af_unix: Don't set err in unix_stream_read_generic unless there was an error
[pandora-kernel.git] / net / unix / af_unix.c
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 static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
119 static DEFINE_SPINLOCK(unix_table_lock);
120 static atomic_long_t unix_nr_socks;
121
122 #define unix_sockets_unbound    (&unix_socket_table[UNIX_HASH_SIZE])
123
124 #define UNIX_ABSTRACT(sk)       (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
125
126 #ifdef CONFIG_SECURITY_NETWORK
127 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
128 {
129         memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
130 }
131
132 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
133 {
134         scm->secid = *UNIXSID(skb);
135 }
136 #else
137 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
138 { }
139
140 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
141 { }
142 #endif /* CONFIG_SECURITY_NETWORK */
143
144 /*
145  *  SMP locking strategy:
146  *    hash table is protected with spinlock unix_table_lock
147  *    each socket state is protected by separate spin lock.
148  */
149
150 static inline unsigned unix_hash_fold(__wsum n)
151 {
152         unsigned int hash = (__force unsigned int)csum_fold(n);
153
154         hash ^= hash>>8;
155         return hash&(UNIX_HASH_SIZE-1);
156 }
157
158 #define unix_peer(sk) (unix_sk(sk)->peer)
159
160 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
161 {
162         return unix_peer(osk) == sk;
163 }
164
165 static inline int unix_may_send(struct sock *sk, struct sock *osk)
166 {
167         return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
168 }
169
170 static inline int unix_recvq_full(struct sock const *sk)
171 {
172         return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
173 }
174
175 static struct sock *unix_peer_get(struct sock *s)
176 {
177         struct sock *peer;
178
179         unix_state_lock(s);
180         peer = unix_peer(s);
181         if (peer)
182                 sock_hold(peer);
183         unix_state_unlock(s);
184         return peer;
185 }
186
187 static inline void unix_release_addr(struct unix_address *addr)
188 {
189         if (atomic_dec_and_test(&addr->refcnt))
190                 kfree(addr);
191 }
192
193 /*
194  *      Check unix socket name:
195  *              - should be not zero length.
196  *              - if started by not zero, should be NULL terminated (FS object)
197  *              - if started by zero, it is abstract name.
198  */
199
200 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned *hashp)
201 {
202         if (len <= sizeof(short) || len > sizeof(*sunaddr))
203                 return -EINVAL;
204         if (!sunaddr || sunaddr->sun_family != AF_UNIX)
205                 return -EINVAL;
206         if (sunaddr->sun_path[0]) {
207                 /*
208                  * This may look like an off by one error but it is a bit more
209                  * subtle. 108 is the longest valid AF_UNIX path for a binding.
210                  * sun_path[108] doesn't as such exist.  However in kernel space
211                  * we are guaranteed that it is a valid memory location in our
212                  * kernel address buffer.
213                  */
214                 ((char *)sunaddr)[len] = 0;
215                 len = strlen(sunaddr->sun_path)+1+sizeof(short);
216                 return len;
217         }
218
219         *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
220         return len;
221 }
222
223 static void __unix_remove_socket(struct sock *sk)
224 {
225         sk_del_node_init(sk);
226 }
227
228 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
229 {
230         WARN_ON(!sk_unhashed(sk));
231         sk_add_node(sk, list);
232 }
233
234 static inline void unix_remove_socket(struct sock *sk)
235 {
236         spin_lock(&unix_table_lock);
237         __unix_remove_socket(sk);
238         spin_unlock(&unix_table_lock);
239 }
240
241 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
242 {
243         spin_lock(&unix_table_lock);
244         __unix_insert_socket(list, sk);
245         spin_unlock(&unix_table_lock);
246 }
247
248 static struct sock *__unix_find_socket_byname(struct net *net,
249                                               struct sockaddr_un *sunname,
250                                               int len, int type, unsigned hash)
251 {
252         struct sock *s;
253         struct hlist_node *node;
254
255         sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
256                 struct unix_sock *u = unix_sk(s);
257
258                 if (!net_eq(sock_net(s), net))
259                         continue;
260
261                 if (u->addr->len == len &&
262                     !memcmp(u->addr->name, sunname, len))
263                         goto found;
264         }
265         s = NULL;
266 found:
267         return s;
268 }
269
270 static inline struct sock *unix_find_socket_byname(struct net *net,
271                                                    struct sockaddr_un *sunname,
272                                                    int len, int type,
273                                                    unsigned hash)
274 {
275         struct sock *s;
276
277         spin_lock(&unix_table_lock);
278         s = __unix_find_socket_byname(net, sunname, len, type, hash);
279         if (s)
280                 sock_hold(s);
281         spin_unlock(&unix_table_lock);
282         return s;
283 }
284
285 static struct sock *unix_find_socket_byinode(struct inode *i)
286 {
287         struct sock *s;
288         struct hlist_node *node;
289
290         spin_lock(&unix_table_lock);
291         sk_for_each(s, node,
292                     &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
293                 struct dentry *dentry = unix_sk(s)->dentry;
294
295                 if (dentry && dentry->d_inode == i) {
296                         sock_hold(s);
297                         goto found;
298                 }
299         }
300         s = NULL;
301 found:
302         spin_unlock(&unix_table_lock);
303         return s;
304 }
305
306 /* Support code for asymmetrically connected dgram sockets
307  *
308  * If a datagram socket is connected to a socket not itself connected
309  * to the first socket (eg, /dev/log), clients may only enqueue more
310  * messages if the present receive queue of the server socket is not
311  * "too large". This means there's a second writeability condition
312  * poll and sendmsg need to test. The dgram recv code will do a wake
313  * up on the peer_wait wait queue of a socket upon reception of a
314  * datagram which needs to be propagated to sleeping would-be writers
315  * since these might not have sent anything so far. This can't be
316  * accomplished via poll_wait because the lifetime of the server
317  * socket might be less than that of its clients if these break their
318  * association with it or if the server socket is closed while clients
319  * are still connected to it and there's no way to inform "a polling
320  * implementation" that it should let go of a certain wait queue
321  *
322  * In order to propagate a wake up, a wait_queue_t of the client
323  * socket is enqueued on the peer_wait queue of the server socket
324  * whose wake function does a wake_up on the ordinary client socket
325  * wait queue. This connection is established whenever a write (or
326  * poll for write) hit the flow control condition and broken when the
327  * association to the server socket is dissolved or after a wake up
328  * was relayed.
329  */
330
331 static int unix_dgram_peer_wake_relay(wait_queue_t *q, unsigned mode, int flags,
332                                       void *key)
333 {
334         struct unix_sock *u;
335         wait_queue_head_t *u_sleep;
336
337         u = container_of(q, struct unix_sock, peer_wake);
338
339         __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
340                             q);
341         u->peer_wake.private = NULL;
342
343         /* relaying can only happen while the wq still exists */
344         u_sleep = sk_sleep(&u->sk);
345         if (u_sleep)
346                 wake_up_interruptible_poll(u_sleep, key);
347
348         return 0;
349 }
350
351 static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
352 {
353         struct unix_sock *u, *u_other;
354         int rc;
355
356         u = unix_sk(sk);
357         u_other = unix_sk(other);
358         rc = 0;
359         spin_lock(&u_other->peer_wait.lock);
360
361         if (!u->peer_wake.private) {
362                 u->peer_wake.private = other;
363                 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
364
365                 rc = 1;
366         }
367
368         spin_unlock(&u_other->peer_wait.lock);
369         return rc;
370 }
371
372 static void unix_dgram_peer_wake_disconnect(struct sock *sk,
373                                             struct sock *other)
374 {
375         struct unix_sock *u, *u_other;
376
377         u = unix_sk(sk);
378         u_other = unix_sk(other);
379         spin_lock(&u_other->peer_wait.lock);
380
381         if (u->peer_wake.private == other) {
382                 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
383                 u->peer_wake.private = NULL;
384         }
385
386         spin_unlock(&u_other->peer_wait.lock);
387 }
388
389 static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
390                                                    struct sock *other)
391 {
392         unix_dgram_peer_wake_disconnect(sk, other);
393         wake_up_interruptible_poll(sk_sleep(sk),
394                                    POLLOUT |
395                                    POLLWRNORM |
396                                    POLLWRBAND);
397 }
398
399 /* preconditions:
400  *      - unix_peer(sk) == other
401  *      - association is stable
402  */
403 static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
404 {
405         int connected;
406
407         connected = unix_dgram_peer_wake_connect(sk, other);
408
409         if (unix_recvq_full(other))
410                 return 1;
411
412         if (connected)
413                 unix_dgram_peer_wake_disconnect(sk, other);
414
415         return 0;
416 }
417
418 static inline int unix_writable(struct sock *sk)
419 {
420         return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
421 }
422
423 static void unix_write_space(struct sock *sk)
424 {
425         struct socket_wq *wq;
426
427         rcu_read_lock();
428         if (unix_writable(sk)) {
429                 wq = rcu_dereference(sk->sk_wq);
430                 if (wq_has_sleeper(wq))
431                         wake_up_interruptible_sync_poll(&wq->wait,
432                                 POLLOUT | POLLWRNORM | POLLWRBAND);
433                 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
434         }
435         rcu_read_unlock();
436 }
437
438 /* When dgram socket disconnects (or changes its peer), we clear its receive
439  * queue of packets arrived from previous peer. First, it allows to do
440  * flow control based only on wmem_alloc; second, sk connected to peer
441  * may receive messages only from that peer. */
442 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
443 {
444         if (!skb_queue_empty(&sk->sk_receive_queue)) {
445                 skb_queue_purge(&sk->sk_receive_queue);
446                 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
447
448                 /* If one link of bidirectional dgram pipe is disconnected,
449                  * we signal error. Messages are lost. Do not make this,
450                  * when peer was not connected to us.
