projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
[NET]: Make /proc/net per network namespace
[deliverable/linux.git] / net / unix / af_unix.c
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(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 (atomic_read(&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 = current->tgid;
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 socket *sock)
598 {
599 struct sock *sk = NULL;
600 struct unix_sock *u;
601
602 if (atomic_read(&unix_nr_socks) >= 2*get_max_files())
603 goto out;
604
605 sk = sk_alloc(PF_UNIX, GFP_KERNEL, &unix_proto, 1);
606 if (!sk)
607 goto out;
608
609 atomic_inc(&unix_nr_socks);
610
611 sock_init_data(sock,sk);
612 lockdep_set_class(&sk->sk_receive_queue.lock,
613 &af_unix_sk_receive_queue_lock_key);
614
615 sk->sk_write_space = unix_write_space;
616 sk->sk_max_ack_backlog = sysctl_unix_max_dgram_qlen;
617 sk->sk_destruct = unix_sock_destructor;
618 u = unix_sk(sk);
619 u->dentry = NULL;
620 u->mnt = NULL;
621 spin_lock_init(&u->lock);
622 atomic_set(&u->inflight, 0);
623 INIT_LIST_HEAD(&u->link);
624 mutex_init(&u->readlock); /* single task reading lock */
625 init_waitqueue_head(&u->peer_wait);
626 unix_insert_socket(unix_sockets_unbound, sk);
627 out:
628 return sk;
629 }
630
631 static int unix_create(struct socket *sock, int protocol)
632 {
633 if (protocol && protocol != PF_UNIX)
634 return -EPROTONOSUPPORT;
635
636 sock->state = SS_UNCONNECTED;
637
638 switch (sock->type) {
639 case SOCK_STREAM:
640 sock->ops = &unix_stream_ops;
641 break;
642 /*
643 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
644 * nothing uses it.
645 */
646 case SOCK_RAW:
647 sock->type=SOCK_DGRAM;
648 case SOCK_DGRAM:
649 sock->ops = &unix_dgram_ops;
650 break;
651 case SOCK_SEQPACKET:
652 sock->ops = &unix_seqpacket_ops;
653 break;
654 default:
655 return -ESOCKTNOSUPPORT;
656 }
657
658 return unix_create1(sock) ? 0 : -ENOMEM;
659 }
660
661 static int unix_release(struct socket *sock)
662 {
663 struct sock *sk = sock->sk;
664
665 if (!sk)
666 return 0;
667
668 sock->sk = NULL;
669
670 return unix_release_sock (sk, 0);
671 }
672
673 static int unix_autobind(struct socket *sock)
674 {
675 struct sock *sk = sock->sk;
676 struct unix_sock *u = unix_sk(sk);
677 static u32 ordernum = 1;
678 struct unix_address * addr;
679 int err;
680
681 mutex_lock(&u->readlock);
682
683 err = 0;
684 if (u->addr)
685 goto out;
686
687 err = -ENOMEM;
688 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
689 if (!addr)
690 goto out;
691
692 addr->name->sun_family = AF_UNIX;
693 atomic_set(&addr->refcnt, 1);
694
695 retry:
696 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
697 addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0));
698
699 spin_lock(&unix_table_lock);
700 ordernum = (ordernum+1)&0xFFFFF;
701
702 if (__unix_find_socket_byname(addr->name, addr->len, sock->type,
703 addr->hash)) {
704 spin_unlock(&unix_table_lock);
705 /* Sanity yield. It is unusual case, but yet... */
706 if (!(ordernum&0xFF))
707 yield();
708 goto retry;
709 }
710 addr->hash ^= sk->sk_type;
711
712 __unix_remove_socket(sk);
713 u->addr = addr;
714 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
715 spin_unlock(&unix_table_lock);
716 err = 0;
717
718 out: mutex_unlock(&u->readlock);
719 return err;
720 }
721
722 static struct sock *unix_find_other(struct sockaddr_un *sunname, int len,
723 int type, unsigned hash, int *error)
724 {
725 struct sock *u;
726 struct nameidata nd;
727 int err = 0;
728
729 if (sunname->sun_path[0]) {
730 err = path_lookup(sunname->sun_path, LOOKUP_FOLLOW, &nd);
731 if (err)
732 goto fail;
733 err = vfs_permission(&nd, MAY_WRITE);
734 if (err)
735 goto put_fail;
736
737 err = -ECONNREFUSED;
738 if (!S_ISSOCK(nd.dentry->d_inode->i_mode))
739 goto put_fail;
740 u=unix_find_socket_byinode(nd.dentry->d_inode);
741 if (!u)
742 goto put_fail;
743
744 if (u->sk_type == type)
745 touch_atime(nd.mnt, nd.dentry);
746
747 path_release(&nd);
748
749 err=-EPROTOTYPE;
750 if (u->sk_type != type) {
751 sock_put(u);
752 goto fail;
753 }
754 } else {
755 err = -ECONNREFUSED;
756 u=unix_find_socket_byname(sunname, len, type, hash);
757 if (u) {
758 struct dentry *dentry;
759 dentry = unix_sk(u)->dentry;
760 if (dentry)
761 touch_atime(unix_sk(u)->mnt, dentry);
762 } else
763 goto fail;
764 }
765 return u;
766
767 put_fail:
768 path_release(&nd);
769 fail:
770 *error=err;
771 return NULL;
772 }
773
774
775 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
776 {
777 struct sock *sk = sock->sk;
778 struct unix_sock *u = unix_sk(sk);
779 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
780 struct dentry * dentry = NULL;
781 struct nameidata nd;
782 int err;
783 unsigned hash;
784 struct unix_address *addr;
785 struct hlist_head *list;
786
787 err = -EINVAL;
788 if (sunaddr->sun_family != AF_UNIX)
789 goto out;
790
791 if (addr_len==sizeof(short)) {
792 err = unix_autobind(sock);
793 goto out;
794 }
795
796 err = unix_mkname(sunaddr, addr_len, &hash);
797 if (err < 0)
798 goto out;
799 addr_len = err;
800
801 mutex_lock(&u->readlock);
802
803 err = -EINVAL;
804 if (u->addr)
805 goto out_up;
806
807 err = -ENOMEM;
808 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
809 if (!addr)
810 goto out_up;
811
812 memcpy(addr->name, sunaddr, addr_len);
813 addr->len = addr_len;
814 addr->hash = hash ^ sk->sk_type;
815 atomic_set(&addr->refcnt, 1);
816
817 if (sunaddr->sun_path[0]) {
818 unsigned int mode;
819 err = 0;
820 /*
821 * Get the parent directory, calculate the hash for last
822 * component.
