af_unix: Allow SO_PEERCRED to work across namespaces.
[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_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 void init_peercred(struct sock *sk)
454 {
455         put_pid(sk->sk_peer_pid);
456         if (sk->sk_peer_cred)
457                 put_cred(sk->sk_peer_cred);
458         sk->sk_peer_pid  = get_pid(task_tgid(current));
459         sk->sk_peer_cred = get_current_cred();
460 }
461
462 static void copy_peercred(struct sock *sk, struct sock *peersk)
463 {
464         put_pid(sk->sk_peer_pid);
465         if (sk->sk_peer_cred)
466                 put_cred(sk->sk_peer_cred);
467         sk->sk_peer_pid  = get_pid(peersk->sk_peer_pid);
468         sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
469 }
470
471 static int unix_listen(struct socket *sock, int backlog)
472 {
473         int err;
474         struct sock *sk = sock->sk;
475         struct unix_sock *u = unix_sk(sk);
476         struct pid *old_pid = NULL;
477         const struct cred *old_cred = NULL;
478
479         err = -EOPNOTSUPP;
480         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
481                 goto out;       /* Only stream/seqpacket sockets accept */
482         err = -EINVAL;
483         if (!u->addr)
484                 goto out;       /* No listens on an unbound socket */
485         unix_state_lock(sk);
486         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
487                 goto out_unlock;
488         if (backlog > sk->sk_max_ack_backlog)
489                 wake_up_interruptible_all(&u->peer_wait);
490         sk->sk_max_ack_backlog  = backlog;
491         sk->sk_state            = TCP_LISTEN;
492         /* set credentials so connect can copy them */
493         init_peercred(sk);
494         err = 0;
495
496 out_unlock:
497         unix_state_unlock(sk);
498         put_pid(old_pid);
499         if (old_cred)
500                 put_cred(old_cred);
501 out:
502         return err;
503 }
504
505 static int unix_release(struct socket *);
506 static int unix_bind(struct socket *, struct sockaddr *, int);
507 static int unix_stream_connect(struct socket *, struct sockaddr *,
508                                int addr_len, int flags);
509 static int unix_socketpair(struct socket *, struct socket *);
510 static int unix_accept(struct socket *, struct socket *, int);
511 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
512 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
513 static unsigned int unix_dgram_poll(struct file *, struct socket *,
514                                     poll_table *);
515 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
516 static int unix_shutdown(struct socket *, int);
517 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
518                                struct msghdr *, size_t);
519 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
520                                struct msghdr *, size_t, int);
521 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
522                               struct msghdr *, size_t);
523 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
524                               struct msghdr *, size_t, int);
525 static int unix_dgram_connect(struct socket *, struct sockaddr *,
526                               int, int);
527 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
528                                   struct msghdr *, size_t);
529
530 static const struct proto_ops unix_stream_ops = {
531         .family =       PF_UNIX,
532         .owner =        THIS_MODULE,
533         .release =      unix_release,
534         .bind =         unix_bind,
535         .connect =      unix_stream_connect,
536         .socketpair =   unix_socketpair,
537         .accept =       unix_accept,
538         .getname =      unix_getname,
539         .poll =         unix_poll,
540         .ioctl =        unix_ioctl,
541         .listen =       unix_listen,
542         .shutdown =     unix_shutdown,
543         .setsockopt =   sock_no_setsockopt,
544         .getsockopt =   sock_no_getsockopt,
545         .sendmsg =      unix_stream_sendmsg,
546         .recvmsg =      unix_stream_recvmsg,
547         .mmap =         sock_no_mmap,
548         .sendpage =     sock_no_sendpage,
549 };
550
551 static const struct proto_ops unix_dgram_ops = {
552         .family =       PF_UNIX,
553         .owner =        THIS_MODULE,
554         .release =      unix_release,
555         .bind =         unix_bind,
556         .connect =      unix_dgram_connect,
557         .socketpair =   unix_socketpair,
558         .accept =       sock_no_accept,
559         .getname =      unix_getname,
560         .poll =         unix_dgram_poll,
561         .ioctl =        unix_ioctl,
562         .listen =       sock_no_listen,
563         .shutdown =     unix_shutdown,
564         .setsockopt =   sock_no_setsockopt,
565         .getsockopt =   sock_no_getsockopt,
566         .sendmsg =      unix_dgram_sendmsg,
567         .recvmsg =      unix_dgram_recvmsg,
568         .mmap =         sock_no_mmap,
569         .sendpage =     sock_no_sendpage,
570 };
571
572 static const struct proto_ops unix_seqpacket_ops = {
573         .family =       PF_UNIX,
574         .owner =        THIS_MODULE,
575         .release =      unix_release,
576         .bind =         unix_bind,
577         .connect =      unix_stream_connect,
578         .socketpair =   unix_socketpair,
579         .accept =       unix_accept,
580         .getname =      unix_getname,
581         .poll =         unix_dgram_poll,
582         .ioctl =        unix_ioctl,
583         .listen =       unix_listen,
584         .shutdown =     unix_shutdown,
585         .setsockopt =   sock_no_setsockopt,
586         .getsockopt =   sock_no_getsockopt,
587         .sendmsg =      unix_seqpacket_sendmsg,
588         .recvmsg =      unix_dgram_recvmsg,
589         .mmap =         sock_no_mmap,
590         .sendpage =     sock_no_sendpage,
591 };
592
593 static struct proto unix_proto = {
594         .name                   = "UNIX",
595         .owner                  = THIS_MODULE,
596         .obj_size               = sizeof(struct unix_sock),
597 };
598
599 /*
600  * AF_UNIX sockets do not interact with hardware, hence they
601  * dont trigger interrupts - so it's safe for them to have
602  * bh-unsafe locking for their sk_receive_queue.lock. Split off
603  * this special lock-class by reinitializing the spinlock key:
604  */
605 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
606
607 static struct sock *unix_create1(struct net *net, struct socket *sock)
608 {
609         struct sock *sk = NULL;
610         struct unix_sock *u;
611
612         atomic_inc(&unix_nr_socks);
613         if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
614                 goto out;
615
616         sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
617         if (!sk)
618                 goto out;
619
620         sock_init_data(sock, sk);
621         lockdep_set_class(&sk->sk_receive_queue.lock,
622                                 &af_unix_sk_receive_queue_lock_key);
623
624         sk->sk_write_space      = unix_write_space;
625         sk->sk_max_ack_backlog  = net->unx.sysctl_max_dgram_qlen;
626         sk->sk_destruct         = unix_sock_destructor;
627         u         = unix_sk(sk);
628         u->dentry = NULL;
629         u->mnt    = NULL;
630         spin_lock_init(&u->lock);
631         atomic_long_set(&u->inflight, 0);
632         INIT_LIST_HEAD(&u->link);
633         mutex_init(&u->readlock); /* single task reading lock */
634         init_waitqueue_head(&u->peer_wait);
635         unix_insert_socket(unix_sockets_unbound, sk);
636 out:
637         if (sk == NULL)
638                 atomic_dec(&unix_nr_socks);
639         else {
640                 local_bh_disable();
641                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
642                 local_bh_enable();
643         }
644         return sk;
645 }
646
647 static int unix_create(struct net *net, struct socket *sock, int protocol,
648                        int kern)
649 {
650         if (protocol && protocol != PF_UNIX)
651                 return -EPROTONOSUPPORT;
652
653         sock->state = SS_UNCONNECTED;
654
655         switch (sock->type) {
656         case SOCK_STREAM:
657                 sock->ops = &unix_stream_ops;
658                 break;
659                 /*
660                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
661                  *      nothing uses it.
