0) {
1674 			kfree_skb(skb);
1675 			goto out_err;
1676 		}
1677 		max_level = err + 1;
1678 		fds_sent = true;
1679 
1680 		err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1681 		if (err) {
1682 			kfree_skb(skb);
1683 			goto out_err;
1684 		}
1685 
1686 		unix_state_lock(other);
1687 
1688 		if (sock_flag(other, SOCK_DEAD) ||
1689 		    (other->sk_shutdown & RCV_SHUTDOWN))
1690 			goto pipe_err_free;
1691 
1692 		maybe_add_creds(skb, sock, other);
1693 		skb_queue_tail(&other->sk_receive_queue, skb);
1694 		if (max_level > unix_sk(other)->recursion_level)
1695 			unix_sk(other)->recursion_level = max_level;
1696 		unix_state_unlock(other);
1697 		other->sk_data_ready(other, size);
1698 		sent += size;
1699 	}
1700 
1701 	scm_destroy(siocb->scm);
1702 	siocb->scm = NULL;
1703 
1704 	return sent;
1705 
1706 pipe_err_free:
1707 	unix_state_unlock(other);
1708 	kfree_skb(skb);
1709 pipe_err:
1710 	if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1711 		send_sig(SIGPIPE, current, 0);
1712 	err = -EPIPE;
1713 out_err:
1714 	scm_destroy(siocb->scm);
1715 	siocb->scm = NULL;
1716 	return sent ? : err;
1717 }
1718 
1719 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1720 				  struct msghdr *msg, size_t len)
1721 {
1722 	int err;
1723 	struct sock *sk = sock->sk;
1724 
1725 	err = sock_error(sk);
1726 	if (err)
1727 		return err;
1728 
1729 	if (sk->sk_state != TCP_ESTABLISHED)
1730 		return -ENOTCONN;
1731 
1732 	if (msg->msg_namelen)
1733 		msg->msg_namelen = 0;
1734 
1735 	return unix_dgram_sendmsg(kiocb, sock, msg, len);
1736 }
1737 
1738 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1739 			      struct msghdr *msg, size_t size,
1740 			      int flags)
1741 {
1742 	struct sock *sk = sock->sk;
1743 
1744 	if (sk->sk_state != TCP_ESTABLISHED)
1745 		return -ENOTCONN;
1746 
1747 	return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1748 }
1749 
1750 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1751 {
1752 	struct unix_sock *u = unix_sk(sk);
1753 
1754 	msg->msg_namelen = 0;
1755 	if (u->addr) {
1756 		msg->msg_namelen = u->addr->len;
1757 		memcpy(msg->msg_name, u->addr->name, u->addr->len);
1758 	}
1759 }
1760 
1761 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1762 			      struct msghdr *msg, size_t size,
1763 			      int flags)
1764 {
1765 	struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1766 	struct scm_cookie tmp_scm;
1767 	struct sock *sk = sock->sk;
1768 	struct unix_sock *u = unix_sk(sk);
1769 	int noblock = flags & MSG_DONTWAIT;
1770 	struct sk_buff *skb;
1771 	int err;
1772 	int peeked, skip;
1773 
1774 	err = -EOPNOTSUPP;
1775 	if (flags&MSG_OOB)
1776 		goto out;
1777 
1778 	msg->msg_namelen = 0;
1779 
1780 	err = mutex_lock_interruptible(&u->readlock);
1781 	if (err) {
1782 		err = sock_intr_errno(sock_rcvtimeo(sk, noblock));
1783 		goto out;
1784 	}
1785 
1786 	skip = sk_peek_offset(sk, flags);
1787 
1788 	skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err);
1789 	if (!skb) {
1790 		unix_state_lock(sk);
1791 		/* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1792 		if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1793 		    (sk->sk_shutdown & RCV_SHUTDOWN))
1794 			err = 0;
1795 		unix_state_unlock(sk);
1796 		goto out_unlock;
1797 	}
1798 
1799 	wake_up_interruptible_sync_poll(&u->peer_wait,
1800 					POLLOUT | POLLWRNORM | POLLWRBAND);
1801 
1802 	if (msg->msg_name)
1803 		unix_copy_addr(msg, skb->sk);
1804 
1805 	if (size > skb->len - skip)
1806 		size = skb->len - skip;
1807 	else if (size < skb->len - skip)
1808 		msg->msg_flags |= MSG_TRUNC;
1809 
1810 	err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size);
1811 	if (err)
1812 		goto out_free;
1813 