823 */
824 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
825 if (err)
826 goto out_mknod_parent;
827
828 dentry = lookup_create(&nd, 0);
829 err = PTR_ERR(dentry);
830 if (IS_ERR(dentry))
831 goto out_mknod_unlock;
832
833 /*
834 * All right, let's create it.
835 */
836 mode = S_IFSOCK |
837 (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
838 err = vfs_mknod(nd.dentry->d_inode, dentry, mode, 0);
839 if (err)
840 goto out_mknod_dput;
841 mutex_unlock(&nd.dentry->d_inode->i_mutex);
842 dput(nd.dentry);
843 nd.dentry = dentry;
844
845 addr->hash = UNIX_HASH_SIZE;
846 }
847
848 spin_lock(&unix_table_lock);
849
850 if (!sunaddr->sun_path[0]) {
851 err = -EADDRINUSE;
852 if (__unix_find_socket_byname(sunaddr, addr_len,
853 sk->sk_type, hash)) {
854 unix_release_addr(addr);
855 goto out_unlock;
856 }
857
858 list = &unix_socket_table[addr->hash];
859 } else {
860 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
861 u->dentry = nd.dentry;
862 u->mnt = nd.mnt;
863 }
864
865 err = 0;
866 __unix_remove_socket(sk);
867 u->addr = addr;
868 __unix_insert_socket(list, sk);
869
870 out_unlock:
871 spin_unlock(&unix_table_lock);
872 out_up:
873 mutex_unlock(&u->readlock);
874 out:
875 return err;
876
877 out_mknod_dput:
878 dput(dentry);
879 out_mknod_unlock:
880 mutex_unlock(&nd.dentry->d_inode->i_mutex);
881 path_release(&nd);
882 out_mknod_parent:
883 if (err==-EEXIST)
884 err=-EADDRINUSE;
885 unix_release_addr(addr);
886 goto out_up;
887 }
888
889 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
890 {
891 if (unlikely(sk1 == sk2) || !sk2) {
892 unix_state_lock(sk1);
893 return;
894 }
895 if (sk1 < sk2) {
896 unix_state_lock(sk1);
897 unix_state_lock_nested(sk2);
898 } else {
899 unix_state_lock(sk2);
900 unix_state_lock_nested(sk1);
901 }
902 }
903
904 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
905 {
906 if (unlikely(sk1 == sk2) || !sk2) {
907 unix_state_unlock(sk1);
908 return;
909 }
910 unix_state_unlock(sk1);
911 unix_state_unlock(sk2);
912 }
913
914 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
915 int alen, int flags)
916 {
917 struct sock *sk = sock->sk;
918 struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr;
919 struct sock *other;
920 unsigned hash;
921 int err;
922
923 if (addr->sa_family != AF_UNSPEC) {
924 err = unix_mkname(sunaddr, alen, &hash);
925 if (err < 0)
926 goto out;
927 alen = err;
928
929 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
930 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
931 goto out;
932
933 restart:
934 other=unix_find_other(sunaddr, alen, sock->type, hash, &err);
935 if (!other)
936 goto out;
937
938 unix_state_double_lock(sk, other);
939
940 /* Apparently VFS overslept socket death. Retry. */
941 if (sock_flag(other, SOCK_DEAD)) {
942 unix_state_double_unlock(sk, other);
943 sock_put(other);
944 goto restart;
945 }
946
947 err = -EPERM;
948 if (!unix_may_send(sk, other))
949 goto out_unlock;
950
951 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
952 if (err)
953 goto out_unlock;
954
955 } else {
956 /*
957 * 1003.1g breaking connected state with AF_UNSPEC
958 */
959 other = NULL;
960 unix_state_double_lock(sk, other);
961 }
962
963 /*
964 * If it was connected, reconnect.
