Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[sfrench/cifs-2.6.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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
84
85 #include <linux/module.h>
86 #include <linux/kernel.h>
87 #include <linux/signal.h>
88 #include <linux/sched/signal.h>
89 #include <linux/errno.h>
90 #include <linux/string.h>
91 #include <linux/stat.h>
92 #include <linux/dcache.h>
93 #include <linux/namei.h>
94 #include <linux/socket.h>
95 #include <linux/un.h>
96 #include <linux/fcntl.h>
97 #include <linux/termios.h>
98 #include <linux/sockios.h>
99 #include <linux/net.h>
100 #include <linux/in.h>
101 #include <linux/fs.h>
102 #include <linux/slab.h>
103 #include <linux/uaccess.h>
104 #include <linux/skbuff.h>
105 #include <linux/netdevice.h>
106 #include <net/net_namespace.h>
107 #include <net/sock.h>
108 #include <net/tcp_states.h>
109 #include <net/af_unix.h>
110 #include <linux/proc_fs.h>
111 #include <linux/seq_file.h>
112 #include <net/scm.h>
113 #include <linux/init.h>
114 #include <linux/poll.h>
115 #include <linux/rtnetlink.h>
116 #include <linux/mount.h>
117 #include <net/checksum.h>
118 #include <linux/security.h>
119 #include <linux/freezer.h>
120 #include <linux/file.h>
121
122 struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
123 EXPORT_SYMBOL_GPL(unix_socket_table);
124 DEFINE_SPINLOCK(unix_table_lock);
125 EXPORT_SYMBOL_GPL(unix_table_lock);
126 static atomic_long_t unix_nr_socks;
127
128
129 static struct hlist_head *unix_sockets_unbound(void *addr)
130 {
131         unsigned long hash = (unsigned long)addr;
132
133         hash ^= hash >> 16;
134         hash ^= hash >> 8;
135         hash %= UNIX_HASH_SIZE;
136         return &unix_socket_table[UNIX_HASH_SIZE + hash];
137 }
138
139 #define UNIX_ABSTRACT(sk)       (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
140
141 #ifdef CONFIG_SECURITY_NETWORK
142 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
143 {
144         UNIXCB(skb).secid = scm->secid;
145 }
146
147 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
148 {
149         scm->secid = UNIXCB(skb).secid;
150 }
151
152 static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
153 {
154         return (scm->secid == UNIXCB(skb).secid);
155 }
156 #else
157 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
158 { }
159
160 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
161 { }
162
163 static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
164 {
165         return true;
166 }
167 #endif /* CONFIG_SECURITY_NETWORK */
168
169 /*
170  *  SMP locking strategy:
171  *    hash table is protected with spinlock unix_table_lock
172  *    each socket state is protected by separate spin lock.
173  */
174
175 static inline unsigned int unix_hash_fold(__wsum n)
176 {
177         unsigned int hash = (__force unsigned int)csum_fold(n);
178
179         hash ^= hash>>8;
180         return hash&(UNIX_HASH_SIZE-1);
181 }
182
183 #define unix_peer(sk) (unix_sk(sk)->peer)
184
185 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
186 {
187         return unix_peer(osk) == sk;
188 }
189
190 static inline int unix_may_send(struct sock *sk, struct sock *osk)
191 {
192         return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
193 }
194
195 static inline int unix_recvq_full(struct sock const *sk)
196 {
197         return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
198 }
199
200 struct sock *unix_peer_get(struct sock *s)
201 {
202         struct sock *peer;
203
204         unix_state_lock(s);
205         peer = unix_peer(s);
206         if (peer)
207                 sock_hold(peer);
208         unix_state_unlock(s);
209         return peer;
210 }
211 EXPORT_SYMBOL_GPL(unix_peer_get);
212
213 static inline void unix_release_addr(struct unix_address *addr)
214 {
215         if (refcount_dec_and_test(&addr->refcnt))
216                 kfree(addr);
217 }
218
219 /*
220  *      Check unix socket name:
221  *              - should be not zero length.
222  *              - if started by not zero, should be NULL terminated (FS object)
223  *              - if started by zero, it is abstract name.
224  */
225
226 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
227 {
228         if (len <= sizeof(short) || len > sizeof(*sunaddr))
229                 return -EINVAL;
230         if (!sunaddr || sunaddr->sun_family != AF_UNIX)
231                 return -EINVAL;
232         if (sunaddr->sun_path[0]) {
233                 /*
234                  * This may look like an off by one error but it is a bit more
235                  * subtle. 108 is the longest valid AF_UNIX path for a binding.
236                  * sun_path[108] doesn't as such exist.  However in kernel space
237                  * we are guaranteed that it is a valid memory location in our
238                  * kernel address buffer.
239                  */
240                 ((char *)sunaddr)[len] = 0;
241                 len = strlen(sunaddr->sun_path)+1+sizeof(short);
242                 return len;
243         }
244
245         *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
246         return len;
247 }
248
249 static void __unix_remove_socket(struct sock *sk)
250 {
251         sk_del_node_init(sk);
252 }
253
254 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
255 {
256         WARN_ON(!sk_unhashed(sk));
257         sk_add_node(sk, list);
258 }
259
260 static inline void unix_remove_socket(struct sock *sk)
261 {
262         spin_lock(&unix_table_lock);
263         __unix_remove_socket(sk);
264         spin_unlock(&unix_table_lock);
265 }
266
267 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
268 {
269         spin_lock(&unix_table_lock);
270         __unix_insert_socket(list, sk);
271         spin_unlock(&unix_table_lock);
272 }
273
274 static struct sock *__unix_find_socket_byname(struct net *net,
275                                               struct sockaddr_un *sunname,
276                                               int len, int type, unsigned int hash)
277 {
278         struct sock *s;
279
280         sk_for_each(s, &unix_socket_table[hash ^ type]) {
281                 struct unix_sock *u = unix_sk(s);
282
283                 if (!net_eq(sock_net(s), net))
284                         continue;
285
286                 if (u->addr->len == len &&
287                     !memcmp(u->addr->name, sunname, len))
288                         goto found;
289         }
290         s = NULL;
291 found:
292         return s;
293 }
294
295 static inline struct sock *unix_find_socket_byname(struct net *net,
296                                                    struct sockaddr_un *sunname,
297                                                    int len, int type,
298                                                    unsigned int hash)
299 {
300         struct sock *s;
301
302         spin_lock(&unix_table_lock);
303         s = __unix_find_socket_byname(net, sunname, len, type, hash);
304         if (s)
305                 sock_hold(s);
306         spin_unlock(&unix_table_lock);
307         return s;
308 }
309
310 static struct sock *unix_find_socket_byinode(struct inode *i)
311 {
312         struct sock *s;
313
314         spin_lock(&unix_table_lock);
315         sk_for_each(s,
316                     &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
317                 struct dentry *dentry = unix_sk(s)->path.dentry;
318
319                 if (dentry && d_backing_inode(dentry) == i) {
320                         sock_hold(s);
321                         goto found;
322                 }
323         }
324         s = NULL;
325 found:
326         spin_unlock(&unix_table_lock);
327         return s;
328 }
329
330 /* Support code for asymmetrically connected dgram sockets
331  *
332  * If a datagram socket is connected to a socket not itself connected
333  * to the first socket (eg, /dev/log), clients may only enqueue more
334  * messages if the present receive queue of the server socket is not
335  * "too large". This means there's a second writeability condition
336  * poll and sendmsg need to test. The dgram recv code will do a wake
337  * up on the peer_wait wait queue of a socket upon reception of a
338  * datagram which needs to be propagated to sleeping would-be writers
339  * since these might not have sent anything so far. This can't be
340  * accomplished via poll_wait because the lifetime of the server
341  * socket might be less than that of its clients if these break their
342  * association with it or if the server socket is closed while clients
343  * are still connected to it and there's no way to inform "a polling
344  * implementation" that it should let go of a certain wait queue
345  *
346  * In order to propagate a wake up, a wait_queue_entry_t of the client
347  * socket is enqueued on the peer_wait queue of the server socket
348  * whose wake function does a wake_up on the ordinary client socket
349  * wait queue. This connection is established whenever a write (or
350  * poll for write) hit the flow control condition and broken when the
351  * association to the server socket is dissolved or after a wake up
352  * was relayed.
353  */
354
355 static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
356                                       void *key)
357 {
358         struct unix_sock *u;
359         wait_queue_head_t *u_sleep;
360
361         u = container_of(q, struct unix_sock, peer_wake);
362
363         __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
364                             q);
365         u->peer_wake.private = NULL;
366
367         /* relaying can only happen while the wq still exists */
368         u_sleep = sk_sleep(&u->sk);
369         if (u_sleep)
370                 wake_up_interruptible_poll(u_sleep, key_to_poll(key));
371
372         return 0;
373 }
374
375 static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
376 {
377         struct unix_sock *u, *u_other;
378         int rc;
379
380         u = unix_sk(sk);
381         u_other = unix_sk(other);
382         rc = 0;
383         spin_lock(&u_other->peer_wait.lock);
384
385         if (!u->peer_wake.private) {
386                 u->peer_wake.private = other;
387                 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
388
389                 rc = 1;
390         }
391
392         spin_unlock(&u_other->peer_wait.lock);
393         return rc;
394 }
395
396 static void unix_dgram_peer_wake_disconnect(struct sock *sk,
397                                             struct sock *other)
398 {
399         struct unix_sock *u, *u_other;
400
401         u = unix_sk(sk);
402         u_other = unix_sk(other);
403         spin_lock(&u_other->peer_wait.lock);
404
405         if (u->peer_wake.private == other) {
406                 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
407                 u->peer_wake.private = NULL;
408         }
409
410         spin_unlock(&u_other->peer_wait.lock);
411 }
412
413 static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
414                                                    struct sock *other)
415 {
416         unix_dgram_peer_wake_disconnect(sk, other);
417         wake_up_interruptible_poll(sk_sleep(sk),
418                                    EPOLLOUT |
419                                    EPOLLWRNORM |
420                                    EPOLLWRBAND);
421 }
422
423 /* preconditions:
424  *      - unix_peer(sk) == other
425  *      - association is stable
426  */
427 static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
428 {
429         int connected;
430
431         connected = unix_dgram_peer_wake_connect(sk, other);
432
433         if (unix_recvq_full(other))
434                 return 1;
435
436         if (connected)
437                 unix_dgram_peer_wake_disconnect(sk, other);
438
439         return 0;
440 }
441
442 static int unix_writable(const struct sock *sk)
443 {
444         return sk->sk_state != TCP_LISTEN &&
445                (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
446 }
447
448 static void unix_write_space(struct sock *sk)
449 {
450         struct socket_wq *wq;
451
452         rcu_read_lock();
453         if (unix_writable(sk)) {
454                 wq = rcu_dereference(sk->sk_wq);
455                 if (skwq_has_sleeper(wq))
456                         wake_up_interruptible_sync_poll(&wq->wait,
457                                 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
458                 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
459         }
460         rcu_read_unlock();
461 }
462
463 /* When dgram socket disconnects (or changes its peer), we clear its receive
464  * queue of packets arrived from previous peer. First, it allows to do
465  * flow control based only on wmem_alloc; second, sk connected to peer
466  * may receive messages only from that peer. */
467 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
468 {
469         if (!skb_queue_empty(&sk->sk_receive_queue)) {
470                 skb_queue_purge(&sk->sk_receive_queue);
471                 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
472
473                 /* If one link of bidirectional dgram pipe is disconnected,
474                  * we signal error. Messages are lost. Do not make this,
475                  * when peer was not connected to us.
