af_unix: implement socket filter
[linux-3.10.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 hash = (__force unsigned)n;
153         hash ^= hash>>16;
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] doesnt 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 static inline int unix_writable(struct sock *sk)
307 {
308         return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
309 }
310
311 static void unix_write_space(struct sock *sk)
312 {
313         struct socket_wq *wq;
314
315         rcu_read_lock();
316         if (unix_writable(sk)) {
317                 wq = rcu_dereference(sk->sk_wq);
318                 if (wq_has_sleeper(wq))
319                         wake_up_interruptible_sync_poll(&wq->wait,
320                                 POLLOUT | POLLWRNORM | POLLWRBAND);
321                 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
322         }
323         rcu_read_unlock();
324 }
325
326 /* When dgram socket disconnects (or changes its peer), we clear its receive
327  * queue of packets arrived from previous peer. First, it allows to do
328  * flow control based only on wmem_alloc; second, sk connected to peer
329  * may receive messages only from that peer. */
330 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
331 {
332         if (!skb_queue_empty(&sk->sk_receive_queue)) {
333                 skb_queue_purge(&sk->sk_receive_queue);
334                 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
335
336                 /* If one link of bidirectional dgram pipe is disconnected,
337                  * we signal error. Messages are lost. Do not make this,
338                  * when peer was not connected to us.
339                  */
340                 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
341                         other->sk_err = ECONNRESET;
342                         other->sk_error_report(other);
343                 }
344         }
345 }
346
347 static void unix_sock_destructor(struct sock *sk)
348 {
349         struct unix_sock *u = unix_sk(sk);
350
351         skb_queue_purge(&sk->sk_receive_queue);
352
353         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
354         WARN_ON(!sk_unhashed(sk));
355         WARN_ON(sk->sk_socket);
356         if (!sock_flag(sk, SOCK_DEAD)) {
357                 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
358                 return;
359         }
360
361         if (u->addr)
362                 unix_release_addr(u->addr);
363
364         atomic_long_dec(&unix_nr_socks);
365         local_bh_disable();
366         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
367         local_bh_enable();
368 #ifdef UNIX_REFCNT_DEBUG
369         printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk,
370                 atomic_long_read(&unix_nr_socks));
371 #endif
372 }
373
374 static int unix_release_sock(struct sock *sk, int embrion)
375 {
376         struct unix_sock *u = unix_sk(sk);
377         struct dentry *dentry;
378         struct vfsmount *mnt;
379         struct sock *skpair;
380         struct sk_buff *skb;
381         int state;
382
383         unix_remove_socket(sk);
384
385         /* Clear state */
386         unix_state_lock(sk);
387         sock_orphan(sk);
388         sk->sk_shutdown = SHUTDOWN_MASK;
389         dentry       = u->dentry;
390         u->dentry    = NULL;
391         mnt          = u->mnt;
392         u->mnt       = NULL;
393         state = sk->sk_state;
394         sk->sk_state = TCP_CLOSE;
395         unix_state_unlock(sk);
396
397         wake_up_interruptible_all(&u->peer_wait);
398
399         skpair = unix_peer(sk);
400
401         if (skpair != NULL) {
402                 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
403                         unix_state_lock(skpair);
404                         /* No more writes */
405                         skpair->sk_shutdown = SHUTDOWN_MASK;
406                         if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
407                                 skpair->sk_err = ECONNRESET;
408                         unix_state_unlock(skpair);
409                         skpair->sk_state_change(skpair);
410                         sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
411                 }
412                 sock_put(skpair); /* It may now die */
413                 unix_peer(sk) = NULL;
414         }
415
416         /* Try to flush out this socket. Throw out buffers at least */
417
418         while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
419                 if (state == TCP_LISTEN)
420                         unix_release_sock(skb->sk, 1);
421                 /* passed fds are erased in the kfree_skb hook        */
422                 kfree_skb(skb);
423         }
424
425         if (dentry) {
426                 dput(dentry);
427                 mntput(mnt);
428         }
429
430         sock_put(sk);
431
432         /* ---- Socket is dead now and most probably destroyed ---- */
433
434         /*
435          * Fixme: BSD difference: In BSD all sockets connected to use get
436          *        ECONNRESET and we die on the spot. In Linux we behave
437          *        like files and pipes do and wait for the last
438          *        dereference.
439          *
440          * Can't we simply set sock->err?
441          *
442          *        What the above comment does talk about? --ANK(980817)
443          */
444
445         if (unix_tot_inflight)
446                 unix_gc();              /* Garbage collect fds */
447
448         return 0;
449 }
450
451 static void init_peercred(struct sock *sk)
452 {
453         put_pid(sk->sk_peer_pid);
454         if (sk->sk_peer_cred)
455                 put_cred(sk->sk_peer_cred);
456         sk->sk_peer_pid  = get_pid(task_tgid(current));
457         sk->sk_peer_cred = get_current_cred();
458 }
459
460 static void copy_peercred(struct sock *sk, struct sock *peersk)
461 {
462         put_pid(sk->sk_peer_pid);
463         if (sk->sk_peer_cred)
464                 put_cred(sk->sk_peer_cred);
465         sk->sk_peer_pid  = get_pid(peersk->sk_peer_pid);
466         sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
467 }
468
469 static int unix_listen(struct socket *sock, int backlog)
470 {
471         int err;
472         struct sock *sk = sock->sk;
473         struct unix_sock *u = unix_sk(sk);
474         struct pid *old_pid = NULL;
475         const struct cred *old_cred = NULL;
476
477         err = -EOPNOTSUPP;
478         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
479                 goto out;       /* Only stream/seqpacket sockets accept */
480         err = -EINVAL;
481         if (!u->addr)
482                 goto out;       /* No listens on an unbound socket */
483         unix_state_lock(sk);
484         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
485                 goto out_unlock;
486         if (backlog > sk->sk_max_ack_backlog)
487                 wake_up_interruptible_all(&u->peer_wait);
488         sk->sk_max_ack_backlog  = backlog;
489         sk->sk_state            = TCP_LISTEN;
490         /* set credentials so connect can copy them */
491         init_peercred(sk);
492         err = 0;
493
494 out_unlock:
495         unix_state_unlock(sk);
496         put_pid(old_pid);
497         if (old_cred)
498                 put_cred(old_cred);
499 out:
500         return err;
501 }
502
503 static int unix_release(struct socket *);
504 static int unix_bind(struct socket *, struct sockaddr *, int);
505 static int unix_stream_connect(struct socket *, struct sockaddr *,
506                                int addr_len, int flags);
507 static int unix_socketpair(struct socket *, struct socket *);
508 static int unix_accept(struct socket *, struct socket *, int);
509 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
510 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
511 static unsigned int unix_dgram_poll(struct file *, struct socket *,
