?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Fix crash with FLAT_MEMORY and ARCH_PFN_OFFSET != 0
[pv_ops_mirror.git] / net / unix / af_unix.c
blob060bba4567d2588d2ceafce255e75ee84aa55b27
1 /*
2 * NET4: Implementation of BSD Unix domain sockets.
3 *
4 * Authors: Alan Cox, <alan.cox@linux.org>
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 * Version: $Id: af_unix.c,v 1.133 2002/02/08 03:57:19 davem Exp $
12 *
13 * Fixes:
14 * Linus Torvalds : Assorted bug cures.
15 * Niibe Yutaka : async I/O support.
16 * Carsten Paeth : PF_UNIX check, address fixes.
17 * Alan Cox : Limit size of allocated blocks.
18 * Alan Cox : Fixed the stupid socketpair bug.
19 * Alan Cox : BSD compatibility fine tuning.
20 * Alan Cox : Fixed a bug in connect when interrupted.
21 * Alan Cox : Sorted out a proper draft version of
22 * file descriptor passing hacked up from
23 * Mike Shaver's work.
24 * Marty Leisner : Fixes to fd passing
25 * Nick Nevin : recvmsg bugfix.
26 * Alan Cox : Started proper garbage collector
27 * Heiko EiBfeldt : Missing verify_area check
28 * Alan Cox : Started POSIXisms
29 * Andreas Schwab : Replace inode by dentry for proper
30 * reference counting
31 * Kirk Petersen : Made this a module
32 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
33 * Lots of bug fixes.
34 * Alexey Kuznetosv : Repaired (I hope) bugs introduces
35 * by above two patches.
36 * Andrea Arcangeli : If possible we block in connect(2)
37 * if the max backlog of the listen socket
38 * is been reached. This won't break
39 * old apps and it will avoid huge amount
40 * of socks hashed (this for unix_gc()
41 * performances reasons).
42 * Security fix that limits the max
43 * number of socks to 2*max_files and
44 * the number of skb queueable in the
45 * dgram receiver.
46 * Artur Skawina : Hash function optimizations
47 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
48 * Malcolm Beattie : Set peercred for socketpair
49 * Michal Ostrowski : Module initialization cleanup.
50 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
51 * the core infrastructure is doing that
52 * for all net proto families now (2.5.69+)
53 *
54 *
55 * Known differences from reference BSD that was tested:
56 *
57 * [TO FIX]
58 * ECONNREFUSED is not returned from one end of a connected() socket to the
59 * other the moment one end closes.
60 * fstat() doesn't return st_dev=0, and give the blksize as high water mark
61 * and a fake inode identifier (nor the BSD first socket fstat twice bug).
62 * [NOT TO FIX]
63 * accept() returns a path name even if the connecting socket has closed
64 * in the meantime (BSD loses the path and gives up).
65 * accept() returns 0 length path for an unbound connector. BSD returns 16
66 * and a null first byte in the path (but not for gethost/peername - BSD bug ??)
67 * socketpair(...SOCK_RAW..) doesn't panic the kernel.
68 * BSD af_unix apparently has connect forgetting to block properly.
69 * (need to check this with the POSIX spec in detail)
70 *
71 * Differences from 2.0.0-11-... (ANK)
72 * Bug fixes and improvements.
73 * - client shutdown killed server socket.
74 * - removed all useless cli/sti pairs.
75 *
76 * Semantic changes/extensions.
77 * - generic control message passing.
78 * - SCM_CREDENTIALS control message.
79 * - "Abstract" (not FS based) socket bindings.
80 * Abstract names are sequences of bytes (not zero terminated)
81 * started by 0, so that this name space does not intersect
82 * with BSD names.
83 */
84
85 #include <linux/module.h>
86 #include <linux/kernel.h>
87 #include <linux/signal.h>
88 #include <linux/sched.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 <asm/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
120 int sysctl_unix_max_dgram_qlen __read_mostly = 10;
121
122 static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
123 static DEFINE_SPINLOCK(unix_table_lock);
124 static atomic_t unix_nr_socks = ATOMIC_INIT(0);
125
126 #define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE])
127
128 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
129
130 static struct sock *first_unix_socket(int *i)
131 {
132 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
133 if (!hlist_empty(&unix_socket_table[*i]))
134 return __sk_head(&unix_socket_table[*i]);
135 }
136 return NULL;
137 }
138
139 static struct sock *next_unix_socket(int *i, struct sock *s)
140 {
141 struct sock *next = sk_next(s);
142 /* More in this chain? */
143 if (next)
144 return next;
145 /* Look for next non-empty chain. */
146 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
147 if (!hlist_empty(&unix_socket_table[*i]))
148 return __sk_head(&unix_socket_table[*i]);
149 }
150 return NULL;
151 }
152
153 #define forall_unix_sockets(i, s) \
154 for (s = first_unix_socket(&(i)); s; s = next_unix_socket(&(i),(s)))
155
156 #ifdef CONFIG_SECURITY_NETWORK
157 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
158 {
159 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
160 }
161
162 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
163 {
164 scm->secid = *UNIXSID(skb);
165 }
166 #else
167 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
168 { }
169
170 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
171 { }
172 #endif /* CONFIG_SECURITY_NETWORK */
173
174 /*
175 * SMP locking strategy:
176 * hash table is protected with spinlock unix_table_lock
177 * each socket state is protected by separate rwlock.
178 */
179
180 static inline unsigned unix_hash_fold(__wsum n)
181 {
182 unsigned hash = (__force unsigned)n;
183 hash ^= hash>>16;
184 hash ^= hash>>8;
185 return hash&(UNIX_HASH_SIZE-1);
186 }
187
188 #define unix_peer(sk) (unix_sk(sk)->peer)
189
190 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
191 {
192 return unix_peer(osk) == sk;
193 }
194
195 static inline int unix_may_send(struct sock *sk, struct sock *osk)
196 {
197 return (unix_peer(osk) == NULL || unix_our_peer(sk, osk));
198 }
199
200 static 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
212 static inline void unix_release_addr(struct unix_address *addr)
213 {
214 if (atomic_dec_and_test(&addr->refcnt))
215 kfree(addr);
216 }
217
218 /*
219 * Check unix socket name:
220 * - should be not zero length.
221 * - if started by not zero, should be NULL terminated (FS object)
222 * - if started by zero, it is abstract name.
223 */
224
225 static int unix_mkname(struct sockaddr_un * sunaddr, int len, unsigned *hashp)
226 {
227 if (len <= sizeof(short) || len > sizeof(*sunaddr))
228 return -EINVAL;
229 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
230 return -EINVAL;
231 if (sunaddr->sun_path[0]) {
232 /*
233 * This may look like an off by one error but it is a bit more
234 * subtle. 108 is the longest valid AF_UNIX path for a binding.
235 * sun_path[108] doesnt as such exist. However in kernel space
236 * we are guaranteed that it is a valid memory location in our
237 * kernel address buffer.
