xref: /openbmc/linux/net/unix/af_unix.c (revision ca481398)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
84 
85 #include <linux/module.h>
86 #include <linux/kernel.h>
87 #include <linux/signal.h>
88 #include <linux/sched/signal.h>
89 #include <linux/errno.h>
90 #include <linux/string.h>
91 #include <linux/stat.h>
92 #include <linux/dcache.h>
93 #include <linux/namei.h>
94 #include <linux/socket.h>
95 #include <linux/un.h>
96 #include <linux/fcntl.h>
97 #include <linux/termios.h>
98 #include <linux/sockios.h>
99 #include <linux/net.h>
100 #include <linux/in.h>
101 #include <linux/fs.h>
102 #include <linux/slab.h>
103 #include <linux/uaccess.h>
104 #include <linux/skbuff.h>
105 #include <linux/netdevice.h>
106 #include <net/net_namespace.h>
107 #include <net/sock.h>
108 #include <net/tcp_states.h>
109 #include <net/af_unix.h>
110 #include <linux/proc_fs.h>
111 #include <linux/seq_file.h>
112 #include <net/scm.h>
113 #include <linux/init.h>
114 #include <linux/poll.h>
115 #include <linux/rtnetlink.h>
116 #include <linux/mount.h>
117 #include <net/checksum.h>
118 #include <linux/security.h>
119 #include <linux/freezer.h>
120 #include <linux/file.h>
121 
122 struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
123 EXPORT_SYMBOL_GPL(unix_socket_table);
124 DEFINE_SPINLOCK(unix_table_lock);
125 EXPORT_SYMBOL_GPL(unix_table_lock);
126 static atomic_long_t unix_nr_socks;
127 
128 
129 static struct hlist_head *unix_sockets_unbound(void *addr)
130 {
131 	unsigned long hash = (unsigned long)addr;
132 
133 	hash ^= hash >> 16;
134 	hash ^= hash >> 8;
135 	hash %= UNIX_HASH_SIZE;
136 	return &unix_socket_table[UNIX_HASH_SIZE + hash];
137 }
138 
139 #define UNIX_ABSTRACT(sk)	(unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
140 
141 #ifdef CONFIG_SECURITY_NETWORK
142 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
143 {
144 	UNIXCB(skb).secid = scm->secid;
145 }
146 
147 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
148 {
149 	scm->secid = UNIXCB(skb).secid;
150 }
151 
152 static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
153 {
154 	return (scm->secid == UNIXCB(skb).secid);
155 }
156 #else
157 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
158 { }
159 
160 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
161 { }
162 
163 static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
164 {
165 	return true;
166 }
167 #endif /* CONFIG_SECURITY_NETWORK */
168 
169 /*
170  *  SMP locking strategy:
171  *    hash table is protected with spinlock unix_table_lock
172  *    each socket state is protected by separate spin lock.
173  */
174 
175 static inline unsigned int unix_hash_fold(__wsum n)
176 {
177 	unsigned int hash = (__force unsigned int)csum_fold(n);
178 
179 	hash ^= hash>>8;
180 	return hash&(UNIX_HASH_SIZE-1);
181 }
182 
183 #define unix_peer(sk) (unix_sk(sk)->peer)
184 
185 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
186 {
187 	return unix_peer(osk) == sk;
188 }
189 
190 static inline int unix_may_send(struct sock *sk, struct sock *osk)
191 {
192 	return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
193 }
194 
195 static inline int unix_recvq_full(struct sock const *sk)
196 {
197 	return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
198 }
199 
200 struct sock *unix_peer_get(struct sock *s)
201 {
202 	struct sock *peer;
203 
204 	unix_state_lock(s);
205 	peer = unix_peer(s);
206 	if (peer)
207 		sock_hold(peer);
208 	unix_state_unlock(s);
209 	return peer;
210 }
211 EXPORT_SYMBOL_GPL(unix_peer_get);
212 
213 static inline void unix_release_addr(struct unix_address *addr)
214 {
215 	if (refcount_dec_and_test(&addr->refcnt))
216 		kfree(addr);
217 }
218 
219 /*
220  *	Check unix socket name:
221  *		- should be not zero length.
222  *	        - if started by not zero, should be NULL terminated (FS object)
223  *		- if started by zero, it is abstract name.
224  */
225 
226 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
227 {
228 	if (len <= sizeof(short) || len > sizeof(*sunaddr))
229 		return -EINVAL;
230 	if (!sunaddr || sunaddr->sun_family != AF_UNIX)
231 		return -EINVAL;
232 	if (sunaddr->sun_path[0]) {
233 		/*
234 		 * This may look like an off by one error but it is a bit more
235 		 * subtle. 108 is the longest valid AF_UNIX path for a binding.
236 		 * sun_path[108] doesn't as such exist.  However in kernel space
237 		 * we are guaranteed that it is a valid memory location in our
238 		 * kernel address buffer.
239 		 */
240 		((char *)sunaddr)[len] = 0;
241 		len = strlen(sunaddr->sun_path)+1+sizeof(short);
242 		return len;
243 	}
244 
245 	*hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
246 	return len;
247 }
248 
249 static void __unix_remove_socket(struct sock *sk)
250 {
251 	sk_del_node_init(sk);
252 }
253 
254 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
255 {
256 	WARN_ON(!sk_unhashed(sk));
257 	sk_add_node(sk, list);
258 }
259 
260 static inline void unix_remove_socket(struct sock *sk)
261 {
262 	spin_lock(&unix_table_lock);
263 	__unix_remove_socket(sk);
264 	spin_unlock(&unix_table_lock);
265 }
266 
267 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
268 {
269 	spin_lock(&unix_table_lock);
270 	__unix_insert_socket(list, sk);
271 	spin_unlock(&unix_table_lock);
272 }
273 
274 static struct sock *__unix_find_socket_byname(struct net *net,
275 					      struct sockaddr_un *sunname,
276 					      int len, int type, unsigned int hash)
277 {
278 	struct sock *s;
279 
280 	sk_for_each(s, &unix_socket_table[hash ^ type]) {
281 		struct unix_sock *u = unix_sk(s);
282 
283 		if (!net_eq(sock_net(s), net))
284 			continue;
285 
286 		if (u->addr->len == len &&
287 		    !memcmp(u->addr->name, sunname, len))
288 			goto found;
289 	}
290 	s = NULL;
291 found:
292 	return s;
293 }
294 
295 static inline struct sock *unix_find_socket_byname(struct net *net,
296 						   struct sockaddr_un *sunname,
297 						   int len, int type,
298 						   unsigned int hash)
299 {
300 	struct sock *s;
301 
302 	spin_lock(&unix_table_lock);
303 	s = __unix_find_socket_byname(net, sunname, len, type, hash);
304 	if (s)
305 		sock_hold(s);
306 	spin_unlock(&unix_table_lock);
307 	return s;
308 }
309 
310 static struct sock *unix_find_socket_byinode(struct inode *i)
311 {
312 	struct sock *s;
313 
314 	spin_lock(&unix_table_lock);
315 	sk_for_each(s,
316 		    &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
317 		struct dentry *dentry = unix_sk(s)->path.dentry;
318 
319 		if (dentry && d_backing_inode(dentry) == i) {
320 			sock_hold(s);
321 			goto found;
322 		}
323 	}
324 	s = NULL;
325 found:
326 	spin_unlock(&unix_table_lock);
327 	return s;
328 }
329 
330 /* Support code for asymmetrically connected dgram sockets
331  *
332  * If a datagram socket is connected to a socket not itself connected
333  * to the first socket (eg, /dev/log), clients may only enqueue more
334  * messages if the present receive queue of the server socket is not
335  * "too large". This means there's a second writeability condition
336  * poll and sendmsg need to test. The dgram recv code will do a wake
337  * up on the peer_wait wait queue of a socket upon reception of a
338  * datagram which needs to be propagated to sleeping would-be writers
339  * since these might not have sent anything so far. This can't be
340  * accomplished via poll_wait because the lifetime of the server
341  * socket might be less than that of its clients if these break their
342  * association with it or if the server socket is closed while clients
343  * are still connected to it and there's no way to inform "a polling
344  * implementation" that it should let go of a certain wait queue
345  *
346  * In order to propagate a wake up, a wait_queue_entry_t of the client
347  * socket is enqueued on the peer_wait queue of the server socket
348  * whose wake function does a wake_up on the ordinary client socket
349  * wait queue. This connection is established whenever a write (or
350  * poll for write) hit the flow control condition and broken when the
351  * association to the server socket is dissolved or after a wake up
352  * was relayed.
353  */
354 
355 static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
356 				      void *key)
357 {
358 	struct unix_sock *u;
359 	wait_queue_head_t *u_sleep;
360 
361 	u = container_of(q, struct unix_sock, peer_wake);
362 
363 	__remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
364 			    q);
365 	u->peer_wake.private = NULL;
366 
367 	/* relaying can only happen while the wq still exists */
368 	u_sleep = sk_sleep(&u->sk);
369 	if (u_sleep)
370 		wake_up_interruptible_poll(u_sleep, key);
371 
372 	return 0;
373 }
374 
375 static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
376 {
377 	struct unix_sock *u, *u_other;
378 	int rc;
379 
380 	u = unix_sk(sk);
381 	u_other = unix_sk(other);
382 	rc = 0;
383 	spin_lock(&u_other->peer_wait.lock);
384 
385 	if (!u->peer_wake.private) {
386 		u->peer_wake.private = other;
387 		__add_wait_queue(&u_other->peer_wait, &u->peer_wake);
388 
389 		rc = 1;
390 	}
391 
392 	spin_unlock(&u_other->peer_wait.lock);
393 	return rc;
394 }
395 
396 static void unix_dgram_peer_wake_disconnect(struct sock *sk,
397 					    struct sock *other)
398 {
399 	struct unix_sock *u, *u_other;
400 
401 	u = unix_sk(sk);
402 	u_other = unix_sk(other);
403 	spin_lock(&u_other->peer_wait.lock);
404 
405 	if (u->peer_wake.private == other) {
406 		__remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
407 		u->peer_wake.private = NULL;
408 	}
409 
410 	spin_unlock(&u_other->peer_wait.lock);
411 }
412 
413 static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
414 						   struct sock *other)
415 {
416 	unix_dgram_peer_wake_disconnect(sk, other);
417 	wake_up_interruptible_poll(sk_sleep(sk),
418 				   POLLOUT |
419 				   POLLWRNORM |
420 				   POLLWRBAND);
421 }
422 
423 /* preconditions:
424  *	- unix_peer(sk) == other
425  *	- association is stable
426  */
427 static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
428 {
429 	int connected;
430 
431 	connected = unix_dgram_peer_wake_connect(sk, other);
432 
433 	if (unix_recvq_full(other))
434 		return 1;
435 
436 	if (connected)
437 		unix_dgram_peer_wake_disconnect(sk, other);
438 
439 	return 0;
440 }
441 
442 static int unix_writable(const struct sock *sk)
443 {
444 	return sk->sk_state != TCP_LISTEN &&
445 	       (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
446 }
447 
448 static void unix_write_space(struct sock *sk)
449 {
450 	struct socket_wq *wq;
451 
452 	rcu_read_lock();
453 	if (unix_writable(sk)) {
454 		wq = rcu_dereference(sk->sk_wq);
455 		if (skwq_has_sleeper(wq))
456 			wake_up_interruptible_sync_poll(&wq->wait,
457 				POLLOUT | POLLWRNORM | POLLWRBAND);
458 		sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
459 	}
460 	rcu_read_unlock();
461 }
462 
463 /* When dgram socket disconnects (or changes its peer), we clear its receive
464  * queue of packets arrived from previous peer. First, it allows to do
465  * flow control based only on wmem_alloc; second, sk connected to peer
466  * may receive messages only from that peer. */
467 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
468 {
469 	if (!skb_queue_empty(&sk->sk_receive_queue)) {
470 		skb_queue_purge(&sk->sk_receive_queue);
471 		wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
472 
473 		/* If one link of bidirectional dgram pipe is disconnected,
474 		 * we signal error. Messages are lost. Do not make this,
475 		 * when peer was not connected to us.