512                                     poll_table *);
513 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
514 static int unix_shutdown(struct socket *, int);
515 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
516                                struct msghdr *, size_t);
517 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
518                                struct msghdr *, size_t, int);
519 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
520                               struct msghdr *, size_t);
521 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
522                               struct msghdr *, size_t, int);
523 static int unix_dgram_connect(struct socket *, struct sockaddr *,
524                               int, int);
525 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
526                                   struct msghdr *, size_t);
527
528 static const struct proto_ops unix_stream_ops = {
529         .family =       PF_UNIX,
530         .owner =        THIS_MODULE,
531         .release =      unix_release,
532         .bind =         unix_bind,
533         .connect =      unix_stream_connect,
534         .socketpair =   unix_socketpair,
535         .accept =       unix_accept,
536         .getname =      unix_getname,
537         .poll =         unix_poll,
538         .ioctl =        unix_ioctl,
539         .listen =       unix_listen,
540         .shutdown =     unix_shutdown,
541         .setsockopt =   sock_no_setsockopt,
542         .getsockopt =   sock_no_getsockopt,
543         .sendmsg =      unix_stream_sendmsg,
544         .recvmsg =      unix_stream_recvmsg,
545         .mmap =         sock_no_mmap,
546         .sendpage =     sock_no_sendpage,
547 };
548
549 static const struct proto_ops unix_dgram_ops = {
550         .family =       PF_UNIX,
551         .owner =        THIS_MODULE,
552         .release =      unix_release,
553         .bind =         unix_bind,
554         .connect =      unix_dgram_connect,
555         .socketpair =   unix_socketpair,
556         .accept =       sock_no_accept,
557         .getname =      unix_getname,
558         .poll =         unix_dgram_poll,
559         .ioctl =        unix_ioctl,
560         .listen =       sock_no_listen,
561         .shutdown =     unix_shutdown,
562         .setsockopt =   sock_no_setsockopt,
563         .getsockopt =   sock_no_getsockopt,
564         .sendmsg =      unix_dgram_sendmsg,
565         .recvmsg =      unix_dgram_recvmsg,
566         .mmap =         sock_no_mmap,
567         .sendpage =     sock_no_sendpage,
568 };
569
570 static const struct proto_ops unix_seqpacket_ops = {
571         .family =       PF_UNIX,
572         .owner =        THIS_MODULE,
573         .release =      unix_release,
574         .bind =         unix_bind,
575         .connect =      unix_stream_connect,
576         .socketpair =   unix_socketpair,
577         .accept =       unix_accept,
578         .getname =      unix_getname,
579         .poll =         unix_dgram_poll,
580         .ioctl =        unix_ioctl,
581         .listen =       unix_listen,
582         .shutdown =     unix_shutdown,
583         .setsockopt =   sock_no_setsockopt,
584         .getsockopt =   sock_no_getsockopt,
585         .sendmsg =      unix_seqpacket_sendmsg,
586         .recvmsg =      unix_dgram_recvmsg,
587         .mmap =         sock_no_mmap,
588         .sendpage =     sock_no_sendpage,
589 };
590
591 static struct proto unix_proto = {
592         .name                   = "UNIX",
593         .owner                  = THIS_MODULE,
594         .obj_size               = sizeof(struct unix_sock),
595 };
596
597 /*
598  * AF_UNIX sockets do not interact with hardware, hence they
599  * dont trigger interrupts - so it's safe for them to have
600  * bh-unsafe locking for their sk_receive_queue.lock. Split off
601  * this special lock-class by reinitializing the spinlock key:
602  */
603 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
604
605 static struct sock *unix_create1(struct net *net, struct socket *sock)
606 {
607         struct sock *sk = NULL;
608         struct unix_sock *u;
609
610         atomic_long_inc(&unix_nr_socks);
611         if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
612                 goto out;
613
614         sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
615         if (!sk)
616                 goto out;
617
618         sock_init_data(sock, sk);
619         lockdep_set_class(&sk->sk_receive_queue.lock,
620                                 &af_unix_sk_receive_queue_lock_key);
621
622         sk->sk_write_space      = unix_write_space;
623         sk->sk_max_ack_backlog  = net->unx.sysctl_max_dgram_qlen;
624         sk->sk_destruct         = unix_sock_destructor;
625         u         = unix_sk(sk);
626         u->dentry = NULL;
627         u->mnt    = NULL;
628         spin_lock_init(&u->lock);
629         atomic_long_set(&u->inflight, 0);
630         INIT_LIST_HEAD(&u->link);
631         mutex_init(&u->readlock); /* single task reading lock */
632         init_waitqueue_head(&u->peer_wait);
633         unix_insert_socket(unix_sockets_unbound, sk);
634 out:
635         if (sk == NULL)
636                 atomic_long_dec(&unix_nr_socks);
637         else {
638                 local_bh_disable();
639                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
640                 local_bh_enable();
641         }
642         return sk;
643 }
644
645 static int unix_create(struct net *net, struct socket *sock, int protocol,
646                        int kern)
647 {
648         if (protocol && protocol != PF_UNIX)
649                 return -EPROTONOSUPPORT;
650
651         sock->state = SS_UNCONNECTED;
652
653         switch (sock->type) {
654         case SOCK_STREAM:
655                 sock->ops = &unix_stream_ops;
656                 break;
657                 /*
658                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
659                  *      nothing uses it.
660                  */
661         case SOCK_RAW:
662                 sock->type = SOCK_DGRAM;
663         case SOCK_DGRAM:
664                 sock->ops = &unix_dgram_ops;
665                 break;
666         case SOCK_SEQPACKET:
667                 sock->ops = &unix_seqpacket_ops;
668                 break;
669         default:
670                 return -ESOCKTNOSUPPORT;
671         }
672
673         return unix_create1(net, sock) ? 0 : -ENOMEM;
674 }
675
676 static int unix_release(struct socket *sock)
677 {
678         struct sock *sk = sock->sk;
679
680         if (!sk)
681                 return 0;
682
683         sock->sk = NULL;
684
685         return unix_release_sock(sk, 0);
686 }
687
688 static int unix_autobind(struct socket *sock)
689 {
690         struct sock *sk = sock->sk;
691         struct net *net = sock_net(sk);
692         struct unix_sock *u = unix_sk(sk);
693         static u32 ordernum = 1;
694         struct unix_address *addr;
695         int err;
696         unsigned int retries = 0;
697
698         mutex_lock(&u->readlock);
699
700         err = 0;
701         if (u->addr)
702                 goto out;
703
704         err = -ENOMEM;
705         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
706         if (!addr)
707                 goto out;
708
709         addr->name->sun_family = AF_UNIX;
710         atomic_set(&addr->refcnt, 1);
711
712 retry:
713         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
714         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
715
716         spin_lock(&unix_table_lock);
717         ordernum = (ordernum+1)&0xFFFFF;
718
719         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
720                                       addr->hash)) {
721                 spin_unlock(&unix_table_lock);
722                 /*
723                  * __unix_find_socket_byname() may take long time if many names
724                  * are already in use.