965 */
966 if (unix_peer(sk)) {
967 struct sock *old_peer = unix_peer(sk);
968 unix_peer(sk)=other;
969 unix_state_double_unlock(sk, other);
970
971 if (other != old_peer)
972 unix_dgram_disconnected(sk, old_peer);
973 sock_put(old_peer);
974 } else {
975 unix_peer(sk)=other;
976 unix_state_double_unlock(sk, other);
977 }
978 return 0;
979
980 out_unlock:
981 unix_state_double_unlock(sk, other);
982 sock_put(other);
983 out:
984 return err;
985 }
986
987 static long unix_wait_for_peer(struct sock *other, long timeo)
988 {
989 struct unix_sock *u = unix_sk(other);
990 int sched;
991 DEFINE_WAIT(wait);
992
993 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
994
995 sched = !sock_flag(other, SOCK_DEAD) &&
996 !(other->sk_shutdown & RCV_SHUTDOWN) &&
997 (skb_queue_len(&other->sk_receive_queue) >
998 other->sk_max_ack_backlog);
999
1000 unix_state_unlock(other);
1001
1002 if (sched)
1003 timeo = schedule_timeout(timeo);
1004
1005 finish_wait(&u->peer_wait, &wait);
1006 return timeo;
1007 }
1008
1009 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1010 int addr_len, int flags)
1011 {
1012 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1013 struct sock *sk = sock->sk;
1014 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1015 struct sock *newsk = NULL;
1016 struct sock *other = NULL;
1017 struct sk_buff *skb = NULL;
1018 unsigned hash;
1019 int st;
1020 int err;
1021 long timeo;
1022
1023 err = unix_mkname(sunaddr, addr_len, &hash);
1024 if (err < 0)
1025 goto out;
1026 addr_len = err;
1027
1028 if (test_bit(SOCK_PASSCRED, &sock->flags)
1029 && !u->addr && (err = unix_autobind(sock)) != 0)
1030 goto out;
1031
1032 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1033
1034 /* First of all allocate resources.
1035 If we will make it after state is locked,
1036 we will have to recheck all again in any case.
1037 */
1038
1039 err = -ENOMEM;
1040
1041 /* create new sock for complete connection */
1042 newsk = unix_create1(NULL);
1043 if (newsk == NULL)
1044 goto out;
1045
1046 /* Allocate skb for sending to listening sock */
1047 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1048 if (skb == NULL)
1049 goto out;
1050
1051 restart:
1052 /* Find listening sock. */
1053 other = unix_find_other(sunaddr, addr_len, sk->sk_type, hash, &err);
1054 if (!other)
1055 goto out;
1056
1057 /* Latch state of peer */
1058 unix_state_lock(other);
1059
1060 /* Apparently VFS overslept socket death. Retry. */
1061 if (sock_flag(other, SOCK_DEAD)) {
1062 unix_state_unlock(other);
1063 sock_put(other);
1064 goto restart;
1065 }
1066
1067 err = -ECONNREFUSED;
1068 if (other->sk_state != TCP_LISTEN)
1069 goto out_unlock;
1070
1071 if (skb_queue_len(&other->sk_receive_queue) >
1072 other->sk_max_ack_backlog) {
1073 err = -EAGAIN;
1074 if (!timeo)
1075 goto out_unlock;
1076
1077 timeo = unix_wait_for_peer(other, timeo);
1078
1079 err = sock_intr_errno(timeo);
1080 if (signal_pending(current))
1081 goto out;
1082 sock_put(other);
1083 goto restart;
1084 }
1085
1086 /* Latch our state.
1087
1088 It is tricky place. We need to grab write lock and cannot
1089 drop lock on peer. It is dangerous because deadlock is
1090 possible. Connect to self case and simultaneous
1091 attempt to connect are eliminated by checking socket
1092 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1093 check this before attempt to grab lock.
1094
1095 Well, and we have to recheck the state after socket locked.
1096 */
1097 st = sk->sk_state;
1098
1099 switch (st) {
1100 case TCP_CLOSE:
1101 /* This is ok... continue with connect */
1102 break;
1103 case TCP_ESTABLISHED:
1104 /* Socket is already connected */
1105 err = -EISCONN;
1106 goto out_unlock;
1107 default:
1108 err = -EINVAL;
1109 goto out_unlock;
1110 }
1111
1112 unix_state_lock_nested(sk);
1113
1114 if (sk->sk_state != st) {
1115 unix_state_unlock(sk);
1116 unix_state_unlock(other);
1117 sock_put(other);
1118 goto restart;
1119 }
1120
1121 err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1122 if (err) {
1123 unix_state_unlock(sk);
1124 goto out_unlock;
1125 }
1126
1127 /* The way is open! Fastly set all the necessary fields... */
1128
1129 sock_hold(sk);
1130 unix_peer(newsk) = sk;
1131 newsk->sk_state = TCP_ESTABLISHED;
1132 newsk->sk_type = sk->sk_type;
1133 newsk->sk_peercred.pid = current->tgid;
1134 newsk->sk_peercred.uid = current->euid;
1135 newsk->sk_peercred.gid = current->egid;
1136 newu = unix_sk(newsk);
1137 newsk->sk_sleep = &newu->peer_wait;
1138 otheru = unix_sk(other);
1139
1140 /* copy address information from listening to new sock*/
1141 if (otheru->addr) {
1142 atomic_inc(&otheru->addr->refcnt);
1143 newu->addr = otheru->addr;
1144 }
1145 if (otheru->dentry) {
1146 newu->dentry = dget(otheru->dentry);
1147 newu->mnt = mntget(otheru->mnt);
1148 }
1149
1150 /* Set credentials */
1151 sk->sk_peercred = other->sk_peercred;
1152
1153 sock->state = SS_CONNECTED;
1154 sk->sk_state = TCP_ESTABLISHED;
1155 sock_hold(newsk);
1156
1157 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1158 unix_peer(sk) = newsk;
1159
1160 unix_state_unlock(sk);
1161
1162 /* take ten and and send info to listening sock */
1163 spin_lock(&other->sk_receive_queue.lock);
1164 __skb_queue_tail(&other->sk_receive_queue, skb);
1165 spin_unlock(&other->sk_receive_queue.lock);
1166 unix_state_unlock(other);
1167 other->sk_data_ready(other, 0);
1168 sock_put(other);
1169 return 0;
1170
1171 out_unlock:
1172 if (other)
1173 unix_state_unlock(other);
1174
1175 out:
1176 if (skb)
1177 kfree_skb(skb);
1178 if (newsk)
1179 unix_release_sock(newsk, 0);
1180 if (other)
1181 sock_put(other);
1182 return err;
1183 }
1184
1185 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1186 {
1187 struct sock *ska=socka->sk, *skb = sockb->sk;
1188
1189 /* Join our sockets back to back */
1190 sock_hold(ska);
1191 sock_hold(skb);
1192 unix_peer(ska)=skb;
1193 unix_peer(skb)=ska;
1194 ska->sk_peercred.pid = skb->sk_peercred.pid = current->tgid;
1195 ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1196 ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1197
1198 if (ska->sk_type != SOCK_DGRAM) {
1199 ska->sk_state = TCP_ESTABLISHED;
1200 skb->sk_state = TCP_ESTABLISHED;
1201 socka->state = SS_CONNECTED;
1202 sockb->state = SS_CONNECTED;
1203 }
1204 return 0;
1205 }
1206
1207 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1208 {
1209 struct sock *sk = sock->sk;
1210 struct sock *tsk;
1211 struct sk_buff *skb;
1212 int err;
1213
1214 err = -EOPNOTSUPP;
1215 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
1216 goto out;
1217
1218 err = -EINVAL;
1219 if (sk->sk_state != TCP_LISTEN)
1220 goto out;
1221
1222 /* If socket state is TCP_LISTEN it cannot change (for now...),
1223 * so that no locks are necessary.