662                  */
663         case SOCK_RAW:
664                 sock->type = SOCK_DGRAM;
665         case SOCK_DGRAM:
666                 sock->ops = &unix_dgram_ops;
667                 break;
668         case SOCK_SEQPACKET:
669                 sock->ops = &unix_seqpacket_ops;
670                 break;
671         default:
672                 return -ESOCKTNOSUPPORT;
673         }
674
675         return unix_create1(net, sock) ? 0 : -ENOMEM;
676 }
677
678 static int unix_release(struct socket *sock)
679 {
680         struct sock *sk = sock->sk;
681
682         if (!sk)
683                 return 0;
684
685         sock->sk = NULL;
686
687         return unix_release_sock(sk, 0);
688 }
689
690 static int unix_autobind(struct socket *sock)
691 {
692         struct sock *sk = sock->sk;
693         struct net *net = sock_net(sk);
694         struct unix_sock *u = unix_sk(sk);
695         static u32 ordernum = 1;
696         struct unix_address *addr;
697         int err;
698
699         mutex_lock(&u->readlock);
700
701         err = 0;
702         if (u->addr)
703                 goto out;
704
705         err = -ENOMEM;
706         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
707         if (!addr)
708                 goto out;
709
710         addr->name->sun_family = AF_UNIX;
711         atomic_set(&addr->refcnt, 1);
712
713 retry:
714         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
715         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
716
717         spin_lock(&unix_table_lock);
718         ordernum = (ordernum+1)&0xFFFFF;
719
720         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
721                                       addr->hash)) {
722                 spin_unlock(&unix_table_lock);
723                 /* Sanity yield. It is unusual case, but yet... */
724                 if (!(ordernum&0xFF))
725                         yield();
726                 goto retry;
727         }
728         addr->hash ^= sk->sk_type;
729
730         __unix_remove_socket(sk);
731         u->addr = addr;
732         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
733         spin_unlock(&unix_table_lock);
734         err = 0;
735
736 out:    mutex_unlock(&u->readlock);
737         return err;
738 }
739
740 static struct sock *unix_find_other(struct net *net,
741                                     struct sockaddr_un *sunname, int len,
742                                     int type, unsigned hash, int *error)
743 {
744         struct sock *u;
745         struct path path;
746         int err = 0;
747
748         if (sunname->sun_path[0]) {
749                 struct inode *inode;
750                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
751                 if (err)
752                         goto fail;
753                 inode = path.dentry->d_inode;
754                 err = inode_permission(inode, MAY_WRITE);
755                 if (err)
756                         goto put_fail;
757
758                 err = -ECONNREFUSED;
759                 if (!S_ISSOCK(inode->i_mode))
760                         goto put_fail;
761                 u = unix_find_socket_byinode(net, inode);
762                 if (!u)
763                         goto put_fail;
764
765                 if (u->sk_type == type)
766                         touch_atime(path.mnt, path.dentry);
767
768                 path_put(&path);
769
770                 err = -EPROTOTYPE;
771                 if (u->sk_type != type) {
772                         sock_put(u);
773                         goto fail;
774                 }
775         } else {
776                 err = -ECONNREFUSED;
777                 u = unix_find_socket_byname(net, sunname, len, type, hash);
778                 if (u) {
779                         struct dentry *dentry;
780                         dentry = unix_sk(u)->dentry;
781                         if (dentry)
782                                 touch_atime(unix_sk(u)->mnt, dentry);
783                 } else
784                         goto fail;
785         }
786         return u;
787
788 put_fail:
789         path_put(&path);
790 fail:
791         *error = err;
792         return NULL;
793 }
794
795
796 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
797 {
798         struct sock *sk = sock->sk;
799         struct net *net = sock_net(sk);
800         struct unix_sock *u = unix_sk(sk);
801         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
802         struct dentry *dentry = NULL;
803         struct nameidata nd;
804         int err;
805         unsigned hash;
806         struct unix_address *addr;
807         struct hlist_head *list;
808
809         err = -EINVAL;
810         if (sunaddr->sun_family != AF_UNIX)
811                 goto out;
812
813         if (addr_len == sizeof(short)) {
814                 err = unix_autobind(sock);
815                 goto out;
816         }
817
818         err = unix_mkname(sunaddr, addr_len, &hash);
819         if (err < 0)
820                 goto out;
821         addr_len = err;
822
823         mutex_lock(&u->readlock);
824
825         err = -EINVAL;
826         if (u->addr)
827                 goto out_up;
828
829         err = -ENOMEM;
830         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
831         if (!addr)
832                 goto out_up;
833
834         memcpy(addr->name, sunaddr, addr_len);
835         addr->len = addr_len;
836         addr->hash = hash ^ sk->sk_type;
837         atomic_set(&addr->refcnt, 1);
838
839         if (sunaddr->sun_path[0]) {
840                 unsigned int mode;
841                 err = 0;
842                 /*
843                  * Get the parent directory, calculate the hash for last
844                  * component.
845                  */
846                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
847                 if (err)
848                         goto out_mknod_parent;
849
850                 dentry = lookup_create(&nd, 0);
851                 err = PTR_ERR(dentry);
852                 if (IS_ERR(dentry))
853                         goto out_mknod_unlock;
854
855                 /*
856                  * All right, let's create it.