725                  */
726                 cond_resched();
727                 /* Give up if all names seems to be in use. */
728                 if (retries++ == 0xFFFFF) {
729                         err = -ENOSPC;
730                         kfree(addr);
731                         goto out;
732                 }
733                 goto retry;
734         }
735         addr->hash ^= sk->sk_type;
736
737         __unix_remove_socket(sk);
738         u->addr = addr;
739         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
740         spin_unlock(&unix_table_lock);
741         err = 0;
742
743 out:    mutex_unlock(&u->readlock);
744         return err;
745 }
746
747 static struct sock *unix_find_other(struct net *net,
748                                     struct sockaddr_un *sunname, int len,
749                                     int type, unsigned hash, int *error)
750 {
751         struct sock *u;
752         struct path path;
753         int err = 0;
754
755         if (sunname->sun_path[0]) {
756                 struct inode *inode;
757                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
758                 if (err)
759                         goto fail;
760                 inode = path.dentry->d_inode;
761                 err = inode_permission(inode, MAY_WRITE);
762                 if (err)
763                         goto put_fail;
764
765                 err = -ECONNREFUSED;
766                 if (!S_ISSOCK(inode->i_mode))
767                         goto put_fail;
768                 u = unix_find_socket_byinode(inode);
769                 if (!u)
770                         goto put_fail;
771
772                 if (u->sk_type == type)
773                         touch_atime(path.mnt, path.dentry);
774
775                 path_put(&path);
776
777                 err = -EPROTOTYPE;
778                 if (u->sk_type != type) {
779                         sock_put(u);
780                         goto fail;
781                 }
782         } else {
783                 err = -ECONNREFUSED;
784                 u = unix_find_socket_byname(net, sunname, len, type, hash);
785                 if (u) {
786                         struct dentry *dentry;
787                         dentry = unix_sk(u)->dentry;
788                         if (dentry)
789                                 touch_atime(unix_sk(u)->mnt, dentry);
790                 } else
791                         goto fail;
792         }
793         return u;
794
795 put_fail:
796         path_put(&path);
797 fail:
798         *error = err;
799         return NULL;
800 }
801
802
803 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
804 {
805         struct sock *sk = sock->sk;
806         struct net *net = sock_net(sk);
807         struct unix_sock *u = unix_sk(sk);
808         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
809         struct dentry *dentry = NULL;
810         struct nameidata nd;
811         int err;
812         unsigned hash;
813         struct unix_address *addr;
814         struct hlist_head *list;
815
816         err = -EINVAL;
817         if (sunaddr->sun_family != AF_UNIX)
818                 goto out;
819
820         if (addr_len == sizeof(short)) {
821                 err = unix_autobind(sock);
822                 goto out;
823         }
824
825         err = unix_mkname(sunaddr, addr_len, &hash);
826         if (err < 0)
827                 goto out;
828         addr_len = err;
829
830         mutex_lock(&u->readlock);
831
832         err = -EINVAL;
833         if (u->addr)
834                 goto out_up;
835
836         err = -ENOMEM;
837         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
838         if (!addr)
839                 goto out_up;
840
841         memcpy(addr->name, sunaddr, addr_len);
842         addr->len = addr_len;
843         addr->hash = hash ^ sk->sk_type;
844         atomic_set(&addr->refcnt, 1);
845
846         if (sunaddr->sun_path[0]) {
847                 unsigned int mode;
848                 err = 0;
849                 /*
850                  * Get the parent directory, calculate the hash for last
851                  * component.
852                  */
853                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
854                 if (err)
855                         goto out_mknod_parent;
856
857                 dentry = lookup_create(&nd, 0);
858                 err = PTR_ERR(dentry);
859                 if (IS_ERR(dentry))
860                         goto out_mknod_unlock;
861
862                 /*
863                  * All right, let's create it.
864                  */
865                 mode = S_IFSOCK |
866                        (SOCK_INODE(sock)->i_mode & ~current_umask());
867                 err = mnt_want_write(nd.path.mnt);
868                 if (err)
869                         goto out_mknod_dput;
870                 err = security_path_mknod(&nd.path, dentry, mode, 0);
871                 if (err)
872                         goto out_mknod_drop_write;
873                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
874 out_mknod_drop_write:
875                 mnt_drop_write(nd.path.mnt);
876                 if (err)
877                         goto out_mknod_dput;
878                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
879                 dput(nd.path.dentry);
880                 nd.path.dentry = dentry;
881
882                 addr->hash = UNIX_HASH_SIZE;
883         }
884
885         spin_lock(&unix_table_lock);
886
887         if (!sunaddr->sun_path[0]) {
888                 err = -EADDRINUSE;
889                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
890                                               sk->sk_type, hash)) {
891                         unix_release_addr(addr);
892                         goto out_unlock;
893                 }
894
895                 list = &unix_socket_table[addr->hash];
896         } else {
897                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
898                 u->dentry = nd.path.dentry;
899                 u->mnt    = nd.path.mnt;
900         }
901
902         err = 0;
903         __unix_remove_socket(sk);
904         u->addr = addr;
905         __unix_insert_socket(list, sk);
906
907 out_unlock:
908         spin_unlock(&unix_table_lock);
909 out_up:
910         mutex_unlock(&u->readlock);
911 out:
912         return err;
913
914 out_mknod_dput:
915         dput(dentry);
916 out_mknod_unlock:
917         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
918         path_put(&nd.path);
919 out_mknod_parent:
920         if (err == -EEXIST)
921                 err = -EADDRINUSE;
922         unix_release_addr(addr);
923         goto out_up;
924 }
925
926 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
927 {
928         if (unlikely(sk1 == sk2) || !sk2) {
929                 unix_state_lock(sk1);
930                 return;
931         }
932         if (sk1 < sk2) {
933                 unix_state_lock(sk1);
934                 unix_state_lock_nested(sk2);
935         } else {
936                 unix_state_lock(sk2);
937                 unix_state_lock_nested(sk1);
938         }
939 }
940
941 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
942 {
943         if (unlikely(sk1 == sk2) || !sk2) {
944                 unix_state_unlock(sk1);
945                 return;
946         }
947         unix_state_unlock(sk1);
948         unix_state_unlock(sk2);
949 }
950
951 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
952                               int alen, int flags)
953 {
954         struct sock *sk = sock->sk;
955         struct net *net = sock_net(sk);
956         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
957         struct sock *other;
958         unsigned hash;
959         int err;
960
961         if (addr->sa_family != AF_UNSPEC) {
962                 err = unix_mkname(sunaddr, alen, &hash);
963                 if (err < 0)
964                         goto out;
965                 alen = err;
966
967                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
968                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
969                         goto out;
970
971 restart:
972                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
973                 if (!other)
974                         goto out;
975
976                 unix_state_double_lock(sk, other);
977
978                 /* Apparently VFS overslept socket death. Retry. */
979                 if (sock_flag(other, SOCK_DEAD)) {
980                         unix_state_double_unlock(sk, other);
981                         sock_put(other);
982                         goto restart;
983                 }
984
985                 err = -EPERM;
986                 if (!unix_may_send(sk, other))
987                         goto out_unlock;
988
989                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
990                 if (err)
991                         goto out_unlock;
992
993         } else {
994                 /*
995                  *      1003.1g breaking connected state with AF_UNSPEC
996                  */
997                 other = NULL;
998                 unix_state_double_lock(sk, other);
999         }
1000
1001         /*
1002          * If it was connected, reconnect.
1003          */
1004         if (unix_peer(sk)) {
1005                 struct sock *old_peer = unix_peer(sk);
1006                 unix_peer(sk) = other;
1007                 unix_state_double_unlock(sk, other);
1008
1009                 if (other != old_peer)
1010                         unix_dgram_disconnected(sk, old_peer);
1011                 sock_put(old_peer);
1012         } else {
1013                 unix_peer(sk) = other;
1014                 unix_state_double_unlock(sk, other);
1015         }
1016         return 0;
1017
1018 out_unlock:
1019         unix_state_double_unlock(sk, other);
1020         sock_put(other);
1021 out:
1022         return err;
1023 }
1024
1025 static long unix_wait_for_peer(struct sock *other, long timeo)
1026 {
1027         struct unix_sock *u = unix_sk(other);
1028         int sched;
1029         DEFINE_WAIT(wait);
1030
1031         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1032
1033         sched = !sock_flag(other, SOCK_DEAD) &&
1034                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1035                 unix_recvq_full(other);
1036
1037         unix_state_unlock(other);
1038
1039         if (sched)
1040                 timeo = schedule_timeout(timeo);
1041
1042         finish_wait(&u->peer_wait, &wait);
1043         return timeo;
1044 }
1045
1046 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1047                                int addr_len, int flags)
1048 {
1049         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1050         struct sock *sk = sock->sk;
1051         struct net *net = sock_net(sk);
1052         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1053         struct sock *newsk = NULL;
1054         struct sock *other = NULL;
1055         struct sk_buff *skb = NULL;
1056         unsigned hash;
1057         int st;
1058         int err;
1059         long timeo;
1060
1061         err = unix_mkname(sunaddr, addr_len, &hash);
1062         if (err < 0)
1063                 goto out;
1064         addr_len = err;
1065
1066         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1067             (err = unix_autobind(sock)) != 0)
1068                 goto out;
1069
1070         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1071
1072         /* First of all allocate resources.