1224 */
1225
1226 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1227 if (!skb) {
1228 /* This means receive shutdown. */
1229 if (err == 0)
1230 err = -EINVAL;
1231 goto out;
1232 }
1233
1234 tsk = skb->sk;
1235 skb_free_datagram(sk, skb);
1236 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1237
1238 /* attach accepted sock to socket */
1239 unix_state_lock(tsk);
1240 newsock->state = SS_CONNECTED;
1241 sock_graft(tsk, newsock);
1242 unix_state_unlock(tsk);
1243 return 0;
1244
1245 out:
1246 return err;
1247 }
1248
1249
1250 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1251 {
1252 struct sock *sk = sock->sk;
1253 struct unix_sock *u;
1254 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1255 int err = 0;
1256
1257 if (peer) {
1258 sk = unix_peer_get(sk);
1259
1260 err = -ENOTCONN;
1261 if (!sk)
1262 goto out;
1263 err = 0;
1264 } else {
1265 sock_hold(sk);
1266 }
1267
1268 u = unix_sk(sk);
1269 unix_state_lock(sk);
1270 if (!u->addr) {
1271 sunaddr->sun_family = AF_UNIX;
1272 sunaddr->sun_path[0] = 0;
1273 *uaddr_len = sizeof(short);
1274 } else {
1275 struct unix_address *addr = u->addr;
1276
1277 *uaddr_len = addr->len;
1278 memcpy(sunaddr, addr->name, *uaddr_len);
1279 }
1280 unix_state_unlock(sk);
1281 sock_put(sk);
1282 out:
1283 return err;
1284 }
1285
1286 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1287 {
1288 int i;
1289
1290 scm->fp = UNIXCB(skb).fp;
1291 skb->destructor = sock_wfree;
1292 UNIXCB(skb).fp = NULL;
1293
1294 for (i=scm->fp->count-1; i>=0; i--)
1295 unix_notinflight(scm->fp->fp[i]);
1296 }
1297
1298 static void unix_destruct_fds(struct sk_buff *skb)
1299 {
1300 struct scm_cookie scm;
1301 memset(&scm, 0, sizeof(scm));
1302 unix_detach_fds(&scm, skb);
1303
1304 /* Alas, it calls VFS */
1305 /* So fscking what? fput() had been SMP-safe since the last Summer */
1306 scm_destroy(&scm);
1307 sock_wfree(skb);
1308 }
1309
1310 static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1311 {
1312 int i;
1313 for (i=scm->fp->count-1; i>=0; i--)
1314 unix_inflight(scm->fp->fp[i]);
1315 UNIXCB(skb).fp = scm->fp;
1316 skb->destructor = unix_destruct_fds;
1317 scm->fp = NULL;
1318 }
1319
1320 /*
1321 * Send AF_UNIX data.