476 		 */
477 		if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
478 			other->sk_err = ECONNRESET;
479 			other->sk_error_report(other);
480 		}
481 	}
482 }
483 
484 static void unix_sock_destructor(struct sock *sk)
485 {
486 	struct unix_sock *u = unix_sk(sk);
487 
488 	skb_queue_purge(&sk->sk_receive_queue);
489 
490 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
491 	WARN_ON(!sk_unhashed(sk));
492 	WARN_ON(sk->sk_socket);
493 	if (!sock_flag(sk, SOCK_DEAD)) {
494 		pr_info("Attempt to release alive unix socket: %p\n", sk);
495 		return;
496 	}
497 
498 	if (u->addr)
499 		unix_release_addr(u->addr);
500 
501 	atomic_long_dec(&unix_nr_socks);
502 	local_bh_disable();
503 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
504 	local_bh_enable();
505 #ifdef UNIX_REFCNT_DEBUG
506 	pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
507 		atomic_long_read(&unix_nr_socks));
508 #endif
509 }
510 
511 static void unix_release_sock(struct sock *sk, int embrion)
512 {
513 	struct unix_sock *u = unix_sk(sk);
514 	struct path path;
515 	struct sock *skpair;
516 	struct sk_buff *skb;
517 	int state;
518 
519 	unix_remove_socket(sk);
520 
521 	/* Clear state */
522 	unix_state_lock(sk);
523 	sock_orphan(sk);
524 	sk->sk_shutdown = SHUTDOWN_MASK;
525 	path	     = u->path;
526 	u->path.dentry = NULL;
527 	u->path.mnt = NULL;
528 	state = sk->sk_state;
529 	sk->sk_state = TCP_CLOSE;
530 	unix_state_unlock(sk);
531 
532 	wake_up_interruptible_all(&u->peer_wait);
533 
534 	skpair = unix_peer(sk);
535 
536 	if (skpair != NULL) {
537 		if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
538 			unix_state_lock(skpair);
539 			/* No more writes */
540 			skpair->sk_shutdown = SHUTDOWN_MASK;
541 			if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
542 				skpair->sk_err = ECONNRESET;
543 			unix_state_unlock(skpair);
544 			skpair->sk_state_change(skpair);
545 			sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
546 		}
547 
548 		unix_dgram_peer_wake_disconnect(sk, skpair);
549 		sock_put(skpair); /* It may now die */
550 		unix_peer(sk) = NULL;
551 	}
552 
553 	/* Try to flush out this socket. Throw out buffers at least */
554 
555 	while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
556 		if (state == TCP_LISTEN)
557 			unix_release_sock(skb->sk, 1);
558 		/* passed fds are erased in the kfree_skb hook	      */
559 		UNIXCB(skb).consumed = skb->len;
560 		kfree_skb(skb);
561 	}
562 
563 	if (path.dentry)
564 		path_put(&path);
565 
566 	sock_put(sk);
567 
568 	/* ---- Socket is dead now and most probably destroyed ---- */
569 
570 	/*
571 	 * Fixme: BSD difference: In BSD all sockets connected to us get
572 	 *	  ECONNRESET and we die on the spot. In Linux we behave
573 	 *	  like files and pipes do and wait for the last
574 	 *	  dereference.
575 	 *
576 	 * Can't we simply set sock->err?
577 	 *
578 	 *	  What the above comment does talk about? --ANK(980817)
579 	 */
580 
581 	if (unix_tot_inflight)
582 		unix_gc();		/* Garbage collect fds */
583 }
584 
585 static void init_peercred(struct sock *sk)
586 {
587 	put_pid(sk->sk_peer_pid);
588 	if (sk->sk_peer_cred)
589 		put_cred(sk->sk_peer_cred);
590 	sk->sk_peer_pid  = get_pid(task_tgid(current));
591 	sk->sk_peer_cred = get_current_cred();
592 }
593 
594 static void copy_peercred(struct sock *sk, struct sock *peersk)
595 {
596 	put_pid(sk->sk_peer_pid);
597 	if (sk->sk_peer_cred)
598 		put_cred(sk->sk_peer_cred);
599 	sk->sk_peer_pid  = get_pid(peersk->sk_peer_pid);
600 	sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
601 }
602 
603 static int unix_listen(struct socket *sock, int backlog)
604 {
605 	int err;
606 	struct sock *sk = sock->sk;
607 	struct unix_sock *u = unix_sk(sk);
608 	struct pid *old_pid = NULL;
609 
610 	err = -EOPNOTSUPP;
611 	if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
612 		goto out;	/* Only stream/seqpacket sockets accept */
613 	err = -EINVAL;
614 	if (!u->addr)
615 		goto out;	/* No listens on an unbound socket */
616 	unix_state_lock(sk);
617 	if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
618 		goto out_unlock;
619 	if (backlog > sk->sk_max_ack_backlog)
620 		wake_up_interruptible_all(&u->peer_wait);
621 	sk->sk_max_ack_backlog	= backlog;
622 	sk->sk_state		= TCP_LISTEN;
623 	/* set credentials so connect can copy them */
624 	init_peercred(sk);
625 	err = 0;
626 
627 out_unlock:
628 	unix_state_unlock(sk);
629 	put_pid(old_pid);
630 out:
631 	return err;
632 }
633 
634 static int unix_release(struct socket *);
635 static int unix_bind(struct socket *, struct sockaddr *, int);
636 static int unix_stream_connect(struct socket *, struct sockaddr *,
637 			       int addr_len, int flags);
638 static int unix_socketpair(struct socket *, struct socket *);
639 static int unix_accept(struct socket *, struct socket *, int, bool);
640 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
641 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
642 static unsigned int unix_dgram_poll(struct file *, struct socket *,
643 				    poll_table *);
644 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
645 static int unix_shutdown(struct socket *, int);
646 static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
647 static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
648 static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
649 				    size_t size, int flags);
650 static ssize_t unix_stream_splice_read(struct socket *,  loff_t *ppos,
651 				       struct pipe_inode_info *, size_t size,
652 				       unsigned int flags);
653 static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
654 static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
655 static int unix_dgram_connect(struct socket *, struct sockaddr *,
656 			      int, int);
657 static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
658 static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
659 				  int);
660 
661 static int unix_set_peek_off(struct sock *sk, int val)
662 {
663 	struct unix_sock *u = unix_sk(sk);
664 
665 	if (mutex_lock_interruptible(&u->iolock))
666 		return -EINTR;
667 
668 	sk->sk_peek_off = val;
669 	mutex_unlock(&u->iolock);
670 
671 	return 0;
672 }
673 
674 
675 static const struct proto_ops unix_stream_ops = {
676 	.family =	PF_UNIX,
677 	.owner =	THIS_MODULE,
678 	.release =	unix_release,
679 	.bind =		unix_bind,
680 	.connect =	unix_stream_connect,
681 	.socketpair =	unix_socketpair,
682 	.accept =	unix_accept,
683 	.getname =	unix_getname,
684 	.poll =		unix_poll,
685 	.ioctl =	unix_ioctl,
686 	.listen =	unix_listen,
687 	.shutdown =	unix_shutdown,
688 	.setsockopt =	sock_no_setsockopt,
689 	.getsockopt =	sock_no_getsockopt,
690 	.sendmsg =	unix_stream_sendmsg,
691 	.recvmsg =	unix_stream_recvmsg,
692 	.mmap =		sock_no_mmap,
693 	.sendpage =	unix_stream_sendpage,
694 	.splice_read =	unix_stream_splice_read,
695 	.set_peek_off =	unix_set_peek_off,
696 };
697 
698 static const struct proto_ops unix_dgram_ops = {
699 	.family =	PF_UNIX,
700 	.owner =	THIS_MODULE,
701 	.release =	unix_release,
702 	.bind =		unix_bind,
703 	.connect =	unix_dgram_connect,
704 	.socketpair =	unix_socketpair,
705 	.accept =	sock_no_accept,
706 	.getname =	unix_getname,
707 	.poll =		unix_dgram_poll,
708 	.ioctl =	unix_ioctl,
709 	.listen =	sock_no_listen,
710 	.shutdown =	unix_shutdown,
711 	.setsockopt =	sock_no_setsockopt,
712 	.getsockopt =	sock_no_getsockopt,
713 	.sendmsg =	unix_dgram_sendmsg,
714 	.recvmsg =	unix_dgram_recvmsg,
715 	.mmap =		sock_no_mmap,
716 	.sendpage =	sock_no_sendpage,
717 	.set_peek_off =	unix_set_peek_off,
718 };
719 
720 static const struct proto_ops unix_seqpacket_ops = {
721 	.family =	PF_UNIX,
722 	.owner =	THIS_MODULE,
723 	.release =	unix_release,
724 	.bind =		unix_bind,
725 	.connect =	unix_stream_connect,
726 	.socketpair =	unix_socketpair,
727 	.accept =	unix_accept,
728 	.getname =	unix_getname,
729 	.poll =		unix_dgram_poll,
730 	.ioctl =	unix_ioctl,
731 	.listen =	unix_listen,
732 	.shutdown =	unix_shutdown,
733 	.setsockopt =	sock_no_setsockopt,
734 	.getsockopt =	sock_no_getsockopt,
735 	.sendmsg =	unix_seqpacket_sendmsg,
736 	.recvmsg =	unix_seqpacket_recvmsg,
737 	.mmap =		sock_no_mmap,
738 	.sendpage =	sock_no_sendpage,
739 	.set_peek_off =	unix_set_peek_off,
740 };
741 
742 static struct proto unix_proto = {
743 	.name			= "UNIX",
744 	.owner			= THIS_MODULE,
745 	.obj_size		= sizeof(struct unix_sock),
746 };
747 
748 /*
749  * AF_UNIX sockets do not interact with hardware, hence they
750  * dont trigger interrupts - so it's safe for them to have
751  * bh-unsafe locking for their sk_receive_queue.lock. Split off
752  * this special lock-class by reinitializing the spinlock key:
753  */
754 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
755 
756 static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
757 {
758 	struct sock *sk = NULL;
759 	struct unix_sock *u;
760 
761 	atomic_long_inc(&unix_nr_socks);
762 	if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
763 		goto out;
764 
765 	sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
766 	if (!sk)
767 		goto out;
768 
769 	sock_init_data(sock, sk);
770 	lockdep_set_class(&sk->sk_receive_queue.lock,
771 				&af_unix_sk_receive_queue_lock_key);
772 
773 	sk->sk_allocation	= GFP_KERNEL_ACCOUNT;
774 	sk->sk_write_space	= unix_write_space;
775 	sk->sk_max_ack_backlog	= net->unx.sysctl_max_dgram_qlen;
776 	sk->sk_destruct		= unix_sock_destructor;
777 	u	  = unix_sk(sk);
778 	u->path.dentry = NULL;
779 	u->path.mnt = NULL;
780 	spin_lock_init(&u->lock);
781 	atomic_long_set(&u->inflight, 0);
782 	INIT_LIST_HEAD(&u->link);
783 	mutex_init(&u->iolock); /* single task reading lock */
784 	mutex_init(&u->bindlock); /* single task binding lock */
785 	init_waitqueue_head(&u->peer_wait);
786 	init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
787 	unix_insert_socket(unix_sockets_unbound(sk), sk);
788 out:
789 	if (sk == NULL)
790 		atomic_long_dec(&unix_nr_socks);
791 	else {
792 		local_bh_disable();
793 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
794 		local_bh_enable();
795 	}
796 	return sk;
797 }
798 
799 static int unix_create(struct net *net, struct socket *sock, int protocol,
800 		       int kern)
801 {
802 	if (protocol && protocol != PF_UNIX)
803 		return -EPROTONOSUPPORT;
804 
805 	sock->state = SS_UNCONNECTED;
806 
807 	switch (sock->type) {
808 	case SOCK_STREAM:
809 		sock->ops = &unix_stream_ops;
810 		break;
811 		/*
812 		 *	Believe it or not BSD has AF_UNIX, SOCK_RAW though
813 		 *	nothing uses it.