1073            If we will make it after state is locked,
1074            we will have to recheck all again in any case.
1075          */
1076
1077         err = -ENOMEM;
1078
1079         /* create new sock for complete connection */
1080         newsk = unix_create1(sock_net(sk), NULL);
1081         if (newsk == NULL)
1082                 goto out;
1083
1084         /* Allocate skb for sending to listening sock */
1085         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1086         if (skb == NULL)
1087                 goto out;
1088
1089 restart:
1090         /*  Find listening sock. */
1091         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1092         if (!other)
1093                 goto out;
1094
1095         /* Latch state of peer */
1096         unix_state_lock(other);
1097
1098         /* Apparently VFS overslept socket death. Retry. */
1099         if (sock_flag(other, SOCK_DEAD)) {
1100                 unix_state_unlock(other);
1101                 sock_put(other);
1102                 goto restart;
1103         }
1104
1105         err = -ECONNREFUSED;
1106         if (other->sk_state != TCP_LISTEN)
1107                 goto out_unlock;
1108         if (other->sk_shutdown & RCV_SHUTDOWN)
1109                 goto out_unlock;
1110
1111         if (unix_recvq_full(other)) {
1112                 err = -EAGAIN;
1113                 if (!timeo)
1114                         goto out_unlock;
1115
1116                 timeo = unix_wait_for_peer(other, timeo);
1117
1118                 err = sock_intr_errno(timeo);
1119                 if (signal_pending(current))
1120                         goto out;
1121                 sock_put(other);
1122                 goto restart;
1123         }
1124
1125         /* Latch our state.
1126
1127            It is tricky place. We need to grab write lock and cannot
1128            drop lock on peer. It is dangerous because deadlock is
1129            possible. Connect to self case and simultaneous
1130            attempt to connect are eliminated by checking socket
1131            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1132            check this before attempt to grab lock.
1133
1134            Well, and we have to recheck the state after socket locked.
1135          */
1136         st = sk->sk_state;
1137
1138         switch (st) {
1139         case TCP_CLOSE:
1140                 /* This is ok... continue with connect */
1141                 break;
1142         case TCP_ESTABLISHED:
1143                 /* Socket is already connected */
1144                 err = -EISCONN;
1145                 goto out_unlock;
1146         default:
1147                 err = -EINVAL;
1148                 goto out_unlock;
1149         }
1150
1151         unix_state_lock_nested(sk);
1152
1153         if (sk->sk_state != st) {
1154                 unix_state_unlock(sk);
1155                 unix_state_unlock(other);
1156                 sock_put(other);
1157                 goto restart;
1158         }
1159
1160         err = security_unix_stream_connect(sk, other, newsk);
1161         if (err) {
1162                 unix_state_unlock(sk);
1163                 goto out_unlock;
1164         }
1165
1166         /* The way is open! Fastly set all the necessary fields... */
1167
1168         sock_hold(sk);
1169         unix_peer(newsk)        = sk;
1170         newsk->sk_state         = TCP_ESTABLISHED;
1171         newsk->sk_type          = sk->sk_type;
1172         init_peercred(newsk);
1173         newu = unix_sk(newsk);
1174         newsk->sk_wq            = &newu->peer_wq;
1175         otheru = unix_sk(other);
1176
1177         /* copy address information from listening to new sock*/
1178         if (otheru->addr) {
1179                 atomic_inc(&otheru->addr->refcnt);
1180                 newu->addr = otheru->addr;
1181         }
1182         if (otheru->dentry) {
1183                 newu->dentry    = dget(otheru->dentry);
1184                 newu->mnt       = mntget(otheru->mnt);
1185         }
1186
1187         /* Set credentials */
1188         copy_peercred(sk, other);
1189
1190         sock->state     = SS_CONNECTED;
1191         sk->sk_state    = TCP_ESTABLISHED;
1192         sock_hold(newsk);
1193
1194         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1195         unix_peer(sk)   = newsk;
1196
1197         unix_state_unlock(sk);
1198
1199         /* take ten and and send info to listening sock */
1200         spin_lock(&other->sk_receive_queue.lock);
1201         __skb_queue_tail(&other->sk_receive_queue, skb);
1202         spin_unlock(&other->sk_receive_queue.lock);
1203         unix_state_unlock(other);
1204         other->sk_data_ready(other, 0);
1205         sock_put(other);
1206         return 0;
1207
1208 out_unlock:
1209         if (other)
1210                 unix_state_unlock(other);
1211
1212 out:
1213         kfree_skb(skb);
1214         if (newsk)
1215                 unix_release_sock(newsk, 0);
1216         if (other)
1217                 sock_put(other);
1218         return err;
1219 }
1220
1221 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1222 {
1223         struct sock *ska = socka->sk, *skb = sockb->sk;
1224
1225         /* Join our sockets back to back */
1226         sock_hold(ska);
1227         sock_hold(skb);
1228         unix_peer(ska) = skb;
1229         unix_peer(skb) = ska;
1230         init_peercred(ska);
1231         init_peercred(skb);
1232
1233         if (ska->sk_type != SOCK_DGRAM) {
1234                 ska->sk_state = TCP_ESTABLISHED;
1235                 skb->sk_state = TCP_ESTABLISHED;
1236                 socka->state  = SS_CONNECTED;
1237                 sockb->state  = SS_CONNECTED;
1238         }
1239         return 0;
1240 }
1241
1242 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1243 {
1244         struct sock *sk = sock->sk;
1245         struct sock *tsk;
1246         struct sk_buff *skb;
1247         int err;
1248
1249         err = -EOPNOTSUPP;
1250         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1251                 goto out;
1252
1253         err = -EINVAL;
1254         if (sk->sk_state != TCP_LISTEN)
1255                 goto out;
1256
1257         /* If socket state is TCP_LISTEN it cannot change (for now...),
1258          * so that no locks are necessary.
1259          */
1260
1261         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1262         if (!skb) {
1263                 /* This means receive shutdown. */
1264                 if (err == 0)
1265                         err = -EINVAL;
1266                 goto out;
1267         }
1268
1269         tsk = skb->sk;
1270         skb_free_datagram(sk, skb);
1271         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1272
1273         /* attach accepted sock to socket */
1274         unix_state_lock(tsk);
1275         newsock->state = SS_CONNECTED;
1276         sock_graft(tsk, newsock);
1277         unix_state_unlock(tsk);
1278         return 0;
1279
1280 out:
1281         return err;
1282 }
1283
1284
1285 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1286 {
1287         struct sock *sk = sock->sk;
1288         struct unix_sock *u;
1289         DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1290         int err = 0;
1291
1292         if (peer) {
1293                 sk = unix_peer_get(sk);
1294
1295                 err = -ENOTCONN;
1296                 if (!sk)
1297                         goto out;
1298                 err = 0;
1299         } else {
1300                 sock_hold(sk);
1301         }
1302
1303         u = unix_sk(sk);
1304         unix_state_lock(sk);
1305         if (!u->addr) {
1306                 sunaddr->sun_family = AF_UNIX;
1307                 sunaddr->sun_path[0] = 0;
1308                 *uaddr_len = sizeof(short);
1309         } else {
1310                 struct unix_address *addr = u->addr;
1311
1312                 *uaddr_len = addr->len;
1313                 memcpy(sunaddr, addr->name, *uaddr_len);
1314         }
1315         unix_state_unlock(sk);
1316         sock_put(sk);
1317 out:
1318         return err;
1319 }
1320
1321 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1322 {
1323         int i;
1324
1325         scm->fp = UNIXCB(skb).fp;
1326         UNIXCB(skb).fp = NULL;
1327
1328         for (i = scm->fp->count-1; i >= 0; i--)
1329                 unix_notinflight(scm->fp->fp[i]);
1330 }
1331
1332 static void unix_destruct_scm(struct sk_buff *skb)
1333 {
1334         struct scm_cookie scm;
1335         memset(&scm, 0, sizeof(scm));
1336         scm.pid  = UNIXCB(skb).pid;
1337         scm.cred = UNIXCB(skb).cred;
1338         if (UNIXCB(skb).fp)
1339                 unix_detach_fds(&scm, skb);
1340
1341         /* Alas, it calls VFS */
1342         /* So fscking what? fput() had been SMP-safe since the last Summer */
1343         scm_destroy(&scm);
1344         sock_wfree(skb);
1345 }
1346
1347 #define MAX_RECURSION_LEVEL 4
1348
1349 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1350 {
1351         int i;
1352         unsigned char max_level = 0;
1353         int unix_sock_count = 0;
1354
1355         for (i = scm->fp->count - 1; i >= 0; i--) {
1356                 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1357
1358                 if (sk) {
1359                         unix_sock_count++;
1360                         max_level = max(max_level,
1361                                         unix_sk(sk)->recursion_level);
1362                 }
1363         }
1364         if (unlikely(max_level > MAX_RECURSION_LEVEL))
1365                 return -ETOOMANYREFS;
1366
1367         /*
1368          * Need to duplicate file references for the sake of garbage
1369          * collection.  Otherwise a socket in the fps might become a
1370          * candidate for GC while the skb is not yet queued.