1322 */
1323
1324 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1325 struct msghdr *msg, size_t len)
1326 {
1327 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1328 struct sock *sk = sock->sk;
1329 struct unix_sock *u = unix_sk(sk);
1330 struct sockaddr_un *sunaddr=msg->msg_name;
1331 struct sock *other = NULL;
1332 int namelen = 0; /* fake GCC */
1333 int err;
1334 unsigned hash;
1335 struct sk_buff *skb;
1336 long timeo;
1337 struct scm_cookie tmp_scm;
1338
1339 if (NULL == siocb->scm)
1340 siocb->scm = &tmp_scm;
1341 err = scm_send(sock, msg, siocb->scm);
1342 if (err < 0)
1343 return err;
1344
1345 err = -EOPNOTSUPP;
1346 if (msg->msg_flags&MSG_OOB)
1347 goto out;
1348
1349 if (msg->msg_namelen) {
1350 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1351 if (err < 0)
1352 goto out;
1353 namelen = err;
1354 } else {
1355 sunaddr = NULL;
1356 err = -ENOTCONN;
1357 other = unix_peer_get(sk);
1358 if (!other)
1359 goto out;
1360 }
1361
1362 if (test_bit(SOCK_PASSCRED, &sock->flags)
1363 && !u->addr && (err = unix_autobind(sock)) != 0)
1364 goto out;
1365
1366 err = -EMSGSIZE;
1367 if (len > sk->sk_sndbuf - 32)
1368 goto out;
1369
1370 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1371 if (skb==NULL)
1372 goto out;
1373
1374 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1375 if (siocb->scm->fp)
1376 unix_attach_fds(siocb->scm, skb);
1377 unix_get_secdata(siocb->scm, skb);
1378
1379 skb_reset_transport_header(skb);
1380 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1381 if (err)
1382 goto out_free;
1383
1384 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1385
1386 restart:
1387 if (!other) {
1388 err = -ECONNRESET;
1389 if (sunaddr == NULL)
1390 goto out_free;
1391
1392 other = unix_find_other(sunaddr, namelen, sk->sk_type,
1393 hash, &err);
1394 if (other==NULL)
1395 goto out_free;
1396 }
1397
1398 unix_state_lock(other);
1399 err = -EPERM;
1400 if (!unix_may_send(sk, other))
1401 goto out_unlock;
1402
1403 if (sock_flag(other, SOCK_DEAD)) {
1404 /*
1405 * Check with 1003.1g - what should
1406 * datagram error
1407 */
1408 unix_state_unlock(other);
1409 sock_put(other);
1410
1411 err = 0;
1412 unix_state_lock(sk);
1413 if (unix_peer(sk) == other) {
1414 unix_peer(sk)=NULL;
1415 unix_state_unlock(sk);
1416
1417 unix_dgram_disconnected(sk, other);
1418 sock_put(other);
1419 err = -ECONNREFUSED;
1420 } else {
1421 unix_state_unlock(sk);
1422 }
1423
1424 other = NULL;
1425 if (err)
1426 goto out_free;
1427 goto restart;
1428 }
1429
1430 err = -EPIPE;
1431 if (other->sk_shutdown & RCV_SHUTDOWN)
1432 goto out_unlock;
1433
1434 if (sk->sk_type != SOCK_SEQPACKET) {
1435 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1436 if (err)
1437 goto out_unlock;
1438 }
1439
1440 if (unix_peer(other) != sk &&
1441 (skb_queue_len(&other->sk_receive_queue) >
1442 other->sk_max_ack_backlog)) {
1443 if (!timeo) {
1444 err = -EAGAIN;
1445 goto out_unlock;
1446 }
1447
1448 timeo = unix_wait_for_peer(other, timeo);
1449
1450 err = sock_intr_errno(timeo);
1451 if (signal_pending(current))
1452 goto out_free;
1453
1454 goto restart;
1455 }
1456
1457 skb_queue_tail(&other->sk_receive_queue, skb);
1458 unix_state_unlock(other);
1459 other->sk_data_ready(other, len);
1460 sock_put(other);
1461 scm_destroy(siocb->scm);
1462 return len;
1463
1464 out_unlock:
1465 unix_state_unlock(other);
1466 out_free:
1467 kfree_skb(skb);
1468 out:
1469 if (other)
1470 sock_put(other);
1471 scm_destroy(siocb->scm);
1472 return err;
1473 }
1474
1475
1476 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1477 struct msghdr *msg, size_t len)
1478 {
1479 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1480 struct sock *sk = sock->sk;
1481 struct sock *other = NULL;
1482 struct sockaddr_un *sunaddr=msg->msg_name;
1483 int err,size;
1484 struct sk_buff *skb;
1485 int sent=0;
1486 struct scm_cookie tmp_scm;
1487
1488 if (NULL == siocb->scm)
1489 siocb->scm = &tmp_scm;
1490 err = scm_send(sock, msg, siocb->scm);
1491 if (err < 0)
1492 return err;
1493
1494 err = -EOPNOTSUPP;
1495 if (msg->msg_flags&MSG_OOB)
1496 goto out_err;
1497
1498 if (msg->msg_namelen) {
1499 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1500 goto out_err;
1501 } else {
1502 sunaddr = NULL;
1503 err = -ENOTCONN;
1504 other = unix_peer(sk);
1505 if (!other)
1506 goto out_err;
1507 }
1508
1509 if (sk->sk_shutdown & SEND_SHUTDOWN)
1510 goto pipe_err;
1511
1512 while(sent < len)
1513 {
1514 /*
1515 * Optimisation for the fact that under 0.01% of X
1516 * messages typically need breaking up.