857                  */
858                 mode = S_IFSOCK |
859                        (SOCK_INODE(sock)->i_mode & ~current_umask());
860                 err = mnt_want_write(nd.path.mnt);
861                 if (err)
862                         goto out_mknod_dput;
863                 err = security_path_mknod(&nd.path, dentry, mode, 0);
864                 if (err)
865                         goto out_mknod_drop_write;
866                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
867 out_mknod_drop_write:
868                 mnt_drop_write(nd.path.mnt);
869                 if (err)
870                         goto out_mknod_dput;
871                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
872                 dput(nd.path.dentry);
873                 nd.path.dentry = dentry;
874
875                 addr->hash = UNIX_HASH_SIZE;
876         }
877
878         spin_lock(&unix_table_lock);
879
880         if (!sunaddr->sun_path[0]) {
881                 err = -EADDRINUSE;
882                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
883                                               sk->sk_type, hash)) {
884                         unix_release_addr(addr);
885                         goto out_unlock;
886                 }
887
888                 list = &unix_socket_table[addr->hash];
889         } else {
890                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
891                 u->dentry = nd.path.dentry;
892                 u->mnt    = nd.path.mnt;
893         }
894
895         err = 0;
896         __unix_remove_socket(sk);
897         u->addr = addr;
898         __unix_insert_socket(list, sk);
899
900 out_unlock:
901         spin_unlock(&unix_table_lock);
902 out_up:
903         mutex_unlock(&u->readlock);
904 out:
905         return err;
906
907 out_mknod_dput:
908         dput(dentry);
909 out_mknod_unlock:
910         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
911         path_put(&nd.path);
912 out_mknod_parent:
913         if (err == -EEXIST)
914                 err = -EADDRINUSE;
915         unix_release_addr(addr);
916         goto out_up;
917 }
918
919 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
920 {
921         if (unlikely(sk1 == sk2) || !sk2) {
922                 unix_state_lock(sk1);
923                 return;
924         }
925         if (sk1 < sk2) {
926                 unix_state_lock(sk1);
927                 unix_state_lock_nested(sk2);
928         } else {
929                 unix_state_lock(sk2);
930                 unix_state_lock_nested(sk1);
931         }
932 }
933
934 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
935 {
936         if (unlikely(sk1 == sk2) || !sk2) {
937                 unix_state_unlock(sk1);
938                 return;
939         }
940         unix_state_unlock(sk1);
941         unix_state_unlock(sk2);
942 }
943
944 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
945                               int alen, int flags)
946 {
947         struct sock *sk = sock->sk;
948         struct net *net = sock_net(sk);
949         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
950         struct sock *other;
951         unsigned hash;
952         int err;
953
954         if (addr->sa_family != AF_UNSPEC) {
955                 err = unix_mkname(sunaddr, alen, &hash);
956                 if (err < 0)
957                         goto out;
958                 alen = err;
959
960                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
961                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
962                         goto out;
963
964 restart:
965                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
966                 if (!other)
967                         goto out;
968
969                 unix_state_double_lock(sk, other);
970
971                 /* Apparently VFS overslept socket death. Retry. */
972                 if (sock_flag(other, SOCK_DEAD)) {
973                         unix_state_double_unlock(sk, other);
974                         sock_put(other);
975                         goto restart;
976                 }
977
978                 err = -EPERM;
979                 if (!unix_may_send(sk, other))
980                         goto out_unlock;
981
982                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
983                 if (err)
984                         goto out_unlock;
985
986         } else {
987                 /*
988                  *      1003.1g breaking connected state with AF_UNSPEC
989                  */
990                 other = NULL;
991                 unix_state_double_lock(sk, other);
992         }
993
994         /*
995          * If it was connected, reconnect.
996          */
997         if (unix_peer(sk)) {
998                 struct sock *old_peer = unix_peer(sk);
999                 unix_peer(sk) = other;
1000                 unix_state_double_unlock(sk, other);
1001
1002                 if (other != old_peer)
1003                         unix_dgram_disconnected(sk, old_peer);
1004                 sock_put(old_peer);
1005         } else {
1006                 unix_peer(sk) = other;
1007                 unix_state_double_unlock(sk, other);
1008         }
1009         return 0;
1010
1011 out_unlock:
1012         unix_state_double_unlock(sk, other);
1013         sock_put(other);
1014 out:
1015         return err;
1016 }
1017
1018 static long unix_wait_for_peer(struct sock *other, long timeo)
1019 {
1020         struct unix_sock *u = unix_sk(other);
1021         int sched;
1022         DEFINE_WAIT(wait);
1023
1024         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1025
1026         sched = !sock_flag(other, SOCK_DEAD) &&
1027                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1028                 unix_recvq_full(other);
1029
1030         unix_state_unlock(other);
1031
1032         if (sched)
1033                 timeo = schedule_timeout(timeo);
1034
1035         finish_wait(&u->peer_wait, &wait);
1036         return timeo;
1037 }
1038
1039 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1040                                int addr_len, int flags)
1041 {
1042         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1043         struct sock *sk = sock->sk;
1044         struct net *net = sock_net(sk);
1045         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1046         struct sock *newsk = NULL;
1047         struct sock *other = NULL;
1048         struct sk_buff *skb = NULL;
1049         unsigned hash;
1050         int st;
1051         int err;
1052         long timeo;
1053
1054         err = unix_mkname(sunaddr, addr_len, &hash);
1055         if (err < 0)
1056                 goto out;
1057         addr_len = err;
1058
1059         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1060             (err = unix_autobind(sock)) != 0)
1061                 goto out;
1062
1063         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1064
1065         /* First of all allocate resources.
1066            If we will make it after state is locked,
1067            we will have to recheck all again in any case.
1068          */
1069
1070         err = -ENOMEM;
1071
1072         /* create new sock for complete connection */
1073         newsk = unix_create1(sock_net(sk), NULL);
1074         if (newsk == NULL)
1075                 goto out;
1076
1077         /* Allocate skb for sending to listening sock */
1078         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1079         if (skb == NULL)
1080                 goto out;
1081
1082 restart:
1083         /*  Find listening sock. */
1084         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1085         if (!other)
1086                 goto out;
1087
1088         /* Latch state of peer */
1089         unix_state_lock(other);
1090
1091         /* Apparently VFS overslept socket death. Retry. */
1092         if (sock_flag(other, SOCK_DEAD)) {
1093                 unix_state_unlock(other);
1094                 sock_put(other);
1095                 goto restart;
1096         }
1097
1098         err = -ECONNREFUSED;
1099         if (other->sk_state != TCP_LISTEN)
1100                 goto out_unlock;
1101         if (other->sk_shutdown & RCV_SHUTDOWN)
1102                 goto out_unlock;
1103
1104         if (unix_recvq_full(other)) {
1105                 err = -EAGAIN;
1106                 if (!timeo)
1107                         goto out_unlock;
1108
1109                 timeo = unix_wait_for_peer(other, timeo);
1110
1111                 err = sock_intr_errno(timeo);
1112                 if (signal_pending(current))
1113                         goto out;
1114                 sock_put(other);
1115                 goto restart;
1116         }
1117
1118         /* Latch our state.
1119
1120            It is tricky place. We need to grab write lock and cannot
1121            drop lock on peer. It is dangerous because deadlock is
1122            possible. Connect to self case and simultaneous
1123            attempt to connect are eliminated by checking socket
1124            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1125            check this before attempt to grab lock.