1371          */
1372         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1373         if (!UNIXCB(skb).fp)
1374                 return -ENOMEM;
1375
1376         if (unix_sock_count) {
1377                 for (i = scm->fp->count - 1; i >= 0; i--)
1378                         unix_inflight(scm->fp->fp[i]);
1379         }
1380         return max_level;
1381 }
1382
1383 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1384 {
1385         int err = 0;
1386         UNIXCB(skb).pid  = get_pid(scm->pid);
1387         UNIXCB(skb).cred = get_cred(scm->cred);
1388         UNIXCB(skb).fp = NULL;
1389         if (scm->fp && send_fds)
1390                 err = unix_attach_fds(scm, skb);
1391
1392         skb->destructor = unix_destruct_scm;
1393         return err;
1394 }
1395
1396 /*
1397  *      Send AF_UNIX data.
1398  */
1399
1400 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1401                               struct msghdr *msg, size_t len)
1402 {
1403         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1404         struct sock *sk = sock->sk;
1405         struct net *net = sock_net(sk);
1406         struct unix_sock *u = unix_sk(sk);
1407         struct sockaddr_un *sunaddr = msg->msg_name;
1408         struct sock *other = NULL;
1409         int namelen = 0; /* fake GCC */
1410         int err;
1411         unsigned hash;
1412         struct sk_buff *skb;
1413         long timeo;
1414         struct scm_cookie tmp_scm;
1415         int max_level;
1416
1417         if (NULL == siocb->scm)
1418                 siocb->scm = &tmp_scm;
1419         wait_for_unix_gc();
1420         err = scm_send(sock, msg, siocb->scm);
1421         if (err < 0)
1422                 return err;
1423
1424         err = -EOPNOTSUPP;
1425         if (msg->msg_flags&MSG_OOB)
1426                 goto out;
1427
1428         if (msg->msg_namelen) {
1429                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1430                 if (err < 0)
1431                         goto out;
1432                 namelen = err;
1433         } else {
1434                 sunaddr = NULL;
1435                 err = -ENOTCONN;
1436                 other = unix_peer_get(sk);
1437                 if (!other)
1438                         goto out;
1439         }
1440
1441         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1442             && (err = unix_autobind(sock)) != 0)
1443                 goto out;
1444
1445         err = -EMSGSIZE;
1446         if (len > sk->sk_sndbuf - 32)
1447                 goto out;
1448
1449         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1450         if (skb == NULL)
1451                 goto out;
1452
1453         err = unix_scm_to_skb(siocb->scm, skb, true);
1454         if (err < 0)
1455                 goto out_free;
1456         max_level = err + 1;
1457         unix_get_secdata(siocb->scm, skb);
1458
1459         skb_reset_transport_header(skb);
1460         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1461         if (err)
1462                 goto out_free;
1463
1464         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1465
1466 restart:
1467         if (!other) {
1468                 err = -ECONNRESET;
1469                 if (sunaddr == NULL)
1470                         goto out_free;
1471
1472                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1473                                         hash, &err);
1474                 if (other == NULL)
1475                         goto out_free;
1476         }
1477
1478         if (sk_filter(other, skb) < 0) {
1479                 /* Toss the packet but do not return any error to the sender */
1480                 err = len;
1481                 goto out_free;
1482         }
1483
1484         unix_state_lock(other);
1485         err = -EPERM;
1486         if (!unix_may_send(sk, other))
1487                 goto out_unlock;
1488
1489         if (sock_flag(other, SOCK_DEAD)) {
1490                 /*
1491                  *      Check with 1003.1g - what should
1492                  *      datagram error
1493                  */
1494                 unix_state_unlock(other);
1495                 sock_put(other);
1496
1497                 err = 0;
1498                 unix_state_lock(sk);
1499                 if (unix_peer(sk) == other) {
1500                         unix_peer(sk) = NULL;
1501                         unix_state_unlock(sk);
1502
1503                         unix_dgram_disconnected(sk, other);
1504                         sock_put(other);
1505                         err = -ECONNREFUSED;
1506                 } else {
1507                         unix_state_unlock(sk);
1508                 }
1509
1510                 other = NULL;
1511                 if (err)
1512                         goto out_free;
1513                 goto restart;
1514         }
1515
1516         err = -EPIPE;
1517         if (other->sk_shutdown & RCV_SHUTDOWN)
1518                 goto out_unlock;
1519
1520         if (sk->sk_type != SOCK_SEQPACKET) {
1521                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1522                 if (err)
1523                         goto out_unlock;
1524         }
1525
1526         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1527                 if (!timeo) {
1528                         err = -EAGAIN;
1529                         goto out_unlock;
1530                 }
1531
1532                 timeo = unix_wait_for_peer(other, timeo);
1533
1534                 err = sock_intr_errno(timeo);
1535                 if (signal_pending(current))
1536                         goto out_free;
1537
1538                 goto restart;
1539         }
1540
1541         if (sock_flag(other, SOCK_RCVTSTAMP))
1542                 __net_timestamp(skb);
1543         skb_queue_tail(&other->sk_receive_queue, skb);
1544         if (max_level > unix_sk(other)->recursion_level)
1545                 unix_sk(other)->recursion_level = max_level;
1546         unix_state_unlock(other);
1547         other->sk_data_ready(other, len);
1548         sock_put(other);
1549         scm_destroy(siocb->scm);
1550         return len;
1551
1552 out_unlock:
1553         unix_state_unlock(other);
1554 out_free:
1555         kfree_skb(skb);
1556 out:
1557         if (other)
1558                 sock_put(other);
1559         scm_destroy(siocb->scm);
1560         return err;
1561 }
1562
1563
1564 static int unix_stream_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 sock *other = NULL;
1570         struct sockaddr_un *sunaddr = msg->msg_name;
1571         int err, size;
1572         struct sk_buff *skb;
1573         int sent = 0;
1574         struct scm_cookie tmp_scm;
1575         bool fds_sent = false;
1576         int max_level;
1577
1578         if (NULL == siocb->scm)
1579                 siocb->scm = &tmp_scm;
1580         wait_for_unix_gc();
1581         err = scm_send(sock, msg, siocb->scm);
1582         if (err < 0)
1583                 return err;
1584
1585         err = -EOPNOTSUPP;
1586         if (msg->msg_flags&MSG_OOB)
1587                 goto out_err;
1588
1589         if (msg->msg_namelen) {
1590                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1591                 goto out_err;
1592         } else {
1593                 sunaddr = NULL;
1594                 err = -ENOTCONN;
1595                 other = unix_peer(sk);
1596                 if (!other)
1597                         goto out_err;
1598         }
1599
1600         if (sk->sk_shutdown & SEND_SHUTDOWN)
1601                 goto pipe_err;
1602
1603         while (sent < len) {
1604                 /*
1605                  *      Optimisation for the fact that under 0.01% of X
1606                  *      messages typically need breaking up.