451                  */
452                 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
453                         other->sk_err = ECONNRESET;
454                         other->sk_error_report(other);
455                 }
456         }
457 }
458
459 static void unix_sock_destructor(struct sock *sk)
460 {
461         struct unix_sock *u = unix_sk(sk);
462
463         skb_queue_purge(&sk->sk_receive_queue);
464
465         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
466         WARN_ON(!sk_unhashed(sk));
467         WARN_ON(sk->sk_socket);
468         if (!sock_flag(sk, SOCK_DEAD)) {
469                 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
470                 return;
471         }
472
473         if (u->addr)
474                 unix_release_addr(u->addr);
475
476         atomic_long_dec(&unix_nr_socks);
477         local_bh_disable();
478         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
479         local_bh_enable();
480 #ifdef UNIX_REFCNT_DEBUG
481         printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk,
482                 atomic_long_read(&unix_nr_socks));
483 #endif
484 }
485
486 static void unix_release_sock(struct sock *sk, int embrion)
487 {
488         struct unix_sock *u = unix_sk(sk);
489         struct dentry *dentry;
490         struct vfsmount *mnt;
491         struct sock *skpair;
492         struct sk_buff *skb;
493         int state;
494
495         unix_remove_socket(sk);
496
497         /* Clear state */
498         unix_state_lock(sk);
499         sock_orphan(sk);
500         sk->sk_shutdown = SHUTDOWN_MASK;
501         dentry       = u->dentry;
502         u->dentry    = NULL;
503         mnt          = u->mnt;
504         u->mnt       = NULL;
505         state = sk->sk_state;
506         sk->sk_state = TCP_CLOSE;
507         unix_state_unlock(sk);
508
509         wake_up_interruptible_all(&u->peer_wait);
510
511         skpair = unix_peer(sk);
512
513         if (skpair != NULL) {
514                 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
515                         unix_state_lock(skpair);
516                         /* No more writes */
517                         skpair->sk_shutdown = SHUTDOWN_MASK;
518                         if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
519                                 skpair->sk_err = ECONNRESET;
520                         unix_state_unlock(skpair);
521                         skpair->sk_state_change(skpair);
522                         sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
523                 }
524
525                 unix_dgram_peer_wake_disconnect(sk, skpair);
526                 sock_put(skpair); /* It may now die */
527                 unix_peer(sk) = NULL;
528         }
529
530         /* Try to flush out this socket. Throw out buffers at least */
531
532         while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
533                 if (state == TCP_LISTEN)
534                         unix_release_sock(skb->sk, 1);
535                 /* passed fds are erased in the kfree_skb hook        */
536                 kfree_skb(skb);
537         }
538
539         if (dentry) {
540                 dput(dentry);
541                 mntput(mnt);
542         }
543
544         sock_put(sk);
545
546         /* ---- Socket is dead now and most probably destroyed ---- */
547
548         /*
549          * Fixme: BSD difference: In BSD all sockets connected to use get
550          *        ECONNRESET and we die on the spot. In Linux we behave
551          *        like files and pipes do and wait for the last
552          *        dereference.
553          *
554          * Can't we simply set sock->err?
555          *
556          *        What the above comment does talk about? --ANK(980817)
557          */
558
559         if (unix_tot_inflight)
560                 unix_gc();              /* Garbage collect fds */
561 }
562
563 static void init_peercred(struct sock *sk)
564 {
565         put_pid(sk->sk_peer_pid);
566         if (sk->sk_peer_cred)
567                 put_cred(sk->sk_peer_cred);
568         sk->sk_peer_pid  = get_pid(task_tgid(current));
569         sk->sk_peer_cred = get_current_cred();
570 }
571
572 static void copy_peercred(struct sock *sk, struct sock *peersk)
573 {
574         put_pid(sk->sk_peer_pid);
575         if (sk->sk_peer_cred)
576                 put_cred(sk->sk_peer_cred);
577         sk->sk_peer_pid  = get_pid(peersk->sk_peer_pid);
578         sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
579 }
580
581 static int unix_listen(struct socket *sock, int backlog)
582 {
583         int err;
584         struct sock *sk = sock->sk;
585         struct unix_sock *u = unix_sk(sk);
586         struct pid *old_pid = NULL;
587         const struct cred *old_cred = NULL;
588
589         err = -EOPNOTSUPP;
590         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
591                 goto out;       /* Only stream/seqpacket sockets accept */
592         err = -EINVAL;
593         if (!u->addr)
594                 goto out;       /* No listens on an unbound socket */
595         unix_state_lock(sk);
596         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
597                 goto out_unlock;
598         if (backlog > sk->sk_max_ack_backlog)
599                 wake_up_interruptible_all(&u->peer_wait);
600         sk->sk_max_ack_backlog  = backlog;
601         sk->sk_state            = TCP_LISTEN;
602         /* set credentials so connect can copy them */
603         init_peercred(sk);
604         err = 0;
605
606 out_unlock:
607         unix_state_unlock(sk);
608         put_pid(old_pid);
609         if (old_cred)
610                 put_cred(old_cred);
611 out:
612         return err;
613 }
614
615 static int unix_release(struct socket *);
616 static int unix_bind(struct socket *, struct sockaddr *, int);
617 static int unix_stream_connect(struct socket *, struct sockaddr *,
618                                int addr_len, int flags);
619 static int unix_socketpair(struct socket *, struct socket *);
620 static int unix_accept(struct socket *, struct socket *, int);
621 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
622 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
623 static unsigned int unix_dgram_poll(struct file *, struct socket *,
624                                     poll_table *);
625 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
626 static int unix_shutdown(struct socket *, int);
627 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
628                                struct msghdr *, size_t);
629 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
630                                struct msghdr *, size_t, int);
631 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
632                               struct msghdr *, size_t);
633 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
634                               struct msghdr *, size_t, int);
635 static int unix_dgram_connect(struct socket *, struct sockaddr *,
636                               int, int);
637 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
638                                   struct msghdr *, size_t);
639 static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *,
640                                   struct msghdr *, size_t, int);
641
642 static const struct proto_ops unix_stream_ops = {
643         .family =       PF_UNIX,
644         .owner =        THIS_MODULE,
645         .release =      unix_release,
646         .bind =         unix_bind,
647         .connect =      unix_stream_connect,
648         .socketpair =   unix_socketpair,
649         .accept =       unix_accept,
650         .getname =      unix_getname,
651         .poll =         unix_poll,
652         .ioctl =        unix_ioctl,
653         .listen =       unix_listen,
654         .shutdown =     unix_shutdown,
655         .setsockopt =   sock_no_setsockopt,
656         .getsockopt =   sock_no_getsockopt,
657         .sendmsg =      unix_stream_sendmsg,
658         .recvmsg =      unix_stream_recvmsg,
659         .mmap =         sock_no_mmap,
660         .sendpage =     sock_no_sendpage,
661 };
662
663 static const struct proto_ops unix_dgram_ops = {
664         .family =       PF_UNIX,
665         .owner =        THIS_MODULE,
666         .release =      unix_release,
667         .bind =         unix_bind,
668         .connect =      unix_dgram_connect,
669         .socketpair =   unix_socketpair,
670         .accept =       sock_no_accept,
671         .getname =      unix_getname,
672         .poll =         unix_dgram_poll,
673         .ioctl =        unix_ioctl,
674         .listen =       sock_no_listen,
675         .shutdown =     unix_shutdown,
676         .setsockopt =   sock_no_setsockopt,
677         .getsockopt =   sock_no_getsockopt,
678         .sendmsg =      unix_dgram_sendmsg,
679         .recvmsg =      unix_dgram_recvmsg,
680         .mmap =         sock_no_mmap,
681         .sendpage =     sock_no_sendpage,
682 };
683
684 static const struct proto_ops unix_seqpacket_ops = {
685         .family =       PF_UNIX,
686         .owner =        THIS_MODULE,
687         .release =      unix_release,
688         .bind =         unix_bind,
689         .connect =      unix_stream_connect,
690         .socketpair =   unix_socketpair,
691         .accept =       unix_accept,
692         .getname =      unix_getname,
693         .poll =         unix_dgram_poll,
694         .ioctl =        unix_ioctl,
695         .listen =       unix_listen,
696         .shutdown =     unix_shutdown,
697         .setsockopt =   sock_no_setsockopt,
698         .getsockopt =   sock_no_getsockopt,
699         .sendmsg =      unix_seqpacket_sendmsg,
700         .recvmsg =      unix_seqpacket_recvmsg,
701         .mmap =         sock_no_mmap,
702         .sendpage =     sock_no_sendpage,
703 };
704
705 static struct proto unix_proto = {
706         .name                   = "UNIX",
707         .owner                  = THIS_MODULE,
708         .obj_size               = sizeof(struct unix_sock),
709 };
710
711 /*
712  * AF_UNIX sockets do not interact with hardware, hence they
713  * dont trigger interrupts - so it's safe for them to have
714  * bh-unsafe locking for their sk_receive_queue.lock. Split off
715  * this special lock-class by reinitializing the spinlock key:
716  */
717 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
718
719 static struct sock *unix_create1(struct net *net, struct socket *sock)
720 {
721         struct sock *sk = NULL;
722         struct unix_sock *u;
723
724         atomic_long_inc(&unix_nr_socks);
725         if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
726                 goto out;
727
728         sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
729         if (!sk)
730                 goto out;
731
732         sock_init_data(sock, sk);
733         lockdep_set_class(&sk->sk_receive_queue.lock,
734                                 &af_unix_sk_receive_queue_lock_key);
735
736         sk->sk_write_space      = unix_write_space;
737         sk->sk_max_ack_backlog  = net->unx.sysctl_max_dgram_qlen;
738         sk->sk_destruct         = unix_sock_destructor;
739         u         = unix_sk(sk);
740         u->dentry = NULL;
741         u->mnt    = NULL;
742         spin_lock_init(&u->lock);
743         atomic_long_set(&u->inflight, 0);
744         INIT_LIST_HEAD(&u->link);
745         mutex_init(&u->readlock); /* single task reading lock */
746         init_waitqueue_head(&u->peer_wait);
747         init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
748         unix_insert_socket(unix_sockets_unbound, sk);
749 out:
750         if (sk == NULL)
751                 atomic_long_dec(&unix_nr_socks);
752         else {
753                 local_bh_disable();
754                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
755                 local_bh_enable();
756         }
757         return sk;
758 }
759
760 static int unix_create(struct net *net, struct socket *sock, int protocol,
761                        int kern)
762 {
763         if (protocol && protocol != PF_UNIX)
764                 return -EPROTONOSUPPORT;
765
766         sock->state = SS_UNCONNECTED;
767
768         switch (sock->type) {
769         case SOCK_STREAM:
770                 sock->ops = &unix_stream_ops;
771                 break;
772                 /*
773                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
774                  *      nothing uses it.