238 */
239 ((char *)sunaddr)[len]=0;
240 len = strlen(sunaddr->sun_path)+1+sizeof(short);
241 return len;
242 }
243
244 *hashp = unix_hash_fold(csum_partial((char*)sunaddr, len, 0));
245 return len;
246 }
247
248 static void __unix_remove_socket(struct sock *sk)
249 {
250 sk_del_node_init(sk);
251 }
252
253 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
254 {
255 BUG_TRAP(sk_unhashed(sk));
256 sk_add_node(sk, list);
257 }
258
259 static inline void unix_remove_socket(struct sock *sk)
260 {
261 spin_lock(&unix_table_lock);
262 __unix_remove_socket(sk);
263 spin_unlock(&unix_table_lock);
264 }
265
266 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
267 {
268 spin_lock(&unix_table_lock);
269 __unix_insert_socket(list, sk);
270 spin_unlock(&unix_table_lock);
271 }
272
273 static struct sock *__unix_find_socket_byname(struct sockaddr_un *sunname,
274 int len, int type, unsigned hash)
275 {
276 struct sock *s;
277 struct hlist_node *node;
278
279 sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
280 struct unix_sock *u = unix_sk(s);
281
282 if (u->addr->len == len &&
283 !memcmp(u->addr->name, sunname, len))
284 goto found;
285 }
286 s = NULL;
287 found:
288 return s;
289 }
290
291 static inline struct sock *unix_find_socket_byname(struct sockaddr_un *sunname,
292 int len, int type,
293 unsigned hash)
294 {
295 struct sock *s;
296
297 spin_lock(&unix_table_lock);
298 s = __unix_find_socket_byname(sunname, len, type, hash);
299 if (s)
300 sock_hold(s);
301 spin_unlock(&unix_table_lock);
302 return s;
303 }
304
305 static struct sock *unix_find_socket_byinode(struct inode *i)
306 {
307 struct sock *s;
308 struct hlist_node *node;
309
310 spin_lock(&unix_table_lock);
311 sk_for_each(s, node,
312 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
313 struct dentry *dentry = unix_sk(s)->dentry;
314
315 if(dentry && dentry->d_inode == i)
316 {
317 sock_hold(s);
318 goto found;
319 }
320 }
321 s = NULL;
322 found:
323 spin_unlock(&unix_table_lock);
324 return s;
325 }
326
327 static inline int unix_writable(struct sock *sk)
328 {
329 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
330 }
331
332 static void unix_write_space(struct sock *sk)
333 {
334 read_lock(&sk->sk_callback_lock);
335 if (unix_writable(sk)) {
336 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
337 wake_up_interruptible_sync(sk->sk_sleep);
338 sk_wake_async(sk, 2, POLL_OUT);
339 }
340 read_unlock(&sk->sk_callback_lock);
341 }
342
343 /* When dgram socket disconnects (or changes its peer), we clear its receive
344 * queue of packets arrived from previous peer. First, it allows to do
345 * flow control based only on wmem_alloc; second, sk connected to peer
346 * may receive messages only from that peer. */
347 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
348 {
349 if (!skb_queue_empty(&sk->sk_receive_queue)) {
350 skb_queue_purge(&sk->sk_receive_queue);
351 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
352
353 /* If one link of bidirectional dgram pipe is disconnected,
354 * we signal error. Messages are lost. Do not make this,
355 * when peer was not connected to us.
356 */
357 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
358 other->sk_err = ECONNRESET;
359 other->sk_error_report(other);
360 }
361 }
362 }
363
364 static void unix_sock_destructor(struct sock *sk)
365 {
366 struct unix_sock *u = unix_sk(sk);
367
368 skb_queue_purge(&sk->sk_receive_queue);
369
370 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
371 BUG_TRAP(sk_unhashed(sk));
372 BUG_TRAP(!sk->sk_socket);
373 if (!sock_flag(sk, SOCK_DEAD)) {
374 printk("Attempt to release alive unix socket: %p\n", sk);
375 return;
376 }
377
378 if (u->addr)
379 unix_release_addr(u->addr);
380
381 atomic_dec(&unix_nr_socks);
382 #ifdef UNIX_REFCNT_DEBUG
383 printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk, atomic_read(&unix_nr_socks));
384 #endif
385 }
386
387 static int unix_release_sock (struct sock *sk, int embrion)
388 {
389 struct unix_sock *u = unix_sk(sk);
390 struct dentry *dentry;
391 struct vfsmount *mnt;
392 struct sock *skpair;
393 struct sk_buff *skb;
394 int state;
395
396 unix_remove_socket(sk);
397
398 /* Clear state */
399 unix_state_lock(sk);
400 sock_orphan(sk);
401 sk->sk_shutdown = SHUTDOWN_MASK;
402 dentry = u->dentry;
403 u->dentry = NULL;
404 mnt = u->mnt;
405 u->mnt = NULL;
406 state = sk->sk_state;
407 sk->sk_state = TCP_CLOSE;
408 unix_state_unlock(sk);
409
410 wake_up_interruptible_all(&u->peer_wait);
411
412 skpair=unix_peer(sk);
413
414 if (skpair!=NULL) {
415 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
416 unix_state_lock(skpair);
417 /* No more writes */
418 skpair->sk_shutdown = SHUTDOWN_MASK;
419 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
420 skpair->sk_err = ECONNRESET;
421 unix_state_unlock(skpair);
422 skpair->sk_state_change(skpair);
423 read_lock(&skpair->sk_callback_lock);
424 sk_wake_async(skpair,1,POLL_HUP);
425 read_unlock(&skpair->sk_callback_lock);
426 }
427 sock_put(skpair); /* It may now die */
428 unix_peer(sk) = NULL;
429 }
430
431 /* Try to flush out this socket. Throw out buffers at least */
432
433 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
434 if (state==TCP_LISTEN)
435 unix_release_sock(skb->sk, 1);
436 /* passed fds are erased in the kfree_skb hook */
437 kfree_skb(skb);
438 }
439
440 if (dentry) {
441 dput(dentry);
442 mntput(mnt);
443 }
444
445 sock_put(sk);
446
447 /* ---- Socket is dead now and most probably destroyed ---- */
448
449 /*
450 * Fixme: BSD difference: In BSD all sockets connected to use get
451 * ECONNRESET and we die on the spot. In Linux we behave
452 * like files and pipes do and wait for the last
453 * dereference.
454 *
455 * Can't we simply set sock->err?