476                  */
477                 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
478                         other->sk_err = ECONNRESET;
479                         other->sk_error_report(other);
480                 }
481         }
482 }
483
484 static void unix_sock_destructor(struct sock *sk)
485 {
486         struct unix_sock *u = unix_sk(sk);
487
488         skb_queue_purge(&sk->sk_receive_queue);
489
490         WARN_ON(refcount_read(&sk->sk_wmem_alloc));
491         WARN_ON(!sk_unhashed(sk));
492         WARN_ON(sk->sk_socket);
493         if (!sock_flag(sk, SOCK_DEAD)) {
494                 pr_info("Attempt to release alive unix socket: %p\n", sk);
495                 return;
496         }
497
498         if (u->addr)
499                 unix_release_addr(u->addr);
500
501         atomic_long_dec(&unix_nr_socks);
502         local_bh_disable();
503         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
504         local_bh_enable();
505 #ifdef UNIX_REFCNT_DEBUG
506         pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
507                 atomic_long_read(&unix_nr_socks));
508 #endif
509 }
510
511 static void unix_release_sock(struct sock *sk, int embrion)
512 {
513         struct unix_sock *u = unix_sk(sk);
514         struct path path;
515         struct sock *skpair;
516         struct sk_buff *skb;
517         int state;
518
519         unix_remove_socket(sk);
520
521         /* Clear state */
522         unix_state_lock(sk);
523         sock_orphan(sk);
524         sk->sk_shutdown = SHUTDOWN_MASK;
525         path         = u->path;
526         u->path.dentry = NULL;
527         u->path.mnt = NULL;
528         state = sk->sk_state;
529         sk->sk_state = TCP_CLOSE;
530         unix_state_unlock(sk);
531
532         wake_up_interruptible_all(&u->peer_wait);
533
534         skpair = unix_peer(sk);
535
536         if (skpair != NULL) {
537                 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
538                         unix_state_lock(skpair);
539                         /* No more writes */
540                         skpair->sk_shutdown = SHUTDOWN_MASK;
541                         if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
542                                 skpair->sk_err = ECONNRESET;
543                         unix_state_unlock(skpair);
544                         skpair->sk_state_change(skpair);
545                         sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
546                 }
547
548                 unix_dgram_peer_wake_disconnect(sk, skpair);
549                 sock_put(skpair); /* It may now die */
550                 unix_peer(sk) = NULL;
551         }
552
553         /* Try to flush out this socket. Throw out buffers at least */
554
555         while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
556                 if (state == TCP_LISTEN)
557                         unix_release_sock(skb->sk, 1);
558                 /* passed fds are erased in the kfree_skb hook        */
559                 UNIXCB(skb).consumed = skb->len;
560                 kfree_skb(skb);
561         }
562
563         if (path.dentry)
564                 path_put(&path);
565
566         sock_put(sk);
567
568         /* ---- Socket is dead now and most probably destroyed ---- */
569
570         /*
571          * Fixme: BSD difference: In BSD all sockets connected to us get
572          *        ECONNRESET and we die on the spot. In Linux we behave
573          *        like files and pipes do and wait for the last
574          *        dereference.
575          *
576          * Can't we simply set sock->err?
577          *
578          *        What the above comment does talk about? --ANK(980817)
579          */
580
581         if (unix_tot_inflight)
582                 unix_gc();              /* Garbage collect fds */
583 }
584
585 static void init_peercred(struct sock *sk)
586 {
587         put_pid(sk->sk_peer_pid);
588         if (sk->sk_peer_cred)
589                 put_cred(sk->sk_peer_cred);
590         sk->sk_peer_pid  = get_pid(task_tgid(current));
591         sk->sk_peer_cred = get_current_cred();
592 }
593
594 static void copy_peercred(struct sock *sk, struct sock *peersk)
595 {
596         put_pid(sk->sk_peer_pid);
597         if (sk->sk_peer_cred)
598                 put_cred(sk->sk_peer_cred);
599         sk->sk_peer_pid  = get_pid(peersk->sk_peer_pid);
600         sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
601 }
602
603 static int unix_listen(struct socket *sock, int backlog)
604 {
605         int err;
606         struct sock *sk = sock->sk;
607         struct unix_sock *u = unix_sk(sk);
608         struct pid *old_pid = NULL;
609
610         err = -EOPNOTSUPP;
611         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
612                 goto out;       /* Only stream/seqpacket sockets accept */
613         err = -EINVAL;
614         if (!u->addr)
615                 goto out;       /* No listens on an unbound socket */
616         unix_state_lock(sk);
617         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
618                 goto out_unlock;
619         if (backlog > sk->sk_max_ack_backlog)
620                 wake_up_interruptible_all(&u->peer_wait);
621         sk->sk_max_ack_backlog  = backlog;
622         sk->sk_state            = TCP_LISTEN;
623         /* set credentials so connect can copy them */
624         init_peercred(sk);
625         err = 0;
626
627 out_unlock:
628         unix_state_unlock(sk);
629         put_pid(old_pid);
630 out:
631         return err;
632 }
633
634 static int unix_release(struct socket *);
635 static int unix_bind(struct socket *, struct sockaddr *, int);
636 static int unix_stream_connect(struct socket *, struct sockaddr *,
637                                int addr_len, int flags);
638 static int unix_socketpair(struct socket *, struct socket *);
639 static int unix_accept(struct socket *, struct socket *, int, bool);
640 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
641 static __poll_t unix_poll(struct file *, struct socket *, poll_table *);
642 static __poll_t unix_dgram_poll(struct file *, struct socket *,
643                                     poll_table *);
644 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
645 static int unix_shutdown(struct socket *, int);
646 static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
647 static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
648 static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
649                                     size_t size, int flags);
650 static ssize_t unix_stream_splice_read(struct socket *,  loff_t *ppos,
651                                        struct pipe_inode_info *, size_t size,
652                                        unsigned int flags);
653 static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
654 static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
655 static int unix_dgram_connect(struct socket *, struct sockaddr *,
656                               int, int);
657 static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
658 static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
659                                   int);
660
661 static int unix_set_peek_off(struct sock *sk, int val)
662 {
663         struct unix_sock *u = unix_sk(sk);
664
665         if (mutex_lock_interruptible(&u->iolock))
666                 return -EINTR;
667
668         sk->sk_peek_off = val;
669         mutex_unlock(&u->iolock);
670
671         return 0;
672 }
673
674
675 static const struct proto_ops unix_stream_ops = {
676         .family =       PF_UNIX,
677         .owner =        THIS_MODULE,
678         .release =      unix_release,
679         .bind =         unix_bind,
680         .connect =      unix_stream_connect,
681         .socketpair =   unix_socketpair,
682         .accept =       unix_accept,
683         .getname =      unix_getname,
684         .poll =         unix_poll,
685         .ioctl =        unix_ioctl,
686         .listen =       unix_listen,
687         .shutdown =     unix_shutdown,
688         .setsockopt =   sock_no_setsockopt,
689         .getsockopt =   sock_no_getsockopt,
690         .sendmsg =      unix_stream_sendmsg,
691         .recvmsg =      unix_stream_recvmsg,
692         .mmap =         sock_no_mmap,
693         .sendpage =     unix_stream_sendpage,
694         .splice_read =  unix_stream_splice_read,
695         .set_peek_off = unix_set_peek_off,
696 };
697
698 static const struct proto_ops unix_dgram_ops = {
699         .family =       PF_UNIX,
700         .owner =        THIS_MODULE,
701         .release =      unix_release,
702         .bind =         unix_bind,
703         .connect =      unix_dgram_connect,
704         .socketpair =   unix_socketpair,
705         .accept =       sock_no_accept,
706         .getname =      unix_getname,
707         .poll =         unix_dgram_poll,
708         .ioctl =        unix_ioctl,
709         .listen =       sock_no_listen,
710         .shutdown =     unix_shutdown,
711         .setsockopt =   sock_no_setsockopt,
712         .getsockopt =   sock_no_getsockopt,
713         .sendmsg =      unix_dgram_sendmsg,
714         .recvmsg =      unix_dgram_recvmsg,
715         .mmap =         sock_no_mmap,
716         .sendpage =     sock_no_sendpage,
717         .set_peek_off = unix_set_peek_off,
718 };
719
720 static const struct proto_ops unix_seqpacket_ops = {
721         .family =       PF_UNIX,
722         .owner =        THIS_MODULE,
723         .release =      unix_release,
724         .bind =         unix_bind,
725         .connect =      unix_stream_connect,
726         .socketpair =   unix_socketpair,
727         .accept =       unix_accept,
728         .getname =      unix_getname,
729         .poll =         unix_dgram_poll,
730         .ioctl =        unix_ioctl,
731         .listen =       unix_listen,
732         .shutdown =     unix_shutdown,
733         .setsockopt =   sock_no_setsockopt,
734         .getsockopt =   sock_no_getsockopt,
735         .sendmsg =      unix_seqpacket_sendmsg,
736         .recvmsg =      unix_seqpacket_recvmsg,
737         .mmap =         sock_no_mmap,
738         .sendpage =     sock_no_sendpage,
739         .set_peek_off = unix_set_peek_off,
740 };
741
742 static struct proto unix_proto = {
743         .name                   = "UNIX",
744         .owner                  = THIS_MODULE,
745         .obj_size               = sizeof(struct unix_sock),
746 };
747
748 /*
749  * AF_UNIX sockets do not interact with hardware, hence they
750  * dont trigger interrupts - so it's safe for them to have
751  * bh-unsafe locking for their sk_receive_queue.lock. Split off
752  * this special lock-class by reinitializing the spinlock key:
753  */
754 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
755
756 static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
757 {
758         struct sock *sk = NULL;
759         struct unix_sock *u;
760
761         atomic_long_inc(&unix_nr_socks);
762         if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
763                 goto out;
764
765         sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
766         if (!sk)
767                 goto out;
768
769         sock_init_data(sock, sk);
770         lockdep_set_class(&sk->sk_receive_queue.lock,
771                                 &af_unix_sk_receive_queue_lock_key);
772
773         sk->sk_allocation       = GFP_KERNEL_ACCOUNT;
774         sk->sk_write_space      = unix_write_space;
775         sk->sk_max_ack_backlog  = net->unx.sysctl_max_dgram_qlen;
776         sk->sk_destruct         = unix_sock_destructor;
777         u         = unix_sk(sk);
778         u->path.dentry = NULL;
779         u->path.mnt = NULL;
780         spin_lock_init(&u->lock);
781         atomic_long_set(&u->inflight, 0);
782         INIT_LIST_HEAD(&u->link);
783         mutex_init(&u->iolock); /* single task reading lock */
784         mutex_init(&u->bindlock); /* single task binding lock */
785         init_waitqueue_head(&u->peer_wait);
786         init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
787         unix_insert_socket(unix_sockets_unbound(sk), sk);
788 out:
789         if (sk == NULL)
790                 atomic_long_dec(&unix_nr_socks);
791         else {
792                 local_bh_disable();
793                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
794                 local_bh_enable();
795         }
796         return sk;
797 }
798
799 static int unix_create(struct net *net, struct socket *sock, int protocol,
800                        int kern)
801 {
802         if (protocol && protocol != PF_UNIX)
803                 return -EPROTONOSUPPORT;
804
805         sock->state = SS_UNCONNECTED;
806
807         switch (sock->type) {
808         case SOCK_STREAM:
809                 sock->ops = &unix_stream_ops;
810                 break;
811                 /*
812                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
813                  *      nothing uses it.