775                  */
776         case SOCK_RAW:
777                 sock->type = SOCK_DGRAM;
778         case SOCK_DGRAM:
779                 sock->ops = &unix_dgram_ops;
780                 break;
781         case SOCK_SEQPACKET:
782                 sock->ops = &unix_seqpacket_ops;
783                 break;
784         default:
785                 return -ESOCKTNOSUPPORT;
786         }
787
788         return unix_create1(net, sock) ? 0 : -ENOMEM;
789 }
790
791 static int unix_release(struct socket *sock)
792 {
793         struct sock *sk = sock->sk;
794
795         if (!sk)
796                 return 0;
797
798         unix_release_sock(sk, 0);
799         sock->sk = NULL;
800
801         return 0;
802 }
803
804 static int unix_autobind(struct socket *sock)
805 {
806         struct sock *sk = sock->sk;
807         struct net *net = sock_net(sk);
808         struct unix_sock *u = unix_sk(sk);
809         static u32 ordernum = 1;
810         struct unix_address *addr;
811         int err;
812         unsigned int retries = 0;
813
814         err = mutex_lock_interruptible(&u->readlock);
815         if (err)
816                 return err;
817
818         err = 0;
819         if (u->addr)
820                 goto out;
821
822         err = -ENOMEM;
823         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
824         if (!addr)
825                 goto out;
826
827         addr->name->sun_family = AF_UNIX;
828         atomic_set(&addr->refcnt, 1);
829
830 retry:
831         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
832         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
833
834         spin_lock(&unix_table_lock);
835         ordernum = (ordernum+1)&0xFFFFF;
836
837         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
838                                       addr->hash)) {
839                 spin_unlock(&unix_table_lock);
840                 /*
841                  * __unix_find_socket_byname() may take long time if many names
842                  * are already in use.
843                  */
844                 cond_resched();
845                 /* Give up if all names seems to be in use. */
846                 if (retries++ == 0xFFFFF) {
847                         err = -ENOSPC;
848                         kfree(addr);
849                         goto out;
850                 }
851                 goto retry;
852         }
853         addr->hash ^= sk->sk_type;
854
855         __unix_remove_socket(sk);
856         u->addr = addr;
857         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
858         spin_unlock(&unix_table_lock);
859         err = 0;
860
861 out:    mutex_unlock(&u->readlock);
862         return err;
863 }
864
865 static struct sock *unix_find_other(struct net *net,
866                                     struct sockaddr_un *sunname, int len,
867                                     int type, unsigned hash, int *error)
868 {
869         struct sock *u;
870         struct path path;
871         int err = 0;
872
873         if (sunname->sun_path[0]) {
874                 struct inode *inode;
875                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
876                 if (err)
877                         goto fail;
878                 inode = path.dentry->d_inode;
879                 err = inode_permission(inode, MAY_WRITE);
880                 if (err)
881                         goto put_fail;
882
883                 err = -ECONNREFUSED;
884                 if (!S_ISSOCK(inode->i_mode))
885                         goto put_fail;
886                 u = unix_find_socket_byinode(inode);
887                 if (!u)
888                         goto put_fail;
889
890                 if (u->sk_type == type)
891                         touch_atime(path.mnt, path.dentry);
892
893                 path_put(&path);
894
895                 err = -EPROTOTYPE;
896                 if (u->sk_type != type) {
897                         sock_put(u);
898                         goto fail;
899                 }
900         } else {
901                 err = -ECONNREFUSED;
902                 u = unix_find_socket_byname(net, sunname, len, type, hash);
903                 if (u) {
904                         struct dentry *dentry;
905                         dentry = unix_sk(u)->dentry;
906                         if (dentry)
907                                 touch_atime(unix_sk(u)->mnt, dentry);
908                 } else
909                         goto fail;
910         }
911         return u;
912
913 put_fail:
914         path_put(&path);
915 fail:
916         *error = err;
917         return NULL;
918 }
919
920
921 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
922 {
923         struct sock *sk = sock->sk;
924         struct net *net = sock_net(sk);
925         struct unix_sock *u = unix_sk(sk);
926         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
927         char *sun_path = sunaddr->sun_path;
928         struct dentry *dentry = NULL;
929         struct path path;
930         int err;
931         unsigned hash;
932         struct unix_address *addr;
933         struct hlist_head *list;
934
935         err = -EINVAL;
936         if (sunaddr->sun_family != AF_UNIX)
937                 goto out;
938
939         if (addr_len == sizeof(short)) {
940                 err = unix_autobind(sock);
941                 goto out;
942         }
943
944         err = unix_mkname(sunaddr, addr_len, &hash);
945         if (err < 0)
946                 goto out;
947         addr_len = err;
948
949         err = mutex_lock_interruptible(&u->readlock);
950         if (err)
951                 goto out;
952
953         err = -EINVAL;
954         if (u->addr)
955                 goto out_up;
956
957         err = -ENOMEM;
958         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
959         if (!addr)
960                 goto out_up;
961
962         memcpy(addr->name, sunaddr, addr_len);
963         addr->len = addr_len;
964         addr->hash = hash ^ sk->sk_type;
965         atomic_set(&addr->refcnt, 1);
966
967         if (sun_path[0]) {
968                 unsigned int mode;
969                 err = 0;
970                 /*
971                  * Get the parent directory, calculate the hash for last
972                  * component.
973                  */
974                 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
975                 err = PTR_ERR(dentry);
976                 if (IS_ERR(dentry))
977                         goto out_mknod_parent;
978
979                 /*
980                  * All right, let's create it.