1126
1127            Well, and we have to recheck the state after socket locked.
1128          */
1129         st = sk->sk_state;
1130
1131         switch (st) {
1132         case TCP_CLOSE:
1133                 /* This is ok... continue with connect */
1134                 break;
1135         case TCP_ESTABLISHED:
1136                 /* Socket is already connected */
1137                 err = -EISCONN;
1138                 goto out_unlock;
1139         default:
1140                 err = -EINVAL;
1141                 goto out_unlock;
1142         }
1143
1144         unix_state_lock_nested(sk);
1145
1146         if (sk->sk_state != st) {
1147                 unix_state_unlock(sk);
1148                 unix_state_unlock(other);
1149                 sock_put(other);
1150                 goto restart;
1151         }
1152
1153         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1154         if (err) {
1155                 unix_state_unlock(sk);
1156                 goto out_unlock;
1157         }
1158
1159         /* The way is open! Fastly set all the necessary fields... */
1160
1161         sock_hold(sk);
1162         unix_peer(newsk)        = sk;
1163         newsk->sk_state         = TCP_ESTABLISHED;
1164         newsk->sk_type          = sk->sk_type;
1165         init_peercred(newsk);
1166         newu = unix_sk(newsk);
1167         newsk->sk_wq            = &newu->peer_wq;
1168         otheru = unix_sk(other);
1169
1170         /* copy address information from listening to new sock*/
1171         if (otheru->addr) {
1172                 atomic_inc(&otheru->addr->refcnt);
1173                 newu->addr = otheru->addr;
1174         }
1175         if (otheru->dentry) {
1176                 newu->dentry    = dget(otheru->dentry);
1177                 newu->mnt       = mntget(otheru->mnt);
1178         }
1179
1180         /* Set credentials */
1181         copy_peercred(sk, other);
1182
1183         sock->state     = SS_CONNECTED;
1184         sk->sk_state    = TCP_ESTABLISHED;
1185         sock_hold(newsk);
1186
1187         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1188         unix_peer(sk)   = newsk;
1189
1190         unix_state_unlock(sk);
1191
1192         /* take ten and and send info to listening sock */
1193         spin_lock(&other->sk_receive_queue.lock);
1194         __skb_queue_tail(&other->sk_receive_queue, skb);
1195         spin_unlock(&other->sk_receive_queue.lock);
1196         unix_state_unlock(other);
1197         other->sk_data_ready(other, 0);
1198         sock_put(other);
1199         return 0;
1200
1201 out_unlock:
1202         if (other)
1203                 unix_state_unlock(other);
1204
1205 out:
1206         kfree_skb(skb);
1207         if (newsk)
1208                 unix_release_sock(newsk, 0);
1209         if (other)
1210                 sock_put(other);
1211         return err;
1212 }
1213
1214 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1215 {
1216         struct sock *ska = socka->sk, *skb = sockb->sk;
1217
1218         /* Join our sockets back to back */
1219         sock_hold(ska);
1220         sock_hold(skb);
1221         unix_peer(ska) = skb;
1222         unix_peer(skb) = ska;
1223         init_peercred(ska);
1224         init_peercred(skb);
1225
1226         if (ska->sk_type != SOCK_DGRAM) {
1227                 ska->sk_state = TCP_ESTABLISHED;
1228                 skb->sk_state = TCP_ESTABLISHED;
1229                 socka->state  = SS_CONNECTED;
1230                 sockb->state  = SS_CONNECTED;
1231         }
1232         return 0;
1233 }
1234
1235 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1236 {
1237         struct sock *sk = sock->sk;
1238         struct sock *tsk;
1239         struct sk_buff *skb;
1240         int err;
1241
1242         err = -EOPNOTSUPP;
1243         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1244                 goto out;
1245
1246         err = -EINVAL;
1247         if (sk->sk_state != TCP_LISTEN)
1248                 goto out;
1249
1250         /* If socket state is TCP_LISTEN it cannot change (for now...),
1251          * so that no locks are necessary.
1252          */
1253
1254         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1255         if (!skb) {
1256                 /* This means receive shutdown. */
1257                 if (err == 0)
1258                         err = -EINVAL;
1259                 goto out;
1260         }
1261
1262         tsk = skb->sk;
1263         skb_free_datagram(sk, skb);
1264         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1265
1266         /* attach accepted sock to socket */
1267         unix_state_lock(tsk);
1268         newsock->state = SS_CONNECTED;
1269         sock_graft(tsk, newsock);
1270         unix_state_unlock(tsk);
1271         return 0;
1272
1273 out:
1274         return err;
1275 }
1276
1277
1278 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1279 {
1280         struct sock *sk = sock->sk;
1281         struct unix_sock *u;
1282         DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1283         int err = 0;
1284
1285         if (peer) {
1286                 sk = unix_peer_get(sk);
1287
1288                 err = -ENOTCONN;
1289                 if (!sk)
1290                         goto out;
1291                 err = 0;
1292         } else {
1293                 sock_hold(sk);
1294         }
1295
1296         u = unix_sk(sk);
1297         unix_state_lock(sk);
1298         if (!u->addr) {
1299                 sunaddr->sun_family = AF_UNIX;
1300                 sunaddr->sun_path[0] = 0;
1301                 *uaddr_len = sizeof(short);
1302         } else {
1303                 struct unix_address *addr = u->addr;
1304
1305                 *uaddr_len = addr->len;
1306                 memcpy(sunaddr, addr->name, *uaddr_len);
1307         }
1308         unix_state_unlock(sk);
1309         sock_put(sk);
1310 out:
1311         return err;
1312 }
1313
1314 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1315 {
1316         int i;
1317
1318         scm->fp = UNIXCB(skb).fp;
1319         skb->destructor = sock_wfree;
1320         UNIXCB(skb).fp = NULL;
1321
1322         for (i = scm->fp->count-1; i >= 0; i--)
1323                 unix_notinflight(scm->fp->fp[i]);
1324 }
1325
1326 static void unix_destruct_fds(struct sk_buff *skb)
1327 {
1328         struct scm_cookie scm;
1329         memset(&scm, 0, sizeof(scm));
1330         unix_detach_fds(&scm, skb);
1331
1332         /* Alas, it calls VFS */
1333         /* So fscking what? fput() had been SMP-safe since the last Summer */
1334         scm_destroy(&scm);
1335         sock_wfree(skb);
1336 }
1337
1338 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1339 {
1340         int i;
1341
1342         /*
1343          * Need to duplicate file references for the sake of garbage
1344          * collection.  Otherwise a socket in the fps might become a
1345          * candidate for GC while the skb is not yet queued.
1346          */
1347         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1348         if (!UNIXCB(skb).fp)
1349                 return -ENOMEM;
1350
1351         for (i = scm->fp->count-1; i >= 0; i--)
1352                 unix_inflight(scm->fp->fp[i]);
1353         skb->destructor = unix_destruct_fds;
1354         return 0;
1355 }
1356
1357 /*
1358  *      Send AF_UNIX data.