814                  */
815         case SOCK_RAW:
816                 sock->type = SOCK_DGRAM;
817                 /* fall through */
818         case SOCK_DGRAM:
819                 sock->ops = &unix_dgram_ops;
820                 break;
821         case SOCK_SEQPACKET:
822                 sock->ops = &unix_seqpacket_ops;
823                 break;
824         default:
825                 return -ESOCKTNOSUPPORT;
826         }
827
828         return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
829 }
830
831 static int unix_release(struct socket *sock)
832 {
833         struct sock *sk = sock->sk;
834
835         if (!sk)
836                 return 0;
837
838         unix_release_sock(sk, 0);
839         sock->sk = NULL;
840
841         return 0;
842 }
843
844 static int unix_autobind(struct socket *sock)
845 {
846         struct sock *sk = sock->sk;
847         struct net *net = sock_net(sk);
848         struct unix_sock *u = unix_sk(sk);
849         static u32 ordernum = 1;
850         struct unix_address *addr;
851         int err;
852         unsigned int retries = 0;
853
854         err = mutex_lock_interruptible(&u->bindlock);
855         if (err)
856                 return err;
857
858         err = 0;
859         if (u->addr)
860                 goto out;
861
862         err = -ENOMEM;
863         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
864         if (!addr)
865                 goto out;
866
867         addr->name->sun_family = AF_UNIX;
868         refcount_set(&addr->refcnt, 1);
869
870 retry:
871         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
872         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
873
874         spin_lock(&unix_table_lock);
875         ordernum = (ordernum+1)&0xFFFFF;
876
877         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
878                                       addr->hash)) {
879                 spin_unlock(&unix_table_lock);
880                 /*
881                  * __unix_find_socket_byname() may take long time if many names
882                  * are already in use.
883                  */
884                 cond_resched();
885                 /* Give up if all names seems to be in use. */
886                 if (retries++ == 0xFFFFF) {
887                         err = -ENOSPC;
888                         kfree(addr);
889                         goto out;
890                 }
891                 goto retry;
892         }
893         addr->hash ^= sk->sk_type;
894
895         __unix_remove_socket(sk);
896         u->addr = addr;
897         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
898         spin_unlock(&unix_table_lock);
899         err = 0;
900
901 out:    mutex_unlock(&u->bindlock);
902         return err;
903 }
904
905 static struct sock *unix_find_other(struct net *net,
906                                     struct sockaddr_un *sunname, int len,
907                                     int type, unsigned int hash, int *error)
908 {
909         struct sock *u;
910         struct path path;
911         int err = 0;
912
913         if (sunname->sun_path[0]) {
914                 struct inode *inode;
915                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
916                 if (err)
917                         goto fail;
918                 inode = d_backing_inode(path.dentry);
919                 err = inode_permission(inode, MAY_WRITE);
920                 if (err)
921                         goto put_fail;
922
923                 err = -ECONNREFUSED;
924                 if (!S_ISSOCK(inode->i_mode))
925                         goto put_fail;
926                 u = unix_find_socket_byinode(inode);
927                 if (!u)
928                         goto put_fail;
929
930                 if (u->sk_type == type)
931                         touch_atime(&path);
932
933                 path_put(&path);
934
935                 err = -EPROTOTYPE;
936                 if (u->sk_type != type) {
937                         sock_put(u);
938                         goto fail;
939                 }
940         } else {
941                 err = -ECONNREFUSED;
942                 u = unix_find_socket_byname(net, sunname, len, type, hash);
943                 if (u) {
944                         struct dentry *dentry;
945                         dentry = unix_sk(u)->path.dentry;
946                         if (dentry)
947                                 touch_atime(&unix_sk(u)->path);
948                 } else
949                         goto fail;
950         }
951         return u;
952
953 put_fail:
954         path_put(&path);
955 fail:
956         *error = err;
957         return NULL;
958 }
959
960 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
961 {
962         struct dentry *dentry;
963         struct path path;
964         int err = 0;
965         /*
966          * Get the parent directory, calculate the hash for last
967          * component.
968          */
969         dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
970         err = PTR_ERR(dentry);
971         if (IS_ERR(dentry))
972                 return err;
973
974         /*
975          * All right, let's create it.
976          */
977         err = security_path_mknod(&path, dentry, mode, 0);
978         if (!err) {
979                 err = vfs_mknod(d_inode(path.dentry), dentry, mode, 0);
980                 if (!err) {
981                         res->mnt = mntget(path.mnt);
982                         res->dentry = dget(dentry);
983                 }
984         }
985         done_path_create(&path, dentry);
986         return err;
987 }
988
989 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
990 {
991         struct sock *sk = sock->sk;
992         struct net *net = sock_net(sk);
993         struct unix_sock *u = unix_sk(sk);
994         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
995         char *sun_path = sunaddr->sun_path;
996         int err;
997         unsigned int hash;
998         struct unix_address *addr;
999         struct hlist_head *list;
1000         struct path path = { };
1001
1002         err = -EINVAL;
1003         if (addr_len < offsetofend(struct sockaddr_un, sun_family) ||
1004             sunaddr->sun_family != AF_UNIX)
1005                 goto out;
1006
1007         if (addr_len == sizeof(short)) {
1008                 err = unix_autobind(sock);
1009                 goto out;
1010         }
1011
1012         err = unix_mkname(sunaddr, addr_len, &hash);
1013         if (err < 0)
1014                 goto out;
1015         addr_len = err;
1016
1017         if (sun_path[0]) {
1018                 umode_t mode = S_IFSOCK |
1019                        (SOCK_INODE(sock)->i_mode & ~current_umask());
1020                 err = unix_mknod(sun_path, mode, &path);
1021                 if (err) {
1022                         if (err == -EEXIST)
1023                                 err = -EADDRINUSE;
1024                         goto out;
1025                 }
1026         }
1027
1028         err = mutex_lock_interruptible(&u->bindlock);
1029         if (err)
1030                 goto out_put;
1031
1032         err = -EINVAL;
1033         if (u->addr)
1034                 goto out_up;
1035
1036         err = -ENOMEM;
1037         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1038         if (!addr)
1039                 goto out_up;
1040
1041         memcpy(addr->name, sunaddr, addr_len);
1042         addr->len = addr_len;
1043         addr->hash = hash ^ sk->sk_type;
1044         refcount_set(&addr->refcnt, 1);
1045
1046         if (sun_path[0]) {
1047                 addr->hash = UNIX_HASH_SIZE;
1048                 hash = d_backing_inode(path.dentry)->i_ino & (UNIX_HASH_SIZE - 1);
1049                 spin_lock(&unix_table_lock);
1050                 u->path = path;
1051                 list = &unix_socket_table[hash];
1052         } else {
1053                 spin_lock(&unix_table_lock);
1054                 err = -EADDRINUSE;
1055                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
1056                                               sk->sk_type, hash)) {
1057                         unix_release_addr(addr);
1058                         goto out_unlock;
1059                 }
1060
1061                 list = &unix_socket_table[addr->hash];
1062         }
1063
1064         err = 0;
1065         __unix_remove_socket(sk);
1066         u->addr = addr;
1067         __unix_insert_socket(list, sk);
1068
1069 out_unlock:
1070         spin_unlock(&unix_table_lock);
1071 out_up:
1072         mutex_unlock(&u->bindlock);
1073 out_put:
1074         if (err)
1075                 path_put(&path);
1076 out:
1077         return err;
1078 }
1079
1080 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1081 {
1082         if (unlikely(sk1 == sk2) || !sk2) {
1083                 unix_state_lock(sk1);
1084                 return;
1085         }
1086         if (sk1 < sk2) {
1087                 unix_state_lock(sk1);
1088                 unix_state_lock_nested(sk2);
1089         } else {
1090                 unix_state_lock(sk2);
1091                 unix_state_lock_nested(sk1);
1092         }
1093 }
1094
1095 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1096 {
1097         if (unlikely(sk1 == sk2) || !sk2) {
1098                 unix_state_unlock(sk1);
1099                 return;
1100         }
1101         unix_state_unlock(sk1);
1102         unix_state_unlock(sk2);
1103 }
1104
1105 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1106                               int alen, int flags)
1107 {
1108         struct sock *sk = sock->sk;
1109         struct net *net = sock_net(sk);
1110         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1111         struct sock *other;
1112         unsigned int hash;
1113         int err;
1114
1115         err = -EINVAL;
1116         if (alen < offsetofend(struct sockaddr, sa_family))
1117                 goto out;
1118
1119         if (addr->sa_family != AF_UNSPEC) {
1120                 err = unix_mkname(sunaddr, alen, &hash);
1121                 if (err < 0)
1122                         goto out;
1123                 alen = err;
1124
1125                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1126                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1127                         goto out;
1128
1129 restart:
1130                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1131                 if (!other)
1132                         goto out;
1133
1134                 unix_state_double_lock(sk, other);
1135
1136                 /* Apparently VFS overslept socket death. Retry. */
1137                 if (sock_flag(other, SOCK_DEAD)) {
1138                         unix_state_double_unlock(sk, other);
1139                         sock_put(other);
1140                         goto restart;
1141                 }
1142
1143                 err = -EPERM;
1144                 if (!unix_may_send(sk, other))
1145                         goto out_unlock;
1146
1147                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1148                 if (err)
1149                         goto out_unlock;
1150
1151         } else {
1152                 /*
1153                  *      1003.1g breaking connected state with AF_UNSPEC
1154                  */
1155                 other = NULL;
1156                 unix_state_double_lock(sk, other);
1157         }
1158
1159         /*
1160          * If it was connected, reconnect.