981                  */
982                 mode = S_IFSOCK |
983                        (SOCK_INODE(sock)->i_mode & ~current_umask());
984                 err = mnt_want_write(path.mnt);
985                 if (err)
986                         goto out_mknod_dput;
987                 err = security_path_mknod(&path, dentry, mode, 0);
988                 if (err)
989                         goto out_mknod_drop_write;
990                 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
991 out_mknod_drop_write:
992                 mnt_drop_write(path.mnt);
993                 if (err)
994                         goto out_mknod_dput;
995                 mutex_unlock(&path.dentry->d_inode->i_mutex);
996                 dput(path.dentry);
997                 path.dentry = dentry;
998
999                 addr->hash = UNIX_HASH_SIZE;
1000         }
1001
1002         spin_lock(&unix_table_lock);
1003
1004         if (!sun_path[0]) {
1005                 err = -EADDRINUSE;
1006                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
1007                                               sk->sk_type, hash)) {
1008                         unix_release_addr(addr);
1009                         goto out_unlock;
1010                 }
1011
1012                 list = &unix_socket_table[addr->hash];
1013         } else {
1014                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
1015                 u->dentry = path.dentry;
1016                 u->mnt    = path.mnt;
1017         }
1018
1019         err = 0;
1020         __unix_remove_socket(sk);
1021         u->addr = addr;
1022         __unix_insert_socket(list, sk);
1023
1024 out_unlock:
1025         spin_unlock(&unix_table_lock);
1026 out_up:
1027         mutex_unlock(&u->readlock);
1028 out:
1029         return err;
1030
1031 out_mknod_dput:
1032         dput(dentry);
1033         mutex_unlock(&path.dentry->d_inode->i_mutex);
1034         path_put(&path);
1035 out_mknod_parent:
1036         if (err == -EEXIST)
1037                 err = -EADDRINUSE;
1038         unix_release_addr(addr);
1039         goto out_up;
1040 }
1041
1042 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1043 {
1044         if (unlikely(sk1 == sk2) || !sk2) {
1045                 unix_state_lock(sk1);
1046                 return;
1047         }
1048         if (sk1 < sk2) {
1049                 unix_state_lock(sk1);
1050                 unix_state_lock_nested(sk2);
1051         } else {
1052                 unix_state_lock(sk2);
1053                 unix_state_lock_nested(sk1);
1054         }
1055 }
1056
1057 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1058 {
1059         if (unlikely(sk1 == sk2) || !sk2) {
1060                 unix_state_unlock(sk1);
1061                 return;
1062         }
1063         unix_state_unlock(sk1);
1064         unix_state_unlock(sk2);
1065 }
1066
1067 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1068                               int alen, int flags)
1069 {
1070         struct sock *sk = sock->sk;
1071         struct net *net = sock_net(sk);
1072         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1073         struct sock *other;
1074         unsigned hash;
1075         int err;
1076
1077         if (addr->sa_family != AF_UNSPEC) {
1078                 err = unix_mkname(sunaddr, alen, &hash);
1079                 if (err < 0)
1080                         goto out;
1081                 alen = err;
1082
1083                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1084                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1085                         goto out;
1086
1087 restart:
1088                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1089                 if (!other)
1090                         goto out;
1091
1092                 unix_state_double_lock(sk, other);
1093
1094                 /* Apparently VFS overslept socket death. Retry. */
1095                 if (sock_flag(other, SOCK_DEAD)) {
1096                         unix_state_double_unlock(sk, other);
1097                         sock_put(other);
1098                         goto restart;
1099                 }
1100
1101                 err = -EPERM;
1102                 if (!unix_may_send(sk, other))
1103                         goto out_unlock;
1104
1105                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1106                 if (err)
1107                         goto out_unlock;
1108
1109         } else {
1110                 /*
1111                  *      1003.1g breaking connected state with AF_UNSPEC
1112                  */
1113                 other = NULL;
1114                 unix_state_double_lock(sk, other);
1115         }
1116
1117         /*
1118          * If it was connected, reconnect.
1119          */
1120         if (unix_peer(sk)) {
1121                 struct sock *old_peer = unix_peer(sk);
1122                 unix_peer(sk) = other;
1123                 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1124
1125                 unix_state_double_unlock(sk, other);
1126
1127                 if (other != old_peer)
1128                         unix_dgram_disconnected(sk, old_peer);
1129                 sock_put(old_peer);
1130         } else {
1131                 unix_peer(sk) = other;
1132                 unix_state_double_unlock(sk, other);
1133         }
1134         return 0;
1135
1136 out_unlock:
1137         unix_state_double_unlock(sk, other);
1138         sock_put(other);
1139 out:
1140         return err;
1141 }
1142
1143 static long unix_wait_for_peer(struct sock *other, long timeo)
1144 {
1145         struct unix_sock *u = unix_sk(other);
1146         int sched;
1147         DEFINE_WAIT(wait);
1148
1149         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1150
1151         sched = !sock_flag(other, SOCK_DEAD) &&
1152                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1153                 unix_recvq_full(other);
1154
1155         unix_state_unlock(other);
1156
1157         if (sched)
1158                 timeo = schedule_timeout(timeo);
1159
1160         finish_wait(&u->peer_wait, &wait);
1161         return timeo;
1162 }
1163
1164 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1165                                int addr_len, int flags)
1166 {
1167         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1168         struct sock *sk = sock->sk;
1169         struct net *net = sock_net(sk);
1170         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1171         struct sock *newsk = NULL;
1172         struct sock *other = NULL;
1173         struct sk_buff *skb = NULL;
1174         unsigned hash;
1175         int st;
1176         int err;
1177         long timeo;
1178
1179         err = unix_mkname(sunaddr, addr_len, &hash);
1180         if (err < 0)
1181                 goto out;
1182         addr_len = err;
1183
1184         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1185             (err = unix_autobind(sock)) != 0)
1186                 goto out;
1187
1188         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1189
1190         /* First of all allocate resources.
1191            If we will make it after state is locked,
1192            we will have to recheck all again in any case.
1193          */
1194
1195         err = -ENOMEM;
1196
1197         /* create new sock for complete connection */
1198         newsk = unix_create1(sock_net(sk), NULL);
1199         if (newsk == NULL)
1200                 goto out;
1201
1202         /* Allocate skb for sending to listening sock */
1203         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1204         if (skb == NULL)
1205                 goto out;
1206
1207 restart:
1208         /*  Find listening sock. */
1209         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1210         if (!other)
1211                 goto out;
1212
1213         /* Latch state of peer */
1214         unix_state_lock(other);
1215
1216         /* Apparently VFS overslept socket death. Retry. */
1217         if (sock_flag(other, SOCK_DEAD)) {
1218                 unix_state_unlock(other);
1219                 sock_put(other);
1220                 goto restart;
1221         }
1222
1223         err = -ECONNREFUSED;
1224         if (other->sk_state != TCP_LISTEN)
1225                 goto out_unlock;
1226         if (other->sk_shutdown & RCV_SHUTDOWN)
1227                 goto out_unlock;
1228
1229         if (unix_recvq_full(other)) {
1230                 err = -EAGAIN;
1231                 if (!timeo)
1232                         goto out_unlock;
1233
1234                 timeo = unix_wait_for_peer(other, timeo);
1235
1236                 err = sock_intr_errno(timeo);
1237                 if (signal_pending(current))
1238                         goto out;
1239                 sock_put(other);
1240                 goto restart;
1241         }
1242
1243         /* Latch our state.
1244
1245            It is tricky place. We need to grab our state lock and cannot
1246            drop lock on peer. It is dangerous because deadlock is
1247            possible. Connect to self case and simultaneous
1248            attempt to connect are eliminated by checking socket
1249            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1250            check this before attempt to grab lock.
1251
1252            Well, and we have to recheck the state after socket locked.
1253          */
1254         st = sk->sk_state;
1255
1256         switch (st) {
1257         case TCP_CLOSE:
1258                 /* This is ok... continue with connect */
1259                 break;
1260         case TCP_ESTABLISHED:
1261                 /* Socket is already connected */
1262                 err = -EISCONN;
1263                 goto out_unlock;
1264         default:
1265                 err = -EINVAL;
1266                 goto out_unlock;
1267         }
1268
1269         unix_state_lock_nested(sk);
1270
1271         if (sk->sk_state != st) {
1272                 unix_state_unlock(sk);
1273                 unix_state_unlock(other);
1274                 sock_put(other);
1275                 goto restart;
1276         }
1277
1278         err = security_unix_stream_connect(sk, other, newsk);
1279         if (err) {
1280                 unix_state_unlock(sk);
1281                 goto out_unlock;
1282         }
1283
1284         /* The way is open! Fastly set all the necessary fields... */
1285
1286         sock_hold(sk);
1287         unix_peer(newsk)        = sk;
1288         newsk->sk_state         = TCP_ESTABLISHED;
1289         newsk->sk_type          = sk->sk_type;
1290         init_peercred(newsk);
1291         newu = unix_sk(newsk);
1292         RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1293         otheru = unix_sk(other);
1294
1295         /* copy address information from listening to new sock*/
1296         if (otheru->addr) {
1297                 atomic_inc(&otheru->addr->refcnt);
1298                 newu->addr = otheru->addr;
1299         }
1300         if (otheru->dentry) {
1301                 newu->dentry    = dget(otheru->dentry);
1302                 newu->mnt       = mntget(otheru->mnt);
1303         }
1304
1305         /* Set credentials */
1306         copy_peercred(sk, other);
1307
1308         sock->state     = SS_CONNECTED;
1309         sk->sk_state    = TCP_ESTABLISHED;
1310         sock_hold(newsk);
1311
1312         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1313         unix_peer(sk)   = newsk;
1314
1315         unix_state_unlock(sk);
1316
1317         /* take ten and and send info to listening sock */
1318         spin_lock(&other->sk_receive_queue.lock);
1319         __skb_queue_tail(&other->sk_receive_queue, skb);
1320         spin_unlock(&other->sk_receive_queue.lock);
1321         unix_state_unlock(other);
1322         other->sk_data_ready(other, 0);
1323         sock_put(other);
1324         return 0;
1325
1326 out_unlock:
1327         if (other)
1328                 unix_state_unlock(other);
1329
1330 out:
1331         kfree_skb(skb);
1332         if (newsk)
1333                 unix_release_sock(newsk, 0);
1334         if (other)
1335                 sock_put(other);
1336         return err;
1337 }
1338
1339 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1340 {
1341         struct sock *ska = socka->sk, *skb = sockb->sk;
1342
1343         /* Join our sockets back to back */
1344         sock_hold(ska);
1345         sock_hold(skb);
1346         unix_peer(ska) = skb;
1347         unix_peer(skb) = ska;
1348         init_peercred(ska);
1349         init_peercred(skb);
1350
1351         if (ska->sk_type != SOCK_DGRAM) {
1352                 ska->sk_state = TCP_ESTABLISHED;
1353                 skb->sk_state = TCP_ESTABLISHED;
1354                 socka->state  = SS_CONNECTED;
1355                 sockb->state  = SS_CONNECTED;
1356         }
1357         return 0;
1358 }
1359
1360 static void unix_sock_inherit_flags(const struct socket *old,
1361                                     struct socket *new)
1362 {
1363         if (test_bit(SOCK_PASSCRED, &old->flags))
1364                 set_bit(SOCK_PASSCRED, &new->flags);
1365         if (test_bit(SOCK_PASSSEC, &old->flags))
1366                 set_bit(SOCK_PASSSEC, &new->flags);
1367 }
1368
1369 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1370 {
1371         struct sock *sk = sock->sk;
1372         struct sock *tsk;
1373         struct sk_buff *skb;
1374         int err;
1375
1376         err = -EOPNOTSUPP;
1377         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1378                 goto out;
1379
1380         err = -EINVAL;
1381         if (sk->sk_state != TCP_LISTEN)
1382                 goto out;
1383
1384         /* If socket state is TCP_LISTEN it cannot change (for now...),
1385          * so that no locks are necessary.