1607                  */
1608
1609                 size = len-sent;
1610
1611                 /* Keep two messages in the pipe so it schedules better */
1612                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1613                         size = (sk->sk_sndbuf >> 1) - 64;
1614
1615                 if (size > SKB_MAX_ALLOC)
1616                         size = SKB_MAX_ALLOC;
1617
1618                 /*
1619                  *      Grab a buffer
1620                  */
1621
1622                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1623                                           &err);
1624
1625                 if (skb == NULL)
1626                         goto out_err;
1627
1628                 /*
1629                  *      If you pass two values to the sock_alloc_send_skb
1630                  *      it tries to grab the large buffer with GFP_NOFS
1631                  *      (which can fail easily), and if it fails grab the
1632                  *      fallback size buffer which is under a page and will
1633                  *      succeed. [Alan]
1634                  */
1635                 size = min_t(int, size, skb_tailroom(skb));
1636
1637
1638                 /* Only send the fds in the first buffer */
1639                 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1640                 if (err < 0) {
1641                         kfree_skb(skb);
1642                         goto out_err;
1643                 }
1644                 max_level = err + 1;
1645                 fds_sent = true;
1646
1647                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1648                 if (err) {
1649                         kfree_skb(skb);
1650                         goto out_err;
1651                 }
1652
1653                 unix_state_lock(other);
1654
1655                 if (sock_flag(other, SOCK_DEAD) ||
1656                     (other->sk_shutdown & RCV_SHUTDOWN))
1657                         goto pipe_err_free;
1658
1659                 skb_queue_tail(&other->sk_receive_queue, skb);
1660                 if (max_level > unix_sk(other)->recursion_level)
1661                         unix_sk(other)->recursion_level = max_level;
1662                 unix_state_unlock(other);
1663                 other->sk_data_ready(other, size);
1664                 sent += size;
1665         }
1666
1667         scm_destroy(siocb->scm);
1668         siocb->scm = NULL;
1669
1670         return sent;
1671
1672 pipe_err_free:
1673         unix_state_unlock(other);
1674         kfree_skb(skb);
1675 pipe_err:
1676         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1677                 send_sig(SIGPIPE, current, 0);
1678         err = -EPIPE;
1679 out_err:
1680         scm_destroy(siocb->scm);
1681         siocb->scm = NULL;
1682         return sent ? : err;
1683 }
1684
1685 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1686                                   struct msghdr *msg, size_t len)
1687 {
1688         int err;
1689         struct sock *sk = sock->sk;
1690
1691         err = sock_error(sk);
1692         if (err)
1693                 return err;
1694
1695         if (sk->sk_state != TCP_ESTABLISHED)
1696                 return -ENOTCONN;
1697
1698         if (msg->msg_namelen)
1699                 msg->msg_namelen = 0;
1700
1701         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1702 }
1703
1704 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1705 {
1706         struct unix_sock *u = unix_sk(sk);
1707
1708         msg->msg_namelen = 0;
1709         if (u->addr) {
1710                 msg->msg_namelen = u->addr->len;
1711                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1712         }
1713 }
1714
1715 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1716                               struct msghdr *msg, size_t size,
1717                               int flags)
1718 {
1719         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1720         struct scm_cookie tmp_scm;
1721         struct sock *sk = sock->sk;
1722         struct unix_sock *u = unix_sk(sk);
1723         int noblock = flags & MSG_DONTWAIT;
1724         struct sk_buff *skb;
1725         int err;
1726
1727         err = -EOPNOTSUPP;
1728         if (flags&MSG_OOB)
1729                 goto out;
1730
1731         msg->msg_namelen = 0;
1732
1733         mutex_lock(&u->readlock);
1734
1735         skb = skb_recv_datagram(sk, flags, noblock, &err);
1736         if (!skb) {
1737                 unix_state_lock(sk);
1738                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1739                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1740                     (sk->sk_shutdown & RCV_SHUTDOWN))
1741                         err = 0;
1742                 unix_state_unlock(sk);
1743                 goto out_unlock;
1744         }
1745
1746         wake_up_interruptible_sync_poll(&u->peer_wait,
1747                                         POLLOUT | POLLWRNORM | POLLWRBAND);
1748
1749         if (msg->msg_name)
1750                 unix_copy_addr(msg, skb->sk);
1751
1752         if (size > skb->len)
1753                 size = skb->len;
1754         else if (size < skb->len)
1755                 msg->msg_flags |= MSG_TRUNC;
1756
1757         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1758         if (err)
1759                 goto out_free;
1760
1761         if (sock_flag(sk, SOCK_RCVTSTAMP))
1762                 __sock_recv_timestamp(msg, sk, skb);
1763
1764         if (!siocb->scm) {
1765                 siocb->scm = &tmp_scm;
1766                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1767         }
1768         scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1769         unix_set_secdata(siocb->scm, skb);
1770
1771         if (!(flags & MSG_PEEK)) {
1772                 if (UNIXCB(skb).fp)
1773                         unix_detach_fds(siocb->scm, skb);
1774         } else {
1775                 /* It is questionable: on PEEK we could:
1776                    - do not return fds - good, but too simple 8)
1777                    - return fds, and do not return them on read (old strategy,
1778                      apparently wrong)
1779                    - clone fds (I chose it for now, it is the most universal
1780                      solution)
1781
1782                    POSIX 1003.1g does not actually define this clearly
1783                    at all. POSIX 1003.1g doesn't define a lot of things
1784                    clearly however!
1785
1786                 */
1787                 if (UNIXCB(skb).fp)
1788                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1789         }
1790         err = size;
1791
1792         scm_recv(sock, msg, siocb->scm, flags);
1793
1794 out_free:
1795         skb_free_datagram(sk, skb);
1796 out_unlock:
1797         mutex_unlock(&u->readlock);
1798 out:
1799         return err;
1800 }
1801
1802 /*
1803  *      Sleep until data has arrive. But check for races..
1804  */
1805
1806 static long unix_stream_data_wait(struct sock *sk, long timeo)
1807 {
1808         DEFINE_WAIT(wait);
1809
1810         unix_state_lock(sk);
1811
1812         for (;;) {
1813                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1814
1815                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1816                     sk->sk_err ||
1817                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1818                     signal_pending(current) ||
1819                     !timeo)
1820                         break;
1821
1822                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1823                 unix_state_unlock(sk);
1824                 timeo = schedule_timeout(timeo);
1825                 unix_state_lock(sk);
1826                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1827         }
1828
1829         finish_wait(sk_sleep(sk), &wait);
1830         unix_state_unlock(sk);
1831         return timeo;
1832 }
1833
1834
1835
1836 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1837                                struct msghdr *msg, size_t size,
1838                                int flags)
1839 {
1840         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1841         struct scm_cookie tmp_scm;
1842         struct sock *sk = sock->sk;
1843         struct unix_sock *u = unix_sk(sk);
1844         struct sockaddr_un *sunaddr = msg->msg_name;
1845         int copied = 0;
1846         int check_creds = 0;
1847         int target;
1848         int err = 0;
1849         long timeo;
1850
1851         err = -EINVAL;
1852         if (sk->sk_state != TCP_ESTABLISHED)
1853                 goto out;
1854
1855         err = -EOPNOTSUPP;
1856         if (flags&MSG_OOB)
1857                 goto out;
1858
1859         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1860         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1861
1862         msg->msg_namelen = 0;
1863
1864         /* Lock the socket to prevent queue disordering
1865          * while sleeps in memcpy_tomsg
1866          */
1867
1868         if (!siocb->scm) {
1869                 siocb->scm = &tmp_scm;
1870                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1871         }
1872
1873         mutex_lock(&u->readlock);
1874
1875         do {
1876                 int chunk;
1877                 struct sk_buff *skb;
1878
1879                 unix_state_lock(sk);
1880                 skb = skb_dequeue(&sk->sk_receive_queue);
1881                 if (skb == NULL) {
1882                         unix_sk(sk)->recursion_level = 0;
1883                         if (copied >= target)
1884                                 goto unlock;
1885
1886                         /*
1887                          *      POSIX 1003.1g mandates this order.