1161          */
1162         if (unix_peer(sk)) {
1163                 struct sock *old_peer = unix_peer(sk);
1164                 unix_peer(sk) = other;
1165                 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1166
1167                 unix_state_double_unlock(sk, other);
1168
1169                 if (other != old_peer)
1170                         unix_dgram_disconnected(sk, old_peer);
1171                 sock_put(old_peer);
1172         } else {
1173                 unix_peer(sk) = other;
1174                 unix_state_double_unlock(sk, other);
1175         }
1176         return 0;
1177
1178 out_unlock:
1179         unix_state_double_unlock(sk, other);
1180         sock_put(other);
1181 out:
1182         return err;
1183 }
1184
1185 static long unix_wait_for_peer(struct sock *other, long timeo)
1186 {
1187         struct unix_sock *u = unix_sk(other);
1188         int sched;
1189         DEFINE_WAIT(wait);
1190
1191         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1192
1193         sched = !sock_flag(other, SOCK_DEAD) &&
1194                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1195                 unix_recvq_full(other);
1196
1197         unix_state_unlock(other);
1198
1199         if (sched)
1200                 timeo = schedule_timeout(timeo);
1201
1202         finish_wait(&u->peer_wait, &wait);
1203         return timeo;
1204 }
1205
1206 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1207                                int addr_len, int flags)
1208 {
1209         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1210         struct sock *sk = sock->sk;
1211         struct net *net = sock_net(sk);
1212         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1213         struct sock *newsk = NULL;
1214         struct sock *other = NULL;
1215         struct sk_buff *skb = NULL;
1216         unsigned int hash;
1217         int st;
1218         int err;
1219         long timeo;
1220
1221         err = unix_mkname(sunaddr, addr_len, &hash);
1222         if (err < 0)
1223                 goto out;
1224         addr_len = err;
1225
1226         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1227             (err = unix_autobind(sock)) != 0)
1228                 goto out;
1229
1230         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1231
1232         /* First of all allocate resources.
1233            If we will make it after state is locked,
1234            we will have to recheck all again in any case.
1235          */
1236
1237         err = -ENOMEM;
1238
1239         /* create new sock for complete connection */
1240         newsk = unix_create1(sock_net(sk), NULL, 0);
1241         if (newsk == NULL)
1242                 goto out;
1243
1244         /* Allocate skb for sending to listening sock */
1245         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1246         if (skb == NULL)
1247                 goto out;
1248
1249 restart:
1250         /*  Find listening sock. */
1251         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1252         if (!other)
1253                 goto out;
1254
1255         /* Latch state of peer */
1256         unix_state_lock(other);
1257
1258         /* Apparently VFS overslept socket death. Retry. */
1259         if (sock_flag(other, SOCK_DEAD)) {
1260                 unix_state_unlock(other);
1261                 sock_put(other);
1262                 goto restart;
1263         }
1264
1265         err = -ECONNREFUSED;
1266         if (other->sk_state != TCP_LISTEN)
1267                 goto out_unlock;
1268         if (other->sk_shutdown & RCV_SHUTDOWN)
1269                 goto out_unlock;
1270
1271         if (unix_recvq_full(other)) {
1272                 err = -EAGAIN;
1273                 if (!timeo)
1274                         goto out_unlock;
1275
1276                 timeo = unix_wait_for_peer(other, timeo);
1277
1278                 err = sock_intr_errno(timeo);
1279                 if (signal_pending(current))
1280                         goto out;
1281                 sock_put(other);
1282                 goto restart;
1283         }
1284
1285         /* Latch our state.
1286
1287            It is tricky place. We need to grab our state lock and cannot
1288            drop lock on peer. It is dangerous because deadlock is
1289            possible. Connect to self case and simultaneous
1290            attempt to connect are eliminated by checking socket
1291            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1292            check this before attempt to grab lock.
1293
1294            Well, and we have to recheck the state after socket locked.
1295          */
1296         st = sk->sk_state;
1297
1298         switch (st) {
1299         case TCP_CLOSE:
1300                 /* This is ok... continue with connect */
1301                 break;
1302         case TCP_ESTABLISHED:
1303                 /* Socket is already connected */
1304                 err = -EISCONN;
1305                 goto out_unlock;
1306         default:
1307                 err = -EINVAL;
1308                 goto out_unlock;
1309         }
1310
1311         unix_state_lock_nested(sk);
1312
1313         if (sk->sk_state != st) {
1314                 unix_state_unlock(sk);
1315                 unix_state_unlock(other);
1316                 sock_put(other);
1317                 goto restart;
1318         }
1319
1320         err = security_unix_stream_connect(sk, other, newsk);
1321         if (err) {
1322                 unix_state_unlock(sk);
1323                 goto out_unlock;
1324         }
1325
1326         /* The way is open! Fastly set all the necessary fields... */
1327
1328         sock_hold(sk);
1329         unix_peer(newsk)        = sk;
1330         newsk->sk_state         = TCP_ESTABLISHED;
1331         newsk->sk_type          = sk->sk_type;
1332         init_peercred(newsk);
1333         newu = unix_sk(newsk);
1334         RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1335         otheru = unix_sk(other);
1336
1337         /* copy address information from listening to new sock*/
1338         if (otheru->addr) {
1339                 refcount_inc(&otheru->addr->refcnt);
1340                 newu->addr = otheru->addr;
1341         }
1342         if (otheru->path.dentry) {
1343                 path_get(&otheru->path);
1344                 newu->path = otheru->path;
1345         }
1346
1347         /* Set credentials */
1348         copy_peercred(sk, other);
1349
1350         sock->state     = SS_CONNECTED;
1351         sk->sk_state    = TCP_ESTABLISHED;
1352         sock_hold(newsk);
1353
1354         smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1355         unix_peer(sk)   = newsk;
1356
1357         unix_state_unlock(sk);
1358
1359         /* take ten and and send info to listening sock */
1360         spin_lock(&other->sk_receive_queue.lock);
1361         __skb_queue_tail(&other->sk_receive_queue, skb);
1362         spin_unlock(&other->sk_receive_queue.lock);
1363         unix_state_unlock(other);
1364         other->sk_data_ready(other);
1365         sock_put(other);
1366         return 0;
1367
1368 out_unlock:
1369         if (other)
1370                 unix_state_unlock(other);
1371
1372 out:
1373         kfree_skb(skb);
1374         if (newsk)
1375                 unix_release_sock(newsk, 0);
1376         if (other)
1377                 sock_put(other);
1378         return err;
1379 }
1380
1381 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1382 {
1383         struct sock *ska = socka->sk, *skb = sockb->sk;
1384
1385         /* Join our sockets back to back */
1386         sock_hold(ska);
1387         sock_hold(skb);
1388         unix_peer(ska) = skb;
1389         unix_peer(skb) = ska;
1390         init_peercred(ska);
1391         init_peercred(skb);
1392
1393         if (ska->sk_type != SOCK_DGRAM) {
1394                 ska->sk_state = TCP_ESTABLISHED;
1395                 skb->sk_state = TCP_ESTABLISHED;
1396                 socka->state  = SS_CONNECTED;
1397                 sockb->state  = SS_CONNECTED;
1398         }
1399         return 0;
1400 }
1401
1402 static void unix_sock_inherit_flags(const struct socket *old,
1403                                     struct socket *new)
1404 {
1405         if (test_bit(SOCK_PASSCRED, &old->flags))
1406                 set_bit(SOCK_PASSCRED, &new->flags);
1407         if (test_bit(SOCK_PASSSEC, &old->flags))
1408                 set_bit(SOCK_PASSSEC, &new->flags);
1409 }
1410
1411 static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1412                        bool kern)
1413 {
1414         struct sock *sk = sock->sk;
1415         struct sock *tsk;
1416         struct sk_buff *skb;
1417         int err;
1418
1419         err = -EOPNOTSUPP;
1420         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1421                 goto out;
1422
1423         err = -EINVAL;
1424         if (sk->sk_state != TCP_LISTEN)
1425                 goto out;
1426
1427         /* If socket state is TCP_LISTEN it cannot change (for now...),
1428          * so that no locks are necessary.
1429          */
1430
1431         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1432         if (!skb) {
1433                 /* This means receive shutdown. */
1434                 if (err == 0)
1435                         err = -EINVAL;
1436                 goto out;
1437         }
1438
1439         tsk = skb->sk;
1440         skb_free_datagram(sk, skb);
1441         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1442
1443         /* attach accepted sock to socket */
1444         unix_state_lock(tsk);
1445         newsock->state = SS_CONNECTED;
1446         unix_sock_inherit_flags(sock, newsock);
1447         sock_graft(tsk, newsock);
1448         unix_state_unlock(tsk);
1449         return 0;
1450
1451 out:
1452         return err;
1453 }
1454
1455
1456 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1457 {
1458         struct sock *sk = sock->sk;
1459         struct unix_sock *u;
1460         DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1461         int err = 0;
1462
1463         if (peer) {
1464                 sk = unix_peer_get(sk);
1465
1466                 err = -ENOTCONN;
1467                 if (!sk)
1468                         goto out;
1469                 err = 0;
1470         } else {
1471                 sock_hold(sk);
1472         }
1473
1474         u = unix_sk(sk);
1475         unix_state_lock(sk);
1476         if (!u->addr) {
1477                 sunaddr->sun_family = AF_UNIX;
1478                 sunaddr->sun_path[0] = 0;
1479                 *uaddr_len = sizeof(short);
1480         } else {
1481                 struct unix_address *addr = u->addr;
1482
1483                 *uaddr_len = addr->len;
1484                 memcpy(sunaddr, addr->name, *uaddr_len);
1485         }
1486         unix_state_unlock(sk);
1487         sock_put(sk);
1488 out:
1489         return err;
1490 }
1491
1492 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1493 {
1494         int i;
1495
1496         scm->fp = UNIXCB(skb).fp;
1497         UNIXCB(skb).fp = NULL;
1498
1499         for (i = scm->fp->count-1; i >= 0; i--)
1500                 unix_notinflight(scm->fp->user, scm->fp->fp[i]);
1501 }
1502
1503 static void unix_destruct_scm(struct sk_buff *skb)
1504 {
1505         struct scm_cookie scm;
1506         memset(&scm, 0, sizeof(scm));
1507         scm.pid  = UNIXCB(skb).pid;
1508         if (UNIXCB(skb).fp)
1509                 unix_detach_fds(&scm, skb);
1510
1511         /* Alas, it calls VFS */
1512         /* So fscking what? fput() had been SMP-safe since the last Summer */
1513         scm_destroy(&scm);
1514         sock_wfree(skb);
1515 }
1516
1517 /*
1518  * The "user->unix_inflight" variable is protected by the garbage
1519  * collection lock, and we just read it locklessly here. If you go
1520  * over the limit, there might be a tiny race in actually noticing
1521  * it across threads. Tough.