814 		 */
815 	case SOCK_RAW:
816 		sock->type = SOCK_DGRAM;
817 	case SOCK_DGRAM:
818 		sock->ops = &unix_dgram_ops;
819 		break;
820 	case SOCK_SEQPACKET:
821 		sock->ops = &unix_seqpacket_ops;
822 		break;
823 	default:
824 		return -ESOCKTNOSUPPORT;
825 	}
826 
827 	return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
828 }
829 
830 static int unix_release(struct socket *sock)
831 {
832 	struct sock *sk = sock->sk;
833 
834 	if (!sk)
835 		return 0;
836 
837 	unix_release_sock(sk, 0);
838 	sock->sk = NULL;
839 
840 	return 0;
841 }
842 
843 static int unix_autobind(struct socket *sock)
844 {
845 	struct sock *sk = sock->sk;
846 	struct net *net = sock_net(sk);
847 	struct unix_sock *u = unix_sk(sk);
848 	static u32 ordernum = 1;
849 	struct unix_address *addr;
850 	int err;
851 	unsigned int retries = 0;
852 
853 	err = mutex_lock_interruptible(&u->bindlock);
854 	if (err)
855 		return err;
856 
857 	err = 0;
858 	if (u->addr)
859 		goto out;
860 
861 	err = -ENOMEM;
862 	addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
863 	if (!addr)
864 		goto out;
865 
866 	addr->name->sun_family = AF_UNIX;
867 	refcount_set(&addr->refcnt, 1);
868 
869 retry:
870 	addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
871 	addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
872 
873 	spin_lock(&unix_table_lock);
874 	ordernum = (ordernum+1)&0xFFFFF;
875 
876 	if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
877 				      addr->hash)) {
878 		spin_unlock(&unix_table_lock);
879 		/*
880 		 * __unix_find_socket_byname() may take long time if many names
881 		 * are already in use.
882 		 */
883 		cond_resched();
884 		/* Give up if all names seems to be in use. */
885 		if (retries++ == 0xFFFFF) {
886 			err = -ENOSPC;
887 			kfree(addr);
888 			goto out;
889 		}
890 		goto retry;
891 	}
892 	addr->hash ^= sk->sk_type;
893 
894 	__unix_remove_socket(sk);
895 	u->addr = addr;
896 	__unix_insert_socket(&unix_socket_table[addr->hash], sk);
897 	spin_unlock(&unix_table_lock);
898 	err = 0;
899 
900 out:	mutex_unlock(&u->bindlock);
901 	return err;
902 }
903 
904 static struct sock *unix_find_other(struct net *net,
905 				    struct sockaddr_un *sunname, int len,
906 				    int type, unsigned int hash, int *error)
907 {
908 	struct sock *u;
909 	struct path path;
910 	int err = 0;
911 
912 	if (sunname->sun_path[0]) {
913 		struct inode *inode;
914 		err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
915 		if (err)
916 			goto fail;
917 		inode = d_backing_inode(path.dentry);
918 		err = inode_permission(inode, MAY_WRITE);
919 		if (err)
920 			goto put_fail;
921 
922 		err = -ECONNREFUSED;
923 		if (!S_ISSOCK(inode->i_mode))
924 			goto put_fail;
925 		u = unix_find_socket_byinode(inode);
926 		if (!u)
927 			goto put_fail;
928 
929 		if (u->sk_type == type)
930 			touch_atime(&path);
931 
932 		path_put(&path);
933 
934 		err = -EPROTOTYPE;
935 		if (u->sk_type != type) {
936 			sock_put(u);
937 			goto fail;
938 		}
939 	} else {
940 		err = -ECONNREFUSED;
941 		u = unix_find_socket_byname(net, sunname, len, type, hash);
942 		if (u) {
943 			struct dentry *dentry;
944 			dentry = unix_sk(u)->path.dentry;
945 			if (dentry)
946 				touch_atime(&unix_sk(u)->path);
947 		} else
948 			goto fail;
949 	}
950 	return u;
951 
952 put_fail:
953 	path_put(&path);
954 fail:
955 	*error = err;
956 	return NULL;
957 }
958 
959 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
960 {
961 	struct dentry *dentry;
962 	struct path path;
963 	int err = 0;
964 	/*
965 	 * Get the parent directory, calculate the hash for last
966 	 * component.
967 	 */
968 	dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
969 	err = PTR_ERR(dentry);
970 	if (IS_ERR(dentry))
971 		return err;
972 
973 	/*
974 	 * All right, let's create it.
975 	 */
976 	err = security_path_mknod(&path, dentry, mode, 0);
977 	if (!err) {
978 		err = vfs_mknod(d_inode(path.dentry), dentry, mode, 0);
979 		if (!err) {
980 			res->mnt = mntget(path.mnt);
981 			res->dentry = dget(dentry);
982 		}
983 	}
984 	done_path_create(&path, dentry);
985 	return err;
986 }
987 
988 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
989 {
990 	struct sock *sk = sock->sk;
991 	struct net *net = sock_net(sk);
992 	struct unix_sock *u = unix_sk(sk);
993 	struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
994 	char *sun_path = sunaddr->sun_path;
995 	int err;
996 	unsigned int hash;
997 	struct unix_address *addr;
998 	struct hlist_head *list;
999 	struct path path = { };
1000 
1001 	err = -EINVAL;
1002 	if (addr_len < offsetofend(struct sockaddr_un, sun_family) ||
1003 	    sunaddr->sun_family != AF_UNIX)
1004 		goto out;
1005 
1006 	if (addr_len == sizeof(short)) {
1007 		err = unix_autobind(sock);
1008 		goto out;
1009 	}
1010 
1011 	err = unix_mkname(sunaddr, addr_len, &hash);
1012 	if (err < 0)
1013 		goto out;
1014 	addr_len = err;
1015 
1016 	if (sun_path[0]) {
1017 		umode_t mode = S_IFSOCK |
1018 		       (SOCK_INODE(sock)->i_mode & ~current_umask());
1019 		err = unix_mknod(sun_path, mode, &path);
1020 		if (err) {
1021 			if (err == -EEXIST)
1022 				err = -EADDRINUSE;
1023 			goto out;
1024 		}
1025 	}
1026 
1027 	err = mutex_lock_interruptible(&u->bindlock);
1028 	if (err)
1029 		goto out_put;
1030 
1031 	err = -EINVAL;
1032 	if (u->addr)
1033 		goto out_up;
1034 
1035 	err = -ENOMEM;
1036 	addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1037 	if (!addr)
1038 		goto out_up;
1039 
1040 	memcpy(addr->name, sunaddr, addr_len);
1041 	addr->len = addr_len;
1042 	addr->hash = hash ^ sk->sk_type;
1043 	refcount_set(&addr->refcnt, 1);
1044 
1045 	if (sun_path[0]) {
1046 		addr->hash = UNIX_HASH_SIZE;
1047 		hash = d_backing_inode(path.dentry)->i_ino & (UNIX_HASH_SIZE - 1);
1048 		spin_lock(&unix_table_lock);
1049 		u->path = path;
1050 		list = &unix_socket_table[hash];
1051 	} else {
1052 		spin_lock(&unix_table_lock);
1053 		err = -EADDRINUSE;
1054 		if (__unix_find_socket_byname(net, sunaddr, addr_len,
1055 					      sk->sk_type, hash)) {
1056 			unix_release_addr(addr);
1057 			goto out_unlock;
1058 		}
1059 
1060 		list = &unix_socket_table[addr->hash];
1061 	}
1062 
1063 	err = 0;
1064 	__unix_remove_socket(sk);
1065 	u->addr = addr;
1066 	__unix_insert_socket(list, sk);
1067 
1068 out_unlock:
1069 	spin_unlock(&unix_table_lock);
1070 out_up:
1071 	mutex_unlock(&u->bindlock);
1072 out_put:
1073 	if (err)
1074 		path_put(&path);
1075 out:
1076 	return err;
1077 }
1078 
1079 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1080 {
1081 	if (unlikely(sk1 == sk2) || !sk2) {
1082 		unix_state_lock(sk1);
1083 		return;
1084 	}
1085 	if (sk1 < sk2) {
1086 		unix_state_lock(sk1);
1087 		unix_state_lock_nested(sk2);
1088 	} else {
1089 		unix_state_lock(sk2);
1090 		unix_state_lock_nested(sk1);
1091 	}
1092 }
1093 
1094 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1095 {
1096 	if (unlikely(sk1 == sk2) || !sk2) {
1097 		unix_state_unlock(sk1);
1098 		return;
1099 	}
1100 	unix_state_unlock(sk1);
1101 	unix_state_unlock(sk2);
1102 }
1103 
1104 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1105 			      int alen, int flags)
1106 {
1107 	struct sock *sk = sock->sk;
1108 	struct net *net = sock_net(sk);
1109 	struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1110 	struct sock *other;
1111 	unsigned int hash;
1112 	int err;
1113 
1114 	err = -EINVAL;
1115 	if (alen < offsetofend(struct sockaddr, sa_family))
1116 		goto out;
1117 
1118 	if (addr->sa_family != AF_UNSPEC) {
1119 		err = unix_mkname(sunaddr, alen, &hash);
1120 		if (err < 0)
1121 			goto out;
1122 		alen = err;
1123 
1124 		if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1125 		    !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1126 			goto out;
1127 
1128 restart:
1129 		other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1130 		if (!other)
1131 			goto out;
1132 
1133 		unix_state_double_lock(sk, other);
1134 
1135 		/* Apparently VFS overslept socket death. Retry. */
1136 		if (sock_flag(other, SOCK_DEAD)) {
1137 			unix_state_double_unlock(sk, other);
1138 			sock_put(other);
1139 			goto restart;
1140 		}
1141 
1142 		err = -EPERM;
1143 		if (!unix_may_send(sk, other))
1144 			goto out_unlock;
1145 
1146 		err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1147 		if (err)
1148 			goto out_unlock;
1149 
1150 	} else {
1151 		/*
1152 		 *	1003.1g breaking connected state with AF_UNSPEC
1153 		 */
1154 		other = NULL;
1155 		unix_state_double_lock(sk, other);
1156 	}
1157 
1158 	/*
1159 	 * If it was connected, reconnect.