1386          */
1387
1388         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1389         if (!skb) {
1390                 /* This means receive shutdown. */
1391                 if (err == 0)
1392                         err = -EINVAL;
1393                 goto out;
1394         }
1395
1396         tsk = skb->sk;
1397         skb_free_datagram(sk, skb);
1398         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1399
1400         /* attach accepted sock to socket */
1401         unix_state_lock(tsk);
1402         newsock->state = SS_CONNECTED;
1403         unix_sock_inherit_flags(sock, newsock);
1404         sock_graft(tsk, newsock);
1405         unix_state_unlock(tsk);
1406         return 0;
1407
1408 out:
1409         return err;
1410 }
1411
1412
1413 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1414 {
1415         struct sock *sk = sock->sk;
1416         struct unix_sock *u;
1417         DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1418         int err = 0;
1419
1420         if (peer) {
1421                 sk = unix_peer_get(sk);
1422
1423                 err = -ENOTCONN;
1424                 if (!sk)
1425                         goto out;
1426                 err = 0;
1427         } else {
1428                 sock_hold(sk);
1429         }
1430
1431         u = unix_sk(sk);
1432         unix_state_lock(sk);
1433         if (!u->addr) {
1434                 sunaddr->sun_family = AF_UNIX;
1435                 sunaddr->sun_path[0] = 0;
1436                 *uaddr_len = sizeof(short);
1437         } else {
1438                 struct unix_address *addr = u->addr;
1439
1440                 *uaddr_len = addr->len;
1441                 memcpy(sunaddr, addr->name, *uaddr_len);
1442         }
1443         unix_state_unlock(sk);
1444         sock_put(sk);
1445 out:
1446         return err;
1447 }
1448
1449 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1450 {
1451         int i;
1452
1453         scm->fp = UNIXCB(skb).fp;
1454         UNIXCB(skb).fp = NULL;
1455
1456         for (i = scm->fp->count-1; i >= 0; i--)
1457                 unix_notinflight(scm->fp->user, scm->fp->fp[i]);
1458 }
1459
1460 static void unix_destruct_scm(struct sk_buff *skb)
1461 {
1462         struct scm_cookie scm;
1463         memset(&scm, 0, sizeof(scm));
1464         scm.pid  = UNIXCB(skb).pid;
1465         scm.cred = UNIXCB(skb).cred;
1466         if (UNIXCB(skb).fp)
1467                 unix_detach_fds(&scm, skb);
1468
1469         /* Alas, it calls VFS */
1470         /* So fscking what? fput() had been SMP-safe since the last Summer */
1471         scm_destroy(&scm);
1472         sock_wfree(skb);
1473 }
1474
1475 /*
1476  * The "user->unix_inflight" variable is protected by the garbage
1477  * collection lock, and we just read it locklessly here. If you go
1478  * over the limit, there might be a tiny race in actually noticing
1479  * it across threads. Tough.
1480  */
1481 static inline bool too_many_unix_fds(struct task_struct *p)
1482 {
1483         struct user_struct *user = current_user();
1484
1485         if (unlikely(user->unix_inflight > task_rlimit(p, RLIMIT_NOFILE)))
1486                 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
1487         return false;
1488 }
1489
1490 #define MAX_RECURSION_LEVEL 4
1491
1492 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1493 {
1494         int i;
1495         unsigned char max_level = 0;
1496         int unix_sock_count = 0;
1497
1498         if (too_many_unix_fds(current))
1499                 return -ETOOMANYREFS;
1500
1501         for (i = scm->fp->count - 1; i >= 0; i--) {
1502                 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1503
1504                 if (sk) {
1505                         unix_sock_count++;
1506                         max_level = max(max_level,
1507                                         unix_sk(sk)->recursion_level);
1508                 }
1509         }
1510         if (unlikely(max_level > MAX_RECURSION_LEVEL))
1511                 return -ETOOMANYREFS;
1512
1513         /*
1514          * Need to duplicate file references for the sake of garbage
1515          * collection.  Otherwise a socket in the fps might become a
1516          * candidate for GC while the skb is not yet queued.
1517          */
1518         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1519         if (!UNIXCB(skb).fp)
1520                 return -ENOMEM;
1521
1522         for (i = scm->fp->count - 1; i >= 0; i--)
1523                 unix_inflight(scm->fp->user, scm->fp->fp[i]);
1524         return max_level;
1525 }
1526
1527 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1528 {
1529         int err = 0;
1530
1531         UNIXCB(skb).pid  = get_pid(scm->pid);
1532         if (scm->cred)
1533                 UNIXCB(skb).cred = get_cred(scm->cred);
1534         UNIXCB(skb).fp = NULL;
1535         if (scm->fp && send_fds)
1536                 err = unix_attach_fds(scm, skb);
1537
1538         skb->destructor = unix_destruct_scm;
1539         return err;
1540 }
1541
1542 /*
1543  * Some apps rely on write() giving SCM_CREDENTIALS
1544  * We include credentials if source or destination socket
1545  * asserted SOCK_PASSCRED.
1546  */
1547 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1548                             const struct sock *other)
1549 {
1550         if (UNIXCB(skb).cred)
1551                 return;
1552         if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1553             !other->sk_socket ||
1554             test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1555                 UNIXCB(skb).pid  = get_pid(task_tgid(current));
1556                 UNIXCB(skb).cred = get_current_cred();
1557         }
1558 }
1559
1560 /*
1561  *      Send AF_UNIX data.