1522  */
1523 static inline bool too_many_unix_fds(struct task_struct *p)
1524 {
1525         struct user_struct *user = current_user();
1526
1527         if (unlikely(user->unix_inflight > task_rlimit(p, RLIMIT_NOFILE)))
1528                 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
1529         return false;
1530 }
1531
1532 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1533 {
1534         int i;
1535
1536         if (too_many_unix_fds(current))
1537                 return -ETOOMANYREFS;
1538
1539         /*
1540          * Need to duplicate file references for the sake of garbage
1541          * collection.  Otherwise a socket in the fps might become a
1542          * candidate for GC while the skb is not yet queued.
1543          */
1544         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1545         if (!UNIXCB(skb).fp)
1546                 return -ENOMEM;
1547
1548         for (i = scm->fp->count - 1; i >= 0; i--)
1549                 unix_inflight(scm->fp->user, scm->fp->fp[i]);
1550         return 0;
1551 }
1552
1553 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1554 {
1555         int err = 0;
1556
1557         UNIXCB(skb).pid  = get_pid(scm->pid);
1558         UNIXCB(skb).uid = scm->creds.uid;
1559         UNIXCB(skb).gid = scm->creds.gid;
1560         UNIXCB(skb).fp = NULL;
1561         unix_get_secdata(scm, skb);
1562         if (scm->fp && send_fds)
1563                 err = unix_attach_fds(scm, skb);
1564
1565         skb->destructor = unix_destruct_scm;
1566         return err;
1567 }
1568
1569 static bool unix_passcred_enabled(const struct socket *sock,
1570                                   const struct sock *other)
1571 {
1572         return test_bit(SOCK_PASSCRED, &sock->flags) ||
1573                !other->sk_socket ||
1574                test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1575 }
1576
1577 /*
1578  * Some apps rely on write() giving SCM_CREDENTIALS
1579  * We include credentials if source or destination socket
1580  * asserted SOCK_PASSCRED.
1581  */
1582 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1583                             const struct sock *other)
1584 {
1585         if (UNIXCB(skb).pid)
1586                 return;
1587         if (unix_passcred_enabled(sock, other)) {
1588                 UNIXCB(skb).pid  = get_pid(task_tgid(current));
1589                 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1590         }
1591 }
1592
1593 static int maybe_init_creds(struct scm_cookie *scm,
1594                             struct socket *socket,
1595                             const struct sock *other)
1596 {
1597         int err;
1598         struct msghdr msg = { .msg_controllen = 0 };
1599
1600         err = scm_send(socket, &msg, scm, false);
1601         if (err)
1602                 return err;
1603
1604         if (unix_passcred_enabled(socket, other)) {
1605                 scm->pid = get_pid(task_tgid(current));
1606                 current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1607         }
1608         return err;
1609 }
1610
1611 static bool unix_skb_scm_eq(struct sk_buff *skb,
1612                             struct scm_cookie *scm)
1613 {
1614         const struct unix_skb_parms *u = &UNIXCB(skb);
1615
1616         return u->pid == scm->pid &&
1617                uid_eq(u->uid, scm->creds.uid) &&
1618                gid_eq(u->gid, scm->creds.gid) &&
1619                unix_secdata_eq(scm, skb);
1620 }
1621
1622 /*
1623  *      Send AF_UNIX data.
1624  */
1625
1626 static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1627                               size_t len)
1628 {
1629         struct sock *sk = sock->sk;
1630         struct net *net = sock_net(sk);
1631         struct unix_sock *u = unix_sk(sk);
1632         DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1633         struct sock *other = NULL;
1634         int namelen = 0; /* fake GCC */
1635         int err;
1636         unsigned int hash;
1637         struct sk_buff *skb;
1638         long timeo;
1639         struct scm_cookie scm;
1640         int data_len = 0;
1641         int sk_locked;
1642
1643         wait_for_unix_gc();
1644         err = scm_send(sock, msg, &scm, false);
1645         if (err < 0)
1646                 return err;
1647
1648         err = -EOPNOTSUPP;
1649         if (msg->msg_flags&MSG_OOB)
1650                 goto out;
1651
1652         if (msg->msg_namelen) {
1653                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1654                 if (err < 0)
1655                         goto out;
1656                 namelen = err;
1657         } else {
1658                 sunaddr = NULL;
1659                 err = -ENOTCONN;
1660                 other = unix_peer_get(sk);
1661                 if (!other)
1662                         goto out;
1663         }
1664
1665         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1666             && (err = unix_autobind(sock)) != 0)
1667                 goto out;
1668
1669         err = -EMSGSIZE;
1670         if (len > sk->sk_sndbuf - 32)
1671                 goto out;
1672
1673         if (len > SKB_MAX_ALLOC) {
1674                 data_len = min_t(size_t,
1675                                  len - SKB_MAX_ALLOC,
1676                                  MAX_SKB_FRAGS * PAGE_SIZE);
1677                 data_len = PAGE_ALIGN(data_len);
1678
1679                 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1680         }
1681
1682         skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1683                                    msg->msg_flags & MSG_DONTWAIT, &err,
1684                                    PAGE_ALLOC_COSTLY_ORDER);
1685         if (skb == NULL)
1686                 goto out;
1687
1688         err = unix_scm_to_skb(&scm, skb, true);
1689         if (err < 0)
1690                 goto out_free;
1691
1692         skb_put(skb, len - data_len);
1693         skb->data_len = data_len;
1694         skb->len = len;
1695         err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1696         if (err)
1697                 goto out_free;
1698
1699         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1700
1701 restart:
1702         if (!other) {
1703                 err = -ECONNRESET;
1704                 if (sunaddr == NULL)
1705                         goto out_free;
1706
1707                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1708                                         hash, &err);
1709                 if (other == NULL)
1710                         goto out_free;
1711         }
1712
1713         if (sk_filter(other, skb) < 0) {
1714                 /* Toss the packet but do not return any error to the sender */
1715                 err = len;
1716                 goto out_free;
1717         }
1718
1719         sk_locked = 0;
1720         unix_state_lock(other);
1721 restart_locked:
1722         err = -EPERM;
1723         if (!unix_may_send(sk, other))
1724                 goto out_unlock;
1725
1726         if (unlikely(sock_flag(other, SOCK_DEAD))) {
1727                 /*
1728                  *      Check with 1003.1g - what should
1729                  *      datagram error
1730                  */
1731                 unix_state_unlock(other);
1732                 sock_put(other);
1733
1734                 if (!sk_locked)
1735                         unix_state_lock(sk);
1736
1737                 err = 0;
1738                 if (unix_peer(sk) == other) {
1739                         unix_peer(sk) = NULL;
1740                         unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1741
1742                         unix_state_unlock(sk);
1743
1744                         unix_dgram_disconnected(sk, other);
1745                         sock_put(other);
1746                         err = -ECONNREFUSED;
1747                 } else {
1748                         unix_state_unlock(sk);
1749                 }
1750
1751                 other = NULL;
1752                 if (err)
1753                         goto out_free;
1754                 goto restart;
1755         }
1756
1757         err = -EPIPE;
1758         if (other->sk_shutdown & RCV_SHUTDOWN)
1759                 goto out_unlock;
1760
1761         if (sk->sk_type != SOCK_SEQPACKET) {
1762                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1763                 if (err)
1764                         goto out_unlock;
1765         }
1766
1767         /* other == sk && unix_peer(other) != sk if
1768          * - unix_peer(sk) == NULL, destination address bound to sk
1769          * - unix_peer(sk) == sk by time of get but disconnected before lock
1770          */
1771         if (other != sk &&
1772             unlikely(unix_peer(other) != sk && unix_recvq_full(other))) {
1773                 if (timeo) {
1774                         timeo = unix_wait_for_peer(other, timeo);
1775
1776                         err = sock_intr_errno(timeo);
1777                         if (signal_pending(current))
1778                                 goto out_free;
1779
1780                         goto restart;
1781                 }
1782
1783                 if (!sk_locked) {
1784                         unix_state_unlock(other);
1785                         unix_state_double_lock(sk, other);
1786                 }
1787
1788                 if (unix_peer(sk) != other ||
1789                     unix_dgram_peer_wake_me(sk, other)) {
1790                         err = -EAGAIN;
1791                         sk_locked = 1;
1792                         goto out_unlock;
1793                 }
1794
1795                 if (!sk_locked) {
1796                         sk_locked = 1;
1797                         goto restart_locked;
1798                 }
1799         }
1800
1801         if (unlikely(sk_locked))
1802                 unix_state_unlock(sk);
1803
1804         if (sock_flag(other, SOCK_RCVTSTAMP))
1805                 __net_timestamp(skb);
1806         maybe_add_creds(skb, sock, other);
1807         skb_queue_tail(&other->sk_receive_queue, skb);
1808         unix_state_unlock(other);
1809         other->sk_data_ready(other);
1810         sock_put(other);
1811         scm_destroy(&scm);
1812         return len;
1813
1814 out_unlock:
1815         if (sk_locked)
1816                 unix_state_unlock(sk);
1817         unix_state_unlock(other);
1818 out_free:
1819         kfree_skb(skb);
1820 out:
1821         if (other)
1822                 sock_put(other);
1823         scm_destroy(&scm);
1824         return err;
1825 }
1826
1827 /* We use paged skbs for stream sockets, and limit occupancy to 32768
1828  * bytes, and a minimum of a full page.