1359  */
1360
1361 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1362                               struct msghdr *msg, size_t len)
1363 {
1364         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1365         struct sock *sk = sock->sk;
1366         struct net *net = sock_net(sk);
1367         struct unix_sock *u = unix_sk(sk);
1368         struct sockaddr_un *sunaddr = msg->msg_name;
1369         struct sock *other = NULL;
1370         int namelen = 0; /* fake GCC */
1371         int err;
1372         unsigned hash;
1373         struct sk_buff *skb;
1374         long timeo;
1375         struct scm_cookie tmp_scm;
1376
1377         if (NULL == siocb->scm)
1378                 siocb->scm = &tmp_scm;
1379         wait_for_unix_gc();
1380         err = scm_send(sock, msg, siocb->scm);
1381         if (err < 0)
1382                 return err;
1383
1384         err = -EOPNOTSUPP;
1385         if (msg->msg_flags&MSG_OOB)
1386                 goto out;
1387
1388         if (msg->msg_namelen) {
1389                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1390                 if (err < 0)
1391                         goto out;
1392                 namelen = err;
1393         } else {
1394                 sunaddr = NULL;
1395                 err = -ENOTCONN;
1396                 other = unix_peer_get(sk);
1397                 if (!other)
1398                         goto out;
1399         }
1400
1401         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1402             && (err = unix_autobind(sock)) != 0)
1403                 goto out;
1404
1405         err = -EMSGSIZE;
1406         if (len > sk->sk_sndbuf - 32)
1407                 goto out;
1408
1409         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1410         if (skb == NULL)
1411                 goto out;
1412
1413         memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1414         if (siocb->scm->fp) {
1415                 err = unix_attach_fds(siocb->scm, skb);
1416                 if (err)
1417                         goto out_free;
1418         }
1419         unix_get_secdata(siocb->scm, skb);
1420
1421         skb_reset_transport_header(skb);
1422         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1423         if (err)
1424                 goto out_free;
1425
1426         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1427
1428 restart:
1429         if (!other) {
1430                 err = -ECONNRESET;
1431                 if (sunaddr == NULL)
1432                         goto out_free;
1433
1434                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1435                                         hash, &err);
1436                 if (other == NULL)
1437                         goto out_free;
1438         }
1439
1440         unix_state_lock(other);
1441         err = -EPERM;
1442         if (!unix_may_send(sk, other))
1443                 goto out_unlock;
1444
1445         if (sock_flag(other, SOCK_DEAD)) {
1446                 /*
1447                  *      Check with 1003.1g - what should
1448                  *      datagram error
1449                  */
1450                 unix_state_unlock(other);
1451                 sock_put(other);
1452
1453                 err = 0;
1454                 unix_state_lock(sk);
1455                 if (unix_peer(sk) == other) {
1456                         unix_peer(sk) = NULL;
1457                         unix_state_unlock(sk);
1458
1459                         unix_dgram_disconnected(sk, other);
1460                         sock_put(other);
1461                         err = -ECONNREFUSED;
1462                 } else {
1463                         unix_state_unlock(sk);
1464                 }
1465
1466                 other = NULL;
1467                 if (err)
1468                         goto out_free;
1469                 goto restart;
1470         }
1471
1472         err = -EPIPE;
1473         if (other->sk_shutdown & RCV_SHUTDOWN)
1474                 goto out_unlock;
1475
1476         if (sk->sk_type != SOCK_SEQPACKET) {
1477                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1478                 if (err)
1479                         goto out_unlock;
1480         }
1481
1482         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1483                 if (!timeo) {
1484                         err = -EAGAIN;
1485                         goto out_unlock;
1486                 }
1487
1488                 timeo = unix_wait_for_peer(other, timeo);
1489
1490                 err = sock_intr_errno(timeo);
1491                 if (signal_pending(current))
1492                         goto out_free;
1493
1494                 goto restart;
1495         }
1496
1497         skb_queue_tail(&other->sk_receive_queue, skb);
1498         unix_state_unlock(other);
1499         other->sk_data_ready(other, len);
1500         sock_put(other);
1501         scm_destroy(siocb->scm);
1502         return len;
1503
1504 out_unlock:
1505         unix_state_unlock(other);
1506 out_free:
1507         kfree_skb(skb);
1508 out:
1509         if (other)
1510                 sock_put(other);
1511         scm_destroy(siocb->scm);
1512         return err;
1513 }
1514
1515
1516 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1517                                struct msghdr *msg, size_t len)
1518 {
1519         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1520         struct sock *sk = sock->sk;
1521         struct sock *other = NULL;
1522         struct sockaddr_un *sunaddr = msg->msg_name;
1523         int err, size;
1524         struct sk_buff *skb;
1525         int sent = 0;
1526         struct scm_cookie tmp_scm;
1527         bool fds_sent = false;
1528
1529         if (NULL == siocb->scm)
1530                 siocb->scm = &tmp_scm;
1531         wait_for_unix_gc();
1532         err = scm_send(sock, msg, siocb->scm);
1533         if (err < 0)
1534                 return err;
1535
1536         err = -EOPNOTSUPP;
1537         if (msg->msg_flags&MSG_OOB)
1538                 goto out_err;
1539
1540         if (msg->msg_namelen) {
1541                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1542                 goto out_err;
1543         } else {
1544                 sunaddr = NULL;
1545                 err = -ENOTCONN;
1546                 other = unix_peer(sk);
1547                 if (!other)
1548                         goto out_err;
1549         }
1550
1551         if (sk->sk_shutdown & SEND_SHUTDOWN)
1552                 goto pipe_err;
1553
1554         while (sent < len) {
1555                 /*
1556                  *      Optimisation for the fact that under 0.01% of X
1557                  *      messages typically need breaking up.