1160 	 */
1161 	if (unix_peer(sk)) {
1162 		struct sock *old_peer = unix_peer(sk);
1163 		unix_peer(sk) = other;
1164 		unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1165 
1166 		unix_state_double_unlock(sk, other);
1167 
1168 		if (other != old_peer)
1169 			unix_dgram_disconnected(sk, old_peer);
1170 		sock_put(old_peer);
1171 	} else {
1172 		unix_peer(sk) = other;
1173 		unix_state_double_unlock(sk, other);
1174 	}
1175 	return 0;
1176 
1177 out_unlock:
1178 	unix_state_double_unlock(sk, other);
1179 	sock_put(other);
1180 out:
1181 	return err;
1182 }
1183 
1184 static long unix_wait_for_peer(struct sock *other, long timeo)
1185 {
1186 	struct unix_sock *u = unix_sk(other);
1187 	int sched;
1188 	DEFINE_WAIT(wait);
1189 
1190 	prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1191 
1192 	sched = !sock_flag(other, SOCK_DEAD) &&
1193 		!(other->sk_shutdown & RCV_SHUTDOWN) &&
1194 		unix_recvq_full(other);
1195 
1196 	unix_state_unlock(other);
1197 
1198 	if (sched)
1199 		timeo = schedule_timeout(timeo);
1200 
1201 	finish_wait(&u->peer_wait, &wait);
1202 	return timeo;
1203 }
1204 
1205 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1206 			       int addr_len, int flags)
1207 {
1208 	struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1209 	struct sock *sk = sock->sk;
1210 	struct net *net = sock_net(sk);
1211 	struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1212 	struct sock *newsk = NULL;
1213 	struct sock *other = NULL;
1214 	struct sk_buff *skb = NULL;
1215 	unsigned int hash;
1216 	int st;
1217 	int err;
1218 	long timeo;
1219 
1220 	err = unix_mkname(sunaddr, addr_len, &hash);
1221 	if (err < 0)
1222 		goto out;
1223 	addr_len = err;
1224 
1225 	if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1226 	    (err = unix_autobind(sock)) != 0)
1227 		goto out;
1228 
1229 	timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1230 
1231 	/* First of all allocate resources.
1232 	   If we will make it after state is locked,
1233 	   we will have to recheck all again in any case.
1234 	 */
1235 
1236 	err = -ENOMEM;
1237 
1238 	/* create new sock for complete connection */
1239 	newsk = unix_create1(sock_net(sk), NULL, 0);
1240 	if (newsk == NULL)
1241 		goto out;
1242 
1243 	/* Allocate skb for sending to listening sock */
1244 	skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1245 	if (skb == NULL)
1246 		goto out;
1247 
1248 restart:
1249 	/*  Find listening sock. */
1250 	other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1251 	if (!other)
1252 		goto out;
1253 
1254 	/* Latch state of peer */
1255 	unix_state_lock(other);
1256 
1257 	/* Apparently VFS overslept socket death. Retry. */
1258 	if (sock_flag(other, SOCK_DEAD)) {
1259 		unix_state_unlock(other);
1260 		sock_put(other);
1261 		goto restart;
1262 	}
1263 
1264 	err = -ECONNREFUSED;
1265 	if (other->sk_state != TCP_LISTEN)
1266 		goto out_unlock;
1267 	if (other->sk_shutdown & RCV_SHUTDOWN)
1268 		goto out_unlock;
1269 
1270 	if (unix_recvq_full(other)) {
1271 		err = -EAGAIN;
1272 		if (!timeo)
1273 			goto out_unlock;
1274 
1275 		timeo = unix_wait_for_peer(other, timeo);
1276 
1277 		err = sock_intr_errno(timeo);
1278 		if (signal_pending(current))
1279 			goto out;
1280 		sock_put(other);
1281 		goto restart;
1282 	}
1283 
1284 	/* Latch our state.
1285 
1286 	   It is tricky place. We need to grab our state lock and cannot
1287 	   drop lock on peer. It is dangerous because deadlock is
1288 	   possible. Connect to self case and simultaneous
1289 	   attempt to connect are eliminated by checking socket
1290 	   state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1291 	   check this before attempt to grab lock.
1292 
1293 	   Well, and we have to recheck the state after socket locked.
1294 	 */
1295 	st = sk->sk_state;
1296 
1297 	switch (st) {
1298 	case TCP_CLOSE:
1299 		/* This is ok... continue with connect */
1300 		break;
1301 	case TCP_ESTABLISHED:
1302 		/* Socket is already connected */
1303 		err = -EISCONN;
1304 		goto out_unlock;
1305 	default:
1306 		err = -EINVAL;
1307 		goto out_unlock;
1308 	}
1309 
1310 	unix_state_lock_nested(sk);
1311 
1312 	if (sk->sk_state != st) {
1313 		unix_state_unlock(sk);
1314 		unix_state_unlock(other);
1315 		sock_put(other);
1316 		goto restart;
1317 	}
1318 
1319 	err = security_unix_stream_connect(sk, other, newsk);
1320 	if (err) {
1321 		unix_state_unlock(sk);
1322 		goto out_unlock;
1323 	}
1324 
1325 	/* The way is open! Fastly set all the necessary fields... */
1326 
1327 	sock_hold(sk);
1328 	unix_peer(newsk)	= sk;
1329 	newsk->sk_state		= TCP_ESTABLISHED;
1330 	newsk->sk_type		= sk->sk_type;
1331 	init_peercred(newsk);
1332 	newu = unix_sk(newsk);
1333 	RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1334 	otheru = unix_sk(other);
1335 
1336 	/* copy address information from listening to new sock*/
1337 	if (otheru->addr) {
1338 		refcount_inc(&otheru->addr->refcnt);
1339 		newu->addr = otheru->addr;
1340 	}
1341 	if (otheru->path.dentry) {
1342 		path_get(&otheru->path);
1343 		newu->path = otheru->path;
1344 	}
1345 
1346 	/* Set credentials */
1347 	copy_peercred(sk, other);
1348 
1349 	sock->state	= SS_CONNECTED;
1350 	sk->sk_state	= TCP_ESTABLISHED;
1351 	sock_hold(newsk);
1352 
1353 	smp_mb__after_atomic();	/* sock_hold() does an atomic_inc() */
1354 	unix_peer(sk)	= newsk;
1355 
1356 	unix_state_unlock(sk);
1357 
1358 	/* take ten and and send info to listening sock */
1359 	spin_lock(&other->sk_receive_queue.lock);
1360 	__skb_queue_tail(&other->sk_receive_queue, skb);
1361 	spin_unlock(&other->sk_receive_queue.lock);
1362 	unix_state_unlock(other);
1363 	other->sk_data_ready(other);
1364 	sock_put(other);
1365 	return 0;
1366 
1367 out_unlock:
1368 	if (other)
1369 		unix_state_unlock(other);
1370 
1371 out:
1372 	kfree_skb(skb);
1373 	if (newsk)
1374 		unix_release_sock(newsk, 0);
1375 	if (other)
1376 		sock_put(other);
1377 	return err;
1378 }
1379 
1380 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1381 {
1382 	struct sock *ska = socka->sk, *skb = sockb->sk;
1383 
1384 	/* Join our sockets back to back */
1385 	sock_hold(ska);
1386 	sock_hold(skb);
1387 	unix_peer(ska) = skb;
1388 	unix_peer(skb) = ska;
1389 	init_peercred(ska);
1390 	init_peercred(skb);
1391 
1392 	if (ska->sk_type != SOCK_DGRAM) {
1393 		ska->sk_state = TCP_ESTABLISHED;
1394 		skb->sk_state = TCP_ESTABLISHED;
1395 		socka->state  = SS_CONNECTED;
1396 		sockb->state  = SS_CONNECTED;
1397 	}
1398 	return 0;
1399 }
1400 
1401 static void unix_sock_inherit_flags(const struct socket *old,
1402 				    struct socket *new)
1403 {
1404 	if (test_bit(SOCK_PASSCRED, &old->flags))
1405 		set_bit(SOCK_PASSCRED, &new->flags);
1406 	if (test_bit(SOCK_PASSSEC, &old->flags))
1407 		set_bit(SOCK_PASSSEC, &new->flags);
1408 }
1409 
1410 static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1411 		       bool kern)
1412 {
1413 	struct sock *sk = sock->sk;
1414 	struct sock *tsk;
1415 	struct sk_buff *skb;
1416 	int err;
1417 
1418 	err = -EOPNOTSUPP;
1419 	if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1420 		goto out;
1421 
1422 	err = -EINVAL;
1423 	if (sk->sk_state != TCP_LISTEN)
1424 		goto out;
1425 
1426 	/* If socket state is TCP_LISTEN it cannot change (for now...),
1427 	 * so that no locks are necessary.
1428 	 */
1429 
1430 	skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1431 	if (!skb) {
1432 		/* This means receive shutdown. */
1433 		if (err == 0)
1434 			err = -EINVAL;
1435 		goto out;
1436 	}
1437 
1438 	tsk = skb->sk;
1439 	skb_free_datagram(sk, skb);
1440 	wake_up_interruptible(&unix_sk(sk)->peer_wait);
1441 
1442 	/* attach accepted sock to socket */
1443 	unix_state_lock(tsk);
1444 	newsock->state = SS_CONNECTED;
1445 	unix_sock_inherit_flags(sock, newsock);
1446 	sock_graft(tsk, newsock);
1447 	unix_state_unlock(tsk);
1448 	return 0;
1449 
1450 out:
1451 	return err;
1452 }
1453 
1454 
1455 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1456 {
1457 	struct sock *sk = sock->sk;
1458 	struct unix_sock *u;
1459 	DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1460 	int err = 0;
1461 
1462 	if (peer) {
1463 		sk = unix_peer_get(sk);
1464 
1465 		err = -ENOTCONN;
1466 		if (!sk)
1467 			goto out;
1468 		err = 0;
1469 	} else {
1470 		sock_hold(sk);
1471 	}
1472 
1473 	u = unix_sk(sk);
1474 	unix_state_lock(sk);
1475 	if (!u->addr) {
1476 		sunaddr->sun_family = AF_UNIX;
1477 		sunaddr->sun_path[0] = 0;
1478 		*uaddr_len = sizeof(short);
1479 	} else {
1480 		struct unix_address *addr = u->addr;
1481 
1482 		*uaddr_len = addr->len;
1483 		memcpy(sunaddr, addr->name, *uaddr_len);
1484 	}
1485 	unix_state_unlock(sk);
1486 	sock_put(sk);
1487 out:
1488 	return err;
1489 }
1490 
1491 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1492 {
1493 	int i;
1494 
1495 	scm->fp = UNIXCB(skb).fp;
1496 	UNIXCB(skb).fp = NULL;
1497 
1498 	for (i = scm->fp->count-1; i >= 0; i--)
1499 		unix_notinflight(scm->fp->user, scm->fp->fp[i]);
1500 }
1501 
1502 static void unix_destruct_scm(struct sk_buff *skb)
1503 {
1504 	struct scm_cookie scm;
1505 	memset(&scm, 0, sizeof(scm));
1506 	scm.pid  = UNIXCB(skb).pid;
1507 	if (UNIXCB(skb).fp)
1508 		unix_detach_fds(&scm, skb);
1509 
1510 	/* Alas, it calls VFS */
1511 	/* So fscking what? fput() had been SMP-safe since the last Summer */
1512 	scm_destroy(&scm);
1513 	sock_wfree(skb);
1514 }
1515 
1516 /*
1517  * The "user->unix_inflight" variable is protected by the garbage
1518  * collection lock, and we just read it locklessly here. If you go
1519  * over the limit, there might be a tiny race in actually noticing
1520  * it across threads. Tough.