1829  */
1830 #define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1831
1832 static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1833                                size_t len)
1834 {
1835         struct sock *sk = sock->sk;
1836         struct sock *other = NULL;
1837         int err, size;
1838         struct sk_buff *skb;
1839         int sent = 0;
1840         struct scm_cookie scm;
1841         bool fds_sent = false;
1842         int data_len;
1843
1844         wait_for_unix_gc();
1845         err = scm_send(sock, msg, &scm, false);
1846         if (err < 0)
1847                 return err;
1848
1849         err = -EOPNOTSUPP;
1850         if (msg->msg_flags&MSG_OOB)
1851                 goto out_err;
1852
1853         if (msg->msg_namelen) {
1854                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1855                 goto out_err;
1856         } else {
1857                 err = -ENOTCONN;
1858                 other = unix_peer(sk);
1859                 if (!other)
1860                         goto out_err;
1861         }
1862
1863         if (sk->sk_shutdown & SEND_SHUTDOWN)
1864                 goto pipe_err;
1865
1866         while (sent < len) {
1867                 size = len - sent;
1868
1869                 /* Keep two messages in the pipe so it schedules better */
1870                 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1871
1872                 /* allow fallback to order-0 allocations */
1873                 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1874
1875                 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1876
1877                 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1878
1879                 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1880                                            msg->msg_flags & MSG_DONTWAIT, &err,
1881                                            get_order(UNIX_SKB_FRAGS_SZ));
1882                 if (!skb)
1883                         goto out_err;
1884
1885                 /* Only send the fds in the first buffer */
1886                 err = unix_scm_to_skb(&scm, skb, !fds_sent);
1887                 if (err < 0) {
1888                         kfree_skb(skb);
1889                         goto out_err;
1890                 }
1891                 fds_sent = true;
1892
1893                 skb_put(skb, size - data_len);
1894                 skb->data_len = data_len;
1895                 skb->len = size;
1896                 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
1897                 if (err) {
1898                         kfree_skb(skb);
1899                         goto out_err;
1900                 }
1901
1902                 unix_state_lock(other);
1903
1904                 if (sock_flag(other, SOCK_DEAD) ||
1905                     (other->sk_shutdown & RCV_SHUTDOWN))
1906                         goto pipe_err_free;
1907
1908                 maybe_add_creds(skb, sock, other);
1909                 skb_queue_tail(&other->sk_receive_queue, skb);
1910                 unix_state_unlock(other);
1911                 other->sk_data_ready(other);
1912                 sent += size;
1913         }
1914
1915         scm_destroy(&scm);
1916
1917         return sent;
1918
1919 pipe_err_free:
1920         unix_state_unlock(other);
1921         kfree_skb(skb);
1922 pipe_err:
1923         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1924                 send_sig(SIGPIPE, current, 0);
1925         err = -EPIPE;
1926 out_err:
1927         scm_destroy(&scm);
1928         return sent ? : err;
1929 }
1930
1931 static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1932                                     int offset, size_t size, int flags)
1933 {
1934         int err;
1935         bool send_sigpipe = false;
1936         bool init_scm = true;
1937         struct scm_cookie scm;
1938         struct sock *other, *sk = socket->sk;
1939         struct sk_buff *skb, *newskb = NULL, *tail = NULL;
1940
1941         if (flags & MSG_OOB)
1942                 return -EOPNOTSUPP;
1943
1944         other = unix_peer(sk);
1945         if (!other || sk->sk_state != TCP_ESTABLISHED)
1946                 return -ENOTCONN;
1947
1948         if (false) {
1949 alloc_skb:
1950                 unix_state_unlock(other);
1951                 mutex_unlock(&unix_sk(other)->iolock);
1952                 newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
1953                                               &err, 0);
1954                 if (!newskb)
1955                         goto err;
1956         }
1957
1958         /* we must acquire iolock as we modify already present
1959          * skbs in the sk_receive_queue and mess with skb->len
1960          */
1961         err = mutex_lock_interruptible(&unix_sk(other)->iolock);
1962         if (err) {
1963                 err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
1964                 goto err;
1965         }
1966
1967         if (sk->sk_shutdown & SEND_SHUTDOWN) {
1968                 err = -EPIPE;
1969                 send_sigpipe = true;
1970                 goto err_unlock;
1971         }
1972
1973         unix_state_lock(other);
1974
1975         if (sock_flag(other, SOCK_DEAD) ||
1976             other->sk_shutdown & RCV_SHUTDOWN) {
1977                 err = -EPIPE;
1978                 send_sigpipe = true;
1979                 goto err_state_unlock;
1980         }
1981
1982         if (init_scm) {
1983                 err = maybe_init_creds(&scm, socket, other);
1984                 if (err)
1985                         goto err_state_unlock;
1986                 init_scm = false;
1987         }
1988
1989         skb = skb_peek_tail(&other->sk_receive_queue);
1990         if (tail && tail == skb) {
1991                 skb = newskb;
1992         } else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
1993                 if (newskb) {
1994                         skb = newskb;
1995                 } else {
1996                         tail = skb;
1997                         goto alloc_skb;
1998                 }
1999         } else if (newskb) {
2000                 /* this is fast path, we don't necessarily need to
2001                  * call to kfree_skb even though with newskb == NULL
2002                  * this - does no harm
2003                  */
2004                 consume_skb(newskb);
2005                 newskb = NULL;
2006         }
2007
2008         if (skb_append_pagefrags(skb, page, offset, size)) {
2009                 tail = skb;
2010                 goto alloc_skb;
2011         }
2012
2013         skb->len += size;
2014         skb->data_len += size;
2015         skb->truesize += size;
2016         refcount_add(size, &sk->sk_wmem_alloc);
2017
2018         if (newskb) {
2019                 err = unix_scm_to_skb(&scm, skb, false);
2020                 if (err)
2021                         goto err_state_unlock;
2022                 spin_lock(&other->sk_receive_queue.lock);
2023                 __skb_queue_tail(&other->sk_receive_queue, newskb);
2024                 spin_unlock(&other->sk_receive_queue.lock);
2025         }
2026
2027         unix_state_unlock(other);
2028         mutex_unlock(&unix_sk(other)->iolock);
2029
2030         other->sk_data_ready(other);
2031         scm_destroy(&scm);
2032         return size;
2033
2034 err_state_unlock:
2035         unix_state_unlock(other);
2036 err_unlock:
2037         mutex_unlock(&unix_sk(other)->iolock);
2038 err:
2039         kfree_skb(newskb);
2040         if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2041                 send_sig(SIGPIPE, current, 0);
2042         if (!init_scm)
2043                 scm_destroy(&scm);
2044         return err;
2045 }
2046
2047 static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2048                                   size_t len)
2049 {
2050         int err;
2051         struct sock *sk = sock->sk;
2052
2053         err = sock_error(sk);
2054         if (err)
2055                 return err;
2056
2057         if (sk->sk_state != TCP_ESTABLISHED)
2058                 return -ENOTCONN;
2059
2060         if (msg->msg_namelen)
2061                 msg->msg_namelen = 0;
2062
2063         return unix_dgram_sendmsg(sock, msg, len);
2064 }
2065
2066 static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2067                                   size_t size, int flags)
2068 {
2069         struct sock *sk = sock->sk;
2070
2071         if (sk->sk_state != TCP_ESTABLISHED)
2072                 return -ENOTCONN;
2073
2074         return unix_dgram_recvmsg(sock, msg, size, flags);
2075 }
2076
2077 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2078 {
2079         struct unix_sock *u = unix_sk(sk);
2080
2081         if (u->addr) {
2082                 msg->msg_namelen = u->addr->len;
2083                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
2084         }
2085 }
2086
2087 static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2088                               size_t size, int flags)
2089 {
2090         struct scm_cookie scm;
2091         struct sock *sk = sock->sk;
2092         struct unix_sock *u = unix_sk(sk);
2093         struct sk_buff *skb, *last;
2094         long timeo;
2095         int err;
2096         int peeked, skip;
2097
2098         err = -EOPNOTSUPP;
2099         if (flags&MSG_OOB)
2100                 goto out;
2101
2102         timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2103
2104         do {
2105                 mutex_lock(&u->iolock);
2106
2107                 skip = sk_peek_offset(sk, flags);
2108                 skb = __skb_try_recv_datagram(sk, flags, NULL, &peeked, &skip,
2109                                               &err, &last);
2110                 if (skb)
2111                         break;
2112
2113                 mutex_unlock(&u->iolock);
2114
2115                 if (err != -EAGAIN)
2116                         break;
2117         } while (timeo &&
2118                  !__skb_wait_for_more_packets(sk, &err, &timeo, last));
2119
2120         if (!skb) { /* implies iolock unlocked */
2121                 unix_state_lock(sk);
2122                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2123                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2124                     (sk->sk_shutdown & RCV_SHUTDOWN))
2125                         err = 0;
2126                 unix_state_unlock(sk);
2127                 goto out;
2128         }
2129
2130         if (wq_has_sleeper(&u->peer_wait))
2131                 wake_up_interruptible_sync_poll(&u->peer_wait,
2132                                                 EPOLLOUT | EPOLLWRNORM |
2133                                                 EPOLLWRBAND);
2134
2135         if (msg->msg_name)
2136                 unix_copy_addr(msg, skb->sk);
2137
2138         if (size > skb->len - skip)
2139                 size = skb->len - skip;
2140         else if (size < skb->len - skip)
2141                 msg->msg_flags |= MSG_TRUNC;
2142
2143         err = skb_copy_datagram_msg(skb, skip, msg, size);
2144         if (err)
2145                 goto out_free;
2146
2147         if (sock_flag(sk, SOCK_RCVTSTAMP))
2148                 __sock_recv_timestamp(msg, sk, skb);
2149
2150         memset(&scm, 0, sizeof(scm));
2151
2152         scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2153         unix_set_secdata(&scm, skb);
2154
2155         if (!(flags & MSG_PEEK)) {
2156                 if (UNIXCB(skb).fp)
2157                         unix_detach_fds(&scm, skb);
2158
2159                 sk_peek_offset_bwd(sk, skb->len);
2160         } else {
2161                 /* It is questionable: on PEEK we could:
2162                    - do not return fds - good, but too simple 8)
2163                    - return fds, and do not return them on read (old strategy,
2164                      apparently wrong)
2165                    - clone fds (I chose it for now, it is the most universal
2166                      solution)
2167
2168                    POSIX 1003.1g does not actually define this clearly
2169                    at all. POSIX 1003.1g doesn't define a lot of things
2170                    clearly however!
2171
2172                 */
2173
2174                 sk_peek_offset_fwd(sk, size);
2175
2176                 if (UNIXCB(skb).fp)
2177                         scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2178         }
2179         err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2180
2181         scm_recv(sock, msg, &scm, flags);
2182
2183 out_free:
2184         skb_free_datagram(sk, skb);
2185         mutex_unlock(&u->iolock);
2186 out:
2187         return err;
2188 }
2189
2190 /*
2191  *      Sleep until more data has arrived. But check for races..