1558                  */
1559
1560                 size = len-sent;
1561
1562                 /* Keep two messages in the pipe so it schedules better */
1563                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1564                         size = (sk->sk_sndbuf >> 1) - 64;
1565
1566                 if (size > SKB_MAX_ALLOC)
1567                         size = SKB_MAX_ALLOC;
1568
1569                 /*
1570                  *      Grab a buffer
1571                  */
1572
1573                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1574                                           &err);
1575
1576                 if (skb == NULL)
1577                         goto out_err;
1578
1579                 /*
1580                  *      If you pass two values to the sock_alloc_send_skb
1581                  *      it tries to grab the large buffer with GFP_NOFS
1582                  *      (which can fail easily), and if it fails grab the
1583                  *      fallback size buffer which is under a page and will
1584                  *      succeed. [Alan]
1585                  */
1586                 size = min_t(int, size, skb_tailroom(skb));
1587
1588                 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1589                 /* Only send the fds in the first buffer */
1590                 if (siocb->scm->fp && !fds_sent) {
1591                         err = unix_attach_fds(siocb->scm, skb);
1592                         if (err) {
1593                                 kfree_skb(skb);
1594                                 goto out_err;
1595                         }
1596                         fds_sent = true;
1597                 }
1598
1599                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1600                 if (err) {
1601                         kfree_skb(skb);
1602                         goto out_err;
1603                 }
1604
1605                 unix_state_lock(other);
1606
1607                 if (sock_flag(other, SOCK_DEAD) ||
1608                     (other->sk_shutdown & RCV_SHUTDOWN))
1609                         goto pipe_err_free;
1610
1611                 skb_queue_tail(&other->sk_receive_queue, skb);
1612                 unix_state_unlock(other);
1613                 other->sk_data_ready(other, size);
1614                 sent += size;
1615         }
1616
1617         scm_destroy(siocb->scm);
1618         siocb->scm = NULL;
1619
1620         return sent;
1621
1622 pipe_err_free:
1623         unix_state_unlock(other);
1624         kfree_skb(skb);
1625 pipe_err:
1626         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1627                 send_sig(SIGPIPE, current, 0);
1628         err = -EPIPE;
1629 out_err:
1630         scm_destroy(siocb->scm);
1631         siocb->scm = NULL;
1632         return sent ? : err;
1633 }
1634
1635 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1636                                   struct msghdr *msg, size_t len)
1637 {
1638         int err;
1639         struct sock *sk = sock->sk;
1640
1641         err = sock_error(sk);
1642         if (err)
1643                 return err;
1644
1645         if (sk->sk_state != TCP_ESTABLISHED)
1646                 return -ENOTCONN;
1647
1648         if (msg->msg_namelen)
1649                 msg->msg_namelen = 0;
1650
1651         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1652 }
1653
1654 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1655 {
1656         struct unix_sock *u = unix_sk(sk);
1657
1658         msg->msg_namelen = 0;
1659         if (u->addr) {
1660                 msg->msg_namelen = u->addr->len;
1661                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1662         }
1663 }
1664
1665 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1666                               struct msghdr *msg, size_t size,
1667                               int flags)
1668 {
1669         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1670         struct scm_cookie tmp_scm;
1671         struct sock *sk = sock->sk;
1672         struct unix_sock *u = unix_sk(sk);
1673         int noblock = flags & MSG_DONTWAIT;
1674         struct sk_buff *skb;
1675         int err;
1676
1677         err = -EOPNOTSUPP;
1678         if (flags&MSG_OOB)
1679                 goto out;
1680
1681         msg->msg_namelen = 0;
1682
1683         mutex_lock(&u->readlock);
1684
1685         skb = skb_recv_datagram(sk, flags, noblock, &err);
1686         if (!skb) {
1687                 unix_state_lock(sk);
1688                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1689                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1690                     (sk->sk_shutdown & RCV_SHUTDOWN))
1691                         err = 0;
1692                 unix_state_unlock(sk);
1693                 goto out_unlock;
1694         }
1695
1696         wake_up_interruptible_sync(&u->peer_wait);
1697
1698         if (msg->msg_name)
1699                 unix_copy_addr(msg, skb->sk);
1700
1701         if (size > skb->len)
1702                 size = skb->len;
1703         else if (size < skb->len)
1704                 msg->msg_flags |= MSG_TRUNC;
1705
1706         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1707         if (err)
1708                 goto out_free;
1709
1710         if (!siocb->scm) {
1711                 siocb->scm = &tmp_scm;
1712                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1713         }
1714         siocb->scm->creds = *UNIXCREDS(skb);
1715         unix_set_secdata(siocb->scm, skb);
1716
1717         if (!(flags & MSG_PEEK)) {
1718                 if (UNIXCB(skb).fp)
1719                         unix_detach_fds(siocb->scm, skb);
1720         } else {
1721                 /* It is questionable: on PEEK we could:
1722                    - do not return fds - good, but too simple 8)
1723                    - return fds, and do not return them on read (old strategy,
1724                      apparently wrong)
1725                    - clone fds (I chose it for now, it is the most universal
1726                      solution)
1727
1728                    POSIX 1003.1g does not actually define this clearly
1729                    at all. POSIX 1003.1g doesn't define a lot of things
1730                    clearly however!
1731
1732                 */
1733                 if (UNIXCB(skb).fp)
1734                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1735         }
1736         err = size;
1737
1738         scm_recv(sock, msg, siocb->scm, flags);
1739
1740 out_free:
1741         skb_free_datagram(sk, skb);
1742 out_unlock:
1743         mutex_unlock(&u->readlock);
1744 out:
1745         return err;
1746 }
1747
1748 /*
1749  *      Sleep until data has arrive. But check for races..
1750  */
1751
1752 static long unix_stream_data_wait(struct sock *sk, long timeo)
1753 {
1754         DEFINE_WAIT(wait);
1755
1756         unix_state_lock(sk);
1757
1758         for (;;) {
1759                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1760
1761                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1762                     sk->sk_err ||
1763                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1764                     signal_pending(current) ||
1765                     !timeo)
1766                         break;
1767
1768                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1769                 unix_state_unlock(sk);
1770                 timeo = schedule_timeout(timeo);
1771                 unix_state_lock(sk);
1772                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1773         }
1774
1775         finish_wait(sk_sleep(sk), &wait);
1776         unix_state_unlock(sk);
1777         return timeo;
1778 }
1779
1780
1781
1782 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1783                                struct msghdr *msg, size_t size,
1784                                int flags)
1785 {
1786         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1787         struct scm_cookie tmp_scm;
1788         struct sock *sk = sock->sk;
1789         struct unix_sock *u = unix_sk(sk);
1790         struct sockaddr_un *sunaddr = msg->msg_name;
1791         int copied = 0;
1792         int check_creds = 0;
1793         int target;
1794         int err = 0;
1795         long timeo;
1796
1797         err = -EINVAL;
1798         if (sk->sk_state != TCP_ESTABLISHED)
1799                 goto out;
1800
1801         err = -EOPNOTSUPP;
1802         if (flags&MSG_OOB)
1803                 goto out;
1804
1805         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1806         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1807
1808         msg->msg_namelen = 0;
1809
1810         /* Lock the socket to prevent queue disordering
1811          * while sleeps in memcpy_tomsg
1812          */
1813
1814         if (!siocb->scm) {
1815                 siocb->scm = &tmp_scm;
1816                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1817         }
1818
1819         mutex_lock(&u->readlock);
1820
1821         do {
1822                 int chunk;
1823                 struct sk_buff *skb;
1824
1825                 unix_state_lock(sk);
1826                 skb = skb_dequeue(&sk->sk_receive_queue);
1827                 if (skb == NULL) {
1828                         if (copied >= target)
1829                                 goto unlock;
1830
1831                         /*
1832                          *      POSIX 1003.1g mandates this order.