456 *
457 * What the above comment does talk about? --ANK(980817)
458 */
459
460 if (unix_tot_inflight)
461 unix_gc(); /* Garbage collect fds */
462
463 return 0;
464 }
465
466 static int unix_listen(struct socket *sock, int backlog)
467 {
468 int err;
469 struct sock *sk = sock->sk;
470 struct unix_sock *u = unix_sk(sk);
471
472 err = -EOPNOTSUPP;
473 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
474 goto out; /* Only stream/seqpacket sockets accept */
475 err = -EINVAL;
476 if (!u->addr)
477 goto out; /* No listens on an unbound socket */
478 unix_state_lock(sk);
479 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
480 goto out_unlock;
481 if (backlog > sk->sk_max_ack_backlog)
482 wake_up_interruptible_all(&u->peer_wait);
483 sk->sk_max_ack_backlog = backlog;
484 sk->sk_state = TCP_LISTEN;
485 /* set credentials so connect can copy them */
486 sk->sk_peercred.pid = task_tgid_vnr(current);
487 sk->sk_peercred.uid = current->euid;
488 sk->sk_peercred.gid = current->egid;
489 err = 0;
490
491 out_unlock:
492 unix_state_unlock(sk);
493 out:
494 return err;
495 }
496
497 static int unix_release(struct socket *);
498 static int unix_bind(struct socket *, struct sockaddr *, int);
499 static int unix_stream_connect(struct socket *, struct sockaddr *,
500 int addr_len, int flags);
501 static int unix_socketpair(struct socket *, struct socket *);
502 static int unix_accept(struct socket *, struct socket *, int);
503 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
504 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
505 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
506 static int unix_shutdown(struct socket *, int);
507 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
508 struct msghdr *, size_t);
509 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
510 struct msghdr *, size_t, int);
511 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
512 struct msghdr *, size_t);
513 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
514 struct msghdr *, size_t, int);
515 static int unix_dgram_connect(struct socket *, struct sockaddr *,
516 int, int);
517 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
518 struct msghdr *, size_t);
519
520 static const struct proto_ops unix_stream_ops = {
521 .family = PF_UNIX,
522 .owner = THIS_MODULE,
523 .release = unix_release,
524 .bind = unix_bind,
525 .connect = unix_stream_connect,
526 .socketpair = unix_socketpair,
527 .accept = unix_accept,
528 .getname = unix_getname,
529 .poll = unix_poll,
530 .ioctl = unix_ioctl,
531 .listen = unix_listen,
532 .shutdown = unix_shutdown,
533 .setsockopt = sock_no_setsockopt,
534 .getsockopt = sock_no_getsockopt,
535 .sendmsg = unix_stream_sendmsg,
536 .recvmsg = unix_stream_recvmsg,
537 .mmap = sock_no_mmap,
538 .sendpage = sock_no_sendpage,
539 };
540
541 static const struct proto_ops unix_dgram_ops = {
542 .family = PF_UNIX,
543 .owner = THIS_MODULE,
544 .release = unix_release,
545 .bind = unix_bind,
546 .connect = unix_dgram_connect,
547 .socketpair = unix_socketpair,
548 .accept = sock_no_accept,
549 .getname = unix_getname,
550 .poll = datagram_poll,
551 .ioctl = unix_ioctl,
552 .listen = sock_no_listen,
553 .shutdown = unix_shutdown,
554 .setsockopt = sock_no_setsockopt,
555 .getsockopt = sock_no_getsockopt,
556 .sendmsg = unix_dgram_sendmsg,
557 .recvmsg = unix_dgram_recvmsg,
558 .mmap = sock_no_mmap,
559 .sendpage = sock_no_sendpage,
560 };
561
562 static const struct proto_ops unix_seqpacket_ops = {
563 .family = PF_UNIX,
564 .owner = THIS_MODULE,
565 .release = unix_release,
566 .bind = unix_bind,
567 .connect = unix_stream_connect,
568 .socketpair = unix_socketpair,
569 .accept = unix_accept,
570 .getname = unix_getname,
571 .poll = datagram_poll,
572 .ioctl = unix_ioctl,
573 .listen = unix_listen,
574 .shutdown = unix_shutdown,
575 .setsockopt = sock_no_setsockopt,
576 .getsockopt = sock_no_getsockopt,
577 .sendmsg = unix_seqpacket_sendmsg,
578 .recvmsg = unix_dgram_recvmsg,
579 .mmap = sock_no_mmap,
580 .sendpage = sock_no_sendpage,
581 };
582
583 static struct proto unix_proto = {
584 .name = "UNIX",
585 .owner = THIS_MODULE,
586 .obj_size = sizeof(struct unix_sock),
587 };
588
589 /*
590 * AF_UNIX sockets do not interact with hardware, hence they
591 * dont trigger interrupts - so it's safe for them to have
592 * bh-unsafe locking for their sk_receive_queue.lock. Split off
593 * this special lock-class by reinitializing the spinlock key:
594 */
595 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
596
597 static struct sock * unix_create1(struct net *net, struct socket *sock)
598 {
599 struct sock *sk = NULL;
600 struct unix_sock *u;
601
602 atomic_inc(&unix_nr_socks);
603 if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
604 goto out;
605
606 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
607 if (!sk)
608 goto out;
609
610 sock_init_data(sock,sk);
611 lockdep_set_class(&sk->sk_receive_queue.lock,
612 &af_unix_sk_receive_queue_lock_key);
613
614 sk->sk_write_space = unix_write_space;
615 sk->sk_max_ack_backlog = sysctl_unix_max_dgram_qlen;
616 sk->sk_destruct = unix_sock_destructor;
617 u = unix_sk(sk);
618 u->dentry = NULL;
619 u->mnt = NULL;
620 spin_lock_init(&u->lock);
621 atomic_set(&u->inflight, 0);
622 INIT_LIST_HEAD(&u->link);
623 mutex_init(&u->readlock); /* single task reading lock */
624 init_waitqueue_head(&u->peer_wait);
625 unix_insert_socket(unix_sockets_unbound, sk);
626 out:
627 if (sk == NULL)
628 atomic_dec(&unix_nr_socks);
629 return sk;
630 }
631
632 static int unix_create(struct net *net, struct socket *sock, int protocol)
633 {
634 if (net != &init_net)
635 return -EAFNOSUPPORT;
636
637 if (protocol && protocol != PF_UNIX)
638 return -EPROTONOSUPPORT;
639
640 sock->state = SS_UNCONNECTED;
641
642 switch (sock->type) {
643 case SOCK_STREAM:
644 sock->ops = &unix_stream_ops;
645 break;
646 /*
647 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
648 * nothing uses it.
649 */
650 case SOCK_RAW:
651 sock->type=SOCK_DGRAM;
652 case SOCK_DGRAM:
653 sock->ops = &unix_dgram_ops;
654 break;
655 case SOCK_SEQPACKET:
656 sock->ops = &unix_seqpacket_ops;
657 break;
658 default:
659 return -ESOCKTNOSUPPORT;
660 }
661
662 return unix_create1(net, sock) ? 0 : -ENOMEM;
663 }
664
665 static int unix_release(struct socket *sock)
666 {
667 struct sock *sk = sock->sk;
668
669 if (!sk)
670 return 0;
671
672 sock->sk = NULL;
673
674 return unix_release_sock (sk, 0);
675 }
676
677 static int unix_autobind(struct socket *sock)
678 {
679 struct sock *sk = sock->sk;
680 struct unix_sock *u = unix_sk(sk);
681 static u32 ordernum = 1;
682 struct unix_address * addr;
683 int err;
684
685 mutex_lock(&u->readlock);
686
687 err = 0;
688 if (u->addr)
689 goto out;
690
691 err = -ENOMEM;
692 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
693 if (!addr)
694 goto out;
695
696 addr->name->sun_family = AF_UNIX;
697 atomic_set(&addr->refcnt, 1);
698
699 retry:
700 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
701 addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0));
702
703 spin_lock(&unix_table_lock);
704 ordernum = (ordernum+1)&0xFFFFF;
705
706 if (__unix_find_socket_byname(addr->name, addr->len, sock->type,
707 addr->hash)) {
708 spin_unlock(&unix_table_lock);
709 /* Sanity yield. It is unusual case, but yet... */
710 if (!(ordernum&0xFF))
711 yield();
712 goto retry;
713 }
714 addr->hash ^= sk->sk_type;
715
716 __unix_remove_socket(sk);
717 u->addr = addr;
718 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
719 spin_unlock(&unix_table_lock);
720 err = 0;
721
722 out: mutex_unlock(&u->readlock);
723 return err;
724 }
725
726 static struct sock *unix_find_other(struct sockaddr_un *sunname, int len,
727 int type, unsigned hash, int *error)
728 {
729 struct sock *u;
730 struct nameidata nd;
731 int err = 0;
732
733 if (sunname->sun_path[0]) {
734 err = path_lookup(sunname->sun_path, LOOKUP_FOLLOW, &nd);
735 if (err)
736 goto fail;
737 err = vfs_permission(&nd, MAY_WRITE);
738 if (err)
739 goto put_fail;
740
741 err = -ECONNREFUSED;
742 if (!S_ISSOCK(nd.dentry->d_inode->i_mode))
743 goto put_fail;
744 u=unix_find_socket_byinode(nd.dentry->d_inode);
745 if (!u)
746 goto put_fail;
747
748 if (u->sk_type == type)
749 touch_atime(nd.mnt, nd.dentry);
750
751 path_release(&nd);
752
753 err=-EPROTOTYPE;
754 if (u->sk_type != type) {
755 sock_put(u);
756 goto fail;
757 }
758 } else {
759 err = -ECONNREFUSED;
760 u=unix_find_socket_byname(sunname, len, type, hash);
761 if (u) {
762 struct dentry *dentry;
763 dentry = unix_sk(u)->dentry;
764 if (dentry)
765 touch_atime(unix_sk(u)->mnt, dentry);
766 } else
767 goto fail;
768 }
769 return u;
770
771 put_fail:
772 path_release(&nd);
773 fail:
774 *error=err;
775 return NULL;
776 }
777
778
779 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
780 {
781 struct sock *sk = sock->sk;
782 struct unix_sock *u = unix_sk(sk);
783 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
784 struct dentry * dentry = NULL;
785 struct nameidata nd;
786 int err;
787 unsigned hash;
788 struct unix_address *addr;
789 struct hlist_head *list;
790
791 err = -EINVAL;
792 if (sunaddr->sun_family != AF_UNIX)
793 goto out;
794
795 if (addr_len==sizeof(short)) {
796 err = unix_autobind(sock);
797 goto out;
798 }
799
800 err = unix_mkname(sunaddr, addr_len, &hash);
801 if (err < 0)
802 goto out;
803 addr_len = err;
804
805 mutex_lock(&u->readlock);
806
807 err = -EINVAL;
808 if (u->addr)
809 goto out_up;
810
811 err = -ENOMEM;
812 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
813 if (!addr)
814 goto out_up;
815
816 memcpy(addr->name, sunaddr, addr_len);
817 addr->len = addr_len;
818 addr->hash = hash ^ sk->sk_type;
819 atomic_set(&addr->refcnt, 1);
820
821 if (sunaddr->sun_path[0]) {
822 unsigned int mode;
823 err = 0;
824 /*
825 * Get the parent directory, calculate the hash for last
826 * component.