1814 	if (sock_flag(sk, SOCK_RCVTSTAMP))
1815 		__sock_recv_timestamp(msg, sk, skb);
1816 
1817 	if (!siocb->scm) {
1818 		siocb->scm = &tmp_scm;
1819 		memset(&tmp_scm, 0, sizeof(tmp_scm));
1820 	}
1821 	scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
1822 	unix_set_secdata(siocb->scm, skb);
1823 
1824 	if (!(flags & MSG_PEEK)) {
1825 		if (UNIXCB(skb).fp)
1826 			unix_detach_fds(siocb->scm, skb);
1827 
1828 		sk_peek_offset_bwd(sk, skb->len);
1829 	} else {
1830 		/* It is questionable: on PEEK we could:
1831 		   - do not return fds - good, but too simple 8)
1832 		   - return fds, and do not return them on read (old strategy,
1833 		     apparently wrong)
1834 		   - clone fds (I chose it for now, it is the most universal
1835 		     solution)
1836 
1837 		   POSIX 1003.1g does not actually define this clearly
1838 		   at all. POSIX 1003.1g doesn't define a lot of things
1839 		   clearly however!
1840 
1841 		*/
1842 
1843 		sk_peek_offset_fwd(sk, size);
1844 
1845 		if (UNIXCB(skb).fp)
1846 			siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1847 	}
1848 	err = (flags & MSG_TRUNC) ? skb->len - skip : size;
1849 
1850 	scm_recv(sock, msg, siocb->scm, flags);
1851 
1852 out_free:
1853 	skb_free_datagram(sk, skb);
1854 out_unlock:
1855 	mutex_unlock(&u->readlock);
1856 out:
1857 	return err;
1858 }
1859 
1860 /*
1861  *	Sleep until more data has arrived. But check for races..
1862  */
1863 static long unix_stream_data_wait(struct sock *sk, long timeo,
1864 				  struct sk_buff *last)
1865 {
1866 	DEFINE_WAIT(wait);
1867 
1868 	unix_state_lock(sk);
1869 
1870 	for (;;) {
1871 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1872 
1873 		if (skb_peek_tail(&sk->sk_receive_queue) != last ||
1874 		    sk->sk_err ||
1875 		    (sk->sk_shutdown & RCV_SHUTDOWN) ||
1876 		    signal_pending(current) ||
1877 		    !timeo)
1878 			break;
1879 
1880 		set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1881 		unix_state_unlock(sk);
1882 		timeo = schedule_timeout(timeo);
1883 		unix_state_lock(sk);
1884 		clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1885 	}
1886 
1887 	finish_wait(sk_sleep(sk), &wait);
1888 	unix_state_unlock(sk);
1889 	return timeo;
1890 }
1891 
1892 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1893 			       struct msghdr *msg, size_t size,
1894 			       int flags)
1895 {
1896 	struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1897 	struct scm_cookie tmp_scm;
1898 	struct sock *sk = sock->sk;
1899 	struct unix_sock *u = unix_sk(sk);
1900 	struct sockaddr_un *sunaddr = msg->msg_name;
1901 	int copied = 0;
1902 	int check_creds = 0;
1903 	int target;
1904 	int err = 0;
1905 	long timeo;
1906 	int skip;
1907 
1908 	err = -EINVAL;
1909 	if (sk->sk_state != TCP_ESTABLISHED)
1910 		goto out;
1911 
1912 	err = -EOPNOTSUPP;
1913 	if (flags&MSG_OOB)
1914 		goto out;
1915 
1916 	target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1917 	timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1918 
1919 	msg->msg_namelen = 0;
1920 
1921 	/* Lock the socket to prevent queue disordering
1922 	 * while sleeps in memcpy_tomsg
1923 	 */
1924 
1925 	if (!siocb->scm) {
1926 		siocb->scm = &tmp_scm;
1927 		memset(&tmp_scm, 0, sizeof(tmp_scm));
1928 	}
1929 
1930 	err = mutex_lock_interruptible(&u->readlock);
1931 	if (err) {
1932 		err = sock_intr_errno(timeo);
1933 		goto out;
1934 	}
1935 
1936 	do {
1937 		int chunk;
1938 		struct sk_buff *skb, *last;
1939 
1940 		unix_state_lock(sk);
1941 		last = skb = skb_peek(&sk->sk_receive_queue);
1942 again:
1943 		if (skb == NULL) {
1944 			unix_sk(sk)->recursion_level = 0;
1945 			if (copied >= target)
1946 				goto unlock;
1947 
1948 			/*
1949 			 *	POSIX 1003.1g mandates this order.