1833                          */
1834
1835                         err = sock_error(sk);
1836                         if (err)
1837                                 goto unlock;
1838                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1839                                 goto unlock;
1840
1841                         unix_state_unlock(sk);
1842                         err = -EAGAIN;
1843                         if (!timeo)
1844                                 break;
1845                         mutex_unlock(&u->readlock);
1846
1847                         timeo = unix_stream_data_wait(sk, timeo);
1848
1849                         if (signal_pending(current)) {
1850                                 err = sock_intr_errno(timeo);
1851                                 goto out;
1852                         }
1853                         mutex_lock(&u->readlock);
1854                         continue;
1855  unlock:
1856                         unix_state_unlock(sk);
1857                         break;
1858                 }
1859                 unix_state_unlock(sk);
1860
1861                 if (check_creds) {
1862                         /* Never glue messages from different writers */
1863                         if (memcmp(UNIXCREDS(skb), &siocb->scm->creds,
1864                                    sizeof(siocb->scm->creds)) != 0) {
1865                                 skb_queue_head(&sk->sk_receive_queue, skb);
1866                                 break;
1867                         }
1868                 } else {
1869                         /* Copy credentials */
1870                         siocb->scm->creds = *UNIXCREDS(skb);
1871                         check_creds = 1;
1872                 }
1873
1874                 /* Copy address just once */
1875                 if (sunaddr) {
1876                         unix_copy_addr(msg, skb->sk);
1877                         sunaddr = NULL;
1878                 }
1879
1880                 chunk = min_t(unsigned int, skb->len, size);
1881                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1882                         skb_queue_head(&sk->sk_receive_queue, skb);
1883                         if (copied == 0)
1884                                 copied = -EFAULT;
1885                         break;
1886                 }
1887                 copied += chunk;
1888                 size -= chunk;
1889
1890                 /* Mark read part of skb as used */
1891                 if (!(flags & MSG_PEEK)) {
1892                         skb_pull(skb, chunk);
1893
1894                         if (UNIXCB(skb).fp)
1895                                 unix_detach_fds(siocb->scm, skb);
1896
1897                         /* put the skb back if we didn't use it up.. */
1898                         if (skb->len) {
1899                                 skb_queue_head(&sk->sk_receive_queue, skb);
1900                                 break;
1901                         }
1902
1903                         kfree_skb(skb);
1904
1905                         if (siocb->scm->fp)
1906                                 break;
1907                 } else {
1908                         /* It is questionable, see note in unix_dgram_recvmsg.
1909                          */
1910                         if (UNIXCB(skb).fp)
1911                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1912
1913                         /* put message back and return */
1914                         skb_queue_head(&sk->sk_receive_queue, skb);
1915                         break;
1916                 }
1917         } while (size);
1918
1919         mutex_unlock(&u->readlock);
1920         scm_recv(sock, msg, siocb->scm, flags);
1921 out:
1922         return copied ? : err;
1923 }
1924
1925 static int unix_shutdown(struct socket *sock, int mode)
1926 {
1927         struct sock *sk = sock->sk;
1928         struct sock *other;
1929
1930         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1931
1932         if (mode) {
1933                 unix_state_lock(sk);
1934                 sk->sk_shutdown |= mode;
1935                 other = unix_peer(sk);
1936                 if (other)
1937                         sock_hold(other);
1938                 unix_state_unlock(sk);
1939                 sk->sk_state_change(sk);
1940
1941                 if (other &&
1942                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1943
1944                         int peer_mode = 0;
1945
1946                         if (mode&RCV_SHUTDOWN)
1947                                 peer_mode |= SEND_SHUTDOWN;
1948                         if (mode&SEND_SHUTDOWN)
1949                                 peer_mode |= RCV_SHUTDOWN;
1950                         unix_state_lock(other);
1951                         other->sk_shutdown |= peer_mode;
1952                         unix_state_unlock(other);
1953                         other->sk_state_change(other);
1954                         if (peer_mode == SHUTDOWN_MASK)
1955                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1956                         else if (peer_mode & RCV_SHUTDOWN)
1957                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1958                 }
1959                 if (other)
1960                         sock_put(other);
1961         }
1962         return 0;
1963 }
1964
1965 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1966 {
1967         struct sock *sk = sock->sk;
1968         long amount = 0;
1969         int err;
1970
1971         switch (cmd) {
1972         case SIOCOUTQ:
1973                 amount = sk_wmem_alloc_get(sk);
1974                 err = put_user(amount, (int __user *)arg);
1975                 break;
1976         case SIOCINQ:
1977                 {
1978                         struct sk_buff *skb;
1979
1980                         if (sk->sk_state == TCP_LISTEN) {
1981                                 err = -EINVAL;
1982                                 break;
1983                         }
1984
1985                         spin_lock(&sk->sk_receive_queue.lock);
1986                         if (sk->sk_type == SOCK_STREAM ||
1987                             sk->sk_type == SOCK_SEQPACKET) {
1988                                 skb_queue_walk(&sk->sk_receive_queue, skb)
1989                                         amount += skb->len;
1990                         } else {
1991                                 skb = skb_peek(&sk->sk_receive_queue);
1992                                 if (skb)
1993                                         amount = skb->len;
1994                         }
1995                         spin_unlock(&sk->sk_receive_queue.lock);
1996                         err = put_user(amount, (int __user *)arg);
1997                         break;
1998                 }
1999
2000         default:
2001                 err = -ENOIOCTLCMD;
2002                 break;
2003         }
2004         return err;
2005 }
2006
2007 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2008 {
2009         struct sock *sk = sock->sk;
2010         unsigned int mask;
2011
2012         sock_poll_wait(file, sk_sleep(sk), wait);
2013         mask = 0;
2014
2015         /* exceptional events? */
2016         if (sk->sk_err)
2017                 mask |= POLLERR;
2018         if (sk->sk_shutdown == SHUTDOWN_MASK)
2019                 mask |= POLLHUP;
2020         if (sk->sk_shutdown & RCV_SHUTDOWN)
2021                 mask |= POLLRDHUP;
2022
2023         /* readable? */
2024         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2025             (sk->sk_shutdown & RCV_SHUTDOWN))
2026                 mask |= POLLIN | POLLRDNORM;
2027
2028         /* Connection-based need to check for termination and startup */
2029         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2030             sk->sk_state == TCP_CLOSE)
2031                 mask |= POLLHUP;
2032
2033         /*
2034          * we set writable also when the other side has shut down the
2035          * connection. This prevents stuck sockets.