1562  */
1563
1564 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1565                               struct msghdr *msg, size_t len)
1566 {
1567         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1568         struct sock *sk = sock->sk;
1569         struct net *net = sock_net(sk);
1570         struct unix_sock *u = unix_sk(sk);
1571         struct sockaddr_un *sunaddr = msg->msg_name;
1572         struct sock *other = NULL;
1573         int namelen = 0; /* fake GCC */
1574         int err;
1575         unsigned hash;
1576         struct sk_buff *skb;
1577         long timeo;
1578         struct scm_cookie tmp_scm;
1579         int max_level;
1580         int sk_locked;
1581
1582         if (NULL == siocb->scm)
1583                 siocb->scm = &tmp_scm;
1584         wait_for_unix_gc();
1585         err = scm_send(sock, msg, siocb->scm, false);
1586         if (err < 0)
1587                 return err;
1588
1589         err = -EOPNOTSUPP;
1590         if (msg->msg_flags&MSG_OOB)
1591                 goto out;
1592
1593         if (msg->msg_namelen) {
1594                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1595                 if (err < 0)
1596                         goto out;
1597                 namelen = err;
1598         } else {
1599                 sunaddr = NULL;
1600                 err = -ENOTCONN;
1601                 other = unix_peer_get(sk);
1602                 if (!other)
1603                         goto out;
1604         }
1605
1606         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1607             && (err = unix_autobind(sock)) != 0)
1608                 goto out;
1609
1610         err = -EMSGSIZE;
1611         if (len > sk->sk_sndbuf - 32)
1612                 goto out;
1613
1614         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1615         if (skb == NULL)
1616                 goto out;
1617
1618         err = unix_scm_to_skb(siocb->scm, skb, true);
1619         if (err < 0)
1620                 goto out_free;
1621         max_level = err + 1;
1622         unix_get_secdata(siocb->scm, skb);
1623
1624         skb_reset_transport_header(skb);
1625         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1626         if (err)
1627                 goto out_free;
1628
1629         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1630
1631 restart:
1632         if (!other) {
1633                 err = -ECONNRESET;
1634                 if (sunaddr == NULL)
1635                         goto out_free;
1636
1637                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1638                                         hash, &err);
1639                 if (other == NULL)
1640                         goto out_free;
1641         }
1642
1643         if (sk_filter(other, skb) < 0) {
1644                 /* Toss the packet but do not return any error to the sender */
1645                 err = len;
1646                 goto out_free;
1647         }
1648
1649         sk_locked = 0;
1650         unix_state_lock(other);
1651 restart_locked:
1652         err = -EPERM;
1653         if (!unix_may_send(sk, other))
1654                 goto out_unlock;
1655
1656         if (unlikely(sock_flag(other, SOCK_DEAD))) {
1657                 /*
1658                  *      Check with 1003.1g - what should
1659                  *      datagram error
1660                  */
1661                 unix_state_unlock(other);
1662                 sock_put(other);
1663
1664                 if (!sk_locked)
1665                         unix_state_lock(sk);
1666
1667                 err = 0;
1668                 if (unix_peer(sk) == other) {
1669                         unix_peer(sk) = NULL;
1670                         unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1671
1672                         unix_state_unlock(sk);
1673
1674                         unix_dgram_disconnected(sk, other);
1675                         sock_put(other);
1676                         err = -ECONNREFUSED;
1677                 } else {
1678                         unix_state_unlock(sk);
1679                 }
1680
1681                 other = NULL;
1682                 if (err)
1683                         goto out_free;
1684                 goto restart;
1685         }
1686
1687         err = -EPIPE;
1688         if (other->sk_shutdown & RCV_SHUTDOWN)
1689                 goto out_unlock;
1690
1691         if (sk->sk_type != SOCK_SEQPACKET) {
1692                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1693                 if (err)
1694                         goto out_unlock;
1695         }
1696
1697         if (unlikely(unix_peer(other) != sk && unix_recvq_full(other))) {
1698                 if (timeo) {
1699                         timeo = unix_wait_for_peer(other, timeo);
1700
1701                         err = sock_intr_errno(timeo);
1702                         if (signal_pending(current))
1703                                 goto out_free;
1704
1705                         goto restart;
1706                 }
1707
1708                 if (!sk_locked) {
1709                         unix_state_unlock(other);
1710                         unix_state_double_lock(sk, other);
1711                 }
1712
1713                 if (unix_peer(sk) != other ||
1714                     unix_dgram_peer_wake_me(sk, other)) {
1715                         err = -EAGAIN;
1716                         sk_locked = 1;
1717                         goto out_unlock;
1718                 }
1719
1720                 if (!sk_locked) {
1721                         sk_locked = 1;
1722                         goto restart_locked;
1723                 }
1724         }
1725
1726         if (unlikely(sk_locked))
1727                 unix_state_unlock(sk);
1728
1729         if (sock_flag(other, SOCK_RCVTSTAMP))
1730                 __net_timestamp(skb);
1731         maybe_add_creds(skb, sock, other);
1732         skb_queue_tail(&other->sk_receive_queue, skb);
1733         if (max_level > unix_sk(other)->recursion_level)
1734                 unix_sk(other)->recursion_level = max_level;
1735         unix_state_unlock(other);
1736         other->sk_data_ready(other, len);
1737         sock_put(other);
1738         scm_destroy(siocb->scm);
1739         return len;
1740
1741 out_unlock:
1742         if (sk_locked)
1743                 unix_state_unlock(sk);
1744         unix_state_unlock(other);
1745 out_free:
1746         kfree_skb(skb);
1747 out:
1748         if (other)
1749                 sock_put(other);
1750         scm_destroy(siocb->scm);
1751         return err;
1752 }
1753
1754
1755 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1756                                struct msghdr *msg, size_t len)
1757 {
1758         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1759         struct sock *sk = sock->sk;
1760         struct sock *other = NULL;
1761         int err, size;
1762         struct sk_buff *skb;
1763         int sent = 0;
1764         struct scm_cookie tmp_scm;
1765         bool fds_sent = false;
1766         int max_level;
1767
1768         if (NULL == siocb->scm)
1769                 siocb->scm = &tmp_scm;
1770         wait_for_unix_gc();
1771         err = scm_send(sock, msg, siocb->scm, false);
1772         if (err < 0)
1773                 return err;
1774
1775         err = -EOPNOTSUPP;
1776         if (msg->msg_flags&MSG_OOB)
1777                 goto out_err;
1778
1779         if (msg->msg_namelen) {
1780                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1781                 goto out_err;
1782         } else {
1783                 err = -ENOTCONN;
1784                 other = unix_peer(sk);
1785                 if (!other)
1786                         goto out_err;
1787         }
1788
1789         if (sk->sk_shutdown & SEND_SHUTDOWN)
1790                 goto pipe_err;
1791
1792         while (sent < len) {
1793                 /*
1794                  *      Optimisation for the fact that under 0.01% of X
1795                  *      messages typically need breaking up.
1796                  */
1797
1798                 size = len-sent;
1799
1800                 /* Keep two messages in the pipe so it schedules better */
1801                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1802                         size = (sk->sk_sndbuf >> 1) - 64;
1803
1804                 if (size > SKB_MAX_ALLOC)
1805                         size = SKB_MAX_ALLOC;
1806
1807                 /*
1808                  *      Grab a buffer
1809                  */
1810
1811                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1812                                           &err);
1813
1814                 if (skb == NULL)
1815                         goto out_err;
1816
1817                 /*
1818                  *      If you pass two values to the sock_alloc_send_skb
1819                  *      it tries to grab the large buffer with GFP_NOFS
1820                  *      (which can fail easily), and if it fails grab the
1821                  *      fallback size buffer which is under a page and will
1822                  *      succeed. [Alan]
1823                  */
1824                 size = min_t(int, size, skb_tailroom(skb));
1825
1826
1827                 /* Only send the fds in the first buffer */
1828                 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1829                 if (err < 0) {
1830                         kfree_skb(skb);
1831                         goto out_err;
1832                 }
1833                 max_level = err + 1;
1834                 fds_sent = true;
1835
1836                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1837                 if (err) {
1838                         kfree_skb(skb);
1839                         goto out_err;
1840                 }
1841
1842                 unix_state_lock(other);
1843
1844                 if (sock_flag(other, SOCK_DEAD) ||
1845                     (other->sk_shutdown & RCV_SHUTDOWN))
1846                         goto pipe_err_free;
1847
1848                 maybe_add_creds(skb, sock, other);
1849                 skb_queue_tail(&other->sk_receive_queue, skb);
1850                 if (max_level > unix_sk(other)->recursion_level)
1851                         unix_sk(other)->recursion_level = max_level;
1852                 unix_state_unlock(other);
1853                 other->sk_data_ready(other, size);
1854                 sent += size;
1855         }
1856
1857         scm_destroy(siocb->scm);
1858         siocb->scm = NULL;
1859
1860         return sent;
1861
1862 pipe_err_free:
1863         unix_state_unlock(other);
1864         kfree_skb(skb);
1865 pipe_err:
1866         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1867                 send_sig(SIGPIPE, current, 0);
1868         err = -EPIPE;
1869 out_err:
1870         scm_destroy(siocb->scm);
1871         siocb->scm = NULL;
1872         return sent ? : err;
1873 }
1874
1875 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1876                                   struct msghdr *msg, size_t len)
1877 {
1878         int err;
1879         struct sock *sk = sock->sk;
1880
1881         err = sock_error(sk);
1882         if (err)
1883                 return err;
1884
1885         if (sk->sk_state != TCP_ESTABLISHED)
1886                 return -ENOTCONN;
1887
1888         if (msg->msg_namelen)
1889                 msg->msg_namelen = 0;
1890
1891         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1892 }
1893
1894 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1895                               struct msghdr *msg, size_t size,
1896                               int flags)
1897 {
1898         struct sock *sk = sock->sk;
1899
1900         if (sk->sk_state != TCP_ESTABLISHED)
1901                 return -ENOTCONN;
1902
1903         return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1904 }
1905
1906 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1907 {
1908         struct unix_sock *u = unix_sk(sk);
1909
1910         if (u->addr) {
1911                 msg->msg_namelen = u->addr->len;
1912                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1913         }
1914 }
1915
1916 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1917                               struct msghdr *msg, size_t size,
1918                               int flags)
1919 {
1920         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1921         struct scm_cookie tmp_scm;
1922         struct sock *sk = sock->sk;
1923         struct unix_sock *u = unix_sk(sk);
1924         int noblock = flags & MSG_DONTWAIT;
1925         struct sk_buff *skb;
1926         int err;
1927
1928         err = -EOPNOTSUPP;
1929         if (flags&MSG_OOB)
1930                 goto out;
1931
1932         err = mutex_lock_interruptible(&u->readlock);
1933         if (unlikely(err)) {
1934                 /* recvmsg() in non blocking mode is supposed to return -EAGAIN
1935                  * sk_rcvtimeo is not honored by mutex_lock_interruptible()
1936                  */
1937                 err = noblock ? -EAGAIN : -ERESTARTSYS;
1938                 goto out;
1939         }
1940
1941         skb = skb_recv_datagram(sk, flags, noblock, &err);
1942         if (!skb) {
1943                 unix_state_lock(sk);
1944                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1945                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1946                     (sk->sk_shutdown & RCV_SHUTDOWN))
1947                         err = 0;
1948                 unix_state_unlock(sk);
1949                 goto out_unlock;
1950         }
1951
1952         wake_up_interruptible_sync_poll(&u->peer_wait,
1953                                         POLLOUT | POLLWRNORM | POLLWRBAND);
1954
1955         if (msg->msg_name)
1956                 unix_copy_addr(msg, skb->sk);
1957
1958         if (size > skb->len)
1959                 size = skb->len;
1960         else if (size < skb->len)
1961                 msg->msg_flags |= MSG_TRUNC;
1962
1963         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1964         if (err)
1965                 goto out_free;
1966
1967         if (sock_flag(sk, SOCK_RCVTSTAMP))
1968                 __sock_recv_timestamp(msg, sk, skb);
1969
1970         if (!siocb->scm) {
1971                 siocb->scm = &tmp_scm;
1972                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1973         }
1974         scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1975         unix_set_secdata(siocb->scm, skb);
1976
1977         if (!(flags & MSG_PEEK)) {
1978                 if (UNIXCB(skb).fp)
1979                         unix_detach_fds(siocb->scm, skb);
1980         } else {
1981                 /* It is questionable: on PEEK we could:
1982                    - do not return fds - good, but too simple 8)
1983                    - return fds, and do not return them on read (old strategy,
1984                      apparently wrong)
1985                    - clone fds (I chose it for now, it is the most universal
1986                      solution)
1987
1988                    POSIX 1003.1g does not actually define this clearly
1989                    at all. POSIX 1003.1g doesn't define a lot of things
1990                    clearly however!