1950 			 */
1951 
1952 			err = sock_error(sk);
1953 			if (err)
1954 				goto unlock;
1955 			if (sk->sk_shutdown & RCV_SHUTDOWN)
1956 				goto unlock;
1957 
1958 			unix_state_unlock(sk);
1959 			err = -EAGAIN;
1960 			if (!timeo)
1961 				break;
1962 			mutex_unlock(&u->readlock);
1963 
1964 			timeo = unix_stream_data_wait(sk, timeo, last);
1965 
1966 			if (signal_pending(current)
1967 			    ||  mutex_lock_interruptible(&u->readlock)) {
1968 				err = sock_intr_errno(timeo);
1969 				goto out;
1970 			}
1971 
1972 			continue;
1973  unlock:
1974 			unix_state_unlock(sk);
1975 			break;
1976 		}
1977 
1978 		skip = sk_peek_offset(sk, flags);
1979 		while (skip >= skb->len) {
1980 			skip -= skb->len;
1981 			last = skb;
1982 			skb = skb_peek_next(skb, &sk->sk_receive_queue);
1983 			if (!skb)
1984 				goto again;
1985 		}
1986 
1987 		unix_state_unlock(sk);
1988 
1989 		if (check_creds) {
1990 			/* Never glue messages from different writers */
1991 			if ((UNIXCB(skb).pid  != siocb->scm->pid) ||
1992 			    !uid_eq(UNIXCB(skb).uid, siocb->scm->creds.uid) ||
1993 			    !gid_eq(UNIXCB(skb).gid, siocb->scm->creds.gid))
1994 				break;
1995 		} else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
1996 			/* Copy credentials */
1997 			scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
1998 			check_creds = 1;
1999 		}
2000 
2001 		/* Copy address just once */
2002 		if (sunaddr) {
2003 			unix_copy_addr(msg, skb->sk);
2004 			sunaddr = NULL;
2005 		}
2006 
2007 		chunk = min_t(unsigned int, skb->len - skip, size);
2008 		if (memcpy_toiovec(msg->msg_iov, skb->data + skip, chunk)) {
2009 			if (copied == 0)
2010 				copied = -EFAULT;
2011 			break;
2012 		}
2013 		copied += chunk;
2014 		size -= chunk;
2015 
2016 		/* Mark read part of skb as used */
2017 		if (!(flags & MSG_PEEK)) {
2018 			skb_pull(skb, chunk);
2019 
2020 			sk_peek_offset_bwd(sk, chunk);
2021 
2022 			if (UNIXCB(skb).fp)
2023 				unix_detach_fds(siocb->scm, skb);
2024 
2025 			if (skb->len)
2026 				break;
2027 
2028 			skb_unlink(skb, &sk->sk_receive_queue);
2029 			consume_skb(skb);
2030 
2031 			if (siocb->scm->fp)
2032 				break;
2033 		} else {
2034 			/* It is questionable, see note in unix_dgram_recvmsg.