827 */
828 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
829 if (err)
830 goto out_mknod_parent;
831
832 dentry = lookup_create(&nd, 0);
833 err = PTR_ERR(dentry);
834 if (IS_ERR(dentry))
835 goto out_mknod_unlock;
836
837 /*
838 * All right, let's create it.
839 */
840 mode = S_IFSOCK |
841 (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
842 err = vfs_mknod(nd.dentry->d_inode, dentry, mode, 0);
843 if (err)
844 goto out_mknod_dput;
845 mutex_unlock(&nd.dentry->d_inode->i_mutex);
846 dput(nd.dentry);
847 nd.dentry = dentry;
848
849 addr->hash = UNIX_HASH_SIZE;
850 }
851
852 spin_lock(&unix_table_lock);
853
854 if (!sunaddr->sun_path[0]) {
855 err = -EADDRINUSE;
856 if (__unix_find_socket_byname(sunaddr, addr_len,
857 sk->sk_type, hash)) {
858 unix_release_addr(addr);
859 goto out_unlock;
860 }
861
862 list = &unix_socket_table[addr->hash];
863 } else {
864 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
865 u->dentry = nd.dentry;
866 u->mnt = nd.mnt;
867 }
868
869 err = 0;
870 __unix_remove_socket(sk);
871 u->addr = addr;
872 __unix_insert_socket(list, sk);
873
874 out_unlock:
875 spin_unlock(&unix_table_lock);
876 out_up:
877 mutex_unlock(&u->readlock);
878 out:
879 return err;
880
881 out_mknod_dput:
882 dput(dentry);
883 out_mknod_unlock:
884 mutex_unlock(&nd.dentry->d_inode->i_mutex);
885 path_release(&nd);
886 out_mknod_parent:
887 if (err==-EEXIST)
888 err=-EADDRINUSE;
889 unix_release_addr(addr);
890 goto out_up;
891 }
892
893 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
894 {
895 if (unlikely(sk1 == sk2) || !sk2) {
896 unix_state_lock(sk1);
897 return;
898 }
899 if (sk1 < sk2) {
900 unix_state_lock(sk1);
901 unix_state_lock_nested(sk2);
902 } else {
903 unix_state_lock(sk2);
904 unix_state_lock_nested(sk1);
905 }
906 }
907
908 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
909 {
910 if (unlikely(sk1 == sk2) || !sk2) {
911 unix_state_unlock(sk1);
912 return;
913 }
914 unix_state_unlock(sk1);
915 unix_state_unlock(sk2);
916 }
917
918 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
919 int alen, int flags)
920 {
921 struct sock *sk = sock->sk;
922 struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr;
923 struct sock *other;
924 unsigned hash;
925 int err;
926
927 if (addr->sa_family != AF_UNSPEC) {
928 err = unix_mkname(sunaddr, alen, &hash);
929 if (err < 0)
930 goto out;
931 alen = err;
932
933 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
934 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
935 goto out;
936
937 restart:
938 other=unix_find_other(sunaddr, alen, sock->type, hash, &err);
939 if (!other)
940 goto out;
941
942 unix_state_double_lock(sk, other);
943
944 /* Apparently VFS overslept socket death. Retry. */
945 if (sock_flag(other, SOCK_DEAD)) {
946 unix_state_double_unlock(sk, other);
947 sock_put(other);
948 goto restart;
949 }
950
951 err = -EPERM;
952 if (!unix_may_send(sk, other))
953 goto out_unlock;
954
955 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
956 if (err)
957 goto out_unlock;
958
959 } else {
960 /*
961 * 1003.1g breaking connected state with AF_UNSPEC
962 */
963 other = NULL;
964 unix_state_double_lock(sk, other);
965 }
966
967 /*
968 * If it was connected, reconnect.
969 */
970 if (unix_peer(sk)) {
971 struct sock *old_peer = unix_peer(sk);
972 unix_peer(sk)=other;
973 unix_state_double_unlock(sk, other);
974
975 if (other != old_peer)
976 unix_dgram_disconnected(sk, old_peer);
977 sock_put(old_peer);
978 } else {
979 unix_peer(sk)=other;
980 unix_state_double_unlock(sk, other);
981 }
982 return 0;
983
984 out_unlock:
985 unix_state_double_unlock(sk, other);
986 sock_put(other);
987 out:
988 return err;
989 }
990
991 static long unix_wait_for_peer(struct sock *other, long timeo)
992 {
993 struct unix_sock *u = unix_sk(other);
994 int sched;
995 DEFINE_WAIT(wait);
996
997 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
998
999 sched = !sock_flag(other, SOCK_DEAD) &&
1000 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1001 (skb_queue_len(&other->sk_receive_queue) >
1002 other->sk_max_ack_backlog);
1003
1004 unix_state_unlock(other);
1005
1006 if (sched)
1007 timeo = schedule_timeout(timeo);
1008
1009 finish_wait(&u->peer_wait, &wait);
1010 return timeo;
1011 }
1012
1013 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1014 int addr_len, int flags)
1015 {
1016 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1017 struct sock *sk = sock->sk;
1018 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1019 struct sock *newsk = NULL;
1020 struct sock *other = NULL;
1021 struct sk_buff *skb = NULL;
1022 unsigned hash;
1023 int st;
1024 int err;
1025 long timeo;
1026
1027 err = unix_mkname(sunaddr, addr_len, &hash);
1028 if (err < 0)
1029 goto out;
1030 addr_len = err;
1031
1032 if (test_bit(SOCK_PASSCRED, &sock->flags)
1033 && !u->addr && (err = unix_autobind(sock)) != 0)
1034 goto out;
1035
1036 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1037
1038 /* First of all allocate resources.
1039 If we will make it after state is locked,
1040 we will have to recheck all again in any case.
1041 */
1042
1043 err = -ENOMEM;
1044
1045 /* create new sock for complete connection */
1046 newsk = unix_create1(sk->sk_net, NULL);
1047 if (newsk == NULL)
1048 goto out;
1049
1050 /* Allocate skb for sending to listening sock */
1051 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1052 if (skb == NULL)
1053 goto out;
1054
1055 restart:
1056 /* Find listening sock. */
1057 other = unix_find_other(sunaddr, addr_len, sk->sk_type, hash, &err);
1058 if (!other)
1059 goto out;
1060
1061 /* Latch state of peer */
1062 unix_state_lock(other);
1063
1064 /* Apparently VFS overslept socket death. Retry. */
1065 if (sock_flag(other, SOCK_DEAD)) {
1066 unix_state_unlock(other);
1067 sock_put(other);
1068 goto restart;
1069 }
1070
1071 err = -ECONNREFUSED;
1072 if (other->sk_state != TCP_LISTEN)
1073 goto out_unlock;
1074
1075 if (skb_queue_len(&other->sk_receive_queue) >
1076 other->sk_max_ack_backlog) {
1077 err = -EAGAIN;
1078 if (!timeo)
1079 goto out_unlock;
1080
1081 timeo = unix_wait_for_peer(other, timeo);
1082
1083 err = sock_intr_errno(timeo);
1084 if (signal_pending(current))
1085 goto out;
1086 sock_put(other);
1087 goto restart;
1088 }
1089
1090 /* Latch our state.