1517 */
1518
1519 size = len-sent;
1520
1521 /* Keep two messages in the pipe so it schedules better */
1522 if (size > ((sk->sk_sndbuf >> 1) - 64))
1523 size = (sk->sk_sndbuf >> 1) - 64;
1524
1525 if (size > SKB_MAX_ALLOC)
1526 size = SKB_MAX_ALLOC;
1527
1528 /*
1529 * Grab a buffer
1530 */
1531
1532 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
1533
1534 if (skb==NULL)
1535 goto out_err;
1536
1537 /*
1538 * If you pass two values to the sock_alloc_send_skb
1539 * it tries to grab the large buffer with GFP_NOFS
1540 * (which can fail easily), and if it fails grab the
1541 * fallback size buffer which is under a page and will
1542 * succeed. [Alan]
1543 */
1544 size = min_t(int, size, skb_tailroom(skb));
1545
1546 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1547 if (siocb->scm->fp)
1548 unix_attach_fds(siocb->scm, skb);
1549
1550 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
1551 kfree_skb(skb);
1552 goto out_err;
1553 }
1554
1555 unix_state_lock(other);
1556
1557 if (sock_flag(other, SOCK_DEAD) ||
1558 (other->sk_shutdown & RCV_SHUTDOWN))
1559 goto pipe_err_free;
1560
1561 skb_queue_tail(&other->sk_receive_queue, skb);
1562 unix_state_unlock(other);
1563 other->sk_data_ready(other, size);
1564 sent+=size;
1565 }
1566
1567 scm_destroy(siocb->scm);
1568 siocb->scm = NULL;
1569
1570 return sent;
1571
1572 pipe_err_free:
1573 unix_state_unlock(other);
1574 kfree_skb(skb);
1575 pipe_err:
1576 if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL))
1577 send_sig(SIGPIPE,current,0);
1578 err = -EPIPE;
1579 out_err:
1580 scm_destroy(siocb->scm);
1581 siocb->scm = NULL;
1582 return sent ? : err;
1583 }
1584
1585 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1586 struct msghdr *msg, size_t len)
1587 {
1588 int err;
1589 struct sock *sk = sock->sk;
1590
1591 err = sock_error(sk);
1592 if (err)
1593 return err;
1594
1595 if (sk->sk_state != TCP_ESTABLISHED)
1596 return -ENOTCONN;
1597
1598 if (msg->msg_namelen)
1599 msg->msg_namelen = 0;
1600
1601 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1602 }
1603
1604 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1605 {
1606 struct unix_sock *u = unix_sk(sk);
1607
1608 msg->msg_namelen = 0;
1609 if (u->addr) {
1610 msg->msg_namelen = u->addr->len;
1611 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1612 }
1613 }
1614
1615 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1616 struct msghdr *msg, size_t size,
1617 int flags)
1618 {
1619 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1620 struct scm_cookie tmp_scm;
1621 struct sock *sk = sock->sk;
1622 struct unix_sock *u = unix_sk(sk);
1623 int noblock = flags & MSG_DONTWAIT;
1624 struct sk_buff *skb;
1625 int err;
1626
1627 err = -EOPNOTSUPP;
1628 if (flags&MSG_OOB)
1629 goto out;
1630
1631 msg->msg_namelen = 0;
1632
1633 mutex_lock(&u->readlock);
1634
1635 skb = skb_recv_datagram(sk, flags, noblock, &err);
1636 if (!skb)
1637 goto out_unlock;
1638
1639 wake_up_interruptible(&u->peer_wait);
1640
1641 if (msg->msg_name)
1642 unix_copy_addr(msg, skb->sk);
1643
1644 if (size > skb->len)
1645 size = skb->len;
1646 else if (size < skb->len)
1647 msg->msg_flags |= MSG_TRUNC;
1648
1649 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1650 if (err)
1651 goto out_free;
1652
1653 if (!siocb->scm) {
1654 siocb->scm = &tmp_scm;
1655 memset(&tmp_scm, 0, sizeof(tmp_scm));
1656 }
1657 siocb->scm->creds = *UNIXCREDS(skb);
1658 unix_set_secdata(siocb->scm, skb);
1659
1660 if (!(flags & MSG_PEEK))
1661 {
1662 if (UNIXCB(skb).fp)
1663 unix_detach_fds(siocb->scm, skb);
1664 }
1665 else
1666 {
1667 /* It is questionable: on PEEK we could:
1668 - do not return fds - good, but too simple 8)
1669 - return fds, and do not return them on read (old strategy,
1670 apparently wrong)
1671 - clone fds (I chose it for now, it is the most universal
1672 solution)
1673
1674 POSIX 1003.1g does not actually define this clearly
1675 at all. POSIX 1003.1g doesn't define a lot of things
1676 clearly however!
1677
1678 */
1679 if (UNIXCB(skb).fp)
1680 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1681 }
1682 err = size;
1683
1684 scm_recv(sock, msg, siocb->scm, flags);
1685
1686 out_free:
1687 skb_free_datagram(sk,skb);
1688 out_unlock:
1689 mutex_unlock(&u->readlock);
1690 out:
1691 return err;
1692 }
1693
1694 /*
1695 * Sleep until data has arrive. But check for races..
1696 */
1697
1698 static long unix_stream_data_wait(struct sock * sk, long timeo)
1699 {
1700 DEFINE_WAIT(wait);
1701
1702 unix_state_lock(sk);
1703
1704 for (;;) {
1705 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1706
1707 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1708 sk->sk_err ||
1709 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1710 signal_pending(current) ||
1711 !timeo)
1712 break;
1713
1714 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1715 unix_state_unlock(sk);
1716 timeo = schedule_timeout(timeo);
1717 unix_state_lock(sk);
1718 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1719 }
1720
1721 finish_wait(sk->sk_sleep, &wait);
1722 unix_state_unlock(sk);
1723 return timeo;
1724 }
1725
1726
1727
1728 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1729 struct msghdr *msg, size_t size,
1730 int flags)
1731 {
1732 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1733 struct scm_cookie tmp_scm;
1734 struct sock *sk = sock->sk;
1735 struct unix_sock *u = unix_sk(sk);
1736 struct sockaddr_un *sunaddr=msg->msg_name;
1737 int copied = 0;
1738 int check_creds = 0;
1739 int target;
1740 int err = 0;
1741 long timeo;
1742
1743 err = -EINVAL;
1744 if (sk->sk_state != TCP_ESTABLISHED)
1745 goto out;
1746
1747 err = -EOPNOTSUPP;
1748 if (flags&MSG_OOB)
1749 goto out;
1750
1751 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1752 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1753
1754 msg->msg_namelen = 0;
1755
1756 /* Lock the socket to prevent queue disordering
1757 * while sleeps in memcpy_tomsg
1758 */
1759
1760 if (!siocb->scm) {
1761 siocb->scm = &tmp_scm;
1762 memset(&tmp_scm, 0, sizeof(tmp_scm));
1763 }
1764
1765 mutex_lock(&u->readlock);
1766
1767 do
1768 {
1769 int chunk;
1770 struct sk_buff *skb;
1771
1772 unix_state_lock(sk);
1773 skb = skb_dequeue(&sk->sk_receive_queue);
1774 if (skb==NULL)
1775 {
1776 if (copied >= target)
1777 goto unlock;
1778
1779 /*
1780 * POSIX 1003.1g mandates this order.