2035 			 */
2036 			if (UNIXCB(skb).fp)
2037 				siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2038 
2039 			sk_peek_offset_fwd(sk, chunk);
2040 
2041 			break;
2042 		}
2043 	} while (size);
2044 
2045 	mutex_unlock(&u->readlock);
2046 	scm_recv(sock, msg, siocb->scm, flags);
2047 out:
2048 	return copied ? : err;
2049 }
2050 
2051 static int unix_shutdown(struct socket *sock, int mode)
2052 {
2053 	struct sock *sk = sock->sk;
2054 	struct sock *other;
2055 
2056 	if (mode < SHUT_RD || mode > SHUT_RDWR)
2057 		return -EINVAL;
2058 	/* This maps:
2059 	 * SHUT_RD   (0) -> RCV_SHUTDOWN  (1)
2060 	 * SHUT_WR   (1) -> SEND_SHUTDOWN (2)
2061 	 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2062 	 */
2063 	++mode;
2064 
2065 	unix_state_lock(sk);
2066 	sk->sk_shutdown |= mode;
2067 	other = unix_peer(sk);
2068 	if (other)
2069 		sock_hold(other);
2070 	unix_state_unlock(sk);
2071 	sk->sk_state_change(sk);
2072 
2073 	if (other &&
2074 		(sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2075 
2076 		int peer_mode = 0;
2077 
2078 		if (mode&RCV_SHUTDOWN)
2079 			peer_mode |= SEND_SHUTDOWN;
2080 		if (mode&SEND_SHUTDOWN)
2081 			peer_mode |= RCV_SHUTDOWN;
2082 		unix_state_lock(other);
2083 		other->sk_shutdown |= peer_mode;
2084 		unix_state_unlock(other);
2085 		other->sk_state_change(other);
2086 		if (peer_mode == SHUTDOWN_MASK)
2087 			sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2088 		else if (peer_mode & RCV_SHUTDOWN)
2089 			sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2090 	}
2091 	if (other)
2092 		sock_put(other);
2093 
2094 	return 0;
2095 }
2096 
2097 long unix_inq_len(struct sock *sk)
2098 {
2099 	struct sk_buff *skb;
2100 	long amount = 0;
2101 
2102 	if (sk->sk_state == TCP_LISTEN)
2103 		return -EINVAL;
2104 
2105 	spin_lock(&sk->sk_receive_queue.lock);
2106 	if (sk->sk_type == SOCK_STREAM ||
2107 	    sk->sk_type == SOCK_SEQPACKET) {
2108 		skb_queue_walk(&sk->sk_receive_queue, skb)
2109 			amount += skb->len;
2110 	} else {
2111 		skb = skb_peek(&sk->sk_receive_queue);
2112 		if (skb)
2113 			amount = skb->len;
2114 	}
2115 	spin_unlock(&sk->sk_receive_queue.lock);
2116 
2117 	return amount;
2118 }
2119 EXPORT_SYMBOL_GPL(unix_inq_len);
2120 
2121 long unix_outq_len(struct sock *sk)
2122 {
2123 	return sk_wmem_alloc_get(sk);
2124 }
2125 EXPORT_SYMBOL_GPL(unix_outq_len);
2126 
2127 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2128 {
2129 	struct sock *sk = sock->sk;
2130 	long amount = 0;
2131 	int err;
2132 
2133 	switch (cmd) {
2134 	case SIOCOUTQ:
2135 		amount = unix_outq_len(sk);
2136 		err = put_user(amount, (int __user *)arg);
2137 		break;
2138 	case SIOCINQ:
2139 		amount = unix_inq_len(sk);
2140 		if (amount < 0)
2141 			err = amount;
2142 		else
2143 			err = put_user(amount, (int __user *)arg);
2144 		break;
2145 	default:
2146 		err = -ENOIOCTLCMD;
2147 		break;
2148 	}
2149 	return err;
2150 }
2151 
2152 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2153 {
2154 	struct sock *sk = sock->sk;
2155 	unsigned int mask;
2156 
2157 	sock_poll_wait(file, sk_sleep(sk), wait);
2158 	mask = 0;
2159 
2160 	/* exceptional events? */
2161 	if (sk->sk_err)
2162 		mask |= POLLERR;
2163 	if (sk->sk_shutdown == SHUTDOWN_MASK)
2164 		mask |= POLLHUP;
2165 	if (sk->sk_shutdown & RCV_SHUTDOWN)
2166 		mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2167 
2168 	/* readable? */
2169 	if (!skb_queue_empty(&sk->sk_receive_queue))
2170 		mask |= POLLIN | POLLRDNORM;
2171 
2172 	/* Connection-based need to check for termination and startup */
2173 	if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2174 	    sk->sk_state == TCP_CLOSE)
2175 		mask |= POLLHUP;
2176 
2177 	/*
2178 	 * we set writable also when the other side has shut down the
2179 	 * connection. This prevents stuck sockets.