1091
1092 It is tricky place. We need to grab write lock and cannot
1093 drop lock on peer. It is dangerous because deadlock is
1094 possible. Connect to self case and simultaneous
1095 attempt to connect are eliminated by checking socket
1096 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1097 check this before attempt to grab lock.
1098
1099 Well, and we have to recheck the state after socket locked.
1100 */
1101 st = sk->sk_state;
1102
1103 switch (st) {
1104 case TCP_CLOSE:
1105 /* This is ok... continue with connect */
1106 break;
1107 case TCP_ESTABLISHED:
1108 /* Socket is already connected */
1109 err = -EISCONN;
1110 goto out_unlock;
1111 default:
1112 err = -EINVAL;
1113 goto out_unlock;
1114 }
1115
1116 unix_state_lock_nested(sk);
1117
1118 if (sk->sk_state != st) {
1119 unix_state_unlock(sk);
1120 unix_state_unlock(other);
1121 sock_put(other);
1122 goto restart;
1123 }
1124
1125 err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1126 if (err) {
1127 unix_state_unlock(sk);
1128 goto out_unlock;
1129 }
1130
1131 /* The way is open! Fastly set all the necessary fields... */
1132
1133 sock_hold(sk);
1134 unix_peer(newsk) = sk;
1135 newsk->sk_state = TCP_ESTABLISHED;
1136 newsk->sk_type = sk->sk_type;
1137 newsk->sk_peercred.pid = task_tgid_vnr(current);
1138 newsk->sk_peercred.uid = current->euid;
1139 newsk->sk_peercred.gid = current->egid;
1140 newu = unix_sk(newsk);
1141 newsk->sk_sleep = &newu->peer_wait;
1142 otheru = unix_sk(other);
1143
1144 /* copy address information from listening to new sock*/
1145 if (otheru->addr) {
1146 atomic_inc(&otheru->addr->refcnt);
1147 newu->addr = otheru->addr;
1148 }
1149 if (otheru->dentry) {
1150 newu->dentry = dget(otheru->dentry);
1151 newu->mnt = mntget(otheru->mnt);
1152 }
1153
1154 /* Set credentials */
1155 sk->sk_peercred = other->sk_peercred;
1156
1157 sock->state = SS_CONNECTED;
1158 sk->sk_state = TCP_ESTABLISHED;
1159 sock_hold(newsk);
1160
1161 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1162 unix_peer(sk) = newsk;
1163
1164 unix_state_unlock(sk);
1165
1166 /* take ten and and send info to listening sock */
1167 spin_lock(&other->sk_receive_queue.lock);
1168 __skb_queue_tail(&other->sk_receive_queue, skb);
1169 spin_unlock(&other->sk_receive_queue.lock);
1170 unix_state_unlock(other);
1171 other->sk_data_ready(other, 0);
1172 sock_put(other);
1173 return 0;
1174
1175 out_unlock:
1176 if (other)
1177 unix_state_unlock(other);
1178
1179 out:
1180 if (skb)
1181 kfree_skb(skb);
1182 if (newsk)
1183 unix_release_sock(newsk, 0);
1184 if (other)
1185 sock_put(other);
1186 return err;
1187 }
1188
1189 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1190 {
1191 struct sock *ska=socka->sk, *skb = sockb->sk;
1192
1193 /* Join our sockets back to back */
1194 sock_hold(ska);
1195 sock_hold(skb);
1196 unix_peer(ska)=skb;
1197 unix_peer(skb)=ska;
1198 ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1199 ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1200 ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1201
1202 if (ska->sk_type != SOCK_DGRAM) {
1203 ska->sk_state = TCP_ESTABLISHED;
1204 skb->sk_state = TCP_ESTABLISHED;
1205 socka->state = SS_CONNECTED;
1206 sockb->state = SS_CONNECTED;
1207 }
1208 return 0;
1209 }
1210
1211 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1212 {
1213 struct sock *sk = sock->sk;
1214 struct sock *tsk;
1215 struct sk_buff *skb;
1216 int err;
1217
1218 err = -EOPNOTSUPP;
1219 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
1220 goto out;
1221
1222 err = -EINVAL;
1223 if (sk->sk_state != TCP_LISTEN)
1224 goto out;
1225
1226 /* If socket state is TCP_LISTEN it cannot change (for now...),
1227 * so that no locks are necessary.
1228 */
1229
1230 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1231 if (!skb) {
1232 /* This means receive shutdown. */
1233 if (err == 0)
1234 err = -EINVAL;
1235 goto out;
1236 }
1237
1238 tsk = skb->sk;
1239 skb_free_datagram(sk, skb);
1240 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1241
1242 /* attach accepted sock to socket */
1243 unix_state_lock(tsk);
1244 newsock->state = SS_CONNECTED;
1245 sock_graft(tsk, newsock);
1246 unix_state_unlock(tsk);
1247 return 0;
1248
1249 out:
1250 return err;
1251 }
1252
1253
1254 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1255 {
1256 struct sock *sk = sock->sk;
1257 struct unix_sock *u;
1258 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1259 int err = 0;
1260
1261 if (peer) {
1262 sk = unix_peer_get(sk);
1263
1264 err = -ENOTCONN;
1265 if (!sk)
1266 goto out;
1267 err = 0;
1268 } else {
1269 sock_hold(sk);
1270 }
1271
1272 u = unix_sk(sk);
1273 unix_state_lock(sk);
1274 if (!u->addr) {
1275 sunaddr->sun_family = AF_UNIX;
1276 sunaddr->sun_path[0] = 0;
1277 *uaddr_len = sizeof(short);
1278 } else {
1279 struct unix_address *addr = u->addr;
1280
1281 *uaddr_len = addr->len;
1282 memcpy(sunaddr, addr->name, *uaddr_len);
1283 }
1284 unix_state_unlock(sk);
1285 sock_put(sk);
1286 out:
1287 return err;
1288 }
1289
1290 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1291 {
1292 int i;
1293
1294 scm->fp = UNIXCB(skb).fp;
1295 skb->destructor = sock_wfree;
1296 UNIXCB(skb).fp = NULL;
1297
1298 for (i=scm->fp->count-1; i>=0; i--)
1299 unix_notinflight(scm->fp->fp[i]);
1300 }
1301
1302 static void unix_destruct_fds(struct sk_buff *skb)
1303 {
1304 struct scm_cookie scm;
1305 memset(&scm, 0, sizeof(scm));
1306 unix_detach_fds(&scm, skb);
1307
1308 /* Alas, it calls VFS */
1309 /* So fscking what? fput() had been SMP-safe since the last Summer */
1310 scm_destroy(&scm);
1311 sock_wfree(skb);
1312 }
1313
1314 static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1315 {
1316 int i;
1317 for (i=scm->fp->count-1; i>=0; i--)
1318 unix_inflight(scm->fp->fp[i]);
1319 UNIXCB(skb).fp = scm->fp;
1320 skb->destructor = unix_destruct_fds;
1321 scm->fp = NULL;
1322 }
1323
1324 /*
1325 * Send AF_UNIX data.