2192  */
2193 static long unix_stream_data_wait(struct sock *sk, long timeo,
2194                                   struct sk_buff *last, unsigned int last_len,
2195                                   bool freezable)
2196 {
2197         struct sk_buff *tail;
2198         DEFINE_WAIT(wait);
2199
2200         unix_state_lock(sk);
2201
2202         for (;;) {
2203                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2204
2205                 tail = skb_peek_tail(&sk->sk_receive_queue);
2206                 if (tail != last ||
2207                     (tail && tail->len != last_len) ||
2208                     sk->sk_err ||
2209                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
2210                     signal_pending(current) ||
2211                     !timeo)
2212                         break;
2213
2214                 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2215                 unix_state_unlock(sk);
2216                 if (freezable)
2217                         timeo = freezable_schedule_timeout(timeo);
2218                 else
2219                         timeo = schedule_timeout(timeo);
2220                 unix_state_lock(sk);
2221
2222                 if (sock_flag(sk, SOCK_DEAD))
2223                         break;
2224
2225                 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2226         }
2227
2228         finish_wait(sk_sleep(sk), &wait);
2229         unix_state_unlock(sk);
2230         return timeo;
2231 }
2232
2233 static unsigned int unix_skb_len(const struct sk_buff *skb)
2234 {
2235         return skb->len - UNIXCB(skb).consumed;
2236 }
2237
2238 struct unix_stream_read_state {
2239         int (*recv_actor)(struct sk_buff *, int, int,
2240                           struct unix_stream_read_state *);
2241         struct socket *socket;
2242         struct msghdr *msg;
2243         struct pipe_inode_info *pipe;
2244         size_t size;
2245         int flags;
2246         unsigned int splice_flags;
2247 };
2248
2249 static int unix_stream_read_generic(struct unix_stream_read_state *state,
2250                                     bool freezable)
2251 {
2252         struct scm_cookie scm;
2253         struct socket *sock = state->socket;
2254         struct sock *sk = sock->sk;
2255         struct unix_sock *u = unix_sk(sk);
2256         int copied = 0;
2257         int flags = state->flags;
2258         int noblock = flags & MSG_DONTWAIT;
2259         bool check_creds = false;
2260         int target;
2261         int err = 0;
2262         long timeo;
2263         int skip;
2264         size_t size = state->size;
2265         unsigned int last_len;
2266
2267         if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2268                 err = -EINVAL;
2269                 goto out;
2270         }
2271
2272         if (unlikely(flags & MSG_OOB)) {
2273                 err = -EOPNOTSUPP;
2274                 goto out;
2275         }
2276
2277         target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2278         timeo = sock_rcvtimeo(sk, noblock);
2279
2280         memset(&scm, 0, sizeof(scm));
2281
2282         /* Lock the socket to prevent queue disordering
2283          * while sleeps in memcpy_tomsg
2284          */
2285         mutex_lock(&u->iolock);
2286
2287         skip = max(sk_peek_offset(sk, flags), 0);
2288
2289         do {
2290                 int chunk;
2291                 bool drop_skb;
2292                 struct sk_buff *skb, *last;
2293
2294 redo:
2295                 unix_state_lock(sk);
2296                 if (sock_flag(sk, SOCK_DEAD)) {
2297                         err = -ECONNRESET;
2298                         goto unlock;
2299                 }
2300                 last = skb = skb_peek(&sk->sk_receive_queue);
2301                 last_len = last ? last->len : 0;
2302 again:
2303                 if (skb == NULL) {
2304                         if (copied >= target)
2305                                 goto unlock;
2306
2307                         /*
2308                          *      POSIX 1003.1g mandates this order.
2309                          */
2310
2311                         err = sock_error(sk);
2312                         if (err)
2313                                 goto unlock;
2314                         if (sk->sk_shutdown & RCV_SHUTDOWN)
2315                                 goto unlock;
2316
2317                         unix_state_unlock(sk);
2318                         if (!timeo) {
2319                                 err = -EAGAIN;
2320                                 break;
2321                         }
2322
2323                         mutex_unlock(&u->iolock);
2324
2325                         timeo = unix_stream_data_wait(sk, timeo, last,
2326                                                       last_len, freezable);
2327
2328                         if (signal_pending(current)) {
2329                                 err = sock_intr_errno(timeo);
2330                                 scm_destroy(&scm);
2331                                 goto out;
2332                         }
2333
2334                         mutex_lock(&u->iolock);
2335                         goto redo;
2336 unlock:
2337                         unix_state_unlock(sk);
2338                         break;
2339                 }
2340
2341                 while (skip >= unix_skb_len(skb)) {
2342                         skip -= unix_skb_len(skb);
2343                         last = skb;
2344                         last_len = skb->len;
2345                         skb = skb_peek_next(skb, &sk->sk_receive_queue);
2346                         if (!skb)
2347                                 goto again;
2348                 }
2349
2350                 unix_state_unlock(sk);
2351
2352                 if (check_creds) {
2353                         /* Never glue messages from different writers */
2354                         if (!unix_skb_scm_eq(skb, &scm))
2355                                 break;
2356                 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2357                         /* Copy credentials */
2358                         scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2359                         unix_set_secdata(&scm, skb);
2360                         check_creds = true;
2361                 }
2362
2363                 /* Copy address just once */
2364                 if (state->msg && state->msg->msg_name) {
2365                         DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2366                                          state->msg->msg_name);
2367                         unix_copy_addr(state->msg, skb->sk);
2368                         sunaddr = NULL;
2369                 }
2370
2371                 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2372                 skb_get(skb);
2373                 chunk = state->recv_actor(skb, skip, chunk, state);
2374                 drop_skb = !unix_skb_len(skb);
2375                 /* skb is only safe to use if !drop_skb */
2376                 consume_skb(skb);
2377                 if (chunk < 0) {
2378                         if (copied == 0)
2379                                 copied = -EFAULT;
2380                         break;
2381                 }
2382                 copied += chunk;
2383                 size -= chunk;
2384
2385                 if (drop_skb) {
2386                         /* the skb was touched by a concurrent reader;
2387                          * we should not expect anything from this skb
2388                          * anymore and assume it invalid - we can be
2389                          * sure it was dropped from the socket queue
2390                          *
2391                          * let's report a short read
2392                          */
2393                         err = 0;
2394                         break;
2395                 }
2396
2397                 /* Mark read part of skb as used */
2398                 if (!(flags & MSG_PEEK)) {
2399                         UNIXCB(skb).consumed += chunk;
2400
2401                         sk_peek_offset_bwd(sk, chunk);
2402
2403                         if (UNIXCB(skb).fp)
2404                                 unix_detach_fds(&scm, skb);
2405
2406                         if (unix_skb_len(skb))
2407                                 break;
2408
2409                         skb_unlink(skb, &sk->sk_receive_queue);
2410                         consume_skb(skb);
2411
2412                         if (scm.fp)
2413                                 break;
2414                 } else {
2415                         /* It is questionable, see note in unix_dgram_recvmsg.
2416                          */
2417                         if (UNIXCB(skb).fp)
2418                                 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2419
2420                         sk_peek_offset_fwd(sk, chunk);
2421
2422                         if (UNIXCB(skb).fp)
2423                                 break;
2424
2425                         skip = 0;
2426                         last = skb;
2427                         last_len = skb->len;
2428                         unix_state_lock(sk);
2429                         skb = skb_peek_next(skb, &sk->sk_receive_queue);
2430                         if (skb)
2431                                 goto again;
2432                         unix_state_unlock(sk);
2433                         break;
2434                 }
2435         } while (size);
2436
2437         mutex_unlock(&u->iolock);
2438         if (state->msg)
2439                 scm_recv(sock, state->msg, &scm, flags);
2440         else
2441                 scm_destroy(&scm);
2442 out:
2443         return copied ? : err;
2444 }
2445
2446 static int unix_stream_read_actor(struct sk_buff *skb,
2447                                   int skip, int chunk,
2448                                   struct unix_stream_read_state *state)
2449 {
2450         int ret;
2451
2452         ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2453                                     state->msg, chunk);
2454         return ret ?: chunk;
2455 }
2456
2457 static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2458                                size_t size, int flags)
2459 {
2460         struct unix_stream_read_state state = {
2461                 .recv_actor = unix_stream_read_actor,
2462                 .socket = sock,
2463                 .msg = msg,
2464                 .size = size,
2465                 .flags = flags
2466         };
2467
2468         return unix_stream_read_generic(&state, true);
2469 }
2470
2471 static int unix_stream_splice_actor(struct sk_buff *skb,
2472                                     int skip, int chunk,
2473                                     struct unix_stream_read_state *state)
2474 {
2475         return skb_splice_bits(skb, state->socket->sk,
2476                                UNIXCB(skb).consumed + skip,
2477                                state->pipe, chunk, state->splice_flags);
2478 }
2479
2480 static ssize_t unix_stream_splice_read(struct socket *sock,  loff_t *ppos,
2481                                        struct pipe_inode_info *pipe,
2482                                        size_t size, unsigned int flags)
2483 {
2484         struct unix_stream_read_state state = {
2485                 .recv_actor = unix_stream_splice_actor,
2486                 .socket = sock,
2487                 .pipe = pipe,
2488                 .size = size,
2489                 .splice_flags = flags,
2490         };
2491
2492         if (unlikely(*ppos))
2493                 return -ESPIPE;
2494
2495         if (sock->file->f_flags & O_NONBLOCK ||
2496             flags & SPLICE_F_NONBLOCK)
2497                 state.flags = MSG_DONTWAIT;
2498
2499         return unix_stream_read_generic(&state, false);
2500 }
2501
2502 static int unix_shutdown(struct socket *sock, int mode)
2503 {
2504         struct sock *sk = sock->sk;
2505         struct sock *other;
2506
2507         if (mode < SHUT_RD || mode > SHUT_RDWR)
2508                 return -EINVAL;
2509         /* This maps:
2510          * SHUT_RD   (0) -> RCV_SHUTDOWN  (1)
2511          * SHUT_WR   (1) -> SEND_SHUTDOWN (2)
2512          * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2513          */
2514         ++mode;
2515
2516         unix_state_lock(sk);
2517         sk->sk_shutdown |= mode;
2518         other = unix_peer(sk);
2519         if (other)
2520                 sock_hold(other);
2521         unix_state_unlock(sk);
2522         sk->sk_state_change(sk);
2523
2524         if (other &&
2525                 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2526
2527                 int peer_mode = 0;
2528
2529                 if (mode&RCV_SHUTDOWN)
2530                         peer_mode |= SEND_SHUTDOWN;
2531                 if (mode&SEND_SHUTDOWN)
2532                         peer_mode |= RCV_SHUTDOWN;
2533                 unix_state_lock(other);
2534                 other->sk_shutdown |= peer_mode;
2535                 unix_state_unlock(other);
2536                 other->sk_state_change(other);
2537                 if (peer_mode == SHUTDOWN_MASK)
2538                         sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2539                 else if (peer_mode & RCV_SHUTDOWN)
2540                         sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2541         }
2542         if (other)
2543                 sock_put(other);
2544
2545         return 0;
2546 }
2547
2548 long unix_inq_len(struct sock *sk)
2549 {
2550         struct sk_buff *skb;
2551         long amount = 0;
2552
2553         if (sk->sk_state == TCP_LISTEN)
2554                 return -EINVAL;
2555
2556         spin_lock(&sk->sk_receive_queue.lock);
2557         if (sk->sk_type == SOCK_STREAM ||
2558             sk->sk_type == SOCK_SEQPACKET) {
2559                 skb_queue_walk(&sk->sk_receive_queue, skb)
2560                         amount += unix_skb_len(skb);
2561         } else {
2562                 skb = skb_peek(&sk->sk_receive_queue);
2563                 if (skb)
2564                         amount = skb->len;
2565         }
2566         spin_unlock(&sk->sk_receive_queue.lock);
2567
2568         return amount;
2569 }
2570 EXPORT_SYMBOL_GPL(unix_inq_len);
2571
2572 long unix_outq_len(struct sock *sk)
2573 {
2574         return sk_wmem_alloc_get(sk);
2575 }
2576 EXPORT_SYMBOL_GPL(unix_outq_len);
2577
2578 static int unix_open_file(struct sock *sk)
2579 {
2580         struct path path;
2581         struct file *f;
2582         int fd;
2583
2584         if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2585                 return -EPERM;
2586
2587         unix_state_lock(sk);
2588         path = unix_sk(sk)->path;
2589         if (!path.dentry) {
2590                 unix_state_unlock(sk);
2591                 return -ENOENT;
2592         }
2593
2594         path_get(&path);
2595         unix_state_unlock(sk);
2596
2597         fd = get_unused_fd_flags(O_CLOEXEC);
2598         if (fd < 0)
2599                 goto out;
2600
2601         f = dentry_open(&path, O_PATH, current_cred());
2602         if (IS_ERR(f)) {
2603                 put_unused_fd(fd);
2604                 fd = PTR_ERR(f);
2605                 goto out;
2606         }
2607
2608         fd_install(fd, f);
2609 out:
2610         path_put(&path);
2611
2612         return fd;
2613 }
2614
2615 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2616 {
2617         struct sock *sk = sock->sk;
2618         long amount = 0;
2619         int err;
2620
2621         switch (cmd) {
2622         case SIOCOUTQ:
2623                 amount = unix_outq_len(sk);
2624                 err = put_user(amount, (int __user *)arg);
2625                 break;
2626         case SIOCINQ:
2627                 amount = unix_inq_len(sk);
2628                 if (amount < 0)
2629                         err = amount;
2630                 else
2631                         err = put_user(amount, (int __user *)arg);
2632                 break;
2633         case SIOCUNIXFILE:
2634                 err = unix_open_file(sk);
2635                 break;
2636         default:
2637                 err = -ENOIOCTLCMD;
2638                 break;
2639         }
2640         return err;
2641 }
2642
2643 static __poll_t unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2644 {
2645         struct sock *sk = sock->sk;
2646         __poll_t mask;
2647
2648         sock_poll_wait(file, sk_sleep(sk), wait);
2649         mask = 0;
2650
2651         /* exceptional events? */
2652         if (sk->sk_err)
2653                 mask |= EPOLLERR;
2654         if (sk->sk_shutdown == SHUTDOWN_MASK)
2655                 mask |= EPOLLHUP;
2656         if (sk->sk_shutdown & RCV_SHUTDOWN)
2657                 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2658
2659         /* readable? */
2660         if (!skb_queue_empty(&sk->sk_receive_queue))
2661                 mask |= EPOLLIN | EPOLLRDNORM;
2662
2663         /* Connection-based need to check for termination and startup */
2664         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2665             sk->sk_state == TCP_CLOSE)
2666                 mask |= EPOLLHUP;
2667
2668         /*
2669          * we set writable also when the other side has shut down the
2670          * connection. This prevents stuck sockets.