1888                          */
1889
1890                         err = sock_error(sk);
1891                         if (err)
1892                                 goto unlock;
1893                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1894                                 goto unlock;
1895
1896                         unix_state_unlock(sk);
1897                         err = -EAGAIN;
1898                         if (!timeo)
1899                                 break;
1900                         mutex_unlock(&u->readlock);
1901
1902                         timeo = unix_stream_data_wait(sk, timeo);
1903
1904                         if (signal_pending(current)) {
1905                                 err = sock_intr_errno(timeo);
1906                                 goto out;
1907                         }
1908                         mutex_lock(&u->readlock);
1909                         continue;
1910  unlock:
1911                         unix_state_unlock(sk);
1912                         break;
1913                 }
1914                 unix_state_unlock(sk);
1915
1916                 if (check_creds) {
1917                         /* Never glue messages from different writers */
1918                         if ((UNIXCB(skb).pid  != siocb->scm->pid) ||
1919                             (UNIXCB(skb).cred != siocb->scm->cred)) {
1920                                 skb_queue_head(&sk->sk_receive_queue, skb);
1921                                 break;
1922                         }
1923                 } else {
1924                         /* Copy credentials */
1925                         scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1926                         check_creds = 1;
1927                 }
1928
1929                 /* Copy address just once */
1930                 if (sunaddr) {
1931                         unix_copy_addr(msg, skb->sk);
1932                         sunaddr = NULL;
1933                 }
1934
1935                 chunk = min_t(unsigned int, skb->len, size);
1936                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1937                         skb_queue_head(&sk->sk_receive_queue, skb);
1938                         if (copied == 0)
1939                                 copied = -EFAULT;
1940                         break;
1941                 }
1942                 copied += chunk;
1943                 size -= chunk;
1944
1945                 /* Mark read part of skb as used */
1946                 if (!(flags & MSG_PEEK)) {
1947                         skb_pull(skb, chunk);
1948
1949                         if (UNIXCB(skb).fp)
1950                                 unix_detach_fds(siocb->scm, skb);
1951
1952                         /* put the skb back if we didn't use it up.. */
1953                         if (skb->len) {
1954                                 skb_queue_head(&sk->sk_receive_queue, skb);
1955                                 break;
1956                         }
1957
1958                         consume_skb(skb);
1959
1960                         if (siocb->scm->fp)
1961                                 break;
1962                 } else {
1963                         /* It is questionable, see note in unix_dgram_recvmsg.
1964                          */
1965                         if (UNIXCB(skb).fp)
1966                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1967
1968                         /* put message back and return */
1969                         skb_queue_head(&sk->sk_receive_queue, skb);
1970                         break;
1971                 }
1972         } while (size);
1973
1974         mutex_unlock(&u->readlock);
1975         scm_recv(sock, msg, siocb->scm, flags);
1976 out:
1977         return copied ? : err;
1978 }
1979
1980 static int unix_shutdown(struct socket *sock, int mode)
1981 {
1982         struct sock *sk = sock->sk;
1983         struct sock *other;
1984
1985         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1986
1987         if (mode) {
1988                 unix_state_lock(sk);
1989                 sk->sk_shutdown |= mode;
1990                 other = unix_peer(sk);
1991                 if (other)
1992                         sock_hold(other);
1993                 unix_state_unlock(sk);
1994                 sk->sk_state_change(sk);
1995
1996                 if (other &&
1997                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1998
1999                         int peer_mode = 0;
2000
2001                         if (mode&RCV_SHUTDOWN)
2002                                 peer_mode |= SEND_SHUTDOWN;
2003                         if (mode&SEND_SHUTDOWN)
2004                                 peer_mode |= RCV_SHUTDOWN;
2005                         unix_state_lock(other);
2006                         other->sk_shutdown |= peer_mode;
2007                         unix_state_unlock(other);
2008                         other->sk_state_change(other);
2009                         if (peer_mode == SHUTDOWN_MASK)
2010                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2011                         else if (peer_mode & RCV_SHUTDOWN)
2012                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2013                 }
2014                 if (other)
2015                         sock_put(other);
2016         }
2017         return 0;
2018 }
2019
2020 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2021 {
2022         struct sock *sk = sock->sk;
2023         long amount = 0;
2024         int err;
2025
2026         switch (cmd) {
2027         case SIOCOUTQ:
2028                 amount = sk_wmem_alloc_get(sk);
2029                 err = put_user(amount, (int __user *)arg);
2030                 break;
2031         case SIOCINQ:
2032                 {
2033                         struct sk_buff *skb;
2034
2035                         if (sk->sk_state == TCP_LISTEN) {
2036                                 err = -EINVAL;
2037                                 break;
2038                         }
2039
2040                         spin_lock(&sk->sk_receive_queue.lock);
2041                         if (sk->sk_type == SOCK_STREAM ||
2042                             sk->sk_type == SOCK_SEQPACKET) {
2043                                 skb_queue_walk(&sk->sk_receive_queue, skb)
2044                                         amount += skb->len;
2045                         } else {
2046                                 skb = skb_peek(&sk->sk_receive_queue);
2047                                 if (skb)
2048                                         amount = skb->len;
2049                         }
2050                         spin_unlock(&sk->sk_receive_queue.lock);
2051                         err = put_user(amount, (int __user *)arg);
2052                         break;
2053                 }
2054
2055         default:
2056                 err = -ENOIOCTLCMD;
2057                 break;
2058         }
2059         return err;
2060 }
2061
2062 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2063 {
2064         struct sock *sk = sock->sk;
2065         unsigned int mask;
2066
2067         sock_poll_wait(file, sk_sleep(sk), wait);
2068         mask = 0;
2069
2070         /* exceptional events? */
2071         if (sk->sk_err)
2072                 mask |= POLLERR;
2073         if (sk->sk_shutdown == SHUTDOWN_MASK)
2074                 mask |= POLLHUP;
2075         if (sk->sk_shutdown & RCV_SHUTDOWN)
2076                 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2077
2078         /* readable? */
2079         if (!skb_queue_empty(&sk->sk_receive_queue))
2080                 mask |= POLLIN | POLLRDNORM;
2081
2082         /* Connection-based need to check for termination and startup */
2083         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2084             sk->sk_state == TCP_CLOSE)
2085                 mask |= POLLHUP;
2086
2087         /*
2088          * we set writable also when the other side has shut down the
2089          * connection. This prevents stuck sockets.