1781 */
1782
1783 if ((err = sock_error(sk)) != 0)
1784 goto unlock;
1785 if (sk->sk_shutdown & RCV_SHUTDOWN)
1786 goto unlock;
1787
1788 unix_state_unlock(sk);
1789 err = -EAGAIN;
1790 if (!timeo)
1791 break;
1792 mutex_unlock(&u->readlock);
1793
1794 timeo = unix_stream_data_wait(sk, timeo);
1795
1796 if (signal_pending(current)) {
1797 err = sock_intr_errno(timeo);
1798 goto out;
1799 }
1800 mutex_lock(&u->readlock);
1801 continue;
1802 unlock:
1803 unix_state_unlock(sk);
1804 break;
1805 }
1806 unix_state_unlock(sk);
1807
1808 if (check_creds) {
1809 /* Never glue messages from different writers */
1810 if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) {
1811 skb_queue_head(&sk->sk_receive_queue, skb);
1812 break;
1813 }
1814 } else {
1815 /* Copy credentials */
1816 siocb->scm->creds = *UNIXCREDS(skb);
1817 check_creds = 1;
1818 }
1819
1820 /* Copy address just once */
1821 if (sunaddr)
1822 {
1823 unix_copy_addr(msg, skb->sk);
1824 sunaddr = NULL;
1825 }
1826
1827 chunk = min_t(unsigned int, skb->len, size);
1828 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1829 skb_queue_head(&sk->sk_receive_queue, skb);
1830 if (copied == 0)
1831 copied = -EFAULT;
1832 break;
1833 }
1834 copied += chunk;
1835 size -= chunk;
1836
1837 /* Mark read part of skb as used */
1838 if (!(flags & MSG_PEEK))
1839 {
1840 skb_pull(skb, chunk);
1841
1842 if (UNIXCB(skb).fp)
1843 unix_detach_fds(siocb->scm, skb);
1844
1845 /* put the skb back if we didn't use it up.. */
1846 if (skb->len)
1847 {
1848 skb_queue_head(&sk->sk_receive_queue, skb);
1849 break;
1850 }
1851
1852 kfree_skb(skb);
1853
1854 if (siocb->scm->fp)
1855 break;
1856 }
1857 else
1858 {
1859 /* It is questionable, see note in unix_dgram_recvmsg.
1860 */
1861 if (UNIXCB(skb).fp)
1862 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1863
1864 /* put message back and return */
1865 skb_queue_head(&sk->sk_receive_queue, skb);
1866 break;
1867 }
1868 } while (size);
1869
1870 mutex_unlock(&u->readlock);
1871 scm_recv(sock, msg, siocb->scm, flags);
1872 out:
1873 return copied ? : err;
1874 }
1875
1876 static int unix_shutdown(struct socket *sock, int mode)
1877 {
1878 struct sock *sk = sock->sk;
1879 struct sock *other;
1880
1881 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1882
1883 if (mode) {
1884 unix_state_lock(sk);
1885 sk->sk_shutdown |= mode;
1886 other=unix_peer(sk);
1887 if (other)
1888 sock_hold(other);
1889 unix_state_unlock(sk);
1890 sk->sk_state_change(sk);
1891
1892 if (other &&
1893 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1894
1895 int peer_mode = 0;
1896
1897 if (mode&RCV_SHUTDOWN)
1898 peer_mode |= SEND_SHUTDOWN;
1899 if (mode&SEND_SHUTDOWN)
1900 peer_mode |= RCV_SHUTDOWN;
1901 unix_state_lock(other);
1902 other->sk_shutdown |= peer_mode;
1903 unix_state_unlock(other);
1904 other->sk_state_change(other);
1905 read_lock(&other->sk_callback_lock);
1906 if (peer_mode == SHUTDOWN_MASK)
1907 sk_wake_async(other,1,POLL_HUP);
1908 else if (peer_mode & RCV_SHUTDOWN)
1909 sk_wake_async(other,1,POLL_IN);
1910 read_unlock(&other->sk_callback_lock);
1911 }
1912 if (other)
1913 sock_put(other);
1914 }
1915 return 0;
1916 }
1917
1918 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1919 {
1920 struct sock *sk = sock->sk;
1921 long amount=0;
1922 int err;
1923
1924 switch(cmd)
1925 {
1926 case SIOCOUTQ:
1927 amount = atomic_read(&sk->sk_wmem_alloc);
1928 err = put_user(amount, (int __user *)arg);
1929 break;
1930 case SIOCINQ:
1931 {
1932 struct sk_buff *skb;
1933
1934 if (sk->sk_state == TCP_LISTEN) {
1935 err = -EINVAL;
1936 break;
1937 }
1938
1939 spin_lock(&sk->sk_receive_queue.lock);
1940 if (sk->sk_type == SOCK_STREAM ||
1941 sk->sk_type == SOCK_SEQPACKET) {
1942 skb_queue_walk(&sk->sk_receive_queue, skb)
1943 amount += skb->len;
1944 } else {
1945 skb = skb_peek(&sk->sk_receive_queue);
1946 if (skb)
1947 amount=skb->len;
1948 }
1949 spin_unlock(&sk->sk_receive_queue.lock);
1950 err = put_user(amount, (int __user *)arg);
1951 break;
1952 }
1953
1954 default:
1955 err = -ENOIOCTLCMD;
1956 break;
1957 }
1958 return err;
1959 }
1960
1961 static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait)
1962 {
1963 struct sock *sk = sock->sk;
1964 unsigned int mask;
1965
1966 poll_wait(file, sk->sk_sleep, wait);
1967 mask = 0;
1968
1969 /* exceptional events? */
1970 if (sk->sk_err)
1971 mask |= POLLERR;
1972 if (sk->sk_shutdown == SHUTDOWN_MASK)
1973 mask |= POLLHUP;
1974 if (sk->sk_shutdown & RCV_SHUTDOWN)
1975 mask |= POLLRDHUP;
1976
1977 /* readable? */
1978 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1979 (sk->sk_shutdown & RCV_SHUTDOWN))
1980 mask |= POLLIN | POLLRDNORM;
1981
1982 /* Connection-based need to check for termination and startup */
1983 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE)
1984 mask |= POLLHUP;
1985
1986 /*
1987 * we set writable also when the other side has shut down the
1988 * connection. This prevents stuck sockets.