1521  */
1522 static inline bool too_many_unix_fds(struct task_struct *p)
1523 {
1524 	struct user_struct *user = current_user();
1525 
1526 	if (unlikely(user->unix_inflight > task_rlimit(p, RLIMIT_NOFILE)))
1527 		return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
1528 	return false;
1529 }
1530 
1531 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1532 {
1533 	int i;
1534 
1535 	if (too_many_unix_fds(current))
1536 		return -ETOOMANYREFS;
1537 
1538 	/*
1539 	 * Need to duplicate file references for the sake of garbage
1540 	 * collection.  Otherwise a socket in the fps might become a
1541 	 * candidate for GC while the skb is not yet queued.
1542 	 */
1543 	UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1544 	if (!UNIXCB(skb).fp)
1545 		return -ENOMEM;
1546 
1547 	for (i = scm->fp->count - 1; i >= 0; i--)
1548 		unix_inflight(scm->fp->user, scm->fp->fp[i]);
1549 	return 0;
1550 }
1551 
1552 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1553 {
1554 	int err = 0;
1555 
1556 	UNIXCB(skb).pid  = get_pid(scm->pid);
1557 	UNIXCB(skb).uid = scm->creds.uid;
1558 	UNIXCB(skb).gid = scm->creds.gid;
1559 	UNIXCB(skb).fp = NULL;
1560 	unix_get_secdata(scm, skb);
1561 	if (scm->fp && send_fds)
1562 		err = unix_attach_fds(scm, skb);
1563 
1564 	skb->destructor = unix_destruct_scm;
1565 	return err;
1566 }
1567 
1568 static bool unix_passcred_enabled(const struct socket *sock,
1569 				  const struct sock *other)
1570 {
1571 	return test_bit(SOCK_PASSCRED, &sock->flags) ||
1572 	       !other->sk_socket ||
1573 	       test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1574 }
1575 
1576 /*
1577  * Some apps rely on write() giving SCM_CREDENTIALS
1578  * We include credentials if source or destination socket
1579  * asserted SOCK_PASSCRED.
1580  */
1581 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1582 			    const struct sock *other)
1583 {
1584 	if (UNIXCB(skb).pid)
1585 		return;
1586 	if (unix_passcred_enabled(sock, other)) {
1587 		UNIXCB(skb).pid  = get_pid(task_tgid(current));
1588 		current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1589 	}
1590 }
1591 
1592 static int maybe_init_creds(struct scm_cookie *scm,
1593 			    struct socket *socket,
1594 			    const struct sock *other)
1595 {
1596 	int err;
1597 	struct msghdr msg = { .msg_controllen = 0 };
1598 
1599 	err = scm_send(socket, &msg, scm, false);
1600 	if (err)
1601 		return err;
1602 
1603 	if (unix_passcred_enabled(socket, other)) {
1604 		scm->pid = get_pid(task_tgid(current));
1605 		current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1606 	}
1607 	return err;
1608 }
1609 
1610 static bool unix_skb_scm_eq(struct sk_buff *skb,
1611 			    struct scm_cookie *scm)
1612 {
1613 	const struct unix_skb_parms *u = &UNIXCB(skb);
1614 
1615 	return u->pid == scm->pid &&
1616 	       uid_eq(u->uid, scm->creds.uid) &&
1617 	       gid_eq(u->gid, scm->creds.gid) &&
1618 	       unix_secdata_eq(scm, skb);
1619 }
1620 
1621 /*
1622  *	Send AF_UNIX data.
1623  */
1624 
1625 static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1626 			      size_t len)
1627 {
1628 	struct sock *sk = sock->sk;
1629 	struct net *net = sock_net(sk);
1630 	struct unix_sock *u = unix_sk(sk);
1631 	DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1632 	struct sock *other = NULL;
1633 	int namelen = 0; /* fake GCC */
1634 	int err;
1635 	unsigned int hash;
1636 	struct sk_buff *skb;
1637 	long timeo;
1638 	struct scm_cookie scm;
1639 	int data_len = 0;
1640 	int sk_locked;
1641 
1642 	wait_for_unix_gc();
1643 	err = scm_send(sock, msg, &scm, false);
1644 	if (err < 0)
1645 		return err;
1646 
1647 	err = -EOPNOTSUPP;
1648 	if (msg->msg_flags&MSG_OOB)
1649 		goto out;
1650 
1651 	if (msg->msg_namelen) {
1652 		err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1653 		if (err < 0)
1654 			goto out;
1655 		namelen = err;
1656 	} else {
1657 		sunaddr = NULL;
1658 		err = -ENOTCONN;
1659 		other = unix_peer_get(sk);
1660 		if (!other)
1661 			goto out;
1662 	}
1663 
1664 	if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1665 	    && (err = unix_autobind(sock)) != 0)
1666 		goto out;
1667 
1668 	err = -EMSGSIZE;
1669 	if (len > sk->sk_sndbuf - 32)
1670 		goto out;
1671 
1672 	if (len > SKB_MAX_ALLOC) {
1673 		data_len = min_t(size_t,
1674 				 len - SKB_MAX_ALLOC,
1675 				 MAX_SKB_FRAGS * PAGE_SIZE);
1676 		data_len = PAGE_ALIGN(data_len);
1677 
1678 		BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1679 	}
1680 
1681 	skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1682 				   msg->msg_flags & MSG_DONTWAIT, &err,
1683 				   PAGE_ALLOC_COSTLY_ORDER);
1684 	if (skb == NULL)
1685 		goto out;
1686 
1687 	err = unix_scm_to_skb(&scm, skb, true);
1688 	if (err < 0)
1689 		goto out_free;
1690 
1691 	skb_put(skb, len - data_len);
1692 	skb->data_len = data_len;
1693 	skb->len = len;
1694 	err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1695 	if (err)
1696 		goto out_free;
1697 
1698 	timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1699 
1700 restart:
1701 	if (!other) {
1702 		err = -ECONNRESET;
1703 		if (sunaddr == NULL)
1704 			goto out_free;
1705 
1706 		other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1707 					hash, &err);
1708 		if (other == NULL)
1709 			goto out_free;
1710 	}
1711 
1712 	if (sk_filter(other, skb) < 0) {
1713 		/* Toss the packet but do not return any error to the sender */
1714 		err = len;
1715 		goto out_free;
1716 	}
1717 
1718 	sk_locked = 0;
1719 	unix_state_lock(other);
1720 restart_locked:
1721 	err = -EPERM;
1722 	if (!unix_may_send(sk, other))
1723 		goto out_unlock;
1724 
1725 	if (unlikely(sock_flag(other, SOCK_DEAD))) {
1726 		/*
1727 		 *	Check with 1003.1g - what should
1728 		 *	datagram error
1729 		 */
1730 		unix_state_unlock(other);
1731 		sock_put(other);
1732 
1733 		if (!sk_locked)
1734 			unix_state_lock(sk);
1735 
1736 		err = 0;
1737 		if (unix_peer(sk) == other) {
1738 			unix_peer(sk) = NULL;
1739 			unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1740 
1741 			unix_state_unlock(sk);
1742 
1743 			unix_dgram_disconnected(sk, other);
1744 			sock_put(other);
1745 			err = -ECONNREFUSED;
1746 		} else {
1747 			unix_state_unlock(sk);
1748 		}
1749 
1750 		other = NULL;
1751 		if (err)
1752 			goto out_free;
1753 		goto restart;
1754 	}
1755 
1756 	err = -EPIPE;
1757 	if (other->sk_shutdown & RCV_SHUTDOWN)
1758 		goto out_unlock;
1759 
1760 	if (sk->sk_type != SOCK_SEQPACKET) {
1761 		err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1762 		if (err)
1763 			goto out_unlock;
1764 	}
1765 
1766 	/* other == sk && unix_peer(other) != sk if
1767 	 * - unix_peer(sk) == NULL, destination address bound to sk
1768 	 * - unix_peer(sk) == sk by time of get but disconnected before lock
1769 	 */
1770 	if (other != sk &&
1771 	    unlikely(unix_peer(other) != sk && unix_recvq_full(other))) {
1772 		if (timeo) {
1773 			timeo = unix_wait_for_peer(other, timeo);
1774 
1775 			err = sock_intr_errno(timeo);
1776 			if (signal_pending(current))
1777 				goto out_free;
1778 
1779 			goto restart;
1780 		}
1781 
1782 		if (!sk_locked) {
1783 			unix_state_unlock(other);
1784 			unix_state_double_lock(sk, other);
1785 		}
1786 
1787 		if (unix_peer(sk) != other ||
1788 		    unix_dgram_peer_wake_me(sk, other)) {
1789 			err = -EAGAIN;
1790 			sk_locked = 1;
1791 			goto out_unlock;
1792 		}
1793 
1794 		if (!sk_locked) {
1795 			sk_locked = 1;
1796 			goto restart_locked;
1797 		}
1798 	}
1799 
1800 	if (unlikely(sk_locked))
1801 		unix_state_unlock(sk);
1802 
1803 	if (sock_flag(other, SOCK_RCVTSTAMP))
1804 		__net_timestamp(skb);
1805 	maybe_add_creds(skb, sock, other);
1806 	skb_queue_tail(&other->sk_receive_queue, skb);
1807 	unix_state_unlock(other);
1808 	other->sk_data_ready(other);
1809 	sock_put(other);
1810 	scm_destroy(&scm);
1811 	return len;
1812 
1813 out_unlock:
1814 	if (sk_locked)
1815 		unix_state_unlock(sk);
1816 	unix_state_unlock(other);
1817 out_free:
1818 	kfree_skb(skb);
1819 out:
1820 	if (other)
1821 		sock_put(other);
1822 	scm_destroy(&scm);
1823 	return err;
1824 }
1825 
1826 /* We use paged skbs for stream sockets, and limit occupancy to 32768
1827  * bytes, and a minimun of a full page.