1326 */
1327
1328 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1329 struct msghdr *msg, size_t len)
1330 {
1331 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1332 struct sock *sk = sock->sk;
1333 struct unix_sock *u = unix_sk(sk);
1334 struct sockaddr_un *sunaddr=msg->msg_name;
1335 struct sock *other = NULL;
1336 int namelen = 0; /* fake GCC */
1337 int err;
1338 unsigned hash;
1339 struct sk_buff *skb;
1340 long timeo;
1341 struct scm_cookie tmp_scm;
1342
1343 if (NULL == siocb->scm)
1344 siocb->scm = &tmp_scm;
1345 err = scm_send(sock, msg, siocb->scm);
1346 if (err < 0)
1347 return err;
1348
1349 err = -EOPNOTSUPP;
1350 if (msg->msg_flags&MSG_OOB)
1351 goto out;
1352
1353 if (msg->msg_namelen) {
1354 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1355 if (err < 0)
1356 goto out;
1357 namelen = err;
1358 } else {
1359 sunaddr = NULL;
1360 err = -ENOTCONN;
1361 other = unix_peer_get(sk);
1362 if (!other)
1363 goto out;
1364 }
1365
1366 if (test_bit(SOCK_PASSCRED, &sock->flags)
1367 && !u->addr && (err = unix_autobind(sock)) != 0)
1368 goto out;
1369
1370 err = -EMSGSIZE;
1371 if (len > sk->sk_sndbuf - 32)
1372 goto out;
1373
1374 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1375 if (skb==NULL)
1376 goto out;
1377
1378 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1379 if (siocb->scm->fp)
1380 unix_attach_fds(siocb->scm, skb);
1381 unix_get_secdata(siocb->scm, skb);
1382
1383 skb_reset_transport_header(skb);
1384 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1385 if (err)
1386 goto out_free;
1387
1388 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1389
1390 restart:
1391 if (!other) {
1392 err = -ECONNRESET;
1393 if (sunaddr == NULL)
1394 goto out_free;
1395
1396 other = unix_find_other(sunaddr, namelen, sk->sk_type,
1397 hash, &err);
1398 if (other==NULL)
1399 goto out_free;
1400 }
1401
1402 unix_state_lock(other);
1403 err = -EPERM;
1404 if (!unix_may_send(sk, other))
1405 goto out_unlock;
1406
1407 if (sock_flag(other, SOCK_DEAD)) {
1408 /*
1409 * Check with 1003.1g - what should
1410 * datagram error
1411 */
1412 unix_state_unlock(other);
1413 sock_put(other);
1414
1415 err = 0;
1416 unix_state_lock(sk);
1417 if (unix_peer(sk) == other) {
1418 unix_peer(sk)=NULL;
1419 unix_state_unlock(sk);
1420
1421 unix_dgram_disconnected(sk, other);
1422 sock_put(other);
1423 err = -ECONNREFUSED;
1424 } else {
1425 unix_state_unlock(sk);
1426 }
1427
1428 other = NULL;
1429 if (err)
1430 goto out_free;
1431 goto restart;
1432 }
1433
1434 err = -EPIPE;
1435 if (other->sk_shutdown & RCV_SHUTDOWN)
1436 goto out_unlock;
1437
1438 if (sk->sk_type != SOCK_SEQPACKET) {
1439 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1440 if (err)
1441 goto out_unlock;
1442 }
1443
1444 if (unix_peer(other) != sk &&
1445 (skb_queue_len(&other->sk_receive_queue) >
1446 other->sk_max_ack_backlog)) {
1447 if (!timeo) {
1448 err = -EAGAIN;
1449 goto out_unlock;
1450 }
1451
1452 timeo = unix_wait_for_peer(other, timeo);
1453
1454 err = sock_intr_errno(timeo);
1455 if (signal_pending(current))
1456 goto out_free;
1457
1458 goto restart;
1459 }
1460
1461 skb_queue_tail(&other->sk_receive_queue, skb);
1462 unix_state_unlock(other);
1463 other->sk_data_ready(other, len);
1464 sock_put(other);
1465 scm_destroy(siocb->scm);
1466 return len;
1467
1468 out_unlock:
1469 unix_state_unlock(other);
1470 out_free:
1471 kfree_skb(skb);
1472 out:
1473 if (other)
1474 sock_put(other);
1475 scm_destroy(siocb->scm);
1476 return err;
1477 }
1478
1479
1480 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1481 struct msghdr *msg, size_t len)
1482 {
1483 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1484 struct sock *sk = sock->sk;
1485 struct sock *other = NULL;
1486 struct sockaddr_un *sunaddr=msg->msg_name;
1487 int err,size;
1488 struct sk_buff *skb;
1489 int sent=0;
1490 struct scm_cookie tmp_scm;
1491
1492 if (NULL == siocb->scm)
1493 siocb->scm = &tmp_scm;
1494 err = scm_send(sock, msg, siocb->scm);
1495 if (err < 0)
1496 return err;
1497
1498 err = -EOPNOTSUPP;
1499 if (msg->msg_flags&MSG_OOB)
1500 goto out_err;
1501
1502 if (msg->msg_namelen) {
1503 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1504 goto out_err;
1505 } else {
1506 sunaddr = NULL;
1507 err = -ENOTCONN;
1508 other = unix_peer(sk);
1509 if (!other)
1510 goto out_err;
1511 }
1512
1513 if (sk->sk_shutdown & SEND_SHUTDOWN)
1514 goto pipe_err;
1515
1516 while(sent < len)
1517 {
1518 /*
1519 * Optimisation for the fact that under 0.01% of X
1520 * messages typically need breaking up.
1521 */
1522
1523 size = len-sent;
1524
1525 /* Keep two messages in the pipe so it schedules better */
1526 if (size > ((sk->sk_sndbuf >> 1) - 64))
1527 size = (sk->sk_sndbuf >> 1) - 64;
1528
1529 if (size > SKB_MAX_ALLOC)
1530 size = SKB_MAX_ALLOC;
1531
1532 /*
1533 * Grab a buffer
1534 */
1535
1536 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
1537
1538 if (skb==NULL)
1539 goto out_err;
1540
1541 /*
1542 * If you pass two values to the sock_alloc_send_skb
1543 * it tries to grab the large buffer with GFP_NOFS
1544 * (which can fail easily), and if it fails grab the
1545 * fallback size buffer which is under a page and will
1546 * succeed. [Alan]
1547 */
1548 size = min_t(int, size, skb_tailroom(skb));
1549
1550 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1551 if (siocb->scm->fp)
1552 unix_attach_fds(siocb->scm, skb);
1553
1554 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
1555 kfree_skb(skb);
1556 goto out_err;
1557 }
1558
1559 unix_state_lock(other);
1560
1561 if (sock_flag(other, SOCK_DEAD) ||
1562 (other->sk_shutdown & RCV_SHUTDOWN))
1563 goto pipe_err_free;
1564
1565 skb_queue_tail(&other->sk_receive_queue, skb);
1566 unix_state_unlock(other);
1567 other->sk_data_ready(other, size);
1568 sent+=size;
1569 }
1570
1571 scm_destroy(siocb->scm);
1572 siocb->scm = NULL;
1573
1574 return sent;
1575
1576 pipe_err_free:
1577 unix_state_unlock(other);
1578 kfree_skb(skb);
1579 pipe_err:
1580 if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL))
1581 send_sig(SIGPIPE,current,0);
1582 err = -EPIPE;
1583 out_err:
1584 scm_destroy(siocb->scm);
1585 siocb->scm = NULL;
1586 return sent ? : err;
1587 }
1588
1589 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1590 struct msghdr *msg, size_t len)
1591 {
1592 int err;
1593 struct sock *sk = sock->sk;
1594
1595 err = sock_error(sk);
1596 if (err)
1597 return err;
1598