1991
1992                 */
1993                 if (UNIXCB(skb).fp)
1994                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1995         }
1996         err = size;
1997
1998         scm_recv(sock, msg, siocb->scm, flags);
1999
2000 out_free:
2001         skb_free_datagram(sk, skb);
2002 out_unlock:
2003         mutex_unlock(&u->readlock);
2004 out:
2005         return err;
2006 }
2007
2008 /*
2009  *      Sleep until data has arrive. But check for races..
2010  */
2011
2012 static long unix_stream_data_wait(struct sock *sk, long timeo)
2013 {
2014         DEFINE_WAIT(wait);
2015
2016         unix_state_lock(sk);
2017
2018         for (;;) {
2019                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2020
2021                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
2022                     sk->sk_err ||
2023                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
2024                     signal_pending(current) ||
2025                     !timeo)
2026                         break;
2027
2028                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2029                 unix_state_unlock(sk);
2030                 timeo = schedule_timeout(timeo);
2031                 unix_state_lock(sk);
2032
2033                 if (sock_flag(sk, SOCK_DEAD))
2034                         break;
2035
2036                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2037         }
2038
2039         finish_wait(sk_sleep(sk), &wait);
2040         unix_state_unlock(sk);
2041         return timeo;
2042 }
2043
2044
2045
2046 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
2047                                struct msghdr *msg, size_t size,
2048                                int flags)
2049 {
2050         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
2051         struct scm_cookie tmp_scm;
2052         struct sock *sk = sock->sk;
2053         struct unix_sock *u = unix_sk(sk);
2054         struct sockaddr_un *sunaddr = msg->msg_name;
2055         int copied = 0;
2056         int noblock = flags & MSG_DONTWAIT;
2057         int check_creds = 0;
2058         int target;
2059         int err = 0;
2060         long timeo;
2061
2062         if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2063                 err = -EINVAL;
2064                 goto out;
2065         }
2066
2067         if (unlikely(flags & MSG_OOB)) {
2068                 err = -EOPNOTSUPP;
2069                 goto out;
2070         }
2071
2072         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
2073         timeo = sock_rcvtimeo(sk, noblock);
2074
2075         /* Lock the socket to prevent queue disordering
2076          * while sleeps in memcpy_tomsg
2077          */
2078
2079         if (!siocb->scm) {
2080                 siocb->scm = &tmp_scm;
2081                 memset(&tmp_scm, 0, sizeof(tmp_scm));
2082         }
2083
2084         mutex_lock(&u->readlock);
2085
2086         do {
2087                 int chunk;
2088                 struct sk_buff *skb;
2089
2090                 unix_state_lock(sk);
2091                 if (sock_flag(sk, SOCK_DEAD)) {
2092                         err = -ECONNRESET;
2093                         goto unlock;
2094                 }
2095                 skb = skb_peek(&sk->sk_receive_queue);
2096                 if (skb == NULL) {
2097                         unix_sk(sk)->recursion_level = 0;
2098                         if (copied >= target)
2099                                 goto unlock;
2100
2101                         /*
2102                          *      POSIX 1003.1g mandates this order.
2103                          */
2104
2105                         err = sock_error(sk);
2106                         if (err)
2107                                 goto unlock;
2108                         if (sk->sk_shutdown & RCV_SHUTDOWN)
2109                                 goto unlock;
2110
2111                         unix_state_unlock(sk);
2112                         if (!timeo) {
2113                                 err = -EAGAIN;
2114                                 break;
2115                         }
2116
2117                         mutex_unlock(&u->readlock);
2118
2119                         timeo = unix_stream_data_wait(sk, timeo);
2120
2121                         if (signal_pending(current)) {
2122                                 err = sock_intr_errno(timeo);
2123                                 scm_destroy(siocb->scm);
2124                                 goto out;
2125                         }
2126
2127                         mutex_lock(&u->readlock);
2128                         continue;
2129  unlock:
2130                         unix_state_unlock(sk);
2131                         break;
2132                 }
2133                 unix_state_unlock(sk);
2134
2135                 if (check_creds) {
2136                         /* Never glue messages from different writers */
2137                         if ((UNIXCB(skb).pid  != siocb->scm->pid) ||
2138                             (UNIXCB(skb).cred != siocb->scm->cred))
2139                                 break;
2140                 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2141                         /* Copy credentials */
2142                         scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
2143                         check_creds = 1;
2144                 }
2145
2146                 /* Copy address just once */
2147                 if (sunaddr) {
2148                         unix_copy_addr(msg, skb->sk);
2149                         sunaddr = NULL;
2150                 }
2151
2152                 chunk = min_t(unsigned int, skb->len, size);
2153                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
2154                         if (copied == 0)
2155                                 copied = -EFAULT;
2156                         break;
2157                 }
2158                 copied += chunk;
2159                 size -= chunk;
2160
2161                 /* Mark read part of skb as used */
2162                 if (!(flags & MSG_PEEK)) {
2163                         skb_pull(skb, chunk);
2164
2165                         if (UNIXCB(skb).fp)
2166                                 unix_detach_fds(siocb->scm, skb);
2167
2168                         if (skb->len)
2169                                 break;
2170
2171                         skb_unlink(skb, &sk->sk_receive_queue);
2172                         consume_skb(skb);
2173
2174                         if (siocb->scm->fp)
2175                                 break;
2176                 } else {
2177                         /* It is questionable, see note in unix_dgram_recvmsg.
2178                          */
2179                         if (UNIXCB(skb).fp)
2180                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2181
2182                         break;
2183                 }
2184         } while (size);
2185
2186         mutex_unlock(&u->readlock);
2187         scm_recv(sock, msg, siocb->scm, flags);
2188 out:
2189         return copied ? : err;
2190 }
2191
2192 static int unix_shutdown(struct socket *sock, int mode)
2193 {
2194         struct sock *sk = sock->sk;
2195         struct sock *other;
2196
2197         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
2198
2199         if (!mode)
2200                 return 0;
2201
2202         unix_state_lock(sk);
2203         sk->sk_shutdown |= mode;
2204         other = unix_peer(sk);
2205         if (other)
2206                 sock_hold(other);
2207         unix_state_unlock(sk);
2208         sk->sk_state_change(sk);
2209
2210         if (other &&
2211                 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2212
2213                 int peer_mode = 0;
2214
2215                 if (mode&RCV_SHUTDOWN)
2216                         peer_mode |= SEND_SHUTDOWN;
2217                 if (mode&SEND_SHUTDOWN)
2218                         peer_mode |= RCV_SHUTDOWN;
2219                 unix_state_lock(other);
2220                 other->sk_shutdown |= peer_mode;
2221                 unix_state_unlock(other);
2222                 other->sk_state_change(other);
2223                 if (peer_mode == SHUTDOWN_MASK)
2224                         sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2225                 else if (peer_mode & RCV_SHUTDOWN)
2226                         sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2227         }
2228         if (other)
2229                 sock_put(other);
2230
2231         return 0;
2232 }
2233
2234 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2235 {
2236         struct sock *sk = sock->sk;
2237         long amount = 0;
2238         int err;
2239
2240         switch (cmd) {
2241         case SIOCOUTQ:
2242                 amount = sk_wmem_alloc_get(sk);
2243                 err = put_user(amount, (int __user *)arg);
2244                 break;
2245         case SIOCINQ:
2246                 {
2247                         struct sk_buff *skb;
2248
2249                         if (sk->sk_state == TCP_LISTEN) {
2250                                 err = -EINVAL;
2251                                 break;
2252                         }
2253
2254                         spin_lock(&sk->sk_receive_queue.lock);
2255                         if (sk->sk_type == SOCK_STREAM ||
2256                             sk->sk_type == SOCK_SEQPACKET) {
2257                                 skb_queue_walk(&sk->sk_receive_queue, skb)
2258                                         amount += skb->len;
2259                         } else {
2260                                 skb = skb_peek(&sk->sk_receive_queue);
2261                                 if (skb)
2262                                         amount = skb->len;
2263                         }
2264                         spin_unlock(&sk->sk_receive_queue.lock);
2265                         err = put_user(amount, (int __user *)arg);
2266                         break;
2267                 }
2268
2269         default:
2270                 err = -ENOIOCTLCMD;
2271                 break;
2272         }
2273         return err;
2274 }
2275
2276 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2277 {
2278         struct sock *sk = sock->sk;
2279         unsigned int mask;
2280
2281         sock_poll_wait(file, sk_sleep(sk), wait);
2282         mask = 0;
2283
2284         /* exceptional events? */
2285         if (sk->sk_err)
2286                 mask |= POLLERR;
2287         if (sk->sk_shutdown == SHUTDOWN_MASK)
2288                 mask |= POLLHUP;
2289         if (sk->sk_shutdown & RCV_SHUTDOWN)
2290                 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2291
2292         /* readable? */
2293         if (!skb_queue_empty(&sk->sk_receive_queue))
2294                 mask |= POLLIN | POLLRDNORM;
2295
2296         /* Connection-based need to check for termination and startup */
2297         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2298             sk->sk_state == TCP_CLOSE)
2299                 mask |= POLLHUP;
2300
2301         /*
2302          * we set writable also when the other side has shut down the
2303          * connection. This prevents stuck sockets.