1828  */
1829 #define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1830 
1831 static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1832 			       size_t len)
1833 {
1834 	struct sock *sk = sock->sk;
1835 	struct sock *other = NULL;
1836 	int err, size;
1837 	struct sk_buff *skb;
1838 	int sent = 0;
1839 	struct scm_cookie scm;
1840 	bool fds_sent = false;
1841 	int data_len;
1842 
1843 	wait_for_unix_gc();
1844 	err = scm_send(sock, msg, &scm, false);
1845 	if (err < 0)
1846 		return err;
1847 
1848 	err = -EOPNOTSUPP;
1849 	if (msg->msg_flags&MSG_OOB)
1850 		goto out_err;
1851 
1852 	if (msg->msg_namelen) {
1853 		err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1854 		goto out_err;
1855 	} else {
1856 		err = -ENOTCONN;
1857 		other = unix_peer(sk);
1858 		if (!other)
1859 			goto out_err;
1860 	}
1861 
1862 	if (sk->sk_shutdown & SEND_SHUTDOWN)
1863 		goto pipe_err;
1864 
1865 	while (sent < len) {
1866 		size = len - sent;
1867 
1868 		/* Keep two messages in the pipe so it schedules better */
1869 		size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1870 
1871 		/* allow fallback to order-0 allocations */
1872 		size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1873 
1874 		data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1875 
1876 		data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1877 
1878 		skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1879 					   msg->msg_flags & MSG_DONTWAIT, &err,
1880 					   get_order(UNIX_SKB_FRAGS_SZ));
1881 		if (!skb)
1882 			goto out_err;
1883 
1884 		/* Only send the fds in the first buffer */
1885 		err = unix_scm_to_skb(&scm, skb, !fds_sent);
1886 		if (err < 0) {
1887 			kfree_skb(skb);
1888 			goto out_err;
1889 		}
1890 		fds_sent = true;
1891 
1892 		skb_put(skb, size - data_len);
1893 		skb->data_len = data_len;
1894 		skb->len = size;
1895 		err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
1896 		if (err) {
1897 			kfree_skb(skb);
1898 			goto out_err;
1899 		}
1900 
1901 		unix_state_lock(other);
1902 
1903 		if (sock_flag(other, SOCK_DEAD) ||
1904 		    (other->sk_shutdown & RCV_SHUTDOWN))
1905 			goto pipe_err_free;
1906 
1907 		maybe_add_creds(skb, sock, other);
1908 		skb_queue_tail(&other->sk_receive_queue, skb);
1909 		unix_state_unlock(other);
1910 		other->sk_data_ready(other);
1911 		sent += size;
1912 	}
1913 
1914 	scm_destroy(&scm);
1915 
1916 	return sent;
1917 
1918 pipe_err_free:
1919 	unix_state_unlock(other);
1920 	kfree_skb(skb);
1921 pipe_err:
1922 	if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1923 		send_sig(SIGPIPE, current, 0);
1924 	err = -EPIPE;
1925 out_err:
1926 	scm_destroy(&scm);
1927 	return sent ? : err;
1928 }
1929 
1930 static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1931 				    int offset, size_t size, int flags)
1932 {
1933 	int err;
1934 	bool send_sigpipe = false;
1935 	bool init_scm = true;
1936 	struct scm_cookie scm;
1937 	struct sock *other, *sk = socket->sk;
1938 	struct sk_buff *skb, *newskb = NULL, *tail = NULL;
1939 
1940 	if (flags & MSG_OOB)
1941 		return -EOPNOTSUPP;
1942 
1943 	other = unix_peer(sk);
1944 	if (!other || sk->sk_state != TCP_ESTABLISHED)
1945 		return -ENOTCONN;
1946 
1947 	if (false) {
1948 alloc_skb:
1949 		unix_state_unlock(other);
1950 		mutex_unlock(&unix_sk(other)->iolock);
1951 		newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
1952 					      &err, 0);
1953 		if (!newskb)
1954 			goto err;
1955 	}
1956 
1957 	/* we must acquire iolock as we modify already present
1958 	 * skbs in the sk_receive_queue and mess with skb->len
1959 	 */
1960 	err = mutex_lock_interruptible(&unix_sk(other)->iolock);
1961 	if (err) {
1962 		err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
1963 		goto err;
1964 	}
1965 
1966 	if (sk->sk_shutdown & SEND_SHUTDOWN) {
1967 		err = -EPIPE;
1968 		send_sigpipe = true;
1969 		goto err_unlock;
1970 	}
1971 
1972 	unix_state_lock(other);
1973 
1974 	if (sock_flag(other, SOCK_DEAD) ||
1975 	    other->sk_shutdown & RCV_SHUTDOWN) {
1976 		err = -EPIPE;
1977 		send_sigpipe = true;
1978 		goto err_state_unlock;
1979 	}
1980 
1981 	if (init_scm) {
1982 		err = maybe_init_creds(&scm, socket, other);
1983 		if (err)
1984 			goto err_state_unlock;
1985 		init_scm = false;
1986 	}
1987 
1988 	skb = skb_peek_tail(&other->sk_receive_queue);
1989 	if (tail && tail == skb) {
1990 		skb = newskb;
1991 	} else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
1992 		if (newskb) {
1993 			skb = newskb;
1994 		} else {
1995 			tail = skb;
1996 			goto alloc_skb;
1997 		}
1998 	} else if (newskb) {
1999 		/* this is fast path, we don't necessarily need to
2000 		 * call to kfree_skb even though with newskb == NULL
2001 		 * this - does no harm
2002 		 */
2003 		consume_skb(newskb);
2004 		newskb = NULL;
2005 	}
2006 
2007 	if (skb_append_pagefrags(skb, page, offset, size)) {
2008 		tail = skb;
2009 		goto alloc_skb;
2010 	}
2011 
2012 	skb->len += size;
2013 	skb->data_len += size;
2014 	skb->truesize += size;
2015 	refcount_add(size, &sk->sk_wmem_alloc);
2016 
2017 	if (newskb) {
2018 		err = unix_scm_to_skb(&scm, skb, false);
2019 		if (err)
2020 			goto err_state_unlock;
2021 		spin_lock(&other->sk_receive_queue.lock);
2022 		__skb_queue_tail(&other->sk_receive_queue, newskb);
2023 		spin_unlock(&other->sk_receive_queue.lock);
2024 	}
2025 
2026 	unix_state_unlock(other);
2027 	mutex_unlock(&unix_sk(other)->iolock);
2028 
2029 	other->sk_data_ready(other);
2030 	scm_destroy(&scm);
2031 	return size;
2032 
2033 err_state_unlock:
2034 	unix_state_unlock(other);
2035 err_unlock:
2036 	mutex_unlock(&unix_sk(other)->iolock);
2037 err:
2038 	kfree_skb(newskb);
2039 	if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2040 		send_sig(SIGPIPE, current, 0);
2041 	if (!init_scm)
2042 		scm_destroy(&scm);
2043 	return err;
2044 }
2045 
2046 static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2047 				  size_t len)
2048 {
2049 	int err;
2050 	struct sock *sk = sock->sk;
2051 
2052 	err = sock_error(sk);
2053 	if (err)
2054 		return err;
2055 
2056 	if (sk->sk_state != TCP_ESTABLISHED)
2057 		return -ENOTCONN;
2058 
2059 	if (msg->msg_namelen)
2060 		msg->msg_namelen = 0;
2061 
2062 	return unix_dgram_sendmsg(sock, msg, len);
2063 }
2064 
2065 static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2066 				  size_t size, int flags)
2067 {
2068 	struct sock *sk = sock->sk;
2069 
2070 	if (sk->sk_state != TCP_ESTABLISHED)
2071 		return -ENOTCONN;
2072 
2073 	return unix_dgram_recvmsg(sock, msg, size, flags);
2074 }
2075 
2076 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2077 {
2078 	struct unix_sock *u = unix_sk(sk);
2079 
2080 	if (u->addr) {
2081 		msg->msg_namelen = u->addr->len;
2082 		memcpy(msg->msg_name, u->addr->name, u->addr->len);
2083 	}
2084 }
2085 
2086 static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2087 			      size_t size, int flags)
2088 {
2089 	struct scm_cookie scm;
2090 	struct sock *sk = sock->sk;
2091 	struct unix_sock *u = unix_sk(sk);
2092 	struct sk_buff *skb, *last;
2093 	long timeo;
2094 	int err;
2095 	int peeked, skip;
2096 
2097 	err = -EOPNOTSUPP;
2098 	if (flags&MSG_OOB)
2099 		goto out;
2100 
2101 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2102 
2103 	do {
2104 		mutex_lock(&u->iolock);
2105 
2106 		skip = sk_peek_offset(sk, flags);
2107 		skb = __skb_try_recv_datagram(sk, flags, NULL, &peeked, &skip,
2108 					      &err, &last);
2109 		if (skb)
2110 			break;
2111 
2112 		mutex_unlock(&u->iolock);
2113 
2114 		if (err != -EAGAIN)
2115 			break;
2116 	} while (timeo &&
2117 		 !__skb_wait_for_more_packets(sk, &err, &timeo, last));
2118 
2119 	if (!skb) { /* implies iolock unlocked */
2120 		unix_state_lock(sk);
2121 		/* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2122 		if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2123 		    (sk->sk_shutdown & RCV_SHUTDOWN))
2124 			err = 0;
2125 		unix_state_unlock(sk);
2126 		goto out;
2127 	}
2128 
2129 	if (wq_has_sleeper(&u->peer_wait))
2130 		wake_up_interruptible_sync_poll(&u->peer_wait,
2131 						POLLOUT | POLLWRNORM |
2132 						POLLWRBAND);
2133 
2134 	if (msg->msg_name)
2135 		unix_copy_addr(msg, skb->sk);
2136 
2137 	if (size > skb->len - skip)
2138 		size = skb->len - skip;
2139 	else if (size < skb->len - skip)
2140 		msg->msg_flags |= MSG_TRUNC;
2141 
2142 	err = skb_copy_datagram_msg(skb, skip, msg, size);
2143 	if (err)
2144 		goto out_free;
2145 
2146 	if (sock_flag(sk, SOCK_RCVTSTAMP))
2147 		__sock_recv_timestamp(msg, sk, skb);
2148 
2149 	memset(&scm, 0, sizeof(scm));
2150 
2151 	scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2152 	unix_set_secdata(&scm, skb);
2153 
2154 	if (!(flags & MSG_PEEK)) {
2155 		if (UNIXCB(skb).fp)
2156 			unix_detach_fds(&scm, skb);
2157 
2158 		sk_peek_offset_bwd(sk, skb->len);
2159 	} else {
2160 		/* It is questionable: on PEEK we could:
2161 		   - do not return fds - good, but too simple 8)
2162 		   - return fds, and do not return them on read (old strategy,
2163 		     apparently wrong)
2164 		   - clone fds (I chose it for now, it is the most universal
2165 		     solution)
2166 
2167 		   POSIX 1003.1g does not actually define this clearly
2168 		   at all. POSIX 1003.1g doesn't define a lot of things
2169 		   clearly however!
2170 
2171 		*/
2172 
2173 		sk_peek_offset_fwd(sk, size);
2174 
2175 		if (UNIXCB(skb).fp)
2176 			scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2177 	}
2178 	err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2179 
2180 	scm_recv(sock, msg, &scm, flags);
2181 
2182 out_free:
2183 	skb_free_datagram(sk, skb);
2184 	mutex_unlock(&u->iolock);
2185 out:
2186 	return err;
2187 }
2188 
2189 /*
2190  *	Sleep until more data has arrived. But check for races..