1989 */
1990 if (unix_writable(sk))
1991 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
1992
1993 return mask;
1994 }
1995
1996
1997 #ifdef CONFIG_PROC_FS
1998 static struct sock *unix_seq_idx(int *iter, loff_t pos)
1999 {
2000 loff_t off = 0;
2001 struct sock *s;
2002
2003 for (s = first_unix_socket(iter); s; s = next_unix_socket(iter, s)) {
2004 if (off == pos)
2005 return s;
2006 ++off;
2007 }
2008 return NULL;
2009 }
2010
2011
2012 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2013 {
2014 spin_lock(&unix_table_lock);
2015 return *pos ? unix_seq_idx(seq->private, *pos - 1) : ((void *) 1);
2016 }
2017
2018 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2019 {
2020 ++*pos;
2021
2022 if (v == (void *)1)
2023 return first_unix_socket(seq->private);
2024 return next_unix_socket(seq->private, v);
2025 }
2026
2027 static void unix_seq_stop(struct seq_file *seq, void *v)
2028 {
2029 spin_unlock(&unix_table_lock);
2030 }
2031
2032 static int unix_seq_show(struct seq_file *seq, void *v)
2033 {
2034
2035 if (v == (void *)1)
2036 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2037 "Inode Path\n");
2038 else {
2039 struct sock *s = v;
2040 struct unix_sock *u = unix_sk(s);
2041 unix_state_lock(s);
2042
2043 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2044 s,
2045 atomic_read(&s->sk_refcnt),
2046 0,
2047 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2048 s->sk_type,
2049 s->sk_socket ?
2050 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2051 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2052 sock_i_ino(s));
2053
2054 if (u->addr) {
2055 int i, len;
2056 seq_putc(seq, ' ');
2057
2058 i = 0;
2059 len = u->addr->len - sizeof(short);
2060 if (!UNIX_ABSTRACT(s))
2061 len--;
2062 else {
2063 seq_putc(seq, '@');
2064 i++;
2065 }
2066 for ( ; i < len; i++)
2067 seq_putc(seq, u->addr->name->sun_path[i]);
2068 }
2069 unix_state_unlock(s);
2070 seq_putc(seq, '\n');
2071 }
2072
2073 return 0;
2074 }
2075
2076 static const struct seq_operations unix_seq_ops = {
2077 .start = unix_seq_start,
2078 .next = unix_seq_next,
2079 .stop = unix_seq_stop,
2080 .show = unix_seq_show,
2081 };
2082
2083
2084 static int unix_seq_open(struct inode *inode, struct file *file)
2085 {
2086 struct seq_file *seq;
2087 int rc = -ENOMEM;
2088 int *iter = kmalloc(sizeof(int), GFP_KERNEL);
2089
2090 if (!iter)
2091 goto out;
2092
2093 rc = seq_open(file, &unix_seq_ops);
2094 if (rc)
2095 goto out_kfree;
2096
2097 seq = file->private_data;
2098 seq->private = iter;
2099 *iter = 0;
2100 out:
2101 return rc;
2102 out_kfree:
2103 kfree(iter);
2104 goto out;
2105 }
2106
2107 static const struct file_operations unix_seq_fops = {
2108 .owner = THIS_MODULE,
2109 .open = unix_seq_open,
2110 .read = seq_read,
2111 .llseek = seq_lseek,
2112 .release = seq_release_private,
2113 };
2114
2115 #endif
2116
2117 static struct net_proto_family unix_family_ops = {
2118 .family = PF_UNIX,
2119 .create = unix_create,
2120 .owner = THIS_MODULE,
2121 };
2122
2123 static int __init af_unix_init(void)
2124 {
2125 int rc = -1;
2126 struct sk_buff *dummy_skb;
2127
2128 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2129
2130 rc = proto_register(&unix_proto, 1);
2131 if (rc != 0) {
2132 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2133 __FUNCTION__);
2134 goto out;
2135 }
2136
2137 sock_register(&unix_family_ops);
2138 #ifdef CONFIG_PROC_FS
2139 proc_net_fops_create(&init_net, "unix", 0, &unix_seq_fops);
2140 #endif
2141 unix_sysctl_register();
2142 out:
2143 return rc;
2144 }
2145
2146 static void __exit af_unix_exit(void)
2147 {
2148 sock_unregister(PF_UNIX);
2149 unix_sysctl_unregister();
2150 proc_net_remove(&init_net, "unix");
2151 proto_unregister(&unix_proto);
2152 }
2153
2154 module_init(af_unix_init);
2155 module_exit(af_unix_exit);
2156
2157 MODULE_LICENSE("GPL");
2158 MODULE_ALIAS_NETPROTO(PF_UNIX);
Linux kernel - Deliverables
This page took 0.084539 seconds and 5 git commands to generate.