2180 	 */
2181 	if (unix_writable(sk))
2182 		mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2183 
2184 	return mask;
2185 }
2186 
2187 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2188 				    poll_table *wait)
2189 {
2190 	struct sock *sk = sock->sk, *other;
2191 	unsigned int mask, writable;
2192 
2193 	sock_poll_wait(file, sk_sleep(sk), wait);
2194 	mask = 0;
2195 
2196 	/* exceptional events? */
2197 	if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2198 		mask |= POLLERR |
2199 			(sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2200 
2201 	if (sk->sk_shutdown & RCV_SHUTDOWN)
2202 		mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2203 	if (sk->sk_shutdown == SHUTDOWN_MASK)
2204 		mask |= POLLHUP;
2205 
2206 	/* readable? */
2207 	if (!skb_queue_empty(&sk->sk_receive_queue))
2208 		mask |= POLLIN | POLLRDNORM;
2209 
2210 	/* Connection-based need to check for termination and startup */
2211 	if (sk->sk_type == SOCK_SEQPACKET) {
2212 		if (sk->sk_state == TCP_CLOSE)
2213 			mask |= POLLHUP;
2214 		/* connection hasn't started yet? */
2215 		if (sk->sk_state == TCP_SYN_SENT)
2216 			return mask;
2217 	}
2218 
2219 	/* No write status requested, avoid expensive OUT tests. */
2220 	if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2221 		return mask;
2222 
2223 	writable = unix_writable(sk);
2224 	other = unix_peer_get(sk);
2225 	if (other) {
2226 		if (unix_peer(other) != sk) {
2227 			sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2228 			if (unix_recvq_full(other))
2229 				writable = 0;
2230 		}
2231 		sock_put(other);
2232 	}
2233 
2234 	if (writable)
2235 		mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2236 	else
2237 		set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2238 
2239 	return mask;
2240 }
2241 
2242 #ifdef CONFIG_PROC_FS
2243 
2244 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2245 
2246 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2247 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2248 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2249 
2250 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2251 {
2252 	unsigned long offset = get_offset(*pos);
2253 	unsigned long bucket = get_bucket(*pos);
2254 	struct sock *sk;
2255 	unsigned long count = 0;
2256 
2257 	for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2258 		if (sock_net(sk) != seq_file_net(seq))
2259 			continue;
2260 		if (++count == offset)
2261 			break;
2262 	}
2263 
2264 	return sk;
2265 }
2266 
2267 static struct sock *unix_next_socket(struct seq_file *seq,
2268 				     struct sock *sk,
2269 				     loff_t *pos)
2270 {
2271 	unsigned long bucket;
2272 
2273 	while (sk > (struct sock *)SEQ_START_TOKEN) {
2274 		sk = sk_next(sk);
2275 		if (!sk)
2276 			goto next_bucket;
2277 		if (sock_net(sk) == seq_file_net(seq))
2278 			return sk;
2279 	}
2280 
2281 	do {
2282 		sk = unix_from_bucket(seq, pos);
2283 		if (sk)
2284 			return sk;
2285 
2286 next_bucket:
2287 		bucket = get_bucket(*pos) + 1;
2288 		*pos = set_bucket_offset(bucket, 1);
2289 	} while (bucket < ARRAY_SIZE(unix_socket_table));
2290 
2291 	return NULL;
2292 }
2293 
2294 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2295 	__acquires(unix_table_lock)
2296 {
2297 	spin_lock(&unix_table_lock);
2298 
2299 	if (!*pos)
2300 		return SEQ_START_TOKEN;
2301 
2302 	if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2303 		return NULL;
2304 
2305 	return unix_next_socket(seq, NULL, pos);
2306 }
2307 
2308 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2309 {
2310 	++*pos;
2311 	return unix_next_socket(seq, v, pos);
2312 }
2313 
2314 static void unix_seq_stop(struct seq_file *seq, void *v)
2315 	__releases(unix_table_lock)
2316 {
2317 	spin_unlock(&unix_table_lock);
2318 }
2319 
2320 static int unix_seq_show(struct seq_file *seq, void *v)
2321 {
2322 
2323 	if (v == SEQ_START_TOKEN)
2324 		seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2325 			 "Inode Path\n");
2326 	else {
2327 		struct sock *s = v;
2328 		struct unix_sock *u = unix_sk(s);
2329 		unix_state_lock(s);
2330 
2331 		seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2332 			s,
2333 			atomic_read(&s->sk_refcnt),
2334 			0,
2335 			s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2336 			s->sk_type,
2337 			s->sk_socket ?