2036          */
2037         if (unix_writable(sk))
2038                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2039
2040         return mask;
2041 }
2042
2043 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2044                                     poll_table *wait)
2045 {
2046         struct sock *sk = sock->sk, *other;
2047         unsigned int mask, writable;
2048
2049         sock_poll_wait(file, sk_sleep(sk), wait);
2050         mask = 0;
2051
2052         /* exceptional events? */
2053         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2054                 mask |= POLLERR;
2055         if (sk->sk_shutdown & RCV_SHUTDOWN)
2056                 mask |= POLLRDHUP;
2057         if (sk->sk_shutdown == SHUTDOWN_MASK)
2058                 mask |= POLLHUP;
2059
2060         /* readable? */
2061         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2062             (sk->sk_shutdown & RCV_SHUTDOWN))
2063                 mask |= POLLIN | POLLRDNORM;
2064
2065         /* Connection-based need to check for termination and startup */
2066         if (sk->sk_type == SOCK_SEQPACKET) {
2067                 if (sk->sk_state == TCP_CLOSE)
2068                         mask |= POLLHUP;
2069                 /* connection hasn't started yet? */
2070                 if (sk->sk_state == TCP_SYN_SENT)
2071                         return mask;
2072         }
2073
2074         /* writable? */
2075         writable = unix_writable(sk);
2076         if (writable) {
2077                 other = unix_peer_get(sk);
2078                 if (other) {
2079                         if (unix_peer(other) != sk) {
2080                                 sock_poll_wait(file, &unix_sk(other)->peer_wait,
2081                                           wait);
2082                                 if (unix_recvq_full(other))
2083                                         writable = 0;
2084                         }
2085
2086                         sock_put(other);
2087                 }
2088         }
2089
2090         if (writable)
2091                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2092         else
2093                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2094
2095         return mask;
2096 }
2097
2098 #ifdef CONFIG_PROC_FS
2099 static struct sock *first_unix_socket(int *i)
2100 {
2101         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2102                 if (!hlist_empty(&unix_socket_table[*i]))
2103                         return __sk_head(&unix_socket_table[*i]);
2104         }
2105         return NULL;
2106 }
2107
2108 static struct sock *next_unix_socket(int *i, struct sock *s)
2109 {
2110         struct sock *next = sk_next(s);
2111         /* More in this chain? */
2112         if (next)
2113                 return next;
2114         /* Look for next non-empty chain. */
2115         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2116                 if (!hlist_empty(&unix_socket_table[*i]))
2117                         return __sk_head(&unix_socket_table[*i]);
2118         }
2119         return NULL;
2120 }
2121
2122 struct unix_iter_state {
2123         struct seq_net_private p;
2124         int i;
2125 };
2126
2127 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2128 {
2129         struct unix_iter_state *iter = seq->private;
2130         loff_t off = 0;
2131         struct sock *s;
2132
2133         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2134                 if (sock_net(s) != seq_file_net(seq))
2135                         continue;
2136                 if (off == pos)
2137                         return s;
2138                 ++off;
2139         }
2140         return NULL;
2141 }
2142
2143 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2144         __acquires(unix_table_lock)
2145 {
2146         spin_lock(&unix_table_lock);
2147         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2148 }
2149
2150 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2151 {
2152         struct unix_iter_state *iter = seq->private;
2153         struct sock *sk = v;
2154         ++*pos;
2155
2156         if (v == SEQ_START_TOKEN)
2157                 sk = first_unix_socket(&iter->i);
2158         else
2159                 sk = next_unix_socket(&iter->i, sk);
2160         while (sk && (sock_net(sk) != seq_file_net(seq)))
2161                 sk = next_unix_socket(&iter->i, sk);
2162         return sk;
2163 }
2164
2165 static void unix_seq_stop(struct seq_file *seq, void *v)
2166         __releases(unix_table_lock)
2167 {
2168         spin_unlock(&unix_table_lock);
2169 }
2170
2171 static int unix_seq_show(struct seq_file *seq, void *v)
2172 {
2173
2174         if (v == SEQ_START_TOKEN)
2175                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2176                          "Inode Path\n");
2177         else {
2178                 struct sock *s = v;
2179                 struct unix_sock *u = unix_sk(s);
2180                 unix_state_lock(s);
2181
2182                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2183                         s,
2184                         atomic_read(&s->sk_refcnt),
2185                         0,
2186                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2187                         s->sk_type,
2188                         s->sk_socket ?
2189                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2190                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2191                         sock_i_ino(s));
2192
2193                 if (u->addr) {
2194                         int i, len;
2195                         seq_putc(seq, ' ');
2196
2197                         i = 0;
2198                         len = u->addr->len - sizeof(short);
2199                         if (!UNIX_ABSTRACT(s))
2200                                 len--;
2201                         else {
2202                                 seq_putc(seq, '@');
2203                                 i++;
2204                         }
2205                         for ( ; i < len; i++)
2206                                 seq_putc(seq, u->addr->name->sun_path[i]);
2207                 }
2208                 unix_state_unlock(s);
2209                 seq_putc(seq, '\n');
2210         }
2211
2212         return 0;
2213 }
2214
2215 static const struct seq_operations unix_seq_ops = {
2216         .start  = unix_seq_start,
2217         .next   = unix_seq_next,
2218         .stop   = unix_seq_stop,
2219         .show   = unix_seq_show,
2220 };
2221
2222 static int unix_seq_open(struct inode *inode, struct file *file)
2223 {
2224         return seq_open_net(inode, file, &unix_seq_ops,
2225                             sizeof(struct unix_iter_state));
2226 }
2227
2228 static const struct file_operations unix_seq_fops = {
2229         .owner          = THIS_MODULE,
2230         .open           = unix_seq_open,
2231         .read           = seq_read,
2232         .llseek         = seq_lseek,
2233         .release        = seq_release_net,
2234 };
2235
2236 #endif
2237
2238 static const struct net_proto_family unix_family_ops = {
2239         .family = PF_UNIX,
2240         .create = unix_create,
2241         .owner  = THIS_MODULE,
2242 };
2243
2244
2245 static int __net_init unix_net_init(struct net *net)
2246 {
2247         int error = -ENOMEM;
2248
2249         net->unx.sysctl_max_dgram_qlen = 10;
2250         if (unix_sysctl_register(net))
2251                 goto out;
2252
2253 #ifdef CONFIG_PROC_FS
2254         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2255                 unix_sysctl_unregister(net);
2256                 goto out;
2257         }
2258 #endif
2259         error = 0;
2260 out:
2261         return error;
2262 }
2263
2264 static void __net_exit unix_net_exit(struct net *net)
2265 {
2266         unix_sysctl_unregister(net);
2267         proc_net_remove(net, "unix");
2268 }
2269
2270 static struct pernet_operations unix_net_ops = {
2271         .init = unix_net_init,
2272         .exit = unix_net_exit,
2273 };
2274
2275 static int __init af_unix_init(void)
2276 {
2277         int rc = -1;
2278         struct sk_buff *dummy_skb;
2279
2280         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2281
2282         rc = proto_register(&unix_proto, 1);
2283         if (rc != 0) {
2284                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2285                        __func__);
2286                 goto out;
2287         }
2288
2289         sock_register(&unix_family_ops);
2290         register_pernet_subsys(&unix_net_ops);
2291 out:
2292         return rc;
2293 }
2294
2295 static void __exit af_unix_exit(void)
2296 {
2297         sock_unregister(PF_UNIX);
2298         proto_unregister(&unix_proto);
2299         unregister_pernet_subsys(&unix_net_ops);
2300 }
2301
2302 /* Earlier than device_initcall() so that other drivers invoking
2303    request_module() don't end up in a loop when modprobe tries
2304    to use a UNIX socket. But later than subsys_initcall() because
2305    we depend on stuff initialised there */
2306 fs_initcall(af_unix_init);
2307 module_exit(af_unix_exit);
2308
2309 MODULE_LICENSE("GPL");
2310 MODULE_ALIAS_NETPROTO(PF_UNIX);