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