2338 			(s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2339 			(s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2340 			sock_i_ino(s));
2341 
2342 		if (u->addr) {
2343 			int i, len;
2344 			seq_putc(seq, ' ');
2345 
2346 			i = 0;
2347 			len = u->addr->len - sizeof(short);
2348 			if (!UNIX_ABSTRACT(s))
2349 				len--;
2350 			else {
2351 				seq_putc(seq, '@');
2352 				i++;
2353 			}
2354 			for ( ; i < len; i++)
2355 				seq_putc(seq, u->addr->name->sun_path[i]);
2356 		}
2357 		unix_state_unlock(s);
2358 		seq_putc(seq, '\n');
2359 	}
2360 
2361 	return 0;
2362 }
2363 
2364 static const struct seq_operations unix_seq_ops = {
2365 	.start  = unix_seq_start,
2366 	.next   = unix_seq_next,
2367 	.stop   = unix_seq_stop,
2368 	.show   = unix_seq_show,
2369 };
2370 
2371 static int unix_seq_open(struct inode *inode, struct file *file)
2372 {
2373 	return seq_open_net(inode, file, &unix_seq_ops,
2374 			    sizeof(struct seq_net_private));
2375 }
2376 
2377 static const struct file_operations unix_seq_fops = {
2378 	.owner		= THIS_MODULE,
2379 	.open		= unix_seq_open,
2380 	.read		= seq_read,
2381 	.llseek		= seq_lseek,
2382 	.release	= seq_release_net,
2383 };
2384 
2385 #endif
2386 
2387 static const struct net_proto_family unix_family_ops = {
2388 	.family = PF_UNIX,
2389 	.create = unix_create,
2390 	.owner	= THIS_MODULE,
2391 };
2392 
2393 
2394 static int __net_init unix_net_init(struct net *net)
2395 {
2396 	int error = -ENOMEM;
2397 
2398 	net->unx.sysctl_max_dgram_qlen = 10;
2399 	if (unix_sysctl_register(net))
2400 		goto out;
2401 
2402 #ifdef CONFIG_PROC_FS
2403 	if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2404 		unix_sysctl_unregister(net);
2405 		goto out;
2406 	}
2407 #endif
2408 	error = 0;
2409 out:
2410 	return error;
2411 }
2412 
2413 static void __net_exit unix_net_exit(struct net *net)
2414 {
2415 	unix_sysctl_unregister(net);
2416 	remove_proc_entry("unix", net->proc_net);
2417 }
2418 
2419 static struct pernet_operations unix_net_ops = {
2420 	.init = unix_net_init,
2421 	.exit = unix_net_exit,
2422 };
2423 
2424 static int __init af_unix_init(void)
2425 {
2426 	int rc = -1;
2427 
2428 	BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2429 
2430 	rc = proto_register(&unix_proto, 1);
2431 	if (rc != 0) {
2432 		printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2433 		       __func__);
2434 		goto out;
2435 	}
2436 
2437 	sock_register(&unix_family_ops);
2438 	register_pernet_subsys(&unix_net_ops);
2439 out:
2440 	return rc;
2441 }
2442 
2443 static void __exit af_unix_exit(void)
2444 {
2445 	sock_unregister(PF_UNIX);
2446 	proto_unregister(&unix_proto);
2447 	unregister_pernet_subsys(&unix_net_ops);
2448 }
2449 
2450 /* Earlier than device_initcall() so that other drivers invoking
2451    request_module() don't end up in a loop when modprobe tries
2452    to use a UNIX socket. But later than subsys_initcall() because
2453    we depend on stuff initialised there */
2454 fs_initcall(af_unix_init);
2455 module_exit(af_unix_exit);
2456 
2457 MODULE_LICENSE("GPL");
2458 MODULE_ALIAS_NETPROTO(PF_UNIX);
2459