2304          */
2305         if (unix_writable(sk))
2306                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2307
2308         return mask;
2309 }
2310
2311 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2312                                     poll_table *wait)
2313 {
2314         struct sock *sk = sock->sk, *other;
2315         unsigned int mask, writable;
2316
2317         sock_poll_wait(file, sk_sleep(sk), wait);
2318         mask = 0;
2319
2320         /* exceptional events? */
2321         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2322                 mask |= POLLERR;
2323         if (sk->sk_shutdown & RCV_SHUTDOWN)
2324                 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2325         if (sk->sk_shutdown == SHUTDOWN_MASK)
2326                 mask |= POLLHUP;
2327
2328         /* readable? */
2329         if (!skb_queue_empty(&sk->sk_receive_queue))
2330                 mask |= POLLIN | POLLRDNORM;
2331
2332         /* Connection-based need to check for termination and startup */
2333         if (sk->sk_type == SOCK_SEQPACKET) {
2334                 if (sk->sk_state == TCP_CLOSE)
2335                         mask |= POLLHUP;
2336                 /* connection hasn't started yet? */
2337                 if (sk->sk_state == TCP_SYN_SENT)
2338                         return mask;
2339         }
2340
2341         /* No write status requested, avoid expensive OUT tests. */
2342         if (wait && !(wait->key & (POLLWRBAND | POLLWRNORM | POLLOUT)))
2343                 return mask;
2344
2345         writable = unix_writable(sk);
2346         if (writable) {
2347                 unix_state_lock(sk);
2348
2349                 other = unix_peer(sk);
2350                 if (other && unix_peer(other) != sk &&
2351                     unix_recvq_full(other) &&
2352                     unix_dgram_peer_wake_me(sk, other))
2353                         writable = 0;
2354
2355                 unix_state_unlock(sk);
2356         }
2357
2358         if (writable)
2359                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2360         else
2361                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2362
2363         return mask;
2364 }
2365
2366 #ifdef CONFIG_PROC_FS
2367 static struct sock *first_unix_socket(int *i)
2368 {
2369         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2370                 if (!hlist_empty(&unix_socket_table[*i]))
2371                         return __sk_head(&unix_socket_table[*i]);
2372         }
2373         return NULL;
2374 }
2375
2376 static struct sock *next_unix_socket(int *i, struct sock *s)
2377 {
2378         struct sock *next = sk_next(s);
2379         /* More in this chain? */
2380         if (next)
2381                 return next;
2382         /* Look for next non-empty chain. */
2383         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2384                 if (!hlist_empty(&unix_socket_table[*i]))
2385                         return __sk_head(&unix_socket_table[*i]);
2386         }
2387         return NULL;
2388 }
2389
2390 struct unix_iter_state {
2391         struct seq_net_private p;
2392         int i;
2393 };
2394
2395 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2396 {
2397         struct unix_iter_state *iter = seq->private;
2398         loff_t off = 0;
2399         struct sock *s;
2400
2401         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2402                 if (sock_net(s) != seq_file_net(seq))
2403                         continue;
2404                 if (off == pos)
2405                         return s;
2406                 ++off;
2407         }
2408         return NULL;
2409 }
2410
2411 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2412         __acquires(unix_table_lock)
2413 {
2414         spin_lock(&unix_table_lock);
2415         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2416 }
2417
2418 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2419 {
2420         struct unix_iter_state *iter = seq->private;
2421         struct sock *sk = v;
2422         ++*pos;
2423
2424         if (v == SEQ_START_TOKEN)
2425                 sk = first_unix_socket(&iter->i);
2426         else
2427                 sk = next_unix_socket(&iter->i, sk);
2428         while (sk && (sock_net(sk) != seq_file_net(seq)))
2429                 sk = next_unix_socket(&iter->i, sk);
2430         return sk;
2431 }
2432
2433 static void unix_seq_stop(struct seq_file *seq, void *v)
2434         __releases(unix_table_lock)
2435 {
2436         spin_unlock(&unix_table_lock);
2437 }
2438
2439 static int unix_seq_show(struct seq_file *seq, void *v)
2440 {
2441
2442         if (v == SEQ_START_TOKEN)
2443                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2444                          "Inode Path\n");
2445         else {
2446                 struct sock *s = v;
2447                 struct unix_sock *u = unix_sk(s);
2448                 unix_state_lock(s);
2449
2450                 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2451                         s,
2452                         atomic_read(&s->sk_refcnt),
2453                         0,
2454                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2455                         s->sk_type,
2456                         s->sk_socket ?
2457                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2458                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2459                         sock_i_ino(s));
2460
2461                 if (u->addr) {
2462                         int i, len;
2463                         seq_putc(seq, ' ');
2464
2465                         i = 0;
2466                         len = u->addr->len - sizeof(short);
2467                         if (!UNIX_ABSTRACT(s))
2468                                 len--;
2469                         else {
2470                                 seq_putc(seq, '@');
2471                                 i++;
2472                         }
2473                         for ( ; i < len; i++)
2474                                 seq_putc(seq, u->addr->name->sun_path[i]);
2475                 }
2476                 unix_state_unlock(s);
2477                 seq_putc(seq, '\n');
2478         }
2479
2480         return 0;
2481 }
2482
2483 static const struct seq_operations unix_seq_ops = {
2484         .start  = unix_seq_start,
2485         .next   = unix_seq_next,
2486         .stop   = unix_seq_stop,
2487         .show   = unix_seq_show,
2488 };
2489
2490 static int unix_seq_open(struct inode *inode, struct file *file)
2491 {
2492         return seq_open_net(inode, file, &unix_seq_ops,
2493                             sizeof(struct unix_iter_state));
2494 }
2495
2496 static const struct file_operations unix_seq_fops = {
2497         .owner          = THIS_MODULE,
2498         .open           = unix_seq_open,
2499         .read           = seq_read,
2500         .llseek         = seq_lseek,
2501         .release        = seq_release_net,
2502 };
2503
2504 #endif
2505
2506 static const struct net_proto_family unix_family_ops = {
2507         .family = PF_UNIX,
2508         .create = unix_create,
2509         .owner  = THIS_MODULE,
2510 };
2511
2512
2513 static int __net_init unix_net_init(struct net *net)
2514 {
2515         int error = -ENOMEM;
2516
2517         net->unx.sysctl_max_dgram_qlen = 10;
2518         if (unix_sysctl_register(net))
2519                 goto out;
2520
2521 #ifdef CONFIG_PROC_FS
2522         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2523                 unix_sysctl_unregister(net);
2524                 goto out;
2525         }
2526 #endif
2527         error = 0;
2528 out:
2529         return error;
2530 }
2531
2532 static void __net_exit unix_net_exit(struct net *net)
2533 {
2534         unix_sysctl_unregister(net);
2535         proc_net_remove(net, "unix");
2536 }
2537
2538 static struct pernet_operations unix_net_ops = {
2539         .init = unix_net_init,
2540         .exit = unix_net_exit,
2541 };
2542
2543 static int __init af_unix_init(void)
2544 {
2545         int rc = -1;
2546         struct sk_buff *dummy_skb;
2547
2548         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2549
2550         rc = proto_register(&unix_proto, 1);
2551         if (rc != 0) {
2552                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2553                        __func__);
2554                 goto out;
2555         }
2556
2557         sock_register(&unix_family_ops);
2558         register_pernet_subsys(&unix_net_ops);
2559 out:
2560         return rc;
2561 }
2562
2563 static void __exit af_unix_exit(void)
2564 {
2565         sock_unregister(PF_UNIX);
2566         proto_unregister(&unix_proto);
2567         unregister_pernet_subsys(&unix_net_ops);
2568 }
2569
2570 /* Earlier than device_initcall() so that other drivers invoking
2571    request_module() don't end up in a loop when modprobe tries
2572    to use a UNIX socket. But later than subsys_initcall() because
2573    we depend on stuff initialised there */
2574 fs_initcall(af_unix_init);
2575 module_exit(af_unix_exit);
2576
2577 MODULE_LICENSE("GPL");
2578 MODULE_ALIAS_NETPROTO(PF_UNIX);