1599 if (sk->sk_state != TCP_ESTABLISHED)
1600 return -ENOTCONN;
1601
1602 if (msg->msg_namelen)
1603 msg->msg_namelen = 0;
1604
1605 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1606 }
1607
1608 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1609 {
1610 struct unix_sock *u = unix_sk(sk);
1611
1612 msg->msg_namelen = 0;
1613 if (u->addr) {
1614 msg->msg_namelen = u->addr->len;
1615 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1616 }
1617 }
1618
1619 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1620 struct msghdr *msg, size_t size,
1621 int flags)
1622 {
1623 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1624 struct scm_cookie tmp_scm;
1625 struct sock *sk = sock->sk;
1626 struct unix_sock *u = unix_sk(sk);
1627 int noblock = flags & MSG_DONTWAIT;
1628 struct sk_buff *skb;
1629 int err;
1630
1631 err = -EOPNOTSUPP;
1632 if (flags&MSG_OOB)
1633 goto out;
1634
1635 msg->msg_namelen = 0;
1636
1637 mutex_lock(&u->readlock);
1638
1639 skb = skb_recv_datagram(sk, flags, noblock, &err);
1640 if (!skb) {
1641 unix_state_lock(sk);
1642 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1643 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1644 (sk->sk_shutdown & RCV_SHUTDOWN))
1645 err = 0;
1646 unix_state_unlock(sk);
1647 goto out_unlock;
1648 }
1649
1650 wake_up_interruptible_sync(&u->peer_wait);
1651
1652 if (msg->msg_name)
1653 unix_copy_addr(msg, skb->sk);
1654
1655 if (size > skb->len)
1656 size = skb->len;
1657 else if (size < skb->len)
1658 msg->msg_flags |= MSG_TRUNC;
1659
1660 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1661 if (err)
1662 goto out_free;
1663
1664 if (!siocb->scm) {
1665 siocb->scm = &tmp_scm;
1666 memset(&tmp_scm, 0, sizeof(tmp_scm));
1667 }
1668 siocb->scm->creds = *UNIXCREDS(skb);
1669 unix_set_secdata(siocb->scm, skb);
1670
1671 if (!(flags & MSG_PEEK))
1672 {
1673 if (UNIXCB(skb).fp)
1674 unix_detach_fds(siocb->scm, skb);
1675 }
1676 else
1677 {
1678 /* It is questionable: on PEEK we could:
1679 - do not return fds - good, but too simple 8)
1680 - return fds, and do not return them on read (old strategy,
1681 apparently wrong)
1682 - clone fds (I chose it for now, it is the most universal
1683 solution)
1684
1685 POSIX 1003.1g does not actually define this clearly
1686 at all. POSIX 1003.1g doesn't define a lot of things
1687 clearly however!
1688
1689 */
1690 if (UNIXCB(skb).fp)
1691 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1692 }
1693 err = size;
1694
1695 scm_recv(sock, msg, siocb->scm, flags);
1696
1697 out_free:
1698 skb_free_datagram(sk,skb);
1699 out_unlock:
1700 mutex_unlock(&u->readlock);
1701 out:
1702 return err;
1703 }
1704
1705 /*
1706 * Sleep until data has arrive. But check for races..
1707 */
1708
1709 static long unix_stream_data_wait(struct sock * sk, long timeo)
1710 {
1711 DEFINE_WAIT(wait);
1712
1713 unix_state_lock(sk);
1714
1715 for (;;) {
1716 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1717
1718 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1719 sk->sk_err ||
1720 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1721 signal_pending(current) ||
1722 !timeo)
1723 break;
1724
1725 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1726 unix_state_unlock(sk);
1727 timeo = schedule_timeout(timeo);
1728 unix_state_lock(sk);
1729 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1730 }
1731
1732 finish_wait(sk->sk_sleep, &wait);
1733 unix_state_unlock(sk);
1734 return timeo;
1735 }
1736
1737
1738
1739 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1740 struct msghdr *msg, size_t size,
1741 int flags)
1742 {
1743 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1744 struct scm_cookie tmp_scm;
1745 struct sock *sk = sock->sk;
1746 struct unix_sock *u = unix_sk(sk);
1747 struct sockaddr_un *sunaddr=msg->msg_name;
1748 int copied = 0;
1749 int check_creds = 0;
1750 int target;
1751 int err = 0;
1752 long timeo;
1753
1754 err = -EINVAL;
1755 if (sk->sk_state != TCP_ESTABLISHED)
1756 goto out;
1757
1758 err = -EOPNOTSUPP;
1759 if (flags&MSG_OOB)
1760 goto out;
1761
1762 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1763 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1764
1765 msg->msg_namelen = 0;
1766
1767 /* Lock the socket to prevent queue disordering
1768 * while sleeps in memcpy_tomsg
1769 */
1770
1771 if (!siocb->scm) {
1772 siocb->scm = &tmp_scm;
1773 memset(&tmp_scm, 0, sizeof(tmp_scm));
1774 }
1775
1776 mutex_lock(&u->readlock);
1777
1778 do
1779 {
1780 int chunk;
1781 struct sk_buff *skb;
1782
1783 unix_state_lock(sk);
1784 skb = skb_dequeue(&sk->sk_receive_queue);
1785 if (skb==NULL)
1786 {
1787 if (copied >= target)
1788 goto unlock;
1789
1790 /*
1791 * POSIX 1003.1g mandates this order.
1792 */
1793
1794 if ((err = sock_error(sk)) != 0)
1795 goto unlock;
1796 if (sk->sk_shutdown & RCV_SHUTDOWN)
1797 goto unlock;
1798
1799 unix_state_unlock(sk);
1800 err = -EAGAIN;
1801 if (!timeo)
1802 break;
1803 mutex_unlock(&u->readlock);
1804
1805 timeo = unix_stream_data_wait(sk, timeo);
1806
1807 if (signal_pending(current)) {
1808 err = sock_intr_errno(timeo);
1809 goto out;
1810 }
1811 mutex_lock(&u->readlock);
1812 continue;
1813 unlock:
1814 unix_state_unlock(sk);
1815 break;
1816 }
1817 unix_state_unlock(sk);
1818
1819 if (check_creds) {
1820 /* Never glue messages from different writers */
1821 if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) {
1822 skb_queue_head(&sk->sk_receive_queue, skb);
1823 break;
1824 }
1825 } else {
1826 /* Copy credentials */
1827 siocb->scm->creds = *UNIXCREDS(skb);
1828 check_creds = 1;
1829 }
1830
1831 /* Copy address just once */
1832 if (sunaddr)
1833 {
1834 unix_copy_addr(msg, skb->sk);
1835 sunaddr = NULL;
1836 }
1837
1838 chunk = min_t(unsigned int, skb->len, size);
1839 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1840 skb_queue_head(&sk->sk_receive_queue, skb);
1841 if (copied == 0)
1842 copied = -EFAULT;
1843 break;
1844 }
1845 copied += chunk;
1846 size -= chunk;
1847
1848 /* Mark read part of skb as used */
1849 if (!(flags & MSG_PEEK))
1850 {
1851 skb_pull(skb, chunk);
1852
1853 if (UNIXCB(skb).fp)
1854 unix_detach_fds(siocb->scm, skb);
1855
1856 /* put the skb back if we didn't use it up.. */
1857 if (skb->len)
1858 {
1859 skb_queue_head(&sk->sk_receive_queue, skb);
1860 break;
1861 }
1862
1863 kfree_skb(skb);
1864
1865 if (siocb->scm->fp)
1866 break;
1867 }
1868 else
1869 {
1870 /* It is questionable, see note in unix_dgram_recvmsg.