2191  */
2192 static long unix_stream_data_wait(struct sock *sk, long timeo,
2193 				  struct sk_buff *last, unsigned int last_len,
2194 				  bool freezable)
2195 {
2196 	struct sk_buff *tail;
2197 	DEFINE_WAIT(wait);
2198 
2199 	unix_state_lock(sk);
2200 
2201 	for (;;) {
2202 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2203 
2204 		tail = skb_peek_tail(&sk->sk_receive_queue);
2205 		if (tail != last ||
2206 		    (tail && tail->len != last_len) ||
2207 		    sk->sk_err ||
2208 		    (sk->sk_shutdown & RCV_SHUTDOWN) ||
2209 		    signal_pending(current) ||
2210 		    !timeo)
2211 			break;
2212 
2213 		sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2214 		unix_state_unlock(sk);
2215 		if (freezable)
2216 			timeo = freezable_schedule_timeout(timeo);
2217 		else
2218 			timeo = schedule_timeout(timeo);
2219 		unix_state_lock(sk);
2220 
2221 		if (sock_flag(sk, SOCK_DEAD))
2222 			break;
2223 
2224 		sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2225 	}
2226 
2227 	finish_wait(sk_sleep(sk), &wait);
2228 	unix_state_unlock(sk);
2229 	return timeo;
2230 }
2231 
2232 static unsigned int unix_skb_len(const struct sk_buff *skb)
2233 {
2234 	return skb->len - UNIXCB(skb).consumed;
2235 }
2236 
2237 struct unix_stream_read_state {
2238 	int (*recv_actor)(struct sk_buff *, int, int,
2239 			  struct unix_stream_read_state *);
2240 	struct socket *socket;
2241 	struct msghdr *msg;
2242 	struct pipe_inode_info *pipe;
2243 	size_t size;
2244 	int flags;
2245 	unsigned int splice_flags;
2246 };
2247 
2248 static int unix_stream_read_generic(struct unix_stream_read_state *state,
2249 				    bool freezable)
2250 {
2251 	struct scm_cookie scm;
2252 	struct socket *sock = state->socket;
2253 	struct sock *sk = sock->sk;
2254 	struct unix_sock *u = unix_sk(sk);
2255 	int copied = 0;
2256 	int flags = state->flags;
2257 	int noblock = flags & MSG_DONTWAIT;
2258 	bool check_creds = false;
2259 	int target;
2260 	int err = 0;
2261 	long timeo;
2262 	int skip;
2263 	size_t size = state->size;
2264 	unsigned int last_len;
2265 
2266 	if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2267 		err = -EINVAL;
2268 		goto out;
2269 	}
2270 
2271 	if (unlikely(flags & MSG_OOB)) {
2272 		err = -EOPNOTSUPP;
2273 		goto out;
2274 	}
2275 
2276 	target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2277 	timeo = sock_rcvtimeo(sk, noblock);
2278 
2279 	memset(&scm, 0, sizeof(scm));
2280 
2281 	/* Lock the socket to prevent queue disordering
2282 	 * while sleeps in memcpy_tomsg
2283 	 */
2284 	mutex_lock(&u->iolock);
2285 
2286 	skip = max(sk_peek_offset(sk, flags), 0);
2287 
2288 	do {
2289 		int chunk;
2290 		bool drop_skb;
2291 		struct sk_buff *skb, *last;
2292 
2293 redo:
2294 		unix_state_lock(sk);
2295 		if (sock_flag(sk, SOCK_DEAD)) {
2296 			err = -ECONNRESET;
2297 			goto unlock;
2298 		}
2299 		last = skb = skb_peek(&sk->sk_receive_queue);
2300 		last_len = last ? last->len : 0;
2301 again:
2302 		if (skb == NULL) {
2303 			if (copied >= target)
2304 				goto unlock;
2305 
2306 			/*
2307 			 *	POSIX 1003.1g mandates this order.
2308 			 */
2309 
2310 			err = sock_error(sk);
2311 			if (err)
2312 				goto unlock;
2313 			if (sk->sk_shutdown & RCV_SHUTDOWN)
2314 				goto unlock;
2315 
2316 			unix_state_unlock(sk);
2317 			if (!timeo) {
2318 				err = -EAGAIN;
2319 				break;
2320 			}
2321 
2322 			mutex_unlock(&u->iolock);
2323 
2324 			timeo = unix_stream_data_wait(sk, timeo, last,
2325 						      last_len, freezable);
2326 
2327 			if (signal_pending(current)) {
2328 				err = sock_intr_errno(timeo);
2329 				scm_destroy(&scm);
2330 				goto out;
2331 			}
2332 
2333 			mutex_lock(&u->iolock);
2334 			goto redo;
2335 unlock:
2336 			unix_state_unlock(sk);
2337 			break;
2338 		}
2339 
2340 		while (skip >= unix_skb_len(skb)) {
2341 			skip -= unix_skb_len(skb);
2342 			last = skb;
2343 			last_len = skb->len;
2344 			skb = skb_peek_next(skb, &sk->sk_receive_queue);
2345 			if (!skb)
2346 				goto again;
2347 		}
2348 
2349 		unix_state_unlock(sk);
2350 
2351 		if (check_creds) {
2352 			/* Never glue messages from different writers */
2353 			if (!unix_skb_scm_eq(skb, &scm))
2354 				break;
2355 		} else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2356 			/* Copy credentials */
2357 			scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2358 			unix_set_secdata(&scm, skb);
2359 			check_creds = true;
2360 		}
2361 
2362 		/* Copy address just once */
2363 		if (state->msg && state->msg->msg_name) {
2364 			DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2365 					 state->msg->msg_name);
2366 			unix_copy_addr(state->msg, skb->sk);
2367 			sunaddr = NULL;
2368 		}
2369 
2370 		chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2371 		skb_get(skb);
2372 		chunk = state->recv_actor(skb, skip, chunk, state);
2373 		drop_skb = !unix_skb_len(skb);
2374 		/* skb is only safe to use if !drop_skb */
2375 		consume_skb(skb);
2376 		if (chunk < 0) {
2377 			if (copied == 0)
2378 				copied = -EFAULT;
2379 			break;
2380 		}
2381 		copied += chunk;
2382 		size -= chunk;
2383 
2384 		if (drop_skb) {
2385 			/* the skb was touched by a concurrent reader;
2386 			 * we should not expect anything from this skb
2387 			 * anymore and assume it invalid - we can be
2388 			 * sure it was dropped from the socket queue
2389 			 *
2390 			 * let's report a short read
2391 			 */
2392 			err = 0;
2393 			break;
2394 		}
2395 
2396 		/* Mark read part of skb as used */
2397 		if (!(flags & MSG_PEEK)) {
2398 			UNIXCB(skb).consumed += chunk;
2399 
2400 			sk_peek_offset_bwd(sk, chunk);
2401 
2402 			if (UNIXCB(skb).fp)
2403 				unix_detach_fds(&scm, skb);
2404 
2405 			if (unix_skb_len(skb))
2406 				break;
2407 
2408 			skb_unlink(skb, &sk->sk_receive_queue);
2409 			consume_skb(skb);
2410 
2411 			if (scm.fp)
2412 				break;
2413 		} else {
2414 			/* It is questionable, see note in unix_dgram_recvmsg.
2415 			 */
2416 			if (UNIXCB(skb).fp)
2417 				scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2418 
2419 			sk_peek_offset_fwd(sk, chunk);
2420 
2421 			if (UNIXCB(skb).fp)
2422 				break;
2423 
2424 			skip = 0;
2425 			last = skb;
2426 			last_len = skb->len;
2427 			unix_state_lock(sk);
2428 			skb = skb_peek_next(skb, &sk->sk_receive_queue);
2429 			if (skb)
2430 				goto again;
2431 			unix_state_unlock(sk);
2432 			break;
2433 		}
2434 	} while (size);
2435 
2436 	mutex_unlock(&u->iolock);
2437 	if (state->msg)
2438 		scm_recv(sock, state->msg, &scm, flags);
2439 	else
2440 		scm_destroy(&scm);
2441 out:
2442 	return copied ? : err;
2443 }
2444 
2445 static int unix_stream_read_actor(struct sk_buff *skb,
2446 				  int skip, int chunk,
2447 				  struct unix_stream_read_state *state)
2448 {
2449 	int ret;
2450 
2451 	ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2452 				    state->msg, chunk);
2453 	return ret ?: chunk;
2454 }
2455 
2456 static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2457 			       size_t size, int flags)
2458 {
2459 	struct unix_stream_read_state state = {
2460 		.recv_actor = unix_stream_read_actor,
2461 		.socket = sock,
2462 		.msg = msg,
2463 		.size = size,
2464 		.flags = flags
2465 	};
2466 
2467 	return unix_stream_read_generic(&state, true);
2468 }
2469 
2470 static int unix_stream_splice_actor(struct sk_buff *skb,
2471 				    int skip, int chunk,
2472 				    struct unix_stream_read_state *state)
2473 {
2474 	return skb_splice_bits(skb, state->socket->sk,
2475 			       UNIXCB(skb).consumed + skip,
2476 			       state->pipe, chunk, state->splice_flags);
2477 }
2478 
2479 static ssize_t unix_stream_splice_read(struct socket *sock,  loff_t *ppos,
2480 				       struct pipe_inode_info *pipe,
2481 				       size_t size, unsigned int flags)
2482 {
2483 	struct unix_stream_read_state state = {
2484 		.recv_actor = unix_stream_splice_actor,
2485 		.socket = sock,
2486 		.pipe = pipe,
2487 		.size = size,
2488 		.splice_flags = flags,
2489 	};
2490 
2491 	if (unlikely(*ppos))
2492 		return -ESPIPE;
2493 
2494 	if (sock->file->f_flags & O_NONBLOCK ||
2495 	    flags & SPLICE_F_NONBLOCK)
2496 		state.flags = MSG_DONTWAIT;
2497 
2498 	return unix_stream_read_generic(&state, false);
2499 }
2500 
2501 static int unix_shutdown(struct socket *sock, int mode)
2502 {
2503 	struct sock *sk = sock->sk;
2504 	struct sock *other;
2505 
2506 	if (mode < SHUT_RD || mode > SHUT_RDWR)
2507 		return -EINVAL;
2508 	/* This maps:
2509 	 * SHUT_RD   (0) -> RCV_SHUTDOWN  (1)
2510 	 * SHUT_WR   (1) -> SEND_SHUTDOWN (2)
2511 	 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2512 	 */
2513 	++mode;
2514 
2515 	unix_state_lock(sk);
2516 	sk->sk_shutdown |= mode;
2517 	other = unix_peer(sk);
2518 	if (other)
2519 		sock_hold(other);
2520 	unix_state_unlock(sk);
2521 	sk->sk_state_change(sk);
2522 
2523 	if (other &&
2524 		(sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2525 
2526 		int peer_mode = 0;
2527 
2528 		if (mode&RCV_SHUTDOWN)
2529 			peer_mode |= SEND_SHUTDOWN;
2530 		if (mode&SEND_SHUTDOWN)
2531 			peer_mode |= RCV_SHUTDOWN;
2532 		unix_state_lock(other);
2533 		other->sk_shutdown |= peer_mode;
2534 		unix_state_unlock(other);
2535 		other->sk_state_change(other);
2536 		if (peer_mode == SHUTDOWN_MASK)
2537 			sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2538 		else if (peer_mode & RCV_SHUTDOWN)
2539 			sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2540 	}
2541 	if (other)
2542 		sock_put(other);
2543 
2544 	return 0;
2545 }
2546 
2547 long unix_inq_len(struct sock *sk)
2548 {
2549 	struct sk_buff *skb;
2550 	long amount = 0;
2551 
2552 	if (sk->sk_state == TCP_LISTEN)
2553 		return -EINVAL;
2554 
2555 	spin_lock(&sk->sk_receive_queue.lock);
2556 	if (sk->sk_type == SOCK_STREAM ||
2557 	    sk->sk_type == SOCK_SEQPACKET) {
2558 		skb_queue_walk(&sk->sk_receive_queue, skb)
2559 			amount += unix_skb_len(skb);
2560 	} else {
2561 		skb = skb_peek(&sk->sk_receive_queue);
2562 		if (skb)
2563 			amount = skb->len;
2564 	}
2565 	spin_unlock(&sk->sk_receive_queue.lock);
2566 
2567 	return amount;
2568 }
2569 EXPORT_SYMBOL_GPL(unix_inq_len);
2570 
2571 long unix_outq_len(struct sock *sk)
2572 {
2573 	return sk_wmem_alloc_get(sk);
2574 }
2575 EXPORT_SYMBOL_GPL(unix_outq_len);
2576 
2577 static int unix_open_file(struct sock *sk)
2578 {
2579 	struct path path;
2580 	struct file *f;
2581 	int fd;
2582 
2583 	if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2584 		return -EPERM;
2585 
2586 	unix_state_lock(sk);
2587 	path = unix_sk(sk)->path;
2588 	if (!path.dentry) {
2589 		unix_state_unlock(sk);
2590 		return -ENOENT;
2591 	}
2592 
2593 	path_get(&path);
2594 	unix_state_unlock(sk);
2595 
2596 	fd = get_unused_fd_flags(O_CLOEXEC);
2597 	if (fd < 0)
2598 		goto out;
2599 
2600 	f = dentry_open(&path, O_PATH, current_cred());
2601 	if (IS_ERR(f)) {
2602 		put_unused_fd(fd);
2603 		fd = PTR_ERR(f);
2604 		goto out;
2605 	}
2606 
2607 	fd_install(fd, f);
2608 out:
2609 	path_put(&path);
2610 
2611 	return fd;
2612 }
2613 
2614 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2615 {
2616 	struct sock *sk = sock->sk;
2617 	long amount = 0;
2618 	int err;
2619 
2620 	switch (cmd) {
2621 	case SIOCOUTQ:
2622 		amount = unix_outq_len(sk);
2623 		err = put_user(amount, (int __user *)arg);
2624 		break;
2625 	case SIOCINQ:
2626 		amount = unix_inq_len(sk);
2627 		if (amount < 0)
2628 			err = amount;
2629 		else
2630 			err = put_user(amount, (int __user *)arg);
2631 		break;
2632 	case SIOCUNIXFILE:
2633 		err = unix_open_file(sk);
2634 		break;
2635 	default:
2636 		err = -ENOIOCTLCMD;
2637 		break;
2638 	}
2639 	return err;
2640 }
2641 
2642 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2643 {
2644 	struct sock *sk = sock->sk;
2645 	unsigned int mask;
2646 
2647 	sock_poll_wait(file, sk_sleep(sk), wait);
2648 	mask = 0;
2649 
2650 	/* exceptional events? */
2651 	if (sk->sk_err)
2652 		mask |= POLLERR;
2653 	if (sk->sk_shutdown == SHUTDOWN_MASK)
2654 		mask |= POLLHUP;
2655 	if (sk->sk_shutdown & RCV_SHUTDOWN)
2656 		mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2657 
2658 	/* readable? */
2659 	if (!skb_queue_empty(&sk->sk_receive_queue))
2660 		mask |= POLLIN | POLLRDNORM;
2661 
2662 	/* Connection-based need to check for termination and startup */
2663 	if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2664 	    sk->sk_state == TCP_CLOSE)
2665 		mask |= POLLHUP;
2666 
2667 	/*
2668 	 * we set writable also when the other side has shut down the
2669 	 * connection. This prevents stuck sockets.