2090          */
2091         if (unix_writable(sk))
2092                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2093
2094         return mask;
2095 }
2096
2097 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2098                                     poll_table *wait)
2099 {
2100         struct sock *sk = sock->sk, *other;
2101         unsigned int mask, writable;
2102
2103         sock_poll_wait(file, sk_sleep(sk), wait);
2104         mask = 0;
2105
2106         /* exceptional events? */
2107         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2108                 mask |= POLLERR;
2109         if (sk->sk_shutdown & RCV_SHUTDOWN)
2110                 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2111         if (sk->sk_shutdown == SHUTDOWN_MASK)
2112                 mask |= POLLHUP;
2113
2114         /* readable? */
2115         if (!skb_queue_empty(&sk->sk_receive_queue))
2116                 mask |= POLLIN | POLLRDNORM;
2117
2118         /* Connection-based need to check for termination and startup */
2119         if (sk->sk_type == SOCK_SEQPACKET) {
2120                 if (sk->sk_state == TCP_CLOSE)
2121                         mask |= POLLHUP;
2122                 /* connection hasn't started yet? */
2123                 if (sk->sk_state == TCP_SYN_SENT)
2124                         return mask;
2125         }
2126
2127         /* No write status requested, avoid expensive OUT tests. */
2128         if (wait && !(wait->key & (POLLWRBAND | POLLWRNORM | POLLOUT)))
2129                 return mask;
2130
2131         writable = unix_writable(sk);
2132         other = unix_peer_get(sk);
2133         if (other) {
2134                 if (unix_peer(other) != sk) {
2135                         sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2136                         if (unix_recvq_full(other))
2137                                 writable = 0;
2138                 }
2139                 sock_put(other);
2140         }
2141
2142         if (writable)
2143                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2144         else
2145                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2146
2147         return mask;
2148 }
2149
2150 #ifdef CONFIG_PROC_FS
2151 static struct sock *first_unix_socket(int *i)
2152 {
2153         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2154                 if (!hlist_empty(&unix_socket_table[*i]))
2155                         return __sk_head(&unix_socket_table[*i]);
2156         }
2157         return NULL;
2158 }
2159
2160 static struct sock *next_unix_socket(int *i, struct sock *s)
2161 {
2162         struct sock *next = sk_next(s);
2163         /* More in this chain? */
2164         if (next)
2165                 return next;
2166         /* Look for next non-empty chain. */
2167         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2168                 if (!hlist_empty(&unix_socket_table[*i]))
2169                         return __sk_head(&unix_socket_table[*i]);
2170         }
2171         return NULL;
2172 }
2173
2174 struct unix_iter_state {
2175         struct seq_net_private p;
2176         int i;
2177 };
2178
2179 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2180 {
2181         struct unix_iter_state *iter = seq->private;
2182         loff_t off = 0;
2183         struct sock *s;
2184
2185         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2186                 if (sock_net(s) != seq_file_net(seq))
2187                         continue;
2188                 if (off == pos)
2189                         return s;
2190                 ++off;
2191         }
2192         return NULL;
2193 }
2194
2195 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2196         __acquires(unix_table_lock)
2197 {
2198         spin_lock(&unix_table_lock);
2199         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2200 }
2201
2202 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2203 {
2204         struct unix_iter_state *iter = seq->private;
2205         struct sock *sk = v;
2206         ++*pos;
2207
2208         if (v == SEQ_START_TOKEN)
2209                 sk = first_unix_socket(&iter->i);
2210         else
2211                 sk = next_unix_socket(&iter->i, sk);
2212         while (sk && (sock_net(sk) != seq_file_net(seq)))
2213                 sk = next_unix_socket(&iter->i, sk);
2214         return sk;
2215 }
2216
2217 static void unix_seq_stop(struct seq_file *seq, void *v)
2218         __releases(unix_table_lock)
2219 {
2220         spin_unlock(&unix_table_lock);
2221 }
2222
2223 static int unix_seq_show(struct seq_file *seq, void *v)
2224 {
2225
2226         if (v == SEQ_START_TOKEN)
2227                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2228                          "Inode Path\n");
2229         else {
2230                 struct sock *s = v;
2231                 struct unix_sock *u = unix_sk(s);
2232                 unix_state_lock(s);
2233
2234                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2235                         s,
2236                         atomic_read(&s->sk_refcnt),
2237                         0,
2238                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2239                         s->sk_type,
2240                         s->sk_socket ?
2241                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2242                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2243                         sock_i_ino(s));
2244
2245                 if (u->addr) {
2246                         int i, len;
2247                         seq_putc(seq, ' ');
2248
2249                         i = 0;
2250                         len = u->addr->len - sizeof(short);
2251                         if (!UNIX_ABSTRACT(s))
2252                                 len--;
2253                         else {
2254                                 seq_putc(seq, '@');
2255                                 i++;
2256                         }
2257                         for ( ; i < len; i++)
2258                                 seq_putc(seq, u->addr->name->sun_path[i]);
2259                 }
2260                 unix_state_unlock(s);
2261                 seq_putc(seq, '\n');
2262         }
2263
2264         return 0;
2265 }
2266
2267 static const struct seq_operations unix_seq_ops = {
2268         .start  = unix_seq_start,
2269         .next   = unix_seq_next,
2270         .stop   = unix_seq_stop,
2271         .show   = unix_seq_show,
2272 };
2273
2274 static int unix_seq_open(struct inode *inode, struct file *file)
2275 {
2276         return seq_open_net(inode, file, &unix_seq_ops,
2277                             sizeof(struct unix_iter_state));
2278 }
2279
2280 static const struct file_operations unix_seq_fops = {
2281         .owner          = THIS_MODULE,
2282         .open           = unix_seq_open,
2283         .read           = seq_read,
2284         .llseek         = seq_lseek,
2285         .release        = seq_release_net,
2286 };
2287
2288 #endif
2289
2290 static const struct net_proto_family unix_family_ops = {
2291         .family = PF_UNIX,
2292         .create = unix_create,
2293         .owner  = THIS_MODULE,
2294 };
2295
2296
2297 static int __net_init unix_net_init(struct net *net)
2298 {
2299         int error = -ENOMEM;
2300
2301         net->unx.sysctl_max_dgram_qlen = 10;
2302         if (unix_sysctl_register(net))
2303                 goto out;
2304
2305 #ifdef CONFIG_PROC_FS
2306         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2307                 unix_sysctl_unregister(net);
2308                 goto out;
2309         }
2310 #endif
2311         error = 0;
2312 out:
2313         return error;
2314 }
2315
2316 static void __net_exit unix_net_exit(struct net *net)
2317 {
2318         unix_sysctl_unregister(net);
2319         proc_net_remove(net, "unix");
2320 }
2321
2322 static struct pernet_operations unix_net_ops = {
2323         .init = unix_net_init,
2324         .exit = unix_net_exit,
2325 };
2326
2327 static int __init af_unix_init(void)
2328 {
2329         int rc = -1;
2330         struct sk_buff *dummy_skb;
2331
2332         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2333
2334         rc = proto_register(&unix_proto, 1);
2335         if (rc != 0) {
2336                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2337                        __func__);
2338                 goto out;
2339         }
2340
2341         sock_register(&unix_family_ops);
2342         register_pernet_subsys(&unix_net_ops);
2343 out:
2344         return rc;
2345 }
2346
2347 static void __exit af_unix_exit(void)
2348 {
2349         sock_unregister(PF_UNIX);
2350         proto_unregister(&unix_proto);
2351         unregister_pernet_subsys(&unix_net_ops);
2352 }
2353
2354 /* Earlier than device_initcall() so that other drivers invoking
2355    request_module() don't end up in a loop when modprobe tries
2356    to use a UNIX socket. But later than subsys_initcall() because
2357    we depend on stuff initialised there */
2358 fs_initcall(af_unix_init);
2359 module_exit(af_unix_exit);
2360
2361 MODULE_LICENSE("GPL");
2362 MODULE_ALIAS_NETPROTO(PF_UNIX);