2671          */
2672         if (unix_writable(sk))
2673                 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2674
2675         return mask;
2676 }
2677
2678 static __poll_t unix_dgram_poll(struct file *file, struct socket *sock,
2679                                     poll_table *wait)
2680 {
2681         struct sock *sk = sock->sk, *other;
2682         unsigned int writable;
2683         __poll_t mask;
2684
2685         sock_poll_wait(file, sk_sleep(sk), wait);
2686         mask = 0;
2687
2688         /* exceptional events? */
2689         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2690                 mask |= EPOLLERR |
2691                         (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
2692
2693         if (sk->sk_shutdown & RCV_SHUTDOWN)
2694                 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2695         if (sk->sk_shutdown == SHUTDOWN_MASK)
2696                 mask |= EPOLLHUP;
2697
2698         /* readable? */
2699         if (!skb_queue_empty(&sk->sk_receive_queue))
2700                 mask |= EPOLLIN | EPOLLRDNORM;
2701
2702         /* Connection-based need to check for termination and startup */
2703         if (sk->sk_type == SOCK_SEQPACKET) {
2704                 if (sk->sk_state == TCP_CLOSE)
2705                         mask |= EPOLLHUP;
2706                 /* connection hasn't started yet? */
2707                 if (sk->sk_state == TCP_SYN_SENT)
2708                         return mask;
2709         }
2710
2711         /* No write status requested, avoid expensive OUT tests. */
2712         if (!(poll_requested_events(wait) & (EPOLLWRBAND|EPOLLWRNORM|EPOLLOUT)))
2713                 return mask;
2714
2715         writable = unix_writable(sk);
2716         if (writable) {
2717                 unix_state_lock(sk);
2718
2719                 other = unix_peer(sk);
2720                 if (other && unix_peer(other) != sk &&
2721                     unix_recvq_full(other) &&
2722                     unix_dgram_peer_wake_me(sk, other))
2723                         writable = 0;
2724
2725                 unix_state_unlock(sk);
2726         }
2727
2728         if (writable)
2729                 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2730         else
2731                 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2732
2733         return mask;
2734 }
2735
2736 #ifdef CONFIG_PROC_FS
2737
2738 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2739
2740 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2741 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2742 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2743
2744 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2745 {
2746         unsigned long offset = get_offset(*pos);
2747         unsigned long bucket = get_bucket(*pos);
2748         struct sock *sk;
2749         unsigned long count = 0;
2750
2751         for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2752                 if (sock_net(sk) != seq_file_net(seq))
2753                         continue;
2754                 if (++count == offset)
2755                         break;
2756         }
2757
2758         return sk;
2759 }
2760
2761 static struct sock *unix_next_socket(struct seq_file *seq,
2762                                      struct sock *sk,
2763                                      loff_t *pos)
2764 {
2765         unsigned long bucket;
2766
2767         while (sk > (struct sock *)SEQ_START_TOKEN) {
2768                 sk = sk_next(sk);
2769                 if (!sk)
2770                         goto next_bucket;
2771                 if (sock_net(sk) == seq_file_net(seq))
2772                         return sk;
2773         }
2774
2775         do {
2776                 sk = unix_from_bucket(seq, pos);
2777                 if (sk)
2778                         return sk;
2779
2780 next_bucket:
2781                 bucket = get_bucket(*pos) + 1;
2782                 *pos = set_bucket_offset(bucket, 1);
2783         } while (bucket < ARRAY_SIZE(unix_socket_table));
2784
2785         return NULL;
2786 }
2787
2788 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2789         __acquires(unix_table_lock)
2790 {
2791         spin_lock(&unix_table_lock);
2792
2793         if (!*pos)
2794                 return SEQ_START_TOKEN;
2795
2796         if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2797                 return NULL;
2798
2799         return unix_next_socket(seq, NULL, pos);
2800 }
2801
2802 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2803 {
2804         ++*pos;
2805         return unix_next_socket(seq, v, pos);
2806 }
2807
2808 static void unix_seq_stop(struct seq_file *seq, void *v)
2809         __releases(unix_table_lock)
2810 {
2811         spin_unlock(&unix_table_lock);
2812 }
2813
2814 static int unix_seq_show(struct seq_file *seq, void *v)
2815 {
2816
2817         if (v == SEQ_START_TOKEN)
2818                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2819                          "Inode Path\n");
2820         else {
2821                 struct sock *s = v;
2822                 struct unix_sock *u = unix_sk(s);
2823                 unix_state_lock(s);
2824
2825                 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2826                         s,
2827                         refcount_read(&s->sk_refcnt),
2828                         0,
2829                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2830                         s->sk_type,
2831                         s->sk_socket ?
2832                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2833                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2834                         sock_i_ino(s));
2835
2836                 if (u->addr) {
2837                         int i, len;
2838                         seq_putc(seq, ' ');
2839
2840                         i = 0;
2841                         len = u->addr->len - sizeof(short);
2842                         if (!UNIX_ABSTRACT(s))
2843                                 len--;
2844                         else {
2845                                 seq_putc(seq, '@');
2846                                 i++;
2847                         }
2848                         for ( ; i < len; i++)
2849                                 seq_putc(seq, u->addr->name->sun_path[i] ?:
2850                                          '@');
2851                 }
2852                 unix_state_unlock(s);
2853                 seq_putc(seq, '\n');
2854         }
2855
2856         return 0;
2857 }
2858
2859 static const struct seq_operations unix_seq_ops = {
2860         .start  = unix_seq_start,
2861         .next   = unix_seq_next,
2862         .stop   = unix_seq_stop,
2863         .show   = unix_seq_show,
2864 };
2865
2866 static int unix_seq_open(struct inode *inode, struct file *file)
2867 {
2868         return seq_open_net(inode, file, &unix_seq_ops,
2869                             sizeof(struct seq_net_private));
2870 }
2871
2872 static const struct file_operations unix_seq_fops = {
2873         .open           = unix_seq_open,
2874         .read           = seq_read,
2875         .llseek         = seq_lseek,
2876         .release        = seq_release_net,
2877 };
2878
2879 #endif
2880
2881 static const struct net_proto_family unix_family_ops = {
2882         .family = PF_UNIX,
2883         .create = unix_create,
2884         .owner  = THIS_MODULE,
2885 };
2886
2887
2888 static int __net_init unix_net_init(struct net *net)
2889 {
2890         int error = -ENOMEM;
2891
2892         net->unx.sysctl_max_dgram_qlen = 10;
2893         if (unix_sysctl_register(net))
2894                 goto out;
2895
2896 #ifdef CONFIG_PROC_FS
2897         if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2898                 unix_sysctl_unregister(net);
2899                 goto out;
2900         }
2901 #endif
2902         error = 0;
2903 out:
2904         return error;
2905 }
2906
2907 static void __net_exit unix_net_exit(struct net *net)
2908 {
2909         unix_sysctl_unregister(net);
2910         remove_proc_entry("unix", net->proc_net);
2911 }
2912
2913 static struct pernet_operations unix_net_ops = {
2914         .init = unix_net_init,
2915         .exit = unix_net_exit,
2916 };
2917
2918 static int __init af_unix_init(void)
2919 {
2920         int rc = -1;
2921
2922         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2923
2924         rc = proto_register(&unix_proto, 1);
2925         if (rc != 0) {
2926                 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2927                 goto out;
2928         }
2929
2930         sock_register(&unix_family_ops);
2931         register_pernet_subsys(&unix_net_ops);
2932 out:
2933         return rc;
2934 }
2935
2936 static void __exit af_unix_exit(void)
2937 {
2938         sock_unregister(PF_UNIX);
2939         proto_unregister(&unix_proto);
2940         unregister_pernet_subsys(&unix_net_ops);
2941 }
2942
2943 /* Earlier than device_initcall() so that other drivers invoking
2944    request_module() don't end up in a loop when modprobe tries
2945    to use a UNIX socket. But later than subsys_initcall() because
2946    we depend on stuff initialised there */
2947 fs_initcall(af_unix_init);
2948 module_exit(af_unix_exit);
2949
2950 MODULE_LICENSE("GPL");
2951 MODULE_ALIAS_NETPROTO(PF_UNIX);