2670 	 */
2671 	if (unix_writable(sk))
2672 		mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2673 
2674 	return mask;
2675 }
2676 
2677 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2678 				    poll_table *wait)
2679 {
2680 	struct sock *sk = sock->sk, *other;
2681 	unsigned int mask, writable;
2682 
2683 	sock_poll_wait(file, sk_sleep(sk), wait);
2684 	mask = 0;
2685 
2686 	/* exceptional events? */
2687 	if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2688 		mask |= POLLERR |
2689 			(sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2690 
2691 	if (sk->sk_shutdown & RCV_SHUTDOWN)
2692 		mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2693 	if (sk->sk_shutdown == SHUTDOWN_MASK)
2694 		mask |= POLLHUP;
2695 
2696 	/* readable? */
2697 	if (!skb_queue_empty(&sk->sk_receive_queue))
2698 		mask |= POLLIN | POLLRDNORM;
2699 
2700 	/* Connection-based need to check for termination and startup */
2701 	if (sk->sk_type == SOCK_SEQPACKET) {
2702 		if (sk->sk_state == TCP_CLOSE)
2703 			mask |= POLLHUP;
2704 		/* connection hasn't started yet? */
2705 		if (sk->sk_state == TCP_SYN_SENT)
2706 			return mask;
2707 	}
2708 
2709 	/* No write status requested, avoid expensive OUT tests. */
2710 	if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2711 		return mask;
2712 
2713 	writable = unix_writable(sk);
2714 	if (writable) {
2715 		unix_state_lock(sk);
2716 
2717 		other = unix_peer(sk);
2718 		if (other && unix_peer(other) != sk &&
2719 		    unix_recvq_full(other) &&
2720 		    unix_dgram_peer_wake_me(sk, other))
2721 			writable = 0;
2722 
2723 		unix_state_unlock(sk);
2724 	}
2725 
2726 	if (writable)
2727 		mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2728 	else
2729 		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2730 
2731 	return mask;
2732 }
2733 
2734 #ifdef CONFIG_PROC_FS
2735 
2736 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2737 
2738 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2739 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2740 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2741 
2742 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2743 {
2744 	unsigned long offset = get_offset(*pos);
2745 	unsigned long bucket = get_bucket(*pos);
2746 	struct sock *sk;
2747 	unsigned long count = 0;
2748 
2749 	for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2750 		if (sock_net(sk) != seq_file_net(seq))
2751 			continue;
2752 		if (++count == offset)
2753 			break;
2754 	}
2755 
2756 	return sk;
2757 }
2758 
2759 static struct sock *unix_next_socket(struct seq_file *seq,
2760 				     struct sock *sk,
2761 				     loff_t *pos)
2762 {
2763 	unsigned long bucket;
2764 
2765 	while (sk > (struct sock *)SEQ_START_TOKEN) {
2766 		sk = sk_next(sk);
2767 		if (!sk)
2768 			goto next_bucket;
2769 		if (sock_net(sk) == seq_file_net(seq))
2770 			return sk;
2771 	}
2772 
2773 	do {
2774 		sk = unix_from_bucket(seq, pos);
2775 		if (sk)
2776 			return sk;
2777 
2778 next_bucket:
2779 		bucket = get_bucket(*pos) + 1;
2780 		*pos = set_bucket_offset(bucket, 1);
2781 	} while (bucket < ARRAY_SIZE(unix_socket_table));
2782 
2783 	return NULL;
2784 }
2785 
2786 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2787 	__acquires(unix_table_lock)
2788 {
2789 	spin_lock(&unix_table_lock);
2790 
2791 	if (!*pos)
2792 		return SEQ_START_TOKEN;
2793 
2794 	if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2795 		return NULL;
2796 
2797 	return unix_next_socket(seq, NULL, pos);
2798 }
2799 
2800 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2801 {
2802 	++*pos;
2803 	return unix_next_socket(seq, v, pos);
2804 }
2805 
2806 static void unix_seq_stop(struct seq_file *seq, void *v)
2807 	__releases(unix_table_lock)
2808 {
2809 	spin_unlock(&unix_table_lock);
2810 }
2811 
2812 static int unix_seq_show(struct seq_file *seq, void *v)
2813 {
2814 
2815 	if (v == SEQ_START_TOKEN)
2816 		seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2817 			 "Inode Path\n");
2818 	else {
2819 		struct sock *s = v;
2820 		struct unix_sock *u = unix_sk(s);
2821 		unix_state_lock(s);
2822 
2823 		seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2824 			s,
2825 			refcount_read(&s->sk_refcnt),
2826 			0,
2827 			s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2828 			s->sk_type,
2829 			s->sk_socket ?
2830 			(s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2831 			(s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2832 			sock_i_ino(s));
2833 
2834 		if (u->addr) {
2835 			int i, len;
2836 			seq_putc(seq, ' ');
2837 
2838 			i = 0;
2839 			len = u->addr->len - sizeof(short);
2840 			if (!UNIX_ABSTRACT(s))
2841 				len--;
2842 			else {
2843 				seq_putc(seq, '@');
2844 				i++;
2845 			}
2846 			for ( ; i < len; i++)
2847 				seq_putc(seq, u->addr->name->sun_path[i] ?:
2848 					 '@');
2849 		}
2850 		unix_state_unlock(s);
2851 		seq_putc(seq, '\n');
2852 	}
2853 
2854 	return 0;
2855 }
2856 
2857 static const struct seq_operations unix_seq_ops = {
2858 	.start  = unix_seq_start,
2859 	.next   = unix_seq_next,
2860 	.stop   = unix_seq_stop,
2861 	.show   = unix_seq_show,
2862 };
2863 
2864 static int unix_seq_open(struct inode *inode, struct file *file)
2865 {
2866 	return seq_open_net(inode, file, &unix_seq_ops,
2867 			    sizeof(struct seq_net_private));
2868 }
2869 
2870 static const struct file_operations unix_seq_fops = {
2871 	.owner		= THIS_MODULE,
2872 	.open		= unix_seq_open,
2873 	.read		= seq_read,
2874 	.llseek		= seq_lseek,
2875 	.release	= seq_release_net,
2876 };
2877 
2878 #endif
2879 
2880 static const struct net_proto_family unix_family_ops = {
2881 	.family = PF_UNIX,
2882 	.create = unix_create,
2883 	.owner	= THIS_MODULE,
2884 };
2885 
2886 
2887 static int __net_init unix_net_init(struct net *net)
2888 {
2889 	int error = -ENOMEM;
2890 
2891 	net->unx.sysctl_max_dgram_qlen = 10;
2892 	if (unix_sysctl_register(net))
2893 		goto out;
2894 
2895 #ifdef CONFIG_PROC_FS
2896 	if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2897 		unix_sysctl_unregister(net);
2898 		goto out;
2899 	}
2900 #endif
2901 	error = 0;
2902 out:
2903 	return error;
2904 }
2905 
2906 static void __net_exit unix_net_exit(struct net *net)
2907 {
2908 	unix_sysctl_unregister(net);
2909 	remove_proc_entry("unix", net->proc_net);
2910 }
2911 
2912 static struct pernet_operations unix_net_ops = {
2913 	.init = unix_net_init,
2914 	.exit = unix_net_exit,
2915 };
2916 
2917 static int __init af_unix_init(void)
2918 {
2919 	int rc = -1;
2920 
2921 	BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2922 
2923 	rc = proto_register(&unix_proto, 1);
2924 	if (rc != 0) {
2925 		pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2926 		goto out;
2927 	}
2928 
2929 	sock_register(&unix_family_ops);
2930 	register_pernet_subsys(&unix_net_ops);
2931 out:
2932 	return rc;
2933 }
2934 
2935 static void __exit af_unix_exit(void)
2936 {
2937 	sock_unregister(PF_UNIX);
2938 	proto_unregister(&unix_proto);
2939 	unregister_pernet_subsys(&unix_net_ops);
2940 }
2941 
2942 /* Earlier than device_initcall() so that other drivers invoking
2943    request_module() don't end up in a loop when modprobe tries
2944    to use a UNIX socket. But later than subsys_initcall() because
2945    we depend on stuff initialised there */
2946 fs_initcall(af_unix_init);
2947 module_exit(af_unix_exit);
2948 
2949 MODULE_LICENSE("GPL");
2950 MODULE_ALIAS_NETPROTO(PF_UNIX);
2951