1871 */
1872 if (UNIXCB(skb).fp)
1873 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1874
1875 /* put message back and return */
1876 skb_queue_head(&sk->sk_receive_queue, skb);
1877 break;
1878 }
1879 } while (size);
1880
1881 mutex_unlock(&u->readlock);
1882 scm_recv(sock, msg, siocb->scm, flags);
1883 out:
1884 return copied ? : err;
1885 }
1886
1887 static int unix_shutdown(struct socket *sock, int mode)
1888 {
1889 struct sock *sk = sock->sk;
1890 struct sock *other;
1891
1892 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1893
1894 if (mode) {
1895 unix_state_lock(sk);
1896 sk->sk_shutdown |= mode;
1897 other=unix_peer(sk);
1898 if (other)
1899 sock_hold(other);
1900 unix_state_unlock(sk);
1901 sk->sk_state_change(sk);
1902
1903 if (other &&
1904 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1905
1906 int peer_mode = 0;
1907
1908 if (mode&RCV_SHUTDOWN)
1909 peer_mode |= SEND_SHUTDOWN;
1910 if (mode&SEND_SHUTDOWN)
1911 peer_mode |= RCV_SHUTDOWN;
1912 unix_state_lock(other);
1913 other->sk_shutdown |= peer_mode;
1914 unix_state_unlock(other);
1915 other->sk_state_change(other);
1916 read_lock(&other->sk_callback_lock);
1917 if (peer_mode == SHUTDOWN_MASK)
1918 sk_wake_async(other,1,POLL_HUP);
1919 else if (peer_mode & RCV_SHUTDOWN)
1920 sk_wake_async(other,1,POLL_IN);
1921 read_unlock(&other->sk_callback_lock);
1922 }
1923 if (other)
1924 sock_put(other);
1925 }
1926 return 0;
1927 }
1928
1929 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1930 {
1931 struct sock *sk = sock->sk;
1932 long amount=0;
1933 int err;
1934
1935 switch(cmd)
1936 {
1937 case SIOCOUTQ:
1938 amount = atomic_read(&sk->sk_wmem_alloc);
1939 err = put_user(amount, (int __user *)arg);
1940 break;
1941 case SIOCINQ:
1942 {
1943 struct sk_buff *skb;
1944
1945 if (sk->sk_state == TCP_LISTEN) {
1946 err = -EINVAL;
1947 break;
1948 }
1949
1950 spin_lock(&sk->sk_receive_queue.lock);
1951 if (sk->sk_type == SOCK_STREAM ||
1952 sk->sk_type == SOCK_SEQPACKET) {
1953 skb_queue_walk(&sk->sk_receive_queue, skb)
1954 amount += skb->len;
1955 } else {
1956 skb = skb_peek(&sk->sk_receive_queue);
1957 if (skb)
1958 amount=skb->len;
1959 }
1960 spin_unlock(&sk->sk_receive_queue.lock);
1961 err = put_user(amount, (int __user *)arg);
1962 break;
1963 }
1964
1965 default:
1966 err = -ENOIOCTLCMD;
1967 break;
1968 }
1969 return err;
1970 }
1971
1972 static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait)
1973 {
1974 struct sock *sk = sock->sk;
1975 unsigned int mask;
1976
1977 poll_wait(file, sk->sk_sleep, wait);
1978 mask = 0;
1979
1980 /* exceptional events? */
1981 if (sk->sk_err)
1982 mask |= POLLERR;
1983 if (sk->sk_shutdown == SHUTDOWN_MASK)
1984 mask |= POLLHUP;
1985 if (sk->sk_shutdown & RCV_SHUTDOWN)
1986 mask |= POLLRDHUP;
1987
1988 /* readable? */
1989 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1990 (sk->sk_shutdown & RCV_SHUTDOWN))
1991 mask |= POLLIN | POLLRDNORM;
1992
1993 /* Connection-based need to check for termination and startup */
1994 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE)
1995 mask |= POLLHUP;
1996
1997 /*
1998 * we set writable also when the other side has shut down the
1999 * connection. This prevents stuck sockets.
2000 */
2001 if (unix_writable(sk))
2002 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2003
2004 return mask;
2005 }
2006
2007
2008 #ifdef CONFIG_PROC_FS
2009 static struct sock *unix_seq_idx(int *iter, loff_t pos)
2010 {
2011 loff_t off = 0;
2012 struct sock *s;
2013
2014 for (s = first_unix_socket(iter); s; s = next_unix_socket(iter, s)) {
2015 if (off == pos)
2016 return s;
2017 ++off;
2018 }
2019 return NULL;
2020 }
2021
2022
2023 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2024 {
2025 spin_lock(&unix_table_lock);
2026 return *pos ? unix_seq_idx(seq->private, *pos - 1) : ((void *) 1);
2027 }
2028
2029 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2030 {
2031 ++*pos;
2032
2033 if (v == (void *)1)
2034 return first_unix_socket(seq->private);
2035 return next_unix_socket(seq->private, v);
2036 }
2037
2038 static void unix_seq_stop(struct seq_file *seq, void *v)
2039 {
2040 spin_unlock(&unix_table_lock);
2041 }
2042
2043 static int unix_seq_show(struct seq_file *seq, void *v)
2044 {
2045
2046 if (v == (void *)1)
2047 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2048 "Inode Path\n");
2049 else {
2050 struct sock *s = v;
2051 struct unix_sock *u = unix_sk(s);
2052 unix_state_lock(s);
2053
2054 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2055 s,
2056 atomic_read(&s->sk_refcnt),
2057 0,
2058 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2059 s->sk_type,
2060 s->sk_socket ?
2061 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2062 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2063 sock_i_ino(s));
2064
2065 if (u->addr) {
2066 int i, len;
2067 seq_putc(seq, ' ');
2068
2069 i = 0;
2070 len = u->addr->len - sizeof(short);
2071 if (!UNIX_ABSTRACT(s))
2072 len--;
2073 else {
2074 seq_putc(seq, '@');
2075 i++;
2076 }
2077 for ( ; i < len; i++)
2078 seq_putc(seq, u->addr->name->sun_path[i]);
2079 }
2080 unix_state_unlock(s);
2081 seq_putc(seq, '\n');
2082 }
2083
2084 return 0;
2085 }
2086
2087 static const struct seq_operations unix_seq_ops = {
2088 .start = unix_seq_start,
2089 .next = unix_seq_next,
2090 .stop = unix_seq_stop,
2091 .show = unix_seq_show,
2092 };
2093
2094
2095 static int unix_seq_open(struct inode *inode, struct file *file)
2096 {
2097 return seq_open_private(file, &unix_seq_ops, sizeof(int));
2098 }
2099
2100 static const struct file_operations unix_seq_fops = {
2101 .owner = THIS_MODULE,
2102 .open = unix_seq_open,
2103 .read = seq_read,
2104 .llseek = seq_lseek,
2105 .release = seq_release_private,
2106 };
2107
2108 #endif
2109
2110 static struct net_proto_family unix_family_ops = {
2111 .family = PF_UNIX,
2112 .create = unix_create,
2113 .owner = THIS_MODULE,
2114 };
2115
2116 static int __init af_unix_init(void)
2117 {
2118 int rc = -1;
2119 struct sk_buff *dummy_skb;
2120
2121 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2122
2123 rc = proto_register(&unix_proto, 1);
2124 if (rc != 0) {
2125 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2126 __FUNCTION__);
2127 goto out;
2128 }
2129
2130 sock_register(&unix_family_ops);
2131 #ifdef CONFIG_PROC_FS
2132 proc_net_fops_create(&init_net, "unix", 0, &unix_seq_fops);
2133 #endif
2134 unix_sysctl_register();
2135 out:
2136 return rc;
2137 }
2138
2139 static void __exit af_unix_exit(void)
2140 {
2141 sock_unregister(PF_UNIX);
2142 unix_sysctl_unregister();
2143 proc_net_remove(&init_net, "unix");
2144 proto_unregister(&unix_proto);
2145 }
2146
2147 module_init(af_unix_init);
2148 module_exit(af_unix_exit);
2149
2150 MODULE_LICENSE("GPL");
2151 MODULE_ALIAS_NETPROTO(PF_UNIX);
Mirror for ia64_